Kernel - Fix issues with vnode_pager_setsize() (effects NFS only)
[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 $
bb7bf7b2 38 * $DragonFly: src/sys/vfs/nfs/nfs_vnops.c,v 1.80 2008/10/18 01:13:54 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>
fad57d0e 58#include <sys/nlookup.h>
984263bc
MD
59#include <sys/socket.h>
60#include <sys/vnode.h>
61#include <sys/dirent.h>
62#include <sys/fcntl.h>
63#include <sys/lockf.h>
64#include <sys/stat.h>
65#include <sys/sysctl.h>
66#include <sys/conf.h>
67
68#include <vm/vm.h>
69#include <vm/vm_extern.h>
70#include <vm/vm_zone.h>
71
3020e3be
MD
72#include <sys/buf2.h>
73
1f2de5d4 74#include <vfs/fifofs/fifo.h>
01f31ab3
JS
75#include <vfs/ufs/dir.h>
76
77#undef DIRBLKSIZ
984263bc 78
1f2de5d4
MD
79#include "rpcv2.h"
80#include "nfsproto.h"
81#include "nfs.h"
1f2de5d4 82#include "nfsmount.h"
c1cf1e59 83#include "nfsnode.h"
1f2de5d4
MD
84#include "xdr_subs.h"
85#include "nfsm_subs.h"
984263bc
MD
86
87#include <net/if.h>
88#include <netinet/in.h>
89#include <netinet/in_var.h>
90
165dba55
DR
91#include <sys/thread2.h>
92
984263bc
MD
93/* Defs */
94#define TRUE 1
95#define FALSE 0
96
a6ee311a
RG
97static int nfsfifo_read (struct vop_read_args *);
98static int nfsfifo_write (struct vop_write_args *);
a6ee311a 99static int nfsfifo_close (struct vop_close_args *);
984263bc 100#define nfs_poll vop_nopoll
a6ee311a 101static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
e62afb5f
MD
102static int nfs_lookup (struct vop_old_lookup_args *);
103static int nfs_create (struct vop_old_create_args *);
104static int nfs_mknod (struct vop_old_mknod_args *);
a6ee311a
RG
105static int nfs_open (struct vop_open_args *);
106static int nfs_close (struct vop_close_args *);
107static int nfs_access (struct vop_access_args *);
108static int nfs_getattr (struct vop_getattr_args *);
109static int nfs_setattr (struct vop_setattr_args *);
110static int nfs_read (struct vop_read_args *);
111static int nfs_mmap (struct vop_mmap_args *);
112static int nfs_fsync (struct vop_fsync_args *);
e62afb5f
MD
113static int nfs_remove (struct vop_old_remove_args *);
114static int nfs_link (struct vop_old_link_args *);
115static int nfs_rename (struct vop_old_rename_args *);
116static int nfs_mkdir (struct vop_old_mkdir_args *);
117static int nfs_rmdir (struct vop_old_rmdir_args *);
118static int nfs_symlink (struct vop_old_symlink_args *);
a6ee311a
RG
119static int nfs_readdir (struct vop_readdir_args *);
120static int nfs_bmap (struct vop_bmap_args *);
121static int nfs_strategy (struct vop_strategy_args *);
122static int nfs_lookitup (struct vnode *, const char *, int,
123 struct ucred *, struct thread *, struct nfsnode **);
124static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
8be7edad 125static int nfs_laccess (struct vop_access_args *);
a6ee311a
RG
126static int nfs_readlink (struct vop_readlink_args *);
127static int nfs_print (struct vop_print_args *);
128static int nfs_advlock (struct vop_advlock_args *);
fad57d0e
MD
129
130static int nfs_nresolve (struct vop_nresolve_args *);
984263bc
MD
131/*
132 * Global vfs data structures for nfs
133 */
66a1ddf5
MD
134struct vop_ops nfsv2_vnode_vops = {
135 .vop_default = vop_defaultop,
136 .vop_access = nfs_access,
137 .vop_advlock = nfs_advlock,
138 .vop_bmap = nfs_bmap,
139 .vop_close = nfs_close,
140 .vop_old_create = nfs_create,
141 .vop_fsync = nfs_fsync,
142 .vop_getattr = nfs_getattr,
143 .vop_getpages = nfs_getpages,
144 .vop_putpages = nfs_putpages,
145 .vop_inactive = nfs_inactive,
66a1ddf5 146 .vop_old_link = nfs_link,
66a1ddf5
MD
147 .vop_old_lookup = nfs_lookup,
148 .vop_old_mkdir = nfs_mkdir,
149 .vop_old_mknod = nfs_mknod,
150 .vop_mmap = nfs_mmap,
151 .vop_open = nfs_open,
152 .vop_poll = nfs_poll,
153 .vop_print = nfs_print,
154 .vop_read = nfs_read,
155 .vop_readdir = nfs_readdir,
156 .vop_readlink = nfs_readlink,
157 .vop_reclaim = nfs_reclaim,
158 .vop_old_remove = nfs_remove,
159 .vop_old_rename = nfs_rename,
160 .vop_old_rmdir = nfs_rmdir,
161 .vop_setattr = nfs_setattr,
162 .vop_strategy = nfs_strategy,
163 .vop_old_symlink = nfs_symlink,
66a1ddf5
MD
164 .vop_write = nfs_write,
165 .vop_nresolve = nfs_nresolve
984263bc 166};
984263bc
MD
167
168/*
169 * Special device vnode ops
170 */
66a1ddf5 171struct vop_ops nfsv2_spec_vops = {
8be7edad
MD
172 .vop_default = vop_defaultop,
173 .vop_access = nfs_laccess,
174 .vop_close = nfs_close,
66a1ddf5
MD
175 .vop_fsync = nfs_fsync,
176 .vop_getattr = nfs_getattr,
177 .vop_inactive = nfs_inactive,
66a1ddf5 178 .vop_print = nfs_print,
8be7edad 179 .vop_read = vop_stdnoread,
66a1ddf5
MD
180 .vop_reclaim = nfs_reclaim,
181 .vop_setattr = nfs_setattr,
8be7edad 182 .vop_write = vop_stdnowrite
984263bc 183};
984263bc 184
66a1ddf5
MD
185struct vop_ops nfsv2_fifo_vops = {
186 .vop_default = fifo_vnoperate,
8be7edad 187 .vop_access = nfs_laccess,
66a1ddf5
MD
188 .vop_close = nfsfifo_close,
189 .vop_fsync = nfs_fsync,
190 .vop_getattr = nfs_getattr,
191 .vop_inactive = nfs_inactive,
66a1ddf5
MD
192 .vop_print = nfs_print,
193 .vop_read = nfsfifo_read,
194 .vop_reclaim = nfs_reclaim,
195 .vop_setattr = nfs_setattr,
66a1ddf5 196 .vop_write = nfsfifo_write
984263bc 197};
984263bc 198
a6ee311a 199static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
984263bc 200 struct componentname *cnp,
a6ee311a
RG
201 struct vattr *vap);
202static int nfs_removerpc (struct vnode *dvp, const char *name,
984263bc 203 int namelen,
a6ee311a
RG
204 struct ucred *cred, struct thread *td);
205static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
984263bc
MD
206 int fnamelen, struct vnode *tdvp,
207 const char *tnameptr, int tnamelen,
a6ee311a
RG
208 struct ucred *cred, struct thread *td);
209static int nfs_renameit (struct vnode *sdvp,
984263bc 210 struct componentname *scnp,
a6ee311a 211 struct sillyrename *sp);
984263bc 212
984263bc
MD
213SYSCTL_DECL(_vfs_nfs);
214
a482a28a
MD
215static int nfs_flush_on_rename = 1;
216SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
217 &nfs_flush_on_rename, 0, "flush fvp prior to rename");
bb7bf7b2
MD
218static int nfs_flush_on_hlink = 0;
219SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
220 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
a482a28a 221
97100839 222static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
984263bc
MD
223SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
224 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
225
4d17b298
MD
226static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
227SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
a15b4eef
MD
228 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
229
230static int nfspos_cache_timeout = NFS_MINATTRTIMO;
231SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
232 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
4d17b298 233
984263bc
MD
234static int nfsv3_commit_on_close = 0;
235SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
236 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
237#if 0
238SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
239 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
240
241SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
242 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
243#endif
244
245#define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
246 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
247 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
248static int
dadab5e9 249nfs3_access_otw(struct vnode *vp, int wmode,
e851b29e 250 struct thread *td, struct ucred *cred)
984263bc 251{
42edf14f
MD
252 struct nfsnode *np = VTONFS(vp);
253 int attrflag;
254 int error = 0;
984263bc 255 u_int32_t *tl;
984263bc 256 u_int32_t rmode;
42edf14f
MD
257 struct nfsm_info info;
258
259 info.mrep = NULL;
260 info.v3 = 1;
984263bc
MD
261
262 nfsstats.rpccnt[NFSPROC_ACCESS]++;
42edf14f
MD
263 nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
264 NFSX_FH(info.v3) + NFSX_UNSIGNED);
265 ERROROUT(nfsm_fhtom(&info, vp));
266 tl = nfsm_build(&info, NFSX_UNSIGNED);
984263bc 267 *tl = txdr_unsigned(wmode);
42edf14f
MD
268 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
269 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
270 if (error == 0) {
271 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
984263bc
MD
272 rmode = fxdr_unsigned(u_int32_t, *tl);
273 np->n_mode = rmode;
274 np->n_modeuid = cred->cr_uid;
3a6f9faf 275 np->n_modestamp = mycpu->gd_time_seconds;
984263bc 276 }
42edf14f
MD
277 m_freem(info.mrep);
278 info.mrep = NULL;
6b08710e 279nfsmout:
984263bc
MD
280 return error;
281}
282
283/*
284 * nfs access vnode op.
285 * For nfs version 2, just return ok. File accesses may fail later.
286 * For nfs version 3, use the access rpc to check accessibility. If file modes
287 * are changed on the server, accesses might still fail later.
e851b29e 288 *
b478fdce 289 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
984263bc
MD
290 */
291static int
e851b29e 292nfs_access(struct vop_access_args *ap)
984263bc 293{
40393ded 294 struct vnode *vp = ap->a_vp;
87de5057 295 thread_t td = curthread;
984263bc
MD
296 int error = 0;
297 u_int32_t mode, wmode;
984263bc 298 struct nfsnode *np = VTONFS(vp);
42edf14f 299 int v3 = NFS_ISV3(vp);
984263bc
MD
300
301 /*
302 * Disallow write attempts on filesystems mounted read-only;
303 * unless the file is a socket, fifo, or a block or character
304 * device resident on the filesystem.
305 */
306 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
307 switch (vp->v_type) {
308 case VREG:
309 case VDIR:
310 case VLNK:
311 return (EROFS);
312 default:
313 break;
314 }
315 }
316 /*
317 * For nfs v3, check to see if we have done this recently, and if
318 * so return our cached result instead of making an ACCESS call.
319 * If not, do an access rpc, otherwise you are stuck emulating
320 * ufs_access() locally using the vattr. This may not be correct,
321 * since the server may apply other access criteria such as
322 * client uid-->server uid mapping that we do not know about.
323 */
324 if (v3) {
325 if (ap->a_mode & VREAD)
326 mode = NFSV3ACCESS_READ;
327 else
328 mode = 0;
329 if (vp->v_type != VDIR) {
330 if (ap->a_mode & VWRITE)
331 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
332 if (ap->a_mode & VEXEC)
333 mode |= NFSV3ACCESS_EXECUTE;
334 } else {
335 if (ap->a_mode & VWRITE)
336 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
337 NFSV3ACCESS_DELETE);
338 if (ap->a_mode & VEXEC)
339 mode |= NFSV3ACCESS_LOOKUP;
340 }
341 /* XXX safety belt, only make blanket request if caching */
342 if (nfsaccess_cache_timeout > 0) {
343 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
344 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
345 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
346 } else {
347 wmode = mode;
348 }
349
350 /*
351 * Does our cached result allow us to give a definite yes to
352 * this request?
353 */
fad57d0e
MD
354 if (np->n_modestamp &&
355 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
356 (ap->a_cred->cr_uid == np->n_modeuid) &&
357 ((np->n_mode & mode) == mode)) {
984263bc
MD
358 nfsstats.accesscache_hits++;
359 } else {
360 /*
361 * Either a no, or a don't know. Go to the wire.
362 */
363 nfsstats.accesscache_misses++;
87de5057 364 error = nfs3_access_otw(vp, wmode, td, ap->a_cred);
984263bc
MD
365 if (!error) {
366 if ((np->n_mode & mode) != mode) {
367 error = EACCES;
368 }
369 }
370 }
984263bc 371 } else {
8be7edad 372 if ((error = nfs_laccess(ap)) != 0)
984263bc
MD
373 return (error);
374
375 /*
376 * Attempt to prevent a mapped root from accessing a file
377 * which it shouldn't. We try to read a byte from the file
378 * if the user is root and the file is not zero length.
8be7edad 379 * After calling nfs_laccess, we should have the correct
984263bc
MD
380 * file size cached.
381 */
382 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
383 && VTONFS(vp)->n_size > 0) {
384 struct iovec aiov;
385 struct uio auio;
386 char buf[1];
387
388 aiov.iov_base = buf;
389 aiov.iov_len = 1;
390 auio.uio_iov = &aiov;
391 auio.uio_iovcnt = 1;
392 auio.uio_offset = 0;
393 auio.uio_resid = 1;
394 auio.uio_segflg = UIO_SYSSPACE;
395 auio.uio_rw = UIO_READ;
87de5057 396 auio.uio_td = td;
984263bc 397
c1cf1e59 398 if (vp->v_type == VREG) {
edb90c22 399 error = nfs_readrpc_uio(vp, &auio);
c1cf1e59 400 } else if (vp->v_type == VDIR) {
984263bc 401 char* bp;
efda3bd0 402 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
984263bc
MD
403 aiov.iov_base = bp;
404 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
cc7d050e 405 error = nfs_readdirrpc_uio(vp, &auio);
efda3bd0 406 kfree(bp, M_TEMP);
c1cf1e59 407 } else if (vp->v_type == VLNK) {
cc7d050e 408 error = nfs_readlinkrpc_uio(vp, &auio);
c1cf1e59 409 } else {
984263bc 410 error = EACCES;
c1cf1e59 411 }
984263bc 412 }
984263bc 413 }
c1cf1e59
MD
414 /*
415 * [re]record creds for reading and/or writing if access
09b1ee9b
MD
416 * was granted. Assume the NFS server will grant read access
417 * for execute requests.
c1cf1e59
MD
418 */
419 if (error == 0) {
09b1ee9b 420 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) {
c1cf1e59
MD
421 crhold(ap->a_cred);
422 if (np->n_rucred)
423 crfree(np->n_rucred);
424 np->n_rucred = ap->a_cred;
425 }
426 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
427 crhold(ap->a_cred);
428 if (np->n_wucred)
429 crfree(np->n_wucred);
430 np->n_wucred = ap->a_cred;
431 }
432 }
433 return(error);
984263bc
MD
434}
435
436/*
437 * nfs open vnode op
438 * Check to see if the type is ok
439 * and that deletion is not in progress.
440 * For paged in text files, you will need to flush the page cache
441 * if consistency is lost.
e851b29e 442 *
b478fdce
SW
443 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
444 * struct file *a_fp)
984263bc
MD
445 */
446/* ARGSUSED */
447static int
e851b29e 448nfs_open(struct vop_open_args *ap)
984263bc 449{
40393ded 450 struct vnode *vp = ap->a_vp;
984263bc 451 struct nfsnode *np = VTONFS(vp);
984263bc
MD
452 struct vattr vattr;
453 int error;
454
455 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
456#ifdef DIAGNOSTIC
086c1d7e 457 kprintf("open eacces vtyp=%d\n",vp->v_type);
984263bc 458#endif
ca3a2b2f 459 return (EOPNOTSUPP);
984263bc 460 }
5a9187cb 461
ed94161f
MD
462 /*
463 * Save valid creds for reading and writing for later RPCs.
464 */
465 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
466 crhold(ap->a_cred);
467 if (np->n_rucred)
468 crfree(np->n_rucred);
469 np->n_rucred = ap->a_cred;
470 }
471 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
472 crhold(ap->a_cred);
473 if (np->n_wucred)
474 crfree(np->n_wucred);
475 np->n_wucred = ap->a_cred;
476 }
477
984263bc 478 /*
5a9187cb
MD
479 * Clear the attribute cache only if opening with write access. It
480 * is unclear if we should do this at all here, but we certainly
481 * should not clear the cache unconditionally simply because a file
482 * is being opened.
984263bc 483 */
5a9187cb
MD
484 if (ap->a_mode & FWRITE)
485 np->n_attrstamp = 0;
486
e07fef60
MD
487 /*
488 * For normal NFS, reconcile changes made locally verses
489 * changes made remotely. Note that VOP_GETATTR only goes
490 * to the wire if the cached attribute has timed out or been
491 * cleared.
492 *
493 * If local modifications have been made clear the attribute
494 * cache to force an attribute and modified time check. If
495 * GETATTR detects that the file has been changed by someone
496 * other then us it will set NRMODIFIED.
497 *
498 * If we are opening a directory and local changes have been
499 * made we have to invalidate the cache in order to ensure
500 * that we get the most up-to-date information from the
501 * server. XXX
502 */
503 if (np->n_flag & NLMODIFIED) {
504 np->n_attrstamp = 0;
505 if (vp->v_type == VDIR) {
87de5057 506 error = nfs_vinvalbuf(vp, V_SAVE, 1);
5a9187cb
MD
507 if (error == EINTR)
508 return (error);
e07fef60 509 nfs_invaldir(vp);
5a9187cb 510 }
984263bc 511 }
87de5057 512 error = VOP_GETATTR(vp, &vattr);
e07fef60
MD
513 if (error)
514 return (error);
515 if (np->n_flag & NRMODIFIED) {
516 if (vp->v_type == VDIR)
517 nfs_invaldir(vp);
87de5057 518 error = nfs_vinvalbuf(vp, V_SAVE, 1);
e07fef60
MD
519 if (error == EINTR)
520 return (error);
521 np->n_flag &= ~NRMODIFIED;
522 }
fad57d0e 523
8ddc6004 524 return (vop_stdopen(ap));
984263bc
MD
525}
526
527/*
528 * nfs close vnode op
529 * What an NFS client should do upon close after writing is a debatable issue.
530 * Most NFS clients push delayed writes to the server upon close, basically for
531 * two reasons:
532 * 1 - So that any write errors may be reported back to the client process
533 * doing the close system call. By far the two most likely errors are
534 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
535 * 2 - To put a worst case upper bound on cache inconsistency between
536 * multiple clients for the file.
537 * There is also a consistency problem for Version 2 of the protocol w.r.t.
538 * not being able to tell if other clients are writing a file concurrently,
539 * since there is no way of knowing if the changed modify time in the reply
540 * is only due to the write for this client.
541 * (NFS Version 3 provides weak cache consistency data in the reply that
542 * should be sufficient to detect and handle this case.)
543 *
544 * The current code does the following:
545 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
546 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
547 * or commit them (this satisfies 1 and 2 except for the
548 * case where the server crashes after this close but
549 * before the commit RPC, which is felt to be "good
550 * enough". Changing the last argument to nfs_flush() to
551 * a 1 would force a commit operation, if it is felt a
552 * commit is necessary now.
553 * for NQNFS - do nothing now, since 2 is dealt with via leases and
554 * 1 should be dealt with via an fsync() system call for
555 * cases where write errors are important.
e851b29e 556 *
b478fdce 557 * nfs_close(struct vnode *a_vp, int a_fflag)
984263bc
MD
558 */
559/* ARGSUSED */
560static int
e851b29e 561nfs_close(struct vop_close_args *ap)
984263bc 562{
40393ded
RG
563 struct vnode *vp = ap->a_vp;
564 struct nfsnode *np = VTONFS(vp);
984263bc 565 int error = 0;
87de5057 566 thread_t td = curthread;
984263bc
MD
567
568 if (vp->v_type == VREG) {
e07fef60 569 if (np->n_flag & NLMODIFIED) {
984263bc
MD
570 if (NFS_ISV3(vp)) {
571 /*
572 * Under NFSv3 we have dirty buffers to dispose of. We
573 * must flush them to the NFS server. We have the option
574 * of waiting all the way through the commit rpc or just
575 * waiting for the initial write. The default is to only
576 * wait through the initial write so the data is in the
577 * server's cache, which is roughly similar to the state
578 * a standard disk subsystem leaves the file in on close().
579 *
5a9187cb 580 * We cannot clear the NLMODIFIED bit in np->n_flag due to
984263bc
MD
581 * potential races with other processes, and certainly
582 * cannot clear it if we don't commit.
583 */
584 int cm = nfsv3_commit_on_close ? 1 : 0;
87de5057 585 error = nfs_flush(vp, MNT_WAIT, td, cm);
5a9187cb 586 /* np->n_flag &= ~NLMODIFIED; */
984263bc 587 } else {
87de5057 588 error = nfs_vinvalbuf(vp, V_SAVE, 1);
984263bc
MD
589 }
590 np->n_attrstamp = 0;
591 }
592 if (np->n_flag & NWRITEERR) {
593 np->n_flag &= ~NWRITEERR;
594 error = np->n_error;
595 }
596 }
8ddc6004 597 vop_stdclose(ap);
984263bc
MD
598 return (error);
599}
600
601/*
602 * nfs getattr call from vfs.
e851b29e 603 *
b478fdce 604 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
984263bc
MD
605 */
606static int
e851b29e 607nfs_getattr(struct vop_getattr_args *ap)
984263bc 608{
40393ded
RG
609 struct vnode *vp = ap->a_vp;
610 struct nfsnode *np = VTONFS(vp);
984263bc 611 int error = 0;
87de5057 612 thread_t td = curthread;
42edf14f
MD
613 struct nfsm_info info;
614
615 info.mrep = NULL;
616 info.v3 = NFS_ISV3(vp);
984263bc
MD
617
618 /*
619 * Update local times for special files.
620 */
621 if (np->n_flag & (NACC | NUPD))
622 np->n_flag |= NCHG;
623 /*
624 * First look in the cache.
625 */
626 if (nfs_getattrcache(vp, ap->a_vap) == 0)
627 return (0);
628
42edf14f 629 if (info.v3 && nfsaccess_cache_timeout > 0) {
984263bc 630 nfsstats.accesscache_misses++;
87de5057 631 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
984263bc
MD
632 if (nfs_getattrcache(vp, ap->a_vap) == 0)
633 return (0);
634 }
635
636 nfsstats.rpccnt[NFSPROC_GETATTR]++;
42edf14f
MD
637 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
638 ERROROUT(nfsm_fhtom(&info, vp));
639 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
640 nfs_vpcred(vp, ND_CHECK), &error));
641 if (error == 0) {
642 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
984263bc 643 }
42edf14f
MD
644 m_freem(info.mrep);
645 info.mrep = NULL;
6b08710e 646nfsmout:
984263bc
MD
647 return (error);
648}
649
650/*
651 * nfs setattr call.
e851b29e 652 *
31bd717a 653 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
984263bc
MD
654 */
655static int
e851b29e 656nfs_setattr(struct vop_setattr_args *ap)
984263bc 657{
40393ded
RG
658 struct vnode *vp = ap->a_vp;
659 struct nfsnode *np = VTONFS(vp);
660 struct vattr *vap = ap->a_vap;
984263bc
MD
661 int error = 0;
662 u_quad_t tsize;
87de5057 663 thread_t td = curthread;
984263bc
MD
664
665#ifndef nolint
666 tsize = (u_quad_t)0;
667#endif
668
669 /*
670 * Setting of flags is not supported.
671 */
672 if (vap->va_flags != VNOVAL)
673 return (EOPNOTSUPP);
674
675 /*
676 * Disallow write attempts if the filesystem is mounted read-only.
677 */
678 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
679 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
680 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
681 (vp->v_mount->mnt_flag & MNT_RDONLY))
682 return (EROFS);
a482a28a 683
984263bc 684 if (vap->va_size != VNOVAL) {
a482a28a
MD
685 /*
686 * truncation requested
687 */
984263bc
MD
688 switch (vp->v_type) {
689 case VDIR:
690 return (EISDIR);
691 case VCHR:
692 case VBLK:
693 case VSOCK:
694 case VFIFO:
695 if (vap->va_mtime.tv_sec == VNOVAL &&
696 vap->va_atime.tv_sec == VNOVAL &&
697 vap->va_mode == (mode_t)VNOVAL &&
698 vap->va_uid == (uid_t)VNOVAL &&
699 vap->va_gid == (gid_t)VNOVAL)
700 return (0);
701 vap->va_size = VNOVAL;
702 break;
703 default:
704 /*
705 * Disallow write attempts if the filesystem is
706 * mounted read-only.
707 */
708 if (vp->v_mount->mnt_flag & MNT_RDONLY)
709 return (EROFS);
710
711 /*
a004bca6
MD
712 * This is nasty. The RPCs we send to flush pending
713 * data often return attribute information which is
714 * cached via a callback to nfs_loadattrcache(), which
715 * has the effect of changing our notion of the file
716 * size. Due to flushed appends and other operations
717 * the file size can be set to virtually anything,
718 * including values that do not match either the old
719 * or intended file size.
720 *
721 * When this condition is detected we must loop to
722 * try the operation again. Hopefully no more
723 * flushing is required on the loop so it works the
724 * second time around. THIS CASE ALMOST ALWAYS
725 * HAPPENS!
984263bc 726 */
984263bc 727 tsize = np->n_size;
a004bca6 728again:
87de5057 729 error = nfs_meta_setsize(vp, td, vap->va_size);
984263bc 730
5a9187cb 731 if (np->n_flag & NLMODIFIED) {
984263bc 732 if (vap->va_size == 0)
87de5057 733 error = nfs_vinvalbuf(vp, 0, 1);
984263bc 734 else
87de5057 735 error = nfs_vinvalbuf(vp, V_SAVE, 1);
984263bc 736 }
a004bca6
MD
737 /*
738 * note: this loop case almost always happens at
739 * least once per truncation.
b07fc55c 740 */
a004bca6
MD
741 if (error == 0 && np->n_size != vap->va_size)
742 goto again;
743 np->n_vattr.va_size = vap->va_size;
5a9187cb
MD
744 break;
745 }
a482a28a
MD
746 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
747 /*
748 * What to do. If we are modifying the mtime we lose
749 * mtime detection of changes made by the server or other
750 * clients. But programs like rsync/rdist/cpdup are going
751 * to call utimes a lot. We don't want to piecemeal sync.
752 *
753 * For now sync if any prior remote changes were detected,
754 * but allow us to lose track of remote changes made during
755 * the utimes operation.
756 */
757 if (np->n_flag & NRMODIFIED)
758 error = nfs_vinvalbuf(vp, V_SAVE, 1);
759 if (error == EINTR)
760 return (error);
761 if (error == 0) {
762 if (vap->va_mtime.tv_sec != VNOVAL) {
763 np->n_mtime = vap->va_mtime.tv_sec;
764 }
765 }
a004bca6 766 }
87de5057 767 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
a004bca6
MD
768
769 /*
770 * Sanity check if a truncation was issued. This should only occur
771 * if multiple processes are racing on the same file.
772 */
773 if (error == 0 && vap->va_size != VNOVAL &&
774 np->n_size != vap->va_size) {
973c11b9
MD
775 kprintf("NFS ftruncate: server disagrees on the file size: "
776 "%lld/%lld/%lld\n",
777 (long long)tsize,
778 (long long)vap->va_size,
779 (long long)np->n_size);
a004bca6
MD
780 goto again;
781 }
984263bc
MD
782 if (error && vap->va_size != VNOVAL) {
783 np->n_size = np->n_vattr.va_size = tsize;
784 vnode_pager_setsize(vp, np->n_size);
785 }
786 return (error);
787}
788
789/*
790 * Do an nfs setattr rpc.
791 */
792static int
dadab5e9 793nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
e851b29e 794 struct ucred *cred, struct thread *td)
984263bc 795{
40393ded 796 struct nfsv2_sattr *sp;
999914df 797 struct nfsnode *np = VTONFS(vp);
984263bc
MD
798 u_int32_t *tl;
799 int error = 0, wccflag = NFSV3_WCCRATTR;
42edf14f
MD
800 struct nfsm_info info;
801
802 info.mrep = NULL;
803 info.v3 = NFS_ISV3(vp);
984263bc
MD
804
805 nfsstats.rpccnt[NFSPROC_SETATTR]++;
42edf14f
MD
806 nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
807 NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
808 ERROROUT(nfsm_fhtom(&info, vp));
809 if (info.v3) {
810 nfsm_v3attrbuild(&info, vap, TRUE);
811 tl = nfsm_build(&info, NFSX_UNSIGNED);
984263bc
MD
812 *tl = nfs_false;
813 } else {
42edf14f 814 sp = nfsm_build(&info, NFSX_V2SATTR);
984263bc
MD
815 if (vap->va_mode == (mode_t)VNOVAL)
816 sp->sa_mode = nfs_xdrneg1;
817 else
818 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
819 if (vap->va_uid == (uid_t)VNOVAL)
820 sp->sa_uid = nfs_xdrneg1;
821 else
822 sp->sa_uid = txdr_unsigned(vap->va_uid);
823 if (vap->va_gid == (gid_t)VNOVAL)
824 sp->sa_gid = nfs_xdrneg1;
825 else
826 sp->sa_gid = txdr_unsigned(vap->va_gid);
827 sp->sa_size = txdr_unsigned(vap->va_size);
828 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
829 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
830 }
42edf14f
MD
831 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
832 if (info.v3) {
999914df 833 np->n_modestamp = 0;
42edf14f
MD
834 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
835 } else {
836 ERROROUT(nfsm_loadattr(&info, vp, NULL));
837 }
838 m_freem(info.mrep);
839 info.mrep = NULL;
6b08710e 840nfsmout:
984263bc
MD
841 return (error);
842}
843
a15b4eef
MD
844static
845void
846nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
847{
848 if (nctimeout == 0)
849 nctimeout = 1;
850 else
851 nctimeout *= hz;
852 cache_setvp(nch, vp);
853 cache_settimeout(nch, nctimeout);
854}
855
fad57d0e
MD
856/*
857 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
858 * nfs_lookup() until all remaining new api calls are implemented.
859 *
860 * Resolve a namecache entry. This function is passed a locked ncp and
a15b4eef 861 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
fad57d0e
MD
862 */
863static int
864nfs_nresolve(struct vop_nresolve_args *ap)
865{
866 struct thread *td = curthread;
867 struct namecache *ncp;
868 struct ucred *cred;
869 struct nfsnode *np;
870 struct vnode *dvp;
871 struct vnode *nvp;
872 nfsfh_t *fhp;
873 int attrflag;
874 int fhsize;
875 int error;
046b7e33 876 int tmp_error;
fad57d0e 877 int len;
42edf14f 878 struct nfsm_info info;
fad57d0e
MD
879
880 cred = ap->a_cred;
dff430ab 881 dvp = ap->a_dvp;
fad57d0e 882
87de5057 883 if ((error = vget(dvp, LK_SHARED)) != 0)
fad57d0e
MD
884 return (error);
885
42edf14f
MD
886 info.mrep = NULL;
887 info.v3 = NFS_ISV3(dvp);
888
fad57d0e 889 nvp = NULL;
fad57d0e
MD
890 nfsstats.lookupcache_misses++;
891 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
dff430ab 892 ncp = ap->a_nch->ncp;
fad57d0e 893 len = ncp->nc_nlen;
42edf14f
MD
894 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
895 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
896 ERROROUT(nfsm_fhtom(&info, dvp));
897 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
898 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
899 ap->a_cred, &error));
fad57d0e
MD
900 if (error) {
901 /*
902 * Cache negatve lookups to reduce NFS traffic, but use
903 * a fast timeout. Otherwise use a timeout of 1 tick.
904 * XXX we should add a namecache flag for no-caching
905 * to uncache the negative hit as soon as possible, but
906 * we cannot simply destroy the entry because it is used
907 * as a placeholder by the caller.
046b7e33
MD
908 *
909 * The refactored nfs code will overwrite a non-zero error
910 * with 0 when we use ERROROUT(), so don't here.
fad57d0e 911 */
a15b4eef
MD
912 if (error == ENOENT)
913 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
046b7e33
MD
914 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
915 NFS_LATTR_NOSHRINK);
916 if (tmp_error) {
917 error = tmp_error;
918 goto nfsmout;
919 }
42edf14f
MD
920 m_freem(info.mrep);
921 info.mrep = NULL;
fad57d0e
MD
922 goto nfsmout;
923 }
924
925 /*
926 * Success, get the file handle, do various checks, and load
927 * post-operation data from the reply packet. Theoretically
928 * we should never be looking up "." so, theoretically, we
929 * should never get the same file handle as our directory. But
930 * we check anyway. XXX
931 *
932 * Note that no timeout is set for the positive cache hit. We
933 * assume, theoretically, that ESTALE returns will be dealt with
934 * properly to handle NFS races and in anycase we cannot depend
935 * on a timeout to deal with NFS open/create/excl issues so instead
936 * of a bad hack here the rest of the NFS client code needs to do
937 * the right thing.
938 */
42edf14f 939 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
fad57d0e
MD
940
941 np = VTONFS(dvp);
942 if (NFS_CMPFH(np, fhp, fhsize)) {
943 vref(dvp);
944 nvp = dvp;
945 } else {
946 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
947 if (error) {
42edf14f
MD
948 m_freem(info.mrep);
949 info.mrep = NULL;
fad57d0e
MD
950 vput(dvp);
951 return (error);
952 }
953 nvp = NFSTOV(np);
954 }
42edf14f
MD
955 if (info.v3) {
956 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
957 NFS_LATTR_NOSHRINK));
958 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
959 NFS_LATTR_NOSHRINK));
fad57d0e 960 } else {
42edf14f 961 ERROROUT(nfsm_loadattr(&info, nvp, NULL));
fad57d0e 962 }
a15b4eef 963 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
42edf14f
MD
964 m_freem(info.mrep);
965 info.mrep = NULL;
fad57d0e
MD
966nfsmout:
967 vput(dvp);
968 if (nvp) {
969 if (nvp == dvp)
970 vrele(nvp);
971 else
972 vput(nvp);
973 }
974 return (error);
975}
976
984263bc 977/*
4d17b298 978 * 'cached' nfs directory lookup
e851b29e 979 *
fad57d0e
MD
980 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
981 *
31bd717a
MD
982 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
983 * struct componentname *a_cnp)
984263bc
MD
984 */
985static int
e62afb5f 986nfs_lookup(struct vop_old_lookup_args *ap)
984263bc
MD
987{
988 struct componentname *cnp = ap->a_cnp;
989 struct vnode *dvp = ap->a_dvp;
990 struct vnode **vpp = ap->a_vpp;
991 int flags = cnp->cn_flags;
992 struct vnode *newvp;
984263bc 993 struct nfsmount *nmp;
984263bc
MD
994 long len;
995 nfsfh_t *fhp;
996 struct nfsnode *np;
046b7e33
MD
997 int lockparent, wantparent, attrflag, fhsize;
998 int error;
999 int tmp_error;
42edf14f
MD
1000 struct nfsm_info info;
1001
1002 info.mrep = NULL;
1003 info.v3 = NFS_ISV3(dvp);
046b7e33 1004 error = 0;
984263bc 1005
4d17b298
MD
1006 /*
1007 * Read-only mount check and directory check.
1008 */
984263bc 1009 *vpp = NULLVP;
fad57d0e 1010 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
2b69e610 1011 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
984263bc 1012 return (EROFS);
4d17b298 1013
984263bc
MD
1014 if (dvp->v_type != VDIR)
1015 return (ENOTDIR);
4d17b298
MD
1016
1017 /*
1018 * Look it up in the cache. Note that ENOENT is only returned if we
1019 * previously entered a negative hit (see later on). The additional
1020 * nfsneg_cache_timeout check causes previously cached results to
1021 * be instantly ignored if the negative caching is turned off.
1022 */
2b69e610
MD
1023 lockparent = flags & CNP_LOCKPARENT;
1024 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
984263bc
MD
1025 nmp = VFSTONFS(dvp->v_mount);
1026 np = VTONFS(dvp);
984263bc 1027
4d17b298 1028 /*
fad57d0e 1029 * Go to the wire.
4d17b298 1030 */
984263bc
MD
1031 error = 0;
1032 newvp = NULLVP;
1033 nfsstats.lookupcache_misses++;
1034 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1035 len = cnp->cn_namelen;
42edf14f
MD
1036 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1037 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1038 ERROROUT(nfsm_fhtom(&info, dvp));
1039 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1040 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1041 cnp->cn_cred, &error));
984263bc 1042 if (error) {
046b7e33
MD
1043 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1044 NFS_LATTR_NOSHRINK);
1045 if (tmp_error) {
1046 error = tmp_error;
1047 goto nfsmout;
1048 }
1049
42edf14f
MD
1050 m_freem(info.mrep);
1051 info.mrep = NULL;
984263bc
MD
1052 goto nfsmout;
1053 }
42edf14f 1054 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
984263bc
MD
1055
1056 /*
1057 * Handle RENAME case...
1058 */
fad57d0e 1059 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
984263bc 1060 if (NFS_CMPFH(np, fhp, fhsize)) {
42edf14f
MD
1061 m_freem(info.mrep);
1062 info.mrep = NULL;
984263bc
MD
1063 return (EISDIR);
1064 }
1065 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1066 if (error) {
42edf14f
MD
1067 m_freem(info.mrep);
1068 info.mrep = NULL;
984263bc
MD
1069 return (error);
1070 }
1071 newvp = NFSTOV(np);
42edf14f
MD
1072 if (info.v3) {
1073 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1074 NFS_LATTR_NOSHRINK));
1075 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1076 NFS_LATTR_NOSHRINK));
1077 } else {
1078 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1079 }
984263bc 1080 *vpp = newvp;
42edf14f
MD
1081 m_freem(info.mrep);
1082 info.mrep = NULL;
7ab77df6 1083 if (!lockparent) {
a11aaa81 1084 vn_unlock(dvp);
7ab77df6
MD
1085 cnp->cn_flags |= CNP_PDIRUNLOCK;
1086 }
984263bc
MD
1087 return (0);
1088 }
1089
2b69e610 1090 if (flags & CNP_ISDOTDOT) {
a11aaa81 1091 vn_unlock(dvp);
7ab77df6 1092 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
1093 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1094 if (error) {
ca466bae 1095 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
7ab77df6
MD
1096 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1097 return (error); /* NOTE: return error from nget */
984263bc
MD
1098 }
1099 newvp = NFSTOV(np);
fad57d0e 1100 if (lockparent) {
ca466bae 1101 error = vn_lock(dvp, LK_EXCLUSIVE);
7ab77df6
MD
1102 if (error) {
1103 vput(newvp);
1104 return (error);
1105 }
1106 cnp->cn_flags |= CNP_PDIRUNLOCK;
984263bc
MD
1107 }
1108 } else if (NFS_CMPFH(np, fhp, fhsize)) {
597aea93 1109 vref(dvp);
984263bc
MD
1110 newvp = dvp;
1111 } else {
1112 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1113 if (error) {
42edf14f
MD
1114 m_freem(info.mrep);
1115 info.mrep = NULL;
984263bc
MD
1116 return (error);
1117 }
fad57d0e 1118 if (!lockparent) {
a11aaa81 1119 vn_unlock(dvp);
7ab77df6
MD
1120 cnp->cn_flags |= CNP_PDIRUNLOCK;
1121 }
984263bc
MD
1122 newvp = NFSTOV(np);
1123 }
42edf14f
MD
1124 if (info.v3) {
1125 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1126 NFS_LATTR_NOSHRINK));
1127 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1128 NFS_LATTR_NOSHRINK));
1129 } else {
1130 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1131 }
fad57d0e
MD
1132#if 0
1133 /* XXX MOVE TO nfs_nremove() */
2b69e610 1134 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
fad57d0e
MD
1135 cnp->cn_nameiop != NAMEI_DELETE) {
1136 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
984263bc 1137 }
fad57d0e 1138#endif
984263bc 1139 *vpp = newvp;
42edf14f
MD
1140 m_freem(info.mrep);
1141 info.mrep = NULL;
6b08710e 1142nfsmout:
984263bc
MD
1143 if (error) {
1144 if (newvp != NULLVP) {
1145 vrele(newvp);
1146 *vpp = NULLVP;
1147 }
fad57d0e
MD
1148 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1149 cnp->cn_nameiop == NAMEI_RENAME) &&
1150 error == ENOENT) {
7ab77df6 1151 if (!lockparent) {
a11aaa81 1152 vn_unlock(dvp);
7ab77df6
MD
1153 cnp->cn_flags |= CNP_PDIRUNLOCK;
1154 }
984263bc
MD
1155 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1156 error = EROFS;
1157 else
1158 error = EJUSTRETURN;
1159 }
984263bc
MD
1160 }
1161 return (error);
1162}
1163
1164/*
1165 * nfs read call.
1166 * Just call nfs_bioread() to do the work.
e851b29e
CP
1167 *
1168 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1169 * struct ucred *a_cred)
984263bc
MD
1170 */
1171static int
e851b29e 1172nfs_read(struct vop_read_args *ap)
984263bc 1173{
40393ded 1174 struct vnode *vp = ap->a_vp;
984263bc 1175
3b568787 1176 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
984263bc
MD
1177}
1178
1179/*
1180 * nfs readlink call
e851b29e
CP
1181 *
1182 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
984263bc
MD
1183 */
1184static int
e851b29e 1185nfs_readlink(struct vop_readlink_args *ap)
984263bc 1186{
40393ded 1187 struct vnode *vp = ap->a_vp;
984263bc
MD
1188
1189 if (vp->v_type != VLNK)
1190 return (EINVAL);
3b568787 1191 return (nfs_bioread(vp, ap->a_uio, 0));
984263bc
MD
1192}
1193
1194/*
1195 * Do a readlink rpc.
1196 * Called by nfs_doio() from below the buffer cache.
1197 */
1198int
cc7d050e 1199nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
984263bc 1200{
984263bc 1201 int error = 0, len, attrflag;
42edf14f
MD
1202 struct nfsm_info info;
1203
1204 info.mrep = NULL;
1205 info.v3 = NFS_ISV3(vp);
984263bc
MD
1206
1207 nfsstats.rpccnt[NFSPROC_READLINK]++;
42edf14f
MD
1208 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1209 ERROROUT(nfsm_fhtom(&info, vp));
1210 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1211 nfs_vpcred(vp, ND_CHECK), &error));
1212 if (info.v3) {
1213 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1214 NFS_LATTR_NOSHRINK));
1215 }
984263bc 1216 if (!error) {
42edf14f 1217 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
984263bc
MD
1218 if (len == NFS_MAXPATHLEN) {
1219 struct nfsnode *np = VTONFS(vp);
1220 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1221 len = np->n_size;
1222 }
42edf14f 1223 ERROROUT(nfsm_mtouio(&info, uiop, len));
984263bc 1224 }
42edf14f
MD
1225 m_freem(info.mrep);
1226 info.mrep = NULL;
6b08710e 1227nfsmout:
984263bc
MD
1228 return (error);
1229}
1230
1231/*
edb90c22
MD
1232 * nfs read rpc.
1233 *
1234 * If bio is non-NULL and asynchronous
984263bc
MD
1235 */
1236int
edb90c22 1237nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
984263bc 1238{
40393ded 1239 u_int32_t *tl;
984263bc
MD
1240 struct nfsmount *nmp;
1241 int error = 0, len, retlen, tsiz, eof, attrflag;
42edf14f
MD
1242 struct nfsm_info info;
1243
1244 info.mrep = NULL;
1245 info.v3 = NFS_ISV3(vp);
984263bc
MD
1246
1247#ifndef nolint
1248 eof = 0;
1249#endif
1250 nmp = VFSTONFS(vp->v_mount);
1251 tsiz = uiop->uio_resid;
1252 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1253 return (EFBIG);
1254 while (tsiz > 0) {
1255 nfsstats.rpccnt[NFSPROC_READ]++;
1256 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
42edf14f
MD
1257 nfsm_reqhead(&info, vp, NFSPROC_READ,
1258 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1259 ERROROUT(nfsm_fhtom(&info, vp));
1260 tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1261 if (info.v3) {
984263bc
MD
1262 txdr_hyper(uiop->uio_offset, tl);
1263 *(tl + 2) = txdr_unsigned(len);
1264 } else {
1265 *tl++ = txdr_unsigned(uiop->uio_offset);
1266 *tl++ = txdr_unsigned(len);
1267 *tl = 0;
1268 }
42edf14f
MD
1269 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1270 nfs_vpcred(vp, ND_READ), &error));
1271 if (info.v3) {
1272 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1273 NFS_LATTR_NOSHRINK));
1274 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
984263bc 1275 eof = fxdr_unsigned(int, *(tl + 1));
42edf14f
MD
1276 } else {
1277 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1278 }
1279 NEGATIVEOUT(retlen = nfsm_strsiz(&info, nmp->nm_rsize));
1280 ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1281 m_freem(info.mrep);
1282 info.mrep = NULL;
984263bc 1283 tsiz -= retlen;
42edf14f 1284 if (info.v3) {
984263bc
MD
1285 if (eof || retlen == 0) {
1286 tsiz = 0;
1287 }
1288 } else if (retlen < len) {
1289 tsiz = 0;
1290 }
1291 }
1292nfsmout:
1293 return (error);
1294}
1295
1296/*
1297 * nfs write call
1298 */
1299int
cc7d050e
MD
1300nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1301 int *iomode, int *must_commit)
984263bc 1302{
40393ded 1303 u_int32_t *tl;
42edf14f 1304 int32_t backup;
984263bc
MD
1305 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1306 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
42edf14f
MD
1307 int committed = NFSV3WRITE_FILESYNC;
1308 struct nfsm_info info;
1309
1310 info.mrep = NULL;
1311 info.v3 = NFS_ISV3(vp);
984263bc
MD
1312
1313#ifndef DIAGNOSTIC
1314 if (uiop->uio_iovcnt != 1)
1315 panic("nfs: writerpc iovcnt > 1");
1316#endif
1317 *must_commit = 0;
1318 tsiz = uiop->uio_resid;
1319 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1320 return (EFBIG);
1321 while (tsiz > 0) {
1322 nfsstats.rpccnt[NFSPROC_WRITE]++;
1323 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
42edf14f
MD
1324 nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1325 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1326 ERROROUT(nfsm_fhtom(&info, vp));
1327 if (info.v3) {
1328 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
984263bc
MD
1329 txdr_hyper(uiop->uio_offset, tl);
1330 tl += 2;
1331 *tl++ = txdr_unsigned(len);
1332 *tl++ = txdr_unsigned(*iomode);
1333 *tl = txdr_unsigned(len);
1334 } else {
40393ded 1335 u_int32_t x;
984263bc 1336
42edf14f 1337 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
984263bc
MD
1338 /* Set both "begin" and "current" to non-garbage. */
1339 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1340 *tl++ = x; /* "begin offset" */
1341 *tl++ = x; /* "current offset" */
1342 x = txdr_unsigned(len);
1343 *tl++ = x; /* total to this offset */
1344 *tl = x; /* size of this write */
1345 }
42edf14f
MD
1346 ERROROUT(nfsm_uiotom(&info, uiop, len));
1347 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1348 nfs_vpcred(vp, ND_WRITE), &error));
1349 if (info.v3) {
5a9187cb
MD
1350 /*
1351 * The write RPC returns a before and after mtime. The
1352 * nfsm_wcc_data() macro checks the before n_mtime
1353 * against the before time and stores the after time
1354 * in the nfsnode's cached vattr and n_mtime field.
1355 * The NRMODIFIED bit will be set if the before
1356 * time did not match the original mtime.
1357 */
984263bc 1358 wccflag = NFSV3_WCCCHK;
42edf14f
MD
1359 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1360 if (error == 0) {
1361 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
984263bc
MD
1362 rlen = fxdr_unsigned(int, *tl++);
1363 if (rlen == 0) {
1364 error = NFSERR_IO;
42edf14f
MD
1365 m_freem(info.mrep);
1366 info.mrep = NULL;
984263bc
MD
1367 break;
1368 } else if (rlen < len) {
1369 backup = len - rlen;
656849c6 1370 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
984263bc
MD
1371 uiop->uio_iov->iov_len += backup;
1372 uiop->uio_offset -= backup;
1373 uiop->uio_resid += backup;
1374 len = rlen;
1375 }
1376 commit = fxdr_unsigned(int, *tl++);
1377
1378 /*
1379 * Return the lowest committment level
1380 * obtained by any of the RPCs.
1381 */
1382 if (committed == NFSV3WRITE_FILESYNC)
1383 committed = commit;
1384 else if (committed == NFSV3WRITE_DATASYNC &&
1385 commit == NFSV3WRITE_UNSTABLE)
1386 committed = commit;
1387 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1388 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1389 NFSX_V3WRITEVERF);
1390 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1391 } else if (bcmp((caddr_t)tl,
1392 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1393 *must_commit = 1;
1394 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1395 NFSX_V3WRITEVERF);
1396 }
1397 }
5a9187cb 1398 } else {
42edf14f 1399 ERROROUT(nfsm_loadattr(&info, vp, NULL));
5a9187cb 1400 }
42edf14f
MD
1401 m_freem(info.mrep);
1402 info.mrep = NULL;
984263bc
MD
1403 if (error)
1404 break;
1405 tsiz -= len;
1406 }
1407nfsmout:
1408 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1409 committed = NFSV3WRITE_FILESYNC;
1410 *iomode = committed;
1411 if (error)
1412 uiop->uio_resid = tsiz;
1413 return (error);
1414}
1415
1416/*
1417 * nfs mknod rpc
1418 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1419 * mode set to specify the file type and the size field for rdev.
1420 */
1421static int
e851b29e
CP
1422nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1423 struct vattr *vap)
984263bc 1424{
40393ded
RG
1425 struct nfsv2_sattr *sp;
1426 u_int32_t *tl;
60233e58
SW
1427 struct vnode *newvp = NULL;
1428 struct nfsnode *np = NULL;
984263bc 1429 struct vattr vattr;
984263bc 1430 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
0e9b9130 1431 int rmajor, rminor;
42edf14f
MD
1432 struct nfsm_info info;
1433
1434 info.mrep = NULL;
1435 info.v3 = NFS_ISV3(dvp);
984263bc 1436
0e9b9130
MD
1437 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1438 rmajor = txdr_unsigned(vap->va_rmajor);
1439 rminor = txdr_unsigned(vap->va_rminor);
1440 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1441 rmajor = nfs_xdrneg1;
1442 rminor = nfs_xdrneg1;
1443 } else {
984263bc
MD
1444 return (EOPNOTSUPP);
1445 }
87de5057 1446 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
984263bc
MD
1447 return (error);
1448 }
1449 nfsstats.rpccnt[NFSPROC_MKNOD]++;
42edf14f
MD
1450 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1451 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1452 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1453 ERROROUT(nfsm_fhtom(&info, dvp));
1454 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1455 NFS_MAXNAMLEN));
1456 if (info.v3) {
1457 tl = nfsm_build(&info, NFSX_UNSIGNED);
984263bc 1458 *tl++ = vtonfsv3_type(vap->va_type);
42edf14f 1459 nfsm_v3attrbuild(&info, vap, FALSE);
984263bc 1460 if (vap->va_type == VCHR || vap->va_type == VBLK) {
42edf14f 1461 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
0e9b9130
MD
1462 *tl++ = txdr_unsigned(vap->va_rmajor);
1463 *tl = txdr_unsigned(vap->va_rminor);
984263bc
MD
1464 }
1465 } else {
42edf14f 1466 sp = nfsm_build(&info, NFSX_V2SATTR);
984263bc
MD
1467 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1468 sp->sa_uid = nfs_xdrneg1;
1469 sp->sa_gid = nfs_xdrneg1;
0e9b9130 1470 sp->sa_size = makeudev(rmajor, rminor);
984263bc
MD
1471 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1472 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1473 }
42edf14f
MD
1474 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1475 cnp->cn_cred, &error));
984263bc 1476 if (!error) {
42edf14f 1477 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
984263bc
MD
1478 if (!gotvp) {
1479 if (newvp) {
1480 vput(newvp);
60233e58 1481 newvp = NULL;
984263bc
MD
1482 }
1483 error = nfs_lookitup(dvp, cnp->cn_nameptr,
dadab5e9 1484 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
984263bc
MD
1485 if (!error)
1486 newvp = NFSTOV(np);
1487 }
1488 }
42edf14f
MD
1489 if (info.v3) {
1490 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1491 }
1492 m_freem(info.mrep);
1493 info.mrep = NULL;
6b08710e 1494nfsmout:
984263bc
MD
1495 if (error) {
1496 if (newvp)
1497 vput(newvp);
1498 } else {
984263bc
MD
1499 *vpp = newvp;
1500 }
5a9187cb 1501 VTONFS(dvp)->n_flag |= NLMODIFIED;
984263bc
MD
1502 if (!wccflag)
1503 VTONFS(dvp)->n_attrstamp = 0;
1504 return (error);
1505}
1506
1507/*
1508 * nfs mknod vop
1509 * just call nfs_mknodrpc() to do the work.
e851b29e
CP
1510 *
1511 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1512 * struct componentname *a_cnp, struct vattr *a_vap)
984263bc
MD
1513 */
1514/* ARGSUSED */
1515static int
e62afb5f 1516nfs_mknod(struct vop_old_mknod_args *ap)
984263bc
MD
1517{
1518 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1519}
1520
1521static u_long create_verf;
1522/*
1523 * nfs file create call
e851b29e
CP
1524 *
1525 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1526 * struct componentname *a_cnp, struct vattr *a_vap)
984263bc
MD
1527 */
1528static int
e62afb5f 1529nfs_create(struct vop_old_create_args *ap)
984263bc 1530{
40393ded
RG
1531 struct vnode *dvp = ap->a_dvp;
1532 struct vattr *vap = ap->a_vap;
1533 struct componentname *cnp = ap->a_cnp;
1534 struct nfsv2_sattr *sp;
1535 u_int32_t *tl;
60233e58
SW
1536 struct nfsnode *np = NULL;
1537 struct vnode *newvp = NULL;
984263bc 1538 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
984263bc 1539 struct vattr vattr;
42edf14f
MD
1540 struct nfsm_info info;
1541
1542 info.mrep = NULL;
1543 info.v3 = NFS_ISV3(dvp);
984263bc
MD
1544
1545 /*
1546 * Oops, not for me..
1547 */
1548 if (vap->va_type == VSOCK)
1549 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1550
87de5057 1551 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
984263bc
MD
1552 return (error);
1553 }
1554 if (vap->va_vaflags & VA_EXCLUSIVE)
1555 fmode |= O_EXCL;
1556again:
1557 nfsstats.rpccnt[NFSPROC_CREATE]++;
42edf14f
MD
1558 nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1559 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1560 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1561 ERROROUT(nfsm_fhtom(&info, dvp));
1562 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1563 NFS_MAXNAMLEN));
1564 if (info.v3) {
1565 tl = nfsm_build(&info, NFSX_UNSIGNED);
984263bc
MD
1566 if (fmode & O_EXCL) {
1567 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
42edf14f 1568 tl = nfsm_build(&info, NFSX_V3CREATEVERF);
984263bc 1569#ifdef INET
1b562c24
SZ
1570 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1571 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
984263bc
MD
1572 else
1573#endif
1574 *tl++ = create_verf;
1575 *tl = ++create_verf;
1576 } else {
1577 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
42edf14f 1578 nfsm_v3attrbuild(&info, vap, FALSE);
984263bc
MD
1579 }
1580 } else {
42edf14f 1581 sp = nfsm_build(&info, NFSX_V2SATTR);
984263bc
MD
1582 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1583 sp->sa_uid = nfs_xdrneg1;
1584 sp->sa_gid = nfs_xdrneg1;
1585 sp->sa_size = 0;
1586 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1587 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1588 }
42edf14f
MD
1589 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1590 cnp->cn_cred, &error));
1591 if (error == 0) {
1592 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
984263bc
MD
1593 if (!gotvp) {
1594 if (newvp) {
1595 vput(newvp);
60233e58 1596 newvp = NULL;
984263bc
MD
1597 }
1598 error = nfs_lookitup(dvp, cnp->cn_nameptr,
dadab5e9 1599 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
984263bc
MD
1600 if (!error)
1601 newvp = NFSTOV(np);
1602 }
1603 }
42edf14f 1604 if (info.v3) {
d64fd6d3
MD
1605 if (error == 0)
1606 error = nfsm_wcc_data(&info, dvp, &wccflag);
1607 else
1608 (void)nfsm_wcc_data(&info, dvp, &wccflag);
42edf14f
MD
1609 }
1610 m_freem(info.mrep);
1611 info.mrep = NULL;
6b08710e 1612nfsmout:
984263bc 1613 if (error) {
42edf14f 1614 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
c5c4cc5d 1615 KKASSERT(newvp == NULL);
984263bc
MD
1616 fmode &= ~O_EXCL;
1617 goto again;
1618 }
42edf14f 1619 } else if (info.v3 && (fmode & O_EXCL)) {
984263bc
MD
1620 /*
1621 * We are normally called with only a partially initialized
1622 * VAP. Since the NFSv3 spec says that server may use the
1623 * file attributes to store the verifier, the spec requires
1624 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1625 * in atime, but we can't really assume that all servers will
1626 * so we ensure that our SETATTR sets both atime and mtime.
1627 */
1628 if (vap->va_mtime.tv_sec == VNOVAL)
1629 vfs_timestamp(&vap->va_mtime);
1630 if (vap->va_atime.tv_sec == VNOVAL)
1631 vap->va_atime = vap->va_mtime;
dadab5e9 1632 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
984263bc 1633 }
c5c4cc5d 1634 if (error == 0) {
c1cf1e59
MD
1635 /*
1636 * The new np may have enough info for access
1637 * checks, make sure rucred and wucred are
1638 * initialized for read and write rpc's.
1639 */
1640 np = VTONFS(newvp);
1641 if (np->n_rucred == NULL)
1642 np->n_rucred = crhold(cnp->cn_cred);
1643 if (np->n_wucred == NULL)
1644 np->n_wucred = crhold(cnp->cn_cred);
984263bc 1645 *ap->a_vpp = newvp;
c5c4cc5d
MD
1646 } else if (newvp) {
1647 vput(newvp);
984263bc 1648 }
5a9187cb 1649 VTONFS(dvp)->n_flag |= NLMODIFIED;
984263bc
MD
1650 if (!wccflag)
1651 VTONFS(dvp)->n_attrstamp = 0;
1652 return (error);
1653}
1654
1655/*
1656 * nfs file remove call
1657 * To try and make nfs semantics closer to ufs semantics, a file that has
1658 * other processes using the vnode is renamed instead of removed and then
1659 * removed later on the last close.
3c37c940 1660 * - If v_sysref.refcnt > 1
984263bc
MD
1661 * If a rename is not already in the works
1662 * call nfs_sillyrename() to set it up
1663 * else
1664 * do the remove rpc
e851b29e 1665 *
31bd717a
MD
1666 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1667 * struct componentname *a_cnp)
984263bc
MD
1668 */
1669static int
e62afb5f 1670nfs_remove(struct vop_old_remove_args *ap)
984263bc 1671{
40393ded
RG
1672 struct vnode *vp = ap->a_vp;
1673 struct vnode *dvp = ap->a_dvp;
1674 struct componentname *cnp = ap->a_cnp;
1675 struct nfsnode *np = VTONFS(vp);
984263bc
MD
1676 int error = 0;
1677 struct vattr vattr;
1678
1679#ifndef DIAGNOSTIC
3c37c940
MD
1680 if (vp->v_sysref.refcnt < 1)
1681 panic("nfs_remove: bad v_sysref.refcnt");
984263bc
MD
1682#endif
1683 if (vp->v_type == VDIR)
1684 error = EPERM;
3c37c940 1685 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
87de5057 1686 VOP_GETATTR(vp, &vattr) == 0 &&
984263bc 1687 vattr.va_nlink > 1)) {
984263bc
MD
1688 /*
1689 * throw away biocache buffers, mainly to avoid
1690 * unnecessary delayed writes later.
1691 */
87de5057 1692 error = nfs_vinvalbuf(vp, 0, 1);
984263bc
MD
1693 /* Do the rpc */
1694 if (error != EINTR)
1695 error = nfs_removerpc(dvp, cnp->cn_nameptr,
dadab5e9 1696 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
984263bc
MD
1697 /*
1698 * Kludge City: If the first reply to the remove rpc is lost..
1699 * the reply to the retransmitted request will be ENOENT
1700 * since the file was in fact removed
1701 * Therefore, we cheat and return success.
1702 */
1703 if (error == ENOENT)
1704 error = 0;
fad57d0e 1705 } else if (!np->n_sillyrename) {
984263bc 1706 error = nfs_sillyrename(dvp, vp, cnp);
fad57d0e 1707 }
984263bc
MD
1708 np->n_attrstamp = 0;
1709 return (error);
1710}
1711
1712/*
1713 * nfs file remove rpc called from nfs_inactive
1714 */
1715int
dadab5e9 1716nfs_removeit(struct sillyrename *sp)
984263bc 1717{
dadab5e9
MD
1718 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1719 sp->s_cred, NULL));
984263bc
MD
1720}
1721
1722/*
1723 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1724 */
1725static int
e851b29e
CP
1726nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1727 struct ucred *cred, struct thread *td)
984263bc 1728{
984263bc 1729 int error = 0, wccflag = NFSV3_WCCRATTR;
42edf14f
MD
1730 struct nfsm_info info;
1731
1732 info.mrep = NULL;
1733 info.v3 = NFS_ISV3(dvp);
984263bc
MD
1734
1735 nfsstats.rpccnt[NFSPROC_REMOVE]++;
42edf14f
MD
1736 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1737 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1738 ERROROUT(nfsm_fhtom(&info, dvp));
1739 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1740 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1741 if (info.v3) {
1742 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1743 }
1744 m_freem(info.mrep);
1745 info.mrep = NULL;
6b08710e 1746nfsmout:
5a9187cb 1747 VTONFS(dvp)->n_flag |= NLMODIFIED;
984263bc
MD
1748 if (!wccflag)
1749 VTONFS(dvp)->n_attrstamp = 0;
1750 return (error);
1751}
1752
1753/*
1754 * nfs file rename call
e851b29e
CP
1755 *
1756 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1757 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1758 * struct vnode *a_tvp, struct componentname *a_tcnp)
984263bc
MD
1759 */
1760static int
e62afb5f 1761nfs_rename(struct vop_old_rename_args *ap)
984263bc 1762{
40393ded
RG
1763 struct vnode *fvp = ap->a_fvp;
1764 struct vnode *tvp = ap->a_tvp;
1765 struct vnode *fdvp = ap->a_fdvp;
1766 struct vnode *tdvp = ap->a_tdvp;
1767 struct componentname *tcnp = ap->a_tcnp;
1768 struct componentname *fcnp = ap->a_fcnp;
984263bc
MD
1769 int error;
1770
984263bc
MD
1771 /* Check for cross-device rename */
1772 if ((fvp->v_mount != tdvp->v_mount) ||
1773 (tvp && (fvp->v_mount != tvp->v_mount))) {
1774 error = EXDEV;
1775 goto out;
1776 }
1777
1778 /*
bb7bf7b2
MD
1779 * We shouldn't have to flush fvp on rename for most server-side
1780 * filesystems as the file handle should not change. Unfortunately
1781 * the inode for some filesystems (msdosfs) might be tied to the
1782 * file name or directory position so to be completely safe
1783 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1784 * performance.
a482a28a
MD
1785 *
1786 * We must flush tvp on rename because it might become stale on the
1787 * server after the rename.
984263bc 1788 */
a482a28a 1789 if (nfs_flush_on_rename)
52174f71 1790 VOP_FSYNC(fvp, MNT_WAIT, 0);
984263bc 1791 if (tvp)
52174f71 1792 VOP_FSYNC(tvp, MNT_WAIT, 0);
984263bc
MD
1793
1794 /*
1795 * If the tvp exists and is in use, sillyrename it before doing the
1796 * rename of the new file over it.
fad57d0e 1797 *
984263bc 1798 * XXX Can't sillyrename a directory.
5fd012e0 1799 *
fad57d0e
MD
1800 * We do not attempt to do any namecache purges in this old API
1801 * routine. The new API compat functions have access to the actual
1802 * namecache structures and will do it for us.
984263bc 1803 */
3c37c940 1804 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
984263bc
MD
1805 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1806 vput(tvp);
1807 tvp = NULL;
5fd012e0 1808 } else if (tvp) {
fad57d0e 1809 ;
984263bc
MD
1810 }
1811
1812 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1813 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
dadab5e9 1814 tcnp->cn_td);
984263bc 1815
984263bc
MD
1816out:
1817 if (tdvp == tvp)
1818 vrele(tdvp);
1819 else
1820 vput(tdvp);
1821 if (tvp)
1822 vput(tvp);
1823 vrele(fdvp);
1824 vrele(fvp);
1825 /*
1826 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1827 */
1828 if (error == ENOENT)
1829 error = 0;
1830 return (error);
1831}
1832
1833/*
1834 * nfs file rename rpc called from nfs_remove() above
1835 */
1836static int
e851b29e
CP
1837nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1838 struct sillyrename *sp)
984263bc
MD
1839{
1840 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
dadab5e9 1841 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
984263bc
MD
1842}
1843
1844/*
1845 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1846 */
1847static int
e851b29e
CP
1848nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1849 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1850 struct ucred *cred, struct thread *td)
984263bc 1851{
984263bc 1852 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
42edf14f
MD
1853 struct nfsm_info info;
1854
1855 info.mrep = NULL;
1856 info.v3 = NFS_ISV3(fdvp);
984263bc
MD
1857
1858 nfsstats.rpccnt[NFSPROC_RENAME]++;
42edf14f
MD
1859 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1860 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1861 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1862 ERROROUT(nfsm_fhtom(&info, fdvp));
1863 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1864 ERROROUT(nfsm_fhtom(&info, tdvp));
1865 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1866 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
1867 if (info.v3) {
1868 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
1869 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
1870 }
1871 m_freem(info.mrep);
1872 info.mrep = NULL;
6b08710e 1873nfsmout:
5a9187cb
MD
1874 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1875 VTONFS(tdvp)->n_flag |= NLMODIFIED;
984263bc
MD
1876 if (!fwccflag)
1877 VTONFS(fdvp)->n_attrstamp = 0;
1878 if (!twccflag)
1879 VTONFS(tdvp)->n_attrstamp = 0;
1880 return (error);
1881}
1882
1883/*
1884 * nfs hard link create call
e851b29e
CP
1885 *
1886 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1887 * struct componentname *a_cnp)
984263bc
MD
1888 */
1889static int
e62afb5f 1890nfs_link(struct vop_old_link_args *ap)
984263bc 1891{
40393ded
RG
1892 struct vnode *vp = ap->a_vp;
1893 struct vnode *tdvp = ap->a_tdvp;
1894 struct componentname *cnp = ap->a_cnp;
984263bc 1895 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
42edf14f 1896 struct nfsm_info info;
984263bc
MD
1897
1898 if (vp->v_mount != tdvp->v_mount) {
1899 return (EXDEV);
1900 }
1901
1902 /*
bb7bf7b2
MD
1903 * The attribute cache may get out of sync with the server on link.
1904 * Pushing writes to the server before handle was inherited from
1905 * long long ago and it is unclear if we still need to do this.
1906 * Defaults to off.
984263bc 1907 */
bb7bf7b2 1908 if (nfs_flush_on_hlink)
52174f71 1909 VOP_FSYNC(vp, MNT_WAIT, 0);
984263bc 1910
42edf14f
MD
1911 info.mrep = NULL;
1912 info.v3 = NFS_ISV3(vp);
1913
984263bc 1914 nfsstats.rpccnt[NFSPROC_LINK]++;
42edf14f
MD
1915 nfsm_reqhead(&info, vp, NFSPROC_LINK,
1916 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
1917 nfsm_rndup(cnp->cn_namelen));
1918 ERROROUT(nfsm_fhtom(&info, vp));
1919 ERROROUT(nfsm_fhtom(&info, tdvp));
1920 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1921 NFS_MAXNAMLEN));
1922 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
1923 cnp->cn_cred, &error));
1924 if (info.v3) {
1925 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1926 NFS_LATTR_NOSHRINK));
1927 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
1928 }
1929 m_freem(info.mrep);
1930 info.mrep = NULL;
6b08710e 1931nfsmout:
5a9187cb 1932 VTONFS(tdvp)->n_flag |= NLMODIFIED;
984263bc
MD
1933 if (!attrflag)
1934 VTONFS(vp)->n_attrstamp = 0;
1935 if (!wccflag)
1936 VTONFS(tdvp)->n_attrstamp = 0;
1937 /*
1938 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1939 */
1940 if (error == EEXIST)
1941 error = 0;
1942 return (error);
1943}
1944
1945/*
1946 * nfs symbolic link create call
e851b29e
CP
1947 *
1948 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1949 * struct componentname *a_cnp, struct vattr *a_vap,
1950 * char *a_target)
984263bc
MD
1951 */
1952static int
e62afb5f 1953nfs_symlink(struct vop_old_symlink_args *ap)
984263bc 1954{
40393ded
RG
1955 struct vnode *dvp = ap->a_dvp;
1956 struct vattr *vap = ap->a_vap;
1957 struct componentname *cnp = ap->a_cnp;
1958 struct nfsv2_sattr *sp;
984263bc 1959 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
60233e58 1960 struct vnode *newvp = NULL;
42edf14f
MD
1961 struct nfsm_info info;
1962
1963 info.mrep = NULL;
1964 info.v3 = NFS_ISV3(dvp);
984263bc
MD
1965
1966 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1967 slen = strlen(ap->a_target);
42edf14f
MD
1968 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
1969 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
1970 nfsm_rndup(cnp->cn_namelen) +
1971 nfsm_rndup(slen) + NFSX_SATTR(info.v3));
1972 ERROROUT(nfsm_fhtom(&info, dvp));
1973 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1974 NFS_MAXNAMLEN));
1975 if (info.v3) {
1976 nfsm_v3attrbuild(&info, vap, FALSE);
1977 }
1978 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
1979 if (info.v3 == 0) {
1980 sp = nfsm_build(&info, NFSX_V2SATTR);
984263bc
MD
1981 sp->sa_mode = vtonfsv2_mode(VLNK, 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 }
1988
1989 /*
1990 * Issue the NFS request and get the rpc response.
1991 *
1992 * Only NFSv3 responses returning an error of 0 actually return
1993 * a file handle that can be converted into newvp without having
1994 * to do an extra lookup rpc.
1995 */
42edf14f
MD
1996 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
1997 cnp->cn_cred, &error));
1998 if (info.v3) {
1999 if (error == 0) {
2000 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2001 }
2002 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
984263bc
MD
2003 }
2004
2005 /*
2006 * out code jumps -> here, mrep is also freed.
2007 */
2008
42edf14f
MD
2009 m_freem(info.mrep);
2010 info.mrep = NULL;
6b08710e 2011nfsmout:
984263bc
MD
2012
2013 /*
2014 * If we get an EEXIST error, silently convert it to no-error
2015 * in case of an NFS retry.
2016 */
2017 if (error == EEXIST)
2018 error = 0;
2019
2020 /*
2021 * If we do not have (or no longer have) an error, and we could
2022 * not extract the newvp from the response due to the request being
2023 * NFSv2 or the error being EEXIST. We have to do a lookup in order
2024 * to obtain a newvp to return.
2025 */
2026 if (error == 0 && newvp == NULL) {
2027 struct nfsnode *np = NULL;
2028
2029 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
dadab5e9 2030 cnp->cn_cred, cnp->cn_td, &np);
984263bc
MD
2031 if (!error)
2032 newvp = NFSTOV(np);
2033 }
2034 if (error) {
2035 if (newvp)
2036 vput(newvp);
2037 } else {
2038 *ap->a_vpp = newvp;
2039 }
5a9187cb 2040 VTONFS(dvp)->n_flag |= NLMODIFIED;
984263bc
MD
2041 if (!wccflag)
2042 VTONFS(dvp)->n_attrstamp = 0;
2043 return (error);
2044}
2045
2046/*
2047 * nfs make dir call
e851b29e
CP
2048 *
2049 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2050 * struct componentname *a_cnp, struct vattr *a_vap)
984263bc
MD
2051 */
2052static int
e62afb5f 2053nfs_mkdir(struct vop_old_mkdir_args *ap)
984263bc 2054{
40393ded
RG
2055 struct vnode *dvp = ap->a_dvp;
2056 struct vattr *vap = ap->a_vap;
2057 struct componentname *cnp = ap->a_cnp;
2058 struct nfsv2_sattr *sp;
60233e58
SW
2059 struct nfsnode *np = NULL;
2060 struct vnode *newvp = NULL;
42edf14f 2061 struct vattr vattr;
984263bc
MD
2062 int error = 0, wccflag = NFSV3_WCCRATTR;
2063 int gotvp = 0;
42edf14f
MD
2064 int len;
2065 struct nfsm_info info;
2066
2067 info.mrep = NULL;
2068 info.v3 = NFS_ISV3(dvp);
984263bc 2069
87de5057 2070 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
984263bc
MD
2071 return (error);
2072 }
2073 len = cnp->cn_namelen;
2074 nfsstats.rpccnt[NFSPROC_MKDIR]++;
42edf14f
MD
2075 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2076 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2077 nfsm_rndup(len) + NFSX_SATTR(info.v3));
2078 ERROROUT(nfsm_fhtom(&info, dvp));
2079 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2080 if (info.v3) {
2081 nfsm_v3attrbuild(&info, vap, FALSE);
984263bc 2082 } else {
42edf14f 2083 sp = nfsm_build(&info, NFSX_V2SATTR);
984263bc
MD
2084 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2085 sp->sa_uid = nfs_xdrneg1;
2086 sp->sa_gid = nfs_xdrneg1;
2087 sp->sa_size = nfs_xdrneg1;
2088 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2089 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2090 }
42edf14f
MD
2091 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2092 cnp->cn_cred, &error));
2093 if (error == 0) {
2094 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2095 }
2096 if (info.v3) {
2097 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2098 }
2099 m_freem(info.mrep);
2100 info.mrep = NULL;
6b08710e 2101nfsmout:
5a9187cb 2102 VTONFS(dvp)->n_flag |= NLMODIFIED;
984263bc
MD
2103 if (!wccflag)
2104 VTONFS(dvp)->n_attrstamp = 0;
2105 /*
2106 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2107 * if we can succeed in looking up the directory.
2108 */
2109 if (error == EEXIST || (!error && !gotvp)) {
2110 if (newvp) {
2111 vrele(newvp);
60233e58 2112 newvp = NULL;
984263bc
MD
2113 }
2114 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
dadab5e9 2115 cnp->cn_td, &np);
984263bc
MD
2116 if (!error) {
2117 newvp = NFSTOV(np);
2118 if (newvp->v_type != VDIR)
2119 error = EEXIST;
2120 }
2121 }
2122 if (error) {
2123 if (newvp)
2124 vrele(newvp);
2125 } else
2126 *ap->a_vpp = newvp;
2127 return (error);
2128}
2129
2130/*
2131 * nfs remove directory call
e851b29e
CP
2132 *
2133 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2134 * struct componentname *a_cnp)
984263bc
MD
2135 */
2136static int
e62afb5f 2137nfs_rmdir(struct vop_old_rmdir_args *ap)
984263bc 2138{
40393ded
RG
2139 struct vnode *vp = ap->a_vp;
2140 struct vnode *dvp = ap->a_dvp;
2141 struct componentname *cnp = ap->a_cnp;
984263bc 2142 int error = 0, wccflag = NFSV3_WCCRATTR;
42edf14f
MD
2143 struct nfsm_info info;
2144
2145 info.mrep = NULL;
2146 info.v3 = NFS_ISV3(dvp);
984263bc
MD
2147
2148 if (dvp == vp)
2149 return (EINVAL);
2150 nfsstats.rpccnt[NFSPROC_RMDIR]++;
42edf14f
MD
2151 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2152 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2153 nfsm_rndup(cnp->cn_namelen));
2154 ERROROUT(nfsm_fhtom(&info, dvp));
2155 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2156 NFS_MAXNAMLEN));
2157 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2158 cnp->cn_cred, &error));
2159 if (info.v3) {
2160 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2161 }
2162 m_freem(info.mrep);
2163 info.mrep = NULL;
6b08710e 2164nfsmout:
5a9187cb 2165 VTONFS(dvp)->n_flag |= NLMODIFIED;
984263bc
MD
2166 if (!wccflag)
2167 VTONFS(dvp)->n_attrstamp = 0;
984263bc
MD
2168 /*
2169 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2170 */
2171 if (error == ENOENT)
2172 error = 0;
2173 return (error);
2174}
2175
2176/*
2177 * nfs readdir call
e851b29e
CP
2178 *
2179 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
984263bc
MD
2180 */
2181static int
e851b29e 2182nfs_readdir(struct vop_readdir_args *ap)
984263bc 2183{
40393ded
RG
2184 struct vnode *vp = ap->a_vp;
2185 struct nfsnode *np = VTONFS(vp);
2186 struct uio *uio = ap->a_uio;
984263bc
MD
2187 int tresid, error;
2188 struct vattr vattr;
2189
2190 if (vp->v_type != VDIR)
2191 return (EPERM);
5a9187cb 2192
885ecb13
MD
2193 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2194 return (error);
2195
984263bc 2196 /*
5a9187cb
MD
2197 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2198 * and then check that is still valid, or if this is an NQNFS mount
2199 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2200 * VOP_GETATTR() does not necessarily go to the wire.
984263bc
MD
2201 */
2202 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
5a9187cb 2203 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
87de5057 2204 if (VOP_GETATTR(vp, &vattr) == 0 &&
e07fef60 2205 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
5a9187cb 2206 ) {
984263bc 2207 nfsstats.direofcache_hits++;
885ecb13 2208 goto done;
984263bc
MD
2209 }
2210 }
2211
2212 /*
5a9187cb
MD
2213 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2214 * own cache coherency checks so we do not have to.
984263bc
MD
2215 */
2216 tresid = uio->uio_resid;
3b568787 2217 error = nfs_bioread(vp, uio, 0);
984263bc
MD
2218
2219 if (!error && uio->uio_resid == tresid)
2220 nfsstats.direofcache_misses++;
885ecb13
MD
2221done:
2222 vn_unlock(vp);
984263bc
MD
2223 return (error);
2224}
2225
2226/*
7d877edf
MD
2227 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2228 *
2229 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2230 * offset/block and converts the nfs formatted directory entries for userland
2231 * consumption as well as deals with offsets into the middle of blocks.
2232 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2233 * be block-bounded. It must convert to cookies for the actual RPC.
984263bc
MD
2234 */
2235int
cc7d050e 2236nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
984263bc 2237{
40393ded 2238 int len, left;
01f31ab3 2239 struct nfs_dirent *dp = NULL;
40393ded 2240 u_int32_t *tl;
40393ded 2241 nfsuint64 *cookiep;
42edf14f 2242 caddr_t cp;
984263bc
MD
2243 nfsuint64 cookie;
2244 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2245 struct nfsnode *dnp = VTONFS(vp);
2246 u_quad_t fileno;
2247 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2248 int attrflag;
42edf14f
MD
2249 struct nfsm_info info;
2250
2251 info.mrep = NULL;
2252 info.v3 = NFS_ISV3(vp);
984263bc
MD
2253
2254#ifndef DIAGNOSTIC
2255 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2256 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2257 panic("nfs readdirrpc bad uio");
2258#endif
2259
2260 /*
2261 * If there is no cookie, assume directory was stale.
2262 */
2263 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2264 if (cookiep)
2265 cookie = *cookiep;
2266 else
2267 return (NFSERR_BAD_COOKIE);
2268 /*
2269 * Loop around doing readdir rpc's of size nm_readdirsize
2270 * truncated to a multiple of DIRBLKSIZ.
2271 * The stopping criteria is EOF or buffer full.
2272 */
2273 while (more_dirs && bigenough) {
2274 nfsstats.rpccnt[NFSPROC_READDIR]++;
42edf14f
MD
2275 nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2276 NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2277 ERROROUT(nfsm_fhtom(&info, vp));
2278 if (info.v3) {
2279 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
984263bc
MD
2280 *tl++ = cookie.nfsuquad[0];
2281 *tl++ = cookie.nfsuquad[1];
2282 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2283 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2284 } else {
42edf14f 2285 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
984263bc
MD
2286 *tl++ = cookie.nfsuquad[0];
2287 }
2288 *tl = txdr_unsigned(nmp->nm_readdirsize);
42edf14f
MD
2289 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2290 uiop->uio_td,
2291 nfs_vpcred(vp, ND_READ), &error));
2292 if (info.v3) {
2293 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2294 NFS_LATTR_NOSHRINK));
2295 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2296 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2297 dnp->n_cookieverf.nfsuquad[1] = *tl;
984263bc 2298 }
42edf14f 2299 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
984263bc
MD
2300 more_dirs = fxdr_unsigned(int, *tl);
2301
7d877edf 2302 /* loop thru the dir entries, converting them to std form */
984263bc 2303 while (more_dirs && bigenough) {
42edf14f
MD
2304 if (info.v3) {
2305 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
984263bc
MD
2306 fileno = fxdr_hyper(tl);
2307 len = fxdr_unsigned(int, *(tl + 2));
2308 } else {
42edf14f 2309 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
984263bc
MD
2310 fileno = fxdr_unsigned(u_quad_t, *tl++);
2311 len = fxdr_unsigned(int, *tl);
2312 }
2313 if (len <= 0 || len > NFS_MAXNAMLEN) {
2314 error = EBADRPC;
42edf14f
MD
2315 m_freem(info.mrep);
2316 info.mrep = NULL;
984263bc
MD
2317 goto nfsmout;
2318 }
7d877edf
MD
2319
2320 /*
2321 * len is the number of bytes in the path element
2322 * name, not including the \0 termination.
2323 *
2324 * tlen is the number of bytes w have to reserve for
2325 * the path element name.
2326 */
984263bc
MD
2327 tlen = nfsm_rndup(len);
2328 if (tlen == len)
2329 tlen += 4; /* To ensure null termination */
7d877edf
MD
2330
2331 /*
2332 * If the entry would cross a DIRBLKSIZ boundary,
2333 * extend the previous nfs_dirent to cover the
2334 * remaining space.
2335 */
984263bc 2336 left = DIRBLKSIZ - blksiz;
01f31ab3
JS
2337 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2338 dp->nfs_reclen += left;
656849c6 2339 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
984263bc
MD
2340 uiop->uio_iov->iov_len -= left;
2341 uiop->uio_offset += left;
2342 uiop->uio_resid -= left;
2343 blksiz = 0;
2344 }
01f31ab3 2345 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
984263bc
MD
2346 bigenough = 0;
2347 if (bigenough) {
01f31ab3
JS
2348 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2349 dp->nfs_ino = fileno;
2350 dp->nfs_namlen = len;
2351 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2352 dp->nfs_type = DT_UNKNOWN;
2353 blksiz += dp->nfs_reclen;
984263bc
MD
2354 if (blksiz == DIRBLKSIZ)
2355 blksiz = 0;
01f31ab3
JS
2356 uiop->uio_offset += sizeof(struct nfs_dirent);
2357 uiop->uio_resid -= sizeof(struct nfs_dirent);
656849c6 2358 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
01f31ab3 2359 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
42edf14f 2360 ERROROUT(nfsm_mtouio(&info, uiop, len));
7d877edf
MD
2361
2362 /*
2363 * The uiop has advanced by nfs_dirent + len
2364 * but really needs to advance by
2365 * nfs_dirent + tlen
2366 */
984263bc
MD
2367 cp = uiop->uio_iov->iov_base;
2368 tlen -= len;
2369 *cp = '\0'; /* null terminate */
656849c6 2370 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
984263bc
MD
2371 uiop->uio_iov->iov_len -= tlen;
2372 uiop->uio_offset += tlen;
2373 uiop->uio_resid -= tlen;
7d877edf
MD
2374 } else {
2375 /*
2376 * NFS strings must be rounded up (nfsm_myouio
2377 * handled that in the bigenough case).
2378 */
42edf14f 2379 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
7d877edf 2380 }
42edf14f
MD
2381 if (info.v3) {
2382 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
984263bc 2383 } else {
42edf14f 2384 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
984263bc 2385 }
7d877edf
MD
2386
2387 /*
2388 * If we were able to accomodate the last entry,
2389 * get the cookie for the next one. Otherwise
2390 * hold-over the cookie for the one we were not
2391 * able to accomodate.
2392 */
984263bc
MD
2393 if (bigenough) {
2394 cookie.nfsuquad[0] = *tl++;
42edf14f 2395 if (info.v3)
984263bc 2396 cookie.nfsuquad[1] = *tl++;
42edf14f 2397 } else if (info.v3) {
984263bc 2398 tl += 2;
7d877edf 2399 } else {
984263bc 2400 tl++;
7d877edf 2401 }
984263bc
MD
2402 more_dirs = fxdr_unsigned(int, *tl);
2403 }
2404 /*
2405 * If at end of rpc data, get the eof boolean
2406 */
2407 if (!more_dirs) {
42edf14f 2408 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
984263bc
MD
2409 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2410 }
42edf14f
MD
2411 m_freem(info.mrep);
2412 info.mrep = NULL;
984263bc
MD
2413 }
2414 /*
2415 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2416 * by increasing d_reclen for the last record.
2417 */
2418 if (blksiz > 0) {
2419 left = DIRBLKSIZ - blksiz;
01f31ab3 2420 dp->nfs_reclen += left;
656849c6 2421 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
984263bc
MD
2422 uiop->uio_iov->iov_len -= left;
2423 uiop->uio_offset += left;
2424 uiop->uio_resid -= left;
2425 }
2426
7d877edf
MD
2427 if (bigenough) {
2428 /*
2429 * We hit the end of the directory, update direofoffset.
2430 */
984263bc 2431 dnp->n_direofoffset = uiop->uio_offset;
7d877edf
MD
2432 } else {
2433 /*
2434 * There is more to go, insert the link cookie so the
2435 * next block can be read.
2436 */
984263bc 2437 if (uiop->uio_resid > 0)
086c1d7e 2438 kprintf("EEK! readdirrpc resid > 0\n");
984263bc
MD
2439 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2440 *cookiep = cookie;
2441 }
2442nfsmout:
2443 return (error);
2444}
2445
2446/*
2447 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2448 */
2449int
cc7d050e 2450nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
984263bc 2451{
40393ded 2452 int len, left;
01f31ab3 2453 struct nfs_dirent *dp;
40393ded 2454 u_int32_t *tl;
40393ded
RG
2455 struct vnode *newvp;
2456 nfsuint64 *cookiep;
42edf14f
MD
2457 caddr_t dpossav1, dpossav2;
2458 caddr_t cp;
2459 struct mbuf *mdsav1, *mdsav2;
984263bc
MD
2460 nfsuint64 cookie;
2461 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2462 struct nfsnode *dnp = VTONFS(vp), *np;
2463 nfsfh_t *fhp;
2464 u_quad_t fileno;
2465 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2466 int attrflag, fhsize;
28623bf9
MD
2467 struct nchandle nch;
2468 struct nchandle dnch;
fad57d0e 2469 struct nlcomponent nlc;
42edf14f
MD
2470 struct nfsm_info info;
2471
2472 info.mrep = NULL;
2473 info.v3 = 1;
984263bc
MD
2474
2475#ifndef nolint
01f31ab3 2476 dp = NULL;
984263bc
MD
2477#endif
2478#ifndef DIAGNOSTIC
2479 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2480 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2481 panic("nfs readdirplusrpc bad uio");
2482#endif
fad57d0e
MD
2483 /*
2484 * Obtain the namecache record for the directory so we have something
2485 * to use as a basis for creating the entries. This function will
2486 * return a held (but not locked) ncp. The ncp may be disconnected
2487 * from the tree and cannot be used for upward traversals, and the
2488 * ncp may be unnamed. Note that other unrelated operations may
2489 * cause the ncp to be named at any time.
2490 */
28623bf9 2491 cache_fromdvp(vp, NULL, 0, &dnch);
fad57d0e 2492 bzero(&nlc, sizeof(nlc));
984263bc
MD
2493 newvp = NULLVP;
2494
2495 /*
2496 * If there is no cookie, assume directory was stale.
2497 */
2498 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2499 if (cookiep)
2500 cookie = *cookiep;
2501 else
2502 return (NFSERR_BAD_COOKIE);
2503 /*
2504 * Loop around doing readdir rpc's of size nm_readdirsize
2505 * truncated to a multiple of DIRBLKSIZ.
2506 * The stopping criteria is EOF or buffer full.
2507 */
2508 while (more_dirs && bigenough) {
2509 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
42edf14f
MD
2510 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2511 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2512 ERROROUT(nfsm_fhtom(&info, vp));
2513 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
984263bc
MD
2514 *tl++ = cookie.nfsuquad[0];
2515 *tl++ = cookie.nfsuquad[1];
2516 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2517 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2518 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2519 *tl = txdr_unsigned(nmp->nm_rsize);
42edf14f
MD
2520 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2521 uiop->uio_td,
2522 nfs_vpcred(vp, ND_READ), &error));
2523 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2524 NFS_LATTR_NOSHRINK));
2525 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
984263bc
MD
2526 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2527 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2528 more_dirs = fxdr_unsigned(int, *tl);
2529
2530 /* loop thru the dir entries, doctoring them to 4bsd form */
2531 while (more_dirs && bigenough) {
42edf14f 2532 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
984263bc
MD
2533 fileno = fxdr_hyper(tl);
2534 len = fxdr_unsigned(int, *(tl + 2));
2535 if (len <= 0 || len > NFS_MAXNAMLEN) {
2536 error = EBADRPC;
42edf14f
MD
2537 m_freem(info.mrep);
2538 info.mrep = NULL;
984263bc
MD
2539 goto nfsmout;
2540 }
2541 tlen = nfsm_rndup(len);
2542 if (tlen == len)
2543 tlen += 4; /* To ensure null termination*/
2544 left = DIRBLKSIZ - blksiz;
01f31ab3
JS
2545 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2546 dp->nfs_reclen += left;
656849c6 2547 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
984263bc
MD
2548 uiop->uio_iov->iov_len -= left;
2549 uiop->uio_offset += left;
2550 uiop->uio_resid -= left;
2551 blksiz = 0;
2552 }
01f31ab3 2553 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
984263bc
MD
2554 bigenough = 0;
2555 if (bigenough) {
01f31ab3
JS
2556 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2557 dp->nfs_ino = fileno;
2558 dp->nfs_namlen = len;
2559 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2560 dp->nfs_type = DT_UNKNOWN;
2561 blksiz += dp->nfs_reclen;
984263bc
MD
2562 if (blksiz == DIRBLKSIZ)
2563 blksiz = 0;
01f31ab3
JS
2564 uiop->uio_offset += sizeof(struct nfs_dirent);
2565 uiop->uio_resid -= sizeof(struct nfs_dirent);
656849c6 2566 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
01f31ab3 2567 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
fad57d0e
MD
2568 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2569 nlc.nlc_namelen = len;
42edf14f 2570 ERROROUT(nfsm_mtouio(&info, uiop, len));
984263bc
MD
2571 cp = uiop->uio_iov->iov_base;
2572 tlen -= len;
2573 *cp = '\0';
656849c6 2574 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
984263bc
MD
2575 uiop->uio_iov->iov_len -= tlen;
2576 uiop->uio_offset += tlen;
2577 uiop->uio_resid -= tlen;
42edf14f
MD
2578 } else {
2579 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2580 }
2581 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
984263bc
MD
2582 if (bigenough) {
2583 cookie.nfsuquad[0] = *tl++;
2584 cookie.nfsuquad[1] = *tl++;
2585 } else
2586 tl += 2;
2587
2588 /*
2589 * Since the attributes are before the file handle
2590 * (sigh), we must skip over the attributes and then
2591 * come back and get them.
2592 */
2593 attrflag = fxdr_unsigned(int, *tl);
2594 if (attrflag) {
42edf14f
MD
2595 dpossav1 = info.dpos;
2596 mdsav1 = info.md;
2597 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2598 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
984263bc
MD
2599 doit = fxdr_unsigned(int, *tl);
2600 if (doit) {
42edf14f 2601 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
984263bc 2602 if (NFS_CMPFH(dnp, fhp, fhsize)) {
597aea93 2603 vref(vp);
984263bc
MD
2604 newvp = vp;
2605 np = dnp;
2606 } else {
2607 error = nfs_nget(vp->v_mount, fhp,
2608 fhsize, &np);
2609 if (error)
2610 doit = 0;
2611 else
2612 newvp = NFSTOV(np);
2613 }
2614 }
2615 if (doit && bigenough) {
42edf14f
MD
2616 dpossav2 = info.dpos;
2617 info.dpos = dpossav1;
2618 mdsav2 = info.md;
2619 info.md = mdsav1;
2620 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2621 info.dpos = dpossav2;
2622 info.md = mdsav2;
01f31ab3 2623 dp->nfs_type =
984263bc 2624 IFTODT(VTTOIF(np->n_vattr.va_type));
28623bf9 2625 if (dnch.ncp) {
086c1d7e 2626 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
fad57d0e
MD
2627 nlc.nlc_namelen, nlc.nlc_namelen,
2628 nlc.nlc_nameptr);
28623bf9
MD
2629 nch = cache_nlookup(&dnch, &nlc);
2630 cache_setunresolved(&nch);
a15b4eef
MD
2631 nfs_cache_setvp(&nch, newvp,
2632 nfspos_cache_timeout);
28623bf9 2633 cache_put(&nch);
fad57d0e 2634 } else {
086c1d7e 2635 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER"
fad57d0e
MD
2636 " %*.*s\n",
2637 nlc.nlc_namelen, nlc.nlc_namelen,
2638 nlc.nlc_nameptr);
2639 }
984263bc
MD
2640 }
2641 } else {
2642 /* Just skip over the file handle */
42edf14f 2643 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
984263bc 2644 i = fxdr_unsigned(int, *tl);
42edf14f 2645 ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
984263bc
MD
2646 }
2647 if (newvp != NULLVP) {
2648 if (newvp == vp)
2649 vrele(newvp);
2650 else
2651 vput(newvp);
2652 newvp = NULLVP;
2653 }
42edf14f 2654 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
984263bc
MD
2655 more_dirs = fxdr_unsigned(int, *tl);
2656 }
2657 /*
2658 * If at end of rpc data, get the eof boolean
2659 */
2660 if (!more_dirs) {
42edf14f 2661 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
984263bc
MD
2662 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2663 }
42edf14f
MD
2664 m_freem(info.mrep);
2665 info.mrep = NULL;
984263bc
MD
2666 }
2667 /*
2668 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2669 * by increasing d_reclen for the last record.
2670 */
2671 if (blksiz > 0) {
2672 left = DIRBLKSIZ - blksiz;
01f31ab3 2673 dp->nfs_reclen += left;
656849c6 2674 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
984263bc
MD
2675 uiop->uio_iov->iov_len -= left;
2676 uiop->uio_offset += left;
2677 uiop->uio_resid -= left;
2678 }
2679
2680 /*
2681 * We are now either at the end of the directory or have filled the
2682 * block.
2683 */
2684 if (bigenough)
2685 dnp->n_direofoffset = uiop->uio_offset;
2686 else {
2687 if (uiop->uio_resid > 0)
086c1d7e 2688 kprintf("EEK! readdirplusrpc resid > 0\n");
984263bc
MD
2689 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2690 *cookiep = cookie;
2691 }
2692nfsmout:
2693 if (newvp != NULLVP) {
2694 if (newvp == vp)
2695 vrele(newvp);
2696 else
2697 vput(newvp);
2698 newvp = NULLVP;
2699 }
28623bf9
MD
2700 if (dnch.ncp)
2701 cache_drop(&dnch);
984263bc
MD
2702 return (error);
2703}
2704
2705/*
2706 * Silly rename. To make the NFS filesystem that is stateless look a little
2707 * more like the "ufs" a remove of an active vnode is translated to a rename
2708 * to a funny looking filename that is removed by nfs_inactive on the
2709 * nfsnode. There is the potential for another process on a different client
2710 * to create the same funny name between the nfs_lookitup() fails and the
2711 * nfs_rename() completes, but...
2712 */
2713static int
e851b29e 2714nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
984263bc 2715{
40393ded 2716 struct sillyrename *sp;
984263bc
MD
2717 struct nfsnode *np;
2718 int error;
984263bc 2719
8c361dda
MD
2720 /*
2721 * We previously purged dvp instead of vp. I don't know why, it
2722 * completely destroys performance. We can't do it anyway with the
2723 * new VFS API since we would be breaking the namecache topology.
2724 */
fad57d0e 2725 cache_purge(vp); /* XXX */
984263bc
MD
2726 np = VTONFS(vp);
2727#ifndef DIAGNOSTIC
2728 if (vp->v_type == VDIR)
2729 panic("nfs: sillyrename dir");
2730#endif
2731 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2732 M_NFSREQ, M_WAITOK);
2733 sp->s_cred = crdup(cnp->cn_cred);
2734 sp->s_dvp = dvp;
597aea93 2735 vref(dvp);
984263bc
MD
2736
2737 /* Fudge together a funny name */
973c11b9
MD
2738 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2739 (int)(intptr_t)cnp->cn_td);
984263bc
MD
2740
2741 /* Try lookitups until we get one that isn't there */
2742 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
60233e58 2743 cnp->cn_td, NULL) == 0) {
984263bc
MD
2744 sp->s_name[4]++;
2745 if (sp->s_name[4] > 'z') {
2746 error = EINVAL;
2747 goto bad;
2748 }
2749 }
2750 error = nfs_renameit(dvp, cnp, sp);
2751 if (error)
2752 goto bad;
2753 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
dadab5e9 2754 cnp->cn_td, &np);
984263bc
MD
2755 np->n_sillyrename = sp;
2756 return (0);
2757bad:
2758 vrele(sp->s_dvp);
2759 crfree(sp->s_cred);
efda3bd0 2760 kfree((caddr_t)sp, M_NFSREQ);
984263bc
MD
2761 return (error);
2762}
2763
2764/*
2765 * Look up a file name and optionally either update the file handle or
2766 * allocate an nfsnode, depending on the value of npp.
2767 * npp == NULL --> just do the lookup
2768 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2769 * handled too
2770 * *npp != NULL --> update the file handle in the vnode
2771 */
2772static int
e851b29e
CP
2773nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2774 struct thread *td, struct nfsnode **npp)
984263bc 2775{
60233e58 2776 struct vnode *newvp = NULL;
984263bc 2777 struct nfsnode *np, *dnp = VTONFS(dvp);
984263bc 2778 int error = 0, fhlen, attrflag;
984263bc 2779 nfsfh_t *nfhp;
42edf14f
MD
2780 struct nfsm_info info;
2781
2782 info.mrep = NULL;
2783 info.v3 = NFS_ISV3(dvp);
984263bc
MD
2784
2785 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
42edf14f
MD
2786 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2787 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2788 ERROROUT(nfsm_fhtom(&info, dvp));
2789 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2790 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
984263bc 2791 if (npp && !error) {
42edf14f 2792 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
984263bc
MD
2793 if (*npp) {
2794 np = *npp;
2795 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
efda3bd0 2796 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
984263bc
MD
2797 np->n_fhp = &np->n_fh;
2798 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
efda3bd0 2799 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
984263bc
MD
2800 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2801 np->n_fhsize = fhlen;
2802 newvp = NFSTOV(np);
2803 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
597aea93 2804 vref(dvp);
984263bc
MD
2805 newvp = dvp;
2806 } else {
2807 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2808 if (error) {
42edf14f
MD
2809 m_freem(info.mrep);
2810 info.mrep = NULL;
984263bc
MD
2811 return (error);
2812 }
2813 newvp = NFSTOV(np);
2814 }
42edf14f
MD
2815 if (info.v3) {
2816 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2817 NFS_LATTR_NOSHRINK));
984263bc 2818 if (!attrflag && *npp == NULL) {
42edf14f
MD
2819 m_freem(info.mrep);
2820 info.mrep = NULL;
984263bc
MD
2821 if (newvp == dvp)
2822 vrele(newvp);
2823 else
2824 vput(newvp);
2825 return (ENOENT);
2826 }
42edf14f
MD
2827 } else {
2828 ERROROUT(error = nfsm_loadattr(&info, newvp, NULL));
2829 }
984263bc 2830 }
42edf14f
MD
2831 m_freem(info.mrep);
2832 info.mrep = NULL;
6b08710e 2833nfsmout:
984263bc
MD
2834 if (npp && *npp == NULL) {
2835 if (error) {
2836 if (newvp) {
2837 if (newvp == dvp)
2838 vrele(newvp);
2839 else
2840 vput(newvp);
2841 }
2842 } else
2843 *npp = np;
2844 }
2845 return (error);
2846}
2847
2848/*
2849 * Nfs Version 3 commit rpc
cc7d050e
MD
2850 *
2851 * We call it 'uio' to distinguish it from 'bio' but there is no real uio
2852 * involved.
984263bc
MD
2853 */
2854int
cc7d050e 2855nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
984263bc 2856{
40393ded 2857 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
984263bc 2858 int error = 0, wccflag = NFSV3_WCCRATTR;
42edf14f
MD
2859 struct nfsm_info info;
2860 u_int32_t *tl;
2861
2862 info.mrep = NULL;
2863 info.v3 = 1;
984263bc
MD
2864
2865 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2866 return (0);
2867 nfsstats.rpccnt[NFSPROC_COMMIT]++;
42edf14f
MD
2868 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
2869 ERROROUT(nfsm_fhtom(&info, vp));
2870 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
984263bc
MD
2871 txdr_hyper(offset, tl);
2872 tl += 2;
2873 *tl = txdr_unsigned(cnt);
42edf14f
MD
2874 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
2875 nfs_vpcred(vp, ND_WRITE), &error));
2876 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
984263bc 2877 if (!error) {
42edf14f 2878 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
984263bc
MD
2879 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2880 NFSX_V3WRITEVERF)) {
2881 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2882 NFSX_V3WRITEVERF);
2883 error = NFSERR_STALEWRITEVERF;
2884 }
2885 }
42edf14f
MD
2886 m_freem(info.mrep);
2887 info.mrep = NULL;
6b08710e 2888nfsmout:
984263bc
MD
2889 return (error);
2890}
2891
2892/*
2893 * Kludge City..
2894 * - make nfs_bmap() essentially a no-op that does no translation
2895 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2896 * (Maybe I could use the process's page mapping, but I was concerned that
2897 * Kernel Write might not be enabled and also figured copyout() would do
2898 * a lot more work than bcopy() and also it currently happens in the
2899 * context of the swapper process (2).
e851b29e 2900 *
08daea96 2901 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
54078292 2902 * off_t *a_doffsetp, int *a_runp, int *a_runb)
984263bc
MD
2903 */
2904static int
e851b29e 2905nfs_bmap(struct vop_bmap_args *ap)
984263bc 2906{
54078292
MD
2907 if (ap->a_doffsetp != NULL)
2908 *ap->a_doffsetp = ap->a_loffset;
984263bc
MD
2909 if (ap->a_runp != NULL)
2910 *ap->a_runp = 0;
2911 if (ap->a_runb != NULL)
2912 *ap->a_runb = 0;
2913 return (0);
2914}
2915
2916/*
2917 * Strategy routine.
984263bc
MD
2918 */
2919static int
e851b29e 2920nfs_strategy(struct vop_strategy_args *ap)
984263bc 2921{
81b5c339
MD
2922 struct bio *bio = ap->a_bio;
2923 struct bio *nbio;
2924 struct buf *bp = bio->bio_buf;
dadab5e9 2925 struct thread *td;
cc7d050e 2926 int error;
984263bc 2927
10f3fee5
MD
2928 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2929 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
81b5c339
MD
2930 KASSERT(BUF_REFCNT(bp) > 0,
2931 ("nfs_strategy: buffer %p not locked", bp));
984263bc 2932
ae8e83e6 2933 if (bio->bio_flags & BIO_SYNC)
dadab5e9 2934 td = curthread; /* XXX */
ae8e83e6
MD
2935 else
2936 td = NULL;
984263bc 2937
81b5c339 2938 /*
54078292
MD
2939 * We probably don't need to push an nbio any more since no
2940 * block conversion is required due to the use of 64 bit byte
2941 * offsets, but do it anyway.
8bfa5617
MD
2942 *
2943 * NOTE: When NFS callers itself via this strategy routines and
2944 * sets up a synchronous I/O, it expects the I/O to run
2945 * synchronously (its bio_done routine just assumes it),
2946 * so for now we have to honor the bit.
81b5c339
MD
2947 */
2948 nbio = push_bio(bio);
54078292 2949 nbio->bio_offset = bio->bio_offset;
8bfa5617 2950 nbio->bio_flags = bio->bio_flags & BIO_SYNC;
81b5c339 2951
984263bc
MD
2952 /*
2953 * If the op is asynchronous and an i/o daemon is waiting
2954 * queue the request, wake it up and wait for completion
2955 * otherwise just do it ourselves.
2956 */
edb90c22 2957 if (bio->bio_flags & BIO_SYNC) {
cc7d050e 2958 error = nfs_doio(ap->a_vp, nbio, td);
edb90c22
MD
2959 } else {
2960 nfs_asyncio(ap->a_vp, nbio);
cc7d050e 2961 error = 0;
edb90c22 2962 }
cc7d050e 2963 return (error);
984263bc
MD
2964}
2965
2966/*
2967 * Mmap a file
2968 *
2969 * NB Currently unsupported.
e851b29e 2970 *
b478fdce 2971 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
984263bc
MD
2972 */
2973/* ARGSUSED */
2974static int
e851b29e 2975nfs_mmap(struct vop_mmap_args *ap)
984263bc 2976{
984263bc
MD
2977 return (EINVAL);
2978}
2979
2980/*
2981 * fsync vnode op. Just call nfs_flush() with commit == 1.
e851b29e 2982 *
b478fdce 2983 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
984263bc
MD
2984 */
2985/* ARGSUSED */
2986static int
e851b29e 2987nfs_fsync(struct vop_fsync_args *ap)
984263bc 2988{
87de5057 2989 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
984263bc
MD
2990}
2991
2992/*
6bae6177
MD
2993 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2994 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2995 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2996 * set the buffer contains data that has already been written to the server
2997 * and which now needs a commit RPC.
2998 *
2999 * If commit is 0 we only take one pass and only flush buffers containing new
3000 * dirty data.
3001 *
3002 * If commit is 1 we take two passes, issuing a commit RPC in the second
3003 * pass.
3004 *
3005 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3006 * to completely flush all pending data.
3007 *
3008 * Note that the RB_SCAN code properly handles the case where the
3009 * callback might block and directly or indirectly (another thread) cause
3010 * the RB tree to change.
984263bc 3011 */
6bae6177
MD
3012
3013#ifndef NFS_COMMITBVECSIZ
3014#define NFS_COMMITBVECSIZ 16
3015#endif
3016
3017struct nfs_flush_info {
3018 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3019 struct thread *td;
3020 struct vnode *vp;
3021 int waitfor;
3022 int slpflag;
3023 int slptimeo;
3024 int loops;
3025 struct buf *bvary[NFS_COMMITBVECSIZ];
3026 int bvsize;
3027 off_t beg_off;
3028 off_t end_off;
3029};
3030
3031static int nfs_flush_bp(struct buf *bp, void *data);
3032static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3033
5a9187cb 3034int
e851b29e 3035nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
984263bc 3036{
40393ded 3037 struct nfsnode *np = VTONFS(vp);
984263bc 3038 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
6bae6177 3039 struct nfs_flush_info info;
0202303b 3040 lwkt_tokref vlock;
6bae6177 3041 int error;
984263bc 3042
6bae6177
MD
3043 bzero(&info, sizeof(info));
3044 info.td = td;
3045 info.vp = vp;
3046 info.waitfor = waitfor;
3047 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3048 info.loops = 0;
0202303b 3049 lwkt_gettoken(&vlock, &vp->v_token);
6bae6177
MD
3050
3051 do {
984263bc 3052 /*
6bae6177 3053 * Flush mode
984263bc 3054 */
6bae6177
MD
3055 info.mode = NFI_FLUSHNEW;
3056 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3057 nfs_flush_bp, &info);
3058
3059 /*
3060 * Take a second pass if committing and no error occured.
3061 * Clean up any left over collection (whether an error
3062 * occurs or not).
3063 */
3064 if (commit && error == 0) {
3065 info.mode = NFI_COMMIT;
3066 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3067 nfs_flush_bp, &info);
3068 if (info.bvsize)
3069 error = nfs_flush_docommit(&info, error);
984263bc 3070 }
6bae6177 3071
984263bc 3072 /*
6bae6177
MD
3073 * Wait for pending I/O to complete before checking whether
3074 * any further dirty buffers exist.
984263bc 3075 */
a9a20f98
MD
3076 while (waitfor == MNT_WAIT &&
3077 bio_track_active(&vp->v_track_write)) {
3078 error = bio_track_wait(&vp->v_track_write,
3079 info.slpflag, info.slptimeo);
6bae6177
MD
3080 if (error) {
3081 /*
3082 * We have to be able to break out if this
3083 * is an 'intr' mount.
3084 */
60233e58 3085 if (nfs_sigintr(nmp, NULL, td)) {
6bae6177
MD
3086 error = -EINTR;
3087 break;
3088 }
3089
3090 /*
3091 * Since we do not process pending signals,
3092 * once we get a PCATCH our tsleep() will no
3093 * longer sleep, switch to a fixed timeout
3094 * instead.
3095 */
3096 if (info.slpflag == PCATCH) {
3097 info.slpflag = 0;
3098 info.slptimeo = 2 * hz;
3099 }
3100 error = 0;
3101 }
3102 }
3103 ++info.loops;
3104 /*
3105 * Loop if we are flushing synchronous as well as committing,
3106 * and dirty buffers are still present. Otherwise we might livelock.
3107 */
3108 } while (waitfor == MNT_WAIT && commit &&
3109 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3110
3111 /*
3112 * The callbacks have to return a negative error to terminate the
3113 * RB scan.
3114 */
3115 if (error < 0)
3116 error = -error;
3117
3118 /*
3119 * Deal with any error collection
3120 */
3121 if (np->n_flag & NWRITEERR) {
3122 error = np->n_error;
3123 np->n_flag &= ~NWRITEERR;
3124 }
0202303b 3125 lwkt_reltoken(&vlock);
6bae6177
MD
3126 return (error);
3127}
3128
6bae6177
MD
3129static
3130int
3131nfs_flush_bp(struct buf *bp, void *data)
3132{
3133 struct nfs_flush_info *info = data;
b1c20cfa 3134 int lkflags;
6bae6177 3135 int error;
b1c20cfa 3136 off_t toff;
6bae6177
MD
3137
3138 error = 0;
3139 switch(info->mode) {
3140 case NFI_FLUSHNEW:
b1c20cfa
MD
3141 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3142 if (error && info->loops && info->waitfor == MNT_WAIT) {
6bae6177
MD
3143 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3144 if (error) {
b1c20cfa 3145 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
f2770c70
MD
3146 if (info->slpflag & PCATCH)
3147 lkflags |= LK_PCATCH;
3148 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3149 info->slptimeo);
6bae6177 3150 }
984263bc 3151 }
984263bc 3152
b1c20cfa
MD
3153 /*
3154 * Ignore locking errors
3155 */
3156 if (error) {
3157 error = 0;
3158 break;
3159 }
3160
3161 /*
3162 * The buffer may have changed out from under us, even if
3163 * we did not block (MPSAFE). Check again now that it is
3164 * locked.
3165 */
3166 if (bp->b_vp == info->vp &&
3167 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3168 bremfree(bp);
a482a28a 3169 bawrite(bp);
6bae6177 3170 } else {
b1c20cfa 3171 BUF_UNLOCK(bp);
6bae6177
MD
3172 }
3173 break;
3174 case NFI_COMMIT:
3175 /*
3176 * Only process buffers in need of a commit which we can
3177 * immediately lock. This may prevent a buffer from being
3178 * committed, but the normal flush loop will block on the
3179 * same buffer so we shouldn't get into an endless loop.
3180 */
6bae6177 3181 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
b1c20cfa 3182 (B_DELWRI | B_NEEDCOMMIT)) {
6bae6177
MD
3183 break;
3184 }
b1c20cfa
MD
3185 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3186 break;
984263bc 3187
b1c20cfa
MD
3188 /*
3189 * We must recheck after successfully locking the buffer.
3190 */
3191 if (bp->b_vp != info->vp ||
3192 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3193 (B_DELWRI | B_NEEDCOMMIT)) {
3194 BUF_UNLOCK(bp);
3195 break;
3196 }
6bae6177
MD
3197
3198 /*
10f3fee5
MD
3199 * NOTE: storing the bp in the bvary[] basically sets
3200 * it up for a commit operation.
3201 *
3202 * We must call vfs_busy_pages() now so the commit operation
3203 * is interlocked with user modifications to memory mapped
3204 * pages.
6bae6177
MD
3205 *
3206 * Note: to avoid loopback deadlocks, we do not
3207 * assign b_runningbufspace.
3208 */
b1c20cfa 3209 bremfree(bp);
10f3fee5
MD
3210 bp->b_cmd = BUF_CMD_WRITE;
3211 vfs_busy_pages(bp->b_vp, bp);
6bae6177 3212 info->bvary[info->bvsize] = bp;
54078292 3213 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
6bae6177
MD
3214 if (info->bvsize == 0 || toff < info->beg_off)
3215 info->beg_off = toff;
54078292 3216 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
6bae6177
MD
3217 if (info->bvsize == 0 || toff > info->end_off)
3218 info->end_off = toff;
3219 ++info->bvsize;
3220 if (info->bvsize == NFS_COMMITBVECSIZ) {
3221 error = nfs_flush_docommit(info, 0);
3222 KKASSERT(info->bvsize == 0);
984263bc 3223 }
984263bc 3224 }
6bae6177
MD
3225 return (error);
3226}
3227
3228static
3229int
3230nfs_flush_docommit(struct nfs_flush_info *info, int error)
3231{
3232 struct vnode *vp;
3233 struct buf *bp;
3234 off_t bytes;
3235 int retv;
3236 int i;
6bae6177
MD
3237
3238 vp = info->vp;
3239
3240 if (info->bvsize > 0) {
984263bc 3241 /*
3b568787
MD
3242 * Commit data on the server, as required. Note that
3243 * nfs_commit will use the vnode's cred for the commit.
6bae6177 3244 * The NFSv3 commit RPC is limited to a 32 bit byte count.
984263bc 3245 */
6bae6177
MD
3246 bytes = info->end_off - info->beg_off;
3247 if (bytes > 0x40000000)
3248 bytes = 0x40000000;
3249 if (error) {
3250 retv = -error;
3251 } else {
cc7d050e
MD
3252 retv = nfs_commitrpc_uio(vp, info->beg_off,
3253 (int)bytes, info->td);
6bae6177
MD
3254 if (retv == NFSERR_STALEWRITEVERF)
3255 nfs_clearcommit(vp->v_mount);
3256 }
984263bc
MD
3257
3258 /*
3259 * Now, either mark the blocks I/O done or mark the
3260 * blocks dirty, depending on whether the commit
3261 * succeeded.
3262 */
6bae6177
MD
3263 for (i = 0; i < info->bvsize; ++i) {
3264 bp = info->bvary[i];
3265 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
984263bc
MD
3266 if (retv) {
3267 /*
3268 * Error, leave B_DELWRI intact
3269 */
3270 vfs_unbusy_pages(bp);
10f3fee5 3271 bp->b_cmd = BUF_CMD_DONE;
984263bc
MD
3272 brelse(bp);
3273 } else {
3274 /*
3275 * Success, remove B_DELWRI ( bundirty() ).
3276 *
3277 * b_dirtyoff/b_dirtyend seem to be NFS
3278 * specific. We should probably move that
3279 * into bundirty(). XXX
81b5c339
MD
3280 *
3281 * We are faking an I/O write, we have to
3282 * start the transaction in order to
3283 * immediately biodone() it.
984263bc 3284 */
984263bc 3285 bundirty(bp);
10f3fee5 3286 bp->b_flags &= ~B_ERROR;
984263bc 3287 bp->b_dirtyoff = bp->b_dirtyend = 0;
81b5c339 3288 biodone(&bp->b_bio1);
984263bc
MD
3289 }
3290 }
6bae6177 3291 info->bvsize = 0;
984263bc 3292 }
984263bc
MD
3293 return (error);
3294}
3295
3296/*
3297 * NFS advisory byte-level locks.
3298 * Currently unsupported.
e851b29e
CP
3299 *
3300 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3301 * int a_flags)
984263bc
MD
3302 */
3303static int
e851b29e 3304nfs_advlock(struct vop_advlock_args *ap)
984263bc 3305{
40393ded 3306 struct nfsnode *np = VTONFS(ap->a_vp);
984263bc
MD
3307
3308 /*
3309 * The following kludge is to allow diskless support to work
3310 * until a real NFS lockd is implemented. Basically, just pretend
3311 * that this is a local lock.
3312 */
3313 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3314}
3315
3316/*
3317 * Print out the contents of an nfsnode.
e851b29e
CP
3318 *
3319 * nfs_print(struct vnode *a_vp)
984263bc
MD
3320 */
3321static int
e851b29e 3322nfs_print(struct vop_print_args *ap)
984263bc 3323{
40393ded
RG
3324 struct vnode *vp = ap->a_vp;
3325 struct nfsnode *np = VTONFS(vp);
984263bc 3326
50626622 3327 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
973c11b9 3328 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
984263bc
MD
3329 if (vp->v_type == VFIFO)
3330 fifo_printinfo(vp);
086c1d7e 3331 kprintf("\n");
984263bc
MD
3332 return (0);
3333}
3334
984263bc
MD
3335/*
3336 * nfs special file access vnode op.
e851b29e 3337 *
8be7edad 3338 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
984263bc
MD
3339 */
3340static int
8be7edad 3341nfs_laccess(struct vop_access_args *ap)
984263bc 3342{
984263bc 3343 struct vattr vattr;
984263bc
MD
3344 int error;
3345
527e25ed
NT
3346 error = VOP_GETATTR(ap->a_vp, &vattr);
3347 if (!error)
3348 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3349 vattr.va_mode, 0);
984263bc
MD
3350 return (error);
3351}
3352
984263bc
MD
3353/*
3354 * Read wrapper for fifos.
e851b29e
CP
3355 *
3356 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3357 * struct ucred *a_cred)
984263bc
MD
3358 */
3359static int
e851b29e 3360nfsfifo_read(struct vop_read_args *ap)
984263bc 3361{
40393ded 3362 struct nfsnode *np = VTONFS(ap->a_vp);
984263bc
MD
3363
3364 /*
3365 * Set access flag.
3366 */
3367 np->n_flag |= NACC;
3368 getnanotime(&np->n_atim);
66a1ddf5 3369 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
984263bc
MD
3370}
3371
3372/*
3373 * Write wrapper for fifos.
e851b29e
CP
3374 *
3375 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3376 * struct ucred *a_cred)
984263bc
MD
3377 */
3378static int
e851b29e 3379nfsfifo_write(struct vop_write_args *ap)
984263bc 3380{
40393ded 3381 struct nfsnode *np = VTONFS(ap->a_vp);
984263bc
MD
3382
3383 /*
3384 * Set update flag.
3385 */