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_extern.h>
67 #include <vm/vm_object.h>
69 #include <fs/nfs/nfsport.h>
70 #include <fs/nfsclient/nfsnode.h>
71 #include <fs/nfsclient/nfsmount.h>
72 #include <fs/nfsclient/nfs.h>
73 #include <fs/nfsclient/nfs_kdtrace.h>
76 #include <netinet/in.h>
77 #include <netinet/in_var.h>
79 #include <nfs/nfs_lock.h>
82 #include <sys/dtrace_bsd.h>
84 dtrace_nfsclient_accesscache_flush_probe_func_t
85 dtrace_nfscl_accesscache_flush_done_probe;
86 uint32_t nfscl_accesscache_flush_done_id;
88 dtrace_nfsclient_accesscache_get_probe_func_t
89 dtrace_nfscl_accesscache_get_hit_probe,
90 dtrace_nfscl_accesscache_get_miss_probe;
91 uint32_t nfscl_accesscache_get_hit_id;
92 uint32_t nfscl_accesscache_get_miss_id;
94 dtrace_nfsclient_accesscache_load_probe_func_t
95 dtrace_nfscl_accesscache_load_done_probe;
96 uint32_t nfscl_accesscache_load_done_id;
97 #endif /* !KDTRACE_HOOKS */
103 extern struct nfsstats newnfsstats;
104 extern int nfsrv_useacl;
105 extern int nfscl_debuglevel;
106 MALLOC_DECLARE(M_NEWNFSREQ);
109 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
110 * calls are not in getblk() and brelse() so that they would not be necessary
114 #define vfs_busy_pages(bp, f)
117 static vop_read_t nfsfifo_read;
118 static vop_write_t nfsfifo_write;
119 static vop_close_t nfsfifo_close;
120 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
122 static vop_lookup_t nfs_lookup;
123 static vop_create_t nfs_create;
124 static vop_mknod_t nfs_mknod;
125 static vop_open_t nfs_open;
126 static vop_pathconf_t nfs_pathconf;
127 static vop_close_t nfs_close;
128 static vop_access_t nfs_access;
129 static vop_getattr_t nfs_getattr;
130 static vop_setattr_t nfs_setattr;
131 static vop_read_t nfs_read;
132 static vop_fsync_t nfs_fsync;
133 static vop_remove_t nfs_remove;
134 static vop_link_t nfs_link;
135 static vop_rename_t nfs_rename;
136 static vop_mkdir_t nfs_mkdir;
137 static vop_rmdir_t nfs_rmdir;
138 static vop_symlink_t nfs_symlink;
139 static vop_readdir_t nfs_readdir;
140 static vop_strategy_t nfs_strategy;
141 static vop_lock1_t nfs_lock1;
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_lock1 = nfs_lock1,
171 .vop_lookup = nfs_lookup,
172 .vop_mkdir = nfs_mkdir,
173 .vop_mknod = nfs_mknod,
174 .vop_open = nfs_open,
175 .vop_pathconf = nfs_pathconf,
176 .vop_print = nfs_print,
177 .vop_read = nfs_read,
178 .vop_readdir = nfs_readdir,
179 .vop_readlink = nfs_readlink,
180 .vop_reclaim = ncl_reclaim,
181 .vop_remove = nfs_remove,
182 .vop_rename = nfs_rename,
183 .vop_rmdir = nfs_rmdir,
184 .vop_setattr = nfs_setattr,
185 .vop_strategy = nfs_strategy,
186 .vop_symlink = nfs_symlink,
187 .vop_write = ncl_write,
188 .vop_getacl = nfs_getacl,
189 .vop_setacl = nfs_setacl,
192 struct vop_vector newnfs_fifoops = {
193 .vop_default = &fifo_specops,
194 .vop_access = nfsspec_access,
195 .vop_close = nfsfifo_close,
196 .vop_fsync = nfs_fsync,
197 .vop_getattr = nfs_getattr,
198 .vop_inactive = ncl_inactive,
199 .vop_print = nfs_print,
200 .vop_read = nfsfifo_read,
201 .vop_reclaim = ncl_reclaim,
202 .vop_setattr = nfs_setattr,
203 .vop_write = nfsfifo_write,
206 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
207 struct componentname *cnp, struct vattr *vap);
208 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
209 int namelen, struct ucred *cred, struct thread *td);
210 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
211 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
212 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
213 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
214 struct componentname *scnp, struct sillyrename *sp);
219 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
221 SYSCTL_DECL(_vfs_nfs);
223 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
224 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
225 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
227 static int nfs_prime_access_cache = 0;
228 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
229 &nfs_prime_access_cache, 0,
230 "Prime NFS ACCESS cache when fetching attributes");
232 static int newnfs_commit_on_close = 0;
233 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
234 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
236 static int nfs_clean_pages_on_close = 1;
237 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
238 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
240 int newnfs_directio_enable = 0;
241 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
242 &newnfs_directio_enable, 0, "Enable NFS directio");
244 int nfs_keep_dirty_on_error;
245 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
246 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
249 * This sysctl allows other processes to mmap a file that has been opened
250 * O_DIRECT by a process. In general, having processes mmap the file while
251 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
252 * this by default to prevent DoS attacks - to prevent a malicious user from
253 * opening up files O_DIRECT preventing other users from mmap'ing these
254 * files. "Protected" environments where stricter consistency guarantees are
255 * required can disable this knob. The process that opened the file O_DIRECT
256 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
259 int newnfs_directio_allow_mmap = 1;
260 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
261 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
264 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
265 &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
267 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
268 &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
271 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
272 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
273 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
277 * The list of locks after the description of the lock is the ordering
278 * of other locks acquired with the lock held.
279 * np->n_mtx : Protects the fields in the nfsnode.
281 VI_MTX (acquired indirectly)
282 * nmp->nm_mtx : Protects the fields in the nfsmount.
284 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
285 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
288 * rep->r_mtx : Protects the fields in an nfsreq.
292 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
293 struct ucred *cred, u_int32_t *retmode)
295 int error = 0, attrflag, i, lrupos;
297 struct nfsnode *np = VTONFS(vp);
298 struct nfsvattr nfsva;
300 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
303 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
306 mtx_lock(&np->n_mtx);
307 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
308 if (np->n_accesscache[i].uid == cred->cr_uid) {
309 np->n_accesscache[i].mode = rmode;
310 np->n_accesscache[i].stamp = time_second;
313 if (i > 0 && np->n_accesscache[i].stamp <
314 np->n_accesscache[lrupos].stamp)
317 if (i == NFS_ACCESSCACHESIZE) {
318 np->n_accesscache[lrupos].uid = cred->cr_uid;
319 np->n_accesscache[lrupos].mode = rmode;
320 np->n_accesscache[lrupos].stamp = time_second;
322 mtx_unlock(&np->n_mtx);
325 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
326 } else if (NFS_ISV4(vp)) {
327 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
331 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
338 * nfs access vnode op.
339 * For nfs version 2, just return ok. File accesses may fail later.
340 * For nfs version 3, use the access rpc to check accessibility. If file modes
341 * are changed on the server, accesses might still fail later.
344 nfs_access(struct vop_access_args *ap)
346 struct vnode *vp = ap->a_vp;
347 int error = 0, i, gotahit;
348 u_int32_t mode, wmode, rmode;
349 int v34 = NFS_ISV34(vp);
350 struct nfsnode *np = VTONFS(vp);
353 * Disallow write attempts on filesystems mounted read-only;
354 * unless the file is a socket, fifo, or a block or character
355 * device resident on the filesystem.
357 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
358 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
359 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
360 switch (vp->v_type) {
370 * For nfs v3 or v4, check to see if we have done this recently, and if
371 * so return our cached result instead of making an ACCESS call.
372 * If not, do an access rpc, otherwise you are stuck emulating
373 * ufs_access() locally using the vattr. This may not be correct,
374 * since the server may apply other access criteria such as
375 * client uid-->server uid mapping that we do not know about.
378 if (ap->a_accmode & VREAD)
379 mode = NFSACCESS_READ;
382 if (vp->v_type != VDIR) {
383 if (ap->a_accmode & VWRITE)
384 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
385 if (ap->a_accmode & VAPPEND)
386 mode |= NFSACCESS_EXTEND;
387 if (ap->a_accmode & VEXEC)
388 mode |= NFSACCESS_EXECUTE;
389 if (ap->a_accmode & VDELETE)
390 mode |= NFSACCESS_DELETE;
392 if (ap->a_accmode & VWRITE)
393 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
394 if (ap->a_accmode & VAPPEND)
395 mode |= NFSACCESS_EXTEND;
396 if (ap->a_accmode & VEXEC)
397 mode |= NFSACCESS_LOOKUP;
398 if (ap->a_accmode & VDELETE)
399 mode |= NFSACCESS_DELETE;
400 if (ap->a_accmode & VDELETE_CHILD)
401 mode |= NFSACCESS_MODIFY;
403 /* XXX safety belt, only make blanket request if caching */
404 if (nfsaccess_cache_timeout > 0) {
405 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
406 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
407 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
413 * Does our cached result allow us to give a definite yes to
417 mtx_lock(&np->n_mtx);
418 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
419 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
420 if (time_second < (np->n_accesscache[i].stamp
421 + nfsaccess_cache_timeout) &&
422 (np->n_accesscache[i].mode & mode) == mode) {
423 NFSINCRGLOBAL(newnfsstats.accesscache_hits);
429 mtx_unlock(&np->n_mtx);
432 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
433 ap->a_cred->cr_uid, mode);
435 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
436 ap->a_cred->cr_uid, mode);
440 * Either a no, or a don't know. Go to the wire.
442 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
443 error = nfs34_access_otw(vp, wmode, ap->a_td,
446 (rmode & mode) != mode)
451 if ((error = nfsspec_access(ap)) != 0) {
455 * Attempt to prevent a mapped root from accessing a file
456 * which it shouldn't. We try to read a byte from the file
457 * if the user is root and the file is not zero length.
458 * After calling nfsspec_access, we should have the correct
461 mtx_lock(&np->n_mtx);
462 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
463 && VTONFS(vp)->n_size > 0) {
468 mtx_unlock(&np->n_mtx);
471 auio.uio_iov = &aiov;
475 auio.uio_segflg = UIO_SYSSPACE;
476 auio.uio_rw = UIO_READ;
477 auio.uio_td = ap->a_td;
479 if (vp->v_type == VREG)
480 error = ncl_readrpc(vp, &auio, ap->a_cred);
481 else if (vp->v_type == VDIR) {
483 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
485 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
486 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
489 } else if (vp->v_type == VLNK)
490 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
494 mtx_unlock(&np->n_mtx);
502 * Check to see if the type is ok
503 * and that deletion is not in progress.
504 * For paged in text files, you will need to flush the page cache
505 * if consistency is lost.
509 nfs_open(struct vop_open_args *ap)
511 struct vnode *vp = ap->a_vp;
512 struct nfsnode *np = VTONFS(vp);
515 int fmode = ap->a_mode;
518 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
522 * For NFSv4, we need to do the Open Op before cache validation,
523 * so that we conform to RFC3530 Sec. 9.3.1.
526 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
528 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
535 * Now, if this Open will be doing reading, re-validate/flush the
536 * cache, so that Close/Open coherency is maintained.
538 mtx_lock(&np->n_mtx);
539 if (np->n_flag & NMODIFIED) {
540 mtx_unlock(&np->n_mtx);
541 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
542 if (error == EINTR || error == EIO) {
544 (void) nfsrpc_close(vp, 0, ap->a_td);
547 mtx_lock(&np->n_mtx);
549 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
550 if (vp->v_type == VDIR)
551 np->n_direofoffset = 0;
552 mtx_unlock(&np->n_mtx);
553 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
556 (void) nfsrpc_close(vp, 0, ap->a_td);
559 mtx_lock(&np->n_mtx);
560 np->n_mtime = vattr.va_mtime;
562 np->n_change = vattr.va_filerev;
564 mtx_unlock(&np->n_mtx);
565 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
568 (void) nfsrpc_close(vp, 0, ap->a_td);
571 mtx_lock(&np->n_mtx);
572 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
573 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
574 if (vp->v_type == VDIR)
575 np->n_direofoffset = 0;
576 mtx_unlock(&np->n_mtx);
577 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
578 if (error == EINTR || error == EIO) {
580 (void) nfsrpc_close(vp, 0, ap->a_td);
583 mtx_lock(&np->n_mtx);
584 np->n_mtime = vattr.va_mtime;
586 np->n_change = vattr.va_filerev;
591 * If the object has >= 1 O_DIRECT active opens, we disable caching.
593 if (newnfs_directio_enable && (fmode & O_DIRECT) &&
594 (vp->v_type == VREG)) {
595 if (np->n_directio_opens == 0) {
596 mtx_unlock(&np->n_mtx);
597 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
600 (void) nfsrpc_close(vp, 0, ap->a_td);
603 mtx_lock(&np->n_mtx);
604 np->n_flag |= NNONCACHE;
606 np->n_directio_opens++;
609 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
610 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
611 np->n_flag |= NWRITEOPENED;
614 * If this is an open for writing, capture a reference to the
615 * credentials, so they can be used by ncl_putpages(). Using
616 * these write credentials is preferable to the credentials of
617 * whatever thread happens to be doing the VOP_PUTPAGES() since
618 * the write RPCs are less likely to fail with EACCES.
620 if ((fmode & FWRITE) != 0) {
621 cred = np->n_writecred;
622 np->n_writecred = crhold(ap->a_cred);
625 mtx_unlock(&np->n_mtx);
629 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
635 * What an NFS client should do upon close after writing is a debatable issue.
636 * Most NFS clients push delayed writes to the server upon close, basically for
638 * 1 - So that any write errors may be reported back to the client process
639 * doing the close system call. By far the two most likely errors are
640 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
641 * 2 - To put a worst case upper bound on cache inconsistency between
642 * multiple clients for the file.
643 * There is also a consistency problem for Version 2 of the protocol w.r.t.
644 * not being able to tell if other clients are writing a file concurrently,
645 * since there is no way of knowing if the changed modify time in the reply
646 * is only due to the write for this client.
647 * (NFS Version 3 provides weak cache consistency data in the reply that
648 * should be sufficient to detect and handle this case.)
650 * The current code does the following:
651 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
652 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
653 * or commit them (this satisfies 1 and 2 except for the
654 * case where the server crashes after this close but
655 * before the commit RPC, which is felt to be "good
656 * enough". Changing the last argument to ncl_flush() to
657 * a 1 would force a commit operation, if it is felt a
658 * commit is necessary now.
659 * for NFS Version 4 - flush the dirty buffers and commit them, if
660 * nfscl_mustflush() says this is necessary.
661 * It is necessary if there is no write delegation held,
662 * in order to satisfy open/close coherency.
663 * If the file isn't cached on local stable storage,
664 * it may be necessary in order to detect "out of space"
665 * errors from the server, if the write delegation
666 * issued by the server doesn't allow the file to grow.
670 nfs_close(struct vop_close_args *ap)
672 struct vnode *vp = ap->a_vp;
673 struct nfsnode *np = VTONFS(vp);
674 struct nfsvattr nfsva;
676 int error = 0, ret, localcred = 0;
677 int fmode = ap->a_fflag;
679 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
682 * During shutdown, a_cred isn't valid, so just use root.
684 if (ap->a_cred == NOCRED) {
685 cred = newnfs_getcred();
690 if (vp->v_type == VREG) {
692 * Examine and clean dirty pages, regardless of NMODIFIED.
693 * This closes a major hole in close-to-open consistency.
694 * We want to push out all dirty pages (and buffers) on
695 * close, regardless of whether they were dirtied by
696 * mmap'ed writes or via write().
698 if (nfs_clean_pages_on_close && vp->v_object) {
699 VM_OBJECT_WLOCK(vp->v_object);
700 vm_object_page_clean(vp->v_object, 0, 0, 0);
701 VM_OBJECT_WUNLOCK(vp->v_object);
703 mtx_lock(&np->n_mtx);
704 if (np->n_flag & NMODIFIED) {
705 mtx_unlock(&np->n_mtx);
708 * Under NFSv3 we have dirty buffers to dispose of. We
709 * must flush them to the NFS server. We have the option
710 * of waiting all the way through the commit rpc or just
711 * waiting for the initial write. The default is to only
712 * wait through the initial write so the data is in the
713 * server's cache, which is roughly similar to the state
714 * a standard disk subsystem leaves the file in on close().
716 * We cannot clear the NMODIFIED bit in np->n_flag due to
717 * potential races with other processes, and certainly
718 * cannot clear it if we don't commit.
719 * These races occur when there is no longer the old
720 * traditional vnode locking implemented for Vnode Ops.
722 int cm = newnfs_commit_on_close ? 1 : 0;
723 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
724 /* np->n_flag &= ~NMODIFIED; */
725 } else if (NFS_ISV4(vp)) {
726 if (nfscl_mustflush(vp) != 0) {
727 int cm = newnfs_commit_on_close ? 1 : 0;
728 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
731 * as above w.r.t races when clearing
733 * np->n_flag &= ~NMODIFIED;
737 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
738 mtx_lock(&np->n_mtx);
741 * Invalidate the attribute cache in all cases.
742 * An open is going to fetch fresh attrs any way, other procs
743 * on this node that have file open will be forced to do an
744 * otw attr fetch, but this is safe.
745 * --> A user found that their RPC count dropped by 20% when
746 * this was commented out and I can't see any requirement
747 * for it, so I've disabled it when negative lookups are
748 * enabled. (What does this have to do with negative lookup
749 * caching? Well nothing, except it was reported by the
750 * same user that needed negative lookup caching and I wanted
751 * there to be a way to disable it to see if it
752 * is the cause of some caching/coherency issue that might
755 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
757 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
759 if (np->n_flag & NWRITEERR) {
760 np->n_flag &= ~NWRITEERR;
763 mtx_unlock(&np->n_mtx);
768 * Get attributes so "change" is up to date.
770 if (error == 0 && nfscl_mustflush(vp) != 0 &&
771 vp->v_type == VREG &&
772 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
773 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
776 np->n_change = nfsva.na_filerev;
777 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
785 ret = nfsrpc_close(vp, 0, ap->a_td);
789 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
792 if (newnfs_directio_enable)
793 KASSERT((np->n_directio_asyncwr == 0),
794 ("nfs_close: dirty unflushed (%d) directio buffers\n",
795 np->n_directio_asyncwr));
796 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
797 mtx_lock(&np->n_mtx);
798 KASSERT((np->n_directio_opens > 0),
799 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
800 np->n_directio_opens--;
801 if (np->n_directio_opens == 0)
802 np->n_flag &= ~NNONCACHE;
803 mtx_unlock(&np->n_mtx);
811 * nfs getattr call from vfs.
814 nfs_getattr(struct vop_getattr_args *ap)
816 struct vnode *vp = ap->a_vp;
817 struct thread *td = curthread; /* XXX */
818 struct nfsnode *np = VTONFS(vp);
820 struct nfsvattr nfsva;
821 struct vattr *vap = ap->a_vap;
825 * Update local times for special files.
827 mtx_lock(&np->n_mtx);
828 if (np->n_flag & (NACC | NUPD))
830 mtx_unlock(&np->n_mtx);
832 * First look in the cache.
834 if (ncl_getattrcache(vp, &vattr) == 0) {
835 vap->va_type = vattr.va_type;
836 vap->va_mode = vattr.va_mode;
837 vap->va_nlink = vattr.va_nlink;
838 vap->va_uid = vattr.va_uid;
839 vap->va_gid = vattr.va_gid;
840 vap->va_fsid = vattr.va_fsid;
841 vap->va_fileid = vattr.va_fileid;
842 vap->va_size = vattr.va_size;
843 vap->va_blocksize = vattr.va_blocksize;
844 vap->va_atime = vattr.va_atime;
845 vap->va_mtime = vattr.va_mtime;
846 vap->va_ctime = vattr.va_ctime;
847 vap->va_gen = vattr.va_gen;
848 vap->va_flags = vattr.va_flags;
849 vap->va_rdev = vattr.va_rdev;
850 vap->va_bytes = vattr.va_bytes;
851 vap->va_filerev = vattr.va_filerev;
853 * Get the local modify time for the case of a write
856 nfscl_deleggetmodtime(vp, &vap->va_mtime);
860 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
861 nfsaccess_cache_timeout > 0) {
862 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
863 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
864 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
865 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
869 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
871 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
874 * Get the local modify time for the case of a write
877 nfscl_deleggetmodtime(vp, &vap->va_mtime);
878 } else if (NFS_ISV4(vp)) {
879 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
888 nfs_setattr(struct vop_setattr_args *ap)
890 struct vnode *vp = ap->a_vp;
891 struct nfsnode *np = VTONFS(vp);
892 struct thread *td = curthread; /* XXX */
893 struct vattr *vap = ap->a_vap;
902 * Setting of flags and marking of atimes are not supported.
904 if (vap->va_flags != VNOVAL)
908 * Disallow write attempts if the filesystem is mounted read-only.
910 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
911 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
912 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
913 (vp->v_mount->mnt_flag & MNT_RDONLY))
915 if (vap->va_size != VNOVAL) {
916 switch (vp->v_type) {
923 if (vap->va_mtime.tv_sec == VNOVAL &&
924 vap->va_atime.tv_sec == VNOVAL &&
925 vap->va_mode == (mode_t)VNOVAL &&
926 vap->va_uid == (uid_t)VNOVAL &&
927 vap->va_gid == (gid_t)VNOVAL)
929 vap->va_size = VNOVAL;
933 * Disallow write attempts if the filesystem is
936 if (vp->v_mount->mnt_flag & MNT_RDONLY)
939 * We run vnode_pager_setsize() early (why?),
940 * we must set np->n_size now to avoid vinvalbuf
941 * V_SAVE races that might setsize a lower
944 mtx_lock(&np->n_mtx);
946 mtx_unlock(&np->n_mtx);
947 error = ncl_meta_setsize(vp, ap->a_cred, td,
949 mtx_lock(&np->n_mtx);
950 if (np->n_flag & NMODIFIED) {
952 mtx_unlock(&np->n_mtx);
953 if (vap->va_size == 0)
954 error = ncl_vinvalbuf(vp, 0, td, 1);
956 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
958 vnode_pager_setsize(vp, tsize);
962 * Call nfscl_delegmodtime() to set the modify time
963 * locally, as required.
965 nfscl_delegmodtime(vp);
967 mtx_unlock(&np->n_mtx);
969 * np->n_size has already been set to vap->va_size
970 * in ncl_meta_setsize(). We must set it again since
971 * nfs_loadattrcache() could be called through
972 * ncl_meta_setsize() and could modify np->n_size.
974 mtx_lock(&np->n_mtx);
975 np->n_vattr.na_size = np->n_size = vap->va_size;
976 mtx_unlock(&np->n_mtx);
979 mtx_lock(&np->n_mtx);
980 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
981 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
982 mtx_unlock(&np->n_mtx);
983 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
984 (error == EINTR || error == EIO))
987 mtx_unlock(&np->n_mtx);
989 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
990 if (error && vap->va_size != VNOVAL) {
991 mtx_lock(&np->n_mtx);
992 np->n_size = np->n_vattr.na_size = tsize;
993 vnode_pager_setsize(vp, tsize);
994 mtx_unlock(&np->n_mtx);
1000 * Do an nfs setattr rpc.
1003 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1006 struct nfsnode *np = VTONFS(vp);
1007 int error, ret, attrflag, i;
1008 struct nfsvattr nfsva;
1010 if (NFS_ISV34(vp)) {
1011 mtx_lock(&np->n_mtx);
1012 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1013 np->n_accesscache[i].stamp = 0;
1014 np->n_flag |= NDELEGMOD;
1015 mtx_unlock(&np->n_mtx);
1016 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1018 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1021 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1025 if (error && NFS_ISV4(vp))
1026 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1031 * nfs lookup call, one step at a time...
1032 * First look in cache
1033 * If not found, unlock the directory nfsnode and do the rpc
1036 nfs_lookup(struct vop_lookup_args *ap)
1038 struct componentname *cnp = ap->a_cnp;
1039 struct vnode *dvp = ap->a_dvp;
1040 struct vnode **vpp = ap->a_vpp;
1041 struct mount *mp = dvp->v_mount;
1042 int flags = cnp->cn_flags;
1043 struct vnode *newvp;
1044 struct nfsmount *nmp;
1045 struct nfsnode *np, *newnp;
1046 int error = 0, attrflag, dattrflag, ltype, ncticks;
1047 struct thread *td = cnp->cn_thread;
1049 struct nfsvattr dnfsva, nfsva;
1051 struct timespec nctime;
1054 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1055 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1057 if (dvp->v_type != VDIR)
1062 /* For NFSv4, wait until any remove is done. */
1063 mtx_lock(&np->n_mtx);
1064 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1065 np->n_flag |= NREMOVEWANT;
1066 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1068 mtx_unlock(&np->n_mtx);
1070 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1072 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1073 if (error > 0 && error != ENOENT)
1077 * Lookups of "." are special and always return the
1078 * current directory. cache_lookup() already handles
1079 * associated locking bookkeeping, etc.
1081 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1082 /* XXX: Is this really correct? */
1083 if (cnp->cn_nameiop != LOOKUP &&
1085 cnp->cn_flags |= SAVENAME;
1090 * We only accept a positive hit in the cache if the
1091 * change time of the file matches our cached copy.
1092 * Otherwise, we discard the cache entry and fallback
1093 * to doing a lookup RPC. We also only trust cache
1094 * entries for less than nm_nametimeo seconds.
1096 * To better handle stale file handles and attributes,
1097 * clear the attribute cache of this node if it is a
1098 * leaf component, part of an open() call, and not
1099 * locally modified before fetching the attributes.
1100 * This should allow stale file handles to be detected
1101 * here where we can fall back to a LOOKUP RPC to
1102 * recover rather than having nfs_open() detect the
1103 * stale file handle and failing open(2) with ESTALE.
1106 newnp = VTONFS(newvp);
1107 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1108 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1109 !(newnp->n_flag & NMODIFIED)) {
1110 mtx_lock(&newnp->n_mtx);
1111 newnp->n_attrstamp = 0;
1112 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1113 mtx_unlock(&newnp->n_mtx);
1115 if (nfscl_nodeleg(newvp, 0) == 0 ||
1116 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1117 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1118 timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1119 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1120 if (cnp->cn_nameiop != LOOKUP &&
1122 cnp->cn_flags |= SAVENAME;
1131 } else if (error == ENOENT) {
1132 if (dvp->v_iflag & VI_DOOMED)
1135 * We only accept a negative hit in the cache if the
1136 * modification time of the parent directory matches
1137 * the cached copy in the name cache entry.
1138 * Otherwise, we discard all of the negative cache
1139 * entries for this directory. We also only trust
1140 * negative cache entries for up to nm_negnametimeo
1143 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1144 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1145 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1146 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1149 cache_purge_negative(dvp);
1154 NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1155 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1156 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1159 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1161 if (newvp != NULLVP) {
1166 if (error != ENOENT) {
1168 error = nfscl_maperr(td, error, (uid_t)0,
1173 /* The requested file was not found. */
1174 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1175 (flags & ISLASTCN)) {
1177 * XXX: UFS does a full VOP_ACCESS(dvp,
1178 * VWRITE) here instead of just checking
1181 if (mp->mnt_flag & MNT_RDONLY)
1183 cnp->cn_flags |= SAVENAME;
1184 return (EJUSTRETURN);
1187 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1189 * Cache the modification time of the parent
1190 * directory from the post-op attributes in
1191 * the name cache entry. The negative cache
1192 * entry will be ignored once the directory
1193 * has changed. Don't bother adding the entry
1194 * if the directory has already changed.
1196 mtx_lock(&np->n_mtx);
1197 if (timespeccmp(&np->n_vattr.na_mtime,
1198 &dnfsva.na_mtime, ==)) {
1199 mtx_unlock(&np->n_mtx);
1200 cache_enter_time(dvp, NULL, cnp,
1201 &dnfsva.na_mtime, NULL);
1203 mtx_unlock(&np->n_mtx);
1209 * Handle RENAME case...
1211 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1212 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1213 FREE((caddr_t)nfhp, M_NFSFH);
1216 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1222 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1225 cnp->cn_flags |= SAVENAME;
1229 if (flags & ISDOTDOT) {
1230 ltype = NFSVOPISLOCKED(dvp);
1231 error = vfs_busy(mp, MBF_NOWAIT);
1234 NFSVOPUNLOCK(dvp, 0);
1235 error = vfs_busy(mp, 0);
1236 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1238 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1245 NFSVOPUNLOCK(dvp, 0);
1246 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1252 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1253 if (dvp->v_iflag & VI_DOOMED) {
1265 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1267 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1268 FREE((caddr_t)nfhp, M_NFSFH);
1272 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1275 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1281 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1283 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1284 !(np->n_flag & NMODIFIED)) {
1286 * Flush the attribute cache when opening a
1287 * leaf node to ensure that fresh attributes
1288 * are fetched in nfs_open() since we did not
1289 * fetch attributes from the LOOKUP reply.
1291 mtx_lock(&np->n_mtx);
1292 np->n_attrstamp = 0;
1293 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1294 mtx_unlock(&np->n_mtx);
1297 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1298 cnp->cn_flags |= SAVENAME;
1299 if ((cnp->cn_flags & MAKEENTRY) &&
1300 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1301 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1302 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1303 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1310 * Just call ncl_bioread() to do the work.
1313 nfs_read(struct vop_read_args *ap)
1315 struct vnode *vp = ap->a_vp;
1317 switch (vp->v_type) {
1319 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1323 return (EOPNOTSUPP);
1331 nfs_readlink(struct vop_readlink_args *ap)
1333 struct vnode *vp = ap->a_vp;
1335 if (vp->v_type != VLNK)
1337 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1341 * Do a readlink rpc.
1342 * Called by ncl_doio() from below the buffer cache.
1345 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1347 int error, ret, attrflag;
1348 struct nfsvattr nfsva;
1350 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1353 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1357 if (error && NFS_ISV4(vp))
1358 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1367 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1369 int error, ret, attrflag;
1370 struct nfsvattr nfsva;
1371 struct nfsmount *nmp;
1373 nmp = VFSTONFS(vnode_mount(vp));
1376 if (NFSHASPNFS(nmp))
1377 error = nfscl_doiods(vp, uiop, NULL, NULL,
1378 NFSV4OPEN_ACCESSREAD, cred, uiop->uio_td);
1379 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1381 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1384 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1388 if (error && NFS_ISV4(vp))
1389 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1397 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1398 int *iomode, int *must_commit, int called_from_strategy)
1400 struct nfsvattr nfsva;
1401 int error, attrflag, ret;
1402 struct nfsmount *nmp;
1404 nmp = VFSTONFS(vnode_mount(vp));
1407 if (NFSHASPNFS(nmp))
1408 error = nfscl_doiods(vp, uiop, iomode, must_commit,
1409 NFSV4OPEN_ACCESSWRITE, cred, uiop->uio_td);
1410 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1412 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1413 uiop->uio_td, &nfsva, &attrflag, NULL,
1414 called_from_strategy);
1416 if (VTONFS(vp)->n_flag & ND_NFSV4)
1417 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1420 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1426 *iomode = NFSWRITE_FILESYNC;
1427 if (error && NFS_ISV4(vp))
1428 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1434 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1435 * mode set to specify the file type and the size field for rdev.
1438 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1441 struct nfsvattr nfsva, dnfsva;
1442 struct vnode *newvp = NULL;
1443 struct nfsnode *np = NULL, *dnp;
1446 int error = 0, attrflag, dattrflag;
1449 if (vap->va_type == VCHR || vap->va_type == VBLK)
1450 rdev = vap->va_rdev;
1451 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1454 return (EOPNOTSUPP);
1455 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1457 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1458 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1459 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1462 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1463 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1464 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1467 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1468 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1471 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1474 if (attrflag != 0) {
1475 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1483 } else if (NFS_ISV4(dvp)) {
1484 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1488 mtx_lock(&dnp->n_mtx);
1489 dnp->n_flag |= NMODIFIED;
1491 dnp->n_attrstamp = 0;
1492 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1494 mtx_unlock(&dnp->n_mtx);
1500 * just call nfs_mknodrpc() to do the work.
1504 nfs_mknod(struct vop_mknod_args *ap)
1506 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1509 static struct mtx nfs_cverf_mtx;
1510 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1516 static nfsquad_t cverf;
1518 static int cverf_initialized = 0;
1520 mtx_lock(&nfs_cverf_mtx);
1521 if (cverf_initialized == 0) {
1522 cverf.lval[0] = arc4random();
1523 cverf.lval[1] = arc4random();
1524 cverf_initialized = 1;
1528 mtx_unlock(&nfs_cverf_mtx);
1534 * nfs file create call
1537 nfs_create(struct vop_create_args *ap)
1539 struct vnode *dvp = ap->a_dvp;
1540 struct vattr *vap = ap->a_vap;
1541 struct componentname *cnp = ap->a_cnp;
1542 struct nfsnode *np = NULL, *dnp;
1543 struct vnode *newvp = NULL;
1544 struct nfsmount *nmp;
1545 struct nfsvattr dnfsva, nfsva;
1548 int error = 0, attrflag, dattrflag, fmode = 0;
1552 * Oops, not for me..
1554 if (vap->va_type == VSOCK)
1555 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1557 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1559 if (vap->va_vaflags & VA_EXCLUSIVE)
1562 nmp = VFSTONFS(vnode_mount(dvp));
1564 /* For NFSv4, wait until any remove is done. */
1565 mtx_lock(&dnp->n_mtx);
1566 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1567 dnp->n_flag |= NREMOVEWANT;
1568 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1570 mtx_unlock(&dnp->n_mtx);
1572 cverf = nfs_get_cverf();
1573 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1574 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1575 &nfhp, &attrflag, &dattrflag, NULL);
1578 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1579 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1580 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1583 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1584 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1587 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1591 error = nfsrpc_getattr(newvp, cnp->cn_cred,
1592 cnp->cn_thread, &nfsva, NULL);
1594 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1598 if (newvp != NULL) {
1602 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1603 error == NFSERR_NOTSUPP) {
1607 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1608 if (nfscl_checksattr(vap, &nfsva)) {
1609 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1610 cnp->cn_thread, &nfsva, &attrflag, NULL);
1611 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1612 vap->va_gid != (gid_t)VNOVAL)) {
1613 /* try again without setting uid/gid */
1614 vap->va_uid = (uid_t)VNOVAL;
1615 vap->va_gid = (uid_t)VNOVAL;
1616 error = nfsrpc_setattr(newvp, vap, NULL,
1617 cnp->cn_cred, cnp->cn_thread, &nfsva,
1621 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1628 if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1629 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1632 } else if (NFS_ISV4(dvp)) {
1633 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1636 mtx_lock(&dnp->n_mtx);
1637 dnp->n_flag |= NMODIFIED;
1639 dnp->n_attrstamp = 0;
1640 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1642 mtx_unlock(&dnp->n_mtx);
1647 * nfs file remove call
1648 * To try and make nfs semantics closer to ufs semantics, a file that has
1649 * other processes using the vnode is renamed instead of removed and then
1650 * removed later on the last close.
1651 * - If v_usecount > 1
1652 * If a rename is not already in the works
1653 * call nfs_sillyrename() to set it up
1658 nfs_remove(struct vop_remove_args *ap)
1660 struct vnode *vp = ap->a_vp;
1661 struct vnode *dvp = ap->a_dvp;
1662 struct componentname *cnp = ap->a_cnp;
1663 struct nfsnode *np = VTONFS(vp);
1667 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1668 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1669 if (vp->v_type == VDIR)
1671 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1672 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1673 vattr.va_nlink > 1)) {
1675 * Purge the name cache so that the chance of a lookup for
1676 * the name succeeding while the remove is in progress is
1677 * minimized. Without node locking it can still happen, such
1678 * that an I/O op returns ESTALE, but since you get this if
1679 * another host removes the file..
1683 * throw away biocache buffers, mainly to avoid
1684 * unnecessary delayed writes later.
1686 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1688 if (error != EINTR && error != EIO)
1689 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1690 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1692 * Kludge City: If the first reply to the remove rpc is lost..
1693 * the reply to the retransmitted request will be ENOENT
1694 * since the file was in fact removed
1695 * Therefore, we cheat and return success.
1697 if (error == ENOENT)
1699 } else if (!np->n_sillyrename)
1700 error = nfs_sillyrename(dvp, vp, cnp);
1701 mtx_lock(&np->n_mtx);
1702 np->n_attrstamp = 0;
1703 mtx_unlock(&np->n_mtx);
1704 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1709 * nfs file remove rpc called from nfs_inactive
1712 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1715 * Make sure that the directory vnode is still valid.
1716 * XXX we should lock sp->s_dvp here.
1718 if (sp->s_dvp->v_type == VBAD)
1720 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1725 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1728 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1729 int namelen, struct ucred *cred, struct thread *td)
1731 struct nfsvattr dnfsva;
1732 struct nfsnode *dnp = VTONFS(dvp);
1733 int error = 0, dattrflag;
1735 mtx_lock(&dnp->n_mtx);
1736 dnp->n_flag |= NREMOVEINPROG;
1737 mtx_unlock(&dnp->n_mtx);
1738 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1740 mtx_lock(&dnp->n_mtx);
1741 if ((dnp->n_flag & NREMOVEWANT)) {
1742 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1743 mtx_unlock(&dnp->n_mtx);
1744 wakeup((caddr_t)dnp);
1746 dnp->n_flag &= ~NREMOVEINPROG;
1747 mtx_unlock(&dnp->n_mtx);
1750 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1751 mtx_lock(&dnp->n_mtx);
1752 dnp->n_flag |= NMODIFIED;
1754 dnp->n_attrstamp = 0;
1755 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1757 mtx_unlock(&dnp->n_mtx);
1758 if (error && NFS_ISV4(dvp))
1759 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1764 * nfs file rename call
1767 nfs_rename(struct vop_rename_args *ap)
1769 struct vnode *fvp = ap->a_fvp;
1770 struct vnode *tvp = ap->a_tvp;
1771 struct vnode *fdvp = ap->a_fdvp;
1772 struct vnode *tdvp = ap->a_tdvp;
1773 struct componentname *tcnp = ap->a_tcnp;
1774 struct componentname *fcnp = ap->a_fcnp;
1775 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1776 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1777 struct nfsv4node *newv4 = NULL;
1780 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1781 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1782 /* Check for cross-device rename */
1783 if ((fvp->v_mount != tdvp->v_mount) ||
1784 (tvp && (fvp->v_mount != tvp->v_mount))) {
1790 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1794 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1798 * We have to flush B_DELWRI data prior to renaming
1799 * the file. If we don't, the delayed-write buffers
1800 * can be flushed out later after the file has gone stale
1801 * under NFSV3. NFSV2 does not have this problem because
1802 * ( as far as I can tell ) it flushes dirty buffers more
1805 * Skip the rename operation if the fsync fails, this can happen
1806 * due to the server's volume being full, when we pushed out data
1807 * that was written back to our cache earlier. Not checking for
1808 * this condition can result in potential (silent) data loss.
1810 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1811 NFSVOPUNLOCK(fvp, 0);
1813 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1818 * If the tvp exists and is in use, sillyrename it before doing the
1819 * rename of the new file over it.
1820 * XXX Can't sillyrename a directory.
1822 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1823 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1828 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1829 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1832 if (error == 0 && NFS_ISV4(tdvp)) {
1834 * For NFSv4, check to see if it is the same name and
1835 * replace the name, if it is different.
1837 MALLOC(newv4, struct nfsv4node *,
1838 sizeof (struct nfsv4node) +
1839 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1840 M_NFSV4NODE, M_WAITOK);
1841 mtx_lock(&tdnp->n_mtx);
1842 mtx_lock(&fnp->n_mtx);
1843 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1844 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1845 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1846 tcnp->cn_namelen) ||
1847 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1848 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1849 tdnp->n_fhp->nfh_len))) {
1851 { char nnn[100]; int nnnl;
1852 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1853 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1855 printf("ren replace=%s\n",nnn);
1858 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1861 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1862 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1863 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1864 tdnp->n_fhp->nfh_len);
1865 NFSBCOPY(tcnp->cn_nameptr,
1866 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1868 mtx_unlock(&tdnp->n_mtx);
1869 mtx_unlock(&fnp->n_mtx);
1871 FREE((caddr_t)newv4, M_NFSV4NODE);
1874 if (fvp->v_type == VDIR) {
1875 if (tvp != NULL && tvp->v_type == VDIR)
1890 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1892 if (error == ENOENT)
1898 * nfs file rename rpc called from nfs_remove() above
1901 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1902 struct sillyrename *sp)
1905 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1906 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1911 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1914 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1915 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1916 int tnamelen, struct ucred *cred, struct thread *td)
1918 struct nfsvattr fnfsva, tnfsva;
1919 struct nfsnode *fdnp = VTONFS(fdvp);
1920 struct nfsnode *tdnp = VTONFS(tdvp);
1921 int error = 0, fattrflag, tattrflag;
1923 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1924 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1925 &tattrflag, NULL, NULL);
1926 mtx_lock(&fdnp->n_mtx);
1927 fdnp->n_flag |= NMODIFIED;
1928 if (fattrflag != 0) {
1929 mtx_unlock(&fdnp->n_mtx);
1930 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1932 fdnp->n_attrstamp = 0;
1933 mtx_unlock(&fdnp->n_mtx);
1934 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1936 mtx_lock(&tdnp->n_mtx);
1937 tdnp->n_flag |= NMODIFIED;
1938 if (tattrflag != 0) {
1939 mtx_unlock(&tdnp->n_mtx);
1940 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1942 tdnp->n_attrstamp = 0;
1943 mtx_unlock(&tdnp->n_mtx);
1944 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1946 if (error && NFS_ISV4(fdvp))
1947 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1952 * nfs hard link create call
1955 nfs_link(struct vop_link_args *ap)
1957 struct vnode *vp = ap->a_vp;
1958 struct vnode *tdvp = ap->a_tdvp;
1959 struct componentname *cnp = ap->a_cnp;
1960 struct nfsnode *np, *tdnp;
1961 struct nfsvattr nfsva, dnfsva;
1962 int error = 0, attrflag, dattrflag;
1965 * Push all writes to the server, so that the attribute cache
1966 * doesn't get "out of sync" with the server.
1967 * XXX There should be a better way!
1969 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1971 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1972 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1974 tdnp = VTONFS(tdvp);
1975 mtx_lock(&tdnp->n_mtx);
1976 tdnp->n_flag |= NMODIFIED;
1977 if (dattrflag != 0) {
1978 mtx_unlock(&tdnp->n_mtx);
1979 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1981 tdnp->n_attrstamp = 0;
1982 mtx_unlock(&tdnp->n_mtx);
1983 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1986 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1989 mtx_lock(&np->n_mtx);
1990 np->n_attrstamp = 0;
1991 mtx_unlock(&np->n_mtx);
1992 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1995 * If negative lookup caching is enabled, I might as well
1996 * add an entry for this node. Not necessary for correctness,
1997 * but if negative caching is enabled, then the system
1998 * must care about lookup caching hit rate, so...
2000 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2001 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2002 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2004 if (error && NFS_ISV4(vp))
2005 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2011 * nfs symbolic link create call
2014 nfs_symlink(struct vop_symlink_args *ap)
2016 struct vnode *dvp = ap->a_dvp;
2017 struct vattr *vap = ap->a_vap;
2018 struct componentname *cnp = ap->a_cnp;
2019 struct nfsvattr nfsva, dnfsva;
2021 struct nfsnode *np = NULL, *dnp;
2022 struct vnode *newvp = NULL;
2023 int error = 0, attrflag, dattrflag, ret;
2025 vap->va_type = VLNK;
2026 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2027 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2028 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2030 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2031 &np, NULL, LK_EXCLUSIVE);
2037 if (newvp != NULL) {
2039 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2041 } else if (!error) {
2043 * If we do not have an error and we could not extract the
2044 * newvp from the response due to the request being NFSv2, we
2045 * have to do a lookup in order to obtain a newvp to return.
2047 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2048 cnp->cn_cred, cnp->cn_thread, &np);
2056 error = nfscl_maperr(cnp->cn_thread, error,
2057 vap->va_uid, vap->va_gid);
2063 mtx_lock(&dnp->n_mtx);
2064 dnp->n_flag |= NMODIFIED;
2065 if (dattrflag != 0) {
2066 mtx_unlock(&dnp->n_mtx);
2067 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2069 dnp->n_attrstamp = 0;
2070 mtx_unlock(&dnp->n_mtx);
2071 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2074 * If negative lookup caching is enabled, I might as well
2075 * add an entry for this node. Not necessary for correctness,
2076 * but if negative caching is enabled, then the system
2077 * must care about lookup caching hit rate, so...
2079 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2080 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2081 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2090 nfs_mkdir(struct vop_mkdir_args *ap)
2092 struct vnode *dvp = ap->a_dvp;
2093 struct vattr *vap = ap->a_vap;
2094 struct componentname *cnp = ap->a_cnp;
2095 struct nfsnode *np = NULL, *dnp;
2096 struct vnode *newvp = NULL;
2099 struct nfsvattr nfsva, dnfsva;
2100 int error = 0, attrflag, dattrflag, ret;
2102 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2104 vap->va_type = VDIR;
2105 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2106 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2107 &attrflag, &dattrflag, NULL);
2109 mtx_lock(&dnp->n_mtx);
2110 dnp->n_flag |= NMODIFIED;
2111 if (dattrflag != 0) {
2112 mtx_unlock(&dnp->n_mtx);
2113 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2115 dnp->n_attrstamp = 0;
2116 mtx_unlock(&dnp->n_mtx);
2117 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2120 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2121 &np, NULL, LK_EXCLUSIVE);
2125 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2130 if (!error && newvp == NULL) {
2131 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2132 cnp->cn_cred, cnp->cn_thread, &np);
2135 if (newvp->v_type != VDIR)
2143 error = nfscl_maperr(cnp->cn_thread, error,
2144 vap->va_uid, vap->va_gid);
2147 * If negative lookup caching is enabled, I might as well
2148 * add an entry for this node. Not necessary for correctness,
2149 * but if negative caching is enabled, then the system
2150 * must care about lookup caching hit rate, so...
2152 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2153 (cnp->cn_flags & MAKEENTRY) &&
2154 attrflag != 0 && dattrflag != 0)
2155 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2163 * nfs remove directory call
2166 nfs_rmdir(struct vop_rmdir_args *ap)
2168 struct vnode *vp = ap->a_vp;
2169 struct vnode *dvp = ap->a_dvp;
2170 struct componentname *cnp = ap->a_cnp;
2171 struct nfsnode *dnp;
2172 struct nfsvattr dnfsva;
2173 int error, dattrflag;
2177 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2178 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2180 mtx_lock(&dnp->n_mtx);
2181 dnp->n_flag |= NMODIFIED;
2182 if (dattrflag != 0) {
2183 mtx_unlock(&dnp->n_mtx);
2184 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2186 dnp->n_attrstamp = 0;
2187 mtx_unlock(&dnp->n_mtx);
2188 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2193 if (error && NFS_ISV4(dvp))
2194 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2197 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2199 if (error == ENOENT)
2208 nfs_readdir(struct vop_readdir_args *ap)
2210 struct vnode *vp = ap->a_vp;
2211 struct nfsnode *np = VTONFS(vp);
2212 struct uio *uio = ap->a_uio;
2217 if (ap->a_eofflag != NULL)
2219 if (vp->v_type != VDIR)
2223 * First, check for hit on the EOF offset cache
2225 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2226 (np->n_flag & NMODIFIED) == 0) {
2227 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2228 mtx_lock(&np->n_mtx);
2229 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2230 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2231 mtx_unlock(&np->n_mtx);
2232 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2233 if (ap->a_eofflag != NULL)
2237 mtx_unlock(&np->n_mtx);
2242 * Call ncl_bioread() to do the real work.
2244 tresid = uio->uio_resid;
2245 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2247 if (!error && uio->uio_resid == tresid) {
2248 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2249 if (ap->a_eofflag != NULL)
2257 * Called from below the buffer cache by ncl_doio().
2260 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2263 struct nfsvattr nfsva;
2264 nfsuint64 *cookiep, cookie;
2265 struct nfsnode *dnp = VTONFS(vp);
2266 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2267 int error = 0, eof, attrflag;
2269 KASSERT(uiop->uio_iovcnt == 1 &&
2270 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2271 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2272 ("nfs readdirrpc bad uio"));
2275 * If there is no cookie, assume directory was stale.
2277 ncl_dircookie_lock(dnp);
2278 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2281 ncl_dircookie_unlock(dnp);
2283 ncl_dircookie_unlock(dnp);
2284 return (NFSERR_BAD_COOKIE);
2287 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2288 (void)ncl_fsinfo(nmp, vp, cred, td);
2290 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2291 &attrflag, &eof, NULL);
2293 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2297 * We are now either at the end of the directory or have filled
2301 dnp->n_direofoffset = uiop->uio_offset;
2303 if (uiop->uio_resid > 0)
2304 ncl_printf("EEK! readdirrpc resid > 0\n");
2305 ncl_dircookie_lock(dnp);
2306 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2308 ncl_dircookie_unlock(dnp);
2310 } else if (NFS_ISV4(vp)) {
2311 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2317 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2320 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2323 struct nfsvattr nfsva;
2324 nfsuint64 *cookiep, cookie;
2325 struct nfsnode *dnp = VTONFS(vp);
2326 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2327 int error = 0, attrflag, eof;
2329 KASSERT(uiop->uio_iovcnt == 1 &&
2330 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2331 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2332 ("nfs readdirplusrpc bad uio"));
2335 * If there is no cookie, assume directory was stale.
2337 ncl_dircookie_lock(dnp);
2338 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2341 ncl_dircookie_unlock(dnp);
2343 ncl_dircookie_unlock(dnp);
2344 return (NFSERR_BAD_COOKIE);
2347 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2348 (void)ncl_fsinfo(nmp, vp, cred, td);
2349 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2350 &attrflag, &eof, NULL);
2352 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2356 * We are now either at end of the directory or have filled the
2360 dnp->n_direofoffset = uiop->uio_offset;
2362 if (uiop->uio_resid > 0)
2363 ncl_printf("EEK! readdirplusrpc resid > 0\n");
2364 ncl_dircookie_lock(dnp);
2365 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2367 ncl_dircookie_unlock(dnp);
2369 } else if (NFS_ISV4(vp)) {
2370 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2376 * Silly rename. To make the NFS filesystem that is stateless look a little
2377 * more like the "ufs" a remove of an active vnode is translated to a rename
2378 * to a funny looking filename that is removed by nfs_inactive on the
2379 * nfsnode. There is the potential for another process on a different client
2380 * to create the same funny name between the nfs_lookitup() fails and the
2381 * nfs_rename() completes, but...
2384 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2386 struct sillyrename *sp;
2390 unsigned int lticks;
2394 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2395 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2396 M_NEWNFSREQ, M_WAITOK);
2397 sp->s_cred = crhold(cnp->cn_cred);
2402 * Fudge together a funny name.
2403 * Changing the format of the funny name to accomodate more
2404 * sillynames per directory.
2405 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2406 * CPU ticks since boot.
2408 pid = cnp->cn_thread->td_proc->p_pid;
2409 lticks = (unsigned int)ticks;
2411 sp->s_namlen = sprintf(sp->s_name,
2412 ".nfs.%08x.%04x4.4", lticks,
2414 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2415 cnp->cn_thread, NULL))
2419 error = nfs_renameit(dvp, vp, cnp, sp);
2422 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2423 cnp->cn_thread, &np);
2424 np->n_sillyrename = sp;
2429 free((caddr_t)sp, M_NEWNFSREQ);
2434 * Look up a file name and optionally either update the file handle or
2435 * allocate an nfsnode, depending on the value of npp.
2436 * npp == NULL --> just do the lookup
2437 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2439 * *npp != NULL --> update the file handle in the vnode
2442 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2443 struct thread *td, struct nfsnode **npp)
2445 struct vnode *newvp = NULL, *vp;
2446 struct nfsnode *np, *dnp = VTONFS(dvp);
2447 struct nfsfh *nfhp, *onfhp;
2448 struct nfsvattr nfsva, dnfsva;
2449 struct componentname cn;
2450 int error = 0, attrflag, dattrflag;
2453 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2454 &nfhp, &attrflag, &dattrflag, NULL);
2456 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2457 if (npp && !error) {
2462 * For NFSv4, check to see if it is the same name and
2463 * replace the name, if it is different.
2465 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2466 (np->n_v4->n4_namelen != len ||
2467 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2468 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2469 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2470 dnp->n_fhp->nfh_len))) {
2472 { char nnn[100]; int nnnl;
2473 nnnl = (len < 100) ? len : 99;
2474 bcopy(name, nnn, nnnl);
2476 printf("replace=%s\n",nnn);
2479 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2480 MALLOC(np->n_v4, struct nfsv4node *,
2481 sizeof (struct nfsv4node) +
2482 dnp->n_fhp->nfh_len + len - 1,
2483 M_NFSV4NODE, M_WAITOK);
2484 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2485 np->n_v4->n4_namelen = len;
2486 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2487 dnp->n_fhp->nfh_len);
2488 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2490 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2494 * Rehash node for new file handle.
2496 vfs_hash_rehash(vp, hash);
2499 FREE((caddr_t)onfhp, M_NFSFH);
2501 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2502 FREE((caddr_t)nfhp, M_NFSFH);
2506 cn.cn_nameptr = name;
2507 cn.cn_namelen = len;
2508 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2509 &np, NULL, LK_EXCLUSIVE);
2514 if (!attrflag && *npp == NULL) {
2522 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2525 if (npp && *npp == NULL) {
2536 if (error && NFS_ISV4(dvp))
2537 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2542 * Nfs Version 3 and 4 commit rpc
2545 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2548 struct nfsvattr nfsva;
2549 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2550 int error, attrflag;
2552 mtx_lock(&nmp->nm_mtx);
2553 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2554 mtx_unlock(&nmp->nm_mtx);
2557 mtx_unlock(&nmp->nm_mtx);
2558 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2561 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2563 if (error != 0 && NFS_ISV4(vp))
2564 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2570 * For async requests when nfsiod(s) are running, queue the request by
2571 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2575 nfs_strategy(struct vop_strategy_args *ap)
2577 struct buf *bp = ap->a_bp;
2580 KASSERT(!(bp->b_flags & B_DONE),
2581 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2582 BUF_ASSERT_HELD(bp);
2584 if (bp->b_iocmd == BIO_READ)
2590 * If the op is asynchronous and an i/o daemon is waiting
2591 * queue the request, wake it up and wait for completion
2592 * otherwise just do it ourselves.
2594 if ((bp->b_flags & B_ASYNC) == 0 ||
2595 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2596 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2601 * fsync vnode op. Just call ncl_flush() with commit == 1.
2605 nfs_fsync(struct vop_fsync_args *ap)
2608 if (ap->a_vp->v_type != VREG) {
2610 * For NFS, metadata is changed synchronously on the server,
2611 * so there is nothing to flush. Also, ncl_flush() clears
2612 * the NMODIFIED flag and that shouldn't be done here for
2617 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2621 * Flush all the blocks associated with a vnode.
2622 * Walk through the buffer pool and push any dirty pages
2623 * associated with the vnode.
2624 * If the called_from_renewthread argument is TRUE, it has been called
2625 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2626 * waiting for a buffer write to complete.
2629 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2630 int commit, int called_from_renewthread)
2632 struct nfsnode *np = VTONFS(vp);
2636 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2637 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2638 int passone = 1, trycnt = 0;
2639 u_quad_t off, endoff, toff;
2640 struct ucred* wcred = NULL;
2641 struct buf **bvec = NULL;
2643 #ifndef NFS_COMMITBVECSIZ
2644 #define NFS_COMMITBVECSIZ 20
2646 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2647 int bvecsize = 0, bveccount;
2649 if (called_from_renewthread != 0)
2651 if (nmp->nm_flag & NFSMNT_INT)
2657 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2658 * server, but has not been committed to stable storage on the server
2659 * yet. On the first pass, the byte range is worked out and the commit
2660 * rpc is done. On the second pass, ncl_writebp() is called to do the
2667 if (NFS_ISV34(vp) && commit) {
2668 if (bvec != NULL && bvec != bvec_on_stack)
2671 * Count up how many buffers waiting for a commit.
2675 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2676 if (!BUF_ISLOCKED(bp) &&
2677 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2678 == (B_DELWRI | B_NEEDCOMMIT))
2682 * Allocate space to remember the list of bufs to commit. It is
2683 * important to use M_NOWAIT here to avoid a race with nfs_write.
2684 * If we can't get memory (for whatever reason), we will end up
2685 * committing the buffers one-by-one in the loop below.
2687 if (bveccount > NFS_COMMITBVECSIZ) {
2689 * Release the vnode interlock to avoid a lock
2693 bvec = (struct buf **)
2694 malloc(bveccount * sizeof(struct buf *),
2698 bvec = bvec_on_stack;
2699 bvecsize = NFS_COMMITBVECSIZ;
2701 bvecsize = bveccount;
2703 bvec = bvec_on_stack;
2704 bvecsize = NFS_COMMITBVECSIZ;
2706 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2707 if (bvecpos >= bvecsize)
2709 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2710 nbp = TAILQ_NEXT(bp, b_bobufs);
2713 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2714 (B_DELWRI | B_NEEDCOMMIT)) {
2716 nbp = TAILQ_NEXT(bp, b_bobufs);
2722 * Work out if all buffers are using the same cred
2723 * so we can deal with them all with one commit.
2725 * NOTE: we are not clearing B_DONE here, so we have
2726 * to do it later on in this routine if we intend to
2727 * initiate I/O on the bp.
2729 * Note: to avoid loopback deadlocks, we do not
2730 * assign b_runningbufspace.
2733 wcred = bp->b_wcred;
2734 else if (wcred != bp->b_wcred)
2736 vfs_busy_pages(bp, 1);
2740 * bp is protected by being locked, but nbp is not
2741 * and vfs_busy_pages() may sleep. We have to
2744 nbp = TAILQ_NEXT(bp, b_bobufs);
2747 * A list of these buffers is kept so that the
2748 * second loop knows which buffers have actually
2749 * been committed. This is necessary, since there
2750 * may be a race between the commit rpc and new
2751 * uncommitted writes on the file.
2753 bvec[bvecpos++] = bp;
2754 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2758 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2766 * Commit data on the server, as required.
2767 * If all bufs are using the same wcred, then use that with
2768 * one call for all of them, otherwise commit each one
2771 if (wcred != NOCRED)
2772 retv = ncl_commit(vp, off, (int)(endoff - off),
2776 for (i = 0; i < bvecpos; i++) {
2779 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2781 size = (u_quad_t)(bp->b_dirtyend
2783 retv = ncl_commit(vp, off, (int)size,
2789 if (retv == NFSERR_STALEWRITEVERF)
2790 ncl_clearcommit(vp->v_mount);
2793 * Now, either mark the blocks I/O done or mark the
2794 * blocks dirty, depending on whether the commit
2797 for (i = 0; i < bvecpos; i++) {
2799 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2802 * Error, leave B_DELWRI intact
2804 vfs_unbusy_pages(bp);
2808 * Success, remove B_DELWRI ( bundirty() ).
2810 * b_dirtyoff/b_dirtyend seem to be NFS
2811 * specific. We should probably move that
2812 * into bundirty(). XXX
2815 bp->b_flags |= B_ASYNC;
2817 bp->b_flags &= ~B_DONE;
2818 bp->b_ioflags &= ~BIO_ERROR;
2819 bp->b_dirtyoff = bp->b_dirtyend = 0;
2826 * Start/do any write(s) that are required.
2830 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2831 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2832 if (waitfor != MNT_WAIT || passone)
2835 error = BUF_TIMELOCK(bp,
2836 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2837 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
2842 if (error == ENOLCK) {
2846 if (called_from_renewthread != 0) {
2848 * Return EIO so the flush will be retried
2854 if (newnfs_sigintr(nmp, td)) {
2858 if (slpflag == PCATCH) {
2864 if ((bp->b_flags & B_DELWRI) == 0)
2865 panic("nfs_fsync: not dirty");
2866 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2872 if (passone || !commit)
2873 bp->b_flags |= B_ASYNC;
2875 bp->b_flags |= B_ASYNC;
2877 if (newnfs_sigintr(nmp, td)) {
2888 if (waitfor == MNT_WAIT) {
2889 while (bo->bo_numoutput) {
2890 error = bufobj_wwait(bo, slpflag, slptimeo);
2893 if (called_from_renewthread != 0) {
2895 * Return EIO so that the flush will be
2901 error = newnfs_sigintr(nmp, td);
2904 if (slpflag == PCATCH) {
2911 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2916 * Wait for all the async IO requests to drain
2919 mtx_lock(&np->n_mtx);
2920 while (np->n_directio_asyncwr > 0) {
2921 np->n_flag |= NFSYNCWAIT;
2922 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2923 &np->n_mtx, slpflag | (PRIBIO + 1),
2926 if (newnfs_sigintr(nmp, td)) {
2927 mtx_unlock(&np->n_mtx);
2933 mtx_unlock(&np->n_mtx);
2936 if (NFSHASPNFS(nmp)) {
2937 nfscl_layoutcommit(vp, td);
2939 * Invalidate the attribute cache, since writes to a DS
2940 * won't update the size attribute.
2942 mtx_lock(&np->n_mtx);
2943 np->n_attrstamp = 0;
2945 mtx_lock(&np->n_mtx);
2946 if (np->n_flag & NWRITEERR) {
2947 error = np->n_error;
2948 np->n_flag &= ~NWRITEERR;
2950 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2951 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2952 np->n_flag &= ~NMODIFIED;
2953 mtx_unlock(&np->n_mtx);
2955 if (bvec != NULL && bvec != bvec_on_stack)
2957 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2958 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2959 np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2960 /* try, try again... */
2965 printf("try%d\n", trycnt);
2972 * NFS advisory byte-level locks.
2975 nfs_advlock(struct vop_advlock_args *ap)
2977 struct vnode *vp = ap->a_vp;
2979 struct nfsnode *np = VTONFS(ap->a_vp);
2980 struct proc *p = (struct proc *)ap->a_id;
2981 struct thread *td = curthread; /* XXX */
2983 int ret, error = EOPNOTSUPP;
2986 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
2987 if (vp->v_type != VREG)
2989 if ((ap->a_flags & F_POSIX) != 0)
2992 cred = td->td_ucred;
2993 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2994 if (vp->v_iflag & VI_DOOMED) {
2995 NFSVOPUNLOCK(vp, 0);
3000 * If this is unlocking a write locked region, flush and
3001 * commit them before unlocking. This is required by
3002 * RFC3530 Sec. 9.3.2.
3004 if (ap->a_op == F_UNLCK &&
3005 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3007 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
3010 * Loop around doing the lock op, while a blocking lock
3011 * must wait for the lock op to succeed.
3014 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3015 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3016 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3017 ap->a_op == F_SETLK) {
3018 NFSVOPUNLOCK(vp, 0);
3019 error = nfs_catnap(PZERO | PCATCH, ret,
3023 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3024 if (vp->v_iflag & VI_DOOMED) {
3025 NFSVOPUNLOCK(vp, 0);
3029 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3030 ap->a_op == F_SETLK);
3031 if (ret == NFSERR_DENIED) {
3032 NFSVOPUNLOCK(vp, 0);
3034 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3035 NFSVOPUNLOCK(vp, 0);
3037 } else if (ret != 0) {
3038 NFSVOPUNLOCK(vp, 0);
3043 * Now, if we just got a lock, invalidate data in the buffer
3044 * cache, as required, so that the coherency conforms with
3045 * RFC3530 Sec. 9.3.2.
3047 if (ap->a_op == F_SETLK) {
3048 if ((np->n_flag & NMODIFIED) == 0) {
3049 np->n_attrstamp = 0;
3050 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3051 ret = VOP_GETATTR(vp, &va, cred);
3053 if ((np->n_flag & NMODIFIED) || ret ||
3054 np->n_change != va.va_filerev) {
3055 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3056 np->n_attrstamp = 0;
3057 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3058 ret = VOP_GETATTR(vp, &va, cred);
3060 np->n_mtime = va.va_mtime;
3061 np->n_change = va.va_filerev;
3064 /* Mark that a file lock has been acquired. */
3065 mtx_lock(&np->n_mtx);
3066 np->n_flag |= NHASBEENLOCKED;
3067 mtx_unlock(&np->n_mtx);
3069 NFSVOPUNLOCK(vp, 0);
3071 } else if (!NFS_ISV4(vp)) {
3072 error = NFSVOPLOCK(vp, LK_SHARED);
3075 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3076 size = VTONFS(vp)->n_size;
3077 NFSVOPUNLOCK(vp, 0);
3078 error = lf_advlock(ap, &(vp->v_lockf), size);
3080 if (nfs_advlock_p != NULL)
3081 error = nfs_advlock_p(ap);
3083 NFSVOPUNLOCK(vp, 0);
3087 if (error == 0 && ap->a_op == F_SETLK) {
3088 /* Mark that a file lock has been acquired. */
3089 mtx_lock(&np->n_mtx);
3090 np->n_flag |= NHASBEENLOCKED;
3091 mtx_unlock(&np->n_mtx);
3098 * NFS advisory byte-level locks.
3101 nfs_advlockasync(struct vop_advlockasync_args *ap)
3103 struct vnode *vp = ap->a_vp;
3108 return (EOPNOTSUPP);
3109 error = NFSVOPLOCK(vp, LK_SHARED);
3112 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3113 size = VTONFS(vp)->n_size;
3114 NFSVOPUNLOCK(vp, 0);
3115 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3117 NFSVOPUNLOCK(vp, 0);
3124 * Print out the contents of an nfsnode.
3127 nfs_print(struct vop_print_args *ap)
3129 struct vnode *vp = ap->a_vp;
3130 struct nfsnode *np = VTONFS(vp);
3132 ncl_printf("\tfileid %ld fsid 0x%x",
3133 np->n_vattr.na_fileid, np->n_vattr.na_fsid);
3134 if (vp->v_type == VFIFO)
3141 * This is the "real" nfs::bwrite(struct buf*).
3142 * We set B_CACHE if this is a VMIO buffer.
3145 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3148 int oldflags = bp->b_flags;
3154 BUF_ASSERT_HELD(bp);
3156 if (bp->b_flags & B_INVAL) {
3161 bp->b_flags |= B_CACHE;
3164 * Undirty the bp. We will redirty it later if the I/O fails.
3169 bp->b_flags &= ~B_DONE;
3170 bp->b_ioflags &= ~BIO_ERROR;
3171 bp->b_iocmd = BIO_WRITE;
3173 bufobj_wref(bp->b_bufobj);
3174 curthread->td_ru.ru_oublock++;
3178 * Note: to avoid loopback deadlocks, we do not
3179 * assign b_runningbufspace.
3181 vfs_busy_pages(bp, 1);
3184 bp->b_iooffset = dbtob(bp->b_blkno);
3187 if( (oldflags & B_ASYNC) == 0) {
3188 int rtval = bufwait(bp);
3190 if (oldflags & B_DELWRI) {
3203 * nfs special file access vnode op.
3204 * Essentially just get vattr and then imitate iaccess() since the device is
3205 * local to the client.
3208 nfsspec_access(struct vop_access_args *ap)
3211 struct ucred *cred = ap->a_cred;
3212 struct vnode *vp = ap->a_vp;
3213 accmode_t accmode = ap->a_accmode;
3218 * Disallow write attempts on filesystems mounted read-only;
3219 * unless the file is a socket, fifo, or a block or character
3220 * device resident on the filesystem.
3222 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3223 switch (vp->v_type) {
3233 error = VOP_GETATTR(vp, vap, cred);
3236 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3237 accmode, cred, NULL);
3243 * Read wrapper for fifos.
3246 nfsfifo_read(struct vop_read_args *ap)
3248 struct nfsnode *np = VTONFS(ap->a_vp);
3254 mtx_lock(&np->n_mtx);
3256 vfs_timestamp(&np->n_atim);
3257 mtx_unlock(&np->n_mtx);
3258 error = fifo_specops.vop_read(ap);
3263 * Write wrapper for fifos.
3266 nfsfifo_write(struct vop_write_args *ap)
3268 struct nfsnode *np = VTONFS(ap->a_vp);
3273 mtx_lock(&np->n_mtx);
3275 vfs_timestamp(&np->n_mtim);
3276 mtx_unlock(&np->n_mtx);
3277 return(fifo_specops.vop_write(ap));
3281 * Close wrapper for fifos.
3283 * Update the times on the nfsnode then do fifo close.
3286 nfsfifo_close(struct vop_close_args *ap)
3288 struct vnode *vp = ap->a_vp;
3289 struct nfsnode *np = VTONFS(vp);
3293 mtx_lock(&np->n_mtx);
3294 if (np->n_flag & (NACC | NUPD)) {
3296 if (np->n_flag & NACC)
3298 if (np->n_flag & NUPD)
3301 if (vrefcnt(vp) == 1 &&
3302 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3304 if (np->n_flag & NACC)
3305 vattr.va_atime = np->n_atim;
3306 if (np->n_flag & NUPD)
3307 vattr.va_mtime = np->n_mtim;
3308 mtx_unlock(&np->n_mtx);
3309 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3313 mtx_unlock(&np->n_mtx);
3315 return (fifo_specops.vop_close(ap));
3319 * Just call ncl_writebp() with the force argument set to 1.
3321 * NOTE: B_DONE may or may not be set in a_bp on call.
3324 nfs_bwrite(struct buf *bp)
3327 return (ncl_writebp(bp, 1, curthread));
3330 struct buf_ops buf_ops_newnfs = {
3331 .bop_name = "buf_ops_nfs",
3332 .bop_write = nfs_bwrite,
3333 .bop_strategy = bufstrategy,
3334 .bop_sync = bufsync,
3335 .bop_bdflush = bufbdflush,
3339 * Cloned from vop_stdlock(), and then the ugly hack added.
3342 nfs_lock1(struct vop_lock1_args *ap)
3344 struct vnode *vp = ap->a_vp;
3348 * Since vfs_hash_get() calls vget() and it will no longer work
3349 * for FreeBSD8 with flags == 0, I can only think of this horrible
3350 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3351 * and then handle it here. All I want for this case is a v_usecount
3352 * on the vnode to use for recovery, while another thread might
3353 * hold a lock on the vnode. I have the other threads blocked, so
3354 * there isn't any race problem.
3356 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3357 if ((ap->a_flags & LK_INTERLOCK) == 0)
3359 if ((vp->v_iflag & VI_DOOMED))
3364 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3365 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3370 nfs_getacl(struct vop_getacl_args *ap)
3374 if (ap->a_type != ACL_TYPE_NFS4)
3375 return (EOPNOTSUPP);
3376 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3378 if (error > NFSERR_STALE) {
3379 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3386 nfs_setacl(struct vop_setacl_args *ap)
3390 if (ap->a_type != ACL_TYPE_NFS4)
3391 return (EOPNOTSUPP);
3392 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3394 if (error > NFSERR_STALE) {
3395 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3402 * Return POSIX pathconf information applicable to nfs filesystems.
3405 nfs_pathconf(struct vop_pathconf_args *ap)
3407 struct nfsv3_pathconf pc;
3408 struct nfsvattr nfsva;
3409 struct vnode *vp = ap->a_vp;
3410 struct thread *td = curthread;
3411 int attrflag, error;
3413 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
3414 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3415 ap->a_name == _PC_NO_TRUNC)) ||
3416 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
3418 * Since only the above 4 a_names are returned by the NFSv3
3419 * Pathconf RPC, there is no point in doing it for others.
3420 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
3421 * be used for _PC_NFS4_ACL as well.
3423 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3426 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3432 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3435 pc.pc_linkmax = LINK_MAX;
3436 pc.pc_namemax = NFS_MAXNAMLEN;
3438 pc.pc_chownrestricted = 1;
3439 pc.pc_caseinsensitive = 0;
3440 pc.pc_casepreserving = 1;
3443 switch (ap->a_name) {
3445 *ap->a_retval = pc.pc_linkmax;
3448 *ap->a_retval = pc.pc_namemax;
3451 *ap->a_retval = PATH_MAX;
3454 *ap->a_retval = PIPE_BUF;
3456 case _PC_CHOWN_RESTRICTED:
3457 *ap->a_retval = pc.pc_chownrestricted;
3460 *ap->a_retval = pc.pc_notrunc;
3462 case _PC_ACL_EXTENDED:
3466 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3467 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3472 case _PC_ACL_PATH_MAX:
3474 *ap->a_retval = ACL_MAX_ENTRIES;
3478 case _PC_MAC_PRESENT:
3482 /* _PC_ASYNC_IO should have been handled by upper layers. */
3483 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3492 case _PC_ALLOC_SIZE_MIN:
3493 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3495 case _PC_FILESIZEBITS:
3501 case _PC_REC_INCR_XFER_SIZE:
3502 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3504 case _PC_REC_MAX_XFER_SIZE:
3505 *ap->a_retval = -1; /* means ``unlimited'' */
3507 case _PC_REC_MIN_XFER_SIZE:
3508 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3510 case _PC_REC_XFER_ALIGN:
3511 *ap->a_retval = PAGE_SIZE;
3513 case _PC_SYMLINK_MAX:
3514 *ap->a_retval = NFS_MAXPATHLEN;