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, directio_enable, CTLFLAG_RW,
223 &newnfs_directio_enable, 0, "Enable NFS directio");
226 * This sysctl allows other processes to mmap a file that has been opened
227 * O_DIRECT by a process. In general, having processes mmap the file while
228 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
229 * this by default to prevent DoS attacks - to prevent a malicious user from
230 * opening up files O_DIRECT preventing other users from mmap'ing these
231 * files. "Protected" environments where stricter consistency guarantees are
232 * required can disable this knob. The process that opened the file O_DIRECT
233 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
236 int newnfs_directio_allow_mmap = 1;
237 SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_allow_mmap, CTLFLAG_RW,
238 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
241 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
242 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
244 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
245 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
248 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
249 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
250 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
254 * The list of locks after the description of the lock is the ordering
255 * of other locks acquired with the lock held.
256 * np->n_mtx : Protects the fields in the nfsnode.
258 VI_MTX (acquired indirectly)
259 * nmp->nm_mtx : Protects the fields in the nfsmount.
261 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
262 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
265 * rep->r_mtx : Protects the fields in an nfsreq.
269 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
270 struct ucred *cred, u_int32_t *retmode)
272 int error = 0, attrflag, i, lrupos;
274 struct nfsnode *np = VTONFS(vp);
275 struct nfsvattr nfsva;
277 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
280 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
283 mtx_lock(&np->n_mtx);
284 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
285 if (np->n_accesscache[i].uid == cred->cr_uid) {
286 np->n_accesscache[i].mode = rmode;
287 np->n_accesscache[i].stamp = time_second;
290 if (i > 0 && np->n_accesscache[i].stamp <
291 np->n_accesscache[lrupos].stamp)
294 if (i == NFS_ACCESSCACHESIZE) {
295 np->n_accesscache[lrupos].uid = cred->cr_uid;
296 np->n_accesscache[lrupos].mode = rmode;
297 np->n_accesscache[lrupos].stamp = time_second;
299 mtx_unlock(&np->n_mtx);
302 } else if (NFS_ISV4(vp)) {
303 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
309 * nfs access vnode op.
310 * For nfs version 2, just return ok. File accesses may fail later.
311 * For nfs version 3, use the access rpc to check accessibility. If file modes
312 * are changed on the server, accesses might still fail later.
315 nfs_access(struct vop_access_args *ap)
317 struct vnode *vp = ap->a_vp;
318 int error = 0, i, gotahit;
319 u_int32_t mode, wmode, rmode;
320 int v34 = NFS_ISV34(vp);
321 struct nfsnode *np = VTONFS(vp);
324 * Disallow write attempts on filesystems mounted read-only;
325 * unless the file is a socket, fifo, or a block or character
326 * device resident on the filesystem.
328 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
329 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
330 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
331 switch (vp->v_type) {
341 * For nfs v3 or v4, check to see if we have done this recently, and if
342 * so return our cached result instead of making an ACCESS call.
343 * If not, do an access rpc, otherwise you are stuck emulating
344 * ufs_access() locally using the vattr. This may not be correct,
345 * since the server may apply other access criteria such as
346 * client uid-->server uid mapping that we do not know about.
349 if (ap->a_accmode & VREAD)
350 mode = NFSACCESS_READ;
353 if (vp->v_type != VDIR) {
354 if (ap->a_accmode & VWRITE)
355 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
356 if (ap->a_accmode & VAPPEND)
357 mode |= NFSACCESS_EXTEND;
358 if (ap->a_accmode & VEXEC)
359 mode |= NFSACCESS_EXECUTE;
360 if (ap->a_accmode & VDELETE)
361 mode |= NFSACCESS_DELETE;
363 if (ap->a_accmode & VWRITE)
364 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
365 if (ap->a_accmode & VAPPEND)
366 mode |= NFSACCESS_EXTEND;
367 if (ap->a_accmode & VEXEC)
368 mode |= NFSACCESS_LOOKUP;
369 if (ap->a_accmode & VDELETE)
370 mode |= NFSACCESS_DELETE;
371 if (ap->a_accmode & VDELETE_CHILD)
372 mode |= NFSACCESS_MODIFY;
374 /* XXX safety belt, only make blanket request if caching */
375 if (nfsaccess_cache_timeout > 0) {
376 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
377 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
378 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
384 * Does our cached result allow us to give a definite yes to
388 mtx_lock(&np->n_mtx);
389 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
390 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
391 if (time_second < (np->n_accesscache[i].stamp
392 + nfsaccess_cache_timeout) &&
393 (np->n_accesscache[i].mode & mode) == mode) {
394 NFSINCRGLOBAL(newnfsstats.accesscache_hits);
400 mtx_unlock(&np->n_mtx);
403 * Either a no, or a don't know. Go to the wire.
405 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
406 error = nfs34_access_otw(vp, wmode, ap->a_td,
409 (rmode & mode) != mode)
414 if ((error = nfsspec_access(ap)) != 0) {
418 * Attempt to prevent a mapped root from accessing a file
419 * which it shouldn't. We try to read a byte from the file
420 * if the user is root and the file is not zero length.
421 * After calling nfsspec_access, we should have the correct
424 mtx_lock(&np->n_mtx);
425 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
426 && VTONFS(vp)->n_size > 0) {
431 mtx_unlock(&np->n_mtx);
434 auio.uio_iov = &aiov;
438 auio.uio_segflg = UIO_SYSSPACE;
439 auio.uio_rw = UIO_READ;
440 auio.uio_td = ap->a_td;
442 if (vp->v_type == VREG)
443 error = ncl_readrpc(vp, &auio, ap->a_cred);
444 else if (vp->v_type == VDIR) {
446 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
448 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
449 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
452 } else if (vp->v_type == VLNK)
453 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
457 mtx_unlock(&np->n_mtx);
465 * Check to see if the type is ok
466 * and that deletion is not in progress.
467 * For paged in text files, you will need to flush the page cache
468 * if consistency is lost.
472 nfs_open(struct vop_open_args *ap)
474 struct vnode *vp = ap->a_vp;
475 struct nfsnode *np = VTONFS(vp);
478 int fmode = ap->a_mode;
480 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
484 * For NFSv4, we need to do the Open Op before cache validation,
485 * so that we conform to RFC3530 Sec. 9.3.1.
488 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
490 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
497 * Now, if this Open will be doing reading, re-validate/flush the
498 * cache, so that Close/Open coherency is maintained.
500 mtx_lock(&np->n_mtx);
501 if (np->n_flag & NMODIFIED) {
502 mtx_unlock(&np->n_mtx);
503 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
504 if (error == EINTR || error == EIO) {
506 (void) nfsrpc_close(vp, 0, ap->a_td);
509 mtx_lock(&np->n_mtx);
511 if (vp->v_type == VDIR)
512 np->n_direofoffset = 0;
513 mtx_unlock(&np->n_mtx);
514 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
517 (void) nfsrpc_close(vp, 0, ap->a_td);
520 mtx_lock(&np->n_mtx);
521 np->n_mtime = vattr.va_mtime;
523 np->n_change = vattr.va_filerev;
525 mtx_unlock(&np->n_mtx);
526 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
529 (void) nfsrpc_close(vp, 0, ap->a_td);
532 mtx_lock(&np->n_mtx);
533 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
534 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
535 if (vp->v_type == VDIR)
536 np->n_direofoffset = 0;
537 mtx_unlock(&np->n_mtx);
538 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
539 if (error == EINTR || error == EIO) {
541 (void) nfsrpc_close(vp, 0, ap->a_td);
544 mtx_lock(&np->n_mtx);
545 np->n_mtime = vattr.va_mtime;
547 np->n_change = vattr.va_filerev;
552 * If the object has >= 1 O_DIRECT active opens, we disable caching.
554 if (newnfs_directio_enable && (fmode & O_DIRECT) &&
555 (vp->v_type == VREG)) {
556 if (np->n_directio_opens == 0) {
557 mtx_unlock(&np->n_mtx);
558 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
561 (void) nfsrpc_close(vp, 0, ap->a_td);
564 mtx_lock(&np->n_mtx);
565 np->n_flag |= NNONCACHE;
567 np->n_directio_opens++;
569 mtx_unlock(&np->n_mtx);
570 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
576 * What an NFS client should do upon close after writing is a debatable issue.
577 * Most NFS clients push delayed writes to the server upon close, basically for
579 * 1 - So that any write errors may be reported back to the client process
580 * doing the close system call. By far the two most likely errors are
581 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
582 * 2 - To put a worst case upper bound on cache inconsistency between
583 * multiple clients for the file.
584 * There is also a consistency problem for Version 2 of the protocol w.r.t.
585 * not being able to tell if other clients are writing a file concurrently,
586 * since there is no way of knowing if the changed modify time in the reply
587 * is only due to the write for this client.
588 * (NFS Version 3 provides weak cache consistency data in the reply that
589 * should be sufficient to detect and handle this case.)
591 * The current code does the following:
592 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
593 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
594 * or commit them (this satisfies 1 and 2 except for the
595 * case where the server crashes after this close but
596 * before the commit RPC, which is felt to be "good
597 * enough". Changing the last argument to ncl_flush() to
598 * a 1 would force a commit operation, if it is felt a
599 * commit is necessary now.
600 * for NFS Version 4 - flush the dirty buffers and commit them, if
601 * nfscl_mustflush() says this is necessary.
602 * It is necessary if there is no write delegation held,
603 * in order to satisfy open/close coherency.
604 * If the file isn't cached on local stable storage,
605 * it may be necessary in order to detect "out of space"
606 * errors from the server, if the write delegation
607 * issued by the server doesn't allow the file to grow.
611 nfs_close(struct vop_close_args *ap)
613 struct vnode *vp = ap->a_vp;
614 struct nfsnode *np = VTONFS(vp);
615 struct nfsvattr nfsva;
617 int error = 0, ret, localcred = 0;
618 int fmode = ap->a_fflag;
620 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
623 * During shutdown, a_cred isn't valid, so just use root.
625 if (ap->a_cred == NOCRED) {
626 cred = newnfs_getcred();
631 if (vp->v_type == VREG) {
633 * Examine and clean dirty pages, regardless of NMODIFIED.
634 * This closes a major hole in close-to-open consistency.
635 * We want to push out all dirty pages (and buffers) on
636 * close, regardless of whether they were dirtied by
637 * mmap'ed writes or via write().
639 if (nfs_clean_pages_on_close && vp->v_object) {
640 VM_OBJECT_LOCK(vp->v_object);
641 vm_object_page_clean(vp->v_object, 0, 0, 0);
642 VM_OBJECT_UNLOCK(vp->v_object);
644 mtx_lock(&np->n_mtx);
645 if (np->n_flag & NMODIFIED) {
646 mtx_unlock(&np->n_mtx);
649 * Under NFSv3 we have dirty buffers to dispose of. We
650 * must flush them to the NFS server. We have the option
651 * of waiting all the way through the commit rpc or just
652 * waiting for the initial write. The default is to only
653 * wait through the initial write so the data is in the
654 * server's cache, which is roughly similar to the state
655 * a standard disk subsystem leaves the file in on close().
657 * We cannot clear the NMODIFIED bit in np->n_flag due to
658 * potential races with other processes, and certainly
659 * cannot clear it if we don't commit.
660 * These races occur when there is no longer the old
661 * traditional vnode locking implemented for Vnode Ops.
663 int cm = newnfs_commit_on_close ? 1 : 0;
664 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
665 /* np->n_flag &= ~NMODIFIED; */
666 } else if (NFS_ISV4(vp)) {
667 if (nfscl_mustflush(vp) != 0) {
668 int cm = newnfs_commit_on_close ? 1 : 0;
669 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
672 * as above w.r.t races when clearing
674 * np->n_flag &= ~NMODIFIED;
678 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
679 mtx_lock(&np->n_mtx);
682 * Invalidate the attribute cache in all cases.
683 * An open is going to fetch fresh attrs any way, other procs
684 * on this node that have file open will be forced to do an
685 * otw attr fetch, but this is safe.
686 * --> A user found that their RPC count dropped by 20% when
687 * this was commented out and I can't see any requirement
688 * for it, so I've disabled it when negative lookups are
689 * enabled. (What does this have to do with negative lookup
690 * caching? Well nothing, except it was reported by the
691 * same user that needed negative lookup caching and I wanted
692 * there to be a way to disable it to see if it
693 * is the cause of some caching/coherency issue that might
696 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0)
698 if (np->n_flag & NWRITEERR) {
699 np->n_flag &= ~NWRITEERR;
702 mtx_unlock(&np->n_mtx);
707 * Get attributes so "change" is up to date.
709 if (error == 0 && nfscl_mustflush(vp) != 0) {
710 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
713 np->n_change = nfsva.na_filerev;
714 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
722 ret = nfsrpc_close(vp, 0, ap->a_td);
726 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
729 if (newnfs_directio_enable)
730 KASSERT((np->n_directio_asyncwr == 0),
731 ("nfs_close: dirty unflushed (%d) directio buffers\n",
732 np->n_directio_asyncwr));
733 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
734 mtx_lock(&np->n_mtx);
735 KASSERT((np->n_directio_opens > 0),
736 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
737 np->n_directio_opens--;
738 if (np->n_directio_opens == 0)
739 np->n_flag &= ~NNONCACHE;
740 mtx_unlock(&np->n_mtx);
748 * nfs getattr call from vfs.
751 nfs_getattr(struct vop_getattr_args *ap)
753 struct vnode *vp = ap->a_vp;
754 struct thread *td = curthread; /* XXX */
755 struct nfsnode *np = VTONFS(vp);
757 struct nfsvattr nfsva;
758 struct vattr *vap = ap->a_vap;
762 * Update local times for special files.
764 mtx_lock(&np->n_mtx);
765 if (np->n_flag & (NACC | NUPD))
767 mtx_unlock(&np->n_mtx);
769 * First look in the cache.
771 if (ncl_getattrcache(vp, &vattr) == 0) {
772 vap->va_type = vattr.va_type;
773 vap->va_mode = vattr.va_mode;
774 vap->va_nlink = vattr.va_nlink;
775 vap->va_uid = vattr.va_uid;
776 vap->va_gid = vattr.va_gid;
777 vap->va_fsid = vattr.va_fsid;
778 vap->va_fileid = vattr.va_fileid;
779 vap->va_size = vattr.va_size;
780 vap->va_blocksize = vattr.va_blocksize;
781 vap->va_atime = vattr.va_atime;
782 vap->va_mtime = vattr.va_mtime;
783 vap->va_ctime = vattr.va_ctime;
784 vap->va_gen = vattr.va_gen;
785 vap->va_flags = vattr.va_flags;
786 vap->va_rdev = vattr.va_rdev;
787 vap->va_bytes = vattr.va_bytes;
788 vap->va_filerev = vattr.va_filerev;
790 * Get the local modify time for the case of a write
793 nfscl_deleggetmodtime(vp, &vap->va_mtime);
797 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
798 nfsaccess_cache_timeout > 0) {
799 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
800 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
801 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
802 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
806 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
808 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
811 * Get the local modify time for the case of a write
814 nfscl_deleggetmodtime(vp, &vap->va_mtime);
815 } else if (NFS_ISV4(vp)) {
816 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
825 nfs_setattr(struct vop_setattr_args *ap)
827 struct vnode *vp = ap->a_vp;
828 struct nfsnode *np = VTONFS(vp);
829 struct thread *td = curthread; /* XXX */
830 struct vattr *vap = ap->a_vap;
839 * Setting of flags and marking of atimes are not supported.
841 if (vap->va_flags != VNOVAL)
845 * Disallow write attempts if the filesystem is mounted read-only.
847 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
848 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
849 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
850 (vp->v_mount->mnt_flag & MNT_RDONLY))
852 if (vap->va_size != VNOVAL) {
853 switch (vp->v_type) {
860 if (vap->va_mtime.tv_sec == VNOVAL &&
861 vap->va_atime.tv_sec == VNOVAL &&
862 vap->va_mode == (mode_t)VNOVAL &&
863 vap->va_uid == (uid_t)VNOVAL &&
864 vap->va_gid == (gid_t)VNOVAL)
866 vap->va_size = VNOVAL;
870 * Disallow write attempts if the filesystem is
873 if (vp->v_mount->mnt_flag & MNT_RDONLY)
876 * We run vnode_pager_setsize() early (why?),
877 * we must set np->n_size now to avoid vinvalbuf
878 * V_SAVE races that might setsize a lower
881 mtx_lock(&np->n_mtx);
883 mtx_unlock(&np->n_mtx);
884 error = ncl_meta_setsize(vp, ap->a_cred, td,
886 mtx_lock(&np->n_mtx);
887 if (np->n_flag & NMODIFIED) {
889 mtx_unlock(&np->n_mtx);
890 if (vap->va_size == 0)
891 error = ncl_vinvalbuf(vp, 0, td, 1);
893 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
895 vnode_pager_setsize(vp, tsize);
899 * Call nfscl_delegmodtime() to set the modify time
900 * locally, as required.
902 nfscl_delegmodtime(vp);
904 mtx_unlock(&np->n_mtx);
906 * np->n_size has already been set to vap->va_size
907 * in ncl_meta_setsize(). We must set it again since
908 * nfs_loadattrcache() could be called through
909 * ncl_meta_setsize() and could modify np->n_size.
911 mtx_lock(&np->n_mtx);
912 np->n_vattr.na_size = np->n_size = vap->va_size;
913 mtx_unlock(&np->n_mtx);
916 mtx_lock(&np->n_mtx);
917 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
918 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
919 mtx_unlock(&np->n_mtx);
920 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
921 (error == EINTR || error == EIO))
924 mtx_unlock(&np->n_mtx);
926 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
927 if (error && vap->va_size != VNOVAL) {
928 mtx_lock(&np->n_mtx);
929 np->n_size = np->n_vattr.na_size = tsize;
930 vnode_pager_setsize(vp, tsize);
931 mtx_unlock(&np->n_mtx);
937 * Do an nfs setattr rpc.
940 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
943 struct nfsnode *np = VTONFS(vp);
944 int error, ret, attrflag, i;
945 struct nfsvattr nfsva;
948 mtx_lock(&np->n_mtx);
949 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
950 np->n_accesscache[i].stamp = 0;
951 np->n_flag |= NDELEGMOD;
952 mtx_unlock(&np->n_mtx);
954 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
957 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
961 if (error && NFS_ISV4(vp))
962 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
967 * nfs lookup call, one step at a time...
968 * First look in cache
969 * If not found, unlock the directory nfsnode and do the rpc
972 nfs_lookup(struct vop_lookup_args *ap)
974 struct componentname *cnp = ap->a_cnp;
975 struct vnode *dvp = ap->a_dvp;
976 struct vnode **vpp = ap->a_vpp;
977 struct mount *mp = dvp->v_mount;
978 int flags = cnp->cn_flags;
980 struct nfsmount *nmp;
981 struct nfsnode *np, *newnp;
982 int error = 0, attrflag, dattrflag, ltype;
983 struct thread *td = cnp->cn_thread;
985 struct nfsvattr dnfsva, nfsva;
987 struct timespec dmtime;
990 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
991 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
993 if (dvp->v_type != VDIR)
998 /* For NFSv4, wait until any remove is done. */
999 mtx_lock(&np->n_mtx);
1000 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1001 np->n_flag |= NREMOVEWANT;
1002 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1004 mtx_unlock(&np->n_mtx);
1006 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1008 error = cache_lookup(dvp, vpp, cnp);
1009 if (error > 0 && error != ENOENT)
1013 * We only accept a positive hit in the cache if the
1014 * change time of the file matches our cached copy.
1015 * Otherwise, we discard the cache entry and fallback
1016 * to doing a lookup RPC.
1018 * To better handle stale file handles and attributes,
1019 * clear the attribute cache of this node if it is a
1020 * leaf component, part of an open() call, and not
1021 * locally modified before fetching the attributes.
1022 * This should allow stale file handles to be detected
1023 * here where we can fall back to a LOOKUP RPC to
1024 * recover rather than having nfs_open() detect the
1025 * stale file handle and failing open(2) with ESTALE.
1028 newnp = VTONFS(newvp);
1029 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1030 !(newnp->n_flag & NMODIFIED)) {
1031 mtx_lock(&newnp->n_mtx);
1032 newnp->n_attrstamp = 0;
1033 mtx_unlock(&newnp->n_mtx);
1035 if (nfscl_nodeleg(newvp, 0) == 0 ||
1036 (VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1037 timespeccmp(&vattr.va_ctime, &newnp->n_ctime, ==))) {
1038 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1039 if (cnp->cn_nameiop != LOOKUP &&
1041 cnp->cn_flags |= SAVENAME;
1050 } else if (error == ENOENT) {
1051 if (dvp->v_iflag & VI_DOOMED)
1054 * We only accept a negative hit in the cache if the
1055 * modification time of the parent directory matches
1056 * our cached copy. Otherwise, we discard all of the
1057 * negative cache entries for this directory. We also
1058 * only trust -ve cache entries for less than
1059 * nm_negative_namecache_timeout seconds.
1061 if ((u_int)(ticks - np->n_dmtime_ticks) <
1062 (nmp->nm_negnametimeo * hz) &&
1063 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1064 timespeccmp(&vattr.va_mtime, &np->n_dmtime, ==)) {
1065 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1068 cache_purge_negative(dvp);
1069 mtx_lock(&np->n_mtx);
1070 timespecclear(&np->n_dmtime);
1071 mtx_unlock(&np->n_mtx);
1075 * Cache the modification time of the parent directory in case
1076 * the lookup fails and results in adding the first negative
1077 * name cache entry for the directory. Since this is reading
1078 * a single time_t, don't bother with locking. The
1079 * modification time may be a bit stale, but it must be read
1080 * before performing the lookup RPC to prevent a race where
1081 * another lookup updates the timestamp on the directory after
1082 * the lookup RPC has been performed on the server but before
1083 * n_dmtime is set at the end of this function.
1085 dmtime = np->n_vattr.na_mtime;
1088 NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1089 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1090 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1093 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1095 if (newvp != NULLVP) {
1100 if (error != ENOENT) {
1102 error = nfscl_maperr(td, error, (uid_t)0,
1107 /* The requested file was not found. */
1108 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1109 (flags & ISLASTCN)) {
1111 * XXX: UFS does a full VOP_ACCESS(dvp,
1112 * VWRITE) here instead of just checking
1115 if (mp->mnt_flag & MNT_RDONLY)
1117 cnp->cn_flags |= SAVENAME;
1118 return (EJUSTRETURN);
1121 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
1123 * Maintain n_dmtime as the modification time
1124 * of the parent directory when the oldest -ve
1125 * name cache entry for this directory was
1126 * added. If a -ve cache entry has already
1127 * been added with a newer modification time
1128 * by a concurrent lookup, then don't bother
1129 * adding a cache entry. The modification
1130 * time of the directory might have changed
1131 * due to the file this lookup failed to find
1132 * being created. In that case a subsequent
1133 * lookup would incorrectly use the entry
1134 * added here instead of doing an extra
1137 mtx_lock(&np->n_mtx);
1138 if (timespeccmp(&np->n_dmtime, &dmtime, <=)) {
1139 if (!timespecisset(&np->n_dmtime)) {
1140 np->n_dmtime = dmtime;
1141 np->n_dmtime_ticks = ticks;
1143 mtx_unlock(&np->n_mtx);
1144 cache_enter(dvp, NULL, cnp);
1146 mtx_unlock(&np->n_mtx);
1152 * Handle RENAME case...
1154 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1155 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1156 FREE((caddr_t)nfhp, M_NFSFH);
1159 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1165 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1168 cnp->cn_flags |= SAVENAME;
1172 if (flags & ISDOTDOT) {
1173 ltype = VOP_ISLOCKED(dvp);
1174 error = vfs_busy(mp, MBF_NOWAIT);
1178 error = vfs_busy(mp, 0);
1179 vn_lock(dvp, ltype | LK_RETRY);
1181 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1189 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1195 vn_lock(dvp, ltype | LK_RETRY);
1196 if (dvp->v_iflag & VI_DOOMED) {
1208 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1210 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1211 FREE((caddr_t)nfhp, M_NFSFH);
1215 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1218 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1224 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1226 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1227 !(np->n_flag & NMODIFIED)) {
1229 * Flush the attribute cache when opening a
1230 * leaf node to ensure that fresh attributes
1231 * are fetched in nfs_open() since we did not
1232 * fetch attributes from the LOOKUP reply.
1234 mtx_lock(&np->n_mtx);
1235 np->n_attrstamp = 0;
1236 mtx_unlock(&np->n_mtx);
1239 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1240 cnp->cn_flags |= SAVENAME;
1241 if ((cnp->cn_flags & MAKEENTRY) &&
1242 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1243 np->n_ctime = np->n_vattr.na_vattr.va_ctime;
1244 cache_enter(dvp, newvp, cnp);
1252 * Just call ncl_bioread() to do the work.
1255 nfs_read(struct vop_read_args *ap)
1257 struct vnode *vp = ap->a_vp;
1259 switch (vp->v_type) {
1261 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1265 return (EOPNOTSUPP);
1273 nfs_readlink(struct vop_readlink_args *ap)
1275 struct vnode *vp = ap->a_vp;
1277 if (vp->v_type != VLNK)
1279 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1283 * Do a readlink rpc.
1284 * Called by ncl_doio() from below the buffer cache.
1287 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1289 int error, ret, attrflag;
1290 struct nfsvattr nfsva;
1292 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1295 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1299 if (error && NFS_ISV4(vp))
1300 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1309 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1311 int error, ret, attrflag;
1312 struct nfsvattr nfsva;
1314 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1317 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1321 if (error && NFS_ISV4(vp))
1322 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1330 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1331 int *iomode, int *must_commit, int called_from_strategy)
1333 struct nfsvattr nfsva;
1334 int error = 0, attrflag, ret;
1335 u_char verf[NFSX_VERF];
1336 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1339 error = nfsrpc_write(vp, uiop, iomode, verf, cred,
1340 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy);
1342 if (!error && NFSHASWRITEVERF(nmp) &&
1343 NFSBCMP(verf, nmp->nm_verf, NFSX_VERF)) {
1345 NFSBCOPY(verf, nmp->nm_verf, NFSX_VERF);
1349 if (VTONFS(vp)->n_flag & ND_NFSV4)
1350 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1353 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1358 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1359 *iomode = NFSWRITE_FILESYNC;
1360 if (error && NFS_ISV4(vp))
1361 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1367 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1368 * mode set to specify the file type and the size field for rdev.
1371 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1374 struct nfsvattr nfsva, dnfsva;
1375 struct vnode *newvp = NULL;
1376 struct nfsnode *np = NULL, *dnp;
1379 int error = 0, attrflag, dattrflag;
1382 if (vap->va_type == VCHR || vap->va_type == VBLK)
1383 rdev = vap->va_rdev;
1384 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1387 return (EOPNOTSUPP);
1388 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1390 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1391 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1392 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1395 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1396 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1397 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1400 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1401 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1404 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1408 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1412 if ((cnp->cn_flags & MAKEENTRY))
1413 cache_enter(dvp, newvp, cnp);
1415 } else if (NFS_ISV4(dvp)) {
1416 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1420 mtx_lock(&dnp->n_mtx);
1421 dnp->n_flag |= NMODIFIED;
1423 dnp->n_attrstamp = 0;
1424 mtx_unlock(&dnp->n_mtx);
1430 * just call nfs_mknodrpc() to do the work.
1434 nfs_mknod(struct vop_mknod_args *ap)
1436 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1439 static u_long create_verf;
1441 * nfs file create call
1444 nfs_create(struct vop_create_args *ap)
1446 struct vnode *dvp = ap->a_dvp;
1447 struct vattr *vap = ap->a_vap;
1448 struct componentname *cnp = ap->a_cnp;
1449 struct nfsnode *np = NULL, *dnp;
1450 struct vnode *newvp = NULL;
1451 struct nfsmount *nmp;
1452 struct nfsvattr dnfsva, nfsva;
1455 int error = 0, attrflag, dattrflag, fmode = 0;
1459 * Oops, not for me..
1461 if (vap->va_type == VSOCK)
1462 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1464 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1466 if (vap->va_vaflags & VA_EXCLUSIVE)
1469 nmp = VFSTONFS(vnode_mount(dvp));
1471 /* For NFSv4, wait until any remove is done. */
1472 mtx_lock(&dnp->n_mtx);
1473 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1474 dnp->n_flag |= NREMOVEWANT;
1475 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1477 mtx_unlock(&dnp->n_mtx);
1480 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1482 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1483 cverf.lval[0] = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1486 cverf.lval[0] = create_verf;
1488 IN_IFADDR_RUNLOCK();
1491 cverf.lval[1] = ++create_verf;
1492 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1493 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1494 &nfhp, &attrflag, &dattrflag, NULL);
1497 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1498 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1499 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1502 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1503 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1506 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1510 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1514 if (newvp != NULL) {
1518 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1519 error == NFSERR_NOTSUPP) {
1523 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1524 if (nfscl_checksattr(vap, &nfsva)) {
1525 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1526 cnp->cn_thread, &nfsva, &attrflag, NULL);
1527 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1528 vap->va_gid != (gid_t)VNOVAL)) {
1529 /* try again without setting uid/gid */
1530 vap->va_uid = (uid_t)VNOVAL;
1531 vap->va_gid = (uid_t)VNOVAL;
1532 error = nfsrpc_setattr(newvp, vap, NULL,
1533 cnp->cn_cred, cnp->cn_thread, &nfsva,
1537 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1542 if (cnp->cn_flags & MAKEENTRY)
1543 cache_enter(dvp, newvp, cnp);
1545 } else if (NFS_ISV4(dvp)) {
1546 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1549 mtx_lock(&dnp->n_mtx);
1550 dnp->n_flag |= NMODIFIED;
1552 dnp->n_attrstamp = 0;
1553 mtx_unlock(&dnp->n_mtx);
1558 * nfs file remove call
1559 * To try and make nfs semantics closer to ufs semantics, a file that has
1560 * other processes using the vnode is renamed instead of removed and then
1561 * removed later on the last close.
1562 * - If v_usecount > 1
1563 * If a rename is not already in the works
1564 * call nfs_sillyrename() to set it up
1569 nfs_remove(struct vop_remove_args *ap)
1571 struct vnode *vp = ap->a_vp;
1572 struct vnode *dvp = ap->a_dvp;
1573 struct componentname *cnp = ap->a_cnp;
1574 struct nfsnode *np = VTONFS(vp);
1578 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1579 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1580 if (vp->v_type == VDIR)
1582 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1583 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1584 vattr.va_nlink > 1)) {
1586 * Purge the name cache so that the chance of a lookup for
1587 * the name succeeding while the remove is in progress is
1588 * minimized. Without node locking it can still happen, such
1589 * that an I/O op returns ESTALE, but since you get this if
1590 * another host removes the file..
1594 * throw away biocache buffers, mainly to avoid
1595 * unnecessary delayed writes later.
1597 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1599 if (error != EINTR && error != EIO)
1600 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1601 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1603 * Kludge City: If the first reply to the remove rpc is lost..
1604 * the reply to the retransmitted request will be ENOENT
1605 * since the file was in fact removed
1606 * Therefore, we cheat and return success.
1608 if (error == ENOENT)
1610 } else if (!np->n_sillyrename)
1611 error = nfs_sillyrename(dvp, vp, cnp);
1612 np->n_attrstamp = 0;
1617 * nfs file remove rpc called from nfs_inactive
1620 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1623 * Make sure that the directory vnode is still valid.
1624 * XXX we should lock sp->s_dvp here.
1626 if (sp->s_dvp->v_type == VBAD)
1628 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1633 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1636 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1637 int namelen, struct ucred *cred, struct thread *td)
1639 struct nfsvattr dnfsva;
1640 struct nfsnode *dnp = VTONFS(dvp);
1641 int error = 0, dattrflag;
1643 mtx_lock(&dnp->n_mtx);
1644 dnp->n_flag |= NREMOVEINPROG;
1645 mtx_unlock(&dnp->n_mtx);
1646 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1648 mtx_lock(&dnp->n_mtx);
1649 if ((dnp->n_flag & NREMOVEWANT)) {
1650 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1651 mtx_unlock(&dnp->n_mtx);
1652 wakeup((caddr_t)dnp);
1654 dnp->n_flag &= ~NREMOVEINPROG;
1655 mtx_unlock(&dnp->n_mtx);
1658 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1659 mtx_lock(&dnp->n_mtx);
1660 dnp->n_flag |= NMODIFIED;
1662 dnp->n_attrstamp = 0;
1663 mtx_unlock(&dnp->n_mtx);
1664 if (error && NFS_ISV4(dvp))
1665 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1670 * nfs file rename call
1673 nfs_rename(struct vop_rename_args *ap)
1675 struct vnode *fvp = ap->a_fvp;
1676 struct vnode *tvp = ap->a_tvp;
1677 struct vnode *fdvp = ap->a_fdvp;
1678 struct vnode *tdvp = ap->a_tdvp;
1679 struct componentname *tcnp = ap->a_tcnp;
1680 struct componentname *fcnp = ap->a_fcnp;
1681 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1682 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1683 struct nfsv4node *newv4 = NULL;
1686 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1687 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1688 /* Check for cross-device rename */
1689 if ((fvp->v_mount != tdvp->v_mount) ||
1690 (tvp && (fvp->v_mount != tvp->v_mount))) {
1696 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1700 if ((error = vn_lock(fvp, LK_EXCLUSIVE)))
1704 * We have to flush B_DELWRI data prior to renaming
1705 * the file. If we don't, the delayed-write buffers
1706 * can be flushed out later after the file has gone stale
1707 * under NFSV3. NFSV2 does not have this problem because
1708 * ( as far as I can tell ) it flushes dirty buffers more
1711 * Skip the rename operation if the fsync fails, this can happen
1712 * due to the server's volume being full, when we pushed out data
1713 * that was written back to our cache earlier. Not checking for
1714 * this condition can result in potential (silent) data loss.
1716 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1719 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1724 * If the tvp exists and is in use, sillyrename it before doing the
1725 * rename of the new file over it.
1726 * XXX Can't sillyrename a directory.
1728 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1729 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1734 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1735 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1740 * For NFSv4, check to see if it is the same name and
1741 * replace the name, if it is different.
1743 MALLOC(newv4, struct nfsv4node *,
1744 sizeof (struct nfsv4node) +
1745 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1746 M_NFSV4NODE, M_WAITOK);
1747 mtx_lock(&tdnp->n_mtx);
1748 mtx_lock(&fnp->n_mtx);
1749 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1750 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1751 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1752 tcnp->cn_namelen) ||
1753 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1754 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1755 tdnp->n_fhp->nfh_len))) {
1757 { char nnn[100]; int nnnl;
1758 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1759 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1761 printf("ren replace=%s\n",nnn);
1764 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1767 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1768 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1769 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1770 tdnp->n_fhp->nfh_len);
1771 NFSBCOPY(tcnp->cn_nameptr,
1772 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1774 mtx_unlock(&tdnp->n_mtx);
1775 mtx_unlock(&fnp->n_mtx);
1777 FREE((caddr_t)newv4, M_NFSV4NODE);
1780 if (fvp->v_type == VDIR) {
1781 if (tvp != NULL && tvp->v_type == VDIR)
1796 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1798 if (error == ENOENT)
1804 * nfs file rename rpc called from nfs_remove() above
1807 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1808 struct sillyrename *sp)
1811 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1812 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1817 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1820 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1821 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1822 int tnamelen, struct ucred *cred, struct thread *td)
1824 struct nfsvattr fnfsva, tnfsva;
1825 struct nfsnode *fdnp = VTONFS(fdvp);
1826 struct nfsnode *tdnp = VTONFS(tdvp);
1827 int error = 0, fattrflag, tattrflag;
1829 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1830 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1831 &tattrflag, NULL, NULL);
1832 mtx_lock(&fdnp->n_mtx);
1833 fdnp->n_flag |= NMODIFIED;
1834 mtx_unlock(&fdnp->n_mtx);
1835 mtx_lock(&tdnp->n_mtx);
1836 tdnp->n_flag |= NMODIFIED;
1837 mtx_unlock(&tdnp->n_mtx);
1839 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1841 fdnp->n_attrstamp = 0;
1843 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1845 tdnp->n_attrstamp = 0;
1846 if (error && NFS_ISV4(fdvp))
1847 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1852 * nfs hard link create call
1855 nfs_link(struct vop_link_args *ap)
1857 struct vnode *vp = ap->a_vp;
1858 struct vnode *tdvp = ap->a_tdvp;
1859 struct componentname *cnp = ap->a_cnp;
1860 struct nfsnode *tdnp;
1861 struct nfsvattr nfsva, dnfsva;
1862 int error = 0, attrflag, dattrflag;
1864 if (vp->v_mount != tdvp->v_mount) {
1869 * Push all writes to the server, so that the attribute cache
1870 * doesn't get "out of sync" with the server.
1871 * XXX There should be a better way!
1873 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1875 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1876 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1878 tdnp = VTONFS(tdvp);
1879 mtx_lock(&tdnp->n_mtx);
1880 tdnp->n_flag |= NMODIFIED;
1881 mtx_unlock(&tdnp->n_mtx);
1883 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1885 VTONFS(vp)->n_attrstamp = 0;
1887 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1889 tdnp->n_attrstamp = 0;
1891 * If negative lookup caching is enabled, I might as well
1892 * add an entry for this node. Not necessary for correctness,
1893 * but if negative caching is enabled, then the system
1894 * must care about lookup caching hit rate, so...
1896 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1897 (cnp->cn_flags & MAKEENTRY))
1898 cache_enter(tdvp, vp, cnp);
1899 if (error && NFS_ISV4(vp))
1900 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1906 * nfs symbolic link create call
1909 nfs_symlink(struct vop_symlink_args *ap)
1911 struct vnode *dvp = ap->a_dvp;
1912 struct vattr *vap = ap->a_vap;
1913 struct componentname *cnp = ap->a_cnp;
1914 struct nfsvattr nfsva, dnfsva;
1916 struct nfsnode *np = NULL, *dnp;
1917 struct vnode *newvp = NULL;
1918 int error = 0, attrflag, dattrflag, ret;
1920 vap->va_type = VLNK;
1921 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1922 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1923 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1925 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
1926 &np, NULL, LK_EXCLUSIVE);
1932 if (newvp != NULL) {
1934 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1936 } else if (!error) {
1938 * If we do not have an error and we could not extract the
1939 * newvp from the response due to the request being NFSv2, we
1940 * have to do a lookup in order to obtain a newvp to return.
1942 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1943 cnp->cn_cred, cnp->cn_thread, &np);
1951 error = nfscl_maperr(cnp->cn_thread, error,
1952 vap->va_uid, vap->va_gid);
1955 * If negative lookup caching is enabled, I might as well
1956 * add an entry for this node. Not necessary for correctness,
1957 * but if negative caching is enabled, then the system
1958 * must care about lookup caching hit rate, so...
1960 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
1961 (cnp->cn_flags & MAKEENTRY))
1962 cache_enter(dvp, newvp, cnp);
1967 mtx_lock(&dnp->n_mtx);
1968 dnp->n_flag |= NMODIFIED;
1969 mtx_unlock(&dnp->n_mtx);
1971 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1973 dnp->n_attrstamp = 0;
1981 nfs_mkdir(struct vop_mkdir_args *ap)
1983 struct vnode *dvp = ap->a_dvp;
1984 struct vattr *vap = ap->a_vap;
1985 struct componentname *cnp = ap->a_cnp;
1986 struct nfsnode *np = NULL, *dnp;
1987 struct vnode *newvp = NULL;
1990 struct nfsvattr nfsva, dnfsva;
1991 int error = 0, attrflag, dattrflag, ret;
1993 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1995 vap->va_type = VDIR;
1996 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1997 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
1998 &attrflag, &dattrflag, NULL);
2000 mtx_lock(&dnp->n_mtx);
2001 dnp->n_flag |= NMODIFIED;
2002 mtx_unlock(&dnp->n_mtx);
2004 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2006 dnp->n_attrstamp = 0;
2008 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2009 &np, NULL, LK_EXCLUSIVE);
2013 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2018 if (!error && newvp == NULL) {
2019 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2020 cnp->cn_cred, cnp->cn_thread, &np);
2023 if (newvp->v_type != VDIR)
2031 error = nfscl_maperr(cnp->cn_thread, error,
2032 vap->va_uid, vap->va_gid);
2035 * If negative lookup caching is enabled, I might as well
2036 * add an entry for this node. Not necessary for correctness,
2037 * but if negative caching is enabled, then the system
2038 * must care about lookup caching hit rate, so...
2040 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2041 (cnp->cn_flags & MAKEENTRY))
2042 cache_enter(dvp, newvp, cnp);
2049 * nfs remove directory call
2052 nfs_rmdir(struct vop_rmdir_args *ap)
2054 struct vnode *vp = ap->a_vp;
2055 struct vnode *dvp = ap->a_dvp;
2056 struct componentname *cnp = ap->a_cnp;
2057 struct nfsnode *dnp;
2058 struct nfsvattr dnfsva;
2059 int error, dattrflag;
2063 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2064 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2066 mtx_lock(&dnp->n_mtx);
2067 dnp->n_flag |= NMODIFIED;
2068 mtx_unlock(&dnp->n_mtx);
2070 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2072 dnp->n_attrstamp = 0;
2076 if (error && NFS_ISV4(dvp))
2077 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2080 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2082 if (error == ENOENT)
2091 nfs_readdir(struct vop_readdir_args *ap)
2093 struct vnode *vp = ap->a_vp;
2094 struct nfsnode *np = VTONFS(vp);
2095 struct uio *uio = ap->a_uio;
2096 int tresid, error = 0;
2099 if (vp->v_type != VDIR)
2103 * First, check for hit on the EOF offset cache
2105 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2106 (np->n_flag & NMODIFIED) == 0) {
2107 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2108 mtx_lock(&np->n_mtx);
2109 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2110 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2111 mtx_unlock(&np->n_mtx);
2112 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2115 mtx_unlock(&np->n_mtx);
2120 * Call ncl_bioread() to do the real work.
2122 tresid = uio->uio_resid;
2123 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2125 if (!error && uio->uio_resid == tresid)
2126 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2132 * Called from below the buffer cache by ncl_doio().
2135 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2138 struct nfsvattr nfsva;
2139 nfsuint64 *cookiep, cookie;
2140 struct nfsnode *dnp = VTONFS(vp);
2141 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2142 int error = 0, eof, attrflag;
2144 KASSERT(uiop->uio_iovcnt == 1 &&
2145 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2146 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2147 ("nfs readdirrpc bad uio"));
2150 * If there is no cookie, assume directory was stale.
2152 ncl_dircookie_lock(dnp);
2153 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2156 ncl_dircookie_unlock(dnp);
2158 ncl_dircookie_unlock(dnp);
2159 return (NFSERR_BAD_COOKIE);
2162 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2163 (void)ncl_fsinfo(nmp, vp, cred, td);
2165 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2166 &attrflag, &eof, NULL);
2168 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2172 * We are now either at the end of the directory or have filled
2176 dnp->n_direofoffset = uiop->uio_offset;
2178 if (uiop->uio_resid > 0)
2179 ncl_printf("EEK! readdirrpc resid > 0\n");
2180 ncl_dircookie_lock(dnp);
2181 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2183 ncl_dircookie_unlock(dnp);
2185 } else if (NFS_ISV4(vp)) {
2186 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2192 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2195 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2198 struct nfsvattr nfsva;
2199 nfsuint64 *cookiep, cookie;
2200 struct nfsnode *dnp = VTONFS(vp);
2201 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2202 int error = 0, attrflag, eof;
2204 KASSERT(uiop->uio_iovcnt == 1 &&
2205 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2206 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2207 ("nfs readdirplusrpc bad uio"));
2210 * If there is no cookie, assume directory was stale.
2212 ncl_dircookie_lock(dnp);
2213 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2216 ncl_dircookie_unlock(dnp);
2218 ncl_dircookie_unlock(dnp);
2219 return (NFSERR_BAD_COOKIE);
2222 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2223 (void)ncl_fsinfo(nmp, vp, cred, td);
2224 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2225 &attrflag, &eof, NULL);
2227 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2231 * We are now either at end of the directory or have filled the
2235 dnp->n_direofoffset = uiop->uio_offset;
2237 if (uiop->uio_resid > 0)
2238 ncl_printf("EEK! readdirplusrpc resid > 0\n");
2239 ncl_dircookie_lock(dnp);
2240 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2242 ncl_dircookie_unlock(dnp);
2244 } else if (NFS_ISV4(vp)) {
2245 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2251 * Silly rename. To make the NFS filesystem that is stateless look a little
2252 * more like the "ufs" a remove of an active vnode is translated to a rename
2253 * to a funny looking filename that is removed by nfs_inactive on the
2254 * nfsnode. There is the potential for another process on a different client
2255 * to create the same funny name between the nfs_lookitup() fails and the
2256 * nfs_rename() completes, but...
2259 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2261 struct sillyrename *sp;
2265 unsigned int lticks;
2269 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2270 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2271 M_NEWNFSREQ, M_WAITOK);
2272 sp->s_cred = crhold(cnp->cn_cred);
2277 * Fudge together a funny name.
2278 * Changing the format of the funny name to accomodate more
2279 * sillynames per directory.
2280 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2281 * CPU ticks since boot.
2283 pid = cnp->cn_thread->td_proc->p_pid;
2284 lticks = (unsigned int)ticks;
2286 sp->s_namlen = sprintf(sp->s_name,
2287 ".nfs.%08x.%04x4.4", lticks,
2289 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2290 cnp->cn_thread, NULL))
2294 error = nfs_renameit(dvp, vp, cnp, sp);
2297 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2298 cnp->cn_thread, &np);
2299 np->n_sillyrename = sp;
2304 free((caddr_t)sp, M_NEWNFSREQ);
2309 * Look up a file name and optionally either update the file handle or
2310 * allocate an nfsnode, depending on the value of npp.
2311 * npp == NULL --> just do the lookup
2312 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2314 * *npp != NULL --> update the file handle in the vnode
2317 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2318 struct thread *td, struct nfsnode **npp)
2320 struct vnode *newvp = NULL, *vp;
2321 struct nfsnode *np, *dnp = VTONFS(dvp);
2322 struct nfsfh *nfhp, *onfhp;
2323 struct nfsvattr nfsva, dnfsva;
2324 struct componentname cn;
2325 int error = 0, attrflag, dattrflag;
2328 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2329 &nfhp, &attrflag, &dattrflag, NULL);
2331 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2332 if (npp && !error) {
2337 * For NFSv4, check to see if it is the same name and
2338 * replace the name, if it is different.
2340 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2341 (np->n_v4->n4_namelen != len ||
2342 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2343 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2344 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2345 dnp->n_fhp->nfh_len))) {
2347 { char nnn[100]; int nnnl;
2348 nnnl = (len < 100) ? len : 99;
2349 bcopy(name, nnn, nnnl);
2351 printf("replace=%s\n",nnn);
2354 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2355 MALLOC(np->n_v4, struct nfsv4node *,
2356 sizeof (struct nfsv4node) +
2357 dnp->n_fhp->nfh_len + len - 1,
2358 M_NFSV4NODE, M_WAITOK);
2359 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2360 np->n_v4->n4_namelen = len;
2361 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2362 dnp->n_fhp->nfh_len);
2363 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2365 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2369 * Rehash node for new file handle.
2371 vfs_hash_rehash(vp, hash);
2374 FREE((caddr_t)onfhp, M_NFSFH);
2376 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2377 FREE((caddr_t)nfhp, M_NFSFH);
2381 cn.cn_nameptr = name;
2382 cn.cn_namelen = len;
2383 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2384 &np, NULL, LK_EXCLUSIVE);
2389 if (!attrflag && *npp == NULL) {
2394 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2397 if (npp && *npp == NULL) {
2408 if (error && NFS_ISV4(dvp))
2409 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2414 * Nfs Version 3 and 4 commit rpc
2417 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2420 struct nfsvattr nfsva;
2421 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2422 int error, attrflag;
2423 u_char verf[NFSX_VERF];
2425 mtx_lock(&nmp->nm_mtx);
2426 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2427 mtx_unlock(&nmp->nm_mtx);
2430 mtx_unlock(&nmp->nm_mtx);
2431 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2434 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2435 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2436 error = NFSERR_STALEWRITEVERF;
2438 if (!error && attrflag)
2439 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2441 } else if (NFS_ISV4(vp)) {
2442 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2449 * For async requests when nfsiod(s) are running, queue the request by
2450 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2454 nfs_strategy(struct vop_strategy_args *ap)
2456 struct buf *bp = ap->a_bp;
2459 KASSERT(!(bp->b_flags & B_DONE),
2460 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2461 BUF_ASSERT_HELD(bp);
2463 if (bp->b_iocmd == BIO_READ)
2469 * If the op is asynchronous and an i/o daemon is waiting
2470 * queue the request, wake it up and wait for completion
2471 * otherwise just do it ourselves.
2473 if ((bp->b_flags & B_ASYNC) == 0 ||
2474 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2475 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2480 * fsync vnode op. Just call ncl_flush() with commit == 1.
2484 nfs_fsync(struct vop_fsync_args *ap)
2486 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2490 * Flush all the blocks associated with a vnode.
2491 * Walk through the buffer pool and push any dirty pages
2492 * associated with the vnode.
2493 * If the called_from_renewthread argument is TRUE, it has been called
2494 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2495 * waiting for a buffer write to complete.
2498 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2499 int commit, int called_from_renewthread)
2501 struct nfsnode *np = VTONFS(vp);
2505 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2506 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2507 int passone = 1, trycnt = 0;
2508 u_quad_t off, endoff, toff;
2509 struct ucred* wcred = NULL;
2510 struct buf **bvec = NULL;
2512 #ifndef NFS_COMMITBVECSIZ
2513 #define NFS_COMMITBVECSIZ 20
2515 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2516 int bvecsize = 0, bveccount;
2518 if (called_from_renewthread != 0)
2520 if (nmp->nm_flag & NFSMNT_INT)
2521 slpflag = NFS_PCATCH;
2526 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2527 * server, but has not been committed to stable storage on the server
2528 * yet. On the first pass, the byte range is worked out and the commit
2529 * rpc is done. On the second pass, ncl_writebp() is called to do the
2536 if (NFS_ISV34(vp) && commit) {
2537 if (bvec != NULL && bvec != bvec_on_stack)
2540 * Count up how many buffers waiting for a commit.
2544 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2545 if (!BUF_ISLOCKED(bp) &&
2546 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2547 == (B_DELWRI | B_NEEDCOMMIT))
2551 * Allocate space to remember the list of bufs to commit. It is
2552 * important to use M_NOWAIT here to avoid a race with nfs_write.
2553 * If we can't get memory (for whatever reason), we will end up
2554 * committing the buffers one-by-one in the loop below.
2556 if (bveccount > NFS_COMMITBVECSIZ) {
2558 * Release the vnode interlock to avoid a lock
2562 bvec = (struct buf **)
2563 malloc(bveccount * sizeof(struct buf *),
2567 bvec = bvec_on_stack;
2568 bvecsize = NFS_COMMITBVECSIZ;
2570 bvecsize = bveccount;
2572 bvec = bvec_on_stack;
2573 bvecsize = NFS_COMMITBVECSIZ;
2575 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2576 if (bvecpos >= bvecsize)
2578 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2579 nbp = TAILQ_NEXT(bp, b_bobufs);
2582 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2583 (B_DELWRI | B_NEEDCOMMIT)) {
2585 nbp = TAILQ_NEXT(bp, b_bobufs);
2591 * Work out if all buffers are using the same cred
2592 * so we can deal with them all with one commit.
2594 * NOTE: we are not clearing B_DONE here, so we have
2595 * to do it later on in this routine if we intend to
2596 * initiate I/O on the bp.
2598 * Note: to avoid loopback deadlocks, we do not
2599 * assign b_runningbufspace.
2602 wcred = bp->b_wcred;
2603 else if (wcred != bp->b_wcred)
2605 vfs_busy_pages(bp, 1);
2609 * bp is protected by being locked, but nbp is not
2610 * and vfs_busy_pages() may sleep. We have to
2613 nbp = TAILQ_NEXT(bp, b_bobufs);
2616 * A list of these buffers is kept so that the
2617 * second loop knows which buffers have actually
2618 * been committed. This is necessary, since there
2619 * may be a race between the commit rpc and new
2620 * uncommitted writes on the file.
2622 bvec[bvecpos++] = bp;
2623 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2627 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2635 * Commit data on the server, as required.
2636 * If all bufs are using the same wcred, then use that with
2637 * one call for all of them, otherwise commit each one
2640 if (wcred != NOCRED)
2641 retv = ncl_commit(vp, off, (int)(endoff - off),
2645 for (i = 0; i < bvecpos; i++) {
2648 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2650 size = (u_quad_t)(bp->b_dirtyend
2652 retv = ncl_commit(vp, off, (int)size,
2658 if (retv == NFSERR_STALEWRITEVERF)
2659 ncl_clearcommit(vp->v_mount);
2662 * Now, either mark the blocks I/O done or mark the
2663 * blocks dirty, depending on whether the commit
2666 for (i = 0; i < bvecpos; i++) {
2668 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2671 * Error, leave B_DELWRI intact
2673 vfs_unbusy_pages(bp);
2677 * Success, remove B_DELWRI ( bundirty() ).
2679 * b_dirtyoff/b_dirtyend seem to be NFS
2680 * specific. We should probably move that
2681 * into bundirty(). XXX
2684 bp->b_flags |= B_ASYNC;
2686 bp->b_flags &= ~B_DONE;
2687 bp->b_ioflags &= ~BIO_ERROR;
2688 bp->b_dirtyoff = bp->b_dirtyend = 0;
2695 * Start/do any write(s) that are required.
2699 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2700 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2701 if (waitfor != MNT_WAIT || passone)
2704 error = BUF_TIMELOCK(bp,
2705 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2706 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2711 if (error == ENOLCK) {
2715 if (called_from_renewthread != 0) {
2717 * Return EIO so the flush will be retried
2723 if (newnfs_sigintr(nmp, td)) {
2727 if (slpflag & PCATCH) {
2733 if ((bp->b_flags & B_DELWRI) == 0)
2734 panic("nfs_fsync: not dirty");
2735 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2741 if (passone || !commit)
2742 bp->b_flags |= B_ASYNC;
2744 bp->b_flags |= B_ASYNC;
2746 if (newnfs_sigintr(nmp, td)) {
2757 if (waitfor == MNT_WAIT) {
2758 while (bo->bo_numoutput) {
2759 error = bufobj_wwait(bo, slpflag, slptimeo);
2762 if (called_from_renewthread != 0) {
2764 * Return EIO so that the flush will be
2770 error = newnfs_sigintr(nmp, td);
2773 if (slpflag & PCATCH) {
2780 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2785 * Wait for all the async IO requests to drain
2788 mtx_lock(&np->n_mtx);
2789 while (np->n_directio_asyncwr > 0) {
2790 np->n_flag |= NFSYNCWAIT;
2791 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2792 &np->n_mtx, slpflag | (PRIBIO + 1),
2795 if (newnfs_sigintr(nmp, td)) {
2796 mtx_unlock(&np->n_mtx);
2802 mtx_unlock(&np->n_mtx);
2805 mtx_lock(&np->n_mtx);
2806 if (np->n_flag & NWRITEERR) {
2807 error = np->n_error;
2808 np->n_flag &= ~NWRITEERR;
2810 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2811 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2812 np->n_flag &= ~NMODIFIED;
2813 mtx_unlock(&np->n_mtx);
2815 if (bvec != NULL && bvec != bvec_on_stack)
2817 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2818 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2819 np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2820 /* try, try again... */
2825 printf("try%d\n", trycnt);
2832 * NFS advisory byte-level locks.
2835 nfs_advlock(struct vop_advlock_args *ap)
2837 struct vnode *vp = ap->a_vp;
2839 struct nfsnode *np = VTONFS(ap->a_vp);
2840 struct proc *p = (struct proc *)ap->a_id;
2841 struct thread *td = curthread; /* XXX */
2843 int ret, error = EOPNOTSUPP;
2846 if (NFS_ISV4(vp) && (ap->a_flags & F_POSIX)) {
2848 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2849 if (vp->v_iflag & VI_DOOMED) {
2855 * If this is unlocking a write locked region, flush and
2856 * commit them before unlocking. This is required by
2857 * RFC3530 Sec. 9.3.2.
2859 if (ap->a_op == F_UNLCK &&
2860 nfscl_checkwritelocked(vp, ap->a_fl, cred, td))
2861 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
2864 * Loop around doing the lock op, while a blocking lock
2865 * must wait for the lock op to succeed.
2868 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
2869 ap->a_fl, 0, cred, td);
2870 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2871 ap->a_op == F_SETLK) {
2873 error = nfs_catnap(PZERO | PCATCH, ret,
2877 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2878 if (vp->v_iflag & VI_DOOMED) {
2883 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2884 ap->a_op == F_SETLK);
2885 if (ret == NFSERR_DENIED) {
2888 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
2891 } else if (ret != 0) {
2897 * Now, if we just got a lock, invalidate data in the buffer
2898 * cache, as required, so that the coherency conforms with
2899 * RFC3530 Sec. 9.3.2.
2901 if (ap->a_op == F_SETLK) {
2902 if ((np->n_flag & NMODIFIED) == 0) {
2903 np->n_attrstamp = 0;
2904 ret = VOP_GETATTR(vp, &va, cred);
2906 if ((np->n_flag & NMODIFIED) || ret ||
2907 np->n_change != va.va_filerev) {
2908 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
2909 np->n_attrstamp = 0;
2910 ret = VOP_GETATTR(vp, &va, cred);
2912 np->n_mtime = va.va_mtime;
2913 np->n_change = va.va_filerev;
2919 } else if (!NFS_ISV4(vp)) {
2920 error = vn_lock(vp, LK_SHARED);
2923 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2924 size = VTONFS(vp)->n_size;
2926 error = lf_advlock(ap, &(vp->v_lockf), size);
2928 if (nfs_advlock_p != NULL)
2929 error = nfs_advlock_p(ap);
2940 * NFS advisory byte-level locks.
2943 nfs_advlockasync(struct vop_advlockasync_args *ap)
2945 struct vnode *vp = ap->a_vp;
2950 return (EOPNOTSUPP);
2951 error = vn_lock(vp, LK_SHARED);
2954 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2955 size = VTONFS(vp)->n_size;
2957 error = lf_advlockasync(ap, &(vp->v_lockf), size);
2966 * Print out the contents of an nfsnode.
2969 nfs_print(struct vop_print_args *ap)
2971 struct vnode *vp = ap->a_vp;
2972 struct nfsnode *np = VTONFS(vp);
2974 ncl_printf("\tfileid %ld fsid 0x%x",
2975 np->n_vattr.na_fileid, np->n_vattr.na_fsid);
2976 if (vp->v_type == VFIFO)
2983 * This is the "real" nfs::bwrite(struct buf*).
2984 * We set B_CACHE if this is a VMIO buffer.
2987 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
2990 int oldflags = bp->b_flags;
2996 BUF_ASSERT_HELD(bp);
2998 if (bp->b_flags & B_INVAL) {
3003 bp->b_flags |= B_CACHE;
3006 * Undirty the bp. We will redirty it later if the I/O fails.
3011 bp->b_flags &= ~B_DONE;
3012 bp->b_ioflags &= ~BIO_ERROR;
3013 bp->b_iocmd = BIO_WRITE;
3015 bufobj_wref(bp->b_bufobj);
3016 curthread->td_ru.ru_oublock++;
3020 * Note: to avoid loopback deadlocks, we do not
3021 * assign b_runningbufspace.
3023 vfs_busy_pages(bp, 1);
3026 bp->b_iooffset = dbtob(bp->b_blkno);
3029 if( (oldflags & B_ASYNC) == 0) {
3030 int rtval = bufwait(bp);
3032 if (oldflags & B_DELWRI) {
3045 * nfs special file access vnode op.
3046 * Essentially just get vattr and then imitate iaccess() since the device is
3047 * local to the client.
3050 nfsspec_access(struct vop_access_args *ap)
3053 struct ucred *cred = ap->a_cred;
3054 struct vnode *vp = ap->a_vp;
3055 accmode_t accmode = ap->a_accmode;
3060 * Disallow write attempts on filesystems mounted read-only;
3061 * unless the file is a socket, fifo, or a block or character
3062 * device resident on the filesystem.
3064 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3065 switch (vp->v_type) {
3075 error = VOP_GETATTR(vp, vap, cred);
3078 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3079 accmode, cred, NULL);
3085 * Read wrapper for fifos.
3088 nfsfifo_read(struct vop_read_args *ap)
3090 struct nfsnode *np = VTONFS(ap->a_vp);
3096 mtx_lock(&np->n_mtx);
3098 getnanotime(&np->n_atim);
3099 mtx_unlock(&np->n_mtx);
3100 error = fifo_specops.vop_read(ap);
3105 * Write wrapper for fifos.
3108 nfsfifo_write(struct vop_write_args *ap)
3110 struct nfsnode *np = VTONFS(ap->a_vp);
3115 mtx_lock(&np->n_mtx);
3117 getnanotime(&np->n_mtim);
3118 mtx_unlock(&np->n_mtx);
3119 return(fifo_specops.vop_write(ap));
3123 * Close wrapper for fifos.
3125 * Update the times on the nfsnode then do fifo close.
3128 nfsfifo_close(struct vop_close_args *ap)
3130 struct vnode *vp = ap->a_vp;
3131 struct nfsnode *np = VTONFS(vp);
3135 mtx_lock(&np->n_mtx);
3136 if (np->n_flag & (NACC | NUPD)) {
3138 if (np->n_flag & NACC)
3140 if (np->n_flag & NUPD)
3143 if (vrefcnt(vp) == 1 &&
3144 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3146 if (np->n_flag & NACC)
3147 vattr.va_atime = np->n_atim;
3148 if (np->n_flag & NUPD)
3149 vattr.va_mtime = np->n_mtim;
3150 mtx_unlock(&np->n_mtx);
3151 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3155 mtx_unlock(&np->n_mtx);
3157 return (fifo_specops.vop_close(ap));
3161 * Just call ncl_writebp() with the force argument set to 1.
3163 * NOTE: B_DONE may or may not be set in a_bp on call.
3166 nfs_bwrite(struct buf *bp)
3169 return (ncl_writebp(bp, 1, curthread));
3172 struct buf_ops buf_ops_newnfs = {
3173 .bop_name = "buf_ops_nfs",
3174 .bop_write = nfs_bwrite,
3175 .bop_strategy = bufstrategy,
3176 .bop_sync = bufsync,
3177 .bop_bdflush = bufbdflush,
3181 * Cloned from vop_stdlock(), and then the ugly hack added.
3184 nfs_lock1(struct vop_lock1_args *ap)
3186 struct vnode *vp = ap->a_vp;
3190 * Since vfs_hash_get() calls vget() and it will no longer work
3191 * for FreeBSD8 with flags == 0, I can only think of this horrible
3192 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3193 * and then handle it here. All I want for this case is a v_usecount
3194 * on the vnode to use for recovery, while another thread might
3195 * hold a lock on the vnode. I have the other threads blocked, so
3196 * there isn't any race problem.
3198 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3199 if ((ap->a_flags & LK_INTERLOCK) == 0)
3201 if ((vp->v_iflag & VI_DOOMED))
3206 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3207 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3212 nfs_getacl(struct vop_getacl_args *ap)
3216 if (ap->a_type != ACL_TYPE_NFS4)
3217 return (EOPNOTSUPP);
3218 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3220 if (error > NFSERR_STALE) {
3221 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3228 nfs_setacl(struct vop_setacl_args *ap)
3232 if (ap->a_type != ACL_TYPE_NFS4)
3233 return (EOPNOTSUPP);
3234 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3236 if (error > NFSERR_STALE) {
3237 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3244 * Return POSIX pathconf information applicable to nfs filesystems.
3247 nfs_pathconf(struct vop_pathconf_args *ap)
3249 struct nfsv3_pathconf pc;
3250 struct nfsvattr nfsva;
3251 struct vnode *vp = ap->a_vp;
3252 struct thread *td = curthread;
3253 int attrflag, error;
3255 if (NFS_ISV34(vp)) {
3256 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3259 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3264 /* For NFSv2, just fake them. */
3265 pc.pc_linkmax = LINK_MAX;
3266 pc.pc_namemax = NFS_MAXNAMLEN;
3268 pc.pc_chownrestricted = 1;
3269 pc.pc_caseinsensitive = 0;
3270 pc.pc_casepreserving = 1;
3273 switch (ap->a_name) {
3275 *ap->a_retval = pc.pc_linkmax;
3278 *ap->a_retval = pc.pc_namemax;
3281 *ap->a_retval = PATH_MAX;
3284 *ap->a_retval = PIPE_BUF;
3286 case _PC_CHOWN_RESTRICTED:
3287 *ap->a_retval = pc.pc_chownrestricted;
3290 *ap->a_retval = pc.pc_notrunc;
3292 case _PC_ACL_EXTENDED:
3296 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3297 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3302 case _PC_ACL_PATH_MAX:
3304 *ap->a_retval = ACL_MAX_ENTRIES;
3308 case _PC_MAC_PRESENT:
3312 /* _PC_ASYNC_IO should have been handled by upper layers. */
3313 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3322 case _PC_ALLOC_SIZE_MIN:
3323 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3325 case _PC_FILESIZEBITS:
3331 case _PC_REC_INCR_XFER_SIZE:
3332 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3334 case _PC_REC_MAX_XFER_SIZE:
3335 *ap->a_retval = -1; /* means ``unlimited'' */
3337 case _PC_REC_MIN_XFER_SIZE:
3338 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3340 case _PC_REC_XFER_ALIGN:
3341 *ap->a_retval = PAGE_SIZE;
3343 case _PC_SYMLINK_MAX:
3344 *ap->a_retval = NFS_MAXPATHLEN;