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("$FreeBSD$");
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 nametimeo = NFS_DEFAULT_NAMETIMEO;
229 SYSCTL_UINT(_vfs_nfs, OID_AUTO, name_timeout, CTLFLAG_RW,
230 &nametimeo, 0, "Positive name cache entry timeout");
231
232 static u_int negnametimeo = NFS_DEFAULT_NEGNAMETIMEO;
233 SYSCTL_UINT(_vfs_nfs, OID_AUTO, negative_name_timeout, CTLFLAG_RW,
234 &negnametimeo, 0, "Negative name cache entry timeout");
235
236 /*
237 * This sysctl allows other processes to mmap a file that has been opened
238 * O_DIRECT by a process. In general, having processes mmap the file while
239 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
240 * this by default to prevent DoS attacks - to prevent a malicious user from
241 * opening up files O_DIRECT preventing other users from mmap'ing these
242 * files. "Protected" environments where stricter consistency guarantees are
243 * required can disable this knob. The process that opened the file O_DIRECT
244 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
245 * meaningful.
246 */
247 int nfs_directio_allow_mmap = 1;
248 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
249 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
250
251 #if 0
252 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
253 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
254
255 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
256 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
257 #endif
258
259 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
260 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
261 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
262
263 /*
264 * SMP Locking Note :
265 * The list of locks after the description of the lock is the ordering
266 * of other locks acquired with the lock held.
267 * np->n_mtx : Protects the fields in the nfsnode.
268 VM Object Lock
269 VI_MTX (acquired indirectly)
270 * nmp->nm_mtx : Protects the fields in the nfsmount.
271 rep->r_mtx
272 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
273 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
274 nmp->nm_mtx
275 rep->r_mtx
276 * rep->r_mtx : Protects the fields in an nfsreq.
277 */
278
279 static int
280 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
281 struct ucred *cred)
282 {
283 const int v3 = 1;
284 u_int32_t *tl;
285 int error = 0, attrflag;
286
287 struct mbuf *mreq, *mrep, *md, *mb;
288 caddr_t bpos, dpos;
289 u_int32_t rmode;
290 struct nfsnode *np = VTONFS(vp);
291
292 nfsstats.rpccnt[NFSPROC_ACCESS]++;
293 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
294 mb = mreq;
295 bpos = mtod(mb, caddr_t);
296 nfsm_fhtom(vp, v3);
297 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
298 *tl = txdr_unsigned(wmode);
299 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
300 nfsm_postop_attr(vp, attrflag);
301 if (!error) {
302 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
303 rmode = fxdr_unsigned(u_int32_t, *tl);
304 mtx_lock(&np->n_mtx);
305 np->n_mode = rmode;
306 np->n_modeuid = cred->cr_uid;
307 np->n_modestamp = time_second;
308 mtx_unlock(&np->n_mtx);
309 }
310 m_freem(mrep);
311 nfsmout:
312 return (error);
313 }
314
315 /*
316 * nfs access vnode op.
317 * For nfs version 2, just return ok. File accesses may fail later.
318 * For nfs version 3, use the access rpc to check accessibility. If file modes
319 * are changed on the server, accesses might still fail later.
320 */
321 static int
322 nfs_access(struct vop_access_args *ap)
323 {
324 struct vnode *vp = ap->a_vp;
325 int error = 0;
326 u_int32_t mode, wmode;
327 int v3 = NFS_ISV3(vp);
328 struct nfsnode *np = VTONFS(vp);
329
330 /*
331 * Disallow write attempts on filesystems mounted read-only;
332 * unless the file is a socket, fifo, or a block or character
333 * device resident on the filesystem.
334 */
335 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
336 switch (vp->v_type) {
337 case VREG:
338 case VDIR:
339 case VLNK:
340 return (EROFS);
341 default:
342 break;
343 }
344 }
345 /*
346 * For nfs v3, check to see if we have done this recently, and if
347 * so return our cached result instead of making an ACCESS call.
348 * If not, do an access rpc, otherwise you are stuck emulating
349 * ufs_access() locally using the vattr. This may not be correct,
350 * since the server may apply other access criteria such as
351 * client uid-->server uid mapping that we do not know about.
352 */
353 if (v3) {
354 if (ap->a_mode & VREAD)
355 mode = NFSV3ACCESS_READ;
356 else
357 mode = 0;
358 if (vp->v_type != VDIR) {
359 if (ap->a_mode & VWRITE)
360 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
361 if (ap->a_mode & VEXEC)
362 mode |= NFSV3ACCESS_EXECUTE;
363 } else {
364 if (ap->a_mode & VWRITE)
365 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
366 NFSV3ACCESS_DELETE);
367 if (ap->a_mode & VEXEC)
368 mode |= NFSV3ACCESS_LOOKUP;
369 }
370 /* XXX safety belt, only make blanket request if caching */
371 if (nfsaccess_cache_timeout > 0) {
372 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
373 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
374 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
375 } else {
376 wmode = mode;
377 }
378
379 /*
380 * Does our cached result allow us to give a definite yes to
381 * this request?
382 */
383 mtx_lock(&np->n_mtx);
384 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
385 (ap->a_cred->cr_uid == np->n_modeuid) &&
386 ((np->n_mode & mode) == mode)) {
387 nfsstats.accesscache_hits++;
388 } else {
389 /*
390 * Either a no, or a don't know. Go to the wire.
391 */
392 nfsstats.accesscache_misses++;
393 mtx_unlock(&np->n_mtx);
394 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
395 mtx_lock(&np->n_mtx);
396 if (!error) {
397 if ((np->n_mode & mode) != mode) {
398 error = EACCES;
399 }
400 }
401 }
402 mtx_unlock(&np->n_mtx);
403 return (error);
404 } else {
405 if ((error = nfsspec_access(ap)) != 0) {
406 return (error);
407 }
408 /*
409 * Attempt to prevent a mapped root from accessing a file
410 * which it shouldn't. We try to read a byte from the file
411 * if the user is root and the file is not zero length.
412 * After calling nfsspec_access, we should have the correct
413 * file size cached.
414 */
415 mtx_lock(&np->n_mtx);
416 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
417 && VTONFS(vp)->n_size > 0) {
418 struct iovec aiov;
419 struct uio auio;
420 char buf[1];
421
422 mtx_unlock(&np->n_mtx);
423 aiov.iov_base = buf;
424 aiov.iov_len = 1;
425 auio.uio_iov = &aiov;
426 auio.uio_iovcnt = 1;
427 auio.uio_offset = 0;
428 auio.uio_resid = 1;
429 auio.uio_segflg = UIO_SYSSPACE;
430 auio.uio_rw = UIO_READ;
431 auio.uio_td = ap->a_td;
432
433 if (vp->v_type == VREG)
434 error = nfs_readrpc(vp, &auio, ap->a_cred);
435 else if (vp->v_type == VDIR) {
436 char* bp;
437 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
438 aiov.iov_base = bp;
439 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
440 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
441 free(bp, M_TEMP);
442 } else if (vp->v_type == VLNK)
443 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
444 else
445 error = EACCES;
446 } else
447 mtx_unlock(&np->n_mtx);
448 return (error);
449 }
450 }
451
452 int nfs_otw_getattr_avoid = 0;
453
454 /*
455 * nfs open vnode op
456 * Check to see if the type is ok
457 * and that deletion is not in progress.
458 * For paged in text files, you will need to flush the page cache
459 * if consistency is lost.
460 */
461 /* ARGSUSED */
462 static int
463 nfs_open(struct vop_open_args *ap)
464 {
465 struct vnode *vp = ap->a_vp;
466 struct nfsnode *np = VTONFS(vp);
467 struct vattr vattr;
468 int error;
469 int fmode = ap->a_mode;
470
471 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
472 return (EOPNOTSUPP);
473
474 /*
475 * Get a valid lease. If cached data is stale, flush it.
476 */
477 mtx_lock(&np->n_mtx);
478 if (np->n_flag & NMODIFIED) {
479 mtx_unlock(&np->n_mtx);
480 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
481 if (error == EINTR || error == EIO)
482 return (error);
483 mtx_lock(&np->n_mtx);
484 np->n_attrstamp = 0;
485 if (vp->v_type == VDIR)
486 np->n_direofoffset = 0;
487 mtx_unlock(&np->n_mtx);
488 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
489 if (error)
490 return (error);
491 mtx_lock(&np->n_mtx);
492 np->n_mtime = vattr.va_mtime;
493 mtx_unlock(&np->n_mtx);
494 } else {
495 mtx_unlock(&np->n_mtx);
496 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
497 if (error)
498 return (error);
499 mtx_lock(&np->n_mtx);
500 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
501 if (vp->v_type == VDIR)
502 np->n_direofoffset = 0;
503 mtx_unlock(&np->n_mtx);
504 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
505 if (error == EINTR || error == EIO) {
506 return (error);
507 }
508 mtx_lock(&np->n_mtx);
509 np->n_mtime = vattr.va_mtime;
510 }
511 mtx_unlock(&np->n_mtx);
512 }
513 /*
514 * If the object has >= 1 O_DIRECT active opens, we disable caching.
515 */
516 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
517 if (np->n_directio_opens == 0) {
518 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
519 if (error)
520 return (error);
521 mtx_lock(&np->n_mtx);
522 np->n_flag |= NNONCACHE;
523 mtx_unlock(&np->n_mtx);
524 }
525 np->n_directio_opens++;
526 }
527 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
528 return (0);
529 }
530
531 /*
532 * nfs close vnode op
533 * What an NFS client should do upon close after writing is a debatable issue.
534 * Most NFS clients push delayed writes to the server upon close, basically for
535 * two reasons:
536 * 1 - So that any write errors may be reported back to the client process
537 * doing the close system call. By far the two most likely errors are
538 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
539 * 2 - To put a worst case upper bound on cache inconsistency between
540 * multiple clients for the file.
541 * There is also a consistency problem for Version 2 of the protocol w.r.t.
542 * not being able to tell if other clients are writing a file concurrently,
543 * since there is no way of knowing if the changed modify time in the reply
544 * is only due to the write for this client.
545 * (NFS Version 3 provides weak cache consistency data in the reply that
546 * should be sufficient to detect and handle this case.)
547 *
548 * The current code does the following:
549 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
550 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
551 * or commit them (this satisfies 1 and 2 except for the
552 * case where the server crashes after this close but
553 * before the commit RPC, which is felt to be "good
554 * enough". Changing the last argument to nfs_flush() to
555 * a 1 would force a commit operation, if it is felt a
556 * commit is necessary now.
557 */
558 /* ARGSUSED */
559 static int
560 nfs_close(struct vop_close_args *ap)
561 {
562 struct vnode *vp = ap->a_vp;
563 struct nfsnode *np = VTONFS(vp);
564 int error = 0;
565 int fmode = ap->a_fflag;
566
567 if (vp->v_type == VREG) {
568 /*
569 * Examine and clean dirty pages, regardless of NMODIFIED.
570 * This closes a major hole in close-to-open consistency.
571 * We want to push out all dirty pages (and buffers) on
572 * close, regardless of whether they were dirtied by
573 * mmap'ed writes or via write().
574 */
575 if (nfs_clean_pages_on_close && vp->v_object) {
576 VM_OBJECT_LOCK(vp->v_object);
577 vm_object_page_clean(vp->v_object, 0, 0, 0);
578 VM_OBJECT_UNLOCK(vp->v_object);
579 }
580 mtx_lock(&np->n_mtx);
581 if (np->n_flag & NMODIFIED) {
582 mtx_unlock(&np->n_mtx);
583 if (NFS_ISV3(vp)) {
584 /*
585 * Under NFSv3 we have dirty buffers to dispose of. We
586 * must flush them to the NFS server. We have the option
587 * of waiting all the way through the commit rpc or just
588 * waiting for the initial write. The default is to only
589 * wait through the initial write so the data is in the
590 * server's cache, which is roughly similar to the state
591 * a standard disk subsystem leaves the file in on close().
592 *
593 * We cannot clear the NMODIFIED bit in np->n_flag due to
594 * potential races with other processes, and certainly
595 * cannot clear it if we don't commit.
596 */
597 int cm = nfsv3_commit_on_close ? 1 : 0;
598 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
599 /* np->n_flag &= ~NMODIFIED; */
600 } else
601 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
602 mtx_lock(&np->n_mtx);
603 }
604 if (np->n_flag & NWRITEERR) {
605 np->n_flag &= ~NWRITEERR;
606 error = np->n_error;
607 }
608 mtx_unlock(&np->n_mtx);
609 }
610 if (nfs_directio_enable)
611 KASSERT((np->n_directio_asyncwr == 0),
612 ("nfs_close: dirty unflushed (%d) directio buffers\n",
613 np->n_directio_asyncwr));
614 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
615 mtx_lock(&np->n_mtx);
616 KASSERT((np->n_directio_opens > 0),
617 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
618 np->n_directio_opens--;
619 if (np->n_directio_opens == 0)
620 np->n_flag &= ~NNONCACHE;
621 mtx_unlock(&np->n_mtx);
622 }
623 return (error);
624 }
625
626 /*
627 * nfs getattr call from vfs.
628 */
629 static int
630 nfs_getattr(struct vop_getattr_args *ap)
631 {
632 struct vnode *vp = ap->a_vp;
633 struct nfsnode *np = VTONFS(vp);
634 struct vattr *vap = ap->a_vap;
635 struct vattr vattr;
636 caddr_t bpos, dpos;
637 int error = 0;
638 struct mbuf *mreq, *mrep, *md, *mb;
639 int v3 = NFS_ISV3(vp);
640
641 /*
642 * Update local times for special files.
643 */
644 mtx_lock(&np->n_mtx);
645 if (np->n_flag & (NACC | NUPD))
646 np->n_flag |= NCHG;
647 mtx_unlock(&np->n_mtx);
648 /*
649 * First look in the cache.
650 */
651 if (nfs_getattrcache(vp, &vattr) == 0)
652 goto nfsmout;
653 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
654 nfsstats.accesscache_misses++;
655 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
656 if (nfs_getattrcache(vp, &vattr) == 0)
657 goto nfsmout;
658 }
659 nfsstats.rpccnt[NFSPROC_GETATTR]++;
660 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
661 mb = mreq;
662 bpos = mtod(mb, caddr_t);
663 nfsm_fhtom(vp, v3);
664 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
665 if (!error) {
666 nfsm_loadattr(vp, &vattr);
667 }
668 m_freem(mrep);
669 nfsmout:
670 vap->va_type = vattr.va_type;
671 vap->va_mode = vattr.va_mode;
672 vap->va_nlink = vattr.va_nlink;
673 vap->va_uid = vattr.va_uid;
674 vap->va_gid = vattr.va_gid;
675 vap->va_fsid = vattr.va_fsid;
676 vap->va_fileid = vattr.va_fileid;
677 vap->va_size = vattr.va_size;
678 vap->va_blocksize = vattr.va_blocksize;
679 vap->va_atime = vattr.va_atime;
680 vap->va_mtime = vattr.va_mtime;
681 vap->va_ctime = vattr.va_ctime;
682 vap->va_gen = vattr.va_gen;
683 vap->va_flags = vattr.va_flags;
684 vap->va_rdev = vattr.va_rdev;
685 vap->va_bytes = vattr.va_bytes;
686 vap->va_filerev = vattr.va_filerev;
687
688 return (error);
689 }
690
691 /*
692 * nfs setattr call.
693 */
694 static int
695 nfs_setattr(struct vop_setattr_args *ap)
696 {
697 struct vnode *vp = ap->a_vp;
698 struct nfsnode *np = VTONFS(vp);
699 struct vattr *vap = ap->a_vap;
700 int error = 0;
701 u_quad_t tsize;
702
703 #ifndef nolint
704 tsize = (u_quad_t)0;
705 #endif
706
707 /*
708 * Setting of flags and marking of atimes are not supported.
709 */
710 if (vap->va_flags != VNOVAL || (vap->va_vaflags & VA_MARK_ATIME))
711 return (EOPNOTSUPP);
712
713 /*
714 * Disallow write attempts if the filesystem is mounted read-only.
715 */
716 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
717 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
718 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
719 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
720 error = EROFS;
721 goto out;
722 }
723 if (vap->va_size != VNOVAL) {
724 switch (vp->v_type) {
725 case VDIR:
726 return (EISDIR);
727 case VCHR:
728 case VBLK:
729 case VSOCK:
730 case VFIFO:
731 if (vap->va_mtime.tv_sec == VNOVAL &&
732 vap->va_atime.tv_sec == VNOVAL &&
733 vap->va_mode == (mode_t)VNOVAL &&
734 vap->va_uid == (uid_t)VNOVAL &&
735 vap->va_gid == (gid_t)VNOVAL)
736 return (0);
737 vap->va_size = VNOVAL;
738 break;
739 default:
740 /*
741 * Disallow write attempts if the filesystem is
742 * mounted read-only.
743 */
744 if (vp->v_mount->mnt_flag & MNT_RDONLY)
745 return (EROFS);
746 /*
747 * We run vnode_pager_setsize() early (why?),
748 * we must set np->n_size now to avoid vinvalbuf
749 * V_SAVE races that might setsize a lower
750 * value.
751 */
752 mtx_lock(&np->n_mtx);
753 tsize = np->n_size;
754 mtx_unlock(&np->n_mtx);
755 error = nfs_meta_setsize(vp, ap->a_cred,
756 ap->a_td, vap->va_size);
757 mtx_lock(&np->n_mtx);
758 if (np->n_flag & NMODIFIED) {
759 tsize = np->n_size;
760 mtx_unlock(&np->n_mtx);
761 if (vap->va_size == 0)
762 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
763 else
764 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
765 if (error) {
766 vnode_pager_setsize(vp, tsize);
767 goto out;
768 }
769 } else
770 mtx_unlock(&np->n_mtx);
771 /*
772 * np->n_size has already been set to vap->va_size
773 * in nfs_meta_setsize(). We must set it again since
774 * nfs_loadattrcache() could be called through
775 * nfs_meta_setsize() and could modify np->n_size.
776 */
777 mtx_lock(&np->n_mtx);
778 np->n_vattr.va_size = np->n_size = vap->va_size;
779 mtx_unlock(&np->n_mtx);
780 };
781 } else {
782 mtx_lock(&np->n_mtx);
783 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
784 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
785 mtx_unlock(&np->n_mtx);
786 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) != 0 &&
787 (error == EINTR || error == EIO))
788 return error;
789 } else
790 mtx_unlock(&np->n_mtx);
791 }
792 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
793 if (error && vap->va_size != VNOVAL) {
794 mtx_lock(&np->n_mtx);
795 np->n_size = np->n_vattr.va_size = tsize;
796 vnode_pager_setsize(vp, tsize);
797 mtx_unlock(&np->n_mtx);
798 }
799 out:
800 return (error);
801 }
802
803 /*
804 * Do an nfs setattr rpc.
805 */
806 static int
807 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
808 struct thread *td)
809 {
810 struct nfsv2_sattr *sp;
811 struct nfsnode *np = VTONFS(vp);
812 caddr_t bpos, dpos;
813 u_int32_t *tl;
814 int error = 0, wccflag = NFSV3_WCCRATTR;
815 struct mbuf *mreq, *mrep, *md, *mb;
816 int v3 = NFS_ISV3(vp);
817
818 nfsstats.rpccnt[NFSPROC_SETATTR]++;
819 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
820 mb = mreq;
821 bpos = mtod(mb, caddr_t);
822 nfsm_fhtom(vp, v3);
823 if (v3) {
824 nfsm_v3attrbuild(vap, TRUE);
825 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
826 *tl = nfs_false;
827 } else {
828 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
829 if (vap->va_mode == (mode_t)VNOVAL)
830 sp->sa_mode = nfs_xdrneg1;
831 else
832 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
833 if (vap->va_uid == (uid_t)VNOVAL)
834 sp->sa_uid = nfs_xdrneg1;
835 else
836 sp->sa_uid = txdr_unsigned(vap->va_uid);
837 if (vap->va_gid == (gid_t)VNOVAL)
838 sp->sa_gid = nfs_xdrneg1;
839 else
840 sp->sa_gid = txdr_unsigned(vap->va_gid);
841 sp->sa_size = txdr_unsigned(vap->va_size);
842 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
843 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
844 }
845 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
846 if (v3) {
847 np->n_modestamp = 0;
848 nfsm_wcc_data(vp, wccflag);
849 } else
850 nfsm_loadattr(vp, NULL);
851 m_freem(mrep);
852 nfsmout:
853 return (error);
854 }
855
856 /*
857 * nfs lookup call, one step at a time...
858 * First look in cache
859 * If not found, unlock the directory nfsnode and do the rpc
860 */
861 static int
862 nfs_lookup(struct vop_lookup_args *ap)
863 {
864 struct componentname *cnp = ap->a_cnp;
865 struct vnode *dvp = ap->a_dvp;
866 struct vnode **vpp = ap->a_vpp;
867 struct mount *mp = dvp->v_mount;
868 struct vattr dvattr, vattr;
869 struct timespec nctime;
870 int flags = cnp->cn_flags;
871 struct vnode *newvp;
872 struct nfsmount *nmp;
873 caddr_t bpos, dpos;
874 struct mbuf *mreq, *mrep, *md, *mb;
875 long len;
876 nfsfh_t *fhp;
877 struct nfsnode *np, *newnp;
878 int error = 0, attrflag, dattrflag, fhsize, ltype, ncticks;
879 int v3 = NFS_ISV3(dvp);
880 struct thread *td = cnp->cn_thread;
881
882 *vpp = NULLVP;
883 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
884 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
885 return (EROFS);
886 if (dvp->v_type != VDIR)
887 return (ENOTDIR);
888 nmp = VFSTONFS(mp);
889 np = VTONFS(dvp);
890 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
891 *vpp = NULLVP;
892 return (error);
893 }
894 error = cache_lookup_times(dvp, vpp, cnp, &nctime, &ncticks);
895 if (error > 0 && error != ENOENT)
896 return (error);
897 if (error == -1) {
898 /*
899 * Lookups of "." are special and always return the
900 * current directory. cache_lookup() already handles
901 * associated locking bookkeeping, etc.
902 */
903 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
904 /* XXX: Is this really correct? */
905 if (cnp->cn_nameiop != LOOKUP &&
906 (flags & ISLASTCN))
907 cnp->cn_flags |= SAVENAME;
908 return (0);
909 }
910
911 /*
912 * We only accept a positive hit in the cache if the
913 * change time of the file matches our cached copy.
914 * Otherwise, we discard the cache entry and fallback
915 * to doing a lookup RPC. We also only trust cache
916 * entries for less than nametimeo seconds.
917 *
918 * To better handle stale file handles and attributes,
919 * clear the attribute cache of this node if it is a
920 * leaf component, part of an open() call, and not
921 * locally modified before fetching the attributes.
922 * This should allow stale file handles to be detected
923 * here where we can fall back to a LOOKUP RPC to
924 * recover rather than having nfs_open() detect the
925 * stale file handle and failing open(2) with ESTALE.
926 */
927 newvp = *vpp;
928 newnp = VTONFS(newvp);
929 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
930 !(newnp->n_flag & NMODIFIED)) {
931 mtx_lock(&newnp->n_mtx);
932 newnp->n_attrstamp = 0;
933 mtx_unlock(&newnp->n_mtx);
934 }
935 if ((u_int)(ticks - ncticks) < (nametimeo * hz) &&
936 VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td) == 0 &&
937 timespeccmp(&vattr.va_ctime, &nctime, ==)) {
938 nfsstats.lookupcache_hits++;
939 if (cnp->cn_nameiop != LOOKUP &&
940 (flags & ISLASTCN))
941 cnp->cn_flags |= SAVENAME;
942 return (0);
943 }
944 cache_purge(newvp);
945 if (dvp != newvp)
946 vput(newvp);
947 else
948 vrele(newvp);
949 *vpp = NULLVP;
950 } else if (error == ENOENT) {
951 /*
952 * We only accept a negative hit in the cache if the
953 * modification time of the parent directory matches
954 * the cached copy in the name cache entry.
955 * Otherwise, we discard all of the negative cache
956 * entries for this directory. We also only trust
957 * negative cache entries for up to negnametimeo
958 * seconds.
959 */
960 if ((u_int)(ticks - ncticks) < (negnametimeo * hz) &&
961 VOP_GETATTR(dvp, &vattr, cnp->cn_cred, td) == 0 &&
962 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
963 nfsstats.lookupcache_hits++;
964 return (ENOENT);
965 }
966 cache_purge_negative(dvp);
967 }
968
969 attrflag = dattrflag = 0;
970 error = 0;
971 newvp = NULLVP;
972 nfsstats.lookupcache_misses++;
973 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
974 len = cnp->cn_namelen;
975 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
976 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
977 mb = mreq;
978 bpos = mtod(mb, caddr_t);
979 nfsm_fhtom(dvp, v3);
980 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
981 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
982 if (error) {
983 if (v3) {
984 nfsm_postop_attr_va(dvp, dattrflag, &vattr);
985 m_freem(mrep);
986 }
987 goto nfsmout;
988 }
989 nfsm_getfh(fhp, fhsize, v3);
990
991 /*
992 * Handle RENAME case...
993 */
994 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
995 if (NFS_CMPFH(np, fhp, fhsize)) {
996 m_freem(mrep);
997 return (EISDIR);
998 }
999 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE);
1000 if (error) {
1001 m_freem(mrep);
1002 return (error);
1003 }
1004 newvp = NFSTOV(np);
1005 if (v3) {
1006 nfsm_postop_attr(newvp, attrflag);
1007 nfsm_postop_attr(dvp, attrflag);
1008 } else
1009 nfsm_loadattr(newvp, NULL);
1010 *vpp = newvp;
1011 m_freem(mrep);
1012 cnp->cn_flags |= SAVENAME;
1013 return (0);
1014 }
1015
1016 if (flags & ISDOTDOT) {
1017 ltype = VOP_ISLOCKED(dvp, td);
1018 error = vfs_busy(mp, LK_NOWAIT, NULL, td);
1019 if (error != 0) {
1020 VOP_UNLOCK(dvp, 0, td);
1021 error = vfs_busy(mp, 0, NULL, td);
1022 vn_lock(dvp, ltype | LK_RETRY, td);
1023 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1024 vfs_unbusy(mp, td);
1025 error = ENOENT;
1026 }
1027 if (error != 0) {
1028 m_freem(mrep);
1029 return (error);
1030 }
1031 }
1032 VOP_UNLOCK(dvp, 0, td);
1033 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1034 if (error == 0)
1035 newvp = NFSTOV(np);
1036 vfs_unbusy(mp, td);
1037 vn_lock(dvp, ltype | LK_RETRY, td);
1038 if (dvp->v_iflag & VI_DOOMED) {
1039 if (error == 0) {
1040 if (newvp == dvp)
1041 vrele(newvp);
1042 else
1043 vput(newvp);
1044 }
1045 error = ENOENT;
1046 }
1047 if (error) {
1048 m_freem(mrep);
1049 return (error);
1050 }
1051 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1052 VREF(dvp);
1053 newvp = dvp;
1054 } else {
1055 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1056 if (error) {
1057 m_freem(mrep);
1058 return (error);
1059 }
1060 newvp = NFSTOV(np);
1061
1062 /*
1063 * Flush the attribute cache when opening a leaf node
1064 * to ensure that fresh attributes are fetched in
1065 * nfs_open() if we are unable to fetch attributes
1066 * from the LOOKUP reply.
1067 */
1068 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1069 !(np->n_flag & NMODIFIED)) {
1070 mtx_lock(&np->n_mtx);
1071 np->n_attrstamp = 0;
1072 mtx_unlock(&np->n_mtx);
1073 }
1074 }
1075 if (v3) {
1076 nfsm_postop_attr_va(newvp, attrflag, &vattr);
1077 nfsm_postop_attr_va(dvp, dattrflag, &dvattr);
1078 } else {
1079 nfsm_loadattr(newvp, &vattr);
1080 attrflag = 1;
1081 }
1082 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1083 cnp->cn_flags |= SAVENAME;
1084 if ((cnp->cn_flags & MAKEENTRY) &&
1085 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1086 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1087 cache_enter_time(dvp, newvp, cnp, &vattr.va_ctime,
1088 newvp->v_type != VDIR ? NULL : &dvattr.va_ctime);
1089 *vpp = newvp;
1090 m_freem(mrep);
1091 nfsmout:
1092 if (error) {
1093 if (newvp != NULLVP) {
1094 vput(newvp);
1095 *vpp = NULLVP;
1096 }
1097
1098 if (error != ENOENT)
1099 goto done;
1100
1101 /* The requested file was not found. */
1102 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1103 (flags & ISLASTCN)) {
1104 /*
1105 * XXX: UFS does a full VOP_ACCESS(dvp,
1106 * VWRITE) here instead of just checking
1107 * MNT_RDONLY.
1108 */
1109 if (mp->mnt_flag & MNT_RDONLY)
1110 return (EROFS);
1111 cnp->cn_flags |= SAVENAME;
1112 return (EJUSTRETURN);
1113 }
1114
1115 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
1116 dattrflag) {
1117 /*
1118 * Cache the modification time of the parent
1119 * directory from the post-op attributes in
1120 * the name cache entry. The negative cache
1121 * entry will be ignored once the directory
1122 * has changed. Don't bother adding the entry
1123 * if the directory has already changed.
1124 */
1125 mtx_lock(&np->n_mtx);
1126 if (timespeccmp(&np->n_vattr.va_mtime,
1127 &vattr.va_mtime, ==)) {
1128 mtx_unlock(&np->n_mtx);
1129 cache_enter_time(dvp, NULL, cnp,
1130 &vattr.va_mtime, NULL);
1131 } else
1132 mtx_unlock(&np->n_mtx);
1133 }
1134 return (ENOENT);
1135 }
1136 done:
1137 return (error);
1138 }
1139
1140 /*
1141 * nfs read call.
1142 * Just call nfs_bioread() to do the work.
1143 */
1144 static int
1145 nfs_read(struct vop_read_args *ap)
1146 {
1147 struct vnode *vp = ap->a_vp;
1148
1149 switch (vp->v_type) {
1150 case VREG:
1151 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1152 case VDIR:
1153 return (EISDIR);
1154 default:
1155 return (EOPNOTSUPP);
1156 }
1157 }
1158
1159 /*
1160 * nfs readlink call
1161 */
1162 static int
1163 nfs_readlink(struct vop_readlink_args *ap)
1164 {
1165 struct vnode *vp = ap->a_vp;
1166
1167 if (vp->v_type != VLNK)
1168 return (EINVAL);
1169 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1170 }
1171
1172 /*
1173 * Do a readlink rpc.
1174 * Called by nfs_doio() from below the buffer cache.
1175 */
1176 int
1177 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1178 {
1179 caddr_t bpos, dpos;
1180 int error = 0, len, attrflag;
1181 struct mbuf *mreq, *mrep, *md, *mb;
1182 int v3 = NFS_ISV3(vp);
1183
1184 nfsstats.rpccnt[NFSPROC_READLINK]++;
1185 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1186 mb = mreq;
1187 bpos = mtod(mb, caddr_t);
1188 nfsm_fhtom(vp, v3);
1189 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1190 if (v3)
1191 nfsm_postop_attr(vp, attrflag);
1192 if (!error) {
1193 nfsm_strsiz(len, NFS_MAXPATHLEN);
1194 if (len == NFS_MAXPATHLEN) {
1195 struct nfsnode *np = VTONFS(vp);
1196 mtx_lock(&np->n_mtx);
1197 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1198 len = np->n_size;
1199 mtx_unlock(&np->n_mtx);
1200 }
1201 nfsm_mtouio(uiop, len);
1202 }
1203 m_freem(mrep);
1204 nfsmout:
1205 return (error);
1206 }
1207
1208 /*
1209 * nfs read rpc call
1210 * Ditto above
1211 */
1212 int
1213 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1214 {
1215 u_int32_t *tl;
1216 caddr_t bpos, dpos;
1217 struct mbuf *mreq, *mrep, *md, *mb;
1218 struct nfsmount *nmp;
1219 int error = 0, len, retlen, tsiz, eof, attrflag;
1220 int v3 = NFS_ISV3(vp);
1221 int rsize;
1222
1223 #ifndef nolint
1224 eof = 0;
1225 #endif
1226 nmp = VFSTONFS(vp->v_mount);
1227 tsiz = uiop->uio_resid;
1228 mtx_lock(&nmp->nm_mtx);
1229 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1230 mtx_unlock(&nmp->nm_mtx);
1231 return (EFBIG);
1232 }
1233 rsize = nmp->nm_rsize;
1234 mtx_unlock(&nmp->nm_mtx);
1235 while (tsiz > 0) {
1236 nfsstats.rpccnt[NFSPROC_READ]++;
1237 len = (tsiz > rsize) ? rsize : tsiz;
1238 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1239 mb = mreq;
1240 bpos = mtod(mb, caddr_t);
1241 nfsm_fhtom(vp, v3);
1242 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1243 if (v3) {
1244 txdr_hyper(uiop->uio_offset, tl);
1245 *(tl + 2) = txdr_unsigned(len);
1246 } else {
1247 *tl++ = txdr_unsigned(uiop->uio_offset);
1248 *tl++ = txdr_unsigned(len);
1249 *tl = 0;
1250 }
1251 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1252 if (v3) {
1253 nfsm_postop_attr(vp, attrflag);
1254 if (error) {
1255 m_freem(mrep);
1256 goto nfsmout;
1257 }
1258 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1259 eof = fxdr_unsigned(int, *(tl + 1));
1260 } else {
1261 nfsm_loadattr(vp, NULL);
1262 }
1263 nfsm_strsiz(retlen, rsize);
1264 nfsm_mtouio(uiop, retlen);
1265 m_freem(mrep);
1266 tsiz -= retlen;
1267 if (v3) {
1268 if (eof || retlen == 0) {
1269 tsiz = 0;
1270 }
1271 } else if (retlen < len) {
1272 tsiz = 0;
1273 }
1274 }
1275 nfsmout:
1276 return (error);
1277 }
1278
1279 /*
1280 * nfs write call
1281 */
1282 int
1283 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1284 int *iomode, int *must_commit)
1285 {
1286 u_int32_t *tl;
1287 int32_t backup;
1288 caddr_t bpos, dpos;
1289 struct mbuf *mreq, *mrep, *md, *mb;
1290 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1291 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1292 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1293 int wsize;
1294
1295 #ifndef DIAGNOSTIC
1296 if (uiop->uio_iovcnt != 1)
1297 panic("nfs: writerpc iovcnt > 1");
1298 #endif
1299 *must_commit = 0;
1300 tsiz = uiop->uio_resid;
1301 mtx_lock(&nmp->nm_mtx);
1302 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1303 mtx_unlock(&nmp->nm_mtx);
1304 return (EFBIG);
1305 }
1306 wsize = nmp->nm_wsize;
1307 mtx_unlock(&nmp->nm_mtx);
1308 while (tsiz > 0) {
1309 nfsstats.rpccnt[NFSPROC_WRITE]++;
1310 len = (tsiz > wsize) ? wsize : tsiz;
1311 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1312 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1313 mb = mreq;
1314 bpos = mtod(mb, caddr_t);
1315 nfsm_fhtom(vp, v3);
1316 if (v3) {
1317 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1318 txdr_hyper(uiop->uio_offset, tl);
1319 tl += 2;
1320 *tl++ = txdr_unsigned(len);
1321 *tl++ = txdr_unsigned(*iomode);
1322 *tl = txdr_unsigned(len);
1323 } else {
1324 u_int32_t x;
1325
1326 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1327 /* Set both "begin" and "current" to non-garbage. */
1328 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1329 *tl++ = x; /* "begin offset" */
1330 *tl++ = x; /* "current offset" */
1331 x = txdr_unsigned(len);
1332 *tl++ = x; /* total to this offset */
1333 *tl = x; /* size of this write */
1334 }
1335 nfsm_uiotom(uiop, len);
1336 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1337 if (v3) {
1338 wccflag = NFSV3_WCCCHK;
1339 nfsm_wcc_data(vp, wccflag);
1340 if (!error) {
1341 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1342 + NFSX_V3WRITEVERF);
1343 rlen = fxdr_unsigned(int, *tl++);
1344 if (rlen == 0) {
1345 error = NFSERR_IO;
1346 m_freem(mrep);
1347 break;
1348 } else if (rlen < len) {
1349 backup = len - rlen;
1350 uiop->uio_iov->iov_base =
1351 (char *)uiop->uio_iov->iov_base -
1352 backup;
1353 uiop->uio_iov->iov_len += backup;
1354 uiop->uio_offset -= backup;
1355 uiop->uio_resid += backup;
1356 len = rlen;
1357 }
1358 commit = fxdr_unsigned(int, *tl++);
1359
1360 /*
1361 * Return the lowest committment level
1362 * obtained by any of the RPCs.
1363 */
1364 if (committed == NFSV3WRITE_FILESYNC)
1365 committed = commit;
1366 else if (committed == NFSV3WRITE_DATASYNC &&
1367 commit == NFSV3WRITE_UNSTABLE)
1368 committed = commit;
1369 mtx_lock(&nmp->nm_mtx);
1370 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1371 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1372 NFSX_V3WRITEVERF);
1373 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1374 } else if (bcmp((caddr_t)tl,
1375 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1376 *must_commit = 1;
1377 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1378 NFSX_V3WRITEVERF);
1379 }
1380 mtx_unlock(&nmp->nm_mtx);
1381 }
1382 } else {
1383 nfsm_loadattr(vp, NULL);
1384 }
1385 if (wccflag) {
1386 mtx_lock(&(VTONFS(vp))->n_mtx);
1387 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1388 mtx_unlock(&(VTONFS(vp))->n_mtx);
1389 }
1390 m_freem(mrep);
1391 if (error)
1392 break;
1393 tsiz -= len;
1394 }
1395 nfsmout:
1396 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1397 committed = NFSV3WRITE_FILESYNC;
1398 *iomode = committed;
1399 if (error)
1400 uiop->uio_resid = tsiz;
1401 return (error);
1402 }
1403
1404 /*
1405 * nfs mknod rpc
1406 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1407 * mode set to specify the file type and the size field for rdev.
1408 */
1409 static int
1410 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1411 struct vattr *vap)
1412 {
1413 struct nfsv2_sattr *sp;
1414 u_int32_t *tl;
1415 struct vnode *newvp = NULL;
1416 struct nfsnode *np = NULL;
1417 struct vattr vattr;
1418 caddr_t bpos, dpos;
1419 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1420 struct mbuf *mreq, *mrep, *md, *mb;
1421 u_int32_t rdev;
1422 int v3 = NFS_ISV3(dvp);
1423
1424 if (vap->va_type == VCHR || vap->va_type == VBLK)
1425 rdev = txdr_unsigned(vap->va_rdev);
1426 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1427 rdev = nfs_xdrneg1;
1428 else {
1429 return (EOPNOTSUPP);
1430 }
1431 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1432 return (error);
1433 }
1434 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1435 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1436 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1437 mb = mreq;
1438 bpos = mtod(mb, caddr_t);
1439 nfsm_fhtom(dvp, v3);
1440 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1441 if (v3) {
1442 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1443 *tl++ = vtonfsv3_type(vap->va_type);
1444 nfsm_v3attrbuild(vap, FALSE);
1445 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1446 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1447 *tl++ = txdr_unsigned(umajor(vap->va_rdev));
1448 *tl = txdr_unsigned(uminor(vap->va_rdev));
1449 }
1450 } else {
1451 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1452 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1453 sp->sa_uid = nfs_xdrneg1;
1454 sp->sa_gid = nfs_xdrneg1;
1455 sp->sa_size = rdev;
1456 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1457 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1458 }
1459 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1460 if (!error) {
1461 nfsm_mtofh(dvp, newvp, v3, gotvp);
1462 if (!gotvp) {
1463 if (newvp) {
1464 vput(newvp);
1465 newvp = NULL;
1466 }
1467 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1468 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1469 if (!error)
1470 newvp = NFSTOV(np);
1471 }
1472 }
1473 if (v3)
1474 nfsm_wcc_data(dvp, wccflag);
1475 m_freem(mrep);
1476 nfsmout:
1477 if (error) {
1478 if (newvp)
1479 vput(newvp);
1480 } else {
1481 *vpp = newvp;
1482 }
1483 mtx_lock(&(VTONFS(dvp))->n_mtx);
1484 VTONFS(dvp)->n_flag |= NMODIFIED;
1485 if (!wccflag)
1486 VTONFS(dvp)->n_attrstamp = 0;
1487 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1488 return (error);
1489 }
1490
1491 /*
1492 * nfs mknod vop
1493 * just call nfs_mknodrpc() to do the work.
1494 */
1495 /* ARGSUSED */
1496 static int
1497 nfs_mknod(struct vop_mknod_args *ap)
1498 {
1499 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1500 }
1501
1502 static u_long create_verf;
1503 /*
1504 * nfs file create call
1505 */
1506 static int
1507 nfs_create(struct vop_create_args *ap)
1508 {
1509 struct vnode *dvp = ap->a_dvp;
1510 struct vattr *vap = ap->a_vap;
1511 struct componentname *cnp = ap->a_cnp;
1512 struct nfsv2_sattr *sp;
1513 u_int32_t *tl;
1514 struct nfsnode *np = NULL;
1515 struct vnode *newvp = NULL;
1516 caddr_t bpos, dpos;
1517 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1518 struct mbuf *mreq, *mrep, *md, *mb;
1519 struct vattr vattr;
1520 int v3 = NFS_ISV3(dvp);
1521
1522 /*
1523 * Oops, not for me..
1524 */
1525 if (vap->va_type == VSOCK)
1526 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1527
1528 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1529 return (error);
1530 }
1531 if (vap->va_vaflags & VA_EXCLUSIVE)
1532 fmode |= O_EXCL;
1533 again:
1534 nfsstats.rpccnt[NFSPROC_CREATE]++;
1535 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1536 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1537 mb = mreq;
1538 bpos = mtod(mb, caddr_t);
1539 nfsm_fhtom(dvp, v3);
1540 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1541 if (v3) {
1542 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1543 if (fmode & O_EXCL) {
1544 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1545 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1546 #ifdef INET
1547 if (!TAILQ_EMPTY(&in_ifaddrhead))
1548 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1549 else
1550 #endif
1551 *tl++ = create_verf;
1552 *tl = ++create_verf;
1553 } else {
1554 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1555 nfsm_v3attrbuild(vap, FALSE);
1556 }
1557 } else {
1558 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1559 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1560 sp->sa_uid = nfs_xdrneg1;
1561 sp->sa_gid = nfs_xdrneg1;
1562 sp->sa_size = 0;
1563 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1564 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1565 }
1566 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1567 if (!error) {
1568 nfsm_mtofh(dvp, newvp, v3, gotvp);
1569 if (!gotvp) {
1570 if (newvp) {
1571 vput(newvp);
1572 newvp = NULL;
1573 }
1574 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1575 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1576 if (!error)
1577 newvp = NFSTOV(np);
1578 }
1579 }
1580 if (v3)
1581 nfsm_wcc_data(dvp, wccflag);
1582 m_freem(mrep);
1583 nfsmout:
1584 if (error) {
1585 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1586 fmode &= ~O_EXCL;
1587 goto again;
1588 }
1589 if (newvp)
1590 vput(newvp);
1591 } else if (v3 && (fmode & O_EXCL)) {
1592 /*
1593 * We are normally called with only a partially initialized
1594 * VAP. Since the NFSv3 spec says that server may use the
1595 * file attributes to store the verifier, the spec requires
1596 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1597 * in atime, but we can't really assume that all servers will
1598 * so we ensure that our SETATTR sets both atime and mtime.
1599 */
1600 if (vap->va_mtime.tv_sec == VNOVAL)
1601 vfs_timestamp(&vap->va_mtime);
1602 if (vap->va_atime.tv_sec == VNOVAL)
1603 vap->va_atime = vap->va_mtime;
1604 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
1605 if (error)
1606 vput(newvp);
1607 }
1608 if (!error) {
1609 *ap->a_vpp = newvp;
1610 }
1611 mtx_lock(&(VTONFS(dvp))->n_mtx);
1612 VTONFS(dvp)->n_flag |= NMODIFIED;
1613 if (!wccflag)
1614 VTONFS(dvp)->n_attrstamp = 0;
1615 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1616 return (error);
1617 }
1618
1619 /*
1620 * nfs file remove call
1621 * To try and make nfs semantics closer to ufs semantics, a file that has
1622 * other processes using the vnode is renamed instead of removed and then
1623 * removed later on the last close.
1624 * - If v_usecount > 1
1625 * If a rename is not already in the works
1626 * call nfs_sillyrename() to set it up
1627 * else
1628 * do the remove rpc
1629 */
1630 static int
1631 nfs_remove(struct vop_remove_args *ap)
1632 {
1633 struct vnode *vp = ap->a_vp;
1634 struct vnode *dvp = ap->a_dvp;
1635 struct componentname *cnp = ap->a_cnp;
1636 struct nfsnode *np = VTONFS(vp);
1637 int error = 0;
1638 struct vattr vattr;
1639
1640 #ifndef DIAGNOSTIC
1641 if ((cnp->cn_flags & HASBUF) == 0)
1642 panic("nfs_remove: no name");
1643 if (vrefcnt(vp) < 1)
1644 panic("nfs_remove: bad v_usecount");
1645 #endif
1646 if (vp->v_type == VDIR)
1647 error = EPERM;
1648 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1649 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
1650 vattr.va_nlink > 1)) {
1651 /*
1652 * Purge the name cache so that the chance of a lookup for
1653 * the name succeeding while the remove is in progress is
1654 * minimized. Without node locking it can still happen, such
1655 * that an I/O op returns ESTALE, but since you get this if
1656 * another host removes the file..
1657 */
1658 cache_purge(vp);
1659 /*
1660 * throw away biocache buffers, mainly to avoid
1661 * unnecessary delayed writes later.
1662 */
1663 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1664 /* Do the rpc */
1665 if (error != EINTR && error != EIO)
1666 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1667 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1668 /*
1669 * Kludge City: If the first reply to the remove rpc is lost..
1670 * the reply to the retransmitted request will be ENOENT
1671 * since the file was in fact removed
1672 * Therefore, we cheat and return success.
1673 */
1674 if (error == ENOENT)
1675 error = 0;
1676 } else if (!np->n_sillyrename)
1677 error = nfs_sillyrename(dvp, vp, cnp);
1678 np->n_attrstamp = 0;
1679 return (error);
1680 }
1681
1682 /*
1683 * nfs file remove rpc called from nfs_inactive
1684 */
1685 int
1686 nfs_removeit(struct sillyrename *sp)
1687 {
1688 /*
1689 * Make sure that the directory vnode is still valid.
1690 * XXX we should lock sp->s_dvp here.
1691 */
1692 if (sp->s_dvp->v_type == VBAD)
1693 return (0);
1694 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1695 NULL));
1696 }
1697
1698 /*
1699 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1700 */
1701 static int
1702 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1703 struct ucred *cred, struct thread *td)
1704 {
1705 caddr_t bpos, dpos;
1706 int error = 0, wccflag = NFSV3_WCCRATTR;
1707 struct mbuf *mreq, *mrep, *md, *mb;
1708 int v3 = NFS_ISV3(dvp);
1709
1710 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1711 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1712 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1713 mb = mreq;
1714 bpos = mtod(mb, caddr_t);
1715 nfsm_fhtom(dvp, v3);
1716 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1717 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1718 if (v3)
1719 nfsm_wcc_data(dvp, wccflag);
1720 m_freem(mrep);
1721 nfsmout:
1722 mtx_lock(&(VTONFS(dvp))->n_mtx);
1723 VTONFS(dvp)->n_flag |= NMODIFIED;
1724 if (!wccflag)
1725 VTONFS(dvp)->n_attrstamp = 0;
1726 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1727 return (error);
1728 }
1729
1730 /*
1731 * nfs file rename call
1732 */
1733 static int
1734 nfs_rename(struct vop_rename_args *ap)
1735 {
1736 struct vnode *fvp = ap->a_fvp;
1737 struct vnode *tvp = ap->a_tvp;
1738 struct vnode *fdvp = ap->a_fdvp;
1739 struct vnode *tdvp = ap->a_tdvp;
1740 struct componentname *tcnp = ap->a_tcnp;
1741 struct componentname *fcnp = ap->a_fcnp;
1742 int error;
1743
1744 #ifndef DIAGNOSTIC
1745 if ((tcnp->cn_flags & HASBUF) == 0 ||
1746 (fcnp->cn_flags & HASBUF) == 0)
1747 panic("nfs_rename: no name");
1748 #endif
1749 /* Check for cross-device rename */
1750 if ((fvp->v_mount != tdvp->v_mount) ||
1751 (tvp && (fvp->v_mount != tvp->v_mount))) {
1752 error = EXDEV;
1753 goto out;
1754 }
1755
1756 if (fvp == tvp) {
1757 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1758 error = 0;
1759 goto out;
1760 }
1761 if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0)
1762 goto out;
1763
1764 /*
1765 * We have to flush B_DELWRI data prior to renaming
1766 * the file. If we don't, the delayed-write buffers
1767 * can be flushed out later after the file has gone stale
1768 * under NFSV3. NFSV2 does not have this problem because
1769 * ( as far as I can tell ) it flushes dirty buffers more
1770 * often.
1771 *
1772 * Skip the rename operation if the fsync fails, this can happen
1773 * due to the server's volume being full, when we pushed out data
1774 * that was written back to our cache earlier. Not checking for
1775 * this condition can result in potential (silent) data loss.
1776 */
1777 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1778 VOP_UNLOCK(fvp, 0, fcnp->cn_thread);
1779 if (!error && tvp)
1780 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1781 if (error)
1782 goto out;
1783
1784 /*
1785 * If the tvp exists and is in use, sillyrename it before doing the
1786 * rename of the new file over it.
1787 * XXX Can't sillyrename a directory.
1788 */
1789 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1790 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1791 vput(tvp);
1792 tvp = NULL;
1793 }
1794
1795 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1796 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1797 tcnp->cn_thread);
1798
1799 if (fvp->v_type == VDIR) {
1800 if (tvp != NULL && tvp->v_type == VDIR)
1801 cache_purge(tdvp);
1802 cache_purge(fdvp);
1803 }
1804
1805 out:
1806 if (tdvp == tvp)
1807 vrele(tdvp);
1808 else
1809 vput(tdvp);
1810 if (tvp)
1811 vput(tvp);
1812 vrele(fdvp);
1813 vrele(fvp);
1814 /*
1815 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1816 */
1817 if (error == ENOENT)
1818 error = 0;
1819 return (error);
1820 }
1821
1822 /*
1823 * nfs file rename rpc called from nfs_remove() above
1824 */
1825 static int
1826 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1827 struct sillyrename *sp)
1828 {
1829
1830 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1831 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1832 }
1833
1834 /*
1835 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1836 */
1837 static int
1838 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1839 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1840 struct thread *td)
1841 {
1842 caddr_t bpos, dpos;
1843 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1844 struct mbuf *mreq, *mrep, *md, *mb;
1845 int v3 = NFS_ISV3(fdvp);
1846
1847 nfsstats.rpccnt[NFSPROC_RENAME]++;
1848 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1849 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1850 nfsm_rndup(tnamelen));
1851 mb = mreq;
1852 bpos = mtod(mb, caddr_t);
1853 nfsm_fhtom(fdvp, v3);
1854 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1855 nfsm_fhtom(tdvp, v3);
1856 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1857 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1858 if (v3) {
1859 nfsm_wcc_data(fdvp, fwccflag);
1860 nfsm_wcc_data(tdvp, twccflag);
1861 }
1862 m_freem(mrep);
1863 nfsmout:
1864 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1865 VTONFS(fdvp)->n_flag |= NMODIFIED;
1866 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1867 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1868 VTONFS(tdvp)->n_flag |= NMODIFIED;
1869 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1870 if (!fwccflag)
1871 VTONFS(fdvp)->n_attrstamp = 0;
1872 if (!twccflag)
1873 VTONFS(tdvp)->n_attrstamp = 0;
1874 return (error);
1875 }
1876
1877 /*
1878 * nfs hard link create call
1879 */
1880 static int
1881 nfs_link(struct vop_link_args *ap)
1882 {
1883 struct vnode *vp = ap->a_vp;
1884 struct vnode *tdvp = ap->a_tdvp;
1885 struct componentname *cnp = ap->a_cnp;
1886 caddr_t bpos, dpos;
1887 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1888 struct mbuf *mreq, *mrep, *md, *mb;
1889 int v3;
1890
1891 if (vp->v_mount != tdvp->v_mount) {
1892 return (EXDEV);
1893 }
1894
1895 /*
1896 * Push all writes to the server, so that the attribute cache
1897 * doesn't get "out of sync" with the server.
1898 * XXX There should be a better way!
1899 */
1900 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1901
1902 v3 = NFS_ISV3(vp);
1903 nfsstats.rpccnt[NFSPROC_LINK]++;
1904 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1905 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1906 mb = mreq;
1907 bpos = mtod(mb, caddr_t);
1908 nfsm_fhtom(vp, v3);
1909 nfsm_fhtom(tdvp, v3);
1910 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1911 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1912 if (v3) {
1913 nfsm_postop_attr(vp, attrflag);
1914 nfsm_wcc_data(tdvp, wccflag);
1915 }
1916 m_freem(mrep);
1917 nfsmout:
1918 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1919 VTONFS(tdvp)->n_flag |= NMODIFIED;
1920 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1921 if (!attrflag)
1922 VTONFS(vp)->n_attrstamp = 0;
1923 if (!wccflag)
1924 VTONFS(tdvp)->n_attrstamp = 0;
1925 return (error);
1926 }
1927
1928 /*
1929 * nfs symbolic link create call
1930 */
1931 static int
1932 nfs_symlink(struct vop_symlink_args *ap)
1933 {
1934 struct vnode *dvp = ap->a_dvp;
1935 struct vattr *vap = ap->a_vap;
1936 struct componentname *cnp = ap->a_cnp;
1937 struct nfsv2_sattr *sp;
1938 caddr_t bpos, dpos;
1939 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1940 struct mbuf *mreq, *mrep, *md, *mb;
1941 struct vnode *newvp = NULL;
1942 int v3 = NFS_ISV3(dvp);
1943
1944 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1945 slen = strlen(ap->a_target);
1946 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1947 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1948 mb = mreq;
1949 bpos = mtod(mb, caddr_t);
1950 nfsm_fhtom(dvp, v3);
1951 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1952 if (v3) {
1953 nfsm_v3attrbuild(vap, FALSE);
1954 }
1955 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1956 if (!v3) {
1957 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1958 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1959 sp->sa_uid = nfs_xdrneg1;
1960 sp->sa_gid = nfs_xdrneg1;
1961 sp->sa_size = nfs_xdrneg1;
1962 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1963 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1964 }
1965
1966 /*
1967 * Issue the NFS request and get the rpc response.
1968 *
1969 * Only NFSv3 responses returning an error of 0 actually return
1970 * a file handle that can be converted into newvp without having
1971 * to do an extra lookup rpc.
1972 */
1973 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
1974 if (v3) {
1975 if (error == 0)
1976 nfsm_mtofh(dvp, newvp, v3, gotvp);
1977 nfsm_wcc_data(dvp, wccflag);
1978 }
1979
1980 /*
1981 * out code jumps -> here, mrep is also freed.
1982 */
1983
1984 m_freem(mrep);
1985 nfsmout:
1986
1987 /*
1988 * If we do not have an error and we could not extract the newvp from
1989 * the response due to the request being NFSv2, we have to do a
1990 * lookup in order to obtain a newvp to return.
1991 */
1992 if (error == 0 && newvp == NULL) {
1993 struct nfsnode *np = NULL;
1994
1995 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1996 cnp->cn_cred, cnp->cn_thread, &np);
1997 if (!error)
1998 newvp = NFSTOV(np);
1999 }
2000 if (error) {
2001 if (newvp)
2002 vput(newvp);
2003 } else {
2004 *ap->a_vpp = newvp;
2005 }
2006 mtx_lock(&(VTONFS(dvp))->n_mtx);
2007 VTONFS(dvp)->n_flag |= NMODIFIED;
2008 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2009 if (!wccflag)
2010 VTONFS(dvp)->n_attrstamp = 0;
2011 return (error);
2012 }
2013
2014 /*
2015 * nfs make dir call
2016 */
2017 static int
2018 nfs_mkdir(struct vop_mkdir_args *ap)
2019 {
2020 struct vnode *dvp = ap->a_dvp;
2021 struct vattr *vap = ap->a_vap;
2022 struct componentname *cnp = ap->a_cnp;
2023 struct nfsv2_sattr *sp;
2024 int len;
2025 struct nfsnode *np = NULL;
2026 struct vnode *newvp = NULL;
2027 caddr_t bpos, dpos;
2028 int error = 0, wccflag = NFSV3_WCCRATTR;
2029 int gotvp = 0;
2030 struct mbuf *mreq, *mrep, *md, *mb;
2031 struct vattr vattr;
2032 int v3 = NFS_ISV3(dvp);
2033
2034 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
2035 return (error);
2036 }
2037 len = cnp->cn_namelen;
2038 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2039 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
2040 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2041 mb = mreq;
2042 bpos = mtod(mb, caddr_t);
2043 nfsm_fhtom(dvp, v3);
2044 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2045 if (v3) {
2046 nfsm_v3attrbuild(vap, FALSE);
2047 } else {
2048 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2049 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2050 sp->sa_uid = nfs_xdrneg1;
2051 sp->sa_gid = nfs_xdrneg1;
2052 sp->sa_size = nfs_xdrneg1;
2053 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2054 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2055 }
2056 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2057 if (!error)
2058 nfsm_mtofh(dvp, newvp, v3, gotvp);
2059 if (v3)
2060 nfsm_wcc_data(dvp, wccflag);
2061 m_freem(mrep);
2062 nfsmout:
2063 mtx_lock(&(VTONFS(dvp))->n_mtx);
2064 VTONFS(dvp)->n_flag |= NMODIFIED;
2065 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2066 if (!wccflag)
2067 VTONFS(dvp)->n_attrstamp = 0;
2068 if (error == 0 && newvp == NULL) {
2069 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2070 cnp->cn_thread, &np);
2071 if (!error) {
2072 newvp = NFSTOV(np);
2073 if (newvp->v_type != VDIR)
2074 error = EEXIST;
2075 }
2076 }
2077 if (error) {
2078 if (newvp)
2079 vput(newvp);
2080 } else
2081 *ap->a_vpp = newvp;
2082 return (error);
2083 }
2084
2085 /*
2086 * nfs remove directory call
2087 */
2088 static int
2089 nfs_rmdir(struct vop_rmdir_args *ap)
2090 {
2091 struct vnode *vp = ap->a_vp;
2092 struct vnode *dvp = ap->a_dvp;
2093 struct componentname *cnp = ap->a_cnp;
2094 caddr_t bpos, dpos;
2095 int error = 0, wccflag = NFSV3_WCCRATTR;
2096 struct mbuf *mreq, *mrep, *md, *mb;
2097 int v3 = NFS_ISV3(dvp);
2098
2099 if (dvp == vp)
2100 return (EINVAL);
2101 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2102 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
2103 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2104 mb = mreq;
2105 bpos = mtod(mb, caddr_t);
2106 nfsm_fhtom(dvp, v3);
2107 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2108 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2109 if (v3)
2110 nfsm_wcc_data(dvp, wccflag);
2111 m_freem(mrep);
2112 nfsmout:
2113 mtx_lock(&(VTONFS(dvp))->n_mtx);
2114 VTONFS(dvp)->n_flag |= NMODIFIED;
2115 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2116 if (!wccflag)
2117 VTONFS(dvp)->n_attrstamp = 0;
2118 cache_purge(dvp);
2119 cache_purge(vp);
2120 /*
2121 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2122 */
2123 if (error == ENOENT)
2124 error = 0;
2125 return (error);
2126 }
2127
2128 /*
2129 * nfs readdir call
2130 */
2131 static int
2132 nfs_readdir(struct vop_readdir_args *ap)
2133 {
2134 struct vnode *vp = ap->a_vp;
2135 struct nfsnode *np = VTONFS(vp);
2136 struct uio *uio = ap->a_uio;
2137 int tresid, error = 0;
2138 struct vattr vattr;
2139
2140 if (vp->v_type != VDIR)
2141 return(EPERM);
2142
2143 /*
2144 * First, check for hit on the EOF offset cache
2145 */
2146 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2147 (np->n_flag & NMODIFIED) == 0) {
2148 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0) {
2149 mtx_lock(&np->n_mtx);
2150 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2151 mtx_unlock(&np->n_mtx);
2152 nfsstats.direofcache_hits++;
2153 goto out;
2154 } else
2155 mtx_unlock(&np->n_mtx);
2156 }
2157 }
2158
2159 /*
2160 * Call nfs_bioread() to do the real work.
2161 */
2162 tresid = uio->uio_resid;
2163 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2164
2165 if (!error && uio->uio_resid == tresid) {
2166 nfsstats.direofcache_misses++;
2167 }
2168 out:
2169 return (error);
2170 }
2171
2172 /*
2173 * Readdir rpc call.
2174 * Called from below the buffer cache by nfs_doio().
2175 */
2176 int
2177 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2178 {
2179 int len, left;
2180 struct dirent *dp = NULL;
2181 u_int32_t *tl;
2182 caddr_t cp;
2183 nfsuint64 *cookiep;
2184 caddr_t bpos, dpos;
2185 struct mbuf *mreq, *mrep, *md, *mb;
2186 nfsuint64 cookie;
2187 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2188 struct nfsnode *dnp = VTONFS(vp);
2189 u_quad_t fileno;
2190 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2191 int attrflag;
2192 int v3 = NFS_ISV3(vp);
2193
2194 #ifndef DIAGNOSTIC
2195 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2196 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2197 panic("nfs readdirrpc bad uio");
2198 #endif
2199
2200 /*
2201 * If there is no cookie, assume directory was stale.
2202 */
2203 nfs_dircookie_lock(dnp);
2204 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2205 if (cookiep) {
2206 cookie = *cookiep;
2207 nfs_dircookie_unlock(dnp);
2208 } else {
2209 nfs_dircookie_unlock(dnp);
2210 return (NFSERR_BAD_COOKIE);
2211 }
2212
2213 /*
2214 * Loop around doing readdir rpc's of size nm_readdirsize
2215 * truncated to a multiple of DIRBLKSIZ.
2216 * The stopping criteria is EOF or buffer full.
2217 */
2218 while (more_dirs && bigenough) {
2219 nfsstats.rpccnt[NFSPROC_READDIR]++;
2220 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2221 NFSX_READDIR(v3));
2222 mb = mreq;
2223 bpos = mtod(mb, caddr_t);
2224 nfsm_fhtom(vp, v3);
2225 if (v3) {
2226 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2227 *tl++ = cookie.nfsuquad[0];
2228 *tl++ = cookie.nfsuquad[1];
2229 mtx_lock(&dnp->n_mtx);
2230 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2231 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2232 mtx_unlock(&dnp->n_mtx);
2233 } else {
2234 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2235 *tl++ = cookie.nfsuquad[0];
2236 }
2237 *tl = txdr_unsigned(nmp->nm_readdirsize);
2238 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2239 if (v3) {
2240 nfsm_postop_attr(vp, attrflag);
2241 if (!error) {
2242 tl = nfsm_dissect(u_int32_t *,
2243 2 * NFSX_UNSIGNED);
2244 mtx_lock(&dnp->n_mtx);
2245 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2246 dnp->n_cookieverf.nfsuquad[1] = *tl;
2247 mtx_unlock(&dnp->n_mtx);
2248 } else {
2249 m_freem(mrep);
2250 goto nfsmout;
2251 }
2252 }
2253 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2254 more_dirs = fxdr_unsigned(int, *tl);
2255
2256 /* loop thru the dir entries, doctoring them to 4bsd form */
2257 while (more_dirs && bigenough) {
2258 if (v3) {
2259 tl = nfsm_dissect(u_int32_t *,
2260 3 * NFSX_UNSIGNED);
2261 fileno = fxdr_hyper(tl);
2262 len = fxdr_unsigned(int, *(tl + 2));
2263 } else {
2264 tl = nfsm_dissect(u_int32_t *,
2265 2 * NFSX_UNSIGNED);
2266 fileno = fxdr_unsigned(u_quad_t, *tl++);
2267 len = fxdr_unsigned(int, *tl);
2268 }
2269 if (len <= 0 || len > NFS_MAXNAMLEN) {
2270 error = EBADRPC;
2271 m_freem(mrep);
2272 goto nfsmout;
2273 }
2274 tlen = nfsm_rndup(len);
2275 if (tlen == len)
2276 tlen += 4; /* To ensure null termination */
2277 left = DIRBLKSIZ - blksiz;
2278 if ((tlen + DIRHDSIZ) > left) {
2279 dp->d_reclen += left;
2280 uiop->uio_iov->iov_base =
2281 (char *)uiop->uio_iov->iov_base + left;
2282 uiop->uio_iov->iov_len -= left;
2283 uiop->uio_offset += left;
2284 uiop->uio_resid -= left;
2285 blksiz = 0;
2286 }
2287 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2288 bigenough = 0;
2289 if (bigenough) {
2290 dp = (struct dirent *)uiop->uio_iov->iov_base;
2291 dp->d_fileno = (int)fileno;
2292 dp->d_namlen = len;
2293 dp->d_reclen = tlen + DIRHDSIZ;
2294 dp->d_type = DT_UNKNOWN;
2295 blksiz += dp->d_reclen;
2296 if (blksiz == DIRBLKSIZ)
2297 blksiz = 0;
2298 uiop->uio_offset += DIRHDSIZ;
2299 uiop->uio_resid -= DIRHDSIZ;
2300 uiop->uio_iov->iov_base =
2301 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2302 uiop->uio_iov->iov_len -= DIRHDSIZ;
2303 nfsm_mtouio(uiop, len);
2304 cp = uiop->uio_iov->iov_base;
2305 tlen -= len;
2306 *cp = '\0'; /* null terminate */
2307 uiop->uio_iov->iov_base =
2308 (char *)uiop->uio_iov->iov_base + tlen;
2309 uiop->uio_iov->iov_len -= tlen;
2310 uiop->uio_offset += tlen;
2311 uiop->uio_resid -= tlen;
2312 } else
2313 nfsm_adv(nfsm_rndup(len));
2314 if (v3) {
2315 tl = nfsm_dissect(u_int32_t *,
2316 3 * NFSX_UNSIGNED);
2317 } else {
2318 tl = nfsm_dissect(u_int32_t *,
2319 2 * NFSX_UNSIGNED);
2320 }
2321 if (bigenough) {
2322 cookie.nfsuquad[0] = *tl++;
2323 if (v3)
2324 cookie.nfsuquad[1] = *tl++;
2325 } else if (v3)
2326 tl += 2;
2327 else
2328 tl++;
2329 more_dirs = fxdr_unsigned(int, *tl);
2330 }
2331 /*
2332 * If at end of rpc data, get the eof boolean
2333 */
2334 if (!more_dirs) {
2335 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2336 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2337 }
2338 m_freem(mrep);
2339 }
2340 /*
2341 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2342 * by increasing d_reclen for the last record.
2343 */
2344 if (blksiz > 0) {
2345 left = DIRBLKSIZ - blksiz;
2346 dp->d_reclen += left;
2347 uiop->uio_iov->iov_base =
2348 (char *)uiop->uio_iov->iov_base + left;
2349 uiop->uio_iov->iov_len -= left;
2350 uiop->uio_offset += left;
2351 uiop->uio_resid -= left;
2352 }
2353
2354 /*
2355 * We are now either at the end of the directory or have filled the
2356 * block.
2357 */
2358 if (bigenough)
2359 dnp->n_direofoffset = uiop->uio_offset;
2360 else {
2361 if (uiop->uio_resid > 0)
2362 nfs_printf("EEK! readdirrpc resid > 0\n");
2363 nfs_dircookie_lock(dnp);
2364 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2365 *cookiep = cookie;
2366 nfs_dircookie_unlock(dnp);
2367 }
2368 nfsmout:
2369 return (error);
2370 }
2371
2372 /*
2373 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2374 */
2375 int
2376 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2377 {
2378 int len, left;
2379 struct dirent *dp;
2380 u_int32_t *tl;
2381 caddr_t cp;
2382 struct vnode *newvp;
2383 nfsuint64 *cookiep;
2384 caddr_t bpos, dpos, dpossav1, dpossav2;
2385 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2386 struct nameidata nami, *ndp = &nami;
2387 struct componentname *cnp = &ndp->ni_cnd;
2388 nfsuint64 cookie;
2389 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2390 struct nfsnode *dnp = VTONFS(vp), *np;
2391 struct vattr vattr, dvattr;
2392 nfsfh_t *fhp;
2393 u_quad_t fileno;
2394 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2395 int attrflag, dattrflag, fhsize;
2396
2397 #ifndef nolint
2398 dp = NULL;
2399 #endif
2400 #ifndef DIAGNOSTIC
2401 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2402 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2403 panic("nfs readdirplusrpc bad uio");
2404 #endif
2405 ndp->ni_dvp = vp;
2406 newvp = NULLVP;
2407
2408 /*
2409 * If there is no cookie, assume directory was stale.
2410 */
2411 nfs_dircookie_lock(dnp);
2412 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2413 if (cookiep) {
2414 cookie = *cookiep;
2415 nfs_dircookie_unlock(dnp);
2416 } else {
2417 nfs_dircookie_unlock(dnp);
2418 return (NFSERR_BAD_COOKIE);
2419 }
2420 /*
2421 * Loop around doing readdir rpc's of size nm_readdirsize
2422 * truncated to a multiple of DIRBLKSIZ.
2423 * The stopping criteria is EOF or buffer full.
2424 */
2425 while (more_dirs && bigenough) {
2426 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2427 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2428 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2429 mb = mreq;
2430 bpos = mtod(mb, caddr_t);
2431 nfsm_fhtom(vp, 1);
2432 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2433 *tl++ = cookie.nfsuquad[0];
2434 *tl++ = cookie.nfsuquad[1];
2435 mtx_lock(&dnp->n_mtx);
2436 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2437 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2438 mtx_unlock(&dnp->n_mtx);
2439 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2440 *tl = txdr_unsigned(nmp->nm_rsize);
2441 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2442 nfsm_postop_attr_va(vp, dattrflag, &dvattr);
2443 if (error) {
2444 m_freem(mrep);
2445 goto nfsmout;
2446 }
2447 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2448 mtx_lock(&dnp->n_mtx);
2449 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2450 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2451 mtx_unlock(&dnp->n_mtx);
2452 more_dirs = fxdr_unsigned(int, *tl);
2453
2454 /* loop thru the dir entries, doctoring them to 4bsd form */
2455 while (more_dirs && bigenough) {
2456 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2457 fileno = fxdr_hyper(tl);
2458 len = fxdr_unsigned(int, *(tl + 2));
2459 if (len <= 0 || len > NFS_MAXNAMLEN) {
2460 error = EBADRPC;
2461 m_freem(mrep);
2462 goto nfsmout;
2463 }
2464 tlen = nfsm_rndup(len);
2465 if (tlen == len)
2466 tlen += 4; /* To ensure null termination*/
2467 left = DIRBLKSIZ - blksiz;
2468 if ((tlen + DIRHDSIZ) > left) {
2469 dp->d_reclen += left;
2470 uiop->uio_iov->iov_base =
2471 (char *)uiop->uio_iov->iov_base + left;
2472 uiop->uio_iov->iov_len -= left;
2473 uiop->uio_offset += left;
2474 uiop->uio_resid -= left;
2475 blksiz = 0;
2476 }
2477 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2478 bigenough = 0;
2479 if (bigenough) {
2480 dp = (struct dirent *)uiop->uio_iov->iov_base;
2481 dp->d_fileno = (int)fileno;
2482 dp->d_namlen = len;
2483 dp->d_reclen = tlen + DIRHDSIZ;
2484 dp->d_type = DT_UNKNOWN;
2485 blksiz += dp->d_reclen;
2486 if (blksiz == DIRBLKSIZ)
2487 blksiz = 0;
2488 uiop->uio_offset += DIRHDSIZ;
2489 uiop->uio_resid -= DIRHDSIZ;
2490 uiop->uio_iov->iov_base =
2491 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2492 uiop->uio_iov->iov_len -= DIRHDSIZ;
2493 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2494 cnp->cn_namelen = len;
2495 nfsm_mtouio(uiop, len);
2496 cp = uiop->uio_iov->iov_base;
2497 tlen -= len;
2498 *cp = '\0';
2499 uiop->uio_iov->iov_base =
2500 (char *)uiop->uio_iov->iov_base + tlen;
2501 uiop->uio_iov->iov_len -= tlen;
2502 uiop->uio_offset += tlen;
2503 uiop->uio_resid -= tlen;
2504 } else
2505 nfsm_adv(nfsm_rndup(len));
2506 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2507 if (bigenough) {
2508 cookie.nfsuquad[0] = *tl++;
2509 cookie.nfsuquad[1] = *tl++;
2510 } else
2511 tl += 2;
2512
2513 /*
2514 * Since the attributes are before the file handle
2515 * (sigh), we must skip over the attributes and then
2516 * come back and get them.
2517 */
2518 attrflag = fxdr_unsigned(int, *tl);
2519 if (attrflag) {
2520 dpossav1 = dpos;
2521 mdsav1 = md;
2522 nfsm_adv(NFSX_V3FATTR);
2523 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2524 doit = fxdr_unsigned(int, *tl);
2525 /*
2526 * Skip loading the attrs for "..". There's a
2527 * race between loading the attrs here and
2528 * lookups that look for the directory currently
2529 * being read (in the parent). We try to acquire
2530 * the exclusive lock on ".." here, owning the
2531 * lock on the directory being read. Lookup will
2532 * hold the lock on ".." and try to acquire the
2533 * lock on the directory being read.
2534 *
2535 * There are other ways of fixing this, one would
2536 * be to do a trylock on the ".." vnode and skip
2537 * loading the attrs on ".." if it happens to be
2538 * locked by another process. But skipping the
2539 * attrload on ".." seems the easiest option.
2540 */
2541 if (strcmp(dp->d_name, "..") == 0) {
2542 doit = 0;
2543 /*
2544 * We've already skipped over the attrs,
2545 * skip over the filehandle. And store d_type
2546 * as VDIR.
2547 */
2548 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2549 i = fxdr_unsigned(int, *tl);
2550 nfsm_adv(nfsm_rndup(i));
2551 dp->d_type = IFTODT(VTTOIF(VDIR));
2552 }
2553 if (doit) {
2554 nfsm_getfh(fhp, fhsize, 1);
2555 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2556 VREF(vp);
2557 newvp = vp;
2558 np = dnp;
2559 } else {
2560 error = nfs_nget(vp->v_mount, fhp,
2561 fhsize, &np, LK_EXCLUSIVE);
2562 if (error)
2563 doit = 0;
2564 else
2565 newvp = NFSTOV(np);
2566 }
2567 }
2568 if (doit && bigenough) {
2569 dpossav2 = dpos;
2570 dpos = dpossav1;
2571 mdsav2 = md;
2572 md = mdsav1;
2573 nfsm_loadattr(newvp, &vattr);
2574 dpos = dpossav2;
2575 md = mdsav2;
2576 dp->d_type = IFTODT(VTTOIF(vattr.va_type));
2577 ndp->ni_vp = newvp;
2578 if (newvp->v_type != VDIR || dattrflag != 0)
2579 cache_enter_time(ndp->ni_dvp, ndp->ni_vp,
2580 cnp, &vattr.va_ctime,
2581 newvp->v_type != VDIR ? NULL :
2582 &dvattr.va_ctime);
2583 }
2584 } else {
2585 /* Just skip over the file handle */
2586 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2587 i = fxdr_unsigned(int, *tl);
2588 if (i) {
2589 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2590 fhsize = fxdr_unsigned(int, *tl);
2591 nfsm_adv(nfsm_rndup(fhsize));
2592 }
2593 }
2594 if (newvp != NULLVP) {
2595 if (newvp == vp)
2596 vrele(newvp);
2597 else
2598 vput(newvp);
2599 newvp = NULLVP;
2600 }
2601 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2602 more_dirs = fxdr_unsigned(int, *tl);
2603 }
2604 /*
2605 * If at end of rpc data, get the eof boolean
2606 */
2607 if (!more_dirs) {
2608 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2609 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2610 }
2611 m_freem(mrep);
2612 }
2613 /*
2614 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2615 * by increasing d_reclen for the last record.
2616 */
2617 if (blksiz > 0) {
2618 left = DIRBLKSIZ - blksiz;
2619 dp->d_reclen += left;
2620 uiop->uio_iov->iov_base =
2621 (char *)uiop->uio_iov->iov_base + left;
2622 uiop->uio_iov->iov_len -= left;
2623 uiop->uio_offset += left;
2624 uiop->uio_resid -= left;
2625 }
2626
2627 /*
2628 * We are now either at the end of the directory or have filled the
2629 * block.
2630 */
2631 if (bigenough)
2632 dnp->n_direofoffset = uiop->uio_offset;
2633 else {
2634 if (uiop->uio_resid > 0)
2635 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2636 nfs_dircookie_lock(dnp);
2637 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2638 *cookiep = cookie;
2639 nfs_dircookie_unlock(dnp);
2640 }
2641 nfsmout:
2642 if (newvp != NULLVP) {
2643 if (newvp == vp)
2644 vrele(newvp);
2645 else
2646 vput(newvp);
2647 newvp = NULLVP;
2648 }
2649 return (error);
2650 }
2651
2652 /*
2653 * Silly rename. To make the NFS filesystem that is stateless look a little
2654 * more like the "ufs" a remove of an active vnode is translated to a rename
2655 * to a funny looking filename that is removed by nfs_inactive on the
2656 * nfsnode. There is the potential for another process on a different client
2657 * to create the same funny name between the nfs_lookitup() fails and the
2658 * nfs_rename() completes, but...
2659 */
2660 static int
2661 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2662 {
2663 struct sillyrename *sp;
2664 struct nfsnode *np;
2665 int error;
2666 short pid;
2667 unsigned int lticks;
2668
2669 cache_purge(dvp);
2670 np = VTONFS(vp);
2671 #ifndef DIAGNOSTIC
2672 if (vp->v_type == VDIR)
2673 panic("nfs: sillyrename dir");
2674 #endif
2675 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2676 M_NFSREQ, M_WAITOK);
2677 sp->s_cred = crhold(cnp->cn_cred);
2678 sp->s_dvp = dvp;
2679 sp->s_removeit = nfs_removeit;
2680 VREF(dvp);
2681
2682 /*
2683 * Fudge together a funny name.
2684 * Changing the format of the funny name to accomodate more
2685 * sillynames per directory.
2686 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2687 * CPU ticks since boot.
2688 */
2689 pid = cnp->cn_thread->td_proc->p_pid;
2690 lticks = (unsigned int)ticks;
2691 for ( ; ; ) {
2692 sp->s_namlen = sprintf(sp->s_name,
2693 ".nfs.%08x.%04x4.4", lticks,
2694 pid);
2695 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2696 cnp->cn_thread, NULL))
2697 break;
2698 lticks++;
2699 }
2700 error = nfs_renameit(dvp, cnp, sp);
2701 if (error)
2702 goto bad;
2703 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2704 cnp->cn_thread, &np);
2705 np->n_sillyrename = sp;
2706 return (0);
2707 bad:
2708 vrele(sp->s_dvp);
2709 crfree(sp->s_cred);
2710 free((caddr_t)sp, M_NFSREQ);
2711 return (error);
2712 }
2713
2714 /*
2715 * Look up a file name and optionally either update the file handle or
2716 * allocate an nfsnode, depending on the value of npp.
2717 * npp == NULL --> just do the lookup
2718 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2719 * handled too
2720 * *npp != NULL --> update the file handle in the vnode
2721 */
2722 static int
2723 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2724 struct thread *td, struct nfsnode **npp)
2725 {
2726 struct vnode *newvp = NULL;
2727 struct nfsnode *np, *dnp = VTONFS(dvp);
2728 caddr_t bpos, dpos;
2729 int error = 0, fhlen, attrflag;
2730 struct mbuf *mreq, *mrep, *md, *mb;
2731 nfsfh_t *nfhp;
2732 int v3 = NFS_ISV3(dvp);
2733
2734 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2735 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2736 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2737 mb = mreq;
2738 bpos = mtod(mb, caddr_t);
2739 nfsm_fhtom(dvp, v3);
2740 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2741 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2742 if (npp && !error) {
2743 nfsm_getfh(nfhp, fhlen, v3);
2744 if (*npp) {
2745 np = *npp;
2746 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2747 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2748 np->n_fhp = &np->n_fh;
2749 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2750 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2751 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2752 np->n_fhsize = fhlen;
2753 newvp = NFSTOV(np);
2754 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2755 VREF(dvp);
2756 newvp = dvp;
2757 } else {
2758 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2759 if (error) {
2760 m_freem(mrep);
2761 return (error);
2762 }
2763 newvp = NFSTOV(np);
2764 }
2765 if (v3) {
2766 nfsm_postop_attr(newvp, attrflag);
2767 if (!attrflag && *npp == NULL) {
2768 m_freem(mrep);
2769 if (newvp == dvp)
2770 vrele(newvp);
2771 else
2772 vput(newvp);
2773 return (ENOENT);
2774 }
2775 } else
2776 nfsm_loadattr(newvp, NULL);
2777 }
2778 m_freem(mrep);
2779 nfsmout:
2780 if (npp && *npp == NULL) {
2781 if (error) {
2782 if (newvp) {
2783 if (newvp == dvp)
2784 vrele(newvp);
2785 else
2786 vput(newvp);
2787 }
2788 } else
2789 *npp = np;
2790 }
2791 return (error);
2792 }
2793
2794 /*
2795 * Nfs Version 3 commit rpc
2796 */
2797 int
2798 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2799 struct thread *td)
2800 {
2801 u_int32_t *tl;
2802 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2803 caddr_t bpos, dpos;
2804 int error = 0, wccflag = NFSV3_WCCRATTR;
2805 struct mbuf *mreq, *mrep, *md, *mb;
2806
2807 mtx_lock(&nmp->nm_mtx);
2808 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2809 mtx_unlock(&nmp->nm_mtx);
2810 return (0);
2811 }
2812 mtx_unlock(&nmp->nm_mtx);
2813 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2814 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2815 mb = mreq;
2816 bpos = mtod(mb, caddr_t);
2817 nfsm_fhtom(vp, 1);
2818 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2819 txdr_hyper(offset, tl);
2820 tl += 2;
2821 *tl = txdr_unsigned(cnt);
2822 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2823 nfsm_wcc_data(vp, wccflag);
2824 if (!error) {
2825 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2826 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2827 NFSX_V3WRITEVERF)) {
2828 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2829 NFSX_V3WRITEVERF);
2830 error = NFSERR_STALEWRITEVERF;
2831 }
2832 }
2833 m_freem(mrep);
2834 nfsmout:
2835 return (error);
2836 }
2837
2838 /*
2839 * Strategy routine.
2840 * For async requests when nfsiod(s) are running, queue the request by
2841 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2842 * request.
2843 */
2844 static int
2845 nfs_strategy(struct vop_strategy_args *ap)
2846 {
2847 struct buf *bp = ap->a_bp;
2848 struct ucred *cr;
2849
2850 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2851 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
2852
2853 if (bp->b_iocmd == BIO_READ)
2854 cr = bp->b_rcred;
2855 else
2856 cr = bp->b_wcred;
2857
2858 /*
2859 * If the op is asynchronous and an i/o daemon is waiting
2860 * queue the request, wake it up and wait for completion
2861 * otherwise just do it ourselves.
2862 */
2863 if ((bp->b_flags & B_ASYNC) == 0 ||
2864 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2865 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2866 return (0);
2867 }
2868
2869 /*
2870 * fsync vnode op. Just call nfs_flush() with commit == 1.
2871 */
2872 /* ARGSUSED */
2873 static int
2874 nfs_fsync(struct vop_fsync_args *ap)
2875 {
2876 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2877 }
2878
2879 /*
2880 * Flush all the blocks associated with a vnode.
2881 * Walk through the buffer pool and push any dirty pages
2882 * associated with the vnode.
2883 */
2884 static int
2885 nfs_flush(struct vnode *vp, int waitfor, struct thread *td,
2886 int commit)
2887 {
2888 struct nfsnode *np = VTONFS(vp);
2889 struct buf *bp;
2890 int i;
2891 struct buf *nbp;
2892 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2893 int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2894 int passone = 1;
2895 u_quad_t off, endoff, toff;
2896 struct ucred* wcred = NULL;
2897 struct buf **bvec = NULL;
2898 #ifndef NFS_COMMITBVECSIZ
2899 #define NFS_COMMITBVECSIZ 20
2900 #endif
2901 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2902 int bvecsize = 0, bveccount;
2903
2904 if (nmp->nm_flag & NFSMNT_INT)
2905 slpflag = PCATCH;
2906 if (!commit)
2907 passone = 0;
2908 /*
2909 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2910 * server, but has not been committed to stable storage on the server
2911 * yet. On the first pass, the byte range is worked out and the commit
2912 * rpc is done. On the second pass, nfs_writebp() is called to do the
2913 * job.
2914 */
2915 again:
2916 off = (u_quad_t)-1;
2917 endoff = 0;
2918 bvecpos = 0;
2919 if (NFS_ISV3(vp) && commit) {
2920 s = splbio();
2921 if (bvec != NULL && bvec != bvec_on_stack)
2922 free(bvec, M_TEMP);
2923 /*
2924 * Count up how many buffers waiting for a commit.
2925 */
2926 bveccount = 0;
2927 VI_LOCK(vp);
2928 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2929 if (BUF_REFCNT(bp) == 0 &&
2930 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2931 == (B_DELWRI | B_NEEDCOMMIT))
2932 bveccount++;
2933 }
2934 /*
2935 * Allocate space to remember the list of bufs to commit. It is
2936 * important to use M_NOWAIT here to avoid a race with nfs_write.
2937 * If we can't get memory (for whatever reason), we will end up
2938 * committing the buffers one-by-one in the loop below.
2939 */
2940 if (bveccount > NFS_COMMITBVECSIZ) {
2941 /*
2942 * Release the vnode interlock to avoid a lock
2943 * order reversal.
2944 */
2945 VI_UNLOCK(vp);
2946 bvec = (struct buf **)
2947 malloc(bveccount * sizeof(struct buf *),
2948 M_TEMP, M_NOWAIT);
2949 VI_LOCK(vp);
2950 if (bvec == NULL) {
2951 bvec = bvec_on_stack;
2952 bvecsize = NFS_COMMITBVECSIZ;
2953 } else
2954 bvecsize = bveccount;
2955 } else {
2956 bvec = bvec_on_stack;
2957 bvecsize = NFS_COMMITBVECSIZ;
2958 }
2959 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2960 if (bvecpos >= bvecsize)
2961 break;
2962 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2963 nbp = TAILQ_NEXT(bp, b_bobufs);
2964 continue;
2965 }
2966 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2967 (B_DELWRI | B_NEEDCOMMIT)) {
2968 BUF_UNLOCK(bp);
2969 nbp = TAILQ_NEXT(bp, b_bobufs);
2970 continue;
2971 }
2972 VI_UNLOCK(vp);
2973 bremfree(bp);
2974 /*
2975 * Work out if all buffers are using the same cred
2976 * so we can deal with them all with one commit.
2977 *
2978 * NOTE: we are not clearing B_DONE here, so we have
2979 * to do it later on in this routine if we intend to
2980 * initiate I/O on the bp.
2981 *
2982 * Note: to avoid loopback deadlocks, we do not
2983 * assign b_runningbufspace.
2984 */
2985 if (wcred == NULL)
2986 wcred = bp->b_wcred;
2987 else if (wcred != bp->b_wcred)
2988 wcred = NOCRED;
2989 vfs_busy_pages(bp, 1);
2990
2991 VI_LOCK(vp);
2992 /*
2993 * bp is protected by being locked, but nbp is not
2994 * and vfs_busy_pages() may sleep. We have to
2995 * recalculate nbp.
2996 */
2997 nbp = TAILQ_NEXT(bp, b_bobufs);
2998
2999 /*
3000 * A list of these buffers is kept so that the
3001 * second loop knows which buffers have actually
3002 * been committed. This is necessary, since there
3003 * may be a race between the commit rpc and new
3004 * uncommitted writes on the file.
3005 */
3006 bvec[bvecpos++] = bp;
3007 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3008 bp->b_dirtyoff;
3009 if (toff < off)
3010 off = toff;
3011 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3012 if (toff > endoff)
3013 endoff = toff;
3014 }
3015 splx(s);
3016 VI_UNLOCK(vp);
3017 }
3018 if (bvecpos > 0) {
3019 /*
3020 * Commit data on the server, as required.
3021 * If all bufs are using the same wcred, then use that with
3022 * one call for all of them, otherwise commit each one
3023 * separately.
3024 */
3025 if (wcred != NOCRED)
3026 retv = nfs_commit(vp, off, (int)(endoff - off),
3027 wcred, td);
3028 else {
3029 retv = 0;
3030 for (i = 0; i < bvecpos; i++) {
3031 off_t off, size;
3032 bp = bvec[i];
3033 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3034 bp->b_dirtyoff;
3035 size = (u_quad_t)(bp->b_dirtyend
3036 - bp->b_dirtyoff);
3037 retv = nfs_commit(vp, off, (int)size,
3038 bp->b_wcred, td);
3039 if (retv) break;
3040 }
3041 }
3042
3043 if (retv == NFSERR_STALEWRITEVERF)
3044 nfs_clearcommit(vp->v_mount);
3045
3046 /*
3047 * Now, either mark the blocks I/O done or mark the
3048 * blocks dirty, depending on whether the commit
3049 * succeeded.
3050 */
3051 for (i = 0; i < bvecpos; i++) {
3052 bp = bvec[i];
3053 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3054 if (retv) {
3055 /*
3056 * Error, leave B_DELWRI intact
3057 */
3058 vfs_unbusy_pages(bp);
3059 brelse(bp);
3060 } else {
3061 /*
3062 * Success, remove B_DELWRI ( bundirty() ).
3063 *
3064 * b_dirtyoff/b_dirtyend seem to be NFS
3065 * specific. We should probably move that
3066 * into bundirty(). XXX
3067 */
3068 s = splbio();
3069 bufobj_wref(&vp->v_bufobj);
3070 bp->b_flags |= B_ASYNC;
3071 bundirty(bp);
3072 bp->b_flags &= ~B_DONE;
3073 bp->b_ioflags &= ~BIO_ERROR;
3074 bp->b_dirtyoff = bp->b_dirtyend = 0;
3075 splx(s);
3076 bufdone(bp);
3077 }
3078 }
3079 }
3080
3081 /*
3082 * Start/do any write(s) that are required.
3083 */
3084 loop:
3085 s = splbio();
3086 VI_LOCK(vp);
3087 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
3088 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3089 if (waitfor != MNT_WAIT || passone)
3090 continue;
3091
3092 error = BUF_TIMELOCK(bp,
3093 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3094 VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
3095 splx(s);
3096 if (error == 0) {
3097 BUF_UNLOCK(bp);
3098 goto loop;
3099 }
3100 if (error == ENOLCK)
3101 goto loop;
3102 if (nfs_sigintr(nmp, NULL, td)) {
3103 error = EINTR;
3104 goto done;
3105 }
3106 if (slpflag == PCATCH) {
3107 slpflag = 0;
3108 slptimeo = 2 * hz;
3109 }
3110 goto loop;
3111 }
3112 if ((bp->b_flags & B_DELWRI) == 0)
3113 panic("nfs_fsync: not dirty");
3114 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3115 BUF_UNLOCK(bp);
3116 continue;
3117 }
3118 VI_UNLOCK(vp);
3119 bremfree(bp);
3120 if (passone || !commit)
3121 bp->b_flags |= B_ASYNC;
3122 else
3123 bp->b_flags |= B_ASYNC;
3124 splx(s);
3125 bwrite(bp);
3126 if (nfs_sigintr(nmp, NULL, td)) {
3127 error = EINTR;
3128 goto done;
3129 }
3130 goto loop;
3131 }
3132 splx(s);
3133 if (passone) {
3134 passone = 0;
3135 VI_UNLOCK(vp);
3136 goto again;
3137 }
3138 if (waitfor == MNT_WAIT) {
3139 while (vp->v_bufobj.bo_numoutput) {
3140 error = bufobj_wwait(&vp->v_bufobj, slpflag, slptimeo);
3141 if (error) {
3142 VI_UNLOCK(vp);
3143 error = nfs_sigintr(nmp, NULL, td);
3144 if (error)
3145 goto done;
3146 if (slpflag == PCATCH) {
3147 slpflag = 0;
3148 slptimeo = 2 * hz;
3149 }
3150 VI_LOCK(vp);
3151 }
3152 }
3153 if (vp->v_bufobj.bo_dirty.bv_cnt != 0 && commit) {
3154 VI_UNLOCK(vp);
3155 goto loop;
3156 }
3157 /*
3158 * Wait for all the async IO requests to drain
3159 */
3160 VI_UNLOCK(vp);
3161 mtx_lock(&np->n_mtx);
3162 while (np->n_directio_asyncwr > 0) {
3163 np->n_flag |= NFSYNCWAIT;
3164 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3165 &np->n_mtx, slpflag | (PRIBIO + 1),
3166 "nfsfsync", 0);
3167 if (error) {
3168 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
3169 mtx_unlock(&np->n_mtx);
3170 error = EINTR;
3171 goto done;
3172 }
3173 }
3174 }
3175 mtx_unlock(&np->n_mtx);
3176 } else
3177 VI_UNLOCK(vp);
3178 mtx_lock(&np->n_mtx);
3179 if (np->n_flag & NWRITEERR) {
3180 error = np->n_error;
3181 np->n_flag &= ~NWRITEERR;
3182 }
3183 if (commit && vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
3184 vp->v_bufobj.bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3185 np->n_flag &= ~NMODIFIED;
3186 mtx_unlock(&np->n_mtx);
3187 done:
3188 if (bvec != NULL && bvec != bvec_on_stack)
3189 free(bvec, M_TEMP);
3190 return (error);
3191 }
3192
3193 /*
3194 * NFS advisory byte-level locks.
3195 */
3196 static int
3197 nfs_advlock(struct vop_advlock_args *ap)
3198 {
3199 struct vnode *vp = ap->a_vp;
3200 u_quad_t size;
3201 int error;
3202
3203 error = vn_lock(vp, LK_SHARED, curthread);
3204 if (error)
3205 return (error);
3206 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3207 size = VTONFS(vp)->n_size;
3208 VOP_UNLOCK(vp, 0, curthread);
3209 error = lf_advlock(ap, &(vp->v_lockf), size);
3210 } else {
3211 if (nfs_advlock_p)
3212 error = nfs_advlock_p(ap);
3213 else
3214 error = ENOLCK;
3215 }
3216
3217 return (error);
3218 }
3219
3220 /*
3221 * NFS advisory byte-level locks.
3222 */
3223 static int
3224 nfs_advlockasync(struct vop_advlockasync_args *ap)
3225 {
3226 struct vnode *vp = ap->a_vp;
3227 u_quad_t size;
3228 int error;
3229
3230 error = vn_lock(vp, LK_SHARED, curthread);
3231 if (error)
3232 return (error);
3233 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3234 size = VTONFS(vp)->n_size;
3235 VOP_UNLOCK(vp, 0, curthread);
3236 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3237 } else {
3238 VOP_UNLOCK(vp, 0, curthread);
3239 error = EOPNOTSUPP;
3240 }
3241 return (error);
3242 }
3243
3244 /*
3245 * Print out the contents of an nfsnode.
3246 */
3247 static int
3248 nfs_print(struct vop_print_args *ap)
3249 {
3250 struct vnode *vp = ap->a_vp;
3251 struct nfsnode *np = VTONFS(vp);
3252
3253 nfs_printf("\tfileid %ld fsid 0x%x",
3254 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3255 if (vp->v_type == VFIFO)
3256 fifo_printinfo(vp);
3257 printf("\n");
3258 return (0);
3259 }
3260
3261 /*
3262 * This is the "real" nfs::bwrite(struct buf*).
3263 * We set B_CACHE if this is a VMIO buffer.
3264 */
3265 int
3266 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3267 {
3268 int s;
3269 int oldflags = bp->b_flags;
3270 #if 0
3271 int retv = 1;
3272 off_t off;
3273 #endif
3274
3275 if (BUF_REFCNT(bp) == 0)
3276 panic("bwrite: buffer is not locked???");
3277
3278 if (bp->b_flags & B_INVAL) {
3279 brelse(bp);
3280 return(0);
3281 }
3282
3283 bp->b_flags |= B_CACHE;
3284
3285 /*
3286 * Undirty the bp. We will redirty it later if the I/O fails.
3287 */
3288
3289 s = splbio();
3290 bundirty(bp);
3291 bp->b_flags &= ~B_DONE;
3292 bp->b_ioflags &= ~BIO_ERROR;
3293 bp->b_iocmd = BIO_WRITE;
3294
3295 bufobj_wref(bp->b_bufobj);
3296 curthread->td_ru.ru_oublock++;
3297 splx(s);
3298
3299 /*
3300 * Note: to avoid loopback deadlocks, we do not
3301 * assign b_runningbufspace.
3302 */
3303 vfs_busy_pages(bp, 1);
3304
3305 BUF_KERNPROC(bp);
3306 bp->b_iooffset = dbtob(bp->b_blkno);
3307 bstrategy(bp);
3308
3309 if( (oldflags & B_ASYNC) == 0) {
3310 int rtval = bufwait(bp);
3311
3312 if (oldflags & B_DELWRI) {
3313 s = splbio();
3314 reassignbuf(bp);
3315 splx(s);
3316 }
3317 brelse(bp);
3318 return (rtval);
3319 }
3320
3321 return (0);
3322 }
3323
3324 /*
3325 * nfs special file access vnode op.
3326 * Essentially just get vattr and then imitate iaccess() since the device is
3327 * local to the client.
3328 */
3329 static int
3330 nfsspec_access(struct vop_access_args *ap)
3331 {
3332 struct vattr *vap;
3333 struct ucred *cred = ap->a_cred;
3334 struct vnode *vp = ap->a_vp;
3335 mode_t mode = ap->a_mode;
3336 struct vattr vattr;
3337 int error;
3338
3339 /*
3340 * Disallow write attempts on filesystems mounted read-only;
3341 * unless the file is a socket, fifo, or a block or character
3342 * device resident on the filesystem.
3343 */
3344 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3345 switch (vp->v_type) {
3346 case VREG:
3347 case VDIR:
3348 case VLNK:
3349 return (EROFS);
3350 default:
3351 break;
3352 }
3353 }
3354 vap = &vattr;
3355 error = VOP_GETATTR(vp, vap, cred, ap->a_td);
3356 if (error)
3357 goto out;
3358 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3359 mode, cred, NULL);
3360 out:
3361 return error;
3362 }
3363
3364 /*
3365 * Read wrapper for fifos.
3366 */
3367 static int
3368 nfsfifo_read(struct vop_read_args *ap)
3369 {
3370 struct nfsnode *np = VTONFS(ap->a_vp);
3371 int error;
3372
3373 /*
3374 * Set access flag.
3375 */
3376 mtx_lock(&np->n_mtx);
3377 np->n_flag |= NACC;
3378 getnanotime(&np->n_atim);
3379 mtx_unlock(&np->n_mtx);
3380 error = fifo_specops.vop_read(ap);
3381 return error;
3382 }
3383
3384 /*
3385 * Write wrapper for fifos.
3386 */
3387 static int
3388 nfsfifo_write(struct vop_write_args *ap)
3389 {
3390 struct nfsnode *np = VTONFS(ap->a_vp);
3391
3392 /*
3393 * Set update flag.
3394 */
3395 mtx_lock(&np->n_mtx);
3396 np->n_flag |= NUPD;
3397 getnanotime(&np->n_mtim);
3398 mtx_unlock(&np->n_mtx);
3399 return(fifo_specops.vop_write(ap));
3400 }
3401
3402 /*
3403 * Close wrapper for fifos.
3404 *
3405 * Update the times on the nfsnode then do fifo close.
3406 */
3407 static int
3408 nfsfifo_close(struct vop_close_args *ap)
3409 {
3410 struct vnode *vp = ap->a_vp;
3411 struct nfsnode *np = VTONFS(vp);
3412 struct vattr vattr;
3413 struct timespec ts;
3414
3415 mtx_lock(&np->n_mtx);
3416 if (np->n_flag & (NACC | NUPD)) {
3417 getnanotime(&ts);
3418 if (np->n_flag & NACC)
3419 np->n_atim = ts;
3420 if (np->n_flag & NUPD)
3421 np->n_mtim = ts;
3422 np->n_flag |= NCHG;
3423 if (vrefcnt(vp) == 1 &&
3424 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3425 VATTR_NULL(&vattr);
3426 if (np->n_flag & NACC)
3427 vattr.va_atime = np->n_atim;
3428 if (np->n_flag & NUPD)
3429 vattr.va_mtime = np->n_mtim;
3430 mtx_unlock(&np->n_mtx);
3431 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
3432 goto out;
3433 }
3434 }
3435 mtx_unlock(&np->n_mtx);
3436 out:
3437 return (fifo_specops.vop_close(ap));
3438 }
3439
3440 /*
3441 * Just call nfs_writebp() with the force argument set to 1.
3442 *
3443 * NOTE: B_DONE may or may not be set in a_bp on call.
3444 */
3445 static int
3446 nfs_bwrite(struct buf *bp)
3447 {
3448
3449 return (nfs_writebp(bp, 1, curthread));
3450 }
3451
3452 struct buf_ops buf_ops_nfs = {
3453 .bop_name = "buf_ops_nfs",
3454 .bop_write = nfs_bwrite,
3455 .bop_strategy = bufstrategy,
3456 .bop_sync = bufsync,
3457 .bop_bdflush = bufbdflush,
3458 };
Cache object: 0789e402b1d5b781181809de3546957f
|