1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 /*
41 * vnode op calls for Sun NFS version 2, 3 and 4
42 */
43
44 #include "opt_inet.h"
45
46 #include <sys/param.h>
47 #include <sys/kernel.h>
48 #include <sys/systm.h>
49 #include <sys/resourcevar.h>
50 #include <sys/proc.h>
51 #include <sys/mount.h>
52 #include <sys/bio.h>
53 #include <sys/buf.h>
54 #include <sys/extattr.h>
55 #include <sys/filio.h>
56 #include <sys/jail.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.h>
59 #include <sys/namei.h>
60 #include <sys/socket.h>
61 #include <sys/vnode.h>
62 #include <sys/dirent.h>
63 #include <sys/fcntl.h>
64 #include <sys/lockf.h>
65 #include <sys/stat.h>
66 #include <sys/sysctl.h>
67 #include <sys/signalvar.h>
68
69 #include <vm/vm.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_object.h>
72
73 #include <fs/nfs/nfsport.h>
74 #include <fs/nfsclient/nfsnode.h>
75 #include <fs/nfsclient/nfsmount.h>
76 #include <fs/nfsclient/nfs.h>
77 #include <fs/nfsclient/nfs_kdtrace.h>
78
79 #include <net/if.h>
80 #include <netinet/in.h>
81 #include <netinet/in_var.h>
82
83 #include <nfs/nfs_lock.h>
84
85 #ifdef KDTRACE_HOOKS
86 #include <sys/dtrace_bsd.h>
87
88 dtrace_nfsclient_accesscache_flush_probe_func_t
89 dtrace_nfscl_accesscache_flush_done_probe;
90 uint32_t nfscl_accesscache_flush_done_id;
91
92 dtrace_nfsclient_accesscache_get_probe_func_t
93 dtrace_nfscl_accesscache_get_hit_probe,
94 dtrace_nfscl_accesscache_get_miss_probe;
95 uint32_t nfscl_accesscache_get_hit_id;
96 uint32_t nfscl_accesscache_get_miss_id;
97
98 dtrace_nfsclient_accesscache_load_probe_func_t
99 dtrace_nfscl_accesscache_load_done_probe;
100 uint32_t nfscl_accesscache_load_done_id;
101 #endif /* !KDTRACE_HOOKS */
102
103 /* Defs */
104 #define TRUE 1
105 #define FALSE 0
106
107 extern struct nfsstatsv1 nfsstatsv1;
108 extern int nfsrv_useacl;
109 extern int nfscl_debuglevel;
110 MALLOC_DECLARE(M_NEWNFSREQ);
111
112 static vop_read_t nfsfifo_read;
113 static vop_write_t nfsfifo_write;
114 static vop_close_t nfsfifo_close;
115 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
116 struct thread *);
117 static vop_lookup_t nfs_lookup;
118 static vop_create_t nfs_create;
119 static vop_mknod_t nfs_mknod;
120 static vop_open_t nfs_open;
121 static vop_pathconf_t nfs_pathconf;
122 static vop_close_t nfs_close;
123 static vop_access_t nfs_access;
124 static vop_getattr_t nfs_getattr;
125 static vop_setattr_t nfs_setattr;
126 static vop_read_t nfs_read;
127 static vop_fsync_t nfs_fsync;
128 static vop_remove_t nfs_remove;
129 static vop_link_t nfs_link;
130 static vop_rename_t nfs_rename;
131 static vop_mkdir_t nfs_mkdir;
132 static vop_rmdir_t nfs_rmdir;
133 static vop_symlink_t nfs_symlink;
134 static vop_readdir_t nfs_readdir;
135 static vop_strategy_t nfs_strategy;
136 static int nfs_lookitup(struct vnode *, char *, int,
137 struct ucred *, struct thread *, struct nfsnode **);
138 static int nfs_sillyrename(struct vnode *, struct vnode *,
139 struct componentname *);
140 static vop_access_t nfsspec_access;
141 static vop_readlink_t nfs_readlink;
142 static vop_print_t nfs_print;
143 static vop_advlock_t nfs_advlock;
144 static vop_advlockasync_t nfs_advlockasync;
145 static vop_getacl_t nfs_getacl;
146 static vop_setacl_t nfs_setacl;
147 static vop_advise_t nfs_advise;
148 static vop_allocate_t nfs_allocate;
149 static vop_deallocate_t nfs_deallocate;
150 static vop_copy_file_range_t nfs_copy_file_range;
151 static vop_ioctl_t nfs_ioctl;
152 static vop_getextattr_t nfs_getextattr;
153 static vop_setextattr_t nfs_setextattr;
154 static vop_listextattr_t nfs_listextattr;
155 static vop_deleteextattr_t nfs_deleteextattr;
156 static vop_lock1_t nfs_lock;
157
158 /*
159 * Global vfs data structures for nfs
160 */
161
162 static struct vop_vector newnfs_vnodeops_nosig = {
163 .vop_default = &default_vnodeops,
164 .vop_access = nfs_access,
165 .vop_advlock = nfs_advlock,
166 .vop_advlockasync = nfs_advlockasync,
167 .vop_close = nfs_close,
168 .vop_create = nfs_create,
169 .vop_fsync = nfs_fsync,
170 .vop_getattr = nfs_getattr,
171 .vop_getpages = ncl_getpages,
172 .vop_putpages = ncl_putpages,
173 .vop_inactive = ncl_inactive,
174 .vop_link = nfs_link,
175 .vop_lock1 = nfs_lock,
176 .vop_lookup = nfs_lookup,
177 .vop_mkdir = nfs_mkdir,
178 .vop_mknod = nfs_mknod,
179 .vop_open = nfs_open,
180 .vop_pathconf = nfs_pathconf,
181 .vop_print = nfs_print,
182 .vop_read = nfs_read,
183 .vop_readdir = nfs_readdir,
184 .vop_readlink = nfs_readlink,
185 .vop_reclaim = ncl_reclaim,
186 .vop_remove = nfs_remove,
187 .vop_rename = nfs_rename,
188 .vop_rmdir = nfs_rmdir,
189 .vop_setattr = nfs_setattr,
190 .vop_strategy = nfs_strategy,
191 .vop_symlink = nfs_symlink,
192 .vop_write = ncl_write,
193 .vop_getacl = nfs_getacl,
194 .vop_setacl = nfs_setacl,
195 .vop_advise = nfs_advise,
196 .vop_allocate = nfs_allocate,
197 .vop_deallocate = nfs_deallocate,
198 .vop_copy_file_range = nfs_copy_file_range,
199 .vop_ioctl = nfs_ioctl,
200 .vop_getextattr = nfs_getextattr,
201 .vop_setextattr = nfs_setextattr,
202 .vop_listextattr = nfs_listextattr,
203 .vop_deleteextattr = nfs_deleteextattr,
204 };
205 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig);
206
207 static int
208 nfs_vnodeops_bypass(struct vop_generic_args *a)
209 {
210
211 return (vop_sigdefer(&newnfs_vnodeops_nosig, a));
212 }
213
214 struct vop_vector newnfs_vnodeops = {
215 .vop_default = &default_vnodeops,
216 .vop_bypass = nfs_vnodeops_bypass,
217 };
218 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops);
219
220 static struct vop_vector newnfs_fifoops_nosig = {
221 .vop_default = &fifo_specops,
222 .vop_access = nfsspec_access,
223 .vop_close = nfsfifo_close,
224 .vop_fsync = nfs_fsync,
225 .vop_getattr = nfs_getattr,
226 .vop_inactive = ncl_inactive,
227 .vop_pathconf = nfs_pathconf,
228 .vop_print = nfs_print,
229 .vop_read = nfsfifo_read,
230 .vop_reclaim = ncl_reclaim,
231 .vop_setattr = nfs_setattr,
232 .vop_write = nfsfifo_write,
233 };
234 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig);
235
236 static int
237 nfs_fifoops_bypass(struct vop_generic_args *a)
238 {
239
240 return (vop_sigdefer(&newnfs_fifoops_nosig, a));
241 }
242
243 struct vop_vector newnfs_fifoops = {
244 .vop_default = &default_vnodeops,
245 .vop_bypass = nfs_fifoops_bypass,
246 };
247 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops);
248
249 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
250 struct componentname *cnp, struct vattr *vap);
251 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
252 int namelen, struct ucred *cred, struct thread *td);
253 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
254 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
255 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
256 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
257 struct componentname *scnp, struct sillyrename *sp);
258
259 /*
260 * Global variables
261 */
262 SYSCTL_DECL(_vfs_nfs);
263
264 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
265 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
266 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
267
268 static int nfs_prime_access_cache = 0;
269 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
270 &nfs_prime_access_cache, 0,
271 "Prime NFS ACCESS cache when fetching attributes");
272
273 static int newnfs_commit_on_close = 0;
274 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
275 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
276
277 static int nfs_clean_pages_on_close = 1;
278 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
279 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
280
281 int newnfs_directio_enable = 0;
282 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
283 &newnfs_directio_enable, 0, "Enable NFS directio");
284
285 int nfs_keep_dirty_on_error;
286 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
287 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
288
289 /*
290 * This sysctl allows other processes to mmap a file that has been opened
291 * O_DIRECT by a process. In general, having processes mmap the file while
292 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
293 * this by default to prevent DoS attacks - to prevent a malicious user from
294 * opening up files O_DIRECT preventing other users from mmap'ing these
295 * files. "Protected" environments where stricter consistency guarantees are
296 * required can disable this knob. The process that opened the file O_DIRECT
297 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
298 * meaningful.
299 */
300 int newnfs_directio_allow_mmap = 1;
301 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
302 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
303
304 static uint64_t nfs_maxalloclen = 64 * 1024 * 1024;
305 SYSCTL_U64(_vfs_nfs, OID_AUTO, maxalloclen, CTLFLAG_RW,
306 &nfs_maxalloclen, 0, "NFS max allocate/deallocate length");
307
308 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
309 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
310 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
311
312 /*
313 * SMP Locking Note :
314 * The list of locks after the description of the lock is the ordering
315 * of other locks acquired with the lock held.
316 * np->n_mtx : Protects the fields in the nfsnode.
317 VM Object Lock
318 VI_MTX (acquired indirectly)
319 * nmp->nm_mtx : Protects the fields in the nfsmount.
320 rep->r_mtx
321 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
322 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
323 nmp->nm_mtx
324 rep->r_mtx
325 * rep->r_mtx : Protects the fields in an nfsreq.
326 */
327
328 static int
329 nfs_lock(struct vop_lock1_args *ap)
330 {
331 struct vnode *vp;
332 struct nfsnode *np;
333 u_quad_t nsize;
334 int error, lktype;
335 bool onfault;
336
337 vp = ap->a_vp;
338 lktype = ap->a_flags & LK_TYPE_MASK;
339 error = VOP_LOCK1_APV(&default_vnodeops, ap);
340 if (error != 0 || vp->v_op != &newnfs_vnodeops)
341 return (error);
342 np = VTONFS(vp);
343 if (np == NULL)
344 return (0);
345 NFSLOCKNODE(np);
346 if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED &&
347 lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE &&
348 lktype != LK_TRYUPGRADE)) {
349 NFSUNLOCKNODE(np);
350 return (0);
351 }
352 onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT &&
353 (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE &&
354 (lktype == LK_SHARED || lktype == LK_EXCLUSIVE);
355 if (onfault && vp->v_vnlock->lk_recurse == 0) {
356 /*
357 * Force retry in vm_fault(), to make the lock request
358 * sleepable, which allows us to piggy-back the
359 * sleepable call to vnode_pager_setsize().
360 */
361 NFSUNLOCKNODE(np);
362 VOP_UNLOCK(vp);
363 return (EBUSY);
364 }
365 if ((ap->a_flags & LK_NOWAIT) != 0 ||
366 (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) {
367 NFSUNLOCKNODE(np);
368 return (0);
369 }
370 if (lktype == LK_SHARED) {
371 NFSUNLOCKNODE(np);
372 VOP_UNLOCK(vp);
373 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
374 ap->a_flags |= LK_EXCLUSIVE;
375 error = VOP_LOCK1_APV(&default_vnodeops, ap);
376 if (error != 0 || vp->v_op != &newnfs_vnodeops)
377 return (error);
378 if (vp->v_data == NULL)
379 goto downgrade;
380 MPASS(vp->v_data == np);
381 NFSLOCKNODE(np);
382 if ((np->n_flag & NVNSETSZSKIP) == 0) {
383 NFSUNLOCKNODE(np);
384 goto downgrade;
385 }
386 }
387 np->n_flag &= ~NVNSETSZSKIP;
388 nsize = np->n_size;
389 NFSUNLOCKNODE(np);
390 vnode_pager_setsize(vp, nsize);
391 downgrade:
392 if (lktype == LK_SHARED) {
393 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
394 ap->a_flags |= LK_DOWNGRADE;
395 (void)VOP_LOCK1_APV(&default_vnodeops, ap);
396 }
397 return (0);
398 }
399
400 static int
401 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
402 struct ucred *cred, u_int32_t *retmode)
403 {
404 int error = 0, attrflag, i, lrupos;
405 u_int32_t rmode;
406 struct nfsnode *np = VTONFS(vp);
407 struct nfsvattr nfsva;
408
409 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
410 &rmode);
411 if (attrflag)
412 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
413 if (!error) {
414 lrupos = 0;
415 NFSLOCKNODE(np);
416 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
417 if (np->n_accesscache[i].uid == cred->cr_uid) {
418 np->n_accesscache[i].mode = rmode;
419 np->n_accesscache[i].stamp = time_second;
420 break;
421 }
422 if (i > 0 && np->n_accesscache[i].stamp <
423 np->n_accesscache[lrupos].stamp)
424 lrupos = i;
425 }
426 if (i == NFS_ACCESSCACHESIZE) {
427 np->n_accesscache[lrupos].uid = cred->cr_uid;
428 np->n_accesscache[lrupos].mode = rmode;
429 np->n_accesscache[lrupos].stamp = time_second;
430 }
431 NFSUNLOCKNODE(np);
432 if (retmode != NULL)
433 *retmode = rmode;
434 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
435 } else if (NFS_ISV4(vp)) {
436 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
437 }
438 #ifdef KDTRACE_HOOKS
439 if (error != 0)
440 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
441 error);
442 #endif
443 return (error);
444 }
445
446 /*
447 * nfs access vnode op.
448 * For nfs version 2, just return ok. File accesses may fail later.
449 * For nfs version 3, use the access rpc to check accessibility. If file modes
450 * are changed on the server, accesses might still fail later.
451 */
452 static int
453 nfs_access(struct vop_access_args *ap)
454 {
455 struct vnode *vp = ap->a_vp;
456 int error = 0, i, gotahit;
457 u_int32_t mode, wmode, rmode;
458 int v34 = NFS_ISV34(vp);
459 struct nfsnode *np = VTONFS(vp);
460
461 /*
462 * Disallow write attempts on filesystems mounted read-only;
463 * unless the file is a socket, fifo, or a block or character
464 * device resident on the filesystem.
465 */
466 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
467 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
468 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
469 switch (vp->v_type) {
470 case VREG:
471 case VDIR:
472 case VLNK:
473 return (EROFS);
474 default:
475 break;
476 }
477 }
478 /*
479 * For nfs v3 or v4, check to see if we have done this recently, and if
480 * so return our cached result instead of making an ACCESS call.
481 * If not, do an access rpc, otherwise you are stuck emulating
482 * ufs_access() locally using the vattr. This may not be correct,
483 * since the server may apply other access criteria such as
484 * client uid-->server uid mapping that we do not know about.
485 */
486 if (v34) {
487 if (ap->a_accmode & VREAD)
488 mode = NFSACCESS_READ;
489 else
490 mode = 0;
491 if (vp->v_type != VDIR) {
492 if (ap->a_accmode & VWRITE)
493 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
494 if (ap->a_accmode & VAPPEND)
495 mode |= NFSACCESS_EXTEND;
496 if (ap->a_accmode & VEXEC)
497 mode |= NFSACCESS_EXECUTE;
498 if (ap->a_accmode & VDELETE)
499 mode |= NFSACCESS_DELETE;
500 } else {
501 if (ap->a_accmode & VWRITE)
502 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
503 if (ap->a_accmode & VAPPEND)
504 mode |= NFSACCESS_EXTEND;
505 if (ap->a_accmode & VEXEC)
506 mode |= NFSACCESS_LOOKUP;
507 if (ap->a_accmode & VDELETE)
508 mode |= NFSACCESS_DELETE;
509 if (ap->a_accmode & VDELETE_CHILD)
510 mode |= NFSACCESS_MODIFY;
511 }
512 /* XXX safety belt, only make blanket request if caching */
513 if (nfsaccess_cache_timeout > 0) {
514 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
515 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
516 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
517 } else {
518 wmode = mode;
519 }
520
521 /*
522 * Does our cached result allow us to give a definite yes to
523 * this request?
524 */
525 gotahit = 0;
526 NFSLOCKNODE(np);
527 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
528 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
529 if (time_second < (np->n_accesscache[i].stamp
530 + nfsaccess_cache_timeout) &&
531 (np->n_accesscache[i].mode & mode) == mode) {
532 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
533 gotahit = 1;
534 }
535 break;
536 }
537 }
538 NFSUNLOCKNODE(np);
539 #ifdef KDTRACE_HOOKS
540 if (gotahit != 0)
541 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
542 ap->a_cred->cr_uid, mode);
543 else
544 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
545 ap->a_cred->cr_uid, mode);
546 #endif
547 if (gotahit == 0) {
548 /*
549 * Either a no, or a don't know. Go to the wire.
550 */
551 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
552 error = nfs34_access_otw(vp, wmode, ap->a_td,
553 ap->a_cred, &rmode);
554 if (!error &&
555 (rmode & mode) != mode)
556 error = EACCES;
557 }
558 return (error);
559 } else {
560 if ((error = nfsspec_access(ap)) != 0) {
561 return (error);
562 }
563 /*
564 * Attempt to prevent a mapped root from accessing a file
565 * which it shouldn't. We try to read a byte from the file
566 * if the user is root and the file is not zero length.
567 * After calling nfsspec_access, we should have the correct
568 * file size cached.
569 */
570 NFSLOCKNODE(np);
571 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
572 && VTONFS(vp)->n_size > 0) {
573 struct iovec aiov;
574 struct uio auio;
575 char buf[1];
576
577 NFSUNLOCKNODE(np);
578 aiov.iov_base = buf;
579 aiov.iov_len = 1;
580 auio.uio_iov = &aiov;
581 auio.uio_iovcnt = 1;
582 auio.uio_offset = 0;
583 auio.uio_resid = 1;
584 auio.uio_segflg = UIO_SYSSPACE;
585 auio.uio_rw = UIO_READ;
586 auio.uio_td = ap->a_td;
587
588 if (vp->v_type == VREG)
589 error = ncl_readrpc(vp, &auio, ap->a_cred);
590 else if (vp->v_type == VDIR) {
591 char* bp;
592 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
593 aiov.iov_base = bp;
594 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
595 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
596 ap->a_td);
597 free(bp, M_TEMP);
598 } else if (vp->v_type == VLNK)
599 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
600 else
601 error = EACCES;
602 } else
603 NFSUNLOCKNODE(np);
604 return (error);
605 }
606 }
607
608 /*
609 * nfs open vnode op
610 * Check to see if the type is ok
611 * and that deletion is not in progress.
612 * For paged in text files, you will need to flush the page cache
613 * if consistency is lost.
614 */
615 /* ARGSUSED */
616 static int
617 nfs_open(struct vop_open_args *ap)
618 {
619 struct vnode *vp = ap->a_vp;
620 struct nfsnode *np = VTONFS(vp);
621 struct vattr vattr;
622 int error;
623 int fmode = ap->a_mode;
624 struct ucred *cred;
625 vm_object_t obj;
626
627 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
628 return (EOPNOTSUPP);
629
630 /*
631 * For NFSv4, we need to do the Open Op before cache validation,
632 * so that we conform to RFC3530 Sec. 9.3.1.
633 */
634 if (NFS_ISV4(vp)) {
635 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
636 if (error) {
637 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
638 (gid_t)0);
639 return (error);
640 }
641 }
642
643 /*
644 * Now, if this Open will be doing reading, re-validate/flush the
645 * cache, so that Close/Open coherency is maintained.
646 */
647 NFSLOCKNODE(np);
648 if (np->n_flag & NMODIFIED) {
649 NFSUNLOCKNODE(np);
650 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
651 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
652 if (VN_IS_DOOMED(vp))
653 return (EBADF);
654 }
655 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
656 if (error == EINTR || error == EIO) {
657 if (NFS_ISV4(vp))
658 (void) nfsrpc_close(vp, 0, ap->a_td);
659 return (error);
660 }
661 NFSLOCKNODE(np);
662 np->n_attrstamp = 0;
663 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
664 if (vp->v_type == VDIR)
665 np->n_direofoffset = 0;
666 NFSUNLOCKNODE(np);
667 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
668 if (error) {
669 if (NFS_ISV4(vp))
670 (void) nfsrpc_close(vp, 0, ap->a_td);
671 return (error);
672 }
673 NFSLOCKNODE(np);
674 np->n_mtime = vattr.va_mtime;
675 if (NFS_ISV4(vp))
676 np->n_change = vattr.va_filerev;
677 } else {
678 NFSUNLOCKNODE(np);
679 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
680 if (error) {
681 if (NFS_ISV4(vp))
682 (void) nfsrpc_close(vp, 0, ap->a_td);
683 return (error);
684 }
685 NFSLOCKNODE(np);
686 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
687 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
688 if (vp->v_type == VDIR)
689 np->n_direofoffset = 0;
690 NFSUNLOCKNODE(np);
691 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
692 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
693 if (VN_IS_DOOMED(vp))
694 return (EBADF);
695 }
696 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
697 if (error == EINTR || error == EIO) {
698 if (NFS_ISV4(vp))
699 (void) nfsrpc_close(vp, 0, ap->a_td);
700 return (error);
701 }
702 NFSLOCKNODE(np);
703 np->n_mtime = vattr.va_mtime;
704 if (NFS_ISV4(vp))
705 np->n_change = vattr.va_filerev;
706 }
707 }
708
709 /*
710 * If the object has >= 1 O_DIRECT active opens, we disable caching.
711 */
712 if (newnfs_directio_enable && (fmode & O_DIRECT) &&
713 (vp->v_type == VREG)) {
714 if (np->n_directio_opens == 0) {
715 NFSUNLOCKNODE(np);
716 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
717 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
718 if (VN_IS_DOOMED(vp))
719 return (EBADF);
720 }
721 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
722 if (error) {
723 if (NFS_ISV4(vp))
724 (void) nfsrpc_close(vp, 0, ap->a_td);
725 return (error);
726 }
727 NFSLOCKNODE(np);
728 np->n_flag |= NNONCACHE;
729 }
730 np->n_directio_opens++;
731 }
732
733 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
734 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
735 np->n_flag |= NWRITEOPENED;
736
737 /*
738 * If this is an open for writing, capture a reference to the
739 * credentials, so they can be used by ncl_putpages(). Using
740 * these write credentials is preferable to the credentials of
741 * whatever thread happens to be doing the VOP_PUTPAGES() since
742 * the write RPCs are less likely to fail with EACCES.
743 */
744 if ((fmode & FWRITE) != 0) {
745 cred = np->n_writecred;
746 np->n_writecred = crhold(ap->a_cred);
747 } else
748 cred = NULL;
749 NFSUNLOCKNODE(np);
750
751 if (cred != NULL)
752 crfree(cred);
753 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
754
755 /*
756 * If the text file has been mmap'd, flush any dirty pages to the
757 * buffer cache and then...
758 * Make sure all writes are pushed to the NFS server. If this is not
759 * done, the modify time of the file can change while the text
760 * file is being executed. This will cause the process that is
761 * executing the text file to be terminated.
762 */
763 if (vp->v_writecount <= -1) {
764 if ((obj = vp->v_object) != NULL &&
765 vm_object_mightbedirty(obj)) {
766 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
767 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
768 if (VN_IS_DOOMED(vp))
769 return (EBADF);
770 }
771 VM_OBJECT_WLOCK(obj);
772 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
773 VM_OBJECT_WUNLOCK(obj);
774 }
775
776 /* Now, flush the buffer cache. */
777 ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
778
779 /* And, finally, make sure that n_mtime is up to date. */
780 np = VTONFS(vp);
781 NFSLOCKNODE(np);
782 np->n_mtime = np->n_vattr.na_mtime;
783 NFSUNLOCKNODE(np);
784 }
785 return (0);
786 }
787
788 /*
789 * nfs close vnode op
790 * What an NFS client should do upon close after writing is a debatable issue.
791 * Most NFS clients push delayed writes to the server upon close, basically for
792 * two reasons:
793 * 1 - So that any write errors may be reported back to the client process
794 * doing the close system call. By far the two most likely errors are
795 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
796 * 2 - To put a worst case upper bound on cache inconsistency between
797 * multiple clients for the file.
798 * There is also a consistency problem for Version 2 of the protocol w.r.t.
799 * not being able to tell if other clients are writing a file concurrently,
800 * since there is no way of knowing if the changed modify time in the reply
801 * is only due to the write for this client.
802 * (NFS Version 3 provides weak cache consistency data in the reply that
803 * should be sufficient to detect and handle this case.)
804 *
805 * The current code does the following:
806 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
807 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
808 * or commit them (this satisfies 1 and 2 except for the
809 * case where the server crashes after this close but
810 * before the commit RPC, which is felt to be "good
811 * enough". Changing the last argument to ncl_flush() to
812 * a 1 would force a commit operation, if it is felt a
813 * commit is necessary now.
814 * for NFS Version 4 - flush the dirty buffers and commit them, if
815 * nfscl_mustflush() says this is necessary.
816 * It is necessary if there is no write delegation held,
817 * in order to satisfy open/close coherency.
818 * If the file isn't cached on local stable storage,
819 * it may be necessary in order to detect "out of space"
820 * errors from the server, if the write delegation
821 * issued by the server doesn't allow the file to grow.
822 */
823 /* ARGSUSED */
824 static int
825 nfs_close(struct vop_close_args *ap)
826 {
827 struct vnode *vp = ap->a_vp;
828 struct nfsnode *np = VTONFS(vp);
829 struct nfsvattr nfsva;
830 struct ucred *cred;
831 int error = 0, ret, localcred = 0;
832 int fmode = ap->a_fflag;
833
834 if (NFSCL_FORCEDISM(vp->v_mount))
835 return (0);
836 /*
837 * During shutdown, a_cred isn't valid, so just use root.
838 */
839 if (ap->a_cred == NOCRED) {
840 cred = newnfs_getcred();
841 localcred = 1;
842 } else {
843 cred = ap->a_cred;
844 }
845 if (vp->v_type == VREG) {
846 /*
847 * Examine and clean dirty pages, regardless of NMODIFIED.
848 * This closes a major hole in close-to-open consistency.
849 * We want to push out all dirty pages (and buffers) on
850 * close, regardless of whether they were dirtied by
851 * mmap'ed writes or via write().
852 */
853 if (nfs_clean_pages_on_close && vp->v_object) {
854 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
855 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
856 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK)
857 return (EBADF);
858 }
859 VM_OBJECT_WLOCK(vp->v_object);
860 vm_object_page_clean(vp->v_object, 0, 0, 0);
861 VM_OBJECT_WUNLOCK(vp->v_object);
862 }
863 NFSLOCKNODE(np);
864 if (np->n_flag & NMODIFIED) {
865 NFSUNLOCKNODE(np);
866 if (NFS_ISV3(vp)) {
867 /*
868 * Under NFSv3 we have dirty buffers to dispose of. We
869 * must flush them to the NFS server. We have the option
870 * of waiting all the way through the commit rpc or just
871 * waiting for the initial write. The default is to only
872 * wait through the initial write so the data is in the
873 * server's cache, which is roughly similar to the state
874 * a standard disk subsystem leaves the file in on close().
875 *
876 * We cannot clear the NMODIFIED bit in np->n_flag due to
877 * potential races with other processes, and certainly
878 * cannot clear it if we don't commit.
879 * These races occur when there is no longer the old
880 * traditional vnode locking implemented for Vnode Ops.
881 */
882 int cm = newnfs_commit_on_close ? 1 : 0;
883 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
884 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
885 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK)
886 return (EBADF);
887 }
888 error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
889 /* np->n_flag &= ~NMODIFIED; */
890 } else if (NFS_ISV4(vp)) {
891 if (nfscl_mustflush(vp) != 0) {
892 int cm = newnfs_commit_on_close ? 1 : 0;
893 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
894 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
895 if (VN_IS_DOOMED(vp) && ap->a_fflag !=
896 FNONBLOCK)
897 return (EBADF);
898 }
899 error = ncl_flush(vp, MNT_WAIT, ap->a_td,
900 cm, 0);
901 /*
902 * as above w.r.t races when clearing
903 * NMODIFIED.
904 * np->n_flag &= ~NMODIFIED;
905 */
906 }
907 } else {
908 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
909 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
910 if (VN_IS_DOOMED(vp) && ap->a_fflag !=
911 FNONBLOCK)
912 return (EBADF);
913 }
914 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
915 }
916 NFSLOCKNODE(np);
917 }
918 /*
919 * Invalidate the attribute cache in all cases.
920 * An open is going to fetch fresh attrs any way, other procs
921 * on this node that have file open will be forced to do an
922 * otw attr fetch, but this is safe.
923 * --> A user found that their RPC count dropped by 20% when
924 * this was commented out and I can't see any requirement
925 * for it, so I've disabled it when negative lookups are
926 * enabled. (What does this have to do with negative lookup
927 * caching? Well nothing, except it was reported by the
928 * same user that needed negative lookup caching and I wanted
929 * there to be a way to disable it to see if it
930 * is the cause of some caching/coherency issue that might
931 * crop up.)
932 */
933 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
934 np->n_attrstamp = 0;
935 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
936 }
937 if (np->n_flag & NWRITEERR) {
938 np->n_flag &= ~NWRITEERR;
939 error = np->n_error;
940 }
941 NFSUNLOCKNODE(np);
942 }
943
944 if (NFS_ISV4(vp)) {
945 /*
946 * Get attributes so "change" is up to date.
947 */
948 if (error == 0 && nfscl_mustflush(vp) != 0 &&
949 vp->v_type == VREG &&
950 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
951 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva);
952 if (!ret) {
953 np->n_change = nfsva.na_filerev;
954 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
955 0, 0);
956 }
957 }
958
959 /*
960 * and do the close.
961 */
962 ret = nfsrpc_close(vp, 0, ap->a_td);
963 if (!error && ret)
964 error = ret;
965 if (error)
966 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
967 (gid_t)0);
968 }
969 if (newnfs_directio_enable)
970 KASSERT((np->n_directio_asyncwr == 0),
971 ("nfs_close: dirty unflushed (%d) directio buffers\n",
972 np->n_directio_asyncwr));
973 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
974 NFSLOCKNODE(np);
975 KASSERT((np->n_directio_opens > 0),
976 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
977 np->n_directio_opens--;
978 if (np->n_directio_opens == 0)
979 np->n_flag &= ~NNONCACHE;
980 NFSUNLOCKNODE(np);
981 }
982 if (localcred)
983 NFSFREECRED(cred);
984 return (error);
985 }
986
987 /*
988 * nfs getattr call from vfs.
989 */
990 static int
991 nfs_getattr(struct vop_getattr_args *ap)
992 {
993 struct vnode *vp = ap->a_vp;
994 struct thread *td = curthread; /* XXX */
995 struct nfsnode *np = VTONFS(vp);
996 int error = 0;
997 struct nfsvattr nfsva;
998 struct vattr *vap = ap->a_vap;
999 struct vattr vattr;
1000
1001 /*
1002 * Update local times for special files.
1003 */
1004 NFSLOCKNODE(np);
1005 if (np->n_flag & (NACC | NUPD))
1006 np->n_flag |= NCHG;
1007 NFSUNLOCKNODE(np);
1008 /*
1009 * First look in the cache.
1010 */
1011 if (ncl_getattrcache(vp, &vattr) == 0) {
1012 ncl_copy_vattr(vap, &vattr);
1013
1014 /*
1015 * Get the local modify time for the case of a write
1016 * delegation.
1017 */
1018 nfscl_deleggetmodtime(vp, &vap->va_mtime);
1019 return (0);
1020 }
1021
1022 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
1023 nfsaccess_cache_timeout > 0) {
1024 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
1025 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
1026 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
1027 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
1028 return (0);
1029 }
1030 }
1031 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva);
1032 if (!error)
1033 error = nfscl_loadattrcache(&vp, &nfsva, vap, 0, 0);
1034 if (!error) {
1035 /*
1036 * Get the local modify time for the case of a write
1037 * delegation.
1038 */
1039 nfscl_deleggetmodtime(vp, &vap->va_mtime);
1040 } else if (NFS_ISV4(vp)) {
1041 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1042 }
1043 return (error);
1044 }
1045
1046 /*
1047 * nfs setattr call.
1048 */
1049 static int
1050 nfs_setattr(struct vop_setattr_args *ap)
1051 {
1052 struct vnode *vp = ap->a_vp;
1053 struct nfsnode *np = VTONFS(vp);
1054 struct thread *td = curthread; /* XXX */
1055 struct vattr *vap = ap->a_vap;
1056 int error = 0;
1057 u_quad_t tsize;
1058 struct timespec ts;
1059
1060 #ifndef nolint
1061 tsize = (u_quad_t)0;
1062 #endif
1063
1064 /*
1065 * Setting of flags and marking of atimes are not supported.
1066 */
1067 if (vap->va_flags != VNOVAL)
1068 return (EOPNOTSUPP);
1069
1070 /*
1071 * Disallow write attempts if the filesystem is mounted read-only.
1072 */
1073 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
1074 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
1075 vap->va_mtime.tv_sec != VNOVAL ||
1076 vap->va_birthtime.tv_sec != VNOVAL ||
1077 vap->va_mode != (mode_t)VNOVAL) &&
1078 (vp->v_mount->mnt_flag & MNT_RDONLY))
1079 return (EROFS);
1080 if (vap->va_size != VNOVAL) {
1081 switch (vp->v_type) {
1082 case VDIR:
1083 return (EISDIR);
1084 case VCHR:
1085 case VBLK:
1086 case VSOCK:
1087 case VFIFO:
1088 if (vap->va_mtime.tv_sec == VNOVAL &&
1089 vap->va_atime.tv_sec == VNOVAL &&
1090 vap->va_birthtime.tv_sec == VNOVAL &&
1091 vap->va_mode == (mode_t)VNOVAL &&
1092 vap->va_uid == (uid_t)VNOVAL &&
1093 vap->va_gid == (gid_t)VNOVAL)
1094 return (0);
1095 vap->va_size = VNOVAL;
1096 break;
1097 default:
1098 /*
1099 * Disallow write attempts if the filesystem is
1100 * mounted read-only.
1101 */
1102 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1103 return (EROFS);
1104 /*
1105 * We run vnode_pager_setsize() early (why?),
1106 * we must set np->n_size now to avoid vinvalbuf
1107 * V_SAVE races that might setsize a lower
1108 * value.
1109 */
1110 NFSLOCKNODE(np);
1111 tsize = np->n_size;
1112 NFSUNLOCKNODE(np);
1113 error = ncl_meta_setsize(vp, td, vap->va_size);
1114 NFSLOCKNODE(np);
1115 if (np->n_flag & NMODIFIED) {
1116 tsize = np->n_size;
1117 NFSUNLOCKNODE(np);
1118 error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
1119 0 : V_SAVE, td, 1);
1120 if (error != 0) {
1121 vnode_pager_setsize(vp, tsize);
1122 return (error);
1123 }
1124 /*
1125 * Call nfscl_delegmodtime() to set the modify time
1126 * locally, as required.
1127 */
1128 nfscl_delegmodtime(vp);
1129 } else
1130 NFSUNLOCKNODE(np);
1131 /*
1132 * np->n_size has already been set to vap->va_size
1133 * in ncl_meta_setsize(). We must set it again since
1134 * nfs_loadattrcache() could be called through
1135 * ncl_meta_setsize() and could modify np->n_size.
1136 */
1137 NFSLOCKNODE(np);
1138 np->n_vattr.na_size = np->n_size = vap->va_size;
1139 NFSUNLOCKNODE(np);
1140 }
1141 } else {
1142 NFSLOCKNODE(np);
1143 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
1144 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
1145 NFSUNLOCKNODE(np);
1146 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
1147 if (error == EINTR || error == EIO)
1148 return (error);
1149 } else
1150 NFSUNLOCKNODE(np);
1151 }
1152 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
1153 if (vap->va_size != VNOVAL) {
1154 if (error == 0) {
1155 nanouptime(&ts);
1156 NFSLOCKNODE(np);
1157 np->n_localmodtime = ts;
1158 NFSUNLOCKNODE(np);
1159 } else {
1160 NFSLOCKNODE(np);
1161 np->n_size = np->n_vattr.na_size = tsize;
1162 vnode_pager_setsize(vp, tsize);
1163 NFSUNLOCKNODE(np);
1164 }
1165 }
1166 return (error);
1167 }
1168
1169 /*
1170 * Do an nfs setattr rpc.
1171 */
1172 static int
1173 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1174 struct thread *td)
1175 {
1176 struct nfsnode *np = VTONFS(vp);
1177 int error, ret, attrflag, i;
1178 struct nfsvattr nfsva;
1179
1180 if (NFS_ISV34(vp)) {
1181 NFSLOCKNODE(np);
1182 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1183 np->n_accesscache[i].stamp = 0;
1184 np->n_flag |= NDELEGMOD;
1185 NFSUNLOCKNODE(np);
1186 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1187 }
1188 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag);
1189 if (attrflag) {
1190 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
1191 if (ret && !error)
1192 error = ret;
1193 }
1194 if (error && NFS_ISV4(vp))
1195 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1196 return (error);
1197 }
1198
1199 /*
1200 * nfs lookup call, one step at a time...
1201 * First look in cache
1202 * If not found, unlock the directory nfsnode and do the rpc
1203 */
1204 static int
1205 nfs_lookup(struct vop_lookup_args *ap)
1206 {
1207 struct componentname *cnp = ap->a_cnp;
1208 struct vnode *dvp = ap->a_dvp;
1209 struct vnode **vpp = ap->a_vpp;
1210 struct mount *mp = dvp->v_mount;
1211 int flags = cnp->cn_flags;
1212 struct vnode *newvp;
1213 struct nfsmount *nmp;
1214 struct nfsnode *np, *newnp;
1215 int error = 0, attrflag, dattrflag, ltype, ncticks;
1216 struct thread *td = curthread;
1217 struct nfsfh *nfhp;
1218 struct nfsvattr dnfsva, nfsva;
1219 struct vattr vattr;
1220 struct timespec nctime, ts;
1221 uint32_t openmode;
1222
1223 *vpp = NULLVP;
1224 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1225 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1226 return (EROFS);
1227 if (dvp->v_type != VDIR)
1228 return (ENOTDIR);
1229 nmp = VFSTONFS(mp);
1230 np = VTONFS(dvp);
1231
1232 /* For NFSv4, wait until any remove is done. */
1233 NFSLOCKNODE(np);
1234 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1235 np->n_flag |= NREMOVEWANT;
1236 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1237 }
1238 NFSUNLOCKNODE(np);
1239
1240 error = vn_dir_check_exec(dvp, cnp);
1241 if (error != 0)
1242 return (error);
1243 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1244 if (error > 0 && error != ENOENT)
1245 return (error);
1246 if (error == -1) {
1247 /*
1248 * Lookups of "." are special and always return the
1249 * current directory. cache_lookup() already handles
1250 * associated locking bookkeeping, etc.
1251 */
1252 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1253 return (0);
1254 }
1255
1256 /*
1257 * We only accept a positive hit in the cache if the
1258 * change time of the file matches our cached copy.
1259 * Otherwise, we discard the cache entry and fallback
1260 * to doing a lookup RPC. We also only trust cache
1261 * entries for less than nm_nametimeo seconds.
1262 *
1263 * To better handle stale file handles and attributes,
1264 * clear the attribute cache of this node if it is a
1265 * leaf component, part of an open() call, and not
1266 * locally modified before fetching the attributes.
1267 * This should allow stale file handles to be detected
1268 * here where we can fall back to a LOOKUP RPC to
1269 * recover rather than having nfs_open() detect the
1270 * stale file handle and failing open(2) with ESTALE.
1271 */
1272 newvp = *vpp;
1273 newnp = VTONFS(newvp);
1274 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1275 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1276 !(newnp->n_flag & NMODIFIED)) {
1277 NFSLOCKNODE(newnp);
1278 newnp->n_attrstamp = 0;
1279 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1280 NFSUNLOCKNODE(newnp);
1281 }
1282 if (nfscl_nodeleg(newvp, 0) == 0 ||
1283 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1284 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1285 timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1286 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1287 return (0);
1288 }
1289 cache_purge(newvp);
1290 if (dvp != newvp)
1291 vput(newvp);
1292 else
1293 vrele(newvp);
1294 *vpp = NULLVP;
1295 } else if (error == ENOENT) {
1296 if (VN_IS_DOOMED(dvp))
1297 return (ENOENT);
1298 /*
1299 * We only accept a negative hit in the cache if the
1300 * modification time of the parent directory matches
1301 * the cached copy in the name cache entry.
1302 * Otherwise, we discard all of the negative cache
1303 * entries for this directory. We also only trust
1304 * negative cache entries for up to nm_negnametimeo
1305 * seconds.
1306 */
1307 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1308 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1309 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1310 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1311 return (ENOENT);
1312 }
1313 cache_purge_negative(dvp);
1314 }
1315
1316 openmode = 0;
1317 /*
1318 * If this an NFSv4.1/4.2 mount using the "oneopenown" mount
1319 * option, it is possible to do the Open operation in the same
1320 * compound as Lookup, so long as delegations are not being
1321 * issued. This saves doing a separate RPC for Open.
1322 * For pnfs, do not do this, since the Open+LayoutGet will
1323 * be needed as a separate RPC.
1324 */
1325 NFSLOCKMNT(nmp);
1326 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && !NFSHASPNFS(nmp) &&
1327 (nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 &&
1328 (!NFSMNT_RDONLY(mp) || (flags & OPENWRITE) == 0) &&
1329 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN)) {
1330 if ((flags & OPENREAD) != 0)
1331 openmode |= NFSV4OPEN_ACCESSREAD;
1332 if ((flags & OPENWRITE) != 0)
1333 openmode |= NFSV4OPEN_ACCESSWRITE;
1334 }
1335 NFSUNLOCKMNT(nmp);
1336
1337 newvp = NULLVP;
1338 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1339 nanouptime(&ts);
1340 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1341 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1342 openmode);
1343 if (dattrflag)
1344 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
1345 if (error) {
1346 if (newvp != NULLVP) {
1347 vput(newvp);
1348 *vpp = NULLVP;
1349 }
1350
1351 if (error != ENOENT) {
1352 if (NFS_ISV4(dvp))
1353 error = nfscl_maperr(td, error, (uid_t)0,
1354 (gid_t)0);
1355 return (error);
1356 }
1357
1358 /* The requested file was not found. */
1359 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1360 (flags & ISLASTCN)) {
1361 /*
1362 * XXX: UFS does a full VOP_ACCESS(dvp,
1363 * VWRITE) here instead of just checking
1364 * MNT_RDONLY.
1365 */
1366 if (mp->mnt_flag & MNT_RDONLY)
1367 return (EROFS);
1368 return (EJUSTRETURN);
1369 }
1370
1371 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1372 /*
1373 * Cache the modification time of the parent
1374 * directory from the post-op attributes in
1375 * the name cache entry. The negative cache
1376 * entry will be ignored once the directory
1377 * has changed. Don't bother adding the entry
1378 * if the directory has already changed.
1379 */
1380 NFSLOCKNODE(np);
1381 if (timespeccmp(&np->n_vattr.na_mtime,
1382 &dnfsva.na_mtime, ==)) {
1383 NFSUNLOCKNODE(np);
1384 cache_enter_time(dvp, NULL, cnp,
1385 &dnfsva.na_mtime, NULL);
1386 } else
1387 NFSUNLOCKNODE(np);
1388 }
1389 return (ENOENT);
1390 }
1391
1392 /*
1393 * Handle RENAME case...
1394 */
1395 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1396 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1397 free(nfhp, M_NFSFH);
1398 return (EISDIR);
1399 }
1400 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, LK_EXCLUSIVE);
1401 if (error)
1402 return (error);
1403 newvp = NFSTOV(np);
1404 /*
1405 * If n_localmodtime >= time before RPC, then
1406 * a file modification operation, such as
1407 * VOP_SETATTR() of size, has occurred while
1408 * the Lookup RPC and acquisition of the vnode
1409 * happened. As such, the attributes might
1410 * be stale, with possibly an incorrect size.
1411 */
1412 NFSLOCKNODE(np);
1413 if (timespecisset(&np->n_localmodtime) &&
1414 timespeccmp(&np->n_localmodtime, &ts, >=)) {
1415 NFSCL_DEBUG(4, "nfs_lookup: rename localmod "
1416 "stale attributes\n");
1417 attrflag = 0;
1418 }
1419 NFSUNLOCKNODE(np);
1420 if (attrflag)
1421 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1422 *vpp = newvp;
1423 return (0);
1424 }
1425
1426 if (flags & ISDOTDOT) {
1427 ltype = NFSVOPISLOCKED(dvp);
1428 error = vfs_busy(mp, MBF_NOWAIT);
1429 if (error != 0) {
1430 vfs_ref(mp);
1431 NFSVOPUNLOCK(dvp);
1432 error = vfs_busy(mp, 0);
1433 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1434 vfs_rel(mp);
1435 if (error == 0 && VN_IS_DOOMED(dvp)) {
1436 vfs_unbusy(mp);
1437 error = ENOENT;
1438 }
1439 if (error != 0)
1440 return (error);
1441 }
1442 NFSVOPUNLOCK(dvp);
1443 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np,
1444 cnp->cn_lkflags);
1445 if (error == 0)
1446 newvp = NFSTOV(np);
1447 vfs_unbusy(mp);
1448 if (newvp != dvp)
1449 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1450 if (VN_IS_DOOMED(dvp)) {
1451 if (error == 0) {
1452 if (newvp == dvp)
1453 vrele(newvp);
1454 else
1455 vput(newvp);
1456 }
1457 error = ENOENT;
1458 }
1459 if (error != 0)
1460 return (error);
1461 if (attrflag)
1462 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1463 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1464 free(nfhp, M_NFSFH);
1465 VREF(dvp);
1466 newvp = dvp;
1467 if (attrflag)
1468 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1469 } else {
1470 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np,
1471 cnp->cn_lkflags);
1472 if (error)
1473 return (error);
1474 newvp = NFSTOV(np);
1475 /*
1476 * If n_localmodtime >= time before RPC, then
1477 * a file modification operation, such as
1478 * VOP_SETATTR() of size, has occurred while
1479 * the Lookup RPC and acquisition of the vnode
1480 * happened. As such, the attributes might
1481 * be stale, with possibly an incorrect size.
1482 */
1483 NFSLOCKNODE(np);
1484 if (timespecisset(&np->n_localmodtime) &&
1485 timespeccmp(&np->n_localmodtime, &ts, >=)) {
1486 NFSCL_DEBUG(4, "nfs_lookup: localmod "
1487 "stale attributes\n");
1488 attrflag = 0;
1489 }
1490 NFSUNLOCKNODE(np);
1491 if (attrflag)
1492 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1493 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1494 !(np->n_flag & NMODIFIED)) {
1495 /*
1496 * Flush the attribute cache when opening a
1497 * leaf node to ensure that fresh attributes
1498 * are fetched in nfs_open() since we did not
1499 * fetch attributes from the LOOKUP reply.
1500 */
1501 NFSLOCKNODE(np);
1502 np->n_attrstamp = 0;
1503 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1504 NFSUNLOCKNODE(np);
1505 }
1506 }
1507 if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp &&
1508 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1509 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1510 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1511 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1512 *vpp = newvp;
1513 return (0);
1514 }
1515
1516 /*
1517 * nfs read call.
1518 * Just call ncl_bioread() to do the work.
1519 */
1520 static int
1521 nfs_read(struct vop_read_args *ap)
1522 {
1523 struct vnode *vp = ap->a_vp;
1524
1525 switch (vp->v_type) {
1526 case VREG:
1527 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1528 case VDIR:
1529 return (EISDIR);
1530 default:
1531 return (EOPNOTSUPP);
1532 }
1533 }
1534
1535 /*
1536 * nfs readlink call
1537 */
1538 static int
1539 nfs_readlink(struct vop_readlink_args *ap)
1540 {
1541 struct vnode *vp = ap->a_vp;
1542
1543 if (vp->v_type != VLNK)
1544 return (EINVAL);
1545 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1546 }
1547
1548 /*
1549 * Do a readlink rpc.
1550 * Called by ncl_doio() from below the buffer cache.
1551 */
1552 int
1553 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1554 {
1555 int error, ret, attrflag;
1556 struct nfsvattr nfsva;
1557
1558 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1559 &attrflag);
1560 if (attrflag) {
1561 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
1562 if (ret && !error)
1563 error = ret;
1564 }
1565 if (error && NFS_ISV4(vp))
1566 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1567 return (error);
1568 }
1569
1570 /*
1571 * nfs read rpc call
1572 * Ditto above
1573 */
1574 int
1575 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1576 {
1577 int error, ret, attrflag;
1578 struct nfsvattr nfsva;
1579 struct nfsmount *nmp;
1580
1581 nmp = VFSTONFS(vp->v_mount);
1582 error = EIO;
1583 attrflag = 0;
1584 if (NFSHASPNFS(nmp))
1585 error = nfscl_doiods(vp, uiop, NULL, NULL,
1586 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1587 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1588 if (error != 0)
1589 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1590 &attrflag);
1591 if (attrflag) {
1592 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
1593 if (ret && !error)
1594 error = ret;
1595 }
1596 if (error && NFS_ISV4(vp))
1597 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1598 return (error);
1599 }
1600
1601 /*
1602 * nfs write call
1603 */
1604 int
1605 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1606 int *iomode, int *must_commit, int called_from_strategy, int ioflag)
1607 {
1608 struct nfsvattr nfsva;
1609 int error, attrflag, ret;
1610 struct nfsmount *nmp;
1611
1612 nmp = VFSTONFS(vp->v_mount);
1613 error = EIO;
1614 attrflag = 0;
1615 if (NFSHASPNFS(nmp))
1616 error = nfscl_doiods(vp, uiop, iomode, must_commit,
1617 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1618 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1619 if (error != 0)
1620 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1621 uiop->uio_td, &nfsva, &attrflag, called_from_strategy,
1622 ioflag);
1623 if (attrflag) {
1624 if (VTONFS(vp)->n_flag & ND_NFSV4)
1625 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 1, 1);
1626 else
1627 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
1628 if (ret && !error)
1629 error = ret;
1630 }
1631 if (DOINGASYNC(vp))
1632 *iomode = NFSWRITE_FILESYNC;
1633 if (error && NFS_ISV4(vp))
1634 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1635 return (error);
1636 }
1637
1638 /*
1639 * nfs mknod rpc
1640 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1641 * mode set to specify the file type and the size field for rdev.
1642 */
1643 static int
1644 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1645 struct vattr *vap)
1646 {
1647 struct nfsvattr nfsva, dnfsva;
1648 struct vnode *newvp = NULL;
1649 struct nfsnode *np = NULL, *dnp;
1650 struct nfsfh *nfhp;
1651 struct vattr vattr;
1652 int error = 0, attrflag, dattrflag;
1653 u_int32_t rdev;
1654
1655 if (vap->va_type == VCHR || vap->va_type == VBLK)
1656 rdev = vap->va_rdev;
1657 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1658 rdev = 0xffffffff;
1659 else
1660 return (EOPNOTSUPP);
1661 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1662 return (error);
1663 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1664 rdev, vap->va_type, cnp->cn_cred, curthread, &dnfsva,
1665 &nfsva, &nfhp, &attrflag, &dattrflag);
1666 if (!error) {
1667 if (!nfhp)
1668 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1669 cnp->cn_namelen, cnp->cn_cred, curthread,
1670 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0);
1671 if (nfhp)
1672 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1673 curthread, &np, LK_EXCLUSIVE);
1674 }
1675 if (dattrflag)
1676 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
1677 if (!error) {
1678 newvp = NFSTOV(np);
1679 if (attrflag != 0) {
1680 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1681 if (error != 0)
1682 vput(newvp);
1683 }
1684 }
1685 if (!error) {
1686 *vpp = newvp;
1687 } else if (NFS_ISV4(dvp)) {
1688 error = nfscl_maperr(curthread, error, vap->va_uid,
1689 vap->va_gid);
1690 }
1691 dnp = VTONFS(dvp);
1692 NFSLOCKNODE(dnp);
1693 dnp->n_flag |= NMODIFIED;
1694 if (!dattrflag) {
1695 dnp->n_attrstamp = 0;
1696 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1697 }
1698 NFSUNLOCKNODE(dnp);
1699 return (error);
1700 }
1701
1702 /*
1703 * nfs mknod vop
1704 * just call nfs_mknodrpc() to do the work.
1705 */
1706 /* ARGSUSED */
1707 static int
1708 nfs_mknod(struct vop_mknod_args *ap)
1709 {
1710 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1711 }
1712
1713 static struct mtx nfs_cverf_mtx;
1714 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1715 MTX_DEF);
1716
1717 static nfsquad_t
1718 nfs_get_cverf(void)
1719 {
1720 static nfsquad_t cverf;
1721 nfsquad_t ret;
1722 static int cverf_initialized = 0;
1723
1724 mtx_lock(&nfs_cverf_mtx);
1725 if (cverf_initialized == 0) {
1726 cverf.lval[0] = arc4random();
1727 cverf.lval[1] = arc4random();
1728 cverf_initialized = 1;
1729 } else
1730 cverf.qval++;
1731 ret = cverf;
1732 mtx_unlock(&nfs_cverf_mtx);
1733
1734 return (ret);
1735 }
1736
1737 /*
1738 * nfs file create call
1739 */
1740 static int
1741 nfs_create(struct vop_create_args *ap)
1742 {
1743 struct vnode *dvp = ap->a_dvp;
1744 struct vattr *vap = ap->a_vap;
1745 struct componentname *cnp = ap->a_cnp;
1746 struct nfsnode *np = NULL, *dnp;
1747 struct vnode *newvp = NULL;
1748 struct nfsmount *nmp;
1749 struct nfsvattr dnfsva, nfsva;
1750 struct nfsfh *nfhp;
1751 nfsquad_t cverf;
1752 int error = 0, attrflag, dattrflag, fmode = 0;
1753 struct vattr vattr;
1754
1755 /*
1756 * Oops, not for me..
1757 */
1758 if (vap->va_type == VSOCK)
1759 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1760
1761 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1762 return (error);
1763 if (vap->va_vaflags & VA_EXCLUSIVE)
1764 fmode |= O_EXCL;
1765 dnp = VTONFS(dvp);
1766 nmp = VFSTONFS(dvp->v_mount);
1767 again:
1768 /* For NFSv4, wait until any remove is done. */
1769 NFSLOCKNODE(dnp);
1770 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1771 dnp->n_flag |= NREMOVEWANT;
1772 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1773 }
1774 NFSUNLOCKNODE(dnp);
1775
1776 cverf = nfs_get_cverf();
1777 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1778 vap, cverf, fmode, cnp->cn_cred, curthread, &dnfsva, &nfsva,
1779 &nfhp, &attrflag, &dattrflag);
1780 if (!error) {
1781 if (nfhp == NULL)
1782 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1783 cnp->cn_namelen, cnp->cn_cred, curthread,
1784 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0);
1785 if (nfhp != NULL)
1786 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1787 curthread, &np, LK_EXCLUSIVE);
1788 }
1789 if (dattrflag)
1790 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
1791 if (!error) {
1792 newvp = NFSTOV(np);
1793 if (attrflag == 0)
1794 error = nfsrpc_getattr(newvp, cnp->cn_cred, curthread,
1795 &nfsva);
1796 if (error == 0)
1797 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1798 }
1799 if (error) {
1800 if (newvp != NULL) {
1801 vput(newvp);
1802 newvp = NULL;
1803 }
1804 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1805 error == NFSERR_NOTSUPP) {
1806 fmode &= ~O_EXCL;
1807 goto again;
1808 }
1809 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1810 if (nfscl_checksattr(vap, &nfsva)) {
1811 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1812 curthread, &nfsva, &attrflag);
1813 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1814 vap->va_gid != (gid_t)VNOVAL)) {
1815 /* try again without setting uid/gid */
1816 vap->va_uid = (uid_t)VNOVAL;
1817 vap->va_gid = (uid_t)VNOVAL;
1818 error = nfsrpc_setattr(newvp, vap, NULL,
1819 cnp->cn_cred, curthread, &nfsva, &attrflag);
1820 }
1821 if (attrflag)
1822 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1823 0, 1);
1824 if (error != 0)
1825 vput(newvp);
1826 }
1827 }
1828 if (!error) {
1829 if ((cnp->cn_flags & MAKEENTRY) && attrflag) {
1830 if (dvp != newvp)
1831 cache_enter_time(dvp, newvp, cnp,
1832 &nfsva.na_ctime, NULL);
1833 else
1834 printf("nfs_create: bogus NFS server returned "
1835 "the directory as the new file object\n");
1836 }
1837 *ap->a_vpp = newvp;
1838 } else if (NFS_ISV4(dvp)) {
1839 error = nfscl_maperr(curthread, error, vap->va_uid,
1840 vap->va_gid);
1841 }
1842 NFSLOCKNODE(dnp);
1843 dnp->n_flag |= NMODIFIED;
1844 if (!dattrflag) {
1845 dnp->n_attrstamp = 0;
1846 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1847 }
1848 NFSUNLOCKNODE(dnp);
1849 return (error);
1850 }
1851
1852 /*
1853 * nfs file remove call
1854 * To try and make nfs semantics closer to ufs semantics, a file that has
1855 * other processes using the vnode is renamed instead of removed and then
1856 * removed later on the last close.
1857 * - If v_usecount > 1
1858 * If a rename is not already in the works
1859 * call nfs_sillyrename() to set it up
1860 * else
1861 * do the remove rpc
1862 */
1863 static int
1864 nfs_remove(struct vop_remove_args *ap)
1865 {
1866 struct vnode *vp = ap->a_vp;
1867 struct vnode *dvp = ap->a_dvp;
1868 struct componentname *cnp = ap->a_cnp;
1869 struct nfsnode *np = VTONFS(vp);
1870 int error = 0;
1871 struct vattr vattr;
1872
1873 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1874 if (vp->v_type == VDIR)
1875 error = EPERM;
1876 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1877 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1878 vattr.va_nlink > 1)) {
1879 /*
1880 * Purge the name cache so that the chance of a lookup for
1881 * the name succeeding while the remove is in progress is
1882 * minimized. Without node locking it can still happen, such
1883 * that an I/O op returns ESTALE, but since you get this if
1884 * another host removes the file..
1885 */
1886 cache_purge(vp);
1887 /*
1888 * throw away biocache buffers, mainly to avoid
1889 * unnecessary delayed writes later.
1890 */
1891 error = ncl_vinvalbuf(vp, 0, curthread, 1);
1892 if (error != EINTR && error != EIO)
1893 /* Do the rpc */
1894 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1895 cnp->cn_namelen, cnp->cn_cred, curthread);
1896 /*
1897 * Kludge City: If the first reply to the remove rpc is lost..
1898 * the reply to the retransmitted request will be ENOENT
1899 * since the file was in fact removed
1900 * Therefore, we cheat and return success.
1901 */
1902 if (error == ENOENT)
1903 error = 0;
1904 } else if (!np->n_sillyrename)
1905 error = nfs_sillyrename(dvp, vp, cnp);
1906 NFSLOCKNODE(np);
1907 np->n_attrstamp = 0;
1908 NFSUNLOCKNODE(np);
1909 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1910 return (error);
1911 }
1912
1913 /*
1914 * nfs file remove rpc called from nfs_inactive
1915 */
1916 int
1917 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1918 {
1919 /*
1920 * Make sure that the directory vnode is still valid.
1921 * XXX we should lock sp->s_dvp here.
1922 */
1923 if (sp->s_dvp->v_type == VBAD)
1924 return (0);
1925 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1926 sp->s_cred, NULL));
1927 }
1928
1929 /*
1930 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1931 */
1932 static int
1933 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1934 int namelen, struct ucred *cred, struct thread *td)
1935 {
1936 struct nfsvattr dnfsva;
1937 struct nfsnode *dnp = VTONFS(dvp);
1938 int error = 0, dattrflag;
1939
1940 NFSLOCKNODE(dnp);
1941 dnp->n_flag |= NREMOVEINPROG;
1942 NFSUNLOCKNODE(dnp);
1943 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1944 &dattrflag);
1945 NFSLOCKNODE(dnp);
1946 if ((dnp->n_flag & NREMOVEWANT)) {
1947 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1948 NFSUNLOCKNODE(dnp);
1949 wakeup((caddr_t)dnp);
1950 } else {
1951 dnp->n_flag &= ~NREMOVEINPROG;
1952 NFSUNLOCKNODE(dnp);
1953 }
1954 if (dattrflag)
1955 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
1956 NFSLOCKNODE(dnp);
1957 dnp->n_flag |= NMODIFIED;
1958 if (!dattrflag) {
1959 dnp->n_attrstamp = 0;
1960 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1961 }
1962 NFSUNLOCKNODE(dnp);
1963 if (error && NFS_ISV4(dvp))
1964 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1965 return (error);
1966 }
1967
1968 /*
1969 * nfs file rename call
1970 */
1971 static int
1972 nfs_rename(struct vop_rename_args *ap)
1973 {
1974 struct vnode *fvp = ap->a_fvp;
1975 struct vnode *tvp = ap->a_tvp;
1976 struct vnode *fdvp = ap->a_fdvp;
1977 struct vnode *tdvp = ap->a_tdvp;
1978 struct componentname *tcnp = ap->a_tcnp;
1979 struct componentname *fcnp = ap->a_fcnp;
1980 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1981 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1982 struct nfsv4node *newv4 = NULL;
1983 int error;
1984
1985 /* Check for cross-device rename */
1986 if ((fvp->v_mount != tdvp->v_mount) ||
1987 (tvp && (fvp->v_mount != tvp->v_mount))) {
1988 error = EXDEV;
1989 goto out;
1990 }
1991
1992 if (fvp == tvp) {
1993 printf("nfs_rename: fvp == tvp (can't happen)\n");
1994 error = 0;
1995 goto out;
1996 }
1997 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1998 goto out;
1999
2000 /*
2001 * We have to flush B_DELWRI data prior to renaming
2002 * the file. If we don't, the delayed-write buffers
2003 * can be flushed out later after the file has gone stale
2004 * under NFSV3. NFSV2 does not have this problem because
2005 * ( as far as I can tell ) it flushes dirty buffers more
2006 * often.
2007 *
2008 * Skip the rename operation if the fsync fails, this can happen
2009 * due to the server's volume being full, when we pushed out data
2010 * that was written back to our cache earlier. Not checking for
2011 * this condition can result in potential (silent) data loss.
2012 */
2013 error = VOP_FSYNC(fvp, MNT_WAIT, curthread);
2014 NFSVOPUNLOCK(fvp);
2015 if (!error && tvp)
2016 error = VOP_FSYNC(tvp, MNT_WAIT, curthread);
2017 if (error)
2018 goto out;
2019
2020 /*
2021 * If the tvp exists and is in use, sillyrename it before doing the
2022 * rename of the new file over it.
2023 * XXX Can't sillyrename a directory.
2024 */
2025 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
2026 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
2027 vput(tvp);
2028 tvp = NULL;
2029 }
2030
2031 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
2032 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
2033 curthread);
2034
2035 if (error == 0 && NFS_ISV4(tdvp)) {
2036 /*
2037 * For NFSv4, check to see if it is the same name and
2038 * replace the name, if it is different.
2039 */
2040 newv4 = malloc(
2041 sizeof (struct nfsv4node) +
2042 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
2043 M_NFSV4NODE, M_WAITOK);
2044 NFSLOCKNODE(tdnp);
2045 NFSLOCKNODE(fnp);
2046 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
2047 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
2048 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
2049 tcnp->cn_namelen) ||
2050 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
2051 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
2052 tdnp->n_fhp->nfh_len))) {
2053 #ifdef notdef
2054 { char nnn[100]; int nnnl;
2055 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
2056 bcopy(tcnp->cn_nameptr, nnn, nnnl);
2057 nnn[nnnl] = '\0';
2058 printf("ren replace=%s\n",nnn);
2059 }
2060 #endif
2061 free(fnp->n_v4, M_NFSV4NODE);
2062 fnp->n_v4 = newv4;
2063 newv4 = NULL;
2064 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
2065 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
2066 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
2067 tdnp->n_fhp->nfh_len);
2068 NFSBCOPY(tcnp->cn_nameptr,
2069 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
2070 }
2071 NFSUNLOCKNODE(tdnp);
2072 NFSUNLOCKNODE(fnp);
2073 if (newv4 != NULL)
2074 free(newv4, M_NFSV4NODE);
2075 }
2076
2077 if (fvp->v_type == VDIR) {
2078 if (tvp != NULL && tvp->v_type == VDIR)
2079 cache_purge(tdvp);
2080 cache_purge(fdvp);
2081 }
2082
2083 out:
2084 if (tdvp == tvp)
2085 vrele(tdvp);
2086 else
2087 vput(tdvp);
2088 if (tvp)
2089 vput(tvp);
2090 vrele(fdvp);
2091 vrele(fvp);
2092 /*
2093 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2094 */
2095 if (error == ENOENT)
2096 error = 0;
2097 return (error);
2098 }
2099
2100 /*
2101 * nfs file rename rpc called from nfs_remove() above
2102 */
2103 static int
2104 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
2105 struct sillyrename *sp)
2106 {
2107
2108 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
2109 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
2110 curthread));
2111 }
2112
2113 /*
2114 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2115 */
2116 static int
2117 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
2118 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
2119 int tnamelen, struct ucred *cred, struct thread *td)
2120 {
2121 struct nfsvattr fnfsva, tnfsva;
2122 struct nfsnode *fdnp = VTONFS(fdvp);
2123 struct nfsnode *tdnp = VTONFS(tdvp);
2124 int error = 0, fattrflag, tattrflag;
2125
2126 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
2127 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
2128 &tattrflag);
2129 NFSLOCKNODE(fdnp);
2130 fdnp->n_flag |= NMODIFIED;
2131 if (fattrflag != 0) {
2132 NFSUNLOCKNODE(fdnp);
2133 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, 0, 1);
2134 } else {
2135 fdnp->n_attrstamp = 0;
2136 NFSUNLOCKNODE(fdnp);
2137 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
2138 }
2139 NFSLOCKNODE(tdnp);
2140 tdnp->n_flag |= NMODIFIED;
2141 if (tattrflag != 0) {
2142 NFSUNLOCKNODE(tdnp);
2143 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, 0, 1);
2144 } else {
2145 tdnp->n_attrstamp = 0;
2146 NFSUNLOCKNODE(tdnp);
2147 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2148 }
2149 if (error && NFS_ISV4(fdvp))
2150 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2151 return (error);
2152 }
2153
2154 /*
2155 * nfs hard link create call
2156 */
2157 static int
2158 nfs_link(struct vop_link_args *ap)
2159 {
2160 struct vnode *vp = ap->a_vp;
2161 struct vnode *tdvp = ap->a_tdvp;
2162 struct componentname *cnp = ap->a_cnp;
2163 struct nfsnode *np, *tdnp;
2164 struct nfsvattr nfsva, dnfsva;
2165 int error = 0, attrflag, dattrflag;
2166
2167 /*
2168 * Push all writes to the server, so that the attribute cache
2169 * doesn't get "out of sync" with the server.
2170 * XXX There should be a better way!
2171 */
2172 VOP_FSYNC(vp, MNT_WAIT, curthread);
2173
2174 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2175 cnp->cn_cred, curthread, &dnfsva, &nfsva, &attrflag, &dattrflag);
2176 tdnp = VTONFS(tdvp);
2177 NFSLOCKNODE(tdnp);
2178 tdnp->n_flag |= NMODIFIED;
2179 if (dattrflag != 0) {
2180 NFSUNLOCKNODE(tdnp);
2181 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, 0, 1);
2182 } else {
2183 tdnp->n_attrstamp = 0;
2184 NFSUNLOCKNODE(tdnp);
2185 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2186 }
2187 if (attrflag)
2188 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
2189 else {
2190 np = VTONFS(vp);
2191 NFSLOCKNODE(np);
2192 np->n_attrstamp = 0;
2193 NFSUNLOCKNODE(np);
2194 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2195 }
2196 /*
2197 * If negative lookup caching is enabled, I might as well
2198 * add an entry for this node. Not necessary for correctness,
2199 * but if negative caching is enabled, then the system
2200 * must care about lookup caching hit rate, so...
2201 */
2202 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2203 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2204 if (tdvp != vp)
2205 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2206 else
2207 printf("nfs_link: bogus NFS server returned "
2208 "the directory as the new link\n");
2209 }
2210 if (error && NFS_ISV4(vp))
2211 error = nfscl_maperr(curthread, error, (uid_t)0,
2212 (gid_t)0);
2213 return (error);
2214 }
2215
2216 /*
2217 * nfs symbolic link create call
2218 */
2219 static int
2220 nfs_symlink(struct vop_symlink_args *ap)
2221 {
2222 struct vnode *dvp = ap->a_dvp;
2223 struct vattr *vap = ap->a_vap;
2224 struct componentname *cnp = ap->a_cnp;
2225 struct nfsvattr nfsva, dnfsva;
2226 struct nfsfh *nfhp;
2227 struct nfsnode *np = NULL, *dnp;
2228 struct vnode *newvp = NULL;
2229 int error = 0, attrflag, dattrflag, ret;
2230
2231 vap->va_type = VLNK;
2232 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2233 ap->a_target, vap, cnp->cn_cred, curthread, &dnfsva,
2234 &nfsva, &nfhp, &attrflag, &dattrflag);
2235 if (nfhp) {
2236 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread,
2237 &np, LK_EXCLUSIVE);
2238 if (!ret)
2239 newvp = NFSTOV(np);
2240 else if (!error)
2241 error = ret;
2242 }
2243 if (newvp != NULL) {
2244 if (attrflag)
2245 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
2246 } else if (!error) {
2247 /*
2248 * If we do not have an error and we could not extract the
2249 * newvp from the response due to the request being NFSv2, we
2250 * have to do a lookup in order to obtain a newvp to return.
2251 */
2252 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2253 cnp->cn_cred, curthread, &np);
2254 if (!error)
2255 newvp = NFSTOV(np);
2256 }
2257 if (error) {
2258 if (newvp)
2259 vput(newvp);
2260 if (NFS_ISV4(dvp))
2261 error = nfscl_maperr(curthread, error,
2262 vap->va_uid, vap->va_gid);
2263 } else {
2264 *ap->a_vpp = newvp;
2265 }
2266
2267 dnp = VTONFS(dvp);
2268 NFSLOCKNODE(dnp);
2269 dnp->n_flag |= NMODIFIED;
2270 if (dattrflag != 0) {
2271 NFSUNLOCKNODE(dnp);
2272 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2273 } else {
2274 dnp->n_attrstamp = 0;
2275 NFSUNLOCKNODE(dnp);
2276 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2277 }
2278 /*
2279 * If negative lookup caching is enabled, I might as well
2280 * add an entry for this node. Not necessary for correctness,
2281 * but if negative caching is enabled, then the system
2282 * must care about lookup caching hit rate, so...
2283 */
2284 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2285 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2286 if (dvp != newvp)
2287 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2288 NULL);
2289 else
2290 printf("nfs_symlink: bogus NFS server returned "
2291 "the directory as the new file object\n");
2292 }
2293 return (error);
2294 }
2295
2296 /*
2297 * nfs make dir call
2298 */
2299 static int
2300 nfs_mkdir(struct vop_mkdir_args *ap)
2301 {
2302 struct vnode *dvp = ap->a_dvp;
2303 struct vattr *vap = ap->a_vap;
2304 struct componentname *cnp = ap->a_cnp;
2305 struct nfsnode *np = NULL, *dnp;
2306 struct vnode *newvp = NULL;
2307 struct vattr vattr;
2308 struct nfsfh *nfhp;
2309 struct nfsvattr nfsva, dnfsva;
2310 int error = 0, attrflag, dattrflag, ret;
2311
2312 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2313 return (error);
2314 vap->va_type = VDIR;
2315 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2316 vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp,
2317 &attrflag, &dattrflag);
2318 dnp = VTONFS(dvp);
2319 NFSLOCKNODE(dnp);
2320 dnp->n_flag |= NMODIFIED;
2321 if (dattrflag != 0) {
2322 NFSUNLOCKNODE(dnp);
2323 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2324 } else {
2325 dnp->n_attrstamp = 0;
2326 NFSUNLOCKNODE(dnp);
2327 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2328 }
2329 if (nfhp) {
2330 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread,
2331 &np, LK_EXCLUSIVE);
2332 if (!ret) {
2333 newvp = NFSTOV(np);
2334 if (attrflag)
2335 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2336 0, 1);
2337 } else if (!error)
2338 error = ret;
2339 }
2340 if (!error && newvp == NULL) {
2341 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2342 cnp->cn_cred, curthread, &np);
2343 if (!error) {
2344 newvp = NFSTOV(np);
2345 if (newvp->v_type != VDIR)
2346 error = EEXIST;
2347 }
2348 }
2349 if (error) {
2350 if (newvp)
2351 vput(newvp);
2352 if (NFS_ISV4(dvp))
2353 error = nfscl_maperr(curthread, error,
2354 vap->va_uid, vap->va_gid);
2355 } else {
2356 /*
2357 * If negative lookup caching is enabled, I might as well
2358 * add an entry for this node. Not necessary for correctness,
2359 * but if negative caching is enabled, then the system
2360 * must care about lookup caching hit rate, so...
2361 */
2362 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2363 (cnp->cn_flags & MAKEENTRY) &&
2364 attrflag != 0 && dattrflag != 0) {
2365 if (dvp != newvp)
2366 cache_enter_time(dvp, newvp, cnp,
2367 &nfsva.na_ctime, &dnfsva.na_ctime);
2368 else
2369 printf("nfs_mkdir: bogus NFS server returned "
2370 "the directory that the directory was "
2371 "created in as the new file object\n");
2372 }
2373 *ap->a_vpp = newvp;
2374 }
2375 return (error);
2376 }
2377
2378 /*
2379 * nfs remove directory call
2380 */
2381 static int
2382 nfs_rmdir(struct vop_rmdir_args *ap)
2383 {
2384 struct vnode *vp = ap->a_vp;
2385 struct vnode *dvp = ap->a_dvp;
2386 struct componentname *cnp = ap->a_cnp;
2387 struct nfsnode *dnp;
2388 struct nfsvattr dnfsva;
2389 int error, dattrflag;
2390
2391 if (dvp == vp)
2392 return (EINVAL);
2393 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2394 cnp->cn_cred, curthread, &dnfsva, &dattrflag);
2395 dnp = VTONFS(dvp);
2396 NFSLOCKNODE(dnp);
2397 dnp->n_flag |= NMODIFIED;
2398 if (dattrflag != 0) {
2399 NFSUNLOCKNODE(dnp);
2400 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2401 } else {
2402 dnp->n_attrstamp = 0;
2403 NFSUNLOCKNODE(dnp);
2404 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2405 }
2406
2407 cache_purge(dvp);
2408 cache_purge(vp);
2409 if (error && NFS_ISV4(dvp))
2410 error = nfscl_maperr(curthread, error, (uid_t)0,
2411 (gid_t)0);
2412 /*
2413 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2414 */
2415 if (error == ENOENT)
2416 error = 0;
2417 return (error);
2418 }
2419
2420 /*
2421 * nfs readdir call
2422 */
2423 static int
2424 nfs_readdir(struct vop_readdir_args *ap)
2425 {
2426 struct vnode *vp = ap->a_vp;
2427 struct nfsnode *np = VTONFS(vp);
2428 struct uio *uio = ap->a_uio;
2429 ssize_t tresid, left;
2430 int error = 0;
2431 struct vattr vattr;
2432
2433 if (ap->a_eofflag != NULL)
2434 *ap->a_eofflag = 0;
2435 if (vp->v_type != VDIR)
2436 return(EPERM);
2437
2438 /*
2439 * First, check for hit on the EOF offset cache
2440 */
2441 NFSLOCKNODE(np);
2442 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2443 (np->n_flag & NMODIFIED) == 0) {
2444 NFSUNLOCKNODE(np);
2445 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2446 NFSLOCKNODE(np);
2447 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2448 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2449 NFSUNLOCKNODE(np);
2450 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2451 if (ap->a_eofflag != NULL)
2452 *ap->a_eofflag = 1;
2453 return (0);
2454 } else
2455 NFSUNLOCKNODE(np);
2456 }
2457 } else
2458 NFSUNLOCKNODE(np);
2459
2460 /*
2461 * NFS always guarantees that directory entries don't straddle
2462 * DIRBLKSIZ boundaries. As such, we need to limit the size
2463 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2464 * directory entry.
2465 */
2466 left = uio->uio_resid % DIRBLKSIZ;
2467 if (left == uio->uio_resid)
2468 return (EINVAL);
2469 uio->uio_resid -= left;
2470
2471 /*
2472 * Call ncl_bioread() to do the real work.
2473 */
2474 tresid = uio->uio_resid;
2475 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2476
2477 if (!error && uio->uio_resid == tresid) {
2478 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2479 if (ap->a_eofflag != NULL)
2480 *ap->a_eofflag = 1;
2481 }
2482
2483 /* Add the partial DIRBLKSIZ (left) back in. */
2484 uio->uio_resid += left;
2485 return (error);
2486 }
2487
2488 /*
2489 * Readdir rpc call.
2490 * Called from below the buffer cache by ncl_doio().
2491 */
2492 int
2493 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2494 struct thread *td)
2495 {
2496 struct nfsvattr nfsva;
2497 nfsuint64 *cookiep, cookie;
2498 struct nfsnode *dnp = VTONFS(vp);
2499 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2500 int error = 0, eof, attrflag;
2501
2502 KASSERT(uiop->uio_iovcnt == 1 &&
2503 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2504 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2505 ("nfs readdirrpc bad uio"));
2506
2507 /*
2508 * If there is no cookie, assume directory was stale.
2509 */
2510 ncl_dircookie_lock(dnp);
2511 NFSUNLOCKNODE(dnp);
2512 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2513 if (cookiep) {
2514 cookie = *cookiep;
2515 ncl_dircookie_unlock(dnp);
2516 } else {
2517 ncl_dircookie_unlock(dnp);
2518 return (NFSERR_BAD_COOKIE);
2519 }
2520
2521 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2522 (void)ncl_fsinfo(nmp, vp, cred, td);
2523
2524 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2525 &attrflag, &eof);
2526 if (attrflag)
2527 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
2528
2529 if (!error) {
2530 /*
2531 * We are now either at the end of the directory or have filled
2532 * the block.
2533 */
2534 if (eof) {
2535 NFSLOCKNODE(dnp);
2536 dnp->n_direofoffset = uiop->uio_offset;
2537 NFSUNLOCKNODE(dnp);
2538 } else {
2539 if (uiop->uio_resid > 0)
2540 printf("EEK! readdirrpc resid > 0\n");
2541 ncl_dircookie_lock(dnp);
2542 NFSUNLOCKNODE(dnp);
2543 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2544 *cookiep = cookie;
2545 ncl_dircookie_unlock(dnp);
2546 }
2547 } else if (NFS_ISV4(vp)) {
2548 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2549 }
2550 return (error);
2551 }
2552
2553 /*
2554 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2555 */
2556 int
2557 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2558 struct thread *td)
2559 {
2560 struct nfsvattr nfsva;
2561 nfsuint64 *cookiep, cookie;
2562 struct nfsnode *dnp = VTONFS(vp);
2563 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2564 int error = 0, attrflag, eof;
2565
2566 KASSERT(uiop->uio_iovcnt == 1 &&
2567 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2568 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2569 ("nfs readdirplusrpc bad uio"));
2570
2571 /*
2572 * If there is no cookie, assume directory was stale.
2573 */
2574 ncl_dircookie_lock(dnp);
2575 NFSUNLOCKNODE(dnp);
2576 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2577 if (cookiep) {
2578 cookie = *cookiep;
2579 ncl_dircookie_unlock(dnp);
2580 } else {
2581 ncl_dircookie_unlock(dnp);
2582 return (NFSERR_BAD_COOKIE);
2583 }
2584
2585 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2586 (void)ncl_fsinfo(nmp, vp, cred, td);
2587 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2588 &attrflag, &eof);
2589 if (attrflag)
2590 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
2591
2592 if (!error) {
2593 /*
2594 * We are now either at end of the directory or have filled the
2595 * the block.
2596 */
2597 if (eof) {
2598 NFSLOCKNODE(dnp);
2599 dnp->n_direofoffset = uiop->uio_offset;
2600 NFSUNLOCKNODE(dnp);
2601 } else {
2602 if (uiop->uio_resid > 0)
2603 printf("EEK! readdirplusrpc resid > 0\n");
2604 ncl_dircookie_lock(dnp);
2605 NFSUNLOCKNODE(dnp);
2606 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2607 *cookiep = cookie;
2608 ncl_dircookie_unlock(dnp);
2609 }
2610 } else if (NFS_ISV4(vp)) {
2611 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2612 }
2613 return (error);
2614 }
2615
2616 /*
2617 * Silly rename. To make the NFS filesystem that is stateless look a little
2618 * more like the "ufs" a remove of an active vnode is translated to a rename
2619 * to a funny looking filename that is removed by nfs_inactive on the
2620 * nfsnode. There is the potential for another process on a different client
2621 * to create the same funny name between the nfs_lookitup() fails and the
2622 * nfs_rename() completes, but...
2623 */
2624 static int
2625 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2626 {
2627 struct sillyrename *sp;
2628 struct nfsnode *np;
2629 int error;
2630 short pid;
2631 unsigned int lticks;
2632
2633 cache_purge(dvp);
2634 np = VTONFS(vp);
2635 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2636 sp = malloc(sizeof (struct sillyrename),
2637 M_NEWNFSREQ, M_WAITOK);
2638 sp->s_cred = crhold(cnp->cn_cred);
2639 sp->s_dvp = dvp;
2640 VREF(dvp);
2641
2642 /*
2643 * Fudge together a funny name.
2644 * Changing the format of the funny name to accommodate more
2645 * sillynames per directory.
2646 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2647 * CPU ticks since boot.
2648 */
2649 pid = curthread->td_proc->p_pid;
2650 lticks = (unsigned int)ticks;
2651 for ( ; ; ) {
2652 sp->s_namlen = sprintf(sp->s_name,
2653 ".nfs.%08x.%04x4.4", lticks,
2654 pid);
2655 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2656 curthread, NULL))
2657 break;
2658 lticks++;
2659 }
2660 error = nfs_renameit(dvp, vp, cnp, sp);
2661 if (error)
2662 goto bad;
2663 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2664 curthread, &np);
2665 np->n_sillyrename = sp;
2666 return (0);
2667 bad:
2668 vrele(sp->s_dvp);
2669 crfree(sp->s_cred);
2670 free(sp, M_NEWNFSREQ);
2671 return (error);
2672 }
2673
2674 /*
2675 * Look up a file name and optionally either update the file handle or
2676 * allocate an nfsnode, depending on the value of npp.
2677 * npp == NULL --> just do the lookup
2678 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2679 * handled too
2680 * *npp != NULL --> update the file handle in the vnode
2681 */
2682 static int
2683 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2684 struct thread *td, struct nfsnode **npp)
2685 {
2686 struct vnode *newvp = NULL, *vp;
2687 struct nfsnode *np, *dnp = VTONFS(dvp);
2688 struct nfsfh *nfhp, *onfhp;
2689 struct nfsvattr nfsva, dnfsva;
2690 struct componentname cn;
2691 int error = 0, attrflag, dattrflag;
2692 u_int hash;
2693 struct timespec ts;
2694
2695 nanouptime(&ts);
2696 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2697 &nfhp, &attrflag, &dattrflag, 0);
2698 if (dattrflag)
2699 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2700 if (npp && !error) {
2701 if (*npp != NULL) {
2702 np = *npp;
2703 vp = NFSTOV(np);
2704 /*
2705 * For NFSv4, check to see if it is the same name and
2706 * replace the name, if it is different.
2707 */
2708 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2709 (np->n_v4->n4_namelen != len ||
2710 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2711 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2712 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2713 dnp->n_fhp->nfh_len))) {
2714 #ifdef notdef
2715 { char nnn[100]; int nnnl;
2716 nnnl = (len < 100) ? len : 99;
2717 bcopy(name, nnn, nnnl);
2718 nnn[nnnl] = '\0';
2719 printf("replace=%s\n",nnn);
2720 }
2721 #endif
2722 free(np->n_v4, M_NFSV4NODE);
2723 np->n_v4 = malloc(
2724 sizeof (struct nfsv4node) +
2725 dnp->n_fhp->nfh_len + len - 1,
2726 M_NFSV4NODE, M_WAITOK);
2727 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2728 np->n_v4->n4_namelen = len;
2729 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2730 dnp->n_fhp->nfh_len);
2731 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2732 }
2733 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2734 FNV1_32_INIT);
2735 onfhp = np->n_fhp;
2736 /*
2737 * Rehash node for new file handle.
2738 */
2739 vfs_hash_rehash(vp, hash);
2740 np->n_fhp = nfhp;
2741 if (onfhp != NULL)
2742 free(onfhp, M_NFSFH);
2743 newvp = NFSTOV(np);
2744 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2745 free(nfhp, M_NFSFH);
2746 VREF(dvp);
2747 newvp = dvp;
2748 } else {
2749 cn.cn_nameptr = name;
2750 cn.cn_namelen = len;
2751 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2752 &np, LK_EXCLUSIVE);
2753 if (error)
2754 return (error);
2755 newvp = NFSTOV(np);
2756 /*
2757 * If n_localmodtime >= time before RPC, then
2758 * a file modification operation, such as
2759 * VOP_SETATTR() of size, has occurred while
2760 * the Lookup RPC and acquisition of the vnode
2761 * happened. As such, the attributes might
2762 * be stale, with possibly an incorrect size.
2763 */
2764 NFSLOCKNODE(np);
2765 if (timespecisset(&np->n_localmodtime) &&
2766 timespeccmp(&np->n_localmodtime, &ts, >=)) {
2767 NFSCL_DEBUG(4, "nfs_lookitup: localmod "
2768 "stale attributes\n");
2769 attrflag = 0;
2770 }
2771 NFSUNLOCKNODE(np);
2772 }
2773 if (!attrflag && *npp == NULL) {
2774 if (newvp == dvp)
2775 vrele(newvp);
2776 else
2777 vput(newvp);
2778 return (ENOENT);
2779 }
2780 if (attrflag)
2781 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
2782 }
2783 if (npp && *npp == NULL) {
2784 if (error) {
2785 if (newvp) {
2786 if (newvp == dvp)
2787 vrele(newvp);
2788 else
2789 vput(newvp);
2790 }
2791 } else
2792 *npp = np;
2793 }
2794 if (error && NFS_ISV4(dvp))
2795 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2796 return (error);
2797 }
2798
2799 /*
2800 * Nfs Version 3 and 4 commit rpc
2801 */
2802 int
2803 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2804 struct thread *td)
2805 {
2806 struct nfsvattr nfsva;
2807 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2808 struct nfsnode *np;
2809 struct uio uio;
2810 int error, attrflag;
2811
2812 np = VTONFS(vp);
2813 error = EIO;
2814 attrflag = 0;
2815 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2816 uio.uio_offset = offset;
2817 uio.uio_resid = cnt;
2818 error = nfscl_doiods(vp, &uio, NULL, NULL,
2819 NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2820 if (error != 0) {
2821 NFSLOCKNODE(np);
2822 np->n_flag &= ~NDSCOMMIT;
2823 NFSUNLOCKNODE(np);
2824 }
2825 }
2826 if (error != 0) {
2827 mtx_lock(&nmp->nm_mtx);
2828 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2829 mtx_unlock(&nmp->nm_mtx);
2830 return (0);
2831 }
2832 mtx_unlock(&nmp->nm_mtx);
2833 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2834 &attrflag);
2835 }
2836 if (attrflag != 0)
2837 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
2838 if (error != 0 && NFS_ISV4(vp))
2839 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2840 return (error);
2841 }
2842
2843 /*
2844 * Strategy routine.
2845 * For async requests when nfsiod(s) are running, queue the request by
2846 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2847 * request.
2848 */
2849 static int
2850 nfs_strategy(struct vop_strategy_args *ap)
2851 {
2852 struct buf *bp;
2853 struct vnode *vp;
2854 struct ucred *cr;
2855
2856 bp = ap->a_bp;
2857 vp = ap->a_vp;
2858 KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2859 KASSERT(!(bp->b_flags & B_DONE),
2860 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2861
2862 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2863 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2864 DEV_BSIZE);
2865 if (bp->b_iocmd == BIO_READ)
2866 cr = bp->b_rcred;
2867 else
2868 cr = bp->b_wcred;
2869
2870 /*
2871 * If the op is asynchronous and an i/o daemon is waiting
2872 * queue the request, wake it up and wait for completion
2873 * otherwise just do it ourselves.
2874 */
2875 if ((bp->b_flags & B_ASYNC) == 0 ||
2876 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2877 (void) ncl_doio(vp, bp, cr, curthread, 1);
2878 return (0);
2879 }
2880
2881 /*
2882 * fsync vnode op. Just call ncl_flush() with commit == 1.
2883 */
2884 /* ARGSUSED */
2885 static int
2886 nfs_fsync(struct vop_fsync_args *ap)
2887 {
2888
2889 if (ap->a_vp->v_type != VREG) {
2890 /*
2891 * For NFS, metadata is changed synchronously on the server,
2892 * so there is nothing to flush. Also, ncl_flush() clears
2893 * the NMODIFIED flag and that shouldn't be done here for
2894 * directories.
2895 */
2896 return (0);
2897 }
2898 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2899 }
2900
2901 /*
2902 * Flush all the blocks associated with a vnode.
2903 * Walk through the buffer pool and push any dirty pages
2904 * associated with the vnode.
2905 * If the called_from_renewthread argument is TRUE, it has been called
2906 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2907 * waiting for a buffer write to complete.
2908 */
2909 int
2910 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2911 int commit, int called_from_renewthread)
2912 {
2913 struct nfsnode *np = VTONFS(vp);
2914 struct buf *bp;
2915 int i;
2916 struct buf *nbp;
2917 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2918 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2919 int passone = 1, trycnt = 0;
2920 u_quad_t off, endoff, toff;
2921 struct ucred* wcred = NULL;
2922 struct buf **bvec = NULL;
2923 struct bufobj *bo;
2924 #ifndef NFS_COMMITBVECSIZ
2925 #define NFS_COMMITBVECSIZ 20
2926 #endif
2927 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2928 u_int bvecsize = 0, bveccount;
2929 struct timespec ts;
2930
2931 if (called_from_renewthread != 0)
2932 slptimeo = hz;
2933 if (nmp->nm_flag & NFSMNT_INT)
2934 slpflag = PCATCH;
2935 if (!commit)
2936 passone = 0;
2937 bo = &vp->v_bufobj;
2938 /*
2939 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2940 * server, but has not been committed to stable storage on the server
2941 * yet. On the first pass, the byte range is worked out and the commit
2942 * rpc is done. On the second pass, ncl_writebp() is called to do the
2943 * job.
2944 */
2945 again:
2946 off = (u_quad_t)-1;
2947 endoff = 0;
2948 bvecpos = 0;
2949 if (NFS_ISV34(vp) && commit) {
2950 if (bvec != NULL && bvec != bvec_on_stack)
2951 free(bvec, M_TEMP);
2952 /*
2953 * Count up how many buffers waiting for a commit.
2954 */
2955 bveccount = 0;
2956 BO_LOCK(bo);
2957 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2958 if (!BUF_ISLOCKED(bp) &&
2959 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2960 == (B_DELWRI | B_NEEDCOMMIT))
2961 bveccount++;
2962 }
2963 /*
2964 * Allocate space to remember the list of bufs to commit. It is
2965 * important to use M_NOWAIT here to avoid a race with nfs_write.
2966 * If we can't get memory (for whatever reason), we will end up
2967 * committing the buffers one-by-one in the loop below.
2968 */
2969 if (bveccount > NFS_COMMITBVECSIZ) {
2970 /*
2971 * Release the vnode interlock to avoid a lock
2972 * order reversal.
2973 */
2974 BO_UNLOCK(bo);
2975 bvec = (struct buf **)
2976 malloc(bveccount * sizeof(struct buf *),
2977 M_TEMP, M_NOWAIT);
2978 BO_LOCK(bo);
2979 if (bvec == NULL) {
2980 bvec = bvec_on_stack;
2981 bvecsize = NFS_COMMITBVECSIZ;
2982 } else
2983 bvecsize = bveccount;
2984 } else {
2985 bvec = bvec_on_stack;
2986 bvecsize = NFS_COMMITBVECSIZ;
2987 }
2988 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2989 if (bvecpos >= bvecsize)
2990 break;
2991 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2992 nbp = TAILQ_NEXT(bp, b_bobufs);
2993 continue;
2994 }
2995 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2996 (B_DELWRI | B_NEEDCOMMIT)) {
2997 BUF_UNLOCK(bp);
2998 nbp = TAILQ_NEXT(bp, b_bobufs);
2999 continue;
3000 }
3001 BO_UNLOCK(bo);
3002 bremfree(bp);
3003 /*
3004 * Work out if all buffers are using the same cred
3005 * so we can deal with them all with one commit.
3006 *
3007 * NOTE: we are not clearing B_DONE here, so we have
3008 * to do it later on in this routine if we intend to
3009 * initiate I/O on the bp.
3010 *
3011 * Note: to avoid loopback deadlocks, we do not
3012 * assign b_runningbufspace.
3013 */
3014 if (wcred == NULL)
3015 wcred = bp->b_wcred;
3016 else if (wcred != bp->b_wcred)
3017 wcred = NOCRED;
3018 vfs_busy_pages(bp, 1);
3019
3020 BO_LOCK(bo);
3021 /*
3022 * bp is protected by being locked, but nbp is not
3023 * and vfs_busy_pages() may sleep. We have to
3024 * recalculate nbp.
3025 */
3026 nbp = TAILQ_NEXT(bp, b_bobufs);
3027
3028 /*
3029 * A list of these buffers is kept so that the
3030 * second loop knows which buffers have actually
3031 * been committed. This is necessary, since there
3032 * may be a race between the commit rpc and new
3033 * uncommitted writes on the file.
3034 */
3035 bvec[bvecpos++] = bp;
3036 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3037 bp->b_dirtyoff;
3038 if (toff < off)
3039 off = toff;
3040 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3041 if (toff > endoff)
3042 endoff = toff;
3043 }
3044 BO_UNLOCK(bo);
3045 }
3046 if (bvecpos > 0) {
3047 /*
3048 * Commit data on the server, as required.
3049 * If all bufs are using the same wcred, then use that with
3050 * one call for all of them, otherwise commit each one
3051 * separately.
3052 */
3053 if (wcred != NOCRED)
3054 retv = ncl_commit(vp, off, (int)(endoff - off),
3055 wcred, td);
3056 else {
3057 retv = 0;
3058 for (i = 0; i < bvecpos; i++) {
3059 off_t off, size;
3060 bp = bvec[i];
3061 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3062 bp->b_dirtyoff;
3063 size = (u_quad_t)(bp->b_dirtyend
3064 - bp->b_dirtyoff);
3065 retv = ncl_commit(vp, off, (int)size,
3066 bp->b_wcred, td);
3067 if (retv) break;
3068 }
3069 }
3070
3071 if (retv == NFSERR_STALEWRITEVERF)
3072 ncl_clearcommit(vp->v_mount);
3073
3074 /*
3075 * Now, either mark the blocks I/O done or mark the
3076 * blocks dirty, depending on whether the commit
3077 * succeeded.
3078 */
3079 for (i = 0; i < bvecpos; i++) {
3080 bp = bvec[i];
3081 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3082 if (!NFSCL_FORCEDISM(vp->v_mount) && retv) {
3083 /*
3084 * Error, leave B_DELWRI intact
3085 */
3086 vfs_unbusy_pages(bp);
3087 brelse(bp);
3088 } else {
3089 /*
3090 * Success, remove B_DELWRI ( bundirty() ).
3091 *
3092 * b_dirtyoff/b_dirtyend seem to be NFS
3093 * specific. We should probably move that
3094 * into bundirty(). XXX
3095 */
3096 bufobj_wref(bo);
3097 bp->b_flags |= B_ASYNC;
3098 bundirty(bp);
3099 bp->b_flags &= ~B_DONE;
3100 bp->b_ioflags &= ~BIO_ERROR;
3101 bp->b_dirtyoff = bp->b_dirtyend = 0;
3102 bufdone(bp);
3103 }
3104 }
3105 }
3106
3107 /*
3108 * Start/do any write(s) that are required.
3109 */
3110 loop:
3111 BO_LOCK(bo);
3112 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3113 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3114 if (waitfor != MNT_WAIT || passone)
3115 continue;
3116
3117 error = BUF_TIMELOCK(bp,
3118 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3119 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
3120 if (error == 0) {
3121 BUF_UNLOCK(bp);
3122 goto loop;
3123 }
3124 if (error == ENOLCK) {
3125 error = 0;
3126 goto loop;
3127 }
3128 if (called_from_renewthread != 0) {
3129 /*
3130 * Return EIO so the flush will be retried
3131 * later.
3132 */
3133 error = EIO;
3134 goto done;
3135 }
3136 if (newnfs_sigintr(nmp, td)) {
3137 error = EINTR;
3138 goto done;
3139 }
3140 if (slpflag == PCATCH) {
3141 slpflag = 0;
3142 slptimeo = 2 * hz;
3143 }
3144 goto loop;
3145 }
3146 if ((bp->b_flags & B_DELWRI) == 0)
3147 panic("nfs_fsync: not dirty");
3148 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3149 BUF_UNLOCK(bp);
3150 continue;
3151 }
3152 BO_UNLOCK(bo);
3153 bremfree(bp);
3154 bp->b_flags |= B_ASYNC;
3155 bwrite(bp);
3156 if (newnfs_sigintr(nmp, td)) {
3157 error = EINTR;
3158 goto done;
3159 }
3160 goto loop;
3161 }
3162 if (passone) {
3163 passone = 0;
3164 BO_UNLOCK(bo);
3165 goto again;
3166 }
3167 if (waitfor == MNT_WAIT) {
3168 while (bo->bo_numoutput) {
3169 error = bufobj_wwait(bo, slpflag, slptimeo);
3170 if (error) {
3171 BO_UNLOCK(bo);
3172 if (called_from_renewthread != 0) {
3173 /*
3174 * Return EIO so that the flush will be
3175 * retried later.
3176 */
3177 error = EIO;
3178 goto done;
3179 }
3180 error = newnfs_sigintr(nmp, td);
3181 if (error)
3182 goto done;
3183 if (slpflag == PCATCH) {
3184 slpflag = 0;
3185 slptimeo = 2 * hz;
3186 }
3187 BO_LOCK(bo);
3188 }
3189 }
3190 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3191 BO_UNLOCK(bo);
3192 goto loop;
3193 }
3194 /*
3195 * Wait for all the async IO requests to drain
3196 */
3197 BO_UNLOCK(bo);
3198 NFSLOCKNODE(np);
3199 while (np->n_directio_asyncwr > 0) {
3200 np->n_flag |= NFSYNCWAIT;
3201 error = newnfs_msleep(td, &np->n_directio_asyncwr,
3202 &np->n_mtx, slpflag | (PRIBIO + 1),
3203 "nfsfsync", 0);
3204 if (error) {
3205 if (newnfs_sigintr(nmp, td)) {
3206 NFSUNLOCKNODE(np);
3207 error = EINTR;
3208 goto done;
3209 }
3210 }
3211 }
3212 NFSUNLOCKNODE(np);
3213 } else
3214 BO_UNLOCK(bo);
3215 if (NFSHASPNFS(nmp)) {
3216 nfscl_layoutcommit(vp, td);
3217 /*
3218 * Invalidate the attribute cache, since writes to a DS
3219 * won't update the size attribute.
3220 */
3221 NFSLOCKNODE(np);
3222 np->n_attrstamp = 0;
3223 } else
3224 NFSLOCKNODE(np);
3225 if (np->n_flag & NWRITEERR) {
3226 error = np->n_error;
3227 np->n_flag &= ~NWRITEERR;
3228 }
3229 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3230 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3231 np->n_flag &= ~NMODIFIED;
3232 NFSUNLOCKNODE(np);
3233 done:
3234 if (bvec != NULL && bvec != bvec_on_stack)
3235 free(bvec, M_TEMP);
3236 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3237 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
3238 np->n_directio_asyncwr != 0)) {
3239 if (trycnt++ < 5) {
3240 /* try, try again... */
3241 passone = 1;
3242 wcred = NULL;
3243 bvec = NULL;
3244 bvecsize = 0;
3245 goto again;
3246 }
3247 vn_printf(vp, "ncl_flush failed");
3248 error = called_from_renewthread != 0 ? EIO : EBUSY;
3249 }
3250 if (error == 0) {
3251 nanouptime(&ts);
3252 NFSLOCKNODE(np);
3253 np->n_localmodtime = ts;
3254 NFSUNLOCKNODE(np);
3255 }
3256 return (error);
3257 }
3258
3259 /*
3260 * NFS advisory byte-level locks.
3261 */
3262 static int
3263 nfs_advlock(struct vop_advlock_args *ap)
3264 {
3265 struct vnode *vp = ap->a_vp;
3266 struct ucred *cred;
3267 struct nfsnode *np = VTONFS(ap->a_vp);
3268 struct proc *p = (struct proc *)ap->a_id;
3269 struct thread *td = curthread; /* XXX */
3270 struct vattr va;
3271 int ret, error;
3272 u_quad_t size;
3273 struct nfsmount *nmp;
3274
3275 error = NFSVOPLOCK(vp, LK_SHARED);
3276 if (error != 0)
3277 return (EBADF);
3278 nmp = VFSTONFS(vp->v_mount);
3279 if (!NFS_ISV4(vp) || (nmp->nm_flag & NFSMNT_NOLOCKD) != 0) {
3280 if ((nmp->nm_flag & NFSMNT_NOLOCKD) != 0) {
3281 size = np->n_size;
3282 NFSVOPUNLOCK(vp);
3283 error = lf_advlock(ap, &(vp->v_lockf), size);
3284 } else {
3285 if (nfs_advlock_p != NULL)
3286 error = nfs_advlock_p(ap);
3287 else {
3288 NFSVOPUNLOCK(vp);
3289 error = ENOLCK;
3290 }
3291 }
3292 if (error == 0 && ap->a_op == F_SETLK) {
3293 error = NFSVOPLOCK(vp, LK_SHARED);
3294 if (error == 0) {
3295 /* Mark that a file lock has been acquired. */
3296 NFSLOCKNODE(np);
3297 np->n_flag |= NHASBEENLOCKED;
3298 NFSUNLOCKNODE(np);
3299 NFSVOPUNLOCK(vp);
3300 }
3301 }
3302 return (error);
3303 } else if ((ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3304 if (vp->v_type != VREG) {
3305 error = EINVAL;
3306 goto out;
3307 }
3308 if ((ap->a_flags & F_POSIX) != 0)
3309 cred = p->p_ucred;
3310 else
3311 cred = td->td_ucred;
3312 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
3313 if (VN_IS_DOOMED(vp)) {
3314 error = EBADF;
3315 goto out;
3316 }
3317
3318 /*
3319 * If this is unlocking a write locked region, flush and
3320 * commit them before unlocking. This is required by
3321 * RFC3530 Sec. 9.3.2.
3322 */
3323 if (ap->a_op == F_UNLCK &&
3324 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3325 ap->a_flags))
3326 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3327
3328 /*
3329 * Mark NFS node as might have acquired a lock.
3330 * This is separate from NHASBEENLOCKED, because it must
3331 * be done before the nfsrpc_advlock() call, which might
3332 * add a nfscllock structure to the client state.
3333 * It is used to check for the case where a nfscllock
3334 * state structure cannot exist for the file.
3335 * Only done for "oneopenown" NFSv4.1/4.2 mounts.
3336 */
3337 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) {
3338 NFSLOCKNODE(np);
3339 np->n_flag |= NMIGHTBELOCKED;
3340 NFSUNLOCKNODE(np);
3341 }
3342
3343 /*
3344 * Loop around doing the lock op, while a blocking lock
3345 * must wait for the lock op to succeed.
3346 */
3347 do {
3348 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3349 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3350 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3351 ap->a_op == F_SETLK) {
3352 NFSVOPUNLOCK(vp);
3353 error = nfs_catnap(PZERO | PCATCH, ret,
3354 "ncladvl");
3355 if (error)
3356 return (EINTR);
3357 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3358 if (VN_IS_DOOMED(vp)) {
3359 error = EBADF;
3360 goto out;
3361 }
3362 }
3363 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3364 ap->a_op == F_SETLK);
3365 if (ret == NFSERR_DENIED) {
3366 error = EAGAIN;
3367 goto out;
3368 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3369 error = ret;
3370 goto out;
3371 } else if (ret != 0) {
3372 error = EACCES;
3373 goto out;
3374 }
3375
3376 /*
3377 * Now, if we just got a lock, invalidate data in the buffer
3378 * cache, as required, so that the coherency conforms with
3379 * RFC3530 Sec. 9.3.2.
3380 */
3381 if (ap->a_op == F_SETLK) {
3382 if ((np->n_flag & NMODIFIED) == 0) {
3383 np->n_attrstamp = 0;
3384 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3385 ret = VOP_GETATTR(vp, &va, cred);
3386 }
3387 if ((np->n_flag & NMODIFIED) || ret ||
3388 np->n_change != va.va_filerev) {
3389 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3390 np->n_attrstamp = 0;
3391 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3392 ret = VOP_GETATTR(vp, &va, cred);
3393 if (!ret) {
3394 np->n_mtime = va.va_mtime;
3395 np->n_change = va.va_filerev;
3396 }
3397 }
3398 /* Mark that a file lock has been acquired. */
3399 NFSLOCKNODE(np);
3400 np->n_flag |= NHASBEENLOCKED;
3401 NFSUNLOCKNODE(np);
3402 }
3403 } else
3404 error = EOPNOTSUPP;
3405 out:
3406 NFSVOPUNLOCK(vp);
3407 return (error);
3408 }
3409
3410 /*
3411 * NFS advisory byte-level locks.
3412 */
3413 static int
3414 nfs_advlockasync(struct vop_advlockasync_args *ap)
3415 {
3416 struct vnode *vp = ap->a_vp;
3417 u_quad_t size;
3418 int error;
3419
3420 error = NFSVOPLOCK(vp, LK_SHARED);
3421 if (error)
3422 return (error);
3423 if (NFS_ISV4(vp)) {
3424 NFSVOPUNLOCK(vp);
3425 return (EOPNOTSUPP);
3426 }
3427 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3428 size = VTONFS(vp)->n_size;
3429 NFSVOPUNLOCK(vp);
3430 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3431 } else {
3432 NFSVOPUNLOCK(vp);
3433 error = EOPNOTSUPP;
3434 }
3435 return (error);
3436 }
3437
3438 /*
3439 * Print out the contents of an nfsnode.
3440 */
3441 static int
3442 nfs_print(struct vop_print_args *ap)
3443 {
3444 struct vnode *vp = ap->a_vp;
3445 struct nfsnode *np = VTONFS(vp);
3446
3447 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3448 (uintmax_t)np->n_vattr.na_fsid);
3449 if (vp->v_type == VFIFO)
3450 fifo_printinfo(vp);
3451 printf("\n");
3452 return (0);
3453 }
3454
3455 /*
3456 * This is the "real" nfs::bwrite(struct buf*).
3457 * We set B_CACHE if this is a VMIO buffer.
3458 */
3459 int
3460 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3461 {
3462 int oldflags, rtval;
3463
3464 if (bp->b_flags & B_INVAL) {
3465 brelse(bp);
3466 return (0);
3467 }
3468
3469 oldflags = bp->b_flags;
3470 bp->b_flags |= B_CACHE;
3471
3472 /*
3473 * Undirty the bp. We will redirty it later if the I/O fails.
3474 */
3475 bundirty(bp);
3476 bp->b_flags &= ~B_DONE;
3477 bp->b_ioflags &= ~BIO_ERROR;
3478 bp->b_iocmd = BIO_WRITE;
3479
3480 bufobj_wref(bp->b_bufobj);
3481 curthread->td_ru.ru_oublock++;
3482
3483 /*
3484 * Note: to avoid loopback deadlocks, we do not
3485 * assign b_runningbufspace.
3486 */
3487 vfs_busy_pages(bp, 1);
3488
3489 BUF_KERNPROC(bp);
3490 bp->b_iooffset = dbtob(bp->b_blkno);
3491 bstrategy(bp);
3492
3493 if ((oldflags & B_ASYNC) != 0)
3494 return (0);
3495
3496 rtval = bufwait(bp);
3497 if (oldflags & B_DELWRI)
3498 reassignbuf(bp);
3499 brelse(bp);
3500 return (rtval);
3501 }
3502
3503 /*
3504 * nfs special file access vnode op.
3505 * Essentially just get vattr and then imitate iaccess() since the device is
3506 * local to the client.
3507 */
3508 static int
3509 nfsspec_access(struct vop_access_args *ap)
3510 {
3511 struct vattr *vap;
3512 struct ucred *cred = ap->a_cred;
3513 struct vnode *vp = ap->a_vp;
3514 accmode_t accmode = ap->a_accmode;
3515 struct vattr vattr;
3516 int error;
3517
3518 /*
3519 * Disallow write attempts on filesystems mounted read-only;
3520 * unless the file is a socket, fifo, or a block or character
3521 * device resident on the filesystem.
3522 */
3523 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3524 switch (vp->v_type) {
3525 case VREG:
3526 case VDIR:
3527 case VLNK:
3528 return (EROFS);
3529 default:
3530 break;
3531 }
3532 }
3533 vap = &vattr;
3534 error = VOP_GETATTR(vp, vap, cred);
3535 if (error)
3536 goto out;
3537 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3538 accmode, cred);
3539 out:
3540 return error;
3541 }
3542
3543 /*
3544 * Read wrapper for fifos.
3545 */
3546 static int
3547 nfsfifo_read(struct vop_read_args *ap)
3548 {
3549 struct nfsnode *np = VTONFS(ap->a_vp);
3550 int error;
3551
3552 /*
3553 * Set access flag.
3554 */
3555 NFSLOCKNODE(np);
3556 np->n_flag |= NACC;
3557 vfs_timestamp(&np->n_atim);
3558 NFSUNLOCKNODE(np);
3559 error = fifo_specops.vop_read(ap);
3560 return error;
3561 }
3562
3563 /*
3564 * Write wrapper for fifos.
3565 */
3566 static int
3567 nfsfifo_write(struct vop_write_args *ap)
3568 {
3569 struct nfsnode *np = VTONFS(ap->a_vp);
3570
3571 /*
3572 * Set update flag.
3573 */
3574 NFSLOCKNODE(np);
3575 np->n_flag |= NUPD;
3576 vfs_timestamp(&np->n_mtim);
3577 NFSUNLOCKNODE(np);
3578 return(fifo_specops.vop_write(ap));
3579 }
3580
3581 /*
3582 * Close wrapper for fifos.
3583 *
3584 * Update the times on the nfsnode then do fifo close.
3585 */
3586 static int
3587 nfsfifo_close(struct vop_close_args *ap)
3588 {
3589 struct vnode *vp = ap->a_vp;
3590 struct nfsnode *np = VTONFS(vp);
3591 struct vattr vattr;
3592 struct timespec ts;
3593
3594 NFSLOCKNODE(np);
3595 if (np->n_flag & (NACC | NUPD)) {
3596 vfs_timestamp(&ts);
3597 if (np->n_flag & NACC)
3598 np->n_atim = ts;
3599 if (np->n_flag & NUPD)
3600 np->n_mtim = ts;
3601 np->n_flag |= NCHG;
3602 if (vrefcnt(vp) == 1 &&
3603 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3604 VATTR_NULL(&vattr);
3605 if (np->n_flag & NACC)
3606 vattr.va_atime = np->n_atim;
3607 if (np->n_flag & NUPD)
3608 vattr.va_mtime = np->n_mtim;
3609 NFSUNLOCKNODE(np);
3610 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3611 goto out;
3612 }
3613 }
3614 NFSUNLOCKNODE(np);
3615 out:
3616 return (fifo_specops.vop_close(ap));
3617 }
3618
3619 /*
3620 * Just call ncl_writebp() with the force argument set to 1.
3621 *
3622 * NOTE: B_DONE may or may not be set in a_bp on call.
3623 */
3624 static int
3625 nfs_bwrite(struct buf *bp)
3626 {
3627
3628 return (ncl_writebp(bp, 1, curthread));
3629 }
3630
3631 struct buf_ops buf_ops_newnfs = {
3632 .bop_name = "buf_ops_nfs",
3633 .bop_write = nfs_bwrite,
3634 .bop_strategy = bufstrategy,
3635 .bop_sync = bufsync,
3636 .bop_bdflush = bufbdflush,
3637 };
3638
3639 static int
3640 nfs_getacl(struct vop_getacl_args *ap)
3641 {
3642 int error;
3643
3644 if (ap->a_type != ACL_TYPE_NFS4)
3645 return (EOPNOTSUPP);
3646 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp);
3647 if (error > NFSERR_STALE) {
3648 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3649 error = EPERM;
3650 }
3651 return (error);
3652 }
3653
3654 static int
3655 nfs_setacl(struct vop_setacl_args *ap)
3656 {
3657 int error;
3658
3659 if (ap->a_type != ACL_TYPE_NFS4)
3660 return (EOPNOTSUPP);
3661 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp);
3662 if (error > NFSERR_STALE) {
3663 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3664 error = EPERM;
3665 }
3666 return (error);
3667 }
3668
3669 /*
3670 * VOP_ADVISE for NFS.
3671 * Just return 0 for any errors, since it is just a hint.
3672 */
3673 static int
3674 nfs_advise(struct vop_advise_args *ap)
3675 {
3676 struct thread *td = curthread;
3677 struct nfsmount *nmp;
3678 uint64_t len;
3679 int error;
3680
3681 /*
3682 * First do vop_stdadvise() to handle the buffer cache.
3683 */
3684 error = vop_stdadvise(ap);
3685 if (error != 0)
3686 return (error);
3687 if (ap->a_start < 0 || ap->a_end < 0)
3688 return (0);
3689 if (ap->a_end == OFF_MAX)
3690 len = 0;
3691 else if (ap->a_end < ap->a_start)
3692 return (0);
3693 else
3694 len = ap->a_end - ap->a_start + 1;
3695 nmp = VFSTONFS(ap->a_vp->v_mount);
3696 mtx_lock(&nmp->nm_mtx);
3697 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3698 (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) ==
3699 0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) {
3700 mtx_unlock(&nmp->nm_mtx);
3701 return (0);
3702 }
3703 mtx_unlock(&nmp->nm_mtx);
3704 error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice,
3705 td->td_ucred, td);
3706 if (error == NFSERR_NOTSUPP) {
3707 mtx_lock(&nmp->nm_mtx);
3708 nmp->nm_privflag |= NFSMNTP_NOADVISE;
3709 mtx_unlock(&nmp->nm_mtx);
3710 }
3711 return (0);
3712 }
3713
3714 /*
3715 * nfs allocate call
3716 */
3717 static int
3718 nfs_allocate(struct vop_allocate_args *ap)
3719 {
3720 struct vnode *vp = ap->a_vp;
3721 struct thread *td = curthread;
3722 struct nfsvattr nfsva;
3723 struct nfsmount *nmp;
3724 struct nfsnode *np;
3725 off_t alen;
3726 int attrflag, error, ret;
3727 struct timespec ts;
3728 struct uio io;
3729
3730 attrflag = 0;
3731 nmp = VFSTONFS(vp->v_mount);
3732 np = VTONFS(vp);
3733 mtx_lock(&nmp->nm_mtx);
3734 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3735 (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) {
3736 mtx_unlock(&nmp->nm_mtx);
3737 alen = *ap->a_len;
3738 if ((uint64_t)alen > nfs_maxalloclen)
3739 alen = nfs_maxalloclen;
3740
3741 /* Check the file size limit. */
3742 io.uio_offset = *ap->a_offset;
3743 io.uio_resid = alen;
3744 error = vn_rlimit_fsize(vp, &io, td);
3745
3746 /*
3747 * Flush first to ensure that the allocate adds to the
3748 * file's allocation on the server.
3749 */
3750 if (error == 0)
3751 error = ncl_flush(vp, MNT_WAIT, td, 1, 0);
3752 if (error == 0)
3753 error = nfsrpc_allocate(vp, *ap->a_offset, alen,
3754 &nfsva, &attrflag, ap->a_cred, td);
3755 if (error == 0) {
3756 *ap->a_offset += alen;
3757 *ap->a_len -= alen;
3758 nanouptime(&ts);
3759 NFSLOCKNODE(np);
3760 np->n_localmodtime = ts;
3761 NFSUNLOCKNODE(np);
3762 } else if (error == NFSERR_NOTSUPP) {
3763 mtx_lock(&nmp->nm_mtx);
3764 nmp->nm_privflag |= NFSMNTP_NOALLOCATE;
3765 mtx_unlock(&nmp->nm_mtx);
3766 error = EINVAL;
3767 }
3768 } else {
3769 mtx_unlock(&nmp->nm_mtx);
3770 error = EINVAL;
3771 }
3772 if (attrflag != 0) {
3773 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
3774 if (error == 0 && ret != 0)
3775 error = ret;
3776 }
3777 if (error != 0)
3778 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3779 return (error);
3780 }
3781
3782 /*
3783 * nfs deallocate call
3784 */
3785 static int
3786 nfs_deallocate(struct vop_deallocate_args *ap)
3787 {
3788 struct vnode *vp = ap->a_vp;
3789 struct thread *td = curthread;
3790 struct nfsvattr nfsva;
3791 struct nfsmount *nmp;
3792 struct nfsnode *np;
3793 off_t tlen, mlen;
3794 int attrflag, error, ret;
3795 bool clipped;
3796 struct timespec ts;
3797
3798 error = 0;
3799 attrflag = 0;
3800 nmp = VFSTONFS(vp->v_mount);
3801 np = VTONFS(vp);
3802 mtx_lock(&nmp->nm_mtx);
3803 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3804 (nmp->nm_privflag & NFSMNTP_NODEALLOCATE) == 0) {
3805 mtx_unlock(&nmp->nm_mtx);
3806 tlen = omin(OFF_MAX - *ap->a_offset, *ap->a_len);
3807 NFSCL_DEBUG(4, "dealloc: off=%jd len=%jd maxfilesize=%ju\n",
3808 (intmax_t)*ap->a_offset, (intmax_t)tlen,
3809 (uintmax_t)nmp->nm_maxfilesize);
3810 if ((uint64_t)*ap->a_offset >= nmp->nm_maxfilesize) {
3811 /* Avoid EFBIG error return from the NFSv4.2 server. */
3812 *ap->a_len = 0;
3813 return (0);
3814 }
3815 clipped = false;
3816 if ((uint64_t)*ap->a_offset + tlen > nmp->nm_maxfilesize)
3817 tlen = nmp->nm_maxfilesize - *ap->a_offset;
3818 if ((uint64_t)*ap->a_offset < np->n_size) {
3819 /* Limit the len to nfs_maxalloclen before EOF. */
3820 mlen = omin((off_t)np->n_size - *ap->a_offset, tlen);
3821 if ((uint64_t)mlen > nfs_maxalloclen) {
3822 NFSCL_DEBUG(4, "dealloc: tlen maxalloclen\n");
3823 tlen = nfs_maxalloclen;
3824 clipped = true;
3825 }
3826 }
3827 if (error == 0)
3828 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
3829 if (error == 0) {
3830 vnode_pager_purge_range(vp, *ap->a_offset,
3831 *ap->a_offset + tlen);
3832 error = nfsrpc_deallocate(vp, *ap->a_offset, tlen,
3833 &nfsva, &attrflag, ap->a_cred, td);
3834 NFSCL_DEBUG(4, "dealloc: rpc=%d\n", error);
3835 }
3836 if (error == 0) {
3837 NFSCL_DEBUG(4, "dealloc: attrflag=%d na_size=%ju\n",
3838 attrflag, (uintmax_t)nfsva.na_size);
3839 nanouptime(&ts);
3840 NFSLOCKNODE(np);
3841 np->n_localmodtime = ts;
3842 NFSUNLOCKNODE(np);
3843 if (attrflag != 0) {
3844 if ((uint64_t)*ap->a_offset < nfsva.na_size)
3845 *ap->a_offset += omin((off_t)
3846 nfsva.na_size - *ap->a_offset,
3847 tlen);
3848 }
3849 if (clipped && tlen < *ap->a_len)
3850 *ap->a_len -= tlen;
3851 else
3852 *ap->a_len = 0;
3853 } else if (error == NFSERR_NOTSUPP) {
3854 mtx_lock(&nmp->nm_mtx);
3855 nmp->nm_privflag |= NFSMNTP_NODEALLOCATE;
3856 mtx_unlock(&nmp->nm_mtx);
3857 }
3858 } else {
3859 mtx_unlock(&nmp->nm_mtx);
3860 error = EIO;
3861 }
3862 /*
3863 * If the NFS server cannot perform the Deallocate operation, just call
3864 * vop_stddeallocate() to perform it.
3865 */
3866 if (error != 0 && error != NFSERR_FBIG && error != NFSERR_INVAL) {
3867 error = vop_stddeallocate(ap);
3868 NFSCL_DEBUG(4, "dealloc: stddeallocate=%d\n", error);
3869 }
3870 if (attrflag != 0) {
3871 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
3872 if (error == 0 && ret != 0)
3873 error = ret;
3874 }
3875 if (error != 0)
3876 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3877 return (error);
3878 }
3879
3880 /*
3881 * nfs copy_file_range call
3882 */
3883 static int
3884 nfs_copy_file_range(struct vop_copy_file_range_args *ap)
3885 {
3886 struct vnode *invp = ap->a_invp;
3887 struct vnode *outvp = ap->a_outvp;
3888 struct mount *mp;
3889 struct nfsvattr innfsva, outnfsva;
3890 struct vattr *vap;
3891 struct uio io;
3892 struct nfsmount *nmp;
3893 size_t len, len2;
3894 ssize_t r;
3895 int error, inattrflag, outattrflag, ret, ret2;
3896 off_t inoff, outoff;
3897 bool consecutive, must_commit, tryoutcred;
3898
3899 /* NFSv4.2 Copy is not permitted for infile == outfile. */
3900 if (invp == outvp) {
3901 generic_copy:
3902 return (vn_generic_copy_file_range(invp, ap->a_inoffp,
3903 outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3904 ap->a_incred, ap->a_outcred, ap->a_fsizetd));
3905 }
3906
3907 /* Lock both vnodes, avoiding risk of deadlock. */
3908 do {
3909 mp = NULL;
3910 error = vn_start_write(outvp, &mp, V_WAIT);
3911 if (error == 0) {
3912 error = vn_lock(outvp, LK_EXCLUSIVE);
3913 if (error == 0) {
3914 error = vn_lock(invp, LK_SHARED | LK_NOWAIT);
3915 if (error == 0)
3916 break;
3917 VOP_UNLOCK(outvp);
3918 if (mp != NULL)
3919 vn_finished_write(mp);
3920 mp = NULL;
3921 error = vn_lock(invp, LK_SHARED);
3922 if (error == 0)
3923 VOP_UNLOCK(invp);
3924 }
3925 }
3926 if (mp != NULL)
3927 vn_finished_write(mp);
3928 } while (error == 0);
3929 if (error != 0)
3930 return (error);
3931
3932 /*
3933 * More reasons to avoid nfs copy: not NFSv4.2, or explicitly
3934 * disabled.
3935 */
3936 nmp = VFSTONFS(invp->v_mount);
3937 mtx_lock(&nmp->nm_mtx);
3938 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3939 (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0) {
3940 mtx_unlock(&nmp->nm_mtx);
3941 VOP_UNLOCK(invp);
3942 VOP_UNLOCK(outvp);
3943 if (mp != NULL)
3944 vn_finished_write(mp);
3945 goto generic_copy;
3946 }
3947 mtx_unlock(&nmp->nm_mtx);
3948
3949 /*
3950 * Do the vn_rlimit_fsize() check. Should this be above the VOP layer?
3951 */
3952 io.uio_offset = *ap->a_outoffp;
3953 io.uio_resid = *ap->a_lenp;
3954 error = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd);
3955 *ap->a_lenp = io.uio_resid;
3956 /*
3957 * No need to call vn_rlimit_fsizex_res before return, since the uio is
3958 * local.
3959 */
3960
3961 /*
3962 * Flush the input file so that the data is up to date before
3963 * the copy. Flush writes for the output file so that they
3964 * do not overwrite the data copied to the output file by the Copy.
3965 * Set the commit argument for both flushes so that the data is on
3966 * stable storage before the Copy RPC. This is done in case the
3967 * server reboots during the Copy and needs to be redone.
3968 */
3969 if (error == 0)
3970 error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0);
3971 if (error == 0)
3972 error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0);
3973
3974 /* Do the actual NFSv4.2 RPC. */
3975 ret = ret2 = 0;
3976 len = *ap->a_lenp;
3977 mtx_lock(&nmp->nm_mtx);
3978 if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0)
3979 consecutive = true;
3980 else
3981 consecutive = false;
3982 mtx_unlock(&nmp->nm_mtx);
3983 inoff = *ap->a_inoffp;
3984 outoff = *ap->a_outoffp;
3985 tryoutcred = true;
3986 must_commit = false;
3987 if (error == 0) {
3988 vap = &VTONFS(invp)->n_vattr.na_vattr;
3989 error = VOP_GETATTR(invp, vap, ap->a_incred);
3990 if (error == 0) {
3991 /*
3992 * Clip "len" at va_size so that RFC compliant servers
3993 * will not reply NFSERR_INVAL.
3994 * Setting "len == 0" for the RPC would be preferred,
3995 * but some Linux servers do not support that.
3996 */
3997 if (inoff >= vap->va_size)
3998 *ap->a_lenp = len = 0;
3999 else if (inoff + len > vap->va_size)
4000 *ap->a_lenp = len = vap->va_size - inoff;
4001 } else
4002 error = 0;
4003 }
4004
4005 /*
4006 * len will be set to 0 upon a successful Copy RPC.
4007 * As such, this only loops when the Copy RPC needs to be retried.
4008 */
4009 while (len > 0 && error == 0) {
4010 inattrflag = outattrflag = 0;
4011 len2 = len;
4012 if (tryoutcred)
4013 error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
4014 outvp, ap->a_outoffp, &len2, ap->a_flags,
4015 &inattrflag, &innfsva, &outattrflag, &outnfsva,
4016 ap->a_outcred, consecutive, &must_commit);
4017 else
4018 error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
4019 outvp, ap->a_outoffp, &len2, ap->a_flags,
4020 &inattrflag, &innfsva, &outattrflag, &outnfsva,
4021 ap->a_incred, consecutive, &must_commit);
4022 if (inattrflag != 0)
4023 ret = nfscl_loadattrcache(&invp, &innfsva, NULL, 0, 1);
4024 if (outattrflag != 0)
4025 ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL,
4026 1, 1);
4027 if (error == 0) {
4028 if (consecutive == false) {
4029 if (len2 == len) {
4030 mtx_lock(&nmp->nm_mtx);
4031 nmp->nm_privflag |=
4032 NFSMNTP_NOCONSECUTIVE;
4033 mtx_unlock(&nmp->nm_mtx);
4034 } else
4035 error = NFSERR_OFFLOADNOREQS;
4036 }
4037 *ap->a_lenp = len2;
4038 len = 0;
4039 if (len2 > 0 && must_commit && error == 0)
4040 error = ncl_commit(outvp, outoff, *ap->a_lenp,
4041 ap->a_outcred, curthread);
4042 if (error == 0 && ret != 0)
4043 error = ret;
4044 if (error == 0 && ret2 != 0)
4045 error = ret2;
4046 } else if (error == NFSERR_OFFLOADNOREQS && consecutive) {
4047 /*
4048 * Try consecutive == false, which is ok only if all
4049 * bytes are copied.
4050 * If only some bytes were copied when consecutive
4051 * is false, there is no way to know which bytes
4052 * still need to be written.
4053 */
4054 consecutive = false;
4055 error = 0;
4056 } else if (error == NFSERR_ACCES && tryoutcred) {
4057 /* Try again with incred. */
4058 tryoutcred = false;
4059 error = 0;
4060 }
4061 if (error == NFSERR_STALEWRITEVERF) {
4062 /*
4063 * Server rebooted, so do it all again.
4064 */
4065 *ap->a_inoffp = inoff;
4066 *ap->a_outoffp = outoff;
4067 len = *ap->a_lenp;
4068 must_commit = false;
4069 error = 0;
4070 }
4071 }
4072 VOP_UNLOCK(invp);
4073 VOP_UNLOCK(outvp);
4074 if (mp != NULL)
4075 vn_finished_write(mp);
4076 if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS ||
4077 error == NFSERR_ACCES) {
4078 /*
4079 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can
4080 * use a_incred for the read and a_outcred for the write, so
4081 * try this for NFSERR_ACCES failures for the Copy.
4082 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can
4083 * never succeed, so disable it.
4084 */
4085 if (error != NFSERR_ACCES) {
4086 /* Can never do Copy on this mount. */
4087 mtx_lock(&nmp->nm_mtx);
4088 nmp->nm_privflag |= NFSMNTP_NOCOPY;
4089 mtx_unlock(&nmp->nm_mtx);
4090 }
4091 *ap->a_inoffp = inoff;
4092 *ap->a_outoffp = outoff;
4093 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
4094 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
4095 ap->a_incred, ap->a_outcred, ap->a_fsizetd);
4096 } else if (error != 0)
4097 *ap->a_lenp = 0;
4098
4099 if (error != 0)
4100 error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0);
4101 return (error);
4102 }
4103
4104 /*
4105 * nfs ioctl call
4106 */
4107 static int
4108 nfs_ioctl(struct vop_ioctl_args *ap)
4109 {
4110 struct vnode *vp = ap->a_vp;
4111 struct nfsvattr nfsva;
4112 struct nfsmount *nmp;
4113 int attrflag, content, error, ret;
4114 bool eof = false; /* shut up compiler. */
4115
4116 /* Do the actual NFSv4.2 RPC. */
4117 switch (ap->a_command) {
4118 case FIOSEEKDATA:
4119 content = NFSV4CONTENT_DATA;
4120 break;
4121 case FIOSEEKHOLE:
4122 content = NFSV4CONTENT_HOLE;
4123 break;
4124 default:
4125 return (ENOTTY);
4126 }
4127
4128 error = vn_lock(vp, LK_SHARED);
4129 if (error != 0)
4130 return (EBADF);
4131
4132 if (vp->v_type != VREG) {
4133 VOP_UNLOCK(vp);
4134 return (ENOTTY);
4135 }
4136 nmp = VFSTONFS(vp->v_mount);
4137 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) {
4138 VOP_UNLOCK(vp);
4139 error = vop_stdioctl(ap);
4140 return (error);
4141 }
4142
4143 attrflag = 0;
4144 if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size)
4145 error = ENXIO;
4146 else {
4147 /*
4148 * Flush all writes, so that the server is up to date.
4149 * Although a Commit is not required, the commit argument
4150 * is set so that, for a pNFS File/Flexible File Layout
4151 * server, the LayoutCommit will be done to ensure the file
4152 * size is up to date on the Metadata Server.
4153 */
4154 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0);
4155 if (error == 0)
4156 error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof,
4157 content, ap->a_cred, &nfsva, &attrflag);
4158 /* If at eof for FIOSEEKDATA, return ENXIO. */
4159 if (eof && error == 0 && content == NFSV4CONTENT_DATA)
4160 error = ENXIO;
4161 }
4162 if (attrflag != 0) {
4163 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4164 if (error == 0 && ret != 0)
4165 error = ret;
4166 }
4167 NFSVOPUNLOCK(vp);
4168
4169 if (error != 0)
4170 error = ENXIO;
4171 return (error);
4172 }
4173
4174 /*
4175 * nfs getextattr call
4176 */
4177 static int
4178 nfs_getextattr(struct vop_getextattr_args *ap)
4179 {
4180 struct vnode *vp = ap->a_vp;
4181 struct nfsmount *nmp;
4182 struct ucred *cred;
4183 struct thread *td = ap->a_td;
4184 struct nfsvattr nfsva;
4185 ssize_t len;
4186 int attrflag, error, ret;
4187
4188 nmp = VFSTONFS(vp->v_mount);
4189 mtx_lock(&nmp->nm_mtx);
4190 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4191 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4192 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4193 mtx_unlock(&nmp->nm_mtx);
4194 return (EOPNOTSUPP);
4195 }
4196 mtx_unlock(&nmp->nm_mtx);
4197
4198 cred = ap->a_cred;
4199 if (cred == NULL)
4200 cred = td->td_ucred;
4201 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4202 attrflag = 0;
4203 error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva,
4204 &attrflag, cred, td);
4205 if (attrflag != 0) {
4206 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4207 if (error == 0 && ret != 0)
4208 error = ret;
4209 }
4210 if (error == 0 && ap->a_size != NULL)
4211 *ap->a_size = len;
4212
4213 switch (error) {
4214 case NFSERR_NOTSUPP:
4215 case NFSERR_OPILLEGAL:
4216 mtx_lock(&nmp->nm_mtx);
4217 nmp->nm_privflag |= NFSMNTP_NOXATTR;
4218 mtx_unlock(&nmp->nm_mtx);
4219 error = EOPNOTSUPP;
4220 break;
4221 case NFSERR_NOXATTR:
4222 case NFSERR_XATTR2BIG:
4223 error = ENOATTR;
4224 break;
4225 default:
4226 error = nfscl_maperr(td, error, 0, 0);
4227 break;
4228 }
4229 return (error);
4230 }
4231
4232 /*
4233 * nfs setextattr call
4234 */
4235 static int
4236 nfs_setextattr(struct vop_setextattr_args *ap)
4237 {
4238 struct vnode *vp = ap->a_vp;
4239 struct nfsmount *nmp;
4240 struct ucred *cred;
4241 struct thread *td = ap->a_td;
4242 struct nfsvattr nfsva;
4243 int attrflag, error, ret;
4244
4245 nmp = VFSTONFS(vp->v_mount);
4246 mtx_lock(&nmp->nm_mtx);
4247 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4248 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4249 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4250 mtx_unlock(&nmp->nm_mtx);
4251 return (EOPNOTSUPP);
4252 }
4253 mtx_unlock(&nmp->nm_mtx);
4254
4255 if (ap->a_uio->uio_resid < 0)
4256 return (EINVAL);
4257 cred = ap->a_cred;
4258 if (cred == NULL)
4259 cred = td->td_ucred;
4260 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4261 attrflag = 0;
4262 error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva,
4263 &attrflag, cred, td);
4264 if (attrflag != 0) {
4265 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4266 if (error == 0 && ret != 0)
4267 error = ret;
4268 }
4269
4270 switch (error) {
4271 case NFSERR_NOTSUPP:
4272 case NFSERR_OPILLEGAL:
4273 mtx_lock(&nmp->nm_mtx);
4274 nmp->nm_privflag |= NFSMNTP_NOXATTR;
4275 mtx_unlock(&nmp->nm_mtx);
4276 error = EOPNOTSUPP;
4277 break;
4278 case NFSERR_NOXATTR:
4279 case NFSERR_XATTR2BIG:
4280 error = ENOATTR;
4281 break;
4282 default:
4283 error = nfscl_maperr(td, error, 0, 0);
4284 break;
4285 }
4286 return (error);
4287 }
4288
4289 /*
4290 * nfs listextattr call
4291 */
4292 static int
4293 nfs_listextattr(struct vop_listextattr_args *ap)
4294 {
4295 struct vnode *vp = ap->a_vp;
4296 struct nfsmount *nmp;
4297 struct ucred *cred;
4298 struct thread *td = ap->a_td;
4299 struct nfsvattr nfsva;
4300 size_t len, len2;
4301 uint64_t cookie;
4302 int attrflag, error, ret;
4303 bool eof;
4304
4305 nmp = VFSTONFS(vp->v_mount);
4306 mtx_lock(&nmp->nm_mtx);
4307 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4308 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4309 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4310 mtx_unlock(&nmp->nm_mtx);
4311 return (EOPNOTSUPP);
4312 }
4313 mtx_unlock(&nmp->nm_mtx);
4314
4315 cred = ap->a_cred;
4316 if (cred == NULL)
4317 cred = td->td_ucred;
4318
4319 /* Loop around doing List Extended Attribute RPCs. */
4320 eof = false;
4321 cookie = 0;
4322 len2 = 0;
4323 error = 0;
4324 while (!eof && error == 0) {
4325 len = nmp->nm_rsize;
4326 attrflag = 0;
4327 error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof,
4328 &nfsva, &attrflag, cred, td);
4329 if (attrflag != 0) {
4330 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4331 if (error == 0 && ret != 0)
4332 error = ret;
4333 }
4334 if (error == 0) {
4335 len2 += len;
4336 if (len2 > SSIZE_MAX)
4337 error = ENOATTR;
4338 }
4339 }
4340 if (error == 0 && ap->a_size != NULL)
4341 *ap->a_size = len2;
4342
4343 switch (error) {
4344 case NFSERR_NOTSUPP:
4345 case NFSERR_OPILLEGAL:
4346 mtx_lock(&nmp->nm_mtx);
4347 nmp->nm_privflag |= NFSMNTP_NOXATTR;
4348 mtx_unlock(&nmp->nm_mtx);
4349 error = EOPNOTSUPP;
4350 break;
4351 case NFSERR_NOXATTR:
4352 case NFSERR_XATTR2BIG:
4353 error = ENOATTR;
4354 break;
4355 default:
4356 error = nfscl_maperr(td, error, 0, 0);
4357 break;
4358 }
4359 return (error);
4360 }
4361
4362 /*
4363 * nfs setextattr call
4364 */
4365 static int
4366 nfs_deleteextattr(struct vop_deleteextattr_args *ap)
4367 {
4368 struct vnode *vp = ap->a_vp;
4369 struct nfsmount *nmp;
4370 struct nfsvattr nfsva;
4371 int attrflag, error, ret;
4372
4373 nmp = VFSTONFS(vp->v_mount);
4374 mtx_lock(&nmp->nm_mtx);
4375 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4376 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4377 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4378 mtx_unlock(&nmp->nm_mtx);
4379 return (EOPNOTSUPP);
4380 }
4381 mtx_unlock(&nmp->nm_mtx);
4382
4383 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4384 attrflag = 0;
4385 error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred,
4386 ap->a_td);
4387 if (attrflag != 0) {
4388 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4389 if (error == 0 && ret != 0)
4390 error = ret;
4391 }
4392
4393 switch (error) {
4394 case NFSERR_NOTSUPP:
4395 case NFSERR_OPILLEGAL:
4396 mtx_lock(&nmp->nm_mtx);
4397 nmp->nm_privflag |= NFSMNTP_NOXATTR;
4398 mtx_unlock(&nmp->nm_mtx);
4399 error = EOPNOTSUPP;
4400 break;
4401 case NFSERR_NOXATTR:
4402 case NFSERR_XATTR2BIG:
4403 error = ENOATTR;
4404 break;
4405 default:
4406 error = nfscl_maperr(ap->a_td, error, 0, 0);
4407 break;
4408 }
4409 return (error);
4410 }
4411
4412 /*
4413 * Return POSIX pathconf information applicable to nfs filesystems.
4414 */
4415 static int
4416 nfs_pathconf(struct vop_pathconf_args *ap)
4417 {
4418 struct nfsv3_pathconf pc;
4419 struct nfsvattr nfsva;
4420 struct vnode *vp = ap->a_vp;
4421 struct nfsmount *nmp;
4422 struct thread *td = curthread;
4423 off_t off;
4424 bool eof;
4425 int attrflag, error;
4426
4427 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
4428 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
4429 ap->a_name == _PC_NO_TRUNC)) ||
4430 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
4431 /*
4432 * Since only the above 4 a_names are returned by the NFSv3
4433 * Pathconf RPC, there is no point in doing it for others.
4434 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
4435 * be used for _PC_NFS4_ACL as well.
4436 */
4437 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
4438 &attrflag);
4439 if (attrflag != 0)
4440 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4441 if (error != 0)
4442 return (error);
4443 } else {
4444 /*
4445 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
4446 * just fake them.
4447 */
4448 pc.pc_linkmax = NFS_LINK_MAX;
4449 pc.pc_namemax = NFS_MAXNAMLEN;
4450 pc.pc_notrunc = 1;
4451 pc.pc_chownrestricted = 1;
4452 pc.pc_caseinsensitive = 0;
4453 pc.pc_casepreserving = 1;
4454 error = 0;
4455 }
4456 switch (ap->a_name) {
4457 case _PC_LINK_MAX:
4458 #ifdef _LP64
4459 *ap->a_retval = pc.pc_linkmax;
4460 #else
4461 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
4462 #endif
4463 break;
4464 case _PC_NAME_MAX:
4465 *ap->a_retval = pc.pc_namemax;
4466 break;
4467 case _PC_PIPE_BUF:
4468 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
4469 *ap->a_retval = PIPE_BUF;
4470 else
4471 error = EINVAL;
4472 break;
4473 case _PC_CHOWN_RESTRICTED:
4474 *ap->a_retval = pc.pc_chownrestricted;
4475 break;
4476 case _PC_NO_TRUNC:
4477 *ap->a_retval = pc.pc_notrunc;
4478 break;
4479 case _PC_ACL_NFS4:
4480 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4481 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
4482 *ap->a_retval = 1;
4483 else
4484 *ap->a_retval = 0;
4485 break;
4486 case _PC_ACL_PATH_MAX:
4487 if (NFS_ISV4(vp))
4488 *ap->a_retval = ACL_MAX_ENTRIES;
4489 else
4490 *ap->a_retval = 3;
4491 break;
4492 case _PC_PRIO_IO:
4493 *ap->a_retval = 0;
4494 break;
4495 case _PC_SYNC_IO:
4496 *ap->a_retval = 0;
4497 break;
4498 case _PC_ALLOC_SIZE_MIN:
4499 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
4500 break;
4501 case _PC_FILESIZEBITS:
4502 if (NFS_ISV34(vp))
4503 *ap->a_retval = 64;
4504 else
4505 *ap->a_retval = 32;
4506 break;
4507 case _PC_REC_INCR_XFER_SIZE:
4508 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4509 break;
4510 case _PC_REC_MAX_XFER_SIZE:
4511 *ap->a_retval = -1; /* means ``unlimited'' */
4512 break;
4513 case _PC_REC_MIN_XFER_SIZE:
4514 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4515 break;
4516 case _PC_REC_XFER_ALIGN:
4517 *ap->a_retval = PAGE_SIZE;
4518 break;
4519 case _PC_SYMLINK_MAX:
4520 *ap->a_retval = NFS_MAXPATHLEN;
4521 break;
4522 case _PC_MIN_HOLE_SIZE:
4523 /* Only some NFSv4.2 servers support Seek for Holes. */
4524 *ap->a_retval = 0;
4525 nmp = VFSTONFS(vp->v_mount);
4526 if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) {
4527 /*
4528 * NFSv4.2 doesn't have an attribute for hole size,
4529 * so all we can do is see if the Seek operation is
4530 * supported and then use f_iosize as a "best guess".
4531 */
4532 mtx_lock(&nmp->nm_mtx);
4533 if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) {
4534 mtx_unlock(&nmp->nm_mtx);
4535 off = 0;
4536 attrflag = 0;
4537 error = nfsrpc_seek(vp, &off, &eof,
4538 NFSV4CONTENT_HOLE, td->td_ucred, &nfsva,
4539 &attrflag);
4540 if (attrflag != 0)
4541 nfscl_loadattrcache(&vp, &nfsva,
4542 NULL, 0, 1);
4543 mtx_lock(&nmp->nm_mtx);
4544 if (error == NFSERR_NOTSUPP)
4545 nmp->nm_privflag |= NFSMNTP_SEEKTESTED;
4546 else
4547 nmp->nm_privflag |= NFSMNTP_SEEKTESTED |
4548 NFSMNTP_SEEK;
4549 error = 0;
4550 }
4551 if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0)
4552 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4553 mtx_unlock(&nmp->nm_mtx);
4554 }
4555 break;
4556
4557 default:
4558 error = vop_stdpathconf(ap);
4559 break;
4560 }
4561 return (error);
4562 }
Cache object: a458e2315755accc427467ca9cf83dc1
|