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
42 #include "opt_kdtrace.h"
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/systm.h>
48 #include <sys/resourcevar.h>
50 #include <sys/mount.h>
54 #include <sys/malloc.h>
56 #include <sys/namei.h>
57 #include <sys/socket.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/fcntl.h>
61 #include <sys/lockf.h>
63 #include <sys/sysctl.h>
64 #include <sys/signalvar.h>
67 #include <vm/vm_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>
74 #include <fs/nfsclient/nfs_kdtrace.h>
77 #include <netinet/in.h>
78 #include <netinet/in_var.h>
80 #include <nfs/nfs_lock.h>
83 #include <sys/dtrace_bsd.h>
85 dtrace_nfsclient_accesscache_flush_probe_func_t
86 dtrace_nfscl_accesscache_flush_done_probe;
87 uint32_t nfscl_accesscache_flush_done_id;
89 dtrace_nfsclient_accesscache_get_probe_func_t
90 dtrace_nfscl_accesscache_get_hit_probe,
91 dtrace_nfscl_accesscache_get_miss_probe;
92 uint32_t nfscl_accesscache_get_hit_id;
93 uint32_t nfscl_accesscache_get_miss_id;
95 dtrace_nfsclient_accesscache_load_probe_func_t
96 dtrace_nfscl_accesscache_load_done_probe;
97 uint32_t nfscl_accesscache_load_done_id;
98 #endif /* !KDTRACE_HOOKS */
104 extern struct nfsstats newnfsstats;
105 extern int nfsrv_useacl;
106 extern int nfscl_debuglevel;
107 MALLOC_DECLARE(M_NEWNFSREQ);
110 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
111 * calls are not in getblk() and brelse() so that they would not be necessary
115 #define vfs_busy_pages(bp, f)
118 static vop_read_t nfsfifo_read;
119 static vop_write_t nfsfifo_write;
120 static vop_close_t nfsfifo_close;
121 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
123 static vop_lookup_t nfs_lookup;
124 static vop_create_t nfs_create;
125 static vop_mknod_t nfs_mknod;
126 static vop_open_t nfs_open;
127 static vop_pathconf_t nfs_pathconf;
128 static vop_close_t nfs_close;
129 static vop_access_t nfs_access;
130 static vop_getattr_t nfs_getattr;
131 static vop_setattr_t nfs_setattr;
132 static vop_read_t nfs_read;
133 static vop_fsync_t nfs_fsync;
134 static vop_remove_t nfs_remove;
135 static vop_link_t nfs_link;
136 static vop_rename_t nfs_rename;
137 static vop_mkdir_t nfs_mkdir;
138 static vop_rmdir_t nfs_rmdir;
139 static vop_symlink_t nfs_symlink;
140 static vop_readdir_t nfs_readdir;
141 static vop_strategy_t nfs_strategy;
142 static int nfs_lookitup(struct vnode *, char *, int,
143 struct ucred *, struct thread *, struct nfsnode **);
144 static int nfs_sillyrename(struct vnode *, struct vnode *,
145 struct componentname *);
146 static vop_access_t nfsspec_access;
147 static vop_readlink_t nfs_readlink;
148 static vop_print_t nfs_print;
149 static vop_advlock_t nfs_advlock;
150 static vop_advlockasync_t nfs_advlockasync;
151 static vop_getacl_t nfs_getacl;
152 static vop_setacl_t nfs_setacl;
155 * Global vfs data structures for nfs
157 struct vop_vector newnfs_vnodeops = {
158 .vop_default = &default_vnodeops,
159 .vop_access = nfs_access,
160 .vop_advlock = nfs_advlock,
161 .vop_advlockasync = nfs_advlockasync,
162 .vop_close = nfs_close,
163 .vop_create = nfs_create,
164 .vop_fsync = nfs_fsync,
165 .vop_getattr = nfs_getattr,
166 .vop_getpages = ncl_getpages,
167 .vop_putpages = ncl_putpages,
168 .vop_inactive = ncl_inactive,
169 .vop_link = nfs_link,
170 .vop_lookup = nfs_lookup,
171 .vop_mkdir = nfs_mkdir,
172 .vop_mknod = nfs_mknod,
173 .vop_open = nfs_open,
174 .vop_pathconf = nfs_pathconf,
175 .vop_print = nfs_print,
176 .vop_read = nfs_read,
177 .vop_readdir = nfs_readdir,
178 .vop_readlink = nfs_readlink,
179 .vop_reclaim = ncl_reclaim,
180 .vop_remove = nfs_remove,
181 .vop_rename = nfs_rename,
182 .vop_rmdir = nfs_rmdir,
183 .vop_setattr = nfs_setattr,
184 .vop_strategy = nfs_strategy,
185 .vop_symlink = nfs_symlink,
186 .vop_write = ncl_write,
187 .vop_getacl = nfs_getacl,
188 .vop_setacl = nfs_setacl,
191 struct vop_vector newnfs_fifoops = {
192 .vop_default = &fifo_specops,
193 .vop_access = nfsspec_access,
194 .vop_close = nfsfifo_close,
195 .vop_fsync = nfs_fsync,
196 .vop_getattr = nfs_getattr,
197 .vop_inactive = ncl_inactive,
198 .vop_print = nfs_print,
199 .vop_read = nfsfifo_read,
200 .vop_reclaim = ncl_reclaim,
201 .vop_setattr = nfs_setattr,
202 .vop_write = nfsfifo_write,
205 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
206 struct componentname *cnp, struct vattr *vap);
207 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
208 int namelen, struct ucred *cred, struct thread *td);
209 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
210 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
211 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
212 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
213 struct componentname *scnp, struct sillyrename *sp);
218 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
220 SYSCTL_DECL(_vfs_nfs);
222 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
223 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
224 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
226 static int nfs_prime_access_cache = 0;
227 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
228 &nfs_prime_access_cache, 0,
229 "Prime NFS ACCESS cache when fetching attributes");
231 static int newnfs_commit_on_close = 0;
232 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
233 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
235 static int nfs_clean_pages_on_close = 1;
236 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
237 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
239 int newnfs_directio_enable = 0;
240 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
241 &newnfs_directio_enable, 0, "Enable NFS directio");
243 int nfs_keep_dirty_on_error;
244 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
245 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
248 * This sysctl allows other processes to mmap a file that has been opened
249 * O_DIRECT by a process. In general, having processes mmap the file while
250 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
251 * this by default to prevent DoS attacks - to prevent a malicious user from
252 * opening up files O_DIRECT preventing other users from mmap'ing these
253 * files. "Protected" environments where stricter consistency guarantees are
254 * required can disable this knob. The process that opened the file O_DIRECT
255 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
258 int newnfs_directio_allow_mmap = 1;
259 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
260 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
263 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
264 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
266 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
267 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
270 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
271 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
272 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
276 * The list of locks after the description of the lock is the ordering
277 * of other locks acquired with the lock held.
278 * np->n_mtx : Protects the fields in the nfsnode.
280 VI_MTX (acquired indirectly)
281 * nmp->nm_mtx : Protects the fields in the nfsmount.
283 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
284 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
287 * rep->r_mtx : Protects the fields in an nfsreq.
291 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
292 struct ucred *cred, u_int32_t *retmode)
294 int error = 0, attrflag, i, lrupos;
296 struct nfsnode *np = VTONFS(vp);
297 struct nfsvattr nfsva;
299 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
302 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
305 mtx_lock(&np->n_mtx);
306 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
307 if (np->n_accesscache[i].uid == cred->cr_uid) {
308 np->n_accesscache[i].mode = rmode;
309 np->n_accesscache[i].stamp = time_second;
312 if (i > 0 && np->n_accesscache[i].stamp <
313 np->n_accesscache[lrupos].stamp)
316 if (i == NFS_ACCESSCACHESIZE) {
317 np->n_accesscache[lrupos].uid = cred->cr_uid;
318 np->n_accesscache[lrupos].mode = rmode;
319 np->n_accesscache[lrupos].stamp = time_second;
321 mtx_unlock(&np->n_mtx);
324 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
325 } else if (NFS_ISV4(vp)) {
326 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
330 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
337 * nfs access vnode op.
338 * For nfs version 2, just return ok. File accesses may fail later.
339 * For nfs version 3, use the access rpc to check accessibility. If file modes
340 * are changed on the server, accesses might still fail later.
343 nfs_access(struct vop_access_args *ap)
345 struct vnode *vp = ap->a_vp;
346 int error = 0, i, gotahit;
347 u_int32_t mode, wmode, rmode;
348 int v34 = NFS_ISV34(vp);
349 struct nfsnode *np = VTONFS(vp);
352 * Disallow write attempts on filesystems mounted read-only;
353 * unless the file is a socket, fifo, or a block or character
354 * device resident on the filesystem.
356 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
357 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
358 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
359 switch (vp->v_type) {
369 * For nfs v3 or v4, check to see if we have done this recently, and if
370 * so return our cached result instead of making an ACCESS call.
371 * If not, do an access rpc, otherwise you are stuck emulating
372 * ufs_access() locally using the vattr. This may not be correct,
373 * since the server may apply other access criteria such as
374 * client uid-->server uid mapping that we do not know about.
377 if (ap->a_accmode & VREAD)
378 mode = NFSACCESS_READ;
381 if (vp->v_type != VDIR) {
382 if (ap->a_accmode & VWRITE)
383 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
384 if (ap->a_accmode & VAPPEND)
385 mode |= NFSACCESS_EXTEND;
386 if (ap->a_accmode & VEXEC)
387 mode |= NFSACCESS_EXECUTE;
388 if (ap->a_accmode & VDELETE)
389 mode |= NFSACCESS_DELETE;
391 if (ap->a_accmode & VWRITE)
392 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
393 if (ap->a_accmode & VAPPEND)
394 mode |= NFSACCESS_EXTEND;
395 if (ap->a_accmode & VEXEC)
396 mode |= NFSACCESS_LOOKUP;
397 if (ap->a_accmode & VDELETE)
398 mode |= NFSACCESS_DELETE;
399 if (ap->a_accmode & VDELETE_CHILD)
400 mode |= NFSACCESS_MODIFY;
402 /* XXX safety belt, only make blanket request if caching */
403 if (nfsaccess_cache_timeout > 0) {
404 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
405 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
406 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
412 * Does our cached result allow us to give a definite yes to
416 mtx_lock(&np->n_mtx);
417 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
418 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
419 if (time_second < (np->n_accesscache[i].stamp
420 + nfsaccess_cache_timeout) &&
421 (np->n_accesscache[i].mode & mode) == mode) {
422 NFSINCRGLOBAL(newnfsstats.accesscache_hits);
428 mtx_unlock(&np->n_mtx);
431 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
432 ap->a_cred->cr_uid, mode);
434 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
435 ap->a_cred->cr_uid, mode);
439 * Either a no, or a don't know. Go to the wire.
441 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
442 error = nfs34_access_otw(vp, wmode, ap->a_td,
445 (rmode & mode) != mode)
450 if ((error = nfsspec_access(ap)) != 0) {
454 * Attempt to prevent a mapped root from accessing a file
455 * which it shouldn't. We try to read a byte from the file
456 * if the user is root and the file is not zero length.
457 * After calling nfsspec_access, we should have the correct
460 mtx_lock(&np->n_mtx);
461 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
462 && VTONFS(vp)->n_size > 0) {
467 mtx_unlock(&np->n_mtx);
470 auio.uio_iov = &aiov;
474 auio.uio_segflg = UIO_SYSSPACE;
475 auio.uio_rw = UIO_READ;
476 auio.uio_td = ap->a_td;
478 if (vp->v_type == VREG)
479 error = ncl_readrpc(vp, &auio, ap->a_cred);
480 else if (vp->v_type == VDIR) {
482 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
484 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
485 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
488 } else if (vp->v_type == VLNK)
489 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
493 mtx_unlock(&np->n_mtx);
501 * Check to see if the type is ok
502 * and that deletion is not in progress.
503 * For paged in text files, you will need to flush the page cache
504 * if consistency is lost.
508 nfs_open(struct vop_open_args *ap)
510 struct vnode *vp = ap->a_vp;
511 struct nfsnode *np = VTONFS(vp);
514 int fmode = ap->a_mode;
517 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
521 * For NFSv4, we need to do the Open Op before cache validation,
522 * so that we conform to RFC3530 Sec. 9.3.1.
525 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
527 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
534 * Now, if this Open will be doing reading, re-validate/flush the
535 * cache, so that Close/Open coherency is maintained.
537 mtx_lock(&np->n_mtx);
538 if (np->n_flag & NMODIFIED) {
539 mtx_unlock(&np->n_mtx);
540 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
541 if (error == EINTR || error == EIO) {
543 (void) nfsrpc_close(vp, 0, ap->a_td);
546 mtx_lock(&np->n_mtx);
548 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
549 if (vp->v_type == VDIR)
550 np->n_direofoffset = 0;
551 mtx_unlock(&np->n_mtx);
552 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
555 (void) nfsrpc_close(vp, 0, ap->a_td);
558 mtx_lock(&np->n_mtx);
559 np->n_mtime = vattr.va_mtime;
561 np->n_change = vattr.va_filerev;
563 mtx_unlock(&np->n_mtx);
564 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
567 (void) nfsrpc_close(vp, 0, ap->a_td);
570 mtx_lock(&np->n_mtx);
571 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
572 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
573 if (vp->v_type == VDIR)
574 np->n_direofoffset = 0;
575 mtx_unlock(&np->n_mtx);
576 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
577 if (error == EINTR || error == EIO) {
579 (void) nfsrpc_close(vp, 0, ap->a_td);
582 mtx_lock(&np->n_mtx);
583 np->n_mtime = vattr.va_mtime;
585 np->n_change = vattr.va_filerev;
590 * If the object has >= 1 O_DIRECT active opens, we disable caching.
592 if (newnfs_directio_enable && (fmode & O_DIRECT) &&
593 (vp->v_type == VREG)) {
594 if (np->n_directio_opens == 0) {
595 mtx_unlock(&np->n_mtx);
596 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
599 (void) nfsrpc_close(vp, 0, ap->a_td);
602 mtx_lock(&np->n_mtx);
603 np->n_flag |= NNONCACHE;
605 np->n_directio_opens++;
608 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
609 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
610 np->n_flag |= NWRITEOPENED;
613 * If this is an open for writing, capture a reference to the
614 * credentials, so they can be used by ncl_putpages(). Using
615 * these write credentials is preferable to the credentials of
616 * whatever thread happens to be doing the VOP_PUTPAGES() since
617 * the write RPCs are less likely to fail with EACCES.
619 if ((fmode & FWRITE) != 0) {
620 cred = np->n_writecred;
621 np->n_writecred = crhold(ap->a_cred);
624 mtx_unlock(&np->n_mtx);
628 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
634 * What an NFS client should do upon close after writing is a debatable issue.
635 * Most NFS clients push delayed writes to the server upon close, basically for
637 * 1 - So that any write errors may be reported back to the client process
638 * doing the close system call. By far the two most likely errors are
639 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
640 * 2 - To put a worst case upper bound on cache inconsistency between
641 * multiple clients for the file.
642 * There is also a consistency problem for Version 2 of the protocol w.r.t.
643 * not being able to tell if other clients are writing a file concurrently,
644 * since there is no way of knowing if the changed modify time in the reply
645 * is only due to the write for this client.
646 * (NFS Version 3 provides weak cache consistency data in the reply that
647 * should be sufficient to detect and handle this case.)
649 * The current code does the following:
650 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
651 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
652 * or commit them (this satisfies 1 and 2 except for the
653 * case where the server crashes after this close but
654 * before the commit RPC, which is felt to be "good
655 * enough". Changing the last argument to ncl_flush() to
656 * a 1 would force a commit operation, if it is felt a
657 * commit is necessary now.
658 * for NFS Version 4 - flush the dirty buffers and commit them, if
659 * nfscl_mustflush() says this is necessary.
660 * It is necessary if there is no write delegation held,
661 * in order to satisfy open/close coherency.
662 * If the file isn't cached on local stable storage,
663 * it may be necessary in order to detect "out of space"
664 * errors from the server, if the write delegation
665 * issued by the server doesn't allow the file to grow.
669 nfs_close(struct vop_close_args *ap)
671 struct vnode *vp = ap->a_vp;
672 struct nfsnode *np = VTONFS(vp);
673 struct nfsvattr nfsva;
675 int error = 0, ret, localcred = 0;
676 int fmode = ap->a_fflag;
678 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
681 * During shutdown, a_cred isn't valid, so just use root.
683 if (ap->a_cred == NOCRED) {
684 cred = newnfs_getcred();
689 if (vp->v_type == VREG) {
691 * Examine and clean dirty pages, regardless of NMODIFIED.
692 * This closes a major hole in close-to-open consistency.
693 * We want to push out all dirty pages (and buffers) on
694 * close, regardless of whether they were dirtied by
695 * mmap'ed writes or via write().
697 if (nfs_clean_pages_on_close && vp->v_object) {
698 VM_OBJECT_WLOCK(vp->v_object);
699 vm_object_page_clean(vp->v_object, 0, 0, 0);
700 VM_OBJECT_WUNLOCK(vp->v_object);
702 mtx_lock(&np->n_mtx);
703 if (np->n_flag & NMODIFIED) {
704 mtx_unlock(&np->n_mtx);
707 * Under NFSv3 we have dirty buffers to dispose of. We
708 * must flush them to the NFS server. We have the option
709 * of waiting all the way through the commit rpc or just
710 * waiting for the initial write. The default is to only
711 * wait through the initial write so the data is in the
712 * server's cache, which is roughly similar to the state
713 * a standard disk subsystem leaves the file in on close().
715 * We cannot clear the NMODIFIED bit in np->n_flag due to
716 * potential races with other processes, and certainly
717 * cannot clear it if we don't commit.
718 * These races occur when there is no longer the old
719 * traditional vnode locking implemented for Vnode Ops.
721 int cm = newnfs_commit_on_close ? 1 : 0;
722 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
723 /* np->n_flag &= ~NMODIFIED; */
724 } else if (NFS_ISV4(vp)) {
725 if (nfscl_mustflush(vp) != 0) {
726 int cm = newnfs_commit_on_close ? 1 : 0;
727 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
730 * as above w.r.t races when clearing
732 * np->n_flag &= ~NMODIFIED;
736 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
737 mtx_lock(&np->n_mtx);
740 * Invalidate the attribute cache in all cases.
741 * An open is going to fetch fresh attrs any way, other procs
742 * on this node that have file open will be forced to do an
743 * otw attr fetch, but this is safe.
744 * --> A user found that their RPC count dropped by 20% when
745 * this was commented out and I can't see any requirement
746 * for it, so I've disabled it when negative lookups are
747 * enabled. (What does this have to do with negative lookup
748 * caching? Well nothing, except it was reported by the
749 * same user that needed negative lookup caching and I wanted
750 * there to be a way to disable it to see if it
751 * is the cause of some caching/coherency issue that might
754 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
756 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
758 if (np->n_flag & NWRITEERR) {
759 np->n_flag &= ~NWRITEERR;
762 mtx_unlock(&np->n_mtx);
767 * Get attributes so "change" is up to date.
769 if (error == 0 && nfscl_mustflush(vp) != 0 &&
770 vp->v_type == VREG &&
771 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
772 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
775 np->n_change = nfsva.na_filerev;
776 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
784 ret = nfsrpc_close(vp, 0, ap->a_td);
788 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
791 if (newnfs_directio_enable)
792 KASSERT((np->n_directio_asyncwr == 0),
793 ("nfs_close: dirty unflushed (%d) directio buffers\n",
794 np->n_directio_asyncwr));
795 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
796 mtx_lock(&np->n_mtx);
797 KASSERT((np->n_directio_opens > 0),
798 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
799 np->n_directio_opens--;
800 if (np->n_directio_opens == 0)
801 np->n_flag &= ~NNONCACHE;
802 mtx_unlock(&np->n_mtx);
810 * nfs getattr call from vfs.
813 nfs_getattr(struct vop_getattr_args *ap)
815 struct vnode *vp = ap->a_vp;
816 struct thread *td = curthread; /* XXX */
817 struct nfsnode *np = VTONFS(vp);
819 struct nfsvattr nfsva;
820 struct vattr *vap = ap->a_vap;
824 * Update local times for special files.
826 mtx_lock(&np->n_mtx);
827 if (np->n_flag & (NACC | NUPD))
829 mtx_unlock(&np->n_mtx);
831 * First look in the cache.
833 if (ncl_getattrcache(vp, &vattr) == 0) {
834 vap->va_type = vattr.va_type;
835 vap->va_mode = vattr.va_mode;
836 vap->va_nlink = vattr.va_nlink;
837 vap->va_uid = vattr.va_uid;
838 vap->va_gid = vattr.va_gid;
839 vap->va_fsid = vattr.va_fsid;
840 vap->va_fileid = vattr.va_fileid;
841 vap->va_size = vattr.va_size;
842 vap->va_blocksize = vattr.va_blocksize;
843 vap->va_atime = vattr.va_atime;
844 vap->va_mtime = vattr.va_mtime;
845 vap->va_ctime = vattr.va_ctime;
846 vap->va_gen = vattr.va_gen;
847 vap->va_flags = vattr.va_flags;
848 vap->va_rdev = vattr.va_rdev;
849 vap->va_bytes = vattr.va_bytes;
850 vap->va_filerev = vattr.va_filerev;
852 * Get the local modify time for the case of a write
855 nfscl_deleggetmodtime(vp, &vap->va_mtime);
859 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
860 nfsaccess_cache_timeout > 0) {
861 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
862 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
863 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
864 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
868 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
870 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
873 * Get the local modify time for the case of a write
876 nfscl_deleggetmodtime(vp, &vap->va_mtime);
877 } else if (NFS_ISV4(vp)) {
878 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
887 nfs_setattr(struct vop_setattr_args *ap)
889 struct vnode *vp = ap->a_vp;
890 struct nfsnode *np = VTONFS(vp);
891 struct thread *td = curthread; /* XXX */
892 struct vattr *vap = ap->a_vap;
901 * Setting of flags and marking of atimes are not supported.
903 if (vap->va_flags != VNOVAL)
907 * Disallow write attempts if the filesystem is mounted read-only.
909 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
910 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
911 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
912 (vp->v_mount->mnt_flag & MNT_RDONLY))
914 if (vap->va_size != VNOVAL) {
915 switch (vp->v_type) {
922 if (vap->va_mtime.tv_sec == VNOVAL &&
923 vap->va_atime.tv_sec == VNOVAL &&
924 vap->va_mode == (mode_t)VNOVAL &&
925 vap->va_uid == (uid_t)VNOVAL &&
926 vap->va_gid == (gid_t)VNOVAL)
928 vap->va_size = VNOVAL;
932 * Disallow write attempts if the filesystem is
935 if (vp->v_mount->mnt_flag & MNT_RDONLY)
938 * We run vnode_pager_setsize() early (why?),
939 * we must set np->n_size now to avoid vinvalbuf
940 * V_SAVE races that might setsize a lower
943 mtx_lock(&np->n_mtx);
945 mtx_unlock(&np->n_mtx);
946 error = ncl_meta_setsize(vp, ap->a_cred, td,
948 mtx_lock(&np->n_mtx);
949 if (np->n_flag & NMODIFIED) {
951 mtx_unlock(&np->n_mtx);
952 if (vap->va_size == 0)
953 error = ncl_vinvalbuf(vp, 0, td, 1);
955 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
957 vnode_pager_setsize(vp, tsize);
961 * Call nfscl_delegmodtime() to set the modify time
962 * locally, as required.
964 nfscl_delegmodtime(vp);
966 mtx_unlock(&np->n_mtx);
968 * np->n_size has already been set to vap->va_size
969 * in ncl_meta_setsize(). We must set it again since
970 * nfs_loadattrcache() could be called through
971 * ncl_meta_setsize() and could modify np->n_size.
973 mtx_lock(&np->n_mtx);
974 np->n_vattr.na_size = np->n_size = vap->va_size;
975 mtx_unlock(&np->n_mtx);
978 mtx_lock(&np->n_mtx);
979 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
980 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
981 mtx_unlock(&np->n_mtx);
982 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
983 (error == EINTR || error == EIO))
986 mtx_unlock(&np->n_mtx);
988 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
989 if (error && vap->va_size != VNOVAL) {
990 mtx_lock(&np->n_mtx);
991 np->n_size = np->n_vattr.na_size = tsize;
992 vnode_pager_setsize(vp, tsize);
993 mtx_unlock(&np->n_mtx);
999 * Do an nfs setattr rpc.
1002 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1005 struct nfsnode *np = VTONFS(vp);
1006 int error, ret, attrflag, i;
1007 struct nfsvattr nfsva;
1009 if (NFS_ISV34(vp)) {
1010 mtx_lock(&np->n_mtx);
1011 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1012 np->n_accesscache[i].stamp = 0;
1013 np->n_flag |= NDELEGMOD;
1014 mtx_unlock(&np->n_mtx);
1015 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1017 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1020 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1024 if (error && NFS_ISV4(vp))
1025 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1030 * nfs lookup call, one step at a time...
1031 * First look in cache
1032 * If not found, unlock the directory nfsnode and do the rpc
1035 nfs_lookup(struct vop_lookup_args *ap)
1037 struct componentname *cnp = ap->a_cnp;
1038 struct vnode *dvp = ap->a_dvp;
1039 struct vnode **vpp = ap->a_vpp;
1040 struct mount *mp = dvp->v_mount;
1041 int flags = cnp->cn_flags;
1042 struct vnode *newvp;
1043 struct nfsmount *nmp;
1044 struct nfsnode *np, *newnp;
1045 int error = 0, attrflag, dattrflag, ltype, ncticks;
1046 struct thread *td = cnp->cn_thread;
1048 struct nfsvattr dnfsva, nfsva;
1050 struct timespec nctime;
1053 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1054 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1056 if (dvp->v_type != VDIR)
1061 /* For NFSv4, wait until any remove is done. */
1062 mtx_lock(&np->n_mtx);
1063 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1064 np->n_flag |= NREMOVEWANT;
1065 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1067 mtx_unlock(&np->n_mtx);
1069 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1071 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1072 if (error > 0 && error != ENOENT)
1076 * Lookups of "." are special and always return the
1077 * current directory. cache_lookup() already handles
1078 * associated locking bookkeeping, etc.
1080 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1081 /* XXX: Is this really correct? */
1082 if (cnp->cn_nameiop != LOOKUP &&
1084 cnp->cn_flags |= SAVENAME;
1089 * We only accept a positive hit in the cache if the
1090 * change time of the file matches our cached copy.
1091 * Otherwise, we discard the cache entry and fallback
1092 * to doing a lookup RPC. We also only trust cache
1093 * entries for less than nm_nametimeo seconds.
1095 * To better handle stale file handles and attributes,
1096 * clear the attribute cache of this node if it is a
1097 * leaf component, part of an open() call, and not
1098 * locally modified before fetching the attributes.
1099 * This should allow stale file handles to be detected
1100 * here where we can fall back to a LOOKUP RPC to
1101 * recover rather than having nfs_open() detect the
1102 * stale file handle and failing open(2) with ESTALE.
1105 newnp = VTONFS(newvp);
1106 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1107 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1108 !(newnp->n_flag & NMODIFIED)) {
1109 mtx_lock(&newnp->n_mtx);
1110 newnp->n_attrstamp = 0;
1111 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1112 mtx_unlock(&newnp->n_mtx);
1114 if (nfscl_nodeleg(newvp, 0) == 0 ||
1115 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1116 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1117 timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1118 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1119 if (cnp->cn_nameiop != LOOKUP &&
1121 cnp->cn_flags |= SAVENAME;
1130 } else if (error == ENOENT) {
1131 if (dvp->v_iflag & VI_DOOMED)
1134 * We only accept a negative hit in the cache if the
1135 * modification time of the parent directory matches
1136 * the cached copy in the name cache entry.
1137 * Otherwise, we discard all of the negative cache
1138 * entries for this directory. We also only trust
1139 * negative cache entries for up to nm_negnametimeo
1142 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1143 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1144 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1145 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1148 cache_purge_negative(dvp);
1153 NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1154 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1155 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1158 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1160 if (newvp != NULLVP) {
1165 if (error != ENOENT) {
1167 error = nfscl_maperr(td, error, (uid_t)0,
1172 /* The requested file was not found. */
1173 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1174 (flags & ISLASTCN)) {
1176 * XXX: UFS does a full VOP_ACCESS(dvp,
1177 * VWRITE) here instead of just checking
1180 if (mp->mnt_flag & MNT_RDONLY)
1182 cnp->cn_flags |= SAVENAME;
1183 return (EJUSTRETURN);
1186 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1188 * Cache the modification time of the parent
1189 * directory from the post-op attributes in
1190 * the name cache entry. The negative cache
1191 * entry will be ignored once the directory
1192 * has changed. Don't bother adding the entry
1193 * if the directory has already changed.
1195 mtx_lock(&np->n_mtx);
1196 if (timespeccmp(&np->n_vattr.na_mtime,
1197 &dnfsva.na_mtime, ==)) {
1198 mtx_unlock(&np->n_mtx);
1199 cache_enter_time(dvp, NULL, cnp,
1200 &dnfsva.na_mtime, NULL);
1202 mtx_unlock(&np->n_mtx);
1208 * Handle RENAME case...
1210 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1211 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1212 FREE((caddr_t)nfhp, M_NFSFH);
1215 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1221 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1224 cnp->cn_flags |= SAVENAME;
1228 if (flags & ISDOTDOT) {
1229 ltype = NFSVOPISLOCKED(dvp);
1230 error = vfs_busy(mp, MBF_NOWAIT);
1233 NFSVOPUNLOCK(dvp, 0);
1234 error = vfs_busy(mp, 0);
1235 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1237 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1244 NFSVOPUNLOCK(dvp, 0);
1245 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1251 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1252 if (dvp->v_iflag & VI_DOOMED) {
1264 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1266 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1267 FREE((caddr_t)nfhp, M_NFSFH);
1271 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1274 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1280 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1282 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1283 !(np->n_flag & NMODIFIED)) {
1285 * Flush the attribute cache when opening a
1286 * leaf node to ensure that fresh attributes
1287 * are fetched in nfs_open() since we did not
1288 * fetch attributes from the LOOKUP reply.
1290 mtx_lock(&np->n_mtx);
1291 np->n_attrstamp = 0;
1292 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1293 mtx_unlock(&np->n_mtx);
1296 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1297 cnp->cn_flags |= SAVENAME;
1298 if ((cnp->cn_flags & MAKEENTRY) &&
1299 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1300 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1301 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1302 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1309 * Just call ncl_bioread() to do the work.
1312 nfs_read(struct vop_read_args *ap)
1314 struct vnode *vp = ap->a_vp;
1316 switch (vp->v_type) {
1318 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1322 return (EOPNOTSUPP);
1330 nfs_readlink(struct vop_readlink_args *ap)
1332 struct vnode *vp = ap->a_vp;
1334 if (vp->v_type != VLNK)
1336 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1340 * Do a readlink rpc.
1341 * Called by ncl_doio() from below the buffer cache.
1344 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1346 int error, ret, attrflag;
1347 struct nfsvattr nfsva;
1349 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1352 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1356 if (error && NFS_ISV4(vp))
1357 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1366 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1368 int error, ret, attrflag;
1369 struct nfsvattr nfsva;
1370 struct nfsmount *nmp;
1372 nmp = VFSTONFS(vnode_mount(vp));
1375 if (NFSHASPNFS(nmp))
1376 error = nfscl_doiods(vp, uiop, NULL, NULL,
1377 NFSV4OPEN_ACCESSREAD, cred, uiop->uio_td);
1378 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1380 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1383 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1387 if (error && NFS_ISV4(vp))
1388 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1396 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1397 int *iomode, int *must_commit, int called_from_strategy)
1399 struct nfsvattr nfsva;
1400 int error, attrflag, ret;
1401 struct nfsmount *nmp;
1403 nmp = VFSTONFS(vnode_mount(vp));
1406 if (NFSHASPNFS(nmp))
1407 error = nfscl_doiods(vp, uiop, iomode, must_commit,
1408 NFSV4OPEN_ACCESSWRITE, cred, uiop->uio_td);
1409 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1411 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1412 uiop->uio_td, &nfsva, &attrflag, NULL,
1413 called_from_strategy);
1415 if (VTONFS(vp)->n_flag & ND_NFSV4)
1416 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1419 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1425 *iomode = NFSWRITE_FILESYNC;
1426 if (error && NFS_ISV4(vp))
1427 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1433 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1434 * mode set to specify the file type and the size field for rdev.
1437 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1440 struct nfsvattr nfsva, dnfsva;
1441 struct vnode *newvp = NULL;
1442 struct nfsnode *np = NULL, *dnp;
1445 int error = 0, attrflag, dattrflag;
1448 if (vap->va_type == VCHR || vap->va_type == VBLK)
1449 rdev = vap->va_rdev;
1450 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1453 return (EOPNOTSUPP);
1454 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1456 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1457 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1458 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1461 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1462 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1463 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1466 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1467 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1470 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1473 if (attrflag != 0) {
1474 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1482 } else if (NFS_ISV4(dvp)) {
1483 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1487 mtx_lock(&dnp->n_mtx);
1488 dnp->n_flag |= NMODIFIED;
1490 dnp->n_attrstamp = 0;
1491 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1493 mtx_unlock(&dnp->n_mtx);
1499 * just call nfs_mknodrpc() to do the work.
1503 nfs_mknod(struct vop_mknod_args *ap)
1505 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1508 static struct mtx nfs_cverf_mtx;
1509 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1515 static nfsquad_t cverf;
1517 static int cverf_initialized = 0;
1519 mtx_lock(&nfs_cverf_mtx);
1520 if (cverf_initialized == 0) {
1521 cverf.lval[0] = arc4random();
1522 cverf.lval[1] = arc4random();
1523 cverf_initialized = 1;
1527 mtx_unlock(&nfs_cverf_mtx);
1533 * nfs file create call
1536 nfs_create(struct vop_create_args *ap)
1538 struct vnode *dvp = ap->a_dvp;
1539 struct vattr *vap = ap->a_vap;
1540 struct componentname *cnp = ap->a_cnp;
1541 struct nfsnode *np = NULL, *dnp;
1542 struct vnode *newvp = NULL;
1543 struct nfsmount *nmp;
1544 struct nfsvattr dnfsva, nfsva;
1547 int error = 0, attrflag, dattrflag, fmode = 0;
1551 * Oops, not for me..
1553 if (vap->va_type == VSOCK)
1554 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1556 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1558 if (vap->va_vaflags & VA_EXCLUSIVE)
1561 nmp = VFSTONFS(vnode_mount(dvp));
1563 /* For NFSv4, wait until any remove is done. */
1564 mtx_lock(&dnp->n_mtx);
1565 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1566 dnp->n_flag |= NREMOVEWANT;
1567 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1569 mtx_unlock(&dnp->n_mtx);
1571 cverf = nfs_get_cverf();
1572 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1573 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1574 &nfhp, &attrflag, &dattrflag, NULL);
1577 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1578 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1579 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1582 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1583 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1586 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1590 error = nfsrpc_getattr(newvp, cnp->cn_cred,
1591 cnp->cn_thread, &nfsva, NULL);
1593 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1597 if (newvp != NULL) {
1601 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1602 error == NFSERR_NOTSUPP) {
1606 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1607 if (nfscl_checksattr(vap, &nfsva)) {
1608 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1609 cnp->cn_thread, &nfsva, &attrflag, NULL);
1610 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1611 vap->va_gid != (gid_t)VNOVAL)) {
1612 /* try again without setting uid/gid */
1613 vap->va_uid = (uid_t)VNOVAL;
1614 vap->va_gid = (uid_t)VNOVAL;
1615 error = nfsrpc_setattr(newvp, vap, NULL,
1616 cnp->cn_cred, cnp->cn_thread, &nfsva,
1620 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1627 if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1628 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1631 } else if (NFS_ISV4(dvp)) {
1632 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1635 mtx_lock(&dnp->n_mtx);
1636 dnp->n_flag |= NMODIFIED;
1638 dnp->n_attrstamp = 0;
1639 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1641 mtx_unlock(&dnp->n_mtx);
1646 * nfs file remove call
1647 * To try and make nfs semantics closer to ufs semantics, a file that has
1648 * other processes using the vnode is renamed instead of removed and then
1649 * removed later on the last close.
1650 * - If v_usecount > 1
1651 * If a rename is not already in the works
1652 * call nfs_sillyrename() to set it up
1657 nfs_remove(struct vop_remove_args *ap)
1659 struct vnode *vp = ap->a_vp;
1660 struct vnode *dvp = ap->a_dvp;
1661 struct componentname *cnp = ap->a_cnp;
1662 struct nfsnode *np = VTONFS(vp);
1666 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1667 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1668 if (vp->v_type == VDIR)
1670 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1671 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1672 vattr.va_nlink > 1)) {
1674 * Purge the name cache so that the chance of a lookup for
1675 * the name succeeding while the remove is in progress is
1676 * minimized. Without node locking it can still happen, such
1677 * that an I/O op returns ESTALE, but since you get this if
1678 * another host removes the file..
1682 * throw away biocache buffers, mainly to avoid
1683 * unnecessary delayed writes later.
1685 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1687 if (error != EINTR && error != EIO)
1688 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1689 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1691 * Kludge City: If the first reply to the remove rpc is lost..
1692 * the reply to the retransmitted request will be ENOENT
1693 * since the file was in fact removed
1694 * Therefore, we cheat and return success.
1696 if (error == ENOENT)
1698 } else if (!np->n_sillyrename)
1699 error = nfs_sillyrename(dvp, vp, cnp);
1700 mtx_lock(&np->n_mtx);
1701 np->n_attrstamp = 0;
1702 mtx_unlock(&np->n_mtx);
1703 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1708 * nfs file remove rpc called from nfs_inactive
1711 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1714 * Make sure that the directory vnode is still valid.
1715 * XXX we should lock sp->s_dvp here.
1717 if (sp->s_dvp->v_type == VBAD)
1719 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1724 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1727 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1728 int namelen, struct ucred *cred, struct thread *td)
1730 struct nfsvattr dnfsva;
1731 struct nfsnode *dnp = VTONFS(dvp);
1732 int error = 0, dattrflag;
1734 mtx_lock(&dnp->n_mtx);
1735 dnp->n_flag |= NREMOVEINPROG;
1736 mtx_unlock(&dnp->n_mtx);
1737 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1739 mtx_lock(&dnp->n_mtx);
1740 if ((dnp->n_flag & NREMOVEWANT)) {
1741 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1742 mtx_unlock(&dnp->n_mtx);
1743 wakeup((caddr_t)dnp);
1745 dnp->n_flag &= ~NREMOVEINPROG;
1746 mtx_unlock(&dnp->n_mtx);
1749 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1750 mtx_lock(&dnp->n_mtx);
1751 dnp->n_flag |= NMODIFIED;
1753 dnp->n_attrstamp = 0;
1754 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1756 mtx_unlock(&dnp->n_mtx);
1757 if (error && NFS_ISV4(dvp))
1758 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1763 * nfs file rename call
1766 nfs_rename(struct vop_rename_args *ap)
1768 struct vnode *fvp = ap->a_fvp;
1769 struct vnode *tvp = ap->a_tvp;
1770 struct vnode *fdvp = ap->a_fdvp;
1771 struct vnode *tdvp = ap->a_tdvp;
1772 struct componentname *tcnp = ap->a_tcnp;
1773 struct componentname *fcnp = ap->a_fcnp;
1774 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1775 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1776 struct nfsv4node *newv4 = NULL;
1779 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1780 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1781 /* Check for cross-device rename */
1782 if ((fvp->v_mount != tdvp->v_mount) ||
1783 (tvp && (fvp->v_mount != tvp->v_mount))) {
1789 printf("nfs_rename: fvp == tvp (can't happen)\n");
1793 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1797 * We have to flush B_DELWRI data prior to renaming
1798 * the file. If we don't, the delayed-write buffers
1799 * can be flushed out later after the file has gone stale
1800 * under NFSV3. NFSV2 does not have this problem because
1801 * ( as far as I can tell ) it flushes dirty buffers more
1804 * Skip the rename operation if the fsync fails, this can happen
1805 * due to the server's volume being full, when we pushed out data
1806 * that was written back to our cache earlier. Not checking for
1807 * this condition can result in potential (silent) data loss.
1809 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1810 NFSVOPUNLOCK(fvp, 0);
1812 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1817 * If the tvp exists and is in use, sillyrename it before doing the
1818 * rename of the new file over it.
1819 * XXX Can't sillyrename a directory.
1821 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1822 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1827 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1828 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1831 if (error == 0 && NFS_ISV4(tdvp)) {
1833 * For NFSv4, check to see if it is the same name and
1834 * replace the name, if it is different.
1836 MALLOC(newv4, struct nfsv4node *,
1837 sizeof (struct nfsv4node) +
1838 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1839 M_NFSV4NODE, M_WAITOK);
1840 mtx_lock(&tdnp->n_mtx);
1841 mtx_lock(&fnp->n_mtx);
1842 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1843 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1844 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1845 tcnp->cn_namelen) ||
1846 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1847 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1848 tdnp->n_fhp->nfh_len))) {
1850 { char nnn[100]; int nnnl;
1851 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1852 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1854 printf("ren replace=%s\n",nnn);
1857 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1860 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1861 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1862 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1863 tdnp->n_fhp->nfh_len);
1864 NFSBCOPY(tcnp->cn_nameptr,
1865 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1867 mtx_unlock(&tdnp->n_mtx);
1868 mtx_unlock(&fnp->n_mtx);
1870 FREE((caddr_t)newv4, M_NFSV4NODE);
1873 if (fvp->v_type == VDIR) {
1874 if (tvp != NULL && tvp->v_type == VDIR)
1889 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1891 if (error == ENOENT)
1897 * nfs file rename rpc called from nfs_remove() above
1900 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1901 struct sillyrename *sp)
1904 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1905 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1910 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1913 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1914 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1915 int tnamelen, struct ucred *cred, struct thread *td)
1917 struct nfsvattr fnfsva, tnfsva;
1918 struct nfsnode *fdnp = VTONFS(fdvp);
1919 struct nfsnode *tdnp = VTONFS(tdvp);
1920 int error = 0, fattrflag, tattrflag;
1922 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1923 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1924 &tattrflag, NULL, NULL);
1925 mtx_lock(&fdnp->n_mtx);
1926 fdnp->n_flag |= NMODIFIED;
1927 if (fattrflag != 0) {
1928 mtx_unlock(&fdnp->n_mtx);
1929 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1931 fdnp->n_attrstamp = 0;
1932 mtx_unlock(&fdnp->n_mtx);
1933 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1935 mtx_lock(&tdnp->n_mtx);
1936 tdnp->n_flag |= NMODIFIED;
1937 if (tattrflag != 0) {
1938 mtx_unlock(&tdnp->n_mtx);
1939 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1941 tdnp->n_attrstamp = 0;
1942 mtx_unlock(&tdnp->n_mtx);
1943 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1945 if (error && NFS_ISV4(fdvp))
1946 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1951 * nfs hard link create call
1954 nfs_link(struct vop_link_args *ap)
1956 struct vnode *vp = ap->a_vp;
1957 struct vnode *tdvp = ap->a_tdvp;
1958 struct componentname *cnp = ap->a_cnp;
1959 struct nfsnode *np, *tdnp;
1960 struct nfsvattr nfsva, dnfsva;
1961 int error = 0, attrflag, dattrflag;
1964 * Push all writes to the server, so that the attribute cache
1965 * doesn't get "out of sync" with the server.
1966 * XXX There should be a better way!
1968 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1970 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1971 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1973 tdnp = VTONFS(tdvp);
1974 mtx_lock(&tdnp->n_mtx);
1975 tdnp->n_flag |= NMODIFIED;
1976 if (dattrflag != 0) {
1977 mtx_unlock(&tdnp->n_mtx);
1978 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1980 tdnp->n_attrstamp = 0;
1981 mtx_unlock(&tdnp->n_mtx);
1982 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1985 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1988 mtx_lock(&np->n_mtx);
1989 np->n_attrstamp = 0;
1990 mtx_unlock(&np->n_mtx);
1991 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1994 * If negative lookup caching is enabled, I might as well
1995 * add an entry for this node. Not necessary for correctness,
1996 * but if negative caching is enabled, then the system
1997 * must care about lookup caching hit rate, so...
1999 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2000 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2001 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2003 if (error && NFS_ISV4(vp))
2004 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2010 * nfs symbolic link create call
2013 nfs_symlink(struct vop_symlink_args *ap)
2015 struct vnode *dvp = ap->a_dvp;
2016 struct vattr *vap = ap->a_vap;
2017 struct componentname *cnp = ap->a_cnp;
2018 struct nfsvattr nfsva, dnfsva;
2020 struct nfsnode *np = NULL, *dnp;
2021 struct vnode *newvp = NULL;
2022 int error = 0, attrflag, dattrflag, ret;
2024 vap->va_type = VLNK;
2025 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2026 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2027 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2029 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2030 &np, NULL, LK_EXCLUSIVE);
2036 if (newvp != NULL) {
2038 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2040 } else if (!error) {
2042 * If we do not have an error and we could not extract the
2043 * newvp from the response due to the request being NFSv2, we
2044 * have to do a lookup in order to obtain a newvp to return.
2046 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2047 cnp->cn_cred, cnp->cn_thread, &np);
2055 error = nfscl_maperr(cnp->cn_thread, error,
2056 vap->va_uid, vap->va_gid);
2062 mtx_lock(&dnp->n_mtx);
2063 dnp->n_flag |= NMODIFIED;
2064 if (dattrflag != 0) {
2065 mtx_unlock(&dnp->n_mtx);
2066 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2068 dnp->n_attrstamp = 0;
2069 mtx_unlock(&dnp->n_mtx);
2070 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2073 * If negative lookup caching is enabled, I might as well
2074 * add an entry for this node. Not necessary for correctness,
2075 * but if negative caching is enabled, then the system
2076 * must care about lookup caching hit rate, so...
2078 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2079 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2080 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2089 nfs_mkdir(struct vop_mkdir_args *ap)
2091 struct vnode *dvp = ap->a_dvp;
2092 struct vattr *vap = ap->a_vap;
2093 struct componentname *cnp = ap->a_cnp;
2094 struct nfsnode *np = NULL, *dnp;
2095 struct vnode *newvp = NULL;
2098 struct nfsvattr nfsva, dnfsva;
2099 int error = 0, attrflag, dattrflag, ret;
2101 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2103 vap->va_type = VDIR;
2104 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2105 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2106 &attrflag, &dattrflag, NULL);
2108 mtx_lock(&dnp->n_mtx);
2109 dnp->n_flag |= NMODIFIED;
2110 if (dattrflag != 0) {
2111 mtx_unlock(&dnp->n_mtx);
2112 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2114 dnp->n_attrstamp = 0;
2115 mtx_unlock(&dnp->n_mtx);
2116 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2119 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2120 &np, NULL, LK_EXCLUSIVE);
2124 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2129 if (!error && newvp == NULL) {
2130 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2131 cnp->cn_cred, cnp->cn_thread, &np);
2134 if (newvp->v_type != VDIR)
2142 error = nfscl_maperr(cnp->cn_thread, error,
2143 vap->va_uid, vap->va_gid);
2146 * If negative lookup caching is enabled, I might as well
2147 * add an entry for this node. Not necessary for correctness,
2148 * but if negative caching is enabled, then the system
2149 * must care about lookup caching hit rate, so...
2151 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2152 (cnp->cn_flags & MAKEENTRY) &&
2153 attrflag != 0 && dattrflag != 0)
2154 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2162 * nfs remove directory call
2165 nfs_rmdir(struct vop_rmdir_args *ap)
2167 struct vnode *vp = ap->a_vp;
2168 struct vnode *dvp = ap->a_dvp;
2169 struct componentname *cnp = ap->a_cnp;
2170 struct nfsnode *dnp;
2171 struct nfsvattr dnfsva;
2172 int error, dattrflag;
2176 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2177 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2179 mtx_lock(&dnp->n_mtx);
2180 dnp->n_flag |= NMODIFIED;
2181 if (dattrflag != 0) {
2182 mtx_unlock(&dnp->n_mtx);
2183 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2185 dnp->n_attrstamp = 0;
2186 mtx_unlock(&dnp->n_mtx);
2187 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2192 if (error && NFS_ISV4(dvp))
2193 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2196 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2198 if (error == ENOENT)
2207 nfs_readdir(struct vop_readdir_args *ap)
2209 struct vnode *vp = ap->a_vp;
2210 struct nfsnode *np = VTONFS(vp);
2211 struct uio *uio = ap->a_uio;
2212 ssize_t tresid, left;
2216 if (ap->a_eofflag != NULL)
2218 if (vp->v_type != VDIR)
2222 * First, check for hit on the EOF offset cache
2224 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2225 (np->n_flag & NMODIFIED) == 0) {
2226 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2227 mtx_lock(&np->n_mtx);
2228 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2229 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2230 mtx_unlock(&np->n_mtx);
2231 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2232 if (ap->a_eofflag != NULL)
2236 mtx_unlock(&np->n_mtx);
2241 * NFS always guarantees that directory entries don't straddle
2242 * DIRBLKSIZ boundaries. As such, we need to limit the size
2243 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2246 left = uio->uio_resid % DIRBLKSIZ;
2247 if (left == uio->uio_resid)
2249 uio->uio_resid -= left;
2252 * Call ncl_bioread() to do the real work.
2254 tresid = uio->uio_resid;
2255 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2257 if (!error && uio->uio_resid == tresid) {
2258 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2259 if (ap->a_eofflag != NULL)
2263 /* Add the partial DIRBLKSIZ (left) back in. */
2264 uio->uio_resid += left;
2270 * Called from below the buffer cache by ncl_doio().
2273 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2276 struct nfsvattr nfsva;
2277 nfsuint64 *cookiep, cookie;
2278 struct nfsnode *dnp = VTONFS(vp);
2279 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2280 int error = 0, eof, attrflag;
2282 KASSERT(uiop->uio_iovcnt == 1 &&
2283 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2284 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2285 ("nfs readdirrpc bad uio"));
2288 * If there is no cookie, assume directory was stale.
2290 ncl_dircookie_lock(dnp);
2291 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2294 ncl_dircookie_unlock(dnp);
2296 ncl_dircookie_unlock(dnp);
2297 return (NFSERR_BAD_COOKIE);
2300 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2301 (void)ncl_fsinfo(nmp, vp, cred, td);
2303 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2304 &attrflag, &eof, NULL);
2306 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2310 * We are now either at the end of the directory or have filled
2314 dnp->n_direofoffset = uiop->uio_offset;
2316 if (uiop->uio_resid > 0)
2317 printf("EEK! readdirrpc resid > 0\n");
2318 ncl_dircookie_lock(dnp);
2319 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2321 ncl_dircookie_unlock(dnp);
2323 } else if (NFS_ISV4(vp)) {
2324 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2330 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2333 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2336 struct nfsvattr nfsva;
2337 nfsuint64 *cookiep, cookie;
2338 struct nfsnode *dnp = VTONFS(vp);
2339 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2340 int error = 0, attrflag, eof;
2342 KASSERT(uiop->uio_iovcnt == 1 &&
2343 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2344 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2345 ("nfs readdirplusrpc bad uio"));
2348 * If there is no cookie, assume directory was stale.
2350 ncl_dircookie_lock(dnp);
2351 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2354 ncl_dircookie_unlock(dnp);
2356 ncl_dircookie_unlock(dnp);
2357 return (NFSERR_BAD_COOKIE);
2360 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2361 (void)ncl_fsinfo(nmp, vp, cred, td);
2362 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2363 &attrflag, &eof, NULL);
2365 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2369 * We are now either at end of the directory or have filled the
2373 dnp->n_direofoffset = uiop->uio_offset;
2375 if (uiop->uio_resid > 0)
2376 printf("EEK! readdirplusrpc resid > 0\n");
2377 ncl_dircookie_lock(dnp);
2378 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2380 ncl_dircookie_unlock(dnp);
2382 } else if (NFS_ISV4(vp)) {
2383 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2389 * Silly rename. To make the NFS filesystem that is stateless look a little
2390 * more like the "ufs" a remove of an active vnode is translated to a rename
2391 * to a funny looking filename that is removed by nfs_inactive on the
2392 * nfsnode. There is the potential for another process on a different client
2393 * to create the same funny name between the nfs_lookitup() fails and the
2394 * nfs_rename() completes, but...
2397 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2399 struct sillyrename *sp;
2403 unsigned int lticks;
2407 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2408 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2409 M_NEWNFSREQ, M_WAITOK);
2410 sp->s_cred = crhold(cnp->cn_cred);
2415 * Fudge together a funny name.
2416 * Changing the format of the funny name to accomodate more
2417 * sillynames per directory.
2418 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2419 * CPU ticks since boot.
2421 pid = cnp->cn_thread->td_proc->p_pid;
2422 lticks = (unsigned int)ticks;
2424 sp->s_namlen = sprintf(sp->s_name,
2425 ".nfs.%08x.%04x4.4", lticks,
2427 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2428 cnp->cn_thread, NULL))
2432 error = nfs_renameit(dvp, vp, cnp, sp);
2435 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2436 cnp->cn_thread, &np);
2437 np->n_sillyrename = sp;
2442 free((caddr_t)sp, M_NEWNFSREQ);
2447 * Look up a file name and optionally either update the file handle or
2448 * allocate an nfsnode, depending on the value of npp.
2449 * npp == NULL --> just do the lookup
2450 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2452 * *npp != NULL --> update the file handle in the vnode
2455 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2456 struct thread *td, struct nfsnode **npp)
2458 struct vnode *newvp = NULL, *vp;
2459 struct nfsnode *np, *dnp = VTONFS(dvp);
2460 struct nfsfh *nfhp, *onfhp;
2461 struct nfsvattr nfsva, dnfsva;
2462 struct componentname cn;
2463 int error = 0, attrflag, dattrflag;
2466 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2467 &nfhp, &attrflag, &dattrflag, NULL);
2469 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2470 if (npp && !error) {
2475 * For NFSv4, check to see if it is the same name and
2476 * replace the name, if it is different.
2478 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2479 (np->n_v4->n4_namelen != len ||
2480 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2481 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2482 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2483 dnp->n_fhp->nfh_len))) {
2485 { char nnn[100]; int nnnl;
2486 nnnl = (len < 100) ? len : 99;
2487 bcopy(name, nnn, nnnl);
2489 printf("replace=%s\n",nnn);
2492 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2493 MALLOC(np->n_v4, struct nfsv4node *,
2494 sizeof (struct nfsv4node) +
2495 dnp->n_fhp->nfh_len + len - 1,
2496 M_NFSV4NODE, M_WAITOK);
2497 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2498 np->n_v4->n4_namelen = len;
2499 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2500 dnp->n_fhp->nfh_len);
2501 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2503 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2507 * Rehash node for new file handle.
2509 vfs_hash_rehash(vp, hash);
2512 FREE((caddr_t)onfhp, M_NFSFH);
2514 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2515 FREE((caddr_t)nfhp, M_NFSFH);
2519 cn.cn_nameptr = name;
2520 cn.cn_namelen = len;
2521 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2522 &np, NULL, LK_EXCLUSIVE);
2527 if (!attrflag && *npp == NULL) {
2535 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2538 if (npp && *npp == NULL) {
2549 if (error && NFS_ISV4(dvp))
2550 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2555 * Nfs Version 3 and 4 commit rpc
2558 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2561 struct nfsvattr nfsva;
2562 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2563 int error, attrflag;
2565 mtx_lock(&nmp->nm_mtx);
2566 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2567 mtx_unlock(&nmp->nm_mtx);
2570 mtx_unlock(&nmp->nm_mtx);
2571 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2574 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2576 if (error != 0 && NFS_ISV4(vp))
2577 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2583 * For async requests when nfsiod(s) are running, queue the request by
2584 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2588 nfs_strategy(struct vop_strategy_args *ap)
2590 struct buf *bp = ap->a_bp;
2593 KASSERT(!(bp->b_flags & B_DONE),
2594 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2595 BUF_ASSERT_HELD(bp);
2597 if (bp->b_iocmd == BIO_READ)
2603 * If the op is asynchronous and an i/o daemon is waiting
2604 * queue the request, wake it up and wait for completion
2605 * otherwise just do it ourselves.
2607 if ((bp->b_flags & B_ASYNC) == 0 ||
2608 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2609 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2614 * fsync vnode op. Just call ncl_flush() with commit == 1.
2618 nfs_fsync(struct vop_fsync_args *ap)
2621 if (ap->a_vp->v_type != VREG) {
2623 * For NFS, metadata is changed synchronously on the server,
2624 * so there is nothing to flush. Also, ncl_flush() clears
2625 * the NMODIFIED flag and that shouldn't be done here for
2630 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2634 * Flush all the blocks associated with a vnode.
2635 * Walk through the buffer pool and push any dirty pages
2636 * associated with the vnode.
2637 * If the called_from_renewthread argument is TRUE, it has been called
2638 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2639 * waiting for a buffer write to complete.
2642 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2643 int commit, int called_from_renewthread)
2645 struct nfsnode *np = VTONFS(vp);
2649 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2650 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2651 int passone = 1, trycnt = 0;
2652 u_quad_t off, endoff, toff;
2653 struct ucred* wcred = NULL;
2654 struct buf **bvec = NULL;
2656 #ifndef NFS_COMMITBVECSIZ
2657 #define NFS_COMMITBVECSIZ 20
2659 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2660 int bvecsize = 0, bveccount;
2662 if (called_from_renewthread != 0)
2664 if (nmp->nm_flag & NFSMNT_INT)
2670 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2671 * server, but has not been committed to stable storage on the server
2672 * yet. On the first pass, the byte range is worked out and the commit
2673 * rpc is done. On the second pass, ncl_writebp() is called to do the
2680 if (NFS_ISV34(vp) && commit) {
2681 if (bvec != NULL && bvec != bvec_on_stack)
2684 * Count up how many buffers waiting for a commit.
2688 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2689 if (!BUF_ISLOCKED(bp) &&
2690 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2691 == (B_DELWRI | B_NEEDCOMMIT))
2695 * Allocate space to remember the list of bufs to commit. It is
2696 * important to use M_NOWAIT here to avoid a race with nfs_write.
2697 * If we can't get memory (for whatever reason), we will end up
2698 * committing the buffers one-by-one in the loop below.
2700 if (bveccount > NFS_COMMITBVECSIZ) {
2702 * Release the vnode interlock to avoid a lock
2706 bvec = (struct buf **)
2707 malloc(bveccount * sizeof(struct buf *),
2711 bvec = bvec_on_stack;
2712 bvecsize = NFS_COMMITBVECSIZ;
2714 bvecsize = bveccount;
2716 bvec = bvec_on_stack;
2717 bvecsize = NFS_COMMITBVECSIZ;
2719 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2720 if (bvecpos >= bvecsize)
2722 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2723 nbp = TAILQ_NEXT(bp, b_bobufs);
2726 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2727 (B_DELWRI | B_NEEDCOMMIT)) {
2729 nbp = TAILQ_NEXT(bp, b_bobufs);
2735 * Work out if all buffers are using the same cred
2736 * so we can deal with them all with one commit.
2738 * NOTE: we are not clearing B_DONE here, so we have
2739 * to do it later on in this routine if we intend to
2740 * initiate I/O on the bp.
2742 * Note: to avoid loopback deadlocks, we do not
2743 * assign b_runningbufspace.
2746 wcred = bp->b_wcred;
2747 else if (wcred != bp->b_wcred)
2749 vfs_busy_pages(bp, 1);
2753 * bp is protected by being locked, but nbp is not
2754 * and vfs_busy_pages() may sleep. We have to
2757 nbp = TAILQ_NEXT(bp, b_bobufs);
2760 * A list of these buffers is kept so that the
2761 * second loop knows which buffers have actually
2762 * been committed. This is necessary, since there
2763 * may be a race between the commit rpc and new
2764 * uncommitted writes on the file.
2766 bvec[bvecpos++] = bp;
2767 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2771 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2779 * Commit data on the server, as required.
2780 * If all bufs are using the same wcred, then use that with
2781 * one call for all of them, otherwise commit each one
2784 if (wcred != NOCRED)
2785 retv = ncl_commit(vp, off, (int)(endoff - off),
2789 for (i = 0; i < bvecpos; i++) {
2792 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2794 size = (u_quad_t)(bp->b_dirtyend
2796 retv = ncl_commit(vp, off, (int)size,
2802 if (retv == NFSERR_STALEWRITEVERF)
2803 ncl_clearcommit(vp->v_mount);
2806 * Now, either mark the blocks I/O done or mark the
2807 * blocks dirty, depending on whether the commit
2810 for (i = 0; i < bvecpos; i++) {
2812 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2815 * Error, leave B_DELWRI intact
2817 vfs_unbusy_pages(bp);
2821 * Success, remove B_DELWRI ( bundirty() ).
2823 * b_dirtyoff/b_dirtyend seem to be NFS
2824 * specific. We should probably move that
2825 * into bundirty(). XXX
2828 bp->b_flags |= B_ASYNC;
2830 bp->b_flags &= ~B_DONE;
2831 bp->b_ioflags &= ~BIO_ERROR;
2832 bp->b_dirtyoff = bp->b_dirtyend = 0;
2839 * Start/do any write(s) that are required.
2843 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2844 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2845 if (waitfor != MNT_WAIT || passone)
2848 error = BUF_TIMELOCK(bp,
2849 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2850 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
2855 if (error == ENOLCK) {
2859 if (called_from_renewthread != 0) {
2861 * Return EIO so the flush will be retried
2867 if (newnfs_sigintr(nmp, td)) {
2871 if (slpflag == PCATCH) {
2877 if ((bp->b_flags & B_DELWRI) == 0)
2878 panic("nfs_fsync: not dirty");
2879 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2885 if (passone || !commit)
2886 bp->b_flags |= B_ASYNC;
2888 bp->b_flags |= B_ASYNC;
2890 if (newnfs_sigintr(nmp, td)) {
2901 if (waitfor == MNT_WAIT) {
2902 while (bo->bo_numoutput) {
2903 error = bufobj_wwait(bo, slpflag, slptimeo);
2906 if (called_from_renewthread != 0) {
2908 * Return EIO so that the flush will be
2914 error = newnfs_sigintr(nmp, td);
2917 if (slpflag == PCATCH) {
2924 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2929 * Wait for all the async IO requests to drain
2932 mtx_lock(&np->n_mtx);
2933 while (np->n_directio_asyncwr > 0) {
2934 np->n_flag |= NFSYNCWAIT;
2935 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2936 &np->n_mtx, slpflag | (PRIBIO + 1),
2939 if (newnfs_sigintr(nmp, td)) {
2940 mtx_unlock(&np->n_mtx);
2946 mtx_unlock(&np->n_mtx);
2949 if (NFSHASPNFS(nmp)) {
2950 nfscl_layoutcommit(vp, td);
2952 * Invalidate the attribute cache, since writes to a DS
2953 * won't update the size attribute.
2955 mtx_lock(&np->n_mtx);
2956 np->n_attrstamp = 0;
2958 mtx_lock(&np->n_mtx);
2959 if (np->n_flag & NWRITEERR) {
2960 error = np->n_error;
2961 np->n_flag &= ~NWRITEERR;
2963 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2964 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2965 np->n_flag &= ~NMODIFIED;
2966 mtx_unlock(&np->n_mtx);
2968 if (bvec != NULL && bvec != bvec_on_stack)
2970 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2971 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2972 np->n_directio_asyncwr != 0)) {
2974 /* try, try again... */
2981 vn_printf(vp, "ncl_flush failed");
2982 error = called_from_renewthread != 0 ? EIO : EBUSY;
2988 * NFS advisory byte-level locks.
2991 nfs_advlock(struct vop_advlock_args *ap)
2993 struct vnode *vp = ap->a_vp;
2995 struct nfsnode *np = VTONFS(ap->a_vp);
2996 struct proc *p = (struct proc *)ap->a_id;
2997 struct thread *td = curthread; /* XXX */
2999 int ret, error = EOPNOTSUPP;
3002 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3003 if (vp->v_type != VREG)
3005 if ((ap->a_flags & F_POSIX) != 0)
3008 cred = td->td_ucred;
3009 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3010 if (vp->v_iflag & VI_DOOMED) {
3011 NFSVOPUNLOCK(vp, 0);
3016 * If this is unlocking a write locked region, flush and
3017 * commit them before unlocking. This is required by
3018 * RFC3530 Sec. 9.3.2.
3020 if (ap->a_op == F_UNLCK &&
3021 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3023 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
3026 * Loop around doing the lock op, while a blocking lock
3027 * must wait for the lock op to succeed.
3030 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3031 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3032 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3033 ap->a_op == F_SETLK) {
3034 NFSVOPUNLOCK(vp, 0);
3035 error = nfs_catnap(PZERO | PCATCH, ret,
3039 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3040 if (vp->v_iflag & VI_DOOMED) {
3041 NFSVOPUNLOCK(vp, 0);
3045 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3046 ap->a_op == F_SETLK);
3047 if (ret == NFSERR_DENIED) {
3048 NFSVOPUNLOCK(vp, 0);
3050 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3051 NFSVOPUNLOCK(vp, 0);
3053 } else if (ret != 0) {
3054 NFSVOPUNLOCK(vp, 0);
3059 * Now, if we just got a lock, invalidate data in the buffer
3060 * cache, as required, so that the coherency conforms with
3061 * RFC3530 Sec. 9.3.2.
3063 if (ap->a_op == F_SETLK) {
3064 if ((np->n_flag & NMODIFIED) == 0) {
3065 np->n_attrstamp = 0;
3066 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3067 ret = VOP_GETATTR(vp, &va, cred);
3069 if ((np->n_flag & NMODIFIED) || ret ||
3070 np->n_change != va.va_filerev) {
3071 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3072 np->n_attrstamp = 0;
3073 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3074 ret = VOP_GETATTR(vp, &va, cred);
3076 np->n_mtime = va.va_mtime;
3077 np->n_change = va.va_filerev;
3080 /* Mark that a file lock has been acquired. */
3081 mtx_lock(&np->n_mtx);
3082 np->n_flag |= NHASBEENLOCKED;
3083 mtx_unlock(&np->n_mtx);
3085 NFSVOPUNLOCK(vp, 0);
3087 } else if (!NFS_ISV4(vp)) {
3088 error = NFSVOPLOCK(vp, LK_SHARED);
3091 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3092 size = VTONFS(vp)->n_size;
3093 NFSVOPUNLOCK(vp, 0);
3094 error = lf_advlock(ap, &(vp->v_lockf), size);
3096 if (nfs_advlock_p != NULL)
3097 error = nfs_advlock_p(ap);
3099 NFSVOPUNLOCK(vp, 0);
3103 if (error == 0 && ap->a_op == F_SETLK) {
3104 error = NFSVOPLOCK(vp, LK_SHARED);
3106 /* Mark that a file lock has been acquired. */
3107 mtx_lock(&np->n_mtx);
3108 np->n_flag |= NHASBEENLOCKED;
3109 mtx_unlock(&np->n_mtx);
3110 NFSVOPUNLOCK(vp, 0);
3118 * NFS advisory byte-level locks.
3121 nfs_advlockasync(struct vop_advlockasync_args *ap)
3123 struct vnode *vp = ap->a_vp;
3128 return (EOPNOTSUPP);
3129 error = NFSVOPLOCK(vp, LK_SHARED);
3132 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3133 size = VTONFS(vp)->n_size;
3134 NFSVOPUNLOCK(vp, 0);
3135 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3137 NFSVOPUNLOCK(vp, 0);
3144 * Print out the contents of an nfsnode.
3147 nfs_print(struct vop_print_args *ap)
3149 struct vnode *vp = ap->a_vp;
3150 struct nfsnode *np = VTONFS(vp);
3152 printf("\tfileid %ld fsid 0x%x", np->n_vattr.na_fileid,
3153 np->n_vattr.na_fsid);
3154 if (vp->v_type == VFIFO)
3161 * This is the "real" nfs::bwrite(struct buf*).
3162 * We set B_CACHE if this is a VMIO buffer.
3165 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3168 int oldflags = bp->b_flags;
3174 BUF_ASSERT_HELD(bp);
3176 if (bp->b_flags & B_INVAL) {
3181 bp->b_flags |= B_CACHE;
3184 * Undirty the bp. We will redirty it later if the I/O fails.
3189 bp->b_flags &= ~B_DONE;
3190 bp->b_ioflags &= ~BIO_ERROR;
3191 bp->b_iocmd = BIO_WRITE;
3193 bufobj_wref(bp->b_bufobj);
3194 curthread->td_ru.ru_oublock++;
3198 * Note: to avoid loopback deadlocks, we do not
3199 * assign b_runningbufspace.
3201 vfs_busy_pages(bp, 1);
3204 bp->b_iooffset = dbtob(bp->b_blkno);
3207 if( (oldflags & B_ASYNC) == 0) {
3208 int rtval = bufwait(bp);
3210 if (oldflags & B_DELWRI) {
3223 * nfs special file access vnode op.
3224 * Essentially just get vattr and then imitate iaccess() since the device is
3225 * local to the client.
3228 nfsspec_access(struct vop_access_args *ap)
3231 struct ucred *cred = ap->a_cred;
3232 struct vnode *vp = ap->a_vp;
3233 accmode_t accmode = ap->a_accmode;
3238 * Disallow write attempts on filesystems mounted read-only;
3239 * unless the file is a socket, fifo, or a block or character
3240 * device resident on the filesystem.
3242 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3243 switch (vp->v_type) {
3253 error = VOP_GETATTR(vp, vap, cred);
3256 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3257 accmode, cred, NULL);
3263 * Read wrapper for fifos.
3266 nfsfifo_read(struct vop_read_args *ap)
3268 struct nfsnode *np = VTONFS(ap->a_vp);
3274 mtx_lock(&np->n_mtx);
3276 vfs_timestamp(&np->n_atim);
3277 mtx_unlock(&np->n_mtx);
3278 error = fifo_specops.vop_read(ap);
3283 * Write wrapper for fifos.
3286 nfsfifo_write(struct vop_write_args *ap)
3288 struct nfsnode *np = VTONFS(ap->a_vp);
3293 mtx_lock(&np->n_mtx);
3295 vfs_timestamp(&np->n_mtim);
3296 mtx_unlock(&np->n_mtx);
3297 return(fifo_specops.vop_write(ap));
3301 * Close wrapper for fifos.
3303 * Update the times on the nfsnode then do fifo close.
3306 nfsfifo_close(struct vop_close_args *ap)
3308 struct vnode *vp = ap->a_vp;
3309 struct nfsnode *np = VTONFS(vp);
3313 mtx_lock(&np->n_mtx);
3314 if (np->n_flag & (NACC | NUPD)) {
3316 if (np->n_flag & NACC)
3318 if (np->n_flag & NUPD)
3321 if (vrefcnt(vp) == 1 &&
3322 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3324 if (np->n_flag & NACC)
3325 vattr.va_atime = np->n_atim;
3326 if (np->n_flag & NUPD)
3327 vattr.va_mtime = np->n_mtim;
3328 mtx_unlock(&np->n_mtx);
3329 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3333 mtx_unlock(&np->n_mtx);
3335 return (fifo_specops.vop_close(ap));
3339 * Just call ncl_writebp() with the force argument set to 1.
3341 * NOTE: B_DONE may or may not be set in a_bp on call.
3344 nfs_bwrite(struct buf *bp)
3347 return (ncl_writebp(bp, 1, curthread));
3350 struct buf_ops buf_ops_newnfs = {
3351 .bop_name = "buf_ops_nfs",
3352 .bop_write = nfs_bwrite,
3353 .bop_strategy = bufstrategy,
3354 .bop_sync = bufsync,
3355 .bop_bdflush = bufbdflush,
3359 nfs_getacl(struct vop_getacl_args *ap)
3363 if (ap->a_type != ACL_TYPE_NFS4)
3364 return (EOPNOTSUPP);
3365 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3367 if (error > NFSERR_STALE) {
3368 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3375 nfs_setacl(struct vop_setacl_args *ap)
3379 if (ap->a_type != ACL_TYPE_NFS4)
3380 return (EOPNOTSUPP);
3381 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3383 if (error > NFSERR_STALE) {
3384 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3391 * Return POSIX pathconf information applicable to nfs filesystems.
3394 nfs_pathconf(struct vop_pathconf_args *ap)
3396 struct nfsv3_pathconf pc;
3397 struct nfsvattr nfsva;
3398 struct vnode *vp = ap->a_vp;
3399 struct thread *td = curthread;
3400 int attrflag, error;
3402 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
3403 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3404 ap->a_name == _PC_NO_TRUNC)) ||
3405 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
3407 * Since only the above 4 a_names are returned by the NFSv3
3408 * Pathconf RPC, there is no point in doing it for others.
3409 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
3410 * be used for _PC_NFS4_ACL as well.
3412 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3415 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3421 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3424 pc.pc_linkmax = LINK_MAX;
3425 pc.pc_namemax = NFS_MAXNAMLEN;
3427 pc.pc_chownrestricted = 1;
3428 pc.pc_caseinsensitive = 0;
3429 pc.pc_casepreserving = 1;
3432 switch (ap->a_name) {
3434 *ap->a_retval = pc.pc_linkmax;
3437 *ap->a_retval = pc.pc_namemax;
3440 *ap->a_retval = PATH_MAX;
3443 *ap->a_retval = PIPE_BUF;
3445 case _PC_CHOWN_RESTRICTED:
3446 *ap->a_retval = pc.pc_chownrestricted;
3449 *ap->a_retval = pc.pc_notrunc;
3451 case _PC_ACL_EXTENDED:
3455 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3456 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3461 case _PC_ACL_PATH_MAX:
3463 *ap->a_retval = ACL_MAX_ENTRIES;
3467 case _PC_MAC_PRESENT:
3471 /* _PC_ASYNC_IO should have been handled by upper layers. */
3472 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3481 case _PC_ALLOC_SIZE_MIN:
3482 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3484 case _PC_FILESIZEBITS:
3490 case _PC_REC_INCR_XFER_SIZE:
3491 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3493 case _PC_REC_MAX_XFER_SIZE:
3494 *ap->a_retval = -1; /* means ``unlimited'' */
3496 case _PC_REC_MIN_XFER_SIZE:
3497 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3499 case _PC_REC_XFER_ALIGN:
3500 *ap->a_retval = PAGE_SIZE;
3502 case _PC_SYMLINK_MAX:
3503 *ap->a_retval = NFS_MAXPATHLEN;