2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * from nfs_vnops.c 8.16 (Berkeley) 5/27/95
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
41 * vnode op calls for Sun NFS version 2, 3 and 4
46 #include <sys/param.h>
47 #include <sys/kernel.h>
48 #include <sys/systm.h>
49 #include <sys/resourcevar.h>
51 #include <sys/mount.h>
54 #include <sys/extattr.h>
55 #include <sys/filio.h>
57 #include <sys/malloc.h>
59 #include <sys/namei.h>
60 #include <sys/socket.h>
61 #include <sys/vnode.h>
62 #include <sys/dirent.h>
63 #include <sys/fcntl.h>
64 #include <sys/lockf.h>
66 #include <sys/sysctl.h>
67 #include <sys/signalvar.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_object.h>
73 #include <fs/nfs/nfsport.h>
74 #include <fs/nfsclient/nfsnode.h>
75 #include <fs/nfsclient/nfsmount.h>
76 #include <fs/nfsclient/nfs.h>
77 #include <fs/nfsclient/nfs_kdtrace.h>
80 #include <netinet/in.h>
81 #include <netinet/in_var.h>
83 #include <nfs/nfs_lock.h>
86 #include <sys/dtrace_bsd.h>
88 dtrace_nfsclient_accesscache_flush_probe_func_t
89 dtrace_nfscl_accesscache_flush_done_probe;
90 uint32_t nfscl_accesscache_flush_done_id;
92 dtrace_nfsclient_accesscache_get_probe_func_t
93 dtrace_nfscl_accesscache_get_hit_probe,
94 dtrace_nfscl_accesscache_get_miss_probe;
95 uint32_t nfscl_accesscache_get_hit_id;
96 uint32_t nfscl_accesscache_get_miss_id;
98 dtrace_nfsclient_accesscache_load_probe_func_t
99 dtrace_nfscl_accesscache_load_done_probe;
100 uint32_t nfscl_accesscache_load_done_id;
101 #endif /* !KDTRACE_HOOKS */
107 extern struct nfsstatsv1 nfsstatsv1;
108 extern int nfsrv_useacl;
109 extern int nfscl_debuglevel;
110 MALLOC_DECLARE(M_NEWNFSREQ);
112 static vop_read_t nfsfifo_read;
113 static vop_write_t nfsfifo_write;
114 static vop_close_t nfsfifo_close;
115 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
117 static vop_lookup_t nfs_lookup;
118 static vop_create_t nfs_create;
119 static vop_mknod_t nfs_mknod;
120 static vop_open_t nfs_open;
121 static vop_pathconf_t nfs_pathconf;
122 static vop_close_t nfs_close;
123 static vop_access_t nfs_access;
124 static vop_getattr_t nfs_getattr;
125 static vop_setattr_t nfs_setattr;
126 static vop_read_t nfs_read;
127 static vop_fsync_t nfs_fsync;
128 static vop_remove_t nfs_remove;
129 static vop_link_t nfs_link;
130 static vop_rename_t nfs_rename;
131 static vop_mkdir_t nfs_mkdir;
132 static vop_rmdir_t nfs_rmdir;
133 static vop_symlink_t nfs_symlink;
134 static vop_readdir_t nfs_readdir;
135 static vop_strategy_t nfs_strategy;
136 static int nfs_lookitup(struct vnode *, char *, int,
137 struct ucred *, struct thread *, struct nfsnode **);
138 static int nfs_sillyrename(struct vnode *, struct vnode *,
139 struct componentname *);
140 static vop_access_t nfsspec_access;
141 static vop_readlink_t nfs_readlink;
142 static vop_print_t nfs_print;
143 static vop_advlock_t nfs_advlock;
144 static vop_advlockasync_t nfs_advlockasync;
145 static vop_getacl_t nfs_getacl;
146 static vop_setacl_t nfs_setacl;
147 static vop_advise_t nfs_advise;
148 static vop_allocate_t nfs_allocate;
149 static vop_copy_file_range_t nfs_copy_file_range;
150 static vop_ioctl_t nfs_ioctl;
151 static vop_getextattr_t nfs_getextattr;
152 static vop_setextattr_t nfs_setextattr;
153 static vop_listextattr_t nfs_listextattr;
154 static vop_deleteextattr_t nfs_deleteextattr;
155 static vop_lock1_t nfs_lock;
158 * Global vfs data structures for nfs
161 static struct vop_vector newnfs_vnodeops_nosig = {
162 .vop_default = &default_vnodeops,
163 .vop_access = nfs_access,
164 .vop_advlock = nfs_advlock,
165 .vop_advlockasync = nfs_advlockasync,
166 .vop_close = nfs_close,
167 .vop_create = nfs_create,
168 .vop_fsync = nfs_fsync,
169 .vop_getattr = nfs_getattr,
170 .vop_getpages = ncl_getpages,
171 .vop_putpages = ncl_putpages,
172 .vop_inactive = ncl_inactive,
173 .vop_link = nfs_link,
174 .vop_lock1 = nfs_lock,
175 .vop_lookup = nfs_lookup,
176 .vop_mkdir = nfs_mkdir,
177 .vop_mknod = nfs_mknod,
178 .vop_open = nfs_open,
179 .vop_pathconf = nfs_pathconf,
180 .vop_print = nfs_print,
181 .vop_read = nfs_read,
182 .vop_readdir = nfs_readdir,
183 .vop_readlink = nfs_readlink,
184 .vop_reclaim = ncl_reclaim,
185 .vop_remove = nfs_remove,
186 .vop_rename = nfs_rename,
187 .vop_rmdir = nfs_rmdir,
188 .vop_setattr = nfs_setattr,
189 .vop_strategy = nfs_strategy,
190 .vop_symlink = nfs_symlink,
191 .vop_write = ncl_write,
192 .vop_getacl = nfs_getacl,
193 .vop_setacl = nfs_setacl,
194 .vop_advise = nfs_advise,
195 .vop_allocate = nfs_allocate,
196 .vop_copy_file_range = nfs_copy_file_range,
197 .vop_ioctl = nfs_ioctl,
198 .vop_getextattr = nfs_getextattr,
199 .vop_setextattr = nfs_setextattr,
200 .vop_listextattr = nfs_listextattr,
201 .vop_deleteextattr = nfs_deleteextattr,
203 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig);
206 nfs_vnodeops_bypass(struct vop_generic_args *a)
209 return (vop_sigdefer(&newnfs_vnodeops_nosig, a));
212 struct vop_vector newnfs_vnodeops = {
213 .vop_default = &default_vnodeops,
214 .vop_bypass = nfs_vnodeops_bypass,
216 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops);
218 static struct vop_vector newnfs_fifoops_nosig = {
219 .vop_default = &fifo_specops,
220 .vop_access = nfsspec_access,
221 .vop_close = nfsfifo_close,
222 .vop_fsync = nfs_fsync,
223 .vop_getattr = nfs_getattr,
224 .vop_inactive = ncl_inactive,
225 .vop_pathconf = nfs_pathconf,
226 .vop_print = nfs_print,
227 .vop_read = nfsfifo_read,
228 .vop_reclaim = ncl_reclaim,
229 .vop_setattr = nfs_setattr,
230 .vop_write = nfsfifo_write,
232 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig);
235 nfs_fifoops_bypass(struct vop_generic_args *a)
238 return (vop_sigdefer(&newnfs_fifoops_nosig, a));
241 struct vop_vector newnfs_fifoops = {
242 .vop_default = &default_vnodeops,
243 .vop_bypass = nfs_fifoops_bypass,
245 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops);
247 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
248 struct componentname *cnp, struct vattr *vap);
249 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
250 int namelen, struct ucred *cred, struct thread *td);
251 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
252 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
253 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
254 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
255 struct componentname *scnp, struct sillyrename *sp);
260 SYSCTL_DECL(_vfs_nfs);
262 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
263 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
264 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
266 static int nfs_prime_access_cache = 0;
267 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
268 &nfs_prime_access_cache, 0,
269 "Prime NFS ACCESS cache when fetching attributes");
271 static int newnfs_commit_on_close = 0;
272 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
273 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
275 static int nfs_clean_pages_on_close = 1;
276 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
277 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
279 int newnfs_directio_enable = 0;
280 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
281 &newnfs_directio_enable, 0, "Enable NFS directio");
283 int nfs_keep_dirty_on_error;
284 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
285 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
288 * This sysctl allows other processes to mmap a file that has been opened
289 * O_DIRECT by a process. In general, having processes mmap the file while
290 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
291 * this by default to prevent DoS attacks - to prevent a malicious user from
292 * opening up files O_DIRECT preventing other users from mmap'ing these
293 * files. "Protected" environments where stricter consistency guarantees are
294 * required can disable this knob. The process that opened the file O_DIRECT
295 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
298 int newnfs_directio_allow_mmap = 1;
299 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
300 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
302 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
303 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
304 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
308 * The list of locks after the description of the lock is the ordering
309 * of other locks acquired with the lock held.
310 * np->n_mtx : Protects the fields in the nfsnode.
312 VI_MTX (acquired indirectly)
313 * nmp->nm_mtx : Protects the fields in the nfsmount.
315 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
316 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
319 * rep->r_mtx : Protects the fields in an nfsreq.
323 nfs_lock(struct vop_lock1_args *ap)
332 lktype = ap->a_flags & LK_TYPE_MASK;
333 error = VOP_LOCK1_APV(&default_vnodeops, ap);
334 if (error != 0 || vp->v_op != &newnfs_vnodeops)
340 if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED &&
341 lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE &&
342 lktype != LK_TRYUPGRADE)) {
346 onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT &&
347 (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE &&
348 (lktype == LK_SHARED || lktype == LK_EXCLUSIVE);
349 if (onfault && vp->v_vnlock->lk_recurse == 0) {
351 * Force retry in vm_fault(), to make the lock request
352 * sleepable, which allows us to piggy-back the
353 * sleepable call to vnode_pager_setsize().
359 if ((ap->a_flags & LK_NOWAIT) != 0 ||
360 (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) {
364 if (lktype == LK_SHARED) {
367 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
368 ap->a_flags |= LK_EXCLUSIVE;
369 error = VOP_LOCK1_APV(&default_vnodeops, ap);
370 if (error != 0 || vp->v_op != &newnfs_vnodeops)
372 if (vp->v_data == NULL)
374 MPASS(vp->v_data == np);
376 if ((np->n_flag & NVNSETSZSKIP) == 0) {
381 np->n_flag &= ~NVNSETSZSKIP;
384 vnode_pager_setsize(vp, nsize);
386 if (lktype == LK_SHARED) {
387 ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
388 ap->a_flags |= LK_DOWNGRADE;
389 (void)VOP_LOCK1_APV(&default_vnodeops, ap);
395 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
396 struct ucred *cred, u_int32_t *retmode)
398 int error = 0, attrflag, i, lrupos;
400 struct nfsnode *np = VTONFS(vp);
401 struct nfsvattr nfsva;
403 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
406 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
410 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
411 if (np->n_accesscache[i].uid == cred->cr_uid) {
412 np->n_accesscache[i].mode = rmode;
413 np->n_accesscache[i].stamp = time_second;
416 if (i > 0 && np->n_accesscache[i].stamp <
417 np->n_accesscache[lrupos].stamp)
420 if (i == NFS_ACCESSCACHESIZE) {
421 np->n_accesscache[lrupos].uid = cred->cr_uid;
422 np->n_accesscache[lrupos].mode = rmode;
423 np->n_accesscache[lrupos].stamp = time_second;
428 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
429 } else if (NFS_ISV4(vp)) {
430 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
434 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
441 * nfs access vnode op.
442 * For nfs version 2, just return ok. File accesses may fail later.
443 * For nfs version 3, use the access rpc to check accessibility. If file modes
444 * are changed on the server, accesses might still fail later.
447 nfs_access(struct vop_access_args *ap)
449 struct vnode *vp = ap->a_vp;
450 int error = 0, i, gotahit;
451 u_int32_t mode, wmode, rmode;
452 int v34 = NFS_ISV34(vp);
453 struct nfsnode *np = VTONFS(vp);
456 * Disallow write attempts on filesystems mounted read-only;
457 * unless the file is a socket, fifo, or a block or character
458 * device resident on the filesystem.
460 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
461 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
462 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
463 switch (vp->v_type) {
473 * For nfs v3 or v4, check to see if we have done this recently, and if
474 * so return our cached result instead of making an ACCESS call.
475 * If not, do an access rpc, otherwise you are stuck emulating
476 * ufs_access() locally using the vattr. This may not be correct,
477 * since the server may apply other access criteria such as
478 * client uid-->server uid mapping that we do not know about.
481 if (ap->a_accmode & VREAD)
482 mode = NFSACCESS_READ;
485 if (vp->v_type != VDIR) {
486 if (ap->a_accmode & VWRITE)
487 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
488 if (ap->a_accmode & VAPPEND)
489 mode |= NFSACCESS_EXTEND;
490 if (ap->a_accmode & VEXEC)
491 mode |= NFSACCESS_EXECUTE;
492 if (ap->a_accmode & VDELETE)
493 mode |= NFSACCESS_DELETE;
495 if (ap->a_accmode & VWRITE)
496 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
497 if (ap->a_accmode & VAPPEND)
498 mode |= NFSACCESS_EXTEND;
499 if (ap->a_accmode & VEXEC)
500 mode |= NFSACCESS_LOOKUP;
501 if (ap->a_accmode & VDELETE)
502 mode |= NFSACCESS_DELETE;
503 if (ap->a_accmode & VDELETE_CHILD)
504 mode |= NFSACCESS_MODIFY;
506 /* XXX safety belt, only make blanket request if caching */
507 if (nfsaccess_cache_timeout > 0) {
508 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
509 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
510 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
516 * Does our cached result allow us to give a definite yes to
521 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
522 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
523 if (time_second < (np->n_accesscache[i].stamp
524 + nfsaccess_cache_timeout) &&
525 (np->n_accesscache[i].mode & mode) == mode) {
526 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
535 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
536 ap->a_cred->cr_uid, mode);
538 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
539 ap->a_cred->cr_uid, mode);
543 * Either a no, or a don't know. Go to the wire.
545 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
546 error = nfs34_access_otw(vp, wmode, ap->a_td,
549 (rmode & mode) != mode)
554 if ((error = nfsspec_access(ap)) != 0) {
558 * Attempt to prevent a mapped root from accessing a file
559 * which it shouldn't. We try to read a byte from the file
560 * if the user is root and the file is not zero length.
561 * After calling nfsspec_access, we should have the correct
565 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
566 && VTONFS(vp)->n_size > 0) {
574 auio.uio_iov = &aiov;
578 auio.uio_segflg = UIO_SYSSPACE;
579 auio.uio_rw = UIO_READ;
580 auio.uio_td = ap->a_td;
582 if (vp->v_type == VREG)
583 error = ncl_readrpc(vp, &auio, ap->a_cred);
584 else if (vp->v_type == VDIR) {
586 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
588 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
589 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
592 } else if (vp->v_type == VLNK)
593 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
604 * Check to see if the type is ok
605 * and that deletion is not in progress.
606 * For paged in text files, you will need to flush the page cache
607 * if consistency is lost.
611 nfs_open(struct vop_open_args *ap)
613 struct vnode *vp = ap->a_vp;
614 struct nfsnode *np = VTONFS(vp);
617 int fmode = ap->a_mode;
621 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
625 * For NFSv4, we need to do the Open Op before cache validation,
626 * so that we conform to RFC3530 Sec. 9.3.1.
629 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
631 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
638 * Now, if this Open will be doing reading, re-validate/flush the
639 * cache, so that Close/Open coherency is maintained.
642 if (np->n_flag & NMODIFIED) {
644 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
645 if (error == EINTR || error == EIO) {
647 (void) nfsrpc_close(vp, 0, ap->a_td);
652 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
653 if (vp->v_type == VDIR)
654 np->n_direofoffset = 0;
656 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
659 (void) nfsrpc_close(vp, 0, ap->a_td);
663 np->n_mtime = vattr.va_mtime;
665 np->n_change = vattr.va_filerev;
668 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
671 (void) nfsrpc_close(vp, 0, ap->a_td);
675 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
676 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
677 if (vp->v_type == VDIR)
678 np->n_direofoffset = 0;
680 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
681 if (error == EINTR || error == EIO) {
683 (void) nfsrpc_close(vp, 0, ap->a_td);
687 np->n_mtime = vattr.va_mtime;
689 np->n_change = vattr.va_filerev;
694 * If the object has >= 1 O_DIRECT active opens, we disable caching.
696 if (newnfs_directio_enable && (fmode & O_DIRECT) &&
697 (vp->v_type == VREG)) {
698 if (np->n_directio_opens == 0) {
700 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
703 (void) nfsrpc_close(vp, 0, ap->a_td);
707 np->n_flag |= NNONCACHE;
709 np->n_directio_opens++;
712 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
713 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
714 np->n_flag |= NWRITEOPENED;
717 * If this is an open for writing, capture a reference to the
718 * credentials, so they can be used by ncl_putpages(). Using
719 * these write credentials is preferable to the credentials of
720 * whatever thread happens to be doing the VOP_PUTPAGES() since
721 * the write RPCs are less likely to fail with EACCES.
723 if ((fmode & FWRITE) != 0) {
724 cred = np->n_writecred;
725 np->n_writecred = crhold(ap->a_cred);
732 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
735 * If the text file has been mmap'd, flush any dirty pages to the
736 * buffer cache and then...
737 * Make sure all writes are pushed to the NFS server. If this is not
738 * done, the modify time of the file can change while the text
739 * file is being executed. This will cause the process that is
740 * executing the text file to be terminated.
742 if (vp->v_writecount <= -1) {
743 if ((obj = vp->v_object) != NULL &&
744 vm_object_mightbedirty(obj)) {
745 VM_OBJECT_WLOCK(obj);
746 vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
747 VM_OBJECT_WUNLOCK(obj);
750 /* Now, flush the buffer cache. */
751 ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
753 /* And, finally, make sure that n_mtime is up to date. */
756 np->n_mtime = np->n_vattr.na_mtime;
764 * What an NFS client should do upon close after writing is a debatable issue.
765 * Most NFS clients push delayed writes to the server upon close, basically for
767 * 1 - So that any write errors may be reported back to the client process
768 * doing the close system call. By far the two most likely errors are
769 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
770 * 2 - To put a worst case upper bound on cache inconsistency between
771 * multiple clients for the file.
772 * There is also a consistency problem for Version 2 of the protocol w.r.t.
773 * not being able to tell if other clients are writing a file concurrently,
774 * since there is no way of knowing if the changed modify time in the reply
775 * is only due to the write for this client.
776 * (NFS Version 3 provides weak cache consistency data in the reply that
777 * should be sufficient to detect and handle this case.)
779 * The current code does the following:
780 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
781 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
782 * or commit them (this satisfies 1 and 2 except for the
783 * case where the server crashes after this close but
784 * before the commit RPC, which is felt to be "good
785 * enough". Changing the last argument to ncl_flush() to
786 * a 1 would force a commit operation, if it is felt a
787 * commit is necessary now.
788 * for NFS Version 4 - flush the dirty buffers and commit them, if
789 * nfscl_mustflush() says this is necessary.
790 * It is necessary if there is no write delegation held,
791 * in order to satisfy open/close coherency.
792 * If the file isn't cached on local stable storage,
793 * it may be necessary in order to detect "out of space"
794 * errors from the server, if the write delegation
795 * issued by the server doesn't allow the file to grow.
799 nfs_close(struct vop_close_args *ap)
801 struct vnode *vp = ap->a_vp;
802 struct nfsnode *np = VTONFS(vp);
803 struct nfsvattr nfsva;
805 int error = 0, ret, localcred = 0;
806 int fmode = ap->a_fflag;
808 if (NFSCL_FORCEDISM(vp->v_mount))
811 * During shutdown, a_cred isn't valid, so just use root.
813 if (ap->a_cred == NOCRED) {
814 cred = newnfs_getcred();
819 if (vp->v_type == VREG) {
821 * Examine and clean dirty pages, regardless of NMODIFIED.
822 * This closes a major hole in close-to-open consistency.
823 * We want to push out all dirty pages (and buffers) on
824 * close, regardless of whether they were dirtied by
825 * mmap'ed writes or via write().
827 if (nfs_clean_pages_on_close && vp->v_object) {
828 VM_OBJECT_WLOCK(vp->v_object);
829 vm_object_page_clean(vp->v_object, 0, 0, 0);
830 VM_OBJECT_WUNLOCK(vp->v_object);
833 if (np->n_flag & NMODIFIED) {
837 * Under NFSv3 we have dirty buffers to dispose of. We
838 * must flush them to the NFS server. We have the option
839 * of waiting all the way through the commit rpc or just
840 * waiting for the initial write. The default is to only
841 * wait through the initial write so the data is in the
842 * server's cache, which is roughly similar to the state
843 * a standard disk subsystem leaves the file in on close().
845 * We cannot clear the NMODIFIED bit in np->n_flag due to
846 * potential races with other processes, and certainly
847 * cannot clear it if we don't commit.
848 * These races occur when there is no longer the old
849 * traditional vnode locking implemented for Vnode Ops.
851 int cm = newnfs_commit_on_close ? 1 : 0;
852 error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
853 /* np->n_flag &= ~NMODIFIED; */
854 } else if (NFS_ISV4(vp)) {
855 if (nfscl_mustflush(vp) != 0) {
856 int cm = newnfs_commit_on_close ? 1 : 0;
857 error = ncl_flush(vp, MNT_WAIT, ap->a_td,
860 * as above w.r.t races when clearing
862 * np->n_flag &= ~NMODIFIED;
866 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
871 * Invalidate the attribute cache in all cases.
872 * An open is going to fetch fresh attrs any way, other procs
873 * on this node that have file open will be forced to do an
874 * otw attr fetch, but this is safe.
875 * --> A user found that their RPC count dropped by 20% when
876 * this was commented out and I can't see any requirement
877 * for it, so I've disabled it when negative lookups are
878 * enabled. (What does this have to do with negative lookup
879 * caching? Well nothing, except it was reported by the
880 * same user that needed negative lookup caching and I wanted
881 * there to be a way to disable it to see if it
882 * is the cause of some caching/coherency issue that might
885 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
887 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
889 if (np->n_flag & NWRITEERR) {
890 np->n_flag &= ~NWRITEERR;
898 * Get attributes so "change" is up to date.
900 if (error == 0 && nfscl_mustflush(vp) != 0 &&
901 vp->v_type == VREG &&
902 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
903 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
906 np->n_change = nfsva.na_filerev;
907 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
915 ret = nfsrpc_close(vp, 0, ap->a_td);
919 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
922 if (newnfs_directio_enable)
923 KASSERT((np->n_directio_asyncwr == 0),
924 ("nfs_close: dirty unflushed (%d) directio buffers\n",
925 np->n_directio_asyncwr));
926 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
928 KASSERT((np->n_directio_opens > 0),
929 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
930 np->n_directio_opens--;
931 if (np->n_directio_opens == 0)
932 np->n_flag &= ~NNONCACHE;
941 * nfs getattr call from vfs.
944 nfs_getattr(struct vop_getattr_args *ap)
946 struct vnode *vp = ap->a_vp;
947 struct thread *td = curthread; /* XXX */
948 struct nfsnode *np = VTONFS(vp);
950 struct nfsvattr nfsva;
951 struct vattr *vap = ap->a_vap;
955 * Update local times for special files.
958 if (np->n_flag & (NACC | NUPD))
962 * First look in the cache.
964 if (ncl_getattrcache(vp, &vattr) == 0) {
965 ncl_copy_vattr(vap, &vattr);
968 * Get the local modify time for the case of a write
971 nfscl_deleggetmodtime(vp, &vap->va_mtime);
975 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
976 nfsaccess_cache_timeout > 0) {
977 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
978 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
979 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
980 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
984 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
986 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
989 * Get the local modify time for the case of a write
992 nfscl_deleggetmodtime(vp, &vap->va_mtime);
993 } else if (NFS_ISV4(vp)) {
994 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1003 nfs_setattr(struct vop_setattr_args *ap)
1005 struct vnode *vp = ap->a_vp;
1006 struct nfsnode *np = VTONFS(vp);
1007 struct thread *td = curthread; /* XXX */
1008 struct vattr *vap = ap->a_vap;
1013 tsize = (u_quad_t)0;
1017 * Setting of flags and marking of atimes are not supported.
1019 if (vap->va_flags != VNOVAL)
1020 return (EOPNOTSUPP);
1023 * Disallow write attempts if the filesystem is mounted read-only.
1025 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
1026 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
1027 vap->va_mtime.tv_sec != VNOVAL ||
1028 vap->va_birthtime.tv_sec != VNOVAL ||
1029 vap->va_mode != (mode_t)VNOVAL) &&
1030 (vp->v_mount->mnt_flag & MNT_RDONLY))
1032 if (vap->va_size != VNOVAL) {
1033 switch (vp->v_type) {
1040 if (vap->va_mtime.tv_sec == VNOVAL &&
1041 vap->va_atime.tv_sec == VNOVAL &&
1042 vap->va_birthtime.tv_sec == VNOVAL &&
1043 vap->va_mode == (mode_t)VNOVAL &&
1044 vap->va_uid == (uid_t)VNOVAL &&
1045 vap->va_gid == (gid_t)VNOVAL)
1047 vap->va_size = VNOVAL;
1051 * Disallow write attempts if the filesystem is
1052 * mounted read-only.
1054 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1057 * We run vnode_pager_setsize() early (why?),
1058 * we must set np->n_size now to avoid vinvalbuf
1059 * V_SAVE races that might setsize a lower
1065 error = ncl_meta_setsize(vp, td, vap->va_size);
1067 if (np->n_flag & NMODIFIED) {
1070 error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
1073 vnode_pager_setsize(vp, tsize);
1077 * Call nfscl_delegmodtime() to set the modify time
1078 * locally, as required.
1080 nfscl_delegmodtime(vp);
1084 * np->n_size has already been set to vap->va_size
1085 * in ncl_meta_setsize(). We must set it again since
1086 * nfs_loadattrcache() could be called through
1087 * ncl_meta_setsize() and could modify np->n_size.
1090 np->n_vattr.na_size = np->n_size = vap->va_size;
1095 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
1096 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
1098 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
1099 if (error == EINTR || error == EIO)
1104 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
1105 if (error && vap->va_size != VNOVAL) {
1107 np->n_size = np->n_vattr.na_size = tsize;
1108 vnode_pager_setsize(vp, tsize);
1115 * Do an nfs setattr rpc.
1118 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1121 struct nfsnode *np = VTONFS(vp);
1122 int error, ret, attrflag, i;
1123 struct nfsvattr nfsva;
1125 if (NFS_ISV34(vp)) {
1127 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1128 np->n_accesscache[i].stamp = 0;
1129 np->n_flag |= NDELEGMOD;
1131 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1133 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1136 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1140 if (error && NFS_ISV4(vp))
1141 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1146 * nfs lookup call, one step at a time...
1147 * First look in cache
1148 * If not found, unlock the directory nfsnode and do the rpc
1151 nfs_lookup(struct vop_lookup_args *ap)
1153 struct componentname *cnp = ap->a_cnp;
1154 struct vnode *dvp = ap->a_dvp;
1155 struct vnode **vpp = ap->a_vpp;
1156 struct mount *mp = dvp->v_mount;
1157 int flags = cnp->cn_flags;
1158 struct vnode *newvp;
1159 struct nfsmount *nmp;
1160 struct nfsnode *np, *newnp;
1161 int error = 0, attrflag, dattrflag, ltype, ncticks;
1162 struct thread *td = cnp->cn_thread;
1164 struct nfsvattr dnfsva, nfsva;
1166 struct timespec nctime;
1169 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1170 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1172 if (dvp->v_type != VDIR)
1177 /* For NFSv4, wait until any remove is done. */
1179 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1180 np->n_flag |= NREMOVEWANT;
1181 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1185 error = vn_dir_check_exec(dvp, cnp);
1188 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1189 if (error > 0 && error != ENOENT)
1193 * Lookups of "." are special and always return the
1194 * current directory. cache_lookup() already handles
1195 * associated locking bookkeeping, etc.
1197 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1198 /* XXX: Is this really correct? */
1199 if (cnp->cn_nameiop != LOOKUP &&
1201 cnp->cn_flags |= SAVENAME;
1206 * We only accept a positive hit in the cache if the
1207 * change time of the file matches our cached copy.
1208 * Otherwise, we discard the cache entry and fallback
1209 * to doing a lookup RPC. We also only trust cache
1210 * entries for less than nm_nametimeo seconds.
1212 * To better handle stale file handles and attributes,
1213 * clear the attribute cache of this node if it is a
1214 * leaf component, part of an open() call, and not
1215 * locally modified before fetching the attributes.
1216 * This should allow stale file handles to be detected
1217 * here where we can fall back to a LOOKUP RPC to
1218 * recover rather than having nfs_open() detect the
1219 * stale file handle and failing open(2) with ESTALE.
1222 newnp = VTONFS(newvp);
1223 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1224 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1225 !(newnp->n_flag & NMODIFIED)) {
1227 newnp->n_attrstamp = 0;
1228 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1229 NFSUNLOCKNODE(newnp);
1231 if (nfscl_nodeleg(newvp, 0) == 0 ||
1232 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1233 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1234 timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1235 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1236 if (cnp->cn_nameiop != LOOKUP &&
1238 cnp->cn_flags |= SAVENAME;
1247 } else if (error == ENOENT) {
1248 if (VN_IS_DOOMED(dvp))
1251 * We only accept a negative hit in the cache if the
1252 * modification time of the parent directory matches
1253 * the cached copy in the name cache entry.
1254 * Otherwise, we discard all of the negative cache
1255 * entries for this directory. We also only trust
1256 * negative cache entries for up to nm_negnametimeo
1259 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1260 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1261 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1262 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1265 cache_purge_negative(dvp);
1269 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1270 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1271 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1274 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1276 if (newvp != NULLVP) {
1281 if (error != ENOENT) {
1283 error = nfscl_maperr(td, error, (uid_t)0,
1288 /* The requested file was not found. */
1289 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1290 (flags & ISLASTCN)) {
1292 * XXX: UFS does a full VOP_ACCESS(dvp,
1293 * VWRITE) here instead of just checking
1296 if (mp->mnt_flag & MNT_RDONLY)
1298 cnp->cn_flags |= SAVENAME;
1299 return (EJUSTRETURN);
1302 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1304 * Cache the modification time of the parent
1305 * directory from the post-op attributes in
1306 * the name cache entry. The negative cache
1307 * entry will be ignored once the directory
1308 * has changed. Don't bother adding the entry
1309 * if the directory has already changed.
1312 if (timespeccmp(&np->n_vattr.na_mtime,
1313 &dnfsva.na_mtime, ==)) {
1315 cache_enter_time(dvp, NULL, cnp,
1316 &dnfsva.na_mtime, NULL);
1324 * Handle RENAME case...
1326 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1327 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1328 free(nfhp, M_NFSFH);
1331 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1337 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1340 cnp->cn_flags |= SAVENAME;
1344 if (flags & ISDOTDOT) {
1345 ltype = NFSVOPISLOCKED(dvp);
1346 error = vfs_busy(mp, MBF_NOWAIT);
1350 error = vfs_busy(mp, 0);
1351 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1353 if (error == 0 && VN_IS_DOOMED(dvp)) {
1361 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1367 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1368 if (VN_IS_DOOMED(dvp)) {
1380 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1382 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1383 free(nfhp, M_NFSFH);
1387 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1390 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1396 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1398 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1399 !(np->n_flag & NMODIFIED)) {
1401 * Flush the attribute cache when opening a
1402 * leaf node to ensure that fresh attributes
1403 * are fetched in nfs_open() since we did not
1404 * fetch attributes from the LOOKUP reply.
1407 np->n_attrstamp = 0;
1408 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1412 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1413 cnp->cn_flags |= SAVENAME;
1414 if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp &&
1415 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1416 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1417 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1418 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1425 * Just call ncl_bioread() to do the work.
1428 nfs_read(struct vop_read_args *ap)
1430 struct vnode *vp = ap->a_vp;
1432 switch (vp->v_type) {
1434 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1438 return (EOPNOTSUPP);
1446 nfs_readlink(struct vop_readlink_args *ap)
1448 struct vnode *vp = ap->a_vp;
1450 if (vp->v_type != VLNK)
1452 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1456 * Do a readlink rpc.
1457 * Called by ncl_doio() from below the buffer cache.
1460 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1462 int error, ret, attrflag;
1463 struct nfsvattr nfsva;
1465 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1468 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1472 if (error && NFS_ISV4(vp))
1473 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1482 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1484 int error, ret, attrflag;
1485 struct nfsvattr nfsva;
1486 struct nfsmount *nmp;
1488 nmp = VFSTONFS(vp->v_mount);
1491 if (NFSHASPNFS(nmp))
1492 error = nfscl_doiods(vp, uiop, NULL, NULL,
1493 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1494 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1496 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1499 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1503 if (error && NFS_ISV4(vp))
1504 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1512 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1513 int *iomode, int *must_commit, int called_from_strategy)
1515 struct nfsvattr nfsva;
1516 int error, attrflag, ret;
1517 struct nfsmount *nmp;
1519 nmp = VFSTONFS(vp->v_mount);
1522 if (NFSHASPNFS(nmp))
1523 error = nfscl_doiods(vp, uiop, iomode, must_commit,
1524 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1525 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1527 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1528 uiop->uio_td, &nfsva, &attrflag, NULL,
1529 called_from_strategy);
1531 if (VTONFS(vp)->n_flag & ND_NFSV4)
1532 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1535 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1541 *iomode = NFSWRITE_FILESYNC;
1542 if (error && NFS_ISV4(vp))
1543 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1549 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1550 * mode set to specify the file type and the size field for rdev.
1553 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1556 struct nfsvattr nfsva, dnfsva;
1557 struct vnode *newvp = NULL;
1558 struct nfsnode *np = NULL, *dnp;
1561 int error = 0, attrflag, dattrflag;
1564 if (vap->va_type == VCHR || vap->va_type == VBLK)
1565 rdev = vap->va_rdev;
1566 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1569 return (EOPNOTSUPP);
1570 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1572 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1573 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1574 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1577 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1578 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1579 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1582 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1583 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1586 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1589 if (attrflag != 0) {
1590 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1598 } else if (NFS_ISV4(dvp)) {
1599 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1604 dnp->n_flag |= NMODIFIED;
1606 dnp->n_attrstamp = 0;
1607 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1615 * just call nfs_mknodrpc() to do the work.
1619 nfs_mknod(struct vop_mknod_args *ap)
1621 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1624 static struct mtx nfs_cverf_mtx;
1625 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1631 static nfsquad_t cverf;
1633 static int cverf_initialized = 0;
1635 mtx_lock(&nfs_cverf_mtx);
1636 if (cverf_initialized == 0) {
1637 cverf.lval[0] = arc4random();
1638 cverf.lval[1] = arc4random();
1639 cverf_initialized = 1;
1643 mtx_unlock(&nfs_cverf_mtx);
1649 * nfs file create call
1652 nfs_create(struct vop_create_args *ap)
1654 struct vnode *dvp = ap->a_dvp;
1655 struct vattr *vap = ap->a_vap;
1656 struct componentname *cnp = ap->a_cnp;
1657 struct nfsnode *np = NULL, *dnp;
1658 struct vnode *newvp = NULL;
1659 struct nfsmount *nmp;
1660 struct nfsvattr dnfsva, nfsva;
1663 int error = 0, attrflag, dattrflag, fmode = 0;
1667 * Oops, not for me..
1669 if (vap->va_type == VSOCK)
1670 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1672 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1674 if (vap->va_vaflags & VA_EXCLUSIVE)
1677 nmp = VFSTONFS(dvp->v_mount);
1679 /* For NFSv4, wait until any remove is done. */
1681 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1682 dnp->n_flag |= NREMOVEWANT;
1683 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1687 cverf = nfs_get_cverf();
1688 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1689 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1690 &nfhp, &attrflag, &dattrflag, NULL);
1693 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1694 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1695 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1698 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1699 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1702 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1706 error = nfsrpc_getattr(newvp, cnp->cn_cred,
1707 cnp->cn_thread, &nfsva, NULL);
1709 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1713 if (newvp != NULL) {
1717 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1718 error == NFSERR_NOTSUPP) {
1722 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1723 if (nfscl_checksattr(vap, &nfsva)) {
1724 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1725 cnp->cn_thread, &nfsva, &attrflag, NULL);
1726 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1727 vap->va_gid != (gid_t)VNOVAL)) {
1728 /* try again without setting uid/gid */
1729 vap->va_uid = (uid_t)VNOVAL;
1730 vap->va_gid = (uid_t)VNOVAL;
1731 error = nfsrpc_setattr(newvp, vap, NULL,
1732 cnp->cn_cred, cnp->cn_thread, &nfsva,
1736 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1743 if ((cnp->cn_flags & MAKEENTRY) && attrflag) {
1745 cache_enter_time(dvp, newvp, cnp,
1746 &nfsva.na_ctime, NULL);
1748 printf("nfs_create: bogus NFS server returned "
1749 "the directory as the new file object\n");
1752 } else if (NFS_ISV4(dvp)) {
1753 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1757 dnp->n_flag |= NMODIFIED;
1759 dnp->n_attrstamp = 0;
1760 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1767 * nfs file remove call
1768 * To try and make nfs semantics closer to ufs semantics, a file that has
1769 * other processes using the vnode is renamed instead of removed and then
1770 * removed later on the last close.
1771 * - If v_usecount > 1
1772 * If a rename is not already in the works
1773 * call nfs_sillyrename() to set it up
1778 nfs_remove(struct vop_remove_args *ap)
1780 struct vnode *vp = ap->a_vp;
1781 struct vnode *dvp = ap->a_dvp;
1782 struct componentname *cnp = ap->a_cnp;
1783 struct nfsnode *np = VTONFS(vp);
1787 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1788 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1789 if (vp->v_type == VDIR)
1791 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1792 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1793 vattr.va_nlink > 1)) {
1795 * Purge the name cache so that the chance of a lookup for
1796 * the name succeeding while the remove is in progress is
1797 * minimized. Without node locking it can still happen, such
1798 * that an I/O op returns ESTALE, but since you get this if
1799 * another host removes the file..
1803 * throw away biocache buffers, mainly to avoid
1804 * unnecessary delayed writes later.
1806 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1807 if (error != EINTR && error != EIO)
1809 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1810 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1812 * Kludge City: If the first reply to the remove rpc is lost..
1813 * the reply to the retransmitted request will be ENOENT
1814 * since the file was in fact removed
1815 * Therefore, we cheat and return success.
1817 if (error == ENOENT)
1819 } else if (!np->n_sillyrename)
1820 error = nfs_sillyrename(dvp, vp, cnp);
1822 np->n_attrstamp = 0;
1824 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1829 * nfs file remove rpc called from nfs_inactive
1832 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1835 * Make sure that the directory vnode is still valid.
1836 * XXX we should lock sp->s_dvp here.
1838 if (sp->s_dvp->v_type == VBAD)
1840 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1845 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1848 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1849 int namelen, struct ucred *cred, struct thread *td)
1851 struct nfsvattr dnfsva;
1852 struct nfsnode *dnp = VTONFS(dvp);
1853 int error = 0, dattrflag;
1856 dnp->n_flag |= NREMOVEINPROG;
1858 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1861 if ((dnp->n_flag & NREMOVEWANT)) {
1862 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1864 wakeup((caddr_t)dnp);
1866 dnp->n_flag &= ~NREMOVEINPROG;
1870 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1872 dnp->n_flag |= NMODIFIED;
1874 dnp->n_attrstamp = 0;
1875 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1878 if (error && NFS_ISV4(dvp))
1879 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1884 * nfs file rename call
1887 nfs_rename(struct vop_rename_args *ap)
1889 struct vnode *fvp = ap->a_fvp;
1890 struct vnode *tvp = ap->a_tvp;
1891 struct vnode *fdvp = ap->a_fdvp;
1892 struct vnode *tdvp = ap->a_tdvp;
1893 struct componentname *tcnp = ap->a_tcnp;
1894 struct componentname *fcnp = ap->a_fcnp;
1895 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1896 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1897 struct nfsv4node *newv4 = NULL;
1900 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1901 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1902 /* Check for cross-device rename */
1903 if ((fvp->v_mount != tdvp->v_mount) ||
1904 (tvp && (fvp->v_mount != tvp->v_mount))) {
1910 printf("nfs_rename: fvp == tvp (can't happen)\n");
1914 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1918 * We have to flush B_DELWRI data prior to renaming
1919 * the file. If we don't, the delayed-write buffers
1920 * can be flushed out later after the file has gone stale
1921 * under NFSV3. NFSV2 does not have this problem because
1922 * ( as far as I can tell ) it flushes dirty buffers more
1925 * Skip the rename operation if the fsync fails, this can happen
1926 * due to the server's volume being full, when we pushed out data
1927 * that was written back to our cache earlier. Not checking for
1928 * this condition can result in potential (silent) data loss.
1930 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1933 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1938 * If the tvp exists and is in use, sillyrename it before doing the
1939 * rename of the new file over it.
1940 * XXX Can't sillyrename a directory.
1942 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1943 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1948 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1949 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1952 if (error == 0 && NFS_ISV4(tdvp)) {
1954 * For NFSv4, check to see if it is the same name and
1955 * replace the name, if it is different.
1958 sizeof (struct nfsv4node) +
1959 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1960 M_NFSV4NODE, M_WAITOK);
1963 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1964 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1965 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1966 tcnp->cn_namelen) ||
1967 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1968 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1969 tdnp->n_fhp->nfh_len))) {
1971 { char nnn[100]; int nnnl;
1972 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1973 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1975 printf("ren replace=%s\n",nnn);
1978 free(fnp->n_v4, M_NFSV4NODE);
1981 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1982 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1983 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1984 tdnp->n_fhp->nfh_len);
1985 NFSBCOPY(tcnp->cn_nameptr,
1986 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1988 NFSUNLOCKNODE(tdnp);
1991 free(newv4, M_NFSV4NODE);
1994 if (fvp->v_type == VDIR) {
1995 if (tvp != NULL && tvp->v_type == VDIR)
2010 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2012 if (error == ENOENT)
2018 * nfs file rename rpc called from nfs_remove() above
2021 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
2022 struct sillyrename *sp)
2025 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
2026 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
2031 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2034 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
2035 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
2036 int tnamelen, struct ucred *cred, struct thread *td)
2038 struct nfsvattr fnfsva, tnfsva;
2039 struct nfsnode *fdnp = VTONFS(fdvp);
2040 struct nfsnode *tdnp = VTONFS(tdvp);
2041 int error = 0, fattrflag, tattrflag;
2043 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
2044 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
2045 &tattrflag, NULL, NULL);
2047 fdnp->n_flag |= NMODIFIED;
2048 if (fattrflag != 0) {
2049 NFSUNLOCKNODE(fdnp);
2050 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
2052 fdnp->n_attrstamp = 0;
2053 NFSUNLOCKNODE(fdnp);
2054 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
2057 tdnp->n_flag |= NMODIFIED;
2058 if (tattrflag != 0) {
2059 NFSUNLOCKNODE(tdnp);
2060 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
2062 tdnp->n_attrstamp = 0;
2063 NFSUNLOCKNODE(tdnp);
2064 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2066 if (error && NFS_ISV4(fdvp))
2067 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2072 * nfs hard link create call
2075 nfs_link(struct vop_link_args *ap)
2077 struct vnode *vp = ap->a_vp;
2078 struct vnode *tdvp = ap->a_tdvp;
2079 struct componentname *cnp = ap->a_cnp;
2080 struct nfsnode *np, *tdnp;
2081 struct nfsvattr nfsva, dnfsva;
2082 int error = 0, attrflag, dattrflag;
2085 * Push all writes to the server, so that the attribute cache
2086 * doesn't get "out of sync" with the server.
2087 * XXX There should be a better way!
2089 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
2091 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2092 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
2094 tdnp = VTONFS(tdvp);
2096 tdnp->n_flag |= NMODIFIED;
2097 if (dattrflag != 0) {
2098 NFSUNLOCKNODE(tdnp);
2099 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
2101 tdnp->n_attrstamp = 0;
2102 NFSUNLOCKNODE(tdnp);
2103 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2106 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2110 np->n_attrstamp = 0;
2112 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2115 * If negative lookup caching is enabled, I might as well
2116 * add an entry for this node. Not necessary for correctness,
2117 * but if negative caching is enabled, then the system
2118 * must care about lookup caching hit rate, so...
2120 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2121 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2123 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2125 printf("nfs_link: bogus NFS server returned "
2126 "the directory as the new link\n");
2128 if (error && NFS_ISV4(vp))
2129 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2135 * nfs symbolic link create call
2138 nfs_symlink(struct vop_symlink_args *ap)
2140 struct vnode *dvp = ap->a_dvp;
2141 struct vattr *vap = ap->a_vap;
2142 struct componentname *cnp = ap->a_cnp;
2143 struct nfsvattr nfsva, dnfsva;
2145 struct nfsnode *np = NULL, *dnp;
2146 struct vnode *newvp = NULL;
2147 int error = 0, attrflag, dattrflag, ret;
2149 vap->va_type = VLNK;
2150 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2151 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2152 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2154 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2155 &np, NULL, LK_EXCLUSIVE);
2161 if (newvp != NULL) {
2163 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2165 } else if (!error) {
2167 * If we do not have an error and we could not extract the
2168 * newvp from the response due to the request being NFSv2, we
2169 * have to do a lookup in order to obtain a newvp to return.
2171 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2172 cnp->cn_cred, cnp->cn_thread, &np);
2180 error = nfscl_maperr(cnp->cn_thread, error,
2181 vap->va_uid, vap->va_gid);
2188 dnp->n_flag |= NMODIFIED;
2189 if (dattrflag != 0) {
2191 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2193 dnp->n_attrstamp = 0;
2195 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2198 * If negative lookup caching is enabled, I might as well
2199 * add an entry for this node. Not necessary for correctness,
2200 * but if negative caching is enabled, then the system
2201 * must care about lookup caching hit rate, so...
2203 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2204 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2206 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2209 printf("nfs_symlink: bogus NFS server returned "
2210 "the directory as the new file object\n");
2219 nfs_mkdir(struct vop_mkdir_args *ap)
2221 struct vnode *dvp = ap->a_dvp;
2222 struct vattr *vap = ap->a_vap;
2223 struct componentname *cnp = ap->a_cnp;
2224 struct nfsnode *np = NULL, *dnp;
2225 struct vnode *newvp = NULL;
2228 struct nfsvattr nfsva, dnfsva;
2229 int error = 0, attrflag, dattrflag, ret;
2231 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2233 vap->va_type = VDIR;
2234 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2235 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2236 &attrflag, &dattrflag, NULL);
2239 dnp->n_flag |= NMODIFIED;
2240 if (dattrflag != 0) {
2242 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2244 dnp->n_attrstamp = 0;
2246 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2249 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2250 &np, NULL, LK_EXCLUSIVE);
2254 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2259 if (!error && newvp == NULL) {
2260 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2261 cnp->cn_cred, cnp->cn_thread, &np);
2264 if (newvp->v_type != VDIR)
2272 error = nfscl_maperr(cnp->cn_thread, error,
2273 vap->va_uid, vap->va_gid);
2276 * If negative lookup caching is enabled, I might as well
2277 * add an entry for this node. Not necessary for correctness,
2278 * but if negative caching is enabled, then the system
2279 * must care about lookup caching hit rate, so...
2281 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2282 (cnp->cn_flags & MAKEENTRY) &&
2283 attrflag != 0 && dattrflag != 0) {
2285 cache_enter_time(dvp, newvp, cnp,
2286 &nfsva.na_ctime, &dnfsva.na_ctime);
2288 printf("nfs_mkdir: bogus NFS server returned "
2289 "the directory that the directory was "
2290 "created in as the new file object\n");
2298 * nfs remove directory call
2301 nfs_rmdir(struct vop_rmdir_args *ap)
2303 struct vnode *vp = ap->a_vp;
2304 struct vnode *dvp = ap->a_dvp;
2305 struct componentname *cnp = ap->a_cnp;
2306 struct nfsnode *dnp;
2307 struct nfsvattr dnfsva;
2308 int error, dattrflag;
2312 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2313 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2316 dnp->n_flag |= NMODIFIED;
2317 if (dattrflag != 0) {
2319 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2321 dnp->n_attrstamp = 0;
2323 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2328 if (error && NFS_ISV4(dvp))
2329 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2332 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2334 if (error == ENOENT)
2343 nfs_readdir(struct vop_readdir_args *ap)
2345 struct vnode *vp = ap->a_vp;
2346 struct nfsnode *np = VTONFS(vp);
2347 struct uio *uio = ap->a_uio;
2348 ssize_t tresid, left;
2352 if (ap->a_eofflag != NULL)
2354 if (vp->v_type != VDIR)
2358 * First, check for hit on the EOF offset cache
2361 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2362 (np->n_flag & NMODIFIED) == 0) {
2364 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2366 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2367 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2369 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2370 if (ap->a_eofflag != NULL)
2380 * NFS always guarantees that directory entries don't straddle
2381 * DIRBLKSIZ boundaries. As such, we need to limit the size
2382 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2385 left = uio->uio_resid % DIRBLKSIZ;
2386 if (left == uio->uio_resid)
2388 uio->uio_resid -= left;
2391 * Call ncl_bioread() to do the real work.
2393 tresid = uio->uio_resid;
2394 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2396 if (!error && uio->uio_resid == tresid) {
2397 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2398 if (ap->a_eofflag != NULL)
2402 /* Add the partial DIRBLKSIZ (left) back in. */
2403 uio->uio_resid += left;
2409 * Called from below the buffer cache by ncl_doio().
2412 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2415 struct nfsvattr nfsva;
2416 nfsuint64 *cookiep, cookie;
2417 struct nfsnode *dnp = VTONFS(vp);
2418 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2419 int error = 0, eof, attrflag;
2421 KASSERT(uiop->uio_iovcnt == 1 &&
2422 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2423 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2424 ("nfs readdirrpc bad uio"));
2427 * If there is no cookie, assume directory was stale.
2429 ncl_dircookie_lock(dnp);
2431 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2434 ncl_dircookie_unlock(dnp);
2436 ncl_dircookie_unlock(dnp);
2437 return (NFSERR_BAD_COOKIE);
2440 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2441 (void)ncl_fsinfo(nmp, vp, cred, td);
2443 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2444 &attrflag, &eof, NULL);
2446 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2450 * We are now either at the end of the directory or have filled
2455 dnp->n_direofoffset = uiop->uio_offset;
2458 if (uiop->uio_resid > 0)
2459 printf("EEK! readdirrpc resid > 0\n");
2460 ncl_dircookie_lock(dnp);
2462 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2464 ncl_dircookie_unlock(dnp);
2466 } else if (NFS_ISV4(vp)) {
2467 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2473 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2476 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2479 struct nfsvattr nfsva;
2480 nfsuint64 *cookiep, cookie;
2481 struct nfsnode *dnp = VTONFS(vp);
2482 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2483 int error = 0, attrflag, eof;
2485 KASSERT(uiop->uio_iovcnt == 1 &&
2486 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2487 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2488 ("nfs readdirplusrpc bad uio"));
2491 * If there is no cookie, assume directory was stale.
2493 ncl_dircookie_lock(dnp);
2495 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2498 ncl_dircookie_unlock(dnp);
2500 ncl_dircookie_unlock(dnp);
2501 return (NFSERR_BAD_COOKIE);
2504 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2505 (void)ncl_fsinfo(nmp, vp, cred, td);
2506 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2507 &attrflag, &eof, NULL);
2509 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2513 * We are now either at end of the directory or have filled the
2518 dnp->n_direofoffset = uiop->uio_offset;
2521 if (uiop->uio_resid > 0)
2522 printf("EEK! readdirplusrpc resid > 0\n");
2523 ncl_dircookie_lock(dnp);
2525 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2527 ncl_dircookie_unlock(dnp);
2529 } else if (NFS_ISV4(vp)) {
2530 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2536 * Silly rename. To make the NFS filesystem that is stateless look a little
2537 * more like the "ufs" a remove of an active vnode is translated to a rename
2538 * to a funny looking filename that is removed by nfs_inactive on the
2539 * nfsnode. There is the potential for another process on a different client
2540 * to create the same funny name between the nfs_lookitup() fails and the
2541 * nfs_rename() completes, but...
2544 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2546 struct sillyrename *sp;
2550 unsigned int lticks;
2554 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2555 sp = malloc(sizeof (struct sillyrename),
2556 M_NEWNFSREQ, M_WAITOK);
2557 sp->s_cred = crhold(cnp->cn_cred);
2562 * Fudge together a funny name.
2563 * Changing the format of the funny name to accommodate more
2564 * sillynames per directory.
2565 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2566 * CPU ticks since boot.
2568 pid = cnp->cn_thread->td_proc->p_pid;
2569 lticks = (unsigned int)ticks;
2571 sp->s_namlen = sprintf(sp->s_name,
2572 ".nfs.%08x.%04x4.4", lticks,
2574 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2575 cnp->cn_thread, NULL))
2579 error = nfs_renameit(dvp, vp, cnp, sp);
2582 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2583 cnp->cn_thread, &np);
2584 np->n_sillyrename = sp;
2589 free(sp, M_NEWNFSREQ);
2594 * Look up a file name and optionally either update the file handle or
2595 * allocate an nfsnode, depending on the value of npp.
2596 * npp == NULL --> just do the lookup
2597 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2599 * *npp != NULL --> update the file handle in the vnode
2602 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2603 struct thread *td, struct nfsnode **npp)
2605 struct vnode *newvp = NULL, *vp;
2606 struct nfsnode *np, *dnp = VTONFS(dvp);
2607 struct nfsfh *nfhp, *onfhp;
2608 struct nfsvattr nfsva, dnfsva;
2609 struct componentname cn;
2610 int error = 0, attrflag, dattrflag;
2613 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2614 &nfhp, &attrflag, &dattrflag, NULL);
2616 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2617 if (npp && !error) {
2622 * For NFSv4, check to see if it is the same name and
2623 * replace the name, if it is different.
2625 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2626 (np->n_v4->n4_namelen != len ||
2627 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2628 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2629 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2630 dnp->n_fhp->nfh_len))) {
2632 { char nnn[100]; int nnnl;
2633 nnnl = (len < 100) ? len : 99;
2634 bcopy(name, nnn, nnnl);
2636 printf("replace=%s\n",nnn);
2639 free(np->n_v4, M_NFSV4NODE);
2641 sizeof (struct nfsv4node) +
2642 dnp->n_fhp->nfh_len + len - 1,
2643 M_NFSV4NODE, M_WAITOK);
2644 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2645 np->n_v4->n4_namelen = len;
2646 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2647 dnp->n_fhp->nfh_len);
2648 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2650 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2654 * Rehash node for new file handle.
2656 vfs_hash_rehash(vp, hash);
2659 free(onfhp, M_NFSFH);
2661 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2662 free(nfhp, M_NFSFH);
2666 cn.cn_nameptr = name;
2667 cn.cn_namelen = len;
2668 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2669 &np, NULL, LK_EXCLUSIVE);
2674 if (!attrflag && *npp == NULL) {
2682 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2685 if (npp && *npp == NULL) {
2696 if (error && NFS_ISV4(dvp))
2697 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2702 * Nfs Version 3 and 4 commit rpc
2705 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2708 struct nfsvattr nfsva;
2709 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2712 int error, attrflag;
2717 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2718 uio.uio_offset = offset;
2719 uio.uio_resid = cnt;
2720 error = nfscl_doiods(vp, &uio, NULL, NULL,
2721 NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2724 np->n_flag &= ~NDSCOMMIT;
2729 mtx_lock(&nmp->nm_mtx);
2730 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2731 mtx_unlock(&nmp->nm_mtx);
2734 mtx_unlock(&nmp->nm_mtx);
2735 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2739 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2741 if (error != 0 && NFS_ISV4(vp))
2742 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2748 * For async requests when nfsiod(s) are running, queue the request by
2749 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2753 nfs_strategy(struct vop_strategy_args *ap)
2761 KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2762 KASSERT(!(bp->b_flags & B_DONE),
2763 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2765 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2766 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2768 if (bp->b_iocmd == BIO_READ)
2774 * If the op is asynchronous and an i/o daemon is waiting
2775 * queue the request, wake it up and wait for completion
2776 * otherwise just do it ourselves.
2778 if ((bp->b_flags & B_ASYNC) == 0 ||
2779 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2780 (void) ncl_doio(vp, bp, cr, curthread, 1);
2785 * fsync vnode op. Just call ncl_flush() with commit == 1.
2789 nfs_fsync(struct vop_fsync_args *ap)
2792 if (ap->a_vp->v_type != VREG) {
2794 * For NFS, metadata is changed synchronously on the server,
2795 * so there is nothing to flush. Also, ncl_flush() clears
2796 * the NMODIFIED flag and that shouldn't be done here for
2801 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2805 * Flush all the blocks associated with a vnode.
2806 * Walk through the buffer pool and push any dirty pages
2807 * associated with the vnode.
2808 * If the called_from_renewthread argument is TRUE, it has been called
2809 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2810 * waiting for a buffer write to complete.
2813 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2814 int commit, int called_from_renewthread)
2816 struct nfsnode *np = VTONFS(vp);
2820 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2821 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2822 int passone = 1, trycnt = 0;
2823 u_quad_t off, endoff, toff;
2824 struct ucred* wcred = NULL;
2825 struct buf **bvec = NULL;
2827 #ifndef NFS_COMMITBVECSIZ
2828 #define NFS_COMMITBVECSIZ 20
2830 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2831 u_int bvecsize = 0, bveccount;
2833 if (called_from_renewthread != 0)
2835 if (nmp->nm_flag & NFSMNT_INT)
2841 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2842 * server, but has not been committed to stable storage on the server
2843 * yet. On the first pass, the byte range is worked out and the commit
2844 * rpc is done. On the second pass, ncl_writebp() is called to do the
2851 if (NFS_ISV34(vp) && commit) {
2852 if (bvec != NULL && bvec != bvec_on_stack)
2855 * Count up how many buffers waiting for a commit.
2859 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2860 if (!BUF_ISLOCKED(bp) &&
2861 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2862 == (B_DELWRI | B_NEEDCOMMIT))
2866 * Allocate space to remember the list of bufs to commit. It is
2867 * important to use M_NOWAIT here to avoid a race with nfs_write.
2868 * If we can't get memory (for whatever reason), we will end up
2869 * committing the buffers one-by-one in the loop below.
2871 if (bveccount > NFS_COMMITBVECSIZ) {
2873 * Release the vnode interlock to avoid a lock
2877 bvec = (struct buf **)
2878 malloc(bveccount * sizeof(struct buf *),
2882 bvec = bvec_on_stack;
2883 bvecsize = NFS_COMMITBVECSIZ;
2885 bvecsize = bveccount;
2887 bvec = bvec_on_stack;
2888 bvecsize = NFS_COMMITBVECSIZ;
2890 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2891 if (bvecpos >= bvecsize)
2893 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2894 nbp = TAILQ_NEXT(bp, b_bobufs);
2897 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2898 (B_DELWRI | B_NEEDCOMMIT)) {
2900 nbp = TAILQ_NEXT(bp, b_bobufs);
2906 * Work out if all buffers are using the same cred
2907 * so we can deal with them all with one commit.
2909 * NOTE: we are not clearing B_DONE here, so we have
2910 * to do it later on in this routine if we intend to
2911 * initiate I/O on the bp.
2913 * Note: to avoid loopback deadlocks, we do not
2914 * assign b_runningbufspace.
2917 wcred = bp->b_wcred;
2918 else if (wcred != bp->b_wcred)
2920 vfs_busy_pages(bp, 1);
2924 * bp is protected by being locked, but nbp is not
2925 * and vfs_busy_pages() may sleep. We have to
2928 nbp = TAILQ_NEXT(bp, b_bobufs);
2931 * A list of these buffers is kept so that the
2932 * second loop knows which buffers have actually
2933 * been committed. This is necessary, since there
2934 * may be a race between the commit rpc and new
2935 * uncommitted writes on the file.
2937 bvec[bvecpos++] = bp;
2938 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2942 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2950 * Commit data on the server, as required.
2951 * If all bufs are using the same wcred, then use that with
2952 * one call for all of them, otherwise commit each one
2955 if (wcred != NOCRED)
2956 retv = ncl_commit(vp, off, (int)(endoff - off),
2960 for (i = 0; i < bvecpos; i++) {
2963 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2965 size = (u_quad_t)(bp->b_dirtyend
2967 retv = ncl_commit(vp, off, (int)size,
2973 if (retv == NFSERR_STALEWRITEVERF)
2974 ncl_clearcommit(vp->v_mount);
2977 * Now, either mark the blocks I/O done or mark the
2978 * blocks dirty, depending on whether the commit
2981 for (i = 0; i < bvecpos; i++) {
2983 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2986 * Error, leave B_DELWRI intact
2988 vfs_unbusy_pages(bp);
2992 * Success, remove B_DELWRI ( bundirty() ).
2994 * b_dirtyoff/b_dirtyend seem to be NFS
2995 * specific. We should probably move that
2996 * into bundirty(). XXX
2999 bp->b_flags |= B_ASYNC;
3001 bp->b_flags &= ~B_DONE;
3002 bp->b_ioflags &= ~BIO_ERROR;
3003 bp->b_dirtyoff = bp->b_dirtyend = 0;
3010 * Start/do any write(s) that are required.
3014 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3015 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3016 if (waitfor != MNT_WAIT || passone)
3019 error = BUF_TIMELOCK(bp,
3020 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3021 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
3026 if (error == ENOLCK) {
3030 if (called_from_renewthread != 0) {
3032 * Return EIO so the flush will be retried
3038 if (newnfs_sigintr(nmp, td)) {
3042 if (slpflag == PCATCH) {
3048 if ((bp->b_flags & B_DELWRI) == 0)
3049 panic("nfs_fsync: not dirty");
3050 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3056 bp->b_flags |= B_ASYNC;
3058 if (newnfs_sigintr(nmp, td)) {
3069 if (waitfor == MNT_WAIT) {
3070 while (bo->bo_numoutput) {
3071 error = bufobj_wwait(bo, slpflag, slptimeo);
3074 if (called_from_renewthread != 0) {
3076 * Return EIO so that the flush will be
3082 error = newnfs_sigintr(nmp, td);
3085 if (slpflag == PCATCH) {
3092 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3097 * Wait for all the async IO requests to drain
3101 while (np->n_directio_asyncwr > 0) {
3102 np->n_flag |= NFSYNCWAIT;
3103 error = newnfs_msleep(td, &np->n_directio_asyncwr,
3104 &np->n_mtx, slpflag | (PRIBIO + 1),
3107 if (newnfs_sigintr(nmp, td)) {
3117 if (NFSHASPNFS(nmp)) {
3118 nfscl_layoutcommit(vp, td);
3120 * Invalidate the attribute cache, since writes to a DS
3121 * won't update the size attribute.
3124 np->n_attrstamp = 0;
3127 if (np->n_flag & NWRITEERR) {
3128 error = np->n_error;
3129 np->n_flag &= ~NWRITEERR;
3131 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3132 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3133 np->n_flag &= ~NMODIFIED;
3136 if (bvec != NULL && bvec != bvec_on_stack)
3138 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3139 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
3140 np->n_directio_asyncwr != 0)) {
3142 /* try, try again... */
3149 vn_printf(vp, "ncl_flush failed");
3150 error = called_from_renewthread != 0 ? EIO : EBUSY;
3156 * NFS advisory byte-level locks.
3159 nfs_advlock(struct vop_advlock_args *ap)
3161 struct vnode *vp = ap->a_vp;
3163 struct nfsnode *np = VTONFS(ap->a_vp);
3164 struct proc *p = (struct proc *)ap->a_id;
3165 struct thread *td = curthread; /* XXX */
3170 error = NFSVOPLOCK(vp, LK_SHARED);
3173 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3174 if (vp->v_type != VREG) {
3178 if ((ap->a_flags & F_POSIX) != 0)
3181 cred = td->td_ucred;
3182 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
3183 if (VN_IS_DOOMED(vp)) {
3189 * If this is unlocking a write locked region, flush and
3190 * commit them before unlocking. This is required by
3191 * RFC3530 Sec. 9.3.2.
3193 if (ap->a_op == F_UNLCK &&
3194 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3196 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3199 * Loop around doing the lock op, while a blocking lock
3200 * must wait for the lock op to succeed.
3203 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3204 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3205 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3206 ap->a_op == F_SETLK) {
3208 error = nfs_catnap(PZERO | PCATCH, ret,
3212 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3213 if (VN_IS_DOOMED(vp)) {
3218 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3219 ap->a_op == F_SETLK);
3220 if (ret == NFSERR_DENIED) {
3223 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3226 } else if (ret != 0) {
3232 * Now, if we just got a lock, invalidate data in the buffer
3233 * cache, as required, so that the coherency conforms with
3234 * RFC3530 Sec. 9.3.2.
3236 if (ap->a_op == F_SETLK) {
3237 if ((np->n_flag & NMODIFIED) == 0) {
3238 np->n_attrstamp = 0;
3239 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3240 ret = VOP_GETATTR(vp, &va, cred);
3242 if ((np->n_flag & NMODIFIED) || ret ||
3243 np->n_change != va.va_filerev) {
3244 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3245 np->n_attrstamp = 0;
3246 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3247 ret = VOP_GETATTR(vp, &va, cred);
3249 np->n_mtime = va.va_mtime;
3250 np->n_change = va.va_filerev;
3253 /* Mark that a file lock has been acquired. */
3255 np->n_flag |= NHASBEENLOCKED;
3258 } else if (!NFS_ISV4(vp)) {
3259 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3260 size = VTONFS(vp)->n_size;
3262 error = lf_advlock(ap, &(vp->v_lockf), size);
3264 if (nfs_advlock_p != NULL)
3265 error = nfs_advlock_p(ap);
3271 if (error == 0 && ap->a_op == F_SETLK) {
3272 error = NFSVOPLOCK(vp, LK_SHARED);
3274 /* Mark that a file lock has been acquired. */
3276 np->n_flag |= NHASBEENLOCKED;
3290 * NFS advisory byte-level locks.
3293 nfs_advlockasync(struct vop_advlockasync_args *ap)
3295 struct vnode *vp = ap->a_vp;
3300 return (EOPNOTSUPP);
3301 error = NFSVOPLOCK(vp, LK_SHARED);
3304 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3305 size = VTONFS(vp)->n_size;
3307 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3316 * Print out the contents of an nfsnode.
3319 nfs_print(struct vop_print_args *ap)
3321 struct vnode *vp = ap->a_vp;
3322 struct nfsnode *np = VTONFS(vp);
3324 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3325 (uintmax_t)np->n_vattr.na_fsid);
3326 if (vp->v_type == VFIFO)
3333 * This is the "real" nfs::bwrite(struct buf*).
3334 * We set B_CACHE if this is a VMIO buffer.
3337 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3339 int oldflags, rtval;
3341 if (bp->b_flags & B_INVAL) {
3346 oldflags = bp->b_flags;
3347 bp->b_flags |= B_CACHE;
3350 * Undirty the bp. We will redirty it later if the I/O fails.
3353 bp->b_flags &= ~B_DONE;
3354 bp->b_ioflags &= ~BIO_ERROR;
3355 bp->b_iocmd = BIO_WRITE;
3357 bufobj_wref(bp->b_bufobj);
3358 curthread->td_ru.ru_oublock++;
3361 * Note: to avoid loopback deadlocks, we do not
3362 * assign b_runningbufspace.
3364 vfs_busy_pages(bp, 1);
3367 bp->b_iooffset = dbtob(bp->b_blkno);
3370 if ((oldflags & B_ASYNC) != 0)
3373 rtval = bufwait(bp);
3374 if (oldflags & B_DELWRI)
3381 * nfs special file access vnode op.
3382 * Essentially just get vattr and then imitate iaccess() since the device is
3383 * local to the client.
3386 nfsspec_access(struct vop_access_args *ap)
3389 struct ucred *cred = ap->a_cred;
3390 struct vnode *vp = ap->a_vp;
3391 accmode_t accmode = ap->a_accmode;
3396 * Disallow write attempts on filesystems mounted read-only;
3397 * unless the file is a socket, fifo, or a block or character
3398 * device resident on the filesystem.
3400 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3401 switch (vp->v_type) {
3411 error = VOP_GETATTR(vp, vap, cred);
3414 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3421 * Read wrapper for fifos.
3424 nfsfifo_read(struct vop_read_args *ap)
3426 struct nfsnode *np = VTONFS(ap->a_vp);
3434 vfs_timestamp(&np->n_atim);
3436 error = fifo_specops.vop_read(ap);
3441 * Write wrapper for fifos.
3444 nfsfifo_write(struct vop_write_args *ap)
3446 struct nfsnode *np = VTONFS(ap->a_vp);
3453 vfs_timestamp(&np->n_mtim);
3455 return(fifo_specops.vop_write(ap));
3459 * Close wrapper for fifos.
3461 * Update the times on the nfsnode then do fifo close.
3464 nfsfifo_close(struct vop_close_args *ap)
3466 struct vnode *vp = ap->a_vp;
3467 struct nfsnode *np = VTONFS(vp);
3472 if (np->n_flag & (NACC | NUPD)) {
3474 if (np->n_flag & NACC)
3476 if (np->n_flag & NUPD)
3479 if (vrefcnt(vp) == 1 &&
3480 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3482 if (np->n_flag & NACC)
3483 vattr.va_atime = np->n_atim;
3484 if (np->n_flag & NUPD)
3485 vattr.va_mtime = np->n_mtim;
3487 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3493 return (fifo_specops.vop_close(ap));
3497 * Just call ncl_writebp() with the force argument set to 1.
3499 * NOTE: B_DONE may or may not be set in a_bp on call.
3502 nfs_bwrite(struct buf *bp)
3505 return (ncl_writebp(bp, 1, curthread));
3508 struct buf_ops buf_ops_newnfs = {
3509 .bop_name = "buf_ops_nfs",
3510 .bop_write = nfs_bwrite,
3511 .bop_strategy = bufstrategy,
3512 .bop_sync = bufsync,
3513 .bop_bdflush = bufbdflush,
3517 nfs_getacl(struct vop_getacl_args *ap)
3521 if (ap->a_type != ACL_TYPE_NFS4)
3522 return (EOPNOTSUPP);
3523 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3525 if (error > NFSERR_STALE) {
3526 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3533 nfs_setacl(struct vop_setacl_args *ap)
3537 if (ap->a_type != ACL_TYPE_NFS4)
3538 return (EOPNOTSUPP);
3539 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3541 if (error > NFSERR_STALE) {
3542 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3549 * VOP_ADVISE for NFS.
3550 * Just return 0 for any errors, since it is just a hint.
3553 nfs_advise(struct vop_advise_args *ap)
3555 struct thread *td = curthread;
3556 struct nfsmount *nmp;
3561 * First do vop_stdadvise() to handle the buffer cache.
3563 error = vop_stdadvise(ap);
3566 if (ap->a_start < 0 || ap->a_end < 0)
3568 if (ap->a_end == OFF_MAX)
3570 else if (ap->a_end < ap->a_start)
3573 len = ap->a_end - ap->a_start + 1;
3574 nmp = VFSTONFS(ap->a_vp->v_mount);
3575 mtx_lock(&nmp->nm_mtx);
3576 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3577 (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) ==
3578 0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) {
3579 mtx_unlock(&nmp->nm_mtx);
3582 mtx_unlock(&nmp->nm_mtx);
3583 error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice,
3585 if (error == NFSERR_NOTSUPP) {
3586 mtx_lock(&nmp->nm_mtx);
3587 nmp->nm_privflag |= NFSMNTP_NOADVISE;
3588 mtx_unlock(&nmp->nm_mtx);
3597 nfs_allocate(struct vop_allocate_args *ap)
3599 struct vnode *vp = ap->a_vp;
3600 struct thread *td = curthread;
3601 struct nfsvattr nfsva;
3602 struct nfsmount *nmp;
3603 int attrflag, error, ret;
3606 nmp = VFSTONFS(vp->v_mount);
3607 mtx_lock(&nmp->nm_mtx);
3608 if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3609 (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) {
3610 mtx_unlock(&nmp->nm_mtx);
3612 * Flush first to ensure that the allocate adds to the
3613 * file's allocation on the server.
3615 error = ncl_flush(vp, MNT_WAIT, td, 1, 0);
3617 error = nfsrpc_allocate(vp, *ap->a_offset, *ap->a_len,
3618 &nfsva, &attrflag, td->td_ucred, td, NULL);
3620 *ap->a_offset += *ap->a_len;
3622 } else if (error == NFSERR_NOTSUPP) {
3623 mtx_lock(&nmp->nm_mtx);
3624 nmp->nm_privflag |= NFSMNTP_NOALLOCATE;
3625 mtx_unlock(&nmp->nm_mtx);
3628 mtx_unlock(&nmp->nm_mtx);
3632 * If the NFS server cannot perform the Allocate operation, just call
3633 * vop_stdallocate() to perform it.
3636 error = vop_stdallocate(ap);
3637 if (attrflag != 0) {
3638 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3639 if (error == 0 && ret != 0)
3643 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3648 * nfs copy_file_range call
3651 nfs_copy_file_range(struct vop_copy_file_range_args *ap)
3653 struct vnode *invp = ap->a_invp;
3654 struct vnode *outvp = ap->a_outvp;
3656 struct nfsvattr innfsva, outnfsva;
3659 struct nfsmount *nmp;
3661 int error, inattrflag, outattrflag, ret, ret2;
3662 off_t inoff, outoff;
3663 bool consecutive, must_commit, tryoutcred;
3666 nmp = VFSTONFS(invp->v_mount);
3667 mtx_lock(&nmp->nm_mtx);
3668 /* NFSv4.2 Copy is not permitted for infile == outfile. */
3669 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3670 (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0 || invp == outvp) {
3671 mtx_unlock(&nmp->nm_mtx);
3672 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3673 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3674 ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3677 mtx_unlock(&nmp->nm_mtx);
3679 /* Lock both vnodes, avoiding risk of deadlock. */
3682 error = vn_start_write(outvp, &mp, V_WAIT);
3684 error = vn_lock(outvp, LK_EXCLUSIVE);
3686 error = vn_lock(invp, LK_SHARED | LK_NOWAIT);
3691 vn_finished_write(mp);
3693 error = vn_lock(invp, LK_SHARED);
3699 vn_finished_write(mp);
3700 } while (error == 0);
3705 * Do the vn_rlimit_fsize() check. Should this be above the VOP layer?
3707 io.uio_offset = *ap->a_outoffp;
3708 io.uio_resid = *ap->a_lenp;
3709 error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd);
3712 * Flush the input file so that the data is up to date before
3713 * the copy. Flush writes for the output file so that they
3714 * do not overwrite the data copied to the output file by the Copy.
3715 * Set the commit argument for both flushes so that the data is on
3716 * stable storage before the Copy RPC. This is done in case the
3717 * server reboots during the Copy and needs to be redone.
3720 error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0);
3722 error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0);
3724 /* Do the actual NFSv4.2 RPC. */
3726 mtx_lock(&nmp->nm_mtx);
3727 if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0)
3730 consecutive = false;
3731 mtx_unlock(&nmp->nm_mtx);
3732 inoff = *ap->a_inoffp;
3733 outoff = *ap->a_outoffp;
3735 must_commit = false;
3737 vap = &VTONFS(invp)->n_vattr.na_vattr;
3738 error = VOP_GETATTR(invp, vap, ap->a_incred);
3741 * Clip "len" at va_size so that RFC compliant servers
3742 * will not reply NFSERR_INVAL.
3743 * Setting "len == 0" for the RPC would be preferred,
3744 * but some Linux servers do not support that.
3746 if (inoff >= vap->va_size)
3747 *ap->a_lenp = len = 0;
3748 else if (inoff + len > vap->va_size)
3749 *ap->a_lenp = len = vap->va_size - inoff;
3755 * len will be set to 0 upon a successful Copy RPC.
3756 * As such, this only loops when the Copy RPC needs to be retried.
3758 while (len > 0 && error == 0) {
3759 inattrflag = outattrflag = 0;
3762 error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3763 outvp, ap->a_outoffp, &len2, ap->a_flags,
3764 &inattrflag, &innfsva, &outattrflag, &outnfsva,
3765 ap->a_outcred, consecutive, &must_commit);
3767 error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3768 outvp, ap->a_outoffp, &len2, ap->a_flags,
3769 &inattrflag, &innfsva, &outattrflag, &outnfsva,
3770 ap->a_incred, consecutive, &must_commit);
3771 if (inattrflag != 0)
3772 ret = nfscl_loadattrcache(&invp, &innfsva, NULL, NULL,
3774 if (outattrflag != 0)
3775 ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL,
3778 if (consecutive == false) {
3780 mtx_lock(&nmp->nm_mtx);
3782 NFSMNTP_NOCONSECUTIVE;
3783 mtx_unlock(&nmp->nm_mtx);
3785 error = NFSERR_OFFLOADNOREQS;
3789 if (len2 > 0 && must_commit && error == 0)
3790 error = ncl_commit(outvp, outoff, *ap->a_lenp,
3791 ap->a_outcred, curthread);
3792 if (error == 0 && ret != 0)
3794 if (error == 0 && ret2 != 0)
3796 } else if (error == NFSERR_OFFLOADNOREQS && consecutive) {
3798 * Try consecutive == false, which is ok only if all
3800 * If only some bytes were copied when consecutive
3801 * is false, there is no way to know which bytes
3802 * still need to be written.
3804 consecutive = false;
3806 } else if (error == NFSERR_ACCES && tryoutcred) {
3807 /* Try again with incred. */
3811 if (error == NFSERR_STALEWRITEVERF) {
3813 * Server rebooted, so do it all again.
3815 *ap->a_inoffp = inoff;
3816 *ap->a_outoffp = outoff;
3818 must_commit = false;
3825 vn_finished_write(mp);
3826 if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS ||
3827 error == NFSERR_ACCES) {
3829 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can
3830 * use a_incred for the read and a_outcred for the write, so
3831 * try this for NFSERR_ACCES failures for the Copy.
3832 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can
3833 * never succeed, so disable it.
3835 if (error != NFSERR_ACCES) {
3836 /* Can never do Copy on this mount. */
3837 mtx_lock(&nmp->nm_mtx);
3838 nmp->nm_privflag |= NFSMNTP_NOCOPY;
3839 mtx_unlock(&nmp->nm_mtx);
3841 *ap->a_inoffp = inoff;
3842 *ap->a_outoffp = outoff;
3843 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3844 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3845 ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3846 } else if (error != 0)
3850 error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0);
3858 nfs_ioctl(struct vop_ioctl_args *ap)
3860 struct vnode *vp = ap->a_vp;
3861 struct nfsvattr nfsva;
3862 struct nfsmount *nmp;
3863 int attrflag, content, error, ret;
3864 bool eof = false; /* shut up compiler. */
3866 if (vp->v_type != VREG)
3868 nmp = VFSTONFS(vp->v_mount);
3869 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) {
3870 error = vop_stdioctl(ap);
3874 /* Do the actual NFSv4.2 RPC. */
3875 switch (ap->a_command) {
3877 content = NFSV4CONTENT_DATA;
3880 content = NFSV4CONTENT_HOLE;
3886 error = vn_lock(vp, LK_SHARED);
3890 if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size)
3894 * Flush all writes, so that the server is up to date.
3895 * Although a Commit is not required, the commit argument
3896 * is set so that, for a pNFS File/Flexible File Layout
3897 * server, the LayoutCommit will be done to ensure the file
3898 * size is up to date on the Metadata Server.
3900 error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0);
3902 error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof,
3903 content, ap->a_cred, &nfsva, &attrflag);
3904 /* If at eof for FIOSEEKDATA, return ENXIO. */
3905 if (eof && error == 0 && content == NFSV4CONTENT_DATA)
3908 if (attrflag != 0) {
3909 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3910 if (error == 0 && ret != 0)
3921 * nfs getextattr call
3924 nfs_getextattr(struct vop_getextattr_args *ap)
3926 struct vnode *vp = ap->a_vp;
3927 struct nfsmount *nmp;
3929 struct thread *td = ap->a_td;
3930 struct nfsvattr nfsva;
3932 int attrflag, error, ret;
3934 nmp = VFSTONFS(vp->v_mount);
3935 mtx_lock(&nmp->nm_mtx);
3936 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3937 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
3938 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
3939 mtx_unlock(&nmp->nm_mtx);
3940 return (EOPNOTSUPP);
3942 mtx_unlock(&nmp->nm_mtx);
3946 cred = td->td_ucred;
3947 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
3949 error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva,
3950 &attrflag, cred, td);
3951 if (attrflag != 0) {
3952 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3953 if (error == 0 && ret != 0)
3956 if (error == 0 && ap->a_size != NULL)
3960 case NFSERR_NOTSUPP:
3961 case NFSERR_OPILLEGAL:
3962 mtx_lock(&nmp->nm_mtx);
3963 nmp->nm_privflag |= NFSMNTP_NOXATTR;
3964 mtx_unlock(&nmp->nm_mtx);
3967 case NFSERR_NOXATTR:
3968 case NFSERR_XATTR2BIG:
3972 error = nfscl_maperr(td, error, 0, 0);
3979 * nfs setextattr call
3982 nfs_setextattr(struct vop_setextattr_args *ap)
3984 struct vnode *vp = ap->a_vp;
3985 struct nfsmount *nmp;
3987 struct thread *td = ap->a_td;
3988 struct nfsvattr nfsva;
3989 int attrflag, error, ret;
3991 nmp = VFSTONFS(vp->v_mount);
3992 mtx_lock(&nmp->nm_mtx);
3993 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3994 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
3995 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
3996 mtx_unlock(&nmp->nm_mtx);
3997 return (EOPNOTSUPP);
3999 mtx_unlock(&nmp->nm_mtx);
4001 if (ap->a_uio->uio_resid < 0)
4005 cred = td->td_ucred;
4006 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4008 error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva,
4009 &attrflag, cred, td);
4010 if (attrflag != 0) {
4011 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4012 if (error == 0 && ret != 0)
4017 case NFSERR_NOTSUPP:
4018 case NFSERR_OPILLEGAL:
4019 mtx_lock(&nmp->nm_mtx);
4020 nmp->nm_privflag |= NFSMNTP_NOXATTR;
4021 mtx_unlock(&nmp->nm_mtx);
4024 case NFSERR_NOXATTR:
4025 case NFSERR_XATTR2BIG:
4029 error = nfscl_maperr(td, error, 0, 0);
4036 * nfs listextattr call
4039 nfs_listextattr(struct vop_listextattr_args *ap)
4041 struct vnode *vp = ap->a_vp;
4042 struct nfsmount *nmp;
4044 struct thread *td = ap->a_td;
4045 struct nfsvattr nfsva;
4048 int attrflag, error, ret;
4051 nmp = VFSTONFS(vp->v_mount);
4052 mtx_lock(&nmp->nm_mtx);
4053 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4054 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4055 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4056 mtx_unlock(&nmp->nm_mtx);
4057 return (EOPNOTSUPP);
4059 mtx_unlock(&nmp->nm_mtx);
4063 cred = td->td_ucred;
4065 /* Loop around doing List Extended Attribute RPCs. */
4070 while (!eof && error == 0) {
4071 len = nmp->nm_rsize;
4073 error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof,
4074 &nfsva, &attrflag, cred, td);
4075 if (attrflag != 0) {
4076 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4078 if (error == 0 && ret != 0)
4083 if (len2 > SSIZE_MAX)
4087 if (error == 0 && ap->a_size != NULL)
4091 case NFSERR_NOTSUPP:
4092 case NFSERR_OPILLEGAL:
4093 mtx_lock(&nmp->nm_mtx);
4094 nmp->nm_privflag |= NFSMNTP_NOXATTR;
4095 mtx_unlock(&nmp->nm_mtx);
4098 case NFSERR_NOXATTR:
4099 case NFSERR_XATTR2BIG:
4103 error = nfscl_maperr(td, error, 0, 0);
4110 * nfs setextattr call
4113 nfs_deleteextattr(struct vop_deleteextattr_args *ap)
4115 struct vnode *vp = ap->a_vp;
4116 struct nfsmount *nmp;
4117 struct nfsvattr nfsva;
4118 int attrflag, error, ret;
4120 nmp = VFSTONFS(vp->v_mount);
4121 mtx_lock(&nmp->nm_mtx);
4122 if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4123 (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4124 ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4125 mtx_unlock(&nmp->nm_mtx);
4126 return (EOPNOTSUPP);
4128 mtx_unlock(&nmp->nm_mtx);
4130 /* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4132 error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred,
4134 if (attrflag != 0) {
4135 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4136 if (error == 0 && ret != 0)
4141 case NFSERR_NOTSUPP:
4142 case NFSERR_OPILLEGAL:
4143 mtx_lock(&nmp->nm_mtx);
4144 nmp->nm_privflag |= NFSMNTP_NOXATTR;
4145 mtx_unlock(&nmp->nm_mtx);
4148 case NFSERR_NOXATTR:
4149 case NFSERR_XATTR2BIG:
4153 error = nfscl_maperr(ap->a_td, error, 0, 0);
4160 * Return POSIX pathconf information applicable to nfs filesystems.
4163 nfs_pathconf(struct vop_pathconf_args *ap)
4165 struct nfsv3_pathconf pc;
4166 struct nfsvattr nfsva;
4167 struct vnode *vp = ap->a_vp;
4168 struct nfsmount *nmp;
4169 struct thread *td = curthread;
4172 int attrflag, error;
4174 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
4175 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
4176 ap->a_name == _PC_NO_TRUNC)) ||
4177 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
4179 * Since only the above 4 a_names are returned by the NFSv3
4180 * Pathconf RPC, there is no point in doing it for others.
4181 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
4182 * be used for _PC_NFS4_ACL as well.
4184 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
4187 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4193 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
4196 pc.pc_linkmax = NFS_LINK_MAX;
4197 pc.pc_namemax = NFS_MAXNAMLEN;
4199 pc.pc_chownrestricted = 1;
4200 pc.pc_caseinsensitive = 0;
4201 pc.pc_casepreserving = 1;
4204 switch (ap->a_name) {
4207 *ap->a_retval = pc.pc_linkmax;
4209 *ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
4213 *ap->a_retval = pc.pc_namemax;
4216 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
4217 *ap->a_retval = PIPE_BUF;
4221 case _PC_CHOWN_RESTRICTED:
4222 *ap->a_retval = pc.pc_chownrestricted;
4225 *ap->a_retval = pc.pc_notrunc;
4228 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4229 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
4234 case _PC_ACL_PATH_MAX:
4236 *ap->a_retval = ACL_MAX_ENTRIES;
4246 case _PC_ALLOC_SIZE_MIN:
4247 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
4249 case _PC_FILESIZEBITS:
4255 case _PC_REC_INCR_XFER_SIZE:
4256 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4258 case _PC_REC_MAX_XFER_SIZE:
4259 *ap->a_retval = -1; /* means ``unlimited'' */
4261 case _PC_REC_MIN_XFER_SIZE:
4262 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4264 case _PC_REC_XFER_ALIGN:
4265 *ap->a_retval = PAGE_SIZE;
4267 case _PC_SYMLINK_MAX:
4268 *ap->a_retval = NFS_MAXPATHLEN;
4270 case _PC_MIN_HOLE_SIZE:
4271 /* Only some NFSv4.2 servers support Seek for Holes. */
4273 nmp = VFSTONFS(vp->v_mount);
4274 if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) {
4276 * NFSv4.2 doesn't have an attribute for hole size,
4277 * so all we can do is see if the Seek operation is
4278 * supported and then use f_iosize as a "best guess".
4280 mtx_lock(&nmp->nm_mtx);
4281 if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) {
4282 mtx_unlock(&nmp->nm_mtx);
4285 error = nfsrpc_seek(vp, &off, &eof,
4286 NFSV4CONTENT_HOLE, td->td_ucred, &nfsva,
4289 nfscl_loadattrcache(&vp, &nfsva,
4291 mtx_lock(&nmp->nm_mtx);
4292 if (error == NFSERR_NOTSUPP)
4293 nmp->nm_privflag |= NFSMNTP_SEEKTESTED;
4295 nmp->nm_privflag |= NFSMNTP_SEEKTESTED |
4299 if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0)
4300 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4301 mtx_unlock(&nmp->nm_mtx);
4306 error = vop_stdpathconf(ap);