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 $
42 * vnode op calls for Sun NFS version 2 and 3
47 #include <sys/param.h>
48 #include <sys/kernel.h>
49 #include <sys/systm.h>
50 #include <sys/resourcevar.h>
52 #include <sys/mount.h>
54 #include <sys/malloc.h>
56 #include <sys/namei.h>
57 #include <sys/nlookup.h>
58 #include <sys/socket.h>
59 #include <sys/vnode.h>
60 #include <sys/dirent.h>
61 #include <sys/fcntl.h>
62 #include <sys/lockf.h>
64 #include <sys/sysctl.h>
68 #include <vm/vm_extern.h>
72 #include <vfs/fifofs/fifo.h>
73 #include <vfs/ufs/dir.h>
83 #include "nfsm_subs.h"
86 #include <netinet/in.h>
87 #include <netinet/in_var.h>
89 #include <sys/thread2.h>
95 static int nfsfifo_read (struct vop_read_args *);
96 static int nfsfifo_write (struct vop_write_args *);
97 static int nfsfifo_close (struct vop_close_args *);
98 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
99 static int nfs_lookup (struct vop_old_lookup_args *);
100 static int nfs_create (struct vop_old_create_args *);
101 static int nfs_mknod (struct vop_old_mknod_args *);
102 static int nfs_open (struct vop_open_args *);
103 static int nfs_close (struct vop_close_args *);
104 static int nfs_access (struct vop_access_args *);
105 static int nfs_getattr (struct vop_getattr_args *);
106 static int nfs_setattr (struct vop_setattr_args *);
107 static int nfs_read (struct vop_read_args *);
108 static int nfs_mmap (struct vop_mmap_args *);
109 static int nfs_fsync (struct vop_fsync_args *);
110 static int nfs_remove (struct vop_old_remove_args *);
111 static int nfs_link (struct vop_old_link_args *);
112 static int nfs_rename (struct vop_old_rename_args *);
113 static int nfs_mkdir (struct vop_old_mkdir_args *);
114 static int nfs_rmdir (struct vop_old_rmdir_args *);
115 static int nfs_symlink (struct vop_old_symlink_args *);
116 static int nfs_readdir (struct vop_readdir_args *);
117 static int nfs_bmap (struct vop_bmap_args *);
118 static int nfs_strategy (struct vop_strategy_args *);
119 static int nfs_lookitup (struct vnode *, const char *, int,
120 struct ucred *, struct thread *, struct nfsnode **);
121 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
122 static int nfs_laccess (struct vop_access_args *);
123 static int nfs_readlink (struct vop_readlink_args *);
124 static int nfs_print (struct vop_print_args *);
125 static int nfs_advlock (struct vop_advlock_args *);
127 static int nfs_nresolve (struct vop_nresolve_args *);
129 * Global vfs data structures for nfs
131 struct vop_ops nfsv2_vnode_vops = {
132 .vop_default = vop_defaultop,
133 .vop_access = nfs_access,
134 .vop_advlock = nfs_advlock,
135 .vop_bmap = nfs_bmap,
136 .vop_close = nfs_close,
137 .vop_old_create = nfs_create,
138 .vop_fsync = nfs_fsync,
139 .vop_getattr = nfs_getattr,
140 .vop_getpages = vop_stdgetpages,
141 .vop_putpages = vop_stdputpages,
142 .vop_inactive = nfs_inactive,
143 .vop_old_link = nfs_link,
144 .vop_old_lookup = nfs_lookup,
145 .vop_old_mkdir = nfs_mkdir,
146 .vop_old_mknod = nfs_mknod,
147 .vop_mmap = nfs_mmap,
148 .vop_open = nfs_open,
149 .vop_print = nfs_print,
150 .vop_read = nfs_read,
151 .vop_readdir = nfs_readdir,
152 .vop_readlink = nfs_readlink,
153 .vop_reclaim = nfs_reclaim,
154 .vop_old_remove = nfs_remove,
155 .vop_old_rename = nfs_rename,
156 .vop_old_rmdir = nfs_rmdir,
157 .vop_setattr = nfs_setattr,
158 .vop_strategy = nfs_strategy,
159 .vop_old_symlink = nfs_symlink,
160 .vop_write = nfs_write,
161 .vop_nresolve = nfs_nresolve
165 * Special device vnode ops
167 struct vop_ops nfsv2_spec_vops = {
168 .vop_default = vop_defaultop,
169 .vop_access = nfs_laccess,
170 .vop_close = nfs_close,
171 .vop_fsync = nfs_fsync,
172 .vop_getattr = nfs_getattr,
173 .vop_inactive = nfs_inactive,
174 .vop_print = nfs_print,
175 .vop_read = vop_stdnoread,
176 .vop_reclaim = nfs_reclaim,
177 .vop_setattr = nfs_setattr,
178 .vop_write = vop_stdnowrite
181 struct vop_ops nfsv2_fifo_vops = {
182 .vop_default = fifo_vnoperate,
183 .vop_access = nfs_laccess,
184 .vop_close = nfsfifo_close,
185 .vop_fsync = nfs_fsync,
186 .vop_getattr = nfs_getattr,
187 .vop_inactive = nfs_inactive,
188 .vop_print = nfs_print,
189 .vop_read = nfsfifo_read,
190 .vop_reclaim = nfs_reclaim,
191 .vop_setattr = nfs_setattr,
192 .vop_write = nfsfifo_write
195 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
196 struct componentname *cnp,
198 static int nfs_removerpc (struct vnode *dvp, const char *name,
200 struct ucred *cred, struct thread *td);
201 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
202 int fnamelen, struct vnode *tdvp,
203 const char *tnameptr, int tnamelen,
204 struct ucred *cred, struct thread *td);
205 static int nfs_renameit (struct vnode *sdvp,
206 struct componentname *scnp,
207 struct sillyrename *sp);
209 SYSCTL_DECL(_vfs_nfs);
211 static int nfs_flush_on_rename = 1;
212 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
213 &nfs_flush_on_rename, 0, "flush fvp prior to rename");
214 static int nfs_flush_on_hlink = 0;
215 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
216 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
218 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
219 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
220 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
222 static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
223 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
224 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
226 static int nfspos_cache_timeout = NFS_MINATTRTIMO;
227 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
228 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
230 static int nfsv3_commit_on_close = 0;
231 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
232 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
235 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
237 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
238 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
241 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
242 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
243 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
246 * Returns whether a name component is a degenerate '.' or '..'.
250 nlcdegenerate(struct nlcomponent *nlc)
252 if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.')
254 if (nlc->nlc_namelen == 2 &&
255 nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.')
261 nfs3_access_otw(struct vnode *vp, int wmode,
262 struct thread *td, struct ucred *cred)
264 struct nfsnode *np = VTONFS(vp);
269 struct nfsm_info info;
274 nfsstats.rpccnt[NFSPROC_ACCESS]++;
275 nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
276 NFSX_FH(info.v3) + NFSX_UNSIGNED);
277 ERROROUT(nfsm_fhtom(&info, vp));
278 tl = nfsm_build(&info, NFSX_UNSIGNED);
279 *tl = txdr_unsigned(wmode);
280 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
281 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
283 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
284 rmode = fxdr_unsigned(u_int32_t, *tl);
286 np->n_modeuid = cred->cr_uid;
287 np->n_modestamp = mycpu->gd_time_seconds;
296 * nfs access vnode op.
297 * For nfs version 2, just return ok. File accesses may fail later.
298 * For nfs version 3, use the access rpc to check accessibility. If file modes
299 * are changed on the server, accesses might still fail later.
301 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
304 nfs_access(struct vop_access_args *ap)
307 struct vnode *vp = ap->a_vp;
308 thread_t td = curthread;
310 u_int32_t mode, wmode;
311 struct nfsnode *np = VTONFS(vp);
312 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
313 int v3 = NFS_ISV3(vp);
315 lwkt_gettoken(&nmp->nm_token);
318 * Disallow write attempts on filesystems mounted read-only;
319 * unless the file is a socket, fifo, or a block or character
320 * device resident on the filesystem.
322 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
323 switch (vp->v_type) {
327 lwkt_reltoken(&nmp->nm_token);
335 * The NFS protocol passes only the effective uid/gid over the wire but
336 * we need to check access against real ids if AT_EACCESS not set.
337 * Handle this case by cloning the credentials and setting the
338 * effective ids to the real ones.
340 if (ap->a_flags & AT_EACCESS) {
341 cred = crhold(ap->a_cred);
343 cred = crdup(ap->a_cred);
344 cred->cr_uid = cred->cr_ruid;
345 cred->cr_gid = cred->cr_rgid;
349 * For nfs v3, check to see if we have done this recently, and if
350 * so return our cached result instead of making an ACCESS call.
351 * If not, do an access rpc, otherwise you are stuck emulating
352 * ufs_access() locally using the vattr. This may not be correct,
353 * since the server may apply other access criteria such as
354 * client uid-->server uid mapping that we do not know about.
357 if (ap->a_mode & VREAD)
358 mode = NFSV3ACCESS_READ;
361 if (vp->v_type != VDIR) {
362 if (ap->a_mode & VWRITE)
363 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
364 if (ap->a_mode & VEXEC)
365 mode |= NFSV3ACCESS_EXECUTE;
367 if (ap->a_mode & VWRITE)
368 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
370 if (ap->a_mode & VEXEC)
371 mode |= NFSV3ACCESS_LOOKUP;
373 /* XXX safety belt, only make blanket request if caching */
374 if (nfsaccess_cache_timeout > 0) {
375 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
376 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
377 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
383 * Does our cached result allow us to give a definite yes to
386 if (np->n_modestamp &&
387 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
388 (cred->cr_uid == np->n_modeuid) &&
389 ((np->n_mode & mode) == mode)) {
390 nfsstats.accesscache_hits++;
393 * Either a no, or a don't know. Go to the wire.
395 nfsstats.accesscache_misses++;
396 error = nfs3_access_otw(vp, wmode, td, cred);
398 if ((np->n_mode & mode) != mode) {
404 if ((error = nfs_laccess(ap)) != 0) {
406 lwkt_reltoken(&nmp->nm_token);
411 * Attempt to prevent a mapped root from accessing a file
412 * which it shouldn't. We try to read a byte from the file
413 * if the user is root and the file is not zero length.
414 * After calling nfs_laccess, we should have the correct
417 if (cred->cr_uid == 0 && (ap->a_mode & VREAD)
418 && VTONFS(vp)->n_size > 0) {
425 auio.uio_iov = &aiov;
429 auio.uio_segflg = UIO_SYSSPACE;
430 auio.uio_rw = UIO_READ;
433 if (vp->v_type == VREG) {
434 error = nfs_readrpc_uio(vp, &auio);
435 } else if (vp->v_type == VDIR) {
437 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
439 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
440 error = nfs_readdirrpc_uio(vp, &auio);
442 } else if (vp->v_type == VLNK) {
443 error = nfs_readlinkrpc_uio(vp, &auio);
450 * [re]record creds for reading and/or writing if access
451 * was granted. Assume the NFS server will grant read access
452 * for execute requests.
455 if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) {
458 crfree(np->n_rucred);
461 if ((ap->a_mode & VWRITE) && cred != np->n_wucred) {
464 crfree(np->n_wucred);
468 lwkt_reltoken(&nmp->nm_token);
475 * Check to see if the type is ok
476 * and that deletion is not in progress.
477 * For paged in text files, you will need to flush the page cache
478 * if consistency is lost.
480 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
485 nfs_open(struct vop_open_args *ap)
487 struct vnode *vp = ap->a_vp;
488 struct nfsnode *np = VTONFS(vp);
489 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
493 lwkt_gettoken(&nmp->nm_token);
495 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
497 kprintf("open eacces vtyp=%d\n",vp->v_type);
499 lwkt_reltoken(&nmp->nm_token);
504 * Save valid creds for reading and writing for later RPCs.
506 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
509 crfree(np->n_rucred);
510 np->n_rucred = ap->a_cred;
512 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
515 crfree(np->n_wucred);
516 np->n_wucred = ap->a_cred;
520 * Clear the attribute cache only if opening with write access. It
521 * is unclear if we should do this at all here, but we certainly
522 * should not clear the cache unconditionally simply because a file
525 if (ap->a_mode & FWRITE)
529 * For normal NFS, reconcile changes made locally verses
530 * changes made remotely. Note that VOP_GETATTR only goes
531 * to the wire if the cached attribute has timed out or been
534 * If local modifications have been made clear the attribute
535 * cache to force an attribute and modified time check. If
536 * GETATTR detects that the file has been changed by someone
537 * other then us it will set NRMODIFIED.
539 * If we are opening a directory and local changes have been
540 * made we have to invalidate the cache in order to ensure
541 * that we get the most up-to-date information from the
544 if (np->n_flag & NLMODIFIED) {
546 if (vp->v_type == VDIR) {
547 error = nfs_vinvalbuf(vp, V_SAVE, 1);
553 error = VOP_GETATTR(vp, &vattr);
555 lwkt_reltoken(&nmp->nm_token);
558 if (np->n_flag & NRMODIFIED) {
559 if (vp->v_type == VDIR)
561 error = nfs_vinvalbuf(vp, V_SAVE, 1);
562 if (error == EINTR) {
563 lwkt_reltoken(&nmp->nm_token);
566 np->n_flag &= ~NRMODIFIED;
568 error = vop_stdopen(ap);
569 lwkt_reltoken(&nmp->nm_token);
576 * What an NFS client should do upon close after writing is a debatable issue.
577 * Most NFS clients push delayed writes to the server upon close, basically for
579 * 1 - So that any write errors may be reported back to the client process
580 * doing the close system call. By far the two most likely errors are
581 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
582 * 2 - To put a worst case upper bound on cache inconsistency between
583 * multiple clients for the file.
584 * There is also a consistency problem for Version 2 of the protocol w.r.t.
585 * not being able to tell if other clients are writing a file concurrently,
586 * since there is no way of knowing if the changed modify time in the reply
587 * is only due to the write for this client.
588 * (NFS Version 3 provides weak cache consistency data in the reply that
589 * should be sufficient to detect and handle this case.)
591 * The current code does the following:
592 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
593 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
594 * or commit them (this satisfies 1 and 2 except for the
595 * case where the server crashes after this close but
596 * before the commit RPC, which is felt to be "good
597 * enough". Changing the last argument to nfs_flush() to
598 * a 1 would force a commit operation, if it is felt a
599 * commit is necessary now.
600 * for NQNFS - do nothing now, since 2 is dealt with via leases and
601 * 1 should be dealt with via an fsync() system call for
602 * cases where write errors are important.
604 * nfs_close(struct vnode *a_vp, int a_fflag)
608 nfs_close(struct vop_close_args *ap)
610 struct vnode *vp = ap->a_vp;
611 struct nfsnode *np = VTONFS(vp);
612 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
614 thread_t td = curthread;
616 lwkt_gettoken(&nmp->nm_token);
618 if (vp->v_type == VREG) {
619 if (np->n_flag & NLMODIFIED) {
622 * Under NFSv3 we have dirty buffers to dispose of. We
623 * must flush them to the NFS server. We have the option
624 * of waiting all the way through the commit rpc or just
625 * waiting for the initial write. The default is to only
626 * wait through the initial write so the data is in the
627 * server's cache, which is roughly similar to the state
628 * a standard disk subsystem leaves the file in on close().
630 * We cannot clear the NLMODIFIED bit in np->n_flag due to
631 * potential races with other processes, and certainly
632 * cannot clear it if we don't commit.
634 int cm = nfsv3_commit_on_close ? 1 : 0;
635 error = nfs_flush(vp, MNT_WAIT, td, cm);
636 /* np->n_flag &= ~NLMODIFIED; */
638 error = nfs_vinvalbuf(vp, V_SAVE, 1);
642 if (np->n_flag & NWRITEERR) {
643 np->n_flag &= ~NWRITEERR;
648 lwkt_reltoken(&nmp->nm_token);
654 * nfs getattr call from vfs.
656 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
659 nfs_getattr(struct vop_getattr_args *ap)
661 struct vnode *vp = ap->a_vp;
662 struct nfsnode *np = VTONFS(vp);
663 struct nfsmount *nmp;
665 thread_t td = curthread;
666 struct nfsm_info info;
669 info.v3 = NFS_ISV3(vp);
670 nmp = VFSTONFS(vp->v_mount);
672 lwkt_gettoken(&nmp->nm_token);
675 * Update local times for special files.
677 if (np->n_flag & (NACC | NUPD))
680 * First look in the cache.
682 if (nfs_getattrcache(vp, ap->a_vap) == 0)
685 if (info.v3 && nfsaccess_cache_timeout > 0) {
686 nfsstats.accesscache_misses++;
687 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
688 if (nfs_getattrcache(vp, ap->a_vap) == 0)
692 nfsstats.rpccnt[NFSPROC_GETATTR]++;
693 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
694 ERROROUT(nfsm_fhtom(&info, vp));
695 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
696 nfs_vpcred(vp, ND_CHECK), &error));
698 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
704 * NFS doesn't support chflags flags. If the nfs mount was
705 * made -o cache set the UF_CACHE bit for swapcache.
707 if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT))
708 ap->a_vap->va_flags |= UF_CACHE;
710 lwkt_reltoken(&nmp->nm_token);
717 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
720 nfs_setattr(struct vop_setattr_args *ap)
722 struct vnode *vp = ap->a_vp;
723 struct nfsnode *np = VTONFS(vp);
724 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
725 struct vattr *vap = ap->a_vap;
726 int biosize = vp->v_mount->mnt_stat.f_iosize;
730 thread_t td = curthread;
736 * Setting of flags is not supported.
738 if (vap->va_flags != VNOVAL)
742 * Disallow write attempts if the filesystem is mounted read-only.
744 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
745 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
746 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
747 (vp->v_mount->mnt_flag & MNT_RDONLY))
750 lwkt_gettoken(&nmp->nm_token);
752 if (vap->va_size != VNOVAL) {
754 * truncation requested
756 switch (vp->v_type) {
758 lwkt_reltoken(&nmp->nm_token);
764 if (vap->va_mtime.tv_sec == VNOVAL &&
765 vap->va_atime.tv_sec == VNOVAL &&
766 vap->va_mode == (mode_t)VNOVAL &&
767 vap->va_uid == (uid_t)VNOVAL &&
768 vap->va_gid == (gid_t)VNOVAL) {
769 lwkt_reltoken(&nmp->nm_token);
772 vap->va_size = VNOVAL;
776 * Disallow write attempts if the filesystem is
779 if (vp->v_mount->mnt_flag & MNT_RDONLY) {
780 lwkt_reltoken(&nmp->nm_token);
786 boff = (int)vap->va_size & (biosize - 1);
787 error = nfs_meta_setsize(vp, td, vap->va_size, 0);
790 if (np->n_flag & NLMODIFIED) {
791 if (vap->va_size == 0)
792 error = nfs_vinvalbuf(vp, 0, 1);
794 error = nfs_vinvalbuf(vp, V_SAVE, 1);
798 * note: this loop case almost always happens at
799 * least once per truncation.
801 if (error == 0 && np->n_size != vap->va_size)
803 np->n_vattr.va_size = vap->va_size;
806 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
808 * What to do. If we are modifying the mtime we lose
809 * mtime detection of changes made by the server or other
810 * clients. But programs like rsync/rdist/cpdup are going
811 * to call utimes a lot. We don't want to piecemeal sync.
813 * For now sync if any prior remote changes were detected,
814 * but allow us to lose track of remote changes made during
815 * the utimes operation.
817 if (np->n_flag & NRMODIFIED)
818 error = nfs_vinvalbuf(vp, V_SAVE, 1);
822 if (vap->va_mtime.tv_sec != VNOVAL) {
823 np->n_mtime = vap->va_mtime.tv_sec;
827 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
830 * Sanity check if a truncation was issued. This should only occur
831 * if multiple processes are racing on the same file.
833 if (error == 0 && vap->va_size != VNOVAL &&
834 np->n_size != vap->va_size) {
835 kprintf("NFS ftruncate: server disagrees on the file size: "
838 (intmax_t)vap->va_size,
839 (intmax_t)np->n_size);
842 if (error && vap->va_size != VNOVAL) {
843 np->n_size = np->n_vattr.va_size = tsize;
844 nfs_meta_setsize(vp, td, np->n_size, 0);
846 lwkt_reltoken(&nmp->nm_token);
852 * Do an nfs setattr rpc.
855 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
856 struct ucred *cred, struct thread *td)
858 struct nfsv2_sattr *sp;
859 struct nfsnode *np = VTONFS(vp);
861 int error = 0, wccflag = NFSV3_WCCRATTR;
862 struct nfsm_info info;
865 info.v3 = NFS_ISV3(vp);
867 nfsstats.rpccnt[NFSPROC_SETATTR]++;
868 nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
869 NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
870 ERROROUT(nfsm_fhtom(&info, vp));
872 nfsm_v3attrbuild(&info, vap, TRUE);
873 tl = nfsm_build(&info, NFSX_UNSIGNED);
876 sp = nfsm_build(&info, NFSX_V2SATTR);
877 if (vap->va_mode == (mode_t)VNOVAL)
878 sp->sa_mode = nfs_xdrneg1;
880 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
881 if (vap->va_uid == (uid_t)VNOVAL)
882 sp->sa_uid = nfs_xdrneg1;
884 sp->sa_uid = txdr_unsigned(vap->va_uid);
885 if (vap->va_gid == (gid_t)VNOVAL)
886 sp->sa_gid = nfs_xdrneg1;
888 sp->sa_gid = txdr_unsigned(vap->va_gid);
889 sp->sa_size = txdr_unsigned(vap->va_size);
890 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
891 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
893 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
896 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
898 ERROROUT(nfsm_loadattr(&info, vp, NULL));
908 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
914 cache_setvp(nch, vp);
915 cache_settimeout(nch, nctimeout);
919 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
920 * nfs_lookup() until all remaining new api calls are implemented.
922 * Resolve a namecache entry. This function is passed a locked ncp and
923 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
926 nfs_nresolve(struct vop_nresolve_args *ap)
928 struct thread *td = curthread;
929 struct namecache *ncp;
930 struct nfsmount *nmp;
941 struct nfsm_info info;
945 nmp = VFSTONFS(dvp->v_mount);
947 lwkt_gettoken(&nmp->nm_token);
949 if ((error = vget(dvp, LK_SHARED)) != 0) {
950 lwkt_reltoken(&nmp->nm_token);
955 info.v3 = NFS_ISV3(dvp);
958 nfsstats.lookupcache_misses++;
959 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
960 ncp = ap->a_nch->ncp;
962 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
963 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
964 ERROROUT(nfsm_fhtom(&info, dvp));
965 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
966 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
967 ap->a_cred, &error));
970 * Cache negatve lookups to reduce NFS traffic, but use
971 * a fast timeout. Otherwise use a timeout of 1 tick.
972 * XXX we should add a namecache flag for no-caching
973 * to uncache the negative hit as soon as possible, but
974 * we cannot simply destroy the entry because it is used
975 * as a placeholder by the caller.
977 * The refactored nfs code will overwrite a non-zero error
978 * with 0 when we use ERROROUT(), so don't here.
981 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
982 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
994 * Success, get the file handle, do various checks, and load
995 * post-operation data from the reply packet. Theoretically
996 * we should never be looking up "." so, theoretically, we
997 * should never get the same file handle as our directory. But
998 * we check anyway. XXX
1000 * Note that no timeout is set for the positive cache hit. We
1001 * assume, theoretically, that ESTALE returns will be dealt with
1002 * properly to handle NFS races and in anycase we cannot depend
1003 * on a timeout to deal with NFS open/create/excl issues so instead
1004 * of a bad hack here the rest of the NFS client code needs to do
1007 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1010 if (NFS_CMPFH(np, fhp, fhsize)) {
1014 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1019 lwkt_reltoken(&nmp->nm_token);
1025 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
1026 NFS_LATTR_NOSHRINK));
1027 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1028 NFS_LATTR_NOSHRINK));
1030 ERROROUT(nfsm_loadattr(&info, nvp, NULL));
1032 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
1036 lwkt_reltoken(&nmp->nm_token);
1048 * 'cached' nfs directory lookup
1050 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
1052 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
1053 * struct componentname *a_cnp)
1056 nfs_lookup(struct vop_old_lookup_args *ap)
1058 struct componentname *cnp = ap->a_cnp;
1059 struct vnode *dvp = ap->a_dvp;
1060 struct vnode **vpp = ap->a_vpp;
1061 int flags = cnp->cn_flags;
1062 struct vnode *newvp;
1063 struct nfsmount *nmp;
1067 int lockparent, wantparent, attrflag, fhsize;
1070 struct nfsm_info info;
1073 info.v3 = NFS_ISV3(dvp);
1077 * Read-only mount check and directory check.
1080 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1081 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1084 if (dvp->v_type != VDIR)
1088 * Look it up in the cache. Note that ENOENT is only returned if we
1089 * previously entered a negative hit (see later on). The additional
1090 * nfsneg_cache_timeout check causes previously cached results to
1091 * be instantly ignored if the negative caching is turned off.
1093 lockparent = flags & CNP_LOCKPARENT;
1094 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1095 nmp = VFSTONFS(dvp->v_mount);
1098 lwkt_gettoken(&nmp->nm_token);
1105 nfsstats.lookupcache_misses++;
1106 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1107 len = cnp->cn_namelen;
1108 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1109 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1110 ERROROUT(nfsm_fhtom(&info, dvp));
1111 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1112 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1113 cnp->cn_cred, &error));
1115 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1116 NFS_LATTR_NOSHRINK);
1126 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1129 * Handle RENAME case...
1131 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1132 if (NFS_CMPFH(np, fhp, fhsize)) {
1135 lwkt_reltoken(&nmp->nm_token);
1138 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1142 lwkt_reltoken(&nmp->nm_token);
1147 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1148 NFS_LATTR_NOSHRINK));
1149 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1150 NFS_LATTR_NOSHRINK));
1152 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1159 cnp->cn_flags |= CNP_PDIRUNLOCK;
1161 lwkt_reltoken(&nmp->nm_token);
1165 if (flags & CNP_ISDOTDOT) {
1167 cnp->cn_flags |= CNP_PDIRUNLOCK;
1168 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1170 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1171 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1172 lwkt_reltoken(&nmp->nm_token);
1173 return (error); /* NOTE: return error from nget */
1177 error = vn_lock(dvp, LK_EXCLUSIVE);
1180 lwkt_reltoken(&nmp->nm_token);
1183 cnp->cn_flags |= CNP_PDIRUNLOCK;
1185 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1189 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1193 lwkt_reltoken(&nmp->nm_token);
1198 cnp->cn_flags |= CNP_PDIRUNLOCK;
1203 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1204 NFS_LATTR_NOSHRINK));
1205 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1206 NFS_LATTR_NOSHRINK));
1208 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1211 /* XXX MOVE TO nfs_nremove() */
1212 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1213 cnp->cn_nameiop != NAMEI_DELETE) {
1214 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1222 if (newvp != NULLVP) {
1226 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1227 cnp->cn_nameiop == NAMEI_RENAME) &&
1231 cnp->cn_flags |= CNP_PDIRUNLOCK;
1233 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1236 error = EJUSTRETURN;
1239 lwkt_reltoken(&nmp->nm_token);
1245 * Just call nfs_bioread() to do the work.
1247 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1248 * struct ucred *a_cred)
1251 nfs_read(struct vop_read_args *ap)
1253 struct vnode *vp = ap->a_vp;
1254 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1257 lwkt_gettoken(&nmp->nm_token);
1258 error = nfs_bioread(vp, ap->a_uio, ap->a_ioflag);
1259 lwkt_reltoken(&nmp->nm_token);
1267 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1270 nfs_readlink(struct vop_readlink_args *ap)
1272 struct vnode *vp = ap->a_vp;
1273 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1276 if (vp->v_type != VLNK)
1279 lwkt_gettoken(&nmp->nm_token);
1280 error = nfs_bioread(vp, ap->a_uio, 0);
1281 lwkt_reltoken(&nmp->nm_token);
1287 * Do a readlink rpc.
1288 * Called by nfs_doio() from below the buffer cache.
1291 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1293 int error = 0, len, attrflag;
1294 struct nfsm_info info;
1297 info.v3 = NFS_ISV3(vp);
1299 nfsstats.rpccnt[NFSPROC_READLINK]++;
1300 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1301 ERROROUT(nfsm_fhtom(&info, vp));
1302 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1303 nfs_vpcred(vp, ND_CHECK), &error));
1305 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1306 NFS_LATTR_NOSHRINK));
1309 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1310 if (len == NFS_MAXPATHLEN) {
1311 struct nfsnode *np = VTONFS(vp);
1312 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1315 ERROROUT(nfsm_mtouio(&info, uiop, len));
1324 * nfs synchronous read rpc using UIO
1327 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1330 struct nfsmount *nmp;
1331 int error = 0, len, retlen, tsiz, eof, attrflag;
1332 struct nfsm_info info;
1336 info.v3 = NFS_ISV3(vp);
1341 nmp = VFSTONFS(vp->v_mount);
1343 tsiz = uiop->uio_resid;
1344 tmp_off = uiop->uio_offset + tsiz;
1345 if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1347 tmp_off = uiop->uio_offset;
1349 nfsstats.rpccnt[NFSPROC_READ]++;
1350 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1351 nfsm_reqhead(&info, vp, NFSPROC_READ,
1352 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1353 ERROROUT(nfsm_fhtom(&info, vp));
1354 tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1356 txdr_hyper(uiop->uio_offset, tl);
1357 *(tl + 2) = txdr_unsigned(len);
1359 *tl++ = txdr_unsigned(uiop->uio_offset);
1360 *tl++ = txdr_unsigned(len);
1363 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1364 nfs_vpcred(vp, ND_READ), &error));
1366 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1367 NFS_LATTR_NOSHRINK));
1368 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1369 eof = fxdr_unsigned(int, *(tl + 1));
1371 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1373 NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1374 ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1379 * Handle short-read from server (NFSv3). If EOF is not
1380 * flagged (and no error occurred), but retlen is less
1381 * then the request size, we must zero-fill the remainder.
1383 if (retlen < len && info.v3 && eof == 0) {
1384 ERROROUT(uiomovez(len - retlen, uiop));
1390 * Terminate loop on EOF or zero-length read.
1392 * For NFSv2 a short-read indicates EOF, not zero-fill,
1393 * and also terminates the loop.
1396 if (eof || retlen == 0)
1398 } else if (retlen < len) {
1410 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1411 int *iomode, int *must_commit)
1415 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1416 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1417 int committed = NFSV3WRITE_FILESYNC;
1418 struct nfsm_info info;
1421 info.v3 = NFS_ISV3(vp);
1424 if (uiop->uio_iovcnt != 1)
1425 panic("nfs: writerpc iovcnt > 1");
1428 tsiz = uiop->uio_resid;
1429 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1432 nfsstats.rpccnt[NFSPROC_WRITE]++;
1433 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1434 nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1435 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED +
1437 ERROROUT(nfsm_fhtom(&info, vp));
1439 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1440 txdr_hyper(uiop->uio_offset, tl);
1442 *tl++ = txdr_unsigned(len);
1443 *tl++ = txdr_unsigned(*iomode);
1444 *tl = txdr_unsigned(len);
1448 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1449 /* Set both "begin" and "current" to non-garbage. */
1450 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1451 *tl++ = x; /* "begin offset" */
1452 *tl++ = x; /* "current offset" */
1453 x = txdr_unsigned(len);
1454 *tl++ = x; /* total to this offset */
1455 *tl = x; /* size of this write */
1457 ERROROUT(nfsm_uiotom(&info, uiop, len));
1458 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1459 nfs_vpcred(vp, ND_WRITE), &error));
1462 * The write RPC returns a before and after mtime. The
1463 * nfsm_wcc_data() macro checks the before n_mtime
1464 * against the before time and stores the after time
1465 * in the nfsnode's cached vattr and n_mtime field.
1466 * The NRMODIFIED bit will be set if the before
1467 * time did not match the original mtime.
1469 wccflag = NFSV3_WCCCHK;
1470 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1472 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1473 rlen = fxdr_unsigned(int, *tl++);
1479 } else if (rlen < len) {
1480 backup = len - rlen;
1481 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1482 uiop->uio_iov->iov_len += backup;
1483 uiop->uio_offset -= backup;
1484 uiop->uio_resid += backup;
1487 commit = fxdr_unsigned(int, *tl++);
1490 * Return the lowest committment level
1491 * obtained by any of the RPCs.
1493 if (committed == NFSV3WRITE_FILESYNC)
1495 else if (committed == NFSV3WRITE_DATASYNC &&
1496 commit == NFSV3WRITE_UNSTABLE)
1498 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1499 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1501 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1502 } else if (bcmp((caddr_t)tl,
1503 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1505 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1510 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1519 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1520 committed = NFSV3WRITE_FILESYNC;
1521 *iomode = committed;
1523 uiop->uio_resid = tsiz;
1529 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1530 * mode set to specify the file type and the size field for rdev.
1533 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1536 struct nfsv2_sattr *sp;
1538 struct vnode *newvp = NULL;
1539 struct nfsnode *np = NULL;
1541 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1543 struct nfsm_info info;
1546 info.v3 = NFS_ISV3(dvp);
1548 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1549 rmajor = txdr_unsigned(vap->va_rmajor);
1550 rminor = txdr_unsigned(vap->va_rminor);
1551 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1552 rmajor = nfs_xdrneg1;
1553 rminor = nfs_xdrneg1;
1555 return (EOPNOTSUPP);
1557 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1560 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1561 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1562 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1563 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1564 ERROROUT(nfsm_fhtom(&info, dvp));
1565 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1568 tl = nfsm_build(&info, NFSX_UNSIGNED);
1569 *tl++ = vtonfsv3_type(vap->va_type);
1570 nfsm_v3attrbuild(&info, vap, FALSE);
1571 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1572 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1573 *tl++ = txdr_unsigned(vap->va_rmajor);
1574 *tl = txdr_unsigned(vap->va_rminor);
1577 sp = nfsm_build(&info, NFSX_V2SATTR);
1578 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1579 sp->sa_uid = nfs_xdrneg1;
1580 sp->sa_gid = nfs_xdrneg1;
1581 sp->sa_size = makeudev(rmajor, rminor);
1582 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1583 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1585 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1586 cnp->cn_cred, &error));
1588 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1594 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1595 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1601 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1612 VTONFS(dvp)->n_flag |= NLMODIFIED;
1614 VTONFS(dvp)->n_attrstamp = 0;
1620 * just call nfs_mknodrpc() to do the work.
1622 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1623 * struct componentname *a_cnp, struct vattr *a_vap)
1627 nfs_mknod(struct vop_old_mknod_args *ap)
1629 struct nfsmount *nmp = VFSTONFS(ap->a_dvp->v_mount);
1632 lwkt_gettoken(&nmp->nm_token);
1633 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1634 lwkt_reltoken(&nmp->nm_token);
1639 static u_long create_verf;
1641 * nfs file create call
1643 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1644 * struct componentname *a_cnp, struct vattr *a_vap)
1647 nfs_create(struct vop_old_create_args *ap)
1649 struct vnode *dvp = ap->a_dvp;
1650 struct vattr *vap = ap->a_vap;
1651 struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1652 struct componentname *cnp = ap->a_cnp;
1653 struct nfsv2_sattr *sp;
1655 struct nfsnode *np = NULL;
1656 struct vnode *newvp = NULL;
1657 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1659 struct nfsm_info info;
1662 info.v3 = NFS_ISV3(dvp);
1663 lwkt_gettoken(&nmp->nm_token);
1666 * Oops, not for me..
1668 if (vap->va_type == VSOCK) {
1669 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1670 lwkt_reltoken(&nmp->nm_token);
1674 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1675 lwkt_reltoken(&nmp->nm_token);
1678 if (vap->va_vaflags & VA_EXCLUSIVE)
1681 nfsstats.rpccnt[NFSPROC_CREATE]++;
1682 nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1683 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1684 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1685 ERROROUT(nfsm_fhtom(&info, dvp));
1686 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1689 tl = nfsm_build(&info, NFSX_UNSIGNED);
1690 if (fmode & O_EXCL) {
1691 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1692 tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1694 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1695 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1698 *tl++ = create_verf;
1699 *tl = ++create_verf;
1701 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1702 nfsm_v3attrbuild(&info, vap, FALSE);
1705 sp = nfsm_build(&info, NFSX_V2SATTR);
1706 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1707 sp->sa_uid = nfs_xdrneg1;
1708 sp->sa_gid = nfs_xdrneg1;
1710 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1711 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1713 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1714 cnp->cn_cred, &error));
1716 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1722 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1723 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1730 error = nfsm_wcc_data(&info, dvp, &wccflag);
1732 (void)nfsm_wcc_data(&info, dvp, &wccflag);
1738 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1739 KKASSERT(newvp == NULL);
1743 } else if (info.v3 && (fmode & O_EXCL)) {
1745 * We are normally called with only a partially initialized
1746 * VAP. Since the NFSv3 spec says that server may use the
1747 * file attributes to store the verifier, the spec requires
1748 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1749 * in atime, but we can't really assume that all servers will
1750 * so we ensure that our SETATTR sets both atime and mtime.
1752 if (vap->va_mtime.tv_sec == VNOVAL)
1753 vfs_timestamp(&vap->va_mtime);
1754 if (vap->va_atime.tv_sec == VNOVAL)
1755 vap->va_atime = vap->va_mtime;
1756 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1760 * The new np may have enough info for access
1761 * checks, make sure rucred and wucred are
1762 * initialized for read and write rpc's.
1765 if (np->n_rucred == NULL)
1766 np->n_rucred = crhold(cnp->cn_cred);
1767 if (np->n_wucred == NULL)
1768 np->n_wucred = crhold(cnp->cn_cred);
1773 VTONFS(dvp)->n_flag |= NLMODIFIED;
1775 VTONFS(dvp)->n_attrstamp = 0;
1776 lwkt_reltoken(&nmp->nm_token);
1781 * nfs file remove call
1782 * To try and make nfs semantics closer to ufs semantics, a file that has
1783 * other processes using the vnode is renamed instead of removed and then
1784 * removed later on the last close.
1785 * - If v_sysref.refcnt > 1
1786 * If a rename is not already in the works
1787 * call nfs_sillyrename() to set it up
1791 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1792 * struct componentname *a_cnp)
1795 nfs_remove(struct vop_old_remove_args *ap)
1797 struct vnode *vp = ap->a_vp;
1798 struct vnode *dvp = ap->a_dvp;
1799 struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1800 struct componentname *cnp = ap->a_cnp;
1801 struct nfsnode *np = VTONFS(vp);
1805 lwkt_gettoken(&nmp->nm_token);
1807 if (vp->v_sysref.refcnt < 1)
1808 panic("nfs_remove: bad v_sysref.refcnt");
1810 if (vp->v_type == VDIR) {
1812 } else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1813 VOP_GETATTR(vp, &vattr) == 0 && vattr.va_nlink > 1)) {
1815 * throw away biocache buffers, mainly to avoid
1816 * unnecessary delayed writes later.
1818 error = nfs_vinvalbuf(vp, 0, 1);
1821 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1822 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1824 * Kludge City: If the first reply to the remove rpc is lost..
1825 * the reply to the retransmitted request will be ENOENT
1826 * since the file was in fact removed
1827 * Therefore, we cheat and return success.
1829 if (error == ENOENT)
1831 } else if (!np->n_sillyrename) {
1832 error = nfs_sillyrename(dvp, vp, cnp);
1834 np->n_attrstamp = 0;
1835 lwkt_reltoken(&nmp->nm_token);
1841 * nfs file remove rpc called from nfs_inactive
1844 nfs_removeit(struct sillyrename *sp)
1846 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1851 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1854 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1855 struct ucred *cred, struct thread *td)
1857 int error = 0, wccflag = NFSV3_WCCRATTR;
1858 struct nfsm_info info;
1861 info.v3 = NFS_ISV3(dvp);
1863 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1864 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1865 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1866 ERROROUT(nfsm_fhtom(&info, dvp));
1867 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1868 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1870 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1875 VTONFS(dvp)->n_flag |= NLMODIFIED;
1877 VTONFS(dvp)->n_attrstamp = 0;
1882 * nfs file rename call
1884 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1885 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1886 * struct vnode *a_tvp, struct componentname *a_tcnp)
1889 nfs_rename(struct vop_old_rename_args *ap)
1891 struct vnode *fvp = ap->a_fvp;
1892 struct vnode *tvp = ap->a_tvp;
1893 struct vnode *fdvp = ap->a_fdvp;
1894 struct vnode *tdvp = ap->a_tdvp;
1895 struct componentname *tcnp = ap->a_tcnp;
1896 struct componentname *fcnp = ap->a_fcnp;
1897 struct nfsmount *nmp = VFSTONFS(fdvp->v_mount);
1900 lwkt_gettoken(&nmp->nm_token);
1902 /* Check for cross-device rename */
1903 if ((fvp->v_mount != tdvp->v_mount) ||
1904 (tvp && (fvp->v_mount != tvp->v_mount))) {
1910 * We shouldn't have to flush fvp on rename for most server-side
1911 * filesystems as the file handle should not change. Unfortunately
1912 * the inode for some filesystems (msdosfs) might be tied to the
1913 * file name or directory position so to be completely safe
1914 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1917 * We must flush tvp on rename because it might become stale on the
1918 * server after the rename.
1920 if (nfs_flush_on_rename)
1921 VOP_FSYNC(fvp, MNT_WAIT, 0);
1923 VOP_FSYNC(tvp, MNT_WAIT, 0);
1926 * If the tvp exists and is in use, sillyrename it before doing the
1927 * rename of the new file over it.
1929 * XXX Can't sillyrename a directory.
1931 * We do not attempt to do any namecache purges in this old API
1932 * routine. The new API compat functions have access to the actual
1933 * namecache structures and will do it for us.
1935 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1936 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1943 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1944 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1948 lwkt_reltoken(&nmp->nm_token);
1958 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1960 if (error == ENOENT)
1966 * nfs file rename rpc called from nfs_remove() above
1969 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1970 struct sillyrename *sp)
1972 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1973 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1977 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1980 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1981 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1982 struct ucred *cred, struct thread *td)
1984 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1985 struct nfsm_info info;
1988 info.v3 = NFS_ISV3(fdvp);
1990 nfsstats.rpccnt[NFSPROC_RENAME]++;
1991 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1992 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1993 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1994 ERROROUT(nfsm_fhtom(&info, fdvp));
1995 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1996 ERROROUT(nfsm_fhtom(&info, tdvp));
1997 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1998 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
2000 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
2001 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
2006 VTONFS(fdvp)->n_flag |= NLMODIFIED;
2007 VTONFS(tdvp)->n_flag |= NLMODIFIED;
2009 VTONFS(fdvp)->n_attrstamp = 0;
2011 VTONFS(tdvp)->n_attrstamp = 0;
2016 * nfs hard link create call
2018 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
2019 * struct componentname *a_cnp)
2022 nfs_link(struct vop_old_link_args *ap)
2024 struct vnode *vp = ap->a_vp;
2025 struct vnode *tdvp = ap->a_tdvp;
2026 struct nfsmount *nmp = VFSTONFS(tdvp->v_mount);
2027 struct componentname *cnp = ap->a_cnp;
2028 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2029 struct nfsm_info info;
2031 if (vp->v_mount != tdvp->v_mount) {
2034 lwkt_gettoken(&nmp->nm_token);
2037 * The attribute cache may get out of sync with the server on link.
2038 * Pushing writes to the server before handle was inherited from
2039 * long long ago and it is unclear if we still need to do this.
2042 if (nfs_flush_on_hlink)
2043 VOP_FSYNC(vp, MNT_WAIT, 0);
2046 info.v3 = NFS_ISV3(vp);
2048 nfsstats.rpccnt[NFSPROC_LINK]++;
2049 nfsm_reqhead(&info, vp, NFSPROC_LINK,
2050 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
2051 nfsm_rndup(cnp->cn_namelen));
2052 ERROROUT(nfsm_fhtom(&info, vp));
2053 ERROROUT(nfsm_fhtom(&info, tdvp));
2054 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2056 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
2057 cnp->cn_cred, &error));
2059 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2060 NFS_LATTR_NOSHRINK));
2061 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
2066 VTONFS(tdvp)->n_flag |= NLMODIFIED;
2068 VTONFS(vp)->n_attrstamp = 0;
2070 VTONFS(tdvp)->n_attrstamp = 0;
2072 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2074 if (error == EEXIST)
2076 lwkt_reltoken(&nmp->nm_token);
2081 * nfs symbolic link create call
2083 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
2084 * struct componentname *a_cnp, struct vattr *a_vap,
2088 nfs_symlink(struct vop_old_symlink_args *ap)
2090 struct vnode *dvp = ap->a_dvp;
2091 struct vattr *vap = ap->a_vap;
2092 struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2093 struct componentname *cnp = ap->a_cnp;
2094 struct nfsv2_sattr *sp;
2095 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2096 struct vnode *newvp = NULL;
2097 struct nfsm_info info;
2100 info.v3 = NFS_ISV3(dvp);
2101 lwkt_gettoken(&nmp->nm_token);
2103 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2104 slen = strlen(ap->a_target);
2105 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
2106 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
2107 nfsm_rndup(cnp->cn_namelen) +
2108 nfsm_rndup(slen) + NFSX_SATTR(info.v3));
2109 ERROROUT(nfsm_fhtom(&info, dvp));
2110 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2113 nfsm_v3attrbuild(&info, vap, FALSE);
2115 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
2117 sp = nfsm_build(&info, NFSX_V2SATTR);
2118 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2119 sp->sa_uid = nfs_xdrneg1;
2120 sp->sa_gid = nfs_xdrneg1;
2121 sp->sa_size = nfs_xdrneg1;
2122 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2123 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2127 * Issue the NFS request and get the rpc response.
2129 * Only NFSv3 responses returning an error of 0 actually return
2130 * a file handle that can be converted into newvp without having
2131 * to do an extra lookup rpc.
2133 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
2134 cnp->cn_cred, &error));
2137 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2139 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2143 * out code jumps -> here, mrep is also freed.
2151 * If we get an EEXIST error, silently convert it to no-error
2152 * in case of an NFS retry.
2154 if (error == EEXIST)
2158 * If we do not have (or no longer have) an error, and we could
2159 * not extract the newvp from the response due to the request being
2160 * NFSv2 or the error being EEXIST. We have to do a lookup in order
2161 * to obtain a newvp to return.
2163 if (error == 0 && newvp == NULL) {
2164 struct nfsnode *np = NULL;
2166 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2167 cnp->cn_cred, cnp->cn_td, &np);
2177 VTONFS(dvp)->n_flag |= NLMODIFIED;
2179 VTONFS(dvp)->n_attrstamp = 0;
2180 lwkt_reltoken(&nmp->nm_token);
2188 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2189 * struct componentname *a_cnp, struct vattr *a_vap)
2192 nfs_mkdir(struct vop_old_mkdir_args *ap)
2194 struct vnode *dvp = ap->a_dvp;
2195 struct vattr *vap = ap->a_vap;
2196 struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2197 struct componentname *cnp = ap->a_cnp;
2198 struct nfsv2_sattr *sp;
2199 struct nfsnode *np = NULL;
2200 struct vnode *newvp = NULL;
2202 int error = 0, wccflag = NFSV3_WCCRATTR;
2205 struct nfsm_info info;
2208 info.v3 = NFS_ISV3(dvp);
2209 lwkt_gettoken(&nmp->nm_token);
2211 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2212 lwkt_reltoken(&nmp->nm_token);
2215 len = cnp->cn_namelen;
2216 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2217 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2218 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2219 nfsm_rndup(len) + NFSX_SATTR(info.v3));
2220 ERROROUT(nfsm_fhtom(&info, dvp));
2221 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2223 nfsm_v3attrbuild(&info, vap, FALSE);
2225 sp = nfsm_build(&info, NFSX_V2SATTR);
2226 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2227 sp->sa_uid = nfs_xdrneg1;
2228 sp->sa_gid = nfs_xdrneg1;
2229 sp->sa_size = nfs_xdrneg1;
2230 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2231 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2233 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2234 cnp->cn_cred, &error));
2236 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2239 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2244 VTONFS(dvp)->n_flag |= NLMODIFIED;
2246 VTONFS(dvp)->n_attrstamp = 0;
2248 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2249 * if we can succeed in looking up the directory.
2251 if (error == EEXIST || (!error && !gotvp)) {
2256 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2260 if (newvp->v_type != VDIR)
2270 lwkt_reltoken(&nmp->nm_token);
2275 * nfs remove directory call
2277 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2278 * struct componentname *a_cnp)
2281 nfs_rmdir(struct vop_old_rmdir_args *ap)
2283 struct vnode *vp = ap->a_vp;
2284 struct vnode *dvp = ap->a_dvp;
2285 struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2286 struct componentname *cnp = ap->a_cnp;
2287 int error = 0, wccflag = NFSV3_WCCRATTR;
2288 struct nfsm_info info;
2291 info.v3 = NFS_ISV3(dvp);
2296 lwkt_gettoken(&nmp->nm_token);
2298 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2299 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2300 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2301 nfsm_rndup(cnp->cn_namelen));
2302 ERROROUT(nfsm_fhtom(&info, dvp));
2303 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2305 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2306 cnp->cn_cred, &error));
2308 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2313 VTONFS(dvp)->n_flag |= NLMODIFIED;
2315 VTONFS(dvp)->n_attrstamp = 0;
2317 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2319 if (error == ENOENT)
2321 lwkt_reltoken(&nmp->nm_token);
2329 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2332 nfs_readdir(struct vop_readdir_args *ap)
2334 struct vnode *vp = ap->a_vp;
2335 struct nfsnode *np = VTONFS(vp);
2336 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2337 struct uio *uio = ap->a_uio;
2341 if (vp->v_type != VDIR)
2344 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2347 lwkt_gettoken(&nmp->nm_token);
2350 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2351 * and then check that is still valid, or if this is an NQNFS mount
2352 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2353 * VOP_GETATTR() does not necessarily go to the wire.
2355 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2356 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2357 if (VOP_GETATTR(vp, &vattr) == 0 &&
2358 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2360 nfsstats.direofcache_hits++;
2366 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2367 * own cache coherency checks so we do not have to.
2369 tresid = uio->uio_resid;
2370 error = nfs_bioread(vp, uio, 0);
2372 if (!error && uio->uio_resid == tresid)
2373 nfsstats.direofcache_misses++;
2375 lwkt_reltoken(&nmp->nm_token);
2382 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2384 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2385 * offset/block and converts the nfs formatted directory entries for userland
2386 * consumption as well as deals with offsets into the middle of blocks.
2387 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2388 * be block-bounded. It must convert to cookies for the actual RPC.
2391 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2394 struct nfs_dirent *dp = NULL;
2399 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2400 struct nfsnode *dnp = VTONFS(vp);
2402 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2404 struct nfsm_info info;
2407 info.v3 = NFS_ISV3(vp);
2410 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2411 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2412 panic("nfs readdirrpc bad uio");
2416 * If there is no cookie, assume directory was stale.
2418 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2422 return (NFSERR_BAD_COOKIE);
2424 * Loop around doing readdir rpc's of size nm_readdirsize
2425 * truncated to a multiple of DIRBLKSIZ.
2426 * The stopping criteria is EOF or buffer full.
2428 while (more_dirs && bigenough) {
2429 nfsstats.rpccnt[NFSPROC_READDIR]++;
2430 nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2431 NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2432 ERROROUT(nfsm_fhtom(&info, vp));
2434 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2435 *tl++ = cookie.nfsuquad[0];
2436 *tl++ = cookie.nfsuquad[1];
2437 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2438 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2441 * WARNING! HAMMER DIRECTORIES WILL NOT WORK WELL
2442 * WITH NFSv2!!! There's nothing I can really do
2443 * about it other than to hope the server supports
2446 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2447 *tl++ = cookie.nfsuquad[0];
2449 *tl = txdr_unsigned(nmp->nm_readdirsize);
2450 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2452 nfs_vpcred(vp, ND_READ), &error));
2454 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2455 NFS_LATTR_NOSHRINK));
2456 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2457 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2458 dnp->n_cookieverf.nfsuquad[1] = *tl;
2460 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2461 more_dirs = fxdr_unsigned(int, *tl);
2463 /* loop thru the dir entries, converting them to std form */
2464 while (more_dirs && bigenough) {
2466 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2467 fileno = fxdr_hyper(tl);
2468 len = fxdr_unsigned(int, *(tl + 2));
2470 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2471 fileno = fxdr_unsigned(u_quad_t, *tl++);
2472 len = fxdr_unsigned(int, *tl);
2474 if (len <= 0 || len > NFS_MAXNAMLEN) {
2482 * len is the number of bytes in the path element
2483 * name, not including the \0 termination.
2485 * tlen is the number of bytes w have to reserve for
2486 * the path element name.
2488 tlen = nfsm_rndup(len);
2490 tlen += 4; /* To ensure null termination */
2493 * If the entry would cross a DIRBLKSIZ boundary,
2494 * extend the previous nfs_dirent to cover the
2497 left = DIRBLKSIZ - blksiz;
2498 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2499 dp->nfs_reclen += left;
2500 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2501 uiop->uio_iov->iov_len -= left;
2502 uiop->uio_offset += left;
2503 uiop->uio_resid -= left;
2506 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2509 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2510 dp->nfs_ino = fileno;
2511 dp->nfs_namlen = len;
2512 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2513 dp->nfs_type = DT_UNKNOWN;
2514 blksiz += dp->nfs_reclen;
2515 if (blksiz == DIRBLKSIZ)
2517 uiop->uio_offset += sizeof(struct nfs_dirent);
2518 uiop->uio_resid -= sizeof(struct nfs_dirent);
2519 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2520 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2521 ERROROUT(nfsm_mtouio(&info, uiop, len));
2524 * The uiop has advanced by nfs_dirent + len
2525 * but really needs to advance by
2528 cp = uiop->uio_iov->iov_base;
2530 *cp = '\0'; /* null terminate */
2531 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2532 uiop->uio_iov->iov_len -= tlen;
2533 uiop->uio_offset += tlen;
2534 uiop->uio_resid -= tlen;
2537 * NFS strings must be rounded up (nfsm_myouio
2538 * handled that in the bigenough case).
2540 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2543 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2545 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2549 * If we were able to accomodate the last entry,
2550 * get the cookie for the next one. Otherwise
2551 * hold-over the cookie for the one we were not
2552 * able to accomodate.
2555 cookie.nfsuquad[0] = *tl++;
2557 cookie.nfsuquad[1] = *tl++;
2558 } else if (info.v3) {
2563 more_dirs = fxdr_unsigned(int, *tl);
2566 * If at end of rpc data, get the eof boolean
2569 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2570 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2576 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2577 * by increasing d_reclen for the last record.
2580 left = DIRBLKSIZ - blksiz;
2581 dp->nfs_reclen += left;
2582 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2583 uiop->uio_iov->iov_len -= left;
2584 uiop->uio_offset += left;
2585 uiop->uio_resid -= left;
2590 * We hit the end of the directory, update direofoffset.
2592 dnp->n_direofoffset = uiop->uio_offset;
2595 * There is more to go, insert the link cookie so the
2596 * next block can be read.
2598 if (uiop->uio_resid > 0)
2599 kprintf("EEK! readdirrpc resid > 0\n");
2600 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2608 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2611 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2614 struct nfs_dirent *dp;
2616 struct vnode *newvp;
2618 caddr_t dpossav1, dpossav2;
2620 struct mbuf *mdsav1, *mdsav2;
2622 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2623 struct nfsnode *dnp = VTONFS(vp), *np;
2626 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2627 int attrflag, fhsize;
2628 struct nchandle nch;
2629 struct nchandle dnch;
2630 struct nlcomponent nlc;
2631 struct nfsm_info info;
2640 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2641 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2642 panic("nfs readdirplusrpc bad uio");
2645 * Obtain the namecache record for the directory so we have something
2646 * to use as a basis for creating the entries. This function will
2647 * return a held (but not locked) ncp. The ncp may be disconnected
2648 * from the tree and cannot be used for upward traversals, and the
2649 * ncp may be unnamed. Note that other unrelated operations may
2650 * cause the ncp to be named at any time.
2652 * We have to lock the ncp to prevent a lock order reversal when
2653 * rdirplus does nlookups of the children, because the vnode is
2654 * locked and has to stay that way.
2656 cache_fromdvp(vp, NULL, 0, &dnch);
2657 bzero(&nlc, sizeof(nlc));
2661 * If there is no cookie, assume directory was stale.
2663 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2669 return (NFSERR_BAD_COOKIE);
2673 * Loop around doing readdir rpc's of size nm_readdirsize
2674 * truncated to a multiple of DIRBLKSIZ.
2675 * The stopping criteria is EOF or buffer full.
2677 while (more_dirs && bigenough) {
2678 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2679 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2680 NFSX_FH(info.v3) + 6 * NFSX_UNSIGNED);
2681 ERROROUT(nfsm_fhtom(&info, vp));
2682 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2683 *tl++ = cookie.nfsuquad[0];
2684 *tl++ = cookie.nfsuquad[1];
2685 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2686 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2687 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2688 *tl = txdr_unsigned(nmp->nm_rsize);
2689 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2691 nfs_vpcred(vp, ND_READ), &error));
2692 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2693 NFS_LATTR_NOSHRINK));
2694 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2695 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2696 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2697 more_dirs = fxdr_unsigned(int, *tl);
2699 /* loop thru the dir entries, doctoring them to 4bsd form */
2700 while (more_dirs && bigenough) {
2701 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2702 fileno = fxdr_hyper(tl);
2703 len = fxdr_unsigned(int, *(tl + 2));
2704 if (len <= 0 || len > NFS_MAXNAMLEN) {
2710 tlen = nfsm_rndup(len);
2712 tlen += 4; /* To ensure null termination*/
2713 left = DIRBLKSIZ - blksiz;
2714 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2715 dp->nfs_reclen += left;
2716 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2717 uiop->uio_iov->iov_len -= left;
2718 uiop->uio_offset += left;
2719 uiop->uio_resid -= left;
2722 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2725 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2726 dp->nfs_ino = fileno;
2727 dp->nfs_namlen = len;
2728 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2729 dp->nfs_type = DT_UNKNOWN;
2730 blksiz += dp->nfs_reclen;
2731 if (blksiz == DIRBLKSIZ)
2733 uiop->uio_offset += sizeof(struct nfs_dirent);
2734 uiop->uio_resid -= sizeof(struct nfs_dirent);
2735 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2736 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2737 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2738 nlc.nlc_namelen = len;
2739 ERROROUT(nfsm_mtouio(&info, uiop, len));
2740 cp = uiop->uio_iov->iov_base;
2743 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2744 uiop->uio_iov->iov_len -= tlen;
2745 uiop->uio_offset += tlen;
2746 uiop->uio_resid -= tlen;
2748 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2750 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2752 cookie.nfsuquad[0] = *tl++;
2753 cookie.nfsuquad[1] = *tl++;
2759 * Since the attributes are before the file handle
2760 * (sigh), we must skip over the attributes and then
2761 * come back and get them.
2763 attrflag = fxdr_unsigned(int, *tl);
2765 dpossav1 = info.dpos;
2767 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2768 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2769 doit = fxdr_unsigned(int, *tl);
2771 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2773 if (doit && bigenough && !nlcdegenerate(&nlc) &&
2774 !NFS_CMPFH(dnp, fhp, fhsize)
2778 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2779 nlc.nlc_namelen, nlc.nlc_namelen,
2783 * This is a bit hokey but there isn't
2784 * much we can do about it. We can't
2785 * hold the directory vp locked while
2786 * doing lookups and gets.
2788 nch = cache_nlookup_nonblock(&dnch, &nlc);
2789 if (nch.ncp == NULL)
2791 cache_setunresolved(&nch);
2792 error = nfs_nget_nonblock(vp->v_mount, fhp,
2799 dpossav2 = info.dpos;
2800 info.dpos = dpossav1;
2803 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2804 info.dpos = dpossav2;
2807 IFTODT(VTTOIF(np->n_vattr.va_type));
2808 nfs_cache_setvp(&nch, newvp,
2809 nfspos_cache_timeout);
2817 kprintf("Warning: NFS/rddirplus, "
2818 "UNABLE TO ENTER %*.*s\n",
2819 nlc.nlc_namelen, nlc.nlc_namelen,
2825 /* Just skip over the file handle */
2826 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2827 i = fxdr_unsigned(int, *tl);
2828 ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2830 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2831 more_dirs = fxdr_unsigned(int, *tl);
2834 * If at end of rpc data, get the eof boolean
2837 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2838 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2844 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2845 * by increasing d_reclen for the last record.
2848 left = DIRBLKSIZ - blksiz;
2849 dp->nfs_reclen += left;
2850 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2851 uiop->uio_iov->iov_len -= left;
2852 uiop->uio_offset += left;
2853 uiop->uio_resid -= left;
2857 * We are now either at the end of the directory or have filled the
2861 dnp->n_direofoffset = uiop->uio_offset;
2863 if (uiop->uio_resid > 0)
2864 kprintf("EEK! readdirplusrpc resid > 0\n");
2865 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2869 if (newvp != NULLVP) {
2882 * Silly rename. To make the NFS filesystem that is stateless look a little
2883 * more like the "ufs" a remove of an active vnode is translated to a rename
2884 * to a funny looking filename that is removed by nfs_inactive on the
2885 * nfsnode. There is the potential for another process on a different client
2886 * to create the same funny name between the nfs_lookitup() fails and the
2887 * nfs_rename() completes, but...
2890 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2892 struct sillyrename *sp;
2897 * We previously purged dvp instead of vp. I don't know why, it
2898 * completely destroys performance. We can't do it anyway with the
2899 * new VFS API since we would be breaking the namecache topology.
2901 cache_purge(vp); /* XXX */
2904 if (vp->v_type == VDIR)
2905 panic("nfs: sillyrename dir");
2907 sp = kmalloc(sizeof(struct sillyrename), M_NFSREQ, M_WAITOK);
2908 sp->s_cred = crdup(cnp->cn_cred);
2912 /* Fudge together a funny name */
2913 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2914 (int)(intptr_t)cnp->cn_td);
2916 /* Try lookitups until we get one that isn't there */
2917 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2918 cnp->cn_td, NULL) == 0) {
2920 if (sp->s_name[4] > 'z') {
2925 error = nfs_renameit(dvp, cnp, sp);
2928 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2930 np->n_sillyrename = sp;
2935 kfree((caddr_t)sp, M_NFSREQ);
2940 * Look up a file name and optionally either update the file handle or
2941 * allocate an nfsnode, depending on the value of npp.
2942 * npp == NULL --> just do the lookup
2943 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2945 * *npp != NULL --> update the file handle in the vnode
2948 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2949 struct thread *td, struct nfsnode **npp)
2951 struct vnode *newvp = NULL;
2952 struct nfsnode *np, *dnp = VTONFS(dvp);
2953 int error = 0, fhlen, attrflag;
2955 struct nfsm_info info;
2958 info.v3 = NFS_ISV3(dvp);
2960 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2961 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2962 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2963 ERROROUT(nfsm_fhtom(&info, dvp));
2964 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2965 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2966 if (npp && !error) {
2967 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2970 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2971 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2972 np->n_fhp = &np->n_fh;
2973 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2974 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2975 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2976 np->n_fhsize = fhlen;
2978 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2982 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2991 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2992 NFS_LATTR_NOSHRINK));
2993 if (!attrflag && *npp == NULL) {
3003 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
3009 if (npp && *npp == NULL) {
3024 * Nfs Version 3 commit rpc
3026 * We call it 'uio' to distinguish it from 'bio' but there is no real uio
3030 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
3032 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3033 int error = 0, wccflag = NFSV3_WCCRATTR;
3034 struct nfsm_info info;
3040 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
3042 nfsstats.rpccnt[NFSPROC_COMMIT]++;
3043 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
3044 ERROROUT(nfsm_fhtom(&info, vp));
3045 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
3046 txdr_hyper(offset, tl);
3048 *tl = txdr_unsigned(cnt);
3049 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
3050 nfs_vpcred(vp, ND_WRITE), &error));
3051 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
3053 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
3054 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
3055 NFSX_V3WRITEVERF)) {
3056 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
3058 error = NFSERR_STALEWRITEVERF;
3069 * - make nfs_bmap() essentially a no-op that does no translation
3070 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3071 * (Maybe I could use the process's page mapping, but I was concerned that
3072 * Kernel Write might not be enabled and also figured copyout() would do
3073 * a lot more work than bcopy() and also it currently happens in the
3074 * context of the swapper process (2).
3076 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
3077 * off_t *a_doffsetp, int *a_runp, int *a_runb)
3080 nfs_bmap(struct vop_bmap_args *ap)
3082 /* no token lock required */
3083 if (ap->a_doffsetp != NULL)
3084 *ap->a_doffsetp = ap->a_loffset;
3085 if (ap->a_runp != NULL)
3087 if (ap->a_runb != NULL)
3096 nfs_strategy(struct vop_strategy_args *ap)
3098 struct bio *bio = ap->a_bio;
3100 struct buf *bp __debugvar = bio->bio_buf;
3101 struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3105 KASSERT(bp->b_cmd != BUF_CMD_DONE,
3106 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
3107 KASSERT(BUF_REFCNT(bp) > 0,
3108 ("nfs_strategy: buffer %p not locked", bp));
3110 if (bio->bio_flags & BIO_SYNC)
3111 td = curthread; /* XXX */
3115 lwkt_gettoken(&nmp->nm_token);
3118 * We probably don't need to push an nbio any more since no
3119 * block conversion is required due to the use of 64 bit byte
3120 * offsets, but do it anyway.
3122 * NOTE: When NFS callers itself via this strategy routines and
3123 * sets up a synchronous I/O, it expects the I/O to run
3124 * synchronously (its bio_done routine just assumes it),
3125 * so for now we have to honor the bit.
3127 nbio = push_bio(bio);
3128 nbio->bio_offset = bio->bio_offset;
3129 nbio->bio_flags = bio->bio_flags & BIO_SYNC;
3132 * If the op is asynchronous and an i/o daemon is waiting
3133 * queue the request, wake it up and wait for completion
3134 * otherwise just do it ourselves.
3136 if (bio->bio_flags & BIO_SYNC) {
3137 error = nfs_doio(ap->a_vp, nbio, td);
3139 nfs_asyncio(ap->a_vp, nbio);
3142 lwkt_reltoken(&nmp->nm_token);
3150 * NB Currently unsupported.
3152 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
3156 nfs_mmap(struct vop_mmap_args *ap)
3158 /* no token lock required */
3163 * fsync vnode op. Just call nfs_flush() with commit == 1.
3165 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3169 nfs_fsync(struct vop_fsync_args *ap)
3171 struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3174 lwkt_gettoken(&nmp->nm_token);
3175 error = nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1);
3176 lwkt_reltoken(&nmp->nm_token);
3182 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
3183 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
3184 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
3185 * set the buffer contains data that has already been written to the server
3186 * and which now needs a commit RPC.
3188 * If commit is 0 we only take one pass and only flush buffers containing new
3191 * If commit is 1 we take two passes, issuing a commit RPC in the second
3194 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3195 * to completely flush all pending data.
3197 * Note that the RB_SCAN code properly handles the case where the
3198 * callback might block and directly or indirectly (another thread) cause
3199 * the RB tree to change.
3202 #ifndef NFS_COMMITBVECSIZ
3203 #define NFS_COMMITBVECSIZ 16
3206 struct nfs_flush_info {
3207 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3214 struct buf *bvary[NFS_COMMITBVECSIZ];
3220 static int nfs_flush_bp(struct buf *bp, void *data);
3221 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3224 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3226 struct nfsnode *np = VTONFS(vp);
3227 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3228 struct nfs_flush_info info;
3231 bzero(&info, sizeof(info));
3234 info.waitfor = waitfor;
3235 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3237 lwkt_gettoken(&vp->v_token);
3243 info.mode = NFI_FLUSHNEW;
3244 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3245 nfs_flush_bp, &info);
3248 * Take a second pass if committing and no error occured.
3249 * Clean up any left over collection (whether an error
3252 if (commit && error == 0) {
3253 info.mode = NFI_COMMIT;
3254 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3255 nfs_flush_bp, &info);
3257 error = nfs_flush_docommit(&info, error);
3261 * Wait for pending I/O to complete before checking whether
3262 * any further dirty buffers exist.
3264 while (waitfor == MNT_WAIT &&
3265 bio_track_active(&vp->v_track_write)) {
3266 error = bio_track_wait(&vp->v_track_write,
3267 info.slpflag, info.slptimeo);
3270 * We have to be able to break out if this
3271 * is an 'intr' mount.
3273 if (nfs_sigintr(nmp, NULL, td)) {
3279 * Since we do not process pending signals,
3280 * once we get a PCATCH our tsleep() will no
3281 * longer sleep, switch to a fixed timeout
3284 if (info.slpflag == PCATCH) {
3286 info.slptimeo = 2 * hz;
3293 * Loop if we are flushing synchronous as well as committing,
3294 * and dirty buffers are still present. Otherwise we might livelock.
3296 } while (waitfor == MNT_WAIT && commit &&
3297 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3300 * The callbacks have to return a negative error to terminate the
3307 * Deal with any error collection
3309 if (np->n_flag & NWRITEERR) {
3310 error = np->n_error;
3311 np->n_flag &= ~NWRITEERR;
3313 lwkt_reltoken(&vp->v_token);
3319 nfs_flush_bp(struct buf *bp, void *data)
3321 struct nfs_flush_info *info = data;
3327 switch(info->mode) {
3329 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3330 if (error && info->loops && info->waitfor == MNT_WAIT) {
3331 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3333 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3334 if (info->slpflag & PCATCH)
3335 lkflags |= LK_PCATCH;
3336 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3342 * Ignore locking errors
3350 * The buffer may have changed out from under us, even if
3351 * we did not block (MPSAFE). Check again now that it is
3354 if (bp->b_vp == info->vp &&
3355 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3364 * Only process buffers in need of a commit which we can
3365 * immediately lock. This may prevent a buffer from being
3366 * committed, but the normal flush loop will block on the
3367 * same buffer so we shouldn't get into an endless loop.
3369 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3370 (B_DELWRI | B_NEEDCOMMIT)) {
3373 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3377 * We must recheck after successfully locking the buffer.
3379 if (bp->b_vp != info->vp ||
3380 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3381 (B_DELWRI | B_NEEDCOMMIT)) {
3387 * NOTE: storing the bp in the bvary[] basically sets
3388 * it up for a commit operation.
3390 * We must call vfs_busy_pages() now so the commit operation
3391 * is interlocked with user modifications to memory mapped
3392 * pages. The b_dirtyoff/b_dirtyend range is not correct
3393 * until after the pages have been busied.
3395 * Note: to avoid loopback deadlocks, we do not
3396 * assign b_runningbufspace.
3399 bp->b_cmd = BUF_CMD_WRITE;
3400 vfs_busy_pages(bp->b_vp, bp);
3401 info->bvary[info->bvsize] = bp;
3402 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3403 if (info->bvsize == 0 || toff < info->beg_off)
3404 info->beg_off = toff;
3405 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3406 if (info->bvsize == 0 || toff > info->end_off)
3407 info->end_off = toff;
3409 if (info->bvsize == NFS_COMMITBVECSIZ) {
3410 error = nfs_flush_docommit(info, 0);
3411 KKASSERT(info->bvsize == 0);
3419 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3429 if (info->bvsize > 0) {
3431 * Commit data on the server, as required. Note that
3432 * nfs_commit will use the vnode's cred for the commit.
3433 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3435 bytes = info->end_off - info->beg_off;
3436 if (bytes > 0x40000000)
3441 retv = nfs_commitrpc_uio(vp, info->beg_off,
3442 (int)bytes, info->td);
3443 if (retv == NFSERR_STALEWRITEVERF)
3444 nfs_clearcommit(vp->v_mount);
3448 * Now, either mark the blocks I/O done or mark the
3449 * blocks dirty, depending on whether the commit
3452 for (i = 0; i < info->bvsize; ++i) {
3453 bp = info->bvary[i];
3454 if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3456 * Either an error or the original
3457 * vfs_busy_pages() cleared B_NEEDCOMMIT
3458 * due to finding new dirty VM pages in
3461 * Leave B_DELWRI intact.
3463 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3464 vfs_unbusy_pages(bp);
3465 bp->b_cmd = BUF_CMD_DONE;
3469 * Success, remove B_DELWRI ( bundirty() ).
3471 * b_dirtyoff/b_dirtyend seem to be NFS
3472 * specific. We should probably move that
3473 * into bundirty(). XXX
3475 * We are faking an I/O write, we have to
3476 * start the transaction in order to
3477 * immediately biodone() it.
3480 bp->b_flags &= ~B_ERROR;
3481 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3482 bp->b_dirtyoff = bp->b_dirtyend = 0;
3483 biodone(&bp->b_bio1);
3492 * NFS advisory byte-level locks.
3493 * Currently unsupported.
3495 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3499 nfs_advlock(struct vop_advlock_args *ap)
3501 struct nfsnode *np = VTONFS(ap->a_vp);
3503 /* no token lock currently required */
3505 * The following kludge is to allow diskless support to work
3506 * until a real NFS lockd is implemented. Basically, just pretend
3507 * that this is a local lock.
3509 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3513 * Print out the contents of an nfsnode.
3515 * nfs_print(struct vnode *a_vp)
3518 nfs_print(struct vop_print_args *ap)
3520 struct vnode *vp = ap->a_vp;
3521 struct nfsnode *np = VTONFS(vp);
3523 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3524 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3525 if (vp->v_type == VFIFO)
3532 * nfs special file access vnode op.
3534 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3537 nfs_laccess(struct vop_access_args *ap)
3539 struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3543 lwkt_gettoken(&nmp->nm_token);
3544 error = VOP_GETATTR(ap->a_vp, &vattr);
3546 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3549 lwkt_reltoken(&nmp->nm_token);
3555 * Read wrapper for fifos.
3557 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3558 * struct ucred *a_cred)
3561 nfsfifo_read(struct vop_read_args *ap)
3563 struct nfsnode *np = VTONFS(ap->a_vp);
3565 /* no token access required */
3570 getnanotime(&np->n_atim);
3571 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3575 * Write wrapper for fifos.
3577 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3578 * struct ucred *a_cred)
3581 nfsfifo_write(struct vop_write_args *ap)
3583 struct nfsnode *np = VTONFS(ap->a_vp);
3585 /* no token access required */
3590 getnanotime(&np->n_mtim);
3591 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3595 * Close wrapper for fifos.
3597 * Update the times on the nfsnode then do fifo close.
3599 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3602 nfsfifo_close(struct vop_close_args *ap)
3604 struct vnode *vp = ap->a_vp;
3605 struct nfsnode *np = VTONFS(vp);
3609 /* no token access required */
3611 if (np->n_flag & (NACC | NUPD)) {
3613 if (np->n_flag & NACC)
3615 if (np->n_flag & NUPD)
3618 if (vp->v_sysref.refcnt == 1 &&
3619 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3621 if (np->n_flag & NACC)
3622 vattr.va_atime = np->n_atim;
3623 if (np->n_flag & NUPD)
3624 vattr.va_mtime = np->n_mtim;
3625 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3628 return (VOCALL(&fifo_vnode_vops, &ap->a_head));