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_object.h>
68 #include <vm/vm_extern.h>
69 #include <vm/vm_object.h>
71 #include <fs/fifofs/fifo.h>
73 #include <nfs/nfsproto.h>
74 #include <nfsclient/nfs.h>
75 #include <nfsclient/nfsnode.h>
76 #include <nfsclient/nfsmount.h>
77 #include <nfsclient/nfs_kdtrace.h>
78 #include <nfsclient/nfs_lock.h>
79 #include <nfs/xdr_subs.h>
80 #include <nfsclient/nfsm_subs.h>
83 #include <netinet/in.h>
84 #include <netinet/in_var.h>
86 #include <machine/stdarg.h>
89 #include <sys/dtrace_bsd.h>
91 dtrace_nfsclient_accesscache_flush_probe_func_t
92 dtrace_nfsclient_accesscache_flush_done_probe;
93 uint32_t nfsclient_accesscache_flush_done_id;
95 dtrace_nfsclient_accesscache_get_probe_func_t
96 dtrace_nfsclient_accesscache_get_hit_probe,
97 dtrace_nfsclient_accesscache_get_miss_probe;
98 uint32_t nfsclient_accesscache_get_hit_id;
99 uint32_t nfsclient_accesscache_get_miss_id;
101 dtrace_nfsclient_accesscache_load_probe_func_t
102 dtrace_nfsclient_accesscache_load_done_probe;
103 uint32_t nfsclient_accesscache_load_done_id;
104 #endif /* !KDTRACE_HOOKS */
111 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
112 * calls are not in getblk() and brelse() so that they would not be necessary
116 #define vfs_busy_pages(bp, f)
119 static vop_read_t nfsfifo_read;
120 static vop_write_t nfsfifo_write;
121 static vop_close_t nfsfifo_close;
122 static int nfs_flush(struct vnode *, int, int);
123 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *);
124 static vop_lookup_t nfs_lookup;
125 static vop_create_t nfs_create;
126 static vop_mknod_t nfs_mknod;
127 static vop_open_t nfs_open;
128 static vop_close_t nfs_close;
129 static vop_access_t nfs_access;
130 static vop_getattr_t nfs_getattr;
131 static vop_setattr_t nfs_setattr;
132 static vop_read_t nfs_read;
133 static vop_fsync_t nfs_fsync;
134 static vop_remove_t nfs_remove;
135 static vop_link_t nfs_link;
136 static vop_rename_t nfs_rename;
137 static vop_mkdir_t nfs_mkdir;
138 static vop_rmdir_t nfs_rmdir;
139 static vop_symlink_t nfs_symlink;
140 static vop_readdir_t nfs_readdir;
141 static vop_strategy_t nfs_strategy;
142 static int nfs_lookitup(struct vnode *, const char *, int,
143 struct ucred *, struct thread *, struct nfsnode **);
144 static int nfs_sillyrename(struct vnode *, struct vnode *,
145 struct componentname *);
146 static vop_access_t nfsspec_access;
147 static vop_readlink_t nfs_readlink;
148 static vop_print_t nfs_print;
149 static vop_advlock_t nfs_advlock;
150 static vop_advlockasync_t nfs_advlockasync;
153 * Global vfs data structures for nfs
155 struct vop_vector nfs_vnodeops = {
156 .vop_default = &default_vnodeops,
157 .vop_access = nfs_access,
158 .vop_advlock = nfs_advlock,
159 .vop_advlockasync = nfs_advlockasync,
160 .vop_close = nfs_close,
161 .vop_create = nfs_create,
162 .vop_fsync = nfs_fsync,
163 .vop_getattr = nfs_getattr,
164 .vop_getpages = nfs_getpages,
165 .vop_putpages = nfs_putpages,
166 .vop_inactive = nfs_inactive,
167 .vop_link = nfs_link,
168 .vop_lookup = nfs_lookup,
169 .vop_mkdir = nfs_mkdir,
170 .vop_mknod = nfs_mknod,
171 .vop_open = nfs_open,
172 .vop_print = nfs_print,
173 .vop_read = nfs_read,
174 .vop_readdir = nfs_readdir,
175 .vop_readlink = nfs_readlink,
176 .vop_reclaim = nfs_reclaim,
177 .vop_remove = nfs_remove,
178 .vop_rename = nfs_rename,
179 .vop_rmdir = nfs_rmdir,
180 .vop_setattr = nfs_setattr,
181 .vop_strategy = nfs_strategy,
182 .vop_symlink = nfs_symlink,
183 .vop_write = nfs_write,
186 struct vop_vector nfs_fifoops = {
187 .vop_default = &fifo_specops,
188 .vop_access = nfsspec_access,
189 .vop_close = nfsfifo_close,
190 .vop_fsync = nfs_fsync,
191 .vop_getattr = nfs_getattr,
192 .vop_inactive = nfs_inactive,
193 .vop_print = nfs_print,
194 .vop_read = nfsfifo_read,
195 .vop_reclaim = nfs_reclaim,
196 .vop_setattr = nfs_setattr,
197 .vop_write = nfsfifo_write,
200 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
201 struct componentname *cnp, struct vattr *vap);
202 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
203 struct ucred *cred, struct thread *td);
204 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
205 int fnamelen, struct vnode *tdvp,
206 const char *tnameptr, int tnamelen,
207 struct ucred *cred, struct thread *td);
208 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
209 struct sillyrename *sp);
214 struct mtx nfs_iod_mtx;
215 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
216 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
217 int nfs_numasync = 0;
218 vop_advlock_t *nfs_advlock_p = nfs_dolock;
219 vop_reclaim_t *nfs_reclaim_p = NULL;
220 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
222 SYSCTL_DECL(_vfs_nfs);
224 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
225 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
226 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
228 static int nfs_prime_access_cache = 0;
229 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
230 &nfs_prime_access_cache, 0,
231 "Prime NFS ACCESS cache when fetching attributes");
233 static int nfsv3_commit_on_close = 0;
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
235 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
237 static int nfs_clean_pages_on_close = 1;
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
239 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
241 int nfs_directio_enable = 0;
242 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
243 &nfs_directio_enable, 0, "Enable NFS directio");
246 * This sysctl allows other processes to mmap a file that has been opened
247 * O_DIRECT by a process. In general, having processes mmap the file while
248 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
249 * this by default to prevent DoS attacks - to prevent a malicious user from
250 * opening up files O_DIRECT preventing other users from mmap'ing these
251 * files. "Protected" environments where stricter consistency guarantees are
252 * required can disable this knob. The process that opened the file O_DIRECT
253 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
256 int nfs_directio_allow_mmap = 1;
257 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
258 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
261 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
262 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
264 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
265 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
268 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
269 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
270 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
274 * The list of locks after the description of the lock is the ordering
275 * of other locks acquired with the lock held.
276 * np->n_mtx : Protects the fields in the nfsnode.
278 VI_MTX (acquired indirectly)
279 * nmp->nm_mtx : Protects the fields in the nfsmount.
281 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
282 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
285 * rep->r_mtx : Protects the fields in an nfsreq.
289 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
290 struct ucred *cred, uint32_t *retmode)
294 int error = 0, attrflag, i, lrupos;
296 struct mbuf *mreq, *mrep, *md, *mb;
299 struct nfsnode *np = VTONFS(vp);
301 nfsstats.rpccnt[NFSPROC_ACCESS]++;
302 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
304 bpos = mtod(mb, caddr_t);
306 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
307 *tl = txdr_unsigned(wmode);
308 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
309 nfsm_postop_attr(vp, attrflag);
312 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
313 rmode = fxdr_unsigned(u_int32_t, *tl);
314 mtx_lock(&np->n_mtx);
315 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
316 if (np->n_accesscache[i].uid == cred->cr_uid) {
317 np->n_accesscache[i].mode = rmode;
318 np->n_accesscache[i].stamp = time_second;
321 if (i > 0 && np->n_accesscache[i].stamp <
322 np->n_accesscache[lrupos].stamp)
325 if (i == NFS_ACCESSCACHESIZE) {
326 np->n_accesscache[lrupos].uid = cred->cr_uid;
327 np->n_accesscache[lrupos].mode = rmode;
328 np->n_accesscache[lrupos].stamp = time_second;
330 mtx_unlock(&np->n_mtx);
333 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
339 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
347 * nfs access vnode op.
348 * For nfs version 2, just return ok. File accesses may fail later.
349 * For nfs version 3, use the access rpc to check accessibility. If file modes
350 * are changed on the server, accesses might still fail later.
353 nfs_access(struct vop_access_args *ap)
355 struct vnode *vp = ap->a_vp;
356 int error = 0, i, gotahit;
357 u_int32_t mode, rmode, wmode;
358 int v3 = NFS_ISV3(vp);
359 struct nfsnode *np = VTONFS(vp);
362 * Disallow write attempts on filesystems mounted read-only;
363 * unless the file is a socket, fifo, or a block or character
364 * device resident on the filesystem.
366 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
367 switch (vp->v_type) {
377 * For nfs v3, check to see if we have done this recently, and if
378 * so return our cached result instead of making an ACCESS call.
379 * If not, do an access rpc, otherwise you are stuck emulating
380 * ufs_access() locally using the vattr. This may not be correct,
381 * since the server may apply other access criteria such as
382 * client uid-->server uid mapping that we do not know about.
385 if (ap->a_accmode & VREAD)
386 mode = NFSV3ACCESS_READ;
389 if (vp->v_type != VDIR) {
390 if (ap->a_accmode & VWRITE)
391 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
392 if (ap->a_accmode & VEXEC)
393 mode |= NFSV3ACCESS_EXECUTE;
395 if (ap->a_accmode & VWRITE)
396 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
398 if (ap->a_accmode & VEXEC)
399 mode |= NFSV3ACCESS_LOOKUP;
401 /* XXX safety belt, only make blanket request if caching */
402 if (nfsaccess_cache_timeout > 0) {
403 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
404 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
405 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
411 * Does our cached result allow us to give a definite yes to
415 mtx_lock(&np->n_mtx);
416 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
417 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
418 if (time_second < (np->n_accesscache[i].stamp +
419 nfsaccess_cache_timeout) &&
420 (np->n_accesscache[i].mode & mode) == mode) {
421 nfsstats.accesscache_hits++;
427 mtx_unlock(&np->n_mtx);
430 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
431 ap->a_cred->cr_uid, mode);
433 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
434 ap->a_cred->cr_uid, mode);
438 * Either a no, or a don't know. Go to the wire.
440 nfsstats.accesscache_misses++;
441 error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred,
444 if ((rmode & mode) != mode)
450 if ((error = nfsspec_access(ap)) != 0) {
454 * Attempt to prevent a mapped root from accessing a file
455 * which it shouldn't. We try to read a byte from the file
456 * if the user is root and the file is not zero length.
457 * After calling nfsspec_access, we should have the correct
460 mtx_lock(&np->n_mtx);
461 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
462 && VTONFS(vp)->n_size > 0) {
467 mtx_unlock(&np->n_mtx);
470 auio.uio_iov = &aiov;
474 auio.uio_segflg = UIO_SYSSPACE;
475 auio.uio_rw = UIO_READ;
476 auio.uio_td = ap->a_td;
478 if (vp->v_type == VREG)
479 error = nfs_readrpc(vp, &auio, ap->a_cred);
480 else if (vp->v_type == VDIR) {
482 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
484 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
485 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
487 } else if (vp->v_type == VLNK)
488 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
492 mtx_unlock(&np->n_mtx);
497 int nfs_otw_getattr_avoid = 0;
501 * Check to see if the type is ok
502 * and that deletion is not in progress.
503 * For paged in text files, you will need to flush the page cache
504 * if consistency is lost.
508 nfs_open(struct vop_open_args *ap)
510 struct vnode *vp = ap->a_vp;
511 struct nfsnode *np = VTONFS(vp);
514 int fmode = ap->a_mode;
516 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
520 * Get a valid lease. If cached data is stale, flush it.
522 mtx_lock(&np->n_mtx);
523 if (np->n_flag & NMODIFIED) {
524 mtx_unlock(&np->n_mtx);
525 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
526 if (error == EINTR || error == EIO)
529 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
530 if (vp->v_type == VDIR)
531 np->n_direofoffset = 0;
532 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
535 mtx_lock(&np->n_mtx);
536 np->n_mtime = vattr.va_mtime;
537 mtx_unlock(&np->n_mtx);
539 struct thread *td = curthread;
541 if (np->n_ac_ts_syscalls != td->td_syscalls ||
542 np->n_ac_ts_tid != td->td_tid ||
543 td->td_proc == NULL ||
544 np->n_ac_ts_pid != td->td_proc->p_pid) {
546 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
548 mtx_unlock(&np->n_mtx);
549 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
552 mtx_lock(&np->n_mtx);
553 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
554 if (vp->v_type == VDIR)
555 np->n_direofoffset = 0;
556 mtx_unlock(&np->n_mtx);
557 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
558 if (error == EINTR || error == EIO) {
561 mtx_lock(&np->n_mtx);
562 np->n_mtime = vattr.va_mtime;
564 mtx_unlock(&np->n_mtx);
567 * If the object has >= 1 O_DIRECT active opens, we disable caching.
569 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
570 if (np->n_directio_opens == 0) {
571 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
574 mtx_lock(&np->n_mtx);
575 np->n_flag |= NNONCACHE;
576 mtx_unlock(&np->n_mtx);
578 np->n_directio_opens++;
580 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
586 * What an NFS client should do upon close after writing is a debatable issue.
587 * Most NFS clients push delayed writes to the server upon close, basically for
589 * 1 - So that any write errors may be reported back to the client process
590 * doing the close system call. By far the two most likely errors are
591 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
592 * 2 - To put a worst case upper bound on cache inconsistency between
593 * multiple clients for the file.
594 * There is also a consistency problem for Version 2 of the protocol w.r.t.
595 * not being able to tell if other clients are writing a file concurrently,
596 * since there is no way of knowing if the changed modify time in the reply
597 * is only due to the write for this client.
598 * (NFS Version 3 provides weak cache consistency data in the reply that
599 * should be sufficient to detect and handle this case.)
601 * The current code does the following:
602 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
603 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
604 * or commit them (this satisfies 1 and 2 except for the
605 * case where the server crashes after this close but
606 * before the commit RPC, which is felt to be "good
607 * enough". Changing the last argument to nfs_flush() to
608 * a 1 would force a commit operation, if it is felt a
609 * commit is necessary now.
613 nfs_close(struct vop_close_args *ap)
615 struct vnode *vp = ap->a_vp;
616 struct nfsnode *np = VTONFS(vp);
618 int fmode = ap->a_fflag;
620 if (vp->v_type == VREG) {
622 * Examine and clean dirty pages, regardless of NMODIFIED.
623 * This closes a major hole in close-to-open consistency.
624 * We want to push out all dirty pages (and buffers) on
625 * close, regardless of whether they were dirtied by
626 * mmap'ed writes or via write().
628 if (nfs_clean_pages_on_close && vp->v_object) {
629 VM_OBJECT_LOCK(vp->v_object);
630 vm_object_page_clean(vp->v_object, 0, 0, 0);
631 VM_OBJECT_UNLOCK(vp->v_object);
633 mtx_lock(&np->n_mtx);
634 if (np->n_flag & NMODIFIED) {
635 mtx_unlock(&np->n_mtx);
638 * Under NFSv3 we have dirty buffers to dispose of. We
639 * must flush them to the NFS server. We have the option
640 * of waiting all the way through the commit rpc or just
641 * waiting for the initial write. The default is to only
642 * wait through the initial write so the data is in the
643 * server's cache, which is roughly similar to the state
644 * a standard disk subsystem leaves the file in on close().
646 * We cannot clear the NMODIFIED bit in np->n_flag due to
647 * potential races with other processes, and certainly
648 * cannot clear it if we don't commit.
650 int cm = nfsv3_commit_on_close ? 1 : 0;
651 error = nfs_flush(vp, MNT_WAIT, cm);
652 /* np->n_flag &= ~NMODIFIED; */
654 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
655 mtx_lock(&np->n_mtx);
657 if (np->n_flag & NWRITEERR) {
658 np->n_flag &= ~NWRITEERR;
661 mtx_unlock(&np->n_mtx);
663 if (nfs_directio_enable)
664 KASSERT((np->n_directio_asyncwr == 0),
665 ("nfs_close: dirty unflushed (%d) directio buffers\n",
666 np->n_directio_asyncwr));
667 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
668 mtx_lock(&np->n_mtx);
669 KASSERT((np->n_directio_opens > 0),
670 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
671 np->n_directio_opens--;
672 if (np->n_directio_opens == 0)
673 np->n_flag &= ~NNONCACHE;
674 mtx_unlock(&np->n_mtx);
680 * nfs getattr call from vfs.
683 nfs_getattr(struct vop_getattr_args *ap)
685 struct vnode *vp = ap->a_vp;
686 struct nfsnode *np = VTONFS(vp);
687 struct thread *td = curthread;
688 struct vattr *vap = ap->a_vap;
692 struct mbuf *mreq, *mrep, *md, *mb;
693 int v3 = NFS_ISV3(vp);
696 * Update local times for special files.
698 mtx_lock(&np->n_mtx);
699 if (np->n_flag & (NACC | NUPD))
701 mtx_unlock(&np->n_mtx);
703 * First look in the cache.
705 if (nfs_getattrcache(vp, &vattr) == 0)
707 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
708 nfsstats.accesscache_misses++;
709 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL);
710 if (nfs_getattrcache(vp, &vattr) == 0)
713 nfsstats.rpccnt[NFSPROC_GETATTR]++;
714 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
716 bpos = mtod(mb, caddr_t);
718 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
720 nfsm_loadattr(vp, &vattr);
724 vap->va_type = vattr.va_type;
725 vap->va_mode = vattr.va_mode;
726 vap->va_nlink = vattr.va_nlink;
727 vap->va_uid = vattr.va_uid;
728 vap->va_gid = vattr.va_gid;
729 vap->va_fsid = vattr.va_fsid;
730 vap->va_fileid = vattr.va_fileid;
731 vap->va_size = vattr.va_size;
732 vap->va_blocksize = vattr.va_blocksize;
733 vap->va_atime = vattr.va_atime;
734 vap->va_mtime = vattr.va_mtime;
735 vap->va_ctime = vattr.va_ctime;
736 vap->va_gen = vattr.va_gen;
737 vap->va_flags = vattr.va_flags;
738 vap->va_rdev = vattr.va_rdev;
739 vap->va_bytes = vattr.va_bytes;
740 vap->va_filerev = vattr.va_filerev;
749 nfs_setattr(struct vop_setattr_args *ap)
751 struct vnode *vp = ap->a_vp;
752 struct nfsnode *np = VTONFS(vp);
753 struct vattr *vap = ap->a_vap;
754 struct thread *td = curthread;
763 * Setting of flags is not supported.
765 if (vap->va_flags != VNOVAL)
769 * Disallow write attempts if the filesystem is mounted read-only.
771 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
772 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
773 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
774 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
778 if (vap->va_size != VNOVAL) {
779 switch (vp->v_type) {
786 if (vap->va_mtime.tv_sec == VNOVAL &&
787 vap->va_atime.tv_sec == VNOVAL &&
788 vap->va_mode == (mode_t)VNOVAL &&
789 vap->va_uid == (uid_t)VNOVAL &&
790 vap->va_gid == (gid_t)VNOVAL)
792 vap->va_size = VNOVAL;
796 * Disallow write attempts if the filesystem is
799 if (vp->v_mount->mnt_flag & MNT_RDONLY)
802 * We run vnode_pager_setsize() early (why?),
803 * we must set np->n_size now to avoid vinvalbuf
804 * V_SAVE races that might setsize a lower
807 mtx_lock(&np->n_mtx);
809 mtx_unlock(&np->n_mtx);
810 error = nfs_meta_setsize(vp, ap->a_cred, td,
812 mtx_lock(&np->n_mtx);
813 if (np->n_flag & NMODIFIED) {
815 mtx_unlock(&np->n_mtx);
816 if (vap->va_size == 0)
817 error = nfs_vinvalbuf(vp, 0, td, 1);
819 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
821 vnode_pager_setsize(vp, tsize);
825 mtx_unlock(&np->n_mtx);
827 * np->n_size has already been set to vap->va_size
828 * in nfs_meta_setsize(). We must set it again since
829 * nfs_loadattrcache() could be called through
830 * nfs_meta_setsize() and could modify np->n_size.
832 mtx_lock(&np->n_mtx);
833 np->n_vattr.va_size = np->n_size = vap->va_size;
834 mtx_unlock(&np->n_mtx);
837 mtx_lock(&np->n_mtx);
838 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
839 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
840 mtx_unlock(&np->n_mtx);
841 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
842 (error == EINTR || error == EIO))
845 mtx_unlock(&np->n_mtx);
847 error = nfs_setattrrpc(vp, vap, ap->a_cred);
848 if (error && vap->va_size != VNOVAL) {
849 mtx_lock(&np->n_mtx);
850 np->n_size = np->n_vattr.va_size = tsize;
851 vnode_pager_setsize(vp, tsize);
852 mtx_unlock(&np->n_mtx);
859 * Do an nfs setattr rpc.
862 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
864 struct nfsv2_sattr *sp;
865 struct nfsnode *np = VTONFS(vp);
868 int error = 0, i, wccflag = NFSV3_WCCRATTR;
869 struct mbuf *mreq, *mrep, *md, *mb;
870 int v3 = NFS_ISV3(vp);
872 nfsstats.rpccnt[NFSPROC_SETATTR]++;
873 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
875 bpos = mtod(mb, caddr_t);
878 nfsm_v3attrbuild(vap, TRUE);
879 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
882 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
883 if (vap->va_mode == (mode_t)VNOVAL)
884 sp->sa_mode = nfs_xdrneg1;
886 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
887 if (vap->va_uid == (uid_t)VNOVAL)
888 sp->sa_uid = nfs_xdrneg1;
890 sp->sa_uid = txdr_unsigned(vap->va_uid);
891 if (vap->va_gid == (gid_t)VNOVAL)
892 sp->sa_gid = nfs_xdrneg1;
894 sp->sa_gid = txdr_unsigned(vap->va_gid);
895 sp->sa_size = txdr_unsigned(vap->va_size);
896 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
897 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
899 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
901 mtx_lock(&np->n_mtx);
902 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
903 np->n_accesscache[i].stamp = 0;
904 mtx_unlock(&np->n_mtx);
905 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
906 nfsm_wcc_data(vp, wccflag);
908 nfsm_loadattr(vp, NULL);
915 * nfs lookup call, one step at a time...
916 * First look in cache
917 * If not found, unlock the directory nfsnode and do the rpc
920 nfs_lookup(struct vop_lookup_args *ap)
922 struct componentname *cnp = ap->a_cnp;
923 struct vnode *dvp = ap->a_dvp;
924 struct vnode **vpp = ap->a_vpp;
925 struct mount *mp = dvp->v_mount;
928 int flags = cnp->cn_flags;
930 struct nfsmount *nmp;
932 struct mbuf *mreq, *mrep, *md, *mb;
936 int error = 0, attrflag, fhsize, ltype;
937 int v3 = NFS_ISV3(dvp);
938 struct thread *td = cnp->cn_thread;
941 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
942 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
944 if (dvp->v_type != VDIR)
948 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
952 error = cache_lookup(dvp, vpp, cnp);
953 if (error > 0 && error != ENOENT)
957 * We only accept a positive hit in the cache if the
958 * change time of the file matches our cached copy.
959 * Otherwise, we discard the cache entry and fallback
960 * to doing a lookup RPC.
963 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
964 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
965 nfsstats.lookupcache_hits++;
966 if (cnp->cn_nameiop != LOOKUP &&
968 cnp->cn_flags |= SAVENAME;
977 } else if (error == ENOENT) {
978 if (dvp->v_iflag & VI_DOOMED)
981 * We only accept a negative hit in the cache if the
982 * modification time of the parent directory matches
983 * our cached copy. Otherwise, we discard all of the
984 * negative cache entries for this directory.
986 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
987 vattr.va_mtime.tv_sec == np->n_dmtime) {
988 nfsstats.lookupcache_hits++;
991 cache_purge_negative(dvp);
992 mtx_lock(&np->n_mtx);
994 mtx_unlock(&np->n_mtx);
998 * Cache the modification time of the parent directory in case
999 * the lookup fails and results in adding the first negative
1000 * name cache entry for the directory. Since this is reading
1001 * a single time_t, don't bother with locking. The
1002 * modification time may be a bit stale, but it must be read
1003 * before performing the lookup RPC to prevent a race where
1004 * another lookup updates the timestamp on the directory after
1005 * the lookup RPC has been performed on the server but before
1006 * n_dmtime is set at the end of this function.
1008 dmtime = np->n_vattr.va_mtime.tv_sec;
1011 nfsstats.lookupcache_misses++;
1012 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1013 len = cnp->cn_namelen;
1014 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1015 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1017 bpos = mtod(mb, caddr_t);
1018 nfsm_fhtom(dvp, v3);
1019 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1020 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
1023 nfsm_postop_attr(dvp, attrflag);
1028 nfsm_getfh(fhp, fhsize, v3);
1031 * Handle RENAME case...
1033 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1034 if (NFS_CMPFH(np, fhp, fhsize)) {
1038 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE);
1045 nfsm_postop_attr(newvp, attrflag);
1046 nfsm_postop_attr(dvp, attrflag);
1048 nfsm_loadattr(newvp, NULL);
1051 cnp->cn_flags |= SAVENAME;
1055 if (flags & ISDOTDOT) {
1056 ltype = VOP_ISLOCKED(dvp);
1057 error = vfs_busy(mp, MBF_NOWAIT);
1061 error = vfs_busy(mp, 0);
1062 vn_lock(dvp, ltype | LK_RETRY);
1064 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1074 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1079 vn_lock(dvp, ltype | LK_RETRY);
1080 if (dvp->v_iflag & VI_DOOMED) {
1093 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1097 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1105 nfsm_postop_attr(newvp, attrflag);
1106 nfsm_postop_attr(dvp, attrflag);
1108 nfsm_loadattr(newvp, NULL);
1109 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1110 cnp->cn_flags |= SAVENAME;
1111 if ((cnp->cn_flags & MAKEENTRY) &&
1112 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1113 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
1114 cache_enter(dvp, newvp, cnp);
1120 if (newvp != NULLVP) {
1125 if (error != ENOENT)
1128 /* The requested file was not found. */
1129 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1130 (flags & ISLASTCN)) {
1132 * XXX: UFS does a full VOP_ACCESS(dvp,
1133 * VWRITE) here instead of just checking
1136 if (mp->mnt_flag & MNT_RDONLY)
1138 cnp->cn_flags |= SAVENAME;
1139 return (EJUSTRETURN);
1142 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1144 * Maintain n_dmtime as the modification time
1145 * of the parent directory when the oldest -ve
1146 * name cache entry for this directory was
1147 * added. If a -ve cache entry has already
1148 * been added with a newer modification time
1149 * by a concurrent lookup, then don't bother
1150 * adding a cache entry. The modification
1151 * time of the directory might have changed
1152 * due to the file this lookup failed to find
1153 * being created. In that case a subsequent
1154 * lookup would incorrectly use the entry
1155 * added here instead of doing an extra
1158 mtx_lock(&np->n_mtx);
1159 if (np->n_dmtime <= dmtime) {
1160 if (np->n_dmtime == 0)
1161 np->n_dmtime = dmtime;
1162 mtx_unlock(&np->n_mtx);
1163 cache_enter(dvp, NULL, cnp);
1165 mtx_unlock(&np->n_mtx);
1175 * Just call nfs_bioread() to do the work.
1178 nfs_read(struct vop_read_args *ap)
1180 struct vnode *vp = ap->a_vp;
1182 switch (vp->v_type) {
1184 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1188 return (EOPNOTSUPP);
1196 nfs_readlink(struct vop_readlink_args *ap)
1198 struct vnode *vp = ap->a_vp;
1200 if (vp->v_type != VLNK)
1202 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1206 * Do a readlink rpc.
1207 * Called by nfs_doio() from below the buffer cache.
1210 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1213 int error = 0, len, attrflag;
1214 struct mbuf *mreq, *mrep, *md, *mb;
1215 int v3 = NFS_ISV3(vp);
1217 nfsstats.rpccnt[NFSPROC_READLINK]++;
1218 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1220 bpos = mtod(mb, caddr_t);
1222 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1224 nfsm_postop_attr(vp, attrflag);
1226 nfsm_strsiz(len, NFS_MAXPATHLEN);
1227 if (len == NFS_MAXPATHLEN) {
1228 struct nfsnode *np = VTONFS(vp);
1229 mtx_lock(&np->n_mtx);
1230 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1232 mtx_unlock(&np->n_mtx);
1234 nfsm_mtouio(uiop, len);
1246 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1250 struct mbuf *mreq, *mrep, *md, *mb;
1251 struct nfsmount *nmp;
1252 int error = 0, len, retlen, tsiz, eof, attrflag;
1253 int v3 = NFS_ISV3(vp);
1259 nmp = VFSTONFS(vp->v_mount);
1260 tsiz = uiop->uio_resid;
1261 mtx_lock(&nmp->nm_mtx);
1262 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1263 mtx_unlock(&nmp->nm_mtx);
1266 rsize = nmp->nm_rsize;
1267 mtx_unlock(&nmp->nm_mtx);
1269 nfsstats.rpccnt[NFSPROC_READ]++;
1270 len = (tsiz > rsize) ? rsize : tsiz;
1271 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1273 bpos = mtod(mb, caddr_t);
1275 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1277 txdr_hyper(uiop->uio_offset, tl);
1278 *(tl + 2) = txdr_unsigned(len);
1280 *tl++ = txdr_unsigned(uiop->uio_offset);
1281 *tl++ = txdr_unsigned(len);
1284 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1286 nfsm_postop_attr(vp, attrflag);
1291 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1292 eof = fxdr_unsigned(int, *(tl + 1));
1294 nfsm_loadattr(vp, NULL);
1296 nfsm_strsiz(retlen, rsize);
1297 nfsm_mtouio(uiop, retlen);
1301 if (eof || retlen == 0) {
1304 } else if (retlen < len) {
1316 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1317 int *iomode, int *must_commit)
1322 struct mbuf *mreq, *mrep, *md, *mb;
1323 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1324 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1325 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1329 if (uiop->uio_iovcnt != 1)
1330 panic("nfs: writerpc iovcnt > 1");
1333 tsiz = uiop->uio_resid;
1334 mtx_lock(&nmp->nm_mtx);
1335 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1336 mtx_unlock(&nmp->nm_mtx);
1339 wsize = nmp->nm_wsize;
1340 mtx_unlock(&nmp->nm_mtx);
1342 nfsstats.rpccnt[NFSPROC_WRITE]++;
1343 len = (tsiz > wsize) ? wsize : tsiz;
1344 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1345 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1347 bpos = mtod(mb, caddr_t);
1350 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1351 txdr_hyper(uiop->uio_offset, tl);
1353 *tl++ = txdr_unsigned(len);
1354 *tl++ = txdr_unsigned(*iomode);
1355 *tl = txdr_unsigned(len);
1359 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1360 /* Set both "begin" and "current" to non-garbage. */
1361 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1362 *tl++ = x; /* "begin offset" */
1363 *tl++ = x; /* "current offset" */
1364 x = txdr_unsigned(len);
1365 *tl++ = x; /* total to this offset */
1366 *tl = x; /* size of this write */
1368 nfsm_uiotom(uiop, len);
1369 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1371 wccflag = NFSV3_WCCCHK;
1372 nfsm_wcc_data(vp, wccflag);
1374 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1375 + NFSX_V3WRITEVERF);
1376 rlen = fxdr_unsigned(int, *tl++);
1381 } else if (rlen < len) {
1382 backup = len - rlen;
1383 uiop->uio_iov->iov_base =
1384 (char *)uiop->uio_iov->iov_base -
1386 uiop->uio_iov->iov_len += backup;
1387 uiop->uio_offset -= backup;
1388 uiop->uio_resid += backup;
1391 commit = fxdr_unsigned(int, *tl++);
1394 * Return the lowest committment level
1395 * obtained by any of the RPCs.
1397 if (committed == NFSV3WRITE_FILESYNC)
1399 else if (committed == NFSV3WRITE_DATASYNC &&
1400 commit == NFSV3WRITE_UNSTABLE)
1402 mtx_lock(&nmp->nm_mtx);
1403 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1404 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1406 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1407 } else if (bcmp((caddr_t)tl,
1408 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1410 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1413 mtx_unlock(&nmp->nm_mtx);
1416 nfsm_loadattr(vp, NULL);
1419 mtx_lock(&(VTONFS(vp))->n_mtx);
1420 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1421 mtx_unlock(&(VTONFS(vp))->n_mtx);
1429 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1430 committed = NFSV3WRITE_FILESYNC;
1431 *iomode = committed;
1433 uiop->uio_resid = tsiz;
1439 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1440 * mode set to specify the file type and the size field for rdev.
1443 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1446 struct nfsv2_sattr *sp;
1448 struct vnode *newvp = NULL;
1449 struct nfsnode *np = NULL;
1452 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1453 struct mbuf *mreq, *mrep, *md, *mb;
1455 int v3 = NFS_ISV3(dvp);
1457 if (vap->va_type == VCHR || vap->va_type == VBLK)
1458 rdev = txdr_unsigned(vap->va_rdev);
1459 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1462 return (EOPNOTSUPP);
1464 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1466 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1467 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1468 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1470 bpos = mtod(mb, caddr_t);
1471 nfsm_fhtom(dvp, v3);
1472 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1474 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1475 *tl++ = vtonfsv3_type(vap->va_type);
1476 nfsm_v3attrbuild(vap, FALSE);
1477 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1478 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1479 *tl++ = txdr_unsigned(major(vap->va_rdev));
1480 *tl = txdr_unsigned(minor(vap->va_rdev));
1483 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1484 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1485 sp->sa_uid = nfs_xdrneg1;
1486 sp->sa_gid = nfs_xdrneg1;
1488 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1489 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1491 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1493 nfsm_mtofh(dvp, newvp, v3, gotvp);
1499 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1500 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1506 nfsm_wcc_data(dvp, wccflag);
1513 if (cnp->cn_flags & MAKEENTRY)
1514 cache_enter(dvp, newvp, cnp);
1517 mtx_lock(&(VTONFS(dvp))->n_mtx);
1518 VTONFS(dvp)->n_flag |= NMODIFIED;
1520 VTONFS(dvp)->n_attrstamp = 0;
1521 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1523 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1529 * just call nfs_mknodrpc() to do the work.
1533 nfs_mknod(struct vop_mknod_args *ap)
1535 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1538 static u_long create_verf;
1540 * nfs file create call
1543 nfs_create(struct vop_create_args *ap)
1545 struct vnode *dvp = ap->a_dvp;
1546 struct vattr *vap = ap->a_vap;
1547 struct componentname *cnp = ap->a_cnp;
1548 struct nfsv2_sattr *sp;
1550 struct nfsnode *np = NULL;
1551 struct vnode *newvp = NULL;
1553 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1554 struct mbuf *mreq, *mrep, *md, *mb;
1556 int v3 = NFS_ISV3(dvp);
1559 * Oops, not for me..
1561 if (vap->va_type == VSOCK) {
1562 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1566 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) {
1569 if (vap->va_vaflags & VA_EXCLUSIVE)
1572 nfsstats.rpccnt[NFSPROC_CREATE]++;
1573 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1574 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1576 bpos = mtod(mb, caddr_t);
1577 nfsm_fhtom(dvp, v3);
1578 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1580 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1581 if (fmode & O_EXCL) {
1582 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1583 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1585 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1587 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1588 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1591 *tl++ = create_verf;
1593 IN_IFADDR_RUNLOCK();
1596 *tl = ++create_verf;
1598 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1599 nfsm_v3attrbuild(vap, FALSE);
1602 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1603 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1604 sp->sa_uid = nfs_xdrneg1;
1605 sp->sa_gid = nfs_xdrneg1;
1607 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1608 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1610 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1612 nfsm_mtofh(dvp, newvp, v3, gotvp);
1618 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1619 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1625 nfsm_wcc_data(dvp, wccflag);
1629 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1635 } else if (v3 && (fmode & O_EXCL)) {
1637 * We are normally called with only a partially initialized
1638 * VAP. Since the NFSv3 spec says that server may use the
1639 * file attributes to store the verifier, the spec requires
1640 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1641 * in atime, but we can't really assume that all servers will
1642 * so we ensure that our SETATTR sets both atime and mtime.
1644 if (vap->va_mtime.tv_sec == VNOVAL)
1645 vfs_timestamp(&vap->va_mtime);
1646 if (vap->va_atime.tv_sec == VNOVAL)
1647 vap->va_atime = vap->va_mtime;
1648 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred);
1653 if (cnp->cn_flags & MAKEENTRY)
1654 cache_enter(dvp, newvp, cnp);
1657 mtx_lock(&(VTONFS(dvp))->n_mtx);
1658 VTONFS(dvp)->n_flag |= NMODIFIED;
1660 VTONFS(dvp)->n_attrstamp = 0;
1661 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1663 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1668 * nfs file remove call
1669 * To try and make nfs semantics closer to ufs semantics, a file that has
1670 * other processes using the vnode is renamed instead of removed and then
1671 * removed later on the last close.
1672 * - If v_usecount > 1
1673 * If a rename is not already in the works
1674 * call nfs_sillyrename() to set it up
1679 nfs_remove(struct vop_remove_args *ap)
1681 struct vnode *vp = ap->a_vp;
1682 struct vnode *dvp = ap->a_dvp;
1683 struct componentname *cnp = ap->a_cnp;
1684 struct nfsnode *np = VTONFS(vp);
1689 if ((cnp->cn_flags & HASBUF) == 0)
1690 panic("nfs_remove: no name");
1691 if (vrefcnt(vp) < 1)
1692 panic("nfs_remove: bad v_usecount");
1694 if (vp->v_type == VDIR)
1696 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1697 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1699 * Purge the name cache so that the chance of a lookup for
1700 * the name succeeding while the remove is in progress is
1701 * minimized. Without node locking it can still happen, such
1702 * that an I/O op returns ESTALE, but since you get this if
1703 * another host removes the file..
1707 * throw away biocache buffers, mainly to avoid
1708 * unnecessary delayed writes later.
1710 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1712 if (error != EINTR && error != EIO)
1713 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1714 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1716 * Kludge City: If the first reply to the remove rpc is lost..
1717 * the reply to the retransmitted request will be ENOENT
1718 * since the file was in fact removed
1719 * Therefore, we cheat and return success.
1721 if (error == ENOENT)
1723 } else if (!np->n_sillyrename)
1724 error = nfs_sillyrename(dvp, vp, cnp);
1725 np->n_attrstamp = 0;
1726 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1731 * nfs file remove rpc called from nfs_inactive
1734 nfs_removeit(struct sillyrename *sp)
1737 * Make sure that the directory vnode is still valid.
1738 * XXX we should lock sp->s_dvp here.
1740 if (sp->s_dvp->v_type == VBAD)
1742 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1747 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1750 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1751 struct ucred *cred, struct thread *td)
1754 int error = 0, wccflag = NFSV3_WCCRATTR;
1755 struct mbuf *mreq, *mrep, *md, *mb;
1756 int v3 = NFS_ISV3(dvp);
1758 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1759 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1760 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1762 bpos = mtod(mb, caddr_t);
1763 nfsm_fhtom(dvp, v3);
1764 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1765 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1767 nfsm_wcc_data(dvp, wccflag);
1770 mtx_lock(&(VTONFS(dvp))->n_mtx);
1771 VTONFS(dvp)->n_flag |= NMODIFIED;
1773 VTONFS(dvp)->n_attrstamp = 0;
1774 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1776 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1781 * nfs file rename call
1784 nfs_rename(struct vop_rename_args *ap)
1786 struct vnode *fvp = ap->a_fvp;
1787 struct vnode *tvp = ap->a_tvp;
1788 struct vnode *fdvp = ap->a_fdvp;
1789 struct vnode *tdvp = ap->a_tdvp;
1790 struct componentname *tcnp = ap->a_tcnp;
1791 struct componentname *fcnp = ap->a_fcnp;
1795 if ((tcnp->cn_flags & HASBUF) == 0 ||
1796 (fcnp->cn_flags & HASBUF) == 0)
1797 panic("nfs_rename: no name");
1799 /* Check for cross-device rename */
1800 if ((fvp->v_mount != tdvp->v_mount) ||
1801 (tvp && (fvp->v_mount != tvp->v_mount))) {
1807 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1811 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1815 * We have to flush B_DELWRI data prior to renaming
1816 * the file. If we don't, the delayed-write buffers
1817 * can be flushed out later after the file has gone stale
1818 * under NFSV3. NFSV2 does not have this problem because
1819 * ( as far as I can tell ) it flushes dirty buffers more
1822 * Skip the rename operation if the fsync fails, this can happen
1823 * due to the server's volume being full, when we pushed out data
1824 * that was written back to our cache earlier. Not checking for
1825 * this condition can result in potential (silent) data loss.
1827 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1830 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1835 * If the tvp exists and is in use, sillyrename it before doing the
1836 * rename of the new file over it.
1837 * XXX Can't sillyrename a directory.
1839 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1840 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1845 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1846 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1849 if (fvp->v_type == VDIR) {
1850 if (tvp != NULL && tvp->v_type == VDIR)
1865 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1867 if (error == ENOENT)
1873 * nfs file rename rpc called from nfs_remove() above
1876 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1877 struct sillyrename *sp)
1880 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1881 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1885 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1888 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1889 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1893 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1894 struct mbuf *mreq, *mrep, *md, *mb;
1895 int v3 = NFS_ISV3(fdvp);
1897 nfsstats.rpccnt[NFSPROC_RENAME]++;
1898 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1899 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1900 nfsm_rndup(tnamelen));
1902 bpos = mtod(mb, caddr_t);
1903 nfsm_fhtom(fdvp, v3);
1904 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1905 nfsm_fhtom(tdvp, v3);
1906 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1907 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1909 nfsm_wcc_data(fdvp, fwccflag);
1910 nfsm_wcc_data(tdvp, twccflag);
1914 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1915 VTONFS(fdvp)->n_flag |= NMODIFIED;
1916 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1917 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1918 VTONFS(tdvp)->n_flag |= NMODIFIED;
1919 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1921 VTONFS(fdvp)->n_attrstamp = 0;
1922 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1925 VTONFS(tdvp)->n_attrstamp = 0;
1926 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1932 * nfs hard link create call
1935 nfs_link(struct vop_link_args *ap)
1937 struct vnode *vp = ap->a_vp;
1938 struct vnode *tdvp = ap->a_tdvp;
1939 struct componentname *cnp = ap->a_cnp;
1941 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1942 struct mbuf *mreq, *mrep, *md, *mb;
1945 if (vp->v_mount != tdvp->v_mount) {
1950 * Push all writes to the server, so that the attribute cache
1951 * doesn't get "out of sync" with the server.
1952 * XXX There should be a better way!
1954 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1957 nfsstats.rpccnt[NFSPROC_LINK]++;
1958 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1959 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1961 bpos = mtod(mb, caddr_t);
1963 nfsm_fhtom(tdvp, v3);
1964 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1965 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1967 nfsm_postop_attr(vp, attrflag);
1968 nfsm_wcc_data(tdvp, wccflag);
1972 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1973 VTONFS(tdvp)->n_flag |= NMODIFIED;
1974 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1976 VTONFS(vp)->n_attrstamp = 0;
1977 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1980 VTONFS(tdvp)->n_attrstamp = 0;
1981 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1987 * nfs symbolic link create call
1990 nfs_symlink(struct vop_symlink_args *ap)
1992 struct vnode *dvp = ap->a_dvp;
1993 struct vattr *vap = ap->a_vap;
1994 struct componentname *cnp = ap->a_cnp;
1995 struct nfsv2_sattr *sp;
1997 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1998 struct mbuf *mreq, *mrep, *md, *mb;
1999 struct vnode *newvp = NULL;
2000 int v3 = NFS_ISV3(dvp);
2002 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2003 slen = strlen(ap->a_target);
2004 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2005 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2007 bpos = mtod(mb, caddr_t);
2008 nfsm_fhtom(dvp, v3);
2009 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2011 nfsm_v3attrbuild(vap, FALSE);
2013 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2015 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2016 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2017 sp->sa_uid = nfs_xdrneg1;
2018 sp->sa_gid = nfs_xdrneg1;
2019 sp->sa_size = nfs_xdrneg1;
2020 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2021 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2025 * Issue the NFS request and get the rpc response.
2027 * Only NFSv3 responses returning an error of 0 actually return
2028 * a file handle that can be converted into newvp without having
2029 * to do an extra lookup rpc.
2031 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
2034 nfsm_mtofh(dvp, newvp, v3, gotvp);
2035 nfsm_wcc_data(dvp, wccflag);
2039 * out code jumps -> here, mrep is also freed.
2046 * If we do not have an error and we could not extract the newvp from
2047 * the response due to the request being NFSv2, we have to do a
2048 * lookup in order to obtain a newvp to return.
2050 if (error == 0 && newvp == NULL) {
2051 struct nfsnode *np = NULL;
2053 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2054 cnp->cn_cred, cnp->cn_thread, &np);
2064 mtx_lock(&(VTONFS(dvp))->n_mtx);
2065 VTONFS(dvp)->n_flag |= NMODIFIED;
2066 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2068 VTONFS(dvp)->n_attrstamp = 0;
2069 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2078 nfs_mkdir(struct vop_mkdir_args *ap)
2080 struct vnode *dvp = ap->a_dvp;
2081 struct vattr *vap = ap->a_vap;
2082 struct componentname *cnp = ap->a_cnp;
2083 struct nfsv2_sattr *sp;
2085 struct nfsnode *np = NULL;
2086 struct vnode *newvp = NULL;
2088 int error = 0, wccflag = NFSV3_WCCRATTR;
2090 struct mbuf *mreq, *mrep, *md, *mb;
2092 int v3 = NFS_ISV3(dvp);
2094 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2096 len = cnp->cn_namelen;
2097 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2098 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
2099 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2101 bpos = mtod(mb, caddr_t);
2102 nfsm_fhtom(dvp, v3);
2103 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2105 nfsm_v3attrbuild(vap, FALSE);
2107 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2108 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2109 sp->sa_uid = nfs_xdrneg1;
2110 sp->sa_gid = nfs_xdrneg1;
2111 sp->sa_size = nfs_xdrneg1;
2112 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2113 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2115 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2117 nfsm_mtofh(dvp, newvp, v3, gotvp);
2119 nfsm_wcc_data(dvp, wccflag);
2122 mtx_lock(&(VTONFS(dvp))->n_mtx);
2123 VTONFS(dvp)->n_flag |= NMODIFIED;
2124 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2126 VTONFS(dvp)->n_attrstamp = 0;
2127 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2129 if (error == 0 && newvp == NULL) {
2130 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2131 cnp->cn_thread, &np);
2134 if (newvp->v_type != VDIR)
2147 * nfs remove directory call
2150 nfs_rmdir(struct vop_rmdir_args *ap)
2152 struct vnode *vp = ap->a_vp;
2153 struct vnode *dvp = ap->a_dvp;
2154 struct componentname *cnp = ap->a_cnp;
2156 int error = 0, wccflag = NFSV3_WCCRATTR;
2157 struct mbuf *mreq, *mrep, *md, *mb;
2158 int v3 = NFS_ISV3(dvp);
2162 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2163 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
2164 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2166 bpos = mtod(mb, caddr_t);
2167 nfsm_fhtom(dvp, v3);
2168 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2169 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2171 nfsm_wcc_data(dvp, wccflag);
2174 mtx_lock(&(VTONFS(dvp))->n_mtx);
2175 VTONFS(dvp)->n_flag |= NMODIFIED;
2176 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2178 VTONFS(dvp)->n_attrstamp = 0;
2179 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2184 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2186 if (error == ENOENT)
2195 nfs_readdir(struct vop_readdir_args *ap)
2197 struct vnode *vp = ap->a_vp;
2198 struct nfsnode *np = VTONFS(vp);
2199 struct uio *uio = ap->a_uio;
2200 int tresid, error = 0;
2203 if (vp->v_type != VDIR)
2207 * First, check for hit on the EOF offset cache
2209 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2210 (np->n_flag & NMODIFIED) == 0) {
2211 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2212 mtx_lock(&np->n_mtx);
2213 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2214 mtx_unlock(&np->n_mtx);
2215 nfsstats.direofcache_hits++;
2218 mtx_unlock(&np->n_mtx);
2223 * Call nfs_bioread() to do the real work.
2225 tresid = uio->uio_resid;
2226 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2228 if (!error && uio->uio_resid == tresid) {
2229 nfsstats.direofcache_misses++;
2237 * Called from below the buffer cache by nfs_doio().
2240 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2243 struct dirent *dp = NULL;
2248 struct mbuf *mreq, *mrep, *md, *mb;
2250 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2251 struct nfsnode *dnp = VTONFS(vp);
2253 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2255 int v3 = NFS_ISV3(vp);
2258 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2259 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2260 panic("nfs readdirrpc bad uio");
2264 * If there is no cookie, assume directory was stale.
2266 nfs_dircookie_lock(dnp);
2267 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2270 nfs_dircookie_unlock(dnp);
2272 nfs_dircookie_unlock(dnp);
2273 return (NFSERR_BAD_COOKIE);
2277 * Loop around doing readdir rpc's of size nm_readdirsize
2278 * truncated to a multiple of DIRBLKSIZ.
2279 * The stopping criteria is EOF or buffer full.
2281 while (more_dirs && bigenough) {
2282 nfsstats.rpccnt[NFSPROC_READDIR]++;
2283 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2286 bpos = mtod(mb, caddr_t);
2289 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2290 *tl++ = cookie.nfsuquad[0];
2291 *tl++ = cookie.nfsuquad[1];
2292 mtx_lock(&dnp->n_mtx);
2293 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2294 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2295 mtx_unlock(&dnp->n_mtx);
2297 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2298 *tl++ = cookie.nfsuquad[0];
2300 *tl = txdr_unsigned(nmp->nm_readdirsize);
2301 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2303 nfsm_postop_attr(vp, attrflag);
2305 tl = nfsm_dissect(u_int32_t *,
2307 mtx_lock(&dnp->n_mtx);
2308 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2309 dnp->n_cookieverf.nfsuquad[1] = *tl;
2310 mtx_unlock(&dnp->n_mtx);
2316 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2317 more_dirs = fxdr_unsigned(int, *tl);
2319 /* loop thru the dir entries, doctoring them to 4bsd form */
2320 while (more_dirs && bigenough) {
2322 tl = nfsm_dissect(u_int32_t *,
2324 fileno = fxdr_hyper(tl);
2325 len = fxdr_unsigned(int, *(tl + 2));
2327 tl = nfsm_dissect(u_int32_t *,
2329 fileno = fxdr_unsigned(u_quad_t, *tl++);
2330 len = fxdr_unsigned(int, *tl);
2332 if (len <= 0 || len > NFS_MAXNAMLEN) {
2337 tlen = nfsm_rndup(len);
2339 tlen += 4; /* To ensure null termination */
2340 left = DIRBLKSIZ - blksiz;
2341 if ((tlen + DIRHDSIZ) > left) {
2342 dp->d_reclen += left;
2343 uiop->uio_iov->iov_base =
2344 (char *)uiop->uio_iov->iov_base + left;
2345 uiop->uio_iov->iov_len -= left;
2346 uiop->uio_offset += left;
2347 uiop->uio_resid -= left;
2350 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2353 dp = (struct dirent *)uiop->uio_iov->iov_base;
2354 dp->d_fileno = (int)fileno;
2356 dp->d_reclen = tlen + DIRHDSIZ;
2357 dp->d_type = DT_UNKNOWN;
2358 blksiz += dp->d_reclen;
2359 if (blksiz == DIRBLKSIZ)
2361 uiop->uio_offset += DIRHDSIZ;
2362 uiop->uio_resid -= DIRHDSIZ;
2363 uiop->uio_iov->iov_base =
2364 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2365 uiop->uio_iov->iov_len -= DIRHDSIZ;
2366 nfsm_mtouio(uiop, len);
2367 cp = uiop->uio_iov->iov_base;
2369 *cp = '\0'; /* null terminate */
2370 uiop->uio_iov->iov_base =
2371 (char *)uiop->uio_iov->iov_base + tlen;
2372 uiop->uio_iov->iov_len -= tlen;
2373 uiop->uio_offset += tlen;
2374 uiop->uio_resid -= tlen;
2376 nfsm_adv(nfsm_rndup(len));
2378 tl = nfsm_dissect(u_int32_t *,
2381 tl = nfsm_dissect(u_int32_t *,
2385 cookie.nfsuquad[0] = *tl++;
2387 cookie.nfsuquad[1] = *tl++;
2392 more_dirs = fxdr_unsigned(int, *tl);
2395 * If at end of rpc data, get the eof boolean
2398 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2399 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2404 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2405 * by increasing d_reclen for the last record.
2408 left = DIRBLKSIZ - blksiz;
2409 dp->d_reclen += left;
2410 uiop->uio_iov->iov_base =
2411 (char *)uiop->uio_iov->iov_base + left;
2412 uiop->uio_iov->iov_len -= left;
2413 uiop->uio_offset += left;
2414 uiop->uio_resid -= left;
2418 * We are now either at the end of the directory or have filled the
2422 dnp->n_direofoffset = uiop->uio_offset;
2424 if (uiop->uio_resid > 0)
2425 nfs_printf("EEK! readdirrpc resid > 0\n");
2426 nfs_dircookie_lock(dnp);
2427 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2429 nfs_dircookie_unlock(dnp);
2436 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2439 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2445 struct vnode *newvp;
2447 caddr_t bpos, dpos, dpossav1, dpossav2;
2448 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2449 struct nameidata nami, *ndp = &nami;
2450 struct componentname *cnp = &ndp->ni_cnd;
2452 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2453 struct nfsnode *dnp = VTONFS(vp), *np;
2456 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2457 int attrflag, fhsize;
2463 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2464 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2465 panic("nfs readdirplusrpc bad uio");
2471 * If there is no cookie, assume directory was stale.
2473 nfs_dircookie_lock(dnp);
2474 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2477 nfs_dircookie_unlock(dnp);
2479 nfs_dircookie_unlock(dnp);
2480 return (NFSERR_BAD_COOKIE);
2483 * Loop around doing readdir rpc's of size nm_readdirsize
2484 * truncated to a multiple of DIRBLKSIZ.
2485 * The stopping criteria is EOF or buffer full.
2487 while (more_dirs && bigenough) {
2488 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2489 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2490 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2492 bpos = mtod(mb, caddr_t);
2494 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2495 *tl++ = cookie.nfsuquad[0];
2496 *tl++ = cookie.nfsuquad[1];
2497 mtx_lock(&dnp->n_mtx);
2498 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2499 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2500 mtx_unlock(&dnp->n_mtx);
2501 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2502 *tl = txdr_unsigned(nmp->nm_rsize);
2503 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2504 nfsm_postop_attr(vp, attrflag);
2509 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2510 mtx_lock(&dnp->n_mtx);
2511 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2512 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2513 mtx_unlock(&dnp->n_mtx);
2514 more_dirs = fxdr_unsigned(int, *tl);
2516 /* loop thru the dir entries, doctoring them to 4bsd form */
2517 while (more_dirs && bigenough) {
2518 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2519 fileno = fxdr_hyper(tl);
2520 len = fxdr_unsigned(int, *(tl + 2));
2521 if (len <= 0 || len > NFS_MAXNAMLEN) {
2526 tlen = nfsm_rndup(len);
2528 tlen += 4; /* To ensure null termination*/
2529 left = DIRBLKSIZ - blksiz;
2530 if ((tlen + DIRHDSIZ) > left) {
2531 dp->d_reclen += left;
2532 uiop->uio_iov->iov_base =
2533 (char *)uiop->uio_iov->iov_base + left;
2534 uiop->uio_iov->iov_len -= left;
2535 uiop->uio_offset += left;
2536 uiop->uio_resid -= left;
2539 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2542 dp = (struct dirent *)uiop->uio_iov->iov_base;
2543 dp->d_fileno = (int)fileno;
2545 dp->d_reclen = tlen + DIRHDSIZ;
2546 dp->d_type = DT_UNKNOWN;
2547 blksiz += dp->d_reclen;
2548 if (blksiz == DIRBLKSIZ)
2550 uiop->uio_offset += DIRHDSIZ;
2551 uiop->uio_resid -= DIRHDSIZ;
2552 uiop->uio_iov->iov_base =
2553 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2554 uiop->uio_iov->iov_len -= DIRHDSIZ;
2555 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2556 cnp->cn_namelen = len;
2557 nfsm_mtouio(uiop, len);
2558 cp = uiop->uio_iov->iov_base;
2561 uiop->uio_iov->iov_base =
2562 (char *)uiop->uio_iov->iov_base + tlen;
2563 uiop->uio_iov->iov_len -= tlen;
2564 uiop->uio_offset += tlen;
2565 uiop->uio_resid -= tlen;
2567 nfsm_adv(nfsm_rndup(len));
2568 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2570 cookie.nfsuquad[0] = *tl++;
2571 cookie.nfsuquad[1] = *tl++;
2576 * Since the attributes are before the file handle
2577 * (sigh), we must skip over the attributes and then
2578 * come back and get them.
2580 attrflag = fxdr_unsigned(int, *tl);
2584 nfsm_adv(NFSX_V3FATTR);
2585 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2586 doit = fxdr_unsigned(int, *tl);
2588 * Skip loading the attrs for "..". There's a
2589 * race between loading the attrs here and
2590 * lookups that look for the directory currently
2591 * being read (in the parent). We try to acquire
2592 * the exclusive lock on ".." here, owning the
2593 * lock on the directory being read. Lookup will
2594 * hold the lock on ".." and try to acquire the
2595 * lock on the directory being read.
2597 * There are other ways of fixing this, one would
2598 * be to do a trylock on the ".." vnode and skip
2599 * loading the attrs on ".." if it happens to be
2600 * locked by another process. But skipping the
2601 * attrload on ".." seems the easiest option.
2603 if (strcmp(dp->d_name, "..") == 0) {
2606 * We've already skipped over the attrs,
2607 * skip over the filehandle. And store d_type
2610 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2611 i = fxdr_unsigned(int, *tl);
2612 nfsm_adv(nfsm_rndup(i));
2613 dp->d_type = IFTODT(VTTOIF(VDIR));
2616 nfsm_getfh(fhp, fhsize, 1);
2617 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2622 error = nfs_nget(vp->v_mount, fhp,
2623 fhsize, &np, LK_EXCLUSIVE);
2630 if (doit && bigenough) {
2635 nfsm_loadattr(newvp, NULL);
2639 IFTODT(VTTOIF(np->n_vattr.va_type));
2641 /* Update n_ctime, so subsequent lookup doesn't purge entry */
2642 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
2643 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2646 /* Just skip over the file handle */
2647 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2648 i = fxdr_unsigned(int, *tl);
2650 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2651 fhsize = fxdr_unsigned(int, *tl);
2652 nfsm_adv(nfsm_rndup(fhsize));
2655 if (newvp != NULLVP) {
2662 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2663 more_dirs = fxdr_unsigned(int, *tl);
2666 * If at end of rpc data, get the eof boolean
2669 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2670 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2675 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2676 * by increasing d_reclen for the last record.
2679 left = DIRBLKSIZ - blksiz;
2680 dp->d_reclen += left;
2681 uiop->uio_iov->iov_base =
2682 (char *)uiop->uio_iov->iov_base + left;
2683 uiop->uio_iov->iov_len -= left;
2684 uiop->uio_offset += left;
2685 uiop->uio_resid -= left;
2689 * We are now either at the end of the directory or have filled the
2693 dnp->n_direofoffset = uiop->uio_offset;
2695 if (uiop->uio_resid > 0)
2696 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2697 nfs_dircookie_lock(dnp);
2698 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2700 nfs_dircookie_unlock(dnp);
2703 if (newvp != NULLVP) {
2714 * Silly rename. To make the NFS filesystem that is stateless look a little
2715 * more like the "ufs" a remove of an active vnode is translated to a rename
2716 * to a funny looking filename that is removed by nfs_inactive on the
2717 * nfsnode. There is the potential for another process on a different client
2718 * to create the same funny name between the nfs_lookitup() fails and the
2719 * nfs_rename() completes, but...
2722 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2724 struct sillyrename *sp;
2728 unsigned int lticks;
2733 if (vp->v_type == VDIR)
2734 panic("nfs: sillyrename dir");
2736 sp = malloc(sizeof (struct sillyrename),
2737 M_NFSREQ, M_WAITOK);
2738 sp->s_cred = crhold(cnp->cn_cred);
2740 sp->s_removeit = nfs_removeit;
2744 * Fudge together a funny name.
2745 * Changing the format of the funny name to accomodate more
2746 * sillynames per directory.
2747 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2748 * CPU ticks since boot.
2750 pid = cnp->cn_thread->td_proc->p_pid;
2751 lticks = (unsigned int)ticks;
2753 sp->s_namlen = sprintf(sp->s_name,
2754 ".nfs.%08x.%04x4.4", lticks,
2756 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2757 cnp->cn_thread, NULL))
2761 error = nfs_renameit(dvp, cnp, sp);
2764 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2765 cnp->cn_thread, &np);
2766 np->n_sillyrename = sp;
2771 free((caddr_t)sp, M_NFSREQ);
2776 * Look up a file name and optionally either update the file handle or
2777 * allocate an nfsnode, depending on the value of npp.
2778 * npp == NULL --> just do the lookup
2779 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2781 * *npp != NULL --> update the file handle in the vnode
2784 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2785 struct thread *td, struct nfsnode **npp)
2787 struct vnode *newvp = NULL;
2788 struct nfsnode *np, *dnp = VTONFS(dvp);
2790 int error = 0, fhlen, attrflag;
2791 struct mbuf *mreq, *mrep, *md, *mb;
2793 int v3 = NFS_ISV3(dvp);
2795 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2796 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2797 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2799 bpos = mtod(mb, caddr_t);
2800 nfsm_fhtom(dvp, v3);
2801 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2802 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2803 if (npp && !error) {
2804 nfsm_getfh(nfhp, fhlen, v3);
2807 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2808 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2809 np->n_fhp = &np->n_fh;
2810 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2811 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2812 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2813 np->n_fhsize = fhlen;
2815 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2819 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2827 nfsm_postop_attr(newvp, attrflag);
2828 if (!attrflag && *npp == NULL) {
2837 nfsm_loadattr(newvp, NULL);
2841 if (npp && *npp == NULL) {
2856 * Nfs Version 3 commit rpc
2859 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2863 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2865 int error = 0, wccflag = NFSV3_WCCRATTR;
2866 struct mbuf *mreq, *mrep, *md, *mb;
2868 mtx_lock(&nmp->nm_mtx);
2869 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2870 mtx_unlock(&nmp->nm_mtx);
2873 mtx_unlock(&nmp->nm_mtx);
2874 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2875 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2877 bpos = mtod(mb, caddr_t);
2879 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2880 txdr_hyper(offset, tl);
2882 *tl = txdr_unsigned(cnt);
2883 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2884 nfsm_wcc_data(vp, wccflag);
2886 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2887 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2888 NFSX_V3WRITEVERF)) {
2889 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2891 error = NFSERR_STALEWRITEVERF;
2901 * For async requests when nfsiod(s) are running, queue the request by
2902 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2906 nfs_strategy(struct vop_strategy_args *ap)
2908 struct buf *bp = ap->a_bp;
2911 KASSERT(!(bp->b_flags & B_DONE),
2912 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2913 BUF_ASSERT_HELD(bp);
2915 if (bp->b_iocmd == BIO_READ)
2921 * If the op is asynchronous and an i/o daemon is waiting
2922 * queue the request, wake it up and wait for completion
2923 * otherwise just do it ourselves.
2925 if ((bp->b_flags & B_ASYNC) == 0 ||
2926 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2927 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2932 * fsync vnode op. Just call nfs_flush() with commit == 1.
2936 nfs_fsync(struct vop_fsync_args *ap)
2939 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1));
2943 * Flush all the blocks associated with a vnode.
2944 * Walk through the buffer pool and push any dirty pages
2945 * associated with the vnode.
2948 nfs_flush(struct vnode *vp, int waitfor, int commit)
2950 struct nfsnode *np = VTONFS(vp);
2954 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2955 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2957 u_quad_t off, endoff, toff;
2958 struct ucred* wcred = NULL;
2959 struct buf **bvec = NULL;
2961 struct thread *td = curthread;
2962 #ifndef NFS_COMMITBVECSIZ
2963 #define NFS_COMMITBVECSIZ 20
2965 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2966 int bvecsize = 0, bveccount;
2968 if (nmp->nm_flag & NFSMNT_INT)
2969 slpflag = NFS_PCATCH;
2974 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2975 * server, but has not been committed to stable storage on the server
2976 * yet. On the first pass, the byte range is worked out and the commit
2977 * rpc is done. On the second pass, nfs_writebp() is called to do the
2984 if (NFS_ISV3(vp) && commit) {
2985 if (bvec != NULL && bvec != bvec_on_stack)
2988 * Count up how many buffers waiting for a commit.
2992 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2993 if (!BUF_ISLOCKED(bp) &&
2994 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2995 == (B_DELWRI | B_NEEDCOMMIT))
2999 * Allocate space to remember the list of bufs to commit. It is
3000 * important to use M_NOWAIT here to avoid a race with nfs_write.
3001 * If we can't get memory (for whatever reason), we will end up
3002 * committing the buffers one-by-one in the loop below.
3004 if (bveccount > NFS_COMMITBVECSIZ) {
3006 * Release the vnode interlock to avoid a lock
3010 bvec = (struct buf **)
3011 malloc(bveccount * sizeof(struct buf *),
3015 bvec = bvec_on_stack;
3016 bvecsize = NFS_COMMITBVECSIZ;
3018 bvecsize = bveccount;
3020 bvec = bvec_on_stack;
3021 bvecsize = NFS_COMMITBVECSIZ;
3023 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3024 if (bvecpos >= bvecsize)
3026 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3027 nbp = TAILQ_NEXT(bp, b_bobufs);
3030 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3031 (B_DELWRI | B_NEEDCOMMIT)) {
3033 nbp = TAILQ_NEXT(bp, b_bobufs);
3039 * Work out if all buffers are using the same cred
3040 * so we can deal with them all with one commit.
3042 * NOTE: we are not clearing B_DONE here, so we have
3043 * to do it later on in this routine if we intend to
3044 * initiate I/O on the bp.
3046 * Note: to avoid loopback deadlocks, we do not
3047 * assign b_runningbufspace.
3050 wcred = bp->b_wcred;
3051 else if (wcred != bp->b_wcred)
3053 vfs_busy_pages(bp, 1);
3057 * bp is protected by being locked, but nbp is not
3058 * and vfs_busy_pages() may sleep. We have to
3061 nbp = TAILQ_NEXT(bp, b_bobufs);
3064 * A list of these buffers is kept so that the
3065 * second loop knows which buffers have actually
3066 * been committed. This is necessary, since there
3067 * may be a race between the commit rpc and new
3068 * uncommitted writes on the file.
3070 bvec[bvecpos++] = bp;
3071 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3075 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3083 * Commit data on the server, as required.
3084 * If all bufs are using the same wcred, then use that with
3085 * one call for all of them, otherwise commit each one
3088 if (wcred != NOCRED)
3089 retv = nfs_commit(vp, off, (int)(endoff - off),
3093 for (i = 0; i < bvecpos; i++) {
3096 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3098 size = (u_quad_t)(bp->b_dirtyend
3100 retv = nfs_commit(vp, off, (int)size,
3106 if (retv == NFSERR_STALEWRITEVERF)
3107 nfs_clearcommit(vp->v_mount);
3110 * Now, either mark the blocks I/O done or mark the
3111 * blocks dirty, depending on whether the commit
3114 for (i = 0; i < bvecpos; i++) {
3116 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3119 * Error, leave B_DELWRI intact
3121 vfs_unbusy_pages(bp);
3125 * Success, remove B_DELWRI ( bundirty() ).
3127 * b_dirtyoff/b_dirtyend seem to be NFS
3128 * specific. We should probably move that
3129 * into bundirty(). XXX
3132 bp->b_flags |= B_ASYNC;
3134 bp->b_flags &= ~B_DONE;
3135 bp->b_ioflags &= ~BIO_ERROR;
3136 bp->b_dirtyoff = bp->b_dirtyend = 0;
3143 * Start/do any write(s) that are required.
3147 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3148 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3149 if (waitfor != MNT_WAIT || passone)
3152 error = BUF_TIMELOCK(bp,
3153 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3154 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
3159 if (error == ENOLCK) {
3163 if (nfs_sigintr(nmp, td)) {
3167 if (slpflag & PCATCH) {
3173 if ((bp->b_flags & B_DELWRI) == 0)
3174 panic("nfs_fsync: not dirty");
3175 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3181 if (passone || !commit)
3182 bp->b_flags |= B_ASYNC;
3184 bp->b_flags |= B_ASYNC;
3186 if (nfs_sigintr(nmp, td)) {
3197 if (waitfor == MNT_WAIT) {
3198 while (bo->bo_numoutput) {
3199 error = bufobj_wwait(bo, slpflag, slptimeo);
3202 error = nfs_sigintr(nmp, td);
3205 if (slpflag & PCATCH) {
3212 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3217 * Wait for all the async IO requests to drain
3220 mtx_lock(&np->n_mtx);
3221 while (np->n_directio_asyncwr > 0) {
3222 np->n_flag |= NFSYNCWAIT;
3223 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3224 &np->n_mtx, slpflag | (PRIBIO + 1),
3227 if (nfs_sigintr(nmp, td)) {
3228 mtx_unlock(&np->n_mtx);
3234 mtx_unlock(&np->n_mtx);
3237 mtx_lock(&np->n_mtx);
3238 if (np->n_flag & NWRITEERR) {
3239 error = np->n_error;
3240 np->n_flag &= ~NWRITEERR;
3242 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3243 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3244 np->n_flag &= ~NMODIFIED;
3245 mtx_unlock(&np->n_mtx);
3247 if (bvec != NULL && bvec != bvec_on_stack)
3253 * NFS advisory byte-level locks.
3256 nfs_advlock(struct vop_advlock_args *ap)
3258 struct vnode *vp = ap->a_vp;
3262 error = vn_lock(vp, LK_SHARED);
3265 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3266 size = VTONFS(vp)->n_size;
3268 error = lf_advlock(ap, &(vp->v_lockf), size);
3271 error = nfs_advlock_p(ap);
3280 * NFS advisory byte-level locks.
3283 nfs_advlockasync(struct vop_advlockasync_args *ap)
3285 struct vnode *vp = ap->a_vp;
3289 error = vn_lock(vp, LK_SHARED);
3292 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3293 size = VTONFS(vp)->n_size;
3295 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3304 * Print out the contents of an nfsnode.
3307 nfs_print(struct vop_print_args *ap)
3309 struct vnode *vp = ap->a_vp;
3310 struct nfsnode *np = VTONFS(vp);
3312 nfs_printf("\tfileid %ld fsid 0x%x",
3313 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3314 if (vp->v_type == VFIFO)
3321 * This is the "real" nfs::bwrite(struct buf*).
3322 * We set B_CACHE if this is a VMIO buffer.
3325 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3328 int oldflags = bp->b_flags;
3334 BUF_ASSERT_HELD(bp);
3336 if (bp->b_flags & B_INVAL) {
3341 bp->b_flags |= B_CACHE;
3344 * Undirty the bp. We will redirty it later if the I/O fails.
3349 bp->b_flags &= ~B_DONE;
3350 bp->b_ioflags &= ~BIO_ERROR;
3351 bp->b_iocmd = BIO_WRITE;
3353 bufobj_wref(bp->b_bufobj);
3354 curthread->td_ru.ru_oublock++;
3358 * Note: to avoid loopback deadlocks, we do not
3359 * assign b_runningbufspace.
3361 vfs_busy_pages(bp, 1);
3364 bp->b_iooffset = dbtob(bp->b_blkno);
3367 if( (oldflags & B_ASYNC) == 0) {
3368 int rtval = bufwait(bp);
3370 if (oldflags & B_DELWRI) {
3383 * nfs special file access vnode op.
3384 * Essentially just get vattr and then imitate iaccess() since the device is
3385 * local to the client.
3388 nfsspec_access(struct vop_access_args *ap)
3391 struct ucred *cred = ap->a_cred;
3392 struct vnode *vp = ap->a_vp;
3393 accmode_t accmode = ap->a_accmode;
3398 * Disallow write attempts on filesystems mounted read-only;
3399 * unless the file is a socket, fifo, or a block or character
3400 * device resident on the filesystem.
3402 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3403 switch (vp->v_type) {
3413 error = VOP_GETATTR(vp, vap, cred);
3416 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3417 accmode, cred, NULL);
3423 * Read wrapper for fifos.
3426 nfsfifo_read(struct vop_read_args *ap)
3428 struct nfsnode *np = VTONFS(ap->a_vp);
3434 mtx_lock(&np->n_mtx);
3436 getnanotime(&np->n_atim);
3437 mtx_unlock(&np->n_mtx);
3438 error = fifo_specops.vop_read(ap);
3443 * Write wrapper for fifos.
3446 nfsfifo_write(struct vop_write_args *ap)
3448 struct nfsnode *np = VTONFS(ap->a_vp);
3453 mtx_lock(&np->n_mtx);
3455 getnanotime(&np->n_mtim);
3456 mtx_unlock(&np->n_mtx);
3457 return(fifo_specops.vop_write(ap));
3461 * Close wrapper for fifos.
3463 * Update the times on the nfsnode then do fifo close.
3466 nfsfifo_close(struct vop_close_args *ap)
3468 struct vnode *vp = ap->a_vp;
3469 struct nfsnode *np = VTONFS(vp);
3473 mtx_lock(&np->n_mtx);
3474 if (np->n_flag & (NACC | NUPD)) {
3476 if (np->n_flag & NACC)
3478 if (np->n_flag & NUPD)
3481 if (vrefcnt(vp) == 1 &&
3482 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3484 if (np->n_flag & NACC)
3485 vattr.va_atime = np->n_atim;
3486 if (np->n_flag & NUPD)
3487 vattr.va_mtime = np->n_mtim;
3488 mtx_unlock(&np->n_mtx);
3489 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3493 mtx_unlock(&np->n_mtx);
3495 return (fifo_specops.vop_close(ap));
3499 * Just call nfs_writebp() with the force argument set to 1.
3501 * NOTE: B_DONE may or may not be set in a_bp on call.
3504 nfs_bwrite(struct buf *bp)
3507 return (nfs_writebp(bp, 1, curthread));
3510 struct buf_ops buf_ops_nfs = {
3511 .bop_name = "buf_ops_nfs",
3512 .bop_write = nfs_bwrite,
3513 .bop_strategy = bufstrategy,
3514 .bop_sync = bufsync,
3515 .bop_bdflush = bufbdflush,