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: releng/7.4/sys/nfsclient/nfs_vnops.c 214986 2010-11-08 15:22:20Z kib $");
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
64 #include <vm/vm.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_extern.h>
67 #include <vm/vm_object.h>
68
69 #include <fs/fifofs/fifo.h>
70
71 #include <rpc/rpcclnt.h>
72
73 #include <nfs/rpcv2.h>
74 #include <nfs/nfsproto.h>
75 #include <nfsclient/nfs.h>
76 #include <nfsclient/nfsnode.h>
77 #include <nfsclient/nfsmount.h>
78 #include <nfsclient/nfs_lock.h>
79 #include <nfs/xdr_subs.h>
80 #include <nfsclient/nfsm_subs.h>
81
82 #include <net/if.h>
83 #include <netinet/in.h>
84 #include <netinet/in_var.h>
85
86 /* Defs */
87 #define TRUE 1
88 #define FALSE 0
89
90 /*
91 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
92 * calls are not in getblk() and brelse() so that they would not be necessary
93 * here.
94 */
95 #ifndef B_VMIO
96 #define vfs_busy_pages(bp, f)
97 #endif
98
99 static vop_read_t nfsfifo_read;
100 static vop_write_t nfsfifo_write;
101 static vop_close_t nfsfifo_close;
102 static int nfs_flush(struct vnode *, int, struct thread *,
103 int);
104 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
105 struct thread *);
106 static vop_lookup_t nfs_lookup;
107 static vop_create_t nfs_create;
108 static vop_mknod_t nfs_mknod;
109 static vop_open_t nfs_open;
110 static vop_close_t nfs_close;
111 static vop_access_t nfs_access;
112 static vop_getattr_t nfs_getattr;
113 static vop_setattr_t nfs_setattr;
114 static vop_read_t nfs_read;
115 static vop_fsync_t nfs_fsync;
116 static vop_remove_t nfs_remove;
117 static vop_link_t nfs_link;
118 static vop_rename_t nfs_rename;
119 static vop_mkdir_t nfs_mkdir;
120 static vop_rmdir_t nfs_rmdir;
121 static vop_symlink_t nfs_symlink;
122 static vop_readdir_t nfs_readdir;
123 static vop_strategy_t nfs_strategy;
124 static int nfs_lookitup(struct vnode *, const char *, int,
125 struct ucred *, struct thread *, struct nfsnode **);
126 static int nfs_sillyrename(struct vnode *, struct vnode *,
127 struct componentname *);
128 static vop_access_t nfsspec_access;
129 static vop_readlink_t nfs_readlink;
130 static vop_print_t nfs_print;
131 static vop_advlock_t nfs_advlock;
132 static vop_advlockasync_t nfs_advlockasync;
133
134 /*
135 * Global vfs data structures for nfs
136 */
137 struct vop_vector nfs_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 = nfs_getpages,
147 .vop_putpages = nfs_putpages,
148 .vop_inactive = nfs_inactive,
149 .vop_lease = VOP_NULL,
150 .vop_link = nfs_link,
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 = nfs_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 = nfs_write,
167 };
168
169 struct vop_vector nfs_fifoops = {
170 .vop_default = &fifo_specops,
171 .vop_access = nfsspec_access,
172 .vop_close = nfsfifo_close,
173 .vop_fsync = nfs_fsync,
174 .vop_getattr = nfs_getattr,
175 .vop_inactive = nfs_inactive,
176 .vop_print = nfs_print,
177 .vop_read = nfsfifo_read,
178 .vop_reclaim = nfs_reclaim,
179 .vop_setattr = nfs_setattr,
180 .vop_write = nfsfifo_write,
181 };
182
183 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
184 struct componentname *cnp, struct vattr *vap);
185 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
186 struct ucred *cred, struct thread *td);
187 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
188 int fnamelen, struct vnode *tdvp,
189 const char *tnameptr, int tnamelen,
190 struct ucred *cred, struct thread *td);
191 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
192 struct sillyrename *sp);
193
194 /*
195 * Global variables
196 */
197 struct mtx nfs_iod_mtx;
198 enum nfsiod_state nfs_iodwant[NFS_MAXASYNCDAEMON];
199 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
200 int nfs_numasync = 0;
201 vop_advlock_t *nfs_advlock_p = nfs_dolock;
202 vop_reclaim_t *nfs_reclaim_p = NULL;
203 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
204
205 SYSCTL_DECL(_vfs_nfs);
206
207 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
208 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
209 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
210
211 static int nfs_prime_access_cache = 1;
212 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
213 &nfs_prime_access_cache, 0,
214 "Prime NFS ACCESS cache when fetching attributes");
215
216 static int nfsv3_commit_on_close = 0;
217 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
218 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
219
220 static int nfs_clean_pages_on_close = 1;
221 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
222 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
223
224 int nfs_directio_enable = 0;
225 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
226 &nfs_directio_enable, 0, "Enable NFS directio");
227
228 static u_int negnametimeo = NFS_DEFAULT_NEGNAMETIMEO;
229 SYSCTL_UINT(_vfs_nfs, OID_AUTO, negative_name_timeout, CTLFLAG_RW,
230 &negnametimeo, 0, "Negative name cache entry timeout");
231
232 /*
233 * This sysctl allows other processes to mmap a file that has been opened
234 * O_DIRECT by a process. In general, having processes mmap the file while
235 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
236 * this by default to prevent DoS attacks - to prevent a malicious user from
237 * opening up files O_DIRECT preventing other users from mmap'ing these
238 * files. "Protected" environments where stricter consistency guarantees are
239 * required can disable this knob. The process that opened the file O_DIRECT
240 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
241 * meaningful.
242 */
243 int nfs_directio_allow_mmap = 1;
244 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
245 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
246
247 #if 0
248 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
249 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
250
251 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
252 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
253 #endif
254
255 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
256 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
257 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
258
259 /*
260 * SMP Locking Note :
261 * The list of locks after the description of the lock is the ordering
262 * of other locks acquired with the lock held.
263 * np->n_mtx : Protects the fields in the nfsnode.
264 VM Object Lock
265 VI_MTX (acquired indirectly)
266 * nmp->nm_mtx : Protects the fields in the nfsmount.
267 rep->r_mtx
268 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
269 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
270 nmp->nm_mtx
271 rep->r_mtx
272 * rep->r_mtx : Protects the fields in an nfsreq.
273 */
274
275 static int
276 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
277 struct ucred *cred)
278 {
279 const int v3 = 1;
280 u_int32_t *tl;
281 int error = 0, attrflag;
282
283 struct mbuf *mreq, *mrep, *md, *mb;
284 caddr_t bpos, dpos;
285 u_int32_t rmode;
286 struct nfsnode *np = VTONFS(vp);
287
288 nfsstats.rpccnt[NFSPROC_ACCESS]++;
289 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
290 mb = mreq;
291 bpos = mtod(mb, caddr_t);
292 nfsm_fhtom(vp, v3);
293 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
294 *tl = txdr_unsigned(wmode);
295 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
296 nfsm_postop_attr(vp, attrflag);
297 if (!error) {
298 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
299 rmode = fxdr_unsigned(u_int32_t, *tl);
300 mtx_lock(&np->n_mtx);
301 np->n_mode = rmode;
302 np->n_modeuid = cred->cr_uid;
303 np->n_modestamp = time_second;
304 mtx_unlock(&np->n_mtx);
305 }
306 m_freem(mrep);
307 nfsmout:
308 return (error);
309 }
310
311 /*
312 * nfs access vnode op.
313 * For nfs version 2, just return ok. File accesses may fail later.
314 * For nfs version 3, use the access rpc to check accessibility. If file modes
315 * are changed on the server, accesses might still fail later.
316 */
317 static int
318 nfs_access(struct vop_access_args *ap)
319 {
320 struct vnode *vp = ap->a_vp;
321 int error = 0;
322 u_int32_t mode, wmode;
323 int v3 = NFS_ISV3(vp);
324 struct nfsnode *np = VTONFS(vp);
325
326 /*
327 * Disallow write attempts on filesystems mounted read-only;
328 * unless the file is a socket, fifo, or a block or character
329 * device resident on the filesystem.
330 */
331 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
332 switch (vp->v_type) {
333 case VREG:
334 case VDIR:
335 case VLNK:
336 return (EROFS);
337 default:
338 break;
339 }
340 }
341 /*
342 * For nfs v3, check to see if we have done this recently, and if
343 * so return our cached result instead of making an ACCESS call.
344 * If not, do an access rpc, otherwise you are stuck emulating
345 * ufs_access() locally using the vattr. This may not be correct,
346 * since the server may apply other access criteria such as
347 * client uid-->server uid mapping that we do not know about.
348 */
349 if (v3) {
350 if (ap->a_mode & VREAD)
351 mode = NFSV3ACCESS_READ;
352 else
353 mode = 0;
354 if (vp->v_type != VDIR) {
355 if (ap->a_mode & VWRITE)
356 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
357 if (ap->a_mode & VEXEC)
358 mode |= NFSV3ACCESS_EXECUTE;
359 } else {
360 if (ap->a_mode & VWRITE)
361 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
362 NFSV3ACCESS_DELETE);
363 if (ap->a_mode & VEXEC)
364 mode |= NFSV3ACCESS_LOOKUP;
365 }
366 /* XXX safety belt, only make blanket request if caching */
367 if (nfsaccess_cache_timeout > 0) {
368 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
369 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
370 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
371 } else {
372 wmode = mode;
373 }
374
375 /*
376 * Does our cached result allow us to give a definite yes to
377 * this request?
378 */
379 mtx_lock(&np->n_mtx);
380 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
381 (ap->a_cred->cr_uid == np->n_modeuid) &&
382 ((np->n_mode & mode) == mode)) {
383 nfsstats.accesscache_hits++;
384 } else {
385 /*
386 * Either a no, or a don't know. Go to the wire.
387 */
388 nfsstats.accesscache_misses++;
389 mtx_unlock(&np->n_mtx);
390 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
391 mtx_lock(&np->n_mtx);
392 if (!error) {
393 if ((np->n_mode & mode) != mode) {
394 error = EACCES;
395 }
396 }
397 }
398 mtx_unlock(&np->n_mtx);
399 return (error);
400 } else {
401 if ((error = nfsspec_access(ap)) != 0) {
402 return (error);
403 }
404 /*
405 * Attempt to prevent a mapped root from accessing a file
406 * which it shouldn't. We try to read a byte from the file
407 * if the user is root and the file is not zero length.
408 * After calling nfsspec_access, we should have the correct
409 * file size cached.
410 */
411 mtx_lock(&np->n_mtx);
412 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
413 && VTONFS(vp)->n_size > 0) {
414 struct iovec aiov;
415 struct uio auio;
416 char buf[1];
417
418 mtx_unlock(&np->n_mtx);
419 aiov.iov_base = buf;
420 aiov.iov_len = 1;
421 auio.uio_iov = &aiov;
422 auio.uio_iovcnt = 1;
423 auio.uio_offset = 0;
424 auio.uio_resid = 1;
425 auio.uio_segflg = UIO_SYSSPACE;
426 auio.uio_rw = UIO_READ;
427 auio.uio_td = ap->a_td;
428
429 if (vp->v_type == VREG)
430 error = nfs_readrpc(vp, &auio, ap->a_cred);
431 else if (vp->v_type == VDIR) {
432 char* bp;
433 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
434 aiov.iov_base = bp;
435 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
436 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
437 free(bp, M_TEMP);
438 } else if (vp->v_type == VLNK)
439 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
440 else
441 error = EACCES;
442 } else
443 mtx_unlock(&np->n_mtx);
444 return (error);
445 }
446 }
447
448 int nfs_otw_getattr_avoid = 0;
449
450 /*
451 * nfs open vnode op
452 * Check to see if the type is ok
453 * and that deletion is not in progress.
454 * For paged in text files, you will need to flush the page cache
455 * if consistency is lost.
456 */
457 /* ARGSUSED */
458 static int
459 nfs_open(struct vop_open_args *ap)
460 {
461 struct vnode *vp = ap->a_vp;
462 struct nfsnode *np = VTONFS(vp);
463 struct vattr vattr;
464 int error;
465 int fmode = ap->a_mode;
466
467 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
468 return (EOPNOTSUPP);
469
470 /*
471 * Get a valid lease. If cached data is stale, flush it.
472 */
473 mtx_lock(&np->n_mtx);
474 if (np->n_flag & NMODIFIED) {
475 mtx_unlock(&np->n_mtx);
476 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
477 if (error == EINTR || error == EIO)
478 return (error);
479 mtx_lock(&np->n_mtx);
480 np->n_attrstamp = 0;
481 if (vp->v_type == VDIR)
482 np->n_direofoffset = 0;
483 mtx_unlock(&np->n_mtx);
484 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
485 if (error)
486 return (error);
487 mtx_lock(&np->n_mtx);
488 np->n_mtime = vattr.va_mtime;
489 mtx_unlock(&np->n_mtx);
490 } else {
491 mtx_unlock(&np->n_mtx);
492 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
493 if (error)
494 return (error);
495 mtx_lock(&np->n_mtx);
496 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
497 if (vp->v_type == VDIR)
498 np->n_direofoffset = 0;
499 mtx_unlock(&np->n_mtx);
500 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
501 if (error == EINTR || error == EIO) {
502 return (error);
503 }
504 mtx_lock(&np->n_mtx);
505 np->n_mtime = vattr.va_mtime;
506 }
507 mtx_unlock(&np->n_mtx);
508 }
509 /*
510 * If the object has >= 1 O_DIRECT active opens, we disable caching.
511 */
512 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
513 if (np->n_directio_opens == 0) {
514 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
515 if (error)
516 return (error);
517 mtx_lock(&np->n_mtx);
518 np->n_flag |= NNONCACHE;
519 mtx_unlock(&np->n_mtx);
520 }
521 np->n_directio_opens++;
522 }
523 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
524 return (0);
525 }
526
527 /*
528 * nfs close vnode op
529 * What an NFS client should do upon close after writing is a debatable issue.
530 * Most NFS clients push delayed writes to the server upon close, basically for
531 * two reasons:
532 * 1 - So that any write errors may be reported back to the client process
533 * doing the close system call. By far the two most likely errors are
534 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
535 * 2 - To put a worst case upper bound on cache inconsistency between
536 * multiple clients for the file.
537 * There is also a consistency problem for Version 2 of the protocol w.r.t.
538 * not being able to tell if other clients are writing a file concurrently,
539 * since there is no way of knowing if the changed modify time in the reply
540 * is only due to the write for this client.
541 * (NFS Version 3 provides weak cache consistency data in the reply that
542 * should be sufficient to detect and handle this case.)
543 *
544 * The current code does the following:
545 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
546 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
547 * or commit them (this satisfies 1 and 2 except for the
548 * case where the server crashes after this close but
549 * before the commit RPC, which is felt to be "good
550 * enough". Changing the last argument to nfs_flush() to
551 * a 1 would force a commit operation, if it is felt a
552 * commit is necessary now.
553 */
554 /* ARGSUSED */
555 static int
556 nfs_close(struct vop_close_args *ap)
557 {
558 struct vnode *vp = ap->a_vp;
559 struct nfsnode *np = VTONFS(vp);
560 int error = 0;
561 int fmode = ap->a_fflag;
562
563 if (vp->v_type == VREG) {
564 /*
565 * Examine and clean dirty pages, regardless of NMODIFIED.
566 * This closes a major hole in close-to-open consistency.
567 * We want to push out all dirty pages (and buffers) on
568 * close, regardless of whether they were dirtied by
569 * mmap'ed writes or via write().
570 */
571 if (nfs_clean_pages_on_close && vp->v_object) {
572 VM_OBJECT_LOCK(vp->v_object);
573 vm_object_page_clean(vp->v_object, 0, 0, 0);
574 VM_OBJECT_UNLOCK(vp->v_object);
575 }
576 mtx_lock(&np->n_mtx);
577 if (np->n_flag & NMODIFIED) {
578 mtx_unlock(&np->n_mtx);
579 if (NFS_ISV3(vp)) {
580 /*
581 * Under NFSv3 we have dirty buffers to dispose of. We
582 * must flush them to the NFS server. We have the option
583 * of waiting all the way through the commit rpc or just
584 * waiting for the initial write. The default is to only
585 * wait through the initial write so the data is in the
586 * server's cache, which is roughly similar to the state
587 * a standard disk subsystem leaves the file in on close().
588 *
589 * We cannot clear the NMODIFIED bit in np->n_flag due to
590 * potential races with other processes, and certainly
591 * cannot clear it if we don't commit.
592 */
593 int cm = nfsv3_commit_on_close ? 1 : 0;
594 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
595 /* np->n_flag &= ~NMODIFIED; */
596 } else
597 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
598 mtx_lock(&np->n_mtx);
599 }
600 if (np->n_flag & NWRITEERR) {
601 np->n_flag &= ~NWRITEERR;
602 error = np->n_error;
603 }
604 mtx_unlock(&np->n_mtx);
605 }
606 if (nfs_directio_enable)
607 KASSERT((np->n_directio_asyncwr == 0),
608 ("nfs_close: dirty unflushed (%d) directio buffers\n",
609 np->n_directio_asyncwr));
610 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
611 mtx_lock(&np->n_mtx);
612 KASSERT((np->n_directio_opens > 0),
613 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
614 np->n_directio_opens--;
615 if (np->n_directio_opens == 0)
616 np->n_flag &= ~NNONCACHE;
617 mtx_unlock(&np->n_mtx);
618 }
619 return (error);
620 }
621
622 /*
623 * nfs getattr call from vfs.
624 */
625 static int
626 nfs_getattr(struct vop_getattr_args *ap)
627 {
628 struct vnode *vp = ap->a_vp;
629 struct nfsnode *np = VTONFS(vp);
630 struct vattr *vap = ap->a_vap;
631 struct vattr vattr;
632 caddr_t bpos, dpos;
633 int error = 0;
634 struct mbuf *mreq, *mrep, *md, *mb;
635 int v3 = NFS_ISV3(vp);
636
637 /*
638 * Update local times for special files.
639 */
640 mtx_lock(&np->n_mtx);
641 if (np->n_flag & (NACC | NUPD))
642 np->n_flag |= NCHG;
643 mtx_unlock(&np->n_mtx);
644 /*
645 * First look in the cache.
646 */
647 if (nfs_getattrcache(vp, &vattr) == 0)
648 goto nfsmout;
649 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
650 nfsstats.accesscache_misses++;
651 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
652 if (nfs_getattrcache(vp, &vattr) == 0)
653 goto nfsmout;
654 }
655 nfsstats.rpccnt[NFSPROC_GETATTR]++;
656 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
657 mb = mreq;
658 bpos = mtod(mb, caddr_t);
659 nfsm_fhtom(vp, v3);
660 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
661 if (!error) {
662 nfsm_loadattr(vp, &vattr);
663 }
664 m_freem(mrep);
665 nfsmout:
666 vap->va_type = vattr.va_type;
667 vap->va_mode = vattr.va_mode;
668 vap->va_nlink = vattr.va_nlink;
669 vap->va_uid = vattr.va_uid;
670 vap->va_gid = vattr.va_gid;
671 vap->va_fsid = vattr.va_fsid;
672 vap->va_fileid = vattr.va_fileid;
673 vap->va_size = vattr.va_size;
674 vap->va_blocksize = vattr.va_blocksize;
675 vap->va_atime = vattr.va_atime;
676 vap->va_mtime = vattr.va_mtime;
677 vap->va_ctime = vattr.va_ctime;
678 vap->va_gen = vattr.va_gen;
679 vap->va_flags = vattr.va_flags;
680 vap->va_rdev = vattr.va_rdev;
681 vap->va_bytes = vattr.va_bytes;
682 vap->va_filerev = vattr.va_filerev;
683
684 return (error);
685 }
686
687 /*
688 * nfs setattr call.
689 */
690 static int
691 nfs_setattr(struct vop_setattr_args *ap)
692 {
693 struct vnode *vp = ap->a_vp;
694 struct nfsnode *np = VTONFS(vp);
695 struct vattr *vap = ap->a_vap;
696 int error = 0;
697 u_quad_t tsize;
698
699 #ifndef nolint
700 tsize = (u_quad_t)0;
701 #endif
702
703 /*
704 * Setting of flags and marking of atimes are not supported.
705 */
706 if (vap->va_flags != VNOVAL || (vap->va_vaflags & VA_MARK_ATIME))
707 return (EOPNOTSUPP);
708
709 /*
710 * Disallow write attempts if the filesystem is mounted read-only.
711 */
712 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
713 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
714 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
715 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
716 error = EROFS;
717 goto out;
718 }
719 if (vap->va_size != VNOVAL) {
720 switch (vp->v_type) {
721 case VDIR:
722 return (EISDIR);
723 case VCHR:
724 case VBLK:
725 case VSOCK:
726 case VFIFO:
727 if (vap->va_mtime.tv_sec == VNOVAL &&
728 vap->va_atime.tv_sec == VNOVAL &&
729 vap->va_mode == (mode_t)VNOVAL &&
730 vap->va_uid == (uid_t)VNOVAL &&
731 vap->va_gid == (gid_t)VNOVAL)
732 return (0);
733 vap->va_size = VNOVAL;
734 break;
735 default:
736 /*
737 * Disallow write attempts if the filesystem is
738 * mounted read-only.
739 */
740 if (vp->v_mount->mnt_flag & MNT_RDONLY)
741 return (EROFS);
742 /*
743 * We run vnode_pager_setsize() early (why?),
744 * we must set np->n_size now to avoid vinvalbuf
745 * V_SAVE races that might setsize a lower
746 * value.
747 */
748 mtx_lock(&np->n_mtx);
749 tsize = np->n_size;
750 mtx_unlock(&np->n_mtx);
751 error = nfs_meta_setsize(vp, ap->a_cred,
752 ap->a_td, vap->va_size);
753 mtx_lock(&np->n_mtx);
754 if (np->n_flag & NMODIFIED) {
755 tsize = np->n_size;
756 mtx_unlock(&np->n_mtx);
757 if (vap->va_size == 0)
758 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
759 else
760 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
761 if (error) {
762 vnode_pager_setsize(vp, tsize);
763 goto out;
764 }
765 } else
766 mtx_unlock(&np->n_mtx);
767 /*
768 * np->n_size has already been set to vap->va_size
769 * in nfs_meta_setsize(). We must set it again since
770 * nfs_loadattrcache() could be called through
771 * nfs_meta_setsize() and could modify np->n_size.
772 */
773 mtx_lock(&np->n_mtx);
774 np->n_vattr.va_size = np->n_size = vap->va_size;
775 mtx_unlock(&np->n_mtx);
776 };
777 } else {
778 mtx_lock(&np->n_mtx);
779 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
780 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
781 mtx_unlock(&np->n_mtx);
782 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) != 0 &&
783 (error == EINTR || error == EIO))
784 return error;
785 } else
786 mtx_unlock(&np->n_mtx);
787 }
788 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
789 if (error && vap->va_size != VNOVAL) {
790 mtx_lock(&np->n_mtx);
791 np->n_size = np->n_vattr.va_size = tsize;
792 vnode_pager_setsize(vp, tsize);
793 mtx_unlock(&np->n_mtx);
794 }
795 out:
796 return (error);
797 }
798
799 /*
800 * Do an nfs setattr rpc.
801 */
802 static int
803 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
804 struct thread *td)
805 {
806 struct nfsv2_sattr *sp;
807 struct nfsnode *np = VTONFS(vp);
808 caddr_t bpos, dpos;
809 u_int32_t *tl;
810 int error = 0, wccflag = NFSV3_WCCRATTR;
811 struct mbuf *mreq, *mrep, *md, *mb;
812 int v3 = NFS_ISV3(vp);
813
814 nfsstats.rpccnt[NFSPROC_SETATTR]++;
815 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
816 mb = mreq;
817 bpos = mtod(mb, caddr_t);
818 nfsm_fhtom(vp, v3);
819 if (v3) {
820 nfsm_v3attrbuild(vap, TRUE);
821 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
822 *tl = nfs_false;
823 } else {
824 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
825 if (vap->va_mode == (mode_t)VNOVAL)
826 sp->sa_mode = nfs_xdrneg1;
827 else
828 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
829 if (vap->va_uid == (uid_t)VNOVAL)
830 sp->sa_uid = nfs_xdrneg1;
831 else
832 sp->sa_uid = txdr_unsigned(vap->va_uid);
833 if (vap->va_gid == (gid_t)VNOVAL)
834 sp->sa_gid = nfs_xdrneg1;
835 else
836 sp->sa_gid = txdr_unsigned(vap->va_gid);
837 sp->sa_size = txdr_unsigned(vap->va_size);
838 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
839 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
840 }
841 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
842 if (v3) {
843 np->n_modestamp = 0;
844 nfsm_wcc_data(vp, wccflag);
845 } else
846 nfsm_loadattr(vp, NULL);
847 m_freem(mrep);
848 nfsmout:
849 return (error);
850 }
851
852 /*
853 * nfs lookup call, one step at a time...
854 * First look in cache
855 * If not found, unlock the directory nfsnode and do the rpc
856 */
857 static int
858 nfs_lookup(struct vop_lookup_args *ap)
859 {
860 struct componentname *cnp = ap->a_cnp;
861 struct vnode *dvp = ap->a_dvp;
862 struct vnode **vpp = ap->a_vpp;
863 struct vattr vattr;
864 struct timespec dmtime;
865 int flags = cnp->cn_flags;
866 struct vnode *newvp;
867 struct nfsmount *nmp;
868 caddr_t bpos, dpos;
869 struct mbuf *mreq, *mrep, *md, *mb;
870 long len;
871 nfsfh_t *fhp;
872 struct nfsnode *np, *newnp;
873 int error = 0, attrflag, fhsize;
874 int v3 = NFS_ISV3(dvp);
875 struct thread *td = cnp->cn_thread;
876
877 *vpp = NULLVP;
878 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
879 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
880 return (EROFS);
881 if (dvp->v_type != VDIR)
882 return (ENOTDIR);
883 nmp = VFSTONFS(dvp->v_mount);
884 np = VTONFS(dvp);
885 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
886 *vpp = NULLVP;
887 return (error);
888 }
889 error = cache_lookup(dvp, vpp, cnp);
890 if (error > 0 && error != ENOENT)
891 return (error);
892 if (error == -1) {
893 /*
894 * We only accept a positive hit in the cache if the
895 * change time of the file matches our cached copy.
896 * Otherwise, we discard the cache entry and fallback
897 * to doing a lookup RPC.
898 *
899 * To better handle stale file handles and attributes,
900 * clear the attribute cache of this node if it is a
901 * leaf component, part of an open() call, and not
902 * locally modified before fetching the attributes.
903 * This should allow stale file handles to be detected
904 * here where we can fall back to a LOOKUP RPC to
905 * recover rather than having nfs_open() detect the
906 * stale file handle and failing open(2) with ESTALE.
907 */
908 newvp = *vpp;
909 newnp = VTONFS(newvp);
910 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
911 !(newnp->n_flag & NMODIFIED)) {
912 mtx_lock(&newnp->n_mtx);
913 newnp->n_attrstamp = 0;
914 mtx_unlock(&newnp->n_mtx);
915 }
916 if (VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td) == 0 &&
917 timespeccmp(&vattr.va_ctime, &newnp->n_ctime, ==)) {
918 nfsstats.lookupcache_hits++;
919 if (cnp->cn_nameiop != LOOKUP &&
920 (flags & ISLASTCN))
921 cnp->cn_flags |= SAVENAME;
922 return (0);
923 }
924 cache_purge(newvp);
925 if (dvp != newvp)
926 vput(newvp);
927 else
928 vrele(newvp);
929 *vpp = NULLVP;
930 } else if (error == ENOENT) {
931 /*
932 * We only accept a negative hit in the cache if the
933 * modification time of the parent directory matches
934 * our cached copy. Otherwise, we discard all of the
935 * negative cache entries for this directory. We also
936 * only trust -ve cache entries for less than
937 * negnametimeo seconds.
938 */
939 if ((u_int)(ticks - np->n_dmtime_ticks) < (negnametimeo * hz) &&
940 VOP_GETATTR(dvp, &vattr, cnp->cn_cred, td) == 0 &&
941 timespeccmp(&vattr.va_mtime, &np->n_dmtime, ==)) {
942 nfsstats.lookupcache_hits++;
943 return (ENOENT);
944 }
945 cache_purge_negative(dvp);
946 mtx_lock(&np->n_mtx);
947 timespecclear(&np->n_dmtime);
948 mtx_unlock(&np->n_mtx);
949 }
950
951 /*
952 * Cache the modification time of the parent directory in case
953 * the lookup fails and results in adding the first negative
954 * name cache entry for the directory. Since this is reading
955 * a single time_t, don't bother with locking. The
956 * modification time may be a bit stale, but it must be read
957 * before performing the lookup RPC to prevent a race where
958 * another lookup updates the timestamp on the directory after
959 * the lookup RPC has been performed on the server but before
960 * n_dmtime is set at the end of this function.
961 */
962 dmtime = np->n_vattr.va_mtime;
963 error = 0;
964 newvp = NULLVP;
965 nfsstats.lookupcache_misses++;
966 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
967 len = cnp->cn_namelen;
968 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
969 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
970 mb = mreq;
971 bpos = mtod(mb, caddr_t);
972 nfsm_fhtom(dvp, v3);
973 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
974 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
975 if (error) {
976 if (v3) {
977 nfsm_postop_attr(dvp, attrflag);
978 m_freem(mrep);
979 }
980 goto nfsmout;
981 }
982 nfsm_getfh(fhp, fhsize, v3);
983
984 /*
985 * Handle RENAME case...
986 */
987 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
988 if (NFS_CMPFH(np, fhp, fhsize)) {
989 m_freem(mrep);
990 return (EISDIR);
991 }
992 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, LK_EXCLUSIVE);
993 if (error) {
994 m_freem(mrep);
995 return (error);
996 }
997 newvp = NFSTOV(np);
998 if (v3) {
999 nfsm_postop_attr(newvp, attrflag);
1000 nfsm_postop_attr(dvp, attrflag);
1001 } else
1002 nfsm_loadattr(newvp, NULL);
1003 *vpp = newvp;
1004 m_freem(mrep);
1005 cnp->cn_flags |= SAVENAME;
1006 return (0);
1007 }
1008
1009 if (flags & ISDOTDOT) {
1010 VOP_UNLOCK(dvp, 0, td);
1011 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, cnp->cn_lkflags);
1012 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
1013 if (error)
1014 return (error);
1015 newvp = NFSTOV(np);
1016 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1017 VREF(dvp);
1018 newvp = dvp;
1019 } else {
1020 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, cnp->cn_lkflags);
1021 if (error) {
1022 m_freem(mrep);
1023 return (error);
1024 }
1025 newvp = NFSTOV(np);
1026
1027 /*
1028 * Flush the attribute cache when opening a leaf node
1029 * to ensure that fresh attributes are fetched in
1030 * nfs_open() if we are unable to fetch attributes
1031 * from the LOOKUP reply.
1032 */
1033 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1034 !(np->n_flag & NMODIFIED)) {
1035 mtx_lock(&np->n_mtx);
1036 np->n_attrstamp = 0;
1037 mtx_unlock(&np->n_mtx);
1038 }
1039 }
1040 if (v3) {
1041 nfsm_postop_attr(newvp, attrflag);
1042 nfsm_postop_attr(dvp, attrflag);
1043 } else
1044 nfsm_loadattr(newvp, NULL);
1045 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1046 cnp->cn_flags |= SAVENAME;
1047 if ((cnp->cn_flags & MAKEENTRY) &&
1048 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1049 np->n_ctime = np->n_vattr.va_ctime;
1050 cache_enter(dvp, newvp, cnp);
1051 }
1052 *vpp = newvp;
1053 m_freem(mrep);
1054 nfsmout:
1055 if (error) {
1056 if (newvp != NULLVP) {
1057 vput(newvp);
1058 *vpp = NULLVP;
1059 }
1060
1061 if (error != ENOENT)
1062 goto done;
1063
1064 /* The requested file was not found. */
1065 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1066 (flags & ISLASTCN)) {
1067 /*
1068 * XXX: UFS does a full VOP_ACCESS(dvp,
1069 * VWRITE) here instead of just checking
1070 * MNT_RDONLY.
1071 */
1072 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1073 return (EROFS);
1074 cnp->cn_flags |= SAVENAME;
1075 return (EJUSTRETURN);
1076 }
1077
1078 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1079 /*
1080 * Maintain n_dmtime as the modification time
1081 * of the parent directory when the oldest -ve
1082 * name cache entry for this directory was
1083 * added. If a -ve cache entry has already
1084 * been added with a newer modification time
1085 * by a concurrent lookup, then don't bother
1086 * adding a cache entry. The modification
1087 * time of the directory might have changed
1088 * due to the file this lookup failed to find
1089 * being created. In that case a subsequent
1090 * lookup would incorrectly use the entry
1091 * added here instead of doing an extra
1092 * lookup.
1093 */
1094 mtx_lock(&np->n_mtx);
1095 if (timespeccmp(&np->n_dmtime, &dmtime, <=)) {
1096 if (!timespecisset(&np->n_dmtime)) {
1097 np->n_dmtime = dmtime;
1098 np->n_dmtime_ticks = ticks;
1099 }
1100 mtx_unlock(&np->n_mtx);
1101 cache_enter(dvp, NULL, cnp);
1102 } else
1103 mtx_unlock(&np->n_mtx);
1104 }
1105 return (ENOENT);
1106 }
1107 done:
1108 return (error);
1109 }
1110
1111 /*
1112 * nfs read call.
1113 * Just call nfs_bioread() to do the work.
1114 */
1115 static int
1116 nfs_read(struct vop_read_args *ap)
1117 {
1118 struct vnode *vp = ap->a_vp;
1119
1120 switch (vp->v_type) {
1121 case VREG:
1122 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1123 case VDIR:
1124 return (EISDIR);
1125 default:
1126 return (EOPNOTSUPP);
1127 }
1128 }
1129
1130 /*
1131 * nfs readlink call
1132 */
1133 static int
1134 nfs_readlink(struct vop_readlink_args *ap)
1135 {
1136 struct vnode *vp = ap->a_vp;
1137
1138 if (vp->v_type != VLNK)
1139 return (EINVAL);
1140 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1141 }
1142
1143 /*
1144 * Do a readlink rpc.
1145 * Called by nfs_doio() from below the buffer cache.
1146 */
1147 int
1148 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1149 {
1150 caddr_t bpos, dpos;
1151 int error = 0, len, attrflag;
1152 struct mbuf *mreq, *mrep, *md, *mb;
1153 int v3 = NFS_ISV3(vp);
1154
1155 nfsstats.rpccnt[NFSPROC_READLINK]++;
1156 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1157 mb = mreq;
1158 bpos = mtod(mb, caddr_t);
1159 nfsm_fhtom(vp, v3);
1160 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1161 if (v3)
1162 nfsm_postop_attr(vp, attrflag);
1163 if (!error) {
1164 nfsm_strsiz(len, NFS_MAXPATHLEN);
1165 if (len == NFS_MAXPATHLEN) {
1166 struct nfsnode *np = VTONFS(vp);
1167 mtx_lock(&np->n_mtx);
1168 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1169 len = np->n_size;
1170 mtx_unlock(&np->n_mtx);
1171 }
1172 nfsm_mtouio(uiop, len);
1173 }
1174 m_freem(mrep);
1175 nfsmout:
1176 return (error);
1177 }
1178
1179 /*
1180 * nfs read rpc call
1181 * Ditto above
1182 */
1183 int
1184 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1185 {
1186 u_int32_t *tl;
1187 caddr_t bpos, dpos;
1188 struct mbuf *mreq, *mrep, *md, *mb;
1189 struct nfsmount *nmp;
1190 int error = 0, len, retlen, tsiz, eof, attrflag;
1191 int v3 = NFS_ISV3(vp);
1192 int rsize;
1193
1194 #ifndef nolint
1195 eof = 0;
1196 #endif
1197 nmp = VFSTONFS(vp->v_mount);
1198 tsiz = uiop->uio_resid;
1199 mtx_lock(&nmp->nm_mtx);
1200 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1201 mtx_unlock(&nmp->nm_mtx);
1202 return (EFBIG);
1203 }
1204 rsize = nmp->nm_rsize;
1205 mtx_unlock(&nmp->nm_mtx);
1206 while (tsiz > 0) {
1207 nfsstats.rpccnt[NFSPROC_READ]++;
1208 len = (tsiz > rsize) ? rsize : tsiz;
1209 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1210 mb = mreq;
1211 bpos = mtod(mb, caddr_t);
1212 nfsm_fhtom(vp, v3);
1213 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1214 if (v3) {
1215 txdr_hyper(uiop->uio_offset, tl);
1216 *(tl + 2) = txdr_unsigned(len);
1217 } else {
1218 *tl++ = txdr_unsigned(uiop->uio_offset);
1219 *tl++ = txdr_unsigned(len);
1220 *tl = 0;
1221 }
1222 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1223 if (v3) {
1224 nfsm_postop_attr(vp, attrflag);
1225 if (error) {
1226 m_freem(mrep);
1227 goto nfsmout;
1228 }
1229 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1230 eof = fxdr_unsigned(int, *(tl + 1));
1231 } else {
1232 nfsm_loadattr(vp, NULL);
1233 }
1234 nfsm_strsiz(retlen, rsize);
1235 nfsm_mtouio(uiop, retlen);
1236 m_freem(mrep);
1237 tsiz -= retlen;
1238 if (v3) {
1239 if (eof || retlen == 0) {
1240 tsiz = 0;
1241 }
1242 } else if (retlen < len) {
1243 tsiz = 0;
1244 }
1245 }
1246 nfsmout:
1247 return (error);
1248 }
1249
1250 /*
1251 * nfs write call
1252 */
1253 int
1254 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1255 int *iomode, int *must_commit)
1256 {
1257 u_int32_t *tl;
1258 int32_t backup;
1259 caddr_t bpos, dpos;
1260 struct mbuf *mreq, *mrep, *md, *mb;
1261 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1262 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1263 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1264 int wsize;
1265
1266 #ifndef DIAGNOSTIC
1267 if (uiop->uio_iovcnt != 1)
1268 panic("nfs: writerpc iovcnt > 1");
1269 #endif
1270 *must_commit = 0;
1271 tsiz = uiop->uio_resid;
1272 mtx_lock(&nmp->nm_mtx);
1273 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1274 mtx_unlock(&nmp->nm_mtx);
1275 return (EFBIG);
1276 }
1277 wsize = nmp->nm_wsize;
1278 mtx_unlock(&nmp->nm_mtx);
1279 while (tsiz > 0) {
1280 nfsstats.rpccnt[NFSPROC_WRITE]++;
1281 len = (tsiz > wsize) ? wsize : tsiz;
1282 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1283 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1284 mb = mreq;
1285 bpos = mtod(mb, caddr_t);
1286 nfsm_fhtom(vp, v3);
1287 if (v3) {
1288 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1289 txdr_hyper(uiop->uio_offset, tl);
1290 tl += 2;
1291 *tl++ = txdr_unsigned(len);
1292 *tl++ = txdr_unsigned(*iomode);
1293 *tl = txdr_unsigned(len);
1294 } else {
1295 u_int32_t x;
1296
1297 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1298 /* Set both "begin" and "current" to non-garbage. */
1299 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1300 *tl++ = x; /* "begin offset" */
1301 *tl++ = x; /* "current offset" */
1302 x = txdr_unsigned(len);
1303 *tl++ = x; /* total to this offset */
1304 *tl = x; /* size of this write */
1305 }
1306 nfsm_uiotom(uiop, len);
1307 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1308 if (v3) {
1309 wccflag = NFSV3_WCCCHK;
1310 nfsm_wcc_data(vp, wccflag);
1311 if (!error) {
1312 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1313 + NFSX_V3WRITEVERF);
1314 rlen = fxdr_unsigned(int, *tl++);
1315 if (rlen == 0) {
1316 error = NFSERR_IO;
1317 m_freem(mrep);
1318 break;
1319 } else if (rlen < len) {
1320 backup = len - rlen;
1321 uiop->uio_iov->iov_base =
1322 (char *)uiop->uio_iov->iov_base -
1323 backup;
1324 uiop->uio_iov->iov_len += backup;
1325 uiop->uio_offset -= backup;
1326 uiop->uio_resid += backup;
1327 len = rlen;
1328 }
1329 commit = fxdr_unsigned(int, *tl++);
1330
1331 /*
1332 * Return the lowest committment level
1333 * obtained by any of the RPCs.
1334 */
1335 if (committed == NFSV3WRITE_FILESYNC)
1336 committed = commit;
1337 else if (committed == NFSV3WRITE_DATASYNC &&
1338 commit == NFSV3WRITE_UNSTABLE)
1339 committed = commit;
1340 mtx_lock(&nmp->nm_mtx);
1341 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1342 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1343 NFSX_V3WRITEVERF);
1344 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1345 } else if (bcmp((caddr_t)tl,
1346 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1347 *must_commit = 1;
1348 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1349 NFSX_V3WRITEVERF);
1350 }
1351 mtx_unlock(&nmp->nm_mtx);
1352 }
1353 } else {
1354 nfsm_loadattr(vp, NULL);
1355 }
1356 if (wccflag) {
1357 mtx_lock(&(VTONFS(vp))->n_mtx);
1358 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1359 mtx_unlock(&(VTONFS(vp))->n_mtx);
1360 }
1361 m_freem(mrep);
1362 if (error)
1363 break;
1364 tsiz -= len;
1365 }
1366 nfsmout:
1367 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1368 committed = NFSV3WRITE_FILESYNC;
1369 *iomode = committed;
1370 if (error)
1371 uiop->uio_resid = tsiz;
1372 return (error);
1373 }
1374
1375 /*
1376 * nfs mknod rpc
1377 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1378 * mode set to specify the file type and the size field for rdev.
1379 */
1380 static int
1381 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1382 struct vattr *vap)
1383 {
1384 struct nfsv2_sattr *sp;
1385 u_int32_t *tl;
1386 struct vnode *newvp = NULL;
1387 struct nfsnode *np = NULL;
1388 struct vattr vattr;
1389 caddr_t bpos, dpos;
1390 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1391 struct mbuf *mreq, *mrep, *md, *mb;
1392 u_int32_t rdev;
1393 int v3 = NFS_ISV3(dvp);
1394
1395 if (vap->va_type == VCHR || vap->va_type == VBLK)
1396 rdev = txdr_unsigned(vap->va_rdev);
1397 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1398 rdev = nfs_xdrneg1;
1399 else {
1400 return (EOPNOTSUPP);
1401 }
1402 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1403 return (error);
1404 }
1405 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1406 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1407 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1408 mb = mreq;
1409 bpos = mtod(mb, caddr_t);
1410 nfsm_fhtom(dvp, v3);
1411 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1412 if (v3) {
1413 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1414 *tl++ = vtonfsv3_type(vap->va_type);
1415 nfsm_v3attrbuild(vap, FALSE);
1416 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1417 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1418 *tl++ = txdr_unsigned(umajor(vap->va_rdev));
1419 *tl = txdr_unsigned(uminor(vap->va_rdev));
1420 }
1421 } else {
1422 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1423 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1424 sp->sa_uid = nfs_xdrneg1;
1425 sp->sa_gid = nfs_xdrneg1;
1426 sp->sa_size = rdev;
1427 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1428 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1429 }
1430 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1431 if (!error) {
1432 nfsm_mtofh(dvp, newvp, v3, gotvp);
1433 if (!gotvp) {
1434 if (newvp) {
1435 vput(newvp);
1436 newvp = NULL;
1437 }
1438 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1439 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1440 if (!error)
1441 newvp = NFSTOV(np);
1442 }
1443 }
1444 if (v3)
1445 nfsm_wcc_data(dvp, wccflag);
1446 m_freem(mrep);
1447 nfsmout:
1448 if (error) {
1449 if (newvp)
1450 vput(newvp);
1451 } else {
1452 if (cnp->cn_flags & MAKEENTRY)
1453 cache_enter(dvp, newvp, cnp);
1454 *vpp = newvp;
1455 }
1456 mtx_lock(&(VTONFS(dvp))->n_mtx);
1457 VTONFS(dvp)->n_flag |= NMODIFIED;
1458 if (!wccflag)
1459 VTONFS(dvp)->n_attrstamp = 0;
1460 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1461 return (error);
1462 }
1463
1464 /*
1465 * nfs mknod vop
1466 * just call nfs_mknodrpc() to do the work.
1467 */
1468 /* ARGSUSED */
1469 static int
1470 nfs_mknod(struct vop_mknod_args *ap)
1471 {
1472 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1473 }
1474
1475 static u_long create_verf;
1476 /*
1477 * nfs file create call
1478 */
1479 static int
1480 nfs_create(struct vop_create_args *ap)
1481 {
1482 struct vnode *dvp = ap->a_dvp;
1483 struct vattr *vap = ap->a_vap;
1484 struct componentname *cnp = ap->a_cnp;
1485 struct nfsv2_sattr *sp;
1486 u_int32_t *tl;
1487 struct nfsnode *np = NULL;
1488 struct vnode *newvp = NULL;
1489 caddr_t bpos, dpos;
1490 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1491 struct mbuf *mreq, *mrep, *md, *mb;
1492 struct vattr vattr;
1493 int v3 = NFS_ISV3(dvp);
1494
1495 /*
1496 * Oops, not for me..
1497 */
1498 if (vap->va_type == VSOCK)
1499 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1500
1501 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1502 return (error);
1503 }
1504 if (vap->va_vaflags & VA_EXCLUSIVE)
1505 fmode |= O_EXCL;
1506 again:
1507 nfsstats.rpccnt[NFSPROC_CREATE]++;
1508 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1509 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1510 mb = mreq;
1511 bpos = mtod(mb, caddr_t);
1512 nfsm_fhtom(dvp, v3);
1513 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1514 if (v3) {
1515 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1516 if (fmode & O_EXCL) {
1517 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1518 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1519 #ifdef INET
1520 if (!TAILQ_EMPTY(&in_ifaddrhead))
1521 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1522 else
1523 #endif
1524 *tl++ = create_verf;
1525 *tl = ++create_verf;
1526 } else {
1527 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1528 nfsm_v3attrbuild(vap, FALSE);
1529 }
1530 } else {
1531 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1532 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1533 sp->sa_uid = nfs_xdrneg1;
1534 sp->sa_gid = nfs_xdrneg1;
1535 sp->sa_size = 0;
1536 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1537 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1538 }
1539 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1540 if (!error) {
1541 nfsm_mtofh(dvp, newvp, v3, gotvp);
1542 if (!gotvp) {
1543 if (newvp) {
1544 vput(newvp);
1545 newvp = NULL;
1546 }
1547 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1548 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1549 if (!error)
1550 newvp = NFSTOV(np);
1551 }
1552 }
1553 if (v3)
1554 nfsm_wcc_data(dvp, wccflag);
1555 m_freem(mrep);
1556 nfsmout:
1557 if (error) {
1558 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1559 fmode &= ~O_EXCL;
1560 goto again;
1561 }
1562 if (newvp)
1563 vput(newvp);
1564 } else if (v3 && (fmode & O_EXCL)) {
1565 /*
1566 * We are normally called with only a partially initialized
1567 * VAP. Since the NFSv3 spec says that server may use the
1568 * file attributes to store the verifier, the spec requires
1569 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1570 * in atime, but we can't really assume that all servers will
1571 * so we ensure that our SETATTR sets both atime and mtime.
1572 */
1573 if (vap->va_mtime.tv_sec == VNOVAL)
1574 vfs_timestamp(&vap->va_mtime);
1575 if (vap->va_atime.tv_sec == VNOVAL)
1576 vap->va_atime = vap->va_mtime;
1577 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
1578 if (error)
1579 vput(newvp);
1580 }
1581 if (!error) {
1582 if (cnp->cn_flags & MAKEENTRY)
1583 cache_enter(dvp, newvp, cnp);
1584 *ap->a_vpp = newvp;
1585 }
1586 mtx_lock(&(VTONFS(dvp))->n_mtx);
1587 VTONFS(dvp)->n_flag |= NMODIFIED;
1588 if (!wccflag)
1589 VTONFS(dvp)->n_attrstamp = 0;
1590 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1591 return (error);
1592 }
1593
1594 /*
1595 * nfs file remove call
1596 * To try and make nfs semantics closer to ufs semantics, a file that has
1597 * other processes using the vnode is renamed instead of removed and then
1598 * removed later on the last close.
1599 * - If v_usecount > 1
1600 * If a rename is not already in the works
1601 * call nfs_sillyrename() to set it up
1602 * else
1603 * do the remove rpc
1604 */
1605 static int
1606 nfs_remove(struct vop_remove_args *ap)
1607 {
1608 struct vnode *vp = ap->a_vp;
1609 struct vnode *dvp = ap->a_dvp;
1610 struct componentname *cnp = ap->a_cnp;
1611 struct nfsnode *np = VTONFS(vp);
1612 int error = 0;
1613 struct vattr vattr;
1614
1615 #ifndef DIAGNOSTIC
1616 if ((cnp->cn_flags & HASBUF) == 0)
1617 panic("nfs_remove: no name");
1618 if (vrefcnt(vp) < 1)
1619 panic("nfs_remove: bad v_usecount");
1620 #endif
1621 if (vp->v_type == VDIR)
1622 error = EPERM;
1623 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1624 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
1625 vattr.va_nlink > 1)) {
1626 /*
1627 * Purge the name cache so that the chance of a lookup for
1628 * the name succeeding while the remove is in progress is
1629 * minimized. Without node locking it can still happen, such
1630 * that an I/O op returns ESTALE, but since you get this if
1631 * another host removes the file..
1632 */
1633 cache_purge(vp);
1634 /*
1635 * throw away biocache buffers, mainly to avoid
1636 * unnecessary delayed writes later.
1637 */
1638 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1639 /* Do the rpc */
1640 if (error != EINTR && error != EIO)
1641 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1642 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1643 /*
1644 * Kludge City: If the first reply to the remove rpc is lost..
1645 * the reply to the retransmitted request will be ENOENT
1646 * since the file was in fact removed
1647 * Therefore, we cheat and return success.
1648 */
1649 if (error == ENOENT)
1650 error = 0;
1651 } else if (!np->n_sillyrename)
1652 error = nfs_sillyrename(dvp, vp, cnp);
1653 np->n_attrstamp = 0;
1654 return (error);
1655 }
1656
1657 /*
1658 * nfs file remove rpc called from nfs_inactive
1659 */
1660 int
1661 nfs_removeit(struct sillyrename *sp)
1662 {
1663 /*
1664 * Make sure that the directory vnode is still valid.
1665 * XXX we should lock sp->s_dvp here.
1666 */
1667 if (sp->s_dvp->v_type == VBAD)
1668 return (0);
1669 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1670 NULL));
1671 }
1672
1673 /*
1674 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1675 */
1676 static int
1677 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1678 struct ucred *cred, struct thread *td)
1679 {
1680 caddr_t bpos, dpos;
1681 int error = 0, wccflag = NFSV3_WCCRATTR;
1682 struct mbuf *mreq, *mrep, *md, *mb;
1683 int v3 = NFS_ISV3(dvp);
1684
1685 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1686 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1687 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1688 mb = mreq;
1689 bpos = mtod(mb, caddr_t);
1690 nfsm_fhtom(dvp, v3);
1691 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1692 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1693 if (v3)
1694 nfsm_wcc_data(dvp, wccflag);
1695 m_freem(mrep);
1696 nfsmout:
1697 mtx_lock(&(VTONFS(dvp))->n_mtx);
1698 VTONFS(dvp)->n_flag |= NMODIFIED;
1699 if (!wccflag)
1700 VTONFS(dvp)->n_attrstamp = 0;
1701 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1702 return (error);
1703 }
1704
1705 /*
1706 * nfs file rename call
1707 */
1708 static int
1709 nfs_rename(struct vop_rename_args *ap)
1710 {
1711 struct vnode *fvp = ap->a_fvp;
1712 struct vnode *tvp = ap->a_tvp;
1713 struct vnode *fdvp = ap->a_fdvp;
1714 struct vnode *tdvp = ap->a_tdvp;
1715 struct componentname *tcnp = ap->a_tcnp;
1716 struct componentname *fcnp = ap->a_fcnp;
1717 int error;
1718
1719 #ifndef DIAGNOSTIC
1720 if ((tcnp->cn_flags & HASBUF) == 0 ||
1721 (fcnp->cn_flags & HASBUF) == 0)
1722 panic("nfs_rename: no name");
1723 #endif
1724 /* Check for cross-device rename */
1725 if ((fvp->v_mount != tdvp->v_mount) ||
1726 (tvp && (fvp->v_mount != tvp->v_mount))) {
1727 error = EXDEV;
1728 goto out;
1729 }
1730
1731 if (fvp == tvp) {
1732 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1733 error = 0;
1734 goto out;
1735 }
1736 if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0)
1737 goto out;
1738
1739 /*
1740 * We have to flush B_DELWRI data prior to renaming
1741 * the file. If we don't, the delayed-write buffers
1742 * can be flushed out later after the file has gone stale
1743 * under NFSV3. NFSV2 does not have this problem because
1744 * ( as far as I can tell ) it flushes dirty buffers more
1745 * often.
1746 *
1747 * Skip the rename operation if the fsync fails, this can happen
1748 * due to the server's volume being full, when we pushed out data
1749 * that was written back to our cache earlier. Not checking for
1750 * this condition can result in potential (silent) data loss.
1751 */
1752 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1753 VOP_UNLOCK(fvp, 0, fcnp->cn_thread);
1754 if (!error && tvp)
1755 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1756 if (error)
1757 goto out;
1758
1759 /*
1760 * If the tvp exists and is in use, sillyrename it before doing the
1761 * rename of the new file over it.
1762 * XXX Can't sillyrename a directory.
1763 */
1764 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1765 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1766 vput(tvp);
1767 tvp = NULL;
1768 }
1769
1770 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1771 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1772 tcnp->cn_thread);
1773
1774 if (fvp->v_type == VDIR) {
1775 if (tvp != NULL && tvp->v_type == VDIR)
1776 cache_purge(tdvp);
1777 cache_purge(fdvp);
1778 }
1779
1780 out:
1781 if (tdvp == tvp)
1782 vrele(tdvp);
1783 else
1784 vput(tdvp);
1785 if (tvp)
1786 vput(tvp);
1787 vrele(fdvp);
1788 vrele(fvp);
1789 /*
1790 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1791 */
1792 if (error == ENOENT)
1793 error = 0;
1794 return (error);
1795 }
1796
1797 /*
1798 * nfs file rename rpc called from nfs_remove() above
1799 */
1800 static int
1801 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1802 struct sillyrename *sp)
1803 {
1804
1805 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1806 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1807 }
1808
1809 /*
1810 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1811 */
1812 static int
1813 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1814 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1815 struct thread *td)
1816 {
1817 caddr_t bpos, dpos;
1818 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1819 struct mbuf *mreq, *mrep, *md, *mb;
1820 int v3 = NFS_ISV3(fdvp);
1821
1822 nfsstats.rpccnt[NFSPROC_RENAME]++;
1823 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1824 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1825 nfsm_rndup(tnamelen));
1826 mb = mreq;
1827 bpos = mtod(mb, caddr_t);
1828 nfsm_fhtom(fdvp, v3);
1829 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1830 nfsm_fhtom(tdvp, v3);
1831 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1832 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1833 if (v3) {
1834 nfsm_wcc_data(fdvp, fwccflag);
1835 nfsm_wcc_data(tdvp, twccflag);
1836 }
1837 m_freem(mrep);
1838 nfsmout:
1839 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1840 VTONFS(fdvp)->n_flag |= NMODIFIED;
1841 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1842 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1843 VTONFS(tdvp)->n_flag |= NMODIFIED;
1844 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1845 if (!fwccflag)
1846 VTONFS(fdvp)->n_attrstamp = 0;
1847 if (!twccflag)
1848 VTONFS(tdvp)->n_attrstamp = 0;
1849 return (error);
1850 }
1851
1852 /*
1853 * nfs hard link create call
1854 */
1855 static int
1856 nfs_link(struct vop_link_args *ap)
1857 {
1858 struct vnode *vp = ap->a_vp;
1859 struct vnode *tdvp = ap->a_tdvp;
1860 struct componentname *cnp = ap->a_cnp;
1861 caddr_t bpos, dpos;
1862 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1863 struct mbuf *mreq, *mrep, *md, *mb;
1864 int v3;
1865
1866 if (vp->v_mount != tdvp->v_mount) {
1867 return (EXDEV);
1868 }
1869
1870 /*
1871 * Push all writes to the server, so that the attribute cache
1872 * doesn't get "out of sync" with the server.
1873 * XXX There should be a better way!
1874 */
1875 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1876
1877 v3 = NFS_ISV3(vp);
1878 nfsstats.rpccnt[NFSPROC_LINK]++;
1879 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1880 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1881 mb = mreq;
1882 bpos = mtod(mb, caddr_t);
1883 nfsm_fhtom(vp, v3);
1884 nfsm_fhtom(tdvp, v3);
1885 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1886 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1887 if (v3) {
1888 nfsm_postop_attr(vp, attrflag);
1889 nfsm_wcc_data(tdvp, wccflag);
1890 }
1891 m_freem(mrep);
1892 nfsmout:
1893 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1894 VTONFS(tdvp)->n_flag |= NMODIFIED;
1895 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1896 if (!attrflag)
1897 VTONFS(vp)->n_attrstamp = 0;
1898 if (!wccflag)
1899 VTONFS(tdvp)->n_attrstamp = 0;
1900 return (error);
1901 }
1902
1903 /*
1904 * nfs symbolic link create call
1905 */
1906 static int
1907 nfs_symlink(struct vop_symlink_args *ap)
1908 {
1909 struct vnode *dvp = ap->a_dvp;
1910 struct vattr *vap = ap->a_vap;
1911 struct componentname *cnp = ap->a_cnp;
1912 struct nfsv2_sattr *sp;
1913 caddr_t bpos, dpos;
1914 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1915 struct mbuf *mreq, *mrep, *md, *mb;
1916 struct vnode *newvp = NULL;
1917 int v3 = NFS_ISV3(dvp);
1918
1919 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1920 slen = strlen(ap->a_target);
1921 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1922 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1923 mb = mreq;
1924 bpos = mtod(mb, caddr_t);
1925 nfsm_fhtom(dvp, v3);
1926 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1927 if (v3) {
1928 nfsm_v3attrbuild(vap, FALSE);
1929 }
1930 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1931 if (!v3) {
1932 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1933 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1934 sp->sa_uid = nfs_xdrneg1;
1935 sp->sa_gid = nfs_xdrneg1;
1936 sp->sa_size = nfs_xdrneg1;
1937 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1938 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1939 }
1940
1941 /*
1942 * Issue the NFS request and get the rpc response.
1943 *
1944 * Only NFSv3 responses returning an error of 0 actually return
1945 * a file handle that can be converted into newvp without having
1946 * to do an extra lookup rpc.
1947 */
1948 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
1949 if (v3) {
1950 if (error == 0)
1951 nfsm_mtofh(dvp, newvp, v3, gotvp);
1952 nfsm_wcc_data(dvp, wccflag);
1953 }
1954
1955 /*
1956 * out code jumps -> here, mrep is also freed.
1957 */
1958
1959 m_freem(mrep);
1960 nfsmout:
1961
1962 /*
1963 * If we do not have an error and we could not extract the newvp from
1964 * the response due to the request being NFSv2, we have to do a
1965 * lookup in order to obtain a newvp to return.
1966 */
1967 if (error == 0 && newvp == NULL) {
1968 struct nfsnode *np = NULL;
1969
1970 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1971 cnp->cn_cred, cnp->cn_thread, &np);
1972 if (!error)
1973 newvp = NFSTOV(np);
1974 }
1975 if (error) {
1976 if (newvp)
1977 vput(newvp);
1978 } else {
1979 *ap->a_vpp = newvp;
1980 }
1981 mtx_lock(&(VTONFS(dvp))->n_mtx);
1982 VTONFS(dvp)->n_flag |= NMODIFIED;
1983 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1984 if (!wccflag)
1985 VTONFS(dvp)->n_attrstamp = 0;
1986 return (error);
1987 }
1988
1989 /*
1990 * nfs make dir call
1991 */
1992 static int
1993 nfs_mkdir(struct vop_mkdir_args *ap)
1994 {
1995 struct vnode *dvp = ap->a_dvp;
1996 struct vattr *vap = ap->a_vap;
1997 struct componentname *cnp = ap->a_cnp;
1998 struct nfsv2_sattr *sp;
1999 int len;
2000 struct nfsnode *np = NULL;
2001 struct vnode *newvp = NULL;
2002 caddr_t bpos, dpos;
2003 int error = 0, wccflag = NFSV3_WCCRATTR;
2004 int gotvp = 0;
2005 struct mbuf *mreq, *mrep, *md, *mb;
2006 struct vattr vattr;
2007 int v3 = NFS_ISV3(dvp);
2008
2009 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
2010 return (error);
2011 }
2012 len = cnp->cn_namelen;
2013 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2014 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
2015 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2016 mb = mreq;
2017 bpos = mtod(mb, caddr_t);
2018 nfsm_fhtom(dvp, v3);
2019 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2020 if (v3) {
2021 nfsm_v3attrbuild(vap, FALSE);
2022 } else {
2023 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2024 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2025 sp->sa_uid = nfs_xdrneg1;
2026 sp->sa_gid = nfs_xdrneg1;
2027 sp->sa_size = nfs_xdrneg1;
2028 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2029 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2030 }
2031 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2032 if (!error)
2033 nfsm_mtofh(dvp, newvp, v3, gotvp);
2034 if (v3)
2035 nfsm_wcc_data(dvp, wccflag);
2036 m_freem(mrep);
2037 nfsmout:
2038 mtx_lock(&(VTONFS(dvp))->n_mtx);
2039 VTONFS(dvp)->n_flag |= NMODIFIED;
2040 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2041 if (!wccflag)
2042 VTONFS(dvp)->n_attrstamp = 0;
2043 if (error == 0 && newvp == NULL) {
2044 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2045 cnp->cn_thread, &np);
2046 if (!error) {
2047 newvp = NFSTOV(np);
2048 if (newvp->v_type != VDIR)
2049 error = EEXIST;
2050 }
2051 }
2052 if (error) {
2053 if (newvp)
2054 vput(newvp);
2055 } else
2056 *ap->a_vpp = newvp;
2057 return (error);
2058 }
2059
2060 /*
2061 * nfs remove directory call
2062 */
2063 static int
2064 nfs_rmdir(struct vop_rmdir_args *ap)
2065 {
2066 struct vnode *vp = ap->a_vp;
2067 struct vnode *dvp = ap->a_dvp;
2068 struct componentname *cnp = ap->a_cnp;
2069 caddr_t bpos, dpos;
2070 int error = 0, wccflag = NFSV3_WCCRATTR;
2071 struct mbuf *mreq, *mrep, *md, *mb;
2072 int v3 = NFS_ISV3(dvp);
2073
2074 if (dvp == vp)
2075 return (EINVAL);
2076 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2077 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
2078 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2079 mb = mreq;
2080 bpos = mtod(mb, caddr_t);
2081 nfsm_fhtom(dvp, v3);
2082 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2083 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2084 if (v3)
2085 nfsm_wcc_data(dvp, wccflag);
2086 m_freem(mrep);
2087 nfsmout:
2088 mtx_lock(&(VTONFS(dvp))->n_mtx);
2089 VTONFS(dvp)->n_flag |= NMODIFIED;
2090 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2091 if (!wccflag)
2092 VTONFS(dvp)->n_attrstamp = 0;
2093 cache_purge(dvp);
2094 cache_purge(vp);
2095 /*
2096 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2097 */
2098 if (error == ENOENT)
2099 error = 0;
2100 return (error);
2101 }
2102
2103 /*
2104 * nfs readdir call
2105 */
2106 static int
2107 nfs_readdir(struct vop_readdir_args *ap)
2108 {
2109 struct vnode *vp = ap->a_vp;
2110 struct nfsnode *np = VTONFS(vp);
2111 struct uio *uio = ap->a_uio;
2112 int tresid, error = 0;
2113 struct vattr vattr;
2114
2115 if (vp->v_type != VDIR)
2116 return(EPERM);
2117
2118 /*
2119 * First, check for hit on the EOF offset cache
2120 */
2121 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2122 (np->n_flag & NMODIFIED) == 0) {
2123 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0) {
2124 mtx_lock(&np->n_mtx);
2125 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2126 mtx_unlock(&np->n_mtx);
2127 nfsstats.direofcache_hits++;
2128 goto out;
2129 } else
2130 mtx_unlock(&np->n_mtx);
2131 }
2132 }
2133
2134 /*
2135 * Call nfs_bioread() to do the real work.
2136 */
2137 tresid = uio->uio_resid;
2138 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2139
2140 if (!error && uio->uio_resid == tresid) {
2141 nfsstats.direofcache_misses++;
2142 }
2143 out:
2144 return (error);
2145 }
2146
2147 /*
2148 * Readdir rpc call.
2149 * Called from below the buffer cache by nfs_doio().
2150 */
2151 int
2152 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2153 {
2154 int len, left;
2155 struct dirent *dp = NULL;
2156 u_int32_t *tl;
2157 caddr_t cp;
2158 nfsuint64 *cookiep;
2159 caddr_t bpos, dpos;
2160 struct mbuf *mreq, *mrep, *md, *mb;
2161 nfsuint64 cookie;
2162 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2163 struct nfsnode *dnp = VTONFS(vp);
2164 u_quad_t fileno;
2165 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2166 int attrflag;
2167 int v3 = NFS_ISV3(vp);
2168
2169 #ifndef DIAGNOSTIC
2170 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2171 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2172 panic("nfs readdirrpc bad uio");
2173 #endif
2174
2175 /*
2176 * If there is no cookie, assume directory was stale.
2177 */
2178 nfs_dircookie_lock(dnp);
2179 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2180 if (cookiep) {
2181 cookie = *cookiep;
2182 nfs_dircookie_unlock(dnp);
2183 } else {
2184 nfs_dircookie_unlock(dnp);
2185 return (NFSERR_BAD_COOKIE);
2186 }
2187
2188 /*
2189 * Loop around doing readdir rpc's of size nm_readdirsize
2190 * truncated to a multiple of DIRBLKSIZ.
2191 * The stopping criteria is EOF or buffer full.
2192 */
2193 while (more_dirs && bigenough) {
2194 nfsstats.rpccnt[NFSPROC_READDIR]++;
2195 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2196 NFSX_READDIR(v3));
2197 mb = mreq;
2198 bpos = mtod(mb, caddr_t);
2199 nfsm_fhtom(vp, v3);
2200 if (v3) {
2201 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2202 *tl++ = cookie.nfsuquad[0];
2203 *tl++ = cookie.nfsuquad[1];
2204 mtx_lock(&dnp->n_mtx);
2205 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2206 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2207 mtx_unlock(&dnp->n_mtx);
2208 } else {
2209 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2210 *tl++ = cookie.nfsuquad[0];
2211 }
2212 *tl = txdr_unsigned(nmp->nm_readdirsize);
2213 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2214 if (v3) {
2215 nfsm_postop_attr(vp, attrflag);
2216 if (!error) {
2217 tl = nfsm_dissect(u_int32_t *,
2218 2 * NFSX_UNSIGNED);
2219 mtx_lock(&dnp->n_mtx);
2220 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2221 dnp->n_cookieverf.nfsuquad[1] = *tl;
2222 mtx_unlock(&dnp->n_mtx);
2223 } else {
2224 m_freem(mrep);
2225 goto nfsmout;
2226 }
2227 }
2228 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2229 more_dirs = fxdr_unsigned(int, *tl);
2230
2231 /* loop thru the dir entries, doctoring them to 4bsd form */
2232 while (more_dirs && bigenough) {
2233 if (v3) {
2234 tl = nfsm_dissect(u_int32_t *,
2235 3 * NFSX_UNSIGNED);
2236 fileno = fxdr_hyper(tl);
2237 len = fxdr_unsigned(int, *(tl + 2));
2238 } else {
2239 tl = nfsm_dissect(u_int32_t *,
2240 2 * NFSX_UNSIGNED);
2241 fileno = fxdr_unsigned(u_quad_t, *tl++);
2242 len = fxdr_unsigned(int, *tl);
2243 }
2244 if (len <= 0 || len > NFS_MAXNAMLEN) {
2245 error = EBADRPC;
2246 m_freem(mrep);
2247 goto nfsmout;
2248 }
2249 tlen = nfsm_rndup(len);
2250 if (tlen == len)
2251 tlen += 4; /* To ensure null termination */
2252 left = DIRBLKSIZ - blksiz;
2253 if ((tlen + DIRHDSIZ) > left) {
2254 dp->d_reclen += left;
2255 uiop->uio_iov->iov_base =
2256 (char *)uiop->uio_iov->iov_base + left;
2257 uiop->uio_iov->iov_len -= left;
2258 uiop->uio_offset += left;
2259 uiop->uio_resid -= left;
2260 blksiz = 0;
2261 }
2262 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2263 bigenough = 0;
2264 if (bigenough) {
2265 dp = (struct dirent *)uiop->uio_iov->iov_base;
2266 dp->d_fileno = (int)fileno;
2267 dp->d_namlen = len;
2268 dp->d_reclen = tlen + DIRHDSIZ;
2269 dp->d_type = DT_UNKNOWN;
2270 blksiz += dp->d_reclen;
2271 if (blksiz == DIRBLKSIZ)
2272 blksiz = 0;
2273 uiop->uio_offset += DIRHDSIZ;
2274 uiop->uio_resid -= DIRHDSIZ;
2275 uiop->uio_iov->iov_base =
2276 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2277 uiop->uio_iov->iov_len -= DIRHDSIZ;
2278 nfsm_mtouio(uiop, len);
2279 cp = uiop->uio_iov->iov_base;
2280 tlen -= len;
2281 *cp = '\0'; /* null terminate */
2282 uiop->uio_iov->iov_base =
2283 (char *)uiop->uio_iov->iov_base + tlen;
2284 uiop->uio_iov->iov_len -= tlen;
2285 uiop->uio_offset += tlen;
2286 uiop->uio_resid -= tlen;
2287 } else
2288 nfsm_adv(nfsm_rndup(len));
2289 if (v3) {
2290 tl = nfsm_dissect(u_int32_t *,
2291 3 * NFSX_UNSIGNED);
2292 } else {
2293 tl = nfsm_dissect(u_int32_t *,
2294 2 * NFSX_UNSIGNED);
2295 }
2296 if (bigenough) {
2297 cookie.nfsuquad[0] = *tl++;
2298 if (v3)
2299 cookie.nfsuquad[1] = *tl++;
2300 } else if (v3)
2301 tl += 2;
2302 else
2303 tl++;
2304 more_dirs = fxdr_unsigned(int, *tl);
2305 }
2306 /*
2307 * If at end of rpc data, get the eof boolean
2308 */
2309 if (!more_dirs) {
2310 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2311 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2312 }
2313 m_freem(mrep);
2314 }
2315 /*
2316 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2317 * by increasing d_reclen for the last record.
2318 */
2319 if (blksiz > 0) {
2320 left = DIRBLKSIZ - blksiz;
2321 dp->d_reclen += left;
2322 uiop->uio_iov->iov_base =
2323 (char *)uiop->uio_iov->iov_base + left;
2324 uiop->uio_iov->iov_len -= left;
2325 uiop->uio_offset += left;
2326 uiop->uio_resid -= left;
2327 }
2328
2329 /*
2330 * We are now either at the end of the directory or have filled the
2331 * block.
2332 */
2333 if (bigenough)
2334 dnp->n_direofoffset = uiop->uio_offset;
2335 else {
2336 if (uiop->uio_resid > 0)
2337 nfs_printf("EEK! readdirrpc resid > 0\n");
2338 nfs_dircookie_lock(dnp);
2339 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2340 *cookiep = cookie;
2341 nfs_dircookie_unlock(dnp);
2342 }
2343 nfsmout:
2344 return (error);
2345 }
2346
2347 /*
2348 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2349 */
2350 int
2351 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2352 {
2353 int len, left;
2354 struct dirent *dp;
2355 u_int32_t *tl;
2356 caddr_t cp;
2357 struct vnode *newvp;
2358 nfsuint64 *cookiep;
2359 caddr_t bpos, dpos, dpossav1, dpossav2;
2360 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2361 struct nameidata nami, *ndp = &nami;
2362 struct componentname *cnp = &ndp->ni_cnd;
2363 nfsuint64 cookie;
2364 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2365 struct nfsnode *dnp = VTONFS(vp), *np;
2366 nfsfh_t *fhp;
2367 u_quad_t fileno;
2368 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2369 int attrflag, fhsize;
2370
2371 #ifndef nolint
2372 dp = NULL;
2373 #endif
2374 #ifndef DIAGNOSTIC
2375 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2376 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2377 panic("nfs readdirplusrpc bad uio");
2378 #endif
2379 ndp->ni_dvp = vp;
2380 newvp = NULLVP;
2381
2382 /*
2383 * If there is no cookie, assume directory was stale.
2384 */
2385 nfs_dircookie_lock(dnp);
2386 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2387 if (cookiep) {
2388 cookie = *cookiep;
2389 nfs_dircookie_unlock(dnp);
2390 } else {
2391 nfs_dircookie_unlock(dnp);
2392 return (NFSERR_BAD_COOKIE);
2393 }
2394 /*
2395 * Loop around doing readdir rpc's of size nm_readdirsize
2396 * truncated to a multiple of DIRBLKSIZ.
2397 * The stopping criteria is EOF or buffer full.
2398 */
2399 while (more_dirs && bigenough) {
2400 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2401 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2402 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2403 mb = mreq;
2404 bpos = mtod(mb, caddr_t);
2405 nfsm_fhtom(vp, 1);
2406 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2407 *tl++ = cookie.nfsuquad[0];
2408 *tl++ = cookie.nfsuquad[1];
2409 mtx_lock(&dnp->n_mtx);
2410 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2411 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2412 mtx_unlock(&dnp->n_mtx);
2413 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2414 *tl = txdr_unsigned(nmp->nm_rsize);
2415 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2416 nfsm_postop_attr(vp, attrflag);
2417 if (error) {
2418 m_freem(mrep);
2419 goto nfsmout;
2420 }
2421 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2422 mtx_lock(&dnp->n_mtx);
2423 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2424 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2425 mtx_unlock(&dnp->n_mtx);
2426 more_dirs = fxdr_unsigned(int, *tl);
2427
2428 /* loop thru the dir entries, doctoring them to 4bsd form */
2429 while (more_dirs && bigenough) {
2430 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2431 fileno = fxdr_hyper(tl);
2432 len = fxdr_unsigned(int, *(tl + 2));
2433 if (len <= 0 || len > NFS_MAXNAMLEN) {
2434 error = EBADRPC;
2435 m_freem(mrep);
2436 goto nfsmout;
2437 }
2438 tlen = nfsm_rndup(len);
2439 if (tlen == len)
2440 tlen += 4; /* To ensure null termination*/
2441 left = DIRBLKSIZ - blksiz;
2442 if ((tlen + DIRHDSIZ) > left) {
2443 dp->d_reclen += left;
2444 uiop->uio_iov->iov_base =
2445 (char *)uiop->uio_iov->iov_base + left;
2446 uiop->uio_iov->iov_len -= left;
2447 uiop->uio_offset += left;
2448 uiop->uio_resid -= left;
2449 blksiz = 0;
2450 }
2451 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2452 bigenough = 0;
2453 if (bigenough) {
2454 dp = (struct dirent *)uiop->uio_iov->iov_base;
2455 dp->d_fileno = (int)fileno;
2456 dp->d_namlen = len;
2457 dp->d_reclen = tlen + DIRHDSIZ;
2458 dp->d_type = DT_UNKNOWN;
2459 blksiz += dp->d_reclen;
2460 if (blksiz == DIRBLKSIZ)
2461 blksiz = 0;
2462 uiop->uio_offset += DIRHDSIZ;
2463 uiop->uio_resid -= DIRHDSIZ;
2464 uiop->uio_iov->iov_base =
2465 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2466 uiop->uio_iov->iov_len -= DIRHDSIZ;
2467 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2468 cnp->cn_namelen = len;
2469 nfsm_mtouio(uiop, len);
2470 cp = uiop->uio_iov->iov_base;
2471 tlen -= len;
2472 *cp = '\0';
2473 uiop->uio_iov->iov_base =
2474 (char *)uiop->uio_iov->iov_base + tlen;
2475 uiop->uio_iov->iov_len -= tlen;
2476 uiop->uio_offset += tlen;
2477 uiop->uio_resid -= tlen;
2478 } else
2479 nfsm_adv(nfsm_rndup(len));
2480 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2481 if (bigenough) {
2482 cookie.nfsuquad[0] = *tl++;
2483 cookie.nfsuquad[1] = *tl++;
2484 } else
2485 tl += 2;
2486
2487 /*
2488 * Since the attributes are before the file handle
2489 * (sigh), we must skip over the attributes and then
2490 * come back and get them.
2491 */
2492 attrflag = fxdr_unsigned(int, *tl);
2493 if (attrflag) {
2494 dpossav1 = dpos;
2495 mdsav1 = md;
2496 nfsm_adv(NFSX_V3FATTR);
2497 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2498 doit = fxdr_unsigned(int, *tl);
2499 /*
2500 * Skip loading the attrs for "..". There's a
2501 * race between loading the attrs here and
2502 * lookups that look for the directory currently
2503 * being read (in the parent). We try to acquire
2504 * the exclusive lock on ".." here, owning the
2505 * lock on the directory being read. Lookup will
2506 * hold the lock on ".." and try to acquire the
2507 * lock on the directory being read.
2508 *
2509 * There are other ways of fixing this, one would
2510 * be to do a trylock on the ".." vnode and skip
2511 * loading the attrs on ".." if it happens to be
2512 * locked by another process. But skipping the
2513 * attrload on ".." seems the easiest option.
2514 */
2515 if (strcmp(dp->d_name, "..") == 0) {
2516 doit = 0;
2517 /*
2518 * We've already skipped over the attrs,
2519 * skip over the filehandle. And store d_type
2520 * as VDIR.
2521 */
2522 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2523 i = fxdr_unsigned(int, *tl);
2524 nfsm_adv(nfsm_rndup(i));
2525 dp->d_type = IFTODT(VTTOIF(VDIR));
2526 }
2527 if (doit) {
2528 nfsm_getfh(fhp, fhsize, 1);
2529 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2530 VREF(vp);
2531 newvp = vp;
2532 np = dnp;
2533 } else {
2534 error = nfs_nget(vp->v_mount, fhp,
2535 fhsize, &np, LK_EXCLUSIVE);
2536 if (error)
2537 doit = 0;
2538 else
2539 newvp = NFSTOV(np);
2540 }
2541 }
2542 if (doit && bigenough) {
2543 dpossav2 = dpos;
2544 dpos = dpossav1;
2545 mdsav2 = md;
2546 md = mdsav1;
2547 nfsm_loadattr(newvp, NULL);
2548 dpos = dpossav2;
2549 md = mdsav2;
2550 dp->d_type =
2551 IFTODT(VTTOIF(np->n_vattr.va_type));
2552 ndp->ni_vp = newvp;
2553 /*
2554 * Update n_ctime so subsequent lookup
2555 * doesn't purge entry.
2556 */
2557 np->n_ctime = np->n_vattr.va_ctime;
2558 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2559 }
2560 } else {
2561 /* Just skip over the file handle */
2562 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2563 i = fxdr_unsigned(int, *tl);
2564 if (i) {
2565 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2566 fhsize = fxdr_unsigned(int, *tl);
2567 nfsm_adv(nfsm_rndup(fhsize));
2568 }
2569 }
2570 if (newvp != NULLVP) {
2571 if (newvp == vp)
2572 vrele(newvp);
2573 else
2574 vput(newvp);
2575 newvp = NULLVP;
2576 }
2577 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2578 more_dirs = fxdr_unsigned(int, *tl);
2579 }
2580 /*
2581 * If at end of rpc data, get the eof boolean
2582 */
2583 if (!more_dirs) {
2584 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2585 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2586 }
2587 m_freem(mrep);
2588 }
2589 /*
2590 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2591 * by increasing d_reclen for the last record.
2592 */
2593 if (blksiz > 0) {
2594 left = DIRBLKSIZ - blksiz;
2595 dp->d_reclen += left;
2596 uiop->uio_iov->iov_base =
2597 (char *)uiop->uio_iov->iov_base + left;
2598 uiop->uio_iov->iov_len -= left;
2599 uiop->uio_offset += left;
2600 uiop->uio_resid -= left;
2601 }
2602
2603 /*
2604 * We are now either at the end of the directory or have filled the
2605 * block.
2606 */
2607 if (bigenough)
2608 dnp->n_direofoffset = uiop->uio_offset;
2609 else {
2610 if (uiop->uio_resid > 0)
2611 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2612 nfs_dircookie_lock(dnp);
2613 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2614 *cookiep = cookie;
2615 nfs_dircookie_unlock(dnp);
2616 }
2617 nfsmout:
2618 if (newvp != NULLVP) {
2619 if (newvp == vp)
2620 vrele(newvp);
2621 else
2622 vput(newvp);
2623 newvp = NULLVP;
2624 }
2625 return (error);
2626 }
2627
2628 /*
2629 * Silly rename. To make the NFS filesystem that is stateless look a little
2630 * more like the "ufs" a remove of an active vnode is translated to a rename
2631 * to a funny looking filename that is removed by nfs_inactive on the
2632 * nfsnode. There is the potential for another process on a different client
2633 * to create the same funny name between the nfs_lookitup() fails and the
2634 * nfs_rename() completes, but...
2635 */
2636 static int
2637 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2638 {
2639 struct sillyrename *sp;
2640 struct nfsnode *np;
2641 int error;
2642 short pid;
2643 unsigned int lticks;
2644
2645 cache_purge(dvp);
2646 np = VTONFS(vp);
2647 #ifndef DIAGNOSTIC
2648 if (vp->v_type == VDIR)
2649 panic("nfs: sillyrename dir");
2650 #endif
2651 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2652 M_NFSREQ, M_WAITOK);
2653 sp->s_cred = crhold(cnp->cn_cred);
2654 sp->s_dvp = dvp;
2655 sp->s_removeit = nfs_removeit;
2656 VREF(dvp);
2657
2658 /*
2659 * Fudge together a funny name.
2660 * Changing the format of the funny name to accomodate more
2661 * sillynames per directory.
2662 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2663 * CPU ticks since boot.
2664 */
2665 pid = cnp->cn_thread->td_proc->p_pid;
2666 lticks = (unsigned int)ticks;
2667 for ( ; ; ) {
2668 sp->s_namlen = sprintf(sp->s_name,
2669 ".nfs.%08x.%04x4.4", lticks,
2670 pid);
2671 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2672 cnp->cn_thread, NULL))
2673 break;
2674 lticks++;
2675 }
2676 error = nfs_renameit(dvp, cnp, sp);
2677 if (error)
2678 goto bad;
2679 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2680 cnp->cn_thread, &np);
2681 np->n_sillyrename = sp;
2682 return (0);
2683 bad:
2684 vrele(sp->s_dvp);
2685 crfree(sp->s_cred);
2686 free((caddr_t)sp, M_NFSREQ);
2687 return (error);
2688 }
2689
2690 /*
2691 * Look up a file name and optionally either update the file handle or
2692 * allocate an nfsnode, depending on the value of npp.
2693 * npp == NULL --> just do the lookup
2694 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2695 * handled too
2696 * *npp != NULL --> update the file handle in the vnode
2697 */
2698 static int
2699 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2700 struct thread *td, struct nfsnode **npp)
2701 {
2702 struct vnode *newvp = NULL;
2703 struct nfsnode *np, *dnp = VTONFS(dvp);
2704 caddr_t bpos, dpos;
2705 int error = 0, fhlen, attrflag;
2706 struct mbuf *mreq, *mrep, *md, *mb;
2707 nfsfh_t *nfhp;
2708 int v3 = NFS_ISV3(dvp);
2709
2710 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2711 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2712 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2713 mb = mreq;
2714 bpos = mtod(mb, caddr_t);
2715 nfsm_fhtom(dvp, v3);
2716 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2717 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2718 if (npp && !error) {
2719 nfsm_getfh(nfhp, fhlen, v3);
2720 if (*npp) {
2721 np = *npp;
2722 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2723 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2724 np->n_fhp = &np->n_fh;
2725 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2726 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2727 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2728 np->n_fhsize = fhlen;
2729 newvp = NFSTOV(np);
2730 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2731 VREF(dvp);
2732 newvp = dvp;
2733 } else {
2734 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2735 if (error) {
2736 m_freem(mrep);
2737 return (error);
2738 }
2739 newvp = NFSTOV(np);
2740 }
2741 if (v3) {
2742 nfsm_postop_attr(newvp, attrflag);
2743 if (!attrflag && *npp == NULL) {
2744 m_freem(mrep);
2745 if (newvp == dvp)
2746 vrele(newvp);
2747 else
2748 vput(newvp);
2749 return (ENOENT);
2750 }
2751 } else
2752 nfsm_loadattr(newvp, NULL);
2753 }
2754 m_freem(mrep);
2755 nfsmout:
2756 if (npp && *npp == NULL) {
2757 if (error) {
2758 if (newvp) {
2759 if (newvp == dvp)
2760 vrele(newvp);
2761 else
2762 vput(newvp);
2763 }
2764 } else
2765 *npp = np;
2766 }
2767 return (error);
2768 }
2769
2770 /*
2771 * Nfs Version 3 commit rpc
2772 */
2773 int
2774 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2775 struct thread *td)
2776 {
2777 u_int32_t *tl;
2778 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2779 caddr_t bpos, dpos;
2780 int error = 0, wccflag = NFSV3_WCCRATTR;
2781 struct mbuf *mreq, *mrep, *md, *mb;
2782
2783 mtx_lock(&nmp->nm_mtx);
2784 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2785 mtx_unlock(&nmp->nm_mtx);
2786 return (0);
2787 }
2788 mtx_unlock(&nmp->nm_mtx);
2789 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2790 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2791 mb = mreq;
2792 bpos = mtod(mb, caddr_t);
2793 nfsm_fhtom(vp, 1);
2794 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2795 txdr_hyper(offset, tl);
2796 tl += 2;
2797 *tl = txdr_unsigned(cnt);
2798 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2799 nfsm_wcc_data(vp, wccflag);
2800 if (!error) {
2801 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2802 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2803 NFSX_V3WRITEVERF)) {
2804 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2805 NFSX_V3WRITEVERF);
2806 error = NFSERR_STALEWRITEVERF;
2807 }
2808 }
2809 m_freem(mrep);
2810 nfsmout:
2811 return (error);
2812 }
2813
2814 /*
2815 * Strategy routine.
2816 * For async requests when nfsiod(s) are running, queue the request by
2817 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2818 * request.
2819 */
2820 static int
2821 nfs_strategy(struct vop_strategy_args *ap)
2822 {
2823 struct buf *bp = ap->a_bp;
2824 struct ucred *cr;
2825
2826 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2827 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
2828
2829 if (bp->b_iocmd == BIO_READ)
2830 cr = bp->b_rcred;
2831 else
2832 cr = bp->b_wcred;
2833
2834 /*
2835 * If the op is asynchronous and an i/o daemon is waiting
2836 * queue the request, wake it up and wait for completion
2837 * otherwise just do it ourselves.
2838 */
2839 if ((bp->b_flags & B_ASYNC) == 0 ||
2840 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2841 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2842 return (0);
2843 }
2844
2845 /*
2846 * fsync vnode op. Just call nfs_flush() with commit == 1.
2847 */
2848 /* ARGSUSED */
2849 static int
2850 nfs_fsync(struct vop_fsync_args *ap)
2851 {
2852 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2853 }
2854
2855 /*
2856 * Flush all the blocks associated with a vnode.
2857 * Walk through the buffer pool and push any dirty pages
2858 * associated with the vnode.
2859 */
2860 static int
2861 nfs_flush(struct vnode *vp, int waitfor, struct thread *td,
2862 int commit)
2863 {
2864 struct nfsnode *np = VTONFS(vp);
2865 struct buf *bp;
2866 int i;
2867 struct buf *nbp;
2868 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2869 int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2870 int passone = 1;
2871 u_quad_t off, endoff, toff;
2872 struct ucred* wcred = NULL;
2873 struct buf **bvec = NULL;
2874 #ifndef NFS_COMMITBVECSIZ
2875 #define NFS_COMMITBVECSIZ 20
2876 #endif
2877 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2878 int bvecsize = 0, bveccount;
2879
2880 if (nmp->nm_flag & NFSMNT_INT)
2881 slpflag = PCATCH;
2882 if (!commit)
2883 passone = 0;
2884 /*
2885 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2886 * server, but has not been committed to stable storage on the server
2887 * yet. On the first pass, the byte range is worked out and the commit
2888 * rpc is done. On the second pass, nfs_writebp() is called to do the
2889 * job.
2890 */
2891 again:
2892 off = (u_quad_t)-1;
2893 endoff = 0;
2894 bvecpos = 0;
2895 if (NFS_ISV3(vp) && commit) {
2896 s = splbio();
2897 if (bvec != NULL && bvec != bvec_on_stack)
2898 free(bvec, M_TEMP);
2899 /*
2900 * Count up how many buffers waiting for a commit.
2901 */
2902 bveccount = 0;
2903 VI_LOCK(vp);
2904 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2905 if (BUF_REFCNT(bp) == 0 &&
2906 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2907 == (B_DELWRI | B_NEEDCOMMIT))
2908 bveccount++;
2909 }
2910 /*
2911 * Allocate space to remember the list of bufs to commit. It is
2912 * important to use M_NOWAIT here to avoid a race with nfs_write.
2913 * If we can't get memory (for whatever reason), we will end up
2914 * committing the buffers one-by-one in the loop below.
2915 */
2916 if (bveccount > NFS_COMMITBVECSIZ) {
2917 /*
2918 * Release the vnode interlock to avoid a lock
2919 * order reversal.
2920 */
2921 VI_UNLOCK(vp);
2922 bvec = (struct buf **)
2923 malloc(bveccount * sizeof(struct buf *),
2924 M_TEMP, M_NOWAIT);
2925 VI_LOCK(vp);
2926 if (bvec == NULL) {
2927 bvec = bvec_on_stack;
2928 bvecsize = NFS_COMMITBVECSIZ;
2929 } else
2930 bvecsize = bveccount;
2931 } else {
2932 bvec = bvec_on_stack;
2933 bvecsize = NFS_COMMITBVECSIZ;
2934 }
2935 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2936 if (bvecpos >= bvecsize)
2937 break;
2938 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2939 nbp = TAILQ_NEXT(bp, b_bobufs);
2940 continue;
2941 }
2942 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2943 (B_DELWRI | B_NEEDCOMMIT)) {
2944 BUF_UNLOCK(bp);
2945 nbp = TAILQ_NEXT(bp, b_bobufs);
2946 continue;
2947 }
2948 VI_UNLOCK(vp);
2949 bremfree(bp);
2950 /*
2951 * Work out if all buffers are using the same cred
2952 * so we can deal with them all with one commit.
2953 *
2954 * NOTE: we are not clearing B_DONE here, so we have
2955 * to do it later on in this routine if we intend to
2956 * initiate I/O on the bp.
2957 *
2958 * Note: to avoid loopback deadlocks, we do not
2959 * assign b_runningbufspace.
2960 */
2961 if (wcred == NULL)
2962 wcred = bp->b_wcred;
2963 else if (wcred != bp->b_wcred)
2964 wcred = NOCRED;
2965 vfs_busy_pages(bp, 1);
2966
2967 VI_LOCK(vp);
2968 /*
2969 * bp is protected by being locked, but nbp is not
2970 * and vfs_busy_pages() may sleep. We have to
2971 * recalculate nbp.
2972 */
2973 nbp = TAILQ_NEXT(bp, b_bobufs);
2974
2975 /*
2976 * A list of these buffers is kept so that the
2977 * second loop knows which buffers have actually
2978 * been committed. This is necessary, since there
2979 * may be a race between the commit rpc and new
2980 * uncommitted writes on the file.
2981 */
2982 bvec[bvecpos++] = bp;
2983 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2984 bp->b_dirtyoff;
2985 if (toff < off)
2986 off = toff;
2987 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2988 if (toff > endoff)
2989 endoff = toff;
2990 }
2991 splx(s);
2992 VI_UNLOCK(vp);
2993 }
2994 if (bvecpos > 0) {
2995 /*
2996 * Commit data on the server, as required.
2997 * If all bufs are using the same wcred, then use that with
2998 * one call for all of them, otherwise commit each one
2999 * separately.
3000 */
3001 if (wcred != NOCRED)
3002 retv = nfs_commit(vp, off, (int)(endoff - off),
3003 wcred, td);
3004 else {
3005 retv = 0;
3006 for (i = 0; i < bvecpos; i++) {
3007 off_t off, size;
3008 bp = bvec[i];
3009 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3010 bp->b_dirtyoff;
3011 size = (u_quad_t)(bp->b_dirtyend
3012 - bp->b_dirtyoff);
3013 retv = nfs_commit(vp, off, (int)size,
3014 bp->b_wcred, td);
3015 if (retv) break;
3016 }
3017 }
3018
3019 if (retv == NFSERR_STALEWRITEVERF)
3020 nfs_clearcommit(vp->v_mount);
3021
3022 /*
3023 * Now, either mark the blocks I/O done or mark the
3024 * blocks dirty, depending on whether the commit
3025 * succeeded.
3026 */
3027 for (i = 0; i < bvecpos; i++) {
3028 bp = bvec[i];
3029 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3030 if (retv) {
3031 /*
3032 * Error, leave B_DELWRI intact
3033 */
3034 vfs_unbusy_pages(bp);
3035 brelse(bp);
3036 } else {
3037 /*
3038 * Success, remove B_DELWRI ( bundirty() ).
3039 *
3040 * b_dirtyoff/b_dirtyend seem to be NFS
3041 * specific. We should probably move that
3042 * into bundirty(). XXX
3043 */
3044 s = splbio();
3045 bufobj_wref(&vp->v_bufobj);
3046 bp->b_flags |= B_ASYNC;
3047 bundirty(bp);
3048 bp->b_flags &= ~B_DONE;
3049 bp->b_ioflags &= ~BIO_ERROR;
3050 bp->b_dirtyoff = bp->b_dirtyend = 0;
3051 splx(s);
3052 bufdone(bp);
3053 }
3054 }
3055 }
3056
3057 /*
3058 * Start/do any write(s) that are required.
3059 */
3060 loop:
3061 s = splbio();
3062 VI_LOCK(vp);
3063 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
3064 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3065 if (waitfor != MNT_WAIT || passone)
3066 continue;
3067
3068 error = BUF_TIMELOCK(bp,
3069 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3070 VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
3071 splx(s);
3072 if (error == 0) {
3073 BUF_UNLOCK(bp);
3074 goto loop;
3075 }
3076 if (error == ENOLCK)
3077 goto loop;
3078 if (nfs_sigintr(nmp, NULL, td)) {
3079 error = EINTR;
3080 goto done;
3081 }
3082 if (slpflag == PCATCH) {
3083 slpflag = 0;
3084 slptimeo = 2 * hz;
3085 }
3086 goto loop;
3087 }
3088 if ((bp->b_flags & B_DELWRI) == 0)
3089 panic("nfs_fsync: not dirty");
3090 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3091 BUF_UNLOCK(bp);
3092 continue;
3093 }
3094 VI_UNLOCK(vp);
3095 bremfree(bp);
3096 if (passone || !commit)
3097 bp->b_flags |= B_ASYNC;
3098 else
3099 bp->b_flags |= B_ASYNC;
3100 splx(s);
3101 bwrite(bp);
3102 if (nfs_sigintr(nmp, NULL, td)) {
3103 error = EINTR;
3104 goto done;
3105 }
3106 goto loop;
3107 }
3108 splx(s);
3109 if (passone) {
3110 passone = 0;
3111 VI_UNLOCK(vp);
3112 goto again;
3113 }
3114 if (waitfor == MNT_WAIT) {
3115 while (vp->v_bufobj.bo_numoutput) {
3116 error = bufobj_wwait(&vp->v_bufobj, slpflag, slptimeo);
3117 if (error) {
3118 VI_UNLOCK(vp);
3119 error = nfs_sigintr(nmp, NULL, td);
3120 if (error)
3121 goto done;
3122 if (slpflag == PCATCH) {
3123 slpflag = 0;
3124 slptimeo = 2 * hz;
3125 }
3126 VI_LOCK(vp);
3127 }
3128 }
3129 if (vp->v_bufobj.bo_dirty.bv_cnt != 0 && commit) {
3130 VI_UNLOCK(vp);
3131 goto loop;
3132 }
3133 /*
3134 * Wait for all the async IO requests to drain
3135 */
3136 VI_UNLOCK(vp);
3137 mtx_lock(&np->n_mtx);
3138 while (np->n_directio_asyncwr > 0) {
3139 np->n_flag |= NFSYNCWAIT;
3140 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3141 &np->n_mtx, slpflag | (PRIBIO + 1),
3142 "nfsfsync", 0);
3143 if (error) {
3144 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
3145 mtx_unlock(&np->n_mtx);
3146 error = EINTR;
3147 goto done;
3148 }
3149 }
3150 }
3151 mtx_unlock(&np->n_mtx);
3152 } else
3153 VI_UNLOCK(vp);
3154 mtx_lock(&np->n_mtx);
3155 if (np->n_flag & NWRITEERR) {
3156 error = np->n_error;
3157 np->n_flag &= ~NWRITEERR;
3158 }
3159 if (commit && vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
3160 vp->v_bufobj.bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3161 np->n_flag &= ~NMODIFIED;
3162 mtx_unlock(&np->n_mtx);
3163 done:
3164 if (bvec != NULL && bvec != bvec_on_stack)
3165 free(bvec, M_TEMP);
3166 return (error);
3167 }
3168
3169 /*
3170 * NFS advisory byte-level locks.
3171 */
3172 static int
3173 nfs_advlock(struct vop_advlock_args *ap)
3174 {
3175 struct vnode *vp = ap->a_vp;
3176 u_quad_t size;
3177 int error;
3178
3179 error = vn_lock(vp, LK_SHARED, curthread);
3180 if (error)
3181 return (error);
3182 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3183 size = VTONFS(vp)->n_size;
3184 VOP_UNLOCK(vp, 0, curthread);
3185 error = lf_advlock(ap, &(vp->v_lockf), size);
3186 } else {
3187 if (nfs_advlock_p)
3188 error = nfs_advlock_p(ap);
3189 else
3190 error = ENOLCK;
3191 }
3192
3193 return (error);
3194 }
3195
3196 /*
3197 * NFS advisory byte-level locks.
3198 */
3199 static int
3200 nfs_advlockasync(struct vop_advlockasync_args *ap)
3201 {
3202 struct vnode *vp = ap->a_vp;
3203 u_quad_t size;
3204 int error;
3205
3206 error = vn_lock(vp, LK_SHARED, curthread);
3207 if (error)
3208 return (error);
3209 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3210 size = VTONFS(vp)->n_size;
3211 VOP_UNLOCK(vp, 0, curthread);
3212 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3213 } else {
3214 VOP_UNLOCK(vp, 0, curthread);
3215 error = EOPNOTSUPP;
3216 }
3217 return (error);
3218 }
3219
3220 /*
3221 * Print out the contents of an nfsnode.
3222 */
3223 static int
3224 nfs_print(struct vop_print_args *ap)
3225 {
3226 struct vnode *vp = ap->a_vp;
3227 struct nfsnode *np = VTONFS(vp);
3228
3229 nfs_printf("\tfileid %ld fsid 0x%x",
3230 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3231 if (vp->v_type == VFIFO)
3232 fifo_printinfo(vp);
3233 printf("\n");
3234 return (0);
3235 }
3236
3237 /*
3238 * This is the "real" nfs::bwrite(struct buf*).
3239 * We set B_CACHE if this is a VMIO buffer.
3240 */
3241 int
3242 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3243 {
3244 int s;
3245 int oldflags = bp->b_flags;
3246 #if 0
3247 int retv = 1;
3248 off_t off;
3249 #endif
3250
3251 if (BUF_REFCNT(bp) == 0)
3252 panic("bwrite: buffer is not locked???");
3253
3254 if (bp->b_flags & B_INVAL) {
3255 brelse(bp);
3256 return(0);
3257 }
3258
3259 bp->b_flags |= B_CACHE;
3260
3261 /*
3262 * Undirty the bp. We will redirty it later if the I/O fails.
3263 */
3264
3265 s = splbio();
3266 bundirty(bp);
3267 bp->b_flags &= ~B_DONE;
3268 bp->b_ioflags &= ~BIO_ERROR;
3269 bp->b_iocmd = BIO_WRITE;
3270
3271 bufobj_wref(bp->b_bufobj);
3272 curthread->td_ru.ru_oublock++;
3273 splx(s);
3274
3275 /*
3276 * Note: to avoid loopback deadlocks, we do not
3277 * assign b_runningbufspace.
3278 */
3279 vfs_busy_pages(bp, 1);
3280
3281 BUF_KERNPROC(bp);
3282 bp->b_iooffset = dbtob(bp->b_blkno);
3283 bstrategy(bp);
3284
3285 if( (oldflags & B_ASYNC) == 0) {
3286 int rtval = bufwait(bp);
3287
3288 if (oldflags & B_DELWRI) {
3289 s = splbio();
3290 reassignbuf(bp);
3291 splx(s);
3292 }
3293 brelse(bp);
3294 return (rtval);
3295 }
3296
3297 return (0);
3298 }
3299
3300 /*
3301 * nfs special file access vnode op.
3302 * Essentially just get vattr and then imitate iaccess() since the device is
3303 * local to the client.
3304 */
3305 static int
3306 nfsspec_access(struct vop_access_args *ap)
3307 {
3308 struct vattr *vap;
3309 struct ucred *cred = ap->a_cred;
3310 struct vnode *vp = ap->a_vp;
3311 mode_t mode = ap->a_mode;
3312 struct vattr vattr;
3313 int error;
3314
3315 /*
3316 * Disallow write attempts on filesystems mounted read-only;
3317 * unless the file is a socket, fifo, or a block or character
3318 * device resident on the filesystem.
3319 */
3320 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3321 switch (vp->v_type) {
3322 case VREG:
3323 case VDIR:
3324 case VLNK:
3325 return (EROFS);
3326 default:
3327 break;
3328 }
3329 }
3330 vap = &vattr;
3331 error = VOP_GETATTR(vp, vap, cred, ap->a_td);
3332 if (error)
3333 goto out;
3334 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3335 mode, cred, NULL);
3336 out:
3337 return error;
3338 }
3339
3340 /*
3341 * Read wrapper for fifos.
3342 */
3343 static int
3344 nfsfifo_read(struct vop_read_args *ap)
3345 {
3346 struct nfsnode *np = VTONFS(ap->a_vp);
3347 int error;
3348
3349 /*
3350 * Set access flag.
3351 */
3352 mtx_lock(&np->n_mtx);
3353 np->n_flag |= NACC;
3354 getnanotime(&np->n_atim);
3355 mtx_unlock(&np->n_mtx);
3356 error = fifo_specops.vop_read(ap);
3357 return error;
3358 }
3359
3360 /*
3361 * Write wrapper for fifos.
3362 */
3363 static int
3364 nfsfifo_write(struct vop_write_args *ap)
3365 {
3366 struct nfsnode *np = VTONFS(ap->a_vp);
3367
3368 /*
3369 * Set update flag.
3370 */
3371 mtx_lock(&np->n_mtx);
3372 np->n_flag |= NUPD;
3373 getnanotime(&np->n_mtim);
3374 mtx_unlock(&np->n_mtx);
3375 return(fifo_specops.vop_write(ap));
3376 }
3377
3378 /*
3379 * Close wrapper for fifos.
3380 *
3381 * Update the times on the nfsnode then do fifo close.
3382 */
3383 static int
3384 nfsfifo_close(struct vop_close_args *ap)
3385 {
3386 struct vnode *vp = ap->a_vp;
3387 struct nfsnode *np = VTONFS(vp);
3388 struct vattr vattr;
3389 struct timespec ts;
3390
3391 mtx_lock(&np->n_mtx);
3392 if (np->n_flag & (NACC | NUPD)) {
3393 getnanotime(&ts);
3394 if (np->n_flag & NACC)
3395 np->n_atim = ts;
3396 if (np->n_flag & NUPD)
3397 np->n_mtim = ts;
3398 np->n_flag |= NCHG;
3399 if (vrefcnt(vp) == 1 &&
3400 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3401 VATTR_NULL(&vattr);
3402 if (np->n_flag & NACC)
3403 vattr.va_atime = np->n_atim;
3404 if (np->n_flag & NUPD)
3405 vattr.va_mtime = np->n_mtim;
3406 mtx_unlock(&np->n_mtx);
3407 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
3408 goto out;
3409 }
3410 }
3411 mtx_unlock(&np->n_mtx);
3412 out:
3413 return (fifo_specops.vop_close(ap));
3414 }
3415
3416 /*
3417 * Just call nfs_writebp() with the force argument set to 1.
3418 *
3419 * NOTE: B_DONE may or may not be set in a_bp on call.
3420 */
3421 static int
3422 nfs_bwrite(struct buf *bp)
3423 {
3424
3425 return (nfs_writebp(bp, 1, curthread));
3426 }
3427
3428 struct buf_ops buf_ops_nfs = {
3429 .bop_name = "buf_ops_nfs",
3430 .bop_write = nfs_bwrite,
3431 .bop_strategy = bufstrategy,
3432 .bop_sync = bufsync,
3433 .bop_bdflush = bufbdflush,
3434 };
Cache object: 620de813945e95065d9a5d664959a3d0
|