1 /*-
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37
38 /*
39 * vnode op calls for Sun NFS version 2, 3 and 4
40 */
41
42 #include "opt_kdtrace.h"
43 #include "opt_inet.h"
44
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/systm.h>
48 #include <sys/resourcevar.h>
49 #include <sys/proc.h>
50 #include <sys/mount.h>
51 #include <sys/bio.h>
52 #include <sys/buf.h>
53 #include <sys/jail.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/namei.h>
57 #include <sys/socket.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/fcntl.h>
61 #include <sys/lockf.h>
62 #include <sys/stat.h>
63 #include <sys/sysctl.h>
64 #include <sys/signalvar.h>
65
66 #include <vm/vm.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
69
70 #include <fs/nfs/nfsport.h>
71 #include <fs/nfsclient/nfsnode.h>
72 #include <fs/nfsclient/nfsmount.h>
73 #include <fs/nfsclient/nfs.h>
74 #include <fs/nfsclient/nfs_kdtrace.h>
75
76 #include <net/if.h>
77 #include <netinet/in.h>
78 #include <netinet/in_var.h>
79
80 #include <nfs/nfs_lock.h>
81
82 #ifdef KDTRACE_HOOKS
83 #include <sys/dtrace_bsd.h>
84
85 dtrace_nfsclient_accesscache_flush_probe_func_t
86 dtrace_nfscl_accesscache_flush_done_probe;
87 uint32_t nfscl_accesscache_flush_done_id;
88
89 dtrace_nfsclient_accesscache_get_probe_func_t
90 dtrace_nfscl_accesscache_get_hit_probe,
91 dtrace_nfscl_accesscache_get_miss_probe;
92 uint32_t nfscl_accesscache_get_hit_id;
93 uint32_t nfscl_accesscache_get_miss_id;
94
95 dtrace_nfsclient_accesscache_load_probe_func_t
96 dtrace_nfscl_accesscache_load_done_probe;
97 uint32_t nfscl_accesscache_load_done_id;
98 #endif /* !KDTRACE_HOOKS */
99
100 /* Defs */
101 #define TRUE 1
102 #define FALSE 0
103
104 extern struct nfsstats newnfsstats;
105 extern int nfsrv_useacl;
106 MALLOC_DECLARE(M_NEWNFSREQ);
107
108 /*
109 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
110 * calls are not in getblk() and brelse() so that they would not be necessary
111 * here.
112 */
113 #ifndef B_VMIO
114 #define vfs_busy_pages(bp, f)
115 #endif
116
117 static vop_read_t nfsfifo_read;
118 static vop_write_t nfsfifo_write;
119 static vop_close_t nfsfifo_close;
120 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
121 struct thread *);
122 static vop_lookup_t nfs_lookup;
123 static vop_create_t nfs_create;
124 static vop_mknod_t nfs_mknod;
125 static vop_open_t nfs_open;
126 static vop_pathconf_t nfs_pathconf;
127 static vop_close_t nfs_close;
128 static vop_access_t nfs_access;
129 static vop_getattr_t nfs_getattr;
130 static vop_setattr_t nfs_setattr;
131 static vop_read_t nfs_read;
132 static vop_fsync_t nfs_fsync;
133 static vop_remove_t nfs_remove;
134 static vop_link_t nfs_link;
135 static vop_rename_t nfs_rename;
136 static vop_mkdir_t nfs_mkdir;
137 static vop_rmdir_t nfs_rmdir;
138 static vop_symlink_t nfs_symlink;
139 static vop_readdir_t nfs_readdir;
140 static vop_strategy_t nfs_strategy;
141 static vop_lock1_t nfs_lock1;
142 static int nfs_lookitup(struct vnode *, char *, int,
143 struct ucred *, struct thread *, struct nfsnode **);
144 static int nfs_sillyrename(struct vnode *, struct vnode *,
145 struct componentname *);
146 static vop_access_t nfsspec_access;
147 static vop_readlink_t nfs_readlink;
148 static vop_print_t nfs_print;
149 static vop_advlock_t nfs_advlock;
150 static vop_advlockasync_t nfs_advlockasync;
151 static vop_getacl_t nfs_getacl;
152 static vop_setacl_t nfs_setacl;
153
154 /*
155 * Global vfs data structures for nfs
156 */
157 struct vop_vector newnfs_vnodeops = {
158 .vop_default = &default_vnodeops,
159 .vop_access = nfs_access,
160 .vop_advlock = nfs_advlock,
161 .vop_advlockasync = nfs_advlockasync,
162 .vop_close = nfs_close,
163 .vop_create = nfs_create,
164 .vop_fsync = nfs_fsync,
165 .vop_getattr = nfs_getattr,
166 .vop_getpages = ncl_getpages,
167 .vop_putpages = ncl_putpages,
168 .vop_inactive = ncl_inactive,
169 .vop_link = nfs_link,
170 .vop_lock1 = nfs_lock1,
171 .vop_lookup = nfs_lookup,
172 .vop_mkdir = nfs_mkdir,
173 .vop_mknod = nfs_mknod,
174 .vop_open = nfs_open,
175 .vop_pathconf = nfs_pathconf,
176 .vop_print = nfs_print,
177 .vop_read = nfs_read,
178 .vop_readdir = nfs_readdir,
179 .vop_readlink = nfs_readlink,
180 .vop_reclaim = ncl_reclaim,
181 .vop_remove = nfs_remove,
182 .vop_rename = nfs_rename,
183 .vop_rmdir = nfs_rmdir,
184 .vop_setattr = nfs_setattr,
185 .vop_strategy = nfs_strategy,
186 .vop_symlink = nfs_symlink,
187 .vop_write = ncl_write,
188 .vop_getacl = nfs_getacl,
189 .vop_setacl = nfs_setacl,
190 };
191
192 struct vop_vector newnfs_fifoops = {
193 .vop_default = &fifo_specops,
194 .vop_access = nfsspec_access,
195 .vop_close = nfsfifo_close,
196 .vop_fsync = nfs_fsync,
197 .vop_getattr = nfs_getattr,
198 .vop_inactive = ncl_inactive,
199 .vop_print = nfs_print,
200 .vop_read = nfsfifo_read,
201 .vop_reclaim = ncl_reclaim,
202 .vop_setattr = nfs_setattr,
203 .vop_write = nfsfifo_write,
204 };
205
206 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
207 struct componentname *cnp, struct vattr *vap);
208 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
209 int namelen, struct ucred *cred, struct thread *td);
210 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
211 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
212 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
213 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
214 struct componentname *scnp, struct sillyrename *sp);
215
216 /*
217 * Global variables
218 */
219 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
220
221 SYSCTL_DECL(_vfs_nfs);
222
223 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
224 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
225 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
226
227 static int nfs_prime_access_cache = 0;
228 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
229 &nfs_prime_access_cache, 0,
230 "Prime NFS ACCESS cache when fetching attributes");
231
232 static int newnfs_commit_on_close = 0;
233 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
234 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
235
236 static int nfs_clean_pages_on_close = 1;
237 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
238 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
239
240 int newnfs_directio_enable = 0;
241 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
242 &newnfs_directio_enable, 0, "Enable NFS directio");
243
244 int nfs_keep_dirty_on_error;
245 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
246 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
247
248 /*
249 * This sysctl allows other processes to mmap a file that has been opened
250 * O_DIRECT by a process. In general, having processes mmap the file while
251 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
252 * this by default to prevent DoS attacks - to prevent a malicious user from
253 * opening up files O_DIRECT preventing other users from mmap'ing these
254 * files. "Protected" environments where stricter consistency guarantees are
255 * required can disable this knob. The process that opened the file O_DIRECT
256 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
257 * meaningful.
258 */
259 int newnfs_directio_allow_mmap = 1;
260 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
261 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
262
263 #if 0
264 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
265 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
266
267 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
268 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
269 #endif
270
271 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
272 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
273 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
274
275 /*
276 * SMP Locking Note :
277 * The list of locks after the description of the lock is the ordering
278 * of other locks acquired with the lock held.
279 * np->n_mtx : Protects the fields in the nfsnode.
280 VM Object Lock
281 VI_MTX (acquired indirectly)
282 * nmp->nm_mtx : Protects the fields in the nfsmount.
283 rep->r_mtx
284 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
285 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
286 nmp->nm_mtx
287 rep->r_mtx
288 * rep->r_mtx : Protects the fields in an nfsreq.
289 */
290
291 static int
292 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
293 struct ucred *cred, u_int32_t *retmode)
294 {
295 int error = 0, attrflag, i, lrupos;
296 u_int32_t rmode;
297 struct nfsnode *np = VTONFS(vp);
298 struct nfsvattr nfsva;
299
300 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
301 &rmode, NULL);
302 if (attrflag)
303 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
304 if (!error) {
305 lrupos = 0;
306 mtx_lock(&np->n_mtx);
307 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
308 if (np->n_accesscache[i].uid == cred->cr_uid) {
309 np->n_accesscache[i].mode = rmode;
310 np->n_accesscache[i].stamp = time_second;
311 break;
312 }
313 if (i > 0 && np->n_accesscache[i].stamp <
314 np->n_accesscache[lrupos].stamp)
315 lrupos = i;
316 }
317 if (i == NFS_ACCESSCACHESIZE) {
318 np->n_accesscache[lrupos].uid = cred->cr_uid;
319 np->n_accesscache[lrupos].mode = rmode;
320 np->n_accesscache[lrupos].stamp = time_second;
321 }
322 mtx_unlock(&np->n_mtx);
323 if (retmode != NULL)
324 *retmode = rmode;
325 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
326 } else if (NFS_ISV4(vp)) {
327 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
328 }
329 #ifdef KDTRACE_HOOKS
330 if (error != 0)
331 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
332 error);
333 #endif
334 return (error);
335 }
336
337 /*
338 * nfs access vnode op.
339 * For nfs version 2, just return ok. File accesses may fail later.
340 * For nfs version 3, use the access rpc to check accessibility. If file modes
341 * are changed on the server, accesses might still fail later.
342 */
343 static int
344 nfs_access(struct vop_access_args *ap)
345 {
346 struct vnode *vp = ap->a_vp;
347 int error = 0, i, gotahit;
348 u_int32_t mode, wmode, rmode;
349 int v34 = NFS_ISV34(vp);
350 struct nfsnode *np = VTONFS(vp);
351
352 /*
353 * Disallow write attempts on filesystems mounted read-only;
354 * unless the file is a socket, fifo, or a block or character
355 * device resident on the filesystem.
356 */
357 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
358 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
359 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
360 switch (vp->v_type) {
361 case VREG:
362 case VDIR:
363 case VLNK:
364 return (EROFS);
365 default:
366 break;
367 }
368 }
369 /*
370 * For nfs v3 or v4, check to see if we have done this recently, and if
371 * so return our cached result instead of making an ACCESS call.
372 * If not, do an access rpc, otherwise you are stuck emulating
373 * ufs_access() locally using the vattr. This may not be correct,
374 * since the server may apply other access criteria such as
375 * client uid-->server uid mapping that we do not know about.
376 */
377 if (v34) {
378 if (ap->a_accmode & VREAD)
379 mode = NFSACCESS_READ;
380 else
381 mode = 0;
382 if (vp->v_type != VDIR) {
383 if (ap->a_accmode & VWRITE)
384 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
385 if (ap->a_accmode & VAPPEND)
386 mode |= NFSACCESS_EXTEND;
387 if (ap->a_accmode & VEXEC)
388 mode |= NFSACCESS_EXECUTE;
389 if (ap->a_accmode & VDELETE)
390 mode |= NFSACCESS_DELETE;
391 } else {
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_LOOKUP;
398 if (ap->a_accmode & VDELETE)
399 mode |= NFSACCESS_DELETE;
400 if (ap->a_accmode & VDELETE_CHILD)
401 mode |= NFSACCESS_MODIFY;
402 }
403 /* XXX safety belt, only make blanket request if caching */
404 if (nfsaccess_cache_timeout > 0) {
405 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
406 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
407 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
408 } else {
409 wmode = mode;
410 }
411
412 /*
413 * Does our cached result allow us to give a definite yes to
414 * this request?
415 */
416 gotahit = 0;
417 mtx_lock(&np->n_mtx);
418 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
419 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
420 if (time_second < (np->n_accesscache[i].stamp
421 + nfsaccess_cache_timeout) &&
422 (np->n_accesscache[i].mode & mode) == mode) {
423 NFSINCRGLOBAL(newnfsstats.accesscache_hits);
424 gotahit = 1;
425 }
426 break;
427 }
428 }
429 mtx_unlock(&np->n_mtx);
430 #ifdef KDTRACE_HOOKS
431 if (gotahit != 0)
432 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
433 ap->a_cred->cr_uid, mode);
434 else
435 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
436 ap->a_cred->cr_uid, mode);
437 #endif
438 if (gotahit == 0) {
439 /*
440 * Either a no, or a don't know. Go to the wire.
441 */
442 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
443 error = nfs34_access_otw(vp, wmode, ap->a_td,
444 ap->a_cred, &rmode);
445 if (!error &&
446 (rmode & mode) != mode)
447 error = EACCES;
448 }
449 return (error);
450 } else {
451 if ((error = nfsspec_access(ap)) != 0) {
452 return (error);
453 }
454 /*
455 * Attempt to prevent a mapped root from accessing a file
456 * which it shouldn't. We try to read a byte from the file
457 * if the user is root and the file is not zero length.
458 * After calling nfsspec_access, we should have the correct
459 * file size cached.
460 */
461 mtx_lock(&np->n_mtx);
462 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
463 && VTONFS(vp)->n_size > 0) {
464 struct iovec aiov;
465 struct uio auio;
466 char buf[1];
467
468 mtx_unlock(&np->n_mtx);
469 aiov.iov_base = buf;
470 aiov.iov_len = 1;
471 auio.uio_iov = &aiov;
472 auio.uio_iovcnt = 1;
473 auio.uio_offset = 0;
474 auio.uio_resid = 1;
475 auio.uio_segflg = UIO_SYSSPACE;
476 auio.uio_rw = UIO_READ;
477 auio.uio_td = ap->a_td;
478
479 if (vp->v_type == VREG)
480 error = ncl_readrpc(vp, &auio, ap->a_cred);
481 else if (vp->v_type == VDIR) {
482 char* bp;
483 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
484 aiov.iov_base = bp;
485 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
486 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
487 ap->a_td);
488 free(bp, M_TEMP);
489 } else if (vp->v_type == VLNK)
490 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
491 else
492 error = EACCES;
493 } else
494 mtx_unlock(&np->n_mtx);
495 return (error);
496 }
497 }
498
499
500 /*
501 * nfs open vnode op
502 * Check to see if the type is ok
503 * and that deletion is not in progress.
504 * For paged in text files, you will need to flush the page cache
505 * if consistency is lost.
506 */
507 /* ARGSUSED */
508 static int
509 nfs_open(struct vop_open_args *ap)
510 {
511 struct vnode *vp = ap->a_vp;
512 struct nfsnode *np = VTONFS(vp);
513 struct vattr vattr;
514 int error;
515 int fmode = ap->a_mode;
516 struct ucred *cred;
517
518 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
519 return (EOPNOTSUPP);
520
521 /*
522 * For NFSv4, we need to do the Open Op before cache validation,
523 * so that we conform to RFC3530 Sec. 9.3.1.
524 */
525 if (NFS_ISV4(vp)) {
526 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
527 if (error) {
528 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
529 (gid_t)0);
530 return (error);
531 }
532 }
533
534 /*
535 * Now, if this Open will be doing reading, re-validate/flush the
536 * cache, so that Close/Open coherency is maintained.
537 */
538 mtx_lock(&np->n_mtx);
539 if (np->n_flag & NMODIFIED) {
540 mtx_unlock(&np->n_mtx);
541 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
542 if (error == EINTR || error == EIO) {
543 if (NFS_ISV4(vp))
544 (void) nfsrpc_close(vp, 0, ap->a_td);
545 return (error);
546 }
547 mtx_lock(&np->n_mtx);
548 np->n_attrstamp = 0;
549 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
550 if (vp->v_type == VDIR)
551 np->n_direofoffset = 0;
552 mtx_unlock(&np->n_mtx);
553 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
554 if (error) {
555 if (NFS_ISV4(vp))
556 (void) nfsrpc_close(vp, 0, ap->a_td);
557 return (error);
558 }
559 mtx_lock(&np->n_mtx);
560 np->n_mtime = vattr.va_mtime;
561 if (NFS_ISV4(vp))
562 np->n_change = vattr.va_filerev;
563 } else {
564 mtx_unlock(&np->n_mtx);
565 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
566 if (error) {
567 if (NFS_ISV4(vp))
568 (void) nfsrpc_close(vp, 0, ap->a_td);
569 return (error);
570 }
571 mtx_lock(&np->n_mtx);
572 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
573 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
574 if (vp->v_type == VDIR)
575 np->n_direofoffset = 0;
576 mtx_unlock(&np->n_mtx);
577 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
578 if (error == EINTR || error == EIO) {
579 if (NFS_ISV4(vp))
580 (void) nfsrpc_close(vp, 0, ap->a_td);
581 return (error);
582 }
583 mtx_lock(&np->n_mtx);
584 np->n_mtime = vattr.va_mtime;
585 if (NFS_ISV4(vp))
586 np->n_change = vattr.va_filerev;
587 }
588 }
589
590 /*
591 * If the object has >= 1 O_DIRECT active opens, we disable caching.
592 */
593 if (newnfs_directio_enable && (fmode & O_DIRECT) &&
594 (vp->v_type == VREG)) {
595 if (np->n_directio_opens == 0) {
596 mtx_unlock(&np->n_mtx);
597 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
598 if (error) {
599 if (NFS_ISV4(vp))
600 (void) nfsrpc_close(vp, 0, ap->a_td);
601 return (error);
602 }
603 mtx_lock(&np->n_mtx);
604 np->n_flag |= NNONCACHE;
605 }
606 np->n_directio_opens++;
607 }
608
609 /*
610 * If this is an open for writing, capture a reference to the
611 * credentials, so they can be used by ncl_putpages(). Using
612 * these write credentials is preferable to the credentials of
613 * whatever thread happens to be doing the VOP_PUTPAGES() since
614 * the write RPCs are less likely to fail with EACCES.
615 */
616 if ((fmode & FWRITE) != 0) {
617 cred = np->n_writecred;
618 np->n_writecred = crhold(ap->a_cred);
619 } else
620 cred = NULL;
621 mtx_unlock(&np->n_mtx);
622 if (cred != NULL)
623 crfree(cred);
624 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
625 return (0);
626 }
627
628 /*
629 * nfs close vnode op
630 * What an NFS client should do upon close after writing is a debatable issue.
631 * Most NFS clients push delayed writes to the server upon close, basically for
632 * two reasons:
633 * 1 - So that any write errors may be reported back to the client process
634 * doing the close system call. By far the two most likely errors are
635 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
636 * 2 - To put a worst case upper bound on cache inconsistency between
637 * multiple clients for the file.
638 * There is also a consistency problem for Version 2 of the protocol w.r.t.
639 * not being able to tell if other clients are writing a file concurrently,
640 * since there is no way of knowing if the changed modify time in the reply
641 * is only due to the write for this client.
642 * (NFS Version 3 provides weak cache consistency data in the reply that
643 * should be sufficient to detect and handle this case.)
644 *
645 * The current code does the following:
646 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
647 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
648 * or commit them (this satisfies 1 and 2 except for the
649 * case where the server crashes after this close but
650 * before the commit RPC, which is felt to be "good
651 * enough". Changing the last argument to ncl_flush() to
652 * a 1 would force a commit operation, if it is felt a
653 * commit is necessary now.
654 * for NFS Version 4 - flush the dirty buffers and commit them, if
655 * nfscl_mustflush() says this is necessary.
656 * It is necessary if there is no write delegation held,
657 * in order to satisfy open/close coherency.
658 * If the file isn't cached on local stable storage,
659 * it may be necessary in order to detect "out of space"
660 * errors from the server, if the write delegation
661 * issued by the server doesn't allow the file to grow.
662 */
663 /* ARGSUSED */
664 static int
665 nfs_close(struct vop_close_args *ap)
666 {
667 struct vnode *vp = ap->a_vp;
668 struct nfsnode *np = VTONFS(vp);
669 struct nfsvattr nfsva;
670 struct ucred *cred;
671 int error = 0, ret, localcred = 0;
672 int fmode = ap->a_fflag;
673
674 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
675 return (0);
676 /*
677 * During shutdown, a_cred isn't valid, so just use root.
678 */
679 if (ap->a_cred == NOCRED) {
680 cred = newnfs_getcred();
681 localcred = 1;
682 } else {
683 cred = ap->a_cred;
684 }
685 if (vp->v_type == VREG) {
686 /*
687 * Examine and clean dirty pages, regardless of NMODIFIED.
688 * This closes a major hole in close-to-open consistency.
689 * We want to push out all dirty pages (and buffers) on
690 * close, regardless of whether they were dirtied by
691 * mmap'ed writes or via write().
692 */
693 if (nfs_clean_pages_on_close && vp->v_object) {
694 VM_OBJECT_LOCK(vp->v_object);
695 vm_object_page_clean(vp->v_object, 0, 0, 0);
696 VM_OBJECT_UNLOCK(vp->v_object);
697 }
698 mtx_lock(&np->n_mtx);
699 if (np->n_flag & NMODIFIED) {
700 mtx_unlock(&np->n_mtx);
701 if (NFS_ISV3(vp)) {
702 /*
703 * Under NFSv3 we have dirty buffers to dispose of. We
704 * must flush them to the NFS server. We have the option
705 * of waiting all the way through the commit rpc or just
706 * waiting for the initial write. The default is to only
707 * wait through the initial write so the data is in the
708 * server's cache, which is roughly similar to the state
709 * a standard disk subsystem leaves the file in on close().
710 *
711 * We cannot clear the NMODIFIED bit in np->n_flag due to
712 * potential races with other processes, and certainly
713 * cannot clear it if we don't commit.
714 * These races occur when there is no longer the old
715 * traditional vnode locking implemented for Vnode Ops.
716 */
717 int cm = newnfs_commit_on_close ? 1 : 0;
718 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
719 /* np->n_flag &= ~NMODIFIED; */
720 } else if (NFS_ISV4(vp)) {
721 if (nfscl_mustflush(vp) != 0) {
722 int cm = newnfs_commit_on_close ? 1 : 0;
723 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
724 cm, 0);
725 /*
726 * as above w.r.t races when clearing
727 * NMODIFIED.
728 * np->n_flag &= ~NMODIFIED;
729 */
730 }
731 } else
732 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
733 mtx_lock(&np->n_mtx);
734 }
735 /*
736 * Invalidate the attribute cache in all cases.
737 * An open is going to fetch fresh attrs any way, other procs
738 * on this node that have file open will be forced to do an
739 * otw attr fetch, but this is safe.
740 * --> A user found that their RPC count dropped by 20% when
741 * this was commented out and I can't see any requirement
742 * for it, so I've disabled it when negative lookups are
743 * enabled. (What does this have to do with negative lookup
744 * caching? Well nothing, except it was reported by the
745 * same user that needed negative lookup caching and I wanted
746 * there to be a way to disable it to see if it
747 * is the cause of some caching/coherency issue that might
748 * crop up.)
749 */
750 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
751 np->n_attrstamp = 0;
752 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
753 }
754 if (np->n_flag & NWRITEERR) {
755 np->n_flag &= ~NWRITEERR;
756 error = np->n_error;
757 }
758 mtx_unlock(&np->n_mtx);
759 }
760
761 if (NFS_ISV4(vp)) {
762 /*
763 * Get attributes so "change" is up to date.
764 */
765 if (error == 0 && nfscl_mustflush(vp) != 0 &&
766 vp->v_type == VREG &&
767 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
768 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
769 NULL);
770 if (!ret) {
771 np->n_change = nfsva.na_filerev;
772 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
773 NULL, 0, 0);
774 }
775 }
776
777 /*
778 * and do the close.
779 */
780 ret = nfsrpc_close(vp, 0, ap->a_td);
781 if (!error && ret)
782 error = ret;
783 if (error)
784 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
785 (gid_t)0);
786 }
787 if (newnfs_directio_enable)
788 KASSERT((np->n_directio_asyncwr == 0),
789 ("nfs_close: dirty unflushed (%d) directio buffers\n",
790 np->n_directio_asyncwr));
791 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
792 mtx_lock(&np->n_mtx);
793 KASSERT((np->n_directio_opens > 0),
794 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
795 np->n_directio_opens--;
796 if (np->n_directio_opens == 0)
797 np->n_flag &= ~NNONCACHE;
798 mtx_unlock(&np->n_mtx);
799 }
800 if (localcred)
801 NFSFREECRED(cred);
802 return (error);
803 }
804
805 /*
806 * nfs getattr call from vfs.
807 */
808 static int
809 nfs_getattr(struct vop_getattr_args *ap)
810 {
811 struct vnode *vp = ap->a_vp;
812 struct thread *td = curthread; /* XXX */
813 struct nfsnode *np = VTONFS(vp);
814 int error = 0;
815 struct nfsvattr nfsva;
816 struct vattr *vap = ap->a_vap;
817 struct vattr vattr;
818
819 /*
820 * Update local times for special files.
821 */
822 mtx_lock(&np->n_mtx);
823 if (np->n_flag & (NACC | NUPD))
824 np->n_flag |= NCHG;
825 mtx_unlock(&np->n_mtx);
826 /*
827 * First look in the cache.
828 */
829 if (ncl_getattrcache(vp, &vattr) == 0) {
830 vap->va_type = vattr.va_type;
831 vap->va_mode = vattr.va_mode;
832 vap->va_nlink = vattr.va_nlink;
833 vap->va_uid = vattr.va_uid;
834 vap->va_gid = vattr.va_gid;
835 vap->va_fsid = vattr.va_fsid;
836 vap->va_fileid = vattr.va_fileid;
837 vap->va_size = vattr.va_size;
838 vap->va_blocksize = vattr.va_blocksize;
839 vap->va_atime = vattr.va_atime;
840 vap->va_mtime = vattr.va_mtime;
841 vap->va_ctime = vattr.va_ctime;
842 vap->va_gen = vattr.va_gen;
843 vap->va_flags = vattr.va_flags;
844 vap->va_rdev = vattr.va_rdev;
845 vap->va_bytes = vattr.va_bytes;
846 vap->va_filerev = vattr.va_filerev;
847 /*
848 * Get the local modify time for the case of a write
849 * delegation.
850 */
851 nfscl_deleggetmodtime(vp, &vap->va_mtime);
852 return (0);
853 }
854
855 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
856 nfsaccess_cache_timeout > 0) {
857 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
858 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
859 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
860 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
861 return (0);
862 }
863 }
864 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
865 if (!error)
866 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
867 if (!error) {
868 /*
869 * Get the local modify time for the case of a write
870 * delegation.
871 */
872 nfscl_deleggetmodtime(vp, &vap->va_mtime);
873 } else if (NFS_ISV4(vp)) {
874 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
875 }
876 return (error);
877 }
878
879 /*
880 * nfs setattr call.
881 */
882 static int
883 nfs_setattr(struct vop_setattr_args *ap)
884 {
885 struct vnode *vp = ap->a_vp;
886 struct nfsnode *np = VTONFS(vp);
887 struct thread *td = curthread; /* XXX */
888 struct vattr *vap = ap->a_vap;
889 int error = 0;
890 u_quad_t tsize;
891
892 #ifndef nolint
893 tsize = (u_quad_t)0;
894 #endif
895
896 /*
897 * Setting of flags and marking of atimes are not supported.
898 */
899 if (vap->va_flags != VNOVAL)
900 return (EOPNOTSUPP);
901
902 /*
903 * Disallow write attempts if the filesystem is mounted read-only.
904 */
905 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
906 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
907 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
908 (vp->v_mount->mnt_flag & MNT_RDONLY))
909 return (EROFS);
910 if (vap->va_size != VNOVAL) {
911 switch (vp->v_type) {
912 case VDIR:
913 return (EISDIR);
914 case VCHR:
915 case VBLK:
916 case VSOCK:
917 case VFIFO:
918 if (vap->va_mtime.tv_sec == VNOVAL &&
919 vap->va_atime.tv_sec == VNOVAL &&
920 vap->va_mode == (mode_t)VNOVAL &&
921 vap->va_uid == (uid_t)VNOVAL &&
922 vap->va_gid == (gid_t)VNOVAL)
923 return (0);
924 vap->va_size = VNOVAL;
925 break;
926 default:
927 /*
928 * Disallow write attempts if the filesystem is
929 * mounted read-only.
930 */
931 if (vp->v_mount->mnt_flag & MNT_RDONLY)
932 return (EROFS);
933 /*
934 * We run vnode_pager_setsize() early (why?),
935 * we must set np->n_size now to avoid vinvalbuf
936 * V_SAVE races that might setsize a lower
937 * value.
938 */
939 mtx_lock(&np->n_mtx);
940 tsize = np->n_size;
941 mtx_unlock(&np->n_mtx);
942 error = ncl_meta_setsize(vp, ap->a_cred, td,
943 vap->va_size);
944 mtx_lock(&np->n_mtx);
945 if (np->n_flag & NMODIFIED) {
946 tsize = np->n_size;
947 mtx_unlock(&np->n_mtx);
948 if (vap->va_size == 0)
949 error = ncl_vinvalbuf(vp, 0, td, 1);
950 else
951 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
952 if (error) {
953 vnode_pager_setsize(vp, tsize);
954 return (error);
955 }
956 /*
957 * Call nfscl_delegmodtime() to set the modify time
958 * locally, as required.
959 */
960 nfscl_delegmodtime(vp);
961 } else
962 mtx_unlock(&np->n_mtx);
963 /*
964 * np->n_size has already been set to vap->va_size
965 * in ncl_meta_setsize(). We must set it again since
966 * nfs_loadattrcache() could be called through
967 * ncl_meta_setsize() and could modify np->n_size.
968 */
969 mtx_lock(&np->n_mtx);
970 np->n_vattr.na_size = np->n_size = vap->va_size;
971 mtx_unlock(&np->n_mtx);
972 };
973 } else {
974 mtx_lock(&np->n_mtx);
975 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
976 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
977 mtx_unlock(&np->n_mtx);
978 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
979 (error == EINTR || error == EIO))
980 return (error);
981 } else
982 mtx_unlock(&np->n_mtx);
983 }
984 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
985 if (error && vap->va_size != VNOVAL) {
986 mtx_lock(&np->n_mtx);
987 np->n_size = np->n_vattr.na_size = tsize;
988 vnode_pager_setsize(vp, tsize);
989 mtx_unlock(&np->n_mtx);
990 }
991 return (error);
992 }
993
994 /*
995 * Do an nfs setattr rpc.
996 */
997 static int
998 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
999 struct thread *td)
1000 {
1001 struct nfsnode *np = VTONFS(vp);
1002 int error, ret, attrflag, i;
1003 struct nfsvattr nfsva;
1004
1005 if (NFS_ISV34(vp)) {
1006 mtx_lock(&np->n_mtx);
1007 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1008 np->n_accesscache[i].stamp = 0;
1009 np->n_flag |= NDELEGMOD;
1010 mtx_unlock(&np->n_mtx);
1011 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1012 }
1013 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1014 NULL);
1015 if (attrflag) {
1016 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1017 if (ret && !error)
1018 error = ret;
1019 }
1020 if (error && NFS_ISV4(vp))
1021 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1022 return (error);
1023 }
1024
1025 /*
1026 * nfs lookup call, one step at a time...
1027 * First look in cache
1028 * If not found, unlock the directory nfsnode and do the rpc
1029 */
1030 static int
1031 nfs_lookup(struct vop_lookup_args *ap)
1032 {
1033 struct componentname *cnp = ap->a_cnp;
1034 struct vnode *dvp = ap->a_dvp;
1035 struct vnode **vpp = ap->a_vpp;
1036 struct mount *mp = dvp->v_mount;
1037 int flags = cnp->cn_flags;
1038 struct vnode *newvp;
1039 struct nfsmount *nmp;
1040 struct nfsnode *np, *newnp;
1041 int error = 0, attrflag, dattrflag, ltype, ncticks;
1042 struct thread *td = cnp->cn_thread;
1043 struct nfsfh *nfhp;
1044 struct nfsvattr dnfsva, nfsva;
1045 struct vattr vattr;
1046 struct timespec nctime;
1047
1048 *vpp = NULLVP;
1049 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1050 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1051 return (EROFS);
1052 if (dvp->v_type != VDIR)
1053 return (ENOTDIR);
1054 nmp = VFSTONFS(mp);
1055 np = VTONFS(dvp);
1056
1057 /* For NFSv4, wait until any remove is done. */
1058 mtx_lock(&np->n_mtx);
1059 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1060 np->n_flag |= NREMOVEWANT;
1061 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1062 }
1063 mtx_unlock(&np->n_mtx);
1064
1065 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1066 return (error);
1067 error = cache_lookup_times(dvp, vpp, cnp, &nctime, &ncticks);
1068 if (error > 0 && error != ENOENT)
1069 return (error);
1070 if (error == -1) {
1071 /*
1072 * Lookups of "." are special and always return the
1073 * current directory. cache_lookup() already handles
1074 * associated locking bookkeeping, etc.
1075 */
1076 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1077 /* XXX: Is this really correct? */
1078 if (cnp->cn_nameiop != LOOKUP &&
1079 (flags & ISLASTCN))
1080 cnp->cn_flags |= SAVENAME;
1081 return (0);
1082 }
1083
1084 /*
1085 * We only accept a positive hit in the cache if the
1086 * change time of the file matches our cached copy.
1087 * Otherwise, we discard the cache entry and fallback
1088 * to doing a lookup RPC. We also only trust cache
1089 * entries for less than nm_nametimeo seconds.
1090 *
1091 * To better handle stale file handles and attributes,
1092 * clear the attribute cache of this node if it is a
1093 * leaf component, part of an open() call, and not
1094 * locally modified before fetching the attributes.
1095 * This should allow stale file handles to be detected
1096 * here where we can fall back to a LOOKUP RPC to
1097 * recover rather than having nfs_open() detect the
1098 * stale file handle and failing open(2) with ESTALE.
1099 */
1100 newvp = *vpp;
1101 newnp = VTONFS(newvp);
1102 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1103 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1104 !(newnp->n_flag & NMODIFIED)) {
1105 mtx_lock(&newnp->n_mtx);
1106 newnp->n_attrstamp = 0;
1107 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1108 mtx_unlock(&newnp->n_mtx);
1109 }
1110 if (nfscl_nodeleg(newvp, 0) == 0 ||
1111 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1112 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1113 timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1114 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1115 if (cnp->cn_nameiop != LOOKUP &&
1116 (flags & ISLASTCN))
1117 cnp->cn_flags |= SAVENAME;
1118 return (0);
1119 }
1120 cache_purge(newvp);
1121 if (dvp != newvp)
1122 vput(newvp);
1123 else
1124 vrele(newvp);
1125 *vpp = NULLVP;
1126 } else if (error == ENOENT) {
1127 if (dvp->v_iflag & VI_DOOMED)
1128 return (ENOENT);
1129 /*
1130 * We only accept a negative hit in the cache if the
1131 * modification time of the parent directory matches
1132 * the cached copy in the name cache entry.
1133 * Otherwise, we discard all of the negative cache
1134 * entries for this directory. We also only trust
1135 * negative cache entries for up to nm_negnametimeo
1136 * seconds.
1137 */
1138 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1139 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1140 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1141 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1142 return (ENOENT);
1143 }
1144 cache_purge_negative(dvp);
1145 }
1146
1147 error = 0;
1148 newvp = NULLVP;
1149 NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1150 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1151 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1152 NULL);
1153 if (dattrflag)
1154 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1155 if (error) {
1156 if (newvp != NULLVP) {
1157 vput(newvp);
1158 *vpp = NULLVP;
1159 }
1160
1161 if (error != ENOENT) {
1162 if (NFS_ISV4(dvp))
1163 error = nfscl_maperr(td, error, (uid_t)0,
1164 (gid_t)0);
1165 return (error);
1166 }
1167
1168 /* The requested file was not found. */
1169 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1170 (flags & ISLASTCN)) {
1171 /*
1172 * XXX: UFS does a full VOP_ACCESS(dvp,
1173 * VWRITE) here instead of just checking
1174 * MNT_RDONLY.
1175 */
1176 if (mp->mnt_flag & MNT_RDONLY)
1177 return (EROFS);
1178 cnp->cn_flags |= SAVENAME;
1179 return (EJUSTRETURN);
1180 }
1181
1182 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
1183 dattrflag) {
1184 /*
1185 * Cache the modification time of the parent
1186 * directory from the post-op attributes in
1187 * the name cache entry. The negative cache
1188 * entry will be ignored once the directory
1189 * has changed. Don't bother adding the entry
1190 * if the directory has already changed.
1191 */
1192 mtx_lock(&np->n_mtx);
1193 if (timespeccmp(&np->n_vattr.na_mtime,
1194 &dnfsva.na_mtime, ==)) {
1195 mtx_unlock(&np->n_mtx);
1196 cache_enter_time(dvp, NULL, cnp,
1197 &dnfsva.na_mtime, NULL);
1198 } else
1199 mtx_unlock(&np->n_mtx);
1200 }
1201 return (ENOENT);
1202 }
1203
1204 /*
1205 * Handle RENAME case...
1206 */
1207 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1208 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1209 FREE((caddr_t)nfhp, M_NFSFH);
1210 return (EISDIR);
1211 }
1212 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1213 LK_EXCLUSIVE);
1214 if (error)
1215 return (error);
1216 newvp = NFSTOV(np);
1217 if (attrflag)
1218 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1219 0, 1);
1220 *vpp = newvp;
1221 cnp->cn_flags |= SAVENAME;
1222 return (0);
1223 }
1224
1225 if (flags & ISDOTDOT) {
1226 ltype = NFSVOPISLOCKED(dvp);
1227 error = vfs_busy(mp, MBF_NOWAIT);
1228 if (error != 0) {
1229 vfs_ref(mp);
1230 NFSVOPUNLOCK(dvp, 0);
1231 error = vfs_busy(mp, 0);
1232 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1233 vfs_rel(mp);
1234 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1235 vfs_unbusy(mp);
1236 error = ENOENT;
1237 }
1238 if (error != 0)
1239 return (error);
1240 }
1241 NFSVOPUNLOCK(dvp, 0);
1242 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1243 cnp->cn_lkflags);
1244 if (error == 0)
1245 newvp = NFSTOV(np);
1246 vfs_unbusy(mp);
1247 if (newvp != dvp)
1248 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1249 if (dvp->v_iflag & VI_DOOMED) {
1250 if (error == 0) {
1251 if (newvp == dvp)
1252 vrele(newvp);
1253 else
1254 vput(newvp);
1255 }
1256 error = ENOENT;
1257 }
1258 if (error != 0)
1259 return (error);
1260 if (attrflag)
1261 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1262 0, 1);
1263 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1264 FREE((caddr_t)nfhp, M_NFSFH);
1265 VREF(dvp);
1266 newvp = dvp;
1267 if (attrflag)
1268 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1269 0, 1);
1270 } else {
1271 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1272 cnp->cn_lkflags);
1273 if (error)
1274 return (error);
1275 newvp = NFSTOV(np);
1276 if (attrflag)
1277 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1278 0, 1);
1279 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1280 !(np->n_flag & NMODIFIED)) {
1281 /*
1282 * Flush the attribute cache when opening a
1283 * leaf node to ensure that fresh attributes
1284 * are fetched in nfs_open() since we did not
1285 * fetch attributes from the LOOKUP reply.
1286 */
1287 mtx_lock(&np->n_mtx);
1288 np->n_attrstamp = 0;
1289 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1290 mtx_unlock(&np->n_mtx);
1291 }
1292 }
1293 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1294 cnp->cn_flags |= SAVENAME;
1295 if ((cnp->cn_flags & MAKEENTRY) &&
1296 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1297 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1298 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1299 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1300 *vpp = newvp;
1301 return (0);
1302 }
1303
1304 /*
1305 * nfs read call.
1306 * Just call ncl_bioread() to do the work.
1307 */
1308 static int
1309 nfs_read(struct vop_read_args *ap)
1310 {
1311 struct vnode *vp = ap->a_vp;
1312
1313 switch (vp->v_type) {
1314 case VREG:
1315 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1316 case VDIR:
1317 return (EISDIR);
1318 default:
1319 return (EOPNOTSUPP);
1320 }
1321 }
1322
1323 /*
1324 * nfs readlink call
1325 */
1326 static int
1327 nfs_readlink(struct vop_readlink_args *ap)
1328 {
1329 struct vnode *vp = ap->a_vp;
1330
1331 if (vp->v_type != VLNK)
1332 return (EINVAL);
1333 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1334 }
1335
1336 /*
1337 * Do a readlink rpc.
1338 * Called by ncl_doio() from below the buffer cache.
1339 */
1340 int
1341 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1342 {
1343 int error, ret, attrflag;
1344 struct nfsvattr nfsva;
1345
1346 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1347 &attrflag, NULL);
1348 if (attrflag) {
1349 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1350 if (ret && !error)
1351 error = ret;
1352 }
1353 if (error && NFS_ISV4(vp))
1354 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1355 return (error);
1356 }
1357
1358 /*
1359 * nfs read rpc call
1360 * Ditto above
1361 */
1362 int
1363 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1364 {
1365 int error, ret, attrflag;
1366 struct nfsvattr nfsva;
1367
1368 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1369 NULL);
1370 if (attrflag) {
1371 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1372 if (ret && !error)
1373 error = ret;
1374 }
1375 if (error && NFS_ISV4(vp))
1376 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1377 return (error);
1378 }
1379
1380 /*
1381 * nfs write call
1382 */
1383 int
1384 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1385 int *iomode, int *must_commit, int called_from_strategy)
1386 {
1387 struct nfsvattr nfsva;
1388 int error = 0, attrflag, ret;
1389
1390 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1391 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy);
1392 if (attrflag) {
1393 if (VTONFS(vp)->n_flag & ND_NFSV4)
1394 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1395 1);
1396 else
1397 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1398 1);
1399 if (ret && !error)
1400 error = ret;
1401 }
1402 if (DOINGASYNC(vp))
1403 *iomode = NFSWRITE_FILESYNC;
1404 if (error && NFS_ISV4(vp))
1405 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1406 return (error);
1407 }
1408
1409 /*
1410 * nfs mknod rpc
1411 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1412 * mode set to specify the file type and the size field for rdev.
1413 */
1414 static int
1415 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1416 struct vattr *vap)
1417 {
1418 struct nfsvattr nfsva, dnfsva;
1419 struct vnode *newvp = NULL;
1420 struct nfsnode *np = NULL, *dnp;
1421 struct nfsfh *nfhp;
1422 struct vattr vattr;
1423 int error = 0, attrflag, dattrflag;
1424 u_int32_t rdev;
1425
1426 if (vap->va_type == VCHR || vap->va_type == VBLK)
1427 rdev = vap->va_rdev;
1428 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1429 rdev = 0xffffffff;
1430 else
1431 return (EOPNOTSUPP);
1432 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1433 return (error);
1434 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1435 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1436 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1437 if (!error) {
1438 if (!nfhp)
1439 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1440 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1441 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1442 NULL);
1443 if (nfhp)
1444 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1445 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1446 }
1447 if (dattrflag)
1448 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1449 if (!error) {
1450 newvp = NFSTOV(np);
1451 if (attrflag != 0) {
1452 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1453 0, 1);
1454 if (error != 0)
1455 vput(newvp);
1456 }
1457 }
1458 if (!error) {
1459 *vpp = newvp;
1460 } else if (NFS_ISV4(dvp)) {
1461 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1462 vap->va_gid);
1463 }
1464 dnp = VTONFS(dvp);
1465 mtx_lock(&dnp->n_mtx);
1466 dnp->n_flag |= NMODIFIED;
1467 if (!dattrflag) {
1468 dnp->n_attrstamp = 0;
1469 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1470 }
1471 mtx_unlock(&dnp->n_mtx);
1472 return (error);
1473 }
1474
1475 /*
1476 * nfs mknod vop
1477 * just call nfs_mknodrpc() to do the work.
1478 */
1479 /* ARGSUSED */
1480 static int
1481 nfs_mknod(struct vop_mknod_args *ap)
1482 {
1483 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1484 }
1485
1486 static struct mtx nfs_cverf_mtx;
1487 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1488 MTX_DEF);
1489
1490 static nfsquad_t
1491 nfs_get_cverf(void)
1492 {
1493 static nfsquad_t cverf;
1494 nfsquad_t ret;
1495 static int cverf_initialized = 0;
1496
1497 mtx_lock(&nfs_cverf_mtx);
1498 if (cverf_initialized == 0) {
1499 cverf.lval[0] = arc4random();
1500 cverf.lval[1] = arc4random();
1501 cverf_initialized = 1;
1502 } else
1503 cverf.qval++;
1504 ret = cverf;
1505 mtx_unlock(&nfs_cverf_mtx);
1506
1507 return (ret);
1508 }
1509
1510 /*
1511 * nfs file create call
1512 */
1513 static int
1514 nfs_create(struct vop_create_args *ap)
1515 {
1516 struct vnode *dvp = ap->a_dvp;
1517 struct vattr *vap = ap->a_vap;
1518 struct componentname *cnp = ap->a_cnp;
1519 struct nfsnode *np = NULL, *dnp;
1520 struct vnode *newvp = NULL;
1521 struct nfsmount *nmp;
1522 struct nfsvattr dnfsva, nfsva;
1523 struct nfsfh *nfhp;
1524 nfsquad_t cverf;
1525 int error = 0, attrflag, dattrflag, fmode = 0;
1526 struct vattr vattr;
1527
1528 /*
1529 * Oops, not for me..
1530 */
1531 if (vap->va_type == VSOCK)
1532 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1533
1534 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1535 return (error);
1536 if (vap->va_vaflags & VA_EXCLUSIVE)
1537 fmode |= O_EXCL;
1538 dnp = VTONFS(dvp);
1539 nmp = VFSTONFS(vnode_mount(dvp));
1540 again:
1541 /* For NFSv4, wait until any remove is done. */
1542 mtx_lock(&dnp->n_mtx);
1543 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1544 dnp->n_flag |= NREMOVEWANT;
1545 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1546 }
1547 mtx_unlock(&dnp->n_mtx);
1548
1549 cverf = nfs_get_cverf();
1550 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1551 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1552 &nfhp, &attrflag, &dattrflag, NULL);
1553 if (!error) {
1554 if (nfhp == NULL)
1555 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1556 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1557 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1558 NULL);
1559 if (nfhp != NULL)
1560 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1561 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1562 }
1563 if (dattrflag)
1564 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1565 if (!error) {
1566 newvp = NFSTOV(np);
1567 if (attrflag == 0)
1568 error = nfsrpc_getattr(newvp, cnp->cn_cred,
1569 cnp->cn_thread, &nfsva, NULL);
1570 if (error == 0)
1571 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1572 0, 1);
1573 }
1574 if (error) {
1575 if (newvp != NULL) {
1576 vput(newvp);
1577 newvp = NULL;
1578 }
1579 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1580 error == NFSERR_NOTSUPP) {
1581 fmode &= ~O_EXCL;
1582 goto again;
1583 }
1584 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1585 if (nfscl_checksattr(vap, &nfsva)) {
1586 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1587 cnp->cn_thread, &nfsva, &attrflag, NULL);
1588 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1589 vap->va_gid != (gid_t)VNOVAL)) {
1590 /* try again without setting uid/gid */
1591 vap->va_uid = (uid_t)VNOVAL;
1592 vap->va_gid = (uid_t)VNOVAL;
1593 error = nfsrpc_setattr(newvp, vap, NULL,
1594 cnp->cn_cred, cnp->cn_thread, &nfsva,
1595 &attrflag, NULL);
1596 }
1597 if (attrflag)
1598 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1599 NULL, 0, 1);
1600 if (error != 0)
1601 vput(newvp);
1602 }
1603 }
1604 if (!error) {
1605 if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1606 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1607 NULL);
1608 *ap->a_vpp = newvp;
1609 } else if (NFS_ISV4(dvp)) {
1610 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1611 vap->va_gid);
1612 }
1613 mtx_lock(&dnp->n_mtx);
1614 dnp->n_flag |= NMODIFIED;
1615 if (!dattrflag) {
1616 dnp->n_attrstamp = 0;
1617 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1618 }
1619 mtx_unlock(&dnp->n_mtx);
1620 return (error);
1621 }
1622
1623 /*
1624 * nfs file remove call
1625 * To try and make nfs semantics closer to ufs semantics, a file that has
1626 * other processes using the vnode is renamed instead of removed and then
1627 * removed later on the last close.
1628 * - If v_usecount > 1
1629 * If a rename is not already in the works
1630 * call nfs_sillyrename() to set it up
1631 * else
1632 * do the remove rpc
1633 */
1634 static int
1635 nfs_remove(struct vop_remove_args *ap)
1636 {
1637 struct vnode *vp = ap->a_vp;
1638 struct vnode *dvp = ap->a_dvp;
1639 struct componentname *cnp = ap->a_cnp;
1640 struct nfsnode *np = VTONFS(vp);
1641 int error = 0;
1642 struct vattr vattr;
1643
1644 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1645 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1646 if (vp->v_type == VDIR)
1647 error = EPERM;
1648 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1649 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1650 vattr.va_nlink > 1)) {
1651 /*
1652 * Purge the name cache so that the chance of a lookup for
1653 * the name succeeding while the remove is in progress is
1654 * minimized. Without node locking it can still happen, such
1655 * that an I/O op returns ESTALE, but since you get this if
1656 * another host removes the file..
1657 */
1658 cache_purge(vp);
1659 /*
1660 * throw away biocache buffers, mainly to avoid
1661 * unnecessary delayed writes later.
1662 */
1663 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1664 /* Do the rpc */
1665 if (error != EINTR && error != EIO)
1666 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1667 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1668 /*
1669 * Kludge City: If the first reply to the remove rpc is lost..
1670 * the reply to the retransmitted request will be ENOENT
1671 * since the file was in fact removed
1672 * Therefore, we cheat and return success.
1673 */
1674 if (error == ENOENT)
1675 error = 0;
1676 } else if (!np->n_sillyrename)
1677 error = nfs_sillyrename(dvp, vp, cnp);
1678 mtx_lock(&np->n_mtx);
1679 np->n_attrstamp = 0;
1680 mtx_unlock(&np->n_mtx);
1681 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1682 return (error);
1683 }
1684
1685 /*
1686 * nfs file remove rpc called from nfs_inactive
1687 */
1688 int
1689 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1690 {
1691 /*
1692 * Make sure that the directory vnode is still valid.
1693 * XXX we should lock sp->s_dvp here.
1694 */
1695 if (sp->s_dvp->v_type == VBAD)
1696 return (0);
1697 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1698 sp->s_cred, NULL));
1699 }
1700
1701 /*
1702 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1703 */
1704 static int
1705 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1706 int namelen, struct ucred *cred, struct thread *td)
1707 {
1708 struct nfsvattr dnfsva;
1709 struct nfsnode *dnp = VTONFS(dvp);
1710 int error = 0, dattrflag;
1711
1712 mtx_lock(&dnp->n_mtx);
1713 dnp->n_flag |= NREMOVEINPROG;
1714 mtx_unlock(&dnp->n_mtx);
1715 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1716 &dattrflag, NULL);
1717 mtx_lock(&dnp->n_mtx);
1718 if ((dnp->n_flag & NREMOVEWANT)) {
1719 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1720 mtx_unlock(&dnp->n_mtx);
1721 wakeup((caddr_t)dnp);
1722 } else {
1723 dnp->n_flag &= ~NREMOVEINPROG;
1724 mtx_unlock(&dnp->n_mtx);
1725 }
1726 if (dattrflag)
1727 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1728 mtx_lock(&dnp->n_mtx);
1729 dnp->n_flag |= NMODIFIED;
1730 if (!dattrflag) {
1731 dnp->n_attrstamp = 0;
1732 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1733 }
1734 mtx_unlock(&dnp->n_mtx);
1735 if (error && NFS_ISV4(dvp))
1736 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1737 return (error);
1738 }
1739
1740 /*
1741 * nfs file rename call
1742 */
1743 static int
1744 nfs_rename(struct vop_rename_args *ap)
1745 {
1746 struct vnode *fvp = ap->a_fvp;
1747 struct vnode *tvp = ap->a_tvp;
1748 struct vnode *fdvp = ap->a_fdvp;
1749 struct vnode *tdvp = ap->a_tdvp;
1750 struct componentname *tcnp = ap->a_tcnp;
1751 struct componentname *fcnp = ap->a_fcnp;
1752 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1753 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1754 struct nfsv4node *newv4 = NULL;
1755 int error;
1756
1757 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1758 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1759 /* Check for cross-device rename */
1760 if ((fvp->v_mount != tdvp->v_mount) ||
1761 (tvp && (fvp->v_mount != tvp->v_mount))) {
1762 error = EXDEV;
1763 goto out;
1764 }
1765
1766 if (fvp == tvp) {
1767 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1768 error = 0;
1769 goto out;
1770 }
1771 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1772 goto out;
1773
1774 /*
1775 * We have to flush B_DELWRI data prior to renaming
1776 * the file. If we don't, the delayed-write buffers
1777 * can be flushed out later after the file has gone stale
1778 * under NFSV3. NFSV2 does not have this problem because
1779 * ( as far as I can tell ) it flushes dirty buffers more
1780 * often.
1781 *
1782 * Skip the rename operation if the fsync fails, this can happen
1783 * due to the server's volume being full, when we pushed out data
1784 * that was written back to our cache earlier. Not checking for
1785 * this condition can result in potential (silent) data loss.
1786 */
1787 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1788 NFSVOPUNLOCK(fvp, 0);
1789 if (!error && tvp)
1790 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1791 if (error)
1792 goto out;
1793
1794 /*
1795 * If the tvp exists and is in use, sillyrename it before doing the
1796 * rename of the new file over it.
1797 * XXX Can't sillyrename a directory.
1798 */
1799 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1800 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1801 vput(tvp);
1802 tvp = NULL;
1803 }
1804
1805 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1806 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1807 tcnp->cn_thread);
1808
1809 if (error == 0 && NFS_ISV4(tdvp)) {
1810 /*
1811 * For NFSv4, check to see if it is the same name and
1812 * replace the name, if it is different.
1813 */
1814 MALLOC(newv4, struct nfsv4node *,
1815 sizeof (struct nfsv4node) +
1816 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1817 M_NFSV4NODE, M_WAITOK);
1818 mtx_lock(&tdnp->n_mtx);
1819 mtx_lock(&fnp->n_mtx);
1820 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1821 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1822 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1823 tcnp->cn_namelen) ||
1824 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1825 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1826 tdnp->n_fhp->nfh_len))) {
1827 #ifdef notdef
1828 { char nnn[100]; int nnnl;
1829 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1830 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1831 nnn[nnnl] = '\0';
1832 printf("ren replace=%s\n",nnn);
1833 }
1834 #endif
1835 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1836 fnp->n_v4 = newv4;
1837 newv4 = NULL;
1838 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1839 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1840 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1841 tdnp->n_fhp->nfh_len);
1842 NFSBCOPY(tcnp->cn_nameptr,
1843 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1844 }
1845 mtx_unlock(&tdnp->n_mtx);
1846 mtx_unlock(&fnp->n_mtx);
1847 if (newv4 != NULL)
1848 FREE((caddr_t)newv4, M_NFSV4NODE);
1849 }
1850
1851 if (fvp->v_type == VDIR) {
1852 if (tvp != NULL && tvp->v_type == VDIR)
1853 cache_purge(tdvp);
1854 cache_purge(fdvp);
1855 }
1856
1857 out:
1858 if (tdvp == tvp)
1859 vrele(tdvp);
1860 else
1861 vput(tdvp);
1862 if (tvp)
1863 vput(tvp);
1864 vrele(fdvp);
1865 vrele(fvp);
1866 /*
1867 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1868 */
1869 if (error == ENOENT)
1870 error = 0;
1871 return (error);
1872 }
1873
1874 /*
1875 * nfs file rename rpc called from nfs_remove() above
1876 */
1877 static int
1878 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1879 struct sillyrename *sp)
1880 {
1881
1882 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1883 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1884 scnp->cn_thread));
1885 }
1886
1887 /*
1888 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1889 */
1890 static int
1891 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1892 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1893 int tnamelen, struct ucred *cred, struct thread *td)
1894 {
1895 struct nfsvattr fnfsva, tnfsva;
1896 struct nfsnode *fdnp = VTONFS(fdvp);
1897 struct nfsnode *tdnp = VTONFS(tdvp);
1898 int error = 0, fattrflag, tattrflag;
1899
1900 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1901 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1902 &tattrflag, NULL, NULL);
1903 mtx_lock(&fdnp->n_mtx);
1904 fdnp->n_flag |= NMODIFIED;
1905 if (fattrflag != 0) {
1906 mtx_unlock(&fdnp->n_mtx);
1907 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1908 } else {
1909 fdnp->n_attrstamp = 0;
1910 mtx_unlock(&fdnp->n_mtx);
1911 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1912 }
1913 mtx_lock(&tdnp->n_mtx);
1914 tdnp->n_flag |= NMODIFIED;
1915 if (tattrflag != 0) {
1916 mtx_unlock(&tdnp->n_mtx);
1917 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1918 } else {
1919 tdnp->n_attrstamp = 0;
1920 mtx_unlock(&tdnp->n_mtx);
1921 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1922 }
1923 if (error && NFS_ISV4(fdvp))
1924 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1925 return (error);
1926 }
1927
1928 /*
1929 * nfs hard link create call
1930 */
1931 static int
1932 nfs_link(struct vop_link_args *ap)
1933 {
1934 struct vnode *vp = ap->a_vp;
1935 struct vnode *tdvp = ap->a_tdvp;
1936 struct componentname *cnp = ap->a_cnp;
1937 struct nfsnode *np, *tdnp;
1938 struct nfsvattr nfsva, dnfsva;
1939 int error = 0, attrflag, dattrflag;
1940
1941 if (vp->v_mount != tdvp->v_mount) {
1942 return (EXDEV);
1943 }
1944
1945 /*
1946 * Push all writes to the server, so that the attribute cache
1947 * doesn't get "out of sync" with the server.
1948 * XXX There should be a better way!
1949 */
1950 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1951
1952 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1953 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1954 &dattrflag, NULL);
1955 tdnp = VTONFS(tdvp);
1956 mtx_lock(&tdnp->n_mtx);
1957 tdnp->n_flag |= NMODIFIED;
1958 if (dattrflag != 0) {
1959 mtx_unlock(&tdnp->n_mtx);
1960 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1961 } else {
1962 tdnp->n_attrstamp = 0;
1963 mtx_unlock(&tdnp->n_mtx);
1964 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1965 }
1966 if (attrflag)
1967 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1968 else {
1969 np = VTONFS(vp);
1970 mtx_lock(&np->n_mtx);
1971 np->n_attrstamp = 0;
1972 mtx_unlock(&np->n_mtx);
1973 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1974 }
1975 /*
1976 * If negative lookup caching is enabled, I might as well
1977 * add an entry for this node. Not necessary for correctness,
1978 * but if negative caching is enabled, then the system
1979 * must care about lookup caching hit rate, so...
1980 */
1981 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1982 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
1983 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
1984 }
1985 if (error && NFS_ISV4(vp))
1986 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1987 (gid_t)0);
1988 return (error);
1989 }
1990
1991 /*
1992 * nfs symbolic link create call
1993 */
1994 static int
1995 nfs_symlink(struct vop_symlink_args *ap)
1996 {
1997 struct vnode *dvp = ap->a_dvp;
1998 struct vattr *vap = ap->a_vap;
1999 struct componentname *cnp = ap->a_cnp;
2000 struct nfsvattr nfsva, dnfsva;
2001 struct nfsfh *nfhp;
2002 struct nfsnode *np = NULL, *dnp;
2003 struct vnode *newvp = NULL;
2004 int error = 0, attrflag, dattrflag, ret;
2005
2006 vap->va_type = VLNK;
2007 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2008 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2009 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2010 if (nfhp) {
2011 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2012 &np, NULL, LK_EXCLUSIVE);
2013 if (!ret)
2014 newvp = NFSTOV(np);
2015 else if (!error)
2016 error = ret;
2017 }
2018 if (newvp != NULL) {
2019 if (attrflag)
2020 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2021 0, 1);
2022 } else if (!error) {
2023 /*
2024 * If we do not have an error and we could not extract the
2025 * newvp from the response due to the request being NFSv2, we
2026 * have to do a lookup in order to obtain a newvp to return.
2027 */
2028 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2029 cnp->cn_cred, cnp->cn_thread, &np);
2030 if (!error)
2031 newvp = NFSTOV(np);
2032 }
2033 if (error) {
2034 if (newvp)
2035 vput(newvp);
2036 if (NFS_ISV4(dvp))
2037 error = nfscl_maperr(cnp->cn_thread, error,
2038 vap->va_uid, vap->va_gid);
2039 } else {
2040 *ap->a_vpp = newvp;
2041 }
2042
2043 dnp = VTONFS(dvp);
2044 mtx_lock(&dnp->n_mtx);
2045 dnp->n_flag |= NMODIFIED;
2046 if (dattrflag != 0) {
2047 mtx_unlock(&dnp->n_mtx);
2048 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2049 } else {
2050 dnp->n_attrstamp = 0;
2051 mtx_unlock(&dnp->n_mtx);
2052 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2053 }
2054 /*
2055 * If negative lookup caching is enabled, I might as well
2056 * add an entry for this node. Not necessary for correctness,
2057 * but if negative caching is enabled, then the system
2058 * must care about lookup caching hit rate, so...
2059 */
2060 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2061 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2062 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2063 }
2064 return (error);
2065 }
2066
2067 /*
2068 * nfs make dir call
2069 */
2070 static int
2071 nfs_mkdir(struct vop_mkdir_args *ap)
2072 {
2073 struct vnode *dvp = ap->a_dvp;
2074 struct vattr *vap = ap->a_vap;
2075 struct componentname *cnp = ap->a_cnp;
2076 struct nfsnode *np = NULL, *dnp;
2077 struct vnode *newvp = NULL;
2078 struct vattr vattr;
2079 struct nfsfh *nfhp;
2080 struct nfsvattr nfsva, dnfsva;
2081 int error = 0, attrflag, dattrflag, ret;
2082
2083 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2084 return (error);
2085 vap->va_type = VDIR;
2086 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2087 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2088 &attrflag, &dattrflag, NULL);
2089 dnp = VTONFS(dvp);
2090 mtx_lock(&dnp->n_mtx);
2091 dnp->n_flag |= NMODIFIED;
2092 if (dattrflag != 0) {
2093 mtx_unlock(&dnp->n_mtx);
2094 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2095 } else {
2096 dnp->n_attrstamp = 0;
2097 mtx_unlock(&dnp->n_mtx);
2098 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2099 }
2100 if (nfhp) {
2101 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2102 &np, NULL, LK_EXCLUSIVE);
2103 if (!ret) {
2104 newvp = NFSTOV(np);
2105 if (attrflag)
2106 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2107 NULL, 0, 1);
2108 } else if (!error)
2109 error = ret;
2110 }
2111 if (!error && newvp == NULL) {
2112 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2113 cnp->cn_cred, cnp->cn_thread, &np);
2114 if (!error) {
2115 newvp = NFSTOV(np);
2116 if (newvp->v_type != VDIR)
2117 error = EEXIST;
2118 }
2119 }
2120 if (error) {
2121 if (newvp)
2122 vput(newvp);
2123 if (NFS_ISV4(dvp))
2124 error = nfscl_maperr(cnp->cn_thread, error,
2125 vap->va_uid, vap->va_gid);
2126 } else {
2127 /*
2128 * If negative lookup caching is enabled, I might as well
2129 * add an entry for this node. Not necessary for correctness,
2130 * but if negative caching is enabled, then the system
2131 * must care about lookup caching hit rate, so...
2132 */
2133 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2134 (cnp->cn_flags & MAKEENTRY) &&
2135 attrflag != 0 && dattrflag != 0)
2136 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2137 &dnfsva.na_ctime);
2138 *ap->a_vpp = newvp;
2139 }
2140 return (error);
2141 }
2142
2143 /*
2144 * nfs remove directory call
2145 */
2146 static int
2147 nfs_rmdir(struct vop_rmdir_args *ap)
2148 {
2149 struct vnode *vp = ap->a_vp;
2150 struct vnode *dvp = ap->a_dvp;
2151 struct componentname *cnp = ap->a_cnp;
2152 struct nfsnode *dnp;
2153 struct nfsvattr dnfsva;
2154 int error, dattrflag;
2155
2156 if (dvp == vp)
2157 return (EINVAL);
2158 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2159 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2160 dnp = VTONFS(dvp);
2161 mtx_lock(&dnp->n_mtx);
2162 dnp->n_flag |= NMODIFIED;
2163 if (dattrflag != 0) {
2164 mtx_unlock(&dnp->n_mtx);
2165 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2166 } else {
2167 dnp->n_attrstamp = 0;
2168 mtx_unlock(&dnp->n_mtx);
2169 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2170 }
2171
2172 cache_purge(dvp);
2173 cache_purge(vp);
2174 if (error && NFS_ISV4(dvp))
2175 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2176 (gid_t)0);
2177 /*
2178 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2179 */
2180 if (error == ENOENT)
2181 error = 0;
2182 return (error);
2183 }
2184
2185 /*
2186 * nfs readdir call
2187 */
2188 static int
2189 nfs_readdir(struct vop_readdir_args *ap)
2190 {
2191 struct vnode *vp = ap->a_vp;
2192 struct nfsnode *np = VTONFS(vp);
2193 struct uio *uio = ap->a_uio;
2194 ssize_t tresid, left;
2195 int error = 0;
2196 struct vattr vattr;
2197
2198 if (ap->a_eofflag != NULL)
2199 *ap->a_eofflag = 0;
2200 if (vp->v_type != VDIR)
2201 return(EPERM);
2202
2203 /*
2204 * First, check for hit on the EOF offset cache
2205 */
2206 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2207 (np->n_flag & NMODIFIED) == 0) {
2208 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2209 mtx_lock(&np->n_mtx);
2210 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2211 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2212 mtx_unlock(&np->n_mtx);
2213 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2214 if (ap->a_eofflag != NULL)
2215 *ap->a_eofflag = 1;
2216 return (0);
2217 } else
2218 mtx_unlock(&np->n_mtx);
2219 }
2220 }
2221
2222 /*
2223 * NFS always guarantees that directory entries don't straddle
2224 * DIRBLKSIZ boundaries. As such, we need to limit the size
2225 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2226 * directory entry.
2227 */
2228 left = uio->uio_resid % DIRBLKSIZ;
2229 if (left == uio->uio_resid)
2230 return (EINVAL);
2231 uio->uio_resid -= left;
2232
2233 /*
2234 * Call ncl_bioread() to do the real work.
2235 */
2236 tresid = uio->uio_resid;
2237 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2238
2239 if (!error && uio->uio_resid == tresid) {
2240 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2241 if (ap->a_eofflag != NULL)
2242 *ap->a_eofflag = 1;
2243 }
2244
2245 /* Add the partial DIRBLKSIZ (left) back in. */
2246 uio->uio_resid += left;
2247 return (error);
2248 }
2249
2250 /*
2251 * Readdir rpc call.
2252 * Called from below the buffer cache by ncl_doio().
2253 */
2254 int
2255 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2256 struct thread *td)
2257 {
2258 struct nfsvattr nfsva;
2259 nfsuint64 *cookiep, cookie;
2260 struct nfsnode *dnp = VTONFS(vp);
2261 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2262 int error = 0, eof, attrflag;
2263
2264 KASSERT(uiop->uio_iovcnt == 1 &&
2265 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2266 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2267 ("nfs readdirrpc bad uio"));
2268
2269 /*
2270 * If there is no cookie, assume directory was stale.
2271 */
2272 ncl_dircookie_lock(dnp);
2273 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2274 if (cookiep) {
2275 cookie = *cookiep;
2276 ncl_dircookie_unlock(dnp);
2277 } else {
2278 ncl_dircookie_unlock(dnp);
2279 return (NFSERR_BAD_COOKIE);
2280 }
2281
2282 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2283 (void)ncl_fsinfo(nmp, vp, cred, td);
2284
2285 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2286 &attrflag, &eof, NULL);
2287 if (attrflag)
2288 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2289
2290 if (!error) {
2291 /*
2292 * We are now either at the end of the directory or have filled
2293 * the block.
2294 */
2295 if (eof)
2296 dnp->n_direofoffset = uiop->uio_offset;
2297 else {
2298 if (uiop->uio_resid > 0)
2299 ncl_printf("EEK! readdirrpc resid > 0\n");
2300 ncl_dircookie_lock(dnp);
2301 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2302 *cookiep = cookie;
2303 ncl_dircookie_unlock(dnp);
2304 }
2305 } else if (NFS_ISV4(vp)) {
2306 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2307 }
2308 return (error);
2309 }
2310
2311 /*
2312 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2313 */
2314 int
2315 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2316 struct thread *td)
2317 {
2318 struct nfsvattr nfsva;
2319 nfsuint64 *cookiep, cookie;
2320 struct nfsnode *dnp = VTONFS(vp);
2321 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2322 int error = 0, attrflag, eof;
2323
2324 KASSERT(uiop->uio_iovcnt == 1 &&
2325 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2326 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2327 ("nfs readdirplusrpc bad uio"));
2328
2329 /*
2330 * If there is no cookie, assume directory was stale.
2331 */
2332 ncl_dircookie_lock(dnp);
2333 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2334 if (cookiep) {
2335 cookie = *cookiep;
2336 ncl_dircookie_unlock(dnp);
2337 } else {
2338 ncl_dircookie_unlock(dnp);
2339 return (NFSERR_BAD_COOKIE);
2340 }
2341
2342 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2343 (void)ncl_fsinfo(nmp, vp, cred, td);
2344 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2345 &attrflag, &eof, NULL);
2346 if (attrflag)
2347 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2348
2349 if (!error) {
2350 /*
2351 * We are now either at end of the directory or have filled the
2352 * the block.
2353 */
2354 if (eof)
2355 dnp->n_direofoffset = uiop->uio_offset;
2356 else {
2357 if (uiop->uio_resid > 0)
2358 ncl_printf("EEK! readdirplusrpc resid > 0\n");
2359 ncl_dircookie_lock(dnp);
2360 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2361 *cookiep = cookie;
2362 ncl_dircookie_unlock(dnp);
2363 }
2364 } else if (NFS_ISV4(vp)) {
2365 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2366 }
2367 return (error);
2368 }
2369
2370 /*
2371 * Silly rename. To make the NFS filesystem that is stateless look a little
2372 * more like the "ufs" a remove of an active vnode is translated to a rename
2373 * to a funny looking filename that is removed by nfs_inactive on the
2374 * nfsnode. There is the potential for another process on a different client
2375 * to create the same funny name between the nfs_lookitup() fails and the
2376 * nfs_rename() completes, but...
2377 */
2378 static int
2379 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2380 {
2381 struct sillyrename *sp;
2382 struct nfsnode *np;
2383 int error;
2384 short pid;
2385 unsigned int lticks;
2386
2387 cache_purge(dvp);
2388 np = VTONFS(vp);
2389 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2390 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2391 M_NEWNFSREQ, M_WAITOK);
2392 sp->s_cred = crhold(cnp->cn_cred);
2393 sp->s_dvp = dvp;
2394 VREF(dvp);
2395
2396 /*
2397 * Fudge together a funny name.
2398 * Changing the format of the funny name to accomodate more
2399 * sillynames per directory.
2400 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2401 * CPU ticks since boot.
2402 */
2403 pid = cnp->cn_thread->td_proc->p_pid;
2404 lticks = (unsigned int)ticks;
2405 for ( ; ; ) {
2406 sp->s_namlen = sprintf(sp->s_name,
2407 ".nfs.%08x.%04x4.4", lticks,
2408 pid);
2409 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2410 cnp->cn_thread, NULL))
2411 break;
2412 lticks++;
2413 }
2414 error = nfs_renameit(dvp, vp, cnp, sp);
2415 if (error)
2416 goto bad;
2417 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2418 cnp->cn_thread, &np);
2419 np->n_sillyrename = sp;
2420 return (0);
2421 bad:
2422 vrele(sp->s_dvp);
2423 crfree(sp->s_cred);
2424 free((caddr_t)sp, M_NEWNFSREQ);
2425 return (error);
2426 }
2427
2428 /*
2429 * Look up a file name and optionally either update the file handle or
2430 * allocate an nfsnode, depending on the value of npp.
2431 * npp == NULL --> just do the lookup
2432 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2433 * handled too
2434 * *npp != NULL --> update the file handle in the vnode
2435 */
2436 static int
2437 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2438 struct thread *td, struct nfsnode **npp)
2439 {
2440 struct vnode *newvp = NULL, *vp;
2441 struct nfsnode *np, *dnp = VTONFS(dvp);
2442 struct nfsfh *nfhp, *onfhp;
2443 struct nfsvattr nfsva, dnfsva;
2444 struct componentname cn;
2445 int error = 0, attrflag, dattrflag;
2446 u_int hash;
2447
2448 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2449 &nfhp, &attrflag, &dattrflag, NULL);
2450 if (dattrflag)
2451 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2452 if (npp && !error) {
2453 if (*npp != NULL) {
2454 np = *npp;
2455 vp = NFSTOV(np);
2456 /*
2457 * For NFSv4, check to see if it is the same name and
2458 * replace the name, if it is different.
2459 */
2460 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2461 (np->n_v4->n4_namelen != len ||
2462 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2463 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2464 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2465 dnp->n_fhp->nfh_len))) {
2466 #ifdef notdef
2467 { char nnn[100]; int nnnl;
2468 nnnl = (len < 100) ? len : 99;
2469 bcopy(name, nnn, nnnl);
2470 nnn[nnnl] = '\0';
2471 printf("replace=%s\n",nnn);
2472 }
2473 #endif
2474 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2475 MALLOC(np->n_v4, struct nfsv4node *,
2476 sizeof (struct nfsv4node) +
2477 dnp->n_fhp->nfh_len + len - 1,
2478 M_NFSV4NODE, M_WAITOK);
2479 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2480 np->n_v4->n4_namelen = len;
2481 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2482 dnp->n_fhp->nfh_len);
2483 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2484 }
2485 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2486 FNV1_32_INIT);
2487 onfhp = np->n_fhp;
2488 /*
2489 * Rehash node for new file handle.
2490 */
2491 vfs_hash_rehash(vp, hash);
2492 np->n_fhp = nfhp;
2493 if (onfhp != NULL)
2494 FREE((caddr_t)onfhp, M_NFSFH);
2495 newvp = NFSTOV(np);
2496 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2497 FREE((caddr_t)nfhp, M_NFSFH);
2498 VREF(dvp);
2499 newvp = dvp;
2500 } else {
2501 cn.cn_nameptr = name;
2502 cn.cn_namelen = len;
2503 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2504 &np, NULL, LK_EXCLUSIVE);
2505 if (error)
2506 return (error);
2507 newvp = NFSTOV(np);
2508 }
2509 if (!attrflag && *npp == NULL) {
2510 if (newvp == dvp)
2511 vrele(newvp);
2512 else
2513 vput(newvp);
2514 return (ENOENT);
2515 }
2516 if (attrflag)
2517 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2518 0, 1);
2519 }
2520 if (npp && *npp == NULL) {
2521 if (error) {
2522 if (newvp) {
2523 if (newvp == dvp)
2524 vrele(newvp);
2525 else
2526 vput(newvp);
2527 }
2528 } else
2529 *npp = np;
2530 }
2531 if (error && NFS_ISV4(dvp))
2532 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2533 return (error);
2534 }
2535
2536 /*
2537 * Nfs Version 3 and 4 commit rpc
2538 */
2539 int
2540 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2541 struct thread *td)
2542 {
2543 struct nfsvattr nfsva;
2544 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2545 int error, attrflag;
2546 u_char verf[NFSX_VERF];
2547
2548 mtx_lock(&nmp->nm_mtx);
2549 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2550 mtx_unlock(&nmp->nm_mtx);
2551 return (0);
2552 }
2553 mtx_unlock(&nmp->nm_mtx);
2554 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2555 &attrflag, NULL);
2556 if (!error) {
2557 mtx_lock(&nmp->nm_mtx);
2558 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2559 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2560 error = NFSERR_STALEWRITEVERF;
2561 }
2562 mtx_unlock(&nmp->nm_mtx);
2563 if (!error && attrflag)
2564 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2565 0, 1);
2566 } else if (NFS_ISV4(vp)) {
2567 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2568 }
2569 return (error);
2570 }
2571
2572 /*
2573 * Strategy routine.
2574 * For async requests when nfsiod(s) are running, queue the request by
2575 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2576 * request.
2577 */
2578 static int
2579 nfs_strategy(struct vop_strategy_args *ap)
2580 {
2581 struct buf *bp = ap->a_bp;
2582 struct ucred *cr;
2583
2584 KASSERT(!(bp->b_flags & B_DONE),
2585 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2586 BUF_ASSERT_HELD(bp);
2587
2588 if (bp->b_iocmd == BIO_READ)
2589 cr = bp->b_rcred;
2590 else
2591 cr = bp->b_wcred;
2592
2593 /*
2594 * If the op is asynchronous and an i/o daemon is waiting
2595 * queue the request, wake it up and wait for completion
2596 * otherwise just do it ourselves.
2597 */
2598 if ((bp->b_flags & B_ASYNC) == 0 ||
2599 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2600 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2601 return (0);
2602 }
2603
2604 /*
2605 * fsync vnode op. Just call ncl_flush() with commit == 1.
2606 */
2607 /* ARGSUSED */
2608 static int
2609 nfs_fsync(struct vop_fsync_args *ap)
2610 {
2611
2612 if (ap->a_vp->v_type != VREG) {
2613 /*
2614 * For NFS, metadata is changed synchronously on the server,
2615 * so there is nothing to flush. Also, ncl_flush() clears
2616 * the NMODIFIED flag and that shouldn't be done here for
2617 * directories.
2618 */
2619 return (0);
2620 }
2621 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2622 }
2623
2624 /*
2625 * Flush all the blocks associated with a vnode.
2626 * Walk through the buffer pool and push any dirty pages
2627 * associated with the vnode.
2628 * If the called_from_renewthread argument is TRUE, it has been called
2629 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2630 * waiting for a buffer write to complete.
2631 */
2632 int
2633 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2634 int commit, int called_from_renewthread)
2635 {
2636 struct nfsnode *np = VTONFS(vp);
2637 struct buf *bp;
2638 int i;
2639 struct buf *nbp;
2640 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2641 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2642 int passone = 1, trycnt = 0;
2643 u_quad_t off, endoff, toff;
2644 struct ucred* wcred = NULL;
2645 struct buf **bvec = NULL;
2646 struct bufobj *bo;
2647 #ifndef NFS_COMMITBVECSIZ
2648 #define NFS_COMMITBVECSIZ 20
2649 #endif
2650 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2651 int bvecsize = 0, bveccount;
2652
2653 if (called_from_renewthread != 0)
2654 slptimeo = hz;
2655 if (nmp->nm_flag & NFSMNT_INT)
2656 slpflag = NFS_PCATCH;
2657 if (!commit)
2658 passone = 0;
2659 bo = &vp->v_bufobj;
2660 /*
2661 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2662 * server, but has not been committed to stable storage on the server
2663 * yet. On the first pass, the byte range is worked out and the commit
2664 * rpc is done. On the second pass, ncl_writebp() is called to do the
2665 * job.
2666 */
2667 again:
2668 off = (u_quad_t)-1;
2669 endoff = 0;
2670 bvecpos = 0;
2671 if (NFS_ISV34(vp) && commit) {
2672 if (bvec != NULL && bvec != bvec_on_stack)
2673 free(bvec, M_TEMP);
2674 /*
2675 * Count up how many buffers waiting for a commit.
2676 */
2677 bveccount = 0;
2678 BO_LOCK(bo);
2679 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2680 if (!BUF_ISLOCKED(bp) &&
2681 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2682 == (B_DELWRI | B_NEEDCOMMIT))
2683 bveccount++;
2684 }
2685 /*
2686 * Allocate space to remember the list of bufs to commit. It is
2687 * important to use M_NOWAIT here to avoid a race with nfs_write.
2688 * If we can't get memory (for whatever reason), we will end up
2689 * committing the buffers one-by-one in the loop below.
2690 */
2691 if (bveccount > NFS_COMMITBVECSIZ) {
2692 /*
2693 * Release the vnode interlock to avoid a lock
2694 * order reversal.
2695 */
2696 BO_UNLOCK(bo);
2697 bvec = (struct buf **)
2698 malloc(bveccount * sizeof(struct buf *),
2699 M_TEMP, M_NOWAIT);
2700 BO_LOCK(bo);
2701 if (bvec == NULL) {
2702 bvec = bvec_on_stack;
2703 bvecsize = NFS_COMMITBVECSIZ;
2704 } else
2705 bvecsize = bveccount;
2706 } else {
2707 bvec = bvec_on_stack;
2708 bvecsize = NFS_COMMITBVECSIZ;
2709 }
2710 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2711 if (bvecpos >= bvecsize)
2712 break;
2713 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2714 nbp = TAILQ_NEXT(bp, b_bobufs);
2715 continue;
2716 }
2717 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2718 (B_DELWRI | B_NEEDCOMMIT)) {
2719 BUF_UNLOCK(bp);
2720 nbp = TAILQ_NEXT(bp, b_bobufs);
2721 continue;
2722 }
2723 BO_UNLOCK(bo);
2724 bremfree(bp);
2725 /*
2726 * Work out if all buffers are using the same cred
2727 * so we can deal with them all with one commit.
2728 *
2729 * NOTE: we are not clearing B_DONE here, so we have
2730 * to do it later on in this routine if we intend to
2731 * initiate I/O on the bp.
2732 *
2733 * Note: to avoid loopback deadlocks, we do not
2734 * assign b_runningbufspace.
2735 */
2736 if (wcred == NULL)
2737 wcred = bp->b_wcred;
2738 else if (wcred != bp->b_wcred)
2739 wcred = NOCRED;
2740 vfs_busy_pages(bp, 1);
2741
2742 BO_LOCK(bo);
2743 /*
2744 * bp is protected by being locked, but nbp is not
2745 * and vfs_busy_pages() may sleep. We have to
2746 * recalculate nbp.
2747 */
2748 nbp = TAILQ_NEXT(bp, b_bobufs);
2749
2750 /*
2751 * A list of these buffers is kept so that the
2752 * second loop knows which buffers have actually
2753 * been committed. This is necessary, since there
2754 * may be a race between the commit rpc and new
2755 * uncommitted writes on the file.
2756 */
2757 bvec[bvecpos++] = bp;
2758 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2759 bp->b_dirtyoff;
2760 if (toff < off)
2761 off = toff;
2762 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2763 if (toff > endoff)
2764 endoff = toff;
2765 }
2766 BO_UNLOCK(bo);
2767 }
2768 if (bvecpos > 0) {
2769 /*
2770 * Commit data on the server, as required.
2771 * If all bufs are using the same wcred, then use that with
2772 * one call for all of them, otherwise commit each one
2773 * separately.
2774 */
2775 if (wcred != NOCRED)
2776 retv = ncl_commit(vp, off, (int)(endoff - off),
2777 wcred, td);
2778 else {
2779 retv = 0;
2780 for (i = 0; i < bvecpos; i++) {
2781 off_t off, size;
2782 bp = bvec[i];
2783 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2784 bp->b_dirtyoff;
2785 size = (u_quad_t)(bp->b_dirtyend
2786 - bp->b_dirtyoff);
2787 retv = ncl_commit(vp, off, (int)size,
2788 bp->b_wcred, td);
2789 if (retv) break;
2790 }
2791 }
2792
2793 if (retv == NFSERR_STALEWRITEVERF)
2794 ncl_clearcommit(vp->v_mount);
2795
2796 /*
2797 * Now, either mark the blocks I/O done or mark the
2798 * blocks dirty, depending on whether the commit
2799 * succeeded.
2800 */
2801 for (i = 0; i < bvecpos; i++) {
2802 bp = bvec[i];
2803 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2804 if (retv) {
2805 /*
2806 * Error, leave B_DELWRI intact
2807 */
2808 vfs_unbusy_pages(bp);
2809 brelse(bp);
2810 } else {
2811 /*
2812 * Success, remove B_DELWRI ( bundirty() ).
2813 *
2814 * b_dirtyoff/b_dirtyend seem to be NFS
2815 * specific. We should probably move that
2816 * into bundirty(). XXX
2817 */
2818 bufobj_wref(bo);
2819 bp->b_flags |= B_ASYNC;
2820 bundirty(bp);
2821 bp->b_flags &= ~B_DONE;
2822 bp->b_ioflags &= ~BIO_ERROR;
2823 bp->b_dirtyoff = bp->b_dirtyend = 0;
2824 bufdone(bp);
2825 }
2826 }
2827 }
2828
2829 /*
2830 * Start/do any write(s) that are required.
2831 */
2832 loop:
2833 BO_LOCK(bo);
2834 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2835 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2836 if (waitfor != MNT_WAIT || passone)
2837 continue;
2838
2839 error = BUF_TIMELOCK(bp,
2840 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2841 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2842 if (error == 0) {
2843 BUF_UNLOCK(bp);
2844 goto loop;
2845 }
2846 if (error == ENOLCK) {
2847 error = 0;
2848 goto loop;
2849 }
2850 if (called_from_renewthread != 0) {
2851 /*
2852 * Return EIO so the flush will be retried
2853 * later.
2854 */
2855 error = EIO;
2856 goto done;
2857 }
2858 if (newnfs_sigintr(nmp, td)) {
2859 error = EINTR;
2860 goto done;
2861 }
2862 if (slpflag & PCATCH) {
2863 slpflag = 0;
2864 slptimeo = 2 * hz;
2865 }
2866 goto loop;
2867 }
2868 if ((bp->b_flags & B_DELWRI) == 0)
2869 panic("nfs_fsync: not dirty");
2870 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2871 BUF_UNLOCK(bp);
2872 continue;
2873 }
2874 BO_UNLOCK(bo);
2875 bremfree(bp);
2876 if (passone || !commit)
2877 bp->b_flags |= B_ASYNC;
2878 else
2879 bp->b_flags |= B_ASYNC;
2880 bwrite(bp);
2881 if (newnfs_sigintr(nmp, td)) {
2882 error = EINTR;
2883 goto done;
2884 }
2885 goto loop;
2886 }
2887 if (passone) {
2888 passone = 0;
2889 BO_UNLOCK(bo);
2890 goto again;
2891 }
2892 if (waitfor == MNT_WAIT) {
2893 while (bo->bo_numoutput) {
2894 error = bufobj_wwait(bo, slpflag, slptimeo);
2895 if (error) {
2896 BO_UNLOCK(bo);
2897 if (called_from_renewthread != 0) {
2898 /*
2899 * Return EIO so that the flush will be
2900 * retried later.
2901 */
2902 error = EIO;
2903 goto done;
2904 }
2905 error = newnfs_sigintr(nmp, td);
2906 if (error)
2907 goto done;
2908 if (slpflag & PCATCH) {
2909 slpflag = 0;
2910 slptimeo = 2 * hz;
2911 }
2912 BO_LOCK(bo);
2913 }
2914 }
2915 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2916 BO_UNLOCK(bo);
2917 goto loop;
2918 }
2919 /*
2920 * Wait for all the async IO requests to drain
2921 */
2922 BO_UNLOCK(bo);
2923 mtx_lock(&np->n_mtx);
2924 while (np->n_directio_asyncwr > 0) {
2925 np->n_flag |= NFSYNCWAIT;
2926 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2927 &np->n_mtx, slpflag | (PRIBIO + 1),
2928 "nfsfsync", 0);
2929 if (error) {
2930 if (newnfs_sigintr(nmp, td)) {
2931 mtx_unlock(&np->n_mtx);
2932 error = EINTR;
2933 goto done;
2934 }
2935 }
2936 }
2937 mtx_unlock(&np->n_mtx);
2938 } else
2939 BO_UNLOCK(bo);
2940 mtx_lock(&np->n_mtx);
2941 if (np->n_flag & NWRITEERR) {
2942 error = np->n_error;
2943 np->n_flag &= ~NWRITEERR;
2944 }
2945 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2946 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2947 np->n_flag &= ~NMODIFIED;
2948 mtx_unlock(&np->n_mtx);
2949 done:
2950 if (bvec != NULL && bvec != bvec_on_stack)
2951 free(bvec, M_TEMP);
2952 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2953 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2954 np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2955 /* try, try again... */
2956 passone = 1;
2957 wcred = NULL;
2958 bvec = NULL;
2959 bvecsize = 0;
2960 printf("try%d\n", trycnt);
2961 goto again;
2962 }
2963 return (error);
2964 }
2965
2966 /*
2967 * NFS advisory byte-level locks.
2968 */
2969 static int
2970 nfs_advlock(struct vop_advlock_args *ap)
2971 {
2972 struct vnode *vp = ap->a_vp;
2973 struct ucred *cred;
2974 struct nfsnode *np = VTONFS(ap->a_vp);
2975 struct proc *p = (struct proc *)ap->a_id;
2976 struct thread *td = curthread; /* XXX */
2977 struct vattr va;
2978 int ret, error = EOPNOTSUPP;
2979 u_quad_t size;
2980
2981 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
2982 if (vp->v_type != VREG)
2983 return (EINVAL);
2984 if ((ap->a_flags & F_POSIX) != 0)
2985 cred = p->p_ucred;
2986 else
2987 cred = td->td_ucred;
2988 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2989 if (vp->v_iflag & VI_DOOMED) {
2990 NFSVOPUNLOCK(vp, 0);
2991 return (EBADF);
2992 }
2993
2994 /*
2995 * If this is unlocking a write locked region, flush and
2996 * commit them before unlocking. This is required by
2997 * RFC3530 Sec. 9.3.2.
2998 */
2999 if (ap->a_op == F_UNLCK &&
3000 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3001 ap->a_flags))
3002 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
3003
3004 /*
3005 * Loop around doing the lock op, while a blocking lock
3006 * must wait for the lock op to succeed.
3007 */
3008 do {
3009 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3010 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3011 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3012 ap->a_op == F_SETLK) {
3013 NFSVOPUNLOCK(vp, 0);
3014 error = nfs_catnap(PZERO | PCATCH, ret,
3015 "ncladvl");
3016 if (error)
3017 return (EINTR);
3018 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3019 if (vp->v_iflag & VI_DOOMED) {
3020 NFSVOPUNLOCK(vp, 0);
3021 return (EBADF);
3022 }
3023 }
3024 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3025 ap->a_op == F_SETLK);
3026 if (ret == NFSERR_DENIED) {
3027 NFSVOPUNLOCK(vp, 0);
3028 return (EAGAIN);
3029 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3030 NFSVOPUNLOCK(vp, 0);
3031 return (ret);
3032 } else if (ret != 0) {
3033 NFSVOPUNLOCK(vp, 0);
3034 return (EACCES);
3035 }
3036
3037 /*
3038 * Now, if we just got a lock, invalidate data in the buffer
3039 * cache, as required, so that the coherency conforms with
3040 * RFC3530 Sec. 9.3.2.
3041 */
3042 if (ap->a_op == F_SETLK) {
3043 if ((np->n_flag & NMODIFIED) == 0) {
3044 np->n_attrstamp = 0;
3045 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3046 ret = VOP_GETATTR(vp, &va, cred);
3047 }
3048 if ((np->n_flag & NMODIFIED) || ret ||
3049 np->n_change != va.va_filerev) {
3050 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3051 np->n_attrstamp = 0;
3052 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3053 ret = VOP_GETATTR(vp, &va, cred);
3054 if (!ret) {
3055 np->n_mtime = va.va_mtime;
3056 np->n_change = va.va_filerev;
3057 }
3058 }
3059 /* Mark that a file lock has been acquired. */
3060 mtx_lock(&np->n_mtx);
3061 np->n_flag |= NHASBEENLOCKED;
3062 mtx_unlock(&np->n_mtx);
3063 }
3064 NFSVOPUNLOCK(vp, 0);
3065 return (0);
3066 } else if (!NFS_ISV4(vp)) {
3067 error = NFSVOPLOCK(vp, LK_SHARED);
3068 if (error)
3069 return (error);
3070 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3071 size = VTONFS(vp)->n_size;
3072 NFSVOPUNLOCK(vp, 0);
3073 error = lf_advlock(ap, &(vp->v_lockf), size);
3074 } else {
3075 if (nfs_advlock_p != NULL)
3076 error = nfs_advlock_p(ap);
3077 else {
3078 NFSVOPUNLOCK(vp, 0);
3079 error = ENOLCK;
3080 }
3081 }
3082 if (error == 0 && ap->a_op == F_SETLK) {
3083 /* Mark that a file lock has been acquired. */
3084 mtx_lock(&np->n_mtx);
3085 np->n_flag |= NHASBEENLOCKED;
3086 mtx_unlock(&np->n_mtx);
3087 }
3088 }
3089 return (error);
3090 }
3091
3092 /*
3093 * NFS advisory byte-level locks.
3094 */
3095 static int
3096 nfs_advlockasync(struct vop_advlockasync_args *ap)
3097 {
3098 struct vnode *vp = ap->a_vp;
3099 u_quad_t size;
3100 int error;
3101
3102 if (NFS_ISV4(vp))
3103 return (EOPNOTSUPP);
3104 error = NFSVOPLOCK(vp, LK_SHARED);
3105 if (error)
3106 return (error);
3107 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3108 size = VTONFS(vp)->n_size;
3109 NFSVOPUNLOCK(vp, 0);
3110 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3111 } else {
3112 NFSVOPUNLOCK(vp, 0);
3113 error = EOPNOTSUPP;
3114 }
3115 return (error);
3116 }
3117
3118 /*
3119 * Print out the contents of an nfsnode.
3120 */
3121 static int
3122 nfs_print(struct vop_print_args *ap)
3123 {
3124 struct vnode *vp = ap->a_vp;
3125 struct nfsnode *np = VTONFS(vp);
3126
3127 ncl_printf("\tfileid %ld fsid 0x%x",
3128 np->n_vattr.na_fileid, np->n_vattr.na_fsid);
3129 if (vp->v_type == VFIFO)
3130 fifo_printinfo(vp);
3131 printf("\n");
3132 return (0);
3133 }
3134
3135 /*
3136 * This is the "real" nfs::bwrite(struct buf*).
3137 * We set B_CACHE if this is a VMIO buffer.
3138 */
3139 int
3140 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3141 {
3142 int s;
3143 int oldflags = bp->b_flags;
3144 #if 0
3145 int retv = 1;
3146 off_t off;
3147 #endif
3148
3149 BUF_ASSERT_HELD(bp);
3150
3151 if (bp->b_flags & B_INVAL) {
3152 brelse(bp);
3153 return(0);
3154 }
3155
3156 bp->b_flags |= B_CACHE;
3157
3158 /*
3159 * Undirty the bp. We will redirty it later if the I/O fails.
3160 */
3161
3162 s = splbio();
3163 bundirty(bp);
3164 bp->b_flags &= ~B_DONE;
3165 bp->b_ioflags &= ~BIO_ERROR;
3166 bp->b_iocmd = BIO_WRITE;
3167
3168 bufobj_wref(bp->b_bufobj);
3169 curthread->td_ru.ru_oublock++;
3170 splx(s);
3171
3172 /*
3173 * Note: to avoid loopback deadlocks, we do not
3174 * assign b_runningbufspace.
3175 */
3176 vfs_busy_pages(bp, 1);
3177
3178 BUF_KERNPROC(bp);
3179 bp->b_iooffset = dbtob(bp->b_blkno);
3180 bstrategy(bp);
3181
3182 if( (oldflags & B_ASYNC) == 0) {
3183 int rtval = bufwait(bp);
3184
3185 if (oldflags & B_DELWRI) {
3186 s = splbio();
3187 reassignbuf(bp);
3188 splx(s);
3189 }
3190 brelse(bp);
3191 return (rtval);
3192 }
3193
3194 return (0);
3195 }
3196
3197 /*
3198 * nfs special file access vnode op.
3199 * Essentially just get vattr and then imitate iaccess() since the device is
3200 * local to the client.
3201 */
3202 static int
3203 nfsspec_access(struct vop_access_args *ap)
3204 {
3205 struct vattr *vap;
3206 struct ucred *cred = ap->a_cred;
3207 struct vnode *vp = ap->a_vp;
3208 accmode_t accmode = ap->a_accmode;
3209 struct vattr vattr;
3210 int error;
3211
3212 /*
3213 * Disallow write attempts on filesystems mounted read-only;
3214 * unless the file is a socket, fifo, or a block or character
3215 * device resident on the filesystem.
3216 */
3217 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3218 switch (vp->v_type) {
3219 case VREG:
3220 case VDIR:
3221 case VLNK:
3222 return (EROFS);
3223 default:
3224 break;
3225 }
3226 }
3227 vap = &vattr;
3228 error = VOP_GETATTR(vp, vap, cred);
3229 if (error)
3230 goto out;
3231 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3232 accmode, cred, NULL);
3233 out:
3234 return error;
3235 }
3236
3237 /*
3238 * Read wrapper for fifos.
3239 */
3240 static int
3241 nfsfifo_read(struct vop_read_args *ap)
3242 {
3243 struct nfsnode *np = VTONFS(ap->a_vp);
3244 int error;
3245
3246 /*
3247 * Set access flag.
3248 */
3249 mtx_lock(&np->n_mtx);
3250 np->n_flag |= NACC;
3251 vfs_timestamp(&np->n_atim);
3252 mtx_unlock(&np->n_mtx);
3253 error = fifo_specops.vop_read(ap);
3254 return error;
3255 }
3256
3257 /*
3258 * Write wrapper for fifos.
3259 */
3260 static int
3261 nfsfifo_write(struct vop_write_args *ap)
3262 {
3263 struct nfsnode *np = VTONFS(ap->a_vp);
3264
3265 /*
3266 * Set update flag.
3267 */
3268 mtx_lock(&np->n_mtx);
3269 np->n_flag |= NUPD;
3270 vfs_timestamp(&np->n_mtim);
3271 mtx_unlock(&np->n_mtx);
3272 return(fifo_specops.vop_write(ap));
3273 }
3274
3275 /*
3276 * Close wrapper for fifos.
3277 *
3278 * Update the times on the nfsnode then do fifo close.
3279 */
3280 static int
3281 nfsfifo_close(struct vop_close_args *ap)
3282 {
3283 struct vnode *vp = ap->a_vp;
3284 struct nfsnode *np = VTONFS(vp);
3285 struct vattr vattr;
3286 struct timespec ts;
3287
3288 mtx_lock(&np->n_mtx);
3289 if (np->n_flag & (NACC | NUPD)) {
3290 vfs_timestamp(&ts);
3291 if (np->n_flag & NACC)
3292 np->n_atim = ts;
3293 if (np->n_flag & NUPD)
3294 np->n_mtim = ts;
3295 np->n_flag |= NCHG;
3296 if (vrefcnt(vp) == 1 &&
3297 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3298 VATTR_NULL(&vattr);
3299 if (np->n_flag & NACC)
3300 vattr.va_atime = np->n_atim;
3301 if (np->n_flag & NUPD)
3302 vattr.va_mtime = np->n_mtim;
3303 mtx_unlock(&np->n_mtx);
3304 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3305 goto out;
3306 }
3307 }
3308 mtx_unlock(&np->n_mtx);
3309 out:
3310 return (fifo_specops.vop_close(ap));
3311 }
3312
3313 /*
3314 * Just call ncl_writebp() with the force argument set to 1.
3315 *
3316 * NOTE: B_DONE may or may not be set in a_bp on call.
3317 */
3318 static int
3319 nfs_bwrite(struct buf *bp)
3320 {
3321
3322 return (ncl_writebp(bp, 1, curthread));
3323 }
3324
3325 struct buf_ops buf_ops_newnfs = {
3326 .bop_name = "buf_ops_nfs",
3327 .bop_write = nfs_bwrite,
3328 .bop_strategy = bufstrategy,
3329 .bop_sync = bufsync,
3330 .bop_bdflush = bufbdflush,
3331 };
3332
3333 /*
3334 * Cloned from vop_stdlock(), and then the ugly hack added.
3335 */
3336 static int
3337 nfs_lock1(struct vop_lock1_args *ap)
3338 {
3339 struct vnode *vp = ap->a_vp;
3340 int error = 0;
3341
3342 /*
3343 * Since vfs_hash_get() calls vget() and it will no longer work
3344 * for FreeBSD8 with flags == 0, I can only think of this horrible
3345 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3346 * and then handle it here. All I want for this case is a v_usecount
3347 * on the vnode to use for recovery, while another thread might
3348 * hold a lock on the vnode. I have the other threads blocked, so
3349 * there isn't any race problem.
3350 */
3351 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3352 if ((ap->a_flags & LK_INTERLOCK) == 0)
3353 panic("ncllock1");
3354 if ((vp->v_iflag & VI_DOOMED))
3355 error = ENOENT;
3356 VI_UNLOCK(vp);
3357 return (error);
3358 }
3359 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3360 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3361 ap->a_line));
3362 }
3363
3364 static int
3365 nfs_getacl(struct vop_getacl_args *ap)
3366 {
3367 int error;
3368
3369 if (ap->a_type != ACL_TYPE_NFS4)
3370 return (EOPNOTSUPP);
3371 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3372 NULL);
3373 if (error > NFSERR_STALE) {
3374 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3375 error = EPERM;
3376 }
3377 return (error);
3378 }
3379
3380 static int
3381 nfs_setacl(struct vop_setacl_args *ap)
3382 {
3383 int error;
3384
3385 if (ap->a_type != ACL_TYPE_NFS4)
3386 return (EOPNOTSUPP);
3387 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3388 NULL);
3389 if (error > NFSERR_STALE) {
3390 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3391 error = EPERM;
3392 }
3393 return (error);
3394 }
3395
3396 /*
3397 * Return POSIX pathconf information applicable to nfs filesystems.
3398 */
3399 static int
3400 nfs_pathconf(struct vop_pathconf_args *ap)
3401 {
3402 struct nfsv3_pathconf pc;
3403 struct nfsvattr nfsva;
3404 struct vnode *vp = ap->a_vp;
3405 struct thread *td = curthread;
3406 int attrflag, error;
3407
3408 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
3409 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3410 ap->a_name == _PC_NO_TRUNC)) ||
3411 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
3412 /*
3413 * Since only the above 4 a_names are returned by the NFSv3
3414 * Pathconf RPC, there is no point in doing it for others.
3415 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
3416 * be used for _PC_NFS4_ACL as well.
3417 */
3418 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3419 &attrflag, NULL);
3420 if (attrflag != 0)
3421 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3422 1);
3423 if (error != 0)
3424 return (error);
3425 } else {
3426 /*
3427 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3428 * just fake them.
3429 */
3430 pc.pc_linkmax = LINK_MAX;
3431 pc.pc_namemax = NFS_MAXNAMLEN;
3432 pc.pc_notrunc = 1;
3433 pc.pc_chownrestricted = 1;
3434 pc.pc_caseinsensitive = 0;
3435 pc.pc_casepreserving = 1;
3436 error = 0;
3437 }
3438 switch (ap->a_name) {
3439 case _PC_LINK_MAX:
3440 *ap->a_retval = pc.pc_linkmax;
3441 break;
3442 case _PC_NAME_MAX:
3443 *ap->a_retval = pc.pc_namemax;
3444 break;
3445 case _PC_PATH_MAX:
3446 *ap->a_retval = PATH_MAX;
3447 break;
3448 case _PC_PIPE_BUF:
3449 *ap->a_retval = PIPE_BUF;
3450 break;
3451 case _PC_CHOWN_RESTRICTED:
3452 *ap->a_retval = pc.pc_chownrestricted;
3453 break;
3454 case _PC_NO_TRUNC:
3455 *ap->a_retval = pc.pc_notrunc;
3456 break;
3457 case _PC_ACL_EXTENDED:
3458 *ap->a_retval = 0;
3459 break;
3460 case _PC_ACL_NFS4:
3461 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3462 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3463 *ap->a_retval = 1;
3464 else
3465 *ap->a_retval = 0;
3466 break;
3467 case _PC_ACL_PATH_MAX:
3468 if (NFS_ISV4(vp))
3469 *ap->a_retval = ACL_MAX_ENTRIES;
3470 else
3471 *ap->a_retval = 3;
3472 break;
3473 case _PC_MAC_PRESENT:
3474 *ap->a_retval = 0;
3475 break;
3476 case _PC_ASYNC_IO:
3477 /* _PC_ASYNC_IO should have been handled by upper layers. */
3478 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3479 error = EINVAL;
3480 break;
3481 case _PC_PRIO_IO:
3482 *ap->a_retval = 0;
3483 break;
3484 case _PC_SYNC_IO:
3485 *ap->a_retval = 0;
3486 break;
3487 case _PC_ALLOC_SIZE_MIN:
3488 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3489 break;
3490 case _PC_FILESIZEBITS:
3491 if (NFS_ISV34(vp))
3492 *ap->a_retval = 64;
3493 else
3494 *ap->a_retval = 32;
3495 break;
3496 case _PC_REC_INCR_XFER_SIZE:
3497 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3498 break;
3499 case _PC_REC_MAX_XFER_SIZE:
3500 *ap->a_retval = -1; /* means ``unlimited'' */
3501 break;
3502 case _PC_REC_MIN_XFER_SIZE:
3503 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3504 break;
3505 case _PC_REC_XFER_ALIGN:
3506 *ap->a_retval = PAGE_SIZE;
3507 break;
3508 case _PC_SYMLINK_MAX:
3509 *ap->a_retval = NFS_MAXPATHLEN;
3510 break;
3511
3512 default:
3513 error = EINVAL;
3514 break;
3515 }
3516 return (error);
3517 }
3518
Cache object: 08ee2aafff1439a335d1d2f9763b70f5
|