2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
37 * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $
38 * $DragonFly: src/sys/vfs/nfs/nfs_vnops.c,v 1.79 2008/07/16 18:20:40 dillon Exp $
43 * vnode op calls for Sun NFS version 2 and 3
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/resourcevar.h>
53 #include <sys/mount.h>
55 #include <sys/malloc.h>
57 #include <sys/namei.h>
58 #include <sys/nlookup.h>
59 #include <sys/socket.h>
60 #include <sys/vnode.h>
61 #include <sys/dirent.h>
62 #include <sys/fcntl.h>
63 #include <sys/lockf.h>
65 #include <sys/sysctl.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vm_zone.h>
74 #include <vfs/fifofs/fifo.h>
75 #include <vfs/ufs/dir.h>
85 #include "nfsm_subs.h"
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
91 #include <sys/thread2.h>
97 static int nfsspec_read (struct vop_read_args
*);
98 static int nfsspec_write (struct vop_write_args
*);
99 static int nfsfifo_read (struct vop_read_args
*);
100 static int nfsfifo_write (struct vop_write_args
*);
101 static int nfsspec_close (struct vop_close_args
*);
102 static int nfsfifo_close (struct vop_close_args
*);
103 #define nfs_poll vop_nopoll
104 static int nfs_setattrrpc (struct vnode
*,struct vattr
*,struct ucred
*,struct thread
*);
105 static int nfs_lookup (struct vop_old_lookup_args
*);
106 static int nfs_create (struct vop_old_create_args
*);
107 static int nfs_mknod (struct vop_old_mknod_args
*);
108 static int nfs_open (struct vop_open_args
*);
109 static int nfs_close (struct vop_close_args
*);
110 static int nfs_access (struct vop_access_args
*);
111 static int nfs_getattr (struct vop_getattr_args
*);
112 static int nfs_setattr (struct vop_setattr_args
*);
113 static int nfs_read (struct vop_read_args
*);
114 static int nfs_mmap (struct vop_mmap_args
*);
115 static int nfs_fsync (struct vop_fsync_args
*);
116 static int nfs_remove (struct vop_old_remove_args
*);
117 static int nfs_link (struct vop_old_link_args
*);
118 static int nfs_rename (struct vop_old_rename_args
*);
119 static int nfs_mkdir (struct vop_old_mkdir_args
*);
120 static int nfs_rmdir (struct vop_old_rmdir_args
*);
121 static int nfs_symlink (struct vop_old_symlink_args
*);
122 static int nfs_readdir (struct vop_readdir_args
*);
123 static int nfs_bmap (struct vop_bmap_args
*);
124 static int nfs_strategy (struct vop_strategy_args
*);
125 static int nfs_lookitup (struct vnode
*, const char *, int,
126 struct ucred
*, struct thread
*, struct nfsnode
**);
127 static int nfs_sillyrename (struct vnode
*,struct vnode
*,struct componentname
*);
128 static int nfsspec_access (struct vop_access_args
*);
129 static int nfs_readlink (struct vop_readlink_args
*);
130 static int nfs_print (struct vop_print_args
*);
131 static int nfs_advlock (struct vop_advlock_args
*);
133 static int nfs_nresolve (struct vop_nresolve_args
*);
135 * Global vfs data structures for nfs
137 struct vop_ops nfsv2_vnode_vops
= {
138 .vop_default
= vop_defaultop
,
139 .vop_access
= nfs_access
,
140 .vop_advlock
= nfs_advlock
,
141 .vop_bmap
= nfs_bmap
,
142 .vop_close
= nfs_close
,
143 .vop_old_create
= nfs_create
,
144 .vop_fsync
= nfs_fsync
,
145 .vop_getattr
= nfs_getattr
,
146 .vop_getpages
= nfs_getpages
,
147 .vop_putpages
= nfs_putpages
,
148 .vop_inactive
= nfs_inactive
,
149 .vop_old_link
= nfs_link
,
150 .vop_old_lookup
= nfs_lookup
,
151 .vop_old_mkdir
= nfs_mkdir
,
152 .vop_old_mknod
= nfs_mknod
,
153 .vop_mmap
= nfs_mmap
,
154 .vop_open
= nfs_open
,
155 .vop_poll
= nfs_poll
,
156 .vop_print
= nfs_print
,
157 .vop_read
= nfs_read
,
158 .vop_readdir
= nfs_readdir
,
159 .vop_readlink
= nfs_readlink
,
160 .vop_reclaim
= nfs_reclaim
,
161 .vop_old_remove
= nfs_remove
,
162 .vop_old_rename
= nfs_rename
,
163 .vop_old_rmdir
= nfs_rmdir
,
164 .vop_setattr
= nfs_setattr
,
165 .vop_strategy
= nfs_strategy
,
166 .vop_old_symlink
= nfs_symlink
,
167 .vop_write
= nfs_write
,
168 .vop_nresolve
= nfs_nresolve
172 * Special device vnode ops
174 struct vop_ops nfsv2_spec_vops
= {
175 .vop_default
= spec_vnoperate
,
176 .vop_access
= nfsspec_access
,
177 .vop_close
= nfsspec_close
,
178 .vop_fsync
= nfs_fsync
,
179 .vop_getattr
= nfs_getattr
,
180 .vop_inactive
= nfs_inactive
,
181 .vop_print
= nfs_print
,
182 .vop_read
= nfsspec_read
,
183 .vop_reclaim
= nfs_reclaim
,
184 .vop_setattr
= nfs_setattr
,
185 .vop_write
= nfsspec_write
188 struct vop_ops nfsv2_fifo_vops
= {
189 .vop_default
= fifo_vnoperate
,
190 .vop_access
= nfsspec_access
,
191 .vop_close
= nfsfifo_close
,
192 .vop_fsync
= nfs_fsync
,
193 .vop_getattr
= nfs_getattr
,
194 .vop_inactive
= nfs_inactive
,
195 .vop_print
= nfs_print
,
196 .vop_read
= nfsfifo_read
,
197 .vop_reclaim
= nfs_reclaim
,
198 .vop_setattr
= nfs_setattr
,
199 .vop_write
= nfsfifo_write
202 static int nfs_mknodrpc (struct vnode
*dvp
, struct vnode
**vpp
,
203 struct componentname
*cnp
,
205 static int nfs_removerpc (struct vnode
*dvp
, const char *name
,
207 struct ucred
*cred
, struct thread
*td
);
208 static int nfs_renamerpc (struct vnode
*fdvp
, const char *fnameptr
,
209 int fnamelen
, struct vnode
*tdvp
,
210 const char *tnameptr
, int tnamelen
,
211 struct ucred
*cred
, struct thread
*td
);
212 static int nfs_renameit (struct vnode
*sdvp
,
213 struct componentname
*scnp
,
214 struct sillyrename
*sp
);
219 extern u_int32_t nfs_true
, nfs_false
;
220 extern u_int32_t nfs_xdrneg1
;
221 extern struct nfsstats nfsstats
;
222 extern nfstype nfsv3_type
[9];
223 struct thread
*nfs_iodwant
[NFS_MAXASYNCDAEMON
];
224 struct nfsmount
*nfs_iodmount
[NFS_MAXASYNCDAEMON
];
225 int nfs_numasync
= 0;
227 SYSCTL_DECL(_vfs_nfs
);
229 static int nfs_flush_on_rename
= 1;
230 SYSCTL_INT(_vfs_nfs
, OID_AUTO
, flush_on_rename
, CTLFLAG_RW
,
231 &nfs_flush_on_rename
, 0, "flush fvp prior to rename");
233 static int nfsaccess_cache_timeout
= NFS_DEFATTRTIMO
;
234 SYSCTL_INT(_vfs_nfs
, OID_AUTO
, access_cache_timeout
, CTLFLAG_RW
,
235 &nfsaccess_cache_timeout
, 0, "NFS ACCESS cache timeout");
237 static int nfsneg_cache_timeout
= NFS_MINATTRTIMO
;
238 SYSCTL_INT(_vfs_nfs
, OID_AUTO
, neg_cache_timeout
, CTLFLAG_RW
,
239 &nfsneg_cache_timeout
, 0, "NFS NEGATIVE NAMECACHE timeout");
241 static int nfspos_cache_timeout
= NFS_MINATTRTIMO
;
242 SYSCTL_INT(_vfs_nfs
, OID_AUTO
, pos_cache_timeout
, CTLFLAG_RW
,
243 &nfspos_cache_timeout
, 0, "NFS POSITIVE NAMECACHE timeout");
245 static int nfsv3_commit_on_close
= 0;
246 SYSCTL_INT(_vfs_nfs
, OID_AUTO
, nfsv3_commit_on_close
, CTLFLAG_RW
,
247 &nfsv3_commit_on_close
, 0, "write+commit on close, else only write");
249 SYSCTL_INT(_vfs_nfs
, OID_AUTO
, access_cache_hits
, CTLFLAG_RD
,
250 &nfsstats
.accesscache_hits
, 0, "NFS ACCESS cache hit count");
252 SYSCTL_INT(_vfs_nfs
, OID_AUTO
, access_cache_misses
, CTLFLAG_RD
,
253 &nfsstats
.accesscache_misses
, 0, "NFS ACCESS cache miss count");
256 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
257 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
258 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
260 nfs3_access_otw(struct vnode
*vp
, int wmode
,
261 struct thread
*td
, struct ucred
*cred
)
265 int error
= 0, attrflag
;
267 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
268 caddr_t bpos
, dpos
, cp2
;
272 struct nfsnode
*np
= VTONFS(vp
);
274 nfsstats
.rpccnt
[NFSPROC_ACCESS
]++;
275 nfsm_reqhead(vp
, NFSPROC_ACCESS
, NFSX_FH(v3
) + NFSX_UNSIGNED
);
277 nfsm_build(tl
, u_int32_t
*, NFSX_UNSIGNED
);
278 *tl
= txdr_unsigned(wmode
);
279 nfsm_request(vp
, NFSPROC_ACCESS
, td
, cred
);
280 nfsm_postop_attr(vp
, attrflag
, NFS_LATTR_NOSHRINK
);
282 nfsm_dissect(tl
, u_int32_t
*, NFSX_UNSIGNED
);
283 rmode
= fxdr_unsigned(u_int32_t
, *tl
);
285 np
->n_modeuid
= cred
->cr_uid
;
286 np
->n_modestamp
= mycpu
->gd_time_seconds
;
294 * nfs access vnode op.
295 * For nfs version 2, just return ok. File accesses may fail later.
296 * For nfs version 3, use the access rpc to check accessibility. If file modes
297 * are changed on the server, accesses might still fail later.
299 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
302 nfs_access(struct vop_access_args
*ap
)
304 struct vnode
*vp
= ap
->a_vp
;
305 thread_t td
= curthread
;
307 u_int32_t mode
, wmode
;
308 int v3
= NFS_ISV3(vp
);
309 struct nfsnode
*np
= VTONFS(vp
);
312 * Disallow write attempts on filesystems mounted read-only;
313 * unless the file is a socket, fifo, or a block or character
314 * device resident on the filesystem.
316 if ((ap
->a_mode
& VWRITE
) && (vp
->v_mount
->mnt_flag
& MNT_RDONLY
)) {
317 switch (vp
->v_type
) {
327 * For nfs v3, check to see if we have done this recently, and if
328 * so return our cached result instead of making an ACCESS call.
329 * If not, do an access rpc, otherwise you are stuck emulating
330 * ufs_access() locally using the vattr. This may not be correct,
331 * since the server may apply other access criteria such as
332 * client uid-->server uid mapping that we do not know about.
335 if (ap
->a_mode
& VREAD
)
336 mode
= NFSV3ACCESS_READ
;
339 if (vp
->v_type
!= VDIR
) {
340 if (ap
->a_mode
& VWRITE
)
341 mode
|= (NFSV3ACCESS_MODIFY
| NFSV3ACCESS_EXTEND
);
342 if (ap
->a_mode
& VEXEC
)
343 mode
|= NFSV3ACCESS_EXECUTE
;
345 if (ap
->a_mode
& VWRITE
)
346 mode
|= (NFSV3ACCESS_MODIFY
| NFSV3ACCESS_EXTEND
|
348 if (ap
->a_mode
& VEXEC
)
349 mode
|= NFSV3ACCESS_LOOKUP
;
351 /* XXX safety belt, only make blanket request if caching */
352 if (nfsaccess_cache_timeout
> 0) {
353 wmode
= NFSV3ACCESS_READ
| NFSV3ACCESS_MODIFY
|
354 NFSV3ACCESS_EXTEND
| NFSV3ACCESS_EXECUTE
|
355 NFSV3ACCESS_DELETE
| NFSV3ACCESS_LOOKUP
;
361 * Does our cached result allow us to give a definite yes to
364 if (np
->n_modestamp
&&
365 (mycpu
->gd_time_seconds
< (np
->n_modestamp
+ nfsaccess_cache_timeout
)) &&
366 (ap
->a_cred
->cr_uid
== np
->n_modeuid
) &&
367 ((np
->n_mode
& mode
) == mode
)) {
368 nfsstats
.accesscache_hits
++;
371 * Either a no, or a don't know. Go to the wire.
373 nfsstats
.accesscache_misses
++;
374 error
= nfs3_access_otw(vp
, wmode
, td
, ap
->a_cred
);
376 if ((np
->n_mode
& mode
) != mode
) {
382 if ((error
= nfsspec_access(ap
)) != 0)
386 * Attempt to prevent a mapped root from accessing a file
387 * which it shouldn't. We try to read a byte from the file
388 * if the user is root and the file is not zero length.
389 * After calling nfsspec_access, we should have the correct
392 if (ap
->a_cred
->cr_uid
== 0 && (ap
->a_mode
& VREAD
)
393 && VTONFS(vp
)->n_size
> 0) {
400 auio
.uio_iov
= &aiov
;
404 auio
.uio_segflg
= UIO_SYSSPACE
;
405 auio
.uio_rw
= UIO_READ
;
408 if (vp
->v_type
== VREG
) {
409 error
= nfs_readrpc(vp
, &auio
);
410 } else if (vp
->v_type
== VDIR
) {
412 bp
= kmalloc(NFS_DIRBLKSIZ
, M_TEMP
, M_WAITOK
);
414 aiov
.iov_len
= auio
.uio_resid
= NFS_DIRBLKSIZ
;
415 error
= nfs_readdirrpc(vp
, &auio
);
417 } else if (vp
->v_type
== VLNK
) {
418 error
= nfs_readlinkrpc(vp
, &auio
);
425 * [re]record creds for reading and/or writing if access
426 * was granted. Assume the NFS server will grant read access
427 * for execute requests.
430 if ((ap
->a_mode
& (VREAD
|VEXEC
)) && ap
->a_cred
!= np
->n_rucred
) {
433 crfree(np
->n_rucred
);
434 np
->n_rucred
= ap
->a_cred
;
436 if ((ap
->a_mode
& VWRITE
) && ap
->a_cred
!= np
->n_wucred
) {
439 crfree(np
->n_wucred
);
440 np
->n_wucred
= ap
->a_cred
;
448 * Check to see if the type is ok
449 * and that deletion is not in progress.
450 * For paged in text files, you will need to flush the page cache
451 * if consistency is lost.
453 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
458 nfs_open(struct vop_open_args
*ap
)
460 struct vnode
*vp
= ap
->a_vp
;
461 struct nfsnode
*np
= VTONFS(vp
);
465 if (vp
->v_type
!= VREG
&& vp
->v_type
!= VDIR
&& vp
->v_type
!= VLNK
) {
467 kprintf("open eacces vtyp=%d\n",vp
->v_type
);
473 * Clear the attribute cache only if opening with write access. It
474 * is unclear if we should do this at all here, but we certainly
475 * should not clear the cache unconditionally simply because a file
478 if (ap
->a_mode
& FWRITE
)
482 * For normal NFS, reconcile changes made locally verses
483 * changes made remotely. Note that VOP_GETATTR only goes
484 * to the wire if the cached attribute has timed out or been
487 * If local modifications have been made clear the attribute
488 * cache to force an attribute and modified time check. If
489 * GETATTR detects that the file has been changed by someone
490 * other then us it will set NRMODIFIED.
492 * If we are opening a directory and local changes have been
493 * made we have to invalidate the cache in order to ensure
494 * that we get the most up-to-date information from the
497 if (np
->n_flag
& NLMODIFIED
) {
499 if (vp
->v_type
== VDIR
) {
500 error
= nfs_vinvalbuf(vp
, V_SAVE
, 1);
506 error
= VOP_GETATTR(vp
, &vattr
);
509 if (np
->n_flag
& NRMODIFIED
) {
510 if (vp
->v_type
== VDIR
)
512 error
= nfs_vinvalbuf(vp
, V_SAVE
, 1);
515 np
->n_flag
&= ~NRMODIFIED
;
518 return (vop_stdopen(ap
));
523 * What an NFS client should do upon close after writing is a debatable issue.
524 * Most NFS clients push delayed writes to the server upon close, basically for
526 * 1 - So that any write errors may be reported back to the client process
527 * doing the close system call. By far the two most likely errors are
528 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
529 * 2 - To put a worst case upper bound on cache inconsistency between
530 * multiple clients for the file.
531 * There is also a consistency problem for Version 2 of the protocol w.r.t.
532 * not being able to tell if other clients are writing a file concurrently,
533 * since there is no way of knowing if the changed modify time in the reply
534 * is only due to the write for this client.
535 * (NFS Version 3 provides weak cache consistency data in the reply that
536 * should be sufficient to detect and handle this case.)
538 * The current code does the following:
539 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
540 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
541 * or commit them (this satisfies 1 and 2 except for the
542 * case where the server crashes after this close but
543 * before the commit RPC, which is felt to be "good
544 * enough". Changing the last argument to nfs_flush() to
545 * a 1 would force a commit operation, if it is felt a
546 * commit is necessary now.
547 * for NQNFS - do nothing now, since 2 is dealt with via leases and
548 * 1 should be dealt with via an fsync() system call for
549 * cases where write errors are important.
551 * nfs_close(struct vnode *a_vp, int a_fflag)
555 nfs_close(struct vop_close_args
*ap
)
557 struct vnode
*vp
= ap
->a_vp
;
558 struct nfsnode
*np
= VTONFS(vp
);
560 thread_t td
= curthread
;
562 if (vp
->v_type
== VREG
) {
563 if (np
->n_flag
& NLMODIFIED
) {
566 * Under NFSv3 we have dirty buffers to dispose of. We
567 * must flush them to the NFS server. We have the option
568 * of waiting all the way through the commit rpc or just
569 * waiting for the initial write. The default is to only
570 * wait through the initial write so the data is in the
571 * server's cache, which is roughly similar to the state
572 * a standard disk subsystem leaves the file in on close().
574 * We cannot clear the NLMODIFIED bit in np->n_flag due to
575 * potential races with other processes, and certainly
576 * cannot clear it if we don't commit.
578 int cm
= nfsv3_commit_on_close
? 1 : 0;
579 error
= nfs_flush(vp
, MNT_WAIT
, td
, cm
);
580 /* np->n_flag &= ~NLMODIFIED; */
582 error
= nfs_vinvalbuf(vp
, V_SAVE
, 1);
586 if (np
->n_flag
& NWRITEERR
) {
587 np
->n_flag
&= ~NWRITEERR
;
596 * nfs getattr call from vfs.
598 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
601 nfs_getattr(struct vop_getattr_args
*ap
)
603 struct vnode
*vp
= ap
->a_vp
;
604 struct nfsnode
*np
= VTONFS(vp
);
610 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
611 int v3
= NFS_ISV3(vp
);
612 thread_t td
= curthread
;
615 * Update local times for special files.
617 if (np
->n_flag
& (NACC
| NUPD
))
620 * First look in the cache.
622 if (nfs_getattrcache(vp
, ap
->a_vap
) == 0)
625 if (v3
&& nfsaccess_cache_timeout
> 0) {
626 nfsstats
.accesscache_misses
++;
627 nfs3_access_otw(vp
, NFSV3ACCESS_ALL
, td
, nfs_vpcred(vp
, ND_CHECK
));
628 if (nfs_getattrcache(vp
, ap
->a_vap
) == 0)
632 nfsstats
.rpccnt
[NFSPROC_GETATTR
]++;
633 nfsm_reqhead(vp
, NFSPROC_GETATTR
, NFSX_FH(v3
));
635 nfsm_request(vp
, NFSPROC_GETATTR
, td
, nfs_vpcred(vp
, ND_CHECK
));
637 nfsm_loadattr(vp
, ap
->a_vap
);
647 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
650 nfs_setattr(struct vop_setattr_args
*ap
)
652 struct vnode
*vp
= ap
->a_vp
;
653 struct nfsnode
*np
= VTONFS(vp
);
654 struct vattr
*vap
= ap
->a_vap
;
657 thread_t td
= curthread
;
664 * Setting of flags is not supported.
666 if (vap
->va_flags
!= VNOVAL
)
670 * Disallow write attempts if the filesystem is mounted read-only.
672 if ((vap
->va_flags
!= VNOVAL
|| vap
->va_uid
!= (uid_t
)VNOVAL
||
673 vap
->va_gid
!= (gid_t
)VNOVAL
|| vap
->va_atime
.tv_sec
!= VNOVAL
||
674 vap
->va_mtime
.tv_sec
!= VNOVAL
|| vap
->va_mode
!= (mode_t
)VNOVAL
) &&
675 (vp
->v_mount
->mnt_flag
& MNT_RDONLY
))
678 if (vap
->va_size
!= VNOVAL
) {
680 * truncation requested
682 switch (vp
->v_type
) {
689 if (vap
->va_mtime
.tv_sec
== VNOVAL
&&
690 vap
->va_atime
.tv_sec
== VNOVAL
&&
691 vap
->va_mode
== (mode_t
)VNOVAL
&&
692 vap
->va_uid
== (uid_t
)VNOVAL
&&
693 vap
->va_gid
== (gid_t
)VNOVAL
)
695 vap
->va_size
= VNOVAL
;
699 * Disallow write attempts if the filesystem is
702 if (vp
->v_mount
->mnt_flag
& MNT_RDONLY
)
706 * This is nasty. The RPCs we send to flush pending
707 * data often return attribute information which is
708 * cached via a callback to nfs_loadattrcache(), which
709 * has the effect of changing our notion of the file
710 * size. Due to flushed appends and other operations
711 * the file size can be set to virtually anything,
712 * including values that do not match either the old
713 * or intended file size.
715 * When this condition is detected we must loop to
716 * try the operation again. Hopefully no more
717 * flushing is required on the loop so it works the
718 * second time around. THIS CASE ALMOST ALWAYS
723 error
= nfs_meta_setsize(vp
, td
, vap
->va_size
);
725 if (np
->n_flag
& NLMODIFIED
) {
726 if (vap
->va_size
== 0)
727 error
= nfs_vinvalbuf(vp
, 0, 1);
729 error
= nfs_vinvalbuf(vp
, V_SAVE
, 1);
732 * note: this loop case almost always happens at
733 * least once per truncation.
735 if (error
== 0 && np
->n_size
!= vap
->va_size
)
737 np
->n_vattr
.va_size
= vap
->va_size
;
740 } else if ((np
->n_flag
& NLMODIFIED
) && vp
->v_type
== VREG
) {
742 * What to do. If we are modifying the mtime we lose
743 * mtime detection of changes made by the server or other
744 * clients. But programs like rsync/rdist/cpdup are going
745 * to call utimes a lot. We don't want to piecemeal sync.
747 * For now sync if any prior remote changes were detected,
748 * but allow us to lose track of remote changes made during
749 * the utimes operation.
751 if (np
->n_flag
& NRMODIFIED
)
752 error
= nfs_vinvalbuf(vp
, V_SAVE
, 1);
756 if (vap
->va_mtime
.tv_sec
!= VNOVAL
) {
757 np
->n_mtime
= vap
->va_mtime
.tv_sec
;
761 error
= nfs_setattrrpc(vp
, vap
, ap
->a_cred
, td
);
764 * Sanity check if a truncation was issued. This should only occur
765 * if multiple processes are racing on the same file.
767 if (error
== 0 && vap
->va_size
!= VNOVAL
&&
768 np
->n_size
!= vap
->va_size
) {
769 kprintf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize
, vap
->va_size
, np
->n_size
);
772 if (error
&& vap
->va_size
!= VNOVAL
) {
773 np
->n_size
= np
->n_vattr
.va_size
= tsize
;
774 vnode_pager_setsize(vp
, np
->n_size
);
780 * Do an nfs setattr rpc.
783 nfs_setattrrpc(struct vnode
*vp
, struct vattr
*vap
,
784 struct ucred
*cred
, struct thread
*td
)
786 struct nfsv2_sattr
*sp
;
787 struct nfsnode
*np
= VTONFS(vp
);
790 caddr_t bpos
, dpos
, cp2
;
792 int error
= 0, wccflag
= NFSV3_WCCRATTR
;
793 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
794 int v3
= NFS_ISV3(vp
);
796 nfsstats
.rpccnt
[NFSPROC_SETATTR
]++;
797 nfsm_reqhead(vp
, NFSPROC_SETATTR
, NFSX_FH(v3
) + NFSX_SATTR(v3
));
800 nfsm_v3attrbuild(vap
, TRUE
);
801 nfsm_build(tl
, u_int32_t
*, NFSX_UNSIGNED
);
804 nfsm_build(sp
, struct nfsv2_sattr
*, NFSX_V2SATTR
);
805 if (vap
->va_mode
== (mode_t
)VNOVAL
)
806 sp
->sa_mode
= nfs_xdrneg1
;
808 sp
->sa_mode
= vtonfsv2_mode(vp
->v_type
, vap
->va_mode
);
809 if (vap
->va_uid
== (uid_t
)VNOVAL
)
810 sp
->sa_uid
= nfs_xdrneg1
;
812 sp
->sa_uid
= txdr_unsigned(vap
->va_uid
);
813 if (vap
->va_gid
== (gid_t
)VNOVAL
)
814 sp
->sa_gid
= nfs_xdrneg1
;
816 sp
->sa_gid
= txdr_unsigned(vap
->va_gid
);
817 sp
->sa_size
= txdr_unsigned(vap
->va_size
);
818 txdr_nfsv2time(&vap
->va_atime
, &sp
->sa_atime
);
819 txdr_nfsv2time(&vap
->va_mtime
, &sp
->sa_mtime
);
821 nfsm_request(vp
, NFSPROC_SETATTR
, td
, cred
);
824 nfsm_wcc_data(vp
, wccflag
);
826 nfsm_loadattr(vp
, (struct vattr
*)0);
834 nfs_cache_setvp(struct nchandle
*nch
, struct vnode
*vp
, int nctimeout
)
840 cache_setvp(nch
, vp
);
841 cache_settimeout(nch
, nctimeout
);
845 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
846 * nfs_lookup() until all remaining new api calls are implemented.
848 * Resolve a namecache entry. This function is passed a locked ncp and
849 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
852 nfs_nresolve(struct vop_nresolve_args
*ap
)
854 struct thread
*td
= curthread
;
855 struct namecache
*ncp
;
866 /******NFSM MACROS********/
867 struct mbuf
*mb
, *mrep
, *mreq
, *mb2
, *md
;
868 caddr_t bpos
, dpos
, cp
, cp2
;
875 if ((error
= vget(dvp
, LK_SHARED
)) != 0)
880 nfsstats
.lookupcache_misses
++;
881 nfsstats
.rpccnt
[NFSPROC_LOOKUP
]++;
882 ncp
= ap
->a_nch
->ncp
;
884 nfsm_reqhead(dvp
, NFSPROC_LOOKUP
,
885 NFSX_FH(v3
) + NFSX_UNSIGNED
+ nfsm_rndup(len
));
887 nfsm_strtom(ncp
->nc_name
, len
, NFS_MAXNAMLEN
);
888 nfsm_request(dvp
, NFSPROC_LOOKUP
, td
, ap
->a_cred
);
891 * Cache negatve lookups to reduce NFS traffic, but use
892 * a fast timeout. Otherwise use a timeout of 1 tick.
893 * XXX we should add a namecache flag for no-caching
894 * to uncache the negative hit as soon as possible, but
895 * we cannot simply destroy the entry because it is used
896 * as a placeholder by the caller.
899 nfs_cache_setvp(ap
->a_nch
, NULL
, nfsneg_cache_timeout
);
900 nfsm_postop_attr(dvp
, attrflag
, NFS_LATTR_NOSHRINK
);
906 * Success, get the file handle, do various checks, and load
907 * post-operation data from the reply packet. Theoretically
908 * we should never be looking up "." so, theoretically, we
909 * should never get the same file handle as our directory. But
910 * we check anyway. XXX
912 * Note that no timeout is set for the positive cache hit. We
913 * assume, theoretically, that ESTALE returns will be dealt with
914 * properly to handle NFS races and in anycase we cannot depend
915 * on a timeout to deal with NFS open/create/excl issues so instead
916 * of a bad hack here the rest of the NFS client code needs to do
919 nfsm_getfh(fhp
, fhsize
, v3
);
922 if (NFS_CMPFH(np
, fhp
, fhsize
)) {
926 error
= nfs_nget(dvp
->v_mount
, fhp
, fhsize
, &np
);
935 nfsm_postop_attr(nvp
, attrflag
, NFS_LATTR_NOSHRINK
);
936 nfsm_postop_attr(dvp
, attrflag
, NFS_LATTR_NOSHRINK
);
938 nfsm_loadattr(nvp
, NULL
);
940 nfs_cache_setvp(ap
->a_nch
, nvp
, nfspos_cache_timeout
);
954 * 'cached' nfs directory lookup
956 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
958 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
959 * struct componentname *a_cnp)
962 nfs_lookup(struct vop_old_lookup_args
*ap
)
964 struct componentname
*cnp
= ap
->a_cnp
;
965 struct vnode
*dvp
= ap
->a_dvp
;
966 struct vnode
**vpp
= ap
->a_vpp
;
967 int flags
= cnp
->cn_flags
;
972 struct nfsmount
*nmp
;
973 caddr_t bpos
, dpos
, cp2
;
974 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
978 int lockparent
, wantparent
, error
= 0, attrflag
, fhsize
;
979 int v3
= NFS_ISV3(dvp
);
982 * Read-only mount check and directory check.
985 if ((dvp
->v_mount
->mnt_flag
& MNT_RDONLY
) &&
986 (cnp
->cn_nameiop
== NAMEI_DELETE
|| cnp
->cn_nameiop
== NAMEI_RENAME
))
989 if (dvp
->v_type
!= VDIR
)
993 * Look it up in the cache. Note that ENOENT is only returned if we
994 * previously entered a negative hit (see later on). The additional
995 * nfsneg_cache_timeout check causes previously cached results to
996 * be instantly ignored if the negative caching is turned off.
998 lockparent
= flags
& CNP_LOCKPARENT
;
999 wantparent
= flags
& (CNP_LOCKPARENT
|CNP_WANTPARENT
);
1000 nmp
= VFSTONFS(dvp
->v_mount
);
1008 nfsstats
.lookupcache_misses
++;
1009 nfsstats
.rpccnt
[NFSPROC_LOOKUP
]++;
1010 len
= cnp
->cn_namelen
;
1011 nfsm_reqhead(dvp
, NFSPROC_LOOKUP
,
1012 NFSX_FH(v3
) + NFSX_UNSIGNED
+ nfsm_rndup(len
));
1013 nfsm_fhtom(dvp
, v3
);
1014 nfsm_strtom(cnp
->cn_nameptr
, len
, NFS_MAXNAMLEN
);
1015 nfsm_request(dvp
, NFSPROC_LOOKUP
, cnp
->cn_td
, cnp
->cn_cred
);
1017 nfsm_postop_attr(dvp
, attrflag
, NFS_LATTR_NOSHRINK
);
1021 nfsm_getfh(fhp
, fhsize
, v3
);
1024 * Handle RENAME case...
1026 if (cnp
->cn_nameiop
== NAMEI_RENAME
&& wantparent
) {
1027 if (NFS_CMPFH(np
, fhp
, fhsize
)) {
1031 error
= nfs_nget(dvp
->v_mount
, fhp
, fhsize
, &np
);
1038 nfsm_postop_attr(newvp
, attrflag
, NFS_LATTR_NOSHRINK
);
1039 nfsm_postop_attr(dvp
, attrflag
, NFS_LATTR_NOSHRINK
);
1041 nfsm_loadattr(newvp
, (struct vattr
*)0);
1046 cnp
->cn_flags
|= CNP_PDIRUNLOCK
;
1051 if (flags
& CNP_ISDOTDOT
) {
1053 cnp
->cn_flags
|= CNP_PDIRUNLOCK
;
1054 error
= nfs_nget(dvp
->v_mount
, fhp
, fhsize
, &np
);
1056 vn_lock(dvp
, LK_EXCLUSIVE
| LK_RETRY
);
1057 cnp
->cn_flags
&= ~CNP_PDIRUNLOCK
;
1058 return (error
); /* NOTE: return error from nget */
1062 error
= vn_lock(dvp
, LK_EXCLUSIVE
);
1067 cnp
->cn_flags
|= CNP_PDIRUNLOCK
;
1069 } else if (NFS_CMPFH(np
, fhp
, fhsize
)) {
1073 error
= nfs_nget(dvp
->v_mount
, fhp
, fhsize
, &np
);
1080 cnp
->cn_flags
|= CNP_PDIRUNLOCK
;
1085 nfsm_postop_attr(newvp
, attrflag
, NFS_LATTR_NOSHRINK
);
1086 nfsm_postop_attr(dvp
, attrflag
, NFS_LATTR_NOSHRINK
);
1088 nfsm_loadattr(newvp
, (struct vattr
*)0);
1090 /* XXX MOVE TO nfs_nremove() */
1091 if ((cnp
->cn_flags
& CNP_MAKEENTRY
) &&
1092 cnp
->cn_nameiop
!= NAMEI_DELETE
) {
1093 np
->n_ctime
= np
->n_vattr
.va_ctime
.tv_sec
; /* XXX */
1100 if (newvp
!= NULLVP
) {
1104 if ((cnp
->cn_nameiop
== NAMEI_CREATE
||
1105 cnp
->cn_nameiop
== NAMEI_RENAME
) &&
1109 cnp
->cn_flags
|= CNP_PDIRUNLOCK
;
1111 if (dvp
->v_mount
->mnt_flag
& MNT_RDONLY
)
1114 error
= EJUSTRETURN
;
1122 * Just call nfs_bioread() to do the work.
1124 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1125 * struct ucred *a_cred)
1128 nfs_read(struct vop_read_args
*ap
)
1130 struct vnode
*vp
= ap
->a_vp
;
1132 return (nfs_bioread(vp
, ap
->a_uio
, ap
->a_ioflag
));
1133 switch (vp
->v_type
) {
1135 return (nfs_bioread(vp
, ap
->a_uio
, ap
->a_ioflag
));
1146 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1149 nfs_readlink(struct vop_readlink_args
*ap
)
1151 struct vnode
*vp
= ap
->a_vp
;
1153 if (vp
->v_type
!= VLNK
)
1155 return (nfs_bioread(vp
, ap
->a_uio
, 0));
1159 * Do a readlink rpc.
1160 * Called by nfs_doio() from below the buffer cache.
1163 nfs_readlinkrpc(struct vnode
*vp
, struct uio
*uiop
)
1168 caddr_t bpos
, dpos
, cp2
;
1169 int error
= 0, len
, attrflag
;
1170 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
1171 int v3
= NFS_ISV3(vp
);
1173 nfsstats
.rpccnt
[NFSPROC_READLINK
]++;
1174 nfsm_reqhead(vp
, NFSPROC_READLINK
, NFSX_FH(v3
));
1176 nfsm_request(vp
, NFSPROC_READLINK
, uiop
->uio_td
, nfs_vpcred(vp
, ND_CHECK
));
1178 nfsm_postop_attr(vp
, attrflag
, NFS_LATTR_NOSHRINK
);
1180 nfsm_strsiz(len
, NFS_MAXPATHLEN
);
1181 if (len
== NFS_MAXPATHLEN
) {
1182 struct nfsnode
*np
= VTONFS(vp
);
1183 if (np
->n_size
&& np
->n_size
< NFS_MAXPATHLEN
)
1186 nfsm_mtouio(uiop
, len
);
1198 nfs_readrpc(struct vnode
*vp
, struct uio
*uiop
)
1203 caddr_t bpos
, dpos
, cp2
;
1204 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
1205 struct nfsmount
*nmp
;
1206 int error
= 0, len
, retlen
, tsiz
, eof
, attrflag
;
1207 int v3
= NFS_ISV3(vp
);
1212 nmp
= VFSTONFS(vp
->v_mount
);
1213 tsiz
= uiop
->uio_resid
;
1214 if (uiop
->uio_offset
+ tsiz
> nmp
->nm_maxfilesize
)
1217 nfsstats
.rpccnt
[NFSPROC_READ
]++;
1218 len
= (tsiz
> nmp
->nm_rsize
) ? nmp
->nm_rsize
: tsiz
;
1219 nfsm_reqhead(vp
, NFSPROC_READ
, NFSX_FH(v3
) + NFSX_UNSIGNED
* 3);
1221 nfsm_build(tl
, u_int32_t
*, NFSX_UNSIGNED
* 3);
1223 txdr_hyper(uiop
->uio_offset
, tl
);
1224 *(tl
+ 2) = txdr_unsigned(len
);
1226 *tl
++ = txdr_unsigned(uiop
->uio_offset
);
1227 *tl
++ = txdr_unsigned(len
);
1230 nfsm_request(vp
, NFSPROC_READ
, uiop
->uio_td
, nfs_vpcred(vp
, ND_READ
));
1232 nfsm_postop_attr(vp
, attrflag
, NFS_LATTR_NOSHRINK
);
1237 nfsm_dissect(tl
, u_int32_t
*, 2 * NFSX_UNSIGNED
);
1238 eof
= fxdr_unsigned(int, *(tl
+ 1));
1240 nfsm_loadattr(vp
, (struct vattr
*)0);
1241 nfsm_strsiz(retlen
, nmp
->nm_rsize
);
1242 nfsm_mtouio(uiop
, retlen
);
1246 if (eof
|| retlen
== 0) {
1249 } else if (retlen
< len
) {
1261 nfs_writerpc(struct vnode
*vp
, struct uio
*uiop
, int *iomode
, int *must_commit
)
1265 int32_t t1
, t2
, backup
;
1266 caddr_t bpos
, dpos
, cp2
;
1267 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
1268 struct nfsmount
*nmp
= VFSTONFS(vp
->v_mount
);
1269 int error
= 0, len
, tsiz
, wccflag
= NFSV3_WCCRATTR
, rlen
, commit
;
1270 int v3
= NFS_ISV3(vp
), committed
= NFSV3WRITE_FILESYNC
;
1273 if (uiop
->uio_iovcnt
!= 1)
1274 panic("nfs: writerpc iovcnt > 1");
1277 tsiz
= uiop
->uio_resid
;
1278 if (uiop
->uio_offset
+ tsiz
> nmp
->nm_maxfilesize
)
1281 nfsstats
.rpccnt
[NFSPROC_WRITE
]++;
1282 len
= (tsiz
> nmp
->nm_wsize
) ? nmp
->nm_wsize
: tsiz
;
1283 nfsm_reqhead(vp
, NFSPROC_WRITE
,
1284 NFSX_FH(v3
) + 5 * NFSX_UNSIGNED
+ nfsm_rndup(len
));
1287 nfsm_build(tl
, u_int32_t
*, 5 * NFSX_UNSIGNED
);
1288 txdr_hyper(uiop
->uio_offset
, tl
);
1290 *tl
++ = txdr_unsigned(len
);
1291 *tl
++ = txdr_unsigned(*iomode
);
1292 *tl
= txdr_unsigned(len
);
1296 nfsm_build(tl
, u_int32_t
*, 4 * NFSX_UNSIGNED
);
1297 /* Set both "begin" and "current" to non-garbage. */
1298 x
= txdr_unsigned((u_int32_t
)uiop
->uio_offset
);
1299 *tl
++ = x
; /* "begin offset" */
1300 *tl
++ = x
; /* "current offset" */
1301 x
= txdr_unsigned(len
);
1302 *tl
++ = x
; /* total to this offset */
1303 *tl
= x
; /* size of this write */
1305 nfsm_uiotom(uiop
, len
);
1306 nfsm_request(vp
, NFSPROC_WRITE
, uiop
->uio_td
, nfs_vpcred(vp
, ND_WRITE
));
1309 * The write RPC returns a before and after mtime. The
1310 * nfsm_wcc_data() macro checks the before n_mtime
1311 * against the before time and stores the after time
1312 * in the nfsnode's cached vattr and n_mtime field.
1313 * The NRMODIFIED bit will be set if the before
1314 * time did not match the original mtime.
1316 wccflag
= NFSV3_WCCCHK
;
1317 nfsm_wcc_data(vp
, wccflag
);
1319 nfsm_dissect(tl
, u_int32_t
*, 2 * NFSX_UNSIGNED
1320 + NFSX_V3WRITEVERF
);
1321 rlen
= fxdr_unsigned(int, *tl
++);
1326 } else if (rlen
< len
) {
1327 backup
= len
- rlen
;
1328 uiop
->uio_iov
->iov_base
-= backup
;
1329 uiop
->uio_iov
->iov_len
+= backup
;
1330 uiop
->uio_offset
-= backup
;
1331 uiop
->uio_resid
+= backup
;
1334 commit
= fxdr_unsigned(int, *tl
++);
1337 * Return the lowest committment level
1338 * obtained by any of the RPCs.
1340 if (committed
== NFSV3WRITE_FILESYNC
)
1342 else if (committed
== NFSV3WRITE_DATASYNC
&&
1343 commit
== NFSV3WRITE_UNSTABLE
)
1345 if ((nmp
->nm_state
& NFSSTA_HASWRITEVERF
) == 0){
1346 bcopy((caddr_t
)tl
, (caddr_t
)nmp
->nm_verf
,
1348 nmp
->nm_state
|= NFSSTA_HASWRITEVERF
;
1349 } else if (bcmp((caddr_t
)tl
,
1350 (caddr_t
)nmp
->nm_verf
, NFSX_V3WRITEVERF
)) {
1352 bcopy((caddr_t
)tl
, (caddr_t
)nmp
->nm_verf
,
1357 nfsm_loadattr(vp
, (struct vattr
*)0);
1365 if (vp
->v_mount
->mnt_flag
& MNT_ASYNC
)
1366 committed
= NFSV3WRITE_FILESYNC
;
1367 *iomode
= committed
;
1369 uiop
->uio_resid
= tsiz
;
1375 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1376 * mode set to specify the file type and the size field for rdev.
1379 nfs_mknodrpc(struct vnode
*dvp
, struct vnode
**vpp
, struct componentname
*cnp
,
1382 struct nfsv2_sattr
*sp
;
1386 struct vnode
*newvp
= (struct vnode
*)0;
1387 struct nfsnode
*np
= (struct nfsnode
*)0;
1391 int error
= 0, wccflag
= NFSV3_WCCRATTR
, gotvp
= 0;
1392 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
1394 int v3
= NFS_ISV3(dvp
);
1396 if (vap
->va_type
== VCHR
|| vap
->va_type
== VBLK
) {
1397 rmajor
= txdr_unsigned(vap
->va_rmajor
);
1398 rminor
= txdr_unsigned(vap
->va_rminor
);
1399 } else if (vap
->va_type
== VFIFO
|| vap
->va_type
== VSOCK
) {
1400 rmajor
= nfs_xdrneg1
;
1401 rminor
= nfs_xdrneg1
;
1403 return (EOPNOTSUPP
);
1405 if ((error
= VOP_GETATTR(dvp
, &vattr
)) != 0) {
1408 nfsstats
.rpccnt
[NFSPROC_MKNOD
]++;
1409 nfsm_reqhead(dvp
, NFSPROC_MKNOD
, NFSX_FH(v3
) + 4 * NFSX_UNSIGNED
+
1410 + nfsm_rndup(cnp
->cn_namelen
) + NFSX_SATTR(v3
));
1411 nfsm_fhtom(dvp
, v3
);
1412 nfsm_strtom(cnp
->cn_nameptr
, cnp
->cn_namelen
, NFS_MAXNAMLEN
);
1414 nfsm_build(tl
, u_int32_t
*, NFSX_UNSIGNED
);
1415 *tl
++ = vtonfsv3_type(vap
->va_type
);
1416 nfsm_v3attrbuild(vap
, FALSE
);
1417 if (vap
->va_type
== VCHR
|| vap
->va_type
== VBLK
) {
1418 nfsm_build(tl
, u_int32_t
*, 2 * NFSX_UNSIGNED
);
1419 *tl
++ = txdr_unsigned(vap
->va_rmajor
);
1420 *tl
= txdr_unsigned(vap
->va_rminor
);
1423 nfsm_build(sp
, struct nfsv2_sattr
*, NFSX_V2SATTR
);
1424 sp
->sa_mode
= vtonfsv2_mode(vap
->va_type
, vap
->va_mode
);
1425 sp
->sa_uid
= nfs_xdrneg1
;
1426 sp
->sa_gid
= nfs_xdrneg1
;
1427 sp
->sa_size
= makeudev(rmajor
, rminor
);
1428 txdr_nfsv2time(&vap
->va_atime
, &sp
->sa_atime
);
1429 txdr_nfsv2time(&vap
->va_mtime
, &sp
->sa_mtime
);
1431 nfsm_request(dvp
, NFSPROC_MKNOD
, cnp
->cn_td
, cnp
->cn_cred
);
1433 nfsm_mtofh(dvp
, newvp
, v3
, gotvp
);
1437 newvp
= (struct vnode
*)0;
1439 error
= nfs_lookitup(dvp
, cnp
->cn_nameptr
,
1440 cnp
->cn_namelen
, cnp
->cn_cred
, cnp
->cn_td
, &np
);
1446 nfsm_wcc_data(dvp
, wccflag
);
1455 VTONFS(dvp
)->n_flag
|= NLMODIFIED
;
1457 VTONFS(dvp
)->n_attrstamp
= 0;
1463 * just call nfs_mknodrpc() to do the work.
1465 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1466 * struct componentname *a_cnp, struct vattr *a_vap)
1470 nfs_mknod(struct vop_old_mknod_args
*ap
)
1472 return nfs_mknodrpc(ap
->a_dvp
, ap
->a_vpp
, ap
->a_cnp
, ap
->a_vap
);
1475 static u_long create_verf
;
1477 * nfs file create call
1479 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1480 * struct componentname *a_cnp, struct vattr *a_vap)
1483 nfs_create(struct vop_old_create_args
*ap
)
1485 struct vnode
*dvp
= ap
->a_dvp
;
1486 struct vattr
*vap
= ap
->a_vap
;
1487 struct componentname
*cnp
= ap
->a_cnp
;
1488 struct nfsv2_sattr
*sp
;
1492 struct nfsnode
*np
= (struct nfsnode
*)0;
1493 struct vnode
*newvp
= (struct vnode
*)0;
1494 caddr_t bpos
, dpos
, cp2
;
1495 int error
= 0, wccflag
= NFSV3_WCCRATTR
, gotvp
= 0, fmode
= 0;
1496 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
1498 int v3
= NFS_ISV3(dvp
);
1501 * Oops, not for me..
1503 if (vap
->va_type
== VSOCK
)
1504 return (nfs_mknodrpc(dvp
, ap
->a_vpp
, cnp
, vap
));
1506 if ((error
= VOP_GETATTR(dvp
, &vattr
)) != 0) {
1509 if (vap
->va_vaflags
& VA_EXCLUSIVE
)
1512 nfsstats
.rpccnt
[NFSPROC_CREATE
]++;
1513 nfsm_reqhead(dvp
, NFSPROC_CREATE
, NFSX_FH(v3
) + 2 * NFSX_UNSIGNED
+
1514 nfsm_rndup(cnp
->cn_namelen
) + NFSX_SATTR(v3
));
1515 nfsm_fhtom(dvp
, v3
);
1516 nfsm_strtom(cnp
->cn_nameptr
, cnp
->cn_namelen
, NFS_MAXNAMLEN
);
1518 nfsm_build(tl
, u_int32_t
*, NFSX_UNSIGNED
);
1519 if (fmode
& O_EXCL
) {
1520 *tl
= txdr_unsigned(NFSV3CREATE_EXCLUSIVE
);
1521 nfsm_build(tl
, u_int32_t
*, NFSX_V3CREATEVERF
);
1523 if (!TAILQ_EMPTY(&in_ifaddrheads
[mycpuid
]))
1524 *tl
++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads
[mycpuid
])->ia
)->sin_addr
.s_addr
;
1527 *tl
++ = create_verf
;
1528 *tl
= ++create_verf
;
1530 *tl
= txdr_unsigned(NFSV3CREATE_UNCHECKED
);
1531 nfsm_v3attrbuild(vap
, FALSE
);
1534 nfsm_build(sp
, struct nfsv2_sattr
*, NFSX_V2SATTR
);
1535 sp
->sa_mode
= vtonfsv2_mode(vap
->va_type
, vap
->va_mode
);
1536 sp
->sa_uid
= nfs_xdrneg1
;
1537 sp
->sa_gid
= nfs_xdrneg1
;
1539 txdr_nfsv2time(&vap
->va_atime
, &sp
->sa_atime
);
1540 txdr_nfsv2time(&vap
->va_mtime
, &sp
->sa_mtime
);
1542 nfsm_request(dvp
, NFSPROC_CREATE
, cnp
->cn_td
, cnp
->cn_cred
);
1544 nfsm_mtofh(dvp
, newvp
, v3
, gotvp
);
1548 newvp
= (struct vnode
*)0;
1550 error
= nfs_lookitup(dvp
, cnp
->cn_nameptr
,
1551 cnp
->cn_namelen
, cnp
->cn_cred
, cnp
->cn_td
, &np
);
1557 nfsm_wcc_data(dvp
, wccflag
);
1561 if (v3
&& (fmode
& O_EXCL
) && error
== NFSERR_NOTSUPP
) {
1562 KKASSERT(newvp
== NULL
);
1566 } else if (v3
&& (fmode
& O_EXCL
)) {
1568 * We are normally called with only a partially initialized
1569 * VAP. Since the NFSv3 spec says that server may use the
1570 * file attributes to store the verifier, the spec requires
1571 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1572 * in atime, but we can't really assume that all servers will
1573 * so we ensure that our SETATTR sets both atime and mtime.
1575 if (vap
->va_mtime
.tv_sec
== VNOVAL
)
1576 vfs_timestamp(&vap
->va_mtime
);
1577 if (vap
->va_atime
.tv_sec
== VNOVAL
)
1578 vap
->va_atime
= vap
->va_mtime
;
1579 error
= nfs_setattrrpc(newvp
, vap
, cnp
->cn_cred
, cnp
->cn_td
);
1583 * The new np may have enough info for access
1584 * checks, make sure rucred and wucred are
1585 * initialized for read and write rpc's.
1588 if (np
->n_rucred
== NULL
)
1589 np
->n_rucred
= crhold(cnp
->cn_cred
);
1590 if (np
->n_wucred
== NULL
)
1591 np
->n_wucred
= crhold(cnp
->cn_cred
);
1596 VTONFS(dvp
)->n_flag
|= NLMODIFIED
;
1598 VTONFS(dvp
)->n_attrstamp
= 0;
1603 * nfs file remove call
1604 * To try and make nfs semantics closer to ufs semantics, a file that has
1605 * other processes using the vnode is renamed instead of removed and then
1606 * removed later on the last close.
1607 * - If v_sysref.refcnt > 1
1608 * If a rename is not already in the works
1609 * call nfs_sillyrename() to set it up
1613 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1614 * struct componentname *a_cnp)
1617 nfs_remove(struct vop_old_remove_args
*ap
)
1619 struct vnode
*vp
= ap
->a_vp
;
1620 struct vnode
*dvp
= ap
->a_dvp
;
1621 struct componentname
*cnp
= ap
->a_cnp
;
1622 struct nfsnode
*np
= VTONFS(vp
);
1627 if (vp
->v_sysref
.refcnt
< 1)
1628 panic("nfs_remove: bad v_sysref.refcnt");
1630 if (vp
->v_type
== VDIR
)
1632 else if (vp
->v_sysref
.refcnt
== 1 || (np
->n_sillyrename
&&
1633 VOP_GETATTR(vp
, &vattr
) == 0 &&
1634 vattr
.va_nlink
> 1)) {
1636 * throw away biocache buffers, mainly to avoid
1637 * unnecessary delayed writes later.
1639 error
= nfs_vinvalbuf(vp
, 0, 1);
1642 error
= nfs_removerpc(dvp
, cnp
->cn_nameptr
,
1643 cnp
->cn_namelen
, cnp
->cn_cred
, cnp
->cn_td
);
1645 * Kludge City: If the first reply to the remove rpc is lost..
1646 * the reply to the retransmitted request will be ENOENT
1647 * since the file was in fact removed
1648 * Therefore, we cheat and return success.
1650 if (error
== ENOENT
)
1652 } else if (!np
->n_sillyrename
) {
1653 error
= nfs_sillyrename(dvp
, vp
, cnp
);
1655 np
->n_attrstamp
= 0;
1660 * nfs file remove rpc called from nfs_inactive
1663 nfs_removeit(struct sillyrename
*sp
)
1665 return (nfs_removerpc(sp
->s_dvp
, sp
->s_name
, sp
->s_namlen
,
1670 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1673 nfs_removerpc(struct vnode
*dvp
, const char *name
, int namelen
,
1674 struct ucred
*cred
, struct thread
*td
)
1679 caddr_t bpos
, dpos
, cp2
;
1680 int error
= 0, wccflag
= NFSV3_WCCRATTR
;
1681 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
1682 int v3
= NFS_ISV3(dvp
);
1684 nfsstats
.rpccnt
[NFSPROC_REMOVE
]++;
1685 nfsm_reqhead(dvp
, NFSPROC_REMOVE
,
1686 NFSX_FH(v3
) + NFSX_UNSIGNED
+ nfsm_rndup(namelen
));
1687 nfsm_fhtom(dvp
, v3
);
1688 nfsm_strtom(name
, namelen
, NFS_MAXNAMLEN
);
1689 nfsm_request(dvp
, NFSPROC_REMOVE
, td
, cred
);
1691 nfsm_wcc_data(dvp
, wccflag
);
1694 VTONFS(dvp
)->n_flag
|= NLMODIFIED
;
1696 VTONFS(dvp
)->n_attrstamp
= 0;
1701 * nfs file rename call
1703 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1704 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1705 * struct vnode *a_tvp, struct componentname *a_tcnp)
1708 nfs_rename(struct vop_old_rename_args
*ap
)
1710 struct vnode
*fvp
= ap
->a_fvp
;
1711 struct vnode
*tvp
= ap
->a_tvp
;
1712 struct vnode
*fdvp
= ap
->a_fdvp
;
1713 struct vnode
*tdvp
= ap
->a_tdvp
;
1714 struct componentname
*tcnp
= ap
->a_tcnp
;
1715 struct componentname
*fcnp
= ap
->a_fcnp
;
1718 /* Check for cross-device rename */
1719 if ((fvp
->v_mount
!= tdvp
->v_mount
) ||
1720 (tvp
&& (fvp
->v_mount
!= tvp
->v_mount
))) {
1726 * We shouldn't have to flush fvp on rename as the file handle should
1727 * not change, but the default is to do so.
1729 * We must flush tvp on rename because it might become stale on the
1730 * server after the rename.
1732 if (nfs_flush_on_rename
)
1733 VOP_FSYNC(fvp
, MNT_WAIT
);
1735 VOP_FSYNC(tvp
, MNT_WAIT
);
1738 * If the tvp exists and is in use, sillyrename it before doing the
1739 * rename of the new file over it.
1741 * XXX Can't sillyrename a directory.
1743 * We do not attempt to do any namecache purges in this old API
1744 * routine. The new API compat functions have access to the actual
1745 * namecache structures and will do it for us.
1747 if (tvp
&& tvp
->v_sysref
.refcnt
> 1 && !VTONFS(tvp
)->n_sillyrename
&&
1748 tvp
->v_type
!= VDIR
&& !nfs_sillyrename(tdvp
, tvp
, tcnp
)) {
1755 error
= nfs_renamerpc(fdvp
, fcnp
->cn_nameptr
, fcnp
->cn_namelen
,
1756 tdvp
, tcnp
->cn_nameptr
, tcnp
->cn_namelen
, tcnp
->cn_cred
,
1769 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1771 if (error
== ENOENT
)
1777 * nfs file rename rpc called from nfs_remove() above
1780 nfs_renameit(struct vnode
*sdvp
, struct componentname
*scnp
,
1781 struct sillyrename
*sp
)
1783 return (nfs_renamerpc(sdvp
, scnp
->cn_nameptr
, scnp
->cn_namelen
,
1784 sdvp
, sp
->s_name
, sp
->s_namlen
, scnp
->cn_cred
, scnp
->cn_td
));
1788 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1791 nfs_renamerpc(struct vnode
*fdvp
, const char *fnameptr
, int fnamelen
,
1792 struct vnode
*tdvp
, const char *tnameptr
, int tnamelen
,
1793 struct ucred
*cred
, struct thread
*td
)
1798 caddr_t bpos
, dpos
, cp2
;
1799 int error
= 0, fwccflag
= NFSV3_WCCRATTR
, twccflag
= NFSV3_WCCRATTR
;
1800 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
1801 int v3
= NFS_ISV3(fdvp
);
1803 nfsstats
.rpccnt
[NFSPROC_RENAME
]++;
1804 nfsm_reqhead(fdvp
, NFSPROC_RENAME
,
1805 (NFSX_FH(v3
) + NFSX_UNSIGNED
)*2 + nfsm_rndup(fnamelen
) +
1806 nfsm_rndup(tnamelen
));
1807 nfsm_fhtom(fdvp
, v3
);
1808 nfsm_strtom(fnameptr
, fnamelen
, NFS_MAXNAMLEN
);
1809 nfsm_fhtom(tdvp
, v3
);
1810 nfsm_strtom(tnameptr
, tnamelen
, NFS_MAXNAMLEN
);
1811 nfsm_request(fdvp
, NFSPROC_RENAME
, td
, cred
);
1813 nfsm_wcc_data(fdvp
, fwccflag
);
1814 nfsm_wcc_data(tdvp
, twccflag
);
1818 VTONFS(fdvp
)->n_flag
|= NLMODIFIED
;
1819 VTONFS(tdvp
)->n_flag
|= NLMODIFIED
;
1821 VTONFS(fdvp
)->n_attrstamp
= 0;
1823 VTONFS(tdvp
)->n_attrstamp
= 0;
1828 * nfs hard link create call
1830 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1831 * struct componentname *a_cnp)
1834 nfs_link(struct vop_old_link_args
*ap
)
1836 struct vnode
*vp
= ap
->a_vp
;
1837 struct vnode
*tdvp
= ap
->a_tdvp
;
1838 struct componentname
*cnp
= ap
->a_cnp
;
1842 caddr_t bpos
, dpos
, cp2
;
1843 int error
= 0, wccflag
= NFSV3_WCCRATTR
, attrflag
= 0;
1844 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
1847 if (vp
->v_mount
!= tdvp
->v_mount
) {
1852 * Push all writes to the server, so that the attribute cache
1853 * doesn't get "out of sync" with the server.
1854 * XXX There should be a better way!
1856 VOP_FSYNC(vp
, MNT_WAIT
);
1859 nfsstats
.rpccnt
[NFSPROC_LINK
]++;
1860 nfsm_reqhead(vp
, NFSPROC_LINK
,
1861 NFSX_FH(v3
)*2 + NFSX_UNSIGNED
+ nfsm_rndup(cnp
->cn_namelen
));
1863 nfsm_fhtom(tdvp
, v3
);
1864 nfsm_strtom(cnp
->cn_nameptr
, cnp
->cn_namelen
, NFS_MAXNAMLEN
);
1865 nfsm_request(vp
, NFSPROC_LINK
, cnp
->cn_td
, cnp
->cn_cred
);
1867 nfsm_postop_attr(vp
, attrflag
, NFS_LATTR_NOSHRINK
);
1868 nfsm_wcc_data(tdvp
, wccflag
);
1872 VTONFS(tdvp
)->n_flag
|= NLMODIFIED
;
1874 VTONFS(vp
)->n_attrstamp
= 0;
1876 VTONFS(tdvp
)->n_attrstamp
= 0;
1878 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1880 if (error
== EEXIST
)
1886 * nfs symbolic link create call
1888 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1889 * struct componentname *a_cnp, struct vattr *a_vap,
1893 nfs_symlink(struct vop_old_symlink_args
*ap
)
1895 struct vnode
*dvp
= ap
->a_dvp
;
1896 struct vattr
*vap
= ap
->a_vap
;
1897 struct componentname
*cnp
= ap
->a_cnp
;
1898 struct nfsv2_sattr
*sp
;
1902 caddr_t bpos
, dpos
, cp2
;
1903 int slen
, error
= 0, wccflag
= NFSV3_WCCRATTR
, gotvp
;
1904 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
1905 struct vnode
*newvp
= (struct vnode
*)0;
1906 int v3
= NFS_ISV3(dvp
);
1908 nfsstats
.rpccnt
[NFSPROC_SYMLINK
]++;
1909 slen
= strlen(ap
->a_target
);
1910 nfsm_reqhead(dvp
, NFSPROC_SYMLINK
, NFSX_FH(v3
) + 2*NFSX_UNSIGNED
+
1911 nfsm_rndup(cnp
->cn_namelen
) + nfsm_rndup(slen
) + NFSX_SATTR(v3
));
1912 nfsm_fhtom(dvp
, v3
);
1913 nfsm_strtom(cnp
->cn_nameptr
, cnp
->cn_namelen
, NFS_MAXNAMLEN
);
1915 nfsm_v3attrbuild(vap
, FALSE
);
1917 nfsm_strtom(ap
->a_target
, slen
, NFS_MAXPATHLEN
);
1919 nfsm_build(sp
, struct nfsv2_sattr
*, NFSX_V2SATTR
);
1920 sp
->sa_mode
= vtonfsv2_mode(VLNK
, vap
->va_mode
);
1921 sp
->sa_uid
= nfs_xdrneg1
;
1922 sp
->sa_gid
= nfs_xdrneg1
;
1923 sp
->sa_size
= nfs_xdrneg1
;
1924 txdr_nfsv2time(&vap
->va_atime
, &sp
->sa_atime
);
1925 txdr_nfsv2time(&vap
->va_mtime
, &sp
->sa_mtime
);
1929 * Issue the NFS request and get the rpc response.
1931 * Only NFSv3 responses returning an error of 0 actually return
1932 * a file handle that can be converted into newvp without having
1933 * to do an extra lookup rpc.
1935 nfsm_request(dvp
, NFSPROC_SYMLINK
, cnp
->cn_td
, cnp
->cn_cred
);
1938 nfsm_mtofh(dvp
, newvp
, v3
, gotvp
);
1939 nfsm_wcc_data(dvp
, wccflag
);
1943 * out code jumps -> here, mrep is also freed.
1950 * If we get an EEXIST error, silently convert it to no-error
1951 * in case of an NFS retry.
1953 if (error
== EEXIST
)
1957 * If we do not have (or no longer have) an error, and we could
1958 * not extract the newvp from the response due to the request being
1959 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1960 * to obtain a newvp to return.
1962 if (error
== 0 && newvp
== NULL
) {
1963 struct nfsnode
*np
= NULL
;
1965 error
= nfs_lookitup(dvp
, cnp
->cn_nameptr
, cnp
->cn_namelen
,
1966 cnp
->cn_cred
, cnp
->cn_td
, &np
);
1976 VTONFS(dvp
)->n_flag
|= NLMODIFIED
;
1978 VTONFS(dvp
)->n_attrstamp
= 0;
1985 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1986 * struct componentname *a_cnp, struct vattr *a_vap)
1989 nfs_mkdir(struct vop_old_mkdir_args
*ap
)
1991 struct vnode
*dvp
= ap
->a_dvp
;
1992 struct vattr
*vap
= ap
->a_vap
;
1993 struct componentname
*cnp
= ap
->a_cnp
;
1994 struct nfsv2_sattr
*sp
;
1999 struct nfsnode
*np
= (struct nfsnode
*)0;
2000 struct vnode
*newvp
= (struct vnode
*)0;
2001 caddr_t bpos
, dpos
, cp2
;
2002 int error
= 0, wccflag
= NFSV3_WCCRATTR
;
2004 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
2006 int v3
= NFS_ISV3(dvp
);
2008 if ((error
= VOP_GETATTR(dvp
, &vattr
)) != 0) {
2011 len
= cnp
->cn_namelen
;
2012 nfsstats
.rpccnt
[NFSPROC_MKDIR
]++;
2013 nfsm_reqhead(dvp
, NFSPROC_MKDIR
,
2014 NFSX_FH(v3
) + NFSX_UNSIGNED
+ nfsm_rndup(len
) + NFSX_SATTR(v3
));
2015 nfsm_fhtom(dvp
, v3
);
2016 nfsm_strtom(cnp
->cn_nameptr
, len
, NFS_MAXNAMLEN
);
2018 nfsm_v3attrbuild(vap
, FALSE
);
2020 nfsm_build(sp
, struct nfsv2_sattr
*, NFSX_V2SATTR
);
2021 sp
->sa_mode
= vtonfsv2_mode(VDIR
, vap
->va_mode
);
2022 sp
->sa_uid
= nfs_xdrneg1
;
2023 sp
->sa_gid
= nfs_xdrneg1
;
2024 sp
->sa_size
= nfs_xdrneg1
;
2025 txdr_nfsv2time(&vap
->va_atime
, &sp
->sa_atime
);
2026 txdr_nfsv2time(&vap
->va_mtime
, &sp
->sa_mtime
);
2028 nfsm_request(dvp
, NFSPROC_MKDIR
, cnp
->cn_td
, cnp
->cn_cred
);
2030 nfsm_mtofh(dvp
, newvp
, v3
, gotvp
);
2032 nfsm_wcc_data(dvp
, wccflag
);
2035 VTONFS(dvp
)->n_flag
|= NLMODIFIED
;
2037 VTONFS(dvp
)->n_attrstamp
= 0;
2039 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2040 * if we can succeed in looking up the directory.
2042 if (error
== EEXIST
|| (!error
&& !gotvp
)) {
2045 newvp
= (struct vnode
*)0;
2047 error
= nfs_lookitup(dvp
, cnp
->cn_nameptr
, len
, cnp
->cn_cred
,
2051 if (newvp
->v_type
!= VDIR
)
2064 * nfs remove directory call
2066 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2067 * struct componentname *a_cnp)
2070 nfs_rmdir(struct vop_old_rmdir_args
*ap
)
2072 struct vnode
*vp
= ap
->a_vp
;
2073 struct vnode
*dvp
= ap
->a_dvp
;
2074 struct componentname
*cnp
= ap
->a_cnp
;
2078 caddr_t bpos
, dpos
, cp2
;
2079 int error
= 0, wccflag
= NFSV3_WCCRATTR
;
2080 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
2081 int v3
= NFS_ISV3(dvp
);
2085 nfsstats
.rpccnt
[NFSPROC_RMDIR
]++;
2086 nfsm_reqhead(dvp
, NFSPROC_RMDIR
,
2087 NFSX_FH(v3
) + NFSX_UNSIGNED
+ nfsm_rndup(cnp
->cn_namelen
));
2088 nfsm_fhtom(dvp
, v3
);
2089 nfsm_strtom(cnp
->cn_nameptr
, cnp
->cn_namelen
, NFS_MAXNAMLEN
);
2090 nfsm_request(dvp
, NFSPROC_RMDIR
, cnp
->cn_td
, cnp
->cn_cred
);
2092 nfsm_wcc_data(dvp
, wccflag
);
2095 VTONFS(dvp
)->n_flag
|= NLMODIFIED
;
2097 VTONFS(dvp
)->n_attrstamp
= 0;
2099 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2101 if (error
== ENOENT
)
2109 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2112 nfs_readdir(struct vop_readdir_args
*ap
)
2114 struct vnode
*vp
= ap
->a_vp
;
2115 struct nfsnode
*np
= VTONFS(vp
);
2116 struct uio
*uio
= ap
->a_uio
;
2120 if (vp
->v_type
!= VDIR
)
2123 if ((error
= vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
)) != 0)
2127 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2128 * and then check that is still valid, or if this is an NQNFS mount
2129 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2130 * VOP_GETATTR() does not necessarily go to the wire.
2132 if (np
->n_direofoffset
> 0 && uio
->uio_offset
>= np
->n_direofoffset
&&
2133 (np
->n_flag
& (NLMODIFIED
|NRMODIFIED
)) == 0) {
2134 if (VOP_GETATTR(vp
, &vattr
) == 0 &&
2135 (np
->n_flag
& (NLMODIFIED
|NRMODIFIED
)) == 0
2137 nfsstats
.direofcache_hits
++;
2143 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2144 * own cache coherency checks so we do not have to.
2146 tresid
= uio
->uio_resid
;
2147 error
= nfs_bioread(vp
, uio
, 0);
2149 if (!error
&& uio
->uio_resid
== tresid
)
2150 nfsstats
.direofcache_misses
++;
2157 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2159 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2160 * offset/block and converts the nfs formatted directory entries for userland
2161 * consumption as well as deals with offsets into the middle of blocks.
2162 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2163 * be block-bounded. It must convert to cookies for the actual RPC.
2166 nfs_readdirrpc(struct vnode
*vp
, struct uio
*uiop
)
2169 struct nfs_dirent
*dp
= NULL
;
2174 caddr_t bpos
, dpos
, cp2
;
2175 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
2177 struct nfsmount
*nmp
= VFSTONFS(vp
->v_mount
);
2178 struct nfsnode
*dnp
= VTONFS(vp
);
2180 int error
= 0, tlen
, more_dirs
= 1, blksiz
= 0, bigenough
= 1;
2182 int v3
= NFS_ISV3(vp
);
2185 if (uiop
->uio_iovcnt
!= 1 || (uiop
->uio_offset
& (DIRBLKSIZ
- 1)) ||
2186 (uiop
->uio_resid
& (DIRBLKSIZ
- 1)))
2187 panic("nfs readdirrpc bad uio");
2191 * If there is no cookie, assume directory was stale.
2193 cookiep
= nfs_getcookie(dnp
, uiop
->uio_offset
, 0);
2197 return (NFSERR_BAD_COOKIE
);
2199 * Loop around doing readdir rpc's of size nm_readdirsize
2200 * truncated to a multiple of DIRBLKSIZ.
2201 * The stopping criteria is EOF or buffer full.
2203 while (more_dirs
&& bigenough
) {
2204 nfsstats
.rpccnt
[NFSPROC_READDIR
]++;
2205 nfsm_reqhead(vp
, NFSPROC_READDIR
, NFSX_FH(v3
) +
2209 nfsm_build(tl
, u_int32_t
*, 5 * NFSX_UNSIGNED
);
2210 *tl
++ = cookie
.nfsuquad
[0];
2211 *tl
++ = cookie
.nfsuquad
[1];
2212 *tl
++ = dnp
->n_cookieverf
.nfsuquad
[0];
2213 *tl
++ = dnp
->n_cookieverf
.nfsuquad
[1];
2215 nfsm_build(tl
, u_int32_t
*, 2 * NFSX_UNSIGNED
);
2216 *tl
++ = cookie
.nfsuquad
[0];
2218 *tl
= txdr_unsigned(nmp
->nm_readdirsize
);
2219 nfsm_request(vp
, NFSPROC_READDIR
, uiop
->uio_td
, nfs_vpcred(vp
, ND_READ
));
2221 nfsm_postop_attr(vp
, attrflag
, NFS_LATTR_NOSHRINK
);
2223 nfsm_dissect(tl
, u_int32_t
*,
2225 dnp
->n_cookieverf
.nfsuquad
[0] = *tl
++;
2226 dnp
->n_cookieverf
.nfsuquad
[1] = *tl
;
2232 nfsm_dissect(tl
, u_int32_t
*, NFSX_UNSIGNED
);
2233 more_dirs
= fxdr_unsigned(int, *tl
);
2235 /* loop thru the dir entries, converting them to std form */
2236 while (more_dirs
&& bigenough
) {
2238 nfsm_dissect(tl
, u_int32_t
*,
2240 fileno
= fxdr_hyper(tl
);
2241 len
= fxdr_unsigned(int, *(tl
+ 2));
2243 nfsm_dissect(tl
, u_int32_t
*,
2245 fileno
= fxdr_unsigned(u_quad_t
, *tl
++);
2246 len
= fxdr_unsigned(int, *tl
);
2248 if (len
<= 0 || len
> NFS_MAXNAMLEN
) {
2255 * len is the number of bytes in the path element
2256 * name, not including the \0 termination.
2258 * tlen is the number of bytes w have to reserve for
2259 * the path element name.
2261 tlen
= nfsm_rndup(len
);
2263 tlen
+= 4; /* To ensure null termination */
2266 * If the entry would cross a DIRBLKSIZ boundary,
2267 * extend the previous nfs_dirent to cover the
2270 left
= DIRBLKSIZ
- blksiz
;
2271 if ((tlen
+ sizeof(struct nfs_dirent
)) > left
) {
2272 dp
->nfs_reclen
+= left
;
2273 uiop
->uio_iov
->iov_base
+= left
;
2274 uiop
->uio_iov
->iov_len
-= left
;
2275 uiop
->uio_offset
+= left
;
2276 uiop
->uio_resid
-= left
;
2279 if ((tlen
+ sizeof(struct nfs_dirent
)) > uiop
->uio_resid
)
2282 dp
= (struct nfs_dirent
*)uiop
->uio_iov
->iov_base
;
2283 dp
->nfs_ino
= fileno
;
2284 dp
->nfs_namlen
= len
;
2285 dp
->nfs_reclen
= tlen
+ sizeof(struct nfs_dirent
);
2286 dp
->nfs_type
= DT_UNKNOWN
;
2287 blksiz
+= dp
->nfs_reclen
;
2288 if (blksiz
== DIRBLKSIZ
)
2290 uiop
->uio_offset
+= sizeof(struct nfs_dirent
);
2291 uiop
->uio_resid
-= sizeof(struct nfs_dirent
);
2292 uiop
->uio_iov
->iov_base
+= sizeof(struct nfs_dirent
);
2293 uiop
->uio_iov
->iov_len
-= sizeof(struct nfs_dirent
);
2294 nfsm_mtouio(uiop
, len
);
2297 * The uiop has advanced by nfs_dirent + len
2298 * but really needs to advance by
2301 cp
= uiop
->uio_iov
->iov_base
;
2303 *cp
= '\0'; /* null terminate */
2304 uiop
->uio_iov
->iov_base
+= tlen
;
2305 uiop
->uio_iov
->iov_len
-= tlen
;
2306 uiop
->uio_offset
+= tlen
;
2307 uiop
->uio_resid
-= tlen
;
2310 * NFS strings must be rounded up (nfsm_myouio
2311 * handled that in the bigenough case).
2313 nfsm_adv(nfsm_rndup(len
));
2316 nfsm_dissect(tl
, u_int32_t
*,
2319 nfsm_dissect(tl
, u_int32_t
*,
2324 * If we were able to accomodate the last entry,
2325 * get the cookie for the next one. Otherwise
2326 * hold-over the cookie for the one we were not
2327 * able to accomodate.
2330 cookie
.nfsuquad
[0] = *tl
++;
2332 cookie
.nfsuquad
[1] = *tl
++;
2338 more_dirs
= fxdr_unsigned(int, *tl
);
2341 * If at end of rpc data, get the eof boolean
2344 nfsm_dissect(tl
, u_int32_t
*, NFSX_UNSIGNED
);
2345 more_dirs
= (fxdr_unsigned(int, *tl
) == 0);
2350 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2351 * by increasing d_reclen for the last record.
2354 left
= DIRBLKSIZ
- blksiz
;
2355 dp
->nfs_reclen
+= left
;
2356 uiop
->uio_iov
->iov_base
+= left
;
2357 uiop
->uio_iov
->iov_len
-= left
;
2358 uiop
->uio_offset
+= left
;
2359 uiop
->uio_resid
-= left
;
2364 * We hit the end of the directory, update direofoffset.
2366 dnp
->n_direofoffset
= uiop
->uio_offset
;
2369 * There is more to go, insert the link cookie so the
2370 * next block can be read.
2372 if (uiop
->uio_resid
> 0)
2373 kprintf("EEK! readdirrpc resid > 0\n");
2374 cookiep
= nfs_getcookie(dnp
, uiop
->uio_offset
, 1);
2382 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2385 nfs_readdirplusrpc(struct vnode
*vp
, struct uio
*uiop
)
2388 struct nfs_dirent
*dp
;
2392 struct vnode
*newvp
;
2394 caddr_t bpos
, dpos
, cp2
, dpossav1
, dpossav2
;
2395 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
, *mdsav1
, *mdsav2
;
2397 struct nfsmount
*nmp
= VFSTONFS(vp
->v_mount
);
2398 struct nfsnode
*dnp
= VTONFS(vp
), *np
;
2401 int error
= 0, tlen
, more_dirs
= 1, blksiz
= 0, doit
, bigenough
= 1, i
;
2402 int attrflag
, fhsize
;
2403 struct nchandle nch
;
2404 struct nchandle dnch
;
2405 struct nlcomponent nlc
;
2411 if (uiop
->uio_iovcnt
!= 1 || (uiop
->uio_offset
& (DIRBLKSIZ
- 1)) ||
2412 (uiop
->uio_resid
& (DIRBLKSIZ
- 1)))
2413 panic("nfs readdirplusrpc bad uio");
2416 * Obtain the namecache record for the directory so we have something
2417 * to use as a basis for creating the entries. This function will
2418 * return a held (but not locked) ncp. The ncp may be disconnected
2419 * from the tree and cannot be used for upward traversals, and the
2420 * ncp may be unnamed. Note that other unrelated operations may
2421 * cause the ncp to be named at any time.
2423 cache_fromdvp(vp
, NULL
, 0, &dnch
);
2424 bzero(&nlc
, sizeof(nlc
));
2428 * If there is no cookie, assume directory was stale.
2430 cookiep
= nfs_getcookie(dnp
, uiop
->uio_offset
, 0);
2434 return (NFSERR_BAD_COOKIE
);
2436 * Loop around doing readdir rpc's of size nm_readdirsize
2437 * truncated to a multiple of DIRBLKSIZ.
2438 * The stopping criteria is EOF or buffer full.
2440 while (more_dirs
&& bigenough
) {
2441 nfsstats
.rpccnt
[NFSPROC_READDIRPLUS
]++;
2442 nfsm_reqhead(vp
, NFSPROC_READDIRPLUS
,
2443 NFSX_FH(1) + 6 * NFSX_UNSIGNED
);
2445 nfsm_build(tl
, u_int32_t
*, 6 * NFSX_UNSIGNED
);
2446 *tl
++ = cookie
.nfsuquad
[0];
2447 *tl
++ = cookie
.nfsuquad
[1];
2448 *tl
++ = dnp
->n_cookieverf
.nfsuquad
[0];
2449 *tl
++ = dnp
->n_cookieverf
.nfsuquad
[1];
2450 *tl
++ = txdr_unsigned(nmp
->nm_readdirsize
);
2451 *tl
= txdr_unsigned(nmp
->nm_rsize
);
2452 nfsm_request(vp
, NFSPROC_READDIRPLUS
, uiop
->uio_td
, nfs_vpcred(vp
, ND_READ
));
2453 nfsm_postop_attr(vp
, attrflag
, NFS_LATTR_NOSHRINK
);
2458 nfsm_dissect(tl
, u_int32_t
*, 3 * NFSX_UNSIGNED
);
2459 dnp
->n_cookieverf
.nfsuquad
[0] = *tl
++;
2460 dnp
->n_cookieverf
.nfsuquad
[1] = *tl
++;
2461 more_dirs
= fxdr_unsigned(int, *tl
);
2463 /* loop thru the dir entries, doctoring them to 4bsd form */
2464 while (more_dirs
&& bigenough
) {
2465 nfsm_dissect(tl
, u_int32_t
*, 3 * NFSX_UNSIGNED
);
2466 fileno
= fxdr_hyper(tl
);
2467 len
= fxdr_unsigned(int, *(tl
+ 2));
2468 if (len
<= 0 || len
> NFS_MAXNAMLEN
) {
2473 tlen
= nfsm_rndup(len
);
2475 tlen
+= 4; /* To ensure null termination*/
2476 left
= DIRBLKSIZ
- blksiz
;
2477 if ((tlen
+ sizeof(struct nfs_dirent
)) > left
) {
2478 dp
->nfs_reclen
+= left
;
2479 uiop
->uio_iov
->iov_base
+= left
;
2480 uiop
->uio_iov
->iov_len
-= left
;
2481 uiop
->uio_offset
+= left
;
2482 uiop
->uio_resid
-= left
;
2485 if ((tlen
+ sizeof(struct nfs_dirent
)) > uiop
->uio_resid
)
2488 dp
= (struct nfs_dirent
*)uiop
->uio_iov
->iov_base
;
2489 dp
->nfs_ino
= fileno
;
2490 dp
->nfs_namlen
= len
;
2491 dp
->nfs_reclen
= tlen
+ sizeof(struct nfs_dirent
);
2492 dp
->nfs_type
= DT_UNKNOWN
;
2493 blksiz
+= dp
->nfs_reclen
;
2494 if (blksiz
== DIRBLKSIZ
)
2496 uiop
->uio_offset
+= sizeof(struct nfs_dirent
);
2497 uiop
->uio_resid
-= sizeof(struct nfs_dirent
);
2498 uiop
->uio_iov
->iov_base
+= sizeof(struct nfs_dirent
);
2499 uiop
->uio_iov
->iov_len
-= sizeof(struct nfs_dirent
);
2500 nlc
.nlc_nameptr
= uiop
->uio_iov
->iov_base
;
2501 nlc
.nlc_namelen
= len
;
2502 nfsm_mtouio(uiop
, len
);
2503 cp
= uiop
->uio_iov
->iov_base
;
2506 uiop
->uio_iov
->iov_base
+= tlen
;
2507 uiop
->uio_iov
->iov_len
-= tlen
;
2508 uiop
->uio_offset
+= tlen
;
2509 uiop
->uio_resid
-= tlen
;
2511 nfsm_adv(nfsm_rndup(len
));
2512 nfsm_dissect(tl
, u_int32_t
*, 3 * NFSX_UNSIGNED
);
2514 cookie
.nfsuquad
[0] = *tl
++;
2515 cookie
.nfsuquad
[1] = *tl
++;
2520 * Since the attributes are before the file handle
2521 * (sigh), we must skip over the attributes and then
2522 * come back and get them.
2524 attrflag
= fxdr_unsigned(int, *tl
);
2528 nfsm_adv(NFSX_V3FATTR
);
2529 nfsm_dissect(tl
, u_int32_t
*, NFSX_UNSIGNED
);
2530 doit
= fxdr_unsigned(int, *tl
);
2532 nfsm_getfh(fhp
, fhsize
, 1);
2533 if (NFS_CMPFH(dnp
, fhp
, fhsize
)) {
2538 error
= nfs_nget(vp
->v_mount
, fhp
,
2546 if (doit
&& bigenough
) {
2551 nfsm_loadattr(newvp
, (struct vattr
*)0);
2555 IFTODT(VTTOIF(np
->n_vattr
.va_type
));
2557 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2558 nlc
.nlc_namelen
, nlc
.nlc_namelen
,
2560 nch
= cache_nlookup(&dnch
, &nlc
);
2561 cache_setunresolved(&nch
);
2562 nfs_cache_setvp(&nch
, newvp
,
2563 nfspos_cache_timeout
);
2566 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER"
2568 nlc
.nlc_namelen
, nlc
.nlc_namelen
,
2573 /* Just skip over the file handle */
2574 nfsm_dissect(tl
, u_int32_t
*, NFSX_UNSIGNED
);
2575 i
= fxdr_unsigned(int, *tl
);
2576 nfsm_adv(nfsm_rndup(i
));
2578 if (newvp
!= NULLVP
) {
2585 nfsm_dissect(tl
, u_int32_t
*, NFSX_UNSIGNED
);
2586 more_dirs
= fxdr_unsigned(int, *tl
);
2589 * If at end of rpc data, get the eof boolean
2592 nfsm_dissect(tl
, u_int32_t
*, NFSX_UNSIGNED
);
2593 more_dirs
= (fxdr_unsigned(int, *tl
) == 0);
2598 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2599 * by increasing d_reclen for the last record.
2602 left
= DIRBLKSIZ
- blksiz
;
2603 dp
->nfs_reclen
+= left
;
2604 uiop
->uio_iov
->iov_base
+= left
;
2605 uiop
->uio_iov
->iov_len
-= left
;
2606 uiop
->uio_offset
+= left
;
2607 uiop
->uio_resid
-= left
;
2611 * We are now either at the end of the directory or have filled the
2615 dnp
->n_direofoffset
= uiop
->uio_offset
;
2617 if (uiop
->uio_resid
> 0)
2618 kprintf("EEK! readdirplusrpc resid > 0\n");
2619 cookiep
= nfs_getcookie(dnp
, uiop
->uio_offset
, 1);
2623 if (newvp
!= NULLVP
) {
2636 * Silly rename. To make the NFS filesystem that is stateless look a little
2637 * more like the "ufs" a remove of an active vnode is translated to a rename
2638 * to a funny looking filename that is removed by nfs_inactive on the
2639 * nfsnode. There is the potential for another process on a different client
2640 * to create the same funny name between the nfs_lookitup() fails and the
2641 * nfs_rename() completes, but...
2644 nfs_sillyrename(struct vnode
*dvp
, struct vnode
*vp
, struct componentname
*cnp
)
2646 struct sillyrename
*sp
;
2651 * We previously purged dvp instead of vp. I don't know why, it
2652 * completely destroys performance. We can't do it anyway with the
2653 * new VFS API since we would be breaking the namecache topology.
2655 cache_purge(vp
); /* XXX */
2658 if (vp
->v_type
== VDIR
)
2659 panic("nfs: sillyrename dir");
2661 MALLOC(sp
, struct sillyrename
*, sizeof (struct sillyrename
),
2662 M_NFSREQ
, M_WAITOK
);
2663 sp
->s_cred
= crdup(cnp
->cn_cred
);
2667 /* Fudge together a funny name */
2668 sp
->s_namlen
= ksprintf(sp
->s_name
, ".nfsA%08x4.4", (int)cnp
->cn_td
);
2670 /* Try lookitups until we get one that isn't there */
2671 while (nfs_lookitup(dvp
, sp
->s_name
, sp
->s_namlen
, sp
->s_cred
,
2672 cnp
->cn_td
, (struct nfsnode
**)0) == 0) {
2674 if (sp
->s_name
[4] > 'z') {
2679 error
= nfs_renameit(dvp
, cnp
, sp
);
2682 error
= nfs_lookitup(dvp
, sp
->s_name
, sp
->s_namlen
, sp
->s_cred
,
2684 np
->n_sillyrename
= sp
;
2689 kfree((caddr_t
)sp
, M_NFSREQ
);
2694 * Look up a file name and optionally either update the file handle or
2695 * allocate an nfsnode, depending on the value of npp.
2696 * npp == NULL --> just do the lookup
2697 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2699 * *npp != NULL --> update the file handle in the vnode
2702 nfs_lookitup(struct vnode
*dvp
, const char *name
, int len
, struct ucred
*cred
,
2703 struct thread
*td
, struct nfsnode
**npp
)
2708 struct vnode
*newvp
= (struct vnode
*)0;
2709 struct nfsnode
*np
, *dnp
= VTONFS(dvp
);
2710 caddr_t bpos
, dpos
, cp2
;
2711 int error
= 0, fhlen
, attrflag
;
2712 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
2714 int v3
= NFS_ISV3(dvp
);
2716 nfsstats
.rpccnt
[NFSPROC_LOOKUP
]++;
2717 nfsm_reqhead(dvp
, NFSPROC_LOOKUP
,
2718 NFSX_FH(v3
) + NFSX_UNSIGNED
+ nfsm_rndup(len
));
2719 nfsm_fhtom(dvp
, v3
);
2720 nfsm_strtom(name
, len
, NFS_MAXNAMLEN
);
2721 nfsm_request(dvp
, NFSPROC_LOOKUP
, td
, cred
);
2722 if (npp
&& !error
) {
2723 nfsm_getfh(nfhp
, fhlen
, v3
);
2726 if (np
->n_fhsize
> NFS_SMALLFH
&& fhlen
<= NFS_SMALLFH
) {
2727 kfree((caddr_t
)np
->n_fhp
, M_NFSBIGFH
);
2728 np
->n_fhp
= &np
->n_fh
;
2729 } else if (np
->n_fhsize
<= NFS_SMALLFH
&& fhlen
>NFS_SMALLFH
)
2730 np
->n_fhp
=(nfsfh_t
*)kmalloc(fhlen
,M_NFSBIGFH
,M_WAITOK
);
2731 bcopy((caddr_t
)nfhp
, (caddr_t
)np
->n_fhp
, fhlen
);
2732 np
->n_fhsize
= fhlen
;
2734 } else if (NFS_CMPFH(dnp
, nfhp
, fhlen
)) {
2738 error
= nfs_nget(dvp
->v_mount
, nfhp
, fhlen
, &np
);
2746 nfsm_postop_attr(newvp
, attrflag
, NFS_LATTR_NOSHRINK
);
2747 if (!attrflag
&& *npp
== NULL
) {
2756 nfsm_loadattr(newvp
, (struct vattr
*)0);
2760 if (npp
&& *npp
== NULL
) {
2775 * Nfs Version 3 commit rpc
2778 nfs_commit(struct vnode
*vp
, u_quad_t offset
, int cnt
, struct thread
*td
)
2783 struct nfsmount
*nmp
= VFSTONFS(vp
->v_mount
);
2784 caddr_t bpos
, dpos
, cp2
;
2785 int error
= 0, wccflag
= NFSV3_WCCRATTR
;
2786 struct mbuf
*mreq
, *mrep
, *md
, *mb
, *mb2
;
2788 if ((nmp
->nm_state
& NFSSTA_HASWRITEVERF
) == 0)
2790 nfsstats
.rpccnt
[NFSPROC_COMMIT
]++;
2791 nfsm_reqhead(vp
, NFSPROC_COMMIT
, NFSX_FH(1));
2793 nfsm_build(tl
, u_int32_t
*, 3 * NFSX_UNSIGNED
);
2794 txdr_hyper(offset
, tl
);
2796 *tl
= txdr_unsigned(cnt
);
2797 nfsm_request(vp
, NFSPROC_COMMIT
, td
, nfs_vpcred(vp
, ND_WRITE
));
2798 nfsm_wcc_data(vp
, wccflag
);
2800 nfsm_dissect(tl
, u_int32_t
*, NFSX_V3WRITEVERF
);
2801 if (bcmp((caddr_t
)nmp
->nm_verf
, (caddr_t
)tl
,
2802 NFSX_V3WRITEVERF
)) {
2803 bcopy((caddr_t
)tl
, (caddr_t
)nmp
->nm_verf
,
2805 error
= NFSERR_STALEWRITEVERF
;
2815 * - make nfs_bmap() essentially a no-op that does no translation
2816 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2817 * (Maybe I could use the process's page mapping, but I was concerned that
2818 * Kernel Write might not be enabled and also figured copyout() would do
2819 * a lot more work than bcopy() and also it currently happens in the
2820 * context of the swapper process (2).
2822 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
2823 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2826 nfs_bmap(struct vop_bmap_args
*ap
)
2828 if (ap
->a_doffsetp
!= NULL
)
2829 *ap
->a_doffsetp
= ap
->a_loffset
;
2830 if (ap
->a_runp
!= NULL
)
2832 if (ap
->a_runb
!= NULL
)
2840 * For async requests when nfsiod(s) are running, queue the request by
2841 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2845 nfs_strategy(struct vop_strategy_args
*ap
)
2847 struct bio
*bio
= ap
->a_bio
;
2849 struct buf
*bp
= bio
->bio_buf
;
2853 KASSERT(bp
->b_cmd
!= BUF_CMD_DONE
,
2854 ("nfs_strategy: buffer %p unexpectedly marked done", bp
));
2855 KASSERT(BUF_REFCNT(bp
) > 0,
2856 ("nfs_strategy: buffer %p not locked", bp
));
2858 if (bp
->b_flags
& B_ASYNC
)
2861 td
= curthread
; /* XXX */
2864 * We probably don't need to push an nbio any more since no
2865 * block conversion is required due to the use of 64 bit byte
2866 * offsets, but do it anyway.
2868 nbio
= push_bio(bio
);
2869 nbio
->bio_offset
= bio
->bio_offset
;
2872 * If the op is asynchronous and an i/o daemon is waiting
2873 * queue the request, wake it up and wait for completion
2874 * otherwise just do it ourselves.
2876 if ((bp
->b_flags
& B_ASYNC
) == 0 || nfs_asyncio(ap
->a_vp
, nbio
, td
))
2877 error
= nfs_doio(ap
->a_vp
, nbio
, td
);
2884 * NB Currently unsupported.
2886 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
2890 nfs_mmap(struct vop_mmap_args
*ap
)
2896 * fsync vnode op. Just call nfs_flush() with commit == 1.
2898 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
2902 nfs_fsync(struct vop_fsync_args
*ap
)
2904 return (nfs_flush(ap
->a_vp
, ap
->a_waitfor
, curthread
, 1));
2908 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2909 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2910 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2911 * set the buffer contains data that has already been written to the server
2912 * and which now needs a commit RPC.
2914 * If commit is 0 we only take one pass and only flush buffers containing new
2917 * If commit is 1 we take two passes, issuing a commit RPC in the second
2920 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
2921 * to completely flush all pending data.
2923 * Note that the RB_SCAN code properly handles the case where the
2924 * callback might block and directly or indirectly (another thread) cause
2925 * the RB tree to change.
2928 #ifndef NFS_COMMITBVECSIZ
2929 #define NFS_COMMITBVECSIZ 16
2932 struct nfs_flush_info
{
2933 enum { NFI_FLUSHNEW
, NFI_COMMIT
} mode
;
2940 struct buf
*bvary
[NFS_COMMITBVECSIZ
];
2946 static int nfs_flush_bp(struct buf
*bp
, void *data
);
2947 static int nfs_flush_docommit(struct nfs_flush_info
*info
, int error
);
2950 nfs_flush(struct vnode
*vp
, int waitfor
, struct thread
*td
, int commit
)
2952 struct nfsnode
*np
= VTONFS(vp
);
2953 struct nfsmount
*nmp
= VFSTONFS(vp
->v_mount
);
2954 struct nfs_flush_info info
;
2957 bzero(&info
, sizeof(info
));
2960 info
.waitfor
= waitfor
;
2961 info
.slpflag
= (nmp
->nm_flag
& NFSMNT_INT
) ? PCATCH
: 0;
2968 info
.mode
= NFI_FLUSHNEW
;
2969 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
2970 nfs_flush_bp
, &info
);
2973 * Take a second pass if committing and no error occured.
2974 * Clean up any left over collection (whether an error
2977 if (commit
&& error
== 0) {
2978 info
.mode
= NFI_COMMIT
;
2979 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
2980 nfs_flush_bp
, &info
);
2982 error
= nfs_flush_docommit(&info
, error
);
2986 * Wait for pending I/O to complete before checking whether
2987 * any further dirty buffers exist.
2989 while (waitfor
== MNT_WAIT
&& vp
->v_track_write
.bk_active
) {
2990 vp
->v_track_write
.bk_waitflag
= 1;
2991 error
= tsleep(&vp
->v_track_write
,
2992 info
.slpflag
, "nfsfsync", info
.slptimeo
);
2995 * We have to be able to break out if this
2996 * is an 'intr' mount.
2998 if (nfs_sigintr(nmp
, (struct nfsreq
*)0, td
)) {
3004 * Since we do not process pending signals,
3005 * once we get a PCATCH our tsleep() will no
3006 * longer sleep, switch to a fixed timeout
3009 if (info
.slpflag
== PCATCH
) {
3011 info
.slptimeo
= 2 * hz
;
3018 * Loop if we are flushing synchronous as well as committing,
3019 * and dirty buffers are still present. Otherwise we might livelock.
3021 } while (waitfor
== MNT_WAIT
&& commit
&&
3022 error
== 0 && !RB_EMPTY(&vp
->v_rbdirty_tree
));
3025 * The callbacks have to return a negative error to terminate the
3032 * Deal with any error collection
3034 if (np
->n_flag
& NWRITEERR
) {
3035 error
= np
->n_error
;
3036 np
->n_flag
&= ~NWRITEERR
;
3044 nfs_flush_bp(struct buf
*bp
, void *data
)
3046 struct nfs_flush_info
*info
= data
;
3051 switch(info
->mode
) {
3054 if (info
->loops
&& info
->waitfor
== MNT_WAIT
) {
3055 error
= BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
);
3057 int lkflags
= LK_EXCLUSIVE
| LK_SLEEPFAIL
;
3058 if (info
->slpflag
& PCATCH
)
3059 lkflags
|= LK_PCATCH
;
3060 error
= BUF_TIMELOCK(bp
, lkflags
, "nfsfsync",
3064 error
= BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
);
3067 KKASSERT(bp
->b_vp
== info
->vp
);
3069 if ((bp
->b_flags
& B_DELWRI
) == 0)
3070 panic("nfs_fsync: not dirty");
3071 if (bp
->b_flags
& B_NEEDCOMMIT
) {
3087 * Only process buffers in need of a commit which we can
3088 * immediately lock. This may prevent a buffer from being
3089 * committed, but the normal flush loop will block on the
3090 * same buffer so we shouldn't get into an endless loop.
3093 if ((bp
->b_flags
& (B_DELWRI
| B_NEEDCOMMIT
)) !=
3094 (B_DELWRI
| B_NEEDCOMMIT
) ||
3095 BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
) != 0) {
3100 KKASSERT(bp
->b_vp
== info
->vp
);
3104 * NOTE: storing the bp in the bvary[] basically sets
3105 * it up for a commit operation.
3107 * We must call vfs_busy_pages() now so the commit operation
3108 * is interlocked with user modifications to memory mapped
3111 * Note: to avoid loopback deadlocks, we do not
3112 * assign b_runningbufspace.
3114 bp
->b_cmd
= BUF_CMD_WRITE
;
3115 vfs_busy_pages(bp
->b_vp
, bp
);
3116 info
->bvary
[info
->bvsize
] = bp
;
3117 toff
= bp
->b_bio2
.bio_offset
+ bp
->b_dirtyoff
;
3118 if (info
->bvsize
== 0 || toff
< info
->beg_off
)
3119 info
->beg_off
= toff
;
3120 toff
+= (off_t
)(bp
->b_dirtyend
- bp
->b_dirtyoff
);
3121 if (info
->bvsize
== 0 || toff
> info
->end_off
)
3122 info
->end_off
= toff
;
3124 if (info
->bvsize
== NFS_COMMITBVECSIZ
) {
3125 error
= nfs_flush_docommit(info
, 0);
3126 KKASSERT(info
->bvsize
== 0);
3135 nfs_flush_docommit(struct nfs_flush_info
*info
, int error
)
3145 if (info
->bvsize
> 0) {
3147 * Commit data on the server, as required. Note that
3148 * nfs_commit will use the vnode's cred for the commit.
3149 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3151 bytes
= info
->end_off
- info
->beg_off
;
3152 if (bytes
> 0x40000000)
3157 retv
= nfs_commit(vp
, info
->beg_off
,
3158 (int)bytes
, info
->td
);
3159 if (retv
== NFSERR_STALEWRITEVERF
)
3160 nfs_clearcommit(vp
->v_mount
);
3164 * Now, either mark the blocks I/O done or mark the
3165 * blocks dirty, depending on whether the commit
3168 for (i
= 0; i
< info
->bvsize
; ++i
) {
3169 bp
= info
->bvary
[i
];
3170 bp
->b_flags
&= ~(B_NEEDCOMMIT
| B_CLUSTEROK
);
3173 * Error, leave B_DELWRI intact
3175 vfs_unbusy_pages(bp
);
3176 bp
->b_cmd
= BUF_CMD_DONE
;
3180 * Success, remove B_DELWRI ( bundirty() ).
3182 * b_dirtyoff/b_dirtyend seem to be NFS
3183 * specific. We should probably move that
3184 * into bundirty(). XXX
3186 * We are faking an I/O write, we have to
3187 * start the transaction in order to
3188 * immediately biodone() it.
3191 bp
->b_flags
|= B_ASYNC
;
3193 bp
->b_flags
&= ~B_ERROR
;
3194 bp
->b_dirtyoff
= bp
->b_dirtyend
= 0;
3196 biodone(&bp
->b_bio1
);
3205 * NFS advisory byte-level locks.
3206 * Currently unsupported.
3208 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3212 nfs_advlock(struct vop_advlock_args
*ap
)
3214 struct nfsnode
*np
= VTONFS(ap
->a_vp
);
3217 * The following kludge is to allow diskless support to work
3218 * until a real NFS lockd is implemented. Basically, just pretend
3219 * that this is a local lock.
3221 return (lf_advlock(ap
, &(np
->n_lockf
), np
->n_size
));
3225 * Print out the contents of an nfsnode.
3227 * nfs_print(struct vnode *a_vp)
3230 nfs_print(struct vop_print_args
*ap
)
3232 struct vnode
*vp
= ap
->a_vp
;
3233 struct nfsnode
*np
= VTONFS(vp
);
3235 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3236 np
->n_vattr
.va_fileid
, np
->n_vattr
.va_fsid
);
3237 if (vp
->v_type
== VFIFO
)
3244 * nfs special file access vnode op.
3245 * Essentially just get vattr and then imitate iaccess() since the device is
3246 * local to the client.
3248 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3251 nfsspec_access(struct vop_access_args
*ap
)
3255 struct ucred
*cred
= ap
->a_cred
;
3256 struct vnode
*vp
= ap
->a_vp
;
3257 mode_t mode
= ap
->a_mode
;
3263 * Disallow write attempts on filesystems mounted read-only;
3264 * unless the file is a socket, fifo, or a block or character
3265 * device resident on the filesystem.
3267 if ((mode
& VWRITE
) && (vp
->v_mount
->mnt_flag
& MNT_RDONLY
)) {
3268 switch (vp
->v_type
) {
3278 * If you're the super-user,
3279 * you always get access.
3281 if (cred
->cr_uid
== 0)
3284 error
= VOP_GETATTR(vp
, vap
);
3288 * Access check is based on only one of owner, group, public.
3289 * If not owner, then check group. If not a member of the
3290 * group, then check public access.
3292 if (cred
->cr_uid
!= vap
->va_uid
) {
3294 gp
= cred
->cr_groups
;
3295 for (i
= 0; i
< cred
->cr_ngroups
; i
++, gp
++)
3296 if (vap
->va_gid
== *gp
)
3302 error
= (vap
->va_mode
& mode
) == mode
? 0 : EACCES
;
3307 * Read wrapper for special devices.
3309 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3310 * struct ucred *a_cred)
3313 nfsspec_read(struct vop_read_args
*ap
)
3315 struct nfsnode
*np
= VTONFS(ap
->a_vp
);
3321 getnanotime(&np
->n_atim
);
3322 return (VOCALL(&spec_vnode_vops
, &ap
->a_head
));
3326 * Write wrapper for special devices.
3328 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3329 * struct ucred *a_cred)
3332 nfsspec_write(struct vop_write_args
*ap
)
3334 struct nfsnode
*np
= VTONFS(ap
->a_vp
);
3340 getnanotime(&np
->n_mtim
);
3341 return (VOCALL(&spec_vnode_vops
, &ap
->a_head
));
3345 * Close wrapper for special devices.
3347 * Update the times on the nfsnode then do device close.
3349 * nfsspec_close(struct vnode *a_vp, int a_fflag)
3352 nfsspec_close(struct vop_close_args
*ap
)
3354 struct vnode
*vp
= ap
->a_vp
;
3355 struct nfsnode
*np
= VTONFS(vp
);
3358 if (np
->n_flag
& (NACC
| NUPD
)) {
3360 if (vp
->v_sysref
.refcnt
== 1 &&
3361 (vp
->v_mount
->mnt_flag
& MNT_RDONLY
) == 0) {
3363 if (np
->n_flag
& NACC
)
3364 vattr
.va_atime
= np
->n_atim
;
3365 if (np
->n_flag
& NUPD
)
3366 vattr
.va_mtime
= np
->n_mtim
;
3367 (void)VOP_SETATTR(vp
, &vattr
, nfs_vpcred(vp
, ND_WRITE
));
3370 return (VOCALL(&spec_vnode_vops
, &ap
->a_head
));
3374 * Read wrapper for fifos.
3376 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3377 * struct ucred *a_cred)
3380 nfsfifo_read(struct vop_read_args
*ap
)
3382 struct nfsnode
*np
= VTONFS(ap
->a_vp
);
3388 getnanotime(&np
->n_atim
);
3389 return (VOCALL(&fifo_vnode_vops
, &ap
->a_head
));
3393 * Write wrapper for fifos.
3395 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3396 * struct ucred *a_cred)
3399 nfsfifo_write(struct vop_write_args
*ap
)
3401 struct nfsnode
*np
= VTONFS(ap
->a_vp
);
3407 getnanotime(&np
->n_mtim
);
3408 return (VOCALL(&fifo_vnode_vops
, &ap
->a_head
));
3412 * Close wrapper for fifos.
3414 * Update the times on the nfsnode then do fifo close.
3416 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3419 nfsfifo_close(struct vop_close_args
*ap
)
3421 struct vnode
*vp
= ap
->a_vp
;
3422 struct nfsnode
*np
= VTONFS(vp
);
3426 if (np
->n_flag
& (NACC
| NUPD
)) {
3428 if (np
->n_flag
& NACC
)
3430 if (np
->n_flag
& NUPD
)
3433 if (vp
->v_sysref
.refcnt
== 1 &&
3434 (vp
->v_mount
->mnt_flag
& MNT_RDONLY
) == 0) {
3436 if (np
->n_flag
& NACC
)
3437 vattr
.va_atime
= np
->n_atim
;
3438 if (np
->n_flag
& NUPD
)
3439 vattr
.va_mtime
= np
->n_mtim
;
3440 (void)VOP_SETATTR(vp
, &vattr
, nfs_vpcred(vp
, ND_WRITE
));
3443 return (VOCALL(&fifo_vnode_vops
, &ap
->a_head
));