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);
647 thread_t td = curthread;
648 struct nfsm_info info;
651 info.v3 = NFS_ISV3(vp);
654 * Update local times for special files.
656 if (np->n_flag & (NACC | NUPD))
659 * First look in the cache.
661 if (nfs_getattrcache(vp, ap->a_vap) == 0)
664 if (info.v3 && nfsaccess_cache_timeout > 0) {
665 nfsstats.accesscache_misses++;
666 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
667 if (nfs_getattrcache(vp, ap->a_vap) == 0)
671 nfsstats.rpccnt[NFSPROC_GETATTR]++;
672 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
673 ERROROUT(nfsm_fhtom(&info, vp));
674 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
675 nfs_vpcred(vp, ND_CHECK), &error));
677 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
688 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
691 nfs_setattr(struct vop_setattr_args *ap)
693 struct vnode *vp = ap->a_vp;
694 struct nfsnode *np = VTONFS(vp);
695 struct vattr *vap = ap->a_vap;
696 int biosize = vp->v_mount->mnt_stat.f_iosize;
700 thread_t td = curthread;
707 * Setting of flags is not supported.
709 if (vap->va_flags != VNOVAL)
713 * Disallow write attempts if the filesystem is mounted read-only.
715 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
716 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
717 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
718 (vp->v_mount->mnt_flag & MNT_RDONLY))
721 if (vap->va_size != VNOVAL) {
723 * truncation requested
725 switch (vp->v_type) {
732 if (vap->va_mtime.tv_sec == VNOVAL &&
733 vap->va_atime.tv_sec == VNOVAL &&
734 vap->va_mode == (mode_t)VNOVAL &&
735 vap->va_uid == (uid_t)VNOVAL &&
736 vap->va_gid == (gid_t)VNOVAL)
738 vap->va_size = VNOVAL;
742 * Disallow write attempts if the filesystem is
745 if (vp->v_mount->mnt_flag & MNT_RDONLY)
750 boff = (int)vap->va_size & (biosize - 1);
751 error = nfs_meta_setsize(vp, td, vap->va_size, 0);
754 if (np->n_flag & NLMODIFIED) {
755 if (vap->va_size == 0)
756 error = nfs_vinvalbuf(vp, 0, 1);
758 error = nfs_vinvalbuf(vp, V_SAVE, 1);
762 * note: this loop case almost always happens at
763 * least once per truncation.
765 if (error == 0 && np->n_size != vap->va_size)
767 np->n_vattr.va_size = vap->va_size;
770 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
772 * What to do. If we are modifying the mtime we lose
773 * mtime detection of changes made by the server or other
774 * clients. But programs like rsync/rdist/cpdup are going
775 * to call utimes a lot. We don't want to piecemeal sync.
777 * For now sync if any prior remote changes were detected,
778 * but allow us to lose track of remote changes made during
779 * the utimes operation.
781 if (np->n_flag & NRMODIFIED)
782 error = nfs_vinvalbuf(vp, V_SAVE, 1);
786 if (vap->va_mtime.tv_sec != VNOVAL) {
787 np->n_mtime = vap->va_mtime.tv_sec;
791 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
794 * Sanity check if a truncation was issued. This should only occur
795 * if multiple processes are racing on the same file.
797 if (error == 0 && vap->va_size != VNOVAL &&
798 np->n_size != vap->va_size) {
799 kprintf("NFS ftruncate: server disagrees on the file size: "
802 (intmax_t)vap->va_size,
803 (intmax_t)np->n_size);
806 if (error && vap->va_size != VNOVAL) {
807 np->n_size = np->n_vattr.va_size = tsize;
808 nfs_meta_setsize(vp, td, np->n_size, 0);
814 * Do an nfs setattr rpc.
817 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
818 struct ucred *cred, struct thread *td)
820 struct nfsv2_sattr *sp;
821 struct nfsnode *np = VTONFS(vp);
823 int error = 0, wccflag = NFSV3_WCCRATTR;
824 struct nfsm_info info;
827 info.v3 = NFS_ISV3(vp);
829 nfsstats.rpccnt[NFSPROC_SETATTR]++;
830 nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
831 NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
832 ERROROUT(nfsm_fhtom(&info, vp));
834 nfsm_v3attrbuild(&info, vap, TRUE);
835 tl = nfsm_build(&info, NFSX_UNSIGNED);
838 sp = nfsm_build(&info, NFSX_V2SATTR);
839 if (vap->va_mode == (mode_t)VNOVAL)
840 sp->sa_mode = nfs_xdrneg1;
842 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
843 if (vap->va_uid == (uid_t)VNOVAL)
844 sp->sa_uid = nfs_xdrneg1;
846 sp->sa_uid = txdr_unsigned(vap->va_uid);
847 if (vap->va_gid == (gid_t)VNOVAL)
848 sp->sa_gid = nfs_xdrneg1;
850 sp->sa_gid = txdr_unsigned(vap->va_gid);
851 sp->sa_size = txdr_unsigned(vap->va_size);
852 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
853 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
855 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
858 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
860 ERROROUT(nfsm_loadattr(&info, vp, NULL));
870 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
876 cache_setvp(nch, vp);
877 cache_settimeout(nch, nctimeout);
881 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
882 * nfs_lookup() until all remaining new api calls are implemented.
884 * Resolve a namecache entry. This function is passed a locked ncp and
885 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
888 nfs_nresolve(struct vop_nresolve_args *ap)
890 struct thread *td = curthread;
891 struct namecache *ncp;
902 struct nfsm_info info;
907 if ((error = vget(dvp, LK_SHARED)) != 0)
911 info.v3 = NFS_ISV3(dvp);
914 nfsstats.lookupcache_misses++;
915 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
916 ncp = ap->a_nch->ncp;
918 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
919 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
920 ERROROUT(nfsm_fhtom(&info, dvp));
921 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
922 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
923 ap->a_cred, &error));
926 * Cache negatve lookups to reduce NFS traffic, but use
927 * a fast timeout. Otherwise use a timeout of 1 tick.
928 * XXX we should add a namecache flag for no-caching
929 * to uncache the negative hit as soon as possible, but
930 * we cannot simply destroy the entry because it is used
931 * as a placeholder by the caller.
933 * The refactored nfs code will overwrite a non-zero error
934 * with 0 when we use ERROROUT(), so don't here.
937 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
938 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
950 * Success, get the file handle, do various checks, and load
951 * post-operation data from the reply packet. Theoretically
952 * we should never be looking up "." so, theoretically, we
953 * should never get the same file handle as our directory. But
954 * we check anyway. XXX
956 * Note that no timeout is set for the positive cache hit. We
957 * assume, theoretically, that ESTALE returns will be dealt with
958 * properly to handle NFS races and in anycase we cannot depend
959 * on a timeout to deal with NFS open/create/excl issues so instead
960 * of a bad hack here the rest of the NFS client code needs to do
963 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
966 if (NFS_CMPFH(np, fhp, fhsize)) {
970 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
980 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
981 NFS_LATTR_NOSHRINK));
982 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
983 NFS_LATTR_NOSHRINK));
985 ERROROUT(nfsm_loadattr(&info, nvp, NULL));
987 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
1002 * 'cached' nfs directory lookup
1004 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
1006 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
1007 * struct componentname *a_cnp)
1010 nfs_lookup(struct vop_old_lookup_args *ap)
1012 struct componentname *cnp = ap->a_cnp;
1013 struct vnode *dvp = ap->a_dvp;
1014 struct vnode **vpp = ap->a_vpp;
1015 int flags = cnp->cn_flags;
1016 struct vnode *newvp;
1017 struct nfsmount *nmp;
1021 int lockparent, wantparent, attrflag, fhsize;
1024 struct nfsm_info info;
1027 info.v3 = NFS_ISV3(dvp);
1031 * Read-only mount check and directory check.
1034 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1035 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1038 if (dvp->v_type != VDIR)
1042 * Look it up in the cache. Note that ENOENT is only returned if we
1043 * previously entered a negative hit (see later on). The additional
1044 * nfsneg_cache_timeout check causes previously cached results to
1045 * be instantly ignored if the negative caching is turned off.
1047 lockparent = flags & CNP_LOCKPARENT;
1048 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1049 nmp = VFSTONFS(dvp->v_mount);
1057 nfsstats.lookupcache_misses++;
1058 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1059 len = cnp->cn_namelen;
1060 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1061 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1062 ERROROUT(nfsm_fhtom(&info, dvp));
1063 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1064 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1065 cnp->cn_cred, &error));
1067 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1068 NFS_LATTR_NOSHRINK);
1078 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1081 * Handle RENAME case...
1083 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1084 if (NFS_CMPFH(np, fhp, fhsize)) {
1089 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1097 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1098 NFS_LATTR_NOSHRINK));
1099 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1100 NFS_LATTR_NOSHRINK));
1102 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1109 cnp->cn_flags |= CNP_PDIRUNLOCK;
1114 if (flags & CNP_ISDOTDOT) {
1116 cnp->cn_flags |= CNP_PDIRUNLOCK;
1117 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1119 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1120 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1121 return (error); /* NOTE: return error from nget */
1125 error = vn_lock(dvp, LK_EXCLUSIVE);
1130 cnp->cn_flags |= CNP_PDIRUNLOCK;
1132 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1136 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1144 cnp->cn_flags |= CNP_PDIRUNLOCK;
1149 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1150 NFS_LATTR_NOSHRINK));
1151 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1152 NFS_LATTR_NOSHRINK));
1154 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1157 /* XXX MOVE TO nfs_nremove() */
1158 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1159 cnp->cn_nameiop != NAMEI_DELETE) {
1160 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1168 if (newvp != NULLVP) {
1172 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1173 cnp->cn_nameiop == NAMEI_RENAME) &&
1177 cnp->cn_flags |= CNP_PDIRUNLOCK;
1179 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1182 error = EJUSTRETURN;
1190 * Just call nfs_bioread() to do the work.
1192 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1193 * struct ucred *a_cred)
1196 nfs_read(struct vop_read_args *ap)
1198 struct vnode *vp = ap->a_vp;
1200 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1206 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1209 nfs_readlink(struct vop_readlink_args *ap)
1211 struct vnode *vp = ap->a_vp;
1213 if (vp->v_type != VLNK)
1215 return (nfs_bioread(vp, ap->a_uio, 0));
1219 * Do a readlink rpc.
1220 * Called by nfs_doio() from below the buffer cache.
1223 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1225 int error = 0, len, attrflag;
1226 struct nfsm_info info;
1229 info.v3 = NFS_ISV3(vp);
1231 nfsstats.rpccnt[NFSPROC_READLINK]++;
1232 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1233 ERROROUT(nfsm_fhtom(&info, vp));
1234 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1235 nfs_vpcred(vp, ND_CHECK), &error));
1237 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1238 NFS_LATTR_NOSHRINK));
1241 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1242 if (len == NFS_MAXPATHLEN) {
1243 struct nfsnode *np = VTONFS(vp);
1244 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1247 ERROROUT(nfsm_mtouio(&info, uiop, len));
1256 * nfs synchronous read rpc using UIO
1259 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1262 struct nfsmount *nmp;
1263 int error = 0, len, retlen, tsiz, eof, attrflag;
1264 struct nfsm_info info;
1268 info.v3 = NFS_ISV3(vp);
1273 nmp = VFSTONFS(vp->v_mount);
1274 tsiz = uiop->uio_resid;
1275 tmp_off = uiop->uio_offset + tsiz;
1276 if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1278 tmp_off = uiop->uio_offset;
1280 nfsstats.rpccnt[NFSPROC_READ]++;
1281 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1282 nfsm_reqhead(&info, vp, NFSPROC_READ,
1283 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1284 ERROROUT(nfsm_fhtom(&info, vp));
1285 tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1287 txdr_hyper(uiop->uio_offset, tl);
1288 *(tl + 2) = txdr_unsigned(len);
1290 *tl++ = txdr_unsigned(uiop->uio_offset);
1291 *tl++ = txdr_unsigned(len);
1294 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1295 nfs_vpcred(vp, ND_READ), &error));
1297 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1298 NFS_LATTR_NOSHRINK));
1299 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1300 eof = fxdr_unsigned(int, *(tl + 1));
1302 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1304 NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1305 ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1310 * Handle short-read from server (NFSv3). If EOF is not
1311 * flagged (and no error occurred), but retlen is less
1312 * then the request size, we must zero-fill the remainder.
1314 if (retlen < len && info.v3 && eof == 0) {
1315 ERROROUT(uiomovez(len - retlen, uiop));
1321 * Terminate loop on EOF or zero-length read.
1323 * For NFSv2 a short-read indicates EOF, not zero-fill,
1324 * and also terminates the loop.
1327 if (eof || retlen == 0)
1329 } else if (retlen < len) {
1341 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1342 int *iomode, int *must_commit)
1346 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1347 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1348 int committed = NFSV3WRITE_FILESYNC;
1349 struct nfsm_info info;
1352 info.v3 = NFS_ISV3(vp);
1355 if (uiop->uio_iovcnt != 1)
1356 panic("nfs: writerpc iovcnt > 1");
1359 tsiz = uiop->uio_resid;
1360 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1363 nfsstats.rpccnt[NFSPROC_WRITE]++;
1364 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1365 nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1366 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1367 ERROROUT(nfsm_fhtom(&info, vp));
1369 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1370 txdr_hyper(uiop->uio_offset, tl);
1372 *tl++ = txdr_unsigned(len);
1373 *tl++ = txdr_unsigned(*iomode);
1374 *tl = txdr_unsigned(len);
1378 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1379 /* Set both "begin" and "current" to non-garbage. */
1380 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1381 *tl++ = x; /* "begin offset" */
1382 *tl++ = x; /* "current offset" */
1383 x = txdr_unsigned(len);
1384 *tl++ = x; /* total to this offset */
1385 *tl = x; /* size of this write */
1387 ERROROUT(nfsm_uiotom(&info, uiop, len));
1388 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1389 nfs_vpcred(vp, ND_WRITE), &error));
1392 * The write RPC returns a before and after mtime. The
1393 * nfsm_wcc_data() macro checks the before n_mtime
1394 * against the before time and stores the after time
1395 * in the nfsnode's cached vattr and n_mtime field.
1396 * The NRMODIFIED bit will be set if the before
1397 * time did not match the original mtime.
1399 wccflag = NFSV3_WCCCHK;
1400 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1402 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1403 rlen = fxdr_unsigned(int, *tl++);
1409 } else if (rlen < len) {
1410 backup = len - rlen;
1411 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1412 uiop->uio_iov->iov_len += backup;
1413 uiop->uio_offset -= backup;
1414 uiop->uio_resid += backup;
1417 commit = fxdr_unsigned(int, *tl++);
1420 * Return the lowest committment level
1421 * obtained by any of the RPCs.
1423 if (committed == NFSV3WRITE_FILESYNC)
1425 else if (committed == NFSV3WRITE_DATASYNC &&
1426 commit == NFSV3WRITE_UNSTABLE)
1428 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1429 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1431 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1432 } else if (bcmp((caddr_t)tl,
1433 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1435 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1440 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1449 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1450 committed = NFSV3WRITE_FILESYNC;
1451 *iomode = committed;
1453 uiop->uio_resid = tsiz;
1459 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1460 * mode set to specify the file type and the size field for rdev.
1463 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1466 struct nfsv2_sattr *sp;
1468 struct vnode *newvp = NULL;
1469 struct nfsnode *np = NULL;
1471 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1473 struct nfsm_info info;
1476 info.v3 = NFS_ISV3(dvp);
1478 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1479 rmajor = txdr_unsigned(vap->va_rmajor);
1480 rminor = txdr_unsigned(vap->va_rminor);
1481 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1482 rmajor = nfs_xdrneg1;
1483 rminor = nfs_xdrneg1;
1485 return (EOPNOTSUPP);
1487 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1490 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1491 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1492 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1493 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1494 ERROROUT(nfsm_fhtom(&info, dvp));
1495 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1498 tl = nfsm_build(&info, NFSX_UNSIGNED);
1499 *tl++ = vtonfsv3_type(vap->va_type);
1500 nfsm_v3attrbuild(&info, vap, FALSE);
1501 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1502 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1503 *tl++ = txdr_unsigned(vap->va_rmajor);
1504 *tl = txdr_unsigned(vap->va_rminor);
1507 sp = nfsm_build(&info, NFSX_V2SATTR);
1508 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1509 sp->sa_uid = nfs_xdrneg1;
1510 sp->sa_gid = nfs_xdrneg1;
1511 sp->sa_size = makeudev(rmajor, rminor);
1512 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1513 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1515 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1516 cnp->cn_cred, &error));
1518 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1524 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1525 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1531 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1542 VTONFS(dvp)->n_flag |= NLMODIFIED;
1544 VTONFS(dvp)->n_attrstamp = 0;
1550 * just call nfs_mknodrpc() to do the work.
1552 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1553 * struct componentname *a_cnp, struct vattr *a_vap)
1557 nfs_mknod(struct vop_old_mknod_args *ap)
1559 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1562 static u_long create_verf;
1564 * nfs file create call
1566 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1567 * struct componentname *a_cnp, struct vattr *a_vap)
1570 nfs_create(struct vop_old_create_args *ap)
1572 struct vnode *dvp = ap->a_dvp;
1573 struct vattr *vap = ap->a_vap;
1574 struct componentname *cnp = ap->a_cnp;
1575 struct nfsv2_sattr *sp;
1577 struct nfsnode *np = NULL;
1578 struct vnode *newvp = NULL;
1579 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1581 struct nfsm_info info;
1584 info.v3 = NFS_ISV3(dvp);
1587 * Oops, not for me..
1589 if (vap->va_type == VSOCK)
1590 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1592 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1595 if (vap->va_vaflags & VA_EXCLUSIVE)
1598 nfsstats.rpccnt[NFSPROC_CREATE]++;
1599 nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1600 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1601 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1602 ERROROUT(nfsm_fhtom(&info, dvp));
1603 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1606 tl = nfsm_build(&info, NFSX_UNSIGNED);
1607 if (fmode & O_EXCL) {
1608 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1609 tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1611 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1612 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1615 *tl++ = create_verf;
1616 *tl = ++create_verf;
1618 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1619 nfsm_v3attrbuild(&info, vap, FALSE);
1622 sp = nfsm_build(&info, NFSX_V2SATTR);
1623 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1624 sp->sa_uid = nfs_xdrneg1;
1625 sp->sa_gid = nfs_xdrneg1;
1627 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1628 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1630 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1631 cnp->cn_cred, &error));
1633 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1639 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1640 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1647 error = nfsm_wcc_data(&info, dvp, &wccflag);
1649 (void)nfsm_wcc_data(&info, dvp, &wccflag);
1655 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1656 KKASSERT(newvp == NULL);
1660 } else if (info.v3 && (fmode & O_EXCL)) {
1662 * We are normally called with only a partially initialized
1663 * VAP. Since the NFSv3 spec says that server may use the
1664 * file attributes to store the verifier, the spec requires
1665 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1666 * in atime, but we can't really assume that all servers will
1667 * so we ensure that our SETATTR sets both atime and mtime.
1669 if (vap->va_mtime.tv_sec == VNOVAL)
1670 vfs_timestamp(&vap->va_mtime);
1671 if (vap->va_atime.tv_sec == VNOVAL)
1672 vap->va_atime = vap->va_mtime;
1673 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1677 * The new np may have enough info for access
1678 * checks, make sure rucred and wucred are
1679 * initialized for read and write rpc's.
1682 if (np->n_rucred == NULL)
1683 np->n_rucred = crhold(cnp->cn_cred);
1684 if (np->n_wucred == NULL)
1685 np->n_wucred = crhold(cnp->cn_cred);
1690 VTONFS(dvp)->n_flag |= NLMODIFIED;
1692 VTONFS(dvp)->n_attrstamp = 0;
1697 * nfs file remove call
1698 * To try and make nfs semantics closer to ufs semantics, a file that has
1699 * other processes using the vnode is renamed instead of removed and then
1700 * removed later on the last close.
1701 * - If v_sysref.refcnt > 1
1702 * If a rename is not already in the works
1703 * call nfs_sillyrename() to set it up
1707 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1708 * struct componentname *a_cnp)
1711 nfs_remove(struct vop_old_remove_args *ap)
1713 struct vnode *vp = ap->a_vp;
1714 struct vnode *dvp = ap->a_dvp;
1715 struct componentname *cnp = ap->a_cnp;
1716 struct nfsnode *np = VTONFS(vp);
1721 if (vp->v_sysref.refcnt < 1)
1722 panic("nfs_remove: bad v_sysref.refcnt");
1724 if (vp->v_type == VDIR)
1726 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1727 VOP_GETATTR(vp, &vattr) == 0 &&
1728 vattr.va_nlink > 1)) {
1730 * throw away biocache buffers, mainly to avoid
1731 * unnecessary delayed writes later.
1733 error = nfs_vinvalbuf(vp, 0, 1);
1736 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1737 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1739 * Kludge City: If the first reply to the remove rpc is lost..
1740 * the reply to the retransmitted request will be ENOENT
1741 * since the file was in fact removed
1742 * Therefore, we cheat and return success.
1744 if (error == ENOENT)
1746 } else if (!np->n_sillyrename) {
1747 error = nfs_sillyrename(dvp, vp, cnp);
1749 np->n_attrstamp = 0;
1754 * nfs file remove rpc called from nfs_inactive
1757 nfs_removeit(struct sillyrename *sp)
1759 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1764 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1767 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1768 struct ucred *cred, struct thread *td)
1770 int error = 0, wccflag = NFSV3_WCCRATTR;
1771 struct nfsm_info info;
1774 info.v3 = NFS_ISV3(dvp);
1776 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1777 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1778 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1779 ERROROUT(nfsm_fhtom(&info, dvp));
1780 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1781 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1783 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1788 VTONFS(dvp)->n_flag |= NLMODIFIED;
1790 VTONFS(dvp)->n_attrstamp = 0;
1795 * nfs file rename call
1797 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1798 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1799 * struct vnode *a_tvp, struct componentname *a_tcnp)
1802 nfs_rename(struct vop_old_rename_args *ap)
1804 struct vnode *fvp = ap->a_fvp;
1805 struct vnode *tvp = ap->a_tvp;
1806 struct vnode *fdvp = ap->a_fdvp;
1807 struct vnode *tdvp = ap->a_tdvp;
1808 struct componentname *tcnp = ap->a_tcnp;
1809 struct componentname *fcnp = ap->a_fcnp;
1812 /* Check for cross-device rename */
1813 if ((fvp->v_mount != tdvp->v_mount) ||
1814 (tvp && (fvp->v_mount != tvp->v_mount))) {
1820 * We shouldn't have to flush fvp on rename for most server-side
1821 * filesystems as the file handle should not change. Unfortunately
1822 * the inode for some filesystems (msdosfs) might be tied to the
1823 * file name or directory position so to be completely safe
1824 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1827 * We must flush tvp on rename because it might become stale on the
1828 * server after the rename.
1830 if (nfs_flush_on_rename)
1831 VOP_FSYNC(fvp, MNT_WAIT, 0);
1833 VOP_FSYNC(tvp, MNT_WAIT, 0);
1836 * If the tvp exists and is in use, sillyrename it before doing the
1837 * rename of the new file over it.
1839 * XXX Can't sillyrename a directory.
1841 * We do not attempt to do any namecache purges in this old API
1842 * routine. The new API compat functions have access to the actual
1843 * namecache structures and will do it for us.
1845 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1846 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1853 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1854 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1867 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1869 if (error == ENOENT)
1875 * nfs file rename rpc called from nfs_remove() above
1878 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1879 struct sillyrename *sp)
1881 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1882 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1886 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1889 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1890 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1891 struct ucred *cred, struct thread *td)
1893 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1894 struct nfsm_info info;
1897 info.v3 = NFS_ISV3(fdvp);
1899 nfsstats.rpccnt[NFSPROC_RENAME]++;
1900 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1901 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1902 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1903 ERROROUT(nfsm_fhtom(&info, fdvp));
1904 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1905 ERROROUT(nfsm_fhtom(&info, tdvp));
1906 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1907 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
1909 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
1910 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
1915 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1916 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1918 VTONFS(fdvp)->n_attrstamp = 0;
1920 VTONFS(tdvp)->n_attrstamp = 0;
1925 * nfs hard link create call
1927 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1928 * struct componentname *a_cnp)
1931 nfs_link(struct vop_old_link_args *ap)
1933 struct vnode *vp = ap->a_vp;
1934 struct vnode *tdvp = ap->a_tdvp;
1935 struct componentname *cnp = ap->a_cnp;
1936 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1937 struct nfsm_info info;
1939 if (vp->v_mount != tdvp->v_mount) {
1944 * The attribute cache may get out of sync with the server on link.
1945 * Pushing writes to the server before handle was inherited from
1946 * long long ago and it is unclear if we still need to do this.
1949 if (nfs_flush_on_hlink)
1950 VOP_FSYNC(vp, MNT_WAIT, 0);
1953 info.v3 = NFS_ISV3(vp);
1955 nfsstats.rpccnt[NFSPROC_LINK]++;
1956 nfsm_reqhead(&info, vp, NFSPROC_LINK,
1957 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
1958 nfsm_rndup(cnp->cn_namelen));
1959 ERROROUT(nfsm_fhtom(&info, vp));
1960 ERROROUT(nfsm_fhtom(&info, tdvp));
1961 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1963 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
1964 cnp->cn_cred, &error));
1966 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1967 NFS_LATTR_NOSHRINK));
1968 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
1973 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1975 VTONFS(vp)->n_attrstamp = 0;
1977 VTONFS(tdvp)->n_attrstamp = 0;
1979 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1981 if (error == EEXIST)
1987 * nfs symbolic link create call
1989 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1990 * struct componentname *a_cnp, struct vattr *a_vap,
1994 nfs_symlink(struct vop_old_symlink_args *ap)
1996 struct vnode *dvp = ap->a_dvp;
1997 struct vattr *vap = ap->a_vap;
1998 struct componentname *cnp = ap->a_cnp;
1999 struct nfsv2_sattr *sp;
2000 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2001 struct vnode *newvp = NULL;
2002 struct nfsm_info info;
2005 info.v3 = NFS_ISV3(dvp);
2007 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2008 slen = strlen(ap->a_target);
2009 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
2010 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
2011 nfsm_rndup(cnp->cn_namelen) +
2012 nfsm_rndup(slen) + NFSX_SATTR(info.v3));
2013 ERROROUT(nfsm_fhtom(&info, dvp));
2014 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2017 nfsm_v3attrbuild(&info, vap, FALSE);
2019 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
2021 sp = nfsm_build(&info, NFSX_V2SATTR);
2022 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2023 sp->sa_uid = nfs_xdrneg1;
2024 sp->sa_gid = nfs_xdrneg1;
2025 sp->sa_size = nfs_xdrneg1;
2026 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2027 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2031 * Issue the NFS request and get the rpc response.
2033 * Only NFSv3 responses returning an error of 0 actually return
2034 * a file handle that can be converted into newvp without having
2035 * to do an extra lookup rpc.
2037 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
2038 cnp->cn_cred, &error));
2041 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2043 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2047 * out code jumps -> here, mrep is also freed.
2055 * If we get an EEXIST error, silently convert it to no-error
2056 * in case of an NFS retry.
2058 if (error == EEXIST)
2062 * If we do not have (or no longer have) an error, and we could
2063 * not extract the newvp from the response due to the request being
2064 * NFSv2 or the error being EEXIST. We have to do a lookup in order
2065 * to obtain a newvp to return.
2067 if (error == 0 && newvp == NULL) {
2068 struct nfsnode *np = NULL;
2070 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2071 cnp->cn_cred, cnp->cn_td, &np);
2081 VTONFS(dvp)->n_flag |= NLMODIFIED;
2083 VTONFS(dvp)->n_attrstamp = 0;
2090 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2091 * struct componentname *a_cnp, struct vattr *a_vap)
2094 nfs_mkdir(struct vop_old_mkdir_args *ap)
2096 struct vnode *dvp = ap->a_dvp;
2097 struct vattr *vap = ap->a_vap;
2098 struct componentname *cnp = ap->a_cnp;
2099 struct nfsv2_sattr *sp;
2100 struct nfsnode *np = NULL;
2101 struct vnode *newvp = NULL;
2103 int error = 0, wccflag = NFSV3_WCCRATTR;
2106 struct nfsm_info info;
2109 info.v3 = NFS_ISV3(dvp);
2111 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2114 len = cnp->cn_namelen;
2115 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2116 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2117 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2118 nfsm_rndup(len) + NFSX_SATTR(info.v3));
2119 ERROROUT(nfsm_fhtom(&info, dvp));
2120 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2122 nfsm_v3attrbuild(&info, vap, FALSE);
2124 sp = nfsm_build(&info, NFSX_V2SATTR);
2125 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2126 sp->sa_uid = nfs_xdrneg1;
2127 sp->sa_gid = nfs_xdrneg1;
2128 sp->sa_size = nfs_xdrneg1;
2129 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2130 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2132 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2133 cnp->cn_cred, &error));
2135 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2138 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2143 VTONFS(dvp)->n_flag |= NLMODIFIED;
2145 VTONFS(dvp)->n_attrstamp = 0;
2147 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2148 * if we can succeed in looking up the directory.
2150 if (error == EEXIST || (!error && !gotvp)) {
2155 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2159 if (newvp->v_type != VDIR)
2172 * nfs remove directory call
2174 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2175 * struct componentname *a_cnp)
2178 nfs_rmdir(struct vop_old_rmdir_args *ap)
2180 struct vnode *vp = ap->a_vp;
2181 struct vnode *dvp = ap->a_dvp;
2182 struct componentname *cnp = ap->a_cnp;
2183 int error = 0, wccflag = NFSV3_WCCRATTR;
2184 struct nfsm_info info;
2187 info.v3 = NFS_ISV3(dvp);
2191 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2192 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2193 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2194 nfsm_rndup(cnp->cn_namelen));
2195 ERROROUT(nfsm_fhtom(&info, dvp));
2196 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2198 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2199 cnp->cn_cred, &error));
2201 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2206 VTONFS(dvp)->n_flag |= NLMODIFIED;
2208 VTONFS(dvp)->n_attrstamp = 0;
2210 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2212 if (error == ENOENT)
2220 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2223 nfs_readdir(struct vop_readdir_args *ap)
2225 struct vnode *vp = ap->a_vp;
2226 struct nfsnode *np = VTONFS(vp);
2227 struct uio *uio = ap->a_uio;
2231 if (vp->v_type != VDIR)
2234 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2238 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2239 * and then check that is still valid, or if this is an NQNFS mount
2240 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2241 * VOP_GETATTR() does not necessarily go to the wire.
2243 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2244 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2245 if (VOP_GETATTR(vp, &vattr) == 0 &&
2246 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2248 nfsstats.direofcache_hits++;
2254 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2255 * own cache coherency checks so we do not have to.
2257 tresid = uio->uio_resid;
2258 error = nfs_bioread(vp, uio, 0);
2260 if (!error && uio->uio_resid == tresid)
2261 nfsstats.direofcache_misses++;
2268 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2270 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2271 * offset/block and converts the nfs formatted directory entries for userland
2272 * consumption as well as deals with offsets into the middle of blocks.
2273 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2274 * be block-bounded. It must convert to cookies for the actual RPC.
2277 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2280 struct nfs_dirent *dp = NULL;
2285 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2286 struct nfsnode *dnp = VTONFS(vp);
2288 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2290 struct nfsm_info info;
2293 info.v3 = NFS_ISV3(vp);
2296 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2297 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2298 panic("nfs readdirrpc bad uio");
2302 * If there is no cookie, assume directory was stale.
2304 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2308 return (NFSERR_BAD_COOKIE);
2310 * Loop around doing readdir rpc's of size nm_readdirsize
2311 * truncated to a multiple of DIRBLKSIZ.
2312 * The stopping criteria is EOF or buffer full.
2314 while (more_dirs && bigenough) {
2315 nfsstats.rpccnt[NFSPROC_READDIR]++;
2316 nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2317 NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2318 ERROROUT(nfsm_fhtom(&info, vp));
2320 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2321 *tl++ = cookie.nfsuquad[0];
2322 *tl++ = cookie.nfsuquad[1];
2323 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2324 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2326 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2327 *tl++ = cookie.nfsuquad[0];
2329 *tl = txdr_unsigned(nmp->nm_readdirsize);
2330 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2332 nfs_vpcred(vp, ND_READ), &error));
2334 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2335 NFS_LATTR_NOSHRINK));
2336 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2337 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2338 dnp->n_cookieverf.nfsuquad[1] = *tl;
2340 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2341 more_dirs = fxdr_unsigned(int, *tl);
2343 /* loop thru the dir entries, converting them to std form */
2344 while (more_dirs && bigenough) {
2346 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2347 fileno = fxdr_hyper(tl);
2348 len = fxdr_unsigned(int, *(tl + 2));
2350 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2351 fileno = fxdr_unsigned(u_quad_t, *tl++);
2352 len = fxdr_unsigned(int, *tl);
2354 if (len <= 0 || len > NFS_MAXNAMLEN) {
2362 * len is the number of bytes in the path element
2363 * name, not including the \0 termination.
2365 * tlen is the number of bytes w have to reserve for
2366 * the path element name.
2368 tlen = nfsm_rndup(len);
2370 tlen += 4; /* To ensure null termination */
2373 * If the entry would cross a DIRBLKSIZ boundary,
2374 * extend the previous nfs_dirent to cover the
2377 left = DIRBLKSIZ - blksiz;
2378 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2379 dp->nfs_reclen += left;
2380 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2381 uiop->uio_iov->iov_len -= left;
2382 uiop->uio_offset += left;
2383 uiop->uio_resid -= left;
2386 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2389 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2390 dp->nfs_ino = fileno;
2391 dp->nfs_namlen = len;
2392 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2393 dp->nfs_type = DT_UNKNOWN;
2394 blksiz += dp->nfs_reclen;
2395 if (blksiz == DIRBLKSIZ)
2397 uiop->uio_offset += sizeof(struct nfs_dirent);
2398 uiop->uio_resid -= sizeof(struct nfs_dirent);
2399 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2400 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2401 ERROROUT(nfsm_mtouio(&info, uiop, len));
2404 * The uiop has advanced by nfs_dirent + len
2405 * but really needs to advance by
2408 cp = uiop->uio_iov->iov_base;
2410 *cp = '\0'; /* null terminate */
2411 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2412 uiop->uio_iov->iov_len -= tlen;
2413 uiop->uio_offset += tlen;
2414 uiop->uio_resid -= tlen;
2417 * NFS strings must be rounded up (nfsm_myouio
2418 * handled that in the bigenough case).
2420 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2423 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2425 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2429 * If we were able to accomodate the last entry,
2430 * get the cookie for the next one. Otherwise
2431 * hold-over the cookie for the one we were not
2432 * able to accomodate.
2435 cookie.nfsuquad[0] = *tl++;
2437 cookie.nfsuquad[1] = *tl++;
2438 } else if (info.v3) {
2443 more_dirs = fxdr_unsigned(int, *tl);
2446 * If at end of rpc data, get the eof boolean
2449 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2450 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2456 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2457 * by increasing d_reclen for the last record.
2460 left = DIRBLKSIZ - blksiz;
2461 dp->nfs_reclen += left;
2462 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2463 uiop->uio_iov->iov_len -= left;
2464 uiop->uio_offset += left;
2465 uiop->uio_resid -= left;
2470 * We hit the end of the directory, update direofoffset.
2472 dnp->n_direofoffset = uiop->uio_offset;
2475 * There is more to go, insert the link cookie so the
2476 * next block can be read.
2478 if (uiop->uio_resid > 0)
2479 kprintf("EEK! readdirrpc resid > 0\n");
2480 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2488 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2491 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2494 struct nfs_dirent *dp;
2496 struct vnode *newvp;
2498 caddr_t dpossav1, dpossav2;
2500 struct mbuf *mdsav1, *mdsav2;
2502 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2503 struct nfsnode *dnp = VTONFS(vp), *np;
2506 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2507 int attrflag, fhsize;
2508 struct nchandle nch;
2509 struct nchandle dnch;
2510 struct nlcomponent nlc;
2511 struct nfsm_info info;
2520 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2521 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2522 panic("nfs readdirplusrpc bad uio");
2525 * Obtain the namecache record for the directory so we have something
2526 * to use as a basis for creating the entries. This function will
2527 * return a held (but not locked) ncp. The ncp may be disconnected
2528 * from the tree and cannot be used for upward traversals, and the
2529 * ncp may be unnamed. Note that other unrelated operations may
2530 * cause the ncp to be named at any time.
2532 * We have to lock the ncp to prevent a lock order reversal when
2533 * rdirplus does nlookups of the children, because the vnode is
2534 * locked and has to stay that way.
2536 cache_fromdvp(vp, NULL, 0, &dnch);
2537 bzero(&nlc, sizeof(nlc));
2541 * If there is no cookie, assume directory was stale.
2543 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2549 return (NFSERR_BAD_COOKIE);
2553 * Loop around doing readdir rpc's of size nm_readdirsize
2554 * truncated to a multiple of DIRBLKSIZ.
2555 * The stopping criteria is EOF or buffer full.
2557 while (more_dirs && bigenough) {
2558 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2559 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2560 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2561 ERROROUT(nfsm_fhtom(&info, vp));
2562 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2563 *tl++ = cookie.nfsuquad[0];
2564 *tl++ = cookie.nfsuquad[1];
2565 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2566 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2567 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2568 *tl = txdr_unsigned(nmp->nm_rsize);
2569 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2571 nfs_vpcred(vp, ND_READ), &error));
2572 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2573 NFS_LATTR_NOSHRINK));
2574 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2575 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2576 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2577 more_dirs = fxdr_unsigned(int, *tl);
2579 /* loop thru the dir entries, doctoring them to 4bsd form */
2580 while (more_dirs && bigenough) {
2581 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2582 fileno = fxdr_hyper(tl);
2583 len = fxdr_unsigned(int, *(tl + 2));
2584 if (len <= 0 || len > NFS_MAXNAMLEN) {
2590 tlen = nfsm_rndup(len);
2592 tlen += 4; /* To ensure null termination*/
2593 left = DIRBLKSIZ - blksiz;
2594 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2595 dp->nfs_reclen += left;
2596 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2597 uiop->uio_iov->iov_len -= left;
2598 uiop->uio_offset += left;
2599 uiop->uio_resid -= left;
2602 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2605 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2606 dp->nfs_ino = fileno;
2607 dp->nfs_namlen = len;
2608 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2609 dp->nfs_type = DT_UNKNOWN;
2610 blksiz += dp->nfs_reclen;
2611 if (blksiz == DIRBLKSIZ)
2613 uiop->uio_offset += sizeof(struct nfs_dirent);
2614 uiop->uio_resid -= sizeof(struct nfs_dirent);
2615 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2616 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2617 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2618 nlc.nlc_namelen = len;
2619 ERROROUT(nfsm_mtouio(&info, uiop, len));
2620 cp = uiop->uio_iov->iov_base;
2623 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2624 uiop->uio_iov->iov_len -= tlen;
2625 uiop->uio_offset += tlen;
2626 uiop->uio_resid -= tlen;
2628 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2630 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2632 cookie.nfsuquad[0] = *tl++;
2633 cookie.nfsuquad[1] = *tl++;
2639 * Since the attributes are before the file handle
2640 * (sigh), we must skip over the attributes and then
2641 * come back and get them.
2643 attrflag = fxdr_unsigned(int, *tl);
2645 dpossav1 = info.dpos;
2647 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2648 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2649 doit = fxdr_unsigned(int, *tl);
2651 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2653 if (doit && bigenough && !nlcdegenerate(&nlc) &&
2654 !NFS_CMPFH(dnp, fhp, fhsize)
2658 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2659 nlc.nlc_namelen, nlc.nlc_namelen,
2663 * This is a bit hokey but there isn't
2664 * much we can do about it. We can't
2665 * hold the directory vp locked while
2666 * doing lookups and gets.
2668 nch = cache_nlookup_nonblock(&dnch, &nlc);
2669 if (nch.ncp == NULL)
2671 cache_setunresolved(&nch);
2672 error = nfs_nget_nonblock(vp->v_mount, fhp,
2679 dpossav2 = info.dpos;
2680 info.dpos = dpossav1;
2683 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2684 info.dpos = dpossav2;
2687 IFTODT(VTTOIF(np->n_vattr.va_type));
2688 nfs_cache_setvp(&nch, newvp,
2689 nfspos_cache_timeout);
2695 kprintf("Warning: NFS/rddirplus, "
2696 "UNABLE TO ENTER %*.*s\n",
2697 nlc.nlc_namelen, nlc.nlc_namelen,
2702 /* Just skip over the file handle */
2703 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2704 i = fxdr_unsigned(int, *tl);
2705 ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2707 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2708 more_dirs = fxdr_unsigned(int, *tl);
2711 * If at end of rpc data, get the eof boolean
2714 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2715 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2721 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2722 * by increasing d_reclen for the last record.
2725 left = DIRBLKSIZ - blksiz;
2726 dp->nfs_reclen += left;
2727 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2728 uiop->uio_iov->iov_len -= left;
2729 uiop->uio_offset += left;
2730 uiop->uio_resid -= left;
2734 * We are now either at the end of the directory or have filled the
2738 dnp->n_direofoffset = uiop->uio_offset;
2740 if (uiop->uio_resid > 0)
2741 kprintf("EEK! readdirplusrpc resid > 0\n");
2742 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2746 if (newvp != NULLVP) {
2759 * Silly rename. To make the NFS filesystem that is stateless look a little
2760 * more like the "ufs" a remove of an active vnode is translated to a rename
2761 * to a funny looking filename that is removed by nfs_inactive on the
2762 * nfsnode. There is the potential for another process on a different client
2763 * to create the same funny name between the nfs_lookitup() fails and the
2764 * nfs_rename() completes, but...
2767 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2769 struct sillyrename *sp;
2774 * We previously purged dvp instead of vp. I don't know why, it
2775 * completely destroys performance. We can't do it anyway with the
2776 * new VFS API since we would be breaking the namecache topology.
2778 cache_purge(vp); /* XXX */
2781 if (vp->v_type == VDIR)
2782 panic("nfs: sillyrename dir");
2784 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2785 M_NFSREQ, M_WAITOK);
2786 sp->s_cred = crdup(cnp->cn_cred);
2790 /* Fudge together a funny name */
2791 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2792 (int)(intptr_t)cnp->cn_td);
2794 /* Try lookitups until we get one that isn't there */
2795 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2796 cnp->cn_td, NULL) == 0) {
2798 if (sp->s_name[4] > 'z') {
2803 error = nfs_renameit(dvp, cnp, sp);
2806 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2808 np->n_sillyrename = sp;
2813 kfree((caddr_t)sp, M_NFSREQ);
2818 * Look up a file name and optionally either update the file handle or
2819 * allocate an nfsnode, depending on the value of npp.
2820 * npp == NULL --> just do the lookup
2821 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2823 * *npp != NULL --> update the file handle in the vnode
2826 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2827 struct thread *td, struct nfsnode **npp)
2829 struct vnode *newvp = NULL;
2830 struct nfsnode *np, *dnp = VTONFS(dvp);
2831 int error = 0, fhlen, attrflag;
2833 struct nfsm_info info;
2836 info.v3 = NFS_ISV3(dvp);
2838 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2839 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2840 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2841 ERROROUT(nfsm_fhtom(&info, dvp));
2842 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2843 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2844 if (npp && !error) {
2845 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2848 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2849 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2850 np->n_fhp = &np->n_fh;
2851 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2852 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2853 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2854 np->n_fhsize = fhlen;
2856 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2860 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2869 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2870 NFS_LATTR_NOSHRINK));
2871 if (!attrflag && *npp == NULL) {
2881 ERROROUT(error = nfsm_loadattr(&info, newvp, NULL));
2887 if (npp && *npp == NULL) {
2902 * Nfs Version 3 commit rpc
2904 * We call it 'uio' to distinguish it from 'bio' but there is no real uio
2908 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2910 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2911 int error = 0, wccflag = NFSV3_WCCRATTR;
2912 struct nfsm_info info;
2918 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2920 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2921 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
2922 ERROROUT(nfsm_fhtom(&info, vp));
2923 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
2924 txdr_hyper(offset, tl);
2926 *tl = txdr_unsigned(cnt);
2927 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
2928 nfs_vpcred(vp, ND_WRITE), &error));
2929 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
2931 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
2932 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2933 NFSX_V3WRITEVERF)) {
2934 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2936 error = NFSERR_STALEWRITEVERF;
2947 * - make nfs_bmap() essentially a no-op that does no translation
2948 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2949 * (Maybe I could use the process's page mapping, but I was concerned that
2950 * Kernel Write might not be enabled and also figured copyout() would do
2951 * a lot more work than bcopy() and also it currently happens in the
2952 * context of the swapper process (2).
2954 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
2955 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2958 nfs_bmap(struct vop_bmap_args *ap)
2960 if (ap->a_doffsetp != NULL)
2961 *ap->a_doffsetp = ap->a_loffset;
2962 if (ap->a_runp != NULL)
2964 if (ap->a_runb != NULL)
2973 nfs_strategy(struct vop_strategy_args *ap)
2975 struct bio *bio = ap->a_bio;
2977 struct buf *bp __debugvar = bio->bio_buf;
2981 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2982 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
2983 KASSERT(BUF_REFCNT(bp) > 0,
2984 ("nfs_strategy: buffer %p not locked", bp));
2986 if (bio->bio_flags & BIO_SYNC)
2987 td = curthread; /* XXX */
2992 * We probably don't need to push an nbio any more since no
2993 * block conversion is required due to the use of 64 bit byte
2994 * offsets, but do it anyway.
2996 * NOTE: When NFS callers itself via this strategy routines and
2997 * sets up a synchronous I/O, it expects the I/O to run
2998 * synchronously (its bio_done routine just assumes it),
2999 * so for now we have to honor the bit.
3001 nbio = push_bio(bio);
3002 nbio->bio_offset = bio->bio_offset;
3003 nbio->bio_flags = bio->bio_flags & BIO_SYNC;
3006 * If the op is asynchronous and an i/o daemon is waiting
3007 * queue the request, wake it up and wait for completion
3008 * otherwise just do it ourselves.
3010 if (bio->bio_flags & BIO_SYNC) {
3011 error = nfs_doio(ap->a_vp, nbio, td);
3013 nfs_asyncio(ap->a_vp, nbio);
3022 * NB Currently unsupported.
3024 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
3028 nfs_mmap(struct vop_mmap_args *ap)
3034 * fsync vnode op. Just call nfs_flush() with commit == 1.
3036 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3040 nfs_fsync(struct vop_fsync_args *ap)
3042 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
3046 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
3047 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
3048 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
3049 * set the buffer contains data that has already been written to the server
3050 * and which now needs a commit RPC.
3052 * If commit is 0 we only take one pass and only flush buffers containing new
3055 * If commit is 1 we take two passes, issuing a commit RPC in the second
3058 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3059 * to completely flush all pending data.
3061 * Note that the RB_SCAN code properly handles the case where the
3062 * callback might block and directly or indirectly (another thread) cause
3063 * the RB tree to change.
3066 #ifndef NFS_COMMITBVECSIZ
3067 #define NFS_COMMITBVECSIZ 16
3070 struct nfs_flush_info {
3071 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3078 struct buf *bvary[NFS_COMMITBVECSIZ];
3084 static int nfs_flush_bp(struct buf *bp, void *data);
3085 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3088 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3090 struct nfsnode *np = VTONFS(vp);
3091 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3092 struct nfs_flush_info info;
3096 bzero(&info, sizeof(info));
3099 info.waitfor = waitfor;
3100 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3102 lwkt_gettoken(&vlock, &vp->v_token);
3108 info.mode = NFI_FLUSHNEW;
3109 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3110 nfs_flush_bp, &info);
3113 * Take a second pass if committing and no error occured.
3114 * Clean up any left over collection (whether an error
3117 if (commit && error == 0) {
3118 info.mode = NFI_COMMIT;
3119 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3120 nfs_flush_bp, &info);
3122 error = nfs_flush_docommit(&info, error);
3126 * Wait for pending I/O to complete before checking whether
3127 * any further dirty buffers exist.
3129 while (waitfor == MNT_WAIT &&
3130 bio_track_active(&vp->v_track_write)) {
3131 error = bio_track_wait(&vp->v_track_write,
3132 info.slpflag, info.slptimeo);
3135 * We have to be able to break out if this
3136 * is an 'intr' mount.
3138 if (nfs_sigintr(nmp, NULL, td)) {
3144 * Since we do not process pending signals,
3145 * once we get a PCATCH our tsleep() will no
3146 * longer sleep, switch to a fixed timeout
3149 if (info.slpflag == PCATCH) {
3151 info.slptimeo = 2 * hz;
3158 * Loop if we are flushing synchronous as well as committing,
3159 * and dirty buffers are still present. Otherwise we might livelock.
3161 } while (waitfor == MNT_WAIT && commit &&
3162 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3165 * The callbacks have to return a negative error to terminate the
3172 * Deal with any error collection
3174 if (np->n_flag & NWRITEERR) {
3175 error = np->n_error;
3176 np->n_flag &= ~NWRITEERR;
3178 lwkt_reltoken(&vlock);
3184 nfs_flush_bp(struct buf *bp, void *data)
3186 struct nfs_flush_info *info = data;
3192 switch(info->mode) {
3194 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3195 if (error && info->loops && info->waitfor == MNT_WAIT) {
3196 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3198 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3199 if (info->slpflag & PCATCH)
3200 lkflags |= LK_PCATCH;
3201 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3207 * Ignore locking errors
3215 * The buffer may have changed out from under us, even if
3216 * we did not block (MPSAFE). Check again now that it is
3219 if (bp->b_vp == info->vp &&
3220 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3229 * Only process buffers in need of a commit which we can
3230 * immediately lock. This may prevent a buffer from being
3231 * committed, but the normal flush loop will block on the
3232 * same buffer so we shouldn't get into an endless loop.
3234 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3235 (B_DELWRI | B_NEEDCOMMIT)) {
3238 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3242 * We must recheck after successfully locking the buffer.
3244 if (bp->b_vp != info->vp ||
3245 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3246 (B_DELWRI | B_NEEDCOMMIT)) {
3252 * NOTE: storing the bp in the bvary[] basically sets
3253 * it up for a commit operation.
3255 * We must call vfs_busy_pages() now so the commit operation
3256 * is interlocked with user modifications to memory mapped
3257 * pages. The b_dirtyoff/b_dirtyend range is not correct
3258 * until after the pages have been busied.
3260 * Note: to avoid loopback deadlocks, we do not
3261 * assign b_runningbufspace.
3264 bp->b_cmd = BUF_CMD_WRITE;
3265 vfs_busy_pages(bp->b_vp, bp);
3266 info->bvary[info->bvsize] = bp;
3267 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3268 if (info->bvsize == 0 || toff < info->beg_off)
3269 info->beg_off = toff;
3270 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3271 if (info->bvsize == 0 || toff > info->end_off)
3272 info->end_off = toff;
3274 if (info->bvsize == NFS_COMMITBVECSIZ) {
3275 error = nfs_flush_docommit(info, 0);
3276 KKASSERT(info->bvsize == 0);
3284 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3294 if (info->bvsize > 0) {
3296 * Commit data on the server, as required. Note that
3297 * nfs_commit will use the vnode's cred for the commit.
3298 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3300 bytes = info->end_off - info->beg_off;
3301 if (bytes > 0x40000000)
3306 retv = nfs_commitrpc_uio(vp, info->beg_off,
3307 (int)bytes, info->td);
3308 if (retv == NFSERR_STALEWRITEVERF)
3309 nfs_clearcommit(vp->v_mount);
3313 * Now, either mark the blocks I/O done or mark the
3314 * blocks dirty, depending on whether the commit
3317 for (i = 0; i < info->bvsize; ++i) {
3318 bp = info->bvary[i];
3319 if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3321 * Either an error or the original
3322 * vfs_busy_pages() cleared B_NEEDCOMMIT
3323 * due to finding new dirty VM pages in
3326 * Leave B_DELWRI intact.
3328 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3329 vfs_unbusy_pages(bp);
3330 bp->b_cmd = BUF_CMD_DONE;
3334 * Success, remove B_DELWRI ( bundirty() ).
3336 * b_dirtyoff/b_dirtyend seem to be NFS
3337 * specific. We should probably move that
3338 * into bundirty(). XXX
3340 * We are faking an I/O write, we have to
3341 * start the transaction in order to
3342 * immediately biodone() it.
3345 bp->b_flags &= ~B_ERROR;
3346 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3347 bp->b_dirtyoff = bp->b_dirtyend = 0;
3348 biodone(&bp->b_bio1);
3357 * NFS advisory byte-level locks.
3358 * Currently unsupported.
3360 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3364 nfs_advlock(struct vop_advlock_args *ap)
3366 struct nfsnode *np = VTONFS(ap->a_vp);
3369 * The following kludge is to allow diskless support to work
3370 * until a real NFS lockd is implemented. Basically, just pretend
3371 * that this is a local lock.
3373 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3377 * Print out the contents of an nfsnode.
3379 * nfs_print(struct vnode *a_vp)
3382 nfs_print(struct vop_print_args *ap)
3384 struct vnode *vp = ap->a_vp;
3385 struct nfsnode *np = VTONFS(vp);
3387 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3388 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3389 if (vp->v_type == VFIFO)
3396 * nfs special file access vnode op.
3398 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3401 nfs_laccess(struct vop_access_args *ap)
3406 error = VOP_GETATTR(ap->a_vp, &vattr);
3408 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3414 * Read wrapper for fifos.
3416 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3417 * struct ucred *a_cred)
3420 nfsfifo_read(struct vop_read_args *ap)
3422 struct nfsnode *np = VTONFS(ap->a_vp);
3428 getnanotime(&np->n_atim);
3429 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3433 * Write wrapper for fifos.
3435 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3436 * struct ucred *a_cred)
3439 nfsfifo_write(struct vop_write_args *ap)
3441 struct nfsnode *np = VTONFS(ap->a_vp);
3447 getnanotime(&np->n_mtim);
3448 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3452 * Close wrapper for fifos.
3454 * Update the times on the nfsnode then do fifo close.
3456 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3459 nfsfifo_close(struct vop_close_args *ap)
3461 struct vnode *vp = ap->a_vp;
3462 struct nfsnode *np = VTONFS(vp);
3466 if (np->n_flag & (NACC | NUPD)) {
3468 if (np->n_flag & NACC)
3470 if (np->n_flag & NUPD)
3473 if (vp->v_sysref.refcnt == 1 &&
3474 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3476 if (np->n_flag & NACC)
3477 vattr.va_atime = np->n_atim;
3478 if (np->n_flag & NUPD)
3479 vattr.va_mtime = np->n_mtim;
3480 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3483 return (VOCALL(&fifo_vnode_vops, &ap->a_head));