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 nfsfifo_read (struct vop_read_args *);
98 static int nfsfifo_write (struct vop_write_args *);
99 static int nfsfifo_close (struct vop_close_args *);
100 #define nfs_poll vop_nopoll
101 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
102 static int nfs_lookup (struct vop_old_lookup_args *);
103 static int nfs_create (struct vop_old_create_args *);
104 static int nfs_mknod (struct vop_old_mknod_args *);
105 static int nfs_open (struct vop_open_args *);
106 static int nfs_close (struct vop_close_args *);
107 static int nfs_access (struct vop_access_args *);
108 static int nfs_getattr (struct vop_getattr_args *);
109 static int nfs_setattr (struct vop_setattr_args *);
110 static int nfs_read (struct vop_read_args *);
111 static int nfs_mmap (struct vop_mmap_args *);
112 static int nfs_fsync (struct vop_fsync_args *);
113 static int nfs_remove (struct vop_old_remove_args *);
114 static int nfs_link (struct vop_old_link_args *);
115 static int nfs_rename (struct vop_old_rename_args *);
116 static int nfs_mkdir (struct vop_old_mkdir_args *);
117 static int nfs_rmdir (struct vop_old_rmdir_args *);
118 static int nfs_symlink (struct vop_old_symlink_args *);
119 static int nfs_readdir (struct vop_readdir_args *);
120 static int nfs_bmap (struct vop_bmap_args *);
121 static int nfs_strategy (struct vop_strategy_args *);
122 static int nfs_lookitup (struct vnode *, const char *, int,
123 struct ucred *, struct thread *, struct nfsnode **);
124 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
125 static int nfs_laccess (struct vop_access_args *);
126 static int nfs_readlink (struct vop_readlink_args *);
127 static int nfs_print (struct vop_print_args *);
128 static int nfs_advlock (struct vop_advlock_args *);
130 static int nfs_nresolve (struct vop_nresolve_args *);
132 * Global vfs data structures for nfs
134 struct vop_ops nfsv2_vnode_vops = {
135 .vop_default = vop_defaultop,
136 .vop_access = nfs_access,
137 .vop_advlock = nfs_advlock,
138 .vop_bmap = nfs_bmap,
139 .vop_close = nfs_close,
140 .vop_old_create = nfs_create,
141 .vop_fsync = nfs_fsync,
142 .vop_getattr = nfs_getattr,
143 .vop_getpages = nfs_getpages,
144 .vop_putpages = nfs_putpages,
145 .vop_inactive = nfs_inactive,
146 .vop_old_link = nfs_link,
147 .vop_old_lookup = nfs_lookup,
148 .vop_old_mkdir = nfs_mkdir,
149 .vop_old_mknod = nfs_mknod,
150 .vop_mmap = nfs_mmap,
151 .vop_open = nfs_open,
152 .vop_poll = nfs_poll,
153 .vop_print = nfs_print,
154 .vop_read = nfs_read,
155 .vop_readdir = nfs_readdir,
156 .vop_readlink = nfs_readlink,
157 .vop_reclaim = nfs_reclaim,
158 .vop_old_remove = nfs_remove,
159 .vop_old_rename = nfs_rename,
160 .vop_old_rmdir = nfs_rmdir,
161 .vop_setattr = nfs_setattr,
162 .vop_strategy = nfs_strategy,
163 .vop_old_symlink = nfs_symlink,
164 .vop_write = nfs_write,
165 .vop_nresolve = nfs_nresolve
169 * Special device vnode ops
171 struct vop_ops nfsv2_spec_vops = {
172 .vop_default = vop_defaultop,
173 .vop_access = nfs_laccess,
174 .vop_close = nfs_close,
175 .vop_fsync = nfs_fsync,
176 .vop_getattr = nfs_getattr,
177 .vop_inactive = nfs_inactive,
178 .vop_print = nfs_print,
179 .vop_read = vop_stdnoread,
180 .vop_reclaim = nfs_reclaim,
181 .vop_setattr = nfs_setattr,
182 .vop_write = vop_stdnowrite
185 struct vop_ops nfsv2_fifo_vops = {
186 .vop_default = fifo_vnoperate,
187 .vop_access = nfs_laccess,
188 .vop_close = nfsfifo_close,
189 .vop_fsync = nfs_fsync,
190 .vop_getattr = nfs_getattr,
191 .vop_inactive = nfs_inactive,
192 .vop_print = nfs_print,
193 .vop_read = nfsfifo_read,
194 .vop_reclaim = nfs_reclaim,
195 .vop_setattr = nfs_setattr,
196 .vop_write = nfsfifo_write
199 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
200 struct componentname *cnp,
202 static int nfs_removerpc (struct vnode *dvp, const char *name,
204 struct ucred *cred, struct thread *td);
205 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
206 int fnamelen, struct vnode *tdvp,
207 const char *tnameptr, int tnamelen,
208 struct ucred *cred, struct thread *td);
209 static int nfs_renameit (struct vnode *sdvp,
210 struct componentname *scnp,
211 struct sillyrename *sp);
213 SYSCTL_DECL(_vfs_nfs);
215 static int nfs_flush_on_rename = 1;
216 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
217 &nfs_flush_on_rename, 0, "flush fvp prior to rename");
218 static int nfs_flush_on_hlink = 0;
219 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
220 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
222 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
223 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
224 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
226 static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
227 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
228 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
230 static int nfspos_cache_timeout = NFS_MINATTRTIMO;
231 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
232 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
234 static int nfsv3_commit_on_close = 0;
235 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
236 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
239 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
241 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
242 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
245 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
246 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
247 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
249 nfs3_access_otw(struct vnode *vp, int wmode,
250 struct thread *td, struct ucred *cred)
252 struct nfsnode *np = VTONFS(vp);
257 struct nfsm_info info;
262 nfsstats.rpccnt[NFSPROC_ACCESS]++;
263 nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
264 NFSX_FH(info.v3) + NFSX_UNSIGNED);
265 ERROROUT(nfsm_fhtom(&info, vp));
266 tl = nfsm_build(&info, NFSX_UNSIGNED);
267 *tl = txdr_unsigned(wmode);
268 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
269 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
271 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
272 rmode = fxdr_unsigned(u_int32_t, *tl);
274 np->n_modeuid = cred->cr_uid;
275 np->n_modestamp = mycpu->gd_time_seconds;
284 * nfs access vnode op.
285 * For nfs version 2, just return ok. File accesses may fail later.
286 * For nfs version 3, use the access rpc to check accessibility. If file modes
287 * are changed on the server, accesses might still fail later.
289 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
292 nfs_access(struct vop_access_args *ap)
294 struct vnode *vp = ap->a_vp;
295 thread_t td = curthread;
297 u_int32_t mode, wmode;
298 struct nfsnode *np = VTONFS(vp);
299 int v3 = NFS_ISV3(vp);
302 * Disallow write attempts on filesystems mounted read-only;
303 * unless the file is a socket, fifo, or a block or character
304 * device resident on the filesystem.
306 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
307 switch (vp->v_type) {
317 * For nfs v3, check to see if we have done this recently, and if
318 * so return our cached result instead of making an ACCESS call.
319 * If not, do an access rpc, otherwise you are stuck emulating
320 * ufs_access() locally using the vattr. This may not be correct,
321 * since the server may apply other access criteria such as
322 * client uid-->server uid mapping that we do not know about.
325 if (ap->a_mode & VREAD)
326 mode = NFSV3ACCESS_READ;
329 if (vp->v_type != VDIR) {
330 if (ap->a_mode & VWRITE)
331 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
332 if (ap->a_mode & VEXEC)
333 mode |= NFSV3ACCESS_EXECUTE;
335 if (ap->a_mode & VWRITE)
336 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
338 if (ap->a_mode & VEXEC)
339 mode |= NFSV3ACCESS_LOOKUP;
341 /* XXX safety belt, only make blanket request if caching */
342 if (nfsaccess_cache_timeout > 0) {
343 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
344 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
345 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
351 * Does our cached result allow us to give a definite yes to
354 if (np->n_modestamp &&
355 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
356 (ap->a_cred->cr_uid == np->n_modeuid) &&
357 ((np->n_mode & mode) == mode)) {
358 nfsstats.accesscache_hits++;
361 * Either a no, or a don't know. Go to the wire.
363 nfsstats.accesscache_misses++;
364 error = nfs3_access_otw(vp, wmode, td, ap->a_cred);
366 if ((np->n_mode & mode) != mode) {
372 if ((error = nfs_laccess(ap)) != 0)
376 * Attempt to prevent a mapped root from accessing a file
377 * which it shouldn't. We try to read a byte from the file
378 * if the user is root and the file is not zero length.
379 * After calling nfs_laccess, we should have the correct
382 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
383 && VTONFS(vp)->n_size > 0) {
390 auio.uio_iov = &aiov;
394 auio.uio_segflg = UIO_SYSSPACE;
395 auio.uio_rw = UIO_READ;
398 if (vp->v_type == VREG) {
399 error = nfs_readrpc_uio(vp, &auio);
400 } else if (vp->v_type == VDIR) {
402 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
404 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
405 error = nfs_readdirrpc_uio(vp, &auio);
407 } else if (vp->v_type == VLNK) {
408 error = nfs_readlinkrpc_uio(vp, &auio);
415 * [re]record creds for reading and/or writing if access
416 * was granted. Assume the NFS server will grant read access
417 * for execute requests.
420 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) {
423 crfree(np->n_rucred);
424 np->n_rucred = ap->a_cred;
426 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
429 crfree(np->n_wucred);
430 np->n_wucred = ap->a_cred;
438 * Check to see if the type is ok
439 * and that deletion is not in progress.
440 * For paged in text files, you will need to flush the page cache
441 * if consistency is lost.
443 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
448 nfs_open(struct vop_open_args *ap)
450 struct vnode *vp = ap->a_vp;
451 struct nfsnode *np = VTONFS(vp);
455 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
457 kprintf("open eacces vtyp=%d\n",vp->v_type);
463 * Save valid creds for reading and writing for later RPCs.
465 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
468 crfree(np->n_rucred);
469 np->n_rucred = ap->a_cred;
471 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
474 crfree(np->n_wucred);
475 np->n_wucred = ap->a_cred;
479 * Clear the attribute cache only if opening with write access. It
480 * is unclear if we should do this at all here, but we certainly
481 * should not clear the cache unconditionally simply because a file
484 if (ap->a_mode & FWRITE)
488 * For normal NFS, reconcile changes made locally verses
489 * changes made remotely. Note that VOP_GETATTR only goes
490 * to the wire if the cached attribute has timed out or been
493 * If local modifications have been made clear the attribute
494 * cache to force an attribute and modified time check. If
495 * GETATTR detects that the file has been changed by someone
496 * other then us it will set NRMODIFIED.
498 * If we are opening a directory and local changes have been
499 * made we have to invalidate the cache in order to ensure
500 * that we get the most up-to-date information from the
503 if (np->n_flag & NLMODIFIED) {
505 if (vp->v_type == VDIR) {
506 error = nfs_vinvalbuf(vp, V_SAVE, 1);
512 error = VOP_GETATTR(vp, &vattr);
515 if (np->n_flag & NRMODIFIED) {
516 if (vp->v_type == VDIR)
518 error = nfs_vinvalbuf(vp, V_SAVE, 1);
521 np->n_flag &= ~NRMODIFIED;
524 return (vop_stdopen(ap));
529 * What an NFS client should do upon close after writing is a debatable issue.
530 * Most NFS clients push delayed writes to the server upon close, basically for
532 * 1 - So that any write errors may be reported back to the client process
533 * doing the close system call. By far the two most likely errors are
534 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
535 * 2 - To put a worst case upper bound on cache inconsistency between
536 * multiple clients for the file.
537 * There is also a consistency problem for Version 2 of the protocol w.r.t.
538 * not being able to tell if other clients are writing a file concurrently,
539 * since there is no way of knowing if the changed modify time in the reply
540 * is only due to the write for this client.
541 * (NFS Version 3 provides weak cache consistency data in the reply that
542 * should be sufficient to detect and handle this case.)
544 * The current code does the following:
545 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
546 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
547 * or commit them (this satisfies 1 and 2 except for the
548 * case where the server crashes after this close but
549 * before the commit RPC, which is felt to be "good
550 * enough". Changing the last argument to nfs_flush() to
551 * a 1 would force a commit operation, if it is felt a
552 * commit is necessary now.
553 * for NQNFS - do nothing now, since 2 is dealt with via leases and
554 * 1 should be dealt with via an fsync() system call for
555 * cases where write errors are important.
557 * nfs_close(struct vnode *a_vp, int a_fflag)
561 nfs_close(struct vop_close_args *ap)
563 struct vnode *vp = ap->a_vp;
564 struct nfsnode *np = VTONFS(vp);
566 thread_t td = curthread;
568 if (vp->v_type == VREG) {
569 if (np->n_flag & NLMODIFIED) {
572 * Under NFSv3 we have dirty buffers to dispose of. We
573 * must flush them to the NFS server. We have the option
574 * of waiting all the way through the commit rpc or just
575 * waiting for the initial write. The default is to only
576 * wait through the initial write so the data is in the
577 * server's cache, which is roughly similar to the state
578 * a standard disk subsystem leaves the file in on close().
580 * We cannot clear the NLMODIFIED bit in np->n_flag due to
581 * potential races with other processes, and certainly
582 * cannot clear it if we don't commit.
584 int cm = nfsv3_commit_on_close ? 1 : 0;
585 error = nfs_flush(vp, MNT_WAIT, td, cm);
586 /* np->n_flag &= ~NLMODIFIED; */
588 error = nfs_vinvalbuf(vp, V_SAVE, 1);
592 if (np->n_flag & NWRITEERR) {
593 np->n_flag &= ~NWRITEERR;
602 * nfs getattr call from vfs.
604 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
607 nfs_getattr(struct vop_getattr_args *ap)
609 struct vnode *vp = ap->a_vp;
610 struct nfsnode *np = VTONFS(vp);
612 thread_t td = curthread;
613 struct nfsm_info info;
616 info.v3 = NFS_ISV3(vp);
619 * Update local times for special files.
621 if (np->n_flag & (NACC | NUPD))
624 * First look in the cache.
626 if (nfs_getattrcache(vp, ap->a_vap) == 0)
629 if (info.v3 && nfsaccess_cache_timeout > 0) {
630 nfsstats.accesscache_misses++;
631 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
632 if (nfs_getattrcache(vp, ap->a_vap) == 0)
636 nfsstats.rpccnt[NFSPROC_GETATTR]++;
637 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
638 ERROROUT(nfsm_fhtom(&info, vp));
639 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
640 nfs_vpcred(vp, ND_CHECK), &error));
642 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
653 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
656 nfs_setattr(struct vop_setattr_args *ap)
658 struct vnode *vp = ap->a_vp;
659 struct nfsnode *np = VTONFS(vp);
660 struct vattr *vap = ap->a_vap;
663 thread_t td = curthread;
670 * Setting of flags is not supported.
672 if (vap->va_flags != VNOVAL)
676 * Disallow write attempts if the filesystem is mounted read-only.
678 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
679 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
680 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
681 (vp->v_mount->mnt_flag & MNT_RDONLY))
684 if (vap->va_size != VNOVAL) {
686 * truncation requested
688 switch (vp->v_type) {
695 if (vap->va_mtime.tv_sec == VNOVAL &&
696 vap->va_atime.tv_sec == VNOVAL &&
697 vap->va_mode == (mode_t)VNOVAL &&
698 vap->va_uid == (uid_t)VNOVAL &&
699 vap->va_gid == (gid_t)VNOVAL)
701 vap->va_size = VNOVAL;
705 * Disallow write attempts if the filesystem is
708 if (vp->v_mount->mnt_flag & MNT_RDONLY)
712 * This is nasty. The RPCs we send to flush pending
713 * data often return attribute information which is
714 * cached via a callback to nfs_loadattrcache(), which
715 * has the effect of changing our notion of the file
716 * size. Due to flushed appends and other operations
717 * the file size can be set to virtually anything,
718 * including values that do not match either the old
719 * or intended file size.
721 * When this condition is detected we must loop to
722 * try the operation again. Hopefully no more
723 * flushing is required on the loop so it works the
724 * second time around. THIS CASE ALMOST ALWAYS
729 error = nfs_meta_setsize(vp, td, vap->va_size);
731 if (np->n_flag & NLMODIFIED) {
732 if (vap->va_size == 0)
733 error = nfs_vinvalbuf(vp, 0, 1);
735 error = nfs_vinvalbuf(vp, V_SAVE, 1);
738 * note: this loop case almost always happens at
739 * least once per truncation.
741 if (error == 0 && np->n_size != vap->va_size)
743 np->n_vattr.va_size = vap->va_size;
746 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
748 * What to do. If we are modifying the mtime we lose
749 * mtime detection of changes made by the server or other
750 * clients. But programs like rsync/rdist/cpdup are going
751 * to call utimes a lot. We don't want to piecemeal sync.
753 * For now sync if any prior remote changes were detected,
754 * but allow us to lose track of remote changes made during
755 * the utimes operation.
757 if (np->n_flag & NRMODIFIED)
758 error = nfs_vinvalbuf(vp, V_SAVE, 1);
762 if (vap->va_mtime.tv_sec != VNOVAL) {
763 np->n_mtime = vap->va_mtime.tv_sec;
767 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
770 * Sanity check if a truncation was issued. This should only occur
771 * if multiple processes are racing on the same file.
773 if (error == 0 && vap->va_size != VNOVAL &&
774 np->n_size != vap->va_size) {
775 kprintf("NFS ftruncate: server disagrees on the file size: "
778 (long long)vap->va_size,
779 (long long)np->n_size);
782 if (error && vap->va_size != VNOVAL) {
783 np->n_size = np->n_vattr.va_size = tsize;
784 vnode_pager_setsize(vp, np->n_size);
790 * Do an nfs setattr rpc.
793 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
794 struct ucred *cred, struct thread *td)
796 struct nfsv2_sattr *sp;
797 struct nfsnode *np = VTONFS(vp);
799 int error = 0, wccflag = NFSV3_WCCRATTR;
800 struct nfsm_info info;
803 info.v3 = NFS_ISV3(vp);
805 nfsstats.rpccnt[NFSPROC_SETATTR]++;
806 nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
807 NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
808 ERROROUT(nfsm_fhtom(&info, vp));
810 nfsm_v3attrbuild(&info, vap, TRUE);
811 tl = nfsm_build(&info, NFSX_UNSIGNED);
814 sp = nfsm_build(&info, NFSX_V2SATTR);
815 if (vap->va_mode == (mode_t)VNOVAL)
816 sp->sa_mode = nfs_xdrneg1;
818 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
819 if (vap->va_uid == (uid_t)VNOVAL)
820 sp->sa_uid = nfs_xdrneg1;
822 sp->sa_uid = txdr_unsigned(vap->va_uid);
823 if (vap->va_gid == (gid_t)VNOVAL)
824 sp->sa_gid = nfs_xdrneg1;
826 sp->sa_gid = txdr_unsigned(vap->va_gid);
827 sp->sa_size = txdr_unsigned(vap->va_size);
828 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
829 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
831 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
834 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
836 ERROROUT(nfsm_loadattr(&info, vp, NULL));
846 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
852 cache_setvp(nch, vp);
853 cache_settimeout(nch, nctimeout);
857 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
858 * nfs_lookup() until all remaining new api calls are implemented.
860 * Resolve a namecache entry. This function is passed a locked ncp and
861 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
864 nfs_nresolve(struct vop_nresolve_args *ap)
866 struct thread *td = curthread;
867 struct namecache *ncp;
878 struct nfsm_info info;
883 if ((error = vget(dvp, LK_SHARED)) != 0)
887 info.v3 = NFS_ISV3(dvp);
890 nfsstats.lookupcache_misses++;
891 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
892 ncp = ap->a_nch->ncp;
894 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
895 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
896 ERROROUT(nfsm_fhtom(&info, dvp));
897 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
898 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
899 ap->a_cred, &error));
902 * Cache negatve lookups to reduce NFS traffic, but use
903 * a fast timeout. Otherwise use a timeout of 1 tick.
904 * XXX we should add a namecache flag for no-caching
905 * to uncache the negative hit as soon as possible, but
906 * we cannot simply destroy the entry because it is used
907 * as a placeholder by the caller.
909 * The refactored nfs code will overwrite a non-zero error
910 * with 0 when we use ERROROUT(), so don't here.
913 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
914 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
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 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
942 if (NFS_CMPFH(np, fhp, fhsize)) {
946 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
956 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
957 NFS_LATTR_NOSHRINK));
958 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
959 NFS_LATTR_NOSHRINK));
961 ERROROUT(nfsm_loadattr(&info, nvp, NULL));
963 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
978 * 'cached' nfs directory lookup
980 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
982 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
983 * struct componentname *a_cnp)
986 nfs_lookup(struct vop_old_lookup_args *ap)
988 struct componentname *cnp = ap->a_cnp;
989 struct vnode *dvp = ap->a_dvp;
990 struct vnode **vpp = ap->a_vpp;
991 int flags = cnp->cn_flags;
993 struct nfsmount *nmp;
997 int lockparent, wantparent, attrflag, fhsize;
1000 struct nfsm_info info;
1003 info.v3 = NFS_ISV3(dvp);
1007 * Read-only mount check and directory check.
1010 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1011 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1014 if (dvp->v_type != VDIR)
1018 * Look it up in the cache. Note that ENOENT is only returned if we
1019 * previously entered a negative hit (see later on). The additional
1020 * nfsneg_cache_timeout check causes previously cached results to
1021 * be instantly ignored if the negative caching is turned off.
1023 lockparent = flags & CNP_LOCKPARENT;
1024 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1025 nmp = VFSTONFS(dvp->v_mount);
1033 nfsstats.lookupcache_misses++;
1034 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1035 len = cnp->cn_namelen;
1036 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1037 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1038 ERROROUT(nfsm_fhtom(&info, dvp));
1039 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1040 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1041 cnp->cn_cred, &error));
1043 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1044 NFS_LATTR_NOSHRINK);
1054 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1057 * Handle RENAME case...
1059 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1060 if (NFS_CMPFH(np, fhp, fhsize)) {
1065 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1073 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1074 NFS_LATTR_NOSHRINK));
1075 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1076 NFS_LATTR_NOSHRINK));
1078 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1085 cnp->cn_flags |= CNP_PDIRUNLOCK;
1090 if (flags & CNP_ISDOTDOT) {
1092 cnp->cn_flags |= CNP_PDIRUNLOCK;
1093 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1095 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1096 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1097 return (error); /* NOTE: return error from nget */
1101 error = vn_lock(dvp, LK_EXCLUSIVE);
1106 cnp->cn_flags |= CNP_PDIRUNLOCK;
1108 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1112 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1120 cnp->cn_flags |= CNP_PDIRUNLOCK;
1125 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1126 NFS_LATTR_NOSHRINK));
1127 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1128 NFS_LATTR_NOSHRINK));
1130 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1133 /* XXX MOVE TO nfs_nremove() */
1134 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1135 cnp->cn_nameiop != NAMEI_DELETE) {
1136 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1144 if (newvp != NULLVP) {
1148 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1149 cnp->cn_nameiop == NAMEI_RENAME) &&
1153 cnp->cn_flags |= CNP_PDIRUNLOCK;
1155 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1158 error = EJUSTRETURN;
1166 * Just call nfs_bioread() to do the work.
1168 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1169 * struct ucred *a_cred)
1172 nfs_read(struct vop_read_args *ap)
1174 struct vnode *vp = ap->a_vp;
1176 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1182 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1185 nfs_readlink(struct vop_readlink_args *ap)
1187 struct vnode *vp = ap->a_vp;
1189 if (vp->v_type != VLNK)
1191 return (nfs_bioread(vp, ap->a_uio, 0));
1195 * Do a readlink rpc.
1196 * Called by nfs_doio() from below the buffer cache.
1199 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1201 int error = 0, len, attrflag;
1202 struct nfsm_info info;
1205 info.v3 = NFS_ISV3(vp);
1207 nfsstats.rpccnt[NFSPROC_READLINK]++;
1208 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1209 ERROROUT(nfsm_fhtom(&info, vp));
1210 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1211 nfs_vpcred(vp, ND_CHECK), &error));
1213 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1214 NFS_LATTR_NOSHRINK));
1217 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1218 if (len == NFS_MAXPATHLEN) {
1219 struct nfsnode *np = VTONFS(vp);
1220 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1223 ERROROUT(nfsm_mtouio(&info, uiop, len));
1234 * If bio is non-NULL and asynchronous
1237 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1240 struct nfsmount *nmp;
1241 int error = 0, len, retlen, tsiz, eof, attrflag;
1242 struct nfsm_info info;
1245 info.v3 = NFS_ISV3(vp);
1250 nmp = VFSTONFS(vp->v_mount);
1251 tsiz = uiop->uio_resid;
1252 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1255 nfsstats.rpccnt[NFSPROC_READ]++;
1256 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1257 nfsm_reqhead(&info, vp, NFSPROC_READ,
1258 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1259 ERROROUT(nfsm_fhtom(&info, vp));
1260 tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1262 txdr_hyper(uiop->uio_offset, tl);
1263 *(tl + 2) = txdr_unsigned(len);
1265 *tl++ = txdr_unsigned(uiop->uio_offset);
1266 *tl++ = txdr_unsigned(len);
1269 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1270 nfs_vpcred(vp, ND_READ), &error));
1272 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1273 NFS_LATTR_NOSHRINK));
1274 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1275 eof = fxdr_unsigned(int, *(tl + 1));
1277 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1279 NEGATIVEOUT(retlen = nfsm_strsiz(&info, nmp->nm_rsize));
1280 ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1285 if (eof || retlen == 0) {
1288 } else if (retlen < len) {
1300 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1301 int *iomode, int *must_commit)
1305 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1306 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1307 int committed = NFSV3WRITE_FILESYNC;
1308 struct nfsm_info info;
1311 info.v3 = NFS_ISV3(vp);
1314 if (uiop->uio_iovcnt != 1)
1315 panic("nfs: writerpc iovcnt > 1");
1318 tsiz = uiop->uio_resid;
1319 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1322 nfsstats.rpccnt[NFSPROC_WRITE]++;
1323 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1324 nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1325 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1326 ERROROUT(nfsm_fhtom(&info, vp));
1328 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1329 txdr_hyper(uiop->uio_offset, tl);
1331 *tl++ = txdr_unsigned(len);
1332 *tl++ = txdr_unsigned(*iomode);
1333 *tl = txdr_unsigned(len);
1337 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1338 /* Set both "begin" and "current" to non-garbage. */
1339 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1340 *tl++ = x; /* "begin offset" */
1341 *tl++ = x; /* "current offset" */
1342 x = txdr_unsigned(len);
1343 *tl++ = x; /* total to this offset */
1344 *tl = x; /* size of this write */
1346 ERROROUT(nfsm_uiotom(&info, uiop, len));
1347 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1348 nfs_vpcred(vp, ND_WRITE), &error));
1351 * The write RPC returns a before and after mtime. The
1352 * nfsm_wcc_data() macro checks the before n_mtime
1353 * against the before time and stores the after time
1354 * in the nfsnode's cached vattr and n_mtime field.
1355 * The NRMODIFIED bit will be set if the before
1356 * time did not match the original mtime.
1358 wccflag = NFSV3_WCCCHK;
1359 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1361 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1362 rlen = fxdr_unsigned(int, *tl++);
1368 } else if (rlen < len) {
1369 backup = len - rlen;
1370 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1371 uiop->uio_iov->iov_len += backup;
1372 uiop->uio_offset -= backup;
1373 uiop->uio_resid += backup;
1376 commit = fxdr_unsigned(int, *tl++);
1379 * Return the lowest committment level
1380 * obtained by any of the RPCs.
1382 if (committed == NFSV3WRITE_FILESYNC)
1384 else if (committed == NFSV3WRITE_DATASYNC &&
1385 commit == NFSV3WRITE_UNSTABLE)
1387 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1388 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1390 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1391 } else if (bcmp((caddr_t)tl,
1392 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1394 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1399 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1408 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1409 committed = NFSV3WRITE_FILESYNC;
1410 *iomode = committed;
1412 uiop->uio_resid = tsiz;
1418 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1419 * mode set to specify the file type and the size field for rdev.
1422 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1425 struct nfsv2_sattr *sp;
1427 struct vnode *newvp = NULL;
1428 struct nfsnode *np = NULL;
1430 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1432 struct nfsm_info info;
1435 info.v3 = NFS_ISV3(dvp);
1437 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1438 rmajor = txdr_unsigned(vap->va_rmajor);
1439 rminor = txdr_unsigned(vap->va_rminor);
1440 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1441 rmajor = nfs_xdrneg1;
1442 rminor = nfs_xdrneg1;
1444 return (EOPNOTSUPP);
1446 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1449 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1450 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1451 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1452 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1453 ERROROUT(nfsm_fhtom(&info, dvp));
1454 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1457 tl = nfsm_build(&info, NFSX_UNSIGNED);
1458 *tl++ = vtonfsv3_type(vap->va_type);
1459 nfsm_v3attrbuild(&info, vap, FALSE);
1460 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1461 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1462 *tl++ = txdr_unsigned(vap->va_rmajor);
1463 *tl = txdr_unsigned(vap->va_rminor);
1466 sp = nfsm_build(&info, NFSX_V2SATTR);
1467 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1468 sp->sa_uid = nfs_xdrneg1;
1469 sp->sa_gid = nfs_xdrneg1;
1470 sp->sa_size = makeudev(rmajor, rminor);
1471 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1472 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1474 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1475 cnp->cn_cred, &error));
1477 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1483 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1484 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1490 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1501 VTONFS(dvp)->n_flag |= NLMODIFIED;
1503 VTONFS(dvp)->n_attrstamp = 0;
1509 * just call nfs_mknodrpc() to do the work.
1511 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1512 * struct componentname *a_cnp, struct vattr *a_vap)
1516 nfs_mknod(struct vop_old_mknod_args *ap)
1518 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1521 static u_long create_verf;
1523 * nfs file create call
1525 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1526 * struct componentname *a_cnp, struct vattr *a_vap)
1529 nfs_create(struct vop_old_create_args *ap)
1531 struct vnode *dvp = ap->a_dvp;
1532 struct vattr *vap = ap->a_vap;
1533 struct componentname *cnp = ap->a_cnp;
1534 struct nfsv2_sattr *sp;
1536 struct nfsnode *np = NULL;
1537 struct vnode *newvp = NULL;
1538 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1540 struct nfsm_info info;
1543 info.v3 = NFS_ISV3(dvp);
1546 * Oops, not for me..
1548 if (vap->va_type == VSOCK)
1549 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1551 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1554 if (vap->va_vaflags & VA_EXCLUSIVE)
1557 nfsstats.rpccnt[NFSPROC_CREATE]++;
1558 nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1559 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1560 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1561 ERROROUT(nfsm_fhtom(&info, dvp));
1562 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1565 tl = nfsm_build(&info, NFSX_UNSIGNED);
1566 if (fmode & O_EXCL) {
1567 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1568 tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1570 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1571 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1574 *tl++ = create_verf;
1575 *tl = ++create_verf;
1577 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1578 nfsm_v3attrbuild(&info, vap, FALSE);
1581 sp = nfsm_build(&info, NFSX_V2SATTR);
1582 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1583 sp->sa_uid = nfs_xdrneg1;
1584 sp->sa_gid = nfs_xdrneg1;
1586 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1587 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1589 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1590 cnp->cn_cred, &error));
1592 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1598 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1599 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1606 error = nfsm_wcc_data(&info, dvp, &wccflag);
1608 (void)nfsm_wcc_data(&info, dvp, &wccflag);
1614 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1615 KKASSERT(newvp == NULL);
1619 } else if (info.v3 && (fmode & O_EXCL)) {
1621 * We are normally called with only a partially initialized
1622 * VAP. Since the NFSv3 spec says that server may use the
1623 * file attributes to store the verifier, the spec requires
1624 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1625 * in atime, but we can't really assume that all servers will
1626 * so we ensure that our SETATTR sets both atime and mtime.
1628 if (vap->va_mtime.tv_sec == VNOVAL)
1629 vfs_timestamp(&vap->va_mtime);
1630 if (vap->va_atime.tv_sec == VNOVAL)
1631 vap->va_atime = vap->va_mtime;
1632 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1636 * The new np may have enough info for access
1637 * checks, make sure rucred and wucred are
1638 * initialized for read and write rpc's.
1641 if (np->n_rucred == NULL)
1642 np->n_rucred = crhold(cnp->cn_cred);
1643 if (np->n_wucred == NULL)
1644 np->n_wucred = crhold(cnp->cn_cred);
1649 VTONFS(dvp)->n_flag |= NLMODIFIED;
1651 VTONFS(dvp)->n_attrstamp = 0;
1656 * nfs file remove call
1657 * To try and make nfs semantics closer to ufs semantics, a file that has
1658 * other processes using the vnode is renamed instead of removed and then
1659 * removed later on the last close.
1660 * - If v_sysref.refcnt > 1
1661 * If a rename is not already in the works
1662 * call nfs_sillyrename() to set it up
1666 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1667 * struct componentname *a_cnp)
1670 nfs_remove(struct vop_old_remove_args *ap)
1672 struct vnode *vp = ap->a_vp;
1673 struct vnode *dvp = ap->a_dvp;
1674 struct componentname *cnp = ap->a_cnp;
1675 struct nfsnode *np = VTONFS(vp);
1680 if (vp->v_sysref.refcnt < 1)
1681 panic("nfs_remove: bad v_sysref.refcnt");
1683 if (vp->v_type == VDIR)
1685 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1686 VOP_GETATTR(vp, &vattr) == 0 &&
1687 vattr.va_nlink > 1)) {
1689 * throw away biocache buffers, mainly to avoid
1690 * unnecessary delayed writes later.
1692 error = nfs_vinvalbuf(vp, 0, 1);
1695 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1696 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1698 * Kludge City: If the first reply to the remove rpc is lost..
1699 * the reply to the retransmitted request will be ENOENT
1700 * since the file was in fact removed
1701 * Therefore, we cheat and return success.
1703 if (error == ENOENT)
1705 } else if (!np->n_sillyrename) {
1706 error = nfs_sillyrename(dvp, vp, cnp);
1708 np->n_attrstamp = 0;
1713 * nfs file remove rpc called from nfs_inactive
1716 nfs_removeit(struct sillyrename *sp)
1718 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1723 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1726 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1727 struct ucred *cred, struct thread *td)
1729 int error = 0, wccflag = NFSV3_WCCRATTR;
1730 struct nfsm_info info;
1733 info.v3 = NFS_ISV3(dvp);
1735 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1736 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1737 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1738 ERROROUT(nfsm_fhtom(&info, dvp));
1739 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1740 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1742 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1747 VTONFS(dvp)->n_flag |= NLMODIFIED;
1749 VTONFS(dvp)->n_attrstamp = 0;
1754 * nfs file rename call
1756 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1757 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1758 * struct vnode *a_tvp, struct componentname *a_tcnp)
1761 nfs_rename(struct vop_old_rename_args *ap)
1763 struct vnode *fvp = ap->a_fvp;
1764 struct vnode *tvp = ap->a_tvp;
1765 struct vnode *fdvp = ap->a_fdvp;
1766 struct vnode *tdvp = ap->a_tdvp;
1767 struct componentname *tcnp = ap->a_tcnp;
1768 struct componentname *fcnp = ap->a_fcnp;
1771 /* Check for cross-device rename */
1772 if ((fvp->v_mount != tdvp->v_mount) ||
1773 (tvp && (fvp->v_mount != tvp->v_mount))) {
1779 * We shouldn't have to flush fvp on rename for most server-side
1780 * filesystems as the file handle should not change. Unfortunately
1781 * the inode for some filesystems (msdosfs) might be tied to the
1782 * file name or directory position so to be completely safe
1783 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1786 * We must flush tvp on rename because it might become stale on the
1787 * server after the rename.
1789 if (nfs_flush_on_rename)
1790 VOP_FSYNC(fvp, MNT_WAIT);
1792 VOP_FSYNC(tvp, MNT_WAIT);
1795 * If the tvp exists and is in use, sillyrename it before doing the
1796 * rename of the new file over it.
1798 * XXX Can't sillyrename a directory.
1800 * We do not attempt to do any namecache purges in this old API
1801 * routine. The new API compat functions have access to the actual
1802 * namecache structures and will do it for us.
1804 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1805 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1812 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1813 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1826 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1828 if (error == ENOENT)
1834 * nfs file rename rpc called from nfs_remove() above
1837 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1838 struct sillyrename *sp)
1840 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1841 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1845 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1848 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1849 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1850 struct ucred *cred, struct thread *td)
1852 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1853 struct nfsm_info info;
1856 info.v3 = NFS_ISV3(fdvp);
1858 nfsstats.rpccnt[NFSPROC_RENAME]++;
1859 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1860 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1861 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1862 ERROROUT(nfsm_fhtom(&info, fdvp));
1863 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1864 ERROROUT(nfsm_fhtom(&info, tdvp));
1865 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1866 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
1868 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
1869 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
1874 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1875 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1877 VTONFS(fdvp)->n_attrstamp = 0;
1879 VTONFS(tdvp)->n_attrstamp = 0;
1884 * nfs hard link create call
1886 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1887 * struct componentname *a_cnp)
1890 nfs_link(struct vop_old_link_args *ap)
1892 struct vnode *vp = ap->a_vp;
1893 struct vnode *tdvp = ap->a_tdvp;
1894 struct componentname *cnp = ap->a_cnp;
1895 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1896 struct nfsm_info info;
1898 if (vp->v_mount != tdvp->v_mount) {
1903 * The attribute cache may get out of sync with the server on link.
1904 * Pushing writes to the server before handle was inherited from
1905 * long long ago and it is unclear if we still need to do this.
1908 if (nfs_flush_on_hlink)
1909 VOP_FSYNC(vp, MNT_WAIT);
1912 info.v3 = NFS_ISV3(vp);
1914 nfsstats.rpccnt[NFSPROC_LINK]++;
1915 nfsm_reqhead(&info, vp, NFSPROC_LINK,
1916 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
1917 nfsm_rndup(cnp->cn_namelen));
1918 ERROROUT(nfsm_fhtom(&info, vp));
1919 ERROROUT(nfsm_fhtom(&info, tdvp));
1920 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1922 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
1923 cnp->cn_cred, &error));
1925 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1926 NFS_LATTR_NOSHRINK));
1927 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
1932 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1934 VTONFS(vp)->n_attrstamp = 0;
1936 VTONFS(tdvp)->n_attrstamp = 0;
1938 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1940 if (error == EEXIST)
1946 * nfs symbolic link create call
1948 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1949 * struct componentname *a_cnp, struct vattr *a_vap,
1953 nfs_symlink(struct vop_old_symlink_args *ap)
1955 struct vnode *dvp = ap->a_dvp;
1956 struct vattr *vap = ap->a_vap;
1957 struct componentname *cnp = ap->a_cnp;
1958 struct nfsv2_sattr *sp;
1959 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1960 struct vnode *newvp = NULL;
1961 struct nfsm_info info;
1964 info.v3 = NFS_ISV3(dvp);
1966 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1967 slen = strlen(ap->a_target);
1968 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
1969 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
1970 nfsm_rndup(cnp->cn_namelen) +
1971 nfsm_rndup(slen) + NFSX_SATTR(info.v3));
1972 ERROROUT(nfsm_fhtom(&info, dvp));
1973 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1976 nfsm_v3attrbuild(&info, vap, FALSE);
1978 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
1980 sp = nfsm_build(&info, NFSX_V2SATTR);
1981 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1982 sp->sa_uid = nfs_xdrneg1;
1983 sp->sa_gid = nfs_xdrneg1;
1984 sp->sa_size = nfs_xdrneg1;
1985 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1986 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1990 * Issue the NFS request and get the rpc response.
1992 * Only NFSv3 responses returning an error of 0 actually return
1993 * a file handle that can be converted into newvp without having
1994 * to do an extra lookup rpc.
1996 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
1997 cnp->cn_cred, &error));
2000 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2002 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2006 * out code jumps -> here, mrep is also freed.
2014 * If we get an EEXIST error, silently convert it to no-error
2015 * in case of an NFS retry.
2017 if (error == EEXIST)
2021 * If we do not have (or no longer have) an error, and we could
2022 * not extract the newvp from the response due to the request being
2023 * NFSv2 or the error being EEXIST. We have to do a lookup in order
2024 * to obtain a newvp to return.
2026 if (error == 0 && newvp == NULL) {
2027 struct nfsnode *np = NULL;
2029 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2030 cnp->cn_cred, cnp->cn_td, &np);
2040 VTONFS(dvp)->n_flag |= NLMODIFIED;
2042 VTONFS(dvp)->n_attrstamp = 0;
2049 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2050 * struct componentname *a_cnp, struct vattr *a_vap)
2053 nfs_mkdir(struct vop_old_mkdir_args *ap)
2055 struct vnode *dvp = ap->a_dvp;
2056 struct vattr *vap = ap->a_vap;
2057 struct componentname *cnp = ap->a_cnp;
2058 struct nfsv2_sattr *sp;
2059 struct nfsnode *np = NULL;
2060 struct vnode *newvp = NULL;
2062 int error = 0, wccflag = NFSV3_WCCRATTR;
2065 struct nfsm_info info;
2068 info.v3 = NFS_ISV3(dvp);
2070 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2073 len = cnp->cn_namelen;
2074 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2075 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2076 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2077 nfsm_rndup(len) + NFSX_SATTR(info.v3));
2078 ERROROUT(nfsm_fhtom(&info, dvp));
2079 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2081 nfsm_v3attrbuild(&info, vap, FALSE);
2083 sp = nfsm_build(&info, NFSX_V2SATTR);
2084 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2085 sp->sa_uid = nfs_xdrneg1;
2086 sp->sa_gid = nfs_xdrneg1;
2087 sp->sa_size = nfs_xdrneg1;
2088 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2089 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2091 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2092 cnp->cn_cred, &error));
2094 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2097 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2102 VTONFS(dvp)->n_flag |= NLMODIFIED;
2104 VTONFS(dvp)->n_attrstamp = 0;
2106 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2107 * if we can succeed in looking up the directory.
2109 if (error == EEXIST || (!error && !gotvp)) {
2114 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2118 if (newvp->v_type != VDIR)
2131 * nfs remove directory call
2133 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2134 * struct componentname *a_cnp)
2137 nfs_rmdir(struct vop_old_rmdir_args *ap)
2139 struct vnode *vp = ap->a_vp;
2140 struct vnode *dvp = ap->a_dvp;
2141 struct componentname *cnp = ap->a_cnp;
2142 int error = 0, wccflag = NFSV3_WCCRATTR;
2143 struct nfsm_info info;
2146 info.v3 = NFS_ISV3(dvp);
2150 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2151 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2152 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2153 nfsm_rndup(cnp->cn_namelen));
2154 ERROROUT(nfsm_fhtom(&info, dvp));
2155 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2157 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2158 cnp->cn_cred, &error));
2160 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2165 VTONFS(dvp)->n_flag |= NLMODIFIED;
2167 VTONFS(dvp)->n_attrstamp = 0;
2169 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2171 if (error == ENOENT)
2179 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2182 nfs_readdir(struct vop_readdir_args *ap)
2184 struct vnode *vp = ap->a_vp;
2185 struct nfsnode *np = VTONFS(vp);
2186 struct uio *uio = ap->a_uio;
2190 if (vp->v_type != VDIR)
2193 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2197 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2198 * and then check that is still valid, or if this is an NQNFS mount
2199 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2200 * VOP_GETATTR() does not necessarily go to the wire.
2202 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2203 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2204 if (VOP_GETATTR(vp, &vattr) == 0 &&
2205 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2207 nfsstats.direofcache_hits++;
2213 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2214 * own cache coherency checks so we do not have to.
2216 tresid = uio->uio_resid;
2217 error = nfs_bioread(vp, uio, 0);
2219 if (!error && uio->uio_resid == tresid)
2220 nfsstats.direofcache_misses++;
2227 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2229 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2230 * offset/block and converts the nfs formatted directory entries for userland
2231 * consumption as well as deals with offsets into the middle of blocks.
2232 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2233 * be block-bounded. It must convert to cookies for the actual RPC.
2236 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2239 struct nfs_dirent *dp = NULL;
2244 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2245 struct nfsnode *dnp = VTONFS(vp);
2247 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2249 struct nfsm_info info;
2252 info.v3 = NFS_ISV3(vp);
2255 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2256 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2257 panic("nfs readdirrpc bad uio");
2261 * If there is no cookie, assume directory was stale.
2263 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2267 return (NFSERR_BAD_COOKIE);
2269 * Loop around doing readdir rpc's of size nm_readdirsize
2270 * truncated to a multiple of DIRBLKSIZ.
2271 * The stopping criteria is EOF or buffer full.
2273 while (more_dirs && bigenough) {
2274 nfsstats.rpccnt[NFSPROC_READDIR]++;
2275 nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2276 NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2277 ERROROUT(nfsm_fhtom(&info, vp));
2279 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2280 *tl++ = cookie.nfsuquad[0];
2281 *tl++ = cookie.nfsuquad[1];
2282 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2283 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2285 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2286 *tl++ = cookie.nfsuquad[0];
2288 *tl = txdr_unsigned(nmp->nm_readdirsize);
2289 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2291 nfs_vpcred(vp, ND_READ), &error));
2293 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2294 NFS_LATTR_NOSHRINK));
2295 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2296 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2297 dnp->n_cookieverf.nfsuquad[1] = *tl;
2299 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2300 more_dirs = fxdr_unsigned(int, *tl);
2302 /* loop thru the dir entries, converting them to std form */
2303 while (more_dirs && bigenough) {
2305 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2306 fileno = fxdr_hyper(tl);
2307 len = fxdr_unsigned(int, *(tl + 2));
2309 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2310 fileno = fxdr_unsigned(u_quad_t, *tl++);
2311 len = fxdr_unsigned(int, *tl);
2313 if (len <= 0 || len > NFS_MAXNAMLEN) {
2321 * len is the number of bytes in the path element
2322 * name, not including the \0 termination.
2324 * tlen is the number of bytes w have to reserve for
2325 * the path element name.
2327 tlen = nfsm_rndup(len);
2329 tlen += 4; /* To ensure null termination */
2332 * If the entry would cross a DIRBLKSIZ boundary,
2333 * extend the previous nfs_dirent to cover the
2336 left = DIRBLKSIZ - blksiz;
2337 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2338 dp->nfs_reclen += left;
2339 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2340 uiop->uio_iov->iov_len -= left;
2341 uiop->uio_offset += left;
2342 uiop->uio_resid -= left;
2345 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2348 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2349 dp->nfs_ino = fileno;
2350 dp->nfs_namlen = len;
2351 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2352 dp->nfs_type = DT_UNKNOWN;
2353 blksiz += dp->nfs_reclen;
2354 if (blksiz == DIRBLKSIZ)
2356 uiop->uio_offset += sizeof(struct nfs_dirent);
2357 uiop->uio_resid -= sizeof(struct nfs_dirent);
2358 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2359 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2360 ERROROUT(nfsm_mtouio(&info, uiop, len));
2363 * The uiop has advanced by nfs_dirent + len
2364 * but really needs to advance by
2367 cp = uiop->uio_iov->iov_base;
2369 *cp = '\0'; /* null terminate */
2370 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2371 uiop->uio_iov->iov_len -= tlen;
2372 uiop->uio_offset += tlen;
2373 uiop->uio_resid -= tlen;
2376 * NFS strings must be rounded up (nfsm_myouio
2377 * handled that in the bigenough case).
2379 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2382 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2384 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2388 * If we were able to accomodate the last entry,
2389 * get the cookie for the next one. Otherwise
2390 * hold-over the cookie for the one we were not
2391 * able to accomodate.
2394 cookie.nfsuquad[0] = *tl++;
2396 cookie.nfsuquad[1] = *tl++;
2397 } else if (info.v3) {
2402 more_dirs = fxdr_unsigned(int, *tl);
2405 * If at end of rpc data, get the eof boolean
2408 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2409 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2415 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2416 * by increasing d_reclen for the last record.
2419 left = DIRBLKSIZ - blksiz;
2420 dp->nfs_reclen += left;
2421 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2422 uiop->uio_iov->iov_len -= left;
2423 uiop->uio_offset += left;
2424 uiop->uio_resid -= left;
2429 * We hit the end of the directory, update direofoffset.
2431 dnp->n_direofoffset = uiop->uio_offset;
2434 * There is more to go, insert the link cookie so the
2435 * next block can be read.
2437 if (uiop->uio_resid > 0)
2438 kprintf("EEK! readdirrpc resid > 0\n");
2439 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2447 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2450 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2453 struct nfs_dirent *dp;
2455 struct vnode *newvp;
2457 caddr_t dpossav1, dpossav2;
2459 struct mbuf *mdsav1, *mdsav2;
2461 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2462 struct nfsnode *dnp = VTONFS(vp), *np;
2465 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2466 int attrflag, fhsize;
2467 struct nchandle nch;
2468 struct nchandle dnch;
2469 struct nlcomponent nlc;
2470 struct nfsm_info info;
2479 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2480 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2481 panic("nfs readdirplusrpc bad uio");
2484 * Obtain the namecache record for the directory so we have something
2485 * to use as a basis for creating the entries. This function will
2486 * return a held (but not locked) ncp. The ncp may be disconnected
2487 * from the tree and cannot be used for upward traversals, and the
2488 * ncp may be unnamed. Note that other unrelated operations may
2489 * cause the ncp to be named at any time.
2491 cache_fromdvp(vp, NULL, 0, &dnch);
2492 bzero(&nlc, sizeof(nlc));
2496 * If there is no cookie, assume directory was stale.
2498 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2502 return (NFSERR_BAD_COOKIE);
2504 * Loop around doing readdir rpc's of size nm_readdirsize
2505 * truncated to a multiple of DIRBLKSIZ.
2506 * The stopping criteria is EOF or buffer full.
2508 while (more_dirs && bigenough) {
2509 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2510 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2511 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2512 ERROROUT(nfsm_fhtom(&info, vp));
2513 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2514 *tl++ = cookie.nfsuquad[0];
2515 *tl++ = cookie.nfsuquad[1];
2516 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2517 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2518 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2519 *tl = txdr_unsigned(nmp->nm_rsize);
2520 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2522 nfs_vpcred(vp, ND_READ), &error));
2523 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2524 NFS_LATTR_NOSHRINK));
2525 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2526 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2527 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2528 more_dirs = fxdr_unsigned(int, *tl);
2530 /* loop thru the dir entries, doctoring them to 4bsd form */
2531 while (more_dirs && bigenough) {
2532 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2533 fileno = fxdr_hyper(tl);
2534 len = fxdr_unsigned(int, *(tl + 2));
2535 if (len <= 0 || len > NFS_MAXNAMLEN) {
2541 tlen = nfsm_rndup(len);
2543 tlen += 4; /* To ensure null termination*/
2544 left = DIRBLKSIZ - blksiz;
2545 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2546 dp->nfs_reclen += left;
2547 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2548 uiop->uio_iov->iov_len -= left;
2549 uiop->uio_offset += left;
2550 uiop->uio_resid -= left;
2553 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2556 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2557 dp->nfs_ino = fileno;
2558 dp->nfs_namlen = len;
2559 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2560 dp->nfs_type = DT_UNKNOWN;
2561 blksiz += dp->nfs_reclen;
2562 if (blksiz == DIRBLKSIZ)
2564 uiop->uio_offset += sizeof(struct nfs_dirent);
2565 uiop->uio_resid -= sizeof(struct nfs_dirent);
2566 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2567 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2568 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2569 nlc.nlc_namelen = len;
2570 ERROROUT(nfsm_mtouio(&info, uiop, len));
2571 cp = uiop->uio_iov->iov_base;
2574 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2575 uiop->uio_iov->iov_len -= tlen;
2576 uiop->uio_offset += tlen;
2577 uiop->uio_resid -= tlen;
2579 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2581 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2583 cookie.nfsuquad[0] = *tl++;
2584 cookie.nfsuquad[1] = *tl++;
2589 * Since the attributes are before the file handle
2590 * (sigh), we must skip over the attributes and then
2591 * come back and get them.
2593 attrflag = fxdr_unsigned(int, *tl);
2595 dpossav1 = info.dpos;
2597 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2598 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2599 doit = fxdr_unsigned(int, *tl);
2601 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2602 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2607 error = nfs_nget(vp->v_mount, fhp,
2615 if (doit && bigenough) {
2616 dpossav2 = info.dpos;
2617 info.dpos = dpossav1;
2620 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2621 info.dpos = dpossav2;
2624 IFTODT(VTTOIF(np->n_vattr.va_type));
2626 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2627 nlc.nlc_namelen, nlc.nlc_namelen,
2629 nch = cache_nlookup(&dnch, &nlc);
2630 cache_setunresolved(&nch);
2631 nfs_cache_setvp(&nch, newvp,
2632 nfspos_cache_timeout);
2635 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER"
2637 nlc.nlc_namelen, nlc.nlc_namelen,
2642 /* Just skip over the file handle */
2643 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2644 i = fxdr_unsigned(int, *tl);
2645 ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2647 if (newvp != NULLVP) {
2654 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2655 more_dirs = fxdr_unsigned(int, *tl);
2658 * If at end of rpc data, get the eof boolean
2661 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2662 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2668 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2669 * by increasing d_reclen for the last record.
2672 left = DIRBLKSIZ - blksiz;
2673 dp->nfs_reclen += left;
2674 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2675 uiop->uio_iov->iov_len -= left;
2676 uiop->uio_offset += left;
2677 uiop->uio_resid -= left;
2681 * We are now either at the end of the directory or have filled the
2685 dnp->n_direofoffset = uiop->uio_offset;
2687 if (uiop->uio_resid > 0)
2688 kprintf("EEK! readdirplusrpc resid > 0\n");
2689 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2693 if (newvp != NULLVP) {
2706 * Silly rename. To make the NFS filesystem that is stateless look a little
2707 * more like the "ufs" a remove of an active vnode is translated to a rename
2708 * to a funny looking filename that is removed by nfs_inactive on the
2709 * nfsnode. There is the potential for another process on a different client
2710 * to create the same funny name between the nfs_lookitup() fails and the
2711 * nfs_rename() completes, but...
2714 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2716 struct sillyrename *sp;
2721 * We previously purged dvp instead of vp. I don't know why, it
2722 * completely destroys performance. We can't do it anyway with the
2723 * new VFS API since we would be breaking the namecache topology.
2725 cache_purge(vp); /* XXX */
2728 if (vp->v_type == VDIR)
2729 panic("nfs: sillyrename dir");
2731 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2732 M_NFSREQ, M_WAITOK);
2733 sp->s_cred = crdup(cnp->cn_cred);
2737 /* Fudge together a funny name */
2738 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2739 (int)(intptr_t)cnp->cn_td);
2741 /* Try lookitups until we get one that isn't there */
2742 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2743 cnp->cn_td, NULL) == 0) {
2745 if (sp->s_name[4] > 'z') {
2750 error = nfs_renameit(dvp, cnp, sp);
2753 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2755 np->n_sillyrename = sp;
2760 kfree((caddr_t)sp, M_NFSREQ);
2765 * Look up a file name and optionally either update the file handle or
2766 * allocate an nfsnode, depending on the value of npp.
2767 * npp == NULL --> just do the lookup
2768 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2770 * *npp != NULL --> update the file handle in the vnode
2773 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2774 struct thread *td, struct nfsnode **npp)
2776 struct vnode *newvp = NULL;
2777 struct nfsnode *np, *dnp = VTONFS(dvp);
2778 int error = 0, fhlen, attrflag;
2780 struct nfsm_info info;
2783 info.v3 = NFS_ISV3(dvp);
2785 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2786 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2787 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2788 ERROROUT(nfsm_fhtom(&info, dvp));
2789 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2790 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2791 if (npp && !error) {
2792 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2795 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2796 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2797 np->n_fhp = &np->n_fh;
2798 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2799 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2800 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2801 np->n_fhsize = fhlen;
2803 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2807 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2816 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2817 NFS_LATTR_NOSHRINK));
2818 if (!attrflag && *npp == NULL) {
2828 ERROROUT(error = nfsm_loadattr(&info, newvp, NULL));
2834 if (npp && *npp == NULL) {
2849 * Nfs Version 3 commit rpc
2851 * We call it 'uio' to distinguish it from 'bio' but there is no real uio
2855 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2857 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2858 int error = 0, wccflag = NFSV3_WCCRATTR;
2859 struct nfsm_info info;
2865 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2867 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2868 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
2869 ERROROUT(nfsm_fhtom(&info, vp));
2870 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
2871 txdr_hyper(offset, tl);
2873 *tl = txdr_unsigned(cnt);
2874 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
2875 nfs_vpcred(vp, ND_WRITE), &error));
2876 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
2878 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
2879 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2880 NFSX_V3WRITEVERF)) {
2881 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2883 error = NFSERR_STALEWRITEVERF;
2894 * - make nfs_bmap() essentially a no-op that does no translation
2895 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2896 * (Maybe I could use the process's page mapping, but I was concerned that
2897 * Kernel Write might not be enabled and also figured copyout() would do
2898 * a lot more work than bcopy() and also it currently happens in the
2899 * context of the swapper process (2).
2901 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
2902 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2905 nfs_bmap(struct vop_bmap_args *ap)
2907 if (ap->a_doffsetp != NULL)
2908 *ap->a_doffsetp = ap->a_loffset;
2909 if (ap->a_runp != NULL)
2911 if (ap->a_runb != NULL)
2920 nfs_strategy(struct vop_strategy_args *ap)
2922 struct bio *bio = ap->a_bio;
2924 struct buf *bp = bio->bio_buf;
2928 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2929 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
2930 KASSERT(BUF_REFCNT(bp) > 0,
2931 ("nfs_strategy: buffer %p not locked", bp));
2933 if (bio->bio_flags & BIO_SYNC)
2934 td = curthread; /* XXX */
2939 * We probably don't need to push an nbio any more since no
2940 * block conversion is required due to the use of 64 bit byte
2941 * offsets, but do it anyway.
2943 * NOTE: When NFS callers itself via this strategy routines and
2944 * sets up a synchronous I/O, it expects the I/O to run
2945 * synchronously (its bio_done routine just assumes it),
2946 * so for now we have to honor the bit.
2948 nbio = push_bio(bio);
2949 nbio->bio_offset = bio->bio_offset;
2950 nbio->bio_flags = bio->bio_flags & BIO_SYNC;
2953 * If the op is asynchronous and an i/o daemon is waiting
2954 * queue the request, wake it up and wait for completion
2955 * otherwise just do it ourselves.
2957 if (bio->bio_flags & BIO_SYNC) {
2958 error = nfs_doio(ap->a_vp, nbio, td);
2960 nfs_asyncio(ap->a_vp, nbio);
2969 * NB Currently unsupported.
2971 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
2975 nfs_mmap(struct vop_mmap_args *ap)
2981 * fsync vnode op. Just call nfs_flush() with commit == 1.
2983 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
2987 nfs_fsync(struct vop_fsync_args *ap)
2989 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
2993 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2994 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2995 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2996 * set the buffer contains data that has already been written to the server
2997 * and which now needs a commit RPC.
2999 * If commit is 0 we only take one pass and only flush buffers containing new
3002 * If commit is 1 we take two passes, issuing a commit RPC in the second
3005 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3006 * to completely flush all pending data.
3008 * Note that the RB_SCAN code properly handles the case where the
3009 * callback might block and directly or indirectly (another thread) cause
3010 * the RB tree to change.
3013 #ifndef NFS_COMMITBVECSIZ
3014 #define NFS_COMMITBVECSIZ 16
3017 struct nfs_flush_info {
3018 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3025 struct buf *bvary[NFS_COMMITBVECSIZ];
3031 static int nfs_flush_bp(struct buf *bp, void *data);
3032 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3035 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3037 struct nfsnode *np = VTONFS(vp);
3038 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3039 struct nfs_flush_info info;
3043 bzero(&info, sizeof(info));
3046 info.waitfor = waitfor;
3047 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3049 lwkt_gettoken(&vlock, &vp->v_token);
3055 info.mode = NFI_FLUSHNEW;
3056 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3057 nfs_flush_bp, &info);
3060 * Take a second pass if committing and no error occured.
3061 * Clean up any left over collection (whether an error
3064 if (commit && error == 0) {
3065 info.mode = NFI_COMMIT;
3066 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3067 nfs_flush_bp, &info);
3069 error = nfs_flush_docommit(&info, error);
3073 * Wait for pending I/O to complete before checking whether
3074 * any further dirty buffers exist.
3076 while (waitfor == MNT_WAIT &&
3077 bio_track_active(&vp->v_track_write)) {
3078 error = bio_track_wait(&vp->v_track_write,
3079 info.slpflag, info.slptimeo);
3082 * We have to be able to break out if this
3083 * is an 'intr' mount.
3085 if (nfs_sigintr(nmp, NULL, td)) {
3091 * Since we do not process pending signals,
3092 * once we get a PCATCH our tsleep() will no
3093 * longer sleep, switch to a fixed timeout
3096 if (info.slpflag == PCATCH) {
3098 info.slptimeo = 2 * hz;
3105 * Loop if we are flushing synchronous as well as committing,
3106 * and dirty buffers are still present. Otherwise we might livelock.
3108 } while (waitfor == MNT_WAIT && commit &&
3109 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3112 * The callbacks have to return a negative error to terminate the
3119 * Deal with any error collection
3121 if (np->n_flag & NWRITEERR) {
3122 error = np->n_error;
3123 np->n_flag &= ~NWRITEERR;
3125 lwkt_reltoken(&vlock);
3131 nfs_flush_bp(struct buf *bp, void *data)
3133 struct nfs_flush_info *info = data;
3139 switch(info->mode) {
3141 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3142 if (error && info->loops && info->waitfor == MNT_WAIT) {
3143 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3145 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3146 if (info->slpflag & PCATCH)
3147 lkflags |= LK_PCATCH;
3148 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3154 * Ignore locking errors
3162 * The buffer may have changed out from under us, even if
3163 * we did not block (MPSAFE). Check again now that it is
3166 if (bp->b_vp == info->vp &&
3167 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3176 * Only process buffers in need of a commit which we can
3177 * immediately lock. This may prevent a buffer from being
3178 * committed, but the normal flush loop will block on the
3179 * same buffer so we shouldn't get into an endless loop.
3181 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3182 (B_DELWRI | B_NEEDCOMMIT)) {
3185 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3189 * We must recheck after successfully locking the buffer.
3191 if (bp->b_vp != info->vp ||
3192 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3193 (B_DELWRI | B_NEEDCOMMIT)) {
3199 * NOTE: storing the bp in the bvary[] basically sets
3200 * it up for a commit operation.
3202 * We must call vfs_busy_pages() now so the commit operation
3203 * is interlocked with user modifications to memory mapped
3206 * Note: to avoid loopback deadlocks, we do not
3207 * assign b_runningbufspace.
3210 bp->b_cmd = BUF_CMD_WRITE;
3211 vfs_busy_pages(bp->b_vp, bp);
3212 info->bvary[info->bvsize] = bp;
3213 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3214 if (info->bvsize == 0 || toff < info->beg_off)
3215 info->beg_off = toff;
3216 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3217 if (info->bvsize == 0 || toff > info->end_off)
3218 info->end_off = toff;
3220 if (info->bvsize == NFS_COMMITBVECSIZ) {
3221 error = nfs_flush_docommit(info, 0);
3222 KKASSERT(info->bvsize == 0);
3230 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3240 if (info->bvsize > 0) {
3242 * Commit data on the server, as required. Note that
3243 * nfs_commit will use the vnode's cred for the commit.
3244 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3246 bytes = info->end_off - info->beg_off;
3247 if (bytes > 0x40000000)
3252 retv = nfs_commitrpc_uio(vp, info->beg_off,
3253 (int)bytes, info->td);
3254 if (retv == NFSERR_STALEWRITEVERF)
3255 nfs_clearcommit(vp->v_mount);
3259 * Now, either mark the blocks I/O done or mark the
3260 * blocks dirty, depending on whether the commit
3263 for (i = 0; i < info->bvsize; ++i) {
3264 bp = info->bvary[i];
3265 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3268 * Error, leave B_DELWRI intact
3270 vfs_unbusy_pages(bp);
3271 bp->b_cmd = BUF_CMD_DONE;
3275 * Success, remove B_DELWRI ( bundirty() ).
3277 * b_dirtyoff/b_dirtyend seem to be NFS
3278 * specific. We should probably move that
3279 * into bundirty(). XXX
3281 * We are faking an I/O write, we have to
3282 * start the transaction in order to
3283 * immediately biodone() it.
3286 bp->b_flags &= ~B_ERROR;
3287 bp->b_dirtyoff = bp->b_dirtyend = 0;
3288 biodone(&bp->b_bio1);
3297 * NFS advisory byte-level locks.
3298 * Currently unsupported.
3300 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3304 nfs_advlock(struct vop_advlock_args *ap)
3306 struct nfsnode *np = VTONFS(ap->a_vp);
3309 * The following kludge is to allow diskless support to work
3310 * until a real NFS lockd is implemented. Basically, just pretend
3311 * that this is a local lock.
3313 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3317 * Print out the contents of an nfsnode.
3319 * nfs_print(struct vnode *a_vp)
3322 nfs_print(struct vop_print_args *ap)
3324 struct vnode *vp = ap->a_vp;
3325 struct nfsnode *np = VTONFS(vp);
3327 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3328 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3329 if (vp->v_type == VFIFO)
3336 * nfs special file access vnode op.
3337 * Essentially just get vattr and then imitate iaccess() since the device is
3338 * local to the client.
3340 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3343 nfs_laccess(struct vop_access_args *ap)
3347 struct ucred *cred = ap->a_cred;
3348 struct vnode *vp = ap->a_vp;
3349 mode_t mode = ap->a_mode;
3355 * Disallow write attempts on filesystems mounted read-only;
3356 * unless the file is a socket, fifo, or a block or character
3357 * device resident on the filesystem.
3359 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3360 switch (vp->v_type) {
3370 * If you're the super-user,
3371 * you always get access.
3373 if (cred->cr_uid == 0)
3376 error = VOP_GETATTR(vp, vap);
3380 * Access check is based on only one of owner, group, public.
3381 * If not owner, then check group. If not a member of the
3382 * group, then check public access.
3384 if (cred->cr_uid != vap->va_uid) {
3386 gp = cred->cr_groups;
3387 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3388 if (vap->va_gid == *gp)
3394 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3399 * Read wrapper for fifos.
3401 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3402 * struct ucred *a_cred)
3405 nfsfifo_read(struct vop_read_args *ap)
3407 struct nfsnode *np = VTONFS(ap->a_vp);
3413 getnanotime(&np->n_atim);
3414 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3418 * Write wrapper for fifos.
3420 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3421 * struct ucred *a_cred)
3424 nfsfifo_write(struct vop_write_args *ap)
3426 struct nfsnode *np = VTONFS(ap->a_vp);
3432 getnanotime(&np->n_mtim);
3433 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3437 * Close wrapper for fifos.
3439 * Update the times on the nfsnode then do fifo close.
3441 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3444 nfsfifo_close(struct vop_close_args *ap)
3446 struct vnode *vp = ap->a_vp;
3447 struct nfsnode *np = VTONFS(vp);
3451 if (np->n_flag & (NACC | NUPD)) {
3453 if (np->n_flag & NACC)
3455 if (np->n_flag & NUPD)
3458 if (vp->v_sysref.refcnt == 1 &&
3459 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3461 if (np->n_flag & NACC)
3462 vattr.va_atime = np->n_atim;
3463 if (np->n_flag & NUPD)
3464 vattr.va_mtime = np->n_mtim;
3465 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3468 return (VOCALL(&fifo_vnode_vops, &ap->a_head));