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_uio(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;
881 struct nfsm_info info;
886 if ((error = vget(dvp, LK_SHARED)) != 0)
890 info.v3 = NFS_ISV3(dvp);
893 nfsstats.lookupcache_misses++;
894 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
895 ncp = ap->a_nch->ncp;
897 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
898 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
899 ERROROUT(nfsm_fhtom(&info, dvp));
900 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
901 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
902 ap->a_cred, &error));
905 * Cache negatve lookups to reduce NFS traffic, but use
906 * a fast timeout. Otherwise use a timeout of 1 tick.
907 * XXX we should add a namecache flag for no-caching
908 * to uncache the negative hit as soon as possible, but
909 * we cannot simply destroy the entry because it is used
910 * as a placeholder by the caller.
912 * The refactored nfs code will overwrite a non-zero error
913 * with 0 when we use ERROROUT(), so don't here.
916 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
917 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
929 * Success, get the file handle, do various checks, and load
930 * post-operation data from the reply packet. Theoretically
931 * we should never be looking up "." so, theoretically, we
932 * should never get the same file handle as our directory. But
933 * we check anyway. XXX
935 * Note that no timeout is set for the positive cache hit. We
936 * assume, theoretically, that ESTALE returns will be dealt with
937 * properly to handle NFS races and in anycase we cannot depend
938 * on a timeout to deal with NFS open/create/excl issues so instead
939 * of a bad hack here the rest of the NFS client code needs to do
942 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
945 if (NFS_CMPFH(np, fhp, fhsize)) {
949 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
959 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
960 NFS_LATTR_NOSHRINK));
961 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
962 NFS_LATTR_NOSHRINK));
964 ERROROUT(nfsm_loadattr(&info, nvp, NULL));
966 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
981 * 'cached' nfs directory lookup
983 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
985 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
986 * struct componentname *a_cnp)
989 nfs_lookup(struct vop_old_lookup_args *ap)
991 struct componentname *cnp = ap->a_cnp;
992 struct vnode *dvp = ap->a_dvp;
993 struct vnode **vpp = ap->a_vpp;
994 int flags = cnp->cn_flags;
996 struct nfsmount *nmp;
1000 int lockparent, wantparent, attrflag, fhsize;
1003 struct nfsm_info info;
1006 info.v3 = NFS_ISV3(dvp);
1010 * Read-only mount check and directory check.
1013 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1014 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1017 if (dvp->v_type != VDIR)
1021 * Look it up in the cache. Note that ENOENT is only returned if we
1022 * previously entered a negative hit (see later on). The additional
1023 * nfsneg_cache_timeout check causes previously cached results to
1024 * be instantly ignored if the negative caching is turned off.
1026 lockparent = flags & CNP_LOCKPARENT;
1027 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1028 nmp = VFSTONFS(dvp->v_mount);
1036 nfsstats.lookupcache_misses++;
1037 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1038 len = cnp->cn_namelen;
1039 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1040 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1041 ERROROUT(nfsm_fhtom(&info, dvp));
1042 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1043 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1044 cnp->cn_cred, &error));
1046 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1047 NFS_LATTR_NOSHRINK);
1057 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1060 * Handle RENAME case...
1062 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1063 if (NFS_CMPFH(np, fhp, fhsize)) {
1068 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1076 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1077 NFS_LATTR_NOSHRINK));
1078 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1079 NFS_LATTR_NOSHRINK));
1081 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1088 cnp->cn_flags |= CNP_PDIRUNLOCK;
1093 if (flags & CNP_ISDOTDOT) {
1095 cnp->cn_flags |= CNP_PDIRUNLOCK;
1096 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1098 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1099 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1100 return (error); /* NOTE: return error from nget */
1104 error = vn_lock(dvp, LK_EXCLUSIVE);
1109 cnp->cn_flags |= CNP_PDIRUNLOCK;
1111 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1115 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1123 cnp->cn_flags |= CNP_PDIRUNLOCK;
1128 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1129 NFS_LATTR_NOSHRINK));
1130 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1131 NFS_LATTR_NOSHRINK));
1133 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1136 /* XXX MOVE TO nfs_nremove() */
1137 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1138 cnp->cn_nameiop != NAMEI_DELETE) {
1139 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1147 if (newvp != NULLVP) {
1151 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1152 cnp->cn_nameiop == NAMEI_RENAME) &&
1156 cnp->cn_flags |= CNP_PDIRUNLOCK;
1158 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1161 error = EJUSTRETURN;
1169 * Just call nfs_bioread() to do the work.
1171 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1172 * struct ucred *a_cred)
1175 nfs_read(struct vop_read_args *ap)
1177 struct vnode *vp = ap->a_vp;
1179 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1185 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1188 nfs_readlink(struct vop_readlink_args *ap)
1190 struct vnode *vp = ap->a_vp;
1192 if (vp->v_type != VLNK)
1194 return (nfs_bioread(vp, ap->a_uio, 0));
1198 * Do a readlink rpc.
1199 * Called by nfs_doio() from below the buffer cache.
1202 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1204 int error = 0, len, attrflag;
1205 struct nfsm_info info;
1208 info.v3 = NFS_ISV3(vp);
1210 nfsstats.rpccnt[NFSPROC_READLINK]++;
1211 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1212 ERROROUT(nfsm_fhtom(&info, vp));
1213 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1214 nfs_vpcred(vp, ND_CHECK), &error));
1216 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1217 NFS_LATTR_NOSHRINK));
1220 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1221 if (len == NFS_MAXPATHLEN) {
1222 struct nfsnode *np = VTONFS(vp);
1223 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1226 ERROROUT(nfsm_mtouio(&info, uiop, len));
1237 * If bio is non-NULL and asynchronous
1240 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1243 struct nfsmount *nmp;
1244 int error = 0, len, retlen, tsiz, eof, attrflag;
1245 struct nfsm_info info;
1248 info.v3 = NFS_ISV3(vp);
1253 nmp = VFSTONFS(vp->v_mount);
1254 tsiz = uiop->uio_resid;
1255 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1258 nfsstats.rpccnt[NFSPROC_READ]++;
1259 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1260 nfsm_reqhead(&info, vp, NFSPROC_READ,
1261 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1262 ERROROUT(nfsm_fhtom(&info, vp));
1263 tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1265 txdr_hyper(uiop->uio_offset, tl);
1266 *(tl + 2) = txdr_unsigned(len);
1268 *tl++ = txdr_unsigned(uiop->uio_offset);
1269 *tl++ = txdr_unsigned(len);
1272 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1273 nfs_vpcred(vp, ND_READ), &error));
1275 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1276 NFS_LATTR_NOSHRINK));
1277 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1278 eof = fxdr_unsigned(int, *(tl + 1));
1280 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1282 NEGATIVEOUT(retlen = nfsm_strsiz(&info, nmp->nm_rsize));
1283 ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1288 if (eof || retlen == 0) {
1291 } else if (retlen < len) {
1303 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1307 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1308 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1309 int committed = NFSV3WRITE_FILESYNC;
1310 struct nfsm_info info;
1313 info.v3 = NFS_ISV3(vp);
1316 if (uiop->uio_iovcnt != 1)
1317 panic("nfs: writerpc iovcnt > 1");
1320 tsiz = uiop->uio_resid;
1321 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1324 nfsstats.rpccnt[NFSPROC_WRITE]++;
1325 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1326 nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1327 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1328 ERROROUT(nfsm_fhtom(&info, vp));
1330 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1331 txdr_hyper(uiop->uio_offset, tl);
1333 *tl++ = txdr_unsigned(len);
1334 *tl++ = txdr_unsigned(*iomode);
1335 *tl = txdr_unsigned(len);
1339 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1340 /* Set both "begin" and "current" to non-garbage. */
1341 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1342 *tl++ = x; /* "begin offset" */
1343 *tl++ = x; /* "current offset" */
1344 x = txdr_unsigned(len);
1345 *tl++ = x; /* total to this offset */
1346 *tl = x; /* size of this write */
1348 ERROROUT(nfsm_uiotom(&info, uiop, len));
1349 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1350 nfs_vpcred(vp, ND_WRITE), &error));
1353 * The write RPC returns a before and after mtime. The
1354 * nfsm_wcc_data() macro checks the before n_mtime
1355 * against the before time and stores the after time
1356 * in the nfsnode's cached vattr and n_mtime field.
1357 * The NRMODIFIED bit will be set if the before
1358 * time did not match the original mtime.
1360 wccflag = NFSV3_WCCCHK;
1361 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1363 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1364 rlen = fxdr_unsigned(int, *tl++);
1370 } else if (rlen < len) {
1371 backup = len - rlen;
1372 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1373 uiop->uio_iov->iov_len += backup;
1374 uiop->uio_offset -= backup;
1375 uiop->uio_resid += backup;
1378 commit = fxdr_unsigned(int, *tl++);
1381 * Return the lowest committment level
1382 * obtained by any of the RPCs.
1384 if (committed == NFSV3WRITE_FILESYNC)
1386 else if (committed == NFSV3WRITE_DATASYNC &&
1387 commit == NFSV3WRITE_UNSTABLE)
1389 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1390 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1392 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1393 } else if (bcmp((caddr_t)tl,
1394 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1396 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1401 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1410 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1411 committed = NFSV3WRITE_FILESYNC;
1412 *iomode = committed;
1414 uiop->uio_resid = tsiz;
1420 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1421 * mode set to specify the file type and the size field for rdev.
1424 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1427 struct nfsv2_sattr *sp;
1429 struct vnode *newvp = NULL;
1430 struct nfsnode *np = NULL;
1432 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1434 struct nfsm_info info;
1437 info.v3 = NFS_ISV3(dvp);
1439 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1440 rmajor = txdr_unsigned(vap->va_rmajor);
1441 rminor = txdr_unsigned(vap->va_rminor);
1442 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1443 rmajor = nfs_xdrneg1;
1444 rminor = nfs_xdrneg1;
1446 return (EOPNOTSUPP);
1448 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1451 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1452 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1453 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1454 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1455 ERROROUT(nfsm_fhtom(&info, dvp));
1456 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1459 tl = nfsm_build(&info, NFSX_UNSIGNED);
1460 *tl++ = vtonfsv3_type(vap->va_type);
1461 nfsm_v3attrbuild(&info, vap, FALSE);
1462 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1463 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1464 *tl++ = txdr_unsigned(vap->va_rmajor);
1465 *tl = txdr_unsigned(vap->va_rminor);
1468 sp = nfsm_build(&info, NFSX_V2SATTR);
1469 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1470 sp->sa_uid = nfs_xdrneg1;
1471 sp->sa_gid = nfs_xdrneg1;
1472 sp->sa_size = makeudev(rmajor, rminor);
1473 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1474 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1476 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1477 cnp->cn_cred, &error));
1479 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1485 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1486 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1492 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1503 VTONFS(dvp)->n_flag |= NLMODIFIED;
1505 VTONFS(dvp)->n_attrstamp = 0;
1511 * just call nfs_mknodrpc() to do the work.
1513 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1514 * struct componentname *a_cnp, struct vattr *a_vap)
1518 nfs_mknod(struct vop_old_mknod_args *ap)
1520 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1523 static u_long create_verf;
1525 * nfs file create call
1527 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1528 * struct componentname *a_cnp, struct vattr *a_vap)
1531 nfs_create(struct vop_old_create_args *ap)
1533 struct vnode *dvp = ap->a_dvp;
1534 struct vattr *vap = ap->a_vap;
1535 struct componentname *cnp = ap->a_cnp;
1536 struct nfsv2_sattr *sp;
1538 struct nfsnode *np = NULL;
1539 struct vnode *newvp = NULL;
1540 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1542 struct nfsm_info info;
1545 info.v3 = NFS_ISV3(dvp);
1548 * Oops, not for me..
1550 if (vap->va_type == VSOCK)
1551 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1553 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1556 if (vap->va_vaflags & VA_EXCLUSIVE)
1559 nfsstats.rpccnt[NFSPROC_CREATE]++;
1560 nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1561 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1562 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1563 ERROROUT(nfsm_fhtom(&info, dvp));
1564 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1567 tl = nfsm_build(&info, NFSX_UNSIGNED);
1568 if (fmode & O_EXCL) {
1569 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1570 tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1572 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1573 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1576 *tl++ = create_verf;
1577 *tl = ++create_verf;
1579 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1580 nfsm_v3attrbuild(&info, vap, FALSE);
1583 sp = nfsm_build(&info, NFSX_V2SATTR);
1584 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1585 sp->sa_uid = nfs_xdrneg1;
1586 sp->sa_gid = nfs_xdrneg1;
1588 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1589 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1591 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1592 cnp->cn_cred, &error));
1594 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1600 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1601 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1607 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1613 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1614 KKASSERT(newvp == NULL);
1618 } else if (info.v3 && (fmode & O_EXCL)) {
1620 * We are normally called with only a partially initialized
1621 * VAP. Since the NFSv3 spec says that server may use the
1622 * file attributes to store the verifier, the spec requires
1623 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1624 * in atime, but we can't really assume that all servers will
1625 * so we ensure that our SETATTR sets both atime and mtime.
1627 if (vap->va_mtime.tv_sec == VNOVAL)
1628 vfs_timestamp(&vap->va_mtime);
1629 if (vap->va_atime.tv_sec == VNOVAL)
1630 vap->va_atime = vap->va_mtime;
1631 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1635 * The new np may have enough info for access
1636 * checks, make sure rucred and wucred are
1637 * initialized for read and write rpc's.
1640 if (np->n_rucred == NULL)
1641 np->n_rucred = crhold(cnp->cn_cred);
1642 if (np->n_wucred == NULL)
1643 np->n_wucred = crhold(cnp->cn_cred);
1648 VTONFS(dvp)->n_flag |= NLMODIFIED;
1650 VTONFS(dvp)->n_attrstamp = 0;
1655 * nfs file remove call
1656 * To try and make nfs semantics closer to ufs semantics, a file that has
1657 * other processes using the vnode is renamed instead of removed and then
1658 * removed later on the last close.
1659 * - If v_sysref.refcnt > 1
1660 * If a rename is not already in the works
1661 * call nfs_sillyrename() to set it up
1665 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1666 * struct componentname *a_cnp)
1669 nfs_remove(struct vop_old_remove_args *ap)
1671 struct vnode *vp = ap->a_vp;
1672 struct vnode *dvp = ap->a_dvp;
1673 struct componentname *cnp = ap->a_cnp;
1674 struct nfsnode *np = VTONFS(vp);
1679 if (vp->v_sysref.refcnt < 1)
1680 panic("nfs_remove: bad v_sysref.refcnt");
1682 if (vp->v_type == VDIR)
1684 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1685 VOP_GETATTR(vp, &vattr) == 0 &&
1686 vattr.va_nlink > 1)) {
1688 * throw away biocache buffers, mainly to avoid
1689 * unnecessary delayed writes later.
1691 error = nfs_vinvalbuf(vp, 0, 1);
1694 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1695 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1697 * Kludge City: If the first reply to the remove rpc is lost..
1698 * the reply to the retransmitted request will be ENOENT
1699 * since the file was in fact removed
1700 * Therefore, we cheat and return success.
1702 if (error == ENOENT)
1704 } else if (!np->n_sillyrename) {
1705 error = nfs_sillyrename(dvp, vp, cnp);
1707 np->n_attrstamp = 0;
1712 * nfs file remove rpc called from nfs_inactive
1715 nfs_removeit(struct sillyrename *sp)
1717 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1722 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1725 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1726 struct ucred *cred, struct thread *td)
1728 int error = 0, wccflag = NFSV3_WCCRATTR;
1729 struct nfsm_info info;
1732 info.v3 = NFS_ISV3(dvp);
1734 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1735 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1736 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1737 ERROROUT(nfsm_fhtom(&info, dvp));
1738 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1739 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1741 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1746 VTONFS(dvp)->n_flag |= NLMODIFIED;
1748 VTONFS(dvp)->n_attrstamp = 0;
1753 * nfs file rename call
1755 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1756 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1757 * struct vnode *a_tvp, struct componentname *a_tcnp)
1760 nfs_rename(struct vop_old_rename_args *ap)
1762 struct vnode *fvp = ap->a_fvp;
1763 struct vnode *tvp = ap->a_tvp;
1764 struct vnode *fdvp = ap->a_fdvp;
1765 struct vnode *tdvp = ap->a_tdvp;
1766 struct componentname *tcnp = ap->a_tcnp;
1767 struct componentname *fcnp = ap->a_fcnp;
1770 /* Check for cross-device rename */
1771 if ((fvp->v_mount != tdvp->v_mount) ||
1772 (tvp && (fvp->v_mount != tvp->v_mount))) {
1778 * We shouldn't have to flush fvp on rename for most server-side
1779 * filesystems as the file handle should not change. Unfortunately
1780 * the inode for some filesystems (msdosfs) might be tied to the
1781 * file name or directory position so to be completely safe
1782 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1785 * We must flush tvp on rename because it might become stale on the
1786 * server after the rename.
1788 if (nfs_flush_on_rename)
1789 VOP_FSYNC(fvp, MNT_WAIT);
1791 VOP_FSYNC(tvp, MNT_WAIT);
1794 * If the tvp exists and is in use, sillyrename it before doing the
1795 * rename of the new file over it.
1797 * XXX Can't sillyrename a directory.
1799 * We do not attempt to do any namecache purges in this old API
1800 * routine. The new API compat functions have access to the actual
1801 * namecache structures and will do it for us.
1803 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1804 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1811 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1812 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1825 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1827 if (error == ENOENT)
1833 * nfs file rename rpc called from nfs_remove() above
1836 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1837 struct sillyrename *sp)
1839 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1840 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1844 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1847 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1848 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1849 struct ucred *cred, struct thread *td)
1851 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1852 struct nfsm_info info;
1855 info.v3 = NFS_ISV3(fdvp);
1857 nfsstats.rpccnt[NFSPROC_RENAME]++;
1858 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1859 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1860 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1861 ERROROUT(nfsm_fhtom(&info, fdvp));
1862 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1863 ERROROUT(nfsm_fhtom(&info, tdvp));
1864 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1865 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
1867 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
1868 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
1873 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1874 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1876 VTONFS(fdvp)->n_attrstamp = 0;
1878 VTONFS(tdvp)->n_attrstamp = 0;
1883 * nfs hard link create call
1885 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1886 * struct componentname *a_cnp)
1889 nfs_link(struct vop_old_link_args *ap)
1891 struct vnode *vp = ap->a_vp;
1892 struct vnode *tdvp = ap->a_tdvp;
1893 struct componentname *cnp = ap->a_cnp;
1894 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1895 struct nfsm_info info;
1897 if (vp->v_mount != tdvp->v_mount) {
1902 * The attribute cache may get out of sync with the server on link.
1903 * Pushing writes to the server before handle was inherited from
1904 * long long ago and it is unclear if we still need to do this.
1907 if (nfs_flush_on_hlink)
1908 VOP_FSYNC(vp, MNT_WAIT);
1911 info.v3 = NFS_ISV3(vp);
1913 nfsstats.rpccnt[NFSPROC_LINK]++;
1914 nfsm_reqhead(&info, vp, NFSPROC_LINK,
1915 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
1916 nfsm_rndup(cnp->cn_namelen));
1917 ERROROUT(nfsm_fhtom(&info, vp));
1918 ERROROUT(nfsm_fhtom(&info, tdvp));
1919 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1921 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
1922 cnp->cn_cred, &error));
1924 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1925 NFS_LATTR_NOSHRINK));
1926 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
1931 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1933 VTONFS(vp)->n_attrstamp = 0;
1935 VTONFS(tdvp)->n_attrstamp = 0;
1937 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1939 if (error == EEXIST)
1945 * nfs symbolic link create call
1947 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1948 * struct componentname *a_cnp, struct vattr *a_vap,
1952 nfs_symlink(struct vop_old_symlink_args *ap)
1954 struct vnode *dvp = ap->a_dvp;
1955 struct vattr *vap = ap->a_vap;
1956 struct componentname *cnp = ap->a_cnp;
1957 struct nfsv2_sattr *sp;
1958 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1959 struct vnode *newvp = NULL;
1960 struct nfsm_info info;
1963 info.v3 = NFS_ISV3(dvp);
1965 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1966 slen = strlen(ap->a_target);
1967 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
1968 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
1969 nfsm_rndup(cnp->cn_namelen) +
1970 nfsm_rndup(slen) + NFSX_SATTR(info.v3));
1971 ERROROUT(nfsm_fhtom(&info, dvp));
1972 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1975 nfsm_v3attrbuild(&info, vap, FALSE);
1977 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
1979 sp = nfsm_build(&info, NFSX_V2SATTR);
1980 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1981 sp->sa_uid = nfs_xdrneg1;
1982 sp->sa_gid = nfs_xdrneg1;
1983 sp->sa_size = nfs_xdrneg1;
1984 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1985 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1989 * Issue the NFS request and get the rpc response.
1991 * Only NFSv3 responses returning an error of 0 actually return
1992 * a file handle that can be converted into newvp without having
1993 * to do an extra lookup rpc.
1995 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
1996 cnp->cn_cred, &error));
1999 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2001 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2005 * out code jumps -> here, mrep is also freed.
2013 * If we get an EEXIST error, silently convert it to no-error
2014 * in case of an NFS retry.
2016 if (error == EEXIST)
2020 * If we do not have (or no longer have) an error, and we could
2021 * not extract the newvp from the response due to the request being
2022 * NFSv2 or the error being EEXIST. We have to do a lookup in order
2023 * to obtain a newvp to return.
2025 if (error == 0 && newvp == NULL) {
2026 struct nfsnode *np = NULL;
2028 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2029 cnp->cn_cred, cnp->cn_td, &np);
2039 VTONFS(dvp)->n_flag |= NLMODIFIED;
2041 VTONFS(dvp)->n_attrstamp = 0;
2048 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2049 * struct componentname *a_cnp, struct vattr *a_vap)
2052 nfs_mkdir(struct vop_old_mkdir_args *ap)
2054 struct vnode *dvp = ap->a_dvp;
2055 struct vattr *vap = ap->a_vap;
2056 struct componentname *cnp = ap->a_cnp;
2057 struct nfsv2_sattr *sp;
2058 struct nfsnode *np = NULL;
2059 struct vnode *newvp = NULL;
2061 int error = 0, wccflag = NFSV3_WCCRATTR;
2064 struct nfsm_info info;
2067 info.v3 = NFS_ISV3(dvp);
2069 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2072 len = cnp->cn_namelen;
2073 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2074 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2075 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2076 nfsm_rndup(len) + NFSX_SATTR(info.v3));
2077 ERROROUT(nfsm_fhtom(&info, dvp));
2078 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2080 nfsm_v3attrbuild(&info, vap, FALSE);
2082 sp = nfsm_build(&info, NFSX_V2SATTR);
2083 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2084 sp->sa_uid = nfs_xdrneg1;
2085 sp->sa_gid = nfs_xdrneg1;
2086 sp->sa_size = nfs_xdrneg1;
2087 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2088 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2090 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2091 cnp->cn_cred, &error));
2093 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2096 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2101 VTONFS(dvp)->n_flag |= NLMODIFIED;
2103 VTONFS(dvp)->n_attrstamp = 0;
2105 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2106 * if we can succeed in looking up the directory.
2108 if (error == EEXIST || (!error && !gotvp)) {
2113 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2117 if (newvp->v_type != VDIR)
2130 * nfs remove directory call
2132 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2133 * struct componentname *a_cnp)
2136 nfs_rmdir(struct vop_old_rmdir_args *ap)
2138 struct vnode *vp = ap->a_vp;
2139 struct vnode *dvp = ap->a_dvp;
2140 struct componentname *cnp = ap->a_cnp;
2141 int error = 0, wccflag = NFSV3_WCCRATTR;
2142 struct nfsm_info info;
2145 info.v3 = NFS_ISV3(dvp);
2149 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2150 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2151 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2152 nfsm_rndup(cnp->cn_namelen));
2153 ERROROUT(nfsm_fhtom(&info, dvp));
2154 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2156 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2157 cnp->cn_cred, &error));
2159 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2164 VTONFS(dvp)->n_flag |= NLMODIFIED;
2166 VTONFS(dvp)->n_attrstamp = 0;
2168 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2170 if (error == ENOENT)
2178 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2181 nfs_readdir(struct vop_readdir_args *ap)
2183 struct vnode *vp = ap->a_vp;
2184 struct nfsnode *np = VTONFS(vp);
2185 struct uio *uio = ap->a_uio;
2189 if (vp->v_type != VDIR)
2192 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2196 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2197 * and then check that is still valid, or if this is an NQNFS mount
2198 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2199 * VOP_GETATTR() does not necessarily go to the wire.
2201 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2202 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2203 if (VOP_GETATTR(vp, &vattr) == 0 &&
2204 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2206 nfsstats.direofcache_hits++;
2212 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2213 * own cache coherency checks so we do not have to.
2215 tresid = uio->uio_resid;
2216 error = nfs_bioread(vp, uio, 0);
2218 if (!error && uio->uio_resid == tresid)
2219 nfsstats.direofcache_misses++;
2226 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2228 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2229 * offset/block and converts the nfs formatted directory entries for userland
2230 * consumption as well as deals with offsets into the middle of blocks.
2231 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2232 * be block-bounded. It must convert to cookies for the actual RPC.
2235 nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2238 struct nfs_dirent *dp = NULL;
2243 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2244 struct nfsnode *dnp = VTONFS(vp);
2246 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2248 struct nfsm_info info;
2251 info.v3 = NFS_ISV3(vp);
2254 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2255 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2256 panic("nfs readdirrpc bad uio");
2260 * If there is no cookie, assume directory was stale.
2262 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2266 return (NFSERR_BAD_COOKIE);
2268 * Loop around doing readdir rpc's of size nm_readdirsize
2269 * truncated to a multiple of DIRBLKSIZ.
2270 * The stopping criteria is EOF or buffer full.
2272 while (more_dirs && bigenough) {
2273 nfsstats.rpccnt[NFSPROC_READDIR]++;
2274 nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2275 NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2276 ERROROUT(nfsm_fhtom(&info, vp));
2278 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2279 *tl++ = cookie.nfsuquad[0];
2280 *tl++ = cookie.nfsuquad[1];
2281 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2282 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2284 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2285 *tl++ = cookie.nfsuquad[0];
2287 *tl = txdr_unsigned(nmp->nm_readdirsize);
2288 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2290 nfs_vpcred(vp, ND_READ), &error));
2292 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2293 NFS_LATTR_NOSHRINK));
2294 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2295 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2296 dnp->n_cookieverf.nfsuquad[1] = *tl;
2298 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2299 more_dirs = fxdr_unsigned(int, *tl);
2301 /* loop thru the dir entries, converting them to std form */
2302 while (more_dirs && bigenough) {
2304 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2305 fileno = fxdr_hyper(tl);
2306 len = fxdr_unsigned(int, *(tl + 2));
2308 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2309 fileno = fxdr_unsigned(u_quad_t, *tl++);
2310 len = fxdr_unsigned(int, *tl);
2312 if (len <= 0 || len > NFS_MAXNAMLEN) {
2320 * len is the number of bytes in the path element
2321 * name, not including the \0 termination.
2323 * tlen is the number of bytes w have to reserve for
2324 * the path element name.
2326 tlen = nfsm_rndup(len);
2328 tlen += 4; /* To ensure null termination */
2331 * If the entry would cross a DIRBLKSIZ boundary,
2332 * extend the previous nfs_dirent to cover the
2335 left = DIRBLKSIZ - blksiz;
2336 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2337 dp->nfs_reclen += left;
2338 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2339 uiop->uio_iov->iov_len -= left;
2340 uiop->uio_offset += left;
2341 uiop->uio_resid -= left;
2344 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2347 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2348 dp->nfs_ino = fileno;
2349 dp->nfs_namlen = len;
2350 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2351 dp->nfs_type = DT_UNKNOWN;
2352 blksiz += dp->nfs_reclen;
2353 if (blksiz == DIRBLKSIZ)
2355 uiop->uio_offset += sizeof(struct nfs_dirent);
2356 uiop->uio_resid -= sizeof(struct nfs_dirent);
2357 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2358 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2359 ERROROUT(nfsm_mtouio(&info, uiop, len));
2362 * The uiop has advanced by nfs_dirent + len
2363 * but really needs to advance by
2366 cp = uiop->uio_iov->iov_base;
2368 *cp = '\0'; /* null terminate */
2369 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2370 uiop->uio_iov->iov_len -= tlen;
2371 uiop->uio_offset += tlen;
2372 uiop->uio_resid -= tlen;
2375 * NFS strings must be rounded up (nfsm_myouio
2376 * handled that in the bigenough case).
2378 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2381 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2383 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2387 * If we were able to accomodate the last entry,
2388 * get the cookie for the next one. Otherwise
2389 * hold-over the cookie for the one we were not
2390 * able to accomodate.
2393 cookie.nfsuquad[0] = *tl++;
2395 cookie.nfsuquad[1] = *tl++;
2396 } else if (info.v3) {
2401 more_dirs = fxdr_unsigned(int, *tl);
2404 * If at end of rpc data, get the eof boolean
2407 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2408 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2414 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2415 * by increasing d_reclen for the last record.
2418 left = DIRBLKSIZ - blksiz;
2419 dp->nfs_reclen += left;
2420 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2421 uiop->uio_iov->iov_len -= left;
2422 uiop->uio_offset += left;
2423 uiop->uio_resid -= left;
2428 * We hit the end of the directory, update direofoffset.
2430 dnp->n_direofoffset = uiop->uio_offset;
2433 * There is more to go, insert the link cookie so the
2434 * next block can be read.
2436 if (uiop->uio_resid > 0)
2437 kprintf("EEK! readdirrpc resid > 0\n");
2438 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2446 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2449 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2452 struct nfs_dirent *dp;
2454 struct vnode *newvp;
2456 caddr_t dpossav1, dpossav2;
2458 struct mbuf *mdsav1, *mdsav2;
2460 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2461 struct nfsnode *dnp = VTONFS(vp), *np;
2464 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2465 int attrflag, fhsize;
2466 struct nchandle nch;
2467 struct nchandle dnch;
2468 struct nlcomponent nlc;
2469 struct nfsm_info info;
2478 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2479 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2480 panic("nfs readdirplusrpc bad uio");
2483 * Obtain the namecache record for the directory so we have something
2484 * to use as a basis for creating the entries. This function will
2485 * return a held (but not locked) ncp. The ncp may be disconnected
2486 * from the tree and cannot be used for upward traversals, and the
2487 * ncp may be unnamed. Note that other unrelated operations may
2488 * cause the ncp to be named at any time.
2490 cache_fromdvp(vp, NULL, 0, &dnch);
2491 bzero(&nlc, sizeof(nlc));
2495 * If there is no cookie, assume directory was stale.
2497 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2501 return (NFSERR_BAD_COOKIE);
2503 * Loop around doing readdir rpc's of size nm_readdirsize
2504 * truncated to a multiple of DIRBLKSIZ.
2505 * The stopping criteria is EOF or buffer full.
2507 while (more_dirs && bigenough) {
2508 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2509 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2510 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2511 ERROROUT(nfsm_fhtom(&info, vp));
2512 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2513 *tl++ = cookie.nfsuquad[0];
2514 *tl++ = cookie.nfsuquad[1];
2515 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2516 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2517 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2518 *tl = txdr_unsigned(nmp->nm_rsize);
2519 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2521 nfs_vpcred(vp, ND_READ), &error));
2522 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2523 NFS_LATTR_NOSHRINK));
2524 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2525 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2526 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2527 more_dirs = fxdr_unsigned(int, *tl);
2529 /* loop thru the dir entries, doctoring them to 4bsd form */
2530 while (more_dirs && bigenough) {
2531 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2532 fileno = fxdr_hyper(tl);
2533 len = fxdr_unsigned(int, *(tl + 2));
2534 if (len <= 0 || len > NFS_MAXNAMLEN) {
2540 tlen = nfsm_rndup(len);
2542 tlen += 4; /* To ensure null termination*/
2543 left = DIRBLKSIZ - blksiz;
2544 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2545 dp->nfs_reclen += left;
2546 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2547 uiop->uio_iov->iov_len -= left;
2548 uiop->uio_offset += left;
2549 uiop->uio_resid -= left;
2552 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2555 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2556 dp->nfs_ino = fileno;
2557 dp->nfs_namlen = len;
2558 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2559 dp->nfs_type = DT_UNKNOWN;
2560 blksiz += dp->nfs_reclen;
2561 if (blksiz == DIRBLKSIZ)
2563 uiop->uio_offset += sizeof(struct nfs_dirent);
2564 uiop->uio_resid -= sizeof(struct nfs_dirent);
2565 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2566 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2567 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2568 nlc.nlc_namelen = len;
2569 ERROROUT(nfsm_mtouio(&info, uiop, len));
2570 cp = uiop->uio_iov->iov_base;
2573 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2574 uiop->uio_iov->iov_len -= tlen;
2575 uiop->uio_offset += tlen;
2576 uiop->uio_resid -= tlen;
2578 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2580 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2582 cookie.nfsuquad[0] = *tl++;
2583 cookie.nfsuquad[1] = *tl++;
2588 * Since the attributes are before the file handle
2589 * (sigh), we must skip over the attributes and then
2590 * come back and get them.
2592 attrflag = fxdr_unsigned(int, *tl);
2594 dpossav1 = info.dpos;
2596 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2597 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2598 doit = fxdr_unsigned(int, *tl);
2600 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2601 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2606 error = nfs_nget(vp->v_mount, fhp,
2614 if (doit && bigenough) {
2615 dpossav2 = info.dpos;
2616 info.dpos = dpossav1;
2619 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2620 info.dpos = dpossav2;
2623 IFTODT(VTTOIF(np->n_vattr.va_type));
2625 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2626 nlc.nlc_namelen, nlc.nlc_namelen,
2628 nch = cache_nlookup(&dnch, &nlc);
2629 cache_setunresolved(&nch);
2630 nfs_cache_setvp(&nch, newvp,
2631 nfspos_cache_timeout);
2634 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER"
2636 nlc.nlc_namelen, nlc.nlc_namelen,
2641 /* Just skip over the file handle */
2642 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2643 i = fxdr_unsigned(int, *tl);
2644 ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2646 if (newvp != NULLVP) {
2653 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2654 more_dirs = fxdr_unsigned(int, *tl);
2657 * If at end of rpc data, get the eof boolean
2660 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2661 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2667 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2668 * by increasing d_reclen for the last record.
2671 left = DIRBLKSIZ - blksiz;
2672 dp->nfs_reclen += left;
2673 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2674 uiop->uio_iov->iov_len -= left;
2675 uiop->uio_offset += left;
2676 uiop->uio_resid -= left;
2680 * We are now either at the end of the directory or have filled the
2684 dnp->n_direofoffset = uiop->uio_offset;
2686 if (uiop->uio_resid > 0)
2687 kprintf("EEK! readdirplusrpc resid > 0\n");
2688 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2692 if (newvp != NULLVP) {
2705 * Silly rename. To make the NFS filesystem that is stateless look a little
2706 * more like the "ufs" a remove of an active vnode is translated to a rename
2707 * to a funny looking filename that is removed by nfs_inactive on the
2708 * nfsnode. There is the potential for another process on a different client
2709 * to create the same funny name between the nfs_lookitup() fails and the
2710 * nfs_rename() completes, but...
2713 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2715 struct sillyrename *sp;
2720 * We previously purged dvp instead of vp. I don't know why, it
2721 * completely destroys performance. We can't do it anyway with the
2722 * new VFS API since we would be breaking the namecache topology.
2724 cache_purge(vp); /* XXX */
2727 if (vp->v_type == VDIR)
2728 panic("nfs: sillyrename dir");
2730 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2731 M_NFSREQ, M_WAITOK);
2732 sp->s_cred = crdup(cnp->cn_cred);
2736 /* Fudge together a funny name */
2737 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2738 (int)(intptr_t)cnp->cn_td);
2740 /* Try lookitups until we get one that isn't there */
2741 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2742 cnp->cn_td, NULL) == 0) {
2744 if (sp->s_name[4] > 'z') {
2749 error = nfs_renameit(dvp, cnp, sp);
2752 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2754 np->n_sillyrename = sp;
2759 kfree((caddr_t)sp, M_NFSREQ);
2764 * Look up a file name and optionally either update the file handle or
2765 * allocate an nfsnode, depending on the value of npp.
2766 * npp == NULL --> just do the lookup
2767 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2769 * *npp != NULL --> update the file handle in the vnode
2772 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2773 struct thread *td, struct nfsnode **npp)
2775 struct vnode *newvp = NULL;
2776 struct nfsnode *np, *dnp = VTONFS(dvp);
2777 int error = 0, fhlen, attrflag;
2779 struct nfsm_info info;
2782 info.v3 = NFS_ISV3(dvp);
2784 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2785 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2786 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2787 ERROROUT(nfsm_fhtom(&info, dvp));
2788 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2789 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2790 if (npp && !error) {
2791 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2794 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2795 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2796 np->n_fhp = &np->n_fh;
2797 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2798 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2799 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2800 np->n_fhsize = fhlen;
2802 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2806 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2815 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2816 NFS_LATTR_NOSHRINK));
2817 if (!attrflag && *npp == NULL) {
2827 ERROROUT(error = nfsm_loadattr(&info, newvp, NULL));
2833 if (npp && *npp == NULL) {
2848 * Nfs Version 3 commit rpc
2851 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2853 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2854 int error = 0, wccflag = NFSV3_WCCRATTR;
2855 struct nfsm_info info;
2861 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2863 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2864 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
2865 ERROROUT(nfsm_fhtom(&info, vp));
2866 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
2867 txdr_hyper(offset, tl);
2869 *tl = txdr_unsigned(cnt);
2870 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
2871 nfs_vpcred(vp, ND_WRITE), &error));
2872 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
2874 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
2875 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2876 NFSX_V3WRITEVERF)) {
2877 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2879 error = NFSERR_STALEWRITEVERF;
2890 * - make nfs_bmap() essentially a no-op that does no translation
2891 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2892 * (Maybe I could use the process's page mapping, but I was concerned that
2893 * Kernel Write might not be enabled and also figured copyout() would do
2894 * a lot more work than bcopy() and also it currently happens in the
2895 * context of the swapper process (2).
2897 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
2898 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2901 nfs_bmap(struct vop_bmap_args *ap)
2903 if (ap->a_doffsetp != NULL)
2904 *ap->a_doffsetp = ap->a_loffset;
2905 if (ap->a_runp != NULL)
2907 if (ap->a_runb != NULL)
2916 nfs_strategy(struct vop_strategy_args *ap)
2918 struct bio *bio = ap->a_bio;
2920 struct buf *bp = bio->bio_buf;
2923 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2924 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
2925 KASSERT(BUF_REFCNT(bp) > 0,
2926 ("nfs_strategy: buffer %p not locked", bp));
2928 if (bio->bio_flags & BIO_SYNC)
2929 td = curthread; /* XXX */
2934 * We probably don't need to push an nbio any more since no
2935 * block conversion is required due to the use of 64 bit byte
2936 * offsets, but do it anyway.
2938 nbio = push_bio(bio);
2939 nbio->bio_offset = bio->bio_offset;
2942 * If the op is asynchronous and an i/o daemon is waiting
2943 * queue the request, wake it up and wait for completion
2944 * otherwise just do it ourselves.
2946 if (bio->bio_flags & BIO_SYNC) {
2947 nfs_startio(ap->a_vp, nbio, td);
2949 nfs_asyncio(ap->a_vp, nbio);
2957 * NB Currently unsupported.
2959 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
2963 nfs_mmap(struct vop_mmap_args *ap)
2969 * fsync vnode op. Just call nfs_flush() with commit == 1.
2971 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
2975 nfs_fsync(struct vop_fsync_args *ap)
2977 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
2981 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2982 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2983 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2984 * set the buffer contains data that has already been written to the server
2985 * and which now needs a commit RPC.
2987 * If commit is 0 we only take one pass and only flush buffers containing new
2990 * If commit is 1 we take two passes, issuing a commit RPC in the second
2993 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
2994 * to completely flush all pending data.
2996 * Note that the RB_SCAN code properly handles the case where the
2997 * callback might block and directly or indirectly (another thread) cause
2998 * the RB tree to change.
3001 #ifndef NFS_COMMITBVECSIZ
3002 #define NFS_COMMITBVECSIZ 16
3005 struct nfs_flush_info {
3006 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3013 struct buf *bvary[NFS_COMMITBVECSIZ];
3019 static int nfs_flush_bp(struct buf *bp, void *data);
3020 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3023 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3025 struct nfsnode *np = VTONFS(vp);
3026 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3027 struct nfs_flush_info info;
3031 bzero(&info, sizeof(info));
3034 info.waitfor = waitfor;
3035 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3037 lwkt_gettoken(&vlock, &vp->v_token);
3043 info.mode = NFI_FLUSHNEW;
3044 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3045 nfs_flush_bp, &info);
3048 * Take a second pass if committing and no error occured.
3049 * Clean up any left over collection (whether an error
3052 if (commit && error == 0) {
3053 info.mode = NFI_COMMIT;
3054 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3055 nfs_flush_bp, &info);
3057 error = nfs_flush_docommit(&info, error);
3061 * Wait for pending I/O to complete before checking whether
3062 * any further dirty buffers exist.
3064 while (waitfor == MNT_WAIT &&
3065 bio_track_active(&vp->v_track_write)) {
3066 error = bio_track_wait(&vp->v_track_write,
3067 info.slpflag, info.slptimeo);
3070 * We have to be able to break out if this
3071 * is an 'intr' mount.
3073 if (nfs_sigintr(nmp, NULL, td)) {
3079 * Since we do not process pending signals,
3080 * once we get a PCATCH our tsleep() will no
3081 * longer sleep, switch to a fixed timeout
3084 if (info.slpflag == PCATCH) {
3086 info.slptimeo = 2 * hz;
3093 * Loop if we are flushing synchronous as well as committing,
3094 * and dirty buffers are still present. Otherwise we might livelock.
3096 } while (waitfor == MNT_WAIT && commit &&
3097 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3100 * The callbacks have to return a negative error to terminate the
3107 * Deal with any error collection
3109 if (np->n_flag & NWRITEERR) {
3110 error = np->n_error;
3111 np->n_flag &= ~NWRITEERR;
3113 lwkt_reltoken(&vlock);
3119 nfs_flush_bp(struct buf *bp, void *data)
3121 struct nfs_flush_info *info = data;
3127 switch(info->mode) {
3129 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3130 if (error && info->loops && info->waitfor == MNT_WAIT) {
3131 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3133 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3134 if (info->slpflag & PCATCH)
3135 lkflags |= LK_PCATCH;
3136 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3142 * Ignore locking errors
3150 * The buffer may have changed out from under us, even if
3151 * we did not block (MPSAFE). Check again now that it is
3154 if (bp->b_vp == info->vp &&
3155 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3164 * Only process buffers in need of a commit which we can
3165 * immediately lock. This may prevent a buffer from being
3166 * committed, but the normal flush loop will block on the
3167 * same buffer so we shouldn't get into an endless loop.
3169 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3170 (B_DELWRI | B_NEEDCOMMIT)) {
3173 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3177 * We must recheck after successfully locking the buffer.
3179 if (bp->b_vp != info->vp ||
3180 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3181 (B_DELWRI | B_NEEDCOMMIT)) {
3187 * NOTE: storing the bp in the bvary[] basically sets
3188 * it up for a commit operation.
3190 * We must call vfs_busy_pages() now so the commit operation
3191 * is interlocked with user modifications to memory mapped
3194 * Note: to avoid loopback deadlocks, we do not
3195 * assign b_runningbufspace.
3198 bp->b_cmd = BUF_CMD_WRITE;
3199 vfs_busy_pages(bp->b_vp, bp);
3200 info->bvary[info->bvsize] = bp;
3201 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3202 if (info->bvsize == 0 || toff < info->beg_off)
3203 info->beg_off = toff;
3204 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3205 if (info->bvsize == 0 || toff > info->end_off)
3206 info->end_off = toff;
3208 if (info->bvsize == NFS_COMMITBVECSIZ) {
3209 error = nfs_flush_docommit(info, 0);
3210 KKASSERT(info->bvsize == 0);
3218 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3228 if (info->bvsize > 0) {
3230 * Commit data on the server, as required. Note that
3231 * nfs_commit will use the vnode's cred for the commit.
3232 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3234 bytes = info->end_off - info->beg_off;
3235 if (bytes > 0x40000000)
3240 retv = nfs_commit(vp, info->beg_off,
3241 (int)bytes, info->td);
3242 if (retv == NFSERR_STALEWRITEVERF)
3243 nfs_clearcommit(vp->v_mount);
3247 * Now, either mark the blocks I/O done or mark the
3248 * blocks dirty, depending on whether the commit
3251 for (i = 0; i < info->bvsize; ++i) {
3252 bp = info->bvary[i];
3253 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3256 * Error, leave B_DELWRI intact
3258 vfs_unbusy_pages(bp);
3259 bp->b_cmd = BUF_CMD_DONE;
3263 * Success, remove B_DELWRI ( bundirty() ).
3265 * b_dirtyoff/b_dirtyend seem to be NFS
3266 * specific. We should probably move that
3267 * into bundirty(). XXX
3269 * We are faking an I/O write, we have to
3270 * start the transaction in order to
3271 * immediately biodone() it.
3274 bp->b_flags &= ~B_ERROR;
3275 bp->b_dirtyoff = bp->b_dirtyend = 0;
3276 biodone(&bp->b_bio1);
3285 * NFS advisory byte-level locks.
3286 * Currently unsupported.
3288 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3292 nfs_advlock(struct vop_advlock_args *ap)
3294 struct nfsnode *np = VTONFS(ap->a_vp);
3297 * The following kludge is to allow diskless support to work
3298 * until a real NFS lockd is implemented. Basically, just pretend
3299 * that this is a local lock.
3301 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3305 * Print out the contents of an nfsnode.
3307 * nfs_print(struct vnode *a_vp)
3310 nfs_print(struct vop_print_args *ap)
3312 struct vnode *vp = ap->a_vp;
3313 struct nfsnode *np = VTONFS(vp);
3315 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3316 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3317 if (vp->v_type == VFIFO)
3324 * nfs special file access vnode op.
3325 * Essentially just get vattr and then imitate iaccess() since the device is
3326 * local to the client.
3328 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3331 nfsspec_access(struct vop_access_args *ap)
3335 struct ucred *cred = ap->a_cred;
3336 struct vnode *vp = ap->a_vp;
3337 mode_t mode = ap->a_mode;
3343 * Disallow write attempts on filesystems mounted read-only;
3344 * unless the file is a socket, fifo, or a block or character
3345 * device resident on the filesystem.
3347 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3348 switch (vp->v_type) {
3358 * If you're the super-user,
3359 * you always get access.
3361 if (cred->cr_uid == 0)
3364 error = VOP_GETATTR(vp, vap);
3368 * Access check is based on only one of owner, group, public.
3369 * If not owner, then check group. If not a member of the
3370 * group, then check public access.
3372 if (cred->cr_uid != vap->va_uid) {
3374 gp = cred->cr_groups;
3375 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3376 if (vap->va_gid == *gp)
3382 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3387 * Read wrapper for special devices.
3389 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3390 * struct ucred *a_cred)
3393 nfsspec_read(struct vop_read_args *ap)
3395 struct nfsnode *np = VTONFS(ap->a_vp);
3401 getnanotime(&np->n_atim);
3402 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3406 * Write wrapper for special devices.
3408 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3409 * struct ucred *a_cred)
3412 nfsspec_write(struct vop_write_args *ap)
3414 struct nfsnode *np = VTONFS(ap->a_vp);
3420 getnanotime(&np->n_mtim);
3421 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3425 * Close wrapper for special devices.
3427 * Update the times on the nfsnode then do device close.
3429 * nfsspec_close(struct vnode *a_vp, int a_fflag)
3432 nfsspec_close(struct vop_close_args *ap)
3434 struct vnode *vp = ap->a_vp;
3435 struct nfsnode *np = VTONFS(vp);
3438 if (np->n_flag & (NACC | NUPD)) {
3440 if (vp->v_sysref.refcnt == 1 &&
3441 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3443 if (np->n_flag & NACC)
3444 vattr.va_atime = np->n_atim;
3445 if (np->n_flag & NUPD)
3446 vattr.va_mtime = np->n_mtim;
3447 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3450 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3454 * Read wrapper for fifos.
3456 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3457 * struct ucred *a_cred)
3460 nfsfifo_read(struct vop_read_args *ap)
3462 struct nfsnode *np = VTONFS(ap->a_vp);
3468 getnanotime(&np->n_atim);
3469 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3473 * Write wrapper for fifos.
3475 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3476 * struct ucred *a_cred)
3479 nfsfifo_write(struct vop_write_args *ap)
3481 struct nfsnode *np = VTONFS(ap->a_vp);
3487 getnanotime(&np->n_mtim);
3488 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3492 * Close wrapper for fifos.
3494 * Update the times on the nfsnode then do fifo close.
3496 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3499 nfsfifo_close(struct vop_close_args *ap)
3501 struct vnode *vp = ap->a_vp;
3502 struct nfsnode *np = VTONFS(vp);
3506 if (np->n_flag & (NACC | NUPD)) {
3508 if (np->n_flag & NACC)
3510 if (np->n_flag & NUPD)
3513 if (vp->v_sysref.refcnt == 1 &&
3514 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3516 if (np->n_flag & NACC)
3517 vattr.va_atime = np->n_atim;
3518 if (np->n_flag & NUPD)
3519 vattr.va_mtime = np->n_mtim;
3520 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3523 return (VOCALL(&fifo_vnode_vops, &ap->a_head));