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 1.73 2007/08/08 00:12:51 swildner 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 NAMECACHE timeout");
237 static int nfspos_cache_timeout = NFS_MINATTRTIMO;
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
239 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
241 static int nfsv3_commit_on_close = 0;
242 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
243 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
245 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
246 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
248 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
249 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
252 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
253 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
254 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
256 nfs3_access_otw(struct vnode *vp, int wmode,
257 struct thread *td, struct ucred *cred)
261 int error = 0, attrflag;
263 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
264 caddr_t bpos, dpos, cp2;
268 struct nfsnode *np = VTONFS(vp);
270 nfsstats.rpccnt[NFSPROC_ACCESS]++;
271 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
273 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
274 *tl = txdr_unsigned(wmode);
275 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
276 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
278 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
279 rmode = fxdr_unsigned(u_int32_t, *tl);
281 np->n_modeuid = cred->cr_uid;
282 np->n_modestamp = mycpu->gd_time_seconds;
290 * nfs access vnode op.
291 * For nfs version 2, just return ok. File accesses may fail later.
292 * For nfs version 3, use the access rpc to check accessibility. If file modes
293 * are changed on the server, accesses might still fail later.
295 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
298 nfs_access(struct vop_access_args *ap)
300 struct vnode *vp = ap->a_vp;
301 thread_t td = curthread;
303 u_int32_t mode, wmode;
304 int v3 = NFS_ISV3(vp);
305 struct nfsnode *np = VTONFS(vp);
308 * Disallow write attempts on filesystems mounted read-only;
309 * unless the file is a socket, fifo, or a block or character
310 * device resident on the filesystem.
312 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
313 switch (vp->v_type) {
323 * For nfs v3, check to see if we have done this recently, and if
324 * so return our cached result instead of making an ACCESS call.
325 * If not, do an access rpc, otherwise you are stuck emulating
326 * ufs_access() locally using the vattr. This may not be correct,
327 * since the server may apply other access criteria such as
328 * client uid-->server uid mapping that we do not know about.
331 if (ap->a_mode & VREAD)
332 mode = NFSV3ACCESS_READ;
335 if (vp->v_type != VDIR) {
336 if (ap->a_mode & VWRITE)
337 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
338 if (ap->a_mode & VEXEC)
339 mode |= NFSV3ACCESS_EXECUTE;
341 if (ap->a_mode & VWRITE)
342 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
344 if (ap->a_mode & VEXEC)
345 mode |= NFSV3ACCESS_LOOKUP;
347 /* XXX safety belt, only make blanket request if caching */
348 if (nfsaccess_cache_timeout > 0) {
349 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
350 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
351 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
357 * Does our cached result allow us to give a definite yes to
360 if (np->n_modestamp &&
361 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
362 (ap->a_cred->cr_uid == np->n_modeuid) &&
363 ((np->n_mode & mode) == mode)) {
364 nfsstats.accesscache_hits++;
367 * Either a no, or a don't know. Go to the wire.
369 nfsstats.accesscache_misses++;
370 error = nfs3_access_otw(vp, wmode, td, ap->a_cred);
372 if ((np->n_mode & mode) != mode) {
378 if ((error = nfsspec_access(ap)) != 0)
382 * Attempt to prevent a mapped root from accessing a file
383 * which it shouldn't. We try to read a byte from the file
384 * if the user is root and the file is not zero length.
385 * After calling nfsspec_access, we should have the correct
388 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
389 && VTONFS(vp)->n_size > 0) {
396 auio.uio_iov = &aiov;
400 auio.uio_segflg = UIO_SYSSPACE;
401 auio.uio_rw = UIO_READ;
404 if (vp->v_type == VREG) {
405 error = nfs_readrpc(vp, &auio);
406 } else if (vp->v_type == VDIR) {
408 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
410 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
411 error = nfs_readdirrpc(vp, &auio);
413 } else if (vp->v_type == VLNK) {
414 error = nfs_readlinkrpc(vp, &auio);
421 * [re]record creds for reading and/or writing if access
422 * was granted. Assume the NFS server will grant read access
423 * for execute requests.
426 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) {
429 crfree(np->n_rucred);
430 np->n_rucred = ap->a_cred;
432 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
435 crfree(np->n_wucred);
436 np->n_wucred = ap->a_cred;
444 * Check to see if the type is ok
445 * and that deletion is not in progress.
446 * For paged in text files, you will need to flush the page cache
447 * if consistency is lost.
449 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
454 nfs_open(struct vop_open_args *ap)
456 struct vnode *vp = ap->a_vp;
457 struct nfsnode *np = VTONFS(vp);
461 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
463 kprintf("open eacces vtyp=%d\n",vp->v_type);
469 * Clear the attribute cache only if opening with write access. It
470 * is unclear if we should do this at all here, but we certainly
471 * should not clear the cache unconditionally simply because a file
474 if (ap->a_mode & FWRITE)
478 * For normal NFS, reconcile changes made locally verses
479 * changes made remotely. Note that VOP_GETATTR only goes
480 * to the wire if the cached attribute has timed out or been
483 * If local modifications have been made clear the attribute
484 * cache to force an attribute and modified time check. If
485 * GETATTR detects that the file has been changed by someone
486 * other then us it will set NRMODIFIED.
488 * If we are opening a directory and local changes have been
489 * made we have to invalidate the cache in order to ensure
490 * that we get the most up-to-date information from the
493 if (np->n_flag & NLMODIFIED) {
495 if (vp->v_type == VDIR) {
496 error = nfs_vinvalbuf(vp, V_SAVE, 1);
502 error = VOP_GETATTR(vp, &vattr);
505 if (np->n_flag & NRMODIFIED) {
506 if (vp->v_type == VDIR)
508 error = nfs_vinvalbuf(vp, V_SAVE, 1);
511 np->n_flag &= ~NRMODIFIED;
514 return (vop_stdopen(ap));
519 * What an NFS client should do upon close after writing is a debatable issue.
520 * Most NFS clients push delayed writes to the server upon close, basically for
522 * 1 - So that any write errors may be reported back to the client process
523 * doing the close system call. By far the two most likely errors are
524 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
525 * 2 - To put a worst case upper bound on cache inconsistency between
526 * multiple clients for the file.
527 * There is also a consistency problem for Version 2 of the protocol w.r.t.
528 * not being able to tell if other clients are writing a file concurrently,
529 * since there is no way of knowing if the changed modify time in the reply
530 * is only due to the write for this client.
531 * (NFS Version 3 provides weak cache consistency data in the reply that
532 * should be sufficient to detect and handle this case.)
534 * The current code does the following:
535 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
536 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
537 * or commit them (this satisfies 1 and 2 except for the
538 * case where the server crashes after this close but
539 * before the commit RPC, which is felt to be "good
540 * enough". Changing the last argument to nfs_flush() to
541 * a 1 would force a commit operation, if it is felt a
542 * commit is necessary now.
543 * for NQNFS - do nothing now, since 2 is dealt with via leases and
544 * 1 should be dealt with via an fsync() system call for
545 * cases where write errors are important.
547 * nfs_close(struct vnode *a_vp, int a_fflag)
551 nfs_close(struct vop_close_args *ap)
553 struct vnode *vp = ap->a_vp;
554 struct nfsnode *np = VTONFS(vp);
556 thread_t td = curthread;
558 if (vp->v_type == VREG) {
559 if (np->n_flag & NLMODIFIED) {
562 * Under NFSv3 we have dirty buffers to dispose of. We
563 * must flush them to the NFS server. We have the option
564 * of waiting all the way through the commit rpc or just
565 * waiting for the initial write. The default is to only
566 * wait through the initial write so the data is in the
567 * server's cache, which is roughly similar to the state
568 * a standard disk subsystem leaves the file in on close().
570 * We cannot clear the NLMODIFIED bit in np->n_flag due to
571 * potential races with other processes, and certainly
572 * cannot clear it if we don't commit.
574 int cm = nfsv3_commit_on_close ? 1 : 0;
575 error = nfs_flush(vp, MNT_WAIT, td, cm);
576 /* np->n_flag &= ~NLMODIFIED; */
578 error = nfs_vinvalbuf(vp, V_SAVE, 1);
582 if (np->n_flag & NWRITEERR) {
583 np->n_flag &= ~NWRITEERR;
592 * nfs getattr call from vfs.
594 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
597 nfs_getattr(struct vop_getattr_args *ap)
599 struct vnode *vp = ap->a_vp;
600 struct nfsnode *np = VTONFS(vp);
606 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
607 int v3 = NFS_ISV3(vp);
608 thread_t td = curthread;
611 * Update local times for special files.
613 if (np->n_flag & (NACC | NUPD))
616 * First look in the cache.
618 if (nfs_getattrcache(vp, ap->a_vap) == 0)
621 if (v3 && nfsaccess_cache_timeout > 0) {
622 nfsstats.accesscache_misses++;
623 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
624 if (nfs_getattrcache(vp, ap->a_vap) == 0)
628 nfsstats.rpccnt[NFSPROC_GETATTR]++;
629 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
631 nfsm_request(vp, NFSPROC_GETATTR, td, nfs_vpcred(vp, ND_CHECK));
633 nfsm_loadattr(vp, ap->a_vap);
643 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
646 nfs_setattr(struct vop_setattr_args *ap)
648 struct vnode *vp = ap->a_vp;
649 struct nfsnode *np = VTONFS(vp);
650 struct vattr *vap = ap->a_vap;
653 thread_t td = curthread;
660 * Setting of flags is not supported.
662 if (vap->va_flags != VNOVAL)
666 * Disallow write attempts if the filesystem is mounted read-only.
668 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
669 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
670 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
671 (vp->v_mount->mnt_flag & MNT_RDONLY))
673 if (vap->va_size != VNOVAL) {
674 switch (vp->v_type) {
681 if (vap->va_mtime.tv_sec == VNOVAL &&
682 vap->va_atime.tv_sec == VNOVAL &&
683 vap->va_mode == (mode_t)VNOVAL &&
684 vap->va_uid == (uid_t)VNOVAL &&
685 vap->va_gid == (gid_t)VNOVAL)
687 vap->va_size = VNOVAL;
691 * Disallow write attempts if the filesystem is
694 if (vp->v_mount->mnt_flag & MNT_RDONLY)
698 * This is nasty. The RPCs we send to flush pending
699 * data often return attribute information which is
700 * cached via a callback to nfs_loadattrcache(), which
701 * has the effect of changing our notion of the file
702 * size. Due to flushed appends and other operations
703 * the file size can be set to virtually anything,
704 * including values that do not match either the old
705 * or intended file size.
707 * When this condition is detected we must loop to
708 * try the operation again. Hopefully no more
709 * flushing is required on the loop so it works the
710 * second time around. THIS CASE ALMOST ALWAYS
715 error = nfs_meta_setsize(vp, td, vap->va_size);
717 if (np->n_flag & NLMODIFIED) {
718 if (vap->va_size == 0)
719 error = nfs_vinvalbuf(vp, 0, 1);
721 error = nfs_vinvalbuf(vp, V_SAVE, 1);
724 * note: this loop case almost always happens at
725 * least once per truncation.
727 if (error == 0 && np->n_size != vap->va_size)
729 np->n_vattr.va_size = vap->va_size;
732 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
733 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) &&
734 vp->v_type == VREG &&
735 (error = nfs_vinvalbuf(vp, V_SAVE, 1)) == EINTR
739 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
742 * Sanity check if a truncation was issued. This should only occur
743 * if multiple processes are racing on the same file.
745 if (error == 0 && vap->va_size != VNOVAL &&
746 np->n_size != vap->va_size) {
747 kprintf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size);
750 if (error && vap->va_size != VNOVAL) {
751 np->n_size = np->n_vattr.va_size = tsize;
752 vnode_pager_setsize(vp, np->n_size);
758 * Do an nfs setattr rpc.
761 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
762 struct ucred *cred, struct thread *td)
764 struct nfsv2_sattr *sp;
765 struct nfsnode *np = VTONFS(vp);
768 caddr_t bpos, dpos, cp2;
770 int error = 0, wccflag = NFSV3_WCCRATTR;
771 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
772 int v3 = NFS_ISV3(vp);
774 nfsstats.rpccnt[NFSPROC_SETATTR]++;
775 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
778 nfsm_v3attrbuild(vap, TRUE);
779 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
782 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
783 if (vap->va_mode == (mode_t)VNOVAL)
784 sp->sa_mode = nfs_xdrneg1;
786 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
787 if (vap->va_uid == (uid_t)VNOVAL)
788 sp->sa_uid = nfs_xdrneg1;
790 sp->sa_uid = txdr_unsigned(vap->va_uid);
791 if (vap->va_gid == (gid_t)VNOVAL)
792 sp->sa_gid = nfs_xdrneg1;
794 sp->sa_gid = txdr_unsigned(vap->va_gid);
795 sp->sa_size = txdr_unsigned(vap->va_size);
796 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
797 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
799 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
802 nfsm_wcc_data(vp, wccflag);
804 nfsm_loadattr(vp, (struct vattr *)0);
812 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
818 cache_setvp(nch, vp);
819 cache_settimeout(nch, nctimeout);
823 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
824 * nfs_lookup() until all remaining new api calls are implemented.
826 * Resolve a namecache entry. This function is passed a locked ncp and
827 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
830 nfs_nresolve(struct vop_nresolve_args *ap)
832 struct thread *td = curthread;
833 struct namecache *ncp;
844 /******NFSM MACROS********/
845 struct mbuf *mb, *mrep, *mreq, *mb2, *md;
846 caddr_t bpos, dpos, cp, cp2;
851 ncp = ap->a_nch->ncp;
853 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp);
854 dvp = ncp->nc_parent->nc_vp;
855 if ((error = vget(dvp, LK_SHARED)) != 0)
860 nfsstats.lookupcache_misses++;
861 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
863 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
864 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
866 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN);
867 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred);
870 * Cache negatve lookups to reduce NFS traffic, but use
871 * a fast timeout. Otherwise use a timeout of 1 tick.
872 * XXX we should add a namecache flag for no-caching
873 * to uncache the negative hit as soon as possible, but
874 * we cannot simply destroy the entry because it is used
875 * as a placeholder by the caller.
878 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
879 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
885 * Success, get the file handle, do various checks, and load
886 * post-operation data from the reply packet. Theoretically
887 * we should never be looking up "." so, theoretically, we
888 * should never get the same file handle as our directory. But
889 * we check anyway. XXX
891 * Note that no timeout is set for the positive cache hit. We
892 * assume, theoretically, that ESTALE returns will be dealt with
893 * properly to handle NFS races and in anycase we cannot depend
894 * on a timeout to deal with NFS open/create/excl issues so instead
895 * of a bad hack here the rest of the NFS client code needs to do
898 nfsm_getfh(fhp, fhsize, v3);
901 if (NFS_CMPFH(np, fhp, fhsize)) {
905 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
914 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK);
915 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
917 nfsm_loadattr(nvp, NULL);
919 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
933 * 'cached' nfs directory lookup
935 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
937 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
938 * struct componentname *a_cnp)
941 nfs_lookup(struct vop_old_lookup_args *ap)
943 struct componentname *cnp = ap->a_cnp;
944 struct vnode *dvp = ap->a_dvp;
945 struct vnode **vpp = ap->a_vpp;
946 int flags = cnp->cn_flags;
951 struct nfsmount *nmp;
952 caddr_t bpos, dpos, cp2;
953 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
957 int lockparent, wantparent, error = 0, attrflag, fhsize;
958 int v3 = NFS_ISV3(dvp);
961 * Read-only mount check and directory check.
964 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
965 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
968 if (dvp->v_type != VDIR)
972 * Look it up in the cache. Note that ENOENT is only returned if we
973 * previously entered a negative hit (see later on). The additional
974 * nfsneg_cache_timeout check causes previously cached results to
975 * be instantly ignored if the negative caching is turned off.
977 lockparent = flags & CNP_LOCKPARENT;
978 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
979 nmp = VFSTONFS(dvp->v_mount);
987 nfsstats.lookupcache_misses++;
988 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
989 len = cnp->cn_namelen;
990 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
991 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
993 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
994 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred);
996 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1000 nfsm_getfh(fhp, fhsize, v3);
1003 * Handle RENAME case...
1005 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1006 if (NFS_CMPFH(np, fhp, fhsize)) {
1010 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1017 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1018 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1020 nfsm_loadattr(newvp, (struct vattr *)0);
1025 cnp->cn_flags |= CNP_PDIRUNLOCK;
1030 if (flags & CNP_ISDOTDOT) {
1032 cnp->cn_flags |= CNP_PDIRUNLOCK;
1033 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1035 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1036 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1037 return (error); /* NOTE: return error from nget */
1041 error = vn_lock(dvp, LK_EXCLUSIVE);
1046 cnp->cn_flags |= CNP_PDIRUNLOCK;
1048 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1052 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1059 cnp->cn_flags |= CNP_PDIRUNLOCK;
1064 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1065 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1067 nfsm_loadattr(newvp, (struct vattr *)0);
1069 /* XXX MOVE TO nfs_nremove() */
1070 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1071 cnp->cn_nameiop != NAMEI_DELETE) {
1072 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1079 if (newvp != NULLVP) {
1083 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1084 cnp->cn_nameiop == NAMEI_RENAME) &&
1088 cnp->cn_flags |= CNP_PDIRUNLOCK;
1090 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1093 error = EJUSTRETURN;
1101 * Just call nfs_bioread() to do the work.
1103 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1104 * struct ucred *a_cred)
1107 nfs_read(struct vop_read_args *ap)
1109 struct vnode *vp = ap->a_vp;
1111 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1112 switch (vp->v_type) {
1114 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1125 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1128 nfs_readlink(struct vop_readlink_args *ap)
1130 struct vnode *vp = ap->a_vp;
1132 if (vp->v_type != VLNK)
1134 return (nfs_bioread(vp, ap->a_uio, 0));
1138 * Do a readlink rpc.
1139 * Called by nfs_doio() from below the buffer cache.
1142 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1147 caddr_t bpos, dpos, cp2;
1148 int error = 0, len, attrflag;
1149 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1150 int v3 = NFS_ISV3(vp);
1152 nfsstats.rpccnt[NFSPROC_READLINK]++;
1153 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1155 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK));
1157 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1159 nfsm_strsiz(len, NFS_MAXPATHLEN);
1160 if (len == NFS_MAXPATHLEN) {
1161 struct nfsnode *np = VTONFS(vp);
1162 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1165 nfsm_mtouio(uiop, len);
1177 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1182 caddr_t bpos, dpos, cp2;
1183 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1184 struct nfsmount *nmp;
1185 int error = 0, len, retlen, tsiz, eof, attrflag;
1186 int v3 = NFS_ISV3(vp);
1191 nmp = VFSTONFS(vp->v_mount);
1192 tsiz = uiop->uio_resid;
1193 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1196 nfsstats.rpccnt[NFSPROC_READ]++;
1197 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1198 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1200 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1202 txdr_hyper(uiop->uio_offset, tl);
1203 *(tl + 2) = txdr_unsigned(len);
1205 *tl++ = txdr_unsigned(uiop->uio_offset);
1206 *tl++ = txdr_unsigned(len);
1209 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ));
1211 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1216 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1217 eof = fxdr_unsigned(int, *(tl + 1));
1219 nfsm_loadattr(vp, (struct vattr *)0);
1220 nfsm_strsiz(retlen, nmp->nm_rsize);
1221 nfsm_mtouio(uiop, retlen);
1225 if (eof || retlen == 0) {
1228 } else if (retlen < len) {
1240 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1244 int32_t t1, t2, backup;
1245 caddr_t bpos, dpos, cp2;
1246 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1247 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1248 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1249 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1252 if (uiop->uio_iovcnt != 1)
1253 panic("nfs: writerpc iovcnt > 1");
1256 tsiz = uiop->uio_resid;
1257 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1260 nfsstats.rpccnt[NFSPROC_WRITE]++;
1261 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1262 nfsm_reqhead(vp, NFSPROC_WRITE,
1263 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1266 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1267 txdr_hyper(uiop->uio_offset, tl);
1269 *tl++ = txdr_unsigned(len);
1270 *tl++ = txdr_unsigned(*iomode);
1271 *tl = txdr_unsigned(len);
1275 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1276 /* Set both "begin" and "current" to non-garbage. */
1277 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1278 *tl++ = x; /* "begin offset" */
1279 *tl++ = x; /* "current offset" */
1280 x = txdr_unsigned(len);
1281 *tl++ = x; /* total to this offset */
1282 *tl = x; /* size of this write */
1284 nfsm_uiotom(uiop, len);
1285 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE));
1288 * The write RPC returns a before and after mtime. The
1289 * nfsm_wcc_data() macro checks the before n_mtime
1290 * against the before time and stores the after time
1291 * in the nfsnode's cached vattr and n_mtime field.
1292 * The NRMODIFIED bit will be set if the before
1293 * time did not match the original mtime.
1295 wccflag = NFSV3_WCCCHK;
1296 nfsm_wcc_data(vp, wccflag);
1298 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1299 + NFSX_V3WRITEVERF);
1300 rlen = fxdr_unsigned(int, *tl++);
1305 } else if (rlen < len) {
1306 backup = len - rlen;
1307 uiop->uio_iov->iov_base -= backup;
1308 uiop->uio_iov->iov_len += backup;
1309 uiop->uio_offset -= backup;
1310 uiop->uio_resid += backup;
1313 commit = fxdr_unsigned(int, *tl++);
1316 * Return the lowest committment level
1317 * obtained by any of the RPCs.
1319 if (committed == NFSV3WRITE_FILESYNC)
1321 else if (committed == NFSV3WRITE_DATASYNC &&
1322 commit == NFSV3WRITE_UNSTABLE)
1324 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1325 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1327 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1328 } else if (bcmp((caddr_t)tl,
1329 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1331 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1336 nfsm_loadattr(vp, (struct vattr *)0);
1344 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1345 committed = NFSV3WRITE_FILESYNC;
1346 *iomode = committed;
1348 uiop->uio_resid = tsiz;
1354 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1355 * mode set to specify the file type and the size field for rdev.
1358 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1361 struct nfsv2_sattr *sp;
1365 struct vnode *newvp = (struct vnode *)0;
1366 struct nfsnode *np = (struct nfsnode *)0;
1370 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1371 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1373 int v3 = NFS_ISV3(dvp);
1375 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1376 rmajor = txdr_unsigned(vap->va_rmajor);
1377 rminor = txdr_unsigned(vap->va_rminor);
1378 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1379 rmajor = nfs_xdrneg1;
1380 rminor = nfs_xdrneg1;
1382 return (EOPNOTSUPP);
1384 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1387 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1388 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1389 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1390 nfsm_fhtom(dvp, v3);
1391 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1393 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1394 *tl++ = vtonfsv3_type(vap->va_type);
1395 nfsm_v3attrbuild(vap, FALSE);
1396 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1397 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1398 *tl++ = txdr_unsigned(vap->va_rmajor);
1399 *tl = txdr_unsigned(vap->va_rminor);
1402 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1403 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1404 sp->sa_uid = nfs_xdrneg1;
1405 sp->sa_gid = nfs_xdrneg1;
1406 sp->sa_size = makeudev(rmajor, rminor);
1407 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1408 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1410 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred);
1412 nfsm_mtofh(dvp, newvp, v3, gotvp);
1416 newvp = (struct vnode *)0;
1418 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1419 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1425 nfsm_wcc_data(dvp, wccflag);
1434 VTONFS(dvp)->n_flag |= NLMODIFIED;
1436 VTONFS(dvp)->n_attrstamp = 0;
1442 * just call nfs_mknodrpc() to do the work.
1444 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1445 * struct componentname *a_cnp, struct vattr *a_vap)
1449 nfs_mknod(struct vop_old_mknod_args *ap)
1451 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1454 static u_long create_verf;
1456 * nfs file create call
1458 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1459 * struct componentname *a_cnp, struct vattr *a_vap)
1462 nfs_create(struct vop_old_create_args *ap)
1464 struct vnode *dvp = ap->a_dvp;
1465 struct vattr *vap = ap->a_vap;
1466 struct componentname *cnp = ap->a_cnp;
1467 struct nfsv2_sattr *sp;
1471 struct nfsnode *np = (struct nfsnode *)0;
1472 struct vnode *newvp = (struct vnode *)0;
1473 caddr_t bpos, dpos, cp2;
1474 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1475 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1477 int v3 = NFS_ISV3(dvp);
1480 * Oops, not for me..
1482 if (vap->va_type == VSOCK)
1483 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1485 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1488 if (vap->va_vaflags & VA_EXCLUSIVE)
1491 nfsstats.rpccnt[NFSPROC_CREATE]++;
1492 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1493 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1494 nfsm_fhtom(dvp, v3);
1495 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1497 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1498 if (fmode & O_EXCL) {
1499 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1500 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1502 if (!TAILQ_EMPTY(&in_ifaddrhead))
1503 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1506 *tl++ = create_verf;
1507 *tl = ++create_verf;
1509 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1510 nfsm_v3attrbuild(vap, FALSE);
1513 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1514 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1515 sp->sa_uid = nfs_xdrneg1;
1516 sp->sa_gid = nfs_xdrneg1;
1518 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1519 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1521 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred);
1523 nfsm_mtofh(dvp, newvp, v3, gotvp);
1527 newvp = (struct vnode *)0;
1529 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1530 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1536 nfsm_wcc_data(dvp, wccflag);
1540 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1546 } else if (v3 && (fmode & O_EXCL)) {
1548 * We are normally called with only a partially initialized
1549 * VAP. Since the NFSv3 spec says that server may use the
1550 * file attributes to store the verifier, the spec requires
1551 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1552 * in atime, but we can't really assume that all servers will
1553 * so we ensure that our SETATTR sets both atime and mtime.
1555 if (vap->va_mtime.tv_sec == VNOVAL)
1556 vfs_timestamp(&vap->va_mtime);
1557 if (vap->va_atime.tv_sec == VNOVAL)
1558 vap->va_atime = vap->va_mtime;
1559 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1563 * The new np may have enough info for access
1564 * checks, make sure rucred and wucred are
1565 * initialized for read and write rpc's.
1568 if (np->n_rucred == NULL)
1569 np->n_rucred = crhold(cnp->cn_cred);
1570 if (np->n_wucred == NULL)
1571 np->n_wucred = crhold(cnp->cn_cred);
1574 VTONFS(dvp)->n_flag |= NLMODIFIED;
1576 VTONFS(dvp)->n_attrstamp = 0;
1581 * nfs file remove call
1582 * To try and make nfs semantics closer to ufs semantics, a file that has
1583 * other processes using the vnode is renamed instead of removed and then
1584 * removed later on the last close.
1585 * - If v_sysref.refcnt > 1
1586 * If a rename is not already in the works
1587 * call nfs_sillyrename() to set it up
1591 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1592 * struct componentname *a_cnp)
1595 nfs_remove(struct vop_old_remove_args *ap)
1597 struct vnode *vp = ap->a_vp;
1598 struct vnode *dvp = ap->a_dvp;
1599 struct componentname *cnp = ap->a_cnp;
1600 struct nfsnode *np = VTONFS(vp);
1605 if (vp->v_sysref.refcnt < 1)
1606 panic("nfs_remove: bad v_sysref.refcnt");
1608 if (vp->v_type == VDIR)
1610 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1611 VOP_GETATTR(vp, &vattr) == 0 &&
1612 vattr.va_nlink > 1)) {
1614 * throw away biocache buffers, mainly to avoid
1615 * unnecessary delayed writes later.
1617 error = nfs_vinvalbuf(vp, 0, 1);
1620 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1621 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1623 * Kludge City: If the first reply to the remove rpc is lost..
1624 * the reply to the retransmitted request will be ENOENT
1625 * since the file was in fact removed
1626 * Therefore, we cheat and return success.
1628 if (error == ENOENT)
1630 } else if (!np->n_sillyrename) {
1631 error = nfs_sillyrename(dvp, vp, cnp);
1633 np->n_attrstamp = 0;
1638 * nfs file remove rpc called from nfs_inactive
1641 nfs_removeit(struct sillyrename *sp)
1643 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1648 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1651 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1652 struct ucred *cred, struct thread *td)
1657 caddr_t bpos, dpos, cp2;
1658 int error = 0, wccflag = NFSV3_WCCRATTR;
1659 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1660 int v3 = NFS_ISV3(dvp);
1662 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1663 nfsm_reqhead(dvp, NFSPROC_REMOVE,
1664 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1665 nfsm_fhtom(dvp, v3);
1666 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1667 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1669 nfsm_wcc_data(dvp, wccflag);
1672 VTONFS(dvp)->n_flag |= NLMODIFIED;
1674 VTONFS(dvp)->n_attrstamp = 0;
1679 * nfs file rename call
1681 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1682 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1683 * struct vnode *a_tvp, struct componentname *a_tcnp)
1686 nfs_rename(struct vop_old_rename_args *ap)
1688 struct vnode *fvp = ap->a_fvp;
1689 struct vnode *tvp = ap->a_tvp;
1690 struct vnode *fdvp = ap->a_fdvp;
1691 struct vnode *tdvp = ap->a_tdvp;
1692 struct componentname *tcnp = ap->a_tcnp;
1693 struct componentname *fcnp = ap->a_fcnp;
1696 /* Check for cross-device rename */
1697 if ((fvp->v_mount != tdvp->v_mount) ||
1698 (tvp && (fvp->v_mount != tvp->v_mount))) {
1704 * We have to flush B_DELWRI data prior to renaming
1705 * the file. If we don't, the delayed-write buffers
1706 * can be flushed out later after the file has gone stale
1707 * under NFSV3. NFSV2 does not have this problem because
1708 * ( as far as I can tell ) it flushes dirty buffers more
1712 VOP_FSYNC(fvp, MNT_WAIT);
1714 VOP_FSYNC(tvp, MNT_WAIT);
1717 * If the tvp exists and is in use, sillyrename it before doing the
1718 * rename of the new file over it.
1720 * XXX Can't sillyrename a directory.
1722 * We do not attempt to do any namecache purges in this old API
1723 * routine. The new API compat functions have access to the actual
1724 * namecache structures and will do it for us.
1726 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1727 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1734 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1735 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1748 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1750 if (error == ENOENT)
1756 * nfs file rename rpc called from nfs_remove() above
1759 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1760 struct sillyrename *sp)
1762 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1763 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1767 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1770 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1771 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1772 struct ucred *cred, struct thread *td)
1777 caddr_t bpos, dpos, cp2;
1778 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1779 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1780 int v3 = NFS_ISV3(fdvp);
1782 nfsstats.rpccnt[NFSPROC_RENAME]++;
1783 nfsm_reqhead(fdvp, NFSPROC_RENAME,
1784 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1785 nfsm_rndup(tnamelen));
1786 nfsm_fhtom(fdvp, v3);
1787 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1788 nfsm_fhtom(tdvp, v3);
1789 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1790 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1792 nfsm_wcc_data(fdvp, fwccflag);
1793 nfsm_wcc_data(tdvp, twccflag);
1797 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1798 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1800 VTONFS(fdvp)->n_attrstamp = 0;
1802 VTONFS(tdvp)->n_attrstamp = 0;
1807 * nfs hard link create call
1809 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1810 * struct componentname *a_cnp)
1813 nfs_link(struct vop_old_link_args *ap)
1815 struct vnode *vp = ap->a_vp;
1816 struct vnode *tdvp = ap->a_tdvp;
1817 struct componentname *cnp = ap->a_cnp;
1821 caddr_t bpos, dpos, cp2;
1822 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1823 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1826 if (vp->v_mount != tdvp->v_mount) {
1831 * Push all writes to the server, so that the attribute cache
1832 * doesn't get "out of sync" with the server.
1833 * XXX There should be a better way!
1835 VOP_FSYNC(vp, MNT_WAIT);
1838 nfsstats.rpccnt[NFSPROC_LINK]++;
1839 nfsm_reqhead(vp, NFSPROC_LINK,
1840 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1842 nfsm_fhtom(tdvp, v3);
1843 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1844 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred);
1846 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1847 nfsm_wcc_data(tdvp, wccflag);
1851 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1853 VTONFS(vp)->n_attrstamp = 0;
1855 VTONFS(tdvp)->n_attrstamp = 0;
1857 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1859 if (error == EEXIST)
1865 * nfs symbolic link create call
1867 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1868 * struct componentname *a_cnp, struct vattr *a_vap,
1872 nfs_symlink(struct vop_old_symlink_args *ap)
1874 struct vnode *dvp = ap->a_dvp;
1875 struct vattr *vap = ap->a_vap;
1876 struct componentname *cnp = ap->a_cnp;
1877 struct nfsv2_sattr *sp;
1881 caddr_t bpos, dpos, cp2;
1882 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1883 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1884 struct vnode *newvp = (struct vnode *)0;
1885 int v3 = NFS_ISV3(dvp);
1887 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1888 slen = strlen(ap->a_target);
1889 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1890 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1891 nfsm_fhtom(dvp, v3);
1892 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1894 nfsm_v3attrbuild(vap, FALSE);
1896 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1898 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1899 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1900 sp->sa_uid = nfs_xdrneg1;
1901 sp->sa_gid = nfs_xdrneg1;
1902 sp->sa_size = nfs_xdrneg1;
1903 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1904 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1908 * Issue the NFS request and get the rpc response.
1910 * Only NFSv3 responses returning an error of 0 actually return
1911 * a file handle that can be converted into newvp without having
1912 * to do an extra lookup rpc.
1914 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred);
1917 nfsm_mtofh(dvp, newvp, v3, gotvp);
1918 nfsm_wcc_data(dvp, wccflag);
1922 * out code jumps -> here, mrep is also freed.
1929 * If we get an EEXIST error, silently convert it to no-error
1930 * in case of an NFS retry.
1932 if (error == EEXIST)
1936 * If we do not have (or no longer have) an error, and we could
1937 * not extract the newvp from the response due to the request being
1938 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1939 * to obtain a newvp to return.
1941 if (error == 0 && newvp == NULL) {
1942 struct nfsnode *np = NULL;
1944 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1945 cnp->cn_cred, cnp->cn_td, &np);
1955 VTONFS(dvp)->n_flag |= NLMODIFIED;
1957 VTONFS(dvp)->n_attrstamp = 0;
1964 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1965 * struct componentname *a_cnp, struct vattr *a_vap)
1968 nfs_mkdir(struct vop_old_mkdir_args *ap)
1970 struct vnode *dvp = ap->a_dvp;
1971 struct vattr *vap = ap->a_vap;
1972 struct componentname *cnp = ap->a_cnp;
1973 struct nfsv2_sattr *sp;
1978 struct nfsnode *np = (struct nfsnode *)0;
1979 struct vnode *newvp = (struct vnode *)0;
1980 caddr_t bpos, dpos, cp2;
1981 int error = 0, wccflag = NFSV3_WCCRATTR;
1983 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1985 int v3 = NFS_ISV3(dvp);
1987 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1990 len = cnp->cn_namelen;
1991 nfsstats.rpccnt[NFSPROC_MKDIR]++;
1992 nfsm_reqhead(dvp, NFSPROC_MKDIR,
1993 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1994 nfsm_fhtom(dvp, v3);
1995 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1997 nfsm_v3attrbuild(vap, FALSE);
1999 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2000 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2001 sp->sa_uid = nfs_xdrneg1;
2002 sp->sa_gid = nfs_xdrneg1;
2003 sp->sa_size = nfs_xdrneg1;
2004 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2005 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2007 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred);
2009 nfsm_mtofh(dvp, newvp, v3, gotvp);
2011 nfsm_wcc_data(dvp, wccflag);
2014 VTONFS(dvp)->n_flag |= NLMODIFIED;
2016 VTONFS(dvp)->n_attrstamp = 0;
2018 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2019 * if we can succeed in looking up the directory.
2021 if (error == EEXIST || (!error && !gotvp)) {
2024 newvp = (struct vnode *)0;
2026 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2030 if (newvp->v_type != VDIR)
2043 * nfs remove directory call
2045 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2046 * struct componentname *a_cnp)
2049 nfs_rmdir(struct vop_old_rmdir_args *ap)
2051 struct vnode *vp = ap->a_vp;
2052 struct vnode *dvp = ap->a_dvp;
2053 struct componentname *cnp = ap->a_cnp;
2057 caddr_t bpos, dpos, cp2;
2058 int error = 0, wccflag = NFSV3_WCCRATTR;
2059 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2060 int v3 = NFS_ISV3(dvp);
2064 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2065 nfsm_reqhead(dvp, NFSPROC_RMDIR,
2066 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2067 nfsm_fhtom(dvp, v3);
2068 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2069 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred);
2071 nfsm_wcc_data(dvp, wccflag);
2074 VTONFS(dvp)->n_flag |= NLMODIFIED;
2076 VTONFS(dvp)->n_attrstamp = 0;
2078 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2080 if (error == ENOENT)
2088 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2091 nfs_readdir(struct vop_readdir_args *ap)
2093 struct vnode *vp = ap->a_vp;
2094 struct nfsnode *np = VTONFS(vp);
2095 struct uio *uio = ap->a_uio;
2099 if (vp->v_type != VDIR)
2102 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2106 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2107 * and then check that is still valid, or if this is an NQNFS mount
2108 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2109 * VOP_GETATTR() does not necessarily go to the wire.
2111 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2112 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2113 if (VOP_GETATTR(vp, &vattr) == 0 &&
2114 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2116 nfsstats.direofcache_hits++;
2122 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2123 * own cache coherency checks so we do not have to.
2125 tresid = uio->uio_resid;
2126 error = nfs_bioread(vp, uio, 0);
2128 if (!error && uio->uio_resid == tresid)
2129 nfsstats.direofcache_misses++;
2136 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2138 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2139 * offset/block and converts the nfs formatted directory entries for userland
2140 * consumption as well as deals with offsets into the middle of blocks.
2141 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2142 * be block-bounded. It must convert to cookies for the actual RPC.
2145 nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2148 struct nfs_dirent *dp = NULL;
2153 caddr_t bpos, dpos, cp2;
2154 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2156 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2157 struct nfsnode *dnp = VTONFS(vp);
2159 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2161 int v3 = NFS_ISV3(vp);
2164 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2165 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2166 panic("nfs readdirrpc bad uio");
2170 * If there is no cookie, assume directory was stale.
2172 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2176 return (NFSERR_BAD_COOKIE);
2178 * Loop around doing readdir rpc's of size nm_readdirsize
2179 * truncated to a multiple of DIRBLKSIZ.
2180 * The stopping criteria is EOF or buffer full.
2182 while (more_dirs && bigenough) {
2183 nfsstats.rpccnt[NFSPROC_READDIR]++;
2184 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2188 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2189 *tl++ = cookie.nfsuquad[0];
2190 *tl++ = cookie.nfsuquad[1];
2191 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2192 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2194 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2195 *tl++ = cookie.nfsuquad[0];
2197 *tl = txdr_unsigned(nmp->nm_readdirsize);
2198 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2200 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2202 nfsm_dissect(tl, u_int32_t *,
2204 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2205 dnp->n_cookieverf.nfsuquad[1] = *tl;
2211 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2212 more_dirs = fxdr_unsigned(int, *tl);
2214 /* loop thru the dir entries, converting them to std form */
2215 while (more_dirs && bigenough) {
2217 nfsm_dissect(tl, u_int32_t *,
2219 fileno = fxdr_hyper(tl);
2220 len = fxdr_unsigned(int, *(tl + 2));
2222 nfsm_dissect(tl, u_int32_t *,
2224 fileno = fxdr_unsigned(u_quad_t, *tl++);
2225 len = fxdr_unsigned(int, *tl);
2227 if (len <= 0 || len > NFS_MAXNAMLEN) {
2234 * len is the number of bytes in the path element
2235 * name, not including the \0 termination.
2237 * tlen is the number of bytes w have to reserve for
2238 * the path element name.
2240 tlen = nfsm_rndup(len);
2242 tlen += 4; /* To ensure null termination */
2245 * If the entry would cross a DIRBLKSIZ boundary,
2246 * extend the previous nfs_dirent to cover the
2249 left = DIRBLKSIZ - blksiz;
2250 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2251 dp->nfs_reclen += left;
2252 uiop->uio_iov->iov_base += left;
2253 uiop->uio_iov->iov_len -= left;
2254 uiop->uio_offset += left;
2255 uiop->uio_resid -= left;
2258 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2261 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2262 dp->nfs_ino = fileno;
2263 dp->nfs_namlen = len;
2264 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2265 dp->nfs_type = DT_UNKNOWN;
2266 blksiz += dp->nfs_reclen;
2267 if (blksiz == DIRBLKSIZ)
2269 uiop->uio_offset += sizeof(struct nfs_dirent);
2270 uiop->uio_resid -= sizeof(struct nfs_dirent);
2271 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent);
2272 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2273 nfsm_mtouio(uiop, len);
2276 * The uiop has advanced by nfs_dirent + len
2277 * but really needs to advance by
2280 cp = uiop->uio_iov->iov_base;
2282 *cp = '\0'; /* null terminate */
2283 uiop->uio_iov->iov_base += tlen;
2284 uiop->uio_iov->iov_len -= tlen;
2285 uiop->uio_offset += tlen;
2286 uiop->uio_resid -= tlen;
2289 * NFS strings must be rounded up (nfsm_myouio
2290 * handled that in the bigenough case).
2292 nfsm_adv(nfsm_rndup(len));
2295 nfsm_dissect(tl, u_int32_t *,
2298 nfsm_dissect(tl, u_int32_t *,
2303 * If we were able to accomodate the last entry,
2304 * get the cookie for the next one. Otherwise
2305 * hold-over the cookie for the one we were not
2306 * able to accomodate.
2309 cookie.nfsuquad[0] = *tl++;
2311 cookie.nfsuquad[1] = *tl++;
2317 more_dirs = fxdr_unsigned(int, *tl);
2320 * If at end of rpc data, get the eof boolean
2323 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2324 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2329 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2330 * by increasing d_reclen for the last record.
2333 left = DIRBLKSIZ - blksiz;
2334 dp->nfs_reclen += left;
2335 uiop->uio_iov->iov_base += left;
2336 uiop->uio_iov->iov_len -= left;
2337 uiop->uio_offset += left;
2338 uiop->uio_resid -= left;
2343 * We hit the end of the directory, update direofoffset.
2345 dnp->n_direofoffset = uiop->uio_offset;
2348 * There is more to go, insert the link cookie so the
2349 * next block can be read.
2351 if (uiop->uio_resid > 0)
2352 kprintf("EEK! readdirrpc resid > 0\n");
2353 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2361 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2364 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2367 struct nfs_dirent *dp;
2371 struct vnode *newvp;
2373 caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2374 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2376 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2377 struct nfsnode *dnp = VTONFS(vp), *np;
2380 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2381 int attrflag, fhsize;
2382 struct nchandle nch;
2383 struct nchandle dnch;
2384 struct nlcomponent nlc;
2390 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2391 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2392 panic("nfs readdirplusrpc bad uio");
2395 * Obtain the namecache record for the directory so we have something
2396 * to use as a basis for creating the entries. This function will
2397 * return a held (but not locked) ncp. The ncp may be disconnected
2398 * from the tree and cannot be used for upward traversals, and the
2399 * ncp may be unnamed. Note that other unrelated operations may
2400 * cause the ncp to be named at any time.
2402 cache_fromdvp(vp, NULL, 0, &dnch);
2403 bzero(&nlc, sizeof(nlc));
2407 * If there is no cookie, assume directory was stale.
2409 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2413 return (NFSERR_BAD_COOKIE);
2415 * Loop around doing readdir rpc's of size nm_readdirsize
2416 * truncated to a multiple of DIRBLKSIZ.
2417 * The stopping criteria is EOF or buffer full.
2419 while (more_dirs && bigenough) {
2420 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2421 nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2422 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2424 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2425 *tl++ = cookie.nfsuquad[0];
2426 *tl++ = cookie.nfsuquad[1];
2427 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2428 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2429 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2430 *tl = txdr_unsigned(nmp->nm_rsize);
2431 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2432 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2437 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2438 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2439 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2440 more_dirs = fxdr_unsigned(int, *tl);
2442 /* loop thru the dir entries, doctoring them to 4bsd form */
2443 while (more_dirs && bigenough) {
2444 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2445 fileno = fxdr_hyper(tl);
2446 len = fxdr_unsigned(int, *(tl + 2));
2447 if (len <= 0 || len > NFS_MAXNAMLEN) {
2452 tlen = nfsm_rndup(len);
2454 tlen += 4; /* To ensure null termination*/
2455 left = DIRBLKSIZ - blksiz;
2456 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2457 dp->nfs_reclen += left;
2458 uiop->uio_iov->iov_base += left;
2459 uiop->uio_iov->iov_len -= left;
2460 uiop->uio_offset += left;
2461 uiop->uio_resid -= left;
2464 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2467 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2468 dp->nfs_ino = fileno;
2469 dp->nfs_namlen = len;
2470 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2471 dp->nfs_type = DT_UNKNOWN;
2472 blksiz += dp->nfs_reclen;
2473 if (blksiz == DIRBLKSIZ)
2475 uiop->uio_offset += sizeof(struct nfs_dirent);
2476 uiop->uio_resid -= sizeof(struct nfs_dirent);
2477 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent);
2478 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2479 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2480 nlc.nlc_namelen = len;
2481 nfsm_mtouio(uiop, len);
2482 cp = uiop->uio_iov->iov_base;
2485 uiop->uio_iov->iov_base += tlen;
2486 uiop->uio_iov->iov_len -= tlen;
2487 uiop->uio_offset += tlen;
2488 uiop->uio_resid -= tlen;
2490 nfsm_adv(nfsm_rndup(len));
2491 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2493 cookie.nfsuquad[0] = *tl++;
2494 cookie.nfsuquad[1] = *tl++;
2499 * Since the attributes are before the file handle
2500 * (sigh), we must skip over the attributes and then
2501 * come back and get them.
2503 attrflag = fxdr_unsigned(int, *tl);
2507 nfsm_adv(NFSX_V3FATTR);
2508 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2509 doit = fxdr_unsigned(int, *tl);
2511 nfsm_getfh(fhp, fhsize, 1);
2512 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2517 error = nfs_nget(vp->v_mount, fhp,
2525 if (doit && bigenough) {
2530 nfsm_loadattr(newvp, (struct vattr *)0);
2534 IFTODT(VTTOIF(np->n_vattr.va_type));
2536 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2537 nlc.nlc_namelen, nlc.nlc_namelen,
2539 nch = cache_nlookup(&dnch, &nlc);
2540 cache_setunresolved(&nch);
2541 nfs_cache_setvp(&nch, newvp,
2542 nfspos_cache_timeout);
2545 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER"
2547 nlc.nlc_namelen, nlc.nlc_namelen,
2552 /* Just skip over the file handle */
2553 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2554 i = fxdr_unsigned(int, *tl);
2555 nfsm_adv(nfsm_rndup(i));
2557 if (newvp != NULLVP) {
2564 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2565 more_dirs = fxdr_unsigned(int, *tl);
2568 * If at end of rpc data, get the eof boolean
2571 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2572 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2577 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2578 * by increasing d_reclen for the last record.
2581 left = DIRBLKSIZ - blksiz;
2582 dp->nfs_reclen += left;
2583 uiop->uio_iov->iov_base += left;
2584 uiop->uio_iov->iov_len -= left;
2585 uiop->uio_offset += left;
2586 uiop->uio_resid -= left;
2590 * We are now either at the end of the directory or have filled the
2594 dnp->n_direofoffset = uiop->uio_offset;
2596 if (uiop->uio_resid > 0)
2597 kprintf("EEK! readdirplusrpc resid > 0\n");
2598 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2602 if (newvp != NULLVP) {
2615 * Silly rename. To make the NFS filesystem that is stateless look a little
2616 * more like the "ufs" a remove of an active vnode is translated to a rename
2617 * to a funny looking filename that is removed by nfs_inactive on the
2618 * nfsnode. There is the potential for another process on a different client
2619 * to create the same funny name between the nfs_lookitup() fails and the
2620 * nfs_rename() completes, but...
2623 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2625 struct sillyrename *sp;
2630 * We previously purged dvp instead of vp. I don't know why, it
2631 * completely destroys performance. We can't do it anyway with the
2632 * new VFS API since we would be breaking the namecache topology.
2634 cache_purge(vp); /* XXX */
2637 if (vp->v_type == VDIR)
2638 panic("nfs: sillyrename dir");
2640 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2641 M_NFSREQ, M_WAITOK);
2642 sp->s_cred = crdup(cnp->cn_cred);
2646 /* Fudge together a funny name */
2647 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td);
2649 /* Try lookitups until we get one that isn't there */
2650 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2651 cnp->cn_td, (struct nfsnode **)0) == 0) {
2653 if (sp->s_name[4] > 'z') {
2658 error = nfs_renameit(dvp, cnp, sp);
2661 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2663 np->n_sillyrename = sp;
2668 kfree((caddr_t)sp, M_NFSREQ);
2673 * Look up a file name and optionally either update the file handle or
2674 * allocate an nfsnode, depending on the value of npp.
2675 * npp == NULL --> just do the lookup
2676 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2678 * *npp != NULL --> update the file handle in the vnode
2681 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2682 struct thread *td, struct nfsnode **npp)
2687 struct vnode *newvp = (struct vnode *)0;
2688 struct nfsnode *np, *dnp = VTONFS(dvp);
2689 caddr_t bpos, dpos, cp2;
2690 int error = 0, fhlen, attrflag;
2691 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2693 int v3 = NFS_ISV3(dvp);
2695 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2696 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2697 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2698 nfsm_fhtom(dvp, v3);
2699 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2700 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2701 if (npp && !error) {
2702 nfsm_getfh(nfhp, fhlen, v3);
2705 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2706 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2707 np->n_fhp = &np->n_fh;
2708 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2709 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2710 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2711 np->n_fhsize = fhlen;
2713 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2717 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2725 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
2726 if (!attrflag && *npp == NULL) {
2735 nfsm_loadattr(newvp, (struct vattr *)0);
2739 if (npp && *npp == NULL) {
2754 * Nfs Version 3 commit rpc
2757 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2762 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2763 caddr_t bpos, dpos, cp2;
2764 int error = 0, wccflag = NFSV3_WCCRATTR;
2765 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2767 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2769 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2770 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2772 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2773 txdr_hyper(offset, tl);
2775 *tl = txdr_unsigned(cnt);
2776 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE));
2777 nfsm_wcc_data(vp, wccflag);
2779 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2780 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2781 NFSX_V3WRITEVERF)) {
2782 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2784 error = NFSERR_STALEWRITEVERF;
2794 * - make nfs_bmap() essentially a no-op that does no translation
2795 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2796 * (Maybe I could use the process's page mapping, but I was concerned that
2797 * Kernel Write might not be enabled and also figured copyout() would do
2798 * a lot more work than bcopy() and also it currently happens in the
2799 * context of the swapper process (2).
2801 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, struct vnode **a_vpp,
2802 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2805 nfs_bmap(struct vop_bmap_args *ap)
2807 struct vnode *vp = ap->a_vp;
2809 if (ap->a_vpp != NULL)
2811 if (ap->a_doffsetp != NULL)
2812 *ap->a_doffsetp = ap->a_loffset;
2813 if (ap->a_runp != NULL)
2815 if (ap->a_runb != NULL)
2823 * For async requests when nfsiod(s) are running, queue the request by
2824 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2828 nfs_strategy(struct vop_strategy_args *ap)
2830 struct bio *bio = ap->a_bio;
2832 struct buf *bp = bio->bio_buf;
2836 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2837 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
2838 KASSERT(BUF_REFCNT(bp) > 0,
2839 ("nfs_strategy: buffer %p not locked", bp));
2841 if (bp->b_flags & B_ASYNC)
2844 td = curthread; /* XXX */
2847 * We probably don't need to push an nbio any more since no
2848 * block conversion is required due to the use of 64 bit byte
2849 * offsets, but do it anyway.
2851 nbio = push_bio(bio);
2852 nbio->bio_offset = bio->bio_offset;
2855 * If the op is asynchronous and an i/o daemon is waiting
2856 * queue the request, wake it up and wait for completion
2857 * otherwise just do it ourselves.
2859 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(ap->a_vp, nbio, td))
2860 error = nfs_doio(ap->a_vp, nbio, td);
2867 * NB Currently unsupported.
2869 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
2873 nfs_mmap(struct vop_mmap_args *ap)
2879 * fsync vnode op. Just call nfs_flush() with commit == 1.
2881 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
2885 nfs_fsync(struct vop_fsync_args *ap)
2887 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
2891 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2892 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2893 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2894 * set the buffer contains data that has already been written to the server
2895 * and which now needs a commit RPC.
2897 * If commit is 0 we only take one pass and only flush buffers containing new
2900 * If commit is 1 we take two passes, issuing a commit RPC in the second
2903 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
2904 * to completely flush all pending data.
2906 * Note that the RB_SCAN code properly handles the case where the
2907 * callback might block and directly or indirectly (another thread) cause
2908 * the RB tree to change.
2911 #ifndef NFS_COMMITBVECSIZ
2912 #define NFS_COMMITBVECSIZ 16
2915 struct nfs_flush_info {
2916 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
2923 struct buf *bvary[NFS_COMMITBVECSIZ];
2929 static int nfs_flush_bp(struct buf *bp, void *data);
2930 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
2933 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
2935 struct nfsnode *np = VTONFS(vp);
2936 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2937 struct nfs_flush_info info;
2940 bzero(&info, sizeof(info));
2943 info.waitfor = waitfor;
2944 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
2951 info.mode = NFI_FLUSHNEW;
2952 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2953 nfs_flush_bp, &info);
2956 * Take a second pass if committing and no error occured.
2957 * Clean up any left over collection (whether an error
2960 if (commit && error == 0) {
2961 info.mode = NFI_COMMIT;
2962 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2963 nfs_flush_bp, &info);
2965 error = nfs_flush_docommit(&info, error);
2969 * Wait for pending I/O to complete before checking whether
2970 * any further dirty buffers exist.
2972 while (waitfor == MNT_WAIT && vp->v_track_write.bk_active) {
2973 vp->v_track_write.bk_waitflag = 1;
2974 error = tsleep(&vp->v_track_write,
2975 info.slpflag, "nfsfsync", info.slptimeo);
2978 * We have to be able to break out if this
2979 * is an 'intr' mount.
2981 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
2987 * Since we do not process pending signals,
2988 * once we get a PCATCH our tsleep() will no
2989 * longer sleep, switch to a fixed timeout
2992 if (info.slpflag == PCATCH) {
2994 info.slptimeo = 2 * hz;
3001 * Loop if we are flushing synchronous as well as committing,
3002 * and dirty buffers are still present. Otherwise we might livelock.
3004 } while (waitfor == MNT_WAIT && commit &&
3005 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3008 * The callbacks have to return a negative error to terminate the
3015 * Deal with any error collection
3017 if (np->n_flag & NWRITEERR) {
3018 error = np->n_error;
3019 np->n_flag &= ~NWRITEERR;
3027 nfs_flush_bp(struct buf *bp, void *data)
3029 struct nfs_flush_info *info = data;
3034 switch(info->mode) {
3037 if (info->loops && info->waitfor == MNT_WAIT) {
3038 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3040 int lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3041 if (info->slpflag & PCATCH)
3042 lkflags |= LK_PCATCH;
3043 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3047 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3050 KKASSERT(bp->b_vp == info->vp);
3052 if ((bp->b_flags & B_DELWRI) == 0)
3053 panic("nfs_fsync: not dirty");
3054 if (bp->b_flags & B_NEEDCOMMIT) {
3061 bp->b_flags |= B_ASYNC;
3071 * Only process buffers in need of a commit which we can
3072 * immediately lock. This may prevent a buffer from being
3073 * committed, but the normal flush loop will block on the
3074 * same buffer so we shouldn't get into an endless loop.
3077 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3078 (B_DELWRI | B_NEEDCOMMIT) ||
3079 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
3084 KKASSERT(bp->b_vp == info->vp);
3088 * NOTE: storing the bp in the bvary[] basically sets
3089 * it up for a commit operation.
3091 * We must call vfs_busy_pages() now so the commit operation
3092 * is interlocked with user modifications to memory mapped
3095 * Note: to avoid loopback deadlocks, we do not
3096 * assign b_runningbufspace.
3098 bp->b_cmd = BUF_CMD_WRITE;
3099 vfs_busy_pages(bp->b_vp, bp);
3100 info->bvary[info->bvsize] = bp;
3101 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3102 if (info->bvsize == 0 || toff < info->beg_off)
3103 info->beg_off = toff;
3104 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3105 if (info->bvsize == 0 || toff > info->end_off)
3106 info->end_off = toff;
3108 if (info->bvsize == NFS_COMMITBVECSIZ) {
3109 error = nfs_flush_docommit(info, 0);
3110 KKASSERT(info->bvsize == 0);
3119 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3129 if (info->bvsize > 0) {
3131 * Commit data on the server, as required. Note that
3132 * nfs_commit will use the vnode's cred for the commit.
3133 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3135 bytes = info->end_off - info->beg_off;
3136 if (bytes > 0x40000000)
3141 retv = nfs_commit(vp, info->beg_off,
3142 (int)bytes, info->td);
3143 if (retv == NFSERR_STALEWRITEVERF)
3144 nfs_clearcommit(vp->v_mount);
3148 * Now, either mark the blocks I/O done or mark the
3149 * blocks dirty, depending on whether the commit
3152 for (i = 0; i < info->bvsize; ++i) {
3153 bp = info->bvary[i];
3154 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3157 * Error, leave B_DELWRI intact
3159 vfs_unbusy_pages(bp);
3160 bp->b_cmd = BUF_CMD_DONE;
3164 * Success, remove B_DELWRI ( bundirty() ).
3166 * b_dirtyoff/b_dirtyend seem to be NFS
3167 * specific. We should probably move that
3168 * into bundirty(). XXX
3170 * We are faking an I/O write, we have to
3171 * start the transaction in order to
3172 * immediately biodone() it.
3175 bp->b_flags |= B_ASYNC;
3177 bp->b_flags &= ~B_ERROR;
3178 bp->b_dirtyoff = bp->b_dirtyend = 0;
3180 biodone(&bp->b_bio1);
3189 * NFS advisory byte-level locks.
3190 * Currently unsupported.
3192 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3196 nfs_advlock(struct vop_advlock_args *ap)
3198 struct nfsnode *np = VTONFS(ap->a_vp);
3201 * The following kludge is to allow diskless support to work
3202 * until a real NFS lockd is implemented. Basically, just pretend
3203 * that this is a local lock.
3205 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3209 * Print out the contents of an nfsnode.
3211 * nfs_print(struct vnode *a_vp)
3214 nfs_print(struct vop_print_args *ap)
3216 struct vnode *vp = ap->a_vp;
3217 struct nfsnode *np = VTONFS(vp);
3219 kprintf("tag VT_NFS, fileid %ld fsid 0x%x",
3220 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3221 if (vp->v_type == VFIFO)
3228 * nfs special file access vnode op.
3229 * Essentially just get vattr and then imitate iaccess() since the device is
3230 * local to the client.
3232 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3235 nfsspec_access(struct vop_access_args *ap)
3239 struct ucred *cred = ap->a_cred;
3240 struct vnode *vp = ap->a_vp;
3241 mode_t mode = ap->a_mode;
3247 * Disallow write attempts on filesystems mounted read-only;
3248 * unless the file is a socket, fifo, or a block or character
3249 * device resident on the filesystem.
3251 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3252 switch (vp->v_type) {
3262 * If you're the super-user,
3263 * you always get access.
3265 if (cred->cr_uid == 0)
3268 error = VOP_GETATTR(vp, vap);
3272 * Access check is based on only one of owner, group, public.
3273 * If not owner, then check group. If not a member of the
3274 * group, then check public access.
3276 if (cred->cr_uid != vap->va_uid) {
3278 gp = cred->cr_groups;
3279 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3280 if (vap->va_gid == *gp)
3286 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3291 * Read wrapper for special devices.
3293 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3294 * struct ucred *a_cred)
3297 nfsspec_read(struct vop_read_args *ap)
3299 struct nfsnode *np = VTONFS(ap->a_vp);
3305 getnanotime(&np->n_atim);
3306 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3310 * Write wrapper for special devices.
3312 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3313 * struct ucred *a_cred)
3316 nfsspec_write(struct vop_write_args *ap)
3318 struct nfsnode *np = VTONFS(ap->a_vp);
3324 getnanotime(&np->n_mtim);
3325 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3329 * Close wrapper for special devices.
3331 * Update the times on the nfsnode then do device close.
3333 * nfsspec_close(struct vnode *a_vp, int a_fflag)
3336 nfsspec_close(struct vop_close_args *ap)
3338 struct vnode *vp = ap->a_vp;
3339 struct nfsnode *np = VTONFS(vp);
3342 if (np->n_flag & (NACC | NUPD)) {
3344 if (vp->v_sysref.refcnt == 1 &&
3345 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3347 if (np->n_flag & NACC)
3348 vattr.va_atime = np->n_atim;
3349 if (np->n_flag & NUPD)
3350 vattr.va_mtime = np->n_mtim;
3351 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3354 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3358 * Read wrapper for fifos.
3360 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3361 * struct ucred *a_cred)
3364 nfsfifo_read(struct vop_read_args *ap)
3366 struct nfsnode *np = VTONFS(ap->a_vp);
3372 getnanotime(&np->n_atim);
3373 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3377 * Write wrapper for fifos.
3379 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3380 * struct ucred *a_cred)
3383 nfsfifo_write(struct vop_write_args *ap)
3385 struct nfsnode *np = VTONFS(ap->a_vp);
3391 getnanotime(&np->n_mtim);
3392 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3396 * Close wrapper for fifos.
3398 * Update the times on the nfsnode then do fifo close.
3400 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3403 nfsfifo_close(struct vop_close_args *ap)
3405 struct vnode *vp = ap->a_vp;
3406 struct nfsnode *np = VTONFS(vp);
3410 if (np->n_flag & (NACC | NUPD)) {
3412 if (np->n_flag & NACC)
3414 if (np->n_flag & NUPD)
3417 if (vp->v_sysref.refcnt == 1 &&
3418 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3420 if (np->n_flag & NACC)
3421 vattr.va_atime = np->n_atim;
3422 if (np->n_flag & NUPD)
3423 vattr.va_mtime = np->n_mtim;
3424 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3427 return (VOCALL(&fifo_vnode_vops, &ap->a_head));