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 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD: releng/9.0/sys/fs/nfsclient/nfs_clvnops.c 224532 2011-07-30 22:57:38Z rmacklem $");
37
38 /*
39 * vnode op calls for Sun NFS version 2, 3 and 4
40 */
41
42 #include "opt_kdtrace.h"
43 #include "opt_inet.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_extern.h>
68 #include <vm/vm_object.h>
69
70 #include <fs/nfs/nfsport.h>
71 #include <fs/nfsclient/nfsnode.h>
72 #include <fs/nfsclient/nfsmount.h>
73 #include <fs/nfsclient/nfs.h>
74 #include <fs/nfsclient/nfs_kdtrace.h>
75
76 #include <net/if.h>
77 #include <netinet/in.h>
78 #include <netinet/in_var.h>
79
80 #include <nfs/nfs_lock.h>
81
82 #ifdef KDTRACE_HOOKS
83 #include <sys/dtrace_bsd.h>
84
85 dtrace_nfsclient_accesscache_flush_probe_func_t
86 dtrace_nfscl_accesscache_flush_done_probe;
87 uint32_t nfscl_accesscache_flush_done_id;
88
89 dtrace_nfsclient_accesscache_get_probe_func_t
90 dtrace_nfscl_accesscache_get_hit_probe,
91 dtrace_nfscl_accesscache_get_miss_probe;
92 uint32_t nfscl_accesscache_get_hit_id;
93 uint32_t nfscl_accesscache_get_miss_id;
94
95 dtrace_nfsclient_accesscache_load_probe_func_t
96 dtrace_nfscl_accesscache_load_done_probe;
97 uint32_t nfscl_accesscache_load_done_id;
98 #endif /* !KDTRACE_HOOKS */
99
100 /* Defs */
101 #define TRUE 1
102 #define FALSE 0
103
104 extern struct nfsstats newnfsstats;
105 extern int nfsrv_useacl;
106 MALLOC_DECLARE(M_NEWNFSREQ);
107
108 /*
109 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
110 * calls are not in getblk() and brelse() so that they would not be necessary
111 * here.
112 */
113 #ifndef B_VMIO
114 #define vfs_busy_pages(bp, f)
115 #endif
116
117 static vop_read_t nfsfifo_read;
118 static vop_write_t nfsfifo_write;
119 static vop_close_t nfsfifo_close;
120 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
121 struct thread *);
122 static vop_lookup_t nfs_lookup;
123 static vop_create_t nfs_create;
124 static vop_mknod_t nfs_mknod;
125 static vop_open_t nfs_open;
126 static vop_pathconf_t nfs_pathconf;
127 static vop_close_t nfs_close;
128 static vop_access_t nfs_access;
129 static vop_getattr_t nfs_getattr;
130 static vop_setattr_t nfs_setattr;
131 static vop_read_t nfs_read;
132 static vop_fsync_t nfs_fsync;
133 static vop_remove_t nfs_remove;
134 static vop_link_t nfs_link;
135 static vop_rename_t nfs_rename;
136 static vop_mkdir_t nfs_mkdir;
137 static vop_rmdir_t nfs_rmdir;
138 static vop_symlink_t nfs_symlink;
139 static vop_readdir_t nfs_readdir;
140 static vop_strategy_t nfs_strategy;
141 static vop_lock1_t nfs_lock1;
142 static int nfs_lookitup(struct vnode *, 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 static vop_getacl_t nfs_getacl;
152 static vop_setacl_t nfs_setacl;
153
154 /*
155 * Global vfs data structures for nfs
156 */
157 struct vop_vector newnfs_vnodeops = {
158 .vop_default = &default_vnodeops,
159 .vop_access = nfs_access,
160 .vop_advlock = nfs_advlock,
161 .vop_advlockasync = nfs_advlockasync,
162 .vop_close = nfs_close,
163 .vop_create = nfs_create,
164 .vop_fsync = nfs_fsync,
165 .vop_getattr = nfs_getattr,
166 .vop_getpages = ncl_getpages,
167 .vop_putpages = ncl_putpages,
168 .vop_inactive = ncl_inactive,
169 .vop_link = nfs_link,
170 .vop_lock1 = nfs_lock1,
171 .vop_lookup = nfs_lookup,
172 .vop_mkdir = nfs_mkdir,
173 .vop_mknod = nfs_mknod,
174 .vop_open = nfs_open,
175 .vop_pathconf = nfs_pathconf,
176 .vop_print = nfs_print,
177 .vop_read = nfs_read,
178 .vop_readdir = nfs_readdir,
179 .vop_readlink = nfs_readlink,
180 .vop_reclaim = ncl_reclaim,
181 .vop_remove = nfs_remove,
182 .vop_rename = nfs_rename,
183 .vop_rmdir = nfs_rmdir,
184 .vop_setattr = nfs_setattr,
185 .vop_strategy = nfs_strategy,
186 .vop_symlink = nfs_symlink,
187 .vop_write = ncl_write,
188 .vop_getacl = nfs_getacl,
189 .vop_setacl = nfs_setacl,
190 };
191
192 struct vop_vector newnfs_fifoops = {
193 .vop_default = &fifo_specops,
194 .vop_access = nfsspec_access,
195 .vop_close = nfsfifo_close,
196 .vop_fsync = nfs_fsync,
197 .vop_getattr = nfs_getattr,
198 .vop_inactive = ncl_inactive,
199 .vop_print = nfs_print,
200 .vop_read = nfsfifo_read,
201 .vop_reclaim = ncl_reclaim,
202 .vop_setattr = nfs_setattr,
203 .vop_write = nfsfifo_write,
204 };
205
206 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
207 struct componentname *cnp, struct vattr *vap);
208 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
209 int namelen, struct ucred *cred, struct thread *td);
210 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
211 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
212 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
213 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
214 struct componentname *scnp, struct sillyrename *sp);
215
216 /*
217 * Global variables
218 */
219 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
220
221 SYSCTL_DECL(_vfs_nfs);
222
223 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
224 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
225 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
226
227 static int nfs_prime_access_cache = 0;
228 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
229 &nfs_prime_access_cache, 0,
230 "Prime NFS ACCESS cache when fetching attributes");
231
232 static int newnfs_commit_on_close = 0;
233 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
234 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
235
236 static int nfs_clean_pages_on_close = 1;
237 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
238 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
239
240 int newnfs_directio_enable = 0;
241 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
242 &newnfs_directio_enable, 0, "Enable NFS directio");
243
244 /*
245 * This sysctl allows other processes to mmap a file that has been opened
246 * O_DIRECT by a process. In general, having processes mmap the file while
247 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
248 * this by default to prevent DoS attacks - to prevent a malicious user from
249 * opening up files O_DIRECT preventing other users from mmap'ing these
250 * files. "Protected" environments where stricter consistency guarantees are
251 * required can disable this knob. The process that opened the file O_DIRECT
252 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
253 * meaningful.
254 */
255 int newnfs_directio_allow_mmap = 1;
256 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
257 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
258
259 #if 0
260 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
261 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
262
263 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
264 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
265 #endif
266
267 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
268 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
269 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
270
271 /*
272 * SMP Locking Note :
273 * The list of locks after the description of the lock is the ordering
274 * of other locks acquired with the lock held.
275 * np->n_mtx : Protects the fields in the nfsnode.
276 VM Object Lock
277 VI_MTX (acquired indirectly)
278 * nmp->nm_mtx : Protects the fields in the nfsmount.
279 rep->r_mtx
280 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
281 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
282 nmp->nm_mtx
283 rep->r_mtx
284 * rep->r_mtx : Protects the fields in an nfsreq.
285 */
286
287 static int
288 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
289 struct ucred *cred, u_int32_t *retmode)
290 {
291 int error = 0, attrflag, i, lrupos;
292 u_int32_t rmode;
293 struct nfsnode *np = VTONFS(vp);
294 struct nfsvattr nfsva;
295
296 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
297 &rmode, NULL);
298 if (attrflag)
299 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
300 if (!error) {
301 lrupos = 0;
302 mtx_lock(&np->n_mtx);
303 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
304 if (np->n_accesscache[i].uid == cred->cr_uid) {
305 np->n_accesscache[i].mode = rmode;
306 np->n_accesscache[i].stamp = time_second;
307 break;
308 }
309 if (i > 0 && np->n_accesscache[i].stamp <
310 np->n_accesscache[lrupos].stamp)
311 lrupos = i;
312 }
313 if (i == NFS_ACCESSCACHESIZE) {
314 np->n_accesscache[lrupos].uid = cred->cr_uid;
315 np->n_accesscache[lrupos].mode = rmode;
316 np->n_accesscache[lrupos].stamp = time_second;
317 }
318 mtx_unlock(&np->n_mtx);
319 if (retmode != NULL)
320 *retmode = rmode;
321 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
322 } else if (NFS_ISV4(vp)) {
323 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
324 }
325 #ifdef KDTRACE_HOOKS
326 if (error != 0)
327 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
328 error);
329 #endif
330 return (error);
331 }
332
333 /*
334 * nfs access vnode op.
335 * For nfs version 2, just return ok. File accesses may fail later.
336 * For nfs version 3, use the access rpc to check accessibility. If file modes
337 * are changed on the server, accesses might still fail later.
338 */
339 static int
340 nfs_access(struct vop_access_args *ap)
341 {
342 struct vnode *vp = ap->a_vp;
343 int error = 0, i, gotahit;
344 u_int32_t mode, wmode, rmode;
345 int v34 = NFS_ISV34(vp);
346 struct nfsnode *np = VTONFS(vp);
347
348 /*
349 * Disallow write attempts on filesystems mounted read-only;
350 * unless the file is a socket, fifo, or a block or character
351 * device resident on the filesystem.
352 */
353 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
354 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
355 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
356 switch (vp->v_type) {
357 case VREG:
358 case VDIR:
359 case VLNK:
360 return (EROFS);
361 default:
362 break;
363 }
364 }
365 /*
366 * For nfs v3 or v4, check to see if we have done this recently, and if
367 * so return our cached result instead of making an ACCESS call.
368 * If not, do an access rpc, otherwise you are stuck emulating
369 * ufs_access() locally using the vattr. This may not be correct,
370 * since the server may apply other access criteria such as
371 * client uid-->server uid mapping that we do not know about.
372 */
373 if (v34) {
374 if (ap->a_accmode & VREAD)
375 mode = NFSACCESS_READ;
376 else
377 mode = 0;
378 if (vp->v_type != VDIR) {
379 if (ap->a_accmode & VWRITE)
380 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
381 if (ap->a_accmode & VAPPEND)
382 mode |= NFSACCESS_EXTEND;
383 if (ap->a_accmode & VEXEC)
384 mode |= NFSACCESS_EXECUTE;
385 if (ap->a_accmode & VDELETE)
386 mode |= NFSACCESS_DELETE;
387 } else {
388 if (ap->a_accmode & VWRITE)
389 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
390 if (ap->a_accmode & VAPPEND)
391 mode |= NFSACCESS_EXTEND;
392 if (ap->a_accmode & VEXEC)
393 mode |= NFSACCESS_LOOKUP;
394 if (ap->a_accmode & VDELETE)
395 mode |= NFSACCESS_DELETE;
396 if (ap->a_accmode & VDELETE_CHILD)
397 mode |= NFSACCESS_MODIFY;
398 }
399 /* XXX safety belt, only make blanket request if caching */
400 if (nfsaccess_cache_timeout > 0) {
401 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
402 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
403 NFSACCESS_DELETE | NFSACCESS_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 NFSINCRGLOBAL(newnfsstats.accesscache_hits);
420 gotahit = 1;
421 }
422 break;
423 }
424 }
425 mtx_unlock(&np->n_mtx);
426 #ifdef KDTRACE_HOOKS
427 if (gotahit != 0)
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 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
439 error = nfs34_access_otw(vp, wmode, ap->a_td,
440 ap->a_cred, &rmode);
441 if (!error &&
442 (rmode & mode) != mode)
443 error = EACCES;
444 }
445 return (error);
446 } else {
447 if ((error = nfsspec_access(ap)) != 0) {
448 return (error);
449 }
450 /*
451 * Attempt to prevent a mapped root from accessing a file
452 * which it shouldn't. We try to read a byte from the file
453 * if the user is root and the file is not zero length.
454 * After calling nfsspec_access, we should have the correct
455 * file size cached.
456 */
457 mtx_lock(&np->n_mtx);
458 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
459 && VTONFS(vp)->n_size > 0) {
460 struct iovec aiov;
461 struct uio auio;
462 char buf[1];
463
464 mtx_unlock(&np->n_mtx);
465 aiov.iov_base = buf;
466 aiov.iov_len = 1;
467 auio.uio_iov = &aiov;
468 auio.uio_iovcnt = 1;
469 auio.uio_offset = 0;
470 auio.uio_resid = 1;
471 auio.uio_segflg = UIO_SYSSPACE;
472 auio.uio_rw = UIO_READ;
473 auio.uio_td = ap->a_td;
474
475 if (vp->v_type == VREG)
476 error = ncl_readrpc(vp, &auio, ap->a_cred);
477 else if (vp->v_type == VDIR) {
478 char* bp;
479 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
480 aiov.iov_base = bp;
481 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
482 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
483 ap->a_td);
484 free(bp, M_TEMP);
485 } else if (vp->v_type == VLNK)
486 error = ncl_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
496 /*
497 * nfs open vnode op
498 * Check to see if the type is ok
499 * and that deletion is not in progress.
500 * For paged in text files, you will need to flush the page cache
501 * if consistency is lost.
502 */
503 /* ARGSUSED */
504 static int
505 nfs_open(struct vop_open_args *ap)
506 {
507 struct vnode *vp = ap->a_vp;
508 struct nfsnode *np = VTONFS(vp);
509 struct vattr vattr;
510 int error;
511 int fmode = ap->a_mode;
512
513 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
514 return (EOPNOTSUPP);
515
516 /*
517 * For NFSv4, we need to do the Open Op before cache validation,
518 * so that we conform to RFC3530 Sec. 9.3.1.
519 */
520 if (NFS_ISV4(vp)) {
521 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
522 if (error) {
523 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
524 (gid_t)0);
525 return (error);
526 }
527 }
528
529 /*
530 * Now, if this Open will be doing reading, re-validate/flush the
531 * cache, so that Close/Open coherency is maintained.
532 */
533 mtx_lock(&np->n_mtx);
534 if (np->n_flag & NMODIFIED) {
535 mtx_unlock(&np->n_mtx);
536 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
537 if (error == EINTR || error == EIO) {
538 if (NFS_ISV4(vp))
539 (void) nfsrpc_close(vp, 0, ap->a_td);
540 return (error);
541 }
542 mtx_lock(&np->n_mtx);
543 np->n_attrstamp = 0;
544 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
545 if (vp->v_type == VDIR)
546 np->n_direofoffset = 0;
547 mtx_unlock(&np->n_mtx);
548 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
549 if (error) {
550 if (NFS_ISV4(vp))
551 (void) nfsrpc_close(vp, 0, ap->a_td);
552 return (error);
553 }
554 mtx_lock(&np->n_mtx);
555 np->n_mtime = vattr.va_mtime;
556 if (NFS_ISV4(vp))
557 np->n_change = vattr.va_filerev;
558 } else {
559 mtx_unlock(&np->n_mtx);
560 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
561 if (error) {
562 if (NFS_ISV4(vp))
563 (void) nfsrpc_close(vp, 0, ap->a_td);
564 return (error);
565 }
566 mtx_lock(&np->n_mtx);
567 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
568 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
569 if (vp->v_type == VDIR)
570 np->n_direofoffset = 0;
571 mtx_unlock(&np->n_mtx);
572 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
573 if (error == EINTR || error == EIO) {
574 if (NFS_ISV4(vp))
575 (void) nfsrpc_close(vp, 0, ap->a_td);
576 return (error);
577 }
578 mtx_lock(&np->n_mtx);
579 np->n_mtime = vattr.va_mtime;
580 if (NFS_ISV4(vp))
581 np->n_change = vattr.va_filerev;
582 }
583 }
584
585 /*
586 * If the object has >= 1 O_DIRECT active opens, we disable caching.
587 */
588 if (newnfs_directio_enable && (fmode & O_DIRECT) &&
589 (vp->v_type == VREG)) {
590 if (np->n_directio_opens == 0) {
591 mtx_unlock(&np->n_mtx);
592 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
593 if (error) {
594 if (NFS_ISV4(vp))
595 (void) nfsrpc_close(vp, 0, ap->a_td);
596 return (error);
597 }
598 mtx_lock(&np->n_mtx);
599 np->n_flag |= NNONCACHE;
600 }
601 np->n_directio_opens++;
602 }
603 mtx_unlock(&np->n_mtx);
604 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
605 return (0);
606 }
607
608 /*
609 * nfs close vnode op
610 * What an NFS client should do upon close after writing is a debatable issue.
611 * Most NFS clients push delayed writes to the server upon close, basically for
612 * two reasons:
613 * 1 - So that any write errors may be reported back to the client process
614 * doing the close system call. By far the two most likely errors are
615 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
616 * 2 - To put a worst case upper bound on cache inconsistency between
617 * multiple clients for the file.
618 * There is also a consistency problem for Version 2 of the protocol w.r.t.
619 * not being able to tell if other clients are writing a file concurrently,
620 * since there is no way of knowing if the changed modify time in the reply
621 * is only due to the write for this client.
622 * (NFS Version 3 provides weak cache consistency data in the reply that
623 * should be sufficient to detect and handle this case.)
624 *
625 * The current code does the following:
626 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
627 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
628 * or commit them (this satisfies 1 and 2 except for the
629 * case where the server crashes after this close but
630 * before the commit RPC, which is felt to be "good
631 * enough". Changing the last argument to ncl_flush() to
632 * a 1 would force a commit operation, if it is felt a
633 * commit is necessary now.
634 * for NFS Version 4 - flush the dirty buffers and commit them, if
635 * nfscl_mustflush() says this is necessary.
636 * It is necessary if there is no write delegation held,
637 * in order to satisfy open/close coherency.
638 * If the file isn't cached on local stable storage,
639 * it may be necessary in order to detect "out of space"
640 * errors from the server, if the write delegation
641 * issued by the server doesn't allow the file to grow.
642 */
643 /* ARGSUSED */
644 static int
645 nfs_close(struct vop_close_args *ap)
646 {
647 struct vnode *vp = ap->a_vp;
648 struct nfsnode *np = VTONFS(vp);
649 struct nfsvattr nfsva;
650 struct ucred *cred;
651 int error = 0, ret, localcred = 0;
652 int fmode = ap->a_fflag;
653
654 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
655 return (0);
656 /*
657 * During shutdown, a_cred isn't valid, so just use root.
658 */
659 if (ap->a_cred == NOCRED) {
660 cred = newnfs_getcred();
661 localcred = 1;
662 } else {
663 cred = ap->a_cred;
664 }
665 if (vp->v_type == VREG) {
666 /*
667 * Examine and clean dirty pages, regardless of NMODIFIED.
668 * This closes a major hole in close-to-open consistency.
669 * We want to push out all dirty pages (and buffers) on
670 * close, regardless of whether they were dirtied by
671 * mmap'ed writes or via write().
672 */
673 if (nfs_clean_pages_on_close && vp->v_object) {
674 VM_OBJECT_LOCK(vp->v_object);
675 vm_object_page_clean(vp->v_object, 0, 0, 0);
676 VM_OBJECT_UNLOCK(vp->v_object);
677 }
678 mtx_lock(&np->n_mtx);
679 if (np->n_flag & NMODIFIED) {
680 mtx_unlock(&np->n_mtx);
681 if (NFS_ISV3(vp)) {
682 /*
683 * Under NFSv3 we have dirty buffers to dispose of. We
684 * must flush them to the NFS server. We have the option
685 * of waiting all the way through the commit rpc or just
686 * waiting for the initial write. The default is to only
687 * wait through the initial write so the data is in the
688 * server's cache, which is roughly similar to the state
689 * a standard disk subsystem leaves the file in on close().
690 *
691 * We cannot clear the NMODIFIED bit in np->n_flag due to
692 * potential races with other processes, and certainly
693 * cannot clear it if we don't commit.
694 * These races occur when there is no longer the old
695 * traditional vnode locking implemented for Vnode Ops.
696 */
697 int cm = newnfs_commit_on_close ? 1 : 0;
698 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
699 /* np->n_flag &= ~NMODIFIED; */
700 } else if (NFS_ISV4(vp)) {
701 if (nfscl_mustflush(vp) != 0) {
702 int cm = newnfs_commit_on_close ? 1 : 0;
703 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
704 cm, 0);
705 /*
706 * as above w.r.t races when clearing
707 * NMODIFIED.
708 * np->n_flag &= ~NMODIFIED;
709 */
710 }
711 } else
712 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
713 mtx_lock(&np->n_mtx);
714 }
715 /*
716 * Invalidate the attribute cache in all cases.
717 * An open is going to fetch fresh attrs any way, other procs
718 * on this node that have file open will be forced to do an
719 * otw attr fetch, but this is safe.
720 * --> A user found that their RPC count dropped by 20% when
721 * this was commented out and I can't see any requirement
722 * for it, so I've disabled it when negative lookups are
723 * enabled. (What does this have to do with negative lookup
724 * caching? Well nothing, except it was reported by the
725 * same user that needed negative lookup caching and I wanted
726 * there to be a way to disable it to see if it
727 * is the cause of some caching/coherency issue that might
728 * crop up.)
729 */
730 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
731 np->n_attrstamp = 0;
732 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
733 }
734 if (np->n_flag & NWRITEERR) {
735 np->n_flag &= ~NWRITEERR;
736 error = np->n_error;
737 }
738 mtx_unlock(&np->n_mtx);
739 }
740
741 if (NFS_ISV4(vp)) {
742 /*
743 * Get attributes so "change" is up to date.
744 */
745 if (error == 0 && nfscl_mustflush(vp) != 0) {
746 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
747 NULL);
748 if (!ret) {
749 np->n_change = nfsva.na_filerev;
750 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
751 NULL, 0, 0);
752 }
753 }
754
755 /*
756 * and do the close.
757 */
758 ret = nfsrpc_close(vp, 0, ap->a_td);
759 if (!error && ret)
760 error = ret;
761 if (error)
762 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
763 (gid_t)0);
764 }
765 if (newnfs_directio_enable)
766 KASSERT((np->n_directio_asyncwr == 0),
767 ("nfs_close: dirty unflushed (%d) directio buffers\n",
768 np->n_directio_asyncwr));
769 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
770 mtx_lock(&np->n_mtx);
771 KASSERT((np->n_directio_opens > 0),
772 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
773 np->n_directio_opens--;
774 if (np->n_directio_opens == 0)
775 np->n_flag &= ~NNONCACHE;
776 mtx_unlock(&np->n_mtx);
777 }
778 if (localcred)
779 NFSFREECRED(cred);
780 return (error);
781 }
782
783 /*
784 * nfs getattr call from vfs.
785 */
786 static int
787 nfs_getattr(struct vop_getattr_args *ap)
788 {
789 struct vnode *vp = ap->a_vp;
790 struct thread *td = curthread; /* XXX */
791 struct nfsnode *np = VTONFS(vp);
792 int error = 0;
793 struct nfsvattr nfsva;
794 struct vattr *vap = ap->a_vap;
795 struct vattr vattr;
796
797 /*
798 * Update local times for special files.
799 */
800 mtx_lock(&np->n_mtx);
801 if (np->n_flag & (NACC | NUPD))
802 np->n_flag |= NCHG;
803 mtx_unlock(&np->n_mtx);
804 /*
805 * First look in the cache.
806 */
807 if (ncl_getattrcache(vp, &vattr) == 0) {
808 vap->va_type = vattr.va_type;
809 vap->va_mode = vattr.va_mode;
810 vap->va_nlink = vattr.va_nlink;
811 vap->va_uid = vattr.va_uid;
812 vap->va_gid = vattr.va_gid;
813 vap->va_fsid = vattr.va_fsid;
814 vap->va_fileid = vattr.va_fileid;
815 vap->va_size = vattr.va_size;
816 vap->va_blocksize = vattr.va_blocksize;
817 vap->va_atime = vattr.va_atime;
818 vap->va_mtime = vattr.va_mtime;
819 vap->va_ctime = vattr.va_ctime;
820 vap->va_gen = vattr.va_gen;
821 vap->va_flags = vattr.va_flags;
822 vap->va_rdev = vattr.va_rdev;
823 vap->va_bytes = vattr.va_bytes;
824 vap->va_filerev = vattr.va_filerev;
825 /*
826 * Get the local modify time for the case of a write
827 * delegation.
828 */
829 nfscl_deleggetmodtime(vp, &vap->va_mtime);
830 return (0);
831 }
832
833 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
834 nfsaccess_cache_timeout > 0) {
835 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
836 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
837 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
838 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
839 return (0);
840 }
841 }
842 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
843 if (!error)
844 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
845 if (!error) {
846 /*
847 * Get the local modify time for the case of a write
848 * delegation.
849 */
850 nfscl_deleggetmodtime(vp, &vap->va_mtime);
851 } else if (NFS_ISV4(vp)) {
852 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
853 }
854 return (error);
855 }
856
857 /*
858 * nfs setattr call.
859 */
860 static int
861 nfs_setattr(struct vop_setattr_args *ap)
862 {
863 struct vnode *vp = ap->a_vp;
864 struct nfsnode *np = VTONFS(vp);
865 struct thread *td = curthread; /* XXX */
866 struct vattr *vap = ap->a_vap;
867 int error = 0;
868 u_quad_t tsize;
869
870 #ifndef nolint
871 tsize = (u_quad_t)0;
872 #endif
873
874 /*
875 * Setting of flags and marking of atimes are not supported.
876 */
877 if (vap->va_flags != VNOVAL)
878 return (EOPNOTSUPP);
879
880 /*
881 * Disallow write attempts if the filesystem is mounted read-only.
882 */
883 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
884 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
885 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
886 (vp->v_mount->mnt_flag & MNT_RDONLY))
887 return (EROFS);
888 if (vap->va_size != VNOVAL) {
889 switch (vp->v_type) {
890 case VDIR:
891 return (EISDIR);
892 case VCHR:
893 case VBLK:
894 case VSOCK:
895 case VFIFO:
896 if (vap->va_mtime.tv_sec == VNOVAL &&
897 vap->va_atime.tv_sec == VNOVAL &&
898 vap->va_mode == (mode_t)VNOVAL &&
899 vap->va_uid == (uid_t)VNOVAL &&
900 vap->va_gid == (gid_t)VNOVAL)
901 return (0);
902 vap->va_size = VNOVAL;
903 break;
904 default:
905 /*
906 * Disallow write attempts if the filesystem is
907 * mounted read-only.
908 */
909 if (vp->v_mount->mnt_flag & MNT_RDONLY)
910 return (EROFS);
911 /*
912 * We run vnode_pager_setsize() early (why?),
913 * we must set np->n_size now to avoid vinvalbuf
914 * V_SAVE races that might setsize a lower
915 * value.
916 */
917 mtx_lock(&np->n_mtx);
918 tsize = np->n_size;
919 mtx_unlock(&np->n_mtx);
920 error = ncl_meta_setsize(vp, ap->a_cred, td,
921 vap->va_size);
922 mtx_lock(&np->n_mtx);
923 if (np->n_flag & NMODIFIED) {
924 tsize = np->n_size;
925 mtx_unlock(&np->n_mtx);
926 if (vap->va_size == 0)
927 error = ncl_vinvalbuf(vp, 0, td, 1);
928 else
929 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
930 if (error) {
931 vnode_pager_setsize(vp, tsize);
932 return (error);
933 }
934 /*
935 * Call nfscl_delegmodtime() to set the modify time
936 * locally, as required.
937 */
938 nfscl_delegmodtime(vp);
939 } else
940 mtx_unlock(&np->n_mtx);
941 /*
942 * np->n_size has already been set to vap->va_size
943 * in ncl_meta_setsize(). We must set it again since
944 * nfs_loadattrcache() could be called through
945 * ncl_meta_setsize() and could modify np->n_size.
946 */
947 mtx_lock(&np->n_mtx);
948 np->n_vattr.na_size = np->n_size = vap->va_size;
949 mtx_unlock(&np->n_mtx);
950 };
951 } else {
952 mtx_lock(&np->n_mtx);
953 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
954 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
955 mtx_unlock(&np->n_mtx);
956 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
957 (error == EINTR || error == EIO))
958 return (error);
959 } else
960 mtx_unlock(&np->n_mtx);
961 }
962 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
963 if (error && vap->va_size != VNOVAL) {
964 mtx_lock(&np->n_mtx);
965 np->n_size = np->n_vattr.na_size = tsize;
966 vnode_pager_setsize(vp, tsize);
967 mtx_unlock(&np->n_mtx);
968 }
969 return (error);
970 }
971
972 /*
973 * Do an nfs setattr rpc.
974 */
975 static int
976 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
977 struct thread *td)
978 {
979 struct nfsnode *np = VTONFS(vp);
980 int error, ret, attrflag, i;
981 struct nfsvattr nfsva;
982
983 if (NFS_ISV34(vp)) {
984 mtx_lock(&np->n_mtx);
985 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
986 np->n_accesscache[i].stamp = 0;
987 np->n_flag |= NDELEGMOD;
988 mtx_unlock(&np->n_mtx);
989 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
990 }
991 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
992 NULL);
993 if (attrflag) {
994 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
995 if (ret && !error)
996 error = ret;
997 }
998 if (error && NFS_ISV4(vp))
999 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1000 return (error);
1001 }
1002
1003 /*
1004 * nfs lookup call, one step at a time...
1005 * First look in cache
1006 * If not found, unlock the directory nfsnode and do the rpc
1007 */
1008 static int
1009 nfs_lookup(struct vop_lookup_args *ap)
1010 {
1011 struct componentname *cnp = ap->a_cnp;
1012 struct vnode *dvp = ap->a_dvp;
1013 struct vnode **vpp = ap->a_vpp;
1014 struct mount *mp = dvp->v_mount;
1015 int flags = cnp->cn_flags;
1016 struct vnode *newvp;
1017 struct nfsmount *nmp;
1018 struct nfsnode *np, *newnp;
1019 int error = 0, attrflag, dattrflag, ltype;
1020 struct thread *td = cnp->cn_thread;
1021 struct nfsfh *nfhp;
1022 struct nfsvattr dnfsva, nfsva;
1023 struct vattr vattr;
1024 struct timespec dmtime;
1025
1026 *vpp = NULLVP;
1027 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1028 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1029 return (EROFS);
1030 if (dvp->v_type != VDIR)
1031 return (ENOTDIR);
1032 nmp = VFSTONFS(mp);
1033 np = VTONFS(dvp);
1034
1035 /* For NFSv4, wait until any remove is done. */
1036 mtx_lock(&np->n_mtx);
1037 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1038 np->n_flag |= NREMOVEWANT;
1039 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1040 }
1041 mtx_unlock(&np->n_mtx);
1042
1043 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1044 return (error);
1045 error = cache_lookup(dvp, vpp, cnp);
1046 if (error > 0 && error != ENOENT)
1047 return (error);
1048 if (error == -1) {
1049 /*
1050 * We only accept a positive hit in the cache if the
1051 * change time of the file matches our cached copy.
1052 * Otherwise, we discard the cache entry and fallback
1053 * to doing a lookup RPC.
1054 *
1055 * To better handle stale file handles and attributes,
1056 * clear the attribute cache of this node if it is a
1057 * leaf component, part of an open() call, and not
1058 * locally modified before fetching the attributes.
1059 * This should allow stale file handles to be detected
1060 * here where we can fall back to a LOOKUP RPC to
1061 * recover rather than having nfs_open() detect the
1062 * stale file handle and failing open(2) with ESTALE.
1063 */
1064 newvp = *vpp;
1065 newnp = VTONFS(newvp);
1066 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1067 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1068 !(newnp->n_flag & NMODIFIED)) {
1069 mtx_lock(&newnp->n_mtx);
1070 newnp->n_attrstamp = 0;
1071 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1072 mtx_unlock(&newnp->n_mtx);
1073 }
1074 if (nfscl_nodeleg(newvp, 0) == 0 ||
1075 (VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1076 timespeccmp(&vattr.va_ctime, &newnp->n_ctime, ==))) {
1077 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1078 if (cnp->cn_nameiop != LOOKUP &&
1079 (flags & ISLASTCN))
1080 cnp->cn_flags |= SAVENAME;
1081 return (0);
1082 }
1083 cache_purge(newvp);
1084 if (dvp != newvp)
1085 vput(newvp);
1086 else
1087 vrele(newvp);
1088 *vpp = NULLVP;
1089 } else if (error == ENOENT) {
1090 if (dvp->v_iflag & VI_DOOMED)
1091 return (ENOENT);
1092 /*
1093 * We only accept a negative hit in the cache if the
1094 * modification time of the parent directory matches
1095 * our cached copy. Otherwise, we discard all of the
1096 * negative cache entries for this directory. We also
1097 * only trust -ve cache entries for less than
1098 * nm_negative_namecache_timeout seconds.
1099 */
1100 if ((u_int)(ticks - np->n_dmtime_ticks) <
1101 (nmp->nm_negnametimeo * hz) &&
1102 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1103 timespeccmp(&vattr.va_mtime, &np->n_dmtime, ==)) {
1104 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1105 return (ENOENT);
1106 }
1107 cache_purge_negative(dvp);
1108 mtx_lock(&np->n_mtx);
1109 timespecclear(&np->n_dmtime);
1110 mtx_unlock(&np->n_mtx);
1111 }
1112
1113 /*
1114 * Cache the modification time of the parent directory in case
1115 * the lookup fails and results in adding the first negative
1116 * name cache entry for the directory. Since this is reading
1117 * a single time_t, don't bother with locking. The
1118 * modification time may be a bit stale, but it must be read
1119 * before performing the lookup RPC to prevent a race where
1120 * another lookup updates the timestamp on the directory after
1121 * the lookup RPC has been performed on the server but before
1122 * n_dmtime is set at the end of this function.
1123 */
1124 dmtime = np->n_vattr.na_mtime;
1125 error = 0;
1126 newvp = NULLVP;
1127 NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1128 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1129 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1130 NULL);
1131 if (dattrflag)
1132 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1133 if (error) {
1134 if (newvp != NULLVP) {
1135 vput(newvp);
1136 *vpp = NULLVP;
1137 }
1138
1139 if (error != ENOENT) {
1140 if (NFS_ISV4(dvp))
1141 error = nfscl_maperr(td, error, (uid_t)0,
1142 (gid_t)0);
1143 return (error);
1144 }
1145
1146 /* The requested file was not found. */
1147 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1148 (flags & ISLASTCN)) {
1149 /*
1150 * XXX: UFS does a full VOP_ACCESS(dvp,
1151 * VWRITE) here instead of just checking
1152 * MNT_RDONLY.
1153 */
1154 if (mp->mnt_flag & MNT_RDONLY)
1155 return (EROFS);
1156 cnp->cn_flags |= SAVENAME;
1157 return (EJUSTRETURN);
1158 }
1159
1160 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1161 /*
1162 * Maintain n_dmtime as the modification time
1163 * of the parent directory when the oldest -ve
1164 * name cache entry for this directory was
1165 * added. If a -ve cache entry has already
1166 * been added with a newer modification time
1167 * by a concurrent lookup, then don't bother
1168 * adding a cache entry. The modification
1169 * time of the directory might have changed
1170 * due to the file this lookup failed to find
1171 * being created. In that case a subsequent
1172 * lookup would incorrectly use the entry
1173 * added here instead of doing an extra
1174 * lookup.
1175 */
1176 mtx_lock(&np->n_mtx);
1177 if (timespeccmp(&np->n_dmtime, &dmtime, <=)) {
1178 if (!timespecisset(&np->n_dmtime)) {
1179 np->n_dmtime = dmtime;
1180 np->n_dmtime_ticks = ticks;
1181 }
1182 mtx_unlock(&np->n_mtx);
1183 cache_enter(dvp, NULL, cnp);
1184 } else
1185 mtx_unlock(&np->n_mtx);
1186 }
1187 return (ENOENT);
1188 }
1189
1190 /*
1191 * Handle RENAME case...
1192 */
1193 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1194 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1195 FREE((caddr_t)nfhp, M_NFSFH);
1196 return (EISDIR);
1197 }
1198 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1199 LK_EXCLUSIVE);
1200 if (error)
1201 return (error);
1202 newvp = NFSTOV(np);
1203 if (attrflag)
1204 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1205 0, 1);
1206 *vpp = newvp;
1207 cnp->cn_flags |= SAVENAME;
1208 return (0);
1209 }
1210
1211 if (flags & ISDOTDOT) {
1212 ltype = NFSVOPISLOCKED(dvp);
1213 error = vfs_busy(mp, MBF_NOWAIT);
1214 if (error != 0) {
1215 vfs_ref(mp);
1216 NFSVOPUNLOCK(dvp, 0);
1217 error = vfs_busy(mp, 0);
1218 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1219 vfs_rel(mp);
1220 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1221 vfs_unbusy(mp);
1222 error = ENOENT;
1223 }
1224 if (error != 0)
1225 return (error);
1226 }
1227 NFSVOPUNLOCK(dvp, 0);
1228 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1229 cnp->cn_lkflags);
1230 if (error == 0)
1231 newvp = NFSTOV(np);
1232 vfs_unbusy(mp);
1233 if (newvp != dvp)
1234 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1235 if (dvp->v_iflag & VI_DOOMED) {
1236 if (error == 0) {
1237 if (newvp == dvp)
1238 vrele(newvp);
1239 else
1240 vput(newvp);
1241 }
1242 error = ENOENT;
1243 }
1244 if (error != 0)
1245 return (error);
1246 if (attrflag)
1247 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1248 0, 1);
1249 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1250 FREE((caddr_t)nfhp, M_NFSFH);
1251 VREF(dvp);
1252 newvp = dvp;
1253 if (attrflag)
1254 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1255 0, 1);
1256 } else {
1257 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1258 cnp->cn_lkflags);
1259 if (error)
1260 return (error);
1261 newvp = NFSTOV(np);
1262 if (attrflag)
1263 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1264 0, 1);
1265 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1266 !(np->n_flag & NMODIFIED)) {
1267 /*
1268 * Flush the attribute cache when opening a
1269 * leaf node to ensure that fresh attributes
1270 * are fetched in nfs_open() since we did not
1271 * fetch attributes from the LOOKUP reply.
1272 */
1273 mtx_lock(&np->n_mtx);
1274 np->n_attrstamp = 0;
1275 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1276 mtx_unlock(&np->n_mtx);
1277 }
1278 }
1279 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1280 cnp->cn_flags |= SAVENAME;
1281 if ((cnp->cn_flags & MAKEENTRY) &&
1282 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1283 np->n_ctime = np->n_vattr.na_vattr.va_ctime;
1284 cache_enter(dvp, newvp, cnp);
1285 }
1286 *vpp = newvp;
1287 return (0);
1288 }
1289
1290 /*
1291 * nfs read call.
1292 * Just call ncl_bioread() to do the work.
1293 */
1294 static int
1295 nfs_read(struct vop_read_args *ap)
1296 {
1297 struct vnode *vp = ap->a_vp;
1298
1299 switch (vp->v_type) {
1300 case VREG:
1301 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1302 case VDIR:
1303 return (EISDIR);
1304 default:
1305 return (EOPNOTSUPP);
1306 }
1307 }
1308
1309 /*
1310 * nfs readlink call
1311 */
1312 static int
1313 nfs_readlink(struct vop_readlink_args *ap)
1314 {
1315 struct vnode *vp = ap->a_vp;
1316
1317 if (vp->v_type != VLNK)
1318 return (EINVAL);
1319 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1320 }
1321
1322 /*
1323 * Do a readlink rpc.
1324 * Called by ncl_doio() from below the buffer cache.
1325 */
1326 int
1327 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1328 {
1329 int error, ret, attrflag;
1330 struct nfsvattr nfsva;
1331
1332 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1333 &attrflag, NULL);
1334 if (attrflag) {
1335 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1336 if (ret && !error)
1337 error = ret;
1338 }
1339 if (error && NFS_ISV4(vp))
1340 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1341 return (error);
1342 }
1343
1344 /*
1345 * nfs read rpc call
1346 * Ditto above
1347 */
1348 int
1349 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1350 {
1351 int error, ret, attrflag;
1352 struct nfsvattr nfsva;
1353
1354 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1355 NULL);
1356 if (attrflag) {
1357 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1358 if (ret && !error)
1359 error = ret;
1360 }
1361 if (error && NFS_ISV4(vp))
1362 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1363 return (error);
1364 }
1365
1366 /*
1367 * nfs write call
1368 */
1369 int
1370 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1371 int *iomode, int *must_commit, int called_from_strategy)
1372 {
1373 struct nfsvattr nfsva;
1374 int error = 0, attrflag, ret;
1375
1376 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1377 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy);
1378 if (attrflag) {
1379 if (VTONFS(vp)->n_flag & ND_NFSV4)
1380 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1381 1);
1382 else
1383 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1384 1);
1385 if (ret && !error)
1386 error = ret;
1387 }
1388 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1389 *iomode = NFSWRITE_FILESYNC;
1390 if (error && NFS_ISV4(vp))
1391 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1392 return (error);
1393 }
1394
1395 /*
1396 * nfs mknod rpc
1397 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1398 * mode set to specify the file type and the size field for rdev.
1399 */
1400 static int
1401 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1402 struct vattr *vap)
1403 {
1404 struct nfsvattr nfsva, dnfsva;
1405 struct vnode *newvp = NULL;
1406 struct nfsnode *np = NULL, *dnp;
1407 struct nfsfh *nfhp;
1408 struct vattr vattr;
1409 int error = 0, attrflag, dattrflag;
1410 u_int32_t rdev;
1411
1412 if (vap->va_type == VCHR || vap->va_type == VBLK)
1413 rdev = vap->va_rdev;
1414 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1415 rdev = 0xffffffff;
1416 else
1417 return (EOPNOTSUPP);
1418 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1419 return (error);
1420 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1421 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1422 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1423 if (!error) {
1424 if (!nfhp)
1425 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1426 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1427 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1428 NULL);
1429 if (nfhp)
1430 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1431 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1432 }
1433 if (dattrflag)
1434 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1435 if (!error) {
1436 newvp = NFSTOV(np);
1437 if (attrflag != 0) {
1438 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1439 0, 1);
1440 if (error != 0)
1441 vput(newvp);
1442 }
1443 }
1444 if (!error) {
1445 if ((cnp->cn_flags & MAKEENTRY))
1446 cache_enter(dvp, newvp, cnp);
1447 *vpp = newvp;
1448 } else if (NFS_ISV4(dvp)) {
1449 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1450 vap->va_gid);
1451 }
1452 dnp = VTONFS(dvp);
1453 mtx_lock(&dnp->n_mtx);
1454 dnp->n_flag |= NMODIFIED;
1455 if (!dattrflag) {
1456 dnp->n_attrstamp = 0;
1457 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1458 }
1459 mtx_unlock(&dnp->n_mtx);
1460 return (error);
1461 }
1462
1463 /*
1464 * nfs mknod vop
1465 * just call nfs_mknodrpc() to do the work.
1466 */
1467 /* ARGSUSED */
1468 static int
1469 nfs_mknod(struct vop_mknod_args *ap)
1470 {
1471 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1472 }
1473
1474 static struct mtx nfs_cverf_mtx;
1475 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1476 MTX_DEF);
1477
1478 static nfsquad_t
1479 nfs_get_cverf(void)
1480 {
1481 static nfsquad_t cverf;
1482 nfsquad_t ret;
1483 static int cverf_initialized = 0;
1484
1485 mtx_lock(&nfs_cverf_mtx);
1486 if (cverf_initialized == 0) {
1487 cverf.lval[0] = arc4random();
1488 cverf.lval[1] = arc4random();
1489 cverf_initialized = 1;
1490 } else
1491 cverf.qval++;
1492 ret = cverf;
1493 mtx_unlock(&nfs_cverf_mtx);
1494
1495 return (ret);
1496 }
1497
1498 /*
1499 * nfs file create call
1500 */
1501 static int
1502 nfs_create(struct vop_create_args *ap)
1503 {
1504 struct vnode *dvp = ap->a_dvp;
1505 struct vattr *vap = ap->a_vap;
1506 struct componentname *cnp = ap->a_cnp;
1507 struct nfsnode *np = NULL, *dnp;
1508 struct vnode *newvp = NULL;
1509 struct nfsmount *nmp;
1510 struct nfsvattr dnfsva, nfsva;
1511 struct nfsfh *nfhp;
1512 nfsquad_t cverf;
1513 int error = 0, attrflag, dattrflag, fmode = 0;
1514 struct vattr vattr;
1515
1516 /*
1517 * Oops, not for me..
1518 */
1519 if (vap->va_type == VSOCK)
1520 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1521
1522 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1523 return (error);
1524 if (vap->va_vaflags & VA_EXCLUSIVE)
1525 fmode |= O_EXCL;
1526 dnp = VTONFS(dvp);
1527 nmp = VFSTONFS(vnode_mount(dvp));
1528 again:
1529 /* For NFSv4, wait until any remove is done. */
1530 mtx_lock(&dnp->n_mtx);
1531 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1532 dnp->n_flag |= NREMOVEWANT;
1533 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1534 }
1535 mtx_unlock(&dnp->n_mtx);
1536
1537 cverf = nfs_get_cverf();
1538 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1539 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1540 &nfhp, &attrflag, &dattrflag, NULL);
1541 if (!error) {
1542 if (nfhp == NULL)
1543 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1544 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1545 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1546 NULL);
1547 if (nfhp != NULL)
1548 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1549 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1550 }
1551 if (dattrflag)
1552 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1553 if (!error) {
1554 newvp = NFSTOV(np);
1555 if (attrflag)
1556 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1557 0, 1);
1558 }
1559 if (error) {
1560 if (newvp != NULL) {
1561 vput(newvp);
1562 newvp = NULL;
1563 }
1564 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1565 error == NFSERR_NOTSUPP) {
1566 fmode &= ~O_EXCL;
1567 goto again;
1568 }
1569 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1570 if (nfscl_checksattr(vap, &nfsva)) {
1571 /*
1572 * We are normally called with only a partially
1573 * initialized VAP. Since the NFSv3 spec says that
1574 * the server may use the file attributes to
1575 * store the verifier, the spec requires us to do a
1576 * SETATTR RPC. FreeBSD servers store the verifier in
1577 * atime, but we can't really assume that all servers
1578 * will so we ensure that our SETATTR sets both atime
1579 * and mtime.
1580 */
1581 if (vap->va_mtime.tv_sec == VNOVAL)
1582 vfs_timestamp(&vap->va_mtime);
1583 if (vap->va_atime.tv_sec == VNOVAL)
1584 vap->va_atime = vap->va_mtime;
1585 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1586 cnp->cn_thread, &nfsva, &attrflag, NULL);
1587 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1588 vap->va_gid != (gid_t)VNOVAL)) {
1589 /* try again without setting uid/gid */
1590 vap->va_uid = (uid_t)VNOVAL;
1591 vap->va_gid = (uid_t)VNOVAL;
1592 error = nfsrpc_setattr(newvp, vap, NULL,
1593 cnp->cn_cred, cnp->cn_thread, &nfsva,
1594 &attrflag, NULL);
1595 }
1596 if (attrflag)
1597 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1598 NULL, 0, 1);
1599 if (error != 0)
1600 vput(newvp);
1601 }
1602 }
1603 if (!error) {
1604 if (cnp->cn_flags & MAKEENTRY)
1605 cache_enter(dvp, newvp, cnp);
1606 *ap->a_vpp = newvp;
1607 } else if (NFS_ISV4(dvp)) {
1608 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1609 vap->va_gid);
1610 }
1611 mtx_lock(&dnp->n_mtx);
1612 dnp->n_flag |= NMODIFIED;
1613 if (!dattrflag) {
1614 dnp->n_attrstamp = 0;
1615 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1616 }
1617 mtx_unlock(&dnp->n_mtx);
1618 return (error);
1619 }
1620
1621 /*
1622 * nfs file remove call
1623 * To try and make nfs semantics closer to ufs semantics, a file that has
1624 * other processes using the vnode is renamed instead of removed and then
1625 * removed later on the last close.
1626 * - If v_usecount > 1
1627 * If a rename is not already in the works
1628 * call nfs_sillyrename() to set it up
1629 * else
1630 * do the remove rpc
1631 */
1632 static int
1633 nfs_remove(struct vop_remove_args *ap)
1634 {
1635 struct vnode *vp = ap->a_vp;
1636 struct vnode *dvp = ap->a_dvp;
1637 struct componentname *cnp = ap->a_cnp;
1638 struct nfsnode *np = VTONFS(vp);
1639 int error = 0;
1640 struct vattr vattr;
1641
1642 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1643 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1644 if (vp->v_type == VDIR)
1645 error = EPERM;
1646 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1647 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1648 vattr.va_nlink > 1)) {
1649 /*
1650 * Purge the name cache so that the chance of a lookup for
1651 * the name succeeding while the remove is in progress is
1652 * minimized. Without node locking it can still happen, such
1653 * that an I/O op returns ESTALE, but since you get this if
1654 * another host removes the file..
1655 */
1656 cache_purge(vp);
1657 /*
1658 * throw away biocache buffers, mainly to avoid
1659 * unnecessary delayed writes later.
1660 */
1661 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1662 /* Do the rpc */
1663 if (error != EINTR && error != EIO)
1664 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1665 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1666 /*
1667 * Kludge City: If the first reply to the remove rpc is lost..
1668 * the reply to the retransmitted request will be ENOENT
1669 * since the file was in fact removed
1670 * Therefore, we cheat and return success.
1671 */
1672 if (error == ENOENT)
1673 error = 0;
1674 } else if (!np->n_sillyrename)
1675 error = nfs_sillyrename(dvp, vp, cnp);
1676 mtx_lock(&np->n_mtx);
1677 np->n_attrstamp = 0;
1678 mtx_unlock(&np->n_mtx);
1679 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1680 return (error);
1681 }
1682
1683 /*
1684 * nfs file remove rpc called from nfs_inactive
1685 */
1686 int
1687 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1688 {
1689 /*
1690 * Make sure that the directory vnode is still valid.
1691 * XXX we should lock sp->s_dvp here.
1692 */
1693 if (sp->s_dvp->v_type == VBAD)
1694 return (0);
1695 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1696 sp->s_cred, NULL));
1697 }
1698
1699 /*
1700 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1701 */
1702 static int
1703 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1704 int namelen, struct ucred *cred, struct thread *td)
1705 {
1706 struct nfsvattr dnfsva;
1707 struct nfsnode *dnp = VTONFS(dvp);
1708 int error = 0, dattrflag;
1709
1710 mtx_lock(&dnp->n_mtx);
1711 dnp->n_flag |= NREMOVEINPROG;
1712 mtx_unlock(&dnp->n_mtx);
1713 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1714 &dattrflag, NULL);
1715 mtx_lock(&dnp->n_mtx);
1716 if ((dnp->n_flag & NREMOVEWANT)) {
1717 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1718 mtx_unlock(&dnp->n_mtx);
1719 wakeup((caddr_t)dnp);
1720 } else {
1721 dnp->n_flag &= ~NREMOVEINPROG;
1722 mtx_unlock(&dnp->n_mtx);
1723 }
1724 if (dattrflag)
1725 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1726 mtx_lock(&dnp->n_mtx);
1727 dnp->n_flag |= NMODIFIED;
1728 if (!dattrflag) {
1729 dnp->n_attrstamp = 0;
1730 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1731 }
1732 mtx_unlock(&dnp->n_mtx);
1733 if (error && NFS_ISV4(dvp))
1734 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1735 return (error);
1736 }
1737
1738 /*
1739 * nfs file rename call
1740 */
1741 static int
1742 nfs_rename(struct vop_rename_args *ap)
1743 {
1744 struct vnode *fvp = ap->a_fvp;
1745 struct vnode *tvp = ap->a_tvp;
1746 struct vnode *fdvp = ap->a_fdvp;
1747 struct vnode *tdvp = ap->a_tdvp;
1748 struct componentname *tcnp = ap->a_tcnp;
1749 struct componentname *fcnp = ap->a_fcnp;
1750 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1751 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1752 struct nfsv4node *newv4 = NULL;
1753 int error;
1754
1755 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1756 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1757 /* Check for cross-device rename */
1758 if ((fvp->v_mount != tdvp->v_mount) ||
1759 (tvp && (fvp->v_mount != tvp->v_mount))) {
1760 error = EXDEV;
1761 goto out;
1762 }
1763
1764 if (fvp == tvp) {
1765 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1766 error = 0;
1767 goto out;
1768 }
1769 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1770 goto out;
1771
1772 /*
1773 * We have to flush B_DELWRI data prior to renaming
1774 * the file. If we don't, the delayed-write buffers
1775 * can be flushed out later after the file has gone stale
1776 * under NFSV3. NFSV2 does not have this problem because
1777 * ( as far as I can tell ) it flushes dirty buffers more
1778 * often.
1779 *
1780 * Skip the rename operation if the fsync fails, this can happen
1781 * due to the server's volume being full, when we pushed out data
1782 * that was written back to our cache earlier. Not checking for
1783 * this condition can result in potential (silent) data loss.
1784 */
1785 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1786 NFSVOPUNLOCK(fvp, 0);
1787 if (!error && tvp)
1788 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1789 if (error)
1790 goto out;
1791
1792 /*
1793 * If the tvp exists and is in use, sillyrename it before doing the
1794 * rename of the new file over it.
1795 * XXX Can't sillyrename a directory.
1796 */
1797 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1798 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1799 vput(tvp);
1800 tvp = NULL;
1801 }
1802
1803 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1804 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1805 tcnp->cn_thread);
1806
1807 if (error == 0 && NFS_ISV4(tdvp)) {
1808 /*
1809 * For NFSv4, check to see if it is the same name and
1810 * replace the name, if it is different.
1811 */
1812 MALLOC(newv4, struct nfsv4node *,
1813 sizeof (struct nfsv4node) +
1814 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1815 M_NFSV4NODE, M_WAITOK);
1816 mtx_lock(&tdnp->n_mtx);
1817 mtx_lock(&fnp->n_mtx);
1818 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1819 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1820 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1821 tcnp->cn_namelen) ||
1822 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1823 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1824 tdnp->n_fhp->nfh_len))) {
1825 #ifdef notdef
1826 { char nnn[100]; int nnnl;
1827 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1828 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1829 nnn[nnnl] = '\0';
1830 printf("ren replace=%s\n",nnn);
1831 }
1832 #endif
1833 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1834 fnp->n_v4 = newv4;
1835 newv4 = NULL;
1836 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1837 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1838 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1839 tdnp->n_fhp->nfh_len);
1840 NFSBCOPY(tcnp->cn_nameptr,
1841 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1842 }
1843 mtx_unlock(&tdnp->n_mtx);
1844 mtx_unlock(&fnp->n_mtx);
1845 if (newv4 != NULL)
1846 FREE((caddr_t)newv4, M_NFSV4NODE);
1847 }
1848
1849 if (fvp->v_type == VDIR) {
1850 if (tvp != NULL && tvp->v_type == VDIR)
1851 cache_purge(tdvp);
1852 cache_purge(fdvp);
1853 }
1854
1855 out:
1856 if (tdvp == tvp)
1857 vrele(tdvp);
1858 else
1859 vput(tdvp);
1860 if (tvp)
1861 vput(tvp);
1862 vrele(fdvp);
1863 vrele(fvp);
1864 /*
1865 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1866 */
1867 if (error == ENOENT)
1868 error = 0;
1869 return (error);
1870 }
1871
1872 /*
1873 * nfs file rename rpc called from nfs_remove() above
1874 */
1875 static int
1876 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1877 struct sillyrename *sp)
1878 {
1879
1880 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1881 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1882 scnp->cn_thread));
1883 }
1884
1885 /*
1886 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1887 */
1888 static int
1889 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1890 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1891 int tnamelen, struct ucred *cred, struct thread *td)
1892 {
1893 struct nfsvattr fnfsva, tnfsva;
1894 struct nfsnode *fdnp = VTONFS(fdvp);
1895 struct nfsnode *tdnp = VTONFS(tdvp);
1896 int error = 0, fattrflag, tattrflag;
1897
1898 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1899 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1900 &tattrflag, NULL, NULL);
1901 mtx_lock(&fdnp->n_mtx);
1902 fdnp->n_flag |= NMODIFIED;
1903 if (fattrflag != 0) {
1904 mtx_unlock(&fdnp->n_mtx);
1905 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1906 } else {
1907 fdnp->n_attrstamp = 0;
1908 mtx_unlock(&fdnp->n_mtx);
1909 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1910 }
1911 mtx_lock(&tdnp->n_mtx);
1912 tdnp->n_flag |= NMODIFIED;
1913 if (tattrflag != 0) {
1914 mtx_unlock(&tdnp->n_mtx);
1915 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1916 } else {
1917 tdnp->n_attrstamp = 0;
1918 mtx_unlock(&tdnp->n_mtx);
1919 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1920 }
1921 if (error && NFS_ISV4(fdvp))
1922 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1923 return (error);
1924 }
1925
1926 /*
1927 * nfs hard link create call
1928 */
1929 static int
1930 nfs_link(struct vop_link_args *ap)
1931 {
1932 struct vnode *vp = ap->a_vp;
1933 struct vnode *tdvp = ap->a_tdvp;
1934 struct componentname *cnp = ap->a_cnp;
1935 struct nfsnode *np, *tdnp;
1936 struct nfsvattr nfsva, dnfsva;
1937 int error = 0, attrflag, dattrflag;
1938
1939 if (vp->v_mount != tdvp->v_mount) {
1940 return (EXDEV);
1941 }
1942
1943 /*
1944 * Push all writes to the server, so that the attribute cache
1945 * doesn't get "out of sync" with the server.
1946 * XXX There should be a better way!
1947 */
1948 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1949
1950 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1951 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1952 &dattrflag, NULL);
1953 tdnp = VTONFS(tdvp);
1954 mtx_lock(&tdnp->n_mtx);
1955 tdnp->n_flag |= NMODIFIED;
1956 if (dattrflag != 0) {
1957 mtx_unlock(&tdnp->n_mtx);
1958 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1959 } else {
1960 tdnp->n_attrstamp = 0;
1961 mtx_unlock(&tdnp->n_mtx);
1962 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1963 }
1964 if (attrflag)
1965 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1966 else {
1967 np = VTONFS(vp);
1968 mtx_lock(&np->n_mtx);
1969 np->n_attrstamp = 0;
1970 mtx_unlock(&np->n_mtx);
1971 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1972 }
1973 /*
1974 * If negative lookup caching is enabled, I might as well
1975 * add an entry for this node. Not necessary for correctness,
1976 * but if negative caching is enabled, then the system
1977 * must care about lookup caching hit rate, so...
1978 */
1979 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1980 (cnp->cn_flags & MAKEENTRY))
1981 cache_enter(tdvp, vp, cnp);
1982 if (error && NFS_ISV4(vp))
1983 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1984 (gid_t)0);
1985 return (error);
1986 }
1987
1988 /*
1989 * nfs symbolic link create call
1990 */
1991 static int
1992 nfs_symlink(struct vop_symlink_args *ap)
1993 {
1994 struct vnode *dvp = ap->a_dvp;
1995 struct vattr *vap = ap->a_vap;
1996 struct componentname *cnp = ap->a_cnp;
1997 struct nfsvattr nfsva, dnfsva;
1998 struct nfsfh *nfhp;
1999 struct nfsnode *np = NULL, *dnp;
2000 struct vnode *newvp = NULL;
2001 int error = 0, attrflag, dattrflag, ret;
2002
2003 vap->va_type = VLNK;
2004 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2005 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2006 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2007 if (nfhp) {
2008 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2009 &np, NULL, LK_EXCLUSIVE);
2010 if (!ret)
2011 newvp = NFSTOV(np);
2012 else if (!error)
2013 error = ret;
2014 }
2015 if (newvp != NULL) {
2016 if (attrflag)
2017 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2018 0, 1);
2019 } else if (!error) {
2020 /*
2021 * If we do not have an error and we could not extract the
2022 * newvp from the response due to the request being NFSv2, we
2023 * have to do a lookup in order to obtain a newvp to return.
2024 */
2025 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2026 cnp->cn_cred, cnp->cn_thread, &np);
2027 if (!error)
2028 newvp = NFSTOV(np);
2029 }
2030 if (error) {
2031 if (newvp)
2032 vput(newvp);
2033 if (NFS_ISV4(dvp))
2034 error = nfscl_maperr(cnp->cn_thread, error,
2035 vap->va_uid, vap->va_gid);
2036 } else {
2037 /*
2038 * If negative lookup caching is enabled, I might as well
2039 * add an entry for this node. Not necessary for correctness,
2040 * but if negative caching is enabled, then the system
2041 * must care about lookup caching hit rate, so...
2042 */
2043 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2044 (cnp->cn_flags & MAKEENTRY))
2045 cache_enter(dvp, newvp, cnp);
2046 *ap->a_vpp = newvp;
2047 }
2048
2049 dnp = VTONFS(dvp);
2050 mtx_lock(&dnp->n_mtx);
2051 dnp->n_flag |= NMODIFIED;
2052 if (dattrflag != 0) {
2053 mtx_unlock(&dnp->n_mtx);
2054 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2055 } else {
2056 dnp->n_attrstamp = 0;
2057 mtx_unlock(&dnp->n_mtx);
2058 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2059 }
2060 return (error);
2061 }
2062
2063 /*
2064 * nfs make dir call
2065 */
2066 static int
2067 nfs_mkdir(struct vop_mkdir_args *ap)
2068 {
2069 struct vnode *dvp = ap->a_dvp;
2070 struct vattr *vap = ap->a_vap;
2071 struct componentname *cnp = ap->a_cnp;
2072 struct nfsnode *np = NULL, *dnp;
2073 struct vnode *newvp = NULL;
2074 struct vattr vattr;
2075 struct nfsfh *nfhp;
2076 struct nfsvattr nfsva, dnfsva;
2077 int error = 0, attrflag, dattrflag, ret;
2078
2079 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2080 return (error);
2081 vap->va_type = VDIR;
2082 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2083 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2084 &attrflag, &dattrflag, NULL);
2085 dnp = VTONFS(dvp);
2086 mtx_lock(&dnp->n_mtx);
2087 dnp->n_flag |= NMODIFIED;
2088 if (dattrflag != 0) {
2089 mtx_unlock(&dnp->n_mtx);
2090 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2091 } else {
2092 dnp->n_attrstamp = 0;
2093 mtx_unlock(&dnp->n_mtx);
2094 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2095 }
2096 if (nfhp) {
2097 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2098 &np, NULL, LK_EXCLUSIVE);
2099 if (!ret) {
2100 newvp = NFSTOV(np);
2101 if (attrflag)
2102 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2103 NULL, 0, 1);
2104 } else if (!error)
2105 error = ret;
2106 }
2107 if (!error && newvp == NULL) {
2108 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2109 cnp->cn_cred, cnp->cn_thread, &np);
2110 if (!error) {
2111 newvp = NFSTOV(np);
2112 if (newvp->v_type != VDIR)
2113 error = EEXIST;
2114 }
2115 }
2116 if (error) {
2117 if (newvp)
2118 vput(newvp);
2119 if (NFS_ISV4(dvp))
2120 error = nfscl_maperr(cnp->cn_thread, error,
2121 vap->va_uid, vap->va_gid);
2122 } else {
2123 /*
2124 * If negative lookup caching is enabled, I might as well
2125 * add an entry for this node. Not necessary for correctness,
2126 * but if negative caching is enabled, then the system
2127 * must care about lookup caching hit rate, so...
2128 */
2129 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2130 (cnp->cn_flags & MAKEENTRY))
2131 cache_enter(dvp, newvp, cnp);
2132 *ap->a_vpp = newvp;
2133 }
2134 return (error);
2135 }
2136
2137 /*
2138 * nfs remove directory call
2139 */
2140 static int
2141 nfs_rmdir(struct vop_rmdir_args *ap)
2142 {
2143 struct vnode *vp = ap->a_vp;
2144 struct vnode *dvp = ap->a_dvp;
2145 struct componentname *cnp = ap->a_cnp;
2146 struct nfsnode *dnp;
2147 struct nfsvattr dnfsva;
2148 int error, dattrflag;
2149
2150 if (dvp == vp)
2151 return (EINVAL);
2152 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2153 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2154 dnp = VTONFS(dvp);
2155 mtx_lock(&dnp->n_mtx);
2156 dnp->n_flag |= NMODIFIED;
2157 if (dattrflag != 0) {
2158 mtx_unlock(&dnp->n_mtx);
2159 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2160 } else {
2161 dnp->n_attrstamp = 0;
2162 mtx_unlock(&dnp->n_mtx);
2163 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2164 }
2165
2166 cache_purge(dvp);
2167 cache_purge(vp);
2168 if (error && NFS_ISV4(dvp))
2169 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2170 (gid_t)0);
2171 /*
2172 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2173 */
2174 if (error == ENOENT)
2175 error = 0;
2176 return (error);
2177 }
2178
2179 /*
2180 * nfs readdir call
2181 */
2182 static int
2183 nfs_readdir(struct vop_readdir_args *ap)
2184 {
2185 struct vnode *vp = ap->a_vp;
2186 struct nfsnode *np = VTONFS(vp);
2187 struct uio *uio = ap->a_uio;
2188 int tresid, error = 0;
2189 struct vattr vattr;
2190
2191 if (vp->v_type != VDIR)
2192 return(EPERM);
2193
2194 /*
2195 * First, check for hit on the EOF offset cache
2196 */
2197 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2198 (np->n_flag & NMODIFIED) == 0) {
2199 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2200 mtx_lock(&np->n_mtx);
2201 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2202 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2203 mtx_unlock(&np->n_mtx);
2204 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2205 return (0);
2206 } else
2207 mtx_unlock(&np->n_mtx);
2208 }
2209 }
2210
2211 /*
2212 * Call ncl_bioread() to do the real work.
2213 */
2214 tresid = uio->uio_resid;
2215 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2216
2217 if (!error && uio->uio_resid == tresid)
2218 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2219 return (error);
2220 }
2221
2222 /*
2223 * Readdir rpc call.
2224 * Called from below the buffer cache by ncl_doio().
2225 */
2226 int
2227 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2228 struct thread *td)
2229 {
2230 struct nfsvattr nfsva;
2231 nfsuint64 *cookiep, cookie;
2232 struct nfsnode *dnp = VTONFS(vp);
2233 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2234 int error = 0, eof, attrflag;
2235
2236 KASSERT(uiop->uio_iovcnt == 1 &&
2237 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2238 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2239 ("nfs readdirrpc bad uio"));
2240
2241 /*
2242 * If there is no cookie, assume directory was stale.
2243 */
2244 ncl_dircookie_lock(dnp);
2245 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2246 if (cookiep) {
2247 cookie = *cookiep;
2248 ncl_dircookie_unlock(dnp);
2249 } else {
2250 ncl_dircookie_unlock(dnp);
2251 return (NFSERR_BAD_COOKIE);
2252 }
2253
2254 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2255 (void)ncl_fsinfo(nmp, vp, cred, td);
2256
2257 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2258 &attrflag, &eof, NULL);
2259 if (attrflag)
2260 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2261
2262 if (!error) {
2263 /*
2264 * We are now either at the end of the directory or have filled
2265 * the block.
2266 */
2267 if (eof)
2268 dnp->n_direofoffset = uiop->uio_offset;
2269 else {
2270 if (uiop->uio_resid > 0)
2271 ncl_printf("EEK! readdirrpc resid > 0\n");
2272 ncl_dircookie_lock(dnp);
2273 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2274 *cookiep = cookie;
2275 ncl_dircookie_unlock(dnp);
2276 }
2277 } else if (NFS_ISV4(vp)) {
2278 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2279 }
2280 return (error);
2281 }
2282
2283 /*
2284 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2285 */
2286 int
2287 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2288 struct thread *td)
2289 {
2290 struct nfsvattr nfsva;
2291 nfsuint64 *cookiep, cookie;
2292 struct nfsnode *dnp = VTONFS(vp);
2293 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2294 int error = 0, attrflag, eof;
2295
2296 KASSERT(uiop->uio_iovcnt == 1 &&
2297 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2298 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2299 ("nfs readdirplusrpc bad uio"));
2300
2301 /*
2302 * If there is no cookie, assume directory was stale.
2303 */
2304 ncl_dircookie_lock(dnp);
2305 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2306 if (cookiep) {
2307 cookie = *cookiep;
2308 ncl_dircookie_unlock(dnp);
2309 } else {
2310 ncl_dircookie_unlock(dnp);
2311 return (NFSERR_BAD_COOKIE);
2312 }
2313
2314 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2315 (void)ncl_fsinfo(nmp, vp, cred, td);
2316 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2317 &attrflag, &eof, NULL);
2318 if (attrflag)
2319 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2320
2321 if (!error) {
2322 /*
2323 * We are now either at end of the directory or have filled the
2324 * the block.
2325 */
2326 if (eof)
2327 dnp->n_direofoffset = uiop->uio_offset;
2328 else {
2329 if (uiop->uio_resid > 0)
2330 ncl_printf("EEK! readdirplusrpc resid > 0\n");
2331 ncl_dircookie_lock(dnp);
2332 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2333 *cookiep = cookie;
2334 ncl_dircookie_unlock(dnp);
2335 }
2336 } else if (NFS_ISV4(vp)) {
2337 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2338 }
2339 return (error);
2340 }
2341
2342 /*
2343 * Silly rename. To make the NFS filesystem that is stateless look a little
2344 * more like the "ufs" a remove of an active vnode is translated to a rename
2345 * to a funny looking filename that is removed by nfs_inactive on the
2346 * nfsnode. There is the potential for another process on a different client
2347 * to create the same funny name between the nfs_lookitup() fails and the
2348 * nfs_rename() completes, but...
2349 */
2350 static int
2351 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2352 {
2353 struct sillyrename *sp;
2354 struct nfsnode *np;
2355 int error;
2356 short pid;
2357 unsigned int lticks;
2358
2359 cache_purge(dvp);
2360 np = VTONFS(vp);
2361 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2362 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2363 M_NEWNFSREQ, M_WAITOK);
2364 sp->s_cred = crhold(cnp->cn_cred);
2365 sp->s_dvp = dvp;
2366 VREF(dvp);
2367
2368 /*
2369 * Fudge together a funny name.
2370 * Changing the format of the funny name to accomodate more
2371 * sillynames per directory.
2372 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2373 * CPU ticks since boot.
2374 */
2375 pid = cnp->cn_thread->td_proc->p_pid;
2376 lticks = (unsigned int)ticks;
2377 for ( ; ; ) {
2378 sp->s_namlen = sprintf(sp->s_name,
2379 ".nfs.%08x.%04x4.4", lticks,
2380 pid);
2381 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2382 cnp->cn_thread, NULL))
2383 break;
2384 lticks++;
2385 }
2386 error = nfs_renameit(dvp, vp, cnp, sp);
2387 if (error)
2388 goto bad;
2389 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2390 cnp->cn_thread, &np);
2391 np->n_sillyrename = sp;
2392 return (0);
2393 bad:
2394 vrele(sp->s_dvp);
2395 crfree(sp->s_cred);
2396 free((caddr_t)sp, M_NEWNFSREQ);
2397 return (error);
2398 }
2399
2400 /*
2401 * Look up a file name and optionally either update the file handle or
2402 * allocate an nfsnode, depending on the value of npp.
2403 * npp == NULL --> just do the lookup
2404 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2405 * handled too
2406 * *npp != NULL --> update the file handle in the vnode
2407 */
2408 static int
2409 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2410 struct thread *td, struct nfsnode **npp)
2411 {
2412 struct vnode *newvp = NULL, *vp;
2413 struct nfsnode *np, *dnp = VTONFS(dvp);
2414 struct nfsfh *nfhp, *onfhp;
2415 struct nfsvattr nfsva, dnfsva;
2416 struct componentname cn;
2417 int error = 0, attrflag, dattrflag;
2418 u_int hash;
2419
2420 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2421 &nfhp, &attrflag, &dattrflag, NULL);
2422 if (dattrflag)
2423 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2424 if (npp && !error) {
2425 if (*npp != NULL) {
2426 np = *npp;
2427 vp = NFSTOV(np);
2428 /*
2429 * For NFSv4, check to see if it is the same name and
2430 * replace the name, if it is different.
2431 */
2432 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2433 (np->n_v4->n4_namelen != len ||
2434 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2435 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2436 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2437 dnp->n_fhp->nfh_len))) {
2438 #ifdef notdef
2439 { char nnn[100]; int nnnl;
2440 nnnl = (len < 100) ? len : 99;
2441 bcopy(name, nnn, nnnl);
2442 nnn[nnnl] = '\0';
2443 printf("replace=%s\n",nnn);
2444 }
2445 #endif
2446 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2447 MALLOC(np->n_v4, struct nfsv4node *,
2448 sizeof (struct nfsv4node) +
2449 dnp->n_fhp->nfh_len + len - 1,
2450 M_NFSV4NODE, M_WAITOK);
2451 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2452 np->n_v4->n4_namelen = len;
2453 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2454 dnp->n_fhp->nfh_len);
2455 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2456 }
2457 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2458 FNV1_32_INIT);
2459 onfhp = np->n_fhp;
2460 /*
2461 * Rehash node for new file handle.
2462 */
2463 vfs_hash_rehash(vp, hash);
2464 np->n_fhp = nfhp;
2465 if (onfhp != NULL)
2466 FREE((caddr_t)onfhp, M_NFSFH);
2467 newvp = NFSTOV(np);
2468 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2469 FREE((caddr_t)nfhp, M_NFSFH);
2470 VREF(dvp);
2471 newvp = dvp;
2472 } else {
2473 cn.cn_nameptr = name;
2474 cn.cn_namelen = len;
2475 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2476 &np, NULL, LK_EXCLUSIVE);
2477 if (error)
2478 return (error);
2479 newvp = NFSTOV(np);
2480 }
2481 if (!attrflag && *npp == NULL) {
2482 if (newvp == dvp)
2483 vrele(newvp);
2484 else
2485 vput(newvp);
2486 return (ENOENT);
2487 }
2488 if (attrflag)
2489 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2490 0, 1);
2491 }
2492 if (npp && *npp == NULL) {
2493 if (error) {
2494 if (newvp) {
2495 if (newvp == dvp)
2496 vrele(newvp);
2497 else
2498 vput(newvp);
2499 }
2500 } else
2501 *npp = np;
2502 }
2503 if (error && NFS_ISV4(dvp))
2504 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2505 return (error);
2506 }
2507
2508 /*
2509 * Nfs Version 3 and 4 commit rpc
2510 */
2511 int
2512 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2513 struct thread *td)
2514 {
2515 struct nfsvattr nfsva;
2516 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2517 int error, attrflag;
2518 u_char verf[NFSX_VERF];
2519
2520 mtx_lock(&nmp->nm_mtx);
2521 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2522 mtx_unlock(&nmp->nm_mtx);
2523 return (0);
2524 }
2525 mtx_unlock(&nmp->nm_mtx);
2526 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2527 &attrflag, NULL);
2528 if (!error) {
2529 mtx_lock(&nmp->nm_mtx);
2530 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2531 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2532 error = NFSERR_STALEWRITEVERF;
2533 }
2534 mtx_unlock(&nmp->nm_mtx);
2535 if (!error && attrflag)
2536 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2537 0, 1);
2538 } else if (NFS_ISV4(vp)) {
2539 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2540 }
2541 return (error);
2542 }
2543
2544 /*
2545 * Strategy routine.
2546 * For async requests when nfsiod(s) are running, queue the request by
2547 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2548 * request.
2549 */
2550 static int
2551 nfs_strategy(struct vop_strategy_args *ap)
2552 {
2553 struct buf *bp = ap->a_bp;
2554 struct ucred *cr;
2555
2556 KASSERT(!(bp->b_flags & B_DONE),
2557 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2558 BUF_ASSERT_HELD(bp);
2559
2560 if (bp->b_iocmd == BIO_READ)
2561 cr = bp->b_rcred;
2562 else
2563 cr = bp->b_wcred;
2564
2565 /*
2566 * If the op is asynchronous and an i/o daemon is waiting
2567 * queue the request, wake it up and wait for completion
2568 * otherwise just do it ourselves.
2569 */
2570 if ((bp->b_flags & B_ASYNC) == 0 ||
2571 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2572 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2573 return (0);
2574 }
2575
2576 /*
2577 * fsync vnode op. Just call ncl_flush() with commit == 1.
2578 */
2579 /* ARGSUSED */
2580 static int
2581 nfs_fsync(struct vop_fsync_args *ap)
2582 {
2583 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2584 }
2585
2586 /*
2587 * Flush all the blocks associated with a vnode.
2588 * Walk through the buffer pool and push any dirty pages
2589 * associated with the vnode.
2590 * If the called_from_renewthread argument is TRUE, it has been called
2591 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2592 * waiting for a buffer write to complete.
2593 */
2594 int
2595 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2596 int commit, int called_from_renewthread)
2597 {
2598 struct nfsnode *np = VTONFS(vp);
2599 struct buf *bp;
2600 int i;
2601 struct buf *nbp;
2602 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2603 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2604 int passone = 1, trycnt = 0;
2605 u_quad_t off, endoff, toff;
2606 struct ucred* wcred = NULL;
2607 struct buf **bvec = NULL;
2608 struct bufobj *bo;
2609 #ifndef NFS_COMMITBVECSIZ
2610 #define NFS_COMMITBVECSIZ 20
2611 #endif
2612 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2613 int bvecsize = 0, bveccount;
2614
2615 if (called_from_renewthread != 0)
2616 slptimeo = hz;
2617 if (nmp->nm_flag & NFSMNT_INT)
2618 slpflag = NFS_PCATCH;
2619 if (!commit)
2620 passone = 0;
2621 bo = &vp->v_bufobj;
2622 /*
2623 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2624 * server, but has not been committed to stable storage on the server
2625 * yet. On the first pass, the byte range is worked out and the commit
2626 * rpc is done. On the second pass, ncl_writebp() is called to do the
2627 * job.
2628 */
2629 again:
2630 off = (u_quad_t)-1;
2631 endoff = 0;
2632 bvecpos = 0;
2633 if (NFS_ISV34(vp) && commit) {
2634 if (bvec != NULL && bvec != bvec_on_stack)
2635 free(bvec, M_TEMP);
2636 /*
2637 * Count up how many buffers waiting for a commit.
2638 */
2639 bveccount = 0;
2640 BO_LOCK(bo);
2641 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2642 if (!BUF_ISLOCKED(bp) &&
2643 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2644 == (B_DELWRI | B_NEEDCOMMIT))
2645 bveccount++;
2646 }
2647 /*
2648 * Allocate space to remember the list of bufs to commit. It is
2649 * important to use M_NOWAIT here to avoid a race with nfs_write.
2650 * If we can't get memory (for whatever reason), we will end up
2651 * committing the buffers one-by-one in the loop below.
2652 */
2653 if (bveccount > NFS_COMMITBVECSIZ) {
2654 /*
2655 * Release the vnode interlock to avoid a lock
2656 * order reversal.
2657 */
2658 BO_UNLOCK(bo);
2659 bvec = (struct buf **)
2660 malloc(bveccount * sizeof(struct buf *),
2661 M_TEMP, M_NOWAIT);
2662 BO_LOCK(bo);
2663 if (bvec == NULL) {
2664 bvec = bvec_on_stack;
2665 bvecsize = NFS_COMMITBVECSIZ;
2666 } else
2667 bvecsize = bveccount;
2668 } else {
2669 bvec = bvec_on_stack;
2670 bvecsize = NFS_COMMITBVECSIZ;
2671 }
2672 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2673 if (bvecpos >= bvecsize)
2674 break;
2675 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2676 nbp = TAILQ_NEXT(bp, b_bobufs);
2677 continue;
2678 }
2679 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2680 (B_DELWRI | B_NEEDCOMMIT)) {
2681 BUF_UNLOCK(bp);
2682 nbp = TAILQ_NEXT(bp, b_bobufs);
2683 continue;
2684 }
2685 BO_UNLOCK(bo);
2686 bremfree(bp);
2687 /*
2688 * Work out if all buffers are using the same cred
2689 * so we can deal with them all with one commit.
2690 *
2691 * NOTE: we are not clearing B_DONE here, so we have
2692 * to do it later on in this routine if we intend to
2693 * initiate I/O on the bp.
2694 *
2695 * Note: to avoid loopback deadlocks, we do not
2696 * assign b_runningbufspace.
2697 */
2698 if (wcred == NULL)
2699 wcred = bp->b_wcred;
2700 else if (wcred != bp->b_wcred)
2701 wcred = NOCRED;
2702 vfs_busy_pages(bp, 1);
2703
2704 BO_LOCK(bo);
2705 /*
2706 * bp is protected by being locked, but nbp is not
2707 * and vfs_busy_pages() may sleep. We have to
2708 * recalculate nbp.
2709 */
2710 nbp = TAILQ_NEXT(bp, b_bobufs);
2711
2712 /*
2713 * A list of these buffers is kept so that the
2714 * second loop knows which buffers have actually
2715 * been committed. This is necessary, since there
2716 * may be a race between the commit rpc and new
2717 * uncommitted writes on the file.
2718 */
2719 bvec[bvecpos++] = bp;
2720 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2721 bp->b_dirtyoff;
2722 if (toff < off)
2723 off = toff;
2724 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2725 if (toff > endoff)
2726 endoff = toff;
2727 }
2728 BO_UNLOCK(bo);
2729 }
2730 if (bvecpos > 0) {
2731 /*
2732 * Commit data on the server, as required.
2733 * If all bufs are using the same wcred, then use that with
2734 * one call for all of them, otherwise commit each one
2735 * separately.
2736 */
2737 if (wcred != NOCRED)
2738 retv = ncl_commit(vp, off, (int)(endoff - off),
2739 wcred, td);
2740 else {
2741 retv = 0;
2742 for (i = 0; i < bvecpos; i++) {
2743 off_t off, size;
2744 bp = bvec[i];
2745 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2746 bp->b_dirtyoff;
2747 size = (u_quad_t)(bp->b_dirtyend
2748 - bp->b_dirtyoff);
2749 retv = ncl_commit(vp, off, (int)size,
2750 bp->b_wcred, td);
2751 if (retv) break;
2752 }
2753 }
2754
2755 if (retv == NFSERR_STALEWRITEVERF)
2756 ncl_clearcommit(vp->v_mount);
2757
2758 /*
2759 * Now, either mark the blocks I/O done or mark the
2760 * blocks dirty, depending on whether the commit
2761 * succeeded.
2762 */
2763 for (i = 0; i < bvecpos; i++) {
2764 bp = bvec[i];
2765 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2766 if (retv) {
2767 /*
2768 * Error, leave B_DELWRI intact
2769 */
2770 vfs_unbusy_pages(bp);
2771 brelse(bp);
2772 } else {
2773 /*
2774 * Success, remove B_DELWRI ( bundirty() ).
2775 *
2776 * b_dirtyoff/b_dirtyend seem to be NFS
2777 * specific. We should probably move that
2778 * into bundirty(). XXX
2779 */
2780 bufobj_wref(bo);
2781 bp->b_flags |= B_ASYNC;
2782 bundirty(bp);
2783 bp->b_flags &= ~B_DONE;
2784 bp->b_ioflags &= ~BIO_ERROR;
2785 bp->b_dirtyoff = bp->b_dirtyend = 0;
2786 bufdone(bp);
2787 }
2788 }
2789 }
2790
2791 /*
2792 * Start/do any write(s) that are required.
2793 */
2794 loop:
2795 BO_LOCK(bo);
2796 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2797 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2798 if (waitfor != MNT_WAIT || passone)
2799 continue;
2800
2801 error = BUF_TIMELOCK(bp,
2802 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2803 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2804 if (error == 0) {
2805 BUF_UNLOCK(bp);
2806 goto loop;
2807 }
2808 if (error == ENOLCK) {
2809 error = 0;
2810 goto loop;
2811 }
2812 if (called_from_renewthread != 0) {
2813 /*
2814 * Return EIO so the flush will be retried
2815 * later.
2816 */
2817 error = EIO;
2818 goto done;
2819 }
2820 if (newnfs_sigintr(nmp, td)) {
2821 error = EINTR;
2822 goto done;
2823 }
2824 if (slpflag & PCATCH) {
2825 slpflag = 0;
2826 slptimeo = 2 * hz;
2827 }
2828 goto loop;
2829 }
2830 if ((bp->b_flags & B_DELWRI) == 0)
2831 panic("nfs_fsync: not dirty");
2832 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2833 BUF_UNLOCK(bp);
2834 continue;
2835 }
2836 BO_UNLOCK(bo);
2837 bremfree(bp);
2838 if (passone || !commit)
2839 bp->b_flags |= B_ASYNC;
2840 else
2841 bp->b_flags |= B_ASYNC;
2842 bwrite(bp);
2843 if (newnfs_sigintr(nmp, td)) {
2844 error = EINTR;
2845 goto done;
2846 }
2847 goto loop;
2848 }
2849 if (passone) {
2850 passone = 0;
2851 BO_UNLOCK(bo);
2852 goto again;
2853 }
2854 if (waitfor == MNT_WAIT) {
2855 while (bo->bo_numoutput) {
2856 error = bufobj_wwait(bo, slpflag, slptimeo);
2857 if (error) {
2858 BO_UNLOCK(bo);
2859 if (called_from_renewthread != 0) {
2860 /*
2861 * Return EIO so that the flush will be
2862 * retried later.
2863 */
2864 error = EIO;
2865 goto done;
2866 }
2867 error = newnfs_sigintr(nmp, td);
2868 if (error)
2869 goto done;
2870 if (slpflag & PCATCH) {
2871 slpflag = 0;
2872 slptimeo = 2 * hz;
2873 }
2874 BO_LOCK(bo);
2875 }
2876 }
2877 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2878 BO_UNLOCK(bo);
2879 goto loop;
2880 }
2881 /*
2882 * Wait for all the async IO requests to drain
2883 */
2884 BO_UNLOCK(bo);
2885 mtx_lock(&np->n_mtx);
2886 while (np->n_directio_asyncwr > 0) {
2887 np->n_flag |= NFSYNCWAIT;
2888 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2889 &np->n_mtx, slpflag | (PRIBIO + 1),
2890 "nfsfsync", 0);
2891 if (error) {
2892 if (newnfs_sigintr(nmp, td)) {
2893 mtx_unlock(&np->n_mtx);
2894 error = EINTR;
2895 goto done;
2896 }
2897 }
2898 }
2899 mtx_unlock(&np->n_mtx);
2900 } else
2901 BO_UNLOCK(bo);
2902 mtx_lock(&np->n_mtx);
2903 if (np->n_flag & NWRITEERR) {
2904 error = np->n_error;
2905 np->n_flag &= ~NWRITEERR;
2906 }
2907 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2908 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2909 np->n_flag &= ~NMODIFIED;
2910 mtx_unlock(&np->n_mtx);
2911 done:
2912 if (bvec != NULL && bvec != bvec_on_stack)
2913 free(bvec, M_TEMP);
2914 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2915 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2916 np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2917 /* try, try again... */
2918 passone = 1;
2919 wcred = NULL;
2920 bvec = NULL;
2921 bvecsize = 0;
2922 printf("try%d\n", trycnt);
2923 goto again;
2924 }
2925 return (error);
2926 }
2927
2928 /*
2929 * NFS advisory byte-level locks.
2930 */
2931 static int
2932 nfs_advlock(struct vop_advlock_args *ap)
2933 {
2934 struct vnode *vp = ap->a_vp;
2935 struct ucred *cred;
2936 struct nfsnode *np = VTONFS(ap->a_vp);
2937 struct proc *p = (struct proc *)ap->a_id;
2938 struct thread *td = curthread; /* XXX */
2939 struct vattr va;
2940 int ret, error = EOPNOTSUPP;
2941 u_quad_t size;
2942
2943 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
2944 if ((ap->a_flags & F_POSIX) != 0)
2945 cred = p->p_ucred;
2946 else
2947 cred = td->td_ucred;
2948 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2949 if (vp->v_iflag & VI_DOOMED) {
2950 NFSVOPUNLOCK(vp, 0);
2951 return (EBADF);
2952 }
2953
2954 /*
2955 * If this is unlocking a write locked region, flush and
2956 * commit them before unlocking. This is required by
2957 * RFC3530 Sec. 9.3.2.
2958 */
2959 if (ap->a_op == F_UNLCK &&
2960 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
2961 ap->a_flags))
2962 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
2963
2964 /*
2965 * Loop around doing the lock op, while a blocking lock
2966 * must wait for the lock op to succeed.
2967 */
2968 do {
2969 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
2970 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
2971 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2972 ap->a_op == F_SETLK) {
2973 NFSVOPUNLOCK(vp, 0);
2974 error = nfs_catnap(PZERO | PCATCH, ret,
2975 "ncladvl");
2976 if (error)
2977 return (EINTR);
2978 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2979 if (vp->v_iflag & VI_DOOMED) {
2980 NFSVOPUNLOCK(vp, 0);
2981 return (EBADF);
2982 }
2983 }
2984 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2985 ap->a_op == F_SETLK);
2986 if (ret == NFSERR_DENIED) {
2987 NFSVOPUNLOCK(vp, 0);
2988 return (EAGAIN);
2989 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
2990 NFSVOPUNLOCK(vp, 0);
2991 return (ret);
2992 } else if (ret != 0) {
2993 NFSVOPUNLOCK(vp, 0);
2994 return (EACCES);
2995 }
2996
2997 /*
2998 * Now, if we just got a lock, invalidate data in the buffer
2999 * cache, as required, so that the coherency conforms with
3000 * RFC3530 Sec. 9.3.2.
3001 */
3002 if (ap->a_op == F_SETLK) {
3003 if ((np->n_flag & NMODIFIED) == 0) {
3004 np->n_attrstamp = 0;
3005 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3006 ret = VOP_GETATTR(vp, &va, cred);
3007 }
3008 if ((np->n_flag & NMODIFIED) || ret ||
3009 np->n_change != va.va_filerev) {
3010 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3011 np->n_attrstamp = 0;
3012 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3013 ret = VOP_GETATTR(vp, &va, cred);
3014 if (!ret) {
3015 np->n_mtime = va.va_mtime;
3016 np->n_change = va.va_filerev;
3017 }
3018 }
3019 }
3020 NFSVOPUNLOCK(vp, 0);
3021 return (0);
3022 } else if (!NFS_ISV4(vp)) {
3023 error = NFSVOPLOCK(vp, LK_SHARED);
3024 if (error)
3025 return (error);
3026 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3027 size = VTONFS(vp)->n_size;
3028 NFSVOPUNLOCK(vp, 0);
3029 error = lf_advlock(ap, &(vp->v_lockf), size);
3030 } else {
3031 if (nfs_advlock_p != NULL)
3032 error = nfs_advlock_p(ap);
3033 else {
3034 NFSVOPUNLOCK(vp, 0);
3035 error = ENOLCK;
3036 }
3037 }
3038 }
3039 return (error);
3040 }
3041
3042 /*
3043 * NFS advisory byte-level locks.
3044 */
3045 static int
3046 nfs_advlockasync(struct vop_advlockasync_args *ap)
3047 {
3048 struct vnode *vp = ap->a_vp;
3049 u_quad_t size;
3050 int error;
3051
3052 if (NFS_ISV4(vp))
3053 return (EOPNOTSUPP);
3054 error = NFSVOPLOCK(vp, LK_SHARED);
3055 if (error)
3056 return (error);
3057 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3058 size = VTONFS(vp)->n_size;
3059 NFSVOPUNLOCK(vp, 0);
3060 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3061 } else {
3062 NFSVOPUNLOCK(vp, 0);
3063 error = EOPNOTSUPP;
3064 }
3065 return (error);
3066 }
3067
3068 /*
3069 * Print out the contents of an nfsnode.
3070 */
3071 static int
3072 nfs_print(struct vop_print_args *ap)
3073 {
3074 struct vnode *vp = ap->a_vp;
3075 struct nfsnode *np = VTONFS(vp);
3076
3077 ncl_printf("\tfileid %ld fsid 0x%x",
3078 np->n_vattr.na_fileid, np->n_vattr.na_fsid);
3079 if (vp->v_type == VFIFO)
3080 fifo_printinfo(vp);
3081 printf("\n");
3082 return (0);
3083 }
3084
3085 /*
3086 * This is the "real" nfs::bwrite(struct buf*).
3087 * We set B_CACHE if this is a VMIO buffer.
3088 */
3089 int
3090 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3091 {
3092 int s;
3093 int oldflags = bp->b_flags;
3094 #if 0
3095 int retv = 1;
3096 off_t off;
3097 #endif
3098
3099 BUF_ASSERT_HELD(bp);
3100
3101 if (bp->b_flags & B_INVAL) {
3102 brelse(bp);
3103 return(0);
3104 }
3105
3106 bp->b_flags |= B_CACHE;
3107
3108 /*
3109 * Undirty the bp. We will redirty it later if the I/O fails.
3110 */
3111
3112 s = splbio();
3113 bundirty(bp);
3114 bp->b_flags &= ~B_DONE;
3115 bp->b_ioflags &= ~BIO_ERROR;
3116 bp->b_iocmd = BIO_WRITE;
3117
3118 bufobj_wref(bp->b_bufobj);
3119 curthread->td_ru.ru_oublock++;
3120 splx(s);
3121
3122 /*
3123 * Note: to avoid loopback deadlocks, we do not
3124 * assign b_runningbufspace.
3125 */
3126 vfs_busy_pages(bp, 1);
3127
3128 BUF_KERNPROC(bp);
3129 bp->b_iooffset = dbtob(bp->b_blkno);
3130 bstrategy(bp);
3131
3132 if( (oldflags & B_ASYNC) == 0) {
3133 int rtval = bufwait(bp);
3134
3135 if (oldflags & B_DELWRI) {
3136 s = splbio();
3137 reassignbuf(bp);
3138 splx(s);
3139 }
3140 brelse(bp);
3141 return (rtval);
3142 }
3143
3144 return (0);
3145 }
3146
3147 /*
3148 * nfs special file access vnode op.
3149 * Essentially just get vattr and then imitate iaccess() since the device is
3150 * local to the client.
3151 */
3152 static int
3153 nfsspec_access(struct vop_access_args *ap)
3154 {
3155 struct vattr *vap;
3156 struct ucred *cred = ap->a_cred;
3157 struct vnode *vp = ap->a_vp;
3158 accmode_t accmode = ap->a_accmode;
3159 struct vattr vattr;
3160 int error;
3161
3162 /*
3163 * Disallow write attempts on filesystems mounted read-only;
3164 * unless the file is a socket, fifo, or a block or character
3165 * device resident on the filesystem.
3166 */
3167 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3168 switch (vp->v_type) {
3169 case VREG:
3170 case VDIR:
3171 case VLNK:
3172 return (EROFS);
3173 default:
3174 break;
3175 }
3176 }
3177 vap = &vattr;
3178 error = VOP_GETATTR(vp, vap, cred);
3179 if (error)
3180 goto out;
3181 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3182 accmode, cred, NULL);
3183 out:
3184 return error;
3185 }
3186
3187 /*
3188 * Read wrapper for fifos.
3189 */
3190 static int
3191 nfsfifo_read(struct vop_read_args *ap)
3192 {
3193 struct nfsnode *np = VTONFS(ap->a_vp);
3194 int error;
3195
3196 /*
3197 * Set access flag.
3198 */
3199 mtx_lock(&np->n_mtx);
3200 np->n_flag |= NACC;
3201 getnanotime(&np->n_atim);
3202 mtx_unlock(&np->n_mtx);
3203 error = fifo_specops.vop_read(ap);
3204 return error;
3205 }
3206
3207 /*
3208 * Write wrapper for fifos.
3209 */
3210 static int
3211 nfsfifo_write(struct vop_write_args *ap)
3212 {
3213 struct nfsnode *np = VTONFS(ap->a_vp);
3214
3215 /*
3216 * Set update flag.
3217 */
3218 mtx_lock(&np->n_mtx);
3219 np->n_flag |= NUPD;
3220 getnanotime(&np->n_mtim);
3221 mtx_unlock(&np->n_mtx);
3222 return(fifo_specops.vop_write(ap));
3223 }
3224
3225 /*
3226 * Close wrapper for fifos.
3227 *
3228 * Update the times on the nfsnode then do fifo close.
3229 */
3230 static int
3231 nfsfifo_close(struct vop_close_args *ap)
3232 {
3233 struct vnode *vp = ap->a_vp;
3234 struct nfsnode *np = VTONFS(vp);
3235 struct vattr vattr;
3236 struct timespec ts;
3237
3238 mtx_lock(&np->n_mtx);
3239 if (np->n_flag & (NACC | NUPD)) {
3240 getnanotime(&ts);
3241 if (np->n_flag & NACC)
3242 np->n_atim = ts;
3243 if (np->n_flag & NUPD)
3244 np->n_mtim = ts;
3245 np->n_flag |= NCHG;
3246 if (vrefcnt(vp) == 1 &&
3247 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3248 VATTR_NULL(&vattr);
3249 if (np->n_flag & NACC)
3250 vattr.va_atime = np->n_atim;
3251 if (np->n_flag & NUPD)
3252 vattr.va_mtime = np->n_mtim;
3253 mtx_unlock(&np->n_mtx);
3254 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3255 goto out;
3256 }
3257 }
3258 mtx_unlock(&np->n_mtx);
3259 out:
3260 return (fifo_specops.vop_close(ap));
3261 }
3262
3263 /*
3264 * Just call ncl_writebp() with the force argument set to 1.
3265 *
3266 * NOTE: B_DONE may or may not be set in a_bp on call.
3267 */
3268 static int
3269 nfs_bwrite(struct buf *bp)
3270 {
3271
3272 return (ncl_writebp(bp, 1, curthread));
3273 }
3274
3275 struct buf_ops buf_ops_newnfs = {
3276 .bop_name = "buf_ops_nfs",
3277 .bop_write = nfs_bwrite,
3278 .bop_strategy = bufstrategy,
3279 .bop_sync = bufsync,
3280 .bop_bdflush = bufbdflush,
3281 };
3282
3283 /*
3284 * Cloned from vop_stdlock(), and then the ugly hack added.
3285 */
3286 static int
3287 nfs_lock1(struct vop_lock1_args *ap)
3288 {
3289 struct vnode *vp = ap->a_vp;
3290 int error = 0;
3291
3292 /*
3293 * Since vfs_hash_get() calls vget() and it will no longer work
3294 * for FreeBSD8 with flags == 0, I can only think of this horrible
3295 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3296 * and then handle it here. All I want for this case is a v_usecount
3297 * on the vnode to use for recovery, while another thread might
3298 * hold a lock on the vnode. I have the other threads blocked, so
3299 * there isn't any race problem.
3300 */
3301 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3302 if ((ap->a_flags & LK_INTERLOCK) == 0)
3303 panic("ncllock1");
3304 if ((vp->v_iflag & VI_DOOMED))
3305 error = ENOENT;
3306 VI_UNLOCK(vp);
3307 return (error);
3308 }
3309 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3310 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3311 ap->a_line));
3312 }
3313
3314 static int
3315 nfs_getacl(struct vop_getacl_args *ap)
3316 {
3317 int error;
3318
3319 if (ap->a_type != ACL_TYPE_NFS4)
3320 return (EOPNOTSUPP);
3321 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3322 NULL);
3323 if (error > NFSERR_STALE) {
3324 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3325 error = EPERM;
3326 }
3327 return (error);
3328 }
3329
3330 static int
3331 nfs_setacl(struct vop_setacl_args *ap)
3332 {
3333 int error;
3334
3335 if (ap->a_type != ACL_TYPE_NFS4)
3336 return (EOPNOTSUPP);
3337 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3338 NULL);
3339 if (error > NFSERR_STALE) {
3340 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3341 error = EPERM;
3342 }
3343 return (error);
3344 }
3345
3346 /*
3347 * Return POSIX pathconf information applicable to nfs filesystems.
3348 */
3349 static int
3350 nfs_pathconf(struct vop_pathconf_args *ap)
3351 {
3352 struct nfsv3_pathconf pc;
3353 struct nfsvattr nfsva;
3354 struct vnode *vp = ap->a_vp;
3355 struct thread *td = curthread;
3356 int attrflag, error;
3357
3358 if (NFS_ISV4(vp) || (NFS_ISV3(vp) && (ap->a_name == _PC_LINK_MAX ||
3359 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3360 ap->a_name == _PC_NO_TRUNC))) {
3361 /*
3362 * Since only the above 4 a_names are returned by the NFSv3
3363 * Pathconf RPC, there is no point in doing it for others.
3364 */
3365 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3366 &attrflag, NULL);
3367 if (attrflag != 0)
3368 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3369 1);
3370 if (error != 0)
3371 return (error);
3372 } else {
3373 /*
3374 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3375 * just fake them.
3376 */
3377 pc.pc_linkmax = LINK_MAX;
3378 pc.pc_namemax = NFS_MAXNAMLEN;
3379 pc.pc_notrunc = 1;
3380 pc.pc_chownrestricted = 1;
3381 pc.pc_caseinsensitive = 0;
3382 pc.pc_casepreserving = 1;
3383 error = 0;
3384 }
3385 switch (ap->a_name) {
3386 case _PC_LINK_MAX:
3387 *ap->a_retval = pc.pc_linkmax;
3388 break;
3389 case _PC_NAME_MAX:
3390 *ap->a_retval = pc.pc_namemax;
3391 break;
3392 case _PC_PATH_MAX:
3393 *ap->a_retval = PATH_MAX;
3394 break;
3395 case _PC_PIPE_BUF:
3396 *ap->a_retval = PIPE_BUF;
3397 break;
3398 case _PC_CHOWN_RESTRICTED:
3399 *ap->a_retval = pc.pc_chownrestricted;
3400 break;
3401 case _PC_NO_TRUNC:
3402 *ap->a_retval = pc.pc_notrunc;
3403 break;
3404 case _PC_ACL_EXTENDED:
3405 *ap->a_retval = 0;
3406 break;
3407 case _PC_ACL_NFS4:
3408 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3409 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3410 *ap->a_retval = 1;
3411 else
3412 *ap->a_retval = 0;
3413 break;
3414 case _PC_ACL_PATH_MAX:
3415 if (NFS_ISV4(vp))
3416 *ap->a_retval = ACL_MAX_ENTRIES;
3417 else
3418 *ap->a_retval = 3;
3419 break;
3420 case _PC_MAC_PRESENT:
3421 *ap->a_retval = 0;
3422 break;
3423 case _PC_ASYNC_IO:
3424 /* _PC_ASYNC_IO should have been handled by upper layers. */
3425 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3426 error = EINVAL;
3427 break;
3428 case _PC_PRIO_IO:
3429 *ap->a_retval = 0;
3430 break;
3431 case _PC_SYNC_IO:
3432 *ap->a_retval = 0;
3433 break;
3434 case _PC_ALLOC_SIZE_MIN:
3435 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3436 break;
3437 case _PC_FILESIZEBITS:
3438 if (NFS_ISV34(vp))
3439 *ap->a_retval = 64;
3440 else
3441 *ap->a_retval = 32;
3442 break;
3443 case _PC_REC_INCR_XFER_SIZE:
3444 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3445 break;
3446 case _PC_REC_MAX_XFER_SIZE:
3447 *ap->a_retval = -1; /* means ``unlimited'' */
3448 break;
3449 case _PC_REC_MIN_XFER_SIZE:
3450 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3451 break;
3452 case _PC_REC_XFER_ALIGN:
3453 *ap->a_retval = PAGE_SIZE;
3454 break;
3455 case _PC_SYMLINK_MAX:
3456 *ap->a_retval = NFS_MAXPATHLEN;
3457 break;
3458
3459 default:
3460 error = EINVAL;
3461 break;
3462 }
3463 return (error);
3464 }
3465
Cache object: 79a67fa6c184aa316305b9f5161af980
|