2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
32 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
39 * vnode op calls for Sun NFS version 2 and 3
43 #include "opt_kdtrace.h"
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/systm.h>
48 #include <sys/resourcevar.h>
50 #include <sys/mount.h>
54 #include <sys/malloc.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>
63 #include <sys/sysctl.h>
64 #include <sys/signalvar.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
70 #include <fs/fifofs/fifo.h>
72 #include <nfs/nfsproto.h>
73 #include <nfsclient/nfs.h>
74 #include <nfsclient/nfsnode.h>
75 #include <nfsclient/nfsmount.h>
76 #include <nfs/nfs_kdtrace.h>
77 #include <nfs/nfs_lock.h>
78 #include <nfs/xdr_subs.h>
79 #include <nfsclient/nfsm_subs.h>
82 #include <netinet/in.h>
83 #include <netinet/in_var.h>
85 #include <machine/stdarg.h>
88 #include <sys/dtrace_bsd.h>
90 dtrace_nfsclient_accesscache_flush_probe_func_t
91 dtrace_nfsclient_accesscache_flush_done_probe;
92 uint32_t nfsclient_accesscache_flush_done_id;
94 dtrace_nfsclient_accesscache_get_probe_func_t
95 dtrace_nfsclient_accesscache_get_hit_probe,
96 dtrace_nfsclient_accesscache_get_miss_probe;
97 uint32_t nfsclient_accesscache_get_hit_id;
98 uint32_t nfsclient_accesscache_get_miss_id;
100 dtrace_nfsclient_accesscache_load_probe_func_t
101 dtrace_nfsclient_accesscache_load_done_probe;
102 uint32_t nfsclient_accesscache_load_done_id;
103 #endif /* !KDTRACE_HOOKS */
110 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
111 * calls are not in getblk() and brelse() so that they would not be necessary
115 #define vfs_busy_pages(bp, f)
118 static vop_read_t nfsfifo_read;
119 static vop_write_t nfsfifo_write;
120 static vop_close_t nfsfifo_close;
121 static int nfs_flush(struct vnode *, int, int);
122 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *);
123 static vop_lookup_t nfs_lookup;
124 static vop_create_t nfs_create;
125 static vop_mknod_t nfs_mknod;
126 static vop_open_t nfs_open;
127 static vop_close_t nfs_close;
128 static vop_access_t nfs_access;
129 static vop_getattr_t nfs_getattr;
130 static vop_setattr_t nfs_setattr;
131 static vop_read_t nfs_read;
132 static vop_fsync_t nfs_fsync;
133 static vop_remove_t nfs_remove;
134 static vop_link_t nfs_link;
135 static vop_rename_t nfs_rename;
136 static vop_mkdir_t nfs_mkdir;
137 static vop_rmdir_t nfs_rmdir;
138 static vop_symlink_t nfs_symlink;
139 static vop_readdir_t nfs_readdir;
140 static vop_strategy_t nfs_strategy;
141 static int nfs_lookitup(struct vnode *, const char *, int,
142 struct ucred *, struct thread *, struct nfsnode **);
143 static int nfs_sillyrename(struct vnode *, struct vnode *,
144 struct componentname *);
145 static vop_access_t nfsspec_access;
146 static vop_readlink_t nfs_readlink;
147 static vop_print_t nfs_print;
148 static vop_advlock_t nfs_advlock;
149 static vop_advlockasync_t nfs_advlockasync;
152 * Global vfs data structures for nfs
154 struct vop_vector nfs_vnodeops = {
155 .vop_default = &default_vnodeops,
156 .vop_access = nfs_access,
157 .vop_advlock = nfs_advlock,
158 .vop_advlockasync = nfs_advlockasync,
159 .vop_close = nfs_close,
160 .vop_create = nfs_create,
161 .vop_fsync = nfs_fsync,
162 .vop_getattr = nfs_getattr,
163 .vop_getpages = nfs_getpages,
164 .vop_putpages = nfs_putpages,
165 .vop_inactive = nfs_inactive,
166 .vop_link = nfs_link,
167 .vop_lookup = nfs_lookup,
168 .vop_mkdir = nfs_mkdir,
169 .vop_mknod = nfs_mknod,
170 .vop_open = nfs_open,
171 .vop_print = nfs_print,
172 .vop_read = nfs_read,
173 .vop_readdir = nfs_readdir,
174 .vop_readlink = nfs_readlink,
175 .vop_reclaim = nfs_reclaim,
176 .vop_remove = nfs_remove,
177 .vop_rename = nfs_rename,
178 .vop_rmdir = nfs_rmdir,
179 .vop_setattr = nfs_setattr,
180 .vop_strategy = nfs_strategy,
181 .vop_symlink = nfs_symlink,
182 .vop_write = nfs_write,
185 struct vop_vector nfs_fifoops = {
186 .vop_default = &fifo_specops,
187 .vop_access = nfsspec_access,
188 .vop_close = nfsfifo_close,
189 .vop_fsync = nfs_fsync,
190 .vop_getattr = nfs_getattr,
191 .vop_inactive = nfs_inactive,
192 .vop_print = nfs_print,
193 .vop_read = nfsfifo_read,
194 .vop_reclaim = nfs_reclaim,
195 .vop_setattr = nfs_setattr,
196 .vop_write = nfsfifo_write,
199 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
200 struct componentname *cnp, struct vattr *vap);
201 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
202 struct ucred *cred, struct thread *td);
203 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
204 int fnamelen, struct vnode *tdvp,
205 const char *tnameptr, int tnamelen,
206 struct ucred *cred, struct thread *td);
207 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
208 struct sillyrename *sp);
213 struct mtx nfs_iod_mtx;
214 enum nfsiod_state nfs_iodwant[NFS_MAXASYNCDAEMON];
215 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
216 int nfs_numasync = 0;
217 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
219 SYSCTL_DECL(_vfs_oldnfs);
221 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
222 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
223 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
225 static int nfs_prime_access_cache = 0;
226 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
227 &nfs_prime_access_cache, 0,
228 "Prime NFS ACCESS cache when fetching attributes");
230 static int nfsv3_commit_on_close = 0;
231 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
232 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
234 static int nfs_clean_pages_on_close = 1;
235 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
236 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
238 int nfs_directio_enable = 0;
239 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
240 &nfs_directio_enable, 0, "Enable NFS directio");
243 * This sysctl allows other processes to mmap a file that has been opened
244 * O_DIRECT by a process. In general, having processes mmap the file while
245 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
246 * this by default to prevent DoS attacks - to prevent a malicious user from
247 * opening up files O_DIRECT preventing other users from mmap'ing these
248 * files. "Protected" environments where stricter consistency guarantees are
249 * required can disable this knob. The process that opened the file O_DIRECT
250 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
253 int nfs_directio_allow_mmap = 1;
254 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
255 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
258 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
259 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
261 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
262 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
265 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
266 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
267 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
271 * The list of locks after the description of the lock is the ordering
272 * of other locks acquired with the lock held.
273 * np->n_mtx : Protects the fields in the nfsnode.
275 VI_MTX (acquired indirectly)
276 * nmp->nm_mtx : Protects the fields in the nfsmount.
278 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
279 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
282 * rep->r_mtx : Protects the fields in an nfsreq.
286 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
287 struct ucred *cred, uint32_t *retmode)
291 int error = 0, attrflag, i, lrupos;
293 struct mbuf *mreq, *mrep, *md, *mb;
296 struct nfsnode *np = VTONFS(vp);
298 nfsstats.rpccnt[NFSPROC_ACCESS]++;
299 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
301 bpos = mtod(mb, caddr_t);
303 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
304 *tl = txdr_unsigned(wmode);
305 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
306 nfsm_postop_attr(vp, attrflag);
309 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
310 rmode = fxdr_unsigned(u_int32_t, *tl);
311 mtx_lock(&np->n_mtx);
312 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
313 if (np->n_accesscache[i].uid == cred->cr_uid) {
314 np->n_accesscache[i].mode = rmode;
315 np->n_accesscache[i].stamp = time_second;
318 if (i > 0 && np->n_accesscache[i].stamp <
319 np->n_accesscache[lrupos].stamp)
322 if (i == NFS_ACCESSCACHESIZE) {
323 np->n_accesscache[lrupos].uid = cred->cr_uid;
324 np->n_accesscache[lrupos].mode = rmode;
325 np->n_accesscache[lrupos].stamp = time_second;
327 mtx_unlock(&np->n_mtx);
330 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
336 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
344 * nfs access vnode op.
345 * For nfs version 2, just return ok. File accesses may fail later.
346 * For nfs version 3, use the access rpc to check accessibility. If file modes
347 * are changed on the server, accesses might still fail later.
350 nfs_access(struct vop_access_args *ap)
352 struct vnode *vp = ap->a_vp;
353 int error = 0, i, gotahit;
354 u_int32_t mode, rmode, wmode;
355 int v3 = NFS_ISV3(vp);
356 struct nfsnode *np = VTONFS(vp);
359 * Disallow write attempts on filesystems mounted read-only;
360 * unless the file is a socket, fifo, or a block or character
361 * device resident on the filesystem.
363 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
364 switch (vp->v_type) {
374 * For nfs v3, check to see if we have done this recently, and if
375 * so return our cached result instead of making an ACCESS call.
376 * If not, do an access rpc, otherwise you are stuck emulating
377 * ufs_access() locally using the vattr. This may not be correct,
378 * since the server may apply other access criteria such as
379 * client uid-->server uid mapping that we do not know about.
382 if (ap->a_accmode & VREAD)
383 mode = NFSV3ACCESS_READ;
386 if (vp->v_type != VDIR) {
387 if (ap->a_accmode & VWRITE)
388 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
389 if (ap->a_accmode & VEXEC)
390 mode |= NFSV3ACCESS_EXECUTE;
392 if (ap->a_accmode & VWRITE)
393 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
395 if (ap->a_accmode & VEXEC)
396 mode |= NFSV3ACCESS_LOOKUP;
398 /* XXX safety belt, only make blanket request if caching */
399 if (nfsaccess_cache_timeout > 0) {
400 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
401 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
402 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
408 * Does our cached result allow us to give a definite yes to
412 mtx_lock(&np->n_mtx);
413 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
414 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
415 if (time_second < (np->n_accesscache[i].stamp +
416 nfsaccess_cache_timeout) &&
417 (np->n_accesscache[i].mode & mode) == mode) {
418 nfsstats.accesscache_hits++;
424 mtx_unlock(&np->n_mtx);
427 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
428 ap->a_cred->cr_uid, mode);
430 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
431 ap->a_cred->cr_uid, mode);
435 * Either a no, or a don't know. Go to the wire.
437 nfsstats.accesscache_misses++;
438 error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred,
441 if ((rmode & mode) != mode)
447 if ((error = nfsspec_access(ap)) != 0) {
451 * Attempt to prevent a mapped root from accessing a file
452 * which it shouldn't. We try to read a byte from the file
453 * if the user is root and the file is not zero length.
454 * After calling nfsspec_access, we should have the correct
457 mtx_lock(&np->n_mtx);
458 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
459 && VTONFS(vp)->n_size > 0) {
464 mtx_unlock(&np->n_mtx);
467 auio.uio_iov = &aiov;
471 auio.uio_segflg = UIO_SYSSPACE;
472 auio.uio_rw = UIO_READ;
473 auio.uio_td = ap->a_td;
475 if (vp->v_type == VREG)
476 error = nfs_readrpc(vp, &auio, ap->a_cred);
477 else if (vp->v_type == VDIR) {
479 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
481 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
482 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
484 } else if (vp->v_type == VLNK)
485 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
489 mtx_unlock(&np->n_mtx);
494 int nfs_otw_getattr_avoid = 0;
498 * Check to see if the type is ok
499 * and that deletion is not in progress.
500 * For paged in text files, you will need to flush the page cache
501 * if consistency is lost.
505 nfs_open(struct vop_open_args *ap)
507 struct vnode *vp = ap->a_vp;
508 struct nfsnode *np = VTONFS(vp);
511 int fmode = ap->a_mode;
513 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
517 * Get a valid lease. If cached data is stale, flush it.
519 mtx_lock(&np->n_mtx);
520 if (np->n_flag & NMODIFIED) {
521 mtx_unlock(&np->n_mtx);
522 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
523 if (error == EINTR || error == EIO)
525 mtx_lock(&np->n_mtx);
527 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
528 if (vp->v_type == VDIR)
529 np->n_direofoffset = 0;
530 mtx_unlock(&np->n_mtx);
531 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
534 mtx_lock(&np->n_mtx);
535 np->n_mtime = vattr.va_mtime;
537 mtx_unlock(&np->n_mtx);
538 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
541 mtx_lock(&np->n_mtx);
542 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
543 if (vp->v_type == VDIR)
544 np->n_direofoffset = 0;
545 mtx_unlock(&np->n_mtx);
546 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
547 if (error == EINTR || error == EIO) {
550 mtx_lock(&np->n_mtx);
551 np->n_mtime = vattr.va_mtime;
555 * If the object has >= 1 O_DIRECT active opens, we disable caching.
557 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
558 if (np->n_directio_opens == 0) {
559 mtx_unlock(&np->n_mtx);
560 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
563 mtx_lock(&np->n_mtx);
564 np->n_flag |= NNONCACHE;
566 np->n_directio_opens++;
568 mtx_unlock(&np->n_mtx);
569 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
575 * What an NFS client should do upon close after writing is a debatable issue.
576 * Most NFS clients push delayed writes to the server upon close, basically for
578 * 1 - So that any write errors may be reported back to the client process
579 * doing the close system call. By far the two most likely errors are
580 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
581 * 2 - To put a worst case upper bound on cache inconsistency between
582 * multiple clients for the file.
583 * There is also a consistency problem for Version 2 of the protocol w.r.t.
584 * not being able to tell if other clients are writing a file concurrently,
585 * since there is no way of knowing if the changed modify time in the reply
586 * is only due to the write for this client.
587 * (NFS Version 3 provides weak cache consistency data in the reply that
588 * should be sufficient to detect and handle this case.)
590 * The current code does the following:
591 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
592 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
593 * or commit them (this satisfies 1 and 2 except for the
594 * case where the server crashes after this close but
595 * before the commit RPC, which is felt to be "good
596 * enough". Changing the last argument to nfs_flush() to
597 * a 1 would force a commit operation, if it is felt a
598 * commit is necessary now.
602 nfs_close(struct vop_close_args *ap)
604 struct vnode *vp = ap->a_vp;
605 struct nfsnode *np = VTONFS(vp);
607 int fmode = ap->a_fflag;
609 if (vp->v_type == VREG) {
611 * Examine and clean dirty pages, regardless of NMODIFIED.
612 * This closes a major hole in close-to-open consistency.
613 * We want to push out all dirty pages (and buffers) on
614 * close, regardless of whether they were dirtied by
615 * mmap'ed writes or via write().
617 if (nfs_clean_pages_on_close && vp->v_object) {
618 VM_OBJECT_LOCK(vp->v_object);
619 vm_object_page_clean(vp->v_object, 0, 0, 0);
620 VM_OBJECT_UNLOCK(vp->v_object);
622 mtx_lock(&np->n_mtx);
623 if (np->n_flag & NMODIFIED) {
624 mtx_unlock(&np->n_mtx);
627 * Under NFSv3 we have dirty buffers to dispose of. We
628 * must flush them to the NFS server. We have the option
629 * of waiting all the way through the commit rpc or just
630 * waiting for the initial write. The default is to only
631 * wait through the initial write so the data is in the
632 * server's cache, which is roughly similar to the state
633 * a standard disk subsystem leaves the file in on close().
635 * We cannot clear the NMODIFIED bit in np->n_flag due to
636 * potential races with other processes, and certainly
637 * cannot clear it if we don't commit.
639 int cm = nfsv3_commit_on_close ? 1 : 0;
640 error = nfs_flush(vp, MNT_WAIT, cm);
641 /* np->n_flag &= ~NMODIFIED; */
643 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
644 mtx_lock(&np->n_mtx);
646 if (np->n_flag & NWRITEERR) {
647 np->n_flag &= ~NWRITEERR;
650 mtx_unlock(&np->n_mtx);
652 if (nfs_directio_enable)
653 KASSERT((np->n_directio_asyncwr == 0),
654 ("nfs_close: dirty unflushed (%d) directio buffers\n",
655 np->n_directio_asyncwr));
656 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
657 mtx_lock(&np->n_mtx);
658 KASSERT((np->n_directio_opens > 0),
659 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
660 np->n_directio_opens--;
661 if (np->n_directio_opens == 0)
662 np->n_flag &= ~NNONCACHE;
663 mtx_unlock(&np->n_mtx);
669 * nfs getattr call from vfs.
672 nfs_getattr(struct vop_getattr_args *ap)
674 struct vnode *vp = ap->a_vp;
675 struct nfsnode *np = VTONFS(vp);
676 struct thread *td = curthread;
677 struct vattr *vap = ap->a_vap;
681 struct mbuf *mreq, *mrep, *md, *mb;
682 int v3 = NFS_ISV3(vp);
685 * Update local times for special files.
687 mtx_lock(&np->n_mtx);
688 if (np->n_flag & (NACC | NUPD))
690 mtx_unlock(&np->n_mtx);
692 * First look in the cache.
694 if (nfs_getattrcache(vp, &vattr) == 0)
696 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
697 nfsstats.accesscache_misses++;
698 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL);
699 if (nfs_getattrcache(vp, &vattr) == 0)
702 nfsstats.rpccnt[NFSPROC_GETATTR]++;
703 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
705 bpos = mtod(mb, caddr_t);
707 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
709 nfsm_loadattr(vp, &vattr);
713 vap->va_type = vattr.va_type;
714 vap->va_mode = vattr.va_mode;
715 vap->va_nlink = vattr.va_nlink;
716 vap->va_uid = vattr.va_uid;
717 vap->va_gid = vattr.va_gid;
718 vap->va_fsid = vattr.va_fsid;
719 vap->va_fileid = vattr.va_fileid;
720 vap->va_size = vattr.va_size;
721 vap->va_blocksize = vattr.va_blocksize;
722 vap->va_atime = vattr.va_atime;
723 vap->va_mtime = vattr.va_mtime;
724 vap->va_ctime = vattr.va_ctime;
725 vap->va_gen = vattr.va_gen;
726 vap->va_flags = vattr.va_flags;
727 vap->va_rdev = vattr.va_rdev;
728 vap->va_bytes = vattr.va_bytes;
729 vap->va_filerev = vattr.va_filerev;
738 nfs_setattr(struct vop_setattr_args *ap)
740 struct vnode *vp = ap->a_vp;
741 struct nfsnode *np = VTONFS(vp);
742 struct vattr *vap = ap->a_vap;
743 struct thread *td = curthread;
752 * Setting of flags is not supported.
754 if (vap->va_flags != VNOVAL)
758 * Disallow write attempts if the filesystem is mounted read-only.
760 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
761 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
762 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
763 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
767 if (vap->va_size != VNOVAL) {
768 switch (vp->v_type) {
775 if (vap->va_mtime.tv_sec == VNOVAL &&
776 vap->va_atime.tv_sec == VNOVAL &&
777 vap->va_mode == (mode_t)VNOVAL &&
778 vap->va_uid == (uid_t)VNOVAL &&
779 vap->va_gid == (gid_t)VNOVAL)
781 vap->va_size = VNOVAL;
785 * Disallow write attempts if the filesystem is
788 if (vp->v_mount->mnt_flag & MNT_RDONLY)
791 * We run vnode_pager_setsize() early (why?),
792 * we must set np->n_size now to avoid vinvalbuf
793 * V_SAVE races that might setsize a lower
796 mtx_lock(&np->n_mtx);
798 mtx_unlock(&np->n_mtx);
799 error = nfs_meta_setsize(vp, ap->a_cred, td,
801 mtx_lock(&np->n_mtx);
802 if (np->n_flag & NMODIFIED) {
804 mtx_unlock(&np->n_mtx);
805 if (vap->va_size == 0)
806 error = nfs_vinvalbuf(vp, 0, td, 1);
808 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
810 vnode_pager_setsize(vp, tsize);
814 mtx_unlock(&np->n_mtx);
816 * np->n_size has already been set to vap->va_size
817 * in nfs_meta_setsize(). We must set it again since
818 * nfs_loadattrcache() could be called through
819 * nfs_meta_setsize() and could modify np->n_size.
821 mtx_lock(&np->n_mtx);
822 np->n_vattr.va_size = np->n_size = vap->va_size;
823 mtx_unlock(&np->n_mtx);
826 mtx_lock(&np->n_mtx);
827 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
828 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
829 mtx_unlock(&np->n_mtx);
830 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
831 (error == EINTR || error == EIO))
834 mtx_unlock(&np->n_mtx);
836 error = nfs_setattrrpc(vp, vap, ap->a_cred);
837 if (error && vap->va_size != VNOVAL) {
838 mtx_lock(&np->n_mtx);
839 np->n_size = np->n_vattr.va_size = tsize;
840 vnode_pager_setsize(vp, tsize);
841 mtx_unlock(&np->n_mtx);
848 * Do an nfs setattr rpc.
851 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
853 struct nfsv2_sattr *sp;
854 struct nfsnode *np = VTONFS(vp);
857 int error = 0, i, wccflag = NFSV3_WCCRATTR;
858 struct mbuf *mreq, *mrep, *md, *mb;
859 int v3 = NFS_ISV3(vp);
861 nfsstats.rpccnt[NFSPROC_SETATTR]++;
862 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
864 bpos = mtod(mb, caddr_t);
867 nfsm_v3attrbuild(vap, TRUE);
868 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
871 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
872 if (vap->va_mode == (mode_t)VNOVAL)
873 sp->sa_mode = nfs_xdrneg1;
875 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
876 if (vap->va_uid == (uid_t)VNOVAL)
877 sp->sa_uid = nfs_xdrneg1;
879 sp->sa_uid = txdr_unsigned(vap->va_uid);
880 if (vap->va_gid == (gid_t)VNOVAL)
881 sp->sa_gid = nfs_xdrneg1;
883 sp->sa_gid = txdr_unsigned(vap->va_gid);
884 sp->sa_size = txdr_unsigned(vap->va_size);
885 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
886 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
888 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
890 mtx_lock(&np->n_mtx);
891 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
892 np->n_accesscache[i].stamp = 0;
893 mtx_unlock(&np->n_mtx);
894 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
895 nfsm_wcc_data(vp, wccflag);
897 nfsm_loadattr(vp, NULL);
904 * nfs lookup call, one step at a time...
905 * First look in cache
906 * If not found, unlock the directory nfsnode and do the rpc
909 nfs_lookup(struct vop_lookup_args *ap)
911 struct componentname *cnp = ap->a_cnp;
912 struct vnode *dvp = ap->a_dvp;
913 struct vnode **vpp = ap->a_vpp;
914 struct mount *mp = dvp->v_mount;
916 struct timespec dmtime;
917 int flags = cnp->cn_flags;
919 struct nfsmount *nmp;
921 struct mbuf *mreq, *mrep, *md, *mb;
924 struct nfsnode *np, *newnp;
925 int error = 0, attrflag, fhsize, ltype;
926 int v3 = NFS_ISV3(dvp);
927 struct thread *td = cnp->cn_thread;
930 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
931 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
933 if (dvp->v_type != VDIR)
937 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
941 error = cache_lookup(dvp, vpp, cnp);
942 if (error > 0 && error != ENOENT)
946 * We only accept a positive hit in the cache if the
947 * change time of the file matches our cached copy.
948 * Otherwise, we discard the cache entry and fallback
949 * to doing a lookup RPC.
951 * To better handle stale file handles and attributes,
952 * clear the attribute cache of this node if it is a
953 * leaf component, part of an open() call, and not
954 * locally modified before fetching the attributes.
955 * This should allow stale file handles to be detected
956 * here where we can fall back to a LOOKUP RPC to
957 * recover rather than having nfs_open() detect the
958 * stale file handle and failing open(2) with ESTALE.
961 newnp = VTONFS(newvp);
962 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
963 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
964 !(newnp->n_flag & NMODIFIED)) {
965 mtx_lock(&newnp->n_mtx);
966 newnp->n_attrstamp = 0;
967 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
968 mtx_unlock(&newnp->n_mtx);
970 if (VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
971 timespeccmp(&vattr.va_ctime, &newnp->n_ctime, ==)) {
972 nfsstats.lookupcache_hits++;
973 if (cnp->cn_nameiop != LOOKUP &&
975 cnp->cn_flags |= SAVENAME;
984 } else if (error == ENOENT) {
985 if (dvp->v_iflag & VI_DOOMED)
988 * We only accept a negative hit in the cache if the
989 * modification time of the parent directory matches
990 * our cached copy. Otherwise, we discard all of the
991 * negative cache entries for this directory. We also
992 * only trust -ve cache entries for less than
993 * nm_negative_namecache_timeout seconds.
995 if ((u_int)(ticks - np->n_dmtime_ticks) <
996 (nmp->nm_negnametimeo * hz) &&
997 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
998 timespeccmp(&vattr.va_mtime, &np->n_dmtime, ==)) {
999 nfsstats.lookupcache_hits++;
1002 cache_purge_negative(dvp);
1003 mtx_lock(&np->n_mtx);
1004 timespecclear(&np->n_dmtime);
1005 mtx_unlock(&np->n_mtx);
1009 * Cache the modification time of the parent directory in case
1010 * the lookup fails and results in adding the first negative
1011 * name cache entry for the directory. Since this is reading
1012 * a single time_t, don't bother with locking. The
1013 * modification time may be a bit stale, but it must be read
1014 * before performing the lookup RPC to prevent a race where
1015 * another lookup updates the timestamp on the directory after
1016 * the lookup RPC has been performed on the server but before
1017 * n_dmtime is set at the end of this function.
1019 dmtime = np->n_vattr.va_mtime;
1022 nfsstats.lookupcache_misses++;
1023 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1024 len = cnp->cn_namelen;
1025 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1026 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1028 bpos = mtod(mb, caddr_t);
1029 nfsm_fhtom(dvp, v3);
1030 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1031 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
1034 nfsm_postop_attr(dvp, attrflag);
1039 nfsm_getfh(fhp, fhsize, v3);
1042 * Handle RENAME case...
1044 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1045 if (NFS_CMPFH(np, fhp, fhsize)) {
1049 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE);
1056 nfsm_postop_attr(newvp, attrflag);
1057 nfsm_postop_attr(dvp, attrflag);
1059 nfsm_loadattr(newvp, NULL);
1062 cnp->cn_flags |= SAVENAME;
1066 if (flags & ISDOTDOT) {
1067 ltype = VOP_ISLOCKED(dvp);
1068 error = vfs_busy(mp, MBF_NOWAIT);
1072 error = vfs_busy(mp, 0);
1073 vn_lock(dvp, ltype | LK_RETRY);
1075 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1085 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1090 vn_lock(dvp, ltype | LK_RETRY);
1091 if (dvp->v_iflag & VI_DOOMED) {
1104 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1108 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1116 * Flush the attribute cache when opening a leaf node
1117 * to ensure that fresh attributes are fetched in
1118 * nfs_open() if we are unable to fetch attributes
1119 * from the LOOKUP reply.
1121 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1122 !(np->n_flag & NMODIFIED)) {
1123 mtx_lock(&np->n_mtx);
1124 np->n_attrstamp = 0;
1125 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1126 mtx_unlock(&np->n_mtx);
1130 nfsm_postop_attr(newvp, attrflag);
1131 nfsm_postop_attr(dvp, attrflag);
1133 nfsm_loadattr(newvp, NULL);
1134 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1135 cnp->cn_flags |= SAVENAME;
1136 if ((cnp->cn_flags & MAKEENTRY) &&
1137 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1138 np->n_ctime = np->n_vattr.va_ctime;
1139 cache_enter(dvp, newvp, cnp);
1145 if (newvp != NULLVP) {
1150 if (error != ENOENT)
1153 /* The requested file was not found. */
1154 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1155 (flags & ISLASTCN)) {
1157 * XXX: UFS does a full VOP_ACCESS(dvp,
1158 * VWRITE) here instead of just checking
1161 if (mp->mnt_flag & MNT_RDONLY)
1163 cnp->cn_flags |= SAVENAME;
1164 return (EJUSTRETURN);
1167 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1169 * Maintain n_dmtime as the modification time
1170 * of the parent directory when the oldest -ve
1171 * name cache entry for this directory was
1172 * added. If a -ve cache entry has already
1173 * been added with a newer modification time
1174 * by a concurrent lookup, then don't bother
1175 * adding a cache entry. The modification
1176 * time of the directory might have changed
1177 * due to the file this lookup failed to find
1178 * being created. In that case a subsequent
1179 * lookup would incorrectly use the entry
1180 * added here instead of doing an extra
1183 mtx_lock(&np->n_mtx);
1184 if (timespeccmp(&np->n_dmtime, &dmtime, <=)) {
1185 if (!timespecisset(&np->n_dmtime)) {
1186 np->n_dmtime = dmtime;
1187 np->n_dmtime_ticks = ticks;
1189 mtx_unlock(&np->n_mtx);
1190 cache_enter(dvp, NULL, cnp);
1192 mtx_unlock(&np->n_mtx);
1202 * Just call nfs_bioread() to do the work.
1205 nfs_read(struct vop_read_args *ap)
1207 struct vnode *vp = ap->a_vp;
1209 switch (vp->v_type) {
1211 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1215 return (EOPNOTSUPP);
1223 nfs_readlink(struct vop_readlink_args *ap)
1225 struct vnode *vp = ap->a_vp;
1227 if (vp->v_type != VLNK)
1229 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1233 * Do a readlink rpc.
1234 * Called by nfs_doio() from below the buffer cache.
1237 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1240 int error = 0, len, attrflag;
1241 struct mbuf *mreq, *mrep, *md, *mb;
1242 int v3 = NFS_ISV3(vp);
1244 nfsstats.rpccnt[NFSPROC_READLINK]++;
1245 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1247 bpos = mtod(mb, caddr_t);
1249 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1251 nfsm_postop_attr(vp, attrflag);
1253 nfsm_strsiz(len, NFS_MAXPATHLEN);
1254 if (len == NFS_MAXPATHLEN) {
1255 struct nfsnode *np = VTONFS(vp);
1256 mtx_lock(&np->n_mtx);
1257 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1259 mtx_unlock(&np->n_mtx);
1261 nfsm_mtouio(uiop, len);
1273 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1277 struct mbuf *mreq, *mrep, *md, *mb;
1278 struct nfsmount *nmp;
1280 int error = 0, len, retlen, tsiz, eof, attrflag;
1281 int v3 = NFS_ISV3(vp);
1287 nmp = VFSTONFS(vp->v_mount);
1288 tsiz = uiop->uio_resid;
1289 mtx_lock(&nmp->nm_mtx);
1290 end = uiop->uio_offset + tsiz;
1291 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) {
1292 mtx_unlock(&nmp->nm_mtx);
1295 rsize = nmp->nm_rsize;
1296 mtx_unlock(&nmp->nm_mtx);
1298 nfsstats.rpccnt[NFSPROC_READ]++;
1299 len = (tsiz > rsize) ? rsize : tsiz;
1300 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1302 bpos = mtod(mb, caddr_t);
1304 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1306 txdr_hyper(uiop->uio_offset, tl);
1307 *(tl + 2) = txdr_unsigned(len);
1309 *tl++ = txdr_unsigned(uiop->uio_offset);
1310 *tl++ = txdr_unsigned(len);
1313 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1315 nfsm_postop_attr(vp, attrflag);
1320 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1321 eof = fxdr_unsigned(int, *(tl + 1));
1323 nfsm_loadattr(vp, NULL);
1325 nfsm_strsiz(retlen, rsize);
1326 nfsm_mtouio(uiop, retlen);
1330 if (eof || retlen == 0) {
1333 } else if (retlen < len) {
1345 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1346 int *iomode, int *must_commit)
1351 struct mbuf *mreq, *mrep, *md, *mb;
1352 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1354 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1355 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1358 KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1"));
1360 tsiz = uiop->uio_resid;
1361 mtx_lock(&nmp->nm_mtx);
1362 end = uiop->uio_offset + tsiz;
1363 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) {
1364 mtx_unlock(&nmp->nm_mtx);
1367 wsize = nmp->nm_wsize;
1368 mtx_unlock(&nmp->nm_mtx);
1370 nfsstats.rpccnt[NFSPROC_WRITE]++;
1371 len = (tsiz > wsize) ? wsize : tsiz;
1372 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1373 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1375 bpos = mtod(mb, caddr_t);
1378 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1379 txdr_hyper(uiop->uio_offset, tl);
1381 *tl++ = txdr_unsigned(len);
1382 *tl++ = txdr_unsigned(*iomode);
1383 *tl = txdr_unsigned(len);
1387 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1388 /* Set both "begin" and "current" to non-garbage. */
1389 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1390 *tl++ = x; /* "begin offset" */
1391 *tl++ = x; /* "current offset" */
1392 x = txdr_unsigned(len);
1393 *tl++ = x; /* total to this offset */
1394 *tl = x; /* size of this write */
1396 nfsm_uiotom(uiop, len);
1397 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1399 wccflag = NFSV3_WCCCHK;
1400 nfsm_wcc_data(vp, wccflag);
1402 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1403 + NFSX_V3WRITEVERF);
1404 rlen = fxdr_unsigned(int, *tl++);
1409 } else if (rlen < len) {
1410 backup = len - rlen;
1411 uiop->uio_iov->iov_base =
1412 (char *)uiop->uio_iov->iov_base -
1414 uiop->uio_iov->iov_len += backup;
1415 uiop->uio_offset -= backup;
1416 uiop->uio_resid += backup;
1419 commit = fxdr_unsigned(int, *tl++);
1422 * Return the lowest committment level
1423 * obtained by any of the RPCs.
1425 if (committed == NFSV3WRITE_FILESYNC)
1427 else if (committed == NFSV3WRITE_DATASYNC &&
1428 commit == NFSV3WRITE_UNSTABLE)
1430 mtx_lock(&nmp->nm_mtx);
1431 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1432 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1434 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1435 } else if (bcmp((caddr_t)tl,
1436 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1438 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1441 mtx_unlock(&nmp->nm_mtx);
1444 nfsm_loadattr(vp, NULL);
1447 mtx_lock(&(VTONFS(vp))->n_mtx);
1448 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1449 mtx_unlock(&(VTONFS(vp))->n_mtx);
1457 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1458 committed = NFSV3WRITE_FILESYNC;
1459 *iomode = committed;
1461 uiop->uio_resid = tsiz;
1467 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1468 * mode set to specify the file type and the size field for rdev.
1471 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1474 struct nfsv2_sattr *sp;
1476 struct vnode *newvp = NULL;
1477 struct nfsnode *np = NULL;
1480 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1481 struct mbuf *mreq, *mrep, *md, *mb;
1483 int v3 = NFS_ISV3(dvp);
1485 if (vap->va_type == VCHR || vap->va_type == VBLK)
1486 rdev = txdr_unsigned(vap->va_rdev);
1487 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1490 return (EOPNOTSUPP);
1492 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1494 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1495 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1496 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1498 bpos = mtod(mb, caddr_t);
1499 nfsm_fhtom(dvp, v3);
1500 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1502 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1503 *tl++ = vtonfsv3_type(vap->va_type);
1504 nfsm_v3attrbuild(vap, FALSE);
1505 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1506 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1507 *tl++ = txdr_unsigned(major(vap->va_rdev));
1508 *tl = txdr_unsigned(minor(vap->va_rdev));
1511 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1512 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1513 sp->sa_uid = nfs_xdrneg1;
1514 sp->sa_gid = nfs_xdrneg1;
1516 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1517 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1519 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1521 nfsm_mtofh(dvp, newvp, v3, gotvp);
1527 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1528 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1534 nfsm_wcc_data(dvp, wccflag);
1541 if (cnp->cn_flags & MAKEENTRY)
1542 cache_enter(dvp, newvp, cnp);
1545 mtx_lock(&(VTONFS(dvp))->n_mtx);
1546 VTONFS(dvp)->n_flag |= NMODIFIED;
1548 VTONFS(dvp)->n_attrstamp = 0;
1549 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1551 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1557 * just call nfs_mknodrpc() to do the work.
1561 nfs_mknod(struct vop_mknod_args *ap)
1563 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1566 static u_long create_verf;
1568 * nfs file create call
1571 nfs_create(struct vop_create_args *ap)
1573 struct vnode *dvp = ap->a_dvp;
1574 struct vattr *vap = ap->a_vap;
1575 struct componentname *cnp = ap->a_cnp;
1576 struct nfsv2_sattr *sp;
1578 struct nfsnode *np = NULL;
1579 struct vnode *newvp = NULL;
1581 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1582 struct mbuf *mreq, *mrep, *md, *mb;
1584 int v3 = NFS_ISV3(dvp);
1587 * Oops, not for me..
1589 if (vap->va_type == VSOCK) {
1590 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1594 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) {
1597 if (vap->va_vaflags & VA_EXCLUSIVE)
1600 nfsstats.rpccnt[NFSPROC_CREATE]++;
1601 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1602 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1604 bpos = mtod(mb, caddr_t);
1605 nfsm_fhtom(dvp, v3);
1606 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1608 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1609 if (fmode & O_EXCL) {
1610 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1611 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1613 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1615 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1616 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1619 *tl++ = create_verf;
1621 IN_IFADDR_RUNLOCK();
1624 *tl = ++create_verf;
1626 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1627 nfsm_v3attrbuild(vap, FALSE);
1630 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1631 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1632 sp->sa_uid = nfs_xdrneg1;
1633 sp->sa_gid = nfs_xdrneg1;
1635 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1636 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1638 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1640 nfsm_mtofh(dvp, newvp, v3, gotvp);
1646 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1647 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1653 nfsm_wcc_data(dvp, wccflag);
1657 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1663 } else if (v3 && (fmode & O_EXCL)) {
1665 * We are normally called with only a partially initialized
1666 * VAP. Since the NFSv3 spec says that server may use the
1667 * file attributes to store the verifier, the spec requires
1668 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1669 * in atime, but we can't really assume that all servers will
1670 * so we ensure that our SETATTR sets both atime and mtime.
1672 if (vap->va_mtime.tv_sec == VNOVAL)
1673 vfs_timestamp(&vap->va_mtime);
1674 if (vap->va_atime.tv_sec == VNOVAL)
1675 vap->va_atime = vap->va_mtime;
1676 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred);
1681 if (cnp->cn_flags & MAKEENTRY)
1682 cache_enter(dvp, newvp, cnp);
1685 mtx_lock(&(VTONFS(dvp))->n_mtx);
1686 VTONFS(dvp)->n_flag |= NMODIFIED;
1688 VTONFS(dvp)->n_attrstamp = 0;
1689 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1691 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1696 * nfs file remove call
1697 * To try and make nfs semantics closer to ufs semantics, a file that has
1698 * other processes using the vnode is renamed instead of removed and then
1699 * removed later on the last close.
1700 * - If v_usecount > 1
1701 * If a rename is not already in the works
1702 * call nfs_sillyrename() to set it up
1707 nfs_remove(struct vop_remove_args *ap)
1709 struct vnode *vp = ap->a_vp;
1710 struct vnode *dvp = ap->a_dvp;
1711 struct componentname *cnp = ap->a_cnp;
1712 struct nfsnode *np = VTONFS(vp);
1716 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1717 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1718 if (vp->v_type == VDIR)
1720 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1721 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1723 * Purge the name cache so that the chance of a lookup for
1724 * the name succeeding while the remove is in progress is
1725 * minimized. Without node locking it can still happen, such
1726 * that an I/O op returns ESTALE, but since you get this if
1727 * another host removes the file..
1731 * throw away biocache buffers, mainly to avoid
1732 * unnecessary delayed writes later.
1734 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1736 if (error != EINTR && error != EIO)
1737 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1738 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1740 * Kludge City: If the first reply to the remove rpc is lost..
1741 * the reply to the retransmitted request will be ENOENT
1742 * since the file was in fact removed
1743 * Therefore, we cheat and return success.
1745 if (error == ENOENT)
1747 } else if (!np->n_sillyrename)
1748 error = nfs_sillyrename(dvp, vp, cnp);
1749 mtx_lock(&np->n_mtx);
1750 np->n_attrstamp = 0;
1751 mtx_unlock(&np->n_mtx);
1752 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1757 * nfs file remove rpc called from nfs_inactive
1760 nfs_removeit(struct sillyrename *sp)
1763 * Make sure that the directory vnode is still valid.
1764 * XXX we should lock sp->s_dvp here.
1766 if (sp->s_dvp->v_type == VBAD)
1768 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1773 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1776 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1777 struct ucred *cred, struct thread *td)
1780 int error = 0, wccflag = NFSV3_WCCRATTR;
1781 struct mbuf *mreq, *mrep, *md, *mb;
1782 int v3 = NFS_ISV3(dvp);
1784 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1785 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1786 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1788 bpos = mtod(mb, caddr_t);
1789 nfsm_fhtom(dvp, v3);
1790 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1791 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1793 nfsm_wcc_data(dvp, wccflag);
1796 mtx_lock(&(VTONFS(dvp))->n_mtx);
1797 VTONFS(dvp)->n_flag |= NMODIFIED;
1799 VTONFS(dvp)->n_attrstamp = 0;
1800 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1802 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1807 * nfs file rename call
1810 nfs_rename(struct vop_rename_args *ap)
1812 struct vnode *fvp = ap->a_fvp;
1813 struct vnode *tvp = ap->a_tvp;
1814 struct vnode *fdvp = ap->a_fdvp;
1815 struct vnode *tdvp = ap->a_tdvp;
1816 struct componentname *tcnp = ap->a_tcnp;
1817 struct componentname *fcnp = ap->a_fcnp;
1820 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1821 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1822 /* Check for cross-device rename */
1823 if ((fvp->v_mount != tdvp->v_mount) ||
1824 (tvp && (fvp->v_mount != tvp->v_mount))) {
1830 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1834 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1838 * We have to flush B_DELWRI data prior to renaming
1839 * the file. If we don't, the delayed-write buffers
1840 * can be flushed out later after the file has gone stale
1841 * under NFSV3. NFSV2 does not have this problem because
1842 * ( as far as I can tell ) it flushes dirty buffers more
1845 * Skip the rename operation if the fsync fails, this can happen
1846 * due to the server's volume being full, when we pushed out data
1847 * that was written back to our cache earlier. Not checking for
1848 * this condition can result in potential (silent) data loss.
1850 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1853 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1858 * If the tvp exists and is in use, sillyrename it before doing the
1859 * rename of the new file over it.
1860 * XXX Can't sillyrename a directory.
1862 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1863 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1868 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1869 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1872 if (fvp->v_type == VDIR) {
1873 if (tvp != NULL && tvp->v_type == VDIR)
1888 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1890 if (error == ENOENT)
1896 * nfs file rename rpc called from nfs_remove() above
1899 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1900 struct sillyrename *sp)
1903 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1904 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1908 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1911 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1912 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1916 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1917 struct mbuf *mreq, *mrep, *md, *mb;
1918 int v3 = NFS_ISV3(fdvp);
1920 nfsstats.rpccnt[NFSPROC_RENAME]++;
1921 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1922 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1923 nfsm_rndup(tnamelen));
1925 bpos = mtod(mb, caddr_t);
1926 nfsm_fhtom(fdvp, v3);
1927 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1928 nfsm_fhtom(tdvp, v3);
1929 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1930 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1932 nfsm_wcc_data(fdvp, fwccflag);
1933 nfsm_wcc_data(tdvp, twccflag);
1937 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1938 VTONFS(fdvp)->n_flag |= NMODIFIED;
1939 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1940 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1941 VTONFS(tdvp)->n_flag |= NMODIFIED;
1942 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1944 VTONFS(fdvp)->n_attrstamp = 0;
1945 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1948 VTONFS(tdvp)->n_attrstamp = 0;
1949 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1955 * nfs hard link create call
1958 nfs_link(struct vop_link_args *ap)
1960 struct vnode *vp = ap->a_vp;
1961 struct vnode *tdvp = ap->a_tdvp;
1962 struct componentname *cnp = ap->a_cnp;
1964 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1965 struct mbuf *mreq, *mrep, *md, *mb;
1968 if (vp->v_mount != tdvp->v_mount) {
1973 * Push all writes to the server, so that the attribute cache
1974 * doesn't get "out of sync" with the server.
1975 * XXX There should be a better way!
1977 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1980 nfsstats.rpccnt[NFSPROC_LINK]++;
1981 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1982 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1984 bpos = mtod(mb, caddr_t);
1986 nfsm_fhtom(tdvp, v3);
1987 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1988 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1990 nfsm_postop_attr(vp, attrflag);
1991 nfsm_wcc_data(tdvp, wccflag);
1995 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1996 VTONFS(tdvp)->n_flag |= NMODIFIED;
1997 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1999 VTONFS(vp)->n_attrstamp = 0;
2000 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2003 VTONFS(tdvp)->n_attrstamp = 0;
2004 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2010 * nfs symbolic link create call
2013 nfs_symlink(struct vop_symlink_args *ap)
2015 struct vnode *dvp = ap->a_dvp;
2016 struct vattr *vap = ap->a_vap;
2017 struct componentname *cnp = ap->a_cnp;
2018 struct nfsv2_sattr *sp;
2020 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2021 struct mbuf *mreq, *mrep, *md, *mb;
2022 struct vnode *newvp = NULL;
2023 int v3 = NFS_ISV3(dvp);
2025 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2026 slen = strlen(ap->a_target);
2027 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2028 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2030 bpos = mtod(mb, caddr_t);
2031 nfsm_fhtom(dvp, v3);
2032 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2034 nfsm_v3attrbuild(vap, FALSE);
2036 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2038 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2039 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2040 sp->sa_uid = nfs_xdrneg1;
2041 sp->sa_gid = nfs_xdrneg1;
2042 sp->sa_size = nfs_xdrneg1;
2043 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2044 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2048 * Issue the NFS request and get the rpc response.
2050 * Only NFSv3 responses returning an error of 0 actually return
2051 * a file handle that can be converted into newvp without having
2052 * to do an extra lookup rpc.
2054 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
2057 nfsm_mtofh(dvp, newvp, v3, gotvp);
2058 nfsm_wcc_data(dvp, wccflag);
2062 * out code jumps -> here, mrep is also freed.
2069 * If we do not have an error and we could not extract the newvp from
2070 * the response due to the request being NFSv2, we have to do a
2071 * lookup in order to obtain a newvp to return.
2073 if (error == 0 && newvp == NULL) {
2074 struct nfsnode *np = NULL;
2076 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2077 cnp->cn_cred, cnp->cn_thread, &np);
2087 mtx_lock(&(VTONFS(dvp))->n_mtx);
2088 VTONFS(dvp)->n_flag |= NMODIFIED;
2089 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2091 VTONFS(dvp)->n_attrstamp = 0;
2092 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2101 nfs_mkdir(struct vop_mkdir_args *ap)
2103 struct vnode *dvp = ap->a_dvp;
2104 struct vattr *vap = ap->a_vap;
2105 struct componentname *cnp = ap->a_cnp;
2106 struct nfsv2_sattr *sp;
2108 struct nfsnode *np = NULL;
2109 struct vnode *newvp = NULL;
2111 int error = 0, wccflag = NFSV3_WCCRATTR;
2113 struct mbuf *mreq, *mrep, *md, *mb;
2115 int v3 = NFS_ISV3(dvp);
2117 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2119 len = cnp->cn_namelen;
2120 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2121 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
2122 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2124 bpos = mtod(mb, caddr_t);
2125 nfsm_fhtom(dvp, v3);
2126 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2128 nfsm_v3attrbuild(vap, FALSE);
2130 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2131 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2132 sp->sa_uid = nfs_xdrneg1;
2133 sp->sa_gid = nfs_xdrneg1;
2134 sp->sa_size = nfs_xdrneg1;
2135 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2136 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2138 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2140 nfsm_mtofh(dvp, newvp, v3, gotvp);
2142 nfsm_wcc_data(dvp, wccflag);
2145 mtx_lock(&(VTONFS(dvp))->n_mtx);
2146 VTONFS(dvp)->n_flag |= NMODIFIED;
2147 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2149 VTONFS(dvp)->n_attrstamp = 0;
2150 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2152 if (error == 0 && newvp == NULL) {
2153 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2154 cnp->cn_thread, &np);
2157 if (newvp->v_type != VDIR)
2170 * nfs remove directory call
2173 nfs_rmdir(struct vop_rmdir_args *ap)
2175 struct vnode *vp = ap->a_vp;
2176 struct vnode *dvp = ap->a_dvp;
2177 struct componentname *cnp = ap->a_cnp;
2179 int error = 0, wccflag = NFSV3_WCCRATTR;
2180 struct mbuf *mreq, *mrep, *md, *mb;
2181 int v3 = NFS_ISV3(dvp);
2185 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2186 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
2187 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2189 bpos = mtod(mb, caddr_t);
2190 nfsm_fhtom(dvp, v3);
2191 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2192 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2194 nfsm_wcc_data(dvp, wccflag);
2197 mtx_lock(&(VTONFS(dvp))->n_mtx);
2198 VTONFS(dvp)->n_flag |= NMODIFIED;
2199 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2201 VTONFS(dvp)->n_attrstamp = 0;
2202 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2207 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2209 if (error == ENOENT)
2218 nfs_readdir(struct vop_readdir_args *ap)
2220 struct vnode *vp = ap->a_vp;
2221 struct nfsnode *np = VTONFS(vp);
2222 struct uio *uio = ap->a_uio;
2223 int tresid, error = 0;
2226 if (vp->v_type != VDIR)
2230 * First, check for hit on the EOF offset cache
2232 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2233 (np->n_flag & NMODIFIED) == 0) {
2234 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2235 mtx_lock(&np->n_mtx);
2236 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2237 mtx_unlock(&np->n_mtx);
2238 nfsstats.direofcache_hits++;
2241 mtx_unlock(&np->n_mtx);
2246 * Call nfs_bioread() to do the real work.
2248 tresid = uio->uio_resid;
2249 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2251 if (!error && uio->uio_resid == tresid) {
2252 nfsstats.direofcache_misses++;
2260 * Called from below the buffer cache by nfs_doio().
2263 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2266 struct dirent *dp = NULL;
2271 struct mbuf *mreq, *mrep, *md, *mb;
2273 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2274 struct nfsnode *dnp = VTONFS(vp);
2276 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2278 int v3 = NFS_ISV3(vp);
2280 KASSERT(uiop->uio_iovcnt == 1 &&
2281 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2282 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2283 ("nfs readdirrpc bad uio"));
2286 * If there is no cookie, assume directory was stale.
2288 nfs_dircookie_lock(dnp);
2289 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2292 nfs_dircookie_unlock(dnp);
2294 nfs_dircookie_unlock(dnp);
2295 return (NFSERR_BAD_COOKIE);
2299 * Loop around doing readdir rpc's of size nm_readdirsize
2300 * truncated to a multiple of DIRBLKSIZ.
2301 * The stopping criteria is EOF or buffer full.
2303 while (more_dirs && bigenough) {
2304 nfsstats.rpccnt[NFSPROC_READDIR]++;
2305 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2308 bpos = mtod(mb, caddr_t);
2311 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2312 *tl++ = cookie.nfsuquad[0];
2313 *tl++ = cookie.nfsuquad[1];
2314 mtx_lock(&dnp->n_mtx);
2315 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2316 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2317 mtx_unlock(&dnp->n_mtx);
2319 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2320 *tl++ = cookie.nfsuquad[0];
2322 *tl = txdr_unsigned(nmp->nm_readdirsize);
2323 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2325 nfsm_postop_attr(vp, attrflag);
2327 tl = nfsm_dissect(u_int32_t *,
2329 mtx_lock(&dnp->n_mtx);
2330 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2331 dnp->n_cookieverf.nfsuquad[1] = *tl;
2332 mtx_unlock(&dnp->n_mtx);
2338 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2339 more_dirs = fxdr_unsigned(int, *tl);
2341 /* loop thru the dir entries, doctoring them to 4bsd form */
2342 while (more_dirs && bigenough) {
2344 tl = nfsm_dissect(u_int32_t *,
2346 fileno = fxdr_hyper(tl);
2347 len = fxdr_unsigned(int, *(tl + 2));
2349 tl = nfsm_dissect(u_int32_t *,
2351 fileno = fxdr_unsigned(u_quad_t, *tl++);
2352 len = fxdr_unsigned(int, *tl);
2354 if (len <= 0 || len > NFS_MAXNAMLEN) {
2359 tlen = nfsm_rndup(len);
2361 tlen += 4; /* To ensure null termination */
2362 left = DIRBLKSIZ - blksiz;
2363 if ((tlen + DIRHDSIZ) > left) {
2364 dp->d_reclen += left;
2365 uiop->uio_iov->iov_base =
2366 (char *)uiop->uio_iov->iov_base + left;
2367 uiop->uio_iov->iov_len -= left;
2368 uiop->uio_offset += left;
2369 uiop->uio_resid -= left;
2372 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2375 dp = (struct dirent *)uiop->uio_iov->iov_base;
2376 dp->d_fileno = (int)fileno;
2378 dp->d_reclen = tlen + DIRHDSIZ;
2379 dp->d_type = DT_UNKNOWN;
2380 blksiz += dp->d_reclen;
2381 if (blksiz == DIRBLKSIZ)
2383 uiop->uio_offset += DIRHDSIZ;
2384 uiop->uio_resid -= DIRHDSIZ;
2385 uiop->uio_iov->iov_base =
2386 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2387 uiop->uio_iov->iov_len -= DIRHDSIZ;
2388 nfsm_mtouio(uiop, len);
2389 cp = uiop->uio_iov->iov_base;
2391 *cp = '\0'; /* null terminate */
2392 uiop->uio_iov->iov_base =
2393 (char *)uiop->uio_iov->iov_base + tlen;
2394 uiop->uio_iov->iov_len -= tlen;
2395 uiop->uio_offset += tlen;
2396 uiop->uio_resid -= tlen;
2398 nfsm_adv(nfsm_rndup(len));
2400 tl = nfsm_dissect(u_int32_t *,
2403 tl = nfsm_dissect(u_int32_t *,
2407 cookie.nfsuquad[0] = *tl++;
2409 cookie.nfsuquad[1] = *tl++;
2414 more_dirs = fxdr_unsigned(int, *tl);
2417 * If at end of rpc data, get the eof boolean
2420 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2421 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2426 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2427 * by increasing d_reclen for the last record.
2430 left = DIRBLKSIZ - blksiz;
2431 dp->d_reclen += left;
2432 uiop->uio_iov->iov_base =
2433 (char *)uiop->uio_iov->iov_base + left;
2434 uiop->uio_iov->iov_len -= left;
2435 uiop->uio_offset += left;
2436 uiop->uio_resid -= left;
2440 * We are now either at the end of the directory or have filled the
2444 dnp->n_direofoffset = uiop->uio_offset;
2446 if (uiop->uio_resid > 0)
2447 nfs_printf("EEK! readdirrpc resid > 0\n");
2448 nfs_dircookie_lock(dnp);
2449 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2451 nfs_dircookie_unlock(dnp);
2458 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2461 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2467 struct vnode *newvp;
2469 caddr_t bpos, dpos, dpossav1, dpossav2;
2470 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2471 struct nameidata nami, *ndp = &nami;
2472 struct componentname *cnp = &ndp->ni_cnd;
2474 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2475 struct nfsnode *dnp = VTONFS(vp), *np;
2478 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2479 int attrflag, fhsize;
2484 KASSERT(uiop->uio_iovcnt == 1 &&
2485 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2486 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2487 ("nfs readdirplusrpc bad uio"));
2492 * If there is no cookie, assume directory was stale.
2494 nfs_dircookie_lock(dnp);
2495 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2498 nfs_dircookie_unlock(dnp);
2500 nfs_dircookie_unlock(dnp);
2501 return (NFSERR_BAD_COOKIE);
2504 * Loop around doing readdir rpc's of size nm_readdirsize
2505 * truncated to a multiple of DIRBLKSIZ.
2506 * The stopping criteria is EOF or buffer full.
2508 while (more_dirs && bigenough) {
2509 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2510 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2511 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2513 bpos = mtod(mb, caddr_t);
2515 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2516 *tl++ = cookie.nfsuquad[0];
2517 *tl++ = cookie.nfsuquad[1];
2518 mtx_lock(&dnp->n_mtx);
2519 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2520 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2521 mtx_unlock(&dnp->n_mtx);
2522 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2523 *tl = txdr_unsigned(nmp->nm_rsize);
2524 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2525 nfsm_postop_attr(vp, attrflag);
2530 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2531 mtx_lock(&dnp->n_mtx);
2532 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2533 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2534 mtx_unlock(&dnp->n_mtx);
2535 more_dirs = fxdr_unsigned(int, *tl);
2537 /* loop thru the dir entries, doctoring them to 4bsd form */
2538 while (more_dirs && bigenough) {
2539 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2540 fileno = fxdr_hyper(tl);
2541 len = fxdr_unsigned(int, *(tl + 2));
2542 if (len <= 0 || len > NFS_MAXNAMLEN) {
2547 tlen = nfsm_rndup(len);
2549 tlen += 4; /* To ensure null termination*/
2550 left = DIRBLKSIZ - blksiz;
2551 if ((tlen + DIRHDSIZ) > left) {
2552 dp->d_reclen += left;
2553 uiop->uio_iov->iov_base =
2554 (char *)uiop->uio_iov->iov_base + left;
2555 uiop->uio_iov->iov_len -= left;
2556 uiop->uio_offset += left;
2557 uiop->uio_resid -= left;
2560 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2563 dp = (struct dirent *)uiop->uio_iov->iov_base;
2564 dp->d_fileno = (int)fileno;
2566 dp->d_reclen = tlen + DIRHDSIZ;
2567 dp->d_type = DT_UNKNOWN;
2568 blksiz += dp->d_reclen;
2569 if (blksiz == DIRBLKSIZ)
2571 uiop->uio_offset += DIRHDSIZ;
2572 uiop->uio_resid -= DIRHDSIZ;
2573 uiop->uio_iov->iov_base =
2574 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2575 uiop->uio_iov->iov_len -= DIRHDSIZ;
2576 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2577 cnp->cn_namelen = len;
2578 nfsm_mtouio(uiop, len);
2579 cp = uiop->uio_iov->iov_base;
2582 uiop->uio_iov->iov_base =
2583 (char *)uiop->uio_iov->iov_base + tlen;
2584 uiop->uio_iov->iov_len -= tlen;
2585 uiop->uio_offset += tlen;
2586 uiop->uio_resid -= tlen;
2588 nfsm_adv(nfsm_rndup(len));
2589 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2591 cookie.nfsuquad[0] = *tl++;
2592 cookie.nfsuquad[1] = *tl++;
2597 * Since the attributes are before the file handle
2598 * (sigh), we must skip over the attributes and then
2599 * come back and get them.
2601 attrflag = fxdr_unsigned(int, *tl);
2605 nfsm_adv(NFSX_V3FATTR);
2606 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2607 doit = fxdr_unsigned(int, *tl);
2609 * Skip loading the attrs for "..". There's a
2610 * race between loading the attrs here and
2611 * lookups that look for the directory currently
2612 * being read (in the parent). We try to acquire
2613 * the exclusive lock on ".." here, owning the
2614 * lock on the directory being read. Lookup will
2615 * hold the lock on ".." and try to acquire the
2616 * lock on the directory being read.
2618 * There are other ways of fixing this, one would
2619 * be to do a trylock on the ".." vnode and skip
2620 * loading the attrs on ".." if it happens to be
2621 * locked by another process. But skipping the
2622 * attrload on ".." seems the easiest option.
2624 if (strcmp(dp->d_name, "..") == 0) {
2627 * We've already skipped over the attrs,
2628 * skip over the filehandle. And store d_type
2631 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2632 i = fxdr_unsigned(int, *tl);
2633 nfsm_adv(nfsm_rndup(i));
2634 dp->d_type = IFTODT(VTTOIF(VDIR));
2637 nfsm_getfh(fhp, fhsize, 1);
2638 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2643 error = nfs_nget(vp->v_mount, fhp,
2644 fhsize, &np, LK_EXCLUSIVE);
2651 if (doit && bigenough) {
2656 nfsm_loadattr(newvp, NULL);
2660 IFTODT(VTTOIF(np->n_vattr.va_type));
2663 * Update n_ctime so subsequent lookup
2664 * doesn't purge entry.
2666 np->n_ctime = np->n_vattr.va_ctime;
2667 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2670 /* Just skip over the file handle */
2671 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2672 i = fxdr_unsigned(int, *tl);
2674 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2675 fhsize = fxdr_unsigned(int, *tl);
2676 nfsm_adv(nfsm_rndup(fhsize));
2679 if (newvp != NULLVP) {
2686 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2687 more_dirs = fxdr_unsigned(int, *tl);
2690 * If at end of rpc data, get the eof boolean
2693 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2694 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2699 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2700 * by increasing d_reclen for the last record.
2703 left = DIRBLKSIZ - blksiz;
2704 dp->d_reclen += left;
2705 uiop->uio_iov->iov_base =
2706 (char *)uiop->uio_iov->iov_base + left;
2707 uiop->uio_iov->iov_len -= left;
2708 uiop->uio_offset += left;
2709 uiop->uio_resid -= left;
2713 * We are now either at the end of the directory or have filled the
2717 dnp->n_direofoffset = uiop->uio_offset;
2719 if (uiop->uio_resid > 0)
2720 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2721 nfs_dircookie_lock(dnp);
2722 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2724 nfs_dircookie_unlock(dnp);
2727 if (newvp != NULLVP) {
2738 * Silly rename. To make the NFS filesystem that is stateless look a little
2739 * more like the "ufs" a remove of an active vnode is translated to a rename
2740 * to a funny looking filename that is removed by nfs_inactive on the
2741 * nfsnode. There is the potential for another process on a different client
2742 * to create the same funny name between the nfs_lookitup() fails and the
2743 * nfs_rename() completes, but...
2746 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2748 struct sillyrename *sp;
2752 unsigned int lticks;
2756 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2757 sp = malloc(sizeof (struct sillyrename),
2758 M_NFSREQ, M_WAITOK);
2759 sp->s_cred = crhold(cnp->cn_cred);
2761 sp->s_removeit = nfs_removeit;
2765 * Fudge together a funny name.
2766 * Changing the format of the funny name to accomodate more
2767 * sillynames per directory.
2768 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2769 * CPU ticks since boot.
2771 pid = cnp->cn_thread->td_proc->p_pid;
2772 lticks = (unsigned int)ticks;
2774 sp->s_namlen = sprintf(sp->s_name,
2775 ".nfs.%08x.%04x4.4", lticks,
2777 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2778 cnp->cn_thread, NULL))
2782 error = nfs_renameit(dvp, cnp, sp);
2785 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2786 cnp->cn_thread, &np);
2787 np->n_sillyrename = sp;
2792 free((caddr_t)sp, M_NFSREQ);
2797 * Look up a file name and optionally either update the file handle or
2798 * allocate an nfsnode, depending on the value of npp.
2799 * npp == NULL --> just do the lookup
2800 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2802 * *npp != NULL --> update the file handle in the vnode
2805 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2806 struct thread *td, struct nfsnode **npp)
2808 struct vnode *newvp = NULL;
2809 struct nfsnode *np, *dnp = VTONFS(dvp);
2811 int error = 0, fhlen, attrflag;
2812 struct mbuf *mreq, *mrep, *md, *mb;
2814 int v3 = NFS_ISV3(dvp);
2816 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2817 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2818 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2820 bpos = mtod(mb, caddr_t);
2821 nfsm_fhtom(dvp, v3);
2822 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2823 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2824 if (npp && !error) {
2825 nfsm_getfh(nfhp, fhlen, v3);
2828 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2829 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2830 np->n_fhp = &np->n_fh;
2831 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2832 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2833 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2834 np->n_fhsize = fhlen;
2836 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2840 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2848 nfsm_postop_attr(newvp, attrflag);
2849 if (!attrflag && *npp == NULL) {
2858 nfsm_loadattr(newvp, NULL);
2862 if (npp && *npp == NULL) {
2877 * Nfs Version 3 commit rpc
2880 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2884 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2886 int error = 0, wccflag = NFSV3_WCCRATTR;
2887 struct mbuf *mreq, *mrep, *md, *mb;
2889 mtx_lock(&nmp->nm_mtx);
2890 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2891 mtx_unlock(&nmp->nm_mtx);
2894 mtx_unlock(&nmp->nm_mtx);
2895 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2896 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2898 bpos = mtod(mb, caddr_t);
2900 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2901 txdr_hyper(offset, tl);
2903 *tl = txdr_unsigned(cnt);
2904 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2905 nfsm_wcc_data(vp, wccflag);
2907 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2908 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2909 NFSX_V3WRITEVERF)) {
2910 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2912 error = NFSERR_STALEWRITEVERF;
2922 * For async requests when nfsiod(s) are running, queue the request by
2923 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2927 nfs_strategy(struct vop_strategy_args *ap)
2929 struct buf *bp = ap->a_bp;
2932 KASSERT(!(bp->b_flags & B_DONE),
2933 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2934 BUF_ASSERT_HELD(bp);
2936 if (bp->b_iocmd == BIO_READ)
2942 * If the op is asynchronous and an i/o daemon is waiting
2943 * queue the request, wake it up and wait for completion
2944 * otherwise just do it ourselves.
2946 if ((bp->b_flags & B_ASYNC) == 0 ||
2947 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2948 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2953 * fsync vnode op. Just call nfs_flush() with commit == 1.
2957 nfs_fsync(struct vop_fsync_args *ap)
2960 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1));
2964 * Flush all the blocks associated with a vnode.
2965 * Walk through the buffer pool and push any dirty pages
2966 * associated with the vnode.
2969 nfs_flush(struct vnode *vp, int waitfor, int commit)
2971 struct nfsnode *np = VTONFS(vp);
2975 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2976 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2978 u_quad_t off, endoff, toff;
2979 struct ucred* wcred = NULL;
2980 struct buf **bvec = NULL;
2982 struct thread *td = curthread;
2983 #ifndef NFS_COMMITBVECSIZ
2984 #define NFS_COMMITBVECSIZ 20
2986 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2987 int bvecsize = 0, bveccount;
2989 if (nmp->nm_flag & NFSMNT_INT)
2990 slpflag = NFS_PCATCH;
2995 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2996 * server, but has not been committed to stable storage on the server
2997 * yet. On the first pass, the byte range is worked out and the commit
2998 * rpc is done. On the second pass, nfs_writebp() is called to do the
3005 if (NFS_ISV3(vp) && commit) {
3006 if (bvec != NULL && bvec != bvec_on_stack)
3009 * Count up how many buffers waiting for a commit.
3013 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3014 if (!BUF_ISLOCKED(bp) &&
3015 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
3016 == (B_DELWRI | B_NEEDCOMMIT))
3020 * Allocate space to remember the list of bufs to commit. It is
3021 * important to use M_NOWAIT here to avoid a race with nfs_write.
3022 * If we can't get memory (for whatever reason), we will end up
3023 * committing the buffers one-by-one in the loop below.
3025 if (bveccount > NFS_COMMITBVECSIZ) {
3027 * Release the vnode interlock to avoid a lock
3031 bvec = (struct buf **)
3032 malloc(bveccount * sizeof(struct buf *),
3036 bvec = bvec_on_stack;
3037 bvecsize = NFS_COMMITBVECSIZ;
3039 bvecsize = bveccount;
3041 bvec = bvec_on_stack;
3042 bvecsize = NFS_COMMITBVECSIZ;
3044 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3045 if (bvecpos >= bvecsize)
3047 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3048 nbp = TAILQ_NEXT(bp, b_bobufs);
3051 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3052 (B_DELWRI | B_NEEDCOMMIT)) {
3054 nbp = TAILQ_NEXT(bp, b_bobufs);
3060 * Work out if all buffers are using the same cred
3061 * so we can deal with them all with one commit.
3063 * NOTE: we are not clearing B_DONE here, so we have
3064 * to do it later on in this routine if we intend to
3065 * initiate I/O on the bp.
3067 * Note: to avoid loopback deadlocks, we do not
3068 * assign b_runningbufspace.
3071 wcred = bp->b_wcred;
3072 else if (wcred != bp->b_wcred)
3074 vfs_busy_pages(bp, 1);
3078 * bp is protected by being locked, but nbp is not
3079 * and vfs_busy_pages() may sleep. We have to
3082 nbp = TAILQ_NEXT(bp, b_bobufs);
3085 * A list of these buffers is kept so that the
3086 * second loop knows which buffers have actually
3087 * been committed. This is necessary, since there
3088 * may be a race between the commit rpc and new
3089 * uncommitted writes on the file.
3091 bvec[bvecpos++] = bp;
3092 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3096 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3104 * Commit data on the server, as required.
3105 * If all bufs are using the same wcred, then use that with
3106 * one call for all of them, otherwise commit each one
3109 if (wcred != NOCRED)
3110 retv = nfs_commit(vp, off, (int)(endoff - off),
3114 for (i = 0; i < bvecpos; i++) {
3117 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3119 size = (u_quad_t)(bp->b_dirtyend
3121 retv = nfs_commit(vp, off, (int)size,
3127 if (retv == NFSERR_STALEWRITEVERF)
3128 nfs_clearcommit(vp->v_mount);
3131 * Now, either mark the blocks I/O done or mark the
3132 * blocks dirty, depending on whether the commit
3135 for (i = 0; i < bvecpos; i++) {
3137 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3140 * Error, leave B_DELWRI intact
3142 vfs_unbusy_pages(bp);
3146 * Success, remove B_DELWRI ( bundirty() ).
3148 * b_dirtyoff/b_dirtyend seem to be NFS
3149 * specific. We should probably move that
3150 * into bundirty(). XXX
3153 bp->b_flags |= B_ASYNC;
3155 bp->b_flags &= ~B_DONE;
3156 bp->b_ioflags &= ~BIO_ERROR;
3157 bp->b_dirtyoff = bp->b_dirtyend = 0;
3164 * Start/do any write(s) that are required.
3168 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3169 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3170 if (waitfor != MNT_WAIT || passone)
3173 error = BUF_TIMELOCK(bp,
3174 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3175 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
3180 if (error == ENOLCK) {
3184 if (nfs_sigintr(nmp, td)) {
3188 if (slpflag & PCATCH) {
3194 if ((bp->b_flags & B_DELWRI) == 0)
3195 panic("nfs_fsync: not dirty");
3196 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3202 if (passone || !commit)
3203 bp->b_flags |= B_ASYNC;
3205 bp->b_flags |= B_ASYNC;
3207 if (nfs_sigintr(nmp, td)) {
3218 if (waitfor == MNT_WAIT) {
3219 while (bo->bo_numoutput) {
3220 error = bufobj_wwait(bo, slpflag, slptimeo);
3223 error = nfs_sigintr(nmp, td);
3226 if (slpflag & PCATCH) {
3233 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3238 * Wait for all the async IO requests to drain
3241 mtx_lock(&np->n_mtx);
3242 while (np->n_directio_asyncwr > 0) {
3243 np->n_flag |= NFSYNCWAIT;
3244 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3245 &np->n_mtx, slpflag | (PRIBIO + 1),
3248 if (nfs_sigintr(nmp, td)) {
3249 mtx_unlock(&np->n_mtx);
3255 mtx_unlock(&np->n_mtx);
3258 mtx_lock(&np->n_mtx);
3259 if (np->n_flag & NWRITEERR) {
3260 error = np->n_error;
3261 np->n_flag &= ~NWRITEERR;
3263 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3264 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3265 np->n_flag &= ~NMODIFIED;
3266 mtx_unlock(&np->n_mtx);
3268 if (bvec != NULL && bvec != bvec_on_stack)
3274 * NFS advisory byte-level locks.
3277 nfs_advlock(struct vop_advlock_args *ap)
3279 struct vnode *vp = ap->a_vp;
3283 error = vn_lock(vp, LK_SHARED);
3286 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3287 size = VTONFS(vp)->n_size;
3289 error = lf_advlock(ap, &(vp->v_lockf), size);
3292 error = nfs_advlock_p(ap);
3301 * NFS advisory byte-level locks.
3304 nfs_advlockasync(struct vop_advlockasync_args *ap)
3306 struct vnode *vp = ap->a_vp;
3310 error = vn_lock(vp, LK_SHARED);
3313 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3314 size = VTONFS(vp)->n_size;
3316 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3325 * Print out the contents of an nfsnode.
3328 nfs_print(struct vop_print_args *ap)
3330 struct vnode *vp = ap->a_vp;
3331 struct nfsnode *np = VTONFS(vp);
3333 nfs_printf("\tfileid %ld fsid 0x%x",
3334 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3335 if (vp->v_type == VFIFO)
3342 * This is the "real" nfs::bwrite(struct buf*).
3343 * We set B_CACHE if this is a VMIO buffer.
3346 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3349 int oldflags = bp->b_flags;
3355 BUF_ASSERT_HELD(bp);
3357 if (bp->b_flags & B_INVAL) {
3362 bp->b_flags |= B_CACHE;
3365 * Undirty the bp. We will redirty it later if the I/O fails.
3370 bp->b_flags &= ~B_DONE;
3371 bp->b_ioflags &= ~BIO_ERROR;
3372 bp->b_iocmd = BIO_WRITE;
3374 bufobj_wref(bp->b_bufobj);
3375 curthread->td_ru.ru_oublock++;
3379 * Note: to avoid loopback deadlocks, we do not
3380 * assign b_runningbufspace.
3382 vfs_busy_pages(bp, 1);
3385 bp->b_iooffset = dbtob(bp->b_blkno);
3388 if( (oldflags & B_ASYNC) == 0) {
3389 int rtval = bufwait(bp);
3391 if (oldflags & B_DELWRI) {
3404 * nfs special file access vnode op.
3405 * Essentially just get vattr and then imitate iaccess() since the device is
3406 * local to the client.
3409 nfsspec_access(struct vop_access_args *ap)
3412 struct ucred *cred = ap->a_cred;
3413 struct vnode *vp = ap->a_vp;
3414 accmode_t accmode = ap->a_accmode;
3419 * Disallow write attempts on filesystems mounted read-only;
3420 * unless the file is a socket, fifo, or a block or character
3421 * device resident on the filesystem.
3423 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3424 switch (vp->v_type) {
3434 error = VOP_GETATTR(vp, vap, cred);
3437 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3438 accmode, cred, NULL);
3444 * Read wrapper for fifos.
3447 nfsfifo_read(struct vop_read_args *ap)
3449 struct nfsnode *np = VTONFS(ap->a_vp);
3455 mtx_lock(&np->n_mtx);
3457 getnanotime(&np->n_atim);
3458 mtx_unlock(&np->n_mtx);
3459 error = fifo_specops.vop_read(ap);
3464 * Write wrapper for fifos.
3467 nfsfifo_write(struct vop_write_args *ap)
3469 struct nfsnode *np = VTONFS(ap->a_vp);
3474 mtx_lock(&np->n_mtx);
3476 getnanotime(&np->n_mtim);
3477 mtx_unlock(&np->n_mtx);
3478 return(fifo_specops.vop_write(ap));
3482 * Close wrapper for fifos.
3484 * Update the times on the nfsnode then do fifo close.
3487 nfsfifo_close(struct vop_close_args *ap)
3489 struct vnode *vp = ap->a_vp;
3490 struct nfsnode *np = VTONFS(vp);
3494 mtx_lock(&np->n_mtx);
3495 if (np->n_flag & (NACC | NUPD)) {
3497 if (np->n_flag & NACC)
3499 if (np->n_flag & NUPD)
3502 if (vrefcnt(vp) == 1 &&
3503 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3505 if (np->n_flag & NACC)
3506 vattr.va_atime = np->n_atim;
3507 if (np->n_flag & NUPD)
3508 vattr.va_mtime = np->n_mtim;
3509 mtx_unlock(&np->n_mtx);
3510 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3514 mtx_unlock(&np->n_mtx);
3516 return (fifo_specops.vop_close(ap));
3520 * Just call nfs_writebp() with the force argument set to 1.
3522 * NOTE: B_DONE may or may not be set in a_bp on call.
3525 nfs_bwrite(struct buf *bp)
3528 return (nfs_writebp(bp, 1, curthread));
3531 struct buf_ops buf_ops_nfs = {
3532 .bop_name = "buf_ops_nfs",
3533 .bop_write = nfs_bwrite,
3534 .bop_strategy = bufstrategy,
3535 .bop_sync = bufsync,
3536 .bop_bdflush = bufbdflush,