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>
| 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: head/sys/fs/nfsclient/nfs_clvnops.c 203303 2010-01-31 19:12:24Z rmacklem $");
| 36__FBSDID("$FreeBSD: head/sys/fs/nfsclient/nfs_clvnops.c 207082 2010-04-22 23:51:01Z rmacklem $");
|
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
71#include <fs/nfs/nfsport.h>
72#include <fs/nfsclient/nfsnode.h>
73#include <fs/nfsclient/nfsmount.h>
74#include <fs/nfsclient/nfs.h>
75#include <fs/nfsclient/nfs_lock.h>
76
77#include <net/if.h>
78#include <netinet/in.h>
79#include <netinet/in_var.h>
80
81/* Defs */
82#define TRUE 1
83#define FALSE 0
84
85extern struct nfsstats newnfsstats;
86MALLOC_DECLARE(M_NEWNFSREQ);
87vop_advlock_t *ncl_advlock_p = ncl_dolock;
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
98static vop_read_t nfsfifo_read;
99static vop_write_t nfsfifo_write;
100static vop_close_t nfsfifo_close;
101static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
102 struct thread *);
103static vop_lookup_t nfs_lookup;
104static vop_create_t nfs_create;
105static vop_mknod_t nfs_mknod;
106static vop_open_t nfs_open;
107static vop_close_t nfs_close;
108static vop_access_t nfs_access;
109static vop_getattr_t nfs_getattr;
110static vop_setattr_t nfs_setattr;
111static vop_read_t nfs_read;
112static vop_fsync_t nfs_fsync;
113static vop_remove_t nfs_remove;
114static vop_link_t nfs_link;
115static vop_rename_t nfs_rename;
116static vop_mkdir_t nfs_mkdir;
117static vop_rmdir_t nfs_rmdir;
118static vop_symlink_t nfs_symlink;
119static vop_readdir_t nfs_readdir;
120static vop_strategy_t nfs_strategy;
121static vop_lock1_t nfs_lock1;
122static int nfs_lookitup(struct vnode *, char *, int,
123 struct ucred *, struct thread *, struct nfsnode **);
124static int nfs_sillyrename(struct vnode *, struct vnode *,
125 struct componentname *);
126static vop_access_t nfsspec_access;
127static vop_readlink_t nfs_readlink;
128static vop_print_t nfs_print;
129static vop_advlock_t nfs_advlock;
130static vop_advlockasync_t nfs_advlockasync;
131static vop_getacl_t nfs_getacl;
132static vop_setacl_t nfs_setacl;
133
134/*
135 * Global vfs data structures for nfs
136 */
137struct vop_vector newnfs_vnodeops = {
138 .vop_default = &default_vnodeops,
139 .vop_access = nfs_access,
140 .vop_advlock = nfs_advlock,
141 .vop_advlockasync = nfs_advlockasync,
142 .vop_close = nfs_close,
143 .vop_create = nfs_create,
144 .vop_fsync = nfs_fsync,
145 .vop_getattr = nfs_getattr,
146 .vop_getpages = ncl_getpages,
147 .vop_putpages = ncl_putpages,
148 .vop_inactive = ncl_inactive,
149 .vop_link = nfs_link,
150 .vop_lock1 = nfs_lock1,
151 .vop_lookup = nfs_lookup,
152 .vop_mkdir = nfs_mkdir,
153 .vop_mknod = nfs_mknod,
154 .vop_open = nfs_open,
155 .vop_print = nfs_print,
156 .vop_read = nfs_read,
157 .vop_readdir = nfs_readdir,
158 .vop_readlink = nfs_readlink,
159 .vop_reclaim = ncl_reclaim,
160 .vop_remove = nfs_remove,
161 .vop_rename = nfs_rename,
162 .vop_rmdir = nfs_rmdir,
163 .vop_setattr = nfs_setattr,
164 .vop_strategy = nfs_strategy,
165 .vop_symlink = nfs_symlink,
166 .vop_write = ncl_write,
167 .vop_getacl = nfs_getacl,
168 .vop_setacl = nfs_setacl,
169};
170
171struct vop_vector newnfs_fifoops = {
172 .vop_default = &fifo_specops,
173 .vop_access = nfsspec_access,
174 .vop_close = nfsfifo_close,
175 .vop_fsync = nfs_fsync,
176 .vop_getattr = nfs_getattr,
177 .vop_inactive = ncl_inactive,
178 .vop_print = nfs_print,
179 .vop_read = nfsfifo_read,
180 .vop_reclaim = ncl_reclaim,
181 .vop_setattr = nfs_setattr,
182 .vop_write = nfsfifo_write,
183};
184
185static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
186 struct componentname *cnp, struct vattr *vap);
187static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
188 int namelen, struct ucred *cred, struct thread *td);
189static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
190 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
191 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
192static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
193 struct componentname *scnp, struct sillyrename *sp);
194
195/*
196 * Global variables
197 */
198#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
199
200SYSCTL_DECL(_vfs_newnfs);
201
202static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
203SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
204 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
205
206static int nfs_prime_access_cache = 0;
207SYSCTL_INT(_vfs_newnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
208 &nfs_prime_access_cache, 0,
209 "Prime NFS ACCESS cache when fetching attributes");
210
211static int newnfs_commit_on_close = 0;
212SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
213 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
214
215static int nfs_clean_pages_on_close = 1;
216SYSCTL_INT(_vfs_newnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
217 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
218
219int newnfs_directio_enable = 0;
220SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_enable, CTLFLAG_RW,
221 &newnfs_directio_enable, 0, "Enable NFS directio");
222
223/*
224 * This sysctl allows other processes to mmap a file that has been opened
225 * O_DIRECT by a process. In general, having processes mmap the file while
226 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
227 * this by default to prevent DoS attacks - to prevent a malicious user from
228 * opening up files O_DIRECT preventing other users from mmap'ing these
229 * files. "Protected" environments where stricter consistency guarantees are
230 * required can disable this knob. The process that opened the file O_DIRECT
231 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
232 * meaningful.
233 */
234int newnfs_directio_allow_mmap = 1;
235SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_allow_mmap, CTLFLAG_RW,
236 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
237
238#if 0
239SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
240 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
241
242SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
243 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
244#endif
245
246#define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
247 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
248 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
249
250/*
251 * SMP Locking Note :
252 * The list of locks after the description of the lock is the ordering
253 * of other locks acquired with the lock held.
254 * np->n_mtx : Protects the fields in the nfsnode.
255 VM Object Lock
256 VI_MTX (acquired indirectly)
257 * nmp->nm_mtx : Protects the fields in the nfsmount.
258 rep->r_mtx
259 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
260 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
261 nmp->nm_mtx
262 rep->r_mtx
263 * rep->r_mtx : Protects the fields in an nfsreq.
264 */
265
266static int
267nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
268 struct ucred *cred, u_int32_t *retmode)
269{
270 int error = 0, attrflag, i, lrupos;
271 u_int32_t rmode;
272 struct nfsnode *np = VTONFS(vp);
273 struct nfsvattr nfsva;
274
275 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
276 &rmode, NULL);
277 if (attrflag)
278 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
279 if (!error) {
280 lrupos = 0;
281 mtx_lock(&np->n_mtx);
282 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
283 if (np->n_accesscache[i].uid == cred->cr_uid) {
284 np->n_accesscache[i].mode = rmode;
285 np->n_accesscache[i].stamp = time_second;
286 break;
287 }
288 if (i > 0 && np->n_accesscache[i].stamp <
289 np->n_accesscache[lrupos].stamp)
290 lrupos = i;
291 }
292 if (i == NFS_ACCESSCACHESIZE) {
293 np->n_accesscache[lrupos].uid = cred->cr_uid;
294 np->n_accesscache[lrupos].mode = rmode;
295 np->n_accesscache[lrupos].stamp = time_second;
296 }
297 mtx_unlock(&np->n_mtx);
298 if (retmode != NULL)
299 *retmode = rmode;
300 } else if (NFS_ISV4(vp)) {
301 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
302 }
303 return (error);
304}
305
306/*
307 * nfs access vnode op.
308 * For nfs version 2, just return ok. File accesses may fail later.
309 * For nfs version 3, use the access rpc to check accessibility. If file modes
310 * are changed on the server, accesses might still fail later.
311 */
312static int
313nfs_access(struct vop_access_args *ap)
314{
315 struct vnode *vp = ap->a_vp;
316 int error = 0, i, gotahit;
317 u_int32_t mode, wmode, rmode;
318 int v34 = NFS_ISV34(vp);
319 struct nfsnode *np = VTONFS(vp);
320
321 /*
322 * Disallow write attempts on filesystems mounted read-only;
323 * unless the file is a socket, fifo, or a block or character
324 * device resident on the filesystem.
325 */
326 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
327 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
328 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
329 switch (vp->v_type) {
330 case VREG:
331 case VDIR:
332 case VLNK:
333 return (EROFS);
334 default:
335 break;
336 }
337 }
338 /*
339 * For nfs v3 or v4, check to see if we have done this recently, and if
340 * so return our cached result instead of making an ACCESS call.
341 * If not, do an access rpc, otherwise you are stuck emulating
342 * ufs_access() locally using the vattr. This may not be correct,
343 * since the server may apply other access criteria such as
344 * client uid-->server uid mapping that we do not know about.
345 */
346 if (v34) {
347 if (ap->a_accmode & VREAD)
348 mode = NFSACCESS_READ;
349 else
350 mode = 0;
351 if (vp->v_type != VDIR) {
352 if (ap->a_accmode & VWRITE)
353 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
354 if (ap->a_accmode & VAPPEND)
355 mode |= NFSACCESS_EXTEND;
356 if (ap->a_accmode & VEXEC)
357 mode |= NFSACCESS_EXECUTE;
358 if (ap->a_accmode & VDELETE)
359 mode |= NFSACCESS_DELETE;
360 } else {
361 if (ap->a_accmode & VWRITE)
362 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
363 if (ap->a_accmode & VAPPEND)
364 mode |= NFSACCESS_EXTEND;
365 if (ap->a_accmode & VEXEC)
366 mode |= NFSACCESS_LOOKUP;
367 if (ap->a_accmode & VDELETE)
368 mode |= NFSACCESS_DELETE;
369 if (ap->a_accmode & VDELETE_CHILD)
370 mode |= NFSACCESS_MODIFY;
371 }
372 /* XXX safety belt, only make blanket request if caching */
373 if (nfsaccess_cache_timeout > 0) {
374 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
375 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
376 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
377 } else {
378 wmode = mode;
379 }
380
381 /*
382 * Does our cached result allow us to give a definite yes to
383 * this request?
384 */
385 gotahit = 0;
386 mtx_lock(&np->n_mtx);
387 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
388 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
389 if (time_second < (np->n_accesscache[i].stamp
390 + nfsaccess_cache_timeout) &&
391 (np->n_accesscache[i].mode & mode) == mode) {
392 NFSINCRGLOBAL(newnfsstats.accesscache_hits);
393 gotahit = 1;
394 }
395 break;
396 }
397 }
398 mtx_unlock(&np->n_mtx);
399 if (gotahit == 0) {
400 /*
401 * Either a no, or a don't know. Go to the wire.
402 */
403 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
404 error = nfs34_access_otw(vp, wmode, ap->a_td,
405 ap->a_cred, &rmode);
406 if (!error &&
407 (rmode & mode) != mode)
408 error = EACCES;
409 }
410 return (error);
411 } else {
412 if ((error = nfsspec_access(ap)) != 0) {
413 return (error);
414 }
415 /*
416 * Attempt to prevent a mapped root from accessing a file
417 * which it shouldn't. We try to read a byte from the file
418 * if the user is root and the file is not zero length.
419 * After calling nfsspec_access, we should have the correct
420 * file size cached.
421 */
422 mtx_lock(&np->n_mtx);
423 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
424 && VTONFS(vp)->n_size > 0) {
425 struct iovec aiov;
426 struct uio auio;
427 char buf[1];
428
429 mtx_unlock(&np->n_mtx);
430 aiov.iov_base = buf;
431 aiov.iov_len = 1;
432 auio.uio_iov = &aiov;
433 auio.uio_iovcnt = 1;
434 auio.uio_offset = 0;
435 auio.uio_resid = 1;
436 auio.uio_segflg = UIO_SYSSPACE;
437 auio.uio_rw = UIO_READ;
438 auio.uio_td = ap->a_td;
439
440 if (vp->v_type == VREG)
441 error = ncl_readrpc(vp, &auio, ap->a_cred);
442 else if (vp->v_type == VDIR) {
443 char* bp;
444 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
445 aiov.iov_base = bp;
446 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
447 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
448 ap->a_td);
449 free(bp, M_TEMP);
450 } else if (vp->v_type == VLNK)
451 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
452 else
453 error = EACCES;
454 } else
455 mtx_unlock(&np->n_mtx);
456 return (error);
457 }
458}
459
460
461/*
462 * nfs open vnode op
463 * Check to see if the type is ok
464 * and that deletion is not in progress.
465 * For paged in text files, you will need to flush the page cache
466 * if consistency is lost.
467 */
468/* ARGSUSED */
469static int
470nfs_open(struct vop_open_args *ap)
471{
472 struct vnode *vp = ap->a_vp;
473 struct nfsnode *np = VTONFS(vp);
474 struct vattr vattr;
475 int error;
476 int fmode = ap->a_mode;
477
478 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
479 return (EOPNOTSUPP);
480
481 /*
482 * For NFSv4, we need to do the Open Op before cache validation,
483 * so that we conform to RFC3530 Sec. 9.3.1.
484 */
485 if (NFS_ISV4(vp)) {
486 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
487 if (error) {
488 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
489 (gid_t)0);
490 return (error);
491 }
492 }
493
494 /*
495 * Now, if this Open will be doing reading, re-validate/flush the
496 * cache, so that Close/Open coherency is maintained.
497 */
498 if ((fmode & FREAD) && (!NFS_ISV4(vp) || nfscl_mustflush(vp))) {
499 mtx_lock(&np->n_mtx);
500 if (np->n_flag & NMODIFIED) {
501 mtx_unlock(&np->n_mtx);
502 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
503 if (error == EINTR || error == EIO) {
504 if (NFS_ISV4(vp))
505 (void) nfsrpc_close(vp, 0, ap->a_td);
506 return (error);
507 }
508 np->n_attrstamp = 0;
509 if (vp->v_type == VDIR)
510 np->n_direofoffset = 0;
511 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
512 if (error) {
513 if (NFS_ISV4(vp))
514 (void) nfsrpc_close(vp, 0, ap->a_td);
515 return (error);
516 }
517 mtx_lock(&np->n_mtx);
518 np->n_mtime = vattr.va_mtime;
519 if (NFS_ISV4(vp))
520 np->n_change = vattr.va_filerev;
521 mtx_unlock(&np->n_mtx);
522 } else {
523 struct thread *td = curthread;
524
525 if (np->n_ac_ts_syscalls != td->td_syscalls ||
526 np->n_ac_ts_tid != td->td_tid ||
527 td->td_proc == NULL ||
528 np->n_ac_ts_pid != td->td_proc->p_pid) {
529 np->n_attrstamp = 0;
530 }
531 mtx_unlock(&np->n_mtx);
532 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
533 if (error) {
534 if (NFS_ISV4(vp))
535 (void) nfsrpc_close(vp, 0, ap->a_td);
536 return (error);
537 }
538 mtx_lock(&np->n_mtx);
539 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
540 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
541 if (vp->v_type == VDIR)
542 np->n_direofoffset = 0;
543 mtx_unlock(&np->n_mtx);
544 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
545 if (error == EINTR || error == EIO) {
546 if (NFS_ISV4(vp))
547 (void) nfsrpc_close(vp, 0,
548 ap->a_td);
549 return (error);
550 }
551 mtx_lock(&np->n_mtx);
552 np->n_mtime = vattr.va_mtime;
553 if (NFS_ISV4(vp))
554 np->n_change = vattr.va_filerev;
555 }
556 mtx_unlock(&np->n_mtx);
557 }
558 }
559
560 /*
561 * If the object has >= 1 O_DIRECT active opens, we disable caching.
562 */
563 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
564 if (np->n_directio_opens == 0) {
565 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
566 if (error) {
567 if (NFS_ISV4(vp))
568 (void) nfsrpc_close(vp, 0, ap->a_td);
569 return (error);
570 }
571 mtx_lock(&np->n_mtx);
572 np->n_flag |= NNONCACHE;
573 } else {
574 mtx_lock(&np->n_mtx);
575 }
576 np->n_directio_opens++;
577 mtx_unlock(&np->n_mtx);
578 }
579 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
580 return (0);
581}
582
583/*
584 * nfs close vnode op
585 * What an NFS client should do upon close after writing is a debatable issue.
586 * Most NFS clients push delayed writes to the server upon close, basically for
587 * two reasons:
588 * 1 - So that any write errors may be reported back to the client process
589 * doing the close system call. By far the two most likely errors are
590 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
591 * 2 - To put a worst case upper bound on cache inconsistency between
592 * multiple clients for the file.
593 * There is also a consistency problem for Version 2 of the protocol w.r.t.
594 * not being able to tell if other clients are writing a file concurrently,
595 * since there is no way of knowing if the changed modify time in the reply
596 * is only due to the write for this client.
597 * (NFS Version 3 provides weak cache consistency data in the reply that
598 * should be sufficient to detect and handle this case.)
599 *
600 * The current code does the following:
601 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
602 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
603 * or commit them (this satisfies 1 and 2 except for the
604 * case where the server crashes after this close but
605 * before the commit RPC, which is felt to be "good
606 * enough". Changing the last argument to ncl_flush() to
607 * a 1 would force a commit operation, if it is felt a
608 * commit is necessary now.
609 * for NFS Version 4 - flush the dirty buffers and commit them, if
610 * nfscl_mustflush() says this is necessary.
611 * It is necessary if there is no write delegation held,
612 * in order to satisfy open/close coherency.
613 * If the file isn't cached on local stable storage,
614 * it may be necessary in order to detect "out of space"
615 * errors from the server, if the write delegation
616 * issued by the server doesn't allow the file to grow.
617 */
618/* ARGSUSED */
619static int
620nfs_close(struct vop_close_args *ap)
621{
622 struct vnode *vp = ap->a_vp;
623 struct nfsnode *np = VTONFS(vp);
624 struct nfsvattr nfsva;
625 struct ucred *cred;
626 int error = 0, ret, localcred = 0;
627 int fmode = ap->a_fflag;
628
629 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
630 return (0);
631 /*
632 * During shutdown, a_cred isn't valid, so just use root.
633 */
634 if (ap->a_cred == NOCRED) {
635 cred = newnfs_getcred();
636 localcred = 1;
637 } else {
638 cred = ap->a_cred;
639 }
640 if (vp->v_type == VREG) {
641 /*
642 * Examine and clean dirty pages, regardless of NMODIFIED.
643 * This closes a major hole in close-to-open consistency.
644 * We want to push out all dirty pages (and buffers) on
645 * close, regardless of whether they were dirtied by
646 * mmap'ed writes or via write().
647 */
648 if (nfs_clean_pages_on_close && vp->v_object) {
649 VM_OBJECT_LOCK(vp->v_object);
650 vm_object_page_clean(vp->v_object, 0, 0, 0);
651 VM_OBJECT_UNLOCK(vp->v_object);
652 }
653 mtx_lock(&np->n_mtx);
654 if (np->n_flag & NMODIFIED) {
655 mtx_unlock(&np->n_mtx);
656 if (NFS_ISV3(vp)) {
657 /*
658 * Under NFSv3 we have dirty buffers to dispose of. We
659 * must flush them to the NFS server. We have the option
660 * of waiting all the way through the commit rpc or just
661 * waiting for the initial write. The default is to only
662 * wait through the initial write so the data is in the
663 * server's cache, which is roughly similar to the state
664 * a standard disk subsystem leaves the file in on close().
665 *
666 * We cannot clear the NMODIFIED bit in np->n_flag due to
667 * potential races with other processes, and certainly
668 * cannot clear it if we don't commit.
669 * These races occur when there is no longer the old
670 * traditional vnode locking implemented for Vnode Ops.
671 */
672 int cm = newnfs_commit_on_close ? 1 : 0;
| 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
71#include <fs/nfs/nfsport.h>
72#include <fs/nfsclient/nfsnode.h>
73#include <fs/nfsclient/nfsmount.h>
74#include <fs/nfsclient/nfs.h>
75#include <fs/nfsclient/nfs_lock.h>
76
77#include <net/if.h>
78#include <netinet/in.h>
79#include <netinet/in_var.h>
80
81/* Defs */
82#define TRUE 1
83#define FALSE 0
84
85extern struct nfsstats newnfsstats;
86MALLOC_DECLARE(M_NEWNFSREQ);
87vop_advlock_t *ncl_advlock_p = ncl_dolock;
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
98static vop_read_t nfsfifo_read;
99static vop_write_t nfsfifo_write;
100static vop_close_t nfsfifo_close;
101static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
102 struct thread *);
103static vop_lookup_t nfs_lookup;
104static vop_create_t nfs_create;
105static vop_mknod_t nfs_mknod;
106static vop_open_t nfs_open;
107static vop_close_t nfs_close;
108static vop_access_t nfs_access;
109static vop_getattr_t nfs_getattr;
110static vop_setattr_t nfs_setattr;
111static vop_read_t nfs_read;
112static vop_fsync_t nfs_fsync;
113static vop_remove_t nfs_remove;
114static vop_link_t nfs_link;
115static vop_rename_t nfs_rename;
116static vop_mkdir_t nfs_mkdir;
117static vop_rmdir_t nfs_rmdir;
118static vop_symlink_t nfs_symlink;
119static vop_readdir_t nfs_readdir;
120static vop_strategy_t nfs_strategy;
121static vop_lock1_t nfs_lock1;
122static int nfs_lookitup(struct vnode *, char *, int,
123 struct ucred *, struct thread *, struct nfsnode **);
124static int nfs_sillyrename(struct vnode *, struct vnode *,
125 struct componentname *);
126static vop_access_t nfsspec_access;
127static vop_readlink_t nfs_readlink;
128static vop_print_t nfs_print;
129static vop_advlock_t nfs_advlock;
130static vop_advlockasync_t nfs_advlockasync;
131static vop_getacl_t nfs_getacl;
132static vop_setacl_t nfs_setacl;
133
134/*
135 * Global vfs data structures for nfs
136 */
137struct vop_vector newnfs_vnodeops = {
138 .vop_default = &default_vnodeops,
139 .vop_access = nfs_access,
140 .vop_advlock = nfs_advlock,
141 .vop_advlockasync = nfs_advlockasync,
142 .vop_close = nfs_close,
143 .vop_create = nfs_create,
144 .vop_fsync = nfs_fsync,
145 .vop_getattr = nfs_getattr,
146 .vop_getpages = ncl_getpages,
147 .vop_putpages = ncl_putpages,
148 .vop_inactive = ncl_inactive,
149 .vop_link = nfs_link,
150 .vop_lock1 = nfs_lock1,
151 .vop_lookup = nfs_lookup,
152 .vop_mkdir = nfs_mkdir,
153 .vop_mknod = nfs_mknod,
154 .vop_open = nfs_open,
155 .vop_print = nfs_print,
156 .vop_read = nfs_read,
157 .vop_readdir = nfs_readdir,
158 .vop_readlink = nfs_readlink,
159 .vop_reclaim = ncl_reclaim,
160 .vop_remove = nfs_remove,
161 .vop_rename = nfs_rename,
162 .vop_rmdir = nfs_rmdir,
163 .vop_setattr = nfs_setattr,
164 .vop_strategy = nfs_strategy,
165 .vop_symlink = nfs_symlink,
166 .vop_write = ncl_write,
167 .vop_getacl = nfs_getacl,
168 .vop_setacl = nfs_setacl,
169};
170
171struct vop_vector newnfs_fifoops = {
172 .vop_default = &fifo_specops,
173 .vop_access = nfsspec_access,
174 .vop_close = nfsfifo_close,
175 .vop_fsync = nfs_fsync,
176 .vop_getattr = nfs_getattr,
177 .vop_inactive = ncl_inactive,
178 .vop_print = nfs_print,
179 .vop_read = nfsfifo_read,
180 .vop_reclaim = ncl_reclaim,
181 .vop_setattr = nfs_setattr,
182 .vop_write = nfsfifo_write,
183};
184
185static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
186 struct componentname *cnp, struct vattr *vap);
187static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
188 int namelen, struct ucred *cred, struct thread *td);
189static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
190 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
191 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
192static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
193 struct componentname *scnp, struct sillyrename *sp);
194
195/*
196 * Global variables
197 */
198#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
199
200SYSCTL_DECL(_vfs_newnfs);
201
202static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
203SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
204 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
205
206static int nfs_prime_access_cache = 0;
207SYSCTL_INT(_vfs_newnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
208 &nfs_prime_access_cache, 0,
209 "Prime NFS ACCESS cache when fetching attributes");
210
211static int newnfs_commit_on_close = 0;
212SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
213 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
214
215static int nfs_clean_pages_on_close = 1;
216SYSCTL_INT(_vfs_newnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
217 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
218
219int newnfs_directio_enable = 0;
220SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_enable, CTLFLAG_RW,
221 &newnfs_directio_enable, 0, "Enable NFS directio");
222
223/*
224 * This sysctl allows other processes to mmap a file that has been opened
225 * O_DIRECT by a process. In general, having processes mmap the file while
226 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
227 * this by default to prevent DoS attacks - to prevent a malicious user from
228 * opening up files O_DIRECT preventing other users from mmap'ing these
229 * files. "Protected" environments where stricter consistency guarantees are
230 * required can disable this knob. The process that opened the file O_DIRECT
231 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
232 * meaningful.
233 */
234int newnfs_directio_allow_mmap = 1;
235SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_allow_mmap, CTLFLAG_RW,
236 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
237
238#if 0
239SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
240 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
241
242SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
243 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
244#endif
245
246#define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
247 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
248 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
249
250/*
251 * SMP Locking Note :
252 * The list of locks after the description of the lock is the ordering
253 * of other locks acquired with the lock held.
254 * np->n_mtx : Protects the fields in the nfsnode.
255 VM Object Lock
256 VI_MTX (acquired indirectly)
257 * nmp->nm_mtx : Protects the fields in the nfsmount.
258 rep->r_mtx
259 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
260 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
261 nmp->nm_mtx
262 rep->r_mtx
263 * rep->r_mtx : Protects the fields in an nfsreq.
264 */
265
266static int
267nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
268 struct ucred *cred, u_int32_t *retmode)
269{
270 int error = 0, attrflag, i, lrupos;
271 u_int32_t rmode;
272 struct nfsnode *np = VTONFS(vp);
273 struct nfsvattr nfsva;
274
275 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
276 &rmode, NULL);
277 if (attrflag)
278 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
279 if (!error) {
280 lrupos = 0;
281 mtx_lock(&np->n_mtx);
282 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
283 if (np->n_accesscache[i].uid == cred->cr_uid) {
284 np->n_accesscache[i].mode = rmode;
285 np->n_accesscache[i].stamp = time_second;
286 break;
287 }
288 if (i > 0 && np->n_accesscache[i].stamp <
289 np->n_accesscache[lrupos].stamp)
290 lrupos = i;
291 }
292 if (i == NFS_ACCESSCACHESIZE) {
293 np->n_accesscache[lrupos].uid = cred->cr_uid;
294 np->n_accesscache[lrupos].mode = rmode;
295 np->n_accesscache[lrupos].stamp = time_second;
296 }
297 mtx_unlock(&np->n_mtx);
298 if (retmode != NULL)
299 *retmode = rmode;
300 } else if (NFS_ISV4(vp)) {
301 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
302 }
303 return (error);
304}
305
306/*
307 * nfs access vnode op.
308 * For nfs version 2, just return ok. File accesses may fail later.
309 * For nfs version 3, use the access rpc to check accessibility. If file modes
310 * are changed on the server, accesses might still fail later.
311 */
312static int
313nfs_access(struct vop_access_args *ap)
314{
315 struct vnode *vp = ap->a_vp;
316 int error = 0, i, gotahit;
317 u_int32_t mode, wmode, rmode;
318 int v34 = NFS_ISV34(vp);
319 struct nfsnode *np = VTONFS(vp);
320
321 /*
322 * Disallow write attempts on filesystems mounted read-only;
323 * unless the file is a socket, fifo, or a block or character
324 * device resident on the filesystem.
325 */
326 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
327 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
328 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
329 switch (vp->v_type) {
330 case VREG:
331 case VDIR:
332 case VLNK:
333 return (EROFS);
334 default:
335 break;
336 }
337 }
338 /*
339 * For nfs v3 or v4, check to see if we have done this recently, and if
340 * so return our cached result instead of making an ACCESS call.
341 * If not, do an access rpc, otherwise you are stuck emulating
342 * ufs_access() locally using the vattr. This may not be correct,
343 * since the server may apply other access criteria such as
344 * client uid-->server uid mapping that we do not know about.
345 */
346 if (v34) {
347 if (ap->a_accmode & VREAD)
348 mode = NFSACCESS_READ;
349 else
350 mode = 0;
351 if (vp->v_type != VDIR) {
352 if (ap->a_accmode & VWRITE)
353 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
354 if (ap->a_accmode & VAPPEND)
355 mode |= NFSACCESS_EXTEND;
356 if (ap->a_accmode & VEXEC)
357 mode |= NFSACCESS_EXECUTE;
358 if (ap->a_accmode & VDELETE)
359 mode |= NFSACCESS_DELETE;
360 } else {
361 if (ap->a_accmode & VWRITE)
362 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
363 if (ap->a_accmode & VAPPEND)
364 mode |= NFSACCESS_EXTEND;
365 if (ap->a_accmode & VEXEC)
366 mode |= NFSACCESS_LOOKUP;
367 if (ap->a_accmode & VDELETE)
368 mode |= NFSACCESS_DELETE;
369 if (ap->a_accmode & VDELETE_CHILD)
370 mode |= NFSACCESS_MODIFY;
371 }
372 /* XXX safety belt, only make blanket request if caching */
373 if (nfsaccess_cache_timeout > 0) {
374 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
375 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
376 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
377 } else {
378 wmode = mode;
379 }
380
381 /*
382 * Does our cached result allow us to give a definite yes to
383 * this request?
384 */
385 gotahit = 0;
386 mtx_lock(&np->n_mtx);
387 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
388 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
389 if (time_second < (np->n_accesscache[i].stamp
390 + nfsaccess_cache_timeout) &&
391 (np->n_accesscache[i].mode & mode) == mode) {
392 NFSINCRGLOBAL(newnfsstats.accesscache_hits);
393 gotahit = 1;
394 }
395 break;
396 }
397 }
398 mtx_unlock(&np->n_mtx);
399 if (gotahit == 0) {
400 /*
401 * Either a no, or a don't know. Go to the wire.
402 */
403 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
404 error = nfs34_access_otw(vp, wmode, ap->a_td,
405 ap->a_cred, &rmode);
406 if (!error &&
407 (rmode & mode) != mode)
408 error = EACCES;
409 }
410 return (error);
411 } else {
412 if ((error = nfsspec_access(ap)) != 0) {
413 return (error);
414 }
415 /*
416 * Attempt to prevent a mapped root from accessing a file
417 * which it shouldn't. We try to read a byte from the file
418 * if the user is root and the file is not zero length.
419 * After calling nfsspec_access, we should have the correct
420 * file size cached.
421 */
422 mtx_lock(&np->n_mtx);
423 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
424 && VTONFS(vp)->n_size > 0) {
425 struct iovec aiov;
426 struct uio auio;
427 char buf[1];
428
429 mtx_unlock(&np->n_mtx);
430 aiov.iov_base = buf;
431 aiov.iov_len = 1;
432 auio.uio_iov = &aiov;
433 auio.uio_iovcnt = 1;
434 auio.uio_offset = 0;
435 auio.uio_resid = 1;
436 auio.uio_segflg = UIO_SYSSPACE;
437 auio.uio_rw = UIO_READ;
438 auio.uio_td = ap->a_td;
439
440 if (vp->v_type == VREG)
441 error = ncl_readrpc(vp, &auio, ap->a_cred);
442 else if (vp->v_type == VDIR) {
443 char* bp;
444 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
445 aiov.iov_base = bp;
446 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
447 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
448 ap->a_td);
449 free(bp, M_TEMP);
450 } else if (vp->v_type == VLNK)
451 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
452 else
453 error = EACCES;
454 } else
455 mtx_unlock(&np->n_mtx);
456 return (error);
457 }
458}
459
460
461/*
462 * nfs open vnode op
463 * Check to see if the type is ok
464 * and that deletion is not in progress.
465 * For paged in text files, you will need to flush the page cache
466 * if consistency is lost.
467 */
468/* ARGSUSED */
469static int
470nfs_open(struct vop_open_args *ap)
471{
472 struct vnode *vp = ap->a_vp;
473 struct nfsnode *np = VTONFS(vp);
474 struct vattr vattr;
475 int error;
476 int fmode = ap->a_mode;
477
478 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
479 return (EOPNOTSUPP);
480
481 /*
482 * For NFSv4, we need to do the Open Op before cache validation,
483 * so that we conform to RFC3530 Sec. 9.3.1.
484 */
485 if (NFS_ISV4(vp)) {
486 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
487 if (error) {
488 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
489 (gid_t)0);
490 return (error);
491 }
492 }
493
494 /*
495 * Now, if this Open will be doing reading, re-validate/flush the
496 * cache, so that Close/Open coherency is maintained.
497 */
498 if ((fmode & FREAD) && (!NFS_ISV4(vp) || nfscl_mustflush(vp))) {
499 mtx_lock(&np->n_mtx);
500 if (np->n_flag & NMODIFIED) {
501 mtx_unlock(&np->n_mtx);
502 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
503 if (error == EINTR || error == EIO) {
504 if (NFS_ISV4(vp))
505 (void) nfsrpc_close(vp, 0, ap->a_td);
506 return (error);
507 }
508 np->n_attrstamp = 0;
509 if (vp->v_type == VDIR)
510 np->n_direofoffset = 0;
511 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
512 if (error) {
513 if (NFS_ISV4(vp))
514 (void) nfsrpc_close(vp, 0, ap->a_td);
515 return (error);
516 }
517 mtx_lock(&np->n_mtx);
518 np->n_mtime = vattr.va_mtime;
519 if (NFS_ISV4(vp))
520 np->n_change = vattr.va_filerev;
521 mtx_unlock(&np->n_mtx);
522 } else {
523 struct thread *td = curthread;
524
525 if (np->n_ac_ts_syscalls != td->td_syscalls ||
526 np->n_ac_ts_tid != td->td_tid ||
527 td->td_proc == NULL ||
528 np->n_ac_ts_pid != td->td_proc->p_pid) {
529 np->n_attrstamp = 0;
530 }
531 mtx_unlock(&np->n_mtx);
532 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
533 if (error) {
534 if (NFS_ISV4(vp))
535 (void) nfsrpc_close(vp, 0, ap->a_td);
536 return (error);
537 }
538 mtx_lock(&np->n_mtx);
539 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
540 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
541 if (vp->v_type == VDIR)
542 np->n_direofoffset = 0;
543 mtx_unlock(&np->n_mtx);
544 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
545 if (error == EINTR || error == EIO) {
546 if (NFS_ISV4(vp))
547 (void) nfsrpc_close(vp, 0,
548 ap->a_td);
549 return (error);
550 }
551 mtx_lock(&np->n_mtx);
552 np->n_mtime = vattr.va_mtime;
553 if (NFS_ISV4(vp))
554 np->n_change = vattr.va_filerev;
555 }
556 mtx_unlock(&np->n_mtx);
557 }
558 }
559
560 /*
561 * If the object has >= 1 O_DIRECT active opens, we disable caching.
562 */
563 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
564 if (np->n_directio_opens == 0) {
565 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
566 if (error) {
567 if (NFS_ISV4(vp))
568 (void) nfsrpc_close(vp, 0, ap->a_td);
569 return (error);
570 }
571 mtx_lock(&np->n_mtx);
572 np->n_flag |= NNONCACHE;
573 } else {
574 mtx_lock(&np->n_mtx);
575 }
576 np->n_directio_opens++;
577 mtx_unlock(&np->n_mtx);
578 }
579 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
580 return (0);
581}
582
583/*
584 * nfs close vnode op
585 * What an NFS client should do upon close after writing is a debatable issue.
586 * Most NFS clients push delayed writes to the server upon close, basically for
587 * two reasons:
588 * 1 - So that any write errors may be reported back to the client process
589 * doing the close system call. By far the two most likely errors are
590 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
591 * 2 - To put a worst case upper bound on cache inconsistency between
592 * multiple clients for the file.
593 * There is also a consistency problem for Version 2 of the protocol w.r.t.
594 * not being able to tell if other clients are writing a file concurrently,
595 * since there is no way of knowing if the changed modify time in the reply
596 * is only due to the write for this client.
597 * (NFS Version 3 provides weak cache consistency data in the reply that
598 * should be sufficient to detect and handle this case.)
599 *
600 * The current code does the following:
601 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
602 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
603 * or commit them (this satisfies 1 and 2 except for the
604 * case where the server crashes after this close but
605 * before the commit RPC, which is felt to be "good
606 * enough". Changing the last argument to ncl_flush() to
607 * a 1 would force a commit operation, if it is felt a
608 * commit is necessary now.
609 * for NFS Version 4 - flush the dirty buffers and commit them, if
610 * nfscl_mustflush() says this is necessary.
611 * It is necessary if there is no write delegation held,
612 * in order to satisfy open/close coherency.
613 * If the file isn't cached on local stable storage,
614 * it may be necessary in order to detect "out of space"
615 * errors from the server, if the write delegation
616 * issued by the server doesn't allow the file to grow.
617 */
618/* ARGSUSED */
619static int
620nfs_close(struct vop_close_args *ap)
621{
622 struct vnode *vp = ap->a_vp;
623 struct nfsnode *np = VTONFS(vp);
624 struct nfsvattr nfsva;
625 struct ucred *cred;
626 int error = 0, ret, localcred = 0;
627 int fmode = ap->a_fflag;
628
629 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
630 return (0);
631 /*
632 * During shutdown, a_cred isn't valid, so just use root.
633 */
634 if (ap->a_cred == NOCRED) {
635 cred = newnfs_getcred();
636 localcred = 1;
637 } else {
638 cred = ap->a_cred;
639 }
640 if (vp->v_type == VREG) {
641 /*
642 * Examine and clean dirty pages, regardless of NMODIFIED.
643 * This closes a major hole in close-to-open consistency.
644 * We want to push out all dirty pages (and buffers) on
645 * close, regardless of whether they were dirtied by
646 * mmap'ed writes or via write().
647 */
648 if (nfs_clean_pages_on_close && vp->v_object) {
649 VM_OBJECT_LOCK(vp->v_object);
650 vm_object_page_clean(vp->v_object, 0, 0, 0);
651 VM_OBJECT_UNLOCK(vp->v_object);
652 }
653 mtx_lock(&np->n_mtx);
654 if (np->n_flag & NMODIFIED) {
655 mtx_unlock(&np->n_mtx);
656 if (NFS_ISV3(vp)) {
657 /*
658 * Under NFSv3 we have dirty buffers to dispose of. We
659 * must flush them to the NFS server. We have the option
660 * of waiting all the way through the commit rpc or just
661 * waiting for the initial write. The default is to only
662 * wait through the initial write so the data is in the
663 * server's cache, which is roughly similar to the state
664 * a standard disk subsystem leaves the file in on close().
665 *
666 * We cannot clear the NMODIFIED bit in np->n_flag due to
667 * potential races with other processes, and certainly
668 * cannot clear it if we don't commit.
669 * These races occur when there is no longer the old
670 * traditional vnode locking implemented for Vnode Ops.
671 */
672 int cm = newnfs_commit_on_close ? 1 : 0;
|
673 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm);
| 673 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
|
674 /* np->n_flag &= ~NMODIFIED; */
675 } else if (NFS_ISV4(vp)) {
676 if (nfscl_mustflush(vp)) {
677 int cm = newnfs_commit_on_close ? 1 : 0;
678 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
| 674 /* np->n_flag &= ~NMODIFIED; */
675 } else if (NFS_ISV4(vp)) {
676 if (nfscl_mustflush(vp)) {
677 int cm = newnfs_commit_on_close ? 1 : 0;
678 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
|
679 cm);
| 679 cm, 0);
|
680 /*
681 * as above w.r.t races when clearing
682 * NMODIFIED.
683 * np->n_flag &= ~NMODIFIED;
684 */
685 }
686 } else
687 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
688 mtx_lock(&np->n_mtx);
689 }
690 /*
691 * Invalidate the attribute cache in all cases.
692 * An open is going to fetch fresh attrs any way, other procs
693 * on this node that have file open will be forced to do an
694 * otw attr fetch, but this is safe.
695 * --> A user found that their RPC count dropped by 20% when
696 * this was commented out and I can't see any requirement
697 * for it, so I've disabled it when negative lookups are
698 * enabled. (What does this have to do with negative lookup
699 * caching? Well nothing, except it was reported by the
700 * same user that needed negative lookup caching and I wanted
701 * there to be a way to disable it to see if it
702 * is the cause of some caching/coherency issue that might
703 * crop up.)
704 */
705 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0)
706 np->n_attrstamp = 0;
707 if (np->n_flag & NWRITEERR) {
708 np->n_flag &= ~NWRITEERR;
709 error = np->n_error;
710 }
711 mtx_unlock(&np->n_mtx);
712 }
713
714 if (NFS_ISV4(vp)) {
715 /*
716 * Get attributes so "change" is up to date.
717 */
718 if (!error) {
719 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
720 NULL);
721 if (!ret) {
722 np->n_change = nfsva.na_filerev;
723 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
724 NULL, 0, 0);
725 }
726 }
727
728 /*
729 * and do the close.
730 */
731 ret = nfsrpc_close(vp, 0, ap->a_td);
732 if (!error && ret)
733 error = ret;
734 if (error)
735 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
736 (gid_t)0);
737 }
738 if (newnfs_directio_enable)
739 KASSERT((np->n_directio_asyncwr == 0),
740 ("nfs_close: dirty unflushed (%d) directio buffers\n",
741 np->n_directio_asyncwr));
742 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
743 mtx_lock(&np->n_mtx);
744 KASSERT((np->n_directio_opens > 0),
745 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
746 np->n_directio_opens--;
747 if (np->n_directio_opens == 0)
748 np->n_flag &= ~NNONCACHE;
749 mtx_unlock(&np->n_mtx);
750 }
751 if (localcred)
752 NFSFREECRED(cred);
753 return (error);
754}
755
756/*
757 * nfs getattr call from vfs.
758 */
759static int
760nfs_getattr(struct vop_getattr_args *ap)
761{
762 struct vnode *vp = ap->a_vp;
763 struct thread *td = curthread; /* XXX */
764 struct nfsnode *np = VTONFS(vp);
765 int error = 0;
766 struct nfsvattr nfsva;
767 struct vattr *vap = ap->a_vap;
768 struct vattr vattr;
769
770 /*
771 * Update local times for special files.
772 */
773 mtx_lock(&np->n_mtx);
774 if (np->n_flag & (NACC | NUPD))
775 np->n_flag |= NCHG;
776 mtx_unlock(&np->n_mtx);
777 /*
778 * First look in the cache.
779 */
780 if (ncl_getattrcache(vp, &vattr) == 0) {
781 vap->va_type = vattr.va_type;
782 vap->va_mode = vattr.va_mode;
783 vap->va_nlink = vattr.va_nlink;
784 vap->va_uid = vattr.va_uid;
785 vap->va_gid = vattr.va_gid;
786 vap->va_fsid = vattr.va_fsid;
787 vap->va_fileid = vattr.va_fileid;
788 vap->va_size = vattr.va_size;
789 vap->va_blocksize = vattr.va_blocksize;
790 vap->va_atime = vattr.va_atime;
791 vap->va_mtime = vattr.va_mtime;
792 vap->va_ctime = vattr.va_ctime;
793 vap->va_gen = vattr.va_gen;
794 vap->va_flags = vattr.va_flags;
795 vap->va_rdev = vattr.va_rdev;
796 vap->va_bytes = vattr.va_bytes;
797 vap->va_filerev = vattr.va_filerev;
798 /*
799 * Get the local modify time for the case of a write
800 * delegation.
801 */
802 nfscl_deleggetmodtime(vp, &vap->va_mtime);
803 return (0);
804 }
805
806 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
807 nfsaccess_cache_timeout > 0) {
808 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
809 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
810 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
811 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
812 return (0);
813 }
814 }
815 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
816 if (!error)
817 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
818 if (!error) {
819 /*
820 * Get the local modify time for the case of a write
821 * delegation.
822 */
823 nfscl_deleggetmodtime(vp, &vap->va_mtime);
824 } else if (NFS_ISV4(vp)) {
825 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
826 }
827 return (error);
828}
829
830/*
831 * nfs setattr call.
832 */
833static int
834nfs_setattr(struct vop_setattr_args *ap)
835{
836 struct vnode *vp = ap->a_vp;
837 struct nfsnode *np = VTONFS(vp);
838 struct thread *td = curthread; /* XXX */
839 struct vattr *vap = ap->a_vap;
840 int error = 0;
841 u_quad_t tsize;
842
843#ifndef nolint
844 tsize = (u_quad_t)0;
845#endif
846
847 /*
848 * Setting of flags and marking of atimes are not supported.
849 */
850 if (vap->va_flags != VNOVAL)
851 return (EOPNOTSUPP);
852
853 /*
854 * Disallow write attempts if the filesystem is mounted read-only.
855 */
856 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
857 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
858 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
859 (vp->v_mount->mnt_flag & MNT_RDONLY))
860 return (EROFS);
861 if (vap->va_size != VNOVAL) {
862 switch (vp->v_type) {
863 case VDIR:
864 return (EISDIR);
865 case VCHR:
866 case VBLK:
867 case VSOCK:
868 case VFIFO:
869 if (vap->va_mtime.tv_sec == VNOVAL &&
870 vap->va_atime.tv_sec == VNOVAL &&
871 vap->va_mode == (mode_t)VNOVAL &&
872 vap->va_uid == (uid_t)VNOVAL &&
873 vap->va_gid == (gid_t)VNOVAL)
874 return (0);
875 vap->va_size = VNOVAL;
876 break;
877 default:
878 /*
879 * Disallow write attempts if the filesystem is
880 * mounted read-only.
881 */
882 if (vp->v_mount->mnt_flag & MNT_RDONLY)
883 return (EROFS);
884 /*
885 * We run vnode_pager_setsize() early (why?),
886 * we must set np->n_size now to avoid vinvalbuf
887 * V_SAVE races that might setsize a lower
888 * value.
889 */
890 mtx_lock(&np->n_mtx);
891 tsize = np->n_size;
892 mtx_unlock(&np->n_mtx);
893 error = ncl_meta_setsize(vp, ap->a_cred, td,
894 vap->va_size);
895 mtx_lock(&np->n_mtx);
896 if (np->n_flag & NMODIFIED) {
897 tsize = np->n_size;
898 mtx_unlock(&np->n_mtx);
899 if (vap->va_size == 0)
900 error = ncl_vinvalbuf(vp, 0, td, 1);
901 else
902 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
903 if (error) {
904 vnode_pager_setsize(vp, tsize);
905 return (error);
906 }
907 /*
908 * Call nfscl_delegmodtime() to set the modify time
909 * locally, as required.
910 */
911 nfscl_delegmodtime(vp);
912 } else
913 mtx_unlock(&np->n_mtx);
914 /*
915 * np->n_size has already been set to vap->va_size
916 * in ncl_meta_setsize(). We must set it again since
917 * nfs_loadattrcache() could be called through
918 * ncl_meta_setsize() and could modify np->n_size.
919 */
920 mtx_lock(&np->n_mtx);
921 np->n_vattr.na_size = np->n_size = vap->va_size;
922 mtx_unlock(&np->n_mtx);
923 };
924 } else {
925 mtx_lock(&np->n_mtx);
926 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
927 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
928 mtx_unlock(&np->n_mtx);
929 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
930 (error == EINTR || error == EIO))
931 return (error);
932 } else
933 mtx_unlock(&np->n_mtx);
934 }
935 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
936 if (error && vap->va_size != VNOVAL) {
937 mtx_lock(&np->n_mtx);
938 np->n_size = np->n_vattr.na_size = tsize;
939 vnode_pager_setsize(vp, tsize);
940 mtx_unlock(&np->n_mtx);
941 }
942 return (error);
943}
944
945/*
946 * Do an nfs setattr rpc.
947 */
948static int
949nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
950 struct thread *td)
951{
952 struct nfsnode *np = VTONFS(vp);
953 int error, ret, attrflag, i;
954 struct nfsvattr nfsva;
955
956 if (NFS_ISV34(vp)) {
957 mtx_lock(&np->n_mtx);
958 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
959 np->n_accesscache[i].stamp = 0;
960 np->n_flag |= NDELEGMOD;
961 mtx_unlock(&np->n_mtx);
962 }
963 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
964 NULL);
965 if (attrflag) {
966 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
967 if (ret && !error)
968 error = ret;
969 }
970 if (error && NFS_ISV4(vp))
971 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
972 return (error);
973}
974
975/*
976 * nfs lookup call, one step at a time...
977 * First look in cache
978 * If not found, unlock the directory nfsnode and do the rpc
979 */
980static int
981nfs_lookup(struct vop_lookup_args *ap)
982{
983 struct componentname *cnp = ap->a_cnp;
984 struct vnode *dvp = ap->a_dvp;
985 struct vnode **vpp = ap->a_vpp;
986 struct mount *mp = dvp->v_mount;
987 int flags = cnp->cn_flags;
988 struct vnode *newvp;
989 struct nfsmount *nmp;
990 struct nfsnode *np;
991 int error = 0, attrflag, dattrflag, ltype;
992 struct thread *td = cnp->cn_thread;
993 struct nfsfh *nfhp;
994 struct nfsvattr dnfsva, nfsva;
995 struct vattr vattr;
996 time_t dmtime;
997
998 *vpp = NULLVP;
999 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1000 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1001 return (EROFS);
1002 if (dvp->v_type != VDIR)
1003 return (ENOTDIR);
1004 nmp = VFSTONFS(mp);
1005 np = VTONFS(dvp);
1006
1007 /* For NFSv4, wait until any remove is done. */
1008 mtx_lock(&np->n_mtx);
1009 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1010 np->n_flag |= NREMOVEWANT;
1011 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1012 }
1013 mtx_unlock(&np->n_mtx);
1014
1015 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1016 return (error);
1017 error = cache_lookup(dvp, vpp, cnp);
1018 if (error > 0 && error != ENOENT)
1019 return (error);
1020 if (error == -1) {
1021 /*
1022 * We only accept a positive hit in the cache if the
1023 * change time of the file matches our cached copy.
1024 * Otherwise, we discard the cache entry and fallback
1025 * to doing a lookup RPC.
1026 */
1027 newvp = *vpp;
1028 if (nfscl_nodeleg(newvp, 0) == 0 ||
1029 (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
1030 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime)) {
1031 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1032 if (cnp->cn_nameiop != LOOKUP &&
1033 (flags & ISLASTCN))
1034 cnp->cn_flags |= SAVENAME;
1035 return (0);
1036 }
1037 cache_purge(newvp);
1038 if (dvp != newvp)
1039 vput(newvp);
1040 else
1041 vrele(newvp);
1042 *vpp = NULLVP;
1043 } else if (error == ENOENT) {
1044 if (dvp->v_iflag & VI_DOOMED)
1045 return (ENOENT);
1046 /*
1047 * We only accept a negative hit in the cache if the
1048 * modification time of the parent directory matches
1049 * our cached copy. Otherwise, we discard all of the
1050 * negative cache entries for this directory. We also
1051 * only trust -ve cache entries for less than
1052 * nm_negative_namecache_timeout seconds.
1053 */
1054 if ((u_int)(ticks - np->n_dmtime_ticks) <
1055 (nmp->nm_negnametimeo * hz) &&
1056 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1057 vattr.va_mtime.tv_sec == np->n_dmtime) {
1058 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1059 return (ENOENT);
1060 }
1061 cache_purge_negative(dvp);
1062 mtx_lock(&np->n_mtx);
1063 np->n_dmtime = 0;
1064 mtx_unlock(&np->n_mtx);
1065 }
1066
1067 /*
1068 * Cache the modification time of the parent directory in case
1069 * the lookup fails and results in adding the first negative
1070 * name cache entry for the directory. Since this is reading
1071 * a single time_t, don't bother with locking. The
1072 * modification time may be a bit stale, but it must be read
1073 * before performing the lookup RPC to prevent a race where
1074 * another lookup updates the timestamp on the directory after
1075 * the lookup RPC has been performed on the server but before
1076 * n_dmtime is set at the end of this function.
1077 */
1078 dmtime = np->n_vattr.na_mtime.tv_sec;
1079 error = 0;
1080 newvp = NULLVP;
1081 NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1082 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1083 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1084 NULL);
1085 if (dattrflag)
1086 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1087 if (error) {
1088 if (newvp != NULLVP) {
1089 vput(newvp);
1090 *vpp = NULLVP;
1091 }
1092
1093 if (error != ENOENT) {
1094 if (NFS_ISV4(dvp))
1095 error = nfscl_maperr(td, error, (uid_t)0,
1096 (gid_t)0);
1097 return (error);
1098 }
1099
1100 /* The requested file was not found. */
1101 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1102 (flags & ISLASTCN)) {
1103 /*
1104 * XXX: UFS does a full VOP_ACCESS(dvp,
1105 * VWRITE) here instead of just checking
1106 * MNT_RDONLY.
1107 */
1108 if (mp->mnt_flag & MNT_RDONLY)
1109 return (EROFS);
1110 cnp->cn_flags |= SAVENAME;
1111 return (EJUSTRETURN);
1112 }
1113
1114 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1115 /*
1116 * Maintain n_dmtime as the modification time
1117 * of the parent directory when the oldest -ve
1118 * name cache entry for this directory was
1119 * added. If a -ve cache entry has already
1120 * been added with a newer modification time
1121 * by a concurrent lookup, then don't bother
1122 * adding a cache entry. The modification
1123 * time of the directory might have changed
1124 * due to the file this lookup failed to find
1125 * being created. In that case a subsequent
1126 * lookup would incorrectly use the entry
1127 * added here instead of doing an extra
1128 * lookup.
1129 */
1130 mtx_lock(&np->n_mtx);
1131 if (np->n_dmtime <= dmtime) {
1132 if (np->n_dmtime == 0) {
1133 np->n_dmtime = dmtime;
1134 np->n_dmtime_ticks = ticks;
1135 }
1136 mtx_unlock(&np->n_mtx);
1137 cache_enter(dvp, NULL, cnp);
1138 } else
1139 mtx_unlock(&np->n_mtx);
1140 }
1141 return (ENOENT);
1142 }
1143
1144 /*
1145 * Handle RENAME case...
1146 */
1147 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1148 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1149 FREE((caddr_t)nfhp, M_NFSFH);
1150 return (EISDIR);
1151 }
1152 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1153 if (error)
1154 return (error);
1155 newvp = NFSTOV(np);
1156 if (attrflag)
1157 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1158 0, 1);
1159 *vpp = newvp;
1160 cnp->cn_flags |= SAVENAME;
1161 return (0);
1162 }
1163
1164 if (flags & ISDOTDOT) {
1165 ltype = VOP_ISLOCKED(dvp);
1166 error = vfs_busy(mp, MBF_NOWAIT);
1167 if (error != 0) {
1168 vfs_ref(mp);
1169 VOP_UNLOCK(dvp, 0);
1170 error = vfs_busy(mp, 0);
1171 vn_lock(dvp, ltype | LK_RETRY);
1172 vfs_rel(mp);
1173 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1174 vfs_unbusy(mp);
1175 error = ENOENT;
1176 }
1177 if (error != 0)
1178 return (error);
1179 }
1180 VOP_UNLOCK(dvp, 0);
1181 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1182 if (error == 0)
1183 newvp = NFSTOV(np);
1184 vfs_unbusy(mp);
1185 if (newvp != dvp)
1186 vn_lock(dvp, ltype | LK_RETRY);
1187 if (dvp->v_iflag & VI_DOOMED) {
1188 if (error == 0) {
1189 if (newvp == dvp)
1190 vrele(newvp);
1191 else
1192 vput(newvp);
1193 }
1194 error = ENOENT;
1195 }
1196 if (error != 0)
1197 return (error);
1198 if (attrflag)
1199 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1200 0, 1);
1201 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1202 FREE((caddr_t)nfhp, M_NFSFH);
1203 VREF(dvp);
1204 newvp = dvp;
1205 if (attrflag)
1206 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1207 0, 1);
1208 } else {
1209 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1210 if (error)
1211 return (error);
1212 newvp = NFSTOV(np);
1213 if (attrflag)
1214 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1215 0, 1);
1216 }
1217 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1218 cnp->cn_flags |= SAVENAME;
1219 if ((cnp->cn_flags & MAKEENTRY) &&
1220 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1221 np->n_ctime = np->n_vattr.na_vattr.va_ctime.tv_sec;
1222 cache_enter(dvp, newvp, cnp);
1223 }
1224 *vpp = newvp;
1225 return (0);
1226}
1227
1228/*
1229 * nfs read call.
1230 * Just call ncl_bioread() to do the work.
1231 */
1232static int
1233nfs_read(struct vop_read_args *ap)
1234{
1235 struct vnode *vp = ap->a_vp;
1236
1237 switch (vp->v_type) {
1238 case VREG:
1239 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1240 case VDIR:
1241 return (EISDIR);
1242 default:
1243 return (EOPNOTSUPP);
1244 }
1245}
1246
1247/*
1248 * nfs readlink call
1249 */
1250static int
1251nfs_readlink(struct vop_readlink_args *ap)
1252{
1253 struct vnode *vp = ap->a_vp;
1254
1255 if (vp->v_type != VLNK)
1256 return (EINVAL);
1257 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1258}
1259
1260/*
1261 * Do a readlink rpc.
1262 * Called by ncl_doio() from below the buffer cache.
1263 */
1264int
1265ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1266{
1267 int error, ret, attrflag;
1268 struct nfsvattr nfsva;
1269
1270 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1271 &attrflag, NULL);
1272 if (attrflag) {
1273 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1274 if (ret && !error)
1275 error = ret;
1276 }
1277 if (error && NFS_ISV4(vp))
1278 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1279 return (error);
1280}
1281
1282/*
1283 * nfs read rpc call
1284 * Ditto above
1285 */
1286int
1287ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1288{
1289 int error, ret, attrflag;
1290 struct nfsvattr nfsva;
1291
1292 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1293 NULL);
1294 if (attrflag) {
1295 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1296 if (ret && !error)
1297 error = ret;
1298 }
1299 if (error && NFS_ISV4(vp))
1300 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1301 return (error);
1302}
1303
1304/*
1305 * nfs write call
1306 */
1307int
1308ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
| 680 /*
681 * as above w.r.t races when clearing
682 * NMODIFIED.
683 * np->n_flag &= ~NMODIFIED;
684 */
685 }
686 } else
687 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
688 mtx_lock(&np->n_mtx);
689 }
690 /*
691 * Invalidate the attribute cache in all cases.
692 * An open is going to fetch fresh attrs any way, other procs
693 * on this node that have file open will be forced to do an
694 * otw attr fetch, but this is safe.
695 * --> A user found that their RPC count dropped by 20% when
696 * this was commented out and I can't see any requirement
697 * for it, so I've disabled it when negative lookups are
698 * enabled. (What does this have to do with negative lookup
699 * caching? Well nothing, except it was reported by the
700 * same user that needed negative lookup caching and I wanted
701 * there to be a way to disable it to see if it
702 * is the cause of some caching/coherency issue that might
703 * crop up.)
704 */
705 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0)
706 np->n_attrstamp = 0;
707 if (np->n_flag & NWRITEERR) {
708 np->n_flag &= ~NWRITEERR;
709 error = np->n_error;
710 }
711 mtx_unlock(&np->n_mtx);
712 }
713
714 if (NFS_ISV4(vp)) {
715 /*
716 * Get attributes so "change" is up to date.
717 */
718 if (!error) {
719 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
720 NULL);
721 if (!ret) {
722 np->n_change = nfsva.na_filerev;
723 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
724 NULL, 0, 0);
725 }
726 }
727
728 /*
729 * and do the close.
730 */
731 ret = nfsrpc_close(vp, 0, ap->a_td);
732 if (!error && ret)
733 error = ret;
734 if (error)
735 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
736 (gid_t)0);
737 }
738 if (newnfs_directio_enable)
739 KASSERT((np->n_directio_asyncwr == 0),
740 ("nfs_close: dirty unflushed (%d) directio buffers\n",
741 np->n_directio_asyncwr));
742 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
743 mtx_lock(&np->n_mtx);
744 KASSERT((np->n_directio_opens > 0),
745 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
746 np->n_directio_opens--;
747 if (np->n_directio_opens == 0)
748 np->n_flag &= ~NNONCACHE;
749 mtx_unlock(&np->n_mtx);
750 }
751 if (localcred)
752 NFSFREECRED(cred);
753 return (error);
754}
755
756/*
757 * nfs getattr call from vfs.
758 */
759static int
760nfs_getattr(struct vop_getattr_args *ap)
761{
762 struct vnode *vp = ap->a_vp;
763 struct thread *td = curthread; /* XXX */
764 struct nfsnode *np = VTONFS(vp);
765 int error = 0;
766 struct nfsvattr nfsva;
767 struct vattr *vap = ap->a_vap;
768 struct vattr vattr;
769
770 /*
771 * Update local times for special files.
772 */
773 mtx_lock(&np->n_mtx);
774 if (np->n_flag & (NACC | NUPD))
775 np->n_flag |= NCHG;
776 mtx_unlock(&np->n_mtx);
777 /*
778 * First look in the cache.
779 */
780 if (ncl_getattrcache(vp, &vattr) == 0) {
781 vap->va_type = vattr.va_type;
782 vap->va_mode = vattr.va_mode;
783 vap->va_nlink = vattr.va_nlink;
784 vap->va_uid = vattr.va_uid;
785 vap->va_gid = vattr.va_gid;
786 vap->va_fsid = vattr.va_fsid;
787 vap->va_fileid = vattr.va_fileid;
788 vap->va_size = vattr.va_size;
789 vap->va_blocksize = vattr.va_blocksize;
790 vap->va_atime = vattr.va_atime;
791 vap->va_mtime = vattr.va_mtime;
792 vap->va_ctime = vattr.va_ctime;
793 vap->va_gen = vattr.va_gen;
794 vap->va_flags = vattr.va_flags;
795 vap->va_rdev = vattr.va_rdev;
796 vap->va_bytes = vattr.va_bytes;
797 vap->va_filerev = vattr.va_filerev;
798 /*
799 * Get the local modify time for the case of a write
800 * delegation.
801 */
802 nfscl_deleggetmodtime(vp, &vap->va_mtime);
803 return (0);
804 }
805
806 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
807 nfsaccess_cache_timeout > 0) {
808 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
809 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
810 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
811 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
812 return (0);
813 }
814 }
815 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
816 if (!error)
817 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
818 if (!error) {
819 /*
820 * Get the local modify time for the case of a write
821 * delegation.
822 */
823 nfscl_deleggetmodtime(vp, &vap->va_mtime);
824 } else if (NFS_ISV4(vp)) {
825 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
826 }
827 return (error);
828}
829
830/*
831 * nfs setattr call.
832 */
833static int
834nfs_setattr(struct vop_setattr_args *ap)
835{
836 struct vnode *vp = ap->a_vp;
837 struct nfsnode *np = VTONFS(vp);
838 struct thread *td = curthread; /* XXX */
839 struct vattr *vap = ap->a_vap;
840 int error = 0;
841 u_quad_t tsize;
842
843#ifndef nolint
844 tsize = (u_quad_t)0;
845#endif
846
847 /*
848 * Setting of flags and marking of atimes are not supported.
849 */
850 if (vap->va_flags != VNOVAL)
851 return (EOPNOTSUPP);
852
853 /*
854 * Disallow write attempts if the filesystem is mounted read-only.
855 */
856 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
857 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
858 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
859 (vp->v_mount->mnt_flag & MNT_RDONLY))
860 return (EROFS);
861 if (vap->va_size != VNOVAL) {
862 switch (vp->v_type) {
863 case VDIR:
864 return (EISDIR);
865 case VCHR:
866 case VBLK:
867 case VSOCK:
868 case VFIFO:
869 if (vap->va_mtime.tv_sec == VNOVAL &&
870 vap->va_atime.tv_sec == VNOVAL &&
871 vap->va_mode == (mode_t)VNOVAL &&
872 vap->va_uid == (uid_t)VNOVAL &&
873 vap->va_gid == (gid_t)VNOVAL)
874 return (0);
875 vap->va_size = VNOVAL;
876 break;
877 default:
878 /*
879 * Disallow write attempts if the filesystem is
880 * mounted read-only.
881 */
882 if (vp->v_mount->mnt_flag & MNT_RDONLY)
883 return (EROFS);
884 /*
885 * We run vnode_pager_setsize() early (why?),
886 * we must set np->n_size now to avoid vinvalbuf
887 * V_SAVE races that might setsize a lower
888 * value.
889 */
890 mtx_lock(&np->n_mtx);
891 tsize = np->n_size;
892 mtx_unlock(&np->n_mtx);
893 error = ncl_meta_setsize(vp, ap->a_cred, td,
894 vap->va_size);
895 mtx_lock(&np->n_mtx);
896 if (np->n_flag & NMODIFIED) {
897 tsize = np->n_size;
898 mtx_unlock(&np->n_mtx);
899 if (vap->va_size == 0)
900 error = ncl_vinvalbuf(vp, 0, td, 1);
901 else
902 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
903 if (error) {
904 vnode_pager_setsize(vp, tsize);
905 return (error);
906 }
907 /*
908 * Call nfscl_delegmodtime() to set the modify time
909 * locally, as required.
910 */
911 nfscl_delegmodtime(vp);
912 } else
913 mtx_unlock(&np->n_mtx);
914 /*
915 * np->n_size has already been set to vap->va_size
916 * in ncl_meta_setsize(). We must set it again since
917 * nfs_loadattrcache() could be called through
918 * ncl_meta_setsize() and could modify np->n_size.
919 */
920 mtx_lock(&np->n_mtx);
921 np->n_vattr.na_size = np->n_size = vap->va_size;
922 mtx_unlock(&np->n_mtx);
923 };
924 } else {
925 mtx_lock(&np->n_mtx);
926 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
927 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
928 mtx_unlock(&np->n_mtx);
929 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
930 (error == EINTR || error == EIO))
931 return (error);
932 } else
933 mtx_unlock(&np->n_mtx);
934 }
935 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
936 if (error && vap->va_size != VNOVAL) {
937 mtx_lock(&np->n_mtx);
938 np->n_size = np->n_vattr.na_size = tsize;
939 vnode_pager_setsize(vp, tsize);
940 mtx_unlock(&np->n_mtx);
941 }
942 return (error);
943}
944
945/*
946 * Do an nfs setattr rpc.
947 */
948static int
949nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
950 struct thread *td)
951{
952 struct nfsnode *np = VTONFS(vp);
953 int error, ret, attrflag, i;
954 struct nfsvattr nfsva;
955
956 if (NFS_ISV34(vp)) {
957 mtx_lock(&np->n_mtx);
958 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
959 np->n_accesscache[i].stamp = 0;
960 np->n_flag |= NDELEGMOD;
961 mtx_unlock(&np->n_mtx);
962 }
963 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
964 NULL);
965 if (attrflag) {
966 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
967 if (ret && !error)
968 error = ret;
969 }
970 if (error && NFS_ISV4(vp))
971 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
972 return (error);
973}
974
975/*
976 * nfs lookup call, one step at a time...
977 * First look in cache
978 * If not found, unlock the directory nfsnode and do the rpc
979 */
980static int
981nfs_lookup(struct vop_lookup_args *ap)
982{
983 struct componentname *cnp = ap->a_cnp;
984 struct vnode *dvp = ap->a_dvp;
985 struct vnode **vpp = ap->a_vpp;
986 struct mount *mp = dvp->v_mount;
987 int flags = cnp->cn_flags;
988 struct vnode *newvp;
989 struct nfsmount *nmp;
990 struct nfsnode *np;
991 int error = 0, attrflag, dattrflag, ltype;
992 struct thread *td = cnp->cn_thread;
993 struct nfsfh *nfhp;
994 struct nfsvattr dnfsva, nfsva;
995 struct vattr vattr;
996 time_t dmtime;
997
998 *vpp = NULLVP;
999 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1000 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1001 return (EROFS);
1002 if (dvp->v_type != VDIR)
1003 return (ENOTDIR);
1004 nmp = VFSTONFS(mp);
1005 np = VTONFS(dvp);
1006
1007 /* For NFSv4, wait until any remove is done. */
1008 mtx_lock(&np->n_mtx);
1009 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1010 np->n_flag |= NREMOVEWANT;
1011 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1012 }
1013 mtx_unlock(&np->n_mtx);
1014
1015 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1016 return (error);
1017 error = cache_lookup(dvp, vpp, cnp);
1018 if (error > 0 && error != ENOENT)
1019 return (error);
1020 if (error == -1) {
1021 /*
1022 * We only accept a positive hit in the cache if the
1023 * change time of the file matches our cached copy.
1024 * Otherwise, we discard the cache entry and fallback
1025 * to doing a lookup RPC.
1026 */
1027 newvp = *vpp;
1028 if (nfscl_nodeleg(newvp, 0) == 0 ||
1029 (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
1030 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime)) {
1031 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1032 if (cnp->cn_nameiop != LOOKUP &&
1033 (flags & ISLASTCN))
1034 cnp->cn_flags |= SAVENAME;
1035 return (0);
1036 }
1037 cache_purge(newvp);
1038 if (dvp != newvp)
1039 vput(newvp);
1040 else
1041 vrele(newvp);
1042 *vpp = NULLVP;
1043 } else if (error == ENOENT) {
1044 if (dvp->v_iflag & VI_DOOMED)
1045 return (ENOENT);
1046 /*
1047 * We only accept a negative hit in the cache if the
1048 * modification time of the parent directory matches
1049 * our cached copy. Otherwise, we discard all of the
1050 * negative cache entries for this directory. We also
1051 * only trust -ve cache entries for less than
1052 * nm_negative_namecache_timeout seconds.
1053 */
1054 if ((u_int)(ticks - np->n_dmtime_ticks) <
1055 (nmp->nm_negnametimeo * hz) &&
1056 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1057 vattr.va_mtime.tv_sec == np->n_dmtime) {
1058 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1059 return (ENOENT);
1060 }
1061 cache_purge_negative(dvp);
1062 mtx_lock(&np->n_mtx);
1063 np->n_dmtime = 0;
1064 mtx_unlock(&np->n_mtx);
1065 }
1066
1067 /*
1068 * Cache the modification time of the parent directory in case
1069 * the lookup fails and results in adding the first negative
1070 * name cache entry for the directory. Since this is reading
1071 * a single time_t, don't bother with locking. The
1072 * modification time may be a bit stale, but it must be read
1073 * before performing the lookup RPC to prevent a race where
1074 * another lookup updates the timestamp on the directory after
1075 * the lookup RPC has been performed on the server but before
1076 * n_dmtime is set at the end of this function.
1077 */
1078 dmtime = np->n_vattr.na_mtime.tv_sec;
1079 error = 0;
1080 newvp = NULLVP;
1081 NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1082 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1083 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1084 NULL);
1085 if (dattrflag)
1086 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1087 if (error) {
1088 if (newvp != NULLVP) {
1089 vput(newvp);
1090 *vpp = NULLVP;
1091 }
1092
1093 if (error != ENOENT) {
1094 if (NFS_ISV4(dvp))
1095 error = nfscl_maperr(td, error, (uid_t)0,
1096 (gid_t)0);
1097 return (error);
1098 }
1099
1100 /* The requested file was not found. */
1101 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1102 (flags & ISLASTCN)) {
1103 /*
1104 * XXX: UFS does a full VOP_ACCESS(dvp,
1105 * VWRITE) here instead of just checking
1106 * MNT_RDONLY.
1107 */
1108 if (mp->mnt_flag & MNT_RDONLY)
1109 return (EROFS);
1110 cnp->cn_flags |= SAVENAME;
1111 return (EJUSTRETURN);
1112 }
1113
1114 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1115 /*
1116 * Maintain n_dmtime as the modification time
1117 * of the parent directory when the oldest -ve
1118 * name cache entry for this directory was
1119 * added. If a -ve cache entry has already
1120 * been added with a newer modification time
1121 * by a concurrent lookup, then don't bother
1122 * adding a cache entry. The modification
1123 * time of the directory might have changed
1124 * due to the file this lookup failed to find
1125 * being created. In that case a subsequent
1126 * lookup would incorrectly use the entry
1127 * added here instead of doing an extra
1128 * lookup.
1129 */
1130 mtx_lock(&np->n_mtx);
1131 if (np->n_dmtime <= dmtime) {
1132 if (np->n_dmtime == 0) {
1133 np->n_dmtime = dmtime;
1134 np->n_dmtime_ticks = ticks;
1135 }
1136 mtx_unlock(&np->n_mtx);
1137 cache_enter(dvp, NULL, cnp);
1138 } else
1139 mtx_unlock(&np->n_mtx);
1140 }
1141 return (ENOENT);
1142 }
1143
1144 /*
1145 * Handle RENAME case...
1146 */
1147 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1148 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1149 FREE((caddr_t)nfhp, M_NFSFH);
1150 return (EISDIR);
1151 }
1152 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1153 if (error)
1154 return (error);
1155 newvp = NFSTOV(np);
1156 if (attrflag)
1157 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1158 0, 1);
1159 *vpp = newvp;
1160 cnp->cn_flags |= SAVENAME;
1161 return (0);
1162 }
1163
1164 if (flags & ISDOTDOT) {
1165 ltype = VOP_ISLOCKED(dvp);
1166 error = vfs_busy(mp, MBF_NOWAIT);
1167 if (error != 0) {
1168 vfs_ref(mp);
1169 VOP_UNLOCK(dvp, 0);
1170 error = vfs_busy(mp, 0);
1171 vn_lock(dvp, ltype | LK_RETRY);
1172 vfs_rel(mp);
1173 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1174 vfs_unbusy(mp);
1175 error = ENOENT;
1176 }
1177 if (error != 0)
1178 return (error);
1179 }
1180 VOP_UNLOCK(dvp, 0);
1181 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1182 if (error == 0)
1183 newvp = NFSTOV(np);
1184 vfs_unbusy(mp);
1185 if (newvp != dvp)
1186 vn_lock(dvp, ltype | LK_RETRY);
1187 if (dvp->v_iflag & VI_DOOMED) {
1188 if (error == 0) {
1189 if (newvp == dvp)
1190 vrele(newvp);
1191 else
1192 vput(newvp);
1193 }
1194 error = ENOENT;
1195 }
1196 if (error != 0)
1197 return (error);
1198 if (attrflag)
1199 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1200 0, 1);
1201 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1202 FREE((caddr_t)nfhp, M_NFSFH);
1203 VREF(dvp);
1204 newvp = dvp;
1205 if (attrflag)
1206 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1207 0, 1);
1208 } else {
1209 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1210 if (error)
1211 return (error);
1212 newvp = NFSTOV(np);
1213 if (attrflag)
1214 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1215 0, 1);
1216 }
1217 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1218 cnp->cn_flags |= SAVENAME;
1219 if ((cnp->cn_flags & MAKEENTRY) &&
1220 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1221 np->n_ctime = np->n_vattr.na_vattr.va_ctime.tv_sec;
1222 cache_enter(dvp, newvp, cnp);
1223 }
1224 *vpp = newvp;
1225 return (0);
1226}
1227
1228/*
1229 * nfs read call.
1230 * Just call ncl_bioread() to do the work.
1231 */
1232static int
1233nfs_read(struct vop_read_args *ap)
1234{
1235 struct vnode *vp = ap->a_vp;
1236
1237 switch (vp->v_type) {
1238 case VREG:
1239 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1240 case VDIR:
1241 return (EISDIR);
1242 default:
1243 return (EOPNOTSUPP);
1244 }
1245}
1246
1247/*
1248 * nfs readlink call
1249 */
1250static int
1251nfs_readlink(struct vop_readlink_args *ap)
1252{
1253 struct vnode *vp = ap->a_vp;
1254
1255 if (vp->v_type != VLNK)
1256 return (EINVAL);
1257 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1258}
1259
1260/*
1261 * Do a readlink rpc.
1262 * Called by ncl_doio() from below the buffer cache.
1263 */
1264int
1265ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1266{
1267 int error, ret, attrflag;
1268 struct nfsvattr nfsva;
1269
1270 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1271 &attrflag, NULL);
1272 if (attrflag) {
1273 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1274 if (ret && !error)
1275 error = ret;
1276 }
1277 if (error && NFS_ISV4(vp))
1278 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1279 return (error);
1280}
1281
1282/*
1283 * nfs read rpc call
1284 * Ditto above
1285 */
1286int
1287ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1288{
1289 int error, ret, attrflag;
1290 struct nfsvattr nfsva;
1291
1292 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1293 NULL);
1294 if (attrflag) {
1295 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1296 if (ret && !error)
1297 error = ret;
1298 }
1299 if (error && NFS_ISV4(vp))
1300 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1301 return (error);
1302}
1303
1304/*
1305 * nfs write call
1306 */
1307int
1308ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
|
1309 int *iomode, int *must_commit)
| 1309 int *iomode, int *must_commit, int called_from_strategy)
|
1310{
1311 struct nfsvattr nfsva;
1312 int error = 0, attrflag, ret;
1313 u_char verf[NFSX_VERF];
1314 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1315
1316 *must_commit = 0;
1317 error = nfsrpc_write(vp, uiop, iomode, verf, cred,
| 1310{
1311 struct nfsvattr nfsva;
1312 int error = 0, attrflag, ret;
1313 u_char verf[NFSX_VERF];
1314 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1315
1316 *must_commit = 0;
1317 error = nfsrpc_write(vp, uiop, iomode, verf, cred,
|
1318 uiop->uio_td, &nfsva, &attrflag, NULL);
| 1318 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy);
|
1319 NFSLOCKMNT(nmp);
1320 if (!error && NFSHASWRITEVERF(nmp) &&
1321 NFSBCMP(verf, nmp->nm_verf, NFSX_VERF)) {
1322 *must_commit = 1;
1323 NFSBCOPY(verf, nmp->nm_verf, NFSX_VERF);
1324 }
1325 NFSUNLOCKMNT(nmp);
1326 if (attrflag) {
1327 if (VTONFS(vp)->n_flag & ND_NFSV4)
1328 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1329 1);
1330 else
1331 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1332 1);
1333 if (ret && !error)
1334 error = ret;
1335 }
1336 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1337 *iomode = NFSWRITE_FILESYNC;
1338 if (error && NFS_ISV4(vp))
1339 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1340 return (error);
1341}
1342
1343/*
1344 * nfs mknod rpc
1345 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1346 * mode set to specify the file type and the size field for rdev.
1347 */
1348static int
1349nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1350 struct vattr *vap)
1351{
1352 struct nfsvattr nfsva, dnfsva;
1353 struct vnode *newvp = NULL;
1354 struct nfsnode *np = NULL, *dnp;
1355 struct nfsfh *nfhp;
1356 struct vattr vattr;
1357 int error = 0, attrflag, dattrflag;
1358 u_int32_t rdev;
1359
1360 if (vap->va_type == VCHR || vap->va_type == VBLK)
1361 rdev = vap->va_rdev;
1362 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1363 rdev = 0xffffffff;
1364 else
1365 return (EOPNOTSUPP);
1366 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1367 return (error);
1368 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1369 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1370 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1371 if (!error) {
1372 if (!nfhp)
1373 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1374 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1375 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1376 NULL);
1377 if (nfhp)
1378 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1379 cnp->cn_thread, &np, NULL);
1380 }
1381 if (dattrflag)
1382 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1383 if (!error) {
1384 newvp = NFSTOV(np);
1385 if (attrflag)
1386 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1387 0, 1);
1388 }
1389 if (!error) {
1390 if ((cnp->cn_flags & MAKEENTRY))
1391 cache_enter(dvp, newvp, cnp);
1392 *vpp = newvp;
1393 } else if (NFS_ISV4(dvp)) {
1394 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1395 vap->va_gid);
1396 }
1397 dnp = VTONFS(dvp);
1398 mtx_lock(&dnp->n_mtx);
1399 dnp->n_flag |= NMODIFIED;
1400 if (!dattrflag)
1401 dnp->n_attrstamp = 0;
1402 mtx_unlock(&dnp->n_mtx);
1403 return (error);
1404}
1405
1406/*
1407 * nfs mknod vop
1408 * just call nfs_mknodrpc() to do the work.
1409 */
1410/* ARGSUSED */
1411static int
1412nfs_mknod(struct vop_mknod_args *ap)
1413{
1414 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1415}
1416
1417static struct mtx nfs_cverf_mtx;
1418MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1419 MTX_DEF);
1420
1421static nfsquad_t
1422nfs_get_cverf(void)
1423{
1424 static nfsquad_t cverf;
1425 nfsquad_t ret;
1426 static int cverf_initialized = 0;
1427
1428 mtx_lock(&nfs_cverf_mtx);
1429 if (cverf_initialized == 0) {
1430 cverf.lval[0] = arc4random();
1431 cverf.lval[1] = arc4random();
1432 cverf_initialized = 1;
1433 } else
1434 cverf.qval++;
1435 ret = cverf;
1436 mtx_unlock(&nfs_cverf_mtx);
1437
1438 return (ret);
1439}
1440
1441/*
1442 * nfs file create call
1443 */
1444static int
1445nfs_create(struct vop_create_args *ap)
1446{
1447 struct vnode *dvp = ap->a_dvp;
1448 struct vattr *vap = ap->a_vap;
1449 struct componentname *cnp = ap->a_cnp;
1450 struct nfsnode *np = NULL, *dnp;
1451 struct vnode *newvp = NULL;
1452 struct nfsmount *nmp;
1453 struct nfsvattr dnfsva, nfsva;
1454 struct nfsfh *nfhp;
1455 nfsquad_t cverf;
1456 int error = 0, attrflag, dattrflag, fmode = 0;
1457 struct vattr vattr;
1458
1459 /*
1460 * Oops, not for me..
1461 */
1462 if (vap->va_type == VSOCK)
1463 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1464
1465 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1466 return (error);
1467 if (vap->va_vaflags & VA_EXCLUSIVE)
1468 fmode |= O_EXCL;
1469 dnp = VTONFS(dvp);
1470 nmp = VFSTONFS(vnode_mount(dvp));
1471again:
1472 /* For NFSv4, wait until any remove is done. */
1473 mtx_lock(&dnp->n_mtx);
1474 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1475 dnp->n_flag |= NREMOVEWANT;
1476 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1477 }
1478 mtx_unlock(&dnp->n_mtx);
1479
1480 cverf = nfs_get_cverf();
1481 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1482 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1483 &nfhp, &attrflag, &dattrflag, NULL);
1484 if (!error) {
1485 if (nfhp == NULL)
1486 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1487 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1488 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1489 NULL);
1490 if (nfhp != NULL)
1491 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1492 cnp->cn_thread, &np, NULL);
1493 }
1494 if (dattrflag)
1495 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1496 if (!error) {
1497 newvp = NFSTOV(np);
1498 if (attrflag)
1499 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1500 0, 1);
1501 }
1502 if (error) {
1503 if (newvp != NULL) {
1504 vrele(newvp);
1505 newvp = NULL;
1506 }
1507 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1508 error == NFSERR_NOTSUPP) {
1509 fmode &= ~O_EXCL;
1510 goto again;
1511 }
1512 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1513 if (nfscl_checksattr(vap, &nfsva)) {
1514 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1515 cnp->cn_thread, &nfsva, &attrflag, NULL);
1516 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1517 vap->va_gid != (gid_t)VNOVAL)) {
1518 /* try again without setting uid/gid */
1519 vap->va_uid = (uid_t)VNOVAL;
1520 vap->va_gid = (uid_t)VNOVAL;
1521 error = nfsrpc_setattr(newvp, vap, NULL,
1522 cnp->cn_cred, cnp->cn_thread, &nfsva,
1523 &attrflag, NULL);
1524 }
1525 if (attrflag)
1526 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1527 NULL, 0, 1);
1528 }
1529 }
1530 if (!error) {
1531 if (cnp->cn_flags & MAKEENTRY)
1532 cache_enter(dvp, newvp, cnp);
1533 *ap->a_vpp = newvp;
1534 } else if (NFS_ISV4(dvp)) {
1535 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1536 vap->va_gid);
1537 }
1538 mtx_lock(&dnp->n_mtx);
1539 dnp->n_flag |= NMODIFIED;
1540 if (!dattrflag)
1541 dnp->n_attrstamp = 0;
1542 mtx_unlock(&dnp->n_mtx);
1543 return (error);
1544}
1545
1546/*
1547 * nfs file remove call
1548 * To try and make nfs semantics closer to ufs semantics, a file that has
1549 * other processes using the vnode is renamed instead of removed and then
1550 * removed later on the last close.
1551 * - If v_usecount > 1
1552 * If a rename is not already in the works
1553 * call nfs_sillyrename() to set it up
1554 * else
1555 * do the remove rpc
1556 */
1557static int
1558nfs_remove(struct vop_remove_args *ap)
1559{
1560 struct vnode *vp = ap->a_vp;
1561 struct vnode *dvp = ap->a_dvp;
1562 struct componentname *cnp = ap->a_cnp;
1563 struct nfsnode *np = VTONFS(vp);
1564 int error = 0;
1565 struct vattr vattr;
1566
1567#ifndef DIAGNOSTIC
1568 if ((cnp->cn_flags & HASBUF) == 0)
1569 panic("nfs_remove: no name");
1570 if (vrefcnt(vp) < 1)
1571 panic("nfs_remove: bad v_usecount");
1572#endif
1573 if (vp->v_type == VDIR)
1574 error = EPERM;
1575 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1576 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1577 vattr.va_nlink > 1)) {
1578 /*
1579 * Purge the name cache so that the chance of a lookup for
1580 * the name succeeding while the remove is in progress is
1581 * minimized. Without node locking it can still happen, such
1582 * that an I/O op returns ESTALE, but since you get this if
1583 * another host removes the file..
1584 */
1585 cache_purge(vp);
1586 /*
1587 * throw away biocache buffers, mainly to avoid
1588 * unnecessary delayed writes later.
1589 */
1590 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1591 /* Do the rpc */
1592 if (error != EINTR && error != EIO)
1593 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1594 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1595 /*
1596 * Kludge City: If the first reply to the remove rpc is lost..
1597 * the reply to the retransmitted request will be ENOENT
1598 * since the file was in fact removed
1599 * Therefore, we cheat and return success.
1600 */
1601 if (error == ENOENT)
1602 error = 0;
1603 } else if (!np->n_sillyrename)
1604 error = nfs_sillyrename(dvp, vp, cnp);
1605 np->n_attrstamp = 0;
1606 return (error);
1607}
1608
1609/*
1610 * nfs file remove rpc called from nfs_inactive
1611 */
1612int
1613ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1614{
1615 /*
1616 * Make sure that the directory vnode is still valid.
1617 * XXX we should lock sp->s_dvp here.
1618 */
1619 if (sp->s_dvp->v_type == VBAD)
1620 return (0);
1621 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1622 sp->s_cred, NULL));
1623}
1624
1625/*
1626 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1627 */
1628static int
1629nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1630 int namelen, struct ucred *cred, struct thread *td)
1631{
1632 struct nfsvattr dnfsva;
1633 struct nfsnode *dnp = VTONFS(dvp);
1634 int error = 0, dattrflag;
1635
1636 mtx_lock(&dnp->n_mtx);
1637 dnp->n_flag |= NREMOVEINPROG;
1638 mtx_unlock(&dnp->n_mtx);
1639 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1640 &dattrflag, NULL);
1641 mtx_lock(&dnp->n_mtx);
1642 if ((dnp->n_flag & NREMOVEWANT)) {
1643 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1644 mtx_unlock(&dnp->n_mtx);
1645 wakeup((caddr_t)dnp);
1646 } else {
1647 dnp->n_flag &= ~NREMOVEINPROG;
1648 mtx_unlock(&dnp->n_mtx);
1649 }
1650 if (dattrflag)
1651 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1652 mtx_lock(&dnp->n_mtx);
1653 dnp->n_flag |= NMODIFIED;
1654 if (!dattrflag)
1655 dnp->n_attrstamp = 0;
1656 mtx_unlock(&dnp->n_mtx);
1657 if (error && NFS_ISV4(dvp))
1658 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1659 return (error);
1660}
1661
1662/*
1663 * nfs file rename call
1664 */
1665static int
1666nfs_rename(struct vop_rename_args *ap)
1667{
1668 struct vnode *fvp = ap->a_fvp;
1669 struct vnode *tvp = ap->a_tvp;
1670 struct vnode *fdvp = ap->a_fdvp;
1671 struct vnode *tdvp = ap->a_tdvp;
1672 struct componentname *tcnp = ap->a_tcnp;
1673 struct componentname *fcnp = ap->a_fcnp;
1674 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1675 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1676 struct nfsv4node *newv4 = NULL;
1677 int error;
1678
1679#ifndef DIAGNOSTIC
1680 if ((tcnp->cn_flags & HASBUF) == 0 ||
1681 (fcnp->cn_flags & HASBUF) == 0)
1682 panic("nfs_rename: no name");
1683#endif
1684 /* Check for cross-device rename */
1685 if ((fvp->v_mount != tdvp->v_mount) ||
1686 (tvp && (fvp->v_mount != tvp->v_mount))) {
1687 error = EXDEV;
1688 goto out;
1689 }
1690
1691 if (fvp == tvp) {
1692 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1693 error = 0;
1694 goto out;
1695 }
1696 if ((error = vn_lock(fvp, LK_EXCLUSIVE)))
1697 goto out;
1698
1699 /*
1700 * We have to flush B_DELWRI data prior to renaming
1701 * the file. If we don't, the delayed-write buffers
1702 * can be flushed out later after the file has gone stale
1703 * under NFSV3. NFSV2 does not have this problem because
1704 * ( as far as I can tell ) it flushes dirty buffers more
1705 * often.
1706 *
1707 * Skip the rename operation if the fsync fails, this can happen
1708 * due to the server's volume being full, when we pushed out data
1709 * that was written back to our cache earlier. Not checking for
1710 * this condition can result in potential (silent) data loss.
1711 */
1712 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1713 VOP_UNLOCK(fvp, 0);
1714 if (!error && tvp)
1715 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1716 if (error)
1717 goto out;
1718
1719 /*
1720 * If the tvp exists and is in use, sillyrename it before doing the
1721 * rename of the new file over it.
1722 * XXX Can't sillyrename a directory.
1723 */
1724 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1725 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1726 vput(tvp);
1727 tvp = NULL;
1728 }
1729
1730 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1731 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1732 tcnp->cn_thread);
1733
1734 if (!error) {
1735 /*
1736 * For NFSv4, check to see if it is the same name and
1737 * replace the name, if it is different.
1738 */
1739 MALLOC(newv4, struct nfsv4node *,
1740 sizeof (struct nfsv4node) +
1741 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1742 M_NFSV4NODE, M_WAITOK);
1743 mtx_lock(&tdnp->n_mtx);
1744 mtx_lock(&fnp->n_mtx);
1745 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1746 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1747 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1748 tcnp->cn_namelen) ||
1749 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1750 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1751 tdnp->n_fhp->nfh_len))) {
1752#ifdef notdef
1753{ char nnn[100]; int nnnl;
1754nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1755bcopy(tcnp->cn_nameptr, nnn, nnnl);
1756nnn[nnnl] = '\0';
1757printf("ren replace=%s\n",nnn);
1758}
1759#endif
1760 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1761 fnp->n_v4 = newv4;
1762 newv4 = NULL;
1763 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1764 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1765 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1766 tdnp->n_fhp->nfh_len);
1767 NFSBCOPY(tcnp->cn_nameptr,
1768 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1769 }
1770 mtx_unlock(&tdnp->n_mtx);
1771 mtx_unlock(&fnp->n_mtx);
1772 if (newv4 != NULL)
1773 FREE((caddr_t)newv4, M_NFSV4NODE);
1774 }
1775
1776 if (fvp->v_type == VDIR) {
1777 if (tvp != NULL && tvp->v_type == VDIR)
1778 cache_purge(tdvp);
1779 cache_purge(fdvp);
1780 }
1781
1782out:
1783 if (tdvp == tvp)
1784 vrele(tdvp);
1785 else
1786 vput(tdvp);
1787 if (tvp)
1788 vput(tvp);
1789 vrele(fdvp);
1790 vrele(fvp);
1791 /*
1792 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1793 */
1794 if (error == ENOENT)
1795 error = 0;
1796 return (error);
1797}
1798
1799/*
1800 * nfs file rename rpc called from nfs_remove() above
1801 */
1802static int
1803nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1804 struct sillyrename *sp)
1805{
1806
1807 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1808 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1809 scnp->cn_thread));
1810}
1811
1812/*
1813 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1814 */
1815static int
1816nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1817 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1818 int tnamelen, struct ucred *cred, struct thread *td)
1819{
1820 struct nfsvattr fnfsva, tnfsva;
1821 struct nfsnode *fdnp = VTONFS(fdvp);
1822 struct nfsnode *tdnp = VTONFS(tdvp);
1823 int error = 0, fattrflag, tattrflag;
1824
1825 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1826 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1827 &tattrflag, NULL, NULL);
1828 mtx_lock(&fdnp->n_mtx);
1829 fdnp->n_flag |= NMODIFIED;
1830 mtx_unlock(&fdnp->n_mtx);
1831 mtx_lock(&tdnp->n_mtx);
1832 tdnp->n_flag |= NMODIFIED;
1833 mtx_unlock(&tdnp->n_mtx);
1834 if (fattrflag)
1835 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1836 else
1837 fdnp->n_attrstamp = 0;
1838 if (tattrflag)
1839 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1840 else
1841 tdnp->n_attrstamp = 0;
1842 if (error && NFS_ISV4(fdvp))
1843 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1844 return (error);
1845}
1846
1847/*
1848 * nfs hard link create call
1849 */
1850static int
1851nfs_link(struct vop_link_args *ap)
1852{
1853 struct vnode *vp = ap->a_vp;
1854 struct vnode *tdvp = ap->a_tdvp;
1855 struct componentname *cnp = ap->a_cnp;
1856 struct nfsnode *tdnp;
1857 struct nfsvattr nfsva, dnfsva;
1858 int error = 0, attrflag, dattrflag;
1859
1860 if (vp->v_mount != tdvp->v_mount) {
1861 return (EXDEV);
1862 }
1863
1864 /*
1865 * Push all writes to the server, so that the attribute cache
1866 * doesn't get "out of sync" with the server.
1867 * XXX There should be a better way!
1868 */
1869 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1870
1871 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1872 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1873 &dattrflag, NULL);
1874 tdnp = VTONFS(tdvp);
1875 mtx_lock(&tdnp->n_mtx);
1876 tdnp->n_flag |= NMODIFIED;
1877 mtx_unlock(&tdnp->n_mtx);
1878 if (attrflag)
1879 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1880 else
1881 VTONFS(vp)->n_attrstamp = 0;
1882 if (dattrflag)
1883 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1884 else
1885 tdnp->n_attrstamp = 0;
1886 /*
1887 * If negative lookup caching is enabled, I might as well
1888 * add an entry for this node. Not necessary for correctness,
1889 * but if negative caching is enabled, then the system
1890 * must care about lookup caching hit rate, so...
1891 */
1892 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1893 (cnp->cn_flags & MAKEENTRY))
1894 cache_enter(tdvp, vp, cnp);
1895 if (error && NFS_ISV4(vp))
1896 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1897 (gid_t)0);
1898 return (error);
1899}
1900
1901/*
1902 * nfs symbolic link create call
1903 */
1904static int
1905nfs_symlink(struct vop_symlink_args *ap)
1906{
1907 struct vnode *dvp = ap->a_dvp;
1908 struct vattr *vap = ap->a_vap;
1909 struct componentname *cnp = ap->a_cnp;
1910 struct nfsvattr nfsva, dnfsva;
1911 struct nfsfh *nfhp;
1912 struct nfsnode *np = NULL, *dnp;
1913 struct vnode *newvp = NULL;
1914 int error = 0, attrflag, dattrflag, ret;
1915
1916 vap->va_type = VLNK;
1917 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1918 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1919 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1920 if (nfhp) {
1921 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
1922 &np, NULL);
1923 if (!ret)
1924 newvp = NFSTOV(np);
1925 else if (!error)
1926 error = ret;
1927 }
1928 if (newvp != NULL) {
1929 if (attrflag)
1930 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1931 0, 1);
1932 } else if (!error) {
1933 /*
1934 * If we do not have an error and we could not extract the
1935 * newvp from the response due to the request being NFSv2, we
1936 * have to do a lookup in order to obtain a newvp to return.
1937 */
1938 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1939 cnp->cn_cred, cnp->cn_thread, &np);
1940 if (!error)
1941 newvp = NFSTOV(np);
1942 }
1943 if (error) {
1944 if (newvp)
1945 vput(newvp);
1946 if (NFS_ISV4(dvp))
1947 error = nfscl_maperr(cnp->cn_thread, error,
1948 vap->va_uid, vap->va_gid);
1949 } else {
1950 /*
1951 * If negative lookup caching is enabled, I might as well
1952 * add an entry for this node. Not necessary for correctness,
1953 * but if negative caching is enabled, then the system
1954 * must care about lookup caching hit rate, so...
1955 */
1956 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
1957 (cnp->cn_flags & MAKEENTRY))
1958 cache_enter(dvp, newvp, cnp);
1959 *ap->a_vpp = newvp;
1960 }
1961
1962 dnp = VTONFS(dvp);
1963 mtx_lock(&dnp->n_mtx);
1964 dnp->n_flag |= NMODIFIED;
1965 mtx_unlock(&dnp->n_mtx);
1966 if (dattrflag)
1967 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1968 else
1969 dnp->n_attrstamp = 0;
1970 return (error);
1971}
1972
1973/*
1974 * nfs make dir call
1975 */
1976static int
1977nfs_mkdir(struct vop_mkdir_args *ap)
1978{
1979 struct vnode *dvp = ap->a_dvp;
1980 struct vattr *vap = ap->a_vap;
1981 struct componentname *cnp = ap->a_cnp;
1982 struct nfsnode *np = NULL, *dnp;
1983 struct vnode *newvp = NULL;
1984 struct vattr vattr;
1985 struct nfsfh *nfhp;
1986 struct nfsvattr nfsva, dnfsva;
1987 int error = 0, attrflag, dattrflag, ret;
1988
1989 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1990 return (error);
1991 vap->va_type = VDIR;
1992 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1993 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
1994 &attrflag, &dattrflag, NULL);
1995 dnp = VTONFS(dvp);
1996 mtx_lock(&dnp->n_mtx);
1997 dnp->n_flag |= NMODIFIED;
1998 mtx_unlock(&dnp->n_mtx);
1999 if (dattrflag)
2000 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2001 else
2002 dnp->n_attrstamp = 0;
2003 if (nfhp) {
2004 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2005 &np, NULL);
2006 if (!ret) {
2007 newvp = NFSTOV(np);
2008 if (attrflag)
2009 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2010 NULL, 0, 1);
2011 } else if (!error)
2012 error = ret;
2013 }
2014 if (!error && newvp == NULL) {
2015 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2016 cnp->cn_cred, cnp->cn_thread, &np);
2017 if (!error) {
2018 newvp = NFSTOV(np);
2019 if (newvp->v_type != VDIR)
2020 error = EEXIST;
2021 }
2022 }
2023 if (error) {
2024 if (newvp)
2025 vput(newvp);
2026 if (NFS_ISV4(dvp))
2027 error = nfscl_maperr(cnp->cn_thread, error,
2028 vap->va_uid, vap->va_gid);
2029 } else {
2030 /*
2031 * If negative lookup caching is enabled, I might as well
2032 * add an entry for this node. Not necessary for correctness,
2033 * but if negative caching is enabled, then the system
2034 * must care about lookup caching hit rate, so...
2035 */
2036 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2037 (cnp->cn_flags & MAKEENTRY))
2038 cache_enter(dvp, newvp, cnp);
2039 *ap->a_vpp = newvp;
2040 }
2041 return (error);
2042}
2043
2044/*
2045 * nfs remove directory call
2046 */
2047static int
2048nfs_rmdir(struct vop_rmdir_args *ap)
2049{
2050 struct vnode *vp = ap->a_vp;
2051 struct vnode *dvp = ap->a_dvp;
2052 struct componentname *cnp = ap->a_cnp;
2053 struct nfsnode *dnp;
2054 struct nfsvattr dnfsva;
2055 int error, dattrflag;
2056
2057 if (dvp == vp)
2058 return (EINVAL);
2059 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2060 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2061 dnp = VTONFS(dvp);
2062 mtx_lock(&dnp->n_mtx);
2063 dnp->n_flag |= NMODIFIED;
2064 mtx_unlock(&dnp->n_mtx);
2065 if (dattrflag)
2066 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2067 else
2068 dnp->n_attrstamp = 0;
2069
2070 cache_purge(dvp);
2071 cache_purge(vp);
2072 if (error && NFS_ISV4(dvp))
2073 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2074 (gid_t)0);
2075 /*
2076 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2077 */
2078 if (error == ENOENT)
2079 error = 0;
2080 return (error);
2081}
2082
2083/*
2084 * nfs readdir call
2085 */
2086static int
2087nfs_readdir(struct vop_readdir_args *ap)
2088{
2089 struct vnode *vp = ap->a_vp;
2090 struct nfsnode *np = VTONFS(vp);
2091 struct uio *uio = ap->a_uio;
2092 int tresid, error = 0;
2093 struct vattr vattr;
2094
2095 if (vp->v_type != VDIR)
2096 return(EPERM);
2097
2098 /*
2099 * First, check for hit on the EOF offset cache
2100 */
2101 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2102 (np->n_flag & NMODIFIED) == 0) {
2103 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2104 mtx_lock(&np->n_mtx);
2105 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2106 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2107 mtx_unlock(&np->n_mtx);
2108 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2109 return (0);
2110 } else
2111 mtx_unlock(&np->n_mtx);
2112 }
2113 }
2114
2115 /*
2116 * Call ncl_bioread() to do the real work.
2117 */
2118 tresid = uio->uio_resid;
2119 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2120
2121 if (!error && uio->uio_resid == tresid)
2122 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2123 return (error);
2124}
2125
2126/*
2127 * Readdir rpc call.
2128 * Called from below the buffer cache by ncl_doio().
2129 */
2130int
2131ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2132 struct thread *td)
2133{
2134 struct nfsvattr nfsva;
2135 nfsuint64 *cookiep, cookie;
2136 struct nfsnode *dnp = VTONFS(vp);
2137 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2138 int error = 0, eof, attrflag;
2139
2140#ifndef DIAGNOSTIC
2141 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2142 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2143 panic("nfs readdirrpc bad uio");
2144#endif
2145
2146 /*
2147 * If there is no cookie, assume directory was stale.
2148 */
2149 ncl_dircookie_lock(dnp);
2150 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2151 if (cookiep) {
2152 cookie = *cookiep;
2153 ncl_dircookie_unlock(dnp);
2154 } else {
2155 ncl_dircookie_unlock(dnp);
2156 return (NFSERR_BAD_COOKIE);
2157 }
2158
2159 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2160 (void)ncl_fsinfo(nmp, vp, cred, td);
2161
2162 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2163 &attrflag, &eof, NULL);
2164 if (attrflag)
2165 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2166
2167 if (!error) {
2168 /*
2169 * We are now either at the end of the directory or have filled
2170 * the block.
2171 */
2172 if (eof)
2173 dnp->n_direofoffset = uiop->uio_offset;
2174 else {
2175 if (uiop->uio_resid > 0)
2176 ncl_printf("EEK! readdirrpc resid > 0\n");
2177 ncl_dircookie_lock(dnp);
2178 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2179 *cookiep = cookie;
2180 ncl_dircookie_unlock(dnp);
2181 }
2182 } else if (NFS_ISV4(vp)) {
2183 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2184 }
2185 return (error);
2186}
2187
2188/*
2189 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2190 */
2191int
2192ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2193 struct thread *td)
2194{
2195 struct nfsvattr nfsva;
2196 nfsuint64 *cookiep, cookie;
2197 struct nfsnode *dnp = VTONFS(vp);
2198 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2199 int error = 0, attrflag, eof;
2200
2201#ifndef DIAGNOSTIC
2202 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2203 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2204 panic("nfs readdirplusrpc bad uio");
2205#endif
2206
2207 /*
2208 * If there is no cookie, assume directory was stale.
2209 */
2210 ncl_dircookie_lock(dnp);
2211 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2212 if (cookiep) {
2213 cookie = *cookiep;
2214 ncl_dircookie_unlock(dnp);
2215 } else {
2216 ncl_dircookie_unlock(dnp);
2217 return (NFSERR_BAD_COOKIE);
2218 }
2219
2220 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2221 (void)ncl_fsinfo(nmp, vp, cred, td);
2222 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2223 &attrflag, &eof, NULL);
2224 if (attrflag)
2225 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2226
2227 if (!error) {
2228 /*
2229 * We are now either at end of the directory or have filled the
2230 * the block.
2231 */
2232 if (eof)
2233 dnp->n_direofoffset = uiop->uio_offset;
2234 else {
2235 if (uiop->uio_resid > 0)
2236 ncl_printf("EEK! readdirplusrpc resid > 0\n");
2237 ncl_dircookie_lock(dnp);
2238 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2239 *cookiep = cookie;
2240 ncl_dircookie_unlock(dnp);
2241 }
2242 } else if (NFS_ISV4(vp)) {
2243 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2244 }
2245 return (error);
2246}
2247
2248/*
2249 * Silly rename. To make the NFS filesystem that is stateless look a little
2250 * more like the "ufs" a remove of an active vnode is translated to a rename
2251 * to a funny looking filename that is removed by nfs_inactive on the
2252 * nfsnode. There is the potential for another process on a different client
2253 * to create the same funny name between the nfs_lookitup() fails and the
2254 * nfs_rename() completes, but...
2255 */
2256static int
2257nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2258{
2259 struct sillyrename *sp;
2260 struct nfsnode *np;
2261 int error;
2262 short pid;
2263 unsigned int lticks;
2264
2265 cache_purge(dvp);
2266 np = VTONFS(vp);
2267#ifndef DIAGNOSTIC
2268 if (vp->v_type == VDIR)
2269 panic("nfs: sillyrename dir");
2270#endif
2271 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2272 M_NEWNFSREQ, M_WAITOK);
2273 sp->s_cred = crhold(cnp->cn_cred);
2274 sp->s_dvp = dvp;
2275 VREF(dvp);
2276
2277 /*
2278 * Fudge together a funny name.
2279 * Changing the format of the funny name to accomodate more
2280 * sillynames per directory.
2281 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2282 * CPU ticks since boot.
2283 */
2284 pid = cnp->cn_thread->td_proc->p_pid;
2285 lticks = (unsigned int)ticks;
2286 for ( ; ; ) {
2287 sp->s_namlen = sprintf(sp->s_name,
2288 ".nfs.%08x.%04x4.4", lticks,
2289 pid);
2290 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2291 cnp->cn_thread, NULL))
2292 break;
2293 lticks++;
2294 }
2295 error = nfs_renameit(dvp, vp, cnp, sp);
2296 if (error)
2297 goto bad;
2298 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2299 cnp->cn_thread, &np);
2300 np->n_sillyrename = sp;
2301 return (0);
2302bad:
2303 vrele(sp->s_dvp);
2304 crfree(sp->s_cred);
2305 free((caddr_t)sp, M_NEWNFSREQ);
2306 return (error);
2307}
2308
2309/*
2310 * Look up a file name and optionally either update the file handle or
2311 * allocate an nfsnode, depending on the value of npp.
2312 * npp == NULL --> just do the lookup
2313 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2314 * handled too
2315 * *npp != NULL --> update the file handle in the vnode
2316 */
2317static int
2318nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2319 struct thread *td, struct nfsnode **npp)
2320{
2321 struct vnode *newvp = NULL, *vp;
2322 struct nfsnode *np, *dnp = VTONFS(dvp);
2323 struct nfsfh *nfhp, *onfhp;
2324 struct nfsvattr nfsva, dnfsva;
2325 struct componentname cn;
2326 int error = 0, attrflag, dattrflag;
2327 u_int hash;
2328
2329 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2330 &nfhp, &attrflag, &dattrflag, NULL);
2331 if (dattrflag)
2332 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2333 if (npp && !error) {
2334 if (*npp != NULL) {
2335 np = *npp;
2336 vp = NFSTOV(np);
2337 /*
2338 * For NFSv4, check to see if it is the same name and
2339 * replace the name, if it is different.
2340 */
2341 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2342 (np->n_v4->n4_namelen != len ||
2343 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2344 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2345 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2346 dnp->n_fhp->nfh_len))) {
2347#ifdef notdef
2348{ char nnn[100]; int nnnl;
2349nnnl = (len < 100) ? len : 99;
2350bcopy(name, nnn, nnnl);
2351nnn[nnnl] = '\0';
2352printf("replace=%s\n",nnn);
2353}
2354#endif
2355 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2356 MALLOC(np->n_v4, struct nfsv4node *,
2357 sizeof (struct nfsv4node) +
2358 dnp->n_fhp->nfh_len + len - 1,
2359 M_NFSV4NODE, M_WAITOK);
2360 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2361 np->n_v4->n4_namelen = len;
2362 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2363 dnp->n_fhp->nfh_len);
2364 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2365 }
2366 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2367 FNV1_32_INIT);
2368 onfhp = np->n_fhp;
2369 /*
2370 * Rehash node for new file handle.
2371 */
2372 vfs_hash_rehash(vp, hash);
2373 np->n_fhp = nfhp;
2374 if (onfhp != NULL)
2375 FREE((caddr_t)onfhp, M_NFSFH);
2376 newvp = NFSTOV(np);
2377 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2378 FREE((caddr_t)nfhp, M_NFSFH);
2379 VREF(dvp);
2380 newvp = dvp;
2381 } else {
2382 cn.cn_nameptr = name;
2383 cn.cn_namelen = len;
2384 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2385 &np, NULL);
2386 if (error)
2387 return (error);
2388 newvp = NFSTOV(np);
2389 }
2390 if (!attrflag && *npp == NULL) {
2391 vrele(newvp);
2392 return (ENOENT);
2393 }
2394 if (attrflag)
2395 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2396 0, 1);
2397 }
2398 if (npp && *npp == NULL) {
2399 if (error) {
2400 if (newvp) {
2401 if (newvp == dvp)
2402 vrele(newvp);
2403 else
2404 vput(newvp);
2405 }
2406 } else
2407 *npp = np;
2408 }
2409 if (error && NFS_ISV4(dvp))
2410 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2411 return (error);
2412}
2413
2414/*
2415 * Nfs Version 3 and 4 commit rpc
2416 */
2417int
2418ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2419 struct thread *td)
2420{
2421 struct nfsvattr nfsva;
2422 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2423 int error, attrflag;
2424 u_char verf[NFSX_VERF];
2425
2426 mtx_lock(&nmp->nm_mtx);
2427 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2428 mtx_unlock(&nmp->nm_mtx);
2429 return (0);
2430 }
2431 mtx_unlock(&nmp->nm_mtx);
2432 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2433 &attrflag, NULL);
2434 if (!error) {
2435 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2436 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2437 error = NFSERR_STALEWRITEVERF;
2438 }
2439 if (!error && attrflag)
2440 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2441 0, 1);
2442 } else if (NFS_ISV4(vp)) {
2443 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2444 }
2445 return (error);
2446}
2447
2448/*
2449 * Strategy routine.
2450 * For async requests when nfsiod(s) are running, queue the request by
2451 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2452 * request.
2453 */
2454static int
2455nfs_strategy(struct vop_strategy_args *ap)
2456{
2457 struct buf *bp = ap->a_bp;
2458 struct ucred *cr;
2459
2460 KASSERT(!(bp->b_flags & B_DONE),
2461 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2462 BUF_ASSERT_HELD(bp);
2463
2464 if (bp->b_iocmd == BIO_READ)
2465 cr = bp->b_rcred;
2466 else
2467 cr = bp->b_wcred;
2468
2469 /*
2470 * If the op is asynchronous and an i/o daemon is waiting
2471 * queue the request, wake it up and wait for completion
2472 * otherwise just do it ourselves.
2473 */
2474 if ((bp->b_flags & B_ASYNC) == 0 ||
2475 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
| 1319 NFSLOCKMNT(nmp);
1320 if (!error && NFSHASWRITEVERF(nmp) &&
1321 NFSBCMP(verf, nmp->nm_verf, NFSX_VERF)) {
1322 *must_commit = 1;
1323 NFSBCOPY(verf, nmp->nm_verf, NFSX_VERF);
1324 }
1325 NFSUNLOCKMNT(nmp);
1326 if (attrflag) {
1327 if (VTONFS(vp)->n_flag & ND_NFSV4)
1328 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1329 1);
1330 else
1331 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1332 1);
1333 if (ret && !error)
1334 error = ret;
1335 }
1336 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1337 *iomode = NFSWRITE_FILESYNC;
1338 if (error && NFS_ISV4(vp))
1339 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1340 return (error);
1341}
1342
1343/*
1344 * nfs mknod rpc
1345 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1346 * mode set to specify the file type and the size field for rdev.
1347 */
1348static int
1349nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1350 struct vattr *vap)
1351{
1352 struct nfsvattr nfsva, dnfsva;
1353 struct vnode *newvp = NULL;
1354 struct nfsnode *np = NULL, *dnp;
1355 struct nfsfh *nfhp;
1356 struct vattr vattr;
1357 int error = 0, attrflag, dattrflag;
1358 u_int32_t rdev;
1359
1360 if (vap->va_type == VCHR || vap->va_type == VBLK)
1361 rdev = vap->va_rdev;
1362 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1363 rdev = 0xffffffff;
1364 else
1365 return (EOPNOTSUPP);
1366 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1367 return (error);
1368 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1369 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1370 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1371 if (!error) {
1372 if (!nfhp)
1373 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1374 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1375 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1376 NULL);
1377 if (nfhp)
1378 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1379 cnp->cn_thread, &np, NULL);
1380 }
1381 if (dattrflag)
1382 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1383 if (!error) {
1384 newvp = NFSTOV(np);
1385 if (attrflag)
1386 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1387 0, 1);
1388 }
1389 if (!error) {
1390 if ((cnp->cn_flags & MAKEENTRY))
1391 cache_enter(dvp, newvp, cnp);
1392 *vpp = newvp;
1393 } else if (NFS_ISV4(dvp)) {
1394 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1395 vap->va_gid);
1396 }
1397 dnp = VTONFS(dvp);
1398 mtx_lock(&dnp->n_mtx);
1399 dnp->n_flag |= NMODIFIED;
1400 if (!dattrflag)
1401 dnp->n_attrstamp = 0;
1402 mtx_unlock(&dnp->n_mtx);
1403 return (error);
1404}
1405
1406/*
1407 * nfs mknod vop
1408 * just call nfs_mknodrpc() to do the work.
1409 */
1410/* ARGSUSED */
1411static int
1412nfs_mknod(struct vop_mknod_args *ap)
1413{
1414 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1415}
1416
1417static struct mtx nfs_cverf_mtx;
1418MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1419 MTX_DEF);
1420
1421static nfsquad_t
1422nfs_get_cverf(void)
1423{
1424 static nfsquad_t cverf;
1425 nfsquad_t ret;
1426 static int cverf_initialized = 0;
1427
1428 mtx_lock(&nfs_cverf_mtx);
1429 if (cverf_initialized == 0) {
1430 cverf.lval[0] = arc4random();
1431 cverf.lval[1] = arc4random();
1432 cverf_initialized = 1;
1433 } else
1434 cverf.qval++;
1435 ret = cverf;
1436 mtx_unlock(&nfs_cverf_mtx);
1437
1438 return (ret);
1439}
1440
1441/*
1442 * nfs file create call
1443 */
1444static int
1445nfs_create(struct vop_create_args *ap)
1446{
1447 struct vnode *dvp = ap->a_dvp;
1448 struct vattr *vap = ap->a_vap;
1449 struct componentname *cnp = ap->a_cnp;
1450 struct nfsnode *np = NULL, *dnp;
1451 struct vnode *newvp = NULL;
1452 struct nfsmount *nmp;
1453 struct nfsvattr dnfsva, nfsva;
1454 struct nfsfh *nfhp;
1455 nfsquad_t cverf;
1456 int error = 0, attrflag, dattrflag, fmode = 0;
1457 struct vattr vattr;
1458
1459 /*
1460 * Oops, not for me..
1461 */
1462 if (vap->va_type == VSOCK)
1463 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1464
1465 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1466 return (error);
1467 if (vap->va_vaflags & VA_EXCLUSIVE)
1468 fmode |= O_EXCL;
1469 dnp = VTONFS(dvp);
1470 nmp = VFSTONFS(vnode_mount(dvp));
1471again:
1472 /* For NFSv4, wait until any remove is done. */
1473 mtx_lock(&dnp->n_mtx);
1474 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1475 dnp->n_flag |= NREMOVEWANT;
1476 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1477 }
1478 mtx_unlock(&dnp->n_mtx);
1479
1480 cverf = nfs_get_cverf();
1481 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1482 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1483 &nfhp, &attrflag, &dattrflag, NULL);
1484 if (!error) {
1485 if (nfhp == NULL)
1486 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1487 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1488 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1489 NULL);
1490 if (nfhp != NULL)
1491 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1492 cnp->cn_thread, &np, NULL);
1493 }
1494 if (dattrflag)
1495 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1496 if (!error) {
1497 newvp = NFSTOV(np);
1498 if (attrflag)
1499 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1500 0, 1);
1501 }
1502 if (error) {
1503 if (newvp != NULL) {
1504 vrele(newvp);
1505 newvp = NULL;
1506 }
1507 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1508 error == NFSERR_NOTSUPP) {
1509 fmode &= ~O_EXCL;
1510 goto again;
1511 }
1512 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1513 if (nfscl_checksattr(vap, &nfsva)) {
1514 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1515 cnp->cn_thread, &nfsva, &attrflag, NULL);
1516 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1517 vap->va_gid != (gid_t)VNOVAL)) {
1518 /* try again without setting uid/gid */
1519 vap->va_uid = (uid_t)VNOVAL;
1520 vap->va_gid = (uid_t)VNOVAL;
1521 error = nfsrpc_setattr(newvp, vap, NULL,
1522 cnp->cn_cred, cnp->cn_thread, &nfsva,
1523 &attrflag, NULL);
1524 }
1525 if (attrflag)
1526 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1527 NULL, 0, 1);
1528 }
1529 }
1530 if (!error) {
1531 if (cnp->cn_flags & MAKEENTRY)
1532 cache_enter(dvp, newvp, cnp);
1533 *ap->a_vpp = newvp;
1534 } else if (NFS_ISV4(dvp)) {
1535 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1536 vap->va_gid);
1537 }
1538 mtx_lock(&dnp->n_mtx);
1539 dnp->n_flag |= NMODIFIED;
1540 if (!dattrflag)
1541 dnp->n_attrstamp = 0;
1542 mtx_unlock(&dnp->n_mtx);
1543 return (error);
1544}
1545
1546/*
1547 * nfs file remove call
1548 * To try and make nfs semantics closer to ufs semantics, a file that has
1549 * other processes using the vnode is renamed instead of removed and then
1550 * removed later on the last close.
1551 * - If v_usecount > 1
1552 * If a rename is not already in the works
1553 * call nfs_sillyrename() to set it up
1554 * else
1555 * do the remove rpc
1556 */
1557static int
1558nfs_remove(struct vop_remove_args *ap)
1559{
1560 struct vnode *vp = ap->a_vp;
1561 struct vnode *dvp = ap->a_dvp;
1562 struct componentname *cnp = ap->a_cnp;
1563 struct nfsnode *np = VTONFS(vp);
1564 int error = 0;
1565 struct vattr vattr;
1566
1567#ifndef DIAGNOSTIC
1568 if ((cnp->cn_flags & HASBUF) == 0)
1569 panic("nfs_remove: no name");
1570 if (vrefcnt(vp) < 1)
1571 panic("nfs_remove: bad v_usecount");
1572#endif
1573 if (vp->v_type == VDIR)
1574 error = EPERM;
1575 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1576 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1577 vattr.va_nlink > 1)) {
1578 /*
1579 * Purge the name cache so that the chance of a lookup for
1580 * the name succeeding while the remove is in progress is
1581 * minimized. Without node locking it can still happen, such
1582 * that an I/O op returns ESTALE, but since you get this if
1583 * another host removes the file..
1584 */
1585 cache_purge(vp);
1586 /*
1587 * throw away biocache buffers, mainly to avoid
1588 * unnecessary delayed writes later.
1589 */
1590 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1591 /* Do the rpc */
1592 if (error != EINTR && error != EIO)
1593 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1594 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1595 /*
1596 * Kludge City: If the first reply to the remove rpc is lost..
1597 * the reply to the retransmitted request will be ENOENT
1598 * since the file was in fact removed
1599 * Therefore, we cheat and return success.
1600 */
1601 if (error == ENOENT)
1602 error = 0;
1603 } else if (!np->n_sillyrename)
1604 error = nfs_sillyrename(dvp, vp, cnp);
1605 np->n_attrstamp = 0;
1606 return (error);
1607}
1608
1609/*
1610 * nfs file remove rpc called from nfs_inactive
1611 */
1612int
1613ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1614{
1615 /*
1616 * Make sure that the directory vnode is still valid.
1617 * XXX we should lock sp->s_dvp here.
1618 */
1619 if (sp->s_dvp->v_type == VBAD)
1620 return (0);
1621 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1622 sp->s_cred, NULL));
1623}
1624
1625/*
1626 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1627 */
1628static int
1629nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1630 int namelen, struct ucred *cred, struct thread *td)
1631{
1632 struct nfsvattr dnfsva;
1633 struct nfsnode *dnp = VTONFS(dvp);
1634 int error = 0, dattrflag;
1635
1636 mtx_lock(&dnp->n_mtx);
1637 dnp->n_flag |= NREMOVEINPROG;
1638 mtx_unlock(&dnp->n_mtx);
1639 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1640 &dattrflag, NULL);
1641 mtx_lock(&dnp->n_mtx);
1642 if ((dnp->n_flag & NREMOVEWANT)) {
1643 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1644 mtx_unlock(&dnp->n_mtx);
1645 wakeup((caddr_t)dnp);
1646 } else {
1647 dnp->n_flag &= ~NREMOVEINPROG;
1648 mtx_unlock(&dnp->n_mtx);
1649 }
1650 if (dattrflag)
1651 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1652 mtx_lock(&dnp->n_mtx);
1653 dnp->n_flag |= NMODIFIED;
1654 if (!dattrflag)
1655 dnp->n_attrstamp = 0;
1656 mtx_unlock(&dnp->n_mtx);
1657 if (error && NFS_ISV4(dvp))
1658 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1659 return (error);
1660}
1661
1662/*
1663 * nfs file rename call
1664 */
1665static int
1666nfs_rename(struct vop_rename_args *ap)
1667{
1668 struct vnode *fvp = ap->a_fvp;
1669 struct vnode *tvp = ap->a_tvp;
1670 struct vnode *fdvp = ap->a_fdvp;
1671 struct vnode *tdvp = ap->a_tdvp;
1672 struct componentname *tcnp = ap->a_tcnp;
1673 struct componentname *fcnp = ap->a_fcnp;
1674 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1675 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1676 struct nfsv4node *newv4 = NULL;
1677 int error;
1678
1679#ifndef DIAGNOSTIC
1680 if ((tcnp->cn_flags & HASBUF) == 0 ||
1681 (fcnp->cn_flags & HASBUF) == 0)
1682 panic("nfs_rename: no name");
1683#endif
1684 /* Check for cross-device rename */
1685 if ((fvp->v_mount != tdvp->v_mount) ||
1686 (tvp && (fvp->v_mount != tvp->v_mount))) {
1687 error = EXDEV;
1688 goto out;
1689 }
1690
1691 if (fvp == tvp) {
1692 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1693 error = 0;
1694 goto out;
1695 }
1696 if ((error = vn_lock(fvp, LK_EXCLUSIVE)))
1697 goto out;
1698
1699 /*
1700 * We have to flush B_DELWRI data prior to renaming
1701 * the file. If we don't, the delayed-write buffers
1702 * can be flushed out later after the file has gone stale
1703 * under NFSV3. NFSV2 does not have this problem because
1704 * ( as far as I can tell ) it flushes dirty buffers more
1705 * often.
1706 *
1707 * Skip the rename operation if the fsync fails, this can happen
1708 * due to the server's volume being full, when we pushed out data
1709 * that was written back to our cache earlier. Not checking for
1710 * this condition can result in potential (silent) data loss.
1711 */
1712 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1713 VOP_UNLOCK(fvp, 0);
1714 if (!error && tvp)
1715 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1716 if (error)
1717 goto out;
1718
1719 /*
1720 * If the tvp exists and is in use, sillyrename it before doing the
1721 * rename of the new file over it.
1722 * XXX Can't sillyrename a directory.
1723 */
1724 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1725 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1726 vput(tvp);
1727 tvp = NULL;
1728 }
1729
1730 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1731 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1732 tcnp->cn_thread);
1733
1734 if (!error) {
1735 /*
1736 * For NFSv4, check to see if it is the same name and
1737 * replace the name, if it is different.
1738 */
1739 MALLOC(newv4, struct nfsv4node *,
1740 sizeof (struct nfsv4node) +
1741 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1742 M_NFSV4NODE, M_WAITOK);
1743 mtx_lock(&tdnp->n_mtx);
1744 mtx_lock(&fnp->n_mtx);
1745 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1746 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1747 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1748 tcnp->cn_namelen) ||
1749 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1750 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1751 tdnp->n_fhp->nfh_len))) {
1752#ifdef notdef
1753{ char nnn[100]; int nnnl;
1754nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1755bcopy(tcnp->cn_nameptr, nnn, nnnl);
1756nnn[nnnl] = '\0';
1757printf("ren replace=%s\n",nnn);
1758}
1759#endif
1760 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1761 fnp->n_v4 = newv4;
1762 newv4 = NULL;
1763 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1764 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1765 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1766 tdnp->n_fhp->nfh_len);
1767 NFSBCOPY(tcnp->cn_nameptr,
1768 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1769 }
1770 mtx_unlock(&tdnp->n_mtx);
1771 mtx_unlock(&fnp->n_mtx);
1772 if (newv4 != NULL)
1773 FREE((caddr_t)newv4, M_NFSV4NODE);
1774 }
1775
1776 if (fvp->v_type == VDIR) {
1777 if (tvp != NULL && tvp->v_type == VDIR)
1778 cache_purge(tdvp);
1779 cache_purge(fdvp);
1780 }
1781
1782out:
1783 if (tdvp == tvp)
1784 vrele(tdvp);
1785 else
1786 vput(tdvp);
1787 if (tvp)
1788 vput(tvp);
1789 vrele(fdvp);
1790 vrele(fvp);
1791 /*
1792 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1793 */
1794 if (error == ENOENT)
1795 error = 0;
1796 return (error);
1797}
1798
1799/*
1800 * nfs file rename rpc called from nfs_remove() above
1801 */
1802static int
1803nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1804 struct sillyrename *sp)
1805{
1806
1807 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1808 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1809 scnp->cn_thread));
1810}
1811
1812/*
1813 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1814 */
1815static int
1816nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1817 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1818 int tnamelen, struct ucred *cred, struct thread *td)
1819{
1820 struct nfsvattr fnfsva, tnfsva;
1821 struct nfsnode *fdnp = VTONFS(fdvp);
1822 struct nfsnode *tdnp = VTONFS(tdvp);
1823 int error = 0, fattrflag, tattrflag;
1824
1825 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1826 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1827 &tattrflag, NULL, NULL);
1828 mtx_lock(&fdnp->n_mtx);
1829 fdnp->n_flag |= NMODIFIED;
1830 mtx_unlock(&fdnp->n_mtx);
1831 mtx_lock(&tdnp->n_mtx);
1832 tdnp->n_flag |= NMODIFIED;
1833 mtx_unlock(&tdnp->n_mtx);
1834 if (fattrflag)
1835 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1836 else
1837 fdnp->n_attrstamp = 0;
1838 if (tattrflag)
1839 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1840 else
1841 tdnp->n_attrstamp = 0;
1842 if (error && NFS_ISV4(fdvp))
1843 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1844 return (error);
1845}
1846
1847/*
1848 * nfs hard link create call
1849 */
1850static int
1851nfs_link(struct vop_link_args *ap)
1852{
1853 struct vnode *vp = ap->a_vp;
1854 struct vnode *tdvp = ap->a_tdvp;
1855 struct componentname *cnp = ap->a_cnp;
1856 struct nfsnode *tdnp;
1857 struct nfsvattr nfsva, dnfsva;
1858 int error = 0, attrflag, dattrflag;
1859
1860 if (vp->v_mount != tdvp->v_mount) {
1861 return (EXDEV);
1862 }
1863
1864 /*
1865 * Push all writes to the server, so that the attribute cache
1866 * doesn't get "out of sync" with the server.
1867 * XXX There should be a better way!
1868 */
1869 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1870
1871 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1872 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1873 &dattrflag, NULL);
1874 tdnp = VTONFS(tdvp);
1875 mtx_lock(&tdnp->n_mtx);
1876 tdnp->n_flag |= NMODIFIED;
1877 mtx_unlock(&tdnp->n_mtx);
1878 if (attrflag)
1879 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1880 else
1881 VTONFS(vp)->n_attrstamp = 0;
1882 if (dattrflag)
1883 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1884 else
1885 tdnp->n_attrstamp = 0;
1886 /*
1887 * If negative lookup caching is enabled, I might as well
1888 * add an entry for this node. Not necessary for correctness,
1889 * but if negative caching is enabled, then the system
1890 * must care about lookup caching hit rate, so...
1891 */
1892 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1893 (cnp->cn_flags & MAKEENTRY))
1894 cache_enter(tdvp, vp, cnp);
1895 if (error && NFS_ISV4(vp))
1896 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1897 (gid_t)0);
1898 return (error);
1899}
1900
1901/*
1902 * nfs symbolic link create call
1903 */
1904static int
1905nfs_symlink(struct vop_symlink_args *ap)
1906{
1907 struct vnode *dvp = ap->a_dvp;
1908 struct vattr *vap = ap->a_vap;
1909 struct componentname *cnp = ap->a_cnp;
1910 struct nfsvattr nfsva, dnfsva;
1911 struct nfsfh *nfhp;
1912 struct nfsnode *np = NULL, *dnp;
1913 struct vnode *newvp = NULL;
1914 int error = 0, attrflag, dattrflag, ret;
1915
1916 vap->va_type = VLNK;
1917 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1918 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1919 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1920 if (nfhp) {
1921 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
1922 &np, NULL);
1923 if (!ret)
1924 newvp = NFSTOV(np);
1925 else if (!error)
1926 error = ret;
1927 }
1928 if (newvp != NULL) {
1929 if (attrflag)
1930 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1931 0, 1);
1932 } else if (!error) {
1933 /*
1934 * If we do not have an error and we could not extract the
1935 * newvp from the response due to the request being NFSv2, we
1936 * have to do a lookup in order to obtain a newvp to return.
1937 */
1938 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1939 cnp->cn_cred, cnp->cn_thread, &np);
1940 if (!error)
1941 newvp = NFSTOV(np);
1942 }
1943 if (error) {
1944 if (newvp)
1945 vput(newvp);
1946 if (NFS_ISV4(dvp))
1947 error = nfscl_maperr(cnp->cn_thread, error,
1948 vap->va_uid, vap->va_gid);
1949 } else {
1950 /*
1951 * If negative lookup caching is enabled, I might as well
1952 * add an entry for this node. Not necessary for correctness,
1953 * but if negative caching is enabled, then the system
1954 * must care about lookup caching hit rate, so...
1955 */
1956 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
1957 (cnp->cn_flags & MAKEENTRY))
1958 cache_enter(dvp, newvp, cnp);
1959 *ap->a_vpp = newvp;
1960 }
1961
1962 dnp = VTONFS(dvp);
1963 mtx_lock(&dnp->n_mtx);
1964 dnp->n_flag |= NMODIFIED;
1965 mtx_unlock(&dnp->n_mtx);
1966 if (dattrflag)
1967 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1968 else
1969 dnp->n_attrstamp = 0;
1970 return (error);
1971}
1972
1973/*
1974 * nfs make dir call
1975 */
1976static int
1977nfs_mkdir(struct vop_mkdir_args *ap)
1978{
1979 struct vnode *dvp = ap->a_dvp;
1980 struct vattr *vap = ap->a_vap;
1981 struct componentname *cnp = ap->a_cnp;
1982 struct nfsnode *np = NULL, *dnp;
1983 struct vnode *newvp = NULL;
1984 struct vattr vattr;
1985 struct nfsfh *nfhp;
1986 struct nfsvattr nfsva, dnfsva;
1987 int error = 0, attrflag, dattrflag, ret;
1988
1989 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1990 return (error);
1991 vap->va_type = VDIR;
1992 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1993 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
1994 &attrflag, &dattrflag, NULL);
1995 dnp = VTONFS(dvp);
1996 mtx_lock(&dnp->n_mtx);
1997 dnp->n_flag |= NMODIFIED;
1998 mtx_unlock(&dnp->n_mtx);
1999 if (dattrflag)
2000 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2001 else
2002 dnp->n_attrstamp = 0;
2003 if (nfhp) {
2004 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2005 &np, NULL);
2006 if (!ret) {
2007 newvp = NFSTOV(np);
2008 if (attrflag)
2009 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2010 NULL, 0, 1);
2011 } else if (!error)
2012 error = ret;
2013 }
2014 if (!error && newvp == NULL) {
2015 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2016 cnp->cn_cred, cnp->cn_thread, &np);
2017 if (!error) {
2018 newvp = NFSTOV(np);
2019 if (newvp->v_type != VDIR)
2020 error = EEXIST;
2021 }
2022 }
2023 if (error) {
2024 if (newvp)
2025 vput(newvp);
2026 if (NFS_ISV4(dvp))
2027 error = nfscl_maperr(cnp->cn_thread, error,
2028 vap->va_uid, vap->va_gid);
2029 } else {
2030 /*
2031 * If negative lookup caching is enabled, I might as well
2032 * add an entry for this node. Not necessary for correctness,
2033 * but if negative caching is enabled, then the system
2034 * must care about lookup caching hit rate, so...
2035 */
2036 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2037 (cnp->cn_flags & MAKEENTRY))
2038 cache_enter(dvp, newvp, cnp);
2039 *ap->a_vpp = newvp;
2040 }
2041 return (error);
2042}
2043
2044/*
2045 * nfs remove directory call
2046 */
2047static int
2048nfs_rmdir(struct vop_rmdir_args *ap)
2049{
2050 struct vnode *vp = ap->a_vp;
2051 struct vnode *dvp = ap->a_dvp;
2052 struct componentname *cnp = ap->a_cnp;
2053 struct nfsnode *dnp;
2054 struct nfsvattr dnfsva;
2055 int error, dattrflag;
2056
2057 if (dvp == vp)
2058 return (EINVAL);
2059 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2060 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2061 dnp = VTONFS(dvp);
2062 mtx_lock(&dnp->n_mtx);
2063 dnp->n_flag |= NMODIFIED;
2064 mtx_unlock(&dnp->n_mtx);
2065 if (dattrflag)
2066 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2067 else
2068 dnp->n_attrstamp = 0;
2069
2070 cache_purge(dvp);
2071 cache_purge(vp);
2072 if (error && NFS_ISV4(dvp))
2073 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2074 (gid_t)0);
2075 /*
2076 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2077 */
2078 if (error == ENOENT)
2079 error = 0;
2080 return (error);
2081}
2082
2083/*
2084 * nfs readdir call
2085 */
2086static int
2087nfs_readdir(struct vop_readdir_args *ap)
2088{
2089 struct vnode *vp = ap->a_vp;
2090 struct nfsnode *np = VTONFS(vp);
2091 struct uio *uio = ap->a_uio;
2092 int tresid, error = 0;
2093 struct vattr vattr;
2094
2095 if (vp->v_type != VDIR)
2096 return(EPERM);
2097
2098 /*
2099 * First, check for hit on the EOF offset cache
2100 */
2101 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2102 (np->n_flag & NMODIFIED) == 0) {
2103 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2104 mtx_lock(&np->n_mtx);
2105 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2106 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2107 mtx_unlock(&np->n_mtx);
2108 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2109 return (0);
2110 } else
2111 mtx_unlock(&np->n_mtx);
2112 }
2113 }
2114
2115 /*
2116 * Call ncl_bioread() to do the real work.
2117 */
2118 tresid = uio->uio_resid;
2119 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2120
2121 if (!error && uio->uio_resid == tresid)
2122 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2123 return (error);
2124}
2125
2126/*
2127 * Readdir rpc call.
2128 * Called from below the buffer cache by ncl_doio().
2129 */
2130int
2131ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2132 struct thread *td)
2133{
2134 struct nfsvattr nfsva;
2135 nfsuint64 *cookiep, cookie;
2136 struct nfsnode *dnp = VTONFS(vp);
2137 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2138 int error = 0, eof, attrflag;
2139
2140#ifndef DIAGNOSTIC
2141 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2142 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2143 panic("nfs readdirrpc bad uio");
2144#endif
2145
2146 /*
2147 * If there is no cookie, assume directory was stale.
2148 */
2149 ncl_dircookie_lock(dnp);
2150 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2151 if (cookiep) {
2152 cookie = *cookiep;
2153 ncl_dircookie_unlock(dnp);
2154 } else {
2155 ncl_dircookie_unlock(dnp);
2156 return (NFSERR_BAD_COOKIE);
2157 }
2158
2159 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2160 (void)ncl_fsinfo(nmp, vp, cred, td);
2161
2162 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2163 &attrflag, &eof, NULL);
2164 if (attrflag)
2165 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2166
2167 if (!error) {
2168 /*
2169 * We are now either at the end of the directory or have filled
2170 * the block.
2171 */
2172 if (eof)
2173 dnp->n_direofoffset = uiop->uio_offset;
2174 else {
2175 if (uiop->uio_resid > 0)
2176 ncl_printf("EEK! readdirrpc resid > 0\n");
2177 ncl_dircookie_lock(dnp);
2178 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2179 *cookiep = cookie;
2180 ncl_dircookie_unlock(dnp);
2181 }
2182 } else if (NFS_ISV4(vp)) {
2183 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2184 }
2185 return (error);
2186}
2187
2188/*
2189 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2190 */
2191int
2192ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2193 struct thread *td)
2194{
2195 struct nfsvattr nfsva;
2196 nfsuint64 *cookiep, cookie;
2197 struct nfsnode *dnp = VTONFS(vp);
2198 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2199 int error = 0, attrflag, eof;
2200
2201#ifndef DIAGNOSTIC
2202 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2203 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2204 panic("nfs readdirplusrpc bad uio");
2205#endif
2206
2207 /*
2208 * If there is no cookie, assume directory was stale.
2209 */
2210 ncl_dircookie_lock(dnp);
2211 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2212 if (cookiep) {
2213 cookie = *cookiep;
2214 ncl_dircookie_unlock(dnp);
2215 } else {
2216 ncl_dircookie_unlock(dnp);
2217 return (NFSERR_BAD_COOKIE);
2218 }
2219
2220 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2221 (void)ncl_fsinfo(nmp, vp, cred, td);
2222 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2223 &attrflag, &eof, NULL);
2224 if (attrflag)
2225 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2226
2227 if (!error) {
2228 /*
2229 * We are now either at end of the directory or have filled the
2230 * the block.
2231 */
2232 if (eof)
2233 dnp->n_direofoffset = uiop->uio_offset;
2234 else {
2235 if (uiop->uio_resid > 0)
2236 ncl_printf("EEK! readdirplusrpc resid > 0\n");
2237 ncl_dircookie_lock(dnp);
2238 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2239 *cookiep = cookie;
2240 ncl_dircookie_unlock(dnp);
2241 }
2242 } else if (NFS_ISV4(vp)) {
2243 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2244 }
2245 return (error);
2246}
2247
2248/*
2249 * Silly rename. To make the NFS filesystem that is stateless look a little
2250 * more like the "ufs" a remove of an active vnode is translated to a rename
2251 * to a funny looking filename that is removed by nfs_inactive on the
2252 * nfsnode. There is the potential for another process on a different client
2253 * to create the same funny name between the nfs_lookitup() fails and the
2254 * nfs_rename() completes, but...
2255 */
2256static int
2257nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2258{
2259 struct sillyrename *sp;
2260 struct nfsnode *np;
2261 int error;
2262 short pid;
2263 unsigned int lticks;
2264
2265 cache_purge(dvp);
2266 np = VTONFS(vp);
2267#ifndef DIAGNOSTIC
2268 if (vp->v_type == VDIR)
2269 panic("nfs: sillyrename dir");
2270#endif
2271 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2272 M_NEWNFSREQ, M_WAITOK);
2273 sp->s_cred = crhold(cnp->cn_cred);
2274 sp->s_dvp = dvp;
2275 VREF(dvp);
2276
2277 /*
2278 * Fudge together a funny name.
2279 * Changing the format of the funny name to accomodate more
2280 * sillynames per directory.
2281 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2282 * CPU ticks since boot.
2283 */
2284 pid = cnp->cn_thread->td_proc->p_pid;
2285 lticks = (unsigned int)ticks;
2286 for ( ; ; ) {
2287 sp->s_namlen = sprintf(sp->s_name,
2288 ".nfs.%08x.%04x4.4", lticks,
2289 pid);
2290 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2291 cnp->cn_thread, NULL))
2292 break;
2293 lticks++;
2294 }
2295 error = nfs_renameit(dvp, vp, cnp, sp);
2296 if (error)
2297 goto bad;
2298 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2299 cnp->cn_thread, &np);
2300 np->n_sillyrename = sp;
2301 return (0);
2302bad:
2303 vrele(sp->s_dvp);
2304 crfree(sp->s_cred);
2305 free((caddr_t)sp, M_NEWNFSREQ);
2306 return (error);
2307}
2308
2309/*
2310 * Look up a file name and optionally either update the file handle or
2311 * allocate an nfsnode, depending on the value of npp.
2312 * npp == NULL --> just do the lookup
2313 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2314 * handled too
2315 * *npp != NULL --> update the file handle in the vnode
2316 */
2317static int
2318nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2319 struct thread *td, struct nfsnode **npp)
2320{
2321 struct vnode *newvp = NULL, *vp;
2322 struct nfsnode *np, *dnp = VTONFS(dvp);
2323 struct nfsfh *nfhp, *onfhp;
2324 struct nfsvattr nfsva, dnfsva;
2325 struct componentname cn;
2326 int error = 0, attrflag, dattrflag;
2327 u_int hash;
2328
2329 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2330 &nfhp, &attrflag, &dattrflag, NULL);
2331 if (dattrflag)
2332 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2333 if (npp && !error) {
2334 if (*npp != NULL) {
2335 np = *npp;
2336 vp = NFSTOV(np);
2337 /*
2338 * For NFSv4, check to see if it is the same name and
2339 * replace the name, if it is different.
2340 */
2341 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2342 (np->n_v4->n4_namelen != len ||
2343 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2344 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2345 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2346 dnp->n_fhp->nfh_len))) {
2347#ifdef notdef
2348{ char nnn[100]; int nnnl;
2349nnnl = (len < 100) ? len : 99;
2350bcopy(name, nnn, nnnl);
2351nnn[nnnl] = '\0';
2352printf("replace=%s\n",nnn);
2353}
2354#endif
2355 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2356 MALLOC(np->n_v4, struct nfsv4node *,
2357 sizeof (struct nfsv4node) +
2358 dnp->n_fhp->nfh_len + len - 1,
2359 M_NFSV4NODE, M_WAITOK);
2360 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2361 np->n_v4->n4_namelen = len;
2362 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2363 dnp->n_fhp->nfh_len);
2364 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2365 }
2366 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2367 FNV1_32_INIT);
2368 onfhp = np->n_fhp;
2369 /*
2370 * Rehash node for new file handle.
2371 */
2372 vfs_hash_rehash(vp, hash);
2373 np->n_fhp = nfhp;
2374 if (onfhp != NULL)
2375 FREE((caddr_t)onfhp, M_NFSFH);
2376 newvp = NFSTOV(np);
2377 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2378 FREE((caddr_t)nfhp, M_NFSFH);
2379 VREF(dvp);
2380 newvp = dvp;
2381 } else {
2382 cn.cn_nameptr = name;
2383 cn.cn_namelen = len;
2384 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2385 &np, NULL);
2386 if (error)
2387 return (error);
2388 newvp = NFSTOV(np);
2389 }
2390 if (!attrflag && *npp == NULL) {
2391 vrele(newvp);
2392 return (ENOENT);
2393 }
2394 if (attrflag)
2395 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2396 0, 1);
2397 }
2398 if (npp && *npp == NULL) {
2399 if (error) {
2400 if (newvp) {
2401 if (newvp == dvp)
2402 vrele(newvp);
2403 else
2404 vput(newvp);
2405 }
2406 } else
2407 *npp = np;
2408 }
2409 if (error && NFS_ISV4(dvp))
2410 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2411 return (error);
2412}
2413
2414/*
2415 * Nfs Version 3 and 4 commit rpc
2416 */
2417int
2418ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2419 struct thread *td)
2420{
2421 struct nfsvattr nfsva;
2422 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2423 int error, attrflag;
2424 u_char verf[NFSX_VERF];
2425
2426 mtx_lock(&nmp->nm_mtx);
2427 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2428 mtx_unlock(&nmp->nm_mtx);
2429 return (0);
2430 }
2431 mtx_unlock(&nmp->nm_mtx);
2432 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2433 &attrflag, NULL);
2434 if (!error) {
2435 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2436 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2437 error = NFSERR_STALEWRITEVERF;
2438 }
2439 if (!error && attrflag)
2440 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2441 0, 1);
2442 } else if (NFS_ISV4(vp)) {
2443 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2444 }
2445 return (error);
2446}
2447
2448/*
2449 * Strategy routine.
2450 * For async requests when nfsiod(s) are running, queue the request by
2451 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2452 * request.
2453 */
2454static int
2455nfs_strategy(struct vop_strategy_args *ap)
2456{
2457 struct buf *bp = ap->a_bp;
2458 struct ucred *cr;
2459
2460 KASSERT(!(bp->b_flags & B_DONE),
2461 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2462 BUF_ASSERT_HELD(bp);
2463
2464 if (bp->b_iocmd == BIO_READ)
2465 cr = bp->b_rcred;
2466 else
2467 cr = bp->b_wcred;
2468
2469 /*
2470 * If the op is asynchronous and an i/o daemon is waiting
2471 * queue the request, wake it up and wait for completion
2472 * otherwise just do it ourselves.
2473 */
2474 if ((bp->b_flags & B_ASYNC) == 0 ||
2475 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
|
2476 (void)ncl_doio(ap->a_vp, bp, cr, curthread);
| 2476 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
|
2477 return (0);
2478}
2479
2480/*
2481 * fsync vnode op. Just call ncl_flush() with commit == 1.
2482 */
2483/* ARGSUSED */
2484static int
2485nfs_fsync(struct vop_fsync_args *ap)
2486{
| 2477 return (0);
2478}
2479
2480/*
2481 * fsync vnode op. Just call ncl_flush() with commit == 1.
2482 */
2483/* ARGSUSED */
2484static int
2485nfs_fsync(struct vop_fsync_args *ap)
2486{
|
2487 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1));
| 2487 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
|
2488}
2489
2490/*
2491 * Flush all the blocks associated with a vnode.
2492 * Walk through the buffer pool and push any dirty pages
2493 * associated with the vnode.
| 2488}
2489
2490/*
2491 * Flush all the blocks associated with a vnode.
2492 * Walk through the buffer pool and push any dirty pages
2493 * associated with the vnode.
|
| 2494 * If the called_from_renewthread argument is TRUE, it has been called
2495 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2496 * waiting for a buffer write to complete.
|
2494 */
2495int
2496ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
| 2497 */
2498int
2499ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
|
2497 int commit)
| 2500 int commit, int called_from_renewthread)
|
2498{
2499 struct nfsnode *np = VTONFS(vp);
2500 struct buf *bp;
2501 int i;
2502 struct buf *nbp;
2503 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2504 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2505 int passone = 1, trycnt = 0;
2506 u_quad_t off, endoff, toff;
2507 struct ucred* wcred = NULL;
2508 struct buf **bvec = NULL;
2509 struct bufobj *bo;
2510#ifndef NFS_COMMITBVECSIZ
2511#define NFS_COMMITBVECSIZ 20
2512#endif
2513 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2514 int bvecsize = 0, bveccount;
2515
| 2501{
2502 struct nfsnode *np = VTONFS(vp);
2503 struct buf *bp;
2504 int i;
2505 struct buf *nbp;
2506 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2507 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2508 int passone = 1, trycnt = 0;
2509 u_quad_t off, endoff, toff;
2510 struct ucred* wcred = NULL;
2511 struct buf **bvec = NULL;
2512 struct bufobj *bo;
2513#ifndef NFS_COMMITBVECSIZ
2514#define NFS_COMMITBVECSIZ 20
2515#endif
2516 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2517 int bvecsize = 0, bveccount;
2518
|
| 2519 if (called_from_renewthread != 0)
2520 slptimeo = hz;
|
2516 if (nmp->nm_flag & NFSMNT_INT)
2517 slpflag = NFS_PCATCH;
2518 if (!commit)
2519 passone = 0;
2520 bo = &vp->v_bufobj;
2521 /*
2522 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2523 * server, but has not been committed to stable storage on the server
2524 * yet. On the first pass, the byte range is worked out and the commit
2525 * rpc is done. On the second pass, ncl_writebp() is called to do the
2526 * job.
2527 */
2528again:
2529 off = (u_quad_t)-1;
2530 endoff = 0;
2531 bvecpos = 0;
2532 if (NFS_ISV34(vp) && commit) {
2533 if (bvec != NULL && bvec != bvec_on_stack)
2534 free(bvec, M_TEMP);
2535 /*
2536 * Count up how many buffers waiting for a commit.
2537 */
2538 bveccount = 0;
2539 BO_LOCK(bo);
2540 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2541 if (!BUF_ISLOCKED(bp) &&
2542 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2543 == (B_DELWRI | B_NEEDCOMMIT))
2544 bveccount++;
2545 }
2546 /*
2547 * Allocate space to remember the list of bufs to commit. It is
2548 * important to use M_NOWAIT here to avoid a race with nfs_write.
2549 * If we can't get memory (for whatever reason), we will end up
2550 * committing the buffers one-by-one in the loop below.
2551 */
2552 if (bveccount > NFS_COMMITBVECSIZ) {
2553 /*
2554 * Release the vnode interlock to avoid a lock
2555 * order reversal.
2556 */
2557 BO_UNLOCK(bo);
2558 bvec = (struct buf **)
2559 malloc(bveccount * sizeof(struct buf *),
2560 M_TEMP, M_NOWAIT);
2561 BO_LOCK(bo);
2562 if (bvec == NULL) {
2563 bvec = bvec_on_stack;
2564 bvecsize = NFS_COMMITBVECSIZ;
2565 } else
2566 bvecsize = bveccount;
2567 } else {
2568 bvec = bvec_on_stack;
2569 bvecsize = NFS_COMMITBVECSIZ;
2570 }
2571 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2572 if (bvecpos >= bvecsize)
2573 break;
2574 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2575 nbp = TAILQ_NEXT(bp, b_bobufs);
2576 continue;
2577 }
2578 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2579 (B_DELWRI | B_NEEDCOMMIT)) {
2580 BUF_UNLOCK(bp);
2581 nbp = TAILQ_NEXT(bp, b_bobufs);
2582 continue;
2583 }
2584 BO_UNLOCK(bo);
2585 bremfree(bp);
2586 /*
2587 * Work out if all buffers are using the same cred
2588 * so we can deal with them all with one commit.
2589 *
2590 * NOTE: we are not clearing B_DONE here, so we have
2591 * to do it later on in this routine if we intend to
2592 * initiate I/O on the bp.
2593 *
2594 * Note: to avoid loopback deadlocks, we do not
2595 * assign b_runningbufspace.
2596 */
2597 if (wcred == NULL)
2598 wcred = bp->b_wcred;
2599 else if (wcred != bp->b_wcred)
2600 wcred = NOCRED;
2601 vfs_busy_pages(bp, 1);
2602
2603 BO_LOCK(bo);
2604 /*
2605 * bp is protected by being locked, but nbp is not
2606 * and vfs_busy_pages() may sleep. We have to
2607 * recalculate nbp.
2608 */
2609 nbp = TAILQ_NEXT(bp, b_bobufs);
2610
2611 /*
2612 * A list of these buffers is kept so that the
2613 * second loop knows which buffers have actually
2614 * been committed. This is necessary, since there
2615 * may be a race between the commit rpc and new
2616 * uncommitted writes on the file.
2617 */
2618 bvec[bvecpos++] = bp;
2619 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2620 bp->b_dirtyoff;
2621 if (toff < off)
2622 off = toff;
2623 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2624 if (toff > endoff)
2625 endoff = toff;
2626 }
2627 BO_UNLOCK(bo);
2628 }
2629 if (bvecpos > 0) {
2630 /*
2631 * Commit data on the server, as required.
2632 * If all bufs are using the same wcred, then use that with
2633 * one call for all of them, otherwise commit each one
2634 * separately.
2635 */
2636 if (wcred != NOCRED)
2637 retv = ncl_commit(vp, off, (int)(endoff - off),
2638 wcred, td);
2639 else {
2640 retv = 0;
2641 for (i = 0; i < bvecpos; i++) {
2642 off_t off, size;
2643 bp = bvec[i];
2644 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2645 bp->b_dirtyoff;
2646 size = (u_quad_t)(bp->b_dirtyend
2647 - bp->b_dirtyoff);
2648 retv = ncl_commit(vp, off, (int)size,
2649 bp->b_wcred, td);
2650 if (retv) break;
2651 }
2652 }
2653
2654 if (retv == NFSERR_STALEWRITEVERF)
2655 ncl_clearcommit(vp->v_mount);
2656
2657 /*
2658 * Now, either mark the blocks I/O done or mark the
2659 * blocks dirty, depending on whether the commit
2660 * succeeded.
2661 */
2662 for (i = 0; i < bvecpos; i++) {
2663 bp = bvec[i];
2664 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2665 if (retv) {
2666 /*
2667 * Error, leave B_DELWRI intact
2668 */
2669 vfs_unbusy_pages(bp);
2670 brelse(bp);
2671 } else {
2672 /*
2673 * Success, remove B_DELWRI ( bundirty() ).
2674 *
2675 * b_dirtyoff/b_dirtyend seem to be NFS
2676 * specific. We should probably move that
2677 * into bundirty(). XXX
2678 */
2679 bufobj_wref(bo);
2680 bp->b_flags |= B_ASYNC;
2681 bundirty(bp);
2682 bp->b_flags &= ~B_DONE;
2683 bp->b_ioflags &= ~BIO_ERROR;
2684 bp->b_dirtyoff = bp->b_dirtyend = 0;
2685 bufdone(bp);
2686 }
2687 }
2688 }
2689
2690 /*
2691 * Start/do any write(s) that are required.
2692 */
2693loop:
2694 BO_LOCK(bo);
2695 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2696 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2697 if (waitfor != MNT_WAIT || passone)
2698 continue;
2699
2700 error = BUF_TIMELOCK(bp,
2701 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2702 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2703 if (error == 0) {
2704 BUF_UNLOCK(bp);
2705 goto loop;
2706 }
2707 if (error == ENOLCK) {
2708 error = 0;
2709 goto loop;
2710 }
| 2521 if (nmp->nm_flag & NFSMNT_INT)
2522 slpflag = NFS_PCATCH;
2523 if (!commit)
2524 passone = 0;
2525 bo = &vp->v_bufobj;
2526 /*
2527 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2528 * server, but has not been committed to stable storage on the server
2529 * yet. On the first pass, the byte range is worked out and the commit
2530 * rpc is done. On the second pass, ncl_writebp() is called to do the
2531 * job.
2532 */
2533again:
2534 off = (u_quad_t)-1;
2535 endoff = 0;
2536 bvecpos = 0;
2537 if (NFS_ISV34(vp) && commit) {
2538 if (bvec != NULL && bvec != bvec_on_stack)
2539 free(bvec, M_TEMP);
2540 /*
2541 * Count up how many buffers waiting for a commit.
2542 */
2543 bveccount = 0;
2544 BO_LOCK(bo);
2545 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2546 if (!BUF_ISLOCKED(bp) &&
2547 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2548 == (B_DELWRI | B_NEEDCOMMIT))
2549 bveccount++;
2550 }
2551 /*
2552 * Allocate space to remember the list of bufs to commit. It is
2553 * important to use M_NOWAIT here to avoid a race with nfs_write.
2554 * If we can't get memory (for whatever reason), we will end up
2555 * committing the buffers one-by-one in the loop below.
2556 */
2557 if (bveccount > NFS_COMMITBVECSIZ) {
2558 /*
2559 * Release the vnode interlock to avoid a lock
2560 * order reversal.
2561 */
2562 BO_UNLOCK(bo);
2563 bvec = (struct buf **)
2564 malloc(bveccount * sizeof(struct buf *),
2565 M_TEMP, M_NOWAIT);
2566 BO_LOCK(bo);
2567 if (bvec == NULL) {
2568 bvec = bvec_on_stack;
2569 bvecsize = NFS_COMMITBVECSIZ;
2570 } else
2571 bvecsize = bveccount;
2572 } else {
2573 bvec = bvec_on_stack;
2574 bvecsize = NFS_COMMITBVECSIZ;
2575 }
2576 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2577 if (bvecpos >= bvecsize)
2578 break;
2579 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2580 nbp = TAILQ_NEXT(bp, b_bobufs);
2581 continue;
2582 }
2583 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2584 (B_DELWRI | B_NEEDCOMMIT)) {
2585 BUF_UNLOCK(bp);
2586 nbp = TAILQ_NEXT(bp, b_bobufs);
2587 continue;
2588 }
2589 BO_UNLOCK(bo);
2590 bremfree(bp);
2591 /*
2592 * Work out if all buffers are using the same cred
2593 * so we can deal with them all with one commit.
2594 *
2595 * NOTE: we are not clearing B_DONE here, so we have
2596 * to do it later on in this routine if we intend to
2597 * initiate I/O on the bp.
2598 *
2599 * Note: to avoid loopback deadlocks, we do not
2600 * assign b_runningbufspace.
2601 */
2602 if (wcred == NULL)
2603 wcred = bp->b_wcred;
2604 else if (wcred != bp->b_wcred)
2605 wcred = NOCRED;
2606 vfs_busy_pages(bp, 1);
2607
2608 BO_LOCK(bo);
2609 /*
2610 * bp is protected by being locked, but nbp is not
2611 * and vfs_busy_pages() may sleep. We have to
2612 * recalculate nbp.
2613 */
2614 nbp = TAILQ_NEXT(bp, b_bobufs);
2615
2616 /*
2617 * A list of these buffers is kept so that the
2618 * second loop knows which buffers have actually
2619 * been committed. This is necessary, since there
2620 * may be a race between the commit rpc and new
2621 * uncommitted writes on the file.
2622 */
2623 bvec[bvecpos++] = bp;
2624 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2625 bp->b_dirtyoff;
2626 if (toff < off)
2627 off = toff;
2628 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2629 if (toff > endoff)
2630 endoff = toff;
2631 }
2632 BO_UNLOCK(bo);
2633 }
2634 if (bvecpos > 0) {
2635 /*
2636 * Commit data on the server, as required.
2637 * If all bufs are using the same wcred, then use that with
2638 * one call for all of them, otherwise commit each one
2639 * separately.
2640 */
2641 if (wcred != NOCRED)
2642 retv = ncl_commit(vp, off, (int)(endoff - off),
2643 wcred, td);
2644 else {
2645 retv = 0;
2646 for (i = 0; i < bvecpos; i++) {
2647 off_t off, size;
2648 bp = bvec[i];
2649 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2650 bp->b_dirtyoff;
2651 size = (u_quad_t)(bp->b_dirtyend
2652 - bp->b_dirtyoff);
2653 retv = ncl_commit(vp, off, (int)size,
2654 bp->b_wcred, td);
2655 if (retv) break;
2656 }
2657 }
2658
2659 if (retv == NFSERR_STALEWRITEVERF)
2660 ncl_clearcommit(vp->v_mount);
2661
2662 /*
2663 * Now, either mark the blocks I/O done or mark the
2664 * blocks dirty, depending on whether the commit
2665 * succeeded.
2666 */
2667 for (i = 0; i < bvecpos; i++) {
2668 bp = bvec[i];
2669 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2670 if (retv) {
2671 /*
2672 * Error, leave B_DELWRI intact
2673 */
2674 vfs_unbusy_pages(bp);
2675 brelse(bp);
2676 } else {
2677 /*
2678 * Success, remove B_DELWRI ( bundirty() ).
2679 *
2680 * b_dirtyoff/b_dirtyend seem to be NFS
2681 * specific. We should probably move that
2682 * into bundirty(). XXX
2683 */
2684 bufobj_wref(bo);
2685 bp->b_flags |= B_ASYNC;
2686 bundirty(bp);
2687 bp->b_flags &= ~B_DONE;
2688 bp->b_ioflags &= ~BIO_ERROR;
2689 bp->b_dirtyoff = bp->b_dirtyend = 0;
2690 bufdone(bp);
2691 }
2692 }
2693 }
2694
2695 /*
2696 * Start/do any write(s) that are required.
2697 */
2698loop:
2699 BO_LOCK(bo);
2700 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2701 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2702 if (waitfor != MNT_WAIT || passone)
2703 continue;
2704
2705 error = BUF_TIMELOCK(bp,
2706 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2707 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2708 if (error == 0) {
2709 BUF_UNLOCK(bp);
2710 goto loop;
2711 }
2712 if (error == ENOLCK) {
2713 error = 0;
2714 goto loop;
2715 }
|
| 2716 if (called_from_renewthread != 0) {
2717 /*
2718 * Return EIO so the flush will be retried
2719 * later.
2720 */
2721 error = EIO;
2722 goto done;
2723 }
|
2711 if (newnfs_sigintr(nmp, td)) {
2712 error = EINTR;
2713 goto done;
2714 }
2715 if (slpflag & PCATCH) {
2716 slpflag = 0;
2717 slptimeo = 2 * hz;
2718 }
2719 goto loop;
2720 }
2721 if ((bp->b_flags & B_DELWRI) == 0)
2722 panic("nfs_fsync: not dirty");
2723 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2724 BUF_UNLOCK(bp);
2725 continue;
2726 }
2727 BO_UNLOCK(bo);
2728 bremfree(bp);
2729 if (passone || !commit)
2730 bp->b_flags |= B_ASYNC;
2731 else
2732 bp->b_flags |= B_ASYNC;
2733 bwrite(bp);
2734 if (newnfs_sigintr(nmp, td)) {
2735 error = EINTR;
2736 goto done;
2737 }
2738 goto loop;
2739 }
2740 if (passone) {
2741 passone = 0;
2742 BO_UNLOCK(bo);
2743 goto again;
2744 }
2745 if (waitfor == MNT_WAIT) {
2746 while (bo->bo_numoutput) {
2747 error = bufobj_wwait(bo, slpflag, slptimeo);
2748 if (error) {
2749 BO_UNLOCK(bo);
| 2724 if (newnfs_sigintr(nmp, td)) {
2725 error = EINTR;
2726 goto done;
2727 }
2728 if (slpflag & PCATCH) {
2729 slpflag = 0;
2730 slptimeo = 2 * hz;
2731 }
2732 goto loop;
2733 }
2734 if ((bp->b_flags & B_DELWRI) == 0)
2735 panic("nfs_fsync: not dirty");
2736 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2737 BUF_UNLOCK(bp);
2738 continue;
2739 }
2740 BO_UNLOCK(bo);
2741 bremfree(bp);
2742 if (passone || !commit)
2743 bp->b_flags |= B_ASYNC;
2744 else
2745 bp->b_flags |= B_ASYNC;
2746 bwrite(bp);
2747 if (newnfs_sigintr(nmp, td)) {
2748 error = EINTR;
2749 goto done;
2750 }
2751 goto loop;
2752 }
2753 if (passone) {
2754 passone = 0;
2755 BO_UNLOCK(bo);
2756 goto again;
2757 }
2758 if (waitfor == MNT_WAIT) {
2759 while (bo->bo_numoutput) {
2760 error = bufobj_wwait(bo, slpflag, slptimeo);
2761 if (error) {
2762 BO_UNLOCK(bo);
|
| 2763 if (called_from_renewthread != 0) {
2764 /*
2765 * Return EIO so that the flush will be
2766 * retried later.
2767 */
2768 error = EIO;
2769 goto done;
2770 }
|
2750 error = newnfs_sigintr(nmp, td);
2751 if (error)
2752 goto done;
2753 if (slpflag & PCATCH) {
2754 slpflag = 0;
2755 slptimeo = 2 * hz;
2756 }
2757 BO_LOCK(bo);
2758 }
2759 }
2760 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2761 BO_UNLOCK(bo);
2762 goto loop;
2763 }
2764 /*
2765 * Wait for all the async IO requests to drain
2766 */
2767 BO_UNLOCK(bo);
2768 mtx_lock(&np->n_mtx);
2769 while (np->n_directio_asyncwr > 0) {
2770 np->n_flag |= NFSYNCWAIT;
2771 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2772 &np->n_mtx, slpflag | (PRIBIO + 1),
2773 "nfsfsync", 0);
2774 if (error) {
2775 if (newnfs_sigintr(nmp, td)) {
2776 mtx_unlock(&np->n_mtx);
2777 error = EINTR;
2778 goto done;
2779 }
2780 }
2781 }
2782 mtx_unlock(&np->n_mtx);
2783 } else
2784 BO_UNLOCK(bo);
2785 mtx_lock(&np->n_mtx);
2786 if (np->n_flag & NWRITEERR) {
2787 error = np->n_error;
2788 np->n_flag &= ~NWRITEERR;
2789 }
2790 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2791 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2792 np->n_flag &= ~NMODIFIED;
2793 mtx_unlock(&np->n_mtx);
2794done:
2795 if (bvec != NULL && bvec != bvec_on_stack)
2796 free(bvec, M_TEMP);
2797 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2798 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2799 np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2800 /* try, try again... */
2801 passone = 1;
2802 wcred = NULL;
2803 bvec = NULL;
2804 bvecsize = 0;
2805printf("try%d\n", trycnt);
2806 goto again;
2807 }
2808 return (error);
2809}
2810
2811/*
2812 * NFS advisory byte-level locks.
2813 */
2814static int
2815nfs_advlock(struct vop_advlock_args *ap)
2816{
2817 struct vnode *vp = ap->a_vp;
2818 struct ucred *cred;
2819 struct nfsnode *np = VTONFS(ap->a_vp);
2820 struct proc *p = (struct proc *)ap->a_id;
2821 struct thread *td = curthread; /* XXX */
2822 struct vattr va;
2823 int ret, error = EOPNOTSUPP;
2824 u_quad_t size;
2825
2826 if (NFS_ISV4(vp) && (ap->a_flags & F_POSIX)) {
2827 cred = p->p_ucred;
2828 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2829 if (vp->v_iflag & VI_DOOMED) {
2830 VOP_UNLOCK(vp, 0);
2831 return (EBADF);
2832 }
2833
2834 /*
2835 * If this is unlocking a write locked region, flush and
2836 * commit them before unlocking. This is required by
2837 * RFC3530 Sec. 9.3.2.
2838 */
2839 if (ap->a_op == F_UNLCK &&
2840 nfscl_checkwritelocked(vp, ap->a_fl, cred, td))
| 2771 error = newnfs_sigintr(nmp, td);
2772 if (error)
2773 goto done;
2774 if (slpflag & PCATCH) {
2775 slpflag = 0;
2776 slptimeo = 2 * hz;
2777 }
2778 BO_LOCK(bo);
2779 }
2780 }
2781 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2782 BO_UNLOCK(bo);
2783 goto loop;
2784 }
2785 /*
2786 * Wait for all the async IO requests to drain
2787 */
2788 BO_UNLOCK(bo);
2789 mtx_lock(&np->n_mtx);
2790 while (np->n_directio_asyncwr > 0) {
2791 np->n_flag |= NFSYNCWAIT;
2792 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2793 &np->n_mtx, slpflag | (PRIBIO + 1),
2794 "nfsfsync", 0);
2795 if (error) {
2796 if (newnfs_sigintr(nmp, td)) {
2797 mtx_unlock(&np->n_mtx);
2798 error = EINTR;
2799 goto done;
2800 }
2801 }
2802 }
2803 mtx_unlock(&np->n_mtx);
2804 } else
2805 BO_UNLOCK(bo);
2806 mtx_lock(&np->n_mtx);
2807 if (np->n_flag & NWRITEERR) {
2808 error = np->n_error;
2809 np->n_flag &= ~NWRITEERR;
2810 }
2811 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2812 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2813 np->n_flag &= ~NMODIFIED;
2814 mtx_unlock(&np->n_mtx);
2815done:
2816 if (bvec != NULL && bvec != bvec_on_stack)
2817 free(bvec, M_TEMP);
2818 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2819 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2820 np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2821 /* try, try again... */
2822 passone = 1;
2823 wcred = NULL;
2824 bvec = NULL;
2825 bvecsize = 0;
2826printf("try%d\n", trycnt);
2827 goto again;
2828 }
2829 return (error);
2830}
2831
2832/*
2833 * NFS advisory byte-level locks.
2834 */
2835static int
2836nfs_advlock(struct vop_advlock_args *ap)
2837{
2838 struct vnode *vp = ap->a_vp;
2839 struct ucred *cred;
2840 struct nfsnode *np = VTONFS(ap->a_vp);
2841 struct proc *p = (struct proc *)ap->a_id;
2842 struct thread *td = curthread; /* XXX */
2843 struct vattr va;
2844 int ret, error = EOPNOTSUPP;
2845 u_quad_t size;
2846
2847 if (NFS_ISV4(vp) && (ap->a_flags & F_POSIX)) {
2848 cred = p->p_ucred;
2849 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2850 if (vp->v_iflag & VI_DOOMED) {
2851 VOP_UNLOCK(vp, 0);
2852 return (EBADF);
2853 }
2854
2855 /*
2856 * If this is unlocking a write locked region, flush and
2857 * commit them before unlocking. This is required by
2858 * RFC3530 Sec. 9.3.2.
2859 */
2860 if (ap->a_op == F_UNLCK &&
2861 nfscl_checkwritelocked(vp, ap->a_fl, cred, td))
|
2841 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1);
| 2862 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
|
2842
2843 /*
2844 * Loop around doing the lock op, while a blocking lock
2845 * must wait for the lock op to succeed.
2846 */
2847 do {
2848 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
2849 ap->a_fl, 0, cred, td);
2850 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2851 ap->a_op == F_SETLK) {
2852 VOP_UNLOCK(vp, 0);
2853 error = nfs_catnap(PZERO | PCATCH, "ncladvl");
2854 if (error)
2855 return (EINTR);
2856 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2857 if (vp->v_iflag & VI_DOOMED) {
2858 VOP_UNLOCK(vp, 0);
2859 return (EBADF);
2860 }
2861 }
2862 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2863 ap->a_op == F_SETLK);
2864 if (ret == NFSERR_DENIED) {
2865 VOP_UNLOCK(vp, 0);
2866 return (EAGAIN);
2867 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
2868 VOP_UNLOCK(vp, 0);
2869 return (ret);
2870 } else if (ret != 0) {
2871 VOP_UNLOCK(vp, 0);
2872 return (EACCES);
2873 }
2874
2875 /*
2876 * Now, if we just got a lock, invalidate data in the buffer
2877 * cache, as required, so that the coherency conforms with
2878 * RFC3530 Sec. 9.3.2.
2879 */
2880 if (ap->a_op == F_SETLK) {
2881 if ((np->n_flag & NMODIFIED) == 0) {
2882 np->n_attrstamp = 0;
2883 ret = VOP_GETATTR(vp, &va, cred);
2884 }
2885 if ((np->n_flag & NMODIFIED) || ret ||
2886 np->n_change != va.va_filerev) {
2887 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
2888 np->n_attrstamp = 0;
2889 ret = VOP_GETATTR(vp, &va, cred);
2890 if (!ret) {
2891 np->n_mtime = va.va_mtime;
2892 np->n_change = va.va_filerev;
2893 }
2894 }
2895 }
2896 VOP_UNLOCK(vp, 0);
2897 return (0);
2898 } else if (!NFS_ISV4(vp)) {
2899 error = vn_lock(vp, LK_SHARED);
2900 if (error)
2901 return (error);
2902 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2903 size = VTONFS(vp)->n_size;
2904 VOP_UNLOCK(vp, 0);
2905 error = lf_advlock(ap, &(vp->v_lockf), size);
2906 } else {
2907 if (ncl_advlock_p)
2908 error = ncl_advlock_p(ap);
2909 else
2910 error = ENOLCK;
2911 }
2912 }
2913 return (error);
2914}
2915
2916/*
2917 * NFS advisory byte-level locks.
2918 */
2919static int
2920nfs_advlockasync(struct vop_advlockasync_args *ap)
2921{
2922 struct vnode *vp = ap->a_vp;
2923 u_quad_t size;
2924 int error;
2925
2926 if (NFS_ISV4(vp))
2927 return (EOPNOTSUPP);
2928 error = vn_lock(vp, LK_SHARED);
2929 if (error)
2930 return (error);
2931 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2932 size = VTONFS(vp)->n_size;
2933 VOP_UNLOCK(vp, 0);
2934 error = lf_advlockasync(ap, &(vp->v_lockf), size);
2935 } else {
2936 VOP_UNLOCK(vp, 0);
2937 error = EOPNOTSUPP;
2938 }
2939 return (error);
2940}
2941
2942/*
2943 * Print out the contents of an nfsnode.
2944 */
2945static int
2946nfs_print(struct vop_print_args *ap)
2947{
2948 struct vnode *vp = ap->a_vp;
2949 struct nfsnode *np = VTONFS(vp);
2950
2951 ncl_printf("\tfileid %ld fsid 0x%x",
2952 np->n_vattr.na_fileid, np->n_vattr.na_fsid);
2953 if (vp->v_type == VFIFO)
2954 fifo_printinfo(vp);
2955 printf("\n");
2956 return (0);
2957}
2958
2959/*
2960 * This is the "real" nfs::bwrite(struct buf*).
2961 * We set B_CACHE if this is a VMIO buffer.
2962 */
2963int
2964ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
2965{
2966 int s;
2967 int oldflags = bp->b_flags;
2968#if 0
2969 int retv = 1;
2970 off_t off;
2971#endif
2972
2973 BUF_ASSERT_HELD(bp);
2974
2975 if (bp->b_flags & B_INVAL) {
2976 brelse(bp);
2977 return(0);
2978 }
2979
2980 bp->b_flags |= B_CACHE;
2981
2982 /*
2983 * Undirty the bp. We will redirty it later if the I/O fails.
2984 */
2985
2986 s = splbio();
2987 bundirty(bp);
2988 bp->b_flags &= ~B_DONE;
2989 bp->b_ioflags &= ~BIO_ERROR;
2990 bp->b_iocmd = BIO_WRITE;
2991
2992 bufobj_wref(bp->b_bufobj);
2993 curthread->td_ru.ru_oublock++;
2994 splx(s);
2995
2996 /*
2997 * Note: to avoid loopback deadlocks, we do not
2998 * assign b_runningbufspace.
2999 */
3000 vfs_busy_pages(bp, 1);
3001
3002 BUF_KERNPROC(bp);
3003 bp->b_iooffset = dbtob(bp->b_blkno);
3004 bstrategy(bp);
3005
3006 if( (oldflags & B_ASYNC) == 0) {
3007 int rtval = bufwait(bp);
3008
3009 if (oldflags & B_DELWRI) {
3010 s = splbio();
3011 reassignbuf(bp);
3012 splx(s);
3013 }
3014 brelse(bp);
3015 return (rtval);
3016 }
3017
3018 return (0);
3019}
3020
3021/*
3022 * nfs special file access vnode op.
3023 * Essentially just get vattr and then imitate iaccess() since the device is
3024 * local to the client.
3025 */
3026static int
3027nfsspec_access(struct vop_access_args *ap)
3028{
3029 struct vattr *vap;
3030 struct ucred *cred = ap->a_cred;
3031 struct vnode *vp = ap->a_vp;
3032 accmode_t accmode = ap->a_accmode;
3033 struct vattr vattr;
3034 int error;
3035
3036 /*
3037 * Disallow write attempts on filesystems mounted read-only;
3038 * unless the file is a socket, fifo, or a block or character
3039 * device resident on the filesystem.
3040 */
3041 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3042 switch (vp->v_type) {
3043 case VREG:
3044 case VDIR:
3045 case VLNK:
3046 return (EROFS);
3047 default:
3048 break;
3049 }
3050 }
3051 vap = &vattr;
3052 error = VOP_GETATTR(vp, vap, cred);
3053 if (error)
3054 goto out;
3055 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3056 accmode, cred, NULL);
3057out:
3058 return error;
3059}
3060
3061/*
3062 * Read wrapper for fifos.
3063 */
3064static int
3065nfsfifo_read(struct vop_read_args *ap)
3066{
3067 struct nfsnode *np = VTONFS(ap->a_vp);
3068 int error;
3069
3070 /*
3071 * Set access flag.
3072 */
3073 mtx_lock(&np->n_mtx);
3074 np->n_flag |= NACC;
3075 getnanotime(&np->n_atim);
3076 mtx_unlock(&np->n_mtx);
3077 error = fifo_specops.vop_read(ap);
3078 return error;
3079}
3080
3081/*
3082 * Write wrapper for fifos.
3083 */
3084static int
3085nfsfifo_write(struct vop_write_args *ap)
3086{
3087 struct nfsnode *np = VTONFS(ap->a_vp);
3088
3089 /*
3090 * Set update flag.
3091 */
3092 mtx_lock(&np->n_mtx);
3093 np->n_flag |= NUPD;
3094 getnanotime(&np->n_mtim);
3095 mtx_unlock(&np->n_mtx);
3096 return(fifo_specops.vop_write(ap));
3097}
3098
3099/*
3100 * Close wrapper for fifos.
3101 *
3102 * Update the times on the nfsnode then do fifo close.
3103 */
3104static int
3105nfsfifo_close(struct vop_close_args *ap)
3106{
3107 struct vnode *vp = ap->a_vp;
3108 struct nfsnode *np = VTONFS(vp);
3109 struct vattr vattr;
3110 struct timespec ts;
3111
3112 mtx_lock(&np->n_mtx);
3113 if (np->n_flag & (NACC | NUPD)) {
3114 getnanotime(&ts);
3115 if (np->n_flag & NACC)
3116 np->n_atim = ts;
3117 if (np->n_flag & NUPD)
3118 np->n_mtim = ts;
3119 np->n_flag |= NCHG;
3120 if (vrefcnt(vp) == 1 &&
3121 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3122 VATTR_NULL(&vattr);
3123 if (np->n_flag & NACC)
3124 vattr.va_atime = np->n_atim;
3125 if (np->n_flag & NUPD)
3126 vattr.va_mtime = np->n_mtim;
3127 mtx_unlock(&np->n_mtx);
3128 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3129 goto out;
3130 }
3131 }
3132 mtx_unlock(&np->n_mtx);
3133out:
3134 return (fifo_specops.vop_close(ap));
3135}
3136
3137/*
3138 * Just call ncl_writebp() with the force argument set to 1.
3139 *
3140 * NOTE: B_DONE may or may not be set in a_bp on call.
3141 */
3142static int
3143nfs_bwrite(struct buf *bp)
3144{
3145
3146 return (ncl_writebp(bp, 1, curthread));
3147}
3148
3149struct buf_ops buf_ops_newnfs = {
3150 .bop_name = "buf_ops_nfs",
3151 .bop_write = nfs_bwrite,
3152 .bop_strategy = bufstrategy,
3153 .bop_sync = bufsync,
3154 .bop_bdflush = bufbdflush,
3155};
3156
3157/*
3158 * Cloned from vop_stdlock(), and then the ugly hack added.
3159 */
3160static int
3161nfs_lock1(struct vop_lock1_args *ap)
3162{
3163 struct vnode *vp = ap->a_vp;
3164 int error = 0;
3165
3166 /*
3167 * Since vfs_hash_get() calls vget() and it will no longer work
3168 * for FreeBSD8 with flags == 0, I can only think of this horrible
3169 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3170 * and then handle it here. All I want for this case is a v_usecount
3171 * on the vnode to use for recovery, while another thread might
3172 * hold a lock on the vnode. I have the other threads blocked, so
3173 * there isn't any race problem.
3174 */
3175 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3176 if ((ap->a_flags & LK_INTERLOCK) == 0)
3177 panic("ncllock1");
3178 if ((vp->v_iflag & VI_DOOMED))
3179 error = ENOENT;
3180 VI_UNLOCK(vp);
3181 return (error);
3182 }
3183 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3184 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3185 ap->a_line));
3186}
3187
3188static int
3189nfs_getacl(struct vop_getacl_args *ap)
3190{
3191 int error;
3192
3193 if (ap->a_type != ACL_TYPE_NFS4)
3194 return (EOPNOTSUPP);
3195 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3196 NULL);
3197 if (error > NFSERR_STALE) {
3198 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3199 error = EPERM;
3200 }
3201 return (error);
3202}
3203
3204static int
3205nfs_setacl(struct vop_setacl_args *ap)
3206{
3207 int error;
3208
3209 if (ap->a_type != ACL_TYPE_NFS4)
3210 return (EOPNOTSUPP);
3211 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3212 NULL);
3213 if (error > NFSERR_STALE) {
3214 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3215 error = EPERM;
3216 }
3217 return (error);
3218}
| 2863
2864 /*
2865 * Loop around doing the lock op, while a blocking lock
2866 * must wait for the lock op to succeed.
2867 */
2868 do {
2869 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
2870 ap->a_fl, 0, cred, td);
2871 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2872 ap->a_op == F_SETLK) {
2873 VOP_UNLOCK(vp, 0);
2874 error = nfs_catnap(PZERO | PCATCH, "ncladvl");
2875 if (error)
2876 return (EINTR);
2877 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2878 if (vp->v_iflag & VI_DOOMED) {
2879 VOP_UNLOCK(vp, 0);
2880 return (EBADF);
2881 }
2882 }
2883 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2884 ap->a_op == F_SETLK);
2885 if (ret == NFSERR_DENIED) {
2886 VOP_UNLOCK(vp, 0);
2887 return (EAGAIN);
2888 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
2889 VOP_UNLOCK(vp, 0);
2890 return (ret);
2891 } else if (ret != 0) {
2892 VOP_UNLOCK(vp, 0);
2893 return (EACCES);
2894 }
2895
2896 /*
2897 * Now, if we just got a lock, invalidate data in the buffer
2898 * cache, as required, so that the coherency conforms with
2899 * RFC3530 Sec. 9.3.2.
2900 */
2901 if (ap->a_op == F_SETLK) {
2902 if ((np->n_flag & NMODIFIED) == 0) {
2903 np->n_attrstamp = 0;
2904 ret = VOP_GETATTR(vp, &va, cred);
2905 }
2906 if ((np->n_flag & NMODIFIED) || ret ||
2907 np->n_change != va.va_filerev) {
2908 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
2909 np->n_attrstamp = 0;
2910 ret = VOP_GETATTR(vp, &va, cred);
2911 if (!ret) {
2912 np->n_mtime = va.va_mtime;
2913 np->n_change = va.va_filerev;
2914 }
2915 }
2916 }
2917 VOP_UNLOCK(vp, 0);
2918 return (0);
2919 } else if (!NFS_ISV4(vp)) {
2920 error = vn_lock(vp, LK_SHARED);
2921 if (error)
2922 return (error);
2923 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2924 size = VTONFS(vp)->n_size;
2925 VOP_UNLOCK(vp, 0);
2926 error = lf_advlock(ap, &(vp->v_lockf), size);
2927 } else {
2928 if (ncl_advlock_p)
2929 error = ncl_advlock_p(ap);
2930 else
2931 error = ENOLCK;
2932 }
2933 }
2934 return (error);
2935}
2936
2937/*
2938 * NFS advisory byte-level locks.
2939 */
2940static int
2941nfs_advlockasync(struct vop_advlockasync_args *ap)
2942{
2943 struct vnode *vp = ap->a_vp;
2944 u_quad_t size;
2945 int error;
2946
2947 if (NFS_ISV4(vp))
2948 return (EOPNOTSUPP);
2949 error = vn_lock(vp, LK_SHARED);
2950 if (error)
2951 return (error);
2952 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2953 size = VTONFS(vp)->n_size;
2954 VOP_UNLOCK(vp, 0);
2955 error = lf_advlockasync(ap, &(vp->v_lockf), size);
2956 } else {
2957 VOP_UNLOCK(vp, 0);
2958 error = EOPNOTSUPP;
2959 }
2960 return (error);
2961}
2962
2963/*
2964 * Print out the contents of an nfsnode.
2965 */
2966static int
2967nfs_print(struct vop_print_args *ap)
2968{
2969 struct vnode *vp = ap->a_vp;
2970 struct nfsnode *np = VTONFS(vp);
2971
2972 ncl_printf("\tfileid %ld fsid 0x%x",
2973 np->n_vattr.na_fileid, np->n_vattr.na_fsid);
2974 if (vp->v_type == VFIFO)
2975 fifo_printinfo(vp);
2976 printf("\n");
2977 return (0);
2978}
2979
2980/*
2981 * This is the "real" nfs::bwrite(struct buf*).
2982 * We set B_CACHE if this is a VMIO buffer.
2983 */
2984int
2985ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
2986{
2987 int s;
2988 int oldflags = bp->b_flags;
2989#if 0
2990 int retv = 1;
2991 off_t off;
2992#endif
2993
2994 BUF_ASSERT_HELD(bp);
2995
2996 if (bp->b_flags & B_INVAL) {
2997 brelse(bp);
2998 return(0);
2999 }
3000
3001 bp->b_flags |= B_CACHE;
3002
3003 /*
3004 * Undirty the bp. We will redirty it later if the I/O fails.
3005 */
3006
3007 s = splbio();
3008 bundirty(bp);
3009 bp->b_flags &= ~B_DONE;
3010 bp->b_ioflags &= ~BIO_ERROR;
3011 bp->b_iocmd = BIO_WRITE;
3012
3013 bufobj_wref(bp->b_bufobj);
3014 curthread->td_ru.ru_oublock++;
3015 splx(s);
3016
3017 /*
3018 * Note: to avoid loopback deadlocks, we do not
3019 * assign b_runningbufspace.
3020 */
3021 vfs_busy_pages(bp, 1);
3022
3023 BUF_KERNPROC(bp);
3024 bp->b_iooffset = dbtob(bp->b_blkno);
3025 bstrategy(bp);
3026
3027 if( (oldflags & B_ASYNC) == 0) {
3028 int rtval = bufwait(bp);
3029
3030 if (oldflags & B_DELWRI) {
3031 s = splbio();
3032 reassignbuf(bp);
3033 splx(s);
3034 }
3035 brelse(bp);
3036 return (rtval);
3037 }
3038
3039 return (0);
3040}
3041
3042/*
3043 * nfs special file access vnode op.
3044 * Essentially just get vattr and then imitate iaccess() since the device is
3045 * local to the client.
3046 */
3047static int
3048nfsspec_access(struct vop_access_args *ap)
3049{
3050 struct vattr *vap;
3051 struct ucred *cred = ap->a_cred;
3052 struct vnode *vp = ap->a_vp;
3053 accmode_t accmode = ap->a_accmode;
3054 struct vattr vattr;
3055 int error;
3056
3057 /*
3058 * Disallow write attempts on filesystems mounted read-only;
3059 * unless the file is a socket, fifo, or a block or character
3060 * device resident on the filesystem.
3061 */
3062 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3063 switch (vp->v_type) {
3064 case VREG:
3065 case VDIR:
3066 case VLNK:
3067 return (EROFS);
3068 default:
3069 break;
3070 }
3071 }
3072 vap = &vattr;
3073 error = VOP_GETATTR(vp, vap, cred);
3074 if (error)
3075 goto out;
3076 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3077 accmode, cred, NULL);
3078out:
3079 return error;
3080}
3081
3082/*
3083 * Read wrapper for fifos.
3084 */
3085static int
3086nfsfifo_read(struct vop_read_args *ap)
3087{
3088 struct nfsnode *np = VTONFS(ap->a_vp);
3089 int error;
3090
3091 /*
3092 * Set access flag.
3093 */
3094 mtx_lock(&np->n_mtx);
3095 np->n_flag |= NACC;
3096 getnanotime(&np->n_atim);
3097 mtx_unlock(&np->n_mtx);
3098 error = fifo_specops.vop_read(ap);
3099 return error;
3100}
3101
3102/*
3103 * Write wrapper for fifos.
3104 */
3105static int
3106nfsfifo_write(struct vop_write_args *ap)
3107{
3108 struct nfsnode *np = VTONFS(ap->a_vp);
3109
3110 /*
3111 * Set update flag.
3112 */
3113 mtx_lock(&np->n_mtx);
3114 np->n_flag |= NUPD;
3115 getnanotime(&np->n_mtim);
3116 mtx_unlock(&np->n_mtx);
3117 return(fifo_specops.vop_write(ap));
3118}
3119
3120/*
3121 * Close wrapper for fifos.
3122 *
3123 * Update the times on the nfsnode then do fifo close.
3124 */
3125static int
3126nfsfifo_close(struct vop_close_args *ap)
3127{
3128 struct vnode *vp = ap->a_vp;
3129 struct nfsnode *np = VTONFS(vp);
3130 struct vattr vattr;
3131 struct timespec ts;
3132
3133 mtx_lock(&np->n_mtx);
3134 if (np->n_flag & (NACC | NUPD)) {
3135 getnanotime(&ts);
3136 if (np->n_flag & NACC)
3137 np->n_atim = ts;
3138 if (np->n_flag & NUPD)
3139 np->n_mtim = ts;
3140 np->n_flag |= NCHG;
3141 if (vrefcnt(vp) == 1 &&
3142 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3143 VATTR_NULL(&vattr);
3144 if (np->n_flag & NACC)
3145 vattr.va_atime = np->n_atim;
3146 if (np->n_flag & NUPD)
3147 vattr.va_mtime = np->n_mtim;
3148 mtx_unlock(&np->n_mtx);
3149 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3150 goto out;
3151 }
3152 }
3153 mtx_unlock(&np->n_mtx);
3154out:
3155 return (fifo_specops.vop_close(ap));
3156}
3157
3158/*
3159 * Just call ncl_writebp() with the force argument set to 1.
3160 *
3161 * NOTE: B_DONE may or may not be set in a_bp on call.
3162 */
3163static int
3164nfs_bwrite(struct buf *bp)
3165{
3166
3167 return (ncl_writebp(bp, 1, curthread));
3168}
3169
3170struct buf_ops buf_ops_newnfs = {
3171 .bop_name = "buf_ops_nfs",
3172 .bop_write = nfs_bwrite,
3173 .bop_strategy = bufstrategy,
3174 .bop_sync = bufsync,
3175 .bop_bdflush = bufbdflush,
3176};
3177
3178/*
3179 * Cloned from vop_stdlock(), and then the ugly hack added.
3180 */
3181static int
3182nfs_lock1(struct vop_lock1_args *ap)
3183{
3184 struct vnode *vp = ap->a_vp;
3185 int error = 0;
3186
3187 /*
3188 * Since vfs_hash_get() calls vget() and it will no longer work
3189 * for FreeBSD8 with flags == 0, I can only think of this horrible
3190 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3191 * and then handle it here. All I want for this case is a v_usecount
3192 * on the vnode to use for recovery, while another thread might
3193 * hold a lock on the vnode. I have the other threads blocked, so
3194 * there isn't any race problem.
3195 */
3196 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3197 if ((ap->a_flags & LK_INTERLOCK) == 0)
3198 panic("ncllock1");
3199 if ((vp->v_iflag & VI_DOOMED))
3200 error = ENOENT;
3201 VI_UNLOCK(vp);
3202 return (error);
3203 }
3204 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3205 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3206 ap->a_line));
3207}
3208
3209static int
3210nfs_getacl(struct vop_getacl_args *ap)
3211{
3212 int error;
3213
3214 if (ap->a_type != ACL_TYPE_NFS4)
3215 return (EOPNOTSUPP);
3216 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3217 NULL);
3218 if (error > NFSERR_STALE) {
3219 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3220 error = EPERM;
3221 }
3222 return (error);
3223}
3224
3225static int
3226nfs_setacl(struct vop_setacl_args *ap)
3227{
3228 int error;
3229
3230 if (ap->a_type != ACL_TYPE_NFS4)
3231 return (EOPNOTSUPP);
3232 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3233 NULL);
3234 if (error > NFSERR_STALE) {
3235 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3236 error = EPERM;
3237 }
3238 return (error);
3239}
|