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: head/sys/fs/nfsclient/nfs_clvnops.c 345024 2019-03-11 20:40:56Z sjg $");
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/jail.h>
55 #include <sys/malloc.h>
56 #include <sys/mbuf.h>
57 #include <sys/namei.h>
58 #include <sys/socket.h>
59 #include <sys/vnode.h>
60 #include <sys/dirent.h>
61 #include <sys/fcntl.h>
62 #include <sys/lockf.h>
63 #include <sys/stat.h>
64 #include <sys/sysctl.h>
65 #include <sys/signalvar.h>
66
67 #include <vm/vm.h>
68 #include <vm/vm_extern.h>
69 #include <vm/vm_object.h>
70
71 #include <fs/nfs/nfsport.h>
72 #include <fs/nfsclient/nfsnode.h>
73 #include <fs/nfsclient/nfsmount.h>
74 #include <fs/nfsclient/nfs.h>
75 #include <fs/nfsclient/nfs_kdtrace.h>
76
77 #include <net/if.h>
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80
81 #include <nfs/nfs_lock.h>
82
83 #ifdef KDTRACE_HOOKS
84 #include <sys/dtrace_bsd.h>
85
86 dtrace_nfsclient_accesscache_flush_probe_func_t
87 dtrace_nfscl_accesscache_flush_done_probe;
88 uint32_t nfscl_accesscache_flush_done_id;
89
90 dtrace_nfsclient_accesscache_get_probe_func_t
91 dtrace_nfscl_accesscache_get_hit_probe,
92 dtrace_nfscl_accesscache_get_miss_probe;
93 uint32_t nfscl_accesscache_get_hit_id;
94 uint32_t nfscl_accesscache_get_miss_id;
95
96 dtrace_nfsclient_accesscache_load_probe_func_t
97 dtrace_nfscl_accesscache_load_done_probe;
98 uint32_t nfscl_accesscache_load_done_id;
99 #endif /* !KDTRACE_HOOKS */
100
101 /* Defs */
102 #define TRUE 1
103 #define FALSE 0
104
105 extern struct nfsstatsv1 nfsstatsv1;
106 extern int nfsrv_useacl;
107 extern int nfscl_debuglevel;
108 MALLOC_DECLARE(M_NEWNFSREQ);
109
110 static vop_read_t nfsfifo_read;
111 static vop_write_t nfsfifo_write;
112 static vop_close_t nfsfifo_close;
113 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
114 struct thread *);
115 static vop_lookup_t nfs_lookup;
116 static vop_create_t nfs_create;
117 static vop_mknod_t nfs_mknod;
118 static vop_open_t nfs_open;
119 static vop_pathconf_t nfs_pathconf;
120 static vop_close_t nfs_close;
121 static vop_access_t nfs_access;
122 static vop_getattr_t nfs_getattr;
123 static vop_setattr_t nfs_setattr;
124 static vop_read_t nfs_read;
125 static vop_fsync_t nfs_fsync;
126 static vop_remove_t nfs_remove;
127 static vop_link_t nfs_link;
128 static vop_rename_t nfs_rename;
129 static vop_mkdir_t nfs_mkdir;
130 static vop_rmdir_t nfs_rmdir;
131 static vop_symlink_t nfs_symlink;
132 static vop_readdir_t nfs_readdir;
133 static vop_strategy_t nfs_strategy;
134 static int nfs_lookitup(struct vnode *, char *, int,
135 struct ucred *, struct thread *, struct nfsnode **);
136 static int nfs_sillyrename(struct vnode *, struct vnode *,
137 struct componentname *);
138 static vop_access_t nfsspec_access;
139 static vop_readlink_t nfs_readlink;
140 static vop_print_t nfs_print;
141 static vop_advlock_t nfs_advlock;
142 static vop_advlockasync_t nfs_advlockasync;
143 static vop_getacl_t nfs_getacl;
144 static vop_setacl_t nfs_setacl;
145 static vop_set_text_t nfs_set_text;
146
147 /*
148 * Global vfs data structures for nfs
149 */
150
151 static struct vop_vector newnfs_vnodeops_nosig = {
152 .vop_default = &default_vnodeops,
153 .vop_access = nfs_access,
154 .vop_advlock = nfs_advlock,
155 .vop_advlockasync = nfs_advlockasync,
156 .vop_close = nfs_close,
157 .vop_create = nfs_create,
158 .vop_fsync = nfs_fsync,
159 .vop_getattr = nfs_getattr,
160 .vop_getpages = ncl_getpages,
161 .vop_putpages = ncl_putpages,
162 .vop_inactive = ncl_inactive,
163 .vop_link = nfs_link,
164 .vop_lookup = nfs_lookup,
165 .vop_mkdir = nfs_mkdir,
166 .vop_mknod = nfs_mknod,
167 .vop_open = nfs_open,
168 .vop_pathconf = nfs_pathconf,
169 .vop_print = nfs_print,
170 .vop_read = nfs_read,
171 .vop_readdir = nfs_readdir,
172 .vop_readlink = nfs_readlink,
173 .vop_reclaim = ncl_reclaim,
174 .vop_remove = nfs_remove,
175 .vop_rename = nfs_rename,
176 .vop_rmdir = nfs_rmdir,
177 .vop_setattr = nfs_setattr,
178 .vop_strategy = nfs_strategy,
179 .vop_symlink = nfs_symlink,
180 .vop_write = ncl_write,
181 .vop_getacl = nfs_getacl,
182 .vop_setacl = nfs_setacl,
183 .vop_set_text = nfs_set_text,
184 };
185
186 static int
187 nfs_vnodeops_bypass(struct vop_generic_args *a)
188 {
189
190 return (vop_sigdefer(&newnfs_vnodeops_nosig, a));
191 }
192
193 struct vop_vector newnfs_vnodeops = {
194 .vop_default = &default_vnodeops,
195 .vop_bypass = nfs_vnodeops_bypass,
196 };
197
198 static struct vop_vector newnfs_fifoops_nosig = {
199 .vop_default = &fifo_specops,
200 .vop_access = nfsspec_access,
201 .vop_close = nfsfifo_close,
202 .vop_fsync = nfs_fsync,
203 .vop_getattr = nfs_getattr,
204 .vop_inactive = ncl_inactive,
205 .vop_pathconf = nfs_pathconf,
206 .vop_print = nfs_print,
207 .vop_read = nfsfifo_read,
208 .vop_reclaim = ncl_reclaim,
209 .vop_setattr = nfs_setattr,
210 .vop_write = nfsfifo_write,
211 };
212
213 static int
214 nfs_fifoops_bypass(struct vop_generic_args *a)
215 {
216
217 return (vop_sigdefer(&newnfs_fifoops_nosig, a));
218 }
219
220 struct vop_vector newnfs_fifoops = {
221 .vop_default = &default_vnodeops,
222 .vop_bypass = nfs_fifoops_bypass,
223 };
224
225 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
226 struct componentname *cnp, struct vattr *vap);
227 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
228 int namelen, struct ucred *cred, struct thread *td);
229 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
230 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
231 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
232 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
233 struct componentname *scnp, struct sillyrename *sp);
234
235 /*
236 * Global variables
237 */
238 SYSCTL_DECL(_vfs_nfs);
239
240 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
241 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
242 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
243
244 static int nfs_prime_access_cache = 0;
245 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
246 &nfs_prime_access_cache, 0,
247 "Prime NFS ACCESS cache when fetching attributes");
248
249 static int newnfs_commit_on_close = 0;
250 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
251 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
252
253 static int nfs_clean_pages_on_close = 1;
254 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
255 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
256
257 int newnfs_directio_enable = 0;
258 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
259 &newnfs_directio_enable, 0, "Enable NFS directio");
260
261 int nfs_keep_dirty_on_error;
262 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
263 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
264
265 /*
266 * This sysctl allows other processes to mmap a file that has been opened
267 * O_DIRECT by a process. In general, having processes mmap the file while
268 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
269 * this by default to prevent DoS attacks - to prevent a malicious user from
270 * opening up files O_DIRECT preventing other users from mmap'ing these
271 * files. "Protected" environments where stricter consistency guarantees are
272 * required can disable this knob. The process that opened the file O_DIRECT
273 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
274 * meaningful.
275 */
276 int newnfs_directio_allow_mmap = 1;
277 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
278 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
279
280 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
281 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
282 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
283
284 /*
285 * SMP Locking Note :
286 * The list of locks after the description of the lock is the ordering
287 * of other locks acquired with the lock held.
288 * np->n_mtx : Protects the fields in the nfsnode.
289 VM Object Lock
290 VI_MTX (acquired indirectly)
291 * nmp->nm_mtx : Protects the fields in the nfsmount.
292 rep->r_mtx
293 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
294 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
295 nmp->nm_mtx
296 rep->r_mtx
297 * rep->r_mtx : Protects the fields in an nfsreq.
298 */
299
300 static int
301 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
302 struct ucred *cred, u_int32_t *retmode)
303 {
304 int error = 0, attrflag, i, lrupos;
305 u_int32_t rmode;
306 struct nfsnode *np = VTONFS(vp);
307 struct nfsvattr nfsva;
308
309 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
310 &rmode, NULL);
311 if (attrflag)
312 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
313 if (!error) {
314 lrupos = 0;
315 mtx_lock(&np->n_mtx);
316 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
317 if (np->n_accesscache[i].uid == cred->cr_uid) {
318 np->n_accesscache[i].mode = rmode;
319 np->n_accesscache[i].stamp = time_second;
320 break;
321 }
322 if (i > 0 && np->n_accesscache[i].stamp <
323 np->n_accesscache[lrupos].stamp)
324 lrupos = i;
325 }
326 if (i == NFS_ACCESSCACHESIZE) {
327 np->n_accesscache[lrupos].uid = cred->cr_uid;
328 np->n_accesscache[lrupos].mode = rmode;
329 np->n_accesscache[lrupos].stamp = time_second;
330 }
331 mtx_unlock(&np->n_mtx);
332 if (retmode != NULL)
333 *retmode = rmode;
334 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
335 } else if (NFS_ISV4(vp)) {
336 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
337 }
338 #ifdef KDTRACE_HOOKS
339 if (error != 0)
340 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
341 error);
342 #endif
343 return (error);
344 }
345
346 /*
347 * nfs access vnode op.
348 * For nfs version 2, just return ok. File accesses may fail later.
349 * For nfs version 3, use the access rpc to check accessibility. If file modes
350 * are changed on the server, accesses might still fail later.
351 */
352 static int
353 nfs_access(struct vop_access_args *ap)
354 {
355 struct vnode *vp = ap->a_vp;
356 int error = 0, i, gotahit;
357 u_int32_t mode, wmode, rmode;
358 int v34 = NFS_ISV34(vp);
359 struct nfsnode *np = VTONFS(vp);
360
361 /*
362 * Disallow write attempts on filesystems mounted read-only;
363 * unless the file is a socket, fifo, or a block or character
364 * device resident on the filesystem.
365 */
366 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
367 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
368 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
369 switch (vp->v_type) {
370 case VREG:
371 case VDIR:
372 case VLNK:
373 return (EROFS);
374 default:
375 break;
376 }
377 }
378 /*
379 * For nfs v3 or v4, check to see if we have done this recently, and if
380 * so return our cached result instead of making an ACCESS call.
381 * If not, do an access rpc, otherwise you are stuck emulating
382 * ufs_access() locally using the vattr. This may not be correct,
383 * since the server may apply other access criteria such as
384 * client uid-->server uid mapping that we do not know about.
385 */
386 if (v34) {
387 if (ap->a_accmode & VREAD)
388 mode = NFSACCESS_READ;
389 else
390 mode = 0;
391 if (vp->v_type != VDIR) {
392 if (ap->a_accmode & VWRITE)
393 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
394 if (ap->a_accmode & VAPPEND)
395 mode |= NFSACCESS_EXTEND;
396 if (ap->a_accmode & VEXEC)
397 mode |= NFSACCESS_EXECUTE;
398 if (ap->a_accmode & VDELETE)
399 mode |= NFSACCESS_DELETE;
400 } else {
401 if (ap->a_accmode & VWRITE)
402 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
403 if (ap->a_accmode & VAPPEND)
404 mode |= NFSACCESS_EXTEND;
405 if (ap->a_accmode & VEXEC)
406 mode |= NFSACCESS_LOOKUP;
407 if (ap->a_accmode & VDELETE)
408 mode |= NFSACCESS_DELETE;
409 if (ap->a_accmode & VDELETE_CHILD)
410 mode |= NFSACCESS_MODIFY;
411 }
412 /* XXX safety belt, only make blanket request if caching */
413 if (nfsaccess_cache_timeout > 0) {
414 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
415 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
416 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
417 } else {
418 wmode = mode;
419 }
420
421 /*
422 * Does our cached result allow us to give a definite yes to
423 * this request?
424 */
425 gotahit = 0;
426 mtx_lock(&np->n_mtx);
427 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
428 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
429 if (time_second < (np->n_accesscache[i].stamp
430 + nfsaccess_cache_timeout) &&
431 (np->n_accesscache[i].mode & mode) == mode) {
432 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
433 gotahit = 1;
434 }
435 break;
436 }
437 }
438 mtx_unlock(&np->n_mtx);
439 #ifdef KDTRACE_HOOKS
440 if (gotahit != 0)
441 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
442 ap->a_cred->cr_uid, mode);
443 else
444 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
445 ap->a_cred->cr_uid, mode);
446 #endif
447 if (gotahit == 0) {
448 /*
449 * Either a no, or a don't know. Go to the wire.
450 */
451 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
452 error = nfs34_access_otw(vp, wmode, ap->a_td,
453 ap->a_cred, &rmode);
454 if (!error &&
455 (rmode & mode) != mode)
456 error = EACCES;
457 }
458 return (error);
459 } else {
460 if ((error = nfsspec_access(ap)) != 0) {
461 return (error);
462 }
463 /*
464 * Attempt to prevent a mapped root from accessing a file
465 * which it shouldn't. We try to read a byte from the file
466 * if the user is root and the file is not zero length.
467 * After calling nfsspec_access, we should have the correct
468 * file size cached.
469 */
470 mtx_lock(&np->n_mtx);
471 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
472 && VTONFS(vp)->n_size > 0) {
473 struct iovec aiov;
474 struct uio auio;
475 char buf[1];
476
477 mtx_unlock(&np->n_mtx);
478 aiov.iov_base = buf;
479 aiov.iov_len = 1;
480 auio.uio_iov = &aiov;
481 auio.uio_iovcnt = 1;
482 auio.uio_offset = 0;
483 auio.uio_resid = 1;
484 auio.uio_segflg = UIO_SYSSPACE;
485 auio.uio_rw = UIO_READ;
486 auio.uio_td = ap->a_td;
487
488 if (vp->v_type == VREG)
489 error = ncl_readrpc(vp, &auio, ap->a_cred);
490 else if (vp->v_type == VDIR) {
491 char* bp;
492 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
493 aiov.iov_base = bp;
494 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
495 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
496 ap->a_td);
497 free(bp, M_TEMP);
498 } else if (vp->v_type == VLNK)
499 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
500 else
501 error = EACCES;
502 } else
503 mtx_unlock(&np->n_mtx);
504 return (error);
505 }
506 }
507
508
509 /*
510 * nfs open vnode op
511 * Check to see if the type is ok
512 * and that deletion is not in progress.
513 * For paged in text files, you will need to flush the page cache
514 * if consistency is lost.
515 */
516 /* ARGSUSED */
517 static int
518 nfs_open(struct vop_open_args *ap)
519 {
520 struct vnode *vp = ap->a_vp;
521 struct nfsnode *np = VTONFS(vp);
522 struct vattr vattr;
523 int error;
524 int fmode = ap->a_mode;
525 struct ucred *cred;
526
527 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
528 return (EOPNOTSUPP);
529
530 /*
531 * For NFSv4, we need to do the Open Op before cache validation,
532 * so that we conform to RFC3530 Sec. 9.3.1.
533 */
534 if (NFS_ISV4(vp)) {
535 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
536 if (error) {
537 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
538 (gid_t)0);
539 return (error);
540 }
541 }
542
543 /*
544 * Now, if this Open will be doing reading, re-validate/flush the
545 * cache, so that Close/Open coherency is maintained.
546 */
547 mtx_lock(&np->n_mtx);
548 if (np->n_flag & NMODIFIED) {
549 mtx_unlock(&np->n_mtx);
550 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
551 if (error == EINTR || error == EIO) {
552 if (NFS_ISV4(vp))
553 (void) nfsrpc_close(vp, 0, ap->a_td);
554 return (error);
555 }
556 mtx_lock(&np->n_mtx);
557 np->n_attrstamp = 0;
558 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
559 if (vp->v_type == VDIR)
560 np->n_direofoffset = 0;
561 mtx_unlock(&np->n_mtx);
562 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
563 if (error) {
564 if (NFS_ISV4(vp))
565 (void) nfsrpc_close(vp, 0, ap->a_td);
566 return (error);
567 }
568 mtx_lock(&np->n_mtx);
569 np->n_mtime = vattr.va_mtime;
570 if (NFS_ISV4(vp))
571 np->n_change = vattr.va_filerev;
572 } else {
573 mtx_unlock(&np->n_mtx);
574 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
575 if (error) {
576 if (NFS_ISV4(vp))
577 (void) nfsrpc_close(vp, 0, ap->a_td);
578 return (error);
579 }
580 mtx_lock(&np->n_mtx);
581 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
582 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
583 if (vp->v_type == VDIR)
584 np->n_direofoffset = 0;
585 mtx_unlock(&np->n_mtx);
586 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
587 if (error == EINTR || error == EIO) {
588 if (NFS_ISV4(vp))
589 (void) nfsrpc_close(vp, 0, ap->a_td);
590 return (error);
591 }
592 mtx_lock(&np->n_mtx);
593 np->n_mtime = vattr.va_mtime;
594 if (NFS_ISV4(vp))
595 np->n_change = vattr.va_filerev;
596 }
597 }
598
599 /*
600 * If the object has >= 1 O_DIRECT active opens, we disable caching.
601 */
602 if (newnfs_directio_enable && (fmode & O_DIRECT) &&
603 (vp->v_type == VREG)) {
604 if (np->n_directio_opens == 0) {
605 mtx_unlock(&np->n_mtx);
606 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
607 if (error) {
608 if (NFS_ISV4(vp))
609 (void) nfsrpc_close(vp, 0, ap->a_td);
610 return (error);
611 }
612 mtx_lock(&np->n_mtx);
613 np->n_flag |= NNONCACHE;
614 }
615 np->n_directio_opens++;
616 }
617
618 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
619 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
620 np->n_flag |= NWRITEOPENED;
621
622 /*
623 * If this is an open for writing, capture a reference to the
624 * credentials, so they can be used by ncl_putpages(). Using
625 * these write credentials is preferable to the credentials of
626 * whatever thread happens to be doing the VOP_PUTPAGES() since
627 * the write RPCs are less likely to fail with EACCES.
628 */
629 if ((fmode & FWRITE) != 0) {
630 cred = np->n_writecred;
631 np->n_writecred = crhold(ap->a_cred);
632 } else
633 cred = NULL;
634 mtx_unlock(&np->n_mtx);
635
636 if (cred != NULL)
637 crfree(cred);
638 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
639 return (0);
640 }
641
642 /*
643 * nfs close vnode op
644 * What an NFS client should do upon close after writing is a debatable issue.
645 * Most NFS clients push delayed writes to the server upon close, basically for
646 * two reasons:
647 * 1 - So that any write errors may be reported back to the client process
648 * doing the close system call. By far the two most likely errors are
649 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
650 * 2 - To put a worst case upper bound on cache inconsistency between
651 * multiple clients for the file.
652 * There is also a consistency problem for Version 2 of the protocol w.r.t.
653 * not being able to tell if other clients are writing a file concurrently,
654 * since there is no way of knowing if the changed modify time in the reply
655 * is only due to the write for this client.
656 * (NFS Version 3 provides weak cache consistency data in the reply that
657 * should be sufficient to detect and handle this case.)
658 *
659 * The current code does the following:
660 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
661 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
662 * or commit them (this satisfies 1 and 2 except for the
663 * case where the server crashes after this close but
664 * before the commit RPC, which is felt to be "good
665 * enough". Changing the last argument to ncl_flush() to
666 * a 1 would force a commit operation, if it is felt a
667 * commit is necessary now.
668 * for NFS Version 4 - flush the dirty buffers and commit them, if
669 * nfscl_mustflush() says this is necessary.
670 * It is necessary if there is no write delegation held,
671 * in order to satisfy open/close coherency.
672 * If the file isn't cached on local stable storage,
673 * it may be necessary in order to detect "out of space"
674 * errors from the server, if the write delegation
675 * issued by the server doesn't allow the file to grow.
676 */
677 /* ARGSUSED */
678 static int
679 nfs_close(struct vop_close_args *ap)
680 {
681 struct vnode *vp = ap->a_vp;
682 struct nfsnode *np = VTONFS(vp);
683 struct nfsvattr nfsva;
684 struct ucred *cred;
685 int error = 0, ret, localcred = 0;
686 int fmode = ap->a_fflag;
687
688 if (NFSCL_FORCEDISM(vp->v_mount))
689 return (0);
690 /*
691 * During shutdown, a_cred isn't valid, so just use root.
692 */
693 if (ap->a_cred == NOCRED) {
694 cred = newnfs_getcred();
695 localcred = 1;
696 } else {
697 cred = ap->a_cred;
698 }
699 if (vp->v_type == VREG) {
700 /*
701 * Examine and clean dirty pages, regardless of NMODIFIED.
702 * This closes a major hole in close-to-open consistency.
703 * We want to push out all dirty pages (and buffers) on
704 * close, regardless of whether they were dirtied by
705 * mmap'ed writes or via write().
706 */
707 if (nfs_clean_pages_on_close && vp->v_object) {
708 VM_OBJECT_WLOCK(vp->v_object);
709 vm_object_page_clean(vp->v_object, 0, 0, 0);
710 VM_OBJECT_WUNLOCK(vp->v_object);
711 }
712 mtx_lock(&np->n_mtx);
713 if (np->n_flag & NMODIFIED) {
714 mtx_unlock(&np->n_mtx);
715 if (NFS_ISV3(vp)) {
716 /*
717 * Under NFSv3 we have dirty buffers to dispose of. We
718 * must flush them to the NFS server. We have the option
719 * of waiting all the way through the commit rpc or just
720 * waiting for the initial write. The default is to only
721 * wait through the initial write so the data is in the
722 * server's cache, which is roughly similar to the state
723 * a standard disk subsystem leaves the file in on close().
724 *
725 * We cannot clear the NMODIFIED bit in np->n_flag due to
726 * potential races with other processes, and certainly
727 * cannot clear it if we don't commit.
728 * These races occur when there is no longer the old
729 * traditional vnode locking implemented for Vnode Ops.
730 */
731 int cm = newnfs_commit_on_close ? 1 : 0;
732 error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
733 /* np->n_flag &= ~NMODIFIED; */
734 } else if (NFS_ISV4(vp)) {
735 if (nfscl_mustflush(vp) != 0) {
736 int cm = newnfs_commit_on_close ? 1 : 0;
737 error = ncl_flush(vp, MNT_WAIT, ap->a_td,
738 cm, 0);
739 /*
740 * as above w.r.t races when clearing
741 * NMODIFIED.
742 * np->n_flag &= ~NMODIFIED;
743 */
744 }
745 } else {
746 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
747 }
748 mtx_lock(&np->n_mtx);
749 }
750 /*
751 * Invalidate the attribute cache in all cases.
752 * An open is going to fetch fresh attrs any way, other procs
753 * on this node that have file open will be forced to do an
754 * otw attr fetch, but this is safe.
755 * --> A user found that their RPC count dropped by 20% when
756 * this was commented out and I can't see any requirement
757 * for it, so I've disabled it when negative lookups are
758 * enabled. (What does this have to do with negative lookup
759 * caching? Well nothing, except it was reported by the
760 * same user that needed negative lookup caching and I wanted
761 * there to be a way to disable it to see if it
762 * is the cause of some caching/coherency issue that might
763 * crop up.)
764 */
765 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
766 np->n_attrstamp = 0;
767 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
768 }
769 if (np->n_flag & NWRITEERR) {
770 np->n_flag &= ~NWRITEERR;
771 error = np->n_error;
772 }
773 mtx_unlock(&np->n_mtx);
774 }
775
776 if (NFS_ISV4(vp)) {
777 /*
778 * Get attributes so "change" is up to date.
779 */
780 if (error == 0 && nfscl_mustflush(vp) != 0 &&
781 vp->v_type == VREG &&
782 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
783 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
784 NULL);
785 if (!ret) {
786 np->n_change = nfsva.na_filerev;
787 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
788 NULL, 0, 0);
789 }
790 }
791
792 /*
793 * and do the close.
794 */
795 ret = nfsrpc_close(vp, 0, ap->a_td);
796 if (!error && ret)
797 error = ret;
798 if (error)
799 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
800 (gid_t)0);
801 }
802 if (newnfs_directio_enable)
803 KASSERT((np->n_directio_asyncwr == 0),
804 ("nfs_close: dirty unflushed (%d) directio buffers\n",
805 np->n_directio_asyncwr));
806 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
807 mtx_lock(&np->n_mtx);
808 KASSERT((np->n_directio_opens > 0),
809 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
810 np->n_directio_opens--;
811 if (np->n_directio_opens == 0)
812 np->n_flag &= ~NNONCACHE;
813 mtx_unlock(&np->n_mtx);
814 }
815 if (localcred)
816 NFSFREECRED(cred);
817 return (error);
818 }
819
820 /*
821 * nfs getattr call from vfs.
822 */
823 static int
824 nfs_getattr(struct vop_getattr_args *ap)
825 {
826 struct vnode *vp = ap->a_vp;
827 struct thread *td = curthread; /* XXX */
828 struct nfsnode *np = VTONFS(vp);
829 int error = 0;
830 struct nfsvattr nfsva;
831 struct vattr *vap = ap->a_vap;
832 struct vattr vattr;
833
834 /*
835 * Update local times for special files.
836 */
837 mtx_lock(&np->n_mtx);
838 if (np->n_flag & (NACC | NUPD))
839 np->n_flag |= NCHG;
840 mtx_unlock(&np->n_mtx);
841 /*
842 * First look in the cache.
843 */
844 if (ncl_getattrcache(vp, &vattr) == 0) {
845 vap->va_type = vattr.va_type;
846 vap->va_mode = vattr.va_mode;
847 vap->va_nlink = vattr.va_nlink;
848 vap->va_uid = vattr.va_uid;
849 vap->va_gid = vattr.va_gid;
850 vap->va_fsid = vattr.va_fsid;
851 vap->va_fileid = vattr.va_fileid;
852 vap->va_size = vattr.va_size;
853 vap->va_blocksize = vattr.va_blocksize;
854 vap->va_atime = vattr.va_atime;
855 vap->va_mtime = vattr.va_mtime;
856 vap->va_ctime = vattr.va_ctime;
857 vap->va_gen = vattr.va_gen;
858 vap->va_flags = vattr.va_flags;
859 vap->va_rdev = vattr.va_rdev;
860 vap->va_bytes = vattr.va_bytes;
861 vap->va_filerev = vattr.va_filerev;
862 /*
863 * Get the local modify time for the case of a write
864 * delegation.
865 */
866 nfscl_deleggetmodtime(vp, &vap->va_mtime);
867 return (0);
868 }
869
870 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
871 nfsaccess_cache_timeout > 0) {
872 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
873 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
874 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
875 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
876 return (0);
877 }
878 }
879 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
880 if (!error)
881 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
882 if (!error) {
883 /*
884 * Get the local modify time for the case of a write
885 * delegation.
886 */
887 nfscl_deleggetmodtime(vp, &vap->va_mtime);
888 } else if (NFS_ISV4(vp)) {
889 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
890 }
891 return (error);
892 }
893
894 /*
895 * nfs setattr call.
896 */
897 static int
898 nfs_setattr(struct vop_setattr_args *ap)
899 {
900 struct vnode *vp = ap->a_vp;
901 struct nfsnode *np = VTONFS(vp);
902 struct thread *td = curthread; /* XXX */
903 struct vattr *vap = ap->a_vap;
904 int error = 0;
905 u_quad_t tsize;
906
907 #ifndef nolint
908 tsize = (u_quad_t)0;
909 #endif
910
911 /*
912 * Setting of flags and marking of atimes are not supported.
913 */
914 if (vap->va_flags != VNOVAL)
915 return (EOPNOTSUPP);
916
917 /*
918 * Disallow write attempts if the filesystem is mounted read-only.
919 */
920 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
921 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
922 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
923 (vp->v_mount->mnt_flag & MNT_RDONLY))
924 return (EROFS);
925 if (vap->va_size != VNOVAL) {
926 switch (vp->v_type) {
927 case VDIR:
928 return (EISDIR);
929 case VCHR:
930 case VBLK:
931 case VSOCK:
932 case VFIFO:
933 if (vap->va_mtime.tv_sec == VNOVAL &&
934 vap->va_atime.tv_sec == VNOVAL &&
935 vap->va_mode == (mode_t)VNOVAL &&
936 vap->va_uid == (uid_t)VNOVAL &&
937 vap->va_gid == (gid_t)VNOVAL)
938 return (0);
939 vap->va_size = VNOVAL;
940 break;
941 default:
942 /*
943 * Disallow write attempts if the filesystem is
944 * mounted read-only.
945 */
946 if (vp->v_mount->mnt_flag & MNT_RDONLY)
947 return (EROFS);
948 /*
949 * We run vnode_pager_setsize() early (why?),
950 * we must set np->n_size now to avoid vinvalbuf
951 * V_SAVE races that might setsize a lower
952 * value.
953 */
954 mtx_lock(&np->n_mtx);
955 tsize = np->n_size;
956 mtx_unlock(&np->n_mtx);
957 error = ncl_meta_setsize(vp, ap->a_cred, td,
958 vap->va_size);
959 mtx_lock(&np->n_mtx);
960 if (np->n_flag & NMODIFIED) {
961 tsize = np->n_size;
962 mtx_unlock(&np->n_mtx);
963 error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
964 0 : V_SAVE, td, 1);
965 if (error != 0) {
966 vnode_pager_setsize(vp, tsize);
967 return (error);
968 }
969 /*
970 * Call nfscl_delegmodtime() to set the modify time
971 * locally, as required.
972 */
973 nfscl_delegmodtime(vp);
974 } else
975 mtx_unlock(&np->n_mtx);
976 /*
977 * np->n_size has already been set to vap->va_size
978 * in ncl_meta_setsize(). We must set it again since
979 * nfs_loadattrcache() could be called through
980 * ncl_meta_setsize() and could modify np->n_size.
981 */
982 mtx_lock(&np->n_mtx);
983 np->n_vattr.na_size = np->n_size = vap->va_size;
984 mtx_unlock(&np->n_mtx);
985 }
986 } else {
987 mtx_lock(&np->n_mtx);
988 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
989 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
990 mtx_unlock(&np->n_mtx);
991 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
992 if (error == EINTR || error == EIO)
993 return (error);
994 } else
995 mtx_unlock(&np->n_mtx);
996 }
997 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
998 if (error && vap->va_size != VNOVAL) {
999 mtx_lock(&np->n_mtx);
1000 np->n_size = np->n_vattr.na_size = tsize;
1001 vnode_pager_setsize(vp, tsize);
1002 mtx_unlock(&np->n_mtx);
1003 }
1004 return (error);
1005 }
1006
1007 /*
1008 * Do an nfs setattr rpc.
1009 */
1010 static int
1011 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1012 struct thread *td)
1013 {
1014 struct nfsnode *np = VTONFS(vp);
1015 int error, ret, attrflag, i;
1016 struct nfsvattr nfsva;
1017
1018 if (NFS_ISV34(vp)) {
1019 mtx_lock(&np->n_mtx);
1020 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1021 np->n_accesscache[i].stamp = 0;
1022 np->n_flag |= NDELEGMOD;
1023 mtx_unlock(&np->n_mtx);
1024 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1025 }
1026 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1027 NULL);
1028 if (attrflag) {
1029 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1030 if (ret && !error)
1031 error = ret;
1032 }
1033 if (error && NFS_ISV4(vp))
1034 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1035 return (error);
1036 }
1037
1038 /*
1039 * nfs lookup call, one step at a time...
1040 * First look in cache
1041 * If not found, unlock the directory nfsnode and do the rpc
1042 */
1043 static int
1044 nfs_lookup(struct vop_lookup_args *ap)
1045 {
1046 struct componentname *cnp = ap->a_cnp;
1047 struct vnode *dvp = ap->a_dvp;
1048 struct vnode **vpp = ap->a_vpp;
1049 struct mount *mp = dvp->v_mount;
1050 int flags = cnp->cn_flags;
1051 struct vnode *newvp;
1052 struct nfsmount *nmp;
1053 struct nfsnode *np, *newnp;
1054 int error = 0, attrflag, dattrflag, ltype, ncticks;
1055 struct thread *td = cnp->cn_thread;
1056 struct nfsfh *nfhp;
1057 struct nfsvattr dnfsva, nfsva;
1058 struct vattr vattr;
1059 struct timespec nctime;
1060
1061 *vpp = NULLVP;
1062 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1063 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1064 return (EROFS);
1065 if (dvp->v_type != VDIR)
1066 return (ENOTDIR);
1067 nmp = VFSTONFS(mp);
1068 np = VTONFS(dvp);
1069
1070 /* For NFSv4, wait until any remove is done. */
1071 mtx_lock(&np->n_mtx);
1072 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1073 np->n_flag |= NREMOVEWANT;
1074 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1075 }
1076 mtx_unlock(&np->n_mtx);
1077
1078 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1079 return (error);
1080 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1081 if (error > 0 && error != ENOENT)
1082 return (error);
1083 if (error == -1) {
1084 /*
1085 * Lookups of "." are special and always return the
1086 * current directory. cache_lookup() already handles
1087 * associated locking bookkeeping, etc.
1088 */
1089 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1090 /* XXX: Is this really correct? */
1091 if (cnp->cn_nameiop != LOOKUP &&
1092 (flags & ISLASTCN))
1093 cnp->cn_flags |= SAVENAME;
1094 return (0);
1095 }
1096
1097 /*
1098 * We only accept a positive hit in the cache if the
1099 * change time of the file matches our cached copy.
1100 * Otherwise, we discard the cache entry and fallback
1101 * to doing a lookup RPC. We also only trust cache
1102 * entries for less than nm_nametimeo seconds.
1103 *
1104 * To better handle stale file handles and attributes,
1105 * clear the attribute cache of this node if it is a
1106 * leaf component, part of an open() call, and not
1107 * locally modified before fetching the attributes.
1108 * This should allow stale file handles to be detected
1109 * here where we can fall back to a LOOKUP RPC to
1110 * recover rather than having nfs_open() detect the
1111 * stale file handle and failing open(2) with ESTALE.
1112 */
1113 newvp = *vpp;
1114 newnp = VTONFS(newvp);
1115 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1116 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1117 !(newnp->n_flag & NMODIFIED)) {
1118 mtx_lock(&newnp->n_mtx);
1119 newnp->n_attrstamp = 0;
1120 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1121 mtx_unlock(&newnp->n_mtx);
1122 }
1123 if (nfscl_nodeleg(newvp, 0) == 0 ||
1124 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1125 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1126 timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1127 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1128 if (cnp->cn_nameiop != LOOKUP &&
1129 (flags & ISLASTCN))
1130 cnp->cn_flags |= SAVENAME;
1131 return (0);
1132 }
1133 cache_purge(newvp);
1134 if (dvp != newvp)
1135 vput(newvp);
1136 else
1137 vrele(newvp);
1138 *vpp = NULLVP;
1139 } else if (error == ENOENT) {
1140 if (dvp->v_iflag & VI_DOOMED)
1141 return (ENOENT);
1142 /*
1143 * We only accept a negative hit in the cache if the
1144 * modification time of the parent directory matches
1145 * the cached copy in the name cache entry.
1146 * Otherwise, we discard all of the negative cache
1147 * entries for this directory. We also only trust
1148 * negative cache entries for up to nm_negnametimeo
1149 * seconds.
1150 */
1151 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1152 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1153 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1154 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1155 return (ENOENT);
1156 }
1157 cache_purge_negative(dvp);
1158 }
1159
1160 error = 0;
1161 newvp = NULLVP;
1162 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1163 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1164 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1165 NULL);
1166 if (dattrflag)
1167 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1168 if (error) {
1169 if (newvp != NULLVP) {
1170 vput(newvp);
1171 *vpp = NULLVP;
1172 }
1173
1174 if (error != ENOENT) {
1175 if (NFS_ISV4(dvp))
1176 error = nfscl_maperr(td, error, (uid_t)0,
1177 (gid_t)0);
1178 return (error);
1179 }
1180
1181 /* The requested file was not found. */
1182 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1183 (flags & ISLASTCN)) {
1184 /*
1185 * XXX: UFS does a full VOP_ACCESS(dvp,
1186 * VWRITE) here instead of just checking
1187 * MNT_RDONLY.
1188 */
1189 if (mp->mnt_flag & MNT_RDONLY)
1190 return (EROFS);
1191 cnp->cn_flags |= SAVENAME;
1192 return (EJUSTRETURN);
1193 }
1194
1195 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1196 /*
1197 * Cache the modification time of the parent
1198 * directory from the post-op attributes in
1199 * the name cache entry. The negative cache
1200 * entry will be ignored once the directory
1201 * has changed. Don't bother adding the entry
1202 * if the directory has already changed.
1203 */
1204 mtx_lock(&np->n_mtx);
1205 if (timespeccmp(&np->n_vattr.na_mtime,
1206 &dnfsva.na_mtime, ==)) {
1207 mtx_unlock(&np->n_mtx);
1208 cache_enter_time(dvp, NULL, cnp,
1209 &dnfsva.na_mtime, NULL);
1210 } else
1211 mtx_unlock(&np->n_mtx);
1212 }
1213 return (ENOENT);
1214 }
1215
1216 /*
1217 * Handle RENAME case...
1218 */
1219 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1220 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1221 free(nfhp, M_NFSFH);
1222 return (EISDIR);
1223 }
1224 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1225 LK_EXCLUSIVE);
1226 if (error)
1227 return (error);
1228 newvp = NFSTOV(np);
1229 if (attrflag)
1230 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1231 0, 1);
1232 *vpp = newvp;
1233 cnp->cn_flags |= SAVENAME;
1234 return (0);
1235 }
1236
1237 if (flags & ISDOTDOT) {
1238 ltype = NFSVOPISLOCKED(dvp);
1239 error = vfs_busy(mp, MBF_NOWAIT);
1240 if (error != 0) {
1241 vfs_ref(mp);
1242 NFSVOPUNLOCK(dvp, 0);
1243 error = vfs_busy(mp, 0);
1244 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1245 vfs_rel(mp);
1246 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1247 vfs_unbusy(mp);
1248 error = ENOENT;
1249 }
1250 if (error != 0)
1251 return (error);
1252 }
1253 NFSVOPUNLOCK(dvp, 0);
1254 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1255 cnp->cn_lkflags);
1256 if (error == 0)
1257 newvp = NFSTOV(np);
1258 vfs_unbusy(mp);
1259 if (newvp != dvp)
1260 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1261 if (dvp->v_iflag & VI_DOOMED) {
1262 if (error == 0) {
1263 if (newvp == dvp)
1264 vrele(newvp);
1265 else
1266 vput(newvp);
1267 }
1268 error = ENOENT;
1269 }
1270 if (error != 0)
1271 return (error);
1272 if (attrflag)
1273 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1274 0, 1);
1275 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1276 free(nfhp, M_NFSFH);
1277 VREF(dvp);
1278 newvp = dvp;
1279 if (attrflag)
1280 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1281 0, 1);
1282 } else {
1283 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1284 cnp->cn_lkflags);
1285 if (error)
1286 return (error);
1287 newvp = NFSTOV(np);
1288 if (attrflag)
1289 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1290 0, 1);
1291 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1292 !(np->n_flag & NMODIFIED)) {
1293 /*
1294 * Flush the attribute cache when opening a
1295 * leaf node to ensure that fresh attributes
1296 * are fetched in nfs_open() since we did not
1297 * fetch attributes from the LOOKUP reply.
1298 */
1299 mtx_lock(&np->n_mtx);
1300 np->n_attrstamp = 0;
1301 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1302 mtx_unlock(&np->n_mtx);
1303 }
1304 }
1305 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1306 cnp->cn_flags |= SAVENAME;
1307 if ((cnp->cn_flags & MAKEENTRY) &&
1308 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1309 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1310 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1311 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1312 *vpp = newvp;
1313 return (0);
1314 }
1315
1316 /*
1317 * nfs read call.
1318 * Just call ncl_bioread() to do the work.
1319 */
1320 static int
1321 nfs_read(struct vop_read_args *ap)
1322 {
1323 struct vnode *vp = ap->a_vp;
1324
1325 switch (vp->v_type) {
1326 case VREG:
1327 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1328 case VDIR:
1329 return (EISDIR);
1330 default:
1331 return (EOPNOTSUPP);
1332 }
1333 }
1334
1335 /*
1336 * nfs readlink call
1337 */
1338 static int
1339 nfs_readlink(struct vop_readlink_args *ap)
1340 {
1341 struct vnode *vp = ap->a_vp;
1342
1343 if (vp->v_type != VLNK)
1344 return (EINVAL);
1345 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1346 }
1347
1348 /*
1349 * Do a readlink rpc.
1350 * Called by ncl_doio() from below the buffer cache.
1351 */
1352 int
1353 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1354 {
1355 int error, ret, attrflag;
1356 struct nfsvattr nfsva;
1357
1358 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1359 &attrflag, NULL);
1360 if (attrflag) {
1361 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1362 if (ret && !error)
1363 error = ret;
1364 }
1365 if (error && NFS_ISV4(vp))
1366 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1367 return (error);
1368 }
1369
1370 /*
1371 * nfs read rpc call
1372 * Ditto above
1373 */
1374 int
1375 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1376 {
1377 int error, ret, attrflag;
1378 struct nfsvattr nfsva;
1379 struct nfsmount *nmp;
1380
1381 nmp = VFSTONFS(vnode_mount(vp));
1382 error = EIO;
1383 attrflag = 0;
1384 if (NFSHASPNFS(nmp))
1385 error = nfscl_doiods(vp, uiop, NULL, NULL,
1386 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1387 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1388 if (error != 0)
1389 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1390 &attrflag, NULL);
1391 if (attrflag) {
1392 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1393 if (ret && !error)
1394 error = ret;
1395 }
1396 if (error && NFS_ISV4(vp))
1397 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1398 return (error);
1399 }
1400
1401 /*
1402 * nfs write call
1403 */
1404 int
1405 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1406 int *iomode, int *must_commit, int called_from_strategy)
1407 {
1408 struct nfsvattr nfsva;
1409 int error, attrflag, ret;
1410 struct nfsmount *nmp;
1411
1412 nmp = VFSTONFS(vnode_mount(vp));
1413 error = EIO;
1414 attrflag = 0;
1415 if (NFSHASPNFS(nmp))
1416 error = nfscl_doiods(vp, uiop, iomode, must_commit,
1417 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1418 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1419 if (error != 0)
1420 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1421 uiop->uio_td, &nfsva, &attrflag, NULL,
1422 called_from_strategy);
1423 if (attrflag) {
1424 if (VTONFS(vp)->n_flag & ND_NFSV4)
1425 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1426 1);
1427 else
1428 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1429 1);
1430 if (ret && !error)
1431 error = ret;
1432 }
1433 if (DOINGASYNC(vp))
1434 *iomode = NFSWRITE_FILESYNC;
1435 if (error && NFS_ISV4(vp))
1436 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1437 return (error);
1438 }
1439
1440 /*
1441 * nfs mknod rpc
1442 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1443 * mode set to specify the file type and the size field for rdev.
1444 */
1445 static int
1446 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1447 struct vattr *vap)
1448 {
1449 struct nfsvattr nfsva, dnfsva;
1450 struct vnode *newvp = NULL;
1451 struct nfsnode *np = NULL, *dnp;
1452 struct nfsfh *nfhp;
1453 struct vattr vattr;
1454 int error = 0, attrflag, dattrflag;
1455 u_int32_t rdev;
1456
1457 if (vap->va_type == VCHR || vap->va_type == VBLK)
1458 rdev = vap->va_rdev;
1459 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1460 rdev = 0xffffffff;
1461 else
1462 return (EOPNOTSUPP);
1463 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1464 return (error);
1465 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1466 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1467 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1468 if (!error) {
1469 if (!nfhp)
1470 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1471 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1472 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1473 NULL);
1474 if (nfhp)
1475 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1476 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1477 }
1478 if (dattrflag)
1479 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1480 if (!error) {
1481 newvp = NFSTOV(np);
1482 if (attrflag != 0) {
1483 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1484 0, 1);
1485 if (error != 0)
1486 vput(newvp);
1487 }
1488 }
1489 if (!error) {
1490 *vpp = newvp;
1491 } else if (NFS_ISV4(dvp)) {
1492 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1493 vap->va_gid);
1494 }
1495 dnp = VTONFS(dvp);
1496 mtx_lock(&dnp->n_mtx);
1497 dnp->n_flag |= NMODIFIED;
1498 if (!dattrflag) {
1499 dnp->n_attrstamp = 0;
1500 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1501 }
1502 mtx_unlock(&dnp->n_mtx);
1503 return (error);
1504 }
1505
1506 /*
1507 * nfs mknod vop
1508 * just call nfs_mknodrpc() to do the work.
1509 */
1510 /* ARGSUSED */
1511 static int
1512 nfs_mknod(struct vop_mknod_args *ap)
1513 {
1514 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1515 }
1516
1517 static struct mtx nfs_cverf_mtx;
1518 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1519 MTX_DEF);
1520
1521 static nfsquad_t
1522 nfs_get_cverf(void)
1523 {
1524 static nfsquad_t cverf;
1525 nfsquad_t ret;
1526 static int cverf_initialized = 0;
1527
1528 mtx_lock(&nfs_cverf_mtx);
1529 if (cverf_initialized == 0) {
1530 cverf.lval[0] = arc4random();
1531 cverf.lval[1] = arc4random();
1532 cverf_initialized = 1;
1533 } else
1534 cverf.qval++;
1535 ret = cverf;
1536 mtx_unlock(&nfs_cverf_mtx);
1537
1538 return (ret);
1539 }
1540
1541 /*
1542 * nfs file create call
1543 */
1544 static int
1545 nfs_create(struct vop_create_args *ap)
1546 {
1547 struct vnode *dvp = ap->a_dvp;
1548 struct vattr *vap = ap->a_vap;
1549 struct componentname *cnp = ap->a_cnp;
1550 struct nfsnode *np = NULL, *dnp;
1551 struct vnode *newvp = NULL;
1552 struct nfsmount *nmp;
1553 struct nfsvattr dnfsva, nfsva;
1554 struct nfsfh *nfhp;
1555 nfsquad_t cverf;
1556 int error = 0, attrflag, dattrflag, fmode = 0;
1557 struct vattr vattr;
1558
1559 /*
1560 * Oops, not for me..
1561 */
1562 if (vap->va_type == VSOCK)
1563 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1564
1565 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1566 return (error);
1567 if (vap->va_vaflags & VA_EXCLUSIVE)
1568 fmode |= O_EXCL;
1569 dnp = VTONFS(dvp);
1570 nmp = VFSTONFS(vnode_mount(dvp));
1571 again:
1572 /* For NFSv4, wait until any remove is done. */
1573 mtx_lock(&dnp->n_mtx);
1574 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1575 dnp->n_flag |= NREMOVEWANT;
1576 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1577 }
1578 mtx_unlock(&dnp->n_mtx);
1579
1580 cverf = nfs_get_cverf();
1581 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1582 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1583 &nfhp, &attrflag, &dattrflag, NULL);
1584 if (!error) {
1585 if (nfhp == NULL)
1586 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1587 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1588 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1589 NULL);
1590 if (nfhp != NULL)
1591 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1592 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1593 }
1594 if (dattrflag)
1595 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1596 if (!error) {
1597 newvp = NFSTOV(np);
1598 if (attrflag == 0)
1599 error = nfsrpc_getattr(newvp, cnp->cn_cred,
1600 cnp->cn_thread, &nfsva, NULL);
1601 if (error == 0)
1602 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1603 0, 1);
1604 }
1605 if (error) {
1606 if (newvp != NULL) {
1607 vput(newvp);
1608 newvp = NULL;
1609 }
1610 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1611 error == NFSERR_NOTSUPP) {
1612 fmode &= ~O_EXCL;
1613 goto again;
1614 }
1615 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1616 if (nfscl_checksattr(vap, &nfsva)) {
1617 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1618 cnp->cn_thread, &nfsva, &attrflag, NULL);
1619 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1620 vap->va_gid != (gid_t)VNOVAL)) {
1621 /* try again without setting uid/gid */
1622 vap->va_uid = (uid_t)VNOVAL;
1623 vap->va_gid = (uid_t)VNOVAL;
1624 error = nfsrpc_setattr(newvp, vap, NULL,
1625 cnp->cn_cred, cnp->cn_thread, &nfsva,
1626 &attrflag, NULL);
1627 }
1628 if (attrflag)
1629 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1630 NULL, 0, 1);
1631 if (error != 0)
1632 vput(newvp);
1633 }
1634 }
1635 if (!error) {
1636 if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1637 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1638 NULL);
1639 *ap->a_vpp = newvp;
1640 } else if (NFS_ISV4(dvp)) {
1641 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1642 vap->va_gid);
1643 }
1644 mtx_lock(&dnp->n_mtx);
1645 dnp->n_flag |= NMODIFIED;
1646 if (!dattrflag) {
1647 dnp->n_attrstamp = 0;
1648 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1649 }
1650 mtx_unlock(&dnp->n_mtx);
1651 return (error);
1652 }
1653
1654 /*
1655 * nfs file remove call
1656 * To try and make nfs semantics closer to ufs semantics, a file that has
1657 * other processes using the vnode is renamed instead of removed and then
1658 * removed later on the last close.
1659 * - If v_usecount > 1
1660 * If a rename is not already in the works
1661 * call nfs_sillyrename() to set it up
1662 * else
1663 * do the remove rpc
1664 */
1665 static int
1666 nfs_remove(struct vop_remove_args *ap)
1667 {
1668 struct vnode *vp = ap->a_vp;
1669 struct vnode *dvp = ap->a_dvp;
1670 struct componentname *cnp = ap->a_cnp;
1671 struct nfsnode *np = VTONFS(vp);
1672 int error = 0;
1673 struct vattr vattr;
1674
1675 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1676 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1677 if (vp->v_type == VDIR)
1678 error = EPERM;
1679 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1680 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1681 vattr.va_nlink > 1)) {
1682 /*
1683 * Purge the name cache so that the chance of a lookup for
1684 * the name succeeding while the remove is in progress is
1685 * minimized. Without node locking it can still happen, such
1686 * that an I/O op returns ESTALE, but since you get this if
1687 * another host removes the file..
1688 */
1689 cache_purge(vp);
1690 /*
1691 * throw away biocache buffers, mainly to avoid
1692 * unnecessary delayed writes later.
1693 */
1694 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1695 if (error != EINTR && error != EIO)
1696 /* Do the rpc */
1697 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1698 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1699 /*
1700 * Kludge City: If the first reply to the remove rpc is lost..
1701 * the reply to the retransmitted request will be ENOENT
1702 * since the file was in fact removed
1703 * Therefore, we cheat and return success.
1704 */
1705 if (error == ENOENT)
1706 error = 0;
1707 } else if (!np->n_sillyrename)
1708 error = nfs_sillyrename(dvp, vp, cnp);
1709 mtx_lock(&np->n_mtx);
1710 np->n_attrstamp = 0;
1711 mtx_unlock(&np->n_mtx);
1712 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1713 return (error);
1714 }
1715
1716 /*
1717 * nfs file remove rpc called from nfs_inactive
1718 */
1719 int
1720 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1721 {
1722 /*
1723 * Make sure that the directory vnode is still valid.
1724 * XXX we should lock sp->s_dvp here.
1725 */
1726 if (sp->s_dvp->v_type == VBAD)
1727 return (0);
1728 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1729 sp->s_cred, NULL));
1730 }
1731
1732 /*
1733 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1734 */
1735 static int
1736 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1737 int namelen, struct ucred *cred, struct thread *td)
1738 {
1739 struct nfsvattr dnfsva;
1740 struct nfsnode *dnp = VTONFS(dvp);
1741 int error = 0, dattrflag;
1742
1743 mtx_lock(&dnp->n_mtx);
1744 dnp->n_flag |= NREMOVEINPROG;
1745 mtx_unlock(&dnp->n_mtx);
1746 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1747 &dattrflag, NULL);
1748 mtx_lock(&dnp->n_mtx);
1749 if ((dnp->n_flag & NREMOVEWANT)) {
1750 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1751 mtx_unlock(&dnp->n_mtx);
1752 wakeup((caddr_t)dnp);
1753 } else {
1754 dnp->n_flag &= ~NREMOVEINPROG;
1755 mtx_unlock(&dnp->n_mtx);
1756 }
1757 if (dattrflag)
1758 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1759 mtx_lock(&dnp->n_mtx);
1760 dnp->n_flag |= NMODIFIED;
1761 if (!dattrflag) {
1762 dnp->n_attrstamp = 0;
1763 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1764 }
1765 mtx_unlock(&dnp->n_mtx);
1766 if (error && NFS_ISV4(dvp))
1767 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1768 return (error);
1769 }
1770
1771 /*
1772 * nfs file rename call
1773 */
1774 static int
1775 nfs_rename(struct vop_rename_args *ap)
1776 {
1777 struct vnode *fvp = ap->a_fvp;
1778 struct vnode *tvp = ap->a_tvp;
1779 struct vnode *fdvp = ap->a_fdvp;
1780 struct vnode *tdvp = ap->a_tdvp;
1781 struct componentname *tcnp = ap->a_tcnp;
1782 struct componentname *fcnp = ap->a_fcnp;
1783 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1784 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1785 struct nfsv4node *newv4 = NULL;
1786 int error;
1787
1788 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1789 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1790 /* Check for cross-device rename */
1791 if ((fvp->v_mount != tdvp->v_mount) ||
1792 (tvp && (fvp->v_mount != tvp->v_mount))) {
1793 error = EXDEV;
1794 goto out;
1795 }
1796
1797 if (fvp == tvp) {
1798 printf("nfs_rename: fvp == tvp (can't happen)\n");
1799 error = 0;
1800 goto out;
1801 }
1802 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1803 goto out;
1804
1805 /*
1806 * We have to flush B_DELWRI data prior to renaming
1807 * the file. If we don't, the delayed-write buffers
1808 * can be flushed out later after the file has gone stale
1809 * under NFSV3. NFSV2 does not have this problem because
1810 * ( as far as I can tell ) it flushes dirty buffers more
1811 * often.
1812 *
1813 * Skip the rename operation if the fsync fails, this can happen
1814 * due to the server's volume being full, when we pushed out data
1815 * that was written back to our cache earlier. Not checking for
1816 * this condition can result in potential (silent) data loss.
1817 */
1818 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1819 NFSVOPUNLOCK(fvp, 0);
1820 if (!error && tvp)
1821 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1822 if (error)
1823 goto out;
1824
1825 /*
1826 * If the tvp exists and is in use, sillyrename it before doing the
1827 * rename of the new file over it.
1828 * XXX Can't sillyrename a directory.
1829 */
1830 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1831 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1832 vput(tvp);
1833 tvp = NULL;
1834 }
1835
1836 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1837 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1838 tcnp->cn_thread);
1839
1840 if (error == 0 && NFS_ISV4(tdvp)) {
1841 /*
1842 * For NFSv4, check to see if it is the same name and
1843 * replace the name, if it is different.
1844 */
1845 newv4 = malloc(
1846 sizeof (struct nfsv4node) +
1847 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1848 M_NFSV4NODE, M_WAITOK);
1849 mtx_lock(&tdnp->n_mtx);
1850 mtx_lock(&fnp->n_mtx);
1851 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1852 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1853 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1854 tcnp->cn_namelen) ||
1855 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1856 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1857 tdnp->n_fhp->nfh_len))) {
1858 #ifdef notdef
1859 { char nnn[100]; int nnnl;
1860 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1861 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1862 nnn[nnnl] = '\0';
1863 printf("ren replace=%s\n",nnn);
1864 }
1865 #endif
1866 free(fnp->n_v4, M_NFSV4NODE);
1867 fnp->n_v4 = newv4;
1868 newv4 = NULL;
1869 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1870 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1871 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1872 tdnp->n_fhp->nfh_len);
1873 NFSBCOPY(tcnp->cn_nameptr,
1874 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1875 }
1876 mtx_unlock(&tdnp->n_mtx);
1877 mtx_unlock(&fnp->n_mtx);
1878 if (newv4 != NULL)
1879 free(newv4, M_NFSV4NODE);
1880 }
1881
1882 if (fvp->v_type == VDIR) {
1883 if (tvp != NULL && tvp->v_type == VDIR)
1884 cache_purge(tdvp);
1885 cache_purge(fdvp);
1886 }
1887
1888 out:
1889 if (tdvp == tvp)
1890 vrele(tdvp);
1891 else
1892 vput(tdvp);
1893 if (tvp)
1894 vput(tvp);
1895 vrele(fdvp);
1896 vrele(fvp);
1897 /*
1898 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1899 */
1900 if (error == ENOENT)
1901 error = 0;
1902 return (error);
1903 }
1904
1905 /*
1906 * nfs file rename rpc called from nfs_remove() above
1907 */
1908 static int
1909 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1910 struct sillyrename *sp)
1911 {
1912
1913 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1914 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1915 scnp->cn_thread));
1916 }
1917
1918 /*
1919 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1920 */
1921 static int
1922 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1923 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1924 int tnamelen, struct ucred *cred, struct thread *td)
1925 {
1926 struct nfsvattr fnfsva, tnfsva;
1927 struct nfsnode *fdnp = VTONFS(fdvp);
1928 struct nfsnode *tdnp = VTONFS(tdvp);
1929 int error = 0, fattrflag, tattrflag;
1930
1931 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1932 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1933 &tattrflag, NULL, NULL);
1934 mtx_lock(&fdnp->n_mtx);
1935 fdnp->n_flag |= NMODIFIED;
1936 if (fattrflag != 0) {
1937 mtx_unlock(&fdnp->n_mtx);
1938 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1939 } else {
1940 fdnp->n_attrstamp = 0;
1941 mtx_unlock(&fdnp->n_mtx);
1942 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1943 }
1944 mtx_lock(&tdnp->n_mtx);
1945 tdnp->n_flag |= NMODIFIED;
1946 if (tattrflag != 0) {
1947 mtx_unlock(&tdnp->n_mtx);
1948 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1949 } else {
1950 tdnp->n_attrstamp = 0;
1951 mtx_unlock(&tdnp->n_mtx);
1952 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1953 }
1954 if (error && NFS_ISV4(fdvp))
1955 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1956 return (error);
1957 }
1958
1959 /*
1960 * nfs hard link create call
1961 */
1962 static int
1963 nfs_link(struct vop_link_args *ap)
1964 {
1965 struct vnode *vp = ap->a_vp;
1966 struct vnode *tdvp = ap->a_tdvp;
1967 struct componentname *cnp = ap->a_cnp;
1968 struct nfsnode *np, *tdnp;
1969 struct nfsvattr nfsva, dnfsva;
1970 int error = 0, attrflag, dattrflag;
1971
1972 /*
1973 * Push all writes to the server, so that the attribute cache
1974 * doesn't get "out of sync" with the server.
1975 * XXX There should be a better way!
1976 */
1977 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1978
1979 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1980 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1981 &dattrflag, NULL);
1982 tdnp = VTONFS(tdvp);
1983 mtx_lock(&tdnp->n_mtx);
1984 tdnp->n_flag |= NMODIFIED;
1985 if (dattrflag != 0) {
1986 mtx_unlock(&tdnp->n_mtx);
1987 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1988 } else {
1989 tdnp->n_attrstamp = 0;
1990 mtx_unlock(&tdnp->n_mtx);
1991 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1992 }
1993 if (attrflag)
1994 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1995 else {
1996 np = VTONFS(vp);
1997 mtx_lock(&np->n_mtx);
1998 np->n_attrstamp = 0;
1999 mtx_unlock(&np->n_mtx);
2000 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2001 }
2002 /*
2003 * If negative lookup caching is enabled, I might as well
2004 * add an entry for this node. Not necessary for correctness,
2005 * but if negative caching is enabled, then the system
2006 * must care about lookup caching hit rate, so...
2007 */
2008 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2009 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2010 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2011 }
2012 if (error && NFS_ISV4(vp))
2013 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2014 (gid_t)0);
2015 return (error);
2016 }
2017
2018 /*
2019 * nfs symbolic link create call
2020 */
2021 static int
2022 nfs_symlink(struct vop_symlink_args *ap)
2023 {
2024 struct vnode *dvp = ap->a_dvp;
2025 struct vattr *vap = ap->a_vap;
2026 struct componentname *cnp = ap->a_cnp;
2027 struct nfsvattr nfsva, dnfsva;
2028 struct nfsfh *nfhp;
2029 struct nfsnode *np = NULL, *dnp;
2030 struct vnode *newvp = NULL;
2031 int error = 0, attrflag, dattrflag, ret;
2032
2033 vap->va_type = VLNK;
2034 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2035 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2036 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2037 if (nfhp) {
2038 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2039 &np, NULL, LK_EXCLUSIVE);
2040 if (!ret)
2041 newvp = NFSTOV(np);
2042 else if (!error)
2043 error = ret;
2044 }
2045 if (newvp != NULL) {
2046 if (attrflag)
2047 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2048 0, 1);
2049 } else if (!error) {
2050 /*
2051 * If we do not have an error and we could not extract the
2052 * newvp from the response due to the request being NFSv2, we
2053 * have to do a lookup in order to obtain a newvp to return.
2054 */
2055 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2056 cnp->cn_cred, cnp->cn_thread, &np);
2057 if (!error)
2058 newvp = NFSTOV(np);
2059 }
2060 if (error) {
2061 if (newvp)
2062 vput(newvp);
2063 if (NFS_ISV4(dvp))
2064 error = nfscl_maperr(cnp->cn_thread, error,
2065 vap->va_uid, vap->va_gid);
2066 } else {
2067 *ap->a_vpp = newvp;
2068 }
2069
2070 dnp = VTONFS(dvp);
2071 mtx_lock(&dnp->n_mtx);
2072 dnp->n_flag |= NMODIFIED;
2073 if (dattrflag != 0) {
2074 mtx_unlock(&dnp->n_mtx);
2075 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2076 } else {
2077 dnp->n_attrstamp = 0;
2078 mtx_unlock(&dnp->n_mtx);
2079 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2080 }
2081 /*
2082 * If negative lookup caching is enabled, I might as well
2083 * add an entry for this node. Not necessary for correctness,
2084 * but if negative caching is enabled, then the system
2085 * must care about lookup caching hit rate, so...
2086 */
2087 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2088 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2089 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2090 }
2091 return (error);
2092 }
2093
2094 /*
2095 * nfs make dir call
2096 */
2097 static int
2098 nfs_mkdir(struct vop_mkdir_args *ap)
2099 {
2100 struct vnode *dvp = ap->a_dvp;
2101 struct vattr *vap = ap->a_vap;
2102 struct componentname *cnp = ap->a_cnp;
2103 struct nfsnode *np = NULL, *dnp;
2104 struct vnode *newvp = NULL;
2105 struct vattr vattr;
2106 struct nfsfh *nfhp;
2107 struct nfsvattr nfsva, dnfsva;
2108 int error = 0, attrflag, dattrflag, ret;
2109
2110 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2111 return (error);
2112 vap->va_type = VDIR;
2113 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2114 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2115 &attrflag, &dattrflag, NULL);
2116 dnp = VTONFS(dvp);
2117 mtx_lock(&dnp->n_mtx);
2118 dnp->n_flag |= NMODIFIED;
2119 if (dattrflag != 0) {
2120 mtx_unlock(&dnp->n_mtx);
2121 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2122 } else {
2123 dnp->n_attrstamp = 0;
2124 mtx_unlock(&dnp->n_mtx);
2125 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2126 }
2127 if (nfhp) {
2128 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2129 &np, NULL, LK_EXCLUSIVE);
2130 if (!ret) {
2131 newvp = NFSTOV(np);
2132 if (attrflag)
2133 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2134 NULL, 0, 1);
2135 } else if (!error)
2136 error = ret;
2137 }
2138 if (!error && newvp == NULL) {
2139 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2140 cnp->cn_cred, cnp->cn_thread, &np);
2141 if (!error) {
2142 newvp = NFSTOV(np);
2143 if (newvp->v_type != VDIR)
2144 error = EEXIST;
2145 }
2146 }
2147 if (error) {
2148 if (newvp)
2149 vput(newvp);
2150 if (NFS_ISV4(dvp))
2151 error = nfscl_maperr(cnp->cn_thread, error,
2152 vap->va_uid, vap->va_gid);
2153 } else {
2154 /*
2155 * If negative lookup caching is enabled, I might as well
2156 * add an entry for this node. Not necessary for correctness,
2157 * but if negative caching is enabled, then the system
2158 * must care about lookup caching hit rate, so...
2159 */
2160 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2161 (cnp->cn_flags & MAKEENTRY) &&
2162 attrflag != 0 && dattrflag != 0)
2163 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2164 &dnfsva.na_ctime);
2165 *ap->a_vpp = newvp;
2166 }
2167 return (error);
2168 }
2169
2170 /*
2171 * nfs remove directory call
2172 */
2173 static int
2174 nfs_rmdir(struct vop_rmdir_args *ap)
2175 {
2176 struct vnode *vp = ap->a_vp;
2177 struct vnode *dvp = ap->a_dvp;
2178 struct componentname *cnp = ap->a_cnp;
2179 struct nfsnode *dnp;
2180 struct nfsvattr dnfsva;
2181 int error, dattrflag;
2182
2183 if (dvp == vp)
2184 return (EINVAL);
2185 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2186 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2187 dnp = VTONFS(dvp);
2188 mtx_lock(&dnp->n_mtx);
2189 dnp->n_flag |= NMODIFIED;
2190 if (dattrflag != 0) {
2191 mtx_unlock(&dnp->n_mtx);
2192 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2193 } else {
2194 dnp->n_attrstamp = 0;
2195 mtx_unlock(&dnp->n_mtx);
2196 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2197 }
2198
2199 cache_purge(dvp);
2200 cache_purge(vp);
2201 if (error && NFS_ISV4(dvp))
2202 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2203 (gid_t)0);
2204 /*
2205 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2206 */
2207 if (error == ENOENT)
2208 error = 0;
2209 return (error);
2210 }
2211
2212 /*
2213 * nfs readdir call
2214 */
2215 static int
2216 nfs_readdir(struct vop_readdir_args *ap)
2217 {
2218 struct vnode *vp = ap->a_vp;
2219 struct nfsnode *np = VTONFS(vp);
2220 struct uio *uio = ap->a_uio;
2221 ssize_t tresid, left;
2222 int error = 0;
2223 struct vattr vattr;
2224
2225 if (ap->a_eofflag != NULL)
2226 *ap->a_eofflag = 0;
2227 if (vp->v_type != VDIR)
2228 return(EPERM);
2229
2230 /*
2231 * First, check for hit on the EOF offset cache
2232 */
2233 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2234 (np->n_flag & NMODIFIED) == 0) {
2235 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2236 mtx_lock(&np->n_mtx);
2237 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2238 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2239 mtx_unlock(&np->n_mtx);
2240 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2241 if (ap->a_eofflag != NULL)
2242 *ap->a_eofflag = 1;
2243 return (0);
2244 } else
2245 mtx_unlock(&np->n_mtx);
2246 }
2247 }
2248
2249 /*
2250 * NFS always guarantees that directory entries don't straddle
2251 * DIRBLKSIZ boundaries. As such, we need to limit the size
2252 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2253 * directory entry.
2254 */
2255 left = uio->uio_resid % DIRBLKSIZ;
2256 if (left == uio->uio_resid)
2257 return (EINVAL);
2258 uio->uio_resid -= left;
2259
2260 /*
2261 * Call ncl_bioread() to do the real work.
2262 */
2263 tresid = uio->uio_resid;
2264 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2265
2266 if (!error && uio->uio_resid == tresid) {
2267 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2268 if (ap->a_eofflag != NULL)
2269 *ap->a_eofflag = 1;
2270 }
2271
2272 /* Add the partial DIRBLKSIZ (left) back in. */
2273 uio->uio_resid += left;
2274 return (error);
2275 }
2276
2277 /*
2278 * Readdir rpc call.
2279 * Called from below the buffer cache by ncl_doio().
2280 */
2281 int
2282 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2283 struct thread *td)
2284 {
2285 struct nfsvattr nfsva;
2286 nfsuint64 *cookiep, cookie;
2287 struct nfsnode *dnp = VTONFS(vp);
2288 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2289 int error = 0, eof, attrflag;
2290
2291 KASSERT(uiop->uio_iovcnt == 1 &&
2292 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2293 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2294 ("nfs readdirrpc bad uio"));
2295
2296 /*
2297 * If there is no cookie, assume directory was stale.
2298 */
2299 ncl_dircookie_lock(dnp);
2300 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2301 if (cookiep) {
2302 cookie = *cookiep;
2303 ncl_dircookie_unlock(dnp);
2304 } else {
2305 ncl_dircookie_unlock(dnp);
2306 return (NFSERR_BAD_COOKIE);
2307 }
2308
2309 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2310 (void)ncl_fsinfo(nmp, vp, cred, td);
2311
2312 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2313 &attrflag, &eof, NULL);
2314 if (attrflag)
2315 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2316
2317 if (!error) {
2318 /*
2319 * We are now either at the end of the directory or have filled
2320 * the block.
2321 */
2322 if (eof)
2323 dnp->n_direofoffset = uiop->uio_offset;
2324 else {
2325 if (uiop->uio_resid > 0)
2326 printf("EEK! readdirrpc resid > 0\n");
2327 ncl_dircookie_lock(dnp);
2328 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2329 *cookiep = cookie;
2330 ncl_dircookie_unlock(dnp);
2331 }
2332 } else if (NFS_ISV4(vp)) {
2333 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2334 }
2335 return (error);
2336 }
2337
2338 /*
2339 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2340 */
2341 int
2342 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2343 struct thread *td)
2344 {
2345 struct nfsvattr nfsva;
2346 nfsuint64 *cookiep, cookie;
2347 struct nfsnode *dnp = VTONFS(vp);
2348 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2349 int error = 0, attrflag, eof;
2350
2351 KASSERT(uiop->uio_iovcnt == 1 &&
2352 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2353 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2354 ("nfs readdirplusrpc bad uio"));
2355
2356 /*
2357 * If there is no cookie, assume directory was stale.
2358 */
2359 ncl_dircookie_lock(dnp);
2360 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2361 if (cookiep) {
2362 cookie = *cookiep;
2363 ncl_dircookie_unlock(dnp);
2364 } else {
2365 ncl_dircookie_unlock(dnp);
2366 return (NFSERR_BAD_COOKIE);
2367 }
2368
2369 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2370 (void)ncl_fsinfo(nmp, vp, cred, td);
2371 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2372 &attrflag, &eof, NULL);
2373 if (attrflag)
2374 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2375
2376 if (!error) {
2377 /*
2378 * We are now either at end of the directory or have filled the
2379 * the block.
2380 */
2381 if (eof)
2382 dnp->n_direofoffset = uiop->uio_offset;
2383 else {
2384 if (uiop->uio_resid > 0)
2385 printf("EEK! readdirplusrpc resid > 0\n");
2386 ncl_dircookie_lock(dnp);
2387 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2388 *cookiep = cookie;
2389 ncl_dircookie_unlock(dnp);
2390 }
2391 } else if (NFS_ISV4(vp)) {
2392 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2393 }
2394 return (error);
2395 }
2396
2397 /*
2398 * Silly rename. To make the NFS filesystem that is stateless look a little
2399 * more like the "ufs" a remove of an active vnode is translated to a rename
2400 * to a funny looking filename that is removed by nfs_inactive on the
2401 * nfsnode. There is the potential for another process on a different client
2402 * to create the same funny name between the nfs_lookitup() fails and the
2403 * nfs_rename() completes, but...
2404 */
2405 static int
2406 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2407 {
2408 struct sillyrename *sp;
2409 struct nfsnode *np;
2410 int error;
2411 short pid;
2412 unsigned int lticks;
2413
2414 cache_purge(dvp);
2415 np = VTONFS(vp);
2416 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2417 sp = malloc(sizeof (struct sillyrename),
2418 M_NEWNFSREQ, M_WAITOK);
2419 sp->s_cred = crhold(cnp->cn_cred);
2420 sp->s_dvp = dvp;
2421 VREF(dvp);
2422
2423 /*
2424 * Fudge together a funny name.
2425 * Changing the format of the funny name to accommodate more
2426 * sillynames per directory.
2427 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2428 * CPU ticks since boot.
2429 */
2430 pid = cnp->cn_thread->td_proc->p_pid;
2431 lticks = (unsigned int)ticks;
2432 for ( ; ; ) {
2433 sp->s_namlen = sprintf(sp->s_name,
2434 ".nfs.%08x.%04x4.4", lticks,
2435 pid);
2436 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2437 cnp->cn_thread, NULL))
2438 break;
2439 lticks++;
2440 }
2441 error = nfs_renameit(dvp, vp, cnp, sp);
2442 if (error)
2443 goto bad;
2444 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2445 cnp->cn_thread, &np);
2446 np->n_sillyrename = sp;
2447 return (0);
2448 bad:
2449 vrele(sp->s_dvp);
2450 crfree(sp->s_cred);
2451 free(sp, M_NEWNFSREQ);
2452 return (error);
2453 }
2454
2455 /*
2456 * Look up a file name and optionally either update the file handle or
2457 * allocate an nfsnode, depending on the value of npp.
2458 * npp == NULL --> just do the lookup
2459 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2460 * handled too
2461 * *npp != NULL --> update the file handle in the vnode
2462 */
2463 static int
2464 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2465 struct thread *td, struct nfsnode **npp)
2466 {
2467 struct vnode *newvp = NULL, *vp;
2468 struct nfsnode *np, *dnp = VTONFS(dvp);
2469 struct nfsfh *nfhp, *onfhp;
2470 struct nfsvattr nfsva, dnfsva;
2471 struct componentname cn;
2472 int error = 0, attrflag, dattrflag;
2473 u_int hash;
2474
2475 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2476 &nfhp, &attrflag, &dattrflag, NULL);
2477 if (dattrflag)
2478 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2479 if (npp && !error) {
2480 if (*npp != NULL) {
2481 np = *npp;
2482 vp = NFSTOV(np);
2483 /*
2484 * For NFSv4, check to see if it is the same name and
2485 * replace the name, if it is different.
2486 */
2487 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2488 (np->n_v4->n4_namelen != len ||
2489 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2490 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2491 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2492 dnp->n_fhp->nfh_len))) {
2493 #ifdef notdef
2494 { char nnn[100]; int nnnl;
2495 nnnl = (len < 100) ? len : 99;
2496 bcopy(name, nnn, nnnl);
2497 nnn[nnnl] = '\0';
2498 printf("replace=%s\n",nnn);
2499 }
2500 #endif
2501 free(np->n_v4, M_NFSV4NODE);
2502 np->n_v4 = malloc(
2503 sizeof (struct nfsv4node) +
2504 dnp->n_fhp->nfh_len + len - 1,
2505 M_NFSV4NODE, M_WAITOK);
2506 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2507 np->n_v4->n4_namelen = len;
2508 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2509 dnp->n_fhp->nfh_len);
2510 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2511 }
2512 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2513 FNV1_32_INIT);
2514 onfhp = np->n_fhp;
2515 /*
2516 * Rehash node for new file handle.
2517 */
2518 vfs_hash_rehash(vp, hash);
2519 np->n_fhp = nfhp;
2520 if (onfhp != NULL)
2521 free(onfhp, M_NFSFH);
2522 newvp = NFSTOV(np);
2523 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2524 free(nfhp, M_NFSFH);
2525 VREF(dvp);
2526 newvp = dvp;
2527 } else {
2528 cn.cn_nameptr = name;
2529 cn.cn_namelen = len;
2530 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2531 &np, NULL, LK_EXCLUSIVE);
2532 if (error)
2533 return (error);
2534 newvp = NFSTOV(np);
2535 }
2536 if (!attrflag && *npp == NULL) {
2537 if (newvp == dvp)
2538 vrele(newvp);
2539 else
2540 vput(newvp);
2541 return (ENOENT);
2542 }
2543 if (attrflag)
2544 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2545 0, 1);
2546 }
2547 if (npp && *npp == NULL) {
2548 if (error) {
2549 if (newvp) {
2550 if (newvp == dvp)
2551 vrele(newvp);
2552 else
2553 vput(newvp);
2554 }
2555 } else
2556 *npp = np;
2557 }
2558 if (error && NFS_ISV4(dvp))
2559 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2560 return (error);
2561 }
2562
2563 /*
2564 * Nfs Version 3 and 4 commit rpc
2565 */
2566 int
2567 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2568 struct thread *td)
2569 {
2570 struct nfsvattr nfsva;
2571 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2572 struct nfsnode *np;
2573 struct uio uio;
2574 int error, attrflag;
2575
2576 np = VTONFS(vp);
2577 error = EIO;
2578 attrflag = 0;
2579 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2580 uio.uio_offset = offset;
2581 uio.uio_resid = cnt;
2582 error = nfscl_doiods(vp, &uio, NULL, NULL,
2583 NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2584 if (error != 0) {
2585 mtx_lock(&np->n_mtx);
2586 np->n_flag &= ~NDSCOMMIT;
2587 mtx_unlock(&np->n_mtx);
2588 }
2589 }
2590 if (error != 0) {
2591 mtx_lock(&nmp->nm_mtx);
2592 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2593 mtx_unlock(&nmp->nm_mtx);
2594 return (0);
2595 }
2596 mtx_unlock(&nmp->nm_mtx);
2597 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2598 &attrflag, NULL);
2599 }
2600 if (attrflag != 0)
2601 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2602 0, 1);
2603 if (error != 0 && NFS_ISV4(vp))
2604 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2605 return (error);
2606 }
2607
2608 /*
2609 * Strategy routine.
2610 * For async requests when nfsiod(s) are running, queue the request by
2611 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2612 * request.
2613 */
2614 static int
2615 nfs_strategy(struct vop_strategy_args *ap)
2616 {
2617 struct buf *bp;
2618 struct vnode *vp;
2619 struct ucred *cr;
2620
2621 bp = ap->a_bp;
2622 vp = ap->a_vp;
2623 KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2624 KASSERT(!(bp->b_flags & B_DONE),
2625 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2626 BUF_ASSERT_HELD(bp);
2627
2628 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2629 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2630 DEV_BSIZE);
2631 if (bp->b_iocmd == BIO_READ)
2632 cr = bp->b_rcred;
2633 else
2634 cr = bp->b_wcred;
2635
2636 /*
2637 * If the op is asynchronous and an i/o daemon is waiting
2638 * queue the request, wake it up and wait for completion
2639 * otherwise just do it ourselves.
2640 */
2641 if ((bp->b_flags & B_ASYNC) == 0 ||
2642 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2643 (void) ncl_doio(vp, bp, cr, curthread, 1);
2644 return (0);
2645 }
2646
2647 /*
2648 * fsync vnode op. Just call ncl_flush() with commit == 1.
2649 */
2650 /* ARGSUSED */
2651 static int
2652 nfs_fsync(struct vop_fsync_args *ap)
2653 {
2654
2655 if (ap->a_vp->v_type != VREG) {
2656 /*
2657 * For NFS, metadata is changed synchronously on the server,
2658 * so there is nothing to flush. Also, ncl_flush() clears
2659 * the NMODIFIED flag and that shouldn't be done here for
2660 * directories.
2661 */
2662 return (0);
2663 }
2664 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2665 }
2666
2667 /*
2668 * Flush all the blocks associated with a vnode.
2669 * Walk through the buffer pool and push any dirty pages
2670 * associated with the vnode.
2671 * If the called_from_renewthread argument is TRUE, it has been called
2672 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2673 * waiting for a buffer write to complete.
2674 */
2675 int
2676 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2677 int commit, int called_from_renewthread)
2678 {
2679 struct nfsnode *np = VTONFS(vp);
2680 struct buf *bp;
2681 int i;
2682 struct buf *nbp;
2683 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2684 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2685 int passone = 1, trycnt = 0;
2686 u_quad_t off, endoff, toff;
2687 struct ucred* wcred = NULL;
2688 struct buf **bvec = NULL;
2689 struct bufobj *bo;
2690 #ifndef NFS_COMMITBVECSIZ
2691 #define NFS_COMMITBVECSIZ 20
2692 #endif
2693 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2694 u_int bvecsize = 0, bveccount;
2695
2696 if (called_from_renewthread != 0)
2697 slptimeo = hz;
2698 if (nmp->nm_flag & NFSMNT_INT)
2699 slpflag = PCATCH;
2700 if (!commit)
2701 passone = 0;
2702 bo = &vp->v_bufobj;
2703 /*
2704 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2705 * server, but has not been committed to stable storage on the server
2706 * yet. On the first pass, the byte range is worked out and the commit
2707 * rpc is done. On the second pass, ncl_writebp() is called to do the
2708 * job.
2709 */
2710 again:
2711 off = (u_quad_t)-1;
2712 endoff = 0;
2713 bvecpos = 0;
2714 if (NFS_ISV34(vp) && commit) {
2715 if (bvec != NULL && bvec != bvec_on_stack)
2716 free(bvec, M_TEMP);
2717 /*
2718 * Count up how many buffers waiting for a commit.
2719 */
2720 bveccount = 0;
2721 BO_LOCK(bo);
2722 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2723 if (!BUF_ISLOCKED(bp) &&
2724 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2725 == (B_DELWRI | B_NEEDCOMMIT))
2726 bveccount++;
2727 }
2728 /*
2729 * Allocate space to remember the list of bufs to commit. It is
2730 * important to use M_NOWAIT here to avoid a race with nfs_write.
2731 * If we can't get memory (for whatever reason), we will end up
2732 * committing the buffers one-by-one in the loop below.
2733 */
2734 if (bveccount > NFS_COMMITBVECSIZ) {
2735 /*
2736 * Release the vnode interlock to avoid a lock
2737 * order reversal.
2738 */
2739 BO_UNLOCK(bo);
2740 bvec = (struct buf **)
2741 malloc(bveccount * sizeof(struct buf *),
2742 M_TEMP, M_NOWAIT);
2743 BO_LOCK(bo);
2744 if (bvec == NULL) {
2745 bvec = bvec_on_stack;
2746 bvecsize = NFS_COMMITBVECSIZ;
2747 } else
2748 bvecsize = bveccount;
2749 } else {
2750 bvec = bvec_on_stack;
2751 bvecsize = NFS_COMMITBVECSIZ;
2752 }
2753 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2754 if (bvecpos >= bvecsize)
2755 break;
2756 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2757 nbp = TAILQ_NEXT(bp, b_bobufs);
2758 continue;
2759 }
2760 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2761 (B_DELWRI | B_NEEDCOMMIT)) {
2762 BUF_UNLOCK(bp);
2763 nbp = TAILQ_NEXT(bp, b_bobufs);
2764 continue;
2765 }
2766 BO_UNLOCK(bo);
2767 bremfree(bp);
2768 /*
2769 * Work out if all buffers are using the same cred
2770 * so we can deal with them all with one commit.
2771 *
2772 * NOTE: we are not clearing B_DONE here, so we have
2773 * to do it later on in this routine if we intend to
2774 * initiate I/O on the bp.
2775 *
2776 * Note: to avoid loopback deadlocks, we do not
2777 * assign b_runningbufspace.
2778 */
2779 if (wcred == NULL)
2780 wcred = bp->b_wcred;
2781 else if (wcred != bp->b_wcred)
2782 wcred = NOCRED;
2783 vfs_busy_pages(bp, 1);
2784
2785 BO_LOCK(bo);
2786 /*
2787 * bp is protected by being locked, but nbp is not
2788 * and vfs_busy_pages() may sleep. We have to
2789 * recalculate nbp.
2790 */
2791 nbp = TAILQ_NEXT(bp, b_bobufs);
2792
2793 /*
2794 * A list of these buffers is kept so that the
2795 * second loop knows which buffers have actually
2796 * been committed. This is necessary, since there
2797 * may be a race between the commit rpc and new
2798 * uncommitted writes on the file.
2799 */
2800 bvec[bvecpos++] = bp;
2801 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2802 bp->b_dirtyoff;
2803 if (toff < off)
2804 off = toff;
2805 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2806 if (toff > endoff)
2807 endoff = toff;
2808 }
2809 BO_UNLOCK(bo);
2810 }
2811 if (bvecpos > 0) {
2812 /*
2813 * Commit data on the server, as required.
2814 * If all bufs are using the same wcred, then use that with
2815 * one call for all of them, otherwise commit each one
2816 * separately.
2817 */
2818 if (wcred != NOCRED)
2819 retv = ncl_commit(vp, off, (int)(endoff - off),
2820 wcred, td);
2821 else {
2822 retv = 0;
2823 for (i = 0; i < bvecpos; i++) {
2824 off_t off, size;
2825 bp = bvec[i];
2826 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2827 bp->b_dirtyoff;
2828 size = (u_quad_t)(bp->b_dirtyend
2829 - bp->b_dirtyoff);
2830 retv = ncl_commit(vp, off, (int)size,
2831 bp->b_wcred, td);
2832 if (retv) break;
2833 }
2834 }
2835
2836 if (retv == NFSERR_STALEWRITEVERF)
2837 ncl_clearcommit(vp->v_mount);
2838
2839 /*
2840 * Now, either mark the blocks I/O done or mark the
2841 * blocks dirty, depending on whether the commit
2842 * succeeded.
2843 */
2844 for (i = 0; i < bvecpos; i++) {
2845 bp = bvec[i];
2846 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2847 if (retv) {
2848 /*
2849 * Error, leave B_DELWRI intact
2850 */
2851 vfs_unbusy_pages(bp);
2852 brelse(bp);
2853 } else {
2854 /*
2855 * Success, remove B_DELWRI ( bundirty() ).
2856 *
2857 * b_dirtyoff/b_dirtyend seem to be NFS
2858 * specific. We should probably move that
2859 * into bundirty(). XXX
2860 */
2861 bufobj_wref(bo);
2862 bp->b_flags |= B_ASYNC;
2863 bundirty(bp);
2864 bp->b_flags &= ~B_DONE;
2865 bp->b_ioflags &= ~BIO_ERROR;
2866 bp->b_dirtyoff = bp->b_dirtyend = 0;
2867 bufdone(bp);
2868 }
2869 }
2870 }
2871
2872 /*
2873 * Start/do any write(s) that are required.
2874 */
2875 loop:
2876 BO_LOCK(bo);
2877 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2878 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2879 if (waitfor != MNT_WAIT || passone)
2880 continue;
2881
2882 error = BUF_TIMELOCK(bp,
2883 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2884 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
2885 if (error == 0) {
2886 BUF_UNLOCK(bp);
2887 goto loop;
2888 }
2889 if (error == ENOLCK) {
2890 error = 0;
2891 goto loop;
2892 }
2893 if (called_from_renewthread != 0) {
2894 /*
2895 * Return EIO so the flush will be retried
2896 * later.
2897 */
2898 error = EIO;
2899 goto done;
2900 }
2901 if (newnfs_sigintr(nmp, td)) {
2902 error = EINTR;
2903 goto done;
2904 }
2905 if (slpflag == PCATCH) {
2906 slpflag = 0;
2907 slptimeo = 2 * hz;
2908 }
2909 goto loop;
2910 }
2911 if ((bp->b_flags & B_DELWRI) == 0)
2912 panic("nfs_fsync: not dirty");
2913 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2914 BUF_UNLOCK(bp);
2915 continue;
2916 }
2917 BO_UNLOCK(bo);
2918 bremfree(bp);
2919 if (passone || !commit)
2920 bp->b_flags |= B_ASYNC;
2921 else
2922 bp->b_flags |= B_ASYNC;
2923 bwrite(bp);
2924 if (newnfs_sigintr(nmp, td)) {
2925 error = EINTR;
2926 goto done;
2927 }
2928 goto loop;
2929 }
2930 if (passone) {
2931 passone = 0;
2932 BO_UNLOCK(bo);
2933 goto again;
2934 }
2935 if (waitfor == MNT_WAIT) {
2936 while (bo->bo_numoutput) {
2937 error = bufobj_wwait(bo, slpflag, slptimeo);
2938 if (error) {
2939 BO_UNLOCK(bo);
2940 if (called_from_renewthread != 0) {
2941 /*
2942 * Return EIO so that the flush will be
2943 * retried later.
2944 */
2945 error = EIO;
2946 goto done;
2947 }
2948 error = newnfs_sigintr(nmp, td);
2949 if (error)
2950 goto done;
2951 if (slpflag == PCATCH) {
2952 slpflag = 0;
2953 slptimeo = 2 * hz;
2954 }
2955 BO_LOCK(bo);
2956 }
2957 }
2958 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2959 BO_UNLOCK(bo);
2960 goto loop;
2961 }
2962 /*
2963 * Wait for all the async IO requests to drain
2964 */
2965 BO_UNLOCK(bo);
2966 mtx_lock(&np->n_mtx);
2967 while (np->n_directio_asyncwr > 0) {
2968 np->n_flag |= NFSYNCWAIT;
2969 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2970 &np->n_mtx, slpflag | (PRIBIO + 1),
2971 "nfsfsync", 0);
2972 if (error) {
2973 if (newnfs_sigintr(nmp, td)) {
2974 mtx_unlock(&np->n_mtx);
2975 error = EINTR;
2976 goto done;
2977 }
2978 }
2979 }
2980 mtx_unlock(&np->n_mtx);
2981 } else
2982 BO_UNLOCK(bo);
2983 if (NFSHASPNFS(nmp)) {
2984 nfscl_layoutcommit(vp, td);
2985 /*
2986 * Invalidate the attribute cache, since writes to a DS
2987 * won't update the size attribute.
2988 */
2989 mtx_lock(&np->n_mtx);
2990 np->n_attrstamp = 0;
2991 } else
2992 mtx_lock(&np->n_mtx);
2993 if (np->n_flag & NWRITEERR) {
2994 error = np->n_error;
2995 np->n_flag &= ~NWRITEERR;
2996 }
2997 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2998 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2999 np->n_flag &= ~NMODIFIED;
3000 mtx_unlock(&np->n_mtx);
3001 done:
3002 if (bvec != NULL && bvec != bvec_on_stack)
3003 free(bvec, M_TEMP);
3004 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3005 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
3006 np->n_directio_asyncwr != 0)) {
3007 if (trycnt++ < 5) {
3008 /* try, try again... */
3009 passone = 1;
3010 wcred = NULL;
3011 bvec = NULL;
3012 bvecsize = 0;
3013 goto again;
3014 }
3015 vn_printf(vp, "ncl_flush failed");
3016 error = called_from_renewthread != 0 ? EIO : EBUSY;
3017 }
3018 return (error);
3019 }
3020
3021 /*
3022 * NFS advisory byte-level locks.
3023 */
3024 static int
3025 nfs_advlock(struct vop_advlock_args *ap)
3026 {
3027 struct vnode *vp = ap->a_vp;
3028 struct ucred *cred;
3029 struct nfsnode *np = VTONFS(ap->a_vp);
3030 struct proc *p = (struct proc *)ap->a_id;
3031 struct thread *td = curthread; /* XXX */
3032 struct vattr va;
3033 int ret, error;
3034 u_quad_t size;
3035
3036 error = NFSVOPLOCK(vp, LK_SHARED);
3037 if (error != 0)
3038 return (EBADF);
3039 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3040 if (vp->v_type != VREG) {
3041 error = EINVAL;
3042 goto out;
3043 }
3044 if ((ap->a_flags & F_POSIX) != 0)
3045 cred = p->p_ucred;
3046 else
3047 cred = td->td_ucred;
3048 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
3049 if (vp->v_iflag & VI_DOOMED) {
3050 error = EBADF;
3051 goto out;
3052 }
3053
3054 /*
3055 * If this is unlocking a write locked region, flush and
3056 * commit them before unlocking. This is required by
3057 * RFC3530 Sec. 9.3.2.
3058 */
3059 if (ap->a_op == F_UNLCK &&
3060 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3061 ap->a_flags))
3062 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3063
3064 /*
3065 * Loop around doing the lock op, while a blocking lock
3066 * must wait for the lock op to succeed.
3067 */
3068 do {
3069 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3070 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3071 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3072 ap->a_op == F_SETLK) {
3073 NFSVOPUNLOCK(vp, 0);
3074 error = nfs_catnap(PZERO | PCATCH, ret,
3075 "ncladvl");
3076 if (error)
3077 return (EINTR);
3078 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3079 if (vp->v_iflag & VI_DOOMED) {
3080 error = EBADF;
3081 goto out;
3082 }
3083 }
3084 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3085 ap->a_op == F_SETLK);
3086 if (ret == NFSERR_DENIED) {
3087 error = EAGAIN;
3088 goto out;
3089 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3090 error = ret;
3091 goto out;
3092 } else if (ret != 0) {
3093 error = EACCES;
3094 goto out;
3095 }
3096
3097 /*
3098 * Now, if we just got a lock, invalidate data in the buffer
3099 * cache, as required, so that the coherency conforms with
3100 * RFC3530 Sec. 9.3.2.
3101 */
3102 if (ap->a_op == F_SETLK) {
3103 if ((np->n_flag & NMODIFIED) == 0) {
3104 np->n_attrstamp = 0;
3105 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3106 ret = VOP_GETATTR(vp, &va, cred);
3107 }
3108 if ((np->n_flag & NMODIFIED) || ret ||
3109 np->n_change != va.va_filerev) {
3110 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3111 np->n_attrstamp = 0;
3112 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3113 ret = VOP_GETATTR(vp, &va, cred);
3114 if (!ret) {
3115 np->n_mtime = va.va_mtime;
3116 np->n_change = va.va_filerev;
3117 }
3118 }
3119 /* Mark that a file lock has been acquired. */
3120 mtx_lock(&np->n_mtx);
3121 np->n_flag |= NHASBEENLOCKED;
3122 mtx_unlock(&np->n_mtx);
3123 }
3124 } else if (!NFS_ISV4(vp)) {
3125 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3126 size = VTONFS(vp)->n_size;
3127 NFSVOPUNLOCK(vp, 0);
3128 error = lf_advlock(ap, &(vp->v_lockf), size);
3129 } else {
3130 if (nfs_advlock_p != NULL)
3131 error = nfs_advlock_p(ap);
3132 else {
3133 NFSVOPUNLOCK(vp, 0);
3134 error = ENOLCK;
3135 }
3136 }
3137 if (error == 0 && ap->a_op == F_SETLK) {
3138 error = NFSVOPLOCK(vp, LK_SHARED);
3139 if (error == 0) {
3140 /* Mark that a file lock has been acquired. */
3141 mtx_lock(&np->n_mtx);
3142 np->n_flag |= NHASBEENLOCKED;
3143 mtx_unlock(&np->n_mtx);
3144 NFSVOPUNLOCK(vp, 0);
3145 }
3146 }
3147 return (error);
3148 } else
3149 error = EOPNOTSUPP;
3150 out:
3151 NFSVOPUNLOCK(vp, 0);
3152 return (error);
3153 }
3154
3155 /*
3156 * NFS advisory byte-level locks.
3157 */
3158 static int
3159 nfs_advlockasync(struct vop_advlockasync_args *ap)
3160 {
3161 struct vnode *vp = ap->a_vp;
3162 u_quad_t size;
3163 int error;
3164
3165 if (NFS_ISV4(vp))
3166 return (EOPNOTSUPP);
3167 error = NFSVOPLOCK(vp, LK_SHARED);
3168 if (error)
3169 return (error);
3170 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3171 size = VTONFS(vp)->n_size;
3172 NFSVOPUNLOCK(vp, 0);
3173 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3174 } else {
3175 NFSVOPUNLOCK(vp, 0);
3176 error = EOPNOTSUPP;
3177 }
3178 return (error);
3179 }
3180
3181 /*
3182 * Print out the contents of an nfsnode.
3183 */
3184 static int
3185 nfs_print(struct vop_print_args *ap)
3186 {
3187 struct vnode *vp = ap->a_vp;
3188 struct nfsnode *np = VTONFS(vp);
3189
3190 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3191 (uintmax_t)np->n_vattr.na_fsid);
3192 if (vp->v_type == VFIFO)
3193 fifo_printinfo(vp);
3194 printf("\n");
3195 return (0);
3196 }
3197
3198 /*
3199 * This is the "real" nfs::bwrite(struct buf*).
3200 * We set B_CACHE if this is a VMIO buffer.
3201 */
3202 int
3203 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3204 {
3205 int oldflags, rtval;
3206
3207 BUF_ASSERT_HELD(bp);
3208
3209 if (bp->b_flags & B_INVAL) {
3210 brelse(bp);
3211 return (0);
3212 }
3213
3214 oldflags = bp->b_flags;
3215 bp->b_flags |= B_CACHE;
3216
3217 /*
3218 * Undirty the bp. We will redirty it later if the I/O fails.
3219 */
3220 bundirty(bp);
3221 bp->b_flags &= ~B_DONE;
3222 bp->b_ioflags &= ~BIO_ERROR;
3223 bp->b_iocmd = BIO_WRITE;
3224
3225 bufobj_wref(bp->b_bufobj);
3226 curthread->td_ru.ru_oublock++;
3227
3228 /*
3229 * Note: to avoid loopback deadlocks, we do not
3230 * assign b_runningbufspace.
3231 */
3232 vfs_busy_pages(bp, 1);
3233
3234 BUF_KERNPROC(bp);
3235 bp->b_iooffset = dbtob(bp->b_blkno);
3236 bstrategy(bp);
3237
3238 if ((oldflags & B_ASYNC) != 0)
3239 return (0);
3240
3241 rtval = bufwait(bp);
3242 if (oldflags & B_DELWRI)
3243 reassignbuf(bp);
3244 brelse(bp);
3245 return (rtval);
3246 }
3247
3248 /*
3249 * nfs special file access vnode op.
3250 * Essentially just get vattr and then imitate iaccess() since the device is
3251 * local to the client.
3252 */
3253 static int
3254 nfsspec_access(struct vop_access_args *ap)
3255 {
3256 struct vattr *vap;
3257 struct ucred *cred = ap->a_cred;
3258 struct vnode *vp = ap->a_vp;
3259 accmode_t accmode = ap->a_accmode;
3260 struct vattr vattr;
3261 int error;
3262
3263 /*
3264 * Disallow write attempts on filesystems mounted read-only;
3265 * unless the file is a socket, fifo, or a block or character
3266 * device resident on the filesystem.
3267 */
3268 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3269 switch (vp->v_type) {
3270 case VREG:
3271 case VDIR:
3272 case VLNK:
3273 return (EROFS);
3274 default:
3275 break;
3276 }
3277 }
3278 vap = &vattr;
3279 error = VOP_GETATTR(vp, vap, cred);
3280 if (error)
3281 goto out;
3282 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3283 accmode, cred, NULL);
3284 out:
3285 return error;
3286 }
3287
3288 /*
3289 * Read wrapper for fifos.
3290 */
3291 static int
3292 nfsfifo_read(struct vop_read_args *ap)
3293 {
3294 struct nfsnode *np = VTONFS(ap->a_vp);
3295 int error;
3296
3297 /*
3298 * Set access flag.
3299 */
3300 mtx_lock(&np->n_mtx);
3301 np->n_flag |= NACC;
3302 vfs_timestamp(&np->n_atim);
3303 mtx_unlock(&np->n_mtx);
3304 error = fifo_specops.vop_read(ap);
3305 return error;
3306 }
3307
3308 /*
3309 * Write wrapper for fifos.
3310 */
3311 static int
3312 nfsfifo_write(struct vop_write_args *ap)
3313 {
3314 struct nfsnode *np = VTONFS(ap->a_vp);
3315
3316 /*
3317 * Set update flag.
3318 */
3319 mtx_lock(&np->n_mtx);
3320 np->n_flag |= NUPD;
3321 vfs_timestamp(&np->n_mtim);
3322 mtx_unlock(&np->n_mtx);
3323 return(fifo_specops.vop_write(ap));
3324 }
3325
3326 /*
3327 * Close wrapper for fifos.
3328 *
3329 * Update the times on the nfsnode then do fifo close.
3330 */
3331 static int
3332 nfsfifo_close(struct vop_close_args *ap)
3333 {
3334 struct vnode *vp = ap->a_vp;
3335 struct nfsnode *np = VTONFS(vp);
3336 struct vattr vattr;
3337 struct timespec ts;
3338
3339 mtx_lock(&np->n_mtx);
3340 if (np->n_flag & (NACC | NUPD)) {
3341 vfs_timestamp(&ts);
3342 if (np->n_flag & NACC)
3343 np->n_atim = ts;
3344 if (np->n_flag & NUPD)
3345 np->n_mtim = ts;
3346 np->n_flag |= NCHG;
3347 if (vrefcnt(vp) == 1 &&
3348 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3349 VATTR_NULL(&vattr);
3350 if (np->n_flag & NACC)
3351 vattr.va_atime = np->n_atim;
3352 if (np->n_flag & NUPD)
3353 vattr.va_mtime = np->n_mtim;
3354 mtx_unlock(&np->n_mtx);
3355 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3356 goto out;
3357 }
3358 }
3359 mtx_unlock(&np->n_mtx);
3360 out:
3361 return (fifo_specops.vop_close(ap));
3362 }
3363
3364 /*
3365 * Just call ncl_writebp() with the force argument set to 1.
3366 *
3367 * NOTE: B_DONE may or may not be set in a_bp on call.
3368 */
3369 static int
3370 nfs_bwrite(struct buf *bp)
3371 {
3372
3373 return (ncl_writebp(bp, 1, curthread));
3374 }
3375
3376 struct buf_ops buf_ops_newnfs = {
3377 .bop_name = "buf_ops_nfs",
3378 .bop_write = nfs_bwrite,
3379 .bop_strategy = bufstrategy,
3380 .bop_sync = bufsync,
3381 .bop_bdflush = bufbdflush,
3382 };
3383
3384 static int
3385 nfs_getacl(struct vop_getacl_args *ap)
3386 {
3387 int error;
3388
3389 if (ap->a_type != ACL_TYPE_NFS4)
3390 return (EOPNOTSUPP);
3391 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3392 NULL);
3393 if (error > NFSERR_STALE) {
3394 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3395 error = EPERM;
3396 }
3397 return (error);
3398 }
3399
3400 static int
3401 nfs_setacl(struct vop_setacl_args *ap)
3402 {
3403 int error;
3404
3405 if (ap->a_type != ACL_TYPE_NFS4)
3406 return (EOPNOTSUPP);
3407 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3408 NULL);
3409 if (error > NFSERR_STALE) {
3410 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3411 error = EPERM;
3412 }
3413 return (error);
3414 }
3415
3416 static int
3417 nfs_set_text(struct vop_set_text_args *ap)
3418 {
3419 struct vnode *vp = ap->a_vp;
3420 struct nfsnode *np;
3421
3422 /*
3423 * If the text file has been mmap'd, flush any dirty pages to the
3424 * buffer cache and then...
3425 * Make sure all writes are pushed to the NFS server. If this is not
3426 * done, the modify time of the file can change while the text
3427 * file is being executed. This will cause the process that is
3428 * executing the text file to be terminated.
3429 */
3430 if (vp->v_object != NULL) {
3431 VM_OBJECT_WLOCK(vp->v_object);
3432 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3433 VM_OBJECT_WUNLOCK(vp->v_object);
3434 }
3435
3436 /* Now, flush the buffer cache. */
3437 ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
3438
3439 /* And, finally, make sure that n_mtime is up to date. */
3440 np = VTONFS(vp);
3441 mtx_lock(&np->n_mtx);
3442 np->n_mtime = np->n_vattr.na_mtime;
3443 mtx_unlock(&np->n_mtx);
3444
3445 vp->v_vflag |= VV_TEXT;
3446 return (0);
3447 }
3448
3449 /*
3450 * Return POSIX pathconf information applicable to nfs filesystems.
3451 */
3452 static int
3453 nfs_pathconf(struct vop_pathconf_args *ap)
3454 {
3455 struct nfsv3_pathconf pc;
3456 struct nfsvattr nfsva;
3457 struct vnode *vp = ap->a_vp;
3458 struct thread *td = curthread;
3459 int attrflag, error;
3460
3461 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
3462 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3463 ap->a_name == _PC_NO_TRUNC)) ||
3464 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
3465 /*
3466 * Since only the above 4 a_names are returned by the NFSv3
3467 * Pathconf RPC, there is no point in doing it for others.
3468 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
3469 * be used for _PC_NFS4_ACL as well.
3470 */
3471 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3472 &attrflag, NULL);
3473 if (attrflag != 0)
3474 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3475 1);
3476 if (error != 0)
3477 return (error);
3478 } else {
3479 /*
3480 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3481 * just fake them.
3482 */
3483 pc.pc_linkmax = NFS_LINK_MAX;
3484 pc.pc_namemax = NFS_MAXNAMLEN;
3485 pc.pc_notrunc = 1;
3486 pc.pc_chownrestricted = 1;
3487 pc.pc_caseinsensitive = 0;
3488 pc.pc_casepreserving = 1;
3489 error = 0;
3490 }
3491 switch (ap->a_name) {
3492 case _PC_LINK_MAX:
3493 #ifdef _LP64
3494 *ap->a_retval = pc.pc_linkmax;
3495 #else
3496 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
3497 #endif
3498 break;
3499 case _PC_NAME_MAX:
3500 *ap->a_retval = pc.pc_namemax;
3501 break;
3502 case _PC_PIPE_BUF:
3503 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
3504 *ap->a_retval = PIPE_BUF;
3505 else
3506 error = EINVAL;
3507 break;
3508 case _PC_CHOWN_RESTRICTED:
3509 *ap->a_retval = pc.pc_chownrestricted;
3510 break;
3511 case _PC_NO_TRUNC:
3512 *ap->a_retval = pc.pc_notrunc;
3513 break;
3514 case _PC_ACL_NFS4:
3515 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3516 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3517 *ap->a_retval = 1;
3518 else
3519 *ap->a_retval = 0;
3520 break;
3521 case _PC_ACL_PATH_MAX:
3522 if (NFS_ISV4(vp))
3523 *ap->a_retval = ACL_MAX_ENTRIES;
3524 else
3525 *ap->a_retval = 3;
3526 break;
3527 case _PC_PRIO_IO:
3528 *ap->a_retval = 0;
3529 break;
3530 case _PC_SYNC_IO:
3531 *ap->a_retval = 0;
3532 break;
3533 case _PC_ALLOC_SIZE_MIN:
3534 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3535 break;
3536 case _PC_FILESIZEBITS:
3537 if (NFS_ISV34(vp))
3538 *ap->a_retval = 64;
3539 else
3540 *ap->a_retval = 32;
3541 break;
3542 case _PC_REC_INCR_XFER_SIZE:
3543 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3544 break;
3545 case _PC_REC_MAX_XFER_SIZE:
3546 *ap->a_retval = -1; /* means ``unlimited'' */
3547 break;
3548 case _PC_REC_MIN_XFER_SIZE:
3549 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3550 break;
3551 case _PC_REC_XFER_ALIGN:
3552 *ap->a_retval = PAGE_SIZE;
3553 break;
3554 case _PC_SYMLINK_MAX:
3555 *ap->a_retval = NFS_MAXPATHLEN;
3556 break;
3557
3558 default:
3559 error = vop_stdpathconf(ap);
3560 break;
3561 }
3562 return (error);
3563 }
3564
Cache object: a458e2315755accc427467ca9cf83dc1
|