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 vop_advlock_t *nfs_advlock_p = nfs_dolock;
219 vop_reclaim_t *nfs_reclaim_p = NULL;
220 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
222 SYSCTL_DECL(_vfs_nfs);
224 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
225 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
226 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
228 static int nfs_prime_access_cache = 0;
229 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
230 &nfs_prime_access_cache, 0,
231 "Prime NFS ACCESS cache when fetching attributes");
233 static int nfsv3_commit_on_close = 0;
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
235 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
237 static int nfs_clean_pages_on_close = 1;
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
239 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
241 int nfs_directio_enable = 0;
242 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
243 &nfs_directio_enable, 0, "Enable NFS directio");
246 * This sysctl allows other processes to mmap a file that has been opened
247 * O_DIRECT by a process. In general, having processes mmap the file while
248 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
249 * this by default to prevent DoS attacks - to prevent a malicious user from
250 * opening up files O_DIRECT preventing other users from mmap'ing these
251 * files. "Protected" environments where stricter consistency guarantees are
252 * required can disable this knob. The process that opened the file O_DIRECT
253 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
256 int nfs_directio_allow_mmap = 1;
257 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
258 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
261 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
262 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
264 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
265 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
268 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
269 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
270 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
274 * The list of locks after the description of the lock is the ordering
275 * of other locks acquired with the lock held.
276 * np->n_mtx : Protects the fields in the nfsnode.
278 VI_MTX (acquired indirectly)
279 * nmp->nm_mtx : Protects the fields in the nfsmount.
281 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
282 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
285 * rep->r_mtx : Protects the fields in an nfsreq.
289 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
290 struct ucred *cred, uint32_t *retmode)
294 int error = 0, attrflag, i, lrupos;
296 struct mbuf *mreq, *mrep, *md, *mb;
299 struct nfsnode *np = VTONFS(vp);
301 nfsstats.rpccnt[NFSPROC_ACCESS]++;
302 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
304 bpos = mtod(mb, caddr_t);
306 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
307 *tl = txdr_unsigned(wmode);
308 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
309 nfsm_postop_attr(vp, attrflag);
312 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
313 rmode = fxdr_unsigned(u_int32_t, *tl);
314 mtx_lock(&np->n_mtx);
315 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
316 if (np->n_accesscache[i].uid == cred->cr_uid) {
317 np->n_accesscache[i].mode = rmode;
318 np->n_accesscache[i].stamp = time_second;
321 if (i > 0 && np->n_accesscache[i].stamp <
322 np->n_accesscache[lrupos].stamp)
325 if (i == NFS_ACCESSCACHESIZE) {
326 np->n_accesscache[lrupos].uid = cred->cr_uid;
327 np->n_accesscache[lrupos].mode = rmode;
328 np->n_accesscache[lrupos].stamp = time_second;
330 mtx_unlock(&np->n_mtx);
333 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
339 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
347 * nfs access vnode op.
348 * For nfs version 2, just return ok. File accesses may fail later.
349 * For nfs version 3, use the access rpc to check accessibility. If file modes
350 * are changed on the server, accesses might still fail later.
353 nfs_access(struct vop_access_args *ap)
355 struct vnode *vp = ap->a_vp;
356 int error = 0, i, gotahit;
357 u_int32_t mode, rmode, wmode;
358 int v3 = NFS_ISV3(vp);
359 struct nfsnode *np = VTONFS(vp);
362 * Disallow write attempts on filesystems mounted read-only;
363 * unless the file is a socket, fifo, or a block or character
364 * device resident on the filesystem.
366 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
367 switch (vp->v_type) {
377 * For nfs v3, check to see if we have done this recently, and if
378 * so return our cached result instead of making an ACCESS call.
379 * If not, do an access rpc, otherwise you are stuck emulating
380 * ufs_access() locally using the vattr. This may not be correct,
381 * since the server may apply other access criteria such as
382 * client uid-->server uid mapping that we do not know about.
385 if (ap->a_accmode & VREAD)
386 mode = NFSV3ACCESS_READ;
389 if (vp->v_type != VDIR) {
390 if (ap->a_accmode & VWRITE)
391 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
392 if (ap->a_accmode & VEXEC)
393 mode |= NFSV3ACCESS_EXECUTE;
395 if (ap->a_accmode & VWRITE)
396 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
398 if (ap->a_accmode & VEXEC)
399 mode |= NFSV3ACCESS_LOOKUP;
401 /* XXX safety belt, only make blanket request if caching */
402 if (nfsaccess_cache_timeout > 0) {
403 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
404 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
405 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
411 * Does our cached result allow us to give a definite yes to
415 mtx_lock(&np->n_mtx);
416 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
417 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
418 if (time_second < (np->n_accesscache[i].stamp +
419 nfsaccess_cache_timeout) &&
420 (np->n_accesscache[i].mode & mode) == mode) {
421 nfsstats.accesscache_hits++;
427 mtx_unlock(&np->n_mtx);
430 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
431 ap->a_cred->cr_uid, mode);
433 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
434 ap->a_cred->cr_uid, mode);
438 * Either a no, or a don't know. Go to the wire.
440 nfsstats.accesscache_misses++;
441 error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred,
444 if ((rmode & mode) != mode)
450 if ((error = nfsspec_access(ap)) != 0) {
454 * Attempt to prevent a mapped root from accessing a file
455 * which it shouldn't. We try to read a byte from the file
456 * if the user is root and the file is not zero length.
457 * After calling nfsspec_access, we should have the correct
460 mtx_lock(&np->n_mtx);
461 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
462 && VTONFS(vp)->n_size > 0) {
467 mtx_unlock(&np->n_mtx);
470 auio.uio_iov = &aiov;
474 auio.uio_segflg = UIO_SYSSPACE;
475 auio.uio_rw = UIO_READ;
476 auio.uio_td = ap->a_td;
478 if (vp->v_type == VREG)
479 error = nfs_readrpc(vp, &auio, ap->a_cred);
480 else if (vp->v_type == VDIR) {
482 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
484 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
485 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
487 } else if (vp->v_type == VLNK)
488 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
492 mtx_unlock(&np->n_mtx);
497 int nfs_otw_getattr_avoid = 0;
501 * Check to see if the type is ok
502 * and that deletion is not in progress.
503 * For paged in text files, you will need to flush the page cache
504 * if consistency is lost.
508 nfs_open(struct vop_open_args *ap)
510 struct vnode *vp = ap->a_vp;
511 struct nfsnode *np = VTONFS(vp);
514 int fmode = ap->a_mode;
516 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
520 * Get a valid lease. If cached data is stale, flush it.
522 mtx_lock(&np->n_mtx);
523 if (np->n_flag & NMODIFIED) {
524 mtx_unlock(&np->n_mtx);
525 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
526 if (error == EINTR || error == EIO)
528 mtx_lock(&np->n_mtx);
530 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
531 if (vp->v_type == VDIR)
532 np->n_direofoffset = 0;
533 mtx_unlock(&np->n_mtx);
534 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
537 mtx_lock(&np->n_mtx);
538 np->n_mtime = vattr.va_mtime;
540 mtx_unlock(&np->n_mtx);
541 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
544 mtx_lock(&np->n_mtx);
545 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
546 if (vp->v_type == VDIR)
547 np->n_direofoffset = 0;
548 mtx_unlock(&np->n_mtx);
549 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
550 if (error == EINTR || error == EIO) {
553 mtx_lock(&np->n_mtx);
554 np->n_mtime = vattr.va_mtime;
558 * If the object has >= 1 O_DIRECT active opens, we disable caching.
560 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
561 if (np->n_directio_opens == 0) {
562 mtx_unlock(&np->n_mtx);
563 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
566 mtx_lock(&np->n_mtx);
567 np->n_flag |= NNONCACHE;
569 np->n_directio_opens++;
571 mtx_unlock(&np->n_mtx);
572 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
578 * What an NFS client should do upon close after writing is a debatable issue.
579 * Most NFS clients push delayed writes to the server upon close, basically for
581 * 1 - So that any write errors may be reported back to the client process
582 * doing the close system call. By far the two most likely errors are
583 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
584 * 2 - To put a worst case upper bound on cache inconsistency between
585 * multiple clients for the file.
586 * There is also a consistency problem for Version 2 of the protocol w.r.t.
587 * not being able to tell if other clients are writing a file concurrently,
588 * since there is no way of knowing if the changed modify time in the reply
589 * is only due to the write for this client.
590 * (NFS Version 3 provides weak cache consistency data in the reply that
591 * should be sufficient to detect and handle this case.)
593 * The current code does the following:
594 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
595 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
596 * or commit them (this satisfies 1 and 2 except for the
597 * case where the server crashes after this close but
598 * before the commit RPC, which is felt to be "good
599 * enough". Changing the last argument to nfs_flush() to
600 * a 1 would force a commit operation, if it is felt a
601 * commit is necessary now.
605 nfs_close(struct vop_close_args *ap)
607 struct vnode *vp = ap->a_vp;
608 struct nfsnode *np = VTONFS(vp);
610 int fmode = ap->a_fflag;
612 if (vp->v_type == VREG) {
614 * Examine and clean dirty pages, regardless of NMODIFIED.
615 * This closes a major hole in close-to-open consistency.
616 * We want to push out all dirty pages (and buffers) on
617 * close, regardless of whether they were dirtied by
618 * mmap'ed writes or via write().
620 if (nfs_clean_pages_on_close && vp->v_object) {
621 VM_OBJECT_LOCK(vp->v_object);
622 vm_object_page_clean(vp->v_object, 0, 0, 0);
623 VM_OBJECT_UNLOCK(vp->v_object);
625 mtx_lock(&np->n_mtx);
626 if (np->n_flag & NMODIFIED) {
627 mtx_unlock(&np->n_mtx);
630 * Under NFSv3 we have dirty buffers to dispose of. We
631 * must flush them to the NFS server. We have the option
632 * of waiting all the way through the commit rpc or just
633 * waiting for the initial write. The default is to only
634 * wait through the initial write so the data is in the
635 * server's cache, which is roughly similar to the state
636 * a standard disk subsystem leaves the file in on close().
638 * We cannot clear the NMODIFIED bit in np->n_flag due to
639 * potential races with other processes, and certainly
640 * cannot clear it if we don't commit.
642 int cm = nfsv3_commit_on_close ? 1 : 0;
643 error = nfs_flush(vp, MNT_WAIT, cm);
644 /* np->n_flag &= ~NMODIFIED; */
646 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
647 mtx_lock(&np->n_mtx);
649 if (np->n_flag & NWRITEERR) {
650 np->n_flag &= ~NWRITEERR;
653 mtx_unlock(&np->n_mtx);
655 if (nfs_directio_enable)
656 KASSERT((np->n_directio_asyncwr == 0),
657 ("nfs_close: dirty unflushed (%d) directio buffers\n",
658 np->n_directio_asyncwr));
659 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
660 mtx_lock(&np->n_mtx);
661 KASSERT((np->n_directio_opens > 0),
662 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
663 np->n_directio_opens--;
664 if (np->n_directio_opens == 0)
665 np->n_flag &= ~NNONCACHE;
666 mtx_unlock(&np->n_mtx);
672 * nfs getattr call from vfs.
675 nfs_getattr(struct vop_getattr_args *ap)
677 struct vnode *vp = ap->a_vp;
678 struct nfsnode *np = VTONFS(vp);
679 struct thread *td = curthread;
680 struct vattr *vap = ap->a_vap;
684 struct mbuf *mreq, *mrep, *md, *mb;
685 int v3 = NFS_ISV3(vp);
688 * Update local times for special files.
690 mtx_lock(&np->n_mtx);
691 if (np->n_flag & (NACC | NUPD))
693 mtx_unlock(&np->n_mtx);
695 * First look in the cache.
697 if (nfs_getattrcache(vp, &vattr) == 0)
699 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
700 nfsstats.accesscache_misses++;
701 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL);
702 if (nfs_getattrcache(vp, &vattr) == 0)
705 nfsstats.rpccnt[NFSPROC_GETATTR]++;
706 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
708 bpos = mtod(mb, caddr_t);
710 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
712 nfsm_loadattr(vp, &vattr);
716 vap->va_type = vattr.va_type;
717 vap->va_mode = vattr.va_mode;
718 vap->va_nlink = vattr.va_nlink;
719 vap->va_uid = vattr.va_uid;
720 vap->va_gid = vattr.va_gid;
721 vap->va_fsid = vattr.va_fsid;
722 vap->va_fileid = vattr.va_fileid;
723 vap->va_size = vattr.va_size;
724 vap->va_blocksize = vattr.va_blocksize;
725 vap->va_atime = vattr.va_atime;
726 vap->va_mtime = vattr.va_mtime;
727 vap->va_ctime = vattr.va_ctime;
728 vap->va_gen = vattr.va_gen;
729 vap->va_flags = vattr.va_flags;
730 vap->va_rdev = vattr.va_rdev;
731 vap->va_bytes = vattr.va_bytes;
732 vap->va_filerev = vattr.va_filerev;
741 nfs_setattr(struct vop_setattr_args *ap)
743 struct vnode *vp = ap->a_vp;
744 struct nfsnode *np = VTONFS(vp);
745 struct vattr *vap = ap->a_vap;
746 struct thread *td = curthread;
755 * Setting of flags is not supported.
757 if (vap->va_flags != VNOVAL)
761 * Disallow write attempts if the filesystem is mounted read-only.
763 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
764 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
765 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
766 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
770 if (vap->va_size != VNOVAL) {
771 switch (vp->v_type) {
778 if (vap->va_mtime.tv_sec == VNOVAL &&
779 vap->va_atime.tv_sec == VNOVAL &&
780 vap->va_mode == (mode_t)VNOVAL &&
781 vap->va_uid == (uid_t)VNOVAL &&
782 vap->va_gid == (gid_t)VNOVAL)
784 vap->va_size = VNOVAL;
788 * Disallow write attempts if the filesystem is
791 if (vp->v_mount->mnt_flag & MNT_RDONLY)
794 * We run vnode_pager_setsize() early (why?),
795 * we must set np->n_size now to avoid vinvalbuf
796 * V_SAVE races that might setsize a lower
799 mtx_lock(&np->n_mtx);
801 mtx_unlock(&np->n_mtx);
802 error = nfs_meta_setsize(vp, ap->a_cred, td,
804 mtx_lock(&np->n_mtx);
805 if (np->n_flag & NMODIFIED) {
807 mtx_unlock(&np->n_mtx);
808 if (vap->va_size == 0)
809 error = nfs_vinvalbuf(vp, 0, td, 1);
811 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
813 vnode_pager_setsize(vp, tsize);
817 mtx_unlock(&np->n_mtx);
819 * np->n_size has already been set to vap->va_size
820 * in nfs_meta_setsize(). We must set it again since
821 * nfs_loadattrcache() could be called through
822 * nfs_meta_setsize() and could modify np->n_size.
824 mtx_lock(&np->n_mtx);
825 np->n_vattr.va_size = np->n_size = vap->va_size;
826 mtx_unlock(&np->n_mtx);
829 mtx_lock(&np->n_mtx);
830 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
831 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
832 mtx_unlock(&np->n_mtx);
833 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
834 (error == EINTR || error == EIO))
837 mtx_unlock(&np->n_mtx);
839 error = nfs_setattrrpc(vp, vap, ap->a_cred);
840 if (error && vap->va_size != VNOVAL) {
841 mtx_lock(&np->n_mtx);
842 np->n_size = np->n_vattr.va_size = tsize;
843 vnode_pager_setsize(vp, tsize);
844 mtx_unlock(&np->n_mtx);
851 * Do an nfs setattr rpc.
854 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
856 struct nfsv2_sattr *sp;
857 struct nfsnode *np = VTONFS(vp);
860 int error = 0, i, wccflag = NFSV3_WCCRATTR;
861 struct mbuf *mreq, *mrep, *md, *mb;
862 int v3 = NFS_ISV3(vp);
864 nfsstats.rpccnt[NFSPROC_SETATTR]++;
865 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
867 bpos = mtod(mb, caddr_t);
870 nfsm_v3attrbuild(vap, TRUE);
871 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
874 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
875 if (vap->va_mode == (mode_t)VNOVAL)
876 sp->sa_mode = nfs_xdrneg1;
878 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
879 if (vap->va_uid == (uid_t)VNOVAL)
880 sp->sa_uid = nfs_xdrneg1;
882 sp->sa_uid = txdr_unsigned(vap->va_uid);
883 if (vap->va_gid == (gid_t)VNOVAL)
884 sp->sa_gid = nfs_xdrneg1;
886 sp->sa_gid = txdr_unsigned(vap->va_gid);
887 sp->sa_size = txdr_unsigned(vap->va_size);
888 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
889 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
891 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
893 mtx_lock(&np->n_mtx);
894 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
895 np->n_accesscache[i].stamp = 0;
896 mtx_unlock(&np->n_mtx);
897 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
898 nfsm_wcc_data(vp, wccflag);
900 nfsm_loadattr(vp, NULL);
907 * nfs lookup call, one step at a time...
908 * First look in cache
909 * If not found, unlock the directory nfsnode and do the rpc
912 nfs_lookup(struct vop_lookup_args *ap)
914 struct componentname *cnp = ap->a_cnp;
915 struct vnode *dvp = ap->a_dvp;
916 struct vnode **vpp = ap->a_vpp;
917 struct mount *mp = dvp->v_mount;
919 struct timespec dmtime;
920 int flags = cnp->cn_flags;
922 struct nfsmount *nmp;
924 struct mbuf *mreq, *mrep, *md, *mb;
927 struct nfsnode *np, *newnp;
928 int error = 0, attrflag, fhsize, ltype;
929 int v3 = NFS_ISV3(dvp);
930 struct thread *td = cnp->cn_thread;
933 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
934 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
936 if (dvp->v_type != VDIR)
940 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
944 error = cache_lookup(dvp, vpp, cnp);
945 if (error > 0 && error != ENOENT)
949 * We only accept a positive hit in the cache if the
950 * change time of the file matches our cached copy.
951 * Otherwise, we discard the cache entry and fallback
952 * to doing a lookup RPC.
954 * To better handle stale file handles and attributes,
955 * clear the attribute cache of this node if it is a
956 * leaf component, part of an open() call, and not
957 * locally modified before fetching the attributes.
958 * This should allow stale file handles to be detected
959 * here where we can fall back to a LOOKUP RPC to
960 * recover rather than having nfs_open() detect the
961 * stale file handle and failing open(2) with ESTALE.
964 newnp = VTONFS(newvp);
965 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
966 !(newnp->n_flag & NMODIFIED)) {
967 mtx_lock(&newnp->n_mtx);
968 newnp->n_attrstamp = 0;
969 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
970 mtx_unlock(&newnp->n_mtx);
972 if (VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
973 timespeccmp(&vattr.va_ctime, &newnp->n_ctime, ==)) {
974 nfsstats.lookupcache_hits++;
975 if (cnp->cn_nameiop != LOOKUP &&
977 cnp->cn_flags |= SAVENAME;
986 } else if (error == ENOENT) {
987 if (dvp->v_iflag & VI_DOOMED)
990 * We only accept a negative hit in the cache if the
991 * modification time of the parent directory matches
992 * our cached copy. Otherwise, we discard all of the
993 * negative cache entries for this directory. We also
994 * only trust -ve cache entries for less than
995 * nm_negative_namecache_timeout seconds.
997 if ((u_int)(ticks - np->n_dmtime_ticks) <
998 (nmp->nm_negnametimeo * hz) &&
999 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1000 timespeccmp(&vattr.va_mtime, &np->n_dmtime, ==)) {
1001 nfsstats.lookupcache_hits++;
1004 cache_purge_negative(dvp);
1005 mtx_lock(&np->n_mtx);
1006 timespecclear(&np->n_dmtime);
1007 mtx_unlock(&np->n_mtx);
1011 * Cache the modification time of the parent directory in case
1012 * the lookup fails and results in adding the first negative
1013 * name cache entry for the directory. Since this is reading
1014 * a single time_t, don't bother with locking. The
1015 * modification time may be a bit stale, but it must be read
1016 * before performing the lookup RPC to prevent a race where
1017 * another lookup updates the timestamp on the directory after
1018 * the lookup RPC has been performed on the server but before
1019 * n_dmtime is set at the end of this function.
1021 dmtime = np->n_vattr.va_mtime;
1024 nfsstats.lookupcache_misses++;
1025 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1026 len = cnp->cn_namelen;
1027 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1028 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1030 bpos = mtod(mb, caddr_t);
1031 nfsm_fhtom(dvp, v3);
1032 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1033 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
1036 nfsm_postop_attr(dvp, attrflag);
1041 nfsm_getfh(fhp, fhsize, v3);
1044 * Handle RENAME case...
1046 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1047 if (NFS_CMPFH(np, fhp, fhsize)) {
1051 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE);
1058 nfsm_postop_attr(newvp, attrflag);
1059 nfsm_postop_attr(dvp, attrflag);
1061 nfsm_loadattr(newvp, NULL);
1064 cnp->cn_flags |= SAVENAME;
1068 if (flags & ISDOTDOT) {
1069 ltype = VOP_ISLOCKED(dvp);
1070 error = vfs_busy(mp, MBF_NOWAIT);
1074 error = vfs_busy(mp, 0);
1075 vn_lock(dvp, ltype | LK_RETRY);
1077 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1087 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1092 vn_lock(dvp, ltype | LK_RETRY);
1093 if (dvp->v_iflag & VI_DOOMED) {
1106 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1110 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1118 * Flush the attribute cache when opening a leaf node
1119 * to ensure that fresh attributes are fetched in
1120 * nfs_open() if we are unable to fetch attributes
1121 * from the LOOKUP reply.
1123 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1124 !(np->n_flag & NMODIFIED)) {
1125 mtx_lock(&np->n_mtx);
1126 np->n_attrstamp = 0;
1127 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1128 mtx_unlock(&np->n_mtx);
1132 nfsm_postop_attr(newvp, attrflag);
1133 nfsm_postop_attr(dvp, attrflag);
1135 nfsm_loadattr(newvp, NULL);
1136 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1137 cnp->cn_flags |= SAVENAME;
1138 if ((cnp->cn_flags & MAKEENTRY) &&
1139 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1140 np->n_ctime = np->n_vattr.va_ctime;
1141 cache_enter(dvp, newvp, cnp);
1147 if (newvp != NULLVP) {
1152 if (error != ENOENT)
1155 /* The requested file was not found. */
1156 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1157 (flags & ISLASTCN)) {
1159 * XXX: UFS does a full VOP_ACCESS(dvp,
1160 * VWRITE) here instead of just checking
1163 if (mp->mnt_flag & MNT_RDONLY)
1165 cnp->cn_flags |= SAVENAME;
1166 return (EJUSTRETURN);
1169 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1171 * Maintain n_dmtime as the modification time
1172 * of the parent directory when the oldest -ve
1173 * name cache entry for this directory was
1174 * added. If a -ve cache entry has already
1175 * been added with a newer modification time
1176 * by a concurrent lookup, then don't bother
1177 * adding a cache entry. The modification
1178 * time of the directory might have changed
1179 * due to the file this lookup failed to find
1180 * being created. In that case a subsequent
1181 * lookup would incorrectly use the entry
1182 * added here instead of doing an extra
1185 mtx_lock(&np->n_mtx);
1186 if (timespeccmp(&np->n_dmtime, &dmtime, <=)) {
1187 if (!timespecisset(&np->n_dmtime)) {
1188 np->n_dmtime = dmtime;
1189 np->n_dmtime_ticks = ticks;
1191 mtx_unlock(&np->n_mtx);
1192 cache_enter(dvp, NULL, cnp);
1194 mtx_unlock(&np->n_mtx);
1204 * Just call nfs_bioread() to do the work.
1207 nfs_read(struct vop_read_args *ap)
1209 struct vnode *vp = ap->a_vp;
1211 switch (vp->v_type) {
1213 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1217 return (EOPNOTSUPP);
1225 nfs_readlink(struct vop_readlink_args *ap)
1227 struct vnode *vp = ap->a_vp;
1229 if (vp->v_type != VLNK)
1231 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1235 * Do a readlink rpc.
1236 * Called by nfs_doio() from below the buffer cache.
1239 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1242 int error = 0, len, attrflag;
1243 struct mbuf *mreq, *mrep, *md, *mb;
1244 int v3 = NFS_ISV3(vp);
1246 nfsstats.rpccnt[NFSPROC_READLINK]++;
1247 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1249 bpos = mtod(mb, caddr_t);
1251 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1253 nfsm_postop_attr(vp, attrflag);
1255 nfsm_strsiz(len, NFS_MAXPATHLEN);
1256 if (len == NFS_MAXPATHLEN) {
1257 struct nfsnode *np = VTONFS(vp);
1258 mtx_lock(&np->n_mtx);
1259 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1261 mtx_unlock(&np->n_mtx);
1263 nfsm_mtouio(uiop, len);
1275 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1279 struct mbuf *mreq, *mrep, *md, *mb;
1280 struct nfsmount *nmp;
1281 int error = 0, len, retlen, tsiz, eof, attrflag;
1282 int v3 = NFS_ISV3(vp);
1288 nmp = VFSTONFS(vp->v_mount);
1289 tsiz = uiop->uio_resid;
1290 mtx_lock(&nmp->nm_mtx);
1291 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1292 mtx_unlock(&nmp->nm_mtx);
1295 rsize = nmp->nm_rsize;
1296 mtx_unlock(&nmp->nm_mtx);
1298 nfsstats.rpccnt[NFSPROC_READ]++;
1299 len = (tsiz > rsize) ? rsize : tsiz;
1300 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1302 bpos = mtod(mb, caddr_t);
1304 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1306 txdr_hyper(uiop->uio_offset, tl);
1307 *(tl + 2) = txdr_unsigned(len);
1309 *tl++ = txdr_unsigned(uiop->uio_offset);
1310 *tl++ = txdr_unsigned(len);
1313 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1315 nfsm_postop_attr(vp, attrflag);
1320 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1321 eof = fxdr_unsigned(int, *(tl + 1));
1323 nfsm_loadattr(vp, NULL);
1325 nfsm_strsiz(retlen, rsize);
1326 nfsm_mtouio(uiop, retlen);
1330 if (eof || retlen == 0) {
1333 } else if (retlen < len) {
1345 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1346 int *iomode, int *must_commit)
1351 struct mbuf *mreq, *mrep, *md, *mb;
1352 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1353 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1354 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1357 KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1"));
1359 tsiz = uiop->uio_resid;
1360 mtx_lock(&nmp->nm_mtx);
1361 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1362 mtx_unlock(&nmp->nm_mtx);
1365 wsize = nmp->nm_wsize;
1366 mtx_unlock(&nmp->nm_mtx);
1368 nfsstats.rpccnt[NFSPROC_WRITE]++;
1369 len = (tsiz > wsize) ? wsize : tsiz;
1370 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1371 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1373 bpos = mtod(mb, caddr_t);
1376 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1377 txdr_hyper(uiop->uio_offset, tl);
1379 *tl++ = txdr_unsigned(len);
1380 *tl++ = txdr_unsigned(*iomode);
1381 *tl = txdr_unsigned(len);
1385 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1386 /* Set both "begin" and "current" to non-garbage. */
1387 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1388 *tl++ = x; /* "begin offset" */
1389 *tl++ = x; /* "current offset" */
1390 x = txdr_unsigned(len);
1391 *tl++ = x; /* total to this offset */
1392 *tl = x; /* size of this write */
1394 nfsm_uiotom(uiop, len);
1395 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1397 wccflag = NFSV3_WCCCHK;
1398 nfsm_wcc_data(vp, wccflag);
1400 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1401 + NFSX_V3WRITEVERF);
1402 rlen = fxdr_unsigned(int, *tl++);
1407 } else if (rlen < len) {
1408 backup = len - rlen;
1409 uiop->uio_iov->iov_base =
1410 (char *)uiop->uio_iov->iov_base -
1412 uiop->uio_iov->iov_len += backup;
1413 uiop->uio_offset -= backup;
1414 uiop->uio_resid += backup;
1417 commit = fxdr_unsigned(int, *tl++);
1420 * Return the lowest committment level
1421 * obtained by any of the RPCs.
1423 if (committed == NFSV3WRITE_FILESYNC)
1425 else if (committed == NFSV3WRITE_DATASYNC &&
1426 commit == NFSV3WRITE_UNSTABLE)
1428 mtx_lock(&nmp->nm_mtx);
1429 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1430 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1432 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1433 } else if (bcmp((caddr_t)tl,
1434 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1436 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1439 mtx_unlock(&nmp->nm_mtx);
1442 nfsm_loadattr(vp, NULL);
1445 mtx_lock(&(VTONFS(vp))->n_mtx);
1446 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1447 mtx_unlock(&(VTONFS(vp))->n_mtx);
1455 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1456 committed = NFSV3WRITE_FILESYNC;
1457 *iomode = committed;
1459 uiop->uio_resid = tsiz;
1465 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1466 * mode set to specify the file type and the size field for rdev.
1469 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1472 struct nfsv2_sattr *sp;
1474 struct vnode *newvp = NULL;
1475 struct nfsnode *np = NULL;
1478 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1479 struct mbuf *mreq, *mrep, *md, *mb;
1481 int v3 = NFS_ISV3(dvp);
1483 if (vap->va_type == VCHR || vap->va_type == VBLK)
1484 rdev = txdr_unsigned(vap->va_rdev);
1485 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1488 return (EOPNOTSUPP);
1490 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1492 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1493 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1494 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1496 bpos = mtod(mb, caddr_t);
1497 nfsm_fhtom(dvp, v3);
1498 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1500 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1501 *tl++ = vtonfsv3_type(vap->va_type);
1502 nfsm_v3attrbuild(vap, FALSE);
1503 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1504 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1505 *tl++ = txdr_unsigned(major(vap->va_rdev));
1506 *tl = txdr_unsigned(minor(vap->va_rdev));
1509 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1510 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1511 sp->sa_uid = nfs_xdrneg1;
1512 sp->sa_gid = nfs_xdrneg1;
1514 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1515 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1517 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1519 nfsm_mtofh(dvp, newvp, v3, gotvp);
1525 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1526 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1532 nfsm_wcc_data(dvp, wccflag);
1539 if (cnp->cn_flags & MAKEENTRY)
1540 cache_enter(dvp, newvp, cnp);
1543 mtx_lock(&(VTONFS(dvp))->n_mtx);
1544 VTONFS(dvp)->n_flag |= NMODIFIED;
1546 VTONFS(dvp)->n_attrstamp = 0;
1547 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1549 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1555 * just call nfs_mknodrpc() to do the work.
1559 nfs_mknod(struct vop_mknod_args *ap)
1561 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1564 static u_long create_verf;
1566 * nfs file create call
1569 nfs_create(struct vop_create_args *ap)
1571 struct vnode *dvp = ap->a_dvp;
1572 struct vattr *vap = ap->a_vap;
1573 struct componentname *cnp = ap->a_cnp;
1574 struct nfsv2_sattr *sp;
1576 struct nfsnode *np = NULL;
1577 struct vnode *newvp = NULL;
1579 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1580 struct mbuf *mreq, *mrep, *md, *mb;
1582 int v3 = NFS_ISV3(dvp);
1585 * Oops, not for me..
1587 if (vap->va_type == VSOCK) {
1588 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1592 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) {
1595 if (vap->va_vaflags & VA_EXCLUSIVE)
1598 nfsstats.rpccnt[NFSPROC_CREATE]++;
1599 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1600 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1602 bpos = mtod(mb, caddr_t);
1603 nfsm_fhtom(dvp, v3);
1604 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1606 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1607 if (fmode & O_EXCL) {
1608 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1609 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1611 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1613 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1614 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1617 *tl++ = create_verf;
1619 IN_IFADDR_RUNLOCK();
1622 *tl = ++create_verf;
1624 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1625 nfsm_v3attrbuild(vap, FALSE);
1628 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1629 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1630 sp->sa_uid = nfs_xdrneg1;
1631 sp->sa_gid = nfs_xdrneg1;
1633 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1634 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1636 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1638 nfsm_mtofh(dvp, newvp, v3, gotvp);
1644 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1645 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1651 nfsm_wcc_data(dvp, wccflag);
1655 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1661 } else if (v3 && (fmode & O_EXCL)) {
1663 * We are normally called with only a partially initialized
1664 * VAP. Since the NFSv3 spec says that server may use the
1665 * file attributes to store the verifier, the spec requires
1666 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1667 * in atime, but we can't really assume that all servers will
1668 * so we ensure that our SETATTR sets both atime and mtime.
1670 if (vap->va_mtime.tv_sec == VNOVAL)
1671 vfs_timestamp(&vap->va_mtime);
1672 if (vap->va_atime.tv_sec == VNOVAL)
1673 vap->va_atime = vap->va_mtime;
1674 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred);
1679 if (cnp->cn_flags & MAKEENTRY)
1680 cache_enter(dvp, newvp, cnp);
1683 mtx_lock(&(VTONFS(dvp))->n_mtx);
1684 VTONFS(dvp)->n_flag |= NMODIFIED;
1686 VTONFS(dvp)->n_attrstamp = 0;
1687 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1689 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1694 * nfs file remove call
1695 * To try and make nfs semantics closer to ufs semantics, a file that has
1696 * other processes using the vnode is renamed instead of removed and then
1697 * removed later on the last close.
1698 * - If v_usecount > 1
1699 * If a rename is not already in the works
1700 * call nfs_sillyrename() to set it up
1705 nfs_remove(struct vop_remove_args *ap)
1707 struct vnode *vp = ap->a_vp;
1708 struct vnode *dvp = ap->a_dvp;
1709 struct componentname *cnp = ap->a_cnp;
1710 struct nfsnode *np = VTONFS(vp);
1714 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1715 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1716 if (vp->v_type == VDIR)
1718 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1719 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1721 * Purge the name cache so that the chance of a lookup for
1722 * the name succeeding while the remove is in progress is
1723 * minimized. Without node locking it can still happen, such
1724 * that an I/O op returns ESTALE, but since you get this if
1725 * another host removes the file..
1729 * throw away biocache buffers, mainly to avoid
1730 * unnecessary delayed writes later.
1732 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1734 if (error != EINTR && error != EIO)
1735 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1736 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1738 * Kludge City: If the first reply to the remove rpc is lost..
1739 * the reply to the retransmitted request will be ENOENT
1740 * since the file was in fact removed
1741 * Therefore, we cheat and return success.
1743 if (error == ENOENT)
1745 } else if (!np->n_sillyrename)
1746 error = nfs_sillyrename(dvp, vp, cnp);
1747 mtx_lock(&np->n_mtx);
1748 np->n_attrstamp = 0;
1749 mtx_unlock(&np->n_mtx);
1750 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1755 * nfs file remove rpc called from nfs_inactive
1758 nfs_removeit(struct sillyrename *sp)
1761 * Make sure that the directory vnode is still valid.
1762 * XXX we should lock sp->s_dvp here.
1764 if (sp->s_dvp->v_type == VBAD)
1766 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1771 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1774 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1775 struct ucred *cred, struct thread *td)
1778 int error = 0, wccflag = NFSV3_WCCRATTR;
1779 struct mbuf *mreq, *mrep, *md, *mb;
1780 int v3 = NFS_ISV3(dvp);
1782 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1783 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1784 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1786 bpos = mtod(mb, caddr_t);
1787 nfsm_fhtom(dvp, v3);
1788 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1789 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1791 nfsm_wcc_data(dvp, wccflag);
1794 mtx_lock(&(VTONFS(dvp))->n_mtx);
1795 VTONFS(dvp)->n_flag |= NMODIFIED;
1797 VTONFS(dvp)->n_attrstamp = 0;
1798 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1800 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1805 * nfs file rename call
1808 nfs_rename(struct vop_rename_args *ap)
1810 struct vnode *fvp = ap->a_fvp;
1811 struct vnode *tvp = ap->a_tvp;
1812 struct vnode *fdvp = ap->a_fdvp;
1813 struct vnode *tdvp = ap->a_tdvp;
1814 struct componentname *tcnp = ap->a_tcnp;
1815 struct componentname *fcnp = ap->a_fcnp;
1818 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1819 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1820 /* Check for cross-device rename */
1821 if ((fvp->v_mount != tdvp->v_mount) ||
1822 (tvp && (fvp->v_mount != tvp->v_mount))) {
1828 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1832 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1836 * We have to flush B_DELWRI data prior to renaming
1837 * the file. If we don't, the delayed-write buffers
1838 * can be flushed out later after the file has gone stale
1839 * under NFSV3. NFSV2 does not have this problem because
1840 * ( as far as I can tell ) it flushes dirty buffers more
1843 * Skip the rename operation if the fsync fails, this can happen
1844 * due to the server's volume being full, when we pushed out data
1845 * that was written back to our cache earlier. Not checking for
1846 * this condition can result in potential (silent) data loss.
1848 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1851 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1856 * If the tvp exists and is in use, sillyrename it before doing the
1857 * rename of the new file over it.
1858 * XXX Can't sillyrename a directory.
1860 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1861 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1866 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1867 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1870 if (fvp->v_type == VDIR) {
1871 if (tvp != NULL && tvp->v_type == VDIR)
1886 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1888 if (error == ENOENT)
1894 * nfs file rename rpc called from nfs_remove() above
1897 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1898 struct sillyrename *sp)
1901 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1902 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1906 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1909 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1910 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1914 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1915 struct mbuf *mreq, *mrep, *md, *mb;
1916 int v3 = NFS_ISV3(fdvp);
1918 nfsstats.rpccnt[NFSPROC_RENAME]++;
1919 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1920 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1921 nfsm_rndup(tnamelen));
1923 bpos = mtod(mb, caddr_t);
1924 nfsm_fhtom(fdvp, v3);
1925 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1926 nfsm_fhtom(tdvp, v3);
1927 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1928 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1930 nfsm_wcc_data(fdvp, fwccflag);
1931 nfsm_wcc_data(tdvp, twccflag);
1935 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1936 VTONFS(fdvp)->n_flag |= NMODIFIED;
1937 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1938 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1939 VTONFS(tdvp)->n_flag |= NMODIFIED;
1940 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1942 VTONFS(fdvp)->n_attrstamp = 0;
1943 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1946 VTONFS(tdvp)->n_attrstamp = 0;
1947 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1953 * nfs hard link create call
1956 nfs_link(struct vop_link_args *ap)
1958 struct vnode *vp = ap->a_vp;
1959 struct vnode *tdvp = ap->a_tdvp;
1960 struct componentname *cnp = ap->a_cnp;
1962 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1963 struct mbuf *mreq, *mrep, *md, *mb;
1966 if (vp->v_mount != tdvp->v_mount) {
1971 * Push all writes to the server, so that the attribute cache
1972 * doesn't get "out of sync" with the server.
1973 * XXX There should be a better way!
1975 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1978 nfsstats.rpccnt[NFSPROC_LINK]++;
1979 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1980 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1982 bpos = mtod(mb, caddr_t);
1984 nfsm_fhtom(tdvp, v3);
1985 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1986 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1988 nfsm_postop_attr(vp, attrflag);
1989 nfsm_wcc_data(tdvp, wccflag);
1993 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1994 VTONFS(tdvp)->n_flag |= NMODIFIED;
1995 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1997 VTONFS(vp)->n_attrstamp = 0;
1998 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2001 VTONFS(tdvp)->n_attrstamp = 0;
2002 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2008 * nfs symbolic link create call
2011 nfs_symlink(struct vop_symlink_args *ap)
2013 struct vnode *dvp = ap->a_dvp;
2014 struct vattr *vap = ap->a_vap;
2015 struct componentname *cnp = ap->a_cnp;
2016 struct nfsv2_sattr *sp;
2018 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2019 struct mbuf *mreq, *mrep, *md, *mb;
2020 struct vnode *newvp = NULL;
2021 int v3 = NFS_ISV3(dvp);
2023 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2024 slen = strlen(ap->a_target);
2025 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2026 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2028 bpos = mtod(mb, caddr_t);
2029 nfsm_fhtom(dvp, v3);
2030 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2032 nfsm_v3attrbuild(vap, FALSE);
2034 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2036 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2037 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2038 sp->sa_uid = nfs_xdrneg1;
2039 sp->sa_gid = nfs_xdrneg1;
2040 sp->sa_size = nfs_xdrneg1;
2041 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2042 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2046 * Issue the NFS request and get the rpc response.
2048 * Only NFSv3 responses returning an error of 0 actually return
2049 * a file handle that can be converted into newvp without having
2050 * to do an extra lookup rpc.
2052 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
2055 nfsm_mtofh(dvp, newvp, v3, gotvp);
2056 nfsm_wcc_data(dvp, wccflag);
2060 * out code jumps -> here, mrep is also freed.
2067 * If we do not have an error and we could not extract the newvp from
2068 * the response due to the request being NFSv2, we have to do a
2069 * lookup in order to obtain a newvp to return.
2071 if (error == 0 && newvp == NULL) {
2072 struct nfsnode *np = NULL;
2074 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2075 cnp->cn_cred, cnp->cn_thread, &np);
2085 mtx_lock(&(VTONFS(dvp))->n_mtx);
2086 VTONFS(dvp)->n_flag |= NMODIFIED;
2087 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2089 VTONFS(dvp)->n_attrstamp = 0;
2090 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2099 nfs_mkdir(struct vop_mkdir_args *ap)
2101 struct vnode *dvp = ap->a_dvp;
2102 struct vattr *vap = ap->a_vap;
2103 struct componentname *cnp = ap->a_cnp;
2104 struct nfsv2_sattr *sp;
2106 struct nfsnode *np = NULL;
2107 struct vnode *newvp = NULL;
2109 int error = 0, wccflag = NFSV3_WCCRATTR;
2111 struct mbuf *mreq, *mrep, *md, *mb;
2113 int v3 = NFS_ISV3(dvp);
2115 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2117 len = cnp->cn_namelen;
2118 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2119 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
2120 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2122 bpos = mtod(mb, caddr_t);
2123 nfsm_fhtom(dvp, v3);
2124 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2126 nfsm_v3attrbuild(vap, FALSE);
2128 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2129 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2130 sp->sa_uid = nfs_xdrneg1;
2131 sp->sa_gid = nfs_xdrneg1;
2132 sp->sa_size = nfs_xdrneg1;
2133 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2134 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2136 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2138 nfsm_mtofh(dvp, newvp, v3, gotvp);
2140 nfsm_wcc_data(dvp, wccflag);
2143 mtx_lock(&(VTONFS(dvp))->n_mtx);
2144 VTONFS(dvp)->n_flag |= NMODIFIED;
2145 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2147 VTONFS(dvp)->n_attrstamp = 0;
2148 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2150 if (error == 0 && newvp == NULL) {
2151 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2152 cnp->cn_thread, &np);
2155 if (newvp->v_type != VDIR)
2168 * nfs remove directory call
2171 nfs_rmdir(struct vop_rmdir_args *ap)
2173 struct vnode *vp = ap->a_vp;
2174 struct vnode *dvp = ap->a_dvp;
2175 struct componentname *cnp = ap->a_cnp;
2177 int error = 0, wccflag = NFSV3_WCCRATTR;
2178 struct mbuf *mreq, *mrep, *md, *mb;
2179 int v3 = NFS_ISV3(dvp);
2183 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2184 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
2185 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2187 bpos = mtod(mb, caddr_t);
2188 nfsm_fhtom(dvp, v3);
2189 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2190 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2192 nfsm_wcc_data(dvp, wccflag);
2195 mtx_lock(&(VTONFS(dvp))->n_mtx);
2196 VTONFS(dvp)->n_flag |= NMODIFIED;
2197 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2199 VTONFS(dvp)->n_attrstamp = 0;
2200 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2205 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2207 if (error == ENOENT)
2216 nfs_readdir(struct vop_readdir_args *ap)
2218 struct vnode *vp = ap->a_vp;
2219 struct nfsnode *np = VTONFS(vp);
2220 struct uio *uio = ap->a_uio;
2221 int tresid, error = 0;
2224 if (vp->v_type != VDIR)
2228 * First, check for hit on the EOF offset cache
2230 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2231 (np->n_flag & NMODIFIED) == 0) {
2232 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2233 mtx_lock(&np->n_mtx);
2234 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2235 mtx_unlock(&np->n_mtx);
2236 nfsstats.direofcache_hits++;
2239 mtx_unlock(&np->n_mtx);
2244 * Call nfs_bioread() to do the real work.
2246 tresid = uio->uio_resid;
2247 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2249 if (!error && uio->uio_resid == tresid) {
2250 nfsstats.direofcache_misses++;
2258 * Called from below the buffer cache by nfs_doio().
2261 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2264 struct dirent *dp = NULL;
2269 struct mbuf *mreq, *mrep, *md, *mb;
2271 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2272 struct nfsnode *dnp = VTONFS(vp);
2274 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2276 int v3 = NFS_ISV3(vp);
2278 KASSERT(uiop->uio_iovcnt == 1 &&
2279 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2280 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2281 ("nfs readdirrpc bad uio"));
2284 * If there is no cookie, assume directory was stale.
2286 nfs_dircookie_lock(dnp);
2287 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2290 nfs_dircookie_unlock(dnp);
2292 nfs_dircookie_unlock(dnp);
2293 return (NFSERR_BAD_COOKIE);
2297 * Loop around doing readdir rpc's of size nm_readdirsize
2298 * truncated to a multiple of DIRBLKSIZ.
2299 * The stopping criteria is EOF or buffer full.
2301 while (more_dirs && bigenough) {
2302 nfsstats.rpccnt[NFSPROC_READDIR]++;
2303 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2306 bpos = mtod(mb, caddr_t);
2309 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2310 *tl++ = cookie.nfsuquad[0];
2311 *tl++ = cookie.nfsuquad[1];
2312 mtx_lock(&dnp->n_mtx);
2313 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2314 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2315 mtx_unlock(&dnp->n_mtx);
2317 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2318 *tl++ = cookie.nfsuquad[0];
2320 *tl = txdr_unsigned(nmp->nm_readdirsize);
2321 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2323 nfsm_postop_attr(vp, attrflag);
2325 tl = nfsm_dissect(u_int32_t *,
2327 mtx_lock(&dnp->n_mtx);
2328 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2329 dnp->n_cookieverf.nfsuquad[1] = *tl;
2330 mtx_unlock(&dnp->n_mtx);
2336 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2337 more_dirs = fxdr_unsigned(int, *tl);
2339 /* loop thru the dir entries, doctoring them to 4bsd form */
2340 while (more_dirs && bigenough) {
2342 tl = nfsm_dissect(u_int32_t *,
2344 fileno = fxdr_hyper(tl);
2345 len = fxdr_unsigned(int, *(tl + 2));
2347 tl = nfsm_dissect(u_int32_t *,
2349 fileno = fxdr_unsigned(u_quad_t, *tl++);
2350 len = fxdr_unsigned(int, *tl);
2352 if (len <= 0 || len > NFS_MAXNAMLEN) {
2357 tlen = nfsm_rndup(len);
2359 tlen += 4; /* To ensure null termination */
2360 left = DIRBLKSIZ - blksiz;
2361 if ((tlen + DIRHDSIZ) > left) {
2362 dp->d_reclen += left;
2363 uiop->uio_iov->iov_base =
2364 (char *)uiop->uio_iov->iov_base + left;
2365 uiop->uio_iov->iov_len -= left;
2366 uiop->uio_offset += left;
2367 uiop->uio_resid -= left;
2370 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2373 dp = (struct dirent *)uiop->uio_iov->iov_base;
2374 dp->d_fileno = (int)fileno;
2376 dp->d_reclen = tlen + DIRHDSIZ;
2377 dp->d_type = DT_UNKNOWN;
2378 blksiz += dp->d_reclen;
2379 if (blksiz == DIRBLKSIZ)
2381 uiop->uio_offset += DIRHDSIZ;
2382 uiop->uio_resid -= DIRHDSIZ;
2383 uiop->uio_iov->iov_base =
2384 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2385 uiop->uio_iov->iov_len -= DIRHDSIZ;
2386 nfsm_mtouio(uiop, len);
2387 cp = uiop->uio_iov->iov_base;
2389 *cp = '\0'; /* null terminate */
2390 uiop->uio_iov->iov_base =
2391 (char *)uiop->uio_iov->iov_base + tlen;
2392 uiop->uio_iov->iov_len -= tlen;
2393 uiop->uio_offset += tlen;
2394 uiop->uio_resid -= tlen;
2396 nfsm_adv(nfsm_rndup(len));
2398 tl = nfsm_dissect(u_int32_t *,
2401 tl = nfsm_dissect(u_int32_t *,
2405 cookie.nfsuquad[0] = *tl++;
2407 cookie.nfsuquad[1] = *tl++;
2412 more_dirs = fxdr_unsigned(int, *tl);
2415 * If at end of rpc data, get the eof boolean
2418 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2419 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2424 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2425 * by increasing d_reclen for the last record.
2428 left = DIRBLKSIZ - blksiz;
2429 dp->d_reclen += left;
2430 uiop->uio_iov->iov_base =
2431 (char *)uiop->uio_iov->iov_base + left;
2432 uiop->uio_iov->iov_len -= left;
2433 uiop->uio_offset += left;
2434 uiop->uio_resid -= left;
2438 * We are now either at the end of the directory or have filled the
2442 dnp->n_direofoffset = uiop->uio_offset;
2444 if (uiop->uio_resid > 0)
2445 nfs_printf("EEK! readdirrpc resid > 0\n");
2446 nfs_dircookie_lock(dnp);
2447 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2449 nfs_dircookie_unlock(dnp);
2456 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2459 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2465 struct vnode *newvp;
2467 caddr_t bpos, dpos, dpossav1, dpossav2;
2468 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2469 struct nameidata nami, *ndp = &nami;
2470 struct componentname *cnp = &ndp->ni_cnd;
2472 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2473 struct nfsnode *dnp = VTONFS(vp), *np;
2476 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2477 int attrflag, fhsize;
2482 KASSERT(uiop->uio_iovcnt == 1 &&
2483 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2484 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2485 ("nfs readdirplusrpc bad uio"));
2490 * If there is no cookie, assume directory was stale.
2492 nfs_dircookie_lock(dnp);
2493 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2496 nfs_dircookie_unlock(dnp);
2498 nfs_dircookie_unlock(dnp);
2499 return (NFSERR_BAD_COOKIE);
2502 * Loop around doing readdir rpc's of size nm_readdirsize
2503 * truncated to a multiple of DIRBLKSIZ.
2504 * The stopping criteria is EOF or buffer full.
2506 while (more_dirs && bigenough) {
2507 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2508 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2509 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2511 bpos = mtod(mb, caddr_t);
2513 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2514 *tl++ = cookie.nfsuquad[0];
2515 *tl++ = cookie.nfsuquad[1];
2516 mtx_lock(&dnp->n_mtx);
2517 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2518 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2519 mtx_unlock(&dnp->n_mtx);
2520 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2521 *tl = txdr_unsigned(nmp->nm_rsize);
2522 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2523 nfsm_postop_attr(vp, attrflag);
2528 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2529 mtx_lock(&dnp->n_mtx);
2530 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2531 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2532 mtx_unlock(&dnp->n_mtx);
2533 more_dirs = fxdr_unsigned(int, *tl);
2535 /* loop thru the dir entries, doctoring them to 4bsd form */
2536 while (more_dirs && bigenough) {
2537 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2538 fileno = fxdr_hyper(tl);
2539 len = fxdr_unsigned(int, *(tl + 2));
2540 if (len <= 0 || len > NFS_MAXNAMLEN) {
2545 tlen = nfsm_rndup(len);
2547 tlen += 4; /* To ensure null termination*/
2548 left = DIRBLKSIZ - blksiz;
2549 if ((tlen + DIRHDSIZ) > left) {
2550 dp->d_reclen += left;
2551 uiop->uio_iov->iov_base =
2552 (char *)uiop->uio_iov->iov_base + left;
2553 uiop->uio_iov->iov_len -= left;
2554 uiop->uio_offset += left;
2555 uiop->uio_resid -= left;
2558 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2561 dp = (struct dirent *)uiop->uio_iov->iov_base;
2562 dp->d_fileno = (int)fileno;
2564 dp->d_reclen = tlen + DIRHDSIZ;
2565 dp->d_type = DT_UNKNOWN;
2566 blksiz += dp->d_reclen;
2567 if (blksiz == DIRBLKSIZ)
2569 uiop->uio_offset += DIRHDSIZ;
2570 uiop->uio_resid -= DIRHDSIZ;
2571 uiop->uio_iov->iov_base =
2572 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2573 uiop->uio_iov->iov_len -= DIRHDSIZ;
2574 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2575 cnp->cn_namelen = len;
2576 nfsm_mtouio(uiop, len);
2577 cp = uiop->uio_iov->iov_base;
2580 uiop->uio_iov->iov_base =
2581 (char *)uiop->uio_iov->iov_base + tlen;
2582 uiop->uio_iov->iov_len -= tlen;
2583 uiop->uio_offset += tlen;
2584 uiop->uio_resid -= tlen;
2586 nfsm_adv(nfsm_rndup(len));
2587 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2589 cookie.nfsuquad[0] = *tl++;
2590 cookie.nfsuquad[1] = *tl++;
2595 * Since the attributes are before the file handle
2596 * (sigh), we must skip over the attributes and then
2597 * come back and get them.
2599 attrflag = fxdr_unsigned(int, *tl);
2603 nfsm_adv(NFSX_V3FATTR);
2604 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2605 doit = fxdr_unsigned(int, *tl);
2607 * Skip loading the attrs for "..". There's a
2608 * race between loading the attrs here and
2609 * lookups that look for the directory currently
2610 * being read (in the parent). We try to acquire
2611 * the exclusive lock on ".." here, owning the
2612 * lock on the directory being read. Lookup will
2613 * hold the lock on ".." and try to acquire the
2614 * lock on the directory being read.
2616 * There are other ways of fixing this, one would
2617 * be to do a trylock on the ".." vnode and skip
2618 * loading the attrs on ".." if it happens to be
2619 * locked by another process. But skipping the
2620 * attrload on ".." seems the easiest option.
2622 if (strcmp(dp->d_name, "..") == 0) {
2625 * We've already skipped over the attrs,
2626 * skip over the filehandle. And store d_type
2629 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2630 i = fxdr_unsigned(int, *tl);
2631 nfsm_adv(nfsm_rndup(i));
2632 dp->d_type = IFTODT(VTTOIF(VDIR));
2635 nfsm_getfh(fhp, fhsize, 1);
2636 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2641 error = nfs_nget(vp->v_mount, fhp,
2642 fhsize, &np, LK_EXCLUSIVE);
2649 if (doit && bigenough) {
2654 nfsm_loadattr(newvp, NULL);
2658 IFTODT(VTTOIF(np->n_vattr.va_type));
2661 * Update n_ctime so subsequent lookup
2662 * doesn't purge entry.
2664 np->n_ctime = np->n_vattr.va_ctime;
2665 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2668 /* Just skip over the file handle */
2669 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2670 i = fxdr_unsigned(int, *tl);
2672 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2673 fhsize = fxdr_unsigned(int, *tl);
2674 nfsm_adv(nfsm_rndup(fhsize));
2677 if (newvp != NULLVP) {
2684 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2685 more_dirs = fxdr_unsigned(int, *tl);
2688 * If at end of rpc data, get the eof boolean
2691 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2692 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2697 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2698 * by increasing d_reclen for the last record.
2701 left = DIRBLKSIZ - blksiz;
2702 dp->d_reclen += left;
2703 uiop->uio_iov->iov_base =
2704 (char *)uiop->uio_iov->iov_base + left;
2705 uiop->uio_iov->iov_len -= left;
2706 uiop->uio_offset += left;
2707 uiop->uio_resid -= left;
2711 * We are now either at the end of the directory or have filled the
2715 dnp->n_direofoffset = uiop->uio_offset;
2717 if (uiop->uio_resid > 0)
2718 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2719 nfs_dircookie_lock(dnp);
2720 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2722 nfs_dircookie_unlock(dnp);
2725 if (newvp != NULLVP) {
2736 * Silly rename. To make the NFS filesystem that is stateless look a little
2737 * more like the "ufs" a remove of an active vnode is translated to a rename
2738 * to a funny looking filename that is removed by nfs_inactive on the
2739 * nfsnode. There is the potential for another process on a different client
2740 * to create the same funny name between the nfs_lookitup() fails and the
2741 * nfs_rename() completes, but...
2744 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2746 struct sillyrename *sp;
2750 unsigned int lticks;
2754 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2755 sp = malloc(sizeof (struct sillyrename),
2756 M_NFSREQ, M_WAITOK);
2757 sp->s_cred = crhold(cnp->cn_cred);
2759 sp->s_removeit = nfs_removeit;
2763 * Fudge together a funny name.
2764 * Changing the format of the funny name to accomodate more
2765 * sillynames per directory.
2766 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2767 * CPU ticks since boot.
2769 pid = cnp->cn_thread->td_proc->p_pid;
2770 lticks = (unsigned int)ticks;
2772 sp->s_namlen = sprintf(sp->s_name,
2773 ".nfs.%08x.%04x4.4", lticks,
2775 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2776 cnp->cn_thread, NULL))
2780 error = nfs_renameit(dvp, cnp, sp);
2783 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2784 cnp->cn_thread, &np);
2785 np->n_sillyrename = sp;
2790 free((caddr_t)sp, M_NFSREQ);
2795 * Look up a file name and optionally either update the file handle or
2796 * allocate an nfsnode, depending on the value of npp.
2797 * npp == NULL --> just do the lookup
2798 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2800 * *npp != NULL --> update the file handle in the vnode
2803 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2804 struct thread *td, struct nfsnode **npp)
2806 struct vnode *newvp = NULL;
2807 struct nfsnode *np, *dnp = VTONFS(dvp);
2809 int error = 0, fhlen, attrflag;
2810 struct mbuf *mreq, *mrep, *md, *mb;
2812 int v3 = NFS_ISV3(dvp);
2814 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2815 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2816 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2818 bpos = mtod(mb, caddr_t);
2819 nfsm_fhtom(dvp, v3);
2820 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2821 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2822 if (npp && !error) {
2823 nfsm_getfh(nfhp, fhlen, v3);
2826 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2827 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2828 np->n_fhp = &np->n_fh;
2829 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2830 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2831 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2832 np->n_fhsize = fhlen;
2834 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2838 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2846 nfsm_postop_attr(newvp, attrflag);
2847 if (!attrflag && *npp == NULL) {
2856 nfsm_loadattr(newvp, NULL);
2860 if (npp && *npp == NULL) {
2875 * Nfs Version 3 commit rpc
2878 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2882 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2884 int error = 0, wccflag = NFSV3_WCCRATTR;
2885 struct mbuf *mreq, *mrep, *md, *mb;
2887 mtx_lock(&nmp->nm_mtx);
2888 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2889 mtx_unlock(&nmp->nm_mtx);
2892 mtx_unlock(&nmp->nm_mtx);
2893 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2894 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2896 bpos = mtod(mb, caddr_t);
2898 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2899 txdr_hyper(offset, tl);
2901 *tl = txdr_unsigned(cnt);
2902 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2903 nfsm_wcc_data(vp, wccflag);
2905 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2906 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2907 NFSX_V3WRITEVERF)) {
2908 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2910 error = NFSERR_STALEWRITEVERF;
2920 * For async requests when nfsiod(s) are running, queue the request by
2921 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2925 nfs_strategy(struct vop_strategy_args *ap)
2927 struct buf *bp = ap->a_bp;
2930 KASSERT(!(bp->b_flags & B_DONE),
2931 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2932 BUF_ASSERT_HELD(bp);
2934 if (bp->b_iocmd == BIO_READ)
2940 * If the op is asynchronous and an i/o daemon is waiting
2941 * queue the request, wake it up and wait for completion
2942 * otherwise just do it ourselves.
2944 if ((bp->b_flags & B_ASYNC) == 0 ||
2945 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2946 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2951 * fsync vnode op. Just call nfs_flush() with commit == 1.
2955 nfs_fsync(struct vop_fsync_args *ap)
2958 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1));
2962 * Flush all the blocks associated with a vnode.
2963 * Walk through the buffer pool and push any dirty pages
2964 * associated with the vnode.
2967 nfs_flush(struct vnode *vp, int waitfor, int commit)
2969 struct nfsnode *np = VTONFS(vp);
2973 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2974 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2976 u_quad_t off, endoff, toff;
2977 struct ucred* wcred = NULL;
2978 struct buf **bvec = NULL;
2980 struct thread *td = curthread;
2981 #ifndef NFS_COMMITBVECSIZ
2982 #define NFS_COMMITBVECSIZ 20
2984 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2985 int bvecsize = 0, bveccount;
2987 if (nmp->nm_flag & NFSMNT_INT)
2988 slpflag = NFS_PCATCH;
2993 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2994 * server, but has not been committed to stable storage on the server
2995 * yet. On the first pass, the byte range is worked out and the commit
2996 * rpc is done. On the second pass, nfs_writebp() is called to do the
3003 if (NFS_ISV3(vp) && commit) {
3004 if (bvec != NULL && bvec != bvec_on_stack)
3007 * Count up how many buffers waiting for a commit.
3011 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3012 if (!BUF_ISLOCKED(bp) &&
3013 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
3014 == (B_DELWRI | B_NEEDCOMMIT))
3018 * Allocate space to remember the list of bufs to commit. It is
3019 * important to use M_NOWAIT here to avoid a race with nfs_write.
3020 * If we can't get memory (for whatever reason), we will end up
3021 * committing the buffers one-by-one in the loop below.
3023 if (bveccount > NFS_COMMITBVECSIZ) {
3025 * Release the vnode interlock to avoid a lock
3029 bvec = (struct buf **)
3030 malloc(bveccount * sizeof(struct buf *),
3034 bvec = bvec_on_stack;
3035 bvecsize = NFS_COMMITBVECSIZ;
3037 bvecsize = bveccount;
3039 bvec = bvec_on_stack;
3040 bvecsize = NFS_COMMITBVECSIZ;
3042 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3043 if (bvecpos >= bvecsize)
3045 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3046 nbp = TAILQ_NEXT(bp, b_bobufs);
3049 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3050 (B_DELWRI | B_NEEDCOMMIT)) {
3052 nbp = TAILQ_NEXT(bp, b_bobufs);
3058 * Work out if all buffers are using the same cred
3059 * so we can deal with them all with one commit.
3061 * NOTE: we are not clearing B_DONE here, so we have
3062 * to do it later on in this routine if we intend to
3063 * initiate I/O on the bp.
3065 * Note: to avoid loopback deadlocks, we do not
3066 * assign b_runningbufspace.
3069 wcred = bp->b_wcred;
3070 else if (wcred != bp->b_wcred)
3072 vfs_busy_pages(bp, 1);
3076 * bp is protected by being locked, but nbp is not
3077 * and vfs_busy_pages() may sleep. We have to
3080 nbp = TAILQ_NEXT(bp, b_bobufs);
3083 * A list of these buffers is kept so that the
3084 * second loop knows which buffers have actually
3085 * been committed. This is necessary, since there
3086 * may be a race between the commit rpc and new
3087 * uncommitted writes on the file.
3089 bvec[bvecpos++] = bp;
3090 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3094 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3102 * Commit data on the server, as required.
3103 * If all bufs are using the same wcred, then use that with
3104 * one call for all of them, otherwise commit each one
3107 if (wcred != NOCRED)
3108 retv = nfs_commit(vp, off, (int)(endoff - off),
3112 for (i = 0; i < bvecpos; i++) {
3115 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3117 size = (u_quad_t)(bp->b_dirtyend
3119 retv = nfs_commit(vp, off, (int)size,
3125 if (retv == NFSERR_STALEWRITEVERF)
3126 nfs_clearcommit(vp->v_mount);
3129 * Now, either mark the blocks I/O done or mark the
3130 * blocks dirty, depending on whether the commit
3133 for (i = 0; i < bvecpos; i++) {
3135 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3138 * Error, leave B_DELWRI intact
3140 vfs_unbusy_pages(bp);
3144 * Success, remove B_DELWRI ( bundirty() ).
3146 * b_dirtyoff/b_dirtyend seem to be NFS
3147 * specific. We should probably move that
3148 * into bundirty(). XXX
3151 bp->b_flags |= B_ASYNC;
3153 bp->b_flags &= ~B_DONE;
3154 bp->b_ioflags &= ~BIO_ERROR;
3155 bp->b_dirtyoff = bp->b_dirtyend = 0;
3162 * Start/do any write(s) that are required.
3166 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3167 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3168 if (waitfor != MNT_WAIT || passone)
3171 error = BUF_TIMELOCK(bp,
3172 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3173 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
3178 if (error == ENOLCK) {
3182 if (nfs_sigintr(nmp, td)) {
3186 if (slpflag & PCATCH) {
3192 if ((bp->b_flags & B_DELWRI) == 0)
3193 panic("nfs_fsync: not dirty");
3194 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3200 if (passone || !commit)
3201 bp->b_flags |= B_ASYNC;
3203 bp->b_flags |= B_ASYNC;
3205 if (nfs_sigintr(nmp, td)) {
3216 if (waitfor == MNT_WAIT) {
3217 while (bo->bo_numoutput) {
3218 error = bufobj_wwait(bo, slpflag, slptimeo);
3221 error = nfs_sigintr(nmp, td);
3224 if (slpflag & PCATCH) {
3231 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3236 * Wait for all the async IO requests to drain
3239 mtx_lock(&np->n_mtx);
3240 while (np->n_directio_asyncwr > 0) {
3241 np->n_flag |= NFSYNCWAIT;
3242 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3243 &np->n_mtx, slpflag | (PRIBIO + 1),
3246 if (nfs_sigintr(nmp, td)) {
3247 mtx_unlock(&np->n_mtx);
3253 mtx_unlock(&np->n_mtx);
3256 mtx_lock(&np->n_mtx);
3257 if (np->n_flag & NWRITEERR) {
3258 error = np->n_error;
3259 np->n_flag &= ~NWRITEERR;
3261 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3262 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3263 np->n_flag &= ~NMODIFIED;
3264 mtx_unlock(&np->n_mtx);
3266 if (bvec != NULL && bvec != bvec_on_stack)
3272 * NFS advisory byte-level locks.
3275 nfs_advlock(struct vop_advlock_args *ap)
3277 struct vnode *vp = ap->a_vp;
3281 error = vn_lock(vp, LK_SHARED);
3284 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3285 size = VTONFS(vp)->n_size;
3287 error = lf_advlock(ap, &(vp->v_lockf), size);
3290 error = nfs_advlock_p(ap);
3299 * NFS advisory byte-level locks.
3302 nfs_advlockasync(struct vop_advlockasync_args *ap)
3304 struct vnode *vp = ap->a_vp;
3308 error = vn_lock(vp, LK_SHARED);
3311 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3312 size = VTONFS(vp)->n_size;
3314 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3323 * Print out the contents of an nfsnode.
3326 nfs_print(struct vop_print_args *ap)
3328 struct vnode *vp = ap->a_vp;
3329 struct nfsnode *np = VTONFS(vp);
3331 nfs_printf("\tfileid %ld fsid 0x%x",
3332 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3333 if (vp->v_type == VFIFO)
3340 * This is the "real" nfs::bwrite(struct buf*).
3341 * We set B_CACHE if this is a VMIO buffer.
3344 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3347 int oldflags = bp->b_flags;
3353 BUF_ASSERT_HELD(bp);
3355 if (bp->b_flags & B_INVAL) {
3360 bp->b_flags |= B_CACHE;
3363 * Undirty the bp. We will redirty it later if the I/O fails.
3368 bp->b_flags &= ~B_DONE;
3369 bp->b_ioflags &= ~BIO_ERROR;
3370 bp->b_iocmd = BIO_WRITE;
3372 bufobj_wref(bp->b_bufobj);
3373 curthread->td_ru.ru_oublock++;
3377 * Note: to avoid loopback deadlocks, we do not
3378 * assign b_runningbufspace.
3380 vfs_busy_pages(bp, 1);
3383 bp->b_iooffset = dbtob(bp->b_blkno);
3386 if( (oldflags & B_ASYNC) == 0) {
3387 int rtval = bufwait(bp);
3389 if (oldflags & B_DELWRI) {
3402 * nfs special file access vnode op.
3403 * Essentially just get vattr and then imitate iaccess() since the device is
3404 * local to the client.
3407 nfsspec_access(struct vop_access_args *ap)
3410 struct ucred *cred = ap->a_cred;
3411 struct vnode *vp = ap->a_vp;
3412 accmode_t accmode = ap->a_accmode;
3417 * Disallow write attempts on filesystems mounted read-only;
3418 * unless the file is a socket, fifo, or a block or character
3419 * device resident on the filesystem.
3421 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3422 switch (vp->v_type) {
3432 error = VOP_GETATTR(vp, vap, cred);
3435 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3436 accmode, cred, NULL);
3442 * Read wrapper for fifos.
3445 nfsfifo_read(struct vop_read_args *ap)
3447 struct nfsnode *np = VTONFS(ap->a_vp);
3453 mtx_lock(&np->n_mtx);
3455 getnanotime(&np->n_atim);
3456 mtx_unlock(&np->n_mtx);
3457 error = fifo_specops.vop_read(ap);
3462 * Write wrapper for fifos.
3465 nfsfifo_write(struct vop_write_args *ap)
3467 struct nfsnode *np = VTONFS(ap->a_vp);
3472 mtx_lock(&np->n_mtx);
3474 getnanotime(&np->n_mtim);
3475 mtx_unlock(&np->n_mtx);
3476 return(fifo_specops.vop_write(ap));
3480 * Close wrapper for fifos.
3482 * Update the times on the nfsnode then do fifo close.
3485 nfsfifo_close(struct vop_close_args *ap)
3487 struct vnode *vp = ap->a_vp;
3488 struct nfsnode *np = VTONFS(vp);
3492 mtx_lock(&np->n_mtx);
3493 if (np->n_flag & (NACC | NUPD)) {
3495 if (np->n_flag & NACC)
3497 if (np->n_flag & NUPD)
3500 if (vrefcnt(vp) == 1 &&
3501 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3503 if (np->n_flag & NACC)
3504 vattr.va_atime = np->n_atim;
3505 if (np->n_flag & NUPD)
3506 vattr.va_mtime = np->n_mtim;
3507 mtx_unlock(&np->n_mtx);
3508 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3512 mtx_unlock(&np->n_mtx);
3514 return (fifo_specops.vop_close(ap));
3518 * Just call nfs_writebp() with the force argument set to 1.
3520 * NOTE: B_DONE may or may not be set in a_bp on call.
3523 nfs_bwrite(struct buf *bp)
3526 return (nfs_writebp(bp, 1, curthread));
3529 struct buf_ops buf_ops_nfs = {
3530 .bop_name = "buf_ops_nfs",
3531 .bop_write = nfs_bwrite,
3532 .bop_strategy = bufstrategy,
3533 .bop_sync = bufsync,
3534 .bop_bdflush = bufbdflush,