1 /*-
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD: releng/8.3/sys/fs/nfsclient/nfs_clvnops.c 231948 2012-02-21 00:32:24Z kib $");
37
38 /*
39 * vnode op calls for Sun NFS version 2, 3 and 4
40 */
41
42 #include "opt_inet.h"
43
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/resourcevar.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/jail.h>
53 #include <sys/malloc.h>
54 #include <sys/mbuf.h>
55 #include <sys/namei.h>
56 #include <sys/socket.h>
57 #include <sys/vnode.h>
58 #include <sys/dirent.h>
59 #include <sys/fcntl.h>
60 #include <sys/lockf.h>
61 #include <sys/stat.h>
62 #include <sys/sysctl.h>
63 #include <sys/signalvar.h>
64
65 #include <vm/vm.h>
66 #include <vm/vm_object.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
75 #include <net/if.h>
76 #include <netinet/in.h>
77 #include <netinet/in_var.h>
78
79 #include <nfs/nfs_lock.h>
80
81 /* Defs */
82 #define TRUE 1
83 #define FALSE 0
84
85 extern struct nfsstats newnfsstats;
86 extern int nfsrv_useacl;
87 MALLOC_DECLARE(M_NEWNFSREQ);
88
89 /*
90 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
91 * calls are not in getblk() and brelse() so that they would not be necessary
92 * here.
93 */
94 #ifndef B_VMIO
95 #define vfs_busy_pages(bp, f)
96 #endif
97
98 static vop_read_t nfsfifo_read;
99 static vop_write_t nfsfifo_write;
100 static vop_close_t nfsfifo_close;
101 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
102 struct thread *);
103 static vop_lookup_t nfs_lookup;
104 static vop_create_t nfs_create;
105 static vop_mknod_t nfs_mknod;
106 static vop_open_t nfs_open;
107 static vop_pathconf_t nfs_pathconf;
108 static vop_close_t nfs_close;
109 static vop_access_t nfs_access;
110 static vop_getattr_t nfs_getattr;
111 static vop_setattr_t nfs_setattr;
112 static vop_read_t nfs_read;
113 static vop_fsync_t nfs_fsync;
114 static vop_remove_t nfs_remove;
115 static vop_link_t nfs_link;
116 static vop_rename_t nfs_rename;
117 static vop_mkdir_t nfs_mkdir;
118 static vop_rmdir_t nfs_rmdir;
119 static vop_symlink_t nfs_symlink;
120 static vop_readdir_t nfs_readdir;
121 static vop_strategy_t nfs_strategy;
122 static vop_lock1_t nfs_lock1;
123 static int nfs_lookitup(struct vnode *, char *, int,
124 struct ucred *, struct thread *, struct nfsnode **);
125 static int nfs_sillyrename(struct vnode *, struct vnode *,
126 struct componentname *);
127 static vop_access_t nfsspec_access;
128 static vop_readlink_t nfs_readlink;
129 static vop_print_t nfs_print;
130 static vop_advlock_t nfs_advlock;
131 static vop_advlockasync_t nfs_advlockasync;
132 static vop_getacl_t nfs_getacl;
133 static vop_setacl_t nfs_setacl;
134
135 /*
136 * Global vfs data structures for nfs
137 */
138 struct vop_vector newnfs_vnodeops = {
139 .vop_default = &default_vnodeops,
140 .vop_access = nfs_access,
141 .vop_advlock = nfs_advlock,
142 .vop_advlockasync = nfs_advlockasync,
143 .vop_close = nfs_close,
144 .vop_create = nfs_create,
145 .vop_fsync = nfs_fsync,
146 .vop_getattr = nfs_getattr,
147 .vop_getpages = ncl_getpages,
148 .vop_putpages = ncl_putpages,
149 .vop_inactive = ncl_inactive,
150 .vop_link = nfs_link,
151 .vop_lock1 = nfs_lock1,
152 .vop_lookup = nfs_lookup,
153 .vop_mkdir = nfs_mkdir,
154 .vop_mknod = nfs_mknod,
155 .vop_open = nfs_open,
156 .vop_pathconf = nfs_pathconf,
157 .vop_print = nfs_print,
158 .vop_read = nfs_read,
159 .vop_readdir = nfs_readdir,
160 .vop_readlink = nfs_readlink,
161 .vop_reclaim = ncl_reclaim,
162 .vop_remove = nfs_remove,
163 .vop_rename = nfs_rename,
164 .vop_rmdir = nfs_rmdir,
165 .vop_setattr = nfs_setattr,
166 .vop_strategy = nfs_strategy,
167 .vop_symlink = nfs_symlink,
168 .vop_write = ncl_write,
169 .vop_getacl = nfs_getacl,
170 .vop_setacl = nfs_setacl,
171 };
172
173 struct vop_vector newnfs_fifoops = {
174 .vop_default = &fifo_specops,
175 .vop_access = nfsspec_access,
176 .vop_close = nfsfifo_close,
177 .vop_fsync = nfs_fsync,
178 .vop_getattr = nfs_getattr,
179 .vop_inactive = ncl_inactive,
180 .vop_print = nfs_print,
181 .vop_read = nfsfifo_read,
182 .vop_reclaim = ncl_reclaim,
183 .vop_setattr = nfs_setattr,
184 .vop_write = nfsfifo_write,
185 };
186
187 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
188 struct componentname *cnp, struct vattr *vap);
189 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
190 int namelen, struct ucred *cred, struct thread *td);
191 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
192 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
193 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
194 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
195 struct componentname *scnp, struct sillyrename *sp);
196
197 /*
198 * Global variables
199 */
200 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
201
202 SYSCTL_DECL(_vfs_newnfs);
203
204 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
205 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
206 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
207
208 static int nfs_prime_access_cache = 0;
209 SYSCTL_INT(_vfs_newnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
210 &nfs_prime_access_cache, 0,
211 "Prime NFS ACCESS cache when fetching attributes");
212
213 static int newnfs_commit_on_close = 0;
214 SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
215 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
216
217 static int nfs_clean_pages_on_close = 1;
218 SYSCTL_INT(_vfs_newnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
219 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
220
221 int newnfs_directio_enable = 0;
222 SYSCTL_INT(_vfs_newnfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
223 &newnfs_directio_enable, 0, "Enable NFS directio");
224
225 /*
226 * This sysctl allows other processes to mmap a file that has been opened
227 * O_DIRECT by a process. In general, having processes mmap the file while
228 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
229 * this by default to prevent DoS attacks - to prevent a malicious user from
230 * opening up files O_DIRECT preventing other users from mmap'ing these
231 * files. "Protected" environments where stricter consistency guarantees are
232 * required can disable this knob. The process that opened the file O_DIRECT
233 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
234 * meaningful.
235 */
236 int newnfs_directio_allow_mmap = 1;
237 SYSCTL_INT(_vfs_newnfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
238 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
239
240 #if 0
241 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
242 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
243
244 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
245 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
246 #endif
247
248 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
249 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
250 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
251
252 /*
253 * SMP Locking Note :
254 * The list of locks after the description of the lock is the ordering
255 * of other locks acquired with the lock held.
256 * np->n_mtx : Protects the fields in the nfsnode.
257 VM Object Lock
258 VI_MTX (acquired indirectly)
259 * nmp->nm_mtx : Protects the fields in the nfsmount.
260 rep->r_mtx
261 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
262 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
263 nmp->nm_mtx
264 rep->r_mtx
265 * rep->r_mtx : Protects the fields in an nfsreq.
266 */
267
268 static int
269 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
270 struct ucred *cred, u_int32_t *retmode)
271 {
272 int error = 0, attrflag, i, lrupos;
273 u_int32_t rmode;
274 struct nfsnode *np = VTONFS(vp);
275 struct nfsvattr nfsva;
276
277 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
278 &rmode, NULL);
279 if (attrflag)
280 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
281 if (!error) {
282 lrupos = 0;
283 mtx_lock(&np->n_mtx);
284 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
285 if (np->n_accesscache[i].uid == cred->cr_uid) {
286 np->n_accesscache[i].mode = rmode;
287 np->n_accesscache[i].stamp = time_second;
288 break;
289 }
290 if (i > 0 && np->n_accesscache[i].stamp <
291 np->n_accesscache[lrupos].stamp)
292 lrupos = i;
293 }
294 if (i == NFS_ACCESSCACHESIZE) {
295 np->n_accesscache[lrupos].uid = cred->cr_uid;
296 np->n_accesscache[lrupos].mode = rmode;
297 np->n_accesscache[lrupos].stamp = time_second;
298 }
299 mtx_unlock(&np->n_mtx);
300 if (retmode != NULL)
301 *retmode = rmode;
302 } else if (NFS_ISV4(vp)) {
303 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
304 }
305 return (error);
306 }
307
308 /*
309 * nfs access vnode op.
310 * For nfs version 2, just return ok. File accesses may fail later.
311 * For nfs version 3, use the access rpc to check accessibility. If file modes
312 * are changed on the server, accesses might still fail later.
313 */
314 static int
315 nfs_access(struct vop_access_args *ap)
316 {
317 struct vnode *vp = ap->a_vp;
318 int error = 0, i, gotahit;
319 u_int32_t mode, wmode, rmode;
320 int v34 = NFS_ISV34(vp);
321 struct nfsnode *np = VTONFS(vp);
322
323 /*
324 * Disallow write attempts on filesystems mounted read-only;
325 * unless the file is a socket, fifo, or a block or character
326 * device resident on the filesystem.
327 */
328 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
329 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
330 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
331 switch (vp->v_type) {
332 case VREG:
333 case VDIR:
334 case VLNK:
335 return (EROFS);
336 default:
337 break;
338 }
339 }
340 /*
341 * For nfs v3 or v4, check to see if we have done this recently, and if
342 * so return our cached result instead of making an ACCESS call.
343 * If not, do an access rpc, otherwise you are stuck emulating
344 * ufs_access() locally using the vattr. This may not be correct,
345 * since the server may apply other access criteria such as
346 * client uid-->server uid mapping that we do not know about.
347 */
348 if (v34) {
349 if (ap->a_accmode & VREAD)
350 mode = NFSACCESS_READ;
351 else
352 mode = 0;
353 if (vp->v_type != VDIR) {
354 if (ap->a_accmode & VWRITE)
355 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
356 if (ap->a_accmode & VAPPEND)
357 mode |= NFSACCESS_EXTEND;
358 if (ap->a_accmode & VEXEC)
359 mode |= NFSACCESS_EXECUTE;
360 if (ap->a_accmode & VDELETE)
361 mode |= NFSACCESS_DELETE;
362 } else {
363 if (ap->a_accmode & VWRITE)
364 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
365 if (ap->a_accmode & VAPPEND)
366 mode |= NFSACCESS_EXTEND;
367 if (ap->a_accmode & VEXEC)
368 mode |= NFSACCESS_LOOKUP;
369 if (ap->a_accmode & VDELETE)
370 mode |= NFSACCESS_DELETE;
371 if (ap->a_accmode & VDELETE_CHILD)
372 mode |= NFSACCESS_MODIFY;
373 }
374 /* XXX safety belt, only make blanket request if caching */
375 if (nfsaccess_cache_timeout > 0) {
376 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
377 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
378 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
379 } else {
380 wmode = mode;
381 }
382
383 /*
384 * Does our cached result allow us to give a definite yes to
385 * this request?
386 */
387 gotahit = 0;
388 mtx_lock(&np->n_mtx);
389 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
390 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
391 if (time_second < (np->n_accesscache[i].stamp
392 + nfsaccess_cache_timeout) &&
393 (np->n_accesscache[i].mode & mode) == mode) {
394 NFSINCRGLOBAL(newnfsstats.accesscache_hits);
395 gotahit = 1;
396 }
397 break;
398 }
399 }
400 mtx_unlock(&np->n_mtx);
401 if (gotahit == 0) {
402 /*
403 * Either a no, or a don't know. Go to the wire.
404 */
405 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
406 error = nfs34_access_otw(vp, wmode, ap->a_td,
407 ap->a_cred, &rmode);
408 if (!error &&
409 (rmode & mode) != mode)
410 error = EACCES;
411 }
412 return (error);
413 } else {
414 if ((error = nfsspec_access(ap)) != 0) {
415 return (error);
416 }
417 /*
418 * Attempt to prevent a mapped root from accessing a file
419 * which it shouldn't. We try to read a byte from the file
420 * if the user is root and the file is not zero length.
421 * After calling nfsspec_access, we should have the correct
422 * file size cached.
423 */
424 mtx_lock(&np->n_mtx);
425 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
426 && VTONFS(vp)->n_size > 0) {
427 struct iovec aiov;
428 struct uio auio;
429 char buf[1];
430
431 mtx_unlock(&np->n_mtx);
432 aiov.iov_base = buf;
433 aiov.iov_len = 1;
434 auio.uio_iov = &aiov;
435 auio.uio_iovcnt = 1;
436 auio.uio_offset = 0;
437 auio.uio_resid = 1;
438 auio.uio_segflg = UIO_SYSSPACE;
439 auio.uio_rw = UIO_READ;
440 auio.uio_td = ap->a_td;
441
442 if (vp->v_type == VREG)
443 error = ncl_readrpc(vp, &auio, ap->a_cred);
444 else if (vp->v_type == VDIR) {
445 char* bp;
446 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
447 aiov.iov_base = bp;
448 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
449 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
450 ap->a_td);
451 free(bp, M_TEMP);
452 } else if (vp->v_type == VLNK)
453 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
454 else
455 error = EACCES;
456 } else
457 mtx_unlock(&np->n_mtx);
458 return (error);
459 }
460 }
461
462
463 /*
464 * nfs open vnode op
465 * Check to see if the type is ok
466 * and that deletion is not in progress.
467 * For paged in text files, you will need to flush the page cache
468 * if consistency is lost.
469 */
470 /* ARGSUSED */
471 static int
472 nfs_open(struct vop_open_args *ap)
473 {
474 struct vnode *vp = ap->a_vp;
475 struct nfsnode *np = VTONFS(vp);
476 struct vattr vattr;
477 int error;
478 int fmode = ap->a_mode;
479
480 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
481 return (EOPNOTSUPP);
482
483 /*
484 * For NFSv4, we need to do the Open Op before cache validation,
485 * so that we conform to RFC3530 Sec. 9.3.1.
486 */
487 if (NFS_ISV4(vp)) {
488 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
489 if (error) {
490 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
491 (gid_t)0);
492 return (error);
493 }
494 }
495
496 /*
497 * Now, if this Open will be doing reading, re-validate/flush the
498 * cache, so that Close/Open coherency is maintained.
499 */
500 mtx_lock(&np->n_mtx);
501 if (np->n_flag & NMODIFIED) {
502 mtx_unlock(&np->n_mtx);
503 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
504 if (error == EINTR || error == EIO) {
505 if (NFS_ISV4(vp))
506 (void) nfsrpc_close(vp, 0, ap->a_td);
507 return (error);
508 }
509 mtx_lock(&np->n_mtx);
510 np->n_attrstamp = 0;
511 if (vp->v_type == VDIR)
512 np->n_direofoffset = 0;
513 mtx_unlock(&np->n_mtx);
514 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
515 if (error) {
516 if (NFS_ISV4(vp))
517 (void) nfsrpc_close(vp, 0, ap->a_td);
518 return (error);
519 }
520 mtx_lock(&np->n_mtx);
521 np->n_mtime = vattr.va_mtime;
522 if (NFS_ISV4(vp))
523 np->n_change = vattr.va_filerev;
524 } else {
525 mtx_unlock(&np->n_mtx);
526 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
527 if (error) {
528 if (NFS_ISV4(vp))
529 (void) nfsrpc_close(vp, 0, ap->a_td);
530 return (error);
531 }
532 mtx_lock(&np->n_mtx);
533 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
534 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
535 if (vp->v_type == VDIR)
536 np->n_direofoffset = 0;
537 mtx_unlock(&np->n_mtx);
538 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
539 if (error == EINTR || error == EIO) {
540 if (NFS_ISV4(vp))
541 (void) nfsrpc_close(vp, 0, ap->a_td);
542 return (error);
543 }
544 mtx_lock(&np->n_mtx);
545 np->n_mtime = vattr.va_mtime;
546 if (NFS_ISV4(vp))
547 np->n_change = vattr.va_filerev;
548 }
549 }
550
551 /*
552 * If the object has >= 1 O_DIRECT active opens, we disable caching.
553 */
554 if (newnfs_directio_enable && (fmode & O_DIRECT) &&
555 (vp->v_type == VREG)) {
556 if (np->n_directio_opens == 0) {
557 mtx_unlock(&np->n_mtx);
558 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
559 if (error) {
560 if (NFS_ISV4(vp))
561 (void) nfsrpc_close(vp, 0, ap->a_td);
562 return (error);
563 }
564 mtx_lock(&np->n_mtx);
565 np->n_flag |= NNONCACHE;
566 }
567 np->n_directio_opens++;
568 }
569 mtx_unlock(&np->n_mtx);
570 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
571 return (0);
572 }
573
574 /*
575 * nfs close vnode op
576 * What an NFS client should do upon close after writing is a debatable issue.
577 * Most NFS clients push delayed writes to the server upon close, basically for
578 * two reasons:
579 * 1 - So that any write errors may be reported back to the client process
580 * doing the close system call. By far the two most likely errors are
581 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
582 * 2 - To put a worst case upper bound on cache inconsistency between
583 * multiple clients for the file.
584 * There is also a consistency problem for Version 2 of the protocol w.r.t.
585 * not being able to tell if other clients are writing a file concurrently,
586 * since there is no way of knowing if the changed modify time in the reply
587 * is only due to the write for this client.
588 * (NFS Version 3 provides weak cache consistency data in the reply that
589 * should be sufficient to detect and handle this case.)
590 *
591 * The current code does the following:
592 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
593 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
594 * or commit them (this satisfies 1 and 2 except for the
595 * case where the server crashes after this close but
596 * before the commit RPC, which is felt to be "good
597 * enough". Changing the last argument to ncl_flush() to
598 * a 1 would force a commit operation, if it is felt a
599 * commit is necessary now.
600 * for NFS Version 4 - flush the dirty buffers and commit them, if
601 * nfscl_mustflush() says this is necessary.
602 * It is necessary if there is no write delegation held,
603 * in order to satisfy open/close coherency.
604 * If the file isn't cached on local stable storage,
605 * it may be necessary in order to detect "out of space"
606 * errors from the server, if the write delegation
607 * issued by the server doesn't allow the file to grow.
608 */
609 /* ARGSUSED */
610 static int
611 nfs_close(struct vop_close_args *ap)
612 {
613 struct vnode *vp = ap->a_vp;
614 struct nfsnode *np = VTONFS(vp);
615 struct nfsvattr nfsva;
616 struct ucred *cred;
617 int error = 0, ret, localcred = 0;
618 int fmode = ap->a_fflag;
619
620 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
621 return (0);
622 /*
623 * During shutdown, a_cred isn't valid, so just use root.
624 */
625 if (ap->a_cred == NOCRED) {
626 cred = newnfs_getcred();
627 localcred = 1;
628 } else {
629 cred = ap->a_cred;
630 }
631 if (vp->v_type == VREG) {
632 /*
633 * Examine and clean dirty pages, regardless of NMODIFIED.
634 * This closes a major hole in close-to-open consistency.
635 * We want to push out all dirty pages (and buffers) on
636 * close, regardless of whether they were dirtied by
637 * mmap'ed writes or via write().
638 */
639 if (nfs_clean_pages_on_close && vp->v_object) {
640 VM_OBJECT_LOCK(vp->v_object);
641 vm_object_page_clean(vp->v_object, 0, 0, 0);
642 VM_OBJECT_UNLOCK(vp->v_object);
643 }
644 mtx_lock(&np->n_mtx);
645 if (np->n_flag & NMODIFIED) {
646 mtx_unlock(&np->n_mtx);
647 if (NFS_ISV3(vp)) {
648 /*
649 * Under NFSv3 we have dirty buffers to dispose of. We
650 * must flush them to the NFS server. We have the option
651 * of waiting all the way through the commit rpc or just
652 * waiting for the initial write. The default is to only
653 * wait through the initial write so the data is in the
654 * server's cache, which is roughly similar to the state
655 * a standard disk subsystem leaves the file in on close().
656 *
657 * We cannot clear the NMODIFIED bit in np->n_flag due to
658 * potential races with other processes, and certainly
659 * cannot clear it if we don't commit.
660 * These races occur when there is no longer the old
661 * traditional vnode locking implemented for Vnode Ops.
662 */
663 int cm = newnfs_commit_on_close ? 1 : 0;
664 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
665 /* np->n_flag &= ~NMODIFIED; */
666 } else if (NFS_ISV4(vp)) {
667 if (nfscl_mustflush(vp) != 0) {
668 int cm = newnfs_commit_on_close ? 1 : 0;
669 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
670 cm, 0);
671 /*
672 * as above w.r.t races when clearing
673 * NMODIFIED.
674 * np->n_flag &= ~NMODIFIED;
675 */
676 }
677 } else
678 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
679 mtx_lock(&np->n_mtx);
680 }
681 /*
682 * Invalidate the attribute cache in all cases.
683 * An open is going to fetch fresh attrs any way, other procs
684 * on this node that have file open will be forced to do an
685 * otw attr fetch, but this is safe.
686 * --> A user found that their RPC count dropped by 20% when
687 * this was commented out and I can't see any requirement
688 * for it, so I've disabled it when negative lookups are
689 * enabled. (What does this have to do with negative lookup
690 * caching? Well nothing, except it was reported by the
691 * same user that needed negative lookup caching and I wanted
692 * there to be a way to disable it to see if it
693 * is the cause of some caching/coherency issue that might
694 * crop up.)
695 */
696 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0)
697 np->n_attrstamp = 0;
698 if (np->n_flag & NWRITEERR) {
699 np->n_flag &= ~NWRITEERR;
700 error = np->n_error;
701 }
702 mtx_unlock(&np->n_mtx);
703 }
704
705 if (NFS_ISV4(vp)) {
706 /*
707 * Get attributes so "change" is up to date.
708 */
709 if (error == 0 && nfscl_mustflush(vp) != 0) {
710 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
711 NULL);
712 if (!ret) {
713 np->n_change = nfsva.na_filerev;
714 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
715 NULL, 0, 0);
716 }
717 }
718
719 /*
720 * and do the close.
721 */
722 ret = nfsrpc_close(vp, 0, ap->a_td);
723 if (!error && ret)
724 error = ret;
725 if (error)
726 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
727 (gid_t)0);
728 }
729 if (newnfs_directio_enable)
730 KASSERT((np->n_directio_asyncwr == 0),
731 ("nfs_close: dirty unflushed (%d) directio buffers\n",
732 np->n_directio_asyncwr));
733 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
734 mtx_lock(&np->n_mtx);
735 KASSERT((np->n_directio_opens > 0),
736 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
737 np->n_directio_opens--;
738 if (np->n_directio_opens == 0)
739 np->n_flag &= ~NNONCACHE;
740 mtx_unlock(&np->n_mtx);
741 }
742 if (localcred)
743 NFSFREECRED(cred);
744 return (error);
745 }
746
747 /*
748 * nfs getattr call from vfs.
749 */
750 static int
751 nfs_getattr(struct vop_getattr_args *ap)
752 {
753 struct vnode *vp = ap->a_vp;
754 struct thread *td = curthread; /* XXX */
755 struct nfsnode *np = VTONFS(vp);
756 int error = 0;
757 struct nfsvattr nfsva;
758 struct vattr *vap = ap->a_vap;
759 struct vattr vattr;
760
761 /*
762 * Update local times for special files.
763 */
764 mtx_lock(&np->n_mtx);
765 if (np->n_flag & (NACC | NUPD))
766 np->n_flag |= NCHG;
767 mtx_unlock(&np->n_mtx);
768 /*
769 * First look in the cache.
770 */
771 if (ncl_getattrcache(vp, &vattr) == 0) {
772 vap->va_type = vattr.va_type;
773 vap->va_mode = vattr.va_mode;
774 vap->va_nlink = vattr.va_nlink;
775 vap->va_uid = vattr.va_uid;
776 vap->va_gid = vattr.va_gid;
777 vap->va_fsid = vattr.va_fsid;
778 vap->va_fileid = vattr.va_fileid;
779 vap->va_size = vattr.va_size;
780 vap->va_blocksize = vattr.va_blocksize;
781 vap->va_atime = vattr.va_atime;
782 vap->va_mtime = vattr.va_mtime;
783 vap->va_ctime = vattr.va_ctime;
784 vap->va_gen = vattr.va_gen;
785 vap->va_flags = vattr.va_flags;
786 vap->va_rdev = vattr.va_rdev;
787 vap->va_bytes = vattr.va_bytes;
788 vap->va_filerev = vattr.va_filerev;
789 /*
790 * Get the local modify time for the case of a write
791 * delegation.
792 */
793 nfscl_deleggetmodtime(vp, &vap->va_mtime);
794 return (0);
795 }
796
797 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
798 nfsaccess_cache_timeout > 0) {
799 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
800 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
801 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
802 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
803 return (0);
804 }
805 }
806 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
807 if (!error)
808 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
809 if (!error) {
810 /*
811 * Get the local modify time for the case of a write
812 * delegation.
813 */
814 nfscl_deleggetmodtime(vp, &vap->va_mtime);
815 } else if (NFS_ISV4(vp)) {
816 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
817 }
818 return (error);
819 }
820
821 /*
822 * nfs setattr call.
823 */
824 static int
825 nfs_setattr(struct vop_setattr_args *ap)
826 {
827 struct vnode *vp = ap->a_vp;
828 struct nfsnode *np = VTONFS(vp);
829 struct thread *td = curthread; /* XXX */
830 struct vattr *vap = ap->a_vap;
831 int error = 0;
832 u_quad_t tsize;
833
834 #ifndef nolint
835 tsize = (u_quad_t)0;
836 #endif
837
838 /*
839 * Setting of flags and marking of atimes are not supported.
840 */
841 if (vap->va_flags != VNOVAL)
842 return (EOPNOTSUPP);
843
844 /*
845 * Disallow write attempts if the filesystem is mounted read-only.
846 */
847 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
848 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
849 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
850 (vp->v_mount->mnt_flag & MNT_RDONLY))
851 return (EROFS);
852 if (vap->va_size != VNOVAL) {
853 switch (vp->v_type) {
854 case VDIR:
855 return (EISDIR);
856 case VCHR:
857 case VBLK:
858 case VSOCK:
859 case VFIFO:
860 if (vap->va_mtime.tv_sec == VNOVAL &&
861 vap->va_atime.tv_sec == VNOVAL &&
862 vap->va_mode == (mode_t)VNOVAL &&
863 vap->va_uid == (uid_t)VNOVAL &&
864 vap->va_gid == (gid_t)VNOVAL)
865 return (0);
866 vap->va_size = VNOVAL;
867 break;
868 default:
869 /*
870 * Disallow write attempts if the filesystem is
871 * mounted read-only.
872 */
873 if (vp->v_mount->mnt_flag & MNT_RDONLY)
874 return (EROFS);
875 /*
876 * We run vnode_pager_setsize() early (why?),
877 * we must set np->n_size now to avoid vinvalbuf
878 * V_SAVE races that might setsize a lower
879 * value.
880 */
881 mtx_lock(&np->n_mtx);
882 tsize = np->n_size;
883 mtx_unlock(&np->n_mtx);
884 error = ncl_meta_setsize(vp, ap->a_cred, td,
885 vap->va_size);
886 mtx_lock(&np->n_mtx);
887 if (np->n_flag & NMODIFIED) {
888 tsize = np->n_size;
889 mtx_unlock(&np->n_mtx);
890 if (vap->va_size == 0)
891 error = ncl_vinvalbuf(vp, 0, td, 1);
892 else
893 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
894 if (error) {
895 vnode_pager_setsize(vp, tsize);
896 return (error);
897 }
898 /*
899 * Call nfscl_delegmodtime() to set the modify time
900 * locally, as required.
901 */
902 nfscl_delegmodtime(vp);
903 } else
904 mtx_unlock(&np->n_mtx);
905 /*
906 * np->n_size has already been set to vap->va_size
907 * in ncl_meta_setsize(). We must set it again since
908 * nfs_loadattrcache() could be called through
909 * ncl_meta_setsize() and could modify np->n_size.
910 */
911 mtx_lock(&np->n_mtx);
912 np->n_vattr.na_size = np->n_size = vap->va_size;
913 mtx_unlock(&np->n_mtx);
914 };
915 } else {
916 mtx_lock(&np->n_mtx);
917 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
918 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
919 mtx_unlock(&np->n_mtx);
920 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
921 (error == EINTR || error == EIO))
922 return (error);
923 } else
924 mtx_unlock(&np->n_mtx);
925 }
926 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
927 if (error && vap->va_size != VNOVAL) {
928 mtx_lock(&np->n_mtx);
929 np->n_size = np->n_vattr.na_size = tsize;
930 vnode_pager_setsize(vp, tsize);
931 mtx_unlock(&np->n_mtx);
932 }
933 return (error);
934 }
935
936 /*
937 * Do an nfs setattr rpc.
938 */
939 static int
940 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
941 struct thread *td)
942 {
943 struct nfsnode *np = VTONFS(vp);
944 int error, ret, attrflag, i;
945 struct nfsvattr nfsva;
946
947 if (NFS_ISV34(vp)) {
948 mtx_lock(&np->n_mtx);
949 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
950 np->n_accesscache[i].stamp = 0;
951 np->n_flag |= NDELEGMOD;
952 mtx_unlock(&np->n_mtx);
953 }
954 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
955 NULL);
956 if (attrflag) {
957 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
958 if (ret && !error)
959 error = ret;
960 }
961 if (error && NFS_ISV4(vp))
962 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
963 return (error);
964 }
965
966 /*
967 * nfs lookup call, one step at a time...
968 * First look in cache
969 * If not found, unlock the directory nfsnode and do the rpc
970 */
971 static int
972 nfs_lookup(struct vop_lookup_args *ap)
973 {
974 struct componentname *cnp = ap->a_cnp;
975 struct vnode *dvp = ap->a_dvp;
976 struct vnode **vpp = ap->a_vpp;
977 struct mount *mp = dvp->v_mount;
978 int flags = cnp->cn_flags;
979 struct vnode *newvp;
980 struct nfsmount *nmp;
981 struct nfsnode *np, *newnp;
982 int error = 0, attrflag, dattrflag, ltype;
983 struct thread *td = cnp->cn_thread;
984 struct nfsfh *nfhp;
985 struct nfsvattr dnfsva, nfsva;
986 struct vattr vattr;
987 struct timespec dmtime;
988
989 *vpp = NULLVP;
990 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
991 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
992 return (EROFS);
993 if (dvp->v_type != VDIR)
994 return (ENOTDIR);
995 nmp = VFSTONFS(mp);
996 np = VTONFS(dvp);
997
998 /* For NFSv4, wait until any remove is done. */
999 mtx_lock(&np->n_mtx);
1000 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1001 np->n_flag |= NREMOVEWANT;
1002 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1003 }
1004 mtx_unlock(&np->n_mtx);
1005
1006 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1007 return (error);
1008 error = cache_lookup(dvp, vpp, cnp);
1009 if (error > 0 && error != ENOENT)
1010 return (error);
1011 if (error == -1) {
1012 /*
1013 * We only accept a positive hit in the cache if the
1014 * change time of the file matches our cached copy.
1015 * Otherwise, we discard the cache entry and fallback
1016 * to doing a lookup RPC.
1017 *
1018 * To better handle stale file handles and attributes,
1019 * clear the attribute cache of this node if it is a
1020 * leaf component, part of an open() call, and not
1021 * locally modified before fetching the attributes.
1022 * This should allow stale file handles to be detected
1023 * here where we can fall back to a LOOKUP RPC to
1024 * recover rather than having nfs_open() detect the
1025 * stale file handle and failing open(2) with ESTALE.
1026 */
1027 newvp = *vpp;
1028 newnp = VTONFS(newvp);
1029 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1030 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1031 !(newnp->n_flag & NMODIFIED)) {
1032 mtx_lock(&newnp->n_mtx);
1033 newnp->n_attrstamp = 0;
1034 mtx_unlock(&newnp->n_mtx);
1035 }
1036 if (nfscl_nodeleg(newvp, 0) == 0 ||
1037 (VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1038 timespeccmp(&vattr.va_ctime, &newnp->n_ctime, ==))) {
1039 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1040 if (cnp->cn_nameiop != LOOKUP &&
1041 (flags & ISLASTCN))
1042 cnp->cn_flags |= SAVENAME;
1043 return (0);
1044 }
1045 cache_purge(newvp);
1046 if (dvp != newvp)
1047 vput(newvp);
1048 else
1049 vrele(newvp);
1050 *vpp = NULLVP;
1051 } else if (error == ENOENT) {
1052 if (dvp->v_iflag & VI_DOOMED)
1053 return (ENOENT);
1054 /*
1055 * We only accept a negative hit in the cache if the
1056 * modification time of the parent directory matches
1057 * our cached copy. Otherwise, we discard all of the
1058 * negative cache entries for this directory. We also
1059 * only trust -ve cache entries for less than
1060 * nm_negative_namecache_timeout seconds.
1061 */
1062 if ((u_int)(ticks - np->n_dmtime_ticks) <
1063 (nmp->nm_negnametimeo * hz) &&
1064 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1065 timespeccmp(&vattr.va_mtime, &np->n_dmtime, ==)) {
1066 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1067 return (ENOENT);
1068 }
1069 cache_purge_negative(dvp);
1070 mtx_lock(&np->n_mtx);
1071 timespecclear(&np->n_dmtime);
1072 mtx_unlock(&np->n_mtx);
1073 }
1074
1075 /*
1076 * Cache the modification time of the parent directory in case
1077 * the lookup fails and results in adding the first negative
1078 * name cache entry for the directory. Since this is reading
1079 * a single time_t, don't bother with locking. The
1080 * modification time may be a bit stale, but it must be read
1081 * before performing the lookup RPC to prevent a race where
1082 * another lookup updates the timestamp on the directory after
1083 * the lookup RPC has been performed on the server but before
1084 * n_dmtime is set at the end of this function.
1085 */
1086 dmtime = np->n_vattr.na_mtime;
1087 error = 0;
1088 newvp = NULLVP;
1089 NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1090 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1091 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1092 NULL);
1093 if (dattrflag)
1094 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1095 if (error) {
1096 if (newvp != NULLVP) {
1097 vput(newvp);
1098 *vpp = NULLVP;
1099 }
1100
1101 if (error != ENOENT) {
1102 if (NFS_ISV4(dvp))
1103 error = nfscl_maperr(td, error, (uid_t)0,
1104 (gid_t)0);
1105 return (error);
1106 }
1107
1108 /* The requested file was not found. */
1109 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1110 (flags & ISLASTCN)) {
1111 /*
1112 * XXX: UFS does a full VOP_ACCESS(dvp,
1113 * VWRITE) here instead of just checking
1114 * MNT_RDONLY.
1115 */
1116 if (mp->mnt_flag & MNT_RDONLY)
1117 return (EROFS);
1118 cnp->cn_flags |= SAVENAME;
1119 return (EJUSTRETURN);
1120 }
1121
1122 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1123 /*
1124 * Maintain n_dmtime as the modification time
1125 * of the parent directory when the oldest -ve
1126 * name cache entry for this directory was
1127 * added. If a -ve cache entry has already
1128 * been added with a newer modification time
1129 * by a concurrent lookup, then don't bother
1130 * adding a cache entry. The modification
1131 * time of the directory might have changed
1132 * due to the file this lookup failed to find
1133 * being created. In that case a subsequent
1134 * lookup would incorrectly use the entry
1135 * added here instead of doing an extra
1136 * lookup.
1137 */
1138 mtx_lock(&np->n_mtx);
1139 if (timespeccmp(&np->n_dmtime, &dmtime, <=)) {
1140 if (!timespecisset(&np->n_dmtime)) {
1141 np->n_dmtime = dmtime;
1142 np->n_dmtime_ticks = ticks;
1143 }
1144 mtx_unlock(&np->n_mtx);
1145 cache_enter(dvp, NULL, cnp);
1146 } else
1147 mtx_unlock(&np->n_mtx);
1148 }
1149 return (ENOENT);
1150 }
1151
1152 /*
1153 * Handle RENAME case...
1154 */
1155 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1156 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1157 FREE((caddr_t)nfhp, M_NFSFH);
1158 return (EISDIR);
1159 }
1160 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1161 LK_EXCLUSIVE);
1162 if (error)
1163 return (error);
1164 newvp = NFSTOV(np);
1165 if (attrflag)
1166 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1167 0, 1);
1168 *vpp = newvp;
1169 cnp->cn_flags |= SAVENAME;
1170 return (0);
1171 }
1172
1173 if (flags & ISDOTDOT) {
1174 ltype = NFSVOPISLOCKED(dvp);
1175 error = vfs_busy(mp, MBF_NOWAIT);
1176 if (error != 0) {
1177 vfs_ref(mp);
1178 NFSVOPUNLOCK(dvp, 0);
1179 error = vfs_busy(mp, 0);
1180 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1181 vfs_rel(mp);
1182 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1183 vfs_unbusy(mp);
1184 error = ENOENT;
1185 }
1186 if (error != 0)
1187 return (error);
1188 }
1189 NFSVOPUNLOCK(dvp, 0);
1190 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1191 cnp->cn_lkflags);
1192 if (error == 0)
1193 newvp = NFSTOV(np);
1194 vfs_unbusy(mp);
1195 if (newvp != dvp)
1196 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1197 if (dvp->v_iflag & VI_DOOMED) {
1198 if (error == 0) {
1199 if (newvp == dvp)
1200 vrele(newvp);
1201 else
1202 vput(newvp);
1203 }
1204 error = ENOENT;
1205 }
1206 if (error != 0)
1207 return (error);
1208 if (attrflag)
1209 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1210 0, 1);
1211 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1212 FREE((caddr_t)nfhp, M_NFSFH);
1213 VREF(dvp);
1214 newvp = dvp;
1215 if (attrflag)
1216 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1217 0, 1);
1218 } else {
1219 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1220 cnp->cn_lkflags);
1221 if (error)
1222 return (error);
1223 newvp = NFSTOV(np);
1224 if (attrflag)
1225 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1226 0, 1);
1227 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1228 !(np->n_flag & NMODIFIED)) {
1229 /*
1230 * Flush the attribute cache when opening a
1231 * leaf node to ensure that fresh attributes
1232 * are fetched in nfs_open() since we did not
1233 * fetch attributes from the LOOKUP reply.
1234 */
1235 mtx_lock(&np->n_mtx);
1236 np->n_attrstamp = 0;
1237 mtx_unlock(&np->n_mtx);
1238 }
1239 }
1240 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1241 cnp->cn_flags |= SAVENAME;
1242 if ((cnp->cn_flags & MAKEENTRY) &&
1243 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1244 np->n_ctime = np->n_vattr.na_vattr.va_ctime;
1245 cache_enter(dvp, newvp, cnp);
1246 }
1247 *vpp = newvp;
1248 return (0);
1249 }
1250
1251 /*
1252 * nfs read call.
1253 * Just call ncl_bioread() to do the work.
1254 */
1255 static int
1256 nfs_read(struct vop_read_args *ap)
1257 {
1258 struct vnode *vp = ap->a_vp;
1259
1260 switch (vp->v_type) {
1261 case VREG:
1262 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1263 case VDIR:
1264 return (EISDIR);
1265 default:
1266 return (EOPNOTSUPP);
1267 }
1268 }
1269
1270 /*
1271 * nfs readlink call
1272 */
1273 static int
1274 nfs_readlink(struct vop_readlink_args *ap)
1275 {
1276 struct vnode *vp = ap->a_vp;
1277
1278 if (vp->v_type != VLNK)
1279 return (EINVAL);
1280 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1281 }
1282
1283 /*
1284 * Do a readlink rpc.
1285 * Called by ncl_doio() from below the buffer cache.
1286 */
1287 int
1288 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1289 {
1290 int error, ret, attrflag;
1291 struct nfsvattr nfsva;
1292
1293 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1294 &attrflag, NULL);
1295 if (attrflag) {
1296 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1297 if (ret && !error)
1298 error = ret;
1299 }
1300 if (error && NFS_ISV4(vp))
1301 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1302 return (error);
1303 }
1304
1305 /*
1306 * nfs read rpc call
1307 * Ditto above
1308 */
1309 int
1310 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1311 {
1312 int error, ret, attrflag;
1313 struct nfsvattr nfsva;
1314
1315 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1316 NULL);
1317 if (attrflag) {
1318 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1319 if (ret && !error)
1320 error = ret;
1321 }
1322 if (error && NFS_ISV4(vp))
1323 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1324 return (error);
1325 }
1326
1327 /*
1328 * nfs write call
1329 */
1330 int
1331 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1332 int *iomode, int *must_commit, int called_from_strategy)
1333 {
1334 struct nfsvattr nfsva;
1335 int error = 0, attrflag, ret;
1336
1337 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1338 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy);
1339 if (attrflag) {
1340 if (VTONFS(vp)->n_flag & ND_NFSV4)
1341 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1342 1);
1343 else
1344 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1345 1);
1346 if (ret && !error)
1347 error = ret;
1348 }
1349 if (DOINGASYNC(vp))
1350 *iomode = NFSWRITE_FILESYNC;
1351 if (error && NFS_ISV4(vp))
1352 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1353 return (error);
1354 }
1355
1356 /*
1357 * nfs mknod rpc
1358 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1359 * mode set to specify the file type and the size field for rdev.
1360 */
1361 static int
1362 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1363 struct vattr *vap)
1364 {
1365 struct nfsvattr nfsva, dnfsva;
1366 struct vnode *newvp = NULL;
1367 struct nfsnode *np = NULL, *dnp;
1368 struct nfsfh *nfhp;
1369 struct vattr vattr;
1370 int error = 0, attrflag, dattrflag;
1371 u_int32_t rdev;
1372
1373 if (vap->va_type == VCHR || vap->va_type == VBLK)
1374 rdev = vap->va_rdev;
1375 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1376 rdev = 0xffffffff;
1377 else
1378 return (EOPNOTSUPP);
1379 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1380 return (error);
1381 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1382 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1383 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1384 if (!error) {
1385 if (!nfhp)
1386 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1387 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1388 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1389 NULL);
1390 if (nfhp)
1391 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1392 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1393 }
1394 if (dattrflag)
1395 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1396 if (!error) {
1397 newvp = NFSTOV(np);
1398 if (attrflag != 0) {
1399 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1400 0, 1);
1401 if (error != 0)
1402 vput(newvp);
1403 }
1404 }
1405 if (!error) {
1406 if ((cnp->cn_flags & MAKEENTRY))
1407 cache_enter(dvp, newvp, cnp);
1408 *vpp = newvp;
1409 } else if (NFS_ISV4(dvp)) {
1410 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1411 vap->va_gid);
1412 }
1413 dnp = VTONFS(dvp);
1414 mtx_lock(&dnp->n_mtx);
1415 dnp->n_flag |= NMODIFIED;
1416 if (!dattrflag)
1417 dnp->n_attrstamp = 0;
1418 mtx_unlock(&dnp->n_mtx);
1419 return (error);
1420 }
1421
1422 /*
1423 * nfs mknod vop
1424 * just call nfs_mknodrpc() to do the work.
1425 */
1426 /* ARGSUSED */
1427 static int
1428 nfs_mknod(struct vop_mknod_args *ap)
1429 {
1430 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1431 }
1432
1433 static u_long create_verf;
1434 /*
1435 * nfs file create call
1436 */
1437 static int
1438 nfs_create(struct vop_create_args *ap)
1439 {
1440 struct vnode *dvp = ap->a_dvp;
1441 struct vattr *vap = ap->a_vap;
1442 struct componentname *cnp = ap->a_cnp;
1443 struct nfsnode *np = NULL, *dnp;
1444 struct vnode *newvp = NULL;
1445 struct nfsmount *nmp;
1446 struct nfsvattr dnfsva, nfsva;
1447 struct nfsfh *nfhp;
1448 nfsquad_t cverf;
1449 int error = 0, attrflag, dattrflag, fmode = 0;
1450 struct vattr vattr;
1451
1452 /*
1453 * Oops, not for me..
1454 */
1455 if (vap->va_type == VSOCK)
1456 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1457
1458 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1459 return (error);
1460 if (vap->va_vaflags & VA_EXCLUSIVE)
1461 fmode |= O_EXCL;
1462 dnp = VTONFS(dvp);
1463 nmp = VFSTONFS(vnode_mount(dvp));
1464 again:
1465 /* For NFSv4, wait until any remove is done. */
1466 mtx_lock(&dnp->n_mtx);
1467 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1468 dnp->n_flag |= NREMOVEWANT;
1469 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1470 }
1471 mtx_unlock(&dnp->n_mtx);
1472
1473 #ifdef INET
1474 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1475 IN_IFADDR_RLOCK();
1476 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1477 cverf.lval[0] = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1478 else
1479 #endif
1480 cverf.lval[0] = create_verf;
1481 #ifdef INET
1482 IN_IFADDR_RUNLOCK();
1483 CURVNET_RESTORE();
1484 #endif
1485 cverf.lval[1] = ++create_verf;
1486 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1487 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1488 &nfhp, &attrflag, &dattrflag, NULL);
1489 if (!error) {
1490 if (nfhp == NULL)
1491 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1492 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1493 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1494 NULL);
1495 if (nfhp != NULL)
1496 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1497 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1498 }
1499 if (dattrflag)
1500 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1501 if (!error) {
1502 newvp = NFSTOV(np);
1503 if (attrflag)
1504 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1505 0, 1);
1506 }
1507 if (error) {
1508 if (newvp != NULL) {
1509 vput(newvp);
1510 newvp = NULL;
1511 }
1512 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1513 error == NFSERR_NOTSUPP) {
1514 fmode &= ~O_EXCL;
1515 goto again;
1516 }
1517 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1518 if (nfscl_checksattr(vap, &nfsva)) {
1519 /*
1520 * We are normally called with only a partially
1521 * initialized VAP. Since the NFSv3 spec says that
1522 * the server may use the file attributes to
1523 * store the verifier, the spec requires us to do a
1524 * SETATTR RPC. FreeBSD servers store the verifier in
1525 * atime, but we can't really assume that all servers
1526 * will so we ensure that our SETATTR sets both atime
1527 * and mtime.
1528 */
1529 if (vap->va_mtime.tv_sec == VNOVAL)
1530 vfs_timestamp(&vap->va_mtime);
1531 if (vap->va_atime.tv_sec == VNOVAL)
1532 vap->va_atime = vap->va_mtime;
1533 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1534 cnp->cn_thread, &nfsva, &attrflag, NULL);
1535 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1536 vap->va_gid != (gid_t)VNOVAL)) {
1537 /* try again without setting uid/gid */
1538 vap->va_uid = (uid_t)VNOVAL;
1539 vap->va_gid = (uid_t)VNOVAL;
1540 error = nfsrpc_setattr(newvp, vap, NULL,
1541 cnp->cn_cred, cnp->cn_thread, &nfsva,
1542 &attrflag, NULL);
1543 }
1544 if (attrflag)
1545 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1546 NULL, 0, 1);
1547 if (error != 0)
1548 vput(newvp);
1549 }
1550 }
1551 if (!error) {
1552 if (cnp->cn_flags & MAKEENTRY)
1553 cache_enter(dvp, newvp, cnp);
1554 *ap->a_vpp = newvp;
1555 } else if (NFS_ISV4(dvp)) {
1556 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1557 vap->va_gid);
1558 }
1559 mtx_lock(&dnp->n_mtx);
1560 dnp->n_flag |= NMODIFIED;
1561 if (!dattrflag)
1562 dnp->n_attrstamp = 0;
1563 mtx_unlock(&dnp->n_mtx);
1564 return (error);
1565 }
1566
1567 /*
1568 * nfs file remove call
1569 * To try and make nfs semantics closer to ufs semantics, a file that has
1570 * other processes using the vnode is renamed instead of removed and then
1571 * removed later on the last close.
1572 * - If v_usecount > 1
1573 * If a rename is not already in the works
1574 * call nfs_sillyrename() to set it up
1575 * else
1576 * do the remove rpc
1577 */
1578 static int
1579 nfs_remove(struct vop_remove_args *ap)
1580 {
1581 struct vnode *vp = ap->a_vp;
1582 struct vnode *dvp = ap->a_dvp;
1583 struct componentname *cnp = ap->a_cnp;
1584 struct nfsnode *np = VTONFS(vp);
1585 int error = 0;
1586 struct vattr vattr;
1587
1588 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1589 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1590 if (vp->v_type == VDIR)
1591 error = EPERM;
1592 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1593 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1594 vattr.va_nlink > 1)) {
1595 /*
1596 * Purge the name cache so that the chance of a lookup for
1597 * the name succeeding while the remove is in progress is
1598 * minimized. Without node locking it can still happen, such
1599 * that an I/O op returns ESTALE, but since you get this if
1600 * another host removes the file..
1601 */
1602 cache_purge(vp);
1603 /*
1604 * throw away biocache buffers, mainly to avoid
1605 * unnecessary delayed writes later.
1606 */
1607 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1608 /* Do the rpc */
1609 if (error != EINTR && error != EIO)
1610 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1611 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1612 /*
1613 * Kludge City: If the first reply to the remove rpc is lost..
1614 * the reply to the retransmitted request will be ENOENT
1615 * since the file was in fact removed
1616 * Therefore, we cheat and return success.
1617 */
1618 if (error == ENOENT)
1619 error = 0;
1620 } else if (!np->n_sillyrename)
1621 error = nfs_sillyrename(dvp, vp, cnp);
1622 mtx_lock(&np->n_mtx);
1623 np->n_attrstamp = 0;
1624 mtx_unlock(&np->n_mtx);
1625 return (error);
1626 }
1627
1628 /*
1629 * nfs file remove rpc called from nfs_inactive
1630 */
1631 int
1632 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1633 {
1634 /*
1635 * Make sure that the directory vnode is still valid.
1636 * XXX we should lock sp->s_dvp here.
1637 */
1638 if (sp->s_dvp->v_type == VBAD)
1639 return (0);
1640 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1641 sp->s_cred, NULL));
1642 }
1643
1644 /*
1645 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1646 */
1647 static int
1648 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1649 int namelen, struct ucred *cred, struct thread *td)
1650 {
1651 struct nfsvattr dnfsva;
1652 struct nfsnode *dnp = VTONFS(dvp);
1653 int error = 0, dattrflag;
1654
1655 mtx_lock(&dnp->n_mtx);
1656 dnp->n_flag |= NREMOVEINPROG;
1657 mtx_unlock(&dnp->n_mtx);
1658 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1659 &dattrflag, NULL);
1660 mtx_lock(&dnp->n_mtx);
1661 if ((dnp->n_flag & NREMOVEWANT)) {
1662 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1663 mtx_unlock(&dnp->n_mtx);
1664 wakeup((caddr_t)dnp);
1665 } else {
1666 dnp->n_flag &= ~NREMOVEINPROG;
1667 mtx_unlock(&dnp->n_mtx);
1668 }
1669 if (dattrflag)
1670 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1671 mtx_lock(&dnp->n_mtx);
1672 dnp->n_flag |= NMODIFIED;
1673 if (!dattrflag)
1674 dnp->n_attrstamp = 0;
1675 mtx_unlock(&dnp->n_mtx);
1676 if (error && NFS_ISV4(dvp))
1677 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1678 return (error);
1679 }
1680
1681 /*
1682 * nfs file rename call
1683 */
1684 static int
1685 nfs_rename(struct vop_rename_args *ap)
1686 {
1687 struct vnode *fvp = ap->a_fvp;
1688 struct vnode *tvp = ap->a_tvp;
1689 struct vnode *fdvp = ap->a_fdvp;
1690 struct vnode *tdvp = ap->a_tdvp;
1691 struct componentname *tcnp = ap->a_tcnp;
1692 struct componentname *fcnp = ap->a_fcnp;
1693 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1694 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1695 struct nfsv4node *newv4 = NULL;
1696 int error;
1697
1698 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1699 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1700 /* Check for cross-device rename */
1701 if ((fvp->v_mount != tdvp->v_mount) ||
1702 (tvp && (fvp->v_mount != tvp->v_mount))) {
1703 error = EXDEV;
1704 goto out;
1705 }
1706
1707 if (fvp == tvp) {
1708 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1709 error = 0;
1710 goto out;
1711 }
1712 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1713 goto out;
1714
1715 /*
1716 * We have to flush B_DELWRI data prior to renaming
1717 * the file. If we don't, the delayed-write buffers
1718 * can be flushed out later after the file has gone stale
1719 * under NFSV3. NFSV2 does not have this problem because
1720 * ( as far as I can tell ) it flushes dirty buffers more
1721 * often.
1722 *
1723 * Skip the rename operation if the fsync fails, this can happen
1724 * due to the server's volume being full, when we pushed out data
1725 * that was written back to our cache earlier. Not checking for
1726 * this condition can result in potential (silent) data loss.
1727 */
1728 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1729 NFSVOPUNLOCK(fvp, 0);
1730 if (!error && tvp)
1731 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1732 if (error)
1733 goto out;
1734
1735 /*
1736 * If the tvp exists and is in use, sillyrename it before doing the
1737 * rename of the new file over it.
1738 * XXX Can't sillyrename a directory.
1739 */
1740 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1741 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1742 vput(tvp);
1743 tvp = NULL;
1744 }
1745
1746 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1747 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1748 tcnp->cn_thread);
1749
1750 if (error == 0 && NFS_ISV4(tdvp)) {
1751 /*
1752 * For NFSv4, check to see if it is the same name and
1753 * replace the name, if it is different.
1754 */
1755 MALLOC(newv4, struct nfsv4node *,
1756 sizeof (struct nfsv4node) +
1757 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1758 M_NFSV4NODE, M_WAITOK);
1759 mtx_lock(&tdnp->n_mtx);
1760 mtx_lock(&fnp->n_mtx);
1761 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1762 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1763 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1764 tcnp->cn_namelen) ||
1765 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1766 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1767 tdnp->n_fhp->nfh_len))) {
1768 #ifdef notdef
1769 { char nnn[100]; int nnnl;
1770 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1771 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1772 nnn[nnnl] = '\0';
1773 printf("ren replace=%s\n",nnn);
1774 }
1775 #endif
1776 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1777 fnp->n_v4 = newv4;
1778 newv4 = NULL;
1779 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1780 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1781 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1782 tdnp->n_fhp->nfh_len);
1783 NFSBCOPY(tcnp->cn_nameptr,
1784 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1785 }
1786 mtx_unlock(&tdnp->n_mtx);
1787 mtx_unlock(&fnp->n_mtx);
1788 if (newv4 != NULL)
1789 FREE((caddr_t)newv4, M_NFSV4NODE);
1790 }
1791
1792 if (fvp->v_type == VDIR) {
1793 if (tvp != NULL && tvp->v_type == VDIR)
1794 cache_purge(tdvp);
1795 cache_purge(fdvp);
1796 }
1797
1798 out:
1799 if (tdvp == tvp)
1800 vrele(tdvp);
1801 else
1802 vput(tdvp);
1803 if (tvp)
1804 vput(tvp);
1805 vrele(fdvp);
1806 vrele(fvp);
1807 /*
1808 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1809 */
1810 if (error == ENOENT)
1811 error = 0;
1812 return (error);
1813 }
1814
1815 /*
1816 * nfs file rename rpc called from nfs_remove() above
1817 */
1818 static int
1819 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1820 struct sillyrename *sp)
1821 {
1822
1823 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1824 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1825 scnp->cn_thread));
1826 }
1827
1828 /*
1829 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1830 */
1831 static int
1832 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1833 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1834 int tnamelen, struct ucred *cred, struct thread *td)
1835 {
1836 struct nfsvattr fnfsva, tnfsva;
1837 struct nfsnode *fdnp = VTONFS(fdvp);
1838 struct nfsnode *tdnp = VTONFS(tdvp);
1839 int error = 0, fattrflag, tattrflag;
1840
1841 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1842 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1843 &tattrflag, NULL, NULL);
1844 mtx_lock(&fdnp->n_mtx);
1845 fdnp->n_flag |= NMODIFIED;
1846 if (fattrflag != 0) {
1847 mtx_unlock(&fdnp->n_mtx);
1848 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1849 } else {
1850 fdnp->n_attrstamp = 0;
1851 mtx_unlock(&fdnp->n_mtx);
1852 }
1853 mtx_lock(&tdnp->n_mtx);
1854 tdnp->n_flag |= NMODIFIED;
1855 if (tattrflag != 0) {
1856 mtx_unlock(&tdnp->n_mtx);
1857 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1858 } else {
1859 tdnp->n_attrstamp = 0;
1860 mtx_unlock(&tdnp->n_mtx);
1861 }
1862 if (error && NFS_ISV4(fdvp))
1863 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1864 return (error);
1865 }
1866
1867 /*
1868 * nfs hard link create call
1869 */
1870 static int
1871 nfs_link(struct vop_link_args *ap)
1872 {
1873 struct vnode *vp = ap->a_vp;
1874 struct vnode *tdvp = ap->a_tdvp;
1875 struct componentname *cnp = ap->a_cnp;
1876 struct nfsnode *np, *tdnp;
1877 struct nfsvattr nfsva, dnfsva;
1878 int error = 0, attrflag, dattrflag;
1879
1880 if (vp->v_mount != tdvp->v_mount) {
1881 return (EXDEV);
1882 }
1883
1884 /*
1885 * Push all writes to the server, so that the attribute cache
1886 * doesn't get "out of sync" with the server.
1887 * XXX There should be a better way!
1888 */
1889 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1890
1891 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1892 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1893 &dattrflag, NULL);
1894 tdnp = VTONFS(tdvp);
1895 mtx_lock(&tdnp->n_mtx);
1896 tdnp->n_flag |= NMODIFIED;
1897 if (dattrflag != 0) {
1898 mtx_unlock(&tdnp->n_mtx);
1899 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1900 } else {
1901 tdnp->n_attrstamp = 0;
1902 mtx_unlock(&tdnp->n_mtx);
1903 }
1904 if (attrflag)
1905 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1906 else {
1907 np = VTONFS(vp);
1908 mtx_lock(&np->n_mtx);
1909 np->n_attrstamp = 0;
1910 mtx_unlock(&np->n_mtx);
1911 }
1912 /*
1913 * If negative lookup caching is enabled, I might as well
1914 * add an entry for this node. Not necessary for correctness,
1915 * but if negative caching is enabled, then the system
1916 * must care about lookup caching hit rate, so...
1917 */
1918 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1919 (cnp->cn_flags & MAKEENTRY))
1920 cache_enter(tdvp, vp, cnp);
1921 if (error && NFS_ISV4(vp))
1922 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1923 (gid_t)0);
1924 return (error);
1925 }
1926
1927 /*
1928 * nfs symbolic link create call
1929 */
1930 static int
1931 nfs_symlink(struct vop_symlink_args *ap)
1932 {
1933 struct vnode *dvp = ap->a_dvp;
1934 struct vattr *vap = ap->a_vap;
1935 struct componentname *cnp = ap->a_cnp;
1936 struct nfsvattr nfsva, dnfsva;
1937 struct nfsfh *nfhp;
1938 struct nfsnode *np = NULL, *dnp;
1939 struct vnode *newvp = NULL;
1940 int error = 0, attrflag, dattrflag, ret;
1941
1942 vap->va_type = VLNK;
1943 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1944 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1945 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1946 if (nfhp) {
1947 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
1948 &np, NULL, LK_EXCLUSIVE);
1949 if (!ret)
1950 newvp = NFSTOV(np);
1951 else if (!error)
1952 error = ret;
1953 }
1954 if (newvp != NULL) {
1955 if (attrflag)
1956 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1957 0, 1);
1958 } else if (!error) {
1959 /*
1960 * If we do not have an error and we could not extract the
1961 * newvp from the response due to the request being NFSv2, we
1962 * have to do a lookup in order to obtain a newvp to return.
1963 */
1964 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1965 cnp->cn_cred, cnp->cn_thread, &np);
1966 if (!error)
1967 newvp = NFSTOV(np);
1968 }
1969 if (error) {
1970 if (newvp)
1971 vput(newvp);
1972 if (NFS_ISV4(dvp))
1973 error = nfscl_maperr(cnp->cn_thread, error,
1974 vap->va_uid, vap->va_gid);
1975 } else {
1976 /*
1977 * If negative lookup caching is enabled, I might as well
1978 * add an entry for this node. Not necessary for correctness,
1979 * but if negative caching is enabled, then the system
1980 * must care about lookup caching hit rate, so...
1981 */
1982 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
1983 (cnp->cn_flags & MAKEENTRY))
1984 cache_enter(dvp, newvp, cnp);
1985 *ap->a_vpp = newvp;
1986 }
1987
1988 dnp = VTONFS(dvp);
1989 mtx_lock(&dnp->n_mtx);
1990 dnp->n_flag |= NMODIFIED;
1991 if (dattrflag != 0) {
1992 mtx_unlock(&dnp->n_mtx);
1993 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1994 } else {
1995 dnp->n_attrstamp = 0;
1996 mtx_unlock(&dnp->n_mtx);
1997 }
1998 return (error);
1999 }
2000
2001 /*
2002 * nfs make dir call
2003 */
2004 static int
2005 nfs_mkdir(struct vop_mkdir_args *ap)
2006 {
2007 struct vnode *dvp = ap->a_dvp;
2008 struct vattr *vap = ap->a_vap;
2009 struct componentname *cnp = ap->a_cnp;
2010 struct nfsnode *np = NULL, *dnp;
2011 struct vnode *newvp = NULL;
2012 struct vattr vattr;
2013 struct nfsfh *nfhp;
2014 struct nfsvattr nfsva, dnfsva;
2015 int error = 0, attrflag, dattrflag, ret;
2016
2017 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2018 return (error);
2019 vap->va_type = VDIR;
2020 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2021 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2022 &attrflag, &dattrflag, NULL);
2023 dnp = VTONFS(dvp);
2024 mtx_lock(&dnp->n_mtx);
2025 dnp->n_flag |= NMODIFIED;
2026 if (dattrflag != 0) {
2027 mtx_unlock(&dnp->n_mtx);
2028 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2029 } else {
2030 dnp->n_attrstamp = 0;
2031 mtx_unlock(&dnp->n_mtx);
2032 }
2033 if (nfhp) {
2034 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2035 &np, NULL, LK_EXCLUSIVE);
2036 if (!ret) {
2037 newvp = NFSTOV(np);
2038 if (attrflag)
2039 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2040 NULL, 0, 1);
2041 } else if (!error)
2042 error = ret;
2043 }
2044 if (!error && newvp == NULL) {
2045 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2046 cnp->cn_cred, cnp->cn_thread, &np);
2047 if (!error) {
2048 newvp = NFSTOV(np);
2049 if (newvp->v_type != VDIR)
2050 error = EEXIST;
2051 }
2052 }
2053 if (error) {
2054 if (newvp)
2055 vput(newvp);
2056 if (NFS_ISV4(dvp))
2057 error = nfscl_maperr(cnp->cn_thread, error,
2058 vap->va_uid, vap->va_gid);
2059 } else {
2060 /*
2061 * If negative lookup caching is enabled, I might as well
2062 * add an entry for this node. Not necessary for correctness,
2063 * but if negative caching is enabled, then the system
2064 * must care about lookup caching hit rate, so...
2065 */
2066 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2067 (cnp->cn_flags & MAKEENTRY))
2068 cache_enter(dvp, newvp, cnp);
2069 *ap->a_vpp = newvp;
2070 }
2071 return (error);
2072 }
2073
2074 /*
2075 * nfs remove directory call
2076 */
2077 static int
2078 nfs_rmdir(struct vop_rmdir_args *ap)
2079 {
2080 struct vnode *vp = ap->a_vp;
2081 struct vnode *dvp = ap->a_dvp;
2082 struct componentname *cnp = ap->a_cnp;
2083 struct nfsnode *dnp;
2084 struct nfsvattr dnfsva;
2085 int error, dattrflag;
2086
2087 if (dvp == vp)
2088 return (EINVAL);
2089 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2090 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2091 dnp = VTONFS(dvp);
2092 mtx_lock(&dnp->n_mtx);
2093 dnp->n_flag |= NMODIFIED;
2094 if (dattrflag != 0) {
2095 mtx_unlock(&dnp->n_mtx);
2096 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2097 } else {
2098 dnp->n_attrstamp = 0;
2099 mtx_unlock(&dnp->n_mtx);
2100 }
2101
2102 cache_purge(dvp);
2103 cache_purge(vp);
2104 if (error && NFS_ISV4(dvp))
2105 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2106 (gid_t)0);
2107 /*
2108 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2109 */
2110 if (error == ENOENT)
2111 error = 0;
2112 return (error);
2113 }
2114
2115 /*
2116 * nfs readdir call
2117 */
2118 static int
2119 nfs_readdir(struct vop_readdir_args *ap)
2120 {
2121 struct vnode *vp = ap->a_vp;
2122 struct nfsnode *np = VTONFS(vp);
2123 struct uio *uio = ap->a_uio;
2124 int tresid, error = 0;
2125 struct vattr vattr;
2126
2127 if (vp->v_type != VDIR)
2128 return(EPERM);
2129
2130 /*
2131 * First, check for hit on the EOF offset cache
2132 */
2133 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2134 (np->n_flag & NMODIFIED) == 0) {
2135 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2136 mtx_lock(&np->n_mtx);
2137 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2138 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2139 mtx_unlock(&np->n_mtx);
2140 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2141 return (0);
2142 } else
2143 mtx_unlock(&np->n_mtx);
2144 }
2145 }
2146
2147 /*
2148 * Call ncl_bioread() to do the real work.
2149 */
2150 tresid = uio->uio_resid;
2151 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2152
2153 if (!error && uio->uio_resid == tresid)
2154 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2155 return (error);
2156 }
2157
2158 /*
2159 * Readdir rpc call.
2160 * Called from below the buffer cache by ncl_doio().
2161 */
2162 int
2163 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2164 struct thread *td)
2165 {
2166 struct nfsvattr nfsva;
2167 nfsuint64 *cookiep, cookie;
2168 struct nfsnode *dnp = VTONFS(vp);
2169 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2170 int error = 0, eof, attrflag;
2171
2172 KASSERT(uiop->uio_iovcnt == 1 &&
2173 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2174 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2175 ("nfs readdirrpc bad uio"));
2176
2177 /*
2178 * If there is no cookie, assume directory was stale.
2179 */
2180 ncl_dircookie_lock(dnp);
2181 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2182 if (cookiep) {
2183 cookie = *cookiep;
2184 ncl_dircookie_unlock(dnp);
2185 } else {
2186 ncl_dircookie_unlock(dnp);
2187 return (NFSERR_BAD_COOKIE);
2188 }
2189
2190 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2191 (void)ncl_fsinfo(nmp, vp, cred, td);
2192
2193 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2194 &attrflag, &eof, NULL);
2195 if (attrflag)
2196 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2197
2198 if (!error) {
2199 /*
2200 * We are now either at the end of the directory or have filled
2201 * the block.
2202 */
2203 if (eof)
2204 dnp->n_direofoffset = uiop->uio_offset;
2205 else {
2206 if (uiop->uio_resid > 0)
2207 ncl_printf("EEK! readdirrpc resid > 0\n");
2208 ncl_dircookie_lock(dnp);
2209 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2210 *cookiep = cookie;
2211 ncl_dircookie_unlock(dnp);
2212 }
2213 } else if (NFS_ISV4(vp)) {
2214 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2215 }
2216 return (error);
2217 }
2218
2219 /*
2220 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2221 */
2222 int
2223 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2224 struct thread *td)
2225 {
2226 struct nfsvattr nfsva;
2227 nfsuint64 *cookiep, cookie;
2228 struct nfsnode *dnp = VTONFS(vp);
2229 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2230 int error = 0, attrflag, eof;
2231
2232 KASSERT(uiop->uio_iovcnt == 1 &&
2233 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2234 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2235 ("nfs readdirplusrpc bad uio"));
2236
2237 /*
2238 * If there is no cookie, assume directory was stale.
2239 */
2240 ncl_dircookie_lock(dnp);
2241 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2242 if (cookiep) {
2243 cookie = *cookiep;
2244 ncl_dircookie_unlock(dnp);
2245 } else {
2246 ncl_dircookie_unlock(dnp);
2247 return (NFSERR_BAD_COOKIE);
2248 }
2249
2250 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2251 (void)ncl_fsinfo(nmp, vp, cred, td);
2252 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2253 &attrflag, &eof, NULL);
2254 if (attrflag)
2255 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2256
2257 if (!error) {
2258 /*
2259 * We are now either at end of the directory or have filled the
2260 * the block.
2261 */
2262 if (eof)
2263 dnp->n_direofoffset = uiop->uio_offset;
2264 else {
2265 if (uiop->uio_resid > 0)
2266 ncl_printf("EEK! readdirplusrpc resid > 0\n");
2267 ncl_dircookie_lock(dnp);
2268 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2269 *cookiep = cookie;
2270 ncl_dircookie_unlock(dnp);
2271 }
2272 } else if (NFS_ISV4(vp)) {
2273 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2274 }
2275 return (error);
2276 }
2277
2278 /*
2279 * Silly rename. To make the NFS filesystem that is stateless look a little
2280 * more like the "ufs" a remove of an active vnode is translated to a rename
2281 * to a funny looking filename that is removed by nfs_inactive on the
2282 * nfsnode. There is the potential for another process on a different client
2283 * to create the same funny name between the nfs_lookitup() fails and the
2284 * nfs_rename() completes, but...
2285 */
2286 static int
2287 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2288 {
2289 struct sillyrename *sp;
2290 struct nfsnode *np;
2291 int error;
2292 short pid;
2293 unsigned int lticks;
2294
2295 cache_purge(dvp);
2296 np = VTONFS(vp);
2297 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2298 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2299 M_NEWNFSREQ, M_WAITOK);
2300 sp->s_cred = crhold(cnp->cn_cred);
2301 sp->s_dvp = dvp;
2302 VREF(dvp);
2303
2304 /*
2305 * Fudge together a funny name.
2306 * Changing the format of the funny name to accomodate more
2307 * sillynames per directory.
2308 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2309 * CPU ticks since boot.
2310 */
2311 pid = cnp->cn_thread->td_proc->p_pid;
2312 lticks = (unsigned int)ticks;
2313 for ( ; ; ) {
2314 sp->s_namlen = sprintf(sp->s_name,
2315 ".nfs.%08x.%04x4.4", lticks,
2316 pid);
2317 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2318 cnp->cn_thread, NULL))
2319 break;
2320 lticks++;
2321 }
2322 error = nfs_renameit(dvp, vp, cnp, sp);
2323 if (error)
2324 goto bad;
2325 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2326 cnp->cn_thread, &np);
2327 np->n_sillyrename = sp;
2328 return (0);
2329 bad:
2330 vrele(sp->s_dvp);
2331 crfree(sp->s_cred);
2332 free((caddr_t)sp, M_NEWNFSREQ);
2333 return (error);
2334 }
2335
2336 /*
2337 * Look up a file name and optionally either update the file handle or
2338 * allocate an nfsnode, depending on the value of npp.
2339 * npp == NULL --> just do the lookup
2340 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2341 * handled too
2342 * *npp != NULL --> update the file handle in the vnode
2343 */
2344 static int
2345 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2346 struct thread *td, struct nfsnode **npp)
2347 {
2348 struct vnode *newvp = NULL, *vp;
2349 struct nfsnode *np, *dnp = VTONFS(dvp);
2350 struct nfsfh *nfhp, *onfhp;
2351 struct nfsvattr nfsva, dnfsva;
2352 struct componentname cn;
2353 int error = 0, attrflag, dattrflag;
2354 u_int hash;
2355
2356 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2357 &nfhp, &attrflag, &dattrflag, NULL);
2358 if (dattrflag)
2359 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2360 if (npp && !error) {
2361 if (*npp != NULL) {
2362 np = *npp;
2363 vp = NFSTOV(np);
2364 /*
2365 * For NFSv4, check to see if it is the same name and
2366 * replace the name, if it is different.
2367 */
2368 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2369 (np->n_v4->n4_namelen != len ||
2370 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2371 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2372 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2373 dnp->n_fhp->nfh_len))) {
2374 #ifdef notdef
2375 { char nnn[100]; int nnnl;
2376 nnnl = (len < 100) ? len : 99;
2377 bcopy(name, nnn, nnnl);
2378 nnn[nnnl] = '\0';
2379 printf("replace=%s\n",nnn);
2380 }
2381 #endif
2382 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2383 MALLOC(np->n_v4, struct nfsv4node *,
2384 sizeof (struct nfsv4node) +
2385 dnp->n_fhp->nfh_len + len - 1,
2386 M_NFSV4NODE, M_WAITOK);
2387 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2388 np->n_v4->n4_namelen = len;
2389 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2390 dnp->n_fhp->nfh_len);
2391 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2392 }
2393 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2394 FNV1_32_INIT);
2395 onfhp = np->n_fhp;
2396 /*
2397 * Rehash node for new file handle.
2398 */
2399 vfs_hash_rehash(vp, hash);
2400 np->n_fhp = nfhp;
2401 if (onfhp != NULL)
2402 FREE((caddr_t)onfhp, M_NFSFH);
2403 newvp = NFSTOV(np);
2404 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2405 FREE((caddr_t)nfhp, M_NFSFH);
2406 VREF(dvp);
2407 newvp = dvp;
2408 } else {
2409 cn.cn_nameptr = name;
2410 cn.cn_namelen = len;
2411 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2412 &np, NULL, LK_EXCLUSIVE);
2413 if (error)
2414 return (error);
2415 newvp = NFSTOV(np);
2416 }
2417 if (!attrflag && *npp == NULL) {
2418 if (newvp == dvp)
2419 vrele(newvp);
2420 else
2421 vput(newvp);
2422 return (ENOENT);
2423 }
2424 if (attrflag)
2425 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2426 0, 1);
2427 }
2428 if (npp && *npp == NULL) {
2429 if (error) {
2430 if (newvp) {
2431 if (newvp == dvp)
2432 vrele(newvp);
2433 else
2434 vput(newvp);
2435 }
2436 } else
2437 *npp = np;
2438 }
2439 if (error && NFS_ISV4(dvp))
2440 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2441 return (error);
2442 }
2443
2444 /*
2445 * Nfs Version 3 and 4 commit rpc
2446 */
2447 int
2448 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2449 struct thread *td)
2450 {
2451 struct nfsvattr nfsva;
2452 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2453 int error, attrflag;
2454 u_char verf[NFSX_VERF];
2455
2456 mtx_lock(&nmp->nm_mtx);
2457 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2458 mtx_unlock(&nmp->nm_mtx);
2459 return (0);
2460 }
2461 mtx_unlock(&nmp->nm_mtx);
2462 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2463 &attrflag, NULL);
2464 if (!error) {
2465 mtx_lock(&nmp->nm_mtx);
2466 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2467 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2468 error = NFSERR_STALEWRITEVERF;
2469 }
2470 mtx_unlock(&nmp->nm_mtx);
2471 if (!error && attrflag)
2472 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2473 0, 1);
2474 } else if (NFS_ISV4(vp)) {
2475 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2476 }
2477 return (error);
2478 }
2479
2480 /*
2481 * Strategy routine.
2482 * For async requests when nfsiod(s) are running, queue the request by
2483 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2484 * request.
2485 */
2486 static int
2487 nfs_strategy(struct vop_strategy_args *ap)
2488 {
2489 struct buf *bp = ap->a_bp;
2490 struct ucred *cr;
2491
2492 KASSERT(!(bp->b_flags & B_DONE),
2493 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2494 BUF_ASSERT_HELD(bp);
2495
2496 if (bp->b_iocmd == BIO_READ)
2497 cr = bp->b_rcred;
2498 else
2499 cr = bp->b_wcred;
2500
2501 /*
2502 * If the op is asynchronous and an i/o daemon is waiting
2503 * queue the request, wake it up and wait for completion
2504 * otherwise just do it ourselves.
2505 */
2506 if ((bp->b_flags & B_ASYNC) == 0 ||
2507 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2508 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2509 return (0);
2510 }
2511
2512 /*
2513 * fsync vnode op. Just call ncl_flush() with commit == 1.
2514 */
2515 /* ARGSUSED */
2516 static int
2517 nfs_fsync(struct vop_fsync_args *ap)
2518 {
2519
2520 if (ap->a_vp->v_type != VREG) {
2521 /*
2522 * For NFS, metadata is changed synchronously on the server,
2523 * so there is nothing to flush. Also, ncl_flush() clears
2524 * the NMODIFIED flag and that shouldn't be done here for
2525 * directories.
2526 */
2527 return (0);
2528 }
2529 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2530 }
2531
2532 /*
2533 * Flush all the blocks associated with a vnode.
2534 * Walk through the buffer pool and push any dirty pages
2535 * associated with the vnode.
2536 * If the called_from_renewthread argument is TRUE, it has been called
2537 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2538 * waiting for a buffer write to complete.
2539 */
2540 int
2541 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2542 int commit, int called_from_renewthread)
2543 {
2544 struct nfsnode *np = VTONFS(vp);
2545 struct buf *bp;
2546 int i;
2547 struct buf *nbp;
2548 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2549 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2550 int passone = 1, trycnt = 0;
2551 u_quad_t off, endoff, toff;
2552 struct ucred* wcred = NULL;
2553 struct buf **bvec = NULL;
2554 struct bufobj *bo;
2555 #ifndef NFS_COMMITBVECSIZ
2556 #define NFS_COMMITBVECSIZ 20
2557 #endif
2558 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2559 int bvecsize = 0, bveccount;
2560
2561 if (called_from_renewthread != 0)
2562 slptimeo = hz;
2563 if (nmp->nm_flag & NFSMNT_INT)
2564 slpflag = NFS_PCATCH;
2565 if (!commit)
2566 passone = 0;
2567 bo = &vp->v_bufobj;
2568 /*
2569 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2570 * server, but has not been committed to stable storage on the server
2571 * yet. On the first pass, the byte range is worked out and the commit
2572 * rpc is done. On the second pass, ncl_writebp() is called to do the
2573 * job.
2574 */
2575 again:
2576 off = (u_quad_t)-1;
2577 endoff = 0;
2578 bvecpos = 0;
2579 if (NFS_ISV34(vp) && commit) {
2580 if (bvec != NULL && bvec != bvec_on_stack)
2581 free(bvec, M_TEMP);
2582 /*
2583 * Count up how many buffers waiting for a commit.
2584 */
2585 bveccount = 0;
2586 BO_LOCK(bo);
2587 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2588 if (!BUF_ISLOCKED(bp) &&
2589 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2590 == (B_DELWRI | B_NEEDCOMMIT))
2591 bveccount++;
2592 }
2593 /*
2594 * Allocate space to remember the list of bufs to commit. It is
2595 * important to use M_NOWAIT here to avoid a race with nfs_write.
2596 * If we can't get memory (for whatever reason), we will end up
2597 * committing the buffers one-by-one in the loop below.
2598 */
2599 if (bveccount > NFS_COMMITBVECSIZ) {
2600 /*
2601 * Release the vnode interlock to avoid a lock
2602 * order reversal.
2603 */
2604 BO_UNLOCK(bo);
2605 bvec = (struct buf **)
2606 malloc(bveccount * sizeof(struct buf *),
2607 M_TEMP, M_NOWAIT);
2608 BO_LOCK(bo);
2609 if (bvec == NULL) {
2610 bvec = bvec_on_stack;
2611 bvecsize = NFS_COMMITBVECSIZ;
2612 } else
2613 bvecsize = bveccount;
2614 } else {
2615 bvec = bvec_on_stack;
2616 bvecsize = NFS_COMMITBVECSIZ;
2617 }
2618 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2619 if (bvecpos >= bvecsize)
2620 break;
2621 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2622 nbp = TAILQ_NEXT(bp, b_bobufs);
2623 continue;
2624 }
2625 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2626 (B_DELWRI | B_NEEDCOMMIT)) {
2627 BUF_UNLOCK(bp);
2628 nbp = TAILQ_NEXT(bp, b_bobufs);
2629 continue;
2630 }
2631 BO_UNLOCK(bo);
2632 bremfree(bp);
2633 /*
2634 * Work out if all buffers are using the same cred
2635 * so we can deal with them all with one commit.
2636 *
2637 * NOTE: we are not clearing B_DONE here, so we have
2638 * to do it later on in this routine if we intend to
2639 * initiate I/O on the bp.
2640 *
2641 * Note: to avoid loopback deadlocks, we do not
2642 * assign b_runningbufspace.
2643 */
2644 if (wcred == NULL)
2645 wcred = bp->b_wcred;
2646 else if (wcred != bp->b_wcred)
2647 wcred = NOCRED;
2648 vfs_busy_pages(bp, 1);
2649
2650 BO_LOCK(bo);
2651 /*
2652 * bp is protected by being locked, but nbp is not
2653 * and vfs_busy_pages() may sleep. We have to
2654 * recalculate nbp.
2655 */
2656 nbp = TAILQ_NEXT(bp, b_bobufs);
2657
2658 /*
2659 * A list of these buffers is kept so that the
2660 * second loop knows which buffers have actually
2661 * been committed. This is necessary, since there
2662 * may be a race between the commit rpc and new
2663 * uncommitted writes on the file.
2664 */
2665 bvec[bvecpos++] = bp;
2666 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2667 bp->b_dirtyoff;
2668 if (toff < off)
2669 off = toff;
2670 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2671 if (toff > endoff)
2672 endoff = toff;
2673 }
2674 BO_UNLOCK(bo);
2675 }
2676 if (bvecpos > 0) {
2677 /*
2678 * Commit data on the server, as required.
2679 * If all bufs are using the same wcred, then use that with
2680 * one call for all of them, otherwise commit each one
2681 * separately.
2682 */
2683 if (wcred != NOCRED)
2684 retv = ncl_commit(vp, off, (int)(endoff - off),
2685 wcred, td);
2686 else {
2687 retv = 0;
2688 for (i = 0; i < bvecpos; i++) {
2689 off_t off, size;
2690 bp = bvec[i];
2691 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2692 bp->b_dirtyoff;
2693 size = (u_quad_t)(bp->b_dirtyend
2694 - bp->b_dirtyoff);
2695 retv = ncl_commit(vp, off, (int)size,
2696 bp->b_wcred, td);
2697 if (retv) break;
2698 }
2699 }
2700
2701 if (retv == NFSERR_STALEWRITEVERF)
2702 ncl_clearcommit(vp->v_mount);
2703
2704 /*
2705 * Now, either mark the blocks I/O done or mark the
2706 * blocks dirty, depending on whether the commit
2707 * succeeded.
2708 */
2709 for (i = 0; i < bvecpos; i++) {
2710 bp = bvec[i];
2711 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2712 if (retv) {
2713 /*
2714 * Error, leave B_DELWRI intact
2715 */
2716 vfs_unbusy_pages(bp);
2717 brelse(bp);
2718 } else {
2719 /*
2720 * Success, remove B_DELWRI ( bundirty() ).
2721 *
2722 * b_dirtyoff/b_dirtyend seem to be NFS
2723 * specific. We should probably move that
2724 * into bundirty(). XXX
2725 */
2726 bufobj_wref(bo);
2727 bp->b_flags |= B_ASYNC;
2728 bundirty(bp);
2729 bp->b_flags &= ~B_DONE;
2730 bp->b_ioflags &= ~BIO_ERROR;
2731 bp->b_dirtyoff = bp->b_dirtyend = 0;
2732 bufdone(bp);
2733 }
2734 }
2735 }
2736
2737 /*
2738 * Start/do any write(s) that are required.
2739 */
2740 loop:
2741 BO_LOCK(bo);
2742 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2743 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2744 if (waitfor != MNT_WAIT || passone)
2745 continue;
2746
2747 error = BUF_TIMELOCK(bp,
2748 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2749 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2750 if (error == 0) {
2751 BUF_UNLOCK(bp);
2752 goto loop;
2753 }
2754 if (error == ENOLCK) {
2755 error = 0;
2756 goto loop;
2757 }
2758 if (called_from_renewthread != 0) {
2759 /*
2760 * Return EIO so the flush will be retried
2761 * later.
2762 */
2763 error = EIO;
2764 goto done;
2765 }
2766 if (newnfs_sigintr(nmp, td)) {
2767 error = EINTR;
2768 goto done;
2769 }
2770 if (slpflag & PCATCH) {
2771 slpflag = 0;
2772 slptimeo = 2 * hz;
2773 }
2774 goto loop;
2775 }
2776 if ((bp->b_flags & B_DELWRI) == 0)
2777 panic("nfs_fsync: not dirty");
2778 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2779 BUF_UNLOCK(bp);
2780 continue;
2781 }
2782 BO_UNLOCK(bo);
2783 bremfree(bp);
2784 if (passone || !commit)
2785 bp->b_flags |= B_ASYNC;
2786 else
2787 bp->b_flags |= B_ASYNC;
2788 bwrite(bp);
2789 if (newnfs_sigintr(nmp, td)) {
2790 error = EINTR;
2791 goto done;
2792 }
2793 goto loop;
2794 }
2795 if (passone) {
2796 passone = 0;
2797 BO_UNLOCK(bo);
2798 goto again;
2799 }
2800 if (waitfor == MNT_WAIT) {
2801 while (bo->bo_numoutput) {
2802 error = bufobj_wwait(bo, slpflag, slptimeo);
2803 if (error) {
2804 BO_UNLOCK(bo);
2805 if (called_from_renewthread != 0) {
2806 /*
2807 * Return EIO so that the flush will be
2808 * retried later.
2809 */
2810 error = EIO;
2811 goto done;
2812 }
2813 error = newnfs_sigintr(nmp, td);
2814 if (error)
2815 goto done;
2816 if (slpflag & PCATCH) {
2817 slpflag = 0;
2818 slptimeo = 2 * hz;
2819 }
2820 BO_LOCK(bo);
2821 }
2822 }
2823 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2824 BO_UNLOCK(bo);
2825 goto loop;
2826 }
2827 /*
2828 * Wait for all the async IO requests to drain
2829 */
2830 BO_UNLOCK(bo);
2831 mtx_lock(&np->n_mtx);
2832 while (np->n_directio_asyncwr > 0) {
2833 np->n_flag |= NFSYNCWAIT;
2834 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2835 &np->n_mtx, slpflag | (PRIBIO + 1),
2836 "nfsfsync", 0);
2837 if (error) {
2838 if (newnfs_sigintr(nmp, td)) {
2839 mtx_unlock(&np->n_mtx);
2840 error = EINTR;
2841 goto done;
2842 }
2843 }
2844 }
2845 mtx_unlock(&np->n_mtx);
2846 } else
2847 BO_UNLOCK(bo);
2848 mtx_lock(&np->n_mtx);
2849 if (np->n_flag & NWRITEERR) {
2850 error = np->n_error;
2851 np->n_flag &= ~NWRITEERR;
2852 }
2853 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2854 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2855 np->n_flag &= ~NMODIFIED;
2856 mtx_unlock(&np->n_mtx);
2857 done:
2858 if (bvec != NULL && bvec != bvec_on_stack)
2859 free(bvec, M_TEMP);
2860 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2861 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2862 np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2863 /* try, try again... */
2864 passone = 1;
2865 wcred = NULL;
2866 bvec = NULL;
2867 bvecsize = 0;
2868 printf("try%d\n", trycnt);
2869 goto again;
2870 }
2871 return (error);
2872 }
2873
2874 /*
2875 * NFS advisory byte-level locks.
2876 */
2877 static int
2878 nfs_advlock(struct vop_advlock_args *ap)
2879 {
2880 struct vnode *vp = ap->a_vp;
2881 struct ucred *cred;
2882 struct nfsnode *np = VTONFS(ap->a_vp);
2883 struct proc *p = (struct proc *)ap->a_id;
2884 struct thread *td = curthread; /* XXX */
2885 struct vattr va;
2886 int ret, error = EOPNOTSUPP;
2887 u_quad_t size;
2888
2889 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
2890 if (vp->v_type != VREG)
2891 return (EINVAL);
2892 if ((ap->a_flags & F_POSIX) != 0)
2893 cred = p->p_ucred;
2894 else
2895 cred = td->td_ucred;
2896 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2897 if (vp->v_iflag & VI_DOOMED) {
2898 NFSVOPUNLOCK(vp, 0);
2899 return (EBADF);
2900 }
2901
2902 /*
2903 * If this is unlocking a write locked region, flush and
2904 * commit them before unlocking. This is required by
2905 * RFC3530 Sec. 9.3.2.
2906 */
2907 if (ap->a_op == F_UNLCK &&
2908 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
2909 ap->a_flags))
2910 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
2911
2912 /*
2913 * Loop around doing the lock op, while a blocking lock
2914 * must wait for the lock op to succeed.
2915 */
2916 do {
2917 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
2918 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
2919 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2920 ap->a_op == F_SETLK) {
2921 NFSVOPUNLOCK(vp, 0);
2922 error = nfs_catnap(PZERO | PCATCH, ret,
2923 "ncladvl");
2924 if (error)
2925 return (EINTR);
2926 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2927 if (vp->v_iflag & VI_DOOMED) {
2928 NFSVOPUNLOCK(vp, 0);
2929 return (EBADF);
2930 }
2931 }
2932 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2933 ap->a_op == F_SETLK);
2934 if (ret == NFSERR_DENIED) {
2935 NFSVOPUNLOCK(vp, 0);
2936 return (EAGAIN);
2937 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
2938 NFSVOPUNLOCK(vp, 0);
2939 return (ret);
2940 } else if (ret != 0) {
2941 NFSVOPUNLOCK(vp, 0);
2942 return (EACCES);
2943 }
2944
2945 /*
2946 * Now, if we just got a lock, invalidate data in the buffer
2947 * cache, as required, so that the coherency conforms with
2948 * RFC3530 Sec. 9.3.2.
2949 */
2950 if (ap->a_op == F_SETLK) {
2951 if ((np->n_flag & NMODIFIED) == 0) {
2952 np->n_attrstamp = 0;
2953 ret = VOP_GETATTR(vp, &va, cred);
2954 }
2955 if ((np->n_flag & NMODIFIED) || ret ||
2956 np->n_change != va.va_filerev) {
2957 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
2958 np->n_attrstamp = 0;
2959 ret = VOP_GETATTR(vp, &va, cred);
2960 if (!ret) {
2961 np->n_mtime = va.va_mtime;
2962 np->n_change = va.va_filerev;
2963 }
2964 }
2965 }
2966 NFSVOPUNLOCK(vp, 0);
2967 return (0);
2968 } else if (!NFS_ISV4(vp)) {
2969 error = NFSVOPLOCK(vp, LK_SHARED);
2970 if (error)
2971 return (error);
2972 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2973 size = VTONFS(vp)->n_size;
2974 NFSVOPUNLOCK(vp, 0);
2975 error = lf_advlock(ap, &(vp->v_lockf), size);
2976 } else {
2977 if (nfs_advlock_p != NULL)
2978 error = nfs_advlock_p(ap);
2979 else {
2980 NFSVOPUNLOCK(vp, 0);
2981 error = ENOLCK;
2982 }
2983 }
2984 }
2985 return (error);
2986 }
2987
2988 /*
2989 * NFS advisory byte-level locks.
2990 */
2991 static int
2992 nfs_advlockasync(struct vop_advlockasync_args *ap)
2993 {
2994 struct vnode *vp = ap->a_vp;
2995 u_quad_t size;
2996 int error;
2997
2998 if (NFS_ISV4(vp))
2999 return (EOPNOTSUPP);
3000 error = NFSVOPLOCK(vp, LK_SHARED);
3001 if (error)
3002 return (error);
3003 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3004 size = VTONFS(vp)->n_size;
3005 NFSVOPUNLOCK(vp, 0);
3006 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3007 } else {
3008 NFSVOPUNLOCK(vp, 0);
3009 error = EOPNOTSUPP;
3010 }
3011 return (error);
3012 }
3013
3014 /*
3015 * Print out the contents of an nfsnode.
3016 */
3017 static int
3018 nfs_print(struct vop_print_args *ap)
3019 {
3020 struct vnode *vp = ap->a_vp;
3021 struct nfsnode *np = VTONFS(vp);
3022
3023 ncl_printf("\tfileid %ld fsid 0x%x",
3024 np->n_vattr.na_fileid, np->n_vattr.na_fsid);
3025 if (vp->v_type == VFIFO)
3026 fifo_printinfo(vp);
3027 printf("\n");
3028 return (0);
3029 }
3030
3031 /*
3032 * This is the "real" nfs::bwrite(struct buf*).
3033 * We set B_CACHE if this is a VMIO buffer.
3034 */
3035 int
3036 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3037 {
3038 int s;
3039 int oldflags = bp->b_flags;
3040 #if 0
3041 int retv = 1;
3042 off_t off;
3043 #endif
3044
3045 BUF_ASSERT_HELD(bp);
3046
3047 if (bp->b_flags & B_INVAL) {
3048 brelse(bp);
3049 return(0);
3050 }
3051
3052 bp->b_flags |= B_CACHE;
3053
3054 /*
3055 * Undirty the bp. We will redirty it later if the I/O fails.
3056 */
3057
3058 s = splbio();
3059 bundirty(bp);
3060 bp->b_flags &= ~B_DONE;
3061 bp->b_ioflags &= ~BIO_ERROR;
3062 bp->b_iocmd = BIO_WRITE;
3063
3064 bufobj_wref(bp->b_bufobj);
3065 curthread->td_ru.ru_oublock++;
3066 splx(s);
3067
3068 /*
3069 * Note: to avoid loopback deadlocks, we do not
3070 * assign b_runningbufspace.
3071 */
3072 vfs_busy_pages(bp, 1);
3073
3074 BUF_KERNPROC(bp);
3075 bp->b_iooffset = dbtob(bp->b_blkno);
3076 bstrategy(bp);
3077
3078 if( (oldflags & B_ASYNC) == 0) {
3079 int rtval = bufwait(bp);
3080
3081 if (oldflags & B_DELWRI) {
3082 s = splbio();
3083 reassignbuf(bp);
3084 splx(s);
3085 }
3086 brelse(bp);
3087 return (rtval);
3088 }
3089
3090 return (0);
3091 }
3092
3093 /*
3094 * nfs special file access vnode op.
3095 * Essentially just get vattr and then imitate iaccess() since the device is
3096 * local to the client.
3097 */
3098 static int
3099 nfsspec_access(struct vop_access_args *ap)
3100 {
3101 struct vattr *vap;
3102 struct ucred *cred = ap->a_cred;
3103 struct vnode *vp = ap->a_vp;
3104 accmode_t accmode = ap->a_accmode;
3105 struct vattr vattr;
3106 int error;
3107
3108 /*
3109 * Disallow write attempts on filesystems mounted read-only;
3110 * unless the file is a socket, fifo, or a block or character
3111 * device resident on the filesystem.
3112 */
3113 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3114 switch (vp->v_type) {
3115 case VREG:
3116 case VDIR:
3117 case VLNK:
3118 return (EROFS);
3119 default:
3120 break;
3121 }
3122 }
3123 vap = &vattr;
3124 error = VOP_GETATTR(vp, vap, cred);
3125 if (error)
3126 goto out;
3127 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3128 accmode, cred, NULL);
3129 out:
3130 return error;
3131 }
3132
3133 /*
3134 * Read wrapper for fifos.
3135 */
3136 static int
3137 nfsfifo_read(struct vop_read_args *ap)
3138 {
3139 struct nfsnode *np = VTONFS(ap->a_vp);
3140 int error;
3141
3142 /*
3143 * Set access flag.
3144 */
3145 mtx_lock(&np->n_mtx);
3146 np->n_flag |= NACC;
3147 getnanotime(&np->n_atim);
3148 mtx_unlock(&np->n_mtx);
3149 error = fifo_specops.vop_read(ap);
3150 return error;
3151 }
3152
3153 /*
3154 * Write wrapper for fifos.
3155 */
3156 static int
3157 nfsfifo_write(struct vop_write_args *ap)
3158 {
3159 struct nfsnode *np = VTONFS(ap->a_vp);
3160
3161 /*
3162 * Set update flag.
3163 */
3164 mtx_lock(&np->n_mtx);
3165 np->n_flag |= NUPD;
3166 getnanotime(&np->n_mtim);
3167 mtx_unlock(&np->n_mtx);
3168 return(fifo_specops.vop_write(ap));
3169 }
3170
3171 /*
3172 * Close wrapper for fifos.
3173 *
3174 * Update the times on the nfsnode then do fifo close.
3175 */
3176 static int
3177 nfsfifo_close(struct vop_close_args *ap)
3178 {
3179 struct vnode *vp = ap->a_vp;
3180 struct nfsnode *np = VTONFS(vp);
3181 struct vattr vattr;
3182 struct timespec ts;
3183
3184 mtx_lock(&np->n_mtx);
3185 if (np->n_flag & (NACC | NUPD)) {
3186 getnanotime(&ts);
3187 if (np->n_flag & NACC)
3188 np->n_atim = ts;
3189 if (np->n_flag & NUPD)
3190 np->n_mtim = ts;
3191 np->n_flag |= NCHG;
3192 if (vrefcnt(vp) == 1 &&
3193 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3194 VATTR_NULL(&vattr);
3195 if (np->n_flag & NACC)
3196 vattr.va_atime = np->n_atim;
3197 if (np->n_flag & NUPD)
3198 vattr.va_mtime = np->n_mtim;
3199 mtx_unlock(&np->n_mtx);
3200 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3201 goto out;
3202 }
3203 }
3204 mtx_unlock(&np->n_mtx);
3205 out:
3206 return (fifo_specops.vop_close(ap));
3207 }
3208
3209 /*
3210 * Just call ncl_writebp() with the force argument set to 1.
3211 *
3212 * NOTE: B_DONE may or may not be set in a_bp on call.
3213 */
3214 static int
3215 nfs_bwrite(struct buf *bp)
3216 {
3217
3218 return (ncl_writebp(bp, 1, curthread));
3219 }
3220
3221 struct buf_ops buf_ops_newnfs = {
3222 .bop_name = "buf_ops_nfs",
3223 .bop_write = nfs_bwrite,
3224 .bop_strategy = bufstrategy,
3225 .bop_sync = bufsync,
3226 .bop_bdflush = bufbdflush,
3227 };
3228
3229 /*
3230 * Cloned from vop_stdlock(), and then the ugly hack added.
3231 */
3232 static int
3233 nfs_lock1(struct vop_lock1_args *ap)
3234 {
3235 struct vnode *vp = ap->a_vp;
3236 int error = 0;
3237
3238 /*
3239 * Since vfs_hash_get() calls vget() and it will no longer work
3240 * for FreeBSD8 with flags == 0, I can only think of this horrible
3241 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3242 * and then handle it here. All I want for this case is a v_usecount
3243 * on the vnode to use for recovery, while another thread might
3244 * hold a lock on the vnode. I have the other threads blocked, so
3245 * there isn't any race problem.
3246 */
3247 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3248 if ((ap->a_flags & LK_INTERLOCK) == 0)
3249 panic("ncllock1");
3250 if ((vp->v_iflag & VI_DOOMED))
3251 error = ENOENT;
3252 VI_UNLOCK(vp);
3253 return (error);
3254 }
3255 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3256 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3257 ap->a_line));
3258 }
3259
3260 static int
3261 nfs_getacl(struct vop_getacl_args *ap)
3262 {
3263 int error;
3264
3265 if (ap->a_type != ACL_TYPE_NFS4)
3266 return (EOPNOTSUPP);
3267 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3268 NULL);
3269 if (error > NFSERR_STALE) {
3270 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3271 error = EPERM;
3272 }
3273 return (error);
3274 }
3275
3276 static int
3277 nfs_setacl(struct vop_setacl_args *ap)
3278 {
3279 int error;
3280
3281 if (ap->a_type != ACL_TYPE_NFS4)
3282 return (EOPNOTSUPP);
3283 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3284 NULL);
3285 if (error > NFSERR_STALE) {
3286 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3287 error = EPERM;
3288 }
3289 return (error);
3290 }
3291
3292 /*
3293 * Return POSIX pathconf information applicable to nfs filesystems.
3294 */
3295 static int
3296 nfs_pathconf(struct vop_pathconf_args *ap)
3297 {
3298 struct nfsv3_pathconf pc;
3299 struct nfsvattr nfsva;
3300 struct vnode *vp = ap->a_vp;
3301 struct thread *td = curthread;
3302 int attrflag, error;
3303
3304 if (NFS_ISV4(vp) || (NFS_ISV3(vp) && (ap->a_name == _PC_LINK_MAX ||
3305 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3306 ap->a_name == _PC_NO_TRUNC))) {
3307 /*
3308 * Since only the above 4 a_names are returned by the NFSv3
3309 * Pathconf RPC, there is no point in doing it for others.
3310 */
3311 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3312 &attrflag, NULL);
3313 if (attrflag != 0)
3314 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3315 1);
3316 if (error != 0)
3317 return (error);
3318 } else {
3319 /*
3320 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3321 * just fake them.
3322 */
3323 pc.pc_linkmax = LINK_MAX;
3324 pc.pc_namemax = NFS_MAXNAMLEN;
3325 pc.pc_notrunc = 1;
3326 pc.pc_chownrestricted = 1;
3327 pc.pc_caseinsensitive = 0;
3328 pc.pc_casepreserving = 1;
3329 error = 0;
3330 }
3331 switch (ap->a_name) {
3332 case _PC_LINK_MAX:
3333 *ap->a_retval = pc.pc_linkmax;
3334 break;
3335 case _PC_NAME_MAX:
3336 *ap->a_retval = pc.pc_namemax;
3337 break;
3338 case _PC_PATH_MAX:
3339 *ap->a_retval = PATH_MAX;
3340 break;
3341 case _PC_PIPE_BUF:
3342 *ap->a_retval = PIPE_BUF;
3343 break;
3344 case _PC_CHOWN_RESTRICTED:
3345 *ap->a_retval = pc.pc_chownrestricted;
3346 break;
3347 case _PC_NO_TRUNC:
3348 *ap->a_retval = pc.pc_notrunc;
3349 break;
3350 case _PC_ACL_EXTENDED:
3351 *ap->a_retval = 0;
3352 break;
3353 case _PC_ACL_NFS4:
3354 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3355 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3356 *ap->a_retval = 1;
3357 else
3358 *ap->a_retval = 0;
3359 break;
3360 case _PC_ACL_PATH_MAX:
3361 if (NFS_ISV4(vp))
3362 *ap->a_retval = ACL_MAX_ENTRIES;
3363 else
3364 *ap->a_retval = 3;
3365 break;
3366 case _PC_MAC_PRESENT:
3367 *ap->a_retval = 0;
3368 break;
3369 case _PC_ASYNC_IO:
3370 /* _PC_ASYNC_IO should have been handled by upper layers. */
3371 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3372 error = EINVAL;
3373 break;
3374 case _PC_PRIO_IO:
3375 *ap->a_retval = 0;
3376 break;
3377 case _PC_SYNC_IO:
3378 *ap->a_retval = 0;
3379 break;
3380 case _PC_ALLOC_SIZE_MIN:
3381 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3382 break;
3383 case _PC_FILESIZEBITS:
3384 if (NFS_ISV34(vp))
3385 *ap->a_retval = 64;
3386 else
3387 *ap->a_retval = 32;
3388 break;
3389 case _PC_REC_INCR_XFER_SIZE:
3390 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3391 break;
3392 case _PC_REC_MAX_XFER_SIZE:
3393 *ap->a_retval = -1; /* means ``unlimited'' */
3394 break;
3395 case _PC_REC_MIN_XFER_SIZE:
3396 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3397 break;
3398 case _PC_REC_XFER_ALIGN:
3399 *ap->a_retval = PAGE_SIZE;
3400 break;
3401 case _PC_SYMLINK_MAX:
3402 *ap->a_retval = NFS_MAXPATHLEN;
3403 break;
3404
3405 default:
3406 error = EINVAL;
3407 break;
3408 }
3409 return (error);
3410 }
3411
Cache object: 1f70073d0c9c628c2454e85082c5250c
|