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 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
769 np->n_change = nfsva.na_filerev;
770 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
778 ret = nfsrpc_close(vp, 0, ap->a_td);
782 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
785 if (newnfs_directio_enable)
786 KASSERT((np->n_directio_asyncwr == 0),
787 ("nfs_close: dirty unflushed (%d) directio buffers\n",
788 np->n_directio_asyncwr));
789 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
790 mtx_lock(&np->n_mtx);
791 KASSERT((np->n_directio_opens > 0),
792 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
793 np->n_directio_opens--;
794 if (np->n_directio_opens == 0)
795 np->n_flag &= ~NNONCACHE;
796 mtx_unlock(&np->n_mtx);
804 * nfs getattr call from vfs.
807 nfs_getattr(struct vop_getattr_args *ap)
809 struct vnode *vp = ap->a_vp;
810 struct thread *td = curthread; /* XXX */
811 struct nfsnode *np = VTONFS(vp);
813 struct nfsvattr nfsva;
814 struct vattr *vap = ap->a_vap;
818 * Update local times for special files.
820 mtx_lock(&np->n_mtx);
821 if (np->n_flag & (NACC | NUPD))
823 mtx_unlock(&np->n_mtx);
825 * First look in the cache.
827 if (ncl_getattrcache(vp, &vattr) == 0) {
828 vap->va_type = vattr.va_type;
829 vap->va_mode = vattr.va_mode;
830 vap->va_nlink = vattr.va_nlink;
831 vap->va_uid = vattr.va_uid;
832 vap->va_gid = vattr.va_gid;
833 vap->va_fsid = vattr.va_fsid;
834 vap->va_fileid = vattr.va_fileid;
835 vap->va_size = vattr.va_size;
836 vap->va_blocksize = vattr.va_blocksize;
837 vap->va_atime = vattr.va_atime;
838 vap->va_mtime = vattr.va_mtime;
839 vap->va_ctime = vattr.va_ctime;
840 vap->va_gen = vattr.va_gen;
841 vap->va_flags = vattr.va_flags;
842 vap->va_rdev = vattr.va_rdev;
843 vap->va_bytes = vattr.va_bytes;
844 vap->va_filerev = vattr.va_filerev;
846 * Get the local modify time for the case of a write
849 nfscl_deleggetmodtime(vp, &vap->va_mtime);
853 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
854 nfsaccess_cache_timeout > 0) {
855 NFSINCRGLOBAL(newnfsstats.accesscache_misses);
856 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
857 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
858 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
862 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
864 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
867 * Get the local modify time for the case of a write
870 nfscl_deleggetmodtime(vp, &vap->va_mtime);
871 } else if (NFS_ISV4(vp)) {
872 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
881 nfs_setattr(struct vop_setattr_args *ap)
883 struct vnode *vp = ap->a_vp;
884 struct nfsnode *np = VTONFS(vp);
885 struct thread *td = curthread; /* XXX */
886 struct vattr *vap = ap->a_vap;
895 * Setting of flags and marking of atimes are not supported.
897 if (vap->va_flags != VNOVAL)
901 * Disallow write attempts if the filesystem is mounted read-only.
903 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
904 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
905 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
906 (vp->v_mount->mnt_flag & MNT_RDONLY))
908 if (vap->va_size != VNOVAL) {
909 switch (vp->v_type) {
916 if (vap->va_mtime.tv_sec == VNOVAL &&
917 vap->va_atime.tv_sec == VNOVAL &&
918 vap->va_mode == (mode_t)VNOVAL &&
919 vap->va_uid == (uid_t)VNOVAL &&
920 vap->va_gid == (gid_t)VNOVAL)
922 vap->va_size = VNOVAL;
926 * Disallow write attempts if the filesystem is
929 if (vp->v_mount->mnt_flag & MNT_RDONLY)
932 * We run vnode_pager_setsize() early (why?),
933 * we must set np->n_size now to avoid vinvalbuf
934 * V_SAVE races that might setsize a lower
937 mtx_lock(&np->n_mtx);
939 mtx_unlock(&np->n_mtx);
940 error = ncl_meta_setsize(vp, ap->a_cred, td,
942 mtx_lock(&np->n_mtx);
943 if (np->n_flag & NMODIFIED) {
945 mtx_unlock(&np->n_mtx);
946 if (vap->va_size == 0)
947 error = ncl_vinvalbuf(vp, 0, td, 1);
949 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
951 vnode_pager_setsize(vp, tsize);
955 * Call nfscl_delegmodtime() to set the modify time
956 * locally, as required.
958 nfscl_delegmodtime(vp);
960 mtx_unlock(&np->n_mtx);
962 * np->n_size has already been set to vap->va_size
963 * in ncl_meta_setsize(). We must set it again since
964 * nfs_loadattrcache() could be called through
965 * ncl_meta_setsize() and could modify np->n_size.
967 mtx_lock(&np->n_mtx);
968 np->n_vattr.na_size = np->n_size = vap->va_size;
969 mtx_unlock(&np->n_mtx);
972 mtx_lock(&np->n_mtx);
973 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
974 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
975 mtx_unlock(&np->n_mtx);
976 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
977 (error == EINTR || error == EIO))
980 mtx_unlock(&np->n_mtx);
982 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
983 if (error && vap->va_size != VNOVAL) {
984 mtx_lock(&np->n_mtx);
985 np->n_size = np->n_vattr.na_size = tsize;
986 vnode_pager_setsize(vp, tsize);
987 mtx_unlock(&np->n_mtx);
993 * Do an nfs setattr rpc.
996 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
999 struct nfsnode *np = VTONFS(vp);
1000 int error, ret, attrflag, i;
1001 struct nfsvattr nfsva;
1003 if (NFS_ISV34(vp)) {
1004 mtx_lock(&np->n_mtx);
1005 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1006 np->n_accesscache[i].stamp = 0;
1007 np->n_flag |= NDELEGMOD;
1008 mtx_unlock(&np->n_mtx);
1009 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1011 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1014 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1018 if (error && NFS_ISV4(vp))
1019 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1024 * nfs lookup call, one step at a time...
1025 * First look in cache
1026 * If not found, unlock the directory nfsnode and do the rpc
1029 nfs_lookup(struct vop_lookup_args *ap)
1031 struct componentname *cnp = ap->a_cnp;
1032 struct vnode *dvp = ap->a_dvp;
1033 struct vnode **vpp = ap->a_vpp;
1034 struct mount *mp = dvp->v_mount;
1035 int flags = cnp->cn_flags;
1036 struct vnode *newvp;
1037 struct nfsmount *nmp;
1038 struct nfsnode *np, *newnp;
1039 int error = 0, attrflag, dattrflag, ltype, ncticks;
1040 struct thread *td = cnp->cn_thread;
1042 struct nfsvattr dnfsva, nfsva;
1044 struct timespec nctime;
1047 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1048 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1050 if (dvp->v_type != VDIR)
1055 /* For NFSv4, wait until any remove is done. */
1056 mtx_lock(&np->n_mtx);
1057 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1058 np->n_flag |= NREMOVEWANT;
1059 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1061 mtx_unlock(&np->n_mtx);
1063 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1065 error = cache_lookup_times(dvp, vpp, cnp, &nctime, &ncticks);
1066 if (error > 0 && error != ENOENT)
1070 * Lookups of "." are special and always return the
1071 * current directory. cache_lookup() already handles
1072 * associated locking bookkeeping, etc.
1074 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1075 /* XXX: Is this really correct? */
1076 if (cnp->cn_nameiop != LOOKUP &&
1078 cnp->cn_flags |= SAVENAME;
1083 * We only accept a positive hit in the cache if the
1084 * change time of the file matches our cached copy.
1085 * Otherwise, we discard the cache entry and fallback
1086 * to doing a lookup RPC. We also only trust cache
1087 * entries for less than nm_nametimeo seconds.
1089 * To better handle stale file handles and attributes,
1090 * clear the attribute cache of this node if it is a
1091 * leaf component, part of an open() call, and not
1092 * locally modified before fetching the attributes.
1093 * This should allow stale file handles to be detected
1094 * here where we can fall back to a LOOKUP RPC to
1095 * recover rather than having nfs_open() detect the
1096 * stale file handle and failing open(2) with ESTALE.
1099 newnp = VTONFS(newvp);
1100 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1101 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1102 !(newnp->n_flag & NMODIFIED)) {
1103 mtx_lock(&newnp->n_mtx);
1104 newnp->n_attrstamp = 0;
1105 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1106 mtx_unlock(&newnp->n_mtx);
1108 if (nfscl_nodeleg(newvp, 0) == 0 ||
1109 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1110 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1111 timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1112 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1113 if (cnp->cn_nameiop != LOOKUP &&
1115 cnp->cn_flags |= SAVENAME;
1124 } else if (error == ENOENT) {
1125 if (dvp->v_iflag & VI_DOOMED)
1128 * We only accept a negative hit in the cache if the
1129 * modification time of the parent directory matches
1130 * the cached copy in the name cache entry.
1131 * Otherwise, we discard all of the negative cache
1132 * entries for this directory. We also only trust
1133 * negative cache entries for up to nm_negnametimeo
1136 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1137 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1138 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1139 NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1142 cache_purge_negative(dvp);
1147 NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1148 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1149 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1152 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1154 if (newvp != NULLVP) {
1159 if (error != ENOENT) {
1161 error = nfscl_maperr(td, error, (uid_t)0,
1166 /* The requested file was not found. */
1167 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1168 (flags & ISLASTCN)) {
1170 * XXX: UFS does a full VOP_ACCESS(dvp,
1171 * VWRITE) here instead of just checking
1174 if (mp->mnt_flag & MNT_RDONLY)
1176 cnp->cn_flags |= SAVENAME;
1177 return (EJUSTRETURN);
1180 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
1183 * Cache the modification time of the parent
1184 * directory from the post-op attributes in
1185 * the name cache entry. The negative cache
1186 * entry will be ignored once the directory
1187 * has changed. Don't bother adding the entry
1188 * if the directory has already changed.
1190 mtx_lock(&np->n_mtx);
1191 if (timespeccmp(&np->n_vattr.na_mtime,
1192 &dnfsva.na_mtime, ==)) {
1193 mtx_unlock(&np->n_mtx);
1194 cache_enter_time(dvp, NULL, cnp,
1195 &dnfsva.na_mtime, NULL);
1197 mtx_unlock(&np->n_mtx);
1203 * Handle RENAME case...
1205 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1206 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1207 FREE((caddr_t)nfhp, M_NFSFH);
1210 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1216 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1219 cnp->cn_flags |= SAVENAME;
1223 if (flags & ISDOTDOT) {
1224 ltype = NFSVOPISLOCKED(dvp);
1225 error = vfs_busy(mp, MBF_NOWAIT);
1228 NFSVOPUNLOCK(dvp, 0);
1229 error = vfs_busy(mp, 0);
1230 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1232 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1239 NFSVOPUNLOCK(dvp, 0);
1240 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1246 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1247 if (dvp->v_iflag & VI_DOOMED) {
1259 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1261 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1262 FREE((caddr_t)nfhp, M_NFSFH);
1266 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1269 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1275 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1277 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1278 !(np->n_flag & NMODIFIED)) {
1280 * Flush the attribute cache when opening a
1281 * leaf node to ensure that fresh attributes
1282 * are fetched in nfs_open() since we did not
1283 * fetch attributes from the LOOKUP reply.
1285 mtx_lock(&np->n_mtx);
1286 np->n_attrstamp = 0;
1287 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1288 mtx_unlock(&np->n_mtx);
1291 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1292 cnp->cn_flags |= SAVENAME;
1293 if ((cnp->cn_flags & MAKEENTRY) &&
1294 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1295 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1296 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1297 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1304 * Just call ncl_bioread() to do the work.
1307 nfs_read(struct vop_read_args *ap)
1309 struct vnode *vp = ap->a_vp;
1311 switch (vp->v_type) {
1313 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1317 return (EOPNOTSUPP);
1325 nfs_readlink(struct vop_readlink_args *ap)
1327 struct vnode *vp = ap->a_vp;
1329 if (vp->v_type != VLNK)
1331 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1335 * Do a readlink rpc.
1336 * Called by ncl_doio() from below the buffer cache.
1339 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1341 int error, ret, attrflag;
1342 struct nfsvattr nfsva;
1344 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1347 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1351 if (error && NFS_ISV4(vp))
1352 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1361 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1363 int error, ret, attrflag;
1364 struct nfsvattr nfsva;
1366 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1369 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1373 if (error && NFS_ISV4(vp))
1374 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1382 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1383 int *iomode, int *must_commit, int called_from_strategy)
1385 struct nfsvattr nfsva;
1386 int error = 0, attrflag, ret;
1388 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1389 uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy);
1391 if (VTONFS(vp)->n_flag & ND_NFSV4)
1392 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1395 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1401 *iomode = NFSWRITE_FILESYNC;
1402 if (error && NFS_ISV4(vp))
1403 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1409 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1410 * mode set to specify the file type and the size field for rdev.
1413 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1416 struct nfsvattr nfsva, dnfsva;
1417 struct vnode *newvp = NULL;
1418 struct nfsnode *np = NULL, *dnp;
1421 int error = 0, attrflag, dattrflag;
1424 if (vap->va_type == VCHR || vap->va_type == VBLK)
1425 rdev = vap->va_rdev;
1426 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1429 return (EOPNOTSUPP);
1430 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1432 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1433 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1434 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1437 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1438 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1439 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1442 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1443 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1446 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1449 if (attrflag != 0) {
1450 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1458 } else if (NFS_ISV4(dvp)) {
1459 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1463 mtx_lock(&dnp->n_mtx);
1464 dnp->n_flag |= NMODIFIED;
1466 dnp->n_attrstamp = 0;
1467 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1469 mtx_unlock(&dnp->n_mtx);
1475 * just call nfs_mknodrpc() to do the work.
1479 nfs_mknod(struct vop_mknod_args *ap)
1481 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1484 static struct mtx nfs_cverf_mtx;
1485 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1491 static nfsquad_t cverf;
1493 static int cverf_initialized = 0;
1495 mtx_lock(&nfs_cverf_mtx);
1496 if (cverf_initialized == 0) {
1497 cverf.lval[0] = arc4random();
1498 cverf.lval[1] = arc4random();
1499 cverf_initialized = 1;
1503 mtx_unlock(&nfs_cverf_mtx);
1509 * nfs file create call
1512 nfs_create(struct vop_create_args *ap)
1514 struct vnode *dvp = ap->a_dvp;
1515 struct vattr *vap = ap->a_vap;
1516 struct componentname *cnp = ap->a_cnp;
1517 struct nfsnode *np = NULL, *dnp;
1518 struct vnode *newvp = NULL;
1519 struct nfsmount *nmp;
1520 struct nfsvattr dnfsva, nfsva;
1523 int error = 0, attrflag, dattrflag, fmode = 0;
1527 * Oops, not for me..
1529 if (vap->va_type == VSOCK)
1530 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1532 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1534 if (vap->va_vaflags & VA_EXCLUSIVE)
1537 nmp = VFSTONFS(vnode_mount(dvp));
1539 /* For NFSv4, wait until any remove is done. */
1540 mtx_lock(&dnp->n_mtx);
1541 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1542 dnp->n_flag |= NREMOVEWANT;
1543 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1545 mtx_unlock(&dnp->n_mtx);
1547 cverf = nfs_get_cverf();
1548 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1549 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1550 &nfhp, &attrflag, &dattrflag, NULL);
1553 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1554 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1555 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1558 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1559 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1562 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1566 error = nfsrpc_getattr(newvp, cnp->cn_cred,
1567 cnp->cn_thread, &nfsva, NULL);
1569 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1573 if (newvp != NULL) {
1577 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1578 error == NFSERR_NOTSUPP) {
1582 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1583 if (nfscl_checksattr(vap, &nfsva)) {
1585 * We are normally called with only a partially
1586 * initialized VAP. Since the NFSv3 spec says that
1587 * the server may use the file attributes to
1588 * store the verifier, the spec requires us to do a
1589 * SETATTR RPC. FreeBSD servers store the verifier in
1590 * atime, but we can't really assume that all servers
1591 * will so we ensure that our SETATTR sets both atime
1594 if (vap->va_mtime.tv_sec == VNOVAL)
1595 vfs_timestamp(&vap->va_mtime);
1596 if (vap->va_atime.tv_sec == VNOVAL)
1597 vap->va_atime = vap->va_mtime;
1598 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1599 cnp->cn_thread, &nfsva, &attrflag, NULL);
1600 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1601 vap->va_gid != (gid_t)VNOVAL)) {
1602 /* try again without setting uid/gid */
1603 vap->va_uid = (uid_t)VNOVAL;
1604 vap->va_gid = (uid_t)VNOVAL;
1605 error = nfsrpc_setattr(newvp, vap, NULL,
1606 cnp->cn_cred, cnp->cn_thread, &nfsva,
1610 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1617 if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1618 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1621 } else if (NFS_ISV4(dvp)) {
1622 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1625 mtx_lock(&dnp->n_mtx);
1626 dnp->n_flag |= NMODIFIED;
1628 dnp->n_attrstamp = 0;
1629 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1631 mtx_unlock(&dnp->n_mtx);
1636 * nfs file remove call
1637 * To try and make nfs semantics closer to ufs semantics, a file that has
1638 * other processes using the vnode is renamed instead of removed and then
1639 * removed later on the last close.
1640 * - If v_usecount > 1
1641 * If a rename is not already in the works
1642 * call nfs_sillyrename() to set it up
1647 nfs_remove(struct vop_remove_args *ap)
1649 struct vnode *vp = ap->a_vp;
1650 struct vnode *dvp = ap->a_dvp;
1651 struct componentname *cnp = ap->a_cnp;
1652 struct nfsnode *np = VTONFS(vp);
1656 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1657 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1658 if (vp->v_type == VDIR)
1660 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1661 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1662 vattr.va_nlink > 1)) {
1664 * Purge the name cache so that the chance of a lookup for
1665 * the name succeeding while the remove is in progress is
1666 * minimized. Without node locking it can still happen, such
1667 * that an I/O op returns ESTALE, but since you get this if
1668 * another host removes the file..
1672 * throw away biocache buffers, mainly to avoid
1673 * unnecessary delayed writes later.
1675 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1677 if (error != EINTR && error != EIO)
1678 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1679 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1681 * Kludge City: If the first reply to the remove rpc is lost..
1682 * the reply to the retransmitted request will be ENOENT
1683 * since the file was in fact removed
1684 * Therefore, we cheat and return success.
1686 if (error == ENOENT)
1688 } else if (!np->n_sillyrename)
1689 error = nfs_sillyrename(dvp, vp, cnp);
1690 mtx_lock(&np->n_mtx);
1691 np->n_attrstamp = 0;
1692 mtx_unlock(&np->n_mtx);
1693 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1698 * nfs file remove rpc called from nfs_inactive
1701 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1704 * Make sure that the directory vnode is still valid.
1705 * XXX we should lock sp->s_dvp here.
1707 if (sp->s_dvp->v_type == VBAD)
1709 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1714 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1717 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1718 int namelen, struct ucred *cred, struct thread *td)
1720 struct nfsvattr dnfsva;
1721 struct nfsnode *dnp = VTONFS(dvp);
1722 int error = 0, dattrflag;
1724 mtx_lock(&dnp->n_mtx);
1725 dnp->n_flag |= NREMOVEINPROG;
1726 mtx_unlock(&dnp->n_mtx);
1727 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1729 mtx_lock(&dnp->n_mtx);
1730 if ((dnp->n_flag & NREMOVEWANT)) {
1731 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1732 mtx_unlock(&dnp->n_mtx);
1733 wakeup((caddr_t)dnp);
1735 dnp->n_flag &= ~NREMOVEINPROG;
1736 mtx_unlock(&dnp->n_mtx);
1739 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1740 mtx_lock(&dnp->n_mtx);
1741 dnp->n_flag |= NMODIFIED;
1743 dnp->n_attrstamp = 0;
1744 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1746 mtx_unlock(&dnp->n_mtx);
1747 if (error && NFS_ISV4(dvp))
1748 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1753 * nfs file rename call
1756 nfs_rename(struct vop_rename_args *ap)
1758 struct vnode *fvp = ap->a_fvp;
1759 struct vnode *tvp = ap->a_tvp;
1760 struct vnode *fdvp = ap->a_fdvp;
1761 struct vnode *tdvp = ap->a_tdvp;
1762 struct componentname *tcnp = ap->a_tcnp;
1763 struct componentname *fcnp = ap->a_fcnp;
1764 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1765 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1766 struct nfsv4node *newv4 = NULL;
1769 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1770 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1771 /* Check for cross-device rename */
1772 if ((fvp->v_mount != tdvp->v_mount) ||
1773 (tvp && (fvp->v_mount != tvp->v_mount))) {
1779 ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1783 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1787 * We have to flush B_DELWRI data prior to renaming
1788 * the file. If we don't, the delayed-write buffers
1789 * can be flushed out later after the file has gone stale
1790 * under NFSV3. NFSV2 does not have this problem because
1791 * ( as far as I can tell ) it flushes dirty buffers more
1794 * Skip the rename operation if the fsync fails, this can happen
1795 * due to the server's volume being full, when we pushed out data
1796 * that was written back to our cache earlier. Not checking for
1797 * this condition can result in potential (silent) data loss.
1799 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1800 NFSVOPUNLOCK(fvp, 0);
1802 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1807 * If the tvp exists and is in use, sillyrename it before doing the
1808 * rename of the new file over it.
1809 * XXX Can't sillyrename a directory.
1811 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1812 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1817 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1818 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1821 if (error == 0 && NFS_ISV4(tdvp)) {
1823 * For NFSv4, check to see if it is the same name and
1824 * replace the name, if it is different.
1826 MALLOC(newv4, struct nfsv4node *,
1827 sizeof (struct nfsv4node) +
1828 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1829 M_NFSV4NODE, M_WAITOK);
1830 mtx_lock(&tdnp->n_mtx);
1831 mtx_lock(&fnp->n_mtx);
1832 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1833 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1834 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1835 tcnp->cn_namelen) ||
1836 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1837 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1838 tdnp->n_fhp->nfh_len))) {
1840 { char nnn[100]; int nnnl;
1841 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1842 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1844 printf("ren replace=%s\n",nnn);
1847 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1850 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1851 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1852 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1853 tdnp->n_fhp->nfh_len);
1854 NFSBCOPY(tcnp->cn_nameptr,
1855 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1857 mtx_unlock(&tdnp->n_mtx);
1858 mtx_unlock(&fnp->n_mtx);
1860 FREE((caddr_t)newv4, M_NFSV4NODE);
1863 if (fvp->v_type == VDIR) {
1864 if (tvp != NULL && tvp->v_type == VDIR)
1879 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1881 if (error == ENOENT)
1887 * nfs file rename rpc called from nfs_remove() above
1890 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1891 struct sillyrename *sp)
1894 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1895 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1900 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1903 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1904 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1905 int tnamelen, struct ucred *cred, struct thread *td)
1907 struct nfsvattr fnfsva, tnfsva;
1908 struct nfsnode *fdnp = VTONFS(fdvp);
1909 struct nfsnode *tdnp = VTONFS(tdvp);
1910 int error = 0, fattrflag, tattrflag;
1912 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1913 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1914 &tattrflag, NULL, NULL);
1915 mtx_lock(&fdnp->n_mtx);
1916 fdnp->n_flag |= NMODIFIED;
1917 if (fattrflag != 0) {
1918 mtx_unlock(&fdnp->n_mtx);
1919 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1921 fdnp->n_attrstamp = 0;
1922 mtx_unlock(&fdnp->n_mtx);
1923 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1925 mtx_lock(&tdnp->n_mtx);
1926 tdnp->n_flag |= NMODIFIED;
1927 if (tattrflag != 0) {
1928 mtx_unlock(&tdnp->n_mtx);
1929 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1931 tdnp->n_attrstamp = 0;
1932 mtx_unlock(&tdnp->n_mtx);
1933 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1935 if (error && NFS_ISV4(fdvp))
1936 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1941 * nfs hard link create call
1944 nfs_link(struct vop_link_args *ap)
1946 struct vnode *vp = ap->a_vp;
1947 struct vnode *tdvp = ap->a_tdvp;
1948 struct componentname *cnp = ap->a_cnp;
1949 struct nfsnode *np, *tdnp;
1950 struct nfsvattr nfsva, dnfsva;
1951 int error = 0, attrflag, dattrflag;
1953 if (vp->v_mount != tdvp->v_mount) {
1958 * Push all writes to the server, so that the attribute cache
1959 * doesn't get "out of sync" with the server.
1960 * XXX There should be a better way!
1962 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1964 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1965 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1967 tdnp = VTONFS(tdvp);
1968 mtx_lock(&tdnp->n_mtx);
1969 tdnp->n_flag |= NMODIFIED;
1970 if (dattrflag != 0) {
1971 mtx_unlock(&tdnp->n_mtx);
1972 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1974 tdnp->n_attrstamp = 0;
1975 mtx_unlock(&tdnp->n_mtx);
1976 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1979 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1982 mtx_lock(&np->n_mtx);
1983 np->n_attrstamp = 0;
1984 mtx_unlock(&np->n_mtx);
1985 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1988 * If negative lookup caching is enabled, I might as well
1989 * add an entry for this node. Not necessary for correctness,
1990 * but if negative caching is enabled, then the system
1991 * must care about lookup caching hit rate, so...
1993 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1994 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
1995 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
1997 if (error && NFS_ISV4(vp))
1998 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2004 * nfs symbolic link create call
2007 nfs_symlink(struct vop_symlink_args *ap)
2009 struct vnode *dvp = ap->a_dvp;
2010 struct vattr *vap = ap->a_vap;
2011 struct componentname *cnp = ap->a_cnp;
2012 struct nfsvattr nfsva, dnfsva;
2014 struct nfsnode *np = NULL, *dnp;
2015 struct vnode *newvp = NULL;
2016 int error = 0, attrflag, dattrflag, ret;
2018 vap->va_type = VLNK;
2019 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2020 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2021 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2023 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2024 &np, NULL, LK_EXCLUSIVE);
2030 if (newvp != NULL) {
2032 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2034 } else if (!error) {
2036 * If we do not have an error and we could not extract the
2037 * newvp from the response due to the request being NFSv2, we
2038 * have to do a lookup in order to obtain a newvp to return.
2040 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2041 cnp->cn_cred, cnp->cn_thread, &np);
2049 error = nfscl_maperr(cnp->cn_thread, error,
2050 vap->va_uid, vap->va_gid);
2056 mtx_lock(&dnp->n_mtx);
2057 dnp->n_flag |= NMODIFIED;
2058 if (dattrflag != 0) {
2059 mtx_unlock(&dnp->n_mtx);
2060 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2062 dnp->n_attrstamp = 0;
2063 mtx_unlock(&dnp->n_mtx);
2064 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2067 * If negative lookup caching is enabled, I might as well
2068 * add an entry for this node. Not necessary for correctness,
2069 * but if negative caching is enabled, then the system
2070 * must care about lookup caching hit rate, so...
2072 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2073 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2074 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2083 nfs_mkdir(struct vop_mkdir_args *ap)
2085 struct vnode *dvp = ap->a_dvp;
2086 struct vattr *vap = ap->a_vap;
2087 struct componentname *cnp = ap->a_cnp;
2088 struct nfsnode *np = NULL, *dnp;
2089 struct vnode *newvp = NULL;
2092 struct nfsvattr nfsva, dnfsva;
2093 int error = 0, attrflag, dattrflag, ret;
2095 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2097 vap->va_type = VDIR;
2098 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2099 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2100 &attrflag, &dattrflag, NULL);
2102 mtx_lock(&dnp->n_mtx);
2103 dnp->n_flag |= NMODIFIED;
2104 if (dattrflag != 0) {
2105 mtx_unlock(&dnp->n_mtx);
2106 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2108 dnp->n_attrstamp = 0;
2109 mtx_unlock(&dnp->n_mtx);
2110 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2113 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2114 &np, NULL, LK_EXCLUSIVE);
2118 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2123 if (!error && newvp == NULL) {
2124 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2125 cnp->cn_cred, cnp->cn_thread, &np);
2128 if (newvp->v_type != VDIR)
2136 error = nfscl_maperr(cnp->cn_thread, error,
2137 vap->va_uid, vap->va_gid);
2140 * If negative lookup caching is enabled, I might as well
2141 * add an entry for this node. Not necessary for correctness,
2142 * but if negative caching is enabled, then the system
2143 * must care about lookup caching hit rate, so...
2145 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2146 (cnp->cn_flags & MAKEENTRY) &&
2147 attrflag != 0 && dattrflag != 0)
2148 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2156 * nfs remove directory call
2159 nfs_rmdir(struct vop_rmdir_args *ap)
2161 struct vnode *vp = ap->a_vp;
2162 struct vnode *dvp = ap->a_dvp;
2163 struct componentname *cnp = ap->a_cnp;
2164 struct nfsnode *dnp;
2165 struct nfsvattr dnfsva;
2166 int error, dattrflag;
2170 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2171 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2173 mtx_lock(&dnp->n_mtx);
2174 dnp->n_flag |= NMODIFIED;
2175 if (dattrflag != 0) {
2176 mtx_unlock(&dnp->n_mtx);
2177 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2179 dnp->n_attrstamp = 0;
2180 mtx_unlock(&dnp->n_mtx);
2181 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2186 if (error && NFS_ISV4(dvp))
2187 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2190 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2192 if (error == ENOENT)
2201 nfs_readdir(struct vop_readdir_args *ap)
2203 struct vnode *vp = ap->a_vp;
2204 struct nfsnode *np = VTONFS(vp);
2205 struct uio *uio = ap->a_uio;
2210 if (ap->a_eofflag != NULL)
2212 if (vp->v_type != VDIR)
2216 * First, check for hit on the EOF offset cache
2218 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2219 (np->n_flag & NMODIFIED) == 0) {
2220 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2221 mtx_lock(&np->n_mtx);
2222 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2223 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2224 mtx_unlock(&np->n_mtx);
2225 NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2226 if (ap->a_eofflag != NULL)
2230 mtx_unlock(&np->n_mtx);
2235 * Call ncl_bioread() to do the real work.
2237 tresid = uio->uio_resid;
2238 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2240 if (!error && uio->uio_resid == tresid) {
2241 NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2242 if (ap->a_eofflag != NULL)
2250 * Called from below the buffer cache by ncl_doio().
2253 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2256 struct nfsvattr nfsva;
2257 nfsuint64 *cookiep, cookie;
2258 struct nfsnode *dnp = VTONFS(vp);
2259 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2260 int error = 0, eof, attrflag;
2262 KASSERT(uiop->uio_iovcnt == 1 &&
2263 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2264 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2265 ("nfs readdirrpc bad uio"));
2268 * If there is no cookie, assume directory was stale.
2270 ncl_dircookie_lock(dnp);
2271 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2274 ncl_dircookie_unlock(dnp);
2276 ncl_dircookie_unlock(dnp);
2277 return (NFSERR_BAD_COOKIE);
2280 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2281 (void)ncl_fsinfo(nmp, vp, cred, td);
2283 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2284 &attrflag, &eof, NULL);
2286 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2290 * We are now either at the end of the directory or have filled
2294 dnp->n_direofoffset = uiop->uio_offset;
2296 if (uiop->uio_resid > 0)
2297 ncl_printf("EEK! readdirrpc resid > 0\n");
2298 ncl_dircookie_lock(dnp);
2299 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2301 ncl_dircookie_unlock(dnp);
2303 } else if (NFS_ISV4(vp)) {
2304 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2310 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2313 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2316 struct nfsvattr nfsva;
2317 nfsuint64 *cookiep, cookie;
2318 struct nfsnode *dnp = VTONFS(vp);
2319 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2320 int error = 0, attrflag, eof;
2322 KASSERT(uiop->uio_iovcnt == 1 &&
2323 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2324 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2325 ("nfs readdirplusrpc bad uio"));
2328 * If there is no cookie, assume directory was stale.
2330 ncl_dircookie_lock(dnp);
2331 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2334 ncl_dircookie_unlock(dnp);
2336 ncl_dircookie_unlock(dnp);
2337 return (NFSERR_BAD_COOKIE);
2340 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2341 (void)ncl_fsinfo(nmp, vp, cred, td);
2342 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2343 &attrflag, &eof, NULL);
2345 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2349 * We are now either at end of the directory or have filled the
2353 dnp->n_direofoffset = uiop->uio_offset;
2355 if (uiop->uio_resid > 0)
2356 ncl_printf("EEK! readdirplusrpc resid > 0\n");
2357 ncl_dircookie_lock(dnp);
2358 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2360 ncl_dircookie_unlock(dnp);
2362 } else if (NFS_ISV4(vp)) {
2363 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2369 * Silly rename. To make the NFS filesystem that is stateless look a little
2370 * more like the "ufs" a remove of an active vnode is translated to a rename
2371 * to a funny looking filename that is removed by nfs_inactive on the
2372 * nfsnode. There is the potential for another process on a different client
2373 * to create the same funny name between the nfs_lookitup() fails and the
2374 * nfs_rename() completes, but...
2377 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2379 struct sillyrename *sp;
2383 unsigned int lticks;
2387 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2388 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2389 M_NEWNFSREQ, M_WAITOK);
2390 sp->s_cred = crhold(cnp->cn_cred);
2395 * Fudge together a funny name.
2396 * Changing the format of the funny name to accomodate more
2397 * sillynames per directory.
2398 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2399 * CPU ticks since boot.
2401 pid = cnp->cn_thread->td_proc->p_pid;
2402 lticks = (unsigned int)ticks;
2404 sp->s_namlen = sprintf(sp->s_name,
2405 ".nfs.%08x.%04x4.4", lticks,
2407 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2408 cnp->cn_thread, NULL))
2412 error = nfs_renameit(dvp, vp, cnp, sp);
2415 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2416 cnp->cn_thread, &np);
2417 np->n_sillyrename = sp;
2422 free((caddr_t)sp, M_NEWNFSREQ);
2427 * Look up a file name and optionally either update the file handle or
2428 * allocate an nfsnode, depending on the value of npp.
2429 * npp == NULL --> just do the lookup
2430 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2432 * *npp != NULL --> update the file handle in the vnode
2435 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2436 struct thread *td, struct nfsnode **npp)
2438 struct vnode *newvp = NULL, *vp;
2439 struct nfsnode *np, *dnp = VTONFS(dvp);
2440 struct nfsfh *nfhp, *onfhp;
2441 struct nfsvattr nfsva, dnfsva;
2442 struct componentname cn;
2443 int error = 0, attrflag, dattrflag;
2446 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2447 &nfhp, &attrflag, &dattrflag, NULL);
2449 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2450 if (npp && !error) {
2455 * For NFSv4, check to see if it is the same name and
2456 * replace the name, if it is different.
2458 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2459 (np->n_v4->n4_namelen != len ||
2460 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2461 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2462 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2463 dnp->n_fhp->nfh_len))) {
2465 { char nnn[100]; int nnnl;
2466 nnnl = (len < 100) ? len : 99;
2467 bcopy(name, nnn, nnnl);
2469 printf("replace=%s\n",nnn);
2472 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2473 MALLOC(np->n_v4, struct nfsv4node *,
2474 sizeof (struct nfsv4node) +
2475 dnp->n_fhp->nfh_len + len - 1,
2476 M_NFSV4NODE, M_WAITOK);
2477 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2478 np->n_v4->n4_namelen = len;
2479 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2480 dnp->n_fhp->nfh_len);
2481 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2483 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2487 * Rehash node for new file handle.
2489 vfs_hash_rehash(vp, hash);
2492 FREE((caddr_t)onfhp, M_NFSFH);
2494 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2495 FREE((caddr_t)nfhp, M_NFSFH);
2499 cn.cn_nameptr = name;
2500 cn.cn_namelen = len;
2501 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2502 &np, NULL, LK_EXCLUSIVE);
2507 if (!attrflag && *npp == NULL) {
2515 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2518 if (npp && *npp == NULL) {
2529 if (error && NFS_ISV4(dvp))
2530 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2535 * Nfs Version 3 and 4 commit rpc
2538 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2541 struct nfsvattr nfsva;
2542 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2543 int error, attrflag;
2544 u_char verf[NFSX_VERF];
2546 mtx_lock(&nmp->nm_mtx);
2547 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2548 mtx_unlock(&nmp->nm_mtx);
2551 mtx_unlock(&nmp->nm_mtx);
2552 error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2555 mtx_lock(&nmp->nm_mtx);
2556 if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2557 NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2558 error = NFSERR_STALEWRITEVERF;
2560 mtx_unlock(&nmp->nm_mtx);
2561 if (!error && attrflag)
2562 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2564 } else if (NFS_ISV4(vp)) {
2565 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2572 * For async requests when nfsiod(s) are running, queue the request by
2573 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2577 nfs_strategy(struct vop_strategy_args *ap)
2579 struct buf *bp = ap->a_bp;
2582 KASSERT(!(bp->b_flags & B_DONE),
2583 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2584 BUF_ASSERT_HELD(bp);
2586 if (bp->b_iocmd == BIO_READ)
2592 * If the op is asynchronous and an i/o daemon is waiting
2593 * queue the request, wake it up and wait for completion
2594 * otherwise just do it ourselves.
2596 if ((bp->b_flags & B_ASYNC) == 0 ||
2597 ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2598 (void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2603 * fsync vnode op. Just call ncl_flush() with commit == 1.
2607 nfs_fsync(struct vop_fsync_args *ap)
2610 if (ap->a_vp->v_type != VREG) {
2612 * For NFS, metadata is changed synchronously on the server,
2613 * so there is nothing to flush. Also, ncl_flush() clears
2614 * the NMODIFIED flag and that shouldn't be done here for
2619 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2623 * Flush all the blocks associated with a vnode.
2624 * Walk through the buffer pool and push any dirty pages
2625 * associated with the vnode.
2626 * If the called_from_renewthread argument is TRUE, it has been called
2627 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2628 * waiting for a buffer write to complete.
2631 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2632 int commit, int called_from_renewthread)
2634 struct nfsnode *np = VTONFS(vp);
2638 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2639 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2640 int passone = 1, trycnt = 0;
2641 u_quad_t off, endoff, toff;
2642 struct ucred* wcred = NULL;
2643 struct buf **bvec = NULL;
2645 #ifndef NFS_COMMITBVECSIZ
2646 #define NFS_COMMITBVECSIZ 20
2648 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2649 int bvecsize = 0, bveccount;
2651 if (called_from_renewthread != 0)
2653 if (nmp->nm_flag & NFSMNT_INT)
2654 slpflag = NFS_PCATCH;
2659 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2660 * server, but has not been committed to stable storage on the server
2661 * yet. On the first pass, the byte range is worked out and the commit
2662 * rpc is done. On the second pass, ncl_writebp() is called to do the
2669 if (NFS_ISV34(vp) && commit) {
2670 if (bvec != NULL && bvec != bvec_on_stack)
2673 * Count up how many buffers waiting for a commit.
2677 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2678 if (!BUF_ISLOCKED(bp) &&
2679 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2680 == (B_DELWRI | B_NEEDCOMMIT))
2684 * Allocate space to remember the list of bufs to commit. It is
2685 * important to use M_NOWAIT here to avoid a race with nfs_write.
2686 * If we can't get memory (for whatever reason), we will end up
2687 * committing the buffers one-by-one in the loop below.
2689 if (bveccount > NFS_COMMITBVECSIZ) {
2691 * Release the vnode interlock to avoid a lock
2695 bvec = (struct buf **)
2696 malloc(bveccount * sizeof(struct buf *),
2700 bvec = bvec_on_stack;
2701 bvecsize = NFS_COMMITBVECSIZ;
2703 bvecsize = bveccount;
2705 bvec = bvec_on_stack;
2706 bvecsize = NFS_COMMITBVECSIZ;
2708 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2709 if (bvecpos >= bvecsize)
2711 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2712 nbp = TAILQ_NEXT(bp, b_bobufs);
2715 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2716 (B_DELWRI | B_NEEDCOMMIT)) {
2718 nbp = TAILQ_NEXT(bp, b_bobufs);
2724 * Work out if all buffers are using the same cred
2725 * so we can deal with them all with one commit.
2727 * NOTE: we are not clearing B_DONE here, so we have
2728 * to do it later on in this routine if we intend to
2729 * initiate I/O on the bp.
2731 * Note: to avoid loopback deadlocks, we do not
2732 * assign b_runningbufspace.
2735 wcred = bp->b_wcred;
2736 else if (wcred != bp->b_wcred)
2738 vfs_busy_pages(bp, 1);
2742 * bp is protected by being locked, but nbp is not
2743 * and vfs_busy_pages() may sleep. We have to
2746 nbp = TAILQ_NEXT(bp, b_bobufs);
2749 * A list of these buffers is kept so that the
2750 * second loop knows which buffers have actually
2751 * been committed. This is necessary, since there
2752 * may be a race between the commit rpc and new
2753 * uncommitted writes on the file.
2755 bvec[bvecpos++] = bp;
2756 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2760 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2768 * Commit data on the server, as required.
2769 * If all bufs are using the same wcred, then use that with
2770 * one call for all of them, otherwise commit each one
2773 if (wcred != NOCRED)
2774 retv = ncl_commit(vp, off, (int)(endoff - off),
2778 for (i = 0; i < bvecpos; i++) {
2781 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2783 size = (u_quad_t)(bp->b_dirtyend
2785 retv = ncl_commit(vp, off, (int)size,
2791 if (retv == NFSERR_STALEWRITEVERF)
2792 ncl_clearcommit(vp->v_mount);
2795 * Now, either mark the blocks I/O done or mark the
2796 * blocks dirty, depending on whether the commit
2799 for (i = 0; i < bvecpos; i++) {
2801 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2804 * Error, leave B_DELWRI intact
2806 vfs_unbusy_pages(bp);
2810 * Success, remove B_DELWRI ( bundirty() ).
2812 * b_dirtyoff/b_dirtyend seem to be NFS
2813 * specific. We should probably move that
2814 * into bundirty(). XXX
2817 bp->b_flags |= B_ASYNC;
2819 bp->b_flags &= ~B_DONE;
2820 bp->b_ioflags &= ~BIO_ERROR;
2821 bp->b_dirtyoff = bp->b_dirtyend = 0;
2828 * Start/do any write(s) that are required.
2832 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2833 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2834 if (waitfor != MNT_WAIT || passone)
2837 error = BUF_TIMELOCK(bp,
2838 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2839 BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2844 if (error == ENOLCK) {
2848 if (called_from_renewthread != 0) {
2850 * Return EIO so the flush will be retried
2856 if (newnfs_sigintr(nmp, td)) {
2860 if (slpflag & PCATCH) {
2866 if ((bp->b_flags & B_DELWRI) == 0)
2867 panic("nfs_fsync: not dirty");
2868 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2874 if (passone || !commit)
2875 bp->b_flags |= B_ASYNC;
2877 bp->b_flags |= B_ASYNC;
2879 if (newnfs_sigintr(nmp, td)) {
2890 if (waitfor == MNT_WAIT) {
2891 while (bo->bo_numoutput) {
2892 error = bufobj_wwait(bo, slpflag, slptimeo);
2895 if (called_from_renewthread != 0) {
2897 * Return EIO so that the flush will be
2903 error = newnfs_sigintr(nmp, td);
2906 if (slpflag & PCATCH) {
2913 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2918 * Wait for all the async IO requests to drain
2921 mtx_lock(&np->n_mtx);
2922 while (np->n_directio_asyncwr > 0) {
2923 np->n_flag |= NFSYNCWAIT;
2924 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2925 &np->n_mtx, slpflag | (PRIBIO + 1),
2928 if (newnfs_sigintr(nmp, td)) {
2929 mtx_unlock(&np->n_mtx);
2935 mtx_unlock(&np->n_mtx);
2938 mtx_lock(&np->n_mtx);
2939 if (np->n_flag & NWRITEERR) {
2940 error = np->n_error;
2941 np->n_flag &= ~NWRITEERR;
2943 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2944 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2945 np->n_flag &= ~NMODIFIED;
2946 mtx_unlock(&np->n_mtx);
2948 if (bvec != NULL && bvec != bvec_on_stack)
2950 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2951 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2952 np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2953 /* try, try again... */
2958 printf("try%d\n", trycnt);
2965 * NFS advisory byte-level locks.
2968 nfs_advlock(struct vop_advlock_args *ap)
2970 struct vnode *vp = ap->a_vp;
2972 struct nfsnode *np = VTONFS(ap->a_vp);
2973 struct proc *p = (struct proc *)ap->a_id;
2974 struct thread *td = curthread; /* XXX */
2976 int ret, error = EOPNOTSUPP;
2979 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
2980 if (vp->v_type != VREG)
2982 if ((ap->a_flags & F_POSIX) != 0)
2985 cred = td->td_ucred;
2986 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2987 if (vp->v_iflag & VI_DOOMED) {
2988 NFSVOPUNLOCK(vp, 0);
2993 * If this is unlocking a write locked region, flush and
2994 * commit them before unlocking. This is required by
2995 * RFC3530 Sec. 9.3.2.
2997 if (ap->a_op == F_UNLCK &&
2998 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3000 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
3003 * Loop around doing the lock op, while a blocking lock
3004 * must wait for the lock op to succeed.
3007 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3008 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3009 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3010 ap->a_op == F_SETLK) {
3011 NFSVOPUNLOCK(vp, 0);
3012 error = nfs_catnap(PZERO | PCATCH, ret,
3016 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3017 if (vp->v_iflag & VI_DOOMED) {
3018 NFSVOPUNLOCK(vp, 0);
3022 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3023 ap->a_op == F_SETLK);
3024 if (ret == NFSERR_DENIED) {
3025 NFSVOPUNLOCK(vp, 0);
3027 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3028 NFSVOPUNLOCK(vp, 0);
3030 } else if (ret != 0) {
3031 NFSVOPUNLOCK(vp, 0);
3036 * Now, if we just got a lock, invalidate data in the buffer
3037 * cache, as required, so that the coherency conforms with
3038 * RFC3530 Sec. 9.3.2.
3040 if (ap->a_op == F_SETLK) {
3041 if ((np->n_flag & NMODIFIED) == 0) {
3042 np->n_attrstamp = 0;
3043 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3044 ret = VOP_GETATTR(vp, &va, cred);
3046 if ((np->n_flag & NMODIFIED) || ret ||
3047 np->n_change != va.va_filerev) {
3048 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3049 np->n_attrstamp = 0;
3050 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3051 ret = VOP_GETATTR(vp, &va, cred);
3053 np->n_mtime = va.va_mtime;
3054 np->n_change = va.va_filerev;
3058 NFSVOPUNLOCK(vp, 0);
3060 } else if (!NFS_ISV4(vp)) {
3061 error = NFSVOPLOCK(vp, LK_SHARED);
3064 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3065 size = VTONFS(vp)->n_size;
3066 NFSVOPUNLOCK(vp, 0);
3067 error = lf_advlock(ap, &(vp->v_lockf), size);
3069 if (nfs_advlock_p != NULL)
3070 error = nfs_advlock_p(ap);
3072 NFSVOPUNLOCK(vp, 0);
3081 * NFS advisory byte-level locks.
3084 nfs_advlockasync(struct vop_advlockasync_args *ap)
3086 struct vnode *vp = ap->a_vp;
3091 return (EOPNOTSUPP);
3092 error = NFSVOPLOCK(vp, LK_SHARED);
3095 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3096 size = VTONFS(vp)->n_size;
3097 NFSVOPUNLOCK(vp, 0);
3098 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3100 NFSVOPUNLOCK(vp, 0);
3107 * Print out the contents of an nfsnode.
3110 nfs_print(struct vop_print_args *ap)
3112 struct vnode *vp = ap->a_vp;
3113 struct nfsnode *np = VTONFS(vp);
3115 ncl_printf("\tfileid %ld fsid 0x%x",
3116 np->n_vattr.na_fileid, np->n_vattr.na_fsid);
3117 if (vp->v_type == VFIFO)
3124 * This is the "real" nfs::bwrite(struct buf*).
3125 * We set B_CACHE if this is a VMIO buffer.
3128 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3131 int oldflags = bp->b_flags;
3137 BUF_ASSERT_HELD(bp);
3139 if (bp->b_flags & B_INVAL) {
3144 bp->b_flags |= B_CACHE;
3147 * Undirty the bp. We will redirty it later if the I/O fails.
3152 bp->b_flags &= ~B_DONE;
3153 bp->b_ioflags &= ~BIO_ERROR;
3154 bp->b_iocmd = BIO_WRITE;
3156 bufobj_wref(bp->b_bufobj);
3157 curthread->td_ru.ru_oublock++;
3161 * Note: to avoid loopback deadlocks, we do not
3162 * assign b_runningbufspace.
3164 vfs_busy_pages(bp, 1);
3167 bp->b_iooffset = dbtob(bp->b_blkno);
3170 if( (oldflags & B_ASYNC) == 0) {
3171 int rtval = bufwait(bp);
3173 if (oldflags & B_DELWRI) {
3186 * nfs special file access vnode op.
3187 * Essentially just get vattr and then imitate iaccess() since the device is
3188 * local to the client.
3191 nfsspec_access(struct vop_access_args *ap)
3194 struct ucred *cred = ap->a_cred;
3195 struct vnode *vp = ap->a_vp;
3196 accmode_t accmode = ap->a_accmode;
3201 * Disallow write attempts on filesystems mounted read-only;
3202 * unless the file is a socket, fifo, or a block or character
3203 * device resident on the filesystem.
3205 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3206 switch (vp->v_type) {
3216 error = VOP_GETATTR(vp, vap, cred);
3219 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3220 accmode, cred, NULL);
3226 * Read wrapper for fifos.
3229 nfsfifo_read(struct vop_read_args *ap)
3231 struct nfsnode *np = VTONFS(ap->a_vp);
3237 mtx_lock(&np->n_mtx);
3239 vfs_timestamp(&np->n_atim);
3240 mtx_unlock(&np->n_mtx);
3241 error = fifo_specops.vop_read(ap);
3246 * Write wrapper for fifos.
3249 nfsfifo_write(struct vop_write_args *ap)
3251 struct nfsnode *np = VTONFS(ap->a_vp);
3256 mtx_lock(&np->n_mtx);
3258 vfs_timestamp(&np->n_mtim);
3259 mtx_unlock(&np->n_mtx);
3260 return(fifo_specops.vop_write(ap));
3264 * Close wrapper for fifos.
3266 * Update the times on the nfsnode then do fifo close.
3269 nfsfifo_close(struct vop_close_args *ap)
3271 struct vnode *vp = ap->a_vp;
3272 struct nfsnode *np = VTONFS(vp);
3276 mtx_lock(&np->n_mtx);
3277 if (np->n_flag & (NACC | NUPD)) {
3279 if (np->n_flag & NACC)
3281 if (np->n_flag & NUPD)
3284 if (vrefcnt(vp) == 1 &&
3285 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3287 if (np->n_flag & NACC)
3288 vattr.va_atime = np->n_atim;
3289 if (np->n_flag & NUPD)
3290 vattr.va_mtime = np->n_mtim;
3291 mtx_unlock(&np->n_mtx);
3292 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3296 mtx_unlock(&np->n_mtx);
3298 return (fifo_specops.vop_close(ap));
3302 * Just call ncl_writebp() with the force argument set to 1.
3304 * NOTE: B_DONE may or may not be set in a_bp on call.
3307 nfs_bwrite(struct buf *bp)
3310 return (ncl_writebp(bp, 1, curthread));
3313 struct buf_ops buf_ops_newnfs = {
3314 .bop_name = "buf_ops_nfs",
3315 .bop_write = nfs_bwrite,
3316 .bop_strategy = bufstrategy,
3317 .bop_sync = bufsync,
3318 .bop_bdflush = bufbdflush,
3322 * Cloned from vop_stdlock(), and then the ugly hack added.
3325 nfs_lock1(struct vop_lock1_args *ap)
3327 struct vnode *vp = ap->a_vp;
3331 * Since vfs_hash_get() calls vget() and it will no longer work
3332 * for FreeBSD8 with flags == 0, I can only think of this horrible
3333 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3334 * and then handle it here. All I want for this case is a v_usecount
3335 * on the vnode to use for recovery, while another thread might
3336 * hold a lock on the vnode. I have the other threads blocked, so
3337 * there isn't any race problem.
3339 if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3340 if ((ap->a_flags & LK_INTERLOCK) == 0)
3342 if ((vp->v_iflag & VI_DOOMED))
3347 return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3348 LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3353 nfs_getacl(struct vop_getacl_args *ap)
3357 if (ap->a_type != ACL_TYPE_NFS4)
3358 return (EOPNOTSUPP);
3359 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3361 if (error > NFSERR_STALE) {
3362 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3369 nfs_setacl(struct vop_setacl_args *ap)
3373 if (ap->a_type != ACL_TYPE_NFS4)
3374 return (EOPNOTSUPP);
3375 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3377 if (error > NFSERR_STALE) {
3378 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3385 * Return POSIX pathconf information applicable to nfs filesystems.
3388 nfs_pathconf(struct vop_pathconf_args *ap)
3390 struct nfsv3_pathconf pc;
3391 struct nfsvattr nfsva;
3392 struct vnode *vp = ap->a_vp;
3393 struct thread *td = curthread;
3394 int attrflag, error;
3396 if (NFS_ISV4(vp) || (NFS_ISV3(vp) && (ap->a_name == _PC_LINK_MAX ||
3397 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3398 ap->a_name == _PC_NO_TRUNC))) {
3400 * Since only the above 4 a_names are returned by the NFSv3
3401 * Pathconf RPC, there is no point in doing it for others.
3403 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3406 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3412 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3415 pc.pc_linkmax = LINK_MAX;
3416 pc.pc_namemax = NFS_MAXNAMLEN;
3418 pc.pc_chownrestricted = 1;
3419 pc.pc_caseinsensitive = 0;
3420 pc.pc_casepreserving = 1;
3423 switch (ap->a_name) {
3425 *ap->a_retval = pc.pc_linkmax;
3428 *ap->a_retval = pc.pc_namemax;
3431 *ap->a_retval = PATH_MAX;
3434 *ap->a_retval = PIPE_BUF;
3436 case _PC_CHOWN_RESTRICTED:
3437 *ap->a_retval = pc.pc_chownrestricted;
3440 *ap->a_retval = pc.pc_notrunc;
3442 case _PC_ACL_EXTENDED:
3446 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3447 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3452 case _PC_ACL_PATH_MAX:
3454 *ap->a_retval = ACL_MAX_ENTRIES;
3458 case _PC_MAC_PRESENT:
3462 /* _PC_ASYNC_IO should have been handled by upper layers. */
3463 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3472 case _PC_ALLOC_SIZE_MIN:
3473 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3475 case _PC_FILESIZEBITS:
3481 case _PC_REC_INCR_XFER_SIZE:
3482 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3484 case _PC_REC_MAX_XFER_SIZE:
3485 *ap->a_retval = -1; /* means ``unlimited'' */
3487 case _PC_REC_MIN_XFER_SIZE:
3488 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3490 case _PC_REC_XFER_ALIGN:
3491 *ap->a_retval = PAGE_SIZE;
3493 case _PC_SYMLINK_MAX:
3494 *ap->a_retval = NFS_MAXPATHLEN;