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 * from 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, 3 and 4
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/resourcevar.h>
49 #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/nfs/nfsport.h>
71 #include <fs/nfsclient/nfsnode.h>
72 #include <fs/nfsclient/nfsmount.h>
73 #include <fs/nfsclient/nfs.h>
76 #include <netinet/in.h>
77 #include <netinet/in_var.h>
79 #include <nfs/nfs_lock.h>
85 extern struct nfsstats newnfsstats;
86 MALLOC_DECLARE(M_NEWNFSREQ);
89 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
90 * calls are not in getblk() and brelse() so that they would not be necessary
94 #define vfs_busy_pages(bp, f)
97 static vop_read_t nfsfifo_read;
98 static vop_write_t nfsfifo_write;
99 static vop_close_t nfsfifo_close;
100 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
102 static vop_lookup_t nfs_lookup;
103 static vop_create_t nfs_create;
104 static vop_mknod_t nfs_mknod;
105 static vop_open_t nfs_open;
106 static vop_close_t nfs_close;
107 static vop_access_t nfs_access;
108 static vop_getattr_t nfs_getattr;
109 static vop_setattr_t nfs_setattr;
110 static vop_read_t nfs_read;
111 static vop_fsync_t nfs_fsync;
112 static vop_remove_t nfs_remove;
113 static vop_link_t nfs_link;
114 static vop_rename_t nfs_rename;
115 static vop_mkdir_t nfs_mkdir;
116 static vop_rmdir_t nfs_rmdir;
117 static vop_symlink_t nfs_symlink;
118 static vop_readdir_t nfs_readdir;
119 static vop_strategy_t nfs_strategy;
120 static vop_lock1_t nfs_lock1;
121 static int nfs_lookitup(struct vnode *, char *, int,
122 struct ucred *, struct thread *, struct nfsnode **);
123 static int nfs_sillyrename(struct vnode *, struct vnode *,
124 struct componentname *);
125 static vop_access_t nfsspec_access;
126 static vop_readlink_t nfs_readlink;
127 static vop_print_t nfs_print;
128 static vop_advlock_t nfs_advlock;
129 static vop_advlockasync_t nfs_advlockasync;
130 #ifdef NFS4_ACL_EXTATTR_NAME
131 static vop_getacl_t nfs_getacl;
132 static vop_setacl_t nfs_setacl;
136 * Global vfs data structures for nfs
138 struct vop_vector newnfs_vnodeops = {
139 .vop_default = &default_vnodeops,
140 .vop_access = nfs_access,
141 .vop_advlock = nfs_advlock,
142 .vop_advlockasync = nfs_advlockasync,
143 .vop_close = nfs_close,
144 .vop_create = nfs_create,
145 .vop_fsync = nfs_fsync,
146 .vop_getattr = nfs_getattr,
147 .vop_getpages = ncl_getpages,
148 .vop_putpages = ncl_putpages,
149 .vop_inactive = ncl_inactive,
150 .vop_link = nfs_link,
151 .vop_lock1 = nfs_lock1,
152 .vop_lookup = nfs_lookup,
153 .vop_mkdir = nfs_mkdir,
154 .vop_mknod = nfs_mknod,
155 .vop_open = nfs_open,
156 .vop_print = nfs_print,
157 .vop_read = nfs_read,
158 .vop_readdir = nfs_readdir,
159 .vop_readlink = nfs_readlink,
160 .vop_reclaim = ncl_reclaim,
161 .vop_remove = nfs_remove,
162 .vop_rename = nfs_rename,
163 .vop_rmdir = nfs_rmdir,
164 .vop_setattr = nfs_setattr,
165 .vop_strategy = nfs_strategy,
166 .vop_symlink = nfs_symlink,
167 .vop_write = ncl_write,
168 #ifdef NFS4_ACL_EXTATTR_NAME
169 .vop_getacl = nfs_getacl,
170 .vop_setacl = nfs_setacl,
174 struct vop_vector newnfs_fifoops = {
175 .vop_default = &fifo_specops,
176 .vop_access = nfsspec_access,
177 .vop_close = nfsfifo_close,
178 .vop_fsync = nfs_fsync,
179 .vop_getattr = nfs_getattr,
180 .vop_inactive = ncl_inactive,
181 .vop_print = nfs_print,
182 .vop_read = nfsfifo_read,
183 .vop_reclaim = ncl_reclaim,
184 .vop_setattr = nfs_setattr,
185 .vop_write = nfsfifo_write,
188 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
189 struct componentname *cnp, struct vattr *vap);
190 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
191 int namelen, struct ucred *cred, struct thread *td);
192 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
193 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
194 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
195 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
196 struct componentname *scnp, struct sillyrename *sp);
201 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
203 SYSCTL_DECL(_vfs_newnfs);
205 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
206 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
207 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
209 static int nfs_prime_access_cache = 0;
210 SYSCTL_INT(_vfs_newnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
211 &nfs_prime_access_cache, 0,
212 "Prime NFS ACCESS cache when fetching attributes");
214 static int newnfs_commit_on_close = 0;
215 SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
216 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
218 static int nfs_clean_pages_on_close = 1;
219 SYSCTL_INT(_vfs_newnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
220 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
222 int newnfs_directio_enable = 0;
223 SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_enable, CTLFLAG_RW,
224 &newnfs_directio_enable, 0, "Enable NFS directio");
227 * This sysctl allows other processes to mmap a file that has been opened
228 * O_DIRECT by a process. In general, having processes mmap the file while
229 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
230 * this by default to prevent DoS attacks - to prevent a malicious user from
231 * opening up files O_DIRECT preventing other users from mmap'ing these
232 * files. "Protected" environments where stricter consistency guarantees are
233 * required can disable this knob. The process that opened the file O_DIRECT
234 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
237 int newnfs_directio_allow_mmap = 1;
238 SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_allow_mmap, CTLFLAG_RW,
239 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
242 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
243 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
245 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
246 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
249 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
250 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
251 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
255 * The list of locks after the description of the lock is the ordering
256 * of other locks acquired with the lock held.
257 * np->n_mtx : Protects the fields in the nfsnode.
259 VI_MTX (acquired indirectly)
260 * nmp->nm_mtx : Protects the fields in the nfsmount.
262 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
263 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
266 * rep->r_mtx : Protects the fields in an nfsreq.
270 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
271 struct ucred *cred, u_int32_t *retmode)
273 int error = 0, attrflag, i, lrupos;
275 struct nfsnode *np = VTONFS(vp);
276 struct nfsvattr nfsva;
278 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
281 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
284 mtx_lock(&np->n_mtx);
285 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
286 if (np->n_accesscache[i].uid == cred->cr_uid) {
287 np->n_accesscache[i].mode = rmode;
288 np->n_accesscache[i].stamp = time_second;
291 if (i > 0 && np->n_accesscache[i].stamp <
292 np->n_accesscache[lrupos].stamp)
295 if (i == NFS_ACCESSCACHESIZE) {
296 np->n_accesscache[lrupos].uid = cred->cr_uid;
297 np->n_accesscache[lrupos].mode = rmode;
298 np->n_accesscache[lrupos].stamp = time_second;
300 mtx_unlock(&np->n_mtx);
303 } else if (NFS_ISV4(vp)) {
304 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
310 * nfs access vnode op.
311 * For nfs version 2, just return ok. File accesses may fail later.
312 * For nfs version 3, use the access rpc to check accessibility. If file modes
313 * are changed on the server, accesses might still fail later.
316 nfs_access(struct vop_access_args *ap)
318 struct vnode *vp = ap->a_vp;
319 int error = 0, i, gotahit;
320 u_int32_t mode, wmode, rmode;
321 int v34 = NFS_ISV34(vp);
322 struct nfsnode *np = VTONFS(vp);
325 * Disallow write attempts on filesystems mounted read-only;
326 * unless the file is a socket, fifo, or a block or character
327 * device resident on the filesystem.
329 if ((ap->a_accmode & (VWRITE | VAPPEND
330 #ifdef NFS4_ACL_EXTATTR_NAME
331 | VWRITE_NAMED_ATTRS | VDELETE_CHILD | VWRITE_ATTRIBUTES |
332 VDELETE | VWRITE_ACL | VWRITE_OWNER
334 )) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
335 switch (vp->v_type) {
345 * For nfs v3 or v4, check to see if we have done this recently, and if
346 * so return our cached result instead of making an ACCESS call.
347 * If not, do an access rpc, otherwise you are stuck emulating
348 * ufs_access() locally using the vattr. This may not be correct,
349 * since the server may apply other access criteria such as
350 * client uid-->server uid mapping that we do not know about.
353 if (ap->a_accmode & VREAD)
354 mode = NFSACCESS_READ;
357 if (vp->v_type != VDIR) {
358 if (ap->a_accmode & VWRITE)
359 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
360 if (ap->a_accmode & VAPPEND)
361 mode |= NFSACCESS_EXTEND;
362 if (ap->a_accmode & VEXEC)
363 mode |= NFSACCESS_EXECUTE;
364 #ifdef NFS4_ACL_EXTATTR_NAME
365 if (ap->a_accmode & VDELETE)
366 mode |= NFSACCESS_DELETE;
369 if (ap->a_accmode & VWRITE)
370 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
371 if (ap->a_accmode & VAPPEND)
372 mode |= NFSACCESS_EXTEND;
373 if (ap->a_accmode & VEXEC)
374 mode |= NFSACCESS_LOOKUP;
375 #ifdef NFS4_ACL_EXTATTR_NAME
376 if (ap->a_accmode & VDELETE)
377 mode |= NFSACCESS_DELETE;
378 if (ap->a_accmode & VDELETE_CHILD)
379 mode |= NFSACCESS_MODIFY;
382 /* XXX safety belt, only make blanket request if caching */
383 if (nfsaccess_cache_timeout > 0) {
384 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
385 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
386 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
392 * Does our cached result allow us to give a definite yes to
396 mtx_lock(&np->n_mtx);
397 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
398 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
399 if (time_second < (np->n_accesscache[i].stamp
400 + nfsaccess_cache_timeout) &&
401 (np->n_accesscache[i].mode & mode) == mode) {
402 NFSINCRGLOBAL(newnfsstats.accesscache_hits);
408 mtx_unlock(&np->n_mtx);
411 * Either a no, or a don't know. Go to the wire.
413 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
414 error = nfs34_access_otw(vp, wmode, ap->a_td,
417 (rmode & mode) != mode)
422 if ((error = nfsspec_access(ap)) != 0) {
426 * Attempt to prevent a mapped root from accessing a file
427 * which it shouldn't. We try to read a byte from the file
428 * if the user is root and the file is not zero length.
429 * After calling nfsspec_access, we should have the correct
432 mtx_lock(&np->n_mtx);
433 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
434 && VTONFS(vp)->n_size > 0) {
439 mtx_unlock(&np->n_mtx);
442 auio.uio_iov = &aiov;
446 auio.uio_segflg = UIO_SYSSPACE;
447 auio.uio_rw = UIO_READ;
448 auio.uio_td = ap->a_td;
450 if (vp->v_type == VREG)
451 error = ncl_readrpc(vp, &auio, ap->a_cred);
452 else if (vp->v_type == VDIR) {
454 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
456 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
457 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
460 } else if (vp->v_type == VLNK)
461 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
465 mtx_unlock(&np->n_mtx);
473 * Check to see if the type is ok
474 * and that deletion is not in progress.
475 * For paged in text files, you will need to flush the page cache
476 * if consistency is lost.
480 nfs_open(struct vop_open_args *ap)
482 struct vnode *vp = ap->a_vp;
483 struct nfsnode *np = VTONFS(vp);
486 int fmode = ap->a_mode;
488 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
492 * For NFSv4, we need to do the Open Op before cache validation,
493 * so that we conform to RFC3530 Sec. 9.3.1.
496 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
498 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
505 * Now, if this Open will be doing reading, re-validate/flush the
506 * cache, so that Close/Open coherency is maintained.
508 if ((fmode & FREAD) != 0 &&
509 (!NFS_ISV4(vp) || nfscl_mustflush(vp) != 0)) {
510 mtx_lock(&np->n_mtx);
511 if (np->n_flag & NMODIFIED) {
512 mtx_unlock(&np->n_mtx);
513 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
514 if (error == EINTR || error == EIO) {
516 (void) nfsrpc_close(vp, 0, ap->a_td);
519 mtx_lock(&np->n_mtx);
521 if (vp->v_type == VDIR)
522 np->n_direofoffset = 0;
523 mtx_unlock(&np->n_mtx);
524 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
527 (void) nfsrpc_close(vp, 0, ap->a_td);
530 mtx_lock(&np->n_mtx);
531 np->n_mtime = vattr.va_mtime;
533 np->n_change = vattr.va_filerev;
534 mtx_unlock(&np->n_mtx);
536 mtx_unlock(&np->n_mtx);
537 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
540 (void) nfsrpc_close(vp, 0, ap->a_td);
543 mtx_lock(&np->n_mtx);
544 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
545 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 = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
550 if (error == EINTR || error == EIO) {
552 (void) nfsrpc_close(vp, 0,
556 mtx_lock(&np->n_mtx);
557 np->n_mtime = vattr.va_mtime;
559 np->n_change = vattr.va_filerev;
561 mtx_unlock(&np->n_mtx);
566 * If the object has >= 1 O_DIRECT active opens, we disable caching.
568 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
569 if (np->n_directio_opens == 0) {
570 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
573 (void) nfsrpc_close(vp, 0, ap->a_td);
576 mtx_lock(&np->n_mtx);
577 np->n_flag |= NNONCACHE;
579 mtx_lock(&np->n_mtx);
581 np->n_directio_opens++;
582 mtx_unlock(&np->n_mtx);
584 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
590 * What an NFS client should do upon close after writing is a debatable issue.
591 * Most NFS clients push delayed writes to the server upon close, basically for
593 * 1 - So that any write errors may be reported back to the client process
594 * doing the close system call. By far the two most likely errors are
595 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
596 * 2 - To put a worst case upper bound on cache inconsistency between
597 * multiple clients for the file.
598 * There is also a consistency problem for Version 2 of the protocol w.r.t.
599 * not being able to tell if other clients are writing a file concurrently,
600 * since there is no way of knowing if the changed modify time in the reply
601 * is only due to the write for this client.
602 * (NFS Version 3 provides weak cache consistency data in the reply that
603 * should be sufficient to detect and handle this case.)
605 * The current code does the following:
606 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
607 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
608 * or commit them (this satisfies 1 and 2 except for the
609 * case where the server crashes after this close but
610 * before the commit RPC, which is felt to be "good
611 * enough". Changing the last argument to ncl_flush() to
612 * a 1 would force a commit operation, if it is felt a
613 * commit is necessary now.
614 * for NFS Version 4 - flush the dirty buffers and commit them, if
615 * nfscl_mustflush() says this is necessary.
616 * It is necessary if there is no write delegation held,
617 * in order to satisfy open/close coherency.
618 * If the file isn't cached on local stable storage,
619 * it may be necessary in order to detect "out of space"
620 * errors from the server, if the write delegation
621 * issued by the server doesn't allow the file to grow.
625 nfs_close(struct vop_close_args *ap)
627 struct vnode *vp = ap->a_vp;
628 struct nfsnode *np = VTONFS(vp);
629 struct nfsvattr nfsva;
631 int error = 0, ret, localcred = 0;
632 int fmode = ap->a_fflag;
634 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
637 * During shutdown, a_cred isn't valid, so just use root.
639 if (ap->a_cred == NOCRED) {
640 cred = newnfs_getcred();
645 if (vp->v_type == VREG) {
647 * Examine and clean dirty pages, regardless of NMODIFIED.
648 * This closes a major hole in close-to-open consistency.
649 * We want to push out all dirty pages (and buffers) on
650 * close, regardless of whether they were dirtied by
651 * mmap'ed writes or via write().
653 if (nfs_clean_pages_on_close && vp->v_object) {
654 VM_OBJECT_LOCK(vp->v_object);
655 vm_object_page_clean(vp->v_object, 0, 0, 0);
656 VM_OBJECT_UNLOCK(vp->v_object);
658 mtx_lock(&np->n_mtx);
659 if (np->n_flag & NMODIFIED) {
660 mtx_unlock(&np->n_mtx);
663 * Under NFSv3 we have dirty buffers to dispose of. We
664 * must flush them to the NFS server. We have the option
665 * of waiting all the way through the commit rpc or just
666 * waiting for the initial write. The default is to only
667 * wait through the initial write so the data is in the
668 * server's cache, which is roughly similar to the state
669 * a standard disk subsystem leaves the file in on close().
671 * We cannot clear the NMODIFIED bit in np->n_flag due to
672 * potential races with other processes, and certainly
673 * cannot clear it if we don't commit.
674 * These races occur when there is no longer the old
675 * traditional vnode locking implemented for Vnode Ops.
677 int cm = newnfs_commit_on_close ? 1 : 0;
678 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
679 /* np->n_flag &= ~NMODIFIED; */
680 } else if (NFS_ISV4(vp)) {
681 if (nfscl_mustflush(vp) != 0) {
682 int cm = newnfs_commit_on_close ? 1 : 0;
683 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
686 * as above w.r.t races when clearing
688 * np->n_flag &= ~NMODIFIED;
692 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
693 mtx_lock(&np->n_mtx);
696 * Invalidate the attribute cache in all cases.
697 * An open is going to fetch fresh attrs any way, other procs
698 * on this node that have file open will be forced to do an
699 * otw attr fetch, but this is safe.
700 * --> A user found that their RPC count dropped by 20% when
701 * this was commented out and I can't see any requirement
702 * for it, so I've disabled it when negative lookups are
703 * enabled. (What does this have to do with negative lookup
704 * caching? Well nothing, except it was reported by the
705 * same user that needed negative lookup caching and I wanted
706 * there to be a way to disable it to see if it
707 * is the cause of some caching/coherency issue that might
710 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0)
712 if (np->n_flag & NWRITEERR) {
713 np->n_flag &= ~NWRITEERR;
716 mtx_unlock(&np->n_mtx);
721 * Get attributes so "change" is up to date.
723 if (error == 0 && nfscl_mustflush(vp) != 0) {
724 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
727 np->n_change = nfsva.na_filerev;
728 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
736 ret = nfsrpc_close(vp, 0, ap->a_td);
740 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
743 if (newnfs_directio_enable)
744 KASSERT((np->n_directio_asyncwr == 0),
745 ("nfs_close: dirty unflushed (%d) directio buffers\n",
746 np->n_directio_asyncwr));
747 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
748 mtx_lock(&np->n_mtx);
749 KASSERT((np->n_directio_opens > 0),
750 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
751 np->n_directio_opens--;
752 if (np->n_directio_opens == 0)
753 np->n_flag &= ~NNONCACHE;
754 mtx_unlock(&np->n_mtx);
762 * nfs getattr call from vfs.
765 nfs_getattr(struct vop_getattr_args *ap)
767 struct vnode *vp = ap->a_vp;
768 struct thread *td = curthread; /* XXX */
769 struct nfsnode *np = VTONFS(vp);
771 struct nfsvattr nfsva;
772 struct vattr *vap = ap->a_vap;
776 * Update local times for special files.
778 mtx_lock(&np->n_mtx);
779 if (np->n_flag & (NACC | NUPD))
781 mtx_unlock(&np->n_mtx);
783 * First look in the cache.
785 if (ncl_getattrcache(vp, &vattr) == 0) {
786 vap->va_type = vattr.va_type;
787 vap->va_mode = vattr.va_mode;
788 vap->va_nlink = vattr.va_nlink;
789 vap->va_uid = vattr.va_uid;
790 vap->va_gid = vattr.va_gid;
791 vap->va_fsid = vattr.va_fsid;
792 vap->va_fileid = vattr.va_fileid;
793 vap->va_size = vattr.va_size;
794 vap->va_blocksize = vattr.va_blocksize;
795 vap->va_atime = vattr.va_atime;
796 vap->va_mtime = vattr.va_mtime;
797 vap->va_ctime = vattr.va_ctime;
798 vap->va_gen = vattr.va_gen;
799 vap->va_flags = vattr.va_flags;
800 vap->va_rdev = vattr.va_rdev;
801 vap->va_bytes = vattr.va_bytes;
802 vap->va_filerev = vattr.va_filerev;
804 * Get the local modify time for the case of a write
807 nfscl_deleggetmodtime(vp, &vap->va_mtime);
811 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
812 nfsaccess_cache_timeout > 0) {
813 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
814 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
815 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
816 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
820 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
822 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
825 * Get the local modify time for the case of a write
828 nfscl_deleggetmodtime(vp, &vap->va_mtime);
829 } else if (NFS_ISV4(vp)) {
830 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
839 nfs_setattr(struct vop_setattr_args *ap)
841 struct vnode *vp = ap->a_vp;
842 struct nfsnode *np = VTONFS(vp);
843 struct thread *td = curthread; /* XXX */
844 struct vattr *vap = ap->a_vap;
853 * Setting of flags and marking of atimes are not supported.
855 if (vap->va_flags != VNOVAL)
859 * Disallow write attempts if the filesystem is mounted read-only.
861 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
862 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
863 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
864 (vp->v_mount->mnt_flag & MNT_RDONLY))
866 if (vap->va_size != VNOVAL) {
867 switch (vp->v_type) {
874 if (vap->va_mtime.tv_sec == VNOVAL &&
875 vap->va_atime.tv_sec == VNOVAL &&
876 vap->va_mode == (mode_t)VNOVAL &&
877 vap->va_uid == (uid_t)VNOVAL &&
878 vap->va_gid == (gid_t)VNOVAL)
880 vap->va_size = VNOVAL;
884 * Disallow write attempts if the filesystem is
887 if (vp->v_mount->mnt_flag & MNT_RDONLY)
890 * We run vnode_pager_setsize() early (why?),
891 * we must set np->n_size now to avoid vinvalbuf
892 * V_SAVE races that might setsize a lower
895 mtx_lock(&np->n_mtx);
897 mtx_unlock(&np->n_mtx);
898 error = ncl_meta_setsize(vp, ap->a_cred, td,
900 mtx_lock(&np->n_mtx);
901 if (np->n_flag & NMODIFIED) {
903 mtx_unlock(&np->n_mtx);
904 if (vap->va_size == 0)
905 error = ncl_vinvalbuf(vp, 0, td, 1);
907 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
909 vnode_pager_setsize(vp, tsize);
913 * Call nfscl_delegmodtime() to set the modify time
914 * locally, as required.
916 nfscl_delegmodtime(vp);
918 mtx_unlock(&np->n_mtx);
920 * np->n_size has already been set to vap->va_size
921 * in ncl_meta_setsize(). We must set it again since
922 * nfs_loadattrcache() could be called through
923 * ncl_meta_setsize() and could modify np->n_size.
925 mtx_lock(&np->n_mtx);
926 np->n_vattr.na_size = np->n_size = vap->va_size;
927 mtx_unlock(&np->n_mtx);
930 mtx_lock(&np->n_mtx);
931 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
932 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
933 mtx_unlock(&np->n_mtx);
934 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
935 (error == EINTR || error == EIO))
938 mtx_unlock(&np->n_mtx);
940 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
941 if (error && vap->va_size != VNOVAL) {
942 mtx_lock(&np->n_mtx);
943 np->n_size = np->n_vattr.na_size = tsize;
944 vnode_pager_setsize(vp, tsize);
945 mtx_unlock(&np->n_mtx);
951 * Do an nfs setattr rpc.
954 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
957 struct nfsnode *np = VTONFS(vp);
958 int error, ret, attrflag, i;
959 struct nfsvattr nfsva;
962 mtx_lock(&np->n_mtx);
963 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
964 np->n_accesscache[i].stamp = 0;
965 np->n_flag |= NDELEGMOD;
966 mtx_unlock(&np->n_mtx);
968 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
971 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
975 if (error && NFS_ISV4(vp))
976 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
981 * nfs lookup call, one step at a time...
982 * First look in cache
983 * If not found, unlock the directory nfsnode and do the rpc
986 nfs_lookup(struct vop_lookup_args *ap)
988 struct componentname *cnp = ap->a_cnp;
989 struct vnode *dvp = ap->a_dvp;
990 struct vnode **vpp = ap->a_vpp;
991 struct mount *mp = dvp->v_mount;
992 int flags = cnp->cn_flags;
994 struct nfsmount *nmp;
995 struct nfsnode *np, *newnp;
996 int error = 0, attrflag, dattrflag, ltype;
997 struct thread *td = cnp->cn_thread;
999 struct nfsvattr dnfsva, nfsva;
1001 struct timespec dmtime;
1004 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1005 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1007 if (dvp->v_type != VDIR)
1012 /* For NFSv4, wait until any remove is done. */
1013 mtx_lock(&np->n_mtx);
1014 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1015 np->n_flag |= NREMOVEWANT;
1016 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1018 mtx_unlock(&np->n_mtx);
1020 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1022 error = cache_lookup(dvp, vpp, cnp);
1023 if (error > 0 && error != ENOENT)
1027 * We only accept a positive hit in the cache if the
1028 * change time of the file matches our cached copy.
1029 * Otherwise, we discard the cache entry and fallback
1030 * to doing a lookup RPC.
1032 * To better handle stale file handles and attributes,
1033 * clear the attribute cache of this node if it is a
1034 * leaf component, part of an open() call, and not
1035 * locally modified before fetching the attributes.
1036 * This should allow stale file handles to be detected
1037 * here where we can fall back to a LOOKUP RPC to
1038 * recover rather than having nfs_open() detect the
1039 * stale file handle and failing open(2) with ESTALE.
1042 newnp = VTONFS(newvp);
1043 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1044 !(newnp->n_flag & NMODIFIED)) {
1045 mtx_lock(&newnp->n_mtx);
1046 newnp->n_attrstamp = 0;
1047 mtx_unlock(&newnp->n_mtx);
1049 if (nfscl_nodeleg(newvp, 0) == 0 ||
1050 (VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1051 timespeccmp(&vattr.va_ctime, &newnp->n_ctime, ==))) {
1052 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1053 if (cnp->cn_nameiop != LOOKUP &&
1055 cnp->cn_flags |= SAVENAME;
1064 } else if (error == ENOENT) {
1065 if (dvp->v_iflag & VI_DOOMED)
1068 * We only accept a negative hit in the cache if the
1069 * modification time of the parent directory matches
1070 * our cached copy. Otherwise, we discard all of the
1071 * negative cache entries for this directory. We also
1072 * only trust -ve cache entries for less than
1073 * nm_negative_namecache_timeout seconds.
1075 if ((u_int)(ticks - np->n_dmtime_ticks) <
1076 (nmp->nm_negnametimeo * hz) &&
1077 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1078 timespeccmp(&vattr.va_mtime, &np->n_dmtime, ==)) {
1079 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1082 cache_purge_negative(dvp);
1083 mtx_lock(&np->n_mtx);
1084 timespecclear(&np->n_dmtime);
1085 mtx_unlock(&np->n_mtx);
1089 * Cache the modification time of the parent directory in case
1090 * the lookup fails and results in adding the first negative
1091 * name cache entry for the directory. Since this is reading
1092 * a single time_t, don't bother with locking. The
1093 * modification time may be a bit stale, but it must be read
1094 * before performing the lookup RPC to prevent a race where
1095 * another lookup updates the timestamp on the directory after
1096 * the lookup RPC has been performed on the server but before
1097 * n_dmtime is set at the end of this function.
1099 dmtime = np->n_vattr.na_mtime;
1102 NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1103 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1104 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1107 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1109 if (newvp != NULLVP) {
1114 if (error != ENOENT) {
1116 error = nfscl_maperr(td, error, (uid_t)0,
1121 /* The requested file was not found. */
1122 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1123 (flags & ISLASTCN)) {
1125 * XXX: UFS does a full VOP_ACCESS(dvp,
1126 * VWRITE) here instead of just checking
1129 if (mp->mnt_flag & MNT_RDONLY)
1131 cnp->cn_flags |= SAVENAME;
1132 return (EJUSTRETURN);
1135 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1137 * Maintain n_dmtime as the modification time
1138 * of the parent directory when the oldest -ve
1139 * name cache entry for this directory was
1140 * added. If a -ve cache entry has already
1141 * been added with a newer modification time
1142 * by a concurrent lookup, then don't bother
1143 * adding a cache entry. The modification
1144 * time of the directory might have changed
1145 * due to the file this lookup failed to find
1146 * being created. In that case a subsequent
1147 * lookup would incorrectly use the entry
1148 * added here instead of doing an extra
1151 mtx_lock(&np->n_mtx);
1152 if (timespeccmp(&np->n_dmtime, &dmtime, <=)) {
1153 if (!timespecisset(&np->n_dmtime)) {
1154 np->n_dmtime = dmtime;
1155 np->n_dmtime_ticks = ticks;
1157 mtx_unlock(&np->n_mtx);
1158 cache_enter(dvp, NULL, cnp);
1160 mtx_unlock(&np->n_mtx);
1166 * Handle RENAME case...
1168 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1169 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1170 FREE((caddr_t)nfhp, M_NFSFH);
1173 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1178 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1181 cnp->cn_flags |= SAVENAME;
1185 if (flags & ISDOTDOT) {
1186 ltype = VOP_ISLOCKED(dvp);
1187 error = vfs_busy(mp, MBF_NOWAIT);
1191 error = vfs_busy(mp, 0);
1192 vn_lock(dvp, ltype | LK_RETRY);
1194 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1202 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1207 vn_lock(dvp, ltype | LK_RETRY);
1208 if (dvp->v_iflag & VI_DOOMED) {
1220 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1222 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1223 FREE((caddr_t)nfhp, M_NFSFH);
1227 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1230 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1235 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1237 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1238 !(np->n_flag & NMODIFIED)) {
1240 * Flush the attribute cache when opening a
1241 * leaf node to ensure that fresh attributes
1242 * are fetched in nfs_open() since we did not
1243 * fetch attributes from the LOOKUP reply.
1245 mtx_lock(&np->n_mtx);
1246 np->n_attrstamp = 0;
1247 mtx_unlock(&np->n_mtx);
1250 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1251 cnp->cn_flags |= SAVENAME;
1252 if ((cnp->cn_flags & MAKEENTRY) &&
1253 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1254 np->n_ctime = np->n_vattr.na_vattr.va_ctime;
1255 cache_enter(dvp, newvp, cnp);
1263 * Just call ncl_bioread() to do the work.
1266 nfs_read(struct vop_read_args *ap)
1268 struct vnode *vp = ap->a_vp;
1270 switch (vp->v_type) {
1272 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1276 return (EOPNOTSUPP);
1284 nfs_readlink(struct vop_readlink_args *ap)
1286 struct vnode *vp = ap->a_vp;
1288 if (vp->v_type != VLNK)
1290 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1294 * Do a readlink rpc.
1295 * Called by ncl_doio() from below the buffer cache.
1298 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1300 int error, ret, attrflag;
1301 struct nfsvattr nfsva;
1303 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1306 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1310 if (error && NFS_ISV4(vp))
1311 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1320 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1322 int error, ret, attrflag;
1323 struct nfsvattr nfsva;
1325 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1328 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1332 if (error && NFS_ISV4(vp))
1333 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1341 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1342 int *iomode, int *must_commit, int called_from_strategy)
1344 struct nfsvattr nfsva;
1345 int error = 0, attrflag, ret;
1346 u_char verf[NFSX_VERF];
1347 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1350 error = nfsrpc_write(vp, uiop, iomode, verf, cred,
1351 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy);
1353 if (!error && NFSHASWRITEVERF(nmp) &&
1354 NFSBCMP(verf, nmp->nm_verf, NFSX_VERF)) {
1356 NFSBCOPY(verf, nmp->nm_verf, NFSX_VERF);
1360 if (VTONFS(vp)->n_flag & ND_NFSV4)
1361 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1364 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1369 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1370 *iomode = NFSWRITE_FILESYNC;
1371 if (error && NFS_ISV4(vp))
1372 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1378 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1379 * mode set to specify the file type and the size field for rdev.
1382 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1385 struct nfsvattr nfsva, dnfsva;
1386 struct vnode *newvp = NULL;
1387 struct nfsnode *np = NULL, *dnp;
1390 int error = 0, attrflag, dattrflag;
1393 if (vap->va_type == VCHR || vap->va_type == VBLK)
1394 rdev = vap->va_rdev;
1395 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1398 return (EOPNOTSUPP);
1399 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1401 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1402 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1403 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1406 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1407 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1408 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1411 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1412 cnp->cn_thread, &np, NULL);
1415 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1419 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1423 if ((cnp->cn_flags & MAKEENTRY))
1424 cache_enter(dvp, newvp, cnp);
1426 } else if (NFS_ISV4(dvp)) {
1427 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1431 mtx_lock(&dnp->n_mtx);
1432 dnp->n_flag |= NMODIFIED;
1434 dnp->n_attrstamp = 0;
1435 mtx_unlock(&dnp->n_mtx);
1441 * just call nfs_mknodrpc() to do the work.
1445 nfs_mknod(struct vop_mknod_args *ap)
1447 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1450 static u_long create_verf;
1452 * nfs file create call
1455 nfs_create(struct vop_create_args *ap)
1457 struct vnode *dvp = ap->a_dvp;
1458 struct vattr *vap = ap->a_vap;
1459 struct componentname *cnp = ap->a_cnp;
1460 struct nfsnode *np = NULL, *dnp;
1461 struct vnode *newvp = NULL;
1462 struct nfsmount *nmp;
1463 struct nfsvattr dnfsva, nfsva;
1466 int error = 0, attrflag, dattrflag, fmode = 0;
1470 * Oops, not for me..
1472 if (vap->va_type == VSOCK)
1473 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1475 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1477 if (vap->va_vaflags & VA_EXCLUSIVE)
1480 nmp = VFSTONFS(vnode_mount(dvp));
1482 /* For NFSv4, wait until any remove is done. */
1483 mtx_lock(&dnp->n_mtx);
1484 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1485 dnp->n_flag |= NREMOVEWANT;
1486 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1488 mtx_unlock(&dnp->n_mtx);
1491 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1493 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1494 cverf.lval[0] = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1497 cverf.lval[0] = create_verf;
1499 IN_IFADDR_RUNLOCK();
1502 cverf.lval[1] = ++create_verf;
1503 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1504 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1505 &nfhp, &attrflag, &dattrflag, NULL);
1508 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1509 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1510 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1513 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1514 cnp->cn_thread, &np, NULL);
1517 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1521 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1525 if (newvp != NULL) {
1529 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1530 error == NFSERR_NOTSUPP) {
1534 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1535 if (nfscl_checksattr(vap, &nfsva)) {
1536 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1537 cnp->cn_thread, &nfsva, &attrflag, NULL);
1538 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1539 vap->va_gid != (gid_t)VNOVAL)) {
1540 /* try again without setting uid/gid */
1541 vap->va_uid = (uid_t)VNOVAL;
1542 vap->va_gid = (uid_t)VNOVAL;
1543 error = nfsrpc_setattr(newvp, vap, NULL,
1544 cnp->cn_cred, cnp->cn_thread, &nfsva,
1548 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1553 if (cnp->cn_flags & MAKEENTRY)
1554 cache_enter(dvp, newvp, cnp);
1556 } else if (NFS_ISV4(dvp)) {
1557 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1560 mtx_lock(&dnp->n_mtx);
1561 dnp->n_flag |= NMODIFIED;
1563 dnp->n_attrstamp = 0;
1564 mtx_unlock(&dnp->n_mtx);
1569 * nfs file remove call
1570 * To try and make nfs semantics closer to ufs semantics, a file that has
1571 * other processes using the vnode is renamed instead of removed and then
1572 * removed later on the last close.
1573 * - If v_usecount > 1
1574 * If a rename is not already in the works
1575 * call nfs_sillyrename() to set it up
1580 nfs_remove(struct vop_remove_args *ap)
1582 struct vnode *vp = ap->a_vp;
1583 struct vnode *dvp = ap->a_dvp;
1584 struct componentname *cnp = ap->a_cnp;
1585 struct nfsnode *np = VTONFS(vp);
1589 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1590 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1591 if (vp->v_type == VDIR)
1593 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1594 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1595 vattr.va_nlink > 1)) {
1597 * Purge the name cache so that the chance of a lookup for
1598 * the name succeeding while the remove is in progress is
1599 * minimized. Without node locking it can still happen, such
1600 * that an I/O op returns ESTALE, but since you get this if
1601 * another host removes the file..
1605 * throw away biocache buffers, mainly to avoid
1606 * unnecessary delayed writes later.
1608 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1610 if (error != EINTR && error != EIO)
1611 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1612 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1614 * Kludge City: If the first reply to the remove rpc is lost..
1615 * the reply to the retransmitted request will be ENOENT
1616 * since the file was in fact removed
1617 * Therefore, we cheat and return success.
1619 if (error == ENOENT)
1621 } else if (!np->n_sillyrename)
1622 error = nfs_sillyrename(dvp, vp, cnp);
1623 np->n_attrstamp = 0;
1628 * nfs file remove rpc called from nfs_inactive
1631 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1634 * Make sure that the directory vnode is still valid.
1635 * XXX we should lock sp->s_dvp here.
1637 if (sp->s_dvp->v_type == VBAD)
1639 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1644 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1647 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1648 int namelen, struct ucred *cred, struct thread *td)
1650 struct nfsvattr dnfsva;
1651 struct nfsnode *dnp = VTONFS(dvp);
1652 int error = 0, dattrflag;
1654 mtx_lock(&dnp->n_mtx);
1655 dnp->n_flag |= NREMOVEINPROG;
1656 mtx_unlock(&dnp->n_mtx);
1657 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1659 mtx_lock(&dnp->n_mtx);
1660 if ((dnp->n_flag & NREMOVEWANT)) {
1661 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1662 mtx_unlock(&dnp->n_mtx);
1663 wakeup((caddr_t)dnp);
1665 dnp->n_flag &= ~NREMOVEINPROG;
1666 mtx_unlock(&dnp->n_mtx);
1669 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1670 mtx_lock(&dnp->n_mtx);
1671 dnp->n_flag |= NMODIFIED;
1673 dnp->n_attrstamp = 0;
1674 mtx_unlock(&dnp->n_mtx);
1675 if (error && NFS_ISV4(dvp))
1676 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1681 * nfs file rename call
1684 nfs_rename(struct vop_rename_args *ap)
1686 struct vnode *fvp = ap->a_fvp;
1687 struct vnode *tvp = ap->a_tvp;
1688 struct vnode *fdvp = ap->a_fdvp;
1689 struct vnode *tdvp = ap->a_tdvp;
1690 struct componentname *tcnp = ap->a_tcnp;
1691 struct componentname *fcnp = ap->a_fcnp;
1692 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1693 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1694 struct nfsv4node *newv4 = NULL;
1697 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1698 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1699 /* Check for cross-device rename */
1700 if ((fvp->v_mount != tdvp->v_mount) ||
1701 (tvp && (fvp->v_mount != tvp->v_mount))) {
1707 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1711 if ((error = vn_lock(fvp, LK_EXCLUSIVE)))
1715 * We have to flush B_DELWRI data prior to renaming
1716 * the file. If we don't, the delayed-write buffers
1717 * can be flushed out later after the file has gone stale
1718 * under NFSV3. NFSV2 does not have this problem because
1719 * ( as far as I can tell ) it flushes dirty buffers more
1722 * Skip the rename operation if the fsync fails, this can happen
1723 * due to the server's volume being full, when we pushed out data
1724 * that was written back to our cache earlier. Not checking for
1725 * this condition can result in potential (silent) data loss.
1727 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1730 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1735 * If the tvp exists and is in use, sillyrename it before doing the
1736 * rename of the new file over it.
1737 * XXX Can't sillyrename a directory.
1739 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1740 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1745 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1746 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1751 * For NFSv4, check to see if it is the same name and
1752 * replace the name, if it is different.
1754 MALLOC(newv4, struct nfsv4node *,
1755 sizeof (struct nfsv4node) +
1756 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1757 M_NFSV4NODE, M_WAITOK);
1758 mtx_lock(&tdnp->n_mtx);
1759 mtx_lock(&fnp->n_mtx);
1760 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1761 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1762 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1763 tcnp->cn_namelen) ||
1764 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1765 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1766 tdnp->n_fhp->nfh_len))) {
1768 { char nnn[100]; int nnnl;
1769 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1770 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1772 printf("ren replace=%s\n",nnn);
1775 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1778 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1779 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1780 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1781 tdnp->n_fhp->nfh_len);
1782 NFSBCOPY(tcnp->cn_nameptr,
1783 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1785 mtx_unlock(&tdnp->n_mtx);
1786 mtx_unlock(&fnp->n_mtx);
1788 FREE((caddr_t)newv4, M_NFSV4NODE);
1791 if (fvp->v_type == VDIR) {
1792 if (tvp != NULL && tvp->v_type == VDIR)
1807 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1809 if (error == ENOENT)
1815 * nfs file rename rpc called from nfs_remove() above
1818 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1819 struct sillyrename *sp)
1822 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1823 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1828 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1831 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1832 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1833 int tnamelen, struct ucred *cred, struct thread *td)
1835 struct nfsvattr fnfsva, tnfsva;
1836 struct nfsnode *fdnp = VTONFS(fdvp);
1837 struct nfsnode *tdnp = VTONFS(tdvp);
1838 int error = 0, fattrflag, tattrflag;
1840 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1841 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1842 &tattrflag, NULL, NULL);
1843 mtx_lock(&fdnp->n_mtx);
1844 fdnp->n_flag |= NMODIFIED;
1845 mtx_unlock(&fdnp->n_mtx);
1846 mtx_lock(&tdnp->n_mtx);
1847 tdnp->n_flag |= NMODIFIED;
1848 mtx_unlock(&tdnp->n_mtx);
1850 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1852 fdnp->n_attrstamp = 0;
1854 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1856 tdnp->n_attrstamp = 0;
1857 if (error && NFS_ISV4(fdvp))
1858 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1863 * nfs hard link create call
1866 nfs_link(struct vop_link_args *ap)
1868 struct vnode *vp = ap->a_vp;
1869 struct vnode *tdvp = ap->a_tdvp;
1870 struct componentname *cnp = ap->a_cnp;
1871 struct nfsnode *tdnp;
1872 struct nfsvattr nfsva, dnfsva;
1873 int error = 0, attrflag, dattrflag;
1875 if (vp->v_mount != tdvp->v_mount) {
1880 * Push all writes to the server, so that the attribute cache
1881 * doesn't get "out of sync" with the server.
1882 * XXX There should be a better way!
1884 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1886 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1887 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1889 tdnp = VTONFS(tdvp);
1890 mtx_lock(&tdnp->n_mtx);
1891 tdnp->n_flag |= NMODIFIED;
1892 mtx_unlock(&tdnp->n_mtx);
1894 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1896 VTONFS(vp)->n_attrstamp = 0;
1898 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1900 tdnp->n_attrstamp = 0;
1902 * If negative lookup caching is enabled, I might as well
1903 * add an entry for this node. Not necessary for correctness,
1904 * but if negative caching is enabled, then the system
1905 * must care about lookup caching hit rate, so...
1907 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1908 (cnp->cn_flags & MAKEENTRY))
1909 cache_enter(tdvp, vp, cnp);
1910 if (error && NFS_ISV4(vp))
1911 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1917 * nfs symbolic link create call
1920 nfs_symlink(struct vop_symlink_args *ap)
1922 struct vnode *dvp = ap->a_dvp;
1923 struct vattr *vap = ap->a_vap;
1924 struct componentname *cnp = ap->a_cnp;
1925 struct nfsvattr nfsva, dnfsva;
1927 struct nfsnode *np = NULL, *dnp;
1928 struct vnode *newvp = NULL;
1929 int error = 0, attrflag, dattrflag, ret;
1931 vap->va_type = VLNK;
1932 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1933 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1934 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1936 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
1943 if (newvp != NULL) {
1945 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1947 } else if (!error) {
1949 * If we do not have an error and we could not extract the
1950 * newvp from the response due to the request being NFSv2, we
1951 * have to do a lookup in order to obtain a newvp to return.
1953 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1954 cnp->cn_cred, cnp->cn_thread, &np);
1962 error = nfscl_maperr(cnp->cn_thread, error,
1963 vap->va_uid, vap->va_gid);
1966 * If negative lookup caching is enabled, I might as well
1967 * add an entry for this node. Not necessary for correctness,
1968 * but if negative caching is enabled, then the system
1969 * must care about lookup caching hit rate, so...
1971 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
1972 (cnp->cn_flags & MAKEENTRY))
1973 cache_enter(dvp, newvp, cnp);
1978 mtx_lock(&dnp->n_mtx);
1979 dnp->n_flag |= NMODIFIED;
1980 mtx_unlock(&dnp->n_mtx);
1982 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1984 dnp->n_attrstamp = 0;
1992 nfs_mkdir(struct vop_mkdir_args *ap)
1994 struct vnode *dvp = ap->a_dvp;
1995 struct vattr *vap = ap->a_vap;
1996 struct componentname *cnp = ap->a_cnp;
1997 struct nfsnode *np = NULL, *dnp;
1998 struct vnode *newvp = NULL;
2001 struct nfsvattr nfsva, dnfsva;
2002 int error = 0, attrflag, dattrflag, ret;
2004 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
2006 vap->va_type = VDIR;
2007 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2008 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2009 &attrflag, &dattrflag, NULL);
2011 mtx_lock(&dnp->n_mtx);
2012 dnp->n_flag |= NMODIFIED;
2013 mtx_unlock(&dnp->n_mtx);
2015 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2017 dnp->n_attrstamp = 0;
2019 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2024 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2029 if (!error && newvp == NULL) {
2030 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2031 cnp->cn_cred, cnp->cn_thread, &np);
2034 if (newvp->v_type != VDIR)
2042 error = nfscl_maperr(cnp->cn_thread, error,
2043 vap->va_uid, vap->va_gid);
2046 * If negative lookup caching is enabled, I might as well
2047 * add an entry for this node. Not necessary for correctness,
2048 * but if negative caching is enabled, then the system
2049 * must care about lookup caching hit rate, so...
2051 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2052 (cnp->cn_flags & MAKEENTRY))
2053 cache_enter(dvp, newvp, cnp);
2060 * nfs remove directory call
2063 nfs_rmdir(struct vop_rmdir_args *ap)
2065 struct vnode *vp = ap->a_vp;
2066 struct vnode *dvp = ap->a_dvp;
2067 struct componentname *cnp = ap->a_cnp;
2068 struct nfsnode *dnp;
2069 struct nfsvattr dnfsva;
2070 int error, dattrflag;
2074 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2075 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2077 mtx_lock(&dnp->n_mtx);
2078 dnp->n_flag |= NMODIFIED;
2079 mtx_unlock(&dnp->n_mtx);
2081 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2083 dnp->n_attrstamp = 0;
2087 if (error && NFS_ISV4(dvp))
2088 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2091 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2093 if (error == ENOENT)
2102 nfs_readdir(struct vop_readdir_args *ap)
2104 struct vnode *vp = ap->a_vp;
2105 struct nfsnode *np = VTONFS(vp);
2106 struct uio *uio = ap->a_uio;
2107 int tresid, error = 0;
2110 if (vp->v_type != VDIR)
2114 * First, check for hit on the EOF offset cache
2116 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2117 (np->n_flag & NMODIFIED) == 0) {
2118 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2119 mtx_lock(&np->n_mtx);
2120 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2121 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2122 mtx_unlock(&np->n_mtx);
2123 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2126 mtx_unlock(&np->n_mtx);
2131 * Call ncl_bioread() to do the real work.
2133 tresid = uio->uio_resid;
2134 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2136 if (!error && uio->uio_resid == tresid)
2137 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2143 * Called from below the buffer cache by ncl_doio().
2146 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2149 struct nfsvattr nfsva;
2150 nfsuint64 *cookiep, cookie;
2151 struct nfsnode *dnp = VTONFS(vp);
2152 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2153 int error = 0, eof, attrflag;
2155 KASSERT(uiop->uio_iovcnt == 1 &&
2156 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2157 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2158 ("nfs readdirrpc bad uio"));
2161 * If there is no cookie, assume directory was stale.
2163 ncl_dircookie_lock(dnp);
2164 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2167 ncl_dircookie_unlock(dnp);
2169 ncl_dircookie_unlock(dnp);
2170 return (NFSERR_BAD_COOKIE);
2173 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2174 (void)ncl_fsinfo(nmp, vp, cred, td);
2176 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2177 &attrflag, &eof, NULL);
2179 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2183 * We are now either at the end of the directory or have filled
2187 dnp->n_direofoffset = uiop->uio_offset;
2189 if (uiop->uio_resid > 0)
2190 ncl_printf("EEK! readdirrpc resid > 0\n");
2191 ncl_dircookie_lock(dnp);
2192 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2194 ncl_dircookie_unlock(dnp);
2196 } else if (NFS_ISV4(vp)) {
2197 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2203 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2206 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2209 struct nfsvattr nfsva;
2210 nfsuint64 *cookiep, cookie;
2211 struct nfsnode *dnp = VTONFS(vp);
2212 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2213 int error = 0, attrflag, eof;
2215 KASSERT(uiop->uio_iovcnt == 1 &&
2216 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2217 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2218 ("nfs readdirplusrpc bad uio"));
2221 * If there is no cookie, assume directory was stale.
2223 ncl_dircookie_lock(dnp);
2224 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2227 ncl_dircookie_unlock(dnp);
2229 ncl_dircookie_unlock(dnp);
2230 return (NFSERR_BAD_COOKIE);
2233 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2234 (void)ncl_fsinfo(nmp, vp, cred, td);
2235 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2236 &attrflag, &eof, NULL);
2238 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2242 * We are now either at end of the directory or have filled the
2246 dnp->n_direofoffset = uiop->uio_offset;
2248 if (uiop->uio_resid > 0)
2249 ncl_printf("EEK! readdirplusrpc resid > 0\n");
2250 ncl_dircookie_lock(dnp);
2251 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2253 ncl_dircookie_unlock(dnp);
2255 } else if (NFS_ISV4(vp)) {
2256 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2262 * Silly rename. To make the NFS filesystem that is stateless look a little
2263 * more like the "ufs" a remove of an active vnode is translated to a rename
2264 * to a funny looking filename that is removed by nfs_inactive on the
2265 * nfsnode. There is the potential for another process on a different client
2266 * to create the same funny name between the nfs_lookitup() fails and the
2267 * nfs_rename() completes, but...
2270 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2272 struct sillyrename *sp;
2276 unsigned int lticks;
2280 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2281 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2282 M_NEWNFSREQ, M_WAITOK);
2283 sp->s_cred = crhold(cnp->cn_cred);
2288 * Fudge together a funny name.
2289 * Changing the format of the funny name to accomodate more
2290 * sillynames per directory.
2291 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2292 * CPU ticks since boot.
2294 pid = cnp->cn_thread->td_proc->p_pid;
2295 lticks = (unsigned int)ticks;
2297 sp->s_namlen = sprintf(sp->s_name,
2298 ".nfs.%08x.%04x4.4", lticks,
2300 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2301 cnp->cn_thread, NULL))
2305 error = nfs_renameit(dvp, vp, cnp, sp);
2308 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2309 cnp->cn_thread, &np);
2310 np->n_sillyrename = sp;
2315 free((caddr_t)sp, M_NEWNFSREQ);
2320 * Look up a file name and optionally either update the file handle or
2321 * allocate an nfsnode, depending on the value of npp.
2322 * npp == NULL --> just do the lookup
2323 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2325 * *npp != NULL --> update the file handle in the vnode
2328 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2329 struct thread *td, struct nfsnode **npp)
2331 struct vnode *newvp = NULL, *vp;
2332 struct nfsnode *np, *dnp = VTONFS(dvp);
2333 struct nfsfh *nfhp, *onfhp;
2334 struct nfsvattr nfsva, dnfsva;
2335 struct componentname cn;
2336 int error = 0, attrflag, dattrflag;
2339 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2340 &nfhp, &attrflag, &dattrflag, NULL);
2342 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2343 if (npp && !error) {
2348 * For NFSv4, check to see if it is the same name and
2349 * replace the name, if it is different.
2351 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2352 (np->n_v4->n4_namelen != len ||
2353 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2354 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2355 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2356 dnp->n_fhp->nfh_len))) {
2358 { char nnn[100]; int nnnl;
2359 nnnl = (len < 100) ? len : 99;
2360 bcopy(name, nnn, nnnl);
2362 printf("replace=%s\n",nnn);
2365 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2366 MALLOC(np->n_v4, struct nfsv4node *,
2367 sizeof (struct nfsv4node) +
2368 dnp->n_fhp->nfh_len + len - 1,
2369 M_NFSV4NODE, M_WAITOK);
2370 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2371 np->n_v4->n4_namelen = len;
2372 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2373 dnp->n_fhp->nfh_len);
2374 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2376 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2380 * Rehash node for new file handle.
2382 vfs_hash_rehash(vp, hash);
2385 FREE((caddr_t)onfhp, M_NFSFH);
2387 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2388 FREE((caddr_t)nfhp, M_NFSFH);
2392 cn.cn_nameptr = name;
2393 cn.cn_namelen = len;
2394 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2400 if (!attrflag && *npp == NULL) {
2405 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2408 if (npp && *npp == NULL) {
2419 if (error && NFS_ISV4(dvp))
2420 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2425 * Nfs Version 3 and 4 commit rpc
2428 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2431 struct nfsvattr nfsva;
2432 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2433 int error, attrflag;
2434 u_char verf[NFSX_VERF];
2436 mtx_lock(&nmp->nm_mtx);
2437 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2438 mtx_unlock(&nmp->nm_mtx);
2441 mtx_unlock(&nmp->nm_mtx);
2442 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2445 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2446 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2447 error = NFSERR_STALEWRITEVERF;
2449 if (!error && attrflag)
2450 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2452 } else if (NFS_ISV4(vp)) {
2453 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2460 * For async requests when nfsiod(s) are running, queue the request by
2461 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2465 nfs_strategy(struct vop_strategy_args *ap)
2467 struct buf *bp = ap->a_bp;
2470 KASSERT(!(bp->b_flags & B_DONE),
2471 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2472 BUF_ASSERT_HELD(bp);
2474 if (bp->b_iocmd == BIO_READ)
2480 * If the op is asynchronous and an i/o daemon is waiting
2481 * queue the request, wake it up and wait for completion
2482 * otherwise just do it ourselves.
2484 if ((bp->b_flags & B_ASYNC) == 0 ||
2485 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2486 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2491 * fsync vnode op. Just call ncl_flush() with commit == 1.
2495 nfs_fsync(struct vop_fsync_args *ap)
2497 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2501 * Flush all the blocks associated with a vnode.
2502 * Walk through the buffer pool and push any dirty pages
2503 * associated with the vnode.
2504 * If the called_from_renewthread argument is TRUE, it has been called
2505 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2506 * waiting for a buffer write to complete.
2509 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2510 int commit, int called_from_renewthread)
2512 struct nfsnode *np = VTONFS(vp);
2516 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2517 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2518 int passone = 1, trycnt = 0;
2519 u_quad_t off, endoff, toff;
2520 struct ucred* wcred = NULL;
2521 struct buf **bvec = NULL;
2523 #ifndef NFS_COMMITBVECSIZ
2524 #define NFS_COMMITBVECSIZ 20
2526 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2527 int bvecsize = 0, bveccount;
2529 if (called_from_renewthread != 0)
2531 if (nmp->nm_flag & NFSMNT_INT)
2532 slpflag = NFS_PCATCH;
2537 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2538 * server, but has not been committed to stable storage on the server
2539 * yet. On the first pass, the byte range is worked out and the commit
2540 * rpc is done. On the second pass, ncl_writebp() is called to do the
2547 if (NFS_ISV34(vp) && commit) {
2548 if (bvec != NULL && bvec != bvec_on_stack)
2551 * Count up how many buffers waiting for a commit.
2555 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2556 if (!BUF_ISLOCKED(bp) &&
2557 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2558 == (B_DELWRI | B_NEEDCOMMIT))
2562 * Allocate space to remember the list of bufs to commit. It is
2563 * important to use M_NOWAIT here to avoid a race with nfs_write.
2564 * If we can't get memory (for whatever reason), we will end up
2565 * committing the buffers one-by-one in the loop below.
2567 if (bveccount > NFS_COMMITBVECSIZ) {
2569 * Release the vnode interlock to avoid a lock
2573 bvec = (struct buf **)
2574 malloc(bveccount * sizeof(struct buf *),
2578 bvec = bvec_on_stack;
2579 bvecsize = NFS_COMMITBVECSIZ;
2581 bvecsize = bveccount;
2583 bvec = bvec_on_stack;
2584 bvecsize = NFS_COMMITBVECSIZ;
2586 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2587 if (bvecpos >= bvecsize)
2589 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2590 nbp = TAILQ_NEXT(bp, b_bobufs);
2593 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2594 (B_DELWRI | B_NEEDCOMMIT)) {
2596 nbp = TAILQ_NEXT(bp, b_bobufs);
2602 * Work out if all buffers are using the same cred
2603 * so we can deal with them all with one commit.
2605 * NOTE: we are not clearing B_DONE here, so we have
2606 * to do it later on in this routine if we intend to
2607 * initiate I/O on the bp.
2609 * Note: to avoid loopback deadlocks, we do not
2610 * assign b_runningbufspace.
2613 wcred = bp->b_wcred;
2614 else if (wcred != bp->b_wcred)
2616 vfs_busy_pages(bp, 1);
2620 * bp is protected by being locked, but nbp is not
2621 * and vfs_busy_pages() may sleep. We have to
2624 nbp = TAILQ_NEXT(bp, b_bobufs);
2627 * A list of these buffers is kept so that the
2628 * second loop knows which buffers have actually
2629 * been committed. This is necessary, since there
2630 * may be a race between the commit rpc and new
2631 * uncommitted writes on the file.
2633 bvec[bvecpos++] = bp;
2634 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2638 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2646 * Commit data on the server, as required.
2647 * If all bufs are using the same wcred, then use that with
2648 * one call for all of them, otherwise commit each one
2651 if (wcred != NOCRED)
2652 retv = ncl_commit(vp, off, (int)(endoff - off),
2656 for (i = 0; i < bvecpos; i++) {
2659 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2661 size = (u_quad_t)(bp->b_dirtyend
2663 retv = ncl_commit(vp, off, (int)size,
2669 if (retv == NFSERR_STALEWRITEVERF)
2670 ncl_clearcommit(vp->v_mount);
2673 * Now, either mark the blocks I/O done or mark the
2674 * blocks dirty, depending on whether the commit
2677 for (i = 0; i < bvecpos; i++) {
2679 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2682 * Error, leave B_DELWRI intact
2684 vfs_unbusy_pages(bp);
2688 * Success, remove B_DELWRI ( bundirty() ).
2690 * b_dirtyoff/b_dirtyend seem to be NFS
2691 * specific. We should probably move that
2692 * into bundirty(). XXX
2695 bp->b_flags |= B_ASYNC;
2697 bp->b_flags &= ~B_DONE;
2698 bp->b_ioflags &= ~BIO_ERROR;
2699 bp->b_dirtyoff = bp->b_dirtyend = 0;
2706 * Start/do any write(s) that are required.
2710 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2711 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2712 if (waitfor != MNT_WAIT || passone)
2715 error = BUF_TIMELOCK(bp,
2716 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2717 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2722 if (error == ENOLCK) {
2726 if (called_from_renewthread != 0) {
2728 * Return EIO so the flush will be retried
2734 if (newnfs_sigintr(nmp, td)) {
2738 if (slpflag & PCATCH) {
2744 if ((bp->b_flags & B_DELWRI) == 0)
2745 panic("nfs_fsync: not dirty");
2746 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2752 if (passone || !commit)
2753 bp->b_flags |= B_ASYNC;
2755 bp->b_flags |= B_ASYNC;
2757 if (newnfs_sigintr(nmp, td)) {
2768 if (waitfor == MNT_WAIT) {
2769 while (bo->bo_numoutput) {
2770 error = bufobj_wwait(bo, slpflag, slptimeo);
2773 if (called_from_renewthread != 0) {
2775 * Return EIO so that the flush will be
2781 error = newnfs_sigintr(nmp, td);
2784 if (slpflag & PCATCH) {
2791 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2796 * Wait for all the async IO requests to drain
2799 mtx_lock(&np->n_mtx);
2800 while (np->n_directio_asyncwr > 0) {
2801 np->n_flag |= NFSYNCWAIT;
2802 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2803 &np->n_mtx, slpflag | (PRIBIO + 1),
2806 if (newnfs_sigintr(nmp, td)) {
2807 mtx_unlock(&np->n_mtx);
2813 mtx_unlock(&np->n_mtx);
2816 mtx_lock(&np->n_mtx);
2817 if (np->n_flag & NWRITEERR) {
2818 error = np->n_error;
2819 np->n_flag &= ~NWRITEERR;
2821 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2822 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2823 np->n_flag &= ~NMODIFIED;
2824 mtx_unlock(&np->n_mtx);
2826 if (bvec != NULL && bvec != bvec_on_stack)
2828 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2829 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2830 np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2831 /* try, try again... */
2836 printf("try%d\n", trycnt);
2843 * NFS advisory byte-level locks.
2846 nfs_advlock(struct vop_advlock_args *ap)
2848 struct vnode *vp = ap->a_vp;
2850 struct nfsnode *np = VTONFS(ap->a_vp);
2851 struct proc *p = (struct proc *)ap->a_id;
2852 struct thread *td = curthread; /* XXX */
2854 int ret, error = EOPNOTSUPP;
2857 if (NFS_ISV4(vp) && (ap->a_flags & F_POSIX)) {
2859 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2860 if (vp->v_iflag & VI_DOOMED) {
2866 * If this is unlocking a write locked region, flush and
2867 * commit them before unlocking. This is required by
2868 * RFC3530 Sec. 9.3.2.
2870 if (ap->a_op == F_UNLCK &&
2871 nfscl_checkwritelocked(vp, ap->a_fl, cred, td))
2872 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
2875 * Loop around doing the lock op, while a blocking lock
2876 * must wait for the lock op to succeed.
2879 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
2880 ap->a_fl, 0, cred, td);
2881 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2882 ap->a_op == F_SETLK) {
2884 error = nfs_catnap(PZERO | PCATCH, ret,
2888 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2889 if (vp->v_iflag & VI_DOOMED) {
2894 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2895 ap->a_op == F_SETLK);
2896 if (ret == NFSERR_DENIED) {
2899 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
2902 } else if (ret != 0) {
2908 * Now, if we just got a lock, invalidate data in the buffer
2909 * cache, as required, so that the coherency conforms with
2910 * RFC3530 Sec. 9.3.2.
2912 if (ap->a_op == F_SETLK) {
2913 if ((np->n_flag & NMODIFIED) == 0) {
2914 np->n_attrstamp = 0;
2915 ret = VOP_GETATTR(vp, &va, cred);
2917 if ((np->n_flag & NMODIFIED) || ret ||
2918 np->n_change != va.va_filerev) {
2919 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
2920 np->n_attrstamp = 0;
2921 ret = VOP_GETATTR(vp, &va, cred);
2923 np->n_mtime = va.va_mtime;
2924 np->n_change = va.va_filerev;
2930 } else if (!NFS_ISV4(vp)) {
2931 error = vn_lock(vp, LK_SHARED);
2934 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2935 size = VTONFS(vp)->n_size;
2937 error = lf_advlock(ap, &(vp->v_lockf), size);
2939 if (nfs_advlock_p != NULL)
2940 error = nfs_advlock_p(ap);
2951 * NFS advisory byte-level locks.
2954 nfs_advlockasync(struct vop_advlockasync_args *ap)
2956 struct vnode *vp = ap->a_vp;
2961 return (EOPNOTSUPP);
2962 error = vn_lock(vp, LK_SHARED);
2965 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2966 size = VTONFS(vp)->n_size;
2968 error = lf_advlockasync(ap, &(vp->v_lockf), size);
2977 * Print out the contents of an nfsnode.
2980 nfs_print(struct vop_print_args *ap)
2982 struct vnode *vp = ap->a_vp;
2983 struct nfsnode *np = VTONFS(vp);
2985 ncl_printf("\tfileid %ld fsid 0x%x",
2986 np->n_vattr.na_fileid, np->n_vattr.na_fsid);
2987 if (vp->v_type == VFIFO)
2994 * This is the "real" nfs::bwrite(struct buf*).
2995 * We set B_CACHE if this is a VMIO buffer.
2998 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3001 int oldflags = bp->b_flags;
3007 BUF_ASSERT_HELD(bp);
3009 if (bp->b_flags & B_INVAL) {
3014 bp->b_flags |= B_CACHE;
3017 * Undirty the bp. We will redirty it later if the I/O fails.
3022 bp->b_flags &= ~B_DONE;
3023 bp->b_ioflags &= ~BIO_ERROR;
3024 bp->b_iocmd = BIO_WRITE;
3026 bufobj_wref(bp->b_bufobj);
3027 curthread->td_ru.ru_oublock++;
3031 * Note: to avoid loopback deadlocks, we do not
3032 * assign b_runningbufspace.
3034 vfs_busy_pages(bp, 1);
3037 bp->b_iooffset = dbtob(bp->b_blkno);
3040 if( (oldflags & B_ASYNC) == 0) {
3041 int rtval = bufwait(bp);
3043 if (oldflags & B_DELWRI) {
3056 * nfs special file access vnode op.
3057 * Essentially just get vattr and then imitate iaccess() since the device is
3058 * local to the client.
3061 nfsspec_access(struct vop_access_args *ap)
3064 struct ucred *cred = ap->a_cred;
3065 struct vnode *vp = ap->a_vp;
3066 accmode_t accmode = ap->a_accmode;
3071 * Disallow write attempts on filesystems mounted read-only;
3072 * unless the file is a socket, fifo, or a block or character
3073 * device resident on the filesystem.
3075 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3076 switch (vp->v_type) {
3086 error = VOP_GETATTR(vp, vap, cred);
3089 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3090 accmode, cred, NULL);
3096 * Read wrapper for fifos.
3099 nfsfifo_read(struct vop_read_args *ap)
3101 struct nfsnode *np = VTONFS(ap->a_vp);
3107 mtx_lock(&np->n_mtx);
3109 getnanotime(&np->n_atim);
3110 mtx_unlock(&np->n_mtx);
3111 error = fifo_specops.vop_read(ap);
3116 * Write wrapper for fifos.
3119 nfsfifo_write(struct vop_write_args *ap)
3121 struct nfsnode *np = VTONFS(ap->a_vp);
3126 mtx_lock(&np->n_mtx);
3128 getnanotime(&np->n_mtim);
3129 mtx_unlock(&np->n_mtx);
3130 return(fifo_specops.vop_write(ap));
3134 * Close wrapper for fifos.
3136 * Update the times on the nfsnode then do fifo close.
3139 nfsfifo_close(struct vop_close_args *ap)
3141 struct vnode *vp = ap->a_vp;
3142 struct nfsnode *np = VTONFS(vp);
3146 mtx_lock(&np->n_mtx);
3147 if (np->n_flag & (NACC | NUPD)) {
3149 if (np->n_flag & NACC)
3151 if (np->n_flag & NUPD)
3154 if (vrefcnt(vp) == 1 &&
3155 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3157 if (np->n_flag & NACC)
3158 vattr.va_atime = np->n_atim;
3159 if (np->n_flag & NUPD)
3160 vattr.va_mtime = np->n_mtim;
3161 mtx_unlock(&np->n_mtx);
3162 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3166 mtx_unlock(&np->n_mtx);
3168 return (fifo_specops.vop_close(ap));
3172 * Just call ncl_writebp() with the force argument set to 1.
3174 * NOTE: B_DONE may or may not be set in a_bp on call.
3177 nfs_bwrite(struct buf *bp)
3180 return (ncl_writebp(bp, 1, curthread));
3183 struct buf_ops buf_ops_newnfs = {
3184 .bop_name = "buf_ops_nfs",
3185 .bop_write = nfs_bwrite,
3186 .bop_strategy = bufstrategy,
3187 .bop_sync = bufsync,
3188 .bop_bdflush = bufbdflush,
3192 * Cloned from vop_stdlock(), and then the ugly hack added.
3195 nfs_lock1(struct vop_lock1_args *ap)
3197 struct vnode *vp = ap->a_vp;
3201 * Since vfs_hash_get() calls vget() and it will no longer work
3202 * for FreeBSD8 with flags == 0, I can only think of this horrible
3203 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3204 * and then handle it here. All I want for this case is a v_usecount
3205 * on the vnode to use for recovery, while another thread might
3206 * hold a lock on the vnode. I have the other threads blocked, so
3207 * there isn't any race problem.
3209 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3210 if ((ap->a_flags & LK_INTERLOCK) == 0)
3212 if ((vp->v_iflag & VI_DOOMED))
3217 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3218 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3222 #ifdef NFS4_ACL_EXTATTR_NAME
3224 nfs_getacl(struct vop_getacl_args *ap)
3228 if (ap->a_type != ACL_TYPE_NFS4)
3229 return (EOPNOTSUPP);
3230 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3232 if (error > NFSERR_STALE) {
3233 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3240 nfs_setacl(struct vop_setacl_args *ap)
3244 if (ap->a_type != ACL_TYPE_NFS4)
3245 return (EOPNOTSUPP);
3246 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3248 if (error > NFSERR_STALE) {
3249 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3255 #endif /* NFS4_ACL_EXTATTR_NAME */