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;
515 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
519 * Get a valid lease. If cached data is stale, flush it.
521 mtx_lock(&np->n_mtx);
522 if (np->n_flag & NMODIFIED) {
523 mtx_unlock(&np->n_mtx);
524 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
525 if (error == EINTR || error == EIO)
527 mtx_lock(&np->n_mtx);
529 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
530 if (vp->v_type == VDIR)
531 np->n_direofoffset = 0;
532 mtx_unlock(&np->n_mtx);
533 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
536 mtx_lock(&np->n_mtx);
537 np->n_mtime = vattr.va_mtime;
539 mtx_unlock(&np->n_mtx);
540 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
543 mtx_lock(&np->n_mtx);
544 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
545 if (vp->v_type == VDIR)
546 np->n_direofoffset = 0;
547 mtx_unlock(&np->n_mtx);
548 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
549 if (error == EINTR || error == EIO) {
552 mtx_lock(&np->n_mtx);
553 np->n_mtime = vattr.va_mtime;
557 * If the object has >= 1 O_DIRECT active opens, we disable caching.
559 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
560 if (np->n_directio_opens == 0) {
561 mtx_unlock(&np->n_mtx);
562 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
565 mtx_lock(&np->n_mtx);
566 np->n_flag |= NNONCACHE;
568 np->n_directio_opens++;
572 * If this is an open for writing, capture a reference to the
573 * credentials, so they can be used by nfs_putpages(). Using
574 * these write credentials is preferable to the credentials of
575 * whatever thread happens to be doing the VOP_PUTPAGES() since
576 * the write RPCs are less likely to fail with EACCES.
578 if ((fmode & FWRITE) != 0) {
579 cred = np->n_writecred;
580 np->n_writecred = crhold(ap->a_cred);
583 mtx_unlock(&np->n_mtx);
586 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
592 * What an NFS client should do upon close after writing is a debatable issue.
593 * Most NFS clients push delayed writes to the server upon close, basically for
595 * 1 - So that any write errors may be reported back to the client process
596 * doing the close system call. By far the two most likely errors are
597 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
598 * 2 - To put a worst case upper bound on cache inconsistency between
599 * multiple clients for the file.
600 * There is also a consistency problem for Version 2 of the protocol w.r.t.
601 * not being able to tell if other clients are writing a file concurrently,
602 * since there is no way of knowing if the changed modify time in the reply
603 * is only due to the write for this client.
604 * (NFS Version 3 provides weak cache consistency data in the reply that
605 * should be sufficient to detect and handle this case.)
607 * The current code does the following:
608 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
609 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
610 * or commit them (this satisfies 1 and 2 except for the
611 * case where the server crashes after this close but
612 * before the commit RPC, which is felt to be "good
613 * enough". Changing the last argument to nfs_flush() to
614 * a 1 would force a commit operation, if it is felt a
615 * commit is necessary now.
619 nfs_close(struct vop_close_args *ap)
621 struct vnode *vp = ap->a_vp;
622 struct nfsnode *np = VTONFS(vp);
624 int fmode = ap->a_fflag;
626 if (vp->v_type == VREG) {
628 * Examine and clean dirty pages, regardless of NMODIFIED.
629 * This closes a major hole in close-to-open consistency.
630 * We want to push out all dirty pages (and buffers) on
631 * close, regardless of whether they were dirtied by
632 * mmap'ed writes or via write().
634 if (nfs_clean_pages_on_close && vp->v_object) {
635 VM_OBJECT_LOCK(vp->v_object);
636 vm_object_page_clean(vp->v_object, 0, 0, 0);
637 VM_OBJECT_UNLOCK(vp->v_object);
639 mtx_lock(&np->n_mtx);
640 if (np->n_flag & NMODIFIED) {
641 mtx_unlock(&np->n_mtx);
644 * Under NFSv3 we have dirty buffers to dispose of. We
645 * must flush them to the NFS server. We have the option
646 * of waiting all the way through the commit rpc or just
647 * waiting for the initial write. The default is to only
648 * wait through the initial write so the data is in the
649 * server's cache, which is roughly similar to the state
650 * a standard disk subsystem leaves the file in on close().
652 * We cannot clear the NMODIFIED bit in np->n_flag due to
653 * potential races with other processes, and certainly
654 * cannot clear it if we don't commit.
656 int cm = nfsv3_commit_on_close ? 1 : 0;
657 error = nfs_flush(vp, MNT_WAIT, cm);
658 /* np->n_flag &= ~NMODIFIED; */
660 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
661 mtx_lock(&np->n_mtx);
663 if (np->n_flag & NWRITEERR) {
664 np->n_flag &= ~NWRITEERR;
667 mtx_unlock(&np->n_mtx);
669 if (nfs_directio_enable)
670 KASSERT((np->n_directio_asyncwr == 0),
671 ("nfs_close: dirty unflushed (%d) directio buffers\n",
672 np->n_directio_asyncwr));
673 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
674 mtx_lock(&np->n_mtx);
675 KASSERT((np->n_directio_opens > 0),
676 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
677 np->n_directio_opens--;
678 if (np->n_directio_opens == 0)
679 np->n_flag &= ~NNONCACHE;
680 mtx_unlock(&np->n_mtx);
686 * nfs getattr call from vfs.
689 nfs_getattr(struct vop_getattr_args *ap)
691 struct vnode *vp = ap->a_vp;
692 struct nfsnode *np = VTONFS(vp);
693 struct thread *td = curthread;
694 struct vattr *vap = ap->a_vap;
698 struct mbuf *mreq, *mrep, *md, *mb;
699 int v3 = NFS_ISV3(vp);
702 * Update local times for special files.
704 mtx_lock(&np->n_mtx);
705 if (np->n_flag & (NACC | NUPD))
707 mtx_unlock(&np->n_mtx);
709 * First look in the cache.
711 if (nfs_getattrcache(vp, &vattr) == 0)
713 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
714 nfsstats.accesscache_misses++;
715 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL);
716 if (nfs_getattrcache(vp, &vattr) == 0)
719 nfsstats.rpccnt[NFSPROC_GETATTR]++;
720 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
722 bpos = mtod(mb, caddr_t);
724 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
726 nfsm_loadattr(vp, &vattr);
730 vap->va_type = vattr.va_type;
731 vap->va_mode = vattr.va_mode;
732 vap->va_nlink = vattr.va_nlink;
733 vap->va_uid = vattr.va_uid;
734 vap->va_gid = vattr.va_gid;
735 vap->va_fsid = vattr.va_fsid;
736 vap->va_fileid = vattr.va_fileid;
737 vap->va_size = vattr.va_size;
738 vap->va_blocksize = vattr.va_blocksize;
739 vap->va_atime = vattr.va_atime;
740 vap->va_mtime = vattr.va_mtime;
741 vap->va_ctime = vattr.va_ctime;
742 vap->va_gen = vattr.va_gen;
743 vap->va_flags = vattr.va_flags;
744 vap->va_rdev = vattr.va_rdev;
745 vap->va_bytes = vattr.va_bytes;
746 vap->va_filerev = vattr.va_filerev;
755 nfs_setattr(struct vop_setattr_args *ap)
757 struct vnode *vp = ap->a_vp;
758 struct nfsnode *np = VTONFS(vp);
759 struct vattr *vap = ap->a_vap;
760 struct thread *td = curthread;
769 * Setting of flags is not supported.
771 if (vap->va_flags != VNOVAL)
775 * Disallow write attempts if the filesystem is mounted read-only.
777 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
778 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
779 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
780 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
784 if (vap->va_size != VNOVAL) {
785 switch (vp->v_type) {
792 if (vap->va_mtime.tv_sec == VNOVAL &&
793 vap->va_atime.tv_sec == VNOVAL &&
794 vap->va_mode == (mode_t)VNOVAL &&
795 vap->va_uid == (uid_t)VNOVAL &&
796 vap->va_gid == (gid_t)VNOVAL)
798 vap->va_size = VNOVAL;
802 * Disallow write attempts if the filesystem is
805 if (vp->v_mount->mnt_flag & MNT_RDONLY)
808 * We run vnode_pager_setsize() early (why?),
809 * we must set np->n_size now to avoid vinvalbuf
810 * V_SAVE races that might setsize a lower
813 mtx_lock(&np->n_mtx);
815 mtx_unlock(&np->n_mtx);
816 error = nfs_meta_setsize(vp, ap->a_cred, td,
818 mtx_lock(&np->n_mtx);
819 if (np->n_flag & NMODIFIED) {
821 mtx_unlock(&np->n_mtx);
822 if (vap->va_size == 0)
823 error = nfs_vinvalbuf(vp, 0, td, 1);
825 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
827 vnode_pager_setsize(vp, tsize);
831 mtx_unlock(&np->n_mtx);
833 * np->n_size has already been set to vap->va_size
834 * in nfs_meta_setsize(). We must set it again since
835 * nfs_loadattrcache() could be called through
836 * nfs_meta_setsize() and could modify np->n_size.
838 mtx_lock(&np->n_mtx);
839 np->n_vattr.va_size = np->n_size = vap->va_size;
840 mtx_unlock(&np->n_mtx);
843 mtx_lock(&np->n_mtx);
844 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
845 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
846 mtx_unlock(&np->n_mtx);
847 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
848 (error == EINTR || error == EIO))
851 mtx_unlock(&np->n_mtx);
853 error = nfs_setattrrpc(vp, vap, ap->a_cred);
854 if (error && vap->va_size != VNOVAL) {
855 mtx_lock(&np->n_mtx);
856 np->n_size = np->n_vattr.va_size = tsize;
857 vnode_pager_setsize(vp, tsize);
858 mtx_unlock(&np->n_mtx);
865 * Do an nfs setattr rpc.
868 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
870 struct nfsv2_sattr *sp;
871 struct nfsnode *np = VTONFS(vp);
874 int error = 0, i, wccflag = NFSV3_WCCRATTR;
875 struct mbuf *mreq, *mrep, *md, *mb;
876 int v3 = NFS_ISV3(vp);
878 nfsstats.rpccnt[NFSPROC_SETATTR]++;
879 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
881 bpos = mtod(mb, caddr_t);
884 nfsm_v3attrbuild(vap, TRUE);
885 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
888 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
889 if (vap->va_mode == (mode_t)VNOVAL)
890 sp->sa_mode = nfs_xdrneg1;
892 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
893 if (vap->va_uid == (uid_t)VNOVAL)
894 sp->sa_uid = nfs_xdrneg1;
896 sp->sa_uid = txdr_unsigned(vap->va_uid);
897 if (vap->va_gid == (gid_t)VNOVAL)
898 sp->sa_gid = nfs_xdrneg1;
900 sp->sa_gid = txdr_unsigned(vap->va_gid);
901 sp->sa_size = txdr_unsigned(vap->va_size);
902 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
903 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
905 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
907 mtx_lock(&np->n_mtx);
908 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
909 np->n_accesscache[i].stamp = 0;
910 mtx_unlock(&np->n_mtx);
911 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
912 nfsm_wcc_data(vp, wccflag);
914 nfsm_loadattr(vp, NULL);
921 * nfs lookup call, one step at a time...
922 * First look in cache
923 * If not found, unlock the directory nfsnode and do the rpc
926 nfs_lookup(struct vop_lookup_args *ap)
928 struct componentname *cnp = ap->a_cnp;
929 struct vnode *dvp = ap->a_dvp;
930 struct vnode **vpp = ap->a_vpp;
931 struct mount *mp = dvp->v_mount;
932 struct vattr dvattr, vattr;
933 struct timespec nctime;
934 int flags = cnp->cn_flags;
936 struct nfsmount *nmp;
938 struct mbuf *mreq, *mrep, *md, *mb;
941 struct nfsnode *np, *newnp;
942 int error = 0, attrflag, dattrflag, fhsize, ltype, ncticks;
943 int v3 = NFS_ISV3(dvp);
944 struct thread *td = cnp->cn_thread;
947 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
948 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
950 if (dvp->v_type != VDIR)
954 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
958 error = cache_lookup_times(dvp, vpp, cnp, &nctime, &ncticks);
959 if (error > 0 && error != ENOENT)
963 * Lookups of "." are special and always return the
964 * current directory. cache_lookup() already handles
965 * associated locking bookkeeping, etc.
967 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
968 /* XXX: Is this really correct? */
969 if (cnp->cn_nameiop != LOOKUP &&
971 cnp->cn_flags |= SAVENAME;
976 * We only accept a positive hit in the cache if the
977 * change time of the file matches our cached copy.
978 * Otherwise, we discard the cache entry and fallback
979 * to doing a lookup RPC. We also only trust cache
980 * entries for less than nm_nametimeo seconds.
982 * To better handle stale file handles and attributes,
983 * clear the attribute cache of this node if it is a
984 * leaf component, part of an open() call, and not
985 * locally modified before fetching the attributes.
986 * This should allow stale file handles to be detected
987 * here where we can fall back to a LOOKUP RPC to
988 * recover rather than having nfs_open() detect the
989 * stale file handle and failing open(2) with ESTALE.
992 newnp = VTONFS(newvp);
993 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
994 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
995 !(newnp->n_flag & NMODIFIED)) {
996 mtx_lock(&newnp->n_mtx);
997 newnp->n_attrstamp = 0;
998 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
999 mtx_unlock(&newnp->n_mtx);
1001 if ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1002 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1003 timespeccmp(&vattr.va_ctime, &nctime, ==)) {
1004 nfsstats.lookupcache_hits++;
1005 if (cnp->cn_nameiop != LOOKUP &&
1007 cnp->cn_flags |= SAVENAME;
1016 } else if (error == ENOENT) {
1017 if (dvp->v_iflag & VI_DOOMED)
1020 * We only accept a negative hit in the cache if the
1021 * modification time of the parent directory matches
1022 * the cached copy in the name cache entry.
1023 * Otherwise, we discard all of the negative cache
1024 * entries for this directory. We also only trust
1025 * negative cache entries for up to nm_negnametimeo
1028 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1029 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1030 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1031 nfsstats.lookupcache_hits++;
1034 cache_purge_negative(dvp);
1037 attrflag = dattrflag = 0;
1040 nfsstats.lookupcache_misses++;
1041 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1042 len = cnp->cn_namelen;
1043 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1044 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1046 bpos = mtod(mb, caddr_t);
1047 nfsm_fhtom(dvp, v3);
1048 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1049 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
1052 nfsm_postop_attr_va(dvp, dattrflag, &vattr);
1057 nfsm_getfh(fhp, fhsize, v3);
1060 * Handle RENAME case...
1062 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1063 if (NFS_CMPFH(np, fhp, fhsize)) {
1067 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE);
1074 nfsm_postop_attr(newvp, attrflag);
1075 nfsm_postop_attr(dvp, attrflag);
1077 nfsm_loadattr(newvp, NULL);
1080 cnp->cn_flags |= SAVENAME;
1084 if (flags & ISDOTDOT) {
1085 ltype = VOP_ISLOCKED(dvp);
1086 error = vfs_busy(mp, MBF_NOWAIT);
1090 error = vfs_busy(mp, 0);
1091 vn_lock(dvp, ltype | LK_RETRY);
1093 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1103 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1108 vn_lock(dvp, ltype | LK_RETRY);
1109 if (dvp->v_iflag & VI_DOOMED) {
1122 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1126 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1134 * Flush the attribute cache when opening a leaf node
1135 * to ensure that fresh attributes are fetched in
1136 * nfs_open() if we are unable to fetch attributes
1137 * from the LOOKUP reply.
1139 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1140 !(np->n_flag & NMODIFIED)) {
1141 mtx_lock(&np->n_mtx);
1142 np->n_attrstamp = 0;
1143 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1144 mtx_unlock(&np->n_mtx);
1148 nfsm_postop_attr_va(newvp, attrflag, &vattr);
1149 nfsm_postop_attr_va(dvp, dattrflag, &dvattr);
1151 nfsm_loadattr(newvp, &vattr);
1154 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1155 cnp->cn_flags |= SAVENAME;
1156 if ((cnp->cn_flags & MAKEENTRY) &&
1157 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1158 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1159 cache_enter_time(dvp, newvp, cnp, &vattr.va_ctime,
1160 newvp->v_type != VDIR ? NULL : &dvattr.va_ctime);
1165 if (newvp != NULLVP) {
1170 if (error != ENOENT)
1173 /* The requested file was not found. */
1174 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1175 (flags & ISLASTCN)) {
1177 * XXX: UFS does a full VOP_ACCESS(dvp,
1178 * VWRITE) here instead of just checking
1181 if (mp->mnt_flag & MNT_RDONLY)
1183 cnp->cn_flags |= SAVENAME;
1184 return (EJUSTRETURN);
1187 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
1190 * Cache the modification time of the parent
1191 * directory from the post-op attributes in
1192 * the name cache entry. The negative cache
1193 * entry will be ignored once the directory
1194 * has changed. Don't bother adding the entry
1195 * if the directory has already changed.
1197 mtx_lock(&np->n_mtx);
1198 if (timespeccmp(&np->n_vattr.va_mtime,
1199 &vattr.va_mtime, ==)) {
1200 mtx_unlock(&np->n_mtx);
1201 cache_enter_time(dvp, NULL, cnp,
1202 &vattr.va_mtime, NULL);
1204 mtx_unlock(&np->n_mtx);
1214 * Just call nfs_bioread() to do the work.
1217 nfs_read(struct vop_read_args *ap)
1219 struct vnode *vp = ap->a_vp;
1221 switch (vp->v_type) {
1223 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1227 return (EOPNOTSUPP);
1235 nfs_readlink(struct vop_readlink_args *ap)
1237 struct vnode *vp = ap->a_vp;
1239 if (vp->v_type != VLNK)
1241 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1245 * Do a readlink rpc.
1246 * Called by nfs_doio() from below the buffer cache.
1249 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1252 int error = 0, len, attrflag;
1253 struct mbuf *mreq, *mrep, *md, *mb;
1254 int v3 = NFS_ISV3(vp);
1256 nfsstats.rpccnt[NFSPROC_READLINK]++;
1257 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1259 bpos = mtod(mb, caddr_t);
1261 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1263 nfsm_postop_attr(vp, attrflag);
1265 nfsm_strsiz(len, NFS_MAXPATHLEN);
1266 if (len == NFS_MAXPATHLEN) {
1267 struct nfsnode *np = VTONFS(vp);
1268 mtx_lock(&np->n_mtx);
1269 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1271 mtx_unlock(&np->n_mtx);
1273 nfsm_mtouio(uiop, len);
1285 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1289 struct mbuf *mreq, *mrep, *md, *mb;
1290 struct nfsmount *nmp;
1292 int error = 0, len, retlen, tsiz, eof, attrflag;
1293 int v3 = NFS_ISV3(vp);
1299 nmp = VFSTONFS(vp->v_mount);
1300 tsiz = uiop->uio_resid;
1301 mtx_lock(&nmp->nm_mtx);
1302 end = uiop->uio_offset + tsiz;
1303 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) {
1304 mtx_unlock(&nmp->nm_mtx);
1307 rsize = nmp->nm_rsize;
1308 mtx_unlock(&nmp->nm_mtx);
1310 nfsstats.rpccnt[NFSPROC_READ]++;
1311 len = (tsiz > rsize) ? rsize : tsiz;
1312 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1314 bpos = mtod(mb, caddr_t);
1316 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1318 txdr_hyper(uiop->uio_offset, tl);
1319 *(tl + 2) = txdr_unsigned(len);
1321 *tl++ = txdr_unsigned(uiop->uio_offset);
1322 *tl++ = txdr_unsigned(len);
1325 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1327 nfsm_postop_attr(vp, attrflag);
1332 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1333 eof = fxdr_unsigned(int, *(tl + 1));
1335 nfsm_loadattr(vp, NULL);
1337 nfsm_strsiz(retlen, rsize);
1338 nfsm_mtouio(uiop, retlen);
1342 if (eof || retlen == 0) {
1345 } else if (retlen < len) {
1357 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1358 int *iomode, int *must_commit)
1363 struct mbuf *mreq, *mrep, *md, *mb;
1364 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1366 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1367 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1370 KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1"));
1372 tsiz = uiop->uio_resid;
1373 mtx_lock(&nmp->nm_mtx);
1374 end = uiop->uio_offset + tsiz;
1375 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) {
1376 mtx_unlock(&nmp->nm_mtx);
1379 wsize = nmp->nm_wsize;
1380 mtx_unlock(&nmp->nm_mtx);
1382 nfsstats.rpccnt[NFSPROC_WRITE]++;
1383 len = (tsiz > wsize) ? wsize : tsiz;
1384 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1385 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1387 bpos = mtod(mb, caddr_t);
1390 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1391 txdr_hyper(uiop->uio_offset, tl);
1393 *tl++ = txdr_unsigned(len);
1394 *tl++ = txdr_unsigned(*iomode);
1395 *tl = txdr_unsigned(len);
1399 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1400 /* Set both "begin" and "current" to non-garbage. */
1401 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1402 *tl++ = x; /* "begin offset" */
1403 *tl++ = x; /* "current offset" */
1404 x = txdr_unsigned(len);
1405 *tl++ = x; /* total to this offset */
1406 *tl = x; /* size of this write */
1408 nfsm_uiotom(uiop, len);
1409 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1411 wccflag = NFSV3_WCCCHK;
1412 nfsm_wcc_data(vp, wccflag);
1414 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1415 + NFSX_V3WRITEVERF);
1416 rlen = fxdr_unsigned(int, *tl++);
1421 } else if (rlen < len) {
1422 backup = len - rlen;
1423 uiop->uio_iov->iov_base =
1424 (char *)uiop->uio_iov->iov_base -
1426 uiop->uio_iov->iov_len += backup;
1427 uiop->uio_offset -= backup;
1428 uiop->uio_resid += backup;
1431 commit = fxdr_unsigned(int, *tl++);
1434 * Return the lowest committment level
1435 * obtained by any of the RPCs.
1437 if (committed == NFSV3WRITE_FILESYNC)
1439 else if (committed == NFSV3WRITE_DATASYNC &&
1440 commit == NFSV3WRITE_UNSTABLE)
1442 mtx_lock(&nmp->nm_mtx);
1443 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1444 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1446 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1447 } else if (bcmp((caddr_t)tl,
1448 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1450 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1453 mtx_unlock(&nmp->nm_mtx);
1456 nfsm_loadattr(vp, NULL);
1459 mtx_lock(&(VTONFS(vp))->n_mtx);
1460 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1461 mtx_unlock(&(VTONFS(vp))->n_mtx);
1470 committed = NFSV3WRITE_FILESYNC;
1471 *iomode = committed;
1473 uiop->uio_resid = tsiz;
1479 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1480 * mode set to specify the file type and the size field for rdev.
1483 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1486 struct nfsv2_sattr *sp;
1488 struct vnode *newvp = NULL;
1489 struct nfsnode *np = NULL;
1492 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1493 struct mbuf *mreq, *mrep, *md, *mb;
1495 int v3 = NFS_ISV3(dvp);
1497 if (vap->va_type == VCHR || vap->va_type == VBLK)
1498 rdev = txdr_unsigned(vap->va_rdev);
1499 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1502 return (EOPNOTSUPP);
1504 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1506 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1507 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1508 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1510 bpos = mtod(mb, caddr_t);
1511 nfsm_fhtom(dvp, v3);
1512 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1514 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1515 *tl++ = vtonfsv3_type(vap->va_type);
1516 nfsm_v3attrbuild(vap, FALSE);
1517 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1518 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1519 *tl++ = txdr_unsigned(major(vap->va_rdev));
1520 *tl = txdr_unsigned(minor(vap->va_rdev));
1523 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1524 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1525 sp->sa_uid = nfs_xdrneg1;
1526 sp->sa_gid = nfs_xdrneg1;
1528 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1529 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1531 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1533 nfsm_mtofh(dvp, newvp, v3, gotvp);
1539 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1540 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1546 nfsm_wcc_data(dvp, wccflag);
1555 mtx_lock(&(VTONFS(dvp))->n_mtx);
1556 VTONFS(dvp)->n_flag |= NMODIFIED;
1558 VTONFS(dvp)->n_attrstamp = 0;
1559 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1561 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1567 * just call nfs_mknodrpc() to do the work.
1571 nfs_mknod(struct vop_mknod_args *ap)
1573 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1576 static u_long create_verf;
1578 * nfs file create call
1581 nfs_create(struct vop_create_args *ap)
1583 struct vnode *dvp = ap->a_dvp;
1584 struct vattr *vap = ap->a_vap;
1585 struct componentname *cnp = ap->a_cnp;
1586 struct nfsv2_sattr *sp;
1588 struct nfsnode *np = NULL;
1589 struct vnode *newvp = NULL;
1591 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1592 struct mbuf *mreq, *mrep, *md, *mb;
1594 int v3 = NFS_ISV3(dvp);
1597 * Oops, not for me..
1599 if (vap->va_type == VSOCK) {
1600 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1604 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) {
1607 if (vap->va_vaflags & VA_EXCLUSIVE)
1610 nfsstats.rpccnt[NFSPROC_CREATE]++;
1611 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1612 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1614 bpos = mtod(mb, caddr_t);
1615 nfsm_fhtom(dvp, v3);
1616 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1618 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1619 if (fmode & O_EXCL) {
1620 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1621 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1623 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1625 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1626 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1629 *tl++ = create_verf;
1631 IN_IFADDR_RUNLOCK();
1634 *tl = ++create_verf;
1636 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1637 nfsm_v3attrbuild(vap, FALSE);
1640 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1641 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1642 sp->sa_uid = nfs_xdrneg1;
1643 sp->sa_gid = nfs_xdrneg1;
1645 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1646 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1648 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1650 nfsm_mtofh(dvp, newvp, v3, gotvp);
1656 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1657 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1663 nfsm_wcc_data(dvp, wccflag);
1667 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1673 } else if (v3 && (fmode & O_EXCL)) {
1675 * We are normally called with only a partially initialized
1676 * VAP. Since the NFSv3 spec says that server may use the
1677 * file attributes to store the verifier, the spec requires
1678 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1679 * in atime, but we can't really assume that all servers will
1680 * so we ensure that our SETATTR sets both atime and mtime.
1682 if (vap->va_mtime.tv_sec == VNOVAL)
1683 vfs_timestamp(&vap->va_mtime);
1684 if (vap->va_atime.tv_sec == VNOVAL)
1685 vap->va_atime = vap->va_mtime;
1686 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred);
1693 mtx_lock(&(VTONFS(dvp))->n_mtx);
1694 VTONFS(dvp)->n_flag |= NMODIFIED;
1696 VTONFS(dvp)->n_attrstamp = 0;
1697 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1699 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1704 * nfs file remove call
1705 * To try and make nfs semantics closer to ufs semantics, a file that has
1706 * other processes using the vnode is renamed instead of removed and then
1707 * removed later on the last close.
1708 * - If v_usecount > 1
1709 * If a rename is not already in the works
1710 * call nfs_sillyrename() to set it up
1715 nfs_remove(struct vop_remove_args *ap)
1717 struct vnode *vp = ap->a_vp;
1718 struct vnode *dvp = ap->a_dvp;
1719 struct componentname *cnp = ap->a_cnp;
1720 struct nfsnode *np = VTONFS(vp);
1724 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1725 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1726 if (vp->v_type == VDIR)
1728 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1729 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1731 * Purge the name cache so that the chance of a lookup for
1732 * the name succeeding while the remove is in progress is
1733 * minimized. Without node locking it can still happen, such
1734 * that an I/O op returns ESTALE, but since you get this if
1735 * another host removes the file..
1739 * throw away biocache buffers, mainly to avoid
1740 * unnecessary delayed writes later.
1742 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1744 if (error != EINTR && error != EIO)
1745 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1746 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1748 * Kludge City: If the first reply to the remove rpc is lost..
1749 * the reply to the retransmitted request will be ENOENT
1750 * since the file was in fact removed
1751 * Therefore, we cheat and return success.
1753 if (error == ENOENT)
1755 } else if (!np->n_sillyrename)
1756 error = nfs_sillyrename(dvp, vp, cnp);
1757 mtx_lock(&np->n_mtx);
1758 np->n_attrstamp = 0;
1759 mtx_unlock(&np->n_mtx);
1760 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1765 * nfs file remove rpc called from nfs_inactive
1768 nfs_removeit(struct sillyrename *sp)
1771 * Make sure that the directory vnode is still valid.
1772 * XXX we should lock sp->s_dvp here.
1774 if (sp->s_dvp->v_type == VBAD)
1776 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1781 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1784 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1785 struct ucred *cred, struct thread *td)
1788 int error = 0, wccflag = NFSV3_WCCRATTR;
1789 struct mbuf *mreq, *mrep, *md, *mb;
1790 int v3 = NFS_ISV3(dvp);
1792 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1793 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1794 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1796 bpos = mtod(mb, caddr_t);
1797 nfsm_fhtom(dvp, v3);
1798 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1799 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1801 nfsm_wcc_data(dvp, wccflag);
1804 mtx_lock(&(VTONFS(dvp))->n_mtx);
1805 VTONFS(dvp)->n_flag |= NMODIFIED;
1807 VTONFS(dvp)->n_attrstamp = 0;
1808 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1810 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1815 * nfs file rename call
1818 nfs_rename(struct vop_rename_args *ap)
1820 struct vnode *fvp = ap->a_fvp;
1821 struct vnode *tvp = ap->a_tvp;
1822 struct vnode *fdvp = ap->a_fdvp;
1823 struct vnode *tdvp = ap->a_tdvp;
1824 struct componentname *tcnp = ap->a_tcnp;
1825 struct componentname *fcnp = ap->a_fcnp;
1828 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1829 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1830 /* Check for cross-device rename */
1831 if ((fvp->v_mount != tdvp->v_mount) ||
1832 (tvp && (fvp->v_mount != tvp->v_mount))) {
1838 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1842 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1846 * We have to flush B_DELWRI data prior to renaming
1847 * the file. If we don't, the delayed-write buffers
1848 * can be flushed out later after the file has gone stale
1849 * under NFSV3. NFSV2 does not have this problem because
1850 * ( as far as I can tell ) it flushes dirty buffers more
1853 * Skip the rename operation if the fsync fails, this can happen
1854 * due to the server's volume being full, when we pushed out data
1855 * that was written back to our cache earlier. Not checking for
1856 * this condition can result in potential (silent) data loss.
1858 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1861 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1866 * If the tvp exists and is in use, sillyrename it before doing the
1867 * rename of the new file over it.
1868 * XXX Can't sillyrename a directory.
1870 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1871 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1876 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1877 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1880 if (fvp->v_type == VDIR) {
1881 if (tvp != NULL && tvp->v_type == VDIR)
1896 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1898 if (error == ENOENT)
1904 * nfs file rename rpc called from nfs_remove() above
1907 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1908 struct sillyrename *sp)
1911 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1912 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1916 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1919 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1920 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1924 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1925 struct mbuf *mreq, *mrep, *md, *mb;
1926 int v3 = NFS_ISV3(fdvp);
1928 nfsstats.rpccnt[NFSPROC_RENAME]++;
1929 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1930 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1931 nfsm_rndup(tnamelen));
1933 bpos = mtod(mb, caddr_t);
1934 nfsm_fhtom(fdvp, v3);
1935 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1936 nfsm_fhtom(tdvp, v3);
1937 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1938 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1940 nfsm_wcc_data(fdvp, fwccflag);
1941 nfsm_wcc_data(tdvp, twccflag);
1945 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1946 VTONFS(fdvp)->n_flag |= NMODIFIED;
1947 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1948 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1949 VTONFS(tdvp)->n_flag |= NMODIFIED;
1950 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1952 VTONFS(fdvp)->n_attrstamp = 0;
1953 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1956 VTONFS(tdvp)->n_attrstamp = 0;
1957 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1963 * nfs hard link create call
1966 nfs_link(struct vop_link_args *ap)
1968 struct vnode *vp = ap->a_vp;
1969 struct vnode *tdvp = ap->a_tdvp;
1970 struct componentname *cnp = ap->a_cnp;
1972 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1973 struct mbuf *mreq, *mrep, *md, *mb;
1976 if (vp->v_mount != tdvp->v_mount) {
1981 * Push all writes to the server, so that the attribute cache
1982 * doesn't get "out of sync" with the server.
1983 * XXX There should be a better way!
1985 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1988 nfsstats.rpccnt[NFSPROC_LINK]++;
1989 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1990 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1992 bpos = mtod(mb, caddr_t);
1994 nfsm_fhtom(tdvp, v3);
1995 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1996 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1998 nfsm_postop_attr(vp, attrflag);
1999 nfsm_wcc_data(tdvp, wccflag);
2003 mtx_lock(&(VTONFS(tdvp))->n_mtx);
2004 VTONFS(tdvp)->n_flag |= NMODIFIED;
2005 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
2007 VTONFS(vp)->n_attrstamp = 0;
2008 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2011 VTONFS(tdvp)->n_attrstamp = 0;
2012 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2018 * nfs symbolic link create call
2021 nfs_symlink(struct vop_symlink_args *ap)
2023 struct vnode *dvp = ap->a_dvp;
2024 struct vattr *vap = ap->a_vap;
2025 struct componentname *cnp = ap->a_cnp;
2026 struct nfsv2_sattr *sp;
2028 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2029 struct mbuf *mreq, *mrep, *md, *mb;
2030 struct vnode *newvp = NULL;
2031 int v3 = NFS_ISV3(dvp);
2033 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2034 slen = strlen(ap->a_target);
2035 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2036 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2038 bpos = mtod(mb, caddr_t);
2039 nfsm_fhtom(dvp, v3);
2040 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2042 nfsm_v3attrbuild(vap, FALSE);
2044 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2046 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2047 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2048 sp->sa_uid = nfs_xdrneg1;
2049 sp->sa_gid = nfs_xdrneg1;
2050 sp->sa_size = nfs_xdrneg1;
2051 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2052 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2056 * Issue the NFS request and get the rpc response.
2058 * Only NFSv3 responses returning an error of 0 actually return
2059 * a file handle that can be converted into newvp without having
2060 * to do an extra lookup rpc.
2062 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
2065 nfsm_mtofh(dvp, newvp, v3, gotvp);
2066 nfsm_wcc_data(dvp, wccflag);
2070 * out code jumps -> here, mrep is also freed.
2077 * If we do not have an error and we could not extract the newvp from
2078 * the response due to the request being NFSv2, we have to do a
2079 * lookup in order to obtain a newvp to return.
2081 if (error == 0 && newvp == NULL) {
2082 struct nfsnode *np = NULL;
2084 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2085 cnp->cn_cred, cnp->cn_thread, &np);
2095 mtx_lock(&(VTONFS(dvp))->n_mtx);
2096 VTONFS(dvp)->n_flag |= NMODIFIED;
2097 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2099 VTONFS(dvp)->n_attrstamp = 0;
2100 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2109 nfs_mkdir(struct vop_mkdir_args *ap)
2111 struct vnode *dvp = ap->a_dvp;
2112 struct vattr *vap = ap->a_vap;
2113 struct componentname *cnp = ap->a_cnp;
2114 struct nfsv2_sattr *sp;
2116 struct nfsnode *np = NULL;
2117 struct vnode *newvp = NULL;
2119 int error = 0, wccflag = NFSV3_WCCRATTR;
2121 struct mbuf *mreq, *mrep, *md, *mb;
2123 int v3 = NFS_ISV3(dvp);
2125 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2127 len = cnp->cn_namelen;
2128 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2129 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
2130 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2132 bpos = mtod(mb, caddr_t);
2133 nfsm_fhtom(dvp, v3);
2134 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2136 nfsm_v3attrbuild(vap, FALSE);
2138 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2139 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2140 sp->sa_uid = nfs_xdrneg1;
2141 sp->sa_gid = nfs_xdrneg1;
2142 sp->sa_size = nfs_xdrneg1;
2143 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2144 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2146 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2148 nfsm_mtofh(dvp, newvp, v3, gotvp);
2150 nfsm_wcc_data(dvp, wccflag);
2153 mtx_lock(&(VTONFS(dvp))->n_mtx);
2154 VTONFS(dvp)->n_flag |= NMODIFIED;
2155 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2157 VTONFS(dvp)->n_attrstamp = 0;
2158 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2160 if (error == 0 && newvp == NULL) {
2161 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2162 cnp->cn_thread, &np);
2165 if (newvp->v_type != VDIR)
2178 * nfs remove directory call
2181 nfs_rmdir(struct vop_rmdir_args *ap)
2183 struct vnode *vp = ap->a_vp;
2184 struct vnode *dvp = ap->a_dvp;
2185 struct componentname *cnp = ap->a_cnp;
2187 int error = 0, wccflag = NFSV3_WCCRATTR;
2188 struct mbuf *mreq, *mrep, *md, *mb;
2189 int v3 = NFS_ISV3(dvp);
2193 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2194 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
2195 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2197 bpos = mtod(mb, caddr_t);
2198 nfsm_fhtom(dvp, v3);
2199 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2200 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2202 nfsm_wcc_data(dvp, wccflag);
2205 mtx_lock(&(VTONFS(dvp))->n_mtx);
2206 VTONFS(dvp)->n_flag |= NMODIFIED;
2207 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2209 VTONFS(dvp)->n_attrstamp = 0;
2210 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2215 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2217 if (error == ENOENT)
2226 nfs_readdir(struct vop_readdir_args *ap)
2228 struct vnode *vp = ap->a_vp;
2229 struct nfsnode *np = VTONFS(vp);
2230 struct uio *uio = ap->a_uio;
2231 int tresid, error = 0;
2234 if (vp->v_type != VDIR)
2238 * First, check for hit on the EOF offset cache
2240 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2241 (np->n_flag & NMODIFIED) == 0) {
2242 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2243 mtx_lock(&np->n_mtx);
2244 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2245 mtx_unlock(&np->n_mtx);
2246 nfsstats.direofcache_hits++;
2249 mtx_unlock(&np->n_mtx);
2254 * Call nfs_bioread() to do the real work.
2256 tresid = uio->uio_resid;
2257 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2259 if (!error && uio->uio_resid == tresid) {
2260 nfsstats.direofcache_misses++;
2268 * Called from below the buffer cache by nfs_doio().
2271 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2274 struct dirent *dp = NULL;
2279 struct mbuf *mreq, *mrep, *md, *mb;
2281 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2282 struct nfsnode *dnp = VTONFS(vp);
2284 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2286 int v3 = NFS_ISV3(vp);
2288 KASSERT(uiop->uio_iovcnt == 1 &&
2289 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2290 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2291 ("nfs readdirrpc bad uio"));
2294 * If there is no cookie, assume directory was stale.
2296 nfs_dircookie_lock(dnp);
2297 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2300 nfs_dircookie_unlock(dnp);
2302 nfs_dircookie_unlock(dnp);
2303 return (NFSERR_BAD_COOKIE);
2307 * Loop around doing readdir rpc's of size nm_readdirsize
2308 * truncated to a multiple of DIRBLKSIZ.
2309 * The stopping criteria is EOF or buffer full.
2311 while (more_dirs && bigenough) {
2312 nfsstats.rpccnt[NFSPROC_READDIR]++;
2313 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2316 bpos = mtod(mb, caddr_t);
2319 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2320 *tl++ = cookie.nfsuquad[0];
2321 *tl++ = cookie.nfsuquad[1];
2322 mtx_lock(&dnp->n_mtx);
2323 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2324 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2325 mtx_unlock(&dnp->n_mtx);
2327 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2328 *tl++ = cookie.nfsuquad[0];
2330 *tl = txdr_unsigned(nmp->nm_readdirsize);
2331 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2333 nfsm_postop_attr(vp, attrflag);
2335 tl = nfsm_dissect(u_int32_t *,
2337 mtx_lock(&dnp->n_mtx);
2338 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2339 dnp->n_cookieverf.nfsuquad[1] = *tl;
2340 mtx_unlock(&dnp->n_mtx);
2346 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2347 more_dirs = fxdr_unsigned(int, *tl);
2349 /* loop thru the dir entries, doctoring them to 4bsd form */
2350 while (more_dirs && bigenough) {
2352 tl = nfsm_dissect(u_int32_t *,
2354 fileno = fxdr_hyper(tl);
2355 len = fxdr_unsigned(int, *(tl + 2));
2357 tl = nfsm_dissect(u_int32_t *,
2359 fileno = fxdr_unsigned(u_quad_t, *tl++);
2360 len = fxdr_unsigned(int, *tl);
2362 if (len <= 0 || len > NFS_MAXNAMLEN) {
2367 tlen = nfsm_rndup(len);
2369 tlen += 4; /* To ensure null termination */
2370 left = DIRBLKSIZ - blksiz;
2371 if ((tlen + DIRHDSIZ) > left) {
2372 dp->d_reclen += left;
2373 uiop->uio_iov->iov_base =
2374 (char *)uiop->uio_iov->iov_base + left;
2375 uiop->uio_iov->iov_len -= left;
2376 uiop->uio_offset += left;
2377 uiop->uio_resid -= left;
2380 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2383 dp = (struct dirent *)uiop->uio_iov->iov_base;
2384 dp->d_fileno = (int)fileno;
2386 dp->d_reclen = tlen + DIRHDSIZ;
2387 dp->d_type = DT_UNKNOWN;
2388 blksiz += dp->d_reclen;
2389 if (blksiz == DIRBLKSIZ)
2391 uiop->uio_offset += DIRHDSIZ;
2392 uiop->uio_resid -= DIRHDSIZ;
2393 uiop->uio_iov->iov_base =
2394 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2395 uiop->uio_iov->iov_len -= DIRHDSIZ;
2396 nfsm_mtouio(uiop, len);
2397 cp = uiop->uio_iov->iov_base;
2399 *cp = '\0'; /* null terminate */
2400 uiop->uio_iov->iov_base =
2401 (char *)uiop->uio_iov->iov_base + tlen;
2402 uiop->uio_iov->iov_len -= tlen;
2403 uiop->uio_offset += tlen;
2404 uiop->uio_resid -= tlen;
2406 nfsm_adv(nfsm_rndup(len));
2408 tl = nfsm_dissect(u_int32_t *,
2411 tl = nfsm_dissect(u_int32_t *,
2415 cookie.nfsuquad[0] = *tl++;
2417 cookie.nfsuquad[1] = *tl++;
2422 more_dirs = fxdr_unsigned(int, *tl);
2425 * If at end of rpc data, get the eof boolean
2428 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2429 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2434 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2435 * by increasing d_reclen for the last record.
2438 left = DIRBLKSIZ - blksiz;
2439 dp->d_reclen += left;
2440 uiop->uio_iov->iov_base =
2441 (char *)uiop->uio_iov->iov_base + left;
2442 uiop->uio_iov->iov_len -= left;
2443 uiop->uio_offset += left;
2444 uiop->uio_resid -= left;
2448 * We are now either at the end of the directory or have filled the
2452 dnp->n_direofoffset = uiop->uio_offset;
2454 if (uiop->uio_resid > 0)
2455 nfs_printf("EEK! readdirrpc resid > 0\n");
2456 nfs_dircookie_lock(dnp);
2457 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2459 nfs_dircookie_unlock(dnp);
2466 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2469 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2475 struct vnode *newvp;
2477 caddr_t bpos, dpos, dpossav1, dpossav2;
2478 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2479 struct nameidata nami, *ndp = &nami;
2480 struct componentname *cnp = &ndp->ni_cnd;
2482 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2483 struct nfsnode *dnp = VTONFS(vp), *np;
2484 struct vattr vattr, dvattr;
2487 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2488 int attrflag, dattrflag, fhsize;
2493 KASSERT(uiop->uio_iovcnt == 1 &&
2494 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2495 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2496 ("nfs readdirplusrpc bad uio"));
2501 * If there is no cookie, assume directory was stale.
2503 nfs_dircookie_lock(dnp);
2504 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2507 nfs_dircookie_unlock(dnp);
2509 nfs_dircookie_unlock(dnp);
2510 return (NFSERR_BAD_COOKIE);
2513 * Loop around doing readdir rpc's of size nm_readdirsize
2514 * truncated to a multiple of DIRBLKSIZ.
2515 * The stopping criteria is EOF or buffer full.
2517 while (more_dirs && bigenough) {
2518 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2519 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2520 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2522 bpos = mtod(mb, caddr_t);
2524 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2525 *tl++ = cookie.nfsuquad[0];
2526 *tl++ = cookie.nfsuquad[1];
2527 mtx_lock(&dnp->n_mtx);
2528 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2529 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2530 mtx_unlock(&dnp->n_mtx);
2531 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2532 *tl = txdr_unsigned(nmp->nm_rsize);
2533 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2534 nfsm_postop_attr_va(vp, dattrflag, &dvattr);
2539 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2540 mtx_lock(&dnp->n_mtx);
2541 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2542 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2543 mtx_unlock(&dnp->n_mtx);
2544 more_dirs = fxdr_unsigned(int, *tl);
2546 /* loop thru the dir entries, doctoring them to 4bsd form */
2547 while (more_dirs && bigenough) {
2548 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2549 fileno = fxdr_hyper(tl);
2550 len = fxdr_unsigned(int, *(tl + 2));
2551 if (len <= 0 || len > NFS_MAXNAMLEN) {
2556 tlen = nfsm_rndup(len);
2558 tlen += 4; /* To ensure null termination*/
2559 left = DIRBLKSIZ - blksiz;
2560 if ((tlen + DIRHDSIZ) > left) {
2561 dp->d_reclen += left;
2562 uiop->uio_iov->iov_base =
2563 (char *)uiop->uio_iov->iov_base + left;
2564 uiop->uio_iov->iov_len -= left;
2565 uiop->uio_offset += left;
2566 uiop->uio_resid -= left;
2569 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2572 dp = (struct dirent *)uiop->uio_iov->iov_base;
2573 dp->d_fileno = (int)fileno;
2575 dp->d_reclen = tlen + DIRHDSIZ;
2576 dp->d_type = DT_UNKNOWN;
2577 blksiz += dp->d_reclen;
2578 if (blksiz == DIRBLKSIZ)
2580 uiop->uio_offset += DIRHDSIZ;
2581 uiop->uio_resid -= DIRHDSIZ;
2582 uiop->uio_iov->iov_base =
2583 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2584 uiop->uio_iov->iov_len -= DIRHDSIZ;
2585 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2586 cnp->cn_namelen = len;
2587 nfsm_mtouio(uiop, len);
2588 cp = uiop->uio_iov->iov_base;
2591 uiop->uio_iov->iov_base =
2592 (char *)uiop->uio_iov->iov_base + tlen;
2593 uiop->uio_iov->iov_len -= tlen;
2594 uiop->uio_offset += tlen;
2595 uiop->uio_resid -= tlen;
2597 nfsm_adv(nfsm_rndup(len));
2598 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2600 cookie.nfsuquad[0] = *tl++;
2601 cookie.nfsuquad[1] = *tl++;
2606 * Since the attributes are before the file handle
2607 * (sigh), we must skip over the attributes and then
2608 * come back and get them.
2610 attrflag = fxdr_unsigned(int, *tl);
2614 nfsm_adv(NFSX_V3FATTR);
2615 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2616 doit = fxdr_unsigned(int, *tl);
2618 * Skip loading the attrs for "..". There's a
2619 * race between loading the attrs here and
2620 * lookups that look for the directory currently
2621 * being read (in the parent). We try to acquire
2622 * the exclusive lock on ".." here, owning the
2623 * lock on the directory being read. Lookup will
2624 * hold the lock on ".." and try to acquire the
2625 * lock on the directory being read.
2627 * There are other ways of fixing this, one would
2628 * be to do a trylock on the ".." vnode and skip
2629 * loading the attrs on ".." if it happens to be
2630 * locked by another process. But skipping the
2631 * attrload on ".." seems the easiest option.
2633 if (strcmp(dp->d_name, "..") == 0) {
2636 * We've already skipped over the attrs,
2637 * skip over the filehandle. And store d_type
2640 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2641 i = fxdr_unsigned(int, *tl);
2642 nfsm_adv(nfsm_rndup(i));
2643 dp->d_type = IFTODT(VTTOIF(VDIR));
2646 nfsm_getfh(fhp, fhsize, 1);
2647 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2652 error = nfs_nget(vp->v_mount, fhp,
2653 fhsize, &np, LK_EXCLUSIVE);
2660 if (doit && bigenough) {
2665 nfsm_loadattr(newvp, &vattr);
2668 dp->d_type = IFTODT(VTTOIF(vattr.va_type));
2670 if (newvp->v_type != VDIR || dattrflag != 0)
2671 cache_enter_time(ndp->ni_dvp, ndp->ni_vp,
2672 cnp, &vattr.va_ctime,
2673 newvp->v_type != VDIR ? NULL :
2677 /* Just skip over the file handle */
2678 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2679 i = fxdr_unsigned(int, *tl);
2681 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2682 fhsize = fxdr_unsigned(int, *tl);
2683 nfsm_adv(nfsm_rndup(fhsize));
2686 if (newvp != NULLVP) {
2693 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2694 more_dirs = fxdr_unsigned(int, *tl);
2697 * If at end of rpc data, get the eof boolean
2700 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2701 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2706 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2707 * by increasing d_reclen for the last record.
2710 left = DIRBLKSIZ - blksiz;
2711 dp->d_reclen += left;
2712 uiop->uio_iov->iov_base =
2713 (char *)uiop->uio_iov->iov_base + left;
2714 uiop->uio_iov->iov_len -= left;
2715 uiop->uio_offset += left;
2716 uiop->uio_resid -= left;
2720 * We are now either at the end of the directory or have filled the
2724 dnp->n_direofoffset = uiop->uio_offset;
2726 if (uiop->uio_resid > 0)
2727 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2728 nfs_dircookie_lock(dnp);
2729 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2731 nfs_dircookie_unlock(dnp);
2734 if (newvp != NULLVP) {
2745 * Silly rename. To make the NFS filesystem that is stateless look a little
2746 * more like the "ufs" a remove of an active vnode is translated to a rename
2747 * to a funny looking filename that is removed by nfs_inactive on the
2748 * nfsnode. There is the potential for another process on a different client
2749 * to create the same funny name between the nfs_lookitup() fails and the
2750 * nfs_rename() completes, but...
2753 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2755 struct sillyrename *sp;
2759 unsigned int lticks;
2763 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2764 sp = malloc(sizeof (struct sillyrename),
2765 M_NFSREQ, M_WAITOK);
2766 sp->s_cred = crhold(cnp->cn_cred);
2768 sp->s_removeit = nfs_removeit;
2772 * Fudge together a funny name.
2773 * Changing the format of the funny name to accomodate more
2774 * sillynames per directory.
2775 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2776 * CPU ticks since boot.
2778 pid = cnp->cn_thread->td_proc->p_pid;
2779 lticks = (unsigned int)ticks;
2781 sp->s_namlen = sprintf(sp->s_name,
2782 ".nfs.%08x.%04x4.4", lticks,
2784 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2785 cnp->cn_thread, NULL))
2789 error = nfs_renameit(dvp, cnp, sp);
2792 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2793 cnp->cn_thread, &np);
2794 np->n_sillyrename = sp;
2799 free((caddr_t)sp, M_NFSREQ);
2804 * Look up a file name and optionally either update the file handle or
2805 * allocate an nfsnode, depending on the value of npp.
2806 * npp == NULL --> just do the lookup
2807 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2809 * *npp != NULL --> update the file handle in the vnode
2812 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2813 struct thread *td, struct nfsnode **npp)
2815 struct vnode *newvp = NULL;
2816 struct nfsnode *np, *dnp = VTONFS(dvp);
2818 int error = 0, fhlen, attrflag;
2819 struct mbuf *mreq, *mrep, *md, *mb;
2821 int v3 = NFS_ISV3(dvp);
2823 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2824 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2825 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2827 bpos = mtod(mb, caddr_t);
2828 nfsm_fhtom(dvp, v3);
2829 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2830 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2831 if (npp && !error) {
2832 nfsm_getfh(nfhp, fhlen, v3);
2835 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2836 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2837 np->n_fhp = &np->n_fh;
2838 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2839 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2840 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2841 np->n_fhsize = fhlen;
2843 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2847 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2855 nfsm_postop_attr(newvp, attrflag);
2856 if (!attrflag && *npp == NULL) {
2865 nfsm_loadattr(newvp, NULL);
2869 if (npp && *npp == NULL) {
2884 * Nfs Version 3 commit rpc
2887 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2891 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2893 int error = 0, wccflag = NFSV3_WCCRATTR;
2894 struct mbuf *mreq, *mrep, *md, *mb;
2896 mtx_lock(&nmp->nm_mtx);
2897 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2898 mtx_unlock(&nmp->nm_mtx);
2901 mtx_unlock(&nmp->nm_mtx);
2902 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2903 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2905 bpos = mtod(mb, caddr_t);
2907 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2908 txdr_hyper(offset, tl);
2910 *tl = txdr_unsigned(cnt);
2911 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2912 nfsm_wcc_data(vp, wccflag);
2914 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2915 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2916 NFSX_V3WRITEVERF)) {
2917 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2919 error = NFSERR_STALEWRITEVERF;
2929 * For async requests when nfsiod(s) are running, queue the request by
2930 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2934 nfs_strategy(struct vop_strategy_args *ap)
2936 struct buf *bp = ap->a_bp;
2939 KASSERT(!(bp->b_flags & B_DONE),
2940 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2941 BUF_ASSERT_HELD(bp);
2943 if (bp->b_iocmd == BIO_READ)
2949 * If the op is asynchronous and an i/o daemon is waiting
2950 * queue the request, wake it up and wait for completion
2951 * otherwise just do it ourselves.
2953 if ((bp->b_flags & B_ASYNC) == 0 ||
2954 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2955 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2960 * fsync vnode op. Just call nfs_flush() with commit == 1.
2964 nfs_fsync(struct vop_fsync_args *ap)
2967 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1));
2971 * Flush all the blocks associated with a vnode.
2972 * Walk through the buffer pool and push any dirty pages
2973 * associated with the vnode.
2976 nfs_flush(struct vnode *vp, int waitfor, int commit)
2978 struct nfsnode *np = VTONFS(vp);
2982 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2983 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2985 u_quad_t off, endoff, toff;
2986 struct ucred* wcred = NULL;
2987 struct buf **bvec = NULL;
2989 struct thread *td = curthread;
2990 #ifndef NFS_COMMITBVECSIZ
2991 #define NFS_COMMITBVECSIZ 20
2993 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2994 int bvecsize = 0, bveccount;
2996 if (nmp->nm_flag & NFSMNT_INT)
2997 slpflag = NFS_PCATCH;
3002 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
3003 * server, but has not been committed to stable storage on the server
3004 * yet. On the first pass, the byte range is worked out and the commit
3005 * rpc is done. On the second pass, nfs_writebp() is called to do the
3012 if (NFS_ISV3(vp) && commit) {
3013 if (bvec != NULL && bvec != bvec_on_stack)
3016 * Count up how many buffers waiting for a commit.
3020 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3021 if (!BUF_ISLOCKED(bp) &&
3022 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
3023 == (B_DELWRI | B_NEEDCOMMIT))
3027 * Allocate space to remember the list of bufs to commit. It is
3028 * important to use M_NOWAIT here to avoid a race with nfs_write.
3029 * If we can't get memory (for whatever reason), we will end up
3030 * committing the buffers one-by-one in the loop below.
3032 if (bveccount > NFS_COMMITBVECSIZ) {
3034 * Release the vnode interlock to avoid a lock
3038 bvec = (struct buf **)
3039 malloc(bveccount * sizeof(struct buf *),
3043 bvec = bvec_on_stack;
3044 bvecsize = NFS_COMMITBVECSIZ;
3046 bvecsize = bveccount;
3048 bvec = bvec_on_stack;
3049 bvecsize = NFS_COMMITBVECSIZ;
3051 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3052 if (bvecpos >= bvecsize)
3054 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3055 nbp = TAILQ_NEXT(bp, b_bobufs);
3058 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3059 (B_DELWRI | B_NEEDCOMMIT)) {
3061 nbp = TAILQ_NEXT(bp, b_bobufs);
3067 * Work out if all buffers are using the same cred
3068 * so we can deal with them all with one commit.
3070 * NOTE: we are not clearing B_DONE here, so we have
3071 * to do it later on in this routine if we intend to
3072 * initiate I/O on the bp.
3074 * Note: to avoid loopback deadlocks, we do not
3075 * assign b_runningbufspace.
3078 wcred = bp->b_wcred;
3079 else if (wcred != bp->b_wcred)
3081 vfs_busy_pages(bp, 1);
3085 * bp is protected by being locked, but nbp is not
3086 * and vfs_busy_pages() may sleep. We have to
3089 nbp = TAILQ_NEXT(bp, b_bobufs);
3092 * A list of these buffers is kept so that the
3093 * second loop knows which buffers have actually
3094 * been committed. This is necessary, since there
3095 * may be a race between the commit rpc and new
3096 * uncommitted writes on the file.
3098 bvec[bvecpos++] = bp;
3099 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3103 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3111 * Commit data on the server, as required.
3112 * If all bufs are using the same wcred, then use that with
3113 * one call for all of them, otherwise commit each one
3116 if (wcred != NOCRED)
3117 retv = nfs_commit(vp, off, (int)(endoff - off),
3121 for (i = 0; i < bvecpos; i++) {
3124 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3126 size = (u_quad_t)(bp->b_dirtyend
3128 retv = nfs_commit(vp, off, (int)size,
3134 if (retv == NFSERR_STALEWRITEVERF)
3135 nfs_clearcommit(vp->v_mount);
3138 * Now, either mark the blocks I/O done or mark the
3139 * blocks dirty, depending on whether the commit
3142 for (i = 0; i < bvecpos; i++) {
3144 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3147 * Error, leave B_DELWRI intact
3149 vfs_unbusy_pages(bp);
3153 * Success, remove B_DELWRI ( bundirty() ).
3155 * b_dirtyoff/b_dirtyend seem to be NFS
3156 * specific. We should probably move that
3157 * into bundirty(). XXX
3160 bp->b_flags |= B_ASYNC;
3162 bp->b_flags &= ~B_DONE;
3163 bp->b_ioflags &= ~BIO_ERROR;
3164 bp->b_dirtyoff = bp->b_dirtyend = 0;
3171 * Start/do any write(s) that are required.
3175 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3176 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3177 if (waitfor != MNT_WAIT || passone)
3180 error = BUF_TIMELOCK(bp,
3181 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3182 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
3187 if (error == ENOLCK) {
3191 if (nfs_sigintr(nmp, td)) {
3195 if (slpflag & PCATCH) {
3201 if ((bp->b_flags & B_DELWRI) == 0)
3202 panic("nfs_fsync: not dirty");
3203 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3209 if (passone || !commit)
3210 bp->b_flags |= B_ASYNC;
3212 bp->b_flags |= B_ASYNC;
3214 if (nfs_sigintr(nmp, td)) {
3225 if (waitfor == MNT_WAIT) {
3226 while (bo->bo_numoutput) {
3227 error = bufobj_wwait(bo, slpflag, slptimeo);
3230 error = nfs_sigintr(nmp, td);
3233 if (slpflag & PCATCH) {
3240 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3245 * Wait for all the async IO requests to drain
3248 mtx_lock(&np->n_mtx);
3249 while (np->n_directio_asyncwr > 0) {
3250 np->n_flag |= NFSYNCWAIT;
3251 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3252 &np->n_mtx, slpflag | (PRIBIO + 1),
3255 if (nfs_sigintr(nmp, td)) {
3256 mtx_unlock(&np->n_mtx);
3262 mtx_unlock(&np->n_mtx);
3265 mtx_lock(&np->n_mtx);
3266 if (np->n_flag & NWRITEERR) {
3267 error = np->n_error;
3268 np->n_flag &= ~NWRITEERR;
3270 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3271 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3272 np->n_flag &= ~NMODIFIED;
3273 mtx_unlock(&np->n_mtx);
3275 if (bvec != NULL && bvec != bvec_on_stack)
3281 * NFS advisory byte-level locks.
3284 nfs_advlock(struct vop_advlock_args *ap)
3286 struct vnode *vp = ap->a_vp;
3290 error = vn_lock(vp, LK_SHARED);
3293 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3294 size = VTONFS(vp)->n_size;
3296 error = lf_advlock(ap, &(vp->v_lockf), size);
3299 error = nfs_advlock_p(ap);
3308 * NFS advisory byte-level locks.
3311 nfs_advlockasync(struct vop_advlockasync_args *ap)
3313 struct vnode *vp = ap->a_vp;
3317 error = vn_lock(vp, LK_SHARED);
3320 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3321 size = VTONFS(vp)->n_size;
3323 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3332 * Print out the contents of an nfsnode.
3335 nfs_print(struct vop_print_args *ap)
3337 struct vnode *vp = ap->a_vp;
3338 struct nfsnode *np = VTONFS(vp);
3340 nfs_printf("\tfileid %ld fsid 0x%x",
3341 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3342 if (vp->v_type == VFIFO)
3349 * This is the "real" nfs::bwrite(struct buf*).
3350 * We set B_CACHE if this is a VMIO buffer.
3353 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3356 int oldflags = bp->b_flags;
3362 BUF_ASSERT_HELD(bp);
3364 if (bp->b_flags & B_INVAL) {
3369 bp->b_flags |= B_CACHE;
3372 * Undirty the bp. We will redirty it later if the I/O fails.
3377 bp->b_flags &= ~B_DONE;
3378 bp->b_ioflags &= ~BIO_ERROR;
3379 bp->b_iocmd = BIO_WRITE;
3381 bufobj_wref(bp->b_bufobj);
3382 curthread->td_ru.ru_oublock++;
3386 * Note: to avoid loopback deadlocks, we do not
3387 * assign b_runningbufspace.
3389 vfs_busy_pages(bp, 1);
3392 bp->b_iooffset = dbtob(bp->b_blkno);
3395 if( (oldflags & B_ASYNC) == 0) {
3396 int rtval = bufwait(bp);
3398 if (oldflags & B_DELWRI) {
3411 * nfs special file access vnode op.
3412 * Essentially just get vattr and then imitate iaccess() since the device is
3413 * local to the client.
3416 nfsspec_access(struct vop_access_args *ap)
3419 struct ucred *cred = ap->a_cred;
3420 struct vnode *vp = ap->a_vp;
3421 accmode_t accmode = ap->a_accmode;
3426 * Disallow write attempts on filesystems mounted read-only;
3427 * unless the file is a socket, fifo, or a block or character
3428 * device resident on the filesystem.
3430 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3431 switch (vp->v_type) {
3441 error = VOP_GETATTR(vp, vap, cred);
3444 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3445 accmode, cred, NULL);
3451 * Read wrapper for fifos.
3454 nfsfifo_read(struct vop_read_args *ap)
3456 struct nfsnode *np = VTONFS(ap->a_vp);
3462 mtx_lock(&np->n_mtx);
3464 vfs_timestamp(&np->n_atim);
3465 mtx_unlock(&np->n_mtx);
3466 error = fifo_specops.vop_read(ap);
3471 * Write wrapper for fifos.
3474 nfsfifo_write(struct vop_write_args *ap)
3476 struct nfsnode *np = VTONFS(ap->a_vp);
3481 mtx_lock(&np->n_mtx);
3483 vfs_timestamp(&np->n_mtim);
3484 mtx_unlock(&np->n_mtx);
3485 return(fifo_specops.vop_write(ap));
3489 * Close wrapper for fifos.
3491 * Update the times on the nfsnode then do fifo close.
3494 nfsfifo_close(struct vop_close_args *ap)
3496 struct vnode *vp = ap->a_vp;
3497 struct nfsnode *np = VTONFS(vp);
3501 mtx_lock(&np->n_mtx);
3502 if (np->n_flag & (NACC | NUPD)) {
3504 if (np->n_flag & NACC)
3506 if (np->n_flag & NUPD)
3509 if (vrefcnt(vp) == 1 &&
3510 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3512 if (np->n_flag & NACC)
3513 vattr.va_atime = np->n_atim;
3514 if (np->n_flag & NUPD)
3515 vattr.va_mtime = np->n_mtim;
3516 mtx_unlock(&np->n_mtx);
3517 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3521 mtx_unlock(&np->n_mtx);
3523 return (fifo_specops.vop_close(ap));
3527 * Just call nfs_writebp() with the force argument set to 1.
3529 * NOTE: B_DONE may or may not be set in a_bp on call.
3532 nfs_bwrite(struct buf *bp)
3535 return (nfs_writebp(bp, 1, curthread));
3538 struct buf_ops buf_ops_nfs = {
3539 .bop_name = "buf_ops_nfs",
3540 .bop_write = nfs_bwrite,
3541 .bop_strategy = bufstrategy,
3542 .bop_sync = bufsync,
3543 .bop_bdflush = bufbdflush,