2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
37 * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $
38 * $DragonFly: src/sys/vfs/nfs/nfs_vnops.c,v 1.80 2008/10/18 01:13:54 dillon Exp $
43 * vnode op calls for Sun NFS version 2 and 3
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/resourcevar.h>
53 #include <sys/mount.h>
55 #include <sys/malloc.h>
57 #include <sys/namei.h>
58 #include <sys/nlookup.h>
59 #include <sys/socket.h>
60 #include <sys/vnode.h>
61 #include <sys/dirent.h>
62 #include <sys/fcntl.h>
63 #include <sys/lockf.h>
65 #include <sys/sysctl.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vm_zone.h>
74 #include <vfs/fifofs/fifo.h>
75 #include <vfs/ufs/dir.h>
85 #include "nfsm_subs.h"
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
91 #include <sys/thread2.h>
97 static int nfsspec_read (struct vop_read_args *);
98 static int nfsspec_write (struct vop_write_args *);
99 static int nfsfifo_read (struct vop_read_args *);
100 static int nfsfifo_write (struct vop_write_args *);
101 static int nfsspec_close (struct vop_close_args *);
102 static int nfsfifo_close (struct vop_close_args *);
103 #define nfs_poll vop_nopoll
104 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
105 static int nfs_lookup (struct vop_old_lookup_args *);
106 static int nfs_create (struct vop_old_create_args *);
107 static int nfs_mknod (struct vop_old_mknod_args *);
108 static int nfs_open (struct vop_open_args *);
109 static int nfs_close (struct vop_close_args *);
110 static int nfs_access (struct vop_access_args *);
111 static int nfs_getattr (struct vop_getattr_args *);
112 static int nfs_setattr (struct vop_setattr_args *);
113 static int nfs_read (struct vop_read_args *);
114 static int nfs_mmap (struct vop_mmap_args *);
115 static int nfs_fsync (struct vop_fsync_args *);
116 static int nfs_remove (struct vop_old_remove_args *);
117 static int nfs_link (struct vop_old_link_args *);
118 static int nfs_rename (struct vop_old_rename_args *);
119 static int nfs_mkdir (struct vop_old_mkdir_args *);
120 static int nfs_rmdir (struct vop_old_rmdir_args *);
121 static int nfs_symlink (struct vop_old_symlink_args *);
122 static int nfs_readdir (struct vop_readdir_args *);
123 static int nfs_bmap (struct vop_bmap_args *);
124 static int nfs_strategy (struct vop_strategy_args *);
125 static int nfs_lookitup (struct vnode *, const char *, int,
126 struct ucred *, struct thread *, struct nfsnode **);
127 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
128 static int nfsspec_access (struct vop_access_args *);
129 static int nfs_readlink (struct vop_readlink_args *);
130 static int nfs_print (struct vop_print_args *);
131 static int nfs_advlock (struct vop_advlock_args *);
133 static int nfs_nresolve (struct vop_nresolve_args *);
135 * Global vfs data structures for nfs
137 struct vop_ops nfsv2_vnode_vops = {
138 .vop_default = vop_defaultop,
139 .vop_access = nfs_access,
140 .vop_advlock = nfs_advlock,
141 .vop_bmap = nfs_bmap,
142 .vop_close = nfs_close,
143 .vop_old_create = nfs_create,
144 .vop_fsync = nfs_fsync,
145 .vop_getattr = nfs_getattr,
146 .vop_getpages = nfs_getpages,
147 .vop_putpages = nfs_putpages,
148 .vop_inactive = nfs_inactive,
149 .vop_old_link = nfs_link,
150 .vop_old_lookup = nfs_lookup,
151 .vop_old_mkdir = nfs_mkdir,
152 .vop_old_mknod = nfs_mknod,
153 .vop_mmap = nfs_mmap,
154 .vop_open = nfs_open,
155 .vop_poll = nfs_poll,
156 .vop_print = nfs_print,
157 .vop_read = nfs_read,
158 .vop_readdir = nfs_readdir,
159 .vop_readlink = nfs_readlink,
160 .vop_reclaim = nfs_reclaim,
161 .vop_old_remove = nfs_remove,
162 .vop_old_rename = nfs_rename,
163 .vop_old_rmdir = nfs_rmdir,
164 .vop_setattr = nfs_setattr,
165 .vop_strategy = nfs_strategy,
166 .vop_old_symlink = nfs_symlink,
167 .vop_write = nfs_write,
168 .vop_nresolve = nfs_nresolve
172 * Special device vnode ops
174 struct vop_ops nfsv2_spec_vops = {
175 .vop_default = spec_vnoperate,
176 .vop_access = nfsspec_access,
177 .vop_close = nfsspec_close,
178 .vop_fsync = nfs_fsync,
179 .vop_getattr = nfs_getattr,
180 .vop_inactive = nfs_inactive,
181 .vop_print = nfs_print,
182 .vop_read = nfsspec_read,
183 .vop_reclaim = nfs_reclaim,
184 .vop_setattr = nfs_setattr,
185 .vop_write = nfsspec_write
188 struct vop_ops nfsv2_fifo_vops = {
189 .vop_default = fifo_vnoperate,
190 .vop_access = nfsspec_access,
191 .vop_close = nfsfifo_close,
192 .vop_fsync = nfs_fsync,
193 .vop_getattr = nfs_getattr,
194 .vop_inactive = nfs_inactive,
195 .vop_print = nfs_print,
196 .vop_read = nfsfifo_read,
197 .vop_reclaim = nfs_reclaim,
198 .vop_setattr = nfs_setattr,
199 .vop_write = nfsfifo_write
202 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
203 struct componentname *cnp,
205 static int nfs_removerpc (struct vnode *dvp, const char *name,
207 struct ucred *cred, struct thread *td);
208 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
209 int fnamelen, struct vnode *tdvp,
210 const char *tnameptr, int tnamelen,
211 struct ucred *cred, struct thread *td);
212 static int nfs_renameit (struct vnode *sdvp,
213 struct componentname *scnp,
214 struct sillyrename *sp);
216 SYSCTL_DECL(_vfs_nfs);
218 static int nfs_flush_on_rename = 1;
219 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
220 &nfs_flush_on_rename, 0, "flush fvp prior to rename");
221 static int nfs_flush_on_hlink = 0;
222 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
223 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
225 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
226 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
227 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
229 static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
230 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
231 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
233 static int nfspos_cache_timeout = NFS_MINATTRTIMO;
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
235 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
237 static int nfsv3_commit_on_close = 0;
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
239 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
241 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
242 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
244 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
245 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
248 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
249 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
250 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
252 nfs3_access_otw(struct vnode *vp, int wmode,
253 struct thread *td, struct ucred *cred)
255 struct nfsnode *np = VTONFS(vp);
260 struct nfsm_info info;
265 nfsstats.rpccnt[NFSPROC_ACCESS]++;
266 nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
267 NFSX_FH(info.v3) + NFSX_UNSIGNED);
268 ERROROUT(nfsm_fhtom(&info, vp));
269 tl = nfsm_build(&info, NFSX_UNSIGNED);
270 *tl = txdr_unsigned(wmode);
271 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
272 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
274 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
275 rmode = fxdr_unsigned(u_int32_t, *tl);
277 np->n_modeuid = cred->cr_uid;
278 np->n_modestamp = mycpu->gd_time_seconds;
287 * nfs access vnode op.
288 * For nfs version 2, just return ok. File accesses may fail later.
289 * For nfs version 3, use the access rpc to check accessibility. If file modes
290 * are changed on the server, accesses might still fail later.
292 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
295 nfs_access(struct vop_access_args *ap)
297 struct vnode *vp = ap->a_vp;
298 thread_t td = curthread;
300 u_int32_t mode, wmode;
301 struct nfsnode *np = VTONFS(vp);
302 int v3 = NFS_ISV3(vp);
305 * Disallow write attempts on filesystems mounted read-only;
306 * unless the file is a socket, fifo, or a block or character
307 * device resident on the filesystem.
309 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
310 switch (vp->v_type) {
320 * For nfs v3, check to see if we have done this recently, and if
321 * so return our cached result instead of making an ACCESS call.
322 * If not, do an access rpc, otherwise you are stuck emulating
323 * ufs_access() locally using the vattr. This may not be correct,
324 * since the server may apply other access criteria such as
325 * client uid-->server uid mapping that we do not know about.
328 if (ap->a_mode & VREAD)
329 mode = NFSV3ACCESS_READ;
332 if (vp->v_type != VDIR) {
333 if (ap->a_mode & VWRITE)
334 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
335 if (ap->a_mode & VEXEC)
336 mode |= NFSV3ACCESS_EXECUTE;
338 if (ap->a_mode & VWRITE)
339 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
341 if (ap->a_mode & VEXEC)
342 mode |= NFSV3ACCESS_LOOKUP;
344 /* XXX safety belt, only make blanket request if caching */
345 if (nfsaccess_cache_timeout > 0) {
346 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
347 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
348 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
354 * Does our cached result allow us to give a definite yes to
357 if (np->n_modestamp &&
358 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
359 (ap->a_cred->cr_uid == np->n_modeuid) &&
360 ((np->n_mode & mode) == mode)) {
361 nfsstats.accesscache_hits++;
364 * Either a no, or a don't know. Go to the wire.
366 nfsstats.accesscache_misses++;
367 error = nfs3_access_otw(vp, wmode, td, ap->a_cred);
369 if ((np->n_mode & mode) != mode) {
375 if ((error = nfsspec_access(ap)) != 0)
379 * Attempt to prevent a mapped root from accessing a file
380 * which it shouldn't. We try to read a byte from the file
381 * if the user is root and the file is not zero length.
382 * After calling nfsspec_access, we should have the correct
385 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
386 && VTONFS(vp)->n_size > 0) {
393 auio.uio_iov = &aiov;
397 auio.uio_segflg = UIO_SYSSPACE;
398 auio.uio_rw = UIO_READ;
401 if (vp->v_type == VREG) {
402 error = nfs_readrpc(vp, &auio);
403 } else if (vp->v_type == VDIR) {
405 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
407 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
408 error = nfs_readdirrpc(vp, &auio);
410 } else if (vp->v_type == VLNK) {
411 error = nfs_readlinkrpc(vp, &auio);
418 * [re]record creds for reading and/or writing if access
419 * was granted. Assume the NFS server will grant read access
420 * for execute requests.
423 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) {
426 crfree(np->n_rucred);
427 np->n_rucred = ap->a_cred;
429 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
432 crfree(np->n_wucred);
433 np->n_wucred = ap->a_cred;
441 * Check to see if the type is ok
442 * and that deletion is not in progress.
443 * For paged in text files, you will need to flush the page cache
444 * if consistency is lost.
446 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
451 nfs_open(struct vop_open_args *ap)
453 struct vnode *vp = ap->a_vp;
454 struct nfsnode *np = VTONFS(vp);
458 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
460 kprintf("open eacces vtyp=%d\n",vp->v_type);
466 * Save valid creds for reading and writing for later RPCs.
468 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
471 crfree(np->n_rucred);
472 np->n_rucred = ap->a_cred;
474 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
477 crfree(np->n_wucred);
478 np->n_wucred = ap->a_cred;
482 * Clear the attribute cache only if opening with write access. It
483 * is unclear if we should do this at all here, but we certainly
484 * should not clear the cache unconditionally simply because a file
487 if (ap->a_mode & FWRITE)
491 * For normal NFS, reconcile changes made locally verses
492 * changes made remotely. Note that VOP_GETATTR only goes
493 * to the wire if the cached attribute has timed out or been
496 * If local modifications have been made clear the attribute
497 * cache to force an attribute and modified time check. If
498 * GETATTR detects that the file has been changed by someone
499 * other then us it will set NRMODIFIED.
501 * If we are opening a directory and local changes have been
502 * made we have to invalidate the cache in order to ensure
503 * that we get the most up-to-date information from the
506 if (np->n_flag & NLMODIFIED) {
508 if (vp->v_type == VDIR) {
509 error = nfs_vinvalbuf(vp, V_SAVE, 1);
515 error = VOP_GETATTR(vp, &vattr);
518 if (np->n_flag & NRMODIFIED) {
519 if (vp->v_type == VDIR)
521 error = nfs_vinvalbuf(vp, V_SAVE, 1);
524 np->n_flag &= ~NRMODIFIED;
527 return (vop_stdopen(ap));
532 * What an NFS client should do upon close after writing is a debatable issue.
533 * Most NFS clients push delayed writes to the server upon close, basically for
535 * 1 - So that any write errors may be reported back to the client process
536 * doing the close system call. By far the two most likely errors are
537 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
538 * 2 - To put a worst case upper bound on cache inconsistency between
539 * multiple clients for the file.
540 * There is also a consistency problem for Version 2 of the protocol w.r.t.
541 * not being able to tell if other clients are writing a file concurrently,
542 * since there is no way of knowing if the changed modify time in the reply
543 * is only due to the write for this client.
544 * (NFS Version 3 provides weak cache consistency data in the reply that
545 * should be sufficient to detect and handle this case.)
547 * The current code does the following:
548 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
549 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
550 * or commit them (this satisfies 1 and 2 except for the
551 * case where the server crashes after this close but
552 * before the commit RPC, which is felt to be "good
553 * enough". Changing the last argument to nfs_flush() to
554 * a 1 would force a commit operation, if it is felt a
555 * commit is necessary now.
556 * for NQNFS - do nothing now, since 2 is dealt with via leases and
557 * 1 should be dealt with via an fsync() system call for
558 * cases where write errors are important.
560 * nfs_close(struct vnode *a_vp, int a_fflag)
564 nfs_close(struct vop_close_args *ap)
566 struct vnode *vp = ap->a_vp;
567 struct nfsnode *np = VTONFS(vp);
569 thread_t td = curthread;
571 if (vp->v_type == VREG) {
572 if (np->n_flag & NLMODIFIED) {
575 * Under NFSv3 we have dirty buffers to dispose of. We
576 * must flush them to the NFS server. We have the option
577 * of waiting all the way through the commit rpc or just
578 * waiting for the initial write. The default is to only
579 * wait through the initial write so the data is in the
580 * server's cache, which is roughly similar to the state
581 * a standard disk subsystem leaves the file in on close().
583 * We cannot clear the NLMODIFIED bit in np->n_flag due to
584 * potential races with other processes, and certainly
585 * cannot clear it if we don't commit.
587 int cm = nfsv3_commit_on_close ? 1 : 0;
588 error = nfs_flush(vp, MNT_WAIT, td, cm);
589 /* np->n_flag &= ~NLMODIFIED; */
591 error = nfs_vinvalbuf(vp, V_SAVE, 1);
595 if (np->n_flag & NWRITEERR) {
596 np->n_flag &= ~NWRITEERR;
605 * nfs getattr call from vfs.
607 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
610 nfs_getattr(struct vop_getattr_args *ap)
612 struct vnode *vp = ap->a_vp;
613 struct nfsnode *np = VTONFS(vp);
615 thread_t td = curthread;
616 struct nfsm_info info;
619 info.v3 = NFS_ISV3(vp);
622 * Update local times for special files.
624 if (np->n_flag & (NACC | NUPD))
627 * First look in the cache.
629 if (nfs_getattrcache(vp, ap->a_vap) == 0)
632 if (info.v3 && nfsaccess_cache_timeout > 0) {
633 nfsstats.accesscache_misses++;
634 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
635 if (nfs_getattrcache(vp, ap->a_vap) == 0)
639 nfsstats.rpccnt[NFSPROC_GETATTR]++;
640 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
641 ERROROUT(nfsm_fhtom(&info, vp));
642 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
643 nfs_vpcred(vp, ND_CHECK), &error));
645 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
656 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
659 nfs_setattr(struct vop_setattr_args *ap)
661 struct vnode *vp = ap->a_vp;
662 struct nfsnode *np = VTONFS(vp);
663 struct vattr *vap = ap->a_vap;
666 thread_t td = curthread;
673 * Setting of flags is not supported.
675 if (vap->va_flags != VNOVAL)
679 * Disallow write attempts if the filesystem is mounted read-only.
681 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
682 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
683 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
684 (vp->v_mount->mnt_flag & MNT_RDONLY))
687 if (vap->va_size != VNOVAL) {
689 * truncation requested
691 switch (vp->v_type) {
698 if (vap->va_mtime.tv_sec == VNOVAL &&
699 vap->va_atime.tv_sec == VNOVAL &&
700 vap->va_mode == (mode_t)VNOVAL &&
701 vap->va_uid == (uid_t)VNOVAL &&
702 vap->va_gid == (gid_t)VNOVAL)
704 vap->va_size = VNOVAL;
708 * Disallow write attempts if the filesystem is
711 if (vp->v_mount->mnt_flag & MNT_RDONLY)
715 * This is nasty. The RPCs we send to flush pending
716 * data often return attribute information which is
717 * cached via a callback to nfs_loadattrcache(), which
718 * has the effect of changing our notion of the file
719 * size. Due to flushed appends and other operations
720 * the file size can be set to virtually anything,
721 * including values that do not match either the old
722 * or intended file size.
724 * When this condition is detected we must loop to
725 * try the operation again. Hopefully no more
726 * flushing is required on the loop so it works the
727 * second time around. THIS CASE ALMOST ALWAYS
732 error = nfs_meta_setsize(vp, td, vap->va_size);
734 if (np->n_flag & NLMODIFIED) {
735 if (vap->va_size == 0)
736 error = nfs_vinvalbuf(vp, 0, 1);
738 error = nfs_vinvalbuf(vp, V_SAVE, 1);
741 * note: this loop case almost always happens at
742 * least once per truncation.
744 if (error == 0 && np->n_size != vap->va_size)
746 np->n_vattr.va_size = vap->va_size;
749 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
751 * What to do. If we are modifying the mtime we lose
752 * mtime detection of changes made by the server or other
753 * clients. But programs like rsync/rdist/cpdup are going
754 * to call utimes a lot. We don't want to piecemeal sync.
756 * For now sync if any prior remote changes were detected,
757 * but allow us to lose track of remote changes made during
758 * the utimes operation.
760 if (np->n_flag & NRMODIFIED)
761 error = nfs_vinvalbuf(vp, V_SAVE, 1);
765 if (vap->va_mtime.tv_sec != VNOVAL) {
766 np->n_mtime = vap->va_mtime.tv_sec;
770 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
773 * Sanity check if a truncation was issued. This should only occur
774 * if multiple processes are racing on the same file.
776 if (error == 0 && vap->va_size != VNOVAL &&
777 np->n_size != vap->va_size) {
778 kprintf("NFS ftruncate: server disagrees on the file size: "
781 (long long)vap->va_size,
782 (long long)np->n_size);
785 if (error && vap->va_size != VNOVAL) {
786 np->n_size = np->n_vattr.va_size = tsize;
787 vnode_pager_setsize(vp, np->n_size);
793 * Do an nfs setattr rpc.
796 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
797 struct ucred *cred, struct thread *td)
799 struct nfsv2_sattr *sp;
800 struct nfsnode *np = VTONFS(vp);
802 int error = 0, wccflag = NFSV3_WCCRATTR;
803 struct nfsm_info info;
806 info.v3 = NFS_ISV3(vp);
808 nfsstats.rpccnt[NFSPROC_SETATTR]++;
809 nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
810 NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
811 ERROROUT(nfsm_fhtom(&info, vp));
813 nfsm_v3attrbuild(&info, vap, TRUE);
814 tl = nfsm_build(&info, NFSX_UNSIGNED);
817 sp = nfsm_build(&info, NFSX_V2SATTR);
818 if (vap->va_mode == (mode_t)VNOVAL)
819 sp->sa_mode = nfs_xdrneg1;
821 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
822 if (vap->va_uid == (uid_t)VNOVAL)
823 sp->sa_uid = nfs_xdrneg1;
825 sp->sa_uid = txdr_unsigned(vap->va_uid);
826 if (vap->va_gid == (gid_t)VNOVAL)
827 sp->sa_gid = nfs_xdrneg1;
829 sp->sa_gid = txdr_unsigned(vap->va_gid);
830 sp->sa_size = txdr_unsigned(vap->va_size);
831 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
832 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
834 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
837 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
839 ERROROUT(nfsm_loadattr(&info, vp, NULL));
849 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
855 cache_setvp(nch, vp);
856 cache_settimeout(nch, nctimeout);
860 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
861 * nfs_lookup() until all remaining new api calls are implemented.
863 * Resolve a namecache entry. This function is passed a locked ncp and
864 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
867 nfs_nresolve(struct vop_nresolve_args *ap)
869 struct thread *td = curthread;
870 struct namecache *ncp;
880 struct nfsm_info info;
885 if ((error = vget(dvp, LK_SHARED)) != 0)
889 info.v3 = NFS_ISV3(dvp);
892 nfsstats.lookupcache_misses++;
893 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
894 ncp = ap->a_nch->ncp;
896 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
897 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
898 ERROROUT(nfsm_fhtom(&info, dvp));
899 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
900 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
901 ap->a_cred, &error));
904 * Cache negatve lookups to reduce NFS traffic, but use
905 * a fast timeout. Otherwise use a timeout of 1 tick.
906 * XXX we should add a namecache flag for no-caching
907 * to uncache the negative hit as soon as possible, but
908 * we cannot simply destroy the entry because it is used
909 * as a placeholder by the caller.
912 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
913 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
914 NFS_LATTR_NOSHRINK));
921 * Success, get the file handle, do various checks, and load
922 * post-operation data from the reply packet. Theoretically
923 * we should never be looking up "." so, theoretically, we
924 * should never get the same file handle as our directory. But
925 * we check anyway. XXX
927 * Note that no timeout is set for the positive cache hit. We
928 * assume, theoretically, that ESTALE returns will be dealt with
929 * properly to handle NFS races and in anycase we cannot depend
930 * on a timeout to deal with NFS open/create/excl issues so instead
931 * of a bad hack here the rest of the NFS client code needs to do
934 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
937 if (NFS_CMPFH(np, fhp, fhsize)) {
941 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
951 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
952 NFS_LATTR_NOSHRINK));
953 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
954 NFS_LATTR_NOSHRINK));
956 ERROROUT(nfsm_loadattr(&info, nvp, NULL));
958 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
973 * 'cached' nfs directory lookup
975 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
977 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
978 * struct componentname *a_cnp)
981 nfs_lookup(struct vop_old_lookup_args *ap)
983 struct componentname *cnp = ap->a_cnp;
984 struct vnode *dvp = ap->a_dvp;
985 struct vnode **vpp = ap->a_vpp;
986 int flags = cnp->cn_flags;
988 struct nfsmount *nmp;
992 int lockparent, wantparent, error = 0, attrflag, fhsize;
993 struct nfsm_info info;
996 info.v3 = NFS_ISV3(dvp);
999 * Read-only mount check and directory check.
1002 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1003 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1006 if (dvp->v_type != VDIR)
1010 * Look it up in the cache. Note that ENOENT is only returned if we
1011 * previously entered a negative hit (see later on). The additional
1012 * nfsneg_cache_timeout check causes previously cached results to
1013 * be instantly ignored if the negative caching is turned off.
1015 lockparent = flags & CNP_LOCKPARENT;
1016 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1017 nmp = VFSTONFS(dvp->v_mount);
1025 nfsstats.lookupcache_misses++;
1026 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1027 len = cnp->cn_namelen;
1028 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1029 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1030 ERROROUT(nfsm_fhtom(&info, dvp));
1031 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1032 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1033 cnp->cn_cred, &error));
1035 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1036 NFS_LATTR_NOSHRINK));
1041 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1044 * Handle RENAME case...
1046 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1047 if (NFS_CMPFH(np, fhp, fhsize)) {
1052 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1060 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1061 NFS_LATTR_NOSHRINK));
1062 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1063 NFS_LATTR_NOSHRINK));
1065 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1072 cnp->cn_flags |= CNP_PDIRUNLOCK;
1077 if (flags & CNP_ISDOTDOT) {
1079 cnp->cn_flags |= CNP_PDIRUNLOCK;
1080 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1082 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1083 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1084 return (error); /* NOTE: return error from nget */
1088 error = vn_lock(dvp, LK_EXCLUSIVE);
1093 cnp->cn_flags |= CNP_PDIRUNLOCK;
1095 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1099 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1107 cnp->cn_flags |= CNP_PDIRUNLOCK;
1112 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1113 NFS_LATTR_NOSHRINK));
1114 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1115 NFS_LATTR_NOSHRINK));
1117 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1120 /* XXX MOVE TO nfs_nremove() */
1121 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1122 cnp->cn_nameiop != NAMEI_DELETE) {
1123 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1131 if (newvp != NULLVP) {
1135 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1136 cnp->cn_nameiop == NAMEI_RENAME) &&
1140 cnp->cn_flags |= CNP_PDIRUNLOCK;
1142 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1145 error = EJUSTRETURN;
1153 * Just call nfs_bioread() to do the work.
1155 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1156 * struct ucred *a_cred)
1159 nfs_read(struct vop_read_args *ap)
1161 struct vnode *vp = ap->a_vp;
1163 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1169 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1172 nfs_readlink(struct vop_readlink_args *ap)
1174 struct vnode *vp = ap->a_vp;
1176 if (vp->v_type != VLNK)
1178 return (nfs_bioread(vp, ap->a_uio, 0));
1182 * Do a readlink rpc.
1183 * Called by nfs_doio() from below the buffer cache.
1186 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1188 int error = 0, len, attrflag;
1189 struct nfsm_info info;
1192 info.v3 = NFS_ISV3(vp);
1194 nfsstats.rpccnt[NFSPROC_READLINK]++;
1195 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1196 ERROROUT(nfsm_fhtom(&info, vp));
1197 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1198 nfs_vpcred(vp, ND_CHECK), &error));
1200 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1201 NFS_LATTR_NOSHRINK));
1204 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1205 if (len == NFS_MAXPATHLEN) {
1206 struct nfsnode *np = VTONFS(vp);
1207 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1210 ERROROUT(nfsm_mtouio(&info, uiop, len));
1223 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1226 struct nfsmount *nmp;
1227 int error = 0, len, retlen, tsiz, eof, attrflag;
1228 struct nfsm_info info;
1231 info.v3 = NFS_ISV3(vp);
1236 nmp = VFSTONFS(vp->v_mount);
1237 tsiz = uiop->uio_resid;
1238 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1241 nfsstats.rpccnt[NFSPROC_READ]++;
1242 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1243 nfsm_reqhead(&info, vp, NFSPROC_READ,
1244 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1245 ERROROUT(nfsm_fhtom(&info, vp));
1246 tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1248 txdr_hyper(uiop->uio_offset, tl);
1249 *(tl + 2) = txdr_unsigned(len);
1251 *tl++ = txdr_unsigned(uiop->uio_offset);
1252 *tl++ = txdr_unsigned(len);
1255 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1256 nfs_vpcred(vp, ND_READ), &error));
1258 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1259 NFS_LATTR_NOSHRINK));
1260 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1261 eof = fxdr_unsigned(int, *(tl + 1));
1263 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1265 NEGATIVEOUT(retlen = nfsm_strsiz(&info, nmp->nm_rsize));
1266 ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1271 if (eof || retlen == 0) {
1274 } else if (retlen < len) {
1286 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1290 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1291 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1292 int committed = NFSV3WRITE_FILESYNC;
1293 struct nfsm_info info;
1296 info.v3 = NFS_ISV3(vp);
1299 if (uiop->uio_iovcnt != 1)
1300 panic("nfs: writerpc iovcnt > 1");
1303 tsiz = uiop->uio_resid;
1304 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1307 nfsstats.rpccnt[NFSPROC_WRITE]++;
1308 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1309 nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1310 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1311 ERROROUT(nfsm_fhtom(&info, vp));
1313 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1314 txdr_hyper(uiop->uio_offset, tl);
1316 *tl++ = txdr_unsigned(len);
1317 *tl++ = txdr_unsigned(*iomode);
1318 *tl = txdr_unsigned(len);
1322 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1323 /* Set both "begin" and "current" to non-garbage. */
1324 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1325 *tl++ = x; /* "begin offset" */
1326 *tl++ = x; /* "current offset" */
1327 x = txdr_unsigned(len);
1328 *tl++ = x; /* total to this offset */
1329 *tl = x; /* size of this write */
1331 ERROROUT(nfsm_uiotom(&info, uiop, len));
1332 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1333 nfs_vpcred(vp, ND_WRITE), &error));
1336 * The write RPC returns a before and after mtime. The
1337 * nfsm_wcc_data() macro checks the before n_mtime
1338 * against the before time and stores the after time
1339 * in the nfsnode's cached vattr and n_mtime field.
1340 * The NRMODIFIED bit will be set if the before
1341 * time did not match the original mtime.
1343 wccflag = NFSV3_WCCCHK;
1344 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1346 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1347 rlen = fxdr_unsigned(int, *tl++);
1353 } else if (rlen < len) {
1354 backup = len - rlen;
1355 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1356 uiop->uio_iov->iov_len += backup;
1357 uiop->uio_offset -= backup;
1358 uiop->uio_resid += backup;
1361 commit = fxdr_unsigned(int, *tl++);
1364 * Return the lowest committment level
1365 * obtained by any of the RPCs.
1367 if (committed == NFSV3WRITE_FILESYNC)
1369 else if (committed == NFSV3WRITE_DATASYNC &&
1370 commit == NFSV3WRITE_UNSTABLE)
1372 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1373 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1375 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1376 } else if (bcmp((caddr_t)tl,
1377 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1379 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1384 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1393 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1394 committed = NFSV3WRITE_FILESYNC;
1395 *iomode = committed;
1397 uiop->uio_resid = tsiz;
1403 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1404 * mode set to specify the file type and the size field for rdev.
1407 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1410 struct nfsv2_sattr *sp;
1412 struct vnode *newvp = NULL;
1413 struct nfsnode *np = NULL;
1415 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1417 struct nfsm_info info;
1420 info.v3 = NFS_ISV3(dvp);
1422 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1423 rmajor = txdr_unsigned(vap->va_rmajor);
1424 rminor = txdr_unsigned(vap->va_rminor);
1425 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1426 rmajor = nfs_xdrneg1;
1427 rminor = nfs_xdrneg1;
1429 return (EOPNOTSUPP);
1431 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1434 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1435 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1436 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1437 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1438 ERROROUT(nfsm_fhtom(&info, dvp));
1439 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1442 tl = nfsm_build(&info, NFSX_UNSIGNED);
1443 *tl++ = vtonfsv3_type(vap->va_type);
1444 nfsm_v3attrbuild(&info, vap, FALSE);
1445 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1446 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1447 *tl++ = txdr_unsigned(vap->va_rmajor);
1448 *tl = txdr_unsigned(vap->va_rminor);
1451 sp = nfsm_build(&info, NFSX_V2SATTR);
1452 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1453 sp->sa_uid = nfs_xdrneg1;
1454 sp->sa_gid = nfs_xdrneg1;
1455 sp->sa_size = makeudev(rmajor, rminor);
1456 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1457 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1459 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1460 cnp->cn_cred, &error));
1462 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1468 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1469 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1475 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1486 VTONFS(dvp)->n_flag |= NLMODIFIED;
1488 VTONFS(dvp)->n_attrstamp = 0;
1494 * just call nfs_mknodrpc() to do the work.
1496 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1497 * struct componentname *a_cnp, struct vattr *a_vap)
1501 nfs_mknod(struct vop_old_mknod_args *ap)
1503 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1506 static u_long create_verf;
1508 * nfs file create call
1510 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1511 * struct componentname *a_cnp, struct vattr *a_vap)
1514 nfs_create(struct vop_old_create_args *ap)
1516 struct vnode *dvp = ap->a_dvp;
1517 struct vattr *vap = ap->a_vap;
1518 struct componentname *cnp = ap->a_cnp;
1519 struct nfsv2_sattr *sp;
1521 struct nfsnode *np = NULL;
1522 struct vnode *newvp = NULL;
1523 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1525 struct nfsm_info info;
1528 info.v3 = NFS_ISV3(dvp);
1531 * Oops, not for me..
1533 if (vap->va_type == VSOCK)
1534 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1536 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1539 if (vap->va_vaflags & VA_EXCLUSIVE)
1542 nfsstats.rpccnt[NFSPROC_CREATE]++;
1543 nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1544 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1545 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1546 ERROROUT(nfsm_fhtom(&info, dvp));
1547 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1550 tl = nfsm_build(&info, NFSX_UNSIGNED);
1551 if (fmode & O_EXCL) {
1552 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1553 tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1555 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1556 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1559 *tl++ = create_verf;
1560 *tl = ++create_verf;
1562 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1563 nfsm_v3attrbuild(&info, vap, FALSE);
1566 sp = nfsm_build(&info, NFSX_V2SATTR);
1567 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1568 sp->sa_uid = nfs_xdrneg1;
1569 sp->sa_gid = nfs_xdrneg1;
1571 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1572 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1574 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1575 cnp->cn_cred, &error));
1577 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1583 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1584 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1590 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1596 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1597 KKASSERT(newvp == NULL);
1601 } else if (info.v3 && (fmode & O_EXCL)) {
1603 * We are normally called with only a partially initialized
1604 * VAP. Since the NFSv3 spec says that server may use the
1605 * file attributes to store the verifier, the spec requires
1606 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1607 * in atime, but we can't really assume that all servers will
1608 * so we ensure that our SETATTR sets both atime and mtime.
1610 if (vap->va_mtime.tv_sec == VNOVAL)
1611 vfs_timestamp(&vap->va_mtime);
1612 if (vap->va_atime.tv_sec == VNOVAL)
1613 vap->va_atime = vap->va_mtime;
1614 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1618 * The new np may have enough info for access
1619 * checks, make sure rucred and wucred are
1620 * initialized for read and write rpc's.
1623 if (np->n_rucred == NULL)
1624 np->n_rucred = crhold(cnp->cn_cred);
1625 if (np->n_wucred == NULL)
1626 np->n_wucred = crhold(cnp->cn_cred);
1631 VTONFS(dvp)->n_flag |= NLMODIFIED;
1633 VTONFS(dvp)->n_attrstamp = 0;
1638 * nfs file remove call
1639 * To try and make nfs semantics closer to ufs semantics, a file that has
1640 * other processes using the vnode is renamed instead of removed and then
1641 * removed later on the last close.
1642 * - If v_sysref.refcnt > 1
1643 * If a rename is not already in the works
1644 * call nfs_sillyrename() to set it up
1648 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1649 * struct componentname *a_cnp)
1652 nfs_remove(struct vop_old_remove_args *ap)
1654 struct vnode *vp = ap->a_vp;
1655 struct vnode *dvp = ap->a_dvp;
1656 struct componentname *cnp = ap->a_cnp;
1657 struct nfsnode *np = VTONFS(vp);
1662 if (vp->v_sysref.refcnt < 1)
1663 panic("nfs_remove: bad v_sysref.refcnt");
1665 if (vp->v_type == VDIR)
1667 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1668 VOP_GETATTR(vp, &vattr) == 0 &&
1669 vattr.va_nlink > 1)) {
1671 * throw away biocache buffers, mainly to avoid
1672 * unnecessary delayed writes later.
1674 error = nfs_vinvalbuf(vp, 0, 1);
1677 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1678 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1680 * Kludge City: If the first reply to the remove rpc is lost..
1681 * the reply to the retransmitted request will be ENOENT
1682 * since the file was in fact removed
1683 * Therefore, we cheat and return success.
1685 if (error == ENOENT)
1687 } else if (!np->n_sillyrename) {
1688 error = nfs_sillyrename(dvp, vp, cnp);
1690 np->n_attrstamp = 0;
1695 * nfs file remove rpc called from nfs_inactive
1698 nfs_removeit(struct sillyrename *sp)
1700 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1705 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1708 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1709 struct ucred *cred, struct thread *td)
1711 int error = 0, wccflag = NFSV3_WCCRATTR;
1712 struct nfsm_info info;
1715 info.v3 = NFS_ISV3(dvp);
1717 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1718 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1719 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1720 ERROROUT(nfsm_fhtom(&info, dvp));
1721 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1722 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1724 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1729 VTONFS(dvp)->n_flag |= NLMODIFIED;
1731 VTONFS(dvp)->n_attrstamp = 0;
1736 * nfs file rename call
1738 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1739 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1740 * struct vnode *a_tvp, struct componentname *a_tcnp)
1743 nfs_rename(struct vop_old_rename_args *ap)
1745 struct vnode *fvp = ap->a_fvp;
1746 struct vnode *tvp = ap->a_tvp;
1747 struct vnode *fdvp = ap->a_fdvp;
1748 struct vnode *tdvp = ap->a_tdvp;
1749 struct componentname *tcnp = ap->a_tcnp;
1750 struct componentname *fcnp = ap->a_fcnp;
1753 /* Check for cross-device rename */
1754 if ((fvp->v_mount != tdvp->v_mount) ||
1755 (tvp && (fvp->v_mount != tvp->v_mount))) {
1761 * We shouldn't have to flush fvp on rename for most server-side
1762 * filesystems as the file handle should not change. Unfortunately
1763 * the inode for some filesystems (msdosfs) might be tied to the
1764 * file name or directory position so to be completely safe
1765 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1768 * We must flush tvp on rename because it might become stale on the
1769 * server after the rename.
1771 if (nfs_flush_on_rename)
1772 VOP_FSYNC(fvp, MNT_WAIT);
1774 VOP_FSYNC(tvp, MNT_WAIT);
1777 * If the tvp exists and is in use, sillyrename it before doing the
1778 * rename of the new file over it.
1780 * XXX Can't sillyrename a directory.
1782 * We do not attempt to do any namecache purges in this old API
1783 * routine. The new API compat functions have access to the actual
1784 * namecache structures and will do it for us.
1786 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1787 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1794 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1795 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1808 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1810 if (error == ENOENT)
1816 * nfs file rename rpc called from nfs_remove() above
1819 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1820 struct sillyrename *sp)
1822 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1823 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1827 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1830 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1831 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1832 struct ucred *cred, struct thread *td)
1834 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1835 struct nfsm_info info;
1838 info.v3 = NFS_ISV3(fdvp);
1840 nfsstats.rpccnt[NFSPROC_RENAME]++;
1841 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1842 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1843 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1844 ERROROUT(nfsm_fhtom(&info, fdvp));
1845 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1846 ERROROUT(nfsm_fhtom(&info, tdvp));
1847 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1848 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
1850 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
1851 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
1856 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1857 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1859 VTONFS(fdvp)->n_attrstamp = 0;
1861 VTONFS(tdvp)->n_attrstamp = 0;
1866 * nfs hard link create call
1868 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1869 * struct componentname *a_cnp)
1872 nfs_link(struct vop_old_link_args *ap)
1874 struct vnode *vp = ap->a_vp;
1875 struct vnode *tdvp = ap->a_tdvp;
1876 struct componentname *cnp = ap->a_cnp;
1877 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1878 struct nfsm_info info;
1880 if (vp->v_mount != tdvp->v_mount) {
1885 * The attribute cache may get out of sync with the server on link.
1886 * Pushing writes to the server before handle was inherited from
1887 * long long ago and it is unclear if we still need to do this.
1890 if (nfs_flush_on_hlink)
1891 VOP_FSYNC(vp, MNT_WAIT);
1894 info.v3 = NFS_ISV3(vp);
1896 nfsstats.rpccnt[NFSPROC_LINK]++;
1897 nfsm_reqhead(&info, vp, NFSPROC_LINK,
1898 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
1899 nfsm_rndup(cnp->cn_namelen));
1900 ERROROUT(nfsm_fhtom(&info, vp));
1901 ERROROUT(nfsm_fhtom(&info, tdvp));
1902 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1904 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
1905 cnp->cn_cred, &error));
1907 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1908 NFS_LATTR_NOSHRINK));
1909 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
1914 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1916 VTONFS(vp)->n_attrstamp = 0;
1918 VTONFS(tdvp)->n_attrstamp = 0;
1920 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1922 if (error == EEXIST)
1928 * nfs symbolic link create call
1930 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1931 * struct componentname *a_cnp, struct vattr *a_vap,
1935 nfs_symlink(struct vop_old_symlink_args *ap)
1937 struct vnode *dvp = ap->a_dvp;
1938 struct vattr *vap = ap->a_vap;
1939 struct componentname *cnp = ap->a_cnp;
1940 struct nfsv2_sattr *sp;
1941 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1942 struct vnode *newvp = NULL;
1943 struct nfsm_info info;
1946 info.v3 = NFS_ISV3(dvp);
1948 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1949 slen = strlen(ap->a_target);
1950 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
1951 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
1952 nfsm_rndup(cnp->cn_namelen) +
1953 nfsm_rndup(slen) + NFSX_SATTR(info.v3));
1954 ERROROUT(nfsm_fhtom(&info, dvp));
1955 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1958 nfsm_v3attrbuild(&info, vap, FALSE);
1960 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
1962 sp = nfsm_build(&info, NFSX_V2SATTR);
1963 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1964 sp->sa_uid = nfs_xdrneg1;
1965 sp->sa_gid = nfs_xdrneg1;
1966 sp->sa_size = nfs_xdrneg1;
1967 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1968 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1972 * Issue the NFS request and get the rpc response.
1974 * Only NFSv3 responses returning an error of 0 actually return
1975 * a file handle that can be converted into newvp without having
1976 * to do an extra lookup rpc.
1978 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
1979 cnp->cn_cred, &error));
1982 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1984 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1988 * out code jumps -> here, mrep is also freed.
1996 * If we get an EEXIST error, silently convert it to no-error
1997 * in case of an NFS retry.
1999 if (error == EEXIST)
2003 * If we do not have (or no longer have) an error, and we could
2004 * not extract the newvp from the response due to the request being
2005 * NFSv2 or the error being EEXIST. We have to do a lookup in order
2006 * to obtain a newvp to return.
2008 if (error == 0 && newvp == NULL) {
2009 struct nfsnode *np = NULL;
2011 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2012 cnp->cn_cred, cnp->cn_td, &np);
2022 VTONFS(dvp)->n_flag |= NLMODIFIED;
2024 VTONFS(dvp)->n_attrstamp = 0;
2031 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2032 * struct componentname *a_cnp, struct vattr *a_vap)
2035 nfs_mkdir(struct vop_old_mkdir_args *ap)
2037 struct vnode *dvp = ap->a_dvp;
2038 struct vattr *vap = ap->a_vap;
2039 struct componentname *cnp = ap->a_cnp;
2040 struct nfsv2_sattr *sp;
2041 struct nfsnode *np = NULL;
2042 struct vnode *newvp = NULL;
2044 int error = 0, wccflag = NFSV3_WCCRATTR;
2047 struct nfsm_info info;
2050 info.v3 = NFS_ISV3(dvp);
2052 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2055 len = cnp->cn_namelen;
2056 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2057 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2058 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2059 nfsm_rndup(len) + NFSX_SATTR(info.v3));
2060 ERROROUT(nfsm_fhtom(&info, dvp));
2061 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2063 nfsm_v3attrbuild(&info, vap, FALSE);
2065 sp = nfsm_build(&info, NFSX_V2SATTR);
2066 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2067 sp->sa_uid = nfs_xdrneg1;
2068 sp->sa_gid = nfs_xdrneg1;
2069 sp->sa_size = nfs_xdrneg1;
2070 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2071 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2073 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2074 cnp->cn_cred, &error));
2076 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2079 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2084 VTONFS(dvp)->n_flag |= NLMODIFIED;
2086 VTONFS(dvp)->n_attrstamp = 0;
2088 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2089 * if we can succeed in looking up the directory.
2091 if (error == EEXIST || (!error && !gotvp)) {
2096 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2100 if (newvp->v_type != VDIR)
2113 * nfs remove directory call
2115 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2116 * struct componentname *a_cnp)
2119 nfs_rmdir(struct vop_old_rmdir_args *ap)
2121 struct vnode *vp = ap->a_vp;
2122 struct vnode *dvp = ap->a_dvp;
2123 struct componentname *cnp = ap->a_cnp;
2124 int error = 0, wccflag = NFSV3_WCCRATTR;
2125 struct nfsm_info info;
2128 info.v3 = NFS_ISV3(dvp);
2132 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2133 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2134 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2135 nfsm_rndup(cnp->cn_namelen));
2136 ERROROUT(nfsm_fhtom(&info, dvp));
2137 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2139 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2140 cnp->cn_cred, &error));
2142 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2147 VTONFS(dvp)->n_flag |= NLMODIFIED;
2149 VTONFS(dvp)->n_attrstamp = 0;
2151 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2153 if (error == ENOENT)
2161 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2164 nfs_readdir(struct vop_readdir_args *ap)
2166 struct vnode *vp = ap->a_vp;
2167 struct nfsnode *np = VTONFS(vp);
2168 struct uio *uio = ap->a_uio;
2172 if (vp->v_type != VDIR)
2175 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2179 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2180 * and then check that is still valid, or if this is an NQNFS mount
2181 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2182 * VOP_GETATTR() does not necessarily go to the wire.
2184 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2185 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2186 if (VOP_GETATTR(vp, &vattr) == 0 &&
2187 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2189 nfsstats.direofcache_hits++;
2195 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2196 * own cache coherency checks so we do not have to.
2198 tresid = uio->uio_resid;
2199 error = nfs_bioread(vp, uio, 0);
2201 if (!error && uio->uio_resid == tresid)
2202 nfsstats.direofcache_misses++;
2209 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2211 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2212 * offset/block and converts the nfs formatted directory entries for userland
2213 * consumption as well as deals with offsets into the middle of blocks.
2214 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2215 * be block-bounded. It must convert to cookies for the actual RPC.
2218 nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2221 struct nfs_dirent *dp = NULL;
2226 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2227 struct nfsnode *dnp = VTONFS(vp);
2229 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2231 struct nfsm_info info;
2234 info.v3 = NFS_ISV3(vp);
2237 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2238 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2239 panic("nfs readdirrpc bad uio");
2243 * If there is no cookie, assume directory was stale.
2245 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2249 return (NFSERR_BAD_COOKIE);
2251 * Loop around doing readdir rpc's of size nm_readdirsize
2252 * truncated to a multiple of DIRBLKSIZ.
2253 * The stopping criteria is EOF or buffer full.
2255 while (more_dirs && bigenough) {
2256 nfsstats.rpccnt[NFSPROC_READDIR]++;
2257 nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2258 NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2259 ERROROUT(nfsm_fhtom(&info, vp));
2261 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2262 *tl++ = cookie.nfsuquad[0];
2263 *tl++ = cookie.nfsuquad[1];
2264 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2265 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2267 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2268 *tl++ = cookie.nfsuquad[0];
2270 *tl = txdr_unsigned(nmp->nm_readdirsize);
2271 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2273 nfs_vpcred(vp, ND_READ), &error));
2275 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2276 NFS_LATTR_NOSHRINK));
2277 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2278 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2279 dnp->n_cookieverf.nfsuquad[1] = *tl;
2281 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2282 more_dirs = fxdr_unsigned(int, *tl);
2284 /* loop thru the dir entries, converting them to std form */
2285 while (more_dirs && bigenough) {
2287 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2288 fileno = fxdr_hyper(tl);
2289 len = fxdr_unsigned(int, *(tl + 2));
2291 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2292 fileno = fxdr_unsigned(u_quad_t, *tl++);
2293 len = fxdr_unsigned(int, *tl);
2295 if (len <= 0 || len > NFS_MAXNAMLEN) {
2303 * len is the number of bytes in the path element
2304 * name, not including the \0 termination.
2306 * tlen is the number of bytes w have to reserve for
2307 * the path element name.
2309 tlen = nfsm_rndup(len);
2311 tlen += 4; /* To ensure null termination */
2314 * If the entry would cross a DIRBLKSIZ boundary,
2315 * extend the previous nfs_dirent to cover the
2318 left = DIRBLKSIZ - blksiz;
2319 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2320 dp->nfs_reclen += left;
2321 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2322 uiop->uio_iov->iov_len -= left;
2323 uiop->uio_offset += left;
2324 uiop->uio_resid -= left;
2327 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2330 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2331 dp->nfs_ino = fileno;
2332 dp->nfs_namlen = len;
2333 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2334 dp->nfs_type = DT_UNKNOWN;
2335 blksiz += dp->nfs_reclen;
2336 if (blksiz == DIRBLKSIZ)
2338 uiop->uio_offset += sizeof(struct nfs_dirent);
2339 uiop->uio_resid -= sizeof(struct nfs_dirent);
2340 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2341 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2342 ERROROUT(nfsm_mtouio(&info, uiop, len));
2345 * The uiop has advanced by nfs_dirent + len
2346 * but really needs to advance by
2349 cp = uiop->uio_iov->iov_base;
2351 *cp = '\0'; /* null terminate */
2352 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2353 uiop->uio_iov->iov_len -= tlen;
2354 uiop->uio_offset += tlen;
2355 uiop->uio_resid -= tlen;
2358 * NFS strings must be rounded up (nfsm_myouio
2359 * handled that in the bigenough case).
2361 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2364 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2366 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2370 * If we were able to accomodate the last entry,
2371 * get the cookie for the next one. Otherwise
2372 * hold-over the cookie for the one we were not
2373 * able to accomodate.
2376 cookie.nfsuquad[0] = *tl++;
2378 cookie.nfsuquad[1] = *tl++;
2379 } else if (info.v3) {
2384 more_dirs = fxdr_unsigned(int, *tl);
2387 * If at end of rpc data, get the eof boolean
2390 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2391 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2397 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2398 * by increasing d_reclen for the last record.
2401 left = DIRBLKSIZ - blksiz;
2402 dp->nfs_reclen += left;
2403 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2404 uiop->uio_iov->iov_len -= left;
2405 uiop->uio_offset += left;
2406 uiop->uio_resid -= left;
2411 * We hit the end of the directory, update direofoffset.
2413 dnp->n_direofoffset = uiop->uio_offset;
2416 * There is more to go, insert the link cookie so the
2417 * next block can be read.
2419 if (uiop->uio_resid > 0)
2420 kprintf("EEK! readdirrpc resid > 0\n");
2421 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2429 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2432 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2435 struct nfs_dirent *dp;
2437 struct vnode *newvp;
2439 caddr_t dpossav1, dpossav2;
2441 struct mbuf *mdsav1, *mdsav2;
2443 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2444 struct nfsnode *dnp = VTONFS(vp), *np;
2447 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2448 int attrflag, fhsize;
2449 struct nchandle nch;
2450 struct nchandle dnch;
2451 struct nlcomponent nlc;
2452 struct nfsm_info info;
2461 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2462 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2463 panic("nfs readdirplusrpc bad uio");
2466 * Obtain the namecache record for the directory so we have something
2467 * to use as a basis for creating the entries. This function will
2468 * return a held (but not locked) ncp. The ncp may be disconnected
2469 * from the tree and cannot be used for upward traversals, and the
2470 * ncp may be unnamed. Note that other unrelated operations may
2471 * cause the ncp to be named at any time.
2473 cache_fromdvp(vp, NULL, 0, &dnch);
2474 bzero(&nlc, sizeof(nlc));
2478 * If there is no cookie, assume directory was stale.
2480 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2484 return (NFSERR_BAD_COOKIE);
2486 * Loop around doing readdir rpc's of size nm_readdirsize
2487 * truncated to a multiple of DIRBLKSIZ.
2488 * The stopping criteria is EOF or buffer full.
2490 while (more_dirs && bigenough) {
2491 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2492 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2493 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2494 ERROROUT(nfsm_fhtom(&info, vp));
2495 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2496 *tl++ = cookie.nfsuquad[0];
2497 *tl++ = cookie.nfsuquad[1];
2498 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2499 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2500 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2501 *tl = txdr_unsigned(nmp->nm_rsize);
2502 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2504 nfs_vpcred(vp, ND_READ), &error));
2505 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2506 NFS_LATTR_NOSHRINK));
2507 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2508 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2509 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2510 more_dirs = fxdr_unsigned(int, *tl);
2512 /* loop thru the dir entries, doctoring them to 4bsd form */
2513 while (more_dirs && bigenough) {
2514 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2515 fileno = fxdr_hyper(tl);
2516 len = fxdr_unsigned(int, *(tl + 2));
2517 if (len <= 0 || len > NFS_MAXNAMLEN) {
2523 tlen = nfsm_rndup(len);
2525 tlen += 4; /* To ensure null termination*/
2526 left = DIRBLKSIZ - blksiz;
2527 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2528 dp->nfs_reclen += left;
2529 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2530 uiop->uio_iov->iov_len -= left;
2531 uiop->uio_offset += left;
2532 uiop->uio_resid -= left;
2535 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2538 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2539 dp->nfs_ino = fileno;
2540 dp->nfs_namlen = len;
2541 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2542 dp->nfs_type = DT_UNKNOWN;
2543 blksiz += dp->nfs_reclen;
2544 if (blksiz == DIRBLKSIZ)
2546 uiop->uio_offset += sizeof(struct nfs_dirent);
2547 uiop->uio_resid -= sizeof(struct nfs_dirent);
2548 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2549 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2550 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2551 nlc.nlc_namelen = len;
2552 ERROROUT(nfsm_mtouio(&info, uiop, len));
2553 cp = uiop->uio_iov->iov_base;
2556 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2557 uiop->uio_iov->iov_len -= tlen;
2558 uiop->uio_offset += tlen;
2559 uiop->uio_resid -= tlen;
2561 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2563 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2565 cookie.nfsuquad[0] = *tl++;
2566 cookie.nfsuquad[1] = *tl++;
2571 * Since the attributes are before the file handle
2572 * (sigh), we must skip over the attributes and then
2573 * come back and get them.
2575 attrflag = fxdr_unsigned(int, *tl);
2577 dpossav1 = info.dpos;
2579 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2580 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2581 doit = fxdr_unsigned(int, *tl);
2583 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2584 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2589 error = nfs_nget(vp->v_mount, fhp,
2597 if (doit && bigenough) {
2598 dpossav2 = info.dpos;
2599 info.dpos = dpossav1;
2602 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2603 info.dpos = dpossav2;
2606 IFTODT(VTTOIF(np->n_vattr.va_type));
2608 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2609 nlc.nlc_namelen, nlc.nlc_namelen,
2611 nch = cache_nlookup(&dnch, &nlc);
2612 cache_setunresolved(&nch);
2613 nfs_cache_setvp(&nch, newvp,
2614 nfspos_cache_timeout);
2617 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER"
2619 nlc.nlc_namelen, nlc.nlc_namelen,
2624 /* Just skip over the file handle */
2625 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2626 i = fxdr_unsigned(int, *tl);
2627 ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2629 if (newvp != NULLVP) {
2636 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2637 more_dirs = fxdr_unsigned(int, *tl);
2640 * If at end of rpc data, get the eof boolean
2643 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2644 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2650 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2651 * by increasing d_reclen for the last record.
2654 left = DIRBLKSIZ - blksiz;
2655 dp->nfs_reclen += left;
2656 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2657 uiop->uio_iov->iov_len -= left;
2658 uiop->uio_offset += left;
2659 uiop->uio_resid -= left;
2663 * We are now either at the end of the directory or have filled the
2667 dnp->n_direofoffset = uiop->uio_offset;
2669 if (uiop->uio_resid > 0)
2670 kprintf("EEK! readdirplusrpc resid > 0\n");
2671 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2675 if (newvp != NULLVP) {
2688 * Silly rename. To make the NFS filesystem that is stateless look a little
2689 * more like the "ufs" a remove of an active vnode is translated to a rename
2690 * to a funny looking filename that is removed by nfs_inactive on the
2691 * nfsnode. There is the potential for another process on a different client
2692 * to create the same funny name between the nfs_lookitup() fails and the
2693 * nfs_rename() completes, but...
2696 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2698 struct sillyrename *sp;
2703 * We previously purged dvp instead of vp. I don't know why, it
2704 * completely destroys performance. We can't do it anyway with the
2705 * new VFS API since we would be breaking the namecache topology.
2707 cache_purge(vp); /* XXX */
2710 if (vp->v_type == VDIR)
2711 panic("nfs: sillyrename dir");
2713 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2714 M_NFSREQ, M_WAITOK);
2715 sp->s_cred = crdup(cnp->cn_cred);
2719 /* Fudge together a funny name */
2720 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2721 (int)(intptr_t)cnp->cn_td);
2723 /* Try lookitups until we get one that isn't there */
2724 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2725 cnp->cn_td, NULL) == 0) {
2727 if (sp->s_name[4] > 'z') {
2732 error = nfs_renameit(dvp, cnp, sp);
2735 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2737 np->n_sillyrename = sp;
2742 kfree((caddr_t)sp, M_NFSREQ);
2747 * Look up a file name and optionally either update the file handle or
2748 * allocate an nfsnode, depending on the value of npp.
2749 * npp == NULL --> just do the lookup
2750 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2752 * *npp != NULL --> update the file handle in the vnode
2755 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2756 struct thread *td, struct nfsnode **npp)
2758 struct vnode *newvp = NULL;
2759 struct nfsnode *np, *dnp = VTONFS(dvp);
2760 int error = 0, fhlen, attrflag;
2762 struct nfsm_info info;
2765 info.v3 = NFS_ISV3(dvp);
2767 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2768 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2769 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2770 ERROROUT(nfsm_fhtom(&info, dvp));
2771 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2772 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2773 if (npp && !error) {
2774 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2777 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2778 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2779 np->n_fhp = &np->n_fh;
2780 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2781 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2782 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2783 np->n_fhsize = fhlen;
2785 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2789 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2798 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2799 NFS_LATTR_NOSHRINK));
2800 if (!attrflag && *npp == NULL) {
2810 ERROROUT(error = nfsm_loadattr(&info, newvp, NULL));
2816 if (npp && *npp == NULL) {
2831 * Nfs Version 3 commit rpc
2834 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2836 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2837 int error = 0, wccflag = NFSV3_WCCRATTR;
2838 struct nfsm_info info;
2844 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2846 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2847 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
2848 ERROROUT(nfsm_fhtom(&info, vp));
2849 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
2850 txdr_hyper(offset, tl);
2852 *tl = txdr_unsigned(cnt);
2853 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
2854 nfs_vpcred(vp, ND_WRITE), &error));
2855 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
2857 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
2858 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2859 NFSX_V3WRITEVERF)) {
2860 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2862 error = NFSERR_STALEWRITEVERF;
2873 * - make nfs_bmap() essentially a no-op that does no translation
2874 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2875 * (Maybe I could use the process's page mapping, but I was concerned that
2876 * Kernel Write might not be enabled and also figured copyout() would do
2877 * a lot more work than bcopy() and also it currently happens in the
2878 * context of the swapper process (2).
2880 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
2881 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2884 nfs_bmap(struct vop_bmap_args *ap)
2886 if (ap->a_doffsetp != NULL)
2887 *ap->a_doffsetp = ap->a_loffset;
2888 if (ap->a_runp != NULL)
2890 if (ap->a_runb != NULL)
2898 * For async requests when nfsiod(s) are running, queue the request by
2899 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2903 nfs_strategy(struct vop_strategy_args *ap)
2905 struct bio *bio = ap->a_bio;
2907 struct buf *bp = bio->bio_buf;
2911 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2912 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
2913 KASSERT(BUF_REFCNT(bp) > 0,
2914 ("nfs_strategy: buffer %p not locked", bp));
2916 if (bio->bio_flags & BIO_SYNC)
2917 td = curthread; /* XXX */
2922 * We probably don't need to push an nbio any more since no
2923 * block conversion is required due to the use of 64 bit byte
2924 * offsets, but do it anyway.
2926 nbio = push_bio(bio);
2927 nbio->bio_offset = bio->bio_offset;
2930 * If the op is asynchronous and an i/o daemon is waiting
2931 * queue the request, wake it up and wait for completion
2932 * otherwise just do it ourselves.
2934 if ((bio->bio_flags & BIO_SYNC) || nfs_asyncio(ap->a_vp, nbio, td))
2935 error = nfs_doio(ap->a_vp, nbio, td);
2942 * NB Currently unsupported.
2944 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
2948 nfs_mmap(struct vop_mmap_args *ap)
2954 * fsync vnode op. Just call nfs_flush() with commit == 1.
2956 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
2960 nfs_fsync(struct vop_fsync_args *ap)
2962 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
2966 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2967 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2968 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2969 * set the buffer contains data that has already been written to the server
2970 * and which now needs a commit RPC.
2972 * If commit is 0 we only take one pass and only flush buffers containing new
2975 * If commit is 1 we take two passes, issuing a commit RPC in the second
2978 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
2979 * to completely flush all pending data.
2981 * Note that the RB_SCAN code properly handles the case where the
2982 * callback might block and directly or indirectly (another thread) cause
2983 * the RB tree to change.
2986 #ifndef NFS_COMMITBVECSIZ
2987 #define NFS_COMMITBVECSIZ 16
2990 struct nfs_flush_info {
2991 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
2998 struct buf *bvary[NFS_COMMITBVECSIZ];
3004 static int nfs_flush_bp(struct buf *bp, void *data);
3005 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3008 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3010 struct nfsnode *np = VTONFS(vp);
3011 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3012 struct nfs_flush_info info;
3016 bzero(&info, sizeof(info));
3019 info.waitfor = waitfor;
3020 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3022 lwkt_gettoken(&vlock, &vp->v_token);
3028 info.mode = NFI_FLUSHNEW;
3029 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3030 nfs_flush_bp, &info);
3033 * Take a second pass if committing and no error occured.
3034 * Clean up any left over collection (whether an error
3037 if (commit && error == 0) {
3038 info.mode = NFI_COMMIT;
3039 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3040 nfs_flush_bp, &info);
3042 error = nfs_flush_docommit(&info, error);
3046 * Wait for pending I/O to complete before checking whether
3047 * any further dirty buffers exist.
3049 while (waitfor == MNT_WAIT &&
3050 bio_track_active(&vp->v_track_write)) {
3051 error = bio_track_wait(&vp->v_track_write,
3052 info.slpflag, info.slptimeo);
3055 * We have to be able to break out if this
3056 * is an 'intr' mount.
3058 if (nfs_sigintr(nmp, NULL, td)) {
3064 * Since we do not process pending signals,
3065 * once we get a PCATCH our tsleep() will no
3066 * longer sleep, switch to a fixed timeout
3069 if (info.slpflag == PCATCH) {
3071 info.slptimeo = 2 * hz;
3078 * Loop if we are flushing synchronous as well as committing,
3079 * and dirty buffers are still present. Otherwise we might livelock.
3081 } while (waitfor == MNT_WAIT && commit &&
3082 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3085 * The callbacks have to return a negative error to terminate the
3092 * Deal with any error collection
3094 if (np->n_flag & NWRITEERR) {
3095 error = np->n_error;
3096 np->n_flag &= ~NWRITEERR;
3098 lwkt_reltoken(&vlock);
3104 nfs_flush_bp(struct buf *bp, void *data)
3106 struct nfs_flush_info *info = data;
3112 switch(info->mode) {
3114 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3115 if (error && info->loops && info->waitfor == MNT_WAIT) {
3116 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3118 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3119 if (info->slpflag & PCATCH)
3120 lkflags |= LK_PCATCH;
3121 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3127 * Ignore locking errors
3135 * The buffer may have changed out from under us, even if
3136 * we did not block (MPSAFE). Check again now that it is
3139 if (bp->b_vp == info->vp &&
3140 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3149 * Only process buffers in need of a commit which we can
3150 * immediately lock. This may prevent a buffer from being
3151 * committed, but the normal flush loop will block on the
3152 * same buffer so we shouldn't get into an endless loop.
3154 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3155 (B_DELWRI | B_NEEDCOMMIT)) {
3158 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3162 * We must recheck after successfully locking the buffer.
3164 if (bp->b_vp != info->vp ||
3165 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3166 (B_DELWRI | B_NEEDCOMMIT)) {
3172 * NOTE: storing the bp in the bvary[] basically sets
3173 * it up for a commit operation.
3175 * We must call vfs_busy_pages() now so the commit operation
3176 * is interlocked with user modifications to memory mapped
3179 * Note: to avoid loopback deadlocks, we do not
3180 * assign b_runningbufspace.
3183 bp->b_cmd = BUF_CMD_WRITE;
3184 vfs_busy_pages(bp->b_vp, bp);
3185 info->bvary[info->bvsize] = bp;
3186 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3187 if (info->bvsize == 0 || toff < info->beg_off)
3188 info->beg_off = toff;
3189 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3190 if (info->bvsize == 0 || toff > info->end_off)
3191 info->end_off = toff;
3193 if (info->bvsize == NFS_COMMITBVECSIZ) {
3194 error = nfs_flush_docommit(info, 0);
3195 KKASSERT(info->bvsize == 0);
3203 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3213 if (info->bvsize > 0) {
3215 * Commit data on the server, as required. Note that
3216 * nfs_commit will use the vnode's cred for the commit.
3217 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3219 bytes = info->end_off - info->beg_off;
3220 if (bytes > 0x40000000)
3225 retv = nfs_commit(vp, info->beg_off,
3226 (int)bytes, info->td);
3227 if (retv == NFSERR_STALEWRITEVERF)
3228 nfs_clearcommit(vp->v_mount);
3232 * Now, either mark the blocks I/O done or mark the
3233 * blocks dirty, depending on whether the commit
3236 for (i = 0; i < info->bvsize; ++i) {
3237 bp = info->bvary[i];
3238 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3241 * Error, leave B_DELWRI intact
3243 vfs_unbusy_pages(bp);
3244 bp->b_cmd = BUF_CMD_DONE;
3248 * Success, remove B_DELWRI ( bundirty() ).
3250 * b_dirtyoff/b_dirtyend seem to be NFS
3251 * specific. We should probably move that
3252 * into bundirty(). XXX
3254 * We are faking an I/O write, we have to
3255 * start the transaction in order to
3256 * immediately biodone() it.
3259 bp->b_flags &= ~B_ERROR;
3260 bp->b_dirtyoff = bp->b_dirtyend = 0;
3261 biodone(&bp->b_bio1);
3270 * NFS advisory byte-level locks.
3271 * Currently unsupported.
3273 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3277 nfs_advlock(struct vop_advlock_args *ap)
3279 struct nfsnode *np = VTONFS(ap->a_vp);
3282 * The following kludge is to allow diskless support to work
3283 * until a real NFS lockd is implemented. Basically, just pretend
3284 * that this is a local lock.
3286 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3290 * Print out the contents of an nfsnode.
3292 * nfs_print(struct vnode *a_vp)
3295 nfs_print(struct vop_print_args *ap)
3297 struct vnode *vp = ap->a_vp;
3298 struct nfsnode *np = VTONFS(vp);
3300 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3301 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3302 if (vp->v_type == VFIFO)
3309 * nfs special file access vnode op.
3310 * Essentially just get vattr and then imitate iaccess() since the device is
3311 * local to the client.
3313 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3316 nfsspec_access(struct vop_access_args *ap)
3320 struct ucred *cred = ap->a_cred;
3321 struct vnode *vp = ap->a_vp;
3322 mode_t mode = ap->a_mode;
3328 * Disallow write attempts on filesystems mounted read-only;
3329 * unless the file is a socket, fifo, or a block or character
3330 * device resident on the filesystem.
3332 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3333 switch (vp->v_type) {
3343 * If you're the super-user,
3344 * you always get access.
3346 if (cred->cr_uid == 0)
3349 error = VOP_GETATTR(vp, vap);
3353 * Access check is based on only one of owner, group, public.
3354 * If not owner, then check group. If not a member of the
3355 * group, then check public access.
3357 if (cred->cr_uid != vap->va_uid) {
3359 gp = cred->cr_groups;
3360 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3361 if (vap->va_gid == *gp)
3367 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3372 * Read wrapper for special devices.
3374 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3375 * struct ucred *a_cred)
3378 nfsspec_read(struct vop_read_args *ap)
3380 struct nfsnode *np = VTONFS(ap->a_vp);
3386 getnanotime(&np->n_atim);
3387 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3391 * Write wrapper for special devices.
3393 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3394 * struct ucred *a_cred)
3397 nfsspec_write(struct vop_write_args *ap)
3399 struct nfsnode *np = VTONFS(ap->a_vp);
3405 getnanotime(&np->n_mtim);
3406 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3410 * Close wrapper for special devices.
3412 * Update the times on the nfsnode then do device close.
3414 * nfsspec_close(struct vnode *a_vp, int a_fflag)
3417 nfsspec_close(struct vop_close_args *ap)
3419 struct vnode *vp = ap->a_vp;
3420 struct nfsnode *np = VTONFS(vp);
3423 if (np->n_flag & (NACC | NUPD)) {
3425 if (vp->v_sysref.refcnt == 1 &&
3426 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3428 if (np->n_flag & NACC)
3429 vattr.va_atime = np->n_atim;
3430 if (np->n_flag & NUPD)
3431 vattr.va_mtime = np->n_mtim;
3432 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3435 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3439 * Read wrapper for fifos.
3441 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3442 * struct ucred *a_cred)
3445 nfsfifo_read(struct vop_read_args *ap)
3447 struct nfsnode *np = VTONFS(ap->a_vp);
3453 getnanotime(&np->n_atim);
3454 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3458 * Write wrapper for fifos.
3460 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3461 * struct ucred *a_cred)
3464 nfsfifo_write(struct vop_write_args *ap)
3466 struct nfsnode *np = VTONFS(ap->a_vp);
3472 getnanotime(&np->n_mtim);
3473 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3477 * Close wrapper for fifos.
3479 * Update the times on the nfsnode then do fifo close.
3481 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3484 nfsfifo_close(struct vop_close_args *ap)
3486 struct vnode *vp = ap->a_vp;
3487 struct nfsnode *np = VTONFS(vp);
3491 if (np->n_flag & (NACC | NUPD)) {
3493 if (np->n_flag & NACC)
3495 if (np->n_flag & NUPD)
3498 if (vp->v_sysref.refcnt == 1 &&
3499 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3501 if (np->n_flag & NACC)
3502 vattr.va_atime = np->n_atim;
3503 if (np->n_flag & NUPD)
3504 vattr.va_mtime = np->n_mtim;
3505 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3508 return (VOCALL(&fifo_vnode_vops, &ap->a_head));