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 1.54 2006/04/01 20:46:53 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 * Regular files are mmapable and we use the buffer cache. We also
490 * use the buffer cache for directories internally, so those
491 * vnodes need a VM object.
493 if (vp->v_type == VREG || vp->v_type == VDIR)
497 * Clear the attribute cache only if opening with write access. It
498 * is unclear if we should do this at all here, but we certainly
499 * should not clear the cache unconditionally simply because a file
502 if (ap->a_mode & FWRITE)
506 * For normal NFS, reconcile changes made locally verses
507 * changes made remotely. Note that VOP_GETATTR only goes
508 * to the wire if the cached attribute has timed out or been
511 * If local modifications have been made clear the attribute
512 * cache to force an attribute and modified time check. If
513 * GETATTR detects that the file has been changed by someone
514 * other then us it will set NRMODIFIED.
516 * If we are opening a directory and local changes have been
517 * made we have to invalidate the cache in order to ensure
518 * that we get the most up-to-date information from the
521 if (np->n_flag & NLMODIFIED) {
523 if (vp->v_type == VDIR) {
524 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
530 error = VOP_GETATTR(vp, &vattr, ap->a_td);
533 if (np->n_flag & NRMODIFIED) {
534 if (vp->v_type == VDIR)
536 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
539 np->n_flag &= ~NRMODIFIED;
542 return (vop_stdopen(ap));
547 * What an NFS client should do upon close after writing is a debatable issue.
548 * Most NFS clients push delayed writes to the server upon close, basically for
550 * 1 - So that any write errors may be reported back to the client process
551 * doing the close system call. By far the two most likely errors are
552 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
553 * 2 - To put a worst case upper bound on cache inconsistency between
554 * multiple clients for the file.
555 * There is also a consistency problem for Version 2 of the protocol w.r.t.
556 * not being able to tell if other clients are writing a file concurrently,
557 * since there is no way of knowing if the changed modify time in the reply
558 * is only due to the write for this client.
559 * (NFS Version 3 provides weak cache consistency data in the reply that
560 * should be sufficient to detect and handle this case.)
562 * The current code does the following:
563 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
564 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
565 * or commit them (this satisfies 1 and 2 except for the
566 * case where the server crashes after this close but
567 * before the commit RPC, which is felt to be "good
568 * enough". Changing the last argument to nfs_flush() to
569 * a 1 would force a commit operation, if it is felt a
570 * commit is necessary now.
571 * for NQNFS - do nothing now, since 2 is dealt with via leases and
572 * 1 should be dealt with via an fsync() system call for
573 * cases where write errors are important.
575 * nfs_close(struct vnodeop_desc *a_desc, struct vnode *a_vp, int a_fflag,
576 * struct ucred *a_cred, struct thread *a_td)
580 nfs_close(struct vop_close_args *ap)
582 struct vnode *vp = ap->a_vp;
583 struct nfsnode *np = VTONFS(vp);
586 if (vp->v_type == VREG) {
587 if (np->n_flag & NLMODIFIED) {
590 * Under NFSv3 we have dirty buffers to dispose of. We
591 * must flush them to the NFS server. We have the option
592 * of waiting all the way through the commit rpc or just
593 * waiting for the initial write. The default is to only
594 * wait through the initial write so the data is in the
595 * server's cache, which is roughly similar to the state
596 * a standard disk subsystem leaves the file in on close().
598 * We cannot clear the NLMODIFIED bit in np->n_flag due to
599 * potential races with other processes, and certainly
600 * cannot clear it if we don't commit.
602 int cm = nfsv3_commit_on_close ? 1 : 0;
603 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
604 /* np->n_flag &= ~NLMODIFIED; */
606 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
610 if (np->n_flag & NWRITEERR) {
611 np->n_flag &= ~NWRITEERR;
620 * nfs getattr call from vfs.
622 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred,
623 * struct thread *a_td)
626 nfs_getattr(struct vop_getattr_args *ap)
628 struct vnode *vp = ap->a_vp;
629 struct nfsnode *np = VTONFS(vp);
635 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
636 int v3 = NFS_ISV3(vp);
639 * Update local times for special files.
641 if (np->n_flag & (NACC | NUPD))
644 * First look in the cache.
646 if (nfs_getattrcache(vp, ap->a_vap) == 0)
649 if (v3 && nfsaccess_cache_timeout > 0) {
650 nfsstats.accesscache_misses++;
651 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK));
652 if (nfs_getattrcache(vp, ap->a_vap) == 0)
656 nfsstats.rpccnt[NFSPROC_GETATTR]++;
657 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
659 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK));
661 nfsm_loadattr(vp, ap->a_vap);
671 * nfs_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp,
672 * struct vattr *a_vap, struct ucred *a_cred,
673 * struct thread *a_td)
676 nfs_setattr(struct vop_setattr_args *ap)
678 struct vnode *vp = ap->a_vp;
679 struct nfsnode *np = VTONFS(vp);
680 struct vattr *vap = ap->a_vap;
689 * Setting of flags is not supported.
691 if (vap->va_flags != VNOVAL)
695 * Disallow write attempts if the filesystem is mounted read-only.
697 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
698 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
699 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
700 (vp->v_mount->mnt_flag & MNT_RDONLY))
702 if (vap->va_size != VNOVAL) {
703 switch (vp->v_type) {
710 if (vap->va_mtime.tv_sec == VNOVAL &&
711 vap->va_atime.tv_sec == VNOVAL &&
712 vap->va_mode == (mode_t)VNOVAL &&
713 vap->va_uid == (uid_t)VNOVAL &&
714 vap->va_gid == (gid_t)VNOVAL)
716 vap->va_size = VNOVAL;
720 * Disallow write attempts if the filesystem is
723 if (vp->v_mount->mnt_flag & MNT_RDONLY)
727 * This is nasty. The RPCs we send to flush pending
728 * data often return attribute information which is
729 * cached via a callback to nfs_loadattrcache(), which
730 * has the effect of changing our notion of the file
731 * size. Due to flushed appends and other operations
732 * the file size can be set to virtually anything,
733 * including values that do not match either the old
734 * or intended file size.
736 * When this condition is detected we must loop to
737 * try the operation again. Hopefully no more
738 * flushing is required on the loop so it works the
739 * second time around. THIS CASE ALMOST ALWAYS
744 error = nfs_meta_setsize(vp, ap->a_td, vap->va_size);
746 if (np->n_flag & NLMODIFIED) {
747 if (vap->va_size == 0)
748 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
750 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
753 * note: this loop case almost always happens at
754 * least once per truncation.
756 if (error == 0 && np->n_size != vap->va_size)
758 np->n_vattr.va_size = vap->va_size;
761 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
762 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) &&
763 vp->v_type == VREG &&
764 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR
768 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
771 * Sanity check if a truncation was issued. This should only occur
772 * if multiple processes are racing on the same file.
774 if (error == 0 && vap->va_size != VNOVAL &&
775 np->n_size != vap->va_size) {
776 printf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size);
779 if (error && vap->va_size != VNOVAL) {
780 np->n_size = np->n_vattr.va_size = tsize;
781 vnode_pager_setsize(vp, np->n_size);
787 * Do an nfs setattr rpc.
790 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
791 struct ucred *cred, struct thread *td)
793 struct nfsv2_sattr *sp;
794 struct nfsnode *np = VTONFS(vp);
797 caddr_t bpos, dpos, cp2;
799 int error = 0, wccflag = NFSV3_WCCRATTR;
800 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
801 int v3 = NFS_ISV3(vp);
803 nfsstats.rpccnt[NFSPROC_SETATTR]++;
804 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
807 nfsm_v3attrbuild(vap, TRUE);
808 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
811 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
812 if (vap->va_mode == (mode_t)VNOVAL)
813 sp->sa_mode = nfs_xdrneg1;
815 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
816 if (vap->va_uid == (uid_t)VNOVAL)
817 sp->sa_uid = nfs_xdrneg1;
819 sp->sa_uid = txdr_unsigned(vap->va_uid);
820 if (vap->va_gid == (gid_t)VNOVAL)
821 sp->sa_gid = nfs_xdrneg1;
823 sp->sa_gid = txdr_unsigned(vap->va_gid);
824 sp->sa_size = txdr_unsigned(vap->va_size);
825 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
826 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
828 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
831 nfsm_wcc_data(vp, wccflag);
833 nfsm_loadattr(vp, (struct vattr *)0);
840 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
841 * nfs_lookup() until all remaining new api calls are implemented.
843 * Resolve a namecache entry. This function is passed a locked ncp and
844 * must call cache_setvp() on it as appropriate to resolve the entry.
847 nfs_nresolve(struct vop_nresolve_args *ap)
849 struct thread *td = curthread;
850 struct namecache *ncp;
861 /******NFSM MACROS********/
862 struct mbuf *mb, *mrep, *mreq, *mb2, *md;
863 caddr_t bpos, dpos, cp, cp2;
870 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp);
871 dvp = ncp->nc_parent->nc_vp;
872 if ((error = vget(dvp, LK_SHARED, td)) != 0)
877 nfsstats.lookupcache_misses++;
878 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
880 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
881 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
883 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN);
884 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred);
887 * Cache negatve lookups to reduce NFS traffic, but use
888 * a fast timeout. Otherwise use a timeout of 1 tick.
889 * XXX we should add a namecache flag for no-caching
890 * to uncache the negative hit as soon as possible, but
891 * we cannot simply destroy the entry because it is used
892 * as a placeholder by the caller.
894 if (error == ENOENT) {
897 if (nfsneg_cache_timeout)
898 nticks = nfsneg_cache_timeout * hz;
901 cache_setvp(ncp, NULL);
902 cache_settimeout(ncp, nticks);
904 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
910 * Success, get the file handle, do various checks, and load
911 * post-operation data from the reply packet. Theoretically
912 * we should never be looking up "." so, theoretically, we
913 * should never get the same file handle as our directory. But
914 * we check anyway. XXX
916 * Note that no timeout is set for the positive cache hit. We
917 * assume, theoretically, that ESTALE returns will be dealt with
918 * properly to handle NFS races and in anycase we cannot depend
919 * on a timeout to deal with NFS open/create/excl issues so instead
920 * of a bad hack here the rest of the NFS client code needs to do
923 nfsm_getfh(fhp, fhsize, v3);
926 if (NFS_CMPFH(np, fhp, fhsize)) {
930 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
939 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK);
940 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
942 nfsm_loadattr(nvp, NULL);
944 cache_setvp(ncp, nvp);
958 * 'cached' nfs directory lookup
960 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
962 * nfs_lookup(struct vnodeop_desc *a_desc, struct vnode *a_dvp,
963 * struct vnode **a_vpp, struct componentname *a_cnp)
966 nfs_lookup(struct vop_old_lookup_args *ap)
968 struct componentname *cnp = ap->a_cnp;
969 struct vnode *dvp = ap->a_dvp;
970 struct vnode **vpp = ap->a_vpp;
971 int flags = cnp->cn_flags;
976 struct nfsmount *nmp;
977 caddr_t bpos, dpos, cp2;
978 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
982 int lockparent, wantparent, error = 0, attrflag, fhsize;
983 int v3 = NFS_ISV3(dvp);
984 struct thread *td = cnp->cn_td;
987 * Read-only mount check and directory check.
990 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
991 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
994 if (dvp->v_type != VDIR)
998 * Look it up in the cache. Note that ENOENT is only returned if we
999 * previously entered a negative hit (see later on). The additional
1000 * nfsneg_cache_timeout check causes previously cached results to
1001 * be instantly ignored if the negative caching is turned off.
1003 lockparent = flags & CNP_LOCKPARENT;
1004 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1005 nmp = VFSTONFS(dvp->v_mount);
1013 nfsstats.lookupcache_misses++;
1014 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1015 len = cnp->cn_namelen;
1016 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1017 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1018 nfsm_fhtom(dvp, v3);
1019 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1020 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred);
1022 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1026 nfsm_getfh(fhp, fhsize, v3);
1029 * Handle RENAME case...
1031 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1032 if (NFS_CMPFH(np, fhp, fhsize)) {
1036 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1043 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1044 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1046 nfsm_loadattr(newvp, (struct vattr *)0);
1050 VOP_UNLOCK(dvp, 0, td);
1051 cnp->cn_flags |= CNP_PDIRUNLOCK;
1056 if (flags & CNP_ISDOTDOT) {
1057 VOP_UNLOCK(dvp, 0, td);
1058 cnp->cn_flags |= CNP_PDIRUNLOCK;
1059 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1061 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
1062 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1063 return (error); /* NOTE: return error from nget */
1067 error = vn_lock(dvp, LK_EXCLUSIVE, td);
1072 cnp->cn_flags |= CNP_PDIRUNLOCK;
1074 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1078 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1084 VOP_UNLOCK(dvp, 0, td);
1085 cnp->cn_flags |= CNP_PDIRUNLOCK;
1090 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1091 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1093 nfsm_loadattr(newvp, (struct vattr *)0);
1095 /* XXX MOVE TO nfs_nremove() */
1096 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1097 cnp->cn_nameiop != NAMEI_DELETE) {
1098 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1105 if (newvp != NULLVP) {
1109 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1110 cnp->cn_nameiop == NAMEI_RENAME) &&
1113 VOP_UNLOCK(dvp, 0, td);
1114 cnp->cn_flags |= CNP_PDIRUNLOCK;
1116 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1119 error = EJUSTRETURN;
1127 * Just call nfs_bioread() to do the work.
1129 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1130 * struct ucred *a_cred)
1133 nfs_read(struct vop_read_args *ap)
1135 struct vnode *vp = ap->a_vp;
1137 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1138 switch (vp->v_type) {
1140 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1151 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1154 nfs_readlink(struct vop_readlink_args *ap)
1156 struct vnode *vp = ap->a_vp;
1158 if (vp->v_type != VLNK)
1160 if (vp->v_object == NULL)
1162 return (nfs_bioread(vp, ap->a_uio, 0));
1166 * Do a readlink rpc.
1167 * Called by nfs_doio() from below the buffer cache.
1170 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1175 caddr_t bpos, dpos, cp2;
1176 int error = 0, len, attrflag;
1177 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1178 int v3 = NFS_ISV3(vp);
1180 nfsstats.rpccnt[NFSPROC_READLINK]++;
1181 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1183 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK));
1185 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1187 nfsm_strsiz(len, NFS_MAXPATHLEN);
1188 if (len == NFS_MAXPATHLEN) {
1189 struct nfsnode *np = VTONFS(vp);
1190 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1193 nfsm_mtouio(uiop, len);
1205 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1210 caddr_t bpos, dpos, cp2;
1211 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1212 struct nfsmount *nmp;
1213 int error = 0, len, retlen, tsiz, eof, attrflag;
1214 int v3 = NFS_ISV3(vp);
1219 nmp = VFSTONFS(vp->v_mount);
1220 tsiz = uiop->uio_resid;
1221 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1224 nfsstats.rpccnt[NFSPROC_READ]++;
1225 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1226 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1228 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1230 txdr_hyper(uiop->uio_offset, tl);
1231 *(tl + 2) = txdr_unsigned(len);
1233 *tl++ = txdr_unsigned(uiop->uio_offset);
1234 *tl++ = txdr_unsigned(len);
1237 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ));
1239 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1244 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1245 eof = fxdr_unsigned(int, *(tl + 1));
1247 nfsm_loadattr(vp, (struct vattr *)0);
1248 nfsm_strsiz(retlen, nmp->nm_rsize);
1249 nfsm_mtouio(uiop, retlen);
1253 if (eof || retlen == 0) {
1256 } else if (retlen < len) {
1268 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1272 int32_t t1, t2, backup;
1273 caddr_t bpos, dpos, cp2;
1274 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1275 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1276 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1277 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1280 if (uiop->uio_iovcnt != 1)
1281 panic("nfs: writerpc iovcnt > 1");
1284 tsiz = uiop->uio_resid;
1285 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1288 nfsstats.rpccnt[NFSPROC_WRITE]++;
1289 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1290 nfsm_reqhead(vp, NFSPROC_WRITE,
1291 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1294 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1295 txdr_hyper(uiop->uio_offset, tl);
1297 *tl++ = txdr_unsigned(len);
1298 *tl++ = txdr_unsigned(*iomode);
1299 *tl = txdr_unsigned(len);
1303 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1304 /* Set both "begin" and "current" to non-garbage. */
1305 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1306 *tl++ = x; /* "begin offset" */
1307 *tl++ = x; /* "current offset" */
1308 x = txdr_unsigned(len);
1309 *tl++ = x; /* total to this offset */
1310 *tl = x; /* size of this write */
1312 nfsm_uiotom(uiop, len);
1313 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE));
1316 * The write RPC returns a before and after mtime. The
1317 * nfsm_wcc_data() macro checks the before n_mtime
1318 * against the before time and stores the after time
1319 * in the nfsnode's cached vattr and n_mtime field.
1320 * The NRMODIFIED bit will be set if the before
1321 * time did not match the original mtime.
1323 wccflag = NFSV3_WCCCHK;
1324 nfsm_wcc_data(vp, wccflag);
1326 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1327 + NFSX_V3WRITEVERF);
1328 rlen = fxdr_unsigned(int, *tl++);
1333 } else if (rlen < len) {
1334 backup = len - rlen;
1335 uiop->uio_iov->iov_base -= backup;
1336 uiop->uio_iov->iov_len += backup;
1337 uiop->uio_offset -= backup;
1338 uiop->uio_resid += backup;
1341 commit = fxdr_unsigned(int, *tl++);
1344 * Return the lowest committment level
1345 * obtained by any of the RPCs.
1347 if (committed == NFSV3WRITE_FILESYNC)
1349 else if (committed == NFSV3WRITE_DATASYNC &&
1350 commit == NFSV3WRITE_UNSTABLE)
1352 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1353 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1355 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1356 } else if (bcmp((caddr_t)tl,
1357 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1359 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1364 nfsm_loadattr(vp, (struct vattr *)0);
1372 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1373 committed = NFSV3WRITE_FILESYNC;
1374 *iomode = committed;
1376 uiop->uio_resid = tsiz;
1382 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1383 * mode set to specify the file type and the size field for rdev.
1386 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1389 struct nfsv2_sattr *sp;
1393 struct vnode *newvp = (struct vnode *)0;
1394 struct nfsnode *np = (struct nfsnode *)0;
1398 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1399 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1401 int v3 = NFS_ISV3(dvp);
1403 if (vap->va_type == VCHR || vap->va_type == VBLK)
1404 rdev = txdr_unsigned(vap->va_rdev);
1405 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1408 return (EOPNOTSUPP);
1410 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1413 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1414 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1415 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1416 nfsm_fhtom(dvp, v3);
1417 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1419 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1420 *tl++ = vtonfsv3_type(vap->va_type);
1421 nfsm_v3attrbuild(vap, FALSE);
1422 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1423 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1424 *tl++ = txdr_unsigned(umajor(vap->va_rdev));
1425 *tl = txdr_unsigned(uminor(vap->va_rdev));
1428 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1429 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1430 sp->sa_uid = nfs_xdrneg1;
1431 sp->sa_gid = nfs_xdrneg1;
1433 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1434 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1436 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred);
1438 nfsm_mtofh(dvp, newvp, v3, gotvp);
1442 newvp = (struct vnode *)0;
1444 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1445 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1451 nfsm_wcc_data(dvp, wccflag);
1460 VTONFS(dvp)->n_flag |= NLMODIFIED;
1462 VTONFS(dvp)->n_attrstamp = 0;
1468 * just call nfs_mknodrpc() to do the work.
1470 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1471 * struct componentname *a_cnp, struct vattr *a_vap)
1475 nfs_mknod(struct vop_old_mknod_args *ap)
1477 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1480 static u_long create_verf;
1482 * nfs file create call
1484 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1485 * struct componentname *a_cnp, struct vattr *a_vap)
1488 nfs_create(struct vop_old_create_args *ap)
1490 struct vnode *dvp = ap->a_dvp;
1491 struct vattr *vap = ap->a_vap;
1492 struct componentname *cnp = ap->a_cnp;
1493 struct nfsv2_sattr *sp;
1497 struct nfsnode *np = (struct nfsnode *)0;
1498 struct vnode *newvp = (struct vnode *)0;
1499 caddr_t bpos, dpos, cp2;
1500 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1501 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1503 int v3 = NFS_ISV3(dvp);
1506 * Oops, not for me..
1508 if (vap->va_type == VSOCK)
1509 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1511 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1514 if (vap->va_vaflags & VA_EXCLUSIVE)
1517 nfsstats.rpccnt[NFSPROC_CREATE]++;
1518 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1519 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1520 nfsm_fhtom(dvp, v3);
1521 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1523 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1524 if (fmode & O_EXCL) {
1525 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1526 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1528 if (!TAILQ_EMPTY(&in_ifaddrhead))
1529 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1532 *tl++ = create_verf;
1533 *tl = ++create_verf;
1535 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1536 nfsm_v3attrbuild(vap, FALSE);
1539 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1540 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1541 sp->sa_uid = nfs_xdrneg1;
1542 sp->sa_gid = nfs_xdrneg1;
1544 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1545 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1547 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred);
1549 nfsm_mtofh(dvp, newvp, v3, gotvp);
1553 newvp = (struct vnode *)0;
1555 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1556 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1562 nfsm_wcc_data(dvp, wccflag);
1566 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1572 } else if (v3 && (fmode & O_EXCL)) {
1574 * We are normally called with only a partially initialized
1575 * VAP. Since the NFSv3 spec says that server may use the
1576 * file attributes to store the verifier, the spec requires
1577 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1578 * in atime, but we can't really assume that all servers will
1579 * so we ensure that our SETATTR sets both atime and mtime.
1581 if (vap->va_mtime.tv_sec == VNOVAL)
1582 vfs_timestamp(&vap->va_mtime);
1583 if (vap->va_atime.tv_sec == VNOVAL)
1584 vap->va_atime = vap->va_mtime;
1585 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1589 * The new np may have enough info for access
1590 * checks, make sure rucred and wucred are
1591 * initialized for read and write rpc's.
1594 if (np->n_rucred == NULL)
1595 np->n_rucred = crhold(cnp->cn_cred);
1596 if (np->n_wucred == NULL)
1597 np->n_wucred = crhold(cnp->cn_cred);
1600 VTONFS(dvp)->n_flag |= NLMODIFIED;
1602 VTONFS(dvp)->n_attrstamp = 0;
1607 * nfs file remove call
1608 * To try and make nfs semantics closer to ufs semantics, a file that has
1609 * other processes using the vnode is renamed instead of removed and then
1610 * removed later on the last close.
1611 * - If v_usecount > 1
1612 * If a rename is not already in the works
1613 * call nfs_sillyrename() to set it up
1617 * nfs_remove(struct vnodeop_desc *a_desc, struct vnode *a_dvp,
1618 * struct vnode *a_vp, struct componentname *a_cnp)
1621 nfs_remove(struct vop_old_remove_args *ap)
1623 struct vnode *vp = ap->a_vp;
1624 struct vnode *dvp = ap->a_dvp;
1625 struct componentname *cnp = ap->a_cnp;
1626 struct nfsnode *np = VTONFS(vp);
1631 if (vp->v_usecount < 1)
1632 panic("nfs_remove: bad v_usecount");
1634 if (vp->v_type == VDIR)
1636 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1637 VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 &&
1638 vattr.va_nlink > 1)) {
1640 * throw away biocache buffers, mainly to avoid
1641 * unnecessary delayed writes later.
1643 error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1);
1646 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1647 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1649 * Kludge City: If the first reply to the remove rpc is lost..
1650 * the reply to the retransmitted request will be ENOENT
1651 * since the file was in fact removed
1652 * Therefore, we cheat and return success.
1654 if (error == ENOENT)
1656 } else if (!np->n_sillyrename) {
1657 error = nfs_sillyrename(dvp, vp, cnp);
1659 np->n_attrstamp = 0;
1664 * nfs file remove rpc called from nfs_inactive
1667 nfs_removeit(struct sillyrename *sp)
1669 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1674 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1677 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1678 struct ucred *cred, struct thread *td)
1683 caddr_t bpos, dpos, cp2;
1684 int error = 0, wccflag = NFSV3_WCCRATTR;
1685 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1686 int v3 = NFS_ISV3(dvp);
1688 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1689 nfsm_reqhead(dvp, NFSPROC_REMOVE,
1690 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1691 nfsm_fhtom(dvp, v3);
1692 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1693 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1695 nfsm_wcc_data(dvp, wccflag);
1698 VTONFS(dvp)->n_flag |= NLMODIFIED;
1700 VTONFS(dvp)->n_attrstamp = 0;
1705 * nfs file rename call
1707 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1708 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1709 * struct vnode *a_tvp, struct componentname *a_tcnp)
1712 nfs_rename(struct vop_old_rename_args *ap)
1714 struct vnode *fvp = ap->a_fvp;
1715 struct vnode *tvp = ap->a_tvp;
1716 struct vnode *fdvp = ap->a_fdvp;
1717 struct vnode *tdvp = ap->a_tdvp;
1718 struct componentname *tcnp = ap->a_tcnp;
1719 struct componentname *fcnp = ap->a_fcnp;
1722 /* Check for cross-device rename */
1723 if ((fvp->v_mount != tdvp->v_mount) ||
1724 (tvp && (fvp->v_mount != tvp->v_mount))) {
1730 * We have to flush B_DELWRI data prior to renaming
1731 * the file. If we don't, the delayed-write buffers
1732 * can be flushed out later after the file has gone stale
1733 * under NFSV3. NFSV2 does not have this problem because
1734 * ( as far as I can tell ) it flushes dirty buffers more
1738 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td);
1740 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td);
1743 * If the tvp exists and is in use, sillyrename it before doing the
1744 * rename of the new file over it.
1746 * XXX Can't sillyrename a directory.
1748 * We do not attempt to do any namecache purges in this old API
1749 * routine. The new API compat functions have access to the actual
1750 * namecache structures and will do it for us.
1752 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1753 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1760 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1761 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1774 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1776 if (error == ENOENT)
1782 * nfs file rename rpc called from nfs_remove() above
1785 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1786 struct sillyrename *sp)
1788 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1789 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1793 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1796 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1797 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1798 struct ucred *cred, struct thread *td)
1803 caddr_t bpos, dpos, cp2;
1804 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1805 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1806 int v3 = NFS_ISV3(fdvp);
1808 nfsstats.rpccnt[NFSPROC_RENAME]++;
1809 nfsm_reqhead(fdvp, NFSPROC_RENAME,
1810 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1811 nfsm_rndup(tnamelen));
1812 nfsm_fhtom(fdvp, v3);
1813 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1814 nfsm_fhtom(tdvp, v3);
1815 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1816 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1818 nfsm_wcc_data(fdvp, fwccflag);
1819 nfsm_wcc_data(tdvp, twccflag);
1823 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1824 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1826 VTONFS(fdvp)->n_attrstamp = 0;
1828 VTONFS(tdvp)->n_attrstamp = 0;
1833 * nfs hard link create call
1835 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1836 * struct componentname *a_cnp)
1839 nfs_link(struct vop_old_link_args *ap)
1841 struct vnode *vp = ap->a_vp;
1842 struct vnode *tdvp = ap->a_tdvp;
1843 struct componentname *cnp = ap->a_cnp;
1847 caddr_t bpos, dpos, cp2;
1848 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1849 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1852 if (vp->v_mount != tdvp->v_mount) {
1857 * Push all writes to the server, so that the attribute cache
1858 * doesn't get "out of sync" with the server.
1859 * XXX There should be a better way!
1861 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td);
1864 nfsstats.rpccnt[NFSPROC_LINK]++;
1865 nfsm_reqhead(vp, NFSPROC_LINK,
1866 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1868 nfsm_fhtom(tdvp, v3);
1869 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1870 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred);
1872 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1873 nfsm_wcc_data(tdvp, wccflag);
1877 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1879 VTONFS(vp)->n_attrstamp = 0;
1881 VTONFS(tdvp)->n_attrstamp = 0;
1883 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1885 if (error == EEXIST)
1891 * nfs symbolic link create call
1893 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1894 * struct componentname *a_cnp, struct vattr *a_vap,
1898 nfs_symlink(struct vop_old_symlink_args *ap)
1900 struct vnode *dvp = ap->a_dvp;
1901 struct vattr *vap = ap->a_vap;
1902 struct componentname *cnp = ap->a_cnp;
1903 struct nfsv2_sattr *sp;
1907 caddr_t bpos, dpos, cp2;
1908 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1909 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1910 struct vnode *newvp = (struct vnode *)0;
1911 int v3 = NFS_ISV3(dvp);
1913 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1914 slen = strlen(ap->a_target);
1915 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1916 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1917 nfsm_fhtom(dvp, v3);
1918 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1920 nfsm_v3attrbuild(vap, FALSE);
1922 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1924 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1925 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1926 sp->sa_uid = nfs_xdrneg1;
1927 sp->sa_gid = nfs_xdrneg1;
1928 sp->sa_size = nfs_xdrneg1;
1929 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1930 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1934 * Issue the NFS request and get the rpc response.
1936 * Only NFSv3 responses returning an error of 0 actually return
1937 * a file handle that can be converted into newvp without having
1938 * to do an extra lookup rpc.
1940 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred);
1943 nfsm_mtofh(dvp, newvp, v3, gotvp);
1944 nfsm_wcc_data(dvp, wccflag);
1948 * out code jumps -> here, mrep is also freed.
1955 * If we get an EEXIST error, silently convert it to no-error
1956 * in case of an NFS retry.
1958 if (error == EEXIST)
1962 * If we do not have (or no longer have) an error, and we could
1963 * not extract the newvp from the response due to the request being
1964 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1965 * to obtain a newvp to return.
1967 if (error == 0 && newvp == NULL) {
1968 struct nfsnode *np = NULL;
1970 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1971 cnp->cn_cred, cnp->cn_td, &np);
1981 VTONFS(dvp)->n_flag |= NLMODIFIED;
1983 VTONFS(dvp)->n_attrstamp = 0;
1990 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1991 * struct componentname *a_cnp, struct vattr *a_vap)
1994 nfs_mkdir(struct vop_old_mkdir_args *ap)
1996 struct vnode *dvp = ap->a_dvp;
1997 struct vattr *vap = ap->a_vap;
1998 struct componentname *cnp = ap->a_cnp;
1999 struct nfsv2_sattr *sp;
2004 struct nfsnode *np = (struct nfsnode *)0;
2005 struct vnode *newvp = (struct vnode *)0;
2006 caddr_t bpos, dpos, cp2;
2007 int error = 0, wccflag = NFSV3_WCCRATTR;
2009 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2011 int v3 = NFS_ISV3(dvp);
2013 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
2016 len = cnp->cn_namelen;
2017 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2018 nfsm_reqhead(dvp, NFSPROC_MKDIR,
2019 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2020 nfsm_fhtom(dvp, v3);
2021 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2023 nfsm_v3attrbuild(vap, FALSE);
2025 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2026 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2027 sp->sa_uid = nfs_xdrneg1;
2028 sp->sa_gid = nfs_xdrneg1;
2029 sp->sa_size = nfs_xdrneg1;
2030 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2031 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2033 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred);
2035 nfsm_mtofh(dvp, newvp, v3, gotvp);
2037 nfsm_wcc_data(dvp, wccflag);
2040 VTONFS(dvp)->n_flag |= NLMODIFIED;
2042 VTONFS(dvp)->n_attrstamp = 0;
2044 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2045 * if we can succeed in looking up the directory.
2047 if (error == EEXIST || (!error && !gotvp)) {
2050 newvp = (struct vnode *)0;
2052 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2056 if (newvp->v_type != VDIR)
2069 * nfs remove directory call
2071 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2072 * struct componentname *a_cnp)
2075 nfs_rmdir(struct vop_old_rmdir_args *ap)
2077 struct vnode *vp = ap->a_vp;
2078 struct vnode *dvp = ap->a_dvp;
2079 struct componentname *cnp = ap->a_cnp;
2083 caddr_t bpos, dpos, cp2;
2084 int error = 0, wccflag = NFSV3_WCCRATTR;
2085 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2086 int v3 = NFS_ISV3(dvp);
2090 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2091 nfsm_reqhead(dvp, NFSPROC_RMDIR,
2092 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2093 nfsm_fhtom(dvp, v3);
2094 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2095 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred);
2097 nfsm_wcc_data(dvp, wccflag);
2100 VTONFS(dvp)->n_flag |= NLMODIFIED;
2102 VTONFS(dvp)->n_attrstamp = 0;
2104 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2106 if (error == ENOENT)
2114 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2117 nfs_readdir(struct vop_readdir_args *ap)
2119 struct vnode *vp = ap->a_vp;
2120 struct nfsnode *np = VTONFS(vp);
2121 struct uio *uio = ap->a_uio;
2125 if (vp->v_type != VDIR)
2129 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2130 * and then check that is still valid, or if this is an NQNFS mount
2131 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2132 * VOP_GETATTR() does not necessarily go to the wire.
2134 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2135 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2136 if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 &&
2137 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2139 nfsstats.direofcache_hits++;
2145 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2146 * own cache coherency checks so we do not have to.
2148 tresid = uio->uio_resid;
2149 error = nfs_bioread(vp, uio, 0);
2151 if (!error && uio->uio_resid == tresid)
2152 nfsstats.direofcache_misses++;
2157 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2159 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2160 * offset/block and converts the nfs formatted directory entries for userland
2161 * consumption as well as deals with offsets into the middle of blocks.
2162 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2163 * be block-bounded. It must convert to cookies for the actual RPC.
2166 nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2169 struct nfs_dirent *dp = NULL;
2174 caddr_t bpos, dpos, cp2;
2175 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2177 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2178 struct nfsnode *dnp = VTONFS(vp);
2180 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2182 int v3 = NFS_ISV3(vp);
2185 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2186 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2187 panic("nfs readdirrpc bad uio");
2191 * If there is no cookie, assume directory was stale.
2193 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2197 return (NFSERR_BAD_COOKIE);
2199 * Loop around doing readdir rpc's of size nm_readdirsize
2200 * truncated to a multiple of DIRBLKSIZ.
2201 * The stopping criteria is EOF or buffer full.
2203 while (more_dirs && bigenough) {
2204 nfsstats.rpccnt[NFSPROC_READDIR]++;
2205 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2209 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2210 *tl++ = cookie.nfsuquad[0];
2211 *tl++ = cookie.nfsuquad[1];
2212 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2213 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2215 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2216 *tl++ = cookie.nfsuquad[0];
2218 *tl = txdr_unsigned(nmp->nm_readdirsize);
2219 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2221 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2223 nfsm_dissect(tl, u_int32_t *,
2225 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2226 dnp->n_cookieverf.nfsuquad[1] = *tl;
2232 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2233 more_dirs = fxdr_unsigned(int, *tl);
2235 /* loop thru the dir entries, converting them to std form */
2236 while (more_dirs && bigenough) {
2238 nfsm_dissect(tl, u_int32_t *,
2240 fileno = fxdr_hyper(tl);
2241 len = fxdr_unsigned(int, *(tl + 2));
2243 nfsm_dissect(tl, u_int32_t *,
2245 fileno = fxdr_unsigned(u_quad_t, *tl++);
2246 len = fxdr_unsigned(int, *tl);
2248 if (len <= 0 || len > NFS_MAXNAMLEN) {
2255 * len is the number of bytes in the path element
2256 * name, not including the \0 termination.
2258 * tlen is the number of bytes w have to reserve for
2259 * the path element name.
2261 tlen = nfsm_rndup(len);
2263 tlen += 4; /* To ensure null termination */
2266 * If the entry would cross a DIRBLKSIZ boundary,
2267 * extend the previous nfs_dirent to cover the
2270 left = DIRBLKSIZ - blksiz;
2271 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2272 dp->nfs_reclen += left;
2273 uiop->uio_iov->iov_base += left;
2274 uiop->uio_iov->iov_len -= left;
2275 uiop->uio_offset += left;
2276 uiop->uio_resid -= left;
2279 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2282 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2283 dp->nfs_ino = fileno;
2284 dp->nfs_namlen = len;
2285 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2286 dp->nfs_type = DT_UNKNOWN;
2287 blksiz += dp->nfs_reclen;
2288 if (blksiz == DIRBLKSIZ)
2290 uiop->uio_offset += sizeof(struct nfs_dirent);
2291 uiop->uio_resid -= sizeof(struct nfs_dirent);
2292 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent);
2293 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2294 nfsm_mtouio(uiop, len);
2297 * The uiop has advanced by nfs_dirent + len
2298 * but really needs to advance by
2301 cp = uiop->uio_iov->iov_base;
2303 *cp = '\0'; /* null terminate */
2304 uiop->uio_iov->iov_base += tlen;
2305 uiop->uio_iov->iov_len -= tlen;
2306 uiop->uio_offset += tlen;
2307 uiop->uio_resid -= tlen;
2310 * NFS strings must be rounded up (nfsm_myouio
2311 * handled that in the bigenough case).
2313 nfsm_adv(nfsm_rndup(len));
2316 nfsm_dissect(tl, u_int32_t *,
2319 nfsm_dissect(tl, u_int32_t *,
2324 * If we were able to accomodate the last entry,
2325 * get the cookie for the next one. Otherwise
2326 * hold-over the cookie for the one we were not
2327 * able to accomodate.
2330 cookie.nfsuquad[0] = *tl++;
2332 cookie.nfsuquad[1] = *tl++;
2338 more_dirs = fxdr_unsigned(int, *tl);
2341 * If at end of rpc data, get the eof boolean
2344 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2345 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2350 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2351 * by increasing d_reclen for the last record.
2354 left = DIRBLKSIZ - blksiz;
2355 dp->nfs_reclen += left;
2356 uiop->uio_iov->iov_base += left;
2357 uiop->uio_iov->iov_len -= left;
2358 uiop->uio_offset += left;
2359 uiop->uio_resid -= left;
2364 * We hit the end of the directory, update direofoffset.
2366 dnp->n_direofoffset = uiop->uio_offset;
2369 * There is more to go, insert the link cookie so the
2370 * next block can be read.
2372 if (uiop->uio_resid > 0)
2373 printf("EEK! readdirrpc resid > 0\n");
2374 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2382 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2385 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2388 struct nfs_dirent *dp;
2392 struct vnode *newvp;
2394 caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2395 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2397 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2398 struct nfsnode *dnp = VTONFS(vp), *np;
2401 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2402 int attrflag, fhsize;
2403 struct namecache *ncp;
2404 struct namecache *dncp;
2405 struct nlcomponent nlc;
2411 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2412 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2413 panic("nfs readdirplusrpc bad uio");
2416 * Obtain the namecache record for the directory so we have something
2417 * to use as a basis for creating the entries. This function will
2418 * return a held (but not locked) ncp. The ncp may be disconnected
2419 * from the tree and cannot be used for upward traversals, and the
2420 * ncp may be unnamed. Note that other unrelated operations may
2421 * cause the ncp to be named at any time.
2423 dncp = cache_fromdvp(vp, NULL, 0);
2424 bzero(&nlc, sizeof(nlc));
2428 * If there is no cookie, assume directory was stale.
2430 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2434 return (NFSERR_BAD_COOKIE);
2436 * Loop around doing readdir rpc's of size nm_readdirsize
2437 * truncated to a multiple of DIRBLKSIZ.
2438 * The stopping criteria is EOF or buffer full.
2440 while (more_dirs && bigenough) {
2441 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2442 nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2443 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2445 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2446 *tl++ = cookie.nfsuquad[0];
2447 *tl++ = cookie.nfsuquad[1];
2448 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2449 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2450 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2451 *tl = txdr_unsigned(nmp->nm_rsize);
2452 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2453 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2458 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2459 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2460 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2461 more_dirs = fxdr_unsigned(int, *tl);
2463 /* loop thru the dir entries, doctoring them to 4bsd form */
2464 while (more_dirs && bigenough) {
2465 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2466 fileno = fxdr_hyper(tl);
2467 len = fxdr_unsigned(int, *(tl + 2));
2468 if (len <= 0 || len > NFS_MAXNAMLEN) {
2473 tlen = nfsm_rndup(len);
2475 tlen += 4; /* To ensure null termination*/
2476 left = DIRBLKSIZ - blksiz;
2477 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2478 dp->nfs_reclen += left;
2479 uiop->uio_iov->iov_base += left;
2480 uiop->uio_iov->iov_len -= left;
2481 uiop->uio_offset += left;
2482 uiop->uio_resid -= left;
2485 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2488 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2489 dp->nfs_ino = fileno;
2490 dp->nfs_namlen = len;
2491 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2492 dp->nfs_type = DT_UNKNOWN;
2493 blksiz += dp->nfs_reclen;
2494 if (blksiz == DIRBLKSIZ)
2496 uiop->uio_offset += sizeof(struct nfs_dirent);
2497 uiop->uio_resid -= sizeof(struct nfs_dirent);
2498 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent);
2499 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2500 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2501 nlc.nlc_namelen = len;
2502 nfsm_mtouio(uiop, len);
2503 cp = uiop->uio_iov->iov_base;
2506 uiop->uio_iov->iov_base += tlen;
2507 uiop->uio_iov->iov_len -= tlen;
2508 uiop->uio_offset += tlen;
2509 uiop->uio_resid -= tlen;
2511 nfsm_adv(nfsm_rndup(len));
2512 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2514 cookie.nfsuquad[0] = *tl++;
2515 cookie.nfsuquad[1] = *tl++;
2520 * Since the attributes are before the file handle
2521 * (sigh), we must skip over the attributes and then
2522 * come back and get them.
2524 attrflag = fxdr_unsigned(int, *tl);
2528 nfsm_adv(NFSX_V3FATTR);
2529 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2530 doit = fxdr_unsigned(int, *tl);
2532 nfsm_getfh(fhp, fhsize, 1);
2533 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2538 error = nfs_nget(vp->v_mount, fhp,
2546 if (doit && bigenough) {
2551 nfsm_loadattr(newvp, (struct vattr *)0);
2555 IFTODT(VTTOIF(np->n_vattr.va_type));
2557 printf("NFS/READDIRPLUS, ENTER %*.*s\n",
2558 nlc.nlc_namelen, nlc.nlc_namelen,
2560 ncp = cache_nlookup(dncp, &nlc);
2561 cache_setunresolved(ncp);
2562 cache_setvp(ncp, newvp);
2565 printf("NFS/READDIRPLUS, UNABLE TO ENTER"
2567 nlc.nlc_namelen, nlc.nlc_namelen,
2572 /* Just skip over the file handle */
2573 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2574 i = fxdr_unsigned(int, *tl);
2575 nfsm_adv(nfsm_rndup(i));
2577 if (newvp != NULLVP) {
2584 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2585 more_dirs = fxdr_unsigned(int, *tl);
2588 * If at end of rpc data, get the eof boolean
2591 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2592 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2597 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2598 * by increasing d_reclen for the last record.
2601 left = DIRBLKSIZ - blksiz;
2602 dp->nfs_reclen += left;
2603 uiop->uio_iov->iov_base += left;
2604 uiop->uio_iov->iov_len -= left;
2605 uiop->uio_offset += left;
2606 uiop->uio_resid -= left;
2610 * We are now either at the end of the directory or have filled the
2614 dnp->n_direofoffset = uiop->uio_offset;
2616 if (uiop->uio_resid > 0)
2617 printf("EEK! readdirplusrpc resid > 0\n");
2618 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2622 if (newvp != NULLVP) {
2635 * Silly rename. To make the NFS filesystem that is stateless look a little
2636 * more like the "ufs" a remove of an active vnode is translated to a rename
2637 * to a funny looking filename that is removed by nfs_inactive on the
2638 * nfsnode. There is the potential for another process on a different client
2639 * to create the same funny name between the nfs_lookitup() fails and the
2640 * nfs_rename() completes, but...
2643 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2645 struct sillyrename *sp;
2650 * We previously purged dvp instead of vp. I don't know why, it
2651 * completely destroys performance. We can't do it anyway with the
2652 * new VFS API since we would be breaking the namecache topology.
2654 cache_purge(vp); /* XXX */
2657 if (vp->v_type == VDIR)
2658 panic("nfs: sillyrename dir");
2660 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2661 M_NFSREQ, M_WAITOK);
2662 sp->s_cred = crdup(cnp->cn_cred);
2666 /* Fudge together a funny name */
2667 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td);
2669 /* Try lookitups until we get one that isn't there */
2670 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2671 cnp->cn_td, (struct nfsnode **)0) == 0) {
2673 if (sp->s_name[4] > 'z') {
2678 error = nfs_renameit(dvp, cnp, sp);
2681 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2683 np->n_sillyrename = sp;
2688 free((caddr_t)sp, M_NFSREQ);
2693 * Look up a file name and optionally either update the file handle or
2694 * allocate an nfsnode, depending on the value of npp.
2695 * npp == NULL --> just do the lookup
2696 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2698 * *npp != NULL --> update the file handle in the vnode
2701 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2702 struct thread *td, struct nfsnode **npp)
2707 struct vnode *newvp = (struct vnode *)0;
2708 struct nfsnode *np, *dnp = VTONFS(dvp);
2709 caddr_t bpos, dpos, cp2;
2710 int error = 0, fhlen, attrflag;
2711 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2713 int v3 = NFS_ISV3(dvp);
2715 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2716 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2717 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2718 nfsm_fhtom(dvp, v3);
2719 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2720 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2721 if (npp && !error) {
2722 nfsm_getfh(nfhp, fhlen, v3);
2725 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2726 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2727 np->n_fhp = &np->n_fh;
2728 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2729 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2730 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2731 np->n_fhsize = fhlen;
2733 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2737 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2745 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
2746 if (!attrflag && *npp == NULL) {
2755 nfsm_loadattr(newvp, (struct vattr *)0);
2759 if (npp && *npp == NULL) {
2774 * Nfs Version 3 commit rpc
2777 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2782 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2783 caddr_t bpos, dpos, cp2;
2784 int error = 0, wccflag = NFSV3_WCCRATTR;
2785 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2787 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2789 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2790 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2792 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2793 txdr_hyper(offset, tl);
2795 *tl = txdr_unsigned(cnt);
2796 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE));
2797 nfsm_wcc_data(vp, wccflag);
2799 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2800 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2801 NFSX_V3WRITEVERF)) {
2802 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2804 error = NFSERR_STALEWRITEVERF;
2814 * - make nfs_bmap() essentially a no-op that does no translation
2815 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2816 * (Maybe I could use the process's page mapping, but I was concerned that
2817 * Kernel Write might not be enabled and also figured copyout() would do
2818 * a lot more work than bcopy() and also it currently happens in the
2819 * context of the swapper process (2).
2821 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, struct vnode **a_vpp,
2822 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2825 nfs_bmap(struct vop_bmap_args *ap)
2827 struct vnode *vp = ap->a_vp;
2829 if (ap->a_vpp != NULL)
2831 if (ap->a_doffsetp != NULL)
2832 *ap->a_doffsetp = ap->a_loffset;
2833 if (ap->a_runp != NULL)
2835 if (ap->a_runb != NULL)
2843 * For async requests when nfsiod(s) are running, queue the request by
2844 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2848 nfs_strategy(struct vop_strategy_args *ap)
2850 struct bio *bio = ap->a_bio;
2852 struct buf *bp = bio->bio_buf;
2856 KASSERT(!(bp->b_flags & B_DONE),
2857 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2858 KASSERT(BUF_REFCNT(bp) > 0,
2859 ("nfs_strategy: buffer %p not locked", bp));
2861 if (bp->b_flags & B_PHYS)
2862 panic("nfs physio");
2864 if (bp->b_flags & B_ASYNC)
2867 td = curthread; /* XXX */
2870 * We probably don't need to push an nbio any more since no
2871 * block conversion is required due to the use of 64 bit byte
2872 * offsets, but do it anyway.
2874 nbio = push_bio(bio);
2875 nbio->bio_offset = bio->bio_offset;
2878 * If the op is asynchronous and an i/o daemon is waiting
2879 * queue the request, wake it up and wait for completion
2880 * otherwise just do it ourselves.
2882 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(ap->a_vp, nbio, td))
2883 error = nfs_doio(ap->a_vp, nbio, td);
2890 * NB Currently unsupported.
2892 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred,
2893 * struct thread *a_td)
2897 nfs_mmap(struct vop_mmap_args *ap)
2903 * fsync vnode op. Just call nfs_flush() with commit == 1.
2905 * nfs_fsync(struct vnodeop_desc *a_desc, struct vnode *a_vp,
2906 * struct ucred * a_cred, int a_waitfor, struct thread *a_td)
2910 nfs_fsync(struct vop_fsync_args *ap)
2912 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2916 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2917 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2918 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2919 * set the buffer contains data that has already been written to the server
2920 * and which now needs a commit RPC.
2922 * If commit is 0 we only take one pass and only flush buffers containing new
2925 * If commit is 1 we take two passes, issuing a commit RPC in the second
2928 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
2929 * to completely flush all pending data.
2931 * Note that the RB_SCAN code properly handles the case where the
2932 * callback might block and directly or indirectly (another thread) cause
2933 * the RB tree to change.
2936 #ifndef NFS_COMMITBVECSIZ
2937 #define NFS_COMMITBVECSIZ 16
2940 struct nfs_flush_info {
2941 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
2948 struct buf *bvary[NFS_COMMITBVECSIZ];
2954 static int nfs_flush_bp(struct buf *bp, void *data);
2955 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
2958 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
2960 struct nfsnode *np = VTONFS(vp);
2961 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2962 struct nfs_flush_info info;
2965 bzero(&info, sizeof(info));
2968 info.waitfor = waitfor;
2969 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
2976 info.mode = NFI_FLUSHNEW;
2977 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2978 nfs_flush_bp, &info);
2981 * Take a second pass if committing and no error occured.
2982 * Clean up any left over collection (whether an error
2985 if (commit && error == 0) {
2986 info.mode = NFI_COMMIT;
2987 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2988 nfs_flush_bp, &info);
2990 error = nfs_flush_docommit(&info, error);
2994 * Wait for pending I/O to complete before checking whether
2995 * any further dirty buffers exist.
2997 while (waitfor == MNT_WAIT && vp->v_track_write.bk_active) {
2998 vp->v_track_write.bk_waitflag = 1;
2999 error = tsleep(&vp->v_track_write,
3000 info.slpflag, "nfsfsync", info.slptimeo);
3003 * We have to be able to break out if this
3004 * is an 'intr' mount.
3006 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
3012 * Since we do not process pending signals,
3013 * once we get a PCATCH our tsleep() will no
3014 * longer sleep, switch to a fixed timeout
3017 if (info.slpflag == PCATCH) {
3019 info.slptimeo = 2 * hz;
3026 * Loop if we are flushing synchronous as well as committing,
3027 * and dirty buffers are still present. Otherwise we might livelock.
3029 } while (waitfor == MNT_WAIT && commit &&
3030 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3033 * The callbacks have to return a negative error to terminate the
3040 * Deal with any error collection
3042 if (np->n_flag & NWRITEERR) {
3043 error = np->n_error;
3044 np->n_flag &= ~NWRITEERR;
3052 nfs_flush_bp(struct buf *bp, void *data)
3054 struct nfs_flush_info *info = data;
3059 switch(info->mode) {
3062 if (info->loops && info->waitfor == MNT_WAIT) {
3063 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3065 int lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3066 if (info->slpflag & PCATCH)
3067 lkflags |= LK_PCATCH;
3068 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3072 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3075 KKASSERT(bp->b_vp == info->vp);
3077 if ((bp->b_flags & B_DELWRI) == 0)
3078 panic("nfs_fsync: not dirty");
3079 if (bp->b_flags & B_NEEDCOMMIT) {
3086 bp->b_flags |= B_ASYNC;
3088 VOP_BWRITE(bp->b_vp, bp);
3096 * Only process buffers in need of a commit which we can
3097 * immediately lock. This may prevent a buffer from being
3098 * committed, but the normal flush loop will block on the
3099 * same buffer so we shouldn't get into an endless loop.
3102 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3103 (B_DELWRI | B_NEEDCOMMIT) ||
3104 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
3109 KKASSERT(bp->b_vp == info->vp);
3113 * NOTE: we are not clearing B_DONE here, so we have
3114 * to do it later on in this routine if we intend to
3115 * initiate I/O on the bp.
3117 * Note: to avoid loopback deadlocks, we do not
3118 * assign b_runningbufspace.
3120 vfs_busy_pages(bp, 1);
3122 info->bvary[info->bvsize] = bp;
3123 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3124 if (info->bvsize == 0 || toff < info->beg_off)
3125 info->beg_off = toff;
3126 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3127 if (info->bvsize == 0 || toff > info->end_off)
3128 info->end_off = toff;
3130 if (info->bvsize == NFS_COMMITBVECSIZ) {
3131 error = nfs_flush_docommit(info, 0);
3132 KKASSERT(info->bvsize == 0);
3141 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3151 if (info->bvsize > 0) {
3153 * Commit data on the server, as required. Note that
3154 * nfs_commit will use the vnode's cred for the commit.
3155 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3157 bytes = info->end_off - info->beg_off;
3158 if (bytes > 0x40000000)
3163 retv = nfs_commit(vp, info->beg_off,
3164 (int)bytes, info->td);
3165 if (retv == NFSERR_STALEWRITEVERF)
3166 nfs_clearcommit(vp->v_mount);
3170 * Now, either mark the blocks I/O done or mark the
3171 * blocks dirty, depending on whether the commit
3174 for (i = 0; i < info->bvsize; ++i) {
3175 bp = info->bvary[i];
3176 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3179 * Error, leave B_DELWRI intact
3181 vfs_unbusy_pages(bp);
3185 * Success, remove B_DELWRI ( bundirty() ).
3187 * b_dirtyoff/b_dirtyend seem to be NFS
3188 * specific. We should probably move that
3189 * into bundirty(). XXX
3191 * We are faking an I/O write, we have to
3192 * start the transaction in order to
3193 * immediately biodone() it.
3196 bp->b_flags |= B_ASYNC;
3198 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3199 bp->b_dirtyoff = bp->b_dirtyend = 0;
3201 biodone(&bp->b_bio1);
3210 * NFS advisory byte-level locks.
3211 * Currently unsupported.
3213 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3217 nfs_advlock(struct vop_advlock_args *ap)
3219 struct nfsnode *np = VTONFS(ap->a_vp);
3222 * The following kludge is to allow diskless support to work
3223 * until a real NFS lockd is implemented. Basically, just pretend
3224 * that this is a local lock.
3226 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3230 * Print out the contents of an nfsnode.
3232 * nfs_print(struct vnode *a_vp)
3235 nfs_print(struct vop_print_args *ap)
3237 struct vnode *vp = ap->a_vp;
3238 struct nfsnode *np = VTONFS(vp);
3240 printf("tag VT_NFS, fileid %ld fsid 0x%x",
3241 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3242 if (vp->v_type == VFIFO)
3249 * Just call nfs_writebp() with the force argument set to 1.
3251 * NOTE: B_DONE may or may not be set in a_bp on call.
3253 * nfs_bwrite(struct vnode *a_bp)
3256 nfs_bwrite(struct vop_bwrite_args *ap)
3258 return (nfs_writebp(ap->a_bp, 1, curthread));
3262 * This is a clone of vn_bwrite(), except that it also handles the
3263 * B_NEEDCOMMIT flag. We set B_CACHE if this is a VMIO buffer.
3266 nfs_writebp(struct buf *bp, int force, struct thread *td)
3270 if (BUF_REFCNT(bp) == 0)
3271 panic("bwrite: buffer is not locked???");
3273 if (bp->b_flags & B_INVAL) {
3278 bp->b_flags |= B_CACHE;
3281 * Undirty the bp. We will redirty it later if the I/O fails.
3285 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3289 * Note: to avoid loopback deadlocks, we do not
3290 * assign b_runningbufspace.
3292 vfs_busy_pages(bp, 1);
3295 if (bp->b_flags & B_ASYNC) {
3296 vn_strategy(bp->b_vp, &bp->b_bio1);
3299 vn_strategy(bp->b_vp, &bp->b_bio1);
3300 error = biowait(bp);
3307 * nfs special file access vnode op.
3308 * Essentially just get vattr and then imitate iaccess() since the device is
3309 * local to the client.
3311 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
3312 * struct thread *a_td)
3315 nfsspec_access(struct vop_access_args *ap)
3319 struct ucred *cred = ap->a_cred;
3320 struct vnode *vp = ap->a_vp;
3321 mode_t mode = ap->a_mode;
3327 * Disallow write attempts on filesystems mounted read-only;
3328 * unless the file is a socket, fifo, or a block or character
3329 * device resident on the filesystem.
3331 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3332 switch (vp->v_type) {
3342 * If you're the super-user,
3343 * you always get access.
3345 if (cred->cr_uid == 0)
3348 error = VOP_GETATTR(vp, vap, ap->a_td);
3352 * Access check is based on only one of owner, group, public.
3353 * If not owner, then check group. If not a member of the
3354 * group, then check public access.
3356 if (cred->cr_uid != vap->va_uid) {
3358 gp = cred->cr_groups;
3359 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3360 if (vap->va_gid == *gp)
3366 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3371 * Read wrapper for special devices.
3373 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3374 * struct ucred *a_cred)
3377 nfsspec_read(struct vop_read_args *ap)
3379 struct nfsnode *np = VTONFS(ap->a_vp);
3385 getnanotime(&np->n_atim);
3386 return (VOCALL(spec_vnode_vops, &ap->a_head));
3390 * Write wrapper for special devices.
3392 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3393 * struct ucred *a_cred)
3396 nfsspec_write(struct vop_write_args *ap)
3398 struct nfsnode *np = VTONFS(ap->a_vp);
3404 getnanotime(&np->n_mtim);
3405 return (VOCALL(spec_vnode_vops, &ap->a_head));
3409 * Close wrapper for special devices.
3411 * Update the times on the nfsnode then do device close.
3413 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
3414 * struct thread *a_td)
3417 nfsspec_close(struct vop_close_args *ap)
3419 struct vnode *vp = ap->a_vp;
3420 struct nfsnode *np = VTONFS(vp);
3423 if (np->n_flag & (NACC | NUPD)) {
3425 if (vp->v_usecount == 1 &&
3426 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3428 if (np->n_flag & NACC)
3429 vattr.va_atime = np->n_atim;
3430 if (np->n_flag & NUPD)
3431 vattr.va_mtime = np->n_mtim;
3432 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
3435 return (VOCALL(spec_vnode_vops, &ap->a_head));
3439 * Read wrapper for fifos.
3441 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3442 * struct ucred *a_cred)
3445 nfsfifo_read(struct vop_read_args *ap)
3447 struct nfsnode *np = VTONFS(ap->a_vp);
3453 getnanotime(&np->n_atim);
3454 return (VOCALL(fifo_vnode_vops, &ap->a_head));
3458 * Write wrapper for fifos.
3460 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3461 * struct ucred *a_cred)
3464 nfsfifo_write(struct vop_write_args *ap)
3466 struct nfsnode *np = VTONFS(ap->a_vp);
3472 getnanotime(&np->n_mtim);
3473 return (VOCALL(fifo_vnode_vops, &ap->a_head));
3477 * Close wrapper for fifos.
3479 * Update the times on the nfsnode then do fifo close.
3481 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td)
3484 nfsfifo_close(struct vop_close_args *ap)
3486 struct vnode *vp = ap->a_vp;
3487 struct nfsnode *np = VTONFS(vp);
3491 if (np->n_flag & (NACC | NUPD)) {
3493 if (np->n_flag & NACC)
3495 if (np->n_flag & NUPD)
3498 if (vp->v_usecount == 1 &&
3499 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3501 if (np->n_flag & NACC)
3502 vattr.va_atime = np->n_atim;
3503 if (np->n_flag & NUPD)
3504 vattr.va_mtime = np->n_mtim;
3505 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
3508 return (VOCALL(fifo_vnode_vops, &ap->a_head));