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 extern int nfsrv_useacl;
87 MALLOC_DECLARE(M_NEWNFSREQ);
90 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
91 * calls are not in getblk() and brelse() so that they would not be necessary
95 #define vfs_busy_pages(bp, f)
98 static vop_read_t nfsfifo_read;
99 static vop_write_t nfsfifo_write;
100 static vop_close_t nfsfifo_close;
101 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
103 static vop_lookup_t nfs_lookup;
104 static vop_create_t nfs_create;
105 static vop_mknod_t nfs_mknod;
106 static vop_open_t nfs_open;
107 static vop_pathconf_t nfs_pathconf;
108 static vop_close_t nfs_close;
109 static vop_access_t nfs_access;
110 static vop_getattr_t nfs_getattr;
111 static vop_setattr_t nfs_setattr;
112 static vop_read_t nfs_read;
113 static vop_fsync_t nfs_fsync;
114 static vop_remove_t nfs_remove;
115 static vop_link_t nfs_link;
116 static vop_rename_t nfs_rename;
117 static vop_mkdir_t nfs_mkdir;
118 static vop_rmdir_t nfs_rmdir;
119 static vop_symlink_t nfs_symlink;
120 static vop_readdir_t nfs_readdir;
121 static vop_strategy_t nfs_strategy;
122 static vop_lock1_t nfs_lock1;
123 static int nfs_lookitup(struct vnode *, char *, int,
124 struct ucred *, struct thread *, struct nfsnode **);
125 static int nfs_sillyrename(struct vnode *, struct vnode *,
126 struct componentname *);
127 static vop_access_t nfsspec_access;
128 static vop_readlink_t nfs_readlink;
129 static vop_print_t nfs_print;
130 static vop_advlock_t nfs_advlock;
131 static vop_advlockasync_t nfs_advlockasync;
132 static vop_getacl_t nfs_getacl;
133 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_pathconf = nfs_pathconf,
157 .vop_print = nfs_print,
158 .vop_read = nfs_read,
159 .vop_readdir = nfs_readdir,
160 .vop_readlink = nfs_readlink,
161 .vop_reclaim = ncl_reclaim,
162 .vop_remove = nfs_remove,
163 .vop_rename = nfs_rename,
164 .vop_rmdir = nfs_rmdir,
165 .vop_setattr = nfs_setattr,
166 .vop_strategy = nfs_strategy,
167 .vop_symlink = nfs_symlink,
168 .vop_write = ncl_write,
169 .vop_getacl = nfs_getacl,
170 .vop_setacl = nfs_setacl,
173 struct vop_vector newnfs_fifoops = {
174 .vop_default = &fifo_specops,
175 .vop_access = nfsspec_access,
176 .vop_close = nfsfifo_close,
177 .vop_fsync = nfs_fsync,
178 .vop_getattr = nfs_getattr,
179 .vop_inactive = ncl_inactive,
180 .vop_print = nfs_print,
181 .vop_read = nfsfifo_read,
182 .vop_reclaim = ncl_reclaim,
183 .vop_setattr = nfs_setattr,
184 .vop_write = nfsfifo_write,
187 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
188 struct componentname *cnp, struct vattr *vap);
189 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
190 int namelen, struct ucred *cred, struct thread *td);
191 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
192 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
193 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
194 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
195 struct componentname *scnp, struct sillyrename *sp);
200 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
202 SYSCTL_DECL(_vfs_newnfs);
204 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
205 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
206 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
208 static int nfs_prime_access_cache = 0;
209 SYSCTL_INT(_vfs_newnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
210 &nfs_prime_access_cache, 0,
211 "Prime NFS ACCESS cache when fetching attributes");
213 static int newnfs_commit_on_close = 0;
214 SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
215 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
217 static int nfs_clean_pages_on_close = 1;
218 SYSCTL_INT(_vfs_newnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
219 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
221 int newnfs_directio_enable = 0;
222 SYSCTL_INT(_vfs_newnfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
223 &newnfs_directio_enable, 0, "Enable NFS directio");
225 int newnfs_keep_dirty_on_error;
226 SYSCTL_INT(_vfs_newnfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
227 &newnfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
230 * This sysctl allows other processes to mmap a file that has been opened
231 * O_DIRECT by a process. In general, having processes mmap the file while
232 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
233 * this by default to prevent DoS attacks - to prevent a malicious user from
234 * opening up files O_DIRECT preventing other users from mmap'ing these
235 * files. "Protected" environments where stricter consistency guarantees are
236 * required can disable this knob. The process that opened the file O_DIRECT
237 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
240 int newnfs_directio_allow_mmap = 1;
241 SYSCTL_INT(_vfs_newnfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
242 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
245 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
246 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
248 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
249 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
252 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
253 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
254 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
258 * The list of locks after the description of the lock is the ordering
259 * of other locks acquired with the lock held.
260 * np->n_mtx : Protects the fields in the nfsnode.
262 VI_MTX (acquired indirectly)
263 * nmp->nm_mtx : Protects the fields in the nfsmount.
265 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
266 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
269 * rep->r_mtx : Protects the fields in an nfsreq.
273 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
274 struct ucred *cred, u_int32_t *retmode)
276 int error = 0, attrflag, i, lrupos;
278 struct nfsnode *np = VTONFS(vp);
279 struct nfsvattr nfsva;
281 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
284 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
287 mtx_lock(&np->n_mtx);
288 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
289 if (np->n_accesscache[i].uid == cred->cr_uid) {
290 np->n_accesscache[i].mode = rmode;
291 np->n_accesscache[i].stamp = time_second;
294 if (i > 0 && np->n_accesscache[i].stamp <
295 np->n_accesscache[lrupos].stamp)
298 if (i == NFS_ACCESSCACHESIZE) {
299 np->n_accesscache[lrupos].uid = cred->cr_uid;
300 np->n_accesscache[lrupos].mode = rmode;
301 np->n_accesscache[lrupos].stamp = time_second;
303 mtx_unlock(&np->n_mtx);
306 } else if (NFS_ISV4(vp)) {
307 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
313 * nfs access vnode op.
314 * For nfs version 2, just return ok. File accesses may fail later.
315 * For nfs version 3, use the access rpc to check accessibility. If file modes
316 * are changed on the server, accesses might still fail later.
319 nfs_access(struct vop_access_args *ap)
321 struct vnode *vp = ap->a_vp;
322 int error = 0, i, gotahit;
323 u_int32_t mode, wmode, rmode;
324 int v34 = NFS_ISV34(vp);
325 struct nfsnode *np = VTONFS(vp);
328 * Disallow write attempts on filesystems mounted read-only;
329 * unless the file is a socket, fifo, or a block or character
330 * device resident on the filesystem.
332 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
333 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
334 VWRITE_OWNER)) != 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 if (ap->a_accmode & VDELETE)
365 mode |= NFSACCESS_DELETE;
367 if (ap->a_accmode & VWRITE)
368 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
369 if (ap->a_accmode & VAPPEND)
370 mode |= NFSACCESS_EXTEND;
371 if (ap->a_accmode & VEXEC)
372 mode |= NFSACCESS_LOOKUP;
373 if (ap->a_accmode & VDELETE)
374 mode |= NFSACCESS_DELETE;
375 if (ap->a_accmode & VDELETE_CHILD)
376 mode |= NFSACCESS_MODIFY;
378 /* XXX safety belt, only make blanket request if caching */
379 if (nfsaccess_cache_timeout > 0) {
380 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
381 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
382 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
388 * Does our cached result allow us to give a definite yes to
392 mtx_lock(&np->n_mtx);
393 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
394 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
395 if (time_second < (np->n_accesscache[i].stamp
396 + nfsaccess_cache_timeout) &&
397 (np->n_accesscache[i].mode & mode) == mode) {
398 NFSINCRGLOBAL(newnfsstats.accesscache_hits);
404 mtx_unlock(&np->n_mtx);
407 * Either a no, or a don't know. Go to the wire.
409 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
410 error = nfs34_access_otw(vp, wmode, ap->a_td,
413 (rmode & mode) != mode)
418 if ((error = nfsspec_access(ap)) != 0) {
422 * Attempt to prevent a mapped root from accessing a file
423 * which it shouldn't. We try to read a byte from the file
424 * if the user is root and the file is not zero length.
425 * After calling nfsspec_access, we should have the correct
428 mtx_lock(&np->n_mtx);
429 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
430 && VTONFS(vp)->n_size > 0) {
435 mtx_unlock(&np->n_mtx);
438 auio.uio_iov = &aiov;
442 auio.uio_segflg = UIO_SYSSPACE;
443 auio.uio_rw = UIO_READ;
444 auio.uio_td = ap->a_td;
446 if (vp->v_type == VREG)
447 error = ncl_readrpc(vp, &auio, ap->a_cred);
448 else if (vp->v_type == VDIR) {
450 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
452 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
453 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
456 } else if (vp->v_type == VLNK)
457 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
461 mtx_unlock(&np->n_mtx);
469 * Check to see if the type is ok
470 * and that deletion is not in progress.
471 * For paged in text files, you will need to flush the page cache
472 * if consistency is lost.
476 nfs_open(struct vop_open_args *ap)
478 struct vnode *vp = ap->a_vp;
479 struct nfsnode *np = VTONFS(vp);
482 int fmode = ap->a_mode;
484 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
488 * For NFSv4, we need to do the Open Op before cache validation,
489 * so that we conform to RFC3530 Sec. 9.3.1.
492 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
494 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
501 * Now, if this Open will be doing reading, re-validate/flush the
502 * cache, so that Close/Open coherency is maintained.
504 mtx_lock(&np->n_mtx);
505 if (np->n_flag & NMODIFIED) {
506 mtx_unlock(&np->n_mtx);
507 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
508 if (error == EINTR || error == EIO) {
510 (void) nfsrpc_close(vp, 0, ap->a_td);
513 mtx_lock(&np->n_mtx);
515 if (vp->v_type == VDIR)
516 np->n_direofoffset = 0;
517 mtx_unlock(&np->n_mtx);
518 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
521 (void) nfsrpc_close(vp, 0, ap->a_td);
524 mtx_lock(&np->n_mtx);
525 np->n_mtime = vattr.va_mtime;
527 np->n_change = vattr.va_filerev;
529 mtx_unlock(&np->n_mtx);
530 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
533 (void) nfsrpc_close(vp, 0, ap->a_td);
536 mtx_lock(&np->n_mtx);
537 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
538 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
539 if (vp->v_type == VDIR)
540 np->n_direofoffset = 0;
541 mtx_unlock(&np->n_mtx);
542 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
543 if (error == EINTR || error == EIO) {
545 (void) nfsrpc_close(vp, 0, ap->a_td);
548 mtx_lock(&np->n_mtx);
549 np->n_mtime = vattr.va_mtime;
551 np->n_change = vattr.va_filerev;
556 * If the object has >= 1 O_DIRECT active opens, we disable caching.
558 if (newnfs_directio_enable && (fmode & O_DIRECT) &&
559 (vp->v_type == VREG)) {
560 if (np->n_directio_opens == 0) {
561 mtx_unlock(&np->n_mtx);
562 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
565 (void) nfsrpc_close(vp, 0, ap->a_td);
568 mtx_lock(&np->n_mtx);
569 np->n_flag |= NNONCACHE;
571 np->n_directio_opens++;
573 mtx_unlock(&np->n_mtx);
574 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
580 * What an NFS client should do upon close after writing is a debatable issue.
581 * Most NFS clients push delayed writes to the server upon close, basically for
583 * 1 - So that any write errors may be reported back to the client process
584 * doing the close system call. By far the two most likely errors are
585 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
586 * 2 - To put a worst case upper bound on cache inconsistency between
587 * multiple clients for the file.
588 * There is also a consistency problem for Version 2 of the protocol w.r.t.
589 * not being able to tell if other clients are writing a file concurrently,
590 * since there is no way of knowing if the changed modify time in the reply
591 * is only due to the write for this client.
592 * (NFS Version 3 provides weak cache consistency data in the reply that
593 * should be sufficient to detect and handle this case.)
595 * The current code does the following:
596 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
597 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
598 * or commit them (this satisfies 1 and 2 except for the
599 * case where the server crashes after this close but
600 * before the commit RPC, which is felt to be "good
601 * enough". Changing the last argument to ncl_flush() to
602 * a 1 would force a commit operation, if it is felt a
603 * commit is necessary now.
604 * for NFS Version 4 - flush the dirty buffers and commit them, if
605 * nfscl_mustflush() says this is necessary.
606 * It is necessary if there is no write delegation held,
607 * in order to satisfy open/close coherency.
608 * If the file isn't cached on local stable storage,
609 * it may be necessary in order to detect "out of space"
610 * errors from the server, if the write delegation
611 * issued by the server doesn't allow the file to grow.
615 nfs_close(struct vop_close_args *ap)
617 struct vnode *vp = ap->a_vp;
618 struct nfsnode *np = VTONFS(vp);
619 struct nfsvattr nfsva;
621 int error = 0, ret, localcred = 0;
622 int fmode = ap->a_fflag;
624 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
627 * During shutdown, a_cred isn't valid, so just use root.
629 if (ap->a_cred == NOCRED) {
630 cred = newnfs_getcred();
635 if (vp->v_type == VREG) {
637 * Examine and clean dirty pages, regardless of NMODIFIED.
638 * This closes a major hole in close-to-open consistency.
639 * We want to push out all dirty pages (and buffers) on
640 * close, regardless of whether they were dirtied by
641 * mmap'ed writes or via write().
643 if (nfs_clean_pages_on_close && vp->v_object) {
644 VM_OBJECT_LOCK(vp->v_object);
645 vm_object_page_clean(vp->v_object, 0, 0, 0);
646 VM_OBJECT_UNLOCK(vp->v_object);
648 mtx_lock(&np->n_mtx);
649 if (np->n_flag & NMODIFIED) {
650 mtx_unlock(&np->n_mtx);
653 * Under NFSv3 we have dirty buffers to dispose of. We
654 * must flush them to the NFS server. We have the option
655 * of waiting all the way through the commit rpc or just
656 * waiting for the initial write. The default is to only
657 * wait through the initial write so the data is in the
658 * server's cache, which is roughly similar to the state
659 * a standard disk subsystem leaves the file in on close().
661 * We cannot clear the NMODIFIED bit in np->n_flag due to
662 * potential races with other processes, and certainly
663 * cannot clear it if we don't commit.
664 * These races occur when there is no longer the old
665 * traditional vnode locking implemented for Vnode Ops.
667 int cm = newnfs_commit_on_close ? 1 : 0;
668 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
669 /* np->n_flag &= ~NMODIFIED; */
670 } else if (NFS_ISV4(vp)) {
671 if (nfscl_mustflush(vp) != 0) {
672 int cm = newnfs_commit_on_close ? 1 : 0;
673 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
676 * as above w.r.t races when clearing
678 * np->n_flag &= ~NMODIFIED;
682 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
683 mtx_lock(&np->n_mtx);
686 * Invalidate the attribute cache in all cases.
687 * An open is going to fetch fresh attrs any way, other procs
688 * on this node that have file open will be forced to do an
689 * otw attr fetch, but this is safe.
690 * --> A user found that their RPC count dropped by 20% when
691 * this was commented out and I can't see any requirement
692 * for it, so I've disabled it when negative lookups are
693 * enabled. (What does this have to do with negative lookup
694 * caching? Well nothing, except it was reported by the
695 * same user that needed negative lookup caching and I wanted
696 * there to be a way to disable it to see if it
697 * is the cause of some caching/coherency issue that might
700 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0)
702 if (np->n_flag & NWRITEERR) {
703 np->n_flag &= ~NWRITEERR;
706 mtx_unlock(&np->n_mtx);
711 * Get attributes so "change" is up to date.
713 if (error == 0 && nfscl_mustflush(vp) != 0) {
714 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
717 np->n_change = nfsva.na_filerev;
718 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
726 ret = nfsrpc_close(vp, 0, ap->a_td);
730 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
733 if (newnfs_directio_enable)
734 KASSERT((np->n_directio_asyncwr == 0),
735 ("nfs_close: dirty unflushed (%d) directio buffers\n",
736 np->n_directio_asyncwr));
737 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
738 mtx_lock(&np->n_mtx);
739 KASSERT((np->n_directio_opens > 0),
740 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
741 np->n_directio_opens--;
742 if (np->n_directio_opens == 0)
743 np->n_flag &= ~NNONCACHE;
744 mtx_unlock(&np->n_mtx);
752 * nfs getattr call from vfs.
755 nfs_getattr(struct vop_getattr_args *ap)
757 struct vnode *vp = ap->a_vp;
758 struct thread *td = curthread; /* XXX */
759 struct nfsnode *np = VTONFS(vp);
761 struct nfsvattr nfsva;
762 struct vattr *vap = ap->a_vap;
766 * Update local times for special files.
768 mtx_lock(&np->n_mtx);
769 if (np->n_flag & (NACC | NUPD))
771 mtx_unlock(&np->n_mtx);
773 * First look in the cache.
775 if (ncl_getattrcache(vp, &vattr) == 0) {
776 vap->va_type = vattr.va_type;
777 vap->va_mode = vattr.va_mode;
778 vap->va_nlink = vattr.va_nlink;
779 vap->va_uid = vattr.va_uid;
780 vap->va_gid = vattr.va_gid;
781 vap->va_fsid = vattr.va_fsid;
782 vap->va_fileid = vattr.va_fileid;
783 vap->va_size = vattr.va_size;
784 vap->va_blocksize = vattr.va_blocksize;
785 vap->va_atime = vattr.va_atime;
786 vap->va_mtime = vattr.va_mtime;
787 vap->va_ctime = vattr.va_ctime;
788 vap->va_gen = vattr.va_gen;
789 vap->va_flags = vattr.va_flags;
790 vap->va_rdev = vattr.va_rdev;
791 vap->va_bytes = vattr.va_bytes;
792 vap->va_filerev = vattr.va_filerev;
794 * Get the local modify time for the case of a write
797 nfscl_deleggetmodtime(vp, &vap->va_mtime);
801 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
802 nfsaccess_cache_timeout > 0) {
803 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
804 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
805 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
806 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
810 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
812 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
815 * Get the local modify time for the case of a write
818 nfscl_deleggetmodtime(vp, &vap->va_mtime);
819 } else if (NFS_ISV4(vp)) {
820 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
829 nfs_setattr(struct vop_setattr_args *ap)
831 struct vnode *vp = ap->a_vp;
832 struct nfsnode *np = VTONFS(vp);
833 struct thread *td = curthread; /* XXX */
834 struct vattr *vap = ap->a_vap;
843 * Setting of flags and marking of atimes are not supported.
845 if (vap->va_flags != VNOVAL)
849 * Disallow write attempts if the filesystem is mounted read-only.
851 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
852 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
853 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
854 (vp->v_mount->mnt_flag & MNT_RDONLY))
856 if (vap->va_size != VNOVAL) {
857 switch (vp->v_type) {
864 if (vap->va_mtime.tv_sec == VNOVAL &&
865 vap->va_atime.tv_sec == VNOVAL &&
866 vap->va_mode == (mode_t)VNOVAL &&
867 vap->va_uid == (uid_t)VNOVAL &&
868 vap->va_gid == (gid_t)VNOVAL)
870 vap->va_size = VNOVAL;
874 * Disallow write attempts if the filesystem is
877 if (vp->v_mount->mnt_flag & MNT_RDONLY)
880 * We run vnode_pager_setsize() early (why?),
881 * we must set np->n_size now to avoid vinvalbuf
882 * V_SAVE races that might setsize a lower
885 mtx_lock(&np->n_mtx);
887 mtx_unlock(&np->n_mtx);
888 error = ncl_meta_setsize(vp, ap->a_cred, td,
890 mtx_lock(&np->n_mtx);
891 if (np->n_flag & NMODIFIED) {
893 mtx_unlock(&np->n_mtx);
894 if (vap->va_size == 0)
895 error = ncl_vinvalbuf(vp, 0, td, 1);
897 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
899 vnode_pager_setsize(vp, tsize);
903 * Call nfscl_delegmodtime() to set the modify time
904 * locally, as required.
906 nfscl_delegmodtime(vp);
908 mtx_unlock(&np->n_mtx);
910 * np->n_size has already been set to vap->va_size
911 * in ncl_meta_setsize(). We must set it again since
912 * nfs_loadattrcache() could be called through
913 * ncl_meta_setsize() and could modify np->n_size.
915 mtx_lock(&np->n_mtx);
916 np->n_vattr.na_size = np->n_size = vap->va_size;
917 mtx_unlock(&np->n_mtx);
920 mtx_lock(&np->n_mtx);
921 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
922 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
923 mtx_unlock(&np->n_mtx);
924 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
925 (error == EINTR || error == EIO))
928 mtx_unlock(&np->n_mtx);
930 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
931 if (error && vap->va_size != VNOVAL) {
932 mtx_lock(&np->n_mtx);
933 np->n_size = np->n_vattr.na_size = tsize;
934 vnode_pager_setsize(vp, tsize);
935 mtx_unlock(&np->n_mtx);
941 * Do an nfs setattr rpc.
944 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
947 struct nfsnode *np = VTONFS(vp);
948 int error, ret, attrflag, i;
949 struct nfsvattr nfsva;
952 mtx_lock(&np->n_mtx);
953 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
954 np->n_accesscache[i].stamp = 0;
955 np->n_flag |= NDELEGMOD;
956 mtx_unlock(&np->n_mtx);
958 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
961 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
965 if (error && NFS_ISV4(vp))
966 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
971 * nfs lookup call, one step at a time...
972 * First look in cache
973 * If not found, unlock the directory nfsnode and do the rpc
976 nfs_lookup(struct vop_lookup_args *ap)
978 struct componentname *cnp = ap->a_cnp;
979 struct vnode *dvp = ap->a_dvp;
980 struct vnode **vpp = ap->a_vpp;
981 struct mount *mp = dvp->v_mount;
982 int flags = cnp->cn_flags;
984 struct nfsmount *nmp;
985 struct nfsnode *np, *newnp;
986 int error = 0, attrflag, dattrflag, ltype, ncticks;
987 struct thread *td = cnp->cn_thread;
989 struct nfsvattr dnfsva, nfsva;
991 struct timespec nctime;
994 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
995 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
997 if (dvp->v_type != VDIR)
1002 /* For NFSv4, wait until any remove is done. */
1003 mtx_lock(&np->n_mtx);
1004 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1005 np->n_flag |= NREMOVEWANT;
1006 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1008 mtx_unlock(&np->n_mtx);
1010 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1012 error = cache_lookup_times(dvp, vpp, cnp, &nctime, &ncticks);
1013 if (error > 0 && error != ENOENT)
1017 * Lookups of "." are special and always return the
1018 * current directory. cache_lookup() already handles
1019 * associated locking bookkeeping, etc.
1021 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1022 /* XXX: Is this really correct? */
1023 if (cnp->cn_nameiop != LOOKUP &&
1025 cnp->cn_flags |= SAVENAME;
1030 * We only accept a positive hit in the cache if the
1031 * change time of the file matches our cached copy.
1032 * Otherwise, we discard the cache entry and fallback
1033 * to doing a lookup RPC. We also only trust cache
1034 * entries for less than nm_nametimeo seconds.
1036 * To better handle stale file handles and attributes,
1037 * clear the attribute cache of this node if it is a
1038 * leaf component, part of an open() call, and not
1039 * locally modified before fetching the attributes.
1040 * This should allow stale file handles to be detected
1041 * here where we can fall back to a LOOKUP RPC to
1042 * recover rather than having nfs_open() detect the
1043 * stale file handle and failing open(2) with ESTALE.
1046 newnp = VTONFS(newvp);
1047 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1048 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1049 !(newnp->n_flag & NMODIFIED)) {
1050 mtx_lock(&newnp->n_mtx);
1051 newnp->n_attrstamp = 0;
1052 mtx_unlock(&newnp->n_mtx);
1054 if (nfscl_nodeleg(newvp, 0) == 0 ||
1055 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1056 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1057 timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1058 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1059 if (cnp->cn_nameiop != LOOKUP &&
1061 cnp->cn_flags |= SAVENAME;
1070 } else if (error == ENOENT) {
1071 if (dvp->v_iflag & VI_DOOMED)
1074 * We only accept a negative hit in the cache if the
1075 * modification time of the parent directory matches
1076 * the cached copy in the name cache entry.
1077 * Otherwise, we discard all of the negative cache
1078 * entries for this directory. We also only trust
1079 * negative cache entries for up to nm_negnametimeo
1082 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1083 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1084 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1085 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1088 cache_purge_negative(dvp);
1093 NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1094 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1095 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1098 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1100 if (newvp != NULLVP) {
1105 if (error != ENOENT) {
1107 error = nfscl_maperr(td, error, (uid_t)0,
1112 /* The requested file was not found. */
1113 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1114 (flags & ISLASTCN)) {
1116 * XXX: UFS does a full VOP_ACCESS(dvp,
1117 * VWRITE) here instead of just checking
1120 if (mp->mnt_flag & MNT_RDONLY)
1122 cnp->cn_flags |= SAVENAME;
1123 return (EJUSTRETURN);
1126 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
1129 * Cache the modification time of the parent
1130 * directory from the post-op attributes in
1131 * the name cache entry. The negative cache
1132 * entry will be ignored once the directory
1133 * has changed. Don't bother adding the entry
1134 * if the directory has already changed.
1136 mtx_lock(&np->n_mtx);
1137 if (timespeccmp(&np->n_vattr.na_mtime,
1138 &dnfsva.na_mtime, ==)) {
1139 mtx_unlock(&np->n_mtx);
1140 cache_enter_time(dvp, NULL, cnp,
1141 &dnfsva.na_mtime, NULL);
1143 mtx_unlock(&np->n_mtx);
1149 * Handle RENAME case...
1151 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1152 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1153 FREE((caddr_t)nfhp, M_NFSFH);
1156 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1162 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1165 cnp->cn_flags |= SAVENAME;
1169 if (flags & ISDOTDOT) {
1170 ltype = NFSVOPISLOCKED(dvp);
1171 error = vfs_busy(mp, MBF_NOWAIT);
1174 NFSVOPUNLOCK(dvp, 0);
1175 error = vfs_busy(mp, 0);
1176 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1178 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1185 NFSVOPUNLOCK(dvp, 0);
1186 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1192 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1193 if (dvp->v_iflag & VI_DOOMED) {
1205 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1207 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1208 FREE((caddr_t)nfhp, M_NFSFH);
1212 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1215 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1221 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1223 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1224 !(np->n_flag & NMODIFIED)) {
1226 * Flush the attribute cache when opening a
1227 * leaf node to ensure that fresh attributes
1228 * are fetched in nfs_open() since we did not
1229 * fetch attributes from the LOOKUP reply.
1231 mtx_lock(&np->n_mtx);
1232 np->n_attrstamp = 0;
1233 mtx_unlock(&np->n_mtx);
1236 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1237 cnp->cn_flags |= SAVENAME;
1238 if ((cnp->cn_flags & MAKEENTRY) &&
1239 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1240 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1241 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1242 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1249 * Just call ncl_bioread() to do the work.
1252 nfs_read(struct vop_read_args *ap)
1254 struct vnode *vp = ap->a_vp;
1256 switch (vp->v_type) {
1258 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1262 return (EOPNOTSUPP);
1270 nfs_readlink(struct vop_readlink_args *ap)
1272 struct vnode *vp = ap->a_vp;
1274 if (vp->v_type != VLNK)
1276 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1280 * Do a readlink rpc.
1281 * Called by ncl_doio() from below the buffer cache.
1284 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1286 int error, ret, attrflag;
1287 struct nfsvattr nfsva;
1289 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1292 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1296 if (error && NFS_ISV4(vp))
1297 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1306 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1308 int error, ret, attrflag;
1309 struct nfsvattr nfsva;
1311 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1314 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1318 if (error && NFS_ISV4(vp))
1319 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1327 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1328 int *iomode, int *must_commit, int called_from_strategy)
1330 struct nfsvattr nfsva;
1331 int error = 0, attrflag, ret;
1333 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1334 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy);
1336 if (VTONFS(vp)->n_flag & ND_NFSV4)
1337 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1340 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1346 *iomode = NFSWRITE_FILESYNC;
1347 if (error && NFS_ISV4(vp))
1348 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1354 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1355 * mode set to specify the file type and the size field for rdev.
1358 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1361 struct nfsvattr nfsva, dnfsva;
1362 struct vnode *newvp = NULL;
1363 struct nfsnode *np = NULL, *dnp;
1366 int error = 0, attrflag, dattrflag;
1369 if (vap->va_type == VCHR || vap->va_type == VBLK)
1370 rdev = vap->va_rdev;
1371 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1374 return (EOPNOTSUPP);
1375 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1377 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1378 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1379 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1382 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1383 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1384 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1387 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1388 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1391 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1394 if (attrflag != 0) {
1395 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1403 } else if (NFS_ISV4(dvp)) {
1404 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1408 mtx_lock(&dnp->n_mtx);
1409 dnp->n_flag |= NMODIFIED;
1411 dnp->n_attrstamp = 0;
1412 mtx_unlock(&dnp->n_mtx);
1418 * just call nfs_mknodrpc() to do the work.
1422 nfs_mknod(struct vop_mknod_args *ap)
1424 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1427 static u_long create_verf;
1429 * nfs file create call
1432 nfs_create(struct vop_create_args *ap)
1434 struct vnode *dvp = ap->a_dvp;
1435 struct vattr *vap = ap->a_vap;
1436 struct componentname *cnp = ap->a_cnp;
1437 struct nfsnode *np = NULL, *dnp;
1438 struct vnode *newvp = NULL;
1439 struct nfsmount *nmp;
1440 struct nfsvattr dnfsva, nfsva;
1443 int error = 0, attrflag, dattrflag, fmode = 0;
1447 * Oops, not for me..
1449 if (vap->va_type == VSOCK)
1450 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1452 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1454 if (vap->va_vaflags & VA_EXCLUSIVE)
1457 nmp = VFSTONFS(vnode_mount(dvp));
1459 /* For NFSv4, wait until any remove is done. */
1460 mtx_lock(&dnp->n_mtx);
1461 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1462 dnp->n_flag |= NREMOVEWANT;
1463 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1465 mtx_unlock(&dnp->n_mtx);
1468 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1470 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1471 cverf.lval[0] = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1474 cverf.lval[0] = create_verf;
1476 IN_IFADDR_RUNLOCK();
1479 cverf.lval[1] = ++create_verf;
1480 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1481 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1482 &nfhp, &attrflag, &dattrflag, NULL);
1485 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1486 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1487 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1490 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1491 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1494 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1498 error = nfsrpc_getattr(newvp, cnp->cn_cred,
1499 cnp->cn_thread, &nfsva, NULL);
1501 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1505 if (newvp != NULL) {
1509 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1510 error == NFSERR_NOTSUPP) {
1514 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1515 if (nfscl_checksattr(vap, &nfsva)) {
1517 * We are normally called with only a partially
1518 * initialized VAP. Since the NFSv3 spec says that
1519 * the server may use the file attributes to
1520 * store the verifier, the spec requires us to do a
1521 * SETATTR RPC. FreeBSD servers store the verifier in
1522 * atime, but we can't really assume that all servers
1523 * will so we ensure that our SETATTR sets both atime
1526 if (vap->va_mtime.tv_sec == VNOVAL)
1527 vfs_timestamp(&vap->va_mtime);
1528 if (vap->va_atime.tv_sec == VNOVAL)
1529 vap->va_atime = vap->va_mtime;
1530 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1531 cnp->cn_thread, &nfsva, &attrflag, NULL);
1532 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1533 vap->va_gid != (gid_t)VNOVAL)) {
1534 /* try again without setting uid/gid */
1535 vap->va_uid = (uid_t)VNOVAL;
1536 vap->va_gid = (uid_t)VNOVAL;
1537 error = nfsrpc_setattr(newvp, vap, NULL,
1538 cnp->cn_cred, cnp->cn_thread, &nfsva,
1542 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1549 if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1550 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1553 } else if (NFS_ISV4(dvp)) {
1554 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1557 mtx_lock(&dnp->n_mtx);
1558 dnp->n_flag |= NMODIFIED;
1560 dnp->n_attrstamp = 0;
1561 mtx_unlock(&dnp->n_mtx);
1566 * nfs file remove call
1567 * To try and make nfs semantics closer to ufs semantics, a file that has
1568 * other processes using the vnode is renamed instead of removed and then
1569 * removed later on the last close.
1570 * - If v_usecount > 1
1571 * If a rename is not already in the works
1572 * call nfs_sillyrename() to set it up
1577 nfs_remove(struct vop_remove_args *ap)
1579 struct vnode *vp = ap->a_vp;
1580 struct vnode *dvp = ap->a_dvp;
1581 struct componentname *cnp = ap->a_cnp;
1582 struct nfsnode *np = VTONFS(vp);
1586 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1587 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1588 if (vp->v_type == VDIR)
1590 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1591 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1592 vattr.va_nlink > 1)) {
1594 * Purge the name cache so that the chance of a lookup for
1595 * the name succeeding while the remove is in progress is
1596 * minimized. Without node locking it can still happen, such
1597 * that an I/O op returns ESTALE, but since you get this if
1598 * another host removes the file..
1602 * throw away biocache buffers, mainly to avoid
1603 * unnecessary delayed writes later.
1605 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1607 if (error != EINTR && error != EIO)
1608 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1609 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1611 * Kludge City: If the first reply to the remove rpc is lost..
1612 * the reply to the retransmitted request will be ENOENT
1613 * since the file was in fact removed
1614 * Therefore, we cheat and return success.
1616 if (error == ENOENT)
1618 } else if (!np->n_sillyrename)
1619 error = nfs_sillyrename(dvp, vp, cnp);
1620 mtx_lock(&np->n_mtx);
1621 np->n_attrstamp = 0;
1622 mtx_unlock(&np->n_mtx);
1627 * nfs file remove rpc called from nfs_inactive
1630 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1633 * Make sure that the directory vnode is still valid.
1634 * XXX we should lock sp->s_dvp here.
1636 if (sp->s_dvp->v_type == VBAD)
1638 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1643 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1646 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1647 int namelen, struct ucred *cred, struct thread *td)
1649 struct nfsvattr dnfsva;
1650 struct nfsnode *dnp = VTONFS(dvp);
1651 int error = 0, dattrflag;
1653 mtx_lock(&dnp->n_mtx);
1654 dnp->n_flag |= NREMOVEINPROG;
1655 mtx_unlock(&dnp->n_mtx);
1656 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1658 mtx_lock(&dnp->n_mtx);
1659 if ((dnp->n_flag & NREMOVEWANT)) {
1660 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1661 mtx_unlock(&dnp->n_mtx);
1662 wakeup((caddr_t)dnp);
1664 dnp->n_flag &= ~NREMOVEINPROG;
1665 mtx_unlock(&dnp->n_mtx);
1668 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1669 mtx_lock(&dnp->n_mtx);
1670 dnp->n_flag |= NMODIFIED;
1672 dnp->n_attrstamp = 0;
1673 mtx_unlock(&dnp->n_mtx);
1674 if (error && NFS_ISV4(dvp))
1675 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1680 * nfs file rename call
1683 nfs_rename(struct vop_rename_args *ap)
1685 struct vnode *fvp = ap->a_fvp;
1686 struct vnode *tvp = ap->a_tvp;
1687 struct vnode *fdvp = ap->a_fdvp;
1688 struct vnode *tdvp = ap->a_tdvp;
1689 struct componentname *tcnp = ap->a_tcnp;
1690 struct componentname *fcnp = ap->a_fcnp;
1691 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1692 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1693 struct nfsv4node *newv4 = NULL;
1696 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1697 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1698 /* Check for cross-device rename */
1699 if ((fvp->v_mount != tdvp->v_mount) ||
1700 (tvp && (fvp->v_mount != tvp->v_mount))) {
1706 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1710 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1714 * We have to flush B_DELWRI data prior to renaming
1715 * the file. If we don't, the delayed-write buffers
1716 * can be flushed out later after the file has gone stale
1717 * under NFSV3. NFSV2 does not have this problem because
1718 * ( as far as I can tell ) it flushes dirty buffers more
1721 * Skip the rename operation if the fsync fails, this can happen
1722 * due to the server's volume being full, when we pushed out data
1723 * that was written back to our cache earlier. Not checking for
1724 * this condition can result in potential (silent) data loss.
1726 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1727 NFSVOPUNLOCK(fvp, 0);
1729 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1734 * If the tvp exists and is in use, sillyrename it before doing the
1735 * rename of the new file over it.
1736 * XXX Can't sillyrename a directory.
1738 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1739 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1744 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1745 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1748 if (error == 0 && NFS_ISV4(tdvp)) {
1750 * For NFSv4, check to see if it is the same name and
1751 * replace the name, if it is different.
1753 MALLOC(newv4, struct nfsv4node *,
1754 sizeof (struct nfsv4node) +
1755 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1756 M_NFSV4NODE, M_WAITOK);
1757 mtx_lock(&tdnp->n_mtx);
1758 mtx_lock(&fnp->n_mtx);
1759 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1760 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1761 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1762 tcnp->cn_namelen) ||
1763 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1764 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1765 tdnp->n_fhp->nfh_len))) {
1767 { char nnn[100]; int nnnl;
1768 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1769 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1771 printf("ren replace=%s\n",nnn);
1774 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1777 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1778 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1779 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1780 tdnp->n_fhp->nfh_len);
1781 NFSBCOPY(tcnp->cn_nameptr,
1782 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1784 mtx_unlock(&tdnp->n_mtx);
1785 mtx_unlock(&fnp->n_mtx);
1787 FREE((caddr_t)newv4, M_NFSV4NODE);
1790 if (fvp->v_type == VDIR) {
1791 if (tvp != NULL && tvp->v_type == VDIR)
1806 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1808 if (error == ENOENT)
1814 * nfs file rename rpc called from nfs_remove() above
1817 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1818 struct sillyrename *sp)
1821 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1822 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1827 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1830 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1831 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1832 int tnamelen, struct ucred *cred, struct thread *td)
1834 struct nfsvattr fnfsva, tnfsva;
1835 struct nfsnode *fdnp = VTONFS(fdvp);
1836 struct nfsnode *tdnp = VTONFS(tdvp);
1837 int error = 0, fattrflag, tattrflag;
1839 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1840 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1841 &tattrflag, NULL, NULL);
1842 mtx_lock(&fdnp->n_mtx);
1843 fdnp->n_flag |= NMODIFIED;
1844 if (fattrflag != 0) {
1845 mtx_unlock(&fdnp->n_mtx);
1846 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1848 fdnp->n_attrstamp = 0;
1849 mtx_unlock(&fdnp->n_mtx);
1851 mtx_lock(&tdnp->n_mtx);
1852 tdnp->n_flag |= NMODIFIED;
1853 if (tattrflag != 0) {
1854 mtx_unlock(&tdnp->n_mtx);
1855 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1857 tdnp->n_attrstamp = 0;
1858 mtx_unlock(&tdnp->n_mtx);
1860 if (error && NFS_ISV4(fdvp))
1861 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1866 * nfs hard link create call
1869 nfs_link(struct vop_link_args *ap)
1871 struct vnode *vp = ap->a_vp;
1872 struct vnode *tdvp = ap->a_tdvp;
1873 struct componentname *cnp = ap->a_cnp;
1874 struct nfsnode *np, *tdnp;
1875 struct nfsvattr nfsva, dnfsva;
1876 int error = 0, attrflag, dattrflag;
1878 if (vp->v_mount != tdvp->v_mount) {
1883 * Push all writes to the server, so that the attribute cache
1884 * doesn't get "out of sync" with the server.
1885 * XXX There should be a better way!
1887 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1889 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1890 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1892 tdnp = VTONFS(tdvp);
1893 mtx_lock(&tdnp->n_mtx);
1894 tdnp->n_flag |= NMODIFIED;
1895 if (dattrflag != 0) {
1896 mtx_unlock(&tdnp->n_mtx);
1897 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1899 tdnp->n_attrstamp = 0;
1900 mtx_unlock(&tdnp->n_mtx);
1903 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1906 mtx_lock(&np->n_mtx);
1907 np->n_attrstamp = 0;
1908 mtx_unlock(&np->n_mtx);
1911 * If negative lookup caching is enabled, I might as well
1912 * add an entry for this node. Not necessary for correctness,
1913 * but if negative caching is enabled, then the system
1914 * must care about lookup caching hit rate, so...
1916 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1917 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
1918 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
1920 if (error && NFS_ISV4(vp))
1921 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1927 * nfs symbolic link create call
1930 nfs_symlink(struct vop_symlink_args *ap)
1932 struct vnode *dvp = ap->a_dvp;
1933 struct vattr *vap = ap->a_vap;
1934 struct componentname *cnp = ap->a_cnp;
1935 struct nfsvattr nfsva, dnfsva;
1937 struct nfsnode *np = NULL, *dnp;
1938 struct vnode *newvp = NULL;
1939 int error = 0, attrflag, dattrflag, ret;
1941 vap->va_type = VLNK;
1942 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1943 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1944 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1946 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
1947 &np, NULL, LK_EXCLUSIVE);
1953 if (newvp != NULL) {
1955 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1957 } else if (!error) {
1959 * If we do not have an error and we could not extract the
1960 * newvp from the response due to the request being NFSv2, we
1961 * have to do a lookup in order to obtain a newvp to return.
1963 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1964 cnp->cn_cred, cnp->cn_thread, &np);
1972 error = nfscl_maperr(cnp->cn_thread, error,
1973 vap->va_uid, vap->va_gid);
1979 mtx_lock(&dnp->n_mtx);
1980 dnp->n_flag |= NMODIFIED;
1981 if (dattrflag != 0) {
1982 mtx_unlock(&dnp->n_mtx);
1983 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1985 dnp->n_attrstamp = 0;
1986 mtx_unlock(&dnp->n_mtx);
1989 * If negative lookup caching is enabled, I might as well
1990 * add an entry for this node. Not necessary for correctness,
1991 * but if negative caching is enabled, then the system
1992 * must care about lookup caching hit rate, so...
1994 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
1995 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
1996 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2005 nfs_mkdir(struct vop_mkdir_args *ap)
2007 struct vnode *dvp = ap->a_dvp;
2008 struct vattr *vap = ap->a_vap;
2009 struct componentname *cnp = ap->a_cnp;
2010 struct nfsnode *np = NULL, *dnp;
2011 struct vnode *newvp = NULL;
2014 struct nfsvattr nfsva, dnfsva;
2015 int error = 0, attrflag, dattrflag, ret;
2017 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2019 vap->va_type = VDIR;
2020 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2021 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2022 &attrflag, &dattrflag, NULL);
2024 mtx_lock(&dnp->n_mtx);
2025 dnp->n_flag |= NMODIFIED;
2026 if (dattrflag != 0) {
2027 mtx_unlock(&dnp->n_mtx);
2028 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2030 dnp->n_attrstamp = 0;
2031 mtx_unlock(&dnp->n_mtx);
2034 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2035 &np, NULL, LK_EXCLUSIVE);
2039 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2044 if (!error && newvp == NULL) {
2045 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2046 cnp->cn_cred, cnp->cn_thread, &np);
2049 if (newvp->v_type != VDIR)
2057 error = nfscl_maperr(cnp->cn_thread, error,
2058 vap->va_uid, vap->va_gid);
2061 * If negative lookup caching is enabled, I might as well
2062 * add an entry for this node. Not necessary for correctness,
2063 * but if negative caching is enabled, then the system
2064 * must care about lookup caching hit rate, so...
2066 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2067 (cnp->cn_flags & MAKEENTRY) &&
2068 attrflag != 0 && dattrflag != 0)
2069 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2077 * nfs remove directory call
2080 nfs_rmdir(struct vop_rmdir_args *ap)
2082 struct vnode *vp = ap->a_vp;
2083 struct vnode *dvp = ap->a_dvp;
2084 struct componentname *cnp = ap->a_cnp;
2085 struct nfsnode *dnp;
2086 struct nfsvattr dnfsva;
2087 int error, dattrflag;
2091 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2092 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2094 mtx_lock(&dnp->n_mtx);
2095 dnp->n_flag |= NMODIFIED;
2096 if (dattrflag != 0) {
2097 mtx_unlock(&dnp->n_mtx);
2098 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2100 dnp->n_attrstamp = 0;
2101 mtx_unlock(&dnp->n_mtx);
2106 if (error && NFS_ISV4(dvp))
2107 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2110 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2112 if (error == ENOENT)
2121 nfs_readdir(struct vop_readdir_args *ap)
2123 struct vnode *vp = ap->a_vp;
2124 struct nfsnode *np = VTONFS(vp);
2125 struct uio *uio = ap->a_uio;
2126 int tresid, error = 0;
2129 if (vp->v_type != VDIR)
2133 * First, check for hit on the EOF offset cache
2135 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2136 (np->n_flag & NMODIFIED) == 0) {
2137 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2138 mtx_lock(&np->n_mtx);
2139 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2140 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2141 mtx_unlock(&np->n_mtx);
2142 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2145 mtx_unlock(&np->n_mtx);
2150 * Call ncl_bioread() to do the real work.
2152 tresid = uio->uio_resid;
2153 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2155 if (!error && uio->uio_resid == tresid)
2156 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2162 * Called from below the buffer cache by ncl_doio().
2165 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2168 struct nfsvattr nfsva;
2169 nfsuint64 *cookiep, cookie;
2170 struct nfsnode *dnp = VTONFS(vp);
2171 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2172 int error = 0, eof, attrflag;
2174 KASSERT(uiop->uio_iovcnt == 1 &&
2175 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2176 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2177 ("nfs readdirrpc bad uio"));
2180 * If there is no cookie, assume directory was stale.
2182 ncl_dircookie_lock(dnp);
2183 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2186 ncl_dircookie_unlock(dnp);
2188 ncl_dircookie_unlock(dnp);
2189 return (NFSERR_BAD_COOKIE);
2192 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2193 (void)ncl_fsinfo(nmp, vp, cred, td);
2195 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2196 &attrflag, &eof, NULL);
2198 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2202 * We are now either at the end of the directory or have filled
2206 dnp->n_direofoffset = uiop->uio_offset;
2208 if (uiop->uio_resid > 0)
2209 ncl_printf("EEK! readdirrpc resid > 0\n");
2210 ncl_dircookie_lock(dnp);
2211 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2213 ncl_dircookie_unlock(dnp);
2215 } else if (NFS_ISV4(vp)) {
2216 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2222 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2225 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2228 struct nfsvattr nfsva;
2229 nfsuint64 *cookiep, cookie;
2230 struct nfsnode *dnp = VTONFS(vp);
2231 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2232 int error = 0, attrflag, eof;
2234 KASSERT(uiop->uio_iovcnt == 1 &&
2235 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2236 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2237 ("nfs readdirplusrpc bad uio"));
2240 * If there is no cookie, assume directory was stale.
2242 ncl_dircookie_lock(dnp);
2243 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2246 ncl_dircookie_unlock(dnp);
2248 ncl_dircookie_unlock(dnp);
2249 return (NFSERR_BAD_COOKIE);
2252 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2253 (void)ncl_fsinfo(nmp, vp, cred, td);
2254 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2255 &attrflag, &eof, NULL);
2257 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2261 * We are now either at end of the directory or have filled the
2265 dnp->n_direofoffset = uiop->uio_offset;
2267 if (uiop->uio_resid > 0)
2268 ncl_printf("EEK! readdirplusrpc resid > 0\n");
2269 ncl_dircookie_lock(dnp);
2270 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2272 ncl_dircookie_unlock(dnp);
2274 } else if (NFS_ISV4(vp)) {
2275 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2281 * Silly rename. To make the NFS filesystem that is stateless look a little
2282 * more like the "ufs" a remove of an active vnode is translated to a rename
2283 * to a funny looking filename that is removed by nfs_inactive on the
2284 * nfsnode. There is the potential for another process on a different client
2285 * to create the same funny name between the nfs_lookitup() fails and the
2286 * nfs_rename() completes, but...
2289 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2291 struct sillyrename *sp;
2295 unsigned int lticks;
2299 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2300 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2301 M_NEWNFSREQ, M_WAITOK);
2302 sp->s_cred = crhold(cnp->cn_cred);
2307 * Fudge together a funny name.
2308 * Changing the format of the funny name to accomodate more
2309 * sillynames per directory.
2310 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2311 * CPU ticks since boot.
2313 pid = cnp->cn_thread->td_proc->p_pid;
2314 lticks = (unsigned int)ticks;
2316 sp->s_namlen = sprintf(sp->s_name,
2317 ".nfs.%08x.%04x4.4", lticks,
2319 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2320 cnp->cn_thread, NULL))
2324 error = nfs_renameit(dvp, vp, cnp, sp);
2327 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2328 cnp->cn_thread, &np);
2329 np->n_sillyrename = sp;
2334 free((caddr_t)sp, M_NEWNFSREQ);
2339 * Look up a file name and optionally either update the file handle or
2340 * allocate an nfsnode, depending on the value of npp.
2341 * npp == NULL --> just do the lookup
2342 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2344 * *npp != NULL --> update the file handle in the vnode
2347 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2348 struct thread *td, struct nfsnode **npp)
2350 struct vnode *newvp = NULL, *vp;
2351 struct nfsnode *np, *dnp = VTONFS(dvp);
2352 struct nfsfh *nfhp, *onfhp;
2353 struct nfsvattr nfsva, dnfsva;
2354 struct componentname cn;
2355 int error = 0, attrflag, dattrflag;
2358 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2359 &nfhp, &attrflag, &dattrflag, NULL);
2361 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2362 if (npp && !error) {
2367 * For NFSv4, check to see if it is the same name and
2368 * replace the name, if it is different.
2370 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2371 (np->n_v4->n4_namelen != len ||
2372 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2373 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2374 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2375 dnp->n_fhp->nfh_len))) {
2377 { char nnn[100]; int nnnl;
2378 nnnl = (len < 100) ? len : 99;
2379 bcopy(name, nnn, nnnl);
2381 printf("replace=%s\n",nnn);
2384 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2385 MALLOC(np->n_v4, struct nfsv4node *,
2386 sizeof (struct nfsv4node) +
2387 dnp->n_fhp->nfh_len + len - 1,
2388 M_NFSV4NODE, M_WAITOK);
2389 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2390 np->n_v4->n4_namelen = len;
2391 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2392 dnp->n_fhp->nfh_len);
2393 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2395 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2399 * Rehash node for new file handle.
2401 vfs_hash_rehash(vp, hash);
2404 FREE((caddr_t)onfhp, M_NFSFH);
2406 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2407 FREE((caddr_t)nfhp, M_NFSFH);
2411 cn.cn_nameptr = name;
2412 cn.cn_namelen = len;
2413 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2414 &np, NULL, LK_EXCLUSIVE);
2419 if (!attrflag && *npp == NULL) {
2427 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2430 if (npp && *npp == NULL) {
2441 if (error && NFS_ISV4(dvp))
2442 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2447 * Nfs Version 3 and 4 commit rpc
2450 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2453 struct nfsvattr nfsva;
2454 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2455 int error, attrflag;
2456 u_char verf[NFSX_VERF];
2458 mtx_lock(&nmp->nm_mtx);
2459 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2460 mtx_unlock(&nmp->nm_mtx);
2463 mtx_unlock(&nmp->nm_mtx);
2464 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2467 mtx_lock(&nmp->nm_mtx);
2468 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2469 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2470 error = NFSERR_STALEWRITEVERF;
2472 mtx_unlock(&nmp->nm_mtx);
2473 if (!error && attrflag)
2474 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2476 } else if (NFS_ISV4(vp)) {
2477 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2484 * For async requests when nfsiod(s) are running, queue the request by
2485 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2489 nfs_strategy(struct vop_strategy_args *ap)
2491 struct buf *bp = ap->a_bp;
2494 KASSERT(!(bp->b_flags & B_DONE),
2495 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2496 BUF_ASSERT_HELD(bp);
2498 if (bp->b_iocmd == BIO_READ)
2504 * If the op is asynchronous and an i/o daemon is waiting
2505 * queue the request, wake it up and wait for completion
2506 * otherwise just do it ourselves.
2508 if ((bp->b_flags & B_ASYNC) == 0 ||
2509 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2510 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2515 * fsync vnode op. Just call ncl_flush() with commit == 1.
2519 nfs_fsync(struct vop_fsync_args *ap)
2522 if (ap->a_vp->v_type != VREG) {
2524 * For NFS, metadata is changed synchronously on the server,
2525 * so there is nothing to flush. Also, ncl_flush() clears
2526 * the NMODIFIED flag and that shouldn't be done here for
2531 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2535 * Flush all the blocks associated with a vnode.
2536 * Walk through the buffer pool and push any dirty pages
2537 * associated with the vnode.
2538 * If the called_from_renewthread argument is TRUE, it has been called
2539 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2540 * waiting for a buffer write to complete.
2543 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2544 int commit, int called_from_renewthread)
2546 struct nfsnode *np = VTONFS(vp);
2550 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2551 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2552 int passone = 1, trycnt = 0;
2553 u_quad_t off, endoff, toff;
2554 struct ucred* wcred = NULL;
2555 struct buf **bvec = NULL;
2557 #ifndef NFS_COMMITBVECSIZ
2558 #define NFS_COMMITBVECSIZ 20
2560 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2561 int bvecsize = 0, bveccount;
2563 if (called_from_renewthread != 0)
2565 if (nmp->nm_flag & NFSMNT_INT)
2566 slpflag = NFS_PCATCH;
2571 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2572 * server, but has not been committed to stable storage on the server
2573 * yet. On the first pass, the byte range is worked out and the commit
2574 * rpc is done. On the second pass, ncl_writebp() is called to do the
2581 if (NFS_ISV34(vp) && commit) {
2582 if (bvec != NULL && bvec != bvec_on_stack)
2585 * Count up how many buffers waiting for a commit.
2589 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2590 if (!BUF_ISLOCKED(bp) &&
2591 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2592 == (B_DELWRI | B_NEEDCOMMIT))
2596 * Allocate space to remember the list of bufs to commit. It is
2597 * important to use M_NOWAIT here to avoid a race with nfs_write.
2598 * If we can't get memory (for whatever reason), we will end up
2599 * committing the buffers one-by-one in the loop below.
2601 if (bveccount > NFS_COMMITBVECSIZ) {
2603 * Release the vnode interlock to avoid a lock
2607 bvec = (struct buf **)
2608 malloc(bveccount * sizeof(struct buf *),
2612 bvec = bvec_on_stack;
2613 bvecsize = NFS_COMMITBVECSIZ;
2615 bvecsize = bveccount;
2617 bvec = bvec_on_stack;
2618 bvecsize = NFS_COMMITBVECSIZ;
2620 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2621 if (bvecpos >= bvecsize)
2623 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2624 nbp = TAILQ_NEXT(bp, b_bobufs);
2627 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2628 (B_DELWRI | B_NEEDCOMMIT)) {
2630 nbp = TAILQ_NEXT(bp, b_bobufs);
2636 * Work out if all buffers are using the same cred
2637 * so we can deal with them all with one commit.
2639 * NOTE: we are not clearing B_DONE here, so we have
2640 * to do it later on in this routine if we intend to
2641 * initiate I/O on the bp.
2643 * Note: to avoid loopback deadlocks, we do not
2644 * assign b_runningbufspace.
2647 wcred = bp->b_wcred;
2648 else if (wcred != bp->b_wcred)
2650 vfs_busy_pages(bp, 1);
2654 * bp is protected by being locked, but nbp is not
2655 * and vfs_busy_pages() may sleep. We have to
2658 nbp = TAILQ_NEXT(bp, b_bobufs);
2661 * A list of these buffers is kept so that the
2662 * second loop knows which buffers have actually
2663 * been committed. This is necessary, since there
2664 * may be a race between the commit rpc and new
2665 * uncommitted writes on the file.
2667 bvec[bvecpos++] = bp;
2668 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2672 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2680 * Commit data on the server, as required.
2681 * If all bufs are using the same wcred, then use that with
2682 * one call for all of them, otherwise commit each one
2685 if (wcred != NOCRED)
2686 retv = ncl_commit(vp, off, (int)(endoff - off),
2690 for (i = 0; i < bvecpos; i++) {
2693 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2695 size = (u_quad_t)(bp->b_dirtyend
2697 retv = ncl_commit(vp, off, (int)size,
2703 if (retv == NFSERR_STALEWRITEVERF)
2704 ncl_clearcommit(vp->v_mount);
2707 * Now, either mark the blocks I/O done or mark the
2708 * blocks dirty, depending on whether the commit
2711 for (i = 0; i < bvecpos; i++) {
2713 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2716 * Error, leave B_DELWRI intact
2718 vfs_unbusy_pages(bp);
2722 * Success, remove B_DELWRI ( bundirty() ).
2724 * b_dirtyoff/b_dirtyend seem to be NFS
2725 * specific. We should probably move that
2726 * into bundirty(). XXX
2729 bp->b_flags |= B_ASYNC;
2731 bp->b_flags &= ~B_DONE;
2732 bp->b_ioflags &= ~BIO_ERROR;
2733 bp->b_dirtyoff = bp->b_dirtyend = 0;
2740 * Start/do any write(s) that are required.
2744 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2745 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2746 if (waitfor != MNT_WAIT || passone)
2749 error = BUF_TIMELOCK(bp,
2750 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2751 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2756 if (error == ENOLCK) {
2760 if (called_from_renewthread != 0) {
2762 * Return EIO so the flush will be retried
2768 if (newnfs_sigintr(nmp, td)) {
2772 if (slpflag & PCATCH) {
2778 if ((bp->b_flags & B_DELWRI) == 0)
2779 panic("nfs_fsync: not dirty");
2780 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2786 if (passone || !commit)
2787 bp->b_flags |= B_ASYNC;
2789 bp->b_flags |= B_ASYNC;
2791 if (newnfs_sigintr(nmp, td)) {
2802 if (waitfor == MNT_WAIT) {
2803 while (bo->bo_numoutput) {
2804 error = bufobj_wwait(bo, slpflag, slptimeo);
2807 if (called_from_renewthread != 0) {
2809 * Return EIO so that the flush will be
2815 error = newnfs_sigintr(nmp, td);
2818 if (slpflag & PCATCH) {
2825 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2830 * Wait for all the async IO requests to drain
2833 mtx_lock(&np->n_mtx);
2834 while (np->n_directio_asyncwr > 0) {
2835 np->n_flag |= NFSYNCWAIT;
2836 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2837 &np->n_mtx, slpflag | (PRIBIO + 1),
2840 if (newnfs_sigintr(nmp, td)) {
2841 mtx_unlock(&np->n_mtx);
2847 mtx_unlock(&np->n_mtx);
2850 mtx_lock(&np->n_mtx);
2851 if (np->n_flag & NWRITEERR) {
2852 error = np->n_error;
2853 np->n_flag &= ~NWRITEERR;
2855 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2856 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2857 np->n_flag &= ~NMODIFIED;
2858 mtx_unlock(&np->n_mtx);
2860 if (bvec != NULL && bvec != bvec_on_stack)
2862 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2863 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2864 np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2865 /* try, try again... */
2870 printf("try%d\n", trycnt);
2877 * NFS advisory byte-level locks.
2880 nfs_advlock(struct vop_advlock_args *ap)
2882 struct vnode *vp = ap->a_vp;
2884 struct nfsnode *np = VTONFS(ap->a_vp);
2885 struct proc *p = (struct proc *)ap->a_id;
2886 struct thread *td = curthread; /* XXX */
2888 int ret, error = EOPNOTSUPP;
2891 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
2892 if (vp->v_type != VREG)
2894 if ((ap->a_flags & F_POSIX) != 0)
2897 cred = td->td_ucred;
2898 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2899 if (vp->v_iflag & VI_DOOMED) {
2900 NFSVOPUNLOCK(vp, 0);
2905 * If this is unlocking a write locked region, flush and
2906 * commit them before unlocking. This is required by
2907 * RFC3530 Sec. 9.3.2.
2909 if (ap->a_op == F_UNLCK &&
2910 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
2912 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
2915 * Loop around doing the lock op, while a blocking lock
2916 * must wait for the lock op to succeed.
2919 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
2920 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
2921 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2922 ap->a_op == F_SETLK) {
2923 NFSVOPUNLOCK(vp, 0);
2924 error = nfs_catnap(PZERO | PCATCH, ret,
2928 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2929 if (vp->v_iflag & VI_DOOMED) {
2930 NFSVOPUNLOCK(vp, 0);
2934 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2935 ap->a_op == F_SETLK);
2936 if (ret == NFSERR_DENIED) {
2937 NFSVOPUNLOCK(vp, 0);
2939 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
2940 NFSVOPUNLOCK(vp, 0);
2942 } else if (ret != 0) {
2943 NFSVOPUNLOCK(vp, 0);
2948 * Now, if we just got a lock, invalidate data in the buffer
2949 * cache, as required, so that the coherency conforms with
2950 * RFC3530 Sec. 9.3.2.
2952 if (ap->a_op == F_SETLK) {
2953 if ((np->n_flag & NMODIFIED) == 0) {
2954 np->n_attrstamp = 0;
2955 ret = VOP_GETATTR(vp, &va, cred);
2957 if ((np->n_flag & NMODIFIED) || ret ||
2958 np->n_change != va.va_filerev) {
2959 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
2960 np->n_attrstamp = 0;
2961 ret = VOP_GETATTR(vp, &va, cred);
2963 np->n_mtime = va.va_mtime;
2964 np->n_change = va.va_filerev;
2968 NFSVOPUNLOCK(vp, 0);
2970 } else if (!NFS_ISV4(vp)) {
2971 error = NFSVOPLOCK(vp, LK_SHARED);
2974 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2975 size = VTONFS(vp)->n_size;
2976 NFSVOPUNLOCK(vp, 0);
2977 error = lf_advlock(ap, &(vp->v_lockf), size);
2979 if (nfs_advlock_p != NULL)
2980 error = nfs_advlock_p(ap);
2982 NFSVOPUNLOCK(vp, 0);
2991 * NFS advisory byte-level locks.
2994 nfs_advlockasync(struct vop_advlockasync_args *ap)
2996 struct vnode *vp = ap->a_vp;
3001 return (EOPNOTSUPP);
3002 error = NFSVOPLOCK(vp, LK_SHARED);
3005 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3006 size = VTONFS(vp)->n_size;
3007 NFSVOPUNLOCK(vp, 0);
3008 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3010 NFSVOPUNLOCK(vp, 0);
3017 * Print out the contents of an nfsnode.
3020 nfs_print(struct vop_print_args *ap)
3022 struct vnode *vp = ap->a_vp;
3023 struct nfsnode *np = VTONFS(vp);
3025 ncl_printf("\tfileid %ld fsid 0x%x",
3026 np->n_vattr.na_fileid, np->n_vattr.na_fsid);
3027 if (vp->v_type == VFIFO)
3034 * This is the "real" nfs::bwrite(struct buf*).
3035 * We set B_CACHE if this is a VMIO buffer.
3038 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3041 int oldflags = bp->b_flags;
3047 BUF_ASSERT_HELD(bp);
3049 if (bp->b_flags & B_INVAL) {
3054 bp->b_flags |= B_CACHE;
3057 * Undirty the bp. We will redirty it later if the I/O fails.
3062 bp->b_flags &= ~B_DONE;
3063 bp->b_ioflags &= ~BIO_ERROR;
3064 bp->b_iocmd = BIO_WRITE;
3066 bufobj_wref(bp->b_bufobj);
3067 curthread->td_ru.ru_oublock++;
3071 * Note: to avoid loopback deadlocks, we do not
3072 * assign b_runningbufspace.
3074 vfs_busy_pages(bp, 1);
3077 bp->b_iooffset = dbtob(bp->b_blkno);
3080 if( (oldflags & B_ASYNC) == 0) {
3081 int rtval = bufwait(bp);
3083 if (oldflags & B_DELWRI) {
3096 * nfs special file access vnode op.
3097 * Essentially just get vattr and then imitate iaccess() since the device is
3098 * local to the client.
3101 nfsspec_access(struct vop_access_args *ap)
3104 struct ucred *cred = ap->a_cred;
3105 struct vnode *vp = ap->a_vp;
3106 accmode_t accmode = ap->a_accmode;
3111 * Disallow write attempts on filesystems mounted read-only;
3112 * unless the file is a socket, fifo, or a block or character
3113 * device resident on the filesystem.
3115 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3116 switch (vp->v_type) {
3126 error = VOP_GETATTR(vp, vap, cred);
3129 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3130 accmode, cred, NULL);
3136 * Read wrapper for fifos.
3139 nfsfifo_read(struct vop_read_args *ap)
3141 struct nfsnode *np = VTONFS(ap->a_vp);
3147 mtx_lock(&np->n_mtx);
3149 getnanotime(&np->n_atim);
3150 mtx_unlock(&np->n_mtx);
3151 error = fifo_specops.vop_read(ap);
3156 * Write wrapper for fifos.
3159 nfsfifo_write(struct vop_write_args *ap)
3161 struct nfsnode *np = VTONFS(ap->a_vp);
3166 mtx_lock(&np->n_mtx);
3168 getnanotime(&np->n_mtim);
3169 mtx_unlock(&np->n_mtx);
3170 return(fifo_specops.vop_write(ap));
3174 * Close wrapper for fifos.
3176 * Update the times on the nfsnode then do fifo close.
3179 nfsfifo_close(struct vop_close_args *ap)
3181 struct vnode *vp = ap->a_vp;
3182 struct nfsnode *np = VTONFS(vp);
3186 mtx_lock(&np->n_mtx);
3187 if (np->n_flag & (NACC | NUPD)) {
3189 if (np->n_flag & NACC)
3191 if (np->n_flag & NUPD)
3194 if (vrefcnt(vp) == 1 &&
3195 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3197 if (np->n_flag & NACC)
3198 vattr.va_atime = np->n_atim;
3199 if (np->n_flag & NUPD)
3200 vattr.va_mtime = np->n_mtim;
3201 mtx_unlock(&np->n_mtx);
3202 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3206 mtx_unlock(&np->n_mtx);
3208 return (fifo_specops.vop_close(ap));
3212 * Just call ncl_writebp() with the force argument set to 1.
3214 * NOTE: B_DONE may or may not be set in a_bp on call.
3217 nfs_bwrite(struct buf *bp)
3220 return (ncl_writebp(bp, 1, curthread));
3223 struct buf_ops buf_ops_newnfs = {
3224 .bop_name = "buf_ops_nfs",
3225 .bop_write = nfs_bwrite,
3226 .bop_strategy = bufstrategy,
3227 .bop_sync = bufsync,
3228 .bop_bdflush = bufbdflush,
3232 * Cloned from vop_stdlock(), and then the ugly hack added.
3235 nfs_lock1(struct vop_lock1_args *ap)
3237 struct vnode *vp = ap->a_vp;
3241 * Since vfs_hash_get() calls vget() and it will no longer work
3242 * for FreeBSD8 with flags == 0, I can only think of this horrible
3243 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3244 * and then handle it here. All I want for this case is a v_usecount
3245 * on the vnode to use for recovery, while another thread might
3246 * hold a lock on the vnode. I have the other threads blocked, so
3247 * there isn't any race problem.
3249 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3250 if ((ap->a_flags & LK_INTERLOCK) == 0)
3252 if ((vp->v_iflag & VI_DOOMED))
3257 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3258 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3263 nfs_getacl(struct vop_getacl_args *ap)
3267 if (ap->a_type != ACL_TYPE_NFS4)
3268 return (EOPNOTSUPP);
3269 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3271 if (error > NFSERR_STALE) {
3272 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3279 nfs_setacl(struct vop_setacl_args *ap)
3283 if (ap->a_type != ACL_TYPE_NFS4)
3284 return (EOPNOTSUPP);
3285 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3287 if (error > NFSERR_STALE) {
3288 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3295 * Return POSIX pathconf information applicable to nfs filesystems.
3298 nfs_pathconf(struct vop_pathconf_args *ap)
3300 struct nfsv3_pathconf pc;
3301 struct nfsvattr nfsva;
3302 struct vnode *vp = ap->a_vp;
3303 struct thread *td = curthread;
3304 int attrflag, error;
3306 if (NFS_ISV4(vp) || (NFS_ISV3(vp) && (ap->a_name == _PC_LINK_MAX ||
3307 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3308 ap->a_name == _PC_NO_TRUNC))) {
3310 * Since only the above 4 a_names are returned by the NFSv3
3311 * Pathconf RPC, there is no point in doing it for others.
3313 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3316 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3322 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3325 pc.pc_linkmax = LINK_MAX;
3326 pc.pc_namemax = NFS_MAXNAMLEN;
3328 pc.pc_chownrestricted = 1;
3329 pc.pc_caseinsensitive = 0;
3330 pc.pc_casepreserving = 1;
3333 switch (ap->a_name) {
3335 *ap->a_retval = pc.pc_linkmax;
3338 *ap->a_retval = pc.pc_namemax;
3341 *ap->a_retval = PATH_MAX;
3344 *ap->a_retval = PIPE_BUF;
3346 case _PC_CHOWN_RESTRICTED:
3347 *ap->a_retval = pc.pc_chownrestricted;
3350 *ap->a_retval = pc.pc_notrunc;
3352 case _PC_ACL_EXTENDED:
3356 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3357 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3362 case _PC_ACL_PATH_MAX:
3364 *ap->a_retval = ACL_MAX_ENTRIES;
3368 case _PC_MAC_PRESENT:
3372 /* _PC_ASYNC_IO should have been handled by upper layers. */
3373 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3382 case _PC_ALLOC_SIZE_MIN:
3383 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3385 case _PC_FILESIZEBITS:
3391 case _PC_REC_INCR_XFER_SIZE:
3392 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3394 case _PC_REC_MAX_XFER_SIZE:
3395 *ap->a_retval = -1; /* means ``unlimited'' */
3397 case _PC_REC_MIN_XFER_SIZE:
3398 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3400 case _PC_REC_XFER_ALIGN:
3401 *ap->a_retval = PAGE_SIZE;
3403 case _PC_SYMLINK_MAX:
3404 *ap->a_retval = NFS_MAXPATHLEN;