kernel - NFS - fix another B_CLUSTEROK / B_NEEDCOMMIT races
[dragonfly.git] / sys / vfs / nfs / nfs_vnops.c
... / ...
CommitLineData
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 $
38 * $DragonFly: src/sys/vfs/nfs/nfs_vnops.c,v 1.80 2008/10/18 01:13:54 dillon Exp $
39 */
40
41
42/*
43 * vnode op calls for Sun NFS version 2 and 3
44 */
45
46#include "opt_inet.h"
47
48#include <sys/param.h>
49#include <sys/kernel.h>
50#include <sys/systm.h>
51#include <sys/resourcevar.h>
52#include <sys/proc.h>
53#include <sys/mount.h>
54#include <sys/buf.h>
55#include <sys/malloc.h>
56#include <sys/mbuf.h>
57#include <sys/namei.h>
58#include <sys/nlookup.h>
59#include <sys/socket.h>
60#include <sys/vnode.h>
61#include <sys/dirent.h>
62#include <sys/fcntl.h>
63#include <sys/lockf.h>
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
72#include <sys/buf2.h>
73
74#include <vfs/fifofs/fifo.h>
75#include <vfs/ufs/dir.h>
76
77#undef DIRBLKSIZ
78
79#include "rpcv2.h"
80#include "nfsproto.h"
81#include "nfs.h"
82#include "nfsmount.h"
83#include "nfsnode.h"
84#include "xdr_subs.h"
85#include "nfsm_subs.h"
86
87#include <net/if.h>
88#include <netinet/in.h>
89#include <netinet/in_var.h>
90
91#include <sys/thread2.h>
92
93/* Defs */
94#define TRUE 1
95#define FALSE 0
96
97static int nfsfifo_read (struct vop_read_args *);
98static int nfsfifo_write (struct vop_write_args *);
99static int nfsfifo_close (struct vop_close_args *);
100#define nfs_poll vop_nopoll
101static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
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 *);
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 *);
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 *);
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 *);
125static int nfs_laccess (struct vop_access_args *);
126static int nfs_readlink (struct vop_readlink_args *);
127static int nfs_print (struct vop_print_args *);
128static int nfs_advlock (struct vop_advlock_args *);
129
130static int nfs_nresolve (struct vop_nresolve_args *);
131/*
132 * Global vfs data structures for nfs
133 */
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 = vop_stdgetpages,
144 .vop_putpages = vop_stdputpages,
145 .vop_inactive = nfs_inactive,
146 .vop_old_link = nfs_link,
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,
164 .vop_write = nfs_write,
165 .vop_nresolve = nfs_nresolve
166};
167
168/*
169 * Special device vnode ops
170 */
171struct vop_ops nfsv2_spec_vops = {
172 .vop_default = vop_defaultop,
173 .vop_access = nfs_laccess,
174 .vop_close = nfs_close,
175 .vop_fsync = nfs_fsync,
176 .vop_getattr = nfs_getattr,
177 .vop_inactive = nfs_inactive,
178 .vop_print = nfs_print,
179 .vop_read = vop_stdnoread,
180 .vop_reclaim = nfs_reclaim,
181 .vop_setattr = nfs_setattr,
182 .vop_write = vop_stdnowrite
183};
184
185struct vop_ops nfsv2_fifo_vops = {
186 .vop_default = fifo_vnoperate,
187 .vop_access = nfs_laccess,
188 .vop_close = nfsfifo_close,
189 .vop_fsync = nfs_fsync,
190 .vop_getattr = nfs_getattr,
191 .vop_inactive = nfs_inactive,
192 .vop_print = nfs_print,
193 .vop_read = nfsfifo_read,
194 .vop_reclaim = nfs_reclaim,
195 .vop_setattr = nfs_setattr,
196 .vop_write = nfsfifo_write
197};
198
199static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
200 struct componentname *cnp,
201 struct vattr *vap);
202static int nfs_removerpc (struct vnode *dvp, const char *name,
203 int namelen,
204 struct ucred *cred, struct thread *td);
205static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
206 int fnamelen, struct vnode *tdvp,
207 const char *tnameptr, int tnamelen,
208 struct ucred *cred, struct thread *td);
209static int nfs_renameit (struct vnode *sdvp,
210 struct componentname *scnp,
211 struct sillyrename *sp);
212
213SYSCTL_DECL(_vfs_nfs);
214
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");
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");
221
222static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
223SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
224 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
225
226static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
227SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
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");
233
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
249nfs3_access_otw(struct vnode *vp, int wmode,
250 struct thread *td, struct ucred *cred)
251{
252 struct nfsnode *np = VTONFS(vp);
253 int attrflag;
254 int error = 0;
255 u_int32_t *tl;
256 u_int32_t rmode;
257 struct nfsm_info info;
258
259 info.mrep = NULL;
260 info.v3 = 1;
261
262 nfsstats.rpccnt[NFSPROC_ACCESS]++;
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);
267 *tl = txdr_unsigned(wmode);
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));
272 rmode = fxdr_unsigned(u_int32_t, *tl);
273 np->n_mode = rmode;
274 np->n_modeuid = cred->cr_uid;
275 np->n_modestamp = mycpu->gd_time_seconds;
276 }
277 m_freem(info.mrep);
278 info.mrep = NULL;
279nfsmout:
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.
288 *
289 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
290 */
291static int
292nfs_access(struct vop_access_args *ap)
293{
294 struct ucred *cred;
295 struct vnode *vp = ap->a_vp;
296 thread_t td = curthread;
297 int error = 0;
298 u_int32_t mode, wmode;
299 struct nfsnode *np = VTONFS(vp);
300 int v3 = NFS_ISV3(vp);
301
302 /*
303 * Disallow write attempts on filesystems mounted read-only;
304 * unless the file is a socket, fifo, or a block or character
305 * device resident on the filesystem.
306 */
307 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
308 switch (vp->v_type) {
309 case VREG:
310 case VDIR:
311 case VLNK:
312 return (EROFS);
313 default:
314 break;
315 }
316 }
317
318 /*
319 * The NFS protocol passes only the effective uid/gid over the wire but
320 * we need to check access against real ids if AT_EACCESS not set.
321 * Handle this case by cloning the credentials and setting the
322 * effective ids to the real ones.
323 */
324 if (ap->a_flags & AT_EACCESS) {
325 cred = crhold(ap->a_cred);
326 } else {
327 cred = crdup(ap->a_cred);
328 cred->cr_uid = cred->cr_ruid;
329 cred->cr_gid = cred->cr_rgid;
330 }
331
332 /*
333 * For nfs v3, check to see if we have done this recently, and if
334 * so return our cached result instead of making an ACCESS call.
335 * If not, do an access rpc, otherwise you are stuck emulating
336 * ufs_access() locally using the vattr. This may not be correct,
337 * since the server may apply other access criteria such as
338 * client uid-->server uid mapping that we do not know about.
339 */
340 if (v3) {
341 if (ap->a_mode & VREAD)
342 mode = NFSV3ACCESS_READ;
343 else
344 mode = 0;
345 if (vp->v_type != VDIR) {
346 if (ap->a_mode & VWRITE)
347 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
348 if (ap->a_mode & VEXEC)
349 mode |= NFSV3ACCESS_EXECUTE;
350 } else {
351 if (ap->a_mode & VWRITE)
352 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
353 NFSV3ACCESS_DELETE);
354 if (ap->a_mode & VEXEC)
355 mode |= NFSV3ACCESS_LOOKUP;
356 }
357 /* XXX safety belt, only make blanket request if caching */
358 if (nfsaccess_cache_timeout > 0) {
359 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
360 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
361 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
362 } else {
363 wmode = mode;
364 }
365
366 /*
367 * Does our cached result allow us to give a definite yes to
368 * this request?
369 */
370 if (np->n_modestamp &&
371 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
372 (cred->cr_uid == np->n_modeuid) &&
373 ((np->n_mode & mode) == mode)) {
374 nfsstats.accesscache_hits++;
375 } else {
376 /*
377 * Either a no, or a don't know. Go to the wire.
378 */
379 nfsstats.accesscache_misses++;
380 error = nfs3_access_otw(vp, wmode, td, cred);
381 if (!error) {
382 if ((np->n_mode & mode) != mode) {
383 error = EACCES;
384 }
385 }
386 }
387 } else {
388 if ((error = nfs_laccess(ap)) != 0) {
389 crfree(cred);
390 return (error);
391 }
392
393 /*
394 * Attempt to prevent a mapped root from accessing a file
395 * which it shouldn't. We try to read a byte from the file
396 * if the user is root and the file is not zero length.
397 * After calling nfs_laccess, we should have the correct
398 * file size cached.
399 */
400 if (cred->cr_uid == 0 && (ap->a_mode & VREAD)
401 && VTONFS(vp)->n_size > 0) {
402 struct iovec aiov;
403 struct uio auio;
404 char buf[1];
405
406 aiov.iov_base = buf;
407 aiov.iov_len = 1;
408 auio.uio_iov = &aiov;
409 auio.uio_iovcnt = 1;
410 auio.uio_offset = 0;
411 auio.uio_resid = 1;
412 auio.uio_segflg = UIO_SYSSPACE;
413 auio.uio_rw = UIO_READ;
414 auio.uio_td = td;
415
416 if (vp->v_type == VREG) {
417 error = nfs_readrpc_uio(vp, &auio);
418 } else if (vp->v_type == VDIR) {
419 char* bp;
420 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
421 aiov.iov_base = bp;
422 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
423 error = nfs_readdirrpc_uio(vp, &auio);
424 kfree(bp, M_TEMP);
425 } else if (vp->v_type == VLNK) {
426 error = nfs_readlinkrpc_uio(vp, &auio);
427 } else {
428 error = EACCES;
429 }
430 }
431 }
432 /*
433 * [re]record creds for reading and/or writing if access
434 * was granted. Assume the NFS server will grant read access
435 * for execute requests.
436 */
437 if (error == 0) {
438 if ((ap->a_mode & (VREAD|VEXEC)) && cred != np->n_rucred) {
439 crhold(cred);
440 if (np->n_rucred)
441 crfree(np->n_rucred);
442 np->n_rucred = cred;
443 }
444 if ((ap->a_mode & VWRITE) && cred != np->n_wucred) {
445 crhold(cred);
446 if (np->n_wucred)
447 crfree(np->n_wucred);
448 np->n_wucred = cred;
449 }
450 }
451 crfree(cred);
452 return(error);
453}
454
455/*
456 * nfs open vnode op
457 * Check to see if the type is ok
458 * and that deletion is not in progress.
459 * For paged in text files, you will need to flush the page cache
460 * if consistency is lost.
461 *
462 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
463 * struct file *a_fp)
464 */
465/* ARGSUSED */
466static int
467nfs_open(struct vop_open_args *ap)
468{
469 struct vnode *vp = ap->a_vp;
470 struct nfsnode *np = VTONFS(vp);
471 struct vattr vattr;
472 int error;
473
474 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
475#ifdef DIAGNOSTIC
476 kprintf("open eacces vtyp=%d\n",vp->v_type);
477#endif
478 return (EOPNOTSUPP);
479 }
480
481 /*
482 * Save valid creds for reading and writing for later RPCs.
483 */
484 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
485 crhold(ap->a_cred);
486 if (np->n_rucred)
487 crfree(np->n_rucred);
488 np->n_rucred = ap->a_cred;
489 }
490 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
491 crhold(ap->a_cred);
492 if (np->n_wucred)
493 crfree(np->n_wucred);
494 np->n_wucred = ap->a_cred;
495 }
496
497 /*
498 * Clear the attribute cache only if opening with write access. It
499 * is unclear if we should do this at all here, but we certainly
500 * should not clear the cache unconditionally simply because a file
501 * is being opened.
502 */
503 if (ap->a_mode & FWRITE)
504 np->n_attrstamp = 0;
505
506 /*
507 * For normal NFS, reconcile changes made locally verses
508 * changes made remotely. Note that VOP_GETATTR only goes
509 * to the wire if the cached attribute has timed out or been
510 * cleared.
511 *
512 * If local modifications have been made clear the attribute
513 * cache to force an attribute and modified time check. If
514 * GETATTR detects that the file has been changed by someone
515 * other then us it will set NRMODIFIED.
516 *
517 * If we are opening a directory and local changes have been
518 * made we have to invalidate the cache in order to ensure
519 * that we get the most up-to-date information from the
520 * server. XXX
521 */
522 if (np->n_flag & NLMODIFIED) {
523 np->n_attrstamp = 0;
524 if (vp->v_type == VDIR) {
525 error = nfs_vinvalbuf(vp, V_SAVE, 1);
526 if (error == EINTR)
527 return (error);
528 nfs_invaldir(vp);
529 }
530 }
531 error = VOP_GETATTR(vp, &vattr);
532 if (error)
533 return (error);
534 if (np->n_flag & NRMODIFIED) {
535 if (vp->v_type == VDIR)
536 nfs_invaldir(vp);
537 error = nfs_vinvalbuf(vp, V_SAVE, 1);
538 if (error == EINTR)
539 return (error);
540 np->n_flag &= ~NRMODIFIED;
541 }
542
543 return (vop_stdopen(ap));
544}
545
546/*
547 * nfs close vnode op
548 * What an NFS client should do upon close after writing is a debatable issue.
549 * Most NFS clients push delayed writes to the server upon close, basically for
550 * two reasons:
551 * 1 - So that any write errors may be reported back to the client process
552 * doing the close system call. By far the two most likely errors are
553 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
554 * 2 - To put a worst case upper bound on cache inconsistency between
555 * multiple clients for the file.
556 * There is also a consistency problem for Version 2 of the protocol w.r.t.
557 * not being able to tell if other clients are writing a file concurrently,
558 * since there is no way of knowing if the changed modify time in the reply
559 * is only due to the write for this client.
560 * (NFS Version 3 provides weak cache consistency data in the reply that
561 * should be sufficient to detect and handle this case.)
562 *
563 * The current code does the following:
564 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
565 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
566 * or commit them (this satisfies 1 and 2 except for the
567 * case where the server crashes after this close but
568 * before the commit RPC, which is felt to be "good
569 * enough". Changing the last argument to nfs_flush() to
570 * a 1 would force a commit operation, if it is felt a
571 * commit is necessary now.
572 * for NQNFS - do nothing now, since 2 is dealt with via leases and
573 * 1 should be dealt with via an fsync() system call for
574 * cases where write errors are important.
575 *
576 * nfs_close(struct vnode *a_vp, int a_fflag)
577 */
578/* ARGSUSED */
579static int
580nfs_close(struct vop_close_args *ap)
581{
582 struct vnode *vp = ap->a_vp;
583 struct nfsnode *np = VTONFS(vp);
584 int error = 0;
585 thread_t td = curthread;
586
587 if (vp->v_type == VREG) {
588 if (np->n_flag & NLMODIFIED) {
589 if (NFS_ISV3(vp)) {
590 /*
591 * Under NFSv3 we have dirty buffers to dispose of. We
592 * must flush them to the NFS server. We have the option
593 * of waiting all the way through the commit rpc or just
594 * waiting for the initial write. The default is to only
595 * wait through the initial write so the data is in the
596 * server's cache, which is roughly similar to the state
597 * a standard disk subsystem leaves the file in on close().
598 *
599 * We cannot clear the NLMODIFIED bit in np->n_flag due to
600 * potential races with other processes, and certainly
601 * cannot clear it if we don't commit.
602 */
603 int cm = nfsv3_commit_on_close ? 1 : 0;
604 error = nfs_flush(vp, MNT_WAIT, td, cm);
605 /* np->n_flag &= ~NLMODIFIED; */
606 } else {
607 error = nfs_vinvalbuf(vp, V_SAVE, 1);
608 }
609 np->n_attrstamp = 0;
610 }
611 if (np->n_flag & NWRITEERR) {
612 np->n_flag &= ~NWRITEERR;
613 error = np->n_error;
614 }
615 }
616 vop_stdclose(ap);
617 return (error);
618}
619
620/*
621 * nfs getattr call from vfs.
622 *
623 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
624 */
625static int
626nfs_getattr(struct vop_getattr_args *ap)
627{
628 struct vnode *vp = ap->a_vp;
629 struct nfsnode *np = VTONFS(vp);
630 int error = 0;
631 thread_t td = curthread;
632 struct nfsm_info info;
633
634 info.mrep = NULL;
635 info.v3 = NFS_ISV3(vp);
636
637 /*
638 * Update local times for special files.
639 */
640 if (np->n_flag & (NACC | NUPD))
641 np->n_flag |= NCHG;
642 /*
643 * First look in the cache.
644 */
645 if (nfs_getattrcache(vp, ap->a_vap) == 0)
646 return (0);
647
648 if (info.v3 && nfsaccess_cache_timeout > 0) {
649 nfsstats.accesscache_misses++;
650 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
651 if (nfs_getattrcache(vp, ap->a_vap) == 0)
652 return (0);
653 }
654
655 nfsstats.rpccnt[NFSPROC_GETATTR]++;
656 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
657 ERROROUT(nfsm_fhtom(&info, vp));
658 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
659 nfs_vpcred(vp, ND_CHECK), &error));
660 if (error == 0) {
661 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
662 }
663 m_freem(info.mrep);
664 info.mrep = NULL;
665nfsmout:
666 return (error);
667}
668
669/*
670 * nfs setattr call.
671 *
672 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
673 */
674static int
675nfs_setattr(struct vop_setattr_args *ap)
676{
677 struct vnode *vp = ap->a_vp;
678 struct nfsnode *np = VTONFS(vp);
679 struct vattr *vap = ap->a_vap;
680 int biosize = vp->v_mount->mnt_stat.f_iosize;
681 int error = 0;
682 int boff;
683 off_t tsize;
684 thread_t td = curthread;
685
686#ifndef nolint
687 tsize = (off_t)0;
688#endif
689
690 /*
691 * Setting of flags is not supported.
692 */
693 if (vap->va_flags != VNOVAL)
694 return (EOPNOTSUPP);
695
696 /*
697 * Disallow write attempts if the filesystem is mounted read-only.
698 */
699 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
700 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
701 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
702 (vp->v_mount->mnt_flag & MNT_RDONLY))
703 return (EROFS);
704
705 if (vap->va_size != VNOVAL) {
706 /*
707 * truncation requested
708 */
709 switch (vp->v_type) {
710 case VDIR:
711 return (EISDIR);
712 case VCHR:
713 case VBLK:
714 case VSOCK:
715 case VFIFO:
716 if (vap->va_mtime.tv_sec == VNOVAL &&
717 vap->va_atime.tv_sec == VNOVAL &&
718 vap->va_mode == (mode_t)VNOVAL &&
719 vap->va_uid == (uid_t)VNOVAL &&
720 vap->va_gid == (gid_t)VNOVAL)
721 return (0);
722 vap->va_size = VNOVAL;
723 break;
724 default:
725 /*
726 * Disallow write attempts if the filesystem is
727 * mounted read-only.
728 */
729 if (vp->v_mount->mnt_flag & MNT_RDONLY)
730 return (EROFS);
731
732 tsize = np->n_size;
733again:
734 boff = (int)vap->va_size & (biosize - 1);
735 error = nfs_meta_setsize(vp, td, vap->va_size, 0);
736
737#if 0
738 if (np->n_flag & NLMODIFIED) {
739 if (vap->va_size == 0)
740 error = nfs_vinvalbuf(vp, 0, 1);
741 else
742 error = nfs_vinvalbuf(vp, V_SAVE, 1);
743 }
744#endif
745 /*
746 * note: this loop case almost always happens at
747 * least once per truncation.
748 */
749 if (error == 0 && np->n_size != vap->va_size)
750 goto again;
751 np->n_vattr.va_size = vap->va_size;
752 break;
753 }
754 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
755 /*
756 * What to do. If we are modifying the mtime we lose
757 * mtime detection of changes made by the server or other
758 * clients. But programs like rsync/rdist/cpdup are going
759 * to call utimes a lot. We don't want to piecemeal sync.
760 *
761 * For now sync if any prior remote changes were detected,
762 * but allow us to lose track of remote changes made during
763 * the utimes operation.
764 */
765 if (np->n_flag & NRMODIFIED)
766 error = nfs_vinvalbuf(vp, V_SAVE, 1);
767 if (error == EINTR)
768 return (error);
769 if (error == 0) {
770 if (vap->va_mtime.tv_sec != VNOVAL) {
771 np->n_mtime = vap->va_mtime.tv_sec;
772 }
773 }
774 }
775 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
776
777 /*
778 * Sanity check if a truncation was issued. This should only occur
779 * if multiple processes are racing on the same file.
780 */
781 if (error == 0 && vap->va_size != VNOVAL &&
782 np->n_size != vap->va_size) {
783 kprintf("NFS ftruncate: server disagrees on the file size: "
784 "%jd/%jd/%jd\n",
785 (intmax_t)tsize,
786 (intmax_t)vap->va_size,
787 (intmax_t)np->n_size);
788 goto again;
789 }
790 if (error && vap->va_size != VNOVAL) {
791 np->n_size = np->n_vattr.va_size = tsize;
792 nfs_meta_setsize(vp, td, np->n_size, 0);
793 }
794 return (error);
795}
796
797/*
798 * Do an nfs setattr rpc.
799 */
800static int
801nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
802 struct ucred *cred, struct thread *td)
803{
804 struct nfsv2_sattr *sp;
805 struct nfsnode *np = VTONFS(vp);
806 u_int32_t *tl;
807 int error = 0, wccflag = NFSV3_WCCRATTR;
808 struct nfsm_info info;
809
810 info.mrep = NULL;
811 info.v3 = NFS_ISV3(vp);
812
813 nfsstats.rpccnt[NFSPROC_SETATTR]++;
814 nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
815 NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
816 ERROROUT(nfsm_fhtom(&info, vp));
817 if (info.v3) {
818 nfsm_v3attrbuild(&info, vap, TRUE);
819 tl = nfsm_build(&info, NFSX_UNSIGNED);
820 *tl = nfs_false;
821 } else {
822 sp = nfsm_build(&info, NFSX_V2SATTR);
823 if (vap->va_mode == (mode_t)VNOVAL)
824 sp->sa_mode = nfs_xdrneg1;
825 else
826 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
827 if (vap->va_uid == (uid_t)VNOVAL)
828 sp->sa_uid = nfs_xdrneg1;
829 else
830 sp->sa_uid = txdr_unsigned(vap->va_uid);
831 if (vap->va_gid == (gid_t)VNOVAL)
832 sp->sa_gid = nfs_xdrneg1;
833 else
834 sp->sa_gid = txdr_unsigned(vap->va_gid);
835 sp->sa_size = txdr_unsigned(vap->va_size);
836 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
837 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
838 }
839 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
840 if (info.v3) {
841 np->n_modestamp = 0;
842 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
843 } else {
844 ERROROUT(nfsm_loadattr(&info, vp, NULL));
845 }
846 m_freem(info.mrep);
847 info.mrep = NULL;
848nfsmout:
849 return (error);
850}
851
852static
853void
854nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
855{
856 if (nctimeout == 0)
857 nctimeout = 1;
858 else
859 nctimeout *= hz;
860 cache_setvp(nch, vp);
861 cache_settimeout(nch, nctimeout);
862}
863
864/*
865 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
866 * nfs_lookup() until all remaining new api calls are implemented.
867 *
868 * Resolve a namecache entry. This function is passed a locked ncp and
869 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
870 */
871static int
872nfs_nresolve(struct vop_nresolve_args *ap)
873{
874 struct thread *td = curthread;
875 struct namecache *ncp;
876 struct ucred *cred;
877 struct nfsnode *np;
878 struct vnode *dvp;
879 struct vnode *nvp;
880 nfsfh_t *fhp;
881 int attrflag;
882 int fhsize;
883 int error;
884 int tmp_error;
885 int len;
886 struct nfsm_info info;
887
888 cred = ap->a_cred;
889 dvp = ap->a_dvp;
890
891 if ((error = vget(dvp, LK_SHARED)) != 0)
892 return (error);
893
894 info.mrep = NULL;
895 info.v3 = NFS_ISV3(dvp);
896
897 nvp = NULL;
898 nfsstats.lookupcache_misses++;
899 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
900 ncp = ap->a_nch->ncp;
901 len = ncp->nc_nlen;
902 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
903 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
904 ERROROUT(nfsm_fhtom(&info, dvp));
905 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
906 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
907 ap->a_cred, &error));
908 if (error) {
909 /*
910 * Cache negatve lookups to reduce NFS traffic, but use
911 * a fast timeout. Otherwise use a timeout of 1 tick.
912 * XXX we should add a namecache flag for no-caching
913 * to uncache the negative hit as soon as possible, but
914 * we cannot simply destroy the entry because it is used
915 * as a placeholder by the caller.
916 *
917 * The refactored nfs code will overwrite a non-zero error
918 * with 0 when we use ERROROUT(), so don't here.
919 */
920 if (error == ENOENT)
921 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
922 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
923 NFS_LATTR_NOSHRINK);
924 if (tmp_error) {
925 error = tmp_error;
926 goto nfsmout;
927 }
928 m_freem(info.mrep);
929 info.mrep = NULL;
930 goto nfsmout;
931 }
932
933 /*
934 * Success, get the file handle, do various checks, and load
935 * post-operation data from the reply packet. Theoretically
936 * we should never be looking up "." so, theoretically, we
937 * should never get the same file handle as our directory. But
938 * we check anyway. XXX
939 *
940 * Note that no timeout is set for the positive cache hit. We
941 * assume, theoretically, that ESTALE returns will be dealt with
942 * properly to handle NFS races and in anycase we cannot depend
943 * on a timeout to deal with NFS open/create/excl issues so instead
944 * of a bad hack here the rest of the NFS client code needs to do
945 * the right thing.
946 */
947 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
948
949 np = VTONFS(dvp);
950 if (NFS_CMPFH(np, fhp, fhsize)) {
951 vref(dvp);
952 nvp = dvp;
953 } else {
954 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
955 if (error) {
956 m_freem(info.mrep);
957 info.mrep = NULL;
958 vput(dvp);
959 return (error);
960 }
961 nvp = NFSTOV(np);
962 }
963 if (info.v3) {
964 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
965 NFS_LATTR_NOSHRINK));
966 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
967 NFS_LATTR_NOSHRINK));
968 } else {
969 ERROROUT(nfsm_loadattr(&info, nvp, NULL));
970 }
971 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
972 m_freem(info.mrep);
973 info.mrep = NULL;
974nfsmout:
975 vput(dvp);
976 if (nvp) {
977 if (nvp == dvp)
978 vrele(nvp);
979 else
980 vput(nvp);
981 }
982 return (error);
983}
984
985/*
986 * 'cached' nfs directory lookup
987 *
988 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
989 *
990 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
991 * struct componentname *a_cnp)
992 */
993static int
994nfs_lookup(struct vop_old_lookup_args *ap)
995{
996 struct componentname *cnp = ap->a_cnp;
997 struct vnode *dvp = ap->a_dvp;
998 struct vnode **vpp = ap->a_vpp;
999 int flags = cnp->cn_flags;
1000 struct vnode *newvp;
1001 struct nfsmount *nmp;
1002 long len;
1003 nfsfh_t *fhp;
1004 struct nfsnode *np;
1005 int lockparent, wantparent, attrflag, fhsize;
1006 int error;
1007 int tmp_error;
1008 struct nfsm_info info;
1009
1010 info.mrep = NULL;
1011 info.v3 = NFS_ISV3(dvp);
1012 error = 0;
1013
1014 /*
1015 * Read-only mount check and directory check.
1016 */
1017 *vpp = NULLVP;
1018 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1019 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1020 return (EROFS);
1021
1022 if (dvp->v_type != VDIR)
1023 return (ENOTDIR);
1024
1025 /*
1026 * Look it up in the cache. Note that ENOENT is only returned if we
1027 * previously entered a negative hit (see later on). The additional
1028 * nfsneg_cache_timeout check causes previously cached results to
1029 * be instantly ignored if the negative caching is turned off.
1030 */
1031 lockparent = flags & CNP_LOCKPARENT;
1032 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1033 nmp = VFSTONFS(dvp->v_mount);
1034 np = VTONFS(dvp);
1035
1036 /*
1037 * Go to the wire.
1038 */
1039 error = 0;
1040 newvp = NULLVP;
1041 nfsstats.lookupcache_misses++;
1042 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1043 len = cnp->cn_namelen;
1044 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1045 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1046 ERROROUT(nfsm_fhtom(&info, dvp));
1047 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1048 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1049 cnp->cn_cred, &error));
1050 if (error) {
1051 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1052 NFS_LATTR_NOSHRINK);
1053 if (tmp_error) {
1054 error = tmp_error;
1055 goto nfsmout;
1056 }
1057
1058 m_freem(info.mrep);
1059 info.mrep = NULL;
1060 goto nfsmout;
1061 }
1062 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1063
1064 /*
1065 * Handle RENAME case...
1066 */
1067 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1068 if (NFS_CMPFH(np, fhp, fhsize)) {
1069 m_freem(info.mrep);
1070 info.mrep = NULL;
1071 return (EISDIR);
1072 }
1073 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1074 if (error) {
1075 m_freem(info.mrep);
1076 info.mrep = NULL;
1077 return (error);
1078 }
1079 newvp = NFSTOV(np);
1080 if (info.v3) {
1081 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1082 NFS_LATTR_NOSHRINK));
1083 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1084 NFS_LATTR_NOSHRINK));
1085 } else {
1086 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1087 }
1088 *vpp = newvp;
1089 m_freem(info.mrep);
1090 info.mrep = NULL;
1091 if (!lockparent) {
1092 vn_unlock(dvp);
1093 cnp->cn_flags |= CNP_PDIRUNLOCK;
1094 }
1095 return (0);
1096 }
1097
1098 if (flags & CNP_ISDOTDOT) {
1099 vn_unlock(dvp);
1100 cnp->cn_flags |= CNP_PDIRUNLOCK;
1101 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1102 if (error) {
1103 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1104 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1105 return (error); /* NOTE: return error from nget */
1106 }
1107 newvp = NFSTOV(np);
1108 if (lockparent) {
1109 error = vn_lock(dvp, LK_EXCLUSIVE);
1110 if (error) {
1111 vput(newvp);
1112 return (error);
1113 }
1114 cnp->cn_flags |= CNP_PDIRUNLOCK;
1115 }
1116 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1117 vref(dvp);
1118 newvp = dvp;
1119 } else {
1120 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1121 if (error) {
1122 m_freem(info.mrep);
1123 info.mrep = NULL;
1124 return (error);
1125 }
1126 if (!lockparent) {
1127 vn_unlock(dvp);
1128 cnp->cn_flags |= CNP_PDIRUNLOCK;
1129 }
1130 newvp = NFSTOV(np);
1131 }
1132 if (info.v3) {
1133 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1134 NFS_LATTR_NOSHRINK));
1135 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1136 NFS_LATTR_NOSHRINK));
1137 } else {
1138 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1139 }
1140#if 0
1141 /* XXX MOVE TO nfs_nremove() */
1142 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1143 cnp->cn_nameiop != NAMEI_DELETE) {
1144 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1145 }
1146#endif
1147 *vpp = newvp;
1148 m_freem(info.mrep);
1149 info.mrep = NULL;
1150nfsmout:
1151 if (error) {
1152 if (newvp != NULLVP) {
1153 vrele(newvp);
1154 *vpp = NULLVP;
1155 }
1156 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1157 cnp->cn_nameiop == NAMEI_RENAME) &&
1158 error == ENOENT) {
1159 if (!lockparent) {
1160 vn_unlock(dvp);
1161 cnp->cn_flags |= CNP_PDIRUNLOCK;
1162 }
1163 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1164 error = EROFS;
1165 else
1166 error = EJUSTRETURN;
1167 }
1168 }
1169 return (error);
1170}
1171
1172/*
1173 * nfs read call.
1174 * Just call nfs_bioread() to do the work.
1175 *
1176 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1177 * struct ucred *a_cred)
1178 */
1179static int
1180nfs_read(struct vop_read_args *ap)
1181{
1182 struct vnode *vp = ap->a_vp;
1183
1184 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1185}
1186
1187/*
1188 * nfs readlink call
1189 *
1190 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1191 */
1192static int
1193nfs_readlink(struct vop_readlink_args *ap)
1194{
1195 struct vnode *vp = ap->a_vp;
1196
1197 if (vp->v_type != VLNK)
1198 return (EINVAL);
1199 return (nfs_bioread(vp, ap->a_uio, 0));
1200}
1201
1202/*
1203 * Do a readlink rpc.
1204 * Called by nfs_doio() from below the buffer cache.
1205 */
1206int
1207nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1208{
1209 int error = 0, len, attrflag;
1210 struct nfsm_info info;
1211
1212 info.mrep = NULL;
1213 info.v3 = NFS_ISV3(vp);
1214
1215 nfsstats.rpccnt[NFSPROC_READLINK]++;
1216 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1217 ERROROUT(nfsm_fhtom(&info, vp));
1218 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1219 nfs_vpcred(vp, ND_CHECK), &error));
1220 if (info.v3) {
1221 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1222 NFS_LATTR_NOSHRINK));
1223 }
1224 if (!error) {
1225 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1226 if (len == NFS_MAXPATHLEN) {
1227 struct nfsnode *np = VTONFS(vp);
1228 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1229 len = np->n_size;
1230 }
1231 ERROROUT(nfsm_mtouio(&info, uiop, len));
1232 }
1233 m_freem(info.mrep);
1234 info.mrep = NULL;
1235nfsmout:
1236 return (error);
1237}
1238
1239/*
1240 * nfs synchronous read rpc using UIO
1241 */
1242int
1243nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1244{
1245 u_int32_t *tl;
1246 struct nfsmount *nmp;
1247 int error = 0, len, retlen, tsiz, eof, attrflag;
1248 struct nfsm_info info;
1249 off_t tmp_off;
1250
1251 info.mrep = NULL;
1252 info.v3 = NFS_ISV3(vp);
1253
1254#ifndef nolint
1255 eof = 0;
1256#endif
1257 nmp = VFSTONFS(vp->v_mount);
1258 tsiz = uiop->uio_resid;
1259 tmp_off = uiop->uio_offset + tsiz;
1260 if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1261 return (EFBIG);
1262 tmp_off = uiop->uio_offset;
1263 while (tsiz > 0) {
1264 nfsstats.rpccnt[NFSPROC_READ]++;
1265 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1266 nfsm_reqhead(&info, vp, NFSPROC_READ,
1267 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1268 ERROROUT(nfsm_fhtom(&info, vp));
1269 tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1270 if (info.v3) {
1271 txdr_hyper(uiop->uio_offset, tl);
1272 *(tl + 2) = txdr_unsigned(len);
1273 } else {
1274 *tl++ = txdr_unsigned(uiop->uio_offset);
1275 *tl++ = txdr_unsigned(len);
1276 *tl = 0;
1277 }
1278 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1279 nfs_vpcred(vp, ND_READ), &error));
1280 if (info.v3) {
1281 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1282 NFS_LATTR_NOSHRINK));
1283 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1284 eof = fxdr_unsigned(int, *(tl + 1));
1285 } else {
1286 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1287 }
1288 NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1289 ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1290 m_freem(info.mrep);
1291 info.mrep = NULL;
1292
1293 /*
1294 * Handle short-read from server (NFSv3). If EOF is not
1295 * flagged (and no error occurred), but retlen is less
1296 * then the request size, we must zero-fill the remainder.
1297 */
1298 if (retlen < len && info.v3 && eof == 0) {
1299 ERROROUT(uiomovez(len - retlen, uiop));
1300 retlen = len;
1301 }
1302 tsiz -= retlen;
1303
1304 /*
1305 * Terminate loop on EOF or zero-length read.
1306 *
1307 * For NFSv2 a short-read indicates EOF, not zero-fill,
1308 * and also terminates the loop.
1309 */
1310 if (info.v3) {
1311 if (eof || retlen == 0)
1312 tsiz = 0;
1313 } else if (retlen < len) {
1314 tsiz = 0;
1315 }
1316 }
1317nfsmout:
1318 return (error);
1319}
1320
1321/*
1322 * nfs write call
1323 */
1324int
1325nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1326 int *iomode, int *must_commit)
1327{
1328 u_int32_t *tl;
1329 int32_t backup;
1330 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1331 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1332 int committed = NFSV3WRITE_FILESYNC;
1333 struct nfsm_info info;
1334
1335 info.mrep = NULL;
1336 info.v3 = NFS_ISV3(vp);
1337
1338#ifndef DIAGNOSTIC
1339 if (uiop->uio_iovcnt != 1)
1340 panic("nfs: writerpc iovcnt > 1");
1341#endif
1342 *must_commit = 0;
1343 tsiz = uiop->uio_resid;
1344 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1345 return (EFBIG);
1346 while (tsiz > 0) {
1347 nfsstats.rpccnt[NFSPROC_WRITE]++;
1348 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1349 nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1350 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1351 ERROROUT(nfsm_fhtom(&info, vp));
1352 if (info.v3) {
1353 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1354 txdr_hyper(uiop->uio_offset, tl);
1355 tl += 2;
1356 *tl++ = txdr_unsigned(len);
1357 *tl++ = txdr_unsigned(*iomode);
1358 *tl = txdr_unsigned(len);
1359 } else {
1360 u_int32_t x;
1361
1362 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1363 /* Set both "begin" and "current" to non-garbage. */
1364 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1365 *tl++ = x; /* "begin offset" */
1366 *tl++ = x; /* "current offset" */
1367 x = txdr_unsigned(len);
1368 *tl++ = x; /* total to this offset */
1369 *tl = x; /* size of this write */
1370 }
1371 ERROROUT(nfsm_uiotom(&info, uiop, len));
1372 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1373 nfs_vpcred(vp, ND_WRITE), &error));
1374 if (info.v3) {
1375 /*
1376 * The write RPC returns a before and after mtime. The
1377 * nfsm_wcc_data() macro checks the before n_mtime
1378 * against the before time and stores the after time
1379 * in the nfsnode's cached vattr and n_mtime field.
1380 * The NRMODIFIED bit will be set if the before
1381 * time did not match the original mtime.
1382 */
1383 wccflag = NFSV3_WCCCHK;
1384 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1385 if (error == 0) {
1386 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1387 rlen = fxdr_unsigned(int, *tl++);
1388 if (rlen == 0) {
1389 error = NFSERR_IO;
1390 m_freem(info.mrep);
1391 info.mrep = NULL;
1392 break;
1393 } else if (rlen < len) {
1394 backup = len - rlen;
1395 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1396 uiop->uio_iov->iov_len += backup;
1397 uiop->uio_offset -= backup;
1398 uiop->uio_resid += backup;
1399 len = rlen;
1400 }
1401 commit = fxdr_unsigned(int, *tl++);
1402
1403 /*
1404 * Return the lowest committment level
1405 * obtained by any of the RPCs.
1406 */
1407 if (committed == NFSV3WRITE_FILESYNC)
1408 committed = commit;
1409 else if (committed == NFSV3WRITE_DATASYNC &&
1410 commit == NFSV3WRITE_UNSTABLE)
1411 committed = commit;
1412 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1413 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1414 NFSX_V3WRITEVERF);
1415 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1416 } else if (bcmp((caddr_t)tl,
1417 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1418 *must_commit = 1;
1419 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1420 NFSX_V3WRITEVERF);
1421 }
1422 }
1423 } else {
1424 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1425 }
1426 m_freem(info.mrep);
1427 info.mrep = NULL;
1428 if (error)
1429 break;
1430 tsiz -= len;
1431 }
1432nfsmout:
1433 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1434 committed = NFSV3WRITE_FILESYNC;
1435 *iomode = committed;
1436 if (error)
1437 uiop->uio_resid = tsiz;
1438 return (error);
1439}
1440
1441/*
1442 * nfs mknod rpc
1443 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1444 * mode set to specify the file type and the size field for rdev.
1445 */
1446static int
1447nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1448 struct vattr *vap)
1449{
1450 struct nfsv2_sattr *sp;
1451 u_int32_t *tl;
1452 struct vnode *newvp = NULL;
1453 struct nfsnode *np = NULL;
1454 struct vattr vattr;
1455 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1456 int rmajor, rminor;
1457 struct nfsm_info info;
1458
1459 info.mrep = NULL;
1460 info.v3 = NFS_ISV3(dvp);
1461
1462 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1463 rmajor = txdr_unsigned(vap->va_rmajor);
1464 rminor = txdr_unsigned(vap->va_rminor);
1465 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1466 rmajor = nfs_xdrneg1;
1467 rminor = nfs_xdrneg1;
1468 } else {
1469 return (EOPNOTSUPP);
1470 }
1471 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1472 return (error);
1473 }
1474 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1475 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1476 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1477 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1478 ERROROUT(nfsm_fhtom(&info, dvp));
1479 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1480 NFS_MAXNAMLEN));
1481 if (info.v3) {
1482 tl = nfsm_build(&info, NFSX_UNSIGNED);
1483 *tl++ = vtonfsv3_type(vap->va_type);
1484 nfsm_v3attrbuild(&info, vap, FALSE);
1485 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1486 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1487 *tl++ = txdr_unsigned(vap->va_rmajor);
1488 *tl = txdr_unsigned(vap->va_rminor);
1489 }
1490 } else {
1491 sp = nfsm_build(&info, NFSX_V2SATTR);
1492 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1493 sp->sa_uid = nfs_xdrneg1;
1494 sp->sa_gid = nfs_xdrneg1;
1495 sp->sa_size = makeudev(rmajor, rminor);
1496 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1497 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1498 }
1499 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1500 cnp->cn_cred, &error));
1501 if (!error) {
1502 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1503 if (!gotvp) {
1504 if (newvp) {
1505 vput(newvp);
1506 newvp = NULL;
1507 }
1508 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1509 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1510 if (!error)
1511 newvp = NFSTOV(np);
1512 }
1513 }
1514 if (info.v3) {
1515 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1516 }
1517 m_freem(info.mrep);
1518 info.mrep = NULL;
1519nfsmout:
1520 if (error) {
1521 if (newvp)
1522 vput(newvp);
1523 } else {
1524 *vpp = newvp;
1525 }
1526 VTONFS(dvp)->n_flag |= NLMODIFIED;
1527 if (!wccflag)
1528 VTONFS(dvp)->n_attrstamp = 0;
1529 return (error);
1530}
1531
1532/*
1533 * nfs mknod vop
1534 * just call nfs_mknodrpc() to do the work.
1535 *
1536 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1537 * struct componentname *a_cnp, struct vattr *a_vap)
1538 */
1539/* ARGSUSED */
1540static int
1541nfs_mknod(struct vop_old_mknod_args *ap)
1542{
1543 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1544}
1545
1546static u_long create_verf;
1547/*
1548 * nfs file create call
1549 *
1550 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1551 * struct componentname *a_cnp, struct vattr *a_vap)
1552 */
1553static int
1554nfs_create(struct vop_old_create_args *ap)
1555{
1556 struct vnode *dvp = ap->a_dvp;
1557 struct vattr *vap = ap->a_vap;
1558 struct componentname *cnp = ap->a_cnp;
1559 struct nfsv2_sattr *sp;
1560 u_int32_t *tl;
1561 struct nfsnode *np = NULL;
1562 struct vnode *newvp = NULL;
1563 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1564 struct vattr vattr;
1565 struct nfsm_info info;
1566
1567 info.mrep = NULL;
1568 info.v3 = NFS_ISV3(dvp);
1569
1570 /*
1571 * Oops, not for me..
1572 */
1573 if (vap->va_type == VSOCK)
1574 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1575
1576 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1577 return (error);
1578 }
1579 if (vap->va_vaflags & VA_EXCLUSIVE)
1580 fmode |= O_EXCL;
1581again:
1582 nfsstats.rpccnt[NFSPROC_CREATE]++;
1583 nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1584 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1585 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1586 ERROROUT(nfsm_fhtom(&info, dvp));
1587 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1588 NFS_MAXNAMLEN));
1589 if (info.v3) {
1590 tl = nfsm_build(&info, NFSX_UNSIGNED);
1591 if (fmode & O_EXCL) {
1592 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1593 tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1594#ifdef INET
1595 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1596 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1597 else
1598#endif
1599 *tl++ = create_verf;
1600 *tl = ++create_verf;
1601 } else {
1602 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1603 nfsm_v3attrbuild(&info, vap, FALSE);
1604 }
1605 } else {
1606 sp = nfsm_build(&info, NFSX_V2SATTR);
1607 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1608 sp->sa_uid = nfs_xdrneg1;
1609 sp->sa_gid = nfs_xdrneg1;
1610 sp->sa_size = 0;
1611 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1612 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1613 }
1614 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1615 cnp->cn_cred, &error));
1616 if (error == 0) {
1617 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1618 if (!gotvp) {
1619 if (newvp) {
1620 vput(newvp);
1621 newvp = NULL;
1622 }
1623 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1624 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1625 if (!error)
1626 newvp = NFSTOV(np);
1627 }
1628 }
1629 if (info.v3) {
1630 if (error == 0)
1631 error = nfsm_wcc_data(&info, dvp, &wccflag);
1632 else
1633 (void)nfsm_wcc_data(&info, dvp, &wccflag);
1634 }
1635 m_freem(info.mrep);
1636 info.mrep = NULL;
1637nfsmout:
1638 if (error) {
1639 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1640 KKASSERT(newvp == NULL);
1641 fmode &= ~O_EXCL;
1642 goto again;
1643 }
1644 } else if (info.v3 && (fmode & O_EXCL)) {
1645 /*
1646 * We are normally called with only a partially initialized
1647 * VAP. Since the NFSv3 spec says that server may use the
1648 * file attributes to store the verifier, the spec requires
1649 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1650 * in atime, but we can't really assume that all servers will
1651 * so we ensure that our SETATTR sets both atime and mtime.
1652 */
1653 if (vap->va_mtime.tv_sec == VNOVAL)
1654 vfs_timestamp(&vap->va_mtime);
1655 if (vap->va_atime.tv_sec == VNOVAL)
1656 vap->va_atime = vap->va_mtime;
1657 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1658 }
1659 if (error == 0) {
1660 /*
1661 * The new np may have enough info for access
1662 * checks, make sure rucred and wucred are
1663 * initialized for read and write rpc's.
1664 */
1665 np = VTONFS(newvp);
1666 if (np->n_rucred == NULL)
1667 np->n_rucred = crhold(cnp->cn_cred);
1668 if (np->n_wucred == NULL)
1669 np->n_wucred = crhold(cnp->cn_cred);
1670 *ap->a_vpp = newvp;
1671 } else if (newvp) {
1672 vput(newvp);
1673 }
1674 VTONFS(dvp)->n_flag |= NLMODIFIED;
1675 if (!wccflag)
1676 VTONFS(dvp)->n_attrstamp = 0;
1677 return (error);
1678}
1679
1680/*
1681 * nfs file remove call
1682 * To try and make nfs semantics closer to ufs semantics, a file that has
1683 * other processes using the vnode is renamed instead of removed and then
1684 * removed later on the last close.
1685 * - If v_sysref.refcnt > 1
1686 * If a rename is not already in the works
1687 * call nfs_sillyrename() to set it up
1688 * else
1689 * do the remove rpc
1690 *
1691 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1692 * struct componentname *a_cnp)
1693 */
1694static int
1695nfs_remove(struct vop_old_remove_args *ap)
1696{
1697 struct vnode *vp = ap->a_vp;
1698 struct vnode *dvp = ap->a_dvp;
1699 struct componentname *cnp = ap->a_cnp;
1700 struct nfsnode *np = VTONFS(vp);
1701 int error = 0;
1702 struct vattr vattr;
1703
1704#ifndef DIAGNOSTIC
1705 if (vp->v_sysref.refcnt < 1)
1706 panic("nfs_remove: bad v_sysref.refcnt");
1707#endif
1708 if (vp->v_type == VDIR)
1709 error = EPERM;
1710 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1711 VOP_GETATTR(vp, &vattr) == 0 &&
1712 vattr.va_nlink > 1)) {
1713 /*
1714 * throw away biocache buffers, mainly to avoid
1715 * unnecessary delayed writes later.
1716 */
1717 error = nfs_vinvalbuf(vp, 0, 1);
1718 /* Do the rpc */
1719 if (error != EINTR)
1720 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1721 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1722 /*
1723 * Kludge City: If the first reply to the remove rpc is lost..
1724 * the reply to the retransmitted request will be ENOENT
1725 * since the file was in fact removed
1726 * Therefore, we cheat and return success.
1727 */
1728 if (error == ENOENT)
1729 error = 0;
1730 } else if (!np->n_sillyrename) {
1731 error = nfs_sillyrename(dvp, vp, cnp);
1732 }
1733 np->n_attrstamp = 0;
1734 return (error);
1735}
1736
1737/*
1738 * nfs file remove rpc called from nfs_inactive
1739 */
1740int
1741nfs_removeit(struct sillyrename *sp)
1742{
1743 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1744 sp->s_cred, NULL));
1745}
1746
1747/*
1748 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1749 */
1750static int
1751nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1752 struct ucred *cred, struct thread *td)
1753{
1754 int error = 0, wccflag = NFSV3_WCCRATTR;
1755 struct nfsm_info info;
1756
1757 info.mrep = NULL;
1758 info.v3 = NFS_ISV3(dvp);
1759
1760 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1761 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1762 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1763 ERROROUT(nfsm_fhtom(&info, dvp));
1764 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1765 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1766 if (info.v3) {
1767 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1768 }
1769 m_freem(info.mrep);
1770 info.mrep = NULL;
1771nfsmout:
1772 VTONFS(dvp)->n_flag |= NLMODIFIED;
1773 if (!wccflag)
1774 VTONFS(dvp)->n_attrstamp = 0;
1775 return (error);
1776}
1777
1778/*
1779 * nfs file rename call
1780 *
1781 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1782 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1783 * struct vnode *a_tvp, struct componentname *a_tcnp)
1784 */
1785static int
1786nfs_rename(struct vop_old_rename_args *ap)
1787{
1788 struct vnode *fvp = ap->a_fvp;
1789 struct vnode *tvp = ap->a_tvp;
1790 struct vnode *fdvp = ap->a_fdvp;
1791 struct vnode *tdvp = ap->a_tdvp;
1792 struct componentname *tcnp = ap->a_tcnp;
1793 struct componentname *fcnp = ap->a_fcnp;
1794 int error;
1795
1796 /* Check for cross-device rename */
1797 if ((fvp->v_mount != tdvp->v_mount) ||
1798 (tvp && (fvp->v_mount != tvp->v_mount))) {
1799 error = EXDEV;
1800 goto out;
1801 }
1802
1803 /*
1804 * We shouldn't have to flush fvp on rename for most server-side
1805 * filesystems as the file handle should not change. Unfortunately
1806 * the inode for some filesystems (msdosfs) might be tied to the
1807 * file name or directory position so to be completely safe
1808 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1809 * performance.
1810 *
1811 * We must flush tvp on rename because it might become stale on the
1812 * server after the rename.
1813 */
1814 if (nfs_flush_on_rename)
1815 VOP_FSYNC(fvp, MNT_WAIT, 0);
1816 if (tvp)
1817 VOP_FSYNC(tvp, MNT_WAIT, 0);
1818
1819 /*
1820 * If the tvp exists and is in use, sillyrename it before doing the
1821 * rename of the new file over it.
1822 *
1823 * XXX Can't sillyrename a directory.
1824 *
1825 * We do not attempt to do any namecache purges in this old API
1826 * routine. The new API compat functions have access to the actual
1827 * namecache structures and will do it for us.
1828 */
1829 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1830 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1831 vput(tvp);
1832 tvp = NULL;
1833 } else if (tvp) {
1834 ;
1835 }
1836
1837 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1838 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1839 tcnp->cn_td);
1840
1841out:
1842 if (tdvp == tvp)
1843 vrele(tdvp);
1844 else
1845 vput(tdvp);
1846 if (tvp)
1847 vput(tvp);
1848 vrele(fdvp);
1849 vrele(fvp);
1850 /*
1851 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1852 */
1853 if (error == ENOENT)
1854 error = 0;
1855 return (error);
1856}
1857
1858/*
1859 * nfs file rename rpc called from nfs_remove() above
1860 */
1861static int
1862nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1863 struct sillyrename *sp)
1864{
1865 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1866 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1867}
1868
1869/*
1870 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1871 */
1872static int
1873nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1874 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1875 struct ucred *cred, struct thread *td)
1876{
1877 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1878 struct nfsm_info info;
1879
1880 info.mrep = NULL;
1881 info.v3 = NFS_ISV3(fdvp);
1882
1883 nfsstats.rpccnt[NFSPROC_RENAME]++;
1884 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
1885 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
1886 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
1887 ERROROUT(nfsm_fhtom(&info, fdvp));
1888 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
1889 ERROROUT(nfsm_fhtom(&info, tdvp));
1890 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
1891 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
1892 if (info.v3) {
1893 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
1894 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
1895 }
1896 m_freem(info.mrep);
1897 info.mrep = NULL;
1898nfsmout:
1899 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1900 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1901 if (!fwccflag)
1902 VTONFS(fdvp)->n_attrstamp = 0;
1903 if (!twccflag)
1904 VTONFS(tdvp)->n_attrstamp = 0;
1905 return (error);
1906}
1907
1908/*
1909 * nfs hard link create call
1910 *
1911 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1912 * struct componentname *a_cnp)
1913 */
1914static int
1915nfs_link(struct vop_old_link_args *ap)
1916{
1917 struct vnode *vp = ap->a_vp;
1918 struct vnode *tdvp = ap->a_tdvp;
1919 struct componentname *cnp = ap->a_cnp;
1920 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1921 struct nfsm_info info;
1922
1923 if (vp->v_mount != tdvp->v_mount) {
1924 return (EXDEV);
1925 }
1926
1927 /*
1928 * The attribute cache may get out of sync with the server on link.
1929 * Pushing writes to the server before handle was inherited from
1930 * long long ago and it is unclear if we still need to do this.
1931 * Defaults to off.
1932 */
1933 if (nfs_flush_on_hlink)
1934 VOP_FSYNC(vp, MNT_WAIT, 0);
1935
1936 info.mrep = NULL;
1937 info.v3 = NFS_ISV3(vp);
1938
1939 nfsstats.rpccnt[NFSPROC_LINK]++;
1940 nfsm_reqhead(&info, vp, NFSPROC_LINK,
1941 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
1942 nfsm_rndup(cnp->cn_namelen));
1943 ERROROUT(nfsm_fhtom(&info, vp));
1944 ERROROUT(nfsm_fhtom(&info, tdvp));
1945 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1946 NFS_MAXNAMLEN));
1947 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
1948 cnp->cn_cred, &error));
1949 if (info.v3) {
1950 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1951 NFS_LATTR_NOSHRINK));
1952 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
1953 }
1954 m_freem(info.mrep);
1955 info.mrep = NULL;
1956nfsmout:
1957 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1958 if (!attrflag)
1959 VTONFS(vp)->n_attrstamp = 0;
1960 if (!wccflag)
1961 VTONFS(tdvp)->n_attrstamp = 0;
1962 /*
1963 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1964 */
1965 if (error == EEXIST)
1966 error = 0;
1967 return (error);
1968}
1969
1970/*
1971 * nfs symbolic link create call
1972 *
1973 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1974 * struct componentname *a_cnp, struct vattr *a_vap,
1975 * char *a_target)
1976 */
1977static int
1978nfs_symlink(struct vop_old_symlink_args *ap)
1979{
1980 struct vnode *dvp = ap->a_dvp;
1981 struct vattr *vap = ap->a_vap;
1982 struct componentname *cnp = ap->a_cnp;
1983 struct nfsv2_sattr *sp;
1984 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1985 struct vnode *newvp = NULL;
1986 struct nfsm_info info;
1987
1988 info.mrep = NULL;
1989 info.v3 = NFS_ISV3(dvp);
1990
1991 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1992 slen = strlen(ap->a_target);
1993 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
1994 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
1995 nfsm_rndup(cnp->cn_namelen) +
1996 nfsm_rndup(slen) + NFSX_SATTR(info.v3));
1997 ERROROUT(nfsm_fhtom(&info, dvp));
1998 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1999 NFS_MAXNAMLEN));
2000 if (info.v3) {
2001 nfsm_v3attrbuild(&info, vap, FALSE);
2002 }
2003 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
2004 if (info.v3 == 0) {
2005 sp = nfsm_build(&info, NFSX_V2SATTR);
2006 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2007 sp->sa_uid = nfs_xdrneg1;
2008 sp->sa_gid = nfs_xdrneg1;
2009 sp->sa_size = nfs_xdrneg1;
2010 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2011 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2012 }
2013
2014 /*
2015 * Issue the NFS request and get the rpc response.
2016 *
2017 * Only NFSv3 responses returning an error of 0 actually return
2018 * a file handle that can be converted into newvp without having
2019 * to do an extra lookup rpc.
2020 */
2021 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
2022 cnp->cn_cred, &error));
2023 if (info.v3) {
2024 if (error == 0) {
2025 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2026 }
2027 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2028 }
2029
2030 /*
2031 * out code jumps -> here, mrep is also freed.
2032 */
2033
2034 m_freem(info.mrep);
2035 info.mrep = NULL;
2036nfsmout:
2037
2038 /*
2039 * If we get an EEXIST error, silently convert it to no-error
2040 * in case of an NFS retry.
2041 */
2042 if (error == EEXIST)
2043 error = 0;
2044
2045 /*
2046 * If we do not have (or no longer have) an error, and we could
2047 * not extract the newvp from the response due to the request being
2048 * NFSv2 or the error being EEXIST. We have to do a lookup in order
2049 * to obtain a newvp to return.
2050 */
2051 if (error == 0 && newvp == NULL) {
2052 struct nfsnode *np = NULL;
2053
2054 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2055 cnp->cn_cred, cnp->cn_td, &np);
2056 if (!error)
2057 newvp = NFSTOV(np);
2058 }
2059 if (error) {
2060 if (newvp)
2061 vput(newvp);
2062 } else {
2063 *ap->a_vpp = newvp;
2064 }
2065 VTONFS(dvp)->n_flag |= NLMODIFIED;
2066 if (!wccflag)
2067 VTONFS(dvp)->n_attrstamp = 0;
2068 return (error);
2069}
2070
2071/*
2072 * nfs make dir call
2073 *
2074 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2075 * struct componentname *a_cnp, struct vattr *a_vap)
2076 */
2077static int
2078nfs_mkdir(struct vop_old_mkdir_args *ap)
2079{
2080 struct vnode *dvp = ap->a_dvp;
2081 struct vattr *vap = ap->a_vap;
2082 struct componentname *cnp = ap->a_cnp;
2083 struct nfsv2_sattr *sp;
2084 struct nfsnode *np = NULL;
2085 struct vnode *newvp = NULL;
2086 struct vattr vattr;
2087 int error = 0, wccflag = NFSV3_WCCRATTR;
2088 int gotvp = 0;
2089 int len;
2090 struct nfsm_info info;
2091
2092 info.mrep = NULL;
2093 info.v3 = NFS_ISV3(dvp);
2094
2095 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2096 return (error);
2097 }
2098 len = cnp->cn_namelen;
2099 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2100 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2101 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2102 nfsm_rndup(len) + NFSX_SATTR(info.v3));
2103 ERROROUT(nfsm_fhtom(&info, dvp));
2104 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2105 if (info.v3) {
2106 nfsm_v3attrbuild(&info, vap, FALSE);
2107 } else {
2108 sp = nfsm_build(&info, NFSX_V2SATTR);
2109 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2110 sp->sa_uid = nfs_xdrneg1;
2111 sp->sa_gid = nfs_xdrneg1;
2112 sp->sa_size = nfs_xdrneg1;
2113 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2114 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2115 }
2116 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2117 cnp->cn_cred, &error));
2118 if (error == 0) {
2119 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2120 }
2121 if (info.v3) {
2122 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2123 }
2124 m_freem(info.mrep);
2125 info.mrep = NULL;
2126nfsmout:
2127 VTONFS(dvp)->n_flag |= NLMODIFIED;
2128 if (!wccflag)
2129 VTONFS(dvp)->n_attrstamp = 0;
2130 /*
2131 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2132 * if we can succeed in looking up the directory.
2133 */
2134 if (error == EEXIST || (!error && !gotvp)) {
2135 if (newvp) {
2136 vrele(newvp);
2137 newvp = NULL;
2138 }
2139 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2140 cnp->cn_td, &np);
2141 if (!error) {
2142 newvp = NFSTOV(np);
2143 if (newvp->v_type != VDIR)
2144 error = EEXIST;
2145 }
2146 }
2147 if (error) {
2148 if (newvp)
2149 vrele(newvp);
2150 } else
2151 *ap->a_vpp = newvp;
2152 return (error);
2153}
2154
2155/*
2156 * nfs remove directory call
2157 *
2158 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2159 * struct componentname *a_cnp)
2160 */
2161static int
2162nfs_rmdir(struct vop_old_rmdir_args *ap)
2163{
2164 struct vnode *vp = ap->a_vp;
2165 struct vnode *dvp = ap->a_dvp;
2166 struct componentname *cnp = ap->a_cnp;
2167 int error = 0, wccflag = NFSV3_WCCRATTR;
2168 struct nfsm_info info;
2169
2170 info.mrep = NULL;
2171 info.v3 = NFS_ISV3(dvp);
2172
2173 if (dvp == vp)
2174 return (EINVAL);
2175 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2176 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2177 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2178 nfsm_rndup(cnp->cn_namelen));
2179 ERROROUT(nfsm_fhtom(&info, dvp));
2180 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2181 NFS_MAXNAMLEN));
2182 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2183 cnp->cn_cred, &error));
2184 if (info.v3) {
2185 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2186 }
2187 m_freem(info.mrep);
2188 info.mrep = NULL;
2189nfsmout:
2190 VTONFS(dvp)->n_flag |= NLMODIFIED;
2191 if (!wccflag)
2192 VTONFS(dvp)->n_attrstamp = 0;
2193 /*
2194 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2195 */
2196 if (error == ENOENT)
2197 error = 0;
2198 return (error);
2199}
2200
2201/*
2202 * nfs readdir call
2203 *
2204 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2205 */
2206static int
2207nfs_readdir(struct vop_readdir_args *ap)
2208{
2209 struct vnode *vp = ap->a_vp;
2210 struct nfsnode *np = VTONFS(vp);
2211 struct uio *uio = ap->a_uio;
2212 int tresid, error;
2213 struct vattr vattr;
2214
2215 if (vp->v_type != VDIR)
2216 return (EPERM);
2217
2218 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2219 return (error);
2220
2221 /*
2222 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2223 * and then check that is still valid, or if this is an NQNFS mount
2224 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2225 * VOP_GETATTR() does not necessarily go to the wire.
2226 */
2227 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2228 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2229 if (VOP_GETATTR(vp, &vattr) == 0 &&
2230 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2231 ) {
2232 nfsstats.direofcache_hits++;
2233 goto done;
2234 }
2235 }
2236
2237 /*
2238 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2239 * own cache coherency checks so we do not have to.
2240 */
2241 tresid = uio->uio_resid;
2242 error = nfs_bioread(vp, uio, 0);
2243
2244 if (!error && uio->uio_resid == tresid)
2245 nfsstats.direofcache_misses++;
2246done:
2247 vn_unlock(vp);
2248 return (error);
2249}
2250
2251/*
2252 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2253 *
2254 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2255 * offset/block and converts the nfs formatted directory entries for userland
2256 * consumption as well as deals with offsets into the middle of blocks.
2257 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2258 * be block-bounded. It must convert to cookies for the actual RPC.
2259 */
2260int
2261nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2262{
2263 int len, left;
2264 struct nfs_dirent *dp = NULL;
2265 u_int32_t *tl;
2266 nfsuint64 *cookiep;
2267 caddr_t cp;
2268 nfsuint64 cookie;
2269 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2270 struct nfsnode *dnp = VTONFS(vp);
2271 u_quad_t fileno;
2272 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2273 int attrflag;
2274 struct nfsm_info info;
2275
2276 info.mrep = NULL;
2277 info.v3 = NFS_ISV3(vp);
2278
2279#ifndef DIAGNOSTIC
2280 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2281 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2282 panic("nfs readdirrpc bad uio");
2283#endif
2284
2285 /*
2286 * If there is no cookie, assume directory was stale.
2287 */
2288 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2289 if (cookiep)
2290 cookie = *cookiep;
2291 else
2292 return (NFSERR_BAD_COOKIE);
2293 /*
2294 * Loop around doing readdir rpc's of size nm_readdirsize
2295 * truncated to a multiple of DIRBLKSIZ.
2296 * The stopping criteria is EOF or buffer full.
2297 */
2298 while (more_dirs && bigenough) {
2299 nfsstats.rpccnt[NFSPROC_READDIR]++;
2300 nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2301 NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2302 ERROROUT(nfsm_fhtom(&info, vp));
2303 if (info.v3) {
2304 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2305 *tl++ = cookie.nfsuquad[0];
2306 *tl++ = cookie.nfsuquad[1];
2307 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2308 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2309 } else {
2310 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2311 *tl++ = cookie.nfsuquad[0];
2312 }
2313 *tl = txdr_unsigned(nmp->nm_readdirsize);
2314 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2315 uiop->uio_td,
2316 nfs_vpcred(vp, ND_READ), &error));
2317 if (info.v3) {
2318 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2319 NFS_LATTR_NOSHRINK));
2320 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2321 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2322 dnp->n_cookieverf.nfsuquad[1] = *tl;
2323 }
2324 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2325 more_dirs = fxdr_unsigned(int, *tl);
2326
2327 /* loop thru the dir entries, converting them to std form */
2328 while (more_dirs && bigenough) {
2329 if (info.v3) {
2330 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2331 fileno = fxdr_hyper(tl);
2332 len = fxdr_unsigned(int, *(tl + 2));
2333 } else {
2334 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2335 fileno = fxdr_unsigned(u_quad_t, *tl++);
2336 len = fxdr_unsigned(int, *tl);
2337 }
2338 if (len <= 0 || len > NFS_MAXNAMLEN) {
2339 error = EBADRPC;
2340 m_freem(info.mrep);
2341 info.mrep = NULL;
2342 goto nfsmout;
2343 }
2344
2345 /*
2346 * len is the number of bytes in the path element
2347 * name, not including the \0 termination.
2348 *
2349 * tlen is the number of bytes w have to reserve for
2350 * the path element name.
2351 */
2352 tlen = nfsm_rndup(len);
2353 if (tlen == len)
2354 tlen += 4; /* To ensure null termination */
2355
2356 /*
2357 * If the entry would cross a DIRBLKSIZ boundary,
2358 * extend the previous nfs_dirent to cover the
2359 * remaining space.
2360 */
2361 left = DIRBLKSIZ - blksiz;
2362 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2363 dp->nfs_reclen += left;
2364 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2365 uiop->uio_iov->iov_len -= left;
2366 uiop->uio_offset += left;
2367 uiop->uio_resid -= left;
2368 blksiz = 0;
2369 }
2370 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2371 bigenough = 0;
2372 if (bigenough) {
2373 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2374 dp->nfs_ino = fileno;
2375 dp->nfs_namlen = len;
2376 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2377 dp->nfs_type = DT_UNKNOWN;
2378 blksiz += dp->nfs_reclen;
2379 if (blksiz == DIRBLKSIZ)
2380 blksiz = 0;
2381 uiop->uio_offset += sizeof(struct nfs_dirent);
2382 uiop->uio_resid -= sizeof(struct nfs_dirent);
2383 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2384 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2385 ERROROUT(nfsm_mtouio(&info, uiop, len));
2386
2387 /*
2388 * The uiop has advanced by nfs_dirent + len
2389 * but really needs to advance by
2390 * nfs_dirent + tlen
2391 */
2392 cp = uiop->uio_iov->iov_base;
2393 tlen -= len;
2394 *cp = '\0'; /* null terminate */
2395 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2396 uiop->uio_iov->iov_len -= tlen;
2397 uiop->uio_offset += tlen;
2398 uiop->uio_resid -= tlen;
2399 } else {
2400 /*
2401 * NFS strings must be rounded up (nfsm_myouio
2402 * handled that in the bigenough case).
2403 */
2404 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2405 }
2406 if (info.v3) {
2407 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2408 } else {
2409 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2410 }
2411
2412 /*
2413 * If we were able to accomodate the last entry,
2414 * get the cookie for the next one. Otherwise
2415 * hold-over the cookie for the one we were not
2416 * able to accomodate.
2417 */
2418 if (bigenough) {
2419 cookie.nfsuquad[0] = *tl++;
2420 if (info.v3)
2421 cookie.nfsuquad[1] = *tl++;
2422 } else if (info.v3) {
2423 tl += 2;
2424 } else {
2425 tl++;
2426 }
2427 more_dirs = fxdr_unsigned(int, *tl);
2428 }
2429 /*
2430 * If at end of rpc data, get the eof boolean
2431 */
2432 if (!more_dirs) {
2433 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2434 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2435 }
2436 m_freem(info.mrep);
2437 info.mrep = NULL;
2438 }
2439 /*
2440 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2441 * by increasing d_reclen for the last record.
2442 */
2443 if (blksiz > 0) {
2444 left = DIRBLKSIZ - blksiz;
2445 dp->nfs_reclen += left;
2446 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2447 uiop->uio_iov->iov_len -= left;
2448 uiop->uio_offset += left;
2449 uiop->uio_resid -= left;
2450 }
2451
2452 if (bigenough) {
2453 /*
2454 * We hit the end of the directory, update direofoffset.
2455 */
2456 dnp->n_direofoffset = uiop->uio_offset;
2457 } else {
2458 /*
2459 * There is more to go, insert the link cookie so the
2460 * next block can be read.
2461 */
2462 if (uiop->uio_resid > 0)
2463 kprintf("EEK! readdirrpc resid > 0\n");
2464 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2465 *cookiep = cookie;
2466 }
2467nfsmout:
2468 return (error);
2469}
2470
2471/*
2472 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2473 */
2474int
2475nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2476{
2477 int len, left;
2478 struct nfs_dirent *dp;
2479 u_int32_t *tl;
2480 struct vnode *newvp;
2481 nfsuint64 *cookiep;
2482 caddr_t dpossav1, dpossav2;
2483 caddr_t cp;
2484 struct mbuf *mdsav1, *mdsav2;
2485 nfsuint64 cookie;
2486 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2487 struct nfsnode *dnp = VTONFS(vp), *np;
2488 nfsfh_t *fhp;
2489 u_quad_t fileno;
2490 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2491 int attrflag, fhsize;
2492 struct nchandle nch;
2493 struct nchandle dnch;
2494 struct nlcomponent nlc;
2495 struct nfsm_info info;
2496
2497 info.mrep = NULL;
2498 info.v3 = 1;
2499
2500#ifndef nolint
2501 dp = NULL;
2502#endif
2503#ifndef DIAGNOSTIC
2504 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2505 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2506 panic("nfs readdirplusrpc bad uio");
2507#endif
2508 /*
2509 * Obtain the namecache record for the directory so we have something
2510 * to use as a basis for creating the entries. This function will
2511 * return a held (but not locked) ncp. The ncp may be disconnected
2512 * from the tree and cannot be used for upward traversals, and the
2513 * ncp may be unnamed. Note that other unrelated operations may
2514 * cause the ncp to be named at any time.
2515 */
2516 cache_fromdvp(vp, NULL, 0, &dnch);
2517 bzero(&nlc, sizeof(nlc));
2518 newvp = NULLVP;
2519
2520 /*
2521 * If there is no cookie, assume directory was stale.
2522 */
2523 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2524 if (cookiep)
2525 cookie = *cookiep;
2526 else
2527 return (NFSERR_BAD_COOKIE);
2528 /*
2529 * Loop around doing readdir rpc's of size nm_readdirsize
2530 * truncated to a multiple of DIRBLKSIZ.
2531 * The stopping criteria is EOF or buffer full.
2532 */
2533 while (more_dirs && bigenough) {
2534 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2535 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2536 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2537 ERROROUT(nfsm_fhtom(&info, vp));
2538 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2539 *tl++ = cookie.nfsuquad[0];
2540 *tl++ = cookie.nfsuquad[1];
2541 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2542 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2543 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2544 *tl = txdr_unsigned(nmp->nm_rsize);
2545 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2546 uiop->uio_td,
2547 nfs_vpcred(vp, ND_READ), &error));
2548 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2549 NFS_LATTR_NOSHRINK));
2550 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2551 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2552 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2553 more_dirs = fxdr_unsigned(int, *tl);
2554
2555 /* loop thru the dir entries, doctoring them to 4bsd form */
2556 while (more_dirs && bigenough) {
2557 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2558 fileno = fxdr_hyper(tl);
2559 len = fxdr_unsigned(int, *(tl + 2));
2560 if (len <= 0 || len > NFS_MAXNAMLEN) {
2561 error = EBADRPC;
2562 m_freem(info.mrep);
2563 info.mrep = NULL;
2564 goto nfsmout;
2565 }
2566 tlen = nfsm_rndup(len);
2567 if (tlen == len)
2568 tlen += 4; /* To ensure null termination*/
2569 left = DIRBLKSIZ - blksiz;
2570 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2571 dp->nfs_reclen += left;
2572 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2573 uiop->uio_iov->iov_len -= left;
2574 uiop->uio_offset += left;
2575 uiop->uio_resid -= left;
2576 blksiz = 0;
2577 }
2578 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2579 bigenough = 0;
2580 if (bigenough) {
2581 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2582 dp->nfs_ino = fileno;
2583 dp->nfs_namlen = len;
2584 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2585 dp->nfs_type = DT_UNKNOWN;
2586 blksiz += dp->nfs_reclen;
2587 if (blksiz == DIRBLKSIZ)
2588 blksiz = 0;
2589 uiop->uio_offset += sizeof(struct nfs_dirent);
2590 uiop->uio_resid -= sizeof(struct nfs_dirent);
2591 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2592 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2593 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2594 nlc.nlc_namelen = len;
2595 ERROROUT(nfsm_mtouio(&info, uiop, len));
2596 cp = uiop->uio_iov->iov_base;
2597 tlen -= len;
2598 *cp = '\0';
2599 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2600 uiop->uio_iov->iov_len -= tlen;
2601 uiop->uio_offset += tlen;
2602 uiop->uio_resid -= tlen;
2603 } else {
2604 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2605 }
2606 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2607 if (bigenough) {
2608 cookie.nfsuquad[0] = *tl++;
2609 cookie.nfsuquad[1] = *tl++;
2610 } else
2611 tl += 2;
2612
2613 /*
2614 * Since the attributes are before the file handle
2615 * (sigh), we must skip over the attributes and then
2616 * come back and get them.
2617 */
2618 attrflag = fxdr_unsigned(int, *tl);
2619 if (attrflag) {
2620 dpossav1 = info.dpos;
2621 mdsav1 = info.md;
2622 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2623 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2624 doit = fxdr_unsigned(int, *tl);
2625 if (doit) {
2626 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2627 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2628 vref(vp);
2629 newvp = vp;
2630 np = dnp;
2631 } else {
2632 error = nfs_nget(vp->v_mount, fhp,
2633 fhsize, &np);
2634 if (error)
2635 doit = 0;
2636 else
2637 newvp = NFSTOV(np);
2638 }
2639 }
2640 if (doit && bigenough) {
2641 dpossav2 = info.dpos;
2642 info.dpos = dpossav1;
2643 mdsav2 = info.md;
2644 info.md = mdsav1;
2645 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2646 info.dpos = dpossav2;
2647 info.md = mdsav2;
2648 dp->nfs_type =
2649 IFTODT(VTTOIF(np->n_vattr.va_type));
2650 if (dnch.ncp) {
2651 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2652 nlc.nlc_namelen, nlc.nlc_namelen,
2653 nlc.nlc_nameptr);
2654 nch = cache_nlookup(&dnch, &nlc);
2655 cache_setunresolved(&nch);
2656 nfs_cache_setvp(&nch, newvp,
2657 nfspos_cache_timeout);
2658 cache_put(&nch);
2659 } else {
2660 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER"
2661 " %*.*s\n",
2662 nlc.nlc_namelen, nlc.nlc_namelen,
2663 nlc.nlc_nameptr);
2664 }
2665 }
2666 } else {
2667 /* Just skip over the file handle */
2668 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2669 i = fxdr_unsigned(int, *tl);
2670 ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2671 }
2672 if (newvp != NULLVP) {
2673 if (newvp == vp)
2674 vrele(newvp);
2675 else
2676 vput(newvp);
2677 newvp = NULLVP;
2678 }
2679 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2680 more_dirs = fxdr_unsigned(int, *tl);
2681 }
2682 /*
2683 * If at end of rpc data, get the eof boolean
2684 */
2685 if (!more_dirs) {
2686 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2687 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2688 }
2689 m_freem(info.mrep);
2690 info.mrep = NULL;
2691 }
2692 /*
2693 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2694 * by increasing d_reclen for the last record.
2695 */
2696 if (blksiz > 0) {
2697 left = DIRBLKSIZ - blksiz;
2698 dp->nfs_reclen += left;
2699 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2700 uiop->uio_iov->iov_len -= left;
2701 uiop->uio_offset += left;
2702 uiop->uio_resid -= left;
2703 }
2704
2705 /*
2706 * We are now either at the end of the directory or have filled the
2707 * block.
2708 */
2709 if (bigenough)
2710 dnp->n_direofoffset = uiop->uio_offset;
2711 else {
2712 if (uiop->uio_resid > 0)
2713 kprintf("EEK! readdirplusrpc resid > 0\n");
2714 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2715 *cookiep = cookie;
2716 }
2717nfsmout:
2718 if (newvp != NULLVP) {
2719 if (newvp == vp)
2720 vrele(newvp);
2721 else
2722 vput(newvp);
2723 newvp = NULLVP;
2724 }
2725 if (dnch.ncp)
2726 cache_drop(&dnch);
2727 return (error);
2728}
2729
2730/*
2731 * Silly rename. To make the NFS filesystem that is stateless look a little
2732 * more like the "ufs" a remove of an active vnode is translated to a rename
2733 * to a funny looking filename that is removed by nfs_inactive on the
2734 * nfsnode. There is the potential for another process on a different client
2735 * to create the same funny name between the nfs_lookitup() fails and the
2736 * nfs_rename() completes, but...
2737 */
2738static int
2739nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2740{
2741 struct sillyrename *sp;
2742 struct nfsnode *np;
2743 int error;
2744
2745 /*
2746 * We previously purged dvp instead of vp. I don't know why, it
2747 * completely destroys performance. We can't do it anyway with the
2748 * new VFS API since we would be breaking the namecache topology.
2749 */
2750 cache_purge(vp); /* XXX */
2751 np = VTONFS(vp);
2752#ifndef DIAGNOSTIC
2753 if (vp->v_type == VDIR)
2754 panic("nfs: sillyrename dir");
2755#endif
2756 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2757 M_NFSREQ, M_WAITOK);
2758 sp->s_cred = crdup(cnp->cn_cred);
2759 sp->s_dvp = dvp;
2760 vref(dvp);
2761
2762 /* Fudge together a funny name */
2763 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2764 (int)(intptr_t)cnp->cn_td);
2765
2766 /* Try lookitups until we get one that isn't there */
2767 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2768 cnp->cn_td, NULL) == 0) {
2769 sp->s_name[4]++;
2770 if (sp->s_name[4] > 'z') {
2771 error = EINVAL;
2772 goto bad;
2773 }
2774 }
2775 error = nfs_renameit(dvp, cnp, sp);
2776 if (error)
2777 goto bad;
2778 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2779 cnp->cn_td, &np);
2780 np->n_sillyrename = sp;
2781 return (0);
2782bad:
2783 vrele(sp->s_dvp);
2784 crfree(sp->s_cred);
2785 kfree((caddr_t)sp, M_NFSREQ);
2786 return (error);
2787}
2788
2789/*
2790 * Look up a file name and optionally either update the file handle or
2791 * allocate an nfsnode, depending on the value of npp.
2792 * npp == NULL --> just do the lookup
2793 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2794 * handled too
2795 * *npp != NULL --> update the file handle in the vnode
2796 */
2797static int
2798nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2799 struct thread *td, struct nfsnode **npp)
2800{
2801 struct vnode *newvp = NULL;
2802 struct nfsnode *np, *dnp = VTONFS(dvp);
2803 int error = 0, fhlen, attrflag;
2804 nfsfh_t *nfhp;
2805 struct nfsm_info info;
2806
2807 info.mrep = NULL;
2808 info.v3 = NFS_ISV3(dvp);
2809
2810 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2811 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
2812 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2813 ERROROUT(nfsm_fhtom(&info, dvp));
2814 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
2815 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
2816 if (npp && !error) {
2817 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
2818 if (*npp) {
2819 np = *npp;
2820 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2821 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2822 np->n_fhp = &np->n_fh;
2823 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2824 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2825 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2826 np->n_fhsize = fhlen;
2827 newvp = NFSTOV(np);
2828 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2829 vref(dvp);
2830 newvp = dvp;
2831 } else {
2832 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2833 if (error) {
2834 m_freem(info.mrep);
2835 info.mrep = NULL;
2836 return (error);
2837 }
2838 newvp = NFSTOV(np);
2839 }
2840 if (info.v3) {
2841 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
2842 NFS_LATTR_NOSHRINK));
2843 if (!attrflag && *npp == NULL) {
2844 m_freem(info.mrep);
2845 info.mrep = NULL;
2846 if (newvp == dvp)
2847 vrele(newvp);
2848 else
2849 vput(newvp);
2850 return (ENOENT);
2851 }
2852 } else {
2853 ERROROUT(error = nfsm_loadattr(&info, newvp, NULL));
2854 }
2855 }
2856 m_freem(info.mrep);
2857 info.mrep = NULL;
2858nfsmout:
2859 if (npp && *npp == NULL) {
2860 if (error) {
2861 if (newvp) {
2862 if (newvp == dvp)
2863 vrele(newvp);
2864 else
2865 vput(newvp);
2866 }
2867 } else
2868 *npp = np;
2869 }
2870 return (error);
2871}
2872
2873/*
2874 * Nfs Version 3 commit rpc
2875 *
2876 * We call it 'uio' to distinguish it from 'bio' but there is no real uio
2877 * involved.
2878 */
2879int
2880nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2881{
2882 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2883 int error = 0, wccflag = NFSV3_WCCRATTR;
2884 struct nfsm_info info;
2885 u_int32_t *tl;
2886
2887 info.mrep = NULL;
2888 info.v3 = 1;
2889
2890 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2891 return (0);
2892 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2893 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
2894 ERROROUT(nfsm_fhtom(&info, vp));
2895 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
2896 txdr_hyper(offset, tl);
2897 tl += 2;
2898 *tl = txdr_unsigned(cnt);
2899 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
2900 nfs_vpcred(vp, ND_WRITE), &error));
2901 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
2902 if (!error) {
2903 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
2904 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2905 NFSX_V3WRITEVERF)) {
2906 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2907 NFSX_V3WRITEVERF);
2908 error = NFSERR_STALEWRITEVERF;
2909 }
2910 }
2911 m_freem(info.mrep);
2912 info.mrep = NULL;
2913nfsmout:
2914 return (error);
2915}
2916
2917/*
2918 * Kludge City..
2919 * - make nfs_bmap() essentially a no-op that does no translation
2920 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2921 * (Maybe I could use the process's page mapping, but I was concerned that
2922 * Kernel Write might not be enabled and also figured copyout() would do
2923 * a lot more work than bcopy() and also it currently happens in the
2924 * context of the swapper process (2).
2925 *
2926 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
2927 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2928 */
2929static int
2930nfs_bmap(struct vop_bmap_args *ap)
2931{
2932 if (ap->a_doffsetp != NULL)
2933 *ap->a_doffsetp = ap->a_loffset;
2934 if (ap->a_runp != NULL)
2935 *ap->a_runp = 0;
2936 if (ap->a_runb != NULL)
2937 *ap->a_runb = 0;
2938 return (0);
2939}
2940
2941/*
2942 * Strategy routine.
2943 */
2944static int
2945nfs_strategy(struct vop_strategy_args *ap)
2946{
2947 struct bio *bio = ap->a_bio;
2948 struct bio *nbio;
2949 struct buf *bp __debugvar = bio->bio_buf;
2950 struct thread *td;
2951 int error;
2952
2953 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2954 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
2955 KASSERT(BUF_REFCNT(bp) > 0,
2956 ("nfs_strategy: buffer %p not locked", bp));
2957
2958 if (bio->bio_flags & BIO_SYNC)
2959 td = curthread; /* XXX */
2960 else
2961 td = NULL;
2962
2963 /*
2964 * We probably don't need to push an nbio any more since no
2965 * block conversion is required due to the use of 64 bit byte
2966 * offsets, but do it anyway.
2967 *
2968 * NOTE: When NFS callers itself via this strategy routines and
2969 * sets up a synchronous I/O, it expects the I/O to run
2970 * synchronously (its bio_done routine just assumes it),
2971 * so for now we have to honor the bit.
2972 */
2973 nbio = push_bio(bio);
2974 nbio->bio_offset = bio->bio_offset;
2975 nbio->bio_flags = bio->bio_flags & BIO_SYNC;
2976
2977 /*
2978 * If the op is asynchronous and an i/o daemon is waiting
2979 * queue the request, wake it up and wait for completion
2980 * otherwise just do it ourselves.
2981 */
2982 if (bio->bio_flags & BIO_SYNC) {
2983 error = nfs_doio(ap->a_vp, nbio, td);
2984 } else {
2985 nfs_asyncio(ap->a_vp, nbio);
2986 error = 0;
2987 }
2988 return (error);
2989}
2990
2991/*
2992 * Mmap a file
2993 *
2994 * NB Currently unsupported.
2995 *
2996 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
2997 */
2998/* ARGSUSED */
2999static int
3000nfs_mmap(struct vop_mmap_args *ap)
3001{
3002 return (EINVAL);
3003}
3004
3005/*
3006 * fsync vnode op. Just call nfs_flush() with commit == 1.
3007 *
3008 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3009 */
3010/* ARGSUSED */
3011static int
3012nfs_fsync(struct vop_fsync_args *ap)
3013{
3014 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
3015}
3016
3017/*
3018 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
3019 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
3020 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
3021 * set the buffer contains data that has already been written to the server
3022 * and which now needs a commit RPC.
3023 *
3024 * If commit is 0 we only take one pass and only flush buffers containing new
3025 * dirty data.
3026 *
3027 * If commit is 1 we take two passes, issuing a commit RPC in the second
3028 * pass.
3029 *
3030 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3031 * to completely flush all pending data.
3032 *
3033 * Note that the RB_SCAN code properly handles the case where the
3034 * callback might block and directly or indirectly (another thread) cause
3035 * the RB tree to change.
3036 */
3037
3038#ifndef NFS_COMMITBVECSIZ
3039#define NFS_COMMITBVECSIZ 16
3040#endif
3041
3042struct nfs_flush_info {
3043 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3044 struct thread *td;
3045 struct vnode *vp;
3046 int waitfor;
3047 int slpflag;
3048 int slptimeo;
3049 int loops;
3050 struct buf *bvary[NFS_COMMITBVECSIZ];
3051 int bvsize;
3052 off_t beg_off;
3053 off_t end_off;
3054};
3055
3056static int nfs_flush_bp(struct buf *bp, void *data);
3057static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3058
3059int
3060nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3061{
3062 struct nfsnode *np = VTONFS(vp);
3063 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3064 struct nfs_flush_info info;
3065 lwkt_tokref vlock;
3066 int error;
3067
3068 bzero(&info, sizeof(info));
3069 info.td = td;
3070 info.vp = vp;
3071 info.waitfor = waitfor;
3072 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3073 info.loops = 0;
3074 lwkt_gettoken(&vlock, &vp->v_token);
3075
3076 do {
3077 /*
3078 * Flush mode
3079 */
3080 info.mode = NFI_FLUSHNEW;
3081 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3082 nfs_flush_bp, &info);
3083
3084 /*
3085 * Take a second pass if committing and no error occured.
3086 * Clean up any left over collection (whether an error
3087 * occurs or not).
3088 */
3089 if (commit && error == 0) {
3090 info.mode = NFI_COMMIT;
3091 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3092 nfs_flush_bp, &info);
3093 if (info.bvsize)
3094 error = nfs_flush_docommit(&info, error);
3095 }
3096
3097 /*
3098 * Wait for pending I/O to complete before checking whether
3099 * any further dirty buffers exist.
3100 */
3101 while (waitfor == MNT_WAIT &&
3102 bio_track_active(&vp->v_track_write)) {
3103 error = bio_track_wait(&vp->v_track_write,
3104 info.slpflag, info.slptimeo);
3105 if (error) {
3106 /*
3107 * We have to be able to break out if this
3108 * is an 'intr' mount.
3109 */
3110 if (nfs_sigintr(nmp, NULL, td)) {
3111 error = -EINTR;
3112 break;
3113 }
3114
3115 /*
3116 * Since we do not process pending signals,
3117 * once we get a PCATCH our tsleep() will no
3118 * longer sleep, switch to a fixed timeout
3119 * instead.
3120 */
3121 if (info.slpflag == PCATCH) {
3122 info.slpflag = 0;
3123 info.slptimeo = 2 * hz;
3124 }
3125 error = 0;
3126 }
3127 }
3128 ++info.loops;
3129 /*
3130 * Loop if we are flushing synchronous as well as committing,
3131 * and dirty buffers are still present. Otherwise we might livelock.
3132 */
3133 } while (waitfor == MNT_WAIT && commit &&
3134 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3135
3136 /*
3137 * The callbacks have to return a negative error to terminate the
3138 * RB scan.
3139 */
3140 if (error < 0)
3141 error = -error;
3142
3143 /*
3144 * Deal with any error collection
3145 */
3146 if (np->n_flag & NWRITEERR) {
3147 error = np->n_error;
3148 np->n_flag &= ~NWRITEERR;
3149 }
3150 lwkt_reltoken(&vlock);
3151 return (error);
3152}
3153
3154static
3155int
3156nfs_flush_bp(struct buf *bp, void *data)
3157{
3158 struct nfs_flush_info *info = data;
3159 int lkflags;
3160 int error;
3161 off_t toff;
3162
3163 error = 0;
3164 switch(info->mode) {
3165 case NFI_FLUSHNEW:
3166 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3167 if (error && info->loops && info->waitfor == MNT_WAIT) {
3168 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3169 if (error) {
3170 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3171 if (info->slpflag & PCATCH)
3172 lkflags |= LK_PCATCH;
3173 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3174 info->slptimeo);
3175 }
3176 }
3177
3178 /*
3179 * Ignore locking errors
3180 */
3181 if (error) {
3182 error = 0;
3183 break;
3184 }
3185
3186 /*
3187 * The buffer may have changed out from under us, even if
3188 * we did not block (MPSAFE). Check again now that it is
3189 * locked.
3190 */
3191 if (bp->b_vp == info->vp &&
3192 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3193 bremfree(bp);
3194 bawrite(bp);
3195 } else {
3196 BUF_UNLOCK(bp);
3197 }
3198 break;
3199 case NFI_COMMIT:
3200 /*
3201 * Only process buffers in need of a commit which we can
3202 * immediately lock. This may prevent a buffer from being
3203 * committed, but the normal flush loop will block on the
3204 * same buffer so we shouldn't get into an endless loop.
3205 */
3206 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3207 (B_DELWRI | B_NEEDCOMMIT)) {
3208 break;
3209 }
3210 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3211 break;
3212
3213 /*
3214 * We must recheck after successfully locking the buffer.
3215 */
3216 if (bp->b_vp != info->vp ||
3217 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3218 (B_DELWRI | B_NEEDCOMMIT)) {
3219 BUF_UNLOCK(bp);
3220 break;
3221 }
3222
3223 /*
3224 * NOTE: storing the bp in the bvary[] basically sets
3225 * it up for a commit operation.
3226 *
3227 * We must call vfs_busy_pages() now so the commit operation
3228 * is interlocked with user modifications to memory mapped
3229 * pages. The b_dirtyoff/b_dirtyend range is not correct
3230 * until after the pages have been busied.
3231 *
3232 * Note: to avoid loopback deadlocks, we do not
3233 * assign b_runningbufspace.
3234 */
3235 bremfree(bp);
3236 bp->b_cmd = BUF_CMD_WRITE;
3237 vfs_busy_pages(bp->b_vp, bp);
3238 info->bvary[info->bvsize] = bp;
3239 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3240 if (info->bvsize == 0 || toff < info->beg_off)
3241 info->beg_off = toff;
3242 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3243 if (info->bvsize == 0 || toff > info->end_off)
3244 info->end_off = toff;
3245 ++info->bvsize;
3246 if (info->bvsize == NFS_COMMITBVECSIZ) {
3247 error = nfs_flush_docommit(info, 0);
3248 KKASSERT(info->bvsize == 0);
3249 }
3250 }
3251 return (error);
3252}
3253
3254static
3255int
3256nfs_flush_docommit(struct nfs_flush_info *info, int error)
3257{
3258 struct vnode *vp;
3259 struct buf *bp;
3260 off_t bytes;
3261 int retv;
3262 int i;
3263
3264 vp = info->vp;
3265
3266 if (info->bvsize > 0) {
3267 /*
3268 * Commit data on the server, as required. Note that
3269 * nfs_commit will use the vnode's cred for the commit.
3270 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3271 */
3272 bytes = info->end_off - info->beg_off;
3273 if (bytes > 0x40000000)
3274 bytes = 0x40000000;
3275 if (error) {
3276 retv = -error;
3277 } else {
3278 retv = nfs_commitrpc_uio(vp, info->beg_off,
3279 (int)bytes, info->td);
3280 if (retv == NFSERR_STALEWRITEVERF)
3281 nfs_clearcommit(vp->v_mount);
3282 }
3283
3284 /*
3285 * Now, either mark the blocks I/O done or mark the
3286 * blocks dirty, depending on whether the commit
3287 * succeeded.
3288 */
3289 for (i = 0; i < info->bvsize; ++i) {
3290 bp = info->bvary[i];
3291 if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3292 /*
3293 * Either an error or the original
3294 * vfs_busy_pages() cleared B_NEEDCOMMIT
3295 * due to finding new dirty VM pages in
3296 * the buffer.
3297 *
3298 * Leave B_DELWRI intact.
3299 */
3300 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3301 vfs_unbusy_pages(bp);
3302 bp->b_cmd = BUF_CMD_DONE;
3303 bqrelse(bp);
3304 } else {
3305 /*
3306 * Success, remove B_DELWRI ( bundirty() ).
3307 *
3308 * b_dirtyoff/b_dirtyend seem to be NFS
3309 * specific. We should probably move that
3310 * into bundirty(). XXX
3311 *
3312 * We are faking an I/O write, we have to
3313 * start the transaction in order to
3314 * immediately biodone() it.
3315 */
3316 bundirty(bp);
3317 bp->b_flags &= ~B_ERROR;
3318 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3319 bp->b_dirtyoff = bp->b_dirtyend = 0;
3320 biodone(&bp->b_bio1);
3321 }
3322 }
3323 info->bvsize = 0;
3324 }
3325 return (error);
3326}
3327
3328/*
3329 * NFS advisory byte-level locks.
3330 * Currently unsupported.
3331 *
3332 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3333 * int a_flags)
3334 */
3335static int
3336nfs_advlock(struct vop_advlock_args *ap)
3337{
3338 struct nfsnode *np = VTONFS(ap->a_vp);
3339
3340 /*
3341 * The following kludge is to allow diskless support to work
3342 * until a real NFS lockd is implemented. Basically, just pretend
3343 * that this is a local lock.
3344 */
3345 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3346}
3347
3348/*
3349 * Print out the contents of an nfsnode.
3350 *
3351 * nfs_print(struct vnode *a_vp)
3352 */
3353static int
3354nfs_print(struct vop_print_args *ap)
3355{
3356 struct vnode *vp = ap->a_vp;
3357 struct nfsnode *np = VTONFS(vp);
3358
3359 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3360 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3361 if (vp->v_type == VFIFO)
3362 fifo_printinfo(vp);
3363 kprintf("\n");
3364 return (0);
3365}
3366
3367/*
3368 * nfs special file access vnode op.
3369 *
3370 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3371 */
3372static int
3373nfs_laccess(struct vop_access_args *ap)
3374{
3375 struct vattr vattr;
3376 int error;
3377
3378 error = VOP_GETATTR(ap->a_vp, &vattr);
3379 if (!error)
3380 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3381 vattr.va_mode, 0);
3382 return (error);
3383}
3384
3385/*
3386 * Read wrapper for fifos.
3387 *
3388 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3389 * struct ucred *a_cred)
3390 */
3391static int
3392nfsfifo_read(struct vop_read_args *ap)
3393{
3394 struct nfsnode *np = VTONFS(ap->a_vp);
3395
3396 /*
3397 * Set access flag.
3398 */
3399 np->n_flag |= NACC;
3400 getnanotime(&np->n_atim);
3401 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3402}
3403
3404/*
3405 * Write wrapper for fifos.
3406 *
3407 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3408 * struct ucred *a_cred)
3409 */
3410static int
3411nfsfifo_write(struct vop_write_args *ap)
3412{
3413 struct nfsnode *np = VTONFS(ap->a_vp);
3414
3415 /*
3416 * Set update flag.
3417 */
3418 np->n_flag |= NUPD;
3419 getnanotime(&np->n_mtim);
3420 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3421}
3422
3423/*
3424 * Close wrapper for fifos.
3425 *
3426 * Update the times on the nfsnode then do fifo close.
3427 *
3428 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3429 */
3430static int
3431nfsfifo_close(struct vop_close_args *ap)
3432{
3433 struct vnode *vp = ap->a_vp;
3434 struct nfsnode *np = VTONFS(vp);
3435 struct vattr vattr;
3436 struct timespec ts;
3437
3438 if (np->n_flag & (NACC | NUPD)) {
3439 getnanotime(&ts);
3440 if (np->n_flag & NACC)
3441 np->n_atim = ts;
3442 if (np->n_flag & NUPD)
3443 np->n_mtim = ts;
3444 np->n_flag |= NCHG;
3445 if (vp->v_sysref.refcnt == 1 &&
3446 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3447 VATTR_NULL(&vattr);
3448 if (np->n_flag & NACC)
3449 vattr.va_atime = np->n_atim;
3450 if (np->n_flag & NUPD)
3451 vattr.va_mtime = np->n_mtim;
3452 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3453 }
3454 }
3455 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3456}
3457