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 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD: src/sys/nfsclient/nfs_vnops.c,v 1.293 2008/10/28 13:44:11 trasz Exp $");
37
38 /*
39 * vnode op calls for Sun NFS version 2 and 3
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/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/namei.h>
55 #include <sys/socket.h>
56 #include <sys/vnode.h>
57 #include <sys/dirent.h>
58 #include <sys/fcntl.h>
59 #include <sys/lockf.h>
60 #include <sys/stat.h>
61 #include <sys/sysctl.h>
62 #include <sys/signalvar.h>
63 #include <sys/vimage.h>
64
65 #include <vm/vm.h>
66 #include <vm/vm_object.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
69
70 #include <fs/fifofs/fifo.h>
71
72 #include <rpc/rpcclnt.h>
73
74 #include <nfs/rpcv2.h>
75 #include <nfs/nfsproto.h>
76 #include <nfsclient/nfs.h>
77 #include <nfsclient/nfsnode.h>
78 #include <nfsclient/nfsmount.h>
79 #include <nfsclient/nfs_lock.h>
80 #include <nfs/xdr_subs.h>
81 #include <nfsclient/nfsm_subs.h>
82
83 #include <net/if.h>
84 #include <netinet/in.h>
85 #include <netinet/in_var.h>
86
87 /* Defs */
88 #define TRUE 1
89 #define FALSE 0
90
91 /*
92 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
93 * calls are not in getblk() and brelse() so that they would not be necessary
94 * here.
95 */
96 #ifndef B_VMIO
97 #define vfs_busy_pages(bp, f)
98 #endif
99
100 static vop_read_t nfsfifo_read;
101 static vop_write_t nfsfifo_write;
102 static vop_close_t nfsfifo_close;
103 static int nfs_flush(struct vnode *, int, int);
104 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *);
105 static vop_lookup_t nfs_lookup;
106 static vop_create_t nfs_create;
107 static vop_mknod_t nfs_mknod;
108 static vop_open_t nfs_open;
109 static vop_close_t nfs_close;
110 static vop_access_t nfs_access;
111 static vop_getattr_t nfs_getattr;
112 static vop_setattr_t nfs_setattr;
113 static vop_read_t nfs_read;
114 static vop_fsync_t nfs_fsync;
115 static vop_remove_t nfs_remove;
116 static vop_link_t nfs_link;
117 static vop_rename_t nfs_rename;
118 static vop_mkdir_t nfs_mkdir;
119 static vop_rmdir_t nfs_rmdir;
120 static vop_symlink_t nfs_symlink;
121 static vop_readdir_t nfs_readdir;
122 static vop_strategy_t nfs_strategy;
123 static int nfs_lookitup(struct vnode *, const char *, int,
124 struct ucred *, struct thread *, struct nfsnode **);
125 static int nfs_sillyrename(struct vnode *, struct vnode *,
126 struct componentname *);
127 static vop_access_t nfsspec_access;
128 static vop_readlink_t nfs_readlink;
129 static vop_print_t nfs_print;
130 static vop_advlock_t nfs_advlock;
131 static vop_advlockasync_t nfs_advlockasync;
132
133 /*
134 * Global vfs data structures for nfs
135 */
136 struct vop_vector nfs_vnodeops = {
137 .vop_default = &default_vnodeops,
138 .vop_access = nfs_access,
139 .vop_advlock = nfs_advlock,
140 .vop_advlockasync = nfs_advlockasync,
141 .vop_close = nfs_close,
142 .vop_create = nfs_create,
143 .vop_fsync = nfs_fsync,
144 .vop_getattr = nfs_getattr,
145 .vop_getpages = nfs_getpages,
146 .vop_putpages = nfs_putpages,
147 .vop_inactive = nfs_inactive,
148 .vop_lease = VOP_NULL,
149 .vop_link = nfs_link,
150 .vop_lookup = nfs_lookup,
151 .vop_mkdir = nfs_mkdir,
152 .vop_mknod = nfs_mknod,
153 .vop_open = nfs_open,
154 .vop_print = nfs_print,
155 .vop_read = nfs_read,
156 .vop_readdir = nfs_readdir,
157 .vop_readlink = nfs_readlink,
158 .vop_reclaim = nfs_reclaim,
159 .vop_remove = nfs_remove,
160 .vop_rename = nfs_rename,
161 .vop_rmdir = nfs_rmdir,
162 .vop_setattr = nfs_setattr,
163 .vop_strategy = nfs_strategy,
164 .vop_symlink = nfs_symlink,
165 .vop_write = nfs_write,
166 };
167
168 struct vop_vector nfs_fifoops = {
169 .vop_default = &fifo_specops,
170 .vop_access = nfsspec_access,
171 .vop_close = nfsfifo_close,
172 .vop_fsync = nfs_fsync,
173 .vop_getattr = nfs_getattr,
174 .vop_inactive = nfs_inactive,
175 .vop_print = nfs_print,
176 .vop_read = nfsfifo_read,
177 .vop_reclaim = nfs_reclaim,
178 .vop_setattr = nfs_setattr,
179 .vop_write = nfsfifo_write,
180 };
181
182 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
183 struct componentname *cnp, struct vattr *vap);
184 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
185 struct ucred *cred, struct thread *td);
186 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
187 int fnamelen, struct vnode *tdvp,
188 const char *tnameptr, int tnamelen,
189 struct ucred *cred, struct thread *td);
190 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
191 struct sillyrename *sp);
192
193 /*
194 * Global variables
195 */
196 struct mtx nfs_iod_mtx;
197 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
198 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
199 int nfs_numasync = 0;
200 vop_advlock_t *nfs_advlock_p = nfs_dolock;
201 vop_reclaim_t *nfs_reclaim_p = NULL;
202 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
203
204 SYSCTL_DECL(_vfs_nfs);
205
206 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
207 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
208 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
209
210 static int nfsv3_commit_on_close = 0;
211 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
212 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
213
214 static int nfs_clean_pages_on_close = 1;
215 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
216 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
217
218 int nfs_directio_enable = 0;
219 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
220 &nfs_directio_enable, 0, "Enable NFS directio");
221
222 /*
223 * This sysctl allows other processes to mmap a file that has been opened
224 * O_DIRECT by a process. In general, having processes mmap the file while
225 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
226 * this by default to prevent DoS attacks - to prevent a malicious user from
227 * opening up files O_DIRECT preventing other users from mmap'ing these
228 * files. "Protected" environments where stricter consistency guarantees are
229 * required can disable this knob. The process that opened the file O_DIRECT
230 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
231 * meaningful.
232 */
233 int nfs_directio_allow_mmap = 1;
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
235 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
236
237 #if 0
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
239 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
240
241 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
242 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
243 #endif
244
245 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
246 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
247 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
248
249 /*
250 * SMP Locking Note :
251 * The list of locks after the description of the lock is the ordering
252 * of other locks acquired with the lock held.
253 * np->n_mtx : Protects the fields in the nfsnode.
254 VM Object Lock
255 VI_MTX (acquired indirectly)
256 * nmp->nm_mtx : Protects the fields in the nfsmount.
257 rep->r_mtx
258 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
259 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
260 nmp->nm_mtx
261 rep->r_mtx
262 * rep->r_mtx : Protects the fields in an nfsreq.
263 */
264
265 static int
266 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
267 struct ucred *cred)
268 {
269 const int v3 = 1;
270 u_int32_t *tl;
271 int error = 0, attrflag;
272
273 struct mbuf *mreq, *mrep, *md, *mb;
274 caddr_t bpos, dpos;
275 u_int32_t rmode;
276 struct nfsnode *np = VTONFS(vp);
277
278 nfsstats.rpccnt[NFSPROC_ACCESS]++;
279 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
280 mb = mreq;
281 bpos = mtod(mb, caddr_t);
282 nfsm_fhtom(vp, v3);
283 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
284 *tl = txdr_unsigned(wmode);
285 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
286 nfsm_postop_attr(vp, attrflag);
287 if (!error) {
288 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
289 rmode = fxdr_unsigned(u_int32_t, *tl);
290 mtx_lock(&np->n_mtx);
291 np->n_mode = rmode;
292 np->n_modeuid = cred->cr_uid;
293 np->n_modestamp = time_second;
294 mtx_unlock(&np->n_mtx);
295 }
296 m_freem(mrep);
297 nfsmout:
298 return (error);
299 }
300
301 /*
302 * nfs access vnode op.
303 * For nfs version 2, just return ok. File accesses may fail later.
304 * For nfs version 3, use the access rpc to check accessibility. If file modes
305 * are changed on the server, accesses might still fail later.
306 */
307 static int
308 nfs_access(struct vop_access_args *ap)
309 {
310 struct vnode *vp = ap->a_vp;
311 int error = 0;
312 u_int32_t mode, wmode;
313 int v3 = NFS_ISV3(vp);
314 struct nfsnode *np = VTONFS(vp);
315
316 /*
317 * Disallow write attempts on filesystems mounted read-only;
318 * unless the file is a socket, fifo, or a block or character
319 * device resident on the filesystem.
320 */
321 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
322 switch (vp->v_type) {
323 case VREG:
324 case VDIR:
325 case VLNK:
326 return (EROFS);
327 default:
328 break;
329 }
330 }
331 /*
332 * For nfs v3, check to see if we have done this recently, and if
333 * so return our cached result instead of making an ACCESS call.
334 * If not, do an access rpc, otherwise you are stuck emulating
335 * ufs_access() locally using the vattr. This may not be correct,
336 * since the server may apply other access criteria such as
337 * client uid-->server uid mapping that we do not know about.
338 */
339 if (v3) {
340 if (ap->a_accmode & VREAD)
341 mode = NFSV3ACCESS_READ;
342 else
343 mode = 0;
344 if (vp->v_type != VDIR) {
345 if (ap->a_accmode & VWRITE)
346 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
347 if (ap->a_accmode & VEXEC)
348 mode |= NFSV3ACCESS_EXECUTE;
349 } else {
350 if (ap->a_accmode & VWRITE)
351 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
352 NFSV3ACCESS_DELETE);
353 if (ap->a_accmode & VEXEC)
354 mode |= NFSV3ACCESS_LOOKUP;
355 }
356 /* XXX safety belt, only make blanket request if caching */
357 if (nfsaccess_cache_timeout > 0) {
358 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
359 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
360 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
361 } else {
362 wmode = mode;
363 }
364
365 /*
366 * Does our cached result allow us to give a definite yes to
367 * this request?
368 */
369 mtx_lock(&np->n_mtx);
370 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
371 (ap->a_cred->cr_uid == np->n_modeuid) &&
372 ((np->n_mode & mode) == mode)) {
373 nfsstats.accesscache_hits++;
374 } else {
375 /*
376 * Either a no, or a don't know. Go to the wire.
377 */
378 nfsstats.accesscache_misses++;
379 mtx_unlock(&np->n_mtx);
380 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
381 mtx_lock(&np->n_mtx);
382 if (!error) {
383 if ((np->n_mode & mode) != mode) {
384 error = EACCES;
385 }
386 }
387 }
388 mtx_unlock(&np->n_mtx);
389 return (error);
390 } else {
391 if ((error = nfsspec_access(ap)) != 0) {
392 return (error);
393 }
394 /*
395 * Attempt to prevent a mapped root from accessing a file
396 * which it shouldn't. We try to read a byte from the file
397 * if the user is root and the file is not zero length.
398 * After calling nfsspec_access, we should have the correct
399 * file size cached.
400 */
401 mtx_lock(&np->n_mtx);
402 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
403 && VTONFS(vp)->n_size > 0) {
404 struct iovec aiov;
405 struct uio auio;
406 char buf[1];
407
408 mtx_unlock(&np->n_mtx);
409 aiov.iov_base = buf;
410 aiov.iov_len = 1;
411 auio.uio_iov = &aiov;
412 auio.uio_iovcnt = 1;
413 auio.uio_offset = 0;
414 auio.uio_resid = 1;
415 auio.uio_segflg = UIO_SYSSPACE;
416 auio.uio_rw = UIO_READ;
417 auio.uio_td = ap->a_td;
418
419 if (vp->v_type == VREG)
420 error = nfs_readrpc(vp, &auio, ap->a_cred);
421 else if (vp->v_type == VDIR) {
422 char* bp;
423 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
424 aiov.iov_base = bp;
425 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
426 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
427 free(bp, M_TEMP);
428 } else if (vp->v_type == VLNK)
429 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
430 else
431 error = EACCES;
432 } else
433 mtx_unlock(&np->n_mtx);
434 return (error);
435 }
436 }
437
438 int nfs_otw_getattr_avoid = 0;
439
440 /*
441 * nfs open vnode op
442 * Check to see if the type is ok
443 * and that deletion is not in progress.
444 * For paged in text files, you will need to flush the page cache
445 * if consistency is lost.
446 */
447 /* ARGSUSED */
448 static int
449 nfs_open(struct vop_open_args *ap)
450 {
451 struct vnode *vp = ap->a_vp;
452 struct nfsnode *np = VTONFS(vp);
453 struct vattr vattr;
454 int error;
455 int fmode = ap->a_mode;
456
457 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
458 return (EOPNOTSUPP);
459
460 /*
461 * Get a valid lease. If cached data is stale, flush it.
462 */
463 mtx_lock(&np->n_mtx);
464 if (np->n_flag & NMODIFIED) {
465 mtx_unlock(&np->n_mtx);
466 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
467 if (error == EINTR || error == EIO)
468 return (error);
469 np->n_attrstamp = 0;
470 if (vp->v_type == VDIR)
471 np->n_direofoffset = 0;
472 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
473 if (error)
474 return (error);
475 mtx_lock(&np->n_mtx);
476 np->n_mtime = vattr.va_mtime;
477 mtx_unlock(&np->n_mtx);
478 } else {
479 struct thread *td = curthread;
480
481 if (np->n_ac_ts_syscalls != td->td_syscalls ||
482 np->n_ac_ts_tid != td->td_tid ||
483 td->td_proc == NULL ||
484 np->n_ac_ts_pid != td->td_proc->p_pid) {
485 np->n_attrstamp = 0;
486 }
487 mtx_unlock(&np->n_mtx);
488 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
489 if (error)
490 return (error);
491 mtx_lock(&np->n_mtx);
492 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
493 if (vp->v_type == VDIR)
494 np->n_direofoffset = 0;
495 mtx_unlock(&np->n_mtx);
496 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
497 if (error == EINTR || error == EIO) {
498 return (error);
499 }
500 mtx_lock(&np->n_mtx);
501 np->n_mtime = vattr.va_mtime;
502 }
503 mtx_unlock(&np->n_mtx);
504 }
505 /*
506 * If the object has >= 1 O_DIRECT active opens, we disable caching.
507 */
508 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
509 if (np->n_directio_opens == 0) {
510 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
511 if (error)
512 return (error);
513 mtx_lock(&np->n_mtx);
514 np->n_flag |= NNONCACHE;
515 mtx_unlock(&np->n_mtx);
516 }
517 np->n_directio_opens++;
518 }
519 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
520 return (0);
521 }
522
523 /*
524 * nfs close vnode op
525 * What an NFS client should do upon close after writing is a debatable issue.
526 * Most NFS clients push delayed writes to the server upon close, basically for
527 * two reasons:
528 * 1 - So that any write errors may be reported back to the client process
529 * doing the close system call. By far the two most likely errors are
530 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
531 * 2 - To put a worst case upper bound on cache inconsistency between
532 * multiple clients for the file.
533 * There is also a consistency problem for Version 2 of the protocol w.r.t.
534 * not being able to tell if other clients are writing a file concurrently,
535 * since there is no way of knowing if the changed modify time in the reply
536 * is only due to the write for this client.
537 * (NFS Version 3 provides weak cache consistency data in the reply that
538 * should be sufficient to detect and handle this case.)
539 *
540 * The current code does the following:
541 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
542 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
543 * or commit them (this satisfies 1 and 2 except for the
544 * case where the server crashes after this close but
545 * before the commit RPC, which is felt to be "good
546 * enough". Changing the last argument to nfs_flush() to
547 * a 1 would force a commit operation, if it is felt a
548 * commit is necessary now.
549 */
550 /* ARGSUSED */
551 static int
552 nfs_close(struct vop_close_args *ap)
553 {
554 struct vnode *vp = ap->a_vp;
555 struct nfsnode *np = VTONFS(vp);
556 int error = 0;
557 int fmode = ap->a_fflag;
558
559 if (vp->v_type == VREG) {
560 /*
561 * Examine and clean dirty pages, regardless of NMODIFIED.
562 * This closes a major hole in close-to-open consistency.
563 * We want to push out all dirty pages (and buffers) on
564 * close, regardless of whether they were dirtied by
565 * mmap'ed writes or via write().
566 */
567 if (nfs_clean_pages_on_close && vp->v_object) {
568 VM_OBJECT_LOCK(vp->v_object);
569 vm_object_page_clean(vp->v_object, 0, 0, 0);
570 VM_OBJECT_UNLOCK(vp->v_object);
571 }
572 mtx_lock(&np->n_mtx);
573 if (np->n_flag & NMODIFIED) {
574 mtx_unlock(&np->n_mtx);
575 if (NFS_ISV3(vp)) {
576 /*
577 * Under NFSv3 we have dirty buffers to dispose of. We
578 * must flush them to the NFS server. We have the option
579 * of waiting all the way through the commit rpc or just
580 * waiting for the initial write. The default is to only
581 * wait through the initial write so the data is in the
582 * server's cache, which is roughly similar to the state
583 * a standard disk subsystem leaves the file in on close().
584 *
585 * We cannot clear the NMODIFIED bit in np->n_flag due to
586 * potential races with other processes, and certainly
587 * cannot clear it if we don't commit.
588 */
589 int cm = nfsv3_commit_on_close ? 1 : 0;
590 error = nfs_flush(vp, MNT_WAIT, cm);
591 /* np->n_flag &= ~NMODIFIED; */
592 } else
593 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
594 mtx_lock(&np->n_mtx);
595 }
596 /*
597 * Invalidate the attribute cache in all cases.
598 * An open is going to fetch fresh attrs any way, other procs
599 * on this node that have file open will be forced to do an
600 * otw attr fetch, but this is safe.
601 */
602 np->n_attrstamp = 0;
603 if (np->n_flag & NWRITEERR) {
604 np->n_flag &= ~NWRITEERR;
605 error = np->n_error;
606 }
607 mtx_unlock(&np->n_mtx);
608 }
609 if (nfs_directio_enable)
610 KASSERT((np->n_directio_asyncwr == 0),
611 ("nfs_close: dirty unflushed (%d) directio buffers\n",
612 np->n_directio_asyncwr));
613 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
614 mtx_lock(&np->n_mtx);
615 KASSERT((np->n_directio_opens > 0),
616 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
617 np->n_directio_opens--;
618 if (np->n_directio_opens == 0)
619 np->n_flag &= ~NNONCACHE;
620 mtx_unlock(&np->n_mtx);
621 }
622 return (error);
623 }
624
625 /*
626 * nfs getattr call from vfs.
627 */
628 static int
629 nfs_getattr(struct vop_getattr_args *ap)
630 {
631 struct vnode *vp = ap->a_vp;
632 struct nfsnode *np = VTONFS(vp);
633 struct thread *td = curthread;
634 struct vattr *vap = ap->a_vap;
635 struct vattr vattr;
636 caddr_t bpos, dpos;
637 int error = 0;
638 struct mbuf *mreq, *mrep, *md, *mb;
639 int v3 = NFS_ISV3(vp);
640
641 /*
642 * Update local times for special files.
643 */
644 mtx_lock(&np->n_mtx);
645 if (np->n_flag & (NACC | NUPD))
646 np->n_flag |= NCHG;
647 mtx_unlock(&np->n_mtx);
648 /*
649 * First look in the cache.
650 */
651 if (nfs_getattrcache(vp, &vattr) == 0)
652 goto nfsmout;
653 if (v3 && nfsaccess_cache_timeout > 0) {
654 nfsstats.accesscache_misses++;
655 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred);
656 if (nfs_getattrcache(vp, &vattr) == 0)
657 goto nfsmout;
658 }
659 nfsstats.rpccnt[NFSPROC_GETATTR]++;
660 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
661 mb = mreq;
662 bpos = mtod(mb, caddr_t);
663 nfsm_fhtom(vp, v3);
664 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
665 if (!error) {
666 nfsm_loadattr(vp, &vattr);
667 }
668 m_freem(mrep);
669 nfsmout:
670 vap->va_type = vattr.va_type;
671 vap->va_mode = vattr.va_mode;
672 vap->va_nlink = vattr.va_nlink;
673 vap->va_uid = vattr.va_uid;
674 vap->va_gid = vattr.va_gid;
675 vap->va_fsid = vattr.va_fsid;
676 vap->va_fileid = vattr.va_fileid;
677 vap->va_size = vattr.va_size;
678 vap->va_blocksize = vattr.va_blocksize;
679 vap->va_atime = vattr.va_atime;
680 vap->va_mtime = vattr.va_mtime;
681 vap->va_ctime = vattr.va_ctime;
682 vap->va_gen = vattr.va_gen;
683 vap->va_flags = vattr.va_flags;
684 vap->va_rdev = vattr.va_rdev;
685 vap->va_bytes = vattr.va_bytes;
686 vap->va_filerev = vattr.va_filerev;
687
688 return (error);
689 }
690
691 /*
692 * nfs setattr call.
693 */
694 static int
695 nfs_setattr(struct vop_setattr_args *ap)
696 {
697 struct vnode *vp = ap->a_vp;
698 struct nfsnode *np = VTONFS(vp);
699 struct vattr *vap = ap->a_vap;
700 struct thread *td = curthread;
701 int error = 0;
702 u_quad_t tsize;
703
704 #ifndef nolint
705 tsize = (u_quad_t)0;
706 #endif
707
708 /*
709 * Setting of flags and marking of atimes are not supported.
710 */
711 if (vap->va_flags != VNOVAL || (vap->va_vaflags & VA_MARK_ATIME))
712 return (EOPNOTSUPP);
713
714 /*
715 * Disallow write attempts if the filesystem is mounted read-only.
716 */
717 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
718 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
719 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
720 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
721 error = EROFS;
722 goto out;
723 }
724 if (vap->va_size != VNOVAL) {
725 switch (vp->v_type) {
726 case VDIR:
727 return (EISDIR);
728 case VCHR:
729 case VBLK:
730 case VSOCK:
731 case VFIFO:
732 if (vap->va_mtime.tv_sec == VNOVAL &&
733 vap->va_atime.tv_sec == VNOVAL &&
734 vap->va_mode == (mode_t)VNOVAL &&
735 vap->va_uid == (uid_t)VNOVAL &&
736 vap->va_gid == (gid_t)VNOVAL)
737 return (0);
738 vap->va_size = VNOVAL;
739 break;
740 default:
741 /*
742 * Disallow write attempts if the filesystem is
743 * mounted read-only.
744 */
745 if (vp->v_mount->mnt_flag & MNT_RDONLY)
746 return (EROFS);
747 /*
748 * We run vnode_pager_setsize() early (why?),
749 * we must set np->n_size now to avoid vinvalbuf
750 * V_SAVE races that might setsize a lower
751 * value.
752 */
753 mtx_lock(&np->n_mtx);
754 tsize = np->n_size;
755 mtx_unlock(&np->n_mtx);
756 error = nfs_meta_setsize(vp, ap->a_cred, td,
757 vap->va_size);
758 mtx_lock(&np->n_mtx);
759 if (np->n_flag & NMODIFIED) {
760 tsize = np->n_size;
761 mtx_unlock(&np->n_mtx);
762 if (vap->va_size == 0)
763 error = nfs_vinvalbuf(vp, 0, td, 1);
764 else
765 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
766 if (error) {
767 vnode_pager_setsize(vp, tsize);
768 goto out;
769 }
770 } else
771 mtx_unlock(&np->n_mtx);
772 /*
773 * np->n_size has already been set to vap->va_size
774 * in nfs_meta_setsize(). We must set it again since
775 * nfs_loadattrcache() could be called through
776 * nfs_meta_setsize() and could modify np->n_size.
777 */
778 mtx_lock(&np->n_mtx);
779 np->n_vattr.va_size = np->n_size = vap->va_size;
780 mtx_unlock(&np->n_mtx);
781 };
782 } else {
783 mtx_lock(&np->n_mtx);
784 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
785 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
786 mtx_unlock(&np->n_mtx);
787 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
788 (error == EINTR || error == EIO))
789 return error;
790 } else
791 mtx_unlock(&np->n_mtx);
792 }
793 error = nfs_setattrrpc(vp, vap, ap->a_cred);
794 if (error && vap->va_size != VNOVAL) {
795 mtx_lock(&np->n_mtx);
796 np->n_size = np->n_vattr.va_size = tsize;
797 vnode_pager_setsize(vp, tsize);
798 mtx_unlock(&np->n_mtx);
799 }
800 out:
801 return (error);
802 }
803
804 /*
805 * Do an nfs setattr rpc.
806 */
807 static int
808 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
809 {
810 struct nfsv2_sattr *sp;
811 struct nfsnode *np = VTONFS(vp);
812 caddr_t bpos, dpos;
813 u_int32_t *tl;
814 int error = 0, wccflag = NFSV3_WCCRATTR;
815 struct mbuf *mreq, *mrep, *md, *mb;
816 int v3 = NFS_ISV3(vp);
817
818 nfsstats.rpccnt[NFSPROC_SETATTR]++;
819 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
820 mb = mreq;
821 bpos = mtod(mb, caddr_t);
822 nfsm_fhtom(vp, v3);
823 if (v3) {
824 nfsm_v3attrbuild(vap, TRUE);
825 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
826 *tl = nfs_false;
827 } else {
828 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
829 if (vap->va_mode == (mode_t)VNOVAL)
830 sp->sa_mode = nfs_xdrneg1;
831 else
832 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
833 if (vap->va_uid == (uid_t)VNOVAL)
834 sp->sa_uid = nfs_xdrneg1;
835 else
836 sp->sa_uid = txdr_unsigned(vap->va_uid);
837 if (vap->va_gid == (gid_t)VNOVAL)
838 sp->sa_gid = nfs_xdrneg1;
839 else
840 sp->sa_gid = txdr_unsigned(vap->va_gid);
841 sp->sa_size = txdr_unsigned(vap->va_size);
842 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
843 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
844 }
845 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
846 if (v3) {
847 np->n_modestamp = 0;
848 nfsm_wcc_data(vp, wccflag);
849 } else
850 nfsm_loadattr(vp, NULL);
851 m_freem(mrep);
852 nfsmout:
853 return (error);
854 }
855
856 /*
857 * nfs lookup call, one step at a time...
858 * First look in cache
859 * If not found, unlock the directory nfsnode and do the rpc
860 */
861 static int
862 nfs_lookup(struct vop_lookup_args *ap)
863 {
864 struct componentname *cnp = ap->a_cnp;
865 struct vnode *dvp = ap->a_dvp;
866 struct vnode **vpp = ap->a_vpp;
867 int flags = cnp->cn_flags;
868 struct vnode *newvp;
869 struct nfsmount *nmp;
870 caddr_t bpos, dpos;
871 struct mbuf *mreq, *mrep, *md, *mb;
872 long len;
873 nfsfh_t *fhp;
874 struct nfsnode *np;
875 int error = 0, attrflag, fhsize;
876 int v3 = NFS_ISV3(dvp);
877 struct thread *td = cnp->cn_thread;
878
879 *vpp = NULLVP;
880 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
881 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
882 return (EROFS);
883 if (dvp->v_type != VDIR)
884 return (ENOTDIR);
885 nmp = VFSTONFS(dvp->v_mount);
886 np = VTONFS(dvp);
887 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
888 *vpp = NULLVP;
889 return (error);
890 }
891 error = cache_lookup(dvp, vpp, cnp);
892</ |