2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
39 * vnode op calls for Sun NFS version 2 and 3
43 #include "opt_kdtrace.h"
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/systm.h>
48 #include <sys/resourcevar.h>
50 #include <sys/mount.h>
54 #include <sys/malloc.h>
56 #include <sys/namei.h>
57 #include <sys/socket.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/fcntl.h>
61 #include <sys/lockf.h>
63 #include <sys/sysctl.h>
64 #include <sys/signalvar.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
70 #include <fs/fifofs/fifo.h>
72 #include <nfs/nfsproto.h>
73 #include <nfsclient/nfs.h>
74 #include <nfsclient/nfsnode.h>
75 #include <nfsclient/nfsmount.h>
76 #include <nfs/nfs_kdtrace.h>
77 #include <nfs/nfs_lock.h>
78 #include <nfs/xdr_subs.h>
79 #include <nfsclient/nfsm_subs.h>
82 #include <netinet/in.h>
83 #include <netinet/in_var.h>
85 #include <machine/stdarg.h>
88 #include <sys/dtrace_bsd.h>
90 dtrace_nfsclient_accesscache_flush_probe_func_t
91 dtrace_nfsclient_accesscache_flush_done_probe;
92 uint32_t nfsclient_accesscache_flush_done_id;
94 dtrace_nfsclient_accesscache_get_probe_func_t
95 dtrace_nfsclient_accesscache_get_hit_probe,
96 dtrace_nfsclient_accesscache_get_miss_probe;
97 uint32_t nfsclient_accesscache_get_hit_id;
98 uint32_t nfsclient_accesscache_get_miss_id;
100 dtrace_nfsclient_accesscache_load_probe_func_t
101 dtrace_nfsclient_accesscache_load_done_probe;
102 uint32_t nfsclient_accesscache_load_done_id;
103 #endif /* !KDTRACE_HOOKS */
110 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
111 * calls are not in getblk() and brelse() so that they would not be necessary
115 #define vfs_busy_pages(bp, f)
118 static vop_read_t nfsfifo_read;
119 static vop_write_t nfsfifo_write;
120 static vop_close_t nfsfifo_close;
121 static int nfs_flush(struct vnode *, int, int);
122 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *);
123 static vop_lookup_t nfs_lookup;
124 static vop_create_t nfs_create;
125 static vop_mknod_t nfs_mknod;
126 static vop_open_t nfs_open;
127 static vop_close_t nfs_close;
128 static vop_access_t nfs_access;
129 static vop_getattr_t nfs_getattr;
130 static vop_setattr_t nfs_setattr;
131 static vop_read_t nfs_read;
132 static vop_fsync_t nfs_fsync;
133 static vop_remove_t nfs_remove;
134 static vop_link_t nfs_link;
135 static vop_rename_t nfs_rename;
136 static vop_mkdir_t nfs_mkdir;
137 static vop_rmdir_t nfs_rmdir;
138 static vop_symlink_t nfs_symlink;
139 static vop_readdir_t nfs_readdir;
140 static vop_strategy_t nfs_strategy;
141 static int nfs_lookitup(struct vnode *, const char *, int,
142 struct ucred *, struct thread *, struct nfsnode **);
143 static int nfs_sillyrename(struct vnode *, struct vnode *,
144 struct componentname *);
145 static vop_access_t nfsspec_access;
146 static vop_readlink_t nfs_readlink;
147 static vop_print_t nfs_print;
148 static vop_advlock_t nfs_advlock;
149 static vop_advlockasync_t nfs_advlockasync;
152 * Global vfs data structures for nfs
154 struct vop_vector nfs_vnodeops = {
155 .vop_default = &default_vnodeops,
156 .vop_access = nfs_access,
157 .vop_advlock = nfs_advlock,
158 .vop_advlockasync = nfs_advlockasync,
159 .vop_close = nfs_close,
160 .vop_create = nfs_create,
161 .vop_fsync = nfs_fsync,
162 .vop_getattr = nfs_getattr,
163 .vop_getpages = nfs_getpages,
164 .vop_putpages = nfs_putpages,
165 .vop_inactive = nfs_inactive,
166 .vop_link = nfs_link,
167 .vop_lookup = nfs_lookup,
168 .vop_mkdir = nfs_mkdir,
169 .vop_mknod = nfs_mknod,
170 .vop_open = nfs_open,
171 .vop_print = nfs_print,
172 .vop_read = nfs_read,
173 .vop_readdir = nfs_readdir,
174 .vop_readlink = nfs_readlink,
175 .vop_reclaim = nfs_reclaim,
176 .vop_remove = nfs_remove,
177 .vop_rename = nfs_rename,
178 .vop_rmdir = nfs_rmdir,
179 .vop_setattr = nfs_setattr,
180 .vop_strategy = nfs_strategy,
181 .vop_symlink = nfs_symlink,
182 .vop_write = nfs_write,
185 struct vop_vector nfs_fifoops = {
186 .vop_default = &fifo_specops,
187 .vop_access = nfsspec_access,
188 .vop_close = nfsfifo_close,
189 .vop_fsync = nfs_fsync,
190 .vop_getattr = nfs_getattr,
191 .vop_inactive = nfs_inactive,
192 .vop_print = nfs_print,
193 .vop_read = nfsfifo_read,
194 .vop_reclaim = nfs_reclaim,
195 .vop_setattr = nfs_setattr,
196 .vop_write = nfsfifo_write,
199 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
200 struct componentname *cnp, struct vattr *vap);
201 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
202 struct ucred *cred, struct thread *td);
203 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
204 int fnamelen, struct vnode *tdvp,
205 const char *tnameptr, int tnamelen,
206 struct ucred *cred, struct thread *td);
207 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
208 struct sillyrename *sp);
213 struct mtx nfs_iod_mtx;
214 enum nfsiod_state nfs_iodwant[NFS_MAXASYNCDAEMON];
215 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
216 int nfs_numasync = 0;
217 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
219 SYSCTL_DECL(_vfs_oldnfs);
221 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
222 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
223 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
225 static int nfs_prime_access_cache = 0;
226 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
227 &nfs_prime_access_cache, 0,
228 "Prime NFS ACCESS cache when fetching attributes");
230 static int nfsv3_commit_on_close = 0;
231 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
232 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
234 static int nfs_clean_pages_on_close = 1;
235 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
236 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
238 int nfs_directio_enable = 0;
239 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
240 &nfs_directio_enable, 0, "Enable NFS directio");
243 * This sysctl allows other processes to mmap a file that has been opened
244 * O_DIRECT by a process. In general, having processes mmap the file while
245 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
246 * this by default to prevent DoS attacks - to prevent a malicious user from
247 * opening up files O_DIRECT preventing other users from mmap'ing these
248 * files. "Protected" environments where stricter consistency guarantees are
249 * required can disable this knob. The process that opened the file O_DIRECT
250 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
253 int nfs_directio_allow_mmap = 1;
254 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
255 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
258 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
259 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
261 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
262 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
265 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
266 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
267 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
271 * The list of locks after the description of the lock is the ordering
272 * of other locks acquired with the lock held.
273 * np->n_mtx : Protects the fields in the nfsnode.
275 VI_MTX (acquired indirectly)
276 * nmp->nm_mtx : Protects the fields in the nfsmount.
278 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
279 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
282 * rep->r_mtx : Protects the fields in an nfsreq.
286 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
287 struct ucred *cred, uint32_t *retmode)
291 int error = 0, attrflag, i, lrupos;
293 struct mbuf *mreq, *mrep, *md, *mb;
296 struct nfsnode *np = VTONFS(vp);
298 nfsstats.rpccnt[NFSPROC_ACCESS]++;
299 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
301 bpos = mtod(mb, caddr_t);
303 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
304 *tl = txdr_unsigned(wmode);
305 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
306 nfsm_postop_attr(vp, attrflag);
309 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
310 rmode = fxdr_unsigned(u_int32_t, *tl);
311 mtx_lock(&np->n_mtx);
312 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
313 if (np->n_accesscache[i].uid == cred->cr_uid) {
314 np->n_accesscache[i].mode = rmode;
315 np->n_accesscache[i].stamp = time_second;
318 if (i > 0 && np->n_accesscache[i].stamp <
319 np->n_accesscache[lrupos].stamp)
322 if (i == NFS_ACCESSCACHESIZE) {
323 np->n_accesscache[lrupos].uid = cred->cr_uid;
324 np->n_accesscache[lrupos].mode = rmode;
325 np->n_accesscache[lrupos].stamp = time_second;
327 mtx_unlock(&np->n_mtx);
330 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
336 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
344 * nfs access vnode op.
345 * For nfs version 2, just return ok. File accesses may fail later.
346 * For nfs version 3, use the access rpc to check accessibility. If file modes
347 * are changed on the server, accesses might still fail later.
350 nfs_access(struct vop_access_args *ap)
352 struct vnode *vp = ap->a_vp;
353 int error = 0, i, gotahit;
354 u_int32_t mode, rmode, wmode;
355 int v3 = NFS_ISV3(vp);
356 struct nfsnode *np = VTONFS(vp);
359 * Disallow write attempts on filesystems mounted read-only;
360 * unless the file is a socket, fifo, or a block or character
361 * device resident on the filesystem.
363 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
364 switch (vp->v_type) {
374 * For nfs v3, check to see if we have done this recently, and if
375 * so return our cached result instead of making an ACCESS call.
376 * If not, do an access rpc, otherwise you are stuck emulating
377 * ufs_access() locally using the vattr. This may not be correct,
378 * since the server may apply other access criteria such as
379 * client uid-->server uid mapping that we do not know about.
382 if (ap->a_accmode & VREAD)
383 mode = NFSV3ACCESS_READ;
386 if (vp->v_type != VDIR) {
387 if (ap->a_accmode & VWRITE)
388 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
389 if (ap->a_accmode & VEXEC)
390 mode |= NFSV3ACCESS_EXECUTE;
392 if (ap->a_accmode & VWRITE)
393 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
395 if (ap->a_accmode & VEXEC)
396 mode |= NFSV3ACCESS_LOOKUP;
398 /* XXX safety belt, only make blanket request if caching */
399 if (nfsaccess_cache_timeout > 0) {
400 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
401 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
402 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
408 * Does our cached result allow us to give a definite yes to
412 mtx_lock(&np->n_mtx);
413 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
414 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
415 if (time_second < (np->n_accesscache[i].stamp +
416 nfsaccess_cache_timeout) &&
417 (np->n_accesscache[i].mode & mode) == mode) {
418 nfsstats.accesscache_hits++;
424 mtx_unlock(&np->n_mtx);
427 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
428 ap->a_cred->cr_uid, mode);
430 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
431 ap->a_cred->cr_uid, mode);
435 * Either a no, or a don't know. Go to the wire.
437 nfsstats.accesscache_misses++;
438 error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred,
441 if ((rmode & mode) != mode)
447 if ((error = nfsspec_access(ap)) != 0) {
451 * Attempt to prevent a mapped root from accessing a file
452 * which it shouldn't. We try to read a byte from the file
453 * if the user is root and the file is not zero length.
454 * After calling nfsspec_access, we should have the correct
457 mtx_lock(&np->n_mtx);
458 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
459 && VTONFS(vp)->n_size > 0) {
464 mtx_unlock(&np->n_mtx);
467 auio.uio_iov = &aiov;
471 auio.uio_segflg = UIO_SYSSPACE;
472 auio.uio_rw = UIO_READ;
473 auio.uio_td = ap->a_td;
475 if (vp->v_type == VREG)
476 error = nfs_readrpc(vp, &auio, ap->a_cred);
477 else if (vp->v_type == VDIR) {
479 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
481 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
482 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
484 } else if (vp->v_type == VLNK)
485 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
489 mtx_unlock(&np->n_mtx);
494 int nfs_otw_getattr_avoid = 0;
498 * Check to see if the type is ok
499 * and that deletion is not in progress.
500 * For paged in text files, you will need to flush the page cache
501 * if consistency is lost.
505 nfs_open(struct vop_open_args *ap)
507 struct vnode *vp = ap->a_vp;
508 struct nfsnode *np = VTONFS(vp);
511 int fmode = ap->a_mode;
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++;
571 * If this is an open for writing, capture a reference to the
572 * credentials, so they can be used by nfs_putpages(). Using
573 * these write credentials is preferable to the credentials of
574 * whatever thread happens to be doing the VOP_PUTPAGES() since
575 * the write RPCs are less likely to fail with EACCES.
577 if ((fmode & FWRITE) != 0) {
578 cred = np->n_writecred;
579 np->n_writecred = crhold(ap->a_cred);
582 mtx_unlock(&np->n_mtx);
585 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
591 * What an NFS client should do upon close after writing is a debatable issue.
592 * Most NFS clients push delayed writes to the server upon close, basically for
594 * 1 - So that any write errors may be reported back to the client process
595 * doing the close system call. By far the two most likely errors are
596 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
597 * 2 - To put a worst case upper bound on cache inconsistency between
598 * multiple clients for the file.
599 * There is also a consistency problem for Version 2 of the protocol w.r.t.
600 * not being able to tell if other clients are writing a file concurrently,
601 * since there is no way of knowing if the changed modify time in the reply
602 * is only due to the write for this client.
603 * (NFS Version 3 provides weak cache consistency data in the reply that
604 * should be sufficient to detect and handle this case.)
606 * The current code does the following:
607 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
608 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
609 * or commit them (this satisfies 1 and 2 except for the
610 * case where the server crashes after this close but
611 * before the commit RPC, which is felt to be "good
612 * enough". Changing the last argument to nfs_flush() to
613 * a 1 would force a commit operation, if it is felt a
614 * commit is necessary now.
618 nfs_close(struct vop_close_args *ap)
620 struct vnode *vp = ap->a_vp;
621 struct nfsnode *np = VTONFS(vp);
623 int fmode = ap->a_fflag;
625 if (vp->v_type == VREG) {
627 * Examine and clean dirty pages, regardless of NMODIFIED.
628 * This closes a major hole in close-to-open consistency.
629 * We want to push out all dirty pages (and buffers) on
630 * close, regardless of whether they were dirtied by
631 * mmap'ed writes or via write().
633 if (nfs_clean_pages_on_close && vp->v_object) {
634 VM_OBJECT_LOCK(vp->v_object);
635 vm_object_page_clean(vp->v_object, 0, 0, 0);
636 VM_OBJECT_UNLOCK(vp->v_object);
638 mtx_lock(&np->n_mtx);
639 if (np->n_flag & NMODIFIED) {
640 mtx_unlock(&np->n_mtx);
643 * Under NFSv3 we have dirty buffers to dispose of. We
644 * must flush them to the NFS server. We have the option
645 * of waiting all the way through the commit rpc or just
646 * waiting for the initial write. The default is to only
647 * wait through the initial write so the data is in the
648 * server's cache, which is roughly similar to the state
649 * a standard disk subsystem leaves the file in on close().
651 * We cannot clear the NMODIFIED bit in np->n_flag due to
652 * potential races with other processes, and certainly
653 * cannot clear it if we don't commit.
655 int cm = nfsv3_commit_on_close ? 1 : 0;
656 error = nfs_flush(vp, MNT_WAIT, cm);
657 /* np->n_flag &= ~NMODIFIED; */
659 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
660 mtx_lock(&np->n_mtx);
662 if (np->n_flag & NWRITEERR) {
663 np->n_flag &= ~NWRITEERR;
666 mtx_unlock(&np->n_mtx);
668 if (nfs_directio_enable)
669 KASSERT((np->n_directio_asyncwr == 0),
670 ("nfs_close: dirty unflushed (%d) directio buffers\n",
671 np->n_directio_asyncwr));
672 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
673 mtx_lock(&np->n_mtx);
674 KASSERT((np->n_directio_opens > 0),
675 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
676 np->n_directio_opens--;
677 if (np->n_directio_opens == 0)
678 np->n_flag &= ~NNONCACHE;
679 mtx_unlock(&np->n_mtx);
685 * nfs getattr call from vfs.
688 nfs_getattr(struct vop_getattr_args *ap)
690 struct vnode *vp = ap->a_vp;
691 struct nfsnode *np = VTONFS(vp);
692 struct thread *td = curthread;
693 struct vattr *vap = ap->a_vap;
697 struct mbuf *mreq, *mrep, *md, *mb;
698 int v3 = NFS_ISV3(vp);
701 * Update local times for special files.
703 mtx_lock(&np->n_mtx);
704 if (np->n_flag & (NACC | NUPD))
706 mtx_unlock(&np->n_mtx);
708 * First look in the cache.
710 if (nfs_getattrcache(vp, &vattr) == 0)
712 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
713 nfsstats.accesscache_misses++;
714 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL);
715 if (nfs_getattrcache(vp, &vattr) == 0)
718 nfsstats.rpccnt[NFSPROC_GETATTR]++;
719 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
721 bpos = mtod(mb, caddr_t);
723 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
725 nfsm_loadattr(vp, &vattr);
729 vap->va_type = vattr.va_type;
730 vap->va_mode = vattr.va_mode;
731 vap->va_nlink = vattr.va_nlink;
732 vap->va_uid = vattr.va_uid;
733 vap->va_gid = vattr.va_gid;
734 vap->va_fsid = vattr.va_fsid;
735 vap->va_fileid = vattr.va_fileid;
736 vap->va_size = vattr.va_size;
737 vap->va_blocksize = vattr.va_blocksize;
738 vap->va_atime = vattr.va_atime;
739 vap->va_mtime = vattr.va_mtime;
740 vap->va_ctime = vattr.va_ctime;
741 vap->va_gen = vattr.va_gen;
742 vap->va_flags = vattr.va_flags;
743 vap->va_rdev = vattr.va_rdev;
744 vap->va_bytes = vattr.va_bytes;
745 vap->va_filerev = vattr.va_filerev;
754 nfs_setattr(struct vop_setattr_args *ap)
756 struct vnode *vp = ap->a_vp;
757 struct nfsnode *np = VTONFS(vp);
758 struct vattr *vap = ap->a_vap;
759 struct thread *td = curthread;
768 * Setting of flags is not supported.
770 if (vap->va_flags != VNOVAL)
774 * Disallow write attempts if the filesystem is mounted read-only.
776 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
777 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
778 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
779 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
783 if (vap->va_size != VNOVAL) {
784 switch (vp->v_type) {
791 if (vap->va_mtime.tv_sec == VNOVAL &&
792 vap->va_atime.tv_sec == VNOVAL &&
793 vap->va_mode == (mode_t)VNOVAL &&
794 vap->va_uid == (uid_t)VNOVAL &&
795 vap->va_gid == (gid_t)VNOVAL)
797 vap->va_size = VNOVAL;
801 * Disallow write attempts if the filesystem is
804 if (vp->v_mount->mnt_flag & MNT_RDONLY)
807 * We run vnode_pager_setsize() early (why?),
808 * we must set np->n_size now to avoid vinvalbuf
809 * V_SAVE races that might setsize a lower
812 mtx_lock(&np->n_mtx);
814 mtx_unlock(&np->n_mtx);
815 error = nfs_meta_setsize(vp, ap->a_cred, td,
817 mtx_lock(&np->n_mtx);
818 if (np->n_flag & NMODIFIED) {
820 mtx_unlock(&np->n_mtx);
821 if (vap->va_size == 0)
822 error = nfs_vinvalbuf(vp, 0, td, 1);
824 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
826 vnode_pager_setsize(vp, tsize);
830 mtx_unlock(&np->n_mtx);
832 * np->n_size has already been set to vap->va_size
833 * in nfs_meta_setsize(). We must set it again since
834 * nfs_loadattrcache() could be called through
835 * nfs_meta_setsize() and could modify np->n_size.
837 mtx_lock(&np->n_mtx);
838 np->n_vattr.va_size = np->n_size = vap->va_size;
839 mtx_unlock(&np->n_mtx);
842 mtx_lock(&np->n_mtx);
843 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
844 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
845 mtx_unlock(&np->n_mtx);
846 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
847 (error == EINTR || error == EIO))
850 mtx_unlock(&np->n_mtx);
852 error = nfs_setattrrpc(vp, vap, ap->a_cred);
853 if (error && vap->va_size != VNOVAL) {
854 mtx_lock(&np->n_mtx);
855 np->n_size = np->n_vattr.va_size = tsize;
856 vnode_pager_setsize(vp, tsize);
857 mtx_unlock(&np->n_mtx);
864 * Do an nfs setattr rpc.
867 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
869 struct nfsv2_sattr *sp;
870 struct nfsnode *np = VTONFS(vp);
873 int error = 0, i, wccflag = NFSV3_WCCRATTR;
874 struct mbuf *mreq, *mrep, *md, *mb;
875 int v3 = NFS_ISV3(vp);
877 nfsstats.rpccnt[NFSPROC_SETATTR]++;
878 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
880 bpos = mtod(mb, caddr_t);
883 nfsm_v3attrbuild(vap, TRUE);
884 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
887 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
888 if (vap->va_mode == (mode_t)VNOVAL)
889 sp->sa_mode = nfs_xdrneg1;
891 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
892 if (vap->va_uid == (uid_t)VNOVAL)
893 sp->sa_uid = nfs_xdrneg1;
895 sp->sa_uid = txdr_unsigned(vap->va_uid);
896 if (vap->va_gid == (gid_t)VNOVAL)
897 sp->sa_gid = nfs_xdrneg1;
899 sp->sa_gid = txdr_unsigned(vap->va_gid);
900 sp->sa_size = txdr_unsigned(vap->va_size);
901 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
902 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
904 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
906 mtx_lock(&np->n_mtx);
907 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
908 np->n_accesscache[i].stamp = 0;
909 mtx_unlock(&np->n_mtx);
910 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
911 nfsm_wcc_data(vp, wccflag);
913 nfsm_loadattr(vp, NULL);
920 * nfs lookup call, one step at a time...
921 * First look in cache
922 * If not found, unlock the directory nfsnode and do the rpc
925 nfs_lookup(struct vop_lookup_args *ap)
927 struct componentname *cnp = ap->a_cnp;
928 struct vnode *dvp = ap->a_dvp;
929 struct vnode **vpp = ap->a_vpp;
930 struct mount *mp = dvp->v_mount;
931 struct vattr dvattr, vattr;
932 struct timespec nctime;
933 int flags = cnp->cn_flags;
935 struct nfsmount *nmp;
937 struct mbuf *mreq, *mrep, *md, *mb;
940 struct nfsnode *np, *newnp;
941 int error = 0, attrflag, dattrflag, fhsize, ltype, ncticks;
942 int v3 = NFS_ISV3(dvp);
943 struct thread *td = cnp->cn_thread;
946 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
947 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
949 if (dvp->v_type != VDIR)
953 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
957 error = cache_lookup_times(dvp, vpp, cnp, &nctime, &ncticks);
958 if (error > 0 && error != ENOENT)
962 * Lookups of "." are special and always return the
963 * current directory. cache_lookup() already handles
964 * associated locking bookkeeping, etc.
966 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
967 /* XXX: Is this really correct? */
968 if (cnp->cn_nameiop != LOOKUP &&
970 cnp->cn_flags |= SAVENAME;
975 * We only accept a positive hit in the cache if the
976 * change time of the file matches our cached copy.
977 * Otherwise, we discard the cache entry and fallback
978 * to doing a lookup RPC. We also only trust cache
979 * entries for less than nm_nametimeo seconds.
981 * To better handle stale file handles and attributes,
982 * clear the attribute cache of this node if it is a
983 * leaf component, part of an open() call, and not
984 * locally modified before fetching the attributes.
985 * This should allow stale file handles to be detected
986 * here where we can fall back to a LOOKUP RPC to
987 * recover rather than having nfs_open() detect the
988 * stale file handle and failing open(2) with ESTALE.
991 newnp = VTONFS(newvp);
992 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
993 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
994 !(newnp->n_flag & NMODIFIED)) {
995 mtx_lock(&newnp->n_mtx);
996 newnp->n_attrstamp = 0;
997 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
998 mtx_unlock(&newnp->n_mtx);
1000 if ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1001 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1002 timespeccmp(&vattr.va_ctime, &nctime, ==)) {
1003 nfsstats.lookupcache_hits++;
1004 if (cnp->cn_nameiop != LOOKUP &&
1006 cnp->cn_flags |= SAVENAME;
1015 } else if (error == ENOENT) {
1016 if (dvp->v_iflag & VI_DOOMED)
1019 * We only accept a negative hit in the cache if the
1020 * modification time of the parent directory matches
1021 * the cached copy in the name cache entry.
1022 * Otherwise, we discard all of the negative cache
1023 * entries for this directory. We also only trust
1024 * negative cache entries for up to nm_negnametimeo
1027 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1028 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1029 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1030 nfsstats.lookupcache_hits++;
1033 cache_purge_negative(dvp);
1036 attrflag = dattrflag = 0;
1039 nfsstats.lookupcache_misses++;
1040 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1041 len = cnp->cn_namelen;
1042 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1043 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1045 bpos = mtod(mb, caddr_t);
1046 nfsm_fhtom(dvp, v3);
1047 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1048 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
1051 nfsm_postop_attr_va(dvp, dattrflag, &vattr);
1056 nfsm_getfh(fhp, fhsize, v3);
1059 * Handle RENAME case...
1061 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1062 if (NFS_CMPFH(np, fhp, fhsize)) {
1066 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE);
1073 nfsm_postop_attr(newvp, attrflag);
1074 nfsm_postop_attr(dvp, attrflag);
1076 nfsm_loadattr(newvp, NULL);
1079 cnp->cn_flags |= SAVENAME;
1083 if (flags & ISDOTDOT) {
1084 ltype = VOP_ISLOCKED(dvp);
1085 error = vfs_busy(mp, MBF_NOWAIT);
1089 error = vfs_busy(mp, 0);
1090 vn_lock(dvp, ltype | LK_RETRY);
1092 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1102 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1107 vn_lock(dvp, ltype | LK_RETRY);
1108 if (dvp->v_iflag & VI_DOOMED) {
1121 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1125 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1133 * Flush the attribute cache when opening a leaf node
1134 * to ensure that fresh attributes are fetched in
1135 * nfs_open() if we are unable to fetch attributes
1136 * from the LOOKUP reply.
1138 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1139 !(np->n_flag & NMODIFIED)) {
1140 mtx_lock(&np->n_mtx);
1141 np->n_attrstamp = 0;
1142 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1143 mtx_unlock(&np->n_mtx);
1147 nfsm_postop_attr_va(newvp, attrflag, &vattr);
1148 nfsm_postop_attr_va(dvp, dattrflag, &dvattr);
1150 nfsm_loadattr(newvp, &vattr);
1153 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1154 cnp->cn_flags |= SAVENAME;
1155 if ((cnp->cn_flags & MAKEENTRY) &&
1156 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1157 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1158 cache_enter_time(dvp, newvp, cnp, &vattr.va_ctime,
1159 newvp->v_type != VDIR ? NULL : &dvattr.va_ctime);
1164 if (newvp != NULLVP) {
1169 if (error != ENOENT)
1172 /* The requested file was not found. */
1173 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1174 (flags & ISLASTCN)) {
1176 * XXX: UFS does a full VOP_ACCESS(dvp,
1177 * VWRITE) here instead of just checking
1180 if (mp->mnt_flag & MNT_RDONLY)
1182 cnp->cn_flags |= SAVENAME;
1183 return (EJUSTRETURN);
1186 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
1189 * Cache the modification time of the parent
1190 * directory from the post-op attributes in
1191 * the name cache entry. The negative cache
1192 * entry will be ignored once the directory
1193 * has changed. Don't bother adding the entry
1194 * if the directory has already changed.
1196 mtx_lock(&np->n_mtx);
1197 if (timespeccmp(&np->n_vattr.va_mtime,
1198 &vattr.va_mtime, ==)) {
1199 mtx_unlock(&np->n_mtx);
1200 cache_enter_time(dvp, NULL, cnp,
1201 &vattr.va_mtime, NULL);
1203 mtx_unlock(&np->n_mtx);
1213 * Just call nfs_bioread() to do the work.
1216 nfs_read(struct vop_read_args *ap)
1218 struct vnode *vp = ap->a_vp;
1220 switch (vp->v_type) {
1222 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1226 return (EOPNOTSUPP);
1234 nfs_readlink(struct vop_readlink_args *ap)
1236 struct vnode *vp = ap->a_vp;
1238 if (vp->v_type != VLNK)
1240 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1244 * Do a readlink rpc.
1245 * Called by nfs_doio() from below the buffer cache.
1248 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1251 int error = 0, len, attrflag;
1252 struct mbuf *mreq, *mrep, *md, *mb;
1253 int v3 = NFS_ISV3(vp);
1255 nfsstats.rpccnt[NFSPROC_READLINK]++;
1256 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1258 bpos = mtod(mb, caddr_t);
1260 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1262 nfsm_postop_attr(vp, attrflag);
1264 nfsm_strsiz(len, NFS_MAXPATHLEN);
1265 if (len == NFS_MAXPATHLEN) {
1266 struct nfsnode *np = VTONFS(vp);
1267 mtx_lock(&np->n_mtx);
1268 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1270 mtx_unlock(&np->n_mtx);
1272 nfsm_mtouio(uiop, len);
1284 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1288 struct mbuf *mreq, *mrep, *md, *mb;
1289 struct nfsmount *nmp;
1291 int error = 0, len, retlen, tsiz, eof, attrflag;
1292 int v3 = NFS_ISV3(vp);
1298 nmp = VFSTONFS(vp->v_mount);
1299 tsiz = uiop->uio_resid;
1300 mtx_lock(&nmp->nm_mtx);
1301 end = uiop->uio_offset + tsiz;
1302 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) {
1303 mtx_unlock(&nmp->nm_mtx);
1306 rsize = nmp->nm_rsize;
1307 mtx_unlock(&nmp->nm_mtx);
1309 nfsstats.rpccnt[NFSPROC_READ]++;
1310 len = (tsiz > rsize) ? rsize : tsiz;
1311 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1313 bpos = mtod(mb, caddr_t);
1315 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1317 txdr_hyper(uiop->uio_offset, tl);
1318 *(tl + 2) = txdr_unsigned(len);
1320 *tl++ = txdr_unsigned(uiop->uio_offset);
1321 *tl++ = txdr_unsigned(len);
1324 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1326 nfsm_postop_attr(vp, attrflag);
1331 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1332 eof = fxdr_unsigned(int, *(tl + 1));
1334 nfsm_loadattr(vp, NULL);
1336 nfsm_strsiz(retlen, rsize);
1337 nfsm_mtouio(uiop, retlen);
1341 if (eof || retlen == 0) {
1344 } else if (retlen < len) {
1356 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1357 int *iomode, int *must_commit)
1362 struct mbuf *mreq, *mrep, *md, *mb;
1363 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1365 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1366 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1369 KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1"));
1371 tsiz = uiop->uio_resid;
1372 mtx_lock(&nmp->nm_mtx);
1373 end = uiop->uio_offset + tsiz;
1374 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) {
1375 mtx_unlock(&nmp->nm_mtx);
1378 wsize = nmp->nm_wsize;
1379 mtx_unlock(&nmp->nm_mtx);
1381 nfsstats.rpccnt[NFSPROC_WRITE]++;
1382 len = (tsiz > wsize) ? wsize : tsiz;
1383 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1384 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1386 bpos = mtod(mb, caddr_t);
1389 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1390 txdr_hyper(uiop->uio_offset, tl);
1392 *tl++ = txdr_unsigned(len);
1393 *tl++ = txdr_unsigned(*iomode);
1394 *tl = txdr_unsigned(len);
1398 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1399 /* Set both "begin" and "current" to non-garbage. */
1400 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1401 *tl++ = x; /* "begin offset" */
1402 *tl++ = x; /* "current offset" */
1403 x = txdr_unsigned(len);
1404 *tl++ = x; /* total to this offset */
1405 *tl = x; /* size of this write */
1407 nfsm_uiotom(uiop, len);
1408 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1410 wccflag = NFSV3_WCCCHK;
1411 nfsm_wcc_data(vp, wccflag);
1413 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1414 + NFSX_V3WRITEVERF);
1415 rlen = fxdr_unsigned(int, *tl++);
1420 } else if (rlen < len) {
1421 backup = len - rlen;
1422 uiop->uio_iov->iov_base =
1423 (char *)uiop->uio_iov->iov_base -
1425 uiop->uio_iov->iov_len += backup;
1426 uiop->uio_offset -= backup;
1427 uiop->uio_resid += backup;
1430 commit = fxdr_unsigned(int, *tl++);
1433 * Return the lowest committment level
1434 * obtained by any of the RPCs.
1436 if (committed == NFSV3WRITE_FILESYNC)
1438 else if (committed == NFSV3WRITE_DATASYNC &&
1439 commit == NFSV3WRITE_UNSTABLE)
1441 mtx_lock(&nmp->nm_mtx);
1442 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1443 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1445 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1446 } else if (bcmp((caddr_t)tl,
1447 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1449 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1452 mtx_unlock(&nmp->nm_mtx);
1455 nfsm_loadattr(vp, NULL);
1458 mtx_lock(&(VTONFS(vp))->n_mtx);
1459 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1460 mtx_unlock(&(VTONFS(vp))->n_mtx);
1469 committed = NFSV3WRITE_FILESYNC;
1470 *iomode = committed;
1472 uiop->uio_resid = tsiz;
1478 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1479 * mode set to specify the file type and the size field for rdev.
1482 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1485 struct nfsv2_sattr *sp;
1487 struct vnode *newvp = NULL;
1488 struct nfsnode *np = NULL;
1491 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1492 struct mbuf *mreq, *mrep, *md, *mb;
1494 int v3 = NFS_ISV3(dvp);
1496 if (vap->va_type == VCHR || vap->va_type == VBLK)
1497 rdev = txdr_unsigned(vap->va_rdev);
1498 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1501 return (EOPNOTSUPP);
1503 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1505 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1506 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1507 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1509 bpos = mtod(mb, caddr_t);
1510 nfsm_fhtom(dvp, v3);
1511 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1513 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1514 *tl++ = vtonfsv3_type(vap->va_type);
1515 nfsm_v3attrbuild(vap, FALSE);
1516 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1517 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1518 *tl++ = txdr_unsigned(major(vap->va_rdev));
1519 *tl = txdr_unsigned(minor(vap->va_rdev));
1522 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1523 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1524 sp->sa_uid = nfs_xdrneg1;
1525 sp->sa_gid = nfs_xdrneg1;
1527 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1528 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1530 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1532 nfsm_mtofh(dvp, newvp, v3, gotvp);
1538 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1539 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1545 nfsm_wcc_data(dvp, wccflag);
1554 mtx_lock(&(VTONFS(dvp))->n_mtx);
1555 VTONFS(dvp)->n_flag |= NMODIFIED;
1557 VTONFS(dvp)->n_attrstamp = 0;
1558 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1560 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1566 * just call nfs_mknodrpc() to do the work.
1570 nfs_mknod(struct vop_mknod_args *ap)
1572 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1575 static u_long create_verf;
1577 * nfs file create call
1580 nfs_create(struct vop_create_args *ap)
1582 struct vnode *dvp = ap->a_dvp;
1583 struct vattr *vap = ap->a_vap;
1584 struct componentname *cnp = ap->a_cnp;
1585 struct nfsv2_sattr *sp;
1587 struct nfsnode *np = NULL;
1588 struct vnode *newvp = NULL;
1590 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1591 struct mbuf *mreq, *mrep, *md, *mb;
1593 int v3 = NFS_ISV3(dvp);
1596 * Oops, not for me..
1598 if (vap->va_type == VSOCK) {
1599 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1603 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) {
1606 if (vap->va_vaflags & VA_EXCLUSIVE)
1609 nfsstats.rpccnt[NFSPROC_CREATE]++;
1610 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1611 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1613 bpos = mtod(mb, caddr_t);
1614 nfsm_fhtom(dvp, v3);
1615 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1617 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1618 if (fmode & O_EXCL) {
1619 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1620 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1622 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1624 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1625 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1628 *tl++ = create_verf;
1630 IN_IFADDR_RUNLOCK();
1633 *tl = ++create_verf;
1635 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1636 nfsm_v3attrbuild(vap, FALSE);
1639 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1640 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1641 sp->sa_uid = nfs_xdrneg1;
1642 sp->sa_gid = nfs_xdrneg1;
1644 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1645 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1647 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1649 nfsm_mtofh(dvp, newvp, v3, gotvp);
1655 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1656 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1662 nfsm_wcc_data(dvp, wccflag);
1666 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1672 } else if (v3 && (fmode & O_EXCL)) {
1674 * We are normally called with only a partially initialized
1675 * VAP. Since the NFSv3 spec says that server may use the
1676 * file attributes to store the verifier, the spec requires
1677 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1678 * in atime, but we can't really assume that all servers will
1679 * so we ensure that our SETATTR sets both atime and mtime.
1681 if (vap->va_mtime.tv_sec == VNOVAL)
1682 vfs_timestamp(&vap->va_mtime);
1683 if (vap->va_atime.tv_sec == VNOVAL)
1684 vap->va_atime = vap->va_mtime;
1685 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred);
1692 mtx_lock(&(VTONFS(dvp))->n_mtx);
1693 VTONFS(dvp)->n_flag |= NMODIFIED;
1695 VTONFS(dvp)->n_attrstamp = 0;
1696 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1698 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1703 * nfs file remove call
1704 * To try and make nfs semantics closer to ufs semantics, a file that has
1705 * other processes using the vnode is renamed instead of removed and then
1706 * removed later on the last close.
1707 * - If v_usecount > 1
1708 * If a rename is not already in the works
1709 * call nfs_sillyrename() to set it up
1714 nfs_remove(struct vop_remove_args *ap)
1716 struct vnode *vp = ap->a_vp;
1717 struct vnode *dvp = ap->a_dvp;
1718 struct componentname *cnp = ap->a_cnp;
1719 struct nfsnode *np = VTONFS(vp);
1723 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1724 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1725 if (vp->v_type == VDIR)
1727 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1728 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1730 * Purge the name cache so that the chance of a lookup for
1731 * the name succeeding while the remove is in progress is
1732 * minimized. Without node locking it can still happen, such
1733 * that an I/O op returns ESTALE, but since you get this if
1734 * another host removes the file..
1738 * throw away biocache buffers, mainly to avoid
1739 * unnecessary delayed writes later.
1741 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1743 if (error != EINTR && error != EIO)
1744 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1745 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1747 * Kludge City: If the first reply to the remove rpc is lost..
1748 * the reply to the retransmitted request will be ENOENT
1749 * since the file was in fact removed
1750 * Therefore, we cheat and return success.
1752 if (error == ENOENT)
1754 } else if (!np->n_sillyrename)
1755 error = nfs_sillyrename(dvp, vp, cnp);
1756 mtx_lock(&np->n_mtx);
1757 np->n_attrstamp = 0;
1758 mtx_unlock(&np->n_mtx);
1759 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1764 * nfs file remove rpc called from nfs_inactive
1767 nfs_removeit(struct sillyrename *sp)
1770 * Make sure that the directory vnode is still valid.
1771 * XXX we should lock sp->s_dvp here.
1773 if (sp->s_dvp->v_type == VBAD)
1775 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1780 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1783 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1784 struct ucred *cred, struct thread *td)
1787 int error = 0, wccflag = NFSV3_WCCRATTR;
1788 struct mbuf *mreq, *mrep, *md, *mb;
1789 int v3 = NFS_ISV3(dvp);
1791 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1792 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1793 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1795 bpos = mtod(mb, caddr_t);
1796 nfsm_fhtom(dvp, v3);
1797 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1798 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1800 nfsm_wcc_data(dvp, wccflag);
1803 mtx_lock(&(VTONFS(dvp))->n_mtx);
1804 VTONFS(dvp)->n_flag |= NMODIFIED;
1806 VTONFS(dvp)->n_attrstamp = 0;
1807 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1809 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1814 * nfs file rename call
1817 nfs_rename(struct vop_rename_args *ap)
1819 struct vnode *fvp = ap->a_fvp;
1820 struct vnode *tvp = ap->a_tvp;
1821 struct vnode *fdvp = ap->a_fdvp;
1822 struct vnode *tdvp = ap->a_tdvp;
1823 struct componentname *tcnp = ap->a_tcnp;
1824 struct componentname *fcnp = ap->a_fcnp;
1827 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1828 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1829 /* Check for cross-device rename */
1830 if ((fvp->v_mount != tdvp->v_mount) ||
1831 (tvp && (fvp->v_mount != tvp->v_mount))) {
1837 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1841 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1845 * We have to flush B_DELWRI data prior to renaming
1846 * the file. If we don't, the delayed-write buffers
1847 * can be flushed out later after the file has gone stale
1848 * under NFSV3. NFSV2 does not have this problem because
1849 * ( as far as I can tell ) it flushes dirty buffers more
1852 * Skip the rename operation if the fsync fails, this can happen
1853 * due to the server's volume being full, when we pushed out data
1854 * that was written back to our cache earlier. Not checking for
1855 * this condition can result in potential (silent) data loss.
1857 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1860 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1865 * If the tvp exists and is in use, sillyrename it before doing the
1866 * rename of the new file over it.
1867 * XXX Can't sillyrename a directory.
1869 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1870 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1875 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1876 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1879 if (fvp->v_type == VDIR) {
1880 if (tvp != NULL && tvp->v_type == VDIR)
1895 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1897 if (error == ENOENT)
1903 * nfs file rename rpc called from nfs_remove() above
1906 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1907 struct sillyrename *sp)
1910 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1911 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1915 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1918 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1919 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1923 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1924 struct mbuf *mreq, *mrep, *md, *mb;
1925 int v3 = NFS_ISV3(fdvp);
1927 nfsstats.rpccnt[NFSPROC_RENAME]++;
1928 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1929 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1930 nfsm_rndup(tnamelen));
1932 bpos = mtod(mb, caddr_t);
1933 nfsm_fhtom(fdvp, v3);
1934 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1935 nfsm_fhtom(tdvp, v3);
1936 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1937 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1939 nfsm_wcc_data(fdvp, fwccflag);
1940 nfsm_wcc_data(tdvp, twccflag);
1944 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1945 VTONFS(fdvp)->n_flag |= NMODIFIED;
1946 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1947 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1948 VTONFS(tdvp)->n_flag |= NMODIFIED;
1949 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1951 VTONFS(fdvp)->n_attrstamp = 0;
1952 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1955 VTONFS(tdvp)->n_attrstamp = 0;
1956 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1962 * nfs hard link create call
1965 nfs_link(struct vop_link_args *ap)
1967 struct vnode *vp = ap->a_vp;
1968 struct vnode *tdvp = ap->a_tdvp;
1969 struct componentname *cnp = ap->a_cnp;
1971 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1972 struct mbuf *mreq, *mrep, *md, *mb;
1975 if (vp->v_mount != tdvp->v_mount) {
1980 * Push all writes to the server, so that the attribute cache
1981 * doesn't get "out of sync" with the server.
1982 * XXX There should be a better way!
1984 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1987 nfsstats.rpccnt[NFSPROC_LINK]++;
1988 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1989 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1991 bpos = mtod(mb, caddr_t);
1993 nfsm_fhtom(tdvp, v3);
1994 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1995 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1997 nfsm_postop_attr(vp, attrflag);
1998 nfsm_wcc_data(tdvp, wccflag);
2002 mtx_lock(&(VTONFS(tdvp))->n_mtx);
2003 VTONFS(tdvp)->n_flag |= NMODIFIED;
2004 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
2006 VTONFS(vp)->n_attrstamp = 0;
2007 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2010 VTONFS(tdvp)->n_attrstamp = 0;
2011 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2017 * nfs symbolic link create call
2020 nfs_symlink(struct vop_symlink_args *ap)
2022 struct vnode *dvp = ap->a_dvp;
2023 struct vattr *vap = ap->a_vap;
2024 struct componentname *cnp = ap->a_cnp;
2025 struct nfsv2_sattr *sp;
2027 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2028 struct mbuf *mreq, *mrep, *md, *mb;
2029 struct vnode *newvp = NULL;
2030 int v3 = NFS_ISV3(dvp);
2032 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2033 slen = strlen(ap->a_target);
2034 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2035 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2037 bpos = mtod(mb, caddr_t);
2038 nfsm_fhtom(dvp, v3);
2039 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2041 nfsm_v3attrbuild(vap, FALSE);
2043 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2045 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2046 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2047 sp->sa_uid = nfs_xdrneg1;
2048 sp->sa_gid = nfs_xdrneg1;
2049 sp->sa_size = nfs_xdrneg1;
2050 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2051 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2055 * Issue the NFS request and get the rpc response.
2057 * Only NFSv3 responses returning an error of 0 actually return
2058 * a file handle that can be converted into newvp without having
2059 * to do an extra lookup rpc.
2061 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
2064 nfsm_mtofh(dvp, newvp, v3, gotvp);
2065 nfsm_wcc_data(dvp, wccflag);
2069 * out code jumps -> here, mrep is also freed.
2076 * If we do not have an error and we could not extract the newvp from
2077 * the response due to the request being NFSv2, we have to do a
2078 * lookup in order to obtain a newvp to return.
2080 if (error == 0 && newvp == NULL) {
2081 struct nfsnode *np = NULL;
2083 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2084 cnp->cn_cred, cnp->cn_thread, &np);
2094 mtx_lock(&(VTONFS(dvp))->n_mtx);
2095 VTONFS(dvp)->n_flag |= NMODIFIED;
2096 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2098 VTONFS(dvp)->n_attrstamp = 0;
2099 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2108 nfs_mkdir(struct vop_mkdir_args *ap)
2110 struct vnode *dvp = ap->a_dvp;
2111 struct vattr *vap = ap->a_vap;
2112 struct componentname *cnp = ap->a_cnp;
2113 struct nfsv2_sattr *sp;
2115 struct nfsnode *np = NULL;
2116 struct vnode *newvp = NULL;
2118 int error = 0, wccflag = NFSV3_WCCRATTR;
2120 struct mbuf *mreq, *mrep, *md, *mb;
2122 int v3 = NFS_ISV3(dvp);
2124 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2126 len = cnp->cn_namelen;
2127 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2128 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
2129 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2131 bpos = mtod(mb, caddr_t);
2132 nfsm_fhtom(dvp, v3);
2133 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2135 nfsm_v3attrbuild(vap, FALSE);
2137 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2138 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2139 sp->sa_uid = nfs_xdrneg1;
2140 sp->sa_gid = nfs_xdrneg1;
2141 sp->sa_size = nfs_xdrneg1;
2142 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2143 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2145 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2147 nfsm_mtofh(dvp, newvp, v3, gotvp);
2149 nfsm_wcc_data(dvp, wccflag);
2152 mtx_lock(&(VTONFS(dvp))->n_mtx);
2153 VTONFS(dvp)->n_flag |= NMODIFIED;
2154 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2156 VTONFS(dvp)->n_attrstamp = 0;
2157 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2159 if (error == 0 && newvp == NULL) {
2160 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2161 cnp->cn_thread, &np);
2164 if (newvp->v_type != VDIR)
2177 * nfs remove directory call
2180 nfs_rmdir(struct vop_rmdir_args *ap)
2182 struct vnode *vp = ap->a_vp;
2183 struct vnode *dvp = ap->a_dvp;
2184 struct componentname *cnp = ap->a_cnp;
2186 int error = 0, wccflag = NFSV3_WCCRATTR;
2187 struct mbuf *mreq, *mrep, *md, *mb;
2188 int v3 = NFS_ISV3(dvp);
2192 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2193 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
2194 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2196 bpos = mtod(mb, caddr_t);
2197 nfsm_fhtom(dvp, v3);
2198 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2199 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2201 nfsm_wcc_data(dvp, wccflag);
2204 mtx_lock(&(VTONFS(dvp))->n_mtx);
2205 VTONFS(dvp)->n_flag |= NMODIFIED;
2206 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2208 VTONFS(dvp)->n_attrstamp = 0;
2209 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2214 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2216 if (error == ENOENT)
2225 nfs_readdir(struct vop_readdir_args *ap)
2227 struct vnode *vp = ap->a_vp;
2228 struct nfsnode *np = VTONFS(vp);
2229 struct uio *uio = ap->a_uio;
2230 int tresid, error = 0;
2233 if (vp->v_type != VDIR)
2237 * First, check for hit on the EOF offset cache
2239 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2240 (np->n_flag & NMODIFIED) == 0) {
2241 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2242 mtx_lock(&np->n_mtx);
2243 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2244 mtx_unlock(&np->n_mtx);
2245 nfsstats.direofcache_hits++;
2248 mtx_unlock(&np->n_mtx);
2253 * Call nfs_bioread() to do the real work.
2255 tresid = uio->uio_resid;
2256 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2258 if (!error && uio->uio_resid == tresid) {
2259 nfsstats.direofcache_misses++;
2267 * Called from below the buffer cache by nfs_doio().
2270 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2273 struct dirent *dp = NULL;
2278 struct mbuf *mreq, *mrep, *md, *mb;
2280 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2281 struct nfsnode *dnp = VTONFS(vp);
2283 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2285 int v3 = NFS_ISV3(vp);
2287 KASSERT(uiop->uio_iovcnt == 1 &&
2288 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2289 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2290 ("nfs readdirrpc bad uio"));
2293 * If there is no cookie, assume directory was stale.
2295 nfs_dircookie_lock(dnp);
2296 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2299 nfs_dircookie_unlock(dnp);
2301 nfs_dircookie_unlock(dnp);
2302 return (NFSERR_BAD_COOKIE);
2306 * Loop around doing readdir rpc's of size nm_readdirsize
2307 * truncated to a multiple of DIRBLKSIZ.
2308 * The stopping criteria is EOF or buffer full.
2310 while (more_dirs && bigenough) {
2311 nfsstats.rpccnt[NFSPROC_READDIR]++;
2312 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2315 bpos = mtod(mb, caddr_t);
2318 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2319 *tl++ = cookie.nfsuquad[0];
2320 *tl++ = cookie.nfsuquad[1];
2321 mtx_lock(&dnp->n_mtx);
2322 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2323 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2324 mtx_unlock(&dnp->n_mtx);
2326 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2327 *tl++ = cookie.nfsuquad[0];
2329 *tl = txdr_unsigned(nmp->nm_readdirsize);
2330 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2332 nfsm_postop_attr(vp, attrflag);
2334 tl = nfsm_dissect(u_int32_t *,
2336 mtx_lock(&dnp->n_mtx);
2337 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2338 dnp->n_cookieverf.nfsuquad[1] = *tl;
2339 mtx_unlock(&dnp->n_mtx);
2345 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2346 more_dirs = fxdr_unsigned(int, *tl);
2348 /* loop thru the dir entries, doctoring them to 4bsd form */
2349 while (more_dirs && bigenough) {
2351 tl = nfsm_dissect(u_int32_t *,
2353 fileno = fxdr_hyper(tl);
2354 len = fxdr_unsigned(int, *(tl + 2));
2356 tl = nfsm_dissect(u_int32_t *,
2358 fileno = fxdr_unsigned(u_quad_t, *tl++);
2359 len = fxdr_unsigned(int, *tl);
2361 if (len <= 0 || len > NFS_MAXNAMLEN) {
2366 tlen = nfsm_rndup(len);
2368 tlen += 4; /* To ensure null termination */
2369 left = DIRBLKSIZ - blksiz;
2370 if ((tlen + DIRHDSIZ) > left) {
2371 dp->d_reclen += left;
2372 uiop->uio_iov->iov_base =
2373 (char *)uiop->uio_iov->iov_base + left;
2374 uiop->uio_iov->iov_len -= left;
2375 uiop->uio_offset += left;
2376 uiop->uio_resid -= left;
2379 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2382 dp = (struct dirent *)uiop->uio_iov->iov_base;
2383 dp->d_fileno = (int)fileno;
2385 dp->d_reclen = tlen + DIRHDSIZ;
2386 dp->d_type = DT_UNKNOWN;
2387 blksiz += dp->d_reclen;
2388 if (blksiz == DIRBLKSIZ)
2390 uiop->uio_offset += DIRHDSIZ;
2391 uiop->uio_resid -= DIRHDSIZ;
2392 uiop->uio_iov->iov_base =
2393 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2394 uiop->uio_iov->iov_len -= DIRHDSIZ;
2395 nfsm_mtouio(uiop, len);
2396 cp = uiop->uio_iov->iov_base;
2398 *cp = '\0'; /* null terminate */
2399 uiop->uio_iov->iov_base =
2400 (char *)uiop->uio_iov->iov_base + tlen;
2401 uiop->uio_iov->iov_len -= tlen;
2402 uiop->uio_offset += tlen;
2403 uiop->uio_resid -= tlen;
2405 nfsm_adv(nfsm_rndup(len));
2407 tl = nfsm_dissect(u_int32_t *,
2410 tl = nfsm_dissect(u_int32_t *,
2414 cookie.nfsuquad[0] = *tl++;
2416 cookie.nfsuquad[1] = *tl++;
2421 more_dirs = fxdr_unsigned(int, *tl);
2424 * If at end of rpc data, get the eof boolean
2427 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2428 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2433 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2434 * by increasing d_reclen for the last record.
2437 left = DIRBLKSIZ - blksiz;
2438 dp->d_reclen += left;
2439 uiop->uio_iov->iov_base =
2440 (char *)uiop->uio_iov->iov_base + left;
2441 uiop->uio_iov->iov_len -= left;
2442 uiop->uio_offset += left;
2443 uiop->uio_resid -= left;
2447 * We are now either at the end of the directory or have filled the
2451 dnp->n_direofoffset = uiop->uio_offset;
2453 if (uiop->uio_resid > 0)
2454 nfs_printf("EEK! readdirrpc resid > 0\n");
2455 nfs_dircookie_lock(dnp);
2456 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2458 nfs_dircookie_unlock(dnp);
2465 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2468 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2474 struct vnode *newvp;
2476 caddr_t bpos, dpos, dpossav1, dpossav2;
2477 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2478 struct nameidata nami, *ndp = &nami;
2479 struct componentname *cnp = &ndp->ni_cnd;
2481 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2482 struct nfsnode *dnp = VTONFS(vp), *np;
2483 struct vattr vattr, dvattr;
2486 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2487 int attrflag, dattrflag, fhsize;
2492 KASSERT(uiop->uio_iovcnt == 1 &&
2493 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2494 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2495 ("nfs readdirplusrpc bad uio"));
2500 * If there is no cookie, assume directory was stale.
2502 nfs_dircookie_lock(dnp);
2503 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2506 nfs_dircookie_unlock(dnp);
2508 nfs_dircookie_unlock(dnp);
2509 return (NFSERR_BAD_COOKIE);
2512 * Loop around doing readdir rpc's of size nm_readdirsize
2513 * truncated to a multiple of DIRBLKSIZ.
2514 * The stopping criteria is EOF or buffer full.
2516 while (more_dirs && bigenough) {
2517 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2518 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2519 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2521 bpos = mtod(mb, caddr_t);
2523 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2524 *tl++ = cookie.nfsuquad[0];
2525 *tl++ = cookie.nfsuquad[1];
2526 mtx_lock(&dnp->n_mtx);
2527 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2528 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2529 mtx_unlock(&dnp->n_mtx);
2530 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2531 *tl = txdr_unsigned(nmp->nm_rsize);
2532 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2533 nfsm_postop_attr_va(vp, dattrflag, &dvattr);
2538 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2539 mtx_lock(&dnp->n_mtx);
2540 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2541 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2542 mtx_unlock(&dnp->n_mtx);
2543 more_dirs = fxdr_unsigned(int, *tl);
2545 /* loop thru the dir entries, doctoring them to 4bsd form */
2546 while (more_dirs && bigenough) {
2547 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2548 fileno = fxdr_hyper(tl);
2549 len = fxdr_unsigned(int, *(tl + 2));
2550 if (len <= 0 || len > NFS_MAXNAMLEN) {
2555 tlen = nfsm_rndup(len);
2557 tlen += 4; /* To ensure null termination*/
2558 left = DIRBLKSIZ - blksiz;
2559 if ((tlen + DIRHDSIZ) > left) {
2560 dp->d_reclen += left;
2561 uiop->uio_iov->iov_base =
2562 (char *)uiop->uio_iov->iov_base + left;
2563 uiop->uio_iov->iov_len -= left;
2564 uiop->uio_offset += left;
2565 uiop->uio_resid -= left;
2568 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2571 dp = (struct dirent *)uiop->uio_iov->iov_base;
2572 dp->d_fileno = (int)fileno;
2574 dp->d_reclen = tlen + DIRHDSIZ;
2575 dp->d_type = DT_UNKNOWN;
2576 blksiz += dp->d_reclen;
2577 if (blksiz == DIRBLKSIZ)
2579 uiop->uio_offset += DIRHDSIZ;
2580 uiop->uio_resid -= DIRHDSIZ;
2581 uiop->uio_iov->iov_base =
2582 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2583 uiop->uio_iov->iov_len -= DIRHDSIZ;
2584 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2585 cnp->cn_namelen = len;
2586 nfsm_mtouio(uiop, len);
2587 cp = uiop->uio_iov->iov_base;
2590 uiop->uio_iov->iov_base =
2591 (char *)uiop->uio_iov->iov_base + tlen;
2592 uiop->uio_iov->iov_len -= tlen;
2593 uiop->uio_offset += tlen;
2594 uiop->uio_resid -= tlen;
2596 nfsm_adv(nfsm_rndup(len));
2597 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2599 cookie.nfsuquad[0] = *tl++;
2600 cookie.nfsuquad[1] = *tl++;
2605 * Since the attributes are before the file handle
2606 * (sigh), we must skip over the attributes and then
2607 * come back and get them.
2609 attrflag = fxdr_unsigned(int, *tl);
2613 nfsm_adv(NFSX_V3FATTR);
2614 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2615 doit = fxdr_unsigned(int, *tl);
2617 * Skip loading the attrs for "..". There's a
2618 * race between loading the attrs here and
2619 * lookups that look for the directory currently
2620 * being read (in the parent). We try to acquire
2621 * the exclusive lock on ".." here, owning the
2622 * lock on the directory being read. Lookup will
2623 * hold the lock on ".." and try to acquire the
2624 * lock on the directory being read.
2626 * There are other ways of fixing this, one would
2627 * be to do a trylock on the ".." vnode and skip
2628 * loading the attrs on ".." if it happens to be
2629 * locked by another process. But skipping the
2630 * attrload on ".." seems the easiest option.
2632 if (strcmp(dp->d_name, "..") == 0) {
2635 * We've already skipped over the attrs,
2636 * skip over the filehandle. And store d_type
2639 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2640 i = fxdr_unsigned(int, *tl);
2641 nfsm_adv(nfsm_rndup(i));
2642 dp->d_type = IFTODT(VTTOIF(VDIR));
2645 nfsm_getfh(fhp, fhsize, 1);
2646 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2651 error = nfs_nget(vp->v_mount, fhp,
2652 fhsize, &np, LK_EXCLUSIVE);
2659 if (doit && bigenough) {
2664 nfsm_loadattr(newvp, &vattr);
2667 dp->d_type = IFTODT(VTTOIF(vattr.va_type));
2669 if (newvp->v_type != VDIR || dattrflag != 0)
2670 cache_enter_time(ndp->ni_dvp, ndp->ni_vp,
2671 cnp, &vattr.va_ctime,
2672 newvp->v_type != VDIR ? NULL :
2676 /* Just skip over the file handle */
2677 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2678 i = fxdr_unsigned(int, *tl);
2680 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2681 fhsize = fxdr_unsigned(int, *tl);
2682 nfsm_adv(nfsm_rndup(fhsize));
2685 if (newvp != NULLVP) {
2692 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2693 more_dirs = fxdr_unsigned(int, *tl);
2696 * If at end of rpc data, get the eof boolean
2699 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2700 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2705 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2706 * by increasing d_reclen for the last record.
2709 left = DIRBLKSIZ - blksiz;
2710 dp->d_reclen += left;
2711 uiop->uio_iov->iov_base =
2712 (char *)uiop->uio_iov->iov_base + left;
2713 uiop->uio_iov->iov_len -= left;
2714 uiop->uio_offset += left;
2715 uiop->uio_resid -= left;
2719 * We are now either at the end of the directory or have filled the
2723 dnp->n_direofoffset = uiop->uio_offset;
2725 if (uiop->uio_resid > 0)
2726 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2727 nfs_dircookie_lock(dnp);
2728 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2730 nfs_dircookie_unlock(dnp);
2733 if (newvp != NULLVP) {
2744 * Silly rename. To make the NFS filesystem that is stateless look a little
2745 * more like the "ufs" a remove of an active vnode is translated to a rename
2746 * to a funny looking filename that is removed by nfs_inactive on the
2747 * nfsnode. There is the potential for another process on a different client
2748 * to create the same funny name between the nfs_lookitup() fails and the
2749 * nfs_rename() completes, but...
2752 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2754 struct sillyrename *sp;
2758 unsigned int lticks;
2762 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2763 sp = malloc(sizeof (struct sillyrename),
2764 M_NFSREQ, M_WAITOK);
2765 sp->s_cred = crhold(cnp->cn_cred);
2767 sp->s_removeit = nfs_removeit;
2771 * Fudge together a funny name.
2772 * Changing the format of the funny name to accomodate more
2773 * sillynames per directory.
2774 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2775 * CPU ticks since boot.
2777 pid = cnp->cn_thread->td_proc->p_pid;
2778 lticks = (unsigned int)ticks;
2780 sp->s_namlen = sprintf(sp->s_name,
2781 ".nfs.%08x.%04x4.4", lticks,
2783 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2784 cnp->cn_thread, NULL))
2788 error = nfs_renameit(dvp, cnp, sp);
2791 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2792 cnp->cn_thread, &np);
2793 np->n_sillyrename = sp;
2798 free((caddr_t)sp, M_NFSREQ);
2803 * Look up a file name and optionally either update the file handle or
2804 * allocate an nfsnode, depending on the value of npp.
2805 * npp == NULL --> just do the lookup
2806 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2808 * *npp != NULL --> update the file handle in the vnode
2811 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2812 struct thread *td, struct nfsnode **npp)
2814 struct vnode *newvp = NULL;
2815 struct nfsnode *np, *dnp = VTONFS(dvp);
2817 int error = 0, fhlen, attrflag;
2818 struct mbuf *mreq, *mrep, *md, *mb;
2820 int v3 = NFS_ISV3(dvp);
2822 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2823 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2824 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2826 bpos = mtod(mb, caddr_t);
2827 nfsm_fhtom(dvp, v3);
2828 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2829 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2830 if (npp && !error) {
2831 nfsm_getfh(nfhp, fhlen, v3);
2834 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2835 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2836 np->n_fhp = &np->n_fh;
2837 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2838 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2839 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2840 np->n_fhsize = fhlen;
2842 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2846 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2854 nfsm_postop_attr(newvp, attrflag);
2855 if (!attrflag && *npp == NULL) {
2864 nfsm_loadattr(newvp, NULL);
2868 if (npp && *npp == NULL) {
2883 * Nfs Version 3 commit rpc
2886 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2890 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2892 int error = 0, wccflag = NFSV3_WCCRATTR;
2893 struct mbuf *mreq, *mrep, *md, *mb;
2895 mtx_lock(&nmp->nm_mtx);
2896 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2897 mtx_unlock(&nmp->nm_mtx);
2900 mtx_unlock(&nmp->nm_mtx);
2901 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2902 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2904 bpos = mtod(mb, caddr_t);
2906 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2907 txdr_hyper(offset, tl);
2909 *tl = txdr_unsigned(cnt);
2910 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2911 nfsm_wcc_data(vp, wccflag);
2913 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2914 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2915 NFSX_V3WRITEVERF)) {
2916 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2918 error = NFSERR_STALEWRITEVERF;
2928 * For async requests when nfsiod(s) are running, queue the request by
2929 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2933 nfs_strategy(struct vop_strategy_args *ap)
2935 struct buf *bp = ap->a_bp;
2938 KASSERT(!(bp->b_flags & B_DONE),
2939 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2940 BUF_ASSERT_HELD(bp);
2942 if (bp->b_iocmd == BIO_READ)
2948 * If the op is asynchronous and an i/o daemon is waiting
2949 * queue the request, wake it up and wait for completion
2950 * otherwise just do it ourselves.
2952 if ((bp->b_flags & B_ASYNC) == 0 ||
2953 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2954 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2959 * fsync vnode op. Just call nfs_flush() with commit == 1.
2963 nfs_fsync(struct vop_fsync_args *ap)
2966 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1));
2970 * Flush all the blocks associated with a vnode.
2971 * Walk through the buffer pool and push any dirty pages
2972 * associated with the vnode.
2975 nfs_flush(struct vnode *vp, int waitfor, int commit)
2977 struct nfsnode *np = VTONFS(vp);
2981 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2982 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2984 u_quad_t off, endoff, toff;
2985 struct ucred* wcred = NULL;
2986 struct buf **bvec = NULL;
2988 struct thread *td = curthread;
2989 #ifndef NFS_COMMITBVECSIZ
2990 #define NFS_COMMITBVECSIZ 20
2992 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2993 int bvecsize = 0, bveccount;
2995 if (nmp->nm_flag & NFSMNT_INT)
2996 slpflag = NFS_PCATCH;
3001 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
3002 * server, but has not been committed to stable storage on the server
3003 * yet. On the first pass, the byte range is worked out and the commit
3004 * rpc is done. On the second pass, nfs_writebp() is called to do the
3011 if (NFS_ISV3(vp) && commit) {
3012 if (bvec != NULL && bvec != bvec_on_stack)
3015 * Count up how many buffers waiting for a commit.
3019 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3020 if (!BUF_ISLOCKED(bp) &&
3021 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
3022 == (B_DELWRI | B_NEEDCOMMIT))
3026 * Allocate space to remember the list of bufs to commit. It is
3027 * important to use M_NOWAIT here to avoid a race with nfs_write.
3028 * If we can't get memory (for whatever reason), we will end up
3029 * committing the buffers one-by-one in the loop below.
3031 if (bveccount > NFS_COMMITBVECSIZ) {
3033 * Release the vnode interlock to avoid a lock
3037 bvec = (struct buf **)
3038 malloc(bveccount * sizeof(struct buf *),
3042 bvec = bvec_on_stack;
3043 bvecsize = NFS_COMMITBVECSIZ;
3045 bvecsize = bveccount;
3047 bvec = bvec_on_stack;
3048 bvecsize = NFS_COMMITBVECSIZ;
3050 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3051 if (bvecpos >= bvecsize)
3053 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3054 nbp = TAILQ_NEXT(bp, b_bobufs);
3057 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3058 (B_DELWRI | B_NEEDCOMMIT)) {
3060 nbp = TAILQ_NEXT(bp, b_bobufs);
3066 * Work out if all buffers are using the same cred
3067 * so we can deal with them all with one commit.
3069 * NOTE: we are not clearing B_DONE here, so we have
3070 * to do it later on in this routine if we intend to
3071 * initiate I/O on the bp.
3073 * Note: to avoid loopback deadlocks, we do not
3074 * assign b_runningbufspace.
3077 wcred = bp->b_wcred;
3078 else if (wcred != bp->b_wcred)
3080 vfs_busy_pages(bp, 1);
3084 * bp is protected by being locked, but nbp is not
3085 * and vfs_busy_pages() may sleep. We have to
3088 nbp = TAILQ_NEXT(bp, b_bobufs);
3091 * A list of these buffers is kept so that the
3092 * second loop knows which buffers have actually
3093 * been committed. This is necessary, since there
3094 * may be a race between the commit rpc and new
3095 * uncommitted writes on the file.
3097 bvec[bvecpos++] = bp;
3098 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3102 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3110 * Commit data on the server, as required.
3111 * If all bufs are using the same wcred, then use that with
3112 * one call for all of them, otherwise commit each one
3115 if (wcred != NOCRED)
3116 retv = nfs_commit(vp, off, (int)(endoff - off),
3120 for (i = 0; i < bvecpos; i++) {
3123 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3125 size = (u_quad_t)(bp->b_dirtyend
3127 retv = nfs_commit(vp, off, (int)size,
3133 if (retv == NFSERR_STALEWRITEVERF)
3134 nfs_clearcommit(vp->v_mount);
3137 * Now, either mark the blocks I/O done or mark the
3138 * blocks dirty, depending on whether the commit
3141 for (i = 0; i < bvecpos; i++) {
3143 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3146 * Error, leave B_DELWRI intact
3148 vfs_unbusy_pages(bp);
3152 * Success, remove B_DELWRI ( bundirty() ).
3154 * b_dirtyoff/b_dirtyend seem to be NFS
3155 * specific. We should probably move that
3156 * into bundirty(). XXX
3159 bp->b_flags |= B_ASYNC;
3161 bp->b_flags &= ~B_DONE;
3162 bp->b_ioflags &= ~BIO_ERROR;
3163 bp->b_dirtyoff = bp->b_dirtyend = 0;
3170 * Start/do any write(s) that are required.
3174 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3175 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3176 if (waitfor != MNT_WAIT || passone)
3179 error = BUF_TIMELOCK(bp,
3180 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3181 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
3186 if (error == ENOLCK) {
3190 if (nfs_sigintr(nmp, td)) {
3194 if (slpflag & PCATCH) {
3200 if ((bp->b_flags & B_DELWRI) == 0)
3201 panic("nfs_fsync: not dirty");
3202 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3208 if (passone || !commit)
3209 bp->b_flags |= B_ASYNC;
3211 bp->b_flags |= B_ASYNC;
3213 if (nfs_sigintr(nmp, td)) {
3224 if (waitfor == MNT_WAIT) {
3225 while (bo->bo_numoutput) {
3226 error = bufobj_wwait(bo, slpflag, slptimeo);
3229 error = nfs_sigintr(nmp, td);
3232 if (slpflag & PCATCH) {
3239 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3244 * Wait for all the async IO requests to drain
3247 mtx_lock(&np->n_mtx);
3248 while (np->n_directio_asyncwr > 0) {
3249 np->n_flag |= NFSYNCWAIT;
3250 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3251 &np->n_mtx, slpflag | (PRIBIO + 1),
3254 if (nfs_sigintr(nmp, td)) {
3255 mtx_unlock(&np->n_mtx);
3261 mtx_unlock(&np->n_mtx);
3264 mtx_lock(&np->n_mtx);
3265 if (np->n_flag & NWRITEERR) {
3266 error = np->n_error;
3267 np->n_flag &= ~NWRITEERR;
3269 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3270 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3271 np->n_flag &= ~NMODIFIED;
3272 mtx_unlock(&np->n_mtx);
3274 if (bvec != NULL && bvec != bvec_on_stack)
3280 * NFS advisory byte-level locks.
3283 nfs_advlock(struct vop_advlock_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_advlock(ap, &(vp->v_lockf), size);
3298 error = nfs_advlock_p(ap);
3307 * NFS advisory byte-level locks.
3310 nfs_advlockasync(struct vop_advlockasync_args *ap)
3312 struct vnode *vp = ap->a_vp;
3316 error = vn_lock(vp, LK_SHARED);
3319 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3320 size = VTONFS(vp)->n_size;
3322 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3331 * Print out the contents of an nfsnode.
3334 nfs_print(struct vop_print_args *ap)
3336 struct vnode *vp = ap->a_vp;
3337 struct nfsnode *np = VTONFS(vp);
3339 nfs_printf("\tfileid %ld fsid 0x%x",
3340 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3341 if (vp->v_type == VFIFO)
3348 * This is the "real" nfs::bwrite(struct buf*).
3349 * We set B_CACHE if this is a VMIO buffer.
3352 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3355 int oldflags = bp->b_flags;
3361 BUF_ASSERT_HELD(bp);
3363 if (bp->b_flags & B_INVAL) {
3368 bp->b_flags |= B_CACHE;
3371 * Undirty the bp. We will redirty it later if the I/O fails.
3376 bp->b_flags &= ~B_DONE;
3377 bp->b_ioflags &= ~BIO_ERROR;
3378 bp->b_iocmd = BIO_WRITE;
3380 bufobj_wref(bp->b_bufobj);
3381 curthread->td_ru.ru_oublock++;
3385 * Note: to avoid loopback deadlocks, we do not
3386 * assign b_runningbufspace.
3388 vfs_busy_pages(bp, 1);
3391 bp->b_iooffset = dbtob(bp->b_blkno);
3394 if( (oldflags & B_ASYNC) == 0) {
3395 int rtval = bufwait(bp);
3397 if (oldflags & B_DELWRI) {
3410 * nfs special file access vnode op.
3411 * Essentially just get vattr and then imitate iaccess() since the device is
3412 * local to the client.
3415 nfsspec_access(struct vop_access_args *ap)
3418 struct ucred *cred = ap->a_cred;
3419 struct vnode *vp = ap->a_vp;
3420 accmode_t accmode = ap->a_accmode;
3425 * Disallow write attempts on filesystems mounted read-only;
3426 * unless the file is a socket, fifo, or a block or character
3427 * device resident on the filesystem.
3429 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3430 switch (vp->v_type) {
3440 error = VOP_GETATTR(vp, vap, cred);
3443 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3444 accmode, cred, NULL);
3450 * Read wrapper for fifos.
3453 nfsfifo_read(struct vop_read_args *ap)
3455 struct nfsnode *np = VTONFS(ap->a_vp);
3461 mtx_lock(&np->n_mtx);
3463 vfs_timestamp(&np->n_atim);
3464 mtx_unlock(&np->n_mtx);
3465 error = fifo_specops.vop_read(ap);
3470 * Write wrapper for fifos.
3473 nfsfifo_write(struct vop_write_args *ap)
3475 struct nfsnode *np = VTONFS(ap->a_vp);
3480 mtx_lock(&np->n_mtx);
3482 vfs_timestamp(&np->n_mtim);
3483 mtx_unlock(&np->n_mtx);
3484 return(fifo_specops.vop_write(ap));
3488 * Close wrapper for fifos.
3490 * Update the times on the nfsnode then do fifo close.
3493 nfsfifo_close(struct vop_close_args *ap)
3495 struct vnode *vp = ap->a_vp;
3496 struct nfsnode *np = VTONFS(vp);
3500 mtx_lock(&np->n_mtx);
3501 if (np->n_flag & (NACC | NUPD)) {
3503 if (np->n_flag & NACC)
3505 if (np->n_flag & NUPD)
3508 if (vrefcnt(vp) == 1 &&
3509 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3511 if (np->n_flag & NACC)
3512 vattr.va_atime = np->n_atim;
3513 if (np->n_flag & NUPD)
3514 vattr.va_mtime = np->n_mtim;
3515 mtx_unlock(&np->n_mtx);
3516 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3520 mtx_unlock(&np->n_mtx);
3522 return (fifo_specops.vop_close(ap));
3526 * Just call nfs_writebp() with the force argument set to 1.
3528 * NOTE: B_DONE may or may not be set in a_bp on call.
3531 nfs_bwrite(struct buf *bp)
3534 return (nfs_writebp(bp, 1, curthread));
3537 struct buf_ops buf_ops_nfs = {
3538 .bop_name = "buf_ops_nfs",
3539 .bop_write = nfs_bwrite,
3540 .bop_strategy = bufstrategy,
3541 .bop_sync = bufsync,
3542 .bop_bdflush = bufbdflush,