4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2016 STRATO AG. All rights reserved.
27 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
31 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
32 * Use is subject to license terms.
36 * Copyright 1983,1984,1985,1986,1987,1988,1989 AT&T.
41 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
44 #include <sys/param.h>
45 #include <sys/types.h>
46 #include <sys/systm.h>
49 #include <sys/vnode.h>
51 #include <sys/vfs_opreg.h>
53 #include <sys/filio.h>
57 #include <sys/pathname.h>
58 #include <sys/dirent.h>
59 #include <sys/debug.h>
60 #include <sys/vmsystm.h>
61 #include <sys/fcntl.h>
62 #include <sys/flock.h>
64 #include <sys/errno.h>
65 #include <sys/strsubr.h>
66 #include <sys/sysmacros.h>
68 #include <sys/cmn_err.h>
69 #include <sys/pathconf.h>
70 #include <sys/utsname.h>
73 #include <sys/systeminfo.h>
74 #include <sys/policy.h>
80 #include <rpc/types.h>
85 #include <nfs/nfs_clnt.h>
86 #include <nfs/nfs_acl.h>
89 #include <nfs/nfs4_kprot.h>
90 #include <nfs/rnode4.h>
91 #include <nfs/nfs4_clnt.h>
98 #include <vm/seg_map.h>
99 #include <vm/seg_kpm.h>
100 #include <vm/seg_vn.h>
102 #include <sys/fs_subr.h>
105 #include <sys/int_fmtio.h>
106 #include <sys/fs/autofs.h>
109 nfs4_ga_res_t
*di_garp
;
111 hrtime_t di_time_call
;
114 typedef enum nfs4_acl_op
{
119 static struct lm_sysid
*nfs4_find_sysid(mntinfo4_t
*);
121 static void nfs4_update_dircaches(change_info4
*, vnode_t
*, vnode_t
*,
122 char *, dirattr_info_t
*);
124 static void nfs4close_otw(rnode4_t
*, cred_t
*, nfs4_open_owner_t
*,
125 nfs4_open_stream_t
*, int *, int *, nfs4_close_type_t
,
126 nfs4_error_t
*, int *);
127 static int nfs4_rdwrlbn(vnode_t
*, page_t
*, uoff_t
, size_t, int,
129 static int nfs4write(vnode_t
*, caddr_t
, uoff_t
, int, cred_t
*,
131 static int nfs4read(vnode_t
*, caddr_t
, offset_t
, int, size_t *,
132 cred_t
*, bool_t
, struct uio
*);
133 static int nfs4setattr(vnode_t
*, struct vattr
*, int, cred_t
*,
135 static int nfs4openattr(vnode_t
*, vnode_t
**, int, cred_t
*);
136 static int nfs4lookup(vnode_t
*, char *, vnode_t
**, cred_t
*, int);
137 static int nfs4lookup_xattr(vnode_t
*, char *, vnode_t
**, int, cred_t
*);
138 static int nfs4lookupvalidate_otw(vnode_t
*, char *, vnode_t
**, cred_t
*);
139 static int nfs4lookupnew_otw(vnode_t
*, char *, vnode_t
**, cred_t
*);
140 static int nfs4mknod(vnode_t
*, char *, struct vattr
*, enum vcexcl
,
141 int, vnode_t
**, cred_t
*);
142 static int nfs4open_otw(vnode_t
*, char *, struct vattr
*, vnode_t
**,
143 cred_t
*, int, int, enum createmode4
, int);
144 static int nfs4rename(vnode_t
*, char *, vnode_t
*, char *, cred_t
*,
146 static int nfs4rename_persistent_fh(vnode_t
*, char *, vnode_t
*,
147 vnode_t
*, char *, cred_t
*, nfsstat4
*);
148 static int nfs4rename_volatile_fh(vnode_t
*, char *, vnode_t
*,
149 vnode_t
*, char *, cred_t
*, nfsstat4
*);
150 static int do_nfs4readdir(vnode_t
*, rddir4_cache
*, cred_t
*);
151 static void nfs4readdir(vnode_t
*, rddir4_cache
*, cred_t
*);
152 static int nfs4_bio(struct buf
*, stable_how4
*, cred_t
*, bool_t
);
153 static int nfs4_getapage(vnode_t
*, uoff_t
, size_t, uint_t
*,
154 page_t
*[], size_t, struct seg
*, caddr_t
,
155 enum seg_rw
, cred_t
*);
156 static void nfs4_readahead(vnode_t
*, uoff_t
, caddr_t
, struct seg
*,
158 static int nfs4_sync_putapage(vnode_t
*, page_t
*, uoff_t
, size_t,
160 static int nfs4_sync_pageio(vnode_t
*, page_t
*, uoff_t
, size_t,
162 static int nfs4_commit(vnode_t
*, offset4
, count4
, cred_t
*);
163 static void nfs4_set_mod(vnode_t
*);
164 static void nfs4_get_commit(vnode_t
*);
165 static void nfs4_get_commit_range(vnode_t
*, uoff_t
, size_t);
166 static int nfs4_putpage_commit(vnode_t
*, offset_t
, size_t, cred_t
*);
167 static int nfs4_commit_vp(vnode_t
*, uoff_t
, size_t, cred_t
*, int);
168 static int nfs4_sync_commit(vnode_t
*, page_t
*, offset3
, count3
,
170 static void do_nfs4_async_commit(vnode_t
*, page_t
*, offset3
, count3
,
172 static int nfs4_update_attrcache(nfsstat4
, nfs4_ga_res_t
*,
173 hrtime_t
, vnode_t
*, cred_t
*);
174 static int nfs4_open_non_reg_file(vnode_t
**, int, cred_t
*);
175 static int nfs4_safelock(vnode_t
*, const struct flock64
*, cred_t
*);
176 static void nfs4_register_lock_locally(vnode_t
*, struct flock64
*, int,
178 static int nfs4_lockrelease(vnode_t
*, int, offset_t
, cred_t
*);
179 static int nfs4_block_and_wait(clock_t *, rnode4_t
*);
180 static cred_t
*state_to_cred(nfs4_open_stream_t
*);
181 static void denied_to_flk(LOCK4denied
*, flock64_t
*, LOCKT4args
*);
182 static pid_t
lo_to_pid(lock_owner4
*);
183 static void nfs4_reinstitute_local_lock_state(vnode_t
*, flock64_t
*,
184 cred_t
*, nfs4_lock_owner_t
*);
185 static void push_reinstate(vnode_t
*, int, flock64_t
*, cred_t
*,
186 nfs4_lock_owner_t
*);
187 static int open_and_get_osp(vnode_t
*, cred_t
*, nfs4_open_stream_t
**);
188 static void nfs4_delmap_callback(struct as
*, void *, uint_t
);
189 static void nfs4_free_delmapcall(nfs4_delmapcall_t
*);
190 static nfs4_delmapcall_t
*nfs4_init_delmapcall();
191 static int nfs4_find_and_delete_delmapcall(rnode4_t
*, int *);
192 static int nfs4_is_acl_mask_valid(uint_t
, nfs4_acl_op_t
);
193 static int nfs4_create_getsecattr_return(vsecattr_t
*, vsecattr_t
*,
197 * Routines that implement the setting of v4 args for the misc. ops
199 static void nfs4args_lock_free(nfs_argop4
*);
200 static void nfs4args_lockt_free(nfs_argop4
*);
201 static void nfs4args_setattr(nfs_argop4
*, vattr_t
*, vsecattr_t
*,
202 int, rnode4_t
*, cred_t
*, bitmap4
, int *,
203 nfs4_stateid_types_t
*);
204 static void nfs4args_setattr_free(nfs_argop4
*);
205 static int nfs4args_verify(nfs_argop4
*, vattr_t
*, enum nfs_opnum4
,
207 static void nfs4args_verify_free(nfs_argop4
*);
208 static void nfs4args_write(nfs_argop4
*, stable_how4
, rnode4_t
*, cred_t
*,
209 WRITE4args
**, nfs4_stateid_types_t
*);
212 * These are the vnode ops functions that implement the vnode interface to
213 * the networked file system. See more comments below at nfs4_vnodeops.
215 static int nfs4_open(vnode_t
**, int, cred_t
*, caller_context_t
*);
216 static int nfs4_close(vnode_t
*, int, int, offset_t
, cred_t
*,
218 static int nfs4_read(vnode_t
*, struct uio
*, int, cred_t
*,
220 static int nfs4_write(vnode_t
*, struct uio
*, int, cred_t
*,
222 static int nfs4_ioctl(vnode_t
*, int, intptr_t, int, cred_t
*, int *,
224 static int nfs4_setattr(vnode_t
*, struct vattr
*, int, cred_t
*,
226 static int nfs4_access(vnode_t
*, int, int, cred_t
*, caller_context_t
*);
227 static int nfs4_readlink(vnode_t
*, struct uio
*, cred_t
*,
229 static int nfs4_fsync(vnode_t
*, int, cred_t
*, caller_context_t
*);
230 static int nfs4_create(vnode_t
*, char *, struct vattr
*, enum vcexcl
,
231 int, vnode_t
**, cred_t
*, int, caller_context_t
*,
233 static int nfs4_remove(vnode_t
*, char *, cred_t
*, caller_context_t
*,
235 static int nfs4_link(vnode_t
*, vnode_t
*, char *, cred_t
*,
236 caller_context_t
*, int);
237 static int nfs4_rename(vnode_t
*, char *, vnode_t
*, char *, cred_t
*,
238 caller_context_t
*, int);
239 static int nfs4_mkdir(vnode_t
*, char *, struct vattr
*, vnode_t
**,
240 cred_t
*, caller_context_t
*, int, vsecattr_t
*);
241 static int nfs4_rmdir(vnode_t
*, char *, vnode_t
*, cred_t
*,
242 caller_context_t
*, int);
243 static int nfs4_symlink(vnode_t
*, char *, struct vattr
*, char *,
244 cred_t
*, caller_context_t
*, int);
245 static int nfs4_readdir(vnode_t
*, struct uio
*, cred_t
*, int *,
246 caller_context_t
*, int);
247 static int nfs4_seek(vnode_t
*, offset_t
, offset_t
*, caller_context_t
*);
248 static int nfs4_getpage(vnode_t
*, offset_t
, size_t, uint_t
*,
249 page_t
*[], size_t, struct seg
*, caddr_t
,
250 enum seg_rw
, cred_t
*, caller_context_t
*);
251 static int nfs4_putpage(vnode_t
*, offset_t
, size_t, int, cred_t
*,
253 static int nfs4_map(vnode_t
*, offset_t
, struct as
*, caddr_t
*, size_t,
254 uchar_t
, uchar_t
, uint_t
, cred_t
*, caller_context_t
*);
255 static int nfs4_addmap(vnode_t
*, offset_t
, struct as
*, caddr_t
, size_t,
256 uchar_t
, uchar_t
, uint_t
, cred_t
*, caller_context_t
*);
257 static int nfs4_cmp(vnode_t
*, vnode_t
*, caller_context_t
*);
258 static int nfs4_frlock(vnode_t
*, int, struct flock64
*, int, offset_t
,
259 struct flk_callback
*, cred_t
*, caller_context_t
*);
260 static int nfs4_space(vnode_t
*, int, struct flock64
*, int, offset_t
,
261 cred_t
*, caller_context_t
*);
262 static int nfs4_delmap(vnode_t
*, offset_t
, struct as
*, caddr_t
, size_t,
263 uint_t
, uint_t
, uint_t
, cred_t
*, caller_context_t
*);
264 static int nfs4_pageio(vnode_t
*, page_t
*, uoff_t
, size_t, int,
265 cred_t
*, caller_context_t
*);
266 static void nfs4_dispose(vnode_t
*, page_t
*, int, int, cred_t
*,
268 static int nfs4_setsecattr(vnode_t
*, vsecattr_t
*, int, cred_t
*,
271 * These vnode ops are required to be called from outside this source file,
272 * e.g. by ephemeral mount stub vnode ops, and so may not be declared
275 int nfs4_getattr(vnode_t
*, struct vattr
*, int, cred_t
*,
277 void nfs4_inactive(vnode_t
*, cred_t
*, caller_context_t
*);
278 int nfs4_lookup(vnode_t
*, char *, vnode_t
**,
279 struct pathname
*, int, vnode_t
*, cred_t
*,
280 caller_context_t
*, int *, pathname_t
*);
281 int nfs4_fid(vnode_t
*, fid_t
*, caller_context_t
*);
282 int nfs4_rwlock(vnode_t
*, int, caller_context_t
*);
283 void nfs4_rwunlock(vnode_t
*, int, caller_context_t
*);
284 int nfs4_realvp(vnode_t
*, vnode_t
**, caller_context_t
*);
285 int nfs4_pathconf(vnode_t
*, int, ulong_t
*, cred_t
*,
287 int nfs4_getsecattr(vnode_t
*, vsecattr_t
*, int, cred_t
*,
289 int nfs4_shrlock(vnode_t
*, int, struct shrlock
*, int, cred_t
*,
293 * Used for nfs4_commit_vp() to indicate if we should
294 * wait on pending writes.
296 #define NFS4_WRITE_NOWAIT 0
297 #define NFS4_WRITE_WAIT 1
299 #define NFS4_BASE_WAIT_TIME 1 /* 1 second */
302 * Error flags used to pass information about certain special errors
303 * which need to be handled specially.
306 #define NFS_VERF_MISMATCH -97
309 * Flags used to differentiate between which operation drove the
310 * potential CLOSE OTW. (see nfs4_close_otw_if_necessary)
312 #define NFS4_CLOSE_OP 0x1
313 #define NFS4_DELMAP_OP 0x2
314 #define NFS4_INACTIVE_OP 0x3
316 #define ISVDEV(t) ((t == VBLK) || (t == VCHR) || (t == VFIFO))
318 /* ALIGN64 aligns the given buffer and adjust buffer size to 64 bit */
319 #define ALIGN64(x, ptr, sz) \
320 x = ((uintptr_t)(ptr)) & (sizeof (uint64_t) - 1); \
322 x = sizeof (uint64_t) - (x); \
328 int nfs4_client_attr_debug
= 0;
329 int nfs4_client_state_debug
= 0;
330 int nfs4_client_shadow_debug
= 0;
331 int nfs4_client_lock_debug
= 0;
332 int nfs4_seqid_sync
= 0;
333 int nfs4_client_map_debug
= 0;
334 static int nfs4_pageio_debug
= 0;
335 int nfs4_client_inactive_debug
= 0;
336 int nfs4_client_recov_debug
= 0;
337 int nfs4_client_failover_debug
= 0;
338 int nfs4_client_call_debug
= 0;
339 int nfs4_client_lookup_debug
= 0;
340 int nfs4_client_zone_debug
= 0;
341 int nfs4_lost_rqst_debug
= 0;
342 int nfs4_rdattrerr_debug
= 0;
343 int nfs4_open_stream_debug
= 0;
345 int nfs4read_error_inject
;
347 static int nfs4_create_misses
= 0;
349 static int nfs4_readdir_cache_shorts
= 0;
350 static int nfs4_readdir_readahead
= 0;
352 static int nfs4_bio_do_stop
= 0;
354 static int nfs4_lostpage
= 0; /* number of times we lost original page */
356 int nfs4_mmap_debug
= 0;
358 static int nfs4_pathconf_cache_hits
= 0;
359 static int nfs4_pathconf_cache_misses
= 0;
361 int nfs4close_all_cnt
;
362 int nfs4close_one_debug
= 0;
363 int nfs4close_notw_debug
= 0;
365 int denied_to_flk_debug
= 0;
366 void *lockt_denied_debug
;
371 * How long to wait before trying again if OPEN_CONFIRM gets ETIMEDOUT
372 * or NFS4ERR_RESOURCE.
374 static int confirm_retry_sec
= 30;
376 static int nfs4_lookup_neg_cache
= 1;
379 * number of pages to read ahead
380 * optimized for 100 base-T.
382 static int nfs4_nra
= 4;
384 static int nfs4_do_symlink_cache
= 1;
386 static int nfs4_pathconf_disable_cache
= 0;
389 * These are the vnode ops routines which implement the vnode interface to
390 * the networked file system. These routines just take their parameters,
391 * make them look networkish by putting the right info into interface structs,
392 * and then calling the appropriate remote routine(s) to do the work.
394 * Note on directory name lookup cacheing: If we detect a stale fhandle,
395 * we purge the directory cache relative to that vnode. This way, the
396 * user won't get burned by the cache repeatedly. See <nfs/rnode4.h> for
397 * more details on rnode locking.
400 struct vnodeops
*nfs4_vnodeops
;
402 const fs_operation_def_t nfs4_vnodeops_template
[] = {
403 VOPNAME_OPEN
, { .vop_open
= nfs4_open
},
404 VOPNAME_CLOSE
, { .vop_close
= nfs4_close
},
405 VOPNAME_READ
, { .vop_read
= nfs4_read
},
406 VOPNAME_WRITE
, { .vop_write
= nfs4_write
},
407 VOPNAME_IOCTL
, { .vop_ioctl
= nfs4_ioctl
},
408 VOPNAME_GETATTR
, { .vop_getattr
= nfs4_getattr
},
409 VOPNAME_SETATTR
, { .vop_setattr
= nfs4_setattr
},
410 VOPNAME_ACCESS
, { .vop_access
= nfs4_access
},
411 VOPNAME_LOOKUP
, { .vop_lookup
= nfs4_lookup
},
412 VOPNAME_CREATE
, { .vop_create
= nfs4_create
},
413 VOPNAME_REMOVE
, { .vop_remove
= nfs4_remove
},
414 VOPNAME_LINK
, { .vop_link
= nfs4_link
},
415 VOPNAME_RENAME
, { .vop_rename
= nfs4_rename
},
416 VOPNAME_MKDIR
, { .vop_mkdir
= nfs4_mkdir
},
417 VOPNAME_RMDIR
, { .vop_rmdir
= nfs4_rmdir
},
418 VOPNAME_READDIR
, { .vop_readdir
= nfs4_readdir
},
419 VOPNAME_SYMLINK
, { .vop_symlink
= nfs4_symlink
},
420 VOPNAME_READLINK
, { .vop_readlink
= nfs4_readlink
},
421 VOPNAME_FSYNC
, { .vop_fsync
= nfs4_fsync
},
422 VOPNAME_INACTIVE
, { .vop_inactive
= nfs4_inactive
},
423 VOPNAME_FID
, { .vop_fid
= nfs4_fid
},
424 VOPNAME_RWLOCK
, { .vop_rwlock
= nfs4_rwlock
},
425 VOPNAME_RWUNLOCK
, { .vop_rwunlock
= nfs4_rwunlock
},
426 VOPNAME_SEEK
, { .vop_seek
= nfs4_seek
},
427 VOPNAME_FRLOCK
, { .vop_frlock
= nfs4_frlock
},
428 VOPNAME_SPACE
, { .vop_space
= nfs4_space
},
429 VOPNAME_REALVP
, { .vop_realvp
= nfs4_realvp
},
430 VOPNAME_GETPAGE
, { .vop_getpage
= nfs4_getpage
},
431 VOPNAME_PUTPAGE
, { .vop_putpage
= nfs4_putpage
},
432 VOPNAME_MAP
, { .vop_map
= nfs4_map
},
433 VOPNAME_ADDMAP
, { .vop_addmap
= nfs4_addmap
},
434 VOPNAME_DELMAP
, { .vop_delmap
= nfs4_delmap
},
435 /* no separate nfs4_dump */
436 VOPNAME_DUMP
, { .vop_dump
= nfs_dump
},
437 VOPNAME_PATHCONF
, { .vop_pathconf
= nfs4_pathconf
},
438 VOPNAME_PAGEIO
, { .vop_pageio
= nfs4_pageio
},
439 VOPNAME_DISPOSE
, { .vop_dispose
= nfs4_dispose
},
440 VOPNAME_SETSECATTR
, { .vop_setsecattr
= nfs4_setsecattr
},
441 VOPNAME_GETSECATTR
, { .vop_getsecattr
= nfs4_getsecattr
},
442 VOPNAME_SHRLOCK
, { .vop_shrlock
= nfs4_shrlock
},
443 VOPNAME_VNEVENT
, { .vop_vnevent
= fs_vnevent_support
},
448 * The following are subroutines and definitions to set args or get res
449 * for the different nfsv4 ops
453 nfs4args_lookup_free(nfs_argop4
*argop
, int arglen
)
457 for (i
= 0; i
< arglen
; i
++) {
458 if (argop
[i
].argop
== OP_LOOKUP
) {
460 argop
[i
].nfs_argop4_u
.oplookup
.
461 objname
.utf8string_val
,
462 argop
[i
].nfs_argop4_u
.oplookup
.
463 objname
.utf8string_len
);
469 nfs4args_lock_free(nfs_argop4
*argop
)
471 locker4
*locker
= &argop
->nfs_argop4_u
.oplock
.locker
;
473 if (locker
->new_lock_owner
== TRUE
) {
474 open_to_lock_owner4
*open_owner
;
476 open_owner
= &locker
->locker4_u
.open_owner
;
477 if (open_owner
->lock_owner
.owner_val
!= NULL
) {
478 kmem_free(open_owner
->lock_owner
.owner_val
,
479 open_owner
->lock_owner
.owner_len
);
485 nfs4args_lockt_free(nfs_argop4
*argop
)
487 lock_owner4
*lowner
= &argop
->nfs_argop4_u
.oplockt
.owner
;
489 if (lowner
->owner_val
!= NULL
) {
490 kmem_free(lowner
->owner_val
, lowner
->owner_len
);
495 nfs4args_setattr(nfs_argop4
*argop
, vattr_t
*vap
, vsecattr_t
*vsap
, int flags
,
496 rnode4_t
*rp
, cred_t
*cr
, bitmap4 supp
, int *error
,
497 nfs4_stateid_types_t
*sid_types
)
499 fattr4
*attr
= &argop
->nfs_argop4_u
.opsetattr
.obj_attributes
;
502 argop
->argop
= OP_SETATTR
;
504 * The stateid is set to 0 if client is not modifying the size
505 * and otherwise to whatever nfs4_get_stateid() returns.
507 * XXX Note: nfs4_get_stateid() returns 0 if no lockowner and/or no
508 * state struct could be found for the process/file pair. We may
509 * want to change this in the future (by OPENing the file). See
512 if (vap
->va_mask
& AT_SIZE
) {
515 mi
= VTOMI4(RTOV4(rp
));
517 argop
->nfs_argop4_u
.opsetattr
.stateid
=
518 nfs4_get_stateid(cr
, rp
, curproc
->p_pidp
->pid_id
, mi
,
519 OP_SETATTR
, sid_types
, FALSE
);
521 bzero(&argop
->nfs_argop4_u
.opsetattr
.stateid
,
525 *error
= vattr_to_fattr4(vap
, vsap
, attr
, flags
, OP_SETATTR
, supp
);
527 bzero(attr
, sizeof (*attr
));
531 nfs4args_setattr_free(nfs_argop4
*argop
)
533 nfs4_fattr4_free(&argop
->nfs_argop4_u
.opsetattr
.obj_attributes
);
537 nfs4args_verify(nfs_argop4
*argop
, vattr_t
*vap
, enum nfs_opnum4 op
,
546 attr
= &argop
->nfs_argop4_u
.opverify
.obj_attributes
;
549 attr
= &argop
->nfs_argop4_u
.opnverify
.obj_attributes
;
555 error
= vattr_to_fattr4(vap
, NULL
, attr
, 0, op
, supp
);
557 bzero(attr
, sizeof (*attr
));
562 nfs4args_verify_free(nfs_argop4
*argop
)
564 switch (argop
->argop
) {
566 nfs4_fattr4_free(&argop
->nfs_argop4_u
.opverify
.obj_attributes
);
569 nfs4_fattr4_free(&argop
->nfs_argop4_u
.opnverify
.obj_attributes
);
577 nfs4args_write(nfs_argop4
*argop
, stable_how4 stable
, rnode4_t
*rp
, cred_t
*cr
,
578 WRITE4args
**wargs_pp
, nfs4_stateid_types_t
*sid_tp
)
580 WRITE4args
*wargs
= &argop
->nfs_argop4_u
.opwrite
;
581 mntinfo4_t
*mi
= VTOMI4(RTOV4(rp
));
583 argop
->argop
= OP_WRITE
;
584 wargs
->stable
= stable
;
585 wargs
->stateid
= nfs4_get_w_stateid(cr
, rp
, curproc
->p_pidp
->pid_id
,
586 mi
, OP_WRITE
, sid_tp
);
592 nfs4args_copen_free(OPEN4cargs
*open_args
)
594 if (open_args
->owner
.owner_val
) {
595 kmem_free(open_args
->owner
.owner_val
,
596 open_args
->owner
.owner_len
);
598 if ((open_args
->opentype
== OPEN4_CREATE
) &&
599 (open_args
->mode
!= EXCLUSIVE4
)) {
600 nfs4_fattr4_free(&open_args
->createhow4_u
.createattrs
);
605 * XXX: This is referenced in modstubs.s
608 nfs4_getvnodeops(void)
610 return (nfs4_vnodeops
);
614 * The OPEN operation opens a regular file.
618 nfs4_open(vnode_t
**vpp
, int flag
, cred_t
*cr
, caller_context_t
*ct
)
623 int just_been_created
;
626 NFS4_DEBUG(nfs4_client_state_debug
, (CE_NOTE
, "nfs4_open: "));
627 if (nfs_zone() != VTOMI4(*vpp
)->mi_zone
)
632 * Check to see if opening something besides a regular file;
633 * if so skip the OTW call
635 if ((*vpp
)->v_type
!= VREG
) {
636 error
= nfs4_open_non_reg_file(vpp
, flag
, cr
);
641 * XXX - would like a check right here to know if the file is
642 * executable or not, so as to skip OTW
645 if ((error
= vtodv(*vpp
, &dvp
, cr
, TRUE
)) != 0)
649 if (nfs_rw_enter_sig(&drp
->r_rwlock
, RW_READER
, INTR4(dvp
)))
652 if ((error
= vtoname(*vpp
, fn
, MAXNAMELEN
)) != 0) {
653 nfs_rw_exit(&drp
->r_rwlock
);
658 * See if this file has just been CREATEd.
659 * If so, clear the flag and update the dnlc, which was previously
660 * skipped in nfs4_create.
661 * XXX need better serilization on this.
662 * XXX move this into the nf4open_otw call, after we have
663 * XXX acquired the open owner seqid sync.
665 mutex_enter(&rp
->r_statev4_lock
);
666 if (rp
->created_v4
) {
668 mutex_exit(&rp
->r_statev4_lock
);
670 dnlc_update(dvp
, fn
, *vpp
);
671 /* This is needed so we don't bump the open ref count */
672 just_been_created
= 1;
674 mutex_exit(&rp
->r_statev4_lock
);
675 just_been_created
= 0;
679 * If caller specified O_TRUNC/FTRUNC, then be sure to set
680 * FWRITE (to drive successful setattr(size=0) after open)
685 error
= nfs4open_otw(dvp
, fn
, NULL
, vpp
, cr
, 0, flag
, 0,
688 if (!error
&& !((*vpp
)->v_flag
& VROOT
))
689 dnlc_update(dvp
, fn
, *vpp
);
691 nfs_rw_exit(&drp
->r_rwlock
);
693 /* release the hold from vtodv */
696 /* exchange the shadow for the master vnode, if needed */
698 if (error
== 0 && IS_SHADOW(*vpp
, rp
))
705 * See if there's a "lost open" request to be saved and recovered.
708 nfs4open_save_lost_rqst(int error
, nfs4_lost_rqst_t
*lost_rqstp
,
709 nfs4_open_owner_t
*oop
, cred_t
*cr
, vnode_t
*vp
,
710 vnode_t
*dvp
, OPEN4cargs
*open_args
)
715 vfsp
= (dvp
? dvp
->v_vfsp
: vp
->v_vfsp
);
717 if (error
!= ETIMEDOUT
&& error
!= EINTR
&&
718 !NFS4_FRC_UNMT_ERR(error
, vfsp
)) {
719 lost_rqstp
->lr_op
= 0;
723 NFS4_DEBUG(nfs4_lost_rqst_debug
, (CE_NOTE
,
724 "nfs4open_save_lost_rqst: error %d", error
));
726 lost_rqstp
->lr_op
= OP_OPEN
;
729 * The vp (if it is not NULL) and dvp are held and rele'd via
730 * the recovery code. See nfs4_save_lost_rqst.
732 lost_rqstp
->lr_vp
= vp
;
733 lost_rqstp
->lr_dvp
= dvp
;
734 lost_rqstp
->lr_oop
= oop
;
735 lost_rqstp
->lr_osp
= NULL
;
736 lost_rqstp
->lr_lop
= NULL
;
737 lost_rqstp
->lr_cr
= cr
;
738 lost_rqstp
->lr_flk
= NULL
;
739 lost_rqstp
->lr_oacc
= open_args
->share_access
;
740 lost_rqstp
->lr_odeny
= open_args
->share_deny
;
741 lost_rqstp
->lr_oclaim
= open_args
->claim
;
742 if (open_args
->claim
== CLAIM_DELEGATE_CUR
) {
743 lost_rqstp
->lr_ostateid
=
744 open_args
->open_claim4_u
.delegate_cur_info
.delegate_stateid
;
745 srccfp
= open_args
->open_claim4_u
.delegate_cur_info
.cfile
;
747 srccfp
= open_args
->open_claim4_u
.cfile
;
749 lost_rqstp
->lr_ofile
.utf8string_len
= 0;
750 lost_rqstp
->lr_ofile
.utf8string_val
= NULL
;
751 (void) str_to_utf8(srccfp
, &lost_rqstp
->lr_ofile
);
752 lost_rqstp
->lr_putfirst
= FALSE
;
755 struct nfs4_excl_time
{
761 * The OPEN operation creates and/or opens a regular file
766 nfs4open_otw(vnode_t
*dvp
, char *file_name
, struct vattr
*in_va
,
767 vnode_t
**vpp
, cred_t
*cr
, int create_flag
, int open_flag
,
768 enum createmode4 createmode
, int file_just_been_created
)
771 rnode4_t
*drp
= VTOR4(dvp
);
774 bool_t needrecov
= FALSE
;
778 COMPOUND4args_clnt args
;
779 COMPOUND4res_clnt res
;
783 int idx_open
, idx_fattr
;
785 GETFH4res
*gf_res
= NULL
;
786 OPEN4res
*op_res
= NULL
;
789 struct nfs4_excl_time verf
;
790 bool_t did_excl_setup
= FALSE
;
793 OPEN4cargs
*open_args
;
794 nfs4_open_owner_t
*oop
= NULL
;
795 nfs4_open_stream_t
*osp
= NULL
;
797 bool_t retry_open
= FALSE
;
798 nfs4_recov_state_t recov_state
;
799 nfs4_lost_rqst_t lost_rqst
;
800 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
803 cred_t
*cred_otw
= NULL
; /* cred used to do the RPC call */
806 nfs4_sharedfh_t
*otw_sfh
;
807 nfs4_sharedfh_t
*orig_sfh
;
809 int numops
, setgid_flag
;
810 int num_bseqid_retry
= NFS4_NUM_RETRY_BAD_SEQID
+ 1;
813 * Make sure we properly deal with setting the right gid on
814 * a newly created file to reflect the parent's setgid bit
817 if (create_flag
&& in_va
) {
820 * If there is grpid mount flag used or
821 * the parent's directory has the setgid bit set
822 * _and_ the client was able to get a valid mapping
823 * for the parent dir's owner_group, we want to
824 * append NVERIFY(owner_group == dva.va_gid) and
825 * SETATTR to the CREATE compound.
827 mutex_enter(&drp
->r_statelock
);
828 if ((VTOMI4(dvp
)->mi_flags
& MI4_GRPID
||
829 drp
->r_attr
.va_mode
& VSGID
) &&
830 drp
->r_attr
.va_gid
!= GID_NOBODY
) {
831 in_va
->va_mask
|= AT_GID
;
832 in_va
->va_gid
= drp
->r_attr
.va_gid
;
835 mutex_exit(&drp
->r_statelock
);
839 * Normal/non-create compound:
840 * PUTFH(dfh) + OPEN(create) + GETFH + GETATTR(new)
842 * Open(create) compound no setgid:
843 * PUTFH(dfh) + SAVEFH + OPEN(create) + GETFH + GETATTR(new) +
844 * RESTOREFH + GETATTR
846 * Open(create) setgid:
847 * PUTFH(dfh) + OPEN(create) + GETFH + GETATTR(new) +
848 * SAVEFH + PUTFH(dfh) + GETATTR(dvp) + RESTOREFH +
849 * NVERIFY(grp) + SETATTR
855 } else if (create_flag
) {
865 args
.array_len
= numops
;
866 argoplist_size
= numops
* sizeof (nfs_argop4
);
867 argop
= kmem_alloc(argoplist_size
, KM_SLEEP
);
869 NFS4_DEBUG(nfs4_client_state_debug
, (CE_NOTE
, "nfs4open_otw: "
870 "open %s open flag 0x%x cred %p", file_name
, open_flag
,
873 ASSERT(nfs_zone() == VTOMI4(dvp
)->mi_zone
);
876 * We are to create a file. Initialize the passed in vnode
882 * Check to see if the client owns a read delegation and is
883 * trying to open for write. If so, then return the delegation
884 * to avoid the server doing a cb_recall and returning DELAY.
885 * NB - we don't use the statev4_lock here because we'd have
886 * to drop the lock anyway and the result would be stale.
888 if ((open_flag
& FWRITE
) &&
889 VTOR4(vpi
)->r_deleg_type
== OPEN_DELEGATE_READ
)
890 (void) nfs4delegreturn(VTOR4(vpi
), NFS4_DR_REOPEN
);
893 * If the file has a delegation, then do an access check up
894 * front. This avoids having to an access check later after
895 * we've already done start_op, which could deadlock.
897 if (VTOR4(vpi
)->r_deleg_type
!= OPEN_DELEGATE_NONE
) {
898 if (open_flag
& FREAD
&&
899 nfs4_access(vpi
, VREAD
, 0, cr
, NULL
) == 0)
901 if (open_flag
& FWRITE
&&
902 nfs4_access(vpi
, VWRITE
, 0, cr
, NULL
) == 0)
909 recov_state
.rs_flags
= 0;
910 recov_state
.rs_num_retry_despite_err
= 0;
915 nfs4_error_zinit(&e
);
917 e
.error
= nfs4_start_op(VTOMI4(dvp
), dvp
, vpi
, &recov_state
);
921 kmem_free(argop
, argoplist_size
);
925 args
.ctag
= TAG_OPEN
;
926 args
.array_len
= numops
;
929 /* putfh directory fh */
930 argop
[0].argop
= OP_CPUTFH
;
931 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= drp
->r_fh
;
933 /* OPEN: either op 1 or op 2 depending upon create/setgid flags */
934 argop
[idx_open
].argop
= OP_COPEN
;
935 open_args
= &argop
[idx_open
].nfs_argop4_u
.opcopen
;
936 open_args
->claim
= CLAIM_NULL
;
939 open_args
->open_claim4_u
.cfile
= file_name
;
940 open_args
->owner
.owner_len
= 0;
941 open_args
->owner
.owner_val
= NULL
;
945 open_args
->opentype
= OPEN4_CREATE
;
946 open_args
->mode
= createmode
;
947 if (createmode
== EXCLUSIVE4
) {
948 if (did_excl_setup
== FALSE
) {
949 verf
.seconds
= zone_get_hostid(NULL
);
950 if (verf
.seconds
!= 0)
951 verf
.nseconds
= newnum();
956 verf
.seconds
= now
.tv_sec
;
957 verf
.nseconds
= now
.tv_nsec
;
960 * Since the server will use this value for the
961 * mtime, make sure that it can't overflow. Zero
962 * out the MSB. The actual value does not matter
963 * here, only its uniqeness.
965 verf
.seconds
&= INT32_MAX
;
966 did_excl_setup
= TRUE
;
969 /* Now copy over verifier to OPEN4args. */
970 open_args
->createhow4_u
.createverf
= *(uint64_t *)&verf
;
976 attr
= &open_args
->createhow4_u
.createattrs
;
979 (void) nfs_rw_enter_sig(&svp
->sv_lock
, RW_READER
, 0);
980 supp_attrs
= svp
->sv_supp_attrs
;
981 nfs_rw_exit(&svp
->sv_lock
);
983 /* GUARDED4 or UNCHECKED4 */
984 v_error
= vattr_to_fattr4(in_va
, NULL
, attr
, 0, OP_OPEN
,
987 bzero(attr
, sizeof (*attr
));
988 nfs4args_copen_free(open_args
);
989 nfs4_end_op(VTOMI4(dvp
), dvp
, vpi
,
990 &recov_state
, FALSE
);
993 kmem_free(argop
, argoplist_size
);
999 open_args
->opentype
= OPEN4_NOCREATE
;
1002 if (recov_state
.rs_sp
!= NULL
) {
1003 mutex_enter(&recov_state
.rs_sp
->s_lock
);
1004 open_args
->owner
.clientid
= recov_state
.rs_sp
->clientid
;
1005 mutex_exit(&recov_state
.rs_sp
->s_lock
);
1007 /* XXX should we just fail here? */
1008 open_args
->owner
.clientid
= 0;
1012 * This increments oop's ref count or creates a temporary 'just_created'
1013 * open owner that will become valid when this OPEN/OPEN_CONFIRM call
1016 mutex_enter(&VTOMI4(dvp
)->mi_lock
);
1018 /* See if a permanent or just created open owner exists */
1019 oop
= find_open_owner_nolock(cr
, NFS4_JUST_CREATED
, VTOMI4(dvp
));
1022 * This open owner does not exist so create a temporary
1025 oop
= create_open_owner(cr
, VTOMI4(dvp
));
1026 ASSERT(oop
!= NULL
);
1028 mutex_exit(&VTOMI4(dvp
)->mi_lock
);
1030 /* this length never changes, do alloc before seqid sync */
1031 open_args
->owner
.owner_len
= sizeof (oop
->oo_name
);
1032 open_args
->owner
.owner_val
=
1033 kmem_alloc(open_args
->owner
.owner_len
, KM_SLEEP
);
1035 e
.error
= nfs4_start_open_seqid_sync(oop
, VTOMI4(dvp
));
1036 if (e
.error
== EAGAIN
) {
1037 open_owner_rele(oop
);
1038 nfs4args_copen_free(open_args
);
1039 nfs4_end_op(VTOMI4(dvp
), dvp
, vpi
, &recov_state
, TRUE
);
1047 /* Check to see if we need to do the OTW call */
1049 if (!nfs4_is_otw_open_necessary(oop
, open_flag
, vpi
,
1050 file_just_been_created
, &e
.error
, acc
, &recov_state
)) {
1053 * The OTW open is not necessary. Either
1054 * the open can succeed without it (eg.
1055 * delegation, error == 0) or the open
1056 * must fail due to an access failure
1057 * (error != 0). In either case, tidy
1061 nfs4_end_open_seqid_sync(oop
);
1062 open_owner_rele(oop
);
1063 nfs4args_copen_free(open_args
);
1064 nfs4_end_op(VTOMI4(dvp
), dvp
, vpi
, &recov_state
, FALSE
);
1067 kmem_free(argop
, argoplist_size
);
1072 bcopy(&oop
->oo_name
, open_args
->owner
.owner_val
,
1073 open_args
->owner
.owner_len
);
1075 seqid
= nfs4_get_open_seqid(oop
) + 1;
1076 open_args
->seqid
= seqid
;
1077 open_args
->share_access
= 0;
1078 if (open_flag
& FREAD
)
1079 open_args
->share_access
|= OPEN4_SHARE_ACCESS_READ
;
1080 if (open_flag
& FWRITE
)
1081 open_args
->share_access
|= OPEN4_SHARE_ACCESS_WRITE
;
1082 open_args
->share_deny
= OPEN4_SHARE_DENY_NONE
;
1087 * getfh w/sanity check for idx_open/idx_fattr
1089 ASSERT((idx_open
+ 1) == (idx_fattr
- 1));
1090 argop
[idx_open
+ 1].argop
= OP_GETFH
;
1093 argop
[idx_fattr
].argop
= OP_GETATTR
;
1094 argop
[idx_fattr
].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
1095 argop
[idx_fattr
].nfs_argop4_u
.opgetattr
.mi
= VTOMI4(dvp
);
1102 svp
= drp
->r_server
;
1103 (void) nfs_rw_enter_sig(&svp
->sv_lock
, RW_READER
, 0);
1104 supp_attrs
= svp
->sv_supp_attrs
;
1105 nfs_rw_exit(&svp
->sv_lock
);
1108 * For setgid case, we need to:
1109 * 4:savefh(new) 5:putfh(dir) 6:getattr(dir) 7:restorefh(new)
1111 argop
[4].argop
= OP_SAVEFH
;
1113 argop
[5].argop
= OP_CPUTFH
;
1114 argop
[5].nfs_argop4_u
.opcputfh
.sfh
= drp
->r_fh
;
1116 argop
[6].argop
= OP_GETATTR
;
1117 argop
[6].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
1118 argop
[6].nfs_argop4_u
.opgetattr
.mi
= VTOMI4(dvp
);
1120 argop
[7].argop
= OP_RESTOREFH
;
1125 _v
.va_mask
= AT_GID
;
1126 _v
.va_gid
= in_va
->va_gid
;
1127 if (!(e
.error
= nfs4args_verify(&argop
[8], &_v
, OP_NVERIFY
,
1133 * We _know_ we're not messing with AT_SIZE or
1134 * AT_XTIME, so no need for stateid or flags.
1135 * Also we specify NULL rp since we're only
1136 * interested in setting owner_group attributes.
1138 nfs4args_setattr(&argop
[9], &_v
, NULL
, 0, NULL
, cr
,
1139 supp_attrs
, &e
.error
, 0);
1141 nfs4args_verify_free(&argop
[8]);
1146 * XXX - Revisit the last argument to nfs4_end_op()
1147 * once 5020486 is fixed.
1149 nfs4_end_open_seqid_sync(oop
);
1150 open_owner_rele(oop
);
1151 nfs4args_copen_free(open_args
);
1152 nfs4_end_op(VTOMI4(dvp
), dvp
, vpi
, &recov_state
, TRUE
);
1155 kmem_free(argop
, argoplist_size
);
1158 } else if (create_flag
) {
1159 argop
[1].argop
= OP_SAVEFH
;
1161 argop
[5].argop
= OP_RESTOREFH
;
1163 argop
[6].argop
= OP_GETATTR
;
1164 argop
[6].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
1165 argop
[6].nfs_argop4_u
.opgetattr
.mi
= VTOMI4(dvp
);
1168 NFS4_DEBUG(nfs4_client_call_debug
, (CE_NOTE
,
1169 "nfs4open_otw: %s call, nm %s, rp %s",
1170 needrecov
? "recov" : "first", file_name
,
1171 rnode4info(VTOR4(dvp
))));
1175 rfs4call(VTOMI4(dvp
), &args
, &res
, cred_otw
, &doqueue
, 0, &e
);
1177 if (!e
.error
&& nfs4_need_to_bump_seqid(&res
))
1178 nfs4_set_open_seqid(seqid
, oop
, args
.ctag
);
1180 needrecov
= nfs4_needs_recovery(&e
, TRUE
, dvp
->v_vfsp
);
1182 if (e
.error
|| needrecov
) {
1183 bool_t abort
= FALSE
;
1186 nfs4_bseqid_entry_t
*bsep
= NULL
;
1188 nfs4open_save_lost_rqst(e
.error
, &lost_rqst
, oop
,
1189 cred_otw
, vpi
, dvp
, open_args
);
1191 if (!e
.error
&& res
.status
== NFS4ERR_BAD_SEQID
) {
1192 bsep
= nfs4_create_bseqid_entry(oop
, NULL
,
1193 vpi
, 0, args
.ctag
, open_args
->seqid
);
1197 abort
= nfs4_start_recovery(&e
, VTOMI4(dvp
), dvp
, vpi
,
1198 NULL
, lost_rqst
.lr_op
== OP_OPEN
?
1199 &lost_rqst
: NULL
, OP_OPEN
, bsep
, NULL
, NULL
);
1202 kmem_free(bsep
, sizeof (*bsep
));
1203 /* give up if we keep getting BAD_SEQID */
1204 if (num_bseqid_retry
== 0)
1206 if (abort
== TRUE
&& e
.error
== 0)
1207 e
.error
= geterrno4(res
.status
);
1209 nfs4_end_open_seqid_sync(oop
);
1210 open_owner_rele(oop
);
1211 nfs4_end_op(VTOMI4(dvp
), dvp
, vpi
, &recov_state
, needrecov
);
1212 nfs4args_copen_free(open_args
);
1214 nfs4args_verify_free(&argop
[8]);
1215 nfs4args_setattr_free(&argop
[9]);
1218 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
1223 if (!needrecov
|| abort
== TRUE
|| e
.error
== EINTR
||
1224 NFS4_FRC_UNMT_ERR(e
.error
, dvp
->v_vfsp
)) {
1225 kmem_free(argop
, argoplist_size
);
1232 * Will check and update lease after checking the rflag for
1233 * OPEN_CONFIRM in the successful OPEN call.
1235 if (res
.status
!= NFS4_OK
&& res
.array_len
<= idx_fattr
+ 1) {
1238 * XXX what if we're crossing mount points from server1:/drp
1239 * to server2:/drp/rp.
1242 /* Signal our end of use of the open seqid */
1243 nfs4_end_open_seqid_sync(oop
);
1246 * This will destroy the open owner if it was just created,
1247 * and no one else has put a reference on it.
1249 open_owner_rele(oop
);
1250 if (create_flag
&& (createmode
!= EXCLUSIVE4
) &&
1251 res
.status
== NFS4ERR_BADOWNER
)
1252 nfs4_log_badowner(VTOMI4(dvp
), OP_OPEN
);
1254 e
.error
= geterrno4(res
.status
);
1255 nfs4args_copen_free(open_args
);
1257 nfs4args_verify_free(&argop
[8]);
1258 nfs4args_setattr_free(&argop
[9]);
1260 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
1261 nfs4_end_op(VTOMI4(dvp
), dvp
, vpi
, &recov_state
, needrecov
);
1263 * If the reply is NFS4ERR_ACCESS, it may be because
1264 * we are root (no root net access). If the real uid
1265 * is not root, then retry with the real uid instead.
1271 if (res
.status
== NFS4ERR_ACCESS
&&
1272 (ncr
= crnetadjust(cred_otw
)) != NULL
) {
1276 kmem_free(argop
, argoplist_size
);
1280 resop
= &res
.array
[idx_open
]; /* open res */
1281 op_res
= &resop
->nfs_resop4_u
.opopen
;
1285 * verify attrset bitmap
1288 (createmode
== UNCHECKED4
|| createmode
== GUARDED4
)) {
1289 /* make sure attrset returned is what we asked for */
1290 /* XXX Ignore this 'error' for now */
1291 if (attr
->attrmask
!= op_res
->attrset
)
1296 if (op_res
->rflags
& OPEN4_RESULT_LOCKTYPE_POSIX
) {
1297 mutex_enter(&VTOMI4(dvp
)->mi_lock
);
1298 VTOMI4(dvp
)->mi_flags
|= MI4_POSIX_LOCK
;
1299 mutex_exit(&VTOMI4(dvp
)->mi_lock
);
1302 resop
= &res
.array
[idx_open
+ 1]; /* getfh res */
1303 gf_res
= &resop
->nfs_resop4_u
.opgetfh
;
1305 otw_sfh
= sfh4_get(&gf_res
->object
, VTOMI4(dvp
));
1308 * The open stateid has been updated on the server but not
1309 * on the client yet. There is a path: makenfs4node->nfs4_attr_cache->
1310 * flush_pages->fop_putpage->...->nfs4write where we will issue an OTW
1311 * WRITE call. That, however, will use the old stateid, so go ahead
1312 * and upate the open stateid now, before any call to makenfs4node.
1315 nfs4_open_stream_t
*tmp_osp
;
1316 rnode4_t
*tmp_rp
= VTOR4(vpi
);
1318 tmp_osp
= find_open_stream(oop
, tmp_rp
);
1320 tmp_osp
->open_stateid
= op_res
->stateid
;
1321 mutex_exit(&tmp_osp
->os_sync_lock
);
1322 open_stream_rele(tmp_osp
, tmp_rp
);
1326 * We must determine if the file handle given by the otw open
1327 * is the same as the file handle which was passed in with
1328 * *vpp. This case can be reached if the file we are trying
1329 * to open has been removed and another file has been created
1330 * having the same file name. The passed in vnode is released
1333 orig_sfh
= VTOR4(vpi
)->r_fh
;
1334 fh_differs
= nfs4cmpfh(&orig_sfh
->sfh_fh
, &otw_sfh
->sfh_fh
);
1337 garp
= &res
.array
[idx_fattr
].nfs_resop4_u
.opgetattr
.ga_res
;
1339 if (create_flag
|| fh_differs
) {
1342 vp
= makenfs4node(otw_sfh
, garp
, dvp
->v_vfsp
, t
, cr
,
1343 dvp
, fn_get(VTOSV(dvp
)->sv_name
, file_name
, otw_sfh
));
1346 PURGE_ATTRCACHE4(vp
);
1348 * For the newly created vp case, make sure the rnode
1349 * isn't bad before using it.
1351 mutex_enter(&(VTOR4(vp
))->r_statelock
);
1352 if (VTOR4(vp
)->r_flags
& R4RECOVERR
)
1354 mutex_exit(&(VTOR4(vp
))->r_statelock
);
1357 nfs4_end_open_seqid_sync(oop
);
1358 nfs4args_copen_free(open_args
);
1360 nfs4args_verify_free(&argop
[8]);
1361 nfs4args_setattr_free(&argop
[9]);
1363 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
1364 nfs4_end_op(VTOMI4(dvp
), dvp
, vpi
, &recov_state
,
1366 open_owner_rele(oop
);
1370 sfh4_rele(&otw_sfh
);
1371 kmem_free(argop
, argoplist_size
);
1377 sfh4_rele(&otw_sfh
);
1380 * It seems odd to get a full set of attrs and then not update
1381 * the object's attrcache in the non-create case. Create case uses
1382 * the attrs since makenfs4node checks to see if the attrs need to
1383 * be updated (and then updates them). The non-create case should
1384 * update attrs also.
1386 if (! create_flag
&& ! fh_differs
&& !e
.error
) {
1387 nfs4_attr_cache(vp
, garp
, t
, cr
, TRUE
, NULL
);
1390 nfs4_error_zinit(&e
);
1391 if (op_res
->rflags
& OPEN4_RESULT_CONFIRM
) {
1392 /* This does not do recovery for vp explicitly. */
1393 nfs4open_confirm(vp
, &seqid
, &op_res
->stateid
, cred_otw
, FALSE
,
1394 &retry_open
, oop
, FALSE
, &e
, &num_bseqid_retry
);
1396 if (e
.error
|| e
.stat
) {
1397 nfs4_end_open_seqid_sync(oop
);
1398 nfs4args_copen_free(open_args
);
1400 nfs4args_verify_free(&argop
[8]);
1401 nfs4args_setattr_free(&argop
[9]);
1403 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
1404 nfs4_end_op(VTOMI4(dvp
), dvp
, vpi
, &recov_state
,
1406 open_owner_rele(oop
);
1407 if (create_flag
|| fh_differs
) {
1408 /* rele the makenfs4node */
1415 if (retry_open
== TRUE
) {
1416 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
,
1417 "nfs4open_otw: retry the open since OPEN "
1418 "CONFIRM failed with error %d stat %d",
1420 if (create_flag
&& createmode
== GUARDED4
) {
1421 NFS4_DEBUG(nfs4_client_recov_debug
,
1422 (CE_NOTE
, "nfs4open_otw: switch "
1423 "createmode from GUARDED4 to "
1425 createmode
= UNCHECKED4
;
1430 if (create_flag
&& (createmode
!= EXCLUSIVE4
) &&
1431 e
.stat
== NFS4ERR_BADOWNER
)
1432 nfs4_log_badowner(VTOMI4(dvp
), OP_OPEN
);
1434 e
.error
= geterrno4(e
.stat
);
1436 kmem_free(argop
, argoplist_size
);
1443 mutex_enter(&rp
->r_statev4_lock
);
1446 mutex_exit(&rp
->r_statev4_lock
);
1448 mutex_enter(&oop
->oo_lock
);
1449 /* Doesn't matter if 'oo_just_created' already was set as this */
1450 oop
->oo_just_created
= NFS4_PERM_CREATED
;
1451 if (oop
->oo_cred_otw
)
1452 crfree(oop
->oo_cred_otw
);
1453 oop
->oo_cred_otw
= cred_otw
;
1454 crhold(oop
->oo_cred_otw
);
1455 mutex_exit(&oop
->oo_lock
);
1457 /* returns with 'os_sync_lock' held */
1458 osp
= find_or_create_open_stream(oop
, rp
, &created_osp
);
1460 NFS4_DEBUG(nfs4_client_state_debug
, (CE_NOTE
,
1461 "nfs4open_otw: failed to create an open stream"));
1462 NFS4_DEBUG(nfs4_seqid_sync
, (CE_NOTE
, "nfs4open_otw: "
1463 "signal our end of use of the open seqid"));
1465 nfs4_end_open_seqid_sync(oop
);
1466 open_owner_rele(oop
);
1467 nfs4args_copen_free(open_args
);
1469 nfs4args_verify_free(&argop
[8]);
1470 nfs4args_setattr_free(&argop
[9]);
1472 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
1473 nfs4_end_op(VTOMI4(dvp
), dvp
, vpi
, &recov_state
, needrecov
);
1474 if (create_flag
|| fh_differs
)
1479 kmem_free(argop
, argoplist_size
);
1484 osp
->open_stateid
= op_res
->stateid
;
1486 if (open_flag
& FREAD
)
1487 osp
->os_share_acc_read
++;
1488 if (open_flag
& FWRITE
)
1489 osp
->os_share_acc_write
++;
1490 osp
->os_share_deny_none
++;
1493 * Need to reset this bitfield for the possible case where we were
1494 * going to OTW CLOSE the file, got a non-recoverable error, and before
1495 * we could retry the CLOSE, OPENed the file again.
1497 ASSERT(osp
->os_open_owner
->oo_seqid_inuse
);
1498 osp
->os_final_close
= 0;
1499 osp
->os_force_close
= 0;
1501 if (osp
->os_failed_reopen
)
1502 NFS4_DEBUG(nfs4_open_stream_debug
, (CE_NOTE
, "nfs4open_otw:"
1503 " clearing os_failed_reopen for osp %p, cr %p, rp %s",
1504 (void *)osp
, (void *)cr
, rnode4info(rp
)));
1506 osp
->os_failed_reopen
= 0;
1508 mutex_exit(&osp
->os_sync_lock
);
1510 nfs4_end_open_seqid_sync(oop
);
1512 if (created_osp
&& recov_state
.rs_sp
!= NULL
) {
1513 mutex_enter(&recov_state
.rs_sp
->s_lock
);
1514 nfs4_inc_state_ref_count_nolock(recov_state
.rs_sp
, VTOMI4(dvp
));
1515 mutex_exit(&recov_state
.rs_sp
->s_lock
);
1518 /* get rid of our reference to find oop */
1519 open_owner_rele(oop
);
1521 open_stream_rele(osp
, rp
);
1523 /* accept delegation, if any */
1524 nfs4_delegation_accept(rp
, CLAIM_NULL
, op_res
, garp
, cred_otw
);
1526 nfs4_end_op(VTOMI4(dvp
), dvp
, vpi
, &recov_state
, needrecov
);
1528 if (createmode
== EXCLUSIVE4
&&
1529 (in_va
->va_mask
& ~(AT_GID
| AT_SIZE
))) {
1530 NFS4_DEBUG(nfs4_client_state_debug
, (CE_NOTE
, "nfs4open_otw:"
1531 " EXCLUSIVE4: sending a SETATTR"));
1533 * If doing an exclusive create, then generate
1534 * a SETATTR to set the initial attributes.
1535 * Try to set the mtime and the atime to the
1536 * server's current time. It is somewhat
1537 * expected that these fields will be used to
1538 * store the exclusive create cookie. If not,
1539 * server implementors will need to know that
1540 * a SETATTR will follow an exclusive create
1541 * and the cookie should be destroyed if
1544 * The AT_GID and AT_SIZE bits are turned off
1545 * so that the SETATTR request will not attempt
1546 * to process these. The gid will be set
1547 * separately if appropriate. The size is turned
1548 * off because it is assumed that a new file will
1549 * be created empty and if the file wasn't empty,
1550 * then the exclusive create will have failed
1551 * because the file must have existed already.
1552 * Therefore, no truncate operation is needed.
1554 in_va
->va_mask
&= ~(AT_GID
| AT_SIZE
);
1555 in_va
->va_mask
|= (AT_MTIME
| AT_ATIME
);
1557 e
.error
= nfs4setattr(vp
, in_va
, 0, cr
, NULL
);
1560 * Couldn't correct the attributes of
1561 * the newly created file and the
1562 * attributes are wrong. Remove the
1563 * file and return an error to the
1566 /* XXX will this take care of client state ? */
1567 NFS4_DEBUG(nfs4_client_state_debug
, (CE_NOTE
,
1568 "nfs4open_otw: EXCLUSIVE4: error %d on SETATTR:"
1569 " remove file", e
.error
));
1571 (void) nfs4_remove(dvp
, file_name
, cr
, NULL
, 0);
1573 * Since we've reled the vnode and removed
1574 * the file we now need to return the error.
1575 * At this point we don't want to update the
1576 * dircaches, call nfs4_waitfor_purge_complete
1577 * or set vpp to vp so we need to skip these
1580 goto skip_update_dircaches
;
1585 * If we created or found the correct vnode, due to create_flag or
1586 * fh_differs being set, then update directory cache attribute, readdir
1589 if (create_flag
|| fh_differs
) {
1590 dirattr_info_t dinfo
, *dinfop
;
1593 * Make sure getattr succeeded before using results.
1594 * note: op 7 is getattr(dir) for both flavors of
1597 if (create_flag
&& res
.status
== NFS4_OK
) {
1598 dinfo
.di_time_call
= t
;
1601 &res
.array
[6].nfs_resop4_u
.opgetattr
.ga_res
;
1607 nfs4_update_dircaches(&op_res
->cinfo
, dvp
, vp
, file_name
,
1612 * If the page cache for this file was flushed from actions
1613 * above, it was done asynchronously and if that is true,
1614 * there is a need to wait here for it to complete. This must
1615 * be done outside of start_fop/end_fop.
1617 (void) nfs4_waitfor_purge_complete(vp
);
1620 * It is implicit that we are in the open case (create_flag == 0) since
1621 * fh_differs can only be set to a non-zero value in the open case.
1623 if (fh_differs
!= 0 && vpi
!= NULL
)
1627 * Be sure to set *vpp to the correct value before returning.
1631 skip_update_dircaches
:
1633 nfs4args_copen_free(open_args
);
1635 nfs4args_verify_free(&argop
[8]);
1636 nfs4args_setattr_free(&argop
[9]);
1638 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
1642 kmem_free(argop
, argoplist_size
);
1647 * Reopen an open instance. cf. nfs4open_otw().
1649 * Errors are returned by the nfs4_error_t parameter.
1650 * - ep->error contains an errno value or zero.
1651 * - if it is zero, ep->stat is set to an NFS status code, if any.
1652 * If the file could not be reopened, but the caller should continue, the
1653 * file is marked dead and no error values are returned. If the caller
1654 * should stop recovering open files and start over, either the ep->error
1655 * value or ep->stat will indicate an error (either something that requires
1656 * recovery or EAGAIN). Note that some recovery (e.g., expired volatile
1657 * filehandles) may be handled silently by this routine.
1658 * - if it is EINTR, ETIMEDOUT, or NFS4_FRC_UNMT_ERR, recovery for lost state
1659 * will be started, so the caller should not do it.
1662 * - kill_file : reopen failed in such a fashion to constitute marking the
1663 * file dead and setting the open stream's 'os_failed_reopen' as 1. This
1664 * is for cases where recovery is not possible.
1665 * - failed_reopen : same as above, except that the file has already been
1666 * marked dead, so no need to do it again.
1667 * - bailout : reopen failed but we are able to recover and retry the reopen -
1668 * either within this function immediately or via the calling function.
1672 nfs4_reopen(vnode_t
*vp
, nfs4_open_stream_t
*osp
, nfs4_error_t
*ep
,
1673 open_claim_type4 claim
, bool_t frc_use_claim_previous
,
1676 COMPOUND4args_clnt args
;
1677 COMPOUND4res_clnt res
;
1678 nfs_argop4 argop
[4];
1680 OPEN4res
*op_res
= NULL
;
1681 OPEN4cargs
*open_args
;
1683 rnode4_t
*rp
= VTOR4(vp
);
1685 cred_t
*cr
= NULL
, *cred_otw
= NULL
;
1686 nfs4_open_owner_t
*oop
= NULL
;
1688 nfs4_ga_res_t
*garp
;
1689 char fn
[MAXNAMELEN
];
1690 nfs4_recov_state_t recov
= {NULL
, 0};
1691 nfs4_lost_rqst_t lost_rqst
;
1692 mntinfo4_t
*mi
= VTOMI4(vp
);
1694 char *failed_msg
= "";
1697 nfs4_bseqid_entry_t
*bsep
= NULL
;
1699 ASSERT(nfs4_consistent_type(vp
));
1700 ASSERT(nfs_zone() == mi
->mi_zone
);
1702 nfs4_error_zinit(ep
);
1704 /* this is the cred used to find the open owner */
1705 cr
= state_to_cred(osp
);
1707 failed_msg
= "Couldn't reopen: no cred";
1710 /* use this cred for OTW operations */
1711 cred_otw
= nfs4_get_otw_cred(cr
, mi
, osp
->os_open_owner
);
1714 nfs4_error_zinit(ep
);
1716 if (mi
->mi_vfsp
->vfs_flag
& VFS_UNMOUNTED
) {
1717 /* File system has been unmounted, quit */
1719 failed_msg
= "Couldn't reopen: file system has been unmounted";
1723 oop
= osp
->os_open_owner
;
1725 ASSERT(oop
!= NULL
);
1726 if (oop
== NULL
) { /* be defensive in non-DEBUG */
1727 failed_msg
= "can't reopen: no open owner";
1730 open_owner_hold(oop
);
1732 ep
->error
= nfs4_start_open_seqid_sync(oop
, mi
);
1734 open_owner_rele(oop
);
1740 * If the rnode has a delegation and the delegation has been
1741 * recovered and the server didn't request a recall and the caller
1742 * didn't specifically ask for CLAIM_PREVIOUS (nfs4frlock during
1743 * recovery) and the rnode hasn't been marked dead, then install
1744 * the delegation stateid in the open stream. Otherwise, proceed
1745 * with a CLAIM_PREVIOUS or CLAIM_NULL OPEN.
1747 mutex_enter(&rp
->r_statev4_lock
);
1748 if (rp
->r_deleg_type
!= OPEN_DELEGATE_NONE
&&
1749 !rp
->r_deleg_return_pending
&&
1750 (rp
->r_deleg_needs_recovery
== OPEN_DELEGATE_NONE
) &&
1751 !rp
->r_deleg_needs_recall
&&
1752 claim
!= CLAIM_DELEGATE_CUR
&& !frc_use_claim_previous
&&
1753 !(rp
->r_flags
& R4RECOVERR
)) {
1754 mutex_enter(&osp
->os_sync_lock
);
1755 osp
->os_delegation
= 1;
1756 osp
->open_stateid
= rp
->r_deleg_stateid
;
1757 mutex_exit(&osp
->os_sync_lock
);
1758 mutex_exit(&rp
->r_statev4_lock
);
1761 mutex_exit(&rp
->r_statev4_lock
);
1764 * If the file failed recovery, just quit. This failure need not
1765 * affect other reopens, so don't return an error.
1767 mutex_enter(&rp
->r_statelock
);
1768 if (rp
->r_flags
& R4RECOVERR
) {
1769 mutex_exit(&rp
->r_statelock
);
1773 mutex_exit(&rp
->r_statelock
);
1776 * argop is empty here
1778 * PUTFH, OPEN, GETATTR
1780 args
.ctag
= TAG_REOPEN
;
1784 NFS4_DEBUG(nfs4_client_failover_debug
, (CE_NOTE
,
1785 "nfs4_reopen: file is type %d, id %s",
1786 vp
->v_type
, rnode4info(VTOR4(vp
))));
1788 argop
[0].argop
= OP_CPUTFH
;
1790 if (claim
!= CLAIM_PREVIOUS
) {
1792 * if this is a file mount then
1793 * use the mntinfo parentfh
1795 argop
[0].nfs_argop4_u
.opcputfh
.sfh
=
1796 (vp
->v_flag
& VROOT
) ? mi
->mi_srvparentfh
:
1799 /* putfh fh to reopen */
1800 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= rp
->r_fh
;
1803 argop
[1].argop
= OP_COPEN
;
1804 open_args
= &argop
[1].nfs_argop4_u
.opcopen
;
1805 open_args
->claim
= claim
;
1807 if (claim
== CLAIM_NULL
) {
1809 if ((ep
->error
= vtoname(vp
, fn
, MAXNAMELEN
)) != 0) {
1810 nfs_cmn_err(ep
->error
, CE_WARN
, "nfs4_reopen: vtoname "
1811 "failed for vp 0x%p for CLAIM_NULL with %m",
1813 failed_msg
= "Couldn't reopen: vtoname failed for "
1815 /* nothing allocated yet */
1819 open_args
->open_claim4_u
.cfile
= fn
;
1820 } else if (claim
== CLAIM_PREVIOUS
) {
1823 * We have two cases to deal with here:
1824 * 1) We're being called to reopen files in order to satisfy
1825 * a lock operation request which requires us to explicitly
1826 * reopen files which were opened under a delegation. If
1827 * we're in recovery, we *must* use CLAIM_PREVIOUS. In
1828 * that case, frc_use_claim_previous is TRUE and we must
1829 * use the rnode's current delegation type (r_deleg_type).
1830 * 2) We're reopening files during some form of recovery.
1831 * In this case, frc_use_claim_previous is FALSE and we
1832 * use the delegation type appropriate for recovery
1833 * (r_deleg_needs_recovery).
1835 mutex_enter(&rp
->r_statev4_lock
);
1836 open_args
->open_claim4_u
.delegate_type
=
1837 frc_use_claim_previous
?
1839 rp
->r_deleg_needs_recovery
;
1840 mutex_exit(&rp
->r_statev4_lock
);
1842 } else if (claim
== CLAIM_DELEGATE_CUR
) {
1844 if ((ep
->error
= vtoname(vp
, fn
, MAXNAMELEN
)) != 0) {
1845 nfs_cmn_err(ep
->error
, CE_WARN
, "nfs4_reopen: vtoname "
1846 "failed for vp 0x%p for CLAIM_DELEGATE_CUR "
1847 "with %m", (void *)vp
);
1848 failed_msg
= "Couldn't reopen: vtoname failed for "
1849 "CLAIM_DELEGATE_CUR";
1850 /* nothing allocated yet */
1854 mutex_enter(&rp
->r_statev4_lock
);
1855 open_args
->open_claim4_u
.delegate_cur_info
.delegate_stateid
=
1856 rp
->r_deleg_stateid
;
1857 mutex_exit(&rp
->r_statev4_lock
);
1859 open_args
->open_claim4_u
.delegate_cur_info
.cfile
= fn
;
1861 open_args
->opentype
= OPEN4_NOCREATE
;
1862 open_args
->owner
.clientid
= mi2clientid(mi
);
1863 open_args
->owner
.owner_len
= sizeof (oop
->oo_name
);
1864 open_args
->owner
.owner_val
=
1865 kmem_alloc(open_args
->owner
.owner_len
, KM_SLEEP
);
1866 bcopy(&oop
->oo_name
, open_args
->owner
.owner_val
,
1867 open_args
->owner
.owner_len
);
1868 open_args
->share_access
= 0;
1869 open_args
->share_deny
= 0;
1871 mutex_enter(&osp
->os_sync_lock
);
1872 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
, "nfs4_reopen: osp %p rp "
1873 "%p: read acc %"PRIu64
" write acc %"PRIu64
": open ref count %d: "
1874 "mmap read %"PRIu64
" mmap write %"PRIu64
" claim %d ",
1875 (void *)osp
, (void *)rp
, osp
->os_share_acc_read
,
1876 osp
->os_share_acc_write
, osp
->os_open_ref_count
,
1877 osp
->os_mmap_read
, osp
->os_mmap_write
, claim
));
1879 if (osp
->os_share_acc_read
|| osp
->os_mmap_read
)
1880 open_args
->share_access
|= OPEN4_SHARE_ACCESS_READ
;
1881 if (osp
->os_share_acc_write
|| osp
->os_mmap_write
)
1882 open_args
->share_access
|= OPEN4_SHARE_ACCESS_WRITE
;
1883 if (osp
->os_share_deny_read
)
1884 open_args
->share_deny
|= OPEN4_SHARE_DENY_READ
;
1885 if (osp
->os_share_deny_write
)
1886 open_args
->share_deny
|= OPEN4_SHARE_DENY_WRITE
;
1887 mutex_exit(&osp
->os_sync_lock
);
1889 seqid
= nfs4_get_open_seqid(oop
) + 1;
1890 open_args
->seqid
= seqid
;
1892 /* Construct the getfh part of the compound */
1893 argop
[2].argop
= OP_GETFH
;
1895 /* Construct the getattr part of the compound */
1896 argop
[3].argop
= OP_GETATTR
;
1897 argop
[3].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
1898 argop
[3].nfs_argop4_u
.opgetattr
.mi
= mi
;
1902 rfs4call(mi
, &args
, &res
, cred_otw
, &doqueue
, 0, ep
);
1905 if (!is_recov
&& !frc_use_claim_previous
&&
1906 (ep
->error
== EINTR
|| ep
->error
== ETIMEDOUT
||
1907 NFS4_FRC_UNMT_ERR(ep
->error
, vp
->v_vfsp
))) {
1908 nfs4open_save_lost_rqst(ep
->error
, &lost_rqst
, oop
,
1909 cred_otw
, vp
, NULL
, open_args
);
1910 abort
= nfs4_start_recovery(ep
,
1911 VTOMI4(vp
), vp
, NULL
, NULL
,
1912 lost_rqst
.lr_op
== OP_OPEN
?
1913 &lost_rqst
: NULL
, OP_OPEN
, NULL
, NULL
, NULL
);
1914 nfs4args_copen_free(open_args
);
1918 nfs4args_copen_free(open_args
);
1920 if (ep
->error
== EACCES
&& cred_otw
!= cr
) {
1924 nfs4_end_open_seqid_sync(oop
);
1925 open_owner_rele(oop
);
1929 if (ep
->error
== ETIMEDOUT
)
1931 failed_msg
= "Couldn't reopen: rpc error";
1935 if (nfs4_need_to_bump_seqid(&res
))
1936 nfs4_set_open_seqid(seqid
, oop
, args
.ctag
);
1938 switch (res
.status
) {
1940 if (recov
.rs_flags
& NFS4_RS_DELAY_MSG
) {
1941 mutex_enter(&rp
->r_statelock
);
1942 rp
->r_delay_interval
= 0;
1943 mutex_exit(&rp
->r_statelock
);
1946 case NFS4ERR_BAD_SEQID
:
1947 bsep
= nfs4_create_bseqid_entry(oop
, NULL
, vp
, 0,
1948 args
.ctag
, open_args
->seqid
);
1950 abort
= nfs4_start_recovery(ep
, VTOMI4(vp
), vp
, NULL
,
1951 NULL
, lost_rqst
.lr_op
== OP_OPEN
? &lost_rqst
:
1952 NULL
, OP_OPEN
, bsep
, NULL
, NULL
);
1954 nfs4args_copen_free(open_args
);
1955 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
1956 nfs4_end_open_seqid_sync(oop
);
1957 open_owner_rele(oop
);
1959 kmem_free(bsep
, sizeof (*bsep
));
1962 case NFS4ERR_NO_GRACE
:
1963 nfs4args_copen_free(open_args
);
1964 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
1965 nfs4_end_open_seqid_sync(oop
);
1966 open_owner_rele(oop
);
1968 if (claim
== CLAIM_PREVIOUS
) {
1970 * Retry as a plain open. We don't need to worry about
1971 * checking the changeinfo: it is acceptable for a
1972 * client to re-open a file and continue processing
1973 * (in the absence of locks).
1975 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
,
1976 "nfs4_reopen: CLAIM_PREVIOUS: NFS4ERR_NO_GRACE; "
1977 "will retry as CLAIM_NULL"));
1979 nfs4_mi_kstat_inc_no_grace(mi
);
1983 "Couldn't reopen: tried reclaim outside grace period. ";
1986 nfs4_set_grace_wait(mi
);
1987 nfs4args_copen_free(open_args
);
1988 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
1989 nfs4_end_open_seqid_sync(oop
);
1990 open_owner_rele(oop
);
1992 ep
->error
= nfs4_wait_for_grace(mi
, &recov
);
1997 nfs4_set_delay_wait(vp
);
1998 nfs4args_copen_free(open_args
);
1999 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
2000 nfs4_end_open_seqid_sync(oop
);
2001 open_owner_rele(oop
);
2003 ep
->error
= nfs4_wait_for_delay(vp
, &recov
);
2004 nfs4_mi_kstat_inc_delay(mi
);
2008 case NFS4ERR_FHEXPIRED
:
2009 /* recover filehandle and retry */
2010 abort
= nfs4_start_recovery(ep
,
2011 mi
, vp
, NULL
, NULL
, NULL
, OP_OPEN
, NULL
, NULL
, NULL
);
2012 nfs4args_copen_free(open_args
);
2013 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
2014 nfs4_end_open_seqid_sync(oop
);
2015 open_owner_rele(oop
);
2019 failed_msg
= "Couldn't reopen: recovery aborted";
2021 case NFS4ERR_RESOURCE
:
2022 case NFS4ERR_STALE_CLIENTID
:
2023 case NFS4ERR_WRONGSEC
:
2024 case NFS4ERR_EXPIRED
:
2026 * Do not mark the file dead and let the calling
2027 * function initiate recovery.
2029 nfs4args_copen_free(open_args
);
2030 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
2031 nfs4_end_open_seqid_sync(oop
);
2032 open_owner_rele(oop
);
2035 case NFS4ERR_ACCESS
:
2036 if (cred_otw
!= cr
) {
2040 nfs4args_copen_free(open_args
);
2041 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
2042 nfs4_end_open_seqid_sync(oop
);
2043 open_owner_rele(oop
);
2049 NFS4_DEBUG(nfs4_client_failover_debug
, (CE_NOTE
,
2050 "nfs4_reopen: r_server 0x%p, mi_curr_serv 0x%p, rnode %s",
2051 (void*)VTOR4(vp
)->r_server
, (void*)mi
->mi_curr_serv
,
2052 rnode4info(VTOR4(vp
))));
2053 failed_msg
= "Couldn't reopen: NFSv4 error";
2054 nfs4args_copen_free(open_args
);
2055 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
2059 resop
= &res
.array
[1]; /* open res */
2060 op_res
= &resop
->nfs_resop4_u
.opopen
;
2062 garp
= &res
.array
[3].nfs_resop4_u
.opgetattr
.ga_res
;
2065 * Check if the path we reopened really is the same
2066 * file. We could end up in a situation where the file
2067 * was removed and a new file created with the same name.
2069 resop
= &res
.array
[2];
2070 gf_res
= &resop
->nfs_resop4_u
.opgetfh
;
2071 (void) nfs_rw_enter_sig(&mi
->mi_fh_lock
, RW_READER
, 0);
2072 fh_different
= (nfs4cmpfh(&rp
->r_fh
->sfh_fh
, &gf_res
->object
) != 0);
2074 if (mi
->mi_fh_expire_type
== FH4_PERSISTENT
||
2075 mi
->mi_fh_expire_type
& FH4_NOEXPIRE_WITH_OPEN
) {
2076 /* Oops, we don't have the same file */
2077 if (mi
->mi_fh_expire_type
== FH4_PERSISTENT
)
2078 failed_msg
= "Couldn't reopen: Persistent "
2079 "file handle changed";
2081 failed_msg
= "Couldn't reopen: Volatile "
2082 "(no expire on open) file handle changed";
2084 nfs4args_copen_free(open_args
);
2085 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
2086 nfs_rw_exit(&mi
->mi_fh_lock
);
2091 * We have volatile file handles that don't compare.
2092 * If the fids are the same then we assume that the
2093 * file handle expired but the rnode still refers to
2094 * the same file object.
2096 * First check that we have fids or not.
2097 * If we don't we have a dumb server so we will
2098 * just assume every thing is ok for now.
2100 if (!ep
->error
&& garp
->n4g_va
.va_mask
& AT_NODEID
&&
2101 rp
->r_attr
.va_mask
& AT_NODEID
&&
2102 rp
->r_attr
.va_nodeid
!= garp
->n4g_va
.va_nodeid
) {
2104 * We have fids, but they don't
2105 * compare. So kill the file.
2108 "Couldn't reopen: file handle changed"
2109 " due to mismatched fids";
2110 nfs4args_copen_free(open_args
);
2111 (void) xdr_free(xdr_COMPOUND4res_clnt
,
2113 nfs_rw_exit(&mi
->mi_fh_lock
);
2117 * We have volatile file handles that refers
2118 * to the same file (at least they have the
2119 * same fid) or we don't have fids so we
2120 * can't tell. :(. We'll be a kind and accepting
2121 * client so we'll update the rnode's file
2122 * handle with the otw handle.
2124 * We need to drop mi->mi_fh_lock since
2125 * sh4_update acquires it. Since there is
2126 * only one recovery thread there is no
2129 nfs_rw_exit(&mi
->mi_fh_lock
);
2130 sfh4_update(rp
->r_fh
, &gf_res
->object
);
2134 nfs_rw_exit(&mi
->mi_fh_lock
);
2137 ASSERT(nfs4_consistent_type(vp
));
2140 * If the server wanted an OPEN_CONFIRM but that fails, just start
2141 * over. Presumably if there is a persistent error it will show up
2142 * when we resend the OPEN.
2144 if (op_res
->rflags
& OPEN4_RESULT_CONFIRM
) {
2145 bool_t retry_open
= FALSE
;
2147 nfs4open_confirm(vp
, &seqid
, &op_res
->stateid
,
2148 cred_otw
, is_recov
, &retry_open
,
2149 oop
, FALSE
, ep
, NULL
);
2150 if (ep
->error
|| ep
->stat
) {
2151 nfs4args_copen_free(open_args
);
2152 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
2153 nfs4_end_open_seqid_sync(oop
);
2154 open_owner_rele(oop
);
2160 mutex_enter(&osp
->os_sync_lock
);
2161 osp
->open_stateid
= op_res
->stateid
;
2162 osp
->os_delegation
= 0;
2164 * Need to reset this bitfield for the possible case where we were
2165 * going to OTW CLOSE the file, got a non-recoverable error, and before
2166 * we could retry the CLOSE, OPENed the file again.
2168 ASSERT(osp
->os_open_owner
->oo_seqid_inuse
);
2169 osp
->os_final_close
= 0;
2170 osp
->os_force_close
= 0;
2171 if (claim
== CLAIM_DELEGATE_CUR
|| claim
== CLAIM_PREVIOUS
)
2172 osp
->os_dc_openacc
= open_args
->share_access
;
2173 mutex_exit(&osp
->os_sync_lock
);
2175 nfs4_end_open_seqid_sync(oop
);
2177 /* accept delegation, if any */
2178 nfs4_delegation_accept(rp
, claim
, op_res
, garp
, cred_otw
);
2180 nfs4args_copen_free(open_args
);
2182 nfs4_attr_cache(vp
, garp
, t
, cr
, TRUE
, NULL
);
2184 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
2186 ASSERT(nfs4_consistent_type(vp
));
2188 open_owner_rele(oop
);
2194 nfs4_fail_recov(vp
, failed_msg
, ep
->error
, ep
->stat
);
2196 NFS4_DEBUG(nfs4_open_stream_debug
, (CE_NOTE
,
2197 "nfs4_reopen: setting os_failed_reopen for osp %p, cr %p, rp %s",
2198 (void *)osp
, (void *)cr
, rnode4info(rp
)));
2199 mutex_enter(&osp
->os_sync_lock
);
2200 osp
->os_failed_reopen
= 1;
2201 mutex_exit(&osp
->os_sync_lock
);
2204 nfs4_end_open_seqid_sync(oop
);
2205 open_owner_rele(oop
);
2209 if (cred_otw
!= NULL
)
2213 /* for . and .. OPENs */
2216 nfs4_open_non_reg_file(vnode_t
**vpp
, int flag
, cred_t
*cr
)
2221 ASSERT(nfs_zone() == VTOMI4(*vpp
)->mi_zone
);
2224 * If close-to-open consistency checking is turned off or
2225 * if there is no cached data, we can avoid
2226 * the over the wire getattr. Otherwise, force a
2227 * call to the server to get fresh attributes and to
2228 * check caches. This is required for close-to-open
2232 if (VTOMI4(*vpp
)->mi_flags
& MI4_NOCTO
||
2233 (rp
->r_dir
== NULL
&& !nfs4_has_pages(*vpp
)))
2236 return (nfs4_getattr_otw(*vpp
, &gar
, cr
, 0));
2244 nfs4_close(vnode_t
*vp
, int flag
, int count
, offset_t offset
, cred_t
*cr
,
2245 caller_context_t
*ct
)
2251 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
2254 * Remove client state for this (lockowner, file) pair.
2255 * Issue otw v4 call to have the server do the same.
2261 * zone_enter(2) prevents processes from changing zones with NFS files
2262 * open; if we happen to get here from the wrong zone we can't do
2263 * anything over the wire.
2265 if (VTOMI4(vp
)->mi_zone
!= nfs_zone()) {
2267 * We could attempt to clean up locks, except we're sure
2268 * that the current process didn't acquire any locks on
2269 * the file: any attempt to lock a file belong to another zone
2270 * will fail, and one can't lock an NFS file and then change
2271 * zones, as that fails too.
2273 * Returning an error here is the sane thing to do. A
2274 * subsequent call to VN_RELE() which translates to a
2275 * nfs4_inactive() will clean up state: if the zone of the
2276 * vnode's origin is still alive and kicking, the inactive
2277 * thread will handle the request (from the correct zone), and
2278 * everything (minus the OTW close call) should be OK. If the
2279 * zone is going away nfs4_async_inactive() will throw away
2280 * delegations, open streams and cached pages inline.
2286 * If we are using local locking for this filesystem, then
2287 * release all of the SYSV style record locks. Otherwise,
2288 * we are doing network locking and we need to release all
2289 * of the network locks. All of the locks held by this
2290 * process on this file are released no matter what the
2291 * incoming reference count is.
2293 if (VTOMI4(vp
)->mi_flags
& MI4_LLOCK
) {
2294 cleanlocks(vp
, ttoproc(curthread
)->p_pid
, 0);
2295 cleanshares(vp
, ttoproc(curthread
)->p_pid
);
2297 e
.error
= nfs4_lockrelease(vp
, flag
, offset
, cr
);
2300 struct lm_sysid
*lmsid
;
2301 lmsid
= nfs4_find_sysid(VTOMI4(vp
));
2302 if (lmsid
== NULL
) {
2303 DTRACE_PROBE2(unknown__sysid
, int, e
.error
,
2306 cleanlocks(vp
, ttoproc(curthread
)->p_pid
,
2307 (lm_sysidt(lmsid
) | LM_SYSID_CLIENT
));
2309 lm_rel_sysid(lmsid
);
2318 * If the file has been `unlinked', then purge the
2319 * DNLC so that this vnode will get reycled quicker
2320 * and the .nfs* file on the server will get removed.
2322 if (rp
->r_unldvp
!= NULL
)
2326 * If the file was open for write and there are pages,
2327 * do a synchronous flush and commit of all of the
2328 * dirty and uncommitted pages.
2331 if ((flag
& FWRITE
) && nfs4_has_pages(vp
))
2332 error
= nfs4_putpage_commit(vp
, 0, 0, cr
);
2334 mutex_enter(&rp
->r_statelock
);
2335 r_error
= rp
->r_error
;
2337 mutex_exit(&rp
->r_statelock
);
2340 * If this file type is one for which no explicit 'open' was
2341 * done, then bail now (ie. no need for protocol 'close'). If
2342 * there was an error w/the vm subsystem, return _that_ error,
2343 * otherwise, return any errors that may've been reported via
2346 if (vp
->v_type
!= VREG
)
2347 return (error
? error
: r_error
);
2350 * The sync putpage commit may have failed above, but since
2351 * we're working w/a regular file, we need to do the protocol
2352 * 'close' (nfs4close_one will figure out if an otw close is
2353 * needed or not). Report any errors _after_ doing the protocol
2356 nfs4close_one(vp
, NULL
, cr
, flag
, NULL
, &e
, CLOSE_NORM
, 0, 0, 0);
2357 n4error
= e
.error
? e
.error
: geterrno4(e
.stat
);
2360 * Error reporting prio (Hi -> Lo)
2362 * i) nfs4_putpage_commit (error)
2363 * ii) rnode's (r_error)
2364 * iii) nfs4close_one (n4error)
2366 return (error
? error
: (r_error
? r_error
: n4error
));
2370 * Initialize *lost_rqstp.
2374 nfs4close_save_lost_rqst(int error
, nfs4_lost_rqst_t
*lost_rqstp
,
2375 nfs4_open_owner_t
*oop
, nfs4_open_stream_t
*osp
, cred_t
*cr
,
2378 if (error
!= ETIMEDOUT
&& error
!= EINTR
&&
2379 !NFS4_FRC_UNMT_ERR(error
, vp
->v_vfsp
)) {
2380 lost_rqstp
->lr_op
= 0;
2384 NFS4_DEBUG(nfs4_lost_rqst_debug
, (CE_NOTE
,
2385 "nfs4close_save_lost_rqst: error %d", error
));
2387 lost_rqstp
->lr_op
= OP_CLOSE
;
2389 * The vp is held and rele'd via the recovery code.
2390 * See nfs4_save_lost_rqst.
2392 lost_rqstp
->lr_vp
= vp
;
2393 lost_rqstp
->lr_dvp
= NULL
;
2394 lost_rqstp
->lr_oop
= oop
;
2395 lost_rqstp
->lr_osp
= osp
;
2396 ASSERT(osp
!= NULL
);
2397 ASSERT(mutex_owned(&osp
->os_sync_lock
));
2398 osp
->os_pending_close
= 1;
2399 lost_rqstp
->lr_lop
= NULL
;
2400 lost_rqstp
->lr_cr
= cr
;
2401 lost_rqstp
->lr_flk
= NULL
;
2402 lost_rqstp
->lr_putfirst
= FALSE
;
2406 * Assumes you already have the open seqid sync grabbed as well as the
2407 * 'os_sync_lock'. Note: this will release the open seqid sync and
2408 * 'os_sync_lock' if client recovery starts. Calling functions have to
2409 * be prepared to handle this.
2411 * 'recov' is returned as 1 if the CLOSE operation detected client recovery
2412 * was needed and was started, and that the calling function should retry
2413 * this function; otherwise it is returned as 0.
2415 * Errors are returned via the nfs4_error_t parameter.
2418 nfs4close_otw(rnode4_t
*rp
, cred_t
*cred_otw
, nfs4_open_owner_t
*oop
,
2419 nfs4_open_stream_t
*osp
, int *recov
, int *did_start_seqid_syncp
,
2420 nfs4_close_type_t close_type
, nfs4_error_t
*ep
, int *have_sync_lockp
)
2422 COMPOUND4args_clnt args
;
2423 COMPOUND4res_clnt res
;
2424 CLOSE4args
*close_args
;
2426 nfs_argop4 argop
[3];
2431 bool_t needrecov
= FALSE
;
2432 nfs4_lost_rqst_t lost_rqst
;
2435 ASSERT(nfs_zone() == VTOMI4(RTOV4(rp
))->mi_zone
);
2437 ASSERT(MUTEX_HELD(&osp
->os_sync_lock
));
2439 NFS4_DEBUG(nfs4_client_state_debug
, (CE_NOTE
, "nfs4close_otw"));
2441 /* Only set this to 1 if recovery is started */
2444 /* do the OTW call to close the file */
2446 if (close_type
== CLOSE_RESEND
)
2447 args
.ctag
= TAG_CLOSE_LOST
;
2448 else if (close_type
== CLOSE_AFTER_RESEND
)
2449 args
.ctag
= TAG_CLOSE_UNDO
;
2451 args
.ctag
= TAG_CLOSE
;
2460 /* putfh target fh */
2461 argop
[0].argop
= OP_CPUTFH
;
2462 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= rp
->r_fh
;
2464 argop
[1].argop
= OP_GETATTR
;
2465 argop
[1].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
2466 argop
[1].nfs_argop4_u
.opgetattr
.mi
= mi
;
2468 argop
[2].argop
= OP_CLOSE
;
2469 close_args
= &argop
[2].nfs_argop4_u
.opclose
;
2471 seqid
= nfs4_get_open_seqid(oop
) + 1;
2473 close_args
->seqid
= seqid
;
2474 close_args
->open_stateid
= osp
->open_stateid
;
2476 NFS4_DEBUG(nfs4_client_call_debug
, (CE_NOTE
,
2477 "nfs4close_otw: %s call, rp %s", needrecov
? "recov" : "first",
2482 rfs4call(mi
, &args
, &res
, cred_otw
, &doqueue
, 0, ep
);
2484 if (!ep
->error
&& nfs4_need_to_bump_seqid(&res
)) {
2485 nfs4_set_open_seqid(seqid
, oop
, args
.ctag
);
2488 needrecov
= nfs4_needs_recovery(ep
, TRUE
, mi
->mi_vfsp
);
2489 if (ep
->error
&& !needrecov
) {
2491 * if there was an error and no recovery is to be done
2492 * then then set up the file to flush its cache if
2493 * needed for the next caller.
2495 mutex_enter(&rp
->r_statelock
);
2496 PURGE_ATTRCACHE4_LOCKED(rp
);
2497 rp
->r_flags
&= ~R4WRITEMODIFIED
;
2498 mutex_exit(&rp
->r_statelock
);
2504 nfs4_bseqid_entry_t
*bsep
= NULL
;
2506 if (close_type
!= CLOSE_RESEND
)
2507 nfs4close_save_lost_rqst(ep
->error
, &lost_rqst
, oop
,
2510 if (!ep
->error
&& res
.status
== NFS4ERR_BAD_SEQID
)
2511 bsep
= nfs4_create_bseqid_entry(oop
, NULL
, vp
,
2512 0, args
.ctag
, close_args
->seqid
);
2514 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
,
2515 "nfs4close_otw: initiating recovery. error %d "
2516 "res.status %d", ep
->error
, res
.status
));
2519 * Drop the 'os_sync_lock' here so we don't hit
2520 * a potential recursive mutex_enter via an
2521 * 'open_stream_hold()'.
2523 mutex_exit(&osp
->os_sync_lock
);
2524 *have_sync_lockp
= 0;
2525 abort
= nfs4_start_recovery(ep
, VTOMI4(vp
), vp
, NULL
, NULL
,
2526 (close_type
!= CLOSE_RESEND
&&
2527 lost_rqst
.lr_op
== OP_CLOSE
) ? &lost_rqst
: NULL
,
2528 OP_CLOSE
, bsep
, NULL
, NULL
);
2530 /* drop open seq sync, and let the calling function regrab it */
2531 nfs4_end_open_seqid_sync(oop
);
2532 *did_start_seqid_syncp
= 0;
2535 kmem_free(bsep
, sizeof (*bsep
));
2537 * For signals, the caller wants to quit, so don't say to
2538 * retry. For forced unmount, if it's a user thread, it
2539 * wants to quit. If it's a recovery thread, the retry
2540 * will happen higher-up on the call stack. Either way,
2541 * don't say to retry.
2543 if (abort
== FALSE
&& ep
->error
!= EINTR
&&
2544 !NFS4_FRC_UNMT_ERR(ep
->error
, mi
->mi_vfsp
) &&
2545 close_type
!= CLOSE_RESEND
&&
2546 close_type
!= CLOSE_AFTER_RESEND
)
2552 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
2557 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
2561 mutex_enter(&rp
->r_statev4_lock
);
2563 mutex_exit(&rp
->r_statev4_lock
);
2565 resop
= &res
.array
[2];
2566 osp
->open_stateid
= resop
->nfs_resop4_u
.opclose
.open_stateid
;
2570 * This removes the reference obtained at OPEN; ie, when the
2571 * open stream structure was created.
2573 * We don't have to worry about calling 'open_stream_rele'
2574 * since we our currently holding a reference to the open
2575 * stream which means the count cannot go to 0 with this
2578 ASSERT(osp
->os_ref_count
>= 2);
2579 osp
->os_ref_count
--;
2581 if (ep
->error
== 0) {
2583 * Avoid a deadlock with the r_serial thread waiting for
2584 * os_sync_lock in nfs4_get_otw_cred_by_osp() which might be
2585 * held by us. We will wait in nfs4_attr_cache() for the
2586 * completion of the r_serial thread.
2588 mutex_exit(&osp
->os_sync_lock
);
2589 *have_sync_lockp
= 0;
2592 &res
.array
[1].nfs_resop4_u
.opgetattr
.ga_res
,
2593 t
, cred_otw
, TRUE
, NULL
);
2596 NFS4_DEBUG(nfs4_client_state_debug
, (CE_NOTE
, "nfs4close_otw:"
2597 " returning %d", ep
->error
));
2599 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
2604 nfs4_read(vnode_t
*vp
, struct uio
*uiop
, int ioflag
, cred_t
*cr
,
2605 caller_context_t
*ct
)
2619 ASSERT(nfs_rw_lock_held(&rp
->r_rwlock
, RW_READER
));
2621 if (IS_SHADOW(vp
, rp
))
2624 if (vp
->v_type
!= VREG
)
2629 if (nfs_zone() != mi
->mi_zone
)
2632 if (uiop
->uio_resid
== 0)
2635 if (uiop
->uio_loffset
< 0 || uiop
->uio_loffset
+ uiop
->uio_resid
< 0)
2638 mutex_enter(&rp
->r_statelock
);
2639 if (rp
->r_flags
& R4RECOVERRP
)
2640 error
= (rp
->r_error
? rp
->r_error
: EIO
);
2643 mutex_exit(&rp
->r_statelock
);
2648 * Bypass VM if caching has been disabled (e.g., locking) or if
2649 * using client-side direct I/O and the file is not mmap'd and
2650 * there are no cached pages.
2652 if ((vp
->v_flag
& VNOCACHE
) ||
2653 (((rp
->r_flags
& R4DIRECTIO
) || (mi
->mi_flags
& MI4_DIRECTIO
)) &&
2654 rp
->r_mapcnt
== 0 && rp
->r_inmap
== 0 && !nfs4_has_pages(vp
))) {
2657 return (nfs4read(vp
, NULL
, uiop
->uio_loffset
,
2658 uiop
->uio_resid
, &resid
, cr
, FALSE
, uiop
));
2664 off
= uiop
->uio_loffset
& MAXBMASK
; /* mapping offset */
2665 on
= uiop
->uio_loffset
& MAXBOFFSET
; /* Relative offset */
2666 n
= MIN(MAXBSIZE
- on
, uiop
->uio_resid
);
2668 if (error
= nfs4_validate_caches(vp
, cr
))
2671 mutex_enter(&rp
->r_statelock
);
2672 while (rp
->r_flags
& R4INCACHEPURGE
) {
2673 if (!cv_wait_sig(&rp
->r_cv
, &rp
->r_statelock
)) {
2674 mutex_exit(&rp
->r_statelock
);
2678 diff
= rp
->r_size
- uiop
->uio_loffset
;
2679 mutex_exit(&rp
->r_statelock
);
2689 error
= vpm_data_copy(vp
, off
+ on
, n
, uiop
,
2690 1, NULL
, 0, S_READ
);
2692 base
= segmap_getmapflt(segkmap
, vp
, off
+ on
, n
, 1,
2695 error
= uiomove(base
+ on
, n
, UIO_READ
, uiop
);
2700 * If read a whole block or read to eof,
2701 * won't need this buffer again soon.
2703 mutex_enter(&rp
->r_statelock
);
2704 if (n
+ on
== MAXBSIZE
||
2705 uiop
->uio_loffset
== rp
->r_size
)
2706 flags
= SM_DONTNEED
;
2709 mutex_exit(&rp
->r_statelock
);
2711 error
= vpm_sync_pages(vp
, off
, n
, flags
);
2713 error
= segmap_release(segkmap
, base
, flags
);
2717 (void) vpm_sync_pages(vp
, off
, n
, 0);
2719 (void) segmap_release(segkmap
, base
, 0);
2722 } while (!error
&& uiop
->uio_resid
> 0);
2729 nfs4_write(vnode_t
*vp
, struct uio
*uiop
, int ioflag
, cred_t
*cr
,
2730 caller_context_t
*ct
)
2732 rlim64_t limit
= uiop
->uio_llimit
;
2748 if (IS_SHADOW(vp
, rp
))
2751 if (vp
->v_type
!= VREG
)
2756 if (nfs_zone() != mi
->mi_zone
)
2759 if (uiop
->uio_resid
== 0)
2762 mutex_enter(&rp
->r_statelock
);
2763 if (rp
->r_flags
& R4RECOVERRP
)
2764 error
= (rp
->r_error
? rp
->r_error
: EIO
);
2767 mutex_exit(&rp
->r_statelock
);
2771 if (ioflag
& FAPPEND
) {
2775 * Must serialize if appending.
2777 if (nfs_rw_lock_held(&rp
->r_rwlock
, RW_READER
)) {
2778 nfs_rw_exit(&rp
->r_rwlock
);
2779 if (nfs_rw_enter_sig(&rp
->r_rwlock
, RW_WRITER
,
2784 va
.va_mask
= AT_SIZE
;
2785 error
= nfs4getattr(vp
, &va
, cr
);
2788 uiop
->uio_loffset
= va
.va_size
;
2791 offset
= uiop
->uio_loffset
+ uiop
->uio_resid
;
2793 if (uiop
->uio_loffset
< (offset_t
)0 || offset
< 0)
2796 if (limit
== RLIM64_INFINITY
|| limit
> MAXOFFSET_T
)
2797 limit
= MAXOFFSET_T
;
2800 * Check to make sure that the process will not exceed
2801 * its limit on file size. It is okay to write up to
2802 * the limit, but not beyond. Thus, the write which
2803 * reaches the limit will be short and the next write
2804 * will return an error.
2807 if (offset
> uiop
->uio_llimit
) {
2808 remainder
= offset
- uiop
->uio_llimit
;
2809 uiop
->uio_resid
= uiop
->uio_llimit
- uiop
->uio_loffset
;
2810 if (uiop
->uio_resid
<= 0) {
2811 proc_t
*p
= ttoproc(curthread
);
2813 uiop
->uio_resid
+= remainder
;
2814 mutex_enter(&p
->p_lock
);
2815 (void) rctl_action(rctlproc_legacy
[RLIMIT_FSIZE
],
2816 p
->p_rctls
, p
, RCA_UNSAFE_SIGINFO
);
2817 mutex_exit(&p
->p_lock
);
2822 /* update the change attribute, if we have a write delegation */
2824 mutex_enter(&rp
->r_statev4_lock
);
2825 if (rp
->r_deleg_type
== OPEN_DELEGATE_WRITE
)
2826 rp
->r_deleg_change
++;
2828 mutex_exit(&rp
->r_statev4_lock
);
2830 if (nfs_rw_enter_sig(&rp
->r_lkserlock
, RW_READER
, INTR4(vp
)))
2834 * Bypass VM if caching has been disabled (e.g., locking) or if
2835 * using client-side direct I/O and the file is not mmap'd and
2836 * there are no cached pages.
2838 if ((vp
->v_flag
& VNOCACHE
) ||
2839 (((rp
->r_flags
& R4DIRECTIO
) || (mi
->mi_flags
& MI4_DIRECTIO
)) &&
2840 rp
->r_mapcnt
== 0 && rp
->r_inmap
== 0 && !nfs4_has_pages(vp
))) {
2844 stable_how4 stab_comm
;
2846 if (rp
->r_flags
& R4STALE
) {
2847 resid
= uiop
->uio_resid
;
2848 offset
= uiop
->uio_loffset
;
2849 error
= rp
->r_error
;
2851 * A close may have cleared r_error, if so,
2852 * propagate ESTALE error return properly
2859 bufsize
= MIN(uiop
->uio_resid
, mi
->mi_stsize
);
2860 base
= kmem_alloc(bufsize
, KM_SLEEP
);
2862 if (ioflag
& FDSYNC
)
2863 stab_comm
= DATA_SYNC4
;
2865 stab_comm
= FILE_SYNC4
;
2866 resid
= uiop
->uio_resid
;
2867 offset
= uiop
->uio_loffset
;
2868 count
= MIN(uiop
->uio_resid
, bufsize
);
2869 org_offset
= uiop
->uio_loffset
;
2870 error
= uiomove(base
, count
, UIO_WRITE
, uiop
);
2872 error
= nfs4write(vp
, base
, org_offset
,
2873 count
, cr
, &stab_comm
);
2875 mutex_enter(&rp
->r_statelock
);
2876 if (rp
->r_size
< uiop
->uio_loffset
)
2877 rp
->r_size
= uiop
->uio_loffset
;
2878 mutex_exit(&rp
->r_statelock
);
2881 } while (!error
&& uiop
->uio_resid
> 0);
2882 kmem_free(base
, bufsize
);
2886 bsize
= vp
->v_vfsp
->vfs_bsize
;
2889 off
= uiop
->uio_loffset
& MAXBMASK
; /* mapping offset */
2890 on
= uiop
->uio_loffset
& MAXBOFFSET
; /* Relative offset */
2891 n
= MIN(MAXBSIZE
- on
, uiop
->uio_resid
);
2893 resid
= uiop
->uio_resid
;
2894 offset
= uiop
->uio_loffset
;
2896 if (rp
->r_flags
& R4STALE
) {
2897 error
= rp
->r_error
;
2899 * A close may have cleared r_error, if so,
2900 * propagate ESTALE error return properly
2908 * Don't create dirty pages faster than they
2909 * can be cleaned so that the system doesn't
2910 * get imbalanced. If the async queue is
2911 * maxed out, then wait for it to drain before
2912 * creating more dirty pages. Also, wait for
2913 * any threads doing pagewalks in the vop_getattr
2914 * entry points so that they don't block for
2917 mutex_enter(&rp
->r_statelock
);
2918 while ((mi
->mi_max_threads
!= 0 &&
2919 rp
->r_awcount
> 2 * mi
->mi_max_threads
) ||
2922 klwp_t
*lwp
= ttolwp(curthread
);
2926 if (!cv_wait_sig(&rp
->r_cv
, &rp
->r_statelock
)) {
2927 mutex_exit(&rp
->r_statelock
);
2936 cv_wait(&rp
->r_cv
, &rp
->r_statelock
);
2938 mutex_exit(&rp
->r_statelock
);
2941 * Touch the page and fault it in if it is not in core
2942 * before segmap_getmapflt or vpm_data_copy can lock it.
2943 * This is to avoid the deadlock if the buffer is mapped
2944 * to the same file through mmap which we want to write.
2946 uio_prefaultpages((long)n
, uiop
);
2950 * It will use kpm mappings, so no need to
2953 error
= writerp4(rp
, NULL
, n
, uiop
, 0);
2956 int pon
= uiop
->uio_loffset
& PAGEOFFSET
;
2957 size_t pn
= MIN(PAGESIZE
- pon
,
2961 mutex_enter(&rp
->r_statelock
);
2962 pagecreate
= (pon
== 0) && (pn
== PAGESIZE
||
2963 uiop
->uio_loffset
+ pn
>= rp
->r_size
);
2964 mutex_exit(&rp
->r_statelock
);
2966 base
= segmap_getmapflt(segkmap
, vp
, off
+ on
,
2967 pn
, !pagecreate
, S_WRITE
);
2969 error
= writerp4(rp
, base
+ pon
, n
, uiop
,
2973 base
= segmap_getmapflt(segkmap
, vp
, off
+ on
,
2975 error
= writerp4(rp
, base
+ on
, n
, uiop
, 0);
2980 if (mi
->mi_flags
& MI4_NOAC
)
2982 else if ((uiop
->uio_loffset
% bsize
) == 0 ||
2985 * Have written a whole block.
2986 * Start an asynchronous write
2987 * and mark the buffer to
2988 * indicate that it won't be
2989 * needed again soon.
2991 flags
= SM_WRITE
| SM_ASYNC
| SM_DONTNEED
;
2994 if ((ioflag
& (FSYNC
|FDSYNC
)) ||
2995 (rp
->r_flags
& R4OUTOFSPACE
)) {
3000 error
= vpm_sync_pages(vp
, off
, n
, flags
);
3002 error
= segmap_release(segkmap
, base
, flags
);
3006 (void) vpm_sync_pages(vp
, off
, n
, 0);
3008 (void) segmap_release(segkmap
, base
, 0);
3011 * In the event that we got an access error while
3012 * faulting in a page for a write-only file just
3015 if (error
== EACCES
)
3018 } while (!error
&& uiop
->uio_resid
> 0);
3022 uiop
->uio_resid
= resid
+ remainder
;
3023 uiop
->uio_loffset
= offset
;
3025 uiop
->uio_resid
+= remainder
;
3027 mutex_enter(&rp
->r_statev4_lock
);
3028 if (rp
->r_deleg_type
== OPEN_DELEGATE_WRITE
) {
3029 gethrestime(&rp
->r_attr
.va_mtime
);
3030 rp
->r_attr
.va_ctime
= rp
->r_attr
.va_mtime
;
3032 mutex_exit(&rp
->r_statev4_lock
);
3035 nfs_rw_exit(&rp
->r_lkserlock
);
3041 * Flags are composed of {B_ASYNC, B_INVAL, B_FREE, B_DONTNEED}
3044 nfs4_rdwrlbn(vnode_t
*vp
, page_t
*pp
, uoff_t off
, size_t len
,
3045 int flags
, cred_t
*cr
)
3051 stable_how4 stab_comm
;
3053 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
3054 bp
= pageio_setup(pp
, len
, vp
, flags
);
3058 * pageio_setup should have set b_addr to 0. This
3059 * is correct since we want to do I/O on a page
3060 * boundary. bp_mapin will use this addr to calculate
3061 * an offset, and then set b_addr to the kernel virtual
3062 * address it allocated for us.
3064 ASSERT(bp
->b_un
.b_addr
== 0);
3068 bp
->b_lblkno
= lbtodb(off
);
3070 bp
->b_offset
= (offset_t
)off
;
3073 if ((flags
& (B_WRITE
|B_ASYNC
)) == (B_WRITE
|B_ASYNC
) &&
3075 stab_comm
= UNSTABLE4
;
3077 stab_comm
= FILE_SYNC4
;
3079 error
= nfs4_bio(bp
, &stab_comm
, cr
, FALSE
);
3084 if (stab_comm
== UNSTABLE4
)
3085 fsdata
= C_DELAYCOMMIT
;
3087 fsdata
= C_NOCOMMIT
;
3091 pp
->p_fsdata
= fsdata
;
3092 } while ((pp
= pp
->p_next
) != savepp
);
3100 nfs4rdwr_check_osid(vnode_t
*vp
, nfs4_error_t
*ep
, cred_t
*cr
)
3102 nfs4_open_owner_t
*oop
;
3103 nfs4_open_stream_t
*osp
;
3104 rnode4_t
*rp
= VTOR4(vp
);
3105 mntinfo4_t
*mi
= VTOMI4(vp
);
3108 ASSERT(nfs_zone() == mi
->mi_zone
);
3111 oop
= find_open_owner(cr
, NFS4_PERM_CREATED
, mi
);
3115 /* returns with 'os_sync_lock' held */
3116 osp
= find_open_stream(oop
, rp
);
3118 open_owner_rele(oop
);
3122 if (osp
->os_failed_reopen
) {
3123 mutex_exit(&osp
->os_sync_lock
);
3124 open_stream_rele(osp
, rp
);
3125 open_owner_rele(oop
);
3130 * Determine whether a reopen is needed. If this
3131 * is a delegation open stream, then the os_delegation bit
3135 reopen_needed
= osp
->os_delegation
;
3137 mutex_exit(&osp
->os_sync_lock
);
3138 open_owner_rele(oop
);
3140 if (reopen_needed
) {
3141 nfs4_error_zinit(ep
);
3142 nfs4_reopen(vp
, osp
, ep
, CLAIM_NULL
, FALSE
, FALSE
);
3143 mutex_enter(&osp
->os_sync_lock
);
3144 if (ep
->error
|| ep
->stat
|| osp
->os_failed_reopen
) {
3145 mutex_exit(&osp
->os_sync_lock
);
3146 open_stream_rele(osp
, rp
);
3149 mutex_exit(&osp
->os_sync_lock
);
3151 open_stream_rele(osp
, rp
);
3157 * Write to file. Writes to remote server in largest size
3158 * chunks that the server can handle. Write is synchronous.
3161 nfs4write(vnode_t
*vp
, caddr_t base
, uoff_t offset
, int count
, cred_t
*cr
,
3162 stable_how4
*stab_comm
)
3165 COMPOUND4args_clnt args
;
3166 COMPOUND4res_clnt res
;
3169 nfs_argop4 argop
[2];
3176 nfs4_recov_state_t recov_state
;
3177 nfs4_stateid_types_t sid_types
;
3178 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
3184 ASSERT(nfs_zone() == mi
->mi_zone
);
3186 stable
= *stab_comm
;
3187 *stab_comm
= FILE_SYNC4
;
3190 recov_state
.rs_flags
= 0;
3191 recov_state
.rs_num_retry_despite_err
= 0;
3192 nfs4_init_stateid_types(&sid_types
);
3194 /* Is curthread the recovery thread? */
3195 mutex_enter(&mi
->mi_lock
);
3196 recov
= (mi
->mi_recovthread
== curthread
);
3197 mutex_exit(&mi
->mi_lock
);
3200 args
.ctag
= TAG_WRITE
;
3205 e
.error
= nfs4_start_fop(VTOMI4(vp
), vp
, NULL
, OH_WRITE
,
3206 &recov_state
, NULL
);
3211 /* 0. putfh target fh */
3212 argop
[0].argop
= OP_CPUTFH
;
3213 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= rp
->r_fh
;
3216 nfs4args_write(&argop
[1], stable
, rp
, cr
, &wargs
, &sid_types
);
3220 wargs
->offset
= (offset4
)offset
;
3221 wargs
->data_val
= base
;
3223 if (mi
->mi_io_kstats
) {
3224 mutex_enter(&mi
->mi_lock
);
3225 kstat_runq_enter(KSTAT_IO_PTR(mi
->mi_io_kstats
));
3226 mutex_exit(&mi
->mi_lock
);
3229 if ((vp
->v_flag
& VNOCACHE
) ||
3230 (rp
->r_flags
& R4DIRECTIO
) ||
3231 (mi
->mi_flags
& MI4_DIRECTIO
))
3232 tsize
= MIN(mi
->mi_stsize
, count
);
3234 tsize
= MIN(mi
->mi_curwrite
, count
);
3235 wargs
->data_len
= (uint_t
)tsize
;
3236 rfs4call(mi
, &args
, &res
, cr
, &doqueue
, 0, &e
);
3238 if (mi
->mi_io_kstats
) {
3239 mutex_enter(&mi
->mi_lock
);
3240 kstat_runq_exit(KSTAT_IO_PTR(mi
->mi_io_kstats
));
3241 mutex_exit(&mi
->mi_lock
);
3245 needrecov
= nfs4_needs_recovery(&e
, FALSE
, mi
->mi_vfsp
);
3246 if (e
.error
&& !needrecov
) {
3247 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_WRITE
,
3248 &recov_state
, needrecov
);
3257 * Do handling of OLD_STATEID outside
3258 * of the normal recovery framework.
3260 * If write receives a BAD stateid error while using a
3261 * delegation stateid, retry using the open stateid (if it
3262 * exists). If it doesn't have an open stateid, reopen the
3263 * file first, then retry.
3265 if (!e
.error
&& res
.status
== NFS4ERR_OLD_STATEID
&&
3266 sid_types
.cur_sid_type
!= SPEC_SID
) {
3267 nfs4_save_stateid(&wargs
->stateid
, &sid_types
);
3269 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_WRITE
,
3270 &recov_state
, needrecov
);
3271 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
3273 } else if (e
.error
== 0 && res
.status
== NFS4ERR_BAD_STATEID
&&
3274 sid_types
.cur_sid_type
== DEL_SID
) {
3275 nfs4_save_stateid(&wargs
->stateid
, &sid_types
);
3276 mutex_enter(&rp
->r_statev4_lock
);
3277 rp
->r_deleg_return_pending
= TRUE
;
3278 mutex_exit(&rp
->r_statev4_lock
);
3279 if (nfs4rdwr_check_osid(vp
, &e
, cr
)) {
3281 nfs4_end_fop(mi
, vp
, NULL
, OH_WRITE
,
3282 &recov_state
, needrecov
);
3283 (void) xdr_free(xdr_COMPOUND4res_clnt
,
3288 nfs4_end_fop(mi
, vp
, NULL
, OH_WRITE
,
3289 &recov_state
, needrecov
);
3290 /* hold needed for nfs4delegreturn_thread */
3292 nfs4delegreturn_async(rp
, (NFS4_DR_PUSH
|NFS4_DR_REOPEN
|
3293 NFS4_DR_DISCARD
), FALSE
);
3294 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
3301 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
,
3302 "nfs4write: client got error %d, res.status %d"
3303 ", so start recovery", e
.error
, res
.status
));
3305 abort
= nfs4_start_recovery(&e
,
3306 VTOMI4(vp
), vp
, NULL
, &wargs
->stateid
,
3307 NULL
, OP_WRITE
, NULL
, NULL
, NULL
);
3309 e
.error
= geterrno4(res
.status
);
3310 (void) xdr_free(xdr_COMPOUND4res_clnt
,
3313 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_WRITE
,
3314 &recov_state
, needrecov
);
3321 e
.error
= geterrno4(res
.status
);
3322 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
3324 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_WRITE
,
3325 &recov_state
, needrecov
);
3329 resop
= &res
.array
[1]; /* write res */
3330 wres
= &resop
->nfs_resop4_u
.opwrite
;
3332 if ((int)wres
->count
> tsize
) {
3333 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
3335 zcmn_err(getzoneid(), CE_WARN
,
3336 "nfs4write: server wrote %u, requested was %u",
3337 (int)wres
->count
, tsize
);
3339 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_WRITE
,
3340 &recov_state
, needrecov
);
3343 if (wres
->committed
== UNSTABLE4
) {
3344 *stab_comm
= UNSTABLE4
;
3345 if (wargs
->stable
== DATA_SYNC4
||
3346 wargs
->stable
== FILE_SYNC4
) {
3347 (void) xdr_free(xdr_COMPOUND4res_clnt
,
3349 zcmn_err(getzoneid(), CE_WARN
,
3350 "nfs4write: server %s did not commit "
3351 "to stable storage",
3352 rp
->r_server
->sv_hostname
);
3354 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
,
3355 OH_WRITE
, &recov_state
, needrecov
);
3360 tsize
= (int)wres
->count
;
3364 if (mi
->mi_io_kstats
) {
3365 mutex_enter(&mi
->mi_lock
);
3366 KSTAT_IO_PTR(mi
->mi_io_kstats
)->writes
++;
3367 KSTAT_IO_PTR(mi
->mi_io_kstats
)->nwritten
+=
3369 mutex_exit(&mi
->mi_lock
);
3371 lwp_stat_update(LWP_STAT_OUBLK
, 1);
3372 mutex_enter(&rp
->r_statelock
);
3373 if (rp
->r_flags
& R4HAVEVERF
) {
3374 if (rp
->r_writeverf
!= wres
->writeverf
) {
3376 rp
->r_writeverf
= wres
->writeverf
;
3379 rp
->r_writeverf
= wres
->writeverf
;
3380 rp
->r_flags
|= R4HAVEVERF
;
3382 PURGE_ATTRCACHE4_LOCKED(rp
);
3383 rp
->r_flags
|= R4WRITEMODIFIED
;
3384 gethrestime(&rp
->r_attr
.va_mtime
);
3385 rp
->r_attr
.va_ctime
= rp
->r_attr
.va_mtime
;
3386 mutex_exit(&rp
->r_statelock
);
3387 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
3391 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_WRITE
, &recov_state
,
3398 * Read from a file. Reads data in largest chunks our interface can handle.
3401 nfs4read(vnode_t
*vp
, caddr_t base
, offset_t offset
, int count
,
3402 size_t *residp
, cred_t
*cr
, bool_t async
, struct uio
*uiop
)
3405 COMPOUND4args_clnt args
;
3406 COMPOUND4res_clnt res
;
3408 nfs_argop4 argop
[2];
3414 bool_t needrecov
= FALSE
;
3415 nfs4_recov_state_t recov_state
;
3416 nfs4_stateid_types_t sid_types
;
3417 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
3423 ASSERT(nfs_zone() == mi
->mi_zone
);
3425 args
.ctag
= async
? TAG_READAHEAD
: TAG_READ
;
3430 nfs4_init_stateid_types(&sid_types
);
3432 recov_state
.rs_flags
= 0;
3433 recov_state
.rs_num_retry_despite_err
= 0;
3436 e
.error
= nfs4_start_fop(mi
, vp
, NULL
, OH_READ
,
3437 &recov_state
, NULL
);
3441 /* putfh target fh */
3442 argop
[0].argop
= OP_CPUTFH
;
3443 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= rp
->r_fh
;
3446 argop
[1].argop
= OP_READ
;
3447 rargs
= &argop
[1].nfs_argop4_u
.opread
;
3448 rargs
->stateid
= nfs4_get_stateid(cr
, rp
, curproc
->p_pidp
->pid_id
, mi
,
3449 OP_READ
, &sid_types
, async
);
3452 if (mi
->mi_io_kstats
) {
3453 mutex_enter(&mi
->mi_lock
);
3454 kstat_runq_enter(KSTAT_IO_PTR(mi
->mi_io_kstats
));
3455 mutex_exit(&mi
->mi_lock
);
3458 NFS4_DEBUG(nfs4_client_call_debug
, (CE_NOTE
,
3459 "nfs4read: %s call, rp %s",
3460 needrecov
? "recov" : "first",
3463 if ((vp
->v_flag
& VNOCACHE
) ||
3464 (rp
->r_flags
& R4DIRECTIO
) ||
3465 (mi
->mi_flags
& MI4_DIRECTIO
))
3466 tsize
= MIN(mi
->mi_tsize
, count
);
3468 tsize
= MIN(mi
->mi_curread
, count
);
3470 rargs
->offset
= (offset4
)offset
;
3471 rargs
->count
= (count4
)tsize
;
3472 rargs
->res_data_val_alt
= NULL
;
3473 rargs
->res_mblk
= NULL
;
3474 rargs
->res_uiop
= NULL
;
3475 rargs
->res_maxsize
= 0;
3476 rargs
->wlist
= NULL
;
3479 rargs
->res_uiop
= uiop
;
3481 rargs
->res_data_val_alt
= base
;
3482 rargs
->res_maxsize
= tsize
;
3484 rfs4call(mi
, &args
, &res
, cr
, &doqueue
, 0, &e
);
3486 if (nfs4read_error_inject
) {
3487 res
.status
= nfs4read_error_inject
;
3488 nfs4read_error_inject
= 0;
3492 if (mi
->mi_io_kstats
) {
3493 mutex_enter(&mi
->mi_lock
);
3494 kstat_runq_exit(KSTAT_IO_PTR(mi
->mi_io_kstats
));
3495 mutex_exit(&mi
->mi_lock
);
3498 needrecov
= nfs4_needs_recovery(&e
, FALSE
, mi
->mi_vfsp
);
3499 if (e
.error
!= 0 && !needrecov
) {
3500 nfs4_end_fop(mi
, vp
, NULL
, OH_READ
,
3501 &recov_state
, needrecov
);
3506 * Do proper retry for OLD and BAD stateid errors outside
3507 * of the normal recovery framework. There are two differences
3508 * between async and sync reads. The first is that we allow
3509 * retry on BAD_STATEID for async reads, but not sync reads.
3510 * The second is that we mark the file dead for a failed
3511 * attempt with a special stateid for sync reads, but just
3512 * return EIO for async reads.
3514 * If a sync read receives a BAD stateid error while using a
3515 * delegation stateid, retry using the open stateid (if it
3516 * exists). If it doesn't have an open stateid, reopen the
3517 * file first, then retry.
3519 if (e
.error
== 0 && (res
.status
== NFS4ERR_OLD_STATEID
||
3520 res
.status
== NFS4ERR_BAD_STATEID
) && async
) {
3521 nfs4_end_fop(mi
, vp
, NULL
, OH_READ
,
3522 &recov_state
, needrecov
);
3523 if (sid_types
.cur_sid_type
== SPEC_SID
) {
3524 (void) xdr_free(xdr_COMPOUND4res_clnt
,
3528 nfs4_save_stateid(&rargs
->stateid
, &sid_types
);
3529 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
3531 } else if (e
.error
== 0 && res
.status
== NFS4ERR_OLD_STATEID
&&
3532 !async
&& sid_types
.cur_sid_type
!= SPEC_SID
) {
3533 nfs4_save_stateid(&rargs
->stateid
, &sid_types
);
3534 nfs4_end_fop(mi
, vp
, NULL
, OH_READ
,
3535 &recov_state
, needrecov
);
3536 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
3538 } else if (e
.error
== 0 && res
.status
== NFS4ERR_BAD_STATEID
&&
3539 sid_types
.cur_sid_type
== DEL_SID
) {
3540 nfs4_save_stateid(&rargs
->stateid
, &sid_types
);
3541 mutex_enter(&rp
->r_statev4_lock
);
3542 rp
->r_deleg_return_pending
= TRUE
;
3543 mutex_exit(&rp
->r_statev4_lock
);
3544 if (nfs4rdwr_check_osid(vp
, &e
, cr
)) {
3545 nfs4_end_fop(mi
, vp
, NULL
, OH_READ
,
3546 &recov_state
, needrecov
);
3547 (void) xdr_free(xdr_COMPOUND4res_clnt
,
3551 nfs4_end_fop(mi
, vp
, NULL
, OH_READ
,
3552 &recov_state
, needrecov
);
3553 /* hold needed for nfs4delegreturn_thread */
3555 nfs4delegreturn_async(rp
, (NFS4_DR_PUSH
|NFS4_DR_REOPEN
|
3556 NFS4_DR_DISCARD
), FALSE
);
3557 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
3563 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
,
3564 "nfs4read: initiating recovery\n"));
3565 abort
= nfs4_start_recovery(&e
,
3566 mi
, vp
, NULL
, &rargs
->stateid
,
3567 NULL
, OP_READ
, NULL
, NULL
, NULL
);
3568 nfs4_end_fop(mi
, vp
, NULL
, OH_READ
,
3569 &recov_state
, needrecov
);
3571 * Do not retry if we got OLD_STATEID using a special
3572 * stateid. This avoids looping with a broken server.
3574 if (e
.error
== 0 && res
.status
== NFS4ERR_OLD_STATEID
&&
3575 sid_types
.cur_sid_type
== SPEC_SID
)
3578 if (abort
== FALSE
) {
3580 * Need to retry all possible stateids in
3581 * case the recovery error wasn't stateid
3582 * related or the stateids have become
3583 * stale (server reboot).
3585 nfs4_init_stateid_types(&sid_types
);
3586 (void) xdr_free(xdr_COMPOUND4res_clnt
,
3592 e
.error
= geterrno4(res
.status
);
3593 (void) xdr_free(xdr_COMPOUND4res_clnt
,
3600 e
.error
= geterrno4(res
.status
);
3601 nfs4_end_fop(mi
, vp
, NULL
, OH_READ
,
3602 &recov_state
, needrecov
);
3603 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
3607 data_len
= res
.array
[1].nfs_resop4_u
.opread
.data_len
;
3612 if (mi
->mi_io_kstats
) {
3613 mutex_enter(&mi
->mi_lock
);
3614 KSTAT_IO_PTR(mi
->mi_io_kstats
)->reads
++;
3615 KSTAT_IO_PTR(mi
->mi_io_kstats
)->nread
+= data_len
;
3616 mutex_exit(&mi
->mi_lock
);
3618 lwp_stat_update(LWP_STAT_INBLK
, 1);
3619 is_eof
= res
.array
[1].nfs_resop4_u
.opread
.eof
;
3620 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
3622 } while (count
&& !is_eof
);
3626 nfs4_end_fop(mi
, vp
, NULL
, OH_READ
, &recov_state
, needrecov
);
3633 nfs4_ioctl(vnode_t
*vp
, int cmd
, intptr_t arg
, int flag
, cred_t
*cr
, int *rvalp
,
3634 caller_context_t
*ct
)
3636 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
3640 return (nfs4_directio(vp
, (int)arg
, cr
));
3648 nfs4_getattr(vnode_t
*vp
, struct vattr
*vap
, int flags
, cred_t
*cr
,
3649 caller_context_t
*ct
)
3652 rnode4_t
*rp
= VTOR4(vp
);
3654 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
3657 * If it has been specified that the return value will
3658 * just be used as a hint, and we are only being asked
3659 * for size, fsid or rdevid, then return the client's
3660 * notion of these values without checking to make sure
3661 * that the attribute cache is up to date.
3662 * The whole point is to avoid an over the wire GETATTR
3665 if (flags
& ATTR_HINT
) {
3666 if (!(vap
->va_mask
& ~(AT_SIZE
| AT_FSID
| AT_RDEV
))) {
3667 mutex_enter(&rp
->r_statelock
);
3668 if (vap
->va_mask
& AT_SIZE
)
3669 vap
->va_size
= rp
->r_size
;
3670 if (vap
->va_mask
& AT_FSID
)
3671 vap
->va_fsid
= rp
->r_attr
.va_fsid
;
3672 if (vap
->va_mask
& AT_RDEV
)
3673 vap
->va_rdev
= rp
->r_attr
.va_rdev
;
3674 mutex_exit(&rp
->r_statelock
);
3680 * Only need to flush pages if asking for the mtime
3681 * and if there any dirty pages or any outstanding
3682 * asynchronous (write) requests for this file.
3684 if (vap
->va_mask
& AT_MTIME
) {
3686 if (nfs4_has_pages(vp
)) {
3687 mutex_enter(&rp
->r_statev4_lock
);
3688 if (rp
->r_deleg_type
!= OPEN_DELEGATE_WRITE
) {
3689 mutex_exit(&rp
->r_statev4_lock
);
3690 if (rp
->r_flags
& R4DIRTY
||
3691 rp
->r_awcount
> 0) {
3692 mutex_enter(&rp
->r_statelock
);
3694 mutex_exit(&rp
->r_statelock
);
3698 mutex_enter(&rp
->r_statelock
);
3699 if (error
&& (error
== ENOSPC
||
3702 rp
->r_error
= error
;
3704 if (--rp
->r_gcount
== 0)
3705 cv_broadcast(&rp
->r_cv
);
3706 mutex_exit(&rp
->r_statelock
);
3709 mutex_exit(&rp
->r_statev4_lock
);
3713 return (nfs4getattr(vp
, vap
, cr
));
3717 nfs4_compare_modes(mode_t from_server
, mode_t on_client
)
3720 * If these are the only two bits cleared
3721 * on the server then return 0 (OK) else
3724 on_client
&= ~(S_ISUID
|S_ISGID
);
3725 if (on_client
== from_server
)
3733 nfs4_setattr(vnode_t
*vp
, struct vattr
*vap
, int flags
, cred_t
*cr
,
3734 caller_context_t
*ct
)
3738 if (vap
->va_mask
& AT_NOSET
)
3741 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
3745 * Don't call secpolicy_vnode_setattr, the client cannot
3746 * use its cached attributes to make security decisions
3747 * as the server may be faking mode bits or mapping uid/gid.
3748 * Always just let the server to the checking.
3749 * If we provide the ability to remove basic priviledges
3750 * to setattr (e.g. basic without chmod) then we will
3751 * need to add a check here before calling the server.
3753 error
= nfs4setattr(vp
, vap
, flags
, cr
, NULL
);
3755 if (error
== 0 && (vap
->va_mask
& AT_SIZE
) && vap
->va_size
== 0)
3756 vnevent_truncate(vp
, ct
);
3762 * To replace the "guarded" version 3 setattr, we use two types of compound
3764 * 1. The "normal" setattr, used when the size of the file isn't being
3765 * changed - { Putfh <fh>; Setattr; Getattr }/
3766 * 2. If the size is changed, precede Setattr with: Getattr; Verify
3767 * with only ctime as the argument. If the server ctime differs from
3768 * what is cached on the client, the verify will fail, but we would
3769 * already have the ctime from the preceding getattr, so just set it
3770 * and retry. Thus the compound here is - { Putfh <fh>; Getattr; Verify;
3771 * Setattr; Getattr }.
3773 * The vsecattr_t * input parameter will be non-NULL if ACLs are being set in
3774 * this setattr and NULL if they are not.
3777 nfs4setattr(vnode_t
*vp
, struct vattr
*vap
, int flags
, cred_t
*cr
,
3780 COMPOUND4args_clnt args
;
3781 COMPOUND4res_clnt res
, *resp
= NULL
;
3782 nfs4_ga_res_t
*garp
= NULL
;
3783 int numops
= 3; /* { Putfh; Setattr; Getattr } */
3784 nfs_argop4 argop
[5];
3785 int verify_argop
= -1;
3786 int setattr_argop
= 1;
3791 uint_t mask
= vap
->va_mask
;
3795 bool_t needrecov
= FALSE
;
3796 nfs4_recov_state_t recov_state
;
3797 nfs4_stateid_types_t sid_types
;
3800 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
3804 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
3806 nfs4_init_stateid_types(&sid_types
);
3809 * Only need to flush pages if there are any pages and
3810 * if the file is marked as dirty in some fashion. The
3811 * file must be flushed so that we can accurately
3812 * determine the size of the file and the cached data
3813 * after the SETATTR returns. A file is considered to
3814 * be dirty if it is either marked with R4DIRTY, has
3815 * outstanding i/o's active, or is mmap'd. In this
3816 * last case, we can't tell whether there are dirty
3817 * pages, so we flush just to be sure.
3819 if (nfs4_has_pages(vp
) &&
3820 ((rp
->r_flags
& R4DIRTY
) ||
3822 rp
->r_mapcnt
> 0)) {
3823 ASSERT(vp
->v_type
!= VCHR
);
3824 e
.error
= nfs4_putpage(vp
, (offset_t
)0, 0, 0, cr
, NULL
);
3825 if (e
.error
&& (e
.error
== ENOSPC
|| e
.error
== EDQUOT
)) {
3826 mutex_enter(&rp
->r_statelock
);
3828 rp
->r_error
= e
.error
;
3829 mutex_exit(&rp
->r_statelock
);
3833 if (mask
& AT_SIZE
) {
3835 * Verification setattr compound for non-deleg AT_SIZE:
3836 * { Putfh; Getattr; Verify; Setattr; Getattr }
3837 * Set ctime local here (outside the do_again label)
3838 * so that subsequent retries (after failed VERIFY)
3839 * will use ctime from GETATTR results (from failed
3840 * verify compound) as VERIFY arg.
3841 * If file has delegation, then VERIFY(time_metadata)
3842 * is of little added value, so don't bother.
3844 mutex_enter(&rp
->r_statev4_lock
);
3845 if (rp
->r_deleg_type
== OPEN_DELEGATE_NONE
||
3846 rp
->r_deleg_return_pending
) {
3848 ctime
= rp
->r_attr
.va_ctime
;
3850 mutex_exit(&rp
->r_statev4_lock
);
3853 recov_state
.rs_flags
= 0;
3854 recov_state
.rs_num_retry_despite_err
= 0;
3856 args
.ctag
= TAG_SETATTR
;
3859 setattr_argop
= numops
- 2;
3862 args
.array_len
= numops
;
3864 e
.error
= nfs4_start_op(VTOMI4(vp
), vp
, NULL
, &recov_state
);
3869 /* putfh target fh */
3870 argop
[0].argop
= OP_CPUTFH
;
3871 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= rp
->r_fh
;
3875 * We only care about the ctime, but need to get mtime
3876 * and size for proper cache update.
3879 argop
[1].argop
= OP_GETATTR
;
3880 argop
[1].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
3881 argop
[1].nfs_argop4_u
.opgetattr
.mi
= VTOMI4(vp
);
3883 /* verify - set later in loop */
3889 (void) nfs_rw_enter_sig(&svp
->sv_lock
, RW_READER
, 0);
3890 supp_attrs
= svp
->sv_supp_attrs
;
3891 nfs_rw_exit(&svp
->sv_lock
);
3893 nfs4args_setattr(&argop
[setattr_argop
], vap
, vsap
, flags
, rp
, cr
,
3894 supp_attrs
, &e
.error
, &sid_types
);
3895 stateid
= argop
[setattr_argop
].nfs_argop4_u
.opsetattr
.stateid
;
3897 /* req time field(s) overflow - return immediately */
3898 nfs4_end_op(VTOMI4(vp
), vp
, NULL
, &recov_state
, needrecov
);
3899 nfs4_fattr4_free(&argop
[setattr_argop
].nfs_argop4_u
.
3900 opsetattr
.obj_attributes
);
3903 omode
= rp
->r_attr
.va_mode
;
3906 argop
[numops
-1].argop
= OP_GETATTR
;
3907 argop
[numops
-1].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
3909 * If we are setting the ACL (indicated only by vsap != NULL), request
3910 * the ACL in this getattr. The ACL returned from this getattr will be
3911 * used in updating the ACL cache.
3914 argop
[numops
-1].nfs_argop4_u
.opgetattr
.attr_request
|=
3916 argop
[numops
-1].nfs_argop4_u
.opgetattr
.mi
= VTOMI4(vp
);
3919 * setattr iterates if the object size is set and the cached ctime
3920 * does not match the file ctime. In that case, verify the ctime first.
3924 if (verify_argop
!= -1) {
3926 * Verify that the ctime match before doing setattr.
3928 va
.va_mask
= AT_CTIME
;
3929 va
.va_ctime
= ctime
;
3931 (void) nfs_rw_enter_sig(&svp
->sv_lock
, RW_READER
, 0);
3932 supp_attrs
= svp
->sv_supp_attrs
;
3933 nfs_rw_exit(&svp
->sv_lock
);
3934 e
.error
= nfs4args_verify(&argop
[verify_argop
], &va
,
3935 OP_VERIFY
, supp_attrs
);
3937 /* req time field(s) overflow - return */
3938 nfs4_end_op(VTOMI4(vp
), vp
, NULL
, &recov_state
,
3948 rfs4call(VTOMI4(vp
), &args
, &res
, cr
, &doqueue
, 0, &e
);
3951 * Purge the access cache and ACL cache if changing either the
3952 * owner of the file, the group owner, or the mode. These may
3953 * change the access permissions of the file, so purge old
3954 * information and start over again.
3956 if (mask
& (AT_UID
| AT_GID
| AT_MODE
)) {
3957 (void) nfs4_access_purge_rp(rp
);
3958 if (rp
->r_secattr
!= NULL
) {
3959 mutex_enter(&rp
->r_statelock
);
3960 vsp
= rp
->r_secattr
;
3961 rp
->r_secattr
= NULL
;
3962 mutex_exit(&rp
->r_statelock
);
3964 nfs4_acl_free_cache(vsp
);
3969 * If res.array_len == numops, then everything succeeded,
3970 * except for possibly the final getattr. If only the
3971 * last getattr failed, give up, and don't try recovery.
3973 if (res
.array_len
== numops
) {
3974 nfs4_end_op(VTOMI4(vp
), vp
, NULL
, &recov_state
,
3982 * if either rpc call failed or completely succeeded - done
3984 needrecov
= nfs4_needs_recovery(&e
, FALSE
, vp
->v_vfsp
);
3986 PURGE_ATTRCACHE4(vp
);
3988 nfs4_end_op(VTOMI4(vp
), vp
, NULL
, &recov_state
,
3995 * Do proper retry for OLD_STATEID outside of the normal
3996 * recovery framework.
3998 if (e
.error
== 0 && res
.status
== NFS4ERR_OLD_STATEID
&&
3999 sid_types
.cur_sid_type
!= SPEC_SID
&&
4000 sid_types
.cur_sid_type
!= NO_SID
) {
4001 nfs4_end_op(VTOMI4(vp
), vp
, NULL
, &recov_state
,
4003 nfs4_save_stateid(&stateid
, &sid_types
);
4004 nfs4_fattr4_free(&argop
[setattr_argop
].nfs_argop4_u
.
4005 opsetattr
.obj_attributes
);
4006 if (verify_argop
!= -1) {
4007 nfs4args_verify_free(&argop
[verify_argop
]);
4010 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
4017 abort
= nfs4_start_recovery(&e
,
4018 VTOMI4(vp
), vp
, NULL
, NULL
, NULL
,
4019 OP_SETATTR
, NULL
, NULL
, NULL
);
4020 nfs4_end_op(VTOMI4(vp
), vp
, NULL
, &recov_state
,
4023 * Do not retry if we failed with OLD_STATEID using
4024 * a special stateid. This is done to avoid looping
4025 * with a broken server.
4027 if (e
.error
== 0 && res
.status
== NFS4ERR_OLD_STATEID
&&
4028 (sid_types
.cur_sid_type
== SPEC_SID
||
4029 sid_types
.cur_sid_type
== NO_SID
))
4032 if (res
.status
== NFS4ERR_BADOWNER
)
4033 nfs4_log_badowner(VTOMI4(vp
),
4036 e
.error
= geterrno4(res
.status
);
4037 (void) xdr_free(xdr_COMPOUND4res_clnt
,
4040 nfs4_fattr4_free(&argop
[setattr_argop
].nfs_argop4_u
.
4041 opsetattr
.obj_attributes
);
4042 if (verify_argop
!= -1) {
4043 nfs4args_verify_free(&argop
[verify_argop
]);
4046 if (abort
== FALSE
) {
4048 * Need to retry all possible stateids in
4049 * case the recovery error wasn't stateid
4050 * related or the stateids have become
4051 * stale (server reboot).
4053 nfs4_init_stateid_types(&sid_types
);
4060 * Need to call nfs4_end_op before nfs4getattr to
4061 * avoid potential nfs4_start_op deadlock. See RFE
4062 * 4777612. Calls to nfs4_invalidate_pages() and
4063 * nfs4_purge_stale_fh() might also generate over the
4064 * wire calls which my cause nfs4_start_op() deadlock.
4066 nfs4_end_op(VTOMI4(vp
), vp
, NULL
, &recov_state
, needrecov
);
4069 * Check to update lease.
4072 if (res
.status
== NFS4_OK
) {
4077 * Check if verify failed to see if try again
4079 if ((verify_argop
== -1) || (res
.array_len
!= 3)) {
4083 if (res
.status
== NFS4ERR_BADOWNER
)
4084 nfs4_log_badowner(VTOMI4(vp
), OP_SETATTR
);
4086 e
.error
= geterrno4(res
.status
);
4089 * When the verify request fails, the client ctime is
4090 * not in sync with the server. This is the same as
4091 * the version 3 "not synchronized" error, and we
4092 * handle it in a similar manner (XXX do we need to???).
4093 * Use the ctime returned in the first getattr for
4094 * the input to the next verify.
4095 * If we couldn't get the attributes, then we give up
4096 * because we can't complete the operation as required.
4098 garp
= &res
.array
[1].nfs_resop4_u
.opgetattr
.ga_res
;
4101 PURGE_ATTRCACHE4(vp
);
4102 nfs4_purge_stale_fh(e
.error
, vp
, cr
);
4105 * retry with a new verify value
4107 ctime
= garp
->n4g_va
.va_ctime
;
4108 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
4112 nfs4_fattr4_free(&argop
[setattr_argop
].nfs_argop4_u
.
4113 opsetattr
.obj_attributes
);
4114 if (verify_argop
!= -1) {
4115 nfs4args_verify_free(&argop
[verify_argop
]);
4118 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
4125 * If we are here, rfs4call has an irrecoverable error - return
4127 nfs4_fattr4_free(&argop
[setattr_argop
].nfs_argop4_u
.
4128 opsetattr
.obj_attributes
);
4129 if (verify_argop
!= -1) {
4130 nfs4args_verify_free(&argop
[verify_argop
]);
4134 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)resp
);
4141 * If changing the size of the file, invalidate
4142 * any local cached data which is no longer part
4143 * of the file. We also possibly invalidate the
4144 * last page in the file. We could use
4145 * pvn_vpzero(), but this would mark the page as
4146 * modified and require it to be written back to
4147 * the server for no particularly good reason.
4148 * This way, if we access it, then we bring it
4149 * back in. A read should be cheaper than a
4152 if (mask
& AT_SIZE
) {
4153 nfs4_invalidate_pages(vp
, (vap
->va_size
& PAGEMASK
), cr
);
4156 /* either no error or one of the postop getattr failed */
4159 * XXX Perform a simplified version of wcc checking. Instead of
4160 * have another getattr to get pre-op, just purge cache if
4161 * any of the ops prior to and including the getattr failed.
4162 * If the getattr succeeded then update the attrcache accordingly.
4166 if (res
.status
== NFS4_OK
) {
4170 resop
= &res
.array
[numops
- 1];
4171 garp
= &resop
->nfs_resop4_u
.opgetattr
.ga_res
;
4174 * In certain cases, nfs4_update_attrcache() will purge the attrcache,
4175 * rather than filling it. See the function itself for details.
4177 e
.error
= nfs4_update_attrcache(res
.status
, garp
, t
, vp
, cr
);
4179 if (garp
->n4g_resbmap
& FATTR4_ACL_MASK
) {
4180 nfs4_acl_fill_cache(rp
, &garp
->n4g_vsa
);
4181 vs_ace4_destroy(&garp
->n4g_vsa
);
4185 * The ACL was supposed to be set and to be
4186 * returned in the last getattr of this
4187 * compound, but for some reason the getattr
4188 * result doesn't contain the ACL. In this
4189 * case, purge the ACL cache.
4191 if (rp
->r_secattr
!= NULL
) {
4192 mutex_enter(&rp
->r_statelock
);
4193 vsp
= rp
->r_secattr
;
4194 rp
->r_secattr
= NULL
;
4195 mutex_exit(&rp
->r_statelock
);
4197 nfs4_acl_free_cache(vsp
);
4203 if (res
.status
== NFS4_OK
&& (mask
& AT_SIZE
)) {
4205 * Set the size, rather than relying on getting it updated
4206 * via a GETATTR. With delegations the client tries to
4207 * suppress GETATTR calls.
4209 mutex_enter(&rp
->r_statelock
);
4210 rp
->r_size
= vap
->va_size
;
4211 mutex_exit(&rp
->r_statelock
);
4215 * Can free up request args and res
4217 nfs4_fattr4_free(&argop
[setattr_argop
].nfs_argop4_u
.
4218 opsetattr
.obj_attributes
);
4219 if (verify_argop
!= -1) {
4220 nfs4args_verify_free(&argop
[verify_argop
]);
4223 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
4226 * Some servers will change the mode to clear the setuid
4227 * and setgid bits when changing the uid or gid. The
4228 * client needs to compensate appropriately.
4230 if (mask
& (AT_UID
| AT_GID
)) {
4231 int terror
, do_setattr
;
4234 va
.va_mask
= AT_MODE
;
4235 terror
= nfs4getattr(vp
, &va
, cr
);
4237 (((mask
& AT_MODE
) && va
.va_mode
!= vap
->va_mode
) ||
4238 (!(mask
& AT_MODE
) && va
.va_mode
!= omode
))) {
4239 va
.va_mask
= AT_MODE
;
4240 if (mask
& AT_MODE
) {
4242 * We asked the mode to be changed and what
4243 * we just got from the server in getattr is
4244 * not what we wanted it to be, so set it now.
4246 va
.va_mode
= vap
->va_mode
;
4250 * We did not ask the mode to be changed,
4251 * Check to see that the server just cleared
4252 * I_SUID and I_GUID from it. If not then
4253 * set mode to omode with UID/GID cleared.
4255 if (nfs4_compare_modes(va
.va_mode
, omode
)) {
4256 omode
&= ~(S_ISUID
|S_ISGID
);
4263 (void) nfs4setattr(vp
, &va
, 0, cr
, NULL
);
4272 nfs4_access(vnode_t
*vp
, int mode
, int flags
, cred_t
*cr
, caller_context_t
*ct
)
4274 COMPOUND4args_clnt args
;
4275 COMPOUND4res_clnt res
;
4277 uint32_t acc
, resacc
, argacc
;
4279 cred_t
*cred
, *ncr
, *ncrfree
= NULL
;
4280 nfs4_access_type_t cacc
;
4282 nfs_argop4 argop
[3];
4284 bool_t needrecov
= FALSE
, do_getattr
;
4285 nfs4_recov_state_t recov_state
;
4288 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
4289 mntinfo4_t
*mi
= VTOMI4(vp
);
4291 if (nfs_zone() != mi
->mi_zone
)
4296 acc
|= ACCESS4_READ
;
4297 if (mode
& VWRITE
) {
4298 if ((vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
) && !ISVDEV(vp
->v_type
))
4300 if (vp
->v_type
== VDIR
)
4301 acc
|= ACCESS4_DELETE
;
4302 acc
|= ACCESS4_MODIFY
| ACCESS4_EXTEND
;
4305 if (vp
->v_type
== VDIR
)
4306 acc
|= ACCESS4_LOOKUP
;
4308 acc
|= ACCESS4_EXECUTE
;
4311 if (VTOR4(vp
)->r_acache
!= NULL
) {
4312 e
.error
= nfs4_validate_caches(vp
, cr
);
4318 if (vp
->v_type
== VDIR
)
4319 argacc
= ACCESS4_READ
| ACCESS4_DELETE
| ACCESS4_MODIFY
|
4320 ACCESS4_EXTEND
| ACCESS4_LOOKUP
;
4322 argacc
= ACCESS4_READ
| ACCESS4_MODIFY
| ACCESS4_EXTEND
|
4324 recov_state
.rs_flags
= 0;
4325 recov_state
.rs_num_retry_despite_err
= 0;
4329 * ncr and ncrfree both initially
4330 * point to the memory area returned
4332 * ncrfree not NULL when exiting means
4333 * that we need to release it
4335 ncr
= crnetadjust(cred
);
4339 cacc
= nfs4_access_check(rp
, acc
, cred
);
4340 if (cacc
== NFS4_ACCESS_ALLOWED
) {
4341 if (ncrfree
!= NULL
)
4345 if (cacc
== NFS4_ACCESS_DENIED
) {
4347 * If the cred can be adjusted, try again
4348 * with the new cred.
4355 if (ncrfree
!= NULL
)
4362 * Don't take with r_statev4_lock here. r_deleg_type could
4363 * change as soon as lock is released. Since it is an int,
4364 * there is no atomicity issue.
4366 do_getattr
= (rp
->r_deleg_type
== OPEN_DELEGATE_NONE
);
4367 num_ops
= do_getattr
? 3 : 2;
4369 args
.ctag
= TAG_ACCESS
;
4371 args
.array_len
= num_ops
;
4374 if (e
.error
= nfs4_start_fop(mi
, vp
, NULL
, OH_ACCESS
,
4375 &recov_state
, NULL
)) {
4376 if (ncrfree
!= NULL
)
4381 /* putfh target fh */
4382 argop
[0].argop
= OP_CPUTFH
;
4383 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= VTOR4(vp
)->r_fh
;
4386 argop
[1].argop
= OP_ACCESS
;
4387 argop
[1].nfs_argop4_u
.opaccess
.access
= argacc
;
4391 argop
[2].argop
= OP_GETATTR
;
4392 argop
[2].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
4393 argop
[2].nfs_argop4_u
.opgetattr
.mi
= mi
;
4396 NFS4_DEBUG(nfs4_client_call_debug
, (CE_NOTE
,
4397 "nfs4_access: %s call, rp %s", needrecov
? "recov" : "first",
4398 rnode4info(VTOR4(vp
))));
4402 rfs4call(VTOMI4(vp
), &args
, &res
, cred
, &doqueue
, 0, &e
);
4403 rpc_error
= e
.error
;
4405 needrecov
= nfs4_needs_recovery(&e
, FALSE
, vp
->v_vfsp
);
4407 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
,
4408 "nfs4_access: initiating recovery\n"));
4410 if (nfs4_start_recovery(&e
, VTOMI4(vp
), vp
, NULL
, NULL
,
4411 NULL
, OP_ACCESS
, NULL
, NULL
, NULL
) == FALSE
) {
4412 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_ACCESS
,
4413 &recov_state
, needrecov
);
4415 (void) xdr_free(xdr_COMPOUND4res_clnt
,
4420 nfs4_end_fop(mi
, vp
, NULL
, OH_ACCESS
, &recov_state
, needrecov
);
4426 e
.error
= geterrno4(res
.status
);
4428 * This might generate over the wire calls throught
4429 * nfs4_invalidate_pages. Hence we need to call nfs4_end_op()
4430 * here to avoid a deadlock.
4432 nfs4_purge_stale_fh(e
.error
, vp
, cr
);
4435 resop
= &res
.array
[1]; /* access res */
4437 resacc
= resop
->nfs_resop4_u
.opaccess
.access
;
4440 resop
++; /* getattr res */
4441 nfs4_attr_cache(vp
, &resop
->nfs_resop4_u
.opgetattr
.ga_res
,
4442 t
, cr
, FALSE
, NULL
);
4446 nfs4_access_cache(rp
, argacc
, resacc
, cred
);
4448 * we just cached results with cred; if cred is the
4449 * adjusted credentials from crnetadjust, we do not want
4450 * to release them before exiting: hence setting ncrfree
4455 /* XXX check the supported bits too? */
4456 if ((acc
& resacc
) != acc
) {
4458 * The following code implements the semantic
4459 * that a setuid root program has *at least* the
4460 * permissions of the user that is running the
4461 * program. See rfs3call() for more portions
4462 * of the implementation of this functionality.
4466 (void) xdr_free(xdr_COMPOUND4res_clnt
,
4478 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
4480 if (ncrfree
!= NULL
)
4488 nfs4_readlink(vnode_t
*vp
, struct uio
*uiop
, cred_t
*cr
, caller_context_t
*ct
)
4490 COMPOUND4args_clnt args
;
4491 COMPOUND4res_clnt res
;
4494 nfs_argop4 argop
[3];
4496 READLINK4res
*lr_res
;
4497 nfs4_ga_res_t
*garp
;
4500 bool_t needrecov
= FALSE
;
4501 nfs4_recov_state_t recov_state
;
4503 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
4505 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
4508 * Can't readlink anything other than a symbolic link.
4510 if (vp
->v_type
!= VLNK
)
4514 if (nfs4_do_symlink_cache
&& rp
->r_symlink
.contents
!= NULL
) {
4515 e
.error
= nfs4_validate_caches(vp
, cr
);
4518 mutex_enter(&rp
->r_statelock
);
4519 if (rp
->r_symlink
.contents
!= NULL
) {
4520 e
.error
= uiomove(rp
->r_symlink
.contents
,
4521 rp
->r_symlink
.len
, UIO_READ
, uiop
);
4522 mutex_exit(&rp
->r_statelock
);
4525 mutex_exit(&rp
->r_statelock
);
4527 recov_state
.rs_flags
= 0;
4528 recov_state
.rs_num_retry_despite_err
= 0;
4533 args
.ctag
= TAG_READLINK
;
4535 e
.error
= nfs4_start_op(VTOMI4(vp
), vp
, NULL
, &recov_state
);
4540 /* 0. putfh symlink fh */
4541 argop
[0].argop
= OP_CPUTFH
;
4542 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= VTOR4(vp
)->r_fh
;
4545 argop
[1].argop
= OP_READLINK
;
4548 argop
[2].argop
= OP_GETATTR
;
4549 argop
[2].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
4550 argop
[2].nfs_argop4_u
.opgetattr
.mi
= VTOMI4(vp
);
4554 NFS4_DEBUG(nfs4_client_call_debug
, (CE_NOTE
,
4555 "nfs4_readlink: %s call, rp %s", needrecov
? "recov" : "first",
4556 rnode4info(VTOR4(vp
))));
4560 rfs4call(VTOMI4(vp
), &args
, &res
, cr
, &doqueue
, 0, &e
);
4562 needrecov
= nfs4_needs_recovery(&e
, FALSE
, vp
->v_vfsp
);
4564 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
,
4565 "nfs4_readlink: initiating recovery\n"));
4567 if (nfs4_start_recovery(&e
, VTOMI4(vp
), vp
, NULL
, NULL
,
4568 NULL
, OP_READLINK
, NULL
, NULL
, NULL
) == FALSE
) {
4570 (void) xdr_free(xdr_COMPOUND4res_clnt
,
4573 nfs4_end_op(VTOMI4(vp
), vp
, NULL
, &recov_state
,
4579 nfs4_end_op(VTOMI4(vp
), vp
, NULL
, &recov_state
, needrecov
);
4585 * There is an path in the code below which calls
4586 * nfs4_purge_stale_fh(), which may generate otw calls through
4587 * nfs4_invalidate_pages. Hence we need to call nfs4_end_op()
4588 * here to avoid nfs4_start_op() deadlock.
4591 if (res
.status
&& (res
.array_len
< args
.array_len
)) {
4593 * either Putfh or Link failed
4595 e
.error
= geterrno4(res
.status
);
4596 nfs4_purge_stale_fh(e
.error
, vp
, cr
);
4597 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
4601 resop
= &res
.array
[1]; /* readlink res */
4602 lr_res
= &resop
->nfs_resop4_u
.opreadlink
;
4605 * treat symlink names as data
4607 linkdata
= utf8_to_str((utf8string
*)&lr_res
->link
, &len
, NULL
);
4608 if (linkdata
!= NULL
) {
4609 int uio_len
= len
- 1;
4610 /* len includes null byte, which we won't uiomove */
4611 e
.error
= uiomove(linkdata
, uio_len
, UIO_READ
, uiop
);
4612 if (nfs4_do_symlink_cache
&& rp
->r_symlink
.contents
== NULL
) {
4613 mutex_enter(&rp
->r_statelock
);
4614 if (rp
->r_symlink
.contents
== NULL
) {
4615 rp
->r_symlink
.contents
= linkdata
;
4616 rp
->r_symlink
.len
= uio_len
;
4617 rp
->r_symlink
.size
= len
;
4618 mutex_exit(&rp
->r_statelock
);
4620 mutex_exit(&rp
->r_statelock
);
4621 kmem_free(linkdata
, len
);
4624 kmem_free(linkdata
, len
);
4627 if (res
.status
== NFS4_OK
) {
4628 resop
++; /* getattr res */
4629 garp
= &resop
->nfs_resop4_u
.opgetattr
.ga_res
;
4631 e
.error
= nfs4_update_attrcache(res
.status
, garp
, t
, vp
, cr
);
4633 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
4636 * The over the wire error for attempting to readlink something
4637 * other than a symbolic link is ENXIO. However, we need to
4638 * return EINVAL instead of ENXIO, so we map it here.
4640 return (e
.error
== ENXIO
? EINVAL
: e
.error
);
4644 * Flush local dirty pages to stable storage on the server.
4646 * If FNODSYNC is specified, then there is nothing to do because
4647 * metadata changes are not cached on the client before being
4648 * sent to the server.
4652 nfs4_fsync(vnode_t
*vp
, int syncflag
, cred_t
*cr
, caller_context_t
*ct
)
4656 if ((syncflag
& FNODSYNC
) || IS_SWAPVP(vp
))
4658 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
4660 error
= nfs4_putpage_commit(vp
, (offset_t
)0, 0, cr
);
4662 error
= VTOR4(vp
)->r_error
;
4667 * Weirdness: if the file was removed or the target of a rename
4668 * operation while it was open, it got renamed instead. Here we
4669 * remove the renamed file.
4673 nfs4_inactive(vnode_t
*vp
, cred_t
*cr
, caller_context_t
*ct
)
4677 ASSERT(vp
!= DNLC_NO_VNODE
);
4681 if (IS_SHADOW(vp
, rp
)) {
4687 * If this is coming from the wrong zone, we let someone in the right
4688 * zone take care of it asynchronously. We can get here due to
4689 * VN_RELE() being called from pageout() or fsflush(). This call may
4690 * potentially turn into an expensive no-op if, for instance, v_count
4691 * gets incremented in the meantime, but it's still correct.
4693 if (nfs_zone() != VTOMI4(vp
)->mi_zone
) {
4694 nfs4_async_inactive(vp
, cr
);
4699 * Some of the cleanup steps might require over-the-wire
4700 * operations. Since fop_inactive can get called as a result of
4701 * other over-the-wire operations (e.g., an attribute cache update
4702 * can lead to a DNLC purge), doing those steps now would lead to a
4703 * nested call to the recovery framework, which can deadlock. So
4704 * do any over-the-wire cleanups asynchronously, in a separate
4708 mutex_enter(&rp
->r_os_lock
);
4709 mutex_enter(&rp
->r_statelock
);
4710 mutex_enter(&rp
->r_statev4_lock
);
4712 if (vp
->v_type
== VREG
&& list_head(&rp
->r_open_streams
) != NULL
) {
4713 mutex_exit(&rp
->r_statev4_lock
);
4714 mutex_exit(&rp
->r_statelock
);
4715 mutex_exit(&rp
->r_os_lock
);
4716 nfs4_async_inactive(vp
, cr
);
4720 if (rp
->r_deleg_type
== OPEN_DELEGATE_READ
||
4721 rp
->r_deleg_type
== OPEN_DELEGATE_WRITE
) {
4722 mutex_exit(&rp
->r_statev4_lock
);
4723 mutex_exit(&rp
->r_statelock
);
4724 mutex_exit(&rp
->r_os_lock
);
4725 nfs4_async_inactive(vp
, cr
);
4729 if (rp
->r_unldvp
!= NULL
) {
4730 mutex_exit(&rp
->r_statev4_lock
);
4731 mutex_exit(&rp
->r_statelock
);
4732 mutex_exit(&rp
->r_os_lock
);
4733 nfs4_async_inactive(vp
, cr
);
4736 mutex_exit(&rp
->r_statev4_lock
);
4737 mutex_exit(&rp
->r_statelock
);
4738 mutex_exit(&rp
->r_os_lock
);
4740 rp4_addfree(rp
, cr
);
4744 * nfs4_inactive_otw - nfs4_inactive, plus over-the-wire calls to free up
4745 * various bits of state. The caller must not refer to vp after this call.
4749 nfs4_inactive_otw(vnode_t
*vp
, cred_t
*cr
)
4751 rnode4_t
*rp
= VTOR4(vp
);
4752 nfs4_recov_state_t recov_state
;
4753 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
4757 COMPOUND4args_clnt args
;
4758 COMPOUND4res_clnt res
, *resp
;
4759 nfs_argop4 argop
[2];
4765 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
4766 ASSERT(!IS_SHADOW(vp
, rp
));
4769 name
= fn_name(VTOSV(vp
)->sv_name
);
4770 NFS4_DEBUG(nfs4_client_inactive_debug
, (CE_NOTE
, "nfs4_inactive_otw: "
4771 "release vnode %s", name
));
4772 kmem_free(name
, MAXNAMELEN
);
4775 if (vp
->v_type
== VREG
) {
4776 bool_t recov_failed
= FALSE
;
4778 e
.error
= nfs4close_all(vp
, cr
);
4780 /* Check to see if recovery failed */
4781 mutex_enter(&(VTOMI4(vp
)->mi_lock
));
4782 if (VTOMI4(vp
)->mi_flags
& MI4_RECOV_FAIL
)
4783 recov_failed
= TRUE
;
4784 mutex_exit(&(VTOMI4(vp
)->mi_lock
));
4785 if (!recov_failed
) {
4786 mutex_enter(&rp
->r_statelock
);
4787 if (rp
->r_flags
& R4RECOVERR
)
4788 recov_failed
= TRUE
;
4789 mutex_exit(&rp
->r_statelock
);
4792 NFS4_DEBUG(nfs4_client_recov_debug
,
4793 (CE_NOTE
, "nfs4_inactive_otw: "
4794 "close failed (recovery failure)"));
4800 if (rp
->r_unldvp
== NULL
) {
4801 rp4_addfree(rp
, cr
);
4806 * Save the vnode pointer for the directory where the
4807 * unlinked-open file got renamed, then set it to NULL
4808 * to prevent another thread from getting here before
4809 * we're done with the remove. While we have the
4810 * statelock, make local copies of the pertinent rnode
4811 * fields. If we weren't to do this in an atomic way, the
4812 * the unl* fields could become inconsistent with respect
4813 * to each other due to a race condition between this
4814 * code and nfs_remove(). See bug report 1034328.
4816 mutex_enter(&rp
->r_statelock
);
4817 if (rp
->r_unldvp
== NULL
) {
4818 mutex_exit(&rp
->r_statelock
);
4819 rp4_addfree(rp
, cr
);
4823 unldvp
= rp
->r_unldvp
;
4824 rp
->r_unldvp
= NULL
;
4825 unlname
= rp
->r_unlname
;
4826 rp
->r_unlname
= NULL
;
4827 unlcred
= rp
->r_unlcred
;
4828 rp
->r_unlcred
= NULL
;
4829 mutex_exit(&rp
->r_statelock
);
4832 * If there are any dirty pages left, then flush
4833 * them. This is unfortunate because they just
4834 * may get thrown away during the remove operation,
4835 * but we have to do this for correctness.
4837 if (nfs4_has_pages(vp
) &&
4838 ((rp
->r_flags
& R4DIRTY
) || rp
->r_count
> 0)) {
4839 ASSERT(vp
->v_type
!= VCHR
);
4840 e
.error
= nfs4_putpage(vp
, 0, 0, 0, cr
, NULL
);
4842 mutex_enter(&rp
->r_statelock
);
4844 rp
->r_error
= e
.error
;
4845 mutex_exit(&rp
->r_statelock
);
4849 recov_state
.rs_flags
= 0;
4850 recov_state
.rs_num_retry_despite_err
= 0;
4853 * Do the remove operation on the renamed file
4855 args
.ctag
= TAG_INACTIVE
;
4858 * Remove ops: putfh dir; remove
4863 e
.error
= nfs4_start_op(VTOMI4(unldvp
), unldvp
, NULL
, &recov_state
);
4865 kmem_free(unlname
, MAXNAMELEN
);
4869 * Try again; this time around r_unldvp will be NULL, so we'll
4870 * just call rp4_addfree() and return.
4875 /* putfh directory */
4876 argop
[0].argop
= OP_CPUTFH
;
4877 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= VTOR4(unldvp
)->r_fh
;
4880 argop
[1].argop
= OP_CREMOVE
;
4881 argop
[1].nfs_argop4_u
.opcremove
.ctarget
= unlname
;
4888 * Can't do this yet. We may be being called from
4889 * dnlc_purge_XXX while that routine is holding a
4890 * mutex lock to the nc_rele list. The calls to
4891 * nfs3_cache_wcc_data may result in calls to
4892 * dnlc_purge_XXX. This will result in a deadlock.
4894 rfs4call(VTOMI4(unldvp
), &args
, &res
, unlcred
, &doqueue
, 0, &e
);
4896 PURGE_ATTRCACHE4(unldvp
);
4898 } else if (res
.status
) {
4899 e
.error
= geterrno4(res
.status
);
4900 PURGE_ATTRCACHE4(unldvp
);
4902 * This code is inactive right now
4903 * but if made active there should
4904 * be a nfs4_end_op() call before
4905 * nfs4_purge_stale_fh to avoid start_op()
4906 * deadlock. See BugId: 4948726
4908 nfs4_purge_stale_fh(error
, unldvp
, cr
);
4913 resop
= &res
.array
[1];
4914 rm_res
= &resop
->nfs_resop4_u
.opremove
;
4916 * Update directory cache attribute,
4917 * readdir and dnlc caches.
4919 nfs4_update_dircaches(&rm_res
->cinfo
, unldvp
, NULL
, NULL
, NULL
);
4922 rfs4call(VTOMI4(unldvp
), &args
, &res
, unlcred
, &doqueue
, 0, &e
);
4924 PURGE_ATTRCACHE4(unldvp
);
4927 if (nfs4_needs_recovery(&e
, FALSE
, unldvp
->v_vfsp
)) {
4928 if (nfs4_start_recovery(&e
, VTOMI4(unldvp
), unldvp
, NULL
,
4929 NULL
, NULL
, OP_REMOVE
, NULL
, NULL
, NULL
) == FALSE
) {
4931 (void) xdr_free(xdr_COMPOUND4res_clnt
,
4933 nfs4_end_op(VTOMI4(unldvp
), unldvp
, NULL
,
4934 &recov_state
, TRUE
);
4935 goto recov_retry_remove
;
4938 nfs4_end_op(VTOMI4(unldvp
), unldvp
, NULL
, &recov_state
, FALSE
);
4941 * Release stuff held for the remove
4944 if (!e
.error
&& resp
)
4945 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)resp
);
4947 kmem_free(unlname
, MAXNAMELEN
);
4953 * Remote file system operations having to do with directory manipulation.
4957 nfs4_lookup(vnode_t
*dvp
, char *nm
, vnode_t
**vpp
, struct pathname
*pnp
,
4958 int flags
, vnode_t
*rdir
, cred_t
*cr
, caller_context_t
*ct
,
4959 int *direntflags
, pathname_t
*realpnp
)
4962 vnode_t
*vp
, *avp
= NULL
;
4966 if (nfs_zone() != VTOMI4(dvp
)->mi_zone
)
4969 * if LOOKUP_XATTR, must replace dvp (object) with
4970 * object's attrdir before continuing with lookup
4972 if (flags
& LOOKUP_XATTR
) {
4973 error
= nfs4lookup_xattr(dvp
, nm
, &avp
, flags
, cr
);
4980 * If lookup is for "", just return dvp now. The attrdir
4981 * has already been activated (from nfs4lookup_xattr), and
4982 * the caller will RELE the original dvp -- not
4983 * the attrdir. So, set vpp and return.
4984 * Currently, when the LOOKUP_XATTR flag is
4985 * passed to fop_lookup, the name is always empty, and
4986 * shortcircuiting here avoids 3 unneeded lock/unlock
4989 * If a non-empty name was provided, then it is the
4990 * attribute name, and it will be looked up below.
4998 * The vfs layer never sends a name when asking for the
4999 * attrdir, so we should never get here (unless of course
5000 * name is passed at some time in future -- at which time
5001 * we'll blow up here).
5007 if (nfs_rw_enter_sig(&drp
->r_rwlock
, RW_READER
, INTR4(dvp
)))
5010 error
= nfs4lookup(dvp
, nm
, vpp
, cr
, 0);
5011 nfs_rw_exit(&drp
->r_rwlock
);
5014 * If vnode is a device, create special vnode.
5016 if (!error
&& ISVDEV((*vpp
)->v_type
)) {
5018 *vpp
= specvp(vp
, vp
->v_rdev
, vp
->v_type
, cr
);
5027 nfs4lookup_xattr(vnode_t
*dvp
, char *nm
, vnode_t
**vpp
, int flags
, cred_t
*cr
)
5031 int cflag
= ((flags
& CREATE_XATTR_DIR
) != 0);
5035 if (!(mi
->mi_vfsp
->vfs_flag
& VFS_XATTR
) &&
5036 !vfs_has_feature(mi
->mi_vfsp
, VFSFT_SYSATTR_VIEWS
))
5040 if (nfs_rw_enter_sig(&drp
->r_rwlock
, RW_READER
, INTR4(dvp
)))
5043 mutex_enter(&drp
->r_statelock
);
5045 * If the server doesn't support xattrs just return EINVAL
5047 if (drp
->r_xattr_dir
== NFS4_XATTR_DIR_NOTSUPP
) {
5048 mutex_exit(&drp
->r_statelock
);
5049 nfs_rw_exit(&drp
->r_rwlock
);
5054 * If there is a cached xattr directory entry,
5055 * use it as long as the attributes are valid. If the
5056 * attributes are not valid, take the simple approach and
5057 * free the cached value and re-fetch a new value.
5059 * We don't negative entry cache for now, if we did we
5060 * would need to check if the file has changed on every
5061 * lookup. But xattrs don't exist very often and failing
5062 * an openattr is not much more expensive than and NVERIFY or GETATTR
5063 * so do an openattr over the wire for now.
5065 if (drp
->r_xattr_dir
!= NULL
) {
5066 if (ATTRCACHE4_VALID(dvp
)) {
5067 VN_HOLD(drp
->r_xattr_dir
);
5068 *vpp
= drp
->r_xattr_dir
;
5069 mutex_exit(&drp
->r_statelock
);
5070 nfs_rw_exit(&drp
->r_rwlock
);
5073 VN_RELE(drp
->r_xattr_dir
);
5074 drp
->r_xattr_dir
= NULL
;
5076 mutex_exit(&drp
->r_statelock
);
5078 error
= nfs4openattr(dvp
, vpp
, cflag
, cr
);
5080 nfs_rw_exit(&drp
->r_rwlock
);
5086 nfs4lookup(vnode_t
*dvp
, char *nm
, vnode_t
**vpp
, cred_t
*cr
, int skipdnlc
)
5091 ASSERT(nfs_zone() == VTOMI4(dvp
)->mi_zone
);
5094 * If lookup is for "", just return dvp. Don't need
5095 * to send it over the wire, look it up in the dnlc,
5096 * or perform any access checks.
5105 * Can't do lookups in non-directories.
5107 if (dvp
->v_type
!= VDIR
)
5111 * If lookup is for ".", just return dvp. Don't need
5112 * to send it over the wire or look it up in the dnlc,
5113 * just need to check access.
5115 if (nm
[0] == '.' && nm
[1] == '\0') {
5116 error
= nfs4_access(dvp
, VEXEC
, 0, cr
, NULL
);
5125 if (!(drp
->r_flags
& R4LOOKUP
)) {
5126 mutex_enter(&drp
->r_statelock
);
5127 drp
->r_flags
|= R4LOOKUP
;
5128 mutex_exit(&drp
->r_statelock
);
5133 * Lookup this name in the DNLC. If there is no entry
5134 * lookup over the wire.
5137 *vpp
= dnlc_lookup(dvp
, nm
);
5140 * We need to go over the wire to lookup the name.
5142 return (nfs4lookupnew_otw(dvp
, nm
, vpp
, cr
));
5146 * We hit on the dnlc
5148 if (*vpp
!= DNLC_NO_VNODE
||
5149 (dvp
->v_vfsp
->vfs_flag
& VFS_RDONLY
)) {
5151 * But our attrs may not be valid.
5153 if (ATTRCACHE4_VALID(dvp
)) {
5154 error
= nfs4_waitfor_purge_complete(dvp
);
5162 * If after the purge completes, check to make sure
5163 * our attrs are still valid.
5165 if (ATTRCACHE4_VALID(dvp
)) {
5167 * If we waited for a purge we may have
5168 * lost our vnode so look it up again.
5171 *vpp
= dnlc_lookup(dvp
, nm
);
5173 return (nfs4lookupnew_otw(dvp
,
5177 * The access cache should almost always hit
5179 error
= nfs4_access(dvp
, VEXEC
, 0, cr
, NULL
);
5186 if (*vpp
== DNLC_NO_VNODE
) {
5196 ASSERT(*vpp
!= NULL
);
5199 * We may have gotten here we have one of the following cases:
5200 * 1) vpp != DNLC_NO_VNODE, our attrs have timed out so we
5201 * need to validate them.
5202 * 2) vpp == DNLC_NO_VNODE, a negative entry that we always
5205 * Go to the server and check if the directory has changed, if
5206 * it hasn't we are done and can use the dnlc entry.
5208 return (nfs4lookupvalidate_otw(dvp
, nm
, vpp
, cr
));
5212 * Go to the server and check if the directory has changed, if
5213 * it hasn't we are done and can use the dnlc entry. If it
5214 * has changed we get a new copy of its attributes and check
5215 * the access for VEXEC, then relookup the filename and
5216 * get its filehandle and attributes.
5218 * PUTFH dfh NVERIFY GETATTR ACCESS LOOKUP GETFH GETATTR
5219 * if the NVERIFY failed we must
5221 * cache new attributes (will set r_time_attr_inval)
5223 * recheck VEXEC access
5224 * add name to dnlc, possibly negative
5225 * if LOOKUP succeeded
5226 * cache new attributes
5228 * set a new r_time_attr_inval for dvp
5229 * check to make sure we have access
5231 * The vpp returned is the vnode passed in if the directory is valid,
5232 * a new vnode if successful lookup, or NULL on error.
5235 nfs4lookupvalidate_otw(vnode_t
*dvp
, char *nm
, vnode_t
**vpp
, cred_t
*cr
)
5237 COMPOUND4args_clnt args
;
5238 COMPOUND4res_clnt res
;
5240 fattr4_change dchange
;
5242 int argoplist_size
= 7 * sizeof (nfs_argop4
);
5246 nfs4_recov_state_t recov_state
;
5251 nfs4_sharedfh_t
*sfhp
;
5252 nfs4_access_type_t cacc
;
5254 rnode4_t
*drp
= VTOR4(dvp
);
5255 nfs4_ga_res_t
*garp
= NULL
;
5256 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
5258 ASSERT(nfs_zone() == VTOMI4(dvp
)->mi_zone
);
5260 ASSERT(nm
[0] != '\0');
5261 ASSERT(dvp
->v_type
== VDIR
);
5262 ASSERT(nm
[0] != '.' || nm
[1] != '\0');
5263 ASSERT(*vpp
!= NULL
);
5265 if (nm
[0] == '.' && nm
[1] == '.' && nm
[2] == '\0') {
5267 args
.ctag
= TAG_LOOKUP_VPARENT
;
5270 * If dvp were a stub, it should have triggered and caused
5271 * a mount for us to get this far.
5273 ASSERT(!RP_ISSTUB(VTOR4(dvp
)));
5276 args
.ctag
= TAG_LOOKUP_VALID
;
5280 recov_state
.rs_flags
= 0;
5281 recov_state
.rs_num_retry_despite_err
= 0;
5285 /* Save the original mount point security information */
5286 (void) save_mnt_secinfo(mi
->mi_curr_serv
);
5289 e
.error
= nfs4_start_fop(mi
, dvp
, NULL
, OH_LOOKUP
,
5290 &recov_state
, NULL
);
5292 (void) check_mnt_secinfo(mi
->mi_curr_serv
, nvp
);
5298 argop
= kmem_alloc(argoplist_size
, KM_SLEEP
);
5300 /* PUTFH dfh NVERIFY GETATTR ACCESS LOOKUP GETFH GETATTR */
5305 argop
[0].argop
= OP_CPUTFH
;
5306 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= VTOR4(dvp
)->r_fh
;
5308 /* 1. nverify the change info */
5309 argop
[1].argop
= OP_NVERIFY
;
5310 ver_fattr
= &argop
[1].nfs_argop4_u
.opnverify
.obj_attributes
;
5311 ver_fattr
->attrmask
= FATTR4_CHANGE_MASK
;
5312 ver_fattr
->attrlist4
= (char *)&dchange
;
5313 ptr
= (int32_t *)&dchange
;
5314 IXDR_PUT_HYPER(ptr
, VTOR4(dvp
)->r_change
);
5315 ver_fattr
->attrlist4_len
= sizeof (fattr4_change
);
5317 /* 2. getattr directory */
5318 argop
[2].argop
= OP_GETATTR
;
5319 argop
[2].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
5320 argop
[2].nfs_argop4_u
.opgetattr
.mi
= VTOMI4(dvp
);
5322 /* 3. access directory */
5323 argop
[3].argop
= OP_ACCESS
;
5324 argop
[3].nfs_argop4_u
.opaccess
.access
= ACCESS4_READ
| ACCESS4_DELETE
|
5325 ACCESS4_MODIFY
| ACCESS4_EXTEND
| ACCESS4_LOOKUP
;
5327 /* 4. lookup name */
5329 argop
[4].argop
= OP_LOOKUPP
;
5331 argop
[4].argop
= OP_CLOOKUP
;
5332 argop
[4].nfs_argop4_u
.opclookup
.cname
= nm
;
5335 /* 5. resulting file handle */
5336 argop
[5].argop
= OP_GETFH
;
5338 /* 6. resulting file attributes */
5339 argop
[6].argop
= OP_GETATTR
;
5340 argop
[6].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
5341 argop
[6].nfs_argop4_u
.opgetattr
.mi
= VTOMI4(dvp
);
5346 rfs4call(VTOMI4(dvp
), &args
, &res
, cr
, &doqueue
, 0, &e
);
5348 if (!isdotdot
&& res
.status
== NFS4ERR_MOVED
) {
5349 e
.error
= nfs4_setup_referral(dvp
, nm
, vpp
, cr
);
5350 if (e
.error
!= 0 && *vpp
!= NULL
)
5352 nfs4_end_fop(mi
, dvp
, NULL
, OH_LOOKUP
,
5353 &recov_state
, FALSE
);
5354 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
5355 kmem_free(argop
, argoplist_size
);
5359 if (nfs4_needs_recovery(&e
, FALSE
, dvp
->v_vfsp
)) {
5361 * For WRONGSEC of a non-dotdot case, send secinfo directly
5362 * from this thread, do not go thru the recovery thread since
5363 * we need the nm information.
5365 * Not doing dotdot case because there is no specification
5366 * for (PUTFH, SECINFO "..") yet.
5368 if (!isdotdot
&& res
.status
== NFS4ERR_WRONGSEC
) {
5369 if ((e
.error
= nfs4_secinfo_vnode_otw(dvp
, nm
, cr
)))
5370 nfs4_end_fop(mi
, dvp
, NULL
, OH_LOOKUP
,
5371 &recov_state
, FALSE
);
5373 nfs4_end_fop(mi
, dvp
, NULL
, OH_LOOKUP
,
5374 &recov_state
, TRUE
);
5375 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
5376 kmem_free(argop
, argoplist_size
);
5379 (void) check_mnt_secinfo(mi
->mi_curr_serv
, nvp
);
5385 if (nfs4_start_recovery(&e
, mi
, dvp
, NULL
, NULL
, NULL
,
5386 OP_LOOKUP
, NULL
, NULL
, NULL
) == FALSE
) {
5387 nfs4_end_fop(mi
, dvp
, NULL
, OH_LOOKUP
,
5388 &recov_state
, TRUE
);
5390 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
5391 kmem_free(argop
, argoplist_size
);
5396 nfs4_end_fop(mi
, dvp
, NULL
, OH_LOOKUP
, &recov_state
, FALSE
);
5398 if (e
.error
|| res
.array_len
== 0) {
5400 * If e.error isn't set, then reply has no ops (or we couldn't
5401 * be here). The only legal way to reply without an op array
5402 * is via NFS4ERR_MINOR_VERS_MISMATCH. An ops array should
5403 * be in the reply for all other status values.
5405 * For valid replies without an ops array, return ENOTSUP
5406 * (geterrno4 xlation of VERS_MISMATCH). For illegal replies,
5407 * return EIO -- don't trust status.
5410 e
.error
= (res
.status
== NFS4ERR_MINOR_VERS_MISMATCH
) ?
5414 kmem_free(argop
, argoplist_size
);
5415 (void) check_mnt_secinfo(mi
->mi_curr_serv
, nvp
);
5419 if (res
.status
!= NFS4ERR_SAME
) {
5420 e
.error
= geterrno4(res
.status
);
5423 * The NVERIFY "failed" so the directory has changed
5424 * First make sure PUTFH succeeded and NVERIFY "failed"
5427 if ((res
.array
[0].nfs_resop4_u
.opputfh
.status
!= NFS4_OK
) ||
5428 (res
.array
[1].nfs_resop4_u
.opnverify
.status
!= NFS4_OK
)) {
5429 nfs4_purge_stale_fh(e
.error
, dvp
, cr
);
5436 * We know the NVERIFY "failed" so we must:
5437 * purge the caches (access and indirectly dnlc if needed)
5439 nfs4_purge_caches(dvp
, NFS4_NOPURGE_DNLC
, cr
, TRUE
);
5441 if (res
.array
[2].nfs_resop4_u
.opgetattr
.status
!= NFS4_OK
) {
5442 nfs4_purge_stale_fh(e
.error
, dvp
, cr
);
5449 * Install new cached attributes for the directory
5451 nfs4_attr_cache(dvp
,
5452 &res
.array
[2].nfs_resop4_u
.opgetattr
.ga_res
,
5453 t
, cr
, FALSE
, NULL
);
5455 if (res
.array
[3].nfs_resop4_u
.opaccess
.status
!= NFS4_OK
) {
5456 nfs4_purge_stale_fh(e
.error
, dvp
, cr
);
5459 e
.error
= geterrno4(res
.status
);
5464 * Now we know the directory is valid,
5465 * cache new directory access
5467 nfs4_access_cache(drp
,
5468 args
.array
[3].nfs_argop4_u
.opaccess
.access
,
5469 res
.array
[3].nfs_resop4_u
.opaccess
.access
, cr
);
5472 * recheck VEXEC access
5474 cacc
= nfs4_access_check(drp
, ACCESS4_LOOKUP
, cr
);
5475 if (cacc
!= NFS4_ACCESS_ALLOWED
) {
5477 * Directory permissions might have been revoked
5479 if (cacc
== NFS4_ACCESS_DENIED
) {
5487 * Somehow we must not have asked for enough
5488 * so try a singleton ACCESS, should never happen.
5490 e
.error
= nfs4_access(dvp
, VEXEC
, 0, cr
, NULL
);
5498 e
.error
= geterrno4(res
.status
);
5499 if (res
.array
[4].nfs_resop4_u
.oplookup
.status
!= NFS4_OK
) {
5501 * The lookup failed, probably no entry
5503 if (e
.error
== ENOENT
&& nfs4_lookup_neg_cache
) {
5504 dnlc_update(dvp
, nm
, DNLC_NO_VNODE
);
5507 * Might be some other error, so remove
5508 * the dnlc entry to make sure we start all
5509 * over again, next time.
5511 dnlc_remove(dvp
, nm
);
5518 if (res
.array
[5].nfs_resop4_u
.opgetfh
.status
!= NFS4_OK
) {
5520 * The file exists but we can't get its fh for
5521 * some unknown reason. Remove it from the dnlc
5522 * and error out to be safe.
5524 dnlc_remove(dvp
, nm
);
5529 fhp
= &res
.array
[5].nfs_resop4_u
.opgetfh
.object
;
5530 if (fhp
->nfs_fh4_len
== 0) {
5532 * The file exists but a bogus fh
5533 * some unknown reason. Remove it from the dnlc
5534 * and error out to be safe.
5537 dnlc_remove(dvp
, nm
);
5542 sfhp
= sfh4_get(fhp
, mi
);
5544 if (res
.array
[6].nfs_resop4_u
.opgetattr
.status
== NFS4_OK
)
5545 garp
= &res
.array
[6].nfs_resop4_u
.opgetattr
.ga_res
;
5548 * Make the new rnode
5551 e
.error
= nfs4_make_dotdot(sfhp
, t
, dvp
, cr
, &nvp
, 1);
5559 * XXX if nfs4_make_dotdot uses an existing rnode
5560 * XXX it doesn't update the attributes.
5561 * XXX for now just save them again to save an OTW
5563 nfs4_attr_cache(nvp
, garp
, t
, cr
, FALSE
, NULL
);
5565 nvp
= makenfs4node(sfhp
, garp
, dvp
->v_vfsp
, t
, cr
,
5566 dvp
, fn_get(VTOSV(dvp
)->sv_name
, nm
, sfhp
));
5568 * If v_type == VNON, then garp was NULL because
5569 * the last op in the compound failed and makenfs4node
5570 * could not find the vnode for sfhp. It created
5571 * a new vnode, so we have nothing to purge here.
5573 if (nvp
->v_type
== VNON
) {
5576 vattr
.va_mask
= AT_TYPE
;
5578 * N.B. We've already called nfs4_end_fop above.
5580 e
.error
= nfs4getattr(nvp
, &vattr
, cr
);
5588 nvp
->v_type
= vattr
.va_type
;
5594 mutex_enter(&nrp
->r_statev4_lock
);
5595 if (!nrp
->created_v4
) {
5596 mutex_exit(&nrp
->r_statev4_lock
);
5597 dnlc_update(dvp
, nm
, nvp
);
5599 mutex_exit(&nrp
->r_statev4_lock
);
5610 * Because the NVERIFY "succeeded" we know that the
5611 * directory attributes are still valid
5612 * so update r_time_attr_inval
5615 mutex_enter(&drp
->r_statelock
);
5616 if (!(mi
->mi_flags
& MI4_NOAC
) && !(dvp
->v_flag
& VNOCACHE
)) {
5617 delta
= now
- drp
->r_time_attr_saved
;
5618 if (delta
< mi
->mi_acdirmin
)
5619 delta
= mi
->mi_acdirmin
;
5620 else if (delta
> mi
->mi_acdirmax
)
5621 delta
= mi
->mi_acdirmax
;
5623 drp
->r_time_attr_inval
= now
+ delta
;
5624 mutex_exit(&drp
->r_statelock
);
5625 dnlc_update(dvp
, nm
, *vpp
);
5628 * Even though we have a valid directory attr cache
5629 * and dnlc entry, we may not have access.
5630 * This should almost always hit the cache.
5632 e
.error
= nfs4_access(dvp
, VEXEC
, 0, cr
, NULL
);
5638 if (*vpp
== DNLC_NO_VNODE
) {
5646 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
5647 kmem_free(argop
, argoplist_size
);
5648 (void) check_mnt_secinfo(mi
->mi_curr_serv
, nvp
);
5653 * We need to go over the wire to lookup the name, but
5654 * while we are there verify the directory has not
5655 * changed but if it has, get new attributes and check access
5657 * PUTFH dfh SAVEFH LOOKUP nm GETFH GETATTR RESTOREFH
5658 * NVERIFY GETATTR ACCESS
5661 * if the NVERIFY failed we must purge the caches, add new attributes,
5662 * and cache new access.
5663 * set a new r_time_attr_inval
5664 * add name to dnlc, possibly negative
5665 * if LOOKUP succeeded
5666 * cache new attributes
5669 nfs4lookupnew_otw(vnode_t
*dvp
, char *nm
, vnode_t
**vpp
, cred_t
*cr
)
5671 COMPOUND4args_clnt args
;
5672 COMPOUND4res_clnt res
;
5674 fattr4_change dchange
;
5676 nfs4_ga_res_t
*garp
= NULL
;
5677 int argoplist_size
= 9 * sizeof (nfs_argop4
);
5681 nfs4_recov_state_t recov_state
;
5686 nfs4_sharedfh_t
*sfhp
;
5687 nfs4_access_type_t cacc
;
5689 rnode4_t
*drp
= VTOR4(dvp
);
5690 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
5692 ASSERT(nfs_zone() == VTOMI4(dvp
)->mi_zone
);
5694 ASSERT(nm
[0] != '\0');
5695 ASSERT(dvp
->v_type
== VDIR
);
5696 ASSERT(nm
[0] != '.' || nm
[1] != '\0');
5697 ASSERT(*vpp
== NULL
);
5699 if (nm
[0] == '.' && nm
[1] == '.' && nm
[2] == '\0') {
5701 args
.ctag
= TAG_LOOKUP_PARENT
;
5704 * If dvp were a stub, it should have triggered and caused
5705 * a mount for us to get this far.
5707 ASSERT(!RP_ISSTUB(VTOR4(dvp
)));
5710 args
.ctag
= TAG_LOOKUP
;
5714 recov_state
.rs_flags
= 0;
5715 recov_state
.rs_num_retry_despite_err
= 0;
5719 /* Save the original mount point security information */
5720 (void) save_mnt_secinfo(mi
->mi_curr_serv
);
5723 e
.error
= nfs4_start_fop(mi
, dvp
, NULL
, OH_LOOKUP
,
5724 &recov_state
, NULL
);
5726 (void) check_mnt_secinfo(mi
->mi_curr_serv
, nvp
);
5730 argop
= kmem_alloc(argoplist_size
, KM_SLEEP
);
5732 /* PUTFH SAVEFH LOOKUP GETFH GETATTR RESTOREFH NVERIFY GETATTR ACCESS */
5737 argop
[0].argop
= OP_CPUTFH
;
5738 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= VTOR4(dvp
)->r_fh
;
5740 /* 1. savefh for the nverify */
5741 argop
[1].argop
= OP_SAVEFH
;
5743 /* 2. lookup name */
5745 argop
[2].argop
= OP_LOOKUPP
;
5747 argop
[2].argop
= OP_CLOOKUP
;
5748 argop
[2].nfs_argop4_u
.opclookup
.cname
= nm
;
5751 /* 3. resulting file handle */
5752 argop
[3].argop
= OP_GETFH
;
5754 /* 4. resulting file attributes */
5755 argop
[4].argop
= OP_GETATTR
;
5756 argop
[4].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
5757 argop
[4].nfs_argop4_u
.opgetattr
.mi
= VTOMI4(dvp
);
5759 /* 5. restorefh back the directory for the nverify */
5760 argop
[5].argop
= OP_RESTOREFH
;
5762 /* 6. nverify the change info */
5763 argop
[6].argop
= OP_NVERIFY
;
5764 ver_fattr
= &argop
[6].nfs_argop4_u
.opnverify
.obj_attributes
;
5765 ver_fattr
->attrmask
= FATTR4_CHANGE_MASK
;
5766 ver_fattr
->attrlist4
= (char *)&dchange
;
5767 ptr
= (int32_t *)&dchange
;
5768 IXDR_PUT_HYPER(ptr
, VTOR4(dvp
)->r_change
);
5769 ver_fattr
->attrlist4_len
= sizeof (fattr4_change
);
5771 /* 7. getattr directory */
5772 argop
[7].argop
= OP_GETATTR
;
5773 argop
[7].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
5774 argop
[7].nfs_argop4_u
.opgetattr
.mi
= VTOMI4(dvp
);
5776 /* 8. access directory */
5777 argop
[8].argop
= OP_ACCESS
;
5778 argop
[8].nfs_argop4_u
.opaccess
.access
= ACCESS4_READ
| ACCESS4_DELETE
|
5779 ACCESS4_MODIFY
| ACCESS4_EXTEND
| ACCESS4_LOOKUP
;
5784 rfs4call(VTOMI4(dvp
), &args
, &res
, cr
, &doqueue
, 0, &e
);
5786 if (!isdotdot
&& res
.status
== NFS4ERR_MOVED
) {
5787 e
.error
= nfs4_setup_referral(dvp
, nm
, vpp
, cr
);
5788 if (e
.error
!= 0 && *vpp
!= NULL
)
5790 nfs4_end_fop(mi
, dvp
, NULL
, OH_LOOKUP
,
5791 &recov_state
, FALSE
);
5792 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
5793 kmem_free(argop
, argoplist_size
);
5797 if (nfs4_needs_recovery(&e
, FALSE
, dvp
->v_vfsp
)) {
5799 * For WRONGSEC of a non-dotdot case, send secinfo directly
5800 * from this thread, do not go thru the recovery thread since
5801 * we need the nm information.
5803 * Not doing dotdot case because there is no specification
5804 * for (PUTFH, SECINFO "..") yet.
5806 if (!isdotdot
&& res
.status
== NFS4ERR_WRONGSEC
) {
5807 if ((e
.error
= nfs4_secinfo_vnode_otw(dvp
, nm
, cr
)))
5808 nfs4_end_fop(mi
, dvp
, NULL
, OH_LOOKUP
,
5809 &recov_state
, FALSE
);
5811 nfs4_end_fop(mi
, dvp
, NULL
, OH_LOOKUP
,
5812 &recov_state
, TRUE
);
5813 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
5814 kmem_free(argop
, argoplist_size
);
5817 (void) check_mnt_secinfo(mi
->mi_curr_serv
, nvp
);
5821 if (nfs4_start_recovery(&e
, mi
, dvp
, NULL
, NULL
, NULL
,
5822 OP_LOOKUP
, NULL
, NULL
, NULL
) == FALSE
) {
5823 nfs4_end_fop(mi
, dvp
, NULL
, OH_LOOKUP
,
5824 &recov_state
, TRUE
);
5826 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
5827 kmem_free(argop
, argoplist_size
);
5832 nfs4_end_fop(mi
, dvp
, NULL
, OH_LOOKUP
, &recov_state
, FALSE
);
5834 if (e
.error
|| res
.array_len
== 0) {
5836 * If e.error isn't set, then reply has no ops (or we couldn't
5837 * be here). The only legal way to reply without an op array
5838 * is via NFS4ERR_MINOR_VERS_MISMATCH. An ops array should
5839 * be in the reply for all other status values.
5841 * For valid replies without an ops array, return ENOTSUP
5842 * (geterrno4 xlation of VERS_MISMATCH). For illegal replies,
5843 * return EIO -- don't trust status.
5846 e
.error
= (res
.status
== NFS4ERR_MINOR_VERS_MISMATCH
) ?
5849 kmem_free(argop
, argoplist_size
);
5850 (void) check_mnt_secinfo(mi
->mi_curr_serv
, nvp
);
5854 e
.error
= geterrno4(res
.status
);
5857 * The PUTFH and SAVEFH may have failed.
5859 if ((res
.array
[0].nfs_resop4_u
.opputfh
.status
!= NFS4_OK
) ||
5860 (res
.array
[1].nfs_resop4_u
.opsavefh
.status
!= NFS4_OK
)) {
5861 nfs4_purge_stale_fh(e
.error
, dvp
, cr
);
5866 * Check if the file exists, if it does delay entering
5867 * into the dnlc until after we update the directory
5868 * attributes so we don't cause it to get purged immediately.
5870 if (res
.array
[2].nfs_resop4_u
.oplookup
.status
!= NFS4_OK
) {
5872 * The lookup failed, probably no entry
5874 if (e
.error
== ENOENT
&& nfs4_lookup_neg_cache
)
5875 dnlc_update(dvp
, nm
, DNLC_NO_VNODE
);
5879 if (res
.array
[3].nfs_resop4_u
.opgetfh
.status
!= NFS4_OK
) {
5881 * The file exists but we can't get its fh for
5882 * some unknown reason. Error out to be safe.
5887 fhp
= &res
.array
[3].nfs_resop4_u
.opgetfh
.object
;
5888 if (fhp
->nfs_fh4_len
== 0) {
5890 * The file exists but a bogus fh
5891 * some unknown reason. Error out to be safe.
5896 sfhp
= sfh4_get(fhp
, mi
);
5898 if (res
.array
[4].nfs_resop4_u
.opgetattr
.status
!= NFS4_OK
) {
5902 garp
= &res
.array
[4].nfs_resop4_u
.opgetattr
.ga_res
;
5905 * The RESTOREFH may have failed
5907 if (res
.array
[5].nfs_resop4_u
.oprestorefh
.status
!= NFS4_OK
) {
5913 if (res
.array
[6].nfs_resop4_u
.opnverify
.status
!= NFS4ERR_SAME
) {
5915 * First make sure the NVERIFY failed as we expected,
5916 * if it didn't then be conservative and error out
5917 * as we can't trust the directory.
5919 if (res
.array
[6].nfs_resop4_u
.opnverify
.status
!= NFS4_OK
) {
5926 * We know the NVERIFY "failed" so the directory has changed,
5928 * purge the caches (access and indirectly dnlc if needed)
5930 nfs4_purge_caches(dvp
, NFS4_NOPURGE_DNLC
, cr
, TRUE
);
5932 if (res
.array
[7].nfs_resop4_u
.opgetattr
.status
!= NFS4_OK
) {
5936 nfs4_attr_cache(dvp
,
5937 &res
.array
[7].nfs_resop4_u
.opgetattr
.ga_res
,
5938 t
, cr
, FALSE
, NULL
);
5940 if (res
.array
[8].nfs_resop4_u
.opaccess
.status
!= NFS4_OK
) {
5941 nfs4_purge_stale_fh(e
.error
, dvp
, cr
);
5943 e
.error
= geterrno4(res
.status
);
5948 * Now we know the directory is valid,
5949 * cache new directory access
5951 nfs4_access_cache(drp
,
5952 args
.array
[8].nfs_argop4_u
.opaccess
.access
,
5953 res
.array
[8].nfs_resop4_u
.opaccess
.access
, cr
);
5956 * recheck VEXEC access
5958 cacc
= nfs4_access_check(drp
, ACCESS4_LOOKUP
, cr
);
5959 if (cacc
!= NFS4_ACCESS_ALLOWED
) {
5961 * Directory permissions might have been revoked
5963 if (cacc
== NFS4_ACCESS_DENIED
) {
5970 * Somehow we must not have asked for enough
5971 * so try a singleton ACCESS should never happen
5973 e
.error
= nfs4_access(dvp
, VEXEC
, 0, cr
, NULL
);
5980 e
.error
= geterrno4(res
.status
);
5988 * Because the NVERIFY "succeeded" we know that the
5989 * directory attributes are still valid
5990 * so update r_time_attr_inval
5993 mutex_enter(&drp
->r_statelock
);
5994 if (!(mi
->mi_flags
& MI4_NOAC
) && !(dvp
->v_flag
& VNOCACHE
)) {
5995 delta
= now
- drp
->r_time_attr_saved
;
5996 if (delta
< mi
->mi_acdirmin
)
5997 delta
= mi
->mi_acdirmin
;
5998 else if (delta
> mi
->mi_acdirmax
)
5999 delta
= mi
->mi_acdirmax
;
6001 drp
->r_time_attr_inval
= now
+ delta
;
6002 mutex_exit(&drp
->r_statelock
);
6005 * Even though we have a valid directory attr cache,
6006 * we may not have access.
6007 * This should almost always hit the cache.
6009 e
.error
= nfs4_access(dvp
, VEXEC
, 0, cr
, NULL
);
6017 * Now we have successfully completed the lookup, if the
6018 * directory has changed we now have the valid attributes.
6019 * We also know we have directory access.
6020 * Create the new rnode and insert it in the dnlc.
6023 e
.error
= nfs4_make_dotdot(sfhp
, t
, dvp
, cr
, &nvp
, 1);
6029 * XXX if nfs4_make_dotdot uses an existing rnode
6030 * XXX it doesn't update the attributes.
6031 * XXX for now just save them again to save an OTW
6033 nfs4_attr_cache(nvp
, garp
, t
, cr
, FALSE
, NULL
);
6035 nvp
= makenfs4node(sfhp
, garp
, dvp
->v_vfsp
, t
, cr
,
6036 dvp
, fn_get(VTOSV(dvp
)->sv_name
, nm
, sfhp
));
6041 mutex_enter(&nrp
->r_statev4_lock
);
6042 if (!nrp
->created_v4
) {
6043 mutex_exit(&nrp
->r_statev4_lock
);
6044 dnlc_update(dvp
, nm
, nvp
);
6046 mutex_exit(&nrp
->r_statev4_lock
);
6051 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
6052 kmem_free(argop
, argoplist_size
);
6053 (void) check_mnt_secinfo(mi
->mi_curr_serv
, nvp
);
6059 nfs4lookup_dump_compound(char *where
, nfs_argop4
*argbase
, int argcnt
)
6062 zoneid_t zoneid
= getzoneid();
6065 zcmn_err(zoneid
, CE_NOTE
, "%s: dumping cmpd", where
);
6066 for (i
= 0; i
< argcnt
; i
++) {
6067 nfs_argop4
*op
= &argbase
[i
];
6068 switch (op
->argop
) {
6071 zcmn_err(zoneid
, CE_NOTE
, "\t op %d, putfh", i
);
6074 zcmn_err(zoneid
, CE_NOTE
, "\t op %d, putrootfh", i
);
6077 s
= op
->nfs_argop4_u
.opclookup
.cname
;
6078 zcmn_err(zoneid
, CE_NOTE
, "\t op %d, lookup %s", i
, s
);
6081 s
= utf8_to_str(&op
->nfs_argop4_u
.oplookup
.objname
,
6083 zcmn_err(zoneid
, CE_NOTE
, "\t op %d, lookup %s", i
, s
);
6087 zcmn_err(zoneid
, CE_NOTE
, "\t op %d, lookupp ..", i
);
6090 zcmn_err(zoneid
, CE_NOTE
, "\t op %d, getfh", i
);
6093 zcmn_err(zoneid
, CE_NOTE
, "\t op %d, getattr", i
);
6096 zcmn_err(zoneid
, CE_NOTE
, "\t op %d, openattr", i
);
6099 zcmn_err(zoneid
, CE_NOTE
, "\t op %d, opcode %d", i
,
6108 * nfs4lookup_setup - constructs a multi-lookup compound request.
6110 * Given the path "nm1/nm2/.../nmn", the following compound requests
6113 * Note: Getfh is not be needed because filehandle attr is mandatory, but it
6114 * is faster, for now.
6116 * l4_getattrs indicates the type of compound requested.
6118 * LKP4_NO_ATTRIBUTE - no attributes (used by secinfo):
6120 * compound { Put*fh; Lookup {nm1}; Lookup {nm2}; ... Lookup {nmn} }
6122 * total number of ops is n + 1.
6124 * LKP4_LAST_NAMED_ATTR - multi-component path for a named
6125 * attribute: create lookups plus one OPENATTR/GETFH/GETATTR
6126 * before the last component, and only get attributes
6127 * for the last component. Note that the second-to-last
6128 * pathname component is XATTR_RPATH, which does NOT go
6129 * over-the-wire as a lookup.
6131 * compound { Put*fh; Lookup {nm1}; Lookup {nm2}; ... Lookup {nmn-2};
6132 * Openattr; Getfh; Getattr; Lookup {nmn}; Getfh; Getattr }
6134 * and total number of ops is n + 5.
6136 * LKP4_LAST_ATTRDIR - multi-component path for the hidden named
6137 * attribute directory: create lookups plus an OPENATTR
6138 * replacing the last lookup. Note that the last pathname
6139 * component is XATTR_RPATH, which does NOT go over-the-wire
6142 * compound { Put*fh; Lookup {nm1}; Lookup {nm2}; ... Getfh; Getattr;
6143 * Openattr; Getfh; Getattr }
6145 * and total number of ops is n + 5.
6147 * LKP4_ALL_ATTRIBUTES - create lookups and get attributes for intermediate
6150 * compound { Put*fh; Lookup {nm1}; Getfh; Getattr;
6151 * Lookup {nm2}; ... Lookup {nmn}; Getfh; Getattr }
6153 * and total number of ops is 3*n + 1.
6155 * All cases: returns the index in the arg array of the final LOOKUP op, or
6156 * -1 if no LOOKUPs were used.
6159 nfs4lookup_setup(char *nm
, lookup4_param_t
*lookupargp
, int needgetfh
)
6161 enum lkp4_attr_setup l4_getattrs
= lookupargp
->l4_getattrs
;
6162 nfs_argop4
*argbase
, *argop
;
6164 int n
= 1; /* number of components */
6165 int nga
= 1; /* number of Getattr's in request */
6166 char c
= '\0', *s
, *p
;
6167 int lookup_idx
= -1;
6170 /* set lookuparg response result to 0 */
6171 lookupargp
->resp
->status
= NFS4_OK
;
6173 /* skip leading "/" or "." e.g. ".//./" if there is */
6175 if (*nm
!= '/' && *nm
!= '.')
6178 /* ".." is counted as 1 component */
6179 if (*nm
== '.' && *(nm
+ 1) != '/')
6184 * Find n = number of components - nm must be null terminated
6185 * Skip "." components.
6188 for (n
= 1, s
= nm
; *s
!= '\0'; s
++) {
6189 if ((*s
== '/') && (*(s
+ 1) != '/') &&
6190 (*(s
+ 1) != '\0') &&
6191 !(*(s
+ 1) == '.' && (*(s
+ 2) == '/' ||
6199 * nga is number of components that need Getfh+Getattr
6201 switch (l4_getattrs
) {
6202 case LKP4_NO_ATTRIBUTES
:
6205 case LKP4_ALL_ATTRIBUTES
:
6208 * Always have at least 1 getfh, getattr pair
6213 case LKP4_LAST_ATTRDIR
:
6214 case LKP4_LAST_NAMED_ATTR
:
6220 * If change to use the filehandle attr instead of getfh
6221 * the following line can be deleted.
6226 * calculate number of ops in request as
6227 * header + trailer + lookups + getattrs
6229 arglen
= lookupargp
->header_len
+ lookupargp
->trailer_len
+ n
+ nga
;
6231 argoplist_size
= arglen
* sizeof (nfs_argop4
);
6232 argop
= argbase
= kmem_alloc(argoplist_size
, KM_SLEEP
);
6233 lookupargp
->argsp
->array
= argop
;
6235 argcnt
= lookupargp
->header_len
;
6239 * loop and create a lookup op and possibly getattr/getfh for
6240 * each component. Skip "." components.
6242 for (s
= nm
; *s
!= '\0'; s
= p
) {
6244 * Set up a pathname struct for each component if needed
6251 for (p
= s
; (*p
!= '/') && (*p
!= '\0'); p
++)
6256 if (s
[0] == '.' && s
[1] == '\0') {
6260 if (l4_getattrs
== LKP4_LAST_ATTRDIR
&&
6261 strcmp(s
, XATTR_RPATH
) == 0) {
6262 /* getfh XXX may not be needed in future */
6263 argop
->argop
= OP_GETFH
;
6268 argop
->argop
= OP_GETATTR
;
6269 argop
->nfs_argop4_u
.opgetattr
.attr_request
=
6270 lookupargp
->ga_bits
;
6271 argop
->nfs_argop4_u
.opgetattr
.mi
=
6277 argop
->argop
= OP_OPENATTR
;
6278 } else if (l4_getattrs
== LKP4_LAST_NAMED_ATTR
&&
6279 strcmp(s
, XATTR_RPATH
) == 0) {
6281 argop
->argop
= OP_OPENATTR
;
6285 /* getfh XXX may not be needed in future */
6286 argop
->argop
= OP_GETFH
;
6291 argop
->argop
= OP_GETATTR
;
6292 argop
->nfs_argop4_u
.opgetattr
.attr_request
=
6293 lookupargp
->ga_bits
;
6294 argop
->nfs_argop4_u
.opgetattr
.mi
=
6300 } else if (s
[0] == '.' && s
[1] == '.' && s
[2] == '\0') {
6302 argop
->argop
= OP_LOOKUPP
;
6305 argop
->argop
= OP_LOOKUP
;
6306 (void) str_to_utf8(s
,
6307 &argop
->nfs_argop4_u
.oplookup
.objname
);
6309 lookup_idx
= argcnt
;
6315 if (l4_getattrs
== LKP4_ALL_ATTRIBUTES
) {
6316 /* getfh XXX may not be needed in future */
6317 argop
->argop
= OP_GETFH
;
6322 argop
->argop
= OP_GETATTR
;
6323 argop
->nfs_argop4_u
.opgetattr
.attr_request
=
6324 lookupargp
->ga_bits
;
6325 argop
->nfs_argop4_u
.opgetattr
.mi
=
6332 if ((l4_getattrs
!= LKP4_NO_ATTRIBUTES
) &&
6333 ((l4_getattrs
!= LKP4_ALL_ATTRIBUTES
) || (lookup_idx
< 0))) {
6335 /* stick in a post-lookup getfh */
6336 argop
->argop
= OP_GETFH
;
6340 /* post-lookup getattr */
6341 argop
->argop
= OP_GETATTR
;
6342 argop
->nfs_argop4_u
.opgetattr
.attr_request
=
6343 lookupargp
->ga_bits
;
6344 argop
->nfs_argop4_u
.opgetattr
.mi
= lookupargp
->mi
;
6347 argcnt
+= lookupargp
->trailer_len
; /* actual op count */
6348 lookupargp
->argsp
->array_len
= argcnt
;
6349 lookupargp
->arglen
= arglen
;
6352 if (nfs4_client_lookup_debug
)
6353 nfs4lookup_dump_compound("nfs4lookup_setup", argbase
, argcnt
);
6356 return (lookup_idx
);
6360 nfs4openattr(vnode_t
*dvp
, vnode_t
**avp
, int cflag
, cred_t
*cr
)
6362 COMPOUND4args_clnt args
;
6363 COMPOUND4res_clnt res
;
6364 GETFH4res
*gf_res
= NULL
;
6365 nfs_argop4 argop
[4];
6366 nfs_resop4
*resop
= NULL
;
6367 nfs4_sharedfh_t
*sfhp
;
6375 nfs4_recov_state_t recov_state
;
6377 ASSERT(nfs_zone() == VTOMI4(dvp
)->mi_zone
);
6380 recov_state
.rs_flags
= 0;
6381 recov_state
.rs_num_retry_despite_err
= 0;
6384 /* COMPOUND: putfh, openattr, getfh, getattr */
6387 args
.ctag
= TAG_OPENATTR
;
6389 e
.error
= nfs4_start_op(VTOMI4(dvp
), dvp
, NULL
, &recov_state
);
6396 argop
[0].argop
= OP_CPUTFH
;
6397 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= drp
->r_fh
;
6400 argop
[1].argop
= OP_OPENATTR
;
6401 argop
[1].nfs_argop4_u
.opopenattr
.createdir
= (cflag
? TRUE
: FALSE
);
6404 argop
[2].argop
= OP_GETFH
;
6407 argop
[3].argop
= OP_GETATTR
;
6408 argop
[3].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
6409 argop
[3].nfs_argop4_u
.opgetattr
.mi
= VTOMI4(dvp
);
6411 NFS4_DEBUG(nfs4_client_call_debug
, (CE_NOTE
,
6412 "nfs4openattr: %s call, drp %s", needrecov
? "recov" : "first",
6417 rfs4call(VTOMI4(dvp
), &args
, &res
, cr
, &doqueue
, 0, &e
);
6419 needrecov
= nfs4_needs_recovery(&e
, FALSE
, dvp
->v_vfsp
);
6423 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
,
6424 "nfs4openattr: initiating recovery\n"));
6426 abort
= nfs4_start_recovery(&e
,
6427 VTOMI4(dvp
), dvp
, NULL
, NULL
, NULL
,
6428 OP_OPENATTR
, NULL
, NULL
, NULL
);
6429 nfs4_end_op(VTOMI4(dvp
), dvp
, NULL
, &recov_state
, needrecov
);
6431 e
.error
= geterrno4(res
.status
);
6432 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
6440 nfs4_end_op(VTOMI4(dvp
), dvp
, NULL
, &recov_state
, needrecov
);
6446 * If OTW errro is NOTSUPP, then it should be
6447 * translated to EINVAL. All Solaris file system
6448 * implementations return EINVAL to the syscall layer
6449 * when the attrdir cannot be created due to an
6450 * implementation restriction or noxattr mount option.
6452 if (res
.status
== NFS4ERR_NOTSUPP
) {
6453 mutex_enter(&drp
->r_statelock
);
6454 if (drp
->r_xattr_dir
)
6455 VN_RELE(drp
->r_xattr_dir
);
6456 VN_HOLD(NFS4_XATTR_DIR_NOTSUPP
);
6457 drp
->r_xattr_dir
= NFS4_XATTR_DIR_NOTSUPP
;
6458 mutex_exit(&drp
->r_statelock
);
6462 e
.error
= geterrno4(res
.status
);
6466 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
6467 nfs4_end_op(VTOMI4(dvp
), dvp
, NULL
, &recov_state
,
6473 resop
= &res
.array
[0]; /* putfh res */
6474 ASSERT(resop
->nfs_resop4_u
.opgetfh
.status
== NFS4_OK
);
6476 resop
= &res
.array
[1]; /* openattr res */
6477 ASSERT(resop
->nfs_resop4_u
.opopenattr
.status
== NFS4_OK
);
6479 resop
= &res
.array
[2]; /* getfh res */
6480 gf_res
= &resop
->nfs_resop4_u
.opgetfh
;
6481 if (gf_res
->object
.nfs_fh4_len
== 0) {
6483 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
6484 nfs4_end_op(VTOMI4(dvp
), dvp
, NULL
, &recov_state
, needrecov
);
6488 sfhp
= sfh4_get(&gf_res
->object
, VTOMI4(dvp
));
6489 vp
= makenfs4node(sfhp
, &res
.array
[3].nfs_resop4_u
.opgetattr
.ga_res
,
6490 dvp
->v_vfsp
, t
, cr
, dvp
,
6491 fn_get(VTOSV(dvp
)->sv_name
, XATTR_RPATH
, sfhp
));
6495 PURGE_ATTRCACHE4(vp
);
6497 mutex_enter(&vp
->v_lock
);
6498 vp
->v_flag
|= V_XATTRDIR
;
6499 mutex_exit(&vp
->v_lock
);
6503 mutex_enter(&drp
->r_statelock
);
6504 if (drp
->r_xattr_dir
)
6505 VN_RELE(drp
->r_xattr_dir
);
6507 drp
->r_xattr_dir
= vp
;
6510 * Invalidate pathconf4 cache because r_xattr_dir is no longer
6511 * NULL. xattrs could be created at any time, and we have no
6512 * way to update pc4_xattr_exists in the base object if/when
6515 drp
->r_pathconf
.pc4_xattr_valid
= 0;
6517 mutex_exit(&drp
->r_statelock
);
6519 nfs4_end_op(VTOMI4(dvp
), dvp
, NULL
, &recov_state
, needrecov
);
6521 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
6528 nfs4_create(vnode_t
*dvp
, char *nm
, struct vattr
*va
, enum vcexcl exclusive
,
6529 int mode
, vnode_t
**vpp
, cred_t
*cr
, int flags
, caller_context_t
*ct
,
6538 enum createmode4 createmode
;
6539 bool_t must_trunc
= FALSE
;
6542 if (nfs_zone() != VTOMI4(dvp
)->mi_zone
)
6544 if (exclusive
== EXCL
&& (dvp
->v_flag
& V_XATTRDIR
)) {
6548 /* . and .. have special meaning in the protocol, reject them. */
6550 if (nm
[0] == '.' && (nm
[1] == '\0' || (nm
[1] == '.' && nm
[2] == '\0')))
6555 if (nfs_rw_enter_sig(&drp
->r_rwlock
, RW_WRITER
, INTR4(dvp
)))
6560 * We make a copy of the attributes because the caller does not
6561 * expect us to change what va points to.
6566 * If the pathname is "", then dvp is the root vnode of
6567 * a remote file mounted over a local directory.
6568 * All that needs to be done is access
6569 * checking and truncation. Note that we avoid doing
6570 * open w/ create because the parent directory might
6571 * be in pseudo-fs and the open would fail.
6580 * We need to go over the wire, just to be sure whether the
6581 * file exists or not. Using the DNLC can be dangerous in
6582 * this case when making a decision regarding existence.
6584 error
= nfs4lookup(dvp
, nm
, &vp
, cr
, 1);
6588 createmode
= EXCLUSIVE4
;
6590 createmode
= GUARDED4
;
6593 * error would be set if the file does not exist on the
6594 * server, so lets go create it.
6601 * File does exist on the server
6603 if (exclusive
== EXCL
)
6605 else if (vp
->v_type
== VDIR
&& (mode
& VWRITE
))
6609 * If vnode is a device, create special vnode.
6611 if (ISVDEV(vp
->v_type
)) {
6613 vp
= specvp(vp
, vp
->v_rdev
, vp
->v_type
, cr
);
6616 if (!(error
= fop_access(vp
, mode
, 0, cr
, ct
))) {
6617 if ((vattr
.va_mask
& AT_SIZE
) &&
6618 vp
->v_type
== VREG
) {
6621 * Check here for large file handled
6622 * by LF-unaware process (as
6623 * ufs_create() does)
6625 if (!(flags
& FOFFMAX
)) {
6626 mutex_enter(&rp
->r_statelock
);
6627 if (rp
->r_size
> MAXOFF32_T
)
6629 mutex_exit(&rp
->r_statelock
);
6632 /* if error is set then we need to return */
6634 nfs_rw_exit(&drp
->r_rwlock
);
6640 vattr
.va_mask
= AT_SIZE
;
6641 error
= nfs4setattr(vp
, &vattr
, 0, cr
,
6645 * we know we have a regular file that already
6646 * exists and we may end up truncating the file
6647 * as a result of the open_otw, so flush out
6648 * any dirty pages for this file first.
6650 if (nfs4_has_pages(vp
) &&
6651 ((rp
->r_flags
& R4DIRTY
) ||
6653 rp
->r_mapcnt
> 0)) {
6654 error
= nfs4_putpage(vp
,
6655 (offset_t
)0, 0, 0, cr
, ct
);
6656 if (error
&& (error
== ENOSPC
||
6667 vattr
.va_mask
= (AT_SIZE
|
6669 vattr
.va_type
= VREG
;
6670 createmode
= UNCHECKED4
;
6677 nfs_rw_exit(&drp
->r_rwlock
);
6684 if (vp
->v_type
== VREG
) {
6686 if (IS_SHADOW(vp
, trp
))
6692 * existing file got truncated, notify.
6694 vnevent_create(tvp
, ct
);
6702 dnlc_remove(dvp
, nm
);
6704 ASSERT(vattr
.va_mask
& AT_TYPE
);
6707 * If not a regular file let nfs4mknod() handle it.
6709 if (vattr
.va_type
!= VREG
) {
6710 error
= nfs4mknod(dvp
, nm
, &vattr
, exclusive
, mode
, vpp
, cr
);
6711 nfs_rw_exit(&drp
->r_rwlock
);
6716 * It _is_ a regular file.
6718 ASSERT(vattr
.va_mask
& AT_MODE
);
6719 if (MANDMODE(vattr
.va_mode
)) {
6720 nfs_rw_exit(&drp
->r_rwlock
);
6725 * If this happens to be a mknod of a regular file, then flags will
6726 * have neither FREAD or FWRITE. However, we must set at least one
6727 * for the call to nfs4open_otw. If it's open(O_CREAT) driving
6728 * nfs4_create, then either FREAD, FWRITE, or FRDWR has already been
6729 * set (based on openmode specified by app).
6731 if ((flags
& (FREAD
|FWRITE
)) == 0)
6732 flags
|= (FREAD
|FWRITE
);
6734 error
= nfs4open_otw(dvp
, nm
, &vattr
, vpp
, cr
, 1, flags
, createmode
, 0);
6737 /* if create was successful, throw away the file's pages */
6738 if (!error
&& (vattr
.va_mask
& AT_SIZE
))
6739 nfs4_invalidate_pages(vp
, (vattr
.va_size
& PAGEMASK
),
6741 /* release the lookup hold */
6747 * validate that we opened a regular file. This handles a misbehaving
6748 * server that returns an incorrect FH.
6750 if ((error
== 0) && *vpp
&& (*vpp
)->v_type
!= VREG
) {
6756 * If this is not an exclusive create, then the CREATE
6757 * request will be made with the GUARDED mode set. This
6758 * means that the server will return EEXIST if the file
6759 * exists. The file could exist because of a retransmitted
6760 * request. In this case, we recover by starting over and
6761 * checking to see whether the file exists. This second
6762 * time through it should and a CREATE request will not be
6765 * This handles the problem of a dangling CREATE request
6766 * which contains attributes which indicate that the file
6767 * should be truncated. This retransmitted request could
6768 * possibly truncate valid data in the file if not caught
6769 * by the duplicate request mechanism on the server or if
6770 * not caught by other means. The scenario is:
6772 * Client transmits CREATE request with size = 0
6773 * Client times out, retransmits request.
6774 * Response to the first request arrives from the server
6775 * and the client proceeds on.
6776 * Client writes data to the file.
6777 * The server now processes retransmitted CREATE request
6778 * and truncates file.
6780 * The use of the GUARDED CREATE request prevents this from
6781 * happening because the retransmitted CREATE would fail
6782 * with EEXIST and would not truncate the file.
6784 if (error
== EEXIST
&& exclusive
== NONEXCL
) {
6786 nfs4_create_misses
++;
6790 nfs_rw_exit(&drp
->r_rwlock
);
6791 if (truncating
&& !error
&& *vpp
) {
6795 * existing file got truncated, notify.
6799 if (IS_SHADOW(tvp
, trp
))
6801 vnevent_create(tvp
, ct
);
6807 * Create compound (for mkdir, mknod, symlink):
6808 * { Putfh <dfh>; Create; Getfh; Getattr }
6809 * It's okay if setattr failed to set gid - this is not considered
6810 * an error, but purge attrs in that case.
6813 call_nfs4_create_req(vnode_t
*dvp
, char *nm
, void *data
, struct vattr
*va
,
6814 vnode_t
**vpp
, cred_t
*cr
, nfs_ftype4 type
)
6816 int need_end_op
= FALSE
;
6817 COMPOUND4args_clnt args
;
6818 COMPOUND4res_clnt res
, *resp
= NULL
;
6823 rnode4_t
*drp
= VTOR4(dvp
);
6824 change_info4
*cinfo
;
6829 bool_t needrecov
= FALSE
;
6830 nfs4_recov_state_t recov_state
;
6831 nfs4_sharedfh_t
*sfhp
= NULL
;
6833 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
6834 int numops
, argoplist_size
, setgid_flag
, idx_create
, idx_fattr
;
6835 dirattr_info_t dinfo
, *dinfop
;
6839 ASSERT(type
== NF4DIR
|| type
== NF4LNK
|| type
== NF4BLK
||
6840 type
== NF4CHR
|| type
== NF4SOCK
|| type
== NF4FIFO
);
6845 * Make sure we properly deal with setting the right gid
6846 * on a new directory to reflect the parent's setgid bit
6849 if (type
== NF4DIR
) {
6852 va
->va_mode
&= ~VSGID
;
6853 dva
.va_mask
= AT_MODE
| AT_GID
;
6854 if (fop_getattr(dvp
, &dva
, 0, cr
, NULL
) == 0) {
6857 * If the parent's directory has the setgid bit set
6858 * _and_ the client was able to get a valid mapping
6859 * for the parent dir's owner_group, we want to
6860 * append NVERIFY(owner_group == dva.va_gid) and
6861 * SETTATTR to the CREATE compound.
6863 if (mi
->mi_flags
& MI4_GRPID
|| dva
.va_mode
& VSGID
) {
6865 va
->va_mode
|= VSGID
;
6866 if (dva
.va_gid
!= GID_NOBODY
) {
6867 va
->va_mask
|= AT_GID
;
6868 va
->va_gid
= dva
.va_gid
;
6876 * 0:putfh(dir) 1:savefh(dir) 2:create 3:getfh(new) 4:getattr(new)
6877 * 5:restorefh(dir) 6:getattr(dir)
6880 * 0:putfh(dir) 1:create 2:getfh(new) 3:getattr(new)
6881 * 4:savefh(new) 5:putfh(dir) 6:getattr(dir) 7:restorefh(new)
6882 * 8:nverify 9:setattr
6894 ASSERT(nfs_zone() == mi
->mi_zone
);
6895 if (nfs_rw_enter_sig(&drp
->r_rwlock
, RW_WRITER
, INTR4(dvp
))) {
6898 recov_state
.rs_flags
= 0;
6899 recov_state
.rs_num_retry_despite_err
= 0;
6901 argoplist_size
= numops
* sizeof (nfs_argop4
);
6902 argop
= kmem_alloc(argoplist_size
, KM_SLEEP
);
6906 args
.ctag
= TAG_SYMLINK
;
6907 else if (type
== NF4DIR
)
6908 args
.ctag
= TAG_MKDIR
;
6910 args
.ctag
= TAG_MKNOD
;
6912 args
.array_len
= numops
;
6915 if (e
.error
= nfs4_start_op(mi
, dvp
, NULL
, &recov_state
)) {
6916 nfs_rw_exit(&drp
->r_rwlock
);
6917 kmem_free(argop
, argoplist_size
);
6923 /* 0: putfh directory */
6924 argop
[0].argop
= OP_CPUTFH
;
6925 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= drp
->r_fh
;
6927 /* 1/2: Create object */
6928 argop
[idx_create
].argop
= OP_CCREATE
;
6929 argop
[idx_create
].nfs_argop4_u
.opccreate
.cname
= nm
;
6930 argop
[idx_create
].nfs_argop4_u
.opccreate
.type
= type
;
6931 if (type
== NF4LNK
) {
6933 * symlink, treat name as data
6935 ASSERT(data
!= NULL
);
6936 argop
[idx_create
].nfs_argop4_u
.opccreate
.ftype4_u
.clinkdata
=
6939 if (type
== NF4BLK
|| type
== NF4CHR
) {
6940 ASSERT(data
!= NULL
);
6941 argop
[idx_create
].nfs_argop4_u
.opccreate
.ftype4_u
.devdata
=
6942 *((specdata4
*)data
);
6945 crattr
= &argop
[idx_create
].nfs_argop4_u
.opccreate
.createattrs
;
6947 svp
= drp
->r_server
;
6948 (void) nfs_rw_enter_sig(&svp
->sv_lock
, RW_READER
, 0);
6949 supp_attrs
= svp
->sv_supp_attrs
;
6950 nfs_rw_exit(&svp
->sv_lock
);
6952 if (vattr_to_fattr4(va
, NULL
, crattr
, 0, OP_CREATE
, supp_attrs
)) {
6953 nfs_rw_exit(&drp
->r_rwlock
);
6954 nfs4_end_op(mi
, dvp
, NULL
, &recov_state
, needrecov
);
6956 kmem_free(argop
, argoplist_size
);
6960 /* 2/3: getfh fh of created object */
6961 ASSERT(idx_create
+ 1 == idx_fattr
- 1);
6962 argop
[idx_create
+ 1].argop
= OP_GETFH
;
6964 /* 3/4: getattr of new object */
6965 argop
[idx_fattr
].argop
= OP_GETATTR
;
6966 argop
[idx_fattr
].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
6967 argop
[idx_fattr
].nfs_argop4_u
.opgetattr
.mi
= mi
;
6972 argop
[4].argop
= OP_SAVEFH
;
6974 argop
[5].argop
= OP_CPUTFH
;
6975 argop
[5].nfs_argop4_u
.opcputfh
.sfh
= drp
->r_fh
;
6977 argop
[6].argop
= OP_GETATTR
;
6978 argop
[6].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
6979 argop
[6].nfs_argop4_u
.opgetattr
.mi
= mi
;
6981 argop
[7].argop
= OP_RESTOREFH
;
6986 * XXX - Revisit the last argument to nfs4_end_op()
6987 * once 5020486 is fixed.
6989 _v
.va_mask
= AT_GID
;
6990 _v
.va_gid
= va
->va_gid
;
6991 if (e
.error
= nfs4args_verify(&argop
[8], &_v
, OP_NVERIFY
,
6993 nfs4_end_op(mi
, dvp
, *vpp
, &recov_state
, TRUE
);
6994 nfs_rw_exit(&drp
->r_rwlock
);
6995 nfs4_fattr4_free(crattr
);
6996 kmem_free(argop
, argoplist_size
);
7003 * We _know_ we're not messing with AT_SIZE or AT_XTIME,
7004 * so no need for stateid or flags. Also we specify NULL
7005 * rp since we're only interested in setting owner_group
7008 nfs4args_setattr(&argop
[9], &_v
, NULL
, 0, NULL
, cr
, supp_attrs
,
7012 nfs4_end_op(mi
, dvp
, *vpp
, &recov_state
, TRUE
);
7013 nfs_rw_exit(&drp
->r_rwlock
);
7014 nfs4_fattr4_free(crattr
);
7015 nfs4args_verify_free(&argop
[8]);
7016 kmem_free(argop
, argoplist_size
);
7020 argop
[1].argop
= OP_SAVEFH
;
7022 argop
[5].argop
= OP_RESTOREFH
;
7024 argop
[6].argop
= OP_GETATTR
;
7025 argop
[6].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
7026 argop
[6].nfs_argop4_u
.opgetattr
.mi
= mi
;
7029 dnlc_remove(dvp
, nm
);
7033 rfs4call(mi
, &args
, &res
, cr
, &doqueue
, 0, &e
);
7035 needrecov
= nfs4_needs_recovery(&e
, FALSE
, mi
->mi_vfsp
);
7037 PURGE_ATTRCACHE4(dvp
);
7043 if (nfs4_start_recovery(&e
, mi
, dvp
, NULL
, NULL
, NULL
,
7044 OP_CREATE
, NULL
, NULL
, NULL
) == FALSE
) {
7045 nfs4_end_op(mi
, dvp
, NULL
, &recov_state
,
7047 need_end_op
= FALSE
;
7048 nfs4_fattr4_free(crattr
);
7050 nfs4args_verify_free(&argop
[8]);
7051 nfs4args_setattr_free(&argop
[9]);
7053 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
7060 if (res
.status
!= NFS4_OK
&& res
.array_len
<= idx_fattr
+ 1) {
7062 if (res
.status
== NFS4ERR_BADOWNER
)
7063 nfs4_log_badowner(mi
, OP_CREATE
);
7065 e
.error
= geterrno4(res
.status
);
7068 * This check is left over from when create was implemented
7069 * using a setattr op (instead of createattrs). If the
7070 * putfh/create/getfh failed, the error was returned. If
7071 * setattr/getattr failed, we keep going.
7073 * It might be better to get rid of the GETFH also, and just
7074 * do PUTFH/CREATE/GETATTR since the FH attr is mandatory.
7075 * Then if any of the operations failed, we could return the
7076 * error now, and remove much of the error code below.
7078 if (res
.array_len
<= idx_fattr
) {
7080 * Either Putfh, Create or Getfh failed.
7082 PURGE_ATTRCACHE4(dvp
);
7084 * nfs4_purge_stale_fh() may generate otw calls through
7085 * nfs4_invalidate_pages. Hence the need to call
7086 * nfs4_end_op() here to avoid nfs4_start_op() deadlock.
7088 nfs4_end_op(mi
, dvp
, NULL
, &recov_state
,
7090 need_end_op
= FALSE
;
7091 nfs4_purge_stale_fh(e
.error
, dvp
, cr
);
7096 resop
= &res
.array
[idx_create
]; /* create res */
7097 cinfo
= &resop
->nfs_resop4_u
.opcreate
.cinfo
;
7099 resop
= &res
.array
[idx_create
+ 1]; /* getfh res */
7100 gf_res
= &resop
->nfs_resop4_u
.opgetfh
;
7102 sfhp
= sfh4_get(&gf_res
->object
, mi
);
7104 *vpp
= vp
= makenfs4node(sfhp
, NULL
, dvp
->v_vfsp
, t
, cr
, dvp
,
7105 fn_get(VTOSV(dvp
)->sv_name
, nm
, sfhp
));
7106 if (vp
->v_type
== VNON
) {
7107 vattr
.va_mask
= AT_TYPE
;
7109 * Need to call nfs4_end_op before nfs4getattr to avoid
7110 * potential nfs4_start_op deadlock. See RFE 4777612.
7112 nfs4_end_op(mi
, dvp
, NULL
, &recov_state
,
7114 need_end_op
= FALSE
;
7115 e
.error
= nfs4getattr(vp
, &vattr
, cr
);
7121 vp
->v_type
= vattr
.va_type
;
7125 *vpp
= vp
= makenfs4node(sfhp
,
7126 &res
.array
[idx_fattr
].nfs_resop4_u
.opgetattr
.ga_res
,
7128 dvp
, fn_get(VTOSV(dvp
)->sv_name
, nm
, sfhp
));
7132 * If compound succeeded, then update dir attrs
7134 if (res
.status
== NFS4_OK
) {
7135 dinfo
.di_garp
= &res
.array
[6].nfs_resop4_u
.opgetattr
.ga_res
;
7137 dinfo
.di_time_call
= t
;
7142 /* Update directory cache attribute, readdir and dnlc caches */
7143 nfs4_update_dircaches(cinfo
, dvp
, vp
, nm
, dinfop
);
7148 nfs_rw_exit(&drp
->r_rwlock
);
7149 nfs4_fattr4_free(crattr
);
7151 nfs4args_verify_free(&argop
[8]);
7152 nfs4args_setattr_free(&argop
[9]);
7155 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)resp
);
7157 nfs4_end_op(mi
, dvp
, NULL
, &recov_state
, needrecov
);
7159 kmem_free(argop
, argoplist_size
);
7165 nfs4mknod(vnode_t
*dvp
, char *nm
, struct vattr
*va
, enum vcexcl exclusive
,
7166 int mode
, vnode_t
**vpp
, cred_t
*cr
)
7171 specdata4 spec
, *specp
= NULL
;
7173 ASSERT(nfs_zone() == VTOMI4(dvp
)->mi_zone
);
7175 switch (va
->va_type
) {
7178 type
= (va
->va_type
== VCHR
) ? NF4CHR
: NF4BLK
;
7179 spec
.specdata1
= getmajor(va
->va_rdev
);
7180 spec
.specdata2
= getminor(va
->va_rdev
);
7195 error
= call_nfs4_create_req(dvp
, nm
, specp
, va
, &vp
, cr
, type
);
7201 * This might not be needed any more; special case to deal
7202 * with problematic v2/v3 servers. Since create was unable
7203 * to set group correctly, not sure what hope setattr has.
7205 if (va
->va_gid
!= VTOR4(vp
)->r_attr
.va_gid
) {
7206 va
->va_mask
= AT_GID
;
7207 (void) nfs4setattr(vp
, va
, 0, cr
, NULL
);
7211 * If vnode is a device create special vnode
7213 if (ISVDEV(vp
->v_type
)) {
7214 *vpp
= specvp(vp
, vp
->v_rdev
, vp
->v_type
, cr
);
7223 * Remove requires that the current fh be the target directory.
7224 * After the operation, the current fh is unchanged.
7225 * The compound op structure is:
7226 * PUTFH(targetdir), REMOVE
7228 * Weirdness: if the vnode to be removed is open
7229 * we rename it instead of removing it and nfs_inactive
7230 * will remove the new name.
7234 nfs4_remove(vnode_t
*dvp
, char *nm
, cred_t
*cr
, caller_context_t
*ct
, int flags
)
7236 COMPOUND4args_clnt args
;
7237 COMPOUND4res_clnt res
, *resp
= NULL
;
7239 nfs_argop4 argop
[3];
7248 nfs4_recov_state_t recov_state
;
7250 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
7251 dirattr_info_t dinfo
;
7253 if (nfs_zone() != VTOMI4(dvp
)->mi_zone
)
7256 if (nfs_rw_enter_sig(&drp
->r_rwlock
, RW_WRITER
, INTR4(dvp
)))
7259 e
.error
= nfs4lookup(dvp
, nm
, &vp
, cr
, 0);
7261 nfs_rw_exit(&drp
->r_rwlock
);
7265 if (vp
->v_type
== VDIR
) {
7267 nfs_rw_exit(&drp
->r_rwlock
);
7272 * First just remove the entry from the name cache, as it
7273 * is most likely the only entry for this vp.
7275 dnlc_remove(dvp
, nm
);
7280 * For regular file types, check to see if the file is open by looking
7281 * at the open streams.
7282 * For all other types, check the reference count on the vnode. Since
7283 * they are not opened OTW they never have an open stream.
7285 * If the file is open, rename it to .nfsXXXX.
7287 if (vp
->v_type
!= VREG
) {
7289 * If the file has a v_count > 1 then there may be more than one
7290 * entry in the name cache due multiple links or an open file,
7291 * but we don't have the real reference count so flush all
7294 if (vp
->v_count
> 1)
7298 * Now we have the real reference count.
7300 isopen
= vp
->v_count
> 1;
7302 mutex_enter(&rp
->r_os_lock
);
7303 isopen
= list_head(&rp
->r_open_streams
) != NULL
;
7304 mutex_exit(&rp
->r_os_lock
);
7307 mutex_enter(&rp
->r_statelock
);
7309 (rp
->r_unldvp
== NULL
|| strcmp(nm
, rp
->r_unlname
) == 0)) {
7310 mutex_exit(&rp
->r_statelock
);
7311 tmpname
= newname();
7312 e
.error
= nfs4rename(dvp
, nm
, dvp
, tmpname
, cr
, ct
);
7314 kmem_free(tmpname
, MAXNAMELEN
);
7316 mutex_enter(&rp
->r_statelock
);
7317 if (rp
->r_unldvp
== NULL
) {
7320 if (rp
->r_unlcred
!= NULL
)
7321 crfree(rp
->r_unlcred
);
7324 rp
->r_unlname
= tmpname
;
7326 kmem_free(rp
->r_unlname
, MAXNAMELEN
);
7327 rp
->r_unlname
= tmpname
;
7329 mutex_exit(&rp
->r_statelock
);
7332 nfs_rw_exit(&drp
->r_rwlock
);
7336 * Actually remove the file/dir
7338 mutex_exit(&rp
->r_statelock
);
7341 * We need to flush any dirty pages which happen to
7342 * be hanging around before removing the file.
7343 * This shouldn't happen very often since in NFSv4
7344 * we should be close to open consistent.
7346 if (nfs4_has_pages(vp
) &&
7347 ((rp
->r_flags
& R4DIRTY
) || rp
->r_count
> 0)) {
7348 e
.error
= nfs4_putpage(vp
, 0, 0, 0, cr
, ct
);
7349 if (e
.error
&& (e
.error
== ENOSPC
|| e
.error
== EDQUOT
)) {
7350 mutex_enter(&rp
->r_statelock
);
7352 rp
->r_error
= e
.error
;
7353 mutex_exit(&rp
->r_statelock
);
7359 (void) nfs4delegreturn(rp
, NFS4_DR_REOPEN
);
7360 recov_state
.rs_flags
= 0;
7361 recov_state
.rs_num_retry_despite_err
= 0;
7365 * Remove ops: putfh dir; remove
7367 args
.ctag
= TAG_REMOVE
;
7371 e
.error
= nfs4_start_op(VTOMI4(dvp
), dvp
, NULL
, &recov_state
);
7373 nfs_rw_exit(&drp
->r_rwlock
);
7378 /* putfh directory */
7379 argop
[0].argop
= OP_CPUTFH
;
7380 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= drp
->r_fh
;
7383 argop
[1].argop
= OP_CREMOVE
;
7384 argop
[1].nfs_argop4_u
.opcremove
.ctarget
= nm
;
7387 argop
[2].argop
= OP_GETATTR
;
7388 argop
[2].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
7389 argop
[2].nfs_argop4_u
.opgetattr
.mi
= mi
;
7392 dinfo
.di_time_call
= gethrtime();
7393 rfs4call(mi
, &args
, &res
, cr
, &doqueue
, 0, &e
);
7395 PURGE_ATTRCACHE4(vp
);
7397 needrecov
= nfs4_needs_recovery(&e
, FALSE
, mi
->mi_vfsp
);
7399 PURGE_ATTRCACHE4(dvp
);
7402 if (nfs4_start_recovery(&e
, VTOMI4(dvp
), dvp
,
7403 NULL
, NULL
, NULL
, OP_REMOVE
, NULL
, NULL
, NULL
) == FALSE
) {
7405 (void) xdr_free(xdr_COMPOUND4res_clnt
,
7407 nfs4_end_op(VTOMI4(dvp
), dvp
, NULL
, &recov_state
,
7414 * Matching nfs4_end_op() for start_op() above.
7415 * There is a path in the code below which calls
7416 * nfs4_purge_stale_fh(), which may generate otw calls through
7417 * nfs4_invalidate_pages. Hence we need to call nfs4_end_op()
7418 * here to avoid nfs4_start_op() deadlock.
7420 nfs4_end_op(VTOMI4(dvp
), dvp
, NULL
, &recov_state
, needrecov
);
7426 e
.error
= geterrno4(res
.status
);
7427 PURGE_ATTRCACHE4(dvp
);
7428 nfs4_purge_stale_fh(e
.error
, dvp
, cr
);
7430 resop
= &res
.array
[1]; /* remove res */
7431 rm_res
= &resop
->nfs_resop4_u
.opremove
;
7434 &res
.array
[2].nfs_resop4_u
.opgetattr
.ga_res
;
7437 /* Update directory attr, readdir and dnlc caches */
7438 nfs4_update_dircaches(&rm_res
->cinfo
, dvp
, NULL
, NULL
,
7442 nfs_rw_exit(&drp
->r_rwlock
);
7444 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)resp
);
7451 if (IS_SHADOW(vp
, trp
))
7453 vnevent_remove(tvp
, dvp
, nm
, ct
);
7460 * Link requires that the current fh be the target directory and the
7461 * saved fh be the source fh. After the operation, the current fh is unchanged.
7462 * Thus the compound op structure is:
7463 * PUTFH(file), SAVEFH, PUTFH(targetdir), LINK, RESTOREFH,
7468 nfs4_link(vnode_t
*tdvp
, vnode_t
*svp
, char *tnm
, cred_t
*cr
,
7469 caller_context_t
*ct
, int flags
)
7471 COMPOUND4args_clnt args
;
7472 COMPOUND4res_clnt res
, *resp
= NULL
;
7474 int argoplist_size
= 7 * sizeof (nfs_argop4
);
7477 vnode_t
*realvp
, *nvp
;
7481 bool_t needrecov
= FALSE
;
7482 nfs4_recov_state_t recov_state
;
7484 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
7485 dirattr_info_t dinfo
;
7487 ASSERT(*tnm
!= '\0');
7488 ASSERT(tdvp
->v_type
== VDIR
);
7489 ASSERT(nfs4_consistent_type(tdvp
));
7490 ASSERT(nfs4_consistent_type(svp
));
7492 if (nfs_zone() != VTOMI4(tdvp
)->mi_zone
)
7494 if (fop_realvp(svp
, &realvp
, ct
) == 0) {
7496 ASSERT(nfs4_consistent_type(svp
));
7502 if (!(mi
->mi_flags
& MI4_LINK
)) {
7503 return (EOPNOTSUPP
);
7505 recov_state
.rs_flags
= 0;
7506 recov_state
.rs_num_retry_despite_err
= 0;
7508 if (nfs_rw_enter_sig(&tdrp
->r_rwlock
, RW_WRITER
, INTR4(tdvp
)))
7512 argop
= kmem_alloc(argoplist_size
, KM_SLEEP
);
7514 args
.ctag
= TAG_LINK
;
7517 * Link ops: putfh fl; savefh; putfh tdir; link; getattr(dir);
7518 * restorefh; getattr(fl)
7523 e
.error
= nfs4_start_op(VTOMI4(svp
), svp
, tdvp
, &recov_state
);
7525 kmem_free(argop
, argoplist_size
);
7526 nfs_rw_exit(&tdrp
->r_rwlock
);
7531 argop
[0].argop
= OP_CPUTFH
;
7532 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= VTOR4(svp
)->r_fh
;
7534 /* 1. save current fh to free up the space for the dir */
7535 argop
[1].argop
= OP_SAVEFH
;
7537 /* 2. putfh targetdir */
7538 argop
[2].argop
= OP_CPUTFH
;
7539 argop
[2].nfs_argop4_u
.opcputfh
.sfh
= tdrp
->r_fh
;
7541 /* 3. link: current_fh is targetdir, saved_fh is source */
7542 argop
[3].argop
= OP_CLINK
;
7543 argop
[3].nfs_argop4_u
.opclink
.cnewname
= tnm
;
7545 /* 4. Get attributes of dir */
7546 argop
[4].argop
= OP_GETATTR
;
7547 argop
[4].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
7548 argop
[4].nfs_argop4_u
.opgetattr
.mi
= mi
;
7550 /* 5. If link was successful, restore current vp to file */
7551 argop
[5].argop
= OP_RESTOREFH
;
7553 /* 6. Get attributes of linked object */
7554 argop
[6].argop
= OP_GETATTR
;
7555 argop
[6].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
7556 argop
[6].nfs_argop4_u
.opgetattr
.mi
= mi
;
7558 dnlc_remove(tdvp
, tnm
);
7563 rfs4call(VTOMI4(svp
), &args
, &res
, cr
, &doqueue
, 0, &e
);
7565 needrecov
= nfs4_needs_recovery(&e
, FALSE
, svp
->v_vfsp
);
7566 if (e
.error
!= 0 && !needrecov
) {
7567 PURGE_ATTRCACHE4(tdvp
);
7568 PURGE_ATTRCACHE4(svp
);
7569 nfs4_end_op(VTOMI4(svp
), svp
, tdvp
, &recov_state
, needrecov
);
7576 abort
= nfs4_start_recovery(&e
, VTOMI4(svp
), svp
, tdvp
,
7577 NULL
, NULL
, OP_LINK
, NULL
, NULL
, NULL
);
7578 if (abort
== FALSE
) {
7579 nfs4_end_op(VTOMI4(svp
), svp
, tdvp
, &recov_state
,
7581 kmem_free(argop
, argoplist_size
);
7583 (void) xdr_free(xdr_COMPOUND4res_clnt
,
7588 PURGE_ATTRCACHE4(tdvp
);
7589 PURGE_ATTRCACHE4(svp
);
7590 nfs4_end_op(VTOMI4(svp
), svp
, tdvp
,
7591 &recov_state
, needrecov
);
7594 /* fall through for res.status case */
7598 nfs4_end_op(VTOMI4(svp
), svp
, tdvp
, &recov_state
, needrecov
);
7602 /* If link succeeded, then don't return error */
7603 e
.error
= geterrno4(res
.status
);
7604 if (res
.array_len
<= 4) {
7606 * Either Putfh, Savefh, Putfh dir, or Link failed
7608 PURGE_ATTRCACHE4(svp
);
7609 PURGE_ATTRCACHE4(tdvp
);
7610 if (e
.error
== EOPNOTSUPP
) {
7611 mutex_enter(&mi
->mi_lock
);
7612 mi
->mi_flags
&= ~MI4_LINK
;
7613 mutex_exit(&mi
->mi_lock
);
7616 if (e
.error
== EISDIR
&& crgetuid(cr
) != 0)
7622 /* either no error or one of the postop getattr failed */
7625 * XXX - if LINK succeeded, but no attrs were returned for link
7626 * file, purge its cache.
7628 * XXX Perform a simplified version of wcc checking. Instead of
7629 * have another getattr to get pre-op, just purge cache if
7630 * any of the ops prior to and including the getattr failed.
7631 * If the getattr succeeded then update the attrcache accordingly.
7635 * update cache with link file postattrs.
7636 * Note: at this point resop points to link res.
7638 resop
= &res
.array
[3]; /* link res */
7639 ln_res
= &resop
->nfs_resop4_u
.oplink
;
7640 if (res
.status
== NFS4_OK
)
7641 e
.error
= nfs4_update_attrcache(res
.status
,
7642 &res
.array
[6].nfs_resop4_u
.opgetattr
.ga_res
,
7646 * Call makenfs4node to create the new shadow vp for tnm.
7647 * We pass NULL attrs because we just cached attrs for
7648 * the src object. All we're trying to accomplish is to
7649 * to create the new shadow vnode.
7651 nvp
= makenfs4node(VTOR4(svp
)->r_fh
, NULL
, tdvp
->v_vfsp
, t
, cr
,
7652 tdvp
, fn_get(VTOSV(tdvp
)->sv_name
, tnm
, VTOR4(svp
)->r_fh
));
7654 /* Update target cache attribute, readdir and dnlc caches */
7655 dinfo
.di_garp
= &res
.array
[4].nfs_resop4_u
.opgetattr
.ga_res
;
7656 dinfo
.di_time_call
= t
;
7659 nfs4_update_dircaches(&ln_res
->cinfo
, tdvp
, nvp
, tnm
, &dinfo
);
7660 ASSERT(nfs4_consistent_type(tdvp
));
7661 ASSERT(nfs4_consistent_type(svp
));
7662 ASSERT(nfs4_consistent_type(nvp
));
7669 * Notify the source file of this link operation.
7673 if (IS_SHADOW(svp
, trp
))
7675 vnevent_link(tvp
, ct
);
7678 kmem_free(argop
, argoplist_size
);
7680 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)resp
);
7682 nfs_rw_exit(&tdrp
->r_rwlock
);
7689 nfs4_rename(vnode_t
*odvp
, char *onm
, vnode_t
*ndvp
, char *nnm
, cred_t
*cr
,
7690 caller_context_t
*ct
, int flags
)
7694 if (nfs_zone() != VTOMI4(odvp
)->mi_zone
)
7696 if (fop_realvp(ndvp
, &realvp
, ct
) == 0)
7699 return (nfs4rename(odvp
, onm
, ndvp
, nnm
, cr
, ct
));
7703 * nfs4rename does the real work of renaming in NFS Version 4.
7705 * A file handle is considered volatile for renaming purposes if either
7706 * of the volatile bits are turned on. However, the compound may differ
7707 * based on the likelihood of the filehandle to change during rename.
7710 nfs4rename(vnode_t
*odvp
, char *onm
, vnode_t
*ndvp
, char *nnm
, cred_t
*cr
,
7711 caller_context_t
*ct
)
7715 vnode_t
*nvp
= NULL
;
7716 vnode_t
*ovp
= NULL
;
7717 char *tmpname
= NULL
;
7723 nfsstat4 stat
= NFS4_OK
;
7725 ASSERT(nfs_zone() == VTOMI4(odvp
)->mi_zone
);
7726 ASSERT(nfs4_consistent_type(odvp
));
7727 ASSERT(nfs4_consistent_type(ndvp
));
7729 if (onm
[0] == '.' && (onm
[1] == '\0' ||
7730 (onm
[1] == '.' && onm
[2] == '\0')))
7733 if (nnm
[0] == '.' && (nnm
[1] == '\0' ||
7734 (nnm
[1] == '.' && nnm
[2] == '\0')))
7739 if ((intptr_t)odrp
< (intptr_t)ndrp
) {
7740 if (nfs_rw_enter_sig(&odrp
->r_rwlock
, RW_WRITER
, INTR4(odvp
)))
7742 if (nfs_rw_enter_sig(&ndrp
->r_rwlock
, RW_WRITER
, INTR4(ndvp
))) {
7743 nfs_rw_exit(&odrp
->r_rwlock
);
7747 if (nfs_rw_enter_sig(&ndrp
->r_rwlock
, RW_WRITER
, INTR4(ndvp
)))
7749 if (nfs_rw_enter_sig(&odrp
->r_rwlock
, RW_WRITER
, INTR4(odvp
))) {
7750 nfs_rw_exit(&ndrp
->r_rwlock
);
7756 * Lookup the target file. If it exists, it needs to be
7757 * checked to see whether it is a mount point and whether
7758 * it is active (open).
7760 error
= nfs4lookup(ndvp
, nnm
, &nvp
, cr
, 0);
7764 ASSERT(nfs4_consistent_type(nvp
));
7766 * If this file has been mounted on, then just
7767 * return busy because renaming to it would remove
7768 * the mounted file system from the name space.
7770 if (vn_ismntpt(nvp
)) {
7772 nfs_rw_exit(&odrp
->r_rwlock
);
7773 nfs_rw_exit(&ndrp
->r_rwlock
);
7778 * First just remove the entry from the name cache, as it
7779 * is most likely the only entry for this vp.
7781 dnlc_remove(ndvp
, nnm
);
7785 if (nvp
->v_type
!= VREG
) {
7787 * Purge the name cache of all references to this vnode
7788 * so that we can check the reference count to infer
7789 * whether it is active or not.
7791 if (nvp
->v_count
> 1)
7794 isactive
= nvp
->v_count
> 1;
7796 mutex_enter(&rp
->r_os_lock
);
7797 isactive
= list_head(&rp
->r_open_streams
) != NULL
;
7798 mutex_exit(&rp
->r_os_lock
);
7802 * If the vnode is active and is not a directory,
7803 * arrange to rename it to a
7804 * temporary file so that it will continue to be
7805 * accessible. This implements the "unlink-open-file"
7806 * semantics for the target of a rename operation.
7807 * Before doing this though, make sure that the
7808 * source and target files are not already the same.
7810 if (isactive
&& nvp
->v_type
!= VDIR
) {
7812 * Lookup the source name.
7814 error
= nfs4lookup(odvp
, onm
, &ovp
, cr
, 0);
7817 * The source name *should* already exist.
7821 nfs_rw_exit(&odrp
->r_rwlock
);
7822 nfs_rw_exit(&ndrp
->r_rwlock
);
7826 ASSERT(nfs4_consistent_type(ovp
));
7829 * Compare the two vnodes. If they are the same,
7830 * just release all held vnodes and return success.
7832 if (VN_CMP(ovp
, nvp
)) {
7835 nfs_rw_exit(&odrp
->r_rwlock
);
7836 nfs_rw_exit(&ndrp
->r_rwlock
);
7841 * Can't mix and match directories and non-
7842 * directories in rename operations. We already
7843 * know that the target is not a directory. If
7844 * the source is a directory, return an error.
7846 if (ovp
->v_type
== VDIR
) {
7849 nfs_rw_exit(&odrp
->r_rwlock
);
7850 nfs_rw_exit(&ndrp
->r_rwlock
);
7855 * The target file exists, is not the same as
7856 * the source file, and is active. We first
7857 * try to Link it to a temporary filename to
7858 * avoid having the server removing the file
7859 * completely (which could cause data loss to
7860 * the user's POV in the event the Rename fails
7861 * -- see bug 1165874).
7864 * The do_link and did_link booleans are
7865 * introduced in the event we get NFS4ERR_FILE_OPEN
7866 * returned for the Rename. Some servers can
7867 * not Rename over an Open file, so they return
7868 * this error. The client needs to Remove the
7869 * newly created Link and do two Renames, just
7870 * as if the server didn't support LINK.
7872 tmpname
= newname();
7876 error
= nfs4_link(ndvp
, nvp
, tmpname
, cr
,
7879 if (error
== EOPNOTSUPP
|| !do_link
) {
7880 error
= nfs4_rename(ndvp
, nnm
, ndvp
, tmpname
,
7887 kmem_free(tmpname
, MAXNAMELEN
);
7890 nfs_rw_exit(&odrp
->r_rwlock
);
7891 nfs_rw_exit(&ndrp
->r_rwlock
);
7895 mutex_enter(&rp
->r_statelock
);
7896 if (rp
->r_unldvp
== NULL
) {
7898 rp
->r_unldvp
= ndvp
;
7899 if (rp
->r_unlcred
!= NULL
)
7900 crfree(rp
->r_unlcred
);
7903 rp
->r_unlname
= tmpname
;
7906 kmem_free(rp
->r_unlname
, MAXNAMELEN
);
7907 rp
->r_unlname
= tmpname
;
7909 mutex_exit(&rp
->r_statelock
);
7912 (void) nfs4delegreturn(VTOR4(nvp
), NFS4_DR_PUSH
|NFS4_DR_REOPEN
);
7914 ASSERT(nfs4_consistent_type(nvp
));
7919 * When renaming directories to be a subdirectory of a
7920 * different parent, the dnlc entry for ".." will no
7921 * longer be valid, so it must be removed.
7923 * We do a lookup here to determine whether we are renaming
7924 * a directory and we need to check if we are renaming
7925 * an unlinked file. This might have already been done
7926 * in previous code, so we check ovp == NULL to avoid
7929 error
= nfs4lookup(odvp
, onm
, &ovp
, cr
, 0);
7931 * The source name *should* already exist.
7934 nfs_rw_exit(&odrp
->r_rwlock
);
7935 nfs_rw_exit(&ndrp
->r_rwlock
);
7941 ASSERT(ovp
!= NULL
);
7942 ASSERT(nfs4_consistent_type(ovp
));
7946 * Is the object being renamed a dir, and if so, is
7947 * it being renamed to a child of itself? The underlying
7948 * fs should ultimately return EINVAL for this case;
7949 * however, buggy beta non-Solaris NFSv4 servers at
7950 * interop testing events have allowed this behavior,
7951 * and it caused our client to panic due to a recursive
7952 * mutex_enter in fn_move.
7954 * The tedious locking in fn_move could be changed to
7955 * deal with this case, and the client could avoid the
7956 * panic; however, the client would just confuse itself
7957 * later and misbehave. A better way to handle the broken
7958 * server is to detect this condition and return EINVAL
7959 * without ever sending the the bogus rename to the server.
7960 * We know the rename is invalid -- just fail it now.
7962 if (ovp
->v_type
== VDIR
&& VN_CMP(ndvp
, ovp
)) {
7964 nfs_rw_exit(&odrp
->r_rwlock
);
7965 nfs_rw_exit(&ndrp
->r_rwlock
);
7972 (void) nfs4delegreturn(VTOR4(ovp
), NFS4_DR_PUSH
|NFS4_DR_REOPEN
);
7975 * If FH4_VOL_RENAME or FH4_VOLATILE_ANY bits are set, it is
7976 * possible for the filehandle to change due to the rename.
7977 * If neither of these bits is set, but FH4_VOL_MIGRATION is set,
7978 * the fh will not change because of the rename, but we still need
7979 * to update its rnode entry with the new name for
7980 * an eventual fh change due to migration. The FH4_NOEXPIRE_ON_OPEN
7981 * has no effect on these for now, but for future improvements,
7982 * we might want to use it too to simplify handling of files
7983 * that are open with that flag on. (XXX)
7986 if (NFS4_VOLATILE_FH(mi
))
7987 error
= nfs4rename_volatile_fh(odvp
, onm
, ovp
, ndvp
, nnm
, cr
,
7990 error
= nfs4rename_persistent_fh(odvp
, onm
, ovp
, ndvp
, nnm
, cr
,
7993 ASSERT(nfs4_consistent_type(odvp
));
7994 ASSERT(nfs4_consistent_type(ndvp
));
7995 ASSERT(nfs4_consistent_type(ovp
));
7997 if (stat
== NFS4ERR_FILE_OPEN
&& did_link
) {
8000 * Before the 'link_call' code, we did a nfs4_lookup
8001 * that puts a VN_HOLD on nvp. After the nfs4_link
8002 * call we call VN_RELE to match that hold. We need
8003 * to place an additional VN_HOLD here since we will
8004 * be hitting that VN_RELE again.
8008 (void) nfs4_remove(ndvp
, tmpname
, cr
, NULL
, 0);
8010 /* Undo the unlinked file naming stuff we just did */
8011 mutex_enter(&rp
->r_statelock
);
8014 rp
->r_unldvp
= NULL
;
8015 if (rp
->r_unlcred
!= NULL
)
8016 crfree(rp
->r_unlcred
);
8017 rp
->r_unlcred
= NULL
;
8018 /* rp->r_unlanme points to tmpname */
8020 kmem_free(rp
->r_unlname
, MAXNAMELEN
);
8021 rp
->r_unlname
= NULL
;
8023 mutex_exit(&rp
->r_statelock
);
8033 nfs_rw_exit(&odrp
->r_rwlock
);
8034 nfs_rw_exit(&ndrp
->r_rwlock
);
8042 * when renaming directories to be a subdirectory of a
8043 * different parent, the dnlc entry for ".." will no
8044 * longer be valid, so it must be removed
8048 if (ovp
->v_type
== VDIR
) {
8049 dnlc_remove(ovp
, "..");
8050 if (rp
->r_dir
!= NULL
)
8051 nfs4_purge_rddir_cache(ovp
);
8056 * If we are renaming the unlinked file, update the
8057 * r_unldvp and r_unlname as needed.
8059 mutex_enter(&rp
->r_statelock
);
8060 if (rp
->r_unldvp
!= NULL
) {
8061 if (strcmp(rp
->r_unlname
, onm
) == 0) {
8062 (void) strncpy(rp
->r_unlname
, nnm
, MAXNAMELEN
);
8063 rp
->r_unlname
[MAXNAMELEN
- 1] = '\0';
8064 if (ndvp
!= rp
->r_unldvp
) {
8065 VN_RELE(rp
->r_unldvp
);
8066 rp
->r_unldvp
= ndvp
;
8071 mutex_exit(&rp
->r_statelock
);
8074 * Notify the rename vnevents to source vnode, and to the target
8075 * vnode if it already existed.
8081 * Notify the vnode. Each links is represented by
8082 * a different vnode, in nfsv4.
8087 if (IS_SHADOW(nvp
, trp
))
8089 vnevent_rename_dest(tvp
, ndvp
, nnm
, ct
);
8093 * if the source and destination directory are not the
8094 * same notify the destination directory.
8096 if (VTOR4(odvp
) != VTOR4(ndvp
)) {
8099 if (IS_SHADOW(ndvp
, trp
))
8101 vnevent_rename_dest_dir(tvp
, ct
);
8106 if (IS_SHADOW(ovp
, trp
))
8108 vnevent_rename_src(tvp
, odvp
, onm
, ct
);
8116 nfs_rw_exit(&odrp
->r_rwlock
);
8117 nfs_rw_exit(&ndrp
->r_rwlock
);
8123 * When the parent directory has changed, sv_dfh must be updated
8126 update_parentdir_sfh(vnode_t
*vp
, vnode_t
*ndvp
)
8128 svnode_t
*sv
= VTOSV(vp
);
8129 nfs4_sharedfh_t
*old_dfh
= sv
->sv_dfh
;
8130 nfs4_sharedfh_t
*new_dfh
= VTOR4(ndvp
)->r_fh
;
8133 sv
->sv_dfh
= new_dfh
;
8134 sfh4_rele(&old_dfh
);
8138 * nfs4rename_persistent does the otw portion of renaming in NFS Version 4,
8139 * when it is known that the filehandle is persistent through rename.
8141 * Rename requires that the current fh be the target directory and the
8142 * saved fh be the source directory. After the operation, the current fh
8144 * The compound op structure for persistent fh rename is:
8145 * PUTFH(sourcdir), SAVEFH, PUTFH(targetdir), RENAME
8146 * Rather than bother with the directory postop args, we'll simply
8147 * update that a change occurred in the cache, so no post-op getattrs.
8150 nfs4rename_persistent_fh(vnode_t
*odvp
, char *onm
, vnode_t
*renvp
,
8151 vnode_t
*ndvp
, char *nnm
, cred_t
*cr
, nfsstat4
*statp
)
8153 COMPOUND4args_clnt args
;
8154 COMPOUND4res_clnt res
, *resp
= NULL
;
8157 int doqueue
, argoplist_size
;
8159 rnode4_t
*odrp
= VTOR4(odvp
);
8160 rnode4_t
*ndrp
= VTOR4(ndvp
);
8163 nfs4_recov_state_t recov_state
;
8164 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
8165 dirattr_info_t dinfo
, *dinfop
;
8167 ASSERT(nfs_zone() == VTOMI4(odvp
)->mi_zone
);
8169 recov_state
.rs_flags
= 0;
8170 recov_state
.rs_num_retry_despite_err
= 0;
8173 * Rename ops: putfh sdir; savefh; putfh tdir; rename; getattr tdir
8175 * If source/target are different dirs, then append putfh(src); getattr
8177 args
.array_len
= (odvp
== ndvp
) ? 5 : 7;
8178 argoplist_size
= args
.array_len
* sizeof (nfs_argop4
);
8179 args
.array
= argop
= kmem_alloc(argoplist_size
, KM_SLEEP
);
8184 /* No need to Lookup the file, persistent fh */
8185 args
.ctag
= TAG_RENAME
;
8188 e
.error
= nfs4_start_op(mi
, odvp
, ndvp
, &recov_state
);
8190 kmem_free(argop
, argoplist_size
);
8194 /* 0: putfh source directory */
8195 argop
[0].argop
= OP_CPUTFH
;
8196 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= odrp
->r_fh
;
8198 /* 1: Save source fh to free up current for target */
8199 argop
[1].argop
= OP_SAVEFH
;
8201 /* 2: putfh targetdir */
8202 argop
[2].argop
= OP_CPUTFH
;
8203 argop
[2].nfs_argop4_u
.opcputfh
.sfh
= ndrp
->r_fh
;
8205 /* 3: current_fh is targetdir, saved_fh is sourcedir */
8206 argop
[3].argop
= OP_CRENAME
;
8207 argop
[3].nfs_argop4_u
.opcrename
.coldname
= onm
;
8208 argop
[3].nfs_argop4_u
.opcrename
.cnewname
= nnm
;
8210 /* 4: getattr (targetdir) */
8211 argop
[4].argop
= OP_GETATTR
;
8212 argop
[4].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
8213 argop
[4].nfs_argop4_u
.opgetattr
.mi
= mi
;
8217 /* 5: putfh (sourcedir) */
8218 argop
[5].argop
= OP_CPUTFH
;
8219 argop
[5].nfs_argop4_u
.opcputfh
.sfh
= ndrp
->r_fh
;
8221 /* 6: getattr (sourcedir) */
8222 argop
[6].argop
= OP_GETATTR
;
8223 argop
[6].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
8224 argop
[6].nfs_argop4_u
.opgetattr
.mi
= mi
;
8227 dnlc_remove(odvp
, onm
);
8228 dnlc_remove(ndvp
, nnm
);
8231 dinfo
.di_time_call
= gethrtime();
8232 rfs4call(mi
, &args
, &res
, cr
, &doqueue
, 0, &e
);
8234 needrecov
= nfs4_needs_recovery(&e
, FALSE
, mi
->mi_vfsp
);
8236 PURGE_ATTRCACHE4(odvp
);
8237 PURGE_ATTRCACHE4(ndvp
);
8239 *statp
= res
.status
;
8243 if (nfs4_start_recovery(&e
, mi
, odvp
, ndvp
, NULL
, NULL
,
8244 OP_RENAME
, NULL
, NULL
, NULL
) == FALSE
) {
8245 nfs4_end_op(mi
, odvp
, ndvp
, &recov_state
, needrecov
);
8247 (void) xdr_free(xdr_COMPOUND4res_clnt
,
8256 * as long as OP_RENAME
8258 if (res
.status
!= NFS4_OK
&& res
.array_len
<= 4) {
8259 e
.error
= geterrno4(res
.status
);
8260 PURGE_ATTRCACHE4(odvp
);
8261 PURGE_ATTRCACHE4(ndvp
);
8263 * System V defines rename to return EEXIST, not
8264 * ENOTEMPTY if the target directory is not empty.
8265 * Over the wire, the error is NFSERR_ENOTEMPTY
8266 * which geterrno4 maps to ENOTEMPTY.
8268 if (e
.error
== ENOTEMPTY
)
8272 resop
= &res
.array
[3]; /* rename res */
8273 rn_res
= &resop
->nfs_resop4_u
.oprename
;
8275 if (res
.status
== NFS4_OK
) {
8277 * Update target attribute, readdir and dnlc
8281 &res
.array
[4].nfs_resop4_u
.opgetattr
.ga_res
;
8287 nfs4_update_dircaches(&rn_res
->target_cinfo
,
8288 ndvp
, NULL
, NULL
, dinfop
);
8291 * Update source attribute, readdir and dnlc caches
8295 update_parentdir_sfh(renvp
, ndvp
);
8299 &(res
.array
[6].nfs_resop4_u
.
8302 nfs4_update_dircaches(&rn_res
->source_cinfo
,
8303 odvp
, NULL
, NULL
, dinfop
);
8306 fn_move(VTOSV(renvp
)->sv_name
, VTOSV(ndvp
)->sv_name
,
8312 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)resp
);
8313 nfs4_end_op(mi
, odvp
, ndvp
, &recov_state
, needrecov
);
8314 kmem_free(argop
, argoplist_size
);
8320 * nfs4rename_volatile_fh does the otw part of renaming in NFS Version 4, when
8321 * it is possible for the filehandle to change due to the rename.
8323 * The compound req in this case includes a post-rename lookup and getattr
8324 * to ensure that we have the correct fh and attributes for the object.
8326 * Rename requires that the current fh be the target directory and the
8327 * saved fh be the source directory. After the operation, the current fh
8330 * We need the new filehandle (hence a LOOKUP and GETFH) so that we can
8331 * update the filehandle for the renamed object. We also get the old
8332 * filehandle for historical reasons; this should be taken out sometime.
8333 * This results in a rather cumbersome compound...
8335 * PUTFH(sourcdir), SAVEFH, LOOKUP(src), GETFH(old),
8336 * PUTFH(targetdir), RENAME, LOOKUP(trgt), GETFH(new), GETATTR
8340 nfs4rename_volatile_fh(vnode_t
*odvp
, char *onm
, vnode_t
*ovp
,
8341 vnode_t
*ndvp
, char *nnm
, cred_t
*cr
, nfsstat4
*statp
)
8343 COMPOUND4args_clnt args
;
8344 COMPOUND4res_clnt res
, *resp
= NULL
;
8350 rnode4_t
*odrp
= VTOR4(odvp
); /* old directory */
8351 rnode4_t
*ndrp
= VTOR4(ndvp
); /* new directory */
8352 rnode4_t
*orp
= VTOR4(ovp
); /* object being renamed */
8356 nfs4_recov_state_t recov_state
;
8358 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
8359 dirattr_info_t dinfo
, *dinfop
= &dinfo
;
8361 ASSERT(nfs_zone() == VTOMI4(odvp
)->mi_zone
);
8363 recov_state
.rs_flags
= 0;
8364 recov_state
.rs_num_retry_despite_err
= 0;
8370 * There is a window between the RPC and updating the path and
8371 * filehandle stored in the rnode. Lock out the FHEXPIRED recovery
8372 * code, so that it doesn't try to use the old path during that
8375 mutex_enter(&orp
->r_statelock
);
8376 while (orp
->r_flags
& R4RECEXPFH
) {
8377 klwp_t
*lwp
= ttolwp(curthread
);
8381 if (cv_wait_sig(&orp
->r_cv
, &orp
->r_statelock
) == 0) {
8382 mutex_exit(&orp
->r_statelock
);
8390 orp
->r_flags
|= R4RECEXPFH
;
8391 mutex_exit(&orp
->r_statelock
);
8395 args
.ctag
= TAG_RENAME_VFH
;
8396 args
.array_len
= (odvp
== ndvp
) ? 10 : 12;
8397 argoplist_size
= args
.array_len
* sizeof (nfs_argop4
);
8398 argop
= kmem_alloc(argoplist_size
, KM_SLEEP
);
8402 * PUTFH(sourcdir), SAVEFH, LOOKUP(src), GETFH(old),
8403 * PUTFH(targetdir), RENAME, GETATTR(targetdir)
8404 * LOOKUP(trgt), GETFH(new), GETATTR,
8407 * add putfh(sourcedir), getattr(sourcedir) }
8411 e
.error
= nfs4_start_fop(mi
, odvp
, ndvp
, OH_VFH_RENAME
,
8412 &recov_state
, NULL
);
8414 kmem_free(argop
, argoplist_size
);
8415 mutex_enter(&orp
->r_statelock
);
8416 orp
->r_flags
&= ~R4RECEXPFH
;
8417 cv_broadcast(&orp
->r_cv
);
8418 mutex_exit(&orp
->r_statelock
);
8422 /* 0: putfh source directory */
8423 argop
[0].argop
= OP_CPUTFH
;
8424 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= odrp
->r_fh
;
8426 /* 1: Save source fh to free up current for target */
8427 argop
[1].argop
= OP_SAVEFH
;
8429 /* 2: Lookup pre-rename fh of renamed object */
8430 argop
[2].argop
= OP_CLOOKUP
;
8431 argop
[2].nfs_argop4_u
.opclookup
.cname
= onm
;
8433 /* 3: getfh fh of renamed object (before rename) */
8434 argop
[3].argop
= OP_GETFH
;
8436 /* 4: putfh targetdir */
8437 argop
[4].argop
= OP_CPUTFH
;
8438 argop
[4].nfs_argop4_u
.opcputfh
.sfh
= ndrp
->r_fh
;
8440 /* 5: current_fh is targetdir, saved_fh is sourcedir */
8441 argop
[5].argop
= OP_CRENAME
;
8442 argop
[5].nfs_argop4_u
.opcrename
.coldname
= onm
;
8443 argop
[5].nfs_argop4_u
.opcrename
.cnewname
= nnm
;
8445 /* 6: getattr of target dir (post op attrs) */
8446 argop
[6].argop
= OP_GETATTR
;
8447 argop
[6].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
8448 argop
[6].nfs_argop4_u
.opgetattr
.mi
= mi
;
8450 /* 7: Lookup post-rename fh of renamed object */
8451 argop
[7].argop
= OP_CLOOKUP
;
8452 argop
[7].nfs_argop4_u
.opclookup
.cname
= nnm
;
8454 /* 8: getfh fh of renamed object (after rename) */
8455 argop
[8].argop
= OP_GETFH
;
8457 /* 9: getattr of renamed object */
8458 argop
[9].argop
= OP_GETATTR
;
8459 argop
[9].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
8460 argop
[9].nfs_argop4_u
.opgetattr
.mi
= mi
;
8463 * If source/target dirs are different, then get new post-op
8464 * attrs for source dir also.
8467 /* 10: putfh (sourcedir) */
8468 argop
[10].argop
= OP_CPUTFH
;
8469 argop
[10].nfs_argop4_u
.opcputfh
.sfh
= ndrp
->r_fh
;
8471 /* 11: getattr (sourcedir) */
8472 argop
[11].argop
= OP_GETATTR
;
8473 argop
[11].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
8474 argop
[11].nfs_argop4_u
.opgetattr
.mi
= mi
;
8477 dnlc_remove(odvp
, onm
);
8478 dnlc_remove(ndvp
, nnm
);
8482 rfs4call(mi
, &args
, &res
, cr
, &doqueue
, 0, &e
);
8484 needrecov
= nfs4_needs_recovery(&e
, FALSE
, mi
->mi_vfsp
);
8486 PURGE_ATTRCACHE4(odvp
);
8487 PURGE_ATTRCACHE4(ndvp
);
8489 nfs4_end_fop(mi
, odvp
, ndvp
, OH_VFH_RENAME
,
8490 &recov_state
, needrecov
);
8494 *statp
= res
.status
;
8500 abort
= nfs4_start_recovery(&e
, mi
, odvp
, ndvp
, NULL
, NULL
,
8501 OP_RENAME
, NULL
, NULL
, NULL
);
8502 if (abort
== FALSE
) {
8503 nfs4_end_fop(mi
, odvp
, ndvp
, OH_VFH_RENAME
,
8504 &recov_state
, needrecov
);
8505 kmem_free(argop
, argoplist_size
);
8507 (void) xdr_free(xdr_COMPOUND4res_clnt
,
8509 mutex_enter(&orp
->r_statelock
);
8510 orp
->r_flags
&= ~R4RECEXPFH
;
8511 cv_broadcast(&orp
->r_cv
);
8512 mutex_exit(&orp
->r_statelock
);
8516 nfs4_end_fop(mi
, odvp
, ndvp
, OH_VFH_RENAME
,
8517 &recov_state
, needrecov
);
8520 /* fall through for res.status case */
8526 * If OP_RENAME (or any prev op) failed, then return an error.
8527 * OP_RENAME is index 5, so if array len <= 6 we return an error.
8529 if ((res
.status
!= NFS4_OK
) && (res
.array_len
<= 6)) {
8531 * Error in an op other than last Getattr
8533 e
.error
= geterrno4(res
.status
);
8534 PURGE_ATTRCACHE4(odvp
);
8535 PURGE_ATTRCACHE4(ndvp
);
8537 * System V defines rename to return EEXIST, not
8538 * ENOTEMPTY if the target directory is not empty.
8539 * Over the wire, the error is NFSERR_ENOTEMPTY
8540 * which geterrno4 maps to ENOTEMPTY.
8542 if (e
.error
== ENOTEMPTY
)
8544 nfs4_end_fop(mi
, odvp
, ndvp
, OH_VFH_RENAME
, &recov_state
,
8549 /* rename results */
8550 rn_res
= &res
.array
[5].nfs_resop4_u
.oprename
;
8552 if (res
.status
== NFS4_OK
) {
8553 /* Update target attribute, readdir and dnlc caches */
8555 &res
.array
[6].nfs_resop4_u
.opgetattr
.ga_res
;
8557 dinfo
.di_time_call
= t
;
8561 /* Update source cache attribute, readdir and dnlc caches */
8562 nfs4_update_dircaches(&rn_res
->target_cinfo
, ndvp
, NULL
, NULL
, dinfop
);
8564 /* Update source cache attribute, readdir and dnlc caches */
8566 update_parentdir_sfh(ovp
, ndvp
);
8569 * If dinfop is non-NULL, then compound succeded, so
8570 * set di_garp to attrs for source dir. dinfop is only
8571 * set to NULL when compound fails.
8575 &res
.array
[11].nfs_resop4_u
.opgetattr
.ga_res
;
8576 nfs4_update_dircaches(&rn_res
->source_cinfo
, odvp
, NULL
, NULL
,
8581 * Update the rnode with the new component name and args,
8582 * and if the file handle changed, also update it with the new fh.
8583 * This is only necessary if the target object has an rnode
8584 * entry and there is no need to create one for it.
8586 resop
= &res
.array
[8]; /* getfh new res */
8587 ngf_res
= &resop
->nfs_resop4_u
.opgetfh
;
8590 * Update the path and filehandle for the renamed object.
8592 nfs4rename_update(ovp
, ndvp
, &ngf_res
->object
, nnm
);
8594 nfs4_end_fop(mi
, odvp
, ndvp
, OH_VFH_RENAME
, &recov_state
, needrecov
);
8596 if (res
.status
== NFS4_OK
) {
8597 resop
++; /* getattr res */
8598 e
.error
= nfs4_update_attrcache(res
.status
,
8599 &resop
->nfs_resop4_u
.opgetattr
.ga_res
,
8604 kmem_free(argop
, argoplist_size
);
8606 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)resp
);
8607 mutex_enter(&orp
->r_statelock
);
8608 orp
->r_flags
&= ~R4RECEXPFH
;
8609 cv_broadcast(&orp
->r_cv
);
8610 mutex_exit(&orp
->r_statelock
);
8617 nfs4_mkdir(vnode_t
*dvp
, char *nm
, struct vattr
*va
, vnode_t
**vpp
, cred_t
*cr
,
8618 caller_context_t
*ct
, int flags
, vsecattr_t
*vsecp
)
8623 if (nfs_zone() != VTOMI4(dvp
)->mi_zone
)
8626 * As ".." has special meaning and rather than send a mkdir
8627 * over the wire to just let the server freak out, we just
8628 * short circuit it here and return EEXIST
8630 if (nm
[0] == '.' && nm
[1] == '.' && nm
[2] == '\0')
8634 * Decision to get the right gid and setgid bit of the
8635 * new directory is now made in call_nfs4_create_req.
8637 va
->va_mask
|= AT_MODE
;
8638 error
= call_nfs4_create_req(dvp
, nm
, NULL
, va
, &vp
, cr
, NF4DIR
);
8648 * rmdir is using the same remove v4 op as does remove.
8649 * Remove requires that the current fh be the target directory.
8650 * After the operation, the current fh is unchanged.
8651 * The compound op structure is:
8652 * PUTFH(targetdir), REMOVE
8656 nfs4_rmdir(vnode_t
*dvp
, char *nm
, vnode_t
*cdir
, cred_t
*cr
,
8657 caller_context_t
*ct
, int flags
)
8659 int need_end_op
= FALSE
;
8660 COMPOUND4args_clnt args
;
8661 COMPOUND4res_clnt res
, *resp
= NULL
;
8663 nfs_argop4 argop
[3];
8669 bool_t needrecov
= FALSE
;
8670 nfs4_recov_state_t recov_state
;
8671 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
8672 dirattr_info_t dinfo
, *dinfop
;
8674 if (nfs_zone() != VTOMI4(dvp
)->mi_zone
)
8677 * As ".." has special meaning and rather than send a rmdir
8678 * over the wire to just let the server freak out, we just
8679 * short circuit it here and return EEXIST
8681 if (nm
[0] == '.' && nm
[1] == '.' && nm
[2] == '\0')
8685 if (nfs_rw_enter_sig(&drp
->r_rwlock
, RW_WRITER
, INTR4(dvp
)))
8689 * Attempt to prevent a rmdir(".") from succeeding.
8691 e
.error
= nfs4lookup(dvp
, nm
, &vp
, cr
, 0);
8693 nfs_rw_exit(&drp
->r_rwlock
);
8698 nfs_rw_exit(&drp
->r_rwlock
);
8703 * Since nfsv4 remove op works on both files and directories,
8704 * check that the removed object is indeed a directory.
8706 if (vp
->v_type
!= VDIR
) {
8708 nfs_rw_exit(&drp
->r_rwlock
);
8713 * First just remove the entry from the name cache, as it
8714 * is most likely an entry for this vp.
8716 dnlc_remove(dvp
, nm
);
8719 * If there vnode reference count is greater than one, then
8720 * there may be additional references in the DNLC which will
8721 * need to be purged. First, trying removing the entry for
8722 * the parent directory and see if that removes the additional
8723 * reference(s). If that doesn't do it, then use dnlc_purge_vp
8724 * to completely remove any references to the directory which
8725 * might still exist in the DNLC.
8727 if (vp
->v_count
> 1) {
8728 dnlc_remove(vp
, "..");
8729 if (vp
->v_count
> 1)
8734 recov_state
.rs_flags
= 0;
8735 recov_state
.rs_num_retry_despite_err
= 0;
8738 args
.ctag
= TAG_RMDIR
;
8741 * Rmdir ops: putfh dir; remove
8746 e
.error
= nfs4_start_op(VTOMI4(dvp
), dvp
, NULL
, &recov_state
);
8748 nfs_rw_exit(&drp
->r_rwlock
);
8753 /* putfh directory */
8754 argop
[0].argop
= OP_CPUTFH
;
8755 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= drp
->r_fh
;
8758 argop
[1].argop
= OP_CREMOVE
;
8759 argop
[1].nfs_argop4_u
.opcremove
.ctarget
= nm
;
8761 /* getattr (postop attrs for dir that contained removed dir) */
8762 argop
[2].argop
= OP_GETATTR
;
8763 argop
[2].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
8764 argop
[2].nfs_argop4_u
.opgetattr
.mi
= mi
;
8766 dinfo
.di_time_call
= gethrtime();
8768 rfs4call(mi
, &args
, &res
, cr
, &doqueue
, 0, &e
);
8770 PURGE_ATTRCACHE4(vp
);
8772 needrecov
= nfs4_needs_recovery(&e
, FALSE
, mi
->mi_vfsp
);
8774 PURGE_ATTRCACHE4(dvp
);
8778 if (nfs4_start_recovery(&e
, VTOMI4(dvp
), dvp
, NULL
, NULL
,
8779 NULL
, OP_REMOVE
, NULL
, NULL
, NULL
) == FALSE
) {
8781 (void) xdr_free(xdr_COMPOUND4res_clnt
,
8784 nfs4_end_op(VTOMI4(dvp
), dvp
, NULL
, &recov_state
,
8786 need_end_op
= FALSE
;
8795 * Only return error if first 2 ops (OP_REMOVE or earlier)
8798 if (res
.status
!= NFS4_OK
&& res
.array_len
<= 2) {
8799 e
.error
= geterrno4(res
.status
);
8800 PURGE_ATTRCACHE4(dvp
);
8801 nfs4_end_op(VTOMI4(dvp
), dvp
, NULL
,
8802 &recov_state
, needrecov
);
8803 need_end_op
= FALSE
;
8804 nfs4_purge_stale_fh(e
.error
, dvp
, cr
);
8806 * System V defines rmdir to return EEXIST, not
8807 * ENOTEMPTY if the directory is not empty. Over
8808 * the wire, the error is NFSERR_ENOTEMPTY which
8809 * geterrno4 maps to ENOTEMPTY.
8811 if (e
.error
== ENOTEMPTY
)
8814 resop
= &res
.array
[1]; /* remove res */
8815 rm_res
= &resop
->nfs_resop4_u
.opremove
;
8817 if (res
.status
== NFS4_OK
) {
8818 resop
= &res
.array
[2]; /* dir attrs */
8820 &resop
->nfs_resop4_u
.opgetattr
.ga_res
;
8826 /* Update dir attribute, readdir and dnlc caches */
8827 nfs4_update_dircaches(&rm_res
->cinfo
, dvp
, NULL
, NULL
,
8830 /* destroy rddir cache for dir that was removed */
8831 if (VTOR4(vp
)->r_dir
!= NULL
)
8832 nfs4_purge_rddir_cache(vp
);
8837 nfs4_end_op(VTOMI4(dvp
), dvp
, NULL
, &recov_state
, needrecov
);
8839 nfs_rw_exit(&drp
->r_rwlock
);
8842 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)resp
);
8849 if (IS_SHADOW(vp
, trp
))
8851 vnevent_rmdir(tvp
, dvp
, nm
, ct
);
8861 nfs4_symlink(vnode_t
*dvp
, char *lnm
, struct vattr
*tva
, char *tnm
, cred_t
*cr
,
8862 caller_context_t
*ct
, int flags
)
8868 mntinfo4_t
*mi
= VTOMI4(dvp
);
8870 if (nfs_zone() != mi
->mi_zone
)
8872 if (!(mi
->mi_flags
& MI4_SYMLINK
))
8873 return (EOPNOTSUPP
);
8875 error
= call_nfs4_create_req(dvp
, lnm
, tnm
, tva
, &vp
, cr
, NF4LNK
);
8879 ASSERT(nfs4_consistent_type(vp
));
8881 if (nfs4_do_symlink_cache
&& rp
->r_symlink
.contents
== NULL
) {
8883 contents
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
8885 if (contents
!= NULL
) {
8886 mutex_enter(&rp
->r_statelock
);
8887 if (rp
->r_symlink
.contents
== NULL
) {
8888 rp
->r_symlink
.len
= strlen(tnm
);
8889 bcopy(tnm
, contents
, rp
->r_symlink
.len
);
8890 rp
->r_symlink
.contents
= contents
;
8891 rp
->r_symlink
.size
= MAXPATHLEN
;
8892 mutex_exit(&rp
->r_statelock
);
8894 mutex_exit(&rp
->r_statelock
);
8895 kmem_free((void *)contents
, MAXPATHLEN
);
8906 * Read directory entries.
8907 * There are some weird things to look out for here. The uio_loffset
8908 * field is either 0 or it is the offset returned from a previous
8909 * readdir. It is an opaque value used by the server to find the
8910 * correct directory block to read. The count field is the number
8911 * of blocks to read on the server. This is advisory only, the server
8912 * may return only one block's worth of entries. Entries may be compressed
8917 nfs4_readdir(vnode_t
*vp
, struct uio
*uiop
, cred_t
*cr
, int *eofp
,
8918 caller_context_t
*ct
, int flags
)
8926 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
8930 ASSERT(nfs_rw_lock_held(&rp
->r_rwlock
, RW_READER
));
8933 * Make sure that the directory cache is valid.
8935 if (rp
->r_dir
!= NULL
) {
8936 if (nfs_disable_rddir_cache
!= 0) {
8938 * Setting nfs_disable_rddir_cache in /etc/system
8939 * allows interoperability with servers that do not
8940 * properly update the attributes of directories.
8941 * Any cached information gets purged before an
8942 * access is made to it.
8944 nfs4_purge_rddir_cache(vp
);
8947 error
= nfs4_validate_caches(vp
, cr
);
8952 count
= MIN(uiop
->uio_iov
->iov_len
, MAXBSIZE
);
8955 * Short circuit last readdir which always returns 0 bytes.
8956 * This can be done after the directory has been read through
8957 * completely at least once. This will set r_direof which
8958 * can be used to find the value of the last cookie.
8960 mutex_enter(&rp
->r_statelock
);
8961 if (rp
->r_direof
!= NULL
&&
8962 uiop
->uio_loffset
== rp
->r_direof
->nfs4_ncookie
) {
8963 mutex_exit(&rp
->r_statelock
);
8965 nfs4_readdir_cache_shorts
++;
8973 * Look for a cache entry. Cache entries are identified
8974 * by the NFS cookie value and the byte count requested.
8976 rdc
= rddir4_cache_lookup(rp
, uiop
->uio_loffset
, count
);
8979 * If rdc is NULL then the lookup resulted in an unrecoverable error.
8982 mutex_exit(&rp
->r_statelock
);
8987 * Check to see if we need to fill this entry in.
8989 if (rdc
->flags
& RDDIRREQ
) {
8990 rdc
->flags
&= ~RDDIRREQ
;
8991 rdc
->flags
|= RDDIR
;
8992 mutex_exit(&rp
->r_statelock
);
8997 nfs4readdir(vp
, rdc
, cr
);
9000 * Reacquire the lock, so that we can continue
9002 mutex_enter(&rp
->r_statelock
);
9004 * The entry is now complete
9006 rdc
->flags
&= ~RDDIR
;
9009 ASSERT(!(rdc
->flags
& RDDIR
));
9012 * If an error occurred while attempting
9013 * to fill the cache entry, mark the entry invalid and
9014 * just return the error.
9018 rdc
->flags
|= RDDIRREQ
;
9019 rddir4_cache_rele(rp
, rdc
);
9020 mutex_exit(&rp
->r_statelock
);
9025 * The cache entry is complete and good,
9026 * copyout the dirent structs to the calling
9029 error
= uiomove(rdc
->entries
, rdc
->actlen
, UIO_READ
, uiop
);
9032 * If no error occurred during the copyout,
9033 * update the offset in the uio struct to
9034 * contain the value of the next NFS 4 cookie
9035 * and set the eof value appropriately.
9038 uiop
->uio_loffset
= rdc
->nfs4_ncookie
;
9044 * Decide whether to do readahead. Don't if we
9045 * have already read to the end of directory.
9049 * Make the entry the direof only if it is cached
9051 if (rdc
->flags
& RDDIRCACHED
)
9053 rddir4_cache_rele(rp
, rdc
);
9054 mutex_exit(&rp
->r_statelock
);
9058 /* Determine if a readdir readahead should be done */
9059 if (!(rp
->r_flags
& R4LOOKUP
)) {
9060 rddir4_cache_rele(rp
, rdc
);
9061 mutex_exit(&rp
->r_statelock
);
9066 * Now look for a readahead entry.
9068 * Check to see whether we found an entry for the readahead.
9069 * If so, we don't need to do anything further, so free the new
9070 * entry if one was allocated. Otherwise, allocate a new entry, add
9071 * it to the cache, and then initiate an asynchronous readdir
9072 * operation to fill it.
9074 rrdc
= rddir4_cache_lookup(rp
, rdc
->nfs4_ncookie
, count
);
9077 * A readdir cache entry could not be obtained for the readahead. In
9078 * this case we skip the readahead and return.
9081 rddir4_cache_rele(rp
, rdc
);
9082 mutex_exit(&rp
->r_statelock
);
9087 * Check to see if we need to fill this entry in.
9089 if (rrdc
->flags
& RDDIRREQ
) {
9090 rrdc
->flags
&= ~RDDIRREQ
;
9091 rrdc
->flags
|= RDDIR
;
9092 rddir4_cache_rele(rp
, rdc
);
9093 mutex_exit(&rp
->r_statelock
);
9095 nfs4_readdir_readahead
++;
9100 nfs4_async_readdir(vp
, rrdc
, cr
, do_nfs4readdir
);
9104 rddir4_cache_rele(rp
, rrdc
);
9105 rddir4_cache_rele(rp
, rdc
);
9106 mutex_exit(&rp
->r_statelock
);
9111 do_nfs4readdir(vnode_t
*vp
, rddir4_cache
*rdc
, cred_t
*cr
)
9116 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
9121 * Obtain the readdir results for the caller.
9123 nfs4readdir(vp
, rdc
, cr
);
9125 mutex_enter(&rp
->r_statelock
);
9127 * The entry is now complete
9129 rdc
->flags
&= ~RDDIR
;
9133 rdc
->flags
|= RDDIRREQ
;
9134 rddir4_cache_rele(rp
, rdc
);
9135 mutex_exit(&rp
->r_statelock
);
9141 * Read directory entries.
9142 * There are some weird things to look out for here. The uio_loffset
9143 * field is either 0 or it is the offset returned from a previous
9144 * readdir. It is an opaque value used by the server to find the
9145 * correct directory block to read. The count field is the number
9146 * of blocks to read on the server. This is advisory only, the server
9147 * may return only one block's worth of entries. Entries may be compressed
9150 * Generates the following compound request:
9151 * 1. If readdir offset is zero and no dnlc entry for parent exists,
9152 * must include a Lookupp as well. In this case, send:
9153 * { Putfh <fh>; Readdir; Lookupp; Getfh; Getattr }
9154 * 2. Otherwise just do: { Putfh <fh>; Readdir }
9156 * Get complete attributes and filehandles for entries if this is the
9157 * first read of the directory. Otherwise, just get fileid's.
9160 nfs4readdir(vnode_t
*vp
, rddir4_cache
*rdc
, cred_t
*cr
)
9162 COMPOUND4args_clnt args
;
9163 COMPOUND4res_clnt res
;
9164 READDIR4args
*rargs
;
9165 READDIR4res_clnt
*rd_res
;
9167 nfs_argop4 argop
[5];
9169 rnode4_t
*rp
= VTOR4(vp
);
9170 mntinfo4_t
*mi
= VTOMI4(vp
);
9172 u_longlong_t nodeid
, pnodeid
; /* id's of dir and its parents */
9174 nfs_cookie4 cookie
= (nfs_cookie4
)rdc
->nfs4_cookie
;
9175 int num_ops
, res_opcnt
;
9176 bool_t needrecov
= FALSE
;
9177 nfs4_recov_state_t recov_state
;
9179 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
9181 ASSERT(nfs_zone() == mi
->mi_zone
);
9182 ASSERT(rdc
->flags
& RDDIR
);
9183 ASSERT(rdc
->entries
== NULL
);
9186 * If rp were a stub, it should have triggered and caused
9187 * a mount for us to get this far.
9189 ASSERT(!RP_ISSTUB(rp
));
9192 if (cookie
== (nfs_cookie4
)0 || cookie
== (nfs_cookie4
)1) {
9194 * Since nfsv4 readdir may not return entries for "." and "..",
9195 * the client must recreate them:
9196 * To find the correct nodeid, do the following:
9197 * For current node, get nodeid from dnlc.
9198 * - if current node is rootvp, set pnodeid to nodeid.
9199 * - else if parent is in the dnlc, get its nodeid from there.
9200 * - else add LOOKUPP+GETATTR to compound.
9202 nodeid
= rp
->r_attr
.va_nodeid
;
9203 if (vp
->v_flag
& VROOT
) {
9204 pnodeid
= nodeid
; /* root of mount point */
9206 dvp
= dnlc_lookup(vp
, "..");
9207 if (dvp
!= NULL
&& dvp
!= DNLC_NO_VNODE
) {
9208 /* parent in dnlc cache - no need for otw */
9209 pnodeid
= VTOR4(dvp
)->r_attr
.va_nodeid
;
9212 * parent not in dnlc cache,
9213 * do lookupp to get its id
9216 pnodeid
= 0; /* set later by getattr parent */
9222 recov_state
.rs_flags
= 0;
9223 recov_state
.rs_num_retry_despite_err
= 0;
9225 /* Save the original mount point security flavor */
9226 (void) save_mnt_secinfo(mi
->mi_curr_serv
);
9229 args
.ctag
= TAG_READDIR
;
9232 args
.array_len
= num_ops
;
9234 if (e
.error
= nfs4_start_fop(VTOMI4(vp
), vp
, NULL
, OH_READDIR
,
9235 &recov_state
, NULL
)) {
9237 * If readdir a node that is a stub for a crossed mount point,
9238 * keep the original secinfo flavor for the current file
9239 * system, not the crossed one.
9241 (void) check_mnt_secinfo(mi
->mi_curr_serv
, vp
);
9242 rdc
->error
= e
.error
;
9247 * Determine which attrs to request for dirents. This code
9248 * must be protected by nfs4_start/end_fop because of r_server
9249 * (which will change during failover recovery).
9252 if (rp
->r_flags
& (R4LOOKUP
| R4READDIRWATTR
)) {
9254 * Get all vattr attrs plus filehandle and rdattr_error
9256 rd_bitsval
= NFS4_VATTR_MASK
|
9257 FATTR4_RDATTR_ERROR_MASK
|
9258 FATTR4_FILEHANDLE_MASK
;
9260 if (rp
->r_flags
& R4READDIRWATTR
) {
9261 mutex_enter(&rp
->r_statelock
);
9262 rp
->r_flags
&= ~R4READDIRWATTR
;
9263 mutex_exit(&rp
->r_statelock
);
9266 servinfo4_t
*svp
= rp
->r_server
;
9269 * Already read directory. Use readdir with
9270 * no attrs (except for mounted_on_fileid) for updates.
9272 rd_bitsval
= FATTR4_RDATTR_ERROR_MASK
;
9275 * request mounted on fileid if supported, else request
9276 * fileid. maybe we should verify that fileid is supported
9277 * and request something else if not.
9279 (void) nfs_rw_enter_sig(&svp
->sv_lock
, RW_READER
, 0);
9280 if (svp
->sv_supp_attrs
& FATTR4_MOUNTED_ON_FILEID_MASK
)
9281 rd_bitsval
|= FATTR4_MOUNTED_ON_FILEID_MASK
;
9282 nfs_rw_exit(&svp
->sv_lock
);
9285 /* putfh directory fh */
9286 argop
[0].argop
= OP_CPUTFH
;
9287 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= rp
->r_fh
;
9289 argop
[1].argop
= OP_READDIR
;
9290 rargs
= &argop
[1].nfs_argop4_u
.opreaddir
;
9292 * 1 and 2 are reserved for client "." and ".." entry offset.
9293 * cookie 0 should be used over-the-wire to start reading at
9294 * the beginning of the directory excluding "." and "..".
9296 if (rdc
->nfs4_cookie
== 0 ||
9297 rdc
->nfs4_cookie
== 1 ||
9298 rdc
->nfs4_cookie
== 2) {
9299 rargs
->cookie
= (nfs_cookie4
)0;
9300 rargs
->cookieverf
= 0;
9302 rargs
->cookie
= (nfs_cookie4
)rdc
->nfs4_cookie
;
9303 mutex_enter(&rp
->r_statelock
);
9304 rargs
->cookieverf
= rp
->r_cookieverf4
;
9305 mutex_exit(&rp
->r_statelock
);
9307 rargs
->dircount
= MIN(rdc
->buflen
, mi
->mi_tsize
);
9308 rargs
->maxcount
= mi
->mi_tsize
;
9309 rargs
->attr_request
= rd_bitsval
;
9317 * If count < than the minimum required, we return no entries
9318 * and fail with EINVAL
9320 if (rargs
->dircount
< (DIRENT64_RECLEN(1) + DIRENT64_RECLEN(2))) {
9321 rdc
->error
= EINVAL
;
9325 if (args
.array_len
== 5) {
9327 * Add lookupp and getattr for parent nodeid.
9329 argop
[2].argop
= OP_LOOKUPP
;
9331 argop
[3].argop
= OP_GETFH
;
9333 /* getattr parent */
9334 argop
[4].argop
= OP_GETATTR
;
9335 argop
[4].nfs_argop4_u
.opgetattr
.attr_request
= NFS4_VATTR_MASK
;
9336 argop
[4].nfs_argop4_u
.opgetattr
.mi
= mi
;
9341 if (mi
->mi_io_kstats
) {
9342 mutex_enter(&mi
->mi_lock
);
9343 kstat_runq_enter(KSTAT_IO_PTR(mi
->mi_io_kstats
));
9344 mutex_exit(&mi
->mi_lock
);
9347 /* capture the time of this call */
9348 rargs
->t
= t
= gethrtime();
9350 rfs4call(mi
, &args
, &res
, cr
, &doqueue
, 0, &e
);
9352 if (mi
->mi_io_kstats
) {
9353 mutex_enter(&mi
->mi_lock
);
9354 kstat_runq_exit(KSTAT_IO_PTR(mi
->mi_io_kstats
));
9355 mutex_exit(&mi
->mi_lock
);
9358 needrecov
= nfs4_needs_recovery(&e
, FALSE
, mi
->mi_vfsp
);
9361 * If RPC error occurred and it isn't an error that
9362 * triggers recovery, then go ahead and fail now.
9364 if (e
.error
!= 0 && !needrecov
) {
9365 rdc
->error
= e
.error
;
9372 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
,
9373 "nfs4readdir: initiating recovery.\n"));
9375 abort
= nfs4_start_recovery(&e
, VTOMI4(vp
), vp
, NULL
, NULL
,
9376 NULL
, OP_READDIR
, NULL
, NULL
, NULL
);
9377 if (abort
== FALSE
) {
9378 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_READDIR
,
9379 &recov_state
, needrecov
);
9381 (void) xdr_free(xdr_COMPOUND4res_clnt
,
9383 if (rdc
->entries
!= NULL
) {
9384 kmem_free(rdc
->entries
, rdc
->entlen
);
9385 rdc
->entries
= NULL
;
9391 rdc
->error
= e
.error
;
9395 /* fall through for res.status case */
9398 res_opcnt
= res
.array_len
;
9401 * If compound failed first 2 ops (PUTFH+READDIR), then return
9402 * failure here. Subsequent ops are for filling out dot-dot
9403 * dirent, and if they fail, we still want to give the caller
9404 * the dirents returned by (the successful) READDIR op, so we need
9405 * to silently ignore failure for subsequent ops (LOOKUPP+GETATTR).
9407 * One example where PUTFH+READDIR ops would succeed but
9408 * LOOKUPP+GETATTR would fail would be a dir that has r perm
9409 * but lacks x. In this case, a POSIX server's fop_readdir
9410 * would succeed; however, fop_lookup(..) would fail since no
9411 * x perm. We need to come up with a non-vendor-specific way
9412 * for a POSIX server to return d_ino from dotdot's dirent if
9413 * client only requests mounted_on_fileid, and just say the
9414 * LOOKUPP succeeded and fill out the GETATTR. However, if
9415 * client requested any mandatory attrs, server would be required
9416 * to fail the GETATTR op because it can't call fop_lookup+fop_getattr
9421 if (res_opcnt
<= 2) {
9422 e
.error
= geterrno4(res
.status
);
9423 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_READDIR
,
9424 &recov_state
, needrecov
);
9425 nfs4_purge_stale_fh(e
.error
, vp
, cr
);
9426 rdc
->error
= e
.error
;
9427 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
9428 if (rdc
->entries
!= NULL
) {
9429 kmem_free(rdc
->entries
, rdc
->entlen
);
9430 rdc
->entries
= NULL
;
9433 * If readdir a node that is a stub for a
9434 * crossed mount point, keep the original
9435 * secinfo flavor for the current file system,
9436 * not the crossed one.
9438 (void) check_mnt_secinfo(mi
->mi_curr_serv
, vp
);
9443 resop
= &res
.array
[1]; /* readdir res */
9444 rd_res
= &resop
->nfs_resop4_u
.opreaddirclnt
;
9446 mutex_enter(&rp
->r_statelock
);
9447 rp
->r_cookieverf4
= rd_res
->cookieverf
;
9448 mutex_exit(&rp
->r_statelock
);
9451 * For "." and ".." entries
9453 * seek(cookie=0) -> "." entry with d_off = 1
9454 * seek(cookie=1) -> ".." entry with d_off = 2
9456 if (cookie
== (nfs_cookie4
) 0) {
9458 rd_res
->dotp
->d_ino
= nodeid
;
9459 if (rd_res
->dotdotp
)
9460 rd_res
->dotdotp
->d_ino
= pnodeid
;
9462 if (cookie
== (nfs_cookie4
) 1) {
9463 if (rd_res
->dotdotp
)
9464 rd_res
->dotdotp
->d_ino
= pnodeid
;
9468 /* LOOKUPP+GETATTR attemped */
9469 if (args
.array_len
== 5 && rd_res
->dotdotp
) {
9470 if (res
.status
== NFS4_OK
&& res_opcnt
== 5) {
9472 nfs4_sharedfh_t
*sfhp
;
9474 nfs4_ga_res_t
*garp
;
9476 resop
++; /* lookupp */
9477 resop
++; /* getfh */
9478 fhp
= &resop
->nfs_resop4_u
.opgetfh
.object
;
9480 resop
++; /* getattr of parent */
9483 * First, take care of finishing the
9486 garp
= &resop
->nfs_resop4_u
.opgetattr
.ga_res
;
9488 * The d_ino of .. must be the inode number
9489 * of the mounted filesystem.
9491 if (garp
->n4g_va
.va_mask
& AT_NODEID
)
9492 rd_res
->dotdotp
->d_ino
=
9493 garp
->n4g_va
.va_nodeid
;
9497 * Next, create the ".." dnlc entry
9499 sfhp
= sfh4_get(fhp
, mi
);
9500 if (!nfs4_make_dotdot(sfhp
, t
, vp
, cr
, &pvp
, 0)) {
9501 dnlc_update(vp
, "..", pvp
);
9508 if (mi
->mi_io_kstats
) {
9509 mutex_enter(&mi
->mi_lock
);
9510 KSTAT_IO_PTR(mi
->mi_io_kstats
)->reads
++;
9511 KSTAT_IO_PTR(mi
->mi_io_kstats
)->nread
+= rdc
->actlen
;
9512 mutex_exit(&mi
->mi_lock
);
9515 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
9519 * If readdir a node that is a stub for a crossed mount point,
9520 * keep the original secinfo flavor for the current file system,
9521 * not the crossed one.
9523 (void) check_mnt_secinfo(mi
->mi_curr_serv
, vp
);
9525 nfs4_end_fop(mi
, vp
, NULL
, OH_READDIR
, &recov_state
, needrecov
);
9530 nfs4_bio(struct buf
*bp
, stable_how4
*stab_comm
, cred_t
*cr
, bool_t readahead
)
9532 rnode4_t
*rp
= VTOR4(bp
->b_vp
);
9535 cred_t
*cred_otw
= NULL
;
9537 nfs4_open_stream_t
*osp
= NULL
;
9538 bool_t first_time
= TRUE
; /* first time getting otw cred */
9539 bool_t last_time
= FALSE
; /* last time getting otw cred */
9541 ASSERT(nfs_zone() == VTOMI4(bp
->b_vp
)->mi_zone
);
9543 DTRACE_IO1(start
, struct buf
*, bp
);
9544 offset
= ldbtob(bp
->b_lblkno
);
9546 if (bp
->b_flags
& B_READ
) {
9549 * Releases the osp, if it is provided.
9550 * Puts a hold on the cred_otw and the new osp (if found).
9552 cred_otw
= nfs4_get_otw_cred_by_osp(rp
, cr
, &osp
,
9553 &first_time
, &last_time
);
9554 error
= bp
->b_error
= nfs4read(bp
->b_vp
, bp
->b_un
.b_addr
,
9555 offset
, bp
->b_bcount
, &bp
->b_resid
, cred_otw
,
9561 * Didn't get it all because we hit EOF,
9562 * zero all the memory beyond the EOF.
9564 /* bzero(rdaddr + */
9565 bzero(bp
->b_un
.b_addr
+
9566 bp
->b_bcount
- bp
->b_resid
, bp
->b_resid
);
9568 mutex_enter(&rp
->r_statelock
);
9569 if (bp
->b_resid
== bp
->b_bcount
&&
9570 offset
>= rp
->r_size
) {
9572 * We didn't read anything at all as we are
9573 * past EOF. Return an error indicator back
9574 * but don't destroy the pages (yet).
9578 mutex_exit(&rp
->r_statelock
);
9579 } else if (error
== EACCES
&& last_time
== FALSE
) {
9583 if (!(rp
->r_flags
& R4STALE
)) {
9586 * Releases the osp, if it is provided.
9587 * Puts a hold on the cred_otw and the new
9590 cred_otw
= nfs4_get_otw_cred_by_osp(rp
, cr
, &osp
,
9591 &first_time
, &last_time
);
9592 mutex_enter(&rp
->r_statelock
);
9593 count
= MIN(bp
->b_bcount
, rp
->r_size
- offset
);
9594 mutex_exit(&rp
->r_statelock
);
9596 cmn_err(CE_PANIC
, "nfs4_bio: write count < 0");
9599 zoneid_t zoneid
= getzoneid();
9601 zcmn_err(zoneid
, CE_WARN
,
9602 "nfs4_bio: zero length write at %lld",
9604 zcmn_err(zoneid
, CE_CONT
, "flags=0x%x, "
9605 "b_bcount=%ld, file size=%lld",
9606 rp
->r_flags
, (long)bp
->b_bcount
,
9608 sfh4_printfhandle(VTOR4(bp
->b_vp
)->r_fh
);
9609 if (nfs4_bio_do_stop
)
9610 debug_enter("nfs4_bio");
9613 error
= nfs4write(bp
->b_vp
, bp
->b_un
.b_addr
, offset
,
9614 count
, cred_otw
, stab_comm
);
9615 if (error
== EACCES
&& last_time
== FALSE
) {
9619 bp
->b_error
= error
;
9620 if (error
&& error
!= EINTR
&&
9621 !(bp
->b_vp
->v_vfsp
->vfs_flag
& VFS_UNMOUNTED
)) {
9623 * Don't print EDQUOT errors on the console.
9624 * Don't print asynchronous EACCES errors.
9625 * Don't print EFBIG errors.
9626 * Print all other write errors.
9628 if (error
!= EDQUOT
&& error
!= EFBIG
&&
9630 !(bp
->b_flags
& B_ASYNC
)))
9631 nfs4_write_error(bp
->b_vp
,
9634 * Update r_error and r_flags as appropriate.
9635 * If the error was ESTALE, then mark the
9636 * rnode as not being writeable and save
9637 * the error status. Otherwise, save any
9638 * errors which occur from asynchronous
9639 * page invalidations. Any errors occurring
9640 * from other operations should be saved
9643 mutex_enter(&rp
->r_statelock
);
9644 if (error
== ESTALE
) {
9645 rp
->r_flags
|= R4STALE
;
9647 rp
->r_error
= error
;
9648 } else if (!rp
->r_error
&&
9650 (B_INVAL
|B_FORCE
|B_ASYNC
)) ==
9651 (B_INVAL
|B_FORCE
|B_ASYNC
)) {
9652 rp
->r_error
= error
;
9654 mutex_exit(&rp
->r_statelock
);
9658 error
= rp
->r_error
;
9660 * A close may have cleared r_error, if so,
9661 * propagate ESTALE error return properly
9668 if (error
!= 0 && error
!= NFS_EOF
)
9669 bp
->b_flags
|= B_ERROR
;
9672 open_stream_rele(osp
, rp
);
9674 DTRACE_IO1(done
, struct buf
*, bp
);
9681 nfs4_fid(vnode_t
*vp
, fid_t
*fidp
, caller_context_t
*ct
)
9688 nfs4_rwlock(vnode_t
*vp
, int write_lock
, caller_context_t
*ctp
)
9690 rnode4_t
*rp
= VTOR4(vp
);
9693 (void) nfs_rw_enter_sig(&rp
->r_rwlock
, RW_READER
, FALSE
);
9694 return (V_WRITELOCK_FALSE
);
9697 if ((rp
->r_flags
& R4DIRECTIO
) ||
9698 (VTOMI4(vp
)->mi_flags
& MI4_DIRECTIO
)) {
9699 (void) nfs_rw_enter_sig(&rp
->r_rwlock
, RW_READER
, FALSE
);
9700 if (rp
->r_mapcnt
== 0 && !nfs4_has_pages(vp
))
9701 return (V_WRITELOCK_FALSE
);
9702 nfs_rw_exit(&rp
->r_rwlock
);
9705 (void) nfs_rw_enter_sig(&rp
->r_rwlock
, RW_WRITER
, FALSE
);
9706 return (V_WRITELOCK_TRUE
);
9711 nfs4_rwunlock(vnode_t
*vp
, int write_lock
, caller_context_t
*ctp
)
9713 rnode4_t
*rp
= VTOR4(vp
);
9715 nfs_rw_exit(&rp
->r_rwlock
);
9720 nfs4_seek(vnode_t
*vp
, offset_t ooff
, offset_t
*noffp
, caller_context_t
*ct
)
9722 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
9726 * Because we stuff the readdir cookie into the offset field
9727 * someone may attempt to do an lseek with the cookie which
9728 * we want to succeed.
9730 if (vp
->v_type
== VDIR
)
9739 * Return all the pages from [off..off+len) in file
9743 nfs4_getpage(vnode_t
*vp
, offset_t off
, size_t len
, uint_t
*protp
,
9744 page_t
*pl
[], size_t plsz
, struct seg
*seg
, caddr_t addr
,
9745 enum seg_rw rw
, cred_t
*cr
, caller_context_t
*ct
)
9751 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
9754 if (IS_SHADOW(vp
, rp
))
9757 if (vp
->v_flag
& VNOMAP
)
9764 * Now validate that the caches are up to date.
9766 if (error
= nfs4_validate_caches(vp
, cr
))
9771 mutex_enter(&rp
->r_statelock
);
9774 * Don't create dirty pages faster than they
9775 * can be cleaned so that the system doesn't
9776 * get imbalanced. If the async queue is
9777 * maxed out, then wait for it to drain before
9778 * creating more dirty pages. Also, wait for
9779 * any threads doing pagewalks in the vop_getattr
9780 * entry points so that they don't block for
9783 if (rw
== S_CREATE
) {
9784 while ((mi
->mi_max_threads
!= 0 &&
9785 rp
->r_awcount
> 2 * mi
->mi_max_threads
) ||
9787 cv_wait(&rp
->r_cv
, &rp
->r_statelock
);
9791 * If we are getting called as a side effect of an nfs_write()
9792 * operation the local file size might not be extended yet.
9793 * In this case we want to be able to return pages of zeroes.
9795 if (off
+ len
> rp
->r_size
+ PAGEOFFSET
&& seg
!= segkmap
) {
9796 NFS4_DEBUG(nfs4_pageio_debug
,
9797 (CE_NOTE
, "getpage beyond EOF: off=%lld, "
9798 "len=%llu, size=%llu, attrsize =%llu", off
,
9799 (u_longlong_t
)len
, rp
->r_size
, rp
->r_attr
.va_size
));
9800 mutex_exit(&rp
->r_statelock
);
9801 return (EFAULT
); /* beyond EOF */
9804 mutex_exit(&rp
->r_statelock
);
9806 error
= pvn_getpages(nfs4_getapage
, vp
, off
, len
, protp
,
9807 pl
, plsz
, seg
, addr
, rw
, cr
);
9808 NFS4_DEBUG(nfs4_pageio_debug
&& error
,
9809 (CE_NOTE
, "getpages error %d; off=%lld, len=%lld",
9810 error
, off
, (u_longlong_t
)len
));
9814 nfs4_purge_caches(vp
, NFS4_NOPURGE_DNLC
, cr
, FALSE
);
9817 nfs4_purge_stale_fh(error
, vp
, cr
);
9824 * Called from pvn_getpages to get a particular page.
9828 nfs4_getapage(vnode_t
*vp
, uoff_t off
, size_t len
, uint_t
*protp
,
9829 page_t
*pl
[], size_t plsz
, struct seg
*seg
, caddr_t addr
,
9830 enum seg_rw rw
, cred_t
*cr
)
9844 int readahead_issued
= 0;
9845 int ra_window
; /* readahead window */
9849 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
9853 ASSERT(!IS_SHADOW(vp
, rp
));
9854 bsize
= MAX(vp
->v_vfsp
->vfs_bsize
, PAGESIZE
);
9866 blkoff
= lbn
* bsize
;
9869 * Queueing up the readahead before doing the synchronous read
9870 * results in a significant increase in read throughput because
9871 * of the increased parallelism between the async threads and
9872 * the process context.
9874 if ((off
& ((vp
->v_vfsp
->vfs_bsize
) - 1)) == 0 &&
9876 !(vp
->v_flag
& VNOCACHE
)) {
9877 mutex_enter(&rp
->r_statelock
);
9880 * Calculate the number of readaheads to do.
9881 * a) No readaheads at offset = 0.
9882 * b) Do maximum(nfs4_nra) readaheads when the readahead
9884 * c) Do readaheads between 1 to (nfs4_nra - 1) depending
9885 * upon how far the readahead window is open or close.
9886 * d) No readaheads if rp->r_nextr is not within the scope
9887 * of the readahead window (random i/o).
9892 else if (blkoff
== rp
->r_nextr
)
9893 readahead
= nfs4_nra
;
9894 else if (rp
->r_nextr
> blkoff
&&
9895 ((ra_window
= (rp
->r_nextr
- blkoff
) / bsize
)
9897 readahead
= nfs4_nra
- ra_window
;
9901 rablkoff
= rp
->r_nextr
;
9902 while (readahead
> 0 && rablkoff
+ bsize
< rp
->r_size
) {
9903 mutex_exit(&rp
->r_statelock
);
9904 if (nfs4_async_readahead(vp
, rablkoff
+ bsize
,
9905 addr
+ (rablkoff
+ bsize
- off
),
9906 seg
, cr
, nfs4_readahead
) < 0) {
9907 mutex_enter(&rp
->r_statelock
);
9913 * Indicate that we did a readahead so
9914 * readahead offset is not updated
9915 * by the synchronous read below.
9917 readahead_issued
= 1;
9918 mutex_enter(&rp
->r_statelock
);
9920 * set readahead offset to
9921 * offset of last async readahead
9924 rp
->r_nextr
= rablkoff
;
9926 mutex_exit(&rp
->r_statelock
);
9930 if ((pagefound
= page_exists(vp
, off
)) == NULL
) {
9932 (void) nfs4_async_readahead(vp
, blkoff
, addr
, seg
, cr
,
9934 } else if (rw
== S_CREATE
) {
9936 * Block for this page is not allocated, or the offset
9937 * is beyond the current allocation size, or we're
9938 * allocating a swap slot and the page was not found,
9939 * so allocate it and return a zero page.
9941 if ((pp
= page_create_va(vp
, off
,
9942 PAGESIZE
, PG_WAIT
, seg
, addr
)) == NULL
)
9943 cmn_err(CE_PANIC
, "nfs4_getapage: page_create");
9945 mutex_enter(&rp
->r_statelock
);
9946 rp
->r_nextr
= off
+ PAGESIZE
;
9947 mutex_exit(&rp
->r_statelock
);
9950 * Need to go to server to get a block
9952 mutex_enter(&rp
->r_statelock
);
9953 if (blkoff
< rp
->r_size
&&
9954 blkoff
+ bsize
> rp
->r_size
) {
9956 * If less than a block left in
9957 * file read less than a block.
9959 if (rp
->r_size
<= off
) {
9961 * Trying to access beyond EOF,
9962 * set up to get at least one page.
9964 blksize
= off
+ PAGESIZE
- blkoff
;
9966 blksize
= rp
->r_size
- blkoff
;
9967 } else if ((off
== 0) ||
9968 (off
!= rp
->r_nextr
&& !readahead_issued
)) {
9970 blkoff
= off
; /* block = page here */
9973 mutex_exit(&rp
->r_statelock
);
9975 pp
= pvn_read_kluster(vp
, off
, seg
, addr
, &io_off
,
9976 &io_len
, blkoff
, blksize
, 0);
9979 * Some other thread has entered the page,
9986 * Now round the request size up to page boundaries.
9987 * This ensures that the entire page will be
9988 * initialized to zeroes if EOF is encountered.
9990 io_len
= ptob(btopr(io_len
));
9992 bp
= pageio_setup(pp
, io_len
, vp
, B_READ
);
9996 * pageio_setup should have set b_addr to 0. This
9997 * is correct since we want to do I/O on a page
9998 * boundary. bp_mapin will use this addr to calculate
9999 * an offset, and then set b_addr to the kernel virtual
10000 * address it allocated for us.
10002 ASSERT(bp
->b_un
.b_addr
== 0);
10006 bp
->b_lblkno
= lbtodb(io_off
);
10008 bp
->b_offset
= (offset_t
)off
;
10012 * If doing a write beyond what we believe is EOF,
10013 * don't bother trying to read the pages from the
10014 * server, we'll just zero the pages here. We
10015 * don't check that the rw flag is S_WRITE here
10016 * because some implementations may attempt a
10017 * read access to the buffer before copying data.
10019 mutex_enter(&rp
->r_statelock
);
10020 if (io_off
>= rp
->r_size
&& seg
== segkmap
) {
10021 mutex_exit(&rp
->r_statelock
);
10022 bzero(bp
->b_un
.b_addr
, io_len
);
10024 mutex_exit(&rp
->r_statelock
);
10025 error
= nfs4_bio(bp
, NULL
, cr
, FALSE
);
10029 * Unmap the buffer before freeing it.
10036 pp
->p_fsdata
= C_NOCOMMIT
;
10037 } while ((pp
= pp
->p_next
) != savepp
);
10039 if (error
== NFS_EOF
) {
10041 * If doing a write system call just return
10042 * zeroed pages, else user tried to get pages
10043 * beyond EOF, return error. We don't check
10044 * that the rw flag is S_WRITE here because
10045 * some implementations may attempt a read
10046 * access to the buffer before copying data.
10048 if (seg
== segkmap
)
10054 if (!readahead_issued
&& !error
) {
10055 mutex_enter(&rp
->r_statelock
);
10056 rp
->r_nextr
= io_off
+ io_len
;
10057 mutex_exit(&rp
->r_statelock
);
10068 pvn_read_done(pp
, B_ERROR
);
10073 se_t se
= (rw
== S_CREATE
? SE_EXCL
: SE_SHARED
);
10076 * Page exists in the cache, acquire the appropriate lock.
10077 * If this fails, start all over again.
10079 if ((pp
= page_lookup(vp
, off
, se
)) == NULL
) {
10091 pvn_plist_init(pp
, pl
, plsz
, off
, io_len
, rw
);
10097 nfs4_readahead(vnode_t
*vp
, uoff_t blkoff
, caddr_t addr
, struct seg
*seg
,
10105 uint_t bsize
, blksize
;
10106 rnode4_t
*rp
= VTOR4(vp
);
10109 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
10111 bsize
= MAX(vp
->v_vfsp
->vfs_bsize
, PAGESIZE
);
10113 mutex_enter(&rp
->r_statelock
);
10114 if (blkoff
< rp
->r_size
&& blkoff
+ bsize
> rp
->r_size
) {
10116 * If less than a block left in file read less
10119 blksize
= rp
->r_size
- blkoff
;
10122 mutex_exit(&rp
->r_statelock
);
10124 pp
= pvn_read_kluster(vp
, blkoff
, segkmap
, addr
,
10125 &io_off
, &io_len
, blkoff
, blksize
, 1);
10127 * The isra flag passed to the kluster function is 1, we may have
10128 * gotten a return value of NULL for a variety of reasons (# of free
10129 * pages < minfree, someone entered the page on the vnode etc). In all
10130 * cases, we want to punt on the readahead.
10136 * Now round the request size up to page boundaries.
10137 * This ensures that the entire page will be
10138 * initialized to zeroes if EOF is encountered.
10140 io_len
= ptob(btopr(io_len
));
10142 bp
= pageio_setup(pp
, io_len
, vp
, B_READ
);
10143 ASSERT(bp
!= NULL
);
10146 * pageio_setup should have set b_addr to 0. This is correct since
10147 * we want to do I/O on a page boundary. bp_mapin() will use this addr
10148 * to calculate an offset, and then set b_addr to the kernel virtual
10149 * address it allocated for us.
10151 ASSERT(bp
->b_un
.b_addr
== 0);
10155 bp
->b_lblkno
= lbtodb(io_off
);
10157 bp
->b_offset
= (offset_t
)blkoff
;
10161 * If doing a write beyond what we believe is EOF, don't bother trying
10162 * to read the pages from the server, we'll just zero the pages here.
10163 * We don't check that the rw flag is S_WRITE here because some
10164 * implementations may attempt a read access to the buffer before
10167 mutex_enter(&rp
->r_statelock
);
10168 if (io_off
>= rp
->r_size
&& seg
== segkmap
) {
10169 mutex_exit(&rp
->r_statelock
);
10170 bzero(bp
->b_un
.b_addr
, io_len
);
10173 mutex_exit(&rp
->r_statelock
);
10174 error
= nfs4_bio(bp
, NULL
, cr
, TRUE
);
10175 if (error
== NFS_EOF
)
10180 * Unmap the buffer before freeing it.
10187 pp
->p_fsdata
= C_NOCOMMIT
;
10188 } while ((pp
= pp
->p_next
) != savepp
);
10190 pvn_read_done(pp
, error
? B_READ
| B_ERROR
: B_READ
);
10193 * In case of error set readahead offset
10194 * to the lowest offset.
10195 * pvn_read_done() calls VN_DISPOSE to destroy the pages
10197 if (error
&& rp
->r_nextr
> io_off
) {
10198 mutex_enter(&rp
->r_statelock
);
10199 if (rp
->r_nextr
> io_off
)
10200 rp
->r_nextr
= io_off
;
10201 mutex_exit(&rp
->r_statelock
);
10206 * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE}
10207 * If len == 0, do from off to EOF.
10209 * The normal cases should be len == 0 && off == 0 (entire vp list) or
10210 * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
10215 nfs4_putpage(vnode_t
*vp
, offset_t off
, size_t len
, int flags
, cred_t
*cr
,
10216 caller_context_t
*ct
)
10221 ASSERT(cr
!= NULL
);
10223 if (!(flags
& B_ASYNC
) && nfs_zone() != VTOMI4(vp
)->mi_zone
)
10227 if (IS_SHADOW(vp
, rp
))
10231 * XXX - Why should this check be made here?
10233 if (vp
->v_flag
& VNOMAP
)
10236 if (len
== 0 && !(flags
& B_INVAL
) &&
10237 (vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
))
10240 mutex_enter(&rp
->r_statelock
);
10242 mutex_exit(&rp
->r_statelock
);
10243 error
= nfs4_putpages(vp
, off
, len
, flags
, cr
);
10244 mutex_enter(&rp
->r_statelock
);
10246 cv_broadcast(&rp
->r_cv
);
10247 mutex_exit(&rp
->r_statelock
);
10253 * Write out a single page, possibly klustering adjacent dirty pages.
10256 nfs4_putapage(vnode_t
*vp
, page_t
*pp
, uoff_t
*offp
, size_t *lenp
,
10257 int flags
, cred_t
*cr
)
10267 ASSERT(!(vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
));
10268 ASSERT(pp
!= NULL
);
10269 ASSERT(cr
!= NULL
);
10270 ASSERT((flags
& B_ASYNC
) || nfs_zone() == VTOMI4(vp
)->mi_zone
);
10273 ASSERT(rp
->r_count
> 0);
10274 ASSERT(!IS_SHADOW(vp
, rp
));
10276 bsize
= MAX(vp
->v_vfsp
->vfs_bsize
, PAGESIZE
);
10277 lbn
= pp
->p_offset
/ bsize
;
10278 lbn_off
= lbn
* bsize
;
10281 * Find a kluster that fits in one block, or in
10282 * one page if pages are bigger than blocks. If
10283 * there is less file space allocated than a whole
10284 * page, we'll shorten the i/o request below.
10286 pp
= pvn_write_kluster(vp
, pp
, &io_off
, &io_len
, lbn_off
,
10287 roundup(bsize
, PAGESIZE
), flags
);
10290 * pvn_write_kluster shouldn't have returned a page with offset
10291 * behind the original page we were given. Verify that.
10293 ASSERT((pp
->p_offset
/ bsize
) >= lbn
);
10296 * Now pp will have the list of kept dirty pages marked for
10297 * write back. It will also handle invalidation and freeing
10298 * of pages that are not dirty. Check for page length rounding
10301 if (io_off
+ io_len
> lbn_off
+ bsize
) {
10302 ASSERT((io_off
+ io_len
) - (lbn_off
+ bsize
) < PAGESIZE
);
10303 io_len
= lbn_off
+ bsize
- io_off
;
10306 * The R4MODINPROGRESS flag makes sure that nfs4_bio() sees a
10307 * consistent value of r_size. R4MODINPROGRESS is set in writerp4().
10308 * When R4MODINPROGRESS is set it indicates that a uiomove() is in
10309 * progress and the r_size has not been made consistent with the
10310 * new size of the file. When the uiomove() completes the r_size is
10311 * updated and the R4MODINPROGRESS flag is cleared.
10313 * The R4MODINPROGRESS flag makes sure that nfs4_bio() sees a
10314 * consistent value of r_size. Without this handshaking, it is
10315 * possible that nfs4_bio() picks up the old value of r_size
10316 * before the uiomove() in writerp4() completes. This will result
10317 * in the write through nfs4_bio() being dropped.
10319 * More precisely, there is a window between the time the uiomove()
10320 * completes and the time the r_size is updated. If a fop_putpage()
10321 * operation intervenes in this window, the page will be picked up,
10322 * because it is dirty (it will be unlocked, unless it was
10323 * pagecreate'd). When the page is picked up as dirty, the dirty
10324 * bit is reset (pvn_getdirty()). In nfs4write(), r_size is
10325 * checked. This will still be the old size. Therefore the page will
10326 * not be written out. When segmap_release() calls fop_putpage(),
10327 * the page will be found to be clean and the write will be dropped.
10329 if (rp
->r_flags
& R4MODINPROGRESS
) {
10330 mutex_enter(&rp
->r_statelock
);
10331 if ((rp
->r_flags
& R4MODINPROGRESS
) &&
10332 rp
->r_modaddr
+ MAXBSIZE
> io_off
&&
10333 rp
->r_modaddr
< io_off
+ io_len
) {
10336 * A write is in progress for this region of the file.
10337 * If we did not detect R4MODINPROGRESS here then this
10338 * path through nfs_putapage() would eventually go to
10339 * nfs4_bio() and may not write out all of the data
10340 * in the pages. We end up losing data. So we decide
10341 * to set the modified bit on each page in the page
10342 * list and mark the rnode with R4DIRTY. This write
10343 * will be restarted at some later time.
10346 while (plist
!= NULL
) {
10348 page_sub(&plist
, pp
);
10350 page_io_unlock(pp
);
10353 rp
->r_flags
|= R4DIRTY
;
10354 mutex_exit(&rp
->r_statelock
);
10361 mutex_exit(&rp
->r_statelock
);
10364 if (flags
& B_ASYNC
) {
10365 error
= nfs4_async_putapage(vp
, pp
, io_off
, io_len
, flags
, cr
,
10366 nfs4_sync_putapage
);
10368 error
= nfs4_sync_putapage(vp
, pp
, io_off
, io_len
, flags
, cr
);
10378 nfs4_sync_putapage(vnode_t
*vp
, page_t
*pp
, uoff_t io_off
, size_t io_len
,
10379 int flags
, cred_t
*cr
)
10384 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
10388 error
= nfs4_rdwrlbn(vp
, pp
, io_off
, io_len
, flags
, cr
);
10392 if ((error
== ENOSPC
|| error
== EDQUOT
|| error
== EFBIG
||
10393 error
== EACCES
) &&
10394 (flags
& (B_INVAL
|B_FORCE
)) != (B_INVAL
|B_FORCE
)) {
10395 if (!(rp
->r_flags
& R4OUTOFSPACE
)) {
10396 mutex_enter(&rp
->r_statelock
);
10397 rp
->r_flags
|= R4OUTOFSPACE
;
10398 mutex_exit(&rp
->r_statelock
);
10401 pvn_write_done(pp
, flags
);
10403 * If this was not an async thread, then try again to
10404 * write out the pages, but this time, also destroy
10405 * them whether or not the write is successful. This
10406 * will prevent memory from filling up with these
10407 * pages and destroying them is the only alternative
10408 * if they can't be written out.
10410 * Don't do this if this is an async thread because
10411 * when the pages are unlocked in pvn_write_done,
10412 * some other thread could have come along, locked
10413 * them, and queued for an async thread. It would be
10414 * possible for all of the async threads to be tied
10415 * up waiting to lock the pages again and they would
10416 * all already be locked and waiting for an async
10417 * thread to handle them. Deadlock.
10419 if (!(flags
& B_ASYNC
)) {
10420 error
= nfs4_putpage(vp
, io_off
, io_len
,
10421 B_INVAL
| B_FORCE
, cr
, NULL
);
10426 else if (rp
->r_flags
& R4OUTOFSPACE
) {
10427 mutex_enter(&rp
->r_statelock
);
10428 rp
->r_flags
&= ~R4OUTOFSPACE
;
10429 mutex_exit(&rp
->r_statelock
);
10431 pvn_write_done(pp
, flags
);
10432 if (freemem
< desfree
)
10433 (void) nfs4_commit_vp(vp
, 0, 0, cr
,
10434 NFS4_WRITE_NOWAIT
);
10441 int nfs4_force_open_before_mmap
= 0;
10446 nfs4_map(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t
*addrp
,
10447 size_t len
, uchar_t prot
, uchar_t maxprot
, uint_t flags
, cred_t
*cr
,
10448 caller_context_t
*ct
)
10450 struct segvn_crargs vn_a
;
10452 rnode4_t
*rp
= VTOR4(vp
);
10453 mntinfo4_t
*mi
= VTOMI4(vp
);
10455 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
10458 if (vp
->v_flag
& VNOMAP
)
10461 if (off
< 0 || (off
+ len
) < 0)
10464 if (vp
->v_type
!= VREG
)
10468 * If the file is delegated to the client don't do anything.
10469 * If the file is not delegated, then validate the data cache.
10471 mutex_enter(&rp
->r_statev4_lock
);
10472 if (rp
->r_deleg_type
== OPEN_DELEGATE_NONE
) {
10473 mutex_exit(&rp
->r_statev4_lock
);
10474 error
= nfs4_validate_caches(vp
, cr
);
10478 mutex_exit(&rp
->r_statev4_lock
);
10482 * Check to see if the vnode is currently marked as not cachable.
10483 * This means portions of the file are locked (through fop_frlock).
10484 * In this case the map request must be refused. We use
10485 * rp->r_lkserlock to avoid a race with concurrent lock requests.
10487 * Atomically increment r_inmap after acquiring r_rwlock. The
10488 * idea here is to acquire r_rwlock to block read/write and
10489 * not to protect r_inmap. r_inmap will inform nfs4_read/write()
10490 * that we are in nfs4_map(). Now, r_rwlock is acquired in order
10491 * and we can prevent the deadlock that would have occurred
10492 * when nfs4_addmap() would have acquired it out of order.
10494 * Since we are not protecting r_inmap by any lock, we do not
10495 * hold any lock when we decrement it. We atomically decrement
10496 * r_inmap after we release r_lkserlock.
10499 if (nfs_rw_enter_sig(&rp
->r_rwlock
, RW_WRITER
, INTR4(vp
)))
10501 atomic_inc_uint(&rp
->r_inmap
);
10502 nfs_rw_exit(&rp
->r_rwlock
);
10504 if (nfs_rw_enter_sig(&rp
->r_lkserlock
, RW_READER
, INTR4(vp
))) {
10505 atomic_dec_uint(&rp
->r_inmap
);
10509 if (vp
->v_flag
& VNOCACHE
) {
10515 * Don't allow concurrent locks and mapping if mandatory locking is
10518 if (flk_has_remote_locks(vp
)) {
10520 va
.va_mask
= AT_MODE
;
10521 error
= nfs4getattr(vp
, &va
, cr
);
10524 if (MANDLOCK(vp
, va
.va_mode
)) {
10531 * It is possible that the rnode has a lost lock request that we
10532 * are still trying to recover, and that the request conflicts with
10533 * this map request.
10535 * An alternative approach would be for nfs4_safemap() to consider
10536 * queued lock requests when deciding whether to set or clear
10537 * VNOCACHE. This would require the frlock code path to call
10538 * nfs4_safemap() after enqueing a lost request.
10540 if (nfs4_map_lost_lock_conflict(vp
)) {
10546 error
= choose_addr(as
, addrp
, len
, off
, ADDR_VACALIGN
, flags
);
10548 as_rangeunlock(as
);
10552 if (vp
->v_type
== VREG
) {
10554 * We need to retrieve the open stream
10556 nfs4_open_stream_t
*osp
= NULL
;
10557 nfs4_open_owner_t
*oop
= NULL
;
10559 oop
= find_open_owner(cr
, NFS4_PERM_CREATED
, mi
);
10561 /* returns with 'os_sync_lock' held */
10562 osp
= find_open_stream(oop
, rp
);
10563 open_owner_rele(oop
);
10567 if (nfs4_force_open_before_mmap
) {
10572 /* returns with 'os_sync_lock' held */
10573 error
= open_and_get_osp(vp
, cr
, &osp
);
10575 NFS4_DEBUG(nfs4_mmap_debug
, (CE_NOTE
,
10576 "nfs4_map: we tried to OPEN the file "
10577 "but again no osp, so fail with EIO"));
10582 if (osp
->os_failed_reopen
) {
10583 mutex_exit(&osp
->os_sync_lock
);
10584 open_stream_rele(osp
, rp
);
10585 NFS4_DEBUG(nfs4_open_stream_debug
, (CE_NOTE
,
10586 "nfs4_map: os_failed_reopen set on "
10587 "osp %p, cr %p, rp %s", (void *)osp
,
10588 (void *)cr
, rnode4info(rp
)));
10592 mutex_exit(&osp
->os_sync_lock
);
10593 open_stream_rele(osp
, rp
);
10598 vn_a
.type
= (flags
& MAP_TYPE
);
10599 vn_a
.prot
= (uchar_t
)prot
;
10600 vn_a
.maxprot
= (uchar_t
)maxprot
;
10601 vn_a
.flags
= (flags
& ~MAP_TYPE
);
10605 vn_a
.lgrp_mem_policy_flags
= 0;
10607 error
= as_map(as
, *addrp
, len
, segvn_create
, &vn_a
);
10608 as_rangeunlock(as
);
10611 nfs_rw_exit(&rp
->r_lkserlock
);
10612 atomic_dec_uint(&rp
->r_inmap
);
10617 * We're most likely dealing with a kernel module that likes to READ
10618 * and mmap without OPENing the file (ie: lookup/read/mmap), so lets
10619 * officially OPEN the file to create the necessary client state
10620 * for bookkeeping of os_mmap_read/write counts.
10622 * Since fop_map only passes in a pointer to the vnode rather than
10623 * a double pointer, we can't handle the case where nfs4open_otw()
10624 * returns a different vnode than the one passed into fop_map (since
10625 * fop_delmap will not see the vnode nfs4open_otw used). In this case,
10626 * we return NULL and let nfs4_map() fail. Note: the only case where
10627 * this should happen is if the file got removed and replaced with the
10628 * same name on the server (in addition to the fact that we're trying
10629 * to fop_map withouth fop_opening the file in the first place).
10632 open_and_get_osp(vnode_t
*map_vp
, cred_t
*cr
, nfs4_open_stream_t
**ospp
)
10634 rnode4_t
*rp
, *drp
;
10635 vnode_t
*dvp
, *open_vp
;
10636 char file_name
[MAXNAMELEN
];
10638 nfs4_open_stream_t
*osp
;
10639 nfs4_open_owner_t
*oop
;
10645 rp
= VTOR4(open_vp
);
10646 if ((error
= vtodv(open_vp
, &dvp
, cr
, TRUE
)) != 0)
10650 if (nfs_rw_enter_sig(&drp
->r_rwlock
, RW_READER
, INTR4(dvp
))) {
10655 if ((error
= vtoname(open_vp
, file_name
, MAXNAMELEN
)) != 0) {
10656 nfs_rw_exit(&drp
->r_rwlock
);
10661 mutex_enter(&rp
->r_statev4_lock
);
10662 if (rp
->created_v4
) {
10663 rp
->created_v4
= 0;
10664 mutex_exit(&rp
->r_statev4_lock
);
10666 dnlc_update(dvp
, file_name
, open_vp
);
10667 /* This is needed so we don't bump the open ref count */
10670 mutex_exit(&rp
->r_statev4_lock
);
10676 error
= nfs4open_otw(dvp
, file_name
, NULL
, &open_vp
, cr
, 0, FREAD
, 0,
10679 nfs_rw_exit(&drp
->r_rwlock
);
10685 nfs_rw_exit(&drp
->r_rwlock
);
10689 * If nfs4open_otw() returned a different vnode then "undo"
10690 * the open and return failure to the caller.
10692 if (!VN_CMP(open_vp
, map_vp
)) {
10695 NFS4_DEBUG(nfs4_mmap_debug
, (CE_NOTE
, "open_and_get_osp: "
10696 "open returned a different vnode"));
10698 * If there's an error, ignore it,
10699 * and let fop_inactive handle it.
10701 (void) nfs4close_one(open_vp
, NULL
, cr
, FREAD
, NULL
, &e
,
10702 CLOSE_NORM
, 0, 0, 0);
10709 oop
= find_open_owner(cr
, NFS4_PERM_CREATED
, VTOMI4(open_vp
));
10713 NFS4_DEBUG(nfs4_mmap_debug
, (CE_NOTE
, "open_and_get_osp: "
10716 * If there's an error, ignore it,
10717 * and let fop_inactive handle it.
10719 (void) nfs4close_one(open_vp
, NULL
, cr
, FREAD
, NULL
, &e
,
10720 CLOSE_NORM
, 0, 0, 0);
10723 osp
= find_open_stream(oop
, rp
);
10724 open_owner_rele(oop
);
10730 * Please be aware that when this function is called, the address space write
10731 * a_lock is held. Do not put over the wire calls in this function.
10735 nfs4_addmap(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t addr
,
10736 size_t len
, uchar_t prot
, uchar_t maxprot
, uint_t flags
, cred_t
*cr
,
10737 caller_context_t
*ct
)
10746 if (nfs_zone() != mi
->mi_zone
)
10748 if (vp
->v_flag
& VNOMAP
)
10752 * Don't need to update the open stream first, since this
10753 * mmap can't add any additional share access that isn't
10754 * already contained in the open stream (for the case where we
10755 * open/mmap/only update rp->r_mapcnt/server reboots/reopen doesn't
10756 * take into account os_mmap_read[write] counts).
10758 atomic_add_long((ulong_t
*)&rp
->r_mapcnt
, btopr(len
));
10760 if (vp
->v_type
== VREG
) {
10762 * We need to retrieve the open stream and update the counts.
10763 * If there is no open stream here, something is wrong.
10765 nfs4_open_stream_t
*osp
= NULL
;
10766 nfs4_open_owner_t
*oop
= NULL
;
10768 oop
= find_open_owner(cr
, NFS4_PERM_CREATED
, mi
);
10770 /* returns with 'os_sync_lock' held */
10771 osp
= find_open_stream(oop
, rp
);
10772 open_owner_rele(oop
);
10775 NFS4_DEBUG(nfs4_mmap_debug
, (CE_NOTE
,
10776 "nfs4_addmap: we should have an osp"
10777 "but we don't, so fail with EIO"));
10782 NFS4_DEBUG(nfs4_mmap_debug
, (CE_NOTE
, "nfs4_addmap: osp %p,"
10783 " pages %ld, prot 0x%x", (void *)osp
, btopr(len
), prot
));
10786 * Update the map count in the open stream.
10787 * This is necessary in the case where we
10788 * open/mmap/close/, then the server reboots, and we
10789 * attempt to reopen. If the mmap doesn't add share
10790 * access then we send an invalid reopen with
10793 * We need to specifically check each PROT_* so a mmap
10794 * call of (PROT_WRITE | PROT_EXEC) will ensure us both
10795 * read and write access. A simple comparison of prot
10796 * to ~PROT_WRITE to determine read access is insufficient
10797 * since prot can be |= with PROT_USER, etc.
10801 * Unless we're MAP_SHARED, no sense in adding os_mmap_write
10803 if ((flags
& MAP_SHARED
) && (maxprot
& PROT_WRITE
))
10804 osp
->os_mmap_write
+= btopr(len
);
10805 if (maxprot
& PROT_READ
)
10806 osp
->os_mmap_read
+= btopr(len
);
10807 if (maxprot
& PROT_EXEC
)
10808 osp
->os_mmap_read
+= btopr(len
);
10810 * Ensure that os_mmap_read gets incremented, even if
10811 * maxprot were to look like PROT_NONE.
10813 if (!(maxprot
& PROT_READ
) && !(maxprot
& PROT_WRITE
) &&
10814 !(maxprot
& PROT_EXEC
))
10815 osp
->os_mmap_read
+= btopr(len
);
10816 osp
->os_mapcnt
+= btopr(len
);
10817 mutex_exit(&osp
->os_sync_lock
);
10818 open_stream_rele(osp
, rp
);
10823 * If we got an error, then undo our
10824 * incrementing of 'r_mapcnt'.
10828 atomic_add_long((ulong_t
*)&rp
->r_mapcnt
, -btopr(len
));
10829 ASSERT(rp
->r_mapcnt
>= 0);
10836 nfs4_cmp(vnode_t
*vp1
, vnode_t
*vp2
, caller_context_t
*ct
)
10839 return (VTOR4(vp1
) == VTOR4(vp2
));
10844 nfs4_frlock(vnode_t
*vp
, int cmd
, struct flock64
*bfp
, int flag
,
10845 offset_t offset
, struct flk_callback
*flk_cbp
, cred_t
*cr
,
10846 caller_context_t
*ct
)
10851 int error
= 0, intr
= INTR4(vp
);
10854 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
10857 /* check for valid cmd parameter */
10858 if (cmd
!= F_GETLK
&& cmd
!= F_SETLK
&& cmd
!= F_SETLKW
)
10861 /* Verify l_type. */
10862 switch (bfp
->l_type
) {
10864 if (cmd
!= F_GETLK
&& !(flag
& FREAD
))
10868 if (cmd
!= F_GETLK
&& !(flag
& FWRITE
))
10879 /* check the validity of the lock range */
10880 if (rc
= flk_convert_lock_data(vp
, bfp
, &start
, &end
, offset
))
10882 if (rc
= flk_check_lock_data(start
, end
, MAXEND
))
10886 * If the filesystem is mounted using local locking, pass the
10887 * request off to the local locking code.
10889 if (VTOMI4(vp
)->mi_flags
& MI4_LLOCK
|| vp
->v_type
!= VREG
) {
10890 if (cmd
== F_SETLK
|| cmd
== F_SETLKW
) {
10892 * For complete safety, we should be holding
10893 * r_lkserlock. However, we can't call
10894 * nfs4_safelock and then fs_frlock while
10895 * holding r_lkserlock, so just invoke
10896 * nfs4_safelock and expect that this will
10897 * catch enough of the cases.
10899 if (!nfs4_safelock(vp
, bfp
, cr
))
10902 return (fs_frlock(vp
, cmd
, bfp
, flag
, offset
, flk_cbp
, cr
, ct
));
10908 * Check whether the given lock request can proceed, given the
10909 * current file mappings.
10911 if (nfs_rw_enter_sig(&rp
->r_lkserlock
, RW_WRITER
, intr
))
10913 if (cmd
== F_SETLK
|| cmd
== F_SETLKW
) {
10914 if (!nfs4_safelock(vp
, bfp
, cr
)) {
10921 * Flush the cache after waiting for async I/O to finish. For new
10922 * locks, this is so that the process gets the latest bits from the
10923 * server. For unlocks, this is so that other clients see the
10924 * latest bits once the file has been unlocked. If currently dirty
10925 * pages can't be flushed, then don't allow a lock to be set. But
10926 * allow unlocks to succeed, to avoid having orphan locks on the
10929 if (cmd
!= F_GETLK
) {
10930 mutex_enter(&rp
->r_statelock
);
10931 while (rp
->r_count
> 0) {
10933 klwp_t
*lwp
= ttolwp(curthread
);
10937 if (cv_wait_sig(&rp
->r_cv
,
10938 &rp
->r_statelock
) == 0) {
10947 cv_wait(&rp
->r_cv
, &rp
->r_statelock
);
10950 mutex_exit(&rp
->r_statelock
);
10953 error
= nfs4_putpage(vp
, (offset_t
)0, 0, B_INVAL
, cr
, ct
);
10955 if (error
== ENOSPC
|| error
== EDQUOT
) {
10956 mutex_enter(&rp
->r_statelock
);
10958 rp
->r_error
= error
;
10959 mutex_exit(&rp
->r_statelock
);
10961 if (bfp
->l_type
!= F_UNLCK
) {
10969 * Call the lock manager to do the real work of contacting
10970 * the server and obtaining the lock.
10972 nfs4frlock(NFS4_LCK_CTYPE_NORM
, vp
, cmd
, bfp
, flag
, offset
,
10973 cr
, &e
, NULL
, NULL
);
10977 nfs4_lockcompletion(vp
, cmd
);
10980 nfs_rw_exit(&rp
->r_lkserlock
);
10986 * Free storage space associated with the specified vnode. The portion
10987 * to be freed is specified by bfp->l_start and bfp->l_len (already
10988 * normalized to a "whence" of 0).
10990 * This is an experimental facility whose continued existence is not
10991 * guaranteed. Currently, we only support the special case
10992 * of l_len == 0, meaning free to end of file.
10996 nfs4_space(vnode_t
*vp
, int cmd
, struct flock64
*bfp
, int flag
,
10997 offset_t offset
, cred_t
*cr
, caller_context_t
*ct
)
11001 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
11003 ASSERT(vp
->v_type
== VREG
);
11004 if (cmd
!= F_FREESP
)
11007 error
= convoff(vp
, bfp
, 0, offset
);
11009 ASSERT(bfp
->l_start
>= 0);
11010 if (bfp
->l_len
== 0) {
11013 va
.va_mask
= AT_SIZE
;
11014 va
.va_size
= bfp
->l_start
;
11015 error
= nfs4setattr(vp
, &va
, 0, cr
, NULL
);
11017 if (error
== 0 && bfp
->l_start
== 0)
11018 vnevent_truncate(vp
, ct
);
11028 nfs4_realvp(vnode_t
*vp
, vnode_t
**vpp
, caller_context_t
*ct
)
11033 if (vp
->v_type
== VREG
&& IS_SHADOW(vp
, rp
)) {
11041 * Setup and add an address space callback to do the work of the delmap call.
11042 * The callback will (and must be) deleted in the actual callback function.
11044 * This is done in order to take care of the problem that we have with holding
11045 * the address space's a_lock for a long period of time (e.g. if the NFS server
11046 * is down). Callbacks will be executed in the address space code while the
11047 * a_lock is not held. Holding the address space's a_lock causes things such
11048 * as ps and fork to hang because they are trying to acquire this lock as well.
11052 nfs4_delmap(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t addr
,
11053 size_t len
, uint_t prot
, uint_t maxprot
, uint_t flags
, cred_t
*cr
,
11054 caller_context_t
*ct
)
11059 nfs4_delmap_args_t
*dmapp
;
11060 nfs4_delmapcall_t
*delmap_call
;
11062 if (vp
->v_flag
& VNOMAP
)
11066 * A process may not change zones if it has NFS pages mmap'ed
11067 * in, so we can't legitimately get here from the wrong zone.
11069 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
11074 * The way that the address space of this process deletes its mapping
11075 * of this file is via the following call chains:
11076 * - as_free()->segop_unmap()/segvn_unmap()->fop_delmap()/nfs4_delmap()
11077 * - as_unmap()->segop_unmap()/segvn_unmap()->fop_delmap()/nfs4_delmap()
11079 * With the use of address space callbacks we are allowed to drop the
11080 * address space lock, a_lock, while executing the NFS operations that
11081 * need to go over the wire. Returning EAGAIN to the caller of this
11082 * function is what drives the execution of the callback that we add
11083 * below. The callback will be executed by the address space code
11084 * after dropping the a_lock. When the callback is finished, since
11085 * we dropped the a_lock, it must be re-acquired and segvn_unmap()
11086 * is called again on the same segment to finish the rest of the work
11087 * that needs to happen during unmapping.
11089 * This action of calling back into the segment driver causes
11090 * nfs4_delmap() to get called again, but since the callback was
11091 * already executed at this point, it already did the work and there
11092 * is nothing left for us to do.
11095 * - The first time nfs4_delmap is called by the current thread is when
11096 * we add the caller associated with this delmap to the delmap caller
11097 * list, add the callback, and return EAGAIN.
11098 * - The second time in this call chain when nfs4_delmap is called we
11099 * will find this caller in the delmap caller list and realize there
11100 * is no more work to do thus removing this caller from the list and
11101 * returning the error that was set in the callback execution.
11103 caller_found
= nfs4_find_and_delete_delmapcall(rp
, &error
);
11104 if (caller_found
) {
11106 * 'error' is from the actual delmap operations. To avoid
11107 * hangs, we need to handle the return of EAGAIN differently
11108 * since this is what drives the callback execution.
11109 * In this case, we don't want to return EAGAIN and do the
11110 * callback execution because there are none to execute.
11112 if (error
== EAGAIN
)
11118 /* current caller was not in the list */
11119 delmap_call
= nfs4_init_delmapcall();
11121 mutex_enter(&rp
->r_statelock
);
11122 list_insert_tail(&rp
->r_indelmap
, delmap_call
);
11123 mutex_exit(&rp
->r_statelock
);
11125 dmapp
= kmem_alloc(sizeof (nfs4_delmap_args_t
), KM_SLEEP
);
11129 dmapp
->addr
= addr
;
11131 dmapp
->prot
= prot
;
11132 dmapp
->maxprot
= maxprot
;
11133 dmapp
->flags
= flags
;
11135 dmapp
->caller
= delmap_call
;
11137 error
= as_add_callback(as
, nfs4_delmap_callback
, dmapp
,
11138 AS_UNMAP_EVENT
, addr
, len
, KM_SLEEP
);
11140 return (error
? error
: EAGAIN
);
11143 static nfs4_delmapcall_t
*
11144 nfs4_init_delmapcall()
11146 nfs4_delmapcall_t
*delmap_call
;
11148 delmap_call
= kmem_alloc(sizeof (nfs4_delmapcall_t
), KM_SLEEP
);
11149 delmap_call
->call_id
= curthread
;
11150 delmap_call
->error
= 0;
11152 return (delmap_call
);
11156 nfs4_free_delmapcall(nfs4_delmapcall_t
*delmap_call
)
11158 kmem_free(delmap_call
, sizeof (nfs4_delmapcall_t
));
11162 * Searches for the current delmap caller (based on curthread) in the list of
11163 * callers. If it is found, we remove it and free the delmap caller.
11165 * 0 if the caller wasn't found
11166 * 1 if the caller was found, removed and freed. *errp will be set
11167 * to what the result of the delmap was.
11170 nfs4_find_and_delete_delmapcall(rnode4_t
*rp
, int *errp
)
11172 nfs4_delmapcall_t
*delmap_call
;
11175 * If the list doesn't exist yet, we create it and return
11176 * that the caller wasn't found. No list = no callers.
11178 mutex_enter(&rp
->r_statelock
);
11179 if (!(rp
->r_flags
& R4DELMAPLIST
)) {
11180 /* The list does not exist */
11181 list_create(&rp
->r_indelmap
, sizeof (nfs4_delmapcall_t
),
11182 offsetof(nfs4_delmapcall_t
, call_node
));
11183 rp
->r_flags
|= R4DELMAPLIST
;
11184 mutex_exit(&rp
->r_statelock
);
11187 /* The list exists so search it */
11188 for (delmap_call
= list_head(&rp
->r_indelmap
);
11189 delmap_call
!= NULL
;
11190 delmap_call
= list_next(&rp
->r_indelmap
, delmap_call
)) {
11191 if (delmap_call
->call_id
== curthread
) {
11192 /* current caller is in the list */
11193 *errp
= delmap_call
->error
;
11194 list_remove(&rp
->r_indelmap
, delmap_call
);
11195 mutex_exit(&rp
->r_statelock
);
11196 nfs4_free_delmapcall(delmap_call
);
11201 mutex_exit(&rp
->r_statelock
);
11206 * Remove some pages from an mmap'd vnode. Just update the
11207 * count of pages. If doing close-to-open, then flush and
11208 * commit all of the pages associated with this file.
11209 * Otherwise, start an asynchronous page flush to write out
11210 * any dirty pages. This will also associate a credential
11211 * with the rnode which can be used to write the pages.
11215 nfs4_delmap_callback(struct as
*as
, void *arg
, uint_t event
)
11217 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
11220 nfs4_delmap_args_t
*dmapp
= (nfs4_delmap_args_t
*)arg
;
11222 rp
= VTOR4(dmapp
->vp
);
11223 mi
= VTOMI4(dmapp
->vp
);
11225 atomic_add_long((ulong_t
*)&rp
->r_mapcnt
, -btopr(dmapp
->len
));
11226 ASSERT(rp
->r_mapcnt
>= 0);
11229 * Initiate a page flush and potential commit if there are
11230 * pages, the file system was not mounted readonly, the segment
11231 * was mapped shared, and the pages themselves were writeable.
11233 if (nfs4_has_pages(dmapp
->vp
) &&
11234 !(dmapp
->vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
) &&
11235 dmapp
->flags
== MAP_SHARED
&& (dmapp
->maxprot
& PROT_WRITE
)) {
11236 mutex_enter(&rp
->r_statelock
);
11237 rp
->r_flags
|= R4DIRTY
;
11238 mutex_exit(&rp
->r_statelock
);
11239 e
.error
= nfs4_putpage_commit(dmapp
->vp
, dmapp
->off
,
11240 dmapp
->len
, dmapp
->cr
);
11242 mutex_enter(&rp
->r_statelock
);
11243 e
.error
= rp
->r_error
;
11245 mutex_exit(&rp
->r_statelock
);
11250 if ((rp
->r_flags
& R4DIRECTIO
) || (mi
->mi_flags
& MI4_DIRECTIO
))
11251 (void) nfs4_putpage(dmapp
->vp
, dmapp
->off
, dmapp
->len
,
11252 B_INVAL
, dmapp
->cr
, NULL
);
11255 e
.stat
= puterrno4(e
.error
);
11256 nfs4_queue_fact(RF_DELMAP_CB_ERR
, mi
, e
.stat
, 0,
11257 OP_COMMIT
, FALSE
, NULL
, 0, dmapp
->vp
);
11258 dmapp
->caller
->error
= e
.error
;
11261 /* Check to see if we need to close the file */
11263 if (dmapp
->vp
->v_type
== VREG
) {
11264 nfs4close_one(dmapp
->vp
, NULL
, dmapp
->cr
, 0, NULL
, &e
,
11265 CLOSE_DELMAP
, dmapp
->len
, dmapp
->maxprot
, dmapp
->flags
);
11267 if (e
.error
!= 0 || e
.stat
!= NFS4_OK
) {
11269 * Since it is possible that e.error == 0 and
11270 * e.stat != NFS4_OK (and vice versa),
11271 * we do the proper checking in order to get both
11272 * e.error and e.stat reporting the correct info.
11274 if (e
.stat
== NFS4_OK
)
11275 e
.stat
= puterrno4(e
.error
);
11277 e
.error
= geterrno4(e
.stat
);
11279 nfs4_queue_fact(RF_DELMAP_CB_ERR
, mi
, e
.stat
, 0,
11280 OP_CLOSE
, FALSE
, NULL
, 0, dmapp
->vp
);
11281 dmapp
->caller
->error
= e
.error
;
11285 (void) as_delete_callback(as
, arg
);
11286 kmem_free(dmapp
, sizeof (nfs4_delmap_args_t
));
11291 fattr4_maxfilesize_to_bits(uint64_t ll
)
11299 if (ll
& 0xffffffff00000000) {
11300 l
+= 32; ll
>>= 32;
11302 if (ll
& 0xffff0000) {
11303 l
+= 16; ll
>>= 16;
11321 nfs4_have_xattrs(vnode_t
*vp
, ulong_t
*valp
, cred_t
*cr
)
11323 vnode_t
*avp
= NULL
;
11326 if ((error
= nfs4lookup_xattr(vp
, "", &avp
,
11327 LOOKUP_XATTR
, cr
)) == 0)
11328 error
= do_xattr_exists_check(avp
, valp
, cr
);
11337 nfs4_pathconf(vnode_t
*vp
, int cmd
, ulong_t
*valp
, cred_t
*cr
,
11338 caller_context_t
*ct
)
11344 nfs4_ga_ext_res_t ger
;
11346 gar
.n4g_ext_res
= &ger
;
11348 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
11350 if (cmd
== _PC_PATH_MAX
|| cmd
== _PC_SYMLINK_MAX
) {
11351 *valp
= MAXPATHLEN
;
11354 if (cmd
== _PC_ACL_ENABLED
) {
11355 *valp
= _ACL_ACE_ENABLED
;
11360 if (cmd
== _PC_XATTR_EXISTS
) {
11362 * The existence of the xattr directory is not sufficient
11363 * for determining whether generic user attributes exists.
11364 * The attribute directory could only be a transient directory
11365 * used for Solaris sysattr support. Do a small readdir
11366 * to verify if the only entries are sysattrs or not.
11368 * pc4_xattr_valid can be only be trusted when r_xattr_dir
11369 * is NULL. Once the xadir vp exists, we can create xattrs,
11370 * and we don't have any way to update the "base" object's
11371 * pc4_xattr_exists from the xattr or xadir. Maybe FEM
11374 if (ATTRCACHE4_VALID(vp
) && rp
->r_pathconf
.pc4_xattr_valid
&&
11375 rp
->r_xattr_dir
== NULL
) {
11376 return (nfs4_have_xattrs(vp
, valp
, cr
));
11378 } else { /* OLD CODE */
11379 if (ATTRCACHE4_VALID(vp
)) {
11380 mutex_enter(&rp
->r_statelock
);
11381 if (rp
->r_pathconf
.pc4_cache_valid
) {
11384 case _PC_FILESIZEBITS
:
11386 rp
->r_pathconf
.pc4_filesizebits
;
11390 rp
->r_pathconf
.pc4_link_max
;
11394 rp
->r_pathconf
.pc4_name_max
;
11396 case _PC_CHOWN_RESTRICTED
:
11398 rp
->r_pathconf
.pc4_chown_restricted
;
11402 rp
->r_pathconf
.pc4_no_trunc
;
11408 mutex_exit(&rp
->r_statelock
);
11410 nfs4_pathconf_cache_hits
++;
11414 mutex_exit(&rp
->r_statelock
);
11418 nfs4_pathconf_cache_misses
++;
11423 error
= nfs4_attr_otw(vp
, TAG_PATHCONF
, &gar
, NFS4_PATHCONF_MASK
, cr
);
11426 mutex_enter(&rp
->r_statelock
);
11427 rp
->r_pathconf
.pc4_cache_valid
= FALSE
;
11428 rp
->r_pathconf
.pc4_xattr_valid
= FALSE
;
11429 mutex_exit(&rp
->r_statelock
);
11433 /* interpret the max filesize */
11434 gar
.n4g_ext_res
->n4g_pc4
.pc4_filesizebits
=
11435 fattr4_maxfilesize_to_bits(gar
.n4g_ext_res
->n4g_maxfilesize
);
11437 /* Store the attributes we just received */
11438 nfs4_attr_cache(vp
, &gar
, t
, cr
, TRUE
, NULL
);
11441 case _PC_FILESIZEBITS
:
11442 *valp
= gar
.n4g_ext_res
->n4g_pc4
.pc4_filesizebits
;
11445 *valp
= gar
.n4g_ext_res
->n4g_pc4
.pc4_link_max
;
11448 *valp
= gar
.n4g_ext_res
->n4g_pc4
.pc4_name_max
;
11450 case _PC_CHOWN_RESTRICTED
:
11451 *valp
= gar
.n4g_ext_res
->n4g_pc4
.pc4_chown_restricted
;
11454 *valp
= gar
.n4g_ext_res
->n4g_pc4
.pc4_no_trunc
;
11456 case _PC_XATTR_EXISTS
:
11457 if (gar
.n4g_ext_res
->n4g_pc4
.pc4_xattr_exists
) {
11458 if (error
= nfs4_have_xattrs(vp
, valp
, cr
))
11470 * Called by async thread to do synchronous pageio. Do the i/o, wait
11471 * for it to complete, and cleanup the page list when done.
11474 nfs4_sync_pageio(vnode_t
*vp
, page_t
*pp
, uoff_t io_off
, size_t io_len
,
11475 int flags
, cred_t
*cr
)
11479 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
11481 error
= nfs4_rdwrlbn(vp
, pp
, io_off
, io_len
, flags
, cr
);
11482 if (flags
& B_READ
)
11483 pvn_read_done(pp
, (error
? B_ERROR
: 0) | flags
);
11485 pvn_write_done(pp
, (error
? B_ERROR
: 0) | flags
);
11491 nfs4_pageio(vnode_t
*vp
, page_t
*pp
, uoff_t io_off
, size_t io_len
,
11492 int flags
, cred_t
*cr
, caller_context_t
*ct
)
11497 if (!(flags
& B_ASYNC
) && nfs_zone() != VTOMI4(vp
)->mi_zone
)
11504 mutex_enter(&rp
->r_statelock
);
11506 mutex_exit(&rp
->r_statelock
);
11508 if (flags
& B_ASYNC
) {
11509 error
= nfs4_async_pageio(vp
, pp
, io_off
, io_len
, flags
, cr
,
11512 error
= nfs4_rdwrlbn(vp
, pp
, io_off
, io_len
, flags
, cr
);
11513 mutex_enter(&rp
->r_statelock
);
11515 cv_broadcast(&rp
->r_cv
);
11516 mutex_exit(&rp
->r_statelock
);
11522 nfs4_dispose(vnode_t
*vp
, page_t
*pp
, int fl
, int dn
, cred_t
*cr
,
11523 caller_context_t
*ct
)
11534 * We should get called with fl equal to either B_FREE or
11535 * B_INVAL. Any other value is illegal.
11537 * The page that we are either supposed to free or destroy
11538 * should be exclusive locked and its io lock should not
11541 ASSERT(fl
== B_FREE
|| fl
== B_INVAL
);
11542 ASSERT((PAGE_EXCL(pp
) && !page_iolock_assert(pp
)) || panicstr
);
11547 * If the page doesn't need to be committed or we shouldn't
11548 * even bother attempting to commit it, then just make sure
11549 * that the p_fsdata byte is clear and then either free or
11550 * destroy the page as appropriate.
11552 if (pp
->p_fsdata
== C_NOCOMMIT
|| (rp
->r_flags
& R4STALE
)) {
11553 pp
->p_fsdata
= C_NOCOMMIT
;
11557 page_destroy(pp
, dn
);
11562 * If there is a page invalidation operation going on, then
11563 * if this is one of the pages being destroyed, then just
11564 * clear the p_fsdata byte and then either free or destroy
11565 * the page as appropriate.
11567 mutex_enter(&rp
->r_statelock
);
11568 if ((rp
->r_flags
& R4TRUNCATE
) && pp
->p_offset
>= rp
->r_truncaddr
) {
11569 mutex_exit(&rp
->r_statelock
);
11570 pp
->p_fsdata
= C_NOCOMMIT
;
11574 page_destroy(pp
, dn
);
11579 * If we are freeing this page and someone else is already
11580 * waiting to do a commit, then just unlock the page and
11581 * return. That other thread will take care of commiting
11582 * this page. The page can be freed sometime after the
11583 * commit has finished. Otherwise, if the page is marked
11584 * as delay commit, then we may be getting called from
11585 * pvn_write_done, one page at a time. This could result
11586 * in one commit per page, so we end up doing lots of small
11587 * commits instead of fewer larger commits. This is bad,
11588 * we want do as few commits as possible.
11590 if (fl
== B_FREE
) {
11591 if (rp
->r_flags
& R4COMMITWAIT
) {
11593 mutex_exit(&rp
->r_statelock
);
11596 if (pp
->p_fsdata
== C_DELAYCOMMIT
) {
11597 pp
->p_fsdata
= C_COMMIT
;
11599 mutex_exit(&rp
->r_statelock
);
11605 * Check to see if there is a signal which would prevent an
11606 * attempt to commit the pages from being successful. If so,
11607 * then don't bother with all of the work to gather pages and
11608 * generate the unsuccessful RPC. Just return from here and
11609 * let the page be committed at some later time.
11611 sigintr(&smask
, VTOMI4(vp
)->mi_flags
& MI4_INT
);
11612 if (ttolwp(curthread
) != NULL
&& ISSIG(curthread
, JUSTLOOKING
)) {
11615 mutex_exit(&rp
->r_statelock
);
11621 * We are starting to need to commit pages, so let's try
11622 * to commit as many as possible at once to reduce the
11625 * Set the `commit inprogress' state bit. We must
11626 * first wait until any current one finishes. Then
11627 * we initialize the c_pages list with this page.
11629 while (rp
->r_flags
& R4COMMIT
) {
11630 rp
->r_flags
|= R4COMMITWAIT
;
11631 cv_wait(&rp
->r_commit
.c_cv
, &rp
->r_statelock
);
11632 rp
->r_flags
&= ~R4COMMITWAIT
;
11634 rp
->r_flags
|= R4COMMIT
;
11635 mutex_exit(&rp
->r_statelock
);
11636 ASSERT(rp
->r_commit
.c_pages
== NULL
);
11637 rp
->r_commit
.c_pages
= pp
;
11638 rp
->r_commit
.c_commbase
= (offset3
)pp
->p_offset
;
11639 rp
->r_commit
.c_commlen
= PAGESIZE
;
11642 * Gather together all other pages which can be committed.
11643 * They will all be chained off r_commit.c_pages.
11645 nfs4_get_commit(vp
);
11648 * Clear the `commit inprogress' status and disconnect
11649 * the list of pages to be committed from the rnode.
11650 * At this same time, we also save the starting offset
11651 * and length of data to be committed on the server.
11653 plist
= rp
->r_commit
.c_pages
;
11654 rp
->r_commit
.c_pages
= NULL
;
11655 offset
= rp
->r_commit
.c_commbase
;
11656 len
= rp
->r_commit
.c_commlen
;
11657 mutex_enter(&rp
->r_statelock
);
11658 rp
->r_flags
&= ~R4COMMIT
;
11659 cv_broadcast(&rp
->r_commit
.c_cv
);
11660 mutex_exit(&rp
->r_statelock
);
11662 if (curproc
== proc_pageout
|| curproc
== proc_fsflush
||
11663 nfs_zone() != VTOMI4(vp
)->mi_zone
) {
11664 nfs4_async_commit(vp
, plist
, offset
, len
,
11665 cr
, do_nfs4_async_commit
);
11670 * Actually generate the COMMIT op over the wire operation.
11672 error
= nfs4_commit(vp
, (offset4
)offset
, (count4
)len
, cr
);
11675 * If we got an error during the commit, just unlock all
11676 * of the pages. The pages will get retransmitted to the
11677 * server during a putpage operation.
11680 while (plist
!= NULL
) {
11682 page_sub(&plist
, pptr
);
11689 * We've tried as hard as we can to commit the data to stable
11690 * storage on the server. We just unlock the rest of the pages
11691 * and clear the commit required state. They will be put
11692 * onto the tail of the cachelist if they are nolonger
11695 while (plist
!= pp
) {
11697 page_sub(&plist
, pptr
);
11698 pptr
->p_fsdata
= C_NOCOMMIT
;
11703 * It is possible that nfs4_commit didn't return error but
11704 * some other thread has modified the page we are going
11706 * In this case we need to rewrite the page. Do an explicit check
11707 * before attempting to free/destroy the page. If modified, needs to
11708 * be rewritten so unlock the page and return.
11710 if (hat_ismod(pp
)) {
11711 pp
->p_fsdata
= C_NOCOMMIT
;
11717 * Now, as appropriate, either free or destroy the page
11718 * that we were called with.
11720 pp
->p_fsdata
= C_NOCOMMIT
;
11724 page_destroy(pp
, dn
);
11728 * Commit requires that the current fh be the file written to.
11729 * The compound op structure is:
11730 * PUTFH(file), COMMIT
11733 nfs4_commit(vnode_t
*vp
, offset4 offset
, count4 count
, cred_t
*cr
)
11735 COMPOUND4args_clnt args
;
11736 COMPOUND4res_clnt res
;
11737 COMMIT4res
*cm_res
;
11738 nfs_argop4 argop
[2];
11743 cred_t
*cred_otw
= NULL
;
11744 bool_t needrecov
= FALSE
;
11745 nfs4_recov_state_t recov_state
;
11746 nfs4_open_stream_t
*osp
= NULL
;
11747 bool_t first_time
= TRUE
; /* first time getting OTW cred */
11748 bool_t last_time
= FALSE
; /* last time getting OTW cred */
11749 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
11751 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
11756 recov_state
.rs_flags
= 0;
11757 recov_state
.rs_num_retry_despite_err
= 0;
11760 * Releases the osp, if a valid open stream is provided.
11761 * Puts a hold on the cred_otw and the new osp (if found).
11763 cred_otw
= nfs4_get_otw_cred_by_osp(rp
, cr
, &osp
,
11764 &first_time
, &last_time
);
11765 args
.ctag
= TAG_COMMIT
;
11768 * Commit ops: putfh file; commit
11770 args
.array_len
= 2;
11771 args
.array
= argop
;
11773 e
.error
= nfs4_start_fop(VTOMI4(vp
), vp
, NULL
, OH_COMMIT
,
11774 &recov_state
, NULL
);
11778 open_stream_rele(osp
, rp
);
11782 /* putfh directory */
11783 argop
[0].argop
= OP_CPUTFH
;
11784 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= rp
->r_fh
;
11787 argop
[1].argop
= OP_COMMIT
;
11788 argop
[1].nfs_argop4_u
.opcommit
.offset
= offset
;
11789 argop
[1].nfs_argop4_u
.opcommit
.count
= count
;
11792 rfs4call(mi
, &args
, &res
, cred_otw
, &doqueue
, 0, &e
);
11794 needrecov
= nfs4_needs_recovery(&e
, FALSE
, mi
->mi_vfsp
);
11795 if (!needrecov
&& e
.error
) {
11796 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_COMMIT
, &recov_state
,
11799 if (e
.error
== EACCES
&& last_time
== FALSE
)
11800 goto get_commit_cred
;
11802 open_stream_rele(osp
, rp
);
11807 if (nfs4_start_recovery(&e
, VTOMI4(vp
), vp
, NULL
, NULL
,
11808 NULL
, OP_COMMIT
, NULL
, NULL
, NULL
) == FALSE
) {
11809 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_COMMIT
,
11810 &recov_state
, needrecov
);
11812 (void) xdr_free(xdr_COMPOUND4res_clnt
,
11817 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_COMMIT
,
11818 &recov_state
, needrecov
);
11821 open_stream_rele(osp
, rp
);
11824 /* fall through for res.status case */
11828 e
.error
= geterrno4(res
.status
);
11829 if (e
.error
== EACCES
&& last_time
== FALSE
) {
11831 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_COMMIT
,
11832 &recov_state
, needrecov
);
11833 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
11834 goto get_commit_cred
;
11837 * Can't do a nfs4_purge_stale_fh here because this
11838 * can cause a deadlock. nfs4_commit can
11839 * be called from nfs4_dispose which can be called
11840 * indirectly via pvn_vplist_dirty. nfs4_purge_stale_fh
11841 * can call back to pvn_vplist_dirty.
11843 if (e
.error
== ESTALE
) {
11844 mutex_enter(&rp
->r_statelock
);
11845 rp
->r_flags
|= R4STALE
;
11847 rp
->r_error
= e
.error
;
11848 mutex_exit(&rp
->r_statelock
);
11849 PURGE_ATTRCACHE4(vp
);
11851 mutex_enter(&rp
->r_statelock
);
11853 rp
->r_error
= e
.error
;
11854 mutex_exit(&rp
->r_statelock
);
11857 ASSERT(rp
->r_flags
& R4HAVEVERF
);
11858 resop
= &res
.array
[1]; /* commit res */
11859 cm_res
= &resop
->nfs_resop4_u
.opcommit
;
11860 mutex_enter(&rp
->r_statelock
);
11861 if (cm_res
->writeverf
== rp
->r_writeverf
) {
11862 mutex_exit(&rp
->r_statelock
);
11863 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
11864 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_COMMIT
,
11865 &recov_state
, needrecov
);
11868 open_stream_rele(osp
, rp
);
11872 rp
->r_writeverf
= cm_res
->writeverf
;
11873 mutex_exit(&rp
->r_statelock
);
11874 e
.error
= NFS_VERF_MISMATCH
;
11877 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
11878 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, OH_COMMIT
, &recov_state
, needrecov
);
11881 open_stream_rele(osp
, rp
);
11887 nfs4_set_mod(vnode_t
*vp
)
11889 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
11891 /* make sure we're looking at the master vnode, not a shadow */
11892 pvn_vplist_setdirty(RTOV4(VTOR4(vp
)), nfs_setmod_check
);
11896 * This function is used to gather a page list of the pages which
11897 * can be committed on the server.
11899 * The calling thread must have set R4COMMIT. This bit is used to
11900 * serialize access to the commit structure in the rnode. As long
11901 * as the thread has set R4COMMIT, then it can manipulate the commit
11902 * structure without requiring any other locks.
11904 * When this function is called from nfs4_dispose() the page passed
11905 * into nfs4_dispose() will be SE_EXCL locked, and so this function
11906 * will skip it. This is not a problem since we initially add the
11907 * page to the r_commit page list.
11911 nfs4_get_commit(vnode_t
*vp
)
11919 ASSERT(rp
->r_flags
& R4COMMIT
);
11921 /* make sure we're looking at the master vnode, not a shadow */
11923 if (IS_SHADOW(vp
, rp
))
11926 vphm
= page_vnode_mutex(vp
);
11930 * Step through all of the pages associated with this vnode
11931 * looking for pages which need to be committed.
11933 for (pp
= vnode_get_head(vp
);
11935 pp
= vnode_get_next(vp
, pp
)) {
11936 /* Skip marker pages. */
11937 if (PP_ISPVN_TAG(pp
))
11941 * First short-cut everything (without the page_lock)
11942 * and see if this page does not need to be committed
11943 * or is modified if so then we'll just skip it.
11945 if (pp
->p_fsdata
== C_NOCOMMIT
|| hat_ismod(pp
))
11949 * Attempt to lock the page. If we can't, then
11950 * someone else is messing with it or we have been
11951 * called from nfs4_dispose and this is the page that
11952 * nfs4_dispose was called with.. anyway just skip it.
11954 if (!page_trylock(pp
, SE_EXCL
))
11958 * Lets check again now that we have the page lock.
11960 if (pp
->p_fsdata
== C_NOCOMMIT
|| hat_ismod(pp
)) {
11965 /* this had better not be a free page */
11966 ASSERT(PP_ISFREE(pp
) == 0);
11969 * The page needs to be committed and we locked it.
11970 * Update the base and length parameters and add it
11973 if (rp
->r_commit
.c_pages
== NULL
) {
11974 rp
->r_commit
.c_commbase
= (offset3
)pp
->p_offset
;
11975 rp
->r_commit
.c_commlen
= PAGESIZE
;
11976 } else if (pp
->p_offset
< rp
->r_commit
.c_commbase
) {
11977 rp
->r_commit
.c_commlen
= rp
->r_commit
.c_commbase
-
11978 (offset3
)pp
->p_offset
+ rp
->r_commit
.c_commlen
;
11979 rp
->r_commit
.c_commbase
= (offset3
)pp
->p_offset
;
11980 } else if ((rp
->r_commit
.c_commbase
+ rp
->r_commit
.c_commlen
)
11982 rp
->r_commit
.c_commlen
= (offset3
)pp
->p_offset
-
11983 rp
->r_commit
.c_commbase
+ PAGESIZE
;
11985 page_add(&rp
->r_commit
.c_pages
, pp
);
11992 * This routine is used to gather together a page list of the pages
11993 * which are to be committed on the server. This routine must not
11994 * be called if the calling thread holds any locked pages.
11996 * The calling thread must have set R4COMMIT. This bit is used to
11997 * serialize access to the commit structure in the rnode. As long
11998 * as the thread has set R4COMMIT, then it can manipulate the commit
11999 * structure without requiring any other locks.
12002 nfs4_get_commit_range(vnode_t
*vp
, uoff_t soff
, size_t len
)
12011 ASSERT(rp
->r_flags
& R4COMMIT
);
12013 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
12015 /* make sure we're looking at the master vnode, not a shadow */
12017 if (IS_SHADOW(vp
, rp
))
12021 * If there are no pages associated with this vnode, then
12024 if (!vn_has_cached_data(vp
))
12027 * Calculate the ending offset.
12030 for (off
= soff
; off
< end
; off
+= PAGESIZE
) {
12032 * Lookup each page by vp, offset.
12034 if ((pp
= page_lookup_nowait(vp
, off
, SE_EXCL
)) == NULL
)
12037 * If this page does not need to be committed or is
12038 * modified, then just skip it.
12040 if (pp
->p_fsdata
== C_NOCOMMIT
|| hat_ismod(pp
)) {
12045 ASSERT(PP_ISFREE(pp
) == 0);
12047 * The page needs to be committed and we locked it.
12048 * Update the base and length parameters and add it
12051 if (rp
->r_commit
.c_pages
== NULL
) {
12052 rp
->r_commit
.c_commbase
= (offset3
)pp
->p_offset
;
12053 rp
->r_commit
.c_commlen
= PAGESIZE
;
12055 rp
->r_commit
.c_commlen
= (offset3
)pp
->p_offset
-
12056 rp
->r_commit
.c_commbase
+ PAGESIZE
;
12058 page_add(&rp
->r_commit
.c_pages
, pp
);
12063 * Called from nfs4_close(), nfs4_fsync() and nfs4_delmap().
12064 * Flushes and commits data to the server.
12067 nfs4_putpage_commit(vnode_t
*vp
, offset_t poff
, size_t plen
, cred_t
*cr
)
12070 verifier4 write_verf
;
12071 rnode4_t
*rp
= VTOR4(vp
);
12073 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
12076 * Flush the data portion of the file and then commit any
12077 * portions which need to be committed. This may need to
12078 * be done twice if the server has changed state since
12079 * data was last written. The data will need to be
12080 * rewritten to the server and then a new commit done.
12082 * In fact, this may need to be done several times if the
12083 * server is having problems and crashing while we are
12084 * attempting to do this.
12089 * Do a flush based on the poff and plen arguments. This
12090 * will synchronously write out any modified pages in the
12091 * range specified by (poff, plen). This starts all of the
12092 * i/o operations which will be waited for in the next
12093 * call to nfs4_putpage
12096 mutex_enter(&rp
->r_statelock
);
12097 write_verf
= rp
->r_writeverf
;
12098 mutex_exit(&rp
->r_statelock
);
12100 error
= nfs4_putpage(vp
, poff
, plen
, B_ASYNC
, cr
, NULL
);
12101 if (error
== EAGAIN
)
12105 * Do a flush based on the poff and plen arguments. This
12106 * will synchronously write out any modified pages in the
12107 * range specified by (poff, plen) and wait until all of
12108 * the asynchronous i/o's in that range are done as well.
12111 error
= nfs4_putpage(vp
, poff
, plen
, 0, cr
, NULL
);
12116 mutex_enter(&rp
->r_statelock
);
12117 if (rp
->r_writeverf
!= write_verf
) {
12118 mutex_exit(&rp
->r_statelock
);
12121 mutex_exit(&rp
->r_statelock
);
12124 * Now commit any pages which might need to be committed.
12125 * If the error, NFS_VERF_MISMATCH, is returned, then
12126 * start over with the flush operation.
12128 error
= nfs4_commit_vp(vp
, poff
, plen
, cr
, NFS4_WRITE_WAIT
);
12130 if (error
== NFS_VERF_MISMATCH
)
12137 * nfs4_commit_vp() will wait for other pending commits and
12138 * will either commit the whole file or a range, plen dictates
12139 * if we commit whole file. a value of zero indicates the whole
12140 * file. Called from nfs4_putpage_commit() or nfs4_sync_putapage()
12143 nfs4_commit_vp(vnode_t
*vp
, uoff_t poff
, size_t plen
,
12144 cred_t
*cr
, int wait_on_writes
)
12151 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
12156 * before we gather commitable pages make
12157 * sure there are no outstanding async writes
12159 if (rp
->r_count
&& wait_on_writes
== NFS4_WRITE_WAIT
) {
12160 mutex_enter(&rp
->r_statelock
);
12161 while (rp
->r_count
> 0) {
12162 cv_wait(&rp
->r_cv
, &rp
->r_statelock
);
12164 mutex_exit(&rp
->r_statelock
);
12168 * Set the `commit inprogress' state bit. We must
12169 * first wait until any current one finishes.
12171 mutex_enter(&rp
->r_statelock
);
12172 while (rp
->r_flags
& R4COMMIT
) {
12173 rp
->r_flags
|= R4COMMITWAIT
;
12174 cv_wait(&rp
->r_commit
.c_cv
, &rp
->r_statelock
);
12175 rp
->r_flags
&= ~R4COMMITWAIT
;
12177 rp
->r_flags
|= R4COMMIT
;
12178 mutex_exit(&rp
->r_statelock
);
12181 * Gather all of the pages which need to be
12185 nfs4_get_commit(vp
);
12187 nfs4_get_commit_range(vp
, poff
, plen
);
12190 * Clear the `commit inprogress' bit and disconnect the
12191 * page list which was gathered by nfs4_get_commit.
12193 plist
= rp
->r_commit
.c_pages
;
12194 rp
->r_commit
.c_pages
= NULL
;
12195 offset
= rp
->r_commit
.c_commbase
;
12196 len
= rp
->r_commit
.c_commlen
;
12197 mutex_enter(&rp
->r_statelock
);
12198 rp
->r_flags
&= ~R4COMMIT
;
12199 cv_broadcast(&rp
->r_commit
.c_cv
);
12200 mutex_exit(&rp
->r_statelock
);
12203 * If any pages need to be committed, commit them and
12204 * then unlock them so that they can be freed some
12211 * No error occurred during the flush portion
12212 * of this operation, so now attempt to commit
12213 * the data to stable storage on the server.
12215 * This will unlock all of the pages on the list.
12217 return (nfs4_sync_commit(vp
, plist
, offset
, len
, cr
));
12221 nfs4_sync_commit(vnode_t
*vp
, page_t
*plist
, offset3 offset
, count3 count
,
12227 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
12229 error
= nfs4_commit(vp
, (offset4
)offset
, (count3
)count
, cr
);
12232 * If we got an error, then just unlock all of the pages
12236 while (plist
!= NULL
) {
12238 page_sub(&plist
, pp
);
12244 * We've tried as hard as we can to commit the data to stable
12245 * storage on the server. We just unlock the pages and clear
12246 * the commit required state. They will get freed later.
12248 while (plist
!= NULL
) {
12250 page_sub(&plist
, pp
);
12251 pp
->p_fsdata
= C_NOCOMMIT
;
12259 do_nfs4_async_commit(vnode_t
*vp
, page_t
*plist
, offset3 offset
, count3 count
,
12263 (void) nfs4_sync_commit(vp
, plist
, offset
, count
, cr
);
12268 nfs4_setsecattr(vnode_t
*vp
, vsecattr_t
*vsecattr
, int flag
, cred_t
*cr
,
12269 caller_context_t
*ct
)
12274 vsecattr_t nfsace4_vsap
;
12277 if (nfs_zone() != mi
->mi_zone
)
12279 if (mi
->mi_flags
& MI4_ACL
) {
12280 /* if we have a delegation, return it */
12281 if (VTOR4(vp
)->r_deleg_type
!= OPEN_DELEGATE_NONE
)
12282 (void) nfs4delegreturn(VTOR4(vp
),
12283 NFS4_DR_REOPEN
|NFS4_DR_PUSH
);
12285 error
= nfs4_is_acl_mask_valid(vsecattr
->vsa_mask
,
12287 if (error
) /* EINVAL */
12290 if (vsecattr
->vsa_mask
& (VSA_ACL
| VSA_DFACL
)) {
12292 * These are aclent_t type entries.
12294 error
= vs_aent_to_ace4(vsecattr
, &nfsace4_vsap
,
12295 vp
->v_type
== VDIR
, FALSE
);
12300 * These are ace_t type entries.
12302 error
= vs_acet_to_ace4(vsecattr
, &nfsace4_vsap
,
12307 bzero(&va
, sizeof (va
));
12308 error
= nfs4setattr(vp
, &va
, flag
, cr
, &nfsace4_vsap
);
12309 vs_ace4_destroy(&nfsace4_vsap
);
12317 nfs4_getsecattr(vnode_t
*vp
, vsecattr_t
*vsecattr
, int flag
, cred_t
*cr
,
12318 caller_context_t
*ct
)
12323 rnode4_t
*rp
= VTOR4(vp
);
12326 if (nfs_zone() != mi
->mi_zone
)
12329 bzero(&gar
, sizeof (gar
));
12330 gar
.n4g_vsa
.vsa_mask
= vsecattr
->vsa_mask
;
12333 * vsecattr->vsa_mask holds the original acl request mask.
12334 * This is needed when determining what to return.
12335 * (See: nfs4_create_getsecattr_return())
12337 error
= nfs4_is_acl_mask_valid(vsecattr
->vsa_mask
, NFS4_ACL_GET
);
12338 if (error
) /* EINVAL */
12342 * If this is a referral stub, don't try to go OTW for an ACL
12344 if (RP_ISSTUB_REFERRAL(VTOR4(vp
)))
12345 return (fs_fab_acl(vp
, vsecattr
, flag
, cr
, ct
));
12347 if (mi
->mi_flags
& MI4_ACL
) {
12349 * Check if the data is cached and the cache is valid. If it
12350 * is we don't go over the wire.
12352 if (rp
->r_secattr
!= NULL
&& ATTRCACHE4_VALID(vp
)) {
12353 mutex_enter(&rp
->r_statelock
);
12354 if (rp
->r_secattr
!= NULL
) {
12355 error
= nfs4_create_getsecattr_return(
12356 rp
->r_secattr
, vsecattr
, rp
->r_attr
.va_uid
,
12358 vp
->v_type
== VDIR
);
12359 if (!error
) { /* error == 0 - Success! */
12360 mutex_exit(&rp
->r_statelock
);
12364 mutex_exit(&rp
->r_statelock
);
12368 * The getattr otw call will always get both the acl, in
12369 * the form of a list of nfsace4's, and the number of acl
12370 * entries; independent of the value of gar.n4g_va.va_mask.
12372 error
= nfs4_getattr_otw(vp
, &gar
, cr
, 1);
12374 vs_ace4_destroy(&gar
.n4g_vsa
);
12375 if (error
== ENOTSUP
|| error
== EOPNOTSUPP
)
12376 error
= fs_fab_acl(vp
, vsecattr
, flag
, cr
, ct
);
12380 if (!(gar
.n4g_resbmap
& FATTR4_ACL_MASK
)) {
12382 * No error was returned, but according to the response
12383 * bitmap, neither was an acl.
12385 vs_ace4_destroy(&gar
.n4g_vsa
);
12386 error
= fs_fab_acl(vp
, vsecattr
, flag
, cr
, ct
);
12391 * Update the cache with the ACL.
12393 nfs4_acl_fill_cache(rp
, &gar
.n4g_vsa
);
12395 error
= nfs4_create_getsecattr_return(&gar
.n4g_vsa
,
12396 vsecattr
, gar
.n4g_va
.va_uid
, gar
.n4g_va
.va_gid
,
12397 vp
->v_type
== VDIR
);
12398 vs_ace4_destroy(&gar
.n4g_vsa
);
12399 if ((error
) && (vsecattr
->vsa_mask
&
12400 (VSA_ACL
| VSA_ACLCNT
| VSA_DFACL
| VSA_DFACLCNT
)) &&
12401 (error
!= EACCES
)) {
12402 error
= fs_fab_acl(vp
, vsecattr
, flag
, cr
, ct
);
12406 error
= fs_fab_acl(vp
, vsecattr
, flag
, cr
, ct
);
12411 * The function returns:
12412 * - 0 (zero) if the passed in "acl_mask" is a valid request.
12413 * - EINVAL if the passed in "acl_mask" is an invalid request.
12415 * In the case of getting an acl (op == NFS4_ACL_GET) the mask is invalid if:
12416 * - We have a mixture of ACE and ACL requests (e.g. VSA_ACL | VSA_ACE)
12418 * In the case of setting an acl (op == NFS4_ACL_SET) the mask is invalid if:
12419 * - We have a mixture of ACE and ACL requests (e.g. VSA_ACL | VSA_ACE)
12420 * - We have a count field set without the corresponding acl field set. (e.g. -
12421 * VSA_ACECNT is set, but VSA_ACE is not)
12424 nfs4_is_acl_mask_valid(uint_t acl_mask
, nfs4_acl_op_t op
)
12426 /* Shortcut the masks that are always valid. */
12427 if (acl_mask
== (VSA_ACE
| VSA_ACECNT
))
12429 if (acl_mask
== (VSA_ACL
| VSA_ACLCNT
| VSA_DFACL
| VSA_DFACLCNT
))
12432 if (acl_mask
& (VSA_ACE
| VSA_ACECNT
)) {
12434 * We can't have any VSA_ACL type stuff in the mask now.
12436 if (acl_mask
& (VSA_ACL
| VSA_ACLCNT
| VSA_DFACL
|
12440 if (op
== NFS4_ACL_SET
) {
12441 if ((acl_mask
& VSA_ACECNT
) && !(acl_mask
& VSA_ACE
))
12446 if (acl_mask
& (VSA_ACL
| VSA_ACLCNT
| VSA_DFACL
| VSA_DFACLCNT
)) {
12448 * We can't have any VSA_ACE type stuff in the mask now.
12450 if (acl_mask
& (VSA_ACE
| VSA_ACECNT
))
12453 if (op
== NFS4_ACL_SET
) {
12454 if ((acl_mask
& VSA_ACLCNT
) && !(acl_mask
& VSA_ACL
))
12457 if ((acl_mask
& VSA_DFACLCNT
) &&
12458 !(acl_mask
& VSA_DFACL
))
12466 * The theory behind creating the correct getsecattr return is simply this:
12467 * "Don't return anything that the caller is not expecting to have to free."
12470 nfs4_create_getsecattr_return(vsecattr_t
*filled_vsap
, vsecattr_t
*vsap
,
12471 uid_t uid
, gid_t gid
, int isdir
)
12474 /* Save the mask since the translators modify it. */
12475 uint_t orig_mask
= vsap
->vsa_mask
;
12477 if (orig_mask
& (VSA_ACE
| VSA_ACECNT
)) {
12478 error
= vs_ace4_to_acet(filled_vsap
, vsap
, uid
, gid
, FALSE
);
12484 * If the caller only asked for the ace count (VSA_ACECNT)
12485 * don't give them the full acl (VSA_ACE), free it.
12487 if (!orig_mask
& VSA_ACE
) {
12488 if (vsap
->vsa_aclentp
!= NULL
) {
12489 kmem_free(vsap
->vsa_aclentp
,
12490 vsap
->vsa_aclcnt
* sizeof (ace_t
));
12491 vsap
->vsa_aclentp
= NULL
;
12494 vsap
->vsa_mask
= orig_mask
;
12496 } else if (orig_mask
& (VSA_ACL
| VSA_ACLCNT
| VSA_DFACL
|
12498 error
= vs_ace4_to_aent(filled_vsap
, vsap
, uid
, gid
,
12505 * If the caller only asked for the acl count (VSA_ACLCNT)
12506 * and/or the default acl count (VSA_DFACLCNT) don't give them
12507 * the acl (VSA_ACL) or default acl (VSA_DFACL), free it.
12509 if (!orig_mask
& VSA_ACL
) {
12510 if (vsap
->vsa_aclentp
!= NULL
) {
12511 kmem_free(vsap
->vsa_aclentp
,
12512 vsap
->vsa_aclcnt
* sizeof (aclent_t
));
12513 vsap
->vsa_aclentp
= NULL
;
12517 if (!orig_mask
& VSA_DFACL
) {
12518 if (vsap
->vsa_dfaclentp
!= NULL
) {
12519 kmem_free(vsap
->vsa_dfaclentp
,
12520 vsap
->vsa_dfaclcnt
* sizeof (aclent_t
));
12521 vsap
->vsa_dfaclentp
= NULL
;
12524 vsap
->vsa_mask
= orig_mask
;
12531 nfs4_shrlock(vnode_t
*vp
, int cmd
, struct shrlock
*shr
, int flag
, cred_t
*cr
,
12532 caller_context_t
*ct
)
12536 if (nfs_zone() != VTOMI4(vp
)->mi_zone
)
12539 * check for valid cmd parameter
12541 if (cmd
!= F_SHARE
&& cmd
!= F_UNSHARE
&& cmd
!= F_HASREMOTELOCKS
)
12545 * Check access permissions
12547 if ((cmd
& F_SHARE
) &&
12548 (((shr
->s_access
& F_RDACC
) && (flag
& FREAD
) == 0) ||
12549 (shr
->s_access
== F_WRACC
&& (flag
& FWRITE
) == 0)))
12553 * If the filesystem is mounted using local locking, pass the
12554 * request off to the local share code.
12556 if (VTOMI4(vp
)->mi_flags
& MI4_LLOCK
)
12557 return (fs_shrlock(vp
, cmd
, shr
, flag
, cr
, ct
));
12563 * This will be properly implemented later,
12564 * see RFE: 4823948 .
12569 case F_HASREMOTELOCKS
:
12571 * NFS client can't store remote locks itself
12586 * Common code called by directory ops to update the attrcache
12589 nfs4_update_attrcache(nfsstat4 status
, nfs4_ga_res_t
*garp
,
12590 hrtime_t t
, vnode_t
*vp
, cred_t
*cr
)
12594 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
12596 if (status
!= NFS4_OK
) {
12597 /* getattr not done or failed */
12598 PURGE_ATTRCACHE4(vp
);
12603 nfs4_attr_cache(vp
, garp
, t
, cr
, FALSE
, NULL
);
12605 PURGE_ATTRCACHE4(vp
);
12611 * Update directory caches for directory modification ops (link, rename, etc.)
12612 * When dinfo is NULL, manage dircaches in the old way.
12615 nfs4_update_dircaches(change_info4
*cinfo
, vnode_t
*dvp
, vnode_t
*vp
, char *nm
,
12616 dirattr_info_t
*dinfo
)
12618 rnode4_t
*drp
= VTOR4(dvp
);
12620 ASSERT(nfs_zone() == VTOMI4(dvp
)->mi_zone
);
12622 /* Purge rddir cache for dir since it changed */
12623 if (drp
->r_dir
!= NULL
)
12624 nfs4_purge_rddir_cache(dvp
);
12627 * If caller provided dinfo, then use it to manage dir caches.
12629 if (dinfo
!= NULL
) {
12631 mutex_enter(&VTOR4(vp
)->r_statev4_lock
);
12632 if (!VTOR4(vp
)->created_v4
) {
12633 mutex_exit(&VTOR4(vp
)->r_statev4_lock
);
12634 dnlc_update(dvp
, nm
, vp
);
12637 * XXX don't update if the created_v4 flag is
12640 mutex_exit(&VTOR4(vp
)->r_statev4_lock
);
12641 NFS4_DEBUG(nfs4_client_state_debug
,
12642 (CE_NOTE
, "nfs4_update_dircaches: "
12643 "don't update dnlc: created_v4 flag"));
12647 nfs4_attr_cache(dvp
, dinfo
->di_garp
, dinfo
->di_time_call
,
12648 dinfo
->di_cred
, FALSE
, cinfo
);
12654 * Caller didn't provide dinfo, then check change_info4 to update DNLC.
12655 * Since caller modified dir but didn't receive post-dirmod-op dir
12656 * attrs, the dir's attrs must be purged.
12658 * XXX this check and dnlc update/purge should really be atomic,
12659 * XXX but can't use rnode statelock because it'll deadlock in
12660 * XXX dnlc_purge_vp, however, the risk is minimal even if a race
12663 * XXX We also may want to check that atomic is true in the
12664 * XXX change_info struct. If it is not, the change_info may
12665 * XXX reflect changes by more than one clients which means that
12666 * XXX our cache may not be valid.
12668 PURGE_ATTRCACHE4(dvp
);
12669 if (drp
->r_change
== cinfo
->before
) {
12670 /* no changes took place in the directory prior to our link */
12672 mutex_enter(&VTOR4(vp
)->r_statev4_lock
);
12673 if (!VTOR4(vp
)->created_v4
) {
12674 mutex_exit(&VTOR4(vp
)->r_statev4_lock
);
12675 dnlc_update(dvp
, nm
, vp
);
12678 * XXX dont' update if the created_v4 flag
12681 mutex_exit(&VTOR4(vp
)->r_statev4_lock
);
12682 NFS4_DEBUG(nfs4_client_state_debug
, (CE_NOTE
,
12683 "nfs4_update_dircaches: don't"
12684 " update dnlc: created_v4 flag"));
12688 /* Another client modified directory - purge its dnlc cache */
12689 dnlc_purge_vp(dvp
);
12694 * The OPEN_CONFIRM operation confirms the sequence number used in OPENing a
12697 * The 'reopening_file' boolean should be set to TRUE if we are reopening this
12698 * file (ie: client recovery) and otherwise set to FALSE.
12700 * 'nfs4_start/end_op' should have been called by the proper (ie: not recovery
12701 * initiated) calling functions.
12703 * 'resend' is set to TRUE if this is a OPEN_CONFIRM issued as a result
12704 * of resending a 'lost' open request.
12706 * 'num_bseqid_retryp' makes sure we don't loop forever on a broken
12707 * server that hands out BAD_SEQID on open confirm.
12709 * Errors are returned via the nfs4_error_t parameter.
12712 nfs4open_confirm(vnode_t
*vp
, seqid4
*seqid
, stateid4
*stateid
, cred_t
*cr
,
12713 bool_t reopening_file
, bool_t
*retry_open
, nfs4_open_owner_t
*oop
,
12714 bool_t resend
, nfs4_error_t
*ep
, int *num_bseqid_retryp
)
12716 COMPOUND4args_clnt args
;
12717 COMPOUND4res_clnt res
;
12718 nfs_argop4 argop
[2];
12722 OPEN_CONFIRM4args
*open_confirm_args
;
12725 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
12727 mutex_enter(&oop
->oo_lock
);
12728 ASSERT(oop
->oo_seqid_inuse
);
12729 mutex_exit(&oop
->oo_lock
);
12732 recov_retry_confirm
:
12733 nfs4_error_zinit(ep
);
12734 *retry_open
= FALSE
;
12737 args
.ctag
= TAG_OPEN_CONFIRM_LOST
;
12739 args
.ctag
= TAG_OPEN_CONFIRM
;
12741 args
.array_len
= 2;
12742 args
.array
= argop
;
12744 /* putfh target fh */
12745 argop
[0].argop
= OP_CPUTFH
;
12746 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= VTOR4(vp
)->r_fh
;
12748 argop
[1].argop
= OP_OPEN_CONFIRM
;
12749 open_confirm_args
= &argop
[1].nfs_argop4_u
.opopen_confirm
;
12752 open_confirm_args
->seqid
= *seqid
;
12753 open_confirm_args
->open_stateid
= *stateid
;
12757 rfs4call(mi
, &args
, &res
, cr
, &doqueue
, 0, ep
);
12759 if (!ep
->error
&& nfs4_need_to_bump_seqid(&res
)) {
12760 nfs4_set_open_seqid((*seqid
), oop
, args
.ctag
);
12763 needrecov
= nfs4_needs_recovery(ep
, FALSE
, mi
->mi_vfsp
);
12764 if (!needrecov
&& ep
->error
)
12768 bool_t abort
= FALSE
;
12770 if (reopening_file
== FALSE
) {
12771 nfs4_bseqid_entry_t
*bsep
= NULL
;
12773 if (!ep
->error
&& res
.status
== NFS4ERR_BAD_SEQID
)
12774 bsep
= nfs4_create_bseqid_entry(oop
, NULL
,
12776 open_confirm_args
->seqid
);
12778 abort
= nfs4_start_recovery(ep
, VTOMI4(vp
), vp
, NULL
,
12779 NULL
, NULL
, OP_OPEN_CONFIRM
, bsep
, NULL
, NULL
);
12781 kmem_free(bsep
, sizeof (*bsep
));
12782 if (num_bseqid_retryp
&&
12783 --(*num_bseqid_retryp
) == 0)
12787 if ((ep
->error
== ETIMEDOUT
||
12788 res
.status
== NFS4ERR_RESOURCE
) &&
12789 abort
== FALSE
&& resend
== FALSE
) {
12791 (void) xdr_free(xdr_COMPOUND4res_clnt
,
12794 delay(SEC_TO_TICK(confirm_retry_sec
));
12795 goto recov_retry_confirm
;
12797 /* State may have changed so retry the entire OPEN op */
12798 if (abort
== FALSE
)
12799 *retry_open
= TRUE
;
12801 *retry_open
= FALSE
;
12803 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
12808 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
12812 resop
= &res
.array
[1]; /* open confirm res */
12813 bcopy(&resop
->nfs_resop4_u
.opopen_confirm
.open_stateid
,
12814 stateid
, sizeof (*stateid
));
12816 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)&res
);
12820 * Return the credentials associated with a client state object. The
12821 * caller is responsible for freeing the credentials.
12825 state_to_cred(nfs4_open_stream_t
*osp
)
12830 * It's ok to not lock the open stream and open owner to get
12831 * the oo_cred since this is only written once (upon creation)
12832 * and will not change.
12834 cr
= osp
->os_open_owner
->oo_cred
;
12843 * Find the sysid for the knetconfig associated with the given mi.
12845 static struct lm_sysid
*
12846 nfs4_find_sysid(mntinfo4_t
*mi
)
12848 ASSERT(nfs_zone() == mi
->mi_zone
);
12851 * Switch from RDMA knconf to original mount knconf
12853 return (lm_get_sysid(ORIG_KNCONF(mi
), &mi
->mi_curr_serv
->sv_addr
,
12854 mi
->mi_curr_serv
->sv_hostname
, NULL
));
12859 * Return a string version of the call type for easy reading.
12862 nfs4frlock_get_call_type(nfs4_lock_call_type_t ctype
)
12865 case NFS4_LCK_CTYPE_NORM
:
12867 case NFS4_LCK_CTYPE_RECLAIM
:
12868 return ("RECLAIM");
12869 case NFS4_LCK_CTYPE_RESEND
:
12871 case NFS4_LCK_CTYPE_REINSTATE
:
12872 return ("REINSTATE");
12874 cmn_err(CE_PANIC
, "nfs4frlock_get_call_type: got illegal "
12882 * Map the frlock cmd and lock type to the NFSv4 over-the-wire lock type
12883 * Unlock requests don't have an over-the-wire locktype, so we just return
12884 * something non-threatening.
12887 static nfs_lock_type4
12888 flk_to_locktype(int cmd
, int l_type
)
12890 ASSERT(l_type
== F_RDLCK
|| l_type
== F_WRLCK
|| l_type
== F_UNLCK
);
12896 if (cmd
== F_SETLK
)
12901 if (cmd
== F_SETLK
)
12904 return (WRITEW_LT
);
12906 panic("flk_to_locktype");
12911 * Do some preliminary checks for nfs4frlock.
12914 nfs4frlock_validate_args(int cmd
, flock64_t
*flk
, int flag
, vnode_t
*vp
,
12920 * If we are setting a lock, check that the file is opened
12921 * with the correct mode.
12923 if (cmd
== F_SETLK
|| cmd
== F_SETLKW
) {
12924 if ((flk
->l_type
== F_RDLCK
&& (flag
& FREAD
) == 0) ||
12925 (flk
->l_type
== F_WRLCK
&& (flag
& FWRITE
) == 0)) {
12926 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
12927 "nfs4frlock_validate_args: file was opened with "
12928 "incorrect mode"));
12933 /* Convert the offset. It may need to be restored before returning. */
12934 if (error
= convoff(vp
, flk
, 0, offset
)) {
12935 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
12936 "nfs4frlock_validate_args: convoff => error= %d\n",
12945 * Set the flock64's lm_sysid for nfs4frlock.
12948 nfs4frlock_get_sysid(struct lm_sysid
**lspp
, vnode_t
*vp
, flock64_t
*flk
)
12950 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
12952 /* Find the lm_sysid */
12953 *lspp
= nfs4_find_sysid(VTOMI4(vp
));
12955 if (*lspp
== NULL
) {
12956 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
12957 "nfs4frlock_get_sysid: no sysid, return ENOLCK"));
12961 flk
->l_sysid
= lm_sysidt(*lspp
);
12967 * Do the remaining preliminary setup for nfs4frlock.
12970 nfs4frlock_pre_setup(clock_t *tick_delayp
, nfs4_recov_state_t
*recov_statep
,
12971 flock64_t
*flk
, short *whencep
, vnode_t
*vp
, cred_t
*search_cr
,
12975 * set tick_delay to the base delay time.
12976 * (NFS4_BASE_WAIT_TIME is in secs)
12979 *tick_delayp
= drv_usectohz(NFS4_BASE_WAIT_TIME
* 1000 * 1000);
12982 * If lock is relative to EOF, we need the newest length of the
12983 * file. Therefore invalidate the ATTR_CACHE.
12986 *whencep
= flk
->l_whence
;
12988 if (*whencep
== 2) /* SEEK_END */
12989 PURGE_ATTRCACHE4(vp
);
12991 recov_statep
->rs_flags
= 0;
12992 recov_statep
->rs_num_retry_despite_err
= 0;
12993 *cred_otw
= nfs4_get_otw_cred(search_cr
, VTOMI4(vp
), NULL
);
12997 * Initialize and allocate the data structures necessary for
12998 * the nfs4frlock call.
12999 * Allocates argsp's op array.
13002 nfs4frlock_call_init(COMPOUND4args_clnt
*argsp
, COMPOUND4args_clnt
**argspp
,
13003 nfs_argop4
**argopp
, nfs4_op_hint_t
*op_hintp
, flock64_t
*flk
, int cmd
,
13004 bool_t
*retry
, bool_t
*did_start_fop
, COMPOUND4res_clnt
**respp
,
13005 bool_t
*skip_get_err
, nfs4_lost_rqst_t
*lost_rqstp
)
13007 int argoplist_size
;
13011 *did_start_fop
= FALSE
;
13012 *skip_get_err
= FALSE
;
13013 lost_rqstp
->lr_op
= 0;
13014 argoplist_size
= num_ops
* sizeof (nfs_argop4
);
13015 /* fill array with zero */
13016 *argopp
= kmem_zalloc(argoplist_size
, KM_SLEEP
);
13021 argsp
->array_len
= num_ops
;
13022 argsp
->array
= *argopp
;
13024 /* initialize in case of error; will get real value down below */
13025 argsp
->ctag
= TAG_NONE
;
13027 if ((cmd
== F_SETLK
|| cmd
== F_SETLKW
) && flk
->l_type
== F_UNLCK
)
13028 *op_hintp
= OH_LOCKU
;
13030 *op_hintp
= OH_OTHER
;
13034 * Call the nfs4_start_fop() for nfs4frlock, if necessary. Assign
13035 * the proper nfs4_server_t for this instance of nfs4frlock.
13036 * Returns 0 (success) or an errno value.
13039 nfs4frlock_start_call(nfs4_lock_call_type_t ctype
, vnode_t
*vp
,
13040 nfs4_op_hint_t op_hint
, nfs4_recov_state_t
*recov_statep
,
13041 bool_t
*did_start_fop
, bool_t
*startrecovp
)
13046 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
13048 if (ctype
== NFS4_LCK_CTYPE_NORM
) {
13049 error
= nfs4_start_fop(VTOMI4(vp
), vp
, NULL
, op_hint
,
13050 recov_statep
, startrecovp
);
13053 *did_start_fop
= TRUE
;
13055 *did_start_fop
= FALSE
;
13056 *startrecovp
= FALSE
;
13062 /* If the file failed recovery, just quit. */
13063 mutex_enter(&rp
->r_statelock
);
13064 if (rp
->r_flags
& R4RECOVERR
) {
13067 mutex_exit(&rp
->r_statelock
);
13074 * Setup the LOCK4/LOCKU4 arguments for resending a lost lock request. A
13075 * resend nfs4frlock call is initiated by the recovery framework.
13076 * Acquires the lop and oop seqid synchronization.
13079 nfs4frlock_setup_resend_lock_args(nfs4_lost_rqst_t
*resend_rqstp
,
13080 COMPOUND4args_clnt
*argsp
, nfs_argop4
*argop
, nfs4_lock_owner_t
**lopp
,
13081 nfs4_open_owner_t
**oopp
, nfs4_open_stream_t
**ospp
,
13082 LOCK4args
**lock_argsp
, LOCKU4args
**locku_argsp
)
13084 mntinfo4_t
*mi
= VTOMI4(resend_rqstp
->lr_vp
);
13087 NFS4_DEBUG((nfs4_lost_rqst_debug
|| nfs4_client_lock_debug
),
13089 "nfs4frlock_setup_resend_lock_args: have lost lock to resend"));
13090 ASSERT(resend_rqstp
!= NULL
);
13091 ASSERT(resend_rqstp
->lr_op
== OP_LOCK
||
13092 resend_rqstp
->lr_op
== OP_LOCKU
);
13094 *oopp
= resend_rqstp
->lr_oop
;
13095 if (resend_rqstp
->lr_oop
) {
13096 open_owner_hold(resend_rqstp
->lr_oop
);
13097 error
= nfs4_start_open_seqid_sync(resend_rqstp
->lr_oop
, mi
);
13098 ASSERT(error
== 0); /* recov thread always succeeds */
13101 /* Must resend this lost lock/locku request. */
13102 ASSERT(resend_rqstp
->lr_lop
!= NULL
);
13103 *lopp
= resend_rqstp
->lr_lop
;
13104 lock_owner_hold(resend_rqstp
->lr_lop
);
13105 error
= nfs4_start_lock_seqid_sync(resend_rqstp
->lr_lop
, mi
);
13106 ASSERT(error
== 0); /* recov thread always succeeds */
13108 *ospp
= resend_rqstp
->lr_osp
;
13110 open_stream_hold(resend_rqstp
->lr_osp
);
13112 if (resend_rqstp
->lr_op
== OP_LOCK
) {
13113 LOCK4args
*lock_args
;
13115 argop
->argop
= OP_LOCK
;
13116 *lock_argsp
= lock_args
= &argop
->nfs_argop4_u
.oplock
;
13117 lock_args
->locktype
= resend_rqstp
->lr_locktype
;
13118 lock_args
->reclaim
=
13119 (resend_rqstp
->lr_ctype
== NFS4_LCK_CTYPE_RECLAIM
);
13120 lock_args
->offset
= resend_rqstp
->lr_flk
->l_start
;
13121 lock_args
->length
= resend_rqstp
->lr_flk
->l_len
;
13122 if (lock_args
->length
== 0)
13123 lock_args
->length
= ~lock_args
->length
;
13124 nfs4_setup_lock_args(*lopp
, *oopp
, *ospp
,
13125 mi2clientid(mi
), &lock_args
->locker
);
13127 switch (resend_rqstp
->lr_ctype
) {
13128 case NFS4_LCK_CTYPE_RESEND
:
13129 argsp
->ctag
= TAG_LOCK_RESEND
;
13131 case NFS4_LCK_CTYPE_REINSTATE
:
13132 argsp
->ctag
= TAG_LOCK_REINSTATE
;
13134 case NFS4_LCK_CTYPE_RECLAIM
:
13135 argsp
->ctag
= TAG_LOCK_RECLAIM
;
13138 argsp
->ctag
= TAG_LOCK_UNKNOWN
;
13142 LOCKU4args
*locku_args
;
13143 nfs4_lock_owner_t
*lop
= resend_rqstp
->lr_lop
;
13145 argop
->argop
= OP_LOCKU
;
13146 *locku_argsp
= locku_args
= &argop
->nfs_argop4_u
.oplocku
;
13147 locku_args
->locktype
= READ_LT
;
13148 locku_args
->seqid
= lop
->lock_seqid
+ 1;
13149 mutex_enter(&lop
->lo_lock
);
13150 locku_args
->lock_stateid
= lop
->lock_stateid
;
13151 mutex_exit(&lop
->lo_lock
);
13152 locku_args
->offset
= resend_rqstp
->lr_flk
->l_start
;
13153 locku_args
->length
= resend_rqstp
->lr_flk
->l_len
;
13154 if (locku_args
->length
== 0)
13155 locku_args
->length
= ~locku_args
->length
;
13157 switch (resend_rqstp
->lr_ctype
) {
13158 case NFS4_LCK_CTYPE_RESEND
:
13159 argsp
->ctag
= TAG_LOCKU_RESEND
;
13161 case NFS4_LCK_CTYPE_REINSTATE
:
13162 argsp
->ctag
= TAG_LOCKU_REINSTATE
;
13165 argsp
->ctag
= TAG_LOCK_UNKNOWN
;
13172 * Setup the LOCKT4 arguments.
13175 nfs4frlock_setup_lockt_args(nfs4_lock_call_type_t ctype
, nfs_argop4
*argop
,
13176 LOCKT4args
**lockt_argsp
, COMPOUND4args_clnt
*argsp
, flock64_t
*flk
,
13179 LOCKT4args
*lockt_args
;
13181 ASSERT(nfs_zone() == VTOMI4(RTOV4(rp
))->mi_zone
);
13182 ASSERT(ctype
== NFS4_LCK_CTYPE_NORM
);
13183 argop
->argop
= OP_LOCKT
;
13184 argsp
->ctag
= TAG_LOCKT
;
13185 lockt_args
= &argop
->nfs_argop4_u
.oplockt
;
13188 * The locktype will be READ_LT unless it's
13189 * a write lock. We do this because the Solaris
13190 * system call allows the combination of
13191 * F_UNLCK and F_GETLK* and so in that case the
13192 * unlock is mapped to a read.
13194 if (flk
->l_type
== F_WRLCK
)
13195 lockt_args
->locktype
= WRITE_LT
;
13197 lockt_args
->locktype
= READ_LT
;
13199 lockt_args
->owner
.clientid
= mi2clientid(VTOMI4(RTOV4(rp
)));
13200 /* set the lock owner4 args */
13201 nfs4_setlockowner_args(&lockt_args
->owner
, rp
,
13202 ctype
== NFS4_LCK_CTYPE_NORM
? curproc
->p_pidp
->pid_id
:
13204 lockt_args
->offset
= flk
->l_start
;
13205 lockt_args
->length
= flk
->l_len
;
13206 if (flk
->l_len
== 0)
13207 lockt_args
->length
= ~lockt_args
->length
;
13209 *lockt_argsp
= lockt_args
;
13213 * If the client is holding a delegation, and the open stream to be used
13214 * with this lock request is a delegation open stream, then re-open the stream.
13215 * Sets the nfs4_error_t to all zeros unless the open stream has already
13216 * failed a reopen or we couldn't find the open stream. NFS4ERR_DELAY
13217 * means the caller should retry (like a recovery retry).
13220 nfs4frlock_check_deleg(vnode_t
*vp
, nfs4_error_t
*ep
, cred_t
*cr
, int lt
)
13222 open_delegation_type4 dt
;
13223 bool_t reopen_needed
, force
;
13224 nfs4_open_stream_t
*osp
;
13225 open_claim_type4 oclaim
;
13226 rnode4_t
*rp
= VTOR4(vp
);
13227 mntinfo4_t
*mi
= VTOMI4(vp
);
13229 ASSERT(nfs_zone() == mi
->mi_zone
);
13231 nfs4_error_zinit(ep
);
13233 mutex_enter(&rp
->r_statev4_lock
);
13234 dt
= rp
->r_deleg_type
;
13235 mutex_exit(&rp
->r_statev4_lock
);
13237 if (dt
!= OPEN_DELEGATE_NONE
) {
13238 nfs4_open_owner_t
*oop
;
13240 oop
= find_open_owner(cr
, NFS4_PERM_CREATED
, mi
);
13242 ep
->stat
= NFS4ERR_IO
;
13245 /* returns with 'os_sync_lock' held */
13246 osp
= find_open_stream(oop
, rp
);
13248 open_owner_rele(oop
);
13249 ep
->stat
= NFS4ERR_IO
;
13253 if (osp
->os_failed_reopen
) {
13254 NFS4_DEBUG((nfs4_open_stream_debug
||
13255 nfs4_client_lock_debug
), (CE_NOTE
,
13256 "nfs4frlock_check_deleg: os_failed_reopen set "
13257 "for osp %p, cr %p, rp %s", (void *)osp
,
13258 (void *)cr
, rnode4info(rp
)));
13259 mutex_exit(&osp
->os_sync_lock
);
13260 open_stream_rele(osp
, rp
);
13261 open_owner_rele(oop
);
13262 ep
->stat
= NFS4ERR_IO
;
13267 * Determine whether a reopen is needed. If this
13268 * is a delegation open stream, then send the open
13269 * to the server to give visibility to the open owner.
13270 * Even if it isn't a delegation open stream, we need
13271 * to check if the previous open CLAIM_DELEGATE_CUR
13275 reopen_needed
= osp
->os_delegation
||
13277 !(osp
->os_dc_openacc
& OPEN4_SHARE_ACCESS_READ
)) ||
13279 !(osp
->os_dc_openacc
& OPEN4_SHARE_ACCESS_WRITE
)));
13281 mutex_exit(&osp
->os_sync_lock
);
13282 open_owner_rele(oop
);
13284 if (reopen_needed
) {
13286 * Always use CLAIM_PREVIOUS after server reboot.
13287 * The server will reject CLAIM_DELEGATE_CUR if
13288 * it is used during the grace period.
13290 mutex_enter(&mi
->mi_lock
);
13291 if (mi
->mi_recovflags
& MI4R_SRV_REBOOT
) {
13292 oclaim
= CLAIM_PREVIOUS
;
13295 oclaim
= CLAIM_DELEGATE_CUR
;
13298 mutex_exit(&mi
->mi_lock
);
13300 nfs4_reopen(vp
, osp
, ep
, oclaim
, force
, FALSE
);
13301 if (ep
->error
== EAGAIN
) {
13302 nfs4_error_zinit(ep
);
13303 ep
->stat
= NFS4ERR_DELAY
;
13306 open_stream_rele(osp
, rp
);
13312 * Setup the LOCKU4 arguments.
13313 * Returns errors via the nfs4_error_t.
13314 * NFS4_OK no problems. *go_otwp is TRUE if call should go
13315 * over-the-wire. The caller must release the
13316 * reference on *lopp.
13317 * NFS4ERR_DELAY caller should retry (like recovery retry)
13318 * (other) unrecoverable error.
13321 nfs4frlock_setup_locku_args(nfs4_lock_call_type_t ctype
, nfs_argop4
*argop
,
13322 LOCKU4args
**locku_argsp
, flock64_t
*flk
,
13323 nfs4_lock_owner_t
**lopp
, nfs4_error_t
*ep
, COMPOUND4args_clnt
*argsp
,
13324 vnode_t
*vp
, int flag
, uoff_t offset
, cred_t
*cr
,
13325 bool_t
*skip_get_err
, bool_t
*go_otwp
)
13327 nfs4_lock_owner_t
*lop
= NULL
;
13328 LOCKU4args
*locku_args
;
13330 bool_t is_spec
= FALSE
;
13331 rnode4_t
*rp
= VTOR4(vp
);
13333 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
13334 ASSERT(ctype
== NFS4_LCK_CTYPE_NORM
);
13336 nfs4frlock_check_deleg(vp
, ep
, cr
, F_UNLCK
);
13337 if (ep
->error
|| ep
->stat
)
13340 argop
->argop
= OP_LOCKU
;
13341 if (ctype
== NFS4_LCK_CTYPE_REINSTATE
)
13342 argsp
->ctag
= TAG_LOCKU_REINSTATE
;
13344 argsp
->ctag
= TAG_LOCKU
;
13345 locku_args
= &argop
->nfs_argop4_u
.oplocku
;
13346 *locku_argsp
= locku_args
;
13348 /* locktype should be set to any legal value */
13349 locku_args
->locktype
= READ_LT
;
13351 pid
= ctype
== NFS4_LCK_CTYPE_NORM
? curproc
->p_pidp
->pid_id
:
13355 * Get the lock owner stateid. If no lock owner
13356 * exists, return success.
13358 lop
= find_lock_owner(rp
, pid
, LOWN_ANY
);
13360 if (lop
&& CLNT_ISSPECIAL(&lop
->lock_stateid
))
13362 if (!lop
|| is_spec
) {
13364 * No lock owner so no locks to unlock.
13365 * Return success. If there was a failed
13366 * reclaim earlier, the lock might still be
13367 * registered with the local locking code,
13368 * so notify it of the unlock.
13370 * If the lockowner is using a special stateid,
13371 * then the original lock request (that created
13372 * this lockowner) was never successful, so we
13373 * have no lock to undo OTW.
13375 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
13376 "nfs4frlock_setup_locku_args: LOCKU: no lock owner "
13377 "(%ld) so return success", (long)pid
));
13379 if (ctype
== NFS4_LCK_CTYPE_NORM
)
13380 flk
->l_pid
= curproc
->p_pid
;
13381 nfs4_register_lock_locally(vp
, flk
, flag
, offset
);
13383 * Release our hold and NULL out so final_cleanup
13384 * doesn't try to end a lock seqid sync we
13388 lock_owner_rele(lop
);
13391 *skip_get_err
= TRUE
;
13396 ep
->error
= nfs4_start_lock_seqid_sync(lop
, VTOMI4(vp
));
13397 if (ep
->error
== EAGAIN
) {
13398 lock_owner_rele(lop
);
13403 mutex_enter(&lop
->lo_lock
);
13404 locku_args
->lock_stateid
= lop
->lock_stateid
;
13405 mutex_exit(&lop
->lo_lock
);
13406 locku_args
->seqid
= lop
->lock_seqid
+ 1;
13408 /* leave the ref count on lop, rele after RPC call */
13410 locku_args
->offset
= flk
->l_start
;
13411 locku_args
->length
= flk
->l_len
;
13412 if (flk
->l_len
== 0)
13413 locku_args
->length
= ~locku_args
->length
;
13419 * Setup the LOCK4 arguments.
13421 * Returns errors via the nfs4_error_t.
13422 * NFS4_OK no problems
13423 * NFS4ERR_DELAY caller should retry (like recovery retry)
13424 * (other) unrecoverable error
13427 nfs4frlock_setup_lock_args(nfs4_lock_call_type_t ctype
, LOCK4args
**lock_argsp
,
13428 nfs4_open_owner_t
**oopp
, nfs4_open_stream_t
**ospp
,
13429 nfs4_lock_owner_t
**lopp
, nfs_argop4
*argop
, COMPOUND4args_clnt
*argsp
,
13430 flock64_t
*flk
, int cmd
, vnode_t
*vp
, cred_t
*cr
, nfs4_error_t
*ep
)
13432 LOCK4args
*lock_args
;
13433 nfs4_open_owner_t
*oop
= NULL
;
13434 nfs4_open_stream_t
*osp
= NULL
;
13435 nfs4_lock_owner_t
*lop
= NULL
;
13437 rnode4_t
*rp
= VTOR4(vp
);
13439 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
13441 nfs4frlock_check_deleg(vp
, ep
, cr
, flk
->l_type
);
13442 if (ep
->error
|| ep
->stat
!= NFS4_OK
)
13445 argop
->argop
= OP_LOCK
;
13446 if (ctype
== NFS4_LCK_CTYPE_NORM
)
13447 argsp
->ctag
= TAG_LOCK
;
13448 else if (ctype
== NFS4_LCK_CTYPE_RECLAIM
)
13449 argsp
->ctag
= TAG_RELOCK
;
13451 argsp
->ctag
= TAG_LOCK_REINSTATE
;
13452 lock_args
= &argop
->nfs_argop4_u
.oplock
;
13453 lock_args
->locktype
= flk_to_locktype(cmd
, flk
->l_type
);
13454 lock_args
->reclaim
= ctype
== NFS4_LCK_CTYPE_RECLAIM
? 1 : 0;
13456 * Get the lock owner. If no lock owner exists,
13457 * create a 'temporary' one and grab the open seqid
13458 * synchronization (which puts a hold on the open
13459 * owner and open stream).
13460 * This also grabs the lock seqid synchronization.
13462 pid
= ctype
== NFS4_LCK_CTYPE_NORM
? curproc
->p_pid
: flk
->l_pid
;
13464 nfs4_find_or_create_lock_owner(pid
, rp
, cr
, &oop
, &osp
, &lop
);
13466 if (ep
->stat
!= NFS4_OK
)
13469 nfs4_setup_lock_args(lop
, oop
, osp
, mi2clientid(VTOMI4(vp
)),
13470 &lock_args
->locker
);
13472 lock_args
->offset
= flk
->l_start
;
13473 lock_args
->length
= flk
->l_len
;
13474 if (flk
->l_len
== 0)
13475 lock_args
->length
= ~lock_args
->length
;
13476 *lock_argsp
= lock_args
;
13484 * After we get the reply from the server, record the proper information
13485 * for possible resend lock requests.
13488 nfs4frlock_save_lost_rqst(nfs4_lock_call_type_t ctype
, int error
,
13489 nfs_lock_type4 locktype
, nfs4_open_owner_t
*oop
,
13490 nfs4_open_stream_t
*osp
, nfs4_lock_owner_t
*lop
, flock64_t
*flk
,
13491 nfs4_lost_rqst_t
*lost_rqstp
, cred_t
*cr
, vnode_t
*vp
)
13493 bool_t unlock
= (flk
->l_type
== F_UNLCK
);
13495 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
13496 ASSERT(ctype
== NFS4_LCK_CTYPE_NORM
||
13497 ctype
== NFS4_LCK_CTYPE_REINSTATE
);
13499 if (error
!= 0 && !unlock
) {
13500 NFS4_DEBUG((nfs4_lost_rqst_debug
||
13501 nfs4_client_lock_debug
), (CE_NOTE
,
13502 "nfs4frlock_save_lost_rqst: set lo_pending_rqsts to 1 "
13503 " for lop %p", (void *)lop
));
13504 ASSERT(lop
!= NULL
);
13505 mutex_enter(&lop
->lo_lock
);
13506 lop
->lo_pending_rqsts
= 1;
13507 mutex_exit(&lop
->lo_lock
);
13510 lost_rqstp
->lr_putfirst
= FALSE
;
13511 lost_rqstp
->lr_op
= 0;
13514 * For lock/locku requests, we treat EINTR as ETIMEDOUT for
13515 * recovery purposes so that the lock request that was sent
13516 * can be saved and re-issued later. Ditto for EIO from a forced
13517 * unmount. This is done to have the client's local locking state
13518 * match the v4 server's state; that is, the request was
13519 * potentially received and accepted by the server but the client
13520 * thinks it was not.
13522 if (error
== ETIMEDOUT
|| error
== EINTR
||
13523 NFS4_FRC_UNMT_ERR(error
, vp
->v_vfsp
)) {
13524 NFS4_DEBUG((nfs4_lost_rqst_debug
||
13525 nfs4_client_lock_debug
), (CE_NOTE
,
13526 "nfs4frlock_save_lost_rqst: got a lost %s lock for "
13527 "lop %p oop %p osp %p", unlock
? "LOCKU" : "LOCK",
13528 (void *)lop
, (void *)oop
, (void *)osp
));
13530 lost_rqstp
->lr_op
= OP_LOCKU
;
13532 lost_rqstp
->lr_op
= OP_LOCK
;
13533 lost_rqstp
->lr_locktype
= locktype
;
13536 * Objects are held and rele'd via the recovery code.
13537 * See nfs4_save_lost_rqst.
13539 lost_rqstp
->lr_vp
= vp
;
13540 lost_rqstp
->lr_dvp
= NULL
;
13541 lost_rqstp
->lr_oop
= oop
;
13542 lost_rqstp
->lr_osp
= osp
;
13543 lost_rqstp
->lr_lop
= lop
;
13544 lost_rqstp
->lr_cr
= cr
;
13546 case NFS4_LCK_CTYPE_NORM
:
13547 flk
->l_pid
= ttoproc(curthread
)->p_pid
;
13548 lost_rqstp
->lr_ctype
= NFS4_LCK_CTYPE_RESEND
;
13550 case NFS4_LCK_CTYPE_REINSTATE
:
13551 lost_rqstp
->lr_putfirst
= TRUE
;
13552 lost_rqstp
->lr_ctype
= ctype
;
13557 lost_rqstp
->lr_flk
= flk
;
13562 * Update lop's seqid. Also update the seqid stored in a resend request,
13563 * if any. (Some recovery errors increment the seqid, and we may have to
13564 * send the resend request again.)
13568 nfs4frlock_bump_seqid(LOCK4args
*lock_args
, LOCKU4args
*locku_args
,
13569 nfs4_open_owner_t
*oop
, nfs4_lock_owner_t
*lop
, nfs4_tag_type_t tag_type
)
13572 if (lock_args
->locker
.new_lock_owner
== TRUE
)
13573 nfs4_get_and_set_next_open_seqid(oop
, tag_type
);
13575 ASSERT(lop
->lo_flags
& NFS4_LOCK_SEQID_INUSE
);
13576 nfs4_set_lock_seqid(lop
->lock_seqid
+ 1, lop
);
13578 } else if (locku_args
) {
13579 ASSERT(lop
->lo_flags
& NFS4_LOCK_SEQID_INUSE
);
13580 nfs4_set_lock_seqid(lop
->lock_seqid
+1, lop
);
13585 * Calls nfs4_end_fop, drops the seqid syncs, and frees up the
13586 * COMPOUND4 args/res for calls that need to retry.
13587 * Switches the *cred_otwp to base_cr.
13590 nfs4frlock_check_access(vnode_t
*vp
, nfs4_op_hint_t op_hint
,
13591 nfs4_recov_state_t
*recov_statep
, int needrecov
, bool_t
*did_start_fop
,
13592 COMPOUND4args_clnt
**argspp
, COMPOUND4res_clnt
**respp
, int error
,
13593 nfs4_lock_owner_t
**lopp
, nfs4_open_owner_t
**oopp
,
13594 nfs4_open_stream_t
**ospp
, cred_t
*base_cr
, cred_t
**cred_otwp
)
13596 nfs4_open_owner_t
*oop
= *oopp
;
13597 nfs4_open_stream_t
*osp
= *ospp
;
13598 nfs4_lock_owner_t
*lop
= *lopp
;
13599 nfs_argop4
*argop
= (*argspp
)->array
;
13601 if (*did_start_fop
) {
13602 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, op_hint
, recov_statep
,
13604 *did_start_fop
= FALSE
;
13606 ASSERT((*argspp
)->array_len
== 2);
13607 if (argop
[1].argop
== OP_LOCK
)
13608 nfs4args_lock_free(&argop
[1]);
13609 else if (argop
[1].argop
== OP_LOCKT
)
13610 nfs4args_lockt_free(&argop
[1]);
13611 kmem_free(argop
, 2 * sizeof (nfs_argop4
));
13613 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)*respp
);
13618 nfs4_end_lock_seqid_sync(lop
);
13619 lock_owner_rele(lop
);
13623 /* need to free up the reference on osp for lock args */
13625 open_stream_rele(osp
, VTOR4(vp
));
13629 /* need to free up the reference on oop for lock args */
13631 nfs4_end_open_seqid_sync(oop
);
13632 open_owner_rele(oop
);
13636 crfree(*cred_otwp
);
13637 *cred_otwp
= base_cr
;
13638 crhold(*cred_otwp
);
13642 * Function to process the client's recovery for nfs4frlock.
13643 * Returns TRUE if we should retry the lock request; FALSE otherwise.
13645 * Calls nfs4_end_fop, drops the seqid syncs, and frees up the
13646 * COMPOUND4 args/res for calls that need to retry.
13648 * Note: the rp's r_lkserlock is *not* dropped during this path.
13651 nfs4frlock_recovery(int needrecov
, nfs4_error_t
*ep
,
13652 COMPOUND4args_clnt
**argspp
, COMPOUND4res_clnt
**respp
,
13653 LOCK4args
*lock_args
, LOCKU4args
*locku_args
,
13654 nfs4_open_owner_t
**oopp
, nfs4_open_stream_t
**ospp
,
13655 nfs4_lock_owner_t
**lopp
, rnode4_t
*rp
, vnode_t
*vp
,
13656 nfs4_recov_state_t
*recov_statep
, nfs4_op_hint_t op_hint
,
13657 bool_t
*did_start_fop
, nfs4_lost_rqst_t
*lost_rqstp
, flock64_t
*flk
)
13659 nfs4_open_owner_t
*oop
= *oopp
;
13660 nfs4_open_stream_t
*osp
= *ospp
;
13661 nfs4_lock_owner_t
*lop
= *lopp
;
13663 bool_t abort
, retry
;
13665 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
13666 ASSERT((*argspp
) != NULL
);
13667 ASSERT((*respp
) != NULL
);
13668 if (lock_args
|| locku_args
)
13669 ASSERT(lop
!= NULL
);
13671 NFS4_DEBUG((nfs4_client_lock_debug
|| nfs4_client_recov_debug
),
13672 (CE_NOTE
, "nfs4frlock_recovery: initiating recovery\n"));
13677 nfs4_bseqid_entry_t
*bsep
= NULL
;
13680 op
= lock_args
? OP_LOCK
: locku_args
? OP_LOCKU
: OP_LOCKT
;
13682 if (!ep
->error
&& ep
->stat
== NFS4ERR_BAD_SEQID
) {
13686 if (lock_args
->locker
.new_lock_owner
== TRUE
)
13687 seqid
= lock_args
->locker
.locker4_u
.
13688 open_owner
.open_seqid
;
13690 seqid
= lock_args
->locker
.locker4_u
.
13691 lock_owner
.lock_seqid
;
13692 } else if (locku_args
) {
13693 seqid
= locku_args
->seqid
;
13698 bsep
= nfs4_create_bseqid_entry(oop
, lop
, vp
,
13699 flk
->l_pid
, (*argspp
)->ctag
, seqid
);
13702 abort
= nfs4_start_recovery(ep
, VTOMI4(vp
), vp
, NULL
, NULL
,
13703 (lost_rqstp
&& (lost_rqstp
->lr_op
== OP_LOCK
||
13704 lost_rqstp
->lr_op
== OP_LOCKU
)) ? lost_rqstp
:
13705 NULL
, op
, bsep
, NULL
, NULL
);
13708 kmem_free(bsep
, sizeof (*bsep
));
13712 * Return that we do not want to retry the request for 3 cases:
13713 * 1. If we received EINTR or are bailing out because of a forced
13714 * unmount, we came into this code path just for the sake of
13715 * initiating recovery, we now need to return the error.
13716 * 2. If we have aborted recovery.
13717 * 3. We received NFS4ERR_BAD_SEQID.
13719 if (ep
->error
== EINTR
|| NFS4_FRC_UNMT_ERR(ep
->error
, vp
->v_vfsp
) ||
13720 abort
== TRUE
|| (ep
->error
== 0 && ep
->stat
== NFS4ERR_BAD_SEQID
))
13723 if (*did_start_fop
== TRUE
) {
13724 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, op_hint
, recov_statep
,
13726 *did_start_fop
= FALSE
;
13729 if (retry
== TRUE
) {
13732 argop
= (*argspp
)->array
;
13733 ASSERT((*argspp
)->array_len
== 2);
13735 if (argop
[1].argop
== OP_LOCK
)
13736 nfs4args_lock_free(&argop
[1]);
13737 else if (argop
[1].argop
== OP_LOCKT
)
13738 nfs4args_lockt_free(&argop
[1]);
13739 kmem_free(argop
, 2 * sizeof (nfs_argop4
));
13741 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)*respp
);
13747 nfs4_end_lock_seqid_sync(lop
);
13748 lock_owner_rele(lop
);
13753 /* need to free up the reference on osp for lock args */
13755 open_stream_rele(osp
, rp
);
13759 /* need to free up the reference on oop for lock args */
13761 nfs4_end_open_seqid_sync(oop
);
13762 open_owner_rele(oop
);
13770 * Handles the successful reply from the server for nfs4frlock.
13773 nfs4frlock_results_ok(nfs4_lock_call_type_t ctype
, int cmd
, flock64_t
*flk
,
13774 vnode_t
*vp
, int flag
, uoff_t offset
,
13775 nfs4_lost_rqst_t
*resend_rqstp
)
13777 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
13778 if ((cmd
== F_SETLK
|| cmd
== F_SETLKW
) &&
13779 (flk
->l_type
== F_RDLCK
|| flk
->l_type
== F_WRLCK
)) {
13780 if (ctype
== NFS4_LCK_CTYPE_NORM
) {
13781 flk
->l_pid
= ttoproc(curthread
)->p_pid
;
13783 * We do not register lost locks locally in
13784 * the 'resend' case since the user/application
13785 * doesn't think we have the lock.
13787 ASSERT(!resend_rqstp
);
13788 nfs4_register_lock_locally(vp
, flk
, flag
, offset
);
13794 * Handle the DENIED reply from the server for nfs4frlock.
13795 * Returns TRUE if we should retry the request; FALSE otherwise.
13797 * Calls nfs4_end_fop, drops the seqid syncs, and frees up the
13798 * COMPOUND4 args/res for calls that need to retry. Can also
13799 * drop and regrab the r_lkserlock.
13802 nfs4frlock_results_denied(nfs4_lock_call_type_t ctype
, LOCK4args
*lock_args
,
13803 LOCKT4args
*lockt_args
, nfs4_open_owner_t
**oopp
,
13804 nfs4_open_stream_t
**ospp
, nfs4_lock_owner_t
**lopp
, int cmd
,
13805 vnode_t
*vp
, flock64_t
*flk
, nfs4_op_hint_t op_hint
,
13806 nfs4_recov_state_t
*recov_statep
, int needrecov
,
13807 COMPOUND4args_clnt
**argspp
, COMPOUND4res_clnt
**respp
,
13808 clock_t *tick_delayp
, short *whencep
, int *errorp
,
13809 nfs_resop4
*resop
, cred_t
*cr
, bool_t
*did_start_fop
,
13810 bool_t
*skip_get_err
)
13812 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
13815 nfs4_open_owner_t
*oop
= *oopp
;
13816 nfs4_open_stream_t
*osp
= *ospp
;
13817 nfs4_lock_owner_t
*lop
= *lopp
;
13821 * Blocking lock needs to sleep and retry from the request.
13823 * Do not block and wait for 'resend' or 'reinstate'
13824 * lock requests, just return the error.
13826 * Note: reclaim requests have cmd == F_SETLK, not F_SETLKW.
13828 if (cmd
== F_SETLKW
) {
13829 rnode4_t
*rp
= VTOR4(vp
);
13830 nfs_argop4
*argop
= (*argspp
)->array
;
13832 ASSERT(ctype
== NFS4_LCK_CTYPE_NORM
);
13834 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, op_hint
,
13835 recov_statep
, needrecov
);
13836 *did_start_fop
= FALSE
;
13837 ASSERT((*argspp
)->array_len
== 2);
13838 if (argop
[1].argop
== OP_LOCK
)
13839 nfs4args_lock_free(&argop
[1]);
13840 else if (argop
[1].argop
== OP_LOCKT
)
13841 nfs4args_lockt_free(&argop
[1]);
13842 kmem_free(argop
, 2 * sizeof (nfs_argop4
));
13844 (void) xdr_free(xdr_COMPOUND4res_clnt
,
13848 nfs4_end_lock_seqid_sync(lop
);
13849 lock_owner_rele(lop
);
13852 open_stream_rele(osp
, rp
);
13856 nfs4_end_open_seqid_sync(oop
);
13857 open_owner_rele(oop
);
13861 nfs_rw_exit(&rp
->r_lkserlock
);
13863 intr
= nfs4_block_and_wait(tick_delayp
, rp
);
13866 (void) nfs_rw_enter_sig(&rp
->r_lkserlock
,
13872 (void) nfs_rw_enter_sig(&rp
->r_lkserlock
,
13876 * Make sure we are still safe to lock with
13877 * regards to mmapping.
13879 if (!nfs4_safelock(vp
, flk
, cr
)) {
13886 if (ctype
== NFS4_LCK_CTYPE_NORM
)
13888 *skip_get_err
= TRUE
;
13892 } else if (lockt_args
) {
13893 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
13894 "nfs4frlock_results_denied: OP_LOCKT DENIED"));
13896 denied_to_flk(&resop
->nfs_resop4_u
.oplockt
.denied
,
13899 /* according to NLM code */
13902 *skip_get_err
= TRUE
;
13909 * Handles all NFS4 errors besides NFS4_OK and NFS4ERR_DENIED for nfs4frlock.
13912 nfs4frlock_results_default(COMPOUND4res_clnt
*resp
, int *errorp
)
13914 switch (resp
->status
) {
13915 case NFS4ERR_ACCESS
:
13916 case NFS4ERR_ADMIN_REVOKED
:
13917 case NFS4ERR_BADHANDLE
:
13918 case NFS4ERR_BAD_RANGE
:
13919 case NFS4ERR_BAD_SEQID
:
13920 case NFS4ERR_BAD_STATEID
:
13921 case NFS4ERR_BADXDR
:
13922 case NFS4ERR_DEADLOCK
:
13923 case NFS4ERR_DELAY
:
13924 case NFS4ERR_EXPIRED
:
13925 case NFS4ERR_FHEXPIRED
:
13926 case NFS4ERR_GRACE
:
13927 case NFS4ERR_INVAL
:
13928 case NFS4ERR_ISDIR
:
13929 case NFS4ERR_LEASE_MOVED
:
13930 case NFS4ERR_LOCK_NOTSUPP
:
13931 case NFS4ERR_LOCK_RANGE
:
13932 case NFS4ERR_MOVED
:
13933 case NFS4ERR_NOFILEHANDLE
:
13934 case NFS4ERR_NO_GRACE
:
13935 case NFS4ERR_OLD_STATEID
:
13936 case NFS4ERR_OPENMODE
:
13937 case NFS4ERR_RECLAIM_BAD
:
13938 case NFS4ERR_RECLAIM_CONFLICT
:
13939 case NFS4ERR_RESOURCE
:
13940 case NFS4ERR_SERVERFAULT
:
13941 case NFS4ERR_STALE
:
13942 case NFS4ERR_STALE_CLIENTID
:
13943 case NFS4ERR_STALE_STATEID
:
13946 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
13947 "nfs4frlock_results_default: got unrecognizable "
13948 "res.status %d", resp
->status
));
13949 *errorp
= NFS4ERR_INVAL
;
13954 * The lock request was successful, so update the client's state.
13957 nfs4frlock_update_state(LOCK4args
*lock_args
, LOCKU4args
*locku_args
,
13958 LOCKT4args
*lockt_args
, nfs_resop4
*resop
, nfs4_lock_owner_t
*lop
,
13959 vnode_t
*vp
, flock64_t
*flk
, cred_t
*cr
,
13960 nfs4_lost_rqst_t
*resend_rqstp
)
13962 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
13965 LOCK4res
*lock_res
;
13967 lock_res
= &resop
->nfs_resop4_u
.oplock
;
13968 /* update the stateid with server's response */
13970 if (lock_args
->locker
.new_lock_owner
== TRUE
) {
13971 mutex_enter(&lop
->lo_lock
);
13972 lop
->lo_just_created
= NFS4_PERM_CREATED
;
13973 mutex_exit(&lop
->lo_lock
);
13976 nfs4_set_lock_stateid(lop
, lock_res
->LOCK4res_u
.lock_stateid
);
13979 * If the lock was the result of a resending a lost
13980 * request, we've synched up the stateid and seqid
13981 * with the server, but now the server might be out of sync
13982 * with what the application thinks it has for locks.
13983 * Clean that up here. It's unclear whether we should do
13984 * this even if the filesystem has been forcibly unmounted.
13985 * For most servers, it's probably wasted effort, but
13986 * RFC 7530 lets servers require that unlocks exactly match
13987 * the locks that are held.
13989 if (resend_rqstp
!= NULL
&&
13990 resend_rqstp
->lr_ctype
!= NFS4_LCK_CTYPE_REINSTATE
) {
13991 nfs4_reinstitute_local_lock_state(vp
, flk
, cr
, lop
);
13995 } else if (locku_args
) {
13996 LOCKU4res
*locku_res
;
13998 locku_res
= &resop
->nfs_resop4_u
.oplocku
;
14000 /* Update the stateid with the server's response */
14001 nfs4_set_lock_stateid(lop
, locku_res
->lock_stateid
);
14002 } else if (lockt_args
) {
14003 /* Switch the lock type to express success, see fcntl */
14004 flk
->l_type
= F_UNLCK
;
14010 * Do final cleanup before exiting nfs4frlock.
14011 * Calls nfs4_end_fop, drops the seqid syncs, and frees up the
14012 * COMPOUND4 args/res for calls that haven't already.
14015 nfs4frlock_final_cleanup(nfs4_lock_call_type_t ctype
, COMPOUND4args_clnt
*argsp
,
14016 COMPOUND4res_clnt
*resp
, vnode_t
*vp
, nfs4_op_hint_t op_hint
,
14017 nfs4_recov_state_t
*recov_statep
, int needrecov
, nfs4_open_owner_t
*oop
,
14018 nfs4_open_stream_t
*osp
, nfs4_lock_owner_t
*lop
, flock64_t
*flk
,
14019 short whence
, uoff_t offset
, struct lm_sysid
*ls
,
14020 int *errorp
, LOCK4args
*lock_args
, LOCKU4args
*locku_args
,
14021 bool_t did_start_fop
, bool_t skip_get_err
,
14022 cred_t
*cred_otw
, cred_t
*cred
)
14024 mntinfo4_t
*mi
= VTOMI4(vp
);
14025 rnode4_t
*rp
= VTOR4(vp
);
14026 int error
= *errorp
;
14028 int do_flush_pages
= 0;
14030 ASSERT(nfs_zone() == mi
->mi_zone
);
14032 * The client recovery code wants the raw status information,
14033 * so don't map the NFS status code to an errno value for
14034 * non-normal call types.
14036 if (ctype
== NFS4_LCK_CTYPE_NORM
) {
14037 if (*errorp
== 0 && resp
!= NULL
&& skip_get_err
== FALSE
)
14038 *errorp
= geterrno4(resp
->status
);
14039 if (did_start_fop
== TRUE
)
14040 nfs4_end_fop(mi
, vp
, NULL
, op_hint
, recov_statep
,
14044 * We've established a new lock on the server, so invalidate
14045 * the pages associated with the vnode to get the most up to
14046 * date pages from the server after acquiring the lock. We
14047 * want to be sure that the read operation gets the newest data.
14049 * We used to do this in nfs4frlock_results_ok but that doesn't
14050 * work since fop_putpage can call nfs4_commit which calls
14051 * nfs4_start_fop. We flush the pages below after calling
14052 * nfs4_end_fop above
14053 * The flush of the page cache must be done after
14054 * nfs4_end_open_seqid_sync() to avoid a 4-way hang.
14056 if (!error
&& resp
&& resp
->status
== NFS4_OK
)
14057 do_flush_pages
= 1;
14060 ASSERT(argsp
->array_len
== 2);
14061 argop
= argsp
->array
;
14062 if (argop
[1].argop
== OP_LOCK
)
14063 nfs4args_lock_free(&argop
[1]);
14064 else if (argop
[1].argop
== OP_LOCKT
)
14065 nfs4args_lockt_free(&argop
[1]);
14066 kmem_free(argop
, 2 * sizeof (nfs_argop4
));
14068 (void) xdr_free(xdr_COMPOUND4res_clnt
, (caddr_t
)resp
);
14071 /* free the reference on the lock owner */
14073 nfs4_end_lock_seqid_sync(lop
);
14074 lock_owner_rele(lop
);
14077 /* need to free up the reference on osp for lock args */
14079 open_stream_rele(osp
, rp
);
14081 /* need to free up the reference on oop for lock args */
14083 nfs4_end_open_seqid_sync(oop
);
14084 open_owner_rele(oop
);
14087 if (do_flush_pages
)
14088 nfs4_flush_pages(vp
, cred
);
14090 (void) convoff(vp
, flk
, whence
, offset
);
14095 * Record debug information in the event we get EINVAL.
14097 mutex_enter(&mi
->mi_lock
);
14098 if (*errorp
== EINVAL
&& (lock_args
|| locku_args
) &&
14099 (!(mi
->mi_flags
& MI4_POSIX_LOCK
))) {
14100 if (!(mi
->mi_flags
& MI4_LOCK_DEBUG
)) {
14101 zcmn_err(getzoneid(), CE_NOTE
,
14102 "%s operation failed with "
14103 "EINVAL probably since the server, %s,"
14104 " doesn't support POSIX style locking",
14105 lock_args
? "LOCK" : "LOCKU",
14106 mi
->mi_curr_serv
->sv_hostname
);
14107 mi
->mi_flags
|= MI4_LOCK_DEBUG
;
14110 mutex_exit(&mi
->mi_lock
);
14117 * This calls the server and the local locking code.
14119 * Client locks are registerred locally by oring the sysid with
14120 * LM_SYSID_CLIENT. The server registers locks locally using just the sysid.
14121 * We need to distinguish between the two to avoid collision in case one
14122 * machine is used as both client and server.
14124 * Blocking lock requests will continually retry to acquire the lock
14127 * The ctype is defined as follows:
14128 * NFS4_LCK_CTYPE_NORM: normal lock request.
14130 * NFS4_LCK_CTYPE_RECLAIM: bypass the usual calls for synchronizing with client
14131 * recovery, get the pid from flk instead of curproc, and don't reregister
14132 * the lock locally.
14134 * NFS4_LCK_CTYPE_RESEND: same as NFS4_LCK_CTYPE_RECLAIM, with the addition
14135 * that we will use the information passed in via resend_rqstp to setup the
14136 * lock/locku request. This resend is the exact same request as the 'lost
14137 * lock', and is initiated by the recovery framework. A successful resend
14138 * request can initiate one or more reinstate requests.
14140 * NFS4_LCK_CTYPE_REINSTATE: same as NFS4_LCK_CTYPE_RESEND, except that it
14141 * does not trigger additional reinstate requests. This lock call type is
14142 * set for setting the v4 server's locking state back to match what the
14143 * client's local locking state is in the event of a received 'lost lock'.
14145 * Errors are returned via the nfs4_error_t parameter.
14148 nfs4frlock(nfs4_lock_call_type_t ctype
, vnode_t
*vp
, int cmd
, flock64_t
*flk
,
14149 int flag
, uoff_t offset
, cred_t
*cr
, nfs4_error_t
*ep
,
14150 nfs4_lost_rqst_t
*resend_rqstp
, int *did_reclaimp
)
14152 COMPOUND4args_clnt args
, *argsp
= NULL
;
14153 COMPOUND4res_clnt res
, *resp
= NULL
;
14158 clock_t tick_delay
; /* delay in clock ticks */
14159 struct lm_sysid
*ls
;
14160 LOCK4args
*lock_args
= NULL
;
14161 LOCKU4args
*locku_args
= NULL
;
14162 LOCKT4args
*lockt_args
= NULL
;
14163 nfs4_open_owner_t
*oop
= NULL
;
14164 nfs4_open_stream_t
*osp
= NULL
;
14165 nfs4_lock_owner_t
*lop
= NULL
;
14166 bool_t needrecov
= FALSE
;
14167 nfs4_recov_state_t recov_state
;
14169 nfs4_op_hint_t op_hint
;
14170 nfs4_lost_rqst_t lost_rqst
;
14171 bool_t retry
= FALSE
;
14172 bool_t did_start_fop
= FALSE
;
14173 bool_t skip_get_err
= FALSE
;
14174 cred_t
*cred_otw
= NULL
;
14175 bool_t recovonly
; /* just queue request */
14176 int frc_no_reclaim
= 0;
14181 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
14184 name
= fn_name(VTOSV(vp
)->sv_name
);
14185 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
, "nfs4frlock: "
14186 "%s: cmd %d, type %d, offset %llu, start %"PRIx64
", "
14187 "length %"PRIu64
", pid %d, sysid %d, call type %s, "
14188 "resend request %s", name
, cmd
, flk
->l_type
, offset
, flk
->l_start
,
14189 flk
->l_len
, ctype
== NFS4_LCK_CTYPE_NORM
? curproc
->p_pid
:
14190 flk
->l_pid
, flk
->l_sysid
, nfs4frlock_get_call_type(ctype
),
14191 resend_rqstp
? "TRUE" : "FALSE"));
14192 kmem_free(name
, MAXNAMELEN
);
14195 nfs4_error_zinit(ep
);
14196 ep
->error
= nfs4frlock_validate_args(cmd
, flk
, flag
, vp
, offset
);
14199 ep
->error
= nfs4frlock_get_sysid(&ls
, vp
, flk
);
14202 nfs4frlock_pre_setup(&tick_delay
, &recov_state
, flk
, &whence
,
14203 vp
, cr
, &cred_otw
);
14206 nfs4frlock_call_init(&args
, &argsp
, &argop
, &op_hint
, flk
, cmd
,
14207 &retry
, &did_start_fop
, &resp
, &skip_get_err
, &lost_rqst
);
14210 ep
->error
= nfs4frlock_start_call(ctype
, vp
, op_hint
, &recov_state
,
14211 &did_start_fop
, &recovonly
);
14218 * Leave the request for the recovery system to deal with.
14220 ASSERT(ctype
== NFS4_LCK_CTYPE_NORM
);
14221 ASSERT(cmd
!= F_GETLK
);
14222 ASSERT(flk
->l_type
== F_UNLCK
);
14224 nfs4_error_init(ep
, EINTR
);
14226 lop
= find_lock_owner(rp
, curproc
->p_pid
, LOWN_ANY
);
14228 nfs4frlock_save_lost_rqst(ctype
, ep
->error
, READ_LT
,
14229 NULL
, NULL
, lop
, flk
, &lost_rqst
, cr
, vp
);
14230 (void) nfs4_start_recovery(ep
,
14231 VTOMI4(vp
), vp
, NULL
, NULL
,
14232 (lost_rqst
.lr_op
== OP_LOCK
||
14233 lost_rqst
.lr_op
== OP_LOCKU
) ?
14234 &lost_rqst
: NULL
, OP_LOCKU
, NULL
, NULL
, NULL
);
14235 lock_owner_rele(lop
);
14238 flk
->l_pid
= curproc
->p_pid
;
14239 nfs4_register_lock_locally(vp
, flk
, flag
, offset
);
14243 /* putfh directory fh */
14244 argop
[0].argop
= OP_CPUTFH
;
14245 argop
[0].nfs_argop4_u
.opcputfh
.sfh
= rp
->r_fh
;
14248 * Set up the over-the-wire arguments and get references to the
14252 if (ctype
== NFS4_LCK_CTYPE_RESEND
||
14253 ctype
== NFS4_LCK_CTYPE_REINSTATE
) {
14254 nfs4frlock_setup_resend_lock_args(resend_rqstp
, argsp
,
14255 &argop
[1], &lop
, &oop
, &osp
, &lock_args
, &locku_args
);
14257 bool_t go_otw
= TRUE
;
14259 ASSERT(resend_rqstp
== NULL
);
14263 nfs4frlock_setup_lockt_args(ctype
, &argop
[1],
14264 &lockt_args
, argsp
, flk
, rp
);
14268 if (flk
->l_type
== F_UNLCK
)
14269 nfs4frlock_setup_locku_args(ctype
,
14270 &argop
[1], &locku_args
, flk
,
14272 vp
, flag
, offset
, cr
,
14273 &skip_get_err
, &go_otw
);
14275 nfs4frlock_setup_lock_args(ctype
,
14276 &lock_args
, &oop
, &osp
, &lop
, &argop
[1],
14277 argsp
, flk
, cmd
, vp
, cr
, ep
);
14282 switch (ep
->stat
) {
14285 case NFS4ERR_DELAY
:
14286 /* recov thread never gets this error */
14287 ASSERT(resend_rqstp
== NULL
);
14288 ASSERT(did_start_fop
);
14290 nfs4_end_fop(VTOMI4(vp
), vp
, NULL
, op_hint
,
14291 &recov_state
, TRUE
);
14292 did_start_fop
= FALSE
;
14293 if (argop
[1].argop
== OP_LOCK
)
14294 nfs4args_lock_free(&argop
[1]);
14295 else if (argop
[1].argop
== OP_LOCKT
)
14296 nfs4args_lockt_free(&argop
[1]);
14297 kmem_free(argop
, 2 * sizeof (nfs_argop4
));
14306 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
14307 "nfs4_frlock: invalid cmd %d", cmd
));
14308 ep
->error
= EINVAL
;
14316 /* XXX should we use the local reclock as a cache ? */
14318 * Unregister the lock with the local locking code before
14319 * contacting the server. This avoids a potential race where
14320 * another process gets notified that it has been granted a lock
14321 * before we can unregister ourselves locally.
14323 if ((cmd
== F_SETLK
|| cmd
== F_SETLKW
) && flk
->l_type
== F_UNLCK
) {
14324 if (ctype
== NFS4_LCK_CTYPE_NORM
)
14325 flk
->l_pid
= ttoproc(curthread
)->p_pid
;
14326 nfs4_register_lock_locally(vp
, flk
, flag
, offset
);
14330 * Send the server the lock request. Continually loop with a delay
14331 * if get error NFS4ERR_DENIED (for blocking locks) or NFS4ERR_GRACE.
14335 NFS4_DEBUG((nfs4_client_call_debug
|| nfs4_client_lock_debug
),
14337 "nfs4frlock: %s call, rp %s", needrecov
? "recov" : "first",
14340 if (lock_args
&& frc_no_reclaim
) {
14341 ASSERT(ctype
== NFS4_LCK_CTYPE_RECLAIM
);
14342 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
14343 "nfs4frlock: frc_no_reclaim: clearing reclaim"));
14344 lock_args
->reclaim
= FALSE
;
14352 rfs4call(VTOMI4(vp
), argsp
, resp
, cred_otw
, &doqueue
, 0, ep
);
14354 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
14355 "nfs4frlock: error %d, status %d", ep
->error
, resp
->status
));
14357 needrecov
= nfs4_needs_recovery(ep
, TRUE
, vp
->v_vfsp
);
14358 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
14359 "nfs4frlock: needrecov %d", needrecov
));
14361 if (ep
->error
== 0 && nfs4_need_to_bump_seqid(resp
))
14362 nfs4frlock_bump_seqid(lock_args
, locku_args
, oop
, lop
,
14366 * Check if one of these mutually exclusive error cases has
14368 * need to swap credentials due to access error
14369 * recovery is needed
14370 * different error (only known case is missing Kerberos ticket)
14373 if ((ep
->error
== EACCES
||
14374 (ep
->error
== 0 && resp
->status
== NFS4ERR_ACCESS
)) &&
14376 nfs4frlock_check_access(vp
, op_hint
, &recov_state
, needrecov
,
14377 &did_start_fop
, &argsp
, &resp
, ep
->error
, &lop
, &oop
, &osp
,
14384 * LOCKT requests don't need to recover from lost
14385 * requests since they don't create/modify state.
14387 if ((ep
->error
== EINTR
||
14388 NFS4_FRC_UNMT_ERR(ep
->error
, vp
->v_vfsp
)) &&
14392 * Do not attempt recovery for requests initiated by
14393 * the recovery framework. Let the framework redrive them.
14395 if (ctype
!= NFS4_LCK_CTYPE_NORM
)
14398 ASSERT(resend_rqstp
== NULL
);
14401 nfs4frlock_save_lost_rqst(ctype
, ep
->error
,
14402 flk_to_locktype(cmd
, flk
->l_type
),
14403 oop
, osp
, lop
, flk
, &lost_rqst
, cred_otw
, vp
);
14405 retry
= nfs4frlock_recovery(needrecov
, ep
, &argsp
,
14406 &resp
, lock_args
, locku_args
, &oop
, &osp
, &lop
,
14407 rp
, vp
, &recov_state
, op_hint
, &did_start_fop
,
14408 cmd
!= F_GETLK
? &lost_rqst
: NULL
, flk
);
14411 ASSERT(oop
== NULL
);
14412 ASSERT(osp
== NULL
);
14413 ASSERT(lop
== NULL
);
14420 * Bail out if have reached this point with ep->error set. Can
14421 * happen if (ep->error == EACCES && !needrecov && cred_otw == cr).
14422 * This happens if Kerberos ticket has expired or has been
14425 if (ep
->error
!= 0)
14429 * Process the reply.
14431 switch (resp
->status
) {
14433 resop
= &resp
->array
[1];
14434 nfs4frlock_results_ok(ctype
, cmd
, flk
, vp
, flag
, offset
,
14437 * Have a successful lock operation, now update state.
14439 nfs4frlock_update_state(lock_args
, locku_args
, lockt_args
,
14440 resop
, lop
, vp
, flk
, cr
, resend_rqstp
);
14443 case NFS4ERR_DENIED
:
14444 resop
= &resp
->array
[1];
14445 retry
= nfs4frlock_results_denied(ctype
, lock_args
, lockt_args
,
14446 &oop
, &osp
, &lop
, cmd
, vp
, flk
, op_hint
,
14447 &recov_state
, needrecov
, &argsp
, &resp
,
14448 &tick_delay
, &whence
, &ep
->error
, resop
, cr
,
14449 &did_start_fop
, &skip_get_err
);
14452 ASSERT(oop
== NULL
);
14453 ASSERT(osp
== NULL
);
14454 ASSERT(lop
== NULL
);
14459 * If the server won't let us reclaim, fall-back to trying to lock
14460 * the file from scratch. Code elsewhere will check the changeinfo
14461 * to ensure the file hasn't been changed.
14463 case NFS4ERR_NO_GRACE
:
14464 if (lock_args
&& lock_args
->reclaim
== TRUE
) {
14465 ASSERT(ctype
== NFS4_LCK_CTYPE_RECLAIM
);
14466 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
14467 "nfs4frlock: reclaim: NFS4ERR_NO_GRACE"));
14468 frc_no_reclaim
= 1;
14469 /* clean up before retrying */
14471 (void) nfs4frlock_recovery(needrecov
, ep
, &argsp
, &resp
,
14472 lock_args
, locku_args
, &oop
, &osp
, &lop
, rp
, vp
,
14473 &recov_state
, op_hint
, &did_start_fop
, NULL
, flk
);
14479 nfs4frlock_results_default(resp
, &ep
->error
);
14484 * Process and cleanup from error. Make interrupted unlock
14485 * requests look successful, since they will be handled by the
14486 * client recovery code.
14488 nfs4frlock_final_cleanup(ctype
, argsp
, resp
, vp
, op_hint
, &recov_state
,
14489 needrecov
, oop
, osp
, lop
, flk
, whence
, offset
, ls
, &ep
->error
,
14490 lock_args
, locku_args
, did_start_fop
,
14491 skip_get_err
, cred_otw
, cr
);
14493 if (ep
->error
== EINTR
&& flk
->l_type
== F_UNLCK
&&
14494 (cmd
== F_SETLK
|| cmd
== F_SETLKW
))
14501 * Return non-zero if the given lock request can be handled without
14502 * violating the constraints on concurrent mapping and locking.
14506 nfs4_safelock(vnode_t
*vp
, const struct flock64
*bfp
, cred_t
*cr
)
14508 rnode4_t
*rp
= VTOR4(vp
);
14512 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
14513 ASSERT(rp
->r_mapcnt
>= 0);
14514 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
, "nfs4_safelock %s: "
14515 "(%"PRIx64
", %"PRIx64
"); mapcnt = %ld", bfp
->l_type
== F_WRLCK
?
14516 "write" : bfp
->l_type
== F_RDLCK
? "read" : "unlock",
14517 bfp
->l_start
, bfp
->l_len
, rp
->r_mapcnt
));
14519 if (rp
->r_mapcnt
== 0)
14520 return (1); /* always safe if not mapped */
14523 * If the file is already mapped and there are locks, then they
14524 * should be all safe locks. So adding or removing a lock is safe
14525 * as long as the new request is safe (i.e., whole-file, meaning
14526 * length and starting offset are both zero).
14529 if (bfp
->l_start
!= 0 || bfp
->l_len
!= 0) {
14530 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
, "nfs4_safelock: "
14531 "cannot lock a memory mapped file unless locking the "
14532 "entire file: start %"PRIx64
", len %"PRIx64
,
14533 bfp
->l_start
, bfp
->l_len
));
14537 /* mandatory locking and mapping don't mix */
14538 va
.va_mask
= AT_MODE
;
14539 error
= fop_getattr(vp
, &va
, 0, cr
, NULL
);
14541 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
, "nfs4_safelock: "
14542 "getattr error %d", error
));
14543 return (0); /* treat errors conservatively */
14545 if (MANDLOCK(vp
, va
.va_mode
)) {
14546 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
, "nfs4_safelock: "
14547 "cannot mandatory lock and mmap a file"));
14556 * Register the lock locally within Solaris.
14557 * As the client, we "or" the sysid with LM_SYSID_CLIENT when
14558 * recording locks locally.
14560 * This should handle conflicts/cooperation with NFS v2/v3 since all locks
14561 * are registered locally.
14564 nfs4_register_lock_locally(vnode_t
*vp
, struct flock64
*flk
, int flag
,
14573 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
14576 name
= fn_name(VTOSV(vp
)->sv_name
);
14577 NFS4_DEBUG(nfs4_client_lock_debug
,
14578 (CE_NOTE
, "nfs4_register_lock_locally: %s: type %d, "
14579 "start %"PRIx64
", length %"PRIx64
", pid %ld, sysid %d",
14580 name
, flk
->l_type
, flk
->l_start
, flk
->l_len
, (long)flk
->l_pid
,
14582 kmem_free(name
, MAXNAMELEN
);
14585 /* register the lock with local locking */
14586 oldsysid
= flk
->l_sysid
;
14587 flk
->l_sysid
|= LM_SYSID_CLIENT
;
14588 error
= reclock(vp
, flk
, SETFLCK
, flag
, offset
, NULL
);
14591 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
14592 "nfs4_register_lock_locally: could not register with"
14593 " local locking"));
14594 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_CONT
,
14595 "error %d, vp 0x%p, pid %d, sysid 0x%x",
14596 error
, (void *)vp
, flk
->l_pid
, flk
->l_sysid
));
14597 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_CONT
,
14598 "type %d off 0x%" PRIx64
" len 0x%" PRIx64
,
14599 flk
->l_type
, flk
->l_start
, flk
->l_len
));
14600 (void) reclock(vp
, flk
, 0, flag
, offset
, NULL
);
14601 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_CONT
,
14602 "blocked by pid %d sysid 0x%x type %d "
14603 "off 0x%" PRIx64
" len 0x%" PRIx64
,
14604 flk
->l_pid
, flk
->l_sysid
, flk
->l_type
, flk
->l_start
,
14608 flk
->l_sysid
= oldsysid
;
14612 * nfs4_lockrelease:
14614 * Release any locks on the given vnode that are held by the current
14615 * process. Also removes the lock owner (if one exists) from the rnode's
14619 nfs4_lockrelease(vnode_t
*vp
, int flag
, offset_t offset
, cred_t
*cr
)
14624 nfs4_lock_owner_t
*lop
;
14625 nfs4_recov_state_t recov_state
;
14627 bool_t possible_orphan
= FALSE
;
14630 ASSERT((uintptr_t)vp
> KERNELBASE
);
14631 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
14637 * If we have not locked anything then we can
14638 * just return since we have no work to do.
14640 if (rp
->r_lo_head
.lo_next_rnode
== &rp
->r_lo_head
) {
14645 * We need to comprehend that another thread may
14646 * kick off recovery and the lock_owner we have stashed
14647 * in lop might be invalid so we should NOT cache it
14650 recov_state
.rs_flags
= 0;
14651 recov_state
.rs_num_retry_despite_err
= 0;
14652 error
= nfs4_start_fop(mi
, vp
, NULL
, OH_LOCKU
, &recov_state
,
14655 mutex_enter(&rp
->r_statelock
);
14656 rp
->r_flags
|= R4LODANGLERS
;
14657 mutex_exit(&rp
->r_statelock
);
14661 lop
= find_lock_owner(rp
, curproc
->p_pid
, LOWN_ANY
);
14664 * Check if the lock owner might have a lock (request was sent but
14665 * no response was received). Also check if there are any remote
14666 * locks on the file. (In theory we shouldn't have to make this
14667 * second check if there's no lock owner, but for now we'll be
14668 * conservative and do it anyway.) If either condition is true,
14669 * send an unlock for the entire file to the server.
14671 * Note that no explicit synchronization is needed here. At worst,
14672 * flk_has_remote_locks() will return a false positive, in which case
14673 * the unlock call wastes time but doesn't harm correctness.
14677 mutex_enter(&lop
->lo_lock
);
14678 possible_orphan
= lop
->lo_pending_rqsts
;
14679 mutex_exit(&lop
->lo_lock
);
14680 lock_owner_rele(lop
);
14683 nfs4_end_fop(mi
, vp
, NULL
, OH_LOCKU
, &recov_state
, 0);
14685 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
14686 "nfs4_lockrelease: possible orphan %d, remote locks %d, for "
14687 "lop %p.", possible_orphan
, flk_has_remote_locks(vp
),
14690 if (possible_orphan
|| flk_has_remote_locks(vp
)) {
14691 ld
.l_type
= F_UNLCK
; /* set to unlock entire file */
14692 ld
.l_whence
= 0; /* unlock from start of file */
14694 ld
.l_len
= 0; /* do entire file */
14696 ret
= fop_frlock(vp
, F_SETLK
, &ld
, flag
, offset
, NULL
,
14701 * If fop_frlock fails, make sure we unregister
14702 * local locks before we continue.
14704 ld
.l_pid
= ttoproc(curthread
)->p_pid
;
14705 nfs4_register_lock_locally(vp
, &ld
, flag
, offset
);
14706 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
14707 "nfs4_lockrelease: lock release error on vp"
14708 " %p: error %d.\n", (void *)vp
, ret
));
14712 recov_state
.rs_flags
= 0;
14713 recov_state
.rs_num_retry_despite_err
= 0;
14714 error
= nfs4_start_fop(mi
, vp
, NULL
, OH_LOCKU
, &recov_state
,
14717 mutex_enter(&rp
->r_statelock
);
14718 rp
->r_flags
|= R4LODANGLERS
;
14719 mutex_exit(&rp
->r_statelock
);
14724 * So, here we're going to need to retrieve the lock-owner
14725 * again (in case recovery has done a switch-a-roo) and
14726 * remove it because we can.
14728 lop
= find_lock_owner(rp
, curproc
->p_pid
, LOWN_ANY
);
14731 nfs4_rnode_remove_lock_owner(rp
, lop
);
14732 lock_owner_rele(lop
);
14735 nfs4_end_fop(mi
, vp
, NULL
, OH_LOCKU
, &recov_state
, 0);
14740 * Wait for 'tick_delay' clock ticks.
14741 * Implement exponential backoff until hit the lease_time of this nfs4_server.
14742 * NOTE: lock_lease_time is in seconds.
14744 * XXX For future improvements, should implement a waiting queue scheme.
14747 nfs4_block_and_wait(clock_t *tick_delay
, rnode4_t
*rp
)
14749 long milliseconds_delay
;
14750 time_t lock_lease_time
;
14752 /* wait tick_delay clock ticks or siginteruptus */
14753 if (delay_sig(*tick_delay
)) {
14756 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
, "nfs4_block_and_wait: "
14757 "reissue the lock request: blocked for %ld clock ticks: %ld "
14758 "milliseconds", *tick_delay
, drv_hztousec(*tick_delay
) / 1000));
14760 /* get the lease time */
14761 lock_lease_time
= r2lease_time(rp
);
14763 /* drv_hztousec converts ticks to microseconds */
14764 milliseconds_delay
= drv_hztousec(*tick_delay
) / 1000;
14765 if (milliseconds_delay
< lock_lease_time
* 1000) {
14766 *tick_delay
= 2 * *tick_delay
;
14767 if (drv_hztousec(*tick_delay
) > lock_lease_time
* 1000 * 1000)
14768 *tick_delay
= drv_usectohz(lock_lease_time
*1000*1000);
14775 nfs4_vnops_init(void)
14780 nfs4_vnops_fini(void)
14785 * Return a reference to the directory (parent) vnode for a given vnode,
14786 * using the saved pathname information and the directory file handle. The
14787 * caller is responsible for disposing of the reference.
14788 * Returns zero or an errno value.
14790 * Caller should set need_start_op to FALSE if it is the recovery
14791 * thread, or if a start_fop has already been done. Otherwise, TRUE.
14794 vtodv(vnode_t
*vp
, vnode_t
**dvpp
, cred_t
*cr
, bool_t need_start_op
)
14797 vnode_t
*dvp
= NULL
;
14799 nfs4_fname_t
*mfname
;
14802 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
14804 if (vp
->v_flag
& VROOT
) {
14805 nfs4_sharedfh_t
*sfh
;
14809 ASSERT(vp
->v_type
== VREG
);
14812 svp
= mi
->mi_curr_serv
;
14813 (void) nfs_rw_enter_sig(&svp
->sv_lock
, RW_READER
, 0);
14814 fh
.nfs_fh4_len
= svp
->sv_pfhandle
.fh_len
;
14815 fh
.nfs_fh4_val
= svp
->sv_pfhandle
.fh_buf
;
14816 sfh
= sfh4_get(&fh
, VTOMI4(vp
));
14817 nfs_rw_exit(&svp
->sv_lock
);
14818 mfname
= mi
->mi_fname
;
14820 dvp
= makenfs4node_by_fh(sfh
, NULL
, &mfname
, NULL
, mi
, cr
, 0);
14823 if (dvp
->v_type
== VNON
)
14824 dvp
->v_type
= VDIR
;
14831 if (svnp
== NULL
) {
14832 NFS4_DEBUG(nfs4_client_shadow_debug
, (CE_NOTE
, "vtodv: "
14833 "shadow node is NULL"));
14837 if (svnp
->sv_name
== NULL
|| svnp
->sv_dfh
== NULL
) {
14838 NFS4_DEBUG(nfs4_client_shadow_debug
, (CE_NOTE
, "vtodv: "
14839 "shadow node name or dfh val == NULL"));
14843 error
= nfs4_make_dotdot(svnp
->sv_dfh
, 0, vp
, cr
, &dvp
,
14844 (int)need_start_op
);
14846 NFS4_DEBUG(nfs4_client_shadow_debug
, (CE_NOTE
, "vtodv: "
14847 "nfs4_make_dotdot returned %d", error
));
14851 NFS4_DEBUG(nfs4_client_shadow_debug
, (CE_NOTE
, "vtodv: "
14852 "nfs4_make_dotdot returned a NULL dvp"));
14855 if (dvp
->v_type
== VNON
)
14856 dvp
->v_type
= VDIR
;
14857 ASSERT(dvp
->v_type
== VDIR
);
14858 if (VTOR4(vp
)->r_flags
& R4ISXATTR
) {
14859 mutex_enter(&dvp
->v_lock
);
14860 dvp
->v_flag
|= V_XATTRDIR
;
14861 mutex_exit(&dvp
->v_lock
);
14868 * Copy the (final) component name of vp to fnamep. maxlen is the maximum
14869 * length that fnamep can accept, including the trailing null.
14870 * Returns 0 if okay, returns an errno value if there was a problem.
14874 vtoname(vnode_t
*vp
, char *fnamep
, ssize_t maxlen
)
14882 * If the file being opened has VROOT set, then this is
14883 * a "file" mount. sv_name will not be interesting, so
14884 * go back to the servinfo4 to get the original mount
14885 * path and strip off all but the final edge. Otherwise
14886 * just return the name from the shadow vnode.
14889 if (vp
->v_flag
& VROOT
) {
14891 svp
= VTOMI4(vp
)->mi_curr_serv
;
14892 (void) nfs_rw_enter_sig(&svp
->sv_lock
, RW_READER
, 0);
14894 fn
= strrchr(svp
->sv_path
, '/');
14901 fn
= fn_name(shvp
->sv_name
);
14905 if (strlen(fn
) < maxlen
)
14906 (void) strcpy(fnamep
, fn
);
14908 err
= ENAMETOOLONG
;
14910 if (vp
->v_flag
& VROOT
)
14911 nfs_rw_exit(&svp
->sv_lock
);
14913 kmem_free(fn
, MAXNAMELEN
);
14919 * Bookkeeping for a close that doesn't need to go over the wire.
14920 * *have_lockp is set to 0 if 'os_sync_lock' is released; otherwise
14924 nfs4close_notw(vnode_t
*vp
, nfs4_open_stream_t
*osp
, int *have_lockp
)
14932 NFS4_DEBUG(nfs4close_notw_debug
, (CE_NOTE
, "nfs4close_notw: "
14933 "rp=%p osp=%p", (void *)rp
, (void *)osp
));
14934 ASSERT(nfs_zone() == mi
->mi_zone
);
14935 ASSERT(mutex_owned(&osp
->os_sync_lock
));
14936 ASSERT(*have_lockp
);
14938 if (!osp
->os_valid
||
14939 osp
->os_open_ref_count
> 0 || osp
->os_mapcnt
> 0) {
14944 * This removes the reference obtained at OPEN; ie,
14945 * when the open stream structure was created.
14947 * We don't have to worry about calling 'open_stream_rele'
14948 * since we our currently holding a reference to this
14949 * open stream which means the count can not go to 0 with
14952 ASSERT(osp
->os_ref_count
>= 2);
14953 osp
->os_ref_count
--;
14955 mutex_exit(&osp
->os_sync_lock
);
14958 nfs4_dec_state_ref_count(mi
);
14962 * Close all remaining open streams on the rnode. These open streams
14963 * could be here because:
14964 * - The close attempted at either close or delmap failed
14965 * - Some kernel entity did fop_open but never did fop_close
14966 * - Someone did mknod on a regular file but never opened it
14969 nfs4close_all(vnode_t
*vp
, cred_t
*cr
)
14971 nfs4_open_stream_t
*osp
;
14973 nfs4_error_t e
= { 0, NFS4_OK
, RPC_SUCCESS
};
14976 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
14982 * At this point, all we know is that the last time
14983 * someone called vn_rele, the count was 1. Since then,
14984 * the vnode could have been re-activated. We want to
14985 * loop through the open streams and close each one, but
14986 * we have to be careful since once we release the rnode
14987 * hash bucket lock, someone else is free to come in and
14988 * re-activate the rnode and add new open streams. The
14989 * strategy is take the rnode hash bucket lock, verify that
14990 * the count is still 1, grab the open stream off the
14991 * head of the list and mark it invalid, then release the
14992 * rnode hash bucket lock and proceed with that open stream.
14993 * This is ok because nfs4close_one() will acquire the proper
14994 * open/create to close/destroy synchronization for open
14995 * streams, and will ensure that if someone has reopened
14996 * the open stream after we've dropped the hash bucket lock
14997 * then we'll just simply return without destroying the
14999 * Repeat until the list is empty.
15004 /* make sure vnode hasn't been reactivated */
15005 rw_enter(&rp
->r_hashq
->r_lock
, RW_READER
);
15006 mutex_enter(&vp
->v_lock
);
15007 if (vp
->v_count
> 1) {
15008 mutex_exit(&vp
->v_lock
);
15009 rw_exit(&rp
->r_hashq
->r_lock
);
15013 * Grabbing r_os_lock before releasing v_lock prevents
15014 * a window where the rnode/open stream could get
15015 * reactivated (and os_force_close set to 0) before we
15016 * had a chance to set os_force_close to 1.
15018 mutex_enter(&rp
->r_os_lock
);
15019 mutex_exit(&vp
->v_lock
);
15021 osp
= list_head(&rp
->r_open_streams
);
15023 /* nothing left to CLOSE OTW, so return */
15024 mutex_exit(&rp
->r_os_lock
);
15025 rw_exit(&rp
->r_hashq
->r_lock
);
15029 mutex_enter(&rp
->r_statev4_lock
);
15030 /* the file can't still be mem mapped */
15031 ASSERT(rp
->r_mapcnt
== 0);
15032 if (rp
->created_v4
)
15033 rp
->created_v4
= 0;
15034 mutex_exit(&rp
->r_statev4_lock
);
15037 * Grab a ref on this open stream; nfs4close_one
15038 * will mark it as invalid
15040 mutex_enter(&osp
->os_sync_lock
);
15041 osp
->os_ref_count
++;
15042 osp
->os_force_close
= 1;
15043 mutex_exit(&osp
->os_sync_lock
);
15044 mutex_exit(&rp
->r_os_lock
);
15045 rw_exit(&rp
->r_hashq
->r_lock
);
15047 nfs4close_one(vp
, osp
, cr
, 0, NULL
, &e
, CLOSE_FORCE
, 0, 0, 0);
15049 /* Update error if it isn't already non-zero */
15054 error
= geterrno4(e
.stat
);
15058 nfs4close_all_cnt
++;
15060 /* Release the ref on osp acquired above. */
15061 open_stream_rele(osp
, rp
);
15063 /* Proceed to the next open stream, if any */
15069 * nfs4close_one - close one open stream for a file if needed.
15071 * "close_type" indicates which close path this is:
15072 * CLOSE_NORM: close initiated via fop_close.
15073 * CLOSE_DELMAP: close initiated via fop_delmap.
15074 * CLOSE_FORCE: close initiated via fop_inactive. This path forces
15075 * the close and release of client state for this open stream
15076 * (unless someone else has the open stream open).
15077 * CLOSE_RESEND: indicates the request is a replay of an earlier request
15078 * (e.g., due to abort because of a signal).
15079 * CLOSE_AFTER_RESEND: close initiated to "undo" a successful resent OPEN.
15081 * CLOSE_RESEND and CLOSE_AFTER_RESEND will not attempt to retry after client
15082 * recovery. Instead, the caller is expected to deal with retries.
15084 * The caller can either pass in the osp ('provided_osp') or not.
15086 * 'access_bits' represents the access we are closing/downgrading.
15088 * 'len', 'prot', and 'mmap_flags' are used for CLOSE_DELMAP. 'len' is the
15089 * number of bytes we are unmapping, 'maxprot' is the mmap protection, and
15090 * 'mmap_flags' tells us the type of sharing (MAP_PRIVATE or MAP_SHARED).
15092 * Errors are returned via the nfs4_error_t.
15095 nfs4close_one(vnode_t
*vp
, nfs4_open_stream_t
*provided_osp
, cred_t
*cr
,
15096 int access_bits
, nfs4_lost_rqst_t
*lrp
, nfs4_error_t
*ep
,
15097 nfs4_close_type_t close_type
, size_t len
, uint_t maxprot
,
15100 nfs4_open_owner_t
*oop
;
15101 nfs4_open_stream_t
*osp
= NULL
;
15103 int num_retries
= NFS4_NUM_RECOV_RETRIES
;
15106 nfs4_recov_state_t recov_state
;
15107 cred_t
*cred_otw
= NULL
;
15108 bool_t recovonly
= FALSE
;
15111 int close_failed
= 0;
15112 int did_dec_count
= 0;
15113 int did_start_op
= 0;
15114 int did_force_recovlock
= 0;
15115 int did_start_seqid_sync
= 0;
15116 int have_sync_lock
= 0;
15118 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
15120 NFS4_DEBUG(nfs4close_one_debug
, (CE_NOTE
, "closing vp %p osp %p, "
15121 "lrp %p, close type %d len %ld prot %x mmap flags %x bits %x",
15122 (void *)vp
, (void *)provided_osp
, (void *)lrp
, close_type
,
15123 len
, maxprot
, mmap_flags
, access_bits
));
15125 nfs4_error_zinit(ep
);
15128 isrecov
= (close_type
== CLOSE_RESEND
||
15129 close_type
== CLOSE_AFTER_RESEND
);
15132 * First get the open owner.
15134 if (!provided_osp
) {
15135 oop
= find_open_owner(cr
, NFS4_PERM_CREATED
, mi
);
15137 oop
= provided_osp
->os_open_owner
;
15138 ASSERT(oop
!= NULL
);
15139 open_owner_hold(oop
);
15143 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
,
15144 "nfs4close_one: no oop, rp %p, mi %p, cr %p, osp %p, "
15145 "close type %d", (void *)rp
, (void *)mi
, (void *)cr
,
15146 (void *)provided_osp
, close_type
));
15151 cred_otw
= nfs4_get_otw_cred(cr
, mi
, oop
);
15155 force_close
= (close_type
== CLOSE_FORCE
);
15158 did_force_recovlock
= 0;
15159 did_start_seqid_sync
= 0;
15160 have_sync_lock
= 0;
15162 recov_state
.rs_flags
= 0;
15163 recov_state
.rs_num_retry_despite_err
= 0;
15166 * Second synchronize with recovery.
15169 ep
->error
= nfs4_start_fop(mi
, vp
, NULL
, OH_CLOSE
,
15170 &recov_state
, &recovonly
);
15176 * If we couldn't get start_fop, but have to
15177 * cleanup state, then at least acquire the
15178 * mi_recovlock so we can synchronize with
15181 if (close_type
== CLOSE_FORCE
) {
15182 (void) nfs_rw_enter_sig(&mi
->mi_recovlock
,
15184 did_force_recovlock
= 1;
15191 * We cannot attempt to get the open seqid sync if nfs4_start_fop
15192 * set 'recovonly' to TRUE since most likely this is due to
15193 * reovery being active (MI4_RECOV_ACTIV). If recovery is active,
15194 * nfs4_start_open_seqid_sync() will fail with EAGAIN asking us
15195 * to retry, causing us to loop until recovery finishes. Plus we
15196 * don't need protection over the open seqid since we're not going
15197 * OTW, hence don't need to use the seqid.
15199 if (recovonly
== FALSE
) {
15200 /* need to grab the open owner sync before 'os_sync_lock' */
15201 ep
->error
= nfs4_start_open_seqid_sync(oop
, mi
);
15202 if (ep
->error
== EAGAIN
) {
15205 nfs4_end_fop(mi
, vp
, NULL
, OH_CLOSE
,
15206 &recov_state
, TRUE
);
15207 if (did_force_recovlock
)
15208 nfs_rw_exit(&mi
->mi_recovlock
);
15211 did_start_seqid_sync
= 1;
15215 * Third get an open stream and acquire 'os_sync_lock' to
15216 * sychronize the opening/creating of an open stream with the
15217 * closing/destroying of an open stream.
15219 if (!provided_osp
) {
15220 /* returns with 'os_sync_lock' held */
15221 osp
= find_open_stream(oop
, rp
);
15227 osp
= provided_osp
;
15228 open_stream_hold(osp
);
15229 mutex_enter(&osp
->os_sync_lock
);
15231 have_sync_lock
= 1;
15233 ASSERT(oop
== osp
->os_open_owner
);
15236 * Fourth, do any special pre-OTW CLOSE processing
15237 * based on the specific close type.
15239 if ((close_type
== CLOSE_NORM
|| close_type
== CLOSE_AFTER_RESEND
) &&
15241 ASSERT(osp
->os_open_ref_count
> 0);
15242 osp
->os_open_ref_count
--;
15244 if (osp
->os_open_ref_count
== 0)
15245 osp
->os_final_close
= 1;
15248 if (close_type
== CLOSE_FORCE
) {
15249 /* see if somebody reopened the open stream. */
15250 if (!osp
->os_force_close
) {
15251 NFS4_DEBUG(nfs4close_one_debug
, (CE_NOTE
,
15252 "nfs4close_one: skip CLOSE_FORCE as osp %p "
15253 "was reopened, vp %p", (void *)osp
, (void *)vp
));
15255 ep
->stat
= NFS4_OK
;
15259 if (!osp
->os_final_close
&& !did_dec_count
) {
15260 osp
->os_open_ref_count
--;
15265 * We can't depend on os_open_ref_count being 0 due to the
15266 * way executables are opened (VN_RELE to match a fop_open).
15269 ASSERT(osp
->os_open_ref_count
== 0);
15271 if (osp
->os_open_ref_count
!= 0) {
15272 NFS4_DEBUG(nfs4close_one_debug
, (CE_NOTE
,
15273 "nfs4close_one: should panic here on an "
15274 "ASSERT(osp->os_open_ref_count == 0). Ignoring "
15275 "since this is probably the exec problem."));
15277 osp
->os_open_ref_count
= 0;
15281 * There is the possibility that nfs4close_one()
15282 * for close_type == CLOSE_DELMAP couldn't find the
15283 * open stream, thus couldn't decrement its os_mapcnt;
15284 * therefore we can't use this ASSERT yet.
15287 ASSERT(osp
->os_mapcnt
== 0);
15289 osp
->os_mapcnt
= 0;
15292 if (close_type
== CLOSE_DELMAP
&& !did_dec_count
) {
15293 ASSERT(osp
->os_mapcnt
>= btopr(len
));
15295 if ((mmap_flags
& MAP_SHARED
) && (maxprot
& PROT_WRITE
))
15296 osp
->os_mmap_write
-= btopr(len
);
15297 if (maxprot
& PROT_READ
)
15298 osp
->os_mmap_read
-= btopr(len
);
15299 if (maxprot
& PROT_EXEC
)
15300 osp
->os_mmap_read
-= btopr(len
);
15301 /* mirror the PROT_NONE check in nfs4_addmap() */
15302 if (!(maxprot
& PROT_READ
) && !(maxprot
& PROT_WRITE
) &&
15303 !(maxprot
& PROT_EXEC
))
15304 osp
->os_mmap_read
-= btopr(len
);
15305 osp
->os_mapcnt
-= btopr(len
);
15310 nfs4_lost_rqst_t lost_rqst
;
15312 /* request should not already be in recovery queue */
15313 ASSERT(lrp
== NULL
);
15314 nfs4_error_init(ep
, EINTR
);
15315 nfs4close_save_lost_rqst(ep
->error
, &lost_rqst
, oop
,
15316 osp
, cred_otw
, vp
);
15317 mutex_exit(&osp
->os_sync_lock
);
15318 have_sync_lock
= 0;
15319 (void) nfs4_start_recovery(ep
, mi
, vp
, NULL
, NULL
,
15320 lost_rqst
.lr_op
== OP_CLOSE
?
15321 &lost_rqst
: NULL
, OP_CLOSE
, NULL
, NULL
, NULL
);
15324 goto close_cleanup
;
15328 * If a previous OTW call got NFS4ERR_BAD_SEQID, then
15329 * we stopped operating on the open owner's <old oo_name, old seqid>
15330 * space, which means we stopped operating on the open stream
15331 * too. So don't go OTW (as the seqid is likely bad, and the
15332 * stateid could be stale, potentially triggering a false
15333 * setclientid), and just clean up the client's internal state.
15335 if (osp
->os_orig_oo_name
!= oop
->oo_name
) {
15336 NFS4_DEBUG(nfs4close_one_debug
|| nfs4_client_recov_debug
,
15337 (CE_NOTE
, "nfs4close_one: skip OTW close for osp %p "
15338 "oop %p due to bad seqid (orig oo_name %" PRIx64
" current "
15339 "oo_name %" PRIx64
")",
15340 (void *)osp
, (void *)oop
, osp
->os_orig_oo_name
,
15345 /* If the file failed recovery, just quit. */
15346 mutex_enter(&rp
->r_statelock
);
15347 if (rp
->r_flags
& R4RECOVERR
) {
15350 mutex_exit(&rp
->r_statelock
);
15353 * If the force close path failed to obtain start_fop
15354 * then skip the OTW close and just remove the state.
15357 goto close_cleanup
;
15360 * Fifth, check to see if there are still mapped pages or other
15361 * opens using this open stream. If there are then we can't
15362 * close yet but we can see if an OPEN_DOWNGRADE is necessary.
15364 if (osp
->os_open_ref_count
> 0 || osp
->os_mapcnt
> 0) {
15365 nfs4_lost_rqst_t new_lost_rqst
;
15366 bool_t needrecov
= FALSE
;
15367 cred_t
*odg_cred_otw
= NULL
;
15368 seqid4 open_dg_seqid
= 0;
15370 if (osp
->os_delegation
) {
15372 * If this open stream was never OPENed OTW then we
15373 * surely can't DOWNGRADE it (especially since the
15374 * osp->open_stateid is really a delegation stateid
15375 * when os_delegation is 1).
15377 if (access_bits
& FREAD
)
15378 osp
->os_share_acc_read
--;
15379 if (access_bits
& FWRITE
)
15380 osp
->os_share_acc_write
--;
15381 osp
->os_share_deny_none
--;
15382 nfs4_error_zinit(ep
);
15385 nfs4_open_downgrade(access_bits
, 0, oop
, osp
, vp
, cr
,
15386 lrp
, ep
, &odg_cred_otw
, &open_dg_seqid
);
15387 needrecov
= nfs4_needs_recovery(ep
, TRUE
, mi
->mi_vfsp
);
15388 if (needrecov
&& !isrecov
) {
15390 nfs4_bseqid_entry_t
*bsep
= NULL
;
15392 if (!ep
->error
&& ep
->stat
== NFS4ERR_BAD_SEQID
)
15393 bsep
= nfs4_create_bseqid_entry(oop
, NULL
,
15395 lrp
? TAG_OPEN_DG_LOST
: TAG_OPEN_DG
,
15398 nfs4open_dg_save_lost_rqst(ep
->error
, &new_lost_rqst
,
15399 oop
, osp
, odg_cred_otw
, vp
, access_bits
, 0);
15400 mutex_exit(&osp
->os_sync_lock
);
15401 have_sync_lock
= 0;
15402 abort
= nfs4_start_recovery(ep
, mi
, vp
, NULL
, NULL
,
15403 new_lost_rqst
.lr_op
== OP_OPEN_DOWNGRADE
?
15404 &new_lost_rqst
: NULL
, OP_OPEN_DOWNGRADE
,
15407 crfree(odg_cred_otw
);
15409 kmem_free(bsep
, sizeof (*bsep
));
15414 if (did_start_seqid_sync
) {
15415 nfs4_end_open_seqid_sync(oop
);
15416 did_start_seqid_sync
= 0;
15418 open_stream_rele(osp
, rp
);
15421 nfs4_end_fop(mi
, vp
, NULL
, OH_CLOSE
,
15422 &recov_state
, FALSE
);
15423 if (did_force_recovlock
)
15424 nfs_rw_exit(&mi
->mi_recovlock
);
15429 crfree(odg_cred_otw
);
15435 * If this open stream was created as the results of an open
15436 * while holding a delegation, then just release it; no need
15437 * to do an OTW close. Otherwise do a "normal" OTW close.
15439 if (osp
->os_delegation
) {
15440 nfs4close_notw(vp
, osp
, &have_sync_lock
);
15441 nfs4_error_zinit(ep
);
15446 * If this stream is not valid, we're done.
15448 if (!osp
->os_valid
) {
15449 nfs4_error_zinit(ep
);
15454 * Last open or mmap ref has vanished, need to do an OTW close.
15455 * First check to see if a close is still necessary.
15457 if (osp
->os_failed_reopen
) {
15458 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
,
15459 "don't close OTW osp %p since reopen failed.",
15462 * Reopen of the open stream failed, hence the
15463 * stateid of the open stream is invalid/stale, and
15464 * sending this OTW would incorrectly cause another
15465 * round of recovery. In this case, we need to set
15466 * the 'os_valid' bit to 0 so another thread doesn't
15467 * come in and re-open this open stream before
15468 * this "closing" thread cleans up state (decrementing
15469 * the nfs4_server_t's state_ref_count and decrementing
15470 * the os_ref_count).
15474 * This removes the reference obtained at OPEN; ie,
15475 * when the open stream structure was created.
15477 * We don't have to worry about calling 'open_stream_rele'
15478 * since we our currently holding a reference to this
15479 * open stream which means the count can not go to 0 with
15482 ASSERT(osp
->os_ref_count
>= 2);
15483 osp
->os_ref_count
--;
15484 nfs4_error_zinit(ep
);
15486 goto close_cleanup
;
15489 ASSERT(osp
->os_ref_count
> 1);
15492 * Sixth, try the CLOSE OTW.
15494 nfs4close_otw(rp
, cred_otw
, oop
, osp
, &retry
, &did_start_seqid_sync
,
15495 close_type
, ep
, &have_sync_lock
);
15497 if (ep
->error
== EINTR
|| NFS4_FRC_UNMT_ERR(ep
->error
, vp
->v_vfsp
)) {
15499 * Let the recovery thread be responsible for
15500 * removing the state for CLOSE.
15507 /* See if we need to retry with a different cred */
15508 if ((ep
->error
== EACCES
||
15509 (ep
->error
== 0 && ep
->stat
== NFS4ERR_ACCESS
)) &&
15517 if (ep
->error
|| ep
->stat
)
15520 if (retry
&& !isrecov
&& num_retries
-- > 0) {
15521 if (have_sync_lock
) {
15522 mutex_exit(&osp
->os_sync_lock
);
15523 have_sync_lock
= 0;
15525 if (did_start_seqid_sync
) {
15526 nfs4_end_open_seqid_sync(oop
);
15527 did_start_seqid_sync
= 0;
15529 open_stream_rele(osp
, rp
);
15532 nfs4_end_fop(mi
, vp
, NULL
, OH_CLOSE
,
15533 &recov_state
, FALSE
);
15534 if (did_force_recovlock
)
15535 nfs_rw_exit(&mi
->mi_recovlock
);
15536 NFS4_DEBUG(nfs4_client_recov_debug
, (CE_NOTE
,
15537 "nfs4close_one: need to retry the close "
15543 * Seventh and lastly, process our results.
15545 if (close_failed
&& force_close
) {
15547 * It's ok to drop and regrab the 'os_sync_lock' since
15548 * nfs4close_notw() will recheck to make sure the
15549 * "close"/removal of state should happen.
15551 if (!have_sync_lock
) {
15552 mutex_enter(&osp
->os_sync_lock
);
15553 have_sync_lock
= 1;
15556 * This is last call, remove the ref on the open
15557 * stream created by open and clean everything up.
15559 osp
->os_pending_close
= 0;
15560 nfs4close_notw(vp
, osp
, &have_sync_lock
);
15561 nfs4_error_zinit(ep
);
15564 if (!close_failed
) {
15565 if (have_sync_lock
) {
15566 osp
->os_pending_close
= 0;
15567 mutex_exit(&osp
->os_sync_lock
);
15568 have_sync_lock
= 0;
15570 mutex_enter(&osp
->os_sync_lock
);
15571 osp
->os_pending_close
= 0;
15572 mutex_exit(&osp
->os_sync_lock
);
15574 if (did_start_op
&& recov_state
.rs_sp
!= NULL
) {
15575 mutex_enter(&recov_state
.rs_sp
->s_lock
);
15576 nfs4_dec_state_ref_count_nolock(recov_state
.rs_sp
, mi
);
15577 mutex_exit(&recov_state
.rs_sp
->s_lock
);
15579 nfs4_dec_state_ref_count(mi
);
15581 nfs4_error_zinit(ep
);
15585 if (have_sync_lock
)
15586 mutex_exit(&osp
->os_sync_lock
);
15588 nfs4_end_fop(mi
, vp
, NULL
, OH_CLOSE
, &recov_state
,
15589 recovonly
? TRUE
: FALSE
);
15590 if (did_force_recovlock
)
15591 nfs_rw_exit(&mi
->mi_recovlock
);
15595 open_stream_rele(osp
, rp
);
15597 if (did_start_seqid_sync
)
15598 nfs4_end_open_seqid_sync(oop
);
15599 open_owner_rele(oop
);
15604 * Convert information returned by the server in the LOCK4denied
15605 * structure to the form required by fcntl.
15608 denied_to_flk(LOCK4denied
*lockt_denied
, flock64_t
*flk
, LOCKT4args
*lockt_args
)
15610 nfs4_lo_name_t
*lo
;
15613 if (denied_to_flk_debug
) {
15614 lockt_denied_debug
= lockt_denied
;
15615 debug_enter("lockt_denied");
15619 flk
->l_type
= lockt_denied
->locktype
== READ_LT
? F_RDLCK
: F_WRLCK
;
15620 flk
->l_whence
= 0; /* aka SEEK_SET */
15621 flk
->l_start
= lockt_denied
->offset
;
15622 flk
->l_len
= lockt_denied
->length
;
15625 * If the blocking clientid matches our client id, then we can
15626 * interpret the lockowner (since we built it). If not, then
15627 * fabricate a sysid and pid. Note that the l_sysid field
15628 * in *flk already has the local sysid.
15631 if (lockt_denied
->owner
.clientid
== lockt_args
->owner
.clientid
) {
15633 if (lockt_denied
->owner
.owner_len
== sizeof (*lo
)) {
15634 lo
= (nfs4_lo_name_t
*)
15635 lockt_denied
->owner
.owner_val
;
15637 flk
->l_pid
= lo
->ln_pid
;
15639 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
15640 "denied_to_flk: bad lock owner length\n"));
15642 flk
->l_pid
= lo_to_pid(&lockt_denied
->owner
);
15645 NFS4_DEBUG(nfs4_client_lock_debug
, (CE_NOTE
,
15646 "denied_to_flk: foreign clientid\n"));
15649 * Construct a new sysid which should be different from
15650 * sysids of other systems.
15654 flk
->l_pid
= lo_to_pid(&lockt_denied
->owner
);
15659 lo_to_pid(lock_owner4
*lop
)
15665 cp
= (uchar_t
*)&lop
->clientid
;
15667 for (i
= 0; i
< sizeof (lop
->clientid
); i
++)
15668 pid
+= (pid_t
)*cp
++;
15670 cp
= (uchar_t
*)lop
->owner_val
;
15672 for (i
= 0; i
< lop
->owner_len
; i
++)
15673 pid
+= (pid_t
)*cp
++;
15679 * Given a lock pointer, returns the length of that lock.
15680 * "end" is the last locked offset the "l_len" covers from
15681 * the start of the lock.
15684 lock_to_end(flock64_t
*lock
)
15688 if (lock
->l_len
== 0)
15689 lock_end
= (off64_t
)MAXEND
;
15691 lock_end
= lock
->l_start
+ lock
->l_len
- 1;
15697 * Given the end of a lock, it will return you the length "l_len" for that lock.
15700 end_to_len(off64_t start
, off64_t end
)
15704 ASSERT(end
>= start
);
15708 lock_len
= end
- start
+ 1;
15714 * On given end for a lock it determines if it is the last locked offset
15715 * or not, if so keeps it as is, else adds one to return the length for
15719 start_check(off64_t x
)
15728 * See if these two locks overlap, and if so return 1;
15729 * otherwise, return 0.
15732 locks_intersect(flock64_t
*llfp
, flock64_t
*curfp
)
15734 off64_t llfp_end
, curfp_end
;
15736 llfp_end
= lock_to_end(llfp
);
15737 curfp_end
= lock_to_end(curfp
);
15739 if (((llfp_end
>= curfp
->l_start
) &&
15740 (llfp
->l_start
<= curfp
->l_start
)) ||
15741 ((curfp
->l_start
<= llfp
->l_start
) && (curfp_end
>= llfp
->l_start
)))
15747 * Determine what the intersecting lock region is, and add that to the
15748 * 'nl_llpp' locklist in increasing order (by l_start).
15751 nfs4_add_lock_range(flock64_t
*lost_flp
, flock64_t
*local_flp
,
15752 locklist_t
**nl_llpp
, vnode_t
*vp
)
15754 locklist_t
*intersect_llp
, *tmp_fllp
, *cur_fllp
;
15755 off64_t lost_flp_end
, local_flp_end
, len
, start
;
15757 NFS4_DEBUG(nfs4_lost_rqst_debug
, (CE_NOTE
, "nfs4_add_lock_range:"));
15759 if (!locks_intersect(lost_flp
, local_flp
))
15762 NFS4_DEBUG(nfs4_lost_rqst_debug
, (CE_NOTE
, "nfs4_add_lock_range: "
15763 "locks intersect"));
15765 lost_flp_end
= lock_to_end(lost_flp
);
15766 local_flp_end
= lock_to_end(local_flp
);
15768 /* Find the starting point of the intersecting region */
15769 if (local_flp
->l_start
> lost_flp
->l_start
)
15770 start
= local_flp
->l_start
;
15772 start
= lost_flp
->l_start
;
15774 /* Find the lenght of the intersecting region */
15775 if (lost_flp_end
< local_flp_end
)
15776 len
= end_to_len(start
, lost_flp_end
);
15778 len
= end_to_len(start
, local_flp_end
);
15781 * Prepare the flock structure for the intersection found and insert
15782 * it into the new list in increasing l_start order. This list contains
15783 * intersections of locks registered by the client with the local host
15784 * and the lost lock.
15785 * The lock type of this lock is the same as that of the local_flp.
15787 intersect_llp
= (locklist_t
*)kmem_alloc(sizeof (locklist_t
), KM_SLEEP
);
15788 intersect_llp
->ll_flock
.l_start
= start
;
15789 intersect_llp
->ll_flock
.l_len
= len
;
15790 intersect_llp
->ll_flock
.l_type
= local_flp
->l_type
;
15791 intersect_llp
->ll_flock
.l_pid
= local_flp
->l_pid
;
15792 intersect_llp
->ll_flock
.l_sysid
= local_flp
->l_sysid
;
15793 intersect_llp
->ll_flock
.l_whence
= 0; /* aka SEEK_SET */
15794 intersect_llp
->ll_vp
= vp
;
15796 tmp_fllp
= *nl_llpp
;
15798 while (tmp_fllp
!= NULL
&& tmp_fllp
->ll_flock
.l_start
<
15799 intersect_llp
->ll_flock
.l_start
) {
15800 cur_fllp
= tmp_fllp
;
15801 tmp_fllp
= tmp_fllp
->ll_next
;
15803 if (cur_fllp
== NULL
) {
15804 /* first on the list */
15805 intersect_llp
->ll_next
= *nl_llpp
;
15806 *nl_llpp
= intersect_llp
;
15808 intersect_llp
->ll_next
= cur_fllp
->ll_next
;
15809 cur_fllp
->ll_next
= intersect_llp
;
15812 NFS4_DEBUG(nfs4_lost_rqst_debug
, (CE_NOTE
, "nfs4_add_lock_range: "
15813 "created lock region: start %"PRIx64
" end %"PRIx64
" : %s\n",
15814 intersect_llp
->ll_flock
.l_start
,
15815 intersect_llp
->ll_flock
.l_start
+ intersect_llp
->ll_flock
.l_len
,
15816 intersect_llp
->ll_flock
.l_type
== F_RDLCK
? "READ" : "WRITE"));
15820 * Our local locking current state is potentially different than
15821 * what the NFSv4 server thinks we have due to a lost lock that was
15822 * resent and then received. We need to reset our "NFSv4" locking
15823 * state to match the current local locking state for this pid since
15824 * that is what the user/application sees as what the world is.
15826 * We cannot afford to drop the open/lock seqid sync since then we can
15827 * get confused about what the current local locking state "is" versus
15830 * If we are unable to fix up the locks, we send SIGLOST to the affected
15831 * process. This is not done if the filesystem has been forcibly
15832 * unmounted, in case the process has already exited and a new process
15833 * exists with the same pid.
15836 nfs4_reinstitute_local_lock_state(vnode_t
*vp
, flock64_t
*lost_flp
, cred_t
*cr
,
15837 nfs4_lock_owner_t
*lop
)
15839 locklist_t
*locks
, *llp
, *ri_llp
, *tmp_llp
;
15840 mntinfo4_t
*mi
= VTOMI4(vp
);
15841 const int cmd
= F_SETLK
;
15842 off64_t cur_start
, llp_ll_flock_end
, lost_flp_end
;
15845 NFS4_DEBUG(nfs4_lost_rqst_debug
, (CE_NOTE
,
15846 "nfs4_reinstitute_local_lock_state"));
15849 * Find active locks for this vp from the local locking code.
15850 * Scan through this list and find out the locks that intersect with
15851 * the lost lock. Once we find the lock that intersects, add the
15852 * intersection area as a new lock to a new list "ri_llp". The lock
15853 * type of the intersection region lock added to ri_llp is the same
15854 * as that found in the active lock list, "list". The intersecting
15855 * region locks are added to ri_llp in increasing l_start order.
15857 ASSERT(nfs_zone() == mi
->mi_zone
);
15859 locks
= flk_active_locks_for_vp(vp
);
15862 for (llp
= locks
; llp
!= NULL
; llp
= llp
->ll_next
) {
15863 ASSERT(llp
->ll_vp
== vp
);
15865 * Pick locks that belong to this pid/lockowner
15867 if (llp
->ll_flock
.l_pid
!= lost_flp
->l_pid
)
15870 nfs4_add_lock_range(lost_flp
, &llp
->ll_flock
, &ri_llp
, vp
);
15874 * Now we have the list of intersections with the lost lock. These are
15875 * the locks that were/are active before the server replied to the
15876 * last/lost lock. Issue these locks to the server here. Playing these
15877 * locks to the server will re-establish our current local locking state
15878 * with the v4 server.
15879 * If we get an error, send SIGLOST to the application for that lock.
15882 for (llp
= ri_llp
; llp
!= NULL
; llp
= llp
->ll_next
) {
15883 NFS4_DEBUG(nfs4_lost_rqst_debug
, (CE_NOTE
,
15884 "nfs4_reinstitute_local_lock_state: need to issue "
15885 "flock: [%"PRIx64
" - %"PRIx64
"] : %s",
15886 llp
->ll_flock
.l_start
,
15887 llp
->ll_flock
.l_start
+ llp
->ll_flock
.l_len
,
15888 llp
->ll_flock
.l_type
== F_RDLCK
? "READ" :
15889 llp
->ll_flock
.l_type
== F_WRLCK
? "WRITE" : "INVALID"));
15891 * No need to relock what we already have
15893 if (llp
->ll_flock
.l_type
== lost_flp
->l_type
)
15896 push_reinstate(vp
, cmd
, &llp
->ll_flock
, cr
, lop
);
15900 * Now keeping the start of the lost lock as our reference parse the
15901 * newly created ri_llp locklist to find the ranges that we have locked
15902 * with the v4 server but not in the current local locking. We need
15903 * to unlock these ranges.
15904 * These ranges can also be reffered to as those ranges, where the lost
15905 * lock does not overlap with the locks in the ri_llp but are locked
15906 * since the server replied to the lost lock.
15908 cur_start
= lost_flp
->l_start
;
15909 lost_flp_end
= lock_to_end(lost_flp
);
15911 ul_fl
.l_type
= F_UNLCK
;
15912 ul_fl
.l_whence
= 0; /* aka SEEK_SET */
15913 ul_fl
.l_sysid
= lost_flp
->l_sysid
;
15914 ul_fl
.l_pid
= lost_flp
->l_pid
;
15916 for (llp
= ri_llp
; llp
!= NULL
; llp
= llp
->ll_next
) {
15917 llp_ll_flock_end
= lock_to_end(&llp
->ll_flock
);
15919 if (llp
->ll_flock
.l_start
<= cur_start
) {
15920 cur_start
= start_check(llp_ll_flock_end
);
15923 NFS4_DEBUG(nfs4_lost_rqst_debug
, (CE_NOTE
,
15924 "nfs4_reinstitute_local_lock_state: "
15925 "UNLOCK [%"PRIx64
" - %"PRIx64
"]",
15926 cur_start
, llp
->ll_flock
.l_start
));
15928 ul_fl
.l_start
= cur_start
;
15929 ul_fl
.l_len
= end_to_len(cur_start
,
15930 (llp
->ll_flock
.l_start
- 1));
15932 push_reinstate(vp
, cmd
, &ul_fl
, cr
, lop
);
15933 cur_start
= start_check(llp_ll_flock_end
);
15937 * In the case where the lost lock ends after all intersecting locks,
15938 * unlock the last part of the lost lock range.
15940 if (cur_start
!= start_check(lost_flp_end
)) {
15941 NFS4_DEBUG(nfs4_lost_rqst_debug
, (CE_NOTE
,
15942 "nfs4_reinstitute_local_lock_state: UNLOCK end of the "
15943 "lost lock region [%"PRIx64
" - %"PRIx64
"]",
15944 cur_start
, lost_flp
->l_start
+ lost_flp
->l_len
));
15946 ul_fl
.l_start
= cur_start
;
15948 * Is it an to-EOF lock? if so unlock till the end
15950 if (lost_flp
->l_len
== 0)
15953 ul_fl
.l_len
= start_check(lost_flp_end
) - cur_start
;
15955 push_reinstate(vp
, cmd
, &ul_fl
, cr
, lop
);
15959 flk_free_locklist(locks
);
15961 /* Free up our newly created locklist */
15962 for (llp
= ri_llp
; llp
!= NULL
; ) {
15963 tmp_llp
= llp
->ll_next
;
15964 kmem_free(llp
, sizeof (locklist_t
));
15969 * Now return back to the original calling nfs4frlock()
15970 * and let us naturally drop our seqid syncs.
15975 * Create a lost state record for the given lock reinstantiation request
15976 * and push it onto the lost state queue.
15979 push_reinstate(vnode_t
*vp
, int cmd
, flock64_t
*flk
, cred_t
*cr
,
15980 nfs4_lock_owner_t
*lop
)
15982 nfs4_lost_rqst_t req
;
15983 nfs_lock_type4 locktype
;
15984 nfs4_error_t e
= { EINTR
, NFS4_OK
, RPC_SUCCESS
};
15986 ASSERT(nfs_zone() == VTOMI4(vp
)->mi_zone
);
15988 locktype
= flk_to_locktype(cmd
, flk
->l_type
);
15989 nfs4frlock_save_lost_rqst(NFS4_LCK_CTYPE_REINSTATE
, EINTR
, locktype
,
15990 NULL
, NULL
, lop
, flk
, &req
, cr
, vp
);
15991 (void) nfs4_start_recovery(&e
, VTOMI4(vp
), vp
, NULL
, NULL
,
15992 (req
.lr_op
== OP_LOCK
|| req
.lr_op
== OP_LOCKU
) ?
15993 &req
: NULL
, flk
->l_type
== F_UNLCK
? OP_LOCKU
: OP_LOCK
,