namecache work stage 4a: Do some minor performance cleanups with negative
[dragonfly.git] / sys / vfs / nfs / nfs_vnops.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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.
23 *
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
34 * SUCH DAMAGE.
35 *
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 $
41a01a4d 38 * $DragonFly: src/sys/vfs/nfs/nfs_vnops.c,v 1.19 2004/03/01 06:33:21 dillon Exp $
984263bc
MD
39 */
40
41
42/*
43 * vnode op calls for Sun NFS version 2 and 3
44 */
45
46#include "opt_inet.h"
47
48#include <sys/param.h>
49#include <sys/kernel.h>
50#include <sys/systm.h>
51#include <sys/resourcevar.h>
52#include <sys/proc.h>
53#include <sys/mount.h>
54#include <sys/buf.h>
55#include <sys/malloc.h>
56#include <sys/mbuf.h>
57#include <sys/namei.h>
58#include <sys/socket.h>
59#include <sys/vnode.h>
60#include <sys/dirent.h>
61#include <sys/fcntl.h>
62#include <sys/lockf.h>
63#include <sys/stat.h>
64#include <sys/sysctl.h>
65#include <sys/conf.h>
66
67#include <vm/vm.h>
68#include <vm/vm_extern.h>
69#include <vm/vm_zone.h>
70
3020e3be
MD
71#include <sys/buf2.h>
72
1f2de5d4 73#include <vfs/fifofs/fifo.h>
984263bc 74
1f2de5d4
MD
75#include "rpcv2.h"
76#include "nfsproto.h"
77#include "nfs.h"
1f2de5d4 78#include "nfsmount.h"
c1cf1e59 79#include "nfsnode.h"
1f2de5d4
MD
80#include "xdr_subs.h"
81#include "nfsm_subs.h"
82#include "nqnfs.h"
984263bc
MD
83
84#include <net/if.h>
85#include <netinet/in.h>
86#include <netinet/in_var.h>
87
88/* Defs */
89#define TRUE 1
90#define FALSE 0
91
92/*
93 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
94 * calls are not in getblk() and brelse() so that they would not be necessary
95 * here.
96 */
97#ifndef B_VMIO
98#define vfs_busy_pages(bp, f)
99#endif
100
a6ee311a
RG
101static int nfsspec_read (struct vop_read_args *);
102static int nfsspec_write (struct vop_write_args *);
103static int nfsfifo_read (struct vop_read_args *);
104static int nfsfifo_write (struct vop_write_args *);
105static int nfsspec_close (struct vop_close_args *);
106static int nfsfifo_close (struct vop_close_args *);
984263bc 107#define nfs_poll vop_nopoll
a6ee311a
RG
108static int nfs_flush (struct vnode *,int,struct thread *,int);
109static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
110static int nfs_lookup (struct vop_lookup_args *);
111static int nfs_create (struct vop_create_args *);
112static int nfs_mknod (struct vop_mknod_args *);
113static int nfs_open (struct vop_open_args *);
114static int nfs_close (struct vop_close_args *);
115static int nfs_access (struct vop_access_args *);
116static int nfs_getattr (struct vop_getattr_args *);
117static int nfs_setattr (struct vop_setattr_args *);
118static int nfs_read (struct vop_read_args *);
119static int nfs_mmap (struct vop_mmap_args *);
120static int nfs_fsync (struct vop_fsync_args *);
121static int nfs_remove (struct vop_remove_args *);
122static int nfs_link (struct vop_link_args *);
123static int nfs_rename (struct vop_rename_args *);
124static int nfs_mkdir (struct vop_mkdir_args *);
125static int nfs_rmdir (struct vop_rmdir_args *);
126static int nfs_symlink (struct vop_symlink_args *);
127static int nfs_readdir (struct vop_readdir_args *);
128static int nfs_bmap (struct vop_bmap_args *);
129static int nfs_strategy (struct vop_strategy_args *);
130static int nfs_lookitup (struct vnode *, const char *, int,
131 struct ucred *, struct thread *, struct nfsnode **);
132static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
133static int nfsspec_access (struct vop_access_args *);
134static int nfs_readlink (struct vop_readlink_args *);
135static int nfs_print (struct vop_print_args *);
136static int nfs_advlock (struct vop_advlock_args *);
137static int nfs_bwrite (struct vop_bwrite_args *);
984263bc
MD
138/*
139 * Global vfs data structures for nfs
140 */
141vop_t **nfsv2_vnodeop_p;
142static struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
143 { &vop_default_desc, (vop_t *) vop_defaultop },
144 { &vop_access_desc, (vop_t *) nfs_access },
145 { &vop_advlock_desc, (vop_t *) nfs_advlock },
146 { &vop_bmap_desc, (vop_t *) nfs_bmap },
147 { &vop_bwrite_desc, (vop_t *) nfs_bwrite },
148 { &vop_close_desc, (vop_t *) nfs_close },
149 { &vop_create_desc, (vop_t *) nfs_create },
150 { &vop_fsync_desc, (vop_t *) nfs_fsync },
151 { &vop_getattr_desc, (vop_t *) nfs_getattr },
152 { &vop_getpages_desc, (vop_t *) nfs_getpages },
153 { &vop_putpages_desc, (vop_t *) nfs_putpages },
154 { &vop_inactive_desc, (vop_t *) nfs_inactive },
155 { &vop_islocked_desc, (vop_t *) vop_stdislocked },
156 { &vop_lease_desc, (vop_t *) vop_null },
157 { &vop_link_desc, (vop_t *) nfs_link },
158 { &vop_lock_desc, (vop_t *) vop_sharedlock },
159 { &vop_lookup_desc, (vop_t *) nfs_lookup },
160 { &vop_mkdir_desc, (vop_t *) nfs_mkdir },
161 { &vop_mknod_desc, (vop_t *) nfs_mknod },
162 { &vop_mmap_desc, (vop_t *) nfs_mmap },
163 { &vop_open_desc, (vop_t *) nfs_open },
164 { &vop_poll_desc, (vop_t *) nfs_poll },
165 { &vop_print_desc, (vop_t *) nfs_print },
166 { &vop_read_desc, (vop_t *) nfs_read },
167 { &vop_readdir_desc, (vop_t *) nfs_readdir },
168 { &vop_readlink_desc, (vop_t *) nfs_readlink },
169 { &vop_reclaim_desc, (vop_t *) nfs_reclaim },
170 { &vop_remove_desc, (vop_t *) nfs_remove },
171 { &vop_rename_desc, (vop_t *) nfs_rename },
172 { &vop_rmdir_desc, (vop_t *) nfs_rmdir },
173 { &vop_setattr_desc, (vop_t *) nfs_setattr },
174 { &vop_strategy_desc, (vop_t *) nfs_strategy },
175 { &vop_symlink_desc, (vop_t *) nfs_symlink },
176 { &vop_unlock_desc, (vop_t *) vop_stdunlock },
177 { &vop_write_desc, (vop_t *) nfs_write },
178 { NULL, NULL }
179};
180static struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
181 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
182VNODEOP_SET(nfsv2_vnodeop_opv_desc);
183
184/*
185 * Special device vnode ops
186 */
187vop_t **spec_nfsv2nodeop_p;
188static struct vnodeopv_entry_desc nfsv2_specop_entries[] = {
189 { &vop_default_desc, (vop_t *) spec_vnoperate },
190 { &vop_access_desc, (vop_t *) nfsspec_access },
191 { &vop_close_desc, (vop_t *) nfsspec_close },
192 { &vop_fsync_desc, (vop_t *) nfs_fsync },
193 { &vop_getattr_desc, (vop_t *) nfs_getattr },
194 { &vop_inactive_desc, (vop_t *) nfs_inactive },
195 { &vop_islocked_desc, (vop_t *) vop_stdislocked },
196 { &vop_lock_desc, (vop_t *) vop_sharedlock },
197 { &vop_print_desc, (vop_t *) nfs_print },
198 { &vop_read_desc, (vop_t *) nfsspec_read },
199 { &vop_reclaim_desc, (vop_t *) nfs_reclaim },
200 { &vop_setattr_desc, (vop_t *) nfs_setattr },
201 { &vop_unlock_desc, (vop_t *) vop_stdunlock },
202 { &vop_write_desc, (vop_t *) nfsspec_write },
203 { NULL, NULL }
204};
205static struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
206 { &spec_nfsv2nodeop_p, nfsv2_specop_entries };
207VNODEOP_SET(spec_nfsv2nodeop_opv_desc);
208
209vop_t **fifo_nfsv2nodeop_p;
210static struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = {
211 { &vop_default_desc, (vop_t *) fifo_vnoperate },
212 { &vop_access_desc, (vop_t *) nfsspec_access },
213 { &vop_close_desc, (vop_t *) nfsfifo_close },
214 { &vop_fsync_desc, (vop_t *) nfs_fsync },
215 { &vop_getattr_desc, (vop_t *) nfs_getattr },
216 { &vop_inactive_desc, (vop_t *) nfs_inactive },
217 { &vop_islocked_desc, (vop_t *) vop_stdislocked },
218 { &vop_lock_desc, (vop_t *) vop_sharedlock },
219 { &vop_print_desc, (vop_t *) nfs_print },
220 { &vop_read_desc, (vop_t *) nfsfifo_read },
221 { &vop_reclaim_desc, (vop_t *) nfs_reclaim },
222 { &vop_setattr_desc, (vop_t *) nfs_setattr },
223 { &vop_unlock_desc, (vop_t *) vop_stdunlock },
224 { &vop_write_desc, (vop_t *) nfsfifo_write },
225 { NULL, NULL }
226};
227static struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
228 { &fifo_nfsv2nodeop_p, nfsv2_fifoop_entries };
229VNODEOP_SET(fifo_nfsv2nodeop_opv_desc);
230
a6ee311a 231static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
984263bc 232 struct componentname *cnp,
a6ee311a
RG
233 struct vattr *vap);
234static int nfs_removerpc (struct vnode *dvp, const char *name,
984263bc 235 int namelen,
a6ee311a
RG
236 struct ucred *cred, struct thread *td);
237static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
984263bc
MD
238 int fnamelen, struct vnode *tdvp,
239 const char *tnameptr, int tnamelen,
a6ee311a
RG
240 struct ucred *cred, struct thread *td);
241static int nfs_renameit (struct vnode *sdvp,
984263bc 242 struct componentname *scnp,
a6ee311a 243 struct sillyrename *sp);
984263bc
MD
244
245/*
246 * Global variables
247 */
248extern u_int32_t nfs_true, nfs_false;
249extern u_int32_t nfs_xdrneg1;
250extern struct nfsstats nfsstats;
251extern nfstype nfsv3_type[9];
dadab5e9 252struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON];
984263bc
MD
253struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
254int nfs_numasync = 0;
255#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
256
257SYSCTL_DECL(_vfs_nfs);
258
259static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
260SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
261 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
262
263static int nfsv3_commit_on_close = 0;
264SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
265 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
266#if 0
267SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
268 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
269
270SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
271 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
272#endif
273
274#define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
275 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
276 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
277static int
dadab5e9
MD
278nfs3_access_otw(struct vnode *vp, int wmode,
279 struct thread *td, struct ucred *cred)
984263bc
MD
280{
281 const int v3 = 1;
282 u_int32_t *tl;
283 int error = 0, attrflag;
284
285 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
286 caddr_t bpos, dpos, cp2;
40393ded
RG
287 int32_t t1, t2;
288 caddr_t cp;
984263bc
MD
289 u_int32_t rmode;
290 struct nfsnode *np = VTONFS(vp);
291
292 nfsstats.rpccnt[NFSPROC_ACCESS]++;
293 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
294 nfsm_fhtom(vp, v3);
295 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
296 *tl = txdr_unsigned(wmode);
dadab5e9 297 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
984263bc
MD
298 nfsm_postop_attr(vp, attrflag);
299 if (!error) {
300 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
301 rmode = fxdr_unsigned(u_int32_t, *tl);
302 np->n_mode = rmode;
303 np->n_modeuid = cred->cr_uid;
3a6f9faf 304 np->n_modestamp = mycpu->gd_time_seconds;
984263bc 305 }
6b08710e
MD
306 m_freem(mrep);
307nfsmout:
984263bc
MD
308 return error;
309}
310
311/*
312 * nfs access vnode op.
313 * For nfs version 2, just return ok. File accesses may fail later.
314 * For nfs version 3, use the access rpc to check accessibility. If file modes
315 * are changed on the server, accesses might still fail later.
316 */
317static int
318nfs_access(ap)
319 struct vop_access_args /* {
320 struct vnode *a_vp;
321 int a_mode;
322 struct ucred *a_cred;
dadab5e9 323 struct thread *a_td;
984263bc
MD
324 } */ *ap;
325{
40393ded 326 struct vnode *vp = ap->a_vp;
984263bc
MD
327 int error = 0;
328 u_int32_t mode, wmode;
329 int v3 = NFS_ISV3(vp);
330 struct nfsnode *np = VTONFS(vp);
331
332 /*
333 * Disallow write attempts on filesystems mounted read-only;
334 * unless the file is a socket, fifo, or a block or character
335 * device resident on the filesystem.
336 */
337 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
338 switch (vp->v_type) {
339 case VREG:
340 case VDIR:
341 case VLNK:
342 return (EROFS);
343 default:
344 break;
345 }
346 }
347 /*
348 * For nfs v3, check to see if we have done this recently, and if
349 * so return our cached result instead of making an ACCESS call.
350 * If not, do an access rpc, otherwise you are stuck emulating
351 * ufs_access() locally using the vattr. This may not be correct,
352 * since the server may apply other access criteria such as
353 * client uid-->server uid mapping that we do not know about.
354 */
355 if (v3) {
356 if (ap->a_mode & VREAD)
357 mode = NFSV3ACCESS_READ;
358 else
359 mode = 0;
360 if (vp->v_type != VDIR) {
361 if (ap->a_mode & VWRITE)
362 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
363 if (ap->a_mode & VEXEC)
364 mode |= NFSV3ACCESS_EXECUTE;
365 } else {
366 if (ap->a_mode & VWRITE)
367 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
368 NFSV3ACCESS_DELETE);
369 if (ap->a_mode & VEXEC)
370 mode |= NFSV3ACCESS_LOOKUP;
371 }
372 /* XXX safety belt, only make blanket request if caching */
373 if (nfsaccess_cache_timeout > 0) {
374 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
375 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
376 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
377 } else {
378 wmode = mode;
379 }
380
381 /*
382 * Does our cached result allow us to give a definite yes to
383 * this request?
384 */
3a6f9faf 385 if ((mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
984263bc
MD
386 (ap->a_cred->cr_uid == np->n_modeuid) &&
387 ((np->n_mode & mode) == mode)) {
388 nfsstats.accesscache_hits++;
389 } else {
390 /*
391 * Either a no, or a don't know. Go to the wire.
392 */
393 nfsstats.accesscache_misses++;
dadab5e9 394 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
984263bc
MD
395 if (!error) {
396 if ((np->n_mode & mode) != mode) {
397 error = EACCES;
398 }
399 }
400 }
984263bc
MD
401 } else {
402 if ((error = nfsspec_access(ap)) != 0)
403 return (error);
404
405 /*
406 * Attempt to prevent a mapped root from accessing a file
407 * which it shouldn't. We try to read a byte from the file
408 * if the user is root and the file is not zero length.
409 * After calling nfsspec_access, we should have the correct
410 * file size cached.
411 */
412 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
413 && VTONFS(vp)->n_size > 0) {
414 struct iovec aiov;
415 struct uio auio;
416 char buf[1];
417
418 aiov.iov_base = buf;
419 aiov.iov_len = 1;
420 auio.uio_iov = &aiov;
421 auio.uio_iovcnt = 1;
422 auio.uio_offset = 0;
423 auio.uio_resid = 1;
424 auio.uio_segflg = UIO_SYSSPACE;
425 auio.uio_rw = UIO_READ;
dadab5e9 426 auio.uio_td = ap->a_td;
984263bc 427
c1cf1e59 428 if (vp->v_type == VREG) {
3b568787 429 error = nfs_readrpc(vp, &auio);
c1cf1e59 430 } else if (vp->v_type == VDIR) {
984263bc
MD
431 char* bp;
432 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
433 aiov.iov_base = bp;
434 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
3b568787 435 error = nfs_readdirrpc(vp, &auio);
984263bc 436 free(bp, M_TEMP);
c1cf1e59 437 } else if (vp->v_type == VLNK) {
3b568787 438 error = nfs_readlinkrpc(vp, &auio);
c1cf1e59 439 } else {
984263bc 440 error = EACCES;
c1cf1e59 441 }
984263bc 442 }
984263bc 443 }
c1cf1e59
MD
444 /*
445 * [re]record creds for reading and/or writing if access
09b1ee9b
MD
446 * was granted. Assume the NFS server will grant read access
447 * for execute requests.
c1cf1e59
MD
448 */
449 if (error == 0) {
09b1ee9b 450 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) {
c1cf1e59
MD
451 crhold(ap->a_cred);
452 if (np->n_rucred)
453 crfree(np->n_rucred);
454 np->n_rucred = ap->a_cred;
455 }
456 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
457 crhold(ap->a_cred);
458 if (np->n_wucred)
459 crfree(np->n_wucred);
460 np->n_wucred = ap->a_cred;
461 }
462 }
463 return(error);
984263bc
MD
464}
465
466/*
467 * nfs open vnode op
468 * Check to see if the type is ok
469 * and that deletion is not in progress.
470 * For paged in text files, you will need to flush the page cache
471 * if consistency is lost.
472 */
473/* ARGSUSED */
474static int
475nfs_open(ap)
476 struct vop_open_args /* {
477 struct vnode *a_vp;
478 int a_mode;
479 struct ucred *a_cred;
dadab5e9 480 struct thread *a_td;
984263bc
MD
481 } */ *ap;
482{
40393ded 483 struct vnode *vp = ap->a_vp;
984263bc
MD
484 struct nfsnode *np = VTONFS(vp);
485 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
486 struct vattr vattr;
487 int error;
488
489 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
490#ifdef DIAGNOSTIC
491 printf("open eacces vtyp=%d\n",vp->v_type);
492#endif
ca3a2b2f 493 return (EOPNOTSUPP);
984263bc
MD
494 }
495 /*
496 * Get a valid lease. If cached data is stale, flush it.
497 */
498 if (nmp->nm_flag & NFSMNT_NQNFS) {
499 if (NQNFS_CKINVALID(vp, np, ND_READ)) {
500 do {
3b568787 501 error = nqnfs_getlease(vp, ND_READ, ap->a_td);
984263bc
MD
502 } while (error == NQNFS_EXPIRED);
503 if (error)
504 return (error);
505 if (np->n_lrev != np->n_brev ||
506 (np->n_flag & NQNFSNONCACHE)) {
3b568787
MD
507 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1))
508 == EINTR) {
984263bc 509 return (error);
3b568787 510 }
984263bc
MD
511 np->n_brev = np->n_lrev;
512 }
513 }
514 } else {
515 if (np->n_flag & NMODIFIED) {
3b568787
MD
516 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1))
517 == EINTR) {
984263bc 518 return (error);
3b568787 519 }
984263bc
MD
520 np->n_attrstamp = 0;
521 if (vp->v_type == VDIR)
522 np->n_direofoffset = 0;
3b568787 523 error = VOP_GETATTR(vp, &vattr, ap->a_td);
984263bc
MD
524 if (error)
525 return (error);
526 np->n_mtime = vattr.va_mtime.tv_sec;
527 } else {
3b568787 528 error = VOP_GETATTR(vp, &vattr, ap->a_td);
984263bc
MD
529 if (error)
530 return (error);
531 if (np->n_mtime != vattr.va_mtime.tv_sec) {
532 if (vp->v_type == VDIR)
533 np->n_direofoffset = 0;
534 if ((error = nfs_vinvalbuf(vp, V_SAVE,
3b568787 535 ap->a_td, 1)) == EINTR) {
984263bc 536 return (error);
3b568787 537 }
984263bc
MD
538 np->n_mtime = vattr.va_mtime.tv_sec;
539 }
540 }
541 }
542 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
543 np->n_attrstamp = 0; /* For Open/Close consistency */
544 return (0);
545}
546
547/*
548 * nfs close vnode op
549 * What an NFS client should do upon close after writing is a debatable issue.
550 * Most NFS clients push delayed writes to the server upon close, basically for
551 * two reasons:
552 * 1 - So that any write errors may be reported back to the client process
553 * doing the close system call. By far the two most likely errors are
554 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
555 * 2 - To put a worst case upper bound on cache inconsistency between
556 * multiple clients for the file.
557 * There is also a consistency problem for Version 2 of the protocol w.r.t.
558 * not being able to tell if other clients are writing a file concurrently,
559 * since there is no way of knowing if the changed modify time in the reply
560 * is only due to the write for this client.
561 * (NFS Version 3 provides weak cache consistency data in the reply that
562 * should be sufficient to detect and handle this case.)
563 *
564 * The current code does the following:
565 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
566 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
567 * or commit them (this satisfies 1 and 2 except for the
568 * case where the server crashes after this close but
569 * before the commit RPC, which is felt to be "good
570 * enough". Changing the last argument to nfs_flush() to
571 * a 1 would force a commit operation, if it is felt a
572 * commit is necessary now.
573 * for NQNFS - do nothing now, since 2 is dealt with via leases and
574 * 1 should be dealt with via an fsync() system call for
575 * cases where write errors are important.
576 */
577/* ARGSUSED */
578static int
579nfs_close(ap)
580 struct vop_close_args /* {
581 struct vnodeop_desc *a_desc;
582 struct vnode *a_vp;
583 int a_fflag;
584 struct ucred *a_cred;
dadab5e9 585 struct thread *a_td;
984263bc
MD
586 } */ *ap;
587{
40393ded
RG
588 struct vnode *vp = ap->a_vp;
589 struct nfsnode *np = VTONFS(vp);
984263bc
MD
590 int error = 0;
591
592 if (vp->v_type == VREG) {
593 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
594 (np->n_flag & NMODIFIED)) {
595 if (NFS_ISV3(vp)) {
596 /*
597 * Under NFSv3 we have dirty buffers to dispose of. We
598 * must flush them to the NFS server. We have the option
599 * of waiting all the way through the commit rpc or just
600 * waiting for the initial write. The default is to only
601 * wait through the initial write so the data is in the
602 * server's cache, which is roughly similar to the state
603 * a standard disk subsystem leaves the file in on close().
604 *
605 * We cannot clear the NMODIFIED bit in np->n_flag due to
606 * potential races with other processes, and certainly
607 * cannot clear it if we don't commit.
608 */
609 int cm = nfsv3_commit_on_close ? 1 : 0;
3b568787 610 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
984263bc
MD
611 /* np->n_flag &= ~NMODIFIED; */
612 } else {
3b568787 613 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
984263bc
MD
614 }
615 np->n_attrstamp = 0;
616 }
617 if (np->n_flag & NWRITEERR) {
618 np->n_flag &= ~NWRITEERR;
619 error = np->n_error;
620 }
621 }
622 return (error);
623}
624
625/*
626 * nfs getattr call from vfs.
627 */
628static int
629nfs_getattr(ap)
630 struct vop_getattr_args /* {
631 struct vnode *a_vp;
632 struct vattr *a_vap;
633 struct ucred *a_cred;
dadab5e9 634 struct thread *a_td;
984263bc
MD
635 } */ *ap;
636{
40393ded
RG
637 struct vnode *vp = ap->a_vp;
638 struct nfsnode *np = VTONFS(vp);
639 caddr_t cp;
640 u_int32_t *tl;
641 int32_t t1, t2;
984263bc
MD
642 caddr_t bpos, dpos;
643 int error = 0;
644 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
645 int v3 = NFS_ISV3(vp);
646
647 /*
648 * Update local times for special files.
649 */
650 if (np->n_flag & (NACC | NUPD))
651 np->n_flag |= NCHG;
652 /*
653 * First look in the cache.
654 */
655 if (nfs_getattrcache(vp, ap->a_vap) == 0)
656 return (0);
657
658 if (v3 && nfsaccess_cache_timeout > 0) {
659 nfsstats.accesscache_misses++;
c1cf1e59 660 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK));
984263bc
MD
661 if (nfs_getattrcache(vp, ap->a_vap) == 0)
662 return (0);
663 }
664
665 nfsstats.rpccnt[NFSPROC_GETATTR]++;
666 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
667 nfsm_fhtom(vp, v3);
c1cf1e59 668 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK));
984263bc
MD
669 if (!error) {
670 nfsm_loadattr(vp, ap->a_vap);
671 }
6b08710e
MD
672 m_freem(mrep);
673nfsmout:
984263bc
MD
674 return (error);
675}
676
677/*
678 * nfs setattr call.
679 */
680static int
681nfs_setattr(ap)
682 struct vop_setattr_args /* {
683 struct vnodeop_desc *a_desc;
684 struct vnode *a_vp;
685 struct vattr *a_vap;
686 struct ucred *a_cred;
dadab5e9 687 struct thread *a_td;
984263bc
MD
688 } */ *ap;
689{
40393ded
RG
690 struct vnode *vp = ap->a_vp;
691 struct nfsnode *np = VTONFS(vp);
692 struct vattr *vap = ap->a_vap;
984263bc
MD
693 int error = 0;
694 u_quad_t tsize;
695
696#ifndef nolint
697 tsize = (u_quad_t)0;
698#endif
699
700 /*
701 * Setting of flags is not supported.
702 */
703 if (vap->va_flags != VNOVAL)
704 return (EOPNOTSUPP);
705
706 /*
707 * Disallow write attempts if the filesystem is mounted read-only.
708 */
709 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
710 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
711 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
712 (vp->v_mount->mnt_flag & MNT_RDONLY))
713 return (EROFS);
714 if (vap->va_size != VNOVAL) {
715 switch (vp->v_type) {
716 case VDIR:
717 return (EISDIR);
718 case VCHR:
719 case VBLK:
720 case VSOCK:
721 case VFIFO:
722 if (vap->va_mtime.tv_sec == VNOVAL &&
723 vap->va_atime.tv_sec == VNOVAL &&
724 vap->va_mode == (mode_t)VNOVAL &&
725 vap->va_uid == (uid_t)VNOVAL &&
726 vap->va_gid == (gid_t)VNOVAL)
727 return (0);
728 vap->va_size = VNOVAL;
729 break;
730 default:
731 /*
732 * Disallow write attempts if the filesystem is
733 * mounted read-only.
734 */
735 if (vp->v_mount->mnt_flag & MNT_RDONLY)
736 return (EROFS);
737
738 /*
739 * We run vnode_pager_setsize() early (why?),
740 * we must set np->n_size now to avoid vinvalbuf
741 * V_SAVE races that might setsize a lower
742 * value.
743 */
744
745 tsize = np->n_size;
3b568787 746 error = nfs_meta_setsize(vp, ap->a_td, vap->va_size);
984263bc
MD
747
748 if (np->n_flag & NMODIFIED) {
749 if (vap->va_size == 0)
3b568787 750 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
984263bc 751 else
3b568787 752 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
984263bc
MD
753 if (error) {
754 np->n_size = tsize;
755 vnode_pager_setsize(vp, np->n_size);
756 return (error);
757 }
758 }
b07fc55c
DR
759 /* np->n_size has already been set to vap->va_size
760 * in nfs_meta_setsize(). We must set it again since
761 * nfs_loadattrcache() could be called through
762 * nfs_meta_setsize() and could modify np->n_size.
763 */
764 np->n_vattr.va_size = np->n_size = vap->va_size;
765 };
984263bc
MD
766 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
767 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
768 vp->v_type == VREG &&
3b568787 769 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR)
984263bc 770 return (error);
dadab5e9 771 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
984263bc
MD
772 if (error && vap->va_size != VNOVAL) {
773 np->n_size = np->n_vattr.va_size = tsize;
774 vnode_pager_setsize(vp, np->n_size);
775 }
776 return (error);
777}
778
779/*
780 * Do an nfs setattr rpc.
781 */
782static int
dadab5e9
MD
783nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
784 struct ucred *cred, struct thread *td)
984263bc 785{
40393ded
RG
786 struct nfsv2_sattr *sp;
787 caddr_t cp;
788 int32_t t1, t2;
984263bc
MD
789 caddr_t bpos, dpos, cp2;
790 u_int32_t *tl;
791 int error = 0, wccflag = NFSV3_WCCRATTR;
792 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
793 int v3 = NFS_ISV3(vp);
794
795 nfsstats.rpccnt[NFSPROC_SETATTR]++;
796 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
797 nfsm_fhtom(vp, v3);
798 if (v3) {
799 nfsm_v3attrbuild(vap, TRUE);
800 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
801 *tl = nfs_false;
802 } else {
803 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
804 if (vap->va_mode == (mode_t)VNOVAL)
805 sp->sa_mode = nfs_xdrneg1;
806 else
807 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
808 if (vap->va_uid == (uid_t)VNOVAL)
809 sp->sa_uid = nfs_xdrneg1;
810 else
811 sp->sa_uid = txdr_unsigned(vap->va_uid);
812 if (vap->va_gid == (gid_t)VNOVAL)
813 sp->sa_gid = nfs_xdrneg1;
814 else
815 sp->sa_gid = txdr_unsigned(vap->va_gid);
816 sp->sa_size = txdr_unsigned(vap->va_size);
817 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
818 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
819 }
dadab5e9 820 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
984263bc
MD
821 if (v3) {
822 nfsm_wcc_data(vp, wccflag);
823 } else
824 nfsm_loadattr(vp, (struct vattr *)0);
6b08710e
MD
825 m_freem(mrep);
826nfsmout:
984263bc
MD
827 return (error);
828}
829
830/*
831 * nfs lookup call, one step at a time...
832 * First look in cache
833 * If not found, unlock the directory nfsnode and do the rpc
834 */
835static int
836nfs_lookup(ap)
837 struct vop_lookup_args /* {
838 struct vnodeop_desc *a_desc;
839 struct vnode *a_dvp;
840 struct vnode **a_vpp;
841 struct componentname *a_cnp;
842 } */ *ap;
843{
844 struct componentname *cnp = ap->a_cnp;
845 struct vnode *dvp = ap->a_dvp;
846 struct vnode **vpp = ap->a_vpp;
847 int flags = cnp->cn_flags;
848 struct vnode *newvp;
849 u_int32_t *tl;
850 caddr_t cp;
851 int32_t t1, t2;
852 struct nfsmount *nmp;
853 caddr_t bpos, dpos, cp2;
854 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
855 long len;
856 nfsfh_t *fhp;
857 struct nfsnode *np;
858 int lockparent, wantparent, error = 0, attrflag, fhsize;
859 int v3 = NFS_ISV3(dvp);
dadab5e9 860 struct thread *td = cnp->cn_td;
984263bc
MD
861
862 *vpp = NULLVP;
2b69e610
MD
863 if ((flags & CNP_ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
864 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
984263bc
MD
865 return (EROFS);
866 if (dvp->v_type != VDIR)
867 return (ENOTDIR);
2b69e610
MD
868 lockparent = flags & CNP_LOCKPARENT;
869 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
984263bc
MD
870 nmp = VFSTONFS(dvp->v_mount);
871 np = VTONFS(dvp);
bc0c094e 872 if ((error = cache_lookup(dvp, NCPNULL, vpp, NCPPNULL, cnp)) && error != ENOENT) {
984263bc
MD
873 struct vattr vattr;
874 int vpid;
875
dadab5e9 876 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
984263bc
MD
877 *vpp = NULLVP;
878 return (error);
879 }
880
881 newvp = *vpp;
882 vpid = newvp->v_id;
883 /*
884 * See the comment starting `Step through' in ufs/ufs_lookup.c
885 * for an explanation of the locking protocol
886 */
887 if (dvp == newvp) {
888 VREF(newvp);
889 error = 0;
2b69e610 890 } else if (flags & CNP_ISDOTDOT) {
41a01a4d
MD
891 VOP_UNLOCK(dvp, NULL, 0, td);
892 error = vget(newvp, NULL, LK_EXCLUSIVE, td);
2b69e610 893 if (!error && lockparent && (flags & CNP_ISLASTCN))
41a01a4d 894 error = vn_lock(dvp, NULL, LK_EXCLUSIVE, td);
984263bc 895 } else {
41a01a4d 896 error = vget(newvp, NULL, LK_EXCLUSIVE, td);
2b69e610 897 if (!lockparent || error || !(flags & CNP_ISLASTCN))
41a01a4d 898 VOP_UNLOCK(dvp, NULL, 0, td);
984263bc
MD
899 }
900 if (!error) {
901 if (vpid == newvp->v_id) {
3b568787 902 if (!VOP_GETATTR(newvp, &vattr, td)
984263bc
MD
903 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
904 nfsstats.lookupcache_hits++;
2b69e610
MD
905 if (cnp->cn_nameiop != NAMEI_LOOKUP &&
906 (flags & CNP_ISLASTCN))
907 cnp->cn_flags |= CNP_SAVENAME;
984263bc
MD
908 return (0);
909 }
910 cache_purge(newvp);
911 }
912 vput(newvp);
2b69e610 913 if (lockparent && dvp != newvp && (flags & CNP_ISLASTCN))
41a01a4d 914 VOP_UNLOCK(dvp, NULL, 0, td);
984263bc 915 }
41a01a4d 916 error = vn_lock(dvp, NULL, LK_EXCLUSIVE, td);
984263bc
MD
917 *vpp = NULLVP;
918 if (error)
919 return (error);
920 }
921 error = 0;
922 newvp = NULLVP;
923 nfsstats.lookupcache_misses++;
924 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
925 len = cnp->cn_namelen;
926 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
927 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
928 nfsm_fhtom(dvp, v3);
929 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
dadab5e9 930 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred);
984263bc
MD
931 if (error) {
932 nfsm_postop_attr(dvp, attrflag);
933 m_freem(mrep);
934 goto nfsmout;
935 }
936 nfsm_getfh(fhp, fhsize, v3);
937
938 /*
939 * Handle RENAME case...
940 */
2b69e610 941 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent && (flags & CNP_ISLASTCN)) {
984263bc
MD
942 if (NFS_CMPFH(np, fhp, fhsize)) {
943 m_freem(mrep);
944 return (EISDIR);
945 }
946 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
947 if (error) {
948 m_freem(mrep);
949 return (error);
950 }
951 newvp = NFSTOV(np);
952 if (v3) {
953 nfsm_postop_attr(newvp, attrflag);
954 nfsm_postop_attr(dvp, attrflag);
955 } else
956 nfsm_loadattr(newvp, (struct vattr *)0);
957 *vpp = newvp;
958 m_freem(mrep);
2b69e610 959 cnp->cn_flags |= CNP_SAVENAME;
984263bc 960 if (!lockparent)
41a01a4d 961 VOP_UNLOCK(dvp, NULL, 0, td);
984263bc
MD
962 return (0);
963 }
964
2b69e610 965 if (flags & CNP_ISDOTDOT) {
41a01a4d 966 VOP_UNLOCK(dvp, NULL, 0, td);
984263bc
MD
967 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
968 if (error) {
41a01a4d 969 vn_lock(dvp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
984263bc
MD
970 return (error);
971 }
972 newvp = NFSTOV(np);
2b69e610 973 if (lockparent && (flags & CNP_ISLASTCN) &&
41a01a4d 974 (error = vn_lock(dvp, NULL, LK_EXCLUSIVE, td))) {
984263bc
MD
975 vput(newvp);
976 return (error);
977 }
978 } else if (NFS_CMPFH(np, fhp, fhsize)) {
979 VREF(dvp);
980 newvp = dvp;
981 } else {
982 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
983 if (error) {
984 m_freem(mrep);
985 return (error);
986 }
2b69e610 987 if (!lockparent || !(flags & CNP_ISLASTCN))
41a01a4d 988 VOP_UNLOCK(dvp, NULL, 0, td);
984263bc
MD
989 newvp = NFSTOV(np);
990 }
991 if (v3) {
992 nfsm_postop_attr(newvp, attrflag);
993 nfsm_postop_attr(dvp, attrflag);
994 } else
995 nfsm_loadattr(newvp, (struct vattr *)0);
2b69e610
MD
996 if (cnp->cn_nameiop != NAMEI_LOOKUP && (flags & CNP_ISLASTCN))
997 cnp->cn_flags |= CNP_SAVENAME;
998 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
999 (cnp->cn_nameiop != NAMEI_DELETE || !(flags & CNP_ISLASTCN))) {
984263bc 1000 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
bc0c094e 1001 cache_enter(dvp, NCPNULL, newvp, cnp);
984263bc
MD
1002 }
1003 *vpp = newvp;
6b08710e
MD
1004 m_freem(mrep);
1005nfsmout:
984263bc
MD
1006 if (error) {
1007 if (newvp != NULLVP) {
1008 vrele(newvp);
1009 *vpp = NULLVP;
1010 }
2b69e610
MD
1011 if ((cnp->cn_nameiop == NAMEI_CREATE || cnp->cn_nameiop == NAMEI_RENAME) &&
1012 (flags & CNP_ISLASTCN) && error == ENOENT) {
984263bc 1013 if (!lockparent)
41a01a4d 1014 VOP_UNLOCK(dvp, NULL, 0, td);
984263bc
MD
1015 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1016 error = EROFS;
1017 else
1018 error = EJUSTRETURN;
1019 }
2b69e610
MD
1020 if (cnp->cn_nameiop != NAMEI_LOOKUP && (flags & CNP_ISLASTCN))
1021 cnp->cn_flags |= CNP_SAVENAME;
984263bc
MD
1022 }
1023 return (error);
1024}
1025
1026/*
1027 * nfs read call.
1028 * Just call nfs_bioread() to do the work.
1029 */
1030static int
1031nfs_read(ap)
1032 struct vop_read_args /* {
1033 struct vnode *a_vp;
1034 struct uio *a_uio;
1035 int a_ioflag;
1036 struct ucred *a_cred;
1037 } */ *ap;
1038{
40393ded 1039 struct vnode *vp = ap->a_vp;
984263bc 1040
3b568787 1041 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
ca3a2b2f
HP
1042 switch (vp->v_type) {
1043 case VREG:
1044 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1045 case VDIR:
1046 return (EISDIR);
1047 default:
1048 return EOPNOTSUPP;
1049 }
984263bc
MD
1050}
1051
1052/*
1053 * nfs readlink call
1054 */
1055static int
1056nfs_readlink(ap)
1057 struct vop_readlink_args /* {
1058 struct vnode *a_vp;
1059 struct uio *a_uio;
1060 struct ucred *a_cred;
1061 } */ *ap;
1062{
40393ded 1063 struct vnode *vp = ap->a_vp;
984263bc
MD
1064
1065 if (vp->v_type != VLNK)
1066 return (EINVAL);
3b568787 1067 return (nfs_bioread(vp, ap->a_uio, 0));
984263bc
MD
1068}
1069
1070/*
1071 * Do a readlink rpc.
1072 * Called by nfs_doio() from below the buffer cache.
1073 */
1074int
3b568787 1075nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
984263bc 1076{
40393ded
RG
1077 u_int32_t *tl;
1078 caddr_t cp;
1079 int32_t t1, t2;
984263bc
MD
1080 caddr_t bpos, dpos, cp2;
1081 int error = 0, len, attrflag;
1082 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1083 int v3 = NFS_ISV3(vp);
1084
1085 nfsstats.rpccnt[NFSPROC_READLINK]++;
1086 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1087 nfsm_fhtom(vp, v3);
c1cf1e59 1088 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK));
984263bc
MD
1089 if (v3)
1090 nfsm_postop_attr(vp, attrflag);
1091 if (!error) {
1092 nfsm_strsiz(len, NFS_MAXPATHLEN);
1093 if (len == NFS_MAXPATHLEN) {
1094 struct nfsnode *np = VTONFS(vp);
1095 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1096 len = np->n_size;
1097 }
1098 nfsm_mtouio(uiop, len);
1099 }
6b08710e
MD
1100 m_freem(mrep);
1101nfsmout:
984263bc
MD
1102 return (error);
1103}
1104
1105/*
1106 * nfs read rpc call
1107 * Ditto above
1108 */
1109int
3b568787 1110nfs_readrpc(struct vnode *vp, struct uio *uiop)
984263bc 1111{
40393ded
RG
1112 u_int32_t *tl;
1113 caddr_t cp;
1114 int32_t t1, t2;
984263bc
MD
1115 caddr_t bpos, dpos, cp2;
1116 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1117 struct nfsmount *nmp;
1118 int error = 0, len, retlen, tsiz, eof, attrflag;
1119 int v3 = NFS_ISV3(vp);
1120
1121#ifndef nolint
1122 eof = 0;
1123#endif
1124 nmp = VFSTONFS(vp->v_mount);
1125 tsiz = uiop->uio_resid;
1126 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1127 return (EFBIG);
1128 while (tsiz > 0) {
1129 nfsstats.rpccnt[NFSPROC_READ]++;
1130 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1131 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1132 nfsm_fhtom(vp, v3);
1133 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1134 if (v3) {
1135 txdr_hyper(uiop->uio_offset, tl);
1136 *(tl + 2) = txdr_unsigned(len);
1137 } else {
1138 *tl++ = txdr_unsigned(uiop->uio_offset);
1139 *tl++ = txdr_unsigned(len);
1140 *tl = 0;
1141 }
c1cf1e59 1142 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ));
984263bc
MD
1143 if (v3) {
1144 nfsm_postop_attr(vp, attrflag);
1145 if (error) {
1146 m_freem(mrep);
1147 goto nfsmout;
1148 }
1149 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1150 eof = fxdr_unsigned(int, *(tl + 1));
1151 } else
1152 nfsm_loadattr(vp, (struct vattr *)0);
1153 nfsm_strsiz(retlen, nmp->nm_rsize);
1154 nfsm_mtouio(uiop, retlen);
1155 m_freem(mrep);
1156 tsiz -= retlen;
1157 if (v3) {
1158 if (eof || retlen == 0) {
1159 tsiz = 0;
1160 }
1161 } else if (retlen < len) {
1162 tsiz = 0;
1163 }
1164 }
1165nfsmout:
1166 return (error);
1167}
1168
1169/*
1170 * nfs write call
1171 */
1172int
3b568787 1173nfs_writerpc(vp, uiop, iomode, must_commit)
40393ded
RG
1174 struct vnode *vp;
1175 struct uio *uiop;
984263bc
MD
1176 int *iomode, *must_commit;
1177{
40393ded
RG
1178 u_int32_t *tl;
1179 caddr_t cp;
1180 int32_t t1, t2, backup;
984263bc
MD
1181 caddr_t bpos, dpos, cp2;
1182 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1183 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1184 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1185 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1186
1187#ifndef DIAGNOSTIC
1188 if (uiop->uio_iovcnt != 1)
1189 panic("nfs: writerpc iovcnt > 1");
1190#endif
1191 *must_commit = 0;
1192 tsiz = uiop->uio_resid;
1193 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1194 return (EFBIG);
1195 while (tsiz > 0) {
1196 nfsstats.rpccnt[NFSPROC_WRITE]++;
1197 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1198 nfsm_reqhead(vp, NFSPROC_WRITE,
1199 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1200 nfsm_fhtom(vp, v3);
1201 if (v3) {
1202 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1203 txdr_hyper(uiop->uio_offset, tl);
1204 tl += 2;
1205 *tl++ = txdr_unsigned(len);
1206 *tl++ = txdr_unsigned(*iomode);
1207 *tl = txdr_unsigned(len);
1208 } else {
40393ded 1209 u_int32_t x;
984263bc
MD
1210
1211 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1212 /* Set both "begin" and "current" to non-garbage. */
1213 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1214 *tl++ = x; /* "begin offset" */
1215 *tl++ = x; /* "current offset" */
1216 x = txdr_unsigned(len);
1217 *tl++ = x; /* total to this offset */
1218 *tl = x; /* size of this write */
1219 }
1220 nfsm_uiotom(uiop, len);
c1cf1e59 1221 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE));
984263bc
MD
1222 if (v3) {
1223 wccflag = NFSV3_WCCCHK;
1224 nfsm_wcc_data(vp, wccflag);
1225 if (!error) {
1226 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1227 + NFSX_V3WRITEVERF);
1228 rlen = fxdr_unsigned(int, *tl++);
1229 if (rlen == 0) {
1230 error = NFSERR_IO;
1231 m_freem(mrep);
1232 break;
1233 } else if (rlen < len) {
1234 backup = len - rlen;
1235 uiop->uio_iov->iov_base -= backup;
1236 uiop->uio_iov->iov_len += backup;
1237 uiop->uio_offset -= backup;
1238 uiop->uio_resid += backup;
1239 len = rlen;
1240 }
1241 commit = fxdr_unsigned(int, *tl++);
1242
1243 /*
1244 * Return the lowest committment level
1245 * obtained by any of the RPCs.
1246 */
1247 if (committed == NFSV3WRITE_FILESYNC)
1248 committed = commit;
1249 else if (committed == NFSV3WRITE_DATASYNC &&
1250 commit == NFSV3WRITE_UNSTABLE)
1251 committed = commit;
1252 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1253 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1254 NFSX_V3WRITEVERF);
1255 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1256 } else if (bcmp((caddr_t)tl,
1257 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1258 *must_commit = 1;
1259 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1260 NFSX_V3WRITEVERF);
1261 }
1262 }
1263 } else
1264 nfsm_loadattr(vp, (struct vattr *)0);
1265 if (wccflag)
1266 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime.tv_sec;
1267 m_freem(mrep);
1268 if (error)
1269 break;
1270 tsiz -= len;
1271 }
1272nfsmout:
1273 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1274 committed = NFSV3WRITE_FILESYNC;
1275 *iomode = committed;
1276 if (error)
1277 uiop->uio_resid = tsiz;
1278 return (error);
1279}
1280
1281/*
1282 * nfs mknod rpc
1283 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1284 * mode set to specify the file type and the size field for rdev.
1285 */
1286static int
1287nfs_mknodrpc(dvp, vpp, cnp, vap)
40393ded
RG
1288 struct vnode *dvp;
1289 struct vnode **vpp;
1290 struct componentname *cnp;
1291 struct vattr *vap;
984263bc 1292{
40393ded
RG
1293 struct nfsv2_sattr *sp;
1294 u_int32_t *tl;
1295 caddr_t cp;
1296 int32_t t1, t2;
984263bc
MD
1297 struct vnode *newvp = (struct vnode *)0;
1298 struct nfsnode *np = (struct nfsnode *)0;
1299 struct vattr vattr;
1300 char *cp2;
1301 caddr_t bpos, dpos;
1302 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1303 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1304 u_int32_t rdev;
1305 int v3 = NFS_ISV3(dvp);
1306
1307 if (vap->va_type == VCHR || vap->va_type == VBLK)
1308 rdev = txdr_unsigned(vap->va_rdev);
1309 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1310 rdev = nfs_xdrneg1;
1311 else {
1312 return (EOPNOTSUPP);
1313 }
3b568787 1314 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
984263bc
MD
1315 return (error);
1316 }
1317 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1318 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1319 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1320 nfsm_fhtom(dvp, v3);
1321 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1322 if (v3) {
1323 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1324 *tl++ = vtonfsv3_type(vap->va_type);
1325 nfsm_v3attrbuild(vap, FALSE);
1326 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1327 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1328 *tl++ = txdr_unsigned(umajor(vap->va_rdev));
1329 *tl = txdr_unsigned(uminor(vap->va_rdev));
1330 }
1331 } else {
1332 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1333 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1334 sp->sa_uid = nfs_xdrneg1;
1335 sp->sa_gid = nfs_xdrneg1;
1336 sp->sa_size = rdev;
1337 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1338 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1339 }
dadab5e9 1340 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred);
984263bc
MD
1341 if (!error) {
1342 nfsm_mtofh(dvp, newvp, v3, gotvp);
1343 if (!gotvp) {
1344 if (newvp) {
1345 vput(newvp);
1346 newvp = (struct vnode *)0;
1347 }
1348 error = nfs_lookitup(dvp, cnp->cn_nameptr,
dadab5e9 1349 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
984263bc
MD
1350 if (!error)
1351 newvp = NFSTOV(np);
1352 }
1353 }
1354 if (v3)
1355 nfsm_wcc_data(dvp, wccflag);
6b08710e
MD
1356 m_freem(mrep);
1357nfsmout:
984263bc
MD
1358 if (error) {
1359 if (newvp)
1360 vput(newvp);
1361 } else {
2b69e610 1362 if (cnp->cn_flags & CNP_MAKEENTRY)
bc0c094e 1363 cache_enter(dvp, NCPNULL, newvp, cnp);
984263bc
MD
1364 *vpp = newvp;
1365 }
1366 VTONFS(dvp)->n_flag |= NMODIFIED;
1367 if (!wccflag)
1368 VTONFS(dvp)->n_attrstamp = 0;
1369 return (error);
1370}
1371
1372/*
1373 * nfs mknod vop
1374 * just call nfs_mknodrpc() to do the work.
1375 */
1376/* ARGSUSED */
1377static int
1378nfs_mknod(ap)
1379 struct vop_mknod_args /* {
1380 struct vnode *a_dvp;
1381 struct vnode **a_vpp;
1382 struct componentname *a_cnp;
1383 struct vattr *a_vap;
1384 } */ *ap;
1385{
1386 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1387}
1388
1389static u_long create_verf;
1390/*
1391 * nfs file create call
1392 */
1393static int
1394nfs_create(ap)
1395 struct vop_create_args /* {
1396 struct vnode *a_dvp;
1397 struct vnode **a_vpp;
1398 struct componentname *a_cnp;
1399 struct vattr *a_vap;
1400 } */ *ap;
1401{
40393ded
RG
1402 struct vnode *dvp = ap->a_dvp;
1403 struct vattr *vap = ap->a_vap;
1404 struct componentname *cnp = ap->a_cnp;
1405 struct nfsv2_sattr *sp;
1406 u_int32_t *tl;
1407 caddr_t cp;
1408 int32_t t1, t2;
984263bc
MD
1409 struct nfsnode *np = (struct nfsnode *)0;
1410 struct vnode *newvp = (struct vnode *)0;
1411 caddr_t bpos, dpos, cp2;
1412 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1413 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1414 struct vattr vattr;
1415 int v3 = NFS_ISV3(dvp);
1416
1417 /*
1418 * Oops, not for me..
1419 */
1420 if (vap->va_type == VSOCK)
1421 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1422
3b568787 1423 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
984263bc
MD
1424 return (error);
1425 }
1426 if (vap->va_vaflags & VA_EXCLUSIVE)
1427 fmode |= O_EXCL;
1428again:
1429 nfsstats.rpccnt[NFSPROC_CREATE]++;
1430 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1431 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1432 nfsm_fhtom(dvp, v3);
1433 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1434 if (v3) {
1435 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1436 if (fmode & O_EXCL) {
1437 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1438 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1439#ifdef INET
1440 if (!TAILQ_EMPTY(&in_ifaddrhead))
1441 *tl++ = IA_SIN(in_ifaddrhead.tqh_first)->sin_addr.s_addr;
1442 else
1443#endif
1444 *tl++ = create_verf;
1445 *tl = ++create_verf;
1446 } else {
1447 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1448 nfsm_v3attrbuild(vap, FALSE);
1449 }
1450 } else {
1451 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1452 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1453 sp->sa_uid = nfs_xdrneg1;
1454 sp->sa_gid = nfs_xdrneg1;
1455 sp->sa_size = 0;
1456 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1457 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1458 }
dadab5e9 1459 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred);
984263bc
MD
1460 if (!error) {
1461 nfsm_mtofh(dvp, newvp, v3, gotvp);
1462 if (!gotvp) {
1463 if (newvp) {
1464 vput(newvp);
1465 newvp = (struct vnode *)0;
1466 }
1467 error = nfs_lookitup(dvp, cnp->cn_nameptr,
dadab5e9 1468 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
984263bc
MD
1469 if (!error)
1470 newvp = NFSTOV(np);
1471 }
1472 }
1473 if (v3)
1474 nfsm_wcc_data(dvp, wccflag);
6b08710e
MD
1475 m_freem(mrep);
1476nfsmout:
984263bc
MD
1477 if (error) {
1478 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1479 fmode &= ~O_EXCL;
1480 goto again;
1481 }
1482 if (newvp)
1483 vput(newvp);
1484 } else if (v3 && (fmode & O_EXCL)) {
1485 /*
1486 * We are normally called with only a partially initialized
1487 * VAP. Since the NFSv3 spec says that server may use the
1488 * file attributes to store the verifier, the spec requires
1489 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1490 * in atime, but we can't really assume that all servers will
1491 * so we ensure that our SETATTR sets both atime and mtime.
1492 */
1493 if (vap->va_mtime.tv_sec == VNOVAL)
1494 vfs_timestamp(&vap->va_mtime);
1495 if (vap->va_atime.tv_sec == VNOVAL)
1496 vap->va_atime = vap->va_mtime;
dadab5e9 1497 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
984263bc
MD
1498 }
1499 if (!error) {
2b69e610 1500 if (cnp->cn_flags & CNP_MAKEENTRY)
bc0c094e 1501 cache_enter(dvp, NCPNULL, newvp, cnp);
c1cf1e59
MD
1502 /*
1503 * The new np may have enough info for access
1504 * checks, make sure rucred and wucred are
1505 * initialized for read and write rpc's.
1506 */
1507 np = VTONFS(newvp);
1508 if (np->n_rucred == NULL)
1509 np->n_rucred = crhold(cnp->cn_cred);
1510 if (np->n_wucred == NULL)
1511 np->n_wucred = crhold(cnp->cn_cred);
984263bc
MD
1512 *ap->a_vpp = newvp;
1513 }
1514 VTONFS(dvp)->n_flag |= NMODIFIED;
1515 if (!wccflag)
1516 VTONFS(dvp)->n_attrstamp = 0;
1517 return (error);
1518}
1519
1520/*
1521 * nfs file remove call
1522 * To try and make nfs semantics closer to ufs semantics, a file that has
1523 * other processes using the vnode is renamed instead of removed and then
1524 * removed later on the last close.
1525 * - If v_usecount > 1
1526 * If a rename is not already in the works
1527 * call nfs_sillyrename() to set it up
1528 * else
1529 * do the remove rpc
1530 */
1531static int
1532nfs_remove(ap)
1533 struct vop_remove_args /* {
1534 struct vnodeop_desc *a_desc;
1535 struct vnode * a_dvp;
1536 struct vnode * a_vp;
1537 struct componentname * a_cnp;
1538 } */ *ap;
1539{
40393ded
RG
1540 struct vnode *vp = ap->a_vp;
1541 struct vnode *dvp = ap->a_dvp;
1542 struct componentname *cnp = ap->a_cnp;
1543 struct nfsnode *np = VTONFS(vp);
984263bc
MD
1544 int error = 0;
1545 struct vattr vattr;
1546
1547#ifndef DIAGNOSTIC
2b69e610 1548 if ((cnp->cn_flags & CNP_HASBUF) == 0)
984263bc
MD
1549 panic("nfs_remove: no name");
1550 if (vp->v_usecount < 1)
1551 panic("nfs_remove: bad v_usecount");
1552#endif
1553 if (vp->v_type == VDIR)
1554 error = EPERM;
1555 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
3b568787 1556 VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 &&
984263bc
MD
1557 vattr.va_nlink > 1)) {
1558 /*
1559 * Purge the name cache so that the chance of a lookup for
1560 * the name succeeding while the remove is in progress is
1561 * minimized. Without node locking it can still happen, such
1562 * that an I/O op returns ESTALE, but since you get this if
1563 * another host removes the file..
1564 */
1565 cache_purge(vp);
1566 /*
1567 * throw away biocache buffers, mainly to avoid
1568 * unnecessary delayed writes later.
1569 */
3b568787 1570 error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1);
984263bc
MD
1571 /* Do the rpc */
1572 if (error != EINTR)
1573 error = nfs_removerpc(dvp, cnp->cn_nameptr,
dadab5e9 1574 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
984263bc
MD
1575 /*
1576 * Kludge City: If the first reply to the remove rpc is lost..
1577 * the reply to the retransmitted request will be ENOENT
1578 * since the file was in fact removed
1579 * Therefore, we cheat and return success.
1580 */
1581 if (error == ENOENT)
1582 error = 0;
1583 } else if (!np->n_sillyrename)
1584 error = nfs_sillyrename(dvp, vp, cnp);
1585 np->n_attrstamp = 0;
1586 return (error);
1587}
1588
1589/*
1590 * nfs file remove rpc called from nfs_inactive
1591 */
1592int
dadab5e9 1593nfs_removeit(struct sillyrename *sp)
984263bc
MD
1594{
1595
dadab5e9
MD
1596 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1597 sp->s_cred, NULL));
984263bc
MD
1598}
1599
1600/*
1601 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1602 */
1603static int
dadab5e9
MD
1604nfs_removerpc(dvp, name, namelen, cred, td)
1605 struct vnode *dvp;
984263bc
MD
1606 const char *name;
1607 int namelen;
1608 struct ucred *cred;
dadab5e9 1609 struct thread *td;
984263bc 1610{
40393ded
RG
1611 u_int32_t *tl;
1612 caddr_t cp;
1613 int32_t t1, t2;
984263bc
MD
1614 caddr_t bpos, dpos, cp2;
1615 int error = 0, wccflag = NFSV3_WCCRATTR;
1616 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1617 int v3 = NFS_ISV3(dvp);
1618
1619 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1620 nfsm_reqhead(dvp, NFSPROC_REMOVE,
1621 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1622 nfsm_fhtom(dvp, v3);
1623 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
dadab5e9 1624 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
984263bc
MD
1625 if (v3)
1626 nfsm_wcc_data(dvp, wccflag);
6b08710e
MD
1627 m_freem(mrep);
1628nfsmout:
984263bc
MD
1629 VTONFS(dvp)->n_flag |= NMODIFIED;
1630 if (!wccflag)
1631 VTONFS(dvp)->n_attrstamp = 0;
1632 return (error);
1633}
1634
1635/*
1636 * nfs file rename call
1637 */
1638static int
1639nfs_rename(ap)
1640 struct vop_rename_args /* {
1641 struct vnode *a_fdvp;
1642 struct vnode *a_fvp;
1643 struct componentname *a_fcnp;
1644 struct vnode *a_tdvp;
1645 struct vnode *a_tvp;
1646 struct componentname *a_tcnp;
1647 } */ *ap;
1648{
40393ded
RG
1649 struct vnode *fvp = ap->a_fvp;
1650 struct vnode *tvp = ap->a_tvp;
1651 struct vnode *fdvp = ap->a_fdvp;
1652 struct vnode *tdvp = ap->a_tdvp;
1653 struct componentname *tcnp = ap->a_tcnp;
1654 struct componentname *fcnp = ap->a_fcnp;
984263bc
MD
1655 int error;
1656
1657#ifndef DIAGNOSTIC
2b69e610
MD
1658 if ((tcnp->cn_flags & CNP_HASBUF) == 0 ||
1659 (fcnp->cn_flags & CNP_HASBUF) == 0)
984263bc
MD
1660 panic("nfs_rename: no name");
1661#endif
1662 /* Check for cross-device rename */
1663 if ((fvp->v_mount != tdvp->v_mount) ||
1664 (tvp && (fvp->v_mount != tvp->v_mount))) {
1665 error = EXDEV;
1666 goto out;
1667 }
1668
1669 /*
1670 * We have to flush B_DELWRI data prior to renaming
1671 * the file. If we don't, the delayed-write buffers
1672 * can be flushed out later after the file has gone stale
1673 * under NFSV3. NFSV2 does not have this problem because
1674 * ( as far as I can tell ) it flushes dirty buffers more
1675 * often.
1676 */
1677
3b568787 1678 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td);
984263bc 1679 if (tvp)
3b568787 1680 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td);
984263bc
MD
1681
1682 /*
1683 * If the tvp exists and is in use, sillyrename it before doing the
1684 * rename of the new file over it.
1685 * XXX Can't sillyrename a directory.
1686 */
1687 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1688 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1689 vput(tvp);
1690 tvp = NULL;
1691 }
1692
1693 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1694 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
dadab5e9 1695 tcnp->cn_td);
984263bc
MD
1696
1697 if (fvp->v_type == VDIR) {
1698 if (tvp != NULL && tvp->v_type == VDIR)
1699 cache_purge(tdvp);
1700 cache_purge(fdvp);
1701 }
1702
1703out:
1704 if (tdvp == tvp)
1705 vrele(tdvp);
1706 else
1707 vput(tdvp);
1708 if (tvp)
1709 vput(tvp);
1710 vrele(fdvp);
1711 vrele(fvp);
1712 /*
1713 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1714 */
1715 if (error == ENOENT)
1716 error = 0;
1717 return (error);
1718}
1719
1720/*
1721 * nfs file rename rpc called from nfs_remove() above
1722 */
1723static int
1724nfs_renameit(sdvp, scnp, sp)
1725 struct vnode *sdvp;
1726 struct componentname *scnp;
40393ded 1727 struct sillyrename *sp;
984263bc
MD
1728{
1729 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
dadab5e9 1730 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
984263bc
MD
1731}
1732
1733/*
1734 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1735 */
1736static int
dadab5e9
MD
1737nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, td)
1738 struct vnode *fdvp;
984263bc
MD
1739 const char *fnameptr;
1740 int fnamelen;
40393ded 1741 struct vnode *tdvp;
984263bc
MD
1742 const char *tnameptr;
1743 int tnamelen;
1744 struct ucred *cred;
dadab5e9 1745 struct thread *td;
984263bc 1746{
40393ded
RG
1747 u_int32_t *tl;
1748 caddr_t cp;
1749 int32_t t1, t2;
984263bc
MD
1750 caddr_t bpos, dpos, cp2;
1751 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1752 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1753 int v3 = NFS_ISV3(fdvp);
1754
1755 nfsstats.rpccnt[NFSPROC_RENAME]++;
1756 nfsm_reqhead(fdvp, NFSPROC_RENAME,
1757 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1758 nfsm_rndup(tnamelen));
1759 nfsm_fhtom(fdvp, v3);
1760 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1761 nfsm_fhtom(tdvp, v3);
1762 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
dadab5e9 1763 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
984263bc
MD
1764 if (v3) {
1765 nfsm_wcc_data(fdvp, fwccflag);
1766 nfsm_wcc_data(tdvp, twccflag);
1767 }
6b08710e
MD
1768 m_freem(mrep);
1769nfsmout:
984263bc
MD
1770 VTONFS(fdvp)->n_flag |= NMODIFIED;
1771 VTONFS(tdvp)->n_flag |= NMODIFIED;
1772 if (!fwccflag)
1773 VTONFS(fdvp)->n_attrstamp = 0;
1774 if (!twccflag)
1775 VTONFS(tdvp)->n_attrstamp = 0;
1776 return (error);
1777}
1778
1779/*
1780 * nfs hard link create call
1781 */
1782static int
1783nfs_link(ap)
1784 struct vop_link_args /* {
1785 struct vnode *a_tdvp;
1786 struct vnode *a_vp;
1787 struct componentname *a_cnp;
1788 } */ *ap;
1789{
40393ded
RG
1790 struct vnode *vp = ap->a_vp;
1791 struct vnode *tdvp = ap->a_tdvp;
1792 struct componentname *cnp = ap->a_cnp;
1793 u_int32_t *tl;
1794 caddr_t cp;
1795 int32_t t1, t2;
984263bc
MD
1796 caddr_t bpos, dpos, cp2;
1797 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1798 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1799 int v3;
1800
1801 if (vp->v_mount != tdvp->v_mount) {
1802 return (EXDEV);
1803 }
1804
1805 /*
1806 * Push all writes to the server, so that the attribute cache
1807 * doesn't get "out of sync" with the server.
1808 * XXX There should be a better way!
1809 */
3b568787 1810 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td);
984263bc
MD
1811
1812 v3 = NFS_ISV3(vp);
1813 nfsstats.rpccnt[NFSPROC_LINK]++;
1814 nfsm_reqhead(vp, NFSPROC_LINK,
1815 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1816 nfsm_fhtom(vp, v3);
1817 nfsm_fhtom(tdvp, v3);
1818 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
dadab5e9 1819 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred);
984263bc
MD
1820 if (v3) {
1821 nfsm_postop_attr(vp, attrflag);
1822 nfsm_wcc_data(tdvp, wccflag);
1823 }
6b08710e
MD
1824 m_freem(mrep);
1825nfsmout:
984263bc
MD
1826 VTONFS(tdvp)->n_flag |= NMODIFIED;
1827 if (!attrflag)
1828 VTONFS(vp)->n_attrstamp = 0;
1829 if (!wccflag)
1830 VTONFS(tdvp)->n_attrstamp = 0;
1831 /*
1832 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1833 */
1834 if (error == EEXIST)
1835 error = 0;
1836 return (error);
1837}
1838
1839/*
1840 * nfs symbolic link create call
1841 */
1842static int
1843nfs_symlink(ap)
1844 struct vop_symlink_args /* {
1845 struct vnode *a_dvp;
1846 struct vnode **a_vpp;
1847 struct componentname *a_cnp;
1848 struct vattr *a_vap;
1849 char *a_target;
1850 } */ *ap;
1851{
40393ded
RG
1852 struct vnode *dvp = ap->a_dvp;
1853 struct vattr *vap = ap->a_vap;
1854 struct componentname *cnp = ap->a_cnp;
1855 struct nfsv2_sattr *sp;
1856 u_int32_t *tl;
1857 caddr_t cp;
1858 int32_t t1, t2;
984263bc
MD
1859 caddr_t bpos, dpos, cp2;
1860 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1861 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1862 struct vnode *newvp = (struct vnode *)0;
1863 int v3 = NFS_ISV3(dvp);
1864
1865 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1866 slen = strlen(ap->a_target);
1867 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1868 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1869 nfsm_fhtom(dvp, v3);
1870 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1871 if (v3) {
1872 nfsm_v3attrbuild(vap, FALSE);
1873 }
1874 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1875 if (!v3) {
1876 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1877 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1878 sp->sa_uid = nfs_xdrneg1;
1879 sp->sa_gid = nfs_xdrneg1;
1880 sp->sa_size = nfs_xdrneg1;
1881 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1882 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1883 }
1884
1885 /*
1886 * Issue the NFS request and get the rpc response.
1887 *
1888 * Only NFSv3 responses returning an error of 0 actually return
1889 * a file handle that can be converted into newvp without having
1890 * to do an extra lookup rpc.
1891 */
dadab5e9 1892 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred);
984263bc
MD
1893 if (v3) {
1894 if (error == 0)
1895 nfsm_mtofh(dvp, newvp, v3, gotvp);
1896 nfsm_wcc_data(dvp, wccflag);
1897 }
1898
1899 /*
1900 * out code jumps -> here, mrep is also freed.
1901 */
1902
6b08710e
MD
1903 m_freem(mrep);
1904nfsmout:
984263bc
MD
1905
1906 /*
1907 * If we get an EEXIST error, silently convert it to no-error
1908 * in case of an NFS retry.
1909 */
1910 if (error == EEXIST)
1911 error = 0;
1912
1913 /*
1914 * If we do not have (or no longer have) an error, and we could
1915 * not extract the newvp from the response due to the request being
1916 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1917 * to obtain a newvp to return.
1918 */
1919 if (error == 0 && newvp == NULL) {
1920 struct nfsnode *np = NULL;
1921
1922 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
dadab5e9 1923 cnp->cn_cred, cnp->cn_td, &np);
984263bc
MD
1924 if (!error)
1925 newvp = NFSTOV(np);
1926 }
1927 if (error) {
1928 if (newvp)
1929 vput(newvp);
1930 } else {
1931 *ap->a_vpp = newvp;
1932 }
1933 VTONFS(dvp)->n_flag |= NMODIFIED;
1934 if (!wccflag)
1935 VTONFS(dvp)->n_attrstamp = 0;
1936 return (error);
1937}
1938
1939/*
1940 * nfs make dir call
1941 */
1942static int
1943nfs_mkdir(ap)
1944 struct vop_mkdir_args /* {
1945 struct vnode *a_dvp;
1946 struct vnode **a_vpp;
1947 struct componentname *a_cnp;
1948 struct vattr *a_vap;
1949 } */ *ap;
1950{
40393ded
RG
1951 struct vnode *dvp = ap->a_dvp;
1952 struct vattr *vap = ap->a_vap;
1953 struct componentname *cnp = ap->a_cnp;
1954 struct nfsv2_sattr *sp;
1955 u_int32_t *tl;
1956 caddr_t cp;
1957 int32_t t1, t2;
1958 int len;
984263bc
MD
1959 struct nfsnode *np = (struct nfsnode *)0;
1960 struct vnode *newvp = (struct vnode *)0;
1961 caddr_t bpos, dpos, cp2;
1962 int error = 0, wccflag = NFSV3_WCCRATTR;
1963 int gotvp = 0;
1964 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1965 struct vattr vattr;
1966 int v3 = NFS_ISV3(dvp);
1967
3b568787 1968 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
984263bc
MD
1969 return (error);
1970 }
1971 len = cnp->cn_namelen;
1972 nfsstats.rpccnt[NFSPROC_MKDIR]++;
1973 nfsm_reqhead(dvp, NFSPROC_MKDIR,
1974 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1975 nfsm_fhtom(dvp, v3);
1976 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1977 if (v3) {
1978 nfsm_v3attrbuild(vap, FALSE);
1979 } else {
1980 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1981 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1982 sp->sa_uid = nfs_xdrneg1;
1983 sp->sa_gid = nfs_xdrneg1;
1984 sp->sa_size = nfs_xdrneg1;
1985 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1986 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1987 }
dadab5e9 1988 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred);
984263bc
MD
1989 if (!error)
1990 nfsm_mtofh(dvp, newvp, v3, gotvp);
1991 if (v3)
1992 nfsm_wcc_data(dvp, wccflag);
6b08710e
MD
1993 m_freem(mrep);
1994nfsmout:
984263bc
MD
1995 VTONFS(dvp)->n_flag |= NMODIFIED;
1996 if (!wccflag)
1997 VTONFS(dvp)->n_attrstamp = 0;
1998 /*
1999 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2000 * if we can succeed in looking up the directory.
2001 */
2002 if (error == EEXIST || (!error && !gotvp)) {
2003 if (newvp) {
2004 vrele(newvp);
2005 newvp = (struct vnode *)0;
2006 }
2007 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
dadab5e9 2008 cnp->cn_td, &np);
984263bc
MD
2009 if (!error) {
2010 newvp = NFSTOV(np);
2011 if (newvp->v_type != VDIR)
2012 error = EEXIST;
2013 }
2014 }
2015 if (error) {
2016 if (newvp)
2017 vrele(newvp);
2018 } else
2019 *ap->a_vpp = newvp;
2020 return (error);
2021}
2022
2023/*
2024 * nfs remove directory call
2025 */
2026static int
2027nfs_rmdir(ap)
2028 struct vop_rmdir_args /* {
2029 struct vnode *a_dvp;
2030 struct vnode *a_vp;
2031 struct componentname *a_cnp;
2032 } */ *ap;
2033{
40393ded
RG
2034 struct vnode *vp = ap->a_vp;
2035 struct vnode *dvp = ap->a_dvp;
2036 struct componentname *cnp = ap->a_cnp;
2037 u_int32_t *tl;
2038 caddr_t cp;
2039 int32_t t1, t2;
984263bc
MD
2040 caddr_t bpos, dpos, cp2;
2041 int error = 0, wccflag = NFSV3_WCCRATTR;
2042 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2043 int v3 = NFS_ISV3(dvp);
2044
2045 if (dvp == vp)
2046 return (EINVAL);
2047 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2048 nfsm_reqhead(dvp, NFSPROC_RMDIR,
2049 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2050 nfsm_fhtom(dvp, v3);
2051 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
dadab5e9 2052 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred);
984263bc
MD
2053 if (v3)
2054 nfsm_wcc_data(dvp, wccflag);
6b08710e
MD
2055 m_freem(mrep);
2056nfsmout:
984263bc
MD
2057 VTONFS(dvp)->n_flag |= NMODIFIED;
2058 if (!wccflag)
2059 VTONFS(dvp)->n_attrstamp = 0;
2060 cache_purge(dvp);
2061 cache_purge(vp);
2062 /*
2063 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2064 */
2065 if (error == ENOENT)
2066 error = 0;
2067 return (error);
2068}
2069
2070/*
2071 * nfs readdir call
2072 */
2073static int
2074nfs_readdir(ap)
2075 struct vop_readdir_args /* {
2076 struct vnode *a_vp;
2077 struct uio *a_uio;
2078 struct ucred *a_cred;
2079 } */ *ap;
2080{
40393ded
RG
2081 struct vnode *vp = ap->a_vp;
2082 struct nfsnode *np = VTONFS(vp);
2083 struct uio *uio = ap->a_uio;
984263bc
MD
2084 int tresid, error;
2085 struct vattr vattr;
2086
2087 if (vp->v_type != VDIR)
2088 return (EPERM);
2089 /*
2090 * First, check for hit on the EOF offset cache
2091 */
2092 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2093 (np->n_flag & NMODIFIED) == 0) {
2094 if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) {
2095 if (NQNFS_CKCACHABLE(vp, ND_READ)) {
2096 nfsstats.direofcache_hits++;
2097 return (0);
2098 }
3b568787 2099 } else if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 &&
984263bc
MD
2100 np->n_mtime == vattr.va_mtime.tv_sec) {
2101 nfsstats.direofcache_hits++;
2102 return (0);
2103 }
2104 }
2105
2106 /*
2107 * Call nfs_bioread() to do the real work.
2108 */
2109 tresid = uio->uio_resid;
3b568787 2110 error = nfs_bioread(vp, uio, 0);
984263bc
MD
2111
2112 if (!error && uio->uio_resid == tresid)
2113 nfsstats.direofcache_misses++;
2114 return (error);
2115}
2116
2117/*
2118 * Readdir rpc call.
2119 * Called from below the buffer cache by nfs_doio().
2120 */
2121int
3b568787 2122nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
984263bc 2123{
40393ded
RG
2124 int len, left;
2125 struct dirent *dp = NULL;
2126 u_int32_t *tl;
2127 caddr_t cp;
2128 int32_t t1, t2;
2129 nfsuint64 *cookiep;
984263bc
MD
2130 caddr_t bpos, dpos, cp2;
2131 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2132 nfsuint64 cookie;
2133 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2134 struct nfsnode *dnp = VTONFS(vp);
2135 u_quad_t fileno;
2136 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2137 int attrflag;
2138 int v3 = NFS_ISV3(vp);
2139
2140#ifndef DIAGNOSTIC
2141 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2142 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2143 panic("nfs readdirrpc bad uio");
2144#endif
2145
2146 /*
2147 * If there is no cookie, assume directory was stale.
2148 */
2149 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2150 if (cookiep)
2151 cookie = *cookiep;
2152 else
2153 return (NFSERR_BAD_COOKIE);
2154 /*
2155 * Loop around doing readdir rpc's of size nm_readdirsize
2156 * truncated to a multiple of DIRBLKSIZ.
2157 * The stopping criteria is EOF or buffer full.
2158 */
2159 while (more_dirs && bigenough) {
2160 nfsstats.rpccnt[NFSPROC_READDIR]++;
2161 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2162 NFSX_READDIR(v3));
2163 nfsm_fhtom(vp, v3);
2164 if (v3) {
2165 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2166 *tl++ = cookie.nfsuquad[0];
2167 *tl++ = cookie.nfsuquad[1];
2168 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2169 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2170 } else {
2171 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2172 *tl++ = cookie.nfsuquad[0];
2173 }
2174 *tl = txdr_unsigned(nmp->nm_readdirsize);
c1cf1e59 2175 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ));
984263bc
MD
2176 if (v3) {
2177 nfsm_postop_attr(vp, attrflag);
2178 if (!error) {
2179 nfsm_dissect(tl, u_int32_t *,
2180 2 * NFSX_UNSIGNED);
2181 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2182 dnp->n_cookieverf.nfsuquad[1] = *tl;
2183 } else {
2184 m_freem(mrep);
2185 goto nfsmout;
2186 }
2187 }
2188 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2189 more_dirs = fxdr_unsigned(int, *tl);
2190
2191 /* loop thru the dir entries, doctoring them to 4bsd form */
2192 while (more_dirs && bigenough) {
2193 if (v3) {
2194 nfsm_dissect(tl, u_int32_t *,
2195 3 * NFSX_UNSIGNED);
2196 fileno = fxdr_hyper(tl);
2197 len = fxdr_unsigned(int, *(tl + 2));
2198 } else {
2199 nfsm_dissect(tl, u_int32_t *,
2200 2 * NFSX_UNSIGNED);
2201 fileno = fxdr_unsigned(u_quad_t, *tl++);
2202 len = fxdr_unsigned(int, *tl);
2203 }
2204 if (len <= 0 || len > NFS_MAXNAMLEN) {
2205 error = EBADRPC;
2206 m_freem(mrep);
2207 goto nfsmout;
2208 }
2209 tlen = nfsm_rndup(len);
2210 if (tlen == len)
2211 tlen += 4; /* To ensure null termination */
2212 left = DIRBLKSIZ - blksiz;
2213 if ((tlen + DIRHDSIZ) > left) {
2214 dp->d_reclen += left;
2215 uiop->uio_iov->iov_base += left;
2216 uiop->uio_iov->iov_len -= left;
2217 uiop->uio_offset += left;
2218 uiop->uio_resid -= left;
2219 blksiz = 0;
2220 }
2221 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2222 bigenough = 0;
2223 if (bigenough) {
2224 dp = (struct dirent *)uiop->uio_iov->iov_base;
2225 dp->d_fileno = (int)fileno;
2226 dp->d_namlen = len;
2227 dp->d_reclen = tlen + DIRHDSIZ;
2228 dp->d_type = DT_UNKNOWN;
2229 blksiz += dp->d_reclen;
2230 if (blksiz == DIRBLKSIZ)
2231 blksiz = 0;
2232 uiop->uio_offset += DIRHDSIZ;
2233 uiop->uio_resid -= DIRHDSIZ;
2234 uiop->uio_iov->iov_base += DIRHDSIZ;
2235 uiop->uio_iov->iov_len -= DIRHDSIZ;
2236 nfsm_mtouio(uiop, len);
2237 cp = uiop->uio_iov->iov_base;
2238 tlen -= len;
2239 *cp = '\0'; /* null terminate */
2240 uiop->uio_iov->iov_base += tlen;
2241 uiop->uio_iov->iov_len -= tlen;
2242 uiop->uio_offset += tlen;
2243 uiop->uio_resid -= tlen;
2244 } else
2245 nfsm_adv(nfsm_rndup(len));
2246 if (v3) {
2247 nfsm_dissect(tl, u_int32_t *,
2248 3 * NFSX_UNSIGNED);
2249 } else {
2250 nfsm_dissect(tl, u_int32_t *,
2251 2 * NFSX_UNSIGNED);
2252 }
2253 if (bigenough) {
2254 cookie.nfsuquad[0] = *tl++;
2255 if (v3)
2256 cookie.nfsuquad[1] = *tl++;
2257 } else if (v3)
2258 tl += 2;
2259 else
2260 tl++;
2261 more_dirs = fxdr_unsigned(int, *tl);
2262 }
2263 /*
2264 * If at end of rpc data, get the eof boolean
2265 */
2266 if (!more_dirs) {
2267 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2268 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2269 }
2270 m_freem(mrep);
2271 }
2272 /*
2273 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2274 * by increasing d_reclen for the last record.
2275 */
2276 if (blksiz > 0) {
2277 left = DIRBLKSIZ - blksiz;
2278 dp->d_reclen += left;
2279 uiop->uio_iov->iov_base += left;
2280 uiop->uio_iov->iov_len -= left;
2281 uiop->uio_offset += left;
2282 uiop->uio_resid -= left;
2283 }
2284
2285 /*
2286 * We are now either at the end of the directory or have filled the
2287 * block.
2288 */
2289 if (bigenough)
2290 dnp->n_direofoffset = uiop->uio_offset;
2291 else {
2292 if (uiop->uio_resid > 0)
2293 printf("EEK! readdirrpc resid > 0\n");
2294 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2295 *cookiep = cookie;
2296 }
2297nfsmout:
2298 return (error);
2299}
2300
2301/*
2302 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2303 */
2304int
3b568787 2305nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
984263bc 2306{
40393ded
RG
2307 int len, left;
2308 struct dirent *dp;
2309 u_int32_t *tl;
2310 caddr_t cp;
2311 int32_t t1, t2;
2312 struct vnode *newvp;
2313 nfsuint64 *cookiep;
984263bc
MD
2314 caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2315 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2316 struct nameidata nami, *ndp = &nami;
2317 struct componentname *cnp = &ndp->ni_cnd;
2318 nfsuint64 cookie;
2319 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2320 struct nfsnode *dnp = VTONFS(vp), *np;
2321 nfsfh_t *fhp;
2322 u_quad_t fileno;
2323 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2324 int attrflag, fhsize;
2325
2326#ifndef nolint
2327 dp = (struct dirent *)0;
2328#endif
2329#ifndef DIAGNOSTIC
2330 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2331 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2332 panic("nfs readdirplusrpc bad uio");
2333#endif
2334 ndp->ni_dvp = vp;
2335 newvp = NULLVP;
2336
2337 /*
2338 * If there is no cookie, assume directory was stale.
2339 */
2340 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2341 if (cookiep)
2342 cookie = *cookiep;
2343 else
2344 return (NFSERR_BAD_COOKIE);
2345 /*
2346 * Loop around doing readdir rpc's of size nm_readdirsize
2347 * truncated to a multiple of DIRBLKSIZ.
2348 * The stopping criteria is EOF or buffer full.
2349 */
2350 while (more_dirs && bigenough) {
2351 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2352 nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2353 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2354 nfsm_fhtom(vp, 1);
2355 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2356 *tl++ = cookie.nfsuquad[0];
2357 *tl++ = cookie.nfsuquad[1];
2358 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2359 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2360 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2361 *tl = txdr_unsigned(nmp->nm_rsize);
c1cf1e59 2362 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ));
984263bc
MD
2363 nfsm_postop_attr(vp, attrflag);
2364 if (error) {
2365 m_freem(mrep);
2366 goto nfsmout;
2367 }
2368 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2369 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2370 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2371 more_dirs = fxdr_unsigned(int, *tl);
2372
2373 /* loop thru the dir entries, doctoring them to 4bsd form */
2374 while (more_dirs && bigenough) {
2375 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2376 fileno = fxdr_hyper(tl);
2377 len = fxdr_unsigned(int, *(tl + 2));
2378 if (len <= 0 || len > NFS_MAXNAMLEN) {
2379 error = EBADRPC;
2380 m_freem(mrep);
2381 goto nfsmout;
2382 }
2383 tlen = nfsm_rndup(len);
2384 if (tlen == len)
2385 tlen += 4; /* To ensure null termination*/
2386 left = DIRBLKSIZ - blksiz;
2387 if ((tlen + DIRHDSIZ) > left) {
2388 dp->d_reclen += left;
2389 uiop->uio_iov->iov_base += left;
2390 uiop->uio_iov->iov_len -= left;
2391 uiop->uio_offset += left;
2392 uiop->uio_resid -= left;
2393 blksiz = 0;
2394 }
2395 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2396 bigenough = 0;
2397 if (bigenough) {
2398 dp = (struct dirent *)uiop->uio_iov->iov_base;
2399 dp->d_fileno = (int)fileno;
2400 dp->d_namlen = len;
2401 dp->d_reclen = tlen + DIRHDSIZ;
2402 dp->d_type = DT_UNKNOWN;
2403 blksiz += dp->d_reclen;
2404 if (blksiz == DIRBLKSIZ)
2405 blksiz = 0;
2406 uiop->uio_offset += DIRHDSIZ;
2407 uiop->uio_resid -= DIRHDSIZ;
2408 uiop->uio_iov->iov_base += DIRHDSIZ;
2409 uiop->uio_iov->iov_len -= DIRHDSIZ;
2410 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2411 cnp->cn_namelen = len;
2412 nfsm_mtouio(uiop, len);
2413 cp = uiop->uio_iov->iov_base;
2414 tlen -= len;
2415 *cp = '\0';
2416 uiop->uio_iov->iov_base += tlen;
2417 uiop->uio_iov->iov_len -= tlen;
2418 uiop->uio_offset += tlen;
2419 uiop->uio_resid -= tlen;
2420 } else
2421 nfsm_adv(nfsm_rndup(len));
2422 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2423 if (bigenough) {
2424 cookie.nfsuquad[0] = *tl++;
2425 cookie.nfsuquad[1] = *tl++;
2426 } else
2427 tl += 2;
2428
2429 /*
2430 * Since the attributes are before the file handle
2431 * (sigh), we must skip over the attributes and then
2432 * come back and get them.
2433 */
2434 attrflag = fxdr_unsigned(int, *tl);
2435 if (attrflag) {
2436 dpossav1 = dpos;
2437 mdsav1 = md;
2438 nfsm_adv(NFSX_V3FATTR);
2439 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2440 doit = fxdr_unsigned(int, *tl);
2441 if (doit) {
2442 nfsm_getfh(fhp, fhsize, 1);
2443 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2444 VREF(vp);
2445 newvp = vp;
2446 np = dnp;
2447 } else {
2448 error = nfs_nget(vp->v_mount, fhp,
2449 fhsize, &np);
2450 if (error)
2451 doit = 0;
2452 else
2453 newvp = NFSTOV(np);
2454 }
2455 }
2456 if (doit && bigenough) {
2457 dpossav2 = dpos;
2458 dpos = dpossav1;
2459 mdsav2 = md;
2460 md = mdsav1;
2461 nfsm_loadattr(newvp, (struct vattr *)0);
2462 dpos = dpossav2;
2463 md = mdsav2;
2464 dp->d_type =
2465 IFTODT(VTTOIF(np->n_vattr.va_type));
2466 ndp->ni_vp = newvp;
bc0c094e 2467 cache_enter(ndp->ni_dvp, NCPNULL, ndp->ni_vp, cnp);
984263bc
MD
2468 }
2469 } else {
2470 /* Just skip over the file handle */
2471 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2472 i = fxdr_unsigned(int, *tl);
2473 nfsm_adv(nfsm_rndup(i));
2474 }
2475 if (newvp != NULLVP) {
2476 if (newvp == vp)
2477 vrele(newvp);
2478 else
2479 vput(newvp);
2480 newvp = NULLVP;
2481 }
2482 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2483 more_dirs = fxdr_unsigned(int, *tl);
2484 }
2485 /*
2486 * If at end of rpc data, get the eof boolean
2487 */
2488 if (!more_dirs) {
2489 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2490 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2491 }
2492 m_freem(mrep);
2493 }
2494 /*
2495 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2496 * by increasing d_reclen for the last record.
2497 */
2498 if (blksiz > 0) {
2499 left = DIRBLKSIZ - blksiz;
2500 dp->d_reclen += left;
2501 uiop->uio_iov->iov_base += left;
2502 uiop->uio_iov->iov_len -= left;
2503 uiop->uio_offset += left;
2504 uiop->uio_resid -= left;
2505 }
2506
2507 /*
2508 * We are now either at the end of the directory or have filled the
2509 * block.
2510 */
2511 if (bigenough)
2512 dnp->n_direofoffset = uiop->uio_offset;
2513 else {
2514 if (uiop->uio_resid > 0)
2515 printf("EEK! readdirplusrpc resid > 0\n");
2516 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2517 *cookiep = cookie;
2518 }
2519nfsmout:
2520 if (newvp != NULLVP) {
2521 if (newvp == vp)
2522 vrele(newvp);
2523 else
2524 vput(newvp);
2525 newvp = NULLVP;
2526 }
2527 return (error);
2528}
2529
2530/*
2531 * Silly rename. To make the NFS filesystem that is stateless look a little
2532 * more like the "ufs" a remove of an active vnode is translated to a rename
2533 * to a funny looking filename that is removed by nfs_inactive on the
2534 * nfsnode. There is the potential for another process on a different client
2535 * to create the same funny name between the nfs_lookitup() fails and the
2536 * nfs_rename() completes, but...
2537 */
2538static int
2539nfs_sillyrename(dvp, vp, cnp)
2540 struct vnode *dvp, *vp;
2541 struct componentname *cnp;
2542{
40393ded 2543 struct sillyrename *sp;
984263bc
MD
2544 struct nfsnode *np;
2545 int error;
984263bc
MD
2546
2547 cache_purge(dvp);
2548 np = VTONFS(vp);
2549#ifndef DIAGNOSTIC
2550 if (vp->v_type == VDIR)
2551 panic("nfs: sillyrename dir");
2552#endif
2553 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2554 M_NFSREQ, M_WAITOK);
2555 sp->s_cred = crdup(cnp->cn_cred);
2556 sp->s_dvp = dvp;
2557 VREF(dvp);
2558
2559 /* Fudge together a funny name */
dadab5e9 2560 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td);
984263bc
MD
2561
2562 /* Try lookitups until we get one that isn't there */
2563 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
dadab5e9 2564 cnp->cn_td, (struct nfsnode **)0) == 0) {
984263bc
MD
2565 sp->s_name[4]++;
2566 if (sp->s_name[4] > 'z') {
2567 error = EINVAL;
2568 goto bad;
2569 }
2570 }
2571 error = nfs_renameit(dvp, cnp, sp);
2572 if (error)
2573 goto bad;
2574 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
dadab5e9 2575 cnp->cn_td, &np);
984263bc
MD
2576 np->n_sillyrename = sp;
2577 return (0);
2578bad:
2579 vrele(sp->s_dvp);
2580 crfree(sp->s_cred);
2581 free((caddr_t)sp, M_NFSREQ);
2582 return (error);
2583}
2584
2585/*
2586 * Look up a file name and optionally either update the file handle or
2587 * allocate an nfsnode, depending on the value of npp.
2588 * npp == NULL --> just do the lookup
2589 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2590 * handled too
2591 * *npp != NULL --> update the file handle in the vnode
2592 */
2593static int
dadab5e9 2594nfs_lookitup(dvp, name, len, cred, td, npp)
40393ded 2595 struct vnode *dvp;
984263bc
MD
2596 const char *name;
2597 int len;
2598 struct ucred *cred;
dadab5e9 2599 struct thread *td;
984263bc
MD
2600 struct nfsnode **npp;
2601{
40393ded
RG
2602 u_int32_t *tl;
2603 caddr_t cp;
2604 int32_t t1, t2;
984263bc
MD
2605 struct vnode *newvp = (struct vnode *)0;
2606 struct nfsnode *np, *dnp = VTONFS(dvp);
2607 caddr_t bpos, dpos, cp2;
2608 int error = 0, fhlen, attrflag;
2609 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2610 nfsfh_t *nfhp;
2611 int v3 = NFS_ISV3(dvp);
2612
2613 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2614 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2615 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2616 nfsm_fhtom(dvp, v3);
2617 nfsm_strtom(name, len, NFS_MAXNAMLEN);
dadab5e9 2618 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
984263bc
MD
2619 if (npp && !error) {
2620 nfsm_getfh(nfhp, fhlen, v3);
2621 if (*npp) {
2622 np = *npp;
2623 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2624 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2625 np->n_fhp = &np->n_fh;
2626 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2627 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2628 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2629 np->n_fhsize = fhlen;
2630 newvp = NFSTOV(np);
2631 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2632 VREF(dvp);
2633 newvp = dvp;
2634 } else {
2635 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2636 if (error) {
2637 m_freem(mrep);
2638 return (error);
2639 }
2640 newvp = NFSTOV(np);
2641 }
2642 if (v3) {
2643 nfsm_postop_attr(newvp, attrflag);
2644 if (!attrflag && *npp == NULL) {
2645 m_freem(mrep);
2646 if (newvp == dvp)
2647 vrele(newvp);
2648 else
2649 vput(newvp);
2650 return (ENOENT);
2651 }
2652 } else
2653 nfsm_loadattr(newvp, (struct vattr *)0);
2654 }
6b08710e
MD
2655 m_freem(mrep);
2656nfsmout:
984263bc
MD
2657 if (npp && *npp == NULL) {
2658 if (error) {
2659 if (newvp) {
2660 if (newvp == dvp)
2661 vrele(newvp);
2662 else
2663 vput(newvp);
2664 }
2665 } else
2666 *npp = np;
2667 }
2668 return (error);
2669}
2670
2671/*
2672 * Nfs Version 3 commit rpc
2673 */
2674int
3b568787 2675nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
984263bc 2676{
40393ded
RG
2677 caddr_t cp;
2678 u_int32_t *tl;
2679 int32_t t1, t2;
2680 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
984263bc
MD
2681 caddr_t bpos, dpos, cp2;
2682 int error = 0, wccflag = NFSV3_WCCRATTR;
2683 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2684
2685 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2686 return (0);
2687 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2688 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2689 nfsm_fhtom(vp, 1);
2690 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2691 txdr_hyper(offset, tl);
2692 tl += 2;
2693 *tl = txdr_unsigned(cnt);
c1cf1e59 2694 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE));
984263bc
MD
2695 nfsm_wcc_data(vp, wccflag);
2696 if (!error) {
2697 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2698 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2699 NFSX_V3WRITEVERF)) {
2700 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2701 NFSX_V3WRITEVERF);
2702 error = NFSERR_STALEWRITEVERF;
2703 }
2704 }
6b08710e
MD
2705 m_freem(mrep);
2706nfsmout:
984263bc
MD
2707 return (error);
2708}
2709
2710/*
2711 * Kludge City..
2712 * - make nfs_bmap() essentially a no-op that does no translation
2713 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2714 * (Maybe I could use the process's page mapping, but I was concerned that
2715 * Kernel Write might not be enabled and also figured copyout() would do
2716 * a lot more work than bcopy() and also it currently happens in the
2717 * context of the swapper process (2).
2718 */
2719static int
2720nfs_bmap(ap)
2721 struct vop_bmap_args /* {
2722 struct vnode *a_vp;
2723 daddr_t a_bn;
2724 struct vnode **a_vpp;
2725 daddr_t *a_bnp;
2726 int *a_runp;
2727 int *a_runb;
2728 } */ *ap;
2729{
40393ded 2730 struct vnode *vp = ap->a_vp;
984263bc
MD
2731
2732 if (ap->a_vpp != NULL)
2733 *ap->a_vpp = vp;
2734 if (ap->a_bnp != NULL)
2735 *ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize);
2736 if (ap->a_runp != NULL)
2737 *ap->a_runp = 0;
2738 if (ap->a_runb != NULL)
2739 *ap->a_runb = 0;
2740 return (0);
2741}
2742
2743/*
2744 * Strategy routine.
2745 * For async requests when nfsiod(s) are running, queue the request by
2746 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2747 * request.
2748 */
2749static int
2750nfs_strategy(ap)
2751 struct vop_strategy_args *ap;
2752{
40393ded 2753 struct buf *bp = ap->a_bp;
dadab5e9 2754 struct thread *td;
984263bc
MD
2755 int error = 0;
2756
2757 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2758 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
2759
2760 if (bp->b_flags & B_PHYS)
2761 panic("nfs physio");
2762
2763 if (bp->b_flags & B_ASYNC)
dadab5e9 2764 td = NULL;
984263bc 2765 else
dadab5e9 2766 td = curthread; /* XXX */
984263bc 2767
984263bc
MD
2768 /*
2769 * If the op is asynchronous and an i/o daemon is waiting
2770 * queue the request, wake it up and wait for completion
2771 * otherwise just do it ourselves.
2772 */
2773 if ((bp->b_flags & B_ASYNC) == 0 ||
3b568787
MD
2774 nfs_asyncio(bp, td))
2775 error = nfs_doio(bp, td);
984263bc
MD
2776 return (error);
2777}
2778
2779/*
2780 * Mmap a file
2781 *
2782 * NB Currently unsupported.
2783 */
2784/* ARGSUSED */
2785static int
2786nfs_mmap(ap)
2787 struct vop_mmap_args /* {
2788 struct vnode *a_vp;
2789 int a_fflags;
2790 struct ucred *a_cred;
dadab5e9 2791 struct thread *a_td;
984263bc
MD
2792 } */ *ap;
2793{
2794
2795 return (EINVAL);
2796}
2797
2798/*
2799 * fsync vnode op. Just call nfs_flush() with commit == 1.
2800 */
2801/* ARGSUSED */
2802static int
2803nfs_fsync(ap)
2804 struct vop_fsync_args /* {
2805 struct vnodeop_desc *a_desc;
2806 struct vnode * a_vp;
2807 struct ucred * a_cred;
2808 int a_waitfor;
dadab5e9 2809 struct thread * a_td;
984263bc
MD
2810 } */ *ap;
2811{
2812
3b568787 2813 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
984263bc
MD
2814}
2815
2816/*
2817 * Flush all the blocks associated with a vnode.
2818 * Walk through the buffer pool and push any dirty pages
2819 * associated with the vnode.
2820 */
2821static int
3b568787 2822nfs_flush(vp, waitfor, td, commit)
40393ded 2823 struct vnode *vp;
984263bc 2824 int waitfor;
dadab5e9 2825 struct thread *td;
984263bc
MD
2826 int commit;
2827{
40393ded
RG
2828 struct nfsnode *np = VTONFS(vp);
2829 struct buf *bp;
2830 int i;
984263bc
MD
2831 struct buf *nbp;
2832 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2833 int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2834 int passone = 1;
2835 u_quad_t off, endoff, toff;
984263bc
MD
2836 struct buf **bvec = NULL;
2837#ifndef NFS_COMMITBVECSIZ
2838#define NFS_COMMITBVECSIZ 20
2839#endif
2840 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2841 int bvecsize = 0, bveccount;
2842
2843 if (nmp->nm_flag & NFSMNT_INT)
2844 slpflag = PCATCH;
2845 if (!commit)
2846 passone = 0;
2847 /*
2848 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2849 * server, but nas not been committed to stable storage on the server
2850 * yet. On the first pass, the byte range is worked out and the commit
2851 * rpc is done. On the second pass, nfs_writebp() is called to do the
2852 * job.
2853 */
2854again:
2855 off = (u_quad_t)-1;
2856 endoff = 0;
2857 bvecpos = 0;
2858 if (NFS_ISV3(vp) && commit) {
2859 s = splbio();
2860 /*
2861 * Count up how many buffers waiting for a commit.
2862 */
2863 bveccount = 0;
2864 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2865 nbp = TAILQ_NEXT(bp, b_vnbufs);
2866 if (BUF_REFCNT(bp) == 0 &&
2867 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2868 == (B_DELWRI | B_NEEDCOMMIT))
2869 bveccount++;
2870 }
2871 /*
2872 * Allocate space to remember the list of bufs to commit. It is
2873 * important to use M_NOWAIT here to avoid a race with nfs_write.
2874 * If we can't get memory (for whatever reason), we will end up
2875 * committing the buffers one-by-one in the loop below.
2876 */
2877 if (bvec != NULL && bvec != bvec_on_stack)
2878 free(bvec, M_TEMP);
2879 if (bveccount > NFS_COMMITBVECSIZ) {
2880 bvec = (struct buf **)
2881 malloc(bveccount * sizeof(struct buf *),
2882 M_TEMP, M_NOWAIT);
2883 if (bvec == NULL) {
2884 bvec = bvec_on_stack;
2885 bvecsize = NFS_COMMITBVECSIZ;
2886 } else
2887 bvecsize = bveccount;
2888 } else {
2889 bvec = bvec_on_stack;
2890 bvecsize = NFS_COMMITBVECSIZ;
2891 }
2892 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2893 nbp = TAILQ_NEXT(bp, b_vnbufs);
2894 if (bvecpos >= bvecsize)
2895 break;
2896 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2897 (B_DELWRI | B_NEEDCOMMIT) ||
2898 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
2899 continue;
2900 bremfree(bp);
2901 /*
984263bc
MD
2902 * NOTE: we are not clearing B_DONE here, so we have
2903 * to do it later on in this routine if we intend to
2904 * initiate I/O on the bp.
2905 *
2906 * Note: to avoid loopback deadlocks, we do not
2907 * assign b_runningbufspace.
2908 */
984263bc
MD
2909 bp->b_flags |= B_WRITEINPROG;
2910 vfs_busy_pages(bp, 1);
2911
2912 /*
2913 * bp is protected by being locked, but nbp is not
2914 * and vfs_busy_pages() may sleep. We have to
2915 * recalculate nbp.
2916 */
2917 nbp = TAILQ_NEXT(bp, b_vnbufs);
2918
2919 /*
2920 * A list of these buffers is kept so that the
2921 * second loop knows which buffers have actually
2922 * been committed. This is necessary, since there
2923 * may be a race between the commit rpc and new
2924 * uncommitted writes on the file.
2925 */
2926 bvec[bvecpos++] = bp;
2927 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2928 bp->b_dirtyoff;
2929 if (toff < off)
2930 off = toff;
2931 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2932 if (toff > endoff)
2933 endoff = toff;
2934 }
2935 splx(s);
2936 }
2937 if (bvecpos > 0) {
2938 /*
3b568787
MD
2939 * Commit data on the server, as required. Note that
2940 * nfs_commit will use the vnode's cred for the commit.
984263bc 2941 */
3b568787 2942 retv = nfs_commit(vp, off, (int)(endoff - off), td);
984263bc
MD
2943
2944 if (retv == NFSERR_STALEWRITEVERF)
2945 nfs_clearcommit(vp->v_mount);
2946
2947 /*
2948 * Now, either mark the blocks I/O done or mark the
2949 * blocks dirty, depending on whether the commit
2950 * succeeded.
2951 */
2952 for (i = 0; i < bvecpos; i++) {
2953 bp = bvec[i];
2954 bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK);
2955 if (retv) {
2956 /*
2957 * Error, leave B_DELWRI intact
2958 */
2959 vfs_unbusy_pages(bp);
2960 brelse(bp);
2961 } else {
2962 /*
2963 * Success, remove B_DELWRI ( bundirty() ).
2964 *
2965 * b_dirtyoff/b_dirtyend seem to be NFS
2966 * specific. We should probably move that
2967 * into bundirty(). XXX
2968 */
2969 s = splbio();
2970 vp->v_numoutput++;
2971 bp->b_flags |= B_ASYNC;
2972 bundirty(bp);
2973 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
2974 bp->b_dirtyoff = bp->b_dirtyend = 0;
2975 splx(s);
2976 biodone(bp);
2977 }
2978 }
2979 }
2980
2981 /*
2982 * Start/do any write(s) that are required.
2983 */
2984loop:
2985 s = splbio();
2986 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2987 nbp = TAILQ_NEXT(bp, b_vnbufs);
2988 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
2989 if (waitfor != MNT_WAIT || passone)
2990 continue;
2991 error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL,
2992 "nfsfsync", slpflag, slptimeo);
2993 splx(s);
2994 if (error == 0)
2995 panic("nfs_fsync: inconsistent lock");
2996 if (error == ENOLCK)
2997 goto loop;
dadab5e9 2998 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
984263bc
MD
2999 error = EINTR;
3000 goto done;
3001 }
3002 if (slpflag == PCATCH) {
3003 slpflag = 0;
3004 slptimeo = 2 * hz;
3005 }
3006 goto loop;
3007 }
3008 if ((bp->b_flags & B_DELWRI) == 0)
3009 panic("nfs_fsync: not dirty");
3010 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3011 BUF_UNLOCK(bp);
3012 continue;
3013 }
3014 bremfree(bp);
3015 if (passone || !commit)
3016 bp->b_flags |= B_ASYNC;
3017 else
3018 bp->b_flags |= B_ASYNC | B_WRITEINPROG;
3019 splx(s);
3020 VOP_BWRITE(bp->b_vp, bp);
3021 goto loop;
3022 }
3023 splx(s);
3024 if (passone) {
3025 passone = 0;
3026 goto again;
3027 }
3028 if (waitfor == MNT_WAIT) {
3029 while (vp->v_numoutput) {
3030 vp->v_flag |= VBWAIT;
3031 error = tsleep((caddr_t)&vp->v_numoutput,
377d4740 3032 slpflag, "nfsfsync", slptimeo);
984263bc 3033 if (error) {
dadab5e9 3034 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
984263bc
MD
3035 error = EINTR;
3036 goto done;
3037 }
3038 if (slpflag == PCATCH) {
3039 slpflag = 0;
3040 slptimeo = 2 * hz;
3041 }
3042 }
3043 }
3044 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) {
3045 goto loop;
3046 }
3047 }
3048 if (np->n_flag & NWRITEERR) {
3049 error = np->n_error;
3050 np->n_flag &= ~NWRITEERR;
3051 }
3052done:
3053 if (bvec != NULL && bvec != bvec_on_stack)
3054 free(bvec, M_TEMP);
3055 return (error);
3056}
3057
3058/*
3059 * NFS advisory byte-level locks.
3060 * Currently unsupported.
3061 */
3062static int
3063nfs_advlock(ap)
3064 struct vop_advlock_args /* {
3065 struct vnode *a_vp;
3066 caddr_t a_id;
3067 int a_op;
3068 struct flock *a_fl;
3069 int a_flags;
3070 } */ *ap;
3071{
40393ded 3072 struct nfsnode *np = VTONFS(ap->a_vp);
984263bc
MD
3073
3074 /*
3075 * The following kludge is to allow diskless support to work
3076 * until a real NFS lockd is implemented. Basically, just pretend
3077 * that this is a local lock.
3078 */
3079 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3080}
3081
3082/*
3083 * Print out the contents of an nfsnode.
3084 */
3085static int
3086nfs_print(ap)
3087 struct vop_print_args /* {
3088 struct vnode *a_vp;
3089 } */ *ap;
3090{
40393ded
RG
3091 struct vnode *vp = ap->a_vp;
3092 struct nfsnode *np = VTONFS(vp);
984263bc
MD
3093
3094 printf("tag VT_NFS, fileid %ld fsid 0x%x",
3095 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3096 if (vp->v_type == VFIFO)
3097 fifo_printinfo(vp);
3098 printf("\n");
3099 return (0);
3100}
3101
3102/*
3103 * Just call nfs_writebp() with the force argument set to 1.
3104 *
3105 * NOTE: B_DONE may or may not be set in a_bp on call.
3106 */
3107static int
3108nfs_bwrite(ap)
3109 struct vop_bwrite_args /* {
3110 struct vnode *a_bp;
3111 } */ *ap;
3112{
dadab5e9 3113 return (nfs_writebp(ap->a_bp, 1, curthread));
984263bc
MD
3114}
3115
3116/*
3117 * This is a clone of vn_bwrite(), except that B_WRITEINPROG isn't set unless
3118 * the force flag is one and it also handles the B_NEEDCOMMIT flag. We set
3119 * B_CACHE if this is a VMIO buffer.
3120 */
3121int
dadab5e9 3122nfs_writebp(bp, force, td)
40393ded 3123 struct buf *bp;
984263bc 3124 int force;
dadab5e9 3125 struct thread *td;
984263bc
MD
3126{
3127 int s;
3128 int oldflags = bp->b_flags;
3129#if 0
3130 int retv = 1;
3131 off_t off;
3132#endif
3133
3134 if (BUF_REFCNT(bp) == 0)
3135 panic("bwrite: buffer is not locked???");
3136
3137 if (bp->b_flags & B_INVAL) {
3138 brelse(bp);
3139 return(0);
3140 }
3141
3142 bp->b_flags |= B_CACHE;
3143
3144 /*
3145 * Undirty the bp. We will redirty it later if the I/O fails.
3146 */
3147
3148 s = splbio();
3149 bundirty(bp);
3150 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3151
3152 bp->b_vp->v_numoutput++;
984263bc
MD
3153 splx(s);
3154
3155 /*
3156 * Note: to avoid loopback deadlocks, we do not
3157 * assign b_runningbufspace.
3158 */
3159 vfs_busy_pages(bp, 1);
3160
3161 if (force)
3162 bp->b_flags |= B_WRITEINPROG;
3163 BUF_KERNPROC(bp);
3164 VOP_STRATEGY(bp->b_vp, bp);
3165
3166 if( (oldflags & B_ASYNC) == 0) {
3167 int rtval = biowait(bp);
3168
3169 if (oldflags & B_DELWRI) {
3170 s = splbio();
3171 reassignbuf(bp, bp->b_vp);
3172 splx(s);
3173 }
3174
3175 brelse(bp);
3176 return (rtval);
3177 }
3178
3179 return (0);
3180}
3181
3182/*
3183 * nfs special file access vnode op.
3184 * Essentially just get vattr and then imitate iaccess() since the device is
3185 * local to the client.
3186 */
3187static int
3188nfsspec_access(ap)
3189 struct vop_access_args /* {
3190 struct vnode *a_vp;
3191 int a_mode;
3192 struct ucred *a_cred;
dadab5e9 3193 struct thread *a_td;
984263bc
MD
3194 } */ *ap;
3195{
40393ded
RG
3196 struct vattr *vap;
3197 gid_t *gp;
3198 struct ucred *cred = ap->a_cred;
984263bc
MD
3199 struct vnode *vp = ap->a_vp;
3200 mode_t mode = ap->a_mode;
3201 struct vattr vattr;
40393ded 3202 int i;
984263bc
MD
3203 int error;
3204
3205 /*
3206 * Disallow write attempts on filesystems mounted read-only;
3207 * unless the file is a socket, fifo, or a block or character
3208 * device resident on the filesystem.
3209 */
3210 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3211 switch (vp->v_type) {
3212 case VREG:
3213 case VDIR:
3214 case VLNK:
3215 return (EROFS);
3216 default:
3217 break;
3218 }
3219 }
3220 /*
3221 * If you're the super-user,
3222 * you always get access.
3223 */
3224 if (cred->cr_uid == 0)
3225 return (0);
3226 vap = &vattr;
3b568787 3227 error = VOP_GETATTR(vp, vap, ap->a_td);
984263bc
MD
3228 if (error)
3229 return (error);
3230 /*
3231 * Access check is based on only one of owner, group, public.
3232 * If not owner, then check group. If not a member of the
3233 * group, then check public access.
3234 */
3235 if (cred->cr_uid != vap->va_uid) {
3236 mode >>= 3;
3237 gp = cred->cr_groups;
3238 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3239 if (vap->va_gid == *gp)
3240 goto found;
3241 mode >>= 3;
3242found:
3243 ;
3244 }
3245 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3246 return (error);
3247}
3248
3249/*
3250 * Read wrapper for special devices.
3251 */
3252static int
3253nfsspec_read(ap)
3254 struct vop_read_args /* {
3255 struct vnode *a_vp;
3256 struct uio *a_uio;
3257 int a_ioflag;
3258 struct ucred *a_cred;
3259 } */ *ap;
3260{
40393ded 3261 struct nfsnode *np = VTONFS(ap->a_vp);
984263bc
MD
3262
3263 /*
3264 * Set access flag.
3265 */
3266 np->n_flag |= NACC;
3267 getnanotime(&np->n_atim);
3268 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3269}
3270
3271/*
3272 * Write wrapper for special devices.
3273 */
3274static int
3275nfsspec_write(ap)
3276 struct vop_write_args /* {
3277 struct vnode *a_vp;
3278 struct uio *a_uio;
3279 int a_ioflag;
3280 struct ucred *a_cred;
3281 } */ *ap;
3282{
40393ded 3283 struct nfsnode *np = VTONFS(ap->a_vp);
984263bc
MD
3284
3285 /*
3286 * Set update flag.
3287 */
3288 np->n_flag |= NUPD;
3289 getnanotime(&np->n_mtim);
3290 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3291}
3292
3293/*
3294 * Close wrapper for special devices.
3295 *
3296 * Update the times on the nfsnode then do device close.
3297 */
3298static int
3299nfsspec_close(ap)
3300 struct vop_close_args /* {
3301 struct vnode *a_vp;
3302 int a_fflag;
3303 struct ucred *a_cred;
dadab5e9 3304 struct thread *a_td;
984263bc
MD
3305 } */ *ap;
3306{
40393ded
RG
3307 struct vnode *vp = ap->a_vp;
3308 struct nfsnode *np = VTONFS(vp);
984263bc
MD
3309 struct vattr vattr;
3310
3311 if (np->n_flag & (NACC | NUPD)) {
3312 np->n_flag |= NCHG;
3313 if (vp->v_usecount == 1 &&
3314 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3315 VATTR_NULL(&vattr);
3316 if (np->n_flag & NACC)
3317 vattr.va_atime = np->n_atim;
3318 if (np->n_flag & NUPD)
3319 vattr.va_mtime = np->n_mtim;
c1cf1e59 3320 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
984263bc
MD
3321 }
3322 }
3323 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3324}
3325
3326/*
3327 * Read wrapper for fifos.
3328 */
3329static int
3330nfsfifo_read(ap)
3331 struct vop_read_args /* {
3332 struct vnode *a_vp;
3333 struct uio *a_uio;
3334 int a_ioflag;
3335 struct ucred *a_cred;
3336 } */ *ap;
3337{
40393ded 3338 struct nfsnode *np = VTONFS(ap->a_vp);
984263bc
MD
3339
3340 /*
3341 * Set access flag.
3342 */
3343 np->n_flag |= NACC;
3344 getnanotime(&np->n_atim);
3345 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3346}
3347
3348/*
3349 * Write wrapper for fifos.
3350 */
3351static int
3352nfsfifo_write(ap)
3353 struct vop_write_args /* {
3354 struct vnode *a_vp;
3355 struct uio *a_uio;
3356 int a_ioflag;
3357 struct ucred *a_cred;
3358 } */ *ap;
3359{
40393ded 3360 struct nfsnode *np = VTONFS(ap->a_vp);
984263bc
MD
3361
3362 /*
3363 * Set update flag.
3364 */
3365 np->n_flag |= NUPD;
3366 getnanotime(&np->n_mtim);
3367 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3368}
3369
3370/*
3371 * Close wrapper for fifos.
3372 *
3373 * Update the times on the nfsnode then do fifo close.
3374 */
3375static int
3376nfsfifo_close(ap)
3377 struct vop_close_args /* {
3378 struct vnode *a_vp;
3379 int a_fflag;
dadab5e9 3380 struct thread *a_td;
984263bc
MD
3381 } */ *ap;
3382{
40393ded
RG
3383 struct vnode *vp = ap->a_vp;
3384 struct nfsnode *np = VTONFS(vp);
984263bc
MD
3385 struct vattr vattr;
3386 struct timespec ts;
3387
3388 if (np->n_flag & (NACC | NUPD)) {
3389 getnanotime(&ts);
3390 if (np->n_flag & NACC)
3391 np->n_atim = ts;
3392 if (np->n_flag & NUPD)
3393 np->n_mtim = ts;
3394 np->n_flag |= NCHG;
3395 if (vp->v_usecount == 1 &&
3396 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3397 VATTR_NULL(&vattr);
3398 if (np->n_flag & NACC)
3399 vattr.va_atime = np->n_atim;
3400 if (np->n_flag & NUPD)
3401 vattr.va_mtime = np->n_mtim;
c1cf1e59 3402 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
984263bc
MD
3403 }
3404 }
3405 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3406}
3407