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_copy_file_range_t nfs_copy_file_range;
150 static vop_ioctl_t nfs_ioctl;
151 static vop_getextattr_t nfs_getextattr;
152 static vop_setextattr_t nfs_setextattr;
153 static vop_listextattr_t nfs_listextattr;
154 static vop_deleteextattr_t nfs_deleteextattr;
155 static vop_lock1_t nfs_lock;
156
157 /*
158 * Global vfs data structures for nfs
159 */
160
161 static struct vop_vector newnfs_vnodeops_nosig = {
162 .vop_default = &default_vnodeops,
163 .vop_access = nfs_access,
164 .vop_advlock = nfs_advlock,
165 .vop_advlockasync = nfs_advlockasync,
166 .vop_close = nfs_close,
167 .vop_create = nfs_create,
168 .vop_fsync = nfs_fsync,
169 .vop_getattr = nfs_getattr,
170 .vop_getpages = ncl_getpages,
171 .vop_putpages = ncl_putpages,
172 .vop_inactive = ncl_inactive,
173 .vop_link = nfs_link,
174 .vop_lock1 = nfs_lock,
175 .vop_lookup = nfs_lookup,
176 .vop_mkdir = nfs_mkdir,
177 .vop_mknod = nfs_mknod,
178 .vop_open = nfs_open,
179 .vop_pathconf = nfs_pathconf,
180 .vop_print = nfs_print,
181 .vop_read = nfs_read,
182 .vop_readdir = nfs_readdir,
183 .vop_readlink = nfs_readlink,
184 .vop_reclaim = ncl_reclaim,
185 .vop_remove = nfs_remove,
186 .vop_rename = nfs_rename,
187 .vop_rmdir = nfs_rmdir,
188 .vop_setattr = nfs_setattr,
189 .vop_strategy = nfs_strategy,
190 .vop_symlink = nfs_symlink,
191 .vop_write = ncl_write,
192 .vop_getacl = nfs_getacl,
193 .vop_setacl = nfs_setacl,
194 .vop_advise = nfs_advise,
195 .vop_allocate = nfs_allocate,
196 .vop_copy_file_range = nfs_copy_file_range,
197 .vop_ioctl = nfs_ioctl,
198 .vop_getextattr = nfs_getextattr,
199 .vop_setextattr = nfs_setextattr,
200 .vop_listextattr = nfs_listextattr,
201 .vop_deleteextattr = nfs_deleteextattr,
202 };
203 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig);
204
205 static int
206 nfs_vnodeops_bypass(struct vop_generic_args *a)
207 {
208
209 return (vop_sigdefer(&newnfs_vnodeops_nosig, a));
210 }
211
212 struct vop_vector newnfs_vnodeops = {
213 .vop_default = &default_vnodeops,
214 .vop_bypass = nfs_vnodeops_bypass,
215 };
216 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops);
217
218 static struct vop_vector newnfs_fifoops_nosig = {
219 .vop_default = &fifo_specops,
220 .vop_access = nfsspec_access,
221 .vop_close = nfsfifo_close,
222 .vop_fsync = nfs_fsync,
223 .vop_getattr = nfs_getattr,
224 .vop_inactive = ncl_inactive,
225 .vop_pathconf = nfs_pathconf,
226 .vop_print = nfs_print,
227 .vop_read = nfsfifo_read,
228 .vop_reclaim = ncl_reclaim,
229 .vop_setattr = nfs_setattr,
230 .vop_write = nfsfifo_write,
231 };
232 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig);
233
234 static int
235 nfs_fifoops_bypass(struct vop_generic_args *a)
236 {
237
238 return (vop_sigdefer(&newnfs_fifoops_nosig, a));
239 }
240
241 struct vop_vector newnfs_fifoops = {
242 .vop_default = &default_vnodeops,
243 .vop_bypass = nfs_fifoops_bypass,
244 };
245 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops);
246
247 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
248 struct componentname *cnp, struct vattr *vap);
249 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
250 int namelen, struct ucred *cred, struct thread *td);
251 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
252 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
253 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
254 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
255 struct componentname *scnp, struct sillyrename *sp);
256
257 /*
258 * Global variables
259 */
260 SYSCTL_DECL(_vfs_nfs);
261
262 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
263 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
264 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
265
266 static int nfs_prime_access_cache = 0;
267 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
268 &nfs_prime_access_cache, 0,
269 "Prime NFS ACCESS cache when fetching attributes");
270
271 static int newnfs_commit_on_close = 0;
272 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
273 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
274
275 static int nfs_clean_pages_on_close = 1;
276 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
277 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
278
279 int newnfs_directio_enable = 0;
280 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
281 &newnfs_directio_enable, 0, "Enable NFS directio");
282
283 int nfs_keep_dirty_on_error;
284 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
285 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
286
287 /*
288 * This sysctl allows other processes to mmap a file that has been opened
289 * O_DIRECT by a process. In general, having processes mmap the file while
290 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
291 * this by default to prevent DoS attacks - to prevent a malicious user from
292 * opening up files O_DIRECT preventing other users from mmap'ing these
293 * files. "Protected" environments where stricter consistency guarantees are
294 * required can disable this knob. The process that opened the file O_DIRECT
295 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
296 * meaningful.
297 */
298 int newnfs_directio_allow_mmap = 1;
299 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
300 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
301
302 static uint64_t nfs_maxalloclen = 64 * 1024 * 1024;
303 SYSCTL_U64(_vfs_nfs, OID_AUTO, maxalloclen, CTLFLAG_RW,
304 &nfs_maxalloclen, 0, "NFS max allocate/deallocate length");
305
306 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
307 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
308 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
309
310 /*
311 * SMP Locking Note :
312 * The list of locks after the description of the lock is the ordering
313 * of other locks acquired with the lock held.
314 * np->n_mtx : Protects the fields in the nfsnode.
315 VM Object Lock
316 VI_MTX (acquired indirectly)
317 * nmp->nm_mtx : Protects the fields in the nfsmount.
318 rep->r_mtx
319 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
320 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
321 nmp->nm_mtx
322 rep->r_mtx
323 * rep->r_mtx : Protects the fields in an nfsreq.
324 */
325
326 static int
327 nfs_lock(struct vop_lock1_args *ap)
328 {
329 struct vnode *vp;
330 struct nfsnode *np;
331 u_quad_t nsize;
332 int error, lktype;
333 bool onfault;
334
335 vp = ap->a_vp;
336 lktype = ap->a_flags & LK_TYPE_MASK;
337 error = VOP_LOCK1_APV(&default_vnodeops, ap);
338 if (error != 0 || vp->v_op != &newnfs_vnodeops)
339 return (error);
340 np = VTONFS(vp);
341 if (np == NULL)
342 return (0);
343 NFSLOCKNODE(np);
344 if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED &&
345 lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE &&
346 lktype != LK_TRYUPGRADE)) {
347 NFSUNLOCKNODE(np);
348 return (0);
349 }
350 onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT &&
351 (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE &&
352 (lktype == LK_SHARED || lktype == LK_EXCLUSIVE);
353 if (onfault && vp->v_vnlock->lk_recurse == 0) {
354 /*
355 * Force retry in vm_fault(), to make the lock request
356 * sleepable, which allows us to piggy-back the
357 * sleepable call to vnode_pager_setsize().
358 */
359 NFSUNLOCKNODE(np);
360 VOP_UNLOCK(vp);
361 return (EBUSY);
362 }
363 if ((ap->a_flags & LK_NOWAIT) != 0 ||
364 (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) {
365 NFSUNLOCKNODE(np);
366 return (0);
367 }
368 if (lktype == LK_SHARED) {
369 NFSUNLOCKNODE(np);
370 VOP_UNLOCK(vp);
371 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
372 ap->a_flags |= LK_EXCLUSIVE;
373 error = VOP_LOCK1_APV(&default_vnodeops, ap);
374 if (error != 0 || vp->v_op != &newnfs_vnodeops)
375 return (error);
376 if (vp->v_data == NULL)
377 goto downgrade;
378 MPASS(vp->v_data == np);
379 NFSLOCKNODE(np);
380 if ((np->n_flag & NVNSETSZSKIP) == 0) {
381 NFSUNLOCKNODE(np);
382 goto downgrade;
383 }
384 }
385 np->n_flag &= ~NVNSETSZSKIP;
386 nsize = np->n_size;
387 NFSUNLOCKNODE(np);
388 vnode_pager_setsize(vp, nsize);
389 downgrade:
390 if (lktype == LK_SHARED) {
391 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
392 ap->a_flags |= LK_DOWNGRADE;
393 (void)VOP_LOCK1_APV(&default_vnodeops, ap);
394 }
395 return (0);
396 }
397
398 static int
399 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
400 struct ucred *cred, u_int32_t *retmode)
401 {
402 int error = 0, attrflag, i, lrupos;
403 u_int32_t rmode;
404 struct nfsnode *np = VTONFS(vp);
405 struct nfsvattr nfsva;
406
407 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
408 &rmode, NULL);
409 if (attrflag)
410 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
411 if (!error) {
412 lrupos = 0;
413 NFSLOCKNODE(np);
414 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
415 if (np->n_accesscache[i].uid == cred->cr_uid) {
416 np->n_accesscache[i].mode = rmode;
417 np->n_accesscache[i].stamp = time_second;
418 break;
419 }
420 if (i > 0 && np->n_accesscache[i].stamp <
421 np->n_accesscache[lrupos].stamp)
422 lrupos = i;
423 }
424 if (i == NFS_ACCESSCACHESIZE) {
425 np->n_accesscache[lrupos].uid = cred->cr_uid;
426 np->n_accesscache[lrupos].mode = rmode;
427 np->n_accesscache[lrupos].stamp = time_second;
428 }
429 NFSUNLOCKNODE(np);
430 if (retmode != NULL)
431 *retmode = rmode;
432 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
433 } else if (NFS_ISV4(vp)) {
434 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
435 }
436 #ifdef KDTRACE_HOOKS
437 if (error != 0)
438 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
439 error);
440 #endif
441 return (error);
442 }
443
444 /*
445 * nfs access vnode op.
446 * For nfs version 2, just return ok. File accesses may fail later.
447 * For nfs version 3, use the access rpc to check accessibility. If file modes
448 * are changed on the server, accesses might still fail later.
449 */
450 static int
451 nfs_access(struct vop_access_args *ap)
452 {
453 struct vnode *vp = ap->a_vp;
454 int error = 0, i, gotahit;
455 u_int32_t mode, wmode, rmode;
456 int v34 = NFS_ISV34(vp);
457 struct nfsnode *np = VTONFS(vp);
458
459 /*
460 * Disallow write attempts on filesystems mounted read-only;
461 * unless the file is a socket, fifo, or a block or character
462 * device resident on the filesystem.
463 */
464 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
465 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
466 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
467 switch (vp->v_type) {
468 case VREG:
469 case VDIR:
470 case VLNK:
471 return (EROFS);
472 default:
473 break;
474 }
475 }
476 /*
477 * For nfs v3 or v4, check to see if we have done this recently, and if
478 * so return our cached result instead of making an ACCESS call.
479 * If not, do an access rpc, otherwise you are stuck emulating
480 * ufs_access() locally using the vattr. This may not be correct,
481 * since the server may apply other access criteria such as
482 * client uid-->server uid mapping that we do not know about.
483 */
484 if (v34) {
485 if (ap->a_accmode & VREAD)
486 mode = NFSACCESS_READ;
487 else
488 mode = 0;
489 if (vp->v_type != VDIR) {
490 if (ap->a_accmode & VWRITE)
491 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
492 if (ap->a_accmode & VAPPEND)
493 mode |= NFSACCESS_EXTEND;
494 if (ap->a_accmode & VEXEC)
495 mode |= NFSACCESS_EXECUTE;
496 if (ap->a_accmode & VDELETE)
497 mode |= NFSACCESS_DELETE;
498 } else {
499 if (ap->a_accmode & VWRITE)
500 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
501 if (ap->a_accmode & VAPPEND)
502 mode |= NFSACCESS_EXTEND;
503 if (ap->a_accmode & VEXEC)
504 mode |= NFSACCESS_LOOKUP;
505 if (ap->a_accmode & VDELETE)
506 mode |= NFSACCESS_DELETE;
507 if (ap->a_accmode & VDELETE_CHILD)
508 mode |= NFSACCESS_MODIFY;
509 }
510 /* XXX safety belt, only make blanket request if caching */
511 if (nfsaccess_cache_timeout > 0) {
512 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
513 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
514 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
515 } else {
516 wmode = mode;
517 }
518
519 /*
520 * Does our cached result allow us to give a definite yes to
521 * this request?
522 */
523 gotahit = 0;
524 NFSLOCKNODE(np);
525 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
526 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
527 if (time_second < (np->n_accesscache[i].stamp
528 + nfsaccess_cache_timeout) &&
529 (np->n_accesscache[i].mode & mode) == mode) {
530 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
531 gotahit = 1;
532 }
533 break;
534 }
535 }
536 NFSUNLOCKNODE(np);
537 #ifdef KDTRACE_HOOKS
538 if (gotahit != 0)
539 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
540 ap->a_cred->cr_uid, mode);
541 else
542 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
543 ap->a_cred->cr_uid, mode);
544 #endif
545 if (gotahit == 0) {
546 /*
547 * Either a no, or a don't know. Go to the wire.
548 */
549 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
550 error = nfs34_access_otw(vp, wmode, ap->a_td,
551 ap->a_cred, &rmode);
552 if (!error &&
553 (rmode & mode) != mode)
554 error = EACCES;
555 }
556 return (error);
557 } else {
558 if ((error = nfsspec_access(ap)) != 0) {
559 return (error);
560 }
561 /*
562 * Attempt to prevent a mapped root from accessing a file
563 * which it shouldn't. We try to read a byte from the file
564 * if the user is root and the file is not zero length.
565 * After calling nfsspec_access, we should have the correct
566 * file size cached.
567 */
568 NFSLOCKNODE(np);
569 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
570 && VTONFS(vp)->n_size > 0) {
571 struct iovec aiov;
572 struct uio auio;
573 char buf[1];
574
575 NFSUNLOCKNODE(np);
576 aiov.iov_base = buf;
577 aiov.iov_len = 1;
578 auio.uio_iov = &aiov;
579 auio.uio_iovcnt = 1;
580 auio.uio_offset = 0;
581 auio.uio_resid = 1;
582 auio.uio_segflg = UIO_SYSSPACE;
583 auio.uio_rw = UIO_READ;
584 auio.uio_td = ap->a_td;
585
586 if (vp->v_type == VREG)
587 error = ncl_readrpc(vp, &auio, ap->a_cred);
588 else if (vp->v_type == VDIR) {
589 char* bp;
590 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
591 aiov.iov_base = bp;
592 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
593 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
594 ap->a_td);
595 free(bp, M_TEMP);
596 } else if (vp->v_type == VLNK)
597 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
598 else
599 error = EACCES;
600 } else
601 NFSUNLOCKNODE(np);
602 return (error);
603 }
604 }
605
606 /*
607 * nfs open vnode op
608 * Check to see if the type is ok
609 * and that deletion is not in progress.
610 * For paged in text files, you will need to flush the page cache
611 * if consistency is lost.
612 */
613 /* ARGSUSED */
614 static int
615 nfs_open(struct vop_open_args *ap)
616 {
617 struct vnode *vp = ap->a_vp;
618 struct nfsnode *np = VTONFS(vp);
619 struct vattr vattr;
620 int error;
621 int fmode = ap->a_mode;
622 struct ucred *cred;
623 vm_object_t obj;
624
625 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
626 return (EOPNOTSUPP);
627
628 /*
629 * For NFSv4, we need to do the Open Op before cache validation,
630 * so that we conform to RFC3530 Sec. 9.3.1.
631 */
632 if (NFS_ISV4(vp)) {
633 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
634 if (error) {
635 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
636 (gid_t)0);
637 return (error);
638 }
639 }
640
641 /*
642 * Now, if this Open will be doing reading, re-validate/flush the
643 * cache, so that Close/Open coherency is maintained.
644 */
645 NFSLOCKNODE(np);
646 if (np->n_flag & NMODIFIED) {
647 NFSUNLOCKNODE(np);
648 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
649 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
650 if (VN_IS_DOOMED(vp))
651 return (EBADF);
652 }
653 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
654 if (error == EINTR || error == EIO) {
655 if (NFS_ISV4(vp))
656 (void) nfsrpc_close(vp, 0, ap->a_td);
657 return (error);
658 }
659 NFSLOCKNODE(np);
660 np->n_attrstamp = 0;
661 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
662 if (vp->v_type == VDIR)
663 np->n_direofoffset = 0;
664 NFSUNLOCKNODE(np);
665 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
666 if (error) {
667 if (NFS_ISV4(vp))
668 (void) nfsrpc_close(vp, 0, ap->a_td);
669 return (error);
670 }
671 NFSLOCKNODE(np);
672 np->n_mtime = vattr.va_mtime;
673 if (NFS_ISV4(vp))
674 np->n_change = vattr.va_filerev;
675 } else {
676 NFSUNLOCKNODE(np);
677 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
678 if (error) {
679 if (NFS_ISV4(vp))
680 (void) nfsrpc_close(vp, 0, ap->a_td);
681 return (error);
682 }
683 NFSLOCKNODE(np);
684 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
685 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
686 if (vp->v_type == VDIR)
687 np->n_direofoffset = 0;
688 NFSUNLOCKNODE(np);
689 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
690 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
691 if (VN_IS_DOOMED(vp))
692 return (EBADF);
693 }
694 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
695 if (error == EINTR || error == EIO) {
696 if (NFS_ISV4(vp))
697 (void) nfsrpc_close(vp, 0, ap->a_td);
698 return (error);
699 }
700 NFSLOCKNODE(np);
701 np->n_mtime = vattr.va_mtime;
702 if (NFS_ISV4(vp))
703 np->n_change = vattr.va_filerev;
704 }
705 }
706
707 /*
708 * If the object has >= 1 O_DIRECT active opens, we disable caching.
709 */
710 if (newnfs_directio_enable && (fmode & O_DIRECT) &&
711 (vp->v_type == VREG)) {
712 if (np->n_directio_opens == 0) {
713 NFSUNLOCKNODE(np);
714 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
715 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
716 if (VN_IS_DOOMED(vp))
717 return (EBADF);
718 }
719 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
720 if (error) {
721 if (NFS_ISV4(vp))
722 (void) nfsrpc_close(vp, 0, ap->a_td);
723 return (error);
724 }
725 NFSLOCKNODE(np);
726 np->n_flag |= NNONCACHE;
727 }
728 np->n_directio_opens++;
729 }
730
731 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
732 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
733 np->n_flag |= NWRITEOPENED;
734
735 /*
736 * If this is an open for writing, capture a reference to the
737 * credentials, so they can be used by ncl_putpages(). Using
738 * these write credentials is preferable to the credentials of
739 * whatever thread happens to be doing the VOP_PUTPAGES() since
740 * the write RPCs are less likely to fail with EACCES.
741 */
742 if ((fmode & FWRITE) != 0) {
743 cred = np->n_writecred;
744 np->n_writecred = crhold(ap->a_cred);
745 } else
746 cred = NULL;
747 NFSUNLOCKNODE(np);
748
749 if (cred != NULL)
750 crfree(cred);
751 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
752
753 /*
754 * If the text file has been mmap'd, flush any dirty pages to the
755 * buffer cache and then...
756 * Make sure all writes are pushed to the NFS server. If this is not
757 * done, the modify time of the file can change while the text
758 * file is being executed. This will cause the process that is
759 * executing the text file to be terminated.
760 */
761 if (vp->v_writecount <= -1) {
762 if ((obj = vp->v_object) != NULL &&
763 vm_object_mightbedirty(obj)) {
764 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
765 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
766 if (VN_IS_DOOMED(vp))
767 return (EBADF);
768 }
769 VM_OBJECT_WLOCK(obj);
770 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
771 VM_OBJECT_WUNLOCK(obj);
772 }
773
774 /* Now, flush the buffer cache. */
775 ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
776
777 /* And, finally, make sure that n_mtime is up to date. */
778 np = VTONFS(vp);
779 NFSLOCKNODE(np);
780 np->n_mtime = np->n_vattr.na_mtime;
781 NFSUNLOCKNODE(np);
782 }
783 return (0);
784 }
785
786 /*
787 * nfs close vnode op
788 * What an NFS client should do upon close after writing is a debatable issue.
789 * Most NFS clients push delayed writes to the server upon close, basically for
790 * two reasons:
791 * 1 - So that any write errors may be reported back to the client process
792 * doing the close system call. By far the two most likely errors are
793 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
794 * 2 - To put a worst case upper bound on cache inconsistency between
795 * multiple clients for the file.
796 * There is also a consistency problem for Version 2 of the protocol w.r.t.
797 * not being able to tell if other clients are writing a file concurrently,
798 * since there is no way of knowing if the changed modify time in the reply
799 * is only due to the write for this client.
800 * (NFS Version 3 provides weak cache consistency data in the reply that
801 * should be sufficient to detect and handle this case.)
802 *
803 * The current code does the following:
804 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
805 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
806 * or commit them (this satisfies 1 and 2 except for the
807 * case where the server crashes after this close but
808 * before the commit RPC, which is felt to be "good
809 * enough". Changing the last argument to ncl_flush() to
810 * a 1 would force a commit operation, if it is felt a
811 * commit is necessary now.
812 * for NFS Version 4 - flush the dirty buffers and commit them, if
813 * nfscl_mustflush() says this is necessary.
814 * It is necessary if there is no write delegation held,
815 * in order to satisfy open/close coherency.
816 * If the file isn't cached on local stable storage,
817 * it may be necessary in order to detect "out of space"
818 * errors from the server, if the write delegation
819 * issued by the server doesn't allow the file to grow.
820 */
821 /* ARGSUSED */
822 static int
823 nfs_close(struct vop_close_args *ap)
824 {
825 struct vnode *vp = ap->a_vp;
826 struct nfsnode *np = VTONFS(vp);
827 struct nfsvattr nfsva;
828 struct ucred *cred;
829 int error = 0, ret, localcred = 0;
830 int fmode = ap->a_fflag;
831
832 if (NFSCL_FORCEDISM(vp->v_mount))
833 return (0);
834 /*
835 * During shutdown, a_cred isn't valid, so just use root.
836 */
837 if (ap->a_cred == NOCRED) {
838 cred = newnfs_getcred();
839 localcred = 1;
840 } else {
841 cred = ap->a_cred;
842 }
843 if (vp->v_type == VREG) {
844 /*
845 * Examine and clean dirty pages, regardless of NMODIFIED.
846 * This closes a major hole in close-to-open consistency.
847 * We want to push out all dirty pages (and buffers) on
848 * close, regardless of whether they were dirtied by
849 * mmap'ed writes or via write().
850 */
851 if (nfs_clean_pages_on_close && vp->v_object) {
852 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
853 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
854 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK)
855 return (EBADF);
856 }
857 VM_OBJECT_WLOCK(vp->v_object);
858 vm_object_page_clean(vp->v_object, 0, 0, 0);
859 VM_OBJECT_WUNLOCK(vp->v_object);
860 }
861 NFSLOCKNODE(np);
862 if (np->n_flag & NMODIFIED) {
863 NFSUNLOCKNODE(np);
864 if (NFS_ISV3(vp)) {
865 /*
866 * Under NFSv3 we have dirty buffers to dispose of. We
867 * must flush them to the NFS server. We have the option
868 * of waiting all the way through the commit rpc or just
869 * waiting for the initial write. The default is to only
870 * wait through the initial write so the data is in the
871 * server's cache, which is roughly similar to the state
872 * a standard disk subsystem leaves the file in on close().
873 *
874 * We cannot clear the NMODIFIED bit in np->n_flag due to
875 * potential races with other processes, and certainly
876 * cannot clear it if we don't commit.
877 * These races occur when there is no longer the old
878 * traditional vnode locking implemented for Vnode Ops.
879 */
880 int cm = newnfs_commit_on_close ? 1 : 0;
881 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
882 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
883 if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK)
884 return (EBADF);
885 }
886 error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
887 /* np->n_flag &= ~NMODIFIED; */
888 } else if (NFS_ISV4(vp)) {
889 if (nfscl_mustflush(vp) != 0) {
890 int cm = newnfs_commit_on_close ? 1 : 0;
891 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
892 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
893 if (VN_IS_DOOMED(vp) && ap->a_fflag !=
894 FNONBLOCK)
895 return (EBADF);
896 }
897 error = ncl_flush(vp, MNT_WAIT, ap->a_td,
898 cm, 0);
899 /*
900 * as above w.r.t races when clearing
901 * NMODIFIED.
902 * np->n_flag &= ~NMODIFIED;
903 */
904 }
905 } else {
906 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
907 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
908 if (VN_IS_DOOMED(vp) && ap->a_fflag !=
909 FNONBLOCK)
910 return (EBADF);
911 }
912 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
913 }
914 NFSLOCKNODE(np);
915 }
916 /*
917 * Invalidate the attribute cache in all cases.
918 * An open is going to fetch fresh attrs any way, other procs
919 * on this node that have file open will be forced to do an
920 * otw attr fetch, but this is safe.
921 * --> A user found that their RPC count dropped by 20% when
922 * this was commented out and I can't see any requirement
923 * for it, so I've disabled it when negative lookups are
924 * enabled. (What does this have to do with negative lookup
925 * caching? Well nothing, except it was reported by the
926 * same user that needed negative lookup caching and I wanted
927 * there to be a way to disable it to see if it
928 * is the cause of some caching/coherency issue that might
929 * crop up.)
930 */
931 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
932 np->n_attrstamp = 0;
933 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
934 }
935 if (np->n_flag & NWRITEERR) {
936 np->n_flag &= ~NWRITEERR;
937 error = np->n_error;
938 }
939 NFSUNLOCKNODE(np);
940 }
941
942 if (NFS_ISV4(vp)) {
943 /*
944 * Get attributes so "change" is up to date.
945 */
946 if (error == 0 && nfscl_mustflush(vp) != 0 &&
947 vp->v_type == VREG &&
948 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
949 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
950 NULL);
951 if (!ret) {
952 np->n_change = nfsva.na_filerev;
953 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
954 NULL, 0, 0);
955 }
956 }
957
958 /*
959 * and do the close.
960 */
961 ret = nfsrpc_close(vp, 0, ap->a_td);
962 if (!error && ret)
963 error = ret;
964 if (error)
965 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
966 (gid_t)0);
967 }
968 if (newnfs_directio_enable)
969 KASSERT((np->n_directio_asyncwr == 0),
970 ("nfs_close: dirty unflushed (%d) directio buffers\n",
971 np->n_directio_asyncwr));
972 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
973 NFSLOCKNODE(np);
974 KASSERT((np->n_directio_opens > 0),
975 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
976 np->n_directio_opens--;
977 if (np->n_directio_opens == 0)
978 np->n_flag &= ~NNONCACHE;
979 NFSUNLOCKNODE(np);
980 }
981 if (localcred)
982 NFSFREECRED(cred);
983 return (error);
984 }
985
986 /*
987 * nfs getattr call from vfs.
988 */
989 static int
990 nfs_getattr(struct vop_getattr_args *ap)
991 {
992 struct vnode *vp = ap->a_vp;
993 struct thread *td = curthread; /* XXX */
994 struct nfsnode *np = VTONFS(vp);
995 int error = 0;
996 struct nfsvattr nfsva;
997 struct vattr *vap = ap->a_vap;
998 struct vattr vattr;
999
1000 /*
1001 * Update local times for special files.
1002 */
1003 NFSLOCKNODE(np);
1004 if (np->n_flag & (NACC | NUPD))
1005 np->n_flag |= NCHG;
1006 NFSUNLOCKNODE(np);
1007 /*
1008 * First look in the cache.
1009 */
1010 if (ncl_getattrcache(vp, &vattr) == 0) {
1011 ncl_copy_vattr(vap, &vattr);
1012
1013 /*
1014 * Get the local modify time for the case of a write
1015 * delegation.
1016 */
1017 nfscl_deleggetmodtime(vp, &vap->va_mtime);
1018 return (0);
1019 }
1020
1021 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
1022 nfsaccess_cache_timeout > 0) {
1023 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
1024 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
1025 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
1026 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
1027 return (0);
1028 }
1029 }
1030 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
1031 if (!error)
1032 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
1033 if (!error) {
1034 /*
1035 * Get the local modify time for the case of a write
1036 * delegation.
1037 */
1038 nfscl_deleggetmodtime(vp, &vap->va_mtime);
1039 } else if (NFS_ISV4(vp)) {
1040 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1041 }
1042 return (error);
1043 }
1044
1045 /*
1046 * nfs setattr call.
1047 */
1048 static int
1049 nfs_setattr(struct vop_setattr_args *ap)
1050 {
1051 struct vnode *vp = ap->a_vp;
1052 struct nfsnode *np = VTONFS(vp);
1053 struct thread *td = curthread; /* XXX */
1054 struct vattr *vap = ap->a_vap;
1055 int error = 0;
1056 u_quad_t tsize;
1057 struct timespec ts;
1058
1059 #ifndef nolint
1060 tsize = (u_quad_t)0;
1061 #endif
1062
1063 /*
1064 * Setting of flags and marking of atimes are not supported.
1065 */
1066 if (vap->va_flags != VNOVAL)
1067 return (EOPNOTSUPP);
1068
1069 /*
1070 * Disallow write attempts if the filesystem is mounted read-only.
1071 */
1072 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
1073 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
1074 vap->va_mtime.tv_sec != VNOVAL ||
1075 vap->va_birthtime.tv_sec != VNOVAL ||
1076 vap->va_mode != (mode_t)VNOVAL) &&
1077 (vp->v_mount->mnt_flag & MNT_RDONLY))
1078 return (EROFS);
1079 if (vap->va_size != VNOVAL) {
1080 switch (vp->v_type) {
1081 case VDIR:
1082 return (EISDIR);
1083 case VCHR:
1084 case VBLK:
1085 case VSOCK:
1086 case VFIFO:
1087 if (vap->va_mtime.tv_sec == VNOVAL &&
1088 vap->va_atime.tv_sec == VNOVAL &&
1089 vap->va_birthtime.tv_sec == VNOVAL &&
1090 vap->va_mode == (mode_t)VNOVAL &&
1091 vap->va_uid == (uid_t)VNOVAL &&
1092 vap->va_gid == (gid_t)VNOVAL)
1093 return (0);
1094 vap->va_size = VNOVAL;
1095 break;
1096 default:
1097 /*
1098 * Disallow write attempts if the filesystem is
1099 * mounted read-only.
1100 */
1101 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1102 return (EROFS);
1103 /*
1104 * We run vnode_pager_setsize() early (why?),
1105 * we must set np->n_size now to avoid vinvalbuf
1106 * V_SAVE races that might setsize a lower
1107 * value.
1108 */
1109 NFSLOCKNODE(np);
1110 tsize = np->n_size;
1111 NFSUNLOCKNODE(np);
1112 error = ncl_meta_setsize(vp, td, vap->va_size);
1113 NFSLOCKNODE(np);
1114 if (np->n_flag & NMODIFIED) {
1115 tsize = np->n_size;
1116 NFSUNLOCKNODE(np);
1117 error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
1118 0 : V_SAVE, td, 1);
1119 if (error != 0) {
1120 vnode_pager_setsize(vp, tsize);
1121 return (error);
1122 }
1123 /*
1124 * Call nfscl_delegmodtime() to set the modify time
1125 * locally, as required.
1126 */
1127 nfscl_delegmodtime(vp);
1128 } else
1129 NFSUNLOCKNODE(np);
1130 /*
1131 * np->n_size has already been set to vap->va_size
1132 * in ncl_meta_setsize(). We must set it again since
1133 * nfs_loadattrcache() could be called through
1134 * ncl_meta_setsize() and could modify np->n_size.
1135 */
1136 NFSLOCKNODE(np);
1137 np->n_vattr.na_size = np->n_size = vap->va_size;
1138 NFSUNLOCKNODE(np);
1139 }
1140 } else {
1141 NFSLOCKNODE(np);
1142 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
1143 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
1144 NFSUNLOCKNODE(np);
1145 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
1146 if (error == EINTR || error == EIO)
1147 return (error);
1148 } else
1149 NFSUNLOCKNODE(np);
1150 }
1151 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
1152 if (vap->va_size != VNOVAL) {
1153 if (error == 0) {
1154 nanouptime(&ts);
1155 NFSLOCKNODE(np);
1156 np->n_localmodtime = ts;
1157 NFSUNLOCKNODE(np);
1158 } else {
1159 NFSLOCKNODE(np);
1160 np->n_size = np->n_vattr.na_size = tsize;
1161 vnode_pager_setsize(vp, tsize);
1162 NFSUNLOCKNODE(np);
1163 }
1164 }
1165 return (error);
1166 }
1167
1168 /*
1169 * Do an nfs setattr rpc.
1170 */
1171 static int
1172 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1173 struct thread *td)
1174 {
1175 struct nfsnode *np = VTONFS(vp);
1176 int error, ret, attrflag, i;
1177 struct nfsvattr nfsva;
1178
1179 if (NFS_ISV34(vp)) {
1180 NFSLOCKNODE(np);
1181 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1182 np->n_accesscache[i].stamp = 0;
1183 np->n_flag |= NDELEGMOD;
1184 NFSUNLOCKNODE(np);
1185 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1186 }
1187 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1188 NULL);
1189 if (attrflag) {
1190 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 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 = cnp->cn_thread;
1217 struct nfsfh *nfhp;
1218 struct nfsvattr dnfsva, nfsva;
1219 struct vattr vattr;
1220 struct timespec nctime, ts;
1221
1222 *vpp = NULLVP;
1223 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1224 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1225 return (EROFS);
1226 if (dvp->v_type != VDIR)
1227 return (ENOTDIR);
1228 nmp = VFSTONFS(mp);
1229 np = VTONFS(dvp);
1230
1231 /* For NFSv4, wait until any remove is done. */
1232 NFSLOCKNODE(np);
1233 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1234 np->n_flag |= NREMOVEWANT;
1235 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1236 }
1237 NFSUNLOCKNODE(np);
1238
1239 error = vn_dir_check_exec(dvp, cnp);
1240 if (error != 0)
1241 return (error);
1242 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1243 if (error > 0 && error != ENOENT)
1244 return (error);
1245 if (error == -1) {
1246 /*
1247 * Lookups of "." are special and always return the
1248 * current directory. cache_lookup() already handles
1249 * associated locking bookkeeping, etc.
1250 */
1251 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1252 /* XXX: Is this really correct? */
1253 if (cnp->cn_nameiop != LOOKUP &&
1254 (flags & ISLASTCN))
1255 cnp->cn_flags |= SAVENAME;
1256 return (0);
1257 }
1258
1259 /*
1260 * We only accept a positive hit in the cache if the
1261 * change time of the file matches our cached copy.
1262 * Otherwise, we discard the cache entry and fallback
1263 * to doing a lookup RPC. We also only trust cache
1264 * entries for less than nm_nametimeo seconds.
1265 *
1266 * To better handle stale file handles and attributes,
1267 * clear the attribute cache of this node if it is a
1268 * leaf component, part of an open() call, and not
1269 * locally modified before fetching the attributes.
1270 * This should allow stale file handles to be detected
1271 * here where we can fall back to a LOOKUP RPC to
1272 * recover rather than having nfs_open() detect the
1273 * stale file handle and failing open(2) with ESTALE.
1274 */
1275 newvp = *vpp;
1276 newnp = VTONFS(newvp);
1277 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1278 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1279 !(newnp->n_flag & NMODIFIED)) {
1280 NFSLOCKNODE(newnp);
1281 newnp->n_attrstamp = 0;
1282 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1283 NFSUNLOCKNODE(newnp);
1284 }
1285 if (nfscl_nodeleg(newvp, 0) == 0 ||
1286 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1287 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1288 timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1289 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1290 if (cnp->cn_nameiop != LOOKUP &&
1291 (flags & ISLASTCN))
1292 cnp->cn_flags |= SAVENAME;
1293 return (0);
1294 }
1295 cache_purge(newvp);
1296 if (dvp != newvp)
1297 vput(newvp);
1298 else
1299 vrele(newvp);
1300 *vpp = NULLVP;
1301 } else if (error == ENOENT) {
1302 if (VN_IS_DOOMED(dvp))
1303 return (ENOENT);
1304 /*
1305 * We only accept a negative hit in the cache if the
1306 * modification time of the parent directory matches
1307 * the cached copy in the name cache entry.
1308 * Otherwise, we discard all of the negative cache
1309 * entries for this directory. We also only trust
1310 * negative cache entries for up to nm_negnametimeo
1311 * seconds.
1312 */
1313 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1314 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1315 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1316 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1317 return (ENOENT);
1318 }
1319 cache_purge_negative(dvp);
1320 }
1321
1322 newvp = NULLVP;
1323 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1324 nanouptime(&ts);
1325 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1326 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1327 NULL);
1328 if (dattrflag)
1329 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1330 if (error) {
1331 if (newvp != NULLVP) {
1332 vput(newvp);
1333 *vpp = NULLVP;
1334 }
1335
1336 if (error != ENOENT) {
1337 if (NFS_ISV4(dvp))
1338 error = nfscl_maperr(td, error, (uid_t)0,
1339 (gid_t)0);
1340 return (error);
1341 }
1342
1343 /* The requested file was not found. */
1344 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1345 (flags & ISLASTCN)) {
1346 /*
1347 * XXX: UFS does a full VOP_ACCESS(dvp,
1348 * VWRITE) here instead of just checking
1349 * MNT_RDONLY.
1350 */
1351 if (mp->mnt_flag & MNT_RDONLY)
1352 return (EROFS);
1353 cnp->cn_flags |= SAVENAME;
1354 return (EJUSTRETURN);
1355 }
1356
1357 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1358 /*
1359 * Cache the modification time of the parent
1360 * directory from the post-op attributes in
1361 * the name cache entry. The negative cache
1362 * entry will be ignored once the directory
1363 * has changed. Don't bother adding the entry
1364 * if the directory has already changed.
1365 */
1366 NFSLOCKNODE(np);
1367 if (timespeccmp(&np->n_vattr.na_mtime,
1368 &dnfsva.na_mtime, ==)) {
1369 NFSUNLOCKNODE(np);
1370 cache_enter_time(dvp, NULL, cnp,
1371 &dnfsva.na_mtime, NULL);
1372 } else
1373 NFSUNLOCKNODE(np);
1374 }
1375 return (ENOENT);
1376 }
1377
1378 /*
1379 * Handle RENAME case...
1380 */
1381 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1382 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1383 free(nfhp, M_NFSFH);
1384 return (EISDIR);
1385 }
1386 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1387 LK_EXCLUSIVE);
1388 if (error)
1389 return (error);
1390 newvp = NFSTOV(np);
1391 /*
1392 * If n_localmodtime >= time before RPC, then
1393 * a file modification operation, such as
1394 * VOP_SETATTR() of size, has occurred while
1395 * the Lookup RPC and acquisition of the vnode
1396 * happened. As such, the attributes might
1397 * be stale, with possibly an incorrect size.
1398 */
1399 NFSLOCKNODE(np);
1400 if (timespecisset(&np->n_localmodtime) &&
1401 timespeccmp(&np->n_localmodtime, &ts, >=)) {
1402 NFSCL_DEBUG(4, "nfs_lookup: rename localmod "
1403 "stale attributes\n");
1404 attrflag = 0;
1405 }
1406 NFSUNLOCKNODE(np);
1407 if (attrflag)
1408 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1409 0, 1);
1410 *vpp = newvp;
1411 cnp->cn_flags |= SAVENAME;
1412 return (0);
1413 }
1414
1415 if (flags & ISDOTDOT) {
1416 ltype = NFSVOPISLOCKED(dvp);
1417 error = vfs_busy(mp, MBF_NOWAIT);
1418 if (error != 0) {
1419 vfs_ref(mp);
1420 NFSVOPUNLOCK(dvp);
1421 error = vfs_busy(mp, 0);
1422 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1423 vfs_rel(mp);
1424 if (error == 0 && VN_IS_DOOMED(dvp)) {
1425 vfs_unbusy(mp);
1426 error = ENOENT;
1427 }
1428 if (error != 0)
1429 return (error);
1430 }
1431 NFSVOPUNLOCK(dvp);
1432 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1433 cnp->cn_lkflags);
1434 if (error == 0)
1435 newvp = NFSTOV(np);
1436 vfs_unbusy(mp);
1437 if (newvp != dvp)
1438 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1439 if (VN_IS_DOOMED(dvp)) {
1440 if (error == 0) {
1441 if (newvp == dvp)
1442 vrele(newvp);
1443 else
1444 vput(newvp);
1445 }
1446 error = ENOENT;
1447 }
1448 if (error != 0)
1449 return (error);
1450 if (attrflag)
1451 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1452 0, 1);
1453 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1454 free(nfhp, M_NFSFH);
1455 VREF(dvp);
1456 newvp = dvp;
1457 if (attrflag)
1458 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1459 0, 1);
1460 } else {
1461 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1462 cnp->cn_lkflags);
1463 if (error)
1464 return (error);
1465 newvp = NFSTOV(np);
1466 /*
1467 * If n_localmodtime >= time before RPC, then
1468 * a file modification operation, such as
1469 * VOP_SETATTR() of size, has occurred while
1470 * the Lookup RPC and acquisition of the vnode
1471 * happened. As such, the attributes might
1472 * be stale, with possibly an incorrect size.
1473 */
1474 NFSLOCKNODE(np);
1475 if (timespecisset(&np->n_localmodtime) &&
1476 timespeccmp(&np->n_localmodtime, &ts, >=)) {
1477 NFSCL_DEBUG(4, "nfs_lookup: localmod "
1478 "stale attributes\n");
1479 attrflag = 0;
1480 }
1481 NFSUNLOCKNODE(np);
1482 if (attrflag)
1483 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1484 0, 1);
1485 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1486 !(np->n_flag & NMODIFIED)) {
1487 /*
1488 * Flush the attribute cache when opening a
1489 * leaf node to ensure that fresh attributes
1490 * are fetched in nfs_open() since we did not
1491 * fetch attributes from the LOOKUP reply.
1492 */
1493 NFSLOCKNODE(np);
1494 np->n_attrstamp = 0;
1495 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1496 NFSUNLOCKNODE(np);
1497 }
1498 }
1499 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1500 cnp->cn_flags |= SAVENAME;
1501 if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp &&
1502 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1503 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1504 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1505 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1506 *vpp = newvp;
1507 return (0);
1508 }
1509
1510 /*
1511 * nfs read call.
1512 * Just call ncl_bioread() to do the work.
1513 */
1514 static int
1515 nfs_read(struct vop_read_args *ap)
1516 {
1517 struct vnode *vp = ap->a_vp;
1518
1519 switch (vp->v_type) {
1520 case VREG:
1521 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1522 case VDIR:
1523 return (EISDIR);
1524 default:
1525 return (EOPNOTSUPP);
1526 }
1527 }
1528
1529 /*
1530 * nfs readlink call
1531 */
1532 static int
1533 nfs_readlink(struct vop_readlink_args *ap)
1534 {
1535 struct vnode *vp = ap->a_vp;
1536
1537 if (vp->v_type != VLNK)
1538 return (EINVAL);
1539 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1540 }
1541
1542 /*
1543 * Do a readlink rpc.
1544 * Called by ncl_doio() from below the buffer cache.
1545 */
1546 int
1547 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1548 {
1549 int error, ret, attrflag;
1550 struct nfsvattr nfsva;
1551
1552 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1553 &attrflag, NULL);
1554 if (attrflag) {
1555 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1556 if (ret && !error)
1557 error = ret;
1558 }
1559 if (error && NFS_ISV4(vp))
1560 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1561 return (error);
1562 }
1563
1564 /*
1565 * nfs read rpc call
1566 * Ditto above
1567 */
1568 int
1569 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1570 {
1571 int error, ret, attrflag;
1572 struct nfsvattr nfsva;
1573 struct nfsmount *nmp;
1574
1575 nmp = VFSTONFS(vp->v_mount);
1576 error = EIO;
1577 attrflag = 0;
1578 if (NFSHASPNFS(nmp))
1579 error = nfscl_doiods(vp, uiop, NULL, NULL,
1580 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1581 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1582 if (error != 0)
1583 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1584 &attrflag, NULL);
1585 if (attrflag) {
1586 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1587 if (ret && !error)
1588 error = ret;
1589 }
1590 if (error && NFS_ISV4(vp))
1591 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1592 return (error);
1593 }
1594
1595 /*
1596 * nfs write call
1597 */
1598 int
1599 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1600 int *iomode, int *must_commit, int called_from_strategy)
1601 {
1602 struct nfsvattr nfsva;
1603 int error, attrflag, ret;
1604 struct nfsmount *nmp;
1605
1606 nmp = VFSTONFS(vp->v_mount);
1607 error = EIO;
1608 attrflag = 0;
1609 if (NFSHASPNFS(nmp))
1610 error = nfscl_doiods(vp, uiop, iomode, must_commit,
1611 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1612 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1613 if (error != 0)
1614 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1615 uiop->uio_td, &nfsva, &attrflag, NULL,
1616 called_from_strategy);
1617 if (attrflag) {
1618 if (VTONFS(vp)->n_flag & ND_NFSV4)
1619 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1620 1);
1621 else
1622 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1623 1);
1624 if (ret && !error)
1625 error = ret;
1626 }
1627 if (DOINGASYNC(vp))
1628 *iomode = NFSWRITE_FILESYNC;
1629 if (error && NFS_ISV4(vp))
1630 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1631 return (error);
1632 }
1633
1634 /*
1635 * nfs mknod rpc
1636 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1637 * mode set to specify the file type and the size field for rdev.
1638 */
1639 static int
1640 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1641 struct vattr *vap)
1642 {
1643 struct nfsvattr nfsva, dnfsva;
1644 struct vnode *newvp = NULL;
1645 struct nfsnode *np = NULL, *dnp;
1646 struct nfsfh *nfhp;
1647 struct vattr vattr;
1648 int error = 0, attrflag, dattrflag;
1649 u_int32_t rdev;
1650
1651 if (vap->va_type == VCHR || vap->va_type == VBLK)
1652 rdev = vap->va_rdev;
1653 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1654 rdev = 0xffffffff;
1655 else
1656 return (EOPNOTSUPP);
1657 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1658 return (error);
1659 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1660 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1661 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1662 if (!error) {
1663 if (!nfhp)
1664 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1665 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1666 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1667 NULL);
1668 if (nfhp)
1669 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1670 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1671 }
1672 if (dattrflag)
1673 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1674 if (!error) {
1675 newvp = NFSTOV(np);
1676 if (attrflag != 0) {
1677 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1678 0, 1);
1679 if (error != 0)
1680 vput(newvp);
1681 }
1682 }
1683 if (!error) {
1684 *vpp = newvp;
1685 } else if (NFS_ISV4(dvp)) {
1686 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1687 vap->va_gid);
1688 }
1689 dnp = VTONFS(dvp);
1690 NFSLOCKNODE(dnp);
1691 dnp->n_flag |= NMODIFIED;
1692 if (!dattrflag) {
1693 dnp->n_attrstamp = 0;
1694 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1695 }
1696 NFSUNLOCKNODE(dnp);
1697 return (error);
1698 }
1699
1700 /*
1701 * nfs mknod vop
1702 * just call nfs_mknodrpc() to do the work.
1703 */
1704 /* ARGSUSED */
1705 static int
1706 nfs_mknod(struct vop_mknod_args *ap)
1707 {
1708 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1709 }
1710
1711 static struct mtx nfs_cverf_mtx;
1712 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1713 MTX_DEF);
1714
1715 static nfsquad_t
1716 nfs_get_cverf(void)
1717 {
1718 static nfsquad_t cverf;
1719 nfsquad_t ret;
1720 static int cverf_initialized = 0;
1721
1722 mtx_lock(&nfs_cverf_mtx);
1723 if (cverf_initialized == 0) {
1724 cverf.lval[0] = arc4random();
1725 cverf.lval[1] = arc4random();
1726 cverf_initialized = 1;
1727 } else
1728 cverf.qval++;
1729 ret = cverf;
1730 mtx_unlock(&nfs_cverf_mtx);
1731
1732 return (ret);
1733 }
1734
1735 /*
1736 * nfs file create call
1737 */
1738 static int
1739 nfs_create(struct vop_create_args *ap)
1740 {
1741 struct vnode *dvp = ap->a_dvp;
1742 struct vattr *vap = ap->a_vap;
1743 struct componentname *cnp = ap->a_cnp;
1744 struct nfsnode *np = NULL, *dnp;
1745 struct vnode *newvp = NULL;
1746 struct nfsmount *nmp;
1747 struct nfsvattr dnfsva, nfsva;
1748 struct nfsfh *nfhp;
1749 nfsquad_t cverf;
1750 int error = 0, attrflag, dattrflag, fmode = 0;
1751 struct vattr vattr;
1752
1753 /*
1754 * Oops, not for me..
1755 */
1756 if (vap->va_type == VSOCK)
1757 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1758
1759 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1760 return (error);
1761 if (vap->va_vaflags & VA_EXCLUSIVE)
1762 fmode |= O_EXCL;
1763 dnp = VTONFS(dvp);
1764 nmp = VFSTONFS(dvp->v_mount);
1765 again:
1766 /* For NFSv4, wait until any remove is done. */
1767 NFSLOCKNODE(dnp);
1768 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1769 dnp->n_flag |= NREMOVEWANT;
1770 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1771 }
1772 NFSUNLOCKNODE(dnp);
1773
1774 cverf = nfs_get_cverf();
1775 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1776 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1777 &nfhp, &attrflag, &dattrflag, NULL);
1778 if (!error) {
1779 if (nfhp == NULL)
1780 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1781 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1782 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1783 NULL);
1784 if (nfhp != NULL)
1785 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1786 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1787 }
1788 if (dattrflag)
1789 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1790 if (!error) {
1791 newvp = NFSTOV(np);
1792 if (attrflag == 0)
1793 error = nfsrpc_getattr(newvp, cnp->cn_cred,
1794 cnp->cn_thread, &nfsva, NULL);
1795 if (error == 0)
1796 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1797 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 cnp->cn_thread, &nfsva, &attrflag, NULL);
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, cnp->cn_thread, &nfsva,
1820 &attrflag, NULL);
1821 }
1822 if (attrflag)
1823 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1824 NULL, 0, 1);
1825 if (error != 0)
1826 vput(newvp);
1827 }
1828 }
1829 if (!error) {
1830 if ((cnp->cn_flags & MAKEENTRY) && attrflag) {
1831 if (dvp != newvp)
1832 cache_enter_time(dvp, newvp, cnp,
1833 &nfsva.na_ctime, NULL);
1834 else
1835 printf("nfs_create: bogus NFS server returned "
1836 "the directory as the new file object\n");
1837 }
1838 *ap->a_vpp = newvp;
1839 } else if (NFS_ISV4(dvp)) {
1840 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1841 vap->va_gid);
1842 }
1843 NFSLOCKNODE(dnp);
1844 dnp->n_flag |= NMODIFIED;
1845 if (!dattrflag) {
1846 dnp->n_attrstamp = 0;
1847 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1848 }
1849 NFSUNLOCKNODE(dnp);
1850 return (error);
1851 }
1852
1853 /*
1854 * nfs file remove call
1855 * To try and make nfs semantics closer to ufs semantics, a file that has
1856 * other processes using the vnode is renamed instead of removed and then
1857 * removed later on the last close.
1858 * - If v_usecount > 1
1859 * If a rename is not already in the works
1860 * call nfs_sillyrename() to set it up
1861 * else
1862 * do the remove rpc
1863 */
1864 static int
1865 nfs_remove(struct vop_remove_args *ap)
1866 {
1867 struct vnode *vp = ap->a_vp;
1868 struct vnode *dvp = ap->a_dvp;
1869 struct componentname *cnp = ap->a_cnp;
1870 struct nfsnode *np = VTONFS(vp);
1871 int error = 0;
1872 struct vattr vattr;
1873
1874 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1875 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1876 if (vp->v_type == VDIR)
1877 error = EPERM;
1878 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1879 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1880 vattr.va_nlink > 1)) {
1881 /*
1882 * Purge the name cache so that the chance of a lookup for
1883 * the name succeeding while the remove is in progress is
1884 * minimized. Without node locking it can still happen, such
1885 * that an I/O op returns ESTALE, but since you get this if
1886 * another host removes the file..
1887 */
1888 cache_purge(vp);
1889 /*
1890 * throw away biocache buffers, mainly to avoid
1891 * unnecessary delayed writes later.
1892 */
1893 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1894 if (error != EINTR && error != EIO)
1895 /* Do the rpc */
1896 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1897 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1898 /*
1899 * Kludge City: If the first reply to the remove rpc is lost..
1900 * the reply to the retransmitted request will be ENOENT
1901 * since the file was in fact removed
1902 * Therefore, we cheat and return success.
1903 */
1904 if (error == ENOENT)
1905 error = 0;
1906 } else if (!np->n_sillyrename)
1907 error = nfs_sillyrename(dvp, vp, cnp);
1908 NFSLOCKNODE(np);
1909 np->n_attrstamp = 0;
1910 NFSUNLOCKNODE(np);
1911 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1912 return (error);
1913 }
1914
1915 /*
1916 * nfs file remove rpc called from nfs_inactive
1917 */
1918 int
1919 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1920 {
1921 /*
1922 * Make sure that the directory vnode is still valid.
1923 * XXX we should lock sp->s_dvp here.
1924 */
1925 if (sp->s_dvp->v_type == VBAD)
1926 return (0);
1927 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1928 sp->s_cred, NULL));
1929 }
1930
1931 /*
1932 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1933 */
1934 static int
1935 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1936 int namelen, struct ucred *cred, struct thread *td)
1937 {
1938 struct nfsvattr dnfsva;
1939 struct nfsnode *dnp = VTONFS(dvp);
1940 int error = 0, dattrflag;
1941
1942 NFSLOCKNODE(dnp);
1943 dnp->n_flag |= NREMOVEINPROG;
1944 NFSUNLOCKNODE(dnp);
1945 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1946 &dattrflag, NULL);
1947 NFSLOCKNODE(dnp);
1948 if ((dnp->n_flag & NREMOVEWANT)) {
1949 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1950 NFSUNLOCKNODE(dnp);
1951 wakeup((caddr_t)dnp);
1952 } else {
1953 dnp->n_flag &= ~NREMOVEINPROG;
1954 NFSUNLOCKNODE(dnp);
1955 }
1956 if (dattrflag)
1957 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1958 NFSLOCKNODE(dnp);
1959 dnp->n_flag |= NMODIFIED;
1960 if (!dattrflag) {
1961 dnp->n_attrstamp = 0;
1962 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1963 }
1964 NFSUNLOCKNODE(dnp);
1965 if (error && NFS_ISV4(dvp))
1966 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1967 return (error);
1968 }
1969
1970 /*
1971 * nfs file rename call
1972 */
1973 static int
1974 nfs_rename(struct vop_rename_args *ap)
1975 {
1976 struct vnode *fvp = ap->a_fvp;
1977 struct vnode *tvp = ap->a_tvp;
1978 struct vnode *fdvp = ap->a_fdvp;
1979 struct vnode *tdvp = ap->a_tdvp;
1980 struct componentname *tcnp = ap->a_tcnp;
1981 struct componentname *fcnp = ap->a_fcnp;
1982 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1983 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1984 struct nfsv4node *newv4 = NULL;
1985 int error;
1986
1987 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1988 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1989 /* Check for cross-device rename */
1990 if ((fvp->v_mount != tdvp->v_mount) ||
1991 (tvp && (fvp->v_mount != tvp->v_mount))) {
1992 error = EXDEV;
1993 goto out;
1994 }
1995
1996 if (fvp == tvp) {
1997 printf("nfs_rename: fvp == tvp (can't happen)\n");
1998 error = 0;
1999 goto out;
2000 }
2001 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
2002 goto out;
2003
2004 /*
2005 * We have to flush B_DELWRI data prior to renaming
2006 * the file. If we don't, the delayed-write buffers
2007 * can be flushed out later after the file has gone stale
2008 * under NFSV3. NFSV2 does not have this problem because
2009 * ( as far as I can tell ) it flushes dirty buffers more
2010 * often.
2011 *
2012 * Skip the rename operation if the fsync fails, this can happen
2013 * due to the server's volume being full, when we pushed out data
2014 * that was written back to our cache earlier. Not checking for
2015 * this condition can result in potential (silent) data loss.
2016 */
2017 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
2018 NFSVOPUNLOCK(fvp);
2019 if (!error && tvp)
2020 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
2021 if (error)
2022 goto out;
2023
2024 /*
2025 * If the tvp exists and is in use, sillyrename it before doing the
2026 * rename of the new file over it.
2027 * XXX Can't sillyrename a directory.
2028 */
2029 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
2030 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
2031 vput(tvp);
2032 tvp = NULL;
2033 }
2034
2035 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
2036 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
2037 tcnp->cn_thread);
2038
2039 if (error == 0 && NFS_ISV4(tdvp)) {
2040 /*
2041 * For NFSv4, check to see if it is the same name and
2042 * replace the name, if it is different.
2043 */
2044 newv4 = malloc(
2045 sizeof (struct nfsv4node) +
2046 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
2047 M_NFSV4NODE, M_WAITOK);
2048 NFSLOCKNODE(tdnp);
2049 NFSLOCKNODE(fnp);
2050 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
2051 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
2052 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
2053 tcnp->cn_namelen) ||
2054 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
2055 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
2056 tdnp->n_fhp->nfh_len))) {
2057 #ifdef notdef
2058 { char nnn[100]; int nnnl;
2059 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
2060 bcopy(tcnp->cn_nameptr, nnn, nnnl);
2061 nnn[nnnl] = '\0';
2062 printf("ren replace=%s\n",nnn);
2063 }
2064 #endif
2065 free(fnp->n_v4, M_NFSV4NODE);
2066 fnp->n_v4 = newv4;
2067 newv4 = NULL;
2068 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
2069 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
2070 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
2071 tdnp->n_fhp->nfh_len);
2072 NFSBCOPY(tcnp->cn_nameptr,
2073 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
2074 }
2075 NFSUNLOCKNODE(tdnp);
2076 NFSUNLOCKNODE(fnp);
2077 if (newv4 != NULL)
2078 free(newv4, M_NFSV4NODE);
2079 }
2080
2081 if (fvp->v_type == VDIR) {
2082 if (tvp != NULL && tvp->v_type == VDIR)
2083 cache_purge(tdvp);
2084 cache_purge(fdvp);
2085 }
2086
2087 out:
2088 if (tdvp == tvp)
2089 vrele(tdvp);
2090 else
2091 vput(tdvp);
2092 if (tvp)
2093 vput(tvp);
2094 vrele(fdvp);
2095 vrele(fvp);
2096 /*
2097 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2098 */
2099 if (error == ENOENT)
2100 error = 0;
2101 return (error);
2102 }
2103
2104 /*
2105 * nfs file rename rpc called from nfs_remove() above
2106 */
2107 static int
2108 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
2109 struct sillyrename *sp)
2110 {
2111
2112 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
2113 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
2114 scnp->cn_thread));
2115 }
2116
2117 /*
2118 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2119 */
2120 static int
2121 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
2122 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
2123 int tnamelen, struct ucred *cred, struct thread *td)
2124 {
2125 struct nfsvattr fnfsva, tnfsva;
2126 struct nfsnode *fdnp = VTONFS(fdvp);
2127 struct nfsnode *tdnp = VTONFS(tdvp);
2128 int error = 0, fattrflag, tattrflag;
2129
2130 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
2131 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
2132 &tattrflag, NULL, NULL);
2133 NFSLOCKNODE(fdnp);
2134 fdnp->n_flag |= NMODIFIED;
2135 if (fattrflag != 0) {
2136 NFSUNLOCKNODE(fdnp);
2137 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
2138 } else {
2139 fdnp->n_attrstamp = 0;
2140 NFSUNLOCKNODE(fdnp);
2141 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
2142 }
2143 NFSLOCKNODE(tdnp);
2144 tdnp->n_flag |= NMODIFIED;
2145 if (tattrflag != 0) {
2146 NFSUNLOCKNODE(tdnp);
2147 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
2148 } else {
2149 tdnp->n_attrstamp = 0;
2150 NFSUNLOCKNODE(tdnp);
2151 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2152 }
2153 if (error && NFS_ISV4(fdvp))
2154 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2155 return (error);
2156 }
2157
2158 /*
2159 * nfs hard link create call
2160 */
2161 static int
2162 nfs_link(struct vop_link_args *ap)
2163 {
2164 struct vnode *vp = ap->a_vp;
2165 struct vnode *tdvp = ap->a_tdvp;
2166 struct componentname *cnp = ap->a_cnp;
2167 struct nfsnode *np, *tdnp;
2168 struct nfsvattr nfsva, dnfsva;
2169 int error = 0, attrflag, dattrflag;
2170
2171 /*
2172 * Push all writes to the server, so that the attribute cache
2173 * doesn't get "out of sync" with the server.
2174 * XXX There should be a better way!
2175 */
2176 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
2177
2178 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2179 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
2180 &dattrflag, NULL);
2181 tdnp = VTONFS(tdvp);
2182 NFSLOCKNODE(tdnp);
2183 tdnp->n_flag |= NMODIFIED;
2184 if (dattrflag != 0) {
2185 NFSUNLOCKNODE(tdnp);
2186 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
2187 } else {
2188 tdnp->n_attrstamp = 0;
2189 NFSUNLOCKNODE(tdnp);
2190 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2191 }
2192 if (attrflag)
2193 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2194 else {
2195 np = VTONFS(vp);
2196 NFSLOCKNODE(np);
2197 np->n_attrstamp = 0;
2198 NFSUNLOCKNODE(np);
2199 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2200 }
2201 /*
2202 * If negative lookup caching is enabled, I might as well
2203 * add an entry for this node. Not necessary for correctness,
2204 * but if negative caching is enabled, then the system
2205 * must care about lookup caching hit rate, so...
2206 */
2207 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2208 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2209 if (tdvp != vp)
2210 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2211 else
2212 printf("nfs_link: bogus NFS server returned "
2213 "the directory as the new link\n");
2214 }
2215 if (error && NFS_ISV4(vp))
2216 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2217 (gid_t)0);
2218 return (error);
2219 }
2220
2221 /*
2222 * nfs symbolic link create call
2223 */
2224 static int
2225 nfs_symlink(struct vop_symlink_args *ap)
2226 {
2227 struct vnode *dvp = ap->a_dvp;
2228 struct vattr *vap = ap->a_vap;
2229 struct componentname *cnp = ap->a_cnp;
2230 struct nfsvattr nfsva, dnfsva;
2231 struct nfsfh *nfhp;
2232 struct nfsnode *np = NULL, *dnp;
2233 struct vnode *newvp = NULL;
2234 int error = 0, attrflag, dattrflag, ret;
2235
2236 vap->va_type = VLNK;
2237 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2238 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2239 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2240 if (nfhp) {
2241 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2242 &np, NULL, LK_EXCLUSIVE);
2243 if (!ret)
2244 newvp = NFSTOV(np);
2245 else if (!error)
2246 error = ret;
2247 }
2248 if (newvp != NULL) {
2249 if (attrflag)
2250 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2251 0, 1);
2252 } else if (!error) {
2253 /*
2254 * If we do not have an error and we could not extract the
2255 * newvp from the response due to the request being NFSv2, we
2256 * have to do a lookup in order to obtain a newvp to return.
2257 */
2258 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2259 cnp->cn_cred, cnp->cn_thread, &np);
2260 if (!error)
2261 newvp = NFSTOV(np);
2262 }
2263 if (error) {
2264 if (newvp)
2265 vput(newvp);
2266 if (NFS_ISV4(dvp))
2267 error = nfscl_maperr(cnp->cn_thread, error,
2268 vap->va_uid, vap->va_gid);
2269 } else {
2270 *ap->a_vpp = newvp;
2271 }
2272
2273 dnp = VTONFS(dvp);
2274 NFSLOCKNODE(dnp);
2275 dnp->n_flag |= NMODIFIED;
2276 if (dattrflag != 0) {
2277 NFSUNLOCKNODE(dnp);
2278 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2279 } else {
2280 dnp->n_attrstamp = 0;
2281 NFSUNLOCKNODE(dnp);
2282 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2283 }
2284 /*
2285 * If negative lookup caching is enabled, I might as well
2286 * add an entry for this node. Not necessary for correctness,
2287 * but if negative caching is enabled, then the system
2288 * must care about lookup caching hit rate, so...
2289 */
2290 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2291 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2292 if (dvp != newvp)
2293 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2294 NULL);
2295 else
2296 printf("nfs_symlink: bogus NFS server returned "
2297 "the directory as the new file object\n");
2298 }
2299 return (error);
2300 }
2301
2302 /*
2303 * nfs make dir call
2304 */
2305 static int
2306 nfs_mkdir(struct vop_mkdir_args *ap)
2307 {
2308 struct vnode *dvp = ap->a_dvp;
2309 struct vattr *vap = ap->a_vap;
2310 struct componentname *cnp = ap->a_cnp;
2311 struct nfsnode *np = NULL, *dnp;
2312 struct vnode *newvp = NULL;
2313 struct vattr vattr;
2314 struct nfsfh *nfhp;
2315 struct nfsvattr nfsva, dnfsva;
2316 int error = 0, attrflag, dattrflag, ret;
2317
2318 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2319 return (error);
2320 vap->va_type = VDIR;
2321 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2322 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2323 &attrflag, &dattrflag, NULL);
2324 dnp = VTONFS(dvp);
2325 NFSLOCKNODE(dnp);
2326 dnp->n_flag |= NMODIFIED;
2327 if (dattrflag != 0) {
2328 NFSUNLOCKNODE(dnp);
2329 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2330 } else {
2331 dnp->n_attrstamp = 0;
2332 NFSUNLOCKNODE(dnp);
2333 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2334 }
2335 if (nfhp) {
2336 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2337 &np, NULL, LK_EXCLUSIVE);
2338 if (!ret) {
2339 newvp = NFSTOV(np);
2340 if (attrflag)
2341 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2342 NULL, 0, 1);
2343 } else if (!error)
2344 error = ret;
2345 }
2346 if (!error && newvp == NULL) {
2347 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2348 cnp->cn_cred, cnp->cn_thread, &np);
2349 if (!error) {
2350 newvp = NFSTOV(np);
2351 if (newvp->v_type != VDIR)
2352 error = EEXIST;
2353 }
2354 }
2355 if (error) {
2356 if (newvp)
2357 vput(newvp);
2358 if (NFS_ISV4(dvp))
2359 error = nfscl_maperr(cnp->cn_thread, error,
2360 vap->va_uid, vap->va_gid);
2361 } else {
2362 /*
2363 * If negative lookup caching is enabled, I might as well
2364 * add an entry for this node. Not necessary for correctness,
2365 * but if negative caching is enabled, then the system
2366 * must care about lookup caching hit rate, so...
2367 */
2368 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2369 (cnp->cn_flags & MAKEENTRY) &&
2370 attrflag != 0 && dattrflag != 0) {
2371 if (dvp != newvp)
2372 cache_enter_time(dvp, newvp, cnp,
2373 &nfsva.na_ctime, &dnfsva.na_ctime);
2374 else
2375 printf("nfs_mkdir: bogus NFS server returned "
2376 "the directory that the directory was "
2377 "created in as the new file object\n");
2378 }
2379 *ap->a_vpp = newvp;
2380 }
2381 return (error);
2382 }
2383
2384 /*
2385 * nfs remove directory call
2386 */
2387 static int
2388 nfs_rmdir(struct vop_rmdir_args *ap)
2389 {
2390 struct vnode *vp = ap->a_vp;
2391 struct vnode *dvp = ap->a_dvp;
2392 struct componentname *cnp = ap->a_cnp;
2393 struct nfsnode *dnp;
2394 struct nfsvattr dnfsva;
2395 int error, dattrflag;
2396
2397 if (dvp == vp)
2398 return (EINVAL);
2399 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2400 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2401 dnp = VTONFS(dvp);
2402 NFSLOCKNODE(dnp);
2403 dnp->n_flag |= NMODIFIED;
2404 if (dattrflag != 0) {
2405 NFSUNLOCKNODE(dnp);
2406 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2407 } else {
2408 dnp->n_attrstamp = 0;
2409 NFSUNLOCKNODE(dnp);
2410 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2411 }
2412
2413 cache_purge(dvp);
2414 cache_purge(vp);
2415 if (error && NFS_ISV4(dvp))
2416 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2417 (gid_t)0);
2418 /*
2419 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2420 */
2421 if (error == ENOENT)
2422 error = 0;
2423 return (error);
2424 }
2425
2426 /*
2427 * nfs readdir call
2428 */
2429 static int
2430 nfs_readdir(struct vop_readdir_args *ap)
2431 {
2432 struct vnode *vp = ap->a_vp;
2433 struct nfsnode *np = VTONFS(vp);
2434 struct uio *uio = ap->a_uio;
2435 ssize_t tresid, left;
2436 int error = 0;
2437 struct vattr vattr;
2438
2439 if (ap->a_eofflag != NULL)
2440 *ap->a_eofflag = 0;
2441 if (vp->v_type != VDIR)
2442 return(EPERM);
2443
2444 /*
2445 * First, check for hit on the EOF offset cache
2446 */
2447 NFSLOCKNODE(np);
2448 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2449 (np->n_flag & NMODIFIED) == 0) {
2450 NFSUNLOCKNODE(np);
2451 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2452 NFSLOCKNODE(np);
2453 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2454 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2455 NFSUNLOCKNODE(np);
2456 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2457 if (ap->a_eofflag != NULL)
2458 *ap->a_eofflag = 1;
2459 return (0);
2460 } else
2461 NFSUNLOCKNODE(np);
2462 }
2463 } else
2464 NFSUNLOCKNODE(np);
2465
2466 /*
2467 * NFS always guarantees that directory entries don't straddle
2468 * DIRBLKSIZ boundaries. As such, we need to limit the size
2469 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2470 * directory entry.
2471 */
2472 left = uio->uio_resid % DIRBLKSIZ;
2473 if (left == uio->uio_resid)
2474 return (EINVAL);
2475 uio->uio_resid -= left;
2476
2477 /*
2478 * Call ncl_bioread() to do the real work.
2479 */
2480 tresid = uio->uio_resid;
2481 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2482
2483 if (!error && uio->uio_resid == tresid) {
2484 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2485 if (ap->a_eofflag != NULL)
2486 *ap->a_eofflag = 1;
2487 }
2488
2489 /* Add the partial DIRBLKSIZ (left) back in. */
2490 uio->uio_resid += left;
2491 return (error);
2492 }
2493
2494 /*
2495 * Readdir rpc call.
2496 * Called from below the buffer cache by ncl_doio().
2497 */
2498 int
2499 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2500 struct thread *td)
2501 {
2502 struct nfsvattr nfsva;
2503 nfsuint64 *cookiep, cookie;
2504 struct nfsnode *dnp = VTONFS(vp);
2505 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2506 int error = 0, eof, attrflag;
2507
2508 KASSERT(uiop->uio_iovcnt == 1 &&
2509 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2510 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2511 ("nfs readdirrpc bad uio"));
2512
2513 /*
2514 * If there is no cookie, assume directory was stale.
2515 */
2516 ncl_dircookie_lock(dnp);
2517 NFSUNLOCKNODE(dnp);
2518 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2519 if (cookiep) {
2520 cookie = *cookiep;
2521 ncl_dircookie_unlock(dnp);
2522 } else {
2523 ncl_dircookie_unlock(dnp);
2524 return (NFSERR_BAD_COOKIE);
2525 }
2526
2527 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2528 (void)ncl_fsinfo(nmp, vp, cred, td);
2529
2530 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2531 &attrflag, &eof, NULL);
2532 if (attrflag)
2533 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2534
2535 if (!error) {
2536 /*
2537 * We are now either at the end of the directory or have filled
2538 * the block.
2539 */
2540 if (eof) {
2541 NFSLOCKNODE(dnp);
2542 dnp->n_direofoffset = uiop->uio_offset;
2543 NFSUNLOCKNODE(dnp);
2544 } else {
2545 if (uiop->uio_resid > 0)
2546 printf("EEK! readdirrpc resid > 0\n");
2547 ncl_dircookie_lock(dnp);
2548 NFSUNLOCKNODE(dnp);
2549 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2550 *cookiep = cookie;
2551 ncl_dircookie_unlock(dnp);
2552 }
2553 } else if (NFS_ISV4(vp)) {
2554 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2555 }
2556 return (error);
2557 }
2558
2559 /*
2560 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2561 */
2562 int
2563 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2564 struct thread *td)
2565 {
2566 struct nfsvattr nfsva;
2567 nfsuint64 *cookiep, cookie;
2568 struct nfsnode *dnp = VTONFS(vp);
2569 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2570 int error = 0, attrflag, eof;
2571
2572 KASSERT(uiop->uio_iovcnt == 1 &&
2573 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2574 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2575 ("nfs readdirplusrpc bad uio"));
2576
2577 /*
2578 * If there is no cookie, assume directory was stale.
2579 */
2580 ncl_dircookie_lock(dnp);
2581 NFSUNLOCKNODE(dnp);
2582 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2583 if (cookiep) {
2584 cookie = *cookiep;
2585 ncl_dircookie_unlock(dnp);
2586 } else {
2587 ncl_dircookie_unlock(dnp);
2588 return (NFSERR_BAD_COOKIE);
2589 }
2590
2591 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2592 (void)ncl_fsinfo(nmp, vp, cred, td);
2593 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2594 &attrflag, &eof, NULL);
2595 if (attrflag)
2596 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2597
2598 if (!error) {
2599 /*
2600 * We are now either at end of the directory or have filled the
2601 * the block.
2602 */
2603 if (eof) {
2604 NFSLOCKNODE(dnp);
2605 dnp->n_direofoffset = uiop->uio_offset;
2606 NFSUNLOCKNODE(dnp);
2607 } else {
2608 if (uiop->uio_resid > 0)
2609 printf("EEK! readdirplusrpc resid > 0\n");
2610 ncl_dircookie_lock(dnp);
2611 NFSUNLOCKNODE(dnp);
2612 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2613 *cookiep = cookie;
2614 ncl_dircookie_unlock(dnp);
2615 }
2616 } else if (NFS_ISV4(vp)) {
2617 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2618 }
2619 return (error);
2620 }
2621
2622 /*
2623 * Silly rename. To make the NFS filesystem that is stateless look a little
2624 * more like the "ufs" a remove of an active vnode is translated to a rename
2625 * to a funny looking filename that is removed by nfs_inactive on the
2626 * nfsnode. There is the potential for another process on a different client
2627 * to create the same funny name between the nfs_lookitup() fails and the
2628 * nfs_rename() completes, but...
2629 */
2630 static int
2631 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2632 {
2633 struct sillyrename *sp;
2634 struct nfsnode *np;
2635 int error;
2636 short pid;
2637 unsigned int lticks;
2638
2639 cache_purge(dvp);
2640 np = VTONFS(vp);
2641 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2642 sp = malloc(sizeof (struct sillyrename),
2643 M_NEWNFSREQ, M_WAITOK);
2644 sp->s_cred = crhold(cnp->cn_cred);
2645 sp->s_dvp = dvp;
2646 VREF(dvp);
2647
2648 /*
2649 * Fudge together a funny name.
2650 * Changing the format of the funny name to accommodate more
2651 * sillynames per directory.
2652 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2653 * CPU ticks since boot.
2654 */
2655 pid = cnp->cn_thread->td_proc->p_pid;
2656 lticks = (unsigned int)ticks;
2657 for ( ; ; ) {
2658 sp->s_namlen = sprintf(sp->s_name,
2659 ".nfs.%08x.%04x4.4", lticks,
2660 pid);
2661 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2662 cnp->cn_thread, NULL))
2663 break;
2664 lticks++;
2665 }
2666 error = nfs_renameit(dvp, vp, cnp, sp);
2667 if (error)
2668 goto bad;
2669 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2670 cnp->cn_thread, &np);
2671 np->n_sillyrename = sp;
2672 return (0);
2673 bad:
2674 vrele(sp->s_dvp);
2675 crfree(sp->s_cred);
2676 free(sp, M_NEWNFSREQ);
2677 return (error);
2678 }
2679
2680 /*
2681 * Look up a file name and optionally either update the file handle or
2682 * allocate an nfsnode, depending on the value of npp.
2683 * npp == NULL --> just do the lookup
2684 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2685 * handled too
2686 * *npp != NULL --> update the file handle in the vnode
2687 */
2688 static int
2689 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2690 struct thread *td, struct nfsnode **npp)
2691 {
2692 struct vnode *newvp = NULL, *vp;
2693 struct nfsnode *np, *dnp = VTONFS(dvp);
2694 struct nfsfh *nfhp, *onfhp;
2695 struct nfsvattr nfsva, dnfsva;
2696 struct componentname cn;
2697 int error = 0, attrflag, dattrflag;
2698 u_int hash;
2699 struct timespec ts;
2700
2701 nanouptime(&ts);
2702 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2703 &nfhp, &attrflag, &dattrflag, NULL);
2704 if (dattrflag)
2705 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2706 if (npp && !error) {
2707 if (*npp != NULL) {
2708 np = *npp;
2709 vp = NFSTOV(np);
2710 /*
2711 * For NFSv4, check to see if it is the same name and
2712 * replace the name, if it is different.
2713 */
2714 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2715 (np->n_v4->n4_namelen != len ||
2716 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2717 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2718 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2719 dnp->n_fhp->nfh_len))) {
2720 #ifdef notdef
2721 { char nnn[100]; int nnnl;
2722 nnnl = (len < 100) ? len : 99;
2723 bcopy(name, nnn, nnnl);
2724 nnn[nnnl] = '\0';
2725 printf("replace=%s\n",nnn);
2726 }
2727 #endif
2728 free(np->n_v4, M_NFSV4NODE);
2729 np->n_v4 = malloc(
2730 sizeof (struct nfsv4node) +
2731 dnp->n_fhp->nfh_len + len - 1,
2732 M_NFSV4NODE, M_WAITOK);
2733 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2734 np->n_v4->n4_namelen = len;
2735 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2736 dnp->n_fhp->nfh_len);
2737 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2738 }
2739 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2740 FNV1_32_INIT);
2741 onfhp = np->n_fhp;
2742 /*
2743 * Rehash node for new file handle.
2744 */
2745 vfs_hash_rehash(vp, hash);
2746 np->n_fhp = nfhp;
2747 if (onfhp != NULL)
2748 free(onfhp, M_NFSFH);
2749 newvp = NFSTOV(np);
2750 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2751 free(nfhp, M_NFSFH);
2752 VREF(dvp);
2753 newvp = dvp;
2754 } else {
2755 cn.cn_nameptr = name;
2756 cn.cn_namelen = len;
2757 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2758 &np, NULL, LK_EXCLUSIVE);
2759 if (error)
2760 return (error);
2761 newvp = NFSTOV(np);
2762 /*
2763 * If n_localmodtime >= time before RPC, then
2764 * a file modification operation, such as
2765 * VOP_SETATTR() of size, has occurred while
2766 * the Lookup RPC and acquisition of the vnode
2767 * happened. As such, the attributes might
2768 * be stale, with possibly an incorrect size.
2769 */
2770 NFSLOCKNODE(np);
2771 if (timespecisset(&np->n_localmodtime) &&
2772 timespeccmp(&np->n_localmodtime, &ts, >=)) {
2773 NFSCL_DEBUG(4, "nfs_lookitup: localmod "
2774 "stale attributes\n");
2775 attrflag = 0;
2776 }
2777 NFSUNLOCKNODE(np);
2778 }
2779 if (!attrflag && *npp == NULL) {
2780 if (newvp == dvp)
2781 vrele(newvp);
2782 else
2783 vput(newvp);
2784 return (ENOENT);
2785 }
2786 if (attrflag)
2787 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2788 0, 1);
2789 }
2790 if (npp && *npp == NULL) {
2791 if (error) {
2792 if (newvp) {
2793 if (newvp == dvp)
2794 vrele(newvp);
2795 else
2796 vput(newvp);
2797 }
2798 } else
2799 *npp = np;
2800 }
2801 if (error && NFS_ISV4(dvp))
2802 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2803 return (error);
2804 }
2805
2806 /*
2807 * Nfs Version 3 and 4 commit rpc
2808 */
2809 int
2810 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2811 struct thread *td)
2812 {
2813 struct nfsvattr nfsva;
2814 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2815 struct nfsnode *np;
2816 struct uio uio;
2817 int error, attrflag;
2818
2819 np = VTONFS(vp);
2820 error = EIO;
2821 attrflag = 0;
2822 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2823 uio.uio_offset = offset;
2824 uio.uio_resid = cnt;
2825 error = nfscl_doiods(vp, &uio, NULL, NULL,
2826 NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2827 if (error != 0) {
2828 NFSLOCKNODE(np);
2829 np->n_flag &= ~NDSCOMMIT;
2830 NFSUNLOCKNODE(np);
2831 }
2832 }
2833 if (error != 0) {
2834 mtx_lock(&nmp->nm_mtx);
2835 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2836 mtx_unlock(&nmp->nm_mtx);
2837 return (0);
2838 }
2839 mtx_unlock(&nmp->nm_mtx);
2840 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2841 &attrflag, NULL);
2842 }
2843 if (attrflag != 0)
2844 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2845 0, 1);
2846 if (error != 0 && NFS_ISV4(vp))
2847 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2848 return (error);
2849 }
2850
2851 /*
2852 * Strategy routine.
2853 * For async requests when nfsiod(s) are running, queue the request by
2854 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2855 * request.
2856 */
2857 static int
2858 nfs_strategy(struct vop_strategy_args *ap)
2859 {
2860 struct buf *bp;
2861 struct vnode *vp;
2862 struct ucred *cr;
2863
2864 bp = ap->a_bp;
2865 vp = ap->a_vp;
2866 KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2867 KASSERT(!(bp->b_flags & B_DONE),
2868 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2869
2870 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2871 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2872 DEV_BSIZE);
2873 if (bp->b_iocmd == BIO_READ)
2874 cr = bp->b_rcred;
2875 else
2876 cr = bp->b_wcred;
2877
2878 /*
2879 * If the op is asynchronous and an i/o daemon is waiting
2880 * queue the request, wake it up and wait for completion
2881 * otherwise just do it ourselves.
2882 */
2883 if ((bp->b_flags & B_ASYNC) == 0 ||
2884 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2885 (void) ncl_doio(vp, bp, cr, curthread, 1);
2886 return (0);
2887 }
2888
2889 /*
2890 * fsync vnode op. Just call ncl_flush() with commit == 1.
2891 */
2892 /* ARGSUSED */
2893 static int
2894 nfs_fsync(struct vop_fsync_args *ap)
2895 {
2896
2897 if (ap->a_vp->v_type != VREG) {
2898 /*
2899 * For NFS, metadata is changed synchronously on the server,
2900 * so there is nothing to flush. Also, ncl_flush() clears
2901 * the NMODIFIED flag and that shouldn't be done here for
2902 * directories.
2903 */
2904 return (0);
2905 }
2906 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2907 }
2908
2909 /*
2910 * Flush all the blocks associated with a vnode.
2911 * Walk through the buffer pool and push any dirty pages
2912 * associated with the vnode.
2913 * If the called_from_renewthread argument is TRUE, it has been called
2914 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2915 * waiting for a buffer write to complete.
2916 */
2917 int
2918 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2919 int commit, int called_from_renewthread)
2920 {
2921 struct nfsnode *np = VTONFS(vp);
2922 struct buf *bp;
2923 int i;
2924 struct buf *nbp;
2925 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2926 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2927 int passone = 1, trycnt = 0;
2928 u_quad_t off, endoff, toff;
2929 struct ucred* wcred = NULL;
2930 struct buf **bvec = NULL;
2931 struct bufobj *bo;
2932 #ifndef NFS_COMMITBVECSIZ
2933 #define NFS_COMMITBVECSIZ 20
2934 #endif
2935 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2936 u_int bvecsize = 0, bveccount;
2937 struct timespec ts;
2938
2939 if (called_from_renewthread != 0)
2940 slptimeo = hz;
2941 if (nmp->nm_flag & NFSMNT_INT)
2942 slpflag = PCATCH;
2943 if (!commit)
2944 passone = 0;
2945 bo = &vp->v_bufobj;
2946 /*
2947 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2948 * server, but has not been committed to stable storage on the server
2949 * yet. On the first pass, the byte range is worked out and the commit
2950 * rpc is done. On the second pass, ncl_writebp() is called to do the
2951 * job.
2952 */
2953 again:
2954 off = (u_quad_t)-1;
2955 endoff = 0;
2956 bvecpos = 0;
2957 if (NFS_ISV34(vp) && commit) {
2958 if (bvec != NULL && bvec != bvec_on_stack)
2959 free(bvec, M_TEMP);
2960 /*
2961 * Count up how many buffers waiting for a commit.
2962 */
2963 bveccount = 0;
2964 BO_LOCK(bo);
2965 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2966 if (!BUF_ISLOCKED(bp) &&
2967 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2968 == (B_DELWRI | B_NEEDCOMMIT))
2969 bveccount++;
2970 }
2971 /*
2972 * Allocate space to remember the list of bufs to commit. It is
2973 * important to use M_NOWAIT here to avoid a race with nfs_write.
2974 * If we can't get memory (for whatever reason), we will end up
2975 * committing the buffers one-by-one in the loop below.
2976 */
2977 if (bveccount > NFS_COMMITBVECSIZ) {
2978 /*
2979 * Release the vnode interlock to avoid a lock
2980 * order reversal.
2981 */
2982 BO_UNLOCK(bo);
2983 bvec = (struct buf **)
2984 malloc(bveccount * sizeof(struct buf *),
2985 M_TEMP, M_NOWAIT);
2986 BO_LOCK(bo);
2987 if (bvec == NULL) {
2988 bvec = bvec_on_stack;
2989 bvecsize = NFS_COMMITBVECSIZ;
2990 } else
2991 bvecsize = bveccount;
2992 } else {
2993 bvec = bvec_on_stack;
2994 bvecsize = NFS_COMMITBVECSIZ;
2995 }
2996 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2997 if (bvecpos >= bvecsize)
2998 break;
2999 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3000 nbp = TAILQ_NEXT(bp, b_bobufs);
3001 continue;
3002 }
3003 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3004 (B_DELWRI | B_NEEDCOMMIT)) {
3005 BUF_UNLOCK(bp);
3006 nbp = TAILQ_NEXT(bp, b_bobufs);
3007 continue;
3008 }
3009 BO_UNLOCK(bo);
3010 bremfree(bp);
3011 /*
3012 * Work out if all buffers are using the same cred
3013 * so we can deal with them all with one commit.
3014 *
3015 * NOTE: we are not clearing B_DONE here, so we have
3016 * to do it later on in this routine if we intend to
3017 * initiate I/O on the bp.
3018 *
3019 * Note: to avoid loopback deadlocks, we do not
3020 * assign b_runningbufspace.
3021 */
3022 if (wcred == NULL)
3023 wcred = bp->b_wcred;
3024 else if (wcred != bp->b_wcred)
3025 wcred = NOCRED;
3026 vfs_busy_pages(bp, 1);
3027
3028 BO_LOCK(bo);
3029 /*
3030 * bp is protected by being locked, but nbp is not
3031 * and vfs_busy_pages() may sleep. We have to
3032 * recalculate nbp.
3033 */
3034 nbp = TAILQ_NEXT(bp, b_bobufs);
3035
3036 /*
3037 * A list of these buffers is kept so that the
3038 * second loop knows which buffers have actually
3039 * been committed. This is necessary, since there
3040 * may be a race between the commit rpc and new
3041 * uncommitted writes on the file.
3042 */
3043 bvec[bvecpos++] = bp;
3044 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3045 bp->b_dirtyoff;
3046 if (toff < off)
3047 off = toff;
3048 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3049 if (toff > endoff)
3050 endoff = toff;
3051 }
3052 BO_UNLOCK(bo);
3053 }
3054 if (bvecpos > 0) {
3055 /*
3056 * Commit data on the server, as required.
3057 * If all bufs are using the same wcred, then use that with
3058 * one call for all of them, otherwise commit each one
3059 * separately.
3060 */
3061 if (wcred != NOCRED)
3062 retv = ncl_commit(vp, off, (int)(endoff - off),
3063 wcred, td);
3064 else {
3065 retv = 0;
3066 for (i = 0; i < bvecpos; i++) {
3067 off_t off, size;
3068 bp = bvec[i];
3069 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3070 bp->b_dirtyoff;
3071 size = (u_quad_t)(bp->b_dirtyend
3072 - bp->b_dirtyoff);
3073 retv = ncl_commit(vp, off, (int)size,
3074 bp->b_wcred, td);
3075 if (retv) break;
3076 }
3077 }
3078
3079 if (retv == NFSERR_STALEWRITEVERF)
3080 ncl_clearcommit(vp->v_mount);
3081
3082 /*
3083 * Now, either mark the blocks I/O done or mark the
3084 * blocks dirty, depending on whether the commit
3085 * succeeded.
3086 */
3087 for (i = 0; i < bvecpos; i++) {
3088 bp = bvec[i];
3089 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3090 if (!NFSCL_FORCEDISM(vp->v_mount) && retv) {
3091 /*
3092 * Error, leave B_DELWRI intact
3093 */
3094 vfs_unbusy_pages(bp);
3095 brelse(bp);
3096 } else {
3097 /*
3098 * Success, remove B_DELWRI ( bundirty() ).
3099 *
3100 * b_dirtyoff/b_dirtyend seem to be NFS
3101 * specific. We should probably move that
3102 * into bundirty(). XXX
3103 */
3104 bufobj_wref(bo);
3105 bp->b_flags |= B_ASYNC;
3106 bundirty(bp);
3107 bp->b_flags &= ~B_DONE;
3108 bp->b_ioflags &= ~BIO_ERROR;
3109 bp->b_dirtyoff = bp->b_dirtyend = 0;
3110 bufdone(bp);
3111 }
3112 }
3113 }
3114
3115 /*
3116 * Start/do any write(s) that are required.
3117 */
3118 loop:
3119 BO_LOCK(bo);
3120 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3121 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3122 if (waitfor != MNT_WAIT || passone)
3123 continue;
3124
3125 error = BUF_TIMELOCK(bp,
3126 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3127 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
3128 if (error == 0) {
3129 BUF_UNLOCK(bp);
3130 goto loop;
3131 }
3132 if (error == ENOLCK) {
3133 error = 0;
3134 goto loop;
3135 }
3136 if (called_from_renewthread != 0) {
3137 /*
3138 * Return EIO so the flush will be retried
3139 * later.
3140 */
3141 error = EIO;
3142 goto done;
3143 }
3144 if (newnfs_sigintr(nmp, td)) {
3145 error = EINTR;
3146 goto done;
3147 }
3148 if (slpflag == PCATCH) {
3149 slpflag = 0;
3150 slptimeo = 2 * hz;
3151 }
3152 goto loop;
3153 }
3154 if ((bp->b_flags & B_DELWRI) == 0)
3155 panic("nfs_fsync: not dirty");
3156 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3157 BUF_UNLOCK(bp);
3158 continue;
3159 }
3160 BO_UNLOCK(bo);
3161 bremfree(bp);
3162 bp->b_flags |= B_ASYNC;
3163 bwrite(bp);
3164 if (newnfs_sigintr(nmp, td)) {
3165 error = EINTR;
3166 goto done;
3167 }
3168 goto loop;
3169 }
3170 if (passone) {
3171 passone = 0;
3172 BO_UNLOCK(bo);
3173 goto again;
3174 }
3175 if (waitfor == MNT_WAIT) {
3176 while (bo->bo_numoutput) {
3177 error = bufobj_wwait(bo, slpflag, slptimeo);
3178 if (error) {
3179 BO_UNLOCK(bo);
3180 if (called_from_renewthread != 0) {
3181 /*
3182 * Return EIO so that the flush will be
3183 * retried later.
3184 */
3185 error = EIO;
3186 goto done;
3187 }
3188 error = newnfs_sigintr(nmp, td);
3189 if (error)
3190 goto done;
3191 if (slpflag == PCATCH) {
3192 slpflag = 0;
3193 slptimeo = 2 * hz;
3194 }
3195 BO_LOCK(bo);
3196 }
3197 }
3198 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3199 BO_UNLOCK(bo);
3200 goto loop;
3201 }
3202 /*
3203 * Wait for all the async IO requests to drain
3204 */
3205 BO_UNLOCK(bo);
3206 NFSLOCKNODE(np);
3207 while (np->n_directio_asyncwr > 0) {
3208 np->n_flag |= NFSYNCWAIT;
3209 error = newnfs_msleep(td, &np->n_directio_asyncwr,
3210 &np->n_mtx, slpflag | (PRIBIO + 1),
3211 "nfsfsync", 0);
3212 if (error) {
3213 if (newnfs_sigintr(nmp, td)) {
3214 NFSUNLOCKNODE(np);
3215 error = EINTR;
3216 goto done;
3217 }
3218 }
3219 }
3220 NFSUNLOCKNODE(np);
3221 } else
3222 BO_UNLOCK(bo);
3223 if (NFSHASPNFS(nmp)) {
3224 nfscl_layoutcommit(vp, td);
3225 /*
3226 * Invalidate the attribute cache, since writes to a DS
3227 * won't update the size attribute.
3228 */
3229 NFSLOCKNODE(np);
3230 np->n_attrstamp = 0;
3231 } else
3232 NFSLOCKNODE(np);
3233 if (np->n_flag & NWRITEERR) {
3234 error = np->n_error;
3235 np->n_flag &= ~NWRITEERR;
3236 }
3237 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3238 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3239 np->n_flag &= ~NMODIFIED;
3240 NFSUNLOCKNODE(np);
3241 done:
3242 if (bvec != NULL && bvec != bvec_on_stack)
3243 free(bvec, M_TEMP);
3244 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3245 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
3246 np->n_directio_asyncwr != 0)) {
3247 if (trycnt++ < 5) {
3248 /* try, try again... */
3249 passone = 1;
3250 wcred = NULL;
3251 bvec = NULL;
3252 bvecsize = 0;
3253 goto again;
3254 }
3255 vn_printf(vp, "ncl_flush failed");
3256 error = called_from_renewthread != 0 ? EIO : EBUSY;
3257 }
3258 if (error == 0) {
3259 nanouptime(&ts);
3260 NFSLOCKNODE(np);
3261 np->n_localmodtime = ts;
3262 NFSUNLOCKNODE(np);
3263 }
3264 return (error);
3265 }
3266
3267 /*
3268 * NFS advisory byte-level locks.
3269 */
3270 static int
3271 nfs_advlock(struct vop_advlock_args *ap)
3272 {
3273 struct vnode *vp = ap->a_vp;
3274 struct ucred *cred;
3275 struct nfsnode *np = VTONFS(ap->a_vp);
3276 struct proc *p = (struct proc *)ap->a_id;
3277 struct thread *td = curthread; /* XXX */
3278 struct vattr va;
3279 int ret, error;
3280 u_quad_t size;
3281 struct nfsmount *nmp;
3282
3283 error = NFSVOPLOCK(vp, LK_SHARED);
3284 if (error != 0)
3285 return (EBADF);
3286 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3287 if (vp->v_type != VREG) {
3288 error = EINVAL;
3289 goto out;
3290 }
3291 if ((ap->a_flags & F_POSIX) != 0)
3292 cred = p->p_ucred;
3293 else
3294 cred = td->td_ucred;
3295 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
3296 if (VN_IS_DOOMED(vp)) {
3297 error = EBADF;
3298 goto out;
3299 }
3300
3301 /*
3302 * If this is unlocking a write locked region, flush and
3303 * commit them before unlocking. This is required by
3304 * RFC3530 Sec. 9.3.2.
3305 */
3306 if (ap->a_op == F_UNLCK &&
3307 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3308 ap->a_flags))
3309 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3310
3311 /*
3312 * Mark NFS node as might have acquired a lock.
3313 * This is separate from NHASBEENLOCKED, because it must
3314 * be done before the nfsrpc_advlock() call, which might
3315 * add a nfscllock structure to the client state.
3316 * It is used to check for the case where a nfscllock
3317 * state structure cannot exist for the file.
3318 * Only done for "oneopenown" NFSv4.1/4.2 mounts.
3319 */
3320 nmp = VFSTONFS(vp->v_mount);
3321 if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) {
3322 NFSLOCKNODE(np);
3323 np->n_flag |= NMIGHTBELOCKED;
3324 NFSUNLOCKNODE(np);
3325 }
3326
3327 /*
3328 * Loop around doing the lock op, while a blocking lock
3329 * must wait for the lock op to succeed.
3330 */
3331 do {
3332 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3333 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3334 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3335 ap->a_op == F_SETLK) {
3336 NFSVOPUNLOCK(vp);
3337 error = nfs_catnap(PZERO | PCATCH, ret,
3338 "ncladvl");
3339 if (error)
3340 return (EINTR);
3341 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3342 if (VN_IS_DOOMED(vp)) {
3343 error = EBADF;
3344 goto out;
3345 }
3346 }
3347 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3348 ap->a_op == F_SETLK);
3349 if (ret == NFSERR_DENIED) {
3350 error = EAGAIN;
3351 goto out;
3352 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3353 error = ret;
3354 goto out;
3355 } else if (ret != 0) {
3356 error = EACCES;
3357 goto out;
3358 }
3359
3360 /*
3361 * Now, if we just got a lock, invalidate data in the buffer
3362 * cache, as required, so that the coherency conforms with
3363 * RFC3530 Sec. 9.3.2.
3364 */
3365 if (ap->a_op == F_SETLK) {
3366 if ((np->n_flag & NMODIFIED) == 0) {
3367 np->n_attrstamp = 0;
3368 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3369 ret = VOP_GETATTR(vp, &va, cred);
3370 }
3371 if ((np->n_flag & NMODIFIED) || ret ||
3372 np->n_change != va.va_filerev) {
3373 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3374 np->n_attrstamp = 0;
3375 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3376 ret = VOP_GETATTR(vp, &va, cred);
3377 if (!ret) {
3378 np->n_mtime = va.va_mtime;
3379 np->n_change = va.va_filerev;
3380 }
3381 }
3382 /* Mark that a file lock has been acquired. */
3383 NFSLOCKNODE(np);
3384 np->n_flag |= NHASBEENLOCKED;
3385 NFSUNLOCKNODE(np);
3386 }
3387 } else if (!NFS_ISV4(vp)) {
3388 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3389 size = VTONFS(vp)->n_size;
3390 NFSVOPUNLOCK(vp);
3391 error = lf_advlock(ap, &(vp->v_lockf), size);
3392 } else {
3393 if (nfs_advlock_p != NULL)
3394 error = nfs_advlock_p(ap);
3395 else {
3396 NFSVOPUNLOCK(vp);
3397 error = ENOLCK;
3398 }
3399 }
3400 if (error == 0 && ap->a_op == F_SETLK) {
3401 error = NFSVOPLOCK(vp, LK_SHARED);
3402 if (error == 0) {
3403 /* Mark that a file lock has been acquired. */
3404 NFSLOCKNODE(np);
3405 np->n_flag |= NHASBEENLOCKED;
3406 NFSUNLOCKNODE(np);
3407 NFSVOPUNLOCK(vp);
3408 }
3409 }
3410 return (error);
3411 } else
3412 error = EOPNOTSUPP;
3413 out:
3414 NFSVOPUNLOCK(vp);
3415 return (error);
3416 }
3417
3418 /*
3419 * NFS advisory byte-level locks.
3420 */
3421 static int
3422 nfs_advlockasync(struct vop_advlockasync_args *ap)
3423 {
3424 struct vnode *vp = ap->a_vp;
3425 u_quad_t size;
3426 int error;
3427
3428 if (NFS_ISV4(vp))
3429 return (EOPNOTSUPP);
3430 error = NFSVOPLOCK(vp, LK_SHARED);
3431 if (error)
3432 return (error);
3433 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3434 size = VTONFS(vp)->n_size;
3435 NFSVOPUNLOCK(vp);
3436 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3437 } else {
3438 NFSVOPUNLOCK(vp);
3439 error = EOPNOTSUPP;
3440 }
3441 return (error);
3442 }
3443
3444 /*
3445 * Print out the contents of an nfsnode.
3446 */
3447 static int
3448 nfs_print(struct vop_print_args *ap)
3449 {
3450 struct vnode *vp = ap->a_vp;
3451 struct nfsnode *np = VTONFS(vp);
3452
3453 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3454 (uintmax_t)np->n_vattr.na_fsid);
3455 if (vp->v_type == VFIFO)
3456 fifo_printinfo(vp);
3457 printf("\n");
3458 return (0);
3459 }
3460
3461 /*
3462 * This is the "real" nfs::bwrite(struct buf*).
3463 * We set B_CACHE if this is a VMIO buffer.
3464 */
3465 int
3466 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3467 {
3468 int oldflags, rtval;
3469
3470 if (bp->b_flags & B_INVAL) {
3471 brelse(bp);
3472 return (0);
3473 }
3474
3475 oldflags = bp->b_flags;
3476 bp->b_flags |= B_CACHE;
3477
3478 /*
3479 * Undirty the bp. We will redirty it later if the I/O fails.
3480 */
3481 bundirty(bp);
3482 bp->b_flags &= ~B_DONE;
3483 bp->b_ioflags &= ~BIO_ERROR;
3484 bp->b_iocmd = BIO_WRITE;
3485
3486 bufobj_wref(bp->b_bufobj);
3487 curthread->td_ru.ru_oublock++;
3488
3489 /*
3490 * Note: to avoid loopback deadlocks, we do not
3491 * assign b_runningbufspace.
3492 */
3493 vfs_busy_pages(bp, 1);
3494
3495 BUF_KERNPROC(bp);
3496 bp->b_iooffset = dbtob(bp->b_blkno);
3497 bstrategy(bp);
3498
3499 if ((oldflags & B_ASYNC) != 0)
3500 return (0);
3501
3502 rtval = bufwait(bp);
3503 if (oldflags & B_DELWRI)
3504 reassignbuf(bp);
3505 brelse(bp);
3506 return (rtval);
3507 }
3508
3509 /*
3510 * nfs special file access vnode op.
3511 * Essentially just get vattr and then imitate iaccess() since the device is
3512 * local to the client.
3513 */
3514 static int
3515 nfsspec_access(struct vop_access_args *ap)
3516 {
3517 struct vattr *vap;
3518 struct ucred *cred = ap->a_cred;
3519 struct vnode *vp = ap->a_vp;
3520 accmode_t accmode = ap->a_accmode;
3521 struct vattr vattr;
3522 int error;
3523
3524 /*
3525 * Disallow write attempts on filesystems mounted read-only;
3526 * unless the file is a socket, fifo, or a block or character
3527 * device resident on the filesystem.
3528 */
3529 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3530 switch (vp->v_type) {
3531 case VREG:
3532 case VDIR:
3533 case VLNK:
3534 return (EROFS);
3535 default:
3536 break;
3537 }
3538 }
3539 vap = &vattr;
3540 error = VOP_GETATTR(vp, vap, cred);
3541 if (error)
3542 goto out;
3543 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3544 accmode, cred);
3545 out:
3546 return error;
3547 }
3548
3549 /*
3550 * Read wrapper for fifos.
3551 */
3552 static int
3553 nfsfifo_read(struct vop_read_args *ap)
3554 {
3555 struct nfsnode *np = VTONFS(ap->a_vp);
3556 int error;
3557
3558 /*
3559 * Set access flag.
3560 */
3561 NFSLOCKNODE(np);
3562 np->n_flag |= NACC;
3563 vfs_timestamp(&np->n_atim);
3564 NFSUNLOCKNODE(np);
3565 error = fifo_specops.vop_read(ap);
3566 return error;
3567 }
3568
3569 /*
3570 * Write wrapper for fifos.
3571 */
3572 static int
3573 nfsfifo_write(struct vop_write_args *ap)
3574 {
3575 struct nfsnode *np = VTONFS(ap->a_vp);
3576
3577 /*
3578 * Set update flag.
3579 */
3580 NFSLOCKNODE(np);
3581 np->n_flag |= NUPD;
3582 vfs_timestamp(&np->n_mtim);
3583 NFSUNLOCKNODE(np);
3584 return(fifo_specops.vop_write(ap));
3585 }
3586
3587 /*
3588 * Close wrapper for fifos.
3589 *
3590 * Update the times on the nfsnode then do fifo close.
3591 */
3592 static int
3593 nfsfifo_close(struct vop_close_args *ap)
3594 {
3595 struct vnode *vp = ap->a_vp;
3596 struct nfsnode *np = VTONFS(vp);
3597 struct vattr vattr;
3598 struct timespec ts;
3599
3600 NFSLOCKNODE(np);
3601 if (np->n_flag & (NACC | NUPD)) {
3602 vfs_timestamp(&ts);
3603 if (np->n_flag & NACC)
3604 np->n_atim = ts;
3605 if (np->n_flag & NUPD)
3606 np->n_mtim = ts;
3607 np->n_flag |= NCHG;
3608 if (vrefcnt(vp) == 1 &&
3609 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3610 VATTR_NULL(&vattr);
3611 if (np->n_flag & NACC)
3612 vattr.va_atime = np->n_atim;
3613 if (np->n_flag & NUPD)
3614 vattr.va_mtime = np->n_mtim;
3615 NFSUNLOCKNODE(np);
3616 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3617 goto out;
3618 }
3619 }
3620 NFSUNLOCKNODE(np);
3621 out:
3622 return (fifo_specops.vop_close(ap));
3623 }
3624
3625 /*
3626 * Just call ncl_writebp() with the force argument set to 1.
3627 *
3628 * NOTE: B_DONE may or may not be set in a_bp on call.
3629 */
3630 static int
3631 nfs_bwrite(struct buf *bp)
3632 {
3633
3634 return (ncl_writebp(bp, 1, curthread));
3635 }
3636
3637 struct buf_ops buf_ops_newnfs = {
3638 .bop_name = "buf_ops_nfs",
3639 .bop_write = nfs_bwrite,
3640 .bop_strategy = bufstrategy,
3641 .bop_sync = bufsync,
3642 .bop_bdflush = bufbdflush,
3643 };
3644
3645 static int
3646 nfs_getacl(struct vop_getacl_args *ap)
3647 {
3648 int error;
3649
3650 if (ap->a_type != ACL_TYPE_NFS4)
3651 return (EOPNOTSUPP);
3652 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3653 NULL);
3654 if (error > NFSERR_STALE) {
3655 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3656 error = EPERM;
3657 }
3658 return (error);
3659 }
3660
3661 static int
3662 nfs_setacl(struct vop_setacl_args *ap)
3663 {
3664 int error;
3665
3666 if (ap->a_type != ACL_TYPE_NFS4)
3667 return (EOPNOTSUPP);
3668 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3669 NULL);
3670 if (error > NFSERR_STALE) {
3671 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3672 error = EPERM;
3673 }
3674 return (error);
3675 }
3676
3677 /*
3678 * VOP_ADVISE for NFS.
3679 * Just return 0 for any errors, since it is just a hint.
3680 */
3681 static int
3682 nfs_advise(struct vop_advise_args *ap)
3683 {
3684 struct thread *td = curthread;
3685 struct nfsmount *nmp;
3686 uint64_t len;
3687 int error;
3688
3689 /*
3690 * First do vop_stdadvise() to handle the buffer cache.
3691 */
3692 error = vop_stdadvise(ap);
3693 if (error != 0)
3694 return (error);
3695 if (ap->a_start < 0 || ap->a_end < 0)
3696 return (0);
3697 if (ap->a_end == OFF_MAX)
3698 len = 0;
3699 else if (ap->a_end < ap->a_start)
3700 return (0);
3701 else
3702 len = ap->a_end - ap->a_start + 1;
3703 nmp = VFSTONFS(ap->a_vp->v_mount);
3704 mtx_lock(&nmp->nm_mtx);
3705 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3706 (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) ==
3707 0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) {
3708 mtx_unlock(&nmp->nm_mtx);
3709 return (0);
3710 }
3711 mtx_unlock(&nmp->nm_mtx);
3712 error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice,
3713 td->td_ucred, td);
3714 if (error == NFSERR_NOTSUPP) {
3715 mtx_lock(&nmp->nm_mtx);
3716 nmp->nm_privflag |= NFSMNTP_NOADVISE;
3717 mtx_unlock(&nmp->nm_mtx);
3718 }
3719 return (0);
3720 }
3721
3722 /*
3723 * nfs allocate call
3724 */
3725 static int
3726 nfs_allocate(struct vop_allocate_args *ap)
3727 {
3728 struct vnode *vp = ap->a_vp;
3729 struct thread *td = curthread;
3730 struct nfsvattr nfsva;
3731 struct nfsmount *nmp;
3732 struct nfsnode *np;
3733 off_t alen;
3734 int attrflag, error, ret;
3735 struct timespec ts;
3736
3737 attrflag = 0;
3738 nmp = VFSTONFS(vp->v_mount);
3739 np = VTONFS(vp);
3740 mtx_lock(&nmp->nm_mtx);
3741 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3742 (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) {
3743 mtx_unlock(&nmp->nm_mtx);
3744 /*
3745 * Flush first to ensure that the allocate adds to the
3746 * file's allocation on the server.
3747 */
3748 error = ncl_flush(vp, MNT_WAIT, td, 1, 0);
3749 if (error == 0) {
3750 alen = *ap->a_len;
3751 if ((uint64_t)alen > nfs_maxalloclen)
3752 alen = nfs_maxalloclen;
3753 error = nfsrpc_allocate(vp, *ap->a_offset, alen,
3754 &nfsva, &attrflag, td->td_ucred, td, NULL);
3755 }
3756 if (error == 0) {
3757 *ap->a_offset += alen;
3758 *ap->a_len -= alen;
3759 nanouptime(&ts);
3760 NFSLOCKNODE(np);
3761 np->n_localmodtime = ts;
3762 NFSUNLOCKNODE(np);
3763 } else if (error == NFSERR_NOTSUPP) {
3764 mtx_lock(&nmp->nm_mtx);
3765 nmp->nm_privflag |= NFSMNTP_NOALLOCATE;
3766 mtx_unlock(&nmp->nm_mtx);
3767 error = EINVAL;
3768 }
3769 } else {
3770 mtx_unlock(&nmp->nm_mtx);
3771 error = EINVAL;
3772 }
3773 if (attrflag != 0) {
3774 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3775 if (error == 0 && ret != 0)
3776 error = ret;
3777 }
3778 if (error != 0)
3779 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3780 return (error);
3781 }
3782
3783 /*
3784 * nfs copy_file_range call
3785 */
3786 static int
3787 nfs_copy_file_range(struct vop_copy_file_range_args *ap)
3788 {
3789 struct vnode *invp = ap->a_invp;
3790 struct vnode *outvp = ap->a_outvp;
3791 struct mount *mp;
3792 struct nfsvattr innfsva, outnfsva;
3793 struct vattr *vap;
3794 struct uio io;
3795 struct nfsmount *nmp;
3796 size_t len, len2;
3797 int error, inattrflag, outattrflag, ret, ret2;
3798 off_t inoff, outoff;
3799 bool consecutive, must_commit, tryoutcred;
3800
3801 ret = ret2 = 0;
3802 nmp = VFSTONFS(invp->v_mount);
3803 mtx_lock(&nmp->nm_mtx);
3804 /* NFSv4.2 Copy is not permitted for infile == outfile. */
3805 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3806 (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0 || invp == outvp) {
3807 mtx_unlock(&nmp->nm_mtx);
3808 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3809 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3810 ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3811 return (error);
3812 }
3813 mtx_unlock(&nmp->nm_mtx);
3814
3815 /* Lock both vnodes, avoiding risk of deadlock. */
3816 do {
3817 mp = NULL;
3818 error = vn_start_write(outvp, &mp, V_WAIT);
3819 if (error == 0) {
3820 error = vn_lock(outvp, LK_EXCLUSIVE);
3821 if (error == 0) {
3822 error = vn_lock(invp, LK_SHARED | LK_NOWAIT);
3823 if (error == 0)
3824 break;
3825 VOP_UNLOCK(outvp);
3826 if (mp != NULL)
3827 vn_finished_write(mp);
3828 mp = NULL;
3829 error = vn_lock(invp, LK_SHARED);
3830 if (error == 0)
3831 VOP_UNLOCK(invp);
3832 }
3833 }
3834 if (mp != NULL)
3835 vn_finished_write(mp);
3836 } while (error == 0);
3837 if (error != 0)
3838 return (error);
3839
3840 /*
3841 * Do the vn_rlimit_fsize() check. Should this be above the VOP layer?
3842 */
3843 io.uio_offset = *ap->a_outoffp;
3844 io.uio_resid = *ap->a_lenp;
3845 error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd);
3846
3847 /*
3848 * Flush the input file so that the data is up to date before
3849 * the copy. Flush writes for the output file so that they
3850 * do not overwrite the data copied to the output file by the Copy.
3851 * Set the commit argument for both flushes so that the data is on
3852 * stable storage before the Copy RPC. This is done in case the
3853 * server reboots during the Copy and needs to be redone.
3854 */
3855 if (error == 0)
3856 error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0);
3857 if (error == 0)
3858 error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0);
3859
3860 /* Do the actual NFSv4.2 RPC. */
3861 len = *ap->a_lenp;
3862 mtx_lock(&nmp->nm_mtx);
3863 if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0)
3864 consecutive = true;
3865 else
3866 consecutive = false;
3867 mtx_unlock(&nmp->nm_mtx);
3868 inoff = *ap->a_inoffp;
3869 outoff = *ap->a_outoffp;
3870 tryoutcred = true;
3871 must_commit = false;
3872 if (error == 0) {
3873 vap = &VTONFS(invp)->n_vattr.na_vattr;
3874 error = VOP_GETATTR(invp, vap, ap->a_incred);
3875 if (error == 0) {
3876 /*
3877 * Clip "len" at va_size so that RFC compliant servers
3878 * will not reply NFSERR_INVAL.
3879 * Setting "len == 0" for the RPC would be preferred,
3880 * but some Linux servers do not support that.
3881 */
3882 if (inoff >= vap->va_size)
3883 *ap->a_lenp = len = 0;
3884 else if (inoff + len > vap->va_size)
3885 *ap->a_lenp = len = vap->va_size - inoff;
3886 } else
3887 error = 0;
3888 }
3889
3890 /*
3891 * len will be set to 0 upon a successful Copy RPC.
3892 * As such, this only loops when the Copy RPC needs to be retried.
3893 */
3894 while (len > 0 && error == 0) {
3895 inattrflag = outattrflag = 0;
3896 len2 = len;
3897 if (tryoutcred)
3898 error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3899 outvp, ap->a_outoffp, &len2, ap->a_flags,
3900 &inattrflag, &innfsva, &outattrflag, &outnfsva,
3901 ap->a_outcred, consecutive, &must_commit);
3902 else
3903 error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3904 outvp, ap->a_outoffp, &len2, ap->a_flags,
3905 &inattrflag, &innfsva, &outattrflag, &outnfsva,
3906 ap->a_incred, consecutive, &must_commit);
3907 if (inattrflag != 0)
3908 ret = nfscl_loadattrcache(&invp, &innfsva, NULL, NULL,
3909 0, 1);
3910 if (outattrflag != 0)
3911 ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL,
3912 NULL, 1, 1);
3913 if (error == 0) {
3914 if (consecutive == false) {
3915 if (len2 == len) {
3916 mtx_lock(&nmp->nm_mtx);
3917 nmp->nm_privflag |=
3918 NFSMNTP_NOCONSECUTIVE;
3919 mtx_unlock(&nmp->nm_mtx);
3920 } else
3921 error = NFSERR_OFFLOADNOREQS;
3922 }
3923 *ap->a_lenp = len2;
3924 len = 0;
3925 if (len2 > 0 && must_commit && error == 0)
3926 error = ncl_commit(outvp, outoff, *ap->a_lenp,
3927 ap->a_outcred, curthread);
3928 if (error == 0 && ret != 0)
3929 error = ret;
3930 if (error == 0 && ret2 != 0)
3931 error = ret2;
3932 } else if (error == NFSERR_OFFLOADNOREQS && consecutive) {
3933 /*
3934 * Try consecutive == false, which is ok only if all
3935 * bytes are copied.
3936 * If only some bytes were copied when consecutive
3937 * is false, there is no way to know which bytes
3938 * still need to be written.
3939 */
3940 consecutive = false;
3941 error = 0;
3942 } else if (error == NFSERR_ACCES && tryoutcred) {
3943 /* Try again with incred. */
3944 tryoutcred = false;
3945 error = 0;
3946 }
3947 if (error == NFSERR_STALEWRITEVERF) {
3948 /*
3949 * Server rebooted, so do it all again.
3950 */
3951 *ap->a_inoffp = inoff;
3952 *ap->a_outoffp = outoff;
3953 len = *ap->a_lenp;
3954 must_commit = false;
3955 error = 0;
3956 }
3957 }
3958 VOP_UNLOCK(invp);
3959 VOP_UNLOCK(outvp);
3960 if (mp != NULL)
3961 vn_finished_write(mp);
3962 if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS ||
3963 error == NFSERR_ACCES) {
3964 /*
3965 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can
3966 * use a_incred for the read and a_outcred for the write, so
3967 * try this for NFSERR_ACCES failures for the Copy.
3968 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can
3969 * never succeed, so disable it.
3970 */
3971 if (error != NFSERR_ACCES) {
3972 /* Can never do Copy on this mount. */
3973 mtx_lock(&nmp->nm_mtx);
3974 nmp->nm_privflag |= NFSMNTP_NOCOPY;
3975 mtx_unlock(&nmp->nm_mtx);
3976 }
3977 *ap->a_inoffp = inoff;
3978 *ap->a_outoffp = outoff;
3979 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3980 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3981 ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3982 } else if (error != 0)
3983 *ap->a_lenp = 0;
3984
3985 if (error != 0)
3986 error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0);
3987 return (error);
3988 }
3989
3990 /*
3991 * nfs ioctl call
3992 */
3993 static int
3994 nfs_ioctl(struct vop_ioctl_args *ap)
3995 {
3996 struct vnode *vp = ap->a_vp;
3997 struct nfsvattr nfsva;
3998 struct nfsmount *nmp;
3999 int attrflag, content, error, ret;
4000 bool eof = false; /* shut up compiler. */
4001
4002 if (vp->v_type != VREG)
4003 return (ENOTTY);
4004 nmp = VFSTONFS(vp->v_mount);
4005 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) {
4006 error = vop_stdioctl(ap);
4007 return (error);
4008 }
4009
4010 /* Do the actual NFSv4.2 RPC. */
4011 switch (ap->a_command) {
4012 case FIOSEEKDATA:
4013 content = NFSV4CONTENT_DATA;
4014 break;
4015 case FIOSEEKHOLE:
4016 content = NFSV4CONTENT_HOLE;
4017 break;
4018 default:
4019 return (ENOTTY);
4020 }
4021
4022 error = vn_lock(vp, LK_SHARED);
4023 if (error != 0)
4024 return (EBADF);
4025 attrflag = 0;
4026 if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size)
4027 error = ENXIO;
4028 else {
4029 /*
4030 * Flush all writes, so that the server is up to date.
4031 * Although a Commit is not required, the commit argument
4032 * is set so that, for a pNFS File/Flexible File Layout
4033 * server, the LayoutCommit will be done to ensure the file
4034 * size is up to date on the Metadata Server.
4035 */
4036 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0);
4037 if (error == 0)
4038 error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof,
4039 content, ap->a_cred, &nfsva, &attrflag);
4040 /* If at eof for FIOSEEKDATA, return ENXIO. */
4041 if (eof && error == 0 && content == NFSV4CONTENT_DATA)
4042 error = ENXIO;
4043 }
4044 if (attrflag != 0) {
4045 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4046 if (error == 0 && ret != 0)
4047 error = ret;
4048 }
4049 NFSVOPUNLOCK(vp);
4050
4051 if (error != 0)
4052 error = ENXIO;
4053 return (error);
4054 }
4055
4056 /*
4057 * nfs getextattr call
4058 */
4059 static int
4060 nfs_getextattr(struct vop_getextattr_args *ap)
4061 {
4062 struct vnode *vp = ap->a_vp;
4063 struct nfsmount *nmp;
4064 struct ucred *cred;
4065 struct thread *td = ap->a_td;
4066 struct nfsvattr nfsva;
4067 ssize_t len;
4068 int attrflag, error, ret;
4069
4070 nmp = VFSTONFS(vp->v_mount);
4071 mtx_lock(&nmp->nm_mtx);
4072 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4073 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4074 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4075 mtx_unlock(&nmp->nm_mtx);
4076 return (EOPNOTSUPP);
4077 }
4078 mtx_unlock(&nmp->nm_mtx);
4079
4080 cred = ap->a_cred;
4081 if (cred == NULL)
4082 cred = td->td_ucred;
4083 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4084 attrflag = 0;
4085 error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva,
4086 &attrflag, cred, td);
4087 if (attrflag != 0) {
4088 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4089 if (error == 0 && ret != 0)
4090 error = ret;
4091 }
4092 if (error == 0 && ap->a_size != NULL)
4093 *ap->a_size = len;
4094
4095 switch (error) {
4096 case NFSERR_NOTSUPP:
4097 case NFSERR_OPILLEGAL:
4098 mtx_lock(&nmp->nm_mtx);
4099 nmp->nm_privflag |= NFSMNTP_NOXATTR;
4100 mtx_unlock(&nmp->nm_mtx);
4101 error = EOPNOTSUPP;
4102 break;
4103 case NFSERR_NOXATTR:
4104 case NFSERR_XATTR2BIG:
4105 error = ENOATTR;
4106 break;
4107 default:
4108 error = nfscl_maperr(td, error, 0, 0);
4109 break;
4110 }
4111 return (error);
4112 }
4113
4114 /*
4115 * nfs setextattr call
4116 */
4117 static int
4118 nfs_setextattr(struct vop_setextattr_args *ap)
4119 {
4120 struct vnode *vp = ap->a_vp;
4121 struct nfsmount *nmp;
4122 struct ucred *cred;
4123 struct thread *td = ap->a_td;
4124 struct nfsvattr nfsva;
4125 int attrflag, error, ret;
4126
4127 nmp = VFSTONFS(vp->v_mount);
4128 mtx_lock(&nmp->nm_mtx);
4129 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4130 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4131 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4132 mtx_unlock(&nmp->nm_mtx);
4133 return (EOPNOTSUPP);
4134 }
4135 mtx_unlock(&nmp->nm_mtx);
4136
4137 if (ap->a_uio->uio_resid < 0)
4138 return (EINVAL);
4139 cred = ap->a_cred;
4140 if (cred == NULL)
4141 cred = td->td_ucred;
4142 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4143 attrflag = 0;
4144 error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva,
4145 &attrflag, cred, td);
4146 if (attrflag != 0) {
4147 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4148 if (error == 0 && ret != 0)
4149 error = ret;
4150 }
4151
4152 switch (error) {
4153 case NFSERR_NOTSUPP:
4154 case NFSERR_OPILLEGAL:
4155 mtx_lock(&nmp->nm_mtx);
4156 nmp->nm_privflag |= NFSMNTP_NOXATTR;
4157 mtx_unlock(&nmp->nm_mtx);
4158 error = EOPNOTSUPP;
4159 break;
4160 case NFSERR_NOXATTR:
4161 case NFSERR_XATTR2BIG:
4162 error = ENOATTR;
4163 break;
4164 default:
4165 error = nfscl_maperr(td, error, 0, 0);
4166 break;
4167 }
4168 return (error);
4169 }
4170
4171 /*
4172 * nfs listextattr call
4173 */
4174 static int
4175 nfs_listextattr(struct vop_listextattr_args *ap)
4176 {
4177 struct vnode *vp = ap->a_vp;
4178 struct nfsmount *nmp;
4179 struct ucred *cred;
4180 struct thread *td = ap->a_td;
4181 struct nfsvattr nfsva;
4182 size_t len, len2;
4183 uint64_t cookie;
4184 int attrflag, error, ret;
4185 bool eof;
4186
4187 nmp = VFSTONFS(vp->v_mount);
4188 mtx_lock(&nmp->nm_mtx);
4189 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4190 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4191 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4192 mtx_unlock(&nmp->nm_mtx);
4193 return (EOPNOTSUPP);
4194 }
4195 mtx_unlock(&nmp->nm_mtx);
4196
4197 cred = ap->a_cred;
4198 if (cred == NULL)
4199 cred = td->td_ucred;
4200
4201 /* Loop around doing List Extended Attribute RPCs. */
4202 eof = false;
4203 cookie = 0;
4204 len2 = 0;
4205 error = 0;
4206 while (!eof && error == 0) {
4207 len = nmp->nm_rsize;
4208 attrflag = 0;
4209 error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof,
4210 &nfsva, &attrflag, cred, td);
4211 if (attrflag != 0) {
4212 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4213 1);
4214 if (error == 0 && ret != 0)
4215 error = ret;
4216 }
4217 if (error == 0) {
4218 len2 += len;
4219 if (len2 > SSIZE_MAX)
4220 error = ENOATTR;
4221 }
4222 }
4223 if (error == 0 && ap->a_size != NULL)
4224 *ap->a_size = len2;
4225
4226 switch (error) {
4227 case NFSERR_NOTSUPP:
4228 case NFSERR_OPILLEGAL:
4229 mtx_lock(&nmp->nm_mtx);
4230 nmp->nm_privflag |= NFSMNTP_NOXATTR;
4231 mtx_unlock(&nmp->nm_mtx);
4232 error = EOPNOTSUPP;
4233 break;
4234 case NFSERR_NOXATTR:
4235 case NFSERR_XATTR2BIG:
4236 error = ENOATTR;
4237 break;
4238 default:
4239 error = nfscl_maperr(td, error, 0, 0);
4240 break;
4241 }
4242 return (error);
4243 }
4244
4245 /*
4246 * nfs setextattr call
4247 */
4248 static int
4249 nfs_deleteextattr(struct vop_deleteextattr_args *ap)
4250 {
4251 struct vnode *vp = ap->a_vp;
4252 struct nfsmount *nmp;
4253 struct nfsvattr nfsva;
4254 int attrflag, error, ret;
4255
4256 nmp = VFSTONFS(vp->v_mount);
4257 mtx_lock(&nmp->nm_mtx);
4258 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4259 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4260 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4261 mtx_unlock(&nmp->nm_mtx);
4262 return (EOPNOTSUPP);
4263 }
4264 mtx_unlock(&nmp->nm_mtx);
4265
4266 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4267 attrflag = 0;
4268 error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred,
4269 ap->a_td);
4270 if (attrflag != 0) {
4271 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4272 if (error == 0 && ret != 0)
4273 error = ret;
4274 }
4275
4276 switch (error) {
4277 case NFSERR_NOTSUPP:
4278 case NFSERR_OPILLEGAL:
4279 mtx_lock(&nmp->nm_mtx);
4280 nmp->nm_privflag |= NFSMNTP_NOXATTR;
4281 mtx_unlock(&nmp->nm_mtx);
4282 error = EOPNOTSUPP;
4283 break;
4284 case NFSERR_NOXATTR:
4285 case NFSERR_XATTR2BIG:
4286 error = ENOATTR;
4287 break;
4288 default:
4289 error = nfscl_maperr(ap->a_td, error, 0, 0);
4290 break;
4291 }
4292 return (error);
4293 }
4294
4295 /*
4296 * Return POSIX pathconf information applicable to nfs filesystems.
4297 */
4298 static int
4299 nfs_pathconf(struct vop_pathconf_args *ap)
4300 {
4301 struct nfsv3_pathconf pc;
4302 struct nfsvattr nfsva;
4303 struct vnode *vp = ap->a_vp;
4304 struct nfsmount *nmp;
4305 struct thread *td = curthread;
4306 off_t off;
4307 bool eof;
4308 int attrflag, error;
4309
4310 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
4311 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
4312 ap->a_name == _PC_NO_TRUNC)) ||
4313 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
4314 /*
4315 * Since only the above 4 a_names are returned by the NFSv3
4316 * Pathconf RPC, there is no point in doing it for others.
4317 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
4318 * be used for _PC_NFS4_ACL as well.
4319 */
4320 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
4321 &attrflag, NULL);
4322 if (attrflag != 0)
4323 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4324 1);
4325 if (error != 0)
4326 return (error);
4327 } else {
4328 /*
4329 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
4330 * just fake them.
4331 */
4332 pc.pc_linkmax = NFS_LINK_MAX;
4333 pc.pc_namemax = NFS_MAXNAMLEN;
4334 pc.pc_notrunc = 1;
4335 pc.pc_chownrestricted = 1;
4336 pc.pc_caseinsensitive = 0;
4337 pc.pc_casepreserving = 1;
4338 error = 0;
4339 }
4340 switch (ap->a_name) {
4341 case _PC_LINK_MAX:
4342 #ifdef _LP64
4343 *ap->a_retval = pc.pc_linkmax;
4344 #else
4345 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
4346 #endif
4347 break;
4348 case _PC_NAME_MAX:
4349 *ap->a_retval = pc.pc_namemax;
4350 break;
4351 case _PC_PIPE_BUF:
4352 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
4353 *ap->a_retval = PIPE_BUF;
4354 else
4355 error = EINVAL;
4356 break;
4357 case _PC_CHOWN_RESTRICTED:
4358 *ap->a_retval = pc.pc_chownrestricted;
4359 break;
4360 case _PC_NO_TRUNC:
4361 *ap->a_retval = pc.pc_notrunc;
4362 break;
4363 case _PC_ACL_NFS4:
4364 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4365 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
4366 *ap->a_retval = 1;
4367 else
4368 *ap->a_retval = 0;
4369 break;
4370 case _PC_ACL_PATH_MAX:
4371 if (NFS_ISV4(vp))
4372 *ap->a_retval = ACL_MAX_ENTRIES;
4373 else
4374 *ap->a_retval = 3;
4375 break;
4376 case _PC_PRIO_IO:
4377 *ap->a_retval = 0;
4378 break;
4379 case _PC_SYNC_IO:
4380 *ap->a_retval = 0;
4381 break;
4382 case _PC_ALLOC_SIZE_MIN:
4383 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
4384 break;
4385 case _PC_FILESIZEBITS:
4386 if (NFS_ISV34(vp))
4387 *ap->a_retval = 64;
4388 else
4389 *ap->a_retval = 32;
4390 break;
4391 case _PC_REC_INCR_XFER_SIZE:
4392 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4393 break;
4394 case _PC_REC_MAX_XFER_SIZE:
4395 *ap->a_retval = -1; /* means ``unlimited'' */
4396 break;
4397 case _PC_REC_MIN_XFER_SIZE:
4398 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4399 break;
4400 case _PC_REC_XFER_ALIGN:
4401 *ap->a_retval = PAGE_SIZE;
4402 break;
4403 case _PC_SYMLINK_MAX:
4404 *ap->a_retval = NFS_MAXPATHLEN;
4405 break;
4406 case _PC_MIN_HOLE_SIZE:
4407 /* Only some NFSv4.2 servers support Seek for Holes. */
4408 *ap->a_retval = 0;
4409 nmp = VFSTONFS(vp->v_mount);
4410 if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) {
4411 /*
4412 * NFSv4.2 doesn't have an attribute for hole size,
4413 * so all we can do is see if the Seek operation is
4414 * supported and then use f_iosize as a "best guess".
4415 */
4416 mtx_lock(&nmp->nm_mtx);
4417 if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) {
4418 mtx_unlock(&nmp->nm_mtx);
4419 off = 0;
4420 attrflag = 0;
4421 error = nfsrpc_seek(vp, &off, &eof,
4422 NFSV4CONTENT_HOLE, td->td_ucred, &nfsva,
4423 &attrflag);
4424 if (attrflag != 0)
4425 nfscl_loadattrcache(&vp, &nfsva,
4426 NULL, NULL, 0, 1);
4427 mtx_lock(&nmp->nm_mtx);
4428 if (error == NFSERR_NOTSUPP)
4429 nmp->nm_privflag |= NFSMNTP_SEEKTESTED;
4430 else
4431 nmp->nm_privflag |= NFSMNTP_SEEKTESTED |
4432 NFSMNTP_SEEK;
4433 error = 0;
4434 }
4435 if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0)
4436 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4437 mtx_unlock(&nmp->nm_mtx);
4438 }
4439 break;
4440
4441 default:
4442 error = vop_stdpathconf(ap);
4443 break;
4444 }
4445 return (error);
4446 }
Cache object: 9cccd75f1b3ea300d5c39668165e0e5a
|