?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
kernel NFS - Fix another deadlock in the readdirplus code
[dragonfly.git] / sys / vfs / nfs / nfs_vnops.c
blobdba0935db743e315857827e33d34117253e2e179
1 /*
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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.
23 *
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
34 * SUCH DAMAGE.
35 *
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 1.80 2008/10/18 01:13:54 dillon Exp $
39 */
40
41
42 /*
43 * vnode op calls for Sun NFS version 2 and 3
44 */
45
46 #include "opt_inet.h"
47
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/resourcevar.h>
52 #include <sys/proc.h>
53 #include <sys/mount.h>
54 #include <sys/buf.h>
55 #include <sys/malloc.h>
56 #include <sys/mbuf.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>
64 #include <sys/stat.h>
65 #include <sys/sysctl.h>
66 #include <sys/conf.h>
67
68 #include <vm/vm.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vm_zone.h>
71
72 #include <sys/buf2.h>
73
74 #include <vfs/fifofs/fifo.h>
75 #include <vfs/ufs/dir.h>
76
77 #undef DIRBLKSIZ
78
79 #include "rpcv2.h"
80 #include "nfsproto.h"
81 #include "nfs.h"
82 #include "nfsmount.h"
83 #include "nfsnode.h"
84 #include "xdr_subs.h"
85 #include "nfsm_subs.h"
86
87 #include <net/if.h>
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
90
91 #include <sys/thread2.h>
92
93 /* Defs */
94 #define TRUE 1
95 #define FALSE 0
96
97 static int nfsfifo_read (struct vop_read_args *);
98 static int nfsfifo_write (struct vop_write_args *);
99 static int nfsfifo_close (struct vop_close_args *);
100 #define nfs_poll vop_nopoll
101 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
102 static int nfs_lookup (struct vop_old_lookup_args *);
103 static int nfs_create (struct vop_old_create_args *);
104 static int nfs_mknod (struct vop_old_mknod_args *);
105 static int nfs_open (struct vop_open_args *);
106 static int nfs_close (struct vop_close_args *);
107 static int nfs_access (struct vop_access_args *);
108 static int nfs_getattr (struct vop_getattr_args *);
109 static int nfs_setattr (struct vop_setattr_args *);
110 static int nfs_read (struct vop_read_args *);
111 static int nfs_mmap (struct vop_mmap_args *);
112 static int nfs_fsync (struct vop_fsync_args *);
113 static int nfs_remove (struct vop_old_remove_args *);
114 static int nfs_link (struct vop_old_link_args *);
115 static int nfs_rename (struct vop_old_rename_args *);
116 static int nfs_mkdir (struct vop_old_mkdir_args *);
117 static int nfs_rmdir (struct vop_old_rmdir_args *);
118 static int nfs_symlink (struct vop_old_symlink_args *);
119 static int nfs_readdir (struct vop_readdir_args *);
120 static int nfs_bmap (struct vop_bmap_args *);
121 static int nfs_strategy (struct vop_strategy_args *);
122 static int nfs_lookitup (struct vnode *, const char *, int,
123 struct ucred *, struct thread *, struct nfsnode **);
124 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
125 static int nfs_laccess (struct vop_access_args *);
126 static int nfs_readlink (struct vop_readlink_args *);
127 static int nfs_print (struct vop_print_args *);
128 static int nfs_advlock (struct vop_advlock_args *);
129
130 static int nfs_nresolve (struct vop_nresolve_args *);
131 /*
132 * Global vfs data structures for nfs
133 */
134 struct vop_ops nfsv2_vnode_vops = {
135 .vop_default = vop_defaultop,
136 .vop_access = nfs_access,
137 .vop_advlock = nfs_advlock,
138 .vop_bmap = nfs_bmap,
139 .vop_close = nfs_close,
140 .vop_old_create = nfs_create,
141 .vop_fsync = nfs_fsync,
142 .vop_getattr = nfs_getattr,
143 .vop_getpages = vop_stdgetpages,
144 .vop_putpages = vop_stdputpages,
145 .vop_inactive = nfs_inactive,
146 .vop_old_link = nfs_link,
147 .vop_old_lookup = nfs_lookup,
148 .vop_old_mkdir = nfs_mkdir,
149 .vop_old_mknod = nfs_mknod,
150 .vop_mmap = nfs_mmap,
151 .vop_open = nfs_open,
152 .vop_poll = nfs_poll,
153 .vop_print = nfs_print,
154 .vop_read = nfs_read,
155 .vop_readdir = nfs_readdir,
156 .vop_readlink = nfs_readlink,
157 .vop_reclaim = nfs_reclaim,
158 .vop_old_remove = nfs_remove,
159 .vop_old_rename = nfs_rename,
160 .vop_old_rmdir = nfs_rmdir,
161 .vop_setattr = nfs_setattr,
162 .vop_strategy = nfs_strategy,
163 .vop_old_symlink = nfs_symlink,
164 .vop_write = nfs_write,
165 .vop_nresolve = nfs_nresolve
166 };
167
168 /*
169 * Special device vnode ops
170 */
171 struct vop_ops nfsv2_spec_vops = {
172 .vop_default = vop_defaultop,
173 .vop_access = nfs_laccess,
174 .vop_close = nfs_close,
175 .vop_fsync = nfs_fsync,
176 .vop_getattr = nfs_getattr,
177 .vop_inactive = nfs_inactive,
178 .vop_print = nfs_print,
179 .vop_read = vop_stdnoread,
180 .vop_reclaim = nfs_reclaim,
181 .vop_setattr = nfs_setattr,
182 .vop_write = vop_stdnowrite
183 };
184
185 struct vop_ops nfsv2_fifo_vops = {
186 .vop_default = fifo_vnoperate,
187 .vop_access = nfs_laccess,
188 .vop_close = nfsfifo_close,
189 .vop_fsync = nfs_fsync,
190 .vop_getattr = nfs_getattr,
191 .vop_inactive = nfs_inactive,
192 .vop_print = nfs_print,
193 .vop_read = nfsfifo_read,
194 .vop_reclaim = nfs_reclaim,
195 .vop_setattr = nfs_setattr,
196 .vop_write = nfsfifo_write
197 };
198
199 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
200 struct componentname *cnp,
201 struct vattr *vap);
202 static int nfs_removerpc (struct vnode *dvp, const char *name,
203 int namelen,
204 struct ucred *cred, struct thread *td);
205 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
206 int fnamelen, struct vnode *tdvp,
207 const char *tnameptr, int tnamelen,
208 struct ucred *cred, struct thread *td);
209 static int nfs_renameit (struct vnode *sdvp,
210 struct componentname *scnp,
211 struct sillyrename *sp);
212
213 SYSCTL_DECL(_vfs_nfs);
214
215 static int nfs_flush_on_rename = 1;
216 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
217 &nfs_flush_on_rename, 0, "flush fvp prior to rename");
218 static int nfs_flush_on_hlink = 0;
219 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
220 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
221
222 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
223 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
224 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
225
226 static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
227 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
228 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
229
230 static int nfspos_cache_timeout = NFS_MINATTRTIMO;
231 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
232 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
233
234 static int nfsv3_commit_on_close = 0;
235 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
236 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
237 #if 0
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
239 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
240
241 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
242 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
243 #endif
244
245 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
246 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
247 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
248
249 /*
250 * Returns whether a name component is a degenerate '.' or '..'.
251 */
252 static __inline
253 int
254 nlcdegenerate(struct nlcomponent *nlc)
255 {
256 if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.')
257 return(1);
258 if (nlc->nlc_namelen == 2 &&
259 nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.')
260 return(1);
261 return(0);
262 }
263
264 static int
265 nfs3_access_otw(struct vnode *vp, int wmode,
266 struct thread *td, struct ucred *cred)
267 {
268 struct nfsnode *np = VTONFS(vp);
269 int attrflag;
270 int error = 0;
271 u_int32_t *tl;
272 u_int32_t rmode;
273 struct nfsm_info info;
274
275 info.mrep = NULL;
276 info.v3 = 1;
277
278 nfsstats.rpccnt[NFSPROC_ACCESS]++;
279 nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
280 NFSX_FH(info.v3) + NFSX_UNSIGNED);
281 ERROROUT(nfsm_fhtom(&info, vp));
282 tl = nfsm_build(&info, NFSX_UNSIGNED);
283 *tl = txdr_unsigned(wmode);
284 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
285 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
286 if (error == 0) {
287 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
288 rmode = fxdr_unsigned(u_int32_t, *tl);
289 np->n_mode = rmode;
290 np->n_modeuid = cred->cr_uid;
291 np->n_modestamp = mycpu->gd_time_seconds;
292 }
293 m_freem(info.mrep);
294 info.mrep = NULL;
295 nfsmout:
296 return error;
297 }
298
299 /*
300 * nfs access vnode op.
301 * For nfs version 2, just return ok. File accesses may fail later.
302 * For nfs version 3, use the access rpc to check accessibility. If file modes
303 * are changed on the server, accesses might still fail later.
304 *
305 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
306 */
307 static int
308 nfs_access(struct vop_access_args *ap)
309 {
310 struct ucred *cred;
311 struct vnode *vp = ap->a_vp;
312 thread_t td = curthread;
313 int error = 0;
314 u_int32_t mode, wmode;
315 struct nfsnode *np = VTONFS(vp);
316 int v3 = NFS_ISV3(vp);
317
318 /*
319 * Disallow write attempts on filesystems mounted read-only;
320 * unless the file is a socket, fifo, or a block or character
321 * device resident on the filesystem.
322 */
323 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
324 switch (vp->v_type) {
325 case VREG:
326 case VDIR:
327 case VLNK:
328 return (EROFS);
329 default:
330 break;
331 }
332 }
333
334 /*
335 * The NFS protocol passes only the effective uid/gid over the wire but
336 * we need to check access against real ids if AT_EACCESS not set.
337 * Handle this case by cloning the credentials and setting the
338 * effective ids to the real ones.
339 */
340 if (ap->a_flags & AT_EACCESS) {
341 cred = crhold(ap->a_cred);
342 } else {
343 cred = crdup(ap->a_cred);
344 cred->cr_uid = cred->cr_ruid;
345 cred->cr_gid = cred->cr_rgid;
346 }
347
348 /*
349 * For nfs v3, check to see if we have done this recently, and if
350 * so return our cached result instead of making an ACCESS call.
351 * If not, do an access rpc, otherwise you are stuck emulating
352 * ufs_access() locally using the vattr. This may not be correct,
353 * since the server may apply other access criteria such as
354 * client uid-->server uid mapping that we do not know about.
355 */
356 if (v3) {
357 if (ap->a_mode & VREAD)
358 mode = NFSV3ACCESS_READ;
359 else
360 mode = 0;
361 if (vp->v_type != VDIR) {
362 if (ap->a_mode & VWRITE)
363 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
364 if (ap->a_mode & VEXEC)
365 mode |= NFSV3ACCESS_EXECUTE;
366 } else {
367 if (ap->a_mode & VWRITE)
368 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
369 NFSV3ACCESS_DELETE);
370 if (ap->a_mode & VEXEC)
371 mode |= NFSV3ACCESS_LOOKUP;
372 }
373 /* XXX safety belt, only make blanket request if caching */
374 if (nfsaccess_cache_timeout > 0) {
375 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
376 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
377 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
378 } else {
379 wmode = mode;
380 }
381
382 /*
383 * Does our cached result allow us to give a definite yes to
384 * this request?
385 */
386 if (np->n_modestamp &&
387 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
388 (cred->cr_uid == np->n_modeuid) &&
389 ((np->n_mode & mode) == mode)) {
390 nfsstats.accesscache_hits++;
391 } else {
392 /*
393 * Either a no, or a don't know. Go to the wire.
394 */
395 nfsstats.accesscache_misses++;
396 error = nfs3_access_otw(vp, wmode, td, cred);
397 if (!error) {
398 if ((np->n_mode & mode) != mode) {
399 error = EACCES;
400 }
401 }
402 }
403 } else {
404 if ((error = nfs_laccess(ap)) != 0) {
405 crfree(cred);
406 return (error);
407 }
408
409 /*
410 * Attempt to prevent a mapped root from accessing a file
411 * which it shouldn't. We try to read a byte from the file
412 * if the user is root and the file is not zero length.
413 * After calling nfs_laccess, we should have the correct
414 * file size cached.
415 */
416 if (cred->cr_uid == 0 && (ap->a_mode & VREAD)
417 && VTONFS(vp)->n_size > 0) {
418 struct iovec aiov;
419 struct uio auio;
420 char buf[1];
421
422 aiov.iov_base = buf;
423 aiov.iov_len = 1;
424 auio.uio_iov = &aiov;
425 auio.uio_iovcnt = 1;
426 auio.uio_offset = 0;
427 auio.uio_resid = 1;
428 auio.uio_segflg = UIO_SYSSPACE;
429 auio.uio_rw = UIO_READ;
430 auio.uio_td = td;
431
432 if (vp->v_type == VREG) {
433 error = nfs_readrpc_uio(vp, &auio);
434 } else if (vp->v_type == VDIR) {
435 char* bp;
436 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
437 aiov.iov_base = bp;
438 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
439 error = nfs_readdirrpc_uio(vp, &auio);
440 kfree(bp, M_TEMP);
441 } else if (vp->v_type == VLNK) {
442 error = nfs_readlinkrpc_uio(vp, &auio);
443 } else {
444 error = EACCES;
445 }
446 }
447 }
448 /*
449 * [re]record creds for reading and/or writing if access
450 * was granted. Assume the NFS server will grant read access
451 * for execute requests.
452 */
453 if (error == 0) {
454 if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) {
455 crhold(cred);
456 if (np->n_rucred)
457 crfree(np->n_rucred);
458 np->n_rucred = cred;
459 }
460 if ((ap->a_mode & VWRITE) && cred != np->n_wucred) {
461 crhold(cred);
462 if (np->n_wucred)
463 crfree(np->n_wucred);
464 np->n_wucred = cred;
465 }
466 }
467 crfree(cred);
468 return(error);
469 }
470
471 /*
472 * nfs open vnode op
473 * Check to see if the type is ok
474 * and that deletion is not in progress.
475 * For paged in text files, you will need to flush the page cache
476 * if consistency is lost.
477 *
478 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
479 * struct file *a_fp)
480 */
481 /* ARGSUSED */
482 static int
483 nfs_open(struct vop_open_args *ap)
484 {
485 struct vnode *vp = ap->a_vp;
486 struct nfsnode *np = VTONFS(vp);
487 struct vattr vattr;
488 int error;
489
490 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
491 #ifdef DIAGNOSTIC
492 kprintf("open eacces vtyp=%d\n",vp->v_type);
493 #endif
494 return (EOPNOTSUPP);
495 }
496
497 /*
498 * Save valid creds for reading and writing for later RPCs.
499 */
500 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
501 crhold(ap->a_cred);
502 if (np->n_rucred)
503 crfree(np->n_rucred);
504 np->n_rucred = ap->a_cred;
505 }
506 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
507 crhold(ap->a_cred);
508 if (np->n_wucred)
509 crfree(np->n_wucred);
510 np->n_wucred = ap->a_cred;
511 }
512
513 /*
514 * Clear the attribute cache only if opening with write access. It
515 * is unclear if we should do this at all here, but we certainly
516 * should not clear the cache unconditionally simply because a file
517 * is being opened.
518 */
519 if (ap->a_mode & FWRITE)
520 np->n_attrstamp = 0;
521
522 /*
523 * For normal NFS, reconcile changes made locally verses
524 * changes made remotely. Note that VOP_GETATTR only goes
525 * to the wire if the cached attribute has timed out or been
526 * cleared.
527 *
528 * If local modifications have been made clear the attribute
529 * cache to force an attribute and modified time check. If
530 * GETATTR detects that the file has been changed by someone
531 * other then us it will set NRMODIFIED.
532 *
533 * If we are opening a directory and local changes have been
534 * made we have to invalidate the cache in order to ensure
535 * that we get the most up-to-date information from the
536 * server. XXX
537 */
538 if (np->n_flag & NLMODIFIED) {
539 np->n_attrstamp = 0;
540 if (vp->v_type == VDIR) {
541 error = nfs_vinvalbuf(vp, V_SAVE, 1);
542 if (error == EINTR)
543 return (error);
544 nfs_invaldir(vp);
545 }
546 }
547 error = VOP_GETATTR(vp, &vattr);
548 if (error)
549 return (error);
550 if (np->n_flag & NRMODIFIED) {
551 if (vp->v_type == VDIR)
552 nfs_invaldir(vp);
553 error = nfs_vinvalbuf(vp, V_SAVE, 1);
554 if (error == EINTR)
555 return (error);
556 np->n_flag &= ~NRMODIFIED;
557 }
558
559 return (vop_stdopen(ap));
560 }
561
562 /*
563 * nfs close vnode op
564 * What an NFS client should do upon close after writing is a debatable issue.
565 * Most NFS clients push delayed writes to the server upon close, basically for
566 * two reasons:
567 * 1 - So that any write errors may be reported back to the client process
568 * doing the close system call. By far the two most likely errors are
569 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
570 * 2 - To put a worst case upper bound on cache inconsistency between
571 * multiple clients for the file.
572 * There is also a consistency problem for Version 2 of the protocol w.r.t.
573 * not being able to tell if other clients are writing a file concurrently,
574 * since there is no way of knowing if the changed modify time in the reply
575 * is only due to the write for this client.
576 * (NFS Version 3 provides weak cache consistency data in the reply that
577 * should be sufficient to detect and handle this case.)
578 *
579 * The current code does the following:
580 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
581 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
582 * or commit them (this satisfies 1 and 2 except for the
583 * case where the server crashes after this close but
584 * before the commit RPC, which is felt to be "good
585 * enough". Changing the last argument to nfs_flush() to
586 * a 1 would force a commit operation, if it is felt a
587 * commit is necessary now.
588 * for NQNFS - do nothing now, since 2 is dealt with via leases and
589 * 1 should be dealt with via an fsync() system call for
590 * cases where write errors are important.
591 *
592 * nfs_close(struct vnode *a_vp, int a_fflag)
593 */
594 /* ARGSUSED */
595 static int
596 nfs_close(struct vop_close_args *ap)
597 {
598 struct vnode *vp = ap->a_vp;
599 struct nfsnode *np = VTONFS(vp);
600 int error = 0;
601 thread_t td = curthread;
602
603 if (vp->v_type == VREG) {
604 if (np->n_flag & NLMODIFIED) {
605 if (NFS_ISV3(vp)) {
606 /*
607 * Under NFSv3 we have dirty buffers to dispose of. We
608 * must flush them to the NFS server. We have the option
609 * of waiting all the way through the commit rpc or just
610 * waiting for the initial write. The default is to only
611 * wait through the initial write so the data is in the
612 * server's cache, which is roughly similar to the state
613 * a standard disk subsystem leaves the file in on close().
614 *
615 * We cannot clear the NLMODIFIED bit in np->n_flag due to
616 * potential races with other processes, and certainly
617 * cannot clear it if we don't commit.
618 */
619 int cm = nfsv3_commit_on_close ? 1 : 0;
620 error = nfs_flush(vp, MNT_WAIT, td, cm);
621 /* np->n_flag &= ~NLMODIFIED; */
622 } else {
623 error = nfs_vinvalbuf(vp, V_SAVE, 1);
624 }
625 np->n_attrstamp = 0;
626 }
627 if (np->n_flag & NWRITEERR) {
628 np->n_flag &= ~NWRITEERR;
629 error = np->n_error;
630 }
631 }
632 vop_stdclose(ap);
633 return (error);
634 }
635
636 /*
637 * nfs getattr call from vfs.
638 *
639 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
640 */
641 static int
642 nfs_getattr(struct vop_getattr_args *ap)
643 {
644 struct vnode *vp = ap->a_vp;
645 struct nfsnode *np = VTONFS(vp);
646 int error = 0;
647 thread_t td = curthread;
648 struct nfsm_info info;
649
650 info.mrep = NULL;
651 info.v3 = NFS_ISV3(vp);
652
653 /*
654 * Update local times for special files.
655 */
656 if (np->n_flag & (NACC | NUPD))
657 np->n_flag |= NCHG;
658 /*
659 * First look in the cache.
660 */
661 if (nfs_getattrcache(vp, ap->a_vap) == 0)
662 return (0);
663
664 if (info.v3 && nfsaccess_cache_timeout > 0) {
665 nfsstats.accesscache_misses++;
666 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
667 if (nfs_getattrcache(vp, ap->a_vap) == 0)
668 return (0);
669 }
670
671 nfsstats.rpccnt[NFSPROC_GETATTR]++;
672 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
673 ERROROUT(nfsm_fhtom(&info, vp));
674 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
675 nfs_vpcred(vp, ND_CHECK), &error));
676 if (error == 0) {
677 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
678 }
679 m_freem(info.mrep);
680 info.mrep = NULL;
681 nfsmout:
682 return (error);
683 }
684
685 /*
686 * nfs setattr call.
687 *
688 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
689 */
690 static int
691 nfs_setattr(struct vop_setattr_args *ap)
692 {
693 struct vnode *vp = ap->a_vp;
694 struct nfsnode *np = VTONFS(vp);
695 struct vattr *vap = ap->a_vap;
696 int biosize = vp->v_mount->mnt_stat.f_iosize;
697 int error = 0;
698 int boff;
699 off_t tsize;
700 thread_t td = curthread;
701
702 #ifndef nolint
703 tsize = (off_t)0;
704 #endif
705
706 /*
707 * Setting of flags is not supported.
708 */
709 if (vap->va_flags != VNOVAL)
710 return (EOPNOTSUPP);
711
712 /*
713 * Disallow write attempts if the filesystem is mounted read-only.
714 */
715 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
716 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
717 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
718 (vp->v_mount->mnt_flag & MNT_RDONLY))
719 return (EROFS);
720
721 if (vap->va_size != VNOVAL) {
722 /*
723 * truncation requested
724 */
725 switch (vp->v_type) {
726 case VDIR:
727 return (EISDIR);
728 case VCHR:
729 case VBLK:
730 case VSOCK:
731 case VFIFO:
732 if (vap->va_mtime.tv_sec == VNOVAL &&
733 vap->va_atime.tv_sec == VNOVAL &&
734 vap->va_mode == (mode_t)VNOVAL &&
735 vap->va_uid == (uid_t)VNOVAL &&
736 vap->va_gid == (gid_t)VNOVAL)
737 return (0);
738 vap->va_size = VNOVAL;
739 break;
740 default:
741 /*
742 * Disallow write attempts if the filesystem is
743 * mounted read-only.
744 */
745 if (vp->v_mount->mnt_flag & MNT_RDONLY)
746 return (EROFS);
747
748 tsize = np->n_size;
749 again:
750 boff = (int)vap->va_size & (biosize - 1);
751 error = nfs_meta_setsize(vp, td, vap->va_size, 0);
752
753 #if 0
754 if (np->n_flag & NLMODIFIED) {
755 if (vap->va_size == 0)
756 error = nfs_vinvalbuf(vp, 0, 1);
757 else
758 error = nfs_vinvalbuf(vp, V_SAVE, 1);
759 }
760 #endif
761 /*
762 * note: this loop case almost always happens at
763 * least once per truncation.
764 */
765 if (error == 0 && np->n_size != vap->va_size)
766 goto again;
767 np->n_vattr.va_size = vap->va_size;
768 break;
769 }
770 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
771 /*
772 * What to do. If we are modifying the mtime we lose
773 * mtime detection of changes made by the server or other
774 * clients. But programs like rsync/rdist/cpdup are going
775 * to call utimes a lot. We don't want to piecemeal sync.
776 *
777 * For now sync if any prior remote changes were detected,
778 * but allow us to lose track of remote changes made during
779 * the utimes operation.
780 */
781 if (np->n_flag & NRMODIFIED)
782 error = nfs_vinvalbuf(vp, V_SAVE, 1);
783 if (error == EINTR)
784 return (error);
785 if (error == 0) {
786 if (vap->va_mtime.tv_sec != VNOVAL) {
787 np->n_mtime = vap->va_mtime.tv_sec;
788 }
789 }
790 }
791 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
792
793 /*
794 * Sanity check if a truncation was issued. This should only occur
795 * if multiple processes are racing on the same file.
796 */
797 if (error == 0 && vap->va_size != VNOVAL &&
798 np->n_size != vap->va_size) {
799 kprintf("NFS ftruncate: server disagrees on the file size: "
800 "%jd/%jd/%jd\n",
801 (intmax_t)tsize,
802 (intmax_t)vap->va_size,
803 (intmax_t)np->n_size);
804 goto again;
805 }
806 if (error && vap->va_size != VNOVAL) {
807 np->n_size = np->n_vattr.va_size = tsize;
808 nfs_meta_setsize(vp, td, np->n_size, 0);
809 }
810 return (error);
811 }
812
813 /*
814 * Do an nfs setattr rpc.
815 */
816 static int
817 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
818 struct ucred *cred, struct thread *td)
819 {
820 struct nfsv2_sattr *sp;
821 struct nfsnode *np = VTONFS(vp);
822 u_int32_t *tl;
823 int error = 0, wccflag = NFSV3_WCCRATTR;
824 struct nfsm_info info;
825
826 info.mrep = NULL;
827 info.v3 = NFS_ISV3(vp);
828
829 nfsstats.rpccnt[NFSPROC_SETATTR]++;
830 nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
831 NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
832 ERROROUT(nfsm_fhtom(&info, vp));
833 if (info.v3) {
834 nfsm_v3attrbuild(&info, vap, TRUE);
835 tl = nfsm_build(&info, NFSX_UNSIGNED);
836 *tl = nfs_false;
837 } else {
838 sp = nfsm_build(&info, NFSX_V2SATTR);
839 if (vap->va_mode == (mode_t)VNOVAL)
840 sp->sa_mode = nfs_xdrneg1;
841 else
842 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
843 if (vap->va_uid == (uid_t)VNOVAL)
844 sp->sa_uid = nfs_xdrneg1;
845 else
846 sp->sa_uid = txdr_unsigned(vap->va_uid);
847 if (vap->va_gid == (gid_t)VNOVAL)
848 sp->sa_gid = nfs_xdrneg1;
849 else
850 sp->sa_gid = txdr_unsigned(vap->va_gid);
851 sp->sa_size = txdr_unsigned(vap->va_size);
852 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
853 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
854 }
855 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
856 if (info.v3) {
857 np->n_modestamp = 0;
858 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
859 } else {
860 ERROROUT(nfsm_loadattr(&info, vp, NULL));
861 }
862 m_freem(info.mrep);
863 info.mrep = NULL;
864 nfsmout:
865 return (error);
866 }
867
868 static
869 void
870 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
871 {
872 if (nctimeout == 0)
873 nctimeout = 1;
874 else
875 nctimeout *= hz;
876 cache_setvp(nch, vp);
877 cache_settimeout(nch, nctimeout);
878 }
879
880 /*
881 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
882 * nfs_lookup() until all remaining new api calls are implemented.
883 *
884 * Resolve a namecache entry. This function is passed a locked ncp and
885 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
886 */
887 static int
888 nfs_nresolve(struct vop_nresolve_args *ap)
889 {
890 struct thread *td = curthread;
891 struct namecache *ncp;
892 struct ucred *cred;
893 struct nfsnode *np;
894 struct vnode *dvp;
895 struct vnode *nvp;
896 nfsfh_t *fhp;
897 int attrflag;
898 int fhsize;
899 int error;
900 int tmp_error;
901 int len;
902 struct nfsm_info info;
903
904 cred = ap->a_cred;
905 dvp = ap->a_dvp;
906
907 if ((error = vget(dvp, LK_SHARED)) != 0)
908 return (error);
909
910 info.mrep = NULL;
911 info.v3 = NFS_ISV3(dvp);
912
913 nvp = NULL;
914 nfsstats.lookupcache_misses++;
915 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
916 ncp = ap->a_nch->ncp;
917 len = ncp->nc_nlen;
918 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
919 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
920 ERROROUT(nfsm_fhtom(&info, dvp));
921 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
922 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
923 ap->a_cred, &error));
924 if (error) {
925 /*
926 * Cache negatve lookups to reduce NFS traffic, but use
927 * a fast timeout. Otherwise use a timeout of 1 tick.
928 * XXX we should add a namecache flag for no-caching
929 * to uncache the negative hit as soon as possible, but
930 * we cannot simply destroy the entry because it is used
931 * as a placeholder by the caller.
932 *
933 * The refactored nfs code will overwrite a non-zero error
934 * with 0 when we use ERROROUT(), so don't here.
935 */
936 if (error == ENOENT)
937 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
938 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
939 NFS_LATTR_NOSHRINK);
940 if (tmp_error) {
941 error = tmp_error;
942 goto nfsmout;
943 }
944 m_freem(info.mrep);
945 info.mrep = NULL;
946 goto nfsmout;
947 }
948
949 /*
950 * Success, get the file handle, do various checks, and load
951 * post-operation data from the reply packet. Theoretically
952 * we should never be looking up "." so, theoretically, we
953 * should never get the same file handle as our directory. But
954 * we check anyway. XXX
955 *
956 * Note that no timeout is set for the positive cache hit. We
957 * assume, theoretically, that ESTALE returns will be dealt with
958 * properly to handle NFS races and in anycase we cannot depend
959 * on a timeout to deal with NFS open/create/excl issues so instead
960 * of a bad hack here the rest of the NFS client code needs to do
961 * the right thing.
962 */
963 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
964
965 np = VTONFS(dvp);
966 if (NFS_CMPFH(np, fhp, fhsize)) {
967 vref(dvp);
968 nvp = dvp;
969 } else {
970 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
971 if (error) {
972 m_freem(info.mrep);
973 info.mrep = NULL;
974 vput(dvp);
975 return (error);
976 }
977 nvp = NFSTOV(np);
978 }
979 if (info.v3) {
980 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
981 NFS_LATTR_NOSHRINK));
982 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
983 NFS_LATTR_NOSHRINK));
984 } else {
985 ERROROUT(nfsm_loadattr(&info, nvp, NULL));
986 }
987 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
988 m_freem(info.mrep);
989 info.mrep = NULL;
990 nfsmout:
991 vput(dvp);
992 if (nvp) {
993 if (nvp == dvp)
994 vrele(nvp);
995 else
996 vput(nvp);
997 }
998 return (error);
999 }
1000
1001 /*
1002 * 'cached' nfs directory lookup
1003 *
1004 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
1005 *
1006 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
1007 * struct componentname *a_cnp)
1008 */
1009 static int
1010 nfs_lookup(struct vop_old_lookup_args *ap)
1011 {
1012 struct componentname *cnp = ap->a_cnp;
1013 struct vnode *dvp = ap->a_dvp;
1014 struct vnode **vpp = ap->a_vpp;
1015 int flags = cnp->cn_flags;
1016 struct vnode *newvp;
1017 struct nfsmount *nmp;
1018 long len;
1019 nfsfh_t *fhp;
1020 struct nfsnode *np;
1021 int lockparent, wantparent, attrflag, fhsize;
1022 int error;
1023 int tmp_error;
1024 struct nfsm_info info;
1025
1026 info.mrep = NULL;
1027 info.v3 = NFS_ISV3(dvp);
1028 error = 0;
1029
1030 /*
1031 * Read-only mount check and directory check.
1032 */
1033 *vpp = NULLVP;
1034 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1035 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1036 return (EROFS);
1037
1038 if (dvp->v_type != VDIR)
1039 return (ENOTDIR);
1040
1041 /*
1042 * Look it up in the cache. Note that ENOENT is only returned if we
1043 * previously entered a negative hit (see later on). The additional
1044 * nfsneg_cache_timeout check causes previously cached results to
1045 * be instantly ignored if the negative caching is turned off.
1046 */
1047 lockparent = flags & CNP_LOCKPARENT;
1048 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1049 nmp = VFSTONFS(dvp->v_mount);
1050 np = VTONFS(dvp);
1051
1052 /*
1053 * Go to the wire.
1054 */
1055 error = 0;
1056 newvp = NULLVP;
1057 nfsstats.lookupcache_misses++;
1058 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1059 len = cnp->cn_namelen;
1060 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1061 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1062 ERROROUT(nfsm_fhtom(&info, dvp));
1063 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1064 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1065 cnp->cn_cred, &error));
1066 if (error) {
1067 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1068 NFS_LATTR_NOSHRINK);
1069 if (tmp_error) {
1070 error = tmp_error;
1071 goto nfsmout;
1072 }
1073
1074 m_freem(info.mrep);
1075 info.mrep = NULL;
1076 goto nfsmout;
1077 }
1078 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1079
1080 /*
1081 * Handle RENAME case...
1082 */
1083 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1084 if (NFS_CMPFH(np, fhp, fhsize)) {
1085 m_freem(info.mrep);
1086 info.mrep = NULL;
1087 return (EISDIR);
1088 }
1089 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1090 if (error) {
1091 m_freem(info.mrep);
1092 info.mrep = NULL;
1093 return (error);
1094 }
1095 newvp = NFSTOV(np);
1096 if (info.v3) {
1097 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1098 NFS_LATTR_NOSHRINK));
1099 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1100 NFS_LATTR_NOSHRINK));
1101 } else {
1102 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1103 }
1104 *vpp = newvp;
1105 m_freem(info.mrep);
1106 info.mrep = NULL;
1107 if (!lockparent) {
1108 vn_unlock(dvp);
1109 cnp->cn_flags |= CNP_PDIRUNLOCK;
1110 }
1111 return (0);
1112 }
1113
1114 if (flags & CNP_ISDOTDOT) {
1115 vn_unlock(dvp);
1116 cnp->cn_flags |= CNP_PDIRUNLOCK;
1117 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1118 if (error) {
1119 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1120 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1121 return (error); /* NOTE: return error from nget */
1122 }
1123 newvp = NFSTOV(np);
1124 if (lockparent) {
1125 error = vn_lock(dvp, LK_EXCLUSIVE);
1126 if (error) {
1127 vput(newvp);
1128 return (error);
1129 }
1130 cnp->cn_flags |= CNP_PDIRUNLOCK;
1131 }
1132 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1133 vref(dvp);
1134 newvp = dvp;
1135 } else {
1136 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1137 if (error) {
1138 m_freem(info.mrep);
1139 info.mrep = NULL;
1140 return (error);
1141 }
1142 if (!lockparent) {
1143 vn_unlock(dvp);
1144 cnp->cn_flags |= CNP_PDIRUNLOCK;
1145 }
1146 newvp = NFSTOV(np);
1147 }
1148 if (info.v3) {
1149 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1150 NFS_LATTR_NOSHRINK));
1151 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1152 NFS_LATTR_NOSHRINK));
1153 } else {
1154 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1155 }
1156 #if 0
1157 /* XXX MOVE TO nfs_nremove() */
1158 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1159 cnp->cn_nameiop != NAMEI_DELETE) {
1160 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1161 }
1162 #endif
1163 *vpp = newvp;
1164 m_freem(info.mrep);
1165 info.mrep = NULL;
1166 nfsmout:
1167 if (error) {
1168 if (newvp != NULLVP) {
1169 vrele(newvp);
1170 *vpp = NULLVP;
1171 }
1172 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1173 cnp->cn_nameiop == NAMEI_RENAME) &&
1174 error == ENOENT) {
1175 if (!lockparent) {
1176 vn_unlock(dvp);
1177 cnp->cn_flags |= CNP_PDIRUNLOCK;
1178 }
1179 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1180 error = EROFS;
1181 else
1182 error = EJUSTRETURN;
1183 }
1184 }
1185 return (error);
1186 }
1187
1188 /*
1189 * nfs read call.
1190 * Just call nfs_bioread() to do the work.
1191 *
1192 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1193 * struct ucred *a_cred)
1194 */
1195 static int
1196 nfs_read(struct vop_read_args *ap)
1197 {
1198 struct vnode *vp = ap->a_vp;
1199
1200 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1201 }
1202
1203 /*
1204 * nfs readlink call
1205 *
1206 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1207 */
1208 static int
1209 nfs_readlink(struct vop_readlink_args *ap)
1210 {
1211 struct vnode *vp = ap->a_vp;
1212
1213 if (vp->v_type != VLNK)
1214 return (EINVAL);
1215 return (nfs_bioread(vp, ap->a_uio, 0));
1216 }
1217
1218 /*
1219 * Do a readlink rpc.
1220 * Called by nfs_doio() from below the buffer cache.
1221 */
1222 int
1223 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1224 {
1225 int error = 0, len, attrflag;
1226 struct nfsm_info info;
1227
1228 info.mrep = NULL;
1229 info.v3 = NFS_ISV3(vp);
1230
1231 nfsstats.rpccnt[NFSPROC_READLINK]++;
1232 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1233 ERROROUT(nfsm_fhtom(&info, vp));
1234 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1235 nfs_vpcred(vp, ND_CHECK), &error));
1236 if (info.v3) {
1237 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1238 NFS_LATTR_NOSHRINK));
1239 }
1240 if (!error) {
1241 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1242 if (len == NFS_MAXPATHLEN) {
1243 struct nfsnode *np = VTONFS(vp);
1244 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1245 len = np->n_size;
1246 }
1247 ERROROUT(nfsm_mtouio(&info, uiop, len));
1248 }
1249 m_freem(info.mrep);
1250 info.mrep = NULL;
1251 nfsmout:
1252 return (error);
1253 }
1254
1255 /*
1256 * nfs synchronous read rpc using UIO
1257 */
1258 int
1259 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1260 {
1261 u_int32_t *tl;
1262 struct nfsmount *nmp;
1263 int error = 0, len, retlen, tsiz, eof, attrflag;
1264 struct nfsm_info info;
1265 off_t tmp_off;
1266
1267 info.mrep = NULL;
1268 info.v3 = NFS_ISV3(vp);
1269
1270 #ifndef nolint
1271 eof = 0;
1272 #endif
1273 nmp = VFSTONFS(vp->v_mount);
1274 tsiz = uiop->uio_resid;
1275 tmp_off = uiop->uio_offset + tsiz;
1276 if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1277 return (EFBIG);
1278 tmp_off = uiop->uio_offset;
1279 while (tsiz > 0) {
1280 nfsstats.rpccnt[NFSPROC_READ]++;
1281 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1282 nfsm_reqhead(&info, vp, NFSPROC_READ,
1283 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1284 ERROROUT(nfsm_fhtom(&info, vp));
1285 tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1286 if (info.v3) {
1287 txdr_hyper(uiop->uio_offset, tl);
1288 *(tl + 2) = txdr_unsigned(len);
1289 } else {
1290 *tl++ = txdr_unsigned(uiop->uio_offset);
1291 *tl++ = txdr_unsigned(len);
1292 *tl = 0;
1293 }
1294 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1295 nfs_vpcred(vp, ND_READ), &error));
1296 if (info.v3) {
1297 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1298 NFS_LATTR_NOSHRINK));
1299 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1300 eof = fxdr_unsigned(int, *(tl + 1));
1301 } else {
1302 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1303 }
1304 NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1305 ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1306 m_freem(info.mrep);
1307 info.mrep = NULL;
1308
1309 /*
1310 * Handle short-read from server (NFSv3). If EOF is not
1311 * flagged (and no error occurred), but retlen is less
1312 * then the request size, we must zero-fill the remainder.
1313 */
1314 if (retlen < len && info.v3 && eof == 0) {
1315 ERROROUT(uiomovez(len - retlen, uiop));
1316 retlen = len;
1317 }
1318 tsiz -= retlen;
1319
1320 /*
1321 * Terminate loop on EOF or zero-length read.
1322 *
1323 * For NFSv2 a short-read indicates EOF, not zero-fill,
1324 * and also terminates the loop.
1325 */
1326 if (info.v3) {
1327 if (eof || retlen == 0)
1328 tsiz = 0;
1329 } else if (retlen < len) {
1330 tsiz = 0;
1331 }
1332 }
1333 nfsmout:
1334 return (error);
1335 }
1336
1337 /*
1338 * nfs write call
1339 */
1340 int
1341 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1342 int *iomode, int *must_commit)
1343 {
1344 u_int32_t *tl;
1345 int32_t backup;
1346 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1347 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1348 int committed = NFSV3WRITE_FILESYNC;
1349 struct nfsm_info info;
1350
1351 info.mrep = NULL;
1352 info.v3 = NFS_ISV3(vp);
1353
1354 #ifndef DIAGNOSTIC
1355 if (uiop->uio_iovcnt != 1)
1356 panic("nfs: writerpc iovcnt > 1");
1357 #endif
1358 *must_commit = 0;
1359 tsiz = uiop->uio_resid;
1360 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1361 return (EFBIG);
1362 while (tsiz > 0) {
1363 nfsstats.rpccnt[NFSPROC_WRITE]++;
1364 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1365 nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1366 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1367 ERROROUT(nfsm_fhtom(&info, vp));
1368 if (info.v3) {
1369 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1370 txdr_hyper(uiop->uio_offset, tl);
1371 tl += 2;
1372 *tl++ = txdr_unsigned(len);
1373 *tl++ = txdr_unsigned(*iomode);
1374 *tl = txdr_unsigned(len);
1375 } else {
1376 u_int32_t x;
1377
1378 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1379 /* Set both "begin" and "current" to non-garbage. */
1380 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1381 *tl++ = x; /* "begin offset" */
1382 *tl++ = x; /* "current offset" */
1383 x = txdr_unsigned(len);
1384 *tl++ = x; /* total to this offset */
1385 *tl = x; /* size of this write */
1386 }
1387 ERROROUT(nfsm_uiotom(&info, uiop, len));
1388 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1389 nfs_vpcred(vp, ND_WRITE), &error));
1390 if (info.v3) {
1391 /*
1392 * The write RPC returns a before and after mtime. The
1393 * nfsm_wcc_data() macro checks the before n_mtime
1394 * against the before time and stores the after time
1395 * in the nfsnode's cached vattr and n_mtime field.
1396 * The NRMODIFIED bit will be set if the before
1397 * time did not match the original mtime.
1398 */
1399 wccflag = NFSV3_WCCCHK;
1400 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1401 if (error == 0) {
1402 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1403 rlen = fxdr_unsigned(int, *tl++);
1404 if (rlen == 0) {
1405 error = NFSERR_IO;
1406 m_freem(info.mrep);
1407 info.mrep = NULL;
1408 break;
1409 } else if (rlen < len) {
1410 backup = len - rlen;
1411 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1412 uiop->uio_iov->iov_len += backup;
1413 uiop->uio_offset -= backup;
1414 uiop->uio_resid += backup;
1415 len = rlen;
1416 }
1417 commit = fxdr_unsigned(int, *tl++);
1418
1419 /*
1420 * Return the lowest committment level
1421 * obtained by any of the RPCs.
1422 */
1423 if (committed == NFSV3WRITE_FILESYNC)
1424 committed = commit;
1425 else if (committed == NFSV3WRITE_DATASYNC &&
1426 commit == NFSV3WRITE_UNSTABLE)
1427 committed = commit;
1428 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1429 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1430 NFSX_V3WRITEVERF);
1431 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1432 } else if (bcmp((caddr_t)tl,
1433 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1434 *must_commit = 1;
1435 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1436 NFSX_V3WRITEVERF);
1437 }
1438 }
1439 } else {
1440 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1441 }
1442 m_freem(info.mrep);
1443 info.mrep = NULL;
1444 if (error)
1445 break;
1446 tsiz -= len;
1447 }
1448 nfsmout:
1449 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1450 committed = NFSV3WRITE_FILESYNC;
1451 *iomode = committed;
1452 if (error)
1453 uiop->uio_resid = tsiz;
1454 return (error);
1455 }
1456
1457 /*
1458 * nfs mknod rpc
1459 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1460 * mode set to specify the file type and the size field for rdev.
1461 */
1462 static int
1463 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1464 struct vattr *vap)
1465 {
1466 struct nfsv2_sattr *sp;
1467 u_int32_t *tl;
1468 struct vnode *newvp = NULL;
1469 struct nfsnode *np = NULL;
1470 struct vattr vattr;
1471 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1472 int rmajor, rminor;
1473 struct nfsm_info info;
1474
1475 info.mrep = NULL;
1476 info.v3 = NFS_ISV3(dvp);
1477
1478 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1479 rmajor = txdr_unsigned(vap->va_rmajor);
1480 rminor = txdr_unsigned(vap->va_rminor);
1481 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1482 rmajor = nfs_xdrneg1;
1483 rminor = nfs_xdrneg1;
1484 } else {
1485 return (EOPNOTSUPP);
1486 }
1487 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1488 return (error);
1489 }
1490 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1491 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1492 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1493 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1494 ERROROUT(nfsm_fhtom(&info, dvp));
1495 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1496 NFS_MAXNAMLEN));
1497 if (info.v3) {
1498 tl = nfsm_build(&info, NFSX_UNSIGNED);
1499 *tl++ = vtonfsv3_type(vap->va_type);
1500 nfsm_v3attrbuild(&info, vap, FALSE);
1501 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1502 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1503 *tl++ = txdr_unsigned(vap->va_rmajor);
1504 *tl = txdr_unsigned(vap->va_rminor);
1505 }
1506 } else {
1507 sp = nfsm_build(&info, NFSX_V2SATTR);
1508 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1509 sp->sa_uid = nfs_xdrneg1;
1510 sp->sa_gid = nfs_xdrneg1;
1511 sp->sa_size = makeudev(rmajor, rminor);
1512 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1513 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1514 }
1515 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1516 cnp->cn_cred, &error));
1517 if (!error) {
1518 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1519 if (!gotvp) {
1520 if (newvp) {
1521 vput(newvp);
1522 newvp = NULL;
1523 }
1524 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1525 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1526 if (!error)
1527 newvp = NFSTOV(np);
1528 }
1529 }
1530 if (info.v3) {
1531 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1532 }
1533 m_freem(info.mrep);
1534 info.mrep = NULL;
1535 nfsmout:
1536 if (error) {
1537 if (newvp)
1538 vput(newvp);
1539 } else {
1540 *vpp = newvp;
1541 }
1542 VTONFS(dvp)->n_flag |= NLMODIFIED;
1543 if (!wccflag)
1544 VTONFS(dvp)->n_attrstamp = 0;
1545 return (error);
1546 }
1547
1548 /*
1549 * nfs mknod vop
1550 * just call nfs_mknodrpc() to do the work.
1551 *
1552 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1553 * struct componentname *a_cnp, struct vattr *a_vap)
1554 */
1555 /* ARGSUSED */
1556 static int
1557 nfs_mknod(struct vop_old_mknod_args *ap)
1558 {
1559 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1560 }
1561
1562 static u_long create_verf;
1563 /*
1564 * nfs file create call
1565 *
1566 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1567 * struct componentname *a_cnp, struct vattr *a_vap)
1568 */
1569 static int
1570 nfs_create(struct vop_old_create_args *ap)
1571 {
1572 struct vnode *dvp = ap->a_dvp;
1573 struct vattr *vap = ap->a_vap;
1574 struct componentname *cnp = ap->a_cnp;
1575 struct nfsv2_sattr *sp;
1576 u_int32_t *tl;
1577 struct nfsnode *np = NULL;
1578 struct vnode *newvp = NULL;
1579 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1580 struct vattr vattr;
1581 struct nfsm_info info;
1582
1583 info.mrep = NULL;
1584 info.v3 = NFS_ISV3(dvp);
1585
1586 /*
1587 * Oops, not for me..
1588 */
1589 if (vap->va_type == VSOCK)
1590 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1591
1592 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1593 return (error);
1594 }
1595 if (vap->va_vaflags & VA_EXCLUSIVE)
1596 fmode |= O_EXCL;
1597 again:
1598 nfsstats.rpccnt[NFSPROC_CREATE]++;
1599 nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1600 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1601 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1602 ERROROUT(nfsm_fhtom(&info, dvp));
1603 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1604 NFS_MAXNAMLEN));
1605 if (info.v3) {
1606 tl = nfsm_build(&info, NFSX_UNSIGNED);
1607 if (fmode & O_EXCL) {
1608 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1609 tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1610 #ifdef INET
1611 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1612 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1613 else
1614 #endif
1615 *tl++ = create_verf;
1616 *tl = ++create_verf;
1617 } else {
1618 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1619 nfsm_v3attrbuild(&info, vap, FALSE);
1620 }
1621 } else {
1622 sp = nfsm_build(&info, NFSX_V2SATTR);
1623 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1624 sp->sa_uid = nfs_xdrneg1;
1625 sp->sa_gid = nfs_xdrneg1;
1626 sp->sa_size = 0;
1627 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1628 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1629 }
1630 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1631 cnp->cn_cred, &error));
1632 if (error == 0) {
1633 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1634 if (!gotvp) {
1635 if (newvp) {
1636 vput(newvp);
1637 newvp = NULL;
1638 }
1639 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1640 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1641 if (!error)
1642 newvp = NFSTOV(np);
1643 }
1644 }
1645 if (info.v3) {
1646 if (error == 0)
1647 error = nfsm_wcc_data(&info, dvp, &wccflag);
1648 else
1649 (void)nfsm_wcc_data(&info, dvp, &wccflag);
1650 }
1651 m_freem(info.mrep);
1652 info.mrep = NULL;
1653 nfsmout:
1654 if (error) {
1655 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1656 KKASSERT(newvp == NULL);
1657 fmode &= ~O_EXCL;
1658 goto again;
1659 }
1660 } else if (info.v3 && (fmode & O_EXCL)) {
1661 /*
1662 * We are normally called with only a partially initialized
1663 * VAP. Since the NFSv3 spec says that server may use the
1664 * file attributes to store the verifier, the spec requires
1665 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1666 * in atime, but we can't really assume that all servers will
1667 * so we ensure that our SETATTR sets both atime and mtime.
1668 */
1669 if (vap->va_mtime.tv_sec == VNOVAL)
1670 vfs_timestamp(&vap->va_mtime);
1671 if (vap->va_atime.tv_sec == VNOVAL)
1672 vap->va_atime = vap->va_mtime;
1673 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1674 }
1675 if (error == 0) {
1676 /*
1677 * The new np may have enough info for access
1678 * checks, make sure rucred and wucred are
1679 * initialized for read and write rpc's.
1680 */
1681 np = VTONFS(newvp);
1682 if (np->n_rucred == NULL)
1683 np->n_rucred = crhold(cnp->cn_cred);
1684 if (np->n_wucred == NULL)
1685 np->n_wucred = crhold(cnp->cn_cred);
1686 *ap->a_vpp = newvp;
1687 } else if (newvp) {
1688 vput(newvp);
1689 }
1690 VTONFS(dvp)->n_flag |= NLMODIFIED;
1691 if (!wccflag)
1692 VTONFS(dvp)->n_attrstamp = 0;
1693 return (error);
1694 }
1695
1696 /*
1697 * nfs file remove call
1698 * To try and make nfs semantics closer to ufs semantics, a file that has
1699 * other processes using the vnode is renamed instead of removed and then
1700 * removed later on the last close.
1701 * - If v_sysref.refcnt > 1
1702 * If a rename is not already in the works
1703 * call nfs_sillyrename() to set it up
1704 * else
1705 * do the remove rpc
1706 *
1707 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1708 * struct componentname *a_cnp)
1709 */
1710 static int
1711 nfs_remove(struct vop_old_remove_args *ap)
1712 {
1713 struct vnode *vp = ap->a_vp;
1714 struct vnode *dvp = ap->a_dvp;
1715 struct componentname *cnp = ap->a_cnp;
1716 struct nfsnode *np = VTONFS(vp);
1717 int error = 0;
1718 struct vattr vattr;
1719
1720 #ifndef DIAGNOSTIC
1721 if (vp->v_sysref.refcnt < 1)
1722 panic("nfs_remove: bad v_sysref.refcnt");
1723 #endif
1724 if (vp->v_type == VDIR)
1725 error = EPERM;
1726 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1727 VOP_GETATTR(vp, &vattr) == 0 &&
1728 vattr.va_nlink > 1)) {
1729 /*
1730 * throw away biocache buffers, mainly to avoid
1731 * unnecessary delayed writes later.
1732 */
1733 error = nfs_vinvalbuf(vp, 0, 1);
1734 /* Do the rpc */
1735 if (error != EINTR)
1736 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1737 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1738 /*
1739 * Kludge City: If the first reply to the remove rpc is lost..
1740 * the reply to the retransmitted request will be ENOENT
1741 * since the file was in fact removed
1742 * Therefore, we cheat and return success.
1743 */
1744 if (error == ENOENT)
1745 error = 0;
1746 } else if (!np->n_sillyrename) {
1747 error = nfs_sillyrename(dvp, vp, cnp);
1748 }
1749 np->n_attrstamp = 0;
1750 return (error);
1751 }
1752
1753 /*
1754 * nfs file remove rpc called from nfs_inactive
1755 */
1756 int
1757 nfs_removeit(struct sillyrename *sp)
1758 {
1759 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1760 sp->s_cred, NULL));
1761 }
1762
1763 /*
1764 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1765 */
1766 static int
1767 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1768 struct ucred *cred, struct thread *td)
1769 {
1770 int error = 0, wccflag = NFSV3_WCCRATTR;
1771 struct nfsm_info info;
1772
1773 info.mrep = NULL;
1774 info.v3 = NFS_ISV3(dvp);
1775
1776 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1777 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1778 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1779 ERROROUT(nfsm_fhtom(&info, dvp));
1780 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1781 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1782 if (info.v3) {
1783 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1784 }
1785 m_freem(info.mrep);
1786 info.mrep = NULL;
1787 nfsmout:
1788 VTONFS(dvp)->n_flag |= NLMODIFIED;
1789 if (!wccflag)
1790 VTONFS(dvp)->n_attrstamp = 0;
1791 return (error);
1792 }
1793
1794 /*
1795 * nfs file rename call
1796 *
1797 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1798 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1799 * struct vnode *a_tvp, struct componentname *a_tcnp)
1800 */
1801 static int
1802 nfs_rename(struct vop_old_rename_args *ap)
1803 {
1804 struct vnode *fvp = ap->a_fvp;
1805 struct vnode *tvp = ap->a_tvp;
1806 struct vnode *fdvp = ap->a_fdvp;
1807 struct vnode *tdvp = ap->a_tdvp;
1808 struct componentname *tcnp = ap->a_tcnp;
1809 struct componentname *fcnp = ap->a_fcnp;
1810 int error;
1811
1812 /* Check for cross-device rename */
1813 if ((fvp->v_mount != tdvp->v_mount) ||
1814 (tvp && (fvp->v_mount != tvp->v_mount))) {
1815 error = EXDEV;
1816 goto out;
1817 }
1818
1819 /*
1820 * We shouldn't have to flush fvp on rename for most server-side
1821 * filesystems as the file handle should not change. Unfortunately
1822 * the inode for some filesystems (msdosfs) might be tied to the
1823 * file name or directory position so to be completely safe
1824 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1825 * performance.
1826 *
1827 * We must flush tvp on rename because it might become stale on the
1828 * server after the rename.
1829 */
1830 if (nfs_flush_on_rename)
1831 VOP_FSYNC(fvp, MNT_WAIT, 0);
1832 if (tvp)
1833 VOP_FSYNC(tvp, MNT_WAIT, 0);
1834
1835 /*
1836 * If the tvp exists and is in use, sillyrename it before doing the
1837 * rename of the new file over it.
1838 *
1839 * XXX Can't sillyrename a directory.
1840 *
1841 * We do not attempt to do any namecache purges in this old API
1842 * routine. The new API compat functions have access to the actual
1843 * namecache structures and will do it for us.
1844 */
1845 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1846 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1847 vput(tvp);
1848 tvp = NULL;
1849 } else if (tvp) {
1850 ;
1851 }
1852
1853 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1854 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1855 tcnp->cn_td);
1856
1857 out:
1858 if (tdvp == tvp)
1859 vrele(tdvp);
1860 else
1861 vput(tdvp);
1862 if (tvp)
1863 vput(tvp);
1864 vrele(fdvp);
1865 vrele(fvp);
1866 /*
1867 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1868 */
1869 if (error == ENOENT)
1870 error = 0;
1871 return (error);
1872 }
1873
1874 /*
1875 * nfs file rename rpc called from nfs_remove() above
1876 */
1877 static int
1878 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1879 struct sillyrename *sp)
1880 {
1881 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1882 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1883 }
1884
1885 /*
1886 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1887 */
1888 static int
1889 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1890 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1891 struct ucred *cred, struct thread *td)
1892 {
1893 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1894 struct nfsm_info info;
1895
1896 info.mrep = NULL;
1897 info.v3 = NFS_ISV3(fdvp);
1898
1899 nfsstats.rpccnt[NFSPROC_RENAME]++;
1900 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1901 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1902 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1903 ERROROUT(nfsm_fhtom(&info, fdvp));
1904 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1905 ERROROUT(nfsm_fhtom(&info, tdvp));
1906 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1907 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
1908 if (info.v3) {
1909 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
1910 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
1911 }
1912 m_freem(info.mrep);
1913 info.mrep = NULL;
1914 nfsmout:
1915 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1916 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1917 if (!fwccflag)
1918 VTONFS(fdvp)->n_attrstamp = 0;
1919 if (!twccflag)
1920 VTONFS(tdvp)->n_attrstamp = 0;
1921 return (error);
1922 }
1923
1924 /*
1925 * nfs hard link create call
1926 *
1927 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1928 * struct componentname *a_cnp)
1929 */
1930 static int
1931 nfs_link(struct vop_old_link_args *ap)
1932 {
1933 struct vnode *vp = ap->a_vp;
1934 struct vnode *tdvp = ap->a_tdvp;
1935 struct componentname *cnp = ap->a_cnp;
1936 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1937 struct nfsm_info info;
1938
1939 if (vp->v_mount != tdvp->v_mount) {
1940 return (EXDEV);
1941 }
1942
1943 /*
1944 * The attribute cache may get out of sync with the server on link.
1945 * Pushing writes to the server before handle was inherited from
1946 * long long ago and it is unclear if we still need to do this.
1947 * Defaults to off.
1948 */
1949 if (nfs_flush_on_hlink)
1950 VOP_FSYNC(vp, MNT_WAIT, 0);
1951
1952 info.mrep = NULL;
1953 info.v3 = NFS_ISV3(vp);
1954
1955 nfsstats.rpccnt[NFSPROC_LINK]++;
1956 nfsm_reqhead(&info, vp, NFSPROC_LINK,
1957 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
1958 nfsm_rndup(cnp->cn_namelen));
1959 ERROROUT(nfsm_fhtom(&info, vp));
1960 ERROROUT(nfsm_fhtom(&info, tdvp));
1961 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1962 NFS_MAXNAMLEN));
1963 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
1964 cnp->cn_cred, &error));
1965 if (info.v3) {
1966 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1967 NFS_LATTR_NOSHRINK));
1968 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
1969 }
1970 m_freem(info.mrep);
1971 info.mrep = NULL;
1972 nfsmout:
1973 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1974 if (!attrflag)
1975 VTONFS(vp)->n_attrstamp = 0;
1976 if (!wccflag)
1977 VTONFS(tdvp)->n_attrstamp = 0;
1978 /*
1979 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1980 */
1981 if (error == EEXIST)
1982 error = 0;
1983 return (error);
1984 }
1985
1986 /*
1987 * nfs symbolic link create call
1988 *
1989 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1990 * struct componentname *a_cnp, struct vattr *a_vap,
1991 * char *a_target)
1992 */
1993 static int
1994 nfs_symlink(struct vop_old_symlink_args *ap)
1995 {
1996 struct vnode *dvp = ap->a_dvp;
1997 struct vattr *vap = ap->a_vap;
1998 struct componentname *cnp = ap->a_cnp;
1999 struct nfsv2_sattr *sp;
2000 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2001 struct vnode *newvp = NULL;
2002 struct nfsm_info info;
2003
2004 info.mrep = NULL;
2005 info.v3 = NFS_ISV3(dvp);
2006
2007 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2008 slen = strlen(ap->a_target);
2009 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
2010 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
2011 nfsm_rndup(cnp->cn_namelen) +
2012 nfsm_rndup(slen) + NFSX_SATTR(info.v3));
2013 ERROROUT(nfsm_fhtom(&info, dvp));
2014 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2015 NFS_MAXNAMLEN));
2016 if (info.v3) {
2017 nfsm_v3attrbuild(&info, vap, FALSE);
2018 }
2019 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
2020 if (info.v3 == 0) {
2021 sp = nfsm_build(&info, NFSX_V2SATTR);
2022 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2023 sp->sa_uid = nfs_xdrneg1;
2024 sp->sa_gid = nfs_xdrneg1;
2025 sp->sa_size = nfs_xdrneg1;
2026 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2027 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2028 }
2029
2030 /*
2031 * Issue the NFS request and get the rpc response.
2032 *
2033 * Only NFSv3 responses returning an error of 0 actually return
2034 * a file handle that can be converted into newvp without having
2035 * to do an extra lookup rpc.
2036 */
2037 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
2038 cnp->cn_cred, &error));
2039 if (info.v3) {
2040 if (error == 0) {
2041 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2042 }
2043 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2044 }
2045
2046 /*
2047 * out code jumps -> here, mrep is also freed.
2048 */
2049
2050 m_freem(info.mrep);
2051 info.mrep = NULL;
2052 nfsmout:
2053
2054 /*
2055 * If we get an EEXIST error, silently convert it to no-error
2056 * in case of an NFS retry.
2057 */
2058 if (error == EEXIST)
2059 error = 0;
2060
2061 /*
2062 * If we do not have (or no longer have) an error, and we could
2063 * not extract the newvp from the response due to the request being
2064 * NFSv2 or the error being EEXIST. We have to do a lookup in order
2065 * to obtain a newvp to return.
2066 */
2067 if (error == 0 && newvp == NULL) {
2068 struct nfsnode *np = NULL;
2069
2070 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2071 cnp->cn_cred, cnp->cn_td, &np);
2072 if (!error)
2073 newvp = NFSTOV(np);
2074 }
2075 if (error) {
2076 if (newvp)
2077 vput(newvp);
2078 } else {
2079 *ap->a_vpp = newvp;
2080 }
2081 VTONFS(dvp)->n_flag |= NLMODIFIED;
2082 if (!wccflag)
2083 VTONFS(dvp)->n_attrstamp = 0;
2084 return (error);
2085 }
2086
2087 /*
2088 * nfs make dir call
2089 *
2090 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2091 * struct componentname *a_cnp, struct vattr *a_vap)
2092 */
2093 static int
2094 nfs_mkdir(struct vop_old_mkdir_args *ap)
2095 {
2096 struct vnode *dvp = ap->a_dvp;
2097 struct vattr *vap = ap->a_vap;
2098 struct componentname *cnp = ap->a_cnp;
2099 struct nfsv2_sattr *sp;
2100 struct nfsnode *np = NULL;
2101 struct vnode *newvp = NULL;
2102 struct vattr vattr;
2103 int error = 0, wccflag = NFSV3_WCCRATTR;
2104 int gotvp = 0;
2105 int len;
2106 struct nfsm_info info;
2107
2108 info.mrep = NULL;
2109 info.v3 = NFS_ISV3(dvp);
2110
2111 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2112 return (error);
2113 }
2114 len = cnp->cn_namelen;
2115 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2116 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2117 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2118 nfsm_rndup(len) + NFSX_SATTR(info.v3));
2119 ERROROUT(nfsm_fhtom(&info, dvp));
2120 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2121 if (info.v3) {
2122 nfsm_v3attrbuild(&info, vap, FALSE);
2123 } else {
2124 sp = nfsm_build(&info, NFSX_V2SATTR);
2125 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2126 sp->sa_uid = nfs_xdrneg1;
2127 sp->sa_gid = nfs_xdrneg1;
2128 sp->sa_size = nfs_xdrneg1;
2129 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2130 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2131 }
2132 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2133 cnp->cn_cred, &error));
2134 if (error == 0) {
2135 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2136 }
2137 if (info.v3) {
2138 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2139 }
2140 m_freem(info.mrep);
2141 info.mrep = NULL;
2142 nfsmout:
2143 VTONFS(dvp)->n_flag |= NLMODIFIED;
2144 if (!wccflag)
2145 VTONFS(dvp)->n_attrstamp = 0;
2146 /*
2147 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2148 * if we can succeed in looking up the directory.
2149 */
2150 if (error == EEXIST || (!error && !gotvp)) {
2151 if (newvp) {
2152 vrele(newvp);
2153 newvp = NULL;
2154 }
2155 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2156 cnp->cn_td, &np);
2157 if (!error) {
2158 newvp = NFSTOV(np);
2159 if (newvp->v_type != VDIR)
2160 error = EEXIST;
2161 }
2162 }
2163 if (error) {
2164 if (newvp)
2165 vrele(newvp);
2166 } else
2167 *ap->a_vpp = newvp;
2168 return (error);
2169 }
2170
2171 /*
2172 * nfs remove directory call
2173 *
2174 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2175 * struct componentname *a_cnp)
2176 */
2177 static int
2178 nfs_rmdir(struct vop_old_rmdir_args *ap)
2179 {
2180 struct vnode *vp = ap->a_vp;
2181 struct vnode *dvp = ap->a_dvp;
2182 struct componentname *cnp = ap->a_cnp;
2183 int error = 0, wccflag = NFSV3_WCCRATTR;
2184 struct nfsm_info info;
2185
2186 info.mrep = NULL;
2187 info.v3 = NFS_ISV3(dvp);
2188
2189 if (dvp == vp)
2190 return (EINVAL);
2191 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2192 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2193 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2194 nfsm_rndup(cnp->cn_namelen));
2195 ERROROUT(nfsm_fhtom(&info, dvp));
2196 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2197 NFS_MAXNAMLEN));
2198 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2199 cnp->cn_cred, &error));
2200 if (info.v3) {
2201 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2202 }
2203 m_freem(info.mrep);
2204 info.mrep = NULL;
2205 nfsmout:
2206 VTONFS(dvp)->n_flag |= NLMODIFIED;
2207 if (!wccflag)
2208 VTONFS(dvp)->n_attrstamp = 0;
2209 /*
2210 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2211 */
2212 if (error == ENOENT)
2213 error = 0;
2214 return (error);
2215 }
2216
2217 /*
2218 * nfs readdir call
2219 *
2220 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2221 */
2222 static int
2223 nfs_readdir(struct vop_readdir_args *ap)
2224 {
2225 struct vnode *vp = ap->a_vp;
2226 struct nfsnode *np = VTONFS(vp);
2227 struct uio *uio = ap->a_uio;
2228 int tresid, error;
2229 struct vattr vattr;
2230
2231 if (vp->v_type != VDIR)
2232 return (EPERM);
2233
2234 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2235 return (error);
2236
2237 /*
2238 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2239 * and then check that is still valid, or if this is an NQNFS mount
2240 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2241 * VOP_GETATTR() does not necessarily go to the wire.
2242 */
2243 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2244 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2245 if (VOP_GETATTR(vp, &vattr) == 0 &&
2246 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2247 ) {
2248 nfsstats.direofcache_hits++;
2249 goto done;
2250 }
2251 }
2252
2253 /*
2254 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2255 * own cache coherency checks so we do not have to.
2256 */
2257 tresid = uio->uio_resid;
2258 error = nfs_bioread(vp, uio, 0);
2259
2260 if (!error && uio->uio_resid == tresid)
2261 nfsstats.direofcache_misses++;
2262 done:
2263 vn_unlock(vp);
2264 return (error);
2265 }
2266
2267 /*
2268 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2269 *
2270 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2271 * offset/block and converts the nfs formatted directory entries for userland
2272 * consumption as well as deals with offsets into the middle of blocks.
2273 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2274 * be block-bounded. It must convert to cookies for the actual RPC.
2275 */
2276 int
2277 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2278 {
2279 int len, left;
2280 struct nfs_dirent *dp = NULL;
2281 u_int32_t *tl;
2282 nfsuint64 *cookiep;
2283 caddr_t cp;
2284 nfsuint64 cookie;
2285 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2286 struct nfsnode *dnp = VTONFS(vp);
2287 u_quad_t fileno;
2288 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2289 int attrflag;
2290 struct nfsm_info info;
2291
2292 info.mrep = NULL;
2293 info.v3 = NFS_ISV3(vp);
2294
2295 #ifndef DIAGNOSTIC
2296 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2297 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2298 panic("nfs readdirrpc bad uio");
2299 #endif
2300
2301 /*
2302 * If there is no cookie, assume directory was stale.
2303 */
2304 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2305 if (cookiep)
2306 cookie = *cookiep;
2307 else
2308 return (NFSERR_BAD_COOKIE);
2309 /*
2310 * Loop around doing readdir rpc's of size nm_readdirsize
2311 * truncated to a multiple of DIRBLKSIZ.
2312 * The stopping criteria is EOF or buffer full.
2313 */
2314 while (more_dirs && bigenough) {
2315 nfsstats.rpccnt[NFSPROC_READDIR]++;
2316 nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2317 NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2318 ERROROUT(nfsm_fhtom(&info, vp));
2319 if (info.v3) {
2320 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2321 *tl++ = cookie.nfsuquad[0];
2322 *tl++ = cookie.nfsuquad[1];
2323 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2324 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2325 } else {
2326 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2327 *tl++ = cookie.nfsuquad[0];
2328 }
2329 *tl = txdr_unsigned(nmp->nm_readdirsize);
2330 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2331 uiop->uio_td,
2332 nfs_vpcred(vp, ND_READ), &error));
2333 if (info.v3) {
2334 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2335 NFS_LATTR_NOSHRINK));
2336 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2337 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2338 dnp->n_cookieverf.nfsuquad[1] = *tl;
2339 }
2340 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2341 more_dirs = fxdr_unsigned(int, *tl);
2342
2343 /* loop thru the dir entries, converting them to std form */
2344 while (more_dirs && bigenough) {
2345 if (info.v3) {
2346 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2347 fileno = fxdr_hyper(tl);
2348 len = fxdr_unsigned(int, *(tl + 2));
2349 } else {
2350 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2351 fileno = fxdr_unsigned(u_quad_t, *tl++);
2352 len = fxdr_unsigned(int, *tl);
2353 }
2354 if (len <= 0 || len > NFS_MAXNAMLEN) {
2355 error = EBADRPC;
2356 m_freem(info.mrep);
2357 info.mrep = NULL;
2358 goto nfsmout;
2359 }
2360
2361 /*
2362 * len is the number of bytes in the path element
2363 * name, not including the \0 termination.
2364 *
2365 * tlen is the number of bytes w have to reserve for
2366 * the path element name.
2367 */
2368 tlen = nfsm_rndup(len);
2369 if (tlen == len)
2370 tlen += 4; /* To ensure null termination */
2371
2372 /*
2373 * If the entry would cross a DIRBLKSIZ boundary,
2374 * extend the previous nfs_dirent to cover the
2375 * remaining space.
2376 */
2377 left = DIRBLKSIZ - blksiz;
2378 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2379 dp->nfs_reclen += left;
2380 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2381 uiop->uio_iov->iov_len -= left;
2382 uiop->uio_offset += left;
2383 uiop->uio_resid -= left;
2384 blksiz = 0;
2385 }
2386 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2387 bigenough = 0;
2388 if (bigenough) {
2389 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2390 dp->nfs_ino = fileno;
2391 dp->nfs_namlen = len;
2392 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2393 dp->nfs_type = DT_UNKNOWN;
2394 blksiz += dp->nfs_reclen;
2395 if (blksiz == DIRBLKSIZ)
2396 blksiz = 0;
2397 uiop->uio_offset += sizeof(struct nfs_dirent);
2398 uiop->uio_resid -= sizeof(struct nfs_dirent);
2399 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2400 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2401 ERROROUT(nfsm_mtouio(&info, uiop, len));
2402
2403 /*
2404 * The uiop has advanced by nfs_dirent + len
2405 * but really needs to advance by
2406 * nfs_dirent + tlen
2407 */
2408 cp = uiop->uio_iov->iov_base;
2409 tlen -= len;
2410 *cp = '\0'; /* null terminate */
2411 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2412 uiop->uio_iov->iov_len -= tlen;
2413 uiop->uio_offset += tlen;
2414 uiop->uio_resid -= tlen;
2415 } else {
2416 /*
2417 * NFS strings must be rounded up (nfsm_myouio
2418 * handled that in the bigenough case).
2419 */
2420 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2421 }
2422 if (info.v3) {
2423 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2424 } else {
2425 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2426 }
2427
2428 /*
2429 * If we were able to accomodate the last entry,
2430 * get the cookie for the next one. Otherwise
2431 * hold-over the cookie for the one we were not
2432 * able to accomodate.
2433 */
2434 if (bigenough) {
2435 cookie.nfsuquad[0] = *tl++;
2436 if (info.v3)
2437 cookie.nfsuquad[1] = *tl++;
2438 } else if (info.v3) {
2439 tl += 2;
2440 } else {
2441 tl++;
2442 }
2443 more_dirs = fxdr_unsigned(int, *tl);
2444 }
2445 /*
2446 * If at end of rpc data, get the eof boolean
2447 */
2448 if (!more_dirs) {
2449 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2450 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2451 }
2452 m_freem(info.mrep);
2453 info.mrep = NULL;
2454 }
2455 /*
2456 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2457 * by increasing d_reclen for the last record.
2458 */
2459 if (blksiz > 0) {
2460 left = DIRBLKSIZ - blksiz;
2461 dp->nfs_reclen += left;
2462 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2463 uiop->uio_iov->iov_len -= left;
2464 uiop->uio_offset += left;
2465 uiop->uio_resid -= left;
2466 }
2467
2468 if (bigenough) {
2469 /*
2470 * We hit the end of the directory, update direofoffset.
2471 */
2472 dnp->n_direofoffset = uiop->uio_offset;
2473 } else {
2474 /*
2475 * There is more to go, insert the link cookie so the
2476 * next block can be read.
2477 */
2478 if (uiop->uio_resid > 0)
2479 kprintf("EEK! readdirrpc resid > 0\n");
2480 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2481 *cookiep = cookie;
2482 }
2483 nfsmout:
2484 return (error);
2485 }
2486
2487 /*
2488 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2489 */
2490 int
2491 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2492 {
2493 int len, left;
2494 struct nfs_dirent *dp;
2495 u_int32_t *tl;
2496 struct vnode *newvp;
2497 nfsuint64 *cookiep;
2498 caddr_t dpossav1, dpossav2;
2499 caddr_t cp;
2500 struct mbuf *mdsav1, *mdsav2;
2501 nfsuint64 cookie;
2502 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2503 struct nfsnode *dnp = VTONFS(vp), *np;
2504 nfsfh_t *fhp;
2505 u_quad_t fileno;
2506 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2507 int attrflag, fhsize;
2508 int vpls;
2509 struct nchandle nch;
2510 struct nchandle dnch;
2511 struct nlcomponent nlc;
2512 struct nfsm_info info;
2513
2514 info.mrep = NULL;
2515 info.v3 = 1;
2516
2517 #ifndef nolint
2518 dp = NULL;
2519 #endif
2520 #ifndef DIAGNOSTIC
2521 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2522 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2523 panic("nfs readdirplusrpc bad uio");
2524 #endif
2525 /*
2526 * Obtain the namecache record for the directory so we have something
2527 * to use as a basis for creating the entries. This function will
2528 * return a held (but not locked) ncp. The ncp may be disconnected
2529 * from the tree and cannot be used for upward traversals, and the
2530 * ncp may be unnamed. Note that other unrelated operations may
2531 * cause the ncp to be named at any time.
2532 *
2533 * We have to lock the ncp to prevent a lock order reversal when
2534 * rdirplus does nlookups of the children, because the vnode is
2535 * locked and has to stay that way.
2536 */
2537 cache_fromdvp(vp, NULL, 0, &dnch);
2538 bzero(&nlc, sizeof(nlc));
2539 newvp = NULLVP;
2540
2541 /*
2542 * If there is no cookie, assume directory was stale.
2543 */
2544 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2545 if (cookiep) {
2546 cookie = *cookiep;
2547 } else {
2548 if (dnch.ncp)
2549 cache_drop(&dnch);
2550 return (NFSERR_BAD_COOKIE);
2551 }
2552
2553 /*
2554 * Loop around doing readdir rpc's of size nm_readdirsize
2555 * truncated to a multiple of DIRBLKSIZ.
2556 * The stopping criteria is EOF or buffer full.
2557 */
2558 while (more_dirs && bigenough) {
2559 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2560 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2561 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2562 ERROROUT(nfsm_fhtom(&info, vp));
2563 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2564 *tl++ = cookie.nfsuquad[0];
2565 *tl++ = cookie.nfsuquad[1];
2566 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2567 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2568 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2569 *tl = txdr_unsigned(nmp->nm_rsize);
2570 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2571 uiop->uio_td,
2572 nfs_vpcred(vp, ND_READ), &error));
2573 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2574 NFS_LATTR_NOSHRINK));
2575 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2576 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2577 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2578 more_dirs = fxdr_unsigned(int, *tl);
2579
2580 /* loop thru the dir entries, doctoring them to 4bsd form */
2581 while (more_dirs && bigenough) {
2582 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2583 fileno = fxdr_hyper(tl);
2584 len = fxdr_unsigned(int, *(tl + 2));
2585 if (len <= 0 || len > NFS_MAXNAMLEN) {
2586 error = EBADRPC;
2587 m_freem(info.mrep);
2588 info.mrep = NULL;
2589 goto nfsmout;
2590 }
2591 tlen = nfsm_rndup(len);
2592 if (tlen == len)
2593 tlen += 4; /* To ensure null termination*/
2594 left = DIRBLKSIZ - blksiz;
2595 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2596 dp->nfs_reclen += left;
2597 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2598 uiop->uio_iov->iov_len -= left;
2599 uiop->uio_offset += left;
2600 uiop->uio_resid -= left;
2601 blksiz = 0;
2602 }
2603 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2604 bigenough = 0;
2605 if (bigenough) {
2606 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2607 dp->nfs_ino = fileno;
2608 dp->nfs_namlen = len;
2609 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2610 dp->nfs_type = DT_UNKNOWN;
2611 blksiz += dp->nfs_reclen;
2612 if (blksiz == DIRBLKSIZ)
2613 blksiz = 0;
2614 uiop->uio_offset += sizeof(struct nfs_dirent);
2615 uiop->uio_resid -= sizeof(struct nfs_dirent);
2616 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2617 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2618 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2619 nlc.nlc_namelen = len;
2620 ERROROUT(nfsm_mtouio(&info, uiop, len));
2621 cp = uiop->uio_iov->iov_base;
2622 tlen -= len;
2623 *cp = '\0';
2624 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2625 uiop->uio_iov->iov_len -= tlen;
2626 uiop->uio_offset += tlen;
2627 uiop->uio_resid -= tlen;
2628 } else {
2629 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2630 }
2631 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2632 if (bigenough) {
2633 cookie.nfsuquad[0] = *tl++;
2634 cookie.nfsuquad[1] = *tl++;
2635 } else {
2636 tl += 2;
2637 }
2638
2639 /*
2640 * Since the attributes are before the file handle
2641 * (sigh), we must skip over the attributes and then
2642 * come back and get them.
2643 */
2644 attrflag = fxdr_unsigned(int, *tl);
2645 if (attrflag) {
2646 dpossav1 = info.dpos;
2647 mdsav1 = info.md;
2648 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2649 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2650 doit = fxdr_unsigned(int, *tl);
2651 if (doit) {
2652 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2653 }
2654 if (doit && bigenough && !nlcdegenerate(&nlc) &&
2655 !NFS_CMPFH(dnp, fhp, fhsize)
2656 ) {
2657 if (dnch.ncp) {
2658 #if 0
2659 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2660 nlc.nlc_namelen, nlc.nlc_namelen,
2661 nlc.nlc_nameptr);
2662 #endif
2663 /*
2664 * This is a bit hokey but there isn't
2665 * much we can do about it. We can't
2666 * hold the directory vp locked while
2667 * doing lookups and gets.
2668 */
2669 vpls = vn_islocked_unlock(vp);
2670 nch = cache_nlookup(&dnch, &nlc);
2671 cache_setunresolved(&nch);
2672 error = nfs_nget(vp->v_mount, fhp,
2673 fhsize, &np);
2674 vn_islocked_relock(vp, vpls);
2675 if (error == 0) {
2676 newvp = NFSTOV(np);
2677 dpossav2 = info.dpos;
2678 info.dpos = dpossav1;
2679 mdsav2 = info.md;
2680 info.md = mdsav1;
2681 ERROROUT(nfsm_loadattr(&info, newvp,
2682 NULL));
2683 info.dpos = dpossav2;
2684 info.md = mdsav2;
2685 dp->nfs_type =
2686 IFTODT(VTTOIF(np->n_vattr.va_type));
2687 nfs_cache_setvp(&nch, newvp,
2688 nfspos_cache_timeout);
2689 vput(newvp);
2690 newvp = NULLVP;
2691 }
2692 cache_put(&nch);
2693 } else {
2694 kprintf("Warning: NFS/rddirplus, "
2695 "UNABLE TO ENTER %*.*s\n",
2696 nlc.nlc_namelen, nlc.nlc_namelen,
2697 nlc.nlc_nameptr);
2698 }
2699 }
2700 } else {
2701 /* Just skip over the file handle */
2702 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2703 i = fxdr_unsigned(int, *tl);
2704 ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2705 }
2706 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2707 more_dirs = fxdr_unsigned(int, *tl);
2708 }
2709 /*
2710 * If at end of rpc data, get the eof boolean
2711 */
2712 if (!more_dirs) {
2713 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2714 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2715 }
2716 m_freem(info.mrep);
2717 info.mrep = NULL;
2718 }
2719 /*
2720 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2721 * by increasing d_reclen for the last record.
2722 */
2723 if (blksiz > 0) {
2724 left = DIRBLKSIZ - blksiz;
2725 dp->nfs_reclen += left;
2726 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2727 uiop->uio_iov->iov_len -= left;
2728 uiop->uio_offset += left;
2729 uiop->uio_resid -= left;
2730 }
2731
2732 /*
2733 * We are now either at the end of the directory or have filled the
2734 * block.
2735 */
2736 if (bigenough) {
2737 dnp->n_direofoffset = uiop->uio_offset;
2738 } else {
2739 if (uiop->uio_resid > 0)
2740 kprintf("EEK! readdirplusrpc resid > 0\n");
2741 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2742 *cookiep = cookie;
2743 }
2744 nfsmout:
2745 if (newvp != NULLVP) {
2746 if (newvp == vp)
2747 vrele(newvp);
2748 else
2749 vput(newvp);
2750 newvp = NULLVP;
2751 }
2752 if (dnch.ncp)
2753 cache_drop(&dnch);
2754 return (error);
2755 }
2756
2757 /*
2758 * Silly rename. To make the NFS filesystem that is stateless look a little
2759 * more like the "ufs" a remove of an active vnode is translated to a rename
2760 * to a funny looking filename that is removed by nfs_inactive on the
2761 * nfsnode. There is the potential for another process on a different client
2762 * to create the same funny name between the nfs_lookitup() fails and the
2763 * nfs_rename() completes, but...
2764 */
2765 static int
2766 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2767 {
2768 struct sillyrename *sp;
2769 struct nfsnode *np;
2770 int error;
2771
2772 /*
2773 * We previously purged dvp instead of vp. I don't know why, it
2774 * completely destroys performance. We can't do it anyway with the
2775 * new VFS API since we would be breaking the namecache topology.
2776 */
2777 cache_purge(vp); /* XXX */
2778 np = VTONFS(vp);
2779 #ifndef DIAGNOSTIC
2780 if (vp->v_type == VDIR)
2781 panic("nfs: sillyrename dir");
2782 #endif
2783 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2784 M_NFSREQ, M_WAITOK);
2785 sp->s_cred = crdup(cnp->cn_cred);
2786 sp->s_dvp = dvp;
2787 vref(dvp);
2788
2789 /* Fudge together a funny name */
2790 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2791 (int)(intptr_t)cnp->cn_td);
2792
2793 /* Try lookitups until we get one that isn't there */
2794 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2795 cnp->cn_td, NULL) == 0) {
2796 sp->s_name[4]++;
2797 if (sp->s_name[4] > 'z') {
2798 error = EINVAL;
2799 goto bad;
2800 }
2801 }
2802 error = nfs_renameit(dvp, cnp, sp);
2803 if (error)
2804 goto bad;
2805 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2806 cnp->cn_td, &np);
2807 np->n_sillyrename = sp;
2808 return (0);
2809 bad:
2810 vrele(sp->s_dvp);
2811 crfree(sp->s_cred);
2812 kfree((caddr_t)sp, M_NFSREQ);
2813 return (error);
2814 }
2815
2816 /*
2817 * Look up a file name and optionally either update the file handle or
2818 * allocate an nfsnode, depending on the value of npp.
2819 * npp == NULL --> just do the lookup
2820 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2821 * handled too
2822 * *npp != NULL --> update the file handle in the vnode
2823 */
2824 static int
2825 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2826 struct thread *td, struct nfsnode **npp)
2827 {
2828 struct vnode *newvp = NULL;
2829 struct nfsnode *np, *dnp = VTONFS(dvp);
2830 int error = 0, fhlen, attrflag;
2831 nfsfh_t *nfhp;
2832 struct nfsm_info info;
2833
2834 info.mrep = NULL;
2835 info.v3 = NFS_ISV3(dvp);
2836
2837 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2838 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2839 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2840 ERROROUT(nfsm_fhtom(&info, dvp));
2841 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2842 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2843 if (npp && !error) {
2844 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2845 if (*npp) {
2846 np = *npp;
2847 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2848 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2849 np->n_fhp = &np->n_fh;
2850 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2851 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2852 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2853 np->n_fhsize = fhlen;
2854 newvp = NFSTOV(np);
2855 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2856 vref(dvp);
2857 newvp = dvp;
2858 } else {
2859 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2860 if (error) {
2861 m_freem(info.mrep);
2862 info.mrep = NULL;
2863 return (error);
2864 }
2865 newvp = NFSTOV(np);
2866 }
2867 if (info.v3) {
2868 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2869 NFS_LATTR_NOSHRINK));
2870 if (!attrflag && *npp == NULL) {
2871 m_freem(info.mrep);
2872 info.mrep = NULL;
2873 if (newvp == dvp)
2874 vrele(newvp);
2875 else
2876 vput(newvp);
2877 return (ENOENT);
2878 }
2879 } else {
2880 ERROROUT(error = nfsm_loadattr(&info, newvp, NULL));
2881 }
2882 }
2883 m_freem(info.mrep);
2884 info.mrep = NULL;
2885 nfsmout:
2886 if (npp && *npp == NULL) {
2887 if (error) {
2888 if (newvp) {
2889 if (newvp == dvp)
2890 vrele(newvp);
2891 else
2892 vput(newvp);
2893 }
2894 } else
2895 *npp = np;
2896 }
2897 return (error);
2898 }
2899
2900 /*
2901 * Nfs Version 3 commit rpc
2902 *
2903 * We call it 'uio' to distinguish it from 'bio' but there is no real uio
2904 * involved.
2905 */
2906 int
2907 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2908 {
2909 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2910 int error = 0, wccflag = NFSV3_WCCRATTR;
2911 struct nfsm_info info;
2912 u_int32_t *tl;
2913
2914 info.mrep = NULL;
2915 info.v3 = 1;
2916
2917 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2918 return (0);
2919 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2920 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
2921 ERROROUT(nfsm_fhtom(&info, vp));
2922 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
2923 txdr_hyper(offset, tl);
2924 tl += 2;
2925 *tl = txdr_unsigned(cnt);
2926 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
2927 nfs_vpcred(vp, ND_WRITE), &error));
2928 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
2929 if (!error) {
2930 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
2931 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2932 NFSX_V3WRITEVERF)) {
2933 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2934 NFSX_V3WRITEVERF);
2935 error = NFSERR_STALEWRITEVERF;
2936 }
2937 }
2938 m_freem(info.mrep);
2939 info.mrep = NULL;
2940 nfsmout:
2941 return (error);
2942 }
2943
2944 /*
2945 * Kludge City..
2946 * - make nfs_bmap() essentially a no-op that does no translation
2947 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2948 * (Maybe I could use the process's page mapping, but I was concerned that
2949 * Kernel Write might not be enabled and also figured copyout() would do
2950 * a lot more work than bcopy() and also it currently happens in the
2951 * context of the swapper process (2).
2952 *
2953 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
2954 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2955 */
2956 static int
2957 nfs_bmap(struct vop_bmap_args *ap)
2958 {
2959 if (ap->a_doffsetp != NULL)
2960 *ap->a_doffsetp = ap->a_loffset;
2961 if (ap->a_runp != NULL)
2962 *ap->a_runp = 0;
2963 if (ap->a_runb != NULL)
2964 *ap->a_runb = 0;
2965 return (0);
2966 }
2967
2968 /*
2969 * Strategy routine.
2970 */
2971 static int
2972 nfs_strategy(struct vop_strategy_args *ap)
2973 {
2974 struct bio *bio = ap->a_bio;
2975 struct bio *nbio;
2976 struct buf *bp __debugvar = bio->bio_buf;
2977 struct thread *td;
2978 int error;
2979
2980 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2981 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
2982 KASSERT(BUF_REFCNT(bp) > 0,
2983 ("nfs_strategy: buffer %p not locked", bp));
2984
2985 if (bio->bio_flags & BIO_SYNC)
2986 td = curthread; /* XXX */
2987 else
2988 td = NULL;
2989
2990 /*
2991 * We probably don't need to push an nbio any more since no
2992 * block conversion is required due to the use of 64 bit byte
2993 * offsets, but do it anyway.
2994 *
2995 * NOTE: When NFS callers itself via this strategy routines and
2996 * sets up a synchronous I/O, it expects the I/O to run
2997 * synchronously (its bio_done routine just assumes it),
2998 * so for now we have to honor the bit.
2999 */
3000 nbio = push_bio(bio);
3001 nbio->bio_offset = bio->bio_offset;
3002 nbio->bio_flags = bio->bio_flags & BIO_SYNC;
3003
3004 /*
3005 * If the op is asynchronous and an i/o daemon is waiting
3006 * queue the request, wake it up and wait for completion
3007 * otherwise just do it ourselves.
3008 */
3009 if (bio->bio_flags & BIO_SYNC) {
3010 error = nfs_doio(ap->a_vp, nbio, td);
3011 } else {
3012 nfs_asyncio(ap->a_vp, nbio);
3013 error = 0;
3014 }
3015 return (error);
3016 }
3017
3018 /*
3019 * Mmap a file
3020 *
3021 * NB Currently unsupported.
3022 *
3023 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
3024 */
3025 /* ARGSUSED */
3026 static int
3027 nfs_mmap(struct vop_mmap_args *ap)
3028 {
3029 return (EINVAL);
3030 }
3031
3032 /*
3033 * fsync vnode op. Just call nfs_flush() with commit == 1.
3034 *
3035 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3036 */
3037 /* ARGSUSED */
3038 static int
3039 nfs_fsync(struct vop_fsync_args *ap)
3040 {
3041 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
3042 }
3043
3044 /*
3045 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
3046 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
3047 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
3048 * set the buffer contains data that has already been written to the server
3049 * and which now needs a commit RPC.
3050 *
3051 * If commit is 0 we only take one pass and only flush buffers containing new
3052 * dirty data.
3053 *
3054 * If commit is 1 we take two passes, issuing a commit RPC in the second
3055 * pass.
3056 *
3057 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3058 * to completely flush all pending data.
3059 *
3060 * Note that the RB_SCAN code properly handles the case where the
3061 * callback might block and directly or indirectly (another thread) cause
3062 * the RB tree to change.
3063 */
3064
3065 #ifndef NFS_COMMITBVECSIZ
3066 #define NFS_COMMITBVECSIZ 16
3067 #endif
3068
3069 struct nfs_flush_info {
3070 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3071 struct thread *td;
3072 struct vnode *vp;
3073 int waitfor;
3074 int slpflag;
3075 int slptimeo;
3076 int loops;
3077 struct buf *bvary[NFS_COMMITBVECSIZ];
3078 int bvsize;
3079 off_t beg_off;
3080 off_t end_off;
3081 };
3082
3083 static int nfs_flush_bp(struct buf *bp, void *data);
3084 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3085
3086 int
3087 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3088 {
3089 struct nfsnode *np = VTONFS(vp);
3090 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3091 struct nfs_flush_info info;
3092 lwkt_tokref vlock;
3093 int error;
3094
3095 bzero(&info, sizeof(info));
3096 info.td = td;
3097 info.vp = vp;
3098 info.waitfor = waitfor;
3099 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3100 info.loops = 0;
3101 lwkt_gettoken(&vlock, &vp->v_token);
3102
3103 do {
3104 /*
3105 * Flush mode
3106 */
3107 info.mode = NFI_FLUSHNEW;
3108 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3109 nfs_flush_bp, &info);
3110
3111 /*
3112 * Take a second pass if committing and no error occured.
3113 * Clean up any left over collection (whether an error
3114 * occurs or not).
3115 */
3116 if (commit && error == 0) {
3117 info.mode = NFI_COMMIT;
3118 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3119 nfs_flush_bp, &info);
3120 if (info.bvsize)
3121 error = nfs_flush_docommit(&info, error);
3122 }
3123
3124 /*
3125 * Wait for pending I/O to complete before checking whether
3126 * any further dirty buffers exist.
3127 */
3128 while (waitfor == MNT_WAIT &&
3129 bio_track_active(&vp->v_track_write)) {
3130 error = bio_track_wait(&vp->v_track_write,
3131 info.slpflag, info.slptimeo);
3132 if (error) {
3133 /*
3134 * We have to be able to break out if this
3135 * is an 'intr' mount.
3136 */
3137 if (nfs_sigintr(nmp, NULL, td)) {
3138 error = -EINTR;
3139 break;
3140 }
3141
3142 /*
3143 * Since we do not process pending signals,
3144 * once we get a PCATCH our tsleep() will no
3145 * longer sleep, switch to a fixed timeout
3146 * instead.
3147 */
3148 if (info.slpflag == PCATCH) {
3149 info.slpflag = 0;
3150 info.slptimeo = 2 * hz;
3151 }
3152 error = 0;
3153 }
3154 }
3155 ++info.loops;
3156 /*
3157 * Loop if we are flushing synchronous as well as committing,
3158 * and dirty buffers are still present. Otherwise we might livelock.
3159 */
3160 } while (waitfor == MNT_WAIT && commit &&
3161 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3162
3163 /*
3164 * The callbacks have to return a negative error to terminate the
3165 * RB scan.
3166 */
3167 if (error < 0)
3168 error = -error;
3169
3170 /*
3171 * Deal with any error collection
3172 */
3173 if (np->n_flag & NWRITEERR) {
3174 error = np->n_error;
3175 np->n_flag &= ~NWRITEERR;
3176 }
3177 lwkt_reltoken(&vlock);
3178 return (error);
3179 }
3180
3181 static
3182 int
3183 nfs_flush_bp(struct buf *bp, void *data)
3184 {
3185 struct nfs_flush_info *info = data;
3186 int lkflags;
3187 int error;
3188 off_t toff;
3189
3190 error = 0;
3191 switch(info->mode) {
3192 case NFI_FLUSHNEW:
3193 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3194 if (error && info->loops && info->waitfor == MNT_WAIT) {
3195 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3196 if (error) {
3197 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3198 if (info->slpflag & PCATCH)
3199 lkflags |= LK_PCATCH;
3200 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3201 info->slptimeo);
3202 }
3203 }
3204
3205 /*
3206 * Ignore locking errors
3207 */
3208 if (error) {
3209 error = 0;
3210 break;
3211 }
3212
3213 /*
3214 * The buffer may have changed out from under us, even if
3215 * we did not block (MPSAFE). Check again now that it is
3216 * locked.
3217 */
3218 if (bp->b_vp == info->vp &&
3219 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3220 bremfree(bp);
3221 bawrite(bp);
3222 } else {
3223 BUF_UNLOCK(bp);
3224 }
3225 break;
3226 case NFI_COMMIT:
3227 /*
3228 * Only process buffers in need of a commit which we can
3229 * immediately lock. This may prevent a buffer from being
3230 * committed, but the normal flush loop will block on the
3231 * same buffer so we shouldn't get into an endless loop.
3232 */
3233 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3234 (B_DELWRI | B_NEEDCOMMIT)) {
3235 break;
3236 }
3237 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3238 break;
3239
3240 /*
3241 * We must recheck after successfully locking the buffer.
3242 */
3243 if (bp->b_vp != info->vp ||
3244 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3245 (B_DELWRI | B_NEEDCOMMIT)) {
3246 BUF_UNLOCK(bp);
3247 break;
3248 }
3249
3250 /*
3251 * NOTE: storing the bp in the bvary[] basically sets
3252 * it up for a commit operation.
3253 *
3254 * We must call vfs_busy_pages() now so the commit operation
3255 * is interlocked with user modifications to memory mapped
3256 * pages. The b_dirtyoff/b_dirtyend range is not correct
3257 * until after the pages have been busied.
3258 *
3259 * Note: to avoid loopback deadlocks, we do not
3260 * assign b_runningbufspace.
3261 */
3262 bremfree(bp);
3263 bp->b_cmd = BUF_CMD_WRITE;
3264 vfs_busy_pages(bp->b_vp, bp);
3265 info->bvary[info->bvsize] = bp;
3266 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3267 if (info->bvsize == 0 || toff < info->beg_off)
3268 info->beg_off = toff;
3269 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3270 if (info->bvsize == 0 || toff > info->end_off)
3271 info->end_off = toff;
3272 ++info->bvsize;
3273 if (info->bvsize == NFS_COMMITBVECSIZ) {
3274 error = nfs_flush_docommit(info, 0);
3275 KKASSERT(info->bvsize == 0);
3276 }
3277 }
3278 return (error);
3279 }
3280
3281 static
3282 int
3283 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3284 {
3285 struct vnode *vp;
3286 struct buf *bp;
3287 off_t bytes;
3288 int retv;
3289 int i;
3290
3291 vp = info->vp;
3292
3293 if (info->bvsize > 0) {
3294 /*
3295 * Commit data on the server, as required. Note that
3296 * nfs_commit will use the vnode's cred for the commit.
3297 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3298 */
3299 bytes = info->end_off - info->beg_off;
3300 if (bytes > 0x40000000)
3301 bytes = 0x40000000;
3302 if (error) {
3303 retv = -error;
3304 } else {
3305 retv = nfs_commitrpc_uio(vp, info->beg_off,
3306 (int)bytes, info->td);
3307 if (retv == NFSERR_STALEWRITEVERF)
3308 nfs_clearcommit(vp->v_mount);
3309 }
3310
3311 /*
3312 * Now, either mark the blocks I/O done or mark the
3313 * blocks dirty, depending on whether the commit
3314 * succeeded.
3315 */
3316 for (i = 0; i < info->bvsize; ++i) {
3317 bp = info->bvary[i];
3318 if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3319 /*
3320 * Either an error or the original
3321 * vfs_busy_pages() cleared B_NEEDCOMMIT
3322 * due to finding new dirty VM pages in
3323 * the buffer.
3324 *
3325 * Leave B_DELWRI intact.
3326 */
3327 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3328 vfs_unbusy_pages(bp);
3329 bp->b_cmd = BUF_CMD_DONE;
3330 bqrelse(bp);
3331 } else {
3332 /*
3333 * Success, remove B_DELWRI ( bundirty() ).
3334 *
3335 * b_dirtyoff/b_dirtyend seem to be NFS
3336 * specific. We should probably move that
3337 * into bundirty(). XXX
3338 *
3339 * We are faking an I/O write, we have to
3340 * start the transaction in order to
3341 * immediately biodone() it.
3342 */
3343 bundirty(bp);
3344 bp->b_flags &= ~B_ERROR;
3345 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3346 bp->b_dirtyoff = bp->b_dirtyend = 0;
3347 biodone(&bp->b_bio1);
3348 }
3349 }
3350 info->bvsize = 0;
3351 }
3352 return (error);
3353 }
3354
3355 /*
3356 * NFS advisory byte-level locks.
3357 * Currently unsupported.
3358 *
3359 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3360 * int a_flags)
3361 */
3362 static int
3363 nfs_advlock(struct vop_advlock_args *ap)
3364 {
3365 struct nfsnode *np = VTONFS(ap->a_vp);
3366
3367 /*
3368 * The following kludge is to allow diskless support to work
3369 * until a real NFS lockd is implemented. Basically, just pretend
3370 * that this is a local lock.
3371 */
3372 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3373 }
3374
3375 /*
3376 * Print out the contents of an nfsnode.
3377 *
3378 * nfs_print(struct vnode *a_vp)
3379 */
3380 static int
3381 nfs_print(struct vop_print_args *ap)
3382 {
3383 struct vnode *vp = ap->a_vp;
3384 struct nfsnode *np = VTONFS(vp);
3385
3386 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3387 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3388 if (vp->v_type == VFIFO)
3389 fifo_printinfo(vp);
3390 kprintf("\n");
3391 return (0);
3392 }
3393
3394 /*
3395 * nfs special file access vnode op.
3396 *
3397 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3398 */
3399 static int
3400 nfs_laccess(struct vop_access_args *ap)
3401 {
3402 struct vattr vattr;
3403 int error;
3404
3405 error = VOP_GETATTR(ap->a_vp, &vattr);
3406 if (!error)
3407 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3408 vattr.va_mode, 0);
3409 return (error);
3410 }
3411
3412 /*
3413 * Read wrapper for fifos.
3414 *
3415 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3416 * struct ucred *a_cred)
3417 */
3418 static int
3419 nfsfifo_read(struct vop_read_args *ap)
3420 {
3421 struct nfsnode *np = VTONFS(ap->a_vp);
3422
3423 /*
3424 * Set access flag.
3425 */
3426 np->n_flag |= NACC;
3427 getnanotime(&np->n_atim);
3428 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3429 }
3430
3431 /*
3432 * Write wrapper for fifos.
3433 *
3434 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3435 * struct ucred *a_cred)
3436 */
3437 static int
3438 nfsfifo_write(struct vop_write_args *ap)
3439 {
3440 struct nfsnode *np = VTONFS(ap->a_vp);
3441
3442 /*
3443 * Set update flag.
3444 */
3445 np->n_flag |= NUPD;
3446 getnanotime(&np->n_mtim);
3447 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3448 }
3449
3450 /*
3451 * Close wrapper for fifos.
3452 *
3453 * Update the times on the nfsnode then do fifo close.
3454 *
3455 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3456 */
3457 static int
3458 nfsfifo_close(struct vop_close_args *ap)
3459 {
3460 struct vnode *vp = ap->a_vp;
3461 struct nfsnode *np = VTONFS(vp);
3462 struct vattr vattr;
3463 struct timespec ts;
3464
3465 if (np->n_flag & (NACC | NUPD)) {
3466 getnanotime(&ts);
3467 if (np->n_flag & NACC)
3468 np->n_atim = ts;
3469 if (np->n_flag & NUPD)
3470 np->n_mtim = ts;
3471 np->n_flag |= NCHG;
3472 if (vp->v_sysref.refcnt == 1 &&
3473 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3474 VATTR_NULL(&vattr);
3475 if (np->n_flag & NACC)
3476 vattr.va_atime = np->n_atim;
3477 if (np->n_flag & NUPD)
3478 vattr.va_mtime = np->n_mtim;
3479 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3480 }
3481 }
3482 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3483 }
3484
DragonFly BSD