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 <nfs/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 enum nfsiod_state nfs_iodwant[NFS_MAXASYNCDAEMON];
216 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
217 int nfs_numasync = 0;
218 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
220 SYSCTL_DECL(_vfs_nfs);
222 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
223 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
224 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
226 static int nfs_prime_access_cache = 0;
227 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
228 &nfs_prime_access_cache, 0,
229 "Prime NFS ACCESS cache when fetching attributes");
231 static int nfsv3_commit_on_close = 0;
232 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
233 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
235 static int nfs_clean_pages_on_close = 1;
236 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
237 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
239 int nfs_directio_enable = 0;
240 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
241 &nfs_directio_enable, 0, "Enable NFS directio");
244 * This sysctl allows other processes to mmap a file that has been opened
245 * O_DIRECT by a process. In general, having processes mmap the file while
246 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
247 * this by default to prevent DoS attacks - to prevent a malicious user from
248 * opening up files O_DIRECT preventing other users from mmap'ing these
249 * files. "Protected" environments where stricter consistency guarantees are
250 * required can disable this knob. The process that opened the file O_DIRECT
251 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
254 int nfs_directio_allow_mmap = 1;
255 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
256 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
259 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
260 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
262 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
263 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
266 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
267 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
268 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
272 * The list of locks after the description of the lock is the ordering
273 * of other locks acquired with the lock held.
274 * np->n_mtx : Protects the fields in the nfsnode.
276 VI_MTX (acquired indirectly)
277 * nmp->nm_mtx : Protects the fields in the nfsmount.
279 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
280 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
283 * rep->r_mtx : Protects the fields in an nfsreq.
287 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
288 struct ucred *cred, uint32_t *retmode)
292 int error = 0, attrflag, i, lrupos;
294 struct mbuf *mreq, *mrep, *md, *mb;
297 struct nfsnode *np = VTONFS(vp);
299 nfsstats.rpccnt[NFSPROC_ACCESS]++;
300 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
302 bpos = mtod(mb, caddr_t);
304 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
305 *tl = txdr_unsigned(wmode);
306 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
307 nfsm_postop_attr(vp, attrflag);
310 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
311 rmode = fxdr_unsigned(u_int32_t, *tl);
312 mtx_lock(&np->n_mtx);
313 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
314 if (np->n_accesscache[i].uid == cred->cr_uid) {
315 np->n_accesscache[i].mode = rmode;
316 np->n_accesscache[i].stamp = time_second;
319 if (i > 0 && np->n_accesscache[i].stamp <
320 np->n_accesscache[lrupos].stamp)
323 if (i == NFS_ACCESSCACHESIZE) {
324 np->n_accesscache[lrupos].uid = cred->cr_uid;
325 np->n_accesscache[lrupos].mode = rmode;
326 np->n_accesscache[lrupos].stamp = time_second;
328 mtx_unlock(&np->n_mtx);
331 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
337 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
345 * nfs access vnode op.
346 * For nfs version 2, just return ok. File accesses may fail later.
347 * For nfs version 3, use the access rpc to check accessibility. If file modes
348 * are changed on the server, accesses might still fail later.
351 nfs_access(struct vop_access_args *ap)
353 struct vnode *vp = ap->a_vp;
354 int error = 0, i, gotahit;
355 u_int32_t mode, rmode, wmode;
356 int v3 = NFS_ISV3(vp);
357 struct nfsnode *np = VTONFS(vp);
360 * Disallow write attempts on filesystems mounted read-only;
361 * unless the file is a socket, fifo, or a block or character
362 * device resident on the filesystem.
364 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
365 switch (vp->v_type) {
375 * For nfs v3, check to see if we have done this recently, and if
376 * so return our cached result instead of making an ACCESS call.
377 * If not, do an access rpc, otherwise you are stuck emulating
378 * ufs_access() locally using the vattr. This may not be correct,
379 * since the server may apply other access criteria such as
380 * client uid-->server uid mapping that we do not know about.
383 if (ap->a_accmode & VREAD)
384 mode = NFSV3ACCESS_READ;
387 if (vp->v_type != VDIR) {
388 if (ap->a_accmode & VWRITE)
389 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
390 if (ap->a_accmode & VEXEC)
391 mode |= NFSV3ACCESS_EXECUTE;
393 if (ap->a_accmode & VWRITE)
394 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
396 if (ap->a_accmode & VEXEC)
397 mode |= NFSV3ACCESS_LOOKUP;
399 /* XXX safety belt, only make blanket request if caching */
400 if (nfsaccess_cache_timeout > 0) {
401 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
402 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
403 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
409 * Does our cached result allow us to give a definite yes to
413 mtx_lock(&np->n_mtx);
414 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
415 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
416 if (time_second < (np->n_accesscache[i].stamp +
417 nfsaccess_cache_timeout) &&
418 (np->n_accesscache[i].mode & mode) == mode) {
419 nfsstats.accesscache_hits++;
425 mtx_unlock(&np->n_mtx);
428 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
429 ap->a_cred->cr_uid, mode);
431 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
432 ap->a_cred->cr_uid, mode);
436 * Either a no, or a don't know. Go to the wire.
438 nfsstats.accesscache_misses++;
439 error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred,
442 if ((rmode & mode) != mode)
448 if ((error = nfsspec_access(ap)) != 0) {
452 * Attempt to prevent a mapped root from accessing a file
453 * which it shouldn't. We try to read a byte from the file
454 * if the user is root and the file is not zero length.
455 * After calling nfsspec_access, we should have the correct
458 mtx_lock(&np->n_mtx);
459 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
460 && VTONFS(vp)->n_size > 0) {
465 mtx_unlock(&np->n_mtx);
468 auio.uio_iov = &aiov;
472 auio.uio_segflg = UIO_SYSSPACE;
473 auio.uio_rw = UIO_READ;
474 auio.uio_td = ap->a_td;
476 if (vp->v_type == VREG)
477 error = nfs_readrpc(vp, &auio, ap->a_cred);
478 else if (vp->v_type == VDIR) {
480 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
482 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
483 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
485 } else if (vp->v_type == VLNK)
486 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
490 mtx_unlock(&np->n_mtx);
495 int nfs_otw_getattr_avoid = 0;
499 * Check to see if the type is ok
500 * and that deletion is not in progress.
501 * For paged in text files, you will need to flush the page cache
502 * if consistency is lost.
506 nfs_open(struct vop_open_args *ap)
508 struct vnode *vp = ap->a_vp;
509 struct nfsnode *np = VTONFS(vp);
512 int fmode = ap->a_mode;
514 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
518 * Get a valid lease. If cached data is stale, flush it.
520 mtx_lock(&np->n_mtx);
521 if (np->n_flag & NMODIFIED) {
522 mtx_unlock(&np->n_mtx);
523 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
524 if (error == EINTR || error == EIO)
526 mtx_lock(&np->n_mtx);
528 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
529 if (vp->v_type == VDIR)
530 np->n_direofoffset = 0;
531 mtx_unlock(&np->n_mtx);
532 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
535 mtx_lock(&np->n_mtx);
536 np->n_mtime = vattr.va_mtime;
538 mtx_unlock(&np->n_mtx);
539 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
542 mtx_lock(&np->n_mtx);
543 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
544 if (vp->v_type == VDIR)
545 np->n_direofoffset = 0;
546 mtx_unlock(&np->n_mtx);
547 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
548 if (error == EINTR || error == EIO) {
551 mtx_lock(&np->n_mtx);
552 np->n_mtime = vattr.va_mtime;
556 * If the object has >= 1 O_DIRECT active opens, we disable caching.
558 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
559 if (np->n_directio_opens == 0) {
560 mtx_unlock(&np->n_mtx);
561 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
564 mtx_lock(&np->n_mtx);
565 np->n_flag |= NNONCACHE;
567 np->n_directio_opens++;
569 mtx_unlock(&np->n_mtx);
570 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
576 * What an NFS client should do upon close after writing is a debatable issue.
577 * Most NFS clients push delayed writes to the server upon close, basically for
579 * 1 - So that any write errors may be reported back to the client process
580 * doing the close system call. By far the two most likely errors are
581 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
582 * 2 - To put a worst case upper bound on cache inconsistency between
583 * multiple clients for the file.
584 * There is also a consistency problem for Version 2 of the protocol w.r.t.
585 * not being able to tell if other clients are writing a file concurrently,
586 * since there is no way of knowing if the changed modify time in the reply
587 * is only due to the write for this client.
588 * (NFS Version 3 provides weak cache consistency data in the reply that
589 * should be sufficient to detect and handle this case.)
591 * The current code does the following:
592 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
593 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
594 * or commit them (this satisfies 1 and 2 except for the
595 * case where the server crashes after this close but
596 * before the commit RPC, which is felt to be "good
597 * enough". Changing the last argument to nfs_flush() to
598 * a 1 would force a commit operation, if it is felt a
599 * commit is necessary now.
603 nfs_close(struct vop_close_args *ap)
605 struct vnode *vp = ap->a_vp;
606 struct nfsnode *np = VTONFS(vp);
608 int fmode = ap->a_fflag;
610 if (vp->v_type == VREG) {
612 * Examine and clean dirty pages, regardless of NMODIFIED.
613 * This closes a major hole in close-to-open consistency.
614 * We want to push out all dirty pages (and buffers) on
615 * close, regardless of whether they were dirtied by
616 * mmap'ed writes or via write().
618 if (nfs_clean_pages_on_close && vp->v_object) {
619 VM_OBJECT_LOCK(vp->v_object);
620 vm_object_page_clean(vp->v_object, 0, 0, 0);
621 VM_OBJECT_UNLOCK(vp->v_object);
623 mtx_lock(&np->n_mtx);
624 if (np->n_flag & NMODIFIED) {
625 mtx_unlock(&np->n_mtx);
628 * Under NFSv3 we have dirty buffers to dispose of. We
629 * must flush them to the NFS server. We have the option
630 * of waiting all the way through the commit rpc or just
631 * waiting for the initial write. The default is to only
632 * wait through the initial write so the data is in the
633 * server's cache, which is roughly similar to the state
634 * a standard disk subsystem leaves the file in on close().
636 * We cannot clear the NMODIFIED bit in np->n_flag due to
637 * potential races with other processes, and certainly
638 * cannot clear it if we don't commit.
640 int cm = nfsv3_commit_on_close ? 1 : 0;
641 error = nfs_flush(vp, MNT_WAIT, cm);
642 /* np->n_flag &= ~NMODIFIED; */
644 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
645 mtx_lock(&np->n_mtx);
647 if (np->n_flag & NWRITEERR) {
648 np->n_flag &= ~NWRITEERR;
651 mtx_unlock(&np->n_mtx);
653 if (nfs_directio_enable)
654 KASSERT((np->n_directio_asyncwr == 0),
655 ("nfs_close: dirty unflushed (%d) directio buffers\n",
656 np->n_directio_asyncwr));
657 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
658 mtx_lock(&np->n_mtx);
659 KASSERT((np->n_directio_opens > 0),
660 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
661 np->n_directio_opens--;
662 if (np->n_directio_opens == 0)
663 np->n_flag &= ~NNONCACHE;
664 mtx_unlock(&np->n_mtx);
670 * nfs getattr call from vfs.
673 nfs_getattr(struct vop_getattr_args *ap)
675 struct vnode *vp = ap->a_vp;
676 struct nfsnode *np = VTONFS(vp);
677 struct thread *td = curthread;
678 struct vattr *vap = ap->a_vap;
682 struct mbuf *mreq, *mrep, *md, *mb;
683 int v3 = NFS_ISV3(vp);
686 * Update local times for special files.
688 mtx_lock(&np->n_mtx);
689 if (np->n_flag & (NACC | NUPD))
691 mtx_unlock(&np->n_mtx);
693 * First look in the cache.
695 if (nfs_getattrcache(vp, &vattr) == 0)
697 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
698 nfsstats.accesscache_misses++;
699 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL);
700 if (nfs_getattrcache(vp, &vattr) == 0)
703 nfsstats.rpccnt[NFSPROC_GETATTR]++;
704 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
706 bpos = mtod(mb, caddr_t);
708 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
710 nfsm_loadattr(vp, &vattr);
714 vap->va_type = vattr.va_type;
715 vap->va_mode = vattr.va_mode;
716 vap->va_nlink = vattr.va_nlink;
717 vap->va_uid = vattr.va_uid;
718 vap->va_gid = vattr.va_gid;
719 vap->va_fsid = vattr.va_fsid;
720 vap->va_fileid = vattr.va_fileid;
721 vap->va_size = vattr.va_size;
722 vap->va_blocksize = vattr.va_blocksize;
723 vap->va_atime = vattr.va_atime;
724 vap->va_mtime = vattr.va_mtime;
725 vap->va_ctime = vattr.va_ctime;
726 vap->va_gen = vattr.va_gen;
727 vap->va_flags = vattr.va_flags;
728 vap->va_rdev = vattr.va_rdev;
729 vap->va_bytes = vattr.va_bytes;
730 vap->va_filerev = vattr.va_filerev;
739 nfs_setattr(struct vop_setattr_args *ap)
741 struct vnode *vp = ap->a_vp;
742 struct nfsnode *np = VTONFS(vp);
743 struct vattr *vap = ap->a_vap;
744 struct thread *td = curthread;
753 * Setting of flags is not supported.
755 if (vap->va_flags != VNOVAL)
759 * Disallow write attempts if the filesystem is mounted read-only.
761 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
762 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
763 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
764 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
768 if (vap->va_size != VNOVAL) {
769 switch (vp->v_type) {
776 if (vap->va_mtime.tv_sec == VNOVAL &&
777 vap->va_atime.tv_sec == VNOVAL &&
778 vap->va_mode == (mode_t)VNOVAL &&
779 vap->va_uid == (uid_t)VNOVAL &&
780 vap->va_gid == (gid_t)VNOVAL)
782 vap->va_size = VNOVAL;
786 * Disallow write attempts if the filesystem is
789 if (vp->v_mount->mnt_flag & MNT_RDONLY)
792 * We run vnode_pager_setsize() early (why?),
793 * we must set np->n_size now to avoid vinvalbuf
794 * V_SAVE races that might setsize a lower
797 mtx_lock(&np->n_mtx);
799 mtx_unlock(&np->n_mtx);
800 error = nfs_meta_setsize(vp, ap->a_cred, td,
802 mtx_lock(&np->n_mtx);
803 if (np->n_flag & NMODIFIED) {
805 mtx_unlock(&np->n_mtx);
806 if (vap->va_size == 0)
807 error = nfs_vinvalbuf(vp, 0, td, 1);
809 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
811 vnode_pager_setsize(vp, tsize);
815 mtx_unlock(&np->n_mtx);
817 * np->n_size has already been set to vap->va_size
818 * in nfs_meta_setsize(). We must set it again since
819 * nfs_loadattrcache() could be called through
820 * nfs_meta_setsize() and could modify np->n_size.
822 mtx_lock(&np->n_mtx);
823 np->n_vattr.va_size = np->n_size = vap->va_size;
824 mtx_unlock(&np->n_mtx);
827 mtx_lock(&np->n_mtx);
828 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
829 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
830 mtx_unlock(&np->n_mtx);
831 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
832 (error == EINTR || error == EIO))
835 mtx_unlock(&np->n_mtx);
837 error = nfs_setattrrpc(vp, vap, ap->a_cred);
838 if (error && vap->va_size != VNOVAL) {
839 mtx_lock(&np->n_mtx);
840 np->n_size = np->n_vattr.va_size = tsize;
841 vnode_pager_setsize(vp, tsize);
842 mtx_unlock(&np->n_mtx);
849 * Do an nfs setattr rpc.
852 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
854 struct nfsv2_sattr *sp;
855 struct nfsnode *np = VTONFS(vp);
858 int error = 0, i, wccflag = NFSV3_WCCRATTR;
859 struct mbuf *mreq, *mrep, *md, *mb;
860 int v3 = NFS_ISV3(vp);
862 nfsstats.rpccnt[NFSPROC_SETATTR]++;
863 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
865 bpos = mtod(mb, caddr_t);
868 nfsm_v3attrbuild(vap, TRUE);
869 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
872 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
873 if (vap->va_mode == (mode_t)VNOVAL)
874 sp->sa_mode = nfs_xdrneg1;
876 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
877 if (vap->va_uid == (uid_t)VNOVAL)
878 sp->sa_uid = nfs_xdrneg1;
880 sp->sa_uid = txdr_unsigned(vap->va_uid);
881 if (vap->va_gid == (gid_t)VNOVAL)
882 sp->sa_gid = nfs_xdrneg1;
884 sp->sa_gid = txdr_unsigned(vap->va_gid);
885 sp->sa_size = txdr_unsigned(vap->va_size);
886 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
887 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
889 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
891 mtx_lock(&np->n_mtx);
892 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
893 np->n_accesscache[i].stamp = 0;
894 mtx_unlock(&np->n_mtx);
895 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
896 nfsm_wcc_data(vp, wccflag);
898 nfsm_loadattr(vp, NULL);
905 * nfs lookup call, one step at a time...
906 * First look in cache
907 * If not found, unlock the directory nfsnode and do the rpc
910 nfs_lookup(struct vop_lookup_args *ap)
912 struct componentname *cnp = ap->a_cnp;
913 struct vnode *dvp = ap->a_dvp;
914 struct vnode **vpp = ap->a_vpp;
915 struct mount *mp = dvp->v_mount;
917 struct timespec dmtime;
918 int flags = cnp->cn_flags;
920 struct nfsmount *nmp;
922 struct mbuf *mreq, *mrep, *md, *mb;
925 struct nfsnode *np, *newnp;
926 int error = 0, attrflag, fhsize, ltype;
927 int v3 = NFS_ISV3(dvp);
928 struct thread *td = cnp->cn_thread;
931 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
932 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
934 if (dvp->v_type != VDIR)
938 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
942 error = cache_lookup(dvp, vpp, cnp);
943 if (error > 0 && error != ENOENT)
947 * We only accept a positive hit in the cache if the
948 * change time of the file matches our cached copy.
949 * Otherwise, we discard the cache entry and fallback
950 * to doing a lookup RPC.
952 * To better handle stale file handles and attributes,
953 * clear the attribute cache of this node if it is a
954 * leaf component, part of an open() call, and not
955 * locally modified before fetching the attributes.
956 * This should allow stale file handles to be detected
957 * here where we can fall back to a LOOKUP RPC to
958 * recover rather than having nfs_open() detect the
959 * stale file handle and failing open(2) with ESTALE.
962 newnp = VTONFS(newvp);
963 if ((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;
1279 int error = 0, len, retlen, tsiz, eof, attrflag;
1280 int v3 = NFS_ISV3(vp);
1286 nmp = VFSTONFS(vp->v_mount);
1287 tsiz = uiop->uio_resid;
1288 mtx_lock(&nmp->nm_mtx);
1289 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1290 mtx_unlock(&nmp->nm_mtx);
1293 rsize = nmp->nm_rsize;
1294 mtx_unlock(&nmp->nm_mtx);
1296 nfsstats.rpccnt[NFSPROC_READ]++;
1297 len = (tsiz > rsize) ? rsize : tsiz;
1298 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1300 bpos = mtod(mb, caddr_t);
1302 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1304 txdr_hyper(uiop->uio_offset, tl);
1305 *(tl + 2) = txdr_unsigned(len);
1307 *tl++ = txdr_unsigned(uiop->uio_offset);
1308 *tl++ = txdr_unsigned(len);
1311 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1313 nfsm_postop_attr(vp, attrflag);
1318 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1319 eof = fxdr_unsigned(int, *(tl + 1));
1321 nfsm_loadattr(vp, NULL);
1323 nfsm_strsiz(retlen, rsize);
1324 nfsm_mtouio(uiop, retlen);
1328 if (eof || retlen == 0) {
1331 } else if (retlen < len) {
1343 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1344 int *iomode, int *must_commit)
1349 struct mbuf *mreq, *mrep, *md, *mb;
1350 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1351 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1352 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1355 KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1"));
1357 tsiz = uiop->uio_resid;
1358 mtx_lock(&nmp->nm_mtx);
1359 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1360 mtx_unlock(&nmp->nm_mtx);
1363 wsize = nmp->nm_wsize;
1364 mtx_unlock(&nmp->nm_mtx);
1366 nfsstats.rpccnt[NFSPROC_WRITE]++;
1367 len = (tsiz > wsize) ? wsize : tsiz;
1368 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1369 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1371 bpos = mtod(mb, caddr_t);
1374 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1375 txdr_hyper(uiop->uio_offset, tl);
1377 *tl++ = txdr_unsigned(len);
1378 *tl++ = txdr_unsigned(*iomode);
1379 *tl = txdr_unsigned(len);
1383 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1384 /* Set both "begin" and "current" to non-garbage. */
1385 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1386 *tl++ = x; /* "begin offset" */
1387 *tl++ = x; /* "current offset" */
1388 x = txdr_unsigned(len);
1389 *tl++ = x; /* total to this offset */
1390 *tl = x; /* size of this write */
1392 nfsm_uiotom(uiop, len);
1393 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1395 wccflag = NFSV3_WCCCHK;
1396 nfsm_wcc_data(vp, wccflag);
1398 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1399 + NFSX_V3WRITEVERF);
1400 rlen = fxdr_unsigned(int, *tl++);
1405 } else if (rlen < len) {
1406 backup = len - rlen;
1407 uiop->uio_iov->iov_base =
1408 (char *)uiop->uio_iov->iov_base -
1410 uiop->uio_iov->iov_len += backup;
1411 uiop->uio_offset -= backup;
1412 uiop->uio_resid += backup;
1415 commit = fxdr_unsigned(int, *tl++);
1418 * Return the lowest committment level
1419 * obtained by any of the RPCs.
1421 if (committed == NFSV3WRITE_FILESYNC)
1423 else if (committed == NFSV3WRITE_DATASYNC &&
1424 commit == NFSV3WRITE_UNSTABLE)
1426 mtx_lock(&nmp->nm_mtx);
1427 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1428 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1430 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1431 } else if (bcmp((caddr_t)tl,
1432 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1434 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1437 mtx_unlock(&nmp->nm_mtx);
1440 nfsm_loadattr(vp, NULL);
1443 mtx_lock(&(VTONFS(vp))->n_mtx);
1444 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1445 mtx_unlock(&(VTONFS(vp))->n_mtx);
1453 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1454 committed = NFSV3WRITE_FILESYNC;
1455 *iomode = committed;
1457 uiop->uio_resid = tsiz;
1463 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1464 * mode set to specify the file type and the size field for rdev.
1467 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1470 struct nfsv2_sattr *sp;
1472 struct vnode *newvp = NULL;
1473 struct nfsnode *np = NULL;
1476 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1477 struct mbuf *mreq, *mrep, *md, *mb;
1479 int v3 = NFS_ISV3(dvp);
1481 if (vap->va_type == VCHR || vap->va_type == VBLK)
1482 rdev = txdr_unsigned(vap->va_rdev);
1483 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1486 return (EOPNOTSUPP);
1488 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1490 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1491 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1492 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1494 bpos = mtod(mb, caddr_t);
1495 nfsm_fhtom(dvp, v3);
1496 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1498 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1499 *tl++ = vtonfsv3_type(vap->va_type);
1500 nfsm_v3attrbuild(vap, FALSE);
1501 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1502 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1503 *tl++ = txdr_unsigned(major(vap->va_rdev));
1504 *tl = txdr_unsigned(minor(vap->va_rdev));
1507 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1508 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1509 sp->sa_uid = nfs_xdrneg1;
1510 sp->sa_gid = nfs_xdrneg1;
1512 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1513 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1515 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1517 nfsm_mtofh(dvp, newvp, v3, gotvp);
1523 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1524 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1530 nfsm_wcc_data(dvp, wccflag);
1537 if (cnp->cn_flags & MAKEENTRY)
1538 cache_enter(dvp, newvp, cnp);
1541 mtx_lock(&(VTONFS(dvp))->n_mtx);
1542 VTONFS(dvp)->n_flag |= NMODIFIED;
1544 VTONFS(dvp)->n_attrstamp = 0;
1545 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1547 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1553 * just call nfs_mknodrpc() to do the work.
1557 nfs_mknod(struct vop_mknod_args *ap)
1559 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1562 static u_long create_verf;
1564 * nfs file create call
1567 nfs_create(struct vop_create_args *ap)
1569 struct vnode *dvp = ap->a_dvp;
1570 struct vattr *vap = ap->a_vap;
1571 struct componentname *cnp = ap->a_cnp;
1572 struct nfsv2_sattr *sp;
1574 struct nfsnode *np = NULL;
1575 struct vnode *newvp = NULL;
1577 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1578 struct mbuf *mreq, *mrep, *md, *mb;
1580 int v3 = NFS_ISV3(dvp);
1583 * Oops, not for me..
1585 if (vap->va_type == VSOCK) {
1586 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1590 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) {
1593 if (vap->va_vaflags & VA_EXCLUSIVE)
1596 nfsstats.rpccnt[NFSPROC_CREATE]++;
1597 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1598 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1600 bpos = mtod(mb, caddr_t);
1601 nfsm_fhtom(dvp, v3);
1602 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1604 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1605 if (fmode & O_EXCL) {
1606 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1607 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1609 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1611 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1612 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1615 *tl++ = create_verf;
1617 IN_IFADDR_RUNLOCK();
1620 *tl = ++create_verf;
1622 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1623 nfsm_v3attrbuild(vap, FALSE);
1626 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1627 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1628 sp->sa_uid = nfs_xdrneg1;
1629 sp->sa_gid = nfs_xdrneg1;
1631 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1632 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1634 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1636 nfsm_mtofh(dvp, newvp, v3, gotvp);
1642 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1643 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1649 nfsm_wcc_data(dvp, wccflag);
1653 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1659 } else if (v3 && (fmode & O_EXCL)) {
1661 * We are normally called with only a partially initialized
1662 * VAP. Since the NFSv3 spec says that server may use the
1663 * file attributes to store the verifier, the spec requires
1664 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1665 * in atime, but we can't really assume that all servers will
1666 * so we ensure that our SETATTR sets both atime and mtime.
1668 if (vap->va_mtime.tv_sec == VNOVAL)
1669 vfs_timestamp(&vap->va_mtime);
1670 if (vap->va_atime.tv_sec == VNOVAL)
1671 vap->va_atime = vap->va_mtime;
1672 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred);
1677 if (cnp->cn_flags & MAKEENTRY)
1678 cache_enter(dvp, newvp, cnp);
1681 mtx_lock(&(VTONFS(dvp))->n_mtx);
1682 VTONFS(dvp)->n_flag |= NMODIFIED;
1684 VTONFS(dvp)->n_attrstamp = 0;
1685 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1687 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1692 * nfs file remove call
1693 * To try and make nfs semantics closer to ufs semantics, a file that has
1694 * other processes using the vnode is renamed instead of removed and then
1695 * removed later on the last close.
1696 * - If v_usecount > 1
1697 * If a rename is not already in the works
1698 * call nfs_sillyrename() to set it up
1703 nfs_remove(struct vop_remove_args *ap)
1705 struct vnode *vp = ap->a_vp;
1706 struct vnode *dvp = ap->a_dvp;
1707 struct componentname *cnp = ap->a_cnp;
1708 struct nfsnode *np = VTONFS(vp);
1712 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1713 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1714 if (vp->v_type == VDIR)
1716 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1717 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1719 * Purge the name cache so that the chance of a lookup for
1720 * the name succeeding while the remove is in progress is
1721 * minimized. Without node locking it can still happen, such
1722 * that an I/O op returns ESTALE, but since you get this if
1723 * another host removes the file..
1727 * throw away biocache buffers, mainly to avoid
1728 * unnecessary delayed writes later.
1730 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1732 if (error != EINTR && error != EIO)
1733 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1734 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1736 * Kludge City: If the first reply to the remove rpc is lost..
1737 * the reply to the retransmitted request will be ENOENT
1738 * since the file was in fact removed
1739 * Therefore, we cheat and return success.
1741 if (error == ENOENT)
1743 } else if (!np->n_sillyrename)
1744 error = nfs_sillyrename(dvp, vp, cnp);
1745 mtx_lock(&np->n_mtx);
1746 np->n_attrstamp = 0;
1747 mtx_unlock(&np->n_mtx);
1748 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1753 * nfs file remove rpc called from nfs_inactive
1756 nfs_removeit(struct sillyrename *sp)
1759 * Make sure that the directory vnode is still valid.
1760 * XXX we should lock sp->s_dvp here.
1762 if (sp->s_dvp->v_type == VBAD)
1764 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1769 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1772 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1773 struct ucred *cred, struct thread *td)
1776 int error = 0, wccflag = NFSV3_WCCRATTR;
1777 struct mbuf *mreq, *mrep, *md, *mb;
1778 int v3 = NFS_ISV3(dvp);
1780 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1781 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1782 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1784 bpos = mtod(mb, caddr_t);
1785 nfsm_fhtom(dvp, v3);
1786 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1787 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1789 nfsm_wcc_data(dvp, wccflag);
1792 mtx_lock(&(VTONFS(dvp))->n_mtx);
1793 VTONFS(dvp)->n_flag |= NMODIFIED;
1795 VTONFS(dvp)->n_attrstamp = 0;
1796 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1798 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1803 * nfs file rename call
1806 nfs_rename(struct vop_rename_args *ap)
1808 struct vnode *fvp = ap->a_fvp;
1809 struct vnode *tvp = ap->a_tvp;
1810 struct vnode *fdvp = ap->a_fdvp;
1811 struct vnode *tdvp = ap->a_tdvp;
1812 struct componentname *tcnp = ap->a_tcnp;
1813 struct componentname *fcnp = ap->a_fcnp;
1816 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1817 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1818 /* Check for cross-device rename */
1819 if ((fvp->v_mount != tdvp->v_mount) ||
1820 (tvp && (fvp->v_mount != tvp->v_mount))) {
1826 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1830 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1834 * We have to flush B_DELWRI data prior to renaming
1835 * the file. If we don't, the delayed-write buffers
1836 * can be flushed out later after the file has gone stale
1837 * under NFSV3. NFSV2 does not have this problem because
1838 * ( as far as I can tell ) it flushes dirty buffers more
1841 * Skip the rename operation if the fsync fails, this can happen
1842 * due to the server's volume being full, when we pushed out data
1843 * that was written back to our cache earlier. Not checking for
1844 * this condition can result in potential (silent) data loss.
1846 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1849 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1854 * If the tvp exists and is in use, sillyrename it before doing the
1855 * rename of the new file over it.
1856 * XXX Can't sillyrename a directory.
1858 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1859 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1864 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1865 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1868 if (fvp->v_type == VDIR) {
1869 if (tvp != NULL && tvp->v_type == VDIR)
1884 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1886 if (error == ENOENT)
1892 * nfs file rename rpc called from nfs_remove() above
1895 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1896 struct sillyrename *sp)
1899 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1900 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1904 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1907 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1908 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1912 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1913 struct mbuf *mreq, *mrep, *md, *mb;
1914 int v3 = NFS_ISV3(fdvp);
1916 nfsstats.rpccnt[NFSPROC_RENAME]++;
1917 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1918 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1919 nfsm_rndup(tnamelen));
1921 bpos = mtod(mb, caddr_t);
1922 nfsm_fhtom(fdvp, v3);
1923 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1924 nfsm_fhtom(tdvp, v3);
1925 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1926 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1928 nfsm_wcc_data(fdvp, fwccflag);
1929 nfsm_wcc_data(tdvp, twccflag);
1933 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1934 VTONFS(fdvp)->n_flag |= NMODIFIED;
1935 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1936 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1937 VTONFS(tdvp)->n_flag |= NMODIFIED;
1938 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1940 VTONFS(fdvp)->n_attrstamp = 0;
1941 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1944 VTONFS(tdvp)->n_attrstamp = 0;
1945 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1951 * nfs hard link create call
1954 nfs_link(struct vop_link_args *ap)
1956 struct vnode *vp = ap->a_vp;
1957 struct vnode *tdvp = ap->a_tdvp;
1958 struct componentname *cnp = ap->a_cnp;
1960 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1961 struct mbuf *mreq, *mrep, *md, *mb;
1964 if (vp->v_mount != tdvp->v_mount) {
1969 * Push all writes to the server, so that the attribute cache
1970 * doesn't get "out of sync" with the server.
1971 * XXX There should be a better way!
1973 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1976 nfsstats.rpccnt[NFSPROC_LINK]++;
1977 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1978 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1980 bpos = mtod(mb, caddr_t);
1982 nfsm_fhtom(tdvp, v3);
1983 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1984 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1986 nfsm_postop_attr(vp, attrflag);
1987 nfsm_wcc_data(tdvp, wccflag);
1991 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1992 VTONFS(tdvp)->n_flag |= NMODIFIED;
1993 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1995 VTONFS(vp)->n_attrstamp = 0;
1996 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1999 VTONFS(tdvp)->n_attrstamp = 0;
2000 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2006 * nfs symbolic link create call
2009 nfs_symlink(struct vop_symlink_args *ap)
2011 struct vnode *dvp = ap->a_dvp;
2012 struct vattr *vap = ap->a_vap;
2013 struct componentname *cnp = ap->a_cnp;
2014 struct nfsv2_sattr *sp;
2016 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2017 struct mbuf *mreq, *mrep, *md, *mb;
2018 struct vnode *newvp = NULL;
2019 int v3 = NFS_ISV3(dvp);
2021 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2022 slen = strlen(ap->a_target);
2023 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2024 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2026 bpos = mtod(mb, caddr_t);
2027 nfsm_fhtom(dvp, v3);
2028 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2030 nfsm_v3attrbuild(vap, FALSE);
2032 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2034 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2035 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2036 sp->sa_uid = nfs_xdrneg1;
2037 sp->sa_gid = nfs_xdrneg1;
2038 sp->sa_size = nfs_xdrneg1;
2039 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2040 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2044 * Issue the NFS request and get the rpc response.
2046 * Only NFSv3 responses returning an error of 0 actually return
2047 * a file handle that can be converted into newvp without having
2048 * to do an extra lookup rpc.
2050 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
2053 nfsm_mtofh(dvp, newvp, v3, gotvp);
2054 nfsm_wcc_data(dvp, wccflag);
2058 * out code jumps -> here, mrep is also freed.
2065 * If we do not have an error and we could not extract the newvp from
2066 * the response due to the request being NFSv2, we have to do a
2067 * lookup in order to obtain a newvp to return.
2069 if (error == 0 && newvp == NULL) {
2070 struct nfsnode *np = NULL;
2072 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2073 cnp->cn_cred, cnp->cn_thread, &np);
2083 mtx_lock(&(VTONFS(dvp))->n_mtx);
2084 VTONFS(dvp)->n_flag |= NMODIFIED;
2085 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2087 VTONFS(dvp)->n_attrstamp = 0;
2088 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2097 nfs_mkdir(struct vop_mkdir_args *ap)
2099 struct vnode *dvp = ap->a_dvp;
2100 struct vattr *vap = ap->a_vap;
2101 struct componentname *cnp = ap->a_cnp;
2102 struct nfsv2_sattr *sp;
2104 struct nfsnode *np = NULL;
2105 struct vnode *newvp = NULL;
2107 int error = 0, wccflag = NFSV3_WCCRATTR;
2109 struct mbuf *mreq, *mrep, *md, *mb;
2111 int v3 = NFS_ISV3(dvp);
2113 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2115 len = cnp->cn_namelen;
2116 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2117 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
2118 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2120 bpos = mtod(mb, caddr_t);
2121 nfsm_fhtom(dvp, v3);
2122 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2124 nfsm_v3attrbuild(vap, FALSE);
2126 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2127 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2128 sp->sa_uid = nfs_xdrneg1;
2129 sp->sa_gid = nfs_xdrneg1;
2130 sp->sa_size = nfs_xdrneg1;
2131 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2132 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2134 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2136 nfsm_mtofh(dvp, newvp, v3, gotvp);
2138 nfsm_wcc_data(dvp, wccflag);
2141 mtx_lock(&(VTONFS(dvp))->n_mtx);
2142 VTONFS(dvp)->n_flag |= NMODIFIED;
2143 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2145 VTONFS(dvp)->n_attrstamp = 0;
2146 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2148 if (error == 0 && newvp == NULL) {
2149 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2150 cnp->cn_thread, &np);
2153 if (newvp->v_type != VDIR)
2166 * nfs remove directory call
2169 nfs_rmdir(struct vop_rmdir_args *ap)
2171 struct vnode *vp = ap->a_vp;
2172 struct vnode *dvp = ap->a_dvp;
2173 struct componentname *cnp = ap->a_cnp;
2175 int error = 0, wccflag = NFSV3_WCCRATTR;
2176 struct mbuf *mreq, *mrep, *md, *mb;
2177 int v3 = NFS_ISV3(dvp);
2181 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2182 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
2183 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2185 bpos = mtod(mb, caddr_t);
2186 nfsm_fhtom(dvp, v3);
2187 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2188 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2190 nfsm_wcc_data(dvp, wccflag);
2193 mtx_lock(&(VTONFS(dvp))->n_mtx);
2194 VTONFS(dvp)->n_flag |= NMODIFIED;
2195 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2197 VTONFS(dvp)->n_attrstamp = 0;
2198 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2203 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2205 if (error == ENOENT)
2214 nfs_readdir(struct vop_readdir_args *ap)
2216 struct vnode *vp = ap->a_vp;
2217 struct nfsnode *np = VTONFS(vp);
2218 struct uio *uio = ap->a_uio;
2219 int tresid, error = 0;
2222 if (vp->v_type != VDIR)
2226 * First, check for hit on the EOF offset cache
2228 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2229 (np->n_flag & NMODIFIED) == 0) {
2230 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2231 mtx_lock(&np->n_mtx);
2232 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2233 mtx_unlock(&np->n_mtx);
2234 nfsstats.direofcache_hits++;
2237 mtx_unlock(&np->n_mtx);
2242 * Call nfs_bioread() to do the real work.
2244 tresid = uio->uio_resid;
2245 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2247 if (!error && uio->uio_resid == tresid) {
2248 nfsstats.direofcache_misses++;
2256 * Called from below the buffer cache by nfs_doio().
2259 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2262 struct dirent *dp = NULL;
2267 struct mbuf *mreq, *mrep, *md, *mb;
2269 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2270 struct nfsnode *dnp = VTONFS(vp);
2272 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2274 int v3 = NFS_ISV3(vp);
2276 KASSERT(uiop->uio_iovcnt == 1 &&
2277 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2278 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2279 ("nfs readdirrpc bad uio"));
2282 * If there is no cookie, assume directory was stale.
2284 nfs_dircookie_lock(dnp);
2285 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2288 nfs_dircookie_unlock(dnp);
2290 nfs_dircookie_unlock(dnp);
2291 return (NFSERR_BAD_COOKIE);
2295 * Loop around doing readdir rpc's of size nm_readdirsize
2296 * truncated to a multiple of DIRBLKSIZ.
2297 * The stopping criteria is EOF or buffer full.
2299 while (more_dirs && bigenough) {
2300 nfsstats.rpccnt[NFSPROC_READDIR]++;
2301 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2304 bpos = mtod(mb, caddr_t);
2307 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2308 *tl++ = cookie.nfsuquad[0];
2309 *tl++ = cookie.nfsuquad[1];
2310 mtx_lock(&dnp->n_mtx);
2311 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2312 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2313 mtx_unlock(&dnp->n_mtx);
2315 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2316 *tl++ = cookie.nfsuquad[0];
2318 *tl = txdr_unsigned(nmp->nm_readdirsize);
2319 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2321 nfsm_postop_attr(vp, attrflag);
2323 tl = nfsm_dissect(u_int32_t *,
2325 mtx_lock(&dnp->n_mtx);
2326 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2327 dnp->n_cookieverf.nfsuquad[1] = *tl;
2328 mtx_unlock(&dnp->n_mtx);
2334 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2335 more_dirs = fxdr_unsigned(int, *tl);
2337 /* loop thru the dir entries, doctoring them to 4bsd form */
2338 while (more_dirs && bigenough) {
2340 tl = nfsm_dissect(u_int32_t *,
2342 fileno = fxdr_hyper(tl);
2343 len = fxdr_unsigned(int, *(tl + 2));
2345 tl = nfsm_dissect(u_int32_t *,
2347 fileno = fxdr_unsigned(u_quad_t, *tl++);
2348 len = fxdr_unsigned(int, *tl);
2350 if (len <= 0 || len > NFS_MAXNAMLEN) {
2355 tlen = nfsm_rndup(len);
2357 tlen += 4; /* To ensure null termination */
2358 left = DIRBLKSIZ - blksiz;
2359 if ((tlen + DIRHDSIZ) > left) {
2360 dp->d_reclen += left;
2361 uiop->uio_iov->iov_base =
2362 (char *)uiop->uio_iov->iov_base + left;
2363 uiop->uio_iov->iov_len -= left;
2364 uiop->uio_offset += left;
2365 uiop->uio_resid -= left;
2368 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2371 dp = (struct dirent *)uiop->uio_iov->iov_base;
2372 dp->d_fileno = (int)fileno;
2374 dp->d_reclen = tlen + DIRHDSIZ;
2375 dp->d_type = DT_UNKNOWN;
2376 blksiz += dp->d_reclen;
2377 if (blksiz == DIRBLKSIZ)
2379 uiop->uio_offset += DIRHDSIZ;
2380 uiop->uio_resid -= DIRHDSIZ;
2381 uiop->uio_iov->iov_base =
2382 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2383 uiop->uio_iov->iov_len -= DIRHDSIZ;
2384 nfsm_mtouio(uiop, len);
2385 cp = uiop->uio_iov->iov_base;
2387 *cp = '\0'; /* null terminate */
2388 uiop->uio_iov->iov_base =
2389 (char *)uiop->uio_iov->iov_base + tlen;
2390 uiop->uio_iov->iov_len -= tlen;
2391 uiop->uio_offset += tlen;
2392 uiop->uio_resid -= tlen;
2394 nfsm_adv(nfsm_rndup(len));
2396 tl = nfsm_dissect(u_int32_t *,
2399 tl = nfsm_dissect(u_int32_t *,
2403 cookie.nfsuquad[0] = *tl++;
2405 cookie.nfsuquad[1] = *tl++;
2410 more_dirs = fxdr_unsigned(int, *tl);
2413 * If at end of rpc data, get the eof boolean
2416 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2417 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2422 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2423 * by increasing d_reclen for the last record.
2426 left = DIRBLKSIZ - blksiz;
2427 dp->d_reclen += left;
2428 uiop->uio_iov->iov_base =
2429 (char *)uiop->uio_iov->iov_base + left;
2430 uiop->uio_iov->iov_len -= left;
2431 uiop->uio_offset += left;
2432 uiop->uio_resid -= left;
2436 * We are now either at the end of the directory or have filled the
2440 dnp->n_direofoffset = uiop->uio_offset;
2442 if (uiop->uio_resid > 0)
2443 nfs_printf("EEK! readdirrpc resid > 0\n");
2444 nfs_dircookie_lock(dnp);
2445 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2447 nfs_dircookie_unlock(dnp);
2454 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2457 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2463 struct vnode *newvp;
2465 caddr_t bpos, dpos, dpossav1, dpossav2;
2466 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2467 struct nameidata nami, *ndp = &nami;
2468 struct componentname *cnp = &ndp->ni_cnd;
2470 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2471 struct nfsnode *dnp = VTONFS(vp), *np;
2474 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2475 int attrflag, fhsize;
2480 KASSERT(uiop->uio_iovcnt == 1 &&
2481 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2482 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2483 ("nfs readdirplusrpc bad uio"));
2488 * If there is no cookie, assume directory was stale.
2490 nfs_dircookie_lock(dnp);
2491 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2494 nfs_dircookie_unlock(dnp);
2496 nfs_dircookie_unlock(dnp);
2497 return (NFSERR_BAD_COOKIE);
2500 * Loop around doing readdir rpc's of size nm_readdirsize
2501 * truncated to a multiple of DIRBLKSIZ.
2502 * The stopping criteria is EOF or buffer full.
2504 while (more_dirs && bigenough) {
2505 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2506 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2507 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2509 bpos = mtod(mb, caddr_t);
2511 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2512 *tl++ = cookie.nfsuquad[0];
2513 *tl++ = cookie.nfsuquad[1];
2514 mtx_lock(&dnp->n_mtx);
2515 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2516 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2517 mtx_unlock(&dnp->n_mtx);
2518 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2519 *tl = txdr_unsigned(nmp->nm_rsize);
2520 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2521 nfsm_postop_attr(vp, attrflag);
2526 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2527 mtx_lock(&dnp->n_mtx);
2528 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2529 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2530 mtx_unlock(&dnp->n_mtx);
2531 more_dirs = fxdr_unsigned(int, *tl);
2533 /* loop thru the dir entries, doctoring them to 4bsd form */
2534 while (more_dirs && bigenough) {
2535 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2536 fileno = fxdr_hyper(tl);
2537 len = fxdr_unsigned(int, *(tl + 2));
2538 if (len <= 0 || len > NFS_MAXNAMLEN) {
2543 tlen = nfsm_rndup(len);
2545 tlen += 4; /* To ensure null termination*/
2546 left = DIRBLKSIZ - blksiz;
2547 if ((tlen + DIRHDSIZ) > left) {
2548 dp->d_reclen += left;
2549 uiop->uio_iov->iov_base =
2550 (char *)uiop->uio_iov->iov_base + left;
2551 uiop->uio_iov->iov_len -= left;
2552 uiop->uio_offset += left;
2553 uiop->uio_resid -= left;
2556 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2559 dp = (struct dirent *)uiop->uio_iov->iov_base;
2560 dp->d_fileno = (int)fileno;
2562 dp->d_reclen = tlen + DIRHDSIZ;
2563 dp->d_type = DT_UNKNOWN;
2564 blksiz += dp->d_reclen;
2565 if (blksiz == DIRBLKSIZ)
2567 uiop->uio_offset += DIRHDSIZ;
2568 uiop->uio_resid -= DIRHDSIZ;
2569 uiop->uio_iov->iov_base =
2570 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2571 uiop->uio_iov->iov_len -= DIRHDSIZ;
2572 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2573 cnp->cn_namelen = len;
2574 nfsm_mtouio(uiop, len);
2575 cp = uiop->uio_iov->iov_base;
2578 uiop->uio_iov->iov_base =
2579 (char *)uiop->uio_iov->iov_base + tlen;
2580 uiop->uio_iov->iov_len -= tlen;
2581 uiop->uio_offset += tlen;
2582 uiop->uio_resid -= tlen;
2584 nfsm_adv(nfsm_rndup(len));
2585 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2587 cookie.nfsuquad[0] = *tl++;
2588 cookie.nfsuquad[1] = *tl++;
2593 * Since the attributes are before the file handle
2594 * (sigh), we must skip over the attributes and then
2595 * come back and get them.
2597 attrflag = fxdr_unsigned(int, *tl);
2601 nfsm_adv(NFSX_V3FATTR);
2602 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2603 doit = fxdr_unsigned(int, *tl);
2605 * Skip loading the attrs for "..". There's a
2606 * race between loading the attrs here and
2607 * lookups that look for the directory currently
2608 * being read (in the parent). We try to acquire
2609 * the exclusive lock on ".." here, owning the
2610 * lock on the directory being read. Lookup will
2611 * hold the lock on ".." and try to acquire the
2612 * lock on the directory being read.
2614 * There are other ways of fixing this, one would
2615 * be to do a trylock on the ".." vnode and skip
2616 * loading the attrs on ".." if it happens to be
2617 * locked by another process. But skipping the
2618 * attrload on ".." seems the easiest option.
2620 if (strcmp(dp->d_name, "..") == 0) {
2623 * We've already skipped over the attrs,
2624 * skip over the filehandle. And store d_type
2627 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2628 i = fxdr_unsigned(int, *tl);
2629 nfsm_adv(nfsm_rndup(i));
2630 dp->d_type = IFTODT(VTTOIF(VDIR));
2633 nfsm_getfh(fhp, fhsize, 1);
2634 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2639 error = nfs_nget(vp->v_mount, fhp,
2640 fhsize, &np, LK_EXCLUSIVE);
2647 if (doit && bigenough) {
2652 nfsm_loadattr(newvp, NULL);
2656 IFTODT(VTTOIF(np->n_vattr.va_type));
2659 * Update n_ctime so subsequent lookup
2660 * doesn't purge entry.
2662 np->n_ctime = np->n_vattr.va_ctime;
2663 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2666 /* Just skip over the file handle */
2667 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2668 i = fxdr_unsigned(int, *tl);
2670 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2671 fhsize = fxdr_unsigned(int, *tl);
2672 nfsm_adv(nfsm_rndup(fhsize));
2675 if (newvp != NULLVP) {
2682 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2683 more_dirs = fxdr_unsigned(int, *tl);
2686 * If at end of rpc data, get the eof boolean
2689 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2690 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2695 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2696 * by increasing d_reclen for the last record.
2699 left = DIRBLKSIZ - blksiz;
2700 dp->d_reclen += left;
2701 uiop->uio_iov->iov_base =
2702 (char *)uiop->uio_iov->iov_base + left;
2703 uiop->uio_iov->iov_len -= left;
2704 uiop->uio_offset += left;
2705 uiop->uio_resid -= left;
2709 * We are now either at the end of the directory or have filled the
2713 dnp->n_direofoffset = uiop->uio_offset;
2715 if (uiop->uio_resid > 0)
2716 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2717 nfs_dircookie_lock(dnp);
2718 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2720 nfs_dircookie_unlock(dnp);
2723 if (newvp != NULLVP) {
2734 * Silly rename. To make the NFS filesystem that is stateless look a little
2735 * more like the "ufs" a remove of an active vnode is translated to a rename
2736 * to a funny looking filename that is removed by nfs_inactive on the
2737 * nfsnode. There is the potential for another process on a different client
2738 * to create the same funny name between the nfs_lookitup() fails and the
2739 * nfs_rename() completes, but...
2742 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2744 struct sillyrename *sp;
2748 unsigned int lticks;
2752 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2753 sp = malloc(sizeof (struct sillyrename),
2754 M_NFSREQ, M_WAITOK);
2755 sp->s_cred = crhold(cnp->cn_cred);
2757 sp->s_removeit = nfs_removeit;
2761 * Fudge together a funny name.
2762 * Changing the format of the funny name to accomodate more
2763 * sillynames per directory.
2764 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2765 * CPU ticks since boot.
2767 pid = cnp->cn_thread->td_proc->p_pid;
2768 lticks = (unsigned int)ticks;
2770 sp->s_namlen = sprintf(sp->s_name,
2771 ".nfs.%08x.%04x4.4", lticks,
2773 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2774 cnp->cn_thread, NULL))
2778 error = nfs_renameit(dvp, cnp, sp);
2781 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2782 cnp->cn_thread, &np);
2783 np->n_sillyrename = sp;
2788 free((caddr_t)sp, M_NFSREQ);
2793 * Look up a file name and optionally either update the file handle or
2794 * allocate an nfsnode, depending on the value of npp.
2795 * npp == NULL --> just do the lookup
2796 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2798 * *npp != NULL --> update the file handle in the vnode
2801 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2802 struct thread *td, struct nfsnode **npp)
2804 struct vnode *newvp = NULL;
2805 struct nfsnode *np, *dnp = VTONFS(dvp);
2807 int error = 0, fhlen, attrflag;
2808 struct mbuf *mreq, *mrep, *md, *mb;
2810 int v3 = NFS_ISV3(dvp);
2812 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2813 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2814 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2816 bpos = mtod(mb, caddr_t);
2817 nfsm_fhtom(dvp, v3);
2818 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2819 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2820 if (npp && !error) {
2821 nfsm_getfh(nfhp, fhlen, v3);
2824 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2825 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2826 np->n_fhp = &np->n_fh;
2827 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2828 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2829 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2830 np->n_fhsize = fhlen;
2832 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2836 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2844 nfsm_postop_attr(newvp, attrflag);
2845 if (!attrflag && *npp == NULL) {
2854 nfsm_loadattr(newvp, NULL);
2858 if (npp && *npp == NULL) {
2873 * Nfs Version 3 commit rpc
2876 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2880 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2882 int error = 0, wccflag = NFSV3_WCCRATTR;
2883 struct mbuf *mreq, *mrep, *md, *mb;
2885 mtx_lock(&nmp->nm_mtx);
2886 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2887 mtx_unlock(&nmp->nm_mtx);
2890 mtx_unlock(&nmp->nm_mtx);
2891 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2892 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2894 bpos = mtod(mb, caddr_t);
2896 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2897 txdr_hyper(offset, tl);
2899 *tl = txdr_unsigned(cnt);
2900 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2901 nfsm_wcc_data(vp, wccflag);
2903 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2904 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2905 NFSX_V3WRITEVERF)) {
2906 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2908 error = NFSERR_STALEWRITEVERF;
2918 * For async requests when nfsiod(s) are running, queue the request by
2919 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2923 nfs_strategy(struct vop_strategy_args *ap)
2925 struct buf *bp = ap->a_bp;
2928 KASSERT(!(bp->b_flags & B_DONE),
2929 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2930 BUF_ASSERT_HELD(bp);
2932 if (bp->b_iocmd == BIO_READ)
2938 * If the op is asynchronous and an i/o daemon is waiting
2939 * queue the request, wake it up and wait for completion
2940 * otherwise just do it ourselves.
2942 if ((bp->b_flags & B_ASYNC) == 0 ||
2943 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2944 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2949 * fsync vnode op. Just call nfs_flush() with commit == 1.
2953 nfs_fsync(struct vop_fsync_args *ap)
2956 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1));
2960 * Flush all the blocks associated with a vnode.
2961 * Walk through the buffer pool and push any dirty pages
2962 * associated with the vnode.
2965 nfs_flush(struct vnode *vp, int waitfor, int commit)
2967 struct nfsnode *np = VTONFS(vp);
2971 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2972 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2974 u_quad_t off, endoff, toff;
2975 struct ucred* wcred = NULL;
2976 struct buf **bvec = NULL;
2978 struct thread *td = curthread;
2979 #ifndef NFS_COMMITBVECSIZ
2980 #define NFS_COMMITBVECSIZ 20
2982 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2983 int bvecsize = 0, bveccount;
2985 if (nmp->nm_flag & NFSMNT_INT)
2986 slpflag = NFS_PCATCH;
2991 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2992 * server, but has not been committed to stable storage on the server
2993 * yet. On the first pass, the byte range is worked out and the commit
2994 * rpc is done. On the second pass, nfs_writebp() is called to do the
3001 if (NFS_ISV3(vp) && commit) {
3002 if (bvec != NULL && bvec != bvec_on_stack)
3005 * Count up how many buffers waiting for a commit.
3009 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3010 if (!BUF_ISLOCKED(bp) &&
3011 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
3012 == (B_DELWRI | B_NEEDCOMMIT))
3016 * Allocate space to remember the list of bufs to commit. It is
3017 * important to use M_NOWAIT here to avoid a race with nfs_write.
3018 * If we can't get memory (for whatever reason), we will end up
3019 * committing the buffers one-by-one in the loop below.
3021 if (bveccount > NFS_COMMITBVECSIZ) {
3023 * Release the vnode interlock to avoid a lock
3027 bvec = (struct buf **)
3028 malloc(bveccount * sizeof(struct buf *),
3032 bvec = bvec_on_stack;
3033 bvecsize = NFS_COMMITBVECSIZ;
3035 bvecsize = bveccount;
3037 bvec = bvec_on_stack;
3038 bvecsize = NFS_COMMITBVECSIZ;
3040 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3041 if (bvecpos >= bvecsize)
3043 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3044 nbp = TAILQ_NEXT(bp, b_bobufs);
3047 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3048 (B_DELWRI | B_NEEDCOMMIT)) {
3050 nbp = TAILQ_NEXT(bp, b_bobufs);
3056 * Work out if all buffers are using the same cred
3057 * so we can deal with them all with one commit.
3059 * NOTE: we are not clearing B_DONE here, so we have
3060 * to do it later on in this routine if we intend to
3061 * initiate I/O on the bp.
3063 * Note: to avoid loopback deadlocks, we do not
3064 * assign b_runningbufspace.
3067 wcred = bp->b_wcred;
3068 else if (wcred != bp->b_wcred)
3070 vfs_busy_pages(bp, 1);
3074 * bp is protected by being locked, but nbp is not
3075 * and vfs_busy_pages() may sleep. We have to
3078 nbp = TAILQ_NEXT(bp, b_bobufs);
3081 * A list of these buffers is kept so that the
3082 * second loop knows which buffers have actually
3083 * been committed. This is necessary, since there
3084 * may be a race between the commit rpc and new
3085 * uncommitted writes on the file.
3087 bvec[bvecpos++] = bp;
3088 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3092 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3100 * Commit data on the server, as required.
3101 * If all bufs are using the same wcred, then use that with
3102 * one call for all of them, otherwise commit each one
3105 if (wcred != NOCRED)
3106 retv = nfs_commit(vp, off, (int)(endoff - off),
3110 for (i = 0; i < bvecpos; i++) {
3113 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3115 size = (u_quad_t)(bp->b_dirtyend
3117 retv = nfs_commit(vp, off, (int)size,
3123 if (retv == NFSERR_STALEWRITEVERF)
3124 nfs_clearcommit(vp->v_mount);
3127 * Now, either mark the blocks I/O done or mark the
3128 * blocks dirty, depending on whether the commit
3131 for (i = 0; i < bvecpos; i++) {
3133 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3136 * Error, leave B_DELWRI intact
3138 vfs_unbusy_pages(bp);
3142 * Success, remove B_DELWRI ( bundirty() ).
3144 * b_dirtyoff/b_dirtyend seem to be NFS
3145 * specific. We should probably move that
3146 * into bundirty(). XXX
3149 bp->b_flags |= B_ASYNC;
3151 bp->b_flags &= ~B_DONE;
3152 bp->b_ioflags &= ~BIO_ERROR;
3153 bp->b_dirtyoff = bp->b_dirtyend = 0;
3160 * Start/do any write(s) that are required.
3164 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3165 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3166 if (waitfor != MNT_WAIT || passone)
3169 error = BUF_TIMELOCK(bp,
3170 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3171 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
3176 if (error == ENOLCK) {
3180 if (nfs_sigintr(nmp, td)) {
3184 if (slpflag & PCATCH) {
3190 if ((bp->b_flags & B_DELWRI) == 0)
3191 panic("nfs_fsync: not dirty");
3192 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3198 if (passone || !commit)
3199 bp->b_flags |= B_ASYNC;
3201 bp->b_flags |= B_ASYNC;
3203 if (nfs_sigintr(nmp, td)) {
3214 if (waitfor == MNT_WAIT) {
3215 while (bo->bo_numoutput) {
3216 error = bufobj_wwait(bo, slpflag, slptimeo);
3219 error = nfs_sigintr(nmp, td);
3222 if (slpflag & PCATCH) {
3229 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3234 * Wait for all the async IO requests to drain
3237 mtx_lock(&np->n_mtx);
3238 while (np->n_directio_asyncwr > 0) {
3239 np->n_flag |= NFSYNCWAIT;
3240 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3241 &np->n_mtx, slpflag | (PRIBIO + 1),
3244 if (nfs_sigintr(nmp, td)) {
3245 mtx_unlock(&np->n_mtx);
3251 mtx_unlock(&np->n_mtx);
3254 mtx_lock(&np->n_mtx);
3255 if (np->n_flag & NWRITEERR) {
3256 error = np->n_error;
3257 np->n_flag &= ~NWRITEERR;
3259 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3260 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3261 np->n_flag &= ~NMODIFIED;
3262 mtx_unlock(&np->n_mtx);
3264 if (bvec != NULL && bvec != bvec_on_stack)
3270 * NFS advisory byte-level locks.
3273 nfs_advlock(struct vop_advlock_args *ap)
3275 struct vnode *vp = ap->a_vp;
3279 error = vn_lock(vp, LK_SHARED);
3282 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3283 size = VTONFS(vp)->n_size;
3285 error = lf_advlock(ap, &(vp->v_lockf), size);
3288 error = nfs_advlock_p(ap);
3297 * NFS advisory byte-level locks.
3300 nfs_advlockasync(struct vop_advlockasync_args *ap)
3302 struct vnode *vp = ap->a_vp;
3306 error = vn_lock(vp, LK_SHARED);
3309 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3310 size = VTONFS(vp)->n_size;
3312 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3321 * Print out the contents of an nfsnode.
3324 nfs_print(struct vop_print_args *ap)
3326 struct vnode *vp = ap->a_vp;
3327 struct nfsnode *np = VTONFS(vp);
3329 nfs_printf("\tfileid %ld fsid 0x%x",
3330 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3331 if (vp->v_type == VFIFO)
3338 * This is the "real" nfs::bwrite(struct buf*).
3339 * We set B_CACHE if this is a VMIO buffer.
3342 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3345 int oldflags = bp->b_flags;
3351 BUF_ASSERT_HELD(bp);
3353 if (bp->b_flags & B_INVAL) {
3358 bp->b_flags |= B_CACHE;
3361 * Undirty the bp. We will redirty it later if the I/O fails.
3366 bp->b_flags &= ~B_DONE;
3367 bp->b_ioflags &= ~BIO_ERROR;
3368 bp->b_iocmd = BIO_WRITE;
3370 bufobj_wref(bp->b_bufobj);
3371 curthread->td_ru.ru_oublock++;
3375 * Note: to avoid loopback deadlocks, we do not
3376 * assign b_runningbufspace.
3378 vfs_busy_pages(bp, 1);
3381 bp->b_iooffset = dbtob(bp->b_blkno);
3384 if( (oldflags & B_ASYNC) == 0) {
3385 int rtval = bufwait(bp);
3387 if (oldflags & B_DELWRI) {
3400 * nfs special file access vnode op.
3401 * Essentially just get vattr and then imitate iaccess() since the device is
3402 * local to the client.
3405 nfsspec_access(struct vop_access_args *ap)
3408 struct ucred *cred = ap->a_cred;
3409 struct vnode *vp = ap->a_vp;
3410 accmode_t accmode = ap->a_accmode;
3415 * Disallow write attempts on filesystems mounted read-only;
3416 * unless the file is a socket, fifo, or a block or character
3417 * device resident on the filesystem.
3419 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3420 switch (vp->v_type) {
3430 error = VOP_GETATTR(vp, vap, cred);
3433 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3434 accmode, cred, NULL);
3440 * Read wrapper for fifos.
3443 nfsfifo_read(struct vop_read_args *ap)
3445 struct nfsnode *np = VTONFS(ap->a_vp);
3451 mtx_lock(&np->n_mtx);
3453 getnanotime(&np->n_atim);
3454 mtx_unlock(&np->n_mtx);
3455 error = fifo_specops.vop_read(ap);
3460 * Write wrapper for fifos.
3463 nfsfifo_write(struct vop_write_args *ap)
3465 struct nfsnode *np = VTONFS(ap->a_vp);
3470 mtx_lock(&np->n_mtx);
3472 getnanotime(&np->n_mtim);
3473 mtx_unlock(&np->n_mtx);
3474 return(fifo_specops.vop_write(ap));
3478 * Close wrapper for fifos.
3480 * Update the times on the nfsnode then do fifo close.
3483 nfsfifo_close(struct vop_close_args *ap)
3485 struct vnode *vp = ap->a_vp;
3486 struct nfsnode *np = VTONFS(vp);
3490 mtx_lock(&np->n_mtx);
3491 if (np->n_flag & (NACC | NUPD)) {
3493 if (np->n_flag & NACC)
3495 if (np->n_flag & NUPD)
3498 if (vrefcnt(vp) == 1 &&
3499 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3501 if (np->n_flag & NACC)
3502 vattr.va_atime = np->n_atim;
3503 if (np->n_flag & NUPD)
3504 vattr.va_mtime = np->n_mtim;
3505 mtx_unlock(&np->n_mtx);
3506 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3510 mtx_unlock(&np->n_mtx);
3512 return (fifo_specops.vop_close(ap));
3516 * Just call nfs_writebp() with the force argument set to 1.
3518 * NOTE: B_DONE may or may not be set in a_bp on call.
3521 nfs_bwrite(struct buf *bp)
3524 return (nfs_writebp(bp, 1, curthread));
3527 struct buf_ops buf_ops_nfs = {
3528 .bop_name = "buf_ops_nfs",
3529 .bop_write = nfs_bwrite,
3530 .bop_strategy = bufstrategy,
3531 .bop_sync = bufsync,
3532 .bop_bdflush = bufbdflush,