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 = vop_stdgetpages,
144 .vop_putpages = vop_stdputpages,
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)
250 * Returns whether a name component is a degenerate '.' or '..'.
254 nlcdegenerate(struct nlcomponent *nlc)
256 if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.')
258 if (nlc->nlc_namelen == 2 &&
259 nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.')
265 nfs3_access_otw(struct vnode *vp, int wmode,
266 struct thread *td, struct ucred *cred)
268 struct nfsnode *np = VTONFS(vp);
273 struct nfsm_info info;
278 nfsstats.rpccnt[NFSPROC_ACCESS]++;
279 nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
280 NFSX_FH(info.v3) + NFSX_UNSIGNED);
281 ERROROUT(nfsm_fhtom(&info, vp));
282 tl = nfsm_build(&info, NFSX_UNSIGNED);
283 *tl = txdr_unsigned(wmode);
284 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
285 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
287 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
288 rmode = fxdr_unsigned(u_int32_t, *tl);
290 np->n_modeuid = cred->cr_uid;
291 np->n_modestamp = mycpu->gd_time_seconds;
300 * nfs access vnode op.
301 * For nfs version 2, just return ok. File accesses may fail later.
302 * For nfs version 3, use the access rpc to check accessibility. If file modes
303 * are changed on the server, accesses might still fail later.
305 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
308 nfs_access(struct vop_access_args *ap)
311 struct vnode *vp = ap->a_vp;
312 thread_t td = curthread;
314 u_int32_t mode, wmode;
315 struct nfsnode *np = VTONFS(vp);
316 int v3 = NFS_ISV3(vp);
319 * Disallow write attempts on filesystems mounted read-only;
320 * unless the file is a socket, fifo, or a block or character
321 * device resident on the filesystem.
323 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
324 switch (vp->v_type) {
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) {
410 * Attempt to prevent a mapped root from accessing a file
411 * which it shouldn't. We try to read a byte from the file
412 * if the user is root and the file is not zero length.
413 * After calling nfs_laccess, we should have the correct
416 if (cred->cr_uid == 0 && (ap->a_mode & VREAD)
417 && VTONFS(vp)->n_size > 0) {
424 auio.uio_iov = &aiov;
428 auio.uio_segflg = UIO_SYSSPACE;
429 auio.uio_rw = UIO_READ;
432 if (vp->v_type == VREG) {
433 error = nfs_readrpc_uio(vp, &auio);
434 } else if (vp->v_type == VDIR) {
436 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
438 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
439 error = nfs_readdirrpc_uio(vp, &auio);
441 } else if (vp->v_type == VLNK) {
442 error = nfs_readlinkrpc_uio(vp, &auio);
449 * [re]record creds for reading and/or writing if access
450 * was granted. Assume the NFS server will grant read access
451 * for execute requests.
454 if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) {
457 crfree(np->n_rucred);
460 if ((ap->a_mode & VWRITE) && cred != np->n_wucred) {
463 crfree(np->n_wucred);
473 * Check to see if the type is ok
474 * and that deletion is not in progress.
475 * For paged in text files, you will need to flush the page cache
476 * if consistency is lost.
478 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
483 nfs_open(struct vop_open_args *ap)
485 struct vnode *vp = ap->a_vp;
486 struct nfsnode *np = VTONFS(vp);
490 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
492 kprintf("open eacces vtyp=%d\n",vp->v_type);
498 * Save valid creds for reading and writing for later RPCs.
500 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
503 crfree(np->n_rucred);
504 np->n_rucred = ap->a_cred;
506 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
509 crfree(np->n_wucred);
510 np->n_wucred = ap->a_cred;
514 * Clear the attribute cache only if opening with write access. It
515 * is unclear if we should do this at all here, but we certainly
516 * should not clear the cache unconditionally simply because a file
519 if (ap->a_mode & FWRITE)
523 * For normal NFS, reconcile changes made locally verses
524 * changes made remotely. Note that VOP_GETATTR only goes
525 * to the wire if the cached attribute has timed out or been
528 * If local modifications have been made clear the attribute
529 * cache to force an attribute and modified time check. If
530 * GETATTR detects that the file has been changed by someone
531 * other then us it will set NRMODIFIED.
533 * If we are opening a directory and local changes have been
534 * made we have to invalidate the cache in order to ensure
535 * that we get the most up-to-date information from the
538 if (np->n_flag & NLMODIFIED) {
540 if (vp->v_type == VDIR) {
541 error = nfs_vinvalbuf(vp, V_SAVE, 1);
547 error = VOP_GETATTR(vp, &vattr);
550 if (np->n_flag & NRMODIFIED) {
551 if (vp->v_type == VDIR)
553 error = nfs_vinvalbuf(vp, V_SAVE, 1);
556 np->n_flag &= ~NRMODIFIED;
559 return (vop_stdopen(ap));
564 * What an NFS client should do upon close after writing is a debatable issue.
565 * Most NFS clients push delayed writes to the server upon close, basically for
567 * 1 - So that any write errors may be reported back to the client process
568 * doing the close system call. By far the two most likely errors are
569 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
570 * 2 - To put a worst case upper bound on cache inconsistency between
571 * multiple clients for the file.
572 * There is also a consistency problem for Version 2 of the protocol w.r.t.
573 * not being able to tell if other clients are writing a file concurrently,
574 * since there is no way of knowing if the changed modify time in the reply
575 * is only due to the write for this client.
576 * (NFS Version 3 provides weak cache consistency data in the reply that
577 * should be sufficient to detect and handle this case.)
579 * The current code does the following:
580 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
581 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
582 * or commit them (this satisfies 1 and 2 except for the
583 * case where the server crashes after this close but
584 * before the commit RPC, which is felt to be "good
585 * enough". Changing the last argument to nfs_flush() to
586 * a 1 would force a commit operation, if it is felt a
587 * commit is necessary now.
588 * for NQNFS - do nothing now, since 2 is dealt with via leases and
589 * 1 should be dealt with via an fsync() system call for
590 * cases where write errors are important.
592 * nfs_close(struct vnode *a_vp, int a_fflag)
596 nfs_close(struct vop_close_args *ap)
598 struct vnode *vp = ap->a_vp;
599 struct nfsnode *np = VTONFS(vp);
601 thread_t td = curthread;
603 if (vp->v_type == VREG) {
604 if (np->n_flag & NLMODIFIED) {
607 * Under NFSv3 we have dirty buffers to dispose of. We
608 * must flush them to the NFS server. We have the option
609 * of waiting all the way through the commit rpc or just
610 * waiting for the initial write. The default is to only
611 * wait through the initial write so the data is in the
612 * server's cache, which is roughly similar to the state
613 * a standard disk subsystem leaves the file in on close().
615 * We cannot clear the NLMODIFIED bit in np->n_flag due to
616 * potential races with other processes, and certainly
617 * cannot clear it if we don't commit.
619 int cm = nfsv3_commit_on_close ? 1 : 0;
620 error = nfs_flush(vp, MNT_WAIT, td, cm);
621 /* np->n_flag &= ~NLMODIFIED; */
623 error = nfs_vinvalbuf(vp, V_SAVE, 1);
627 if (np->n_flag & NWRITEERR) {
628 np->n_flag &= ~NWRITEERR;
637 * nfs getattr call from vfs.
639 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
642 nfs_getattr(struct vop_getattr_args *ap)
644 struct vnode *vp = ap->a_vp;
645 struct nfsnode *np = VTONFS(vp);
646 struct nfsmount *nmp;
648 thread_t td = curthread;
649 struct nfsm_info info;
652 info.v3 = NFS_ISV3(vp);
653 nmp = VFSTONFS(vp->v_mount);
656 * Update local times for special files.
658 if (np->n_flag & (NACC | NUPD))
661 * First look in the cache.
663 if (nfs_getattrcache(vp, ap->a_vap) == 0)
666 if (info.v3 && nfsaccess_cache_timeout > 0) {
667 nfsstats.accesscache_misses++;
668 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
669 if (nfs_getattrcache(vp, ap->a_vap) == 0)
673 nfsstats.rpccnt[NFSPROC_GETATTR]++;
674 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
675 ERROROUT(nfsm_fhtom(&info, vp));
676 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
677 nfs_vpcred(vp, ND_CHECK), &error));
679 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
685 * NFS doesn't support chflags flags. If the nfs mount was
686 * made -o cache set the UF_CACHE bit for swapcache.
688 if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT))
689 ap->a_vap->va_flags |= UF_CACHE;
697 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
700 nfs_setattr(struct vop_setattr_args *ap)
702 struct vnode *vp = ap->a_vp;
703 struct nfsnode *np = VTONFS(vp);
704 struct vattr *vap = ap->a_vap;
705 int biosize = vp->v_mount->mnt_stat.f_iosize;
709 thread_t td = curthread;
716 * Setting of flags is not supported.
718 if (vap->va_flags != VNOVAL)
722 * Disallow write attempts if the filesystem is mounted read-only.
724 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
725 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
726 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
727 (vp->v_mount->mnt_flag & MNT_RDONLY))
730 if (vap->va_size != VNOVAL) {
732 * truncation requested
734 switch (vp->v_type) {
741 if (vap->va_mtime.tv_sec == VNOVAL &&
742 vap->va_atime.tv_sec == VNOVAL &&
743 vap->va_mode == (mode_t)VNOVAL &&
744 vap->va_uid == (uid_t)VNOVAL &&
745 vap->va_gid == (gid_t)VNOVAL)
747 vap->va_size = VNOVAL;
751 * Disallow write attempts if the filesystem is
754 if (vp->v_mount->mnt_flag & MNT_RDONLY)
759 boff = (int)vap->va_size & (biosize - 1);
760 error = nfs_meta_setsize(vp, td, vap->va_size, 0);
763 if (np->n_flag & NLMODIFIED) {
764 if (vap->va_size == 0)
765 error = nfs_vinvalbuf(vp, 0, 1);
767 error = nfs_vinvalbuf(vp, V_SAVE, 1);
771 * note: this loop case almost always happens at
772 * least once per truncation.
774 if (error == 0 && np->n_size != vap->va_size)
776 np->n_vattr.va_size = vap->va_size;
779 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
781 * What to do. If we are modifying the mtime we lose
782 * mtime detection of changes made by the server or other
783 * clients. But programs like rsync/rdist/cpdup are going
784 * to call utimes a lot. We don't want to piecemeal sync.
786 * For now sync if any prior remote changes were detected,
787 * but allow us to lose track of remote changes made during
788 * the utimes operation.
790 if (np->n_flag & NRMODIFIED)
791 error = nfs_vinvalbuf(vp, V_SAVE, 1);
795 if (vap->va_mtime.tv_sec != VNOVAL) {
796 np->n_mtime = vap->va_mtime.tv_sec;
800 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
803 * Sanity check if a truncation was issued. This should only occur
804 * if multiple processes are racing on the same file.
806 if (error == 0 && vap->va_size != VNOVAL &&
807 np->n_size != vap->va_size) {
808 kprintf("NFS ftruncate: server disagrees on the file size: "
811 (intmax_t)vap->va_size,
812 (intmax_t)np->n_size);
815 if (error && vap->va_size != VNOVAL) {
816 np->n_size = np->n_vattr.va_size = tsize;
817 nfs_meta_setsize(vp, td, np->n_size, 0);
823 * Do an nfs setattr rpc.
826 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
827 struct ucred *cred, struct thread *td)
829 struct nfsv2_sattr *sp;
830 struct nfsnode *np = VTONFS(vp);
832 int error = 0, wccflag = NFSV3_WCCRATTR;
833 struct nfsm_info info;
836 info.v3 = NFS_ISV3(vp);
838 nfsstats.rpccnt[NFSPROC_SETATTR]++;
839 nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
840 NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
841 ERROROUT(nfsm_fhtom(&info, vp));
843 nfsm_v3attrbuild(&info, vap, TRUE);
844 tl = nfsm_build(&info, NFSX_UNSIGNED);
847 sp = nfsm_build(&info, NFSX_V2SATTR);
848 if (vap->va_mode == (mode_t)VNOVAL)
849 sp->sa_mode = nfs_xdrneg1;
851 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
852 if (vap->va_uid == (uid_t)VNOVAL)
853 sp->sa_uid = nfs_xdrneg1;
855 sp->sa_uid = txdr_unsigned(vap->va_uid);
856 if (vap->va_gid == (gid_t)VNOVAL)
857 sp->sa_gid = nfs_xdrneg1;
859 sp->sa_gid = txdr_unsigned(vap->va_gid);
860 sp->sa_size = txdr_unsigned(vap->va_size);
861 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
862 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
864 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
867 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
869 ERROROUT(nfsm_loadattr(&info, vp, NULL));
879 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
885 cache_setvp(nch, vp);
886 cache_settimeout(nch, nctimeout);
890 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
891 * nfs_lookup() until all remaining new api calls are implemented.
893 * Resolve a namecache entry. This function is passed a locked ncp and
894 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
897 nfs_nresolve(struct vop_nresolve_args *ap)
899 struct thread *td = curthread;
900 struct namecache *ncp;
911 struct nfsm_info info;
916 if ((error = vget(dvp, LK_SHARED)) != 0)
920 info.v3 = NFS_ISV3(dvp);
923 nfsstats.lookupcache_misses++;
924 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
925 ncp = ap->a_nch->ncp;
927 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
928 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
929 ERROROUT(nfsm_fhtom(&info, dvp));
930 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
931 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
932 ap->a_cred, &error));
935 * Cache negatve lookups to reduce NFS traffic, but use
936 * a fast timeout. Otherwise use a timeout of 1 tick.
937 * XXX we should add a namecache flag for no-caching
938 * to uncache the negative hit as soon as possible, but
939 * we cannot simply destroy the entry because it is used
940 * as a placeholder by the caller.
942 * The refactored nfs code will overwrite a non-zero error
943 * with 0 when we use ERROROUT(), so don't here.
946 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
947 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
959 * Success, get the file handle, do various checks, and load
960 * post-operation data from the reply packet. Theoretically
961 * we should never be looking up "." so, theoretically, we
962 * should never get the same file handle as our directory. But
963 * we check anyway. XXX
965 * Note that no timeout is set for the positive cache hit. We
966 * assume, theoretically, that ESTALE returns will be dealt with
967 * properly to handle NFS races and in anycase we cannot depend
968 * on a timeout to deal with NFS open/create/excl issues so instead
969 * of a bad hack here the rest of the NFS client code needs to do
972 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
975 if (NFS_CMPFH(np, fhp, fhsize)) {
979 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
989 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
990 NFS_LATTR_NOSHRINK));
991 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
992 NFS_LATTR_NOSHRINK));
994 ERROROUT(nfsm_loadattr(&info, nvp, NULL));
996 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
1011 * 'cached' nfs directory lookup
1013 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
1015 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
1016 * struct componentname *a_cnp)
1019 nfs_lookup(struct vop_old_lookup_args *ap)
1021 struct componentname *cnp = ap->a_cnp;
1022 struct vnode *dvp = ap->a_dvp;
1023 struct vnode **vpp = ap->a_vpp;
1024 int flags = cnp->cn_flags;
1025 struct vnode *newvp;
1026 struct nfsmount *nmp;
1030 int lockparent, wantparent, attrflag, fhsize;
1033 struct nfsm_info info;
1036 info.v3 = NFS_ISV3(dvp);
1040 * Read-only mount check and directory check.
1043 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1044 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1047 if (dvp->v_type != VDIR)
1051 * Look it up in the cache. Note that ENOENT is only returned if we
1052 * previously entered a negative hit (see later on). The additional
1053 * nfsneg_cache_timeout check causes previously cached results to
1054 * be instantly ignored if the negative caching is turned off.
1056 lockparent = flags & CNP_LOCKPARENT;
1057 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1058 nmp = VFSTONFS(dvp->v_mount);
1066 nfsstats.lookupcache_misses++;
1067 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1068 len = cnp->cn_namelen;
1069 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1070 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1071 ERROROUT(nfsm_fhtom(&info, dvp));
1072 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1073 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1074 cnp->cn_cred, &error));
1076 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1077 NFS_LATTR_NOSHRINK);
1087 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1090 * Handle RENAME case...
1092 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1093 if (NFS_CMPFH(np, fhp, fhsize)) {
1098 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1106 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1107 NFS_LATTR_NOSHRINK));
1108 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1109 NFS_LATTR_NOSHRINK));
1111 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1118 cnp->cn_flags |= CNP_PDIRUNLOCK;
1123 if (flags & CNP_ISDOTDOT) {
1125 cnp->cn_flags |= CNP_PDIRUNLOCK;
1126 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1128 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1129 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1130 return (error); /* NOTE: return error from nget */
1134 error = vn_lock(dvp, LK_EXCLUSIVE);
1139 cnp->cn_flags |= CNP_PDIRUNLOCK;
1141 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1145 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1153 cnp->cn_flags |= CNP_PDIRUNLOCK;
1158 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1159 NFS_LATTR_NOSHRINK));
1160 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1161 NFS_LATTR_NOSHRINK));
1163 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1166 /* XXX MOVE TO nfs_nremove() */
1167 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1168 cnp->cn_nameiop != NAMEI_DELETE) {
1169 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1177 if (newvp != NULLVP) {
1181 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1182 cnp->cn_nameiop == NAMEI_RENAME) &&
1186 cnp->cn_flags |= CNP_PDIRUNLOCK;
1188 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1191 error = EJUSTRETURN;
1199 * Just call nfs_bioread() to do the work.
1201 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1202 * struct ucred *a_cred)
1205 nfs_read(struct vop_read_args *ap)
1207 struct vnode *vp = ap->a_vp;
1209 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1215 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1218 nfs_readlink(struct vop_readlink_args *ap)
1220 struct vnode *vp = ap->a_vp;
1222 if (vp->v_type != VLNK)
1224 return (nfs_bioread(vp, ap->a_uio, 0));
1228 * Do a readlink rpc.
1229 * Called by nfs_doio() from below the buffer cache.
1232 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1234 int error = 0, len, attrflag;
1235 struct nfsm_info info;
1238 info.v3 = NFS_ISV3(vp);
1240 nfsstats.rpccnt[NFSPROC_READLINK]++;
1241 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1242 ERROROUT(nfsm_fhtom(&info, vp));
1243 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1244 nfs_vpcred(vp, ND_CHECK), &error));
1246 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1247 NFS_LATTR_NOSHRINK));
1250 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1251 if (len == NFS_MAXPATHLEN) {
1252 struct nfsnode *np = VTONFS(vp);
1253 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1256 ERROROUT(nfsm_mtouio(&info, uiop, len));
1265 * nfs synchronous read rpc using UIO
1268 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1271 struct nfsmount *nmp;
1272 int error = 0, len, retlen, tsiz, eof, attrflag;
1273 struct nfsm_info info;
1277 info.v3 = NFS_ISV3(vp);
1282 nmp = VFSTONFS(vp->v_mount);
1283 tsiz = uiop->uio_resid;
1284 tmp_off = uiop->uio_offset + tsiz;
1285 if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1287 tmp_off = uiop->uio_offset;
1289 nfsstats.rpccnt[NFSPROC_READ]++;
1290 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1291 nfsm_reqhead(&info, vp, NFSPROC_READ,
1292 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1293 ERROROUT(nfsm_fhtom(&info, vp));
1294 tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1296 txdr_hyper(uiop->uio_offset, tl);
1297 *(tl + 2) = txdr_unsigned(len);
1299 *tl++ = txdr_unsigned(uiop->uio_offset);
1300 *tl++ = txdr_unsigned(len);
1303 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1304 nfs_vpcred(vp, ND_READ), &error));
1306 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1307 NFS_LATTR_NOSHRINK));
1308 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1309 eof = fxdr_unsigned(int, *(tl + 1));
1311 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1313 NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1314 ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1319 * Handle short-read from server (NFSv3). If EOF is not
1320 * flagged (and no error occurred), but retlen is less
1321 * then the request size, we must zero-fill the remainder.
1323 if (retlen < len && info.v3 && eof == 0) {
1324 ERROROUT(uiomovez(len - retlen, uiop));
1330 * Terminate loop on EOF or zero-length read.
1332 * For NFSv2 a short-read indicates EOF, not zero-fill,
1333 * and also terminates the loop.
1336 if (eof || retlen == 0)
1338 } else if (retlen < len) {
1350 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1351 int *iomode, int *must_commit)
1355 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1356 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1357 int committed = NFSV3WRITE_FILESYNC;
1358 struct nfsm_info info;
1361 info.v3 = NFS_ISV3(vp);
1364 if (uiop->uio_iovcnt != 1)
1365 panic("nfs: writerpc iovcnt > 1");
1368 tsiz = uiop->uio_resid;
1369 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1372 nfsstats.rpccnt[NFSPROC_WRITE]++;
1373 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1374 nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1375 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1376 ERROROUT(nfsm_fhtom(&info, vp));
1378 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1379 txdr_hyper(uiop->uio_offset, tl);
1381 *tl++ = txdr_unsigned(len);
1382 *tl++ = txdr_unsigned(*iomode);
1383 *tl = txdr_unsigned(len);
1387 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1388 /* Set both "begin" and "current" to non-garbage. */
1389 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1390 *tl++ = x; /* "begin offset" */
1391 *tl++ = x; /* "current offset" */
1392 x = txdr_unsigned(len);
1393 *tl++ = x; /* total to this offset */
1394 *tl = x; /* size of this write */
1396 ERROROUT(nfsm_uiotom(&info, uiop, len));
1397 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1398 nfs_vpcred(vp, ND_WRITE), &error));
1401 * The write RPC returns a before and after mtime. The
1402 * nfsm_wcc_data() macro checks the before n_mtime
1403 * against the before time and stores the after time
1404 * in the nfsnode's cached vattr and n_mtime field.
1405 * The NRMODIFIED bit will be set if the before
1406 * time did not match the original mtime.
1408 wccflag = NFSV3_WCCCHK;
1409 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1411 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1412 rlen = fxdr_unsigned(int, *tl++);
1418 } else if (rlen < len) {
1419 backup = len - rlen;
1420 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1421 uiop->uio_iov->iov_len += backup;
1422 uiop->uio_offset -= backup;
1423 uiop->uio_resid += backup;
1426 commit = fxdr_unsigned(int, *tl++);
1429 * Return the lowest committment level
1430 * obtained by any of the RPCs.
1432 if (committed == NFSV3WRITE_FILESYNC)
1434 else if (committed == NFSV3WRITE_DATASYNC &&
1435 commit == NFSV3WRITE_UNSTABLE)
1437 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1438 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1440 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1441 } else if (bcmp((caddr_t)tl,
1442 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1444 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1449 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1458 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1459 committed = NFSV3WRITE_FILESYNC;
1460 *iomode = committed;
1462 uiop->uio_resid = tsiz;
1468 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1469 * mode set to specify the file type and the size field for rdev.
1472 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1475 struct nfsv2_sattr *sp;
1477 struct vnode *newvp = NULL;
1478 struct nfsnode *np = NULL;
1480 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1482 struct nfsm_info info;
1485 info.v3 = NFS_ISV3(dvp);
1487 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1488 rmajor = txdr_unsigned(vap->va_rmajor);
1489 rminor = txdr_unsigned(vap->va_rminor);
1490 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1491 rmajor = nfs_xdrneg1;
1492 rminor = nfs_xdrneg1;
1494 return (EOPNOTSUPP);
1496 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1499 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1500 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1501 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1502 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1503 ERROROUT(nfsm_fhtom(&info, dvp));
1504 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1507 tl = nfsm_build(&info, NFSX_UNSIGNED);
1508 *tl++ = vtonfsv3_type(vap->va_type);
1509 nfsm_v3attrbuild(&info, vap, FALSE);
1510 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1511 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1512 *tl++ = txdr_unsigned(vap->va_rmajor);
1513 *tl = txdr_unsigned(vap->va_rminor);
1516 sp = nfsm_build(&info, NFSX_V2SATTR);
1517 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1518 sp->sa_uid = nfs_xdrneg1;
1519 sp->sa_gid = nfs_xdrneg1;
1520 sp->sa_size = makeudev(rmajor, rminor);
1521 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1522 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1524 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1525 cnp->cn_cred, &error));
1527 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1533 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1534 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1540 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1551 VTONFS(dvp)->n_flag |= NLMODIFIED;
1553 VTONFS(dvp)->n_attrstamp = 0;
1559 * just call nfs_mknodrpc() to do the work.
1561 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1562 * struct componentname *a_cnp, struct vattr *a_vap)
1566 nfs_mknod(struct vop_old_mknod_args *ap)
1568 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1571 static u_long create_verf;
1573 * nfs file create call
1575 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1576 * struct componentname *a_cnp, struct vattr *a_vap)
1579 nfs_create(struct vop_old_create_args *ap)
1581 struct vnode *dvp = ap->a_dvp;
1582 struct vattr *vap = ap->a_vap;
1583 struct componentname *cnp = ap->a_cnp;
1584 struct nfsv2_sattr *sp;
1586 struct nfsnode *np = NULL;
1587 struct vnode *newvp = NULL;
1588 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1590 struct nfsm_info info;
1593 info.v3 = NFS_ISV3(dvp);
1596 * Oops, not for me..
1598 if (vap->va_type == VSOCK)
1599 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1601 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1604 if (vap->va_vaflags & VA_EXCLUSIVE)
1607 nfsstats.rpccnt[NFSPROC_CREATE]++;
1608 nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1609 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1610 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1611 ERROROUT(nfsm_fhtom(&info, dvp));
1612 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1615 tl = nfsm_build(&info, NFSX_UNSIGNED);
1616 if (fmode & O_EXCL) {
1617 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1618 tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1620 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1621 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1624 *tl++ = create_verf;
1625 *tl = ++create_verf;
1627 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1628 nfsm_v3attrbuild(&info, vap, FALSE);
1631 sp = nfsm_build(&info, NFSX_V2SATTR);
1632 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1633 sp->sa_uid = nfs_xdrneg1;
1634 sp->sa_gid = nfs_xdrneg1;
1636 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1637 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1639 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1640 cnp->cn_cred, &error));
1642 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1648 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1649 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1656 error = nfsm_wcc_data(&info, dvp, &wccflag);
1658 (void)nfsm_wcc_data(&info, dvp, &wccflag);
1664 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1665 KKASSERT(newvp == NULL);
1669 } else if (info.v3 && (fmode & O_EXCL)) {
1671 * We are normally called with only a partially initialized
1672 * VAP. Since the NFSv3 spec says that server may use the
1673 * file attributes to store the verifier, the spec requires
1674 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1675 * in atime, but we can't really assume that all servers will
1676 * so we ensure that our SETATTR sets both atime and mtime.
1678 if (vap->va_mtime.tv_sec == VNOVAL)
1679 vfs_timestamp(&vap->va_mtime);
1680 if (vap->va_atime.tv_sec == VNOVAL)
1681 vap->va_atime = vap->va_mtime;
1682 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1686 * The new np may have enough info for access
1687 * checks, make sure rucred and wucred are
1688 * initialized for read and write rpc's.
1691 if (np->n_rucred == NULL)
1692 np->n_rucred = crhold(cnp->cn_cred);
1693 if (np->n_wucred == NULL)
1694 np->n_wucred = crhold(cnp->cn_cred);
1699 VTONFS(dvp)->n_flag |= NLMODIFIED;
1701 VTONFS(dvp)->n_attrstamp = 0;
1706 * nfs file remove call
1707 * To try and make nfs semantics closer to ufs semantics, a file that has
1708 * other processes using the vnode is renamed instead of removed and then
1709 * removed later on the last close.
1710 * - If v_sysref.refcnt > 1
1711 * If a rename is not already in the works
1712 * call nfs_sillyrename() to set it up
1716 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1717 * struct componentname *a_cnp)
1720 nfs_remove(struct vop_old_remove_args *ap)
1722 struct vnode *vp = ap->a_vp;
1723 struct vnode *dvp = ap->a_dvp;
1724 struct componentname *cnp = ap->a_cnp;
1725 struct nfsnode *np = VTONFS(vp);
1730 if (vp->v_sysref.refcnt < 1)
1731 panic("nfs_remove: bad v_sysref.refcnt");
1733 if (vp->v_type == VDIR)
1735 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1736 VOP_GETATTR(vp, &vattr) == 0 &&
1737 vattr.va_nlink > 1)) {
1739 * throw away biocache buffers, mainly to avoid
1740 * unnecessary delayed writes later.
1742 error = nfs_vinvalbuf(vp, 0, 1);
1745 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1746 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1748 * Kludge City: If the first reply to the remove rpc is lost..
1749 * the reply to the retransmitted request will be ENOENT
1750 * since the file was in fact removed
1751 * Therefore, we cheat and return success.
1753 if (error == ENOENT)
1755 } else if (!np->n_sillyrename) {
1756 error = nfs_sillyrename(dvp, vp, cnp);
1758 np->n_attrstamp = 0;
1763 * nfs file remove rpc called from nfs_inactive
1766 nfs_removeit(struct sillyrename *sp)
1768 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1773 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1776 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1777 struct ucred *cred, struct thread *td)
1779 int error = 0, wccflag = NFSV3_WCCRATTR;
1780 struct nfsm_info info;
1783 info.v3 = NFS_ISV3(dvp);
1785 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1786 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1787 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1788 ERROROUT(nfsm_fhtom(&info, dvp));
1789 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1790 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1792 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1797 VTONFS(dvp)->n_flag |= NLMODIFIED;
1799 VTONFS(dvp)->n_attrstamp = 0;
1804 * nfs file rename call
1806 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1807 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1808 * struct vnode *a_tvp, struct componentname *a_tcnp)
1811 nfs_rename(struct vop_old_rename_args *ap)
1813 struct vnode *fvp = ap->a_fvp;
1814 struct vnode *tvp = ap->a_tvp;
1815 struct vnode *fdvp = ap->a_fdvp;
1816 struct vnode *tdvp = ap->a_tdvp;
1817 struct componentname *tcnp = ap->a_tcnp;
1818 struct componentname *fcnp = ap->a_fcnp;
1821 /* Check for cross-device rename */
1822 if ((fvp->v_mount != tdvp->v_mount) ||
1823 (tvp && (fvp->v_mount != tvp->v_mount))) {
1829 * We shouldn't have to flush fvp on rename for most server-side
1830 * filesystems as the file handle should not change. Unfortunately
1831 * the inode for some filesystems (msdosfs) might be tied to the
1832 * file name or directory position so to be completely safe
1833 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1836 * We must flush tvp on rename because it might become stale on the
1837 * server after the rename.
1839 if (nfs_flush_on_rename)
1840 VOP_FSYNC(fvp, MNT_WAIT, 0);
1842 VOP_FSYNC(tvp, MNT_WAIT, 0);
1845 * If the tvp exists and is in use, sillyrename it before doing the
1846 * rename of the new file over it.
1848 * XXX Can't sillyrename a directory.
1850 * We do not attempt to do any namecache purges in this old API
1851 * routine. The new API compat functions have access to the actual
1852 * namecache structures and will do it for us.
1854 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1855 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1862 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1863 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1876 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1878 if (error == ENOENT)
1884 * nfs file rename rpc called from nfs_remove() above
1887 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1888 struct sillyrename *sp)
1890 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1891 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1895 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1898 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1899 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1900 struct ucred *cred, struct thread *td)
1902 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1903 struct nfsm_info info;
1906 info.v3 = NFS_ISV3(fdvp);
1908 nfsstats.rpccnt[NFSPROC_RENAME]++;
1909 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1910 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1911 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1912 ERROROUT(nfsm_fhtom(&info, fdvp));
1913 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1914 ERROROUT(nfsm_fhtom(&info, tdvp));
1915 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1916 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
1918 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
1919 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
1924 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1925 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1927 VTONFS(fdvp)->n_attrstamp = 0;
1929 VTONFS(tdvp)->n_attrstamp = 0;
1934 * nfs hard link create call
1936 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1937 * struct componentname *a_cnp)
1940 nfs_link(struct vop_old_link_args *ap)
1942 struct vnode *vp = ap->a_vp;
1943 struct vnode *tdvp = ap->a_tdvp;
1944 struct componentname *cnp = ap->a_cnp;
1945 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1946 struct nfsm_info info;
1948 if (vp->v_mount != tdvp->v_mount) {
1953 * The attribute cache may get out of sync with the server on link.
1954 * Pushing writes to the server before handle was inherited from
1955 * long long ago and it is unclear if we still need to do this.
1958 if (nfs_flush_on_hlink)
1959 VOP_FSYNC(vp, MNT_WAIT, 0);
1962 info.v3 = NFS_ISV3(vp);
1964 nfsstats.rpccnt[NFSPROC_LINK]++;
1965 nfsm_reqhead(&info, vp, NFSPROC_LINK,
1966 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
1967 nfsm_rndup(cnp->cn_namelen));
1968 ERROROUT(nfsm_fhtom(&info, vp));
1969 ERROROUT(nfsm_fhtom(&info, tdvp));
1970 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1972 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
1973 cnp->cn_cred, &error));
1975 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1976 NFS_LATTR_NOSHRINK));
1977 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
1982 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1984 VTONFS(vp)->n_attrstamp = 0;
1986 VTONFS(tdvp)->n_attrstamp = 0;
1988 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1990 if (error == EEXIST)
1996 * nfs symbolic link create call
1998 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1999 * struct componentname *a_cnp, struct vattr *a_vap,
2003 nfs_symlink(struct vop_old_symlink_args *ap)
2005 struct vnode *dvp = ap->a_dvp;
2006 struct vattr *vap = ap->a_vap;
2007 struct componentname *cnp = ap->a_cnp;
2008 struct nfsv2_sattr *sp;
2009 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2010 struct vnode *newvp = NULL;
2011 struct nfsm_info info;
2014 info.v3 = NFS_ISV3(dvp);
2016 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2017 slen = strlen(ap->a_target);
2018 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
2019 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
2020 nfsm_rndup(cnp->cn_namelen) +
2021 nfsm_rndup(slen) + NFSX_SATTR(info.v3));
2022 ERROROUT(nfsm_fhtom(&info, dvp));
2023 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2026 nfsm_v3attrbuild(&info, vap, FALSE);
2028 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
2030 sp = nfsm_build(&info, NFSX_V2SATTR);
2031 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2032 sp->sa_uid = nfs_xdrneg1;
2033 sp->sa_gid = nfs_xdrneg1;
2034 sp->sa_size = nfs_xdrneg1;
2035 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2036 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2040 * Issue the NFS request and get the rpc response.
2042 * Only NFSv3 responses returning an error of 0 actually return
2043 * a file handle that can be converted into newvp without having
2044 * to do an extra lookup rpc.
2046 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
2047 cnp->cn_cred, &error));
2050 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2052 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2056 * out code jumps -> here, mrep is also freed.
2064 * If we get an EEXIST error, silently convert it to no-error
2065 * in case of an NFS retry.
2067 if (error == EEXIST)
2071 * If we do not have (or no longer have) an error, and we could
2072 * not extract the newvp from the response due to the request being
2073 * NFSv2 or the error being EEXIST. We have to do a lookup in order
2074 * to obtain a newvp to return.
2076 if (error == 0 && newvp == NULL) {
2077 struct nfsnode *np = NULL;
2079 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2080 cnp->cn_cred, cnp->cn_td, &np);
2090 VTONFS(dvp)->n_flag |= NLMODIFIED;
2092 VTONFS(dvp)->n_attrstamp = 0;
2099 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2100 * struct componentname *a_cnp, struct vattr *a_vap)
2103 nfs_mkdir(struct vop_old_mkdir_args *ap)
2105 struct vnode *dvp = ap->a_dvp;
2106 struct vattr *vap = ap->a_vap;
2107 struct componentname *cnp = ap->a_cnp;
2108 struct nfsv2_sattr *sp;
2109 struct nfsnode *np = NULL;
2110 struct vnode *newvp = NULL;
2112 int error = 0, wccflag = NFSV3_WCCRATTR;
2115 struct nfsm_info info;
2118 info.v3 = NFS_ISV3(dvp);
2120 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2123 len = cnp->cn_namelen;
2124 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2125 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2126 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2127 nfsm_rndup(len) + NFSX_SATTR(info.v3));
2128 ERROROUT(nfsm_fhtom(&info, dvp));
2129 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2131 nfsm_v3attrbuild(&info, vap, FALSE);
2133 sp = nfsm_build(&info, NFSX_V2SATTR);
2134 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2135 sp->sa_uid = nfs_xdrneg1;
2136 sp->sa_gid = nfs_xdrneg1;
2137 sp->sa_size = nfs_xdrneg1;
2138 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2139 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2141 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2142 cnp->cn_cred, &error));
2144 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2147 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2152 VTONFS(dvp)->n_flag |= NLMODIFIED;
2154 VTONFS(dvp)->n_attrstamp = 0;
2156 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2157 * if we can succeed in looking up the directory.
2159 if (error == EEXIST || (!error && !gotvp)) {
2164 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2168 if (newvp->v_type != VDIR)
2181 * nfs remove directory call
2183 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2184 * struct componentname *a_cnp)
2187 nfs_rmdir(struct vop_old_rmdir_args *ap)
2189 struct vnode *vp = ap->a_vp;
2190 struct vnode *dvp = ap->a_dvp;
2191 struct componentname *cnp = ap->a_cnp;
2192 int error = 0, wccflag = NFSV3_WCCRATTR;
2193 struct nfsm_info info;
2196 info.v3 = NFS_ISV3(dvp);
2200 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2201 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2202 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2203 nfsm_rndup(cnp->cn_namelen));
2204 ERROROUT(nfsm_fhtom(&info, dvp));
2205 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2207 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2208 cnp->cn_cred, &error));
2210 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2215 VTONFS(dvp)->n_flag |= NLMODIFIED;
2217 VTONFS(dvp)->n_attrstamp = 0;
2219 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2221 if (error == ENOENT)
2229 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2232 nfs_readdir(struct vop_readdir_args *ap)
2234 struct vnode *vp = ap->a_vp;
2235 struct nfsnode *np = VTONFS(vp);
2236 struct uio *uio = ap->a_uio;
2240 if (vp->v_type != VDIR)
2243 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2247 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2248 * and then check that is still valid, or if this is an NQNFS mount
2249 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2250 * VOP_GETATTR() does not necessarily go to the wire.
2252 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2253 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2254 if (VOP_GETATTR(vp, &vattr) == 0 &&
2255 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2257 nfsstats.direofcache_hits++;
2263 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2264 * own cache coherency checks so we do not have to.
2266 tresid = uio->uio_resid;
2267 error = nfs_bioread(vp, uio, 0);
2269 if (!error && uio->uio_resid == tresid)
2270 nfsstats.direofcache_misses++;
2277 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2279 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2280 * offset/block and converts the nfs formatted directory entries for userland
2281 * consumption as well as deals with offsets into the middle of blocks.
2282 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2283 * be block-bounded. It must convert to cookies for the actual RPC.
2286 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2289 struct nfs_dirent *dp = NULL;
2294 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2295 struct nfsnode *dnp = VTONFS(vp);
2297 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2299 struct nfsm_info info;
2302 info.v3 = NFS_ISV3(vp);
2305 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2306 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2307 panic("nfs readdirrpc bad uio");
2311 * If there is no cookie, assume directory was stale.
2313 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2317 return (NFSERR_BAD_COOKIE);
2319 * Loop around doing readdir rpc's of size nm_readdirsize
2320 * truncated to a multiple of DIRBLKSIZ.
2321 * The stopping criteria is EOF or buffer full.
2323 while (more_dirs && bigenough) {
2324 nfsstats.rpccnt[NFSPROC_READDIR]++;
2325 nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2326 NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2327 ERROROUT(nfsm_fhtom(&info, vp));
2329 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2330 *tl++ = cookie.nfsuquad[0];
2331 *tl++ = cookie.nfsuquad[1];
2332 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2333 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2335 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2336 *tl++ = cookie.nfsuquad[0];
2338 *tl = txdr_unsigned(nmp->nm_readdirsize);
2339 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2341 nfs_vpcred(vp, ND_READ), &error));
2343 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2344 NFS_LATTR_NOSHRINK));
2345 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2346 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2347 dnp->n_cookieverf.nfsuquad[1] = *tl;
2349 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2350 more_dirs = fxdr_unsigned(int, *tl);
2352 /* loop thru the dir entries, converting them to std form */
2353 while (more_dirs && bigenough) {
2355 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2356 fileno = fxdr_hyper(tl);
2357 len = fxdr_unsigned(int, *(tl + 2));
2359 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2360 fileno = fxdr_unsigned(u_quad_t, *tl++);
2361 len = fxdr_unsigned(int, *tl);
2363 if (len <= 0 || len > NFS_MAXNAMLEN) {
2371 * len is the number of bytes in the path element
2372 * name, not including the \0 termination.
2374 * tlen is the number of bytes w have to reserve for
2375 * the path element name.
2377 tlen = nfsm_rndup(len);
2379 tlen += 4; /* To ensure null termination */
2382 * If the entry would cross a DIRBLKSIZ boundary,
2383 * extend the previous nfs_dirent to cover the
2386 left = DIRBLKSIZ - blksiz;
2387 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2388 dp->nfs_reclen += left;
2389 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2390 uiop->uio_iov->iov_len -= left;
2391 uiop->uio_offset += left;
2392 uiop->uio_resid -= left;
2395 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2398 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2399 dp->nfs_ino = fileno;
2400 dp->nfs_namlen = len;
2401 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2402 dp->nfs_type = DT_UNKNOWN;
2403 blksiz += dp->nfs_reclen;
2404 if (blksiz == DIRBLKSIZ)
2406 uiop->uio_offset += sizeof(struct nfs_dirent);
2407 uiop->uio_resid -= sizeof(struct nfs_dirent);
2408 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2409 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2410 ERROROUT(nfsm_mtouio(&info, uiop, len));
2413 * The uiop has advanced by nfs_dirent + len
2414 * but really needs to advance by
2417 cp = uiop->uio_iov->iov_base;
2419 *cp = '\0'; /* null terminate */
2420 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2421 uiop->uio_iov->iov_len -= tlen;
2422 uiop->uio_offset += tlen;
2423 uiop->uio_resid -= tlen;
2426 * NFS strings must be rounded up (nfsm_myouio
2427 * handled that in the bigenough case).
2429 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2432 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2434 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2438 * If we were able to accomodate the last entry,
2439 * get the cookie for the next one. Otherwise
2440 * hold-over the cookie for the one we were not
2441 * able to accomodate.
2444 cookie.nfsuquad[0] = *tl++;
2446 cookie.nfsuquad[1] = *tl++;
2447 } else if (info.v3) {
2452 more_dirs = fxdr_unsigned(int, *tl);
2455 * If at end of rpc data, get the eof boolean
2458 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2459 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2465 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2466 * by increasing d_reclen for the last record.
2469 left = DIRBLKSIZ - blksiz;
2470 dp->nfs_reclen += left;
2471 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2472 uiop->uio_iov->iov_len -= left;
2473 uiop->uio_offset += left;
2474 uiop->uio_resid -= left;
2479 * We hit the end of the directory, update direofoffset.
2481 dnp->n_direofoffset = uiop->uio_offset;
2484 * There is more to go, insert the link cookie so the
2485 * next block can be read.
2487 if (uiop->uio_resid > 0)
2488 kprintf("EEK! readdirrpc resid > 0\n");
2489 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2497 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2500 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2503 struct nfs_dirent *dp;
2505 struct vnode *newvp;
2507 caddr_t dpossav1, dpossav2;
2509 struct mbuf *mdsav1, *mdsav2;
2511 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2512 struct nfsnode *dnp = VTONFS(vp), *np;
2515 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2516 int attrflag, fhsize;
2517 struct nchandle nch;
2518 struct nchandle dnch;
2519 struct nlcomponent nlc;
2520 struct nfsm_info info;
2529 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2530 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2531 panic("nfs readdirplusrpc bad uio");
2534 * Obtain the namecache record for the directory so we have something
2535 * to use as a basis for creating the entries. This function will
2536 * return a held (but not locked) ncp. The ncp may be disconnected
2537 * from the tree and cannot be used for upward traversals, and the
2538 * ncp may be unnamed. Note that other unrelated operations may
2539 * cause the ncp to be named at any time.
2541 * We have to lock the ncp to prevent a lock order reversal when
2542 * rdirplus does nlookups of the children, because the vnode is
2543 * locked and has to stay that way.
2545 cache_fromdvp(vp, NULL, 0, &dnch);
2546 bzero(&nlc, sizeof(nlc));
2550 * If there is no cookie, assume directory was stale.
2552 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2558 return (NFSERR_BAD_COOKIE);
2562 * Loop around doing readdir rpc's of size nm_readdirsize
2563 * truncated to a multiple of DIRBLKSIZ.
2564 * The stopping criteria is EOF or buffer full.
2566 while (more_dirs && bigenough) {
2567 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2568 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2569 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2570 ERROROUT(nfsm_fhtom(&info, vp));
2571 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2572 *tl++ = cookie.nfsuquad[0];
2573 *tl++ = cookie.nfsuquad[1];
2574 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2575 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2576 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2577 *tl = txdr_unsigned(nmp->nm_rsize);
2578 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2580 nfs_vpcred(vp, ND_READ), &error));
2581 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2582 NFS_LATTR_NOSHRINK));
2583 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2584 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2585 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2586 more_dirs = fxdr_unsigned(int, *tl);
2588 /* loop thru the dir entries, doctoring them to 4bsd form */
2589 while (more_dirs && bigenough) {
2590 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2591 fileno = fxdr_hyper(tl);
2592 len = fxdr_unsigned(int, *(tl + 2));
2593 if (len <= 0 || len > NFS_MAXNAMLEN) {
2599 tlen = nfsm_rndup(len);
2601 tlen += 4; /* To ensure null termination*/
2602 left = DIRBLKSIZ - blksiz;
2603 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2604 dp->nfs_reclen += left;
2605 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2606 uiop->uio_iov->iov_len -= left;
2607 uiop->uio_offset += left;
2608 uiop->uio_resid -= left;
2611 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2614 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2615 dp->nfs_ino = fileno;
2616 dp->nfs_namlen = len;
2617 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2618 dp->nfs_type = DT_UNKNOWN;
2619 blksiz += dp->nfs_reclen;
2620 if (blksiz == DIRBLKSIZ)
2622 uiop->uio_offset += sizeof(struct nfs_dirent);
2623 uiop->uio_resid -= sizeof(struct nfs_dirent);
2624 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2625 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2626 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2627 nlc.nlc_namelen = len;
2628 ERROROUT(nfsm_mtouio(&info, uiop, len));
2629 cp = uiop->uio_iov->iov_base;
2632 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2633 uiop->uio_iov->iov_len -= tlen;
2634 uiop->uio_offset += tlen;
2635 uiop->uio_resid -= tlen;
2637 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2639 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2641 cookie.nfsuquad[0] = *tl++;
2642 cookie.nfsuquad[1] = *tl++;
2648 * Since the attributes are before the file handle
2649 * (sigh), we must skip over the attributes and then
2650 * come back and get them.
2652 attrflag = fxdr_unsigned(int, *tl);
2654 dpossav1 = info.dpos;
2656 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2657 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2658 doit = fxdr_unsigned(int, *tl);
2660 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2662 if (doit && bigenough && !nlcdegenerate(&nlc) &&
2663 !NFS_CMPFH(dnp, fhp, fhsize)
2667 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2668 nlc.nlc_namelen, nlc.nlc_namelen,
2672 * This is a bit hokey but there isn't
2673 * much we can do about it. We can't
2674 * hold the directory vp locked while
2675 * doing lookups and gets.
2677 nch = cache_nlookup_nonblock(&dnch, &nlc);
2678 if (nch.ncp == NULL)
2680 cache_setunresolved(&nch);
2681 error = nfs_nget_nonblock(vp->v_mount, fhp,
2688 dpossav2 = info.dpos;
2689 info.dpos = dpossav1;
2692 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2693 info.dpos = dpossav2;
2696 IFTODT(VTTOIF(np->n_vattr.va_type));
2697 nfs_cache_setvp(&nch, newvp,
2698 nfspos_cache_timeout);
2706 kprintf("Warning: NFS/rddirplus, "
2707 "UNABLE TO ENTER %*.*s\n",
2708 nlc.nlc_namelen, nlc.nlc_namelen,
2714 /* Just skip over the file handle */
2715 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2716 i = fxdr_unsigned(int, *tl);
2717 ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2719 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2720 more_dirs = fxdr_unsigned(int, *tl);
2723 * If at end of rpc data, get the eof boolean
2726 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2727 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2733 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2734 * by increasing d_reclen for the last record.
2737 left = DIRBLKSIZ - blksiz;
2738 dp->nfs_reclen += left;
2739 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2740 uiop->uio_iov->iov_len -= left;
2741 uiop->uio_offset += left;
2742 uiop->uio_resid -= left;
2746 * We are now either at the end of the directory or have filled the
2750 dnp->n_direofoffset = uiop->uio_offset;
2752 if (uiop->uio_resid > 0)
2753 kprintf("EEK! readdirplusrpc resid > 0\n");
2754 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2758 if (newvp != NULLVP) {
2771 * Silly rename. To make the NFS filesystem that is stateless look a little
2772 * more like the "ufs" a remove of an active vnode is translated to a rename
2773 * to a funny looking filename that is removed by nfs_inactive on the
2774 * nfsnode. There is the potential for another process on a different client
2775 * to create the same funny name between the nfs_lookitup() fails and the
2776 * nfs_rename() completes, but...
2779 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2781 struct sillyrename *sp;
2786 * We previously purged dvp instead of vp. I don't know why, it
2787 * completely destroys performance. We can't do it anyway with the
2788 * new VFS API since we would be breaking the namecache topology.
2790 cache_purge(vp); /* XXX */
2793 if (vp->v_type == VDIR)
2794 panic("nfs: sillyrename dir");
2796 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2797 M_NFSREQ, M_WAITOK);
2798 sp->s_cred = crdup(cnp->cn_cred);
2802 /* Fudge together a funny name */
2803 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2804 (int)(intptr_t)cnp->cn_td);
2806 /* Try lookitups until we get one that isn't there */
2807 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2808 cnp->cn_td, NULL) == 0) {
2810 if (sp->s_name[4] > 'z') {
2815 error = nfs_renameit(dvp, cnp, sp);
2818 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2820 np->n_sillyrename = sp;
2825 kfree((caddr_t)sp, M_NFSREQ);
2830 * Look up a file name and optionally either update the file handle or
2831 * allocate an nfsnode, depending on the value of npp.
2832 * npp == NULL --> just do the lookup
2833 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2835 * *npp != NULL --> update the file handle in the vnode
2838 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2839 struct thread *td, struct nfsnode **npp)
2841 struct vnode *newvp = NULL;
2842 struct nfsnode *np, *dnp = VTONFS(dvp);
2843 int error = 0, fhlen, attrflag;
2845 struct nfsm_info info;
2848 info.v3 = NFS_ISV3(dvp);
2850 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2851 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2852 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2853 ERROROUT(nfsm_fhtom(&info, dvp));
2854 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2855 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2856 if (npp && !error) {
2857 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2860 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2861 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2862 np->n_fhp = &np->n_fh;
2863 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2864 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2865 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2866 np->n_fhsize = fhlen;
2868 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2872 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2881 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2882 NFS_LATTR_NOSHRINK));
2883 if (!attrflag && *npp == NULL) {
2893 ERROROUT(error = nfsm_loadattr(&info, newvp, NULL));
2899 if (npp && *npp == NULL) {
2914 * Nfs Version 3 commit rpc
2916 * We call it 'uio' to distinguish it from 'bio' but there is no real uio
2920 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2922 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2923 int error = 0, wccflag = NFSV3_WCCRATTR;
2924 struct nfsm_info info;
2930 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2932 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2933 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
2934 ERROROUT(nfsm_fhtom(&info, vp));
2935 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
2936 txdr_hyper(offset, tl);
2938 *tl = txdr_unsigned(cnt);
2939 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
2940 nfs_vpcred(vp, ND_WRITE), &error));
2941 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
2943 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
2944 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2945 NFSX_V3WRITEVERF)) {
2946 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2948 error = NFSERR_STALEWRITEVERF;
2959 * - make nfs_bmap() essentially a no-op that does no translation
2960 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2961 * (Maybe I could use the process's page mapping, but I was concerned that
2962 * Kernel Write might not be enabled and also figured copyout() would do
2963 * a lot more work than bcopy() and also it currently happens in the
2964 * context of the swapper process (2).
2966 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
2967 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2970 nfs_bmap(struct vop_bmap_args *ap)
2972 if (ap->a_doffsetp != NULL)
2973 *ap->a_doffsetp = ap->a_loffset;
2974 if (ap->a_runp != NULL)
2976 if (ap->a_runb != NULL)
2985 nfs_strategy(struct vop_strategy_args *ap)
2987 struct bio *bio = ap->a_bio;
2989 struct buf *bp __debugvar = bio->bio_buf;
2993 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2994 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
2995 KASSERT(BUF_REFCNT(bp) > 0,
2996 ("nfs_strategy: buffer %p not locked", bp));
2998 if (bio->bio_flags & BIO_SYNC)
2999 td = curthread; /* XXX */
3004 * We probably don't need to push an nbio any more since no
3005 * block conversion is required due to the use of 64 bit byte
3006 * offsets, but do it anyway.
3008 * NOTE: When NFS callers itself via this strategy routines and
3009 * sets up a synchronous I/O, it expects the I/O to run
3010 * synchronously (its bio_done routine just assumes it),
3011 * so for now we have to honor the bit.
3013 nbio = push_bio(bio);
3014 nbio->bio_offset = bio->bio_offset;
3015 nbio->bio_flags = bio->bio_flags & BIO_SYNC;
3018 * If the op is asynchronous and an i/o daemon is waiting
3019 * queue the request, wake it up and wait for completion
3020 * otherwise just do it ourselves.
3022 if (bio->bio_flags & BIO_SYNC) {
3023 error = nfs_doio(ap->a_vp, nbio, td);
3025 nfs_asyncio(ap->a_vp, nbio);
3034 * NB Currently unsupported.
3036 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
3040 nfs_mmap(struct vop_mmap_args *ap)
3046 * fsync vnode op. Just call nfs_flush() with commit == 1.
3048 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3052 nfs_fsync(struct vop_fsync_args *ap)
3054 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
3058 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
3059 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
3060 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
3061 * set the buffer contains data that has already been written to the server
3062 * and which now needs a commit RPC.
3064 * If commit is 0 we only take one pass and only flush buffers containing new
3067 * If commit is 1 we take two passes, issuing a commit RPC in the second
3070 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3071 * to completely flush all pending data.
3073 * Note that the RB_SCAN code properly handles the case where the
3074 * callback might block and directly or indirectly (another thread) cause
3075 * the RB tree to change.
3078 #ifndef NFS_COMMITBVECSIZ
3079 #define NFS_COMMITBVECSIZ 16
3082 struct nfs_flush_info {
3083 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3090 struct buf *bvary[NFS_COMMITBVECSIZ];
3096 static int nfs_flush_bp(struct buf *bp, void *data);
3097 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3100 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3102 struct nfsnode *np = VTONFS(vp);
3103 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3104 struct nfs_flush_info info;
3108 bzero(&info, sizeof(info));
3111 info.waitfor = waitfor;
3112 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3114 lwkt_gettoken(&vlock, &vp->v_token);
3120 info.mode = NFI_FLUSHNEW;
3121 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3122 nfs_flush_bp, &info);
3125 * Take a second pass if committing and no error occured.
3126 * Clean up any left over collection (whether an error
3129 if (commit && error == 0) {
3130 info.mode = NFI_COMMIT;
3131 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3132 nfs_flush_bp, &info);
3134 error = nfs_flush_docommit(&info, error);
3138 * Wait for pending I/O to complete before checking whether
3139 * any further dirty buffers exist.
3141 while (waitfor == MNT_WAIT &&
3142 bio_track_active(&vp->v_track_write)) {
3143 error = bio_track_wait(&vp->v_track_write,
3144 info.slpflag, info.slptimeo);
3147 * We have to be able to break out if this
3148 * is an 'intr' mount.
3150 if (nfs_sigintr(nmp, NULL, td)) {
3156 * Since we do not process pending signals,
3157 * once we get a PCATCH our tsleep() will no
3158 * longer sleep, switch to a fixed timeout
3161 if (info.slpflag == PCATCH) {
3163 info.slptimeo = 2 * hz;
3170 * Loop if we are flushing synchronous as well as committing,
3171 * and dirty buffers are still present. Otherwise we might livelock.
3173 } while (waitfor == MNT_WAIT && commit &&
3174 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3177 * The callbacks have to return a negative error to terminate the
3184 * Deal with any error collection
3186 if (np->n_flag & NWRITEERR) {
3187 error = np->n_error;
3188 np->n_flag &= ~NWRITEERR;
3190 lwkt_reltoken(&vlock);
3196 nfs_flush_bp(struct buf *bp, void *data)
3198 struct nfs_flush_info *info = data;
3204 switch(info->mode) {
3206 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3207 if (error && info->loops && info->waitfor == MNT_WAIT) {
3208 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3210 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3211 if (info->slpflag & PCATCH)
3212 lkflags |= LK_PCATCH;
3213 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3219 * Ignore locking errors
3227 * The buffer may have changed out from under us, even if
3228 * we did not block (MPSAFE). Check again now that it is
3231 if (bp->b_vp == info->vp &&
3232 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3241 * Only process buffers in need of a commit which we can
3242 * immediately lock. This may prevent a buffer from being
3243 * committed, but the normal flush loop will block on the
3244 * same buffer so we shouldn't get into an endless loop.
3246 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3247 (B_DELWRI | B_NEEDCOMMIT)) {
3250 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3254 * We must recheck after successfully locking the buffer.
3256 if (bp->b_vp != info->vp ||
3257 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3258 (B_DELWRI | B_NEEDCOMMIT)) {
3264 * NOTE: storing the bp in the bvary[] basically sets
3265 * it up for a commit operation.
3267 * We must call vfs_busy_pages() now so the commit operation
3268 * is interlocked with user modifications to memory mapped
3269 * pages. The b_dirtyoff/b_dirtyend range is not correct
3270 * until after the pages have been busied.
3272 * Note: to avoid loopback deadlocks, we do not
3273 * assign b_runningbufspace.
3276 bp->b_cmd = BUF_CMD_WRITE;
3277 vfs_busy_pages(bp->b_vp, bp);
3278 info->bvary[info->bvsize] = bp;
3279 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3280 if (info->bvsize == 0 || toff < info->beg_off)
3281 info->beg_off = toff;
3282 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3283 if (info->bvsize == 0 || toff > info->end_off)
3284 info->end_off = toff;
3286 if (info->bvsize == NFS_COMMITBVECSIZ) {
3287 error = nfs_flush_docommit(info, 0);
3288 KKASSERT(info->bvsize == 0);
3296 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3306 if (info->bvsize > 0) {
3308 * Commit data on the server, as required. Note that
3309 * nfs_commit will use the vnode's cred for the commit.
3310 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3312 bytes = info->end_off - info->beg_off;
3313 if (bytes > 0x40000000)
3318 retv = nfs_commitrpc_uio(vp, info->beg_off,
3319 (int)bytes, info->td);
3320 if (retv == NFSERR_STALEWRITEVERF)
3321 nfs_clearcommit(vp->v_mount);
3325 * Now, either mark the blocks I/O done or mark the
3326 * blocks dirty, depending on whether the commit
3329 for (i = 0; i < info->bvsize; ++i) {
3330 bp = info->bvary[i];
3331 if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3333 * Either an error or the original
3334 * vfs_busy_pages() cleared B_NEEDCOMMIT
3335 * due to finding new dirty VM pages in
3338 * Leave B_DELWRI intact.
3340 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3341 vfs_unbusy_pages(bp);
3342 bp->b_cmd = BUF_CMD_DONE;
3346 * Success, remove B_DELWRI ( bundirty() ).
3348 * b_dirtyoff/b_dirtyend seem to be NFS
3349 * specific. We should probably move that
3350 * into bundirty(). XXX
3352 * We are faking an I/O write, we have to
3353 * start the transaction in order to
3354 * immediately biodone() it.
3357 bp->b_flags &= ~B_ERROR;
3358 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3359 bp->b_dirtyoff = bp->b_dirtyend = 0;
3360 biodone(&bp->b_bio1);
3369 * NFS advisory byte-level locks.
3370 * Currently unsupported.
3372 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3376 nfs_advlock(struct vop_advlock_args *ap)
3378 struct nfsnode *np = VTONFS(ap->a_vp);
3381 * The following kludge is to allow diskless support to work
3382 * until a real NFS lockd is implemented. Basically, just pretend
3383 * that this is a local lock.
3385 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3389 * Print out the contents of an nfsnode.
3391 * nfs_print(struct vnode *a_vp)
3394 nfs_print(struct vop_print_args *ap)
3396 struct vnode *vp = ap->a_vp;
3397 struct nfsnode *np = VTONFS(vp);
3399 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3400 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3401 if (vp->v_type == VFIFO)
3408 * nfs special file access vnode op.
3410 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3413 nfs_laccess(struct vop_access_args *ap)
3418 error = VOP_GETATTR(ap->a_vp, &vattr);
3420 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3426 * Read wrapper for fifos.
3428 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3429 * struct ucred *a_cred)
3432 nfsfifo_read(struct vop_read_args *ap)
3434 struct nfsnode *np = VTONFS(ap->a_vp);
3440 getnanotime(&np->n_atim);
3441 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3445 * Write wrapper for fifos.
3447 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3448 * struct ucred *a_cred)
3451 nfsfifo_write(struct vop_write_args *ap)
3453 struct nfsnode *np = VTONFS(ap->a_vp);
3459 getnanotime(&np->n_mtim);
3460 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3464 * Close wrapper for fifos.
3466 * Update the times on the nfsnode then do fifo close.
3468 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3471 nfsfifo_close(struct vop_close_args *ap)
3473 struct vnode *vp = ap->a_vp;
3474 struct nfsnode *np = VTONFS(vp);
3478 if (np->n_flag & (NACC | NUPD)) {
3480 if (np->n_flag & NACC)
3482 if (np->n_flag & NUPD)
3485 if (vp->v_sysref.refcnt == 1 &&
3486 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3488 if (np->n_flag & NACC)
3489 vattr.va_atime = np->n_atim;
3490 if (np->n_flag & NUPD)
3491 vattr.va_mtime = np->n_mtim;
3492 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3495 return (VOCALL(&fifo_vnode_vops, &ap->a_head));