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 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
101 static int nfs_lookup (struct vop_old_lookup_args *);
102 static int nfs_create (struct vop_old_create_args *);
103 static int nfs_mknod (struct vop_old_mknod_args *);
104 static int nfs_open (struct vop_open_args *);
105 static int nfs_close (struct vop_close_args *);
106 static int nfs_access (struct vop_access_args *);
107 static int nfs_getattr (struct vop_getattr_args *);
108 static int nfs_setattr (struct vop_setattr_args *);
109 static int nfs_read (struct vop_read_args *);
110 static int nfs_mmap (struct vop_mmap_args *);
111 static int nfs_fsync (struct vop_fsync_args *);
112 static int nfs_remove (struct vop_old_remove_args *);
113 static int nfs_link (struct vop_old_link_args *);
114 static int nfs_rename (struct vop_old_rename_args *);
115 static int nfs_mkdir (struct vop_old_mkdir_args *);
116 static int nfs_rmdir (struct vop_old_rmdir_args *);
117 static int nfs_symlink (struct vop_old_symlink_args *);
118 static int nfs_readdir (struct vop_readdir_args *);
119 static int nfs_bmap (struct vop_bmap_args *);
120 static int nfs_strategy (struct vop_strategy_args *);
121 static int nfs_lookitup (struct vnode *, const char *, int,
122 struct ucred *, struct thread *, struct nfsnode **);
123 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
124 static int nfs_laccess (struct vop_access_args *);
125 static int nfs_readlink (struct vop_readlink_args *);
126 static int nfs_print (struct vop_print_args *);
127 static int nfs_advlock (struct vop_advlock_args *);
129 static int nfs_nresolve (struct vop_nresolve_args *);
131 * Global vfs data structures for nfs
133 struct vop_ops nfsv2_vnode_vops = {
134 .vop_default = vop_defaultop,
135 .vop_access = nfs_access,
136 .vop_advlock = nfs_advlock,
137 .vop_bmap = nfs_bmap,
138 .vop_close = nfs_close,
139 .vop_old_create = nfs_create,
140 .vop_fsync = nfs_fsync,
141 .vop_getattr = nfs_getattr,
142 .vop_getpages = vop_stdgetpages,
143 .vop_putpages = vop_stdputpages,
144 .vop_inactive = nfs_inactive,
145 .vop_old_link = nfs_link,
146 .vop_old_lookup = nfs_lookup,
147 .vop_old_mkdir = nfs_mkdir,
148 .vop_old_mknod = nfs_mknod,
149 .vop_mmap = nfs_mmap,
150 .vop_open = nfs_open,
151 .vop_print = nfs_print,
152 .vop_read = nfs_read,
153 .vop_readdir = nfs_readdir,
154 .vop_readlink = nfs_readlink,
155 .vop_reclaim = nfs_reclaim,
156 .vop_old_remove = nfs_remove,
157 .vop_old_rename = nfs_rename,
158 .vop_old_rmdir = nfs_rmdir,
159 .vop_setattr = nfs_setattr,
160 .vop_strategy = nfs_strategy,
161 .vop_old_symlink = nfs_symlink,
162 .vop_write = nfs_write,
163 .vop_nresolve = nfs_nresolve
167 * Special device vnode ops
169 struct vop_ops nfsv2_spec_vops = {
170 .vop_default = vop_defaultop,
171 .vop_access = nfs_laccess,
172 .vop_close = nfs_close,
173 .vop_fsync = nfs_fsync,
174 .vop_getattr = nfs_getattr,
175 .vop_inactive = nfs_inactive,
176 .vop_print = nfs_print,
177 .vop_read = vop_stdnoread,
178 .vop_reclaim = nfs_reclaim,
179 .vop_setattr = nfs_setattr,
180 .vop_write = vop_stdnowrite
183 struct vop_ops nfsv2_fifo_vops = {
184 .vop_default = fifo_vnoperate,
185 .vop_access = nfs_laccess,
186 .vop_close = nfsfifo_close,
187 .vop_fsync = nfs_fsync,
188 .vop_getattr = nfs_getattr,
189 .vop_inactive = nfs_inactive,
190 .vop_print = nfs_print,
191 .vop_read = nfsfifo_read,
192 .vop_reclaim = nfs_reclaim,
193 .vop_setattr = nfs_setattr,
194 .vop_write = nfsfifo_write
197 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
198 struct componentname *cnp,
200 static int nfs_removerpc (struct vnode *dvp, const char *name,
202 struct ucred *cred, struct thread *td);
203 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
204 int fnamelen, struct vnode *tdvp,
205 const char *tnameptr, int tnamelen,
206 struct ucred *cred, struct thread *td);
207 static int nfs_renameit (struct vnode *sdvp,
208 struct componentname *scnp,
209 struct sillyrename *sp);
211 SYSCTL_DECL(_vfs_nfs);
213 static int nfs_flush_on_rename = 1;
214 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
215 &nfs_flush_on_rename, 0, "flush fvp prior to rename");
216 static int nfs_flush_on_hlink = 0;
217 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
218 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
220 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
221 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
222 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
224 static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
225 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
226 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
228 static int nfspos_cache_timeout = NFS_MINATTRTIMO;
229 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
230 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
232 static int nfsv3_commit_on_close = 0;
233 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
234 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
236 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
237 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
239 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
240 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
243 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
244 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
245 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
248 * Returns whether a name component is a degenerate '.' or '..'.
252 nlcdegenerate(struct nlcomponent *nlc)
254 if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.')
256 if (nlc->nlc_namelen == 2 &&
257 nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.')
263 nfs3_access_otw(struct vnode *vp, int wmode,
264 struct thread *td, struct ucred *cred)
266 struct nfsnode *np = VTONFS(vp);
271 struct nfsm_info info;
276 nfsstats.rpccnt[NFSPROC_ACCESS]++;
277 nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
278 NFSX_FH(info.v3) + NFSX_UNSIGNED);
279 ERROROUT(nfsm_fhtom(&info, vp));
280 tl = nfsm_build(&info, NFSX_UNSIGNED);
281 *tl = txdr_unsigned(wmode);
282 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
283 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
285 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
286 rmode = fxdr_unsigned(u_int32_t, *tl);
288 np->n_modeuid = cred->cr_uid;
289 np->n_modestamp = mycpu->gd_time_seconds;
298 * nfs access vnode op.
299 * For nfs version 2, just return ok. File accesses may fail later.
300 * For nfs version 3, use the access rpc to check accessibility. If file modes
301 * are changed on the server, accesses might still fail later.
303 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
306 nfs_access(struct vop_access_args *ap)
309 struct vnode *vp = ap->a_vp;
310 thread_t td = curthread;
312 u_int32_t mode, wmode;
313 struct nfsnode *np = VTONFS(vp);
314 int v3 = NFS_ISV3(vp);
317 * Disallow write attempts on filesystems mounted read-only;
318 * unless the file is a socket, fifo, or a block or character
319 * device resident on the filesystem.
321 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
322 switch (vp->v_type) {
333 * The NFS protocol passes only the effective uid/gid over the wire but
334 * we need to check access against real ids if AT_EACCESS not set.
335 * Handle this case by cloning the credentials and setting the
336 * effective ids to the real ones.
338 if (ap->a_flags & AT_EACCESS) {
339 cred = crhold(ap->a_cred);
341 cred = crdup(ap->a_cred);
342 cred->cr_uid = cred->cr_ruid;
343 cred->cr_gid = cred->cr_rgid;
347 * For nfs v3, check to see if we have done this recently, and if
348 * so return our cached result instead of making an ACCESS call.
349 * If not, do an access rpc, otherwise you are stuck emulating
350 * ufs_access() locally using the vattr. This may not be correct,
351 * since the server may apply other access criteria such as
352 * client uid-->server uid mapping that we do not know about.
355 if (ap->a_mode & VREAD)
356 mode = NFSV3ACCESS_READ;
359 if (vp->v_type != VDIR) {
360 if (ap->a_mode & VWRITE)
361 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
362 if (ap->a_mode & VEXEC)
363 mode |= NFSV3ACCESS_EXECUTE;
365 if (ap->a_mode & VWRITE)
366 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
368 if (ap->a_mode & VEXEC)
369 mode |= NFSV3ACCESS_LOOKUP;
371 /* XXX safety belt, only make blanket request if caching */
372 if (nfsaccess_cache_timeout > 0) {
373 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
374 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
375 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
381 * Does our cached result allow us to give a definite yes to
384 if (np->n_modestamp &&
385 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
386 (cred->cr_uid == np->n_modeuid) &&
387 ((np->n_mode & mode) == mode)) {
388 nfsstats.accesscache_hits++;
391 * Either a no, or a don't know. Go to the wire.
393 nfsstats.accesscache_misses++;
394 error = nfs3_access_otw(vp, wmode, td, cred);
396 if ((np->n_mode & mode) != mode) {
402 if ((error = nfs_laccess(ap)) != 0) {
408 * Attempt to prevent a mapped root from accessing a file
409 * which it shouldn't. We try to read a byte from the file
410 * if the user is root and the file is not zero length.
411 * After calling nfs_laccess, we should have the correct
414 if (cred->cr_uid == 0 && (ap->a_mode & VREAD)
415 && VTONFS(vp)->n_size > 0) {
422 auio.uio_iov = &aiov;
426 auio.uio_segflg = UIO_SYSSPACE;
427 auio.uio_rw = UIO_READ;
430 if (vp->v_type == VREG) {
431 error = nfs_readrpc_uio(vp, &auio);
432 } else if (vp->v_type == VDIR) {
434 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
436 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
437 error = nfs_readdirrpc_uio(vp, &auio);
439 } else if (vp->v_type == VLNK) {
440 error = nfs_readlinkrpc_uio(vp, &auio);
447 * [re]record creds for reading and/or writing if access
448 * was granted. Assume the NFS server will grant read access
449 * for execute requests.
452 if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) {
455 crfree(np->n_rucred);
458 if ((ap->a_mode & VWRITE) && cred != np->n_wucred) {
461 crfree(np->n_wucred);
471 * Check to see if the type is ok
472 * and that deletion is not in progress.
473 * For paged in text files, you will need to flush the page cache
474 * if consistency is lost.
476 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
481 nfs_open(struct vop_open_args *ap)
483 struct vnode *vp = ap->a_vp;
484 struct nfsnode *np = VTONFS(vp);
488 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
490 kprintf("open eacces vtyp=%d\n",vp->v_type);
496 * Save valid creds for reading and writing for later RPCs.
498 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
501 crfree(np->n_rucred);
502 np->n_rucred = ap->a_cred;
504 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
507 crfree(np->n_wucred);
508 np->n_wucred = ap->a_cred;
512 * Clear the attribute cache only if opening with write access. It
513 * is unclear if we should do this at all here, but we certainly
514 * should not clear the cache unconditionally simply because a file
517 if (ap->a_mode & FWRITE)
521 * For normal NFS, reconcile changes made locally verses
522 * changes made remotely. Note that VOP_GETATTR only goes
523 * to the wire if the cached attribute has timed out or been
526 * If local modifications have been made clear the attribute
527 * cache to force an attribute and modified time check. If
528 * GETATTR detects that the file has been changed by someone
529 * other then us it will set NRMODIFIED.
531 * If we are opening a directory and local changes have been
532 * made we have to invalidate the cache in order to ensure
533 * that we get the most up-to-date information from the
536 if (np->n_flag & NLMODIFIED) {
538 if (vp->v_type == VDIR) {
539 error = nfs_vinvalbuf(vp, V_SAVE, 1);
545 error = VOP_GETATTR(vp, &vattr);
548 if (np->n_flag & NRMODIFIED) {
549 if (vp->v_type == VDIR)
551 error = nfs_vinvalbuf(vp, V_SAVE, 1);
554 np->n_flag &= ~NRMODIFIED;
557 return (vop_stdopen(ap));
562 * What an NFS client should do upon close after writing is a debatable issue.
563 * Most NFS clients push delayed writes to the server upon close, basically for
565 * 1 - So that any write errors may be reported back to the client process
566 * doing the close system call. By far the two most likely errors are
567 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
568 * 2 - To put a worst case upper bound on cache inconsistency between
569 * multiple clients for the file.
570 * There is also a consistency problem for Version 2 of the protocol w.r.t.
571 * not being able to tell if other clients are writing a file concurrently,
572 * since there is no way of knowing if the changed modify time in the reply
573 * is only due to the write for this client.
574 * (NFS Version 3 provides weak cache consistency data in the reply that
575 * should be sufficient to detect and handle this case.)
577 * The current code does the following:
578 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
579 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
580 * or commit them (this satisfies 1 and 2 except for the
581 * case where the server crashes after this close but
582 * before the commit RPC, which is felt to be "good
583 * enough". Changing the last argument to nfs_flush() to
584 * a 1 would force a commit operation, if it is felt a
585 * commit is necessary now.
586 * for NQNFS - do nothing now, since 2 is dealt with via leases and
587 * 1 should be dealt with via an fsync() system call for
588 * cases where write errors are important.
590 * nfs_close(struct vnode *a_vp, int a_fflag)
594 nfs_close(struct vop_close_args *ap)
596 struct vnode *vp = ap->a_vp;
597 struct nfsnode *np = VTONFS(vp);
599 thread_t td = curthread;
601 if (vp->v_type == VREG) {
602 if (np->n_flag & NLMODIFIED) {
605 * Under NFSv3 we have dirty buffers to dispose of. We
606 * must flush them to the NFS server. We have the option
607 * of waiting all the way through the commit rpc or just
608 * waiting for the initial write. The default is to only
609 * wait through the initial write so the data is in the
610 * server's cache, which is roughly similar to the state
611 * a standard disk subsystem leaves the file in on close().
613 * We cannot clear the NLMODIFIED bit in np->n_flag due to
614 * potential races with other processes, and certainly
615 * cannot clear it if we don't commit.
617 int cm = nfsv3_commit_on_close ? 1 : 0;
618 error = nfs_flush(vp, MNT_WAIT, td, cm);
619 /* np->n_flag &= ~NLMODIFIED; */
621 error = nfs_vinvalbuf(vp, V_SAVE, 1);
625 if (np->n_flag & NWRITEERR) {
626 np->n_flag &= ~NWRITEERR;
635 * nfs getattr call from vfs.
637 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
640 nfs_getattr(struct vop_getattr_args *ap)
642 struct vnode *vp = ap->a_vp;
643 struct nfsnode *np = VTONFS(vp);
644 struct nfsmount *nmp;
646 thread_t td = curthread;
647 struct nfsm_info info;
650 info.v3 = NFS_ISV3(vp);
651 nmp = VFSTONFS(vp->v_mount);
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));
683 * NFS doesn't support chflags flags. If the nfs mount was
684 * made -o cache set the UF_CACHE bit for swapcache.
686 if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT))
687 ap->a_vap->va_flags |= UF_CACHE;
695 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
698 nfs_setattr(struct vop_setattr_args *ap)
700 struct vnode *vp = ap->a_vp;
701 struct nfsnode *np = VTONFS(vp);
702 struct vattr *vap = ap->a_vap;
703 int biosize = vp->v_mount->mnt_stat.f_iosize;
707 thread_t td = curthread;
714 * Setting of flags is not supported.
716 if (vap->va_flags != VNOVAL)
720 * Disallow write attempts if the filesystem is mounted read-only.
722 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
723 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
724 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
725 (vp->v_mount->mnt_flag & MNT_RDONLY))
728 if (vap->va_size != VNOVAL) {
730 * truncation requested
732 switch (vp->v_type) {
739 if (vap->va_mtime.tv_sec == VNOVAL &&
740 vap->va_atime.tv_sec == VNOVAL &&
741 vap->va_mode == (mode_t)VNOVAL &&
742 vap->va_uid == (uid_t)VNOVAL &&
743 vap->va_gid == (gid_t)VNOVAL)
745 vap->va_size = VNOVAL;
749 * Disallow write attempts if the filesystem is
752 if (vp->v_mount->mnt_flag & MNT_RDONLY)
757 boff = (int)vap->va_size & (biosize - 1);
758 error = nfs_meta_setsize(vp, td, vap->va_size, 0);
761 if (np->n_flag & NLMODIFIED) {
762 if (vap->va_size == 0)
763 error = nfs_vinvalbuf(vp, 0, 1);
765 error = nfs_vinvalbuf(vp, V_SAVE, 1);
769 * note: this loop case almost always happens at
770 * least once per truncation.
772 if (error == 0 && np->n_size != vap->va_size)
774 np->n_vattr.va_size = vap->va_size;
777 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
779 * What to do. If we are modifying the mtime we lose
780 * mtime detection of changes made by the server or other
781 * clients. But programs like rsync/rdist/cpdup are going
782 * to call utimes a lot. We don't want to piecemeal sync.
784 * For now sync if any prior remote changes were detected,
785 * but allow us to lose track of remote changes made during
786 * the utimes operation.
788 if (np->n_flag & NRMODIFIED)
789 error = nfs_vinvalbuf(vp, V_SAVE, 1);
793 if (vap->va_mtime.tv_sec != VNOVAL) {
794 np->n_mtime = vap->va_mtime.tv_sec;
798 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
801 * Sanity check if a truncation was issued. This should only occur
802 * if multiple processes are racing on the same file.
804 if (error == 0 && vap->va_size != VNOVAL &&
805 np->n_size != vap->va_size) {
806 kprintf("NFS ftruncate: server disagrees on the file size: "
809 (intmax_t)vap->va_size,
810 (intmax_t)np->n_size);
813 if (error && vap->va_size != VNOVAL) {
814 np->n_size = np->n_vattr.va_size = tsize;
815 nfs_meta_setsize(vp, td, np->n_size, 0);
821 * Do an nfs setattr rpc.
824 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
825 struct ucred *cred, struct thread *td)
827 struct nfsv2_sattr *sp;
828 struct nfsnode *np = VTONFS(vp);
830 int error = 0, wccflag = NFSV3_WCCRATTR;
831 struct nfsm_info info;
834 info.v3 = NFS_ISV3(vp);
836 nfsstats.rpccnt[NFSPROC_SETATTR]++;
837 nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
838 NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
839 ERROROUT(nfsm_fhtom(&info, vp));
841 nfsm_v3attrbuild(&info, vap, TRUE);
842 tl = nfsm_build(&info, NFSX_UNSIGNED);
845 sp = nfsm_build(&info, NFSX_V2SATTR);
846 if (vap->va_mode == (mode_t)VNOVAL)
847 sp->sa_mode = nfs_xdrneg1;
849 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
850 if (vap->va_uid == (uid_t)VNOVAL)
851 sp->sa_uid = nfs_xdrneg1;
853 sp->sa_uid = txdr_unsigned(vap->va_uid);
854 if (vap->va_gid == (gid_t)VNOVAL)
855 sp->sa_gid = nfs_xdrneg1;
857 sp->sa_gid = txdr_unsigned(vap->va_gid);
858 sp->sa_size = txdr_unsigned(vap->va_size);
859 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
860 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
862 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
865 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
867 ERROROUT(nfsm_loadattr(&info, vp, NULL));
877 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
883 cache_setvp(nch, vp);
884 cache_settimeout(nch, nctimeout);
888 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
889 * nfs_lookup() until all remaining new api calls are implemented.
891 * Resolve a namecache entry. This function is passed a locked ncp and
892 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
895 nfs_nresolve(struct vop_nresolve_args *ap)
897 struct thread *td = curthread;
898 struct namecache *ncp;
909 struct nfsm_info info;
914 if ((error = vget(dvp, LK_SHARED)) != 0)
918 info.v3 = NFS_ISV3(dvp);
921 nfsstats.lookupcache_misses++;
922 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
923 ncp = ap->a_nch->ncp;
925 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
926 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
927 ERROROUT(nfsm_fhtom(&info, dvp));
928 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
929 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
930 ap->a_cred, &error));
933 * Cache negatve lookups to reduce NFS traffic, but use
934 * a fast timeout. Otherwise use a timeout of 1 tick.
935 * XXX we should add a namecache flag for no-caching
936 * to uncache the negative hit as soon as possible, but
937 * we cannot simply destroy the entry because it is used
938 * as a placeholder by the caller.
940 * The refactored nfs code will overwrite a non-zero error
941 * with 0 when we use ERROROUT(), so don't here.
944 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
945 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
957 * Success, get the file handle, do various checks, and load
958 * post-operation data from the reply packet. Theoretically
959 * we should never be looking up "." so, theoretically, we
960 * should never get the same file handle as our directory. But
961 * we check anyway. XXX
963 * Note that no timeout is set for the positive cache hit. We
964 * assume, theoretically, that ESTALE returns will be dealt with
965 * properly to handle NFS races and in anycase we cannot depend
966 * on a timeout to deal with NFS open/create/excl issues so instead
967 * of a bad hack here the rest of the NFS client code needs to do
970 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
973 if (NFS_CMPFH(np, fhp, fhsize)) {
977 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
987 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
988 NFS_LATTR_NOSHRINK));
989 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
990 NFS_LATTR_NOSHRINK));
992 ERROROUT(nfsm_loadattr(&info, nvp, NULL));
994 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
1009 * 'cached' nfs directory lookup
1011 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
1013 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
1014 * struct componentname *a_cnp)
1017 nfs_lookup(struct vop_old_lookup_args *ap)
1019 struct componentname *cnp = ap->a_cnp;
1020 struct vnode *dvp = ap->a_dvp;
1021 struct vnode **vpp = ap->a_vpp;
1022 int flags = cnp->cn_flags;
1023 struct vnode *newvp;
1024 struct nfsmount *nmp;
1028 int lockparent, wantparent, attrflag, fhsize;
1031 struct nfsm_info info;
1034 info.v3 = NFS_ISV3(dvp);
1038 * Read-only mount check and directory check.
1041 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1042 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1045 if (dvp->v_type != VDIR)
1049 * Look it up in the cache. Note that ENOENT is only returned if we
1050 * previously entered a negative hit (see later on). The additional
1051 * nfsneg_cache_timeout check causes previously cached results to
1052 * be instantly ignored if the negative caching is turned off.
1054 lockparent = flags & CNP_LOCKPARENT;
1055 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1056 nmp = VFSTONFS(dvp->v_mount);
1064 nfsstats.lookupcache_misses++;
1065 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1066 len = cnp->cn_namelen;
1067 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1068 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1069 ERROROUT(nfsm_fhtom(&info, dvp));
1070 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1071 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1072 cnp->cn_cred, &error));
1074 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1075 NFS_LATTR_NOSHRINK);
1085 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1088 * Handle RENAME case...
1090 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1091 if (NFS_CMPFH(np, fhp, fhsize)) {
1096 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1104 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1105 NFS_LATTR_NOSHRINK));
1106 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1107 NFS_LATTR_NOSHRINK));
1109 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1116 cnp->cn_flags |= CNP_PDIRUNLOCK;
1121 if (flags & CNP_ISDOTDOT) {
1123 cnp->cn_flags |= CNP_PDIRUNLOCK;
1124 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1126 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1127 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1128 return (error); /* NOTE: return error from nget */
1132 error = vn_lock(dvp, LK_EXCLUSIVE);
1137 cnp->cn_flags |= CNP_PDIRUNLOCK;
1139 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1143 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1151 cnp->cn_flags |= CNP_PDIRUNLOCK;
1156 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1157 NFS_LATTR_NOSHRINK));
1158 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1159 NFS_LATTR_NOSHRINK));
1161 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1164 /* XXX MOVE TO nfs_nremove() */
1165 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1166 cnp->cn_nameiop != NAMEI_DELETE) {
1167 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1175 if (newvp != NULLVP) {
1179 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1180 cnp->cn_nameiop == NAMEI_RENAME) &&
1184 cnp->cn_flags |= CNP_PDIRUNLOCK;
1186 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1189 error = EJUSTRETURN;
1197 * Just call nfs_bioread() to do the work.
1199 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1200 * struct ucred *a_cred)
1203 nfs_read(struct vop_read_args *ap)
1205 struct vnode *vp = ap->a_vp;
1207 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1213 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1216 nfs_readlink(struct vop_readlink_args *ap)
1218 struct vnode *vp = ap->a_vp;
1220 if (vp->v_type != VLNK)
1222 return (nfs_bioread(vp, ap->a_uio, 0));
1226 * Do a readlink rpc.
1227 * Called by nfs_doio() from below the buffer cache.
1230 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1232 int error = 0, len, attrflag;
1233 struct nfsm_info info;
1236 info.v3 = NFS_ISV3(vp);
1238 nfsstats.rpccnt[NFSPROC_READLINK]++;
1239 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1240 ERROROUT(nfsm_fhtom(&info, vp));
1241 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1242 nfs_vpcred(vp, ND_CHECK), &error));
1244 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1245 NFS_LATTR_NOSHRINK));
1248 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1249 if (len == NFS_MAXPATHLEN) {
1250 struct nfsnode *np = VTONFS(vp);
1251 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1254 ERROROUT(nfsm_mtouio(&info, uiop, len));
1263 * nfs synchronous read rpc using UIO
1266 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1269 struct nfsmount *nmp;
1270 int error = 0, len, retlen, tsiz, eof, attrflag;
1271 struct nfsm_info info;
1275 info.v3 = NFS_ISV3(vp);
1280 nmp = VFSTONFS(vp->v_mount);
1281 tsiz = uiop->uio_resid;
1282 tmp_off = uiop->uio_offset + tsiz;
1283 if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1285 tmp_off = uiop->uio_offset;
1287 nfsstats.rpccnt[NFSPROC_READ]++;
1288 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1289 nfsm_reqhead(&info, vp, NFSPROC_READ,
1290 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1291 ERROROUT(nfsm_fhtom(&info, vp));
1292 tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1294 txdr_hyper(uiop->uio_offset, tl);
1295 *(tl + 2) = txdr_unsigned(len);
1297 *tl++ = txdr_unsigned(uiop->uio_offset);
1298 *tl++ = txdr_unsigned(len);
1301 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1302 nfs_vpcred(vp, ND_READ), &error));
1304 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1305 NFS_LATTR_NOSHRINK));
1306 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1307 eof = fxdr_unsigned(int, *(tl + 1));
1309 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1311 NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1312 ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1317 * Handle short-read from server (NFSv3). If EOF is not
1318 * flagged (and no error occurred), but retlen is less
1319 * then the request size, we must zero-fill the remainder.
1321 if (retlen < len && info.v3 && eof == 0) {
1322 ERROROUT(uiomovez(len - retlen, uiop));
1328 * Terminate loop on EOF or zero-length read.
1330 * For NFSv2 a short-read indicates EOF, not zero-fill,
1331 * and also terminates the loop.
1334 if (eof || retlen == 0)
1336 } else if (retlen < len) {
1348 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1349 int *iomode, int *must_commit)
1353 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1354 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1355 int committed = NFSV3WRITE_FILESYNC;
1356 struct nfsm_info info;
1359 info.v3 = NFS_ISV3(vp);
1362 if (uiop->uio_iovcnt != 1)
1363 panic("nfs: writerpc iovcnt > 1");
1366 tsiz = uiop->uio_resid;
1367 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1370 nfsstats.rpccnt[NFSPROC_WRITE]++;
1371 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1372 nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1373 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1374 ERROROUT(nfsm_fhtom(&info, vp));
1376 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1377 txdr_hyper(uiop->uio_offset, tl);
1379 *tl++ = txdr_unsigned(len);
1380 *tl++ = txdr_unsigned(*iomode);
1381 *tl = txdr_unsigned(len);
1385 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1386 /* Set both "begin" and "current" to non-garbage. */
1387 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1388 *tl++ = x; /* "begin offset" */
1389 *tl++ = x; /* "current offset" */
1390 x = txdr_unsigned(len);
1391 *tl++ = x; /* total to this offset */
1392 *tl = x; /* size of this write */
1394 ERROROUT(nfsm_uiotom(&info, uiop, len));
1395 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1396 nfs_vpcred(vp, ND_WRITE), &error));
1399 * The write RPC returns a before and after mtime. The
1400 * nfsm_wcc_data() macro checks the before n_mtime
1401 * against the before time and stores the after time
1402 * in the nfsnode's cached vattr and n_mtime field.
1403 * The NRMODIFIED bit will be set if the before
1404 * time did not match the original mtime.
1406 wccflag = NFSV3_WCCCHK;
1407 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1409 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1410 rlen = fxdr_unsigned(int, *tl++);
1416 } else if (rlen < len) {
1417 backup = len - rlen;
1418 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1419 uiop->uio_iov->iov_len += backup;
1420 uiop->uio_offset -= backup;
1421 uiop->uio_resid += backup;
1424 commit = fxdr_unsigned(int, *tl++);
1427 * Return the lowest committment level
1428 * obtained by any of the RPCs.
1430 if (committed == NFSV3WRITE_FILESYNC)
1432 else if (committed == NFSV3WRITE_DATASYNC &&
1433 commit == NFSV3WRITE_UNSTABLE)
1435 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1436 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1438 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1439 } else if (bcmp((caddr_t)tl,
1440 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1442 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1447 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1456 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1457 committed = NFSV3WRITE_FILESYNC;
1458 *iomode = committed;
1460 uiop->uio_resid = tsiz;
1466 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1467 * mode set to specify the file type and the size field for rdev.
1470 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1473 struct nfsv2_sattr *sp;
1475 struct vnode *newvp = NULL;
1476 struct nfsnode *np = NULL;
1478 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1480 struct nfsm_info info;
1483 info.v3 = NFS_ISV3(dvp);
1485 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1486 rmajor = txdr_unsigned(vap->va_rmajor);
1487 rminor = txdr_unsigned(vap->va_rminor);
1488 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1489 rmajor = nfs_xdrneg1;
1490 rminor = nfs_xdrneg1;
1492 return (EOPNOTSUPP);
1494 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1497 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1498 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1499 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1500 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1501 ERROROUT(nfsm_fhtom(&info, dvp));
1502 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1505 tl = nfsm_build(&info, NFSX_UNSIGNED);
1506 *tl++ = vtonfsv3_type(vap->va_type);
1507 nfsm_v3attrbuild(&info, vap, FALSE);
1508 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1509 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1510 *tl++ = txdr_unsigned(vap->va_rmajor);
1511 *tl = txdr_unsigned(vap->va_rminor);
1514 sp = nfsm_build(&info, NFSX_V2SATTR);
1515 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1516 sp->sa_uid = nfs_xdrneg1;
1517 sp->sa_gid = nfs_xdrneg1;
1518 sp->sa_size = makeudev(rmajor, rminor);
1519 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1520 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1522 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1523 cnp->cn_cred, &error));
1525 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1531 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1532 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1538 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1549 VTONFS(dvp)->n_flag |= NLMODIFIED;
1551 VTONFS(dvp)->n_attrstamp = 0;
1557 * just call nfs_mknodrpc() to do the work.
1559 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1560 * struct componentname *a_cnp, struct vattr *a_vap)
1564 nfs_mknod(struct vop_old_mknod_args *ap)
1566 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1569 static u_long create_verf;
1571 * nfs file create call
1573 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1574 * struct componentname *a_cnp, struct vattr *a_vap)
1577 nfs_create(struct vop_old_create_args *ap)
1579 struct vnode *dvp = ap->a_dvp;
1580 struct vattr *vap = ap->a_vap;
1581 struct componentname *cnp = ap->a_cnp;
1582 struct nfsv2_sattr *sp;
1584 struct nfsnode *np = NULL;
1585 struct vnode *newvp = NULL;
1586 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1588 struct nfsm_info info;
1591 info.v3 = NFS_ISV3(dvp);
1594 * Oops, not for me..
1596 if (vap->va_type == VSOCK)
1597 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1599 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1602 if (vap->va_vaflags & VA_EXCLUSIVE)
1605 nfsstats.rpccnt[NFSPROC_CREATE]++;
1606 nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1607 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1608 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1609 ERROROUT(nfsm_fhtom(&info, dvp));
1610 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1613 tl = nfsm_build(&info, NFSX_UNSIGNED);
1614 if (fmode & O_EXCL) {
1615 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1616 tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1618 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1619 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1622 *tl++ = create_verf;
1623 *tl = ++create_verf;
1625 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1626 nfsm_v3attrbuild(&info, vap, FALSE);
1629 sp = nfsm_build(&info, NFSX_V2SATTR);
1630 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1631 sp->sa_uid = nfs_xdrneg1;
1632 sp->sa_gid = nfs_xdrneg1;
1634 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1635 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1637 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1638 cnp->cn_cred, &error));
1640 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1646 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1647 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1654 error = nfsm_wcc_data(&info, dvp, &wccflag);
1656 (void)nfsm_wcc_data(&info, dvp, &wccflag);
1662 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1663 KKASSERT(newvp == NULL);
1667 } else if (info.v3 && (fmode & O_EXCL)) {
1669 * We are normally called with only a partially initialized
1670 * VAP. Since the NFSv3 spec says that server may use the
1671 * file attributes to store the verifier, the spec requires
1672 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1673 * in atime, but we can't really assume that all servers will
1674 * so we ensure that our SETATTR sets both atime and mtime.
1676 if (vap->va_mtime.tv_sec == VNOVAL)
1677 vfs_timestamp(&vap->va_mtime);
1678 if (vap->va_atime.tv_sec == VNOVAL)
1679 vap->va_atime = vap->va_mtime;
1680 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1684 * The new np may have enough info for access
1685 * checks, make sure rucred and wucred are
1686 * initialized for read and write rpc's.
1689 if (np->n_rucred == NULL)
1690 np->n_rucred = crhold(cnp->cn_cred);
1691 if (np->n_wucred == NULL)
1692 np->n_wucred = crhold(cnp->cn_cred);
1697 VTONFS(dvp)->n_flag |= NLMODIFIED;
1699 VTONFS(dvp)->n_attrstamp = 0;
1704 * nfs file remove call
1705 * To try and make nfs semantics closer to ufs semantics, a file that has
1706 * other processes using the vnode is renamed instead of removed and then
1707 * removed later on the last close.
1708 * - If v_sysref.refcnt > 1
1709 * If a rename is not already in the works
1710 * call nfs_sillyrename() to set it up
1714 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1715 * struct componentname *a_cnp)
1718 nfs_remove(struct vop_old_remove_args *ap)
1720 struct vnode *vp = ap->a_vp;
1721 struct vnode *dvp = ap->a_dvp;
1722 struct componentname *cnp = ap->a_cnp;
1723 struct nfsnode *np = VTONFS(vp);
1728 if (vp->v_sysref.refcnt < 1)
1729 panic("nfs_remove: bad v_sysref.refcnt");
1731 if (vp->v_type == VDIR)
1733 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1734 VOP_GETATTR(vp, &vattr) == 0 &&
1735 vattr.va_nlink > 1)) {
1737 * throw away biocache buffers, mainly to avoid
1738 * unnecessary delayed writes later.
1740 error = nfs_vinvalbuf(vp, 0, 1);
1743 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1744 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1746 * Kludge City: If the first reply to the remove rpc is lost..
1747 * the reply to the retransmitted request will be ENOENT
1748 * since the file was in fact removed
1749 * Therefore, we cheat and return success.
1751 if (error == ENOENT)
1753 } else if (!np->n_sillyrename) {
1754 error = nfs_sillyrename(dvp, vp, cnp);
1756 np->n_attrstamp = 0;
1761 * nfs file remove rpc called from nfs_inactive
1764 nfs_removeit(struct sillyrename *sp)
1766 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1771 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1774 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1775 struct ucred *cred, struct thread *td)
1777 int error = 0, wccflag = NFSV3_WCCRATTR;
1778 struct nfsm_info info;
1781 info.v3 = NFS_ISV3(dvp);
1783 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1784 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1785 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1786 ERROROUT(nfsm_fhtom(&info, dvp));
1787 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1788 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1790 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1795 VTONFS(dvp)->n_flag |= NLMODIFIED;
1797 VTONFS(dvp)->n_attrstamp = 0;
1802 * nfs file rename call
1804 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1805 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1806 * struct vnode *a_tvp, struct componentname *a_tcnp)
1809 nfs_rename(struct vop_old_rename_args *ap)
1811 struct vnode *fvp = ap->a_fvp;
1812 struct vnode *tvp = ap->a_tvp;
1813 struct vnode *fdvp = ap->a_fdvp;
1814 struct vnode *tdvp = ap->a_tdvp;
1815 struct componentname *tcnp = ap->a_tcnp;
1816 struct componentname *fcnp = ap->a_fcnp;
1819 /* Check for cross-device rename */
1820 if ((fvp->v_mount != tdvp->v_mount) ||
1821 (tvp && (fvp->v_mount != tvp->v_mount))) {
1827 * We shouldn't have to flush fvp on rename for most server-side
1828 * filesystems as the file handle should not change. Unfortunately
1829 * the inode for some filesystems (msdosfs) might be tied to the
1830 * file name or directory position so to be completely safe
1831 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1834 * We must flush tvp on rename because it might become stale on the
1835 * server after the rename.
1837 if (nfs_flush_on_rename)
1838 VOP_FSYNC(fvp, MNT_WAIT, 0);
1840 VOP_FSYNC(tvp, MNT_WAIT, 0);
1843 * If the tvp exists and is in use, sillyrename it before doing the
1844 * rename of the new file over it.
1846 * XXX Can't sillyrename a directory.
1848 * We do not attempt to do any namecache purges in this old API
1849 * routine. The new API compat functions have access to the actual
1850 * namecache structures and will do it for us.
1852 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1853 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1860 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1861 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1874 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1876 if (error == ENOENT)
1882 * nfs file rename rpc called from nfs_remove() above
1885 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1886 struct sillyrename *sp)
1888 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1889 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1893 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1896 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1897 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1898 struct ucred *cred, struct thread *td)
1900 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1901 struct nfsm_info info;
1904 info.v3 = NFS_ISV3(fdvp);
1906 nfsstats.rpccnt[NFSPROC_RENAME]++;
1907 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1908 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1909 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1910 ERROROUT(nfsm_fhtom(&info, fdvp));
1911 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1912 ERROROUT(nfsm_fhtom(&info, tdvp));
1913 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1914 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
1916 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
1917 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
1922 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1923 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1925 VTONFS(fdvp)->n_attrstamp = 0;
1927 VTONFS(tdvp)->n_attrstamp = 0;
1932 * nfs hard link create call
1934 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1935 * struct componentname *a_cnp)
1938 nfs_link(struct vop_old_link_args *ap)
1940 struct vnode *vp = ap->a_vp;
1941 struct vnode *tdvp = ap->a_tdvp;
1942 struct componentname *cnp = ap->a_cnp;
1943 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1944 struct nfsm_info info;
1946 if (vp->v_mount != tdvp->v_mount) {
1951 * The attribute cache may get out of sync with the server on link.
1952 * Pushing writes to the server before handle was inherited from
1953 * long long ago and it is unclear if we still need to do this.
1956 if (nfs_flush_on_hlink)
1957 VOP_FSYNC(vp, MNT_WAIT, 0);
1960 info.v3 = NFS_ISV3(vp);
1962 nfsstats.rpccnt[NFSPROC_LINK]++;
1963 nfsm_reqhead(&info, vp, NFSPROC_LINK,
1964 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
1965 nfsm_rndup(cnp->cn_namelen));
1966 ERROROUT(nfsm_fhtom(&info, vp));
1967 ERROROUT(nfsm_fhtom(&info, tdvp));
1968 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1970 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
1971 cnp->cn_cred, &error));
1973 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1974 NFS_LATTR_NOSHRINK));
1975 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
1980 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1982 VTONFS(vp)->n_attrstamp = 0;
1984 VTONFS(tdvp)->n_attrstamp = 0;
1986 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1988 if (error == EEXIST)
1994 * nfs symbolic link create call
1996 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1997 * struct componentname *a_cnp, struct vattr *a_vap,
2001 nfs_symlink(struct vop_old_symlink_args *ap)
2003 struct vnode *dvp = ap->a_dvp;
2004 struct vattr *vap = ap->a_vap;
2005 struct componentname *cnp = ap->a_cnp;
2006 struct nfsv2_sattr *sp;
2007 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2008 struct vnode *newvp = NULL;
2009 struct nfsm_info info;
2012 info.v3 = NFS_ISV3(dvp);
2014 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2015 slen = strlen(ap->a_target);
2016 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
2017 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
2018 nfsm_rndup(cnp->cn_namelen) +
2019 nfsm_rndup(slen) + NFSX_SATTR(info.v3));
2020 ERROROUT(nfsm_fhtom(&info, dvp));
2021 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2024 nfsm_v3attrbuild(&info, vap, FALSE);
2026 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
2028 sp = nfsm_build(&info, NFSX_V2SATTR);
2029 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2030 sp->sa_uid = nfs_xdrneg1;
2031 sp->sa_gid = nfs_xdrneg1;
2032 sp->sa_size = nfs_xdrneg1;
2033 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2034 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2038 * Issue the NFS request and get the rpc response.
2040 * Only NFSv3 responses returning an error of 0 actually return
2041 * a file handle that can be converted into newvp without having
2042 * to do an extra lookup rpc.
2044 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
2045 cnp->cn_cred, &error));
2048 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2050 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2054 * out code jumps -> here, mrep is also freed.
2062 * If we get an EEXIST error, silently convert it to no-error
2063 * in case of an NFS retry.
2065 if (error == EEXIST)
2069 * If we do not have (or no longer have) an error, and we could
2070 * not extract the newvp from the response due to the request being
2071 * NFSv2 or the error being EEXIST. We have to do a lookup in order
2072 * to obtain a newvp to return.
2074 if (error == 0 && newvp == NULL) {
2075 struct nfsnode *np = NULL;
2077 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2078 cnp->cn_cred, cnp->cn_td, &np);
2088 VTONFS(dvp)->n_flag |= NLMODIFIED;
2090 VTONFS(dvp)->n_attrstamp = 0;
2097 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2098 * struct componentname *a_cnp, struct vattr *a_vap)
2101 nfs_mkdir(struct vop_old_mkdir_args *ap)
2103 struct vnode *dvp = ap->a_dvp;
2104 struct vattr *vap = ap->a_vap;
2105 struct componentname *cnp = ap->a_cnp;
2106 struct nfsv2_sattr *sp;
2107 struct nfsnode *np = NULL;
2108 struct vnode *newvp = NULL;
2110 int error = 0, wccflag = NFSV3_WCCRATTR;
2113 struct nfsm_info info;
2116 info.v3 = NFS_ISV3(dvp);
2118 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2121 len = cnp->cn_namelen;
2122 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2123 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2124 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2125 nfsm_rndup(len) + NFSX_SATTR(info.v3));
2126 ERROROUT(nfsm_fhtom(&info, dvp));
2127 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2129 nfsm_v3attrbuild(&info, vap, FALSE);
2131 sp = nfsm_build(&info, NFSX_V2SATTR);
2132 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2133 sp->sa_uid = nfs_xdrneg1;
2134 sp->sa_gid = nfs_xdrneg1;
2135 sp->sa_size = nfs_xdrneg1;
2136 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2137 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2139 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2140 cnp->cn_cred, &error));
2142 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2145 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2150 VTONFS(dvp)->n_flag |= NLMODIFIED;
2152 VTONFS(dvp)->n_attrstamp = 0;
2154 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2155 * if we can succeed in looking up the directory.
2157 if (error == EEXIST || (!error && !gotvp)) {
2162 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2166 if (newvp->v_type != VDIR)
2179 * nfs remove directory call
2181 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2182 * struct componentname *a_cnp)
2185 nfs_rmdir(struct vop_old_rmdir_args *ap)
2187 struct vnode *vp = ap->a_vp;
2188 struct vnode *dvp = ap->a_dvp;
2189 struct componentname *cnp = ap->a_cnp;
2190 int error = 0, wccflag = NFSV3_WCCRATTR;
2191 struct nfsm_info info;
2194 info.v3 = NFS_ISV3(dvp);
2198 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2199 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2200 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2201 nfsm_rndup(cnp->cn_namelen));
2202 ERROROUT(nfsm_fhtom(&info, dvp));
2203 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2205 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2206 cnp->cn_cred, &error));
2208 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2213 VTONFS(dvp)->n_flag |= NLMODIFIED;
2215 VTONFS(dvp)->n_attrstamp = 0;
2217 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2219 if (error == ENOENT)
2227 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2230 nfs_readdir(struct vop_readdir_args *ap)
2232 struct vnode *vp = ap->a_vp;
2233 struct nfsnode *np = VTONFS(vp);
2234 struct uio *uio = ap->a_uio;
2238 if (vp->v_type != VDIR)
2241 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2245 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2246 * and then check that is still valid, or if this is an NQNFS mount
2247 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2248 * VOP_GETATTR() does not necessarily go to the wire.
2250 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2251 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2252 if (VOP_GETATTR(vp, &vattr) == 0 &&
2253 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2255 nfsstats.direofcache_hits++;
2261 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2262 * own cache coherency checks so we do not have to.
2264 tresid = uio->uio_resid;
2265 error = nfs_bioread(vp, uio, 0);
2267 if (!error && uio->uio_resid == tresid)
2268 nfsstats.direofcache_misses++;
2275 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2277 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2278 * offset/block and converts the nfs formatted directory entries for userland
2279 * consumption as well as deals with offsets into the middle of blocks.
2280 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2281 * be block-bounded. It must convert to cookies for the actual RPC.
2284 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2287 struct nfs_dirent *dp = NULL;
2292 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2293 struct nfsnode *dnp = VTONFS(vp);
2295 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2297 struct nfsm_info info;
2300 info.v3 = NFS_ISV3(vp);
2303 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2304 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2305 panic("nfs readdirrpc bad uio");
2309 * If there is no cookie, assume directory was stale.
2311 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2315 return (NFSERR_BAD_COOKIE);
2317 * Loop around doing readdir rpc's of size nm_readdirsize
2318 * truncated to a multiple of DIRBLKSIZ.
2319 * The stopping criteria is EOF or buffer full.
2321 while (more_dirs && bigenough) {
2322 nfsstats.rpccnt[NFSPROC_READDIR]++;
2323 nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2324 NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2325 ERROROUT(nfsm_fhtom(&info, vp));
2327 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2328 *tl++ = cookie.nfsuquad[0];
2329 *tl++ = cookie.nfsuquad[1];
2330 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2331 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2333 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2334 *tl++ = cookie.nfsuquad[0];
2336 *tl = txdr_unsigned(nmp->nm_readdirsize);
2337 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2339 nfs_vpcred(vp, ND_READ), &error));
2341 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2342 NFS_LATTR_NOSHRINK));
2343 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2344 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2345 dnp->n_cookieverf.nfsuquad[1] = *tl;
2347 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2348 more_dirs = fxdr_unsigned(int, *tl);
2350 /* loop thru the dir entries, converting them to std form */
2351 while (more_dirs && bigenough) {
2353 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2354 fileno = fxdr_hyper(tl);
2355 len = fxdr_unsigned(int, *(tl + 2));
2357 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2358 fileno = fxdr_unsigned(u_quad_t, *tl++);
2359 len = fxdr_unsigned(int, *tl);
2361 if (len <= 0 || len > NFS_MAXNAMLEN) {
2369 * len is the number of bytes in the path element
2370 * name, not including the \0 termination.
2372 * tlen is the number of bytes w have to reserve for
2373 * the path element name.
2375 tlen = nfsm_rndup(len);
2377 tlen += 4; /* To ensure null termination */
2380 * If the entry would cross a DIRBLKSIZ boundary,
2381 * extend the previous nfs_dirent to cover the
2384 left = DIRBLKSIZ - blksiz;
2385 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2386 dp->nfs_reclen += left;
2387 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2388 uiop->uio_iov->iov_len -= left;
2389 uiop->uio_offset += left;
2390 uiop->uio_resid -= left;
2393 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2396 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2397 dp->nfs_ino = fileno;
2398 dp->nfs_namlen = len;
2399 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2400 dp->nfs_type = DT_UNKNOWN;
2401 blksiz += dp->nfs_reclen;
2402 if (blksiz == DIRBLKSIZ)
2404 uiop->uio_offset += sizeof(struct nfs_dirent);
2405 uiop->uio_resid -= sizeof(struct nfs_dirent);
2406 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2407 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2408 ERROROUT(nfsm_mtouio(&info, uiop, len));
2411 * The uiop has advanced by nfs_dirent + len
2412 * but really needs to advance by
2415 cp = uiop->uio_iov->iov_base;
2417 *cp = '\0'; /* null terminate */
2418 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2419 uiop->uio_iov->iov_len -= tlen;
2420 uiop->uio_offset += tlen;
2421 uiop->uio_resid -= tlen;
2424 * NFS strings must be rounded up (nfsm_myouio
2425 * handled that in the bigenough case).
2427 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2430 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2432 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2436 * If we were able to accomodate the last entry,
2437 * get the cookie for the next one. Otherwise
2438 * hold-over the cookie for the one we were not
2439 * able to accomodate.
2442 cookie.nfsuquad[0] = *tl++;
2444 cookie.nfsuquad[1] = *tl++;
2445 } else if (info.v3) {
2450 more_dirs = fxdr_unsigned(int, *tl);
2453 * If at end of rpc data, get the eof boolean
2456 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2457 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2463 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2464 * by increasing d_reclen for the last record.
2467 left = DIRBLKSIZ - blksiz;
2468 dp->nfs_reclen += left;
2469 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2470 uiop->uio_iov->iov_len -= left;
2471 uiop->uio_offset += left;
2472 uiop->uio_resid -= left;
2477 * We hit the end of the directory, update direofoffset.
2479 dnp->n_direofoffset = uiop->uio_offset;
2482 * There is more to go, insert the link cookie so the
2483 * next block can be read.
2485 if (uiop->uio_resid > 0)
2486 kprintf("EEK! readdirrpc resid > 0\n");
2487 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2495 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2498 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2501 struct nfs_dirent *dp;
2503 struct vnode *newvp;
2505 caddr_t dpossav1, dpossav2;
2507 struct mbuf *mdsav1, *mdsav2;
2509 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2510 struct nfsnode *dnp = VTONFS(vp), *np;
2513 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2514 int attrflag, fhsize;
2515 struct nchandle nch;
2516 struct nchandle dnch;
2517 struct nlcomponent nlc;
2518 struct nfsm_info info;
2527 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2528 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2529 panic("nfs readdirplusrpc bad uio");
2532 * Obtain the namecache record for the directory so we have something
2533 * to use as a basis for creating the entries. This function will
2534 * return a held (but not locked) ncp. The ncp may be disconnected
2535 * from the tree and cannot be used for upward traversals, and the
2536 * ncp may be unnamed. Note that other unrelated operations may
2537 * cause the ncp to be named at any time.
2539 * We have to lock the ncp to prevent a lock order reversal when
2540 * rdirplus does nlookups of the children, because the vnode is
2541 * locked and has to stay that way.
2543 cache_fromdvp(vp, NULL, 0, &dnch);
2544 bzero(&nlc, sizeof(nlc));
2548 * If there is no cookie, assume directory was stale.
2550 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2556 return (NFSERR_BAD_COOKIE);
2560 * Loop around doing readdir rpc's of size nm_readdirsize
2561 * truncated to a multiple of DIRBLKSIZ.
2562 * The stopping criteria is EOF or buffer full.
2564 while (more_dirs && bigenough) {
2565 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2566 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2567 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2568 ERROROUT(nfsm_fhtom(&info, vp));
2569 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2570 *tl++ = cookie.nfsuquad[0];
2571 *tl++ = cookie.nfsuquad[1];
2572 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2573 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2574 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2575 *tl = txdr_unsigned(nmp->nm_rsize);
2576 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2578 nfs_vpcred(vp, ND_READ), &error));
2579 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2580 NFS_LATTR_NOSHRINK));
2581 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2582 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2583 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2584 more_dirs = fxdr_unsigned(int, *tl);
2586 /* loop thru the dir entries, doctoring them to 4bsd form */
2587 while (more_dirs && bigenough) {
2588 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2589 fileno = fxdr_hyper(tl);
2590 len = fxdr_unsigned(int, *(tl + 2));
2591 if (len <= 0 || len > NFS_MAXNAMLEN) {
2597 tlen = nfsm_rndup(len);
2599 tlen += 4; /* To ensure null termination*/
2600 left = DIRBLKSIZ - blksiz;
2601 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2602 dp->nfs_reclen += left;
2603 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2604 uiop->uio_iov->iov_len -= left;
2605 uiop->uio_offset += left;
2606 uiop->uio_resid -= left;
2609 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2612 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2613 dp->nfs_ino = fileno;
2614 dp->nfs_namlen = len;
2615 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2616 dp->nfs_type = DT_UNKNOWN;
2617 blksiz += dp->nfs_reclen;
2618 if (blksiz == DIRBLKSIZ)
2620 uiop->uio_offset += sizeof(struct nfs_dirent);
2621 uiop->uio_resid -= sizeof(struct nfs_dirent);
2622 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2623 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2624 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2625 nlc.nlc_namelen = len;
2626 ERROROUT(nfsm_mtouio(&info, uiop, len));
2627 cp = uiop->uio_iov->iov_base;
2630 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2631 uiop->uio_iov->iov_len -= tlen;
2632 uiop->uio_offset += tlen;
2633 uiop->uio_resid -= tlen;
2635 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2637 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2639 cookie.nfsuquad[0] = *tl++;
2640 cookie.nfsuquad[1] = *tl++;
2646 * Since the attributes are before the file handle
2647 * (sigh), we must skip over the attributes and then
2648 * come back and get them.
2650 attrflag = fxdr_unsigned(int, *tl);
2652 dpossav1 = info.dpos;
2654 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2655 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2656 doit = fxdr_unsigned(int, *tl);
2658 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2660 if (doit && bigenough && !nlcdegenerate(&nlc) &&
2661 !NFS_CMPFH(dnp, fhp, fhsize)
2665 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2666 nlc.nlc_namelen, nlc.nlc_namelen,
2670 * This is a bit hokey but there isn't
2671 * much we can do about it. We can't
2672 * hold the directory vp locked while
2673 * doing lookups and gets.
2675 nch = cache_nlookup_nonblock(&dnch, &nlc);
2676 if (nch.ncp == NULL)
2678 cache_setunresolved(&nch);
2679 error = nfs_nget_nonblock(vp->v_mount, fhp,
2686 dpossav2 = info.dpos;
2687 info.dpos = dpossav1;
2690 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2691 info.dpos = dpossav2;
2694 IFTODT(VTTOIF(np->n_vattr.va_type));
2695 nfs_cache_setvp(&nch, newvp,
2696 nfspos_cache_timeout);
2704 kprintf("Warning: NFS/rddirplus, "
2705 "UNABLE TO ENTER %*.*s\n",
2706 nlc.nlc_namelen, nlc.nlc_namelen,
2712 /* Just skip over the file handle */
2713 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2714 i = fxdr_unsigned(int, *tl);
2715 ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2717 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2718 more_dirs = fxdr_unsigned(int, *tl);
2721 * If at end of rpc data, get the eof boolean
2724 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2725 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2731 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2732 * by increasing d_reclen for the last record.
2735 left = DIRBLKSIZ - blksiz;
2736 dp->nfs_reclen += left;
2737 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2738 uiop->uio_iov->iov_len -= left;
2739 uiop->uio_offset += left;
2740 uiop->uio_resid -= left;
2744 * We are now either at the end of the directory or have filled the
2748 dnp->n_direofoffset = uiop->uio_offset;
2750 if (uiop->uio_resid > 0)
2751 kprintf("EEK! readdirplusrpc resid > 0\n");
2752 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2756 if (newvp != NULLVP) {
2769 * Silly rename. To make the NFS filesystem that is stateless look a little
2770 * more like the "ufs" a remove of an active vnode is translated to a rename
2771 * to a funny looking filename that is removed by nfs_inactive on the
2772 * nfsnode. There is the potential for another process on a different client
2773 * to create the same funny name between the nfs_lookitup() fails and the
2774 * nfs_rename() completes, but...
2777 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2779 struct sillyrename *sp;
2784 * We previously purged dvp instead of vp. I don't know why, it
2785 * completely destroys performance. We can't do it anyway with the
2786 * new VFS API since we would be breaking the namecache topology.
2788 cache_purge(vp); /* XXX */
2791 if (vp->v_type == VDIR)
2792 panic("nfs: sillyrename dir");
2794 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2795 M_NFSREQ, M_WAITOK);
2796 sp->s_cred = crdup(cnp->cn_cred);
2800 /* Fudge together a funny name */
2801 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2802 (int)(intptr_t)cnp->cn_td);
2804 /* Try lookitups until we get one that isn't there */
2805 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2806 cnp->cn_td, NULL) == 0) {
2808 if (sp->s_name[4] > 'z') {
2813 error = nfs_renameit(dvp, cnp, sp);
2816 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2818 np->n_sillyrename = sp;
2823 kfree((caddr_t)sp, M_NFSREQ);
2828 * Look up a file name and optionally either update the file handle or
2829 * allocate an nfsnode, depending on the value of npp.
2830 * npp == NULL --> just do the lookup
2831 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2833 * *npp != NULL --> update the file handle in the vnode
2836 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2837 struct thread *td, struct nfsnode **npp)
2839 struct vnode *newvp = NULL;
2840 struct nfsnode *np, *dnp = VTONFS(dvp);
2841 int error = 0, fhlen, attrflag;
2843 struct nfsm_info info;
2846 info.v3 = NFS_ISV3(dvp);
2848 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2849 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2850 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2851 ERROROUT(nfsm_fhtom(&info, dvp));
2852 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2853 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2854 if (npp && !error) {
2855 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2858 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2859 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2860 np->n_fhp = &np->n_fh;
2861 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2862 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2863 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2864 np->n_fhsize = fhlen;
2866 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2870 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2879 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2880 NFS_LATTR_NOSHRINK));
2881 if (!attrflag && *npp == NULL) {
2891 ERROROUT(error = nfsm_loadattr(&info, newvp, NULL));
2897 if (npp && *npp == NULL) {
2912 * Nfs Version 3 commit rpc
2914 * We call it 'uio' to distinguish it from 'bio' but there is no real uio
2918 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2920 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2921 int error = 0, wccflag = NFSV3_WCCRATTR;
2922 struct nfsm_info info;
2928 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2930 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2931 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
2932 ERROROUT(nfsm_fhtom(&info, vp));
2933 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
2934 txdr_hyper(offset, tl);
2936 *tl = txdr_unsigned(cnt);
2937 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
2938 nfs_vpcred(vp, ND_WRITE), &error));
2939 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
2941 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
2942 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2943 NFSX_V3WRITEVERF)) {
2944 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2946 error = NFSERR_STALEWRITEVERF;
2957 * - make nfs_bmap() essentially a no-op that does no translation
2958 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2959 * (Maybe I could use the process's page mapping, but I was concerned that
2960 * Kernel Write might not be enabled and also figured copyout() would do
2961 * a lot more work than bcopy() and also it currently happens in the
2962 * context of the swapper process (2).
2964 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
2965 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2968 nfs_bmap(struct vop_bmap_args *ap)
2970 if (ap->a_doffsetp != NULL)
2971 *ap->a_doffsetp = ap->a_loffset;
2972 if (ap->a_runp != NULL)
2974 if (ap->a_runb != NULL)
2983 nfs_strategy(struct vop_strategy_args *ap)
2985 struct bio *bio = ap->a_bio;
2987 struct buf *bp __debugvar = bio->bio_buf;
2991 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2992 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
2993 KASSERT(BUF_REFCNT(bp) > 0,
2994 ("nfs_strategy: buffer %p not locked", bp));
2996 if (bio->bio_flags & BIO_SYNC)
2997 td = curthread; /* XXX */
3002 * We probably don't need to push an nbio any more since no
3003 * block conversion is required due to the use of 64 bit byte
3004 * offsets, but do it anyway.
3006 * NOTE: When NFS callers itself via this strategy routines and
3007 * sets up a synchronous I/O, it expects the I/O to run
3008 * synchronously (its bio_done routine just assumes it),
3009 * so for now we have to honor the bit.
3011 nbio = push_bio(bio);
3012 nbio->bio_offset = bio->bio_offset;
3013 nbio->bio_flags = bio->bio_flags & BIO_SYNC;
3016 * If the op is asynchronous and an i/o daemon is waiting
3017 * queue the request, wake it up and wait for completion
3018 * otherwise just do it ourselves.
3020 if (bio->bio_flags & BIO_SYNC) {
3021 error = nfs_doio(ap->a_vp, nbio, td);
3023 nfs_asyncio(ap->a_vp, nbio);
3032 * NB Currently unsupported.
3034 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
3038 nfs_mmap(struct vop_mmap_args *ap)
3044 * fsync vnode op. Just call nfs_flush() with commit == 1.
3046 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3050 nfs_fsync(struct vop_fsync_args *ap)
3052 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
3056 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
3057 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
3058 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
3059 * set the buffer contains data that has already been written to the server
3060 * and which now needs a commit RPC.
3062 * If commit is 0 we only take one pass and only flush buffers containing new
3065 * If commit is 1 we take two passes, issuing a commit RPC in the second
3068 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3069 * to completely flush all pending data.
3071 * Note that the RB_SCAN code properly handles the case where the
3072 * callback might block and directly or indirectly (another thread) cause
3073 * the RB tree to change.
3076 #ifndef NFS_COMMITBVECSIZ
3077 #define NFS_COMMITBVECSIZ 16
3080 struct nfs_flush_info {
3081 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3088 struct buf *bvary[NFS_COMMITBVECSIZ];
3094 static int nfs_flush_bp(struct buf *bp, void *data);
3095 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3098 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3100 struct nfsnode *np = VTONFS(vp);
3101 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3102 struct nfs_flush_info info;
3105 bzero(&info, sizeof(info));
3108 info.waitfor = waitfor;
3109 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3111 lwkt_gettoken(&vp->v_token);
3117 info.mode = NFI_FLUSHNEW;
3118 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3119 nfs_flush_bp, &info);
3122 * Take a second pass if committing and no error occured.
3123 * Clean up any left over collection (whether an error
3126 if (commit && error == 0) {
3127 info.mode = NFI_COMMIT;
3128 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3129 nfs_flush_bp, &info);
3131 error = nfs_flush_docommit(&info, error);
3135 * Wait for pending I/O to complete before checking whether
3136 * any further dirty buffers exist.
3138 while (waitfor == MNT_WAIT &&
3139 bio_track_active(&vp->v_track_write)) {
3140 error = bio_track_wait(&vp->v_track_write,
3141 info.slpflag, info.slptimeo);
3144 * We have to be able to break out if this
3145 * is an 'intr' mount.
3147 if (nfs_sigintr(nmp, NULL, td)) {
3153 * Since we do not process pending signals,
3154 * once we get a PCATCH our tsleep() will no
3155 * longer sleep, switch to a fixed timeout
3158 if (info.slpflag == PCATCH) {
3160 info.slptimeo = 2 * hz;
3167 * Loop if we are flushing synchronous as well as committing,
3168 * and dirty buffers are still present. Otherwise we might livelock.
3170 } while (waitfor == MNT_WAIT && commit &&
3171 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3174 * The callbacks have to return a negative error to terminate the
3181 * Deal with any error collection
3183 if (np->n_flag & NWRITEERR) {
3184 error = np->n_error;
3185 np->n_flag &= ~NWRITEERR;
3187 lwkt_reltoken(&vp->v_token);
3193 nfs_flush_bp(struct buf *bp, void *data)
3195 struct nfs_flush_info *info = data;
3201 switch(info->mode) {
3203 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3204 if (error && info->loops && info->waitfor == MNT_WAIT) {
3205 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3207 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3208 if (info->slpflag & PCATCH)
3209 lkflags |= LK_PCATCH;
3210 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3216 * Ignore locking errors
3224 * The buffer may have changed out from under us, even if
3225 * we did not block (MPSAFE). Check again now that it is
3228 if (bp->b_vp == info->vp &&
3229 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3238 * Only process buffers in need of a commit which we can
3239 * immediately lock. This may prevent a buffer from being
3240 * committed, but the normal flush loop will block on the
3241 * same buffer so we shouldn't get into an endless loop.
3243 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3244 (B_DELWRI | B_NEEDCOMMIT)) {
3247 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3251 * We must recheck after successfully locking the buffer.
3253 if (bp->b_vp != info->vp ||
3254 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3255 (B_DELWRI | B_NEEDCOMMIT)) {
3261 * NOTE: storing the bp in the bvary[] basically sets
3262 * it up for a commit operation.
3264 * We must call vfs_busy_pages() now so the commit operation
3265 * is interlocked with user modifications to memory mapped
3266 * pages. The b_dirtyoff/b_dirtyend range is not correct
3267 * until after the pages have been busied.
3269 * Note: to avoid loopback deadlocks, we do not
3270 * assign b_runningbufspace.
3273 bp->b_cmd = BUF_CMD_WRITE;
3274 vfs_busy_pages(bp->b_vp, bp);
3275 info->bvary[info->bvsize] = bp;
3276 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3277 if (info->bvsize == 0 || toff < info->beg_off)
3278 info->beg_off = toff;
3279 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3280 if (info->bvsize == 0 || toff > info->end_off)
3281 info->end_off = toff;
3283 if (info->bvsize == NFS_COMMITBVECSIZ) {
3284 error = nfs_flush_docommit(info, 0);
3285 KKASSERT(info->bvsize == 0);
3293 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3303 if (info->bvsize > 0) {
3305 * Commit data on the server, as required. Note that
3306 * nfs_commit will use the vnode's cred for the commit.
3307 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3309 bytes = info->end_off - info->beg_off;
3310 if (bytes > 0x40000000)
3315 retv = nfs_commitrpc_uio(vp, info->beg_off,
3316 (int)bytes, info->td);
3317 if (retv == NFSERR_STALEWRITEVERF)
3318 nfs_clearcommit(vp->v_mount);
3322 * Now, either mark the blocks I/O done or mark the
3323 * blocks dirty, depending on whether the commit
3326 for (i = 0; i < info->bvsize; ++i) {
3327 bp = info->bvary[i];
3328 if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3330 * Either an error or the original
3331 * vfs_busy_pages() cleared B_NEEDCOMMIT
3332 * due to finding new dirty VM pages in
3335 * Leave B_DELWRI intact.
3337 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3338 vfs_unbusy_pages(bp);
3339 bp->b_cmd = BUF_CMD_DONE;
3343 * Success, remove B_DELWRI ( bundirty() ).
3345 * b_dirtyoff/b_dirtyend seem to be NFS
3346 * specific. We should probably move that
3347 * into bundirty(). XXX
3349 * We are faking an I/O write, we have to
3350 * start the transaction in order to
3351 * immediately biodone() it.
3354 bp->b_flags &= ~B_ERROR;
3355 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3356 bp->b_dirtyoff = bp->b_dirtyend = 0;
3357 biodone(&bp->b_bio1);
3366 * NFS advisory byte-level locks.
3367 * Currently unsupported.
3369 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3373 nfs_advlock(struct vop_advlock_args *ap)
3375 struct nfsnode *np = VTONFS(ap->a_vp);
3378 * The following kludge is to allow diskless support to work
3379 * until a real NFS lockd is implemented. Basically, just pretend
3380 * that this is a local lock.
3382 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3386 * Print out the contents of an nfsnode.
3388 * nfs_print(struct vnode *a_vp)
3391 nfs_print(struct vop_print_args *ap)
3393 struct vnode *vp = ap->a_vp;
3394 struct nfsnode *np = VTONFS(vp);
3396 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3397 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3398 if (vp->v_type == VFIFO)
3405 * nfs special file access vnode op.
3407 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3410 nfs_laccess(struct vop_access_args *ap)
3415 error = VOP_GETATTR(ap->a_vp, &vattr);
3417 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3423 * Read wrapper for fifos.
3425 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3426 * struct ucred *a_cred)
3429 nfsfifo_read(struct vop_read_args *ap)
3431 struct nfsnode *np = VTONFS(ap->a_vp);
3437 getnanotime(&np->n_atim);
3438 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3442 * Write wrapper for fifos.
3444 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3445 * struct ucred *a_cred)
3448 nfsfifo_write(struct vop_write_args *ap)
3450 struct nfsnode *np = VTONFS(ap->a_vp);
3456 getnanotime(&np->n_mtim);
3457 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3461 * Close wrapper for fifos.
3463 * Update the times on the nfsnode then do fifo close.
3465 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3468 nfsfifo_close(struct vop_close_args *ap)
3470 struct vnode *vp = ap->a_vp;
3471 struct nfsnode *np = VTONFS(vp);
3475 if (np->n_flag & (NACC | NUPD)) {
3477 if (np->n_flag & NACC)
3479 if (np->n_flag & NUPD)
3482 if (vp->v_sysref.refcnt == 1 &&
3483 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3485 if (np->n_flag & NACC)
3486 vattr.va_atime = np->n_atim;
3487 if (np->n_flag & NUPD)
3488 vattr.va_mtime = np->n_mtim;
3489 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3492 return (VOCALL(&fifo_vnode_vops, &ap->a_head));