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 1.56 2006/04/28 00:24:46 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>
98 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
99 * calls are not in getblk() and brelse() so that they would not be necessary
103 #define vfs_busy_pages(bp, f)
106 static int nfsspec_read (struct vop_read_args *);
107 static int nfsspec_write (struct vop_write_args *);
108 static int nfsfifo_read (struct vop_read_args *);
109 static int nfsfifo_write (struct vop_write_args *);
110 static int nfsspec_close (struct vop_close_args *);
111 static int nfsfifo_close (struct vop_close_args *);
112 #define nfs_poll vop_nopoll
113 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
114 static int nfs_lookup (struct vop_old_lookup_args *);
115 static int nfs_create (struct vop_old_create_args *);
116 static int nfs_mknod (struct vop_old_mknod_args *);
117 static int nfs_open (struct vop_open_args *);
118 static int nfs_close (struct vop_close_args *);
119 static int nfs_access (struct vop_access_args *);
120 static int nfs_getattr (struct vop_getattr_args *);
121 static int nfs_setattr (struct vop_setattr_args *);
122 static int nfs_read (struct vop_read_args *);
123 static int nfs_mmap (struct vop_mmap_args *);
124 static int nfs_fsync (struct vop_fsync_args *);
125 static int nfs_remove (struct vop_old_remove_args *);
126 static int nfs_link (struct vop_old_link_args *);
127 static int nfs_rename (struct vop_old_rename_args *);
128 static int nfs_mkdir (struct vop_old_mkdir_args *);
129 static int nfs_rmdir (struct vop_old_rmdir_args *);
130 static int nfs_symlink (struct vop_old_symlink_args *);
131 static int nfs_readdir (struct vop_readdir_args *);
132 static int nfs_bmap (struct vop_bmap_args *);
133 static int nfs_strategy (struct vop_strategy_args *);
134 static int nfs_lookitup (struct vnode *, const char *, int,
135 struct ucred *, struct thread *, struct nfsnode **);
136 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
137 static int nfsspec_access (struct vop_access_args *);
138 static int nfs_readlink (struct vop_readlink_args *);
139 static int nfs_print (struct vop_print_args *);
140 static int nfs_advlock (struct vop_advlock_args *);
141 static int nfs_bwrite (struct vop_bwrite_args *);
143 static int nfs_nresolve (struct vop_nresolve_args *);
145 * Global vfs data structures for nfs
147 struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
148 { &vop_default_desc, vop_defaultop },
149 { &vop_access_desc, (vnodeopv_entry_t) nfs_access },
150 { &vop_advlock_desc, (vnodeopv_entry_t) nfs_advlock },
151 { &vop_bmap_desc, (vnodeopv_entry_t) nfs_bmap },
152 { &vop_bwrite_desc, (vnodeopv_entry_t) nfs_bwrite },
153 { &vop_close_desc, (vnodeopv_entry_t) nfs_close },
154 { &vop_old_create_desc, (vnodeopv_entry_t) nfs_create },
155 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync },
156 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr },
157 { &vop_getpages_desc, (vnodeopv_entry_t) nfs_getpages },
158 { &vop_putpages_desc, (vnodeopv_entry_t) nfs_putpages },
159 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive },
160 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked },
161 { &vop_old_link_desc, (vnodeopv_entry_t) nfs_link },
162 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock },
163 { &vop_old_lookup_desc, (vnodeopv_entry_t) nfs_lookup },
164 { &vop_old_mkdir_desc, (vnodeopv_entry_t) nfs_mkdir },
165 { &vop_old_mknod_desc, (vnodeopv_entry_t) nfs_mknod },
166 { &vop_mmap_desc, (vnodeopv_entry_t) nfs_mmap },
167 { &vop_open_desc, (vnodeopv_entry_t) nfs_open },
168 { &vop_poll_desc, (vnodeopv_entry_t) nfs_poll },
169 { &vop_print_desc, (vnodeopv_entry_t) nfs_print },
170 { &vop_read_desc, (vnodeopv_entry_t) nfs_read },
171 { &vop_readdir_desc, (vnodeopv_entry_t) nfs_readdir },
172 { &vop_readlink_desc, (vnodeopv_entry_t) nfs_readlink },
173 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim },
174 { &vop_old_remove_desc, (vnodeopv_entry_t) nfs_remove },
175 { &vop_old_rename_desc, (vnodeopv_entry_t) nfs_rename },
176 { &vop_old_rmdir_desc, (vnodeopv_entry_t) nfs_rmdir },
177 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr },
178 { &vop_strategy_desc, (vnodeopv_entry_t) nfs_strategy },
179 { &vop_old_symlink_desc, (vnodeopv_entry_t) nfs_symlink },
180 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock },
181 { &vop_write_desc, (vnodeopv_entry_t) nfs_write },
183 { &vop_nresolve_desc, (vnodeopv_entry_t) nfs_nresolve },
188 * Special device vnode ops
190 struct vnodeopv_entry_desc nfsv2_specop_entries[] = {
191 { &vop_default_desc, (vnodeopv_entry_t) spec_vnoperate },
192 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access },
193 { &vop_close_desc, (vnodeopv_entry_t) nfsspec_close },
194 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync },
195 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr },
196 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive },
197 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked },
198 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock },
199 { &vop_print_desc, (vnodeopv_entry_t) nfs_print },
200 { &vop_read_desc, (vnodeopv_entry_t) nfsspec_read },
201 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim },
202 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr },
203 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock },
204 { &vop_write_desc, (vnodeopv_entry_t) nfsspec_write },
208 struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = {
209 { &vop_default_desc, (vnodeopv_entry_t) fifo_vnoperate },
210 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access },
211 { &vop_close_desc, (vnodeopv_entry_t) nfsfifo_close },
212 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync },
213 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr },
214 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive },
215 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked },
216 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock },
217 { &vop_print_desc, (vnodeopv_entry_t) nfs_print },
218 { &vop_read_desc, (vnodeopv_entry_t) nfsfifo_read },
219 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim },
220 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr },
221 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock },
222 { &vop_write_desc, (vnodeopv_entry_t) nfsfifo_write },
226 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
227 struct componentname *cnp,
229 static int nfs_removerpc (struct vnode *dvp, const char *name,
231 struct ucred *cred, struct thread *td);
232 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
233 int fnamelen, struct vnode *tdvp,
234 const char *tnameptr, int tnamelen,
235 struct ucred *cred, struct thread *td);
236 static int nfs_renameit (struct vnode *sdvp,
237 struct componentname *scnp,
238 struct sillyrename *sp);
243 extern u_int32_t nfs_true, nfs_false;
244 extern u_int32_t nfs_xdrneg1;
245 extern struct nfsstats nfsstats;
246 extern nfstype nfsv3_type[9];
247 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON];
248 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
249 int nfs_numasync = 0;
251 SYSCTL_DECL(_vfs_nfs);
253 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
254 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
255 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
257 static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
258 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
259 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache timeout");
261 static int nfsv3_commit_on_close = 0;
262 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
263 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
265 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
266 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
268 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
269 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
272 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
273 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
274 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
276 nfs3_access_otw(struct vnode *vp, int wmode,
277 struct thread *td, struct ucred *cred)
281 int error = 0, attrflag;
283 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
284 caddr_t bpos, dpos, cp2;
288 struct nfsnode *np = VTONFS(vp);
290 nfsstats.rpccnt[NFSPROC_ACCESS]++;
291 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
293 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
294 *tl = txdr_unsigned(wmode);
295 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
296 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
298 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
299 rmode = fxdr_unsigned(u_int32_t, *tl);
301 np->n_modeuid = cred->cr_uid;
302 np->n_modestamp = mycpu->gd_time_seconds;
310 * nfs access vnode op.
311 * For nfs version 2, just return ok. File accesses may fail later.
312 * For nfs version 3, use the access rpc to check accessibility. If file modes
313 * are changed on the server, accesses might still fail later.
315 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
316 * struct thread *a_td)
319 nfs_access(struct vop_access_args *ap)
321 struct vnode *vp = ap->a_vp;
323 u_int32_t mode, wmode;
324 int v3 = NFS_ISV3(vp);
325 struct nfsnode *np = VTONFS(vp);
328 * Disallow write attempts on filesystems mounted read-only;
329 * unless the file is a socket, fifo, or a block or character
330 * device resident on the filesystem.
332 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
333 switch (vp->v_type) {
343 * For nfs v3, check to see if we have done this recently, and if
344 * so return our cached result instead of making an ACCESS call.
345 * If not, do an access rpc, otherwise you are stuck emulating
346 * ufs_access() locally using the vattr. This may not be correct,
347 * since the server may apply other access criteria such as
348 * client uid-->server uid mapping that we do not know about.
351 if (ap->a_mode & VREAD)
352 mode = NFSV3ACCESS_READ;
355 if (vp->v_type != VDIR) {
356 if (ap->a_mode & VWRITE)
357 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
358 if (ap->a_mode & VEXEC)
359 mode |= NFSV3ACCESS_EXECUTE;
361 if (ap->a_mode & VWRITE)
362 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
364 if (ap->a_mode & VEXEC)
365 mode |= NFSV3ACCESS_LOOKUP;
367 /* XXX safety belt, only make blanket request if caching */
368 if (nfsaccess_cache_timeout > 0) {
369 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
370 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
371 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
377 * Does our cached result allow us to give a definite yes to
380 if (np->n_modestamp &&
381 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
382 (ap->a_cred->cr_uid == np->n_modeuid) &&
383 ((np->n_mode & mode) == mode)) {
384 nfsstats.accesscache_hits++;
387 * Either a no, or a don't know. Go to the wire.
389 nfsstats.accesscache_misses++;
390 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
392 if ((np->n_mode & mode) != mode) {
398 if ((error = nfsspec_access(ap)) != 0)
402 * Attempt to prevent a mapped root from accessing a file
403 * which it shouldn't. We try to read a byte from the file
404 * if the user is root and the file is not zero length.
405 * After calling nfsspec_access, we should have the correct
408 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
409 && VTONFS(vp)->n_size > 0) {
416 auio.uio_iov = &aiov;
420 auio.uio_segflg = UIO_SYSSPACE;
421 auio.uio_rw = UIO_READ;
422 auio.uio_td = ap->a_td;
424 if (vp->v_type == VREG) {
425 error = nfs_readrpc(vp, &auio);
426 } else if (vp->v_type == VDIR) {
428 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
430 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
431 error = nfs_readdirrpc(vp, &auio);
433 } else if (vp->v_type == VLNK) {
434 error = nfs_readlinkrpc(vp, &auio);
441 * [re]record creds for reading and/or writing if access
442 * was granted. Assume the NFS server will grant read access
443 * for execute requests.
446 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) {
449 crfree(np->n_rucred);
450 np->n_rucred = ap->a_cred;
452 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
455 crfree(np->n_wucred);
456 np->n_wucred = ap->a_cred;
464 * Check to see if the type is ok
465 * and that deletion is not in progress.
466 * For paged in text files, you will need to flush the page cache
467 * if consistency is lost.
469 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
470 * struct thread *a_td)
474 nfs_open(struct vop_open_args *ap)
476 struct vnode *vp = ap->a_vp;
477 struct nfsnode *np = VTONFS(vp);
481 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
483 printf("open eacces vtyp=%d\n",vp->v_type);
489 * Clear the attribute cache only if opening with write access. It
490 * is unclear if we should do this at all here, but we certainly
491 * should not clear the cache unconditionally simply because a file
494 if (ap->a_mode & FWRITE)
498 * For normal NFS, reconcile changes made locally verses
499 * changes made remotely. Note that VOP_GETATTR only goes
500 * to the wire if the cached attribute has timed out or been
503 * If local modifications have been made clear the attribute
504 * cache to force an attribute and modified time check. If
505 * GETATTR detects that the file has been changed by someone
506 * other then us it will set NRMODIFIED.
508 * If we are opening a directory and local changes have been
509 * made we have to invalidate the cache in order to ensure
510 * that we get the most up-to-date information from the
513 if (np->n_flag & NLMODIFIED) {
515 if (vp->v_type == VDIR) {
516 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
522 error = VOP_GETATTR(vp, &vattr, ap->a_td);
525 if (np->n_flag & NRMODIFIED) {
526 if (vp->v_type == VDIR)
528 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
531 np->n_flag &= ~NRMODIFIED;
534 return (vop_stdopen(ap));
539 * What an NFS client should do upon close after writing is a debatable issue.
540 * Most NFS clients push delayed writes to the server upon close, basically for
542 * 1 - So that any write errors may be reported back to the client process
543 * doing the close system call. By far the two most likely errors are
544 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
545 * 2 - To put a worst case upper bound on cache inconsistency between
546 * multiple clients for the file.
547 * There is also a consistency problem for Version 2 of the protocol w.r.t.
548 * not being able to tell if other clients are writing a file concurrently,
549 * since there is no way of knowing if the changed modify time in the reply
550 * is only due to the write for this client.
551 * (NFS Version 3 provides weak cache consistency data in the reply that
552 * should be sufficient to detect and handle this case.)
554 * The current code does the following:
555 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
556 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
557 * or commit them (this satisfies 1 and 2 except for the
558 * case where the server crashes after this close but
559 * before the commit RPC, which is felt to be "good
560 * enough". Changing the last argument to nfs_flush() to
561 * a 1 would force a commit operation, if it is felt a
562 * commit is necessary now.
563 * for NQNFS - do nothing now, since 2 is dealt with via leases and
564 * 1 should be dealt with via an fsync() system call for
565 * cases where write errors are important.
567 * nfs_close(struct vnodeop_desc *a_desc, struct vnode *a_vp, int a_fflag,
568 * struct ucred *a_cred, struct thread *a_td)
572 nfs_close(struct vop_close_args *ap)
574 struct vnode *vp = ap->a_vp;
575 struct nfsnode *np = VTONFS(vp);
578 if (vp->v_type == VREG) {
579 if (np->n_flag & NLMODIFIED) {
582 * Under NFSv3 we have dirty buffers to dispose of. We
583 * must flush them to the NFS server. We have the option
584 * of waiting all the way through the commit rpc or just
585 * waiting for the initial write. The default is to only
586 * wait through the initial write so the data is in the
587 * server's cache, which is roughly similar to the state
588 * a standard disk subsystem leaves the file in on close().
590 * We cannot clear the NLMODIFIED bit in np->n_flag due to
591 * potential races with other processes, and certainly
592 * cannot clear it if we don't commit.
594 int cm = nfsv3_commit_on_close ? 1 : 0;
595 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
596 /* np->n_flag &= ~NLMODIFIED; */
598 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
602 if (np->n_flag & NWRITEERR) {
603 np->n_flag &= ~NWRITEERR;
612 * nfs getattr call from vfs.
614 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred,
615 * struct thread *a_td)
618 nfs_getattr(struct vop_getattr_args *ap)
620 struct vnode *vp = ap->a_vp;
621 struct nfsnode *np = VTONFS(vp);
627 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
628 int v3 = NFS_ISV3(vp);
631 * Update local times for special files.
633 if (np->n_flag & (NACC | NUPD))
636 * First look in the cache.
638 if (nfs_getattrcache(vp, ap->a_vap) == 0)
641 if (v3 && nfsaccess_cache_timeout > 0) {
642 nfsstats.accesscache_misses++;
643 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK));
644 if (nfs_getattrcache(vp, ap->a_vap) == 0)
648 nfsstats.rpccnt[NFSPROC_GETATTR]++;
649 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
651 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK));
653 nfsm_loadattr(vp, ap->a_vap);
663 * nfs_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp,
664 * struct vattr *a_vap, struct ucred *a_cred,
665 * struct thread *a_td)
668 nfs_setattr(struct vop_setattr_args *ap)
670 struct vnode *vp = ap->a_vp;
671 struct nfsnode *np = VTONFS(vp);
672 struct vattr *vap = ap->a_vap;
681 * Setting of flags is not supported.
683 if (vap->va_flags != VNOVAL)
687 * Disallow write attempts if the filesystem is mounted read-only.
689 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
690 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
691 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
692 (vp->v_mount->mnt_flag & MNT_RDONLY))
694 if (vap->va_size != VNOVAL) {
695 switch (vp->v_type) {
702 if (vap->va_mtime.tv_sec == VNOVAL &&
703 vap->va_atime.tv_sec == VNOVAL &&
704 vap->va_mode == (mode_t)VNOVAL &&
705 vap->va_uid == (uid_t)VNOVAL &&
706 vap->va_gid == (gid_t)VNOVAL)
708 vap->va_size = VNOVAL;
712 * Disallow write attempts if the filesystem is
715 if (vp->v_mount->mnt_flag & MNT_RDONLY)
719 * This is nasty. The RPCs we send to flush pending
720 * data often return attribute information which is
721 * cached via a callback to nfs_loadattrcache(), which
722 * has the effect of changing our notion of the file
723 * size. Due to flushed appends and other operations
724 * the file size can be set to virtually anything,
725 * including values that do not match either the old
726 * or intended file size.
728 * When this condition is detected we must loop to
729 * try the operation again. Hopefully no more
730 * flushing is required on the loop so it works the
731 * second time around. THIS CASE ALMOST ALWAYS
736 error = nfs_meta_setsize(vp, ap->a_td, vap->va_size);
738 if (np->n_flag & NLMODIFIED) {
739 if (vap->va_size == 0)
740 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
742 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
745 * note: this loop case almost always happens at
746 * least once per truncation.
748 if (error == 0 && np->n_size != vap->va_size)
750 np->n_vattr.va_size = vap->va_size;
753 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
754 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) &&
755 vp->v_type == VREG &&
756 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR
760 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
763 * Sanity check if a truncation was issued. This should only occur
764 * if multiple processes are racing on the same file.
766 if (error == 0 && vap->va_size != VNOVAL &&
767 np->n_size != vap->va_size) {
768 printf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size);
771 if (error && vap->va_size != VNOVAL) {
772 np->n_size = np->n_vattr.va_size = tsize;
773 vnode_pager_setsize(vp, np->n_size);
779 * Do an nfs setattr rpc.
782 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
783 struct ucred *cred, struct thread *td)
785 struct nfsv2_sattr *sp;
786 struct nfsnode *np = VTONFS(vp);
789 caddr_t bpos, dpos, cp2;
791 int error = 0, wccflag = NFSV3_WCCRATTR;
792 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
793 int v3 = NFS_ISV3(vp);
795 nfsstats.rpccnt[NFSPROC_SETATTR]++;
796 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
799 nfsm_v3attrbuild(vap, TRUE);
800 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
803 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
804 if (vap->va_mode == (mode_t)VNOVAL)
805 sp->sa_mode = nfs_xdrneg1;
807 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
808 if (vap->va_uid == (uid_t)VNOVAL)
809 sp->sa_uid = nfs_xdrneg1;
811 sp->sa_uid = txdr_unsigned(vap->va_uid);
812 if (vap->va_gid == (gid_t)VNOVAL)
813 sp->sa_gid = nfs_xdrneg1;
815 sp->sa_gid = txdr_unsigned(vap->va_gid);
816 sp->sa_size = txdr_unsigned(vap->va_size);
817 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
818 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
820 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
823 nfsm_wcc_data(vp, wccflag);
825 nfsm_loadattr(vp, (struct vattr *)0);
832 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
833 * nfs_lookup() until all remaining new api calls are implemented.
835 * Resolve a namecache entry. This function is passed a locked ncp and
836 * must call cache_setvp() on it as appropriate to resolve the entry.
839 nfs_nresolve(struct vop_nresolve_args *ap)
841 struct thread *td = curthread;
842 struct namecache *ncp;
853 /******NFSM MACROS********/
854 struct mbuf *mb, *mrep, *mreq, *mb2, *md;
855 caddr_t bpos, dpos, cp, cp2;
862 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp);
863 dvp = ncp->nc_parent->nc_vp;
864 if ((error = vget(dvp, LK_SHARED, td)) != 0)
869 nfsstats.lookupcache_misses++;
870 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
872 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
873 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
875 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN);
876 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred);
879 * Cache negatve lookups to reduce NFS traffic, but use
880 * a fast timeout. Otherwise use a timeout of 1 tick.
881 * XXX we should add a namecache flag for no-caching
882 * to uncache the negative hit as soon as possible, but
883 * we cannot simply destroy the entry because it is used
884 * as a placeholder by the caller.
886 if (error == ENOENT) {
889 if (nfsneg_cache_timeout)
890 nticks = nfsneg_cache_timeout * hz;
893 cache_setvp(ncp, NULL);
894 cache_settimeout(ncp, nticks);
896 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
902 * Success, get the file handle, do various checks, and load
903 * post-operation data from the reply packet. Theoretically
904 * we should never be looking up "." so, theoretically, we
905 * should never get the same file handle as our directory. But
906 * we check anyway. XXX
908 * Note that no timeout is set for the positive cache hit. We
909 * assume, theoretically, that ESTALE returns will be dealt with
910 * properly to handle NFS races and in anycase we cannot depend
911 * on a timeout to deal with NFS open/create/excl issues so instead
912 * of a bad hack here the rest of the NFS client code needs to do
915 nfsm_getfh(fhp, fhsize, v3);
918 if (NFS_CMPFH(np, fhp, fhsize)) {
922 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
931 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK);
932 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
934 nfsm_loadattr(nvp, NULL);
936 cache_setvp(ncp, nvp);
950 * 'cached' nfs directory lookup
952 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
954 * nfs_lookup(struct vnodeop_desc *a_desc, struct vnode *a_dvp,
955 * struct vnode **a_vpp, struct componentname *a_cnp)
958 nfs_lookup(struct vop_old_lookup_args *ap)
960 struct componentname *cnp = ap->a_cnp;
961 struct vnode *dvp = ap->a_dvp;
962 struct vnode **vpp = ap->a_vpp;
963 int flags = cnp->cn_flags;
968 struct nfsmount *nmp;
969 caddr_t bpos, dpos, cp2;
970 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
974 int lockparent, wantparent, error = 0, attrflag, fhsize;
975 int v3 = NFS_ISV3(dvp);
976 struct thread *td = cnp->cn_td;
979 * Read-only mount check and directory check.
982 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
983 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
986 if (dvp->v_type != VDIR)
990 * Look it up in the cache. Note that ENOENT is only returned if we
991 * previously entered a negative hit (see later on). The additional
992 * nfsneg_cache_timeout check causes previously cached results to
993 * be instantly ignored if the negative caching is turned off.
995 lockparent = flags & CNP_LOCKPARENT;
996 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
997 nmp = VFSTONFS(dvp->v_mount);
1005 nfsstats.lookupcache_misses++;
1006 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1007 len = cnp->cn_namelen;
1008 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1009 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1010 nfsm_fhtom(dvp, v3);
1011 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1012 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred);
1014 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1018 nfsm_getfh(fhp, fhsize, v3);
1021 * Handle RENAME case...
1023 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1024 if (NFS_CMPFH(np, fhp, fhsize)) {
1028 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1035 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1036 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1038 nfsm_loadattr(newvp, (struct vattr *)0);
1042 VOP_UNLOCK(dvp, 0, td);
1043 cnp->cn_flags |= CNP_PDIRUNLOCK;
1048 if (flags & CNP_ISDOTDOT) {
1049 VOP_UNLOCK(dvp, 0, td);
1050 cnp->cn_flags |= CNP_PDIRUNLOCK;
1051 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1053 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
1054 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1055 return (error); /* NOTE: return error from nget */
1059 error = vn_lock(dvp, LK_EXCLUSIVE, td);
1064 cnp->cn_flags |= CNP_PDIRUNLOCK;
1066 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1070 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1076 VOP_UNLOCK(dvp, 0, td);
1077 cnp->cn_flags |= CNP_PDIRUNLOCK;
1082 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1083 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1085 nfsm_loadattr(newvp, (struct vattr *)0);
1087 /* XXX MOVE TO nfs_nremove() */
1088 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1089 cnp->cn_nameiop != NAMEI_DELETE) {
1090 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1097 if (newvp != NULLVP) {
1101 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1102 cnp->cn_nameiop == NAMEI_RENAME) &&
1105 VOP_UNLOCK(dvp, 0, td);
1106 cnp->cn_flags |= CNP_PDIRUNLOCK;
1108 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1111 error = EJUSTRETURN;
1119 * Just call nfs_bioread() to do the work.
1121 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1122 * struct ucred *a_cred)
1125 nfs_read(struct vop_read_args *ap)
1127 struct vnode *vp = ap->a_vp;
1129 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1130 switch (vp->v_type) {
1132 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1143 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1146 nfs_readlink(struct vop_readlink_args *ap)
1148 struct vnode *vp = ap->a_vp;
1150 if (vp->v_type != VLNK)
1152 return (nfs_bioread(vp, ap->a_uio, 0));
1156 * Do a readlink rpc.
1157 * Called by nfs_doio() from below the buffer cache.
1160 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1165 caddr_t bpos, dpos, cp2;
1166 int error = 0, len, attrflag;
1167 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1168 int v3 = NFS_ISV3(vp);
1170 nfsstats.rpccnt[NFSPROC_READLINK]++;
1171 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1173 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK));
1175 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1177 nfsm_strsiz(len, NFS_MAXPATHLEN);
1178 if (len == NFS_MAXPATHLEN) {
1179 struct nfsnode *np = VTONFS(vp);
1180 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1183 nfsm_mtouio(uiop, len);
1195 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1200 caddr_t bpos, dpos, cp2;
1201 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1202 struct nfsmount *nmp;
1203 int error = 0, len, retlen, tsiz, eof, attrflag;
1204 int v3 = NFS_ISV3(vp);
1209 nmp = VFSTONFS(vp->v_mount);
1210 tsiz = uiop->uio_resid;
1211 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1214 nfsstats.rpccnt[NFSPROC_READ]++;
1215 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1216 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1218 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1220 txdr_hyper(uiop->uio_offset, tl);
1221 *(tl + 2) = txdr_unsigned(len);
1223 *tl++ = txdr_unsigned(uiop->uio_offset);
1224 *tl++ = txdr_unsigned(len);
1227 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ));
1229 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1234 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1235 eof = fxdr_unsigned(int, *(tl + 1));
1237 nfsm_loadattr(vp, (struct vattr *)0);
1238 nfsm_strsiz(retlen, nmp->nm_rsize);
1239 nfsm_mtouio(uiop, retlen);
1243 if (eof || retlen == 0) {
1246 } else if (retlen < len) {
1258 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1262 int32_t t1, t2, backup;
1263 caddr_t bpos, dpos, cp2;
1264 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1265 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1266 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1267 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1270 if (uiop->uio_iovcnt != 1)
1271 panic("nfs: writerpc iovcnt > 1");
1274 tsiz = uiop->uio_resid;
1275 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1278 nfsstats.rpccnt[NFSPROC_WRITE]++;
1279 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1280 nfsm_reqhead(vp, NFSPROC_WRITE,
1281 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1284 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1285 txdr_hyper(uiop->uio_offset, tl);
1287 *tl++ = txdr_unsigned(len);
1288 *tl++ = txdr_unsigned(*iomode);
1289 *tl = txdr_unsigned(len);
1293 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1294 /* Set both "begin" and "current" to non-garbage. */
1295 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1296 *tl++ = x; /* "begin offset" */
1297 *tl++ = x; /* "current offset" */
1298 x = txdr_unsigned(len);
1299 *tl++ = x; /* total to this offset */
1300 *tl = x; /* size of this write */
1302 nfsm_uiotom(uiop, len);
1303 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE));
1306 * The write RPC returns a before and after mtime. The
1307 * nfsm_wcc_data() macro checks the before n_mtime
1308 * against the before time and stores the after time
1309 * in the nfsnode's cached vattr and n_mtime field.
1310 * The NRMODIFIED bit will be set if the before
1311 * time did not match the original mtime.
1313 wccflag = NFSV3_WCCCHK;
1314 nfsm_wcc_data(vp, wccflag);
1316 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1317 + NFSX_V3WRITEVERF);
1318 rlen = fxdr_unsigned(int, *tl++);
1323 } else if (rlen < len) {
1324 backup = len - rlen;
1325 uiop->uio_iov->iov_base -= backup;
1326 uiop->uio_iov->iov_len += backup;
1327 uiop->uio_offset -= backup;
1328 uiop->uio_resid += backup;
1331 commit = fxdr_unsigned(int, *tl++);
1334 * Return the lowest committment level
1335 * obtained by any of the RPCs.
1337 if (committed == NFSV3WRITE_FILESYNC)
1339 else if (committed == NFSV3WRITE_DATASYNC &&
1340 commit == NFSV3WRITE_UNSTABLE)
1342 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1343 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1345 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1346 } else if (bcmp((caddr_t)tl,
1347 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1349 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1354 nfsm_loadattr(vp, (struct vattr *)0);
1362 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1363 committed = NFSV3WRITE_FILESYNC;
1364 *iomode = committed;
1366 uiop->uio_resid = tsiz;
1372 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1373 * mode set to specify the file type and the size field for rdev.
1376 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1379 struct nfsv2_sattr *sp;
1383 struct vnode *newvp = (struct vnode *)0;
1384 struct nfsnode *np = (struct nfsnode *)0;
1388 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1389 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1391 int v3 = NFS_ISV3(dvp);
1393 if (vap->va_type == VCHR || vap->va_type == VBLK)
1394 rdev = txdr_unsigned(vap->va_rdev);
1395 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1398 return (EOPNOTSUPP);
1400 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1403 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1404 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1405 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1406 nfsm_fhtom(dvp, v3);
1407 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1409 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1410 *tl++ = vtonfsv3_type(vap->va_type);
1411 nfsm_v3attrbuild(vap, FALSE);
1412 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1413 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1414 *tl++ = txdr_unsigned(umajor(vap->va_rdev));
1415 *tl = txdr_unsigned(uminor(vap->va_rdev));
1418 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1419 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1420 sp->sa_uid = nfs_xdrneg1;
1421 sp->sa_gid = nfs_xdrneg1;
1423 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1424 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1426 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred);
1428 nfsm_mtofh(dvp, newvp, v3, gotvp);
1432 newvp = (struct vnode *)0;
1434 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1435 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1441 nfsm_wcc_data(dvp, wccflag);
1450 VTONFS(dvp)->n_flag |= NLMODIFIED;
1452 VTONFS(dvp)->n_attrstamp = 0;
1458 * just call nfs_mknodrpc() to do the work.
1460 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1461 * struct componentname *a_cnp, struct vattr *a_vap)
1465 nfs_mknod(struct vop_old_mknod_args *ap)
1467 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1470 static u_long create_verf;
1472 * nfs file create call
1474 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1475 * struct componentname *a_cnp, struct vattr *a_vap)
1478 nfs_create(struct vop_old_create_args *ap)
1480 struct vnode *dvp = ap->a_dvp;
1481 struct vattr *vap = ap->a_vap;
1482 struct componentname *cnp = ap->a_cnp;
1483 struct nfsv2_sattr *sp;
1487 struct nfsnode *np = (struct nfsnode *)0;
1488 struct vnode *newvp = (struct vnode *)0;
1489 caddr_t bpos, dpos, cp2;
1490 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1491 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1493 int v3 = NFS_ISV3(dvp);
1496 * Oops, not for me..
1498 if (vap->va_type == VSOCK)
1499 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1501 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1504 if (vap->va_vaflags & VA_EXCLUSIVE)
1507 nfsstats.rpccnt[NFSPROC_CREATE]++;
1508 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1509 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1510 nfsm_fhtom(dvp, v3);
1511 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1513 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1514 if (fmode & O_EXCL) {
1515 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1516 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1518 if (!TAILQ_EMPTY(&in_ifaddrhead))
1519 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1522 *tl++ = create_verf;
1523 *tl = ++create_verf;
1525 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1526 nfsm_v3attrbuild(vap, FALSE);
1529 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1530 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1531 sp->sa_uid = nfs_xdrneg1;
1532 sp->sa_gid = nfs_xdrneg1;
1534 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1535 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1537 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred);
1539 nfsm_mtofh(dvp, newvp, v3, gotvp);
1543 newvp = (struct vnode *)0;
1545 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1546 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1552 nfsm_wcc_data(dvp, wccflag);
1556 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1562 } else if (v3 && (fmode & O_EXCL)) {
1564 * We are normally called with only a partially initialized
1565 * VAP. Since the NFSv3 spec says that server may use the
1566 * file attributes to store the verifier, the spec requires
1567 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1568 * in atime, but we can't really assume that all servers will
1569 * so we ensure that our SETATTR sets both atime and mtime.
1571 if (vap->va_mtime.tv_sec == VNOVAL)
1572 vfs_timestamp(&vap->va_mtime);
1573 if (vap->va_atime.tv_sec == VNOVAL)
1574 vap->va_atime = vap->va_mtime;
1575 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1579 * The new np may have enough info for access
1580 * checks, make sure rucred and wucred are
1581 * initialized for read and write rpc's.
1584 if (np->n_rucred == NULL)
1585 np->n_rucred = crhold(cnp->cn_cred);
1586 if (np->n_wucred == NULL)
1587 np->n_wucred = crhold(cnp->cn_cred);
1590 VTONFS(dvp)->n_flag |= NLMODIFIED;
1592 VTONFS(dvp)->n_attrstamp = 0;
1597 * nfs file remove call
1598 * To try and make nfs semantics closer to ufs semantics, a file that has
1599 * other processes using the vnode is renamed instead of removed and then
1600 * removed later on the last close.
1601 * - If v_usecount > 1
1602 * If a rename is not already in the works
1603 * call nfs_sillyrename() to set it up
1607 * nfs_remove(struct vnodeop_desc *a_desc, struct vnode *a_dvp,
1608 * struct vnode *a_vp, struct componentname *a_cnp)
1611 nfs_remove(struct vop_old_remove_args *ap)
1613 struct vnode *vp = ap->a_vp;
1614 struct vnode *dvp = ap->a_dvp;
1615 struct componentname *cnp = ap->a_cnp;
1616 struct nfsnode *np = VTONFS(vp);
1621 if (vp->v_usecount < 1)
1622 panic("nfs_remove: bad v_usecount");
1624 if (vp->v_type == VDIR)
1626 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1627 VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 &&
1628 vattr.va_nlink > 1)) {
1630 * throw away biocache buffers, mainly to avoid
1631 * unnecessary delayed writes later.
1633 error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1);
1636 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1637 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1639 * Kludge City: If the first reply to the remove rpc is lost..
1640 * the reply to the retransmitted request will be ENOENT
1641 * since the file was in fact removed
1642 * Therefore, we cheat and return success.
1644 if (error == ENOENT)
1646 } else if (!np->n_sillyrename) {
1647 error = nfs_sillyrename(dvp, vp, cnp);
1649 np->n_attrstamp = 0;
1654 * nfs file remove rpc called from nfs_inactive
1657 nfs_removeit(struct sillyrename *sp)
1659 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1664 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1667 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1668 struct ucred *cred, struct thread *td)
1673 caddr_t bpos, dpos, cp2;
1674 int error = 0, wccflag = NFSV3_WCCRATTR;
1675 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1676 int v3 = NFS_ISV3(dvp);
1678 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1679 nfsm_reqhead(dvp, NFSPROC_REMOVE,
1680 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1681 nfsm_fhtom(dvp, v3);
1682 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1683 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1685 nfsm_wcc_data(dvp, wccflag);
1688 VTONFS(dvp)->n_flag |= NLMODIFIED;
1690 VTONFS(dvp)->n_attrstamp = 0;
1695 * nfs file rename call
1697 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1698 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1699 * struct vnode *a_tvp, struct componentname *a_tcnp)
1702 nfs_rename(struct vop_old_rename_args *ap)
1704 struct vnode *fvp = ap->a_fvp;
1705 struct vnode *tvp = ap->a_tvp;
1706 struct vnode *fdvp = ap->a_fdvp;
1707 struct vnode *tdvp = ap->a_tdvp;
1708 struct componentname *tcnp = ap->a_tcnp;
1709 struct componentname *fcnp = ap->a_fcnp;
1712 /* Check for cross-device rename */
1713 if ((fvp->v_mount != tdvp->v_mount) ||
1714 (tvp && (fvp->v_mount != tvp->v_mount))) {
1720 * We have to flush B_DELWRI data prior to renaming
1721 * the file. If we don't, the delayed-write buffers
1722 * can be flushed out later after the file has gone stale
1723 * under NFSV3. NFSV2 does not have this problem because
1724 * ( as far as I can tell ) it flushes dirty buffers more
1728 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td);
1730 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td);
1733 * If the tvp exists and is in use, sillyrename it before doing the
1734 * rename of the new file over it.
1736 * XXX Can't sillyrename a directory.
1738 * We do not attempt to do any namecache purges in this old API
1739 * routine. The new API compat functions have access to the actual
1740 * namecache structures and will do it for us.
1742 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1743 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1750 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1751 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1764 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1766 if (error == ENOENT)
1772 * nfs file rename rpc called from nfs_remove() above
1775 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1776 struct sillyrename *sp)
1778 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1779 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1783 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1786 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1787 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1788 struct ucred *cred, struct thread *td)
1793 caddr_t bpos, dpos, cp2;
1794 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1795 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1796 int v3 = NFS_ISV3(fdvp);
1798 nfsstats.rpccnt[NFSPROC_RENAME]++;
1799 nfsm_reqhead(fdvp, NFSPROC_RENAME,
1800 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1801 nfsm_rndup(tnamelen));
1802 nfsm_fhtom(fdvp, v3);
1803 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1804 nfsm_fhtom(tdvp, v3);
1805 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1806 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1808 nfsm_wcc_data(fdvp, fwccflag);
1809 nfsm_wcc_data(tdvp, twccflag);
1813 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1814 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1816 VTONFS(fdvp)->n_attrstamp = 0;
1818 VTONFS(tdvp)->n_attrstamp = 0;
1823 * nfs hard link create call
1825 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1826 * struct componentname *a_cnp)
1829 nfs_link(struct vop_old_link_args *ap)
1831 struct vnode *vp = ap->a_vp;
1832 struct vnode *tdvp = ap->a_tdvp;
1833 struct componentname *cnp = ap->a_cnp;
1837 caddr_t bpos, dpos, cp2;
1838 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1839 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1842 if (vp->v_mount != tdvp->v_mount) {
1847 * Push all writes to the server, so that the attribute cache
1848 * doesn't get "out of sync" with the server.
1849 * XXX There should be a better way!
1851 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td);
1854 nfsstats.rpccnt[NFSPROC_LINK]++;
1855 nfsm_reqhead(vp, NFSPROC_LINK,
1856 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1858 nfsm_fhtom(tdvp, v3);
1859 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1860 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred);
1862 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1863 nfsm_wcc_data(tdvp, wccflag);
1867 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1869 VTONFS(vp)->n_attrstamp = 0;
1871 VTONFS(tdvp)->n_attrstamp = 0;
1873 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1875 if (error == EEXIST)
1881 * nfs symbolic link create call
1883 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1884 * struct componentname *a_cnp, struct vattr *a_vap,
1888 nfs_symlink(struct vop_old_symlink_args *ap)
1890 struct vnode *dvp = ap->a_dvp;
1891 struct vattr *vap = ap->a_vap;
1892 struct componentname *cnp = ap->a_cnp;
1893 struct nfsv2_sattr *sp;
1897 caddr_t bpos, dpos, cp2;
1898 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1899 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1900 struct vnode *newvp = (struct vnode *)0;
1901 int v3 = NFS_ISV3(dvp);
1903 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1904 slen = strlen(ap->a_target);
1905 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1906 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1907 nfsm_fhtom(dvp, v3);
1908 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1910 nfsm_v3attrbuild(vap, FALSE);
1912 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1914 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1915 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1916 sp->sa_uid = nfs_xdrneg1;
1917 sp->sa_gid = nfs_xdrneg1;
1918 sp->sa_size = nfs_xdrneg1;
1919 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1920 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1924 * Issue the NFS request and get the rpc response.
1926 * Only NFSv3 responses returning an error of 0 actually return
1927 * a file handle that can be converted into newvp without having
1928 * to do an extra lookup rpc.
1930 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred);
1933 nfsm_mtofh(dvp, newvp, v3, gotvp);
1934 nfsm_wcc_data(dvp, wccflag);
1938 * out code jumps -> here, mrep is also freed.
1945 * If we get an EEXIST error, silently convert it to no-error
1946 * in case of an NFS retry.
1948 if (error == EEXIST)
1952 * If we do not have (or no longer have) an error, and we could
1953 * not extract the newvp from the response due to the request being
1954 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1955 * to obtain a newvp to return.
1957 if (error == 0 && newvp == NULL) {
1958 struct nfsnode *np = NULL;
1960 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1961 cnp->cn_cred, cnp->cn_td, &np);
1971 VTONFS(dvp)->n_flag |= NLMODIFIED;
1973 VTONFS(dvp)->n_attrstamp = 0;
1980 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1981 * struct componentname *a_cnp, struct vattr *a_vap)
1984 nfs_mkdir(struct vop_old_mkdir_args *ap)
1986 struct vnode *dvp = ap->a_dvp;
1987 struct vattr *vap = ap->a_vap;
1988 struct componentname *cnp = ap->a_cnp;
1989 struct nfsv2_sattr *sp;
1994 struct nfsnode *np = (struct nfsnode *)0;
1995 struct vnode *newvp = (struct vnode *)0;
1996 caddr_t bpos, dpos, cp2;
1997 int error = 0, wccflag = NFSV3_WCCRATTR;
1999 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2001 int v3 = NFS_ISV3(dvp);
2003 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
2006 len = cnp->cn_namelen;
2007 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2008 nfsm_reqhead(dvp, NFSPROC_MKDIR,
2009 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2010 nfsm_fhtom(dvp, v3);
2011 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2013 nfsm_v3attrbuild(vap, FALSE);
2015 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2016 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2017 sp->sa_uid = nfs_xdrneg1;
2018 sp->sa_gid = nfs_xdrneg1;
2019 sp->sa_size = nfs_xdrneg1;
2020 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2021 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2023 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred);
2025 nfsm_mtofh(dvp, newvp, v3, gotvp);
2027 nfsm_wcc_data(dvp, wccflag);
2030 VTONFS(dvp)->n_flag |= NLMODIFIED;
2032 VTONFS(dvp)->n_attrstamp = 0;
2034 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2035 * if we can succeed in looking up the directory.
2037 if (error == EEXIST || (!error && !gotvp)) {
2040 newvp = (struct vnode *)0;
2042 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2046 if (newvp->v_type != VDIR)
2059 * nfs remove directory call
2061 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2062 * struct componentname *a_cnp)
2065 nfs_rmdir(struct vop_old_rmdir_args *ap)
2067 struct vnode *vp = ap->a_vp;
2068 struct vnode *dvp = ap->a_dvp;
2069 struct componentname *cnp = ap->a_cnp;
2073 caddr_t bpos, dpos, cp2;
2074 int error = 0, wccflag = NFSV3_WCCRATTR;
2075 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2076 int v3 = NFS_ISV3(dvp);
2080 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2081 nfsm_reqhead(dvp, NFSPROC_RMDIR,
2082 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2083 nfsm_fhtom(dvp, v3);
2084 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2085 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred);
2087 nfsm_wcc_data(dvp, wccflag);
2090 VTONFS(dvp)->n_flag |= NLMODIFIED;
2092 VTONFS(dvp)->n_attrstamp = 0;
2094 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2096 if (error == ENOENT)
2104 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2107 nfs_readdir(struct vop_readdir_args *ap)
2109 struct vnode *vp = ap->a_vp;
2110 struct nfsnode *np = VTONFS(vp);
2111 struct uio *uio = ap->a_uio;
2115 if (vp->v_type != VDIR)
2119 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2120 * and then check that is still valid, or if this is an NQNFS mount
2121 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2122 * VOP_GETATTR() does not necessarily go to the wire.
2124 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2125 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2126 if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 &&
2127 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2129 nfsstats.direofcache_hits++;
2135 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2136 * own cache coherency checks so we do not have to.
2138 tresid = uio->uio_resid;
2139 error = nfs_bioread(vp, uio, 0);
2141 if (!error && uio->uio_resid == tresid)
2142 nfsstats.direofcache_misses++;
2147 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2149 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2150 * offset/block and converts the nfs formatted directory entries for userland
2151 * consumption as well as deals with offsets into the middle of blocks.
2152 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2153 * be block-bounded. It must convert to cookies for the actual RPC.
2156 nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2159 struct nfs_dirent *dp = NULL;
2164 caddr_t bpos, dpos, cp2;
2165 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2167 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2168 struct nfsnode *dnp = VTONFS(vp);
2170 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2172 int v3 = NFS_ISV3(vp);
2175 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2176 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2177 panic("nfs readdirrpc bad uio");
2181 * If there is no cookie, assume directory was stale.
2183 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2187 return (NFSERR_BAD_COOKIE);
2189 * Loop around doing readdir rpc's of size nm_readdirsize
2190 * truncated to a multiple of DIRBLKSIZ.
2191 * The stopping criteria is EOF or buffer full.
2193 while (more_dirs && bigenough) {
2194 nfsstats.rpccnt[NFSPROC_READDIR]++;
2195 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2199 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2200 *tl++ = cookie.nfsuquad[0];
2201 *tl++ = cookie.nfsuquad[1];
2202 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2203 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2205 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2206 *tl++ = cookie.nfsuquad[0];
2208 *tl = txdr_unsigned(nmp->nm_readdirsize);
2209 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2211 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2213 nfsm_dissect(tl, u_int32_t *,
2215 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2216 dnp->n_cookieverf.nfsuquad[1] = *tl;
2222 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2223 more_dirs = fxdr_unsigned(int, *tl);
2225 /* loop thru the dir entries, converting them to std form */
2226 while (more_dirs && bigenough) {
2228 nfsm_dissect(tl, u_int32_t *,
2230 fileno = fxdr_hyper(tl);
2231 len = fxdr_unsigned(int, *(tl + 2));
2233 nfsm_dissect(tl, u_int32_t *,
2235 fileno = fxdr_unsigned(u_quad_t, *tl++);
2236 len = fxdr_unsigned(int, *tl);
2238 if (len <= 0 || len > NFS_MAXNAMLEN) {
2245 * len is the number of bytes in the path element
2246 * name, not including the \0 termination.
2248 * tlen is the number of bytes w have to reserve for
2249 * the path element name.
2251 tlen = nfsm_rndup(len);
2253 tlen += 4; /* To ensure null termination */
2256 * If the entry would cross a DIRBLKSIZ boundary,
2257 * extend the previous nfs_dirent to cover the
2260 left = DIRBLKSIZ - blksiz;
2261 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2262 dp->nfs_reclen += left;
2263 uiop->uio_iov->iov_base += left;
2264 uiop->uio_iov->iov_len -= left;
2265 uiop->uio_offset += left;
2266 uiop->uio_resid -= left;
2269 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2272 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2273 dp->nfs_ino = fileno;
2274 dp->nfs_namlen = len;
2275 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2276 dp->nfs_type = DT_UNKNOWN;
2277 blksiz += dp->nfs_reclen;
2278 if (blksiz == DIRBLKSIZ)
2280 uiop->uio_offset += sizeof(struct nfs_dirent);
2281 uiop->uio_resid -= sizeof(struct nfs_dirent);
2282 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent);
2283 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2284 nfsm_mtouio(uiop, len);
2287 * The uiop has advanced by nfs_dirent + len
2288 * but really needs to advance by
2291 cp = uiop->uio_iov->iov_base;
2293 *cp = '\0'; /* null terminate */
2294 uiop->uio_iov->iov_base += tlen;
2295 uiop->uio_iov->iov_len -= tlen;
2296 uiop->uio_offset += tlen;
2297 uiop->uio_resid -= tlen;
2300 * NFS strings must be rounded up (nfsm_myouio
2301 * handled that in the bigenough case).
2303 nfsm_adv(nfsm_rndup(len));
2306 nfsm_dissect(tl, u_int32_t *,
2309 nfsm_dissect(tl, u_int32_t *,
2314 * If we were able to accomodate the last entry,
2315 * get the cookie for the next one. Otherwise
2316 * hold-over the cookie for the one we were not
2317 * able to accomodate.
2320 cookie.nfsuquad[0] = *tl++;
2322 cookie.nfsuquad[1] = *tl++;
2328 more_dirs = fxdr_unsigned(int, *tl);
2331 * If at end of rpc data, get the eof boolean
2334 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2335 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2340 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2341 * by increasing d_reclen for the last record.
2344 left = DIRBLKSIZ - blksiz;
2345 dp->nfs_reclen += left;
2346 uiop->uio_iov->iov_base += left;
2347 uiop->uio_iov->iov_len -= left;
2348 uiop->uio_offset += left;
2349 uiop->uio_resid -= left;
2354 * We hit the end of the directory, update direofoffset.
2356 dnp->n_direofoffset = uiop->uio_offset;
2359 * There is more to go, insert the link cookie so the
2360 * next block can be read.
2362 if (uiop->uio_resid > 0)
2363 printf("EEK! readdirrpc resid > 0\n");
2364 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2372 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2375 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2378 struct nfs_dirent *dp;
2382 struct vnode *newvp;
2384 caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2385 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2387 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2388 struct nfsnode *dnp = VTONFS(vp), *np;
2391 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2392 int attrflag, fhsize;
2393 struct namecache *ncp;
2394 struct namecache *dncp;
2395 struct nlcomponent nlc;
2401 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2402 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2403 panic("nfs readdirplusrpc bad uio");
2406 * Obtain the namecache record for the directory so we have something
2407 * to use as a basis for creating the entries. This function will
2408 * return a held (but not locked) ncp. The ncp may be disconnected
2409 * from the tree and cannot be used for upward traversals, and the
2410 * ncp may be unnamed. Note that other unrelated operations may
2411 * cause the ncp to be named at any time.
2413 dncp = cache_fromdvp(vp, NULL, 0);
2414 bzero(&nlc, sizeof(nlc));
2418 * If there is no cookie, assume directory was stale.
2420 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2424 return (NFSERR_BAD_COOKIE);
2426 * Loop around doing readdir rpc's of size nm_readdirsize
2427 * truncated to a multiple of DIRBLKSIZ.
2428 * The stopping criteria is EOF or buffer full.
2430 while (more_dirs && bigenough) {
2431 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2432 nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2433 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2435 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2436 *tl++ = cookie.nfsuquad[0];
2437 *tl++ = cookie.nfsuquad[1];
2438 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2439 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2440 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2441 *tl = txdr_unsigned(nmp->nm_rsize);
2442 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2443 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2448 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2449 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2450 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2451 more_dirs = fxdr_unsigned(int, *tl);
2453 /* loop thru the dir entries, doctoring them to 4bsd form */
2454 while (more_dirs && bigenough) {
2455 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2456 fileno = fxdr_hyper(tl);
2457 len = fxdr_unsigned(int, *(tl + 2));
2458 if (len <= 0 || len > NFS_MAXNAMLEN) {
2463 tlen = nfsm_rndup(len);
2465 tlen += 4; /* To ensure null termination*/
2466 left = DIRBLKSIZ - blksiz;
2467 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2468 dp->nfs_reclen += left;
2469 uiop->uio_iov->iov_base += left;
2470 uiop->uio_iov->iov_len -= left;
2471 uiop->uio_offset += left;
2472 uiop->uio_resid -= left;
2475 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2478 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2479 dp->nfs_ino = fileno;
2480 dp->nfs_namlen = len;
2481 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2482 dp->nfs_type = DT_UNKNOWN;
2483 blksiz += dp->nfs_reclen;
2484 if (blksiz == DIRBLKSIZ)
2486 uiop->uio_offset += sizeof(struct nfs_dirent);
2487 uiop->uio_resid -= sizeof(struct nfs_dirent);
2488 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent);
2489 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2490 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2491 nlc.nlc_namelen = len;
2492 nfsm_mtouio(uiop, len);
2493 cp = uiop->uio_iov->iov_base;
2496 uiop->uio_iov->iov_base += tlen;
2497 uiop->uio_iov->iov_len -= tlen;
2498 uiop->uio_offset += tlen;
2499 uiop->uio_resid -= tlen;
2501 nfsm_adv(nfsm_rndup(len));
2502 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2504 cookie.nfsuquad[0] = *tl++;
2505 cookie.nfsuquad[1] = *tl++;
2510 * Since the attributes are before the file handle
2511 * (sigh), we must skip over the attributes and then
2512 * come back and get them.
2514 attrflag = fxdr_unsigned(int, *tl);
2518 nfsm_adv(NFSX_V3FATTR);
2519 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2520 doit = fxdr_unsigned(int, *tl);
2522 nfsm_getfh(fhp, fhsize, 1);
2523 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2528 error = nfs_nget(vp->v_mount, fhp,
2536 if (doit && bigenough) {
2541 nfsm_loadattr(newvp, (struct vattr *)0);
2545 IFTODT(VTTOIF(np->n_vattr.va_type));
2547 printf("NFS/READDIRPLUS, ENTER %*.*s\n",
2548 nlc.nlc_namelen, nlc.nlc_namelen,
2550 ncp = cache_nlookup(dncp, &nlc);
2551 cache_setunresolved(ncp);
2552 cache_setvp(ncp, newvp);
2555 printf("NFS/READDIRPLUS, UNABLE TO ENTER"
2557 nlc.nlc_namelen, nlc.nlc_namelen,
2562 /* Just skip over the file handle */
2563 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2564 i = fxdr_unsigned(int, *tl);
2565 nfsm_adv(nfsm_rndup(i));
2567 if (newvp != NULLVP) {
2574 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2575 more_dirs = fxdr_unsigned(int, *tl);
2578 * If at end of rpc data, get the eof boolean
2581 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2582 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2587 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2588 * by increasing d_reclen for the last record.
2591 left = DIRBLKSIZ - blksiz;
2592 dp->nfs_reclen += left;
2593 uiop->uio_iov->iov_base += left;
2594 uiop->uio_iov->iov_len -= left;
2595 uiop->uio_offset += left;
2596 uiop->uio_resid -= left;
2600 * We are now either at the end of the directory or have filled the
2604 dnp->n_direofoffset = uiop->uio_offset;
2606 if (uiop->uio_resid > 0)
2607 printf("EEK! readdirplusrpc resid > 0\n");
2608 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2612 if (newvp != NULLVP) {
2625 * Silly rename. To make the NFS filesystem that is stateless look a little
2626 * more like the "ufs" a remove of an active vnode is translated to a rename
2627 * to a funny looking filename that is removed by nfs_inactive on the
2628 * nfsnode. There is the potential for another process on a different client
2629 * to create the same funny name between the nfs_lookitup() fails and the
2630 * nfs_rename() completes, but...
2633 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2635 struct sillyrename *sp;
2640 * We previously purged dvp instead of vp. I don't know why, it
2641 * completely destroys performance. We can't do it anyway with the
2642 * new VFS API since we would be breaking the namecache topology.
2644 cache_purge(vp); /* XXX */
2647 if (vp->v_type == VDIR)
2648 panic("nfs: sillyrename dir");
2650 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2651 M_NFSREQ, M_WAITOK);
2652 sp->s_cred = crdup(cnp->cn_cred);
2656 /* Fudge together a funny name */
2657 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td);
2659 /* Try lookitups until we get one that isn't there */
2660 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2661 cnp->cn_td, (struct nfsnode **)0) == 0) {
2663 if (sp->s_name[4] > 'z') {
2668 error = nfs_renameit(dvp, cnp, sp);
2671 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2673 np->n_sillyrename = sp;
2678 free((caddr_t)sp, M_NFSREQ);
2683 * Look up a file name and optionally either update the file handle or
2684 * allocate an nfsnode, depending on the value of npp.
2685 * npp == NULL --> just do the lookup
2686 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2688 * *npp != NULL --> update the file handle in the vnode
2691 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2692 struct thread *td, struct nfsnode **npp)
2697 struct vnode *newvp = (struct vnode *)0;
2698 struct nfsnode *np, *dnp = VTONFS(dvp);
2699 caddr_t bpos, dpos, cp2;
2700 int error = 0, fhlen, attrflag;
2701 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2703 int v3 = NFS_ISV3(dvp);
2705 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2706 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2707 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2708 nfsm_fhtom(dvp, v3);
2709 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2710 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2711 if (npp && !error) {
2712 nfsm_getfh(nfhp, fhlen, v3);
2715 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2716 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2717 np->n_fhp = &np->n_fh;
2718 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2719 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2720 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2721 np->n_fhsize = fhlen;
2723 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2727 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2735 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
2736 if (!attrflag && *npp == NULL) {
2745 nfsm_loadattr(newvp, (struct vattr *)0);
2749 if (npp && *npp == NULL) {
2764 * Nfs Version 3 commit rpc
2767 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2772 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2773 caddr_t bpos, dpos, cp2;
2774 int error = 0, wccflag = NFSV3_WCCRATTR;
2775 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2777 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2779 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2780 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2782 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2783 txdr_hyper(offset, tl);
2785 *tl = txdr_unsigned(cnt);
2786 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE));
2787 nfsm_wcc_data(vp, wccflag);
2789 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2790 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2791 NFSX_V3WRITEVERF)) {
2792 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2794 error = NFSERR_STALEWRITEVERF;
2804 * - make nfs_bmap() essentially a no-op that does no translation
2805 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2806 * (Maybe I could use the process's page mapping, but I was concerned that
2807 * Kernel Write might not be enabled and also figured copyout() would do
2808 * a lot more work than bcopy() and also it currently happens in the
2809 * context of the swapper process (2).
2811 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, struct vnode **a_vpp,
2812 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2815 nfs_bmap(struct vop_bmap_args *ap)
2817 struct vnode *vp = ap->a_vp;
2819 if (ap->a_vpp != NULL)
2821 if (ap->a_doffsetp != NULL)
2822 *ap->a_doffsetp = ap->a_loffset;
2823 if (ap->a_runp != NULL)
2825 if (ap->a_runb != NULL)
2833 * For async requests when nfsiod(s) are running, queue the request by
2834 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2838 nfs_strategy(struct vop_strategy_args *ap)
2840 struct bio *bio = ap->a_bio;
2842 struct buf *bp = bio->bio_buf;
2846 KASSERT(!(bp->b_flags & B_DONE),
2847 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2848 KASSERT(BUF_REFCNT(bp) > 0,
2849 ("nfs_strategy: buffer %p not locked", bp));
2851 if (bp->b_flags & B_ASYNC)
2854 td = curthread; /* XXX */
2857 * We probably don't need to push an nbio any more since no
2858 * block conversion is required due to the use of 64 bit byte
2859 * offsets, but do it anyway.
2861 nbio = push_bio(bio);
2862 nbio->bio_offset = bio->bio_offset;
2865 * If the op is asynchronous and an i/o daemon is waiting
2866 * queue the request, wake it up and wait for completion
2867 * otherwise just do it ourselves.
2869 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(ap->a_vp, nbio, td))
2870 error = nfs_doio(ap->a_vp, nbio, td);
2877 * NB Currently unsupported.
2879 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred,
2880 * struct thread *a_td)
2884 nfs_mmap(struct vop_mmap_args *ap)
2890 * fsync vnode op. Just call nfs_flush() with commit == 1.
2892 * nfs_fsync(struct vnodeop_desc *a_desc, struct vnode *a_vp,
2893 * struct ucred * a_cred, int a_waitfor, struct thread *a_td)
2897 nfs_fsync(struct vop_fsync_args *ap)
2899 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2903 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2904 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2905 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2906 * set the buffer contains data that has already been written to the server
2907 * and which now needs a commit RPC.
2909 * If commit is 0 we only take one pass and only flush buffers containing new
2912 * If commit is 1 we take two passes, issuing a commit RPC in the second
2915 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
2916 * to completely flush all pending data.
2918 * Note that the RB_SCAN code properly handles the case where the
2919 * callback might block and directly or indirectly (another thread) cause
2920 * the RB tree to change.
2923 #ifndef NFS_COMMITBVECSIZ
2924 #define NFS_COMMITBVECSIZ 16
2927 struct nfs_flush_info {
2928 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
2935 struct buf *bvary[NFS_COMMITBVECSIZ];
2941 static int nfs_flush_bp(struct buf *bp, void *data);
2942 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
2945 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
2947 struct nfsnode *np = VTONFS(vp);
2948 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2949 struct nfs_flush_info info;
2952 bzero(&info, sizeof(info));
2955 info.waitfor = waitfor;
2956 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
2963 info.mode = NFI_FLUSHNEW;
2964 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2965 nfs_flush_bp, &info);
2968 * Take a second pass if committing and no error occured.
2969 * Clean up any left over collection (whether an error
2972 if (commit && error == 0) {
2973 info.mode = NFI_COMMIT;
2974 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2975 nfs_flush_bp, &info);
2977 error = nfs_flush_docommit(&info, error);
2981 * Wait for pending I/O to complete before checking whether
2982 * any further dirty buffers exist.
2984 while (waitfor == MNT_WAIT && vp->v_track_write.bk_active) {
2985 vp->v_track_write.bk_waitflag = 1;
2986 error = tsleep(&vp->v_track_write,
2987 info.slpflag, "nfsfsync", info.slptimeo);
2990 * We have to be able to break out if this
2991 * is an 'intr' mount.
2993 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
2999 * Since we do not process pending signals,
3000 * once we get a PCATCH our tsleep() will no
3001 * longer sleep, switch to a fixed timeout
3004 if (info.slpflag == PCATCH) {
3006 info.slptimeo = 2 * hz;
3013 * Loop if we are flushing synchronous as well as committing,
3014 * and dirty buffers are still present. Otherwise we might livelock.
3016 } while (waitfor == MNT_WAIT && commit &&
3017 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3020 * The callbacks have to return a negative error to terminate the
3027 * Deal with any error collection
3029 if (np->n_flag & NWRITEERR) {
3030 error = np->n_error;
3031 np->n_flag &= ~NWRITEERR;
3039 nfs_flush_bp(struct buf *bp, void *data)
3041 struct nfs_flush_info *info = data;
3046 switch(info->mode) {
3049 if (info->loops && info->waitfor == MNT_WAIT) {
3050 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3052 int lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3053 if (info->slpflag & PCATCH)
3054 lkflags |= LK_PCATCH;
3055 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3059 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3062 KKASSERT(bp->b_vp == info->vp);
3064 if ((bp->b_flags & B_DELWRI) == 0)
3065 panic("nfs_fsync: not dirty");
3066 if (bp->b_flags & B_NEEDCOMMIT) {
3073 bp->b_flags |= B_ASYNC;
3075 VOP_BWRITE(bp->b_vp, bp);
3083 * Only process buffers in need of a commit which we can
3084 * immediately lock. This may prevent a buffer from being
3085 * committed, but the normal flush loop will block on the
3086 * same buffer so we shouldn't get into an endless loop.
3089 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3090 (B_DELWRI | B_NEEDCOMMIT) ||
3091 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
3096 KKASSERT(bp->b_vp == info->vp);
3100 * NOTE: we are not clearing B_DONE here, so we have
3101 * to do it later on in this routine if we intend to
3102 * initiate I/O on the bp.
3104 * Note: to avoid loopback deadlocks, we do not
3105 * assign b_runningbufspace.
3107 vfs_busy_pages(bp, 1);
3109 info->bvary[info->bvsize] = bp;
3110 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3111 if (info->bvsize == 0 || toff < info->beg_off)
3112 info->beg_off = toff;
3113 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3114 if (info->bvsize == 0 || toff > info->end_off)
3115 info->end_off = toff;
3117 if (info->bvsize == NFS_COMMITBVECSIZ) {
3118 error = nfs_flush_docommit(info, 0);
3119 KKASSERT(info->bvsize == 0);
3128 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3138 if (info->bvsize > 0) {
3140 * Commit data on the server, as required. Note that
3141 * nfs_commit will use the vnode's cred for the commit.
3142 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3144 bytes = info->end_off - info->beg_off;
3145 if (bytes > 0x40000000)
3150 retv = nfs_commit(vp, info->beg_off,
3151 (int)bytes, info->td);
3152 if (retv == NFSERR_STALEWRITEVERF)
3153 nfs_clearcommit(vp->v_mount);
3157 * Now, either mark the blocks I/O done or mark the
3158 * blocks dirty, depending on whether the commit
3161 for (i = 0; i < info->bvsize; ++i) {
3162 bp = info->bvary[i];
3163 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3166 * Error, leave B_DELWRI intact
3168 vfs_unbusy_pages(bp);
3172 * Success, remove B_DELWRI ( bundirty() ).
3174 * b_dirtyoff/b_dirtyend seem to be NFS
3175 * specific. We should probably move that
3176 * into bundirty(). XXX
3178 * We are faking an I/O write, we have to
3179 * start the transaction in order to
3180 * immediately biodone() it.
3183 bp->b_flags |= B_ASYNC;
3185 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3186 bp->b_dirtyoff = bp->b_dirtyend = 0;
3188 biodone(&bp->b_bio1);
3197 * NFS advisory byte-level locks.
3198 * Currently unsupported.
3200 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3204 nfs_advlock(struct vop_advlock_args *ap)
3206 struct nfsnode *np = VTONFS(ap->a_vp);
3209 * The following kludge is to allow diskless support to work
3210 * until a real NFS lockd is implemented. Basically, just pretend
3211 * that this is a local lock.
3213 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3217 * Print out the contents of an nfsnode.
3219 * nfs_print(struct vnode *a_vp)
3222 nfs_print(struct vop_print_args *ap)
3224 struct vnode *vp = ap->a_vp;
3225 struct nfsnode *np = VTONFS(vp);
3227 printf("tag VT_NFS, fileid %ld fsid 0x%x",
3228 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3229 if (vp->v_type == VFIFO)
3236 * Just call nfs_writebp() with the force argument set to 1.
3238 * NOTE: B_DONE may or may not be set in a_bp on call.
3240 * nfs_bwrite(struct vnode *a_bp)
3243 nfs_bwrite(struct vop_bwrite_args *ap)
3245 return (nfs_writebp(ap->a_bp, 1, curthread));
3249 * This is a clone of vn_bwrite(), except that it also handles the
3250 * B_NEEDCOMMIT flag. We set B_CACHE if this is a VMIO buffer.
3253 nfs_writebp(struct buf *bp, int force, struct thread *td)
3257 if (BUF_REFCNT(bp) == 0)
3258 panic("bwrite: buffer is not locked???");
3260 if (bp->b_flags & B_INVAL) {
3265 bp->b_flags |= B_CACHE;
3268 * Undirty the bp. We will redirty it later if the I/O fails.
3272 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3276 * Note: to avoid loopback deadlocks, we do not
3277 * assign b_runningbufspace.
3279 vfs_busy_pages(bp, 1);
3282 if (bp->b_flags & B_ASYNC) {
3283 vn_strategy(bp->b_vp, &bp->b_bio1);
3286 vn_strategy(bp->b_vp, &bp->b_bio1);
3287 error = biowait(bp);
3294 * nfs special file access vnode op.
3295 * Essentially just get vattr and then imitate iaccess() since the device is
3296 * local to the client.
3298 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
3299 * struct thread *a_td)
3302 nfsspec_access(struct vop_access_args *ap)
3306 struct ucred *cred = ap->a_cred;
3307 struct vnode *vp = ap->a_vp;
3308 mode_t mode = ap->a_mode;
3314 * Disallow write attempts on filesystems mounted read-only;
3315 * unless the file is a socket, fifo, or a block or character
3316 * device resident on the filesystem.
3318 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3319 switch (vp->v_type) {
3329 * If you're the super-user,
3330 * you always get access.
3332 if (cred->cr_uid == 0)
3335 error = VOP_GETATTR(vp, vap, ap->a_td);
3339 * Access check is based on only one of owner, group, public.
3340 * If not owner, then check group. If not a member of the
3341 * group, then check public access.
3343 if (cred->cr_uid != vap->va_uid) {
3345 gp = cred->cr_groups;
3346 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3347 if (vap->va_gid == *gp)
3353 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3358 * Read wrapper for special devices.
3360 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3361 * struct ucred *a_cred)
3364 nfsspec_read(struct vop_read_args *ap)
3366 struct nfsnode *np = VTONFS(ap->a_vp);
3372 getnanotime(&np->n_atim);
3373 return (VOCALL(spec_vnode_vops, &ap->a_head));
3377 * Write wrapper for special devices.
3379 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3380 * struct ucred *a_cred)
3383 nfsspec_write(struct vop_write_args *ap)
3385 struct nfsnode *np = VTONFS(ap->a_vp);
3391 getnanotime(&np->n_mtim);
3392 return (VOCALL(spec_vnode_vops, &ap->a_head));
3396 * Close wrapper for special devices.
3398 * Update the times on the nfsnode then do device close.
3400 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
3401 * struct thread *a_td)
3404 nfsspec_close(struct vop_close_args *ap)
3406 struct vnode *vp = ap->a_vp;
3407 struct nfsnode *np = VTONFS(vp);
3410 if (np->n_flag & (NACC | NUPD)) {
3412 if (vp->v_usecount == 1 &&
3413 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3415 if (np->n_flag & NACC)
3416 vattr.va_atime = np->n_atim;
3417 if (np->n_flag & NUPD)
3418 vattr.va_mtime = np->n_mtim;
3419 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
3422 return (VOCALL(spec_vnode_vops, &ap->a_head));
3426 * Read wrapper for fifos.
3428 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3429 * struct ucred *a_cred)
3432 nfsfifo_read(struct vop_read_args *ap)
3434 struct nfsnode *np = VTONFS(ap->a_vp);
3440 getnanotime(&np->n_atim);
3441 return (VOCALL(fifo_vnode_vops, &ap->a_head));
3445 * Write wrapper for fifos.
3447 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3448 * struct ucred *a_cred)
3451 nfsfifo_write(struct vop_write_args *ap)
3453 struct nfsnode *np = VTONFS(ap->a_vp);
3459 getnanotime(&np->n_mtim);
3460 return (VOCALL(fifo_vnode_vops, &ap->a_head));
3464 * Close wrapper for fifos.
3466 * Update the times on the nfsnode then do fifo close.
3468 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td)
3471 nfsfifo_close(struct vop_close_args *ap)
3473 struct vnode *vp = ap->a_vp;
3474 struct nfsnode *np = VTONFS(vp);
3478 if (np->n_flag & (NACC | NUPD)) {
3480 if (np->n_flag & NACC)
3482 if (np->n_flag & NUPD)
3485 if (vp->v_usecount == 1 &&
3486 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3488 if (np->n_flag & NACC)
3489 vattr.va_atime = np->n_atim;
3490 if (np->n_flag & NUPD)
3491 vattr.va_mtime = np->n_mtim;
3492 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
3495 return (VOCALL(fifo_vnode_vops, &ap->a_head));