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 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++;
610 * If this is an open for writing, capture a reference to the
611 * credentials, so they can be used by ncl_putpages(). Using
612 * these write credentials is preferable to the credentials of
613 * whatever thread happens to be doing the VOP_PUTPAGES() since
614 * the write RPCs are less likely to fail with EACCES.
616 if ((fmode & FWRITE) != 0) {
617 cred = np->n_writecred;
618 np->n_writecred = crhold(ap->a_cred);
621 mtx_unlock(&np->n_mtx);
624 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
630 * What an NFS client should do upon close after writing is a debatable issue.
631 * Most NFS clients push delayed writes to the server upon close, basically for
633 * 1 - So that any write errors may be reported back to the client process
634 * doing the close system call. By far the two most likely errors are
635 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
636 * 2 - To put a worst case upper bound on cache inconsistency between
637 * multiple clients for the file.
638 * There is also a consistency problem for Version 2 of the protocol w.r.t.
639 * not being able to tell if other clients are writing a file concurrently,
640 * since there is no way of knowing if the changed modify time in the reply
641 * is only due to the write for this client.
642 * (NFS Version 3 provides weak cache consistency data in the reply that
643 * should be sufficient to detect and handle this case.)
645 * The current code does the following:
646 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
647 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
648 * or commit them (this satisfies 1 and 2 except for the
649 * case where the server crashes after this close but
650 * before the commit RPC, which is felt to be "good
651 * enough". Changing the last argument to ncl_flush() to
652 * a 1 would force a commit operation, if it is felt a
653 * commit is necessary now.
654 * for NFS Version 4 - flush the dirty buffers and commit them, if
655 * nfscl_mustflush() says this is necessary.
656 * It is necessary if there is no write delegation held,
657 * in order to satisfy open/close coherency.
658 * If the file isn't cached on local stable storage,
659 * it may be necessary in order to detect "out of space"
660 * errors from the server, if the write delegation
661 * issued by the server doesn't allow the file to grow.
665 nfs_close(struct vop_close_args *ap)
667 struct vnode *vp = ap->a_vp;
668 struct nfsnode *np = VTONFS(vp);
669 struct nfsvattr nfsva;
671 int error = 0, ret, localcred = 0;
672 int fmode = ap->a_fflag;
674 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
677 * During shutdown, a_cred isn't valid, so just use root.
679 if (ap->a_cred == NOCRED) {
680 cred = newnfs_getcred();
685 if (vp->v_type == VREG) {
687 * Examine and clean dirty pages, regardless of NMODIFIED.
688 * This closes a major hole in close-to-open consistency.
689 * We want to push out all dirty pages (and buffers) on
690 * close, regardless of whether they were dirtied by
691 * mmap'ed writes or via write().
693 if (nfs_clean_pages_on_close && vp->v_object) {
694 VM_OBJECT_LOCK(vp->v_object);
695 vm_object_page_clean(vp->v_object, 0, 0, 0);
696 VM_OBJECT_UNLOCK(vp->v_object);
698 mtx_lock(&np->n_mtx);
699 if (np->n_flag & NMODIFIED) {
700 mtx_unlock(&np->n_mtx);
703 * Under NFSv3 we have dirty buffers to dispose of. We
704 * must flush them to the NFS server. We have the option
705 * of waiting all the way through the commit rpc or just
706 * waiting for the initial write. The default is to only
707 * wait through the initial write so the data is in the
708 * server's cache, which is roughly similar to the state
709 * a standard disk subsystem leaves the file in on close().
711 * We cannot clear the NMODIFIED bit in np->n_flag due to
712 * potential races with other processes, and certainly
713 * cannot clear it if we don't commit.
714 * These races occur when there is no longer the old
715 * traditional vnode locking implemented for Vnode Ops.
717 int cm = newnfs_commit_on_close ? 1 : 0;
718 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
719 /* np->n_flag &= ~NMODIFIED; */
720 } else if (NFS_ISV4(vp)) {
721 if (nfscl_mustflush(vp) != 0) {
722 int cm = newnfs_commit_on_close ? 1 : 0;
723 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
726 * as above w.r.t races when clearing
728 * np->n_flag &= ~NMODIFIED;
732 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
733 mtx_lock(&np->n_mtx);
736 * Invalidate the attribute cache in all cases.
737 * An open is going to fetch fresh attrs any way, other procs
738 * on this node that have file open will be forced to do an
739 * otw attr fetch, but this is safe.
740 * --> A user found that their RPC count dropped by 20% when
741 * this was commented out and I can't see any requirement
742 * for it, so I've disabled it when negative lookups are
743 * enabled. (What does this have to do with negative lookup
744 * caching? Well nothing, except it was reported by the
745 * same user that needed negative lookup caching and I wanted
746 * there to be a way to disable it to see if it
747 * is the cause of some caching/coherency issue that might
750 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
752 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
754 if (np->n_flag & NWRITEERR) {
755 np->n_flag &= ~NWRITEERR;
758 mtx_unlock(&np->n_mtx);
763 * Get attributes so "change" is up to date.
765 if (error == 0 && nfscl_mustflush(vp) != 0 &&
766 vp->v_type == VREG &&
767 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
768 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
771 np->n_change = nfsva.na_filerev;
772 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
780 ret = nfsrpc_close(vp, 0, ap->a_td);
784 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
787 if (newnfs_directio_enable)
788 KASSERT((np->n_directio_asyncwr == 0),
789 ("nfs_close: dirty unflushed (%d) directio buffers\n",
790 np->n_directio_asyncwr));
791 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
792 mtx_lock(&np->n_mtx);
793 KASSERT((np->n_directio_opens > 0),
794 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
795 np->n_directio_opens--;
796 if (np->n_directio_opens == 0)
797 np->n_flag &= ~NNONCACHE;
798 mtx_unlock(&np->n_mtx);
806 * nfs getattr call from vfs.
809 nfs_getattr(struct vop_getattr_args *ap)
811 struct vnode *vp = ap->a_vp;
812 struct thread *td = curthread; /* XXX */
813 struct nfsnode *np = VTONFS(vp);
815 struct nfsvattr nfsva;
816 struct vattr *vap = ap->a_vap;
820 * Update local times for special files.
822 mtx_lock(&np->n_mtx);
823 if (np->n_flag & (NACC | NUPD))
825 mtx_unlock(&np->n_mtx);
827 * First look in the cache.
829 if (ncl_getattrcache(vp, &vattr) == 0) {
830 vap->va_type = vattr.va_type;
831 vap->va_mode = vattr.va_mode;
832 vap->va_nlink = vattr.va_nlink;
833 vap->va_uid = vattr.va_uid;
834 vap->va_gid = vattr.va_gid;
835 vap->va_fsid = vattr.va_fsid;
836 vap->va_fileid = vattr.va_fileid;
837 vap->va_size = vattr.va_size;
838 vap->va_blocksize = vattr.va_blocksize;
839 vap->va_atime = vattr.va_atime;
840 vap->va_mtime = vattr.va_mtime;
841 vap->va_ctime = vattr.va_ctime;
842 vap->va_gen = vattr.va_gen;
843 vap->va_flags = vattr.va_flags;
844 vap->va_rdev = vattr.va_rdev;
845 vap->va_bytes = vattr.va_bytes;
846 vap->va_filerev = vattr.va_filerev;
848 * Get the local modify time for the case of a write
851 nfscl_deleggetmodtime(vp, &vap->va_mtime);
855 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
856 nfsaccess_cache_timeout > 0) {
857 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
858 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
859 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
860 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
864 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
866 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
869 * Get the local modify time for the case of a write
872 nfscl_deleggetmodtime(vp, &vap->va_mtime);
873 } else if (NFS_ISV4(vp)) {
874 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
883 nfs_setattr(struct vop_setattr_args *ap)
885 struct vnode *vp = ap->a_vp;
886 struct nfsnode *np = VTONFS(vp);
887 struct thread *td = curthread; /* XXX */
888 struct vattr *vap = ap->a_vap;
897 * Setting of flags and marking of atimes are not supported.
899 if (vap->va_flags != VNOVAL)
903 * Disallow write attempts if the filesystem is mounted read-only.
905 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
906 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
907 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
908 (vp->v_mount->mnt_flag & MNT_RDONLY))
910 if (vap->va_size != VNOVAL) {
911 switch (vp->v_type) {
918 if (vap->va_mtime.tv_sec == VNOVAL &&
919 vap->va_atime.tv_sec == VNOVAL &&
920 vap->va_mode == (mode_t)VNOVAL &&
921 vap->va_uid == (uid_t)VNOVAL &&
922 vap->va_gid == (gid_t)VNOVAL)
924 vap->va_size = VNOVAL;
928 * Disallow write attempts if the filesystem is
931 if (vp->v_mount->mnt_flag & MNT_RDONLY)
934 * We run vnode_pager_setsize() early (why?),
935 * we must set np->n_size now to avoid vinvalbuf
936 * V_SAVE races that might setsize a lower
939 mtx_lock(&np->n_mtx);
941 mtx_unlock(&np->n_mtx);
942 error = ncl_meta_setsize(vp, ap->a_cred, td,
944 mtx_lock(&np->n_mtx);
945 if (np->n_flag & NMODIFIED) {
947 mtx_unlock(&np->n_mtx);
948 if (vap->va_size == 0)
949 error = ncl_vinvalbuf(vp, 0, td, 1);
951 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
953 vnode_pager_setsize(vp, tsize);
957 * Call nfscl_delegmodtime() to set the modify time
958 * locally, as required.
960 nfscl_delegmodtime(vp);
962 mtx_unlock(&np->n_mtx);
964 * np->n_size has already been set to vap->va_size
965 * in ncl_meta_setsize(). We must set it again since
966 * nfs_loadattrcache() could be called through
967 * ncl_meta_setsize() and could modify np->n_size.
969 mtx_lock(&np->n_mtx);
970 np->n_vattr.na_size = np->n_size = vap->va_size;
971 mtx_unlock(&np->n_mtx);
974 mtx_lock(&np->n_mtx);
975 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
976 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
977 mtx_unlock(&np->n_mtx);
978 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
979 (error == EINTR || error == EIO))
982 mtx_unlock(&np->n_mtx);
984 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
985 if (error && vap->va_size != VNOVAL) {
986 mtx_lock(&np->n_mtx);
987 np->n_size = np->n_vattr.na_size = tsize;
988 vnode_pager_setsize(vp, tsize);
989 mtx_unlock(&np->n_mtx);
995 * Do an nfs setattr rpc.
998 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1001 struct nfsnode *np = VTONFS(vp);
1002 int error, ret, attrflag, i;
1003 struct nfsvattr nfsva;
1005 if (NFS_ISV34(vp)) {
1006 mtx_lock(&np->n_mtx);
1007 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1008 np->n_accesscache[i].stamp = 0;
1009 np->n_flag |= NDELEGMOD;
1010 mtx_unlock(&np->n_mtx);
1011 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1013 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1016 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1020 if (error && NFS_ISV4(vp))
1021 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1026 * nfs lookup call, one step at a time...
1027 * First look in cache
1028 * If not found, unlock the directory nfsnode and do the rpc
1031 nfs_lookup(struct vop_lookup_args *ap)
1033 struct componentname *cnp = ap->a_cnp;
1034 struct vnode *dvp = ap->a_dvp;
1035 struct vnode **vpp = ap->a_vpp;
1036 struct mount *mp = dvp->v_mount;
1037 int flags = cnp->cn_flags;
1038 struct vnode *newvp;
1039 struct nfsmount *nmp;
1040 struct nfsnode *np, *newnp;
1041 int error = 0, attrflag, dattrflag, ltype, ncticks;
1042 struct thread *td = cnp->cn_thread;
1044 struct nfsvattr dnfsva, nfsva;
1046 struct timespec nctime;
1049 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1050 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1052 if (dvp->v_type != VDIR)
1057 /* For NFSv4, wait until any remove is done. */
1058 mtx_lock(&np->n_mtx);
1059 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1060 np->n_flag |= NREMOVEWANT;
1061 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1063 mtx_unlock(&np->n_mtx);
1065 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1067 error = cache_lookup_times(dvp, vpp, cnp, &nctime, &ncticks);
1068 if (error > 0 && error != ENOENT)
1072 * Lookups of "." are special and always return the
1073 * current directory. cache_lookup() already handles
1074 * associated locking bookkeeping, etc.
1076 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1077 /* XXX: Is this really correct? */
1078 if (cnp->cn_nameiop != LOOKUP &&
1080 cnp->cn_flags |= SAVENAME;
1085 * We only accept a positive hit in the cache if the
1086 * change time of the file matches our cached copy.
1087 * Otherwise, we discard the cache entry and fallback
1088 * to doing a lookup RPC. We also only trust cache
1089 * entries for less than nm_nametimeo seconds.
1091 * To better handle stale file handles and attributes,
1092 * clear the attribute cache of this node if it is a
1093 * leaf component, part of an open() call, and not
1094 * locally modified before fetching the attributes.
1095 * This should allow stale file handles to be detected
1096 * here where we can fall back to a LOOKUP RPC to
1097 * recover rather than having nfs_open() detect the
1098 * stale file handle and failing open(2) with ESTALE.
1101 newnp = VTONFS(newvp);
1102 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1103 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1104 !(newnp->n_flag & NMODIFIED)) {
1105 mtx_lock(&newnp->n_mtx);
1106 newnp->n_attrstamp = 0;
1107 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1108 mtx_unlock(&newnp->n_mtx);
1110 if (nfscl_nodeleg(newvp, 0) == 0 ||
1111 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1112 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1113 timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1114 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1115 if (cnp->cn_nameiop != LOOKUP &&
1117 cnp->cn_flags |= SAVENAME;
1126 } else if (error == ENOENT) {
1127 if (dvp->v_iflag & VI_DOOMED)
1130 * We only accept a negative hit in the cache if the
1131 * modification time of the parent directory matches
1132 * the cached copy in the name cache entry.
1133 * Otherwise, we discard all of the negative cache
1134 * entries for this directory. We also only trust
1135 * negative cache entries for up to nm_negnametimeo
1138 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1139 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1140 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1141 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1144 cache_purge_negative(dvp);
1149 NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1150 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1151 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1154 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1156 if (newvp != NULLVP) {
1161 if (error != ENOENT) {
1163 error = nfscl_maperr(td, error, (uid_t)0,
1168 /* The requested file was not found. */
1169 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1170 (flags & ISLASTCN)) {
1172 * XXX: UFS does a full VOP_ACCESS(dvp,
1173 * VWRITE) here instead of just checking
1176 if (mp->mnt_flag & MNT_RDONLY)
1178 cnp->cn_flags |= SAVENAME;
1179 return (EJUSTRETURN);
1182 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
1185 * Cache the modification time of the parent
1186 * directory from the post-op attributes in
1187 * the name cache entry. The negative cache
1188 * entry will be ignored once the directory
1189 * has changed. Don't bother adding the entry
1190 * if the directory has already changed.
1192 mtx_lock(&np->n_mtx);
1193 if (timespeccmp(&np->n_vattr.na_mtime,
1194 &dnfsva.na_mtime, ==)) {
1195 mtx_unlock(&np->n_mtx);
1196 cache_enter_time(dvp, NULL, cnp,
1197 &dnfsva.na_mtime, NULL);
1199 mtx_unlock(&np->n_mtx);
1205 * Handle RENAME case...
1207 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1208 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1209 FREE((caddr_t)nfhp, M_NFSFH);
1212 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1218 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1221 cnp->cn_flags |= SAVENAME;
1225 if (flags & ISDOTDOT) {
1226 ltype = NFSVOPISLOCKED(dvp);
1227 error = vfs_busy(mp, MBF_NOWAIT);
1230 NFSVOPUNLOCK(dvp, 0);
1231 error = vfs_busy(mp, 0);
1232 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1234 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1241 NFSVOPUNLOCK(dvp, 0);
1242 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1248 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1249 if (dvp->v_iflag & VI_DOOMED) {
1261 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1263 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1264 FREE((caddr_t)nfhp, M_NFSFH);
1268 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1271 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1277 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1279 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1280 !(np->n_flag & NMODIFIED)) {
1282 * Flush the attribute cache when opening a
1283 * leaf node to ensure that fresh attributes
1284 * are fetched in nfs_open() since we did not
1285 * fetch attributes from the LOOKUP reply.
1287 mtx_lock(&np->n_mtx);
1288 np->n_attrstamp = 0;
1289 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1290 mtx_unlock(&np->n_mtx);
1293 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1294 cnp->cn_flags |= SAVENAME;
1295 if ((cnp->cn_flags & MAKEENTRY) &&
1296 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1297 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1298 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1299 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1306 * Just call ncl_bioread() to do the work.
1309 nfs_read(struct vop_read_args *ap)
1311 struct vnode *vp = ap->a_vp;
1313 switch (vp->v_type) {
1315 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1319 return (EOPNOTSUPP);
1327 nfs_readlink(struct vop_readlink_args *ap)
1329 struct vnode *vp = ap->a_vp;
1331 if (vp->v_type != VLNK)
1333 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1337 * Do a readlink rpc.
1338 * Called by ncl_doio() from below the buffer cache.
1341 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1343 int error, ret, attrflag;
1344 struct nfsvattr nfsva;
1346 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1349 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1353 if (error && NFS_ISV4(vp))
1354 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1363 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1365 int error, ret, attrflag;
1366 struct nfsvattr nfsva;
1368 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1371 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1375 if (error && NFS_ISV4(vp))
1376 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1384 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1385 int *iomode, int *must_commit, int called_from_strategy)
1387 struct nfsvattr nfsva;
1388 int error = 0, attrflag, ret;
1390 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1391 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy);
1393 if (VTONFS(vp)->n_flag & ND_NFSV4)
1394 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1397 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1403 *iomode = NFSWRITE_FILESYNC;
1404 if (error && NFS_ISV4(vp))
1405 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1411 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1412 * mode set to specify the file type and the size field for rdev.
1415 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1418 struct nfsvattr nfsva, dnfsva;
1419 struct vnode *newvp = NULL;
1420 struct nfsnode *np = NULL, *dnp;
1423 int error = 0, attrflag, dattrflag;
1426 if (vap->va_type == VCHR || vap->va_type == VBLK)
1427 rdev = vap->va_rdev;
1428 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1431 return (EOPNOTSUPP);
1432 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1434 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1435 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1436 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1439 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1440 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1441 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1444 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1445 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1448 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1451 if (attrflag != 0) {
1452 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1460 } else if (NFS_ISV4(dvp)) {
1461 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1465 mtx_lock(&dnp->n_mtx);
1466 dnp->n_flag |= NMODIFIED;
1468 dnp->n_attrstamp = 0;
1469 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1471 mtx_unlock(&dnp->n_mtx);
1477 * just call nfs_mknodrpc() to do the work.
1481 nfs_mknod(struct vop_mknod_args *ap)
1483 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1486 static struct mtx nfs_cverf_mtx;
1487 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1493 static nfsquad_t cverf;
1495 static int cverf_initialized = 0;
1497 mtx_lock(&nfs_cverf_mtx);
1498 if (cverf_initialized == 0) {
1499 cverf.lval[0] = arc4random();
1500 cverf.lval[1] = arc4random();
1501 cverf_initialized = 1;
1505 mtx_unlock(&nfs_cverf_mtx);
1511 * nfs file create call
1514 nfs_create(struct vop_create_args *ap)
1516 struct vnode *dvp = ap->a_dvp;
1517 struct vattr *vap = ap->a_vap;
1518 struct componentname *cnp = ap->a_cnp;
1519 struct nfsnode *np = NULL, *dnp;
1520 struct vnode *newvp = NULL;
1521 struct nfsmount *nmp;
1522 struct nfsvattr dnfsva, nfsva;
1525 int error = 0, attrflag, dattrflag, fmode = 0;
1529 * Oops, not for me..
1531 if (vap->va_type == VSOCK)
1532 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1534 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1536 if (vap->va_vaflags & VA_EXCLUSIVE)
1539 nmp = VFSTONFS(vnode_mount(dvp));
1541 /* For NFSv4, wait until any remove is done. */
1542 mtx_lock(&dnp->n_mtx);
1543 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1544 dnp->n_flag |= NREMOVEWANT;
1545 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1547 mtx_unlock(&dnp->n_mtx);
1549 cverf = nfs_get_cverf();
1550 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1551 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1552 &nfhp, &attrflag, &dattrflag, NULL);
1555 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1556 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1557 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1560 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1561 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1564 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1568 error = nfsrpc_getattr(newvp, cnp->cn_cred,
1569 cnp->cn_thread, &nfsva, NULL);
1571 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1575 if (newvp != NULL) {
1579 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1580 error == NFSERR_NOTSUPP) {
1584 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1585 if (nfscl_checksattr(vap, &nfsva)) {
1587 * We are normally called with only a partially
1588 * initialized VAP. Since the NFSv3 spec says that
1589 * the server may use the file attributes to
1590 * store the verifier, the spec requires us to do a
1591 * SETATTR RPC. FreeBSD servers store the verifier in
1592 * atime, but we can't really assume that all servers
1593 * will so we ensure that our SETATTR sets both atime
1596 if (vap->va_mtime.tv_sec == VNOVAL)
1597 vfs_timestamp(&vap->va_mtime);
1598 if (vap->va_atime.tv_sec == VNOVAL)
1599 vap->va_atime = vap->va_mtime;
1600 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1601 cnp->cn_thread, &nfsva, &attrflag, NULL);
1602 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1603 vap->va_gid != (gid_t)VNOVAL)) {
1604 /* try again without setting uid/gid */
1605 vap->va_uid = (uid_t)VNOVAL;
1606 vap->va_gid = (uid_t)VNOVAL;
1607 error = nfsrpc_setattr(newvp, vap, NULL,
1608 cnp->cn_cred, cnp->cn_thread, &nfsva,
1612 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1619 if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1620 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1623 } else if (NFS_ISV4(dvp)) {
1624 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1627 mtx_lock(&dnp->n_mtx);
1628 dnp->n_flag |= NMODIFIED;
1630 dnp->n_attrstamp = 0;
1631 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1633 mtx_unlock(&dnp->n_mtx);
1638 * nfs file remove call
1639 * To try and make nfs semantics closer to ufs semantics, a file that has
1640 * other processes using the vnode is renamed instead of removed and then
1641 * removed later on the last close.
1642 * - If v_usecount > 1
1643 * If a rename is not already in the works
1644 * call nfs_sillyrename() to set it up
1649 nfs_remove(struct vop_remove_args *ap)
1651 struct vnode *vp = ap->a_vp;
1652 struct vnode *dvp = ap->a_dvp;
1653 struct componentname *cnp = ap->a_cnp;
1654 struct nfsnode *np = VTONFS(vp);
1658 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1659 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1660 if (vp->v_type == VDIR)
1662 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1663 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1664 vattr.va_nlink > 1)) {
1666 * Purge the name cache so that the chance of a lookup for
1667 * the name succeeding while the remove is in progress is
1668 * minimized. Without node locking it can still happen, such
1669 * that an I/O op returns ESTALE, but since you get this if
1670 * another host removes the file..
1674 * throw away biocache buffers, mainly to avoid
1675 * unnecessary delayed writes later.
1677 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1679 if (error != EINTR && error != EIO)
1680 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1681 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1683 * Kludge City: If the first reply to the remove rpc is lost..
1684 * the reply to the retransmitted request will be ENOENT
1685 * since the file was in fact removed
1686 * Therefore, we cheat and return success.
1688 if (error == ENOENT)
1690 } else if (!np->n_sillyrename)
1691 error = nfs_sillyrename(dvp, vp, cnp);
1692 mtx_lock(&np->n_mtx);
1693 np->n_attrstamp = 0;
1694 mtx_unlock(&np->n_mtx);
1695 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1700 * nfs file remove rpc called from nfs_inactive
1703 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1706 * Make sure that the directory vnode is still valid.
1707 * XXX we should lock sp->s_dvp here.
1709 if (sp->s_dvp->v_type == VBAD)
1711 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1716 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1719 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1720 int namelen, struct ucred *cred, struct thread *td)
1722 struct nfsvattr dnfsva;
1723 struct nfsnode *dnp = VTONFS(dvp);
1724 int error = 0, dattrflag;
1726 mtx_lock(&dnp->n_mtx);
1727 dnp->n_flag |= NREMOVEINPROG;
1728 mtx_unlock(&dnp->n_mtx);
1729 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1731 mtx_lock(&dnp->n_mtx);
1732 if ((dnp->n_flag & NREMOVEWANT)) {
1733 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1734 mtx_unlock(&dnp->n_mtx);
1735 wakeup((caddr_t)dnp);
1737 dnp->n_flag &= ~NREMOVEINPROG;
1738 mtx_unlock(&dnp->n_mtx);
1741 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1742 mtx_lock(&dnp->n_mtx);
1743 dnp->n_flag |= NMODIFIED;
1745 dnp->n_attrstamp = 0;
1746 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1748 mtx_unlock(&dnp->n_mtx);
1749 if (error && NFS_ISV4(dvp))
1750 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1755 * nfs file rename call
1758 nfs_rename(struct vop_rename_args *ap)
1760 struct vnode *fvp = ap->a_fvp;
1761 struct vnode *tvp = ap->a_tvp;
1762 struct vnode *fdvp = ap->a_fdvp;
1763 struct vnode *tdvp = ap->a_tdvp;
1764 struct componentname *tcnp = ap->a_tcnp;
1765 struct componentname *fcnp = ap->a_fcnp;
1766 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1767 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1768 struct nfsv4node *newv4 = NULL;
1771 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1772 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1773 /* Check for cross-device rename */
1774 if ((fvp->v_mount != tdvp->v_mount) ||
1775 (tvp && (fvp->v_mount != tvp->v_mount))) {
1781 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1785 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1789 * We have to flush B_DELWRI data prior to renaming
1790 * the file. If we don't, the delayed-write buffers
1791 * can be flushed out later after the file has gone stale
1792 * under NFSV3. NFSV2 does not have this problem because
1793 * ( as far as I can tell ) it flushes dirty buffers more
1796 * Skip the rename operation if the fsync fails, this can happen
1797 * due to the server's volume being full, when we pushed out data
1798 * that was written back to our cache earlier. Not checking for
1799 * this condition can result in potential (silent) data loss.
1801 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1802 NFSVOPUNLOCK(fvp, 0);
1804 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1809 * If the tvp exists and is in use, sillyrename it before doing the
1810 * rename of the new file over it.
1811 * XXX Can't sillyrename a directory.
1813 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1814 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1819 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1820 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1823 if (error == 0 && NFS_ISV4(tdvp)) {
1825 * For NFSv4, check to see if it is the same name and
1826 * replace the name, if it is different.
1828 MALLOC(newv4, struct nfsv4node *,
1829 sizeof (struct nfsv4node) +
1830 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1831 M_NFSV4NODE, M_WAITOK);
1832 mtx_lock(&tdnp->n_mtx);
1833 mtx_lock(&fnp->n_mtx);
1834 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1835 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1836 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1837 tcnp->cn_namelen) ||
1838 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1839 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1840 tdnp->n_fhp->nfh_len))) {
1842 { char nnn[100]; int nnnl;
1843 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1844 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1846 printf("ren replace=%s\n",nnn);
1849 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1852 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1853 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1854 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1855 tdnp->n_fhp->nfh_len);
1856 NFSBCOPY(tcnp->cn_nameptr,
1857 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1859 mtx_unlock(&tdnp->n_mtx);
1860 mtx_unlock(&fnp->n_mtx);
1862 FREE((caddr_t)newv4, M_NFSV4NODE);
1865 if (fvp->v_type == VDIR) {
1866 if (tvp != NULL && tvp->v_type == VDIR)
1881 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1883 if (error == ENOENT)
1889 * nfs file rename rpc called from nfs_remove() above
1892 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1893 struct sillyrename *sp)
1896 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1897 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1902 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1905 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1906 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1907 int tnamelen, struct ucred *cred, struct thread *td)
1909 struct nfsvattr fnfsva, tnfsva;
1910 struct nfsnode *fdnp = VTONFS(fdvp);
1911 struct nfsnode *tdnp = VTONFS(tdvp);
1912 int error = 0, fattrflag, tattrflag;
1914 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1915 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1916 &tattrflag, NULL, NULL);
1917 mtx_lock(&fdnp->n_mtx);
1918 fdnp->n_flag |= NMODIFIED;
1919 if (fattrflag != 0) {
1920 mtx_unlock(&fdnp->n_mtx);
1921 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1923 fdnp->n_attrstamp = 0;
1924 mtx_unlock(&fdnp->n_mtx);
1925 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1927 mtx_lock(&tdnp->n_mtx);
1928 tdnp->n_flag |= NMODIFIED;
1929 if (tattrflag != 0) {
1930 mtx_unlock(&tdnp->n_mtx);
1931 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1933 tdnp->n_attrstamp = 0;
1934 mtx_unlock(&tdnp->n_mtx);
1935 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1937 if (error && NFS_ISV4(fdvp))
1938 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1943 * nfs hard link create call
1946 nfs_link(struct vop_link_args *ap)
1948 struct vnode *vp = ap->a_vp;
1949 struct vnode *tdvp = ap->a_tdvp;
1950 struct componentname *cnp = ap->a_cnp;
1951 struct nfsnode *np, *tdnp;
1952 struct nfsvattr nfsva, dnfsva;
1953 int error = 0, attrflag, dattrflag;
1955 if (vp->v_mount != tdvp->v_mount) {
1960 * Push all writes to the server, so that the attribute cache
1961 * doesn't get "out of sync" with the server.
1962 * XXX There should be a better way!
1964 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1966 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1967 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1969 tdnp = VTONFS(tdvp);
1970 mtx_lock(&tdnp->n_mtx);
1971 tdnp->n_flag |= NMODIFIED;
1972 if (dattrflag != 0) {
1973 mtx_unlock(&tdnp->n_mtx);
1974 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1976 tdnp->n_attrstamp = 0;
1977 mtx_unlock(&tdnp->n_mtx);
1978 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1981 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1984 mtx_lock(&np->n_mtx);
1985 np->n_attrstamp = 0;
1986 mtx_unlock(&np->n_mtx);
1987 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1990 * If negative lookup caching is enabled, I might as well
1991 * add an entry for this node. Not necessary for correctness,
1992 * but if negative caching is enabled, then the system
1993 * must care about lookup caching hit rate, so...
1995 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1996 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
1997 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
1999 if (error && NFS_ISV4(vp))
2000 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2006 * nfs symbolic link create call
2009 nfs_symlink(struct vop_symlink_args *ap)
2011 struct vnode *dvp = ap->a_dvp;
2012 struct vattr *vap = ap->a_vap;
2013 struct componentname *cnp = ap->a_cnp;
2014 struct nfsvattr nfsva, dnfsva;
2016 struct nfsnode *np = NULL, *dnp;
2017 struct vnode *newvp = NULL;
2018 int error = 0, attrflag, dattrflag, ret;
2020 vap->va_type = VLNK;
2021 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2022 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2023 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2025 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2026 &np, NULL, LK_EXCLUSIVE);
2032 if (newvp != NULL) {
2034 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2036 } else if (!error) {
2038 * If we do not have an error and we could not extract the
2039 * newvp from the response due to the request being NFSv2, we
2040 * have to do a lookup in order to obtain a newvp to return.
2042 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2043 cnp->cn_cred, cnp->cn_thread, &np);
2051 error = nfscl_maperr(cnp->cn_thread, error,
2052 vap->va_uid, vap->va_gid);
2058 mtx_lock(&dnp->n_mtx);
2059 dnp->n_flag |= NMODIFIED;
2060 if (dattrflag != 0) {
2061 mtx_unlock(&dnp->n_mtx);
2062 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2064 dnp->n_attrstamp = 0;
2065 mtx_unlock(&dnp->n_mtx);
2066 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2069 * If negative lookup caching is enabled, I might as well
2070 * add an entry for this node. Not necessary for correctness,
2071 * but if negative caching is enabled, then the system
2072 * must care about lookup caching hit rate, so...
2074 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2075 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2076 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2085 nfs_mkdir(struct vop_mkdir_args *ap)
2087 struct vnode *dvp = ap->a_dvp;
2088 struct vattr *vap = ap->a_vap;
2089 struct componentname *cnp = ap->a_cnp;
2090 struct nfsnode *np = NULL, *dnp;
2091 struct vnode *newvp = NULL;
2094 struct nfsvattr nfsva, dnfsva;
2095 int error = 0, attrflag, dattrflag, ret;
2097 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2099 vap->va_type = VDIR;
2100 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2101 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2102 &attrflag, &dattrflag, NULL);
2104 mtx_lock(&dnp->n_mtx);
2105 dnp->n_flag |= NMODIFIED;
2106 if (dattrflag != 0) {
2107 mtx_unlock(&dnp->n_mtx);
2108 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2110 dnp->n_attrstamp = 0;
2111 mtx_unlock(&dnp->n_mtx);
2112 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2115 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2116 &np, NULL, LK_EXCLUSIVE);
2120 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2125 if (!error && newvp == NULL) {
2126 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2127 cnp->cn_cred, cnp->cn_thread, &np);
2130 if (newvp->v_type != VDIR)
2138 error = nfscl_maperr(cnp->cn_thread, error,
2139 vap->va_uid, vap->va_gid);
2142 * If negative lookup caching is enabled, I might as well
2143 * add an entry for this node. Not necessary for correctness,
2144 * but if negative caching is enabled, then the system
2145 * must care about lookup caching hit rate, so...
2147 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2148 (cnp->cn_flags & MAKEENTRY) &&
2149 attrflag != 0 && dattrflag != 0)
2150 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2158 * nfs remove directory call
2161 nfs_rmdir(struct vop_rmdir_args *ap)
2163 struct vnode *vp = ap->a_vp;
2164 struct vnode *dvp = ap->a_dvp;
2165 struct componentname *cnp = ap->a_cnp;
2166 struct nfsnode *dnp;
2167 struct nfsvattr dnfsva;
2168 int error, dattrflag;
2172 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2173 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2175 mtx_lock(&dnp->n_mtx);
2176 dnp->n_flag |= NMODIFIED;
2177 if (dattrflag != 0) {
2178 mtx_unlock(&dnp->n_mtx);
2179 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2181 dnp->n_attrstamp = 0;
2182 mtx_unlock(&dnp->n_mtx);
2183 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2188 if (error && NFS_ISV4(dvp))
2189 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2192 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2194 if (error == ENOENT)
2203 nfs_readdir(struct vop_readdir_args *ap)
2205 struct vnode *vp = ap->a_vp;
2206 struct nfsnode *np = VTONFS(vp);
2207 struct uio *uio = ap->a_uio;
2212 if (ap->a_eofflag != NULL)
2214 if (vp->v_type != VDIR)
2218 * First, check for hit on the EOF offset cache
2220 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2221 (np->n_flag & NMODIFIED) == 0) {
2222 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2223 mtx_lock(&np->n_mtx);
2224 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2225 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2226 mtx_unlock(&np->n_mtx);
2227 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2228 if (ap->a_eofflag != NULL)
2232 mtx_unlock(&np->n_mtx);
2237 * Call ncl_bioread() to do the real work.
2239 tresid = uio->uio_resid;
2240 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2242 if (!error && uio->uio_resid == tresid) {
2243 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2244 if (ap->a_eofflag != NULL)
2252 * Called from below the buffer cache by ncl_doio().
2255 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2258 struct nfsvattr nfsva;
2259 nfsuint64 *cookiep, cookie;
2260 struct nfsnode *dnp = VTONFS(vp);
2261 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2262 int error = 0, eof, attrflag;
2264 KASSERT(uiop->uio_iovcnt == 1 &&
2265 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2266 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2267 ("nfs readdirrpc bad uio"));
2270 * If there is no cookie, assume directory was stale.
2272 ncl_dircookie_lock(dnp);
2273 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2276 ncl_dircookie_unlock(dnp);
2278 ncl_dircookie_unlock(dnp);
2279 return (NFSERR_BAD_COOKIE);
2282 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2283 (void)ncl_fsinfo(nmp, vp, cred, td);
2285 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2286 &attrflag, &eof, NULL);
2288 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2292 * We are now either at the end of the directory or have filled
2296 dnp->n_direofoffset = uiop->uio_offset;
2298 if (uiop->uio_resid > 0)
2299 ncl_printf("EEK! readdirrpc resid > 0\n");
2300 ncl_dircookie_lock(dnp);
2301 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2303 ncl_dircookie_unlock(dnp);
2305 } else if (NFS_ISV4(vp)) {
2306 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2312 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2315 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2318 struct nfsvattr nfsva;
2319 nfsuint64 *cookiep, cookie;
2320 struct nfsnode *dnp = VTONFS(vp);
2321 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2322 int error = 0, attrflag, eof;
2324 KASSERT(uiop->uio_iovcnt == 1 &&
2325 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2326 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2327 ("nfs readdirplusrpc bad uio"));
2330 * If there is no cookie, assume directory was stale.
2332 ncl_dircookie_lock(dnp);
2333 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2336 ncl_dircookie_unlock(dnp);
2338 ncl_dircookie_unlock(dnp);
2339 return (NFSERR_BAD_COOKIE);
2342 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2343 (void)ncl_fsinfo(nmp, vp, cred, td);
2344 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2345 &attrflag, &eof, NULL);
2347 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2351 * We are now either at end of the directory or have filled the
2355 dnp->n_direofoffset = uiop->uio_offset;
2357 if (uiop->uio_resid > 0)
2358 ncl_printf("EEK! readdirplusrpc resid > 0\n");
2359 ncl_dircookie_lock(dnp);
2360 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2362 ncl_dircookie_unlock(dnp);
2364 } else if (NFS_ISV4(vp)) {
2365 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2371 * Silly rename. To make the NFS filesystem that is stateless look a little
2372 * more like the "ufs" a remove of an active vnode is translated to a rename
2373 * to a funny looking filename that is removed by nfs_inactive on the
2374 * nfsnode. There is the potential for another process on a different client
2375 * to create the same funny name between the nfs_lookitup() fails and the
2376 * nfs_rename() completes, but...
2379 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2381 struct sillyrename *sp;
2385 unsigned int lticks;
2389 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2390 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2391 M_NEWNFSREQ, M_WAITOK);
2392 sp->s_cred = crhold(cnp->cn_cred);
2397 * Fudge together a funny name.
2398 * Changing the format of the funny name to accomodate more
2399 * sillynames per directory.
2400 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2401 * CPU ticks since boot.
2403 pid = cnp->cn_thread->td_proc->p_pid;
2404 lticks = (unsigned int)ticks;
2406 sp->s_namlen = sprintf(sp->s_name,
2407 ".nfs.%08x.%04x4.4", lticks,
2409 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2410 cnp->cn_thread, NULL))
2414 error = nfs_renameit(dvp, vp, cnp, sp);
2417 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2418 cnp->cn_thread, &np);
2419 np->n_sillyrename = sp;
2424 free((caddr_t)sp, M_NEWNFSREQ);
2429 * Look up a file name and optionally either update the file handle or
2430 * allocate an nfsnode, depending on the value of npp.
2431 * npp == NULL --> just do the lookup
2432 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2434 * *npp != NULL --> update the file handle in the vnode
2437 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2438 struct thread *td, struct nfsnode **npp)
2440 struct vnode *newvp = NULL, *vp;
2441 struct nfsnode *np, *dnp = VTONFS(dvp);
2442 struct nfsfh *nfhp, *onfhp;
2443 struct nfsvattr nfsva, dnfsva;
2444 struct componentname cn;
2445 int error = 0, attrflag, dattrflag;
2448 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2449 &nfhp, &attrflag, &dattrflag, NULL);
2451 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2452 if (npp && !error) {
2457 * For NFSv4, check to see if it is the same name and
2458 * replace the name, if it is different.
2460 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2461 (np->n_v4->n4_namelen != len ||
2462 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2463 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2464 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2465 dnp->n_fhp->nfh_len))) {
2467 { char nnn[100]; int nnnl;
2468 nnnl = (len < 100) ? len : 99;
2469 bcopy(name, nnn, nnnl);
2471 printf("replace=%s\n",nnn);
2474 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2475 MALLOC(np->n_v4, struct nfsv4node *,
2476 sizeof (struct nfsv4node) +
2477 dnp->n_fhp->nfh_len + len - 1,
2478 M_NFSV4NODE, M_WAITOK);
2479 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2480 np->n_v4->n4_namelen = len;
2481 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2482 dnp->n_fhp->nfh_len);
2483 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2485 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2489 * Rehash node for new file handle.
2491 vfs_hash_rehash(vp, hash);
2494 FREE((caddr_t)onfhp, M_NFSFH);
2496 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2497 FREE((caddr_t)nfhp, M_NFSFH);
2501 cn.cn_nameptr = name;
2502 cn.cn_namelen = len;
2503 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2504 &np, NULL, LK_EXCLUSIVE);
2509 if (!attrflag && *npp == NULL) {
2517 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2520 if (npp && *npp == NULL) {
2531 if (error && NFS_ISV4(dvp))
2532 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2537 * Nfs Version 3 and 4 commit rpc
2540 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2543 struct nfsvattr nfsva;
2544 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2545 int error, attrflag;
2546 u_char verf[NFSX_VERF];
2548 mtx_lock(&nmp->nm_mtx);
2549 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2550 mtx_unlock(&nmp->nm_mtx);
2553 mtx_unlock(&nmp->nm_mtx);
2554 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2557 mtx_lock(&nmp->nm_mtx);
2558 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2559 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2560 error = NFSERR_STALEWRITEVERF;
2562 mtx_unlock(&nmp->nm_mtx);
2563 if (!error && attrflag)
2564 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2566 } else if (NFS_ISV4(vp)) {
2567 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2574 * For async requests when nfsiod(s) are running, queue the request by
2575 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2579 nfs_strategy(struct vop_strategy_args *ap)
2581 struct buf *bp = ap->a_bp;
2584 KASSERT(!(bp->b_flags & B_DONE),
2585 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2586 BUF_ASSERT_HELD(bp);
2588 if (bp->b_iocmd == BIO_READ)
2594 * If the op is asynchronous and an i/o daemon is waiting
2595 * queue the request, wake it up and wait for completion
2596 * otherwise just do it ourselves.
2598 if ((bp->b_flags & B_ASYNC) == 0 ||
2599 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2600 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2605 * fsync vnode op. Just call ncl_flush() with commit == 1.
2609 nfs_fsync(struct vop_fsync_args *ap)
2612 if (ap->a_vp->v_type != VREG) {
2614 * For NFS, metadata is changed synchronously on the server,
2615 * so there is nothing to flush. Also, ncl_flush() clears
2616 * the NMODIFIED flag and that shouldn't be done here for
2621 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2625 * Flush all the blocks associated with a vnode.
2626 * Walk through the buffer pool and push any dirty pages
2627 * associated with the vnode.
2628 * If the called_from_renewthread argument is TRUE, it has been called
2629 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2630 * waiting for a buffer write to complete.
2633 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2634 int commit, int called_from_renewthread)
2636 struct nfsnode *np = VTONFS(vp);
2640 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2641 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2642 int passone = 1, trycnt = 0;
2643 u_quad_t off, endoff, toff;
2644 struct ucred* wcred = NULL;
2645 struct buf **bvec = NULL;
2647 #ifndef NFS_COMMITBVECSIZ
2648 #define NFS_COMMITBVECSIZ 20
2650 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2651 int bvecsize = 0, bveccount;
2653 if (called_from_renewthread != 0)
2655 if (nmp->nm_flag & NFSMNT_INT)
2656 slpflag = NFS_PCATCH;
2661 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2662 * server, but has not been committed to stable storage on the server
2663 * yet. On the first pass, the byte range is worked out and the commit
2664 * rpc is done. On the second pass, ncl_writebp() is called to do the
2671 if (NFS_ISV34(vp) && commit) {
2672 if (bvec != NULL && bvec != bvec_on_stack)
2675 * Count up how many buffers waiting for a commit.
2679 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2680 if (!BUF_ISLOCKED(bp) &&
2681 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2682 == (B_DELWRI | B_NEEDCOMMIT))
2686 * Allocate space to remember the list of bufs to commit. It is
2687 * important to use M_NOWAIT here to avoid a race with nfs_write.
2688 * If we can't get memory (for whatever reason), we will end up
2689 * committing the buffers one-by-one in the loop below.
2691 if (bveccount > NFS_COMMITBVECSIZ) {
2693 * Release the vnode interlock to avoid a lock
2697 bvec = (struct buf **)
2698 malloc(bveccount * sizeof(struct buf *),
2702 bvec = bvec_on_stack;
2703 bvecsize = NFS_COMMITBVECSIZ;
2705 bvecsize = bveccount;
2707 bvec = bvec_on_stack;
2708 bvecsize = NFS_COMMITBVECSIZ;
2710 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2711 if (bvecpos >= bvecsize)
2713 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2714 nbp = TAILQ_NEXT(bp, b_bobufs);
2717 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2718 (B_DELWRI | B_NEEDCOMMIT)) {
2720 nbp = TAILQ_NEXT(bp, b_bobufs);
2726 * Work out if all buffers are using the same cred
2727 * so we can deal with them all with one commit.
2729 * NOTE: we are not clearing B_DONE here, so we have
2730 * to do it later on in this routine if we intend to
2731 * initiate I/O on the bp.
2733 * Note: to avoid loopback deadlocks, we do not
2734 * assign b_runningbufspace.
2737 wcred = bp->b_wcred;
2738 else if (wcred != bp->b_wcred)
2740 vfs_busy_pages(bp, 1);
2744 * bp is protected by being locked, but nbp is not
2745 * and vfs_busy_pages() may sleep. We have to
2748 nbp = TAILQ_NEXT(bp, b_bobufs);
2751 * A list of these buffers is kept so that the
2752 * second loop knows which buffers have actually
2753 * been committed. This is necessary, since there
2754 * may be a race between the commit rpc and new
2755 * uncommitted writes on the file.
2757 bvec[bvecpos++] = bp;
2758 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2762 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2770 * Commit data on the server, as required.
2771 * If all bufs are using the same wcred, then use that with
2772 * one call for all of them, otherwise commit each one
2775 if (wcred != NOCRED)
2776 retv = ncl_commit(vp, off, (int)(endoff - off),
2780 for (i = 0; i < bvecpos; i++) {
2783 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2785 size = (u_quad_t)(bp->b_dirtyend
2787 retv = ncl_commit(vp, off, (int)size,
2793 if (retv == NFSERR_STALEWRITEVERF)
2794 ncl_clearcommit(vp->v_mount);
2797 * Now, either mark the blocks I/O done or mark the
2798 * blocks dirty, depending on whether the commit
2801 for (i = 0; i < bvecpos; i++) {
2803 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2806 * Error, leave B_DELWRI intact
2808 vfs_unbusy_pages(bp);
2812 * Success, remove B_DELWRI ( bundirty() ).
2814 * b_dirtyoff/b_dirtyend seem to be NFS
2815 * specific. We should probably move that
2816 * into bundirty(). XXX
2819 bp->b_flags |= B_ASYNC;
2821 bp->b_flags &= ~B_DONE;
2822 bp->b_ioflags &= ~BIO_ERROR;
2823 bp->b_dirtyoff = bp->b_dirtyend = 0;
2830 * Start/do any write(s) that are required.
2834 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2835 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2836 if (waitfor != MNT_WAIT || passone)
2839 error = BUF_TIMELOCK(bp,
2840 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2841 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2846 if (error == ENOLCK) {
2850 if (called_from_renewthread != 0) {
2852 * Return EIO so the flush will be retried
2858 if (newnfs_sigintr(nmp, td)) {
2862 if (slpflag & PCATCH) {
2868 if ((bp->b_flags & B_DELWRI) == 0)
2869 panic("nfs_fsync: not dirty");
2870 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2876 if (passone || !commit)
2877 bp->b_flags |= B_ASYNC;
2879 bp->b_flags |= B_ASYNC;
2881 if (newnfs_sigintr(nmp, td)) {
2892 if (waitfor == MNT_WAIT) {
2893 while (bo->bo_numoutput) {
2894 error = bufobj_wwait(bo, slpflag, slptimeo);
2897 if (called_from_renewthread != 0) {
2899 * Return EIO so that the flush will be
2905 error = newnfs_sigintr(nmp, td);
2908 if (slpflag & PCATCH) {
2915 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2920 * Wait for all the async IO requests to drain
2923 mtx_lock(&np->n_mtx);
2924 while (np->n_directio_asyncwr > 0) {
2925 np->n_flag |= NFSYNCWAIT;
2926 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2927 &np->n_mtx, slpflag | (PRIBIO + 1),
2930 if (newnfs_sigintr(nmp, td)) {
2931 mtx_unlock(&np->n_mtx);
2937 mtx_unlock(&np->n_mtx);
2940 mtx_lock(&np->n_mtx);
2941 if (np->n_flag & NWRITEERR) {
2942 error = np->n_error;
2943 np->n_flag &= ~NWRITEERR;
2945 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2946 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2947 np->n_flag &= ~NMODIFIED;
2948 mtx_unlock(&np->n_mtx);
2950 if (bvec != NULL && bvec != bvec_on_stack)
2952 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2953 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2954 np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2955 /* try, try again... */
2960 printf("try%d\n", trycnt);
2967 * NFS advisory byte-level locks.
2970 nfs_advlock(struct vop_advlock_args *ap)
2972 struct vnode *vp = ap->a_vp;
2974 struct nfsnode *np = VTONFS(ap->a_vp);
2975 struct proc *p = (struct proc *)ap->a_id;
2976 struct thread *td = curthread; /* XXX */
2978 int ret, error = EOPNOTSUPP;
2981 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
2982 if (vp->v_type != VREG)
2984 if ((ap->a_flags & F_POSIX) != 0)
2987 cred = td->td_ucred;
2988 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2989 if (vp->v_iflag & VI_DOOMED) {
2990 NFSVOPUNLOCK(vp, 0);
2995 * If this is unlocking a write locked region, flush and
2996 * commit them before unlocking. This is required by
2997 * RFC3530 Sec. 9.3.2.
2999 if (ap->a_op == F_UNLCK &&
3000 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3002 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
3005 * Loop around doing the lock op, while a blocking lock
3006 * must wait for the lock op to succeed.
3009 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3010 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3011 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3012 ap->a_op == F_SETLK) {
3013 NFSVOPUNLOCK(vp, 0);
3014 error = nfs_catnap(PZERO | PCATCH, ret,
3018 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3019 if (vp->v_iflag & VI_DOOMED) {
3020 NFSVOPUNLOCK(vp, 0);
3024 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3025 ap->a_op == F_SETLK);
3026 if (ret == NFSERR_DENIED) {
3027 NFSVOPUNLOCK(vp, 0);
3029 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3030 NFSVOPUNLOCK(vp, 0);
3032 } else if (ret != 0) {
3033 NFSVOPUNLOCK(vp, 0);
3038 * Now, if we just got a lock, invalidate data in the buffer
3039 * cache, as required, so that the coherency conforms with
3040 * RFC3530 Sec. 9.3.2.
3042 if (ap->a_op == F_SETLK) {
3043 if ((np->n_flag & NMODIFIED) == 0) {
3044 np->n_attrstamp = 0;
3045 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3046 ret = VOP_GETATTR(vp, &va, cred);
3048 if ((np->n_flag & NMODIFIED) || ret ||
3049 np->n_change != va.va_filerev) {
3050 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3051 np->n_attrstamp = 0;
3052 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3053 ret = VOP_GETATTR(vp, &va, cred);
3055 np->n_mtime = va.va_mtime;
3056 np->n_change = va.va_filerev;
3059 /* Mark that a file lock has been acquired. */
3060 mtx_lock(&np->n_mtx);
3061 np->n_flag |= NHASBEENLOCKED;
3062 mtx_unlock(&np->n_mtx);
3064 NFSVOPUNLOCK(vp, 0);
3066 } else if (!NFS_ISV4(vp)) {
3067 error = NFSVOPLOCK(vp, LK_SHARED);
3070 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3071 size = VTONFS(vp)->n_size;
3072 NFSVOPUNLOCK(vp, 0);
3073 error = lf_advlock(ap, &(vp->v_lockf), size);
3075 if (nfs_advlock_p != NULL)
3076 error = nfs_advlock_p(ap);
3078 NFSVOPUNLOCK(vp, 0);
3082 if (error == 0 && ap->a_op == F_SETLK) {
3083 /* Mark that a file lock has been acquired. */
3084 mtx_lock(&np->n_mtx);
3085 np->n_flag |= NHASBEENLOCKED;
3086 mtx_unlock(&np->n_mtx);
3093 * NFS advisory byte-level locks.
3096 nfs_advlockasync(struct vop_advlockasync_args *ap)
3098 struct vnode *vp = ap->a_vp;
3103 return (EOPNOTSUPP);
3104 error = NFSVOPLOCK(vp, LK_SHARED);
3107 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3108 size = VTONFS(vp)->n_size;
3109 NFSVOPUNLOCK(vp, 0);
3110 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3112 NFSVOPUNLOCK(vp, 0);
3119 * Print out the contents of an nfsnode.
3122 nfs_print(struct vop_print_args *ap)
3124 struct vnode *vp = ap->a_vp;
3125 struct nfsnode *np = VTONFS(vp);
3127 ncl_printf("\tfileid %ld fsid 0x%x",
3128 np->n_vattr.na_fileid, np->n_vattr.na_fsid);
3129 if (vp->v_type == VFIFO)
3136 * This is the "real" nfs::bwrite(struct buf*).
3137 * We set B_CACHE if this is a VMIO buffer.
3140 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3143 int oldflags = bp->b_flags;
3149 BUF_ASSERT_HELD(bp);
3151 if (bp->b_flags & B_INVAL) {
3156 bp->b_flags |= B_CACHE;
3159 * Undirty the bp. We will redirty it later if the I/O fails.
3164 bp->b_flags &= ~B_DONE;
3165 bp->b_ioflags &= ~BIO_ERROR;
3166 bp->b_iocmd = BIO_WRITE;
3168 bufobj_wref(bp->b_bufobj);
3169 curthread->td_ru.ru_oublock++;
3173 * Note: to avoid loopback deadlocks, we do not
3174 * assign b_runningbufspace.
3176 vfs_busy_pages(bp, 1);
3179 bp->b_iooffset = dbtob(bp->b_blkno);
3182 if( (oldflags & B_ASYNC) == 0) {
3183 int rtval = bufwait(bp);
3185 if (oldflags & B_DELWRI) {
3198 * nfs special file access vnode op.
3199 * Essentially just get vattr and then imitate iaccess() since the device is
3200 * local to the client.
3203 nfsspec_access(struct vop_access_args *ap)
3206 struct ucred *cred = ap->a_cred;
3207 struct vnode *vp = ap->a_vp;
3208 accmode_t accmode = ap->a_accmode;
3213 * Disallow write attempts on filesystems mounted read-only;
3214 * unless the file is a socket, fifo, or a block or character
3215 * device resident on the filesystem.
3217 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3218 switch (vp->v_type) {
3228 error = VOP_GETATTR(vp, vap, cred);
3231 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3232 accmode, cred, NULL);
3238 * Read wrapper for fifos.
3241 nfsfifo_read(struct vop_read_args *ap)
3243 struct nfsnode *np = VTONFS(ap->a_vp);
3249 mtx_lock(&np->n_mtx);
3251 vfs_timestamp(&np->n_atim);
3252 mtx_unlock(&np->n_mtx);
3253 error = fifo_specops.vop_read(ap);
3258 * Write wrapper for fifos.
3261 nfsfifo_write(struct vop_write_args *ap)
3263 struct nfsnode *np = VTONFS(ap->a_vp);
3268 mtx_lock(&np->n_mtx);
3270 vfs_timestamp(&np->n_mtim);
3271 mtx_unlock(&np->n_mtx);
3272 return(fifo_specops.vop_write(ap));
3276 * Close wrapper for fifos.
3278 * Update the times on the nfsnode then do fifo close.
3281 nfsfifo_close(struct vop_close_args *ap)
3283 struct vnode *vp = ap->a_vp;
3284 struct nfsnode *np = VTONFS(vp);
3288 mtx_lock(&np->n_mtx);
3289 if (np->n_flag & (NACC | NUPD)) {
3291 if (np->n_flag & NACC)
3293 if (np->n_flag & NUPD)
3296 if (vrefcnt(vp) == 1 &&
3297 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3299 if (np->n_flag & NACC)
3300 vattr.va_atime = np->n_atim;
3301 if (np->n_flag & NUPD)
3302 vattr.va_mtime = np->n_mtim;
3303 mtx_unlock(&np->n_mtx);
3304 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3308 mtx_unlock(&np->n_mtx);
3310 return (fifo_specops.vop_close(ap));
3314 * Just call ncl_writebp() with the force argument set to 1.
3316 * NOTE: B_DONE may or may not be set in a_bp on call.
3319 nfs_bwrite(struct buf *bp)
3322 return (ncl_writebp(bp, 1, curthread));
3325 struct buf_ops buf_ops_newnfs = {
3326 .bop_name = "buf_ops_nfs",
3327 .bop_write = nfs_bwrite,
3328 .bop_strategy = bufstrategy,
3329 .bop_sync = bufsync,
3330 .bop_bdflush = bufbdflush,
3334 * Cloned from vop_stdlock(), and then the ugly hack added.
3337 nfs_lock1(struct vop_lock1_args *ap)
3339 struct vnode *vp = ap->a_vp;
3343 * Since vfs_hash_get() calls vget() and it will no longer work
3344 * for FreeBSD8 with flags == 0, I can only think of this horrible
3345 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3346 * and then handle it here. All I want for this case is a v_usecount
3347 * on the vnode to use for recovery, while another thread might
3348 * hold a lock on the vnode. I have the other threads blocked, so
3349 * there isn't any race problem.
3351 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3352 if ((ap->a_flags & LK_INTERLOCK) == 0)
3354 if ((vp->v_iflag & VI_DOOMED))
3359 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3360 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3365 nfs_getacl(struct vop_getacl_args *ap)
3369 if (ap->a_type != ACL_TYPE_NFS4)
3370 return (EOPNOTSUPP);
3371 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3373 if (error > NFSERR_STALE) {
3374 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3381 nfs_setacl(struct vop_setacl_args *ap)
3385 if (ap->a_type != ACL_TYPE_NFS4)
3386 return (EOPNOTSUPP);
3387 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3389 if (error > NFSERR_STALE) {
3390 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3397 * Return POSIX pathconf information applicable to nfs filesystems.
3400 nfs_pathconf(struct vop_pathconf_args *ap)
3402 struct nfsv3_pathconf pc;
3403 struct nfsvattr nfsva;
3404 struct vnode *vp = ap->a_vp;
3405 struct thread *td = curthread;
3406 int attrflag, error;
3408 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
3409 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3410 ap->a_name == _PC_NO_TRUNC)) ||
3411 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
3413 * Since only the above 4 a_names are returned by the NFSv3
3414 * Pathconf RPC, there is no point in doing it for others.
3415 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
3416 * be used for _PC_NFS4_ACL as well.
3418 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3421 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3427 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3430 pc.pc_linkmax = LINK_MAX;
3431 pc.pc_namemax = NFS_MAXNAMLEN;
3433 pc.pc_chownrestricted = 1;
3434 pc.pc_caseinsensitive = 0;
3435 pc.pc_casepreserving = 1;
3438 switch (ap->a_name) {
3440 *ap->a_retval = pc.pc_linkmax;
3443 *ap->a_retval = pc.pc_namemax;
3446 *ap->a_retval = PATH_MAX;
3449 *ap->a_retval = PIPE_BUF;
3451 case _PC_CHOWN_RESTRICTED:
3452 *ap->a_retval = pc.pc_chownrestricted;
3455 *ap->a_retval = pc.pc_notrunc;
3457 case _PC_ACL_EXTENDED:
3461 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3462 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3467 case _PC_ACL_PATH_MAX:
3469 *ap->a_retval = ACL_MAX_ENTRIES;
3473 case _PC_MAC_PRESENT:
3477 /* _PC_ASYNC_IO should have been handled by upper layers. */
3478 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3487 case _PC_ALLOC_SIZE_MIN:
3488 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3490 case _PC_FILESIZEBITS:
3496 case _PC_REC_INCR_XFER_SIZE:
3497 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3499 case _PC_REC_MAX_XFER_SIZE:
3500 *ap->a_retval = -1; /* means ``unlimited'' */
3502 case _PC_REC_MIN_XFER_SIZE:
3503 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3505 case _PC_REC_XFER_ALIGN:
3506 *ap->a_retval = PAGE_SIZE;
3508 case _PC_SYMLINK_MAX:
3509 *ap->a_retval = NFS_MAXPATHLEN;