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>
53 #include <sys/malloc.h>
55 #include <sys/namei.h>
56 #include <sys/socket.h>
57 #include <sys/vnode.h>
58 #include <sys/dirent.h>
59 #include <sys/fcntl.h>
60 #include <sys/lockf.h>
62 #include <sys/sysctl.h>
63 #include <sys/signalvar.h>
66 #include <vm/vm_object.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
70 #include <fs/fifofs/fifo.h>
72 #include <nfs/nfsproto.h>
73 #include <nfsclient/nfs.h>
74 #include <nfsclient/nfsnode.h>
75 #include <nfsclient/nfsmount.h>
76 #include <nfsclient/nfs_kdtrace.h>
77 #include <nfsclient/nfs_lock.h>
78 #include <nfs/xdr_subs.h>
79 #include <nfsclient/nfsm_subs.h>
82 #include <netinet/in.h>
83 #include <netinet/in_var.h>
85 #include <machine/stdarg.h>
88 #include <sys/dtrace_bsd.h>
90 dtrace_nfsclient_accesscache_flush_probe_func_t
91 dtrace_nfsclient_accesscache_flush_done_probe;
92 uint32_t nfsclient_accesscache_flush_done_id;
94 dtrace_nfsclient_accesscache_get_probe_func_t
95 dtrace_nfsclient_accesscache_get_hit_probe,
96 dtrace_nfsclient_accesscache_get_miss_probe;
97 uint32_t nfsclient_accesscache_get_hit_id;
98 uint32_t nfsclient_accesscache_get_miss_id;
100 dtrace_nfsclient_accesscache_load_probe_func_t
101 dtrace_nfsclient_accesscache_load_done_probe;
102 uint32_t nfsclient_accesscache_load_done_id;
103 #endif /* !KDTRACE_HOOKS */
110 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
111 * calls are not in getblk() and brelse() so that they would not be necessary
115 #define vfs_busy_pages(bp, f)
118 static vop_read_t nfsfifo_read;
119 static vop_write_t nfsfifo_write;
120 static vop_close_t nfsfifo_close;
121 static int nfs_flush(struct vnode *, int, int);
122 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *);
123 static vop_lookup_t nfs_lookup;
124 static vop_create_t nfs_create;
125 static vop_mknod_t nfs_mknod;
126 static vop_open_t nfs_open;
127 static vop_close_t nfs_close;
128 static vop_access_t nfs_access;
129 static vop_getattr_t nfs_getattr;
130 static vop_setattr_t nfs_setattr;
131 static vop_read_t nfs_read;
132 static vop_fsync_t nfs_fsync;
133 static vop_remove_t nfs_remove;
134 static vop_link_t nfs_link;
135 static vop_rename_t nfs_rename;
136 static vop_mkdir_t nfs_mkdir;
137 static vop_rmdir_t nfs_rmdir;
138 static vop_symlink_t nfs_symlink;
139 static vop_readdir_t nfs_readdir;
140 static vop_strategy_t nfs_strategy;
141 static int nfs_lookitup(struct vnode *, const char *, int,
142 struct ucred *, struct thread *, struct nfsnode **);
143 static int nfs_sillyrename(struct vnode *, struct vnode *,
144 struct componentname *);
145 static vop_access_t nfsspec_access;
146 static vop_readlink_t nfs_readlink;
147 static vop_print_t nfs_print;
148 static vop_advlock_t nfs_advlock;
149 static vop_advlockasync_t nfs_advlockasync;
152 * Global vfs data structures for nfs
154 struct vop_vector nfs_vnodeops = {
155 .vop_default = &default_vnodeops,
156 .vop_access = nfs_access,
157 .vop_advlock = nfs_advlock,
158 .vop_advlockasync = nfs_advlockasync,
159 .vop_close = nfs_close,
160 .vop_create = nfs_create,
161 .vop_fsync = nfs_fsync,
162 .vop_getattr = nfs_getattr,
163 .vop_getpages = nfs_getpages,
164 .vop_putpages = nfs_putpages,
165 .vop_inactive = nfs_inactive,
166 .vop_link = nfs_link,
167 .vop_lookup = nfs_lookup,
168 .vop_mkdir = nfs_mkdir,
169 .vop_mknod = nfs_mknod,
170 .vop_open = nfs_open,
171 .vop_print = nfs_print,
172 .vop_read = nfs_read,
173 .vop_readdir = nfs_readdir,
174 .vop_readlink = nfs_readlink,
175 .vop_reclaim = nfs_reclaim,
176 .vop_remove = nfs_remove,
177 .vop_rename = nfs_rename,
178 .vop_rmdir = nfs_rmdir,
179 .vop_setattr = nfs_setattr,
180 .vop_strategy = nfs_strategy,
181 .vop_symlink = nfs_symlink,
182 .vop_write = nfs_write,
185 struct vop_vector nfs_fifoops = {
186 .vop_default = &fifo_specops,
187 .vop_access = nfsspec_access,
188 .vop_close = nfsfifo_close,
189 .vop_fsync = nfs_fsync,
190 .vop_getattr = nfs_getattr,
191 .vop_inactive = nfs_inactive,
192 .vop_print = nfs_print,
193 .vop_read = nfsfifo_read,
194 .vop_reclaim = nfs_reclaim,
195 .vop_setattr = nfs_setattr,
196 .vop_write = nfsfifo_write,
199 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
200 struct componentname *cnp, struct vattr *vap);
201 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
202 struct ucred *cred, struct thread *td);
203 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
204 int fnamelen, struct vnode *tdvp,
205 const char *tnameptr, int tnamelen,
206 struct ucred *cred, struct thread *td);
207 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
208 struct sillyrename *sp);
213 struct mtx nfs_iod_mtx;
214 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
215 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
216 int nfs_numasync = 0;
217 vop_advlock_t *nfs_advlock_p = nfs_dolock;
218 vop_reclaim_t *nfs_reclaim_p = NULL;
219 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
221 SYSCTL_DECL(_vfs_nfs);
223 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
224 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
225 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
227 static int nfs_prime_access_cache = 0;
228 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
229 &nfs_prime_access_cache, 0,
230 "Prime NFS ACCESS cache when fetching attributes");
232 static int nfsv3_commit_on_close = 0;
233 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
234 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
236 static int nfs_clean_pages_on_close = 1;
237 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
238 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
240 int nfs_directio_enable = 0;
241 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
242 &nfs_directio_enable, 0, "Enable NFS directio");
245 * This sysctl allows other processes to mmap a file that has been opened
246 * O_DIRECT by a process. In general, having processes mmap the file while
247 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
248 * this by default to prevent DoS attacks - to prevent a malicious user from
249 * opening up files O_DIRECT preventing other users from mmap'ing these
250 * files. "Protected" environments where stricter consistency guarantees are
251 * required can disable this knob. The process that opened the file O_DIRECT
252 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
255 int nfs_directio_allow_mmap = 1;
256 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
257 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
260 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
261 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
263 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
264 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
267 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
268 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
269 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
273 * The list of locks after the description of the lock is the ordering
274 * of other locks acquired with the lock held.
275 * np->n_mtx : Protects the fields in the nfsnode.
277 VI_MTX (acquired indirectly)
278 * nmp->nm_mtx : Protects the fields in the nfsmount.
280 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
281 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
284 * rep->r_mtx : Protects the fields in an nfsreq.
288 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
289 struct ucred *cred, uint32_t *retmode)
293 int error = 0, attrflag, i, lrupos;
295 struct mbuf *mreq, *mrep, *md, *mb;
298 struct nfsnode *np = VTONFS(vp);
300 nfsstats.rpccnt[NFSPROC_ACCESS]++;
301 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
303 bpos = mtod(mb, caddr_t);
305 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
306 *tl = txdr_unsigned(wmode);
307 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
308 nfsm_postop_attr(vp, attrflag);
311 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
312 rmode = fxdr_unsigned(u_int32_t, *tl);
313 mtx_lock(&np->n_mtx);
314 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
315 if (np->n_accesscache[i].uid == cred->cr_uid) {
316 np->n_accesscache[i].mode = rmode;
317 np->n_accesscache[i].stamp = time_second;
320 if (i > 0 && np->n_accesscache[i].stamp <
321 np->n_accesscache[lrupos].stamp)
324 if (i == NFS_ACCESSCACHESIZE) {
325 np->n_accesscache[lrupos].uid = cred->cr_uid;
326 np->n_accesscache[lrupos].mode = rmode;
327 np->n_accesscache[lrupos].stamp = time_second;
329 mtx_unlock(&np->n_mtx);
332 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
338 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
346 * nfs access vnode op.
347 * For nfs version 2, just return ok. File accesses may fail later.
348 * For nfs version 3, use the access rpc to check accessibility. If file modes
349 * are changed on the server, accesses might still fail later.
352 nfs_access(struct vop_access_args *ap)
354 struct vnode *vp = ap->a_vp;
355 int error = 0, i, gotahit;
356 u_int32_t mode, rmode, wmode;
357 int v3 = NFS_ISV3(vp);
358 struct nfsnode *np = VTONFS(vp);
361 * Disallow write attempts on filesystems mounted read-only;
362 * unless the file is a socket, fifo, or a block or character
363 * device resident on the filesystem.
365 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
366 switch (vp->v_type) {
376 * For nfs v3, check to see if we have done this recently, and if
377 * so return our cached result instead of making an ACCESS call.
378 * If not, do an access rpc, otherwise you are stuck emulating
379 * ufs_access() locally using the vattr. This may not be correct,
380 * since the server may apply other access criteria such as
381 * client uid-->server uid mapping that we do not know about.
384 if (ap->a_accmode & VREAD)
385 mode = NFSV3ACCESS_READ;
388 if (vp->v_type != VDIR) {
389 if (ap->a_accmode & VWRITE)
390 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
391 if (ap->a_accmode & VEXEC)
392 mode |= NFSV3ACCESS_EXECUTE;
394 if (ap->a_accmode & VWRITE)
395 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
397 if (ap->a_accmode & VEXEC)
398 mode |= NFSV3ACCESS_LOOKUP;
400 /* XXX safety belt, only make blanket request if caching */
401 if (nfsaccess_cache_timeout > 0) {
402 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
403 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
404 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
410 * Does our cached result allow us to give a definite yes to
414 mtx_lock(&np->n_mtx);
415 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
416 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
417 if (time_second < (np->n_accesscache[i].stamp +
418 nfsaccess_cache_timeout) &&
419 (np->n_accesscache[i].mode & mode) == mode) {
420 nfsstats.accesscache_hits++;
426 mtx_unlock(&np->n_mtx);
429 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
430 ap->a_cred->cr_uid, mode);
432 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
433 ap->a_cred->cr_uid, mode);
437 * Either a no, or a don't know. Go to the wire.
439 nfsstats.accesscache_misses++;
440 error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred,
443 if ((rmode & mode) != mode)
449 if ((error = nfsspec_access(ap)) != 0) {
453 * Attempt to prevent a mapped root from accessing a file
454 * which it shouldn't. We try to read a byte from the file
455 * if the user is root and the file is not zero length.
456 * After calling nfsspec_access, we should have the correct
459 mtx_lock(&np->n_mtx);
460 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
461 && VTONFS(vp)->n_size > 0) {
466 mtx_unlock(&np->n_mtx);
469 auio.uio_iov = &aiov;
473 auio.uio_segflg = UIO_SYSSPACE;
474 auio.uio_rw = UIO_READ;
475 auio.uio_td = ap->a_td;
477 if (vp->v_type == VREG)
478 error = nfs_readrpc(vp, &auio, ap->a_cred);
479 else if (vp->v_type == VDIR) {
481 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
483 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
484 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
486 } else if (vp->v_type == VLNK)
487 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
491 mtx_unlock(&np->n_mtx);
496 int nfs_otw_getattr_avoid = 0;
500 * Check to see if the type is ok
501 * and that deletion is not in progress.
502 * For paged in text files, you will need to flush the page cache
503 * if consistency is lost.
507 nfs_open(struct vop_open_args *ap)
509 struct vnode *vp = ap->a_vp;
510 struct nfsnode *np = VTONFS(vp);
513 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)
528 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
529 if (vp->v_type == VDIR)
530 np->n_direofoffset = 0;
531 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
534 mtx_lock(&np->n_mtx);
535 np->n_mtime = vattr.va_mtime;
536 mtx_unlock(&np->n_mtx);
538 struct thread *td = curthread;
540 if (np->n_ac_ts_syscalls != td->td_syscalls ||
541 np->n_ac_ts_tid != td->td_tid ||
542 td->td_proc == NULL ||
543 np->n_ac_ts_pid != td->td_proc->p_pid) {
545 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
547 mtx_unlock(&np->n_mtx);
548 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
551 mtx_lock(&np->n_mtx);
552 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
553 if (vp->v_type == VDIR)
554 np->n_direofoffset = 0;
555 mtx_unlock(&np->n_mtx);
556 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
557 if (error == EINTR || error == EIO) {
560 mtx_lock(&np->n_mtx);
561 np->n_mtime = vattr.va_mtime;
563 mtx_unlock(&np->n_mtx);
566 * If the object has >= 1 O_DIRECT active opens, we disable caching.
568 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
569 if (np->n_directio_opens == 0) {
570 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
573 mtx_lock(&np->n_mtx);
574 np->n_flag |= NNONCACHE;
575 mtx_unlock(&np->n_mtx);
577 np->n_directio_opens++;
579 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
585 * What an NFS client should do upon close after writing is a debatable issue.
586 * Most NFS clients push delayed writes to the server upon close, basically for
588 * 1 - So that any write errors may be reported back to the client process
589 * doing the close system call. By far the two most likely errors are
590 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
591 * 2 - To put a worst case upper bound on cache inconsistency between
592 * multiple clients for the file.
593 * There is also a consistency problem for Version 2 of the protocol w.r.t.
594 * not being able to tell if other clients are writing a file concurrently,
595 * since there is no way of knowing if the changed modify time in the reply
596 * is only due to the write for this client.
597 * (NFS Version 3 provides weak cache consistency data in the reply that
598 * should be sufficient to detect and handle this case.)
600 * The current code does the following:
601 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
602 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
603 * or commit them (this satisfies 1 and 2 except for the
604 * case where the server crashes after this close but
605 * before the commit RPC, which is felt to be "good
606 * enough". Changing the last argument to nfs_flush() to
607 * a 1 would force a commit operation, if it is felt a
608 * commit is necessary now.
612 nfs_close(struct vop_close_args *ap)
614 struct vnode *vp = ap->a_vp;
615 struct nfsnode *np = VTONFS(vp);
617 int fmode = ap->a_fflag;
619 if (vp->v_type == VREG) {
621 * Examine and clean dirty pages, regardless of NMODIFIED.
622 * This closes a major hole in close-to-open consistency.
623 * We want to push out all dirty pages (and buffers) on
624 * close, regardless of whether they were dirtied by
625 * mmap'ed writes or via write().
627 if (nfs_clean_pages_on_close && vp->v_object) {
628 VM_OBJECT_LOCK(vp->v_object);
629 vm_object_page_clean(vp->v_object, 0, 0, 0);
630 VM_OBJECT_UNLOCK(vp->v_object);
632 mtx_lock(&np->n_mtx);
633 if (np->n_flag & NMODIFIED) {
634 mtx_unlock(&np->n_mtx);
637 * Under NFSv3 we have dirty buffers to dispose of. We
638 * must flush them to the NFS server. We have the option
639 * of waiting all the way through the commit rpc or just
640 * waiting for the initial write. The default is to only
641 * wait through the initial write so the data is in the
642 * server's cache, which is roughly similar to the state
643 * a standard disk subsystem leaves the file in on close().
645 * We cannot clear the NMODIFIED bit in np->n_flag due to
646 * potential races with other processes, and certainly
647 * cannot clear it if we don't commit.
649 int cm = nfsv3_commit_on_close ? 1 : 0;
650 error = nfs_flush(vp, MNT_WAIT, cm);
651 /* np->n_flag &= ~NMODIFIED; */
653 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
654 mtx_lock(&np->n_mtx);
656 if (np->n_flag & NWRITEERR) {
657 np->n_flag &= ~NWRITEERR;
660 mtx_unlock(&np->n_mtx);
662 if (nfs_directio_enable)
663 KASSERT((np->n_directio_asyncwr == 0),
664 ("nfs_close: dirty unflushed (%d) directio buffers\n",
665 np->n_directio_asyncwr));
666 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
667 mtx_lock(&np->n_mtx);
668 KASSERT((np->n_directio_opens > 0),
669 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
670 np->n_directio_opens--;
671 if (np->n_directio_opens == 0)
672 np->n_flag &= ~NNONCACHE;
673 mtx_unlock(&np->n_mtx);
679 * nfs getattr call from vfs.
682 nfs_getattr(struct vop_getattr_args *ap)
684 struct vnode *vp = ap->a_vp;
685 struct nfsnode *np = VTONFS(vp);
686 struct thread *td = curthread;
687 struct vattr *vap = ap->a_vap;
691 struct mbuf *mreq, *mrep, *md, *mb;
692 int v3 = NFS_ISV3(vp);
695 * Update local times for special files.
697 mtx_lock(&np->n_mtx);
698 if (np->n_flag & (NACC | NUPD))
700 mtx_unlock(&np->n_mtx);
702 * First look in the cache.
704 if (nfs_getattrcache(vp, &vattr) == 0)
706 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
707 nfsstats.accesscache_misses++;
708 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL);
709 if (nfs_getattrcache(vp, &vattr) == 0)
712 nfsstats.rpccnt[NFSPROC_GETATTR]++;
713 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
715 bpos = mtod(mb, caddr_t);
717 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
719 nfsm_loadattr(vp, &vattr);
723 vap->va_type = vattr.va_type;
724 vap->va_mode = vattr.va_mode;
725 vap->va_nlink = vattr.va_nlink;
726 vap->va_uid = vattr.va_uid;
727 vap->va_gid = vattr.va_gid;
728 vap->va_fsid = vattr.va_fsid;
729 vap->va_fileid = vattr.va_fileid;
730 vap->va_size = vattr.va_size;
731 vap->va_blocksize = vattr.va_blocksize;
732 vap->va_atime = vattr.va_atime;
733 vap->va_mtime = vattr.va_mtime;
734 vap->va_ctime = vattr.va_ctime;
735 vap->va_gen = vattr.va_gen;
736 vap->va_flags = vattr.va_flags;
737 vap->va_rdev = vattr.va_rdev;
738 vap->va_bytes = vattr.va_bytes;
739 vap->va_filerev = vattr.va_filerev;
748 nfs_setattr(struct vop_setattr_args *ap)
750 struct vnode *vp = ap->a_vp;
751 struct nfsnode *np = VTONFS(vp);
752 struct vattr *vap = ap->a_vap;
753 struct thread *td = curthread;
762 * Setting of flags is not supported.
764 if (vap->va_flags != VNOVAL)
768 * Disallow write attempts if the filesystem is mounted read-only.
770 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
771 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
772 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
773 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
777 if (vap->va_size != VNOVAL) {
778 switch (vp->v_type) {
785 if (vap->va_mtime.tv_sec == VNOVAL &&
786 vap->va_atime.tv_sec == VNOVAL &&
787 vap->va_mode == (mode_t)VNOVAL &&
788 vap->va_uid == (uid_t)VNOVAL &&
789 vap->va_gid == (gid_t)VNOVAL)
791 vap->va_size = VNOVAL;
795 * Disallow write attempts if the filesystem is
798 if (vp->v_mount->mnt_flag & MNT_RDONLY)
801 * We run vnode_pager_setsize() early (why?),
802 * we must set np->n_size now to avoid vinvalbuf
803 * V_SAVE races that might setsize a lower
806 mtx_lock(&np->n_mtx);
808 mtx_unlock(&np->n_mtx);
809 error = nfs_meta_setsize(vp, ap->a_cred, td,
811 mtx_lock(&np->n_mtx);
812 if (np->n_flag & NMODIFIED) {
814 mtx_unlock(&np->n_mtx);
815 if (vap->va_size == 0)
816 error = nfs_vinvalbuf(vp, 0, td, 1);
818 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
820 vnode_pager_setsize(vp, tsize);
824 mtx_unlock(&np->n_mtx);
826 * np->n_size has already been set to vap->va_size
827 * in nfs_meta_setsize(). We must set it again since
828 * nfs_loadattrcache() could be called through
829 * nfs_meta_setsize() and could modify np->n_size.
831 mtx_lock(&np->n_mtx);
832 np->n_vattr.va_size = np->n_size = vap->va_size;
833 mtx_unlock(&np->n_mtx);
836 mtx_lock(&np->n_mtx);
837 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
838 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
839 mtx_unlock(&np->n_mtx);
840 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
841 (error == EINTR || error == EIO))
844 mtx_unlock(&np->n_mtx);
846 error = nfs_setattrrpc(vp, vap, ap->a_cred);
847 if (error && vap->va_size != VNOVAL) {
848 mtx_lock(&np->n_mtx);
849 np->n_size = np->n_vattr.va_size = tsize;
850 vnode_pager_setsize(vp, tsize);
851 mtx_unlock(&np->n_mtx);
858 * Do an nfs setattr rpc.
861 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
863 struct nfsv2_sattr *sp;
864 struct nfsnode *np = VTONFS(vp);
867 int error = 0, i, wccflag = NFSV3_WCCRATTR;
868 struct mbuf *mreq, *mrep, *md, *mb;
869 int v3 = NFS_ISV3(vp);
871 nfsstats.rpccnt[NFSPROC_SETATTR]++;
872 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
874 bpos = mtod(mb, caddr_t);
877 nfsm_v3attrbuild(vap, TRUE);
878 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
881 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
882 if (vap->va_mode == (mode_t)VNOVAL)
883 sp->sa_mode = nfs_xdrneg1;
885 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
886 if (vap->va_uid == (uid_t)VNOVAL)
887 sp->sa_uid = nfs_xdrneg1;
889 sp->sa_uid = txdr_unsigned(vap->va_uid);
890 if (vap->va_gid == (gid_t)VNOVAL)
891 sp->sa_gid = nfs_xdrneg1;
893 sp->sa_gid = txdr_unsigned(vap->va_gid);
894 sp->sa_size = txdr_unsigned(vap->va_size);
895 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
896 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
898 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
900 mtx_lock(&np->n_mtx);
901 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
902 np->n_accesscache[i].stamp = 0;
903 mtx_unlock(&np->n_mtx);
904 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
905 nfsm_wcc_data(vp, wccflag);
907 nfsm_loadattr(vp, NULL);
914 * nfs lookup call, one step at a time...
915 * First look in cache
916 * If not found, unlock the directory nfsnode and do the rpc
919 nfs_lookup(struct vop_lookup_args *ap)
921 struct componentname *cnp = ap->a_cnp;
922 struct vnode *dvp = ap->a_dvp;
923 struct vnode **vpp = ap->a_vpp;
924 struct mount *mp = dvp->v_mount;
926 int flags = cnp->cn_flags;
928 struct nfsmount *nmp;
930 struct mbuf *mreq, *mrep, *md, *mb;
934 int error = 0, attrflag, fhsize, ltype;
935 int v3 = NFS_ISV3(dvp);
936 struct thread *td = cnp->cn_thread;
939 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
940 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
942 if (dvp->v_type != VDIR)
946 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
950 error = cache_lookup(dvp, vpp, cnp);
951 if (error > 0 && error != ENOENT)
955 * We only accept a positive hit in the cache if the
956 * change time of the file matches our cached copy.
957 * Otherwise, we discard the cache entry and fallback
958 * to doing a lookup RPC.
961 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
962 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
963 nfsstats.lookupcache_hits++;
964 if (cnp->cn_nameiop != LOOKUP &&
966 cnp->cn_flags |= SAVENAME;
975 } else if (error == ENOENT) {
976 if (dvp->v_iflag & VI_DOOMED)
979 * We only accept a negative hit in the cache if the
980 * modification time of the parent directory matches
981 * our cached copy. Otherwise, we discard all of the
982 * negative cache entries for this directory.
984 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
985 vattr.va_mtime.tv_sec == np->n_dmtime) {
986 nfsstats.lookupcache_hits++;
989 cache_purge_negative(dvp);
990 mtx_lock(&np->n_mtx);
992 mtx_unlock(&np->n_mtx);
996 nfsstats.lookupcache_misses++;
997 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
998 len = cnp->cn_namelen;
999 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1000 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1002 bpos = mtod(mb, caddr_t);
1003 nfsm_fhtom(dvp, v3);
1004 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1005 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
1008 nfsm_postop_attr(dvp, attrflag);
1013 nfsm_getfh(fhp, fhsize, v3);
1016 * Handle RENAME case...
1018 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1019 if (NFS_CMPFH(np, fhp, fhsize)) {
1023 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE);
1030 nfsm_postop_attr(newvp, attrflag);
1031 nfsm_postop_attr(dvp, attrflag);
1033 nfsm_loadattr(newvp, NULL);
1036 cnp->cn_flags |= SAVENAME;
1040 if (flags & ISDOTDOT) {
1041 ltype = VOP_ISLOCKED(dvp);
1042 error = vfs_busy(mp, MBF_NOWAIT);
1046 error = vfs_busy(mp, 0);
1047 vn_lock(dvp, ltype | LK_RETRY);
1049 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1059 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1064 vn_lock(dvp, ltype | LK_RETRY);
1065 if (dvp->v_iflag & VI_DOOMED) {
1078 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1082 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1090 nfsm_postop_attr(newvp, attrflag);
1091 nfsm_postop_attr(dvp, attrflag);
1093 nfsm_loadattr(newvp, NULL);
1094 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1095 cnp->cn_flags |= SAVENAME;
1096 if ((cnp->cn_flags & MAKEENTRY) &&
1097 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1098 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
1099 cache_enter(dvp, newvp, cnp);
1105 if (newvp != NULLVP) {
1110 if (error != ENOENT)
1113 /* The requested file was not found. */
1114 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1115 (flags & ISLASTCN)) {
1117 * XXX: UFS does a full VOP_ACCESS(dvp,
1118 * VWRITE) here instead of just checking
1121 if (mp->mnt_flag & MNT_RDONLY)
1123 cnp->cn_flags |= SAVENAME;
1124 return (EJUSTRETURN);
1127 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1129 * Maintain n_dmtime as the modification time
1130 * of the parent directory when the oldest -ve
1131 * name cache entry for this directory was
1134 mtx_lock(&np->n_mtx);
1135 if (np->n_dmtime == 0)
1136 np->n_dmtime = np->n_vattr.va_mtime.tv_sec;
1137 mtx_unlock(&np->n_mtx);
1138 cache_enter(dvp, NULL, cnp);
1148 * Just call nfs_bioread() to do the work.
1151 nfs_read(struct vop_read_args *ap)
1153 struct vnode *vp = ap->a_vp;
1155 switch (vp->v_type) {
1157 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1161 return (EOPNOTSUPP);
1169 nfs_readlink(struct vop_readlink_args *ap)
1171 struct vnode *vp = ap->a_vp;
1173 if (vp->v_type != VLNK)
1175 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1179 * Do a readlink rpc.
1180 * Called by nfs_doio() from below the buffer cache.
1183 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1186 int error = 0, len, attrflag;
1187 struct mbuf *mreq, *mrep, *md, *mb;
1188 int v3 = NFS_ISV3(vp);
1190 nfsstats.rpccnt[NFSPROC_READLINK]++;
1191 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1193 bpos = mtod(mb, caddr_t);
1195 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1197 nfsm_postop_attr(vp, attrflag);
1199 nfsm_strsiz(len, NFS_MAXPATHLEN);
1200 if (len == NFS_MAXPATHLEN) {
1201 struct nfsnode *np = VTONFS(vp);
1202 mtx_lock(&np->n_mtx);
1203 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1205 mtx_unlock(&np->n_mtx);
1207 nfsm_mtouio(uiop, len);
1219 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1223 struct mbuf *mreq, *mrep, *md, *mb;
1224 struct nfsmount *nmp;
1225 int error = 0, len, retlen, tsiz, eof, attrflag;
1226 int v3 = NFS_ISV3(vp);
1232 nmp = VFSTONFS(vp->v_mount);
1233 tsiz = uiop->uio_resid;
1234 mtx_lock(&nmp->nm_mtx);
1235 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1236 mtx_unlock(&nmp->nm_mtx);
1239 rsize = nmp->nm_rsize;
1240 mtx_unlock(&nmp->nm_mtx);
1242 nfsstats.rpccnt[NFSPROC_READ]++;
1243 len = (tsiz > rsize) ? rsize : tsiz;
1244 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1246 bpos = mtod(mb, caddr_t);
1248 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1250 txdr_hyper(uiop->uio_offset, tl);
1251 *(tl + 2) = txdr_unsigned(len);
1253 *tl++ = txdr_unsigned(uiop->uio_offset);
1254 *tl++ = txdr_unsigned(len);
1257 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1259 nfsm_postop_attr(vp, attrflag);
1264 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1265 eof = fxdr_unsigned(int, *(tl + 1));
1267 nfsm_loadattr(vp, NULL);
1269 nfsm_strsiz(retlen, rsize);
1270 nfsm_mtouio(uiop, retlen);
1274 if (eof || retlen == 0) {
1277 } else if (retlen < len) {
1289 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1290 int *iomode, int *must_commit)
1295 struct mbuf *mreq, *mrep, *md, *mb;
1296 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1297 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1298 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1302 if (uiop->uio_iovcnt != 1)
1303 panic("nfs: writerpc iovcnt > 1");
1306 tsiz = uiop->uio_resid;
1307 mtx_lock(&nmp->nm_mtx);
1308 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1309 mtx_unlock(&nmp->nm_mtx);
1312 wsize = nmp->nm_wsize;
1313 mtx_unlock(&nmp->nm_mtx);
1315 nfsstats.rpccnt[NFSPROC_WRITE]++;
1316 len = (tsiz > wsize) ? wsize : tsiz;
1317 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1318 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1320 bpos = mtod(mb, caddr_t);
1323 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1324 txdr_hyper(uiop->uio_offset, tl);
1326 *tl++ = txdr_unsigned(len);
1327 *tl++ = txdr_unsigned(*iomode);
1328 *tl = txdr_unsigned(len);
1332 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1333 /* Set both "begin" and "current" to non-garbage. */
1334 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1335 *tl++ = x; /* "begin offset" */
1336 *tl++ = x; /* "current offset" */
1337 x = txdr_unsigned(len);
1338 *tl++ = x; /* total to this offset */
1339 *tl = x; /* size of this write */
1341 nfsm_uiotom(uiop, len);
1342 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1344 wccflag = NFSV3_WCCCHK;
1345 nfsm_wcc_data(vp, wccflag);
1347 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1348 + NFSX_V3WRITEVERF);
1349 rlen = fxdr_unsigned(int, *tl++);
1354 } else if (rlen < len) {
1355 backup = len - rlen;
1356 uiop->uio_iov->iov_base =
1357 (char *)uiop->uio_iov->iov_base -
1359 uiop->uio_iov->iov_len += backup;
1360 uiop->uio_offset -= backup;
1361 uiop->uio_resid += backup;
1364 commit = fxdr_unsigned(int, *tl++);
1367 * Return the lowest committment level
1368 * obtained by any of the RPCs.
1370 if (committed == NFSV3WRITE_FILESYNC)
1372 else if (committed == NFSV3WRITE_DATASYNC &&
1373 commit == NFSV3WRITE_UNSTABLE)
1375 mtx_lock(&nmp->nm_mtx);
1376 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1377 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1379 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1380 } else if (bcmp((caddr_t)tl,
1381 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1383 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1386 mtx_unlock(&nmp->nm_mtx);
1389 nfsm_loadattr(vp, NULL);
1392 mtx_lock(&(VTONFS(vp))->n_mtx);
1393 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1394 mtx_unlock(&(VTONFS(vp))->n_mtx);
1402 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1403 committed = NFSV3WRITE_FILESYNC;
1404 *iomode = committed;
1406 uiop->uio_resid = tsiz;
1412 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1413 * mode set to specify the file type and the size field for rdev.
1416 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1419 struct nfsv2_sattr *sp;
1421 struct vnode *newvp = NULL;
1422 struct nfsnode *np = NULL;
1425 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1426 struct mbuf *mreq, *mrep, *md, *mb;
1428 int v3 = NFS_ISV3(dvp);
1430 if (vap->va_type == VCHR || vap->va_type == VBLK)
1431 rdev = txdr_unsigned(vap->va_rdev);
1432 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1435 return (EOPNOTSUPP);
1437 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1439 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1440 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1441 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1443 bpos = mtod(mb, caddr_t);
1444 nfsm_fhtom(dvp, v3);
1445 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1447 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1448 *tl++ = vtonfsv3_type(vap->va_type);
1449 nfsm_v3attrbuild(vap, FALSE);
1450 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1451 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1452 *tl++ = txdr_unsigned(major(vap->va_rdev));
1453 *tl = txdr_unsigned(minor(vap->va_rdev));
1456 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1457 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1458 sp->sa_uid = nfs_xdrneg1;
1459 sp->sa_gid = nfs_xdrneg1;
1461 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1462 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1464 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1466 nfsm_mtofh(dvp, newvp, v3, gotvp);
1472 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1473 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1479 nfsm_wcc_data(dvp, wccflag);
1486 if (cnp->cn_flags & MAKEENTRY)
1487 cache_enter(dvp, newvp, cnp);
1490 mtx_lock(&(VTONFS(dvp))->n_mtx);
1491 VTONFS(dvp)->n_flag |= NMODIFIED;
1493 VTONFS(dvp)->n_attrstamp = 0;
1494 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1496 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1502 * just call nfs_mknodrpc() to do the work.
1506 nfs_mknod(struct vop_mknod_args *ap)
1508 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1511 static u_long create_verf;
1513 * nfs file create call
1516 nfs_create(struct vop_create_args *ap)
1518 struct vnode *dvp = ap->a_dvp;
1519 struct vattr *vap = ap->a_vap;
1520 struct componentname *cnp = ap->a_cnp;
1521 struct nfsv2_sattr *sp;
1523 struct nfsnode *np = NULL;
1524 struct vnode *newvp = NULL;
1526 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1527 struct mbuf *mreq, *mrep, *md, *mb;
1529 int v3 = NFS_ISV3(dvp);
1532 * Oops, not for me..
1534 if (vap->va_type == VSOCK)
1535 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1537 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1539 if (vap->va_vaflags & VA_EXCLUSIVE)
1542 nfsstats.rpccnt[NFSPROC_CREATE]++;
1543 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1544 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1546 bpos = mtod(mb, caddr_t);
1547 nfsm_fhtom(dvp, v3);
1548 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1550 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1551 if (fmode & O_EXCL) {
1552 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1553 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1556 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1557 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1560 *tl++ = create_verf;
1562 IN_IFADDR_RUNLOCK();
1564 *tl = ++create_verf;
1566 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1567 nfsm_v3attrbuild(vap, FALSE);
1570 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1571 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1572 sp->sa_uid = nfs_xdrneg1;
1573 sp->sa_gid = nfs_xdrneg1;
1575 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1576 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1578 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1580 nfsm_mtofh(dvp, newvp, v3, gotvp);
1586 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1587 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1593 nfsm_wcc_data(dvp, wccflag);
1597 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1603 } else if (v3 && (fmode & O_EXCL)) {
1605 * We are normally called with only a partially initialized
1606 * VAP. Since the NFSv3 spec says that server may use the
1607 * file attributes to store the verifier, the spec requires
1608 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1609 * in atime, but we can't really assume that all servers will
1610 * so we ensure that our SETATTR sets both atime and mtime.
1612 if (vap->va_mtime.tv_sec == VNOVAL)
1613 vfs_timestamp(&vap->va_mtime);
1614 if (vap->va_atime.tv_sec == VNOVAL)
1615 vap->va_atime = vap->va_mtime;
1616 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred);
1621 if (cnp->cn_flags & MAKEENTRY)
1622 cache_enter(dvp, newvp, cnp);
1625 mtx_lock(&(VTONFS(dvp))->n_mtx);
1626 VTONFS(dvp)->n_flag |= NMODIFIED;
1628 VTONFS(dvp)->n_attrstamp = 0;
1629 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1631 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1636 * nfs file remove call
1637 * To try and make nfs semantics closer to ufs semantics, a file that has
1638 * other processes using the vnode is renamed instead of removed and then
1639 * removed later on the last close.
1640 * - If v_usecount > 1
1641 * If a rename is not already in the works
1642 * call nfs_sillyrename() to set it up
1647 nfs_remove(struct vop_remove_args *ap)
1649 struct vnode *vp = ap->a_vp;
1650 struct vnode *dvp = ap->a_dvp;
1651 struct componentname *cnp = ap->a_cnp;
1652 struct nfsnode *np = VTONFS(vp);
1657 if ((cnp->cn_flags & HASBUF) == 0)
1658 panic("nfs_remove: no name");
1659 if (vrefcnt(vp) < 1)
1660 panic("nfs_remove: bad v_usecount");
1662 if (vp->v_type == VDIR)
1664 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1665 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1667 * Purge the name cache so that the chance of a lookup for
1668 * the name succeeding while the remove is in progress is
1669 * minimized. Without node locking it can still happen, such
1670 * that an I/O op returns ESTALE, but since you get this if
1671 * another host removes the file..
1675 * throw away biocache buffers, mainly to avoid
1676 * unnecessary delayed writes later.
1678 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1680 if (error != EINTR && error != EIO)
1681 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1682 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1684 * Kludge City: If the first reply to the remove rpc is lost..
1685 * the reply to the retransmitted request will be ENOENT
1686 * since the file was in fact removed
1687 * Therefore, we cheat and return success.
1689 if (error == ENOENT)
1691 } else if (!np->n_sillyrename)
1692 error = nfs_sillyrename(dvp, vp, cnp);
1693 np->n_attrstamp = 0;
1694 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1699 * nfs file remove rpc called from nfs_inactive
1702 nfs_removeit(struct sillyrename *sp)
1705 * Make sure that the directory vnode is still valid.
1706 * XXX we should lock sp->s_dvp here.
1708 if (sp->s_dvp->v_type == VBAD)
1710 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1715 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1718 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1719 struct ucred *cred, struct thread *td)
1722 int error = 0, wccflag = NFSV3_WCCRATTR;
1723 struct mbuf *mreq, *mrep, *md, *mb;
1724 int v3 = NFS_ISV3(dvp);
1726 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1727 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1728 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1730 bpos = mtod(mb, caddr_t);
1731 nfsm_fhtom(dvp, v3);
1732 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1733 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1735 nfsm_wcc_data(dvp, wccflag);
1738 mtx_lock(&(VTONFS(dvp))->n_mtx);
1739 VTONFS(dvp)->n_flag |= NMODIFIED;
1741 VTONFS(dvp)->n_attrstamp = 0;
1742 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1744 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1749 * nfs file rename call
1752 nfs_rename(struct vop_rename_args *ap)
1754 struct vnode *fvp = ap->a_fvp;
1755 struct vnode *tvp = ap->a_tvp;
1756 struct vnode *fdvp = ap->a_fdvp;
1757 struct vnode *tdvp = ap->a_tdvp;
1758 struct componentname *tcnp = ap->a_tcnp;
1759 struct componentname *fcnp = ap->a_fcnp;
1763 if ((tcnp->cn_flags & HASBUF) == 0 ||
1764 (fcnp->cn_flags & HASBUF) == 0)
1765 panic("nfs_rename: no name");
1767 /* Check for cross-device rename */
1768 if ((fvp->v_mount != tdvp->v_mount) ||
1769 (tvp && (fvp->v_mount != tvp->v_mount))) {
1775 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1779 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1783 * We have to flush B_DELWRI data prior to renaming
1784 * the file. If we don't, the delayed-write buffers
1785 * can be flushed out later after the file has gone stale
1786 * under NFSV3. NFSV2 does not have this problem because
1787 * ( as far as I can tell ) it flushes dirty buffers more
1790 * Skip the rename operation if the fsync fails, this can happen
1791 * due to the server's volume being full, when we pushed out data
1792 * that was written back to our cache earlier. Not checking for
1793 * this condition can result in potential (silent) data loss.
1795 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1798 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1803 * If the tvp exists and is in use, sillyrename it before doing the
1804 * rename of the new file over it.
1805 * XXX Can't sillyrename a directory.
1807 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1808 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1813 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1814 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1817 if (fvp->v_type == VDIR) {
1818 if (tvp != NULL && tvp->v_type == VDIR)
1833 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1835 if (error == ENOENT)
1841 * nfs file rename rpc called from nfs_remove() above
1844 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1845 struct sillyrename *sp)
1848 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1849 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1853 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1856 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1857 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1861 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1862 struct mbuf *mreq, *mrep, *md, *mb;
1863 int v3 = NFS_ISV3(fdvp);
1865 nfsstats.rpccnt[NFSPROC_RENAME]++;
1866 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1867 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1868 nfsm_rndup(tnamelen));
1870 bpos = mtod(mb, caddr_t);
1871 nfsm_fhtom(fdvp, v3);
1872 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1873 nfsm_fhtom(tdvp, v3);
1874 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1875 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1877 nfsm_wcc_data(fdvp, fwccflag);
1878 nfsm_wcc_data(tdvp, twccflag);
1882 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1883 VTONFS(fdvp)->n_flag |= NMODIFIED;
1884 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1885 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1886 VTONFS(tdvp)->n_flag |= NMODIFIED;
1887 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1889 VTONFS(fdvp)->n_attrstamp = 0;
1890 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1893 VTONFS(tdvp)->n_attrstamp = 0;
1894 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1900 * nfs hard link create call
1903 nfs_link(struct vop_link_args *ap)
1905 struct vnode *vp = ap->a_vp;
1906 struct vnode *tdvp = ap->a_tdvp;
1907 struct componentname *cnp = ap->a_cnp;
1909 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1910 struct mbuf *mreq, *mrep, *md, *mb;
1913 if (vp->v_mount != tdvp->v_mount) {
1918 * Push all writes to the server, so that the attribute cache
1919 * doesn't get "out of sync" with the server.
1920 * XXX There should be a better way!
1922 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1925 nfsstats.rpccnt[NFSPROC_LINK]++;
1926 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1927 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1929 bpos = mtod(mb, caddr_t);
1931 nfsm_fhtom(tdvp, v3);
1932 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1933 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1935 nfsm_postop_attr(vp, attrflag);
1936 nfsm_wcc_data(tdvp, wccflag);
1940 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1941 VTONFS(tdvp)->n_flag |= NMODIFIED;
1942 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1944 VTONFS(vp)->n_attrstamp = 0;
1945 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1948 VTONFS(tdvp)->n_attrstamp = 0;
1949 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1955 * nfs symbolic link create call
1958 nfs_symlink(struct vop_symlink_args *ap)
1960 struct vnode *dvp = ap->a_dvp;
1961 struct vattr *vap = ap->a_vap;
1962 struct componentname *cnp = ap->a_cnp;
1963 struct nfsv2_sattr *sp;
1965 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1966 struct mbuf *mreq, *mrep, *md, *mb;
1967 struct vnode *newvp = NULL;
1968 int v3 = NFS_ISV3(dvp);
1970 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1971 slen = strlen(ap->a_target);
1972 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1973 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1975 bpos = mtod(mb, caddr_t);
1976 nfsm_fhtom(dvp, v3);
1977 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1979 nfsm_v3attrbuild(vap, FALSE);
1981 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1983 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1984 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1985 sp->sa_uid = nfs_xdrneg1;
1986 sp->sa_gid = nfs_xdrneg1;
1987 sp->sa_size = nfs_xdrneg1;
1988 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1989 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1993 * Issue the NFS request and get the rpc response.
1995 * Only NFSv3 responses returning an error of 0 actually return
1996 * a file handle that can be converted into newvp without having
1997 * to do an extra lookup rpc.
1999 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
2002 nfsm_mtofh(dvp, newvp, v3, gotvp);
2003 nfsm_wcc_data(dvp, wccflag);
2007 * out code jumps -> here, mrep is also freed.
2014 * If we do not have an error and we could not extract the newvp from
2015 * the response due to the request being NFSv2, we have to do a
2016 * lookup in order to obtain a newvp to return.
2018 if (error == 0 && newvp == NULL) {
2019 struct nfsnode *np = NULL;
2021 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2022 cnp->cn_cred, cnp->cn_thread, &np);
2032 mtx_lock(&(VTONFS(dvp))->n_mtx);
2033 VTONFS(dvp)->n_flag |= NMODIFIED;
2034 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2036 VTONFS(dvp)->n_attrstamp = 0;
2037 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2046 nfs_mkdir(struct vop_mkdir_args *ap)
2048 struct vnode *dvp = ap->a_dvp;
2049 struct vattr *vap = ap->a_vap;
2050 struct componentname *cnp = ap->a_cnp;
2051 struct nfsv2_sattr *sp;
2053 struct nfsnode *np = NULL;
2054 struct vnode *newvp = NULL;
2056 int error = 0, wccflag = NFSV3_WCCRATTR;
2058 struct mbuf *mreq, *mrep, *md, *mb;
2060 int v3 = NFS_ISV3(dvp);
2062 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2064 len = cnp->cn_namelen;
2065 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2066 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
2067 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2069 bpos = mtod(mb, caddr_t);
2070 nfsm_fhtom(dvp, v3);
2071 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2073 nfsm_v3attrbuild(vap, FALSE);
2075 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2076 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2077 sp->sa_uid = nfs_xdrneg1;
2078 sp->sa_gid = nfs_xdrneg1;
2079 sp->sa_size = nfs_xdrneg1;
2080 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2081 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2083 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2085 nfsm_mtofh(dvp, newvp, v3, gotvp);
2087 nfsm_wcc_data(dvp, wccflag);
2090 mtx_lock(&(VTONFS(dvp))->n_mtx);
2091 VTONFS(dvp)->n_flag |= NMODIFIED;
2092 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2094 VTONFS(dvp)->n_attrstamp = 0;
2095 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2097 if (error == 0 && newvp == NULL) {
2098 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2099 cnp->cn_thread, &np);
2102 if (newvp->v_type != VDIR)
2115 * nfs remove directory call
2118 nfs_rmdir(struct vop_rmdir_args *ap)
2120 struct vnode *vp = ap->a_vp;
2121 struct vnode *dvp = ap->a_dvp;
2122 struct componentname *cnp = ap->a_cnp;
2124 int error = 0, wccflag = NFSV3_WCCRATTR;
2125 struct mbuf *mreq, *mrep, *md, *mb;
2126 int v3 = NFS_ISV3(dvp);
2130 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2131 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
2132 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2134 bpos = mtod(mb, caddr_t);
2135 nfsm_fhtom(dvp, v3);
2136 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2137 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2139 nfsm_wcc_data(dvp, wccflag);
2142 mtx_lock(&(VTONFS(dvp))->n_mtx);
2143 VTONFS(dvp)->n_flag |= NMODIFIED;
2144 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2146 VTONFS(dvp)->n_attrstamp = 0;
2147 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2152 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2154 if (error == ENOENT)
2163 nfs_readdir(struct vop_readdir_args *ap)
2165 struct vnode *vp = ap->a_vp;
2166 struct nfsnode *np = VTONFS(vp);
2167 struct uio *uio = ap->a_uio;
2168 int tresid, error = 0;
2171 if (vp->v_type != VDIR)
2175 * First, check for hit on the EOF offset cache
2177 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2178 (np->n_flag & NMODIFIED) == 0) {
2179 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2180 mtx_lock(&np->n_mtx);
2181 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2182 mtx_unlock(&np->n_mtx);
2183 nfsstats.direofcache_hits++;
2186 mtx_unlock(&np->n_mtx);
2191 * Call nfs_bioread() to do the real work.
2193 tresid = uio->uio_resid;
2194 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2196 if (!error && uio->uio_resid == tresid) {
2197 nfsstats.direofcache_misses++;
2205 * Called from below the buffer cache by nfs_doio().
2208 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2211 struct dirent *dp = NULL;
2216 struct mbuf *mreq, *mrep, *md, *mb;
2218 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2219 struct nfsnode *dnp = VTONFS(vp);
2221 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2223 int v3 = NFS_ISV3(vp);
2226 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2227 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2228 panic("nfs readdirrpc bad uio");
2232 * If there is no cookie, assume directory was stale.
2234 nfs_dircookie_lock(dnp);
2235 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2238 nfs_dircookie_unlock(dnp);
2240 nfs_dircookie_unlock(dnp);
2241 return (NFSERR_BAD_COOKIE);
2245 * Loop around doing readdir rpc's of size nm_readdirsize
2246 * truncated to a multiple of DIRBLKSIZ.
2247 * The stopping criteria is EOF or buffer full.
2249 while (more_dirs && bigenough) {
2250 nfsstats.rpccnt[NFSPROC_READDIR]++;
2251 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2254 bpos = mtod(mb, caddr_t);
2257 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2258 *tl++ = cookie.nfsuquad[0];
2259 *tl++ = cookie.nfsuquad[1];
2260 mtx_lock(&dnp->n_mtx);
2261 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2262 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2263 mtx_unlock(&dnp->n_mtx);
2265 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2266 *tl++ = cookie.nfsuquad[0];
2268 *tl = txdr_unsigned(nmp->nm_readdirsize);
2269 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2271 nfsm_postop_attr(vp, attrflag);
2273 tl = nfsm_dissect(u_int32_t *,
2275 mtx_lock(&dnp->n_mtx);
2276 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2277 dnp->n_cookieverf.nfsuquad[1] = *tl;
2278 mtx_unlock(&dnp->n_mtx);
2284 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2285 more_dirs = fxdr_unsigned(int, *tl);
2287 /* loop thru the dir entries, doctoring them to 4bsd form */
2288 while (more_dirs && bigenough) {
2290 tl = nfsm_dissect(u_int32_t *,
2292 fileno = fxdr_hyper(tl);
2293 len = fxdr_unsigned(int, *(tl + 2));
2295 tl = nfsm_dissect(u_int32_t *,
2297 fileno = fxdr_unsigned(u_quad_t, *tl++);
2298 len = fxdr_unsigned(int, *tl);
2300 if (len <= 0 || len > NFS_MAXNAMLEN) {
2305 tlen = nfsm_rndup(len);
2307 tlen += 4; /* To ensure null termination */
2308 left = DIRBLKSIZ - blksiz;
2309 if ((tlen + DIRHDSIZ) > left) {
2310 dp->d_reclen += left;
2311 uiop->uio_iov->iov_base =
2312 (char *)uiop->uio_iov->iov_base + left;
2313 uiop->uio_iov->iov_len -= left;
2314 uiop->uio_offset += left;
2315 uiop->uio_resid -= left;
2318 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2321 dp = (struct dirent *)uiop->uio_iov->iov_base;
2322 dp->d_fileno = (int)fileno;
2324 dp->d_reclen = tlen + DIRHDSIZ;
2325 dp->d_type = DT_UNKNOWN;
2326 blksiz += dp->d_reclen;
2327 if (blksiz == DIRBLKSIZ)
2329 uiop->uio_offset += DIRHDSIZ;
2330 uiop->uio_resid -= DIRHDSIZ;
2331 uiop->uio_iov->iov_base =
2332 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2333 uiop->uio_iov->iov_len -= DIRHDSIZ;
2334 nfsm_mtouio(uiop, len);
2335 cp = uiop->uio_iov->iov_base;
2337 *cp = '\0'; /* null terminate */
2338 uiop->uio_iov->iov_base =
2339 (char *)uiop->uio_iov->iov_base + tlen;
2340 uiop->uio_iov->iov_len -= tlen;
2341 uiop->uio_offset += tlen;
2342 uiop->uio_resid -= tlen;
2344 nfsm_adv(nfsm_rndup(len));
2346 tl = nfsm_dissect(u_int32_t *,
2349 tl = nfsm_dissect(u_int32_t *,
2353 cookie.nfsuquad[0] = *tl++;
2355 cookie.nfsuquad[1] = *tl++;
2360 more_dirs = fxdr_unsigned(int, *tl);
2363 * If at end of rpc data, get the eof boolean
2366 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2367 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2372 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2373 * by increasing d_reclen for the last record.
2376 left = DIRBLKSIZ - blksiz;
2377 dp->d_reclen += left;
2378 uiop->uio_iov->iov_base =
2379 (char *)uiop->uio_iov->iov_base + left;
2380 uiop->uio_iov->iov_len -= left;
2381 uiop->uio_offset += left;
2382 uiop->uio_resid -= left;
2386 * We are now either at the end of the directory or have filled the
2390 dnp->n_direofoffset = uiop->uio_offset;
2392 if (uiop->uio_resid > 0)
2393 nfs_printf("EEK! readdirrpc resid > 0\n");
2394 nfs_dircookie_lock(dnp);
2395 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2397 nfs_dircookie_unlock(dnp);
2404 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2407 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2413 struct vnode *newvp;
2415 caddr_t bpos, dpos, dpossav1, dpossav2;
2416 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2417 struct nameidata nami, *ndp = &nami;
2418 struct componentname *cnp = &ndp->ni_cnd;
2420 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2421 struct nfsnode *dnp = VTONFS(vp), *np;
2424 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2425 int attrflag, fhsize;
2431 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2432 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2433 panic("nfs readdirplusrpc bad uio");
2439 * If there is no cookie, assume directory was stale.
2441 nfs_dircookie_lock(dnp);
2442 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2445 nfs_dircookie_unlock(dnp);
2447 nfs_dircookie_unlock(dnp);
2448 return (NFSERR_BAD_COOKIE);
2451 * Loop around doing readdir rpc's of size nm_readdirsize
2452 * truncated to a multiple of DIRBLKSIZ.
2453 * The stopping criteria is EOF or buffer full.
2455 while (more_dirs && bigenough) {
2456 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2457 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2458 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2460 bpos = mtod(mb, caddr_t);
2462 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2463 *tl++ = cookie.nfsuquad[0];
2464 *tl++ = cookie.nfsuquad[1];
2465 mtx_lock(&dnp->n_mtx);
2466 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2467 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2468 mtx_unlock(&dnp->n_mtx);
2469 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2470 *tl = txdr_unsigned(nmp->nm_rsize);
2471 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2472 nfsm_postop_attr(vp, attrflag);
2477 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2478 mtx_lock(&dnp->n_mtx);
2479 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2480 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2481 mtx_unlock(&dnp->n_mtx);
2482 more_dirs = fxdr_unsigned(int, *tl);
2484 /* loop thru the dir entries, doctoring them to 4bsd form */
2485 while (more_dirs && bigenough) {
2486 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2487 fileno = fxdr_hyper(tl);
2488 len = fxdr_unsigned(int, *(tl + 2));
2489 if (len <= 0 || len > NFS_MAXNAMLEN) {
2494 tlen = nfsm_rndup(len);
2496 tlen += 4; /* To ensure null termination*/
2497 left = DIRBLKSIZ - blksiz;
2498 if ((tlen + DIRHDSIZ) > left) {
2499 dp->d_reclen += left;
2500 uiop->uio_iov->iov_base =
2501 (char *)uiop->uio_iov->iov_base + left;
2502 uiop->uio_iov->iov_len -= left;
2503 uiop->uio_offset += left;
2504 uiop->uio_resid -= left;
2507 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2510 dp = (struct dirent *)uiop->uio_iov->iov_base;
2511 dp->d_fileno = (int)fileno;
2513 dp->d_reclen = tlen + DIRHDSIZ;
2514 dp->d_type = DT_UNKNOWN;
2515 blksiz += dp->d_reclen;
2516 if (blksiz == DIRBLKSIZ)
2518 uiop->uio_offset += DIRHDSIZ;
2519 uiop->uio_resid -= DIRHDSIZ;
2520 uiop->uio_iov->iov_base =
2521 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2522 uiop->uio_iov->iov_len -= DIRHDSIZ;
2523 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2524 cnp->cn_namelen = len;
2525 nfsm_mtouio(uiop, len);
2526 cp = uiop->uio_iov->iov_base;
2529 uiop->uio_iov->iov_base =
2530 (char *)uiop->uio_iov->iov_base + tlen;
2531 uiop->uio_iov->iov_len -= tlen;
2532 uiop->uio_offset += tlen;
2533 uiop->uio_resid -= tlen;
2535 nfsm_adv(nfsm_rndup(len));
2536 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2538 cookie.nfsuquad[0] = *tl++;
2539 cookie.nfsuquad[1] = *tl++;
2544 * Since the attributes are before the file handle
2545 * (sigh), we must skip over the attributes and then
2546 * come back and get them.
2548 attrflag = fxdr_unsigned(int, *tl);
2552 nfsm_adv(NFSX_V3FATTR);
2553 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2554 doit = fxdr_unsigned(int, *tl);
2556 * Skip loading the attrs for "..". There's a
2557 * race between loading the attrs here and
2558 * lookups that look for the directory currently
2559 * being read (in the parent). We try to acquire
2560 * the exclusive lock on ".." here, owning the
2561 * lock on the directory being read. Lookup will
2562 * hold the lock on ".." and try to acquire the
2563 * lock on the directory being read.
2565 * There are other ways of fixing this, one would
2566 * be to do a trylock on the ".." vnode and skip
2567 * loading the attrs on ".." if it happens to be
2568 * locked by another process. But skipping the
2569 * attrload on ".." seems the easiest option.
2571 if (strcmp(dp->d_name, "..") == 0) {
2574 * We've already skipped over the attrs,
2575 * skip over the filehandle. And store d_type
2578 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2579 i = fxdr_unsigned(int, *tl);
2580 nfsm_adv(nfsm_rndup(i));
2581 dp->d_type = IFTODT(VTTOIF(VDIR));
2584 nfsm_getfh(fhp, fhsize, 1);
2585 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2590 error = nfs_nget(vp->v_mount, fhp,
2591 fhsize, &np, LK_EXCLUSIVE);
2598 if (doit && bigenough) {
2603 nfsm_loadattr(newvp, NULL);
2607 IFTODT(VTTOIF(np->n_vattr.va_type));
2609 /* Update n_ctime, so subsequent lookup doesn't purge entry */
2610 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
2611 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2614 /* Just skip over the file handle */
2615 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2616 i = fxdr_unsigned(int, *tl);
2618 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2619 fhsize = fxdr_unsigned(int, *tl);
2620 nfsm_adv(nfsm_rndup(fhsize));
2623 if (newvp != NULLVP) {
2630 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2631 more_dirs = fxdr_unsigned(int, *tl);
2634 * If at end of rpc data, get the eof boolean
2637 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2638 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2643 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2644 * by increasing d_reclen for the last record.
2647 left = DIRBLKSIZ - blksiz;
2648 dp->d_reclen += left;
2649 uiop->uio_iov->iov_base =
2650 (char *)uiop->uio_iov->iov_base + left;
2651 uiop->uio_iov->iov_len -= left;
2652 uiop->uio_offset += left;
2653 uiop->uio_resid -= left;
2657 * We are now either at the end of the directory or have filled the
2661 dnp->n_direofoffset = uiop->uio_offset;
2663 if (uiop->uio_resid > 0)
2664 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2665 nfs_dircookie_lock(dnp);
2666 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2668 nfs_dircookie_unlock(dnp);
2671 if (newvp != NULLVP) {
2682 * Silly rename. To make the NFS filesystem that is stateless look a little
2683 * more like the "ufs" a remove of an active vnode is translated to a rename
2684 * to a funny looking filename that is removed by nfs_inactive on the
2685 * nfsnode. There is the potential for another process on a different client
2686 * to create the same funny name between the nfs_lookitup() fails and the
2687 * nfs_rename() completes, but...
2690 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2692 struct sillyrename *sp;
2696 unsigned int lticks;
2701 if (vp->v_type == VDIR)
2702 panic("nfs: sillyrename dir");
2704 sp = malloc(sizeof (struct sillyrename),
2705 M_NFSREQ, M_WAITOK);
2706 sp->s_cred = crhold(cnp->cn_cred);
2708 sp->s_removeit = nfs_removeit;
2712 * Fudge together a funny name.
2713 * Changing the format of the funny name to accomodate more
2714 * sillynames per directory.
2715 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2716 * CPU ticks since boot.
2718 pid = cnp->cn_thread->td_proc->p_pid;
2719 lticks = (unsigned int)ticks;
2721 sp->s_namlen = sprintf(sp->s_name,
2722 ".nfs.%08x.%04x4.4", lticks,
2724 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2725 cnp->cn_thread, NULL))
2729 error = nfs_renameit(dvp, cnp, sp);
2732 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2733 cnp->cn_thread, &np);
2734 np->n_sillyrename = sp;
2739 free((caddr_t)sp, M_NFSREQ);
2744 * Look up a file name and optionally either update the file handle or
2745 * allocate an nfsnode, depending on the value of npp.
2746 * npp == NULL --> just do the lookup
2747 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2749 * *npp != NULL --> update the file handle in the vnode
2752 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2753 struct thread *td, struct nfsnode **npp)
2755 struct vnode *newvp = NULL;
2756 struct nfsnode *np, *dnp = VTONFS(dvp);
2758 int error = 0, fhlen, attrflag;
2759 struct mbuf *mreq, *mrep, *md, *mb;
2761 int v3 = NFS_ISV3(dvp);
2763 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2764 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2765 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2767 bpos = mtod(mb, caddr_t);
2768 nfsm_fhtom(dvp, v3);
2769 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2770 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2771 if (npp && !error) {
2772 nfsm_getfh(nfhp, fhlen, v3);
2775 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2776 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2777 np->n_fhp = &np->n_fh;
2778 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2779 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2780 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2781 np->n_fhsize = fhlen;
2783 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2787 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2795 nfsm_postop_attr(newvp, attrflag);
2796 if (!attrflag && *npp == NULL) {
2805 nfsm_loadattr(newvp, NULL);
2809 if (npp && *npp == NULL) {
2824 * Nfs Version 3 commit rpc
2827 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2831 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2833 int error = 0, wccflag = NFSV3_WCCRATTR;
2834 struct mbuf *mreq, *mrep, *md, *mb;
2836 mtx_lock(&nmp->nm_mtx);
2837 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2838 mtx_unlock(&nmp->nm_mtx);
2841 mtx_unlock(&nmp->nm_mtx);
2842 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2843 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2845 bpos = mtod(mb, caddr_t);
2847 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2848 txdr_hyper(offset, tl);
2850 *tl = txdr_unsigned(cnt);
2851 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2852 nfsm_wcc_data(vp, wccflag);
2854 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2855 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2856 NFSX_V3WRITEVERF)) {
2857 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2859 error = NFSERR_STALEWRITEVERF;
2869 * For async requests when nfsiod(s) are running, queue the request by
2870 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2874 nfs_strategy(struct vop_strategy_args *ap)
2876 struct buf *bp = ap->a_bp;
2879 KASSERT(!(bp->b_flags & B_DONE),
2880 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2881 BUF_ASSERT_HELD(bp);
2883 if (bp->b_iocmd == BIO_READ)
2889 * If the op is asynchronous and an i/o daemon is waiting
2890 * queue the request, wake it up and wait for completion
2891 * otherwise just do it ourselves.
2893 if ((bp->b_flags & B_ASYNC) == 0 ||
2894 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2895 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2900 * fsync vnode op. Just call nfs_flush() with commit == 1.
2904 nfs_fsync(struct vop_fsync_args *ap)
2907 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1));
2911 * Flush all the blocks associated with a vnode.
2912 * Walk through the buffer pool and push any dirty pages
2913 * associated with the vnode.
2916 nfs_flush(struct vnode *vp, int waitfor, int commit)
2918 struct nfsnode *np = VTONFS(vp);
2922 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2923 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2925 u_quad_t off, endoff, toff;
2926 struct ucred* wcred = NULL;
2927 struct buf **bvec = NULL;
2929 struct thread *td = curthread;
2930 #ifndef NFS_COMMITBVECSIZ
2931 #define NFS_COMMITBVECSIZ 20
2933 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2934 int bvecsize = 0, bveccount;
2936 if (nmp->nm_flag & NFSMNT_INT)
2937 slpflag = NFS_PCATCH;
2942 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2943 * server, but has not been committed to stable storage on the server
2944 * yet. On the first pass, the byte range is worked out and the commit
2945 * rpc is done. On the second pass, nfs_writebp() is called to do the
2952 if (NFS_ISV3(vp) && commit) {
2953 if (bvec != NULL && bvec != bvec_on_stack)
2956 * Count up how many buffers waiting for a commit.
2960 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2961 if (!BUF_ISLOCKED(bp) &&
2962 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2963 == (B_DELWRI | B_NEEDCOMMIT))
2967 * Allocate space to remember the list of bufs to commit. It is
2968 * important to use M_NOWAIT here to avoid a race with nfs_write.
2969 * If we can't get memory (for whatever reason), we will end up
2970 * committing the buffers one-by-one in the loop below.
2972 if (bveccount > NFS_COMMITBVECSIZ) {
2974 * Release the vnode interlock to avoid a lock
2978 bvec = (struct buf **)
2979 malloc(bveccount * sizeof(struct buf *),
2983 bvec = bvec_on_stack;
2984 bvecsize = NFS_COMMITBVECSIZ;
2986 bvecsize = bveccount;
2988 bvec = bvec_on_stack;
2989 bvecsize = NFS_COMMITBVECSIZ;
2991 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2992 if (bvecpos >= bvecsize)
2994 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2995 nbp = TAILQ_NEXT(bp, b_bobufs);
2998 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2999 (B_DELWRI | B_NEEDCOMMIT)) {
3001 nbp = TAILQ_NEXT(bp, b_bobufs);
3007 * Work out if all buffers are using the same cred
3008 * so we can deal with them all with one commit.
3010 * NOTE: we are not clearing B_DONE here, so we have
3011 * to do it later on in this routine if we intend to
3012 * initiate I/O on the bp.
3014 * Note: to avoid loopback deadlocks, we do not
3015 * assign b_runningbufspace.
3018 wcred = bp->b_wcred;
3019 else if (wcred != bp->b_wcred)
3021 vfs_busy_pages(bp, 1);
3025 * bp is protected by being locked, but nbp is not
3026 * and vfs_busy_pages() may sleep. We have to
3029 nbp = TAILQ_NEXT(bp, b_bobufs);
3032 * A list of these buffers is kept so that the
3033 * second loop knows which buffers have actually
3034 * been committed. This is necessary, since there
3035 * may be a race between the commit rpc and new
3036 * uncommitted writes on the file.
3038 bvec[bvecpos++] = bp;
3039 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3043 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3051 * Commit data on the server, as required.
3052 * If all bufs are using the same wcred, then use that with
3053 * one call for all of them, otherwise commit each one
3056 if (wcred != NOCRED)
3057 retv = nfs_commit(vp, off, (int)(endoff - off),
3061 for (i = 0; i < bvecpos; i++) {
3064 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3066 size = (u_quad_t)(bp->b_dirtyend
3068 retv = nfs_commit(vp, off, (int)size,
3074 if (retv == NFSERR_STALEWRITEVERF)
3075 nfs_clearcommit(vp->v_mount);
3078 * Now, either mark the blocks I/O done or mark the
3079 * blocks dirty, depending on whether the commit
3082 for (i = 0; i < bvecpos; i++) {
3084 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3087 * Error, leave B_DELWRI intact
3089 vfs_unbusy_pages(bp);
3093 * Success, remove B_DELWRI ( bundirty() ).
3095 * b_dirtyoff/b_dirtyend seem to be NFS
3096 * specific. We should probably move that
3097 * into bundirty(). XXX
3100 bp->b_flags |= B_ASYNC;
3102 bp->b_flags &= ~B_DONE;
3103 bp->b_ioflags &= ~BIO_ERROR;
3104 bp->b_dirtyoff = bp->b_dirtyend = 0;
3111 * Start/do any write(s) that are required.
3115 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3116 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3117 if (waitfor != MNT_WAIT || passone)
3120 error = BUF_TIMELOCK(bp,
3121 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3122 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
3127 if (error == ENOLCK) {
3131 if (nfs_sigintr(nmp, td)) {
3135 if (slpflag & PCATCH) {
3141 if ((bp->b_flags & B_DELWRI) == 0)
3142 panic("nfs_fsync: not dirty");
3143 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3149 if (passone || !commit)
3150 bp->b_flags |= B_ASYNC;
3152 bp->b_flags |= B_ASYNC;
3154 if (nfs_sigintr(nmp, td)) {
3165 if (waitfor == MNT_WAIT) {
3166 while (bo->bo_numoutput) {
3167 error = bufobj_wwait(bo, slpflag, slptimeo);
3170 error = nfs_sigintr(nmp, td);
3173 if (slpflag & PCATCH) {
3180 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3185 * Wait for all the async IO requests to drain
3188 mtx_lock(&np->n_mtx);
3189 while (np->n_directio_asyncwr > 0) {
3190 np->n_flag |= NFSYNCWAIT;
3191 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3192 &np->n_mtx, slpflag | (PRIBIO + 1),
3195 if (nfs_sigintr(nmp, td)) {
3196 mtx_unlock(&np->n_mtx);
3202 mtx_unlock(&np->n_mtx);
3205 mtx_lock(&np->n_mtx);
3206 if (np->n_flag & NWRITEERR) {
3207 error = np->n_error;
3208 np->n_flag &= ~NWRITEERR;
3210 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3211 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3212 np->n_flag &= ~NMODIFIED;
3213 mtx_unlock(&np->n_mtx);
3215 if (bvec != NULL && bvec != bvec_on_stack)
3221 * NFS advisory byte-level locks.
3224 nfs_advlock(struct vop_advlock_args *ap)
3226 struct vnode *vp = ap->a_vp;
3230 error = vn_lock(vp, LK_SHARED);
3233 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3234 size = VTONFS(vp)->n_size;
3236 error = lf_advlock(ap, &(vp->v_lockf), size);
3239 error = nfs_advlock_p(ap);
3248 * NFS advisory byte-level locks.
3251 nfs_advlockasync(struct vop_advlockasync_args *ap)
3253 struct vnode *vp = ap->a_vp;
3257 error = vn_lock(vp, LK_SHARED);
3260 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3261 size = VTONFS(vp)->n_size;
3263 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3272 * Print out the contents of an nfsnode.
3275 nfs_print(struct vop_print_args *ap)
3277 struct vnode *vp = ap->a_vp;
3278 struct nfsnode *np = VTONFS(vp);
3280 nfs_printf("\tfileid %ld fsid 0x%x",
3281 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3282 if (vp->v_type == VFIFO)
3289 * This is the "real" nfs::bwrite(struct buf*).
3290 * We set B_CACHE if this is a VMIO buffer.
3293 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3296 int oldflags = bp->b_flags;
3302 BUF_ASSERT_HELD(bp);
3304 if (bp->b_flags & B_INVAL) {
3309 bp->b_flags |= B_CACHE;
3312 * Undirty the bp. We will redirty it later if the I/O fails.
3317 bp->b_flags &= ~B_DONE;
3318 bp->b_ioflags &= ~BIO_ERROR;
3319 bp->b_iocmd = BIO_WRITE;
3321 bufobj_wref(bp->b_bufobj);
3322 curthread->td_ru.ru_oublock++;
3326 * Note: to avoid loopback deadlocks, we do not
3327 * assign b_runningbufspace.
3329 vfs_busy_pages(bp, 1);
3332 bp->b_iooffset = dbtob(bp->b_blkno);
3335 if( (oldflags & B_ASYNC) == 0) {
3336 int rtval = bufwait(bp);
3338 if (oldflags & B_DELWRI) {
3351 * nfs special file access vnode op.
3352 * Essentially just get vattr and then imitate iaccess() since the device is
3353 * local to the client.
3356 nfsspec_access(struct vop_access_args *ap)
3359 struct ucred *cred = ap->a_cred;
3360 struct vnode *vp = ap->a_vp;
3361 accmode_t accmode = ap->a_accmode;
3366 * Disallow write attempts on filesystems mounted read-only;
3367 * unless the file is a socket, fifo, or a block or character
3368 * device resident on the filesystem.
3370 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3371 switch (vp->v_type) {
3381 error = VOP_GETATTR(vp, vap, cred);
3384 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3385 accmode, cred, NULL);
3391 * Read wrapper for fifos.
3394 nfsfifo_read(struct vop_read_args *ap)
3396 struct nfsnode *np = VTONFS(ap->a_vp);
3402 mtx_lock(&np->n_mtx);
3404 getnanotime(&np->n_atim);
3405 mtx_unlock(&np->n_mtx);
3406 error = fifo_specops.vop_read(ap);
3411 * Write wrapper for fifos.
3414 nfsfifo_write(struct vop_write_args *ap)
3416 struct nfsnode *np = VTONFS(ap->a_vp);
3421 mtx_lock(&np->n_mtx);
3423 getnanotime(&np->n_mtim);
3424 mtx_unlock(&np->n_mtx);
3425 return(fifo_specops.vop_write(ap));
3429 * Close wrapper for fifos.
3431 * Update the times on the nfsnode then do fifo close.
3434 nfsfifo_close(struct vop_close_args *ap)
3436 struct vnode *vp = ap->a_vp;
3437 struct nfsnode *np = VTONFS(vp);
3441 mtx_lock(&np->n_mtx);
3442 if (np->n_flag & (NACC | NUPD)) {
3444 if (np->n_flag & NACC)
3446 if (np->n_flag & NUPD)
3449 if (vrefcnt(vp) == 1 &&
3450 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3452 if (np->n_flag & NACC)
3453 vattr.va_atime = np->n_atim;
3454 if (np->n_flag & NUPD)
3455 vattr.va_mtime = np->n_mtim;
3456 mtx_unlock(&np->n_mtx);
3457 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3461 mtx_unlock(&np->n_mtx);
3463 return (fifo_specops.vop_close(ap));
3467 * Just call nfs_writebp() with the force argument set to 1.
3469 * NOTE: B_DONE may or may not be set in a_bp on call.
3472 nfs_bwrite(struct buf *bp)
3475 return (nfs_writebp(bp, 1, curthread));
3478 struct buf_ops buf_ops_nfs = {
3479 .bop_name = "buf_ops_nfs",
3480 .bop_write = nfs_bwrite,
3481 .bop_strategy = bufstrategy,
3482 .bop_sync = bufsync,
3483 .bop_bdflush = bufbdflush,