1 /*-
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD: releng/8.0/sys/nfsclient/nfs_vnops.c 198424 2009-10-23 19:52:29Z jhb $");
37
38 /*
39 * vnode op calls for Sun NFS version 2 and 3
40 */
41
42 #include "opt_inet.h"
43 #include "opt_kdtrace.h"
44
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/systm.h>
48 #include <sys/resourcevar.h>
49 #include <sys/proc.h>
50 #include <sys/mount.h>
51 #include <sys/bio.h>
52 #include <sys/buf.h>
53 #include <sys/jail.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/namei.h>
57 #include <sys/socket.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/fcntl.h>
61 #include <sys/lockf.h>
62 #include <sys/stat.h>
63 #include <sys/sysctl.h>
64 #include <sys/signalvar.h>
65
66 #include <vm/vm.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_extern.h>
69 #include <vm/vm_object.h>
70
71 #include <fs/fifofs/fifo.h>
72
73 #include <nfs/nfsproto.h>
74 #include <nfsclient/nfs.h>
75 #include <nfsclient/nfsnode.h>
76 #include <nfsclient/nfsmount.h>
77 #include <nfsclient/nfs_kdtrace.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 #include <machine/stdarg.h>
87
88 #ifdef KDTRACE_HOOKS
89 #include <sys/dtrace_bsd.h>
90
91 dtrace_nfsclient_accesscache_flush_probe_func_t
92 dtrace_nfsclient_accesscache_flush_done_probe;
93 uint32_t nfsclient_accesscache_flush_done_id;
94
95 dtrace_nfsclient_accesscache_get_probe_func_t
96 dtrace_nfsclient_accesscache_get_hit_probe,
97 dtrace_nfsclient_accesscache_get_miss_probe;
98 uint32_t nfsclient_accesscache_get_hit_id;
99 uint32_t nfsclient_accesscache_get_miss_id;
100
101 dtrace_nfsclient_accesscache_load_probe_func_t
102 dtrace_nfsclient_accesscache_load_done_probe;
103 uint32_t nfsclient_accesscache_load_done_id;
104 #endif /* !KDTRACE_HOOKS */
105
106 /* Defs */
107 #define TRUE 1
108 #define FALSE 0
109
110 /*
111 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
112 * calls are not in getblk() and brelse() so that they would not be necessary
113 * here.
114 */
115 #ifndef B_VMIO
116 #define vfs_busy_pages(bp, f)
117 #endif
118
119 static vop_read_t nfsfifo_read;
120 static vop_write_t nfsfifo_write;
121 static vop_close_t nfsfifo_close;
122 static int nfs_flush(struct vnode *, int, int);
123 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *);
124 static vop_lookup_t nfs_lookup;
125 static vop_create_t nfs_create;
126 static vop_mknod_t nfs_mknod;
127 static vop_open_t nfs_open;
128 static vop_close_t nfs_close;
129 static vop_access_t nfs_access;
130 static vop_getattr_t nfs_getattr;
131 static vop_setattr_t nfs_setattr;
132 static vop_read_t nfs_read;
133 static vop_fsync_t nfs_fsync;
134 static vop_remove_t nfs_remove;
135 static vop_link_t nfs_link;
136 static vop_rename_t nfs_rename;
137 static vop_mkdir_t nfs_mkdir;
138 static vop_rmdir_t nfs_rmdir;
139 static vop_symlink_t nfs_symlink;
140 static vop_readdir_t nfs_readdir;
141 static vop_strategy_t nfs_strategy;
142 static int nfs_lookitup(struct vnode *, const char *, int,
143 struct ucred *, struct thread *, struct nfsnode **);
144 static int nfs_sillyrename(struct vnode *, struct vnode *,
145 struct componentname *);
146 static vop_access_t nfsspec_access;
147 static vop_readlink_t nfs_readlink;
148 static vop_print_t nfs_print;
149 static vop_advlock_t nfs_advlock;
150 static vop_advlockasync_t nfs_advlockasync;
151
152 /*
153 * Global vfs data structures for nfs
154 */
155 struct vop_vector nfs_vnodeops = {
156 .vop_default = &default_vnodeops,
157 .vop_access = nfs_access,
158 .vop_advlock = nfs_advlock,
159 .vop_advlockasync = nfs_advlockasync,
160 .vop_close = nfs_close,
161 .vop_create = nfs_create,
162 .vop_fsync = nfs_fsync,
163 .vop_getattr = nfs_getattr,
164 .vop_getpages = nfs_getpages,
165 .vop_putpages = nfs_putpages,
166 .vop_inactive = nfs_inactive,
167 .vop_link = nfs_link,
168 .vop_lookup = nfs_lookup,
169 .vop_mkdir = nfs_mkdir,
170 .vop_mknod = nfs_mknod,
171 .vop_open = nfs_open,
172 .vop_print = nfs_print,
173 .vop_read = nfs_read,
174 .vop_readdir = nfs_readdir,
175 .vop_readlink = nfs_readlink,
176 .vop_reclaim = nfs_reclaim,
177 .vop_remove = nfs_remove,
178 .vop_rename = nfs_rename,
179 .vop_rmdir = nfs_rmdir,
180 .vop_setattr = nfs_setattr,
181 .vop_strategy = nfs_strategy,
182 .vop_symlink = nfs_symlink,
183 .vop_write = nfs_write,
184 };
185
186 struct vop_vector nfs_fifoops = {
187 .vop_default = &fifo_specops,
188 .vop_access = nfsspec_access,
189 .vop_close = nfsfifo_close,
190 .vop_fsync = nfs_fsync,
191 .vop_getattr = nfs_getattr,
192 .vop_inactive = nfs_inactive,
193 .vop_print = nfs_print,
194 .vop_read = nfsfifo_read,
195 .vop_reclaim = nfs_reclaim,
196 .vop_setattr = nfs_setattr,
197 .vop_write = nfsfifo_write,
198 };
199
200 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
201 struct componentname *cnp, struct vattr *vap);
202 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
203 struct ucred *cred, struct thread *td);
204 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
205 int fnamelen, struct vnode *tdvp,
206 const char *tnameptr, int tnamelen,
207 struct ucred *cred, struct thread *td);
208 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
209 struct sillyrename *sp);
210
211 /*
212 * Global variables
213 */
214 struct mtx nfs_iod_mtx;
215 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
216 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
217 int nfs_numasync = 0;
218 vop_advlock_t *nfs_advlock_p = nfs_dolock;
219 vop_reclaim_t *nfs_reclaim_p = NULL;
220 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
221
222 SYSCTL_DECL(_vfs_nfs);
223
224 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
225 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
226 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
227
228 static int nfs_prime_access_cache = 0;
229 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
230 &nfs_prime_access_cache, 0,
231 "Prime NFS ACCESS cache when fetching attributes");
232
233 static int nfsv3_commit_on_close = 0;
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
235 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
236
237 static int nfs_clean_pages_on_close = 1;
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
239 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
240
241 int nfs_directio_enable = 0;
242 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
243 &nfs_directio_enable, 0, "Enable NFS directio");
244
245 /*
246 * This sysctl allows other processes to mmap a file that has been opened
247 * O_DIRECT by a process. In general, having processes mmap the file while
248 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
249 * this by default to prevent DoS attacks - to prevent a malicious user from
250 * opening up files O_DIRECT preventing other users from mmap'ing these
251 * files. "Protected" environments where stricter consistency guarantees are
252 * required can disable this knob. The process that opened the file O_DIRECT
253 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
254 * meaningful.
255 */
256 int nfs_directio_allow_mmap = 1;
257 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
258 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
259
260 #if 0
261 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
262 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
263
264 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
265 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
266 #endif
267
268 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
269 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
270 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
271
272 /*
273 * SMP Locking Note :
274 * The list of locks after the description of the lock is the ordering
275 * of other locks acquired with the lock held.
276 * np->n_mtx : Protects the fields in the nfsnode.
277 VM Object Lock
278 VI_MTX (acquired indirectly)
279 * nmp->nm_mtx : Protects the fields in the nfsmount.
280 rep->r_mtx
281 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
282 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
283 nmp->nm_mtx
284 rep->r_mtx
285 * rep->r_mtx : Protects the fields in an nfsreq.
286 */
287
288 static int
289 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
290 struct ucred *cred, uint32_t *retmode)
291 {
292 const int v3 = 1;
293 u_int32_t *tl;
294 int error = 0, attrflag, i, lrupos;
295
296 struct mbuf *mreq, *mrep, *md, *mb;
297 caddr_t bpos, dpos;
298 u_int32_t rmode;
299 struct nfsnode *np = VTONFS(vp);
300
301 nfsstats.rpccnt[NFSPROC_ACCESS]++;
302 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
303 mb = mreq;
304 bpos = mtod(mb, caddr_t);
305 nfsm_fhtom(vp, v3);
306 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
307 *tl = txdr_unsigned(wmode);
308 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
309 nfsm_postop_attr(vp, attrflag);
310 if (!error) {
311 lrupos = 0;
312 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
313 rmode = fxdr_unsigned(u_int32_t, *tl);
314 mtx_lock(&np->n_mtx);
315 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
316 if (np->n_accesscache[i].uid == cred->cr_uid) {
317 np->n_accesscache[i].mode = rmode;
318 np->n_accesscache[i].stamp = time_second;
319 break;
320 }
321 if (i > 0 && np->n_accesscache[i].stamp <
322 np->n_accesscache[lrupos].stamp)
323 lrupos = i;
324 }
325 if (i == NFS_ACCESSCACHESIZE) {
326 np->n_accesscache[lrupos].uid = cred->cr_uid;
327 np->n_accesscache[lrupos].mode = rmode;
328 np->n_accesscache[lrupos].stamp = time_second;
329 }
330 mtx_unlock(&np->n_mtx);
331 if (retmode != NULL)
332 *retmode = rmode;
333 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
334 }
335 m_freem(mrep);
336 nfsmout:
337 #ifdef KDTRACE_HOOKS
338 if (error) {
339 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
340 error);
341 }
342 #endif
343 return (error);
344 }
345
346 /*
347 * nfs access vnode op.
348 * For nfs version 2, just return ok. File accesses may fail later.
349 * For nfs version 3, use the access rpc to check accessibility. If file modes
350 * are changed on the server, accesses might still fail later.
351 */
352 static int
353 nfs_access(struct vop_access_args *ap)
354 {
355 struct vnode *vp = ap->a_vp;
356 int error = 0, i, gotahit;
357 u_int32_t mode, rmode, wmode;
358 int v3 = NFS_ISV3(vp);
359 struct nfsnode *np = VTONFS(vp);
360
361 /*
362 * Disallow write attempts on filesystems mounted read-only;
363 * unless the file is a socket, fifo, or a block or character
364 * device resident on the filesystem.
365 */
366 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
367 switch (vp->v_type) {
368 case VREG:
369 case VDIR:
370 case VLNK:
371 return (EROFS);
372 default:
373 break;
374 }
375 }
376 /*
377 * For nfs v3, check to see if we have done this recently, and if
378 * so return our cached result instead of making an ACCESS call.
379 * If not, do an access rpc, otherwise you are stuck emulating
380 * ufs_access() locally using the vattr. This may not be correct,
381 * since the server may apply other access criteria such as
382 * client uid-->server uid mapping that we do not know about.
383 */
384 if (v3) {
385 if (ap->a_accmode & VREAD)
386 mode = NFSV3ACCESS_READ;
387 else
388 mode = 0;
389 if (vp->v_type != VDIR) {
390 if (ap->a_accmode & VWRITE)
391 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
392 if (ap->a_accmode & VEXEC)
393 mode |= NFSV3ACCESS_EXECUTE;
394 } else {
395 if (ap->a_accmode & VWRITE)
396 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
397 NFSV3ACCESS_DELETE);
398 if (ap->a_accmode & VEXEC)
399 mode |= NFSV3ACCESS_LOOKUP;
400 }
401 /* XXX safety belt, only make blanket request if caching */
402 if (nfsaccess_cache_timeout > 0) {
403 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
404 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
405 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
406 } else {
407 wmode = mode;
408 }
409
410 /*
411 * Does our cached result allow us to give a definite yes to
412 * this request?
413 */
414 gotahit = 0;
415 mtx_lock(&np->n_mtx);
416 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
417 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
418 if (time_second < (np->n_accesscache[i].stamp +
419 nfsaccess_cache_timeout) &&
420 (np->n_accesscache[i].mode & mode) == mode) {
421 nfsstats.accesscache_hits++;
422 gotahit = 1;
423 }
424 break;
425 }
426 }
427 mtx_unlock(&np->n_mtx);
428 #ifdef KDTRACE_HOOKS
429 if (gotahit)
430 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
431 ap->a_cred->cr_uid, mode);
432 else
433 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
434 ap->a_cred->cr_uid, mode);
435 #endif
436 if (gotahit == 0) {
437 /*
438 * Either a no, or a don't know. Go to the wire.
439 */
440 nfsstats.accesscache_misses++;
441 error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred,
442 &rmode);
443 if (!error) {
444 if ((rmode & mode) != mode)
445 error = EACCES;
446 }
447 }
448 return (error);
449 } else {
450 if ((error = nfsspec_access(ap)) != 0) {
451 return (error);
452 }
453 /*
454 * Attempt to prevent a mapped root from accessing a file
455 * which it shouldn't. We try to read a byte from the file
456 * if the user is root and the file is not zero length.
457 * After calling nfsspec_access, we should have the correct
458 * file size cached.
459 */
460 mtx_lock(&np->n_mtx);
461 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
462 && VTONFS(vp)->n_size > 0) {
463 struct iovec aiov;
464 struct uio auio;
465 char buf[1];
466
467 mtx_unlock(&np->n_mtx);
468 aiov.iov_base = buf;
469 aiov.iov_len = 1;
470 auio.uio_iov = &aiov;
471 auio.uio_iovcnt = 1;
472 auio.uio_offset = 0;
473 auio.uio_resid = 1;
474 auio.uio_segflg = UIO_SYSSPACE;
475 auio.uio_rw = UIO_READ;
476 auio.uio_td = ap->a_td;
477
478 if (vp->v_type == VREG)
479 error = nfs_readrpc(vp, &auio, ap->a_cred);
480 else if (vp->v_type == VDIR) {
481 char* bp;
482 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
483 aiov.iov_base = bp;
484 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
485 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
486 free(bp, M_TEMP);
487 } else if (vp->v_type == VLNK)
488 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
489 else
490 error = EACCES;
491 } else
492 mtx_unlock(&np->n_mtx);
493 return (error);
494 }
495 }
496
497 int nfs_otw_getattr_avoid = 0;
498
499 /*
500 * nfs open vnode op
501 * Check to see if the type is ok
502 * and that deletion is not in progress.
503 * For paged in text files, you will need to flush the page cache
504 * if consistency is lost.
505 */
506 /* ARGSUSED */
507 static int
508 nfs_open(struct vop_open_args *ap)
509 {
510 struct vnode *vp = ap->a_vp;
511 struct nfsnode *np = VTONFS(vp);
512 struct vattr vattr;
513 int error;
514 int fmode = ap->a_mode;
515
516 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
517 return (EOPNOTSUPP);
518
519 /*
520 * Get a valid lease. If cached data is stale, flush it.
521 */
522 mtx_lock(&np->n_mtx);
523 if (np->n_flag & NMODIFIED) {
524 mtx_unlock(&np->n_mtx);
525 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
526 if (error == EINTR || error == EIO)
527 return (error);
528 np->n_attrstamp = 0;
529 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
530 if (vp->v_type == VDIR)
531 np->n_direofoffset = 0;
532 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
533 if (error)
534 return (error);
535 mtx_lock(&np->n_mtx);
536 np->n_mtime = vattr.va_mtime;
537 mtx_unlock(&np->n_mtx);
538 } else {
539 struct thread *td = curthread;
540
541 if (np->n_ac_ts_syscalls != td->td_syscalls ||
542 np->n_ac_ts_tid != td->td_tid ||
543 td->td_proc == NULL ||
544 np->n_ac_ts_pid != td->td_proc->p_pid) {
545 np->n_attrstamp = 0;
546 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
547 }
548 mtx_unlock(&np->n_mtx);
549 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
550 if (error)
551 return (error);
552 mtx_lock(&np->n_mtx);
553 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
554 if (vp->v_type == VDIR)
555 np->n_direofoffset = 0;
556 mtx_unlock(&np->n_mtx);
557 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
558 if (error == EINTR || error == EIO) {
559 return (error);
560 }
561 mtx_lock(&np->n_mtx);
562 np->n_mtime = vattr.va_mtime;
563 }
564 mtx_unlock(&np->n_mtx);
565 }
566 /*
567 * If the object has >= 1 O_DIRECT active opens, we disable caching.
568 */
569 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
570 if (np->n_directio_opens == 0) {
571 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
572 if (error)
573 return (error);
574 mtx_lock(&np->n_mtx);
575 np->n_flag |= NNONCACHE;
576 mtx_unlock(&np->n_mtx);
577 }
578 np->n_directio_opens++;
579 }
580 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
581 return (0);
582 }
583
584 /*
585 * nfs close vnode op
586 * What an NFS client should do upon close after writing is a debatable issue.
587 * Most NFS clients push delayed writes to the server upon close, basically for
588 * two reasons:
589 * 1 - So that any write errors may be reported back to the client process
590 * doing the close system call. By far the two most likely errors are
591 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
592 * 2 - To put a worst case upper bound on cache inconsistency between
593 * multiple clients for the file.
594 * There is also a consistency problem for Version 2 of the protocol w.r.t.
595 * not being able to tell if other clients are writing a file concurrently,
596 * since there is no way of knowing if the changed modify time in the reply
597 * is only due to the write for this client.
598 * (NFS Version 3 provides weak cache consistency data in the reply that
599 * should be sufficient to detect and handle this case.)
600 *
601 * The current code does the following:
602 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
603 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
604 * or commit them (this satisfies 1 and 2 except for the
605 * case where the server crashes after this close but
606 * before the commit RPC, which is felt to be "good
607 * enough". Changing the last argument to nfs_flush() to
608 * a 1 would force a commit operation, if it is felt a
609 * commit is necessary now.
610 */
611 /* ARGSUSED */
612 static int
613 nfs_close(struct vop_close_args *ap)
614 {
615 struct vnode *vp = ap->a_vp;
616 struct nfsnode *np = VTONFS(vp);
617 int error = 0;
618 int fmode = ap->a_fflag;
619
620 if (vp->v_type == VREG) {
621 /*
622 * Examine and clean dirty pages, regardless of NMODIFIED.
623 * This closes a major hole in close-to-open consistency.
624 * We want to push out all dirty pages (and buffers) on
625 * close, regardless of whether they were dirtied by
626 * mmap'ed writes or via write().
627 */
628 if (nfs_clean_pages_on_close && vp->v_object) {
629 VM_OBJECT_LOCK(vp->v_object);
630 vm_object_page_clean(vp->v_object, 0, 0, 0);
631 VM_OBJECT_UNLOCK(vp->v_object);
632 }
633 mtx_lock(&np->n_mtx);
634 if (np->n_flag & NMODIFIED) {
635 mtx_unlock(&np->n_mtx);
636 if (NFS_ISV3(vp)) {
637 /*
638 * Under NFSv3 we have dirty buffers to dispose of. We
639 * must flush them to the NFS server. We have the option
640 * of waiting all the way through the commit rpc or just
641 * waiting for the initial write. The default is to only
642 * wait through the initial write so the data is in the
643 * server's cache, which is roughly similar to the state
644 * a standard disk subsystem leaves the file in on close().
645 *
646 * We cannot clear the NMODIFIED bit in np->n_flag due to
647 * potential races with other processes, and certainly
648 * cannot clear it if we don't commit.
649 */
650 int cm = nfsv3_commit_on_close ? 1 : 0;
651 error = nfs_flush(vp, MNT_WAIT, cm);
652 /* np->n_flag &= ~NMODIFIED; */
653 } else
654 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
655 mtx_lock(&np->n_mtx);
656 }
657 if (np->n_flag & NWRITEERR) {
658 np->n_flag &= ~NWRITEERR;
659 error = np->n_error;
660 }
661 mtx_unlock(&np->n_mtx);
662 }
663 if (nfs_directio_enable)
664 KASSERT((np->n_directio_asyncwr == 0),
665 ("nfs_close: dirty unflushed (%d) directio buffers\n",
666 np->n_directio_asyncwr));
667 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
668 mtx_lock(&np->n_mtx);
669 KASSERT((np->n_directio_opens > 0),
670 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
671 np->n_directio_opens--;
672 if (np->n_directio_opens == 0)
673 np->n_flag &= ~NNONCACHE;
674 mtx_unlock(&np->n_mtx);
675 }
676 return (error);
677 }
678
679 /*
680 * nfs getattr call from vfs.
681 */
682 static int
683 nfs_getattr(struct vop_getattr_args *ap)
684 {
685 struct vnode *vp = ap->a_vp;
686 struct nfsnode *np = VTONFS(vp);
687 struct thread *td = curthread;
688 struct vattr *vap = ap->a_vap;
689 struct vattr vattr;
690 caddr_t bpos, dpos;
691 int error = 0;
692 struct mbuf *mreq, *mrep, *md, *mb;
693 int v3 = NFS_ISV3(vp);
694
695 /*
696 * Update local times for special files.
697 */
698 mtx_lock(&np->n_mtx);
699 if (np->n_flag & (NACC | NUPD))
700 np->n_flag |= NCHG;
701 mtx_unlock(&np->n_mtx);
702 /*
703 * First look in the cache.
704 */
705 if (nfs_getattrcache(vp, &vattr) == 0)
706 goto nfsmout;
707 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
708 nfsstats.accesscache_misses++;
709 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL);
710 if (nfs_getattrcache(vp, &vattr) == 0)
711 goto nfsmout;
712 }
713 nfsstats.rpccnt[NFSPROC_GETATTR]++;
714 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
715 mb = mreq;
716 bpos = mtod(mb, caddr_t);
717 nfsm_fhtom(vp, v3);
718 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
719 if (!error) {
720 nfsm_loadattr(vp, &vattr);
721 }
722 m_freem(mrep);
723 nfsmout:
724 vap->va_type = vattr.va_type;
725 vap->va_mode = vattr.va_mode;
726 vap->va_nlink = vattr.va_nlink;
727 vap->va_uid = vattr.va_uid;
728 vap->va_gid = vattr.va_gid;
729 vap->va_fsid = vattr.va_fsid;
730 vap->va_fileid = vattr.va_fileid;
731 vap->va_size = vattr.va_size;
732 vap->va_blocksize = vattr.va_blocksize;
733 vap->va_atime = vattr.va_atime;
734 vap->va_mtime = vattr.va_mtime;
735 vap->va_ctime = vattr.va_ctime;
736 vap->va_gen = vattr.va_gen;
737 vap->va_flags = vattr.va_flags;
738 vap->va_rdev = vattr.va_rdev;
739 vap->va_bytes = vattr.va_bytes;
740 vap->va_filerev = vattr.va_filerev;
741
742 return (error);
743 }
744
745 /*
746 * nfs setattr call.
747 */
748 static int
749 nfs_setattr(struct vop_setattr_args *ap)
750 {
751 struct vnode *vp = ap->a_vp;
752 struct nfsnode *np = VTONFS(vp);
753 struct vattr *vap = ap->a_vap;
754 struct thread *td = curthread;
755 int error = 0;
756 u_quad_t tsize;
757
758 #ifndef nolint
759 tsize = (u_quad_t)0;
760 #endif
761
762 /*
763 * Setting of flags is not supported.
764 */
765 if (vap->va_flags != VNOVAL)
766 return (EOPNOTSUPP);
767
768 /*
769 * Disallow write attempts if the filesystem is mounted read-only.
770 */
771 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
772 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
773 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
774 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
775 error = EROFS;
776 goto out;
777 }
778 if (vap->va_size != VNOVAL) {
779 switch (vp->v_type) {
780 case VDIR:
781 return (EISDIR);
782 case VCHR:
783 case VBLK:
784 case VSOCK:
785 case VFIFO:
786 if (vap->va_mtime.tv_sec == VNOVAL &&
787 vap->va_atime.tv_sec == VNOVAL &&
788 vap->va_mode == (mode_t)VNOVAL &&
789 vap->va_uid == (uid_t)VNOVAL &&
790 vap->va_gid == (gid_t)VNOVAL)
791 return (0);
792 vap->va_size = VNOVAL;
793 break;
794 default:
795 /*
796 * Disallow write attempts if the filesystem is
797 * mounted read-only.
798 */
799 if (vp->v_mount->mnt_flag & MNT_RDONLY)
800 return (EROFS);
801 /*
802 * We run vnode_pager_setsize() early (why?),
803 * we must set np->n_size now to avoid vinvalbuf
804 * V_SAVE races that might setsize a lower
805 * value.
806 */
807 mtx_lock(&np->n_mtx);
808 tsize = np->n_size;
809 mtx_unlock(&np->n_mtx);
810 error = nfs_meta_setsize(vp, ap->a_cred, td,
811 vap->va_size);
812 mtx_lock(&np->n_mtx);
813 if (np->n_flag & NMODIFIED) {
814 tsize = np->n_size;
815 mtx_unlock(&np->n_mtx);
816 if (vap->va_size == 0)
817 error = nfs_vinvalbuf(vp, 0, td, 1);
818 else
819 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
820 if (error) {
821 vnode_pager_setsize(vp, tsize);
822 goto out;
823 }
824 } else
825 mtx_unlock(&np->n_mtx);
826 /*
827 * np->n_size has already been set to vap->va_size
828 * in nfs_meta_setsize(). We must set it again since
829 * nfs_loadattrcache() could be called through
830 * nfs_meta_setsize() and could modify np->n_size.
831 */
832 mtx_lock(&np->n_mtx);
833 np->n_vattr.va_size = np->n_size = vap->va_size;
834 mtx_unlock(&np->n_mtx);
835 };
836 } else {
837 mtx_lock(&np->n_mtx);
838 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
839 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
840 mtx_unlock(&np->n_mtx);
841 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
842 (error == EINTR || error == EIO))
843 return error;
844 } else
845 mtx_unlock(&np->n_mtx);
846 }
847 error = nfs_setattrrpc(vp, vap, ap->a_cred);
848 if (error && vap->va_size != VNOVAL) {
849 mtx_lock(&np->n_mtx);
850 np->n_size = np->n_vattr.va_size = tsize;
851 vnode_pager_setsize(vp, tsize);
852 mtx_unlock(&np->n_mtx);
853 }
854 out:
855 return (error);
856 }
857
858 /*
859 * Do an nfs setattr rpc.
860 */
861 static int
862 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
863 {
864 struct nfsv2_sattr *sp;
865 struct nfsnode *np = VTONFS(vp);
866 caddr_t bpos, dpos;
867 u_int32_t *tl;
868 int error = 0, i, wccflag = NFSV3_WCCRATTR;
869 struct mbuf *mreq, *mrep, *md, *mb;
870 int v3 = NFS_ISV3(vp);
871
872 nfsstats.rpccnt[NFSPROC_SETATTR]++;
873 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
874 mb = mreq;
875 bpos = mtod(mb, caddr_t);
876 nfsm_fhtom(vp, v3);
877 if (v3) {
878 nfsm_v3attrbuild(vap, TRUE);
879 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
880 *tl = nfs_false;
881 } else {
882 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
883 if (vap->va_mode == (mode_t)VNOVAL)
884 sp->sa_mode = nfs_xdrneg1;
885 else
886 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
887 if (vap->va_uid == (uid_t)VNOVAL)
888 sp->sa_uid = nfs_xdrneg1;
889 else
890 sp->sa_uid = txdr_unsigned(vap->va_uid);
891 if (vap->va_gid == (gid_t)VNOVAL)
892 sp->sa_gid = nfs_xdrneg1;
893 else
894 sp->sa_gid = txdr_unsigned(vap->va_gid);
895 sp->sa_size = txdr_unsigned(vap->va_size);
896 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
897 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
898 }
899 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
900 if (v3) {
901 mtx_lock(&np->n_mtx);
902 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
903 np->n_accesscache[i].stamp = 0;
904 mtx_unlock(&np->n_mtx);
905 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
906 nfsm_wcc_data(vp, wccflag);
907 } else
908 nfsm_loadattr(vp, NULL);
909 m_freem(mrep);
910 nfsmout:
911 return (error);
912 }
913
914 /*
915 * nfs lookup call, one step at a time...
916 * First look in cache
917 * If not found, unlock the directory nfsnode and do the rpc
918 */
919 static int
920 nfs_lookup(struct vop_lookup_args *ap)
921 {
922 struct componentname *cnp = ap->a_cnp;
923 struct vnode *dvp = ap->a_dvp;
924 struct vnode **vpp = ap->a_vpp;
925 struct mount *mp = dvp->v_mount;
926 struct vattr vattr;
927 time_t dmtime;
928 int flags = cnp->cn_flags;
929 struct vnode *newvp;
930 struct nfsmount *nmp;
931 caddr_t bpos, dpos;
932 struct mbuf *mreq, *mrep, *md, *mb;
933 long len;
934 nfsfh_t *fhp;
935 struct nfsnode *np;
936 int error = 0, attrflag, fhsize, ltype;
937 int v3 = NFS_ISV3(dvp);
938 struct thread *td = cnp->cn_thread;
939
940 *vpp = NULLVP;
941 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
942 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
943 return (EROFS);
944 if (dvp->v_type != VDIR)
945 return (ENOTDIR);
946 nmp = VFSTONFS(mp);
947 np = VTONFS(dvp);
948 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
949 *vpp = NULLVP;
950 return (error);
951 }
952 error = cache_lookup(dvp, vpp, cnp);
953 if (error > 0 && error != ENOENT)
954 return (error);
955 if (error == -1) {
956 /*
957 * We only accept a positive hit in the cache if the
958 * change time of the file matches our cached copy.
959 * Otherwise, we discard the cache entry and fallback
960 * to doing a lookup RPC.
961 */
962 newvp = *vpp;
963 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
964 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
965 nfsstats.lookupcache_hits++;
966 if (cnp->cn_nameiop != LOOKUP &&
967 (flags & ISLASTCN))
968 cnp->cn_flags |= SAVENAME;
969 return (0);
970 }
971 cache_purge(newvp);
972 if (dvp != newvp)
973 vput(newvp);
974 else
975 vrele(newvp);
976 *vpp = NULLVP;
977 } else if (error == ENOENT) {
978 if (dvp->v_iflag & VI_DOOMED)
979 return (ENOENT);
980 /*
981 * We only accept a negative hit in the cache if the
982 * modification time of the parent directory matches
983 * our cached copy. Otherwise, we discard all of the
984 * negative cache entries for this directory.
985 */
986 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
987 vattr.va_mtime.tv_sec == np->n_dmtime) {
988 nfsstats.lookupcache_hits++;
989 return (ENOENT);
990 }
991 cache_purge_negative(dvp);
992 mtx_lock(&np->n_mtx);
993 np->n_dmtime = 0;
994 mtx_unlock(&np->n_mtx);
995 }
996
997 /*
998 * Cache the modification time of the parent directory in case
999 * the lookup fails and results in adding the first negative
1000 * name cache entry for the directory. Since this is reading
1001 * a single time_t, don't bother with locking. The
1002 * modification time may be a bit stale, but it must be read
1003 * before performing the lookup RPC to prevent a race where
1004 * another lookup updates the timestamp on the directory after
1005 * the lookup RPC has been performed on the server but before
1006 * n_dmtime is set at the end of this function.
1007 */
1008 dmtime = np->n_vattr.va_mtime.tv_sec;
1009 error = 0;
1010 newvp = NULLVP;
1011 nfsstats.lookupcache_misses++;
1012 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1013 len = cnp->cn_namelen;
1014 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1015 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1016 mb = mreq;
1017 bpos = mtod(mb, caddr_t);
1018 nfsm_fhtom(dvp, v3);
1019 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1020 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
1021 if (error) {
1022 if (v3) {
1023 nfsm_postop_attr(dvp, attrflag);
1024 m_freem(mrep);
1025 }
1026 goto nfsmout;
1027 }
1028 nfsm_getfh(fhp, fhsize, v3);
1029
1030 /*
1031 * Handle RENAME case...
1032 */
1033 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1034 if (NFS_CMPFH(np, fhp, fhsize)) {
1035 m_freem(mrep);
1036 return (EISDIR);
1037 }
1038 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE);
1039 if (error) {
1040 m_freem(mrep);
1041 return (error);
1042 }
1043 newvp = NFSTOV(np);
1044 if (v3) {
1045 nfsm_postop_attr(newvp, attrflag);
1046 nfsm_postop_attr(dvp, attrflag);
1047 } else
1048 nfsm_loadattr(newvp, NULL);
1049 *vpp = newvp;
1050 m_freem(mrep);
1051 cnp->cn_flags |= SAVENAME;
1052 return (0);
1053 }
1054
1055 if (flags & ISDOTDOT) {
1056 ltype = VOP_ISLOCKED(dvp);
1057 error = vfs_busy(mp, MBF_NOWAIT);
1058 if (error != 0) {
1059 vfs_ref(mp);
1060 VOP_UNLOCK(dvp, 0);
1061 error = vfs_busy(mp, 0);
1062 vn_lock(dvp, ltype | LK_RETRY);
1063 vfs_rel(mp);
1064 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1065 vfs_unbusy(mp);
1066 error = ENOENT;
1067 }
1068 if (error != 0) {
1069 m_freem(mrep);
1070 return (error);
1071 }
1072 }
1073 VOP_UNLOCK(dvp, 0);
1074 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1075 if (error == 0)
1076 newvp = NFSTOV(np);
1077 vfs_unbusy(mp);
1078 if (newvp != dvp)
1079 vn_lock(dvp, ltype | LK_RETRY);
1080 if (dvp->v_iflag & VI_DOOMED) {
1081 if (error == 0) {
1082 if (newvp == dvp)
1083 vrele(newvp);
1084 else
1085 vput(newvp);
1086 }
1087 error = ENOENT;
1088 }
1089 if (error) {
1090 m_freem(mrep);
1091 return (error);
1092 }
1093 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1094 VREF(dvp);
1095 newvp = dvp;
1096 } else {
1097 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1098 if (error) {
1099 m_freem(mrep);
1100 return (error);
1101 }
1102 newvp = NFSTOV(np);
1103 }
1104 if (v3) {
1105 nfsm_postop_attr(newvp, attrflag);
1106 nfsm_postop_attr(dvp, attrflag);
1107 } else
1108 nfsm_loadattr(newvp, NULL);
1109 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1110 cnp->cn_flags |= SAVENAME;
1111 if ((cnp->cn_flags & MAKEENTRY) &&
1112 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1113 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
1114 cache_enter(dvp, newvp, cnp);
1115 }
1116 *vpp = newvp;
1117 m_freem(mrep);
1118 nfsmout:
1119 if (error) {
1120 if (newvp != NULLVP) {
1121 vput(newvp);
1122 *vpp = NULLVP;
1123 }
1124
1125 if (error != ENOENT)
1126 goto done;
1127
1128 /* The requested file was not found. */
1129 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1130 (flags & ISLASTCN)) {
1131 /*
1132 * XXX: UFS does a full VOP_ACCESS(dvp,
1133 * VWRITE) here instead of just checking
1134 * MNT_RDONLY.
1135 */
1136 if (mp->mnt_flag & MNT_RDONLY)
1137 return (EROFS);
1138 cnp->cn_flags |= SAVENAME;
1139 return (EJUSTRETURN);
1140 }
1141
1142 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1143 /*
1144 * Maintain n_dmtime as the modification time
1145 * of the parent directory when the oldest -ve
1146 * name cache entry for this directory was
1147 * added. If a -ve cache entry has already
1148 * been added with a newer modification time
1149 * by a concurrent lookup, then don't bother
1150 * adding a cache entry. The modification
1151 * time of the directory might have changed
1152 * due to the file this lookup failed to find
1153 * being created. In that case a subsequent
1154 * lookup would incorrectly use the entry
1155 * added here instead of doing an extra
1156 * lookup.
1157 */
1158 mtx_lock(&np->n_mtx);
1159 if (np->n_dmtime <= dmtime) {
1160 if (np->n_dmtime == 0)
1161 np->n_dmtime = dmtime;
1162 mtx_unlock(&np->n_mtx);
1163 cache_enter(dvp, NULL, cnp);
1164 } else
1165 mtx_unlock(&np->n_mtx);
1166 }
1167 return (ENOENT);
1168 }
1169 done:
1170 return (error);
1171 }
1172
1173 /*
1174 * nfs read call.
1175 * Just call nfs_bioread() to do the work.
1176 */
1177 static int
1178 nfs_read(struct vop_read_args *ap)
1179 {
1180 struct vnode *vp = ap->a_vp;
1181
1182 switch (vp->v_type) {
1183 case VREG:
1184 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1185 case VDIR:
1186 return (EISDIR);
1187 default:
1188 return (EOPNOTSUPP);
1189 }
1190 }
1191
1192 /*
1193 * nfs readlink call
1194 */
1195 static int
1196 nfs_readlink(struct vop_readlink_args *ap)
1197 {
1198 struct vnode *vp = ap->a_vp;
1199
1200 if (vp->v_type != VLNK)
1201 return (EINVAL);
1202 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1203 }
1204
1205 /*
1206 * Do a readlink rpc.
1207 * Called by nfs_doio() from below the buffer cache.
1208 */
1209 int
1210 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1211 {
1212 caddr_t bpos, dpos;
1213 int error = 0, len, attrflag;
1214 struct mbuf *mreq, *mrep, *md, *mb;
1215 int v3 = NFS_ISV3(vp);
1216
1217 nfsstats.rpccnt[NFSPROC_READLINK]++;
1218 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1219 mb = mreq;
1220 bpos = mtod(mb, caddr_t);
1221 nfsm_fhtom(vp, v3);
1222 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1223 if (v3)
1224 nfsm_postop_attr(vp, attrflag);
1225 if (!error) {
1226 nfsm_strsiz(len, NFS_MAXPATHLEN);
1227 if (len == NFS_MAXPATHLEN) {
1228 struct nfsnode *np = VTONFS(vp);
1229 mtx_lock(&np->n_mtx);
1230 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1231 len = np->n_size;
1232 mtx_unlock(&np->n_mtx);
1233 }
1234 nfsm_mtouio(uiop, len);
1235 }
1236 m_freem(mrep);
1237 nfsmout:
1238 return (error);
1239 }
1240
1241 /*
1242 * nfs read rpc call
1243 * Ditto above
1244 */
1245 int
1246 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1247 {
1248 u_int32_t *tl;
1249 caddr_t bpos, dpos;
1250 struct mbuf *mreq, *mrep, *md, *mb;
1251 struct nfsmount *nmp;
1252 int error = 0, len, retlen, tsiz, eof, attrflag;
1253 int v3 = NFS_ISV3(vp);
1254 int rsize;
1255
1256 #ifndef nolint
1257 eof = 0;
1258 #endif
1259 nmp = VFSTONFS(vp->v_mount);
1260 tsiz = uiop->uio_resid;
1261 mtx_lock(&nmp->nm_mtx);
1262 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1263 mtx_unlock(&nmp->nm_mtx);
1264 return (EFBIG);
1265 }
1266 rsize = nmp->nm_rsize;
1267 mtx_unlock(&nmp->nm_mtx);
1268 while (tsiz > 0) {
1269 nfsstats.rpccnt[NFSPROC_READ]++;
1270 len = (tsiz > rsize) ? rsize : tsiz;
1271 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1272 mb = mreq;
1273 bpos = mtod(mb, caddr_t);
1274 nfsm_fhtom(vp, v3);
1275 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1276 if (v3) {
1277 txdr_hyper(uiop->uio_offset, tl);
1278 *(tl + 2) = txdr_unsigned(len);
1279 } else {
1280 *tl++ = txdr_unsigned(uiop->uio_offset);
1281 *tl++ = txdr_unsigned(len);
1282 *tl = 0;
1283 }
1284 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1285 if (v3) {
1286 nfsm_postop_attr(vp, attrflag);
1287 if (error) {
1288 m_freem(mrep);
1289 goto nfsmout;
1290 }
1291 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1292 eof = fxdr_unsigned(int, *(tl + 1));
1293 } else {
1294 nfsm_loadattr(vp, NULL);
1295 }
1296 nfsm_strsiz(retlen, rsize);
1297 nfsm_mtouio(uiop, retlen);
1298 m_freem(mrep);
1299 tsiz -= retlen;
1300 if (v3) {
1301 if (eof || retlen == 0) {
1302 tsiz = 0;
1303 }
1304 } else if (retlen < len) {
1305 tsiz = 0;
1306 }
1307 }
1308 nfsmout:
1309 return (error);
1310 }
1311
1312 /*
1313 * nfs write call
1314 */
1315 int
1316 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1317 int *iomode, int *must_commit)
1318 {
1319 u_int32_t *tl;
1320 int32_t backup;
1321 caddr_t bpos, dpos;
1322 struct mbuf *mreq, *mrep, *md, *mb;
1323 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1324 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1325 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1326 int wsize;
1327
1328 #ifndef DIAGNOSTIC
1329 if (uiop->uio_iovcnt != 1)
1330 panic("nfs: writerpc iovcnt > 1");
1331 #endif
1332 *must_commit = 0;
1333 tsiz = uiop->uio_resid;
1334 mtx_lock(&nmp->nm_mtx);
1335 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1336 mtx_unlock(&nmp->nm_mtx);
1337 return (EFBIG);
1338 }
1339 wsize = nmp->nm_wsize;
1340 mtx_unlock(&nmp->nm_mtx);
1341 while (tsiz > 0) {
1342 nfsstats.rpccnt[NFSPROC_WRITE]++;
1343 len = (tsiz > wsize) ? wsize : tsiz;
1344 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1345 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1346 mb = mreq;
1347 bpos = mtod(mb, caddr_t);
1348 nfsm_fhtom(vp, v3);
1349 if (v3) {
1350 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1351 txdr_hyper(uiop->uio_offset, tl);
1352 tl += 2;
1353 *tl++ = txdr_unsigned(len);
1354 *tl++ = txdr_unsigned(*iomode);
1355 *tl = txdr_unsigned(len);
1356 } else {
1357 u_int32_t x;
1358
1359 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1360 /* Set both "begin" and "current" to non-garbage. */
1361 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1362 *tl++ = x; /* "begin offset" */
1363 *tl++ = x; /* "current offset" */
1364 x = txdr_unsigned(len);
1365 *tl++ = x; /* total to this offset */
1366 *tl = x; /* size of this write */
1367 }
1368 nfsm_uiotom(uiop, len);
1369 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1370 if (v3) {
1371 wccflag = NFSV3_WCCCHK;
1372 nfsm_wcc_data(vp, wccflag);
1373 if (!error) {
1374 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1375 + NFSX_V3WRITEVERF);
1376 rlen = fxdr_unsigned(int, *tl++);
1377 if (rlen == 0) {
1378 error = NFSERR_IO;
1379 m_freem(mrep);
1380 break;
1381 } else if (rlen < len) {
1382 backup = len - rlen;
1383 uiop->uio_iov->iov_base =
1384 (char *)uiop->uio_iov->iov_base -
1385 backup;
1386 uiop->uio_iov->iov_len += backup;
1387 uiop->uio_offset -= backup;
1388 uiop->uio_resid += backup;
1389 len = rlen;
1390 }
1391 commit = fxdr_unsigned(int, *tl++);
1392
1393 /*
1394 * Return the lowest committment level
1395 * obtained by any of the RPCs.
1396 */
1397 if (committed == NFSV3WRITE_FILESYNC)
1398 committed = commit;
1399 else if (committed == NFSV3WRITE_DATASYNC &&
1400 commit == NFSV3WRITE_UNSTABLE)
1401 committed = commit;
1402 mtx_lock(&nmp->nm_mtx);
1403 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1404 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1405 NFSX_V3WRITEVERF);
1406 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1407 } else if (bcmp((caddr_t)tl,
1408 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1409 *must_commit = 1;
1410 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1411 NFSX_V3WRITEVERF);
1412 }
1413 mtx_unlock(&nmp->nm_mtx);
1414 }
1415 } else {
1416 nfsm_loadattr(vp, NULL);
1417 }
1418 if (wccflag) {
1419 mtx_lock(&(VTONFS(vp))->n_mtx);
1420 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1421 mtx_unlock(&(VTONFS(vp))->n_mtx);
1422 }
1423 m_freem(mrep);
1424 if (error)
1425 break;
1426 tsiz -= len;
1427 }
1428 nfsmout:
1429 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1430 committed = NFSV3WRITE_FILESYNC;
1431 *iomode = committed;
1432 if (error)
1433 uiop->uio_resid = tsiz;
1434 return (error);
1435 }
1436
1437 /*
1438 * nfs mknod rpc
1439 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1440 * mode set to specify the file type and the size field for rdev.
1441 */
1442 static int
1443 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1444 struct vattr *vap)
1445 {
1446 struct nfsv2_sattr *sp;
1447 u_int32_t *tl;
1448 struct vnode *newvp = NULL;
1449 struct nfsnode *np = NULL;
1450 struct vattr vattr;
1451 caddr_t bpos, dpos;
1452 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1453 struct mbuf *mreq, *mrep, *md, *mb;
1454 u_int32_t rdev;
1455 int v3 = NFS_ISV3(dvp);
1456
1457 if (vap->va_type == VCHR || vap->va_type == VBLK)
1458 rdev = txdr_unsigned(vap->va_rdev);
1459 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1460 rdev = nfs_xdrneg1;
1461 else {
1462 return (EOPNOTSUPP);
1463 }
1464 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1465 return (error);
1466 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1467 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1468 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1469 mb = mreq;
1470 bpos = mtod(mb, caddr_t);
1471 nfsm_fhtom(dvp, v3);
1472 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1473 if (v3) {
1474 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1475 *tl++ = vtonfsv3_type(vap->va_type);
1476 nfsm_v3attrbuild(vap, FALSE);
1477 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1478 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1479 *tl++ = txdr_unsigned(major(vap->va_rdev));
1480 *tl = txdr_unsigned(minor(vap->va_rdev));
1481 }
1482 } else {
1483 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1484 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1485 sp->sa_uid = nfs_xdrneg1;
1486 sp->sa_gid = nfs_xdrneg1;
1487 sp->sa_size = rdev;
1488 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1489 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1490 }
1491 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1492 if (!error) {
1493 nfsm_mtofh(dvp, newvp, v3, gotvp);
1494 if (!gotvp) {
1495 if (newvp) {
1496 vput(newvp);
1497 newvp = NULL;
1498 }
1499 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1500 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1501 if (!error)
1502 newvp = NFSTOV(np);
1503 }
1504 }
1505 if (v3)
1506 nfsm_wcc_data(dvp, wccflag);
1507 m_freem(mrep);
1508 nfsmout:
1509 if (error) {
1510 if (newvp)
1511 vput(newvp);
1512 } else {
1513 if (cnp->cn_flags & MAKEENTRY)
1514 cache_enter(dvp, newvp, cnp);
1515 *vpp = newvp;
1516 }
1517 mtx_lock(&(VTONFS(dvp))->n_mtx);
1518 VTONFS(dvp)->n_flag |= NMODIFIED;
1519 if (!wccflag) {
1520 VTONFS(dvp)->n_attrstamp = 0;
1521 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1522 }
1523 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1524 return (error);
1525 }
1526
1527 /*
1528 * nfs mknod vop
1529 * just call nfs_mknodrpc() to do the work.
1530 */
1531 /* ARGSUSED */
1532 static int
1533 nfs_mknod(struct vop_mknod_args *ap)
1534 {
1535 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1536 }
1537
1538 static u_long create_verf;
1539 /*
1540 * nfs file create call
1541 */
1542 static int
1543 nfs_create(struct vop_create_args *ap)
1544 {
1545 struct vnode *dvp = ap->a_dvp;
1546 struct vattr *vap = ap->a_vap;
1547 struct componentname *cnp = ap->a_cnp;
1548 struct nfsv2_sattr *sp;
1549 u_int32_t *tl;
1550 struct nfsnode *np = NULL;
1551 struct vnode *newvp = NULL;
1552 caddr_t bpos, dpos;
1553 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1554 struct mbuf *mreq, *mrep, *md, *mb;
1555 struct vattr vattr;
1556 int v3 = NFS_ISV3(dvp);
1557
1558 /*
1559 * Oops, not for me..
1560 */
1561 if (vap->va_type == VSOCK)
1562 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1563
1564 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1565 return (error);
1566 if (vap->va_vaflags & VA_EXCLUSIVE)
1567 fmode |= O_EXCL;
1568 again:
1569 nfsstats.rpccnt[NFSPROC_CREATE]++;
1570 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1571 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1572 mb = mreq;
1573 bpos = mtod(mb, caddr_t);
1574 nfsm_fhtom(dvp, v3);
1575 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1576 if (v3) {
1577 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1578 if (fmode & O_EXCL) {
1579 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1580 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1581 #ifdef INET
1582 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1583 IN_IFADDR_RLOCK();
1584 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1585 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1586 else
1587 #endif
1588 *tl++ = create_verf;
1589 #ifdef INET
1590 IN_IFADDR_RUNLOCK();
1591 CURVNET_RESTORE();
1592 #endif
1593 *tl = ++create_verf;
1594 } else {
1595 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1596 nfsm_v3attrbuild(vap, FALSE);
1597 }
1598 } else {
1599 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1600 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1601 sp->sa_uid = nfs_xdrneg1;
1602 sp->sa_gid = nfs_xdrneg1;
1603 sp->sa_size = 0;
1604 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1605 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1606 }
1607 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1608 if (!error) {
1609 nfsm_mtofh(dvp, newvp, v3, gotvp);
1610 if (!gotvp) {
1611 if (newvp) {
1612 vput(newvp);
1613 newvp = NULL;
1614 }
1615 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1616 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1617 if (!error)
1618 newvp = NFSTOV(np);
1619 }
1620 }
1621 if (v3)
1622 nfsm_wcc_data(dvp, wccflag);
1623 m_freem(mrep);
1624 nfsmout:
1625 if (error) {
1626 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1627 fmode &= ~O_EXCL;
1628 goto again;
1629 }
1630 if (newvp)
1631 vput(newvp);
1632 } else if (v3 && (fmode & O_EXCL)) {
1633 /*
1634 * We are normally called with only a partially initialized
1635 * VAP. Since the NFSv3 spec says that server may use the
1636 * file attributes to store the verifier, the spec requires
1637 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1638 * in atime, but we can't really assume that all servers will
1639 * so we ensure that our SETATTR sets both atime and mtime.
1640 */
1641 if (vap->va_mtime.tv_sec == VNOVAL)
1642 vfs_timestamp(&vap->va_mtime);
1643 if (vap->va_atime.tv_sec == VNOVAL)
1644 vap->va_atime = vap->va_mtime;
1645 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred);
1646 if (error)
1647 vput(newvp);
1648 }
1649 if (!error) {
1650 if (cnp->cn_flags & MAKEENTRY)
1651 cache_enter(dvp, newvp, cnp);
1652 *ap->a_vpp = newvp;
1653 }
1654 mtx_lock(&(VTONFS(dvp))->n_mtx);
1655 VTONFS(dvp)->n_flag |= NMODIFIED;
1656 if (!wccflag) {
1657 VTONFS(dvp)->n_attrstamp = 0;
1658 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1659 }
1660 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1661 return (error);
1662 }
1663
1664 /*
1665 * nfs file remove call
1666 * To try and make nfs semantics closer to ufs semantics, a file that has
1667 * other processes using the vnode is renamed instead of removed and then
1668 * removed later on the last close.
1669 * - If v_usecount > 1
1670 * If a rename is not already in the works
1671 * call nfs_sillyrename() to set it up
1672 * else
1673 * do the remove rpc
1674 */
1675 static int
1676 nfs_remove(struct vop_remove_args *ap)
1677 {
1678 struct vnode *vp = ap->a_vp;
1679 struct vnode *dvp = ap->a_dvp;
1680 struct componentname *cnp = ap->a_cnp;
1681 struct nfsnode *np = VTONFS(vp);
1682 int error = 0;
1683 struct vattr vattr;
1684
1685 #ifndef DIAGNOSTIC
1686 if ((cnp->cn_flags & HASBUF) == 0)
1687 panic("nfs_remove: no name");
1688 if (vrefcnt(vp) < 1)
1689 panic("nfs_remove: bad v_usecount");
1690 #endif
1691 if (vp->v_type == VDIR)
1692 error = EPERM;
1693 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1694 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1695 /*
1696 * Purge the name cache so that the chance of a lookup for
1697 * the name succeeding while the remove is in progress is
1698 * minimized. Without node locking it can still happen, such
1699 * that an I/O op returns ESTALE, but since you get this if
1700 * another host removes the file..
1701 */
1702 cache_purge(vp);
1703 /*
1704 * throw away biocache buffers, mainly to avoid
1705 * unnecessary delayed writes later.
1706 */
1707 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1708 /* Do the rpc */
1709 if (error != EINTR && error != EIO)
1710 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1711 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1712 /*
1713 * Kludge City: If the first reply to the remove rpc is lost..
1714 * the reply to the retransmitted request will be ENOENT
1715 * since the file was in fact removed
1716 * Therefore, we cheat and return success.
1717 */
1718 if (error == ENOENT)
1719 error = 0;
1720 } else if (!np->n_sillyrename)
1721 error = nfs_sillyrename(dvp, vp, cnp);
1722 np->n_attrstamp = 0;
1723 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1724 return (error);
1725 }
1726
1727 /*
1728 * nfs file remove rpc called from nfs_inactive
1729 */
1730 int
1731 nfs_removeit(struct sillyrename *sp)
1732 {
1733 /*
1734 * Make sure that the directory vnode is still valid.
1735 * XXX we should lock sp->s_dvp here.
1736 */
1737 if (sp->s_dvp->v_type == VBAD)
1738 return (0);
1739 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1740 NULL));
1741 }
1742
1743 /*
1744 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1745 */
1746 static int
1747 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1748 struct ucred *cred, struct thread *td)
1749 {
1750 caddr_t bpos, dpos;
1751 int error = 0, wccflag = NFSV3_WCCRATTR;
1752 struct mbuf *mreq, *mrep, *md, *mb;
1753 int v3 = NFS_ISV3(dvp);
1754
1755 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1756 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1757 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1758 mb = mreq;
1759 bpos = mtod(mb, caddr_t);
1760 nfsm_fhtom(dvp, v3);
1761 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1762 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1763 if (v3)
1764 nfsm_wcc_data(dvp, wccflag);
1765 m_freem(mrep);
1766 nfsmout:
1767 mtx_lock(&(VTONFS(dvp))->n_mtx);
1768 VTONFS(dvp)->n_flag |= NMODIFIED;
1769 if (!wccflag) {
1770 VTONFS(dvp)->n_attrstamp = 0;
1771 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1772 }
1773 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1774 return (error);
1775 }
1776
1777 /*
1778 * nfs file rename call
1779 */
1780 static int
1781 nfs_rename(struct vop_rename_args *ap)
1782 {
1783 struct vnode *fvp = ap->a_fvp;
1784 struct vnode *tvp = ap->a_tvp;
1785 struct vnode *fdvp = ap->a_fdvp;
1786 struct vnode *tdvp = ap->a_tdvp;
1787 struct componentname *tcnp = ap->a_tcnp;
1788 struct componentname *fcnp = ap->a_fcnp;
1789 int error;
1790
1791 #ifndef DIAGNOSTIC
1792 if ((tcnp->cn_flags & HASBUF) == 0 ||
1793 (fcnp->cn_flags & HASBUF) == 0)
1794 panic("nfs_rename: no name");
1795 #endif
1796 /* Check for cross-device rename */
1797 if ((fvp->v_mount != tdvp->v_mount) ||
1798 (tvp && (fvp->v_mount != tvp->v_mount))) {
1799 error = EXDEV;
1800 goto out;
1801 }
1802
1803 if (fvp == tvp) {
1804 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1805 error = 0;
1806 goto out;
1807 }
1808 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1809 goto out;
1810
1811 /*
1812 * We have to flush B_DELWRI data prior to renaming
1813 * the file. If we don't, the delayed-write buffers
1814 * can be flushed out later after the file has gone stale
1815 * under NFSV3. NFSV2 does not have this problem because
1816 * ( as far as I can tell ) it flushes dirty buffers more
1817 * often.
1818 *
1819 * Skip the rename operation if the fsync fails, this can happen
1820 * due to the server's volume being full, when we pushed out data
1821 * that was written back to our cache earlier. Not checking for
1822 * this condition can result in potential (silent) data loss.
1823 */
1824 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1825 VOP_UNLOCK(fvp, 0);
1826 if (!error && tvp)
1827 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1828 if (error)
1829 goto out;
1830
1831 /*
1832 * If the tvp exists and is in use, sillyrename it before doing the
1833 * rename of the new file over it.
1834 * XXX Can't sillyrename a directory.
1835 */
1836 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1837 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1838 vput(tvp);
1839 tvp = NULL;
1840 }
1841
1842 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1843 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1844 tcnp->cn_thread);
1845
1846 if (fvp->v_type == VDIR) {
1847 if (tvp != NULL && tvp->v_type == VDIR)
1848 cache_purge(tdvp);
1849 cache_purge(fdvp);
1850 }
1851
1852 out:
1853 if (tdvp == tvp)
1854 vrele(tdvp);
1855 else
1856 vput(tdvp);
1857 if (tvp)
1858 vput(tvp);
1859 vrele(fdvp);
1860 vrele(fvp);
1861 /*
1862 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1863 */
1864 if (error == ENOENT)
1865 error = 0;
1866 return (error);
1867 }
1868
1869 /*
1870 * nfs file rename rpc called from nfs_remove() above
1871 */
1872 static int
1873 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1874 struct sillyrename *sp)
1875 {
1876
1877 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1878 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1879 }
1880
1881 /*
1882 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1883 */
1884 static int
1885 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1886 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1887 struct thread *td)
1888 {
1889 caddr_t bpos, dpos;
1890 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1891 struct mbuf *mreq, *mrep, *md, *mb;
1892 int v3 = NFS_ISV3(fdvp);
1893
1894 nfsstats.rpccnt[NFSPROC_RENAME]++;
1895 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1896 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1897 nfsm_rndup(tnamelen));
1898 mb = mreq;
1899 bpos = mtod(mb, caddr_t);
1900 nfsm_fhtom(fdvp, v3);
1901 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1902 nfsm_fhtom(tdvp, v3);
1903 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1904 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1905 if (v3) {
1906 nfsm_wcc_data(fdvp, fwccflag);
1907 nfsm_wcc_data(tdvp, twccflag);
1908 }
1909 m_freem(mrep);
1910 nfsmout:
1911 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1912 VTONFS(fdvp)->n_flag |= NMODIFIED;
1913 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1914 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1915 VTONFS(tdvp)->n_flag |= NMODIFIED;
1916 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1917 if (!fwccflag) {
1918 VTONFS(fdvp)->n_attrstamp = 0;
1919 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1920 }
1921 if (!twccflag) {
1922 VTONFS(tdvp)->n_attrstamp = 0;
1923 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1924 }
1925 return (error);
1926 }
1927
1928 /*
1929 * nfs hard link create call
1930 */
1931 static int
1932 nfs_link(struct vop_link_args *ap)
1933 {
1934 struct vnode *vp = ap->a_vp;
1935 struct vnode *tdvp = ap->a_tdvp;
1936 struct componentname *cnp = ap->a_cnp;
1937 caddr_t bpos, dpos;
1938 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1939 struct mbuf *mreq, *mrep, *md, *mb;
1940 int v3;
1941
1942 if (vp->v_mount != tdvp->v_mount) {
1943 return (EXDEV);
1944 }
1945
1946 /*
1947 * Push all writes to the server, so that the attribute cache
1948 * doesn't get "out of sync" with the server.
1949 * XXX There should be a better way!
1950 */
1951 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1952
1953 v3 = NFS_ISV3(vp);
1954 nfsstats.rpccnt[NFSPROC_LINK]++;
1955 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1956 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1957 mb = mreq;
1958 bpos = mtod(mb, caddr_t);
1959 nfsm_fhtom(vp, v3);
1960 nfsm_fhtom(tdvp, v3);
1961 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1962 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1963 if (v3) {
1964 nfsm_postop_attr(vp, attrflag);
1965 nfsm_wcc_data(tdvp, wccflag);
1966 }
1967 m_freem(mrep);
1968 nfsmout:
1969 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1970 VTONFS(tdvp)->n_flag |= NMODIFIED;
1971 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1972 if (!attrflag) {
1973 VTONFS(vp)->n_attrstamp = 0;
1974 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1975 }
1976 if (!wccflag) {
1977 VTONFS(tdvp)->n_attrstamp = 0;
1978 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1979 }
1980 return (error);
1981 }
1982
1983 /*
1984 * nfs symbolic link create call
1985 */
1986 static int
1987 nfs_symlink(struct vop_symlink_args *ap)
1988 {
1989 struct vnode *dvp = ap->a_dvp;
1990 struct vattr *vap = ap->a_vap;
1991 struct componentname *cnp = ap->a_cnp;
1992 struct nfsv2_sattr *sp;
1993 caddr_t bpos, dpos;
1994 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1995 struct mbuf *mreq, *mrep, *md, *mb;
1996 struct vnode *newvp = NULL;
1997 int v3 = NFS_ISV3(dvp);
1998
1999 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2000 slen = strlen(ap->a_target);
2001 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2002 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2003 mb = mreq;
2004 bpos = mtod(mb, caddr_t);
2005 nfsm_fhtom(dvp, v3);
2006 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2007 if (v3) {
2008 nfsm_v3attrbuild(vap, FALSE);
2009 }
2010 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2011 if (!v3) {
2012 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2013 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2014 sp->sa_uid = nfs_xdrneg1;
2015 sp->sa_gid = nfs_xdrneg1;
2016 sp->sa_size = nfs_xdrneg1;
2017 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2018 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2019 }
2020
2021 /*
2022 * Issue the NFS request and get the rpc response.
2023 *
2024 * Only NFSv3 responses returning an error of 0 actually return
2025 * a file handle that can be converted into newvp without having
2026 * to do an extra lookup rpc.
2027 */
2028 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
2029 if (v3) {
2030 if (error == 0)
2031 nfsm_mtofh(dvp, newvp, v3, gotvp);
2032 nfsm_wcc_data(dvp, wccflag);
2033 }
2034
2035 /*
2036 * out code jumps -> here, mrep is also freed.
2037 */
2038
2039 m_freem(mrep);
2040 nfsmout:
2041
2042 /*
2043 * If we do not have an error and we could not extract the newvp from
2044 * the response due to the request being NFSv2, we have to do a
2045 * lookup in order to obtain a newvp to return.
2046 */
2047 if (error == 0 && newvp == NULL) {
2048 struct nfsnode *np = NULL;
2049
2050 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2051 cnp->cn_cred, cnp->cn_thread, &np);
2052 if (!error)
2053 newvp = NFSTOV(np);
2054 }
2055 if (error) {
2056 if (newvp)
2057 vput(newvp);
2058 } else {
2059 *ap->a_vpp = newvp;
2060 }
2061 mtx_lock(&(VTONFS(dvp))->n_mtx);
2062 VTONFS(dvp)->n_flag |= NMODIFIED;
2063 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2064 if (!wccflag) {
2065 VTONFS(dvp)->n_attrstamp = 0;
2066 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2067 }
2068 return (error);
2069 }
2070
2071 /*
2072 * nfs make dir call
2073 */
2074 static int
2075 nfs_mkdir(struct vop_mkdir_args *ap)
2076 {
2077 struct vnode *dvp = ap->a_dvp;
2078 struct vattr *vap = ap->a_vap;
2079 struct componentname *cnp = ap->a_cnp;
2080 struct nfsv2_sattr *sp;
2081 int len;
2082 struct nfsnode *np = NULL;
2083 struct vnode *newvp = NULL;
2084 caddr_t bpos, dpos;
2085 int error = 0, wccflag = NFSV3_WCCRATTR;
2086 int gotvp = 0;
2087 struct mbuf *mreq, *mrep, *md, *mb;
2088 struct vattr vattr;
2089 int v3 = NFS_ISV3(dvp);
2090
2091 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2092 return (error);
2093 len = cnp->cn_namelen;
2094 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2095 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
2096 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2097 mb = mreq;
2098 bpos = mtod(mb, caddr_t);
2099 nfsm_fhtom(dvp, v3);
2100 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2101 if (v3) {
2102 nfsm_v3attrbuild(vap, FALSE);
2103 } else {
2104 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2105 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2106 sp->sa_uid = nfs_xdrneg1;
2107 sp->sa_gid = nfs_xdrneg1;
2108 sp->sa_size = nfs_xdrneg1;
2109 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2110 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2111 }
2112 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2113 if (!error)
2114 nfsm_mtofh(dvp, newvp, v3, gotvp);
2115 if (v3)
2116 nfsm_wcc_data(dvp, wccflag);
2117 m_freem(mrep);
2118 nfsmout:
2119 mtx_lock(&(VTONFS(dvp))->n_mtx);
2120 VTONFS(dvp)->n_flag |= NMODIFIED;
2121 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2122 if (!wccflag) {
2123 VTONFS(dvp)->n_attrstamp = 0;
2124 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2125 }
2126 if (error == 0 && newvp == NULL) {
2127 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2128 cnp->cn_thread, &np);
2129 if (!error) {
2130 newvp = NFSTOV(np);
2131 if (newvp->v_type != VDIR)
2132 error = EEXIST;
2133 }
2134 }
2135 if (error) {
2136 if (newvp)
2137 vput(newvp);
2138 } else
2139 *ap->a_vpp = newvp;
2140 return (error);
2141 }
2142
2143 /*
2144 * nfs remove directory call
2145 */
2146 static int
2147 nfs_rmdir(struct vop_rmdir_args *ap)
2148 {
2149 struct vnode *vp = ap->a_vp;
2150 struct vnode *dvp = ap->a_dvp;
2151 struct componentname *cnp = ap->a_cnp;
2152 caddr_t bpos, dpos;
2153 int error = 0, wccflag = NFSV3_WCCRATTR;
2154 struct mbuf *mreq, *mrep, *md, *mb;
2155 int v3 = NFS_ISV3(dvp);
2156
2157 if (dvp == vp)
2158 return (EINVAL);
2159 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2160 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
2161 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2162 mb = mreq;
2163 bpos = mtod(mb, caddr_t);
2164 nfsm_fhtom(dvp, v3);
2165 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2166 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2167 if (v3)
2168 nfsm_wcc_data(dvp, wccflag);
2169 m_freem(mrep);
2170 nfsmout:
2171 mtx_lock(&(VTONFS(dvp))->n_mtx);
2172 VTONFS(dvp)->n_flag |= NMODIFIED;
2173 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2174 if (!wccflag) {
2175 VTONFS(dvp)->n_attrstamp = 0;
2176 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2177 }
2178 cache_purge(dvp);
2179 cache_purge(vp);
2180 /*
2181 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2182 */
2183 if (error == ENOENT)
2184 error = 0;
2185 return (error);
2186 }
2187
2188 /*
2189 * nfs readdir call
2190 */
2191 static int
2192 nfs_readdir(struct vop_readdir_args *ap)
2193 {
2194 struct vnode *vp = ap->a_vp;
2195 struct nfsnode *np = VTONFS(vp);
2196 struct uio *uio = ap->a_uio;
2197 int tresid, error = 0;
2198 struct vattr vattr;
2199
2200 if (vp->v_type != VDIR)
2201 return(EPERM);
2202
2203 /*
2204 * First, check for hit on the EOF offset cache
2205 */
2206 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2207 (np->n_flag & NMODIFIED) == 0) {
2208 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2209 mtx_lock(&np->n_mtx);
2210 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2211 mtx_unlock(&np->n_mtx);
2212 nfsstats.direofcache_hits++;
2213 goto out;
2214 } else
2215 mtx_unlock(&np->n_mtx);
2216 }
2217 }
2218
2219 /*
2220 * Call nfs_bioread() to do the real work.
2221 */
2222 tresid = uio->uio_resid;
2223 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2224
2225 if (!error && uio->uio_resid == tresid) {
2226 nfsstats.direofcache_misses++;
2227 }
2228 out:
2229 return (error);
2230 }
2231
2232 /*
2233 * Readdir rpc call.
2234 * Called from below the buffer cache by nfs_doio().
2235 */
2236 int
2237 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2238 {
2239 int len, left;
2240 struct dirent *dp = NULL;
2241 u_int32_t *tl;
2242 caddr_t cp;
2243 nfsuint64 *cookiep;
2244 caddr_t bpos, dpos;
2245 struct mbuf *mreq, *mrep, *md, *mb;
2246 nfsuint64 cookie;
2247 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2248 struct nfsnode *dnp = VTONFS(vp);
2249 u_quad_t fileno;
2250 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2251 int attrflag;
2252 int v3 = NFS_ISV3(vp);
2253
2254 #ifndef DIAGNOSTIC
2255 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2256 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2257 panic("nfs readdirrpc bad uio");
2258 #endif
2259
2260 /*
2261 * If there is no cookie, assume directory was stale.
2262 */
2263 nfs_dircookie_lock(dnp);
2264 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2265 if (cookiep) {
2266 cookie = *cookiep;
2267 nfs_dircookie_unlock(dnp);
2268 } else {
2269 nfs_dircookie_unlock(dnp);
2270 return (NFSERR_BAD_COOKIE);
2271 }
2272
2273 /*
2274 * Loop around doing readdir rpc's of size nm_readdirsize
2275 * truncated to a multiple of DIRBLKSIZ.
2276 * The stopping criteria is EOF or buffer full.
2277 */
2278 while (more_dirs && bigenough) {
2279 nfsstats.rpccnt[NFSPROC_READDIR]++;
2280 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2281 NFSX_READDIR(v3));
2282 mb = mreq;
2283 bpos = mtod(mb, caddr_t);
2284 nfsm_fhtom(vp, v3);
2285 if (v3) {
2286 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2287 *tl++ = cookie.nfsuquad[0];
2288 *tl++ = cookie.nfsuquad[1];
2289 mtx_lock(&dnp->n_mtx);
2290 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2291 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2292 mtx_unlock(&dnp->n_mtx);
2293 } else {
2294 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2295 *tl++ = cookie.nfsuquad[0];
2296 }
2297 *tl = txdr_unsigned(nmp->nm_readdirsize);
2298 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2299 if (v3) {
2300 nfsm_postop_attr(vp, attrflag);
2301 if (!error) {
2302 tl = nfsm_dissect(u_int32_t *,
2303 2 * NFSX_UNSIGNED);
2304 mtx_lock(&dnp->n_mtx);
2305 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2306 dnp->n_cookieverf.nfsuquad[1] = *tl;
2307 mtx_unlock(&dnp->n_mtx);
2308 } else {
2309 m_freem(mrep);
2310 goto nfsmout;
2311 }
2312 }
2313 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2314 more_dirs = fxdr_unsigned(int, *tl);
2315
2316 /* loop thru the dir entries, doctoring them to 4bsd form */
2317 while (more_dirs && bigenough) {
2318 if (v3) {
2319 tl = nfsm_dissect(u_int32_t *,
2320 3 * NFSX_UNSIGNED);
2321 fileno = fxdr_hyper(tl);
2322 len = fxdr_unsigned(int, *(tl + 2));
2323 } else {
2324 tl = nfsm_dissect(u_int32_t *,
2325 2 * NFSX_UNSIGNED);
2326 fileno = fxdr_unsigned(u_quad_t, *tl++);
2327 len = fxdr_unsigned(int, *tl);
2328 }
2329 if (len <= 0 || len > NFS_MAXNAMLEN) {
2330 error = EBADRPC;
2331 m_freem(mrep);
2332 goto nfsmout;
2333 }
2334 tlen = nfsm_rndup(len);
2335 if (tlen == len)
2336 tlen += 4; /* To ensure null termination */
2337 left = DIRBLKSIZ - blksiz;
2338 if ((tlen + DIRHDSIZ) > left) {
2339 dp->d_reclen += left;
2340 uiop->uio_iov->iov_base =
2341 (char *)uiop->uio_iov->iov_base + left;
2342 uiop->uio_iov->iov_len -= left;
2343 uiop->uio_offset += left;
2344 uiop->uio_resid -= left;
2345 blksiz = 0;
2346 }
2347 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2348 bigenough = 0;
2349 if (bigenough) {
2350 dp = (struct dirent *)uiop->uio_iov->iov_base;
2351 dp->d_fileno = (int)fileno;
2352 dp->d_namlen = len;
2353 dp->d_reclen = tlen + DIRHDSIZ;
2354 dp->d_type = DT_UNKNOWN;
2355 blksiz += dp->d_reclen;
2356 if (blksiz == DIRBLKSIZ)
2357 blksiz = 0;
2358 uiop->uio_offset += DIRHDSIZ;
2359 uiop->uio_resid -= DIRHDSIZ;
2360 uiop->uio_iov->iov_base =
2361 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2362 uiop->uio_iov->iov_len -= DIRHDSIZ;
2363 nfsm_mtouio(uiop, len);
2364 cp = uiop->uio_iov->iov_base;
2365 tlen -= len;
2366 *cp = '\0'; /* null terminate */
2367 uiop->uio_iov->iov_base =
2368 (char *)uiop->uio_iov->iov_base + tlen;
2369 uiop->uio_iov->iov_len -= tlen;
2370 uiop->uio_offset += tlen;
2371 uiop->uio_resid -= tlen;
2372 } else
2373 nfsm_adv(nfsm_rndup(len));
2374 if (v3) {
2375 tl = nfsm_dissect(u_int32_t *,
2376 3 * NFSX_UNSIGNED);
2377 } else {
2378 tl = nfsm_dissect(u_int32_t *,
2379 2 * NFSX_UNSIGNED);
2380 }
2381 if (bigenough) {
2382 cookie.nfsuquad[0] = *tl++;
2383 if (v3)
2384 cookie.nfsuquad[1] = *tl++;
2385 } else if (v3)
2386 tl += 2;
2387 else
2388 tl++;
2389 more_dirs = fxdr_unsigned(int, *tl);
2390 }
2391 /*
2392 * If at end of rpc data, get the eof boolean
2393 */
2394 if (!more_dirs) {
2395 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2396 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2397 }
2398 m_freem(mrep);
2399 }
2400 /*
2401 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2402 * by increasing d_reclen for the last record.
2403 */
2404 if (blksiz > 0) {
2405 left = DIRBLKSIZ - blksiz;
2406 dp->d_reclen += left;
2407 uiop->uio_iov->iov_base =
2408 (char *)uiop->uio_iov->iov_base + left;
2409 uiop->uio_iov->iov_len -= left;
2410 uiop->uio_offset += left;
2411 uiop->uio_resid -= left;
2412 }
2413
2414 /*
2415 * We are now either at the end of the directory or have filled the
2416 * block.
2417 */
2418 if (bigenough)
2419 dnp->n_direofoffset = uiop->uio_offset;
2420 else {
2421 if (uiop->uio_resid > 0)
2422 nfs_printf("EEK! readdirrpc resid > 0\n");
2423 nfs_dircookie_lock(dnp);
2424 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2425 *cookiep = cookie;
2426 nfs_dircookie_unlock(dnp);
2427 }
2428 nfsmout:
2429 return (error);
2430 }
2431
2432 /*
2433 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2434 */
2435 int
2436 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2437 {
2438 int len, left;
2439 struct dirent *dp;
2440 u_int32_t *tl;
2441 caddr_t cp;
2442 struct vnode *newvp;
2443 nfsuint64 *cookiep;
2444 caddr_t bpos, dpos, dpossav1, dpossav2;
2445 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2446 struct nameidata nami, *ndp = &nami;
2447 struct componentname *cnp = &ndp->ni_cnd;
2448 nfsuint64 cookie;
2449 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2450 struct nfsnode *dnp = VTONFS(vp), *np;
2451 nfsfh_t *fhp;
2452 u_quad_t fileno;
2453 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2454 int attrflag, fhsize;
2455
2456 #ifndef nolint
2457 dp = NULL;
2458 #endif
2459 #ifndef DIAGNOSTIC
2460 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2461 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2462 panic("nfs readdirplusrpc bad uio");
2463 #endif
2464 ndp->ni_dvp = vp;
2465 newvp = NULLVP;
2466
2467 /*
2468 * If there is no cookie, assume directory was stale.
2469 */
2470 nfs_dircookie_lock(dnp);
2471 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2472 if (cookiep) {
2473 cookie = *cookiep;
2474 nfs_dircookie_unlock(dnp);
2475 } else {
2476 nfs_dircookie_unlock(dnp);
2477 return (NFSERR_BAD_COOKIE);
2478 }
2479 /*
2480 * Loop around doing readdir rpc's of size nm_readdirsize
2481 * truncated to a multiple of DIRBLKSIZ.
2482 * The stopping criteria is EOF or buffer full.
2483 */
2484 while (more_dirs && bigenough) {
2485 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2486 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2487 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2488 mb = mreq;
2489 bpos = mtod(mb, caddr_t);
2490 nfsm_fhtom(vp, 1);
2491 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2492 *tl++ = cookie.nfsuquad[0];
2493 *tl++ = cookie.nfsuquad[1];
2494 mtx_lock(&dnp->n_mtx);
2495 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2496 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2497 mtx_unlock(&dnp->n_mtx);
2498 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2499 *tl = txdr_unsigned(nmp->nm_rsize);
2500 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2501 nfsm_postop_attr(vp, attrflag);
2502 if (error) {
2503 m_freem(mrep);
2504 goto nfsmout;
2505 }
2506 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2507 mtx_lock(&dnp->n_mtx);
2508 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2509 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2510 mtx_unlock(&dnp->n_mtx);
2511 more_dirs = fxdr_unsigned(int, *tl);
2512
2513 /* loop thru the dir entries, doctoring them to 4bsd form */
2514 while (more_dirs && bigenough) {
2515 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2516 fileno = fxdr_hyper(tl);
2517 len = fxdr_unsigned(int, *(tl + 2));
2518 if (len <= 0 || len > NFS_MAXNAMLEN) {
2519 error = EBADRPC;
2520 m_freem(mrep);
2521 goto nfsmout;
2522 }
2523 tlen = nfsm_rndup(len);
2524 if (tlen == len)
2525 tlen += 4; /* To ensure null termination*/
2526 left = DIRBLKSIZ - blksiz;
2527 if ((tlen + DIRHDSIZ) > left) {
2528 dp->d_reclen += left;
2529 uiop->uio_iov->iov_base =
2530 (char *)uiop->uio_iov->iov_base + left;
2531 uiop->uio_iov->iov_len -= left;
2532 uiop->uio_offset += left;
2533 uiop->uio_resid -= left;
2534 blksiz = 0;
2535 }
2536 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2537 bigenough = 0;
2538 if (bigenough) {
2539 dp = (struct dirent *)uiop->uio_iov->iov_base;
2540 dp->d_fileno = (int)fileno;
2541 dp->d_namlen = len;
2542 dp->d_reclen = tlen + DIRHDSIZ;
2543 dp->d_type = DT_UNKNOWN;
2544 blksiz += dp->d_reclen;
2545 if (blksiz == DIRBLKSIZ)
2546 blksiz = 0;
2547 uiop->uio_offset += DIRHDSIZ;
2548 uiop->uio_resid -= DIRHDSIZ;
2549 uiop->uio_iov->iov_base =
2550 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2551 uiop->uio_iov->iov_len -= DIRHDSIZ;
2552 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2553 cnp->cn_namelen = len;
2554 nfsm_mtouio(uiop, len);
2555 cp = uiop->uio_iov->iov_base;
2556 tlen -= len;
2557 *cp = '\0';
2558 uiop->uio_iov->iov_base =
2559 (char *)uiop->uio_iov->iov_base + tlen;
2560 uiop->uio_iov->iov_len -= tlen;
2561 uiop->uio_offset += tlen;
2562 uiop->uio_resid -= tlen;
2563 } else
2564 nfsm_adv(nfsm_rndup(len));
2565 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2566 if (bigenough) {
2567 cookie.nfsuquad[0] = *tl++;
2568 cookie.nfsuquad[1] = *tl++;
2569 } else
2570 tl += 2;
2571
2572 /*
2573 * Since the attributes are before the file handle
2574 * (sigh), we must skip over the attributes and then
2575 * come back and get them.
2576 */
2577 attrflag = fxdr_unsigned(int, *tl);
2578 if (attrflag) {
2579 dpossav1 = dpos;
2580 mdsav1 = md;
2581 nfsm_adv(NFSX_V3FATTR);
2582 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2583 doit = fxdr_unsigned(int, *tl);
2584 /*
2585 * Skip loading the attrs for "..". There's a
2586 * race between loading the attrs here and
2587 * lookups that look for the directory currently
2588 * being read (in the parent). We try to acquire
2589 * the exclusive lock on ".." here, owning the
2590 * lock on the directory being read. Lookup will
2591 * hold the lock on ".." and try to acquire the
2592 * lock on the directory being read.
2593 *
2594 * There are other ways of fixing this, one would
2595 * be to do a trylock on the ".." vnode and skip
2596 * loading the attrs on ".." if it happens to be
2597 * locked by another process. But skipping the
2598 * attrload on ".." seems the easiest option.
2599 */
2600 if (strcmp(dp->d_name, "..") == 0) {
2601 doit = 0;
2602 /*
2603 * We've already skipped over the attrs,
2604 * skip over the filehandle. And store d_type
2605 * as VDIR.
2606 */
2607 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2608 i = fxdr_unsigned(int, *tl);
2609 nfsm_adv(nfsm_rndup(i));
2610 dp->d_type = IFTODT(VTTOIF(VDIR));
2611 }
2612 if (doit) {
2613 nfsm_getfh(fhp, fhsize, 1);
2614 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2615 VREF(vp);
2616 newvp = vp;
2617 np = dnp;
2618 } else {
2619 error = nfs_nget(vp->v_mount, fhp,
2620 fhsize, &np, LK_EXCLUSIVE);
2621 if (error)
2622 doit = 0;
2623 else
2624 newvp = NFSTOV(np);
2625 }
2626 }
2627 if (doit && bigenough) {
2628 dpossav2 = dpos;
2629 dpos = dpossav1;
2630 mdsav2 = md;
2631 md = mdsav1;
2632 nfsm_loadattr(newvp, NULL);
2633 dpos = dpossav2;
2634 md = mdsav2;
2635 dp->d_type =
2636 IFTODT(VTTOIF(np->n_vattr.va_type));
2637 ndp->ni_vp = newvp;
2638 /* Update n_ctime, so subsequent lookup doesn't purge entry */
2639 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
2640 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2641 }
2642 } else {
2643 /* Just skip over the file handle */
2644 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2645 i = fxdr_unsigned(int, *tl);
2646 if (i) {
2647 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2648 fhsize = fxdr_unsigned(int, *tl);
2649 nfsm_adv(nfsm_rndup(fhsize));
2650 }
2651 }
2652 if (newvp != NULLVP) {
2653 if (newvp == vp)
2654 vrele(newvp);
2655 else
2656 vput(newvp);
2657 newvp = NULLVP;
2658 }
2659 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2660 more_dirs = fxdr_unsigned(int, *tl);
2661 }
2662 /*
2663 * If at end of rpc data, get the eof boolean
2664 */
2665 if (!more_dirs) {
2666 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2667 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2668 }
2669 m_freem(mrep);
2670 }
2671 /*
2672 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2673 * by increasing d_reclen for the last record.
2674 */
2675 if (blksiz > 0) {
2676 left = DIRBLKSIZ - blksiz;
2677 dp->d_reclen += left;
2678 uiop->uio_iov->iov_base =
2679 (char *)uiop->uio_iov->iov_base + left;
2680 uiop->uio_iov->iov_len -= left;
2681 uiop->uio_offset += left;
2682 uiop->uio_resid -= left;
2683 }
2684
2685 /*
2686 * We are now either at the end of the directory or have filled the
2687 * block.
2688 */
2689 if (bigenough)
2690 dnp->n_direofoffset = uiop->uio_offset;
2691 else {
2692 if (uiop->uio_resid > 0)
2693 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2694 nfs_dircookie_lock(dnp);
2695 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2696 *cookiep = cookie;
2697 nfs_dircookie_unlock(dnp);
2698 }
2699 nfsmout:
2700 if (newvp != NULLVP) {
2701 if (newvp == vp)
2702 vrele(newvp);
2703 else
2704 vput(newvp);
2705 newvp = NULLVP;
2706 }
2707 return (error);
2708 }
2709
2710 /*
2711 * Silly rename. To make the NFS filesystem that is stateless look a little
2712 * more like the "ufs" a remove of an active vnode is translated to a rename
2713 * to a funny looking filename that is removed by nfs_inactive on the
2714 * nfsnode. There is the potential for another process on a different client
2715 * to create the same funny name between the nfs_lookitup() fails and the
2716 * nfs_rename() completes, but...
2717 */
2718 static int
2719 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2720 {
2721 struct sillyrename *sp;
2722 struct nfsnode *np;
2723 int error;
2724 short pid;
2725 unsigned int lticks;
2726
2727 cache_purge(dvp);
2728 np = VTONFS(vp);
2729 #ifndef DIAGNOSTIC
2730 if (vp->v_type == VDIR)
2731 panic("nfs: sillyrename dir");
2732 #endif
2733 sp = malloc(sizeof (struct sillyrename),
2734 M_NFSREQ, M_WAITOK);
2735 sp->s_cred = crhold(cnp->cn_cred);
2736 sp->s_dvp = dvp;
2737 sp->s_removeit = nfs_removeit;
2738 VREF(dvp);
2739
2740 /*
2741 * Fudge together a funny name.
2742 * Changing the format of the funny name to accomodate more
2743 * sillynames per directory.
2744 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2745 * CPU ticks since boot.
2746 */
2747 pid = cnp->cn_thread->td_proc->p_pid;
2748 lticks = (unsigned int)ticks;
2749 for ( ; ; ) {
2750 sp->s_namlen = sprintf(sp->s_name,
2751 ".nfs.%08x.%04x4.4", lticks,
2752 pid);
2753 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2754 cnp->cn_thread, NULL))
2755 break;
2756 lticks++;
2757 }
2758 error = nfs_renameit(dvp, cnp, sp);
2759 if (error)
2760 goto bad;
2761 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2762 cnp->cn_thread, &np);
2763 np->n_sillyrename = sp;
2764 return (0);
2765 bad:
2766 vrele(sp->s_dvp);
2767 crfree(sp->s_cred);
2768 free((caddr_t)sp, M_NFSREQ);
2769 return (error);
2770 }
2771
2772 /*
2773 * Look up a file name and optionally either update the file handle or
2774 * allocate an nfsnode, depending on the value of npp.
2775 * npp == NULL --> just do the lookup
2776 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2777 * handled too
2778 * *npp != NULL --> update the file handle in the vnode
2779 */
2780 static int
2781 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2782 struct thread *td, struct nfsnode **npp)
2783 {
2784 struct vnode *newvp = NULL;
2785 struct nfsnode *np, *dnp = VTONFS(dvp);
2786 caddr_t bpos, dpos;
2787 int error = 0, fhlen, attrflag;
2788 struct mbuf *mreq, *mrep, *md, *mb;
2789 nfsfh_t *nfhp;
2790 int v3 = NFS_ISV3(dvp);
2791
2792 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2793 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2794 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2795 mb = mreq;
2796 bpos = mtod(mb, caddr_t);
2797 nfsm_fhtom(dvp, v3);
2798 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2799 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2800 if (npp && !error) {
2801 nfsm_getfh(nfhp, fhlen, v3);
2802 if (*npp) {
2803 np = *npp;
2804 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2805 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2806 np->n_fhp = &np->n_fh;
2807 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2808 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2809 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2810 np->n_fhsize = fhlen;
2811 newvp = NFSTOV(np);
2812 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2813 VREF(dvp);
2814 newvp = dvp;
2815 } else {
2816 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2817 if (error) {
2818 m_freem(mrep);
2819 return (error);
2820 }
2821 newvp = NFSTOV(np);
2822 }
2823 if (v3) {
2824 nfsm_postop_attr(newvp, attrflag);
2825 if (!attrflag && *npp == NULL) {
2826 m_freem(mrep);
2827 if (newvp == dvp)
2828 vrele(newvp);
2829 else
2830 vput(newvp);
2831 return (ENOENT);
2832 }
2833 } else
2834 nfsm_loadattr(newvp, NULL);
2835 }
2836 m_freem(mrep);
2837 nfsmout:
2838 if (npp && *npp == NULL) {
2839 if (error) {
2840 if (newvp) {
2841 if (newvp == dvp)
2842 vrele(newvp);
2843 else
2844 vput(newvp);
2845 }
2846 } else
2847 *npp = np;
2848 }
2849 return (error);
2850 }
2851
2852 /*
2853 * Nfs Version 3 commit rpc
2854 */
2855 int
2856 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2857 struct thread *td)
2858 {
2859 u_int32_t *tl;
2860 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2861 caddr_t bpos, dpos;
2862 int error = 0, wccflag = NFSV3_WCCRATTR;
2863 struct mbuf *mreq, *mrep, *md, *mb;
2864
2865 mtx_lock(&nmp->nm_mtx);
2866 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2867 mtx_unlock(&nmp->nm_mtx);
2868 return (0);
2869 }
2870 mtx_unlock(&nmp->nm_mtx);
2871 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2872 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2873 mb = mreq;
2874 bpos = mtod(mb, caddr_t);
2875 nfsm_fhtom(vp, 1);
2876 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2877 txdr_hyper(offset, tl);
2878 tl += 2;
2879 *tl = txdr_unsigned(cnt);
2880 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2881 nfsm_wcc_data(vp, wccflag);
2882 if (!error) {
2883 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2884 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2885 NFSX_V3WRITEVERF)) {
2886 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2887 NFSX_V3WRITEVERF);
2888 error = NFSERR_STALEWRITEVERF;
2889 }
2890 }
2891 m_freem(mrep);
2892 nfsmout:
2893 return (error);
2894 }
2895
2896 /*
2897 * Strategy routine.
2898 * For async requests when nfsiod(s) are running, queue the request by
2899 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2900 * request.
2901 */
2902 static int
2903 nfs_strategy(struct vop_strategy_args *ap)
2904 {
2905 struct buf *bp = ap->a_bp;
2906 struct ucred *cr;
2907
2908 KASSERT(!(bp->b_flags & B_DONE),
2909 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2910 BUF_ASSERT_HELD(bp);
2911
2912 if (bp->b_iocmd == BIO_READ)
2913 cr = bp->b_rcred;
2914 else
2915 cr = bp->b_wcred;
2916
2917 /*
2918 * If the op is asynchronous and an i/o daemon is waiting
2919 * queue the request, wake it up and wait for completion
2920 * otherwise just do it ourselves.
2921 */
2922 if ((bp->b_flags & B_ASYNC) == 0 ||
2923 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2924 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2925 return (0);
2926 }
2927
2928 /*
2929 * fsync vnode op. Just call nfs_flush() with commit == 1.
2930 */
2931 /* ARGSUSED */
2932 static int
2933 nfs_fsync(struct vop_fsync_args *ap)
2934 {
2935
2936 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1));
2937 }
2938
2939 /*
2940 * Flush all the blocks associated with a vnode.
2941 * Walk through the buffer pool and push any dirty pages
2942 * associated with the vnode.
2943 */
2944 static int
2945 nfs_flush(struct vnode *vp, int waitfor, int commit)
2946 {
2947 struct nfsnode *np = VTONFS(vp);
2948 struct buf *bp;
2949 int i;
2950 struct buf *nbp;
2951 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2952 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2953 int passone = 1;
2954 u_quad_t off, endoff, toff;
2955 struct ucred* wcred = NULL;
2956 struct buf **bvec = NULL;
2957 struct bufobj *bo;
2958 struct thread *td = curthread;
2959 #ifndef NFS_COMMITBVECSIZ
2960 #define NFS_COMMITBVECSIZ 20
2961 #endif
2962 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2963 int bvecsize = 0, bveccount;
2964
2965 if (nmp->nm_flag & NFSMNT_INT)
2966 slpflag = NFS_PCATCH;
2967 if (!commit)
2968 passone = 0;
2969 bo = &vp->v_bufobj;
2970 /*
2971 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2972 * server, but has not been committed to stable storage on the server
2973 * yet. On the first pass, the byte range is worked out and the commit
2974 * rpc is done. On the second pass, nfs_writebp() is called to do the
2975 * job.
2976 */
2977 again:
2978 off = (u_quad_t)-1;
2979 endoff = 0;
2980 bvecpos = 0;
2981 if (NFS_ISV3(vp) && commit) {
2982 if (bvec != NULL && bvec != bvec_on_stack)
2983 free(bvec, M_TEMP);
2984 /*
2985 * Count up how many buffers waiting for a commit.
2986 */
2987 bveccount = 0;
2988 BO_LOCK(bo);
2989 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2990 if (!BUF_ISLOCKED(bp) &&
2991 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2992 == (B_DELWRI | B_NEEDCOMMIT))
2993 bveccount++;
2994 }
2995 /*
2996 * Allocate space to remember the list of bufs to commit. It is
2997 * important to use M_NOWAIT here to avoid a race with nfs_write.
2998 * If we can't get memory (for whatever reason), we will end up
2999 * committing the buffers one-by-one in the loop below.
3000 */
3001 if (bveccount > NFS_COMMITBVECSIZ) {
3002 /*
3003 * Release the vnode interlock to avoid a lock
3004 * order reversal.
3005 */
3006 BO_UNLOCK(bo);
3007 bvec = (struct buf **)
3008 malloc(bveccount * sizeof(struct buf *),
3009 M_TEMP, M_NOWAIT);
3010 BO_LOCK(bo);
3011 if (bvec == NULL) {
3012 bvec = bvec_on_stack;
3013 bvecsize = NFS_COMMITBVECSIZ;
3014 } else
3015 bvecsize = bveccount;
3016 } else {
3017 bvec = bvec_on_stack;
3018 bvecsize = NFS_COMMITBVECSIZ;
3019 }
3020 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3021 if (bvecpos >= bvecsize)
3022 break;
3023 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3024 nbp = TAILQ_NEXT(bp, b_bobufs);
3025 continue;
3026 }
3027 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3028 (B_DELWRI | B_NEEDCOMMIT)) {
3029 BUF_UNLOCK(bp);
3030 nbp = TAILQ_NEXT(bp, b_bobufs);
3031 continue;
3032 }
3033 BO_UNLOCK(bo);
3034 bremfree(bp);
3035 /*
3036 * Work out if all buffers are using the same cred
3037 * so we can deal with them all with one commit.
3038 *
3039 * NOTE: we are not clearing B_DONE here, so we have
3040 * to do it later on in this routine if we intend to
3041 * initiate I/O on the bp.
3042 *
3043 * Note: to avoid loopback deadlocks, we do not
3044 * assign b_runningbufspace.
3045 */
3046 if (wcred == NULL)
3047 wcred = bp->b_wcred;
3048 else if (wcred != bp->b_wcred)
3049 wcred = NOCRED;
3050 vfs_busy_pages(bp, 1);
3051
3052 BO_LOCK(bo);
3053 /*
3054 * bp is protected by being locked, but nbp is not
3055 * and vfs_busy_pages() may sleep. We have to
3056 * recalculate nbp.
3057 */
3058 nbp = TAILQ_NEXT(bp, b_bobufs);
3059
3060 /*
3061 * A list of these buffers is kept so that the
3062 * second loop knows which buffers have actually
3063 * been committed. This is necessary, since there
3064 * may be a race between the commit rpc and new
3065 * uncommitted writes on the file.
3066 */
3067 bvec[bvecpos++] = bp;
3068 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3069 bp->b_dirtyoff;
3070 if (toff < off)
3071 off = toff;
3072 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3073 if (toff > endoff)
3074 endoff = toff;
3075 }
3076 BO_UNLOCK(bo);
3077 }
3078 if (bvecpos > 0) {
3079 /*
3080 * Commit data on the server, as required.
3081 * If all bufs are using the same wcred, then use that with
3082 * one call for all of them, otherwise commit each one
3083 * separately.
3084 */
3085 if (wcred != NOCRED)
3086 retv = nfs_commit(vp, off, (int)(endoff - off),
3087 wcred, td);
3088 else {
3089 retv = 0;
3090 for (i = 0; i < bvecpos; i++) {
3091 off_t off, size;
3092 bp = bvec[i];
3093 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3094 bp->b_dirtyoff;
3095 size = (u_quad_t)(bp->b_dirtyend
3096 - bp->b_dirtyoff);
3097 retv = nfs_commit(vp, off, (int)size,
3098 bp->b_wcred, td);
3099 if (retv) break;
3100 }
3101 }
3102
3103 if (retv == NFSERR_STALEWRITEVERF)
3104 nfs_clearcommit(vp->v_mount);
3105
3106 /*
3107 * Now, either mark the blocks I/O done or mark the
3108 * blocks dirty, depending on whether the commit
3109 * succeeded.
3110 */
3111 for (i = 0; i < bvecpos; i++) {
3112 bp = bvec[i];
3113 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3114 if (retv) {
3115 /*
3116 * Error, leave B_DELWRI intact
3117 */
3118 vfs_unbusy_pages(bp);
3119 brelse(bp);
3120 } else {
3121 /*
3122 * Success, remove B_DELWRI ( bundirty() ).
3123 *
3124 * b_dirtyoff/b_dirtyend seem to be NFS
3125 * specific. We should probably move that
3126 * into bundirty(). XXX
3127 */
3128 bufobj_wref(bo);
3129 bp->b_flags |= B_ASYNC;
3130 bundirty(bp);
3131 bp->b_flags &= ~B_DONE;
3132 bp->b_ioflags &= ~BIO_ERROR;
3133 bp->b_dirtyoff = bp->b_dirtyend = 0;
3134 bufdone(bp);
3135 }
3136 }
3137 }
3138
3139 /*
3140 * Start/do any write(s) that are required.
3141 */
3142 loop:
3143 BO_LOCK(bo);
3144 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3145 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3146 if (waitfor != MNT_WAIT || passone)
3147 continue;
3148
3149 error = BUF_TIMELOCK(bp,
3150 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3151 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
3152 if (error == 0) {
3153 BUF_UNLOCK(bp);
3154 goto loop;
3155 }
3156 if (error == ENOLCK) {
3157 error = 0;
3158 goto loop;
3159 }
3160 if (nfs_sigintr(nmp, td)) {
3161 error = EINTR;
3162 goto done;
3163 }
3164 if (slpflag & PCATCH) {
3165 slpflag = 0;
3166 slptimeo = 2 * hz;
3167 }
3168 goto loop;
3169 }
3170 if ((bp->b_flags & B_DELWRI) == 0)
3171 panic("nfs_fsync: not dirty");
3172 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3173 BUF_UNLOCK(bp);
3174 continue;
3175 }
3176 BO_UNLOCK(bo);
3177 bremfree(bp);
3178 if (passone || !commit)
3179 bp->b_flags |= B_ASYNC;
3180 else
3181 bp->b_flags |= B_ASYNC;
3182 bwrite(bp);
3183 if (nfs_sigintr(nmp, td)) {
3184 error = EINTR;
3185 goto done;
3186 }
3187 goto loop;
3188 }
3189 if (passone) {
3190 passone = 0;
3191 BO_UNLOCK(bo);
3192 goto again;
3193 }
3194 if (waitfor == MNT_WAIT) {
3195 while (bo->bo_numoutput) {
3196 error = bufobj_wwait(bo, slpflag, slptimeo);
3197 if (error) {
3198 BO_UNLOCK(bo);
3199 error = nfs_sigintr(nmp, td);
3200 if (error)
3201 goto done;
3202 if (slpflag & PCATCH) {
3203 slpflag = 0;
3204 slptimeo = 2 * hz;
3205 }
3206 BO_LOCK(bo);
3207 }
3208 }
3209 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3210 BO_UNLOCK(bo);
3211 goto loop;
3212 }
3213 /*
3214 * Wait for all the async IO requests to drain
3215 */
3216 BO_UNLOCK(bo);
3217 mtx_lock(&np->n_mtx);
3218 while (np->n_directio_asyncwr > 0) {
3219 np->n_flag |= NFSYNCWAIT;
3220 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3221 &np->n_mtx, slpflag | (PRIBIO + 1),
3222 "nfsfsync", 0);
3223 if (error) {
3224 if (nfs_sigintr(nmp, td)) {
3225 mtx_unlock(&np->n_mtx);
3226 error = EINTR;
3227 goto done;
3228 }
3229 }
3230 }
3231 mtx_unlock(&np->n_mtx);
3232 } else
3233 BO_UNLOCK(bo);
3234 mtx_lock(&np->n_mtx);
3235 if (np->n_flag & NWRITEERR) {
3236 error = np->n_error;
3237 np->n_flag &= ~NWRITEERR;
3238 }
3239 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3240 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3241 np->n_flag &= ~NMODIFIED;
3242 mtx_unlock(&np->n_mtx);
3243 done:
3244 if (bvec != NULL && bvec != bvec_on_stack)
3245 free(bvec, M_TEMP);
3246 return (error);
3247 }
3248
3249 /*
3250 * NFS advisory byte-level locks.
3251 */
3252 static int
3253 nfs_advlock(struct vop_advlock_args *ap)
3254 {
3255 struct vnode *vp = ap->a_vp;
3256 u_quad_t size;
3257 int error;
3258
3259 error = vn_lock(vp, LK_SHARED);
3260 if (error)
3261 return (error);
3262 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3263 size = VTONFS(vp)->n_size;
3264 VOP_UNLOCK(vp, 0);
3265 error = lf_advlock(ap, &(vp->v_lockf), size);
3266 } else {
3267 if (nfs_advlock_p)
3268 error = nfs_advlock_p(ap);
3269 else
3270 error = ENOLCK;
3271 }
3272
3273 return (error);
3274 }
3275
3276 /*
3277 * NFS advisory byte-level locks.
3278 */
3279 static int
3280 nfs_advlockasync(struct vop_advlockasync_args *ap)
3281 {
3282 struct vnode *vp = ap->a_vp;
3283 u_quad_t size;
3284 int error;
3285
3286 error = vn_lock(vp, LK_SHARED);
3287 if (error)
3288 return (error);
3289 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3290 size = VTONFS(vp)->n_size;
3291 VOP_UNLOCK(vp, 0);
3292 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3293 } else {
3294 VOP_UNLOCK(vp, 0);
3295 error = EOPNOTSUPP;
3296 }
3297 return (error);
3298 }
3299
3300 /*
3301 * Print out the contents of an nfsnode.
3302 */
3303 static int
3304 nfs_print(struct vop_print_args *ap)
3305 {
3306 struct vnode *vp = ap->a_vp;
3307 struct nfsnode *np = VTONFS(vp);
3308
3309 nfs_printf("\tfileid %ld fsid 0x%x",
3310 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3311 if (vp->v_type == VFIFO)
3312 fifo_printinfo(vp);
3313 printf("\n");
3314 return (0);
3315 }
3316
3317 /*
3318 * This is the "real" nfs::bwrite(struct buf*).
3319 * We set B_CACHE if this is a VMIO buffer.
3320 */
3321 int
3322 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3323 {
3324 int s;
3325 int oldflags = bp->b_flags;
3326 #if 0
3327 int retv = 1;
3328 off_t off;
3329 #endif
3330
3331 BUF_ASSERT_HELD(bp);
3332
3333 if (bp->b_flags & B_INVAL) {
3334 brelse(bp);
3335 return(0);
3336 }
3337
3338 bp->b_flags |= B_CACHE;
3339
3340 /*
3341 * Undirty the bp. We will redirty it later if the I/O fails.
3342 */
3343
3344 s = splbio();
3345 bundirty(bp);
3346 bp->b_flags &= ~B_DONE;
3347 bp->b_ioflags &= ~BIO_ERROR;
3348 bp->b_iocmd = BIO_WRITE;
3349
3350 bufobj_wref(bp->b_bufobj);
3351 curthread->td_ru.ru_oublock++;
3352 splx(s);
3353
3354 /*
3355 * Note: to avoid loopback deadlocks, we do not
3356 * assign b_runningbufspace.
3357 */
3358 vfs_busy_pages(bp, 1);
3359
3360 BUF_KERNPROC(bp);
3361 bp->b_iooffset = dbtob(bp->b_blkno);
3362 bstrategy(bp);
3363
3364 if( (oldflags & B_ASYNC) == 0) {
3365 int rtval = bufwait(bp);
3366
3367 if (oldflags & B_DELWRI) {
3368 s = splbio();
3369 reassignbuf(bp);
3370 splx(s);
3371 }
3372 brelse(bp);
3373 return (rtval);
3374 }
3375
3376 return (0);
3377 }
3378
3379 /*
3380 * nfs special file access vnode op.
3381 * Essentially just get vattr and then imitate iaccess() since the device is
3382 * local to the client.
3383 */
3384 static int
3385 nfsspec_access(struct vop_access_args *ap)
3386 {
3387 struct vattr *vap;
3388 struct ucred *cred = ap->a_cred;
3389 struct vnode *vp = ap->a_vp;
3390 accmode_t accmode = ap->a_accmode;
3391 struct vattr vattr;
3392 int error;
3393
3394 /*
3395 * Disallow write attempts on filesystems mounted read-only;
3396 * unless the file is a socket, fifo, or a block or character
3397 * device resident on the filesystem.
3398 */
3399 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3400 switch (vp->v_type) {
3401 case VREG:
3402 case VDIR:
3403 case VLNK:
3404 return (EROFS);
3405 default:
3406 break;
3407 }
3408 }
3409 vap = &vattr;
3410 error = VOP_GETATTR(vp, vap, cred);
3411 if (error)
3412 goto out;
3413 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3414 accmode, cred, NULL);
3415 out:
3416 return error;
3417 }
3418
3419 /*
3420 * Read wrapper for fifos.
3421 */
3422 static int
3423 nfsfifo_read(struct vop_read_args *ap)
3424 {
3425 struct nfsnode *np = VTONFS(ap->a_vp);
3426 int error;
3427
3428 /*
3429 * Set access flag.
3430 */
3431 mtx_lock(&np->n_mtx);
3432 np->n_flag |= NACC;
3433 getnanotime(&np->n_atim);
3434 mtx_unlock(&np->n_mtx);
3435 error = fifo_specops.vop_read(ap);
3436 return error;
3437 }
3438
3439 /*
3440 * Write wrapper for fifos.
3441 */
3442 static int
3443 nfsfifo_write(struct vop_write_args *ap)
3444 {
3445 struct nfsnode *np = VTONFS(ap->a_vp);
3446
3447 /*
3448 * Set update flag.
3449 */
3450 mtx_lock(&np->n_mtx);
3451 np->n_flag |= NUPD;
3452 getnanotime(&np->n_mtim);
3453 mtx_unlock(&np->n_mtx);
3454 return(fifo_specops.vop_write(ap));
3455 }
3456
3457 /*
3458 * Close wrapper for fifos.
3459 *
3460 * Update the times on the nfsnode then do fifo close.
3461 */
3462 static int
3463 nfsfifo_close(struct vop_close_args *ap)
3464 {
3465 struct vnode *vp = ap->a_vp;
3466 struct nfsnode *np = VTONFS(vp);
3467 struct vattr vattr;
3468 struct timespec ts;
3469
3470 mtx_lock(&np->n_mtx);
3471 if (np->n_flag & (NACC | NUPD)) {
3472 getnanotime(&ts);
3473 if (np->n_flag & NACC)
3474 np->n_atim = ts;
3475 if (np->n_flag & NUPD)
3476 np->n_mtim = ts;
3477 np->n_flag |= NCHG;
3478 if (vrefcnt(vp) == 1 &&
3479 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3480 VATTR_NULL(&vattr);
3481 if (np->n_flag & NACC)
3482 vattr.va_atime = np->n_atim;
3483 if (np->n_flag & NUPD)
3484 vattr.va_mtime = np->n_mtim;
3485 mtx_unlock(&np->n_mtx);
3486 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3487 goto out;
3488 }
3489 }
3490 mtx_unlock(&np->n_mtx);
3491 out:
3492 return (fifo_specops.vop_close(ap));
3493 }
3494
3495 /*
3496 * Just call nfs_writebp() with the force argument set to 1.
3497 *
3498 * NOTE: B_DONE may or may not be set in a_bp on call.
3499 */
3500 static int
3501 nfs_bwrite(struct buf *bp)
3502 {
3503
3504 return (nfs_writebp(bp, 1, curthread));
3505 }
3506
3507 struct buf_ops buf_ops_nfs = {
3508 .bop_name = "buf_ops_nfs",
3509 .bop_write = nfs_bwrite,
3510 .bop_strategy = bufstrategy,
3511 .bop_sync = bufsync,
3512 .bop_bdflush = bufbdflush,
3513 };
Cache object: fbe795f1f272936630fa753e2dfda117
|