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/6.3/sys/nfsclient/nfs_vnops.c 170553 2007-06-11 10:53:48Z kib $");
37
38 /*
39 * vnode op calls for Sun NFS version 2 and 3
40 */
41
42 #include "opt_inet.h"
43
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/resourcevar.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/namei.h>
55 #include <sys/socket.h>
56 #include <sys/vnode.h>
57 #include <sys/dirent.h>
58 #include <sys/fcntl.h>
59 #include <sys/lockf.h>
60 #include <sys/stat.h>
61 #include <sys/sysctl.h>
62 #include <sys/signalvar.h>
63
64 #include <vm/vm.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_extern.h>
67 #include <vm/vm_object.h>
68
69 #include <fs/fifofs/fifo.h>
70
71 #include <rpc/rpcclnt.h>
72
73 #include <nfs/rpcv2.h>
74 #include <nfs/nfsproto.h>
75 #include <nfsclient/nfs.h>
76 #include <nfsclient/nfsnode.h>
77 #include <nfsclient/nfsmount.h>
78 #include <nfsclient/nfs_lock.h>
79 #include <nfs/xdr_subs.h>
80 #include <nfsclient/nfsm_subs.h>
81
82 #include <net/if.h>
83 #include <netinet/in.h>
84 #include <netinet/in_var.h>
85
86 /* Defs */
87 #define TRUE 1
88 #define FALSE 0
89
90 /*
91 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
92 * calls are not in getblk() and brelse() so that they would not be necessary
93 * here.
94 */
95 #ifndef B_VMIO
96 #define vfs_busy_pages(bp, f)
97 #endif
98
99 static vop_read_t nfsfifo_read;
100 static vop_write_t nfsfifo_write;
101 static vop_close_t nfsfifo_close;
102 static int nfs_flush(struct vnode *, int, struct thread *,
103 int);
104 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
105 struct thread *);
106 static vop_lookup_t nfs_lookup;
107 static vop_create_t nfs_create;
108 static vop_mknod_t nfs_mknod;
109 static vop_open_t nfs_open;
110 static vop_close_t nfs_close;
111 static vop_access_t nfs_access;
112 static vop_getattr_t nfs_getattr;
113 static vop_setattr_t nfs_setattr;
114 static vop_read_t nfs_read;
115 static vop_fsync_t nfs_fsync;
116 static vop_remove_t nfs_remove;
117 static vop_link_t nfs_link;
118 static vop_rename_t nfs_rename;
119 static vop_mkdir_t nfs_mkdir;
120 static vop_rmdir_t nfs_rmdir;
121 static vop_symlink_t nfs_symlink;
122 static vop_readdir_t nfs_readdir;
123 static vop_strategy_t nfs_strategy;
124 static int nfs_lookitup(struct vnode *, const char *, int,
125 struct ucred *, struct thread *, struct nfsnode **);
126 static int nfs_sillyrename(struct vnode *, struct vnode *,
127 struct componentname *);
128 static vop_access_t nfsspec_access;
129 static vop_readlink_t nfs_readlink;
130 static vop_print_t nfs_print;
131 static vop_advlock_t nfs_advlock;
132
133 /*
134 * Global vfs data structures for nfs
135 */
136 struct vop_vector nfs_vnodeops = {
137 .vop_default = &default_vnodeops,
138 .vop_access = nfs_access,
139 .vop_advlock = nfs_advlock,
140 .vop_close = nfs_close,
141 .vop_create = nfs_create,
142 .vop_fsync = nfs_fsync,
143 .vop_getattr = nfs_getattr,
144 .vop_getpages = nfs_getpages,
145 .vop_putpages = nfs_putpages,
146 .vop_inactive = nfs_inactive,
147 .vop_lease = VOP_NULL,
148 .vop_link = nfs_link,
149 .vop_lookup = nfs_lookup,
150 .vop_mkdir = nfs_mkdir,
151 .vop_mknod = nfs_mknod,
152 .vop_open = nfs_open,
153 .vop_print = nfs_print,
154 .vop_read = nfs_read,
155 .vop_readdir = nfs_readdir,
156 .vop_readlink = nfs_readlink,
157 .vop_reclaim = nfs_reclaim,
158 .vop_remove = nfs_remove,
159 .vop_rename = nfs_rename,
160 .vop_rmdir = nfs_rmdir,
161 .vop_setattr = nfs_setattr,
162 .vop_strategy = nfs_strategy,
163 .vop_symlink = nfs_symlink,
164 .vop_write = nfs_write,
165 };
166
167 struct vop_vector nfs_fifoops = {
168 .vop_default = &fifo_specops,
169 .vop_access = nfsspec_access,
170 .vop_close = nfsfifo_close,
171 .vop_fsync = nfs_fsync,
172 .vop_getattr = nfs_getattr,
173 .vop_inactive = nfs_inactive,
174 .vop_print = nfs_print,
175 .vop_read = nfsfifo_read,
176 .vop_reclaim = nfs_reclaim,
177 .vop_setattr = nfs_setattr,
178 .vop_write = nfsfifo_write,
179 };
180
181 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
182 struct componentname *cnp, struct vattr *vap);
183 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
184 struct ucred *cred, struct thread *td);
185 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
186 int fnamelen, struct vnode *tdvp,
187 const char *tnameptr, int tnamelen,
188 struct ucred *cred, struct thread *td);
189 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
190 struct sillyrename *sp);
191
192 /*
193 * Global variables
194 */
195 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
196 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
197 int nfs_numasync = 0;
198 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
199
200 SYSCTL_DECL(_vfs_nfs);
201
202 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
203 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
204 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
205
206 static int nfsv3_commit_on_close = 0;
207 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
208 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
209
210 static int nfs_clean_pages_on_close = 1;
211 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
212 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
213
214 int nfs_directio_enable = 0;
215 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
216 &nfs_directio_enable, 0, "Enable NFS directio");
217
218 /*
219 * This sysctl allows other processes to mmap a file that has been opened
220 * O_DIRECT by a process. In general, having processes mmap the file while
221 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
222 * this by default to prevent DoS attacks - to prevent a malicious user from
223 * opening up files O_DIRECT preventing other users from mmap'ing these
224 * files. "Protected" environments where stricter consistency guarantees are
225 * required can disable this knob. The process that opened the file O_DIRECT
226 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
227 * meaningful.
228 */
229 int nfs_directio_allow_mmap = 1;
230 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
231 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
232
233 #if 0
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
235 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
236
237 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
238 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
239 #endif
240
241 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
242 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
243 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
244 static int
245 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
246 struct ucred *cred)
247 {
248 const int v3 = 1;
249 u_int32_t *tl;
250 int error = 0, attrflag;
251
252 struct mbuf *mreq, *mrep, *md, *mb;
253 caddr_t bpos, dpos;
254 u_int32_t rmode;
255 struct nfsnode *np = VTONFS(vp);
256
257 nfsstats.rpccnt[NFSPROC_ACCESS]++;
258 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
259 mb = mreq;
260 bpos = mtod(mb, caddr_t);
261 nfsm_fhtom(vp, v3);
262 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
263 *tl = txdr_unsigned(wmode);
264 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
265 nfsm_postop_attr(vp, attrflag);
266 if (!error) {
267 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
268 rmode = fxdr_unsigned(u_int32_t, *tl);
269 np->n_mode = rmode;
270 np->n_modeuid = cred->cr_uid;
271 np->n_modestamp = time_second;
272 }
273 m_freem(mrep);
274 nfsmout:
275 return (error);
276 }
277
278 /*
279 * nfs access vnode op.
280 * For nfs version 2, just return ok. File accesses may fail later.
281 * For nfs version 3, use the access rpc to check accessibility. If file modes
282 * are changed on the server, accesses might still fail later.
283 */
284 static int
285 nfs_access(struct vop_access_args *ap)
286 {
287 struct vnode *vp = ap->a_vp;
288 int error = 0;
289 u_int32_t mode, wmode;
290 int v3 = NFS_ISV3(vp);
291 struct nfsnode *np = VTONFS(vp);
292
293 /*
294 * Disallow write attempts on filesystems mounted read-only;
295 * unless the file is a socket, fifo, or a block or character
296 * device resident on the filesystem.
297 */
298 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
299 switch (vp->v_type) {
300 case VREG:
301 case VDIR:
302 case VLNK:
303 return (EROFS);
304 default:
305 break;
306 }
307 }
308 /*
309 * For nfs v3, check to see if we have done this recently, and if
310 * so return our cached result instead of making an ACCESS call.
311 * If not, do an access rpc, otherwise you are stuck emulating
312 * ufs_access() locally using the vattr. This may not be correct,
313 * since the server may apply other access criteria such as
314 * client uid-->server uid mapping that we do not know about.
315 */
316 if (v3) {
317 if (ap->a_mode & VREAD)
318 mode = NFSV3ACCESS_READ;
319 else
320 mode = 0;
321 if (vp->v_type != VDIR) {
322 if (ap->a_mode & VWRITE)
323 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
324 if (ap->a_mode & VEXEC)
325 mode |= NFSV3ACCESS_EXECUTE;
326 } else {
327 if (ap->a_mode & VWRITE)
328 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
329 NFSV3ACCESS_DELETE);
330 if (ap->a_mode & VEXEC)
331 mode |= NFSV3ACCESS_LOOKUP;
332 }
333 /* XXX safety belt, only make blanket request if caching */
334 if (nfsaccess_cache_timeout > 0) {
335 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
336 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
337 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
338 } else {
339 wmode = mode;
340 }
341
342 /*
343 * Does our cached result allow us to give a definite yes to
344 * this request?
345 */
346 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
347 (ap->a_cred->cr_uid == np->n_modeuid) &&
348 ((np->n_mode & mode) == mode)) {
349 nfsstats.accesscache_hits++;
350 } else {
351 /*
352 * Either a no, or a don't know. Go to the wire.
353 */
354 nfsstats.accesscache_misses++;
355 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
356 if (!error) {
357 if ((np->n_mode & mode) != mode) {
358 error = EACCES;
359 }
360 }
361 }
362 return (error);
363 } else {
364 if ((error = nfsspec_access(ap)) != 0)
365 return (error);
366
367 /*
368 * Attempt to prevent a mapped root from accessing a file
369 * which it shouldn't. We try to read a byte from the file
370 * if the user is root and the file is not zero length.
371 * After calling nfsspec_access, we should have the correct
372 * file size cached.
373 */
374 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
375 && VTONFS(vp)->n_size > 0) {
376 struct iovec aiov;
377 struct uio auio;
378 char buf[1];
379
380 aiov.iov_base = buf;
381 aiov.iov_len = 1;
382 auio.uio_iov = &aiov;
383 auio.uio_iovcnt = 1;
384 auio.uio_offset = 0;
385 auio.uio_resid = 1;
386 auio.uio_segflg = UIO_SYSSPACE;
387 auio.uio_rw = UIO_READ;
388 auio.uio_td = ap->a_td;
389
390 if (vp->v_type == VREG)
391 error = nfs_readrpc(vp, &auio, ap->a_cred);
392 else if (vp->v_type == VDIR) {
393 char* bp;
394 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
395 aiov.iov_base = bp;
396 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
397 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
398 free(bp, M_TEMP);
399 } else if (vp->v_type == VLNK)
400 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
401 else
402 error = EACCES;
403 }
404 return (error);
405 }
406 }
407
408 /*
409 * nfs open vnode op
410 * Check to see if the type is ok
411 * and that deletion is not in progress.
412 * For paged in text files, you will need to flush the page cache
413 * if consistency is lost.
414 */
415 /* ARGSUSED */
416 static int
417 nfs_open(struct vop_open_args *ap)
418 {
419 struct vnode *vp = ap->a_vp;
420 struct nfsnode *np = VTONFS(vp);
421 struct vattr vattr;
422 int error;
423 int fmode = ap->a_mode;
424
425 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
426 return (EOPNOTSUPP);
427
428 /*
429 * Get a valid lease. If cached data is stale, flush it.
430 */
431 if (np->n_flag & NMODIFIED) {
432 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
433 if (error == EINTR || error == EIO)
434 return (error);
435 np->n_attrstamp = 0;
436 if (vp->v_type == VDIR)
437 np->n_direofoffset = 0;
438 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
439 if (error)
440 return (error);
441 np->n_mtime = vattr.va_mtime;
442 } else {
443 np->n_attrstamp = 0;
444 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
445 if (error)
446 return (error);
447 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
448 if (vp->v_type == VDIR)
449 np->n_direofoffset = 0;
450 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
451 if (error == EINTR || error == EIO)
452 return (error);
453 np->n_mtime = vattr.va_mtime;
454 }
455 }
456 /*
457 * If the object has >= 1 O_DIRECT active opens, we disable caching.
458 */
459 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
460 if (np->n_directio_opens == 0) {
461 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
462 if (error)
463 return (error);
464 np->n_flag |= NNONCACHE;
465 }
466 np->n_directio_opens++;
467 }
468 np->ra_expect_lbn = 0;
469 vnode_create_vobject_off(vp, vattr.va_size, ap->a_td);
470 return (0);
471 }
472
473 /*
474 * nfs close vnode op
475 * What an NFS client should do upon close after writing is a debatable issue.
476 * Most NFS clients push delayed writes to the server upon close, basically for
477 * two reasons:
478 * 1 - So that any write errors may be reported back to the client process
479 * doing the close system call. By far the two most likely errors are
480 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
481 * 2 - To put a worst case upper bound on cache inconsistency between
482 * multiple clients for the file.
483 * There is also a consistency problem for Version 2 of the protocol w.r.t.
484 * not being able to tell if other clients are writing a file concurrently,
485 * since there is no way of knowing if the changed modify time in the reply
486 * is only due to the write for this client.
487 * (NFS Version 3 provides weak cache consistency data in the reply that
488 * should be sufficient to detect and handle this case.)
489 *
490 * The current code does the following:
491 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
492 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
493 * or commit them (this satisfies 1 and 2 except for the
494 * case where the server crashes after this close but
495 * before the commit RPC, which is felt to be "good
496 * enough". Changing the last argument to nfs_flush() to
497 * a 1 would force a commit operation, if it is felt a
498 * commit is necessary now.
499 */
500 /* ARGSUSED */
501 static int
502 nfs_close(struct vop_close_args *ap)
503 {
504 struct vnode *vp = ap->a_vp;
505 struct nfsnode *np = VTONFS(vp);
506 int error = 0;
507 int fmode = ap->a_fflag;
508
509 if (vp->v_type == VREG) {
510 /*
511 * Examine and clean dirty pages, regardless of NMODIFIED.
512 * This closes a major hole in close-to-open consistency.
513 * We want to push out all dirty pages (and buffers) on
514 * close, regardless of whether they were dirtied by
515 * mmap'ed writes or via write().
516 */
517 if (nfs_clean_pages_on_close && vp->v_object) {
518 VM_OBJECT_LOCK(vp->v_object);
519 vm_object_page_clean(vp->v_object, 0, 0, 0);
520 VM_OBJECT_UNLOCK(vp->v_object);
521 }
522 if (np->n_flag & NMODIFIED) {
523 if (NFS_ISV3(vp)) {
524 /*
525 * Under NFSv3 we have dirty buffers to dispose of. We
526 * must flush them to the NFS server. We have the option
527 * of waiting all the way through the commit rpc or just
528 * waiting for the initial write. The default is to only
529 * wait through the initial write so the data is in the
530 * server's cache, which is roughly similar to the state
531 * a standard disk subsystem leaves the file in on close().
532 *
533 * We cannot clear the NMODIFIED bit in np->n_flag due to
534 * potential races with other processes, and certainly
535 * cannot clear it if we don't commit.
536 */
537 int cm = nfsv3_commit_on_close ? 1 : 0;
538 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
539 /* np->n_flag &= ~NMODIFIED; */
540 } else
541 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
542 }
543 /*
544 * Invalidate the attribute cache in all cases.
545 * An open is going to fetch fresh attrs any way, other procs
546 * on this node that have file open will be forced to do an
547 * otw attr fetch, but this is safe.
548 */
549 np->n_attrstamp = 0;
550 if (np->n_flag & NWRITEERR) {
551 np->n_flag &= ~NWRITEERR;
552 error = np->n_error;
553 }
554 }
555 if (nfs_directio_enable)
556 KASSERT((np->n_directio_asyncwr == 0),
557 ("nfs_close: dirty unflushed (%d) directio buffers\n",
558 np->n_directio_asyncwr));
559 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
560 KASSERT((np->n_directio_opens > 0),
561 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
562 np->n_directio_opens--;
563 if (np->n_directio_opens == 0)
564 np->n_flag &= ~NNONCACHE;
565 }
566 return (error);
567 }
568
569 /*
570 * nfs getattr call from vfs.
571 */
572 static int
573 nfs_getattr(struct vop_getattr_args *ap)
574 {
575 struct vnode *vp = ap->a_vp;
576 struct nfsnode *np = VTONFS(vp);
577 caddr_t bpos, dpos;
578 int error = 0;
579 struct mbuf *mreq, *mrep, *md, *mb;
580 int v3 = NFS_ISV3(vp);
581
582 /*
583 * Update local times for special files.
584 */
585 if (np->n_flag & (NACC | NUPD))
586 np->n_flag |= NCHG;
587 /*
588 * First look in the cache.
589 */
590 if (nfs_getattrcache(vp, ap->a_vap) == 0)
591 return (0);
592
593 if (v3 && nfsaccess_cache_timeout > 0) {
594 nfsstats.accesscache_misses++;
595 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
596 if (nfs_getattrcache(vp, ap->a_vap) == 0)
597 return (0);
598 }
599
600 nfsstats.rpccnt[NFSPROC_GETATTR]++;
601 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
602 mb = mreq;
603 bpos = mtod(mb, caddr_t);
604 nfsm_fhtom(vp, v3);
605 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
606 if (!error) {
607 nfsm_loadattr(vp, ap->a_vap);
608 }
609 m_freem(mrep);
610 nfsmout:
611 return (error);
612 }
613
614 /*
615 * nfs setattr call.
616 */
617 static int
618 nfs_setattr(struct vop_setattr_args *ap)
619 {
620 struct vnode *vp = ap->a_vp;
621 struct nfsnode *np = VTONFS(vp);
622 struct vattr *vap = ap->a_vap;
623 int error = 0;
624 u_quad_t tsize;
625
626 #ifndef nolint
627 tsize = (u_quad_t)0;
628 #endif
629
630 /*
631 * Setting of flags and marking of atimes are not supported.
632 */
633 if (vap->va_flags != VNOVAL || (vap->va_vaflags & VA_MARK_ATIME))
634 return (EOPNOTSUPP);
635
636 /*
637 * Disallow write attempts if the filesystem is mounted read-only.
638 */
639 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
640 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
641 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
642 (vp->v_mount->mnt_flag & MNT_RDONLY))
643 return (EROFS);
644 if (vap->va_size != VNOVAL) {
645 switch (vp->v_type) {
646 case VDIR:
647 return (EISDIR);
648 case VCHR:
649 case VBLK:
650 case VSOCK:
651 case VFIFO:
652 if (vap->va_mtime.tv_sec == VNOVAL &&
653 vap->va_atime.tv_sec == VNOVAL &&
654 vap->va_mode == (mode_t)VNOVAL &&
655 vap->va_uid == (uid_t)VNOVAL &&
656 vap->va_gid == (gid_t)VNOVAL)
657 return (0);
658 vap->va_size = VNOVAL;
659 break;
660 default:
661 /*
662 * Disallow write attempts if the filesystem is
663 * mounted read-only.
664 */
665 if (vp->v_mount->mnt_flag & MNT_RDONLY)
666 return (EROFS);
667
668 /*
669 * We run vnode_pager_setsize() early (why?),
670 * we must set np->n_size now to avoid vinvalbuf
671 * V_SAVE races that might setsize a lower
672 * value.
673 */
674
675 tsize = np->n_size;
676 error = nfs_meta_setsize(vp, ap->a_cred,
677 ap->a_td, vap->va_size);
678
679 if (np->n_flag & NMODIFIED) {
680 if (vap->va_size == 0)
681 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
682 else
683 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
684 if (error) {
685 vnode_pager_setsize(vp, np->n_size);
686 return (error);
687 }
688 }
689 /*
690 * np->n_size has already been set to vap->va_size
691 * in nfs_meta_setsize(). We must set it again since
692 * nfs_loadattrcache() could be called through
693 * nfs_meta_setsize() and could modify np->n_size.
694 */
695 np->n_vattr.va_size = np->n_size = vap->va_size;
696 };
697 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
698 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
699 vp->v_type == VREG &&
700 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) != 0 &&
701 (error == EINTR || error == EIO))
702 return (error);
703 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
704 if (error && vap->va_size != VNOVAL) {
705 np->n_size = np->n_vattr.va_size = tsize;
706 vnode_pager_setsize(vp, np->n_size);
707 }
708 return (error);
709 }
710
711 /*
712 * Do an nfs setattr rpc.
713 */
714 static int
715 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
716 struct thread *td)
717 {
718 struct nfsv2_sattr *sp;
719 struct nfsnode *np = VTONFS(vp);
720 caddr_t bpos, dpos;
721 u_int32_t *tl;
722 int error = 0, wccflag = NFSV3_WCCRATTR;
723 struct mbuf *mreq, *mrep, *md, *mb;
724 int v3 = NFS_ISV3(vp);
725
726 nfsstats.rpccnt[NFSPROC_SETATTR]++;
727 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
728 mb = mreq;
729 bpos = mtod(mb, caddr_t);
730 nfsm_fhtom(vp, v3);
731 if (v3) {
732 nfsm_v3attrbuild(vap, TRUE);
733 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
734 *tl = nfs_false;
735 } else {
736 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
737 if (vap->va_mode == (mode_t)VNOVAL)
738 sp->sa_mode = nfs_xdrneg1;
739 else
740 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
741 if (vap->va_uid == (uid_t)VNOVAL)
742 sp->sa_uid = nfs_xdrneg1;
743 else
744 sp->sa_uid = txdr_unsigned(vap->va_uid);
745 if (vap->va_gid == (gid_t)VNOVAL)
746 sp->sa_gid = nfs_xdrneg1;
747 else
748 sp->sa_gid = txdr_unsigned(vap->va_gid);
749 sp->sa_size = txdr_unsigned(vap->va_size);
750 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
751 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
752 }
753 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
754 if (v3) {
755 np->n_modestamp = 0;
756 nfsm_wcc_data(vp, wccflag);
757 } else
758 nfsm_loadattr(vp, NULL);
759 m_freem(mrep);
760 nfsmout:
761 return (error);
762 }
763
764 /*
765 * nfs lookup call, one step at a time...
766 * First look in cache
767 * If not found, unlock the directory nfsnode and do the rpc
768 */
769 static int
770 nfs_lookup(struct vop_lookup_args *ap)
771 {
772 struct componentname *cnp = ap->a_cnp;
773 struct vnode *dvp = ap->a_dvp;
774 struct vnode **vpp = ap->a_vpp;
775 int flags = cnp->cn_flags;
776 struct vnode *newvp;
777 struct nfsmount *nmp;
778 caddr_t bpos, dpos;
779 struct mbuf *mreq, *mrep, *md, *mb;
780 long len;
781 nfsfh_t *fhp;
782 struct nfsnode *np;
783 int error = 0, attrflag, fhsize;
784 int v3 = NFS_ISV3(dvp);
785 struct thread *td = cnp->cn_thread;
786
787 *vpp = NULLVP;
788 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
789 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
790 return (EROFS);
791 if (dvp->v_type != VDIR)
792 return (ENOTDIR);
793 nmp = VFSTONFS(dvp->v_mount);
794 np = VTONFS(dvp);
795 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
796 *vpp = NULLVP;
797 return (error);
798 }
799 if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
800 struct vattr vattr;
801
802 newvp = *vpp;
803 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
804 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
805 nfsstats.lookupcache_hits++;
806 if (cnp->cn_nameiop != LOOKUP &&
807 (flags & ISLASTCN))
808 cnp->cn_flags |= SAVENAME;
809 return (0);
810 }
811 cache_purge(newvp);
812 if (dvp != newvp)
813 vput(newvp);
814 else
815 vrele(newvp);
816 *vpp = NULLVP;
817 }
818 error = 0;
819 newvp = NULLVP;
820 nfsstats.lookupcache_misses++;
821 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
822 len = cnp->cn_namelen;
823 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
824 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
825 mb = mreq;
826 bpos = mtod(mb, caddr_t);
827 nfsm_fhtom(dvp, v3);
828 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
829 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
830 if (error) {
831 if (v3) {
832 nfsm_postop_attr(dvp, attrflag);
833 m_freem(mrep);
834 }
835 goto nfsmout;
836 }
837 nfsm_getfh(fhp, fhsize, v3);
838
839 /*
840 * Handle RENAME case...
841 */
842 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
843 if (NFS_CMPFH(np, fhp, fhsize)) {
844 m_freem(mrep);
845 return (EISDIR);
846 }
847 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, LK_EXCLUSIVE);
848 if (error) {
849 m_freem(mrep);
850 return (error);
851 }
852 newvp = NFSTOV(np);
853 if (v3) {
854 nfsm_postop_attr(newvp, attrflag);
855 nfsm_postop_attr(dvp, attrflag);
856 } else
857 nfsm_loadattr(newvp, NULL);
858 *vpp = newvp;
859 m_freem(mrep);
860 cnp->cn_flags |= SAVENAME;
861 return (0);
862 }
863
864 if (flags & ISDOTDOT) {
865 VOP_UNLOCK(dvp, 0, td);
866 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, cnp->cn_lkflags);
867 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
868 if (error)
869 return (error);
870 newvp = NFSTOV(np);
871 } else if (NFS_CMPFH(np, fhp, fhsize)) {
872 VREF(dvp);
873 newvp = dvp;
874 } else {
875 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, cnp->cn_lkflags);
876 if (error) {
877 m_freem(mrep);
878 return (error);
879 }
880 newvp = NFSTOV(np);
881 }
882 if (v3) {
883 nfsm_postop_attr(newvp, attrflag);
884 nfsm_postop_attr(dvp, attrflag);
885 } else
886 nfsm_loadattr(newvp, NULL);
887 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
888 cnp->cn_flags |= SAVENAME;
889 if ((cnp->cn_flags & MAKEENTRY) &&
890 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
891 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
892 cache_enter(dvp, newvp, cnp);
893 }
894 *vpp = newvp;
895 m_freem(mrep);
896 nfsmout:
897 if (error) {
898 if (newvp != NULLVP) {
899 vput(newvp);
900 *vpp = NULLVP;
901 }
902 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
903 (flags & ISLASTCN) && error == ENOENT) {
904 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
905 error = EROFS;
906 else
907 error = EJUSTRETURN;
908 }
909 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
910 cnp->cn_flags |= SAVENAME;
911 }
912 return (error);
913 }
914
915 /*
916 * nfs read call.
917 * Just call nfs_bioread() to do the work.
918 */
919 static int
920 nfs_read(struct vop_read_args *ap)
921 {
922 struct vnode *vp = ap->a_vp;
923
924 switch (vp->v_type) {
925 case VREG:
926 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
927 case VDIR:
928 return (EISDIR);
929 default:
930 return (EOPNOTSUPP);
931 }
932 }
933
934 /*
935 * nfs readlink call
936 */
937 static int
938 nfs_readlink(struct vop_readlink_args *ap)
939 {
940 struct vnode *vp = ap->a_vp;
941
942 if (vp->v_type != VLNK)
943 return (EINVAL);
944 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
945 }
946
947 /*
948 * Do a readlink rpc.
949 * Called by nfs_doio() from below the buffer cache.
950 */
951 int
952 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
953 {
954 caddr_t bpos, dpos;
955 int error = 0, len, attrflag;
956 struct mbuf *mreq, *mrep, *md, *mb;
957 int v3 = NFS_ISV3(vp);
958
959 nfsstats.rpccnt[NFSPROC_READLINK]++;
960 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
961 mb = mreq;
962 bpos = mtod(mb, caddr_t);
963 nfsm_fhtom(vp, v3);
964 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
965 if (v3)
966 nfsm_postop_attr(vp, attrflag);
967 if (!error) {
968 nfsm_strsiz(len, NFS_MAXPATHLEN);
969 if (len == NFS_MAXPATHLEN) {
970 struct nfsnode *np = VTONFS(vp);
971 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
972 len = np->n_size;
973 }
974 nfsm_mtouio(uiop, len);
975 }
976 m_freem(mrep);
977 nfsmout:
978 return (error);
979 }
980
981 /*
982 * nfs read rpc call
983 * Ditto above
984 */
985 int
986 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
987 {
988 u_int32_t *tl;
989 caddr_t bpos, dpos;
990 struct mbuf *mreq, *mrep, *md, *mb;
991 struct nfsmount *nmp;
992 int error = 0, len, retlen, tsiz, eof, attrflag;
993 int v3 = NFS_ISV3(vp);
994
995 #ifndef nolint
996 eof = 0;
997 #endif
998 nmp = VFSTONFS(vp->v_mount);
999 tsiz = uiop->uio_resid;
1000 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1001 return (EFBIG);
1002 while (tsiz > 0) {
1003 nfsstats.rpccnt[NFSPROC_READ]++;
1004 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1005 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1006 mb = mreq;
1007 bpos = mtod(mb, caddr_t);
1008 nfsm_fhtom(vp, v3);
1009 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1010 if (v3) {
1011 txdr_hyper(uiop->uio_offset, tl);
1012 *(tl + 2) = txdr_unsigned(len);
1013 } else {
1014 *tl++ = txdr_unsigned(uiop->uio_offset);
1015 *tl++ = txdr_unsigned(len);
1016 *tl = 0;
1017 }
1018 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1019 if (v3) {
1020 nfsm_postop_attr(vp, attrflag);
1021 if (error) {
1022 m_freem(mrep);
1023 goto nfsmout;
1024 }
1025 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1026 eof = fxdr_unsigned(int, *(tl + 1));
1027 } else
1028 nfsm_loadattr(vp, NULL);
1029 nfsm_strsiz(retlen, nmp->nm_rsize);
1030 nfsm_mtouio(uiop, retlen);
1031 m_freem(mrep);
1032 tsiz -= retlen;
1033 if (v3) {
1034 if (eof || retlen == 0) {
1035 tsiz = 0;
1036 }
1037 } else if (retlen < len) {
1038 tsiz = 0;
1039 }
1040 }
1041 nfsmout:
1042 return (error);
1043 }
1044
1045 /*
1046 * nfs write call
1047 */
1048 int
1049 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1050 int *iomode, int *must_commit)
1051 {
1052 u_int32_t *tl;
1053 int32_t backup;
1054 caddr_t bpos, dpos;
1055 struct mbuf *mreq, *mrep, *md, *mb;
1056 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1057 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1058 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1059
1060 #ifndef DIAGNOSTIC
1061 if (uiop->uio_iovcnt != 1)
1062 panic("nfs: writerpc iovcnt > 1");
1063 #endif
1064 *must_commit = 0;
1065 tsiz = uiop->uio_resid;
1066 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1067 return (EFBIG);
1068 while (tsiz > 0) {
1069 nfsstats.rpccnt[NFSPROC_WRITE]++;
1070 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1071 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1072 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1073 mb = mreq;
1074 bpos = mtod(mb, caddr_t);
1075 nfsm_fhtom(vp, v3);
1076 if (v3) {
1077 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1078 txdr_hyper(uiop->uio_offset, tl);
1079 tl += 2;
1080 *tl++ = txdr_unsigned(len);
1081 *tl++ = txdr_unsigned(*iomode);
1082 *tl = txdr_unsigned(len);
1083 } else {
1084 u_int32_t x;
1085
1086 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1087 /* Set both "begin" and "current" to non-garbage. */
1088 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1089 *tl++ = x; /* "begin offset" */
1090 *tl++ = x; /* "current offset" */
1091 x = txdr_unsigned(len);
1092 *tl++ = x; /* total to this offset */
1093 *tl = x; /* size of this write */
1094 }
1095 nfsm_uiotom(uiop, len);
1096 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1097 if (v3) {
1098 wccflag = NFSV3_WCCCHK;
1099 nfsm_wcc_data(vp, wccflag);
1100 if (!error) {
1101 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1102 + NFSX_V3WRITEVERF);
1103 rlen = fxdr_unsigned(int, *tl++);
1104 if (rlen == 0) {
1105 error = NFSERR_IO;
1106 m_freem(mrep);
1107 break;
1108 } else if (rlen < len) {
1109 backup = len - rlen;
1110 uiop->uio_iov->iov_base =
1111 (char *)uiop->uio_iov->iov_base -
1112 backup;
1113 uiop->uio_iov->iov_len += backup;
1114 uiop->uio_offset -= backup;
1115 uiop->uio_resid += backup;
1116 len = rlen;
1117 }
1118 commit = fxdr_unsigned(int, *tl++);
1119
1120 /*
1121 * Return the lowest committment level
1122 * obtained by any of the RPCs.
1123 */
1124 if (committed == NFSV3WRITE_FILESYNC)
1125 committed = commit;
1126 else if (committed == NFSV3WRITE_DATASYNC &&
1127 commit == NFSV3WRITE_UNSTABLE)
1128 committed = commit;
1129 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1130 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1131 NFSX_V3WRITEVERF);
1132 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1133 } else if (bcmp((caddr_t)tl,
1134 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1135 *must_commit = 1;
1136 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1137 NFSX_V3WRITEVERF);
1138 }
1139 }
1140 } else
1141 nfsm_loadattr(vp, NULL);
1142 if (wccflag)
1143 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1144 m_freem(mrep);
1145 if (error)
1146 break;
1147 tsiz -= len;
1148 }
1149 nfsmout:
1150 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1151 committed = NFSV3WRITE_FILESYNC;
1152 *iomode = committed;
1153 if (error)
1154 uiop->uio_resid = tsiz;
1155 return (error);
1156 }
1157
1158 /*
1159 * nfs mknod rpc
1160 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1161 * mode set to specify the file type and the size field for rdev.
1162 */
1163 static int
1164 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1165 struct vattr *vap)
1166 {
1167 struct nfsv2_sattr *sp;
1168 u_int32_t *tl;
1169 struct vnode *newvp = NULL;
1170 struct nfsnode *np = NULL;
1171 struct vattr vattr;
1172 caddr_t bpos, dpos;
1173 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1174 struct mbuf *mreq, *mrep, *md, *mb;
1175 u_int32_t rdev;
1176 int v3 = NFS_ISV3(dvp);
1177
1178 if (vap->va_type == VCHR || vap->va_type == VBLK)
1179 rdev = txdr_unsigned(vap->va_rdev);
1180 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1181 rdev = nfs_xdrneg1;
1182 else {
1183 return (EOPNOTSUPP);
1184 }
1185 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1186 return (error);
1187 }
1188 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1189 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1190 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1191 mb = mreq;
1192 bpos = mtod(mb, caddr_t);
1193 nfsm_fhtom(dvp, v3);
1194 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1195 if (v3) {
1196 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1197 *tl++ = vtonfsv3_type(vap->va_type);
1198 nfsm_v3attrbuild(vap, FALSE);
1199 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1200 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1201 *tl++ = txdr_unsigned(umajor(vap->va_rdev));
1202 *tl = txdr_unsigned(uminor(vap->va_rdev));
1203 }
1204 } else {
1205 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1206 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1207 sp->sa_uid = nfs_xdrneg1;
1208 sp->sa_gid = nfs_xdrneg1;
1209 sp->sa_size = rdev;
1210 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1211 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1212 }
1213 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1214 if (!error) {
1215 nfsm_mtofh(dvp, newvp, v3, gotvp);
1216 if (!gotvp) {
1217 if (newvp) {
1218 vput(newvp);
1219 newvp = NULL;
1220 }
1221 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1222 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1223 if (!error)
1224 newvp = NFSTOV(np);
1225 }
1226 }
1227 if (v3)
1228 nfsm_wcc_data(dvp, wccflag);
1229 m_freem(mrep);
1230 nfsmout:
1231 if (error) {
1232 if (newvp)
1233 vput(newvp);
1234 } else {
1235 if (cnp->cn_flags & MAKEENTRY)
1236 cache_enter(dvp, newvp, cnp);
1237 *vpp = newvp;
1238 }
1239 VTONFS(dvp)->n_flag |= NMODIFIED;
1240 if (!wccflag)
1241 VTONFS(dvp)->n_attrstamp = 0;
1242 return (error);
1243 }
1244
1245 /*
1246 * nfs mknod vop
1247 * just call nfs_mknodrpc() to do the work.
1248 */
1249 /* ARGSUSED */
1250 static int
1251 nfs_mknod(struct vop_mknod_args *ap)
1252 {
1253
1254 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1255 }
1256
1257 static u_long create_verf;
1258 /*
1259 * nfs file create call
1260 */
1261 static int
1262 nfs_create(struct vop_create_args *ap)
1263 {
1264 struct vnode *dvp = ap->a_dvp;
1265 struct vattr *vap = ap->a_vap;
1266 struct componentname *cnp = ap->a_cnp;
1267 struct nfsv2_sattr *sp;
1268 u_int32_t *tl;
1269 struct nfsnode *np = NULL;
1270 struct vnode *newvp = NULL;
1271 caddr_t bpos, dpos;
1272 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1273 struct mbuf *mreq, *mrep, *md, *mb;
1274 struct vattr vattr;
1275 int v3 = NFS_ISV3(dvp);
1276
1277 /*
1278 * Oops, not for me..
1279 */
1280 if (vap->va_type == VSOCK)
1281 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1282
1283 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1284 return (error);
1285 }
1286 if (vap->va_vaflags & VA_EXCLUSIVE)
1287 fmode |= O_EXCL;
1288 again:
1289 nfsstats.rpccnt[NFSPROC_CREATE]++;
1290 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1291 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1292 mb = mreq;
1293 bpos = mtod(mb, caddr_t);
1294 nfsm_fhtom(dvp, v3);
1295 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1296 if (v3) {
1297 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1298 if (fmode & O_EXCL) {
1299 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1300 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1301 #ifdef INET
1302 if (!TAILQ_EMPTY(&in_ifaddrhead))
1303 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1304 else
1305 #endif
1306 *tl++ = create_verf;
1307 *tl = ++create_verf;
1308 } else {
1309 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1310 nfsm_v3attrbuild(vap, FALSE);
1311 }
1312 } else {
1313 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1314 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1315 sp->sa_uid = nfs_xdrneg1;
1316 sp->sa_gid = nfs_xdrneg1;
1317 sp->sa_size = 0;
1318 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1319 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1320 }
1321 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1322 if (!error) {
1323 nfsm_mtofh(dvp, newvp, v3, gotvp);
1324 if (!gotvp) {
1325 if (newvp) {
1326 vput(newvp);
1327 newvp = NULL;
1328 }
1329 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1330 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1331 if (!error)
1332 newvp = NFSTOV(np);
1333 }
1334 }
1335 if (v3)
1336 nfsm_wcc_data(dvp, wccflag);
1337 m_freem(mrep);
1338 nfsmout:
1339 if (error) {
1340 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1341 fmode &= ~O_EXCL;
1342 goto again;
1343 }
1344 if (newvp)
1345 vput(newvp);
1346 } else if (v3 && (fmode & O_EXCL)) {
1347 /*
1348 * We are normally called with only a partially initialized
1349 * VAP. Since the NFSv3 spec says that server may use the
1350 * file attributes to store the verifier, the spec requires
1351 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1352 * in atime, but we can't really assume that all servers will
1353 * so we ensure that our SETATTR sets both atime and mtime.
1354 */
1355 if (vap->va_mtime.tv_sec == VNOVAL)
1356 vfs_timestamp(&vap->va_mtime);
1357 if (vap->va_atime.tv_sec == VNOVAL)
1358 vap->va_atime = vap->va_mtime;
1359 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
1360 if (error)
1361 vput(newvp);
1362 }
1363 if (!error) {
1364 if (cnp->cn_flags & MAKEENTRY)
1365 cache_enter(dvp, newvp, cnp);
1366 *ap->a_vpp = newvp;
1367 }
1368 VTONFS(dvp)->n_flag |= NMODIFIED;
1369 if (!wccflag)
1370 VTONFS(dvp)->n_attrstamp = 0;
1371 return (error);
1372 }
1373
1374 /*
1375 * nfs file remove call
1376 * To try and make nfs semantics closer to ufs semantics, a file that has
1377 * other processes using the vnode is renamed instead of removed and then
1378 * removed later on the last close.
1379 * - If v_usecount > 1
1380 * If a rename is not already in the works
1381 * call nfs_sillyrename() to set it up
1382 * else
1383 * do the remove rpc
1384 */
1385 static int
1386 nfs_remove(struct vop_remove_args *ap)
1387 {
1388 struct vnode *vp = ap->a_vp;
1389 struct vnode *dvp = ap->a_dvp;
1390 struct componentname *cnp = ap->a_cnp;
1391 struct nfsnode *np = VTONFS(vp);
1392 int error = 0;
1393 struct vattr vattr;
1394
1395 #ifndef DIAGNOSTIC
1396 if ((cnp->cn_flags & HASBUF) == 0)
1397 panic("nfs_remove: no name");
1398 if (vrefcnt(vp) < 1)
1399 panic("nfs_remove: bad v_usecount");
1400 #endif
1401 if (vp->v_type == VDIR)
1402 error = EPERM;
1403 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1404 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
1405 vattr.va_nlink > 1)) {
1406 /*
1407 * Purge the name cache so that the chance of a lookup for
1408 * the name succeeding while the remove is in progress is
1409 * minimized. Without node locking it can still happen, such
1410 * that an I/O op returns ESTALE, but since you get this if
1411 * another host removes the file..
1412 */
1413 cache_purge(vp);
1414 /*
1415 * throw away biocache buffers, mainly to avoid
1416 * unnecessary delayed writes later.
1417 */
1418 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1419 /* Do the rpc */
1420 if (error != EINTR && error != EIO)
1421 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1422 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1423 /*
1424 * Kludge City: If the first reply to the remove rpc is lost..
1425 * the reply to the retransmitted request will be ENOENT
1426 * since the file was in fact removed
1427 * Therefore, we cheat and return success.
1428 */
1429 if (error == ENOENT)
1430 error = 0;
1431 } else if (!np->n_sillyrename)
1432 error = nfs_sillyrename(dvp, vp, cnp);
1433 np->n_attrstamp = 0;
1434 return (error);
1435 }
1436
1437 /*
1438 * nfs file remove rpc called from nfs_inactive
1439 */
1440 int
1441 nfs_removeit(struct sillyrename *sp)
1442 {
1443
1444 /*
1445 * Make sure that the directory vnode is still valid.
1446 * XXX we should lock sp->s_dvp here.
1447 */
1448 if (sp->s_dvp->v_type == VBAD)
1449 return (0);
1450 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1451 NULL));
1452 }
1453
1454 /*
1455 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1456 */
1457 static int
1458 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1459 struct ucred *cred, struct thread *td)
1460 {
1461 caddr_t bpos, dpos;
1462 int error = 0, wccflag = NFSV3_WCCRATTR;
1463 struct mbuf *mreq, *mrep, *md, *mb;
1464 int v3 = NFS_ISV3(dvp);
1465
1466 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1467 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1468 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1469 mb = mreq;
1470 bpos = mtod(mb, caddr_t);
1471 nfsm_fhtom(dvp, v3);
1472 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1473 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1474 if (v3)
1475 nfsm_wcc_data(dvp, wccflag);
1476 m_freem(mrep);
1477 nfsmout:
1478 VTONFS(dvp)->n_flag |= NMODIFIED;
1479 if (!wccflag)
1480 VTONFS(dvp)->n_attrstamp = 0;
1481 return (error);
1482 }
1483
1484 /*
1485 * nfs file rename call
1486 */
1487 static int
1488 nfs_rename(struct vop_rename_args *ap)
1489 {
1490 struct vnode *fvp = ap->a_fvp;
1491 struct vnode *tvp = ap->a_tvp;
1492 struct vnode *fdvp = ap->a_fdvp;
1493 struct vnode *tdvp = ap->a_tdvp;
1494 struct componentname *tcnp = ap->a_tcnp;
1495 struct componentname *fcnp = ap->a_fcnp;
1496 int error;
1497
1498 #ifndef DIAGNOSTIC
1499 if ((tcnp->cn_flags & HASBUF) == 0 ||
1500 (fcnp->cn_flags & HASBUF) == 0)
1501 panic("nfs_rename: no name");
1502 #endif
1503 /* Check for cross-device rename */
1504 if ((fvp->v_mount != tdvp->v_mount) ||
1505 (tvp && (fvp->v_mount != tvp->v_mount))) {
1506 error = EXDEV;
1507 goto out;
1508 }
1509
1510 if (fvp == tvp) {
1511 printf("nfs_rename: fvp == tvp (can't happen)\n");
1512 error = 0;
1513 goto out;
1514 }
1515 if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0)
1516 goto out;
1517
1518 /*
1519 * We have to flush B_DELWRI data prior to renaming
1520 * the file. If we don't, the delayed-write buffers
1521 * can be flushed out later after the file has gone stale
1522 * under NFSV3. NFSV2 does not have this problem because
1523 * ( as far as I can tell ) it flushes dirty buffers more
1524 * often.
1525 *
1526 * Skip the rename operation if the fsync fails, this can happen
1527 * due to the server's volume being full, when we pushed out data
1528 * that was written back to our cache earlier. Not checking for
1529 * this condition can result in potential (silent) data loss.
1530 */
1531 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1532 VOP_UNLOCK(fvp, 0, fcnp->cn_thread);
1533 if (!error && tvp)
1534 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1535 if (error)
1536 goto out;
1537
1538 /*
1539 * If the tvp exists and is in use, sillyrename it before doing the
1540 * rename of the new file over it.
1541 * XXX Can't sillyrename a directory.
1542 */
1543 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1544 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1545 vput(tvp);
1546 tvp = NULL;
1547 }
1548
1549 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1550 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1551 tcnp->cn_thread);
1552
1553 if (fvp->v_type == VDIR) {
1554 if (tvp != NULL && tvp->v_type == VDIR)
1555 cache_purge(tdvp);
1556 cache_purge(fdvp);
1557 }
1558
1559 out:
1560 if (tdvp == tvp)
1561 vrele(tdvp);
1562 else
1563 vput(tdvp);
1564 if (tvp)
1565 vput(tvp);
1566 vrele(fdvp);
1567 vrele(fvp);
1568 /*
1569 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1570 */
1571 if (error == ENOENT)
1572 error = 0;
1573 return (error);
1574 }
1575
1576 /*
1577 * nfs file rename rpc called from nfs_remove() above
1578 */
1579 static int
1580 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1581 struct sillyrename *sp)
1582 {
1583
1584 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1585 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1586 }
1587
1588 /*
1589 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1590 */
1591 static int
1592 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1593 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1594 struct thread *td)
1595 {
1596 caddr_t bpos, dpos;
1597 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1598 struct mbuf *mreq, *mrep, *md, *mb;
1599 int v3 = NFS_ISV3(fdvp);
1600
1601 nfsstats.rpccnt[NFSPROC_RENAME]++;
1602 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1603 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1604 nfsm_rndup(tnamelen));
1605 mb = mreq;
1606 bpos = mtod(mb, caddr_t);
1607 nfsm_fhtom(fdvp, v3);
1608 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1609 nfsm_fhtom(tdvp, v3);
1610 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1611 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1612 if (v3) {
1613 nfsm_wcc_data(fdvp, fwccflag);
1614 nfsm_wcc_data(tdvp, twccflag);
1615 }
1616 m_freem(mrep);
1617 nfsmout:
1618 VTONFS(fdvp)->n_flag |= NMODIFIED;
1619 VTONFS(tdvp)->n_flag |= NMODIFIED;
1620 if (!fwccflag)
1621 VTONFS(fdvp)->n_attrstamp = 0;
1622 if (!twccflag)
1623 VTONFS(tdvp)->n_attrstamp = 0;
1624 return (error);
1625 }
1626
1627 /*
1628 * nfs hard link create call
1629 */
1630 static int
1631 nfs_link(struct vop_link_args *ap)
1632 {
1633 struct vnode *vp = ap->a_vp;
1634 struct vnode *tdvp = ap->a_tdvp;
1635 struct componentname *cnp = ap->a_cnp;
1636 caddr_t bpos, dpos;
1637 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1638 struct mbuf *mreq, *mrep, *md, *mb;
1639 int v3;
1640
1641 if (vp->v_mount != tdvp->v_mount) {
1642 return (EXDEV);
1643 }
1644
1645 /*
1646 * Push all writes to the server, so that the attribute cache
1647 * doesn't get "out of sync" with the server.
1648 * XXX There should be a better way!
1649 */
1650 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1651
1652 v3 = NFS_ISV3(vp);
1653 nfsstats.rpccnt[NFSPROC_LINK]++;
1654 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1655 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1656 mb = mreq;
1657 bpos = mtod(mb, caddr_t);
1658 nfsm_fhtom(vp, v3);
1659 nfsm_fhtom(tdvp, v3);
1660 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1661 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1662 if (v3) {
1663 nfsm_postop_attr(vp, attrflag);
1664 nfsm_wcc_data(tdvp, wccflag);
1665 }
1666 m_freem(mrep);
1667 nfsmout:
1668 VTONFS(tdvp)->n_flag |= NMODIFIED;
1669 if (!attrflag)
1670 VTONFS(vp)->n_attrstamp = 0;
1671 if (!wccflag)
1672 VTONFS(tdvp)->n_attrstamp = 0;
1673 /*
1674 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1675 */
1676 if (error == EEXIST)
1677 error = 0;
1678 return (error);
1679 }
1680
1681 /*
1682 * nfs symbolic link create call
1683 */
1684 static int
1685 nfs_symlink(struct vop_symlink_args *ap)
1686 {
1687 struct vnode *dvp = ap->a_dvp;
1688 struct vattr *vap = ap->a_vap;
1689 struct componentname *cnp = ap->a_cnp;
1690 struct nfsv2_sattr *sp;
1691 caddr_t bpos, dpos;
1692 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1693 struct mbuf *mreq, *mrep, *md, *mb;
1694 struct vnode *newvp = NULL;
1695 int v3 = NFS_ISV3(dvp);
1696
1697 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1698 slen = strlen(ap->a_target);
1699 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1700 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1701 mb = mreq;
1702 bpos = mtod(mb, caddr_t);
1703 nfsm_fhtom(dvp, v3);
1704 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1705 if (v3) {
1706 nfsm_v3attrbuild(vap, FALSE);
1707 }
1708 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1709 if (!v3) {
1710 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1711 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1712 sp->sa_uid = nfs_xdrneg1;
1713 sp->sa_gid = nfs_xdrneg1;
1714 sp->sa_size = nfs_xdrneg1;
1715 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1716 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1717 }
1718
1719 /*
1720 * Issue the NFS request and get the rpc response.
1721 *
1722 * Only NFSv3 responses returning an error of 0 actually return
1723 * a file handle that can be converted into newvp without having
1724 * to do an extra lookup rpc.
1725 */
1726 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
1727 if (v3) {
1728 if (error == 0)
1729 nfsm_mtofh(dvp, newvp, v3, gotvp);
1730 nfsm_wcc_data(dvp, wccflag);
1731 }
1732
1733 /*
1734 * out code jumps -> here, mrep is also freed.
1735 */
1736
1737 m_freem(mrep);
1738 nfsmout:
1739
1740 /*
1741 * If we get an EEXIST error, silently convert it to no-error
1742 * in case of an NFS retry.
1743 */
1744 if (error == EEXIST)
1745 error = 0;
1746
1747 /*
1748 * If we do not have (or no longer have) an error, and we could
1749 * not extract the newvp from the response due to the request being
1750 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1751 * to obtain a newvp to return.
1752 */
1753 if (error == 0 && newvp == NULL) {
1754 struct nfsnode *np = NULL;
1755
1756 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1757 cnp->cn_cred, cnp->cn_thread, &np);
1758 if (!error)
1759 newvp = NFSTOV(np);
1760 }
1761 if (error) {
1762 if (newvp)
1763 vput(newvp);
1764 } else {
1765 *ap->a_vpp = newvp;
1766 }
1767 VTONFS(dvp)->n_flag |= NMODIFIED;
1768 if (!wccflag)
1769 VTONFS(dvp)->n_attrstamp = 0;
1770 return (error);
1771 }
1772
1773 /*
1774 * nfs make dir call
1775 */
1776 static int
1777 nfs_mkdir(struct vop_mkdir_args *ap)
1778 {
1779 struct vnode *dvp = ap->a_dvp;
1780 struct vattr *vap = ap->a_vap;
1781 struct componentname *cnp = ap->a_cnp;
1782 struct nfsv2_sattr *sp;
1783 int len;
1784 struct nfsnode *np = NULL;
1785 struct vnode *newvp = NULL;
1786 caddr_t bpos, dpos;
1787 int error = 0, wccflag = NFSV3_WCCRATTR;
1788 int gotvp = 0;
1789 struct mbuf *mreq, *mrep, *md, *mb;
1790 struct vattr vattr;
1791 int v3 = NFS_ISV3(dvp);
1792
1793 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1794 return (error);
1795 }
1796 len = cnp->cn_namelen;
1797 nfsstats.rpccnt[NFSPROC_MKDIR]++;
1798 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
1799 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1800 mb = mreq;
1801 bpos = mtod(mb, caddr_t);
1802 nfsm_fhtom(dvp, v3);
1803 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1804 if (v3) {
1805 nfsm_v3attrbuild(vap, FALSE);
1806 } else {
1807 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1808 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1809 sp->sa_uid = nfs_xdrneg1;
1810 sp->sa_gid = nfs_xdrneg1;
1811 sp->sa_size = nfs_xdrneg1;
1812 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1813 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1814 }
1815 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
1816 if (!error)
1817 nfsm_mtofh(dvp, newvp, v3, gotvp);
1818 if (v3)
1819 nfsm_wcc_data(dvp, wccflag);
1820 m_freem(mrep);
1821 nfsmout:
1822 VTONFS(dvp)->n_flag |= NMODIFIED;
1823 if (!wccflag)
1824 VTONFS(dvp)->n_attrstamp = 0;
1825 /*
1826 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1827 * if we can succeed in looking up the directory.
1828 */
1829 if (error == EEXIST || (!error && !gotvp)) {
1830 if (newvp) {
1831 vput(newvp);
1832 newvp = NULL;
1833 }
1834 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1835 cnp->cn_thread, &np);
1836 if (!error) {
1837 newvp = NFSTOV(np);
1838 if (newvp->v_type != VDIR)
1839 error = EEXIST;
1840 }
1841 }
1842 if (error) {
1843 if (newvp)
1844 vput(newvp);
1845 } else
1846 *ap->a_vpp = newvp;
1847 return (error);
1848 }
1849
1850 /*
1851 * nfs remove directory call
1852 */
1853 static int
1854 nfs_rmdir(struct vop_rmdir_args *ap)
1855 {
1856 struct vnode *vp = ap->a_vp;
1857 struct vnode *dvp = ap->a_dvp;
1858 struct componentname *cnp = ap->a_cnp;
1859 caddr_t bpos, dpos;
1860 int error = 0, wccflag = NFSV3_WCCRATTR;
1861 struct mbuf *mreq, *mrep, *md, *mb;
1862 int v3 = NFS_ISV3(dvp);
1863
1864 if (dvp == vp)
1865 return (EINVAL);
1866 nfsstats.rpccnt[NFSPROC_RMDIR]++;
1867 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
1868 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1869 mb = mreq;
1870 bpos = mtod(mb, caddr_t);
1871 nfsm_fhtom(dvp, v3);
1872 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1873 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
1874 if (v3)
1875 nfsm_wcc_data(dvp, wccflag);
1876 m_freem(mrep);
1877 nfsmout:
1878 VTONFS(dvp)->n_flag |= NMODIFIED;
1879 if (!wccflag)
1880 VTONFS(dvp)->n_attrstamp = 0;
1881 cache_purge(dvp);
1882 cache_purge(vp);
1883 /*
1884 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
1885 */
1886 if (error == ENOENT)
1887 error = 0;
1888 return (error);
1889 }
1890
1891 /*
1892 * nfs readdir call
1893 */
1894 static int
1895 nfs_readdir(struct vop_readdir_args *ap)
1896 {
1897 struct vnode *vp = ap->a_vp;
1898 struct nfsnode *np = VTONFS(vp);
1899 struct uio *uio = ap->a_uio;
1900 int tresid, error;
1901 struct vattr vattr;
1902
1903 if (vp->v_type != VDIR)
1904 return (EPERM);
1905 /*
1906 * First, check for hit on the EOF offset cache
1907 */
1908 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
1909 (np->n_flag & NMODIFIED) == 0) {
1910 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 &&
1911 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
1912 nfsstats.direofcache_hits++;
1913 return (0);
1914 }
1915 }
1916
1917 /*
1918 * Call nfs_bioread() to do the real work.
1919 */
1920 tresid = uio->uio_resid;
1921 error = nfs_bioread(vp, uio, 0, ap->a_cred);
1922
1923 if (!error && uio->uio_resid == tresid)
1924 nfsstats.direofcache_misses++;
1925 return (error);
1926 }
1927
1928 /*
1929 * Readdir rpc call.
1930 * Called from below the buffer cache by nfs_doio().
1931 */
1932 int
1933 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1934 {
1935 int len, left;
1936 struct dirent *dp = NULL;
1937 u_int32_t *tl;
1938 caddr_t cp;
1939 nfsuint64 *cookiep;
1940 caddr_t bpos, dpos;
1941 struct mbuf *mreq, *mrep, *md, *mb;
1942 nfsuint64 cookie;
1943 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1944 struct nfsnode *dnp = VTONFS(vp);
1945 u_quad_t fileno;
1946 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
1947 int attrflag;
1948 int v3 = NFS_ISV3(vp);
1949
1950 #ifndef DIAGNOSTIC
1951 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
1952 (uiop->uio_resid & (DIRBLKSIZ - 1)))
1953 panic("nfs readdirrpc bad uio");
1954 #endif
1955
1956 /*
1957 * If there is no cookie, assume directory was stale.
1958 */
1959 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
1960 if (cookiep)
1961 cookie = *cookiep;
1962 else
1963 return (NFSERR_BAD_COOKIE);
1964 /*
1965 * Loop around doing readdir rpc's of size nm_readdirsize
1966 * truncated to a multiple of DIRBLKSIZ.
1967 * The stopping criteria is EOF or buffer full.
1968 */
1969 while (more_dirs && bigenough) {
1970 nfsstats.rpccnt[NFSPROC_READDIR]++;
1971 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
1972 NFSX_READDIR(v3));
1973 mb = mreq;
1974 bpos = mtod(mb, caddr_t);
1975 nfsm_fhtom(vp, v3);
1976 if (v3) {
1977 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1978 *tl++ = cookie.nfsuquad[0];
1979 *tl++ = cookie.nfsuquad[1];
1980 *tl++ = dnp->n_cookieverf.nfsuquad[0];
1981 *tl++ = dnp->n_cookieverf.nfsuquad[1];
1982 } else {
1983 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1984 *tl++ = cookie.nfsuquad[0];
1985 }
1986 *tl = txdr_unsigned(nmp->nm_readdirsize);
1987 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
1988 if (v3) {
1989 nfsm_postop_attr(vp, attrflag);
1990 if (!error) {
1991 tl = nfsm_dissect(u_int32_t *,
1992 2 * NFSX_UNSIGNED);
1993 dnp->n_cookieverf.nfsuquad[0] = *tl++;
1994 dnp->n_cookieverf.nfsuquad[1] = *tl;
1995 } else {
1996 m_freem(mrep);
1997 goto nfsmout;
1998 }
1999 }
2000 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2001 more_dirs = fxdr_unsigned(int, *tl);
2002
2003 /* loop thru the dir entries, doctoring them to 4bsd form */
2004 while (more_dirs && bigenough) {
2005 if (v3) {
2006 tl = nfsm_dissect(u_int32_t *,
2007 3 * NFSX_UNSIGNED);
2008 fileno = fxdr_hyper(tl);
2009 len = fxdr_unsigned(int, *(tl + 2));
2010 } else {
2011 tl = nfsm_dissect(u_int32_t *,
2012 2 * NFSX_UNSIGNED);
2013 fileno = fxdr_unsigned(u_quad_t, *tl++);
2014 len = fxdr_unsigned(int, *tl);
2015 }
2016 if (len <= 0 || len > NFS_MAXNAMLEN) {
2017 error = EBADRPC;
2018 m_freem(mrep);
2019 goto nfsmout;
2020 }
2021 tlen = nfsm_rndup(len);
2022 if (tlen == len)
2023 tlen += 4; /* To ensure null termination */
2024 left = DIRBLKSIZ - blksiz;
2025 if ((tlen + DIRHDSIZ) > left) {
2026 dp->d_reclen += left;
2027 uiop->uio_iov->iov_base =
2028 (char *)uiop->uio_iov->iov_base + left;
2029 uiop->uio_iov->iov_len -= left;
2030 uiop->uio_offset += left;
2031 uiop->uio_resid -= left;
2032 blksiz = 0;
2033 }
2034 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2035 bigenough = 0;
2036 if (bigenough) {
2037 dp = (struct dirent *)uiop->uio_iov->iov_base;
2038 dp->d_fileno = (int)fileno;
2039 dp->d_namlen = len;
2040 dp->d_reclen = tlen + DIRHDSIZ;
2041 dp->d_type = DT_UNKNOWN;
2042 blksiz += dp->d_reclen;
2043 if (blksiz == DIRBLKSIZ)
2044 blksiz = 0;
2045 uiop->uio_offset += DIRHDSIZ;
2046 uiop->uio_resid -= DIRHDSIZ;
2047 uiop->uio_iov->iov_base =
2048 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2049 uiop->uio_iov->iov_len -= DIRHDSIZ;
2050 nfsm_mtouio(uiop, len);
2051 cp = uiop->uio_iov->iov_base;
2052 tlen -= len;
2053 *cp = '\0'; /* null terminate */
2054 uiop->uio_iov->iov_base =
2055 (char *)uiop->uio_iov->iov_base + tlen;
2056 uiop->uio_iov->iov_len -= tlen;
2057 uiop->uio_offset += tlen;
2058 uiop->uio_resid -= tlen;
2059 } else
2060 nfsm_adv(nfsm_rndup(len));
2061 if (v3) {
2062 tl = nfsm_dissect(u_int32_t *,
2063 3 * NFSX_UNSIGNED);
2064 } else {
2065 tl = nfsm_dissect(u_int32_t *,
2066 2 * NFSX_UNSIGNED);
2067 }
2068 if (bigenough) {
2069 cookie.nfsuquad[0] = *tl++;
2070 if (v3)
2071 cookie.nfsuquad[1] = *tl++;
2072 } else if (v3)
2073 tl += 2;
2074 else
2075 tl++;
2076 more_dirs = fxdr_unsigned(int, *tl);
2077 }
2078 /*
2079 * If at end of rpc data, get the eof boolean
2080 */
2081 if (!more_dirs) {
2082 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2083 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2084 }
2085 m_freem(mrep);
2086 }
2087 /*
2088 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2089 * by increasing d_reclen for the last record.
2090 */
2091 if (blksiz > 0) {
2092 left = DIRBLKSIZ - blksiz;
2093 dp->d_reclen += left;
2094 uiop->uio_iov->iov_base =
2095 (char *)uiop->uio_iov->iov_base + left;
2096 uiop->uio_iov->iov_len -= left;
2097 uiop->uio_offset += left;
2098 uiop->uio_resid -= left;
2099 }
2100
2101 /*
2102 * We are now either at the end of the directory or have filled the
2103 * block.
2104 */
2105 if (bigenough)
2106 dnp->n_direofoffset = uiop->uio_offset;
2107 else {
2108 if (uiop->uio_resid > 0)
2109 printf("EEK! readdirrpc resid > 0\n");
2110 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2111 *cookiep = cookie;
2112 }
2113 nfsmout:
2114 return (error);
2115 }
2116
2117 /*
2118 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2119 */
2120 int
2121 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2122 {
2123 int len, left;
2124 struct dirent *dp;
2125 u_int32_t *tl;
2126 caddr_t cp;
2127 struct vnode *newvp;
2128 nfsuint64 *cookiep;
2129 caddr_t bpos, dpos, dpossav1, dpossav2;
2130 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2131 struct nameidata nami, *ndp = &nami;
2132 struct componentname *cnp = &ndp->ni_cnd;
2133 nfsuint64 cookie;
2134 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2135 struct nfsnode *dnp = VTONFS(vp), *np;
2136 nfsfh_t *fhp;
2137 u_quad_t fileno;
2138 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2139 int attrflag, fhsize;
2140
2141 #ifndef nolint
2142 dp = NULL;
2143 #endif
2144 #ifndef DIAGNOSTIC
2145 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2146 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2147 panic("nfs readdirplusrpc bad uio");
2148 #endif
2149 ndp->ni_dvp = vp;
2150 newvp = NULLVP;
2151
2152 /*
2153 * If there is no cookie, assume directory was stale.
2154 */
2155 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2156 if (cookiep)
2157 cookie = *cookiep;
2158 else
2159 return (NFSERR_BAD_COOKIE);
2160 /*
2161 * Loop around doing readdir rpc's of size nm_readdirsize
2162 * truncated to a multiple of DIRBLKSIZ.
2163 * The stopping criteria is EOF or buffer full.
2164 */
2165 while (more_dirs && bigenough) {
2166 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2167 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2168 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2169 mb = mreq;
2170 bpos = mtod(mb, caddr_t);
2171 nfsm_fhtom(vp, 1);
2172 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2173 *tl++ = cookie.nfsuquad[0];
2174 *tl++ = cookie.nfsuquad[1];
2175 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2176 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2177 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2178 *tl = txdr_unsigned(nmp->nm_rsize);
2179 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2180 nfsm_postop_attr(vp, attrflag);
2181 if (error) {
2182 m_freem(mrep);
2183 goto nfsmout;
2184 }
2185 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2186 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2187 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2188 more_dirs = fxdr_unsigned(int, *tl);
2189
2190 /* loop thru the dir entries, doctoring them to 4bsd form */
2191 while (more_dirs && bigenough) {
2192 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2193 fileno = fxdr_hyper(tl);
2194 len = fxdr_unsigned(int, *(tl + 2));
2195 if (len <= 0 || len > NFS_MAXNAMLEN) {
2196 error = EBADRPC;
2197 m_freem(mrep);
2198 goto nfsmout;
2199 }
2200 tlen = nfsm_rndup(len);
2201 if (tlen == len)
2202 tlen += 4; /* To ensure null termination*/
2203 left = DIRBLKSIZ - blksiz;
2204 if ((tlen + DIRHDSIZ) > left) {
2205 dp->d_reclen += left;
2206 uiop->uio_iov->iov_base =
2207 (char *)uiop->uio_iov->iov_base + left;
2208 uiop->uio_iov->iov_len -= left;
2209 uiop->uio_offset += left;
2210 uiop->uio_resid -= left;
2211 blksiz = 0;
2212 }
2213 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2214 bigenough = 0;
2215 if (bigenough) {
2216 dp = (struct dirent *)uiop->uio_iov->iov_base;
2217 dp->d_fileno = (int)fileno;
2218 dp->d_namlen = len;
2219 dp->d_reclen = tlen + DIRHDSIZ;
2220 dp->d_type = DT_UNKNOWN;
2221 blksiz += dp->d_reclen;
2222 if (blksiz == DIRBLKSIZ)
2223 blksiz = 0;
2224 uiop->uio_offset += DIRHDSIZ;
2225 uiop->uio_resid -= DIRHDSIZ;
2226 uiop->uio_iov->iov_base =
2227 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2228 uiop->uio_iov->iov_len -= DIRHDSIZ;
2229 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2230 cnp->cn_namelen = len;
2231 nfsm_mtouio(uiop, len);
2232 cp = uiop->uio_iov->iov_base;
2233 tlen -= len;
2234 *cp = '\0';
2235 uiop->uio_iov->iov_base =
2236 (char *)uiop->uio_iov->iov_base + tlen;
2237 uiop->uio_iov->iov_len -= tlen;
2238 uiop->uio_offset += tlen;
2239 uiop->uio_resid -= tlen;
2240 } else
2241 nfsm_adv(nfsm_rndup(len));
2242 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2243 if (bigenough) {
2244 cookie.nfsuquad[0] = *tl++;
2245 cookie.nfsuquad[1] = *tl++;
2246 } else
2247 tl += 2;
2248
2249 /*
2250 * Since the attributes are before the file handle
2251 * (sigh), we must skip over the attributes and then
2252 * come back and get them.
2253 */
2254 attrflag = fxdr_unsigned(int, *tl);
2255 if (attrflag) {
2256 dpossav1 = dpos;
2257 mdsav1 = md;
2258 nfsm_adv(NFSX_V3FATTR);
2259 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2260 doit = fxdr_unsigned(int, *tl);
2261 /*
2262 * Skip loading the attrs for "..". There's a
2263 * race between loading the attrs here and
2264 * lookups that look for the directory currently
2265 * being read (in the parent). We try to acquire
2266 * the exclusive lock on ".." here, owning the
2267 * lock on the directory being read. Lookup will
2268 * hold the lock on ".." and try to acquire the
2269 * lock on the directory being read.
2270 *
2271 * There are other ways of fixing this, one would
2272 * be to do a trylock on the ".." vnode and skip
2273 * loading the attrs on ".." if it happens to be
2274 * locked by another process. But skipping the
2275 * attrload on ".." seems the easiest option.
2276 */
2277 if (strcmp(dp->d_name, "..") == 0) {
2278 doit = 0;
2279 /*
2280 * We've already skipped over the attrs,
2281 * skip over the filehandle. And store d_type
2282 * as VDIR.
2283 */
2284 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2285 i = fxdr_unsigned(int, *tl);
2286 nfsm_adv(nfsm_rndup(i));
2287 dp->d_type = IFTODT(VTTOIF(VDIR));
2288 }
2289 if (doit) {
2290 nfsm_getfh(fhp, fhsize, 1);
2291 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2292 VREF(vp);
2293 newvp = vp;
2294 np = dnp;
2295 } else {
2296 error = nfs_nget(vp->v_mount, fhp,
2297 fhsize, &np, LK_EXCLUSIVE);
2298 if (error)
2299 doit = 0;
2300 else
2301 newvp = NFSTOV(np);
2302 }
2303 }
2304 if (doit && bigenough) {
2305 dpossav2 = dpos;
2306 dpos = dpossav1;
2307 mdsav2 = md;
2308 md = mdsav1;
2309 nfsm_loadattr(newvp, NULL);
2310 dpos = dpossav2;
2311 md = mdsav2;
2312 dp->d_type =
2313 IFTODT(VTTOIF(np->n_vattr.va_type));
2314 ndp->ni_vp = newvp;
2315 /* Update n_ctime, so subsequent lookup doesn't purge entry */
2316 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
2317 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2318 }
2319 } else {
2320 /* Just skip over the file handle */
2321 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2322 i = fxdr_unsigned(int, *tl);
2323 if (i) {
2324 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2325 fhsize = fxdr_unsigned(int, *tl);
2326 nfsm_adv(nfsm_rndup(fhsize));
2327 }
2328 }
2329 if (newvp != NULLVP) {
2330 if (newvp == vp)
2331 vrele(newvp);
2332 else
2333 vput(newvp);
2334 newvp = NULLVP;
2335 }
2336 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2337 more_dirs = fxdr_unsigned(int, *tl);
2338 }
2339 /*
2340 * If at end of rpc data, get the eof boolean
2341 */
2342 if (!more_dirs) {
2343 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2344 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2345 }
2346 m_freem(mrep);
2347 }
2348 /*
2349 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2350 * by increasing d_reclen for the last record.
2351 */
2352 if (blksiz > 0) {
2353 left = DIRBLKSIZ - blksiz;
2354 dp->d_reclen += left;
2355 uiop->uio_iov->iov_base =
2356 (char *)uiop->uio_iov->iov_base + left;
2357 uiop->uio_iov->iov_len -= left;
2358 uiop->uio_offset += left;
2359 uiop->uio_resid -= left;
2360 }
2361
2362 /*
2363 * We are now either at the end of the directory or have filled the
2364 * block.
2365 */
2366 if (bigenough)
2367 dnp->n_direofoffset = uiop->uio_offset;
2368 else {
2369 if (uiop->uio_resid > 0)
2370 printf("EEK! readdirplusrpc resid > 0\n");
2371 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2372 *cookiep = cookie;
2373 }
2374 nfsmout:
2375 if (newvp != NULLVP) {
2376 if (newvp == vp)
2377 vrele(newvp);
2378 else
2379 vput(newvp);
2380 newvp = NULLVP;
2381 }
2382 return (error);
2383 }
2384
2385 /*
2386 * Silly rename. To make the NFS filesystem that is stateless look a little
2387 * more like the "ufs" a remove of an active vnode is translated to a rename
2388 * to a funny looking filename that is removed by nfs_inactive on the
2389 * nfsnode. There is the potential for another process on a different client
2390 * to create the same funny name between the nfs_lookitup() fails and the
2391 * nfs_rename() completes, but...
2392 */
2393 static int
2394 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2395 {
2396 struct sillyrename *sp;
2397 struct nfsnode *np;
2398 int error;
2399 short pid;
2400 unsigned int lticks;
2401
2402 cache_purge(dvp);
2403 np = VTONFS(vp);
2404 #ifndef DIAGNOSTIC
2405 if (vp->v_type == VDIR)
2406 panic("nfs: sillyrename dir");
2407 #endif
2408 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2409 M_NFSREQ, M_WAITOK);
2410 sp->s_cred = crhold(cnp->cn_cred);
2411 sp->s_dvp = dvp;
2412 sp->s_removeit = nfs_removeit;
2413 VREF(dvp);
2414
2415 /*
2416 * Fudge together a funny name.
2417 * Changing the format of the funny name to accomodate more
2418 * sillynames per directory.
2419 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2420 * CPU ticks since boot.
2421 */
2422 pid = cnp->cn_thread->td_proc->p_pid;
2423 lticks = (unsigned int)ticks;
2424 for ( ; ; ) {
2425 sp->s_namlen = sprintf(sp->s_name,
2426 ".nfs.%08x.%04x4.4", lticks,
2427 pid);
2428 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2429 cnp->cn_thread, NULL))
2430 break;
2431 lticks++;
2432 }
2433 error = nfs_renameit(dvp, cnp, sp);
2434 if (error)
2435 goto bad;
2436 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2437 cnp->cn_thread, &np);
2438 np->n_sillyrename = sp;
2439 return (0);
2440 bad:
2441 vrele(sp->s_dvp);
2442 crfree(sp->s_cred);
2443 free((caddr_t)sp, M_NFSREQ);
2444 return (error);
2445 }
2446
2447 /*
2448 * Look up a file name and optionally either update the file handle or
2449 * allocate an nfsnode, depending on the value of npp.
2450 * npp == NULL --> just do the lookup
2451 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2452 * handled too
2453 * *npp != NULL --> update the file handle in the vnode
2454 */
2455 static int
2456 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2457 struct thread *td, struct nfsnode **npp)
2458 {
2459 struct vnode *newvp = NULL;
2460 struct nfsnode *np, *dnp = VTONFS(dvp);
2461 caddr_t bpos, dpos;
2462 int error = 0, fhlen, attrflag;
2463 struct mbuf *mreq, *mrep, *md, *mb;
2464 nfsfh_t *nfhp;
2465 int v3 = NFS_ISV3(dvp);
2466
2467 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2468 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2469 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2470 mb = mreq;
2471 bpos = mtod(mb, caddr_t);
2472 nfsm_fhtom(dvp, v3);
2473 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2474 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2475 if (npp && !error) {
2476 nfsm_getfh(nfhp, fhlen, v3);
2477 if (*npp) {
2478 np = *npp;
2479 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2480 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2481 np->n_fhp = &np->n_fh;
2482 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2483 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2484 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2485 np->n_fhsize = fhlen;
2486 newvp = NFSTOV(np);
2487 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2488 VREF(dvp);
2489 newvp = dvp;
2490 } else {
2491 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2492 if (error) {
2493 m_freem(mrep);
2494 return (error);
2495 }
2496 newvp = NFSTOV(np);
2497 }
2498 if (v3) {
2499 nfsm_postop_attr(newvp, attrflag);
2500 if (!attrflag && *npp == NULL) {
2501 m_freem(mrep);
2502 if (newvp == dvp)
2503 vrele(newvp);
2504 else
2505 vput(newvp);
2506 return (ENOENT);
2507 }
2508 } else
2509 nfsm_loadattr(newvp, NULL);
2510 }
2511 m_freem(mrep);
2512 nfsmout:
2513 if (npp && *npp == NULL) {
2514 if (error) {
2515 if (newvp) {
2516 if (newvp == dvp)
2517 vrele(newvp);
2518 else
2519 vput(newvp);
2520 }
2521 } else
2522 *npp = np;
2523 }
2524 return (error);
2525 }
2526
2527 /*
2528 * Nfs Version 3 commit rpc
2529 */
2530 int
2531 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2532 struct thread *td)
2533 {
2534 u_int32_t *tl;
2535 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2536 caddr_t bpos, dpos;
2537 int error = 0, wccflag = NFSV3_WCCRATTR;
2538 struct mbuf *mreq, *mrep, *md, *mb;
2539
2540 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2541 return (0);
2542 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2543 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2544 mb = mreq;
2545 bpos = mtod(mb, caddr_t);
2546 nfsm_fhtom(vp, 1);
2547 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2548 txdr_hyper(offset, tl);
2549 tl += 2;
2550 *tl = txdr_unsigned(cnt);
2551 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2552 nfsm_wcc_data(vp, wccflag);
2553 if (!error) {
2554 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2555 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2556 NFSX_V3WRITEVERF)) {
2557 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2558 NFSX_V3WRITEVERF);
2559 error = NFSERR_STALEWRITEVERF;
2560 }
2561 }
2562 m_freem(mrep);
2563 nfsmout:
2564 return (error);
2565 }
2566
2567 /*
2568 * Strategy routine.
2569 * For async requests when nfsiod(s) are running, queue the request by
2570 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2571 * request.
2572 */
2573 static int
2574 nfs_strategy(struct vop_strategy_args *ap)
2575 {
2576 struct buf *bp = ap->a_bp;
2577 struct ucred *cr;
2578
2579 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2580 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
2581
2582 if (bp->b_iocmd == BIO_READ)
2583 cr = bp->b_rcred;
2584 else
2585 cr = bp->b_wcred;
2586
2587 /*
2588 * If the op is asynchronous and an i/o daemon is waiting
2589 * queue the request, wake it up and wait for completion
2590 * otherwise just do it ourselves.
2591 */
2592 if ((bp->b_flags & B_ASYNC) == 0 ||
2593 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2594 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2595 return (0);
2596 }
2597
2598 /*
2599 * fsync vnode op. Just call nfs_flush() with commit == 1.
2600 */
2601 /* ARGSUSED */
2602 static int
2603 nfs_fsync(struct vop_fsync_args *ap)
2604 {
2605
2606 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2607 }
2608
2609 /*
2610 * Flush all the blocks associated with a vnode.
2611 * Walk through the buffer pool and push any dirty pages
2612 * associated with the vnode.
2613 */
2614 static int
2615 nfs_flush(struct vnode *vp, int waitfor, struct thread *td,
2616 int commit)
2617 {
2618 struct nfsnode *np = VTONFS(vp);
2619 struct buf *bp;
2620 int i;
2621 struct buf *nbp;
2622 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2623 int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2624 int passone = 1;
2625 u_quad_t off, endoff, toff;
2626 struct ucred* wcred = NULL;
2627 struct buf **bvec = NULL;
2628 #ifndef NFS_COMMITBVECSIZ
2629 #define NFS_COMMITBVECSIZ 20
2630 #endif
2631 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2632 int bvecsize = 0, bveccount;
2633
2634 if (nmp->nm_flag & NFSMNT_INT)
2635 slpflag = PCATCH;
2636 if (!commit)
2637 passone = 0;
2638 /*
2639 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2640 * server, but has not been committed to stable storage on the server
2641 * yet. On the first pass, the byte range is worked out and the commit
2642 * rpc is done. On the second pass, nfs_writebp() is called to do the
2643 * job.
2644 */
2645 again:
2646 off = (u_quad_t)-1;
2647 endoff = 0;
2648 bvecpos = 0;
2649 if (NFS_ISV3(vp) && commit) {
2650 s = splbio();
2651 if (bvec != NULL && bvec != bvec_on_stack)
2652 free(bvec, M_TEMP);
2653 /*
2654 * Count up how many buffers waiting for a commit.
2655 */
2656 bveccount = 0;
2657 VI_LOCK(vp);
2658 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2659 if (BUF_REFCNT(bp) == 0 &&
2660 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2661 == (B_DELWRI | B_NEEDCOMMIT))
2662 bveccount++;
2663 }
2664 /*
2665 * Allocate space to remember the list of bufs to commit. It is
2666 * important to use M_NOWAIT here to avoid a race with nfs_write.
2667 * If we can't get memory (for whatever reason), we will end up
2668 * committing the buffers one-by-one in the loop below.
2669 */
2670 if (bveccount > NFS_COMMITBVECSIZ) {
2671 /*
2672 * Release the vnode interlock to avoid a lock
2673 * order reversal.
2674 */
2675 VI_UNLOCK(vp);
2676 bvec = (struct buf **)
2677 malloc(bveccount * sizeof(struct buf *),
2678 M_TEMP, M_NOWAIT);
2679 VI_LOCK(vp);
2680 if (bvec == NULL) {
2681 bvec = bvec_on_stack;
2682 bvecsize = NFS_COMMITBVECSIZ;
2683 } else
2684 bvecsize = bveccount;
2685 } else {
2686 bvec = bvec_on_stack;
2687 bvecsize = NFS_COMMITBVECSIZ;
2688 }
2689 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2690 if (bvecpos >= bvecsize)
2691 break;
2692 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2693 nbp = TAILQ_NEXT(bp, b_bobufs);
2694 continue;
2695 }
2696 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2697 (B_DELWRI | B_NEEDCOMMIT)) {
2698 BUF_UNLOCK(bp);
2699 nbp = TAILQ_NEXT(bp, b_bobufs);
2700 continue;
2701 }
2702 VI_UNLOCK(vp);
2703 bremfree(bp);
2704 /*
2705 * Work out if all buffers are using the same cred
2706 * so we can deal with them all with one commit.
2707 *
2708 * NOTE: we are not clearing B_DONE here, so we have
2709 * to do it later on in this routine if we intend to
2710 * initiate I/O on the bp.
2711 *
2712 * Note: to avoid loopback deadlocks, we do not
2713 * assign b_runningbufspace.
2714 */
2715 if (wcred == NULL)
2716 wcred = bp->b_wcred;
2717 else if (wcred != bp->b_wcred)
2718 wcred = NOCRED;
2719 vfs_busy_pages(bp, 1);
2720
2721 VI_LOCK(vp);
2722 /*
2723 * bp is protected by being locked, but nbp is not
2724 * and vfs_busy_pages() may sleep. We have to
2725 * recalculate nbp.
2726 */
2727 nbp = TAILQ_NEXT(bp, b_bobufs);
2728
2729 /*
2730 * A list of these buffers is kept so that the
2731 * second loop knows which buffers have actually
2732 * been committed. This is necessary, since there
2733 * may be a race between the commit rpc and new
2734 * uncommitted writes on the file.
2735 */
2736 bvec[bvecpos++] = bp;
2737 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2738 bp->b_dirtyoff;
2739 if (toff < off)
2740 off = toff;
2741 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2742 if (toff > endoff)
2743 endoff = toff;
2744 }
2745 splx(s);
2746 VI_UNLOCK(vp);
2747 }
2748 if (bvecpos > 0) {
2749 /*
2750 * Commit data on the server, as required.
2751 * If all bufs are using the same wcred, then use that with
2752 * one call for all of them, otherwise commit each one
2753 * separately.
2754 */
2755 if (wcred != NOCRED)
2756 retv = nfs_commit(vp, off, (int)(endoff - off),
2757 wcred, td);
2758 else {
2759 retv = 0;
2760 for (i = 0; i < bvecpos; i++) {
2761 off_t off, size;
2762 bp = bvec[i];
2763 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2764 bp->b_dirtyoff;
2765 size = (u_quad_t)(bp->b_dirtyend
2766 - bp->b_dirtyoff);
2767 retv = nfs_commit(vp, off, (int)size,
2768 bp->b_wcred, td);
2769 if (retv) break;
2770 }
2771 }
2772
2773 if (retv == NFSERR_STALEWRITEVERF)
2774 nfs_clearcommit(vp->v_mount);
2775
2776 /*
2777 * Now, either mark the blocks I/O done or mark the
2778 * blocks dirty, depending on whether the commit
2779 * succeeded.
2780 */
2781 for (i = 0; i < bvecpos; i++) {
2782 bp = bvec[i];
2783 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2784 if (retv) {
2785 /*
2786 * Error, leave B_DELWRI intact
2787 */
2788 vfs_unbusy_pages(bp);
2789 brelse(bp);
2790 } else {
2791 /*
2792 * Success, remove B_DELWRI ( bundirty() ).
2793 *
2794 * b_dirtyoff/b_dirtyend seem to be NFS
2795 * specific. We should probably move that
2796 * into bundirty(). XXX
2797 */
2798 s = splbio();
2799 bufobj_wref(&vp->v_bufobj);
2800 bp->b_flags |= B_ASYNC;
2801 bundirty(bp);
2802 bp->b_flags &= ~B_DONE;
2803 bp->b_ioflags &= ~BIO_ERROR;
2804 bp->b_dirtyoff = bp->b_dirtyend = 0;
2805 splx(s);
2806 bufdone(bp);
2807 }
2808 }
2809 }
2810
2811 /*
2812 * Start/do any write(s) that are required.
2813 */
2814 loop:
2815 s = splbio();
2816 VI_LOCK(vp);
2817 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2818 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2819 if (waitfor != MNT_WAIT || passone)
2820 continue;
2821
2822 error = BUF_TIMELOCK(bp,
2823 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2824 VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
2825 splx(s);
2826 if (error == 0)
2827 panic("nfs_fsync: inconsistent lock");
2828 if (error == ENOLCK)
2829 goto loop;
2830 if (nfs_sigintr(nmp, NULL, td)) {
2831 error = EINTR;
2832 goto done;
2833 }
2834 if (slpflag == PCATCH) {
2835 slpflag = 0;
2836 slptimeo = 2 * hz;
2837 }
2838 goto loop;
2839 }
2840 if ((bp->b_flags & B_DELWRI) == 0)
2841 panic("nfs_fsync: not dirty");
2842 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2843 BUF_UNLOCK(bp);
2844 continue;
2845 }
2846 VI_UNLOCK(vp);
2847 bremfree(bp);
2848 if (passone || !commit)
2849 bp->b_flags |= B_ASYNC;
2850 else
2851 bp->b_flags |= B_ASYNC;
2852 splx(s);
2853 bwrite(bp);
2854 if (nfs_sigintr(nmp, NULL, td)) {
2855 error = EINTR;
2856 goto done;
2857 }
2858 goto loop;
2859 }
2860 splx(s);
2861 if (passone) {
2862 passone = 0;
2863 VI_UNLOCK(vp);
2864 goto again;
2865 }
2866 if (waitfor == MNT_WAIT) {
2867 while (vp->v_bufobj.bo_numoutput) {
2868 error = bufobj_wwait(&vp->v_bufobj, slpflag, slptimeo);
2869 if (error) {
2870 VI_UNLOCK(vp);
2871 error = nfs_sigintr(nmp, NULL, td);
2872 if (error)
2873 goto done;
2874 if (slpflag == PCATCH) {
2875 slpflag = 0;
2876 slptimeo = 2 * hz;
2877 }
2878 VI_LOCK(vp);
2879 }
2880 }
2881 if (vp->v_bufobj.bo_dirty.bv_cnt != 0 && commit) {
2882 VI_UNLOCK(vp);
2883 goto loop;
2884 }
2885 VI_UNLOCK(vp);
2886 /*
2887 * Wait for all the async IO requests to drain
2888 */
2889 while (np->n_directio_asyncwr > 0) {
2890 np->n_flag |= NFSYNCWAIT;
2891 error = nfs_tsleep(td, (caddr_t)&np->n_directio_asyncwr,
2892 slpflag | (PRIBIO + 1), "nfsfsync", 0);
2893 if (error) {
2894 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
2895 error = EINTR;
2896 goto done;
2897 }
2898 }
2899 }
2900
2901 } else
2902 VI_UNLOCK(vp);
2903 if (np->n_flag & NWRITEERR) {
2904 error = np->n_error;
2905 np->n_flag &= ~NWRITEERR;
2906 }
2907 if (commit && vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
2908 vp->v_bufobj.bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2909 np->n_flag &= ~NMODIFIED;
2910 done:
2911 if (bvec != NULL && bvec != bvec_on_stack)
2912 free(bvec, M_TEMP);
2913 return (error);
2914 }
2915
2916 /*
2917 * NFS advisory byte-level locks.
2918 */
2919 static int
2920 nfs_advlock(struct vop_advlock_args *ap)
2921 {
2922
2923 if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2924 struct nfsnode *np = VTONFS(ap->a_vp);
2925
2926 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
2927 }
2928 return (nfs_dolock(ap));
2929 }
2930
2931 /*
2932 * Print out the contents of an nfsnode.
2933 */
2934 static int
2935 nfs_print(struct vop_print_args *ap)
2936 {
2937 struct vnode *vp = ap->a_vp;
2938 struct nfsnode *np = VTONFS(vp);
2939
2940 printf("\tfileid %ld fsid 0x%x",
2941 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
2942 if (vp->v_type == VFIFO)
2943 fifo_printinfo(vp);
2944 printf("\n");
2945 return (0);
2946 }
2947
2948 /*
2949 * This is the "real" nfs::bwrite(struct buf*).
2950 * We set B_CACHE if this is a VMIO buffer.
2951 */
2952 int
2953 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
2954 {
2955 int s;
2956 int oldflags = bp->b_flags;
2957 #if 0
2958 int retv = 1;
2959 off_t off;
2960 #endif
2961
2962 if (BUF_REFCNT(bp) == 0)
2963 panic("bwrite: buffer is not locked???");
2964
2965 if (bp->b_flags & B_INVAL) {
2966 brelse(bp);
2967 return(0);
2968 }
2969
2970 bp->b_flags |= B_CACHE;
2971
2972 /*
2973 * Undirty the bp. We will redirty it later if the I/O fails.
2974 */
2975
2976 s = splbio();
2977 bundirty(bp);
2978 bp->b_flags &= ~B_DONE;
2979 bp->b_ioflags &= ~BIO_ERROR;
2980 bp->b_iocmd = BIO_WRITE;
2981
2982 bufobj_wref(bp->b_bufobj);
2983 curthread->td_proc->p_stats->p_ru.ru_oublock++;
2984 splx(s);
2985
2986 /*
2987 * Note: to avoid loopback deadlocks, we do not
2988 * assign b_runningbufspace.
2989 */
2990 vfs_busy_pages(bp, 1);
2991
2992 BUF_KERNPROC(bp);
2993 bp->b_iooffset = dbtob(bp->b_blkno);
2994 bstrategy(bp);
2995
2996 if( (oldflags & B_ASYNC) == 0) {
2997 int rtval = bufwait(bp);
2998
2999 if (oldflags & B_DELWRI) {
3000 s = splbio();
3001 reassignbuf(bp);
3002 splx(s);
3003 }
3004
3005 brelse(bp);
3006 return (rtval);
3007 }
3008
3009 return (0);
3010 }
3011
3012 /*
3013 * nfs special file access vnode op.
3014 * Essentially just get vattr and then imitate iaccess() since the device is
3015 * local to the client.
3016 */
3017 static int
3018 nfsspec_access(struct vop_access_args *ap)
3019 {
3020 struct vattr *vap;
3021 struct ucred *cred = ap->a_cred;
3022 struct vnode *vp = ap->a_vp;
3023 mode_t mode = ap->a_mode;
3024 struct vattr vattr;
3025 int error;
3026
3027 /*
3028 * Disallow write attempts on filesystems mounted read-only;
3029 * unless the file is a socket, fifo, or a block or character
3030 * device resident on the filesystem.
3031 */
3032 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3033 switch (vp->v_type) {
3034 case VREG:
3035 case VDIR:
3036 case VLNK:
3037 return (EROFS);
3038 default:
3039 break;
3040 }
3041 }
3042 vap = &vattr;
3043 error = VOP_GETATTR(vp, vap, cred, ap->a_td);
3044 if (error)
3045 return (error);
3046 return (vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3047 mode, cred, NULL));
3048 }
3049
3050 /*
3051 * Read wrapper for fifos.
3052 */
3053 static int
3054 nfsfifo_read(struct vop_read_args *ap)
3055 {
3056 struct nfsnode *np = VTONFS(ap->a_vp);
3057
3058 /*
3059 * Set access flag.
3060 */
3061 np->n_flag |= NACC;
3062 getnanotime(&np->n_atim);
3063 return (fifo_specops.vop_read(ap));
3064 }
3065
3066 /*
3067 * Write wrapper for fifos.
3068 */
3069 static int
3070 nfsfifo_write(struct vop_write_args *ap)
3071 {
3072 struct nfsnode *np = VTONFS(ap->a_vp);
3073
3074 /*
3075 * Set update flag.
3076 */
3077 np->n_flag |= NUPD;
3078 getnanotime(&np->n_mtim);
3079 return (fifo_specops.vop_write(ap));
3080 }
3081
3082 /*
3083 * Close wrapper for fifos.
3084 *
3085 * Update the times on the nfsnode then do fifo close.
3086 */
3087 static int
3088 nfsfifo_close(struct vop_close_args *ap)
3089 {
3090 struct vnode *vp = ap->a_vp;
3091 struct nfsnode *np = VTONFS(vp);
3092 struct vattr vattr;
3093 struct timespec ts;
3094
3095 if (np->n_flag & (NACC | NUPD)) {
3096 getnanotime(&ts);
3097 if (np->n_flag & NACC)
3098 np->n_atim = ts;
3099 if (np->n_flag & NUPD)
3100 np->n_mtim = ts;
3101 np->n_flag |= NCHG;
3102 if (vrefcnt(vp) == 1 &&
3103 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3104 VATTR_NULL(&vattr);
3105 if (np->n_flag & NACC)
3106 vattr.va_atime = np->n_atim;
3107 if (np->n_flag & NUPD)
3108 vattr.va_mtime = np->n_mtim;
3109 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
3110 }
3111 }
3112 return (fifo_specops.vop_close(ap));
3113 }
3114
3115 /*
3116 * Just call nfs_writebp() with the force argument set to 1.
3117 *
3118 * NOTE: B_DONE may or may not be set in a_bp on call.
3119 */
3120 static int
3121 nfs_bwrite(struct buf *bp)
3122 {
3123
3124 return (nfs_writebp(bp, 1, curthread));
3125 }
3126
3127 struct buf_ops buf_ops_nfs = {
3128 .bop_name = "buf_ops_nfs",
3129 .bop_write = nfs_bwrite,
3130 .bop_strategy = bufstrategy,
3131 .bop_sync = bufsync,
3132 .bop_bdflush = bufbdflush,
3133 };
Cache object: 96e8411ed689381cc2b23a9348678c03
|