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/8.4/sys/nfsclient/nfs_vnops.c 247503 2013-02-28 21:58:07Z jhb $");
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 struct ucred *cred;
514
515 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
516 return (EOPNOTSUPP);
517
518 /*
519 * Get a valid lease. If cached data is stale, flush it.
520 */
521 mtx_lock(&np->n_mtx);
522 if (np->n_flag & NMODIFIED) {
523 mtx_unlock(&np->n_mtx);
524 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
525 if (error == EINTR || error == EIO)
526 return (error);
527 mtx_lock(&np->n_mtx);
528 np->n_attrstamp = 0;
529 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
530 if (vp->v_type == VDIR)
531 np->n_direofoffset = 0;
532 mtx_unlock(&np->n_mtx);
533 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
534 if (error)
535 return (error);
536 mtx_lock(&np->n_mtx);
537 np->n_mtime = vattr.va_mtime;
538 } else {
539 mtx_unlock(&np->n_mtx);
540 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
541 if (error)
542 return (error);
543 mtx_lock(&np->n_mtx);
544 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
545 if (vp->v_type == VDIR)
546 np->n_direofoffset = 0;
547 mtx_unlock(&np->n_mtx);
548 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
549 if (error == EINTR || error == EIO) {
550 return (error);
551 }
552 mtx_lock(&np->n_mtx);
553 np->n_mtime = vattr.va_mtime;
554 }
555 }
556 /*
557 * If the object has >= 1 O_DIRECT active opens, we disable caching.
558 */
559 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
560 if (np->n_directio_opens == 0) {
561 mtx_unlock(&np->n_mtx);
562 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
563 if (error)
564 return (error);
565 mtx_lock(&np->n_mtx);
566 np->n_flag |= NNONCACHE;
567 }
568 np->n_directio_opens++;
569 }
570
571 /*
572 * If this is an open for writing, capture a reference to the
573 * credentials, so they can be used by nfs_putpages(). Using
574 * these write credentials is preferable to the credentials of
575 * whatever thread happens to be doing the VOP_PUTPAGES() since
576 * the write RPCs are less likely to fail with EACCES.
577 */
578 if ((fmode & FWRITE) != 0) {
579 cred = np->n_writecred;
580 np->n_writecred = crhold(ap->a_cred);
581 } else
582 cred = NULL;
583 mtx_unlock(&np->n_mtx);
584 if (cred != NULL)
585 crfree(cred);
586 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
587 return (0);
588 }
589
590 /*
591 * nfs close vnode op
592 * What an NFS client should do upon close after writing is a debatable issue.
593 * Most NFS clients push delayed writes to the server upon close, basically for
594 * two reasons:
595 * 1 - So that any write errors may be reported back to the client process
596 * doing the close system call. By far the two most likely errors are
597 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
598 * 2 - To put a worst case upper bound on cache inconsistency between
599 * multiple clients for the file.
600 * There is also a consistency problem for Version 2 of the protocol w.r.t.
601 * not being able to tell if other clients are writing a file concurrently,
602 * since there is no way of knowing if the changed modify time in the reply
603 * is only due to the write for this client.
604 * (NFS Version 3 provides weak cache consistency data in the reply that
605 * should be sufficient to detect and handle this case.)
606 *
607 * The current code does the following:
608 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
609 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
610 * or commit them (this satisfies 1 and 2 except for the
611 * case where the server crashes after this close but
612 * before the commit RPC, which is felt to be "good
613 * enough". Changing the last argument to nfs_flush() to
614 * a 1 would force a commit operation, if it is felt a
615 * commit is necessary now.
616 */
617 /* ARGSUSED */
618 static int
619 nfs_close(struct vop_close_args *ap)
620 {
621 struct vnode *vp = ap->a_vp;
622 struct nfsnode *np = VTONFS(vp);
623 int error = 0;
624 int fmode = ap->a_fflag;
625
626 if (vp->v_type == VREG) {
627 /*
628 * Examine and clean dirty pages, regardless of NMODIFIED.
629 * This closes a major hole in close-to-open consistency.
630 * We want to push out all dirty pages (and buffers) on
631 * close, regardless of whether they were dirtied by
632 * mmap'ed writes or via write().
633 */
634 if (nfs_clean_pages_on_close && vp->v_object) {
635 VM_OBJECT_LOCK(vp->v_object);
636 vm_object_page_clean(vp->v_object, 0, 0, 0);
637 VM_OBJECT_UNLOCK(vp->v_object);
638 }
639 mtx_lock(&np->n_mtx);
640 if (np->n_flag & NMODIFIED) {
641 mtx_unlock(&np->n_mtx);
642 if (NFS_ISV3(vp)) {
643 /*
644 * Under NFSv3 we have dirty buffers to dispose of. We
645 * must flush them to the NFS server. We have the option
646 * of waiting all the way through the commit rpc or just
647 * waiting for the initial write. The default is to only
648 * wait through the initial write so the data is in the
649 * server's cache, which is roughly similar to the state
650 * a standard disk subsystem leaves the file in on close().
651 *
652 * We cannot clear the NMODIFIED bit in np->n_flag due to
653 * potential races with other processes, and certainly
654 * cannot clear it if we don't commit.
655 */
656 int cm = nfsv3_commit_on_close ? 1 : 0;
657 error = nfs_flush(vp, MNT_WAIT, cm);
658 /* np->n_flag &= ~NMODIFIED; */
659 } else
660 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
661 mtx_lock(&np->n_mtx);
662 }
663 if (np->n_flag & NWRITEERR) {
664 np->n_flag &= ~NWRITEERR;
665 error = np->n_error;
666 }
667 mtx_unlock(&np->n_mtx);
668 }
669 if (nfs_directio_enable)
670 KASSERT((np->n_directio_asyncwr == 0),
671 ("nfs_close: dirty unflushed (%d) directio buffers\n",
672 np->n_directio_asyncwr));
673 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
674 mtx_lock(&np->n_mtx);
675 KASSERT((np->n_directio_opens > 0),
676 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
677 np->n_directio_opens--;
678 if (np->n_directio_opens == 0)
679 np->n_flag &= ~NNONCACHE;
680 mtx_unlock(&np->n_mtx);
681 }
682 return (error);
683 }
684
685 /*
686 * nfs getattr call from vfs.
687 */
688 static int
689 nfs_getattr(struct vop_getattr_args *ap)
690 {
691 struct vnode *vp = ap->a_vp;
692 struct nfsnode *np = VTONFS(vp);
693 struct thread *td = curthread;
694 struct vattr *vap = ap->a_vap;
695 struct vattr vattr;
696 caddr_t bpos, dpos;
697 int error = 0;
698 struct mbuf *mreq, *mrep, *md, *mb;
699 int v3 = NFS_ISV3(vp);
700
701 /*
702 * Update local times for special files.
703 */
704 mtx_lock(&np->n_mtx);
705 if (np->n_flag & (NACC | NUPD))
706 np->n_flag |= NCHG;
707 mtx_unlock(&np->n_mtx);
708 /*
709 * First look in the cache.
710 */
711 if (nfs_getattrcache(vp, &vattr) == 0)
712 goto nfsmout;
713 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
714 nfsstats.accesscache_misses++;
715 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL);
716 if (nfs_getattrcache(vp, &vattr) == 0)
717 goto nfsmout;
718 }
719 nfsstats.rpccnt[NFSPROC_GETATTR]++;
720 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
721 mb = mreq;
722 bpos = mtod(mb, caddr_t);
723 nfsm_fhtom(vp, v3);
724 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
725 if (!error) {
726 nfsm_loadattr(vp, &vattr);
727 }
728 m_freem(mrep);
729 nfsmout:
730 vap->va_type = vattr.va_type;
731 vap->va_mode = vattr.va_mode;
732 vap->va_nlink = vattr.va_nlink;
733 vap->va_uid = vattr.va_uid;
734 vap->va_gid = vattr.va_gid;
735 vap->va_fsid = vattr.va_fsid;
736 vap->va_fileid = vattr.va_fileid;
737 vap->va_size = vattr.va_size;
738 vap->va_blocksize = vattr.va_blocksize;
739 vap->va_atime = vattr.va_atime;
740 vap->va_mtime = vattr.va_mtime;
741 vap->va_ctime = vattr.va_ctime;
742 vap->va_gen = vattr.va_gen;
743 vap->va_flags = vattr.va_flags;
744 vap->va_rdev = vattr.va_rdev;
745 vap->va_bytes = vattr.va_bytes;
746 vap->va_filerev = vattr.va_filerev;
747
748 return (error);
749 }
750
751 /*
752 * nfs setattr call.
753 */
754 static int
755 nfs_setattr(struct vop_setattr_args *ap)
756 {
757 struct vnode *vp = ap->a_vp;
758 struct nfsnode *np = VTONFS(vp);
759 struct vattr *vap = ap->a_vap;
760 struct thread *td = curthread;
761 int error = 0;
762 u_quad_t tsize;
763
764 #ifndef nolint
765 tsize = (u_quad_t)0;
766 #endif
767
768 /*
769 * Setting of flags is not supported.
770 */
771 if (vap->va_flags != VNOVAL)
772 return (EOPNOTSUPP);
773
774 /*
775 * Disallow write attempts if the filesystem is mounted read-only.
776 */
777 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
778 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
779 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
780 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
781 error = EROFS;
782 goto out;
783 }
784 if (vap->va_size != VNOVAL) {
785 switch (vp->v_type) {
786 case VDIR:
787 return (EISDIR);
788 case VCHR:
789 case VBLK:
790 case VSOCK:
791 case VFIFO:
792 if (vap->va_mtime.tv_sec == VNOVAL &&
793 vap->va_atime.tv_sec == VNOVAL &&
794 vap->va_mode == (mode_t)VNOVAL &&
795 vap->va_uid == (uid_t)VNOVAL &&
796 vap->va_gid == (gid_t)VNOVAL)
797 return (0);
798 vap->va_size = VNOVAL;
799 break;
800 default:
801 /*
802 * Disallow write attempts if the filesystem is
803 * mounted read-only.
804 */
805 if (vp->v_mount->mnt_flag & MNT_RDONLY)
806 return (EROFS);
807 /*
808 * We run vnode_pager_setsize() early (why?),
809 * we must set np->n_size now to avoid vinvalbuf
810 * V_SAVE races that might setsize a lower
811 * value.
812 */
813 mtx_lock(&np->n_mtx);
814 tsize = np->n_size;
815 mtx_unlock(&np->n_mtx);
816 error = nfs_meta_setsize(vp, ap->a_cred, td,
817 vap->va_size);
818 mtx_lock(&np->n_mtx);
819 if (np->n_flag & NMODIFIED) {
820 tsize = np->n_size;
821 mtx_unlock(&np->n_mtx);
822 if (vap->va_size == 0)
823 error = nfs_vinvalbuf(vp, 0, td, 1);
824 else
825 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
826 if (error) {
827 vnode_pager_setsize(vp, tsize);
828 goto out;
829 }
830 } else
831 mtx_unlock(&np->n_mtx);
832 /*
833 * np->n_size has already been set to vap->va_size
834 * in nfs_meta_setsize(). We must set it again since
835 * nfs_loadattrcache() could be called through
836 * nfs_meta_setsize() and could modify np->n_size.
837 */
838 mtx_lock(&np->n_mtx);
839 np->n_vattr.va_size = np->n_size = vap->va_size;
840 mtx_unlock(&np->n_mtx);
841 };
842 } else {
843 mtx_lock(&np->n_mtx);
844 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
845 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
846 mtx_unlock(&np->n_mtx);
847 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
848 (error == EINTR || error == EIO))
849 return error;
850 } else
851 mtx_unlock(&np->n_mtx);
852 }
853 error = nfs_setattrrpc(vp, vap, ap->a_cred);
854 if (error && vap->va_size != VNOVAL) {
855 mtx_lock(&np->n_mtx);
856 np->n_size = np->n_vattr.va_size = tsize;
857 vnode_pager_setsize(vp, tsize);
858 mtx_unlock(&np->n_mtx);
859 }
860 out:
861 return (error);
862 }
863
864 /*
865 * Do an nfs setattr rpc.
866 */
867 static int
868 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
869 {
870 struct nfsv2_sattr *sp;
871 struct nfsnode *np = VTONFS(vp);
872 caddr_t bpos, dpos;
873 u_int32_t *tl;
874 int error = 0, i, wccflag = NFSV3_WCCRATTR;
875 struct mbuf *mreq, *mrep, *md, *mb;
876 int v3 = NFS_ISV3(vp);
877
878 nfsstats.rpccnt[NFSPROC_SETATTR]++;
879 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
880 mb = mreq;
881 bpos = mtod(mb, caddr_t);
882 nfsm_fhtom(vp, v3);
883 if (v3) {
884 nfsm_v3attrbuild(vap, TRUE);
885 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
886 *tl = nfs_false;
887 } else {
888 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
889 if (vap->va_mode == (mode_t)VNOVAL)
890 sp->sa_mode = nfs_xdrneg1;
891 else
892 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
893 if (vap->va_uid == (uid_t)VNOVAL)
894 sp->sa_uid = nfs_xdrneg1;
895 else
896 sp->sa_uid = txdr_unsigned(vap->va_uid);
897 if (vap->va_gid == (gid_t)VNOVAL)
898 sp->sa_gid = nfs_xdrneg1;
899 else
900 sp->sa_gid = txdr_unsigned(vap->va_gid);
901 sp->sa_size = txdr_unsigned(vap->va_size);
902 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
903 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
904 }
905 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
906 if (v3) {
907 mtx_lock(&np->n_mtx);
908 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
909 np->n_accesscache[i].stamp = 0;
910 mtx_unlock(&np->n_mtx);
911 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
912 nfsm_wcc_data(vp, wccflag);
913 } else
914 nfsm_loadattr(vp, NULL);
915 m_freem(mrep);
916 nfsmout:
917 return (error);
918 }
919
920 /*
921 * nfs lookup call, one step at a time...
922 * First look in cache
923 * If not found, unlock the directory nfsnode and do the rpc
924 */
925 static int
926 nfs_lookup(struct vop_lookup_args *ap)
927 {
928 struct componentname *cnp = ap->a_cnp;
929 struct vnode *dvp = ap->a_dvp;
930 struct vnode **vpp = ap->a_vpp;
931 struct mount *mp = dvp->v_mount;
932 struct vattr dvattr, vattr;
933 struct timespec nctime;
934 int flags = cnp->cn_flags;
935 struct vnode *newvp;
936 struct nfsmount *nmp;
937 caddr_t bpos, dpos;
938 struct mbuf *mreq, *mrep, *md, *mb;
939 long len;
940 nfsfh_t *fhp;
941 struct nfsnode *np, *newnp;
942 int error = 0, attrflag, dattrflag, fhsize, ltype, ncticks;
943 int v3 = NFS_ISV3(dvp);
944 struct thread *td = cnp->cn_thread;
945
946 *vpp = NULLVP;
947 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
948 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
949 return (EROFS);
950 if (dvp->v_type != VDIR)
951 return (ENOTDIR);
952 nmp = VFSTONFS(mp);
953 np = VTONFS(dvp);
954 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
955 *vpp = NULLVP;
956 return (error);
957 }
958 error = cache_lookup_times(dvp, vpp, cnp, &nctime, &ncticks);
959 if (error > 0 && error != ENOENT)
960 return (error);
961 if (error == -1) {
962 /*
963 * Lookups of "." are special and always return the
964 * current directory. cache_lookup() already handles
965 * associated locking bookkeeping, etc.
966 */
967 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
968 /* XXX: Is this really correct? */
969 if (cnp->cn_nameiop != LOOKUP &&
970 (flags & ISLASTCN))
971 cnp->cn_flags |= SAVENAME;
972 return (0);
973 }
974
975 /*
976 * We only accept a positive hit in the cache if the
977 * change time of the file matches our cached copy.
978 * Otherwise, we discard the cache entry and fallback
979 * to doing a lookup RPC. We also only trust cache
980 * entries for less than nm_nametimeo seconds.
981 *
982 * To better handle stale file handles and attributes,
983 * clear the attribute cache of this node if it is a
984 * leaf component, part of an open() call, and not
985 * locally modified before fetching the attributes.
986 * This should allow stale file handles to be detected
987 * here where we can fall back to a LOOKUP RPC to
988 * recover rather than having nfs_open() detect the
989 * stale file handle and failing open(2) with ESTALE.
990 */
991 newvp = *vpp;
992 newnp = VTONFS(newvp);
993 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
994 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
995 !(newnp->n_flag & NMODIFIED)) {
996 mtx_lock(&newnp->n_mtx);
997 newnp->n_attrstamp = 0;
998 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
999 mtx_unlock(&newnp->n_mtx);
1000 }
1001 if ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1002 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1003 timespeccmp(&vattr.va_ctime, &nctime, ==)) {
1004 nfsstats.lookupcache_hits++;
1005 if (cnp->cn_nameiop != LOOKUP &&
1006 (flags & ISLASTCN))
1007 cnp->cn_flags |= SAVENAME;
1008 return (0);
1009 }
1010 cache_purge(newvp);
1011 if (dvp != newvp)
1012 vput(newvp);
1013 else
1014 vrele(newvp);
1015 *vpp = NULLVP;
1016 } else if (error == ENOENT) {
1017 if (dvp->v_iflag & VI_DOOMED)
1018 return (ENOENT);
1019 /*
1020 * We only accept a negative hit in the cache if the
1021 * modification time of the parent directory matches
1022 * the cached copy in the name cache entry.
1023 * Otherwise, we discard all of the negative cache
1024 * entries for this directory. We also only trust
1025 * negative cache entries for up to nm_negnametimeo
1026 * seconds.
1027 */
1028 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1029 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1030 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1031 nfsstats.lookupcache_hits++;
1032 return (ENOENT);
1033 }
1034 cache_purge_negative(dvp);
1035 }
1036
1037 attrflag = dattrflag = 0;
1038 error = 0;
1039 newvp = NULLVP;
1040 nfsstats.lookupcache_misses++;
1041 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1042 len = cnp->cn_namelen;
1043 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1044 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1045 mb = mreq;
1046 bpos = mtod(mb, caddr_t);
1047 nfsm_fhtom(dvp, v3);
1048 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1049 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
1050 if (error) {
1051 if (v3) {
1052 nfsm_postop_attr_va(dvp, dattrflag, &vattr);
1053 m_freem(mrep);
1054 }
1055 goto nfsmout;
1056 }
1057 nfsm_getfh(fhp, fhsize, v3);
1058
1059 /*
1060 * Handle RENAME case...
1061 */
1062 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1063 if (NFS_CMPFH(np, fhp, fhsize)) {
1064 m_freem(mrep);
1065 return (EISDIR);
1066 }
1067 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE);
1068 if (error) {
1069 m_freem(mrep);
1070 return (error);
1071 }
1072 newvp = NFSTOV(np);
1073 if (v3) {
1074 nfsm_postop_attr(newvp, attrflag);
1075 nfsm_postop_attr(dvp, attrflag);
1076 } else
1077 nfsm_loadattr(newvp, NULL);
1078 *vpp = newvp;
1079 m_freem(mrep);
1080 cnp->cn_flags |= SAVENAME;
1081 return (0);
1082 }
1083
1084 if (flags & ISDOTDOT) {
1085 ltype = VOP_ISLOCKED(dvp);
1086 error = vfs_busy(mp, MBF_NOWAIT);
1087 if (error != 0) {
1088 vfs_ref(mp);
1089 VOP_UNLOCK(dvp, 0);
1090 error = vfs_busy(mp, 0);
1091 vn_lock(dvp, ltype | LK_RETRY);
1092 vfs_rel(mp);
1093 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1094 vfs_unbusy(mp);
1095 error = ENOENT;
1096 }
1097 if (error != 0) {
1098 m_freem(mrep);
1099 return (error);
1100 }
1101 }
1102 VOP_UNLOCK(dvp, 0);
1103 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1104 if (error == 0)
1105 newvp = NFSTOV(np);
1106 vfs_unbusy(mp);
1107 if (newvp != dvp)
1108 vn_lock(dvp, ltype | LK_RETRY);
1109 if (dvp->v_iflag & VI_DOOMED) {
1110 if (error == 0) {
1111 if (newvp == dvp)
1112 vrele(newvp);
1113 else
1114 vput(newvp);
1115 }
1116 error = ENOENT;
1117 }
1118 if (error) {
1119 m_freem(mrep);
1120 return (error);
1121 }
1122 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1123 VREF(dvp);
1124 newvp = dvp;
1125 } else {
1126 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1127 if (error) {
1128 m_freem(mrep);
1129 return (error);
1130 }
1131 newvp = NFSTOV(np);
1132
1133 /*
1134 * Flush the attribute cache when opening a leaf node
1135 * to ensure that fresh attributes are fetched in
1136 * nfs_open() if we are unable to fetch attributes
1137 * from the LOOKUP reply.
1138 */
1139 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1140 !(np->n_flag & NMODIFIED)) {
1141 mtx_lock(&np->n_mtx);
1142 np->n_attrstamp = 0;
1143 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1144 mtx_unlock(&np->n_mtx);
1145 }
1146 }
1147 if (v3) {
1148 nfsm_postop_attr_va(newvp, attrflag, &vattr);
1149 nfsm_postop_attr_va(dvp, dattrflag, &dvattr);
1150 } else {
1151 nfsm_loadattr(newvp, &vattr);
1152 attrflag = 1;
1153 }
1154 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1155 cnp->cn_flags |= SAVENAME;
1156 if ((cnp->cn_flags & MAKEENTRY) &&
1157 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1158 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1159 cache_enter_time(dvp, newvp, cnp, &vattr.va_ctime,
1160 newvp->v_type != VDIR ? NULL : &dvattr.va_ctime);
1161 *vpp = newvp;
1162 m_freem(mrep);
1163 nfsmout:
1164 if (error) {
1165 if (newvp != NULLVP) {
1166 vput(newvp);
1167 *vpp = NULLVP;
1168 }
1169
1170 if (error != ENOENT)
1171 goto done;
1172
1173 /* The requested file was not found. */
1174 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1175 (flags & ISLASTCN)) {
1176 /*
1177 * XXX: UFS does a full VOP_ACCESS(dvp,
1178 * VWRITE) here instead of just checking
1179 * MNT_RDONLY.
1180 */
1181 if (mp->mnt_flag & MNT_RDONLY)
1182 return (EROFS);
1183 cnp->cn_flags |= SAVENAME;
1184 return (EJUSTRETURN);
1185 }
1186
1187 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
1188 dattrflag) {
1189 /*
1190 * Cache the modification time of the parent
1191 * directory from the post-op attributes in
1192 * the name cache entry. The negative cache
1193 * entry will be ignored once the directory
1194 * has changed. Don't bother adding the entry
1195 * if the directory has already changed.
1196 */
1197 mtx_lock(&np->n_mtx);
1198 if (timespeccmp(&np->n_vattr.va_mtime,
1199 &vattr.va_mtime, ==)) {
1200 mtx_unlock(&np->n_mtx);
1201 cache_enter_time(dvp, NULL, cnp,
1202 &vattr.va_mtime, NULL);
1203 } else
1204 mtx_unlock(&np->n_mtx);
1205 }
1206 return (ENOENT);
1207 }
1208 done:
1209 return (error);
1210 }
1211
1212 /*
1213 * nfs read call.
1214 * Just call nfs_bioread() to do the work.
1215 */
1216 static int
1217 nfs_read(struct vop_read_args *ap)
1218 {
1219 struct vnode *vp = ap->a_vp;
1220
1221 switch (vp->v_type) {
1222 case VREG:
1223 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1224 case VDIR:
1225 return (EISDIR);
1226 default:
1227 return (EOPNOTSUPP);
1228 }
1229 }
1230
1231 /*
1232 * nfs readlink call
1233 */
1234 static int
1235 nfs_readlink(struct vop_readlink_args *ap)
1236 {
1237 struct vnode *vp = ap->a_vp;
1238
1239 if (vp->v_type != VLNK)
1240 return (EINVAL);
1241 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1242 }
1243
1244 /*
1245 * Do a readlink rpc.
1246 * Called by nfs_doio() from below the buffer cache.
1247 */
1248 int
1249 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1250 {
1251 caddr_t bpos, dpos;
1252 int error = 0, len, attrflag;
1253 struct mbuf *mreq, *mrep, *md, *mb;
1254 int v3 = NFS_ISV3(vp);
1255
1256 nfsstats.rpccnt[NFSPROC_READLINK]++;
1257 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1258 mb = mreq;
1259 bpos = mtod(mb, caddr_t);
1260 nfsm_fhtom(vp, v3);
1261 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1262 if (v3)
1263 nfsm_postop_attr(vp, attrflag);
1264 if (!error) {
1265 nfsm_strsiz(len, NFS_MAXPATHLEN);
1266 if (len == NFS_MAXPATHLEN) {
1267 struct nfsnode *np = VTONFS(vp);
1268 mtx_lock(&np->n_mtx);
1269 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1270 len = np->n_size;
1271 mtx_unlock(&np->n_mtx);
1272 }
1273 nfsm_mtouio(uiop, len);
1274 }
1275 m_freem(mrep);
1276 nfsmout:
1277 return (error);
1278 }
1279
1280 /*
1281 * nfs read rpc call
1282 * Ditto above
1283 */
1284 int
1285 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1286 {
1287 u_int32_t *tl;
1288 caddr_t bpos, dpos;
1289 struct mbuf *mreq, *mrep, *md, *mb;
1290 struct nfsmount *nmp;
1291 off_t end;
1292 int error = 0, len, retlen, tsiz, eof, attrflag;
1293 int v3 = NFS_ISV3(vp);
1294 int rsize;
1295
1296 #ifndef nolint
1297 eof = 0;
1298 #endif
1299 nmp = VFSTONFS(vp->v_mount);
1300 tsiz = uiop->uio_resid;
1301 mtx_lock(&nmp->nm_mtx);
1302 end = uiop->uio_offset + tsiz;
1303 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) {
1304 mtx_unlock(&nmp->nm_mtx);
1305 return (EFBIG);
1306 }
1307 rsize = nmp->nm_rsize;
1308 mtx_unlock(&nmp->nm_mtx);
1309 while (tsiz > 0) {
1310 nfsstats.rpccnt[NFSPROC_READ]++;
1311 len = (tsiz > rsize) ? rsize : tsiz;
1312 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1313 mb = mreq;
1314 bpos = mtod(mb, caddr_t);
1315 nfsm_fhtom(vp, v3);
1316 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1317 if (v3) {
1318 txdr_hyper(uiop->uio_offset, tl);
1319 *(tl + 2) = txdr_unsigned(len);
1320 } else {
1321 *tl++ = txdr_unsigned(uiop->uio_offset);
1322 *tl++ = txdr_unsigned(len);
1323 *tl = 0;
1324 }
1325 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1326 if (v3) {
1327 nfsm_postop_attr(vp, attrflag);
1328 if (error) {
1329 m_freem(mrep);
1330 goto nfsmout;
1331 }
1332 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1333 eof = fxdr_unsigned(int, *(tl + 1));
1334 } else {
1335 nfsm_loadattr(vp, NULL);
1336 }
1337 nfsm_strsiz(retlen, rsize);
1338 nfsm_mtouio(uiop, retlen);
1339 m_freem(mrep);
1340 tsiz -= retlen;
1341 if (v3) {
1342 if (eof || retlen == 0) {
1343 tsiz = 0;
1344 }
1345 } else if (retlen < len) {
1346 tsiz = 0;
1347 }
1348 }
1349 nfsmout:
1350 return (error);
1351 }
1352
1353 /*
1354 * nfs write call
1355 */
1356 int
1357 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1358 int *iomode, int *must_commit)
1359 {
1360 u_int32_t *tl;
1361 int32_t backup;
1362 caddr_t bpos, dpos;
1363 struct mbuf *mreq, *mrep, *md, *mb;
1364 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1365 off_t end;
1366 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1367 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1368 int wsize;
1369
1370 KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1"));
1371 *must_commit = 0;
1372 tsiz = uiop->uio_resid;
1373 mtx_lock(&nmp->nm_mtx);
1374 end = uiop->uio_offset + tsiz;
1375 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) {
1376 mtx_unlock(&nmp->nm_mtx);
1377 return (EFBIG);
1378 }
1379 wsize = nmp->nm_wsize;
1380 mtx_unlock(&nmp->nm_mtx);
1381 while (tsiz > 0) {
1382 nfsstats.rpccnt[NFSPROC_WRITE]++;
1383 len = (tsiz > wsize) ? wsize : tsiz;
1384 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1385 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1386 mb = mreq;
1387 bpos = mtod(mb, caddr_t);
1388 nfsm_fhtom(vp, v3);
1389 if (v3) {
1390 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1391 txdr_hyper(uiop->uio_offset, tl);
1392 tl += 2;
1393 *tl++ = txdr_unsigned(len);
1394 *tl++ = txdr_unsigned(*iomode);
1395 *tl = txdr_unsigned(len);
1396 } else {
1397 u_int32_t x;
1398
1399 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1400 /* Set both "begin" and "current" to non-garbage. */
1401 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1402 *tl++ = x; /* "begin offset" */
1403 *tl++ = x; /* "current offset" */
1404 x = txdr_unsigned(len);
1405 *tl++ = x; /* total to this offset */
1406 *tl = x; /* size of this write */
1407 }
1408 nfsm_uiotom(uiop, len);
1409 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1410 if (v3) {
1411 wccflag = NFSV3_WCCCHK;
1412 nfsm_wcc_data(vp, wccflag);
1413 if (!error) {
1414 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1415 + NFSX_V3WRITEVERF);
1416 rlen = fxdr_unsigned(int, *tl++);
1417 if (rlen == 0) {
1418 error = NFSERR_IO;
1419 m_freem(mrep);
1420 break;
1421 } else if (rlen < len) {
1422 backup = len - rlen;
1423 uiop->uio_iov->iov_base =
1424 (char *)uiop->uio_iov->iov_base -
1425 backup;
1426 uiop->uio_iov->iov_len += backup;
1427 uiop->uio_offset -= backup;
1428 uiop->uio_resid += backup;
1429 len = rlen;
1430 }
1431 commit = fxdr_unsigned(int, *tl++);
1432
1433 /*
1434 * Return the lowest committment level
1435 * obtained by any of the RPCs.
1436 */
1437 if (committed == NFSV3WRITE_FILESYNC)
1438 committed = commit;
1439 else if (committed == NFSV3WRITE_DATASYNC &&
1440 commit == NFSV3WRITE_UNSTABLE)
1441 committed = commit;
1442 mtx_lock(&nmp->nm_mtx);
1443 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1444 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1445 NFSX_V3WRITEVERF);
1446 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1447 } else if (bcmp((caddr_t)tl,
1448 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1449 *must_commit = 1;
1450 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1451 NFSX_V3WRITEVERF);
1452 }
1453 mtx_unlock(&nmp->nm_mtx);
1454 }
1455 } else {
1456 nfsm_loadattr(vp, NULL);
1457 }
1458 if (wccflag) {
1459 mtx_lock(&(VTONFS(vp))->n_mtx);
1460 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1461 mtx_unlock(&(VTONFS(vp))->n_mtx);
1462 }
1463 m_freem(mrep);
1464 if (error)
1465 break;
1466 tsiz -= len;
1467 }
1468 nfsmout:
1469 if (DOINGASYNC(vp))
1470 committed = NFSV3WRITE_FILESYNC;
1471 *iomode = committed;
1472 if (error)
1473 uiop->uio_resid = tsiz;
1474 return (error);
1475 }
1476
1477 /*
1478 * nfs mknod rpc
1479 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1480 * mode set to specify the file type and the size field for rdev.
1481 */
1482 static int
1483 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1484 struct vattr *vap)
1485 {
1486 struct nfsv2_sattr *sp;
1487 u_int32_t *tl;
1488 struct vnode *newvp = NULL;
1489 struct nfsnode *np = NULL;
1490 struct vattr vattr;
1491 caddr_t bpos, dpos;
1492 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1493 struct mbuf *mreq, *mrep, *md, *mb;
1494 u_int32_t rdev;
1495 int v3 = NFS_ISV3(dvp);
1496
1497 if (vap->va_type == VCHR || vap->va_type == VBLK)
1498 rdev = txdr_unsigned(vap->va_rdev);
1499 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1500 rdev = nfs_xdrneg1;
1501 else {
1502 return (EOPNOTSUPP);
1503 }
1504 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1505 return (error);
1506 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1507 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1508 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1509 mb = mreq;
1510 bpos = mtod(mb, caddr_t);
1511 nfsm_fhtom(dvp, v3);
1512 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1513 if (v3) {
1514 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1515 *tl++ = vtonfsv3_type(vap->va_type);
1516 nfsm_v3attrbuild(vap, FALSE);
1517 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1518 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1519 *tl++ = txdr_unsigned(major(vap->va_rdev));
1520 *tl = txdr_unsigned(minor(vap->va_rdev));
1521 }
1522 } else {
1523 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1524 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1525 sp->sa_uid = nfs_xdrneg1;
1526 sp->sa_gid = nfs_xdrneg1;
1527 sp->sa_size = rdev;
1528 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1529 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1530 }
1531 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1532 if (!error) {
1533 nfsm_mtofh(dvp, newvp, v3, gotvp);
1534 if (!gotvp) {
1535 if (newvp) {
1536 vput(newvp);
1537 newvp = NULL;
1538 }
1539 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1540 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1541 if (!error)
1542 newvp = NFSTOV(np);
1543 }
1544 }
1545 if (v3)
1546 nfsm_wcc_data(dvp, wccflag);
1547 m_freem(mrep);
1548 nfsmout:
1549 if (error) {
1550 if (newvp)
1551 vput(newvp);
1552 } else {
1553 *vpp = newvp;
1554 }
1555 mtx_lock(&(VTONFS(dvp))->n_mtx);
1556 VTONFS(dvp)->n_flag |= NMODIFIED;
1557 if (!wccflag) {
1558 VTONFS(dvp)->n_attrstamp = 0;
1559 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1560 }
1561 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1562 return (error);
1563 }
1564
1565 /*
1566 * nfs mknod vop
1567 * just call nfs_mknodrpc() to do the work.
1568 */
1569 /* ARGSUSED */
1570 static int
1571 nfs_mknod(struct vop_mknod_args *ap)
1572 {
1573 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1574 }
1575
1576 static u_long create_verf;
1577 /*
1578 * nfs file create call
1579 */
1580 static int
1581 nfs_create(struct vop_create_args *ap)
1582 {
1583 struct vnode *dvp = ap->a_dvp;
1584 struct vattr *vap = ap->a_vap;
1585 struct componentname *cnp = ap->a_cnp;
1586 struct nfsv2_sattr *sp;
1587 u_int32_t *tl;
1588 struct nfsnode *np = NULL;
1589 struct vnode *newvp = NULL;
1590 caddr_t bpos, dpos;
1591 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1592 struct mbuf *mreq, *mrep, *md, *mb;
1593 struct vattr vattr;
1594 int v3 = NFS_ISV3(dvp);
1595
1596 /*
1597 * Oops, not for me..
1598 */
1599 if (vap->va_type == VSOCK) {
1600 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1601 return (error);
1602 }
1603
1604 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) {
1605 return (error);
1606 }
1607 if (vap->va_vaflags & VA_EXCLUSIVE)
1608 fmode |= O_EXCL;
1609 again:
1610 nfsstats.rpccnt[NFSPROC_CREATE]++;
1611 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1612 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1613 mb = mreq;
1614 bpos = mtod(mb, caddr_t);
1615 nfsm_fhtom(dvp, v3);
1616 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1617 if (v3) {
1618 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1619 if (fmode & O_EXCL) {
1620 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1621 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1622 #ifdef INET
1623 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1624 IN_IFADDR_RLOCK();
1625 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1626 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1627 else
1628 #endif
1629 *tl++ = create_verf;
1630 #ifdef INET
1631 IN_IFADDR_RUNLOCK();
1632 CURVNET_RESTORE();
1633 #endif
1634 *tl = ++create_verf;
1635 } else {
1636 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1637 nfsm_v3attrbuild(vap, FALSE);
1638 }
1639 } else {
1640 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1641 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1642 sp->sa_uid = nfs_xdrneg1;
1643 sp->sa_gid = nfs_xdrneg1;
1644 sp->sa_size = 0;
1645 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1646 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1647 }
1648 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1649 if (!error) {
1650 nfsm_mtofh(dvp, newvp, v3, gotvp);
1651 if (!gotvp) {
1652 if (newvp) {
1653 vput(newvp);
1654 newvp = NULL;
1655 }
1656 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1657 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1658 if (!error)
1659 newvp = NFSTOV(np);
1660 }
1661 }
1662 if (v3)
1663 nfsm_wcc_data(dvp, wccflag);
1664 m_freem(mrep);
1665 nfsmout:
1666 if (error) {
1667 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1668 fmode &= ~O_EXCL;
1669 goto again;
1670 }
1671 if (newvp)
1672 vput(newvp);
1673 } else if (v3 && (fmode & O_EXCL)) {
1674 /*
1675 * We are normally called with only a partially initialized
1676 * VAP. Since the NFSv3 spec says that server may use the
1677 * file attributes to store the verifier, the spec requires
1678 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1679 * in atime, but we can't really assume that all servers will
1680 * so we ensure that our SETATTR sets both atime and mtime.
1681 */
1682 if (vap->va_mtime.tv_sec == VNOVAL)
1683 vfs_timestamp(&vap->va_mtime);
1684 if (vap->va_atime.tv_sec == VNOVAL)
1685 vap->va_atime = vap->va_mtime;
1686 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred);
1687 if (error)
1688 vput(newvp);
1689 }
1690 if (!error) {
1691 *ap->a_vpp = newvp;
1692 }
1693 mtx_lock(&(VTONFS(dvp))->n_mtx);
1694 VTONFS(dvp)->n_flag |= NMODIFIED;
1695 if (!wccflag) {
1696 VTONFS(dvp)->n_attrstamp = 0;
1697 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1698 }
1699 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1700 return (error);
1701 }
1702
1703 /*
1704 * nfs file remove call
1705 * To try and make nfs semantics closer to ufs semantics, a file that has
1706 * other processes using the vnode is renamed instead of removed and then
1707 * removed later on the last close.
1708 * - If v_usecount > 1
1709 * If a rename is not already in the works
1710 * call nfs_sillyrename() to set it up
1711 * else
1712 * do the remove rpc
1713 */
1714 static int
1715 nfs_remove(struct vop_remove_args *ap)
1716 {
1717 struct vnode *vp = ap->a_vp;
1718 struct vnode *dvp = ap->a_dvp;
1719 struct componentname *cnp = ap->a_cnp;
1720 struct nfsnode *np = VTONFS(vp);
1721 int error = 0;
1722 struct vattr vattr;
1723
1724 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1725 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1726 if (vp->v_type == VDIR)
1727 error = EPERM;
1728 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1729 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1730 /*
1731 * Purge the name cache so that the chance of a lookup for
1732 * the name succeeding while the remove is in progress is
1733 * minimized. Without node locking it can still happen, such
1734 * that an I/O op returns ESTALE, but since you get this if
1735 * another host removes the file..
1736 */
1737 cache_purge(vp);
1738 /*
1739 * throw away biocache buffers, mainly to avoid
1740 * unnecessary delayed writes later.
1741 */
1742 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1743 /* Do the rpc */
1744 if (error != EINTR && error != EIO)
1745 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1746 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1747 /*
1748 * Kludge City: If the first reply to the remove rpc is lost..
1749 * the reply to the retransmitted request will be ENOENT
1750 * since the file was in fact removed
1751 * Therefore, we cheat and return success.
1752 */
1753 if (error == ENOENT)
1754 error = 0;
1755 } else if (!np->n_sillyrename)
1756 error = nfs_sillyrename(dvp, vp, cnp);
1757 mtx_lock(&np->n_mtx);
1758 np->n_attrstamp = 0;
1759 mtx_unlock(&np->n_mtx);
1760 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1761 return (error);
1762 }
1763
1764 /*
1765 * nfs file remove rpc called from nfs_inactive
1766 */
1767 int
1768 nfs_removeit(struct sillyrename *sp)
1769 {
1770 /*
1771 * Make sure that the directory vnode is still valid.
1772 * XXX we should lock sp->s_dvp here.
1773 */
1774 if (sp->s_dvp->v_type == VBAD)
1775 return (0);
1776 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1777 NULL));
1778 }
1779
1780 /*
1781 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1782 */
1783 static int
1784 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1785 struct ucred *cred, struct thread *td)
1786 {
1787 caddr_t bpos, dpos;
1788 int error = 0, wccflag = NFSV3_WCCRATTR;
1789 struct mbuf *mreq, *mrep, *md, *mb;
1790 int v3 = NFS_ISV3(dvp);
1791
1792 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1793 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1794 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1795 mb = mreq;
1796 bpos = mtod(mb, caddr_t);
1797 nfsm_fhtom(dvp, v3);
1798 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1799 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1800 if (v3)
1801 nfsm_wcc_data(dvp, wccflag);
1802 m_freem(mrep);
1803 nfsmout:
1804 mtx_lock(&(VTONFS(dvp))->n_mtx);
1805 VTONFS(dvp)->n_flag |= NMODIFIED;
1806 if (!wccflag) {
1807 VTONFS(dvp)->n_attrstamp = 0;
1808 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1809 }
1810 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1811 return (error);
1812 }
1813
1814 /*
1815 * nfs file rename call
1816 */
1817 static int
1818 nfs_rename(struct vop_rename_args *ap)
1819 {
1820 struct vnode *fvp = ap->a_fvp;
1821 struct vnode *tvp = ap->a_tvp;
1822 struct vnode *fdvp = ap->a_fdvp;
1823 struct vnode *tdvp = ap->a_tdvp;
1824 struct componentname *tcnp = ap->a_tcnp;
1825 struct componentname *fcnp = ap->a_fcnp;
1826 int error;
1827
1828 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1829 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1830 /* Check for cross-device rename */
1831 if ((fvp->v_mount != tdvp->v_mount) ||
1832 (tvp && (fvp->v_mount != tvp->v_mount))) {
1833 error = EXDEV;
1834 goto out;
1835 }
1836
1837 if (fvp == tvp) {
1838 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1839 error = 0;
1840 goto out;
1841 }
1842 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1843 goto out;
1844
1845 /*
1846 * We have to flush B_DELWRI data prior to renaming
1847 * the file. If we don't, the delayed-write buffers
1848 * can be flushed out later after the file has gone stale
1849 * under NFSV3. NFSV2 does not have this problem because
1850 * ( as far as I can tell ) it flushes dirty buffers more
1851 * often.
1852 *
1853 * Skip the rename operation if the fsync fails, this can happen
1854 * due to the server's volume being full, when we pushed out data
1855 * that was written back to our cache earlier. Not checking for
1856 * this condition can result in potential (silent) data loss.
1857 */
1858 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1859 VOP_UNLOCK(fvp, 0);
1860 if (!error && tvp)
1861 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1862 if (error)
1863 goto out;
1864
1865 /*
1866 * If the tvp exists and is in use, sillyrename it before doing the
1867 * rename of the new file over it.
1868 * XXX Can't sillyrename a directory.
1869 */
1870 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1871 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1872 vput(tvp);
1873 tvp = NULL;
1874 }
1875
1876 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1877 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1878 tcnp->cn_thread);
1879
1880 if (fvp->v_type == VDIR) {
1881 if (tvp != NULL && tvp->v_type == VDIR)
1882 cache_purge(tdvp);
1883 cache_purge(fdvp);
1884 }
1885
1886 out:
1887 if (tdvp == tvp)
1888 vrele(tdvp);
1889 else
1890 vput(tdvp);
1891 if (tvp)
1892 vput(tvp);
1893 vrele(fdvp);
1894 vrele(fvp);
1895 /*
1896 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1897 */
1898 if (error == ENOENT)
1899 error = 0;
1900 return (error);
1901 }
1902
1903 /*
1904 * nfs file rename rpc called from nfs_remove() above
1905 */
1906 static int
1907 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1908 struct sillyrename *sp)
1909 {
1910
1911 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1912 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1913 }
1914
1915 /*
1916 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1917 */
1918 static int
1919 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1920 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1921 struct thread *td)
1922 {
1923 caddr_t bpos, dpos;
1924 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1925 struct mbuf *mreq, *mrep, *md, *mb;
1926 int v3 = NFS_ISV3(fdvp);
1927
1928 nfsstats.rpccnt[NFSPROC_RENAME]++;
1929 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1930 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1931 nfsm_rndup(tnamelen));
1932 mb = mreq;
1933 bpos = mtod(mb, caddr_t);
1934 nfsm_fhtom(fdvp, v3);
1935 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1936 nfsm_fhtom(tdvp, v3);
1937 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1938 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1939 if (v3) {
1940 nfsm_wcc_data(fdvp, fwccflag);
1941 nfsm_wcc_data(tdvp, twccflag);
1942 }
1943 m_freem(mrep);
1944 nfsmout:
1945 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1946 VTONFS(fdvp)->n_flag |= NMODIFIED;
1947 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1948 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1949 VTONFS(tdvp)->n_flag |= NMODIFIED;
1950 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1951 if (!fwccflag) {
1952 VTONFS(fdvp)->n_attrstamp = 0;
1953 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1954 }
1955 if (!twccflag) {
1956 VTONFS(tdvp)->n_attrstamp = 0;
1957 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1958 }
1959 return (error);
1960 }
1961
1962 /*
1963 * nfs hard link create call
1964 */
1965 static int
1966 nfs_link(struct vop_link_args *ap)
1967 {
1968 struct vnode *vp = ap->a_vp;
1969 struct vnode *tdvp = ap->a_tdvp;
1970 struct componentname *cnp = ap->a_cnp;
1971 caddr_t bpos, dpos;
1972 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1973 struct mbuf *mreq, *mrep, *md, *mb;
1974 int v3;
1975
1976 if (vp->v_mount != tdvp->v_mount) {
1977 return (EXDEV);
1978 }
1979
1980 /*
1981 * Push all writes to the server, so that the attribute cache
1982 * doesn't get "out of sync" with the server.
1983 * XXX There should be a better way!
1984 */
1985 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1986
1987 v3 = NFS_ISV3(vp);
1988 nfsstats.rpccnt[NFSPROC_LINK]++;
1989 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1990 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1991 mb = mreq;
1992 bpos = mtod(mb, caddr_t);
1993 nfsm_fhtom(vp, v3);
1994 nfsm_fhtom(tdvp, v3);
1995 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1996 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1997 if (v3) {
1998 nfsm_postop_attr(vp, attrflag);
1999 nfsm_wcc_data(tdvp, wccflag);
2000 }
2001 m_freem(mrep);
2002 nfsmout:
2003 mtx_lock(&(VTONFS(tdvp))->n_mtx);
2004 VTONFS(tdvp)->n_flag |= NMODIFIED;
2005 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
2006 if (!attrflag) {
2007 VTONFS(vp)->n_attrstamp = 0;
2008 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2009 }
2010 if (!wccflag) {
2011 VTONFS(tdvp)->n_attrstamp = 0;
2012 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2013 }
2014 return (error);
2015 }
2016
2017 /*
2018 * nfs symbolic link create call
2019 */
2020 static int
2021 nfs_symlink(struct vop_symlink_args *ap)
2022 {
2023 struct vnode *dvp = ap->a_dvp;
2024 struct vattr *vap = ap->a_vap;
2025 struct componentname *cnp = ap->a_cnp;
2026 struct nfsv2_sattr *sp;
2027 caddr_t bpos, dpos;
2028 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2029 struct mbuf *mreq, *mrep, *md, *mb;
2030 struct vnode *newvp = NULL;
2031 int v3 = NFS_ISV3(dvp);
2032
2033 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2034 slen = strlen(ap->a_target);
2035 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2036 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2037 mb = mreq;
2038 bpos = mtod(mb, caddr_t);
2039 nfsm_fhtom(dvp, v3);
2040 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2041 if (v3) {
2042 nfsm_v3attrbuild(vap, FALSE);
2043 }
2044 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2045 if (!v3) {
2046 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2047 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2048 sp->sa_uid = nfs_xdrneg1;
2049 sp->sa_gid = nfs_xdrneg1;
2050 sp->sa_size = nfs_xdrneg1;
2051 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2052 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2053 }
2054
2055 /*
2056 * Issue the NFS request and get the rpc response.
2057 *
2058 * Only NFSv3 responses returning an error of 0 actually return
2059 * a file handle that can be converted into newvp without having
2060 * to do an extra lookup rpc.
2061 */
2062 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
2063 if (v3) {
2064 if (error == 0)
2065 nfsm_mtofh(dvp, newvp, v3, gotvp);
2066 nfsm_wcc_data(dvp, wccflag);
2067 }
2068
2069 /*
2070 * out code jumps -> here, mrep is also freed.
2071 */
2072
2073 m_freem(mrep);
2074 nfsmout:
2075
2076 /*
2077 * If we do not have an error and we could not extract the newvp from
2078 * the response due to the request being NFSv2, we have to do a
2079 * lookup in order to obtain a newvp to return.
2080 */
2081 if (error == 0 && newvp == NULL) {
2082 struct nfsnode *np = NULL;
2083
2084 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2085 cnp->cn_cred, cnp->cn_thread, &np);
2086 if (!error)
2087 newvp = NFSTOV(np);
2088 }
2089 if (error) {
2090 if (newvp)
2091 vput(newvp);
2092 } else {
2093 *ap->a_vpp = newvp;
2094 }
2095 mtx_lock(&(VTONFS(dvp))->n_mtx);
2096 VTONFS(dvp)->n_flag |= NMODIFIED;
2097 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2098 if (!wccflag) {
2099 VTONFS(dvp)->n_attrstamp = 0;
2100 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2101 }
2102 return (error);
2103 }
2104
2105 /*
2106 * nfs make dir call
2107 */
2108 static int
2109 nfs_mkdir(struct vop_mkdir_args *ap)
2110 {
2111 struct vnode *dvp = ap->a_dvp;
2112 struct vattr *vap = ap->a_vap;
2113 struct componentname *cnp = ap->a_cnp;
2114 struct nfsv2_sattr *sp;
2115 int len;
2116 struct nfsnode *np = NULL;
2117 struct vnode *newvp = NULL;
2118 caddr_t bpos, dpos;
2119 int error = 0, wccflag = NFSV3_WCCRATTR;
2120 int gotvp = 0;
2121 struct mbuf *mreq, *mrep, *md, *mb;
2122 struct vattr vattr;
2123 int v3 = NFS_ISV3(dvp);
2124
2125 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2126 return (error);
2127 len = cnp->cn_namelen;
2128 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2129 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
2130 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2131 mb = mreq;
2132 bpos = mtod(mb, caddr_t);
2133 nfsm_fhtom(dvp, v3);
2134 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2135 if (v3) {
2136 nfsm_v3attrbuild(vap, FALSE);
2137 } else {
2138 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2139 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2140 sp->sa_uid = nfs_xdrneg1;
2141 sp->sa_gid = nfs_xdrneg1;
2142 sp->sa_size = nfs_xdrneg1;
2143 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2144 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2145 }
2146 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2147 if (!error)
2148 nfsm_mtofh(dvp, newvp, v3, gotvp);
2149 if (v3)
2150 nfsm_wcc_data(dvp, wccflag);
2151 m_freem(mrep);
2152 nfsmout:
2153 mtx_lock(&(VTONFS(dvp))->n_mtx);
2154 VTONFS(dvp)->n_flag |= NMODIFIED;
2155 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2156 if (!wccflag) {
2157 VTONFS(dvp)->n_attrstamp = 0;
2158 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2159 }
2160 if (error == 0 && newvp == NULL) {
2161 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2162 cnp->cn_thread, &np);
2163 if (!error) {
2164 newvp = NFSTOV(np);
2165 if (newvp->v_type != VDIR)
2166 error = EEXIST;
2167 }
2168 }
2169 if (error) {
2170 if (newvp)
2171 vput(newvp);
2172 } else
2173 *ap->a_vpp = newvp;
2174 return (error);
2175 }
2176
2177 /*
2178 * nfs remove directory call
2179 */
2180 static int
2181 nfs_rmdir(struct vop_rmdir_args *ap)
2182 {
2183 struct vnode *vp = ap->a_vp;
2184 struct vnode *dvp = ap->a_dvp;
2185 struct componentname *cnp = ap->a_cnp;
2186 caddr_t bpos, dpos;
2187 int error = 0, wccflag = NFSV3_WCCRATTR;
2188 struct mbuf *mreq, *mrep, *md, *mb;
2189 int v3 = NFS_ISV3(dvp);
2190
2191 if (dvp == vp)
2192 return (EINVAL);
2193 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2194 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
2195 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2196 mb = mreq;
2197 bpos = mtod(mb, caddr_t);
2198 nfsm_fhtom(dvp, v3);
2199 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2200 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2201 if (v3)
2202 nfsm_wcc_data(dvp, wccflag);
2203 m_freem(mrep);
2204 nfsmout:
2205 mtx_lock(&(VTONFS(dvp))->n_mtx);
2206 VTONFS(dvp)->n_flag |= NMODIFIED;
2207 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2208 if (!wccflag) {
2209 VTONFS(dvp)->n_attrstamp = 0;
2210 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2211 }
2212 cache_purge(dvp);
2213 cache_purge(vp);
2214 /*
2215 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2216 */
2217 if (error == ENOENT)
2218 error = 0;
2219 return (error);
2220 }
2221
2222 /*
2223 * nfs readdir call
2224 */
2225 static int
2226 nfs_readdir(struct vop_readdir_args *ap)
2227 {
2228 struct vnode *vp = ap->a_vp;
2229 struct nfsnode *np = VTONFS(vp);
2230 struct uio *uio = ap->a_uio;
2231 int tresid, error = 0;
2232 struct vattr vattr;
2233
2234 if (vp->v_type != VDIR)
2235 return(EPERM);
2236
2237 /*
2238 * First, check for hit on the EOF offset cache
2239 */
2240 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2241 (np->n_flag & NMODIFIED) == 0) {
2242 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2243 mtx_lock(&np->n_mtx);
2244 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2245 mtx_unlock(&np->n_mtx);
2246 nfsstats.direofcache_hits++;
2247 goto out;
2248 } else
2249 mtx_unlock(&np->n_mtx);
2250 }
2251 }
2252
2253 /*
2254 * Call nfs_bioread() to do the real work.
2255 */
2256 tresid = uio->uio_resid;
2257 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2258
2259 if (!error && uio->uio_resid == tresid) {
2260 nfsstats.direofcache_misses++;
2261 }
2262 out:
2263 return (error);
2264 }
2265
2266 /*
2267 * Readdir rpc call.
2268 * Called from below the buffer cache by nfs_doio().
2269 */
2270 int
2271 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2272 {
2273 int len, left;
2274 struct dirent *dp = NULL;
2275 u_int32_t *tl;
2276 caddr_t cp;
2277 nfsuint64 *cookiep;
2278 caddr_t bpos, dpos;
2279 struct mbuf *mreq, *mrep, *md, *mb;
2280 nfsuint64 cookie;
2281 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2282 struct nfsnode *dnp = VTONFS(vp);
2283 u_quad_t fileno;
2284 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2285 int attrflag;
2286 int v3 = NFS_ISV3(vp);
2287
2288 KASSERT(uiop->uio_iovcnt == 1 &&
2289 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2290 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2291 ("nfs readdirrpc bad uio"));
2292
2293 /*
2294 * If there is no cookie, assume directory was stale.
2295 */
2296 nfs_dircookie_lock(dnp);
2297 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2298 if (cookiep) {
2299 cookie = *cookiep;
2300 nfs_dircookie_unlock(dnp);
2301 } else {
2302 nfs_dircookie_unlock(dnp);
2303 return (NFSERR_BAD_COOKIE);
2304 }
2305
2306 /*
2307 * Loop around doing readdir rpc's of size nm_readdirsize
2308 * truncated to a multiple of DIRBLKSIZ.
2309 * The stopping criteria is EOF or buffer full.
2310 */
2311 while (more_dirs && bigenough) {
2312 nfsstats.rpccnt[NFSPROC_READDIR]++;
2313 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2314 NFSX_READDIR(v3));
2315 mb = mreq;
2316 bpos = mtod(mb, caddr_t);
2317 nfsm_fhtom(vp, v3);
2318 if (v3) {
2319 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2320 *tl++ = cookie.nfsuquad[0];
2321 *tl++ = cookie.nfsuquad[1];
2322 mtx_lock(&dnp->n_mtx);
2323 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2324 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2325 mtx_unlock(&dnp->n_mtx);
2326 } else {
2327 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2328 *tl++ = cookie.nfsuquad[0];
2329 }
2330 *tl = txdr_unsigned(nmp->nm_readdirsize);
2331 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2332 if (v3) {
2333 nfsm_postop_attr(vp, attrflag);
2334 if (!error) {
2335 tl = nfsm_dissect(u_int32_t *,
2336 2 * NFSX_UNSIGNED);
2337 mtx_lock(&dnp->n_mtx);
2338 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2339 dnp->n_cookieverf.nfsuquad[1] = *tl;
2340 mtx_unlock(&dnp->n_mtx);
2341 } else {
2342 m_freem(mrep);
2343 goto nfsmout;
2344 }
2345 }
2346 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2347 more_dirs = fxdr_unsigned(int, *tl);
2348
2349 /* loop thru the dir entries, doctoring them to 4bsd form */
2350 while (more_dirs && bigenough) {
2351 if (v3) {
2352 tl = nfsm_dissect(u_int32_t *,
2353 3 * NFSX_UNSIGNED);
2354 fileno = fxdr_hyper(tl);
2355 len = fxdr_unsigned(int, *(tl + 2));
2356 } else {
2357 tl = nfsm_dissect(u_int32_t *,
2358 2 * NFSX_UNSIGNED);
2359 fileno = fxdr_unsigned(u_quad_t, *tl++);
2360 len = fxdr_unsigned(int, *tl);
2361 }
2362 if (len <= 0 || len > NFS_MAXNAMLEN) {
2363 error = EBADRPC;
2364 m_freem(mrep);
2365 goto nfsmout;
2366 }
2367 tlen = nfsm_rndup(len);
2368 if (tlen == len)
2369 tlen += 4; /* To ensure null termination */
2370 left = DIRBLKSIZ - blksiz;
2371 if ((tlen + DIRHDSIZ) > left) {
2372 dp->d_reclen += left;
2373 uiop->uio_iov->iov_base =
2374 (char *)uiop->uio_iov->iov_base + left;
2375 uiop->uio_iov->iov_len -= left;
2376 uiop->uio_offset += left;
2377 uiop->uio_resid -= left;
2378 blksiz = 0;
2379 }
2380 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2381 bigenough = 0;
2382 if (bigenough) {
2383 dp = (struct dirent *)uiop->uio_iov->iov_base;
2384 dp->d_fileno = (int)fileno;
2385 dp->d_namlen = len;
2386 dp->d_reclen = tlen + DIRHDSIZ;
2387 dp->d_type = DT_UNKNOWN;
2388 blksiz += dp->d_reclen;
2389 if (blksiz == DIRBLKSIZ)
2390 blksiz = 0;
2391 uiop->uio_offset += DIRHDSIZ;
2392 uiop->uio_resid -= DIRHDSIZ;
2393 uiop->uio_iov->iov_base =
2394 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2395 uiop->uio_iov->iov_len -= DIRHDSIZ;
2396 nfsm_mtouio(uiop, len);
2397 cp = uiop->uio_iov->iov_base;
2398 tlen -= len;
2399 *cp = '\0'; /* null terminate */
2400 uiop->uio_iov->iov_base =
2401 (char *)uiop->uio_iov->iov_base + tlen;
2402 uiop->uio_iov->iov_len -= tlen;
2403 uiop->uio_offset += tlen;
2404 uiop->uio_resid -= tlen;
2405 } else
2406 nfsm_adv(nfsm_rndup(len));
2407 if (v3) {
2408 tl = nfsm_dissect(u_int32_t *,
2409 3 * NFSX_UNSIGNED);
2410 } else {
2411 tl = nfsm_dissect(u_int32_t *,
2412 2 * NFSX_UNSIGNED);
2413 }
2414 if (bigenough) {
2415 cookie.nfsuquad[0] = *tl++;
2416 if (v3)
2417 cookie.nfsuquad[1] = *tl++;
2418 } else if (v3)
2419 tl += 2;
2420 else
2421 tl++;
2422 more_dirs = fxdr_unsigned(int, *tl);
2423 }
2424 /*
2425 * If at end of rpc data, get the eof boolean
2426 */
2427 if (!more_dirs) {
2428 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2429 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2430 }
2431 m_freem(mrep);
2432 }
2433 /*
2434 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2435 * by increasing d_reclen for the last record.
2436 */
2437 if (blksiz > 0) {
2438 left = DIRBLKSIZ - blksiz;
2439 dp->d_reclen += left;
2440 uiop->uio_iov->iov_base =
2441 (char *)uiop->uio_iov->iov_base + left;
2442 uiop->uio_iov->iov_len -= left;
2443 uiop->uio_offset += left;
2444 uiop->uio_resid -= left;
2445 }
2446
2447 /*
2448 * We are now either at the end of the directory or have filled the
2449 * block.
2450 */
2451 if (bigenough)
2452 dnp->n_direofoffset = uiop->uio_offset;
2453 else {
2454 if (uiop->uio_resid > 0)
2455 nfs_printf("EEK! readdirrpc resid > 0\n");
2456 nfs_dircookie_lock(dnp);
2457 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2458 *cookiep = cookie;
2459 nfs_dircookie_unlock(dnp);
2460 }
2461 nfsmout:
2462 return (error);
2463 }
2464
2465 /*
2466 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2467 */
2468 int
2469 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2470 {
2471 int len, left;
2472 struct dirent *dp;
2473 u_int32_t *tl;
2474 caddr_t cp;
2475 struct vnode *newvp;
2476 nfsuint64 *cookiep;
2477 caddr_t bpos, dpos, dpossav1, dpossav2;
2478 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2479 struct nameidata nami, *ndp = &nami;
2480 struct componentname *cnp = &ndp->ni_cnd;
2481 nfsuint64 cookie;
2482 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2483 struct nfsnode *dnp = VTONFS(vp), *np;
2484 struct vattr vattr, dvattr;
2485 nfsfh_t *fhp;
2486 u_quad_t fileno;
2487 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2488 int attrflag, dattrflag, fhsize;
2489
2490 #ifndef nolint
2491 dp = NULL;
2492 #endif
2493 KASSERT(uiop->uio_iovcnt == 1 &&
2494 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2495 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2496 ("nfs readdirplusrpc bad uio"));
2497 ndp->ni_dvp = vp;
2498 newvp = NULLVP;
2499
2500 /*
2501 * If there is no cookie, assume directory was stale.
2502 */
2503 nfs_dircookie_lock(dnp);
2504 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2505 if (cookiep) {
2506 cookie = *cookiep;
2507 nfs_dircookie_unlock(dnp);
2508 } else {
2509 nfs_dircookie_unlock(dnp);
2510 return (NFSERR_BAD_COOKIE);
2511 }
2512 /*
2513 * Loop around doing readdir rpc's of size nm_readdirsize
2514 * truncated to a multiple of DIRBLKSIZ.
2515 * The stopping criteria is EOF or buffer full.
2516 */
2517 while (more_dirs && bigenough) {
2518 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2519 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2520 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2521 mb = mreq;
2522 bpos = mtod(mb, caddr_t);
2523 nfsm_fhtom(vp, 1);
2524 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2525 *tl++ = cookie.nfsuquad[0];
2526 *tl++ = cookie.nfsuquad[1];
2527 mtx_lock(&dnp->n_mtx);
2528 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2529 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2530 mtx_unlock(&dnp->n_mtx);
2531 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2532 *tl = txdr_unsigned(nmp->nm_rsize);
2533 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2534 nfsm_postop_attr_va(vp, dattrflag, &dvattr);
2535 if (error) {
2536 m_freem(mrep);
2537 goto nfsmout;
2538 }
2539 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2540 mtx_lock(&dnp->n_mtx);
2541 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2542 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2543 mtx_unlock(&dnp->n_mtx);
2544 more_dirs = fxdr_unsigned(int, *tl);
2545
2546 /* loop thru the dir entries, doctoring them to 4bsd form */
2547 while (more_dirs && bigenough) {
2548 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2549 fileno = fxdr_hyper(tl);
2550 len = fxdr_unsigned(int, *(tl + 2));
2551 if (len <= 0 || len > NFS_MAXNAMLEN) {
2552 error = EBADRPC;
2553 m_freem(mrep);
2554 goto nfsmout;
2555 }
2556 tlen = nfsm_rndup(len);
2557 if (tlen == len)
2558 tlen += 4; /* To ensure null termination*/
2559 left = DIRBLKSIZ - blksiz;
2560 if ((tlen + DIRHDSIZ) > left) {
2561 dp->d_reclen += left;
2562 uiop->uio_iov->iov_base =
2563 (char *)uiop->uio_iov->iov_base + left;
2564 uiop->uio_iov->iov_len -= left;
2565 uiop->uio_offset += left;
2566 uiop->uio_resid -= left;
2567 blksiz = 0;
2568 }
2569 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2570 bigenough = 0;
2571 if (bigenough) {
2572 dp = (struct dirent *)uiop->uio_iov->iov_base;
2573 dp->d_fileno = (int)fileno;
2574 dp->d_namlen = len;
2575 dp->d_reclen = tlen + DIRHDSIZ;
2576 dp->d_type = DT_UNKNOWN;
2577 blksiz += dp->d_reclen;
2578 if (blksiz == DIRBLKSIZ)
2579 blksiz = 0;
2580 uiop->uio_offset += DIRHDSIZ;
2581 uiop->uio_resid -= DIRHDSIZ;
2582 uiop->uio_iov->iov_base =
2583 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2584 uiop->uio_iov->iov_len -= DIRHDSIZ;
2585 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2586 cnp->cn_namelen = len;
2587 nfsm_mtouio(uiop, len);
2588 cp = uiop->uio_iov->iov_base;
2589 tlen -= len;
2590 *cp = '\0';
2591 uiop->uio_iov->iov_base =
2592 (char *)uiop->uio_iov->iov_base + tlen;
2593 uiop->uio_iov->iov_len -= tlen;
2594 uiop->uio_offset += tlen;
2595 uiop->uio_resid -= tlen;
2596 } else
2597 nfsm_adv(nfsm_rndup(len));
2598 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2599 if (bigenough) {
2600 cookie.nfsuquad[0] = *tl++;
2601 cookie.nfsuquad[1] = *tl++;
2602 } else
2603 tl += 2;
2604
2605 /*
2606 * Since the attributes are before the file handle
2607 * (sigh), we must skip over the attributes and then
2608 * come back and get them.
2609 */
2610 attrflag = fxdr_unsigned(int, *tl);
2611 if (attrflag) {
2612 dpossav1 = dpos;
2613 mdsav1 = md;
2614 nfsm_adv(NFSX_V3FATTR);
2615 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2616 doit = fxdr_unsigned(int, *tl);
2617 /*
2618 * Skip loading the attrs for "..". There's a
2619 * race between loading the attrs here and
2620 * lookups that look for the directory currently
2621 * being read (in the parent). We try to acquire
2622 * the exclusive lock on ".." here, owning the
2623 * lock on the directory being read. Lookup will
2624 * hold the lock on ".." and try to acquire the
2625 * lock on the directory being read.
2626 *
2627 * There are other ways of fixing this, one would
2628 * be to do a trylock on the ".." vnode and skip
2629 * loading the attrs on ".." if it happens to be
2630 * locked by another process. But skipping the
2631 * attrload on ".." seems the easiest option.
2632 */
2633 if (strcmp(dp->d_name, "..") == 0) {
2634 doit = 0;
2635 /*
2636 * We've already skipped over the attrs,
2637 * skip over the filehandle. And store d_type
2638 * as VDIR.
2639 */
2640 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2641 i = fxdr_unsigned(int, *tl);
2642 nfsm_adv(nfsm_rndup(i));
2643 dp->d_type = IFTODT(VTTOIF(VDIR));
2644 }
2645 if (doit) {
2646 nfsm_getfh(fhp, fhsize, 1);
2647 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2648 VREF(vp);
2649 newvp = vp;
2650 np = dnp;
2651 } else {
2652 error = nfs_nget(vp->v_mount, fhp,
2653 fhsize, &np, LK_EXCLUSIVE);
2654 if (error)
2655 doit = 0;
2656 else
2657 newvp = NFSTOV(np);
2658 }
2659 }
2660 if (doit && bigenough) {
2661 dpossav2 = dpos;
2662 dpos = dpossav1;
2663 mdsav2 = md;
2664 md = mdsav1;
2665 nfsm_loadattr(newvp, &vattr);
2666 dpos = dpossav2;
2667 md = mdsav2;
2668 dp->d_type = IFTODT(VTTOIF(vattr.va_type));
2669 ndp->ni_vp = newvp;
2670 if (newvp->v_type != VDIR || dattrflag != 0)
2671 cache_enter_time(ndp->ni_dvp, ndp->ni_vp,
2672 cnp, &vattr.va_ctime,
2673 newvp->v_type != VDIR ? NULL :
2674 &dvattr.va_ctime);
2675 }
2676 } else {
2677 /* Just skip over the file handle */
2678 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2679 i = fxdr_unsigned(int, *tl);
2680 if (i) {
2681 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2682 fhsize = fxdr_unsigned(int, *tl);
2683 nfsm_adv(nfsm_rndup(fhsize));
2684 }
2685 }
2686 if (newvp != NULLVP) {
2687 if (newvp == vp)
2688 vrele(newvp);
2689 else
2690 vput(newvp);
2691 newvp = NULLVP;
2692 }
2693 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2694 more_dirs = fxdr_unsigned(int, *tl);
2695 }
2696 /*
2697 * If at end of rpc data, get the eof boolean
2698 */
2699 if (!more_dirs) {
2700 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2701 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2702 }
2703 m_freem(mrep);
2704 }
2705 /*
2706 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2707 * by increasing d_reclen for the last record.
2708 */
2709 if (blksiz > 0) {
2710 left = DIRBLKSIZ - blksiz;
2711 dp->d_reclen += left;
2712 uiop->uio_iov->iov_base =
2713 (char *)uiop->uio_iov->iov_base + left;
2714 uiop->uio_iov->iov_len -= left;
2715 uiop->uio_offset += left;
2716 uiop->uio_resid -= left;
2717 }
2718
2719 /*
2720 * We are now either at the end of the directory or have filled the
2721 * block.
2722 */
2723 if (bigenough)
2724 dnp->n_direofoffset = uiop->uio_offset;
2725 else {
2726 if (uiop->uio_resid > 0)
2727 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2728 nfs_dircookie_lock(dnp);
2729 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2730 *cookiep = cookie;
2731 nfs_dircookie_unlock(dnp);
2732 }
2733 nfsmout:
2734 if (newvp != NULLVP) {
2735 if (newvp == vp)
2736 vrele(newvp);
2737 else
2738 vput(newvp);
2739 newvp = NULLVP;
2740 }
2741 return (error);
2742 }
2743
2744 /*
2745 * Silly rename. To make the NFS filesystem that is stateless look a little
2746 * more like the "ufs" a remove of an active vnode is translated to a rename
2747 * to a funny looking filename that is removed by nfs_inactive on the
2748 * nfsnode. There is the potential for another process on a different client
2749 * to create the same funny name between the nfs_lookitup() fails and the
2750 * nfs_rename() completes, but...
2751 */
2752 static int
2753 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2754 {
2755 struct sillyrename *sp;
2756 struct nfsnode *np;
2757 int error;
2758 short pid;
2759 unsigned int lticks;
2760
2761 cache_purge(dvp);
2762 np = VTONFS(vp);
2763 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2764 sp = malloc(sizeof (struct sillyrename),
2765 M_NFSREQ, M_WAITOK);
2766 sp->s_cred = crhold(cnp->cn_cred);
2767 sp->s_dvp = dvp;
2768 sp->s_removeit = nfs_removeit;
2769 VREF(dvp);
2770
2771 /*
2772 * Fudge together a funny name.
2773 * Changing the format of the funny name to accomodate more
2774 * sillynames per directory.
2775 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2776 * CPU ticks since boot.
2777 */
2778 pid = cnp->cn_thread->td_proc->p_pid;
2779 lticks = (unsigned int)ticks;
2780 for ( ; ; ) {
2781 sp->s_namlen = sprintf(sp->s_name,
2782 ".nfs.%08x.%04x4.4", lticks,
2783 pid);
2784 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2785 cnp->cn_thread, NULL))
2786 break;
2787 lticks++;
2788 }
2789 error = nfs_renameit(dvp, cnp, sp);
2790 if (error)
2791 goto bad;
2792 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2793 cnp->cn_thread, &np);
2794 np->n_sillyrename = sp;
2795 return (0);
2796 bad:
2797 vrele(sp->s_dvp);
2798 crfree(sp->s_cred);
2799 free((caddr_t)sp, M_NFSREQ);
2800 return (error);
2801 }
2802
2803 /*
2804 * Look up a file name and optionally either update the file handle or
2805 * allocate an nfsnode, depending on the value of npp.
2806 * npp == NULL --> just do the lookup
2807 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2808 * handled too
2809 * *npp != NULL --> update the file handle in the vnode
2810 */
2811 static int
2812 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2813 struct thread *td, struct nfsnode **npp)
2814 {
2815 struct vnode *newvp = NULL;
2816 struct nfsnode *np, *dnp = VTONFS(dvp);
2817 caddr_t bpos, dpos;
2818 int error = 0, fhlen, attrflag;
2819 struct mbuf *mreq, *mrep, *md, *mb;
2820 nfsfh_t *nfhp;
2821 int v3 = NFS_ISV3(dvp);
2822
2823 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2824 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2825 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2826 mb = mreq;
2827 bpos = mtod(mb, caddr_t);
2828 nfsm_fhtom(dvp, v3);
2829 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2830 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2831 if (npp && !error) {
2832 nfsm_getfh(nfhp, fhlen, v3);
2833 if (*npp) {
2834 np = *npp;
2835 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2836 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2837 np->n_fhp = &np->n_fh;
2838 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2839 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2840 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2841 np->n_fhsize = fhlen;
2842 newvp = NFSTOV(np);
2843 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2844 VREF(dvp);
2845 newvp = dvp;
2846 } else {
2847 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2848 if (error) {
2849 m_freem(mrep);
2850 return (error);
2851 }
2852 newvp = NFSTOV(np);
2853 }
2854 if (v3) {
2855 nfsm_postop_attr(newvp, attrflag);
2856 if (!attrflag && *npp == NULL) {
2857 m_freem(mrep);
2858 if (newvp == dvp)
2859 vrele(newvp);
2860 else
2861 vput(newvp);
2862 return (ENOENT);
2863 }
2864 } else
2865 nfsm_loadattr(newvp, NULL);
2866 }
2867 m_freem(mrep);
2868 nfsmout:
2869 if (npp && *npp == NULL) {
2870 if (error) {
2871 if (newvp) {
2872 if (newvp == dvp)
2873 vrele(newvp);
2874 else
2875 vput(newvp);
2876 }
2877 } else
2878 *npp = np;
2879 }
2880 return (error);
2881 }
2882
2883 /*
2884 * Nfs Version 3 commit rpc
2885 */
2886 int
2887 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2888 struct thread *td)
2889 {
2890 u_int32_t *tl;
2891 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2892 caddr_t bpos, dpos;
2893 int error = 0, wccflag = NFSV3_WCCRATTR;
2894 struct mbuf *mreq, *mrep, *md, *mb;
2895
2896 mtx_lock(&nmp->nm_mtx);
2897 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2898 mtx_unlock(&nmp->nm_mtx);
2899 return (0);
2900 }
2901 mtx_unlock(&nmp->nm_mtx);
2902 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2903 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2904 mb = mreq;
2905 bpos = mtod(mb, caddr_t);
2906 nfsm_fhtom(vp, 1);
2907 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2908 txdr_hyper(offset, tl);
2909 tl += 2;
2910 *tl = txdr_unsigned(cnt);
2911 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2912 nfsm_wcc_data(vp, wccflag);
2913 if (!error) {
2914 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2915 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2916 NFSX_V3WRITEVERF)) {
2917 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2918 NFSX_V3WRITEVERF);
2919 error = NFSERR_STALEWRITEVERF;
2920 }
2921 }
2922 m_freem(mrep);
2923 nfsmout:
2924 return (error);
2925 }
2926
2927 /*
2928 * Strategy routine.
2929 * For async requests when nfsiod(s) are running, queue the request by
2930 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2931 * request.
2932 */
2933 static int
2934 nfs_strategy(struct vop_strategy_args *ap)
2935 {
2936 struct buf *bp = ap->a_bp;
2937 struct ucred *cr;
2938
2939 KASSERT(!(bp->b_flags & B_DONE),
2940 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2941 BUF_ASSERT_HELD(bp);
2942
2943 if (bp->b_iocmd == BIO_READ)
2944 cr = bp->b_rcred;
2945 else
2946 cr = bp->b_wcred;
2947
2948 /*
2949 * If the op is asynchronous and an i/o daemon is waiting
2950 * queue the request, wake it up and wait for completion
2951 * otherwise just do it ourselves.
2952 */
2953 if ((bp->b_flags & B_ASYNC) == 0 ||
2954 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2955 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2956 return (0);
2957 }
2958
2959 /*
2960 * fsync vnode op. Just call nfs_flush() with commit == 1.
2961 */
2962 /* ARGSUSED */
2963 static int
2964 nfs_fsync(struct vop_fsync_args *ap)
2965 {
2966
2967 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1));
2968 }
2969
2970 /*
2971 * Flush all the blocks associated with a vnode.
2972 * Walk through the buffer pool and push any dirty pages
2973 * associated with the vnode.
2974 */
2975 static int
2976 nfs_flush(struct vnode *vp, int waitfor, int commit)
2977 {
2978 struct nfsnode *np = VTONFS(vp);
2979 struct buf *bp;
2980 int i;
2981 struct buf *nbp;
2982 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2983 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2984 int passone = 1;
2985 u_quad_t off, endoff, toff;
2986 struct ucred* wcred = NULL;
2987 struct buf **bvec = NULL;
2988 struct bufobj *bo;
2989 struct thread *td = curthread;
2990 #ifndef NFS_COMMITBVECSIZ
2991 #define NFS_COMMITBVECSIZ 20
2992 #endif
2993 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2994 int bvecsize = 0, bveccount;
2995
2996 if (nmp->nm_flag & NFSMNT_INT)
2997 slpflag = NFS_PCATCH;
2998 if (!commit)
2999 passone = 0;
3000 bo = &vp->v_bufobj;
3001 /*
3002 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
3003 * server, but has not been committed to stable storage on the server
3004 * yet. On the first pass, the byte range is worked out and the commit
3005 * rpc is done. On the second pass, nfs_writebp() is called to do the
3006 * job.
3007 */
3008 again:
3009 off = (u_quad_t)-1;
3010 endoff = 0;
3011 bvecpos = 0;
3012 if (NFS_ISV3(vp) && commit) {
3013 if (bvec != NULL && bvec != bvec_on_stack)
3014 free(bvec, M_TEMP);
3015 /*
3016 * Count up how many buffers waiting for a commit.
3017 */
3018 bveccount = 0;
3019 BO_LOCK(bo);
3020 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3021 if (!BUF_ISLOCKED(bp) &&
3022 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
3023 == (B_DELWRI | B_NEEDCOMMIT))
3024 bveccount++;
3025 }
3026 /*
3027 * Allocate space to remember the list of bufs to commit. It is
3028 * important to use M_NOWAIT here to avoid a race with nfs_write.
3029 * If we can't get memory (for whatever reason), we will end up
3030 * committing the buffers one-by-one in the loop below.
3031 */
3032 if (bveccount > NFS_COMMITBVECSIZ) {
3033 /*
3034 * Release the vnode interlock to avoid a lock
3035 * order reversal.
3036 */
3037 BO_UNLOCK(bo);
3038 bvec = (struct buf **)
3039 malloc(bveccount * sizeof(struct buf *),
3040 M_TEMP, M_NOWAIT);
3041 BO_LOCK(bo);
3042 if (bvec == NULL) {
3043 bvec = bvec_on_stack;
3044 bvecsize = NFS_COMMITBVECSIZ;
3045 } else
3046 bvecsize = bveccount;
3047 } else {
3048 bvec = bvec_on_stack;
3049 bvecsize = NFS_COMMITBVECSIZ;
3050 }
3051 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3052 if (bvecpos >= bvecsize)
3053 break;
3054 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3055 nbp = TAILQ_NEXT(bp, b_bobufs);
3056 continue;
3057 }
3058 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3059 (B_DELWRI | B_NEEDCOMMIT)) {
3060 BUF_UNLOCK(bp);
3061 nbp = TAILQ_NEXT(bp, b_bobufs);
3062 continue;
3063 }
3064 BO_UNLOCK(bo);
3065 bremfree(bp);
3066 /*
3067 * Work out if all buffers are using the same cred
3068 * so we can deal with them all with one commit.
3069 *
3070 * NOTE: we are not clearing B_DONE here, so we have
3071 * to do it later on in this routine if we intend to
3072 * initiate I/O on the bp.
3073 *
3074 * Note: to avoid loopback deadlocks, we do not
3075 * assign b_runningbufspace.
3076 */
3077 if (wcred == NULL)
3078 wcred = bp->b_wcred;
3079 else if (wcred != bp->b_wcred)
3080 wcred = NOCRED;
3081 vfs_busy_pages(bp, 1);
3082
3083 BO_LOCK(bo);
3084 /*
3085 * bp is protected by being locked, but nbp is not
3086 * and vfs_busy_pages() may sleep. We have to
3087 * recalculate nbp.
3088 */
3089 nbp = TAILQ_NEXT(bp, b_bobufs);
3090
3091 /*
3092 * A list of these buffers is kept so that the
3093 * second loop knows which buffers have actually
3094 * been committed. This is necessary, since there
3095 * may be a race between the commit rpc and new
3096 * uncommitted writes on the file.
3097 */
3098 bvec[bvecpos++] = bp;
3099 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3100 bp->b_dirtyoff;
3101 if (toff < off)
3102 off = toff;
3103 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3104 if (toff > endoff)
3105 endoff = toff;
3106 }
3107 BO_UNLOCK(bo);
3108 }
3109 if (bvecpos > 0) {
3110 /*
3111 * Commit data on the server, as required.
3112 * If all bufs are using the same wcred, then use that with
3113 * one call for all of them, otherwise commit each one
3114 * separately.
3115 */
3116 if (wcred != NOCRED)
3117 retv = nfs_commit(vp, off, (int)(endoff - off),
3118 wcred, td);
3119 else {
3120 retv = 0;
3121 for (i = 0; i < bvecpos; i++) {
3122 off_t off, size;
3123 bp = bvec[i];
3124 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3125 bp->b_dirtyoff;
3126 size = (u_quad_t)(bp->b_dirtyend
3127 - bp->b_dirtyoff);
3128 retv = nfs_commit(vp, off, (int)size,
3129 bp->b_wcred, td);
3130 if (retv) break;
3131 }
3132 }
3133
3134 if (retv == NFSERR_STALEWRITEVERF)
3135 nfs_clearcommit(vp->v_mount);
3136
3137 /*
3138 * Now, either mark the blocks I/O done or mark the
3139 * blocks dirty, depending on whether the commit
3140 * succeeded.
3141 */
3142 for (i = 0; i < bvecpos; i++) {
3143 bp = bvec[i];
3144 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3145 if (retv) {
3146 /*
3147 * Error, leave B_DELWRI intact
3148 */
3149 vfs_unbusy_pages(bp);
3150 brelse(bp);
3151 } else {
3152 /*
3153 * Success, remove B_DELWRI ( bundirty() ).
3154 *
3155 * b_dirtyoff/b_dirtyend seem to be NFS
3156 * specific. We should probably move that
3157 * into bundirty(). XXX
3158 */
3159 bufobj_wref(bo);
3160 bp->b_flags |= B_ASYNC;
3161 bundirty(bp);
3162 bp->b_flags &= ~B_DONE;
3163 bp->b_ioflags &= ~BIO_ERROR;
3164 bp->b_dirtyoff = bp->b_dirtyend = 0;
3165 bufdone(bp);
3166 }
3167 }
3168 }
3169
3170 /*
3171 * Start/do any write(s) that are required.
3172 */
3173 loop:
3174 BO_LOCK(bo);
3175 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3176 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3177 if (waitfor != MNT_WAIT || passone)
3178 continue;
3179
3180 error = BUF_TIMELOCK(bp,
3181 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3182 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
3183 if (error == 0) {
3184 BUF_UNLOCK(bp);
3185 goto loop;
3186 }
3187 if (error == ENOLCK) {
3188 error = 0;
3189 goto loop;
3190 }
3191 if (nfs_sigintr(nmp, td)) {
3192 error = EINTR;
3193 goto done;
3194 }
3195 if (slpflag & PCATCH) {
3196 slpflag = 0;
3197 slptimeo = 2 * hz;
3198 }
3199 goto loop;
3200 }
3201 if ((bp->b_flags & B_DELWRI) == 0)
3202 panic("nfs_fsync: not dirty");
3203 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3204 BUF_UNLOCK(bp);
3205 continue;
3206 }
3207 BO_UNLOCK(bo);
3208 bremfree(bp);
3209 if (passone || !commit)
3210 bp->b_flags |= B_ASYNC;
3211 else
3212 bp->b_flags |= B_ASYNC;
3213 bwrite(bp);
3214 if (nfs_sigintr(nmp, td)) {
3215 error = EINTR;
3216 goto done;
3217 }
3218 goto loop;
3219 }
3220 if (passone) {
3221 passone = 0;
3222 BO_UNLOCK(bo);
3223 goto again;
3224 }
3225 if (waitfor == MNT_WAIT) {
3226 while (bo->bo_numoutput) {
3227 error = bufobj_wwait(bo, slpflag, slptimeo);
3228 if (error) {
3229 BO_UNLOCK(bo);
3230 error = nfs_sigintr(nmp, td);
3231 if (error)
3232 goto done;
3233 if (slpflag & PCATCH) {
3234 slpflag = 0;
3235 slptimeo = 2 * hz;
3236 }
3237 BO_LOCK(bo);
3238 }
3239 }
3240 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3241 BO_UNLOCK(bo);
3242 goto loop;
3243 }
3244 /*
3245 * Wait for all the async IO requests to drain
3246 */
3247 BO_UNLOCK(bo);
3248 mtx_lock(&np->n_mtx);
3249 while (np->n_directio_asyncwr > 0) {
3250 np->n_flag |= NFSYNCWAIT;
3251 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3252 &np->n_mtx, slpflag | (PRIBIO + 1),
3253 "nfsfsync", 0);
3254 if (error) {
3255 if (nfs_sigintr(nmp, td)) {
3256 mtx_unlock(&np->n_mtx);
3257 error = EINTR;
3258 goto done;
3259 }
3260 }
3261 }
3262 mtx_unlock(&np->n_mtx);
3263 } else
3264 BO_UNLOCK(bo);
3265 mtx_lock(&np->n_mtx);
3266 if (np->n_flag & NWRITEERR) {
3267 error = np->n_error;
3268 np->n_flag &= ~NWRITEERR;
3269 }
3270 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3271 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3272 np->n_flag &= ~NMODIFIED;
3273 mtx_unlock(&np->n_mtx);
3274 done:
3275 if (bvec != NULL && bvec != bvec_on_stack)
3276 free(bvec, M_TEMP);
3277 return (error);
3278 }
3279
3280 /*
3281 * NFS advisory byte-level locks.
3282 */
3283 static int
3284 nfs_advlock(struct vop_advlock_args *ap)
3285 {
3286 struct vnode *vp = ap->a_vp;
3287 u_quad_t size;
3288 int error;
3289
3290 error = vn_lock(vp, LK_SHARED);
3291 if (error)
3292 return (error);
3293 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3294 size = VTONFS(vp)->n_size;
3295 VOP_UNLOCK(vp, 0);
3296 error = lf_advlock(ap, &(vp->v_lockf), size);
3297 } else {
3298 if (nfs_advlock_p)
3299 error = nfs_advlock_p(ap);
3300 else
3301 error = ENOLCK;
3302 }
3303
3304 return (error);
3305 }
3306
3307 /*
3308 * NFS advisory byte-level locks.
3309 */
3310 static int
3311 nfs_advlockasync(struct vop_advlockasync_args *ap)
3312 {
3313 struct vnode *vp = ap->a_vp;
3314 u_quad_t size;
3315 int error;
3316
3317 error = vn_lock(vp, LK_SHARED);
3318 if (error)
3319 return (error);
3320 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3321 size = VTONFS(vp)->n_size;
3322 VOP_UNLOCK(vp, 0);
3323 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3324 } else {
3325 VOP_UNLOCK(vp, 0);
3326 error = EOPNOTSUPP;
3327 }
3328 return (error);
3329 }
3330
3331 /*
3332 * Print out the contents of an nfsnode.
3333 */
3334 static int
3335 nfs_print(struct vop_print_args *ap)
3336 {
3337 struct vnode *vp = ap->a_vp;
3338 struct nfsnode *np = VTONFS(vp);
3339
3340 nfs_printf("\tfileid %ld fsid 0x%x",
3341 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3342 if (vp->v_type == VFIFO)
3343 fifo_printinfo(vp);
3344 printf("\n");
3345 return (0);
3346 }
3347
3348 /*
3349 * This is the "real" nfs::bwrite(struct buf*).
3350 * We set B_CACHE if this is a VMIO buffer.
3351 */
3352 int
3353 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3354 {
3355 int s;
3356 int oldflags = bp->b_flags;
3357 #if 0
3358 int retv = 1;
3359 off_t off;
3360 #endif
3361
3362 BUF_ASSERT_HELD(bp);
3363
3364 if (bp->b_flags & B_INVAL) {
3365 brelse(bp);
3366 return(0);
3367 }
3368
3369 bp->b_flags |= B_CACHE;
3370
3371 /*
3372 * Undirty the bp. We will redirty it later if the I/O fails.
3373 */
3374
3375 s = splbio();
3376 bundirty(bp);
3377 bp->b_flags &= ~B_DONE;
3378 bp->b_ioflags &= ~BIO_ERROR;
3379 bp->b_iocmd = BIO_WRITE;
3380
3381 bufobj_wref(bp->b_bufobj);
3382 curthread->td_ru.ru_oublock++;
3383 splx(s);
3384
3385 /*
3386 * Note: to avoid loopback deadlocks, we do not
3387 * assign b_runningbufspace.
3388 */
3389 vfs_busy_pages(bp, 1);
3390
3391 BUF_KERNPROC(bp);
3392 bp->b_iooffset = dbtob(bp->b_blkno);
3393 bstrategy(bp);
3394
3395 if( (oldflags & B_ASYNC) == 0) {
3396 int rtval = bufwait(bp);
3397
3398 if (oldflags & B_DELWRI) {
3399 s = splbio();
3400 reassignbuf(bp);
3401 splx(s);
3402 }
3403 brelse(bp);
3404 return (rtval);
3405 }
3406
3407 return (0);
3408 }
3409
3410 /*
3411 * nfs special file access vnode op.
3412 * Essentially just get vattr and then imitate iaccess() since the device is
3413 * local to the client.
3414 */
3415 static int
3416 nfsspec_access(struct vop_access_args *ap)
3417 {
3418 struct vattr *vap;
3419 struct ucred *cred = ap->a_cred;
3420 struct vnode *vp = ap->a_vp;
3421 accmode_t accmode = ap->a_accmode;
3422 struct vattr vattr;
3423 int error;
3424
3425 /*
3426 * Disallow write attempts on filesystems mounted read-only;
3427 * unless the file is a socket, fifo, or a block or character
3428 * device resident on the filesystem.
3429 */
3430 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3431 switch (vp->v_type) {
3432 case VREG:
3433 case VDIR:
3434 case VLNK:
3435 return (EROFS);
3436 default:
3437 break;
3438 }
3439 }
3440 vap = &vattr;
3441 error = VOP_GETATTR(vp, vap, cred);
3442 if (error)
3443 goto out;
3444 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3445 accmode, cred, NULL);
3446 out:
3447 return error;
3448 }
3449
3450 /*
3451 * Read wrapper for fifos.
3452 */
3453 static int
3454 nfsfifo_read(struct vop_read_args *ap)
3455 {
3456 struct nfsnode *np = VTONFS(ap->a_vp);
3457 int error;
3458
3459 /*
3460 * Set access flag.
3461 */
3462 mtx_lock(&np->n_mtx);
3463 np->n_flag |= NACC;
3464 vfs_timestamp(&np->n_atim);
3465 mtx_unlock(&np->n_mtx);
3466 error = fifo_specops.vop_read(ap);
3467 return error;
3468 }
3469
3470 /*
3471 * Write wrapper for fifos.
3472 */
3473 static int
3474 nfsfifo_write(struct vop_write_args *ap)
3475 {
3476 struct nfsnode *np = VTONFS(ap->a_vp);
3477
3478 /*
3479 * Set update flag.
3480 */
3481 mtx_lock(&np->n_mtx);
3482 np->n_flag |= NUPD;
3483 vfs_timestamp(&np->n_mtim);
3484 mtx_unlock(&np->n_mtx);
3485 return(fifo_specops.vop_write(ap));
3486 }
3487
3488 /*
3489 * Close wrapper for fifos.
3490 *
3491 * Update the times on the nfsnode then do fifo close.
3492 */
3493 static int
3494 nfsfifo_close(struct vop_close_args *ap)
3495 {
3496 struct vnode *vp = ap->a_vp;
3497 struct nfsnode *np = VTONFS(vp);
3498 struct vattr vattr;
3499 struct timespec ts;
3500
3501 mtx_lock(&np->n_mtx);
3502 if (np->n_flag & (NACC | NUPD)) {
3503 vfs_timestamp(&ts);
3504 if (np->n_flag & NACC)
3505 np->n_atim = ts;
3506 if (np->n_flag & NUPD)
3507 np->n_mtim = ts;
3508 np->n_flag |= NCHG;
3509 if (vrefcnt(vp) == 1 &&
3510 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3511 VATTR_NULL(&vattr);
3512 if (np->n_flag & NACC)
3513 vattr.va_atime = np->n_atim;
3514 if (np->n_flag & NUPD)
3515 vattr.va_mtime = np->n_mtim;
3516 mtx_unlock(&np->n_mtx);
3517 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3518 goto out;
3519 }
3520 }
3521 mtx_unlock(&np->n_mtx);
3522 out:
3523 return (fifo_specops.vop_close(ap));
3524 }
3525
3526 /*
3527 * Just call nfs_writebp() with the force argument set to 1.
3528 *
3529 * NOTE: B_DONE may or may not be set in a_bp on call.
3530 */
3531 static int
3532 nfs_bwrite(struct buf *bp)
3533 {
3534
3535 return (nfs_writebp(bp, 1, curthread));
3536 }
3537
3538 struct buf_ops buf_ops_nfs = {
3539 .bop_name = "buf_ops_nfs",
3540 .bop_write = nfs_bwrite,
3541 .bop_strategy = bufstrategy,
3542 .bop_sync = bufsync,
3543 .bop_bdflush = bufbdflush,
3544 };
Cache object: 9e3c733f1d0f2152ab19e58dd49db77d
|