2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
37 * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $
38 * $DragonFly: src/sys/vfs/nfs/nfs_vnops.c,v 1.80 2008/10/18 01:13:54 dillon Exp $
43 * vnode op calls for Sun NFS version 2 and 3
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/resourcevar.h>
53 #include <sys/mount.h>
55 #include <sys/malloc.h>
57 #include <sys/namei.h>
58 #include <sys/nlookup.h>
59 #include <sys/socket.h>
60 #include <sys/vnode.h>
61 #include <sys/dirent.h>
62 #include <sys/fcntl.h>
63 #include <sys/lockf.h>
65 #include <sys/sysctl.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vm_zone.h>
74 #include <vfs/fifofs/fifo.h>
75 #include <vfs/ufs/dir.h>
85 #include "nfsm_subs.h"
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
91 #include <sys/thread2.h>
97 static int nfsspec_read (struct vop_read_args *);
98 static int nfsspec_write (struct vop_write_args *);
99 static int nfsfifo_read (struct vop_read_args *);
100 static int nfsfifo_write (struct vop_write_args *);
101 static int nfsspec_close (struct vop_close_args *);
102 static int nfsfifo_close (struct vop_close_args *);
103 #define nfs_poll vop_nopoll
104 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
105 static int nfs_lookup (struct vop_old_lookup_args *);
106 static int nfs_create (struct vop_old_create_args *);
107 static int nfs_mknod (struct vop_old_mknod_args *);
108 static int nfs_open (struct vop_open_args *);
109 static int nfs_close (struct vop_close_args *);
110 static int nfs_access (struct vop_access_args *);
111 static int nfs_getattr (struct vop_getattr_args *);
112 static int nfs_setattr (struct vop_setattr_args *);
113 static int nfs_read (struct vop_read_args *);
114 static int nfs_mmap (struct vop_mmap_args *);
115 static int nfs_fsync (struct vop_fsync_args *);
116 static int nfs_remove (struct vop_old_remove_args *);
117 static int nfs_link (struct vop_old_link_args *);
118 static int nfs_rename (struct vop_old_rename_args *);
119 static int nfs_mkdir (struct vop_old_mkdir_args *);
120 static int nfs_rmdir (struct vop_old_rmdir_args *);
121 static int nfs_symlink (struct vop_old_symlink_args *);
122 static int nfs_readdir (struct vop_readdir_args *);
123 static int nfs_bmap (struct vop_bmap_args *);
124 static int nfs_strategy (struct vop_strategy_args *);
125 static int nfs_lookitup (struct vnode *, const char *, int,
126 struct ucred *, struct thread *, struct nfsnode **);
127 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
128 static int nfsspec_access (struct vop_access_args *);
129 static int nfs_readlink (struct vop_readlink_args *);
130 static int nfs_print (struct vop_print_args *);
131 static int nfs_advlock (struct vop_advlock_args *);
133 static int nfs_nresolve (struct vop_nresolve_args *);
135 * Global vfs data structures for nfs
137 struct vop_ops nfsv2_vnode_vops = {
138 .vop_default = vop_defaultop,
139 .vop_access = nfs_access,
140 .vop_advlock = nfs_advlock,
141 .vop_bmap = nfs_bmap,
142 .vop_close = nfs_close,
143 .vop_old_create = nfs_create,
144 .vop_fsync = nfs_fsync,
145 .vop_getattr = nfs_getattr,
146 .vop_getpages = nfs_getpages,
147 .vop_putpages = nfs_putpages,
148 .vop_inactive = nfs_inactive,
149 .vop_old_link = nfs_link,
150 .vop_old_lookup = nfs_lookup,
151 .vop_old_mkdir = nfs_mkdir,
152 .vop_old_mknod = nfs_mknod,
153 .vop_mmap = nfs_mmap,
154 .vop_open = nfs_open,
155 .vop_poll = nfs_poll,
156 .vop_print = nfs_print,
157 .vop_read = nfs_read,
158 .vop_readdir = nfs_readdir,
159 .vop_readlink = nfs_readlink,
160 .vop_reclaim = nfs_reclaim,
161 .vop_old_remove = nfs_remove,
162 .vop_old_rename = nfs_rename,
163 .vop_old_rmdir = nfs_rmdir,
164 .vop_setattr = nfs_setattr,
165 .vop_strategy = nfs_strategy,
166 .vop_old_symlink = nfs_symlink,
167 .vop_write = nfs_write,
168 .vop_nresolve = nfs_nresolve
172 * Special device vnode ops
174 struct vop_ops nfsv2_spec_vops = {
175 .vop_default = spec_vnoperate,
176 .vop_access = nfsspec_access,
177 .vop_close = nfsspec_close,
178 .vop_fsync = nfs_fsync,
179 .vop_getattr = nfs_getattr,
180 .vop_inactive = nfs_inactive,
181 .vop_print = nfs_print,
182 .vop_read = nfsspec_read,
183 .vop_reclaim = nfs_reclaim,
184 .vop_setattr = nfs_setattr,
185 .vop_write = nfsspec_write
188 struct vop_ops nfsv2_fifo_vops = {
189 .vop_default = fifo_vnoperate,
190 .vop_access = nfsspec_access,
191 .vop_close = nfsfifo_close,
192 .vop_fsync = nfs_fsync,
193 .vop_getattr = nfs_getattr,
194 .vop_inactive = nfs_inactive,
195 .vop_print = nfs_print,
196 .vop_read = nfsfifo_read,
197 .vop_reclaim = nfs_reclaim,
198 .vop_setattr = nfs_setattr,
199 .vop_write = nfsfifo_write
202 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
203 struct componentname *cnp,
205 static int nfs_removerpc (struct vnode *dvp, const char *name,
207 struct ucred *cred, struct thread *td);
208 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
209 int fnamelen, struct vnode *tdvp,
210 const char *tnameptr, int tnamelen,
211 struct ucred *cred, struct thread *td);
212 static int nfs_renameit (struct vnode *sdvp,
213 struct componentname *scnp,
214 struct sillyrename *sp);
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];
224 SYSCTL_DECL(_vfs_nfs);
226 static int nfs_flush_on_rename = 1;
227 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
228 &nfs_flush_on_rename, 0, "flush fvp prior to rename");
229 static int nfs_flush_on_hlink = 0;
230 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
231 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
233 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
235 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
237 static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
239 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
241 static int nfspos_cache_timeout = NFS_MINATTRTIMO;
242 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
243 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
245 static int nfsv3_commit_on_close = 0;
246 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
247 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
249 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
250 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
252 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
253 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
256 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
257 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
258 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
260 nfs3_access_otw(struct vnode *vp, int wmode,
261 struct thread *td, struct ucred *cred)
265 int error = 0, attrflag;
267 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
268 caddr_t bpos, dpos, cp2;
272 struct nfsnode *np = VTONFS(vp);
274 nfsstats.rpccnt[NFSPROC_ACCESS]++;
275 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
277 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
278 *tl = txdr_unsigned(wmode);
279 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
280 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
282 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
283 rmode = fxdr_unsigned(u_int32_t, *tl);
285 np->n_modeuid = cred->cr_uid;
286 np->n_modestamp = mycpu->gd_time_seconds;
294 * nfs access vnode op.
295 * For nfs version 2, just return ok. File accesses may fail later.
296 * For nfs version 3, use the access rpc to check accessibility. If file modes
297 * are changed on the server, accesses might still fail later.
299 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
302 nfs_access(struct vop_access_args *ap)
304 struct vnode *vp = ap->a_vp;
305 thread_t td = curthread;
307 u_int32_t mode, wmode;
308 int v3 = NFS_ISV3(vp);
309 struct nfsnode *np = VTONFS(vp);
312 * Disallow write attempts on filesystems mounted read-only;
313 * unless the file is a socket, fifo, or a block or character
314 * device resident on the filesystem.
316 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
317 switch (vp->v_type) {
327 * For nfs v3, check to see if we have done this recently, and if
328 * so return our cached result instead of making an ACCESS call.
329 * If not, do an access rpc, otherwise you are stuck emulating
330 * ufs_access() locally using the vattr. This may not be correct,
331 * since the server may apply other access criteria such as
332 * client uid-->server uid mapping that we do not know about.
335 if (ap->a_mode & VREAD)
336 mode = NFSV3ACCESS_READ;
339 if (vp->v_type != VDIR) {
340 if (ap->a_mode & VWRITE)
341 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
342 if (ap->a_mode & VEXEC)
343 mode |= NFSV3ACCESS_EXECUTE;
345 if (ap->a_mode & VWRITE)
346 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
348 if (ap->a_mode & VEXEC)
349 mode |= NFSV3ACCESS_LOOKUP;
351 /* XXX safety belt, only make blanket request if caching */
352 if (nfsaccess_cache_timeout > 0) {
353 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
354 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
355 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
361 * Does our cached result allow us to give a definite yes to
364 if (np->n_modestamp &&
365 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
366 (ap->a_cred->cr_uid == np->n_modeuid) &&
367 ((np->n_mode & mode) == mode)) {
368 nfsstats.accesscache_hits++;
371 * Either a no, or a don't know. Go to the wire.
373 nfsstats.accesscache_misses++;
374 error = nfs3_access_otw(vp, wmode, td, ap->a_cred);
376 if ((np->n_mode & mode) != mode) {
382 if ((error = nfsspec_access(ap)) != 0)
386 * Attempt to prevent a mapped root from accessing a file
387 * which it shouldn't. We try to read a byte from the file
388 * if the user is root and the file is not zero length.
389 * After calling nfsspec_access, we should have the correct
392 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
393 && VTONFS(vp)->n_size > 0) {
400 auio.uio_iov = &aiov;
404 auio.uio_segflg = UIO_SYSSPACE;
405 auio.uio_rw = UIO_READ;
408 if (vp->v_type == VREG) {
409 error = nfs_readrpc(vp, &auio);
410 } else if (vp->v_type == VDIR) {
412 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
414 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
415 error = nfs_readdirrpc(vp, &auio);
417 } else if (vp->v_type == VLNK) {
418 error = nfs_readlinkrpc(vp, &auio);
425 * [re]record creds for reading and/or writing if access
426 * was granted. Assume the NFS server will grant read access
427 * for execute requests.
430 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) {
433 crfree(np->n_rucred);
434 np->n_rucred = ap->a_cred;
436 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
439 crfree(np->n_wucred);
440 np->n_wucred = ap->a_cred;
448 * Check to see if the type is ok
449 * and that deletion is not in progress.
450 * For paged in text files, you will need to flush the page cache
451 * if consistency is lost.
453 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
458 nfs_open(struct vop_open_args *ap)
460 struct vnode *vp = ap->a_vp;
461 struct nfsnode *np = VTONFS(vp);
465 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
467 kprintf("open eacces vtyp=%d\n",vp->v_type);
473 * Save valid creds for reading and writing for later RPCs.
475 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
478 crfree(np->n_rucred);
479 np->n_rucred = ap->a_cred;
481 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
484 crfree(np->n_wucred);
485 np->n_wucred = ap->a_cred;
489 * Clear the attribute cache only if opening with write access. It
490 * is unclear if we should do this at all here, but we certainly
491 * should not clear the cache unconditionally simply because a file
494 if (ap->a_mode & FWRITE)
498 * For normal NFS, reconcile changes made locally verses
499 * changes made remotely. Note that VOP_GETATTR only goes
500 * to the wire if the cached attribute has timed out or been
503 * If local modifications have been made clear the attribute
504 * cache to force an attribute and modified time check. If
505 * GETATTR detects that the file has been changed by someone
506 * other then us it will set NRMODIFIED.
508 * If we are opening a directory and local changes have been
509 * made we have to invalidate the cache in order to ensure
510 * that we get the most up-to-date information from the
513 if (np->n_flag & NLMODIFIED) {
515 if (vp->v_type == VDIR) {
516 error = nfs_vinvalbuf(vp, V_SAVE, 1);
522 error = VOP_GETATTR(vp, &vattr);
525 if (np->n_flag & NRMODIFIED) {
526 if (vp->v_type == VDIR)
528 error = nfs_vinvalbuf(vp, V_SAVE, 1);
531 np->n_flag &= ~NRMODIFIED;
534 return (vop_stdopen(ap));
539 * What an NFS client should do upon close after writing is a debatable issue.
540 * Most NFS clients push delayed writes to the server upon close, basically for
542 * 1 - So that any write errors may be reported back to the client process
543 * doing the close system call. By far the two most likely errors are
544 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
545 * 2 - To put a worst case upper bound on cache inconsistency between
546 * multiple clients for the file.
547 * There is also a consistency problem for Version 2 of the protocol w.r.t.
548 * not being able to tell if other clients are writing a file concurrently,
549 * since there is no way of knowing if the changed modify time in the reply
550 * is only due to the write for this client.
551 * (NFS Version 3 provides weak cache consistency data in the reply that
552 * should be sufficient to detect and handle this case.)
554 * The current code does the following:
555 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
556 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
557 * or commit them (this satisfies 1 and 2 except for the
558 * case where the server crashes after this close but
559 * before the commit RPC, which is felt to be "good
560 * enough". Changing the last argument to nfs_flush() to
561 * a 1 would force a commit operation, if it is felt a
562 * commit is necessary now.
563 * for NQNFS - do nothing now, since 2 is dealt with via leases and
564 * 1 should be dealt with via an fsync() system call for
565 * cases where write errors are important.
567 * nfs_close(struct vnode *a_vp, int a_fflag)
571 nfs_close(struct vop_close_args *ap)
573 struct vnode *vp = ap->a_vp;
574 struct nfsnode *np = VTONFS(vp);
576 thread_t td = curthread;
578 if (vp->v_type == VREG) {
579 if (np->n_flag & NLMODIFIED) {
582 * Under NFSv3 we have dirty buffers to dispose of. We
583 * must flush them to the NFS server. We have the option
584 * of waiting all the way through the commit rpc or just
585 * waiting for the initial write. The default is to only
586 * wait through the initial write so the data is in the
587 * server's cache, which is roughly similar to the state
588 * a standard disk subsystem leaves the file in on close().
590 * We cannot clear the NLMODIFIED bit in np->n_flag due to
591 * potential races with other processes, and certainly
592 * cannot clear it if we don't commit.
594 int cm = nfsv3_commit_on_close ? 1 : 0;
595 error = nfs_flush(vp, MNT_WAIT, td, cm);
596 /* np->n_flag &= ~NLMODIFIED; */
598 error = nfs_vinvalbuf(vp, V_SAVE, 1);
602 if (np->n_flag & NWRITEERR) {
603 np->n_flag &= ~NWRITEERR;
612 * nfs getattr call from vfs.
614 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
617 nfs_getattr(struct vop_getattr_args *ap)
619 struct vnode *vp = ap->a_vp;
620 struct nfsnode *np = VTONFS(vp);
626 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
627 int v3 = NFS_ISV3(vp);
628 thread_t td = curthread;
631 * Update local times for special files.
633 if (np->n_flag & (NACC | NUPD))
636 * First look in the cache.
638 if (nfs_getattrcache(vp, ap->a_vap) == 0)
641 if (v3 && nfsaccess_cache_timeout > 0) {
642 nfsstats.accesscache_misses++;
643 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
644 if (nfs_getattrcache(vp, ap->a_vap) == 0)
648 nfsstats.rpccnt[NFSPROC_GETATTR]++;
649 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
651 nfsm_request(vp, NFSPROC_GETATTR, td, nfs_vpcred(vp, ND_CHECK));
653 nfsm_loadattr(vp, ap->a_vap);
663 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
666 nfs_setattr(struct vop_setattr_args *ap)
668 struct vnode *vp = ap->a_vp;
669 struct nfsnode *np = VTONFS(vp);
670 struct vattr *vap = ap->a_vap;
673 thread_t td = curthread;
680 * Setting of flags is not supported.
682 if (vap->va_flags != VNOVAL)
686 * Disallow write attempts if the filesystem is mounted read-only.
688 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
689 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
690 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
691 (vp->v_mount->mnt_flag & MNT_RDONLY))
694 if (vap->va_size != VNOVAL) {
696 * truncation requested
698 switch (vp->v_type) {
705 if (vap->va_mtime.tv_sec == VNOVAL &&
706 vap->va_atime.tv_sec == VNOVAL &&
707 vap->va_mode == (mode_t)VNOVAL &&
708 vap->va_uid == (uid_t)VNOVAL &&
709 vap->va_gid == (gid_t)VNOVAL)
711 vap->va_size = VNOVAL;
715 * Disallow write attempts if the filesystem is
718 if (vp->v_mount->mnt_flag & MNT_RDONLY)
722 * This is nasty. The RPCs we send to flush pending
723 * data often return attribute information which is
724 * cached via a callback to nfs_loadattrcache(), which
725 * has the effect of changing our notion of the file
726 * size. Due to flushed appends and other operations
727 * the file size can be set to virtually anything,
728 * including values that do not match either the old
729 * or intended file size.
731 * When this condition is detected we must loop to
732 * try the operation again. Hopefully no more
733 * flushing is required on the loop so it works the
734 * second time around. THIS CASE ALMOST ALWAYS
739 error = nfs_meta_setsize(vp, td, vap->va_size);
741 if (np->n_flag & NLMODIFIED) {
742 if (vap->va_size == 0)
743 error = nfs_vinvalbuf(vp, 0, 1);
745 error = nfs_vinvalbuf(vp, V_SAVE, 1);
748 * note: this loop case almost always happens at
749 * least once per truncation.
751 if (error == 0 && np->n_size != vap->va_size)
753 np->n_vattr.va_size = vap->va_size;
756 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
758 * What to do. If we are modifying the mtime we lose
759 * mtime detection of changes made by the server or other
760 * clients. But programs like rsync/rdist/cpdup are going
761 * to call utimes a lot. We don't want to piecemeal sync.
763 * For now sync if any prior remote changes were detected,
764 * but allow us to lose track of remote changes made during
765 * the utimes operation.
767 if (np->n_flag & NRMODIFIED)
768 error = nfs_vinvalbuf(vp, V_SAVE, 1);
772 if (vap->va_mtime.tv_sec != VNOVAL) {
773 np->n_mtime = vap->va_mtime.tv_sec;
777 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
780 * Sanity check if a truncation was issued. This should only occur
781 * if multiple processes are racing on the same file.
783 if (error == 0 && vap->va_size != VNOVAL &&
784 np->n_size != vap->va_size) {
785 kprintf("NFS ftruncate: server disagrees on the file size: "
788 (long long)vap->va_size,
789 (long long)np->n_size);
792 if (error && vap->va_size != VNOVAL) {
793 np->n_size = np->n_vattr.va_size = tsize;
794 vnode_pager_setsize(vp, np->n_size);
800 * Do an nfs setattr rpc.
803 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
804 struct ucred *cred, struct thread *td)
806 struct nfsv2_sattr *sp;
807 struct nfsnode *np = VTONFS(vp);
810 caddr_t bpos, dpos, cp2;
812 int error = 0, wccflag = NFSV3_WCCRATTR;
813 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
814 int v3 = NFS_ISV3(vp);
816 nfsstats.rpccnt[NFSPROC_SETATTR]++;
817 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
820 nfsm_v3attrbuild(vap, TRUE);
821 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
824 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
825 if (vap->va_mode == (mode_t)VNOVAL)
826 sp->sa_mode = nfs_xdrneg1;
828 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
829 if (vap->va_uid == (uid_t)VNOVAL)
830 sp->sa_uid = nfs_xdrneg1;
832 sp->sa_uid = txdr_unsigned(vap->va_uid);
833 if (vap->va_gid == (gid_t)VNOVAL)
834 sp->sa_gid = nfs_xdrneg1;
836 sp->sa_gid = txdr_unsigned(vap->va_gid);
837 sp->sa_size = txdr_unsigned(vap->va_size);
838 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
839 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
841 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
844 nfsm_wcc_data(vp, wccflag);
846 nfsm_loadattr(vp, NULL);
854 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
860 cache_setvp(nch, vp);
861 cache_settimeout(nch, nctimeout);
865 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
866 * nfs_lookup() until all remaining new api calls are implemented.
868 * Resolve a namecache entry. This function is passed a locked ncp and
869 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
872 nfs_nresolve(struct vop_nresolve_args *ap)
874 struct thread *td = curthread;
875 struct namecache *ncp;
886 /******NFSM MACROS********/
887 struct mbuf *mb, *mrep, *mreq, *mb2, *md;
888 caddr_t bpos, dpos, cp, cp2;
895 if ((error = vget(dvp, LK_SHARED)) != 0)
900 nfsstats.lookupcache_misses++;
901 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
902 ncp = ap->a_nch->ncp;
904 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
905 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
907 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN);
908 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred);
911 * Cache negatve lookups to reduce NFS traffic, but use
912 * a fast timeout. Otherwise use a timeout of 1 tick.
913 * XXX we should add a namecache flag for no-caching
914 * to uncache the negative hit as soon as possible, but
915 * we cannot simply destroy the entry because it is used
916 * as a placeholder by the caller.
919 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
920 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
926 * Success, get the file handle, do various checks, and load
927 * post-operation data from the reply packet. Theoretically
928 * we should never be looking up "." so, theoretically, we
929 * should never get the same file handle as our directory. But
930 * we check anyway. XXX
932 * Note that no timeout is set for the positive cache hit. We
933 * assume, theoretically, that ESTALE returns will be dealt with
934 * properly to handle NFS races and in anycase we cannot depend
935 * on a timeout to deal with NFS open/create/excl issues so instead
936 * of a bad hack here the rest of the NFS client code needs to do
939 nfsm_getfh(fhp, fhsize, v3);
942 if (NFS_CMPFH(np, fhp, fhsize)) {
946 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
955 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK);
956 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
958 nfsm_loadattr(nvp, NULL);
960 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
974 * 'cached' nfs directory lookup
976 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
978 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
979 * struct componentname *a_cnp)
982 nfs_lookup(struct vop_old_lookup_args *ap)
984 struct componentname *cnp = ap->a_cnp;
985 struct vnode *dvp = ap->a_dvp;
986 struct vnode **vpp = ap->a_vpp;
987 int flags = cnp->cn_flags;
992 struct nfsmount *nmp;
993 caddr_t bpos, dpos, cp2;
994 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
998 int lockparent, wantparent, error = 0, attrflag, fhsize;
999 int v3 = NFS_ISV3(dvp);
1002 * Read-only mount check and directory check.
1005 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1006 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1009 if (dvp->v_type != VDIR)
1013 * Look it up in the cache. Note that ENOENT is only returned if we
1014 * previously entered a negative hit (see later on). The additional
1015 * nfsneg_cache_timeout check causes previously cached results to
1016 * be instantly ignored if the negative caching is turned off.
1018 lockparent = flags & CNP_LOCKPARENT;
1019 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1020 nmp = VFSTONFS(dvp->v_mount);
1028 nfsstats.lookupcache_misses++;
1029 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1030 len = cnp->cn_namelen;
1031 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1032 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1033 nfsm_fhtom(dvp, v3);
1034 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1035 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred);
1037 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1041 nfsm_getfh(fhp, fhsize, v3);
1044 * Handle RENAME case...
1046 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1047 if (NFS_CMPFH(np, fhp, fhsize)) {
1051 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1058 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1059 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1061 nfsm_loadattr(newvp, NULL);
1066 cnp->cn_flags |= CNP_PDIRUNLOCK;
1071 if (flags & CNP_ISDOTDOT) {
1073 cnp->cn_flags |= CNP_PDIRUNLOCK;
1074 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1076 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1077 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1078 return (error); /* NOTE: return error from nget */
1082 error = vn_lock(dvp, LK_EXCLUSIVE);
1087 cnp->cn_flags |= CNP_PDIRUNLOCK;
1089 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1093 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1100 cnp->cn_flags |= CNP_PDIRUNLOCK;
1105 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1106 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1108 nfsm_loadattr(newvp, NULL);
1110 /* XXX MOVE TO nfs_nremove() */
1111 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1112 cnp->cn_nameiop != NAMEI_DELETE) {
1113 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1120 if (newvp != NULLVP) {
1124 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1125 cnp->cn_nameiop == NAMEI_RENAME) &&
1129 cnp->cn_flags |= CNP_PDIRUNLOCK;
1131 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1134 error = EJUSTRETURN;
1142 * Just call nfs_bioread() to do the work.
1144 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1145 * struct ucred *a_cred)
1148 nfs_read(struct vop_read_args *ap)
1150 struct vnode *vp = ap->a_vp;
1152 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1158 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1161 nfs_readlink(struct vop_readlink_args *ap)
1163 struct vnode *vp = ap->a_vp;
1165 if (vp->v_type != VLNK)
1167 return (nfs_bioread(vp, ap->a_uio, 0));
1171 * Do a readlink rpc.
1172 * Called by nfs_doio() from below the buffer cache.
1175 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1180 caddr_t bpos, dpos, cp2;
1181 int error = 0, len, attrflag;
1182 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1183 int v3 = NFS_ISV3(vp);
1185 nfsstats.rpccnt[NFSPROC_READLINK]++;
1186 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1188 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK));
1190 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1192 nfsm_strsiz(len, NFS_MAXPATHLEN);
1193 if (len == NFS_MAXPATHLEN) {
1194 struct nfsnode *np = VTONFS(vp);
1195 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1198 nfsm_mtouio(uiop, len);
1210 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1215 caddr_t bpos, dpos, cp2;
1216 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1217 struct nfsmount *nmp;
1218 int error = 0, len, retlen, tsiz, eof, attrflag;
1219 int v3 = NFS_ISV3(vp);
1224 nmp = VFSTONFS(vp->v_mount);
1225 tsiz = uiop->uio_resid;
1226 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1229 nfsstats.rpccnt[NFSPROC_READ]++;
1230 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1231 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1233 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1235 txdr_hyper(uiop->uio_offset, tl);
1236 *(tl + 2) = txdr_unsigned(len);
1238 *tl++ = txdr_unsigned(uiop->uio_offset);
1239 *tl++ = txdr_unsigned(len);
1242 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ));
1244 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1249 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1250 eof = fxdr_unsigned(int, *(tl + 1));
1252 nfsm_loadattr(vp, NULL);
1253 nfsm_strsiz(retlen, nmp->nm_rsize);
1254 nfsm_mtouio(uiop, retlen);
1258 if (eof || retlen == 0) {
1261 } else if (retlen < len) {
1273 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1277 int32_t t1, t2, backup;
1278 caddr_t bpos, dpos, cp2;
1279 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1280 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1281 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1282 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1285 if (uiop->uio_iovcnt != 1)
1286 panic("nfs: writerpc iovcnt > 1");
1289 tsiz = uiop->uio_resid;
1290 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1293 nfsstats.rpccnt[NFSPROC_WRITE]++;
1294 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1295 nfsm_reqhead(vp, NFSPROC_WRITE,
1296 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1299 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1300 txdr_hyper(uiop->uio_offset, tl);
1302 *tl++ = txdr_unsigned(len);
1303 *tl++ = txdr_unsigned(*iomode);
1304 *tl = txdr_unsigned(len);
1308 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1309 /* Set both "begin" and "current" to non-garbage. */
1310 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1311 *tl++ = x; /* "begin offset" */
1312 *tl++ = x; /* "current offset" */
1313 x = txdr_unsigned(len);
1314 *tl++ = x; /* total to this offset */
1315 *tl = x; /* size of this write */
1317 nfsm_uiotom(uiop, len);
1318 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE));
1321 * The write RPC returns a before and after mtime. The
1322 * nfsm_wcc_data() macro checks the before n_mtime
1323 * against the before time and stores the after time
1324 * in the nfsnode's cached vattr and n_mtime field.
1325 * The NRMODIFIED bit will be set if the before
1326 * time did not match the original mtime.
1328 wccflag = NFSV3_WCCCHK;
1329 nfsm_wcc_data(vp, wccflag);
1331 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1332 + NFSX_V3WRITEVERF);
1333 rlen = fxdr_unsigned(int, *tl++);
1338 } else if (rlen < len) {
1339 backup = len - rlen;
1340 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1341 uiop->uio_iov->iov_len += backup;
1342 uiop->uio_offset -= backup;
1343 uiop->uio_resid += backup;
1346 commit = fxdr_unsigned(int, *tl++);
1349 * Return the lowest committment level
1350 * obtained by any of the RPCs.
1352 if (committed == NFSV3WRITE_FILESYNC)
1354 else if (committed == NFSV3WRITE_DATASYNC &&
1355 commit == NFSV3WRITE_UNSTABLE)
1357 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1358 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1360 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1361 } else if (bcmp((caddr_t)tl,
1362 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1364 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1369 nfsm_loadattr(vp, NULL);
1377 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1378 committed = NFSV3WRITE_FILESYNC;
1379 *iomode = committed;
1381 uiop->uio_resid = tsiz;
1387 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1388 * mode set to specify the file type and the size field for rdev.
1391 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1394 struct nfsv2_sattr *sp;
1398 struct vnode *newvp = NULL;
1399 struct nfsnode *np = NULL;
1403 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1404 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1406 int v3 = NFS_ISV3(dvp);
1408 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1409 rmajor = txdr_unsigned(vap->va_rmajor);
1410 rminor = txdr_unsigned(vap->va_rminor);
1411 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1412 rmajor = nfs_xdrneg1;
1413 rminor = nfs_xdrneg1;
1415 return (EOPNOTSUPP);
1417 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1420 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1421 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1422 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1423 nfsm_fhtom(dvp, v3);
1424 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1426 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1427 *tl++ = vtonfsv3_type(vap->va_type);
1428 nfsm_v3attrbuild(vap, FALSE);
1429 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1430 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1431 *tl++ = txdr_unsigned(vap->va_rmajor);
1432 *tl = txdr_unsigned(vap->va_rminor);
1435 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1436 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1437 sp->sa_uid = nfs_xdrneg1;
1438 sp->sa_gid = nfs_xdrneg1;
1439 sp->sa_size = makeudev(rmajor, rminor);
1440 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1441 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1443 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred);
1445 nfsm_mtofh(dvp, newvp, v3, gotvp);
1451 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1452 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1458 nfsm_wcc_data(dvp, wccflag);
1467 VTONFS(dvp)->n_flag |= NLMODIFIED;
1469 VTONFS(dvp)->n_attrstamp = 0;
1475 * just call nfs_mknodrpc() to do the work.
1477 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1478 * struct componentname *a_cnp, struct vattr *a_vap)
1482 nfs_mknod(struct vop_old_mknod_args *ap)
1484 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1487 static u_long create_verf;
1489 * nfs file create call
1491 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1492 * struct componentname *a_cnp, struct vattr *a_vap)
1495 nfs_create(struct vop_old_create_args *ap)
1497 struct vnode *dvp = ap->a_dvp;
1498 struct vattr *vap = ap->a_vap;
1499 struct componentname *cnp = ap->a_cnp;
1500 struct nfsv2_sattr *sp;
1504 struct nfsnode *np = NULL;
1505 struct vnode *newvp = NULL;
1506 caddr_t bpos, dpos, cp2;
1507 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1508 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1510 int v3 = NFS_ISV3(dvp);
1513 * Oops, not for me..
1515 if (vap->va_type == VSOCK)
1516 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1518 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1521 if (vap->va_vaflags & VA_EXCLUSIVE)
1524 nfsstats.rpccnt[NFSPROC_CREATE]++;
1525 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1526 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1527 nfsm_fhtom(dvp, v3);
1528 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1530 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1531 if (fmode & O_EXCL) {
1532 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1533 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1535 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1536 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1539 *tl++ = create_verf;
1540 *tl = ++create_verf;
1542 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1543 nfsm_v3attrbuild(vap, FALSE);
1546 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1547 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1548 sp->sa_uid = nfs_xdrneg1;
1549 sp->sa_gid = nfs_xdrneg1;
1551 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1552 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1554 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred);
1556 nfsm_mtofh(dvp, newvp, v3, gotvp);
1562 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1563 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1569 nfsm_wcc_data(dvp, wccflag);
1573 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1574 KKASSERT(newvp == NULL);
1578 } else if (v3 && (fmode & O_EXCL)) {
1580 * We are normally called with only a partially initialized
1581 * VAP. Since the NFSv3 spec says that server may use the
1582 * file attributes to store the verifier, the spec requires
1583 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1584 * in atime, but we can't really assume that all servers will
1585 * so we ensure that our SETATTR sets both atime and mtime.
1587 if (vap->va_mtime.tv_sec == VNOVAL)
1588 vfs_timestamp(&vap->va_mtime);
1589 if (vap->va_atime.tv_sec == VNOVAL)
1590 vap->va_atime = vap->va_mtime;
1591 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1595 * The new np may have enough info for access
1596 * checks, make sure rucred and wucred are
1597 * initialized for read and write rpc's.
1600 if (np->n_rucred == NULL)
1601 np->n_rucred = crhold(cnp->cn_cred);
1602 if (np->n_wucred == NULL)
1603 np->n_wucred = crhold(cnp->cn_cred);
1608 VTONFS(dvp)->n_flag |= NLMODIFIED;
1610 VTONFS(dvp)->n_attrstamp = 0;
1615 * nfs file remove call
1616 * To try and make nfs semantics closer to ufs semantics, a file that has
1617 * other processes using the vnode is renamed instead of removed and then
1618 * removed later on the last close.
1619 * - If v_sysref.refcnt > 1
1620 * If a rename is not already in the works
1621 * call nfs_sillyrename() to set it up
1625 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1626 * struct componentname *a_cnp)
1629 nfs_remove(struct vop_old_remove_args *ap)
1631 struct vnode *vp = ap->a_vp;
1632 struct vnode *dvp = ap->a_dvp;
1633 struct componentname *cnp = ap->a_cnp;
1634 struct nfsnode *np = VTONFS(vp);
1639 if (vp->v_sysref.refcnt < 1)
1640 panic("nfs_remove: bad v_sysref.refcnt");
1642 if (vp->v_type == VDIR)
1644 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1645 VOP_GETATTR(vp, &vattr) == 0 &&
1646 vattr.va_nlink > 1)) {
1648 * throw away biocache buffers, mainly to avoid
1649 * unnecessary delayed writes later.
1651 error = nfs_vinvalbuf(vp, 0, 1);
1654 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1655 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1657 * Kludge City: If the first reply to the remove rpc is lost..
1658 * the reply to the retransmitted request will be ENOENT
1659 * since the file was in fact removed
1660 * Therefore, we cheat and return success.
1662 if (error == ENOENT)
1664 } else if (!np->n_sillyrename) {
1665 error = nfs_sillyrename(dvp, vp, cnp);
1667 np->n_attrstamp = 0;
1672 * nfs file remove rpc called from nfs_inactive
1675 nfs_removeit(struct sillyrename *sp)
1677 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1682 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1685 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1686 struct ucred *cred, struct thread *td)
1691 caddr_t bpos, dpos, cp2;
1692 int error = 0, wccflag = NFSV3_WCCRATTR;
1693 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1694 int v3 = NFS_ISV3(dvp);
1696 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1697 nfsm_reqhead(dvp, NFSPROC_REMOVE,
1698 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1699 nfsm_fhtom(dvp, v3);
1700 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1701 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1703 nfsm_wcc_data(dvp, wccflag);
1706 VTONFS(dvp)->n_flag |= NLMODIFIED;
1708 VTONFS(dvp)->n_attrstamp = 0;
1713 * nfs file rename call
1715 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1716 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1717 * struct vnode *a_tvp, struct componentname *a_tcnp)
1720 nfs_rename(struct vop_old_rename_args *ap)
1722 struct vnode *fvp = ap->a_fvp;
1723 struct vnode *tvp = ap->a_tvp;
1724 struct vnode *fdvp = ap->a_fdvp;
1725 struct vnode *tdvp = ap->a_tdvp;
1726 struct componentname *tcnp = ap->a_tcnp;
1727 struct componentname *fcnp = ap->a_fcnp;
1730 /* Check for cross-device rename */
1731 if ((fvp->v_mount != tdvp->v_mount) ||
1732 (tvp && (fvp->v_mount != tvp->v_mount))) {
1738 * We shouldn't have to flush fvp on rename for most server-side
1739 * filesystems as the file handle should not change. Unfortunately
1740 * the inode for some filesystems (msdosfs) might be tied to the
1741 * file name or directory position so to be completely safe
1742 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1745 * We must flush tvp on rename because it might become stale on the
1746 * server after the rename.
1748 if (nfs_flush_on_rename)
1749 VOP_FSYNC(fvp, MNT_WAIT);
1751 VOP_FSYNC(tvp, MNT_WAIT);
1754 * If the tvp exists and is in use, sillyrename it before doing the
1755 * rename of the new file over it.
1757 * XXX Can't sillyrename a directory.
1759 * We do not attempt to do any namecache purges in this old API
1760 * routine. The new API compat functions have access to the actual
1761 * namecache structures and will do it for us.
1763 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1764 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1771 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1772 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1785 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1787 if (error == ENOENT)
1793 * nfs file rename rpc called from nfs_remove() above
1796 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1797 struct sillyrename *sp)
1799 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1800 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1804 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1807 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1808 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1809 struct ucred *cred, struct thread *td)
1814 caddr_t bpos, dpos, cp2;
1815 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1816 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1817 int v3 = NFS_ISV3(fdvp);
1819 nfsstats.rpccnt[NFSPROC_RENAME]++;
1820 nfsm_reqhead(fdvp, NFSPROC_RENAME,
1821 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1822 nfsm_rndup(tnamelen));
1823 nfsm_fhtom(fdvp, v3);
1824 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1825 nfsm_fhtom(tdvp, v3);
1826 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1827 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1829 nfsm_wcc_data(fdvp, fwccflag);
1830 nfsm_wcc_data(tdvp, twccflag);
1834 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1835 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1837 VTONFS(fdvp)->n_attrstamp = 0;
1839 VTONFS(tdvp)->n_attrstamp = 0;
1844 * nfs hard link create call
1846 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1847 * struct componentname *a_cnp)
1850 nfs_link(struct vop_old_link_args *ap)
1852 struct vnode *vp = ap->a_vp;
1853 struct vnode *tdvp = ap->a_tdvp;
1854 struct componentname *cnp = ap->a_cnp;
1858 caddr_t bpos, dpos, cp2;
1859 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1860 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1863 if (vp->v_mount != tdvp->v_mount) {
1868 * The attribute cache may get out of sync with the server on link.
1869 * Pushing writes to the server before handle was inherited from
1870 * long long ago and it is unclear if we still need to do this.
1873 if (nfs_flush_on_hlink)
1874 VOP_FSYNC(vp, MNT_WAIT);
1877 nfsstats.rpccnt[NFSPROC_LINK]++;
1878 nfsm_reqhead(vp, NFSPROC_LINK,
1879 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1881 nfsm_fhtom(tdvp, v3);
1882 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1883 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred);
1885 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1886 nfsm_wcc_data(tdvp, wccflag);
1890 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1892 VTONFS(vp)->n_attrstamp = 0;
1894 VTONFS(tdvp)->n_attrstamp = 0;
1896 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1898 if (error == EEXIST)
1904 * nfs symbolic link create call
1906 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1907 * struct componentname *a_cnp, struct vattr *a_vap,
1911 nfs_symlink(struct vop_old_symlink_args *ap)
1913 struct vnode *dvp = ap->a_dvp;
1914 struct vattr *vap = ap->a_vap;
1915 struct componentname *cnp = ap->a_cnp;
1916 struct nfsv2_sattr *sp;
1920 caddr_t bpos, dpos, cp2;
1921 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1922 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1923 struct vnode *newvp = NULL;
1924 int v3 = NFS_ISV3(dvp);
1926 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1927 slen = strlen(ap->a_target);
1928 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1929 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1930 nfsm_fhtom(dvp, v3);
1931 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1933 nfsm_v3attrbuild(vap, FALSE);
1935 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1937 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1938 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1939 sp->sa_uid = nfs_xdrneg1;
1940 sp->sa_gid = nfs_xdrneg1;
1941 sp->sa_size = nfs_xdrneg1;
1942 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1943 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1947 * Issue the NFS request and get the rpc response.
1949 * Only NFSv3 responses returning an error of 0 actually return
1950 * a file handle that can be converted into newvp without having
1951 * to do an extra lookup rpc.
1953 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred);
1956 nfsm_mtofh(dvp, newvp, v3, gotvp);
1957 nfsm_wcc_data(dvp, wccflag);
1961 * out code jumps -> here, mrep is also freed.
1968 * If we get an EEXIST error, silently convert it to no-error
1969 * in case of an NFS retry.
1971 if (error == EEXIST)
1975 * If we do not have (or no longer have) an error, and we could
1976 * not extract the newvp from the response due to the request being
1977 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1978 * to obtain a newvp to return.
1980 if (error == 0 && newvp == NULL) {
1981 struct nfsnode *np = NULL;
1983 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1984 cnp->cn_cred, cnp->cn_td, &np);
1994 VTONFS(dvp)->n_flag |= NLMODIFIED;
1996 VTONFS(dvp)->n_attrstamp = 0;
2003 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2004 * struct componentname *a_cnp, struct vattr *a_vap)
2007 nfs_mkdir(struct vop_old_mkdir_args *ap)
2009 struct vnode *dvp = ap->a_dvp;
2010 struct vattr *vap = ap->a_vap;
2011 struct componentname *cnp = ap->a_cnp;
2012 struct nfsv2_sattr *sp;
2017 struct nfsnode *np = NULL;
2018 struct vnode *newvp = NULL;
2019 caddr_t bpos, dpos, cp2;
2020 int error = 0, wccflag = NFSV3_WCCRATTR;
2022 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2024 int v3 = NFS_ISV3(dvp);
2026 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2029 len = cnp->cn_namelen;
2030 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2031 nfsm_reqhead(dvp, NFSPROC_MKDIR,
2032 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2033 nfsm_fhtom(dvp, v3);
2034 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2036 nfsm_v3attrbuild(vap, FALSE);
2038 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2039 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2040 sp->sa_uid = nfs_xdrneg1;
2041 sp->sa_gid = nfs_xdrneg1;
2042 sp->sa_size = nfs_xdrneg1;
2043 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2044 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2046 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred);
2048 nfsm_mtofh(dvp, newvp, v3, gotvp);
2050 nfsm_wcc_data(dvp, wccflag);
2053 VTONFS(dvp)->n_flag |= NLMODIFIED;
2055 VTONFS(dvp)->n_attrstamp = 0;
2057 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2058 * if we can succeed in looking up the directory.
2060 if (error == EEXIST || (!error && !gotvp)) {
2065 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2069 if (newvp->v_type != VDIR)
2082 * nfs remove directory call
2084 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2085 * struct componentname *a_cnp)
2088 nfs_rmdir(struct vop_old_rmdir_args *ap)
2090 struct vnode *vp = ap->a_vp;
2091 struct vnode *dvp = ap->a_dvp;
2092 struct componentname *cnp = ap->a_cnp;
2096 caddr_t bpos, dpos, cp2;
2097 int error = 0, wccflag = NFSV3_WCCRATTR;
2098 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2099 int v3 = NFS_ISV3(dvp);
2103 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2104 nfsm_reqhead(dvp, NFSPROC_RMDIR,
2105 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2106 nfsm_fhtom(dvp, v3);
2107 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2108 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred);
2110 nfsm_wcc_data(dvp, wccflag);
2113 VTONFS(dvp)->n_flag |= NLMODIFIED;
2115 VTONFS(dvp)->n_attrstamp = 0;
2117 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2119 if (error == ENOENT)
2127 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2130 nfs_readdir(struct vop_readdir_args *ap)
2132 struct vnode *vp = ap->a_vp;
2133 struct nfsnode *np = VTONFS(vp);
2134 struct uio *uio = ap->a_uio;
2138 if (vp->v_type != VDIR)
2141 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2145 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2146 * and then check that is still valid, or if this is an NQNFS mount
2147 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2148 * VOP_GETATTR() does not necessarily go to the wire.
2150 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2151 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2152 if (VOP_GETATTR(vp, &vattr) == 0 &&
2153 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2155 nfsstats.direofcache_hits++;
2161 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2162 * own cache coherency checks so we do not have to.
2164 tresid = uio->uio_resid;
2165 error = nfs_bioread(vp, uio, 0);
2167 if (!error && uio->uio_resid == tresid)
2168 nfsstats.direofcache_misses++;
2175 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2177 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2178 * offset/block and converts the nfs formatted directory entries for userland
2179 * consumption as well as deals with offsets into the middle of blocks.
2180 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2181 * be block-bounded. It must convert to cookies for the actual RPC.
2184 nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2187 struct nfs_dirent *dp = NULL;
2192 caddr_t bpos, dpos, cp2;
2193 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2195 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2196 struct nfsnode *dnp = VTONFS(vp);
2198 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2200 int v3 = NFS_ISV3(vp);
2203 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2204 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2205 panic("nfs readdirrpc bad uio");
2209 * If there is no cookie, assume directory was stale.
2211 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2215 return (NFSERR_BAD_COOKIE);
2217 * Loop around doing readdir rpc's of size nm_readdirsize
2218 * truncated to a multiple of DIRBLKSIZ.
2219 * The stopping criteria is EOF or buffer full.
2221 while (more_dirs && bigenough) {
2222 nfsstats.rpccnt[NFSPROC_READDIR]++;
2223 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2227 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2228 *tl++ = cookie.nfsuquad[0];
2229 *tl++ = cookie.nfsuquad[1];
2230 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2231 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2233 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2234 *tl++ = cookie.nfsuquad[0];
2236 *tl = txdr_unsigned(nmp->nm_readdirsize);
2237 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2239 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2241 nfsm_dissect(tl, u_int32_t *,
2243 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2244 dnp->n_cookieverf.nfsuquad[1] = *tl;
2250 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2251 more_dirs = fxdr_unsigned(int, *tl);
2253 /* loop thru the dir entries, converting them to std form */
2254 while (more_dirs && bigenough) {
2256 nfsm_dissect(tl, u_int32_t *,
2258 fileno = fxdr_hyper(tl);
2259 len = fxdr_unsigned(int, *(tl + 2));
2261 nfsm_dissect(tl, u_int32_t *,
2263 fileno = fxdr_unsigned(u_quad_t, *tl++);
2264 len = fxdr_unsigned(int, *tl);
2266 if (len <= 0 || len > NFS_MAXNAMLEN) {
2273 * len is the number of bytes in the path element
2274 * name, not including the \0 termination.
2276 * tlen is the number of bytes w have to reserve for
2277 * the path element name.
2279 tlen = nfsm_rndup(len);
2281 tlen += 4; /* To ensure null termination */
2284 * If the entry would cross a DIRBLKSIZ boundary,
2285 * extend the previous nfs_dirent to cover the
2288 left = DIRBLKSIZ - blksiz;
2289 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2290 dp->nfs_reclen += left;
2291 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2292 uiop->uio_iov->iov_len -= left;
2293 uiop->uio_offset += left;
2294 uiop->uio_resid -= left;
2297 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2300 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2301 dp->nfs_ino = fileno;
2302 dp->nfs_namlen = len;
2303 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2304 dp->nfs_type = DT_UNKNOWN;
2305 blksiz += dp->nfs_reclen;
2306 if (blksiz == DIRBLKSIZ)
2308 uiop->uio_offset += sizeof(struct nfs_dirent);
2309 uiop->uio_resid -= sizeof(struct nfs_dirent);
2310 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2311 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2312 nfsm_mtouio(uiop, len);
2315 * The uiop has advanced by nfs_dirent + len
2316 * but really needs to advance by
2319 cp = uiop->uio_iov->iov_base;
2321 *cp = '\0'; /* null terminate */
2322 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2323 uiop->uio_iov->iov_len -= tlen;
2324 uiop->uio_offset += tlen;
2325 uiop->uio_resid -= tlen;
2328 * NFS strings must be rounded up (nfsm_myouio
2329 * handled that in the bigenough case).
2331 nfsm_adv(nfsm_rndup(len));
2334 nfsm_dissect(tl, u_int32_t *,
2337 nfsm_dissect(tl, u_int32_t *,
2342 * If we were able to accomodate the last entry,
2343 * get the cookie for the next one. Otherwise
2344 * hold-over the cookie for the one we were not
2345 * able to accomodate.
2348 cookie.nfsuquad[0] = *tl++;
2350 cookie.nfsuquad[1] = *tl++;
2356 more_dirs = fxdr_unsigned(int, *tl);
2359 * If at end of rpc data, get the eof boolean
2362 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2363 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2368 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2369 * by increasing d_reclen for the last record.
2372 left = DIRBLKSIZ - blksiz;
2373 dp->nfs_reclen += left;
2374 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2375 uiop->uio_iov->iov_len -= left;
2376 uiop->uio_offset += left;
2377 uiop->uio_resid -= left;
2382 * We hit the end of the directory, update direofoffset.
2384 dnp->n_direofoffset = uiop->uio_offset;
2387 * There is more to go, insert the link cookie so the
2388 * next block can be read.
2390 if (uiop->uio_resid > 0)
2391 kprintf("EEK! readdirrpc resid > 0\n");
2392 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2400 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2403 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2406 struct nfs_dirent *dp;
2410 struct vnode *newvp;
2412 caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2413 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2415 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2416 struct nfsnode *dnp = VTONFS(vp), *np;
2419 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2420 int attrflag, fhsize;
2421 struct nchandle nch;
2422 struct nchandle dnch;
2423 struct nlcomponent nlc;
2429 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2430 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2431 panic("nfs readdirplusrpc bad uio");
2434 * Obtain the namecache record for the directory so we have something
2435 * to use as a basis for creating the entries. This function will
2436 * return a held (but not locked) ncp. The ncp may be disconnected
2437 * from the tree and cannot be used for upward traversals, and the
2438 * ncp may be unnamed. Note that other unrelated operations may
2439 * cause the ncp to be named at any time.
2441 cache_fromdvp(vp, NULL, 0, &dnch);
2442 bzero(&nlc, sizeof(nlc));
2446 * If there is no cookie, assume directory was stale.
2448 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2452 return (NFSERR_BAD_COOKIE);
2454 * Loop around doing readdir rpc's of size nm_readdirsize
2455 * truncated to a multiple of DIRBLKSIZ.
2456 * The stopping criteria is EOF or buffer full.
2458 while (more_dirs && bigenough) {
2459 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2460 nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2461 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2463 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2464 *tl++ = cookie.nfsuquad[0];
2465 *tl++ = cookie.nfsuquad[1];
2466 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2467 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2468 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2469 *tl = txdr_unsigned(nmp->nm_rsize);
2470 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2471 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2476 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2477 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2478 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2479 more_dirs = fxdr_unsigned(int, *tl);
2481 /* loop thru the dir entries, doctoring them to 4bsd form */
2482 while (more_dirs && bigenough) {
2483 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2484 fileno = fxdr_hyper(tl);
2485 len = fxdr_unsigned(int, *(tl + 2));
2486 if (len <= 0 || len > NFS_MAXNAMLEN) {
2491 tlen = nfsm_rndup(len);
2493 tlen += 4; /* To ensure null termination*/
2494 left = DIRBLKSIZ - blksiz;
2495 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2496 dp->nfs_reclen += left;
2497 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2498 uiop->uio_iov->iov_len -= left;
2499 uiop->uio_offset += left;
2500 uiop->uio_resid -= left;
2503 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2506 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2507 dp->nfs_ino = fileno;
2508 dp->nfs_namlen = len;
2509 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2510 dp->nfs_type = DT_UNKNOWN;
2511 blksiz += dp->nfs_reclen;
2512 if (blksiz == DIRBLKSIZ)
2514 uiop->uio_offset += sizeof(struct nfs_dirent);
2515 uiop->uio_resid -= sizeof(struct nfs_dirent);
2516 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2517 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2518 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2519 nlc.nlc_namelen = len;
2520 nfsm_mtouio(uiop, len);
2521 cp = uiop->uio_iov->iov_base;
2524 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2525 uiop->uio_iov->iov_len -= tlen;
2526 uiop->uio_offset += tlen;
2527 uiop->uio_resid -= tlen;
2529 nfsm_adv(nfsm_rndup(len));
2530 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2532 cookie.nfsuquad[0] = *tl++;
2533 cookie.nfsuquad[1] = *tl++;
2538 * Since the attributes are before the file handle
2539 * (sigh), we must skip over the attributes and then
2540 * come back and get them.
2542 attrflag = fxdr_unsigned(int, *tl);
2546 nfsm_adv(NFSX_V3FATTR);
2547 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2548 doit = fxdr_unsigned(int, *tl);
2550 nfsm_getfh(fhp, fhsize, 1);
2551 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2556 error = nfs_nget(vp->v_mount, fhp,
2564 if (doit && bigenough) {
2569 nfsm_loadattr(newvp, NULL);
2573 IFTODT(VTTOIF(np->n_vattr.va_type));
2575 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2576 nlc.nlc_namelen, nlc.nlc_namelen,
2578 nch = cache_nlookup(&dnch, &nlc);
2579 cache_setunresolved(&nch);
2580 nfs_cache_setvp(&nch, newvp,
2581 nfspos_cache_timeout);
2584 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER"
2586 nlc.nlc_namelen, nlc.nlc_namelen,
2591 /* Just skip over the file handle */
2592 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2593 i = fxdr_unsigned(int, *tl);
2594 nfsm_adv(nfsm_rndup(i));
2596 if (newvp != NULLVP) {
2603 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2604 more_dirs = fxdr_unsigned(int, *tl);
2607 * If at end of rpc data, get the eof boolean
2610 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2611 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2616 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2617 * by increasing d_reclen for the last record.
2620 left = DIRBLKSIZ - blksiz;
2621 dp->nfs_reclen += left;
2622 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2623 uiop->uio_iov->iov_len -= left;
2624 uiop->uio_offset += left;
2625 uiop->uio_resid -= left;
2629 * We are now either at the end of the directory or have filled the
2633 dnp->n_direofoffset = uiop->uio_offset;
2635 if (uiop->uio_resid > 0)
2636 kprintf("EEK! readdirplusrpc resid > 0\n");
2637 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2641 if (newvp != NULLVP) {
2654 * Silly rename. To make the NFS filesystem that is stateless look a little
2655 * more like the "ufs" a remove of an active vnode is translated to a rename
2656 * to a funny looking filename that is removed by nfs_inactive on the
2657 * nfsnode. There is the potential for another process on a different client
2658 * to create the same funny name between the nfs_lookitup() fails and the
2659 * nfs_rename() completes, but...
2662 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2664 struct sillyrename *sp;
2669 * We previously purged dvp instead of vp. I don't know why, it
2670 * completely destroys performance. We can't do it anyway with the
2671 * new VFS API since we would be breaking the namecache topology.
2673 cache_purge(vp); /* XXX */
2676 if (vp->v_type == VDIR)
2677 panic("nfs: sillyrename dir");
2679 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2680 M_NFSREQ, M_WAITOK);
2681 sp->s_cred = crdup(cnp->cn_cred);
2685 /* Fudge together a funny name */
2686 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2687 (int)(intptr_t)cnp->cn_td);
2689 /* Try lookitups until we get one that isn't there */
2690 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2691 cnp->cn_td, NULL) == 0) {
2693 if (sp->s_name[4] > 'z') {
2698 error = nfs_renameit(dvp, cnp, sp);
2701 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2703 np->n_sillyrename = sp;
2708 kfree((caddr_t)sp, M_NFSREQ);
2713 * Look up a file name and optionally either update the file handle or
2714 * allocate an nfsnode, depending on the value of npp.
2715 * npp == NULL --> just do the lookup
2716 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2718 * *npp != NULL --> update the file handle in the vnode
2721 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2722 struct thread *td, struct nfsnode **npp)
2727 struct vnode *newvp = NULL;
2728 struct nfsnode *np, *dnp = VTONFS(dvp);
2729 caddr_t bpos, dpos, cp2;
2730 int error = 0, fhlen, attrflag;
2731 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2733 int v3 = NFS_ISV3(dvp);
2735 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2736 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2737 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2738 nfsm_fhtom(dvp, v3);
2739 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2740 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2741 if (npp && !error) {
2742 nfsm_getfh(nfhp, fhlen, v3);
2745 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2746 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2747 np->n_fhp = &np->n_fh;
2748 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2749 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2750 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2751 np->n_fhsize = fhlen;
2753 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2757 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2765 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
2766 if (!attrflag && *npp == NULL) {
2775 nfsm_loadattr(newvp, NULL);
2779 if (npp && *npp == NULL) {
2794 * Nfs Version 3 commit rpc
2797 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2802 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2803 caddr_t bpos, dpos, cp2;
2804 int error = 0, wccflag = NFSV3_WCCRATTR;
2805 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2807 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2809 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2810 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2812 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2813 txdr_hyper(offset, tl);
2815 *tl = txdr_unsigned(cnt);
2816 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE));
2817 nfsm_wcc_data(vp, wccflag);
2819 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2820 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2821 NFSX_V3WRITEVERF)) {
2822 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2824 error = NFSERR_STALEWRITEVERF;
2834 * - make nfs_bmap() essentially a no-op that does no translation
2835 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2836 * (Maybe I could use the process's page mapping, but I was concerned that
2837 * Kernel Write might not be enabled and also figured copyout() would do
2838 * a lot more work than bcopy() and also it currently happens in the
2839 * context of the swapper process (2).
2841 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
2842 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2845 nfs_bmap(struct vop_bmap_args *ap)
2847 if (ap->a_doffsetp != NULL)
2848 *ap->a_doffsetp = ap->a_loffset;
2849 if (ap->a_runp != NULL)
2851 if (ap->a_runb != NULL)
2859 * For async requests when nfsiod(s) are running, queue the request by
2860 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2864 nfs_strategy(struct vop_strategy_args *ap)
2866 struct bio *bio = ap->a_bio;
2868 struct buf *bp = bio->bio_buf;
2872 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2873 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
2874 KASSERT(BUF_REFCNT(bp) > 0,
2875 ("nfs_strategy: buffer %p not locked", bp));
2877 if (bio->bio_flags & BIO_SYNC)
2878 td = curthread; /* XXX */
2883 * We probably don't need to push an nbio any more since no
2884 * block conversion is required due to the use of 64 bit byte
2885 * offsets, but do it anyway.
2887 nbio = push_bio(bio);
2888 nbio->bio_offset = bio->bio_offset;
2891 * If the op is asynchronous and an i/o daemon is waiting
2892 * queue the request, wake it up and wait for completion
2893 * otherwise just do it ourselves.
2895 if ((bio->bio_flags & BIO_SYNC) || nfs_asyncio(ap->a_vp, nbio, td))
2896 error = nfs_doio(ap->a_vp, nbio, td);
2903 * NB Currently unsupported.
2905 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
2909 nfs_mmap(struct vop_mmap_args *ap)
2915 * fsync vnode op. Just call nfs_flush() with commit == 1.
2917 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
2921 nfs_fsync(struct vop_fsync_args *ap)
2923 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
2927 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2928 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2929 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2930 * set the buffer contains data that has already been written to the server
2931 * and which now needs a commit RPC.
2933 * If commit is 0 we only take one pass and only flush buffers containing new
2936 * If commit is 1 we take two passes, issuing a commit RPC in the second
2939 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
2940 * to completely flush all pending data.
2942 * Note that the RB_SCAN code properly handles the case where the
2943 * callback might block and directly or indirectly (another thread) cause
2944 * the RB tree to change.
2947 #ifndef NFS_COMMITBVECSIZ
2948 #define NFS_COMMITBVECSIZ 16
2951 struct nfs_flush_info {
2952 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
2959 struct buf *bvary[NFS_COMMITBVECSIZ];
2965 static int nfs_flush_bp(struct buf *bp, void *data);
2966 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
2969 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
2971 struct nfsnode *np = VTONFS(vp);
2972 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2973 struct nfs_flush_info info;
2977 bzero(&info, sizeof(info));
2980 info.waitfor = waitfor;
2981 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
2983 lwkt_gettoken(&vlock, &vp->v_token);
2989 info.mode = NFI_FLUSHNEW;
2990 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2991 nfs_flush_bp, &info);
2994 * Take a second pass if committing and no error occured.
2995 * Clean up any left over collection (whether an error
2998 if (commit && error == 0) {
2999 info.mode = NFI_COMMIT;
3000 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3001 nfs_flush_bp, &info);
3003 error = nfs_flush_docommit(&info, error);
3007 * Wait for pending I/O to complete before checking whether
3008 * any further dirty buffers exist.
3010 while (waitfor == MNT_WAIT &&
3011 bio_track_active(&vp->v_track_write)) {
3012 error = bio_track_wait(&vp->v_track_write,
3013 info.slpflag, info.slptimeo);
3016 * We have to be able to break out if this
3017 * is an 'intr' mount.
3019 if (nfs_sigintr(nmp, NULL, td)) {
3025 * Since we do not process pending signals,
3026 * once we get a PCATCH our tsleep() will no
3027 * longer sleep, switch to a fixed timeout
3030 if (info.slpflag == PCATCH) {
3032 info.slptimeo = 2 * hz;
3039 * Loop if we are flushing synchronous as well as committing,
3040 * and dirty buffers are still present. Otherwise we might livelock.
3042 } while (waitfor == MNT_WAIT && commit &&
3043 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3046 * The callbacks have to return a negative error to terminate the
3053 * Deal with any error collection
3055 if (np->n_flag & NWRITEERR) {
3056 error = np->n_error;
3057 np->n_flag &= ~NWRITEERR;
3059 lwkt_reltoken(&vlock);
3065 nfs_flush_bp(struct buf *bp, void *data)
3067 struct nfs_flush_info *info = data;
3073 switch(info->mode) {
3075 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3076 if (error && info->loops && info->waitfor == MNT_WAIT) {
3077 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3079 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3080 if (info->slpflag & PCATCH)
3081 lkflags |= LK_PCATCH;
3082 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3088 * Ignore locking errors
3096 * The buffer may have changed out from under us, even if
3097 * we did not block (MPSAFE). Check again now that it is
3100 if (bp->b_vp == info->vp &&
3101 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3110 * Only process buffers in need of a commit which we can
3111 * immediately lock. This may prevent a buffer from being
3112 * committed, but the normal flush loop will block on the
3113 * same buffer so we shouldn't get into an endless loop.
3115 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3116 (B_DELWRI | B_NEEDCOMMIT)) {
3119 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3123 * We must recheck after successfully locking the buffer.
3125 if (bp->b_vp != info->vp ||
3126 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3127 (B_DELWRI | B_NEEDCOMMIT)) {
3133 * NOTE: storing the bp in the bvary[] basically sets
3134 * it up for a commit operation.
3136 * We must call vfs_busy_pages() now so the commit operation
3137 * is interlocked with user modifications to memory mapped
3140 * Note: to avoid loopback deadlocks, we do not
3141 * assign b_runningbufspace.
3144 bp->b_cmd = BUF_CMD_WRITE;
3145 vfs_busy_pages(bp->b_vp, bp);
3146 info->bvary[info->bvsize] = bp;
3147 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3148 if (info->bvsize == 0 || toff < info->beg_off)
3149 info->beg_off = toff;
3150 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3151 if (info->bvsize == 0 || toff > info->end_off)
3152 info->end_off = toff;
3154 if (info->bvsize == NFS_COMMITBVECSIZ) {
3155 error = nfs_flush_docommit(info, 0);
3156 KKASSERT(info->bvsize == 0);
3164 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3174 if (info->bvsize > 0) {
3176 * Commit data on the server, as required. Note that
3177 * nfs_commit will use the vnode's cred for the commit.
3178 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3180 bytes = info->end_off - info->beg_off;
3181 if (bytes > 0x40000000)
3186 retv = nfs_commit(vp, info->beg_off,
3187 (int)bytes, info->td);
3188 if (retv == NFSERR_STALEWRITEVERF)
3189 nfs_clearcommit(vp->v_mount);
3193 * Now, either mark the blocks I/O done or mark the
3194 * blocks dirty, depending on whether the commit
3197 for (i = 0; i < info->bvsize; ++i) {
3198 bp = info->bvary[i];
3199 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3202 * Error, leave B_DELWRI intact
3204 vfs_unbusy_pages(bp);
3205 bp->b_cmd = BUF_CMD_DONE;
3209 * Success, remove B_DELWRI ( bundirty() ).
3211 * b_dirtyoff/b_dirtyend seem to be NFS
3212 * specific. We should probably move that
3213 * into bundirty(). XXX
3215 * We are faking an I/O write, we have to
3216 * start the transaction in order to
3217 * immediately biodone() it.
3220 bp->b_flags &= ~B_ERROR;
3221 bp->b_dirtyoff = bp->b_dirtyend = 0;
3222 biodone(&bp->b_bio1);
3231 * NFS advisory byte-level locks.
3232 * Currently unsupported.
3234 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3238 nfs_advlock(struct vop_advlock_args *ap)
3240 struct nfsnode *np = VTONFS(ap->a_vp);
3243 * The following kludge is to allow diskless support to work
3244 * until a real NFS lockd is implemented. Basically, just pretend
3245 * that this is a local lock.
3247 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3251 * Print out the contents of an nfsnode.
3253 * nfs_print(struct vnode *a_vp)
3256 nfs_print(struct vop_print_args *ap)
3258 struct vnode *vp = ap->a_vp;
3259 struct nfsnode *np = VTONFS(vp);
3261 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3262 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3263 if (vp->v_type == VFIFO)
3270 * nfs special file access vnode op.
3271 * Essentially just get vattr and then imitate iaccess() since the device is
3272 * local to the client.
3274 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3277 nfsspec_access(struct vop_access_args *ap)
3281 struct ucred *cred = ap->a_cred;
3282 struct vnode *vp = ap->a_vp;
3283 mode_t mode = ap->a_mode;
3289 * Disallow write attempts on filesystems mounted read-only;
3290 * unless the file is a socket, fifo, or a block or character
3291 * device resident on the filesystem.
3293 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3294 switch (vp->v_type) {
3304 * If you're the super-user,
3305 * you always get access.
3307 if (cred->cr_uid == 0)
3310 error = VOP_GETATTR(vp, vap);
3314 * Access check is based on only one of owner, group, public.
3315 * If not owner, then check group. If not a member of the
3316 * group, then check public access.
3318 if (cred->cr_uid != vap->va_uid) {
3320 gp = cred->cr_groups;
3321 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3322 if (vap->va_gid == *gp)
3328 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3333 * Read wrapper for special devices.
3335 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3336 * struct ucred *a_cred)
3339 nfsspec_read(struct vop_read_args *ap)
3341 struct nfsnode *np = VTONFS(ap->a_vp);
3347 getnanotime(&np->n_atim);
3348 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3352 * Write wrapper for special devices.
3354 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3355 * struct ucred *a_cred)
3358 nfsspec_write(struct vop_write_args *ap)
3360 struct nfsnode *np = VTONFS(ap->a_vp);
3366 getnanotime(&np->n_mtim);
3367 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3371 * Close wrapper for special devices.
3373 * Update the times on the nfsnode then do device close.
3375 * nfsspec_close(struct vnode *a_vp, int a_fflag)
3378 nfsspec_close(struct vop_close_args *ap)
3380 struct vnode *vp = ap->a_vp;
3381 struct nfsnode *np = VTONFS(vp);
3384 if (np->n_flag & (NACC | NUPD)) {
3386 if (vp->v_sysref.refcnt == 1 &&
3387 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3389 if (np->n_flag & NACC)
3390 vattr.va_atime = np->n_atim;
3391 if (np->n_flag & NUPD)
3392 vattr.va_mtime = np->n_mtim;
3393 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3396 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3400 * Read wrapper for fifos.
3402 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3403 * struct ucred *a_cred)
3406 nfsfifo_read(struct vop_read_args *ap)
3408 struct nfsnode *np = VTONFS(ap->a_vp);
3414 getnanotime(&np->n_atim);
3415 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3419 * Write wrapper for fifos.
3421 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3422 * struct ucred *a_cred)
3425 nfsfifo_write(struct vop_write_args *ap)
3427 struct nfsnode *np = VTONFS(ap->a_vp);
3433 getnanotime(&np->n_mtim);
3434 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3438 * Close wrapper for fifos.
3440 * Update the times on the nfsnode then do fifo close.
3442 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3445 nfsfifo_close(struct vop_close_args *ap)
3447 struct vnode *vp = ap->a_vp;
3448 struct nfsnode *np = VTONFS(vp);
3452 if (np->n_flag & (NACC | NUPD)) {
3454 if (np->n_flag & NACC)
3456 if (np->n_flag & NUPD)
3459 if (vp->v_sysref.refcnt == 1 &&
3460 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3462 if (np->n_flag & NACC)
3463 vattr.va_atime = np->n_atim;
3464 if (np->n_flag & NUPD)
3465 vattr.va_mtime = np->n_mtim;
3466 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3469 return (VOCALL(&fifo_vnode_vops, &ap->a_head));