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 1.71 2007/05/09 00:53:35 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);
219 extern u_int32_t nfs_true, nfs_false;
220 extern u_int32_t nfs_xdrneg1;
221 extern struct nfsstats nfsstats;
222 extern nfstype nfsv3_type[9];
223 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON];
224 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
225 int nfs_numasync = 0;
227 SYSCTL_DECL(_vfs_nfs);
229 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
230 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
231 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
233 static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
235 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache 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)
257 int error = 0, attrflag;
259 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
260 caddr_t bpos, dpos, cp2;
264 struct nfsnode *np = VTONFS(vp);
266 nfsstats.rpccnt[NFSPROC_ACCESS]++;
267 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
269 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
270 *tl = txdr_unsigned(wmode);
271 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
272 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
274 nfsm_dissect(tl, u_int32_t *, 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;
286 * nfs access vnode op.
287 * For nfs version 2, just return ok. File accesses may fail later.
288 * For nfs version 3, use the access rpc to check accessibility. If file modes
289 * are changed on the server, accesses might still fail later.
291 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
292 * struct thread *a_td)
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 int v3 = NFS_ISV3(vp);
302 struct nfsnode *np = VTONFS(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)
450 nfs_open(struct vop_open_args *ap)
452 struct vnode *vp = ap->a_vp;
453 struct nfsnode *np = VTONFS(vp);
457 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
459 kprintf("open eacces vtyp=%d\n",vp->v_type);
465 * Clear the attribute cache only if opening with write access. It
466 * is unclear if we should do this at all here, but we certainly
467 * should not clear the cache unconditionally simply because a file
470 if (ap->a_mode & FWRITE)
474 * For normal NFS, reconcile changes made locally verses
475 * changes made remotely. Note that VOP_GETATTR only goes
476 * to the wire if the cached attribute has timed out or been
479 * If local modifications have been made clear the attribute
480 * cache to force an attribute and modified time check. If
481 * GETATTR detects that the file has been changed by someone
482 * other then us it will set NRMODIFIED.
484 * If we are opening a directory and local changes have been
485 * made we have to invalidate the cache in order to ensure
486 * that we get the most up-to-date information from the
489 if (np->n_flag & NLMODIFIED) {
491 if (vp->v_type == VDIR) {
492 error = nfs_vinvalbuf(vp, V_SAVE, 1);
498 error = VOP_GETATTR(vp, &vattr);
501 if (np->n_flag & NRMODIFIED) {
502 if (vp->v_type == VDIR)
504 error = nfs_vinvalbuf(vp, V_SAVE, 1);
507 np->n_flag &= ~NRMODIFIED;
510 return (vop_stdopen(ap));
515 * What an NFS client should do upon close after writing is a debatable issue.
516 * Most NFS clients push delayed writes to the server upon close, basically for
518 * 1 - So that any write errors may be reported back to the client process
519 * doing the close system call. By far the two most likely errors are
520 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
521 * 2 - To put a worst case upper bound on cache inconsistency between
522 * multiple clients for the file.
523 * There is also a consistency problem for Version 2 of the protocol w.r.t.
524 * not being able to tell if other clients are writing a file concurrently,
525 * since there is no way of knowing if the changed modify time in the reply
526 * is only due to the write for this client.
527 * (NFS Version 3 provides weak cache consistency data in the reply that
528 * should be sufficient to detect and handle this case.)
530 * The current code does the following:
531 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
532 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
533 * or commit them (this satisfies 1 and 2 except for the
534 * case where the server crashes after this close but
535 * before the commit RPC, which is felt to be "good
536 * enough". Changing the last argument to nfs_flush() to
537 * a 1 would force a commit operation, if it is felt a
538 * commit is necessary now.
539 * for NQNFS - do nothing now, since 2 is dealt with via leases and
540 * 1 should be dealt with via an fsync() system call for
541 * cases where write errors are important.
543 * nfs_close(struct vnode *a_vp, int a_fflag,
544 * struct ucred *a_cred, struct thread *a_td)
548 nfs_close(struct vop_close_args *ap)
550 struct vnode *vp = ap->a_vp;
551 struct nfsnode *np = VTONFS(vp);
553 thread_t td = curthread;
555 if (vp->v_type == VREG) {
556 if (np->n_flag & NLMODIFIED) {
559 * Under NFSv3 we have dirty buffers to dispose of. We
560 * must flush them to the NFS server. We have the option
561 * of waiting all the way through the commit rpc or just
562 * waiting for the initial write. The default is to only
563 * wait through the initial write so the data is in the
564 * server's cache, which is roughly similar to the state
565 * a standard disk subsystem leaves the file in on close().
567 * We cannot clear the NLMODIFIED bit in np->n_flag due to
568 * potential races with other processes, and certainly
569 * cannot clear it if we don't commit.
571 int cm = nfsv3_commit_on_close ? 1 : 0;
572 error = nfs_flush(vp, MNT_WAIT, td, cm);
573 /* np->n_flag &= ~NLMODIFIED; */
575 error = nfs_vinvalbuf(vp, V_SAVE, 1);
579 if (np->n_flag & NWRITEERR) {
580 np->n_flag &= ~NWRITEERR;
589 * nfs getattr call from vfs.
591 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred,
592 * struct thread *a_td)
595 nfs_getattr(struct vop_getattr_args *ap)
597 struct vnode *vp = ap->a_vp;
598 struct nfsnode *np = VTONFS(vp);
604 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
605 int v3 = NFS_ISV3(vp);
606 thread_t td = curthread;
609 * Update local times for special files.
611 if (np->n_flag & (NACC | NUPD))
614 * First look in the cache.
616 if (nfs_getattrcache(vp, ap->a_vap) == 0)
619 if (v3 && nfsaccess_cache_timeout > 0) {
620 nfsstats.accesscache_misses++;
621 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
622 if (nfs_getattrcache(vp, ap->a_vap) == 0)
626 nfsstats.rpccnt[NFSPROC_GETATTR]++;
627 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
629 nfsm_request(vp, NFSPROC_GETATTR, td, nfs_vpcred(vp, ND_CHECK));
631 nfsm_loadattr(vp, ap->a_vap);
641 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
644 nfs_setattr(struct vop_setattr_args *ap)
646 struct vnode *vp = ap->a_vp;
647 struct nfsnode *np = VTONFS(vp);
648 struct vattr *vap = ap->a_vap;
651 thread_t td = curthread;
658 * Setting of flags is not supported.
660 if (vap->va_flags != VNOVAL)
664 * Disallow write attempts if the filesystem is mounted read-only.
666 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
667 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
668 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
669 (vp->v_mount->mnt_flag & MNT_RDONLY))
671 if (vap->va_size != VNOVAL) {
672 switch (vp->v_type) {
679 if (vap->va_mtime.tv_sec == VNOVAL &&
680 vap->va_atime.tv_sec == VNOVAL &&
681 vap->va_mode == (mode_t)VNOVAL &&
682 vap->va_uid == (uid_t)VNOVAL &&
683 vap->va_gid == (gid_t)VNOVAL)
685 vap->va_size = VNOVAL;
689 * Disallow write attempts if the filesystem is
692 if (vp->v_mount->mnt_flag & MNT_RDONLY)
696 * This is nasty. The RPCs we send to flush pending
697 * data often return attribute information which is
698 * cached via a callback to nfs_loadattrcache(), which
699 * has the effect of changing our notion of the file
700 * size. Due to flushed appends and other operations
701 * the file size can be set to virtually anything,
702 * including values that do not match either the old
703 * or intended file size.
705 * When this condition is detected we must loop to
706 * try the operation again. Hopefully no more
707 * flushing is required on the loop so it works the
708 * second time around. THIS CASE ALMOST ALWAYS
713 error = nfs_meta_setsize(vp, td, vap->va_size);
715 if (np->n_flag & NLMODIFIED) {
716 if (vap->va_size == 0)
717 error = nfs_vinvalbuf(vp, 0, 1);
719 error = nfs_vinvalbuf(vp, V_SAVE, 1);
722 * note: this loop case almost always happens at
723 * least once per truncation.
725 if (error == 0 && np->n_size != vap->va_size)
727 np->n_vattr.va_size = vap->va_size;
730 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
731 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) &&
732 vp->v_type == VREG &&
733 (error = nfs_vinvalbuf(vp, V_SAVE, 1)) == EINTR
737 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
740 * Sanity check if a truncation was issued. This should only occur
741 * if multiple processes are racing on the same file.
743 if (error == 0 && vap->va_size != VNOVAL &&
744 np->n_size != vap->va_size) {
745 kprintf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size);
748 if (error && vap->va_size != VNOVAL) {
749 np->n_size = np->n_vattr.va_size = tsize;
750 vnode_pager_setsize(vp, np->n_size);
756 * Do an nfs setattr rpc.
759 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
760 struct ucred *cred, struct thread *td)
762 struct nfsv2_sattr *sp;
763 struct nfsnode *np = VTONFS(vp);
766 caddr_t bpos, dpos, cp2;
768 int error = 0, wccflag = NFSV3_WCCRATTR;
769 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
770 int v3 = NFS_ISV3(vp);
772 nfsstats.rpccnt[NFSPROC_SETATTR]++;
773 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
776 nfsm_v3attrbuild(vap, TRUE);
777 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
780 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
781 if (vap->va_mode == (mode_t)VNOVAL)
782 sp->sa_mode = nfs_xdrneg1;
784 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
785 if (vap->va_uid == (uid_t)VNOVAL)
786 sp->sa_uid = nfs_xdrneg1;
788 sp->sa_uid = txdr_unsigned(vap->va_uid);
789 if (vap->va_gid == (gid_t)VNOVAL)
790 sp->sa_gid = nfs_xdrneg1;
792 sp->sa_gid = txdr_unsigned(vap->va_gid);
793 sp->sa_size = txdr_unsigned(vap->va_size);
794 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
795 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
797 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
800 nfsm_wcc_data(vp, wccflag);
802 nfsm_loadattr(vp, (struct vattr *)0);
809 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
810 * nfs_lookup() until all remaining new api calls are implemented.
812 * Resolve a namecache entry. This function is passed a locked ncp and
813 * must call cache_setvp() on it as appropriate to resolve the entry.
816 nfs_nresolve(struct vop_nresolve_args *ap)
818 struct thread *td = curthread;
819 struct namecache *ncp;
830 /******NFSM MACROS********/
831 struct mbuf *mb, *mrep, *mreq, *mb2, *md;
832 caddr_t bpos, dpos, cp, cp2;
837 ncp = ap->a_nch->ncp;
839 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp);
840 dvp = ncp->nc_parent->nc_vp;
841 if ((error = vget(dvp, LK_SHARED)) != 0)
846 nfsstats.lookupcache_misses++;
847 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
849 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
850 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
852 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN);
853 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred);
856 * Cache negatve lookups to reduce NFS traffic, but use
857 * a fast timeout. Otherwise use a timeout of 1 tick.
858 * XXX we should add a namecache flag for no-caching
859 * to uncache the negative hit as soon as possible, but
860 * we cannot simply destroy the entry because it is used
861 * as a placeholder by the caller.
863 if (error == ENOENT) {
866 if (nfsneg_cache_timeout)
867 nticks = nfsneg_cache_timeout * hz;
870 cache_setvp(ap->a_nch, NULL);
871 cache_settimeout(ap->a_nch, nticks);
873 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
879 * Success, get the file handle, do various checks, and load
880 * post-operation data from the reply packet. Theoretically
881 * we should never be looking up "." so, theoretically, we
882 * should never get the same file handle as our directory. But
883 * we check anyway. XXX
885 * Note that no timeout is set for the positive cache hit. We
886 * assume, theoretically, that ESTALE returns will be dealt with
887 * properly to handle NFS races and in anycase we cannot depend
888 * on a timeout to deal with NFS open/create/excl issues so instead
889 * of a bad hack here the rest of the NFS client code needs to do
892 nfsm_getfh(fhp, fhsize, v3);
895 if (NFS_CMPFH(np, fhp, fhsize)) {
899 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
908 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK);
909 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
911 nfsm_loadattr(nvp, NULL);
913 cache_setvp(ap->a_nch, nvp);
927 * 'cached' nfs directory lookup
929 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
931 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
932 * struct componentname *a_cnp)
935 nfs_lookup(struct vop_old_lookup_args *ap)
937 struct componentname *cnp = ap->a_cnp;
938 struct vnode *dvp = ap->a_dvp;
939 struct vnode **vpp = ap->a_vpp;
940 int flags = cnp->cn_flags;
945 struct nfsmount *nmp;
946 caddr_t bpos, dpos, cp2;
947 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
951 int lockparent, wantparent, error = 0, attrflag, fhsize;
952 int v3 = NFS_ISV3(dvp);
955 * Read-only mount check and directory check.
958 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
959 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
962 if (dvp->v_type != VDIR)
966 * Look it up in the cache. Note that ENOENT is only returned if we
967 * previously entered a negative hit (see later on). The additional
968 * nfsneg_cache_timeout check causes previously cached results to
969 * be instantly ignored if the negative caching is turned off.
971 lockparent = flags & CNP_LOCKPARENT;
972 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
973 nmp = VFSTONFS(dvp->v_mount);
981 nfsstats.lookupcache_misses++;
982 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
983 len = cnp->cn_namelen;
984 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
985 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
987 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
988 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred);
990 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
994 nfsm_getfh(fhp, fhsize, v3);
997 * Handle RENAME case...
999 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1000 if (NFS_CMPFH(np, fhp, fhsize)) {
1004 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1011 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1012 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1014 nfsm_loadattr(newvp, (struct vattr *)0);
1019 cnp->cn_flags |= CNP_PDIRUNLOCK;
1024 if (flags & CNP_ISDOTDOT) {
1026 cnp->cn_flags |= CNP_PDIRUNLOCK;
1027 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1029 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1030 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1031 return (error); /* NOTE: return error from nget */
1035 error = vn_lock(dvp, LK_EXCLUSIVE);
1040 cnp->cn_flags |= CNP_PDIRUNLOCK;
1042 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1046 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1053 cnp->cn_flags |= CNP_PDIRUNLOCK;
1058 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1059 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1061 nfsm_loadattr(newvp, (struct vattr *)0);
1063 /* XXX MOVE TO nfs_nremove() */
1064 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1065 cnp->cn_nameiop != NAMEI_DELETE) {
1066 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1073 if (newvp != NULLVP) {
1077 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1078 cnp->cn_nameiop == NAMEI_RENAME) &&
1082 cnp->cn_flags |= CNP_PDIRUNLOCK;
1084 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1087 error = EJUSTRETURN;
1095 * Just call nfs_bioread() to do the work.
1097 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1098 * struct ucred *a_cred)
1101 nfs_read(struct vop_read_args *ap)
1103 struct vnode *vp = ap->a_vp;
1105 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1106 switch (vp->v_type) {
1108 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1119 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1122 nfs_readlink(struct vop_readlink_args *ap)
1124 struct vnode *vp = ap->a_vp;
1126 if (vp->v_type != VLNK)
1128 return (nfs_bioread(vp, ap->a_uio, 0));
1132 * Do a readlink rpc.
1133 * Called by nfs_doio() from below the buffer cache.
1136 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1141 caddr_t bpos, dpos, cp2;
1142 int error = 0, len, attrflag;
1143 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1144 int v3 = NFS_ISV3(vp);
1146 nfsstats.rpccnt[NFSPROC_READLINK]++;
1147 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1149 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK));
1151 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1153 nfsm_strsiz(len, NFS_MAXPATHLEN);
1154 if (len == NFS_MAXPATHLEN) {
1155 struct nfsnode *np = VTONFS(vp);
1156 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1159 nfsm_mtouio(uiop, len);
1171 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1176 caddr_t bpos, dpos, cp2;
1177 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1178 struct nfsmount *nmp;
1179 int error = 0, len, retlen, tsiz, eof, attrflag;
1180 int v3 = NFS_ISV3(vp);
1185 nmp = VFSTONFS(vp->v_mount);
1186 tsiz = uiop->uio_resid;
1187 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1190 nfsstats.rpccnt[NFSPROC_READ]++;
1191 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1192 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1194 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1196 txdr_hyper(uiop->uio_offset, tl);
1197 *(tl + 2) = txdr_unsigned(len);
1199 *tl++ = txdr_unsigned(uiop->uio_offset);
1200 *tl++ = txdr_unsigned(len);
1203 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ));
1205 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1210 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1211 eof = fxdr_unsigned(int, *(tl + 1));
1213 nfsm_loadattr(vp, (struct vattr *)0);
1214 nfsm_strsiz(retlen, nmp->nm_rsize);
1215 nfsm_mtouio(uiop, retlen);
1219 if (eof || retlen == 0) {
1222 } else if (retlen < len) {
1234 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1238 int32_t t1, t2, backup;
1239 caddr_t bpos, dpos, cp2;
1240 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1241 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1242 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1243 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1246 if (uiop->uio_iovcnt != 1)
1247 panic("nfs: writerpc iovcnt > 1");
1250 tsiz = uiop->uio_resid;
1251 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1254 nfsstats.rpccnt[NFSPROC_WRITE]++;
1255 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1256 nfsm_reqhead(vp, NFSPROC_WRITE,
1257 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1260 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1261 txdr_hyper(uiop->uio_offset, tl);
1263 *tl++ = txdr_unsigned(len);
1264 *tl++ = txdr_unsigned(*iomode);
1265 *tl = txdr_unsigned(len);
1269 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1270 /* Set both "begin" and "current" to non-garbage. */
1271 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1272 *tl++ = x; /* "begin offset" */
1273 *tl++ = x; /* "current offset" */
1274 x = txdr_unsigned(len);
1275 *tl++ = x; /* total to this offset */
1276 *tl = x; /* size of this write */
1278 nfsm_uiotom(uiop, len);
1279 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE));
1282 * The write RPC returns a before and after mtime. The
1283 * nfsm_wcc_data() macro checks the before n_mtime
1284 * against the before time and stores the after time
1285 * in the nfsnode's cached vattr and n_mtime field.
1286 * The NRMODIFIED bit will be set if the before
1287 * time did not match the original mtime.
1289 wccflag = NFSV3_WCCCHK;
1290 nfsm_wcc_data(vp, wccflag);
1292 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1293 + NFSX_V3WRITEVERF);
1294 rlen = fxdr_unsigned(int, *tl++);
1299 } else if (rlen < len) {
1300 backup = len - rlen;
1301 uiop->uio_iov->iov_base -= backup;
1302 uiop->uio_iov->iov_len += backup;
1303 uiop->uio_offset -= backup;
1304 uiop->uio_resid += backup;
1307 commit = fxdr_unsigned(int, *tl++);
1310 * Return the lowest committment level
1311 * obtained by any of the RPCs.
1313 if (committed == NFSV3WRITE_FILESYNC)
1315 else if (committed == NFSV3WRITE_DATASYNC &&
1316 commit == NFSV3WRITE_UNSTABLE)
1318 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1319 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1321 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1322 } else if (bcmp((caddr_t)tl,
1323 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1325 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1330 nfsm_loadattr(vp, (struct vattr *)0);
1338 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1339 committed = NFSV3WRITE_FILESYNC;
1340 *iomode = committed;
1342 uiop->uio_resid = tsiz;
1348 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1349 * mode set to specify the file type and the size field for rdev.
1352 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1355 struct nfsv2_sattr *sp;
1359 struct vnode *newvp = (struct vnode *)0;
1360 struct nfsnode *np = (struct nfsnode *)0;
1364 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1365 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1367 int v3 = NFS_ISV3(dvp);
1369 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1370 rmajor = txdr_unsigned(vap->va_rmajor);
1371 rminor = txdr_unsigned(vap->va_rminor);
1372 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1373 rmajor = nfs_xdrneg1;
1374 rminor = nfs_xdrneg1;
1376 return (EOPNOTSUPP);
1378 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1381 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1382 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1383 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1384 nfsm_fhtom(dvp, v3);
1385 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1387 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1388 *tl++ = vtonfsv3_type(vap->va_type);
1389 nfsm_v3attrbuild(vap, FALSE);
1390 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1391 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1392 *tl++ = txdr_unsigned(vap->va_rmajor);
1393 *tl = txdr_unsigned(vap->va_rminor);
1396 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1397 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1398 sp->sa_uid = nfs_xdrneg1;
1399 sp->sa_gid = nfs_xdrneg1;
1400 sp->sa_size = makeudev(rmajor, rminor);
1401 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1402 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1404 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred);
1406 nfsm_mtofh(dvp, newvp, v3, gotvp);
1410 newvp = (struct vnode *)0;
1412 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1413 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1419 nfsm_wcc_data(dvp, wccflag);
1428 VTONFS(dvp)->n_flag |= NLMODIFIED;
1430 VTONFS(dvp)->n_attrstamp = 0;
1436 * just call nfs_mknodrpc() to do the work.
1438 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1439 * struct componentname *a_cnp, struct vattr *a_vap)
1443 nfs_mknod(struct vop_old_mknod_args *ap)
1445 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1448 static u_long create_verf;
1450 * nfs file create call
1452 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1453 * struct componentname *a_cnp, struct vattr *a_vap)
1456 nfs_create(struct vop_old_create_args *ap)
1458 struct vnode *dvp = ap->a_dvp;
1459 struct vattr *vap = ap->a_vap;
1460 struct componentname *cnp = ap->a_cnp;
1461 struct nfsv2_sattr *sp;
1465 struct nfsnode *np = (struct nfsnode *)0;
1466 struct vnode *newvp = (struct vnode *)0;
1467 caddr_t bpos, dpos, cp2;
1468 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1469 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1471 int v3 = NFS_ISV3(dvp);
1474 * Oops, not for me..
1476 if (vap->va_type == VSOCK)
1477 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1479 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1482 if (vap->va_vaflags & VA_EXCLUSIVE)
1485 nfsstats.rpccnt[NFSPROC_CREATE]++;
1486 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1487 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1488 nfsm_fhtom(dvp, v3);
1489 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1491 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1492 if (fmode & O_EXCL) {
1493 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1494 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1496 if (!TAILQ_EMPTY(&in_ifaddrhead))
1497 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1500 *tl++ = create_verf;
1501 *tl = ++create_verf;
1503 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1504 nfsm_v3attrbuild(vap, FALSE);
1507 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1508 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1509 sp->sa_uid = nfs_xdrneg1;
1510 sp->sa_gid = nfs_xdrneg1;
1512 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1513 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1515 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred);
1517 nfsm_mtofh(dvp, newvp, v3, gotvp);
1521 newvp = (struct vnode *)0;
1523 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1524 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1530 nfsm_wcc_data(dvp, wccflag);
1534 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1540 } else if (v3 && (fmode & O_EXCL)) {
1542 * We are normally called with only a partially initialized
1543 * VAP. Since the NFSv3 spec says that server may use the
1544 * file attributes to store the verifier, the spec requires
1545 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1546 * in atime, but we can't really assume that all servers will
1547 * so we ensure that our SETATTR sets both atime and mtime.
1549 if (vap->va_mtime.tv_sec == VNOVAL)
1550 vfs_timestamp(&vap->va_mtime);
1551 if (vap->va_atime.tv_sec == VNOVAL)
1552 vap->va_atime = vap->va_mtime;
1553 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1557 * The new np may have enough info for access
1558 * checks, make sure rucred and wucred are
1559 * initialized for read and write rpc's.
1562 if (np->n_rucred == NULL)
1563 np->n_rucred = crhold(cnp->cn_cred);
1564 if (np->n_wucred == NULL)
1565 np->n_wucred = crhold(cnp->cn_cred);
1568 VTONFS(dvp)->n_flag |= NLMODIFIED;
1570 VTONFS(dvp)->n_attrstamp = 0;
1575 * nfs file remove call
1576 * To try and make nfs semantics closer to ufs semantics, a file that has
1577 * other processes using the vnode is renamed instead of removed and then
1578 * removed later on the last close.
1579 * - If v_sysref.refcnt > 1
1580 * If a rename is not already in the works
1581 * call nfs_sillyrename() to set it up
1585 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1586 * struct componentname *a_cnp)
1589 nfs_remove(struct vop_old_remove_args *ap)
1591 struct vnode *vp = ap->a_vp;
1592 struct vnode *dvp = ap->a_dvp;
1593 struct componentname *cnp = ap->a_cnp;
1594 struct nfsnode *np = VTONFS(vp);
1599 if (vp->v_sysref.refcnt < 1)
1600 panic("nfs_remove: bad v_sysref.refcnt");
1602 if (vp->v_type == VDIR)
1604 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1605 VOP_GETATTR(vp, &vattr) == 0 &&
1606 vattr.va_nlink > 1)) {
1608 * throw away biocache buffers, mainly to avoid
1609 * unnecessary delayed writes later.
1611 error = nfs_vinvalbuf(vp, 0, 1);
1614 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1615 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1617 * Kludge City: If the first reply to the remove rpc is lost..
1618 * the reply to the retransmitted request will be ENOENT
1619 * since the file was in fact removed
1620 * Therefore, we cheat and return success.
1622 if (error == ENOENT)
1624 } else if (!np->n_sillyrename) {
1625 error = nfs_sillyrename(dvp, vp, cnp);
1627 np->n_attrstamp = 0;
1632 * nfs file remove rpc called from nfs_inactive
1635 nfs_removeit(struct sillyrename *sp)
1637 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1642 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1645 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1646 struct ucred *cred, struct thread *td)
1651 caddr_t bpos, dpos, cp2;
1652 int error = 0, wccflag = NFSV3_WCCRATTR;
1653 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1654 int v3 = NFS_ISV3(dvp);
1656 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1657 nfsm_reqhead(dvp, NFSPROC_REMOVE,
1658 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1659 nfsm_fhtom(dvp, v3);
1660 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1661 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1663 nfsm_wcc_data(dvp, wccflag);
1666 VTONFS(dvp)->n_flag |= NLMODIFIED;
1668 VTONFS(dvp)->n_attrstamp = 0;
1673 * nfs file rename call
1675 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1676 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1677 * struct vnode *a_tvp, struct componentname *a_tcnp)
1680 nfs_rename(struct vop_old_rename_args *ap)
1682 struct vnode *fvp = ap->a_fvp;
1683 struct vnode *tvp = ap->a_tvp;
1684 struct vnode *fdvp = ap->a_fdvp;
1685 struct vnode *tdvp = ap->a_tdvp;
1686 struct componentname *tcnp = ap->a_tcnp;
1687 struct componentname *fcnp = ap->a_fcnp;
1690 /* Check for cross-device rename */
1691 if ((fvp->v_mount != tdvp->v_mount) ||
1692 (tvp && (fvp->v_mount != tvp->v_mount))) {
1698 * We have to flush B_DELWRI data prior to renaming
1699 * the file. If we don't, the delayed-write buffers
1700 * can be flushed out later after the file has gone stale
1701 * under NFSV3. NFSV2 does not have this problem because
1702 * ( as far as I can tell ) it flushes dirty buffers more
1706 VOP_FSYNC(fvp, MNT_WAIT);
1708 VOP_FSYNC(tvp, MNT_WAIT);
1711 * If the tvp exists and is in use, sillyrename it before doing the
1712 * rename of the new file over it.
1714 * XXX Can't sillyrename a directory.
1716 * We do not attempt to do any namecache purges in this old API
1717 * routine. The new API compat functions have access to the actual
1718 * namecache structures and will do it for us.
1720 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1721 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1728 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1729 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1742 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1744 if (error == ENOENT)
1750 * nfs file rename rpc called from nfs_remove() above
1753 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1754 struct sillyrename *sp)
1756 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1757 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1761 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1764 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1765 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1766 struct ucred *cred, struct thread *td)
1771 caddr_t bpos, dpos, cp2;
1772 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1773 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1774 int v3 = NFS_ISV3(fdvp);
1776 nfsstats.rpccnt[NFSPROC_RENAME]++;
1777 nfsm_reqhead(fdvp, NFSPROC_RENAME,
1778 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1779 nfsm_rndup(tnamelen));
1780 nfsm_fhtom(fdvp, v3);
1781 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1782 nfsm_fhtom(tdvp, v3);
1783 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1784 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1786 nfsm_wcc_data(fdvp, fwccflag);
1787 nfsm_wcc_data(tdvp, twccflag);
1791 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1792 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1794 VTONFS(fdvp)->n_attrstamp = 0;
1796 VTONFS(tdvp)->n_attrstamp = 0;
1801 * nfs hard link create call
1803 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1804 * struct componentname *a_cnp)
1807 nfs_link(struct vop_old_link_args *ap)
1809 struct vnode *vp = ap->a_vp;
1810 struct vnode *tdvp = ap->a_tdvp;
1811 struct componentname *cnp = ap->a_cnp;
1815 caddr_t bpos, dpos, cp2;
1816 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1817 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1820 if (vp->v_mount != tdvp->v_mount) {
1825 * Push all writes to the server, so that the attribute cache
1826 * doesn't get "out of sync" with the server.
1827 * XXX There should be a better way!
1829 VOP_FSYNC(vp, MNT_WAIT);
1832 nfsstats.rpccnt[NFSPROC_LINK]++;
1833 nfsm_reqhead(vp, NFSPROC_LINK,
1834 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1836 nfsm_fhtom(tdvp, v3);
1837 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1838 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred);
1840 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1841 nfsm_wcc_data(tdvp, wccflag);
1845 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1847 VTONFS(vp)->n_attrstamp = 0;
1849 VTONFS(tdvp)->n_attrstamp = 0;
1851 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1853 if (error == EEXIST)
1859 * nfs symbolic link create call
1861 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1862 * struct componentname *a_cnp, struct vattr *a_vap,
1866 nfs_symlink(struct vop_old_symlink_args *ap)
1868 struct vnode *dvp = ap->a_dvp;
1869 struct vattr *vap = ap->a_vap;
1870 struct componentname *cnp = ap->a_cnp;
1871 struct nfsv2_sattr *sp;
1875 caddr_t bpos, dpos, cp2;
1876 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1877 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1878 struct vnode *newvp = (struct vnode *)0;
1879 int v3 = NFS_ISV3(dvp);
1881 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1882 slen = strlen(ap->a_target);
1883 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1884 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1885 nfsm_fhtom(dvp, v3);
1886 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1888 nfsm_v3attrbuild(vap, FALSE);
1890 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1892 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1893 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1894 sp->sa_uid = nfs_xdrneg1;
1895 sp->sa_gid = nfs_xdrneg1;
1896 sp->sa_size = nfs_xdrneg1;
1897 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1898 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1902 * Issue the NFS request and get the rpc response.
1904 * Only NFSv3 responses returning an error of 0 actually return
1905 * a file handle that can be converted into newvp without having
1906 * to do an extra lookup rpc.
1908 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred);
1911 nfsm_mtofh(dvp, newvp, v3, gotvp);
1912 nfsm_wcc_data(dvp, wccflag);
1916 * out code jumps -> here, mrep is also freed.
1923 * If we get an EEXIST error, silently convert it to no-error
1924 * in case of an NFS retry.
1926 if (error == EEXIST)
1930 * If we do not have (or no longer have) an error, and we could
1931 * not extract the newvp from the response due to the request being
1932 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1933 * to obtain a newvp to return.
1935 if (error == 0 && newvp == NULL) {
1936 struct nfsnode *np = NULL;
1938 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1939 cnp->cn_cred, cnp->cn_td, &np);
1949 VTONFS(dvp)->n_flag |= NLMODIFIED;
1951 VTONFS(dvp)->n_attrstamp = 0;
1958 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1959 * struct componentname *a_cnp, struct vattr *a_vap)
1962 nfs_mkdir(struct vop_old_mkdir_args *ap)
1964 struct vnode *dvp = ap->a_dvp;
1965 struct vattr *vap = ap->a_vap;
1966 struct componentname *cnp = ap->a_cnp;
1967 struct nfsv2_sattr *sp;
1972 struct nfsnode *np = (struct nfsnode *)0;
1973 struct vnode *newvp = (struct vnode *)0;
1974 caddr_t bpos, dpos, cp2;
1975 int error = 0, wccflag = NFSV3_WCCRATTR;
1977 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1979 int v3 = NFS_ISV3(dvp);
1981 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1984 len = cnp->cn_namelen;
1985 nfsstats.rpccnt[NFSPROC_MKDIR]++;
1986 nfsm_reqhead(dvp, NFSPROC_MKDIR,
1987 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1988 nfsm_fhtom(dvp, v3);
1989 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1991 nfsm_v3attrbuild(vap, FALSE);
1993 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1994 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1995 sp->sa_uid = nfs_xdrneg1;
1996 sp->sa_gid = nfs_xdrneg1;
1997 sp->sa_size = nfs_xdrneg1;
1998 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1999 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2001 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred);
2003 nfsm_mtofh(dvp, newvp, v3, gotvp);
2005 nfsm_wcc_data(dvp, wccflag);
2008 VTONFS(dvp)->n_flag |= NLMODIFIED;
2010 VTONFS(dvp)->n_attrstamp = 0;
2012 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2013 * if we can succeed in looking up the directory.
2015 if (error == EEXIST || (!error && !gotvp)) {
2018 newvp = (struct vnode *)0;
2020 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2024 if (newvp->v_type != VDIR)
2037 * nfs remove directory call
2039 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2040 * struct componentname *a_cnp)
2043 nfs_rmdir(struct vop_old_rmdir_args *ap)
2045 struct vnode *vp = ap->a_vp;
2046 struct vnode *dvp = ap->a_dvp;
2047 struct componentname *cnp = ap->a_cnp;
2051 caddr_t bpos, dpos, cp2;
2052 int error = 0, wccflag = NFSV3_WCCRATTR;
2053 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2054 int v3 = NFS_ISV3(dvp);
2058 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2059 nfsm_reqhead(dvp, NFSPROC_RMDIR,
2060 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2061 nfsm_fhtom(dvp, v3);
2062 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2063 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred);
2065 nfsm_wcc_data(dvp, wccflag);
2068 VTONFS(dvp)->n_flag |= NLMODIFIED;
2070 VTONFS(dvp)->n_attrstamp = 0;
2072 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2074 if (error == ENOENT)
2082 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2085 nfs_readdir(struct vop_readdir_args *ap)
2087 struct vnode *vp = ap->a_vp;
2088 struct nfsnode *np = VTONFS(vp);
2089 struct uio *uio = ap->a_uio;
2093 if (vp->v_type != VDIR)
2096 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2100 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2101 * and then check that is still valid, or if this is an NQNFS mount
2102 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2103 * VOP_GETATTR() does not necessarily go to the wire.
2105 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2106 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2107 if (VOP_GETATTR(vp, &vattr) == 0 &&
2108 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2110 nfsstats.direofcache_hits++;
2116 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2117 * own cache coherency checks so we do not have to.
2119 tresid = uio->uio_resid;
2120 error = nfs_bioread(vp, uio, 0);
2122 if (!error && uio->uio_resid == tresid)
2123 nfsstats.direofcache_misses++;
2130 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2132 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2133 * offset/block and converts the nfs formatted directory entries for userland
2134 * consumption as well as deals with offsets into the middle of blocks.
2135 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2136 * be block-bounded. It must convert to cookies for the actual RPC.
2139 nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2142 struct nfs_dirent *dp = NULL;
2147 caddr_t bpos, dpos, cp2;
2148 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2150 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2151 struct nfsnode *dnp = VTONFS(vp);
2153 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2155 int v3 = NFS_ISV3(vp);
2158 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2159 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2160 panic("nfs readdirrpc bad uio");
2164 * If there is no cookie, assume directory was stale.
2166 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2170 return (NFSERR_BAD_COOKIE);
2172 * Loop around doing readdir rpc's of size nm_readdirsize
2173 * truncated to a multiple of DIRBLKSIZ.
2174 * The stopping criteria is EOF or buffer full.
2176 while (more_dirs && bigenough) {
2177 nfsstats.rpccnt[NFSPROC_READDIR]++;
2178 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2182 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2183 *tl++ = cookie.nfsuquad[0];
2184 *tl++ = cookie.nfsuquad[1];
2185 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2186 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2188 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2189 *tl++ = cookie.nfsuquad[0];
2191 *tl = txdr_unsigned(nmp->nm_readdirsize);
2192 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2194 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2196 nfsm_dissect(tl, u_int32_t *,
2198 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2199 dnp->n_cookieverf.nfsuquad[1] = *tl;
2205 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2206 more_dirs = fxdr_unsigned(int, *tl);
2208 /* loop thru the dir entries, converting them to std form */
2209 while (more_dirs && bigenough) {
2211 nfsm_dissect(tl, u_int32_t *,
2213 fileno = fxdr_hyper(tl);
2214 len = fxdr_unsigned(int, *(tl + 2));
2216 nfsm_dissect(tl, u_int32_t *,
2218 fileno = fxdr_unsigned(u_quad_t, *tl++);
2219 len = fxdr_unsigned(int, *tl);
2221 if (len <= 0 || len > NFS_MAXNAMLEN) {
2228 * len is the number of bytes in the path element
2229 * name, not including the \0 termination.
2231 * tlen is the number of bytes w have to reserve for
2232 * the path element name.
2234 tlen = nfsm_rndup(len);
2236 tlen += 4; /* To ensure null termination */
2239 * If the entry would cross a DIRBLKSIZ boundary,
2240 * extend the previous nfs_dirent to cover the
2243 left = DIRBLKSIZ - blksiz;
2244 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2245 dp->nfs_reclen += left;
2246 uiop->uio_iov->iov_base += left;
2247 uiop->uio_iov->iov_len -= left;
2248 uiop->uio_offset += left;
2249 uiop->uio_resid -= left;
2252 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2255 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2256 dp->nfs_ino = fileno;
2257 dp->nfs_namlen = len;
2258 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2259 dp->nfs_type = DT_UNKNOWN;
2260 blksiz += dp->nfs_reclen;
2261 if (blksiz == DIRBLKSIZ)
2263 uiop->uio_offset += sizeof(struct nfs_dirent);
2264 uiop->uio_resid -= sizeof(struct nfs_dirent);
2265 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent);
2266 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2267 nfsm_mtouio(uiop, len);
2270 * The uiop has advanced by nfs_dirent + len
2271 * but really needs to advance by
2274 cp = uiop->uio_iov->iov_base;
2276 *cp = '\0'; /* null terminate */
2277 uiop->uio_iov->iov_base += tlen;
2278 uiop->uio_iov->iov_len -= tlen;
2279 uiop->uio_offset += tlen;
2280 uiop->uio_resid -= tlen;
2283 * NFS strings must be rounded up (nfsm_myouio
2284 * handled that in the bigenough case).
2286 nfsm_adv(nfsm_rndup(len));
2289 nfsm_dissect(tl, u_int32_t *,
2292 nfsm_dissect(tl, u_int32_t *,
2297 * If we were able to accomodate the last entry,
2298 * get the cookie for the next one. Otherwise
2299 * hold-over the cookie for the one we were not
2300 * able to accomodate.
2303 cookie.nfsuquad[0] = *tl++;
2305 cookie.nfsuquad[1] = *tl++;
2311 more_dirs = fxdr_unsigned(int, *tl);
2314 * If at end of rpc data, get the eof boolean
2317 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2318 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2323 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2324 * by increasing d_reclen for the last record.
2327 left = DIRBLKSIZ - blksiz;
2328 dp->nfs_reclen += left;
2329 uiop->uio_iov->iov_base += left;
2330 uiop->uio_iov->iov_len -= left;
2331 uiop->uio_offset += left;
2332 uiop->uio_resid -= left;
2337 * We hit the end of the directory, update direofoffset.
2339 dnp->n_direofoffset = uiop->uio_offset;
2342 * There is more to go, insert the link cookie so the
2343 * next block can be read.
2345 if (uiop->uio_resid > 0)
2346 kprintf("EEK! readdirrpc resid > 0\n");
2347 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2355 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2358 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2361 struct nfs_dirent *dp;
2365 struct vnode *newvp;
2367 caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2368 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2370 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2371 struct nfsnode *dnp = VTONFS(vp), *np;
2374 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2375 int attrflag, fhsize;
2376 struct nchandle nch;
2377 struct nchandle dnch;
2378 struct nlcomponent nlc;
2384 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2385 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2386 panic("nfs readdirplusrpc bad uio");
2389 * Obtain the namecache record for the directory so we have something
2390 * to use as a basis for creating the entries. This function will
2391 * return a held (but not locked) ncp. The ncp may be disconnected
2392 * from the tree and cannot be used for upward traversals, and the
2393 * ncp may be unnamed. Note that other unrelated operations may
2394 * cause the ncp to be named at any time.
2396 cache_fromdvp(vp, NULL, 0, &dnch);
2397 bzero(&nlc, sizeof(nlc));
2401 * If there is no cookie, assume directory was stale.
2403 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2407 return (NFSERR_BAD_COOKIE);
2409 * Loop around doing readdir rpc's of size nm_readdirsize
2410 * truncated to a multiple of DIRBLKSIZ.
2411 * The stopping criteria is EOF or buffer full.
2413 while (more_dirs && bigenough) {
2414 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2415 nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2416 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2418 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2419 *tl++ = cookie.nfsuquad[0];
2420 *tl++ = cookie.nfsuquad[1];
2421 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2422 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2423 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2424 *tl = txdr_unsigned(nmp->nm_rsize);
2425 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2426 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2431 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2432 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2433 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2434 more_dirs = fxdr_unsigned(int, *tl);
2436 /* loop thru the dir entries, doctoring them to 4bsd form */
2437 while (more_dirs && bigenough) {
2438 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2439 fileno = fxdr_hyper(tl);
2440 len = fxdr_unsigned(int, *(tl + 2));
2441 if (len <= 0 || len > NFS_MAXNAMLEN) {
2446 tlen = nfsm_rndup(len);
2448 tlen += 4; /* To ensure null termination*/
2449 left = DIRBLKSIZ - blksiz;
2450 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2451 dp->nfs_reclen += left;
2452 uiop->uio_iov->iov_base += left;
2453 uiop->uio_iov->iov_len -= left;
2454 uiop->uio_offset += left;
2455 uiop->uio_resid -= left;
2458 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2461 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2462 dp->nfs_ino = fileno;
2463 dp->nfs_namlen = len;
2464 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2465 dp->nfs_type = DT_UNKNOWN;
2466 blksiz += dp->nfs_reclen;
2467 if (blksiz == DIRBLKSIZ)
2469 uiop->uio_offset += sizeof(struct nfs_dirent);
2470 uiop->uio_resid -= sizeof(struct nfs_dirent);
2471 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent);
2472 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2473 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2474 nlc.nlc_namelen = len;
2475 nfsm_mtouio(uiop, len);
2476 cp = uiop->uio_iov->iov_base;
2479 uiop->uio_iov->iov_base += tlen;
2480 uiop->uio_iov->iov_len -= tlen;
2481 uiop->uio_offset += tlen;
2482 uiop->uio_resid -= tlen;
2484 nfsm_adv(nfsm_rndup(len));
2485 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2487 cookie.nfsuquad[0] = *tl++;
2488 cookie.nfsuquad[1] = *tl++;
2493 * Since the attributes are before the file handle
2494 * (sigh), we must skip over the attributes and then
2495 * come back and get them.
2497 attrflag = fxdr_unsigned(int, *tl);
2501 nfsm_adv(NFSX_V3FATTR);
2502 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2503 doit = fxdr_unsigned(int, *tl);
2505 nfsm_getfh(fhp, fhsize, 1);
2506 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2511 error = nfs_nget(vp->v_mount, fhp,
2519 if (doit && bigenough) {
2524 nfsm_loadattr(newvp, (struct vattr *)0);
2528 IFTODT(VTTOIF(np->n_vattr.va_type));
2530 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2531 nlc.nlc_namelen, nlc.nlc_namelen,
2533 nch = cache_nlookup(&dnch, &nlc);
2534 cache_setunresolved(&nch);
2535 cache_setvp(&nch, newvp);
2538 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER"
2540 nlc.nlc_namelen, nlc.nlc_namelen,
2545 /* Just skip over the file handle */
2546 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2547 i = fxdr_unsigned(int, *tl);
2548 nfsm_adv(nfsm_rndup(i));
2550 if (newvp != NULLVP) {
2557 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2558 more_dirs = fxdr_unsigned(int, *tl);
2561 * If at end of rpc data, get the eof boolean
2564 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2565 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2570 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2571 * by increasing d_reclen for the last record.
2574 left = DIRBLKSIZ - blksiz;
2575 dp->nfs_reclen += left;
2576 uiop->uio_iov->iov_base += left;
2577 uiop->uio_iov->iov_len -= left;
2578 uiop->uio_offset += left;
2579 uiop->uio_resid -= left;
2583 * We are now either at the end of the directory or have filled the
2587 dnp->n_direofoffset = uiop->uio_offset;
2589 if (uiop->uio_resid > 0)
2590 kprintf("EEK! readdirplusrpc resid > 0\n");
2591 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2595 if (newvp != NULLVP) {
2608 * Silly rename. To make the NFS filesystem that is stateless look a little
2609 * more like the "ufs" a remove of an active vnode is translated to a rename
2610 * to a funny looking filename that is removed by nfs_inactive on the
2611 * nfsnode. There is the potential for another process on a different client
2612 * to create the same funny name between the nfs_lookitup() fails and the
2613 * nfs_rename() completes, but...
2616 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2618 struct sillyrename *sp;
2623 * We previously purged dvp instead of vp. I don't know why, it
2624 * completely destroys performance. We can't do it anyway with the
2625 * new VFS API since we would be breaking the namecache topology.
2627 cache_purge(vp); /* XXX */
2630 if (vp->v_type == VDIR)
2631 panic("nfs: sillyrename dir");
2633 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2634 M_NFSREQ, M_WAITOK);
2635 sp->s_cred = crdup(cnp->cn_cred);
2639 /* Fudge together a funny name */
2640 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td);
2642 /* Try lookitups until we get one that isn't there */
2643 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2644 cnp->cn_td, (struct nfsnode **)0) == 0) {
2646 if (sp->s_name[4] > 'z') {
2651 error = nfs_renameit(dvp, cnp, sp);
2654 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2656 np->n_sillyrename = sp;
2661 kfree((caddr_t)sp, M_NFSREQ);
2666 * Look up a file name and optionally either update the file handle or
2667 * allocate an nfsnode, depending on the value of npp.
2668 * npp == NULL --> just do the lookup
2669 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2671 * *npp != NULL --> update the file handle in the vnode
2674 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2675 struct thread *td, struct nfsnode **npp)
2680 struct vnode *newvp = (struct vnode *)0;
2681 struct nfsnode *np, *dnp = VTONFS(dvp);
2682 caddr_t bpos, dpos, cp2;
2683 int error = 0, fhlen, attrflag;
2684 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2686 int v3 = NFS_ISV3(dvp);
2688 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2689 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2690 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2691 nfsm_fhtom(dvp, v3);
2692 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2693 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2694 if (npp && !error) {
2695 nfsm_getfh(nfhp, fhlen, v3);
2698 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2699 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2700 np->n_fhp = &np->n_fh;
2701 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2702 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2703 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2704 np->n_fhsize = fhlen;
2706 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2710 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2718 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
2719 if (!attrflag && *npp == NULL) {
2728 nfsm_loadattr(newvp, (struct vattr *)0);
2732 if (npp && *npp == NULL) {
2747 * Nfs Version 3 commit rpc
2750 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2755 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2756 caddr_t bpos, dpos, cp2;
2757 int error = 0, wccflag = NFSV3_WCCRATTR;
2758 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2760 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2762 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2763 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2765 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2766 txdr_hyper(offset, tl);
2768 *tl = txdr_unsigned(cnt);
2769 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE));
2770 nfsm_wcc_data(vp, wccflag);
2772 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2773 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2774 NFSX_V3WRITEVERF)) {
2775 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2777 error = NFSERR_STALEWRITEVERF;
2787 * - make nfs_bmap() essentially a no-op that does no translation
2788 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2789 * (Maybe I could use the process's page mapping, but I was concerned that
2790 * Kernel Write might not be enabled and also figured copyout() would do
2791 * a lot more work than bcopy() and also it currently happens in the
2792 * context of the swapper process (2).
2794 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, struct vnode **a_vpp,
2795 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2798 nfs_bmap(struct vop_bmap_args *ap)
2800 struct vnode *vp = ap->a_vp;
2802 if (ap->a_vpp != NULL)
2804 if (ap->a_doffsetp != NULL)
2805 *ap->a_doffsetp = ap->a_loffset;
2806 if (ap->a_runp != NULL)
2808 if (ap->a_runb != NULL)
2816 * For async requests when nfsiod(s) are running, queue the request by
2817 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2821 nfs_strategy(struct vop_strategy_args *ap)
2823 struct bio *bio = ap->a_bio;
2825 struct buf *bp = bio->bio_buf;
2829 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2830 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
2831 KASSERT(BUF_REFCNT(bp) > 0,
2832 ("nfs_strategy: buffer %p not locked", bp));
2834 if (bp->b_flags & B_ASYNC)
2837 td = curthread; /* XXX */
2840 * We probably don't need to push an nbio any more since no
2841 * block conversion is required due to the use of 64 bit byte
2842 * offsets, but do it anyway.
2844 nbio = push_bio(bio);
2845 nbio->bio_offset = bio->bio_offset;
2848 * If the op is asynchronous and an i/o daemon is waiting
2849 * queue the request, wake it up and wait for completion
2850 * otherwise just do it ourselves.
2852 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(ap->a_vp, nbio, td))
2853 error = nfs_doio(ap->a_vp, nbio, td);
2860 * NB Currently unsupported.
2862 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred,
2863 * struct thread *a_td)
2867 nfs_mmap(struct vop_mmap_args *ap)
2873 * fsync vnode op. Just call nfs_flush() with commit == 1.
2875 * nfs_fsync(struct vnode *a_vp, struct ucred * a_cred, int a_waitfor,
2876 * struct thread *a_td)
2880 nfs_fsync(struct vop_fsync_args *ap)
2882 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
2886 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2887 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2888 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2889 * set the buffer contains data that has already been written to the server
2890 * and which now needs a commit RPC.
2892 * If commit is 0 we only take one pass and only flush buffers containing new
2895 * If commit is 1 we take two passes, issuing a commit RPC in the second
2898 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
2899 * to completely flush all pending data.
2901 * Note that the RB_SCAN code properly handles the case where the
2902 * callback might block and directly or indirectly (another thread) cause
2903 * the RB tree to change.
2906 #ifndef NFS_COMMITBVECSIZ
2907 #define NFS_COMMITBVECSIZ 16
2910 struct nfs_flush_info {
2911 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
2918 struct buf *bvary[NFS_COMMITBVECSIZ];
2924 static int nfs_flush_bp(struct buf *bp, void *data);
2925 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
2928 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
2930 struct nfsnode *np = VTONFS(vp);
2931 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2932 struct nfs_flush_info info;
2935 bzero(&info, sizeof(info));
2938 info.waitfor = waitfor;
2939 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
2946 info.mode = NFI_FLUSHNEW;
2947 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2948 nfs_flush_bp, &info);
2951 * Take a second pass if committing and no error occured.
2952 * Clean up any left over collection (whether an error
2955 if (commit && error == 0) {
2956 info.mode = NFI_COMMIT;
2957 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2958 nfs_flush_bp, &info);
2960 error = nfs_flush_docommit(&info, error);
2964 * Wait for pending I/O to complete before checking whether
2965 * any further dirty buffers exist.
2967 while (waitfor == MNT_WAIT && vp->v_track_write.bk_active) {
2968 vp->v_track_write.bk_waitflag = 1;
2969 error = tsleep(&vp->v_track_write,
2970 info.slpflag, "nfsfsync", info.slptimeo);
2973 * We have to be able to break out if this
2974 * is an 'intr' mount.
2976 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
2982 * Since we do not process pending signals,
2983 * once we get a PCATCH our tsleep() will no
2984 * longer sleep, switch to a fixed timeout
2987 if (info.slpflag == PCATCH) {
2989 info.slptimeo = 2 * hz;
2996 * Loop if we are flushing synchronous as well as committing,
2997 * and dirty buffers are still present. Otherwise we might livelock.
2999 } while (waitfor == MNT_WAIT && commit &&
3000 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3003 * The callbacks have to return a negative error to terminate the
3010 * Deal with any error collection
3012 if (np->n_flag & NWRITEERR) {
3013 error = np->n_error;
3014 np->n_flag &= ~NWRITEERR;
3022 nfs_flush_bp(struct buf *bp, void *data)
3024 struct nfs_flush_info *info = data;
3029 switch(info->mode) {
3032 if (info->loops && info->waitfor == MNT_WAIT) {
3033 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3035 int lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3036 if (info->slpflag & PCATCH)
3037 lkflags |= LK_PCATCH;
3038 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3042 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3045 KKASSERT(bp->b_vp == info->vp);
3047 if ((bp->b_flags & B_DELWRI) == 0)
3048 panic("nfs_fsync: not dirty");
3049 if (bp->b_flags & B_NEEDCOMMIT) {
3056 bp->b_flags |= B_ASYNC;
3066 * Only process buffers in need of a commit which we can
3067 * immediately lock. This may prevent a buffer from being
3068 * committed, but the normal flush loop will block on the
3069 * same buffer so we shouldn't get into an endless loop.
3072 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3073 (B_DELWRI | B_NEEDCOMMIT) ||
3074 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
3079 KKASSERT(bp->b_vp == info->vp);
3083 * NOTE: storing the bp in the bvary[] basically sets
3084 * it up for a commit operation.
3086 * We must call vfs_busy_pages() now so the commit operation
3087 * is interlocked with user modifications to memory mapped
3090 * Note: to avoid loopback deadlocks, we do not
3091 * assign b_runningbufspace.
3093 bp->b_cmd = BUF_CMD_WRITE;
3094 vfs_busy_pages(bp->b_vp, bp);
3095 info->bvary[info->bvsize] = bp;
3096 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3097 if (info->bvsize == 0 || toff < info->beg_off)
3098 info->beg_off = toff;
3099 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3100 if (info->bvsize == 0 || toff > info->end_off)
3101 info->end_off = toff;
3103 if (info->bvsize == NFS_COMMITBVECSIZ) {
3104 error = nfs_flush_docommit(info, 0);
3105 KKASSERT(info->bvsize == 0);
3114 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3124 if (info->bvsize > 0) {
3126 * Commit data on the server, as required. Note that
3127 * nfs_commit will use the vnode's cred for the commit.
3128 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3130 bytes = info->end_off - info->beg_off;
3131 if (bytes > 0x40000000)
3136 retv = nfs_commit(vp, info->beg_off,
3137 (int)bytes, info->td);
3138 if (retv == NFSERR_STALEWRITEVERF)
3139 nfs_clearcommit(vp->v_mount);
3143 * Now, either mark the blocks I/O done or mark the
3144 * blocks dirty, depending on whether the commit
3147 for (i = 0; i < info->bvsize; ++i) {
3148 bp = info->bvary[i];
3149 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3152 * Error, leave B_DELWRI intact
3154 vfs_unbusy_pages(bp);
3155 bp->b_cmd = BUF_CMD_DONE;
3159 * Success, remove B_DELWRI ( bundirty() ).
3161 * b_dirtyoff/b_dirtyend seem to be NFS
3162 * specific. We should probably move that
3163 * into bundirty(). XXX
3165 * We are faking an I/O write, we have to
3166 * start the transaction in order to
3167 * immediately biodone() it.
3170 bp->b_flags |= B_ASYNC;
3172 bp->b_flags &= ~B_ERROR;
3173 bp->b_dirtyoff = bp->b_dirtyend = 0;
3175 biodone(&bp->b_bio1);
3184 * NFS advisory byte-level locks.
3185 * Currently unsupported.
3187 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3191 nfs_advlock(struct vop_advlock_args *ap)
3193 struct nfsnode *np = VTONFS(ap->a_vp);
3196 * The following kludge is to allow diskless support to work
3197 * until a real NFS lockd is implemented. Basically, just pretend
3198 * that this is a local lock.
3200 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3204 * Print out the contents of an nfsnode.
3206 * nfs_print(struct vnode *a_vp)
3209 nfs_print(struct vop_print_args *ap)
3211 struct vnode *vp = ap->a_vp;
3212 struct nfsnode *np = VTONFS(vp);
3214 kprintf("tag VT_NFS, fileid %ld fsid 0x%x",
3215 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3216 if (vp->v_type == VFIFO)
3223 * nfs special file access vnode op.
3224 * Essentially just get vattr and then imitate iaccess() since the device is
3225 * local to the client.
3227 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
3228 * struct thread *a_td)
3231 nfsspec_access(struct vop_access_args *ap)
3235 struct ucred *cred = ap->a_cred;
3236 struct vnode *vp = ap->a_vp;
3237 mode_t mode = ap->a_mode;
3243 * Disallow write attempts on filesystems mounted read-only;
3244 * unless the file is a socket, fifo, or a block or character
3245 * device resident on the filesystem.
3247 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3248 switch (vp->v_type) {
3258 * If you're the super-user,
3259 * you always get access.
3261 if (cred->cr_uid == 0)
3264 error = VOP_GETATTR(vp, vap);
3268 * Access check is based on only one of owner, group, public.
3269 * If not owner, then check group. If not a member of the
3270 * group, then check public access.
3272 if (cred->cr_uid != vap->va_uid) {
3274 gp = cred->cr_groups;
3275 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3276 if (vap->va_gid == *gp)
3282 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3287 * Read wrapper for special devices.
3289 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3290 * struct ucred *a_cred)
3293 nfsspec_read(struct vop_read_args *ap)
3295 struct nfsnode *np = VTONFS(ap->a_vp);
3301 getnanotime(&np->n_atim);
3302 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3306 * Write wrapper for special devices.
3308 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3309 * struct ucred *a_cred)
3312 nfsspec_write(struct vop_write_args *ap)
3314 struct nfsnode *np = VTONFS(ap->a_vp);
3320 getnanotime(&np->n_mtim);
3321 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3325 * Close wrapper for special devices.
3327 * Update the times on the nfsnode then do device close.
3329 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
3330 * struct thread *a_td)
3333 nfsspec_close(struct vop_close_args *ap)
3335 struct vnode *vp = ap->a_vp;
3336 struct nfsnode *np = VTONFS(vp);
3339 if (np->n_flag & (NACC | NUPD)) {
3341 if (vp->v_sysref.refcnt == 1 &&
3342 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3344 if (np->n_flag & NACC)
3345 vattr.va_atime = np->n_atim;
3346 if (np->n_flag & NUPD)
3347 vattr.va_mtime = np->n_mtim;
3348 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3351 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3355 * Read wrapper for fifos.
3357 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3358 * struct ucred *a_cred)
3361 nfsfifo_read(struct vop_read_args *ap)
3363 struct nfsnode *np = VTONFS(ap->a_vp);
3369 getnanotime(&np->n_atim);
3370 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3374 * Write wrapper for fifos.
3376 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3377 * struct ucred *a_cred)
3380 nfsfifo_write(struct vop_write_args *ap)
3382 struct nfsnode *np = VTONFS(ap->a_vp);
3388 getnanotime(&np->n_mtim);
3389 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3393 * Close wrapper for fifos.
3395 * Update the times on the nfsnode then do fifo close.
3397 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td)
3400 nfsfifo_close(struct vop_close_args *ap)
3402 struct vnode *vp = ap->a_vp;
3403 struct nfsnode *np = VTONFS(vp);
3407 if (np->n_flag & (NACC | NUPD)) {
3409 if (np->n_flag & NACC)
3411 if (np->n_flag & NUPD)
3414 if (vp->v_sysref.refcnt == 1 &&
3415 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3417 if (np->n_flag & NACC)
3418 vattr.va_atime = np->n_atim;
3419 if (np->n_flag & NUPD)
3420 vattr.va_mtime = np->n_mtim;
3421 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3424 return (VOCALL(&fifo_vnode_vops, &ap->a_head));