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 1.57 2006/04/28 16:34:01 dillon Exp $
43 * vnode op calls for Sun NFS version 2 and 3
48 #include <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/resourcevar.h>
53 #include <sys/mount.h>
55 #include <sys/malloc.h>
57 #include <sys/namei.h>
58 #include <sys/nlookup.h>
59 #include <sys/socket.h>
60 #include <sys/vnode.h>
61 #include <sys/dirent.h>
62 #include <sys/fcntl.h>
63 #include <sys/lockf.h>
65 #include <sys/sysctl.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vm_zone.h>
74 #include <vfs/fifofs/fifo.h>
75 #include <vfs/ufs/dir.h>
85 #include "nfsm_subs.h"
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
91 #include <sys/thread2.h>
97 static int nfsspec_read (struct vop_read_args *);
98 static int nfsspec_write (struct vop_write_args *);
99 static int nfsfifo_read (struct vop_read_args *);
100 static int nfsfifo_write (struct vop_write_args *);
101 static int nfsspec_close (struct vop_close_args *);
102 static int nfsfifo_close (struct vop_close_args *);
103 #define nfs_poll vop_nopoll
104 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
105 static int nfs_lookup (struct vop_old_lookup_args *);
106 static int nfs_create (struct vop_old_create_args *);
107 static int nfs_mknod (struct vop_old_mknod_args *);
108 static int nfs_open (struct vop_open_args *);
109 static int nfs_close (struct vop_close_args *);
110 static int nfs_access (struct vop_access_args *);
111 static int nfs_getattr (struct vop_getattr_args *);
112 static int nfs_setattr (struct vop_setattr_args *);
113 static int nfs_read (struct vop_read_args *);
114 static int nfs_mmap (struct vop_mmap_args *);
115 static int nfs_fsync (struct vop_fsync_args *);
116 static int nfs_remove (struct vop_old_remove_args *);
117 static int nfs_link (struct vop_old_link_args *);
118 static int nfs_rename (struct vop_old_rename_args *);
119 static int nfs_mkdir (struct vop_old_mkdir_args *);
120 static int nfs_rmdir (struct vop_old_rmdir_args *);
121 static int nfs_symlink (struct vop_old_symlink_args *);
122 static int nfs_readdir (struct vop_readdir_args *);
123 static int nfs_bmap (struct vop_bmap_args *);
124 static int nfs_strategy (struct vop_strategy_args *);
125 static int nfs_lookitup (struct vnode *, const char *, int,
126 struct ucred *, struct thread *, struct nfsnode **);
127 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
128 static int nfsspec_access (struct vop_access_args *);
129 static int nfs_readlink (struct vop_readlink_args *);
130 static int nfs_print (struct vop_print_args *);
131 static int nfs_advlock (struct vop_advlock_args *);
132 static int nfs_bwrite (struct vop_bwrite_args *);
134 static int nfs_nresolve (struct vop_nresolve_args *);
136 * Global vfs data structures for nfs
138 struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
139 { &vop_default_desc, vop_defaultop },
140 { &vop_access_desc, (vnodeopv_entry_t) nfs_access },
141 { &vop_advlock_desc, (vnodeopv_entry_t) nfs_advlock },
142 { &vop_bmap_desc, (vnodeopv_entry_t) nfs_bmap },
143 { &vop_bwrite_desc, (vnodeopv_entry_t) nfs_bwrite },
144 { &vop_close_desc, (vnodeopv_entry_t) nfs_close },
145 { &vop_old_create_desc, (vnodeopv_entry_t) nfs_create },
146 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync },
147 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr },
148 { &vop_getpages_desc, (vnodeopv_entry_t) nfs_getpages },
149 { &vop_putpages_desc, (vnodeopv_entry_t) nfs_putpages },
150 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive },
151 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked },
152 { &vop_old_link_desc, (vnodeopv_entry_t) nfs_link },
153 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock },
154 { &vop_old_lookup_desc, (vnodeopv_entry_t) nfs_lookup },
155 { &vop_old_mkdir_desc, (vnodeopv_entry_t) nfs_mkdir },
156 { &vop_old_mknod_desc, (vnodeopv_entry_t) nfs_mknod },
157 { &vop_mmap_desc, (vnodeopv_entry_t) nfs_mmap },
158 { &vop_open_desc, (vnodeopv_entry_t) nfs_open },
159 { &vop_poll_desc, (vnodeopv_entry_t) nfs_poll },
160 { &vop_print_desc, (vnodeopv_entry_t) nfs_print },
161 { &vop_read_desc, (vnodeopv_entry_t) nfs_read },
162 { &vop_readdir_desc, (vnodeopv_entry_t) nfs_readdir },
163 { &vop_readlink_desc, (vnodeopv_entry_t) nfs_readlink },
164 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim },
165 { &vop_old_remove_desc, (vnodeopv_entry_t) nfs_remove },
166 { &vop_old_rename_desc, (vnodeopv_entry_t) nfs_rename },
167 { &vop_old_rmdir_desc, (vnodeopv_entry_t) nfs_rmdir },
168 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr },
169 { &vop_strategy_desc, (vnodeopv_entry_t) nfs_strategy },
170 { &vop_old_symlink_desc, (vnodeopv_entry_t) nfs_symlink },
171 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock },
172 { &vop_write_desc, (vnodeopv_entry_t) nfs_write },
174 { &vop_nresolve_desc, (vnodeopv_entry_t) nfs_nresolve },
179 * Special device vnode ops
181 struct vnodeopv_entry_desc nfsv2_specop_entries[] = {
182 { &vop_default_desc, (vnodeopv_entry_t) spec_vnoperate },
183 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access },
184 { &vop_close_desc, (vnodeopv_entry_t) nfsspec_close },
185 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync },
186 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr },
187 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive },
188 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked },
189 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock },
190 { &vop_print_desc, (vnodeopv_entry_t) nfs_print },
191 { &vop_read_desc, (vnodeopv_entry_t) nfsspec_read },
192 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim },
193 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr },
194 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock },
195 { &vop_write_desc, (vnodeopv_entry_t) nfsspec_write },
199 struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = {
200 { &vop_default_desc, (vnodeopv_entry_t) fifo_vnoperate },
201 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access },
202 { &vop_close_desc, (vnodeopv_entry_t) nfsfifo_close },
203 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync },
204 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr },
205 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive },
206 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked },
207 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock },
208 { &vop_print_desc, (vnodeopv_entry_t) nfs_print },
209 { &vop_read_desc, (vnodeopv_entry_t) nfsfifo_read },
210 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim },
211 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr },
212 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock },
213 { &vop_write_desc, (vnodeopv_entry_t) nfsfifo_write },
217 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
218 struct componentname *cnp,
220 static int nfs_removerpc (struct vnode *dvp, const char *name,
222 struct ucred *cred, struct thread *td);
223 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
224 int fnamelen, struct vnode *tdvp,
225 const char *tnameptr, int tnamelen,
226 struct ucred *cred, struct thread *td);
227 static int nfs_renameit (struct vnode *sdvp,
228 struct componentname *scnp,
229 struct sillyrename *sp);
234 extern u_int32_t nfs_true, nfs_false;
235 extern u_int32_t nfs_xdrneg1;
236 extern struct nfsstats nfsstats;
237 extern nfstype nfsv3_type[9];
238 struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON];
239 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
240 int nfs_numasync = 0;
242 SYSCTL_DECL(_vfs_nfs);
244 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
245 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
246 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
248 static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
249 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
250 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache timeout");
252 static int nfsv3_commit_on_close = 0;
253 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
254 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
256 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
257 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
259 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
260 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
263 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
264 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
265 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
267 nfs3_access_otw(struct vnode *vp, int wmode,
268 struct thread *td, struct ucred *cred)
272 int error = 0, attrflag;
274 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
275 caddr_t bpos, dpos, cp2;
279 struct nfsnode *np = VTONFS(vp);
281 nfsstats.rpccnt[NFSPROC_ACCESS]++;
282 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
284 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
285 *tl = txdr_unsigned(wmode);
286 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
287 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
289 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
290 rmode = fxdr_unsigned(u_int32_t, *tl);
292 np->n_modeuid = cred->cr_uid;
293 np->n_modestamp = mycpu->gd_time_seconds;
301 * nfs access vnode op.
302 * For nfs version 2, just return ok. File accesses may fail later.
303 * For nfs version 3, use the access rpc to check accessibility. If file modes
304 * are changed on the server, accesses might still fail later.
306 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
307 * struct thread *a_td)
310 nfs_access(struct vop_access_args *ap)
312 struct vnode *vp = ap->a_vp;
314 u_int32_t mode, wmode;
315 int v3 = NFS_ISV3(vp);
316 struct nfsnode *np = VTONFS(vp);
319 * Disallow write attempts on filesystems mounted read-only;
320 * unless the file is a socket, fifo, or a block or character
321 * device resident on the filesystem.
323 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
324 switch (vp->v_type) {
334 * For nfs v3, check to see if we have done this recently, and if
335 * so return our cached result instead of making an ACCESS call.
336 * If not, do an access rpc, otherwise you are stuck emulating
337 * ufs_access() locally using the vattr. This may not be correct,
338 * since the server may apply other access criteria such as
339 * client uid-->server uid mapping that we do not know about.
342 if (ap->a_mode & VREAD)
343 mode = NFSV3ACCESS_READ;
346 if (vp->v_type != VDIR) {
347 if (ap->a_mode & VWRITE)
348 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
349 if (ap->a_mode & VEXEC)
350 mode |= NFSV3ACCESS_EXECUTE;
352 if (ap->a_mode & VWRITE)
353 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
355 if (ap->a_mode & VEXEC)
356 mode |= NFSV3ACCESS_LOOKUP;
358 /* XXX safety belt, only make blanket request if caching */
359 if (nfsaccess_cache_timeout > 0) {
360 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
361 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
362 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
368 * Does our cached result allow us to give a definite yes to
371 if (np->n_modestamp &&
372 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
373 (ap->a_cred->cr_uid == np->n_modeuid) &&
374 ((np->n_mode & mode) == mode)) {
375 nfsstats.accesscache_hits++;
378 * Either a no, or a don't know. Go to the wire.
380 nfsstats.accesscache_misses++;
381 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
383 if ((np->n_mode & mode) != mode) {
389 if ((error = nfsspec_access(ap)) != 0)
393 * Attempt to prevent a mapped root from accessing a file
394 * which it shouldn't. We try to read a byte from the file
395 * if the user is root and the file is not zero length.
396 * After calling nfsspec_access, we should have the correct
399 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
400 && VTONFS(vp)->n_size > 0) {
407 auio.uio_iov = &aiov;
411 auio.uio_segflg = UIO_SYSSPACE;
412 auio.uio_rw = UIO_READ;
413 auio.uio_td = ap->a_td;
415 if (vp->v_type == VREG) {
416 error = nfs_readrpc(vp, &auio);
417 } else if (vp->v_type == VDIR) {
419 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
421 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
422 error = nfs_readdirrpc(vp, &auio);
424 } else if (vp->v_type == VLNK) {
425 error = nfs_readlinkrpc(vp, &auio);
432 * [re]record creds for reading and/or writing if access
433 * was granted. Assume the NFS server will grant read access
434 * for execute requests.
437 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) {
440 crfree(np->n_rucred);
441 np->n_rucred = ap->a_cred;
443 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
446 crfree(np->n_wucred);
447 np->n_wucred = ap->a_cred;
455 * Check to see if the type is ok
456 * and that deletion is not in progress.
457 * For paged in text files, you will need to flush the page cache
458 * if consistency is lost.
460 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
461 * struct thread *a_td)
465 nfs_open(struct vop_open_args *ap)
467 struct vnode *vp = ap->a_vp;
468 struct nfsnode *np = VTONFS(vp);
472 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
474 printf("open eacces vtyp=%d\n",vp->v_type);
480 * Clear the attribute cache only if opening with write access. It
481 * is unclear if we should do this at all here, but we certainly
482 * should not clear the cache unconditionally simply because a file
485 if (ap->a_mode & FWRITE)
489 * For normal NFS, reconcile changes made locally verses
490 * changes made remotely. Note that VOP_GETATTR only goes
491 * to the wire if the cached attribute has timed out or been
494 * If local modifications have been made clear the attribute
495 * cache to force an attribute and modified time check. If
496 * GETATTR detects that the file has been changed by someone
497 * other then us it will set NRMODIFIED.
499 * If we are opening a directory and local changes have been
500 * made we have to invalidate the cache in order to ensure
501 * that we get the most up-to-date information from the
504 if (np->n_flag & NLMODIFIED) {
506 if (vp->v_type == VDIR) {
507 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
513 error = VOP_GETATTR(vp, &vattr, ap->a_td);
516 if (np->n_flag & NRMODIFIED) {
517 if (vp->v_type == VDIR)
519 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
522 np->n_flag &= ~NRMODIFIED;
525 return (vop_stdopen(ap));
530 * What an NFS client should do upon close after writing is a debatable issue.
531 * Most NFS clients push delayed writes to the server upon close, basically for
533 * 1 - So that any write errors may be reported back to the client process
534 * doing the close system call. By far the two most likely errors are
535 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
536 * 2 - To put a worst case upper bound on cache inconsistency between
537 * multiple clients for the file.
538 * There is also a consistency problem for Version 2 of the protocol w.r.t.
539 * not being able to tell if other clients are writing a file concurrently,
540 * since there is no way of knowing if the changed modify time in the reply
541 * is only due to the write for this client.
542 * (NFS Version 3 provides weak cache consistency data in the reply that
543 * should be sufficient to detect and handle this case.)
545 * The current code does the following:
546 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
547 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
548 * or commit them (this satisfies 1 and 2 except for the
549 * case where the server crashes after this close but
550 * before the commit RPC, which is felt to be "good
551 * enough". Changing the last argument to nfs_flush() to
552 * a 1 would force a commit operation, if it is felt a
553 * commit is necessary now.
554 * for NQNFS - do nothing now, since 2 is dealt with via leases and
555 * 1 should be dealt with via an fsync() system call for
556 * cases where write errors are important.
558 * nfs_close(struct vnodeop_desc *a_desc, struct vnode *a_vp, int a_fflag,
559 * struct ucred *a_cred, struct thread *a_td)
563 nfs_close(struct vop_close_args *ap)
565 struct vnode *vp = ap->a_vp;
566 struct nfsnode *np = VTONFS(vp);
569 if (vp->v_type == VREG) {
570 if (np->n_flag & NLMODIFIED) {
573 * Under NFSv3 we have dirty buffers to dispose of. We
574 * must flush them to the NFS server. We have the option
575 * of waiting all the way through the commit rpc or just
576 * waiting for the initial write. The default is to only
577 * wait through the initial write so the data is in the
578 * server's cache, which is roughly similar to the state
579 * a standard disk subsystem leaves the file in on close().
581 * We cannot clear the NLMODIFIED bit in np->n_flag due to
582 * potential races with other processes, and certainly
583 * cannot clear it if we don't commit.
585 int cm = nfsv3_commit_on_close ? 1 : 0;
586 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
587 /* np->n_flag &= ~NLMODIFIED; */
589 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
593 if (np->n_flag & NWRITEERR) {
594 np->n_flag &= ~NWRITEERR;
603 * nfs getattr call from vfs.
605 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred,
606 * struct thread *a_td)
609 nfs_getattr(struct vop_getattr_args *ap)
611 struct vnode *vp = ap->a_vp;
612 struct nfsnode *np = VTONFS(vp);
618 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
619 int v3 = NFS_ISV3(vp);
622 * Update local times for special files.
624 if (np->n_flag & (NACC | NUPD))
627 * First look in the cache.
629 if (nfs_getattrcache(vp, ap->a_vap) == 0)
632 if (v3 && nfsaccess_cache_timeout > 0) {
633 nfsstats.accesscache_misses++;
634 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK));
635 if (nfs_getattrcache(vp, ap->a_vap) == 0)
639 nfsstats.rpccnt[NFSPROC_GETATTR]++;
640 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
642 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK));
644 nfsm_loadattr(vp, ap->a_vap);
654 * nfs_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp,
655 * struct vattr *a_vap, struct ucred *a_cred,
656 * struct thread *a_td)
659 nfs_setattr(struct vop_setattr_args *ap)
661 struct vnode *vp = ap->a_vp;
662 struct nfsnode *np = VTONFS(vp);
663 struct vattr *vap = ap->a_vap;
672 * Setting of flags is not supported.
674 if (vap->va_flags != VNOVAL)
678 * Disallow write attempts if the filesystem is mounted read-only.
680 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
681 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
682 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
683 (vp->v_mount->mnt_flag & MNT_RDONLY))
685 if (vap->va_size != VNOVAL) {
686 switch (vp->v_type) {
693 if (vap->va_mtime.tv_sec == VNOVAL &&
694 vap->va_atime.tv_sec == VNOVAL &&
695 vap->va_mode == (mode_t)VNOVAL &&
696 vap->va_uid == (uid_t)VNOVAL &&
697 vap->va_gid == (gid_t)VNOVAL)
699 vap->va_size = VNOVAL;
703 * Disallow write attempts if the filesystem is
706 if (vp->v_mount->mnt_flag & MNT_RDONLY)
710 * This is nasty. The RPCs we send to flush pending
711 * data often return attribute information which is
712 * cached via a callback to nfs_loadattrcache(), which
713 * has the effect of changing our notion of the file
714 * size. Due to flushed appends and other operations
715 * the file size can be set to virtually anything,
716 * including values that do not match either the old
717 * or intended file size.
719 * When this condition is detected we must loop to
720 * try the operation again. Hopefully no more
721 * flushing is required on the loop so it works the
722 * second time around. THIS CASE ALMOST ALWAYS
727 error = nfs_meta_setsize(vp, ap->a_td, vap->va_size);
729 if (np->n_flag & NLMODIFIED) {
730 if (vap->va_size == 0)
731 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
733 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
736 * note: this loop case almost always happens at
737 * least once per truncation.
739 if (error == 0 && np->n_size != vap->va_size)
741 np->n_vattr.va_size = vap->va_size;
744 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
745 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) &&
746 vp->v_type == VREG &&
747 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR
751 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
754 * Sanity check if a truncation was issued. This should only occur
755 * if multiple processes are racing on the same file.
757 if (error == 0 && vap->va_size != VNOVAL &&
758 np->n_size != vap->va_size) {
759 printf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size);
762 if (error && vap->va_size != VNOVAL) {
763 np->n_size = np->n_vattr.va_size = tsize;
764 vnode_pager_setsize(vp, np->n_size);
770 * Do an nfs setattr rpc.
773 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
774 struct ucred *cred, struct thread *td)
776 struct nfsv2_sattr *sp;
777 struct nfsnode *np = VTONFS(vp);
780 caddr_t bpos, dpos, cp2;
782 int error = 0, wccflag = NFSV3_WCCRATTR;
783 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
784 int v3 = NFS_ISV3(vp);
786 nfsstats.rpccnt[NFSPROC_SETATTR]++;
787 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
790 nfsm_v3attrbuild(vap, TRUE);
791 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
794 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
795 if (vap->va_mode == (mode_t)VNOVAL)
796 sp->sa_mode = nfs_xdrneg1;
798 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
799 if (vap->va_uid == (uid_t)VNOVAL)
800 sp->sa_uid = nfs_xdrneg1;
802 sp->sa_uid = txdr_unsigned(vap->va_uid);
803 if (vap->va_gid == (gid_t)VNOVAL)
804 sp->sa_gid = nfs_xdrneg1;
806 sp->sa_gid = txdr_unsigned(vap->va_gid);
807 sp->sa_size = txdr_unsigned(vap->va_size);
808 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
809 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
811 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
814 nfsm_wcc_data(vp, wccflag);
816 nfsm_loadattr(vp, (struct vattr *)0);
823 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
824 * nfs_lookup() until all remaining new api calls are implemented.
826 * Resolve a namecache entry. This function is passed a locked ncp and
827 * must call cache_setvp() on it as appropriate to resolve the entry.
830 nfs_nresolve(struct vop_nresolve_args *ap)
832 struct thread *td = curthread;
833 struct namecache *ncp;
844 /******NFSM MACROS********/
845 struct mbuf *mb, *mrep, *mreq, *mb2, *md;
846 caddr_t bpos, dpos, cp, cp2;
853 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp);
854 dvp = ncp->nc_parent->nc_vp;
855 if ((error = vget(dvp, LK_SHARED, td)) != 0)
860 nfsstats.lookupcache_misses++;
861 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
863 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
864 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
866 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN);
867 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred);
870 * Cache negatve lookups to reduce NFS traffic, but use
871 * a fast timeout. Otherwise use a timeout of 1 tick.
872 * XXX we should add a namecache flag for no-caching
873 * to uncache the negative hit as soon as possible, but
874 * we cannot simply destroy the entry because it is used
875 * as a placeholder by the caller.
877 if (error == ENOENT) {
880 if (nfsneg_cache_timeout)
881 nticks = nfsneg_cache_timeout * hz;
884 cache_setvp(ncp, NULL);
885 cache_settimeout(ncp, nticks);
887 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
893 * Success, get the file handle, do various checks, and load
894 * post-operation data from the reply packet. Theoretically
895 * we should never be looking up "." so, theoretically, we
896 * should never get the same file handle as our directory. But
897 * we check anyway. XXX
899 * Note that no timeout is set for the positive cache hit. We
900 * assume, theoretically, that ESTALE returns will be dealt with
901 * properly to handle NFS races and in anycase we cannot depend
902 * on a timeout to deal with NFS open/create/excl issues so instead
903 * of a bad hack here the rest of the NFS client code needs to do
906 nfsm_getfh(fhp, fhsize, v3);
909 if (NFS_CMPFH(np, fhp, fhsize)) {
913 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
922 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK);
923 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
925 nfsm_loadattr(nvp, NULL);
927 cache_setvp(ncp, nvp);
941 * 'cached' nfs directory lookup
943 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
945 * nfs_lookup(struct vnodeop_desc *a_desc, struct vnode *a_dvp,
946 * struct vnode **a_vpp, struct componentname *a_cnp)
949 nfs_lookup(struct vop_old_lookup_args *ap)
951 struct componentname *cnp = ap->a_cnp;
952 struct vnode *dvp = ap->a_dvp;
953 struct vnode **vpp = ap->a_vpp;
954 int flags = cnp->cn_flags;
959 struct nfsmount *nmp;
960 caddr_t bpos, dpos, cp2;
961 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
965 int lockparent, wantparent, error = 0, attrflag, fhsize;
966 int v3 = NFS_ISV3(dvp);
967 struct thread *td = cnp->cn_td;
970 * Read-only mount check and directory check.
973 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
974 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
977 if (dvp->v_type != VDIR)
981 * Look it up in the cache. Note that ENOENT is only returned if we
982 * previously entered a negative hit (see later on). The additional
983 * nfsneg_cache_timeout check causes previously cached results to
984 * be instantly ignored if the negative caching is turned off.
986 lockparent = flags & CNP_LOCKPARENT;
987 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
988 nmp = VFSTONFS(dvp->v_mount);
996 nfsstats.lookupcache_misses++;
997 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
998 len = cnp->cn_namelen;
999 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1000 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1001 nfsm_fhtom(dvp, v3);
1002 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1003 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred);
1005 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1009 nfsm_getfh(fhp, fhsize, v3);
1012 * Handle RENAME case...
1014 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1015 if (NFS_CMPFH(np, fhp, fhsize)) {
1019 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1026 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1027 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1029 nfsm_loadattr(newvp, (struct vattr *)0);
1033 VOP_UNLOCK(dvp, 0, td);
1034 cnp->cn_flags |= CNP_PDIRUNLOCK;
1039 if (flags & CNP_ISDOTDOT) {
1040 VOP_UNLOCK(dvp, 0, td);
1041 cnp->cn_flags |= CNP_PDIRUNLOCK;
1042 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1044 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
1045 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1046 return (error); /* NOTE: return error from nget */
1050 error = vn_lock(dvp, LK_EXCLUSIVE, td);
1055 cnp->cn_flags |= CNP_PDIRUNLOCK;
1057 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1061 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1067 VOP_UNLOCK(dvp, 0, td);
1068 cnp->cn_flags |= CNP_PDIRUNLOCK;
1073 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1074 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1076 nfsm_loadattr(newvp, (struct vattr *)0);
1078 /* XXX MOVE TO nfs_nremove() */
1079 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1080 cnp->cn_nameiop != NAMEI_DELETE) {
1081 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1088 if (newvp != NULLVP) {
1092 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1093 cnp->cn_nameiop == NAMEI_RENAME) &&
1096 VOP_UNLOCK(dvp, 0, td);
1097 cnp->cn_flags |= CNP_PDIRUNLOCK;
1099 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1102 error = EJUSTRETURN;
1110 * Just call nfs_bioread() to do the work.
1112 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1113 * struct ucred *a_cred)
1116 nfs_read(struct vop_read_args *ap)
1118 struct vnode *vp = ap->a_vp;
1120 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1121 switch (vp->v_type) {
1123 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1134 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1137 nfs_readlink(struct vop_readlink_args *ap)
1139 struct vnode *vp = ap->a_vp;
1141 if (vp->v_type != VLNK)
1143 return (nfs_bioread(vp, ap->a_uio, 0));
1147 * Do a readlink rpc.
1148 * Called by nfs_doio() from below the buffer cache.
1151 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1156 caddr_t bpos, dpos, cp2;
1157 int error = 0, len, attrflag;
1158 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1159 int v3 = NFS_ISV3(vp);
1161 nfsstats.rpccnt[NFSPROC_READLINK]++;
1162 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1164 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK));
1166 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1168 nfsm_strsiz(len, NFS_MAXPATHLEN);
1169 if (len == NFS_MAXPATHLEN) {
1170 struct nfsnode *np = VTONFS(vp);
1171 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1174 nfsm_mtouio(uiop, len);
1186 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1191 caddr_t bpos, dpos, cp2;
1192 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1193 struct nfsmount *nmp;
1194 int error = 0, len, retlen, tsiz, eof, attrflag;
1195 int v3 = NFS_ISV3(vp);
1200 nmp = VFSTONFS(vp->v_mount);
1201 tsiz = uiop->uio_resid;
1202 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1205 nfsstats.rpccnt[NFSPROC_READ]++;
1206 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1207 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1209 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1211 txdr_hyper(uiop->uio_offset, tl);
1212 *(tl + 2) = txdr_unsigned(len);
1214 *tl++ = txdr_unsigned(uiop->uio_offset);
1215 *tl++ = txdr_unsigned(len);
1218 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ));
1220 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1225 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1226 eof = fxdr_unsigned(int, *(tl + 1));
1228 nfsm_loadattr(vp, (struct vattr *)0);
1229 nfsm_strsiz(retlen, nmp->nm_rsize);
1230 nfsm_mtouio(uiop, retlen);
1234 if (eof || retlen == 0) {
1237 } else if (retlen < len) {
1249 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1253 int32_t t1, t2, backup;
1254 caddr_t bpos, dpos, cp2;
1255 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1256 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1257 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1258 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1261 if (uiop->uio_iovcnt != 1)
1262 panic("nfs: writerpc iovcnt > 1");
1265 tsiz = uiop->uio_resid;
1266 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1269 nfsstats.rpccnt[NFSPROC_WRITE]++;
1270 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1271 nfsm_reqhead(vp, NFSPROC_WRITE,
1272 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1275 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1276 txdr_hyper(uiop->uio_offset, tl);
1278 *tl++ = txdr_unsigned(len);
1279 *tl++ = txdr_unsigned(*iomode);
1280 *tl = txdr_unsigned(len);
1284 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1285 /* Set both "begin" and "current" to non-garbage. */
1286 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1287 *tl++ = x; /* "begin offset" */
1288 *tl++ = x; /* "current offset" */
1289 x = txdr_unsigned(len);
1290 *tl++ = x; /* total to this offset */
1291 *tl = x; /* size of this write */
1293 nfsm_uiotom(uiop, len);
1294 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE));
1297 * The write RPC returns a before and after mtime. The
1298 * nfsm_wcc_data() macro checks the before n_mtime
1299 * against the before time and stores the after time
1300 * in the nfsnode's cached vattr and n_mtime field.
1301 * The NRMODIFIED bit will be set if the before
1302 * time did not match the original mtime.
1304 wccflag = NFSV3_WCCCHK;
1305 nfsm_wcc_data(vp, wccflag);
1307 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1308 + NFSX_V3WRITEVERF);
1309 rlen = fxdr_unsigned(int, *tl++);
1314 } else if (rlen < len) {
1315 backup = len - rlen;
1316 uiop->uio_iov->iov_base -= backup;
1317 uiop->uio_iov->iov_len += backup;
1318 uiop->uio_offset -= backup;
1319 uiop->uio_resid += backup;
1322 commit = fxdr_unsigned(int, *tl++);
1325 * Return the lowest committment level
1326 * obtained by any of the RPCs.
1328 if (committed == NFSV3WRITE_FILESYNC)
1330 else if (committed == NFSV3WRITE_DATASYNC &&
1331 commit == NFSV3WRITE_UNSTABLE)
1333 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1334 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1336 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1337 } else if (bcmp((caddr_t)tl,
1338 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1340 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1345 nfsm_loadattr(vp, (struct vattr *)0);
1353 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1354 committed = NFSV3WRITE_FILESYNC;
1355 *iomode = committed;
1357 uiop->uio_resid = tsiz;
1363 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1364 * mode set to specify the file type and the size field for rdev.
1367 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1370 struct nfsv2_sattr *sp;
1374 struct vnode *newvp = (struct vnode *)0;
1375 struct nfsnode *np = (struct nfsnode *)0;
1379 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1380 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1382 int v3 = NFS_ISV3(dvp);
1384 if (vap->va_type == VCHR || vap->va_type == VBLK)
1385 rdev = txdr_unsigned(vap->va_rdev);
1386 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1389 return (EOPNOTSUPP);
1391 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1394 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1395 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1396 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1397 nfsm_fhtom(dvp, v3);
1398 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1400 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1401 *tl++ = vtonfsv3_type(vap->va_type);
1402 nfsm_v3attrbuild(vap, FALSE);
1403 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1404 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1405 *tl++ = txdr_unsigned(umajor(vap->va_rdev));
1406 *tl = txdr_unsigned(uminor(vap->va_rdev));
1409 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1410 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1411 sp->sa_uid = nfs_xdrneg1;
1412 sp->sa_gid = nfs_xdrneg1;
1414 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1415 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1417 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred);
1419 nfsm_mtofh(dvp, newvp, v3, gotvp);
1423 newvp = (struct vnode *)0;
1425 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1426 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1432 nfsm_wcc_data(dvp, wccflag);
1441 VTONFS(dvp)->n_flag |= NLMODIFIED;
1443 VTONFS(dvp)->n_attrstamp = 0;
1449 * just call nfs_mknodrpc() to do the work.
1451 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1452 * struct componentname *a_cnp, struct vattr *a_vap)
1456 nfs_mknod(struct vop_old_mknod_args *ap)
1458 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1461 static u_long create_verf;
1463 * nfs file create call
1465 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1466 * struct componentname *a_cnp, struct vattr *a_vap)
1469 nfs_create(struct vop_old_create_args *ap)
1471 struct vnode *dvp = ap->a_dvp;
1472 struct vattr *vap = ap->a_vap;
1473 struct componentname *cnp = ap->a_cnp;
1474 struct nfsv2_sattr *sp;
1478 struct nfsnode *np = (struct nfsnode *)0;
1479 struct vnode *newvp = (struct vnode *)0;
1480 caddr_t bpos, dpos, cp2;
1481 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1482 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1484 int v3 = NFS_ISV3(dvp);
1487 * Oops, not for me..
1489 if (vap->va_type == VSOCK)
1490 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1492 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1495 if (vap->va_vaflags & VA_EXCLUSIVE)
1498 nfsstats.rpccnt[NFSPROC_CREATE]++;
1499 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1500 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1501 nfsm_fhtom(dvp, v3);
1502 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1504 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1505 if (fmode & O_EXCL) {
1506 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1507 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1509 if (!TAILQ_EMPTY(&in_ifaddrhead))
1510 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1513 *tl++ = create_verf;
1514 *tl = ++create_verf;
1516 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1517 nfsm_v3attrbuild(vap, FALSE);
1520 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1521 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1522 sp->sa_uid = nfs_xdrneg1;
1523 sp->sa_gid = nfs_xdrneg1;
1525 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1526 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1528 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred);
1530 nfsm_mtofh(dvp, newvp, v3, gotvp);
1534 newvp = (struct vnode *)0;
1536 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1537 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1543 nfsm_wcc_data(dvp, wccflag);
1547 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1553 } else if (v3 && (fmode & O_EXCL)) {
1555 * We are normally called with only a partially initialized
1556 * VAP. Since the NFSv3 spec says that server may use the
1557 * file attributes to store the verifier, the spec requires
1558 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1559 * in atime, but we can't really assume that all servers will
1560 * so we ensure that our SETATTR sets both atime and mtime.
1562 if (vap->va_mtime.tv_sec == VNOVAL)
1563 vfs_timestamp(&vap->va_mtime);
1564 if (vap->va_atime.tv_sec == VNOVAL)
1565 vap->va_atime = vap->va_mtime;
1566 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1570 * The new np may have enough info for access
1571 * checks, make sure rucred and wucred are
1572 * initialized for read and write rpc's.
1575 if (np->n_rucred == NULL)
1576 np->n_rucred = crhold(cnp->cn_cred);
1577 if (np->n_wucred == NULL)
1578 np->n_wucred = crhold(cnp->cn_cred);
1581 VTONFS(dvp)->n_flag |= NLMODIFIED;
1583 VTONFS(dvp)->n_attrstamp = 0;
1588 * nfs file remove call
1589 * To try and make nfs semantics closer to ufs semantics, a file that has
1590 * other processes using the vnode is renamed instead of removed and then
1591 * removed later on the last close.
1592 * - If v_usecount > 1
1593 * If a rename is not already in the works
1594 * call nfs_sillyrename() to set it up
1598 * nfs_remove(struct vnodeop_desc *a_desc, struct vnode *a_dvp,
1599 * struct vnode *a_vp, struct componentname *a_cnp)
1602 nfs_remove(struct vop_old_remove_args *ap)
1604 struct vnode *vp = ap->a_vp;
1605 struct vnode *dvp = ap->a_dvp;
1606 struct componentname *cnp = ap->a_cnp;
1607 struct nfsnode *np = VTONFS(vp);
1612 if (vp->v_usecount < 1)
1613 panic("nfs_remove: bad v_usecount");
1615 if (vp->v_type == VDIR)
1617 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1618 VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 &&
1619 vattr.va_nlink > 1)) {
1621 * throw away biocache buffers, mainly to avoid
1622 * unnecessary delayed writes later.
1624 error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1);
1627 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1628 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1630 * Kludge City: If the first reply to the remove rpc is lost..
1631 * the reply to the retransmitted request will be ENOENT
1632 * since the file was in fact removed
1633 * Therefore, we cheat and return success.
1635 if (error == ENOENT)
1637 } else if (!np->n_sillyrename) {
1638 error = nfs_sillyrename(dvp, vp, cnp);
1640 np->n_attrstamp = 0;
1645 * nfs file remove rpc called from nfs_inactive
1648 nfs_removeit(struct sillyrename *sp)
1650 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1655 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1658 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1659 struct ucred *cred, struct thread *td)
1664 caddr_t bpos, dpos, cp2;
1665 int error = 0, wccflag = NFSV3_WCCRATTR;
1666 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1667 int v3 = NFS_ISV3(dvp);
1669 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1670 nfsm_reqhead(dvp, NFSPROC_REMOVE,
1671 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1672 nfsm_fhtom(dvp, v3);
1673 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1674 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1676 nfsm_wcc_data(dvp, wccflag);
1679 VTONFS(dvp)->n_flag |= NLMODIFIED;
1681 VTONFS(dvp)->n_attrstamp = 0;
1686 * nfs file rename call
1688 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1689 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1690 * struct vnode *a_tvp, struct componentname *a_tcnp)
1693 nfs_rename(struct vop_old_rename_args *ap)
1695 struct vnode *fvp = ap->a_fvp;
1696 struct vnode *tvp = ap->a_tvp;
1697 struct vnode *fdvp = ap->a_fdvp;
1698 struct vnode *tdvp = ap->a_tdvp;
1699 struct componentname *tcnp = ap->a_tcnp;
1700 struct componentname *fcnp = ap->a_fcnp;
1703 /* Check for cross-device rename */
1704 if ((fvp->v_mount != tdvp->v_mount) ||
1705 (tvp && (fvp->v_mount != tvp->v_mount))) {
1711 * We have to flush B_DELWRI data prior to renaming
1712 * the file. If we don't, the delayed-write buffers
1713 * can be flushed out later after the file has gone stale
1714 * under NFSV3. NFSV2 does not have this problem because
1715 * ( as far as I can tell ) it flushes dirty buffers more
1719 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td);
1721 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td);
1724 * If the tvp exists and is in use, sillyrename it before doing the
1725 * rename of the new file over it.
1727 * XXX Can't sillyrename a directory.
1729 * We do not attempt to do any namecache purges in this old API
1730 * routine. The new API compat functions have access to the actual
1731 * namecache structures and will do it for us.
1733 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1734 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1741 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1742 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1755 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1757 if (error == ENOENT)
1763 * nfs file rename rpc called from nfs_remove() above
1766 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1767 struct sillyrename *sp)
1769 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1770 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1774 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1777 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1778 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1779 struct ucred *cred, struct thread *td)
1784 caddr_t bpos, dpos, cp2;
1785 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1786 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1787 int v3 = NFS_ISV3(fdvp);
1789 nfsstats.rpccnt[NFSPROC_RENAME]++;
1790 nfsm_reqhead(fdvp, NFSPROC_RENAME,
1791 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1792 nfsm_rndup(tnamelen));
1793 nfsm_fhtom(fdvp, v3);
1794 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1795 nfsm_fhtom(tdvp, v3);
1796 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1797 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1799 nfsm_wcc_data(fdvp, fwccflag);
1800 nfsm_wcc_data(tdvp, twccflag);
1804 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1805 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1807 VTONFS(fdvp)->n_attrstamp = 0;
1809 VTONFS(tdvp)->n_attrstamp = 0;
1814 * nfs hard link create call
1816 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1817 * struct componentname *a_cnp)
1820 nfs_link(struct vop_old_link_args *ap)
1822 struct vnode *vp = ap->a_vp;
1823 struct vnode *tdvp = ap->a_tdvp;
1824 struct componentname *cnp = ap->a_cnp;
1828 caddr_t bpos, dpos, cp2;
1829 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1830 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1833 if (vp->v_mount != tdvp->v_mount) {
1838 * Push all writes to the server, so that the attribute cache
1839 * doesn't get "out of sync" with the server.
1840 * XXX There should be a better way!
1842 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td);
1845 nfsstats.rpccnt[NFSPROC_LINK]++;
1846 nfsm_reqhead(vp, NFSPROC_LINK,
1847 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1849 nfsm_fhtom(tdvp, v3);
1850 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1851 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred);
1853 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1854 nfsm_wcc_data(tdvp, wccflag);
1858 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1860 VTONFS(vp)->n_attrstamp = 0;
1862 VTONFS(tdvp)->n_attrstamp = 0;
1864 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1866 if (error == EEXIST)
1872 * nfs symbolic link create call
1874 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1875 * struct componentname *a_cnp, struct vattr *a_vap,
1879 nfs_symlink(struct vop_old_symlink_args *ap)
1881 struct vnode *dvp = ap->a_dvp;
1882 struct vattr *vap = ap->a_vap;
1883 struct componentname *cnp = ap->a_cnp;
1884 struct nfsv2_sattr *sp;
1888 caddr_t bpos, dpos, cp2;
1889 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1890 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1891 struct vnode *newvp = (struct vnode *)0;
1892 int v3 = NFS_ISV3(dvp);
1894 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1895 slen = strlen(ap->a_target);
1896 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1897 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1898 nfsm_fhtom(dvp, v3);
1899 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1901 nfsm_v3attrbuild(vap, FALSE);
1903 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1905 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1906 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1907 sp->sa_uid = nfs_xdrneg1;
1908 sp->sa_gid = nfs_xdrneg1;
1909 sp->sa_size = nfs_xdrneg1;
1910 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1911 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1915 * Issue the NFS request and get the rpc response.
1917 * Only NFSv3 responses returning an error of 0 actually return
1918 * a file handle that can be converted into newvp without having
1919 * to do an extra lookup rpc.
1921 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred);
1924 nfsm_mtofh(dvp, newvp, v3, gotvp);
1925 nfsm_wcc_data(dvp, wccflag);
1929 * out code jumps -> here, mrep is also freed.
1936 * If we get an EEXIST error, silently convert it to no-error
1937 * in case of an NFS retry.
1939 if (error == EEXIST)
1943 * If we do not have (or no longer have) an error, and we could
1944 * not extract the newvp from the response due to the request being
1945 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1946 * to obtain a newvp to return.
1948 if (error == 0 && newvp == NULL) {
1949 struct nfsnode *np = NULL;
1951 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1952 cnp->cn_cred, cnp->cn_td, &np);
1962 VTONFS(dvp)->n_flag |= NLMODIFIED;
1964 VTONFS(dvp)->n_attrstamp = 0;
1971 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1972 * struct componentname *a_cnp, struct vattr *a_vap)
1975 nfs_mkdir(struct vop_old_mkdir_args *ap)
1977 struct vnode *dvp = ap->a_dvp;
1978 struct vattr *vap = ap->a_vap;
1979 struct componentname *cnp = ap->a_cnp;
1980 struct nfsv2_sattr *sp;
1985 struct nfsnode *np = (struct nfsnode *)0;
1986 struct vnode *newvp = (struct vnode *)0;
1987 caddr_t bpos, dpos, cp2;
1988 int error = 0, wccflag = NFSV3_WCCRATTR;
1990 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1992 int v3 = NFS_ISV3(dvp);
1994 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1997 len = cnp->cn_namelen;
1998 nfsstats.rpccnt[NFSPROC_MKDIR]++;
1999 nfsm_reqhead(dvp, NFSPROC_MKDIR,
2000 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2001 nfsm_fhtom(dvp, v3);
2002 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2004 nfsm_v3attrbuild(vap, FALSE);
2006 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2007 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2008 sp->sa_uid = nfs_xdrneg1;
2009 sp->sa_gid = nfs_xdrneg1;
2010 sp->sa_size = nfs_xdrneg1;
2011 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2012 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2014 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred);
2016 nfsm_mtofh(dvp, newvp, v3, gotvp);
2018 nfsm_wcc_data(dvp, wccflag);
2021 VTONFS(dvp)->n_flag |= NLMODIFIED;
2023 VTONFS(dvp)->n_attrstamp = 0;
2025 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2026 * if we can succeed in looking up the directory.
2028 if (error == EEXIST || (!error && !gotvp)) {
2031 newvp = (struct vnode *)0;
2033 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2037 if (newvp->v_type != VDIR)
2050 * nfs remove directory call
2052 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2053 * struct componentname *a_cnp)
2056 nfs_rmdir(struct vop_old_rmdir_args *ap)
2058 struct vnode *vp = ap->a_vp;
2059 struct vnode *dvp = ap->a_dvp;
2060 struct componentname *cnp = ap->a_cnp;
2064 caddr_t bpos, dpos, cp2;
2065 int error = 0, wccflag = NFSV3_WCCRATTR;
2066 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2067 int v3 = NFS_ISV3(dvp);
2071 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2072 nfsm_reqhead(dvp, NFSPROC_RMDIR,
2073 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2074 nfsm_fhtom(dvp, v3);
2075 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2076 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred);
2078 nfsm_wcc_data(dvp, wccflag);
2081 VTONFS(dvp)->n_flag |= NLMODIFIED;
2083 VTONFS(dvp)->n_attrstamp = 0;
2085 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2087 if (error == ENOENT)
2095 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2098 nfs_readdir(struct vop_readdir_args *ap)
2100 struct vnode *vp = ap->a_vp;
2101 struct nfsnode *np = VTONFS(vp);
2102 struct uio *uio = ap->a_uio;
2106 if (vp->v_type != VDIR)
2110 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2111 * and then check that is still valid, or if this is an NQNFS mount
2112 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2113 * VOP_GETATTR() does not necessarily go to the wire.
2115 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2116 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2117 if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 &&
2118 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2120 nfsstats.direofcache_hits++;
2126 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2127 * own cache coherency checks so we do not have to.
2129 tresid = uio->uio_resid;
2130 error = nfs_bioread(vp, uio, 0);
2132 if (!error && uio->uio_resid == tresid)
2133 nfsstats.direofcache_misses++;
2138 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2140 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2141 * offset/block and converts the nfs formatted directory entries for userland
2142 * consumption as well as deals with offsets into the middle of blocks.
2143 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2144 * be block-bounded. It must convert to cookies for the actual RPC.
2147 nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2150 struct nfs_dirent *dp = NULL;
2155 caddr_t bpos, dpos, cp2;
2156 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2158 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2159 struct nfsnode *dnp = VTONFS(vp);
2161 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2163 int v3 = NFS_ISV3(vp);
2166 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2167 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2168 panic("nfs readdirrpc bad uio");
2172 * If there is no cookie, assume directory was stale.
2174 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2178 return (NFSERR_BAD_COOKIE);
2180 * Loop around doing readdir rpc's of size nm_readdirsize
2181 * truncated to a multiple of DIRBLKSIZ.
2182 * The stopping criteria is EOF or buffer full.
2184 while (more_dirs && bigenough) {
2185 nfsstats.rpccnt[NFSPROC_READDIR]++;
2186 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2190 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2191 *tl++ = cookie.nfsuquad[0];
2192 *tl++ = cookie.nfsuquad[1];
2193 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2194 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2196 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2197 *tl++ = cookie.nfsuquad[0];
2199 *tl = txdr_unsigned(nmp->nm_readdirsize);
2200 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2202 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2204 nfsm_dissect(tl, u_int32_t *,
2206 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2207 dnp->n_cookieverf.nfsuquad[1] = *tl;
2213 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2214 more_dirs = fxdr_unsigned(int, *tl);
2216 /* loop thru the dir entries, converting them to std form */
2217 while (more_dirs && bigenough) {
2219 nfsm_dissect(tl, u_int32_t *,
2221 fileno = fxdr_hyper(tl);
2222 len = fxdr_unsigned(int, *(tl + 2));
2224 nfsm_dissect(tl, u_int32_t *,
2226 fileno = fxdr_unsigned(u_quad_t, *tl++);
2227 len = fxdr_unsigned(int, *tl);
2229 if (len <= 0 || len > NFS_MAXNAMLEN) {
2236 * len is the number of bytes in the path element
2237 * name, not including the \0 termination.
2239 * tlen is the number of bytes w have to reserve for
2240 * the path element name.
2242 tlen = nfsm_rndup(len);
2244 tlen += 4; /* To ensure null termination */
2247 * If the entry would cross a DIRBLKSIZ boundary,
2248 * extend the previous nfs_dirent to cover the
2251 left = DIRBLKSIZ - blksiz;
2252 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2253 dp->nfs_reclen += left;
2254 uiop->uio_iov->iov_base += left;
2255 uiop->uio_iov->iov_len -= left;
2256 uiop->uio_offset += left;
2257 uiop->uio_resid -= left;
2260 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2263 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2264 dp->nfs_ino = fileno;
2265 dp->nfs_namlen = len;
2266 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2267 dp->nfs_type = DT_UNKNOWN;
2268 blksiz += dp->nfs_reclen;
2269 if (blksiz == DIRBLKSIZ)
2271 uiop->uio_offset += sizeof(struct nfs_dirent);
2272 uiop->uio_resid -= sizeof(struct nfs_dirent);
2273 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent);
2274 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2275 nfsm_mtouio(uiop, len);
2278 * The uiop has advanced by nfs_dirent + len
2279 * but really needs to advance by
2282 cp = uiop->uio_iov->iov_base;
2284 *cp = '\0'; /* null terminate */
2285 uiop->uio_iov->iov_base += tlen;
2286 uiop->uio_iov->iov_len -= tlen;
2287 uiop->uio_offset += tlen;
2288 uiop->uio_resid -= tlen;
2291 * NFS strings must be rounded up (nfsm_myouio
2292 * handled that in the bigenough case).
2294 nfsm_adv(nfsm_rndup(len));
2297 nfsm_dissect(tl, u_int32_t *,
2300 nfsm_dissect(tl, u_int32_t *,
2305 * If we were able to accomodate the last entry,
2306 * get the cookie for the next one. Otherwise
2307 * hold-over the cookie for the one we were not
2308 * able to accomodate.
2311 cookie.nfsuquad[0] = *tl++;
2313 cookie.nfsuquad[1] = *tl++;
2319 more_dirs = fxdr_unsigned(int, *tl);
2322 * If at end of rpc data, get the eof boolean
2325 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2326 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2331 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2332 * by increasing d_reclen for the last record.
2335 left = DIRBLKSIZ - blksiz;
2336 dp->nfs_reclen += left;
2337 uiop->uio_iov->iov_base += left;
2338 uiop->uio_iov->iov_len -= left;
2339 uiop->uio_offset += left;
2340 uiop->uio_resid -= left;
2345 * We hit the end of the directory, update direofoffset.
2347 dnp->n_direofoffset = uiop->uio_offset;
2350 * There is more to go, insert the link cookie so the
2351 * next block can be read.
2353 if (uiop->uio_resid > 0)
2354 printf("EEK! readdirrpc resid > 0\n");
2355 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2363 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2366 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2369 struct nfs_dirent *dp;
2373 struct vnode *newvp;
2375 caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2376 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2378 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2379 struct nfsnode *dnp = VTONFS(vp), *np;
2382 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2383 int attrflag, fhsize;
2384 struct namecache *ncp;
2385 struct namecache *dncp;
2386 struct nlcomponent nlc;
2392 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2393 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2394 panic("nfs readdirplusrpc bad uio");
2397 * Obtain the namecache record for the directory so we have something
2398 * to use as a basis for creating the entries. This function will
2399 * return a held (but not locked) ncp. The ncp may be disconnected
2400 * from the tree and cannot be used for upward traversals, and the
2401 * ncp may be unnamed. Note that other unrelated operations may
2402 * cause the ncp to be named at any time.
2404 dncp = cache_fromdvp(vp, NULL, 0);
2405 bzero(&nlc, sizeof(nlc));
2409 * If there is no cookie, assume directory was stale.
2411 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2415 return (NFSERR_BAD_COOKIE);
2417 * Loop around doing readdir rpc's of size nm_readdirsize
2418 * truncated to a multiple of DIRBLKSIZ.
2419 * The stopping criteria is EOF or buffer full.
2421 while (more_dirs && bigenough) {
2422 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2423 nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2424 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2426 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2427 *tl++ = cookie.nfsuquad[0];
2428 *tl++ = cookie.nfsuquad[1];
2429 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2430 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2431 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2432 *tl = txdr_unsigned(nmp->nm_rsize);
2433 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2434 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2439 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2440 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2441 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2442 more_dirs = fxdr_unsigned(int, *tl);
2444 /* loop thru the dir entries, doctoring them to 4bsd form */
2445 while (more_dirs && bigenough) {
2446 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2447 fileno = fxdr_hyper(tl);
2448 len = fxdr_unsigned(int, *(tl + 2));
2449 if (len <= 0 || len > NFS_MAXNAMLEN) {
2454 tlen = nfsm_rndup(len);
2456 tlen += 4; /* To ensure null termination*/
2457 left = DIRBLKSIZ - blksiz;
2458 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2459 dp->nfs_reclen += left;
2460 uiop->uio_iov->iov_base += left;
2461 uiop->uio_iov->iov_len -= left;
2462 uiop->uio_offset += left;
2463 uiop->uio_resid -= left;
2466 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2469 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2470 dp->nfs_ino = fileno;
2471 dp->nfs_namlen = len;
2472 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2473 dp->nfs_type = DT_UNKNOWN;
2474 blksiz += dp->nfs_reclen;
2475 if (blksiz == DIRBLKSIZ)
2477 uiop->uio_offset += sizeof(struct nfs_dirent);
2478 uiop->uio_resid -= sizeof(struct nfs_dirent);
2479 uiop->uio_iov->iov_base += sizeof(struct nfs_dirent);
2480 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2481 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2482 nlc.nlc_namelen = len;
2483 nfsm_mtouio(uiop, len);
2484 cp = uiop->uio_iov->iov_base;
2487 uiop->uio_iov->iov_base += tlen;
2488 uiop->uio_iov->iov_len -= tlen;
2489 uiop->uio_offset += tlen;
2490 uiop->uio_resid -= tlen;
2492 nfsm_adv(nfsm_rndup(len));
2493 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2495 cookie.nfsuquad[0] = *tl++;
2496 cookie.nfsuquad[1] = *tl++;
2501 * Since the attributes are before the file handle
2502 * (sigh), we must skip over the attributes and then
2503 * come back and get them.
2505 attrflag = fxdr_unsigned(int, *tl);
2509 nfsm_adv(NFSX_V3FATTR);
2510 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2511 doit = fxdr_unsigned(int, *tl);
2513 nfsm_getfh(fhp, fhsize, 1);
2514 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2519 error = nfs_nget(vp->v_mount, fhp,
2527 if (doit && bigenough) {
2532 nfsm_loadattr(newvp, (struct vattr *)0);
2536 IFTODT(VTTOIF(np->n_vattr.va_type));
2538 printf("NFS/READDIRPLUS, ENTER %*.*s\n",
2539 nlc.nlc_namelen, nlc.nlc_namelen,
2541 ncp = cache_nlookup(dncp, &nlc);
2542 cache_setunresolved(ncp);
2543 cache_setvp(ncp, newvp);
2546 printf("NFS/READDIRPLUS, UNABLE TO ENTER"
2548 nlc.nlc_namelen, nlc.nlc_namelen,
2553 /* Just skip over the file handle */
2554 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2555 i = fxdr_unsigned(int, *tl);
2556 nfsm_adv(nfsm_rndup(i));
2558 if (newvp != NULLVP) {
2565 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2566 more_dirs = fxdr_unsigned(int, *tl);
2569 * If at end of rpc data, get the eof boolean
2572 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2573 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2578 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2579 * by increasing d_reclen for the last record.
2582 left = DIRBLKSIZ - blksiz;
2583 dp->nfs_reclen += left;
2584 uiop->uio_iov->iov_base += left;
2585 uiop->uio_iov->iov_len -= left;
2586 uiop->uio_offset += left;
2587 uiop->uio_resid -= left;
2591 * We are now either at the end of the directory or have filled the
2595 dnp->n_direofoffset = uiop->uio_offset;
2597 if (uiop->uio_resid > 0)
2598 printf("EEK! readdirplusrpc resid > 0\n");
2599 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2603 if (newvp != NULLVP) {
2616 * Silly rename. To make the NFS filesystem that is stateless look a little
2617 * more like the "ufs" a remove of an active vnode is translated to a rename
2618 * to a funny looking filename that is removed by nfs_inactive on the
2619 * nfsnode. There is the potential for another process on a different client
2620 * to create the same funny name between the nfs_lookitup() fails and the
2621 * nfs_rename() completes, but...
2624 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2626 struct sillyrename *sp;
2631 * We previously purged dvp instead of vp. I don't know why, it
2632 * completely destroys performance. We can't do it anyway with the
2633 * new VFS API since we would be breaking the namecache topology.
2635 cache_purge(vp); /* XXX */
2638 if (vp->v_type == VDIR)
2639 panic("nfs: sillyrename dir");
2641 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2642 M_NFSREQ, M_WAITOK);
2643 sp->s_cred = crdup(cnp->cn_cred);
2647 /* Fudge together a funny name */
2648 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td);
2650 /* Try lookitups until we get one that isn't there */
2651 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2652 cnp->cn_td, (struct nfsnode **)0) == 0) {
2654 if (sp->s_name[4] > 'z') {
2659 error = nfs_renameit(dvp, cnp, sp);
2662 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2664 np->n_sillyrename = sp;
2669 free((caddr_t)sp, M_NFSREQ);
2674 * Look up a file name and optionally either update the file handle or
2675 * allocate an nfsnode, depending on the value of npp.
2676 * npp == NULL --> just do the lookup
2677 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2679 * *npp != NULL --> update the file handle in the vnode
2682 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2683 struct thread *td, struct nfsnode **npp)
2688 struct vnode *newvp = (struct vnode *)0;
2689 struct nfsnode *np, *dnp = VTONFS(dvp);
2690 caddr_t bpos, dpos, cp2;
2691 int error = 0, fhlen, attrflag;
2692 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2694 int v3 = NFS_ISV3(dvp);
2696 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2697 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2698 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2699 nfsm_fhtom(dvp, v3);
2700 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2701 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2702 if (npp && !error) {
2703 nfsm_getfh(nfhp, fhlen, v3);
2706 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2707 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2708 np->n_fhp = &np->n_fh;
2709 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2710 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2711 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2712 np->n_fhsize = fhlen;
2714 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2718 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2726 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
2727 if (!attrflag && *npp == NULL) {
2736 nfsm_loadattr(newvp, (struct vattr *)0);
2740 if (npp && *npp == NULL) {
2755 * Nfs Version 3 commit rpc
2758 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2763 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2764 caddr_t bpos, dpos, cp2;
2765 int error = 0, wccflag = NFSV3_WCCRATTR;
2766 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2768 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2770 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2771 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2773 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2774 txdr_hyper(offset, tl);
2776 *tl = txdr_unsigned(cnt);
2777 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE));
2778 nfsm_wcc_data(vp, wccflag);
2780 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2781 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2782 NFSX_V3WRITEVERF)) {
2783 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2785 error = NFSERR_STALEWRITEVERF;
2795 * - make nfs_bmap() essentially a no-op that does no translation
2796 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2797 * (Maybe I could use the process's page mapping, but I was concerned that
2798 * Kernel Write might not be enabled and also figured copyout() would do
2799 * a lot more work than bcopy() and also it currently happens in the
2800 * context of the swapper process (2).
2802 * nfs_bmap(struct vnode *a_vp, off_t a_loffset, struct vnode **a_vpp,
2803 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2806 nfs_bmap(struct vop_bmap_args *ap)
2808 struct vnode *vp = ap->a_vp;
2810 if (ap->a_vpp != NULL)
2812 if (ap->a_doffsetp != NULL)
2813 *ap->a_doffsetp = ap->a_loffset;
2814 if (ap->a_runp != NULL)
2816 if (ap->a_runb != NULL)
2824 * For async requests when nfsiod(s) are running, queue the request by
2825 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2829 nfs_strategy(struct vop_strategy_args *ap)
2831 struct bio *bio = ap->a_bio;
2833 struct buf *bp = bio->bio_buf;
2837 KASSERT(!(bp->b_flags & B_DONE),
2838 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2839 KASSERT(BUF_REFCNT(bp) > 0,
2840 ("nfs_strategy: buffer %p not locked", bp));
2842 if (bp->b_flags & B_ASYNC)
2845 td = curthread; /* XXX */
2848 * We probably don't need to push an nbio any more since no
2849 * block conversion is required due to the use of 64 bit byte
2850 * offsets, but do it anyway.
2852 nbio = push_bio(bio);
2853 nbio->bio_offset = bio->bio_offset;
2856 * If the op is asynchronous and an i/o daemon is waiting
2857 * queue the request, wake it up and wait for completion
2858 * otherwise just do it ourselves.
2860 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(ap->a_vp, nbio, td))
2861 error = nfs_doio(ap->a_vp, nbio, td);
2868 * NB Currently unsupported.
2870 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred,
2871 * struct thread *a_td)
2875 nfs_mmap(struct vop_mmap_args *ap)
2881 * fsync vnode op. Just call nfs_flush() with commit == 1.
2883 * nfs_fsync(struct vnodeop_desc *a_desc, struct vnode *a_vp,
2884 * struct ucred * a_cred, int a_waitfor, struct thread *a_td)
2888 nfs_fsync(struct vop_fsync_args *ap)
2890 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2894 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2895 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2896 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2897 * set the buffer contains data that has already been written to the server
2898 * and which now needs a commit RPC.
2900 * If commit is 0 we only take one pass and only flush buffers containing new
2903 * If commit is 1 we take two passes, issuing a commit RPC in the second
2906 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
2907 * to completely flush all pending data.
2909 * Note that the RB_SCAN code properly handles the case where the
2910 * callback might block and directly or indirectly (another thread) cause
2911 * the RB tree to change.
2914 #ifndef NFS_COMMITBVECSIZ
2915 #define NFS_COMMITBVECSIZ 16
2918 struct nfs_flush_info {
2919 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
2926 struct buf *bvary[NFS_COMMITBVECSIZ];
2932 static int nfs_flush_bp(struct buf *bp, void *data);
2933 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
2936 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
2938 struct nfsnode *np = VTONFS(vp);
2939 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2940 struct nfs_flush_info info;
2943 bzero(&info, sizeof(info));
2946 info.waitfor = waitfor;
2947 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
2954 info.mode = NFI_FLUSHNEW;
2955 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2956 nfs_flush_bp, &info);
2959 * Take a second pass if committing and no error occured.
2960 * Clean up any left over collection (whether an error
2963 if (commit && error == 0) {
2964 info.mode = NFI_COMMIT;
2965 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2966 nfs_flush_bp, &info);
2968 error = nfs_flush_docommit(&info, error);
2972 * Wait for pending I/O to complete before checking whether
2973 * any further dirty buffers exist.
2975 while (waitfor == MNT_WAIT && vp->v_track_write.bk_active) {
2976 vp->v_track_write.bk_waitflag = 1;
2977 error = tsleep(&vp->v_track_write,
2978 info.slpflag, "nfsfsync", info.slptimeo);
2981 * We have to be able to break out if this
2982 * is an 'intr' mount.
2984 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
2990 * Since we do not process pending signals,
2991 * once we get a PCATCH our tsleep() will no
2992 * longer sleep, switch to a fixed timeout
2995 if (info.slpflag == PCATCH) {
2997 info.slptimeo = 2 * hz;
3004 * Loop if we are flushing synchronous as well as committing,
3005 * and dirty buffers are still present. Otherwise we might livelock.
3007 } while (waitfor == MNT_WAIT && commit &&
3008 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3011 * The callbacks have to return a negative error to terminate the
3018 * Deal with any error collection
3020 if (np->n_flag & NWRITEERR) {
3021 error = np->n_error;
3022 np->n_flag &= ~NWRITEERR;
3030 nfs_flush_bp(struct buf *bp, void *data)
3032 struct nfs_flush_info *info = data;
3037 switch(info->mode) {
3040 if (info->loops && info->waitfor == MNT_WAIT) {
3041 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3043 int lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3044 if (info->slpflag & PCATCH)
3045 lkflags |= LK_PCATCH;
3046 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3050 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3053 KKASSERT(bp->b_vp == info->vp);
3055 if ((bp->b_flags & B_DELWRI) == 0)
3056 panic("nfs_fsync: not dirty");
3057 if (bp->b_flags & B_NEEDCOMMIT) {
3064 bp->b_flags |= B_ASYNC;
3066 VOP_BWRITE(bp->b_vp, bp);
3074 * Only process buffers in need of a commit which we can
3075 * immediately lock. This may prevent a buffer from being
3076 * committed, but the normal flush loop will block on the
3077 * same buffer so we shouldn't get into an endless loop.
3080 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3081 (B_DELWRI | B_NEEDCOMMIT) ||
3082 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
3087 KKASSERT(bp->b_vp == info->vp);
3091 * NOTE: we are not clearing B_DONE here, so we have
3092 * to do it later on in this routine if we intend to
3093 * initiate I/O on the bp.
3095 * Note: to avoid loopback deadlocks, we do not
3096 * assign b_runningbufspace.
3098 vfs_busy_pages(bp->b_vp, bp, 1);
3100 info->bvary[info->bvsize] = bp;
3101 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3102 if (info->bvsize == 0 || toff < info->beg_off)
3103 info->beg_off = toff;
3104 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3105 if (info->bvsize == 0 || toff > info->end_off)
3106 info->end_off = toff;
3108 if (info->bvsize == NFS_COMMITBVECSIZ) {
3109 error = nfs_flush_docommit(info, 0);
3110 KKASSERT(info->bvsize == 0);
3119 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3129 if (info->bvsize > 0) {
3131 * Commit data on the server, as required. Note that
3132 * nfs_commit will use the vnode's cred for the commit.
3133 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3135 bytes = info->end_off - info->beg_off;
3136 if (bytes > 0x40000000)
3141 retv = nfs_commit(vp, info->beg_off,
3142 (int)bytes, info->td);
3143 if (retv == NFSERR_STALEWRITEVERF)
3144 nfs_clearcommit(vp->v_mount);
3148 * Now, either mark the blocks I/O done or mark the
3149 * blocks dirty, depending on whether the commit
3152 for (i = 0; i < info->bvsize; ++i) {
3153 bp = info->bvary[i];
3154 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3157 * Error, leave B_DELWRI intact
3159 vfs_unbusy_pages(bp);
3163 * Success, remove B_DELWRI ( bundirty() ).
3165 * b_dirtyoff/b_dirtyend seem to be NFS
3166 * specific. We should probably move that
3167 * into bundirty(). XXX
3169 * We are faking an I/O write, we have to
3170 * start the transaction in order to
3171 * immediately biodone() it.
3174 bp->b_flags |= B_ASYNC;
3176 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3177 bp->b_dirtyoff = bp->b_dirtyend = 0;
3179 biodone(&bp->b_bio1);
3188 * NFS advisory byte-level locks.
3189 * Currently unsupported.
3191 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3195 nfs_advlock(struct vop_advlock_args *ap)
3197 struct nfsnode *np = VTONFS(ap->a_vp);
3200 * The following kludge is to allow diskless support to work
3201 * until a real NFS lockd is implemented. Basically, just pretend
3202 * that this is a local lock.
3204 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3208 * Print out the contents of an nfsnode.
3210 * nfs_print(struct vnode *a_vp)
3213 nfs_print(struct vop_print_args *ap)
3215 struct vnode *vp = ap->a_vp;
3216 struct nfsnode *np = VTONFS(vp);
3218 printf("tag VT_NFS, fileid %ld fsid 0x%x",
3219 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3220 if (vp->v_type == VFIFO)
3227 * Just call nfs_writebp() with the force argument set to 1.
3229 * NOTE: B_DONE may or may not be set in a_bp on call.
3231 * nfs_bwrite(struct vnode *a_bp)
3234 nfs_bwrite(struct vop_bwrite_args *ap)
3236 return (nfs_writebp(ap->a_bp, 1, curthread));
3240 * This is a clone of vn_bwrite(), except that it also handles the
3241 * B_NEEDCOMMIT flag. We set B_CACHE if this is a VMIO buffer.
3244 nfs_writebp(struct buf *bp, int force, struct thread *td)
3248 if (BUF_REFCNT(bp) == 0)
3249 panic("bwrite: buffer is not locked???");
3251 if (bp->b_flags & B_INVAL) {
3256 bp->b_flags |= B_CACHE;
3259 * Undirty the bp. We will redirty it later if the I/O fails.
3263 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3267 * Note: to avoid loopback deadlocks, we do not
3268 * assign b_runningbufspace.
3270 vfs_busy_pages(bp->b_vp, bp, 1);
3273 if (bp->b_flags & B_ASYNC) {
3274 vn_strategy(bp->b_vp, &bp->b_bio1);
3277 vn_strategy(bp->b_vp, &bp->b_bio1);
3278 error = biowait(bp);
3285 * nfs special file access vnode op.
3286 * Essentially just get vattr and then imitate iaccess() since the device is
3287 * local to the client.
3289 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
3290 * struct thread *a_td)
3293 nfsspec_access(struct vop_access_args *ap)
3297 struct ucred *cred = ap->a_cred;
3298 struct vnode *vp = ap->a_vp;
3299 mode_t mode = ap->a_mode;
3305 * Disallow write attempts on filesystems mounted read-only;
3306 * unless the file is a socket, fifo, or a block or character
3307 * device resident on the filesystem.
3309 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3310 switch (vp->v_type) {
3320 * If you're the super-user,
3321 * you always get access.
3323 if (cred->cr_uid == 0)
3326 error = VOP_GETATTR(vp, vap, ap->a_td);
3330 * Access check is based on only one of owner, group, public.
3331 * If not owner, then check group. If not a member of the
3332 * group, then check public access.
3334 if (cred->cr_uid != vap->va_uid) {
3336 gp = cred->cr_groups;
3337 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3338 if (vap->va_gid == *gp)
3344 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3349 * Read wrapper for special devices.
3351 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3352 * struct ucred *a_cred)
3355 nfsspec_read(struct vop_read_args *ap)
3357 struct nfsnode *np = VTONFS(ap->a_vp);
3363 getnanotime(&np->n_atim);
3364 return (VOCALL(spec_vnode_vops, &ap->a_head));
3368 * Write wrapper for special devices.
3370 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3371 * struct ucred *a_cred)
3374 nfsspec_write(struct vop_write_args *ap)
3376 struct nfsnode *np = VTONFS(ap->a_vp);
3382 getnanotime(&np->n_mtim);
3383 return (VOCALL(spec_vnode_vops, &ap->a_head));
3387 * Close wrapper for special devices.
3389 * Update the times on the nfsnode then do device close.
3391 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
3392 * struct thread *a_td)
3395 nfsspec_close(struct vop_close_args *ap)
3397 struct vnode *vp = ap->a_vp;
3398 struct nfsnode *np = VTONFS(vp);
3401 if (np->n_flag & (NACC | NUPD)) {
3403 if (vp->v_usecount == 1 &&
3404 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3406 if (np->n_flag & NACC)
3407 vattr.va_atime = np->n_atim;
3408 if (np->n_flag & NUPD)
3409 vattr.va_mtime = np->n_mtim;
3410 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
3413 return (VOCALL(spec_vnode_vops, &ap->a_head));
3417 * Read wrapper for fifos.
3419 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3420 * struct ucred *a_cred)
3423 nfsfifo_read(struct vop_read_args *ap)
3425 struct nfsnode *np = VTONFS(ap->a_vp);
3431 getnanotime(&np->n_atim);
3432 return (VOCALL(fifo_vnode_vops, &ap->a_head));
3436 * Write wrapper for fifos.
3438 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3439 * struct ucred *a_cred)
3442 nfsfifo_write(struct vop_write_args *ap)
3444 struct nfsnode *np = VTONFS(ap->a_vp);
3450 getnanotime(&np->n_mtim);
3451 return (VOCALL(fifo_vnode_vops, &ap->a_head));
3455 * Close wrapper for fifos.
3457 * Update the times on the nfsnode then do fifo close.
3459 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td)
3462 nfsfifo_close(struct vop_close_args *ap)
3464 struct vnode *vp = ap->a_vp;
3465 struct nfsnode *np = VTONFS(vp);
3469 if (np->n_flag & (NACC | NUPD)) {
3471 if (np->n_flag & NACC)
3473 if (np->n_flag & NUPD)
3476 if (vp->v_usecount == 1 &&
3477 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3479 if (np->n_flag & NACC)
3480 vattr.va_atime = np->n_atim;
3481 if (np->n_flag & NUPD)
3482 vattr.va_mtime = np->n_mtim;
3483 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
3486 return (VOCALL(fifo_vnode_vops, &ap->a_head));