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 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD: releng/8.2/sys/nfsclient/nfs_vnops.c 214612 2010-11-01 01:03:05Z rmacklem $");
37
38 /*
39 * vnode op calls for Sun NFS version 2 and 3
40 */
41
42 #include "opt_inet.h"
43 #include "opt_kdtrace.h"
44
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/systm.h>
48 #include <sys/resourcevar.h>
49 #include <sys/proc.h>
50 #include <sys/mount.h>
51 #include <sys/bio.h>
52 #include <sys/buf.h>
53 #include <sys/jail.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/namei.h>
57 #include <sys/socket.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/fcntl.h>
61 #include <sys/lockf.h>
62 #include <sys/stat.h>
63 #include <sys/sysctl.h>
64 #include <sys/signalvar.h>
65
66 #include <vm/vm.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_extern.h>
69 #include <vm/vm_object.h>
70
71 #include <fs/fifofs/fifo.h>
72
73 #include <nfs/nfsproto.h>
74 #include <nfsclient/nfs.h>
75 #include <nfsclient/nfsnode.h>
76 #include <nfsclient/nfsmount.h>
77 #include <nfsclient/nfs_kdtrace.h>
78 #include <nfs/nfs_lock.h>
79 #include <nfs/xdr_subs.h>
80 #include <nfsclient/nfsm_subs.h>
81
82 #include <net/if.h>
83 #include <netinet/in.h>
84 #include <netinet/in_var.h>
85
86 #include <machine/stdarg.h>
87
88 #ifdef KDTRACE_HOOKS
89 #include <sys/dtrace_bsd.h>
90
91 dtrace_nfsclient_accesscache_flush_probe_func_t
92 dtrace_nfsclient_accesscache_flush_done_probe;
93 uint32_t nfsclient_accesscache_flush_done_id;
94
95 dtrace_nfsclient_accesscache_get_probe_func_t
96 dtrace_nfsclient_accesscache_get_hit_probe,
97 dtrace_nfsclient_accesscache_get_miss_probe;
98 uint32_t nfsclient_accesscache_get_hit_id;
99 uint32_t nfsclient_accesscache_get_miss_id;
100
101 dtrace_nfsclient_accesscache_load_probe_func_t
102 dtrace_nfsclient_accesscache_load_done_probe;
103 uint32_t nfsclient_accesscache_load_done_id;
104 #endif /* !KDTRACE_HOOKS */
105
106 /* Defs */
107 #define TRUE 1
108 #define FALSE 0
109
110 /*
111 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
112 * calls are not in getblk() and brelse() so that they would not be necessary
113 * here.
114 */
115 #ifndef B_VMIO
116 #define vfs_busy_pages(bp, f)
117 #endif
118
119 static vop_read_t nfsfifo_read;
120 static vop_write_t nfsfifo_write;
121 static vop_close_t nfsfifo_close;
122 static int nfs_flush(struct vnode *, int, int);
123 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *);
124 static vop_lookup_t nfs_lookup;
125 static vop_create_t nfs_create;
126 static vop_mknod_t nfs_mknod;
127 static vop_open_t nfs_open;
128 static vop_close_t nfs_close;
129 static vop_access_t nfs_access;
130 static vop_getattr_t nfs_getattr;
131 static vop_setattr_t nfs_setattr;
132 static vop_read_t nfs_read;
133 static vop_fsync_t nfs_fsync;
134 static vop_remove_t nfs_remove;
135 static vop_link_t nfs_link;
136 static vop_rename_t nfs_rename;
137 static vop_mkdir_t nfs_mkdir;
138 static vop_rmdir_t nfs_rmdir;
139 static vop_symlink_t nfs_symlink;
140 static vop_readdir_t nfs_readdir;
141 static vop_strategy_t nfs_strategy;
142 static int nfs_lookitup(struct vnode *, const char *, int,
143 struct ucred *, struct thread *, struct nfsnode **);
144 static int nfs_sillyrename(struct vnode *, struct vnode *,
145 struct componentname *);
146 static vop_access_t nfsspec_access;
147 static vop_readlink_t nfs_readlink;
148 static vop_print_t nfs_print;
149 static vop_advlock_t nfs_advlock;
150 static vop_advlockasync_t nfs_advlockasync;
151
152 /*
153 * Global vfs data structures for nfs
154 */
155 struct vop_vector nfs_vnodeops = {
156 .vop_default = &default_vnodeops,
157 .vop_access = nfs_access,
158 .vop_advlock = nfs_advlock,
159 .vop_advlockasync = nfs_advlockasync,
160 .vop_close = nfs_close,
161 .vop_create = nfs_create,
162 .vop_fsync = nfs_fsync,
163 .vop_getattr = nfs_getattr,
164 .vop_getpages = nfs_getpages,
165 .vop_putpages = nfs_putpages,
166 .vop_inactive = nfs_inactive,
167 .vop_link = nfs_link,
168 .vop_lookup = nfs_lookup,
169 .vop_mkdir = nfs_mkdir,
170 .vop_mknod = nfs_mknod,
171 .vop_open = nfs_open,
172 .vop_print = nfs_print,
173 .vop_read = nfs_read,
174 .vop_readdir = nfs_readdir,
175 .vop_readlink = nfs_readlink,
176 .vop_reclaim = nfs_reclaim,
177 .vop_remove = nfs_remove,
178 .vop_rename = nfs_rename,
179 .vop_rmdir = nfs_rmdir,
180 .vop_setattr = nfs_setattr,
181 .vop_strategy = nfs_strategy,
182 .vop_symlink = nfs_symlink,
183 .vop_write = nfs_write,
184 };
185
186 struct vop_vector nfs_fifoops = {
187 .vop_default = &fifo_specops,
188 .vop_access = nfsspec_access,
189 .vop_close = nfsfifo_close,
190 .vop_fsync = nfs_fsync,
191 .vop_getattr = nfs_getattr,
192 .vop_inactive = nfs_inactive,
193 .vop_print = nfs_print,
194 .vop_read = nfsfifo_read,
195 .vop_reclaim = nfs_reclaim,
196 .vop_setattr = nfs_setattr,
197 .vop_write = nfsfifo_write,
198 };
199
200 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
201 struct componentname *cnp, struct vattr *vap);
202 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
203 struct ucred *cred, struct thread *td);
204 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
205 int fnamelen, struct vnode *tdvp,
206 const char *tnameptr, int tnamelen,
207 struct ucred *cred, struct thread *td);
208 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
209 struct sillyrename *sp);
210
211 /*
212 * Global variables
213 */
214 struct mtx nfs_iod_mtx;
215 enum nfsiod_state nfs_iodwant[NFS_MAXASYNCDAEMON];
216 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
217 int nfs_numasync = 0;
218 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
219
220 SYSCTL_DECL(_vfs_nfs);
221
222 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
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 nfs_prime_access_cache = 0;
227 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
228 &nfs_prime_access_cache, 0,
229 "Prime NFS ACCESS cache when fetching attributes");
230
231 static int nfsv3_commit_on_close = 0;
232 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
233 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
234
235 static int nfs_clean_pages_on_close = 1;
236 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
237 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
238
239 int nfs_directio_enable = 0;
240 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
241 &nfs_directio_enable, 0, "Enable NFS directio");
242
243 /*
244 * This sysctl allows other processes to mmap a file that has been opened
245 * O_DIRECT by a process. In general, having processes mmap the file while
246 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
247 * this by default to prevent DoS attacks - to prevent a malicious user from
248 * opening up files O_DIRECT preventing other users from mmap'ing these
249 * files. "Protected" environments where stricter consistency guarantees are
250 * required can disable this knob. The process that opened the file O_DIRECT
251 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
252 * meaningful.
253 */
254 int nfs_directio_allow_mmap = 1;
255 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
256 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
257
258 #if 0
259 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
260 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
261
262 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
263 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
264 #endif
265
266 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
267 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
268 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
269
270 /*
271 * SMP Locking Note :
272 * The list of locks after the description of the lock is the ordering
273 * of other locks acquired with the lock held.
274 * np->n_mtx : Protects the fields in the nfsnode.
275 VM Object Lock
276 VI_MTX (acquired indirectly)
277 * nmp->nm_mtx : Protects the fields in the nfsmount.
278 rep->r_mtx
279 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
280 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
281 nmp->nm_mtx
282 rep->r_mtx
283 * rep->r_mtx : Protects the fields in an nfsreq.
284 */
285
286 static int
287 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
288 struct ucred *cred, uint32_t *retmode)
289 {
290 const int v3 = 1;
291 u_int32_t *tl;
292 int error = 0, attrflag, i, lrupos;
293
294 struct mbuf *mreq, *mrep, *md, *mb;
295 caddr_t bpos, dpos;
296 u_int32_t rmode;
297 struct nfsnode *np = VTONFS(vp);
298
299 nfsstats.rpccnt[NFSPROC_ACCESS]++;
300 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
301 mb = mreq;
302 bpos = mtod(mb, caddr_t);
303 nfsm_fhtom(vp, v3);
304 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
305 *tl = txdr_unsigned(wmode);
306 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
307 nfsm_postop_attr(vp, attrflag);
308 if (!error) {
309 lrupos = 0;
310 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
311 rmode = fxdr_unsigned(u_int32_t, *tl);
312 mtx_lock(&np->n_mtx);
313 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
314 if (np->n_accesscache[i].uid == cred->cr_uid) {
315 np->n_accesscache[i].mode = rmode;
316 np->n_accesscache[i].stamp = time_second;
317 break;
318 }
319 if (i > 0 && np->n_accesscache[i].stamp <
320 np->n_accesscache[lrupos].stamp)
321 lrupos = i;
322 }
323 if (i == NFS_ACCESSCACHESIZE) {
324 np->n_accesscache[lrupos].uid = cred->cr_uid;
325 np->n_accesscache[lrupos].mode = rmode;
326 np->n_accesscache[lrupos].stamp = time_second;
327 }
328 mtx_unlock(&np->n_mtx);
329 if (retmode != NULL)
330 *retmode = rmode;
331 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
332 }
333 m_freem(mrep);
334 nfsmout:
335 #ifdef KDTRACE_HOOKS
336 if (error) {
337 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
338 error);
339 }
340 #endif
341 return (error);
342 }
343
344 /*
345 * nfs access vnode op.
346 * For nfs version 2, just return ok. File accesses may fail later.
347 * For nfs version 3, use the access rpc to check accessibility. If file modes
348 * are changed on the server, accesses might still fail later.
349 */
350 static int
351 nfs_access(struct vop_access_args *ap)
352 {
353 struct vnode *vp = ap->a_vp;
354 int error = 0, i, gotahit;
355 u_int32_t mode, rmode, wmode;
356 int v3 = NFS_ISV3(vp);
357 struct nfsnode *np = VTONFS(vp);
358
359 /*
360 * Disallow write attempts on filesystems mounted read-only;
361 * unless the file is a socket, fifo, or a block or character
362 * device resident on the filesystem.
363 */
364 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
365 switch (vp->v_type) {
366 case VREG:
367 case VDIR:
368 case VLNK:
369 return (EROFS);
370 default:
371 break;
372 }
373 }
374 /*
375 * For nfs v3, check to see if we have done this recently, and if
376 * so return our cached result instead of making an ACCESS call.
377 * If not, do an access rpc, otherwise you are stuck emulating
378 * ufs_access() locally using the vattr. This may not be correct,
379 * since the server may apply other access criteria such as
380 * client uid-->server uid mapping that we do not know about.
381 */
382 if (v3) {
383 if (ap->a_accmode & VREAD)
384 mode = NFSV3ACCESS_READ;
385 else
386 mode = 0;
387 if (vp->v_type != VDIR) {
388 if (ap->a_accmode & VWRITE)
389 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
390 if (ap->a_accmode & VEXEC)
391 mode |= NFSV3ACCESS_EXECUTE;
392 } else {
393 if (ap->a_accmode & VWRITE)
394 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
395 NFSV3ACCESS_DELETE);
396 if (ap->a_accmode & VEXEC)
397 mode |= NFSV3ACCESS_LOOKUP;
398 }
399 /* XXX safety belt, only make blanket request if caching */
400 if (nfsaccess_cache_timeout > 0) {
401 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
402 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
403 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
404 } else {
405 wmode = mode;
406 }
407
408 /*
409 * Does our cached result allow us to give a definite yes to
410 * this request?
411 */
412 gotahit = 0;
413 mtx_lock(&np->n_mtx);
414 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
415 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
416 if (time_second < (np->n_accesscache[i].stamp +
417 nfsaccess_cache_timeout) &&
418 (np->n_accesscache[i].mode & mode) == mode) {
419 nfsstats.accesscache_hits++;
420 gotahit = 1;
421 }
422 break;
423 }
424 }
425 mtx_unlock(&np->n_mtx);
426 #ifdef KDTRACE_HOOKS
427 if (gotahit)
428 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
429 ap->a_cred->cr_uid, mode);
430 else
431 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
432 ap->a_cred->cr_uid, mode);
433 #endif
434 if (gotahit == 0) {
435 /*
436 * Either a no, or a don't know. Go to the wire.
437 */
438 nfsstats.accesscache_misses++;
439 error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred,
440 &rmode);
441 if (!error) {
442 if ((rmode & mode) != mode)
443 error = EACCES;
444 }
445 }
446 return (error);
447 } else {
448 if ((error = nfsspec_access(ap)) != 0) {
449 return (error);
450 }
451 /*
452 * Attempt to prevent a mapped root from accessing a file
453 * which it shouldn't. We try to read a byte from the file
454 * if the user is root and the file is not zero length.
455 * After calling nfsspec_access, we should have the correct
456 * file size cached.
457 */
458 mtx_lock(&np->n_mtx);
459 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
460 && VTONFS(vp)->n_size > 0) {
461 struct iovec aiov;
462 struct uio auio;
463 char buf[1];
464
465 mtx_unlock(&np->n_mtx);
466 aiov.iov_base = buf;
467 aiov.iov_len = 1;
468 auio.uio_iov = &aiov;
469 auio.uio_iovcnt = 1;
470 auio.uio_offset = 0;
471 auio.uio_resid = 1;
472 auio.uio_segflg = UIO_SYSSPACE;
473 auio.uio_rw = UIO_READ;
474 auio.uio_td = ap->a_td;
475
476 if (vp->v_type == VREG)
477 error = nfs_readrpc(vp, &auio, ap->a_cred);
478 else if (vp->v_type == VDIR) {
479 char* bp;
480 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
481 aiov.iov_base = bp;
482 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
483 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
484 free(bp, M_TEMP);
485 } else if (vp->v_type == VLNK)
486 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
487 else
488 error = EACCES;
489 } else
490 mtx_unlock(&np->n_mtx);
491 return (error);
492 }
493 }
494
495 int nfs_otw_getattr_avoid = 0;
496
497 /*
498 * nfs open vnode op
499 * Check to see if the type is ok
500 * and that deletion is not in progress.
501 * For paged in text files, you will need to flush the page cache
502 * if consistency is lost.
503 */
504 /* ARGSUSED */
505 static int
506 nfs_open(struct vop_open_args *ap)
507 {
508 struct vnode *vp = ap->a_vp;
509 struct nfsnode *np = VTONFS(vp);
510 struct vattr vattr;
511 int error;
512 int fmode = ap->a_mode;
513
514 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
515 return (EOPNOTSUPP);
516
517 /*
518 * Get a valid lease. If cached data is stale, flush it.
519 */
520 mtx_lock(&np->n_mtx);
521 if (np->n_flag & NMODIFIED) {
522 mtx_unlock(&np->n_mtx);
523 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
524 if (error == EINTR || error == EIO)
525 return (error);
526 mtx_lock(&np->n_mtx);
527 np->n_attrstamp = 0;
528 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
529 if (vp->v_type == VDIR)
530 np->n_direofoffset = 0;
531 mtx_unlock(&np->n_mtx);
532 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
533 if (error)
534 return (error);
535 mtx_lock(&np->n_mtx);
536 np->n_mtime = vattr.va_mtime;
537 } else {
538 mtx_unlock(&np->n_mtx);
539 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
540 if (error)
541 return (error);
542 mtx_lock(&np->n_mtx);
543 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
544 if (vp->v_type == VDIR)
545 np->n_direofoffset = 0;
546 mtx_unlock(&np->n_mtx);
547 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
548 if (error == EINTR || error == EIO) {
549 return (error);
550 }
551 mtx_lock(&np->n_mtx);
552 np->n_mtime = vattr.va_mtime;
553 }
554 }
555 /*
556 * If the object has >= 1 O_DIRECT active opens, we disable caching.
557 */
558 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
559 if (np->n_directio_opens == 0) {
560 mtx_unlock(&np->n_mtx);
561 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
562 if (error)
563 return (error);
564 mtx_lock(&np->n_mtx);
565 np->n_flag |= NNONCACHE;
566 }
567 np->n_directio_opens++;
568 }
569 mtx_unlock(&np->n_mtx);
570 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
571 return (0);
572 }
573
574 /*
575 * nfs close vnode op
576 * What an NFS client should do upon close after writing is a debatable issue.
577 * Most NFS clients push delayed writes to the server upon close, basically for
578 * two reasons:
579 * 1 - So that any write errors may be reported back to the client process
580 * doing the close system call. By far the two most likely errors are
581 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
582 * 2 - To put a worst case upper bound on cache inconsistency between
583 * multiple clients for the file.
584 * There is also a consistency problem for Version 2 of the protocol w.r.t.
585 * not being able to tell if other clients are writing a file concurrently,
586 * since there is no way of knowing if the changed modify time in the reply
587 * is only due to the write for this client.
588 * (NFS Version 3 provides weak cache consistency data in the reply that
589 * should be sufficient to detect and handle this case.)
590 *
591 * The current code does the following:
592 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
593 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
594 * or commit them (this satisfies 1 and 2 except for the
595 * case where the server crashes after this close but
596 * before the commit RPC, which is felt to be "good
597 * enough". Changing the last argument to nfs_flush() to
598 * a 1 would force a commit operation, if it is felt a
599 * commit is necessary now.
600 */
601 /* ARGSUSED */
602 static int
603 nfs_close(struct vop_close_args *ap)
604 {
605 struct vnode *vp = ap->a_vp;
606 struct nfsnode *np = VTONFS(vp);
607 int error = 0;
608 int fmode = ap->a_fflag;
609
610 if (vp->v_type == VREG) {
611 /*
612 * Examine and clean dirty pages, regardless of NMODIFIED.
613 * This closes a major hole in close-to-open consistency.
614 * We want to push out all dirty pages (and buffers) on
615 * close, regardless of whether they were dirtied by
616 * mmap'ed writes or via write().
617 */
618 if (nfs_clean_pages_on_close && vp->v_object) {
619 VM_OBJECT_LOCK(vp->v_object);
620 vm_object_page_clean(vp->v_object, 0, 0, 0);
621 VM_OBJECT_UNLOCK(vp->v_object);
622 }
623 mtx_lock(&np->n_mtx);
624 if (np->n_flag & NMODIFIED) {
625 mtx_unlock(&np->n_mtx);
626 if (NFS_ISV3(vp)) {
627 /*
628 * Under NFSv3 we have dirty buffers to dispose of. We
629 * must flush them to the NFS server. We have the option
630 * of waiting all the way through the commit rpc or just
631 * waiting for the initial write. The default is to only
632 * wait through the initial write so the data is in the
633 * server's cache, which is roughly similar to the state
634 * a standard disk subsystem leaves the file in on close().
635 *
636 * We cannot clear the NMODIFIED bit in np->n_flag due to
637 * potential races with other processes, and certainly
638 * cannot clear it if we don't commit.
639 */
640 int cm = nfsv3_commit_on_close ? 1 : 0;
641 error = nfs_flush(vp, MNT_WAIT, cm);
642 /* np->n_flag &= ~NMODIFIED; */
643 } else
644 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
645 mtx_lock(&np->n_mtx);
646 }
647 if (np->n_flag & NWRITEERR) {
648 np->n_flag &= ~NWRITEERR;
649 error = np->n_error;
650 }
651 mtx_unlock(&np->n_mtx);
652 }
653 if (nfs_directio_enable)
654 KASSERT((np->n_directio_asyncwr == 0),
655 ("nfs_close: dirty unflushed (%d) directio buffers\n",
656 np->n_directio_asyncwr));
657 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
658 mtx_lock(&np->n_mtx);
659 KASSERT((np->n_directio_opens > 0),
660 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
661 np->n_directio_opens--;
662 if (np->n_directio_opens == 0)
663 np->n_flag &= ~NNONCACHE;
664 mtx_unlock(&np->n_mtx);
665 }
666 return (error);
667 }
668
669 /*
670 * nfs getattr call from vfs.
671 */
672 static int
673 nfs_getattr(struct vop_getattr_args *ap)
674 {
675 struct vnode *vp = ap->a_vp;
676 struct nfsnode *np = VTONFS(vp);
677 struct thread *td = curthread;
678 struct vattr *vap = ap->a_vap;
679 struct vattr vattr;
680 caddr_t bpos, dpos;
681 int error = 0;
682 struct mbuf *mreq, *mrep, *md, *mb;
683 int v3 = NFS_ISV3(vp);
684
685 /*
686 * Update local times for special files.
687 */
688 mtx_lock(&np->n_mtx);
689 if (np->n_flag & (NACC | NUPD))
690 np->n_flag |= NCHG;
691 mtx_unlock(&np->n_mtx);
692 /*
693 * First look in the cache.
694 */
695 if (nfs_getattrcache(vp, &vattr) == 0)
696 goto nfsmout;
697 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
698 nfsstats.accesscache_misses++;
699 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL);
700 if (nfs_getattrcache(vp, &vattr) == 0)
701 goto nfsmout;
702 }
703 nfsstats.rpccnt[NFSPROC_GETATTR]++;
704 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
705 mb = mreq;
706 bpos = mtod(mb, caddr_t);
707 nfsm_fhtom(vp, v3);
708 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
709 if (!error) {
710 nfsm_loadattr(vp, &vattr);
711 }
712 m_freem(mrep);
713 nfsmout:
714 vap->va_type = vattr.va_type;
715 vap->va_mode = vattr.va_mode;
716 vap->va_nlink = vattr.va_nlink;
717 vap->va_uid = vattr.va_uid;
718 vap->va_gid = vattr.va_gid;
719 vap->va_fsid = vattr.va_fsid;
720 vap->va_fileid = vattr.va_fileid;
721 vap->va_size = vattr.va_size;
722 vap->va_blocksize = vattr.va_blocksize;
723 vap->va_atime = vattr.va_atime;
724 vap->va_mtime = vattr.va_mtime;
725 vap->va_ctime = vattr.va_ctime;
726 vap->va_gen = vattr.va_gen;
727 vap->va_flags = vattr.va_flags;
728 vap->va_rdev = vattr.va_rdev;
729 vap->va_bytes = vattr.va_bytes;
730 vap->va_filerev = vattr.va_filerev;
731
732 return (error);
733 }
734
735 /*
736 * nfs setattr call.
737 */
738 static int
739 nfs_setattr(struct vop_setattr_args *ap)
740 {
741 struct vnode *vp = ap->a_vp;
742 struct nfsnode *np = VTONFS(vp);
743 struct vattr *vap = ap->a_vap;
744 struct thread *td = curthread;
745 int error = 0;
746 u_quad_t tsize;
747
748 #ifndef nolint
749 tsize = (u_quad_t)0;
750 #endif
751
752 /*
753 * Setting of flags is not supported.
754 */
755 if (vap->va_flags != VNOVAL)
756 return (EOPNOTSUPP);
757
758 /*
759 * Disallow write attempts if the filesystem is mounted read-only.
760 */
761 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
762 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
763 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
764 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
765 error = EROFS;
766 goto out;
767 }
768 if (vap->va_size != VNOVAL) {
769 switch (vp->v_type) {
770 case VDIR:
771 return (EISDIR);
772 case VCHR:
773 case VBLK:
774 case VSOCK:
775 case VFIFO:
776 if (vap->va_mtime.tv_sec == VNOVAL &&
777 vap->va_atime.tv_sec == VNOVAL &&
778 vap->va_mode == (mode_t)VNOVAL &&
779 vap->va_uid == (uid_t)VNOVAL &&
780 vap->va_gid == (gid_t)VNOVAL)
781 return (0);
782 vap->va_size = VNOVAL;
783 break;
784 default:
785 /*
786 * Disallow write attempts if the filesystem is
787 * mounted read-only.
788 */
789 if (vp->v_mount->mnt_flag & MNT_RDONLY)
790 return (EROFS);
791 /*
792 * We run vnode_pager_setsize() early (why?),
793 * we must set np->n_size now to avoid vinvalbuf
794 * V_SAVE races that might setsize a lower
795 * value.
796 */
797 mtx_lock(&np->n_mtx);
798 tsize = np->n_size;
799 mtx_unlock(&np->n_mtx);
800 error = nfs_meta_setsize(vp, ap->a_cred, td,
801 vap->va_size);
802 mtx_lock(&np->n_mtx);
803 if (np->n_flag & NMODIFIED) {
804 tsize = np->n_size;
805 mtx_unlock(&np->n_mtx);
806 if (vap->va_size == 0)
807 error = nfs_vinvalbuf(vp, 0, td, 1);
808 else
809 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
810 if (error) {
811 vnode_pager_setsize(vp, tsize);
812 goto out;
813 }
814 } else
815 mtx_unlock(&np->n_mtx);
816 /*
817 * np->n_size has already been set to vap->va_size
818 * in nfs_meta_setsize(). We must set it again since
819 * nfs_loadattrcache() could be called through
820 * nfs_meta_setsize() and could modify np->n_size.
821 */
822 mtx_lock(&np->n_mtx);
823 np->n_vattr.va_size = np->n_size = vap->va_size;
824 mtx_unlock(&np->n_mtx);
825 };
826 } else {
827 mtx_lock(&np->n_mtx);
828 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
829 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
830 mtx_unlock(&np->n_mtx);
831 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
832 (error == EINTR || error == EIO))
833 return error;
834 } else
835 mtx_unlock(&np->n_mtx);
836 }
837 error = nfs_setattrrpc(vp, vap, ap->a_cred);
838 if (error && vap->va_size != VNOVAL) {
839 mtx_lock(&np->n_mtx);
840 np->n_size = np->n_vattr.va_size = tsize;
841 vnode_pager_setsize(vp, tsize);
842 mtx_unlock(&np->n_mtx);
843 }
844 out:
845 return (error);
846 }
847
848 /*
849 * Do an nfs setattr rpc.
850 */
851 static int
852 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
853 {
854 struct nfsv2_sattr *sp;
855 struct nfsnode *np = VTONFS(vp);
856 caddr_t bpos, dpos;
857 u_int32_t *tl;
858 int error = 0, i, wccflag = NFSV3_WCCRATTR;
859 struct mbuf *mreq, *mrep, *md, *mb;
860 int v3 = NFS_ISV3(vp);
861
862 nfsstats.rpccnt[NFSPROC_SETATTR]++;
863 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
864 mb = mreq;
865 bpos = mtod(mb, caddr_t);
866 nfsm_fhtom(vp, v3);
867 if (v3) {
868 nfsm_v3attrbuild(vap, TRUE);
869 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
870 *tl = nfs_false;
871 } else {
872 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
873 if (vap->va_mode == (mode_t)VNOVAL)
874 sp->sa_mode = nfs_xdrneg1;
875 else
876 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
877 if (vap->va_uid == (uid_t)VNOVAL)
878 sp->sa_uid = nfs_xdrneg1;
879 else
880 sp->sa_uid = txdr_unsigned(vap->va_uid);
881 if (vap->va_gid == (gid_t)VNOVAL)
882 sp->sa_gid = nfs_xdrneg1;
883 else
884 sp->sa_gid = txdr_unsigned(vap->va_gid);
885 sp->sa_size = txdr_unsigned(vap->va_size);
886 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
887 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
888 }
889 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
890 if (v3) {
891 mtx_lock(&np->n_mtx);
892 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
893 np->n_accesscache[i].stamp = 0;
894 mtx_unlock(&np->n_mtx);
895 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
896 nfsm_wcc_data(vp, wccflag);
897 } else
898 nfsm_loadattr(vp, NULL);
899 m_freem(mrep);
900 nfsmout:
901 return (error);
902 }
903
904 /*
905 * nfs lookup call, one step at a time...
906 * First look in cache
907 * If not found, unlock the directory nfsnode and do the rpc
908 */
909 static int
910 nfs_lookup(struct vop_lookup_args *ap)
911 {
912 struct componentname *cnp = ap->a_cnp;
913 struct vnode *dvp = ap->a_dvp;
914 struct vnode **vpp = ap->a_vpp;
915 struct mount *mp = dvp->v_mount;
916 struct vattr vattr;
917 struct timespec dmtime;
918 int flags = cnp->cn_flags;
919 struct vnode *newvp;
920 struct nfsmount *nmp;
921 caddr_t bpos, dpos;
922 struct mbuf *mreq, *mrep, *md, *mb;
923 long len;
924 nfsfh_t *fhp;
925 struct nfsnode *np, *newnp;
926 int error = 0, attrflag, fhsize, ltype;
927 int v3 = NFS_ISV3(dvp);
928 struct thread *td = cnp->cn_thread;
929
930 *vpp = NULLVP;
931 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
932 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
933 return (EROFS);
934 if (dvp->v_type != VDIR)
935 return (ENOTDIR);
936 nmp = VFSTONFS(mp);
937 np = VTONFS(dvp);
938 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
939 *vpp = NULLVP;
940 return (error);
941 }
942 error = cache_lookup(dvp, vpp, cnp);
943 if (error > 0 && error != ENOENT)
944 return (error);
945 if (error == -1) {
946 /*
947 * We only accept a positive hit in the cache if the
948 * change time of the file matches our cached copy.
949 * Otherwise, we discard the cache entry and fallback
950 * to doing a lookup RPC.
951 *
952 * To better handle stale file handles and attributes,
953 * clear the attribute cache of this node if it is a
954 * leaf component, part of an open() call, and not
955 * locally modified before fetching the attributes.
956 * This should allow stale file handles to be detected
957 * here where we can fall back to a LOOKUP RPC to
958 * recover rather than having nfs_open() detect the
959 * stale file handle and failing open(2) with ESTALE.
960 */
961 newvp = *vpp;
962 newnp = VTONFS(newvp);
963 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
964 !(newnp->n_flag & NMODIFIED)) {
965 mtx_lock(&newnp->n_mtx);
966 newnp->n_attrstamp = 0;
967 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
968 mtx_unlock(&newnp->n_mtx);
969 }
970 if (VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
971 timespeccmp(&vattr.va_ctime, &newnp->n_ctime, ==)) {
972 nfsstats.lookupcache_hits++;
973 if (cnp->cn_nameiop != LOOKUP &&
974 (flags & ISLASTCN))
975 cnp->cn_flags |= SAVENAME;
976 return (0);
977 }
978 cache_purge(newvp);
979 if (dvp != newvp)
980 vput(newvp);
981 else
982 vrele(newvp);
983 *vpp = NULLVP;
984 } else if (error == ENOENT) {
985 if (dvp->v_iflag & VI_DOOMED)
986 return (ENOENT);
987 /*
988 * We only accept a negative hit in the cache if the
989 * modification time of the parent directory matches
990 * our cached copy. Otherwise, we discard all of the
991 * negative cache entries for this directory. We also
992 * only trust -ve cache entries for less than
993 * nm_negative_namecache_timeout seconds.
994 */
995 if ((u_int)(ticks - np->n_dmtime_ticks) <
996 (nmp->nm_negnametimeo * hz) &&
997 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
998 timespeccmp(&vattr.va_mtime, &np->n_dmtime, ==)) {
999 nfsstats.lookupcache_hits++;
1000 return (ENOENT);
1001 }
1002 cache_purge_negative(dvp);
1003 mtx_lock(&np->n_mtx);
1004 timespecclear(&np->n_dmtime);
1005 mtx_unlock(&np->n_mtx);
1006 }
1007
1008 /*
1009 * Cache the modification time of the parent directory in case
1010 * the lookup fails and results in adding the first negative
1011 * name cache entry for the directory. Since this is reading
1012 * a single time_t, don't bother with locking. The
1013 * modification time may be a bit stale, but it must be read
1014 * before performing the lookup RPC to prevent a race where
1015 * another lookup updates the timestamp on the directory after
1016 * the lookup RPC has been performed on the server but before
1017 * n_dmtime is set at the end of this function.
1018 */
1019 dmtime = np->n_vattr.va_mtime;
1020 error = 0;
1021 newvp = NULLVP;
1022 nfsstats.lookupcache_misses++;
1023 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1024 len = cnp->cn_namelen;
1025 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1026 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1027 mb = mreq;
1028 bpos = mtod(mb, caddr_t);
1029 nfsm_fhtom(dvp, v3);
1030 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1031 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
1032 if (error) {
1033 if (v3) {
1034 nfsm_postop_attr(dvp, attrflag);
1035 m_freem(mrep);
1036 }
1037 goto nfsmout;
1038 }
1039 nfsm_getfh(fhp, fhsize, v3);
1040
1041 /*
1042 * Handle RENAME case...
1043 */
1044 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1045 if (NFS_CMPFH(np, fhp, fhsize)) {
1046 m_freem(mrep);
1047 return (EISDIR);
1048 }
1049 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE);
1050 if (error) {
1051 m_freem(mrep);
1052 return (error);
1053 }
1054 newvp = NFSTOV(np);
1055 if (v3) {
1056 nfsm_postop_attr(newvp, attrflag);
1057 nfsm_postop_attr(dvp, attrflag);
1058 } else
1059 nfsm_loadattr(newvp, NULL);
1060 *vpp = newvp;
1061 m_freem(mrep);
1062 cnp->cn_flags |= SAVENAME;
1063 return (0);
1064 }
1065
1066 if (flags & ISDOTDOT) {
1067 ltype = VOP_ISLOCKED(dvp);
1068 error = vfs_busy(mp, MBF_NOWAIT);
1069 if (error != 0) {
1070 vfs_ref(mp);
1071 VOP_UNLOCK(dvp, 0);
1072 error = vfs_busy(mp, 0);
1073 vn_lock(dvp, ltype | LK_RETRY);
1074 vfs_rel(mp);
1075 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1076 vfs_unbusy(mp);
1077 error = ENOENT;
1078 }
1079 if (error != 0) {
1080 m_freem(mrep);
1081 return (error);
1082 }
1083 }
1084 VOP_UNLOCK(dvp, 0);
1085 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1086 if (error == 0)
1087 newvp = NFSTOV(np);
1088 vfs_unbusy(mp);
1089 if (newvp != dvp)
1090 vn_lock(dvp, ltype | LK_RETRY);
1091 if (dvp->v_iflag & VI_DOOMED) {
1092 if (error == 0) {
1093 if (newvp == dvp)
1094 vrele(newvp);
1095 else
1096 vput(newvp);
1097 }
1098 error = ENOENT;
1099 }
1100 if (error) {
1101 m_freem(mrep);
1102 return (error);
1103 }
1104 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1105 VREF(dvp);
1106 newvp = dvp;
1107 } else {
1108 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1109 if (error) {
1110 m_freem(mrep);
1111 return (error);
1112 }
1113 newvp = NFSTOV(np);
1114
1115 /*
1116 * Flush the attribute cache when opening a leaf node
1117 * to ensure that fresh attributes are fetched in
1118 * nfs_open() if we are unable to fetch attributes
1119 * from the LOOKUP reply.
1120 */
1121 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1122 !(np->n_flag & NMODIFIED)) {
1123 mtx_lock(&np->n_mtx);
1124 np->n_attrstamp = 0;
1125 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1126 mtx_unlock(&np->n_mtx);
1127 }
1128 }
1129 if (v3) {
1130 nfsm_postop_attr(newvp, attrflag);
1131 nfsm_postop_attr(dvp, attrflag);
1132 } else
1133 nfsm_loadattr(newvp, NULL);
1134 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1135 cnp->cn_flags |= SAVENAME;
1136 if ((cnp->cn_flags & MAKEENTRY) &&
1137 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1138 np->n_ctime = np->n_vattr.va_ctime;
1139 cache_enter(dvp, newvp, cnp);
1140 }
1141 *vpp = newvp;
1142 m_freem(mrep);
1143 nfsmout:
1144 if (error) {
1145 if (newvp != NULLVP) {
1146 vput(newvp);
1147 *vpp = NULLVP;
1148 }
1149
1150 if (error != ENOENT)
1151 goto done;
1152
1153 /* The requested file was not found. */
1154 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1155 (flags & ISLASTCN)) {
1156 /*
1157 * XXX: UFS does a full VOP_ACCESS(dvp,
1158 * VWRITE) here instead of just checking
1159 * MNT_RDONLY.
1160 */
1161 if (mp->mnt_flag & MNT_RDONLY)
1162 return (EROFS);
1163 cnp->cn_flags |= SAVENAME;
1164 return (EJUSTRETURN);
1165 }
1166
1167 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1168 /*
1169 * Maintain n_dmtime as the modification time
1170 * of the parent directory when the oldest -ve
1171 * name cache entry for this directory was
1172 * added. If a -ve cache entry has already
1173 * been added with a newer modification time
1174 * by a concurrent lookup, then don't bother
1175 * adding a cache entry. The modification
1176 * time of the directory might have changed
1177 * due to the file this lookup failed to find
1178 * being created. In that case a subsequent
1179 * lookup would incorrectly use the entry
1180 * added here instead of doing an extra
1181 * lookup.
1182 */
1183 mtx_lock(&np->n_mtx);
1184 if (timespeccmp(&np->n_dmtime, &dmtime, <=)) {
1185 if (!timespecisset(&np->n_dmtime)) {
1186 np->n_dmtime = dmtime;
1187 np->n_dmtime_ticks = ticks;
1188 }
1189 mtx_unlock(&np->n_mtx);
1190 cache_enter(dvp, NULL, cnp);
1191 } else
1192 mtx_unlock(&np->n_mtx);
1193 }
1194 return (ENOENT);
1195 }
1196 done:
1197 return (error);
1198 }
1199
1200 /*
1201 * nfs read call.
1202 * Just call nfs_bioread() to do the work.
1203 */
1204 static int
1205 nfs_read(struct vop_read_args *ap)
1206 {
1207 struct vnode *vp = ap->a_vp;
1208
1209 switch (vp->v_type) {
1210 case VREG:
1211 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1212 case VDIR:
1213 return (EISDIR);
1214 default:
1215 return (EOPNOTSUPP);
1216 }
1217 }
1218
1219 /*
1220 * nfs readlink call
1221 */
1222 static int
1223 nfs_readlink(struct vop_readlink_args *ap)
1224 {
1225 struct vnode *vp = ap->a_vp;
1226
1227 if (vp->v_type != VLNK)
1228 return (EINVAL);
1229 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1230 }
1231
1232 /*
1233 * Do a readlink rpc.
1234 * Called by nfs_doio() from below the buffer cache.
1235 */
1236 int
1237 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1238 {
1239 caddr_t bpos, dpos;
1240 int error = 0, len, attrflag;
1241 struct mbuf *mreq, *mrep, *md, *mb;
1242 int v3 = NFS_ISV3(vp);
1243
1244 nfsstats.rpccnt[NFSPROC_READLINK]++;
1245 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1246 mb = mreq;
1247 bpos = mtod(mb, caddr_t);
1248 nfsm_fhtom(vp, v3);
1249 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1250 if (v3)
1251 nfsm_postop_attr(vp, attrflag);
1252 if (!error) {
1253 nfsm_strsiz(len, NFS_MAXPATHLEN);
1254 if (len == NFS_MAXPATHLEN) {
1255 struct nfsnode *np = VTONFS(vp);
1256 mtx_lock(&np->n_mtx);
1257 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1258 len = np->n_size;
1259 mtx_unlock(&np->n_mtx);
1260 }
1261 nfsm_mtouio(uiop, len);
1262 }
1263 m_freem(mrep);
1264 nfsmout:
1265 return (error);
1266 }
1267
1268 /*
1269 * nfs read rpc call
1270 * Ditto above
1271 */
1272 int
1273 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1274 {
1275 u_int32_t *tl;
1276 caddr_t bpos, dpos;
1277 struct mbuf *mreq, *mrep, *md, *mb;
1278 struct nfsmount *nmp;
1279 int error = 0, len, retlen, tsiz, eof, attrflag;
1280 int v3 = NFS_ISV3(vp);
1281 int rsize;
1282
1283 #ifndef nolint
1284 eof = 0;
1285 #endif
1286 nmp = VFSTONFS(vp->v_mount);
1287 tsiz = uiop->uio_resid;
1288 mtx_lock(&nmp->nm_mtx);
1289 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1290 mtx_unlock(&nmp->nm_mtx);
1291 return (EFBIG);
1292 }
1293 rsize = nmp->nm_rsize;
1294 mtx_unlock(&nmp->nm_mtx);
1295 while (tsiz > 0) {
1296 nfsstats.rpccnt[NFSPROC_READ]++;
1297 len = (tsiz > rsize) ? rsize : tsiz;
1298 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1299 mb = mreq;
1300 bpos = mtod(mb, caddr_t);
1301 nfsm_fhtom(vp, v3);
1302 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1303 if (v3) {
1304 txdr_hyper(uiop->uio_offset, tl);
1305 *(tl + 2) = txdr_unsigned(len);
1306 } else {
1307 *tl++ = txdr_unsigned(uiop->uio_offset);
1308 *tl++ = txdr_unsigned(len);
1309 *tl = 0;
1310 }
1311 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1312 if (v3) {
1313 nfsm_postop_attr(vp, attrflag);
1314 if (error) {
1315 m_freem(mrep);
1316 goto nfsmout;
1317 }
1318 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1319 eof = fxdr_unsigned(int, *(tl + 1));
1320 } else {
1321 nfsm_loadattr(vp, NULL);
1322 }
1323 nfsm_strsiz(retlen, rsize);
1324 nfsm_mtouio(uiop, retlen);
1325 m_freem(mrep);
1326 tsiz -= retlen;
1327 if (v3) {
1328 if (eof || retlen == 0) {
1329 tsiz = 0;
1330 }
1331 } else if (retlen < len) {
1332 tsiz = 0;
1333 }
1334 }
1335 nfsmout:
1336 return (error);
1337 }
1338
1339 /*
1340 * nfs write call
1341 */
1342 int
1343 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1344 int *iomode, int *must_commit)
1345 {
1346 u_int32_t *tl;
1347 int32_t backup;
1348 caddr_t bpos, dpos;
1349 struct mbuf *mreq, *mrep, *md, *mb;
1350 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1351 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1352 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1353 int wsize;
1354
1355 KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1"));
1356 *must_commit = 0;
1357 tsiz = uiop->uio_resid;
1358 mtx_lock(&nmp->nm_mtx);
1359 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1360 mtx_unlock(&nmp->nm_mtx);
1361 return (EFBIG);
1362 }
1363 wsize = nmp->nm_wsize;
1364 mtx_unlock(&nmp->nm_mtx);
1365 while (tsiz > 0) {
1366 nfsstats.rpccnt[NFSPROC_WRITE]++;
1367 len = (tsiz > wsize) ? wsize : tsiz;
1368 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1369 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1370 mb = mreq;
1371 bpos = mtod(mb, caddr_t);
1372 nfsm_fhtom(vp, v3);
1373 if (v3) {
1374 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1375 txdr_hyper(uiop->uio_offset, tl);
1376 tl += 2;
1377 *tl++ = txdr_unsigned(len);
1378 *tl++ = txdr_unsigned(*iomode);
1379 *tl = txdr_unsigned(len);
1380 } else {
1381 u_int32_t x;
1382
1383 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1384 /* Set both "begin" and "current" to non-garbage. */
1385 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1386 *tl++ = x; /* "begin offset" */
1387 *tl++ = x; /* "current offset" */
1388 x = txdr_unsigned(len);
1389 *tl++ = x; /* total to this offset */
1390 *tl = x; /* size of this write */
1391 }
1392 nfsm_uiotom(uiop, len);
1393 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1394 if (v3) {
1395 wccflag = NFSV3_WCCCHK;
1396 nfsm_wcc_data(vp, wccflag);
1397 if (!error) {
1398 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1399 + NFSX_V3WRITEVERF);
1400 rlen = fxdr_unsigned(int, *tl++);
1401 if (rlen == 0) {
1402 error = NFSERR_IO;
1403 m_freem(mrep);
1404 break;
1405 } else if (rlen < len) {
1406 backup = len - rlen;
1407 uiop->uio_iov->iov_base =
1408 (char *)uiop->uio_iov->iov_base -
1409 backup;
1410 uiop->uio_iov->iov_len += backup;
1411 uiop->uio_offset -= backup;
1412 uiop->uio_resid += backup;
1413 len = rlen;
1414 }
1415 commit = fxdr_unsigned(int, *tl++);
1416
1417 /*
1418 * Return the lowest committment level
1419 * obtained by any of the RPCs.
1420 */
1421 if (committed == NFSV3WRITE_FILESYNC)
1422 committed = commit;
1423 else if (committed == NFSV3WRITE_DATASYNC &&
1424 commit == NFSV3WRITE_UNSTABLE)
1425 committed = commit;
1426 mtx_lock(&nmp->nm_mtx);
1427 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1428 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1429 NFSX_V3WRITEVERF);
1430 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1431 } else if (bcmp((caddr_t)tl,
1432 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1433 *must_commit = 1;
1434 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1435 NFSX_V3WRITEVERF);
1436 }
1437 mtx_unlock(&nmp->nm_mtx);
1438 }
1439 } else {
1440 nfsm_loadattr(vp, NULL);
1441 }
1442 if (wccflag) {
1443 mtx_lock(&(VTONFS(vp))->n_mtx);
1444 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1445 mtx_unlock(&(VTONFS(vp))->n_mtx);
1446 }
1447 m_freem(mrep);
1448 if (error)
1449 break;
1450 tsiz -= len;
1451 }
1452 nfsmout:
1453 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1454 committed = NFSV3WRITE_FILESYNC;
1455 *iomode = committed;
1456 if (error)
1457 uiop->uio_resid = tsiz;
1458 return (error);
1459 }
1460
1461 /*
1462 * nfs mknod rpc
1463 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1464 * mode set to specify the file type and the size field for rdev.
1465 */
1466 static int
1467 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1468 struct vattr *vap)
1469 {
1470 struct nfsv2_sattr *sp;
1471 u_int32_t *tl;
1472 struct vnode *newvp = NULL;
1473 struct nfsnode *np = NULL;
1474 struct vattr vattr;
1475 caddr_t bpos, dpos;
1476 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1477 struct mbuf *mreq, *mrep, *md, *mb;
1478 u_int32_t rdev;
1479 int v3 = NFS_ISV3(dvp);
1480
1481 if (vap->va_type == VCHR || vap->va_type == VBLK)
1482 rdev = txdr_unsigned(vap->va_rdev);
1483 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1484 rdev = nfs_xdrneg1;
1485 else {
1486 return (EOPNOTSUPP);
1487 }
1488 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1489 return (error);
1490 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1491 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1492 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1493 mb = mreq;
1494 bpos = mtod(mb, caddr_t);
1495 nfsm_fhtom(dvp, v3);
1496 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1497 if (v3) {
1498 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1499 *tl++ = vtonfsv3_type(vap->va_type);
1500 nfsm_v3attrbuild(vap, FALSE);
1501 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1502 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1503 *tl++ = txdr_unsigned(major(vap->va_rdev));
1504 *tl = txdr_unsigned(minor(vap->va_rdev));
1505 }
1506 } else {
1507 sp = nfsm_build(struct nfsv2_sattr *, 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 = rdev;
1512 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1513 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1514 }
1515 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1516 if (!error) {
1517 nfsm_mtofh(dvp, newvp, v3, gotvp);
1518 if (!gotvp) {
1519 if (newvp) {
1520 vput(newvp);
1521 newvp = NULL;
1522 }
1523 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1524 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1525 if (!error)
1526 newvp = NFSTOV(np);
1527 }
1528 }
1529 if (v3)
1530 nfsm_wcc_data(dvp, wccflag);
1531 m_freem(mrep);
1532 nfsmout:
1533 if (error) {
1534 if (newvp)
1535 vput(newvp);
1536 } else {
1537 if (cnp->cn_flags & MAKEENTRY)
1538 cache_enter(dvp, newvp, cnp);
1539 *vpp = newvp;
1540 }
1541 mtx_lock(&(VTONFS(dvp))->n_mtx);
1542 VTONFS(dvp)->n_flag |= NMODIFIED;
1543 if (!wccflag) {
1544 VTONFS(dvp)->n_attrstamp = 0;
1545 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1546 }
1547 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1548 return (error);
1549 }
1550
1551 /*
1552 * nfs mknod vop
1553 * just call nfs_mknodrpc() to do the work.
1554 */
1555 /* ARGSUSED */
1556 static int
1557 nfs_mknod(struct vop_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 static int
1567 nfs_create(struct vop_create_args *ap)
1568 {
1569 struct vnode *dvp = ap->a_dvp;
1570 struct vattr *vap = ap->a_vap;
1571 struct componentname *cnp = ap->a_cnp;
1572 struct nfsv2_sattr *sp;
1573 u_int32_t *tl;
1574 struct nfsnode *np = NULL;
1575 struct vnode *newvp = NULL;
1576 caddr_t bpos, dpos;
1577 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1578 struct mbuf *mreq, *mrep, *md, *mb;
1579 struct vattr vattr;
1580 int v3 = NFS_ISV3(dvp);
1581
1582 /*
1583 * Oops, not for me..
1584 */
1585 if (vap->va_type == VSOCK) {
1586 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1587 return (error);
1588 }
1589
1590 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) {
1591 return (error);
1592 }
1593 if (vap->va_vaflags & VA_EXCLUSIVE)
1594 fmode |= O_EXCL;
1595 again:
1596 nfsstats.rpccnt[NFSPROC_CREATE]++;
1597 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1598 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1599 mb = mreq;
1600 bpos = mtod(mb, caddr_t);
1601 nfsm_fhtom(dvp, v3);
1602 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1603 if (v3) {
1604 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1605 if (fmode & O_EXCL) {
1606 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1607 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1608 #ifdef INET
1609 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1610 IN_IFADDR_RLOCK();
1611 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1612 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1613 else
1614 #endif
1615 *tl++ = create_verf;
1616 #ifdef INET
1617 IN_IFADDR_RUNLOCK();
1618 CURVNET_RESTORE();
1619 #endif
1620 *tl = ++create_verf;
1621 } else {
1622 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1623 nfsm_v3attrbuild(vap, FALSE);
1624 }
1625 } else {
1626 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1627 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1628 sp->sa_uid = nfs_xdrneg1;
1629 sp->sa_gid = nfs_xdrneg1;
1630 sp->sa_size = 0;
1631 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1632 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1633 }
1634 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1635 if (!error) {
1636 nfsm_mtofh(dvp, newvp, v3, gotvp);
1637 if (!gotvp) {
1638 if (newvp) {
1639 vput(newvp);
1640 newvp = NULL;
1641 }
1642 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1643 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1644 if (!error)
1645 newvp = NFSTOV(np);
1646 }
1647 }
1648 if (v3)
1649 nfsm_wcc_data(dvp, wccflag);
1650 m_freem(mrep);
1651 nfsmout:
1652 if (error) {
1653 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1654 fmode &= ~O_EXCL;
1655 goto again;
1656 }
1657 if (newvp)
1658 vput(newvp);
1659 } else if (v3 && (fmode & O_EXCL)) {
1660 /*
1661 * We are normally called with only a partially initialized
1662 * VAP. Since the NFSv3 spec says that server may use the
1663 * file attributes to store the verifier, the spec requires
1664 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1665 * in atime, but we can't really assume that all servers will
1666 * so we ensure that our SETATTR sets both atime and mtime.
1667 */
1668 if (vap->va_mtime.tv_sec == VNOVAL)
1669 vfs_timestamp(&vap->va_mtime);
1670 if (vap->va_atime.tv_sec == VNOVAL)
1671 vap->va_atime = vap->va_mtime;
1672 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred);
1673 if (error)
1674 vput(newvp);
1675 }
1676 if (!error) {
1677 if (cnp->cn_flags & MAKEENTRY)
1678 cache_enter(dvp, newvp, cnp);
1679 *ap->a_vpp = newvp;
1680 }
1681 mtx_lock(&(VTONFS(dvp))->n_mtx);
1682 VTONFS(dvp)->n_flag |= NMODIFIED;
1683 if (!wccflag) {
1684 VTONFS(dvp)->n_attrstamp = 0;
1685 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1686 }
1687 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1688 return (error);
1689 }
1690
1691 /*
1692 * nfs file remove call
1693 * To try and make nfs semantics closer to ufs semantics, a file that has
1694 * other processes using the vnode is renamed instead of removed and then
1695 * removed later on the last close.
1696 * - If v_usecount > 1
1697 * If a rename is not already in the works
1698 * call nfs_sillyrename() to set it up
1699 * else
1700 * do the remove rpc
1701 */
1702 static int
1703 nfs_remove(struct vop_remove_args *ap)
1704 {
1705 struct vnode *vp = ap->a_vp;
1706 struct vnode *dvp = ap->a_dvp;
1707 struct componentname *cnp = ap->a_cnp;
1708 struct nfsnode *np = VTONFS(vp);
1709 int error = 0;
1710 struct vattr vattr;
1711
1712 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1713 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1714 if (vp->v_type == VDIR)
1715 error = EPERM;
1716 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1717 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1718 /*
1719 * Purge the name cache so that the chance of a lookup for
1720 * the name succeeding while the remove is in progress is
1721 * minimized. Without node locking it can still happen, such
1722 * that an I/O op returns ESTALE, but since you get this if
1723 * another host removes the file..
1724 */
1725 cache_purge(vp);
1726 /*
1727 * throw away biocache buffers, mainly to avoid
1728 * unnecessary delayed writes later.
1729 */
1730 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1731 /* Do the rpc */
1732 if (error != EINTR && error != EIO)
1733 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1734 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1735 /*
1736 * Kludge City: If the first reply to the remove rpc is lost..
1737 * the reply to the retransmitted request will be ENOENT
1738 * since the file was in fact removed
1739 * Therefore, we cheat and return success.
1740 */
1741 if (error == ENOENT)
1742 error = 0;
1743 } else if (!np->n_sillyrename)
1744 error = nfs_sillyrename(dvp, vp, cnp);
1745 mtx_lock(&np->n_mtx);
1746 np->n_attrstamp = 0;
1747 mtx_unlock(&np->n_mtx);
1748 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1749 return (error);
1750 }
1751
1752 /*
1753 * nfs file remove rpc called from nfs_inactive
1754 */
1755 int
1756 nfs_removeit(struct sillyrename *sp)
1757 {
1758 /*
1759 * Make sure that the directory vnode is still valid.
1760 * XXX we should lock sp->s_dvp here.
1761 */
1762 if (sp->s_dvp->v_type == VBAD)
1763 return (0);
1764 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1765 NULL));
1766 }
1767
1768 /*
1769 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1770 */
1771 static int
1772 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1773 struct ucred *cred, struct thread *td)
1774 {
1775 caddr_t bpos, dpos;
1776 int error = 0, wccflag = NFSV3_WCCRATTR;
1777 struct mbuf *mreq, *mrep, *md, *mb;
1778 int v3 = NFS_ISV3(dvp);
1779
1780 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1781 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1782 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1783 mb = mreq;
1784 bpos = mtod(mb, caddr_t);
1785 nfsm_fhtom(dvp, v3);
1786 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1787 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1788 if (v3)
1789 nfsm_wcc_data(dvp, wccflag);
1790 m_freem(mrep);
1791 nfsmout:
1792 mtx_lock(&(VTONFS(dvp))->n_mtx);
1793 VTONFS(dvp)->n_flag |= NMODIFIED;
1794 if (!wccflag) {
1795 VTONFS(dvp)->n_attrstamp = 0;
1796 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1797 }
1798 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1799 return (error);
1800 }
1801
1802 /*
1803 * nfs file rename call
1804 */
1805 static int
1806 nfs_rename(struct vop_rename_args *ap)
1807 {
1808 struct vnode *fvp = ap->a_fvp;
1809 struct vnode *tvp = ap->a_tvp;
1810 struct vnode *fdvp = ap->a_fdvp;
1811 struct vnode *tdvp = ap->a_tdvp;
1812 struct componentname *tcnp = ap->a_tcnp;
1813 struct componentname *fcnp = ap->a_fcnp;
1814 int error;
1815
1816 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1817 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1818 /* Check for cross-device rename */
1819 if ((fvp->v_mount != tdvp->v_mount) ||
1820 (tvp && (fvp->v_mount != tvp->v_mount))) {
1821 error = EXDEV;
1822 goto out;
1823 }
1824
1825 if (fvp == tvp) {
1826 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1827 error = 0;
1828 goto out;
1829 }
1830 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1831 goto out;
1832
1833 /*
1834 * We have to flush B_DELWRI data prior to renaming
1835 * the file. If we don't, the delayed-write buffers
1836 * can be flushed out later after the file has gone stale
1837 * under NFSV3. NFSV2 does not have this problem because
1838 * ( as far as I can tell ) it flushes dirty buffers more
1839 * often.
1840 *
1841 * Skip the rename operation if the fsync fails, this can happen
1842 * due to the server's volume being full, when we pushed out data
1843 * that was written back to our cache earlier. Not checking for
1844 * this condition can result in potential (silent) data loss.
1845 */
1846 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1847 VOP_UNLOCK(fvp, 0);
1848 if (!error && tvp)
1849 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1850 if (error)
1851 goto out;
1852
1853 /*
1854 * If the tvp exists and is in use, sillyrename it before doing the
1855 * rename of the new file over it.
1856 * XXX Can't sillyrename a directory.
1857 */
1858 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1859 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1860 vput(tvp);
1861 tvp = NULL;
1862 }
1863
1864 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1865 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1866 tcnp->cn_thread);
1867
1868 if (fvp->v_type == VDIR) {
1869 if (tvp != NULL && tvp->v_type == VDIR)
1870 cache_purge(tdvp);
1871 cache_purge(fdvp);
1872 }
1873
1874 out:
1875 if (tdvp == tvp)
1876 vrele(tdvp);
1877 else
1878 vput(tdvp);
1879 if (tvp)
1880 vput(tvp);
1881 vrele(fdvp);
1882 vrele(fvp);
1883 /*
1884 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1885 */
1886 if (error == ENOENT)
1887 error = 0;
1888 return (error);
1889 }
1890
1891 /*
1892 * nfs file rename rpc called from nfs_remove() above
1893 */
1894 static int
1895 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1896 struct sillyrename *sp)
1897 {
1898
1899 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1900 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1901 }
1902
1903 /*
1904 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1905 */
1906 static int
1907 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1908 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1909 struct thread *td)
1910 {
1911 caddr_t bpos, dpos;
1912 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1913 struct mbuf *mreq, *mrep, *md, *mb;
1914 int v3 = NFS_ISV3(fdvp);
1915
1916 nfsstats.rpccnt[NFSPROC_RENAME]++;
1917 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1918 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1919 nfsm_rndup(tnamelen));
1920 mb = mreq;
1921 bpos = mtod(mb, caddr_t);
1922 nfsm_fhtom(fdvp, v3);
1923 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1924 nfsm_fhtom(tdvp, v3);
1925 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1926 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1927 if (v3) {
1928 nfsm_wcc_data(fdvp, fwccflag);
1929 nfsm_wcc_data(tdvp, twccflag);
1930 }
1931 m_freem(mrep);
1932 nfsmout:
1933 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1934 VTONFS(fdvp)->n_flag |= NMODIFIED;
1935 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1936 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1937 VTONFS(tdvp)->n_flag |= NMODIFIED;
1938 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1939 if (!fwccflag) {
1940 VTONFS(fdvp)->n_attrstamp = 0;
1941 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1942 }
1943 if (!twccflag) {
1944 VTONFS(tdvp)->n_attrstamp = 0;
1945 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1946 }
1947 return (error);
1948 }
1949
1950 /*
1951 * nfs hard link create call
1952 */
1953 static int
1954 nfs_link(struct vop_link_args *ap)
1955 {
1956 struct vnode *vp = ap->a_vp;
1957 struct vnode *tdvp = ap->a_tdvp;
1958 struct componentname *cnp = ap->a_cnp;
1959 caddr_t bpos, dpos;
1960 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1961 struct mbuf *mreq, *mrep, *md, *mb;
1962 int v3;
1963
1964 if (vp->v_mount != tdvp->v_mount) {
1965 return (EXDEV);
1966 }
1967
1968 /*
1969 * Push all writes to the server, so that the attribute cache
1970 * doesn't get "out of sync" with the server.
1971 * XXX There should be a better way!
1972 */
1973 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1974
1975 v3 = NFS_ISV3(vp);
1976 nfsstats.rpccnt[NFSPROC_LINK]++;
1977 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1978 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1979 mb = mreq;
1980 bpos = mtod(mb, caddr_t);
1981 nfsm_fhtom(vp, v3);
1982 nfsm_fhtom(tdvp, v3);
1983 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1984 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1985 if (v3) {
1986 nfsm_postop_attr(vp, attrflag);
1987 nfsm_wcc_data(tdvp, wccflag);
1988 }
1989 m_freem(mrep);
1990 nfsmout:
1991 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1992 VTONFS(tdvp)->n_flag |= NMODIFIED;
1993 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1994 if (!attrflag) {
1995 VTONFS(vp)->n_attrstamp = 0;
1996 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1997 }
1998 if (!wccflag) {
1999 VTONFS(tdvp)->n_attrstamp = 0;
2000 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2001 }
2002 return (error);
2003 }
2004
2005 /*
2006 * nfs symbolic link create call
2007 */
2008 static int
2009 nfs_symlink(struct vop_symlink_args *ap)
2010 {
2011 struct vnode *dvp = ap->a_dvp;
2012 struct vattr *vap = ap->a_vap;
2013 struct componentname *cnp = ap->a_cnp;
2014 struct nfsv2_sattr *sp;
2015 caddr_t bpos, dpos;
2016 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2017 struct mbuf *mreq, *mrep, *md, *mb;
2018 struct vnode *newvp = NULL;
2019 int v3 = NFS_ISV3(dvp);
2020
2021 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2022 slen = strlen(ap->a_target);
2023 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2024 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2025 mb = mreq;
2026 bpos = mtod(mb, caddr_t);
2027 nfsm_fhtom(dvp, v3);
2028 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2029 if (v3) {
2030 nfsm_v3attrbuild(vap, FALSE);
2031 }
2032 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2033 if (!v3) {
2034 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2035 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2036 sp->sa_uid = nfs_xdrneg1;
2037 sp->sa_gid = nfs_xdrneg1;
2038 sp->sa_size = nfs_xdrneg1;
2039 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2040 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2041 }
2042
2043 /*
2044 * Issue the NFS request and get the rpc response.
2045 *
2046 * Only NFSv3 responses returning an error of 0 actually return
2047 * a file handle that can be converted into newvp without having
2048 * to do an extra lookup rpc.
2049 */
2050 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
2051 if (v3) {
2052 if (error == 0)
2053 nfsm_mtofh(dvp, newvp, v3, gotvp);
2054 nfsm_wcc_data(dvp, wccflag);
2055 }
2056
2057 /*
2058 * out code jumps -> here, mrep is also freed.
2059 */
2060
2061 m_freem(mrep);
2062 nfsmout:
2063
2064 /*
2065 * If we do not have an error and we could not extract the newvp from
2066 * the response due to the request being NFSv2, we have to do a
2067 * lookup in order to obtain a newvp to return.
2068 */
2069 if (error == 0 && newvp == NULL) {
2070 struct nfsnode *np = NULL;
2071
2072 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2073 cnp->cn_cred, cnp->cn_thread, &np);
2074 if (!error)
2075 newvp = NFSTOV(np);
2076 }
2077 if (error) {
2078 if (newvp)
2079 vput(newvp);
2080 } else {
2081 *ap->a_vpp = newvp;
2082 }
2083 mtx_lock(&(VTONFS(dvp))->n_mtx);
2084 VTONFS(dvp)->n_flag |= NMODIFIED;
2085 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2086 if (!wccflag) {
2087 VTONFS(dvp)->n_attrstamp = 0;
2088 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2089 }
2090 return (error);
2091 }
2092
2093 /*
2094 * nfs make dir call
2095 */
2096 static int
2097 nfs_mkdir(struct vop_mkdir_args *ap)
2098 {
2099 struct vnode *dvp = ap->a_dvp;
2100 struct vattr *vap = ap->a_vap;
2101 struct componentname *cnp = ap->a_cnp;
2102 struct nfsv2_sattr *sp;
2103 int len;
2104 struct nfsnode *np = NULL;
2105 struct vnode *newvp = NULL;
2106 caddr_t bpos, dpos;
2107 int error = 0, wccflag = NFSV3_WCCRATTR;
2108 int gotvp = 0;
2109 struct mbuf *mreq, *mrep, *md, *mb;
2110 struct vattr vattr;
2111 int v3 = NFS_ISV3(dvp);
2112
2113 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2114 return (error);
2115 len = cnp->cn_namelen;
2116 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2117 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
2118 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2119 mb = mreq;
2120 bpos = mtod(mb, caddr_t);
2121 nfsm_fhtom(dvp, v3);
2122 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2123 if (v3) {
2124 nfsm_v3attrbuild(vap, FALSE);
2125 } else {
2126 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2127 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2128 sp->sa_uid = nfs_xdrneg1;
2129 sp->sa_gid = nfs_xdrneg1;
2130 sp->sa_size = nfs_xdrneg1;
2131 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2132 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2133 }
2134 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2135 if (!error)
2136 nfsm_mtofh(dvp, newvp, v3, gotvp);
2137 if (v3)
2138 nfsm_wcc_data(dvp, wccflag);
2139 m_freem(mrep);
2140 nfsmout:
2141 mtx_lock(&(VTONFS(dvp))->n_mtx);
2142 VTONFS(dvp)->n_flag |= NMODIFIED;
2143 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2144 if (!wccflag) {
2145 VTONFS(dvp)->n_attrstamp = 0;
2146 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2147 }
2148 if (error == 0 && newvp == NULL) {
2149 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2150 cnp->cn_thread, &np);
2151 if (!error) {
2152 newvp = NFSTOV(np);
2153 if (newvp->v_type != VDIR)
2154 error = EEXIST;
2155 }
2156 }
2157 if (error) {
2158 if (newvp)
2159 vput(newvp);
2160 } else
2161 *ap->a_vpp = newvp;
2162 return (error);
2163 }
2164
2165 /*
2166 * nfs remove directory call
2167 */
2168 static int
2169 nfs_rmdir(struct vop_rmdir_args *ap)
2170 {
2171 struct vnode *vp = ap->a_vp;
2172 struct vnode *dvp = ap->a_dvp;
2173 struct componentname *cnp = ap->a_cnp;
2174 caddr_t bpos, dpos;
2175 int error = 0, wccflag = NFSV3_WCCRATTR;
2176 struct mbuf *mreq, *mrep, *md, *mb;
2177 int v3 = NFS_ISV3(dvp);
2178
2179 if (dvp == vp)
2180 return (EINVAL);
2181 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2182 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
2183 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2184 mb = mreq;
2185 bpos = mtod(mb, caddr_t);
2186 nfsm_fhtom(dvp, v3);
2187 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2188 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2189 if (v3)
2190 nfsm_wcc_data(dvp, wccflag);
2191 m_freem(mrep);
2192 nfsmout:
2193 mtx_lock(&(VTONFS(dvp))->n_mtx);
2194 VTONFS(dvp)->n_flag |= NMODIFIED;
2195 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2196 if (!wccflag) {
2197 VTONFS(dvp)->n_attrstamp = 0;
2198 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2199 }
2200 cache_purge(dvp);
2201 cache_purge(vp);
2202 /*
2203 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2204 */
2205 if (error == ENOENT)
2206 error = 0;
2207 return (error);
2208 }
2209
2210 /*
2211 * nfs readdir call
2212 */
2213 static int
2214 nfs_readdir(struct vop_readdir_args *ap)
2215 {
2216 struct vnode *vp = ap->a_vp;
2217 struct nfsnode *np = VTONFS(vp);
2218 struct uio *uio = ap->a_uio;
2219 int tresid, error = 0;
2220 struct vattr vattr;
2221
2222 if (vp->v_type != VDIR)
2223 return(EPERM);
2224
2225 /*
2226 * First, check for hit on the EOF offset cache
2227 */
2228 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2229 (np->n_flag & NMODIFIED) == 0) {
2230 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2231 mtx_lock(&np->n_mtx);
2232 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2233 mtx_unlock(&np->n_mtx);
2234 nfsstats.direofcache_hits++;
2235 goto out;
2236 } else
2237 mtx_unlock(&np->n_mtx);
2238 }
2239 }
2240
2241 /*
2242 * Call nfs_bioread() to do the real work.
2243 */
2244 tresid = uio->uio_resid;
2245 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2246
2247 if (!error && uio->uio_resid == tresid) {
2248 nfsstats.direofcache_misses++;
2249 }
2250 out:
2251 return (error);
2252 }
2253
2254 /*
2255 * Readdir rpc call.
2256 * Called from below the buffer cache by nfs_doio().
2257 */
2258 int
2259 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2260 {
2261 int len, left;
2262 struct dirent *dp = NULL;
2263 u_int32_t *tl;
2264 caddr_t cp;
2265 nfsuint64 *cookiep;
2266 caddr_t bpos, dpos;
2267 struct mbuf *mreq, *mrep, *md, *mb;
2268 nfsuint64 cookie;
2269 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2270 struct nfsnode *dnp = VTONFS(vp);
2271 u_quad_t fileno;
2272 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2273 int attrflag;
2274 int v3 = NFS_ISV3(vp);
2275
2276 KASSERT(uiop->uio_iovcnt == 1 &&
2277 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2278 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2279 ("nfs readdirrpc bad uio"));
2280
2281 /*
2282 * If there is no cookie, assume directory was stale.
2283 */
2284 nfs_dircookie_lock(dnp);
2285 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2286 if (cookiep) {
2287 cookie = *cookiep;
2288 nfs_dircookie_unlock(dnp);
2289 } else {
2290 nfs_dircookie_unlock(dnp);
2291 return (NFSERR_BAD_COOKIE);
2292 }
2293
2294 /*
2295 * Loop around doing readdir rpc's of size nm_readdirsize
2296 * truncated to a multiple of DIRBLKSIZ.
2297 * The stopping criteria is EOF or buffer full.
2298 */
2299 while (more_dirs && bigenough) {
2300 nfsstats.rpccnt[NFSPROC_READDIR]++;
2301 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2302 NFSX_READDIR(v3));
2303 mb = mreq;
2304 bpos = mtod(mb, caddr_t);
2305 nfsm_fhtom(vp, v3);
2306 if (v3) {
2307 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2308 *tl++ = cookie.nfsuquad[0];
2309 *tl++ = cookie.nfsuquad[1];
2310 mtx_lock(&dnp->n_mtx);
2311 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2312 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2313 mtx_unlock(&dnp->n_mtx);
2314 } else {
2315 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2316 *tl++ = cookie.nfsuquad[0];
2317 }
2318 *tl = txdr_unsigned(nmp->nm_readdirsize);
2319 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2320 if (v3) {
2321 nfsm_postop_attr(vp, attrflag);
2322 if (!error) {
2323 tl = nfsm_dissect(u_int32_t *,
2324 2 * NFSX_UNSIGNED);
2325 mtx_lock(&dnp->n_mtx);
2326 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2327 dnp->n_cookieverf.nfsuquad[1] = *tl;
2328 mtx_unlock(&dnp->n_mtx);
2329 } else {
2330 m_freem(mrep);
2331 goto nfsmout;
2332 }
2333 }
2334 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2335 more_dirs = fxdr_unsigned(int, *tl);
2336
2337 /* loop thru the dir entries, doctoring them to 4bsd form */
2338 while (more_dirs && bigenough) {
2339 if (v3) {
2340 tl = nfsm_dissect(u_int32_t *,
2341 3 * NFSX_UNSIGNED);
2342 fileno = fxdr_hyper(tl);
2343 len = fxdr_unsigned(int, *(tl + 2));
2344 } else {
2345 tl = nfsm_dissect(u_int32_t *,
2346 2 * NFSX_UNSIGNED);
2347 fileno = fxdr_unsigned(u_quad_t, *tl++);
2348 len = fxdr_unsigned(int, *tl);
2349 }
2350 if (len <= 0 || len > NFS_MAXNAMLEN) {
2351 error = EBADRPC;
2352 m_freem(mrep);
2353 goto nfsmout;
2354 }
2355 tlen = nfsm_rndup(len);
2356 if (tlen == len)
2357 tlen += 4; /* To ensure null termination */
2358 left = DIRBLKSIZ - blksiz;
2359 if ((tlen + DIRHDSIZ) > left) {
2360 dp->d_reclen += left;
2361 uiop->uio_iov->iov_base =
2362 (char *)uiop->uio_iov->iov_base + left;
2363 uiop->uio_iov->iov_len -= left;
2364 uiop->uio_offset += left;
2365 uiop->uio_resid -= left;
2366 blksiz = 0;
2367 }
2368 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2369 bigenough = 0;
2370 if (bigenough) {
2371 dp = (struct dirent *)uiop->uio_iov->iov_base;
2372 dp->d_fileno = (int)fileno;
2373 dp->d_namlen = len;
2374 dp->d_reclen = tlen + DIRHDSIZ;
2375 dp->d_type = DT_UNKNOWN;
2376 blksiz += dp->d_reclen;
2377 if (blksiz == DIRBLKSIZ)
2378 blksiz = 0;
2379 uiop->uio_offset += DIRHDSIZ;
2380 uiop->uio_resid -= DIRHDSIZ;
2381 uiop->uio_iov->iov_base =
2382 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2383 uiop->uio_iov->iov_len -= DIRHDSIZ;
2384 nfsm_mtouio(uiop, len);
2385 cp = uiop->uio_iov->iov_base;
2386 tlen -= len;
2387 *cp = '\0'; /* null terminate */
2388 uiop->uio_iov->iov_base =
2389 (char *)uiop->uio_iov->iov_base + tlen;
2390 uiop->uio_iov->iov_len -= tlen;
2391 uiop->uio_offset += tlen;
2392 uiop->uio_resid -= tlen;
2393 } else
2394 nfsm_adv(nfsm_rndup(len));
2395 if (v3) {
2396 tl = nfsm_dissect(u_int32_t *,
2397 3 * NFSX_UNSIGNED);
2398 } else {
2399 tl = nfsm_dissect(u_int32_t *,
2400 2 * NFSX_UNSIGNED);
2401 }
2402 if (bigenough) {
2403 cookie.nfsuquad[0] = *tl++;
2404 if (v3)
2405 cookie.nfsuquad[1] = *tl++;
2406 } else if (v3)
2407 tl += 2;
2408 else
2409 tl++;
2410 more_dirs = fxdr_unsigned(int, *tl);
2411 }
2412 /*
2413 * If at end of rpc data, get the eof boolean
2414 */
2415 if (!more_dirs) {
2416 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2417 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2418 }
2419 m_freem(mrep);
2420 }
2421 /*
2422 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2423 * by increasing d_reclen for the last record.
2424 */
2425 if (blksiz > 0) {
2426 left = DIRBLKSIZ - blksiz;
2427 dp->d_reclen += left;
2428 uiop->uio_iov->iov_base =
2429 (char *)uiop->uio_iov->iov_base + left;
2430 uiop->uio_iov->iov_len -= left;
2431 uiop->uio_offset += left;
2432 uiop->uio_resid -= left;
2433 }
2434
2435 /*
2436 * We are now either at the end of the directory or have filled the
2437 * block.
2438 */
2439 if (bigenough)
2440 dnp->n_direofoffset = uiop->uio_offset;
2441 else {
2442 if (uiop->uio_resid > 0)
2443 nfs_printf("EEK! readdirrpc resid > 0\n");
2444 nfs_dircookie_lock(dnp);
2445 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2446 *cookiep = cookie;
2447 nfs_dircookie_unlock(dnp);
2448 }
2449 nfsmout:
2450 return (error);
2451 }
2452
2453 /*
2454 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2455 */
2456 int
2457 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2458 {
2459 int len, left;
2460 struct dirent *dp;
2461 u_int32_t *tl;
2462 caddr_t cp;
2463 struct vnode *newvp;
2464 nfsuint64 *cookiep;
2465 caddr_t bpos, dpos, dpossav1, dpossav2;
2466 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2467 struct nameidata nami, *ndp = &nami;
2468 struct componentname *cnp = &ndp->ni_cnd;
2469 nfsuint64 cookie;
2470 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2471 struct nfsnode *dnp = VTONFS(vp), *np;
2472 nfsfh_t *fhp;
2473 u_quad_t fileno;
2474 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2475 int attrflag, fhsize;
2476
2477 #ifndef nolint
2478 dp = NULL;
2479 #endif
2480 KASSERT(uiop->uio_iovcnt == 1 &&
2481 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2482 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2483 ("nfs readdirplusrpc bad uio"));
2484 ndp->ni_dvp = vp;
2485 newvp = NULLVP;
2486
2487 /*
2488 * If there is no cookie, assume directory was stale.
2489 */
2490 nfs_dircookie_lock(dnp);
2491 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2492 if (cookiep) {
2493 cookie = *cookiep;
2494 nfs_dircookie_unlock(dnp);
2495 } else {
2496 nfs_dircookie_unlock(dnp);
2497 return (NFSERR_BAD_COOKIE);
2498 }
2499 /*
2500 * Loop around doing readdir rpc's of size nm_readdirsize
2501 * truncated to a multiple of DIRBLKSIZ.
2502 * The stopping criteria is EOF or buffer full.
2503 */
2504 while (more_dirs && bigenough) {
2505 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2506 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2507 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2508 mb = mreq;
2509 bpos = mtod(mb, caddr_t);
2510 nfsm_fhtom(vp, 1);
2511 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2512 *tl++ = cookie.nfsuquad[0];
2513 *tl++ = cookie.nfsuquad[1];
2514 mtx_lock(&dnp->n_mtx);
2515 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2516 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2517 mtx_unlock(&dnp->n_mtx);
2518 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2519 *tl = txdr_unsigned(nmp->nm_rsize);
2520 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2521 nfsm_postop_attr(vp, attrflag);
2522 if (error) {
2523 m_freem(mrep);
2524 goto nfsmout;
2525 }
2526 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2527 mtx_lock(&dnp->n_mtx);
2528 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2529 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2530 mtx_unlock(&dnp->n_mtx);
2531 more_dirs = fxdr_unsigned(int, *tl);
2532
2533 /* loop thru the dir entries, doctoring them to 4bsd form */
2534 while (more_dirs && bigenough) {
2535 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2536 fileno = fxdr_hyper(tl);
2537 len = fxdr_unsigned(int, *(tl + 2));
2538 if (len <= 0 || len > NFS_MAXNAMLEN) {
2539 error = EBADRPC;
2540 m_freem(mrep);
2541 goto nfsmout;
2542 }
2543 tlen = nfsm_rndup(len);
2544 if (tlen == len)
2545 tlen += 4; /* To ensure null termination*/
2546 left = DIRBLKSIZ - blksiz;
2547 if ((tlen + DIRHDSIZ) > left) {
2548 dp->d_reclen += left;
2549 uiop->uio_iov->iov_base =
2550 (char *)uiop->uio_iov->iov_base + left;
2551 uiop->uio_iov->iov_len -= left;
2552 uiop->uio_offset += left;
2553 uiop->uio_resid -= left;
2554 blksiz = 0;
2555 }
2556 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2557 bigenough = 0;
2558 if (bigenough) {
2559 dp = (struct dirent *)uiop->uio_iov->iov_base;
2560 dp->d_fileno = (int)fileno;
2561 dp->d_namlen = len;
2562 dp->d_reclen = tlen + DIRHDSIZ;
2563 dp->d_type = DT_UNKNOWN;
2564 blksiz += dp->d_reclen;
2565 if (blksiz == DIRBLKSIZ)
2566 blksiz = 0;
2567 uiop->uio_offset += DIRHDSIZ;
2568 uiop->uio_resid -= DIRHDSIZ;
2569 uiop->uio_iov->iov_base =
2570 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2571 uiop->uio_iov->iov_len -= DIRHDSIZ;
2572 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2573 cnp->cn_namelen = len;
2574 nfsm_mtouio(uiop, len);
2575 cp = uiop->uio_iov->iov_base;
2576 tlen -= len;
2577 *cp = '\0';
2578 uiop->uio_iov->iov_base =
2579 (char *)uiop->uio_iov->iov_base + tlen;
2580 uiop->uio_iov->iov_len -= tlen;
2581 uiop->uio_offset += tlen;
2582 uiop->uio_resid -= tlen;
2583 } else
2584 nfsm_adv(nfsm_rndup(len));
2585 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2586 if (bigenough) {
2587 cookie.nfsuquad[0] = *tl++;
2588 cookie.nfsuquad[1] = *tl++;
2589 } else
2590 tl += 2;
2591
2592 /*
2593 * Since the attributes are before the file handle
2594 * (sigh), we must skip over the attributes and then
2595 * come back and get them.
2596 */
2597 attrflag = fxdr_unsigned(int, *tl);
2598 if (attrflag) {
2599 dpossav1 = dpos;
2600 mdsav1 = md;
2601 nfsm_adv(NFSX_V3FATTR);
2602 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2603 doit = fxdr_unsigned(int, *tl);
2604 /*
2605 * Skip loading the attrs for "..". There's a
2606 * race between loading the attrs here and
2607 * lookups that look for the directory currently
2608 * being read (in the parent). We try to acquire
2609 * the exclusive lock on ".." here, owning the
2610 * lock on the directory being read. Lookup will
2611 * hold the lock on ".." and try to acquire the
2612 * lock on the directory being read.
2613 *
2614 * There are other ways of fixing this, one would
2615 * be to do a trylock on the ".." vnode and skip
2616 * loading the attrs on ".." if it happens to be
2617 * locked by another process. But skipping the
2618 * attrload on ".." seems the easiest option.
2619 */
2620 if (strcmp(dp->d_name, "..") == 0) {
2621 doit = 0;
2622 /*
2623 * We've already skipped over the attrs,
2624 * skip over the filehandle. And store d_type
2625 * as VDIR.
2626 */
2627 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2628 i = fxdr_unsigned(int, *tl);
2629 nfsm_adv(nfsm_rndup(i));
2630 dp->d_type = IFTODT(VTTOIF(VDIR));
2631 }
2632 if (doit) {
2633 nfsm_getfh(fhp, fhsize, 1);
2634 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2635 VREF(vp);
2636 newvp = vp;
2637 np = dnp;
2638 } else {
2639 error = nfs_nget(vp->v_mount, fhp,
2640 fhsize, &np, LK_EXCLUSIVE);
2641 if (error)
2642 doit = 0;
2643 else
2644 newvp = NFSTOV(np);
2645 }
2646 }
2647 if (doit && bigenough) {
2648 dpossav2 = dpos;
2649 dpos = dpossav1;
2650 mdsav2 = md;
2651 md = mdsav1;
2652 nfsm_loadattr(newvp, NULL);
2653 dpos = dpossav2;
2654 md = mdsav2;
2655 dp->d_type =
2656 IFTODT(VTTOIF(np->n_vattr.va_type));
2657 ndp->ni_vp = newvp;
2658 /*
2659 * Update n_ctime so subsequent lookup
2660 * doesn't purge entry.
2661 */
2662 np->n_ctime = np->n_vattr.va_ctime;
2663 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2664 }
2665 } else {
2666 /* Just skip over the file handle */
2667 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2668 i = fxdr_unsigned(int, *tl);
2669 if (i) {
2670 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2671 fhsize = fxdr_unsigned(int, *tl);
2672 nfsm_adv(nfsm_rndup(fhsize));
2673 }
2674 }
2675 if (newvp != NULLVP) {
2676 if (newvp == vp)
2677 vrele(newvp);
2678 else
2679 vput(newvp);
2680 newvp = NULLVP;
2681 }
2682 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2683 more_dirs = fxdr_unsigned(int, *tl);
2684 }
2685 /*
2686 * If at end of rpc data, get the eof boolean
2687 */
2688 if (!more_dirs) {
2689 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2690 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2691 }
2692 m_freem(mrep);
2693 }
2694 /*
2695 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2696 * by increasing d_reclen for the last record.
2697 */
2698 if (blksiz > 0) {
2699 left = DIRBLKSIZ - blksiz;
2700 dp->d_reclen += left;
2701 uiop->uio_iov->iov_base =
2702 (char *)uiop->uio_iov->iov_base + left;
2703 uiop->uio_iov->iov_len -= left;
2704 uiop->uio_offset += left;
2705 uiop->uio_resid -= left;
2706 }
2707
2708 /*
2709 * We are now either at the end of the directory or have filled the
2710 * block.
2711 */
2712 if (bigenough)
2713 dnp->n_direofoffset = uiop->uio_offset;
2714 else {
2715 if (uiop->uio_resid > 0)
2716 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2717 nfs_dircookie_lock(dnp);
2718 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2719 *cookiep = cookie;
2720 nfs_dircookie_unlock(dnp);
2721 }
2722 nfsmout:
2723 if (newvp != NULLVP) {
2724 if (newvp == vp)
2725 vrele(newvp);
2726 else
2727 vput(newvp);
2728 newvp = NULLVP;
2729 }
2730 return (error);
2731 }
2732
2733 /*
2734 * Silly rename. To make the NFS filesystem that is stateless look a little
2735 * more like the "ufs" a remove of an active vnode is translated to a rename
2736 * to a funny looking filename that is removed by nfs_inactive on the
2737 * nfsnode. There is the potential for another process on a different client
2738 * to create the same funny name between the nfs_lookitup() fails and the
2739 * nfs_rename() completes, but...
2740 */
2741 static int
2742 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2743 {
2744 struct sillyrename *sp;
2745 struct nfsnode *np;
2746 int error;
2747 short pid;
2748 unsigned int lticks;
2749
2750 cache_purge(dvp);
2751 np = VTONFS(vp);
2752 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2753 sp = malloc(sizeof (struct sillyrename),
2754 M_NFSREQ, M_WAITOK);
2755 sp->s_cred = crhold(cnp->cn_cred);
2756 sp->s_dvp = dvp;
2757 sp->s_removeit = nfs_removeit;
2758 VREF(dvp);
2759
2760 /*
2761 * Fudge together a funny name.
2762 * Changing the format of the funny name to accomodate more
2763 * sillynames per directory.
2764 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2765 * CPU ticks since boot.
2766 */
2767 pid = cnp->cn_thread->td_proc->p_pid;
2768 lticks = (unsigned int)ticks;
2769 for ( ; ; ) {
2770 sp->s_namlen = sprintf(sp->s_name,
2771 ".nfs.%08x.%04x4.4", lticks,
2772 pid);
2773 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2774 cnp->cn_thread, NULL))
2775 break;
2776 lticks++;
2777 }
2778 error = nfs_renameit(dvp, cnp, sp);
2779 if (error)
2780 goto bad;
2781 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2782 cnp->cn_thread, &np);
2783 np->n_sillyrename = sp;
2784 return (0);
2785 bad:
2786 vrele(sp->s_dvp);
2787 crfree(sp->s_cred);
2788 free((caddr_t)sp, M_NFSREQ);
2789 return (error);
2790 }
2791
2792 /*
2793 * Look up a file name and optionally either update the file handle or
2794 * allocate an nfsnode, depending on the value of npp.
2795 * npp == NULL --> just do the lookup
2796 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2797 * handled too
2798 * *npp != NULL --> update the file handle in the vnode
2799 */
2800 static int
2801 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2802 struct thread *td, struct nfsnode **npp)
2803 {
2804 struct vnode *newvp = NULL;
2805 struct nfsnode *np, *dnp = VTONFS(dvp);
2806 caddr_t bpos, dpos;
2807 int error = 0, fhlen, attrflag;
2808 struct mbuf *mreq, *mrep, *md, *mb;
2809 nfsfh_t *nfhp;
2810 int v3 = NFS_ISV3(dvp);
2811
2812 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2813 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2814 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2815 mb = mreq;
2816 bpos = mtod(mb, caddr_t);
2817 nfsm_fhtom(dvp, v3);
2818 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2819 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2820 if (npp && !error) {
2821 nfsm_getfh(nfhp, fhlen, v3);
2822 if (*npp) {
2823 np = *npp;
2824 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2825 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2826 np->n_fhp = &np->n_fh;
2827 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2828 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2829 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2830 np->n_fhsize = fhlen;
2831 newvp = NFSTOV(np);
2832 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2833 VREF(dvp);
2834 newvp = dvp;
2835 } else {
2836 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2837 if (error) {
2838 m_freem(mrep);
2839 return (error);
2840 }
2841 newvp = NFSTOV(np);
2842 }
2843 if (v3) {
2844 nfsm_postop_attr(newvp, attrflag);
2845 if (!attrflag && *npp == NULL) {
2846 m_freem(mrep);
2847 if (newvp == dvp)
2848 vrele(newvp);
2849 else
2850 vput(newvp);
2851 return (ENOENT);
2852 }
2853 } else
2854 nfsm_loadattr(newvp, NULL);
2855 }
2856 m_freem(mrep);
2857 nfsmout:
2858 if (npp && *npp == NULL) {
2859 if (error) {
2860 if (newvp) {
2861 if (newvp == dvp)
2862 vrele(newvp);
2863 else
2864 vput(newvp);
2865 }
2866 } else
2867 *npp = np;
2868 }
2869 return (error);
2870 }
2871
2872 /*
2873 * Nfs Version 3 commit rpc
2874 */
2875 int
2876 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2877 struct thread *td)
2878 {
2879 u_int32_t *tl;
2880 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2881 caddr_t bpos, dpos;
2882 int error = 0, wccflag = NFSV3_WCCRATTR;
2883 struct mbuf *mreq, *mrep, *md, *mb;
2884
2885 mtx_lock(&nmp->nm_mtx);
2886 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2887 mtx_unlock(&nmp->nm_mtx);
2888 return (0);
2889 }
2890 mtx_unlock(&nmp->nm_mtx);
2891 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2892 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2893 mb = mreq;
2894 bpos = mtod(mb, caddr_t);
2895 nfsm_fhtom(vp, 1);
2896 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2897 txdr_hyper(offset, tl);
2898 tl += 2;
2899 *tl = txdr_unsigned(cnt);
2900 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2901 nfsm_wcc_data(vp, wccflag);
2902 if (!error) {
2903 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2904 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2905 NFSX_V3WRITEVERF)) {
2906 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2907 NFSX_V3WRITEVERF);
2908 error = NFSERR_STALEWRITEVERF;
2909 }
2910 }
2911 m_freem(mrep);
2912 nfsmout:
2913 return (error);
2914 }
2915
2916 /*
2917 * Strategy routine.
2918 * For async requests when nfsiod(s) are running, queue the request by
2919 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2920 * request.
2921 */
2922 static int
2923 nfs_strategy(struct vop_strategy_args *ap)
2924 {
2925 struct buf *bp = ap->a_bp;
2926 struct ucred *cr;
2927
2928 KASSERT(!(bp->b_flags & B_DONE),
2929 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2930 BUF_ASSERT_HELD(bp);
2931
2932 if (bp->b_iocmd == BIO_READ)
2933 cr = bp->b_rcred;
2934 else
2935 cr = bp->b_wcred;
2936
2937 /*
2938 * If the op is asynchronous and an i/o daemon is waiting
2939 * queue the request, wake it up and wait for completion
2940 * otherwise just do it ourselves.
2941 */
2942 if ((bp->b_flags & B_ASYNC) == 0 ||
2943 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2944 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2945 return (0);
2946 }
2947
2948 /*
2949 * fsync vnode op. Just call nfs_flush() with commit == 1.
2950 */
2951 /* ARGSUSED */
2952 static int
2953 nfs_fsync(struct vop_fsync_args *ap)
2954 {
2955
2956 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1));
2957 }
2958
2959 /*
2960 * Flush all the blocks associated with a vnode.
2961 * Walk through the buffer pool and push any dirty pages
2962 * associated with the vnode.
2963 */
2964 static int
2965 nfs_flush(struct vnode *vp, int waitfor, int commit)
2966 {
2967 struct nfsnode *np = VTONFS(vp);
2968 struct buf *bp;
2969 int i;
2970 struct buf *nbp;
2971 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2972 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2973 int passone = 1;
2974 u_quad_t off, endoff, toff;
2975 struct ucred* wcred = NULL;
2976 struct buf **bvec = NULL;
2977 struct bufobj *bo;
2978 struct thread *td = curthread;
2979 #ifndef NFS_COMMITBVECSIZ
2980 #define NFS_COMMITBVECSIZ 20
2981 #endif
2982 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2983 int bvecsize = 0, bveccount;
2984
2985 if (nmp->nm_flag & NFSMNT_INT)
2986 slpflag = NFS_PCATCH;
2987 if (!commit)
2988 passone = 0;
2989 bo = &vp->v_bufobj;
2990 /*
2991 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2992 * server, but has not been committed to stable storage on the server
2993 * yet. On the first pass, the byte range is worked out and the commit
2994 * rpc is done. On the second pass, nfs_writebp() is called to do the
2995 * job.
2996 */
2997 again:
2998 off = (u_quad_t)-1;
2999 endoff = 0;
3000 bvecpos = 0;
3001 if (NFS_ISV3(vp) && commit) {
3002 if (bvec != NULL && bvec != bvec_on_stack)
3003 free(bvec, M_TEMP);
3004 /*
3005 * Count up how many buffers waiting for a commit.
3006 */
3007 bveccount = 0;
3008 BO_LOCK(bo);
3009 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3010 if (!BUF_ISLOCKED(bp) &&
3011 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
3012 == (B_DELWRI | B_NEEDCOMMIT))
3013 bveccount++;
3014 }
3015 /*
3016 * Allocate space to remember the list of bufs to commit. It is
3017 * important to use M_NOWAIT here to avoid a race with nfs_write.
3018 * If we can't get memory (for whatever reason), we will end up
3019 * committing the buffers one-by-one in the loop below.
3020 */
3021 if (bveccount > NFS_COMMITBVECSIZ) {
3022 /*
3023 * Release the vnode interlock to avoid a lock
3024 * order reversal.
3025 */
3026 BO_UNLOCK(bo);
3027 bvec = (struct buf **)
3028 malloc(bveccount * sizeof(struct buf *),
3029 M_TEMP, M_NOWAIT);
3030 BO_LOCK(bo);
3031 if (bvec == NULL) {
3032 bvec = bvec_on_stack;
3033 bvecsize = NFS_COMMITBVECSIZ;
3034 } else
3035 bvecsize = bveccount;
3036 } else {
3037 bvec = bvec_on_stack;
3038 bvecsize = NFS_COMMITBVECSIZ;
3039 }
3040 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3041 if (bvecpos >= bvecsize)
3042 break;
3043 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3044 nbp = TAILQ_NEXT(bp, b_bobufs);
3045 continue;
3046 }
3047 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3048 (B_DELWRI | B_NEEDCOMMIT)) {
3049 BUF_UNLOCK(bp);
3050 nbp = TAILQ_NEXT(bp, b_bobufs);
3051 continue;
3052 }
3053 BO_UNLOCK(bo);
3054 bremfree(bp);
3055 /*
3056 * Work out if all buffers are using the same cred
3057 * so we can deal with them all with one commit.
3058 *
3059 * NOTE: we are not clearing B_DONE here, so we have
3060 * to do it later on in this routine if we intend to
3061 * initiate I/O on the bp.
3062 *
3063 * Note: to avoid loopback deadlocks, we do not
3064 * assign b_runningbufspace.
3065 */
3066 if (wcred == NULL)
3067 wcred = bp->b_wcred;
3068 else if (wcred != bp->b_wcred)
3069 wcred = NOCRED;
3070 vfs_busy_pages(bp, 1);
3071
3072 BO_LOCK(bo);
3073 /*
3074 * bp is protected by being locked, but nbp is not
3075 * and vfs_busy_pages() may sleep. We have to
3076 * recalculate nbp.
3077 */
3078 nbp = TAILQ_NEXT(bp, b_bobufs);
3079
3080 /*
3081 * A list of these buffers is kept so that the
3082 * second loop knows which buffers have actually
3083 * been committed. This is necessary, since there
3084 * may be a race between the commit rpc and new
3085 * uncommitted writes on the file.
3086 */
3087 bvec[bvecpos++] = bp;
3088 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3089 bp->b_dirtyoff;
3090 if (toff < off)
3091 off = toff;
3092 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3093 if (toff > endoff)
3094 endoff = toff;
3095 }
3096 BO_UNLOCK(bo);
3097 }
3098 if (bvecpos > 0) {
3099 /*
3100 * Commit data on the server, as required.
3101 * If all bufs are using the same wcred, then use that with
3102 * one call for all of them, otherwise commit each one
3103 * separately.
3104 */
3105 if (wcred != NOCRED)
3106 retv = nfs_commit(vp, off, (int)(endoff - off),
3107 wcred, td);
3108 else {
3109 retv = 0;
3110 for (i = 0; i < bvecpos; i++) {
3111 off_t off, size;
3112 bp = bvec[i];
3113 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3114 bp->b_dirtyoff;
3115 size = (u_quad_t)(bp->b_dirtyend
3116 - bp->b_dirtyoff);
3117 retv = nfs_commit(vp, off, (int)size,
3118 bp->b_wcred, td);
3119 if (retv) break;
3120 }
3121 }
3122
3123 if (retv == NFSERR_STALEWRITEVERF)
3124 nfs_clearcommit(vp->v_mount);
3125
3126 /*
3127 * Now, either mark the blocks I/O done or mark the
3128 * blocks dirty, depending on whether the commit
3129 * succeeded.
3130 */
3131 for (i = 0; i < bvecpos; i++) {
3132 bp = bvec[i];
3133 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3134 if (retv) {
3135 /*
3136 * Error, leave B_DELWRI intact
3137 */
3138 vfs_unbusy_pages(bp);
3139 brelse(bp);
3140 } else {
3141 /*
3142 * Success, remove B_DELWRI ( bundirty() ).
3143 *
3144 * b_dirtyoff/b_dirtyend seem to be NFS
3145 * specific. We should probably move that
3146 * into bundirty(). XXX
3147 */
3148 bufobj_wref(bo);
3149 bp->b_flags |= B_ASYNC;
3150 bundirty(bp);
3151 bp->b_flags &= ~B_DONE;
3152 bp->b_ioflags &= ~BIO_ERROR;
3153 bp->b_dirtyoff = bp->b_dirtyend = 0;
3154 bufdone(bp);
3155 }
3156 }
3157 }
3158
3159 /*
3160 * Start/do any write(s) that are required.
3161 */
3162 loop:
3163 BO_LOCK(bo);
3164 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3165 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3166 if (waitfor != MNT_WAIT || passone)
3167 continue;
3168
3169 error = BUF_TIMELOCK(bp,
3170 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3171 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
3172 if (error == 0) {
3173 BUF_UNLOCK(bp);
3174 goto loop;
3175 }
3176 if (error == ENOLCK) {
3177 error = 0;
3178 goto loop;
3179 }
3180 if (nfs_sigintr(nmp, td)) {
3181 error = EINTR;
3182 goto done;
3183 }
3184 if (slpflag & PCATCH) {
3185 slpflag = 0;
3186 slptimeo = 2 * hz;
3187 }
3188 goto loop;
3189 }
3190 if ((bp->b_flags & B_DELWRI) == 0)
3191 panic("nfs_fsync: not dirty");
3192 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3193 BUF_UNLOCK(bp);
3194 continue;
3195 }
3196 BO_UNLOCK(bo);
3197 bremfree(bp);
3198 if (passone || !commit)
3199 bp->b_flags |= B_ASYNC;
3200 else
3201 bp->b_flags |= B_ASYNC;
3202 bwrite(bp);
3203 if (nfs_sigintr(nmp, td)) {
3204 error = EINTR;
3205 goto done;
3206 }
3207 goto loop;
3208 }
3209 if (passone) {
3210 passone = 0;
3211 BO_UNLOCK(bo);
3212 goto again;
3213 }
3214 if (waitfor == MNT_WAIT) {
3215 while (bo->bo_numoutput) {
3216 error = bufobj_wwait(bo, slpflag, slptimeo);
3217 if (error) {
3218 BO_UNLOCK(bo);
3219 error = nfs_sigintr(nmp, td);
3220 if (error)
3221 goto done;
3222 if (slpflag & PCATCH) {
3223 slpflag = 0;
3224 slptimeo = 2 * hz;
3225 }
3226 BO_LOCK(bo);
3227 }
3228 }
3229 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3230 BO_UNLOCK(bo);
3231 goto loop;
3232 }
3233 /*
3234 * Wait for all the async IO requests to drain
3235 */
3236 BO_UNLOCK(bo);
3237 mtx_lock(&np->n_mtx);
3238 while (np->n_directio_asyncwr > 0) {
3239 np->n_flag |= NFSYNCWAIT;
3240 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3241 &np->n_mtx, slpflag | (PRIBIO + 1),
3242 "nfsfsync", 0);
3243 if (error) {
3244 if (nfs_sigintr(nmp, td)) {
3245 mtx_unlock(&np->n_mtx);
3246 error = EINTR;
3247 goto done;
3248 }
3249 }
3250 }
3251 mtx_unlock(&np->n_mtx);
3252 } else
3253 BO_UNLOCK(bo);
3254 mtx_lock(&np->n_mtx);
3255 if (np->n_flag & NWRITEERR) {
3256 error = np->n_error;
3257 np->n_flag &= ~NWRITEERR;
3258 }
3259 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3260 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3261 np->n_flag &= ~NMODIFIED;
3262 mtx_unlock(&np->n_mtx);
3263 done:
3264 if (bvec != NULL && bvec != bvec_on_stack)
3265 free(bvec, M_TEMP);
3266 return (error);
3267 }
3268
3269 /*
3270 * NFS advisory byte-level locks.
3271 */
3272 static int
3273 nfs_advlock(struct vop_advlock_args *ap)
3274 {
3275 struct vnode *vp = ap->a_vp;
3276 u_quad_t size;
3277 int error;
3278
3279 error = vn_lock(vp, LK_SHARED);
3280 if (error)
3281 return (error);
3282 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3283 size = VTONFS(vp)->n_size;
3284 VOP_UNLOCK(vp, 0);
3285 error = lf_advlock(ap, &(vp->v_lockf), size);
3286 } else {
3287 if (nfs_advlock_p)
3288 error = nfs_advlock_p(ap);
3289 else
3290 error = ENOLCK;
3291 }
3292
3293 return (error);
3294 }
3295
3296 /*
3297 * NFS advisory byte-level locks.
3298 */
3299 static int
3300 nfs_advlockasync(struct vop_advlockasync_args *ap)
3301 {
3302 struct vnode *vp = ap->a_vp;
3303 u_quad_t size;
3304 int error;
3305
3306 error = vn_lock(vp, LK_SHARED);
3307 if (error)
3308 return (error);
3309 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3310 size = VTONFS(vp)->n_size;
3311 VOP_UNLOCK(vp, 0);
3312 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3313 } else {
3314 VOP_UNLOCK(vp, 0);
3315 error = EOPNOTSUPP;
3316 }
3317 return (error);
3318 }
3319
3320 /*
3321 * Print out the contents of an nfsnode.
3322 */
3323 static int
3324 nfs_print(struct vop_print_args *ap)
3325 {
3326 struct vnode *vp = ap->a_vp;
3327 struct nfsnode *np = VTONFS(vp);
3328
3329 nfs_printf("\tfileid %ld fsid 0x%x",
3330 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3331 if (vp->v_type == VFIFO)
3332 fifo_printinfo(vp);
3333 printf("\n");
3334 return (0);
3335 }
3336
3337 /*
3338 * This is the "real" nfs::bwrite(struct buf*).
3339 * We set B_CACHE if this is a VMIO buffer.
3340 */
3341 int
3342 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3343 {
3344 int s;
3345 int oldflags = bp->b_flags;
3346 #if 0
3347 int retv = 1;
3348 off_t off;
3349 #endif
3350
3351 BUF_ASSERT_HELD(bp);
3352
3353 if (bp->b_flags & B_INVAL) {
3354 brelse(bp);
3355 return(0);
3356 }
3357
3358 bp->b_flags |= B_CACHE;
3359
3360 /*
3361 * Undirty the bp. We will redirty it later if the I/O fails.
3362 */
3363
3364 s = splbio();
3365 bundirty(bp);
3366 bp->b_flags &= ~B_DONE;
3367 bp->b_ioflags &= ~BIO_ERROR;
3368 bp->b_iocmd = BIO_WRITE;
3369
3370 bufobj_wref(bp->b_bufobj);
3371 curthread->td_ru.ru_oublock++;
3372 splx(s);
3373
3374 /*
3375 * Note: to avoid loopback deadlocks, we do not
3376 * assign b_runningbufspace.
3377 */
3378 vfs_busy_pages(bp, 1);
3379
3380 BUF_KERNPROC(bp);
3381 bp->b_iooffset = dbtob(bp->b_blkno);
3382 bstrategy(bp);
3383
3384 if( (oldflags & B_ASYNC) == 0) {
3385 int rtval = bufwait(bp);
3386
3387 if (oldflags & B_DELWRI) {
3388 s = splbio();
3389 reassignbuf(bp);
3390 splx(s);
3391 }
3392 brelse(bp);
3393 return (rtval);
3394 }
3395
3396 return (0);
3397 }
3398
3399 /*
3400 * nfs special file access vnode op.
3401 * Essentially just get vattr and then imitate iaccess() since the device is
3402 * local to the client.
3403 */
3404 static int
3405 nfsspec_access(struct vop_access_args *ap)
3406 {
3407 struct vattr *vap;
3408 struct ucred *cred = ap->a_cred;
3409 struct vnode *vp = ap->a_vp;
3410 accmode_t accmode = ap->a_accmode;
3411 struct vattr vattr;
3412 int error;
3413
3414 /*
3415 * Disallow write attempts on filesystems mounted read-only;
3416 * unless the file is a socket, fifo, or a block or character
3417 * device resident on the filesystem.
3418 */
3419 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3420 switch (vp->v_type) {
3421 case VREG:
3422 case VDIR:
3423 case VLNK:
3424 return (EROFS);
3425 default:
3426 break;
3427 }
3428 }
3429 vap = &vattr;
3430 error = VOP_GETATTR(vp, vap, cred);
3431 if (error)
3432 goto out;
3433 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3434 accmode, cred, NULL);
3435 out:
3436 return error;
3437 }
3438
3439 /*
3440 * Read wrapper for fifos.
3441 */
3442 static int
3443 nfsfifo_read(struct vop_read_args *ap)
3444 {
3445 struct nfsnode *np = VTONFS(ap->a_vp);
3446 int error;
3447
3448 /*
3449 * Set access flag.
3450 */
3451 mtx_lock(&np->n_mtx);
3452 np->n_flag |= NACC;
3453 getnanotime(&np->n_atim);
3454 mtx_unlock(&np->n_mtx);
3455 error = fifo_specops.vop_read(ap);
3456 return error;
3457 }
3458
3459 /*
3460 * Write wrapper for fifos.
3461 */
3462 static int
3463 nfsfifo_write(struct vop_write_args *ap)
3464 {
3465 struct nfsnode *np = VTONFS(ap->a_vp);
3466
3467 /*
3468 * Set update flag.
3469 */
3470 mtx_lock(&np->n_mtx);
3471 np->n_flag |= NUPD;
3472 getnanotime(&np->n_mtim);
3473 mtx_unlock(&np->n_mtx);
3474 return(fifo_specops.vop_write(ap));
3475 }
3476
3477 /*
3478 * Close wrapper for fifos.
3479 *
3480 * Update the times on the nfsnode then do fifo close.
3481 */
3482 static int
3483 nfsfifo_close(struct vop_close_args *ap)
3484 {
3485 struct vnode *vp = ap->a_vp;
3486 struct nfsnode *np = VTONFS(vp);
3487 struct vattr vattr;
3488 struct timespec ts;
3489
3490 mtx_lock(&np->n_mtx);
3491 if (np->n_flag & (NACC | NUPD)) {
3492 getnanotime(&ts);
3493 if (np->n_flag & NACC)
3494 np->n_atim = ts;
3495 if (np->n_flag & NUPD)
3496 np->n_mtim = ts;
3497 np->n_flag |= NCHG;
3498 if (vrefcnt(vp) == 1 &&
3499 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3500 VATTR_NULL(&vattr);
3501 if (np->n_flag & NACC)
3502 vattr.va_atime = np->n_atim;
3503 if (np->n_flag & NUPD)
3504 vattr.va_mtime = np->n_mtim;
3505 mtx_unlock(&np->n_mtx);
3506 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3507 goto out;
3508 }
3509 }
3510 mtx_unlock(&np->n_mtx);
3511 out:
3512 return (fifo_specops.vop_close(ap));
3513 }
3514
3515 /*
3516 * Just call nfs_writebp() with the force argument set to 1.
3517 *
3518 * NOTE: B_DONE may or may not be set in a_bp on call.
3519 */
3520 static int
3521 nfs_bwrite(struct buf *bp)
3522 {
3523
3524 return (nfs_writebp(bp, 1, curthread));
3525 }
3526
3527 struct buf_ops buf_ops_nfs = {
3528 .bop_name = "buf_ops_nfs",
3529 .bop_write = nfs_bwrite,
3530 .bop_strategy = bufstrategy,
3531 .bop_sync = bufsync,
3532 .bop_bdflush = bufbdflush,
3533 };
Cache object: 3fb0ace6c2176c6cdb46b4304374fe7e
|