AMD64 - Fix many compile-time warnings. int/ptr type mismatches, %llx, etc.
[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 nfsspec_read (struct vop_read_args *);
98static int nfsspec_write (struct vop_write_args *);
99static int nfsfifo_read (struct vop_read_args *);
100static int nfsfifo_write (struct vop_write_args *);
101static int nfsspec_close (struct vop_close_args *);
102static int nfsfifo_close (struct vop_close_args *);
103#define nfs_poll vop_nopoll
104static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
105static int nfs_lookup (struct vop_old_lookup_args *);
106static int nfs_create (struct vop_old_create_args *);
107static int nfs_mknod (struct vop_old_mknod_args *);
108static int nfs_open (struct vop_open_args *);
109static int nfs_close (struct vop_close_args *);
110static int nfs_access (struct vop_access_args *);
111static int nfs_getattr (struct vop_getattr_args *);
112static int nfs_setattr (struct vop_setattr_args *);
113static int nfs_read (struct vop_read_args *);
114static int nfs_mmap (struct vop_mmap_args *);
115static int nfs_fsync (struct vop_fsync_args *);
116static int nfs_remove (struct vop_old_remove_args *);
117static int nfs_link (struct vop_old_link_args *);
118static int nfs_rename (struct vop_old_rename_args *);
119static int nfs_mkdir (struct vop_old_mkdir_args *);
120static int nfs_rmdir (struct vop_old_rmdir_args *);
121static int nfs_symlink (struct vop_old_symlink_args *);
122static int nfs_readdir (struct vop_readdir_args *);
123static int nfs_bmap (struct vop_bmap_args *);
124static int nfs_strategy (struct vop_strategy_args *);
125static int nfs_lookitup (struct vnode *, const char *, int,
126 struct ucred *, struct thread *, struct nfsnode **);
127static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
128static int nfsspec_access (struct vop_access_args *);
129static int nfs_readlink (struct vop_readlink_args *);
130static int nfs_print (struct vop_print_args *);
131static int nfs_advlock (struct vop_advlock_args *);
132
133static int nfs_nresolve (struct vop_nresolve_args *);
134/*
135 * Global vfs data structures for nfs
136 */
137struct vop_ops nfsv2_vnode_vops = {
138 .vop_default = vop_defaultop,
139 .vop_access = nfs_access,
140 .vop_advlock = nfs_advlock,
141 .vop_bmap = nfs_bmap,
142 .vop_close = nfs_close,
143 .vop_old_create = nfs_create,
144 .vop_fsync = nfs_fsync,
145 .vop_getattr = nfs_getattr,
146 .vop_getpages = nfs_getpages,
147 .vop_putpages = nfs_putpages,
148 .vop_inactive = nfs_inactive,
149 .vop_old_link = nfs_link,
150 .vop_old_lookup = nfs_lookup,
151 .vop_old_mkdir = nfs_mkdir,
152 .vop_old_mknod = nfs_mknod,
153 .vop_mmap = nfs_mmap,
154 .vop_open = nfs_open,
155 .vop_poll = nfs_poll,
156 .vop_print = nfs_print,
157 .vop_read = nfs_read,
158 .vop_readdir = nfs_readdir,
159 .vop_readlink = nfs_readlink,
160 .vop_reclaim = nfs_reclaim,
161 .vop_old_remove = nfs_remove,
162 .vop_old_rename = nfs_rename,
163 .vop_old_rmdir = nfs_rmdir,
164 .vop_setattr = nfs_setattr,
165 .vop_strategy = nfs_strategy,
166 .vop_old_symlink = nfs_symlink,
167 .vop_write = nfs_write,
168 .vop_nresolve = nfs_nresolve
169};
170
171/*
172 * Special device vnode ops
173 */
174struct vop_ops nfsv2_spec_vops = {
175 .vop_default = spec_vnoperate,
176 .vop_access = nfsspec_access,
177 .vop_close = nfsspec_close,
178 .vop_fsync = nfs_fsync,
179 .vop_getattr = nfs_getattr,
180 .vop_inactive = nfs_inactive,
181 .vop_print = nfs_print,
182 .vop_read = nfsspec_read,
183 .vop_reclaim = nfs_reclaim,
184 .vop_setattr = nfs_setattr,
185 .vop_write = nfsspec_write
186};
187
188struct vop_ops nfsv2_fifo_vops = {
189 .vop_default = fifo_vnoperate,
190 .vop_access = nfsspec_access,
191 .vop_close = nfsfifo_close,
192 .vop_fsync = nfs_fsync,
193 .vop_getattr = nfs_getattr,
194 .vop_inactive = nfs_inactive,
195 .vop_print = nfs_print,
196 .vop_read = nfsfifo_read,
197 .vop_reclaim = nfs_reclaim,
198 .vop_setattr = nfs_setattr,
199 .vop_write = nfsfifo_write
200};
201
202static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
203 struct componentname *cnp,
204 struct vattr *vap);
205static int nfs_removerpc (struct vnode *dvp, const char *name,
206 int namelen,
207 struct ucred *cred, struct thread *td);
208static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
209 int fnamelen, struct vnode *tdvp,
210 const char *tnameptr, int tnamelen,
211 struct ucred *cred, struct thread *td);
212static int nfs_renameit (struct vnode *sdvp,
213 struct componentname *scnp,
214 struct sillyrename *sp);
215
216/*
217 * Global variables
218 */
219extern u_int32_t nfs_true, nfs_false;
220extern u_int32_t nfs_xdrneg1;
221extern struct nfsstats nfsstats;
222extern nfstype nfsv3_type[9];
223struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON];
224struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
225int nfs_numasync = 0;
226
227SYSCTL_DECL(_vfs_nfs);
228
229static int nfs_flush_on_rename = 1;
230SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
231 &nfs_flush_on_rename, 0, "flush fvp prior to rename");
232static int nfs_flush_on_hlink = 0;
233SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
234 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
235
236static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
237SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
238 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
239
240static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
241SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
242 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
243
244static int nfspos_cache_timeout = NFS_MINATTRTIMO;
245SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
246 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
247
248static int nfsv3_commit_on_close = 0;
249SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
250 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
251#if 0
252SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
253 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
254
255SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
256 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
257#endif
258
259#define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
260 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
261 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
262static int
263nfs3_access_otw(struct vnode *vp, int wmode,
264 struct thread *td, struct ucred *cred)
265{
266 const int v3 = 1;
267 u_int32_t *tl;
268 int error = 0, attrflag;
269
270 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
271 caddr_t bpos, dpos, cp2;
272 int32_t t1, t2;
273 caddr_t cp;
274 u_int32_t rmode;
275 struct nfsnode *np = VTONFS(vp);
276
277 nfsstats.rpccnt[NFSPROC_ACCESS]++;
278 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
279 nfsm_fhtom(vp, v3);
280 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
281 *tl = txdr_unsigned(wmode);
282 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
283 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
284 if (!error) {
285 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
286 rmode = fxdr_unsigned(u_int32_t, *tl);
287 np->n_mode = rmode;
288 np->n_modeuid = cred->cr_uid;
289 np->n_modestamp = mycpu->gd_time_seconds;
290 }
291 m_freem(mrep);
292nfsmout:
293 return error;
294}
295
296/*
297 * nfs access vnode op.
298 * For nfs version 2, just return ok. File accesses may fail later.
299 * For nfs version 3, use the access rpc to check accessibility. If file modes
300 * are changed on the server, accesses might still fail later.
301 *
302 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
303 */
304static int
305nfs_access(struct vop_access_args *ap)
306{
307 struct vnode *vp = ap->a_vp;
308 thread_t td = curthread;
309 int error = 0;
310 u_int32_t mode, wmode;
311 int v3 = NFS_ISV3(vp);
312 struct nfsnode *np = VTONFS(vp);
313
314 /*
315 * Disallow write attempts on filesystems mounted read-only;
316 * unless the file is a socket, fifo, or a block or character
317 * device resident on the filesystem.
318 */
319 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
320 switch (vp->v_type) {
321 case VREG:
322 case VDIR:
323 case VLNK:
324 return (EROFS);
325 default:
326 break;
327 }
328 }
329 /*
330 * For nfs v3, check to see if we have done this recently, and if
331 * so return our cached result instead of making an ACCESS call.
332 * If not, do an access rpc, otherwise you are stuck emulating
333 * ufs_access() locally using the vattr. This may not be correct,
334 * since the server may apply other access criteria such as
335 * client uid-->server uid mapping that we do not know about.
336 */
337 if (v3) {
338 if (ap->a_mode & VREAD)
339 mode = NFSV3ACCESS_READ;
340 else
341 mode = 0;
342 if (vp->v_type != VDIR) {
343 if (ap->a_mode & VWRITE)
344 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
345 if (ap->a_mode & VEXEC)
346 mode |= NFSV3ACCESS_EXECUTE;
347 } else {
348 if (ap->a_mode & VWRITE)
349 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
350 NFSV3ACCESS_DELETE);
351 if (ap->a_mode & VEXEC)
352 mode |= NFSV3ACCESS_LOOKUP;
353 }
354 /* XXX safety belt, only make blanket request if caching */
355 if (nfsaccess_cache_timeout > 0) {
356 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
357 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
358 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
359 } else {
360 wmode = mode;
361 }
362
363 /*
364 * Does our cached result allow us to give a definite yes to
365 * this request?
366 */
367 if (np->n_modestamp &&
368 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
369 (ap->a_cred->cr_uid == np->n_modeuid) &&
370 ((np->n_mode & mode) == mode)) {
371 nfsstats.accesscache_hits++;
372 } else {
373 /*
374 * Either a no, or a don't know. Go to the wire.
375 */
376 nfsstats.accesscache_misses++;
377 error = nfs3_access_otw(vp, wmode, td, ap->a_cred);
378 if (!error) {
379 if ((np->n_mode & mode) != mode) {
380 error = EACCES;
381 }
382 }
383 }
384 } else {
385 if ((error = nfsspec_access(ap)) != 0)
386 return (error);
387
388 /*
389 * Attempt to prevent a mapped root from accessing a file
390 * which it shouldn't. We try to read a byte from the file
391 * if the user is root and the file is not zero length.
392 * After calling nfsspec_access, we should have the correct
393 * file size cached.
394 */
395 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
396 && VTONFS(vp)->n_size > 0) {
397 struct iovec aiov;
398 struct uio auio;
399 char buf[1];
400
401 aiov.iov_base = buf;
402 aiov.iov_len = 1;
403 auio.uio_iov = &aiov;
404 auio.uio_iovcnt = 1;
405 auio.uio_offset = 0;
406 auio.uio_resid = 1;
407 auio.uio_segflg = UIO_SYSSPACE;
408 auio.uio_rw = UIO_READ;
409 auio.uio_td = td;
410
411 if (vp->v_type == VREG) {
412 error = nfs_readrpc(vp, &auio);
413 } else if (vp->v_type == VDIR) {
414 char* bp;
415 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
416 aiov.iov_base = bp;
417 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
418 error = nfs_readdirrpc(vp, &auio);
419 kfree(bp, M_TEMP);
420 } else if (vp->v_type == VLNK) {
421 error = nfs_readlinkrpc(vp, &auio);
422 } else {
423 error = EACCES;
424 }
425 }
426 }
427 /*
428 * [re]record creds for reading and/or writing if access
429 * was granted. Assume the NFS server will grant read access
430 * for execute requests.
431 */
432 if (error == 0) {
433 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) {
434 crhold(ap->a_cred);
435 if (np->n_rucred)
436 crfree(np->n_rucred);
437 np->n_rucred = ap->a_cred;
438 }
439 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
440 crhold(ap->a_cred);
441 if (np->n_wucred)
442 crfree(np->n_wucred);
443 np->n_wucred = ap->a_cred;
444 }
445 }
446 return(error);
447}
448
449/*
450 * nfs open vnode op
451 * Check to see if the type is ok
452 * and that deletion is not in progress.
453 * For paged in text files, you will need to flush the page cache
454 * if consistency is lost.
455 *
456 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
457 * struct file *a_fp)
458 */
459/* ARGSUSED */
460static int
461nfs_open(struct vop_open_args *ap)
462{
463 struct vnode *vp = ap->a_vp;
464 struct nfsnode *np = VTONFS(vp);
465 struct vattr vattr;
466 int error;
467
468 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
469#ifdef DIAGNOSTIC
470 kprintf("open eacces vtyp=%d\n",vp->v_type);
471#endif
472 return (EOPNOTSUPP);
473 }
474
475 /*
476 * Save valid creds for reading and writing for later RPCs.
477 */
478 if ((ap->a_mode & FREAD) && ap->a_cred != np->n_rucred) {
479 crhold(ap->a_cred);
480 if (np->n_rucred)
481 crfree(np->n_rucred);
482 np->n_rucred = ap->a_cred;
483 }
484 if ((ap->a_mode & FWRITE) && ap->a_cred != np->n_wucred) {
485 crhold(ap->a_cred);
486 if (np->n_wucred)
487 crfree(np->n_wucred);
488 np->n_wucred = ap->a_cred;
489 }
490
491 /*
492 * Clear the attribute cache only if opening with write access. It
493 * is unclear if we should do this at all here, but we certainly
494 * should not clear the cache unconditionally simply because a file
495 * is being opened.
496 */
497 if (ap->a_mode & FWRITE)
498 np->n_attrstamp = 0;
499
500 /*
501 * For normal NFS, reconcile changes made locally verses
502 * changes made remotely. Note that VOP_GETATTR only goes
503 * to the wire if the cached attribute has timed out or been
504 * cleared.
505 *
506 * If local modifications have been made clear the attribute
507 * cache to force an attribute and modified time check. If
508 * GETATTR detects that the file has been changed by someone
509 * other then us it will set NRMODIFIED.
510 *
511 * If we are opening a directory and local changes have been
512 * made we have to invalidate the cache in order to ensure
513 * that we get the most up-to-date information from the
514 * server. XXX
515 */
516 if (np->n_flag & NLMODIFIED) {
517 np->n_attrstamp = 0;
518 if (vp->v_type == VDIR) {
519 error = nfs_vinvalbuf(vp, V_SAVE, 1);
520 if (error == EINTR)
521 return (error);
522 nfs_invaldir(vp);
523 }
524 }
525 error = VOP_GETATTR(vp, &vattr);
526 if (error)
527 return (error);
528 if (np->n_flag & NRMODIFIED) {
529 if (vp->v_type == VDIR)
530 nfs_invaldir(vp);
531 error = nfs_vinvalbuf(vp, V_SAVE, 1);
532 if (error == EINTR)
533 return (error);
534 np->n_flag &= ~NRMODIFIED;
535 }
536
537 return (vop_stdopen(ap));
538}
539
540/*
541 * nfs close vnode op
542 * What an NFS client should do upon close after writing is a debatable issue.
543 * Most NFS clients push delayed writes to the server upon close, basically for
544 * two reasons:
545 * 1 - So that any write errors may be reported back to the client process
546 * doing the close system call. By far the two most likely errors are
547 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
548 * 2 - To put a worst case upper bound on cache inconsistency between
549 * multiple clients for the file.
550 * There is also a consistency problem for Version 2 of the protocol w.r.t.
551 * not being able to tell if other clients are writing a file concurrently,
552 * since there is no way of knowing if the changed modify time in the reply
553 * is only due to the write for this client.
554 * (NFS Version 3 provides weak cache consistency data in the reply that
555 * should be sufficient to detect and handle this case.)
556 *
557 * The current code does the following:
558 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
559 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
560 * or commit them (this satisfies 1 and 2 except for the
561 * case where the server crashes after this close but
562 * before the commit RPC, which is felt to be "good
563 * enough". Changing the last argument to nfs_flush() to
564 * a 1 would force a commit operation, if it is felt a
565 * commit is necessary now.
566 * for NQNFS - do nothing now, since 2 is dealt with via leases and
567 * 1 should be dealt with via an fsync() system call for
568 * cases where write errors are important.
569 *
570 * nfs_close(struct vnode *a_vp, int a_fflag)
571 */
572/* ARGSUSED */
573static int
574nfs_close(struct vop_close_args *ap)
575{
576 struct vnode *vp = ap->a_vp;
577 struct nfsnode *np = VTONFS(vp);
578 int error = 0;
579 thread_t td = curthread;
580
581 if (vp->v_type == VREG) {
582 if (np->n_flag & NLMODIFIED) {
583 if (NFS_ISV3(vp)) {
584 /*
585 * Under NFSv3 we have dirty buffers to dispose of. We
586 * must flush them to the NFS server. We have the option
587 * of waiting all the way through the commit rpc or just
588 * waiting for the initial write. The default is to only
589 * wait through the initial write so the data is in the
590 * server's cache, which is roughly similar to the state
591 * a standard disk subsystem leaves the file in on close().
592 *
593 * We cannot clear the NLMODIFIED bit in np->n_flag due to
594 * potential races with other processes, and certainly
595 * cannot clear it if we don't commit.
596 */
597 int cm = nfsv3_commit_on_close ? 1 : 0;
598 error = nfs_flush(vp, MNT_WAIT, td, cm);
599 /* np->n_flag &= ~NLMODIFIED; */
600 } else {
601 error = nfs_vinvalbuf(vp, V_SAVE, 1);
602 }
603 np->n_attrstamp = 0;
604 }
605 if (np->n_flag & NWRITEERR) {
606 np->n_flag &= ~NWRITEERR;
607 error = np->n_error;
608 }
609 }
610 vop_stdclose(ap);
611 return (error);
612}
613
614/*
615 * nfs getattr call from vfs.
616 *
617 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
618 */
619static int
620nfs_getattr(struct vop_getattr_args *ap)
621{
622 struct vnode *vp = ap->a_vp;
623 struct nfsnode *np = VTONFS(vp);
624 caddr_t cp;
625 u_int32_t *tl;
626 int32_t t1, t2;
627 caddr_t bpos, dpos;
628 int error = 0;
629 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
630 int v3 = NFS_ISV3(vp);
631 thread_t td = curthread;
632
633 /*
634 * Update local times for special files.
635 */
636 if (np->n_flag & (NACC | NUPD))
637 np->n_flag |= NCHG;
638 /*
639 * First look in the cache.
640 */
641 if (nfs_getattrcache(vp, ap->a_vap) == 0)
642 return (0);
643
644 if (v3 && nfsaccess_cache_timeout > 0) {
645 nfsstats.accesscache_misses++;
646 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
647 if (nfs_getattrcache(vp, ap->a_vap) == 0)
648 return (0);
649 }
650
651 nfsstats.rpccnt[NFSPROC_GETATTR]++;
652 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
653 nfsm_fhtom(vp, v3);
654 nfsm_request(vp, NFSPROC_GETATTR, td, nfs_vpcred(vp, ND_CHECK));
655 if (!error) {
656 nfsm_loadattr(vp, ap->a_vap);
657 }
658 m_freem(mrep);
659nfsmout:
660 return (error);
661}
662
663/*
664 * nfs setattr call.
665 *
666 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
667 */
668static int
669nfs_setattr(struct vop_setattr_args *ap)
670{
671 struct vnode *vp = ap->a_vp;
672 struct nfsnode *np = VTONFS(vp);
673 struct vattr *vap = ap->a_vap;
674 int error = 0;
675 u_quad_t tsize;
676 thread_t td = curthread;
677
678#ifndef nolint
679 tsize = (u_quad_t)0;
680#endif
681
682 /*
683 * Setting of flags is not supported.
684 */
685 if (vap->va_flags != VNOVAL)
686 return (EOPNOTSUPP);
687
688 /*
689 * Disallow write attempts if the filesystem is mounted read-only.
690 */
691 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
692 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
693 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
694 (vp->v_mount->mnt_flag & MNT_RDONLY))
695 return (EROFS);
696
697 if (vap->va_size != VNOVAL) {
698 /*
699 * truncation requested
700 */
701 switch (vp->v_type) {
702 case VDIR:
703 return (EISDIR);
704 case VCHR:
705 case VBLK:
706 case VSOCK:
707 case VFIFO:
708 if (vap->va_mtime.tv_sec == VNOVAL &&
709 vap->va_atime.tv_sec == VNOVAL &&
710 vap->va_mode == (mode_t)VNOVAL &&
711 vap->va_uid == (uid_t)VNOVAL &&
712 vap->va_gid == (gid_t)VNOVAL)
713 return (0);
714 vap->va_size = VNOVAL;
715 break;
716 default:
717 /*
718 * Disallow write attempts if the filesystem is
719 * mounted read-only.
720 */
721 if (vp->v_mount->mnt_flag & MNT_RDONLY)
722 return (EROFS);
723
724 /*
725 * This is nasty. The RPCs we send to flush pending
726 * data often return attribute information which is
727 * cached via a callback to nfs_loadattrcache(), which
728 * has the effect of changing our notion of the file
729 * size. Due to flushed appends and other operations
730 * the file size can be set to virtually anything,
731 * including values that do not match either the old
732 * or intended file size.
733 *
734 * When this condition is detected we must loop to
735 * try the operation again. Hopefully no more
736 * flushing is required on the loop so it works the
737 * second time around. THIS CASE ALMOST ALWAYS
738 * HAPPENS!
739 */
740 tsize = np->n_size;
741again:
742 error = nfs_meta_setsize(vp, td, vap->va_size);
743
744 if (np->n_flag & NLMODIFIED) {
745 if (vap->va_size == 0)
746 error = nfs_vinvalbuf(vp, 0, 1);
747 else
748 error = nfs_vinvalbuf(vp, V_SAVE, 1);
749 }
750 /*
751 * note: this loop case almost always happens at
752 * least once per truncation.
753 */
754 if (error == 0 && np->n_size != vap->va_size)
755 goto again;
756 np->n_vattr.va_size = vap->va_size;
757 break;
758 }
759 } else if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
760 /*
761 * What to do. If we are modifying the mtime we lose
762 * mtime detection of changes made by the server or other
763 * clients. But programs like rsync/rdist/cpdup are going
764 * to call utimes a lot. We don't want to piecemeal sync.
765 *
766 * For now sync if any prior remote changes were detected,
767 * but allow us to lose track of remote changes made during
768 * the utimes operation.
769 */
770 if (np->n_flag & NRMODIFIED)
771 error = nfs_vinvalbuf(vp, V_SAVE, 1);
772 if (error == EINTR)
773 return (error);
774 if (error == 0) {
775 if (vap->va_mtime.tv_sec != VNOVAL) {
776 np->n_mtime = vap->va_mtime.tv_sec;
777 }
778 }
779 }
780 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
781
782 /*
783 * Sanity check if a truncation was issued. This should only occur
784 * if multiple processes are racing on the same file.
785 */
786 if (error == 0 && vap->va_size != VNOVAL &&
787 np->n_size != vap->va_size) {
788 kprintf("NFS ftruncate: server disagrees on the file size: "
789 "%lld/%lld/%lld\n",
790 (long long)tsize,
791 (long long)vap->va_size,
792 (long long)np->n_size);
793 goto again;
794 }
795 if (error && vap->va_size != VNOVAL) {
796 np->n_size = np->n_vattr.va_size = tsize;
797 vnode_pager_setsize(vp, np->n_size);
798 }
799 return (error);
800}
801
802/*
803 * Do an nfs setattr rpc.
804 */
805static int
806nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
807 struct ucred *cred, struct thread *td)
808{
809 struct nfsv2_sattr *sp;
810 struct nfsnode *np = VTONFS(vp);
811 caddr_t cp;
812 int32_t t1, t2;
813 caddr_t bpos, dpos, cp2;
814 u_int32_t *tl;
815 int error = 0, wccflag = NFSV3_WCCRATTR;
816 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
817 int v3 = NFS_ISV3(vp);
818
819 nfsstats.rpccnt[NFSPROC_SETATTR]++;
820 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
821 nfsm_fhtom(vp, v3);
822 if (v3) {
823 nfsm_v3attrbuild(vap, TRUE);
824 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
825 *tl = nfs_false;
826 } else {
827 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
828 if (vap->va_mode == (mode_t)VNOVAL)
829 sp->sa_mode = nfs_xdrneg1;
830 else
831 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
832 if (vap->va_uid == (uid_t)VNOVAL)
833 sp->sa_uid = nfs_xdrneg1;
834 else
835 sp->sa_uid = txdr_unsigned(vap->va_uid);
836 if (vap->va_gid == (gid_t)VNOVAL)
837 sp->sa_gid = nfs_xdrneg1;
838 else
839 sp->sa_gid = txdr_unsigned(vap->va_gid);
840 sp->sa_size = txdr_unsigned(vap->va_size);
841 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
842 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
843 }
844 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
845 if (v3) {
846 np->n_modestamp = 0;
847 nfsm_wcc_data(vp, wccflag);
848 } else
849 nfsm_loadattr(vp, NULL);
850 m_freem(mrep);
851nfsmout:
852 return (error);
853}
854
855static
856void
857nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
858{
859 if (nctimeout == 0)
860 nctimeout = 1;
861 else
862 nctimeout *= hz;
863 cache_setvp(nch, vp);
864 cache_settimeout(nch, nctimeout);
865}
866
867/*
868 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
869 * nfs_lookup() until all remaining new api calls are implemented.
870 *
871 * Resolve a namecache entry. This function is passed a locked ncp and
872 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
873 */
874static int
875nfs_nresolve(struct vop_nresolve_args *ap)
876{
877 struct thread *td = curthread;
878 struct namecache *ncp;
879 struct ucred *cred;
880 struct nfsnode *np;
881 struct vnode *dvp;
882 struct vnode *nvp;
883 nfsfh_t *fhp;
884 int attrflag;
885 int fhsize;
886 int error;
887 int len;
888 int v3;
889 /******NFSM MACROS********/
890 struct mbuf *mb, *mrep, *mreq, *mb2, *md;
891 caddr_t bpos, dpos, cp, cp2;
892 u_int32_t *tl;
893 int32_t t1, t2;
894
895 cred = ap->a_cred;
896 dvp = ap->a_dvp;
897
898 if ((error = vget(dvp, LK_SHARED)) != 0)
899 return (error);
900
901 nvp = NULL;
902 v3 = NFS_ISV3(dvp);
903 nfsstats.lookupcache_misses++;
904 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
905 ncp = ap->a_nch->ncp;
906 len = ncp->nc_nlen;
907 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
908 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
909 nfsm_fhtom(dvp, v3);
910 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN);
911 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred);
912 if (error) {
913 /*
914 * Cache negatve lookups to reduce NFS traffic, but use
915 * a fast timeout. Otherwise use a timeout of 1 tick.
916 * XXX we should add a namecache flag for no-caching
917 * to uncache the negative hit as soon as possible, but
918 * we cannot simply destroy the entry because it is used
919 * as a placeholder by the caller.
920 */
921 if (error == ENOENT)
922 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
923 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
924 m_freem(mrep);
925 goto nfsmout;
926 }
927
928 /*
929 * Success, get the file handle, do various checks, and load
930 * post-operation data from the reply packet. Theoretically
931 * we should never be looking up "." so, theoretically, we
932 * should never get the same file handle as our directory. But
933 * we check anyway. XXX
934 *
935 * Note that no timeout is set for the positive cache hit. We
936 * assume, theoretically, that ESTALE returns will be dealt with
937 * properly to handle NFS races and in anycase we cannot depend
938 * on a timeout to deal with NFS open/create/excl issues so instead
939 * of a bad hack here the rest of the NFS client code needs to do
940 * the right thing.
941 */
942 nfsm_getfh(fhp, fhsize, v3);
943
944 np = VTONFS(dvp);
945 if (NFS_CMPFH(np, fhp, fhsize)) {
946 vref(dvp);
947 nvp = dvp;
948 } else {
949 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
950 if (error) {
951 m_freem(mrep);
952 vput(dvp);
953 return (error);
954 }
955 nvp = NFSTOV(np);
956 }
957 if (v3) {
958 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK);
959 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
960 } else {
961 nfsm_loadattr(nvp, NULL);
962 }
963 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
964 m_freem(mrep);
965nfsmout:
966 vput(dvp);
967 if (nvp) {
968 if (nvp == dvp)
969 vrele(nvp);
970 else
971 vput(nvp);
972 }
973 return (error);
974}
975
976/*
977 * 'cached' nfs directory lookup
978 *
979 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
980 *
981 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
982 * struct componentname *a_cnp)
983 */
984static int
985nfs_lookup(struct vop_old_lookup_args *ap)
986{
987 struct componentname *cnp = ap->a_cnp;
988 struct vnode *dvp = ap->a_dvp;
989 struct vnode **vpp = ap->a_vpp;
990 int flags = cnp->cn_flags;
991 struct vnode *newvp;
992 u_int32_t *tl;
993 caddr_t cp;
994 int32_t t1, t2;
995 struct nfsmount *nmp;
996 caddr_t bpos, dpos, cp2;
997 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
998 long len;
999 nfsfh_t *fhp;
1000 struct nfsnode *np;
1001 int lockparent, wantparent, error = 0, attrflag, fhsize;
1002 int v3 = NFS_ISV3(dvp);
1003
1004 /*
1005 * Read-only mount check and directory check.
1006 */
1007 *vpp = NULLVP;
1008 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1009 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1010 return (EROFS);
1011
1012 if (dvp->v_type != VDIR)
1013 return (ENOTDIR);
1014
1015 /*
1016 * Look it up in the cache. Note that ENOENT is only returned if we
1017 * previously entered a negative hit (see later on). The additional
1018 * nfsneg_cache_timeout check causes previously cached results to
1019 * be instantly ignored if the negative caching is turned off.
1020 */
1021 lockparent = flags & CNP_LOCKPARENT;
1022 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1023 nmp = VFSTONFS(dvp->v_mount);
1024 np = VTONFS(dvp);
1025
1026 /*
1027 * Go to the wire.
1028 */
1029 error = 0;
1030 newvp = NULLVP;
1031 nfsstats.lookupcache_misses++;
1032 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1033 len = cnp->cn_namelen;
1034 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1035 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1036 nfsm_fhtom(dvp, v3);
1037 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1038 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred);
1039 if (error) {
1040 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1041 m_freem(mrep);
1042 goto nfsmout;
1043 }
1044 nfsm_getfh(fhp, fhsize, v3);
1045
1046 /*
1047 * Handle RENAME case...
1048 */
1049 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1050 if (NFS_CMPFH(np, fhp, fhsize)) {
1051 m_freem(mrep);
1052 return (EISDIR);
1053 }
1054 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1055 if (error) {
1056 m_freem(mrep);
1057 return (error);
1058 }
1059 newvp = NFSTOV(np);
1060 if (v3) {
1061 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1062 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1063 } else
1064 nfsm_loadattr(newvp, NULL);
1065 *vpp = newvp;
1066 m_freem(mrep);
1067 if (!lockparent) {
1068 vn_unlock(dvp);
1069 cnp->cn_flags |= CNP_PDIRUNLOCK;
1070 }
1071 return (0);
1072 }
1073
1074 if (flags & CNP_ISDOTDOT) {
1075 vn_unlock(dvp);
1076 cnp->cn_flags |= CNP_PDIRUNLOCK;
1077 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1078 if (error) {
1079 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1080 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1081 return (error); /* NOTE: return error from nget */
1082 }
1083 newvp = NFSTOV(np);
1084 if (lockparent) {
1085 error = vn_lock(dvp, LK_EXCLUSIVE);
1086 if (error) {
1087 vput(newvp);
1088 return (error);
1089 }
1090 cnp->cn_flags |= CNP_PDIRUNLOCK;
1091 }
1092 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1093 vref(dvp);
1094 newvp = dvp;
1095 } else {
1096 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1097 if (error) {
1098 m_freem(mrep);
1099 return (error);
1100 }
1101 if (!lockparent) {
1102 vn_unlock(dvp);
1103 cnp->cn_flags |= CNP_PDIRUNLOCK;
1104 }
1105 newvp = NFSTOV(np);
1106 }
1107 if (v3) {
1108 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1109 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1110 } else
1111 nfsm_loadattr(newvp, NULL);
1112#if 0
1113 /* XXX MOVE TO nfs_nremove() */
1114 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1115 cnp->cn_nameiop != NAMEI_DELETE) {
1116 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1117 }
1118#endif
1119 *vpp = newvp;
1120 m_freem(mrep);
1121nfsmout:
1122 if (error) {
1123 if (newvp != NULLVP) {
1124 vrele(newvp);
1125 *vpp = NULLVP;
1126 }
1127 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1128 cnp->cn_nameiop == NAMEI_RENAME) &&
1129 error == ENOENT) {
1130 if (!lockparent) {
1131 vn_unlock(dvp);
1132 cnp->cn_flags |= CNP_PDIRUNLOCK;
1133 }
1134 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1135 error = EROFS;
1136 else
1137 error = EJUSTRETURN;
1138 }
1139 }
1140 return (error);
1141}
1142
1143/*
1144 * nfs read call.
1145 * Just call nfs_bioread() to do the work.
1146 *
1147 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1148 * struct ucred *a_cred)
1149 */
1150static int
1151nfs_read(struct vop_read_args *ap)
1152{
1153 struct vnode *vp = ap->a_vp;
1154
1155 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1156}
1157
1158/*
1159 * nfs readlink call
1160 *
1161 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1162 */
1163static int
1164nfs_readlink(struct vop_readlink_args *ap)
1165{
1166 struct vnode *vp = ap->a_vp;
1167
1168 if (vp->v_type != VLNK)
1169 return (EINVAL);
1170 return (nfs_bioread(vp, ap->a_uio, 0));
1171}
1172
1173/*
1174 * Do a readlink rpc.
1175 * Called by nfs_doio() from below the buffer cache.
1176 */
1177int
1178nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1179{
1180 u_int32_t *tl;
1181 caddr_t cp;
1182 int32_t t1, t2;
1183 caddr_t bpos, dpos, cp2;
1184 int error = 0, len, attrflag;
1185 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1186 int v3 = NFS_ISV3(vp);
1187
1188 nfsstats.rpccnt[NFSPROC_READLINK]++;
1189 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1190 nfsm_fhtom(vp, v3);
1191 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK));
1192 if (v3)
1193 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1194 if (!error) {
1195 nfsm_strsiz(len, NFS_MAXPATHLEN);
1196 if (len == NFS_MAXPATHLEN) {
1197 struct nfsnode *np = VTONFS(vp);
1198 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1199 len = np->n_size;
1200 }
1201 nfsm_mtouio(uiop, len);
1202 }
1203 m_freem(mrep);
1204nfsmout:
1205 return (error);
1206}
1207
1208/*
1209 * nfs read rpc call
1210 * Ditto above
1211 */
1212int
1213nfs_readrpc(struct vnode *vp, struct uio *uiop)
1214{
1215 u_int32_t *tl;
1216 caddr_t cp;
1217 int32_t t1, t2;
1218 caddr_t bpos, dpos, cp2;
1219 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1220 struct nfsmount *nmp;
1221 int error = 0, len, retlen, tsiz, eof, attrflag;
1222 int v3 = NFS_ISV3(vp);
1223
1224#ifndef nolint
1225 eof = 0;
1226#endif
1227 nmp = VFSTONFS(vp->v_mount);
1228 tsiz = uiop->uio_resid;
1229 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1230 return (EFBIG);
1231 while (tsiz > 0) {
1232 nfsstats.rpccnt[NFSPROC_READ]++;
1233 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1234 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1235 nfsm_fhtom(vp, v3);
1236 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1237 if (v3) {
1238 txdr_hyper(uiop->uio_offset, tl);
1239 *(tl + 2) = txdr_unsigned(len);
1240 } else {
1241 *tl++ = txdr_unsigned(uiop->uio_offset);
1242 *tl++ = txdr_unsigned(len);
1243 *tl = 0;
1244 }
1245 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ));
1246 if (v3) {
1247 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1248 if (error) {
1249 m_freem(mrep);
1250 goto nfsmout;
1251 }
1252 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1253 eof = fxdr_unsigned(int, *(tl + 1));
1254 } else
1255 nfsm_loadattr(vp, NULL);
1256 nfsm_strsiz(retlen, nmp->nm_rsize);
1257 nfsm_mtouio(uiop, retlen);
1258 m_freem(mrep);
1259 tsiz -= retlen;
1260 if (v3) {
1261 if (eof || retlen == 0) {
1262 tsiz = 0;
1263 }
1264 } else if (retlen < len) {
1265 tsiz = 0;
1266 }
1267 }
1268nfsmout:
1269 return (error);
1270}
1271
1272/*
1273 * nfs write call
1274 */
1275int
1276nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1277{
1278 u_int32_t *tl;
1279 caddr_t cp;
1280 int32_t t1, t2, backup;
1281 caddr_t bpos, dpos, cp2;
1282 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1283 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1284 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1285 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1286
1287#ifndef DIAGNOSTIC
1288 if (uiop->uio_iovcnt != 1)
1289 panic("nfs: writerpc iovcnt > 1");
1290#endif
1291 *must_commit = 0;
1292 tsiz = uiop->uio_resid;
1293 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1294 return (EFBIG);
1295 while (tsiz > 0) {
1296 nfsstats.rpccnt[NFSPROC_WRITE]++;
1297 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1298 nfsm_reqhead(vp, NFSPROC_WRITE,
1299 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1300 nfsm_fhtom(vp, v3);
1301 if (v3) {
1302 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1303 txdr_hyper(uiop->uio_offset, tl);
1304 tl += 2;
1305 *tl++ = txdr_unsigned(len);
1306 *tl++ = txdr_unsigned(*iomode);
1307 *tl = txdr_unsigned(len);
1308 } else {
1309 u_int32_t x;
1310
1311 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1312 /* Set both "begin" and "current" to non-garbage. */
1313 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1314 *tl++ = x; /* "begin offset" */
1315 *tl++ = x; /* "current offset" */
1316 x = txdr_unsigned(len);
1317 *tl++ = x; /* total to this offset */
1318 *tl = x; /* size of this write */
1319 }
1320 nfsm_uiotom(uiop, len);
1321 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE));
1322 if (v3) {
1323 /*
1324 * The write RPC returns a before and after mtime. The
1325 * nfsm_wcc_data() macro checks the before n_mtime
1326 * against the before time and stores the after time
1327 * in the nfsnode's cached vattr and n_mtime field.
1328 * The NRMODIFIED bit will be set if the before
1329 * time did not match the original mtime.
1330 */
1331 wccflag = NFSV3_WCCCHK;
1332 nfsm_wcc_data(vp, wccflag);
1333 if (!error) {
1334 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1335 + NFSX_V3WRITEVERF);
1336 rlen = fxdr_unsigned(int, *tl++);
1337 if (rlen == 0) {
1338 error = NFSERR_IO;
1339 m_freem(mrep);
1340 break;
1341 } else if (rlen < len) {
1342 backup = len - rlen;
1343 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1344 uiop->uio_iov->iov_len += backup;
1345 uiop->uio_offset -= backup;
1346 uiop->uio_resid += backup;
1347 len = rlen;
1348 }
1349 commit = fxdr_unsigned(int, *tl++);
1350
1351 /*
1352 * Return the lowest committment level
1353 * obtained by any of the RPCs.
1354 */
1355 if (committed == NFSV3WRITE_FILESYNC)
1356 committed = commit;
1357 else if (committed == NFSV3WRITE_DATASYNC &&
1358 commit == NFSV3WRITE_UNSTABLE)
1359 committed = commit;
1360 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1361 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1362 NFSX_V3WRITEVERF);
1363 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1364 } else if (bcmp((caddr_t)tl,
1365 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1366 *must_commit = 1;
1367 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1368 NFSX_V3WRITEVERF);
1369 }
1370 }
1371 } else {
1372 nfsm_loadattr(vp, NULL);
1373 }
1374 m_freem(mrep);
1375 if (error)
1376 break;
1377 tsiz -= len;
1378 }
1379nfsmout:
1380 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1381 committed = NFSV3WRITE_FILESYNC;
1382 *iomode = committed;
1383 if (error)
1384 uiop->uio_resid = tsiz;
1385 return (error);
1386}
1387
1388/*
1389 * nfs mknod rpc
1390 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1391 * mode set to specify the file type and the size field for rdev.
1392 */
1393static int
1394nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1395 struct vattr *vap)
1396{
1397 struct nfsv2_sattr *sp;
1398 u_int32_t *tl;
1399 caddr_t cp;
1400 int32_t t1, t2;
1401 struct vnode *newvp = NULL;
1402 struct nfsnode *np = NULL;
1403 struct vattr vattr;
1404 char *cp2;
1405 caddr_t bpos, dpos;
1406 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1407 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1408 int rmajor, rminor;
1409 int v3 = NFS_ISV3(dvp);
1410
1411 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1412 rmajor = txdr_unsigned(vap->va_rmajor);
1413 rminor = txdr_unsigned(vap->va_rminor);
1414 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1415 rmajor = nfs_xdrneg1;
1416 rminor = nfs_xdrneg1;
1417 } else {
1418 return (EOPNOTSUPP);
1419 }
1420 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1421 return (error);
1422 }
1423 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1424 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1425 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1426 nfsm_fhtom(dvp, v3);
1427 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1428 if (v3) {
1429 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1430 *tl++ = vtonfsv3_type(vap->va_type);
1431 nfsm_v3attrbuild(vap, FALSE);
1432 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1433 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1434 *tl++ = txdr_unsigned(vap->va_rmajor);
1435 *tl = txdr_unsigned(vap->va_rminor);
1436 }
1437 } else {
1438 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1439 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1440 sp->sa_uid = nfs_xdrneg1;
1441 sp->sa_gid = nfs_xdrneg1;
1442 sp->sa_size = makeudev(rmajor, rminor);
1443 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1444 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1445 }
1446 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred);
1447 if (!error) {
1448 nfsm_mtofh(dvp, newvp, v3, gotvp);
1449 if (!gotvp) {
1450 if (newvp) {
1451 vput(newvp);
1452 newvp = NULL;
1453 }
1454 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1455 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1456 if (!error)
1457 newvp = NFSTOV(np);
1458 }
1459 }
1460 if (v3)
1461 nfsm_wcc_data(dvp, wccflag);
1462 m_freem(mrep);
1463nfsmout:
1464 if (error) {
1465 if (newvp)
1466 vput(newvp);
1467 } else {
1468 *vpp = newvp;
1469 }
1470 VTONFS(dvp)->n_flag |= NLMODIFIED;
1471 if (!wccflag)
1472 VTONFS(dvp)->n_attrstamp = 0;
1473 return (error);
1474}
1475
1476/*
1477 * nfs mknod vop
1478 * just call nfs_mknodrpc() to do the work.
1479 *
1480 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1481 * struct componentname *a_cnp, struct vattr *a_vap)
1482 */
1483/* ARGSUSED */
1484static int
1485nfs_mknod(struct vop_old_mknod_args *ap)
1486{
1487 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1488}
1489
1490static u_long create_verf;
1491/*
1492 * nfs file create call
1493 *
1494 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1495 * struct componentname *a_cnp, struct vattr *a_vap)
1496 */
1497static int
1498nfs_create(struct vop_old_create_args *ap)
1499{
1500 struct vnode *dvp = ap->a_dvp;
1501 struct vattr *vap = ap->a_vap;
1502 struct componentname *cnp = ap->a_cnp;
1503 struct nfsv2_sattr *sp;
1504 u_int32_t *tl;
1505 caddr_t cp;
1506 int32_t t1, t2;
1507 struct nfsnode *np = NULL;
1508 struct vnode *newvp = NULL;
1509 caddr_t bpos, dpos, cp2;
1510 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1511 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1512 struct vattr vattr;
1513 int v3 = NFS_ISV3(dvp);
1514
1515 /*
1516 * Oops, not for me..
1517 */
1518 if (vap->va_type == VSOCK)
1519 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1520
1521 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1522 return (error);
1523 }
1524 if (vap->va_vaflags & VA_EXCLUSIVE)
1525 fmode |= O_EXCL;
1526again:
1527 nfsstats.rpccnt[NFSPROC_CREATE]++;
1528 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1529 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1530 nfsm_fhtom(dvp, v3);
1531 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1532 if (v3) {
1533 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1534 if (fmode & O_EXCL) {
1535 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1536 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1537#ifdef INET
1538 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1539 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1540 else
1541#endif
1542 *tl++ = create_verf;
1543 *tl = ++create_verf;
1544 } else {
1545 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1546 nfsm_v3attrbuild(vap, FALSE);
1547 }
1548 } else {
1549 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1550 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1551 sp->sa_uid = nfs_xdrneg1;
1552 sp->sa_gid = nfs_xdrneg1;
1553 sp->sa_size = 0;
1554 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1555 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1556 }
1557 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred);
1558 if (!error) {
1559 nfsm_mtofh(dvp, newvp, v3, gotvp);
1560 if (!gotvp) {
1561 if (newvp) {
1562 vput(newvp);
1563 newvp = NULL;
1564 }
1565 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1566 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1567 if (!error)
1568 newvp = NFSTOV(np);
1569 }
1570 }
1571 if (v3)
1572 nfsm_wcc_data(dvp, wccflag);
1573 m_freem(mrep);
1574nfsmout:
1575 if (error) {
1576 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1577 KKASSERT(newvp == NULL);
1578 fmode &= ~O_EXCL;
1579 goto again;
1580 }
1581 } else if (v3 && (fmode & O_EXCL)) {
1582 /*
1583 * We are normally called with only a partially initialized
1584 * VAP. Since the NFSv3 spec says that server may use the
1585 * file attributes to store the verifier, the spec requires
1586 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1587 * in atime, but we can't really assume that all servers will
1588 * so we ensure that our SETATTR sets both atime and mtime.
1589 */
1590 if (vap->va_mtime.tv_sec == VNOVAL)
1591 vfs_timestamp(&vap->va_mtime);
1592 if (vap->va_atime.tv_sec == VNOVAL)
1593 vap->va_atime = vap->va_mtime;
1594 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1595 }
1596 if (error == 0) {
1597 /*
1598 * The new np may have enough info for access
1599 * checks, make sure rucred and wucred are
1600 * initialized for read and write rpc's.
1601 */
1602 np = VTONFS(newvp);
1603 if (np->n_rucred == NULL)
1604 np->n_rucred = crhold(cnp->cn_cred);
1605 if (np->n_wucred == NULL)
1606 np->n_wucred = crhold(cnp->cn_cred);
1607 *ap->a_vpp = newvp;
1608 } else if (newvp) {
1609 vput(newvp);
1610 }
1611 VTONFS(dvp)->n_flag |= NLMODIFIED;
1612 if (!wccflag)
1613 VTONFS(dvp)->n_attrstamp = 0;
1614 return (error);
1615}
1616
1617/*
1618 * nfs file remove call
1619 * To try and make nfs semantics closer to ufs semantics, a file that has
1620 * other processes using the vnode is renamed instead of removed and then
1621 * removed later on the last close.
1622 * - If v_sysref.refcnt > 1
1623 * If a rename is not already in the works
1624 * call nfs_sillyrename() to set it up
1625 * else
1626 * do the remove rpc
1627 *
1628 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1629 * struct componentname *a_cnp)
1630 */
1631static int
1632nfs_remove(struct vop_old_remove_args *ap)
1633{
1634 struct vnode *vp = ap->a_vp;
1635 struct vnode *dvp = ap->a_dvp;
1636 struct componentname *cnp = ap->a_cnp;
1637 struct nfsnode *np = VTONFS(vp);
1638 int error = 0;
1639 struct vattr vattr;
1640
1641#ifndef DIAGNOSTIC
1642 if (vp->v_sysref.refcnt < 1)
1643 panic("nfs_remove: bad v_sysref.refcnt");
1644#endif
1645 if (vp->v_type == VDIR)
1646 error = EPERM;
1647 else if (vp->v_sysref.refcnt == 1 || (np->n_sillyrename &&
1648 VOP_GETATTR(vp, &vattr) == 0 &&
1649 vattr.va_nlink > 1)) {
1650 /*
1651 * throw away biocache buffers, mainly to avoid
1652 * unnecessary delayed writes later.
1653 */
1654 error = nfs_vinvalbuf(vp, 0, 1);
1655 /* Do the rpc */
1656 if (error != EINTR)
1657 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1658 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1659 /*
1660 * Kludge City: If the first reply to the remove rpc is lost..
1661 * the reply to the retransmitted request will be ENOENT
1662 * since the file was in fact removed
1663 * Therefore, we cheat and return success.
1664 */
1665 if (error == ENOENT)
1666 error = 0;
1667 } else if (!np->n_sillyrename) {
1668 error = nfs_sillyrename(dvp, vp, cnp);
1669 }
1670 np->n_attrstamp = 0;
1671 return (error);
1672}
1673
1674/*
1675 * nfs file remove rpc called from nfs_inactive
1676 */
1677int
1678nfs_removeit(struct sillyrename *sp)
1679{
1680 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1681 sp->s_cred, NULL));
1682}
1683
1684/*
1685 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1686 */
1687static int
1688nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1689 struct ucred *cred, struct thread *td)
1690{
1691 u_int32_t *tl;
1692 caddr_t cp;
1693 int32_t t1, t2;
1694 caddr_t bpos, dpos, cp2;
1695 int error = 0, wccflag = NFSV3_WCCRATTR;
1696 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1697 int v3 = NFS_ISV3(dvp);
1698
1699 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1700 nfsm_reqhead(dvp, NFSPROC_REMOVE,
1701 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1702 nfsm_fhtom(dvp, v3);
1703 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1704 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1705 if (v3)
1706 nfsm_wcc_data(dvp, wccflag);
1707 m_freem(mrep);
1708nfsmout:
1709 VTONFS(dvp)->n_flag |= NLMODIFIED;
1710 if (!wccflag)
1711 VTONFS(dvp)->n_attrstamp = 0;
1712 return (error);
1713}
1714
1715/*
1716 * nfs file rename call
1717 *
1718 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1719 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1720 * struct vnode *a_tvp, struct componentname *a_tcnp)
1721 */
1722static int
1723nfs_rename(struct vop_old_rename_args *ap)
1724{
1725 struct vnode *fvp = ap->a_fvp;
1726 struct vnode *tvp = ap->a_tvp;
1727 struct vnode *fdvp = ap->a_fdvp;
1728 struct vnode *tdvp = ap->a_tdvp;
1729 struct componentname *tcnp = ap->a_tcnp;
1730 struct componentname *fcnp = ap->a_fcnp;
1731 int error;
1732
1733 /* Check for cross-device rename */
1734 if ((fvp->v_mount != tdvp->v_mount) ||
1735 (tvp && (fvp->v_mount != tvp->v_mount))) {
1736 error = EXDEV;
1737 goto out;
1738 }
1739
1740 /*
1741 * We shouldn't have to flush fvp on rename for most server-side
1742 * filesystems as the file handle should not change. Unfortunately
1743 * the inode for some filesystems (msdosfs) might be tied to the
1744 * file name or directory position so to be completely safe
1745 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1746 * performance.
1747 *
1748 * We must flush tvp on rename because it might become stale on the
1749 * server after the rename.
1750 */
1751 if (nfs_flush_on_rename)
1752 VOP_FSYNC(fvp, MNT_WAIT);
1753 if (tvp)
1754 VOP_FSYNC(tvp, MNT_WAIT);
1755
1756 /*
1757 * If the tvp exists and is in use, sillyrename it before doing the
1758 * rename of the new file over it.
1759 *
1760 * XXX Can't sillyrename a directory.
1761 *
1762 * We do not attempt to do any namecache purges in this old API
1763 * routine. The new API compat functions have access to the actual
1764 * namecache structures and will do it for us.
1765 */
1766 if (tvp && tvp->v_sysref.refcnt > 1 && !VTONFS(tvp)->n_sillyrename &&
1767 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1768 vput(tvp);
1769 tvp = NULL;
1770 } else if (tvp) {
1771 ;
1772 }
1773
1774 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1775 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1776 tcnp->cn_td);
1777
1778out:
1779 if (tdvp == tvp)
1780 vrele(tdvp);
1781 else
1782 vput(tdvp);
1783 if (tvp)
1784 vput(tvp);
1785 vrele(fdvp);
1786 vrele(fvp);
1787 /*
1788 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1789 */
1790 if (error == ENOENT)
1791 error = 0;
1792 return (error);
1793}
1794
1795/*
1796 * nfs file rename rpc called from nfs_remove() above
1797 */
1798static int
1799nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1800 struct sillyrename *sp)
1801{
1802 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1803 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1804}
1805
1806/*
1807 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1808 */
1809static int
1810nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1811 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1812 struct ucred *cred, struct thread *td)
1813{
1814 u_int32_t *tl;
1815 caddr_t cp;
1816 int32_t t1, t2;
1817 caddr_t bpos, dpos, cp2;
1818 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1819 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1820 int v3 = NFS_ISV3(fdvp);
1821
1822 nfsstats.rpccnt[NFSPROC_RENAME]++;
1823 nfsm_reqhead(fdvp, NFSPROC_RENAME,
1824 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1825 nfsm_rndup(tnamelen));
1826 nfsm_fhtom(fdvp, v3);
1827 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1828 nfsm_fhtom(tdvp, v3);
1829 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1830 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1831 if (v3) {
1832 nfsm_wcc_data(fdvp, fwccflag);
1833 nfsm_wcc_data(tdvp, twccflag);
1834 }
1835 m_freem(mrep);
1836nfsmout:
1837 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1838 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1839 if (!fwccflag)
1840 VTONFS(fdvp)->n_attrstamp = 0;
1841 if (!twccflag)
1842 VTONFS(tdvp)->n_attrstamp = 0;
1843 return (error);
1844}
1845
1846/*
1847 * nfs hard link create call
1848 *
1849 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1850 * struct componentname *a_cnp)
1851 */
1852static int
1853nfs_link(struct vop_old_link_args *ap)
1854{
1855 struct vnode *vp = ap->a_vp;
1856 struct vnode *tdvp = ap->a_tdvp;
1857 struct componentname *cnp = ap->a_cnp;
1858 u_int32_t *tl;
1859 caddr_t cp;
1860 int32_t t1, t2;
1861 caddr_t bpos, dpos, cp2;
1862 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1863 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1864 int v3;
1865
1866 if (vp->v_mount != tdvp->v_mount) {
1867 return (EXDEV);
1868 }
1869
1870 /*
1871 * The attribute cache may get out of sync with the server on link.
1872 * Pushing writes to the server before handle was inherited from
1873 * long long ago and it is unclear if we still need to do this.
1874 * Defaults to off.
1875 */
1876 if (nfs_flush_on_hlink)
1877 VOP_FSYNC(vp, MNT_WAIT);
1878
1879 v3 = NFS_ISV3(vp);
1880 nfsstats.rpccnt[NFSPROC_LINK]++;
1881 nfsm_reqhead(vp, NFSPROC_LINK,
1882 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1883 nfsm_fhtom(vp, v3);
1884 nfsm_fhtom(tdvp, v3);
1885 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1886 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred);
1887 if (v3) {
1888 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1889 nfsm_wcc_data(tdvp, wccflag);
1890 }
1891 m_freem(mrep);
1892nfsmout:
1893 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1894 if (!attrflag)
1895 VTONFS(vp)->n_attrstamp = 0;
1896 if (!wccflag)
1897 VTONFS(tdvp)->n_attrstamp = 0;
1898 /*
1899 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1900 */
1901 if (error == EEXIST)
1902 error = 0;
1903 return (error);
1904}
1905
1906/*
1907 * nfs symbolic link create call
1908 *
1909 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1910 * struct componentname *a_cnp, struct vattr *a_vap,
1911 * char *a_target)
1912 */
1913static int
1914nfs_symlink(struct vop_old_symlink_args *ap)
1915{
1916 struct vnode *dvp = ap->a_dvp;
1917 struct vattr *vap = ap->a_vap;
1918 struct componentname *cnp = ap->a_cnp;
1919 struct nfsv2_sattr *sp;
1920 u_int32_t *tl;
1921 caddr_t cp;
1922 int32_t t1, t2;
1923 caddr_t bpos, dpos, cp2;
1924 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1925 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1926 struct vnode *newvp = NULL;
1927 int v3 = NFS_ISV3(dvp);
1928
1929 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1930 slen = strlen(ap->a_target);
1931 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1932 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1933 nfsm_fhtom(dvp, v3);
1934 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1935 if (v3) {
1936 nfsm_v3attrbuild(vap, FALSE);
1937 }
1938 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1939 if (!v3) {
1940 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1941 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1942 sp->sa_uid = nfs_xdrneg1;
1943 sp->sa_gid = nfs_xdrneg1;
1944 sp->sa_size = nfs_xdrneg1;
1945 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1946 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1947 }
1948
1949 /*
1950 * Issue the NFS request and get the rpc response.
1951 *
1952 * Only NFSv3 responses returning an error of 0 actually return
1953 * a file handle that can be converted into newvp without having
1954 * to do an extra lookup rpc.
1955 */
1956 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred);
1957 if (v3) {
1958 if (error == 0)
1959 nfsm_mtofh(dvp, newvp, v3, gotvp);
1960 nfsm_wcc_data(dvp, wccflag);
1961 }
1962
1963 /*
1964 * out code jumps -> here, mrep is also freed.
1965 */
1966
1967 m_freem(mrep);
1968nfsmout:
1969
1970 /*
1971 * If we get an EEXIST error, silently convert it to no-error
1972 * in case of an NFS retry.
1973 */
1974 if (error == EEXIST)
1975 error = 0;
1976
1977 /*
1978 * If we do not have (or no longer have) an error, and we could
1979 * not extract the newvp from the response due to the request being
1980 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1981 * to obtain a newvp to return.
1982 */
1983 if (error == 0 && newvp == NULL) {
1984 struct nfsnode *np = NULL;
1985
1986 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1987 cnp->cn_cred, cnp->cn_td, &np);
1988 if (!error)
1989 newvp = NFSTOV(np);
1990 }
1991 if (error) {
1992 if (newvp)
1993 vput(newvp);
1994 } else {
1995 *ap->a_vpp = newvp;
1996 }
1997 VTONFS(dvp)->n_flag |= NLMODIFIED;
1998 if (!wccflag)
1999 VTONFS(dvp)->n_attrstamp = 0;
2000 return (error);
2001}
2002
2003/*
2004 * nfs make dir call
2005 *
2006 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2007 * struct componentname *a_cnp, struct vattr *a_vap)
2008 */
2009static int
2010nfs_mkdir(struct vop_old_mkdir_args *ap)
2011{
2012 struct vnode *dvp = ap->a_dvp;
2013 struct vattr *vap = ap->a_vap;
2014 struct componentname *cnp = ap->a_cnp;
2015 struct nfsv2_sattr *sp;
2016 u_int32_t *tl;
2017 caddr_t cp;
2018 int32_t t1, t2;
2019 int len;
2020 struct nfsnode *np = NULL;
2021 struct vnode *newvp = NULL;
2022 caddr_t bpos, dpos, cp2;
2023 int error = 0, wccflag = NFSV3_WCCRATTR;
2024 int gotvp = 0;
2025 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2026 struct vattr vattr;
2027 int v3 = NFS_ISV3(dvp);
2028
2029 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2030 return (error);
2031 }
2032 len = cnp->cn_namelen;
2033 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2034 nfsm_reqhead(dvp, NFSPROC_MKDIR,
2035 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2036 nfsm_fhtom(dvp, v3);
2037 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2038 if (v3) {
2039 nfsm_v3attrbuild(vap, FALSE);
2040 } else {
2041 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2042 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2043 sp->sa_uid = nfs_xdrneg1;
2044 sp->sa_gid = nfs_xdrneg1;
2045 sp->sa_size = nfs_xdrneg1;
2046 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2047 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2048 }
2049 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred);
2050 if (!error)
2051 nfsm_mtofh(dvp, newvp, v3, gotvp);
2052 if (v3)
2053 nfsm_wcc_data(dvp, wccflag);
2054 m_freem(mrep);
2055nfsmout:
2056 VTONFS(dvp)->n_flag |= NLMODIFIED;
2057 if (!wccflag)
2058 VTONFS(dvp)->n_attrstamp = 0;
2059 /*
2060 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2061 * if we can succeed in looking up the directory.
2062 */
2063 if (error == EEXIST || (!error && !gotvp)) {
2064 if (newvp) {
2065 vrele(newvp);
2066 newvp = NULL;
2067 }
2068 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2069 cnp->cn_td, &np);
2070 if (!error) {
2071 newvp = NFSTOV(np);
2072 if (newvp->v_type != VDIR)
2073 error = EEXIST;
2074 }
2075 }
2076 if (error) {
2077 if (newvp)
2078 vrele(newvp);
2079 } else
2080 *ap->a_vpp = newvp;
2081 return (error);
2082}
2083
2084/*
2085 * nfs remove directory call
2086 *
2087 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2088 * struct componentname *a_cnp)
2089 */
2090static int
2091nfs_rmdir(struct vop_old_rmdir_args *ap)
2092{
2093 struct vnode *vp = ap->a_vp;
2094 struct vnode *dvp = ap->a_dvp;
2095 struct componentname *cnp = ap->a_cnp;
2096 u_int32_t *tl;
2097 caddr_t cp;
2098 int32_t t1, t2;
2099 caddr_t bpos, dpos, cp2;
2100 int error = 0, wccflag = NFSV3_WCCRATTR;
2101 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2102 int v3 = NFS_ISV3(dvp);
2103
2104 if (dvp == vp)
2105 return (EINVAL);
2106 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2107 nfsm_reqhead(dvp, NFSPROC_RMDIR,
2108 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2109 nfsm_fhtom(dvp, v3);
2110 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2111 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred);
2112 if (v3)
2113 nfsm_wcc_data(dvp, wccflag);
2114 m_freem(mrep);
2115nfsmout:
2116 VTONFS(dvp)->n_flag |= NLMODIFIED;
2117 if (!wccflag)
2118 VTONFS(dvp)->n_attrstamp = 0;
2119 /*
2120 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2121 */
2122 if (error == ENOENT)
2123 error = 0;
2124 return (error);
2125}
2126
2127/*
2128 * nfs readdir call
2129 *
2130 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2131 */
2132static int
2133nfs_readdir(struct vop_readdir_args *ap)
2134{
2135 struct vnode *vp = ap->a_vp;
2136 struct nfsnode *np = VTONFS(vp);
2137 struct uio *uio = ap->a_uio;
2138 int tresid, error;
2139 struct vattr vattr;
2140
2141 if (vp->v_type != VDIR)
2142 return (EPERM);
2143
2144 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
2145 return (error);
2146
2147 /*
2148 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2149 * and then check that is still valid, or if this is an NQNFS mount
2150 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2151 * VOP_GETATTR() does not necessarily go to the wire.
2152 */
2153 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2154 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2155 if (VOP_GETATTR(vp, &vattr) == 0 &&
2156 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2157 ) {
2158 nfsstats.direofcache_hits++;
2159 goto done;
2160 }
2161 }
2162
2163 /*
2164 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2165 * own cache coherency checks so we do not have to.
2166 */
2167 tresid = uio->uio_resid;
2168 error = nfs_bioread(vp, uio, 0);
2169
2170 if (!error && uio->uio_resid == tresid)
2171 nfsstats.direofcache_misses++;
2172done:
2173 vn_unlock(vp);
2174 return (error);
2175}
2176
2177/*
2178 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2179 *
2180 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2181 * offset/block and converts the nfs formatted directory entries for userland
2182 * consumption as well as deals with offsets into the middle of blocks.
2183 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2184 * be block-bounded. It must convert to cookies for the actual RPC.
2185 */
2186int
2187nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2188{
2189 int len, left;
2190 struct nfs_dirent *dp = NULL;
2191 u_int32_t *tl;
2192 caddr_t cp;
2193 int32_t t1, t2;
2194 nfsuint64 *cookiep;
2195 caddr_t bpos, dpos, cp2;
2196 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2197 nfsuint64 cookie;
2198 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2199 struct nfsnode *dnp = VTONFS(vp);
2200 u_quad_t fileno;
2201 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2202 int attrflag;
2203 int v3 = NFS_ISV3(vp);
2204
2205#ifndef DIAGNOSTIC
2206 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2207 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2208 panic("nfs readdirrpc bad uio");
2209#endif
2210
2211 /*
2212 * If there is no cookie, assume directory was stale.
2213 */
2214 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2215 if (cookiep)
2216 cookie = *cookiep;
2217 else
2218 return (NFSERR_BAD_COOKIE);
2219 /*
2220 * Loop around doing readdir rpc's of size nm_readdirsize
2221 * truncated to a multiple of DIRBLKSIZ.
2222 * The stopping criteria is EOF or buffer full.
2223 */
2224 while (more_dirs && bigenough) {
2225 nfsstats.rpccnt[NFSPROC_READDIR]++;
2226 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2227 NFSX_READDIR(v3));
2228 nfsm_fhtom(vp, v3);
2229 if (v3) {
2230 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2231 *tl++ = cookie.nfsuquad[0];
2232 *tl++ = cookie.nfsuquad[1];
2233 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2234 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2235 } else {
2236 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2237 *tl++ = cookie.nfsuquad[0];
2238 }
2239 *tl = txdr_unsigned(nmp->nm_readdirsize);
2240 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2241 if (v3) {
2242 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2243 if (!error) {
2244 nfsm_dissect(tl, u_int32_t *,
2245 2 * NFSX_UNSIGNED);
2246 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2247 dnp->n_cookieverf.nfsuquad[1] = *tl;
2248 } else {
2249 m_freem(mrep);
2250 goto nfsmout;
2251 }
2252 }
2253 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2254 more_dirs = fxdr_unsigned(int, *tl);
2255
2256 /* loop thru the dir entries, converting them to std form */
2257 while (more_dirs && bigenough) {
2258 if (v3) {
2259 nfsm_dissect(tl, u_int32_t *,
2260 3 * NFSX_UNSIGNED);
2261 fileno = fxdr_hyper(tl);
2262 len = fxdr_unsigned(int, *(tl + 2));
2263 } else {
2264 nfsm_dissect(tl, u_int32_t *,
2265 2 * NFSX_UNSIGNED);
2266 fileno = fxdr_unsigned(u_quad_t, *tl++);
2267 len = fxdr_unsigned(int, *tl);
2268 }
2269 if (len <= 0 || len > NFS_MAXNAMLEN) {
2270 error = EBADRPC;
2271 m_freem(mrep);
2272 goto nfsmout;
2273 }
2274
2275 /*
2276 * len is the number of bytes in the path element
2277 * name, not including the \0 termination.
2278 *
2279 * tlen is the number of bytes w have to reserve for
2280 * the path element name.
2281 */
2282 tlen = nfsm_rndup(len);
2283 if (tlen == len)
2284 tlen += 4; /* To ensure null termination */
2285
2286 /*
2287 * If the entry would cross a DIRBLKSIZ boundary,
2288 * extend the previous nfs_dirent to cover the
2289 * remaining space.
2290 */
2291 left = DIRBLKSIZ - blksiz;
2292 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2293 dp->nfs_reclen += left;
2294 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2295 uiop->uio_iov->iov_len -= left;
2296 uiop->uio_offset += left;
2297 uiop->uio_resid -= left;
2298 blksiz = 0;
2299 }
2300 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2301 bigenough = 0;
2302 if (bigenough) {
2303 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2304 dp->nfs_ino = fileno;
2305 dp->nfs_namlen = len;
2306 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2307 dp->nfs_type = DT_UNKNOWN;
2308 blksiz += dp->nfs_reclen;
2309 if (blksiz == DIRBLKSIZ)
2310 blksiz = 0;
2311 uiop->uio_offset += sizeof(struct nfs_dirent);
2312 uiop->uio_resid -= sizeof(struct nfs_dirent);
2313 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2314 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2315 nfsm_mtouio(uiop, len);
2316
2317 /*
2318 * The uiop has advanced by nfs_dirent + len
2319 * but really needs to advance by
2320 * nfs_dirent + tlen
2321 */
2322 cp = uiop->uio_iov->iov_base;
2323 tlen -= len;
2324 *cp = '\0'; /* null terminate */
2325 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2326 uiop->uio_iov->iov_len -= tlen;
2327 uiop->uio_offset += tlen;
2328 uiop->uio_resid -= tlen;
2329 } else {
2330 /*
2331 * NFS strings must be rounded up (nfsm_myouio
2332 * handled that in the bigenough case).
2333 */
2334 nfsm_adv(nfsm_rndup(len));
2335 }
2336 if (v3) {
2337 nfsm_dissect(tl, u_int32_t *,
2338 3 * NFSX_UNSIGNED);
2339 } else {
2340 nfsm_dissect(tl, u_int32_t *,
2341 2 * NFSX_UNSIGNED);
2342 }
2343
2344 /*
2345 * If we were able to accomodate the last entry,
2346 * get the cookie for the next one. Otherwise
2347 * hold-over the cookie for the one we were not
2348 * able to accomodate.
2349 */
2350 if (bigenough) {
2351 cookie.nfsuquad[0] = *tl++;
2352 if (v3)
2353 cookie.nfsuquad[1] = *tl++;
2354 } else if (v3) {
2355 tl += 2;
2356 } else {
2357 tl++;
2358 }
2359 more_dirs = fxdr_unsigned(int, *tl);
2360 }
2361 /*
2362 * If at end of rpc data, get the eof boolean
2363 */
2364 if (!more_dirs) {
2365 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2366 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2367 }
2368 m_freem(mrep);
2369 }
2370 /*
2371 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2372 * by increasing d_reclen for the last record.
2373 */
2374 if (blksiz > 0) {
2375 left = DIRBLKSIZ - blksiz;
2376 dp->nfs_reclen += left;
2377 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2378 uiop->uio_iov->iov_len -= left;
2379 uiop->uio_offset += left;
2380 uiop->uio_resid -= left;
2381 }
2382
2383 if (bigenough) {
2384 /*
2385 * We hit the end of the directory, update direofoffset.
2386 */
2387 dnp->n_direofoffset = uiop->uio_offset;
2388 } else {
2389 /*
2390 * There is more to go, insert the link cookie so the
2391 * next block can be read.
2392 */
2393 if (uiop->uio_resid > 0)
2394 kprintf("EEK! readdirrpc resid > 0\n");
2395 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2396 *cookiep = cookie;
2397 }
2398nfsmout:
2399 return (error);
2400}
2401
2402/*
2403 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2404 */
2405int
2406nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2407{
2408 int len, left;
2409 struct nfs_dirent *dp;
2410 u_int32_t *tl;
2411 caddr_t cp;
2412 int32_t t1, t2;
2413 struct vnode *newvp;
2414 nfsuint64 *cookiep;
2415 caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2416 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2417 nfsuint64 cookie;
2418 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2419 struct nfsnode *dnp = VTONFS(vp), *np;
2420 nfsfh_t *fhp;
2421 u_quad_t fileno;
2422 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2423 int attrflag, fhsize;
2424 struct nchandle nch;
2425 struct nchandle dnch;
2426 struct nlcomponent nlc;
2427
2428#ifndef nolint
2429 dp = NULL;
2430#endif
2431#ifndef DIAGNOSTIC
2432 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2433 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2434 panic("nfs readdirplusrpc bad uio");
2435#endif
2436 /*
2437 * Obtain the namecache record for the directory so we have something
2438 * to use as a basis for creating the entries. This function will
2439 * return a held (but not locked) ncp. The ncp may be disconnected
2440 * from the tree and cannot be used for upward traversals, and the
2441 * ncp may be unnamed. Note that other unrelated operations may
2442 * cause the ncp to be named at any time.
2443 */
2444 cache_fromdvp(vp, NULL, 0, &dnch);
2445 bzero(&nlc, sizeof(nlc));
2446 newvp = NULLVP;
2447
2448 /*
2449 * If there is no cookie, assume directory was stale.
2450 */
2451 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2452 if (cookiep)
2453 cookie = *cookiep;
2454 else
2455 return (NFSERR_BAD_COOKIE);
2456 /*
2457 * Loop around doing readdir rpc's of size nm_readdirsize
2458 * truncated to a multiple of DIRBLKSIZ.
2459 * The stopping criteria is EOF or buffer full.
2460 */
2461 while (more_dirs && bigenough) {
2462 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2463 nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2464 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2465 nfsm_fhtom(vp, 1);
2466 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2467 *tl++ = cookie.nfsuquad[0];
2468 *tl++ = cookie.nfsuquad[1];
2469 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2470 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2471 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2472 *tl = txdr_unsigned(nmp->nm_rsize);
2473 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2474 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2475 if (error) {
2476 m_freem(mrep);
2477 goto nfsmout;
2478 }
2479 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2480 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2481 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2482 more_dirs = fxdr_unsigned(int, *tl);
2483
2484 /* loop thru the dir entries, doctoring them to 4bsd form */
2485 while (more_dirs && bigenough) {
2486 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2487 fileno = fxdr_hyper(tl);
2488 len = fxdr_unsigned(int, *(tl + 2));
2489 if (len <= 0 || len > NFS_MAXNAMLEN) {
2490 error = EBADRPC;
2491 m_freem(mrep);
2492 goto nfsmout;
2493 }
2494 tlen = nfsm_rndup(len);
2495 if (tlen == len)
2496 tlen += 4; /* To ensure null termination*/
2497 left = DIRBLKSIZ - blksiz;
2498 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2499 dp->nfs_reclen += left;
2500 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2501 uiop->uio_iov->iov_len -= left;
2502 uiop->uio_offset += left;
2503 uiop->uio_resid -= left;
2504 blksiz = 0;
2505 }
2506 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2507 bigenough = 0;
2508 if (bigenough) {
2509 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2510 dp->nfs_ino = fileno;
2511 dp->nfs_namlen = len;
2512 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2513 dp->nfs_type = DT_UNKNOWN;
2514 blksiz += dp->nfs_reclen;
2515 if (blksiz == DIRBLKSIZ)
2516 blksiz = 0;
2517 uiop->uio_offset += sizeof(struct nfs_dirent);
2518 uiop->uio_resid -= sizeof(struct nfs_dirent);
2519 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2520 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2521 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2522 nlc.nlc_namelen = len;
2523 nfsm_mtouio(uiop, len);
2524 cp = uiop->uio_iov->iov_base;
2525 tlen -= len;
2526 *cp = '\0';
2527 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2528 uiop->uio_iov->iov_len -= tlen;
2529 uiop->uio_offset += tlen;
2530 uiop->uio_resid -= tlen;
2531 } else
2532 nfsm_adv(nfsm_rndup(len));
2533 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2534 if (bigenough) {
2535 cookie.nfsuquad[0] = *tl++;
2536 cookie.nfsuquad[1] = *tl++;
2537 } else
2538 tl += 2;
2539
2540 /*
2541 * Since the attributes are before the file handle
2542 * (sigh), we must skip over the attributes and then
2543 * come back and get them.
2544 */
2545 attrflag = fxdr_unsigned(int, *tl);
2546 if (attrflag) {
2547 dpossav1 = dpos;
2548 mdsav1 = md;
2549 nfsm_adv(NFSX_V3FATTR);
2550 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2551 doit = fxdr_unsigned(int, *tl);
2552 if (doit) {
2553 nfsm_getfh(fhp, fhsize, 1);
2554 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2555 vref(vp);
2556 newvp = vp;
2557 np = dnp;
2558 } else {
2559 error = nfs_nget(vp->v_mount, fhp,
2560 fhsize, &np);
2561 if (error)
2562 doit = 0;
2563 else
2564 newvp = NFSTOV(np);
2565 }
2566 }
2567 if (doit && bigenough) {
2568 dpossav2 = dpos;
2569 dpos = dpossav1;
2570 mdsav2 = md;
2571 md = mdsav1;
2572 nfsm_loadattr(newvp, NULL);
2573 dpos = dpossav2;
2574 md = mdsav2;
2575 dp->nfs_type =
2576 IFTODT(VTTOIF(np->n_vattr.va_type));
2577 if (dnch.ncp) {
2578 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2579 nlc.nlc_namelen, nlc.nlc_namelen,
2580 nlc.nlc_nameptr);
2581 nch = cache_nlookup(&dnch, &nlc);
2582 cache_setunresolved(&nch);
2583 nfs_cache_setvp(&nch, newvp,
2584 nfspos_cache_timeout);
2585 cache_put(&nch);
2586 } else {
2587 kprintf("NFS/READDIRPLUS, UNABLE TO ENTER"
2588 " %*.*s\n",
2589 nlc.nlc_namelen, nlc.nlc_namelen,
2590 nlc.nlc_nameptr);
2591 }
2592 }
2593 } else {
2594 /* Just skip over the file handle */
2595 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2596 i = fxdr_unsigned(int, *tl);
2597 nfsm_adv(nfsm_rndup(i));
2598 }
2599 if (newvp != NULLVP) {
2600 if (newvp == vp)
2601 vrele(newvp);
2602 else
2603 vput(newvp);
2604 newvp = NULLVP;
2605 }
2606 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2607 more_dirs = fxdr_unsigned(int, *tl);
2608 }
2609 /*
2610 * If at end of rpc data, get the eof boolean
2611 */
2612 if (!more_dirs) {
2613 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2614 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2615 }
2616 m_freem(mrep);
2617 }
2618 /*
2619 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2620 * by increasing d_reclen for the last record.
2621 */
2622 if (blksiz > 0) {
2623 left = DIRBLKSIZ - blksiz;
2624 dp->nfs_reclen += left;
2625 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2626 uiop->uio_iov->iov_len -= left;
2627 uiop->uio_offset += left;
2628 uiop->uio_resid -= left;
2629 }
2630
2631 /*
2632 * We are now either at the end of the directory or have filled the
2633 * block.
2634 */
2635 if (bigenough)
2636 dnp->n_direofoffset = uiop->uio_offset;
2637 else {
2638 if (uiop->uio_resid > 0)
2639 kprintf("EEK! readdirplusrpc resid > 0\n");
2640 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2641 *cookiep = cookie;
2642 }
2643nfsmout:
2644 if (newvp != NULLVP) {
2645 if (newvp == vp)
2646 vrele(newvp);
2647 else
2648 vput(newvp);
2649 newvp = NULLVP;
2650 }
2651 if (dnch.ncp)
2652 cache_drop(&dnch);
2653 return (error);
2654}
2655
2656/*
2657 * Silly rename. To make the NFS filesystem that is stateless look a little
2658 * more like the "ufs" a remove of an active vnode is translated to a rename
2659 * to a funny looking filename that is removed by nfs_inactive on the
2660 * nfsnode. There is the potential for another process on a different client
2661 * to create the same funny name between the nfs_lookitup() fails and the
2662 * nfs_rename() completes, but...
2663 */
2664static int
2665nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2666{
2667 struct sillyrename *sp;
2668 struct nfsnode *np;
2669 int error;
2670
2671 /*
2672 * We previously purged dvp instead of vp. I don't know why, it
2673 * completely destroys performance. We can't do it anyway with the
2674 * new VFS API since we would be breaking the namecache topology.
2675 */
2676 cache_purge(vp); /* XXX */
2677 np = VTONFS(vp);
2678#ifndef DIAGNOSTIC
2679 if (vp->v_type == VDIR)
2680 panic("nfs: sillyrename dir");
2681#endif
2682 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2683 M_NFSREQ, M_WAITOK);
2684 sp->s_cred = crdup(cnp->cn_cred);
2685 sp->s_dvp = dvp;
2686 vref(dvp);
2687
2688 /* Fudge together a funny name */
2689 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
2690 (int)(intptr_t)cnp->cn_td);
2691
2692 /* Try lookitups until we get one that isn't there */
2693 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2694 cnp->cn_td, NULL) == 0) {
2695 sp->s_name[4]++;
2696 if (sp->s_name[4] > 'z') {
2697 error = EINVAL;
2698 goto bad;
2699 }
2700 }
2701 error = nfs_renameit(dvp, cnp, sp);
2702 if (error)
2703 goto bad;
2704 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2705 cnp->cn_td, &np);
2706 np->n_sillyrename = sp;
2707 return (0);
2708bad:
2709 vrele(sp->s_dvp);
2710 crfree(sp->s_cred);
2711 kfree((caddr_t)sp, M_NFSREQ);
2712 return (error);
2713}
2714
2715/*
2716 * Look up a file name and optionally either update the file handle or
2717 * allocate an nfsnode, depending on the value of npp.
2718 * npp == NULL --> just do the lookup
2719 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2720 * handled too
2721 * *npp != NULL --> update the file handle in the vnode
2722 */
2723static int
2724nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2725 struct thread *td, struct nfsnode **npp)
2726{
2727 u_int32_t *tl;
2728 caddr_t cp;
2729 int32_t t1, t2;
2730 struct vnode *newvp = NULL;
2731 struct nfsnode *np, *dnp = VTONFS(dvp);
2732 caddr_t bpos, dpos, cp2;
2733 int error = 0, fhlen, attrflag;
2734 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2735 nfsfh_t *nfhp;
2736 int v3 = NFS_ISV3(dvp);
2737
2738 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2739 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2740 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2741 nfsm_fhtom(dvp, v3);
2742 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2743 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2744 if (npp && !error) {
2745 nfsm_getfh(nfhp, fhlen, v3);
2746 if (*npp) {
2747 np = *npp;
2748 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2749 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
2750 np->n_fhp = &np->n_fh;
2751 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2752 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
2753 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2754 np->n_fhsize = fhlen;
2755 newvp = NFSTOV(np);
2756 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2757 vref(dvp);
2758 newvp = dvp;
2759 } else {
2760 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2761 if (error) {
2762 m_freem(mrep);
2763 return (error);
2764 }
2765 newvp = NFSTOV(np);
2766 }
2767 if (v3) {
2768 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
2769 if (!attrflag && *npp == NULL) {
2770 m_freem(mrep);
2771 if (newvp == dvp)
2772 vrele(newvp);
2773 else
2774 vput(newvp);
2775 return (ENOENT);
2776 }
2777 } else
2778 nfsm_loadattr(newvp, NULL);
2779 }
2780 m_freem(mrep);
2781nfsmout:
2782 if (npp && *npp == NULL) {
2783 if (error) {
2784 if (newvp) {
2785 if (newvp == dvp)
2786 vrele(newvp);
2787 else
2788 vput(newvp);
2789 }
2790 } else
2791 *npp = np;
2792 }
2793 return (error);
2794}
2795
2796/*
2797 * Nfs Version 3 commit rpc
2798 */
2799int
2800nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2801{
2802 caddr_t cp;
2803 u_int32_t *tl;
2804 int32_t t1, t2;
2805 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2806 caddr_t bpos, dpos, cp2;
2807 int error = 0, wccflag = NFSV3_WCCRATTR;
2808 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2809
2810 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2811 return (0);
2812 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2813 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2814 nfsm_fhtom(vp, 1);
2815 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2816 txdr_hyper(offset, tl);
2817 tl += 2;
2818 *tl = txdr_unsigned(cnt);
2819 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE));
2820 nfsm_wcc_data(vp, wccflag);
2821 if (!error) {
2822 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2823 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2824 NFSX_V3WRITEVERF)) {
2825 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2826 NFSX_V3WRITEVERF);
2827 error = NFSERR_STALEWRITEVERF;
2828 }
2829 }
2830 m_freem(mrep);
2831nfsmout:
2832 return (error);
2833}
2834
2835/*
2836 * Kludge City..
2837 * - make nfs_bmap() essentially a no-op that does no translation
2838 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2839 * (Maybe I could use the process's page mapping, but I was concerned that
2840 * Kernel Write might not be enabled and also figured copyout() would do
2841 * a lot more work than bcopy() and also it currently happens in the
2842 * context of the swapper process (2).
2843 *
2844 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
2845 * off_t *a_doffsetp, int *a_runp, int *a_runb)
2846 */
2847static int
2848nfs_bmap(struct vop_bmap_args *ap)
2849{
2850 if (ap->a_doffsetp != NULL)
2851 *ap->a_doffsetp = ap->a_loffset;
2852 if (ap->a_runp != NULL)
2853 *ap->a_runp = 0;
2854 if (ap->a_runb != NULL)
2855 *ap->a_runb = 0;
2856 return (0);
2857}
2858
2859/*
2860 * Strategy routine.
2861 *
2862 * For async requests when nfsiod(s) are running, queue the request by
2863 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2864 * request.
2865 */
2866static int
2867nfs_strategy(struct vop_strategy_args *ap)
2868{
2869 struct bio *bio = ap->a_bio;
2870 struct bio *nbio;
2871 struct buf *bp = bio->bio_buf;
2872 struct thread *td;
2873 int error = 0;
2874
2875 KASSERT(bp->b_cmd != BUF_CMD_DONE,
2876 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
2877 KASSERT(BUF_REFCNT(bp) > 0,
2878 ("nfs_strategy: buffer %p not locked", bp));
2879
2880 if (bp->b_flags & B_ASYNC)
2881 td = NULL;
2882 else
2883 td = curthread; /* XXX */
2884
2885 /*
2886 * We probably don't need to push an nbio any more since no
2887 * block conversion is required due to the use of 64 bit byte
2888 * offsets, but do it anyway.
2889 */
2890 nbio = push_bio(bio);
2891 nbio->bio_offset = bio->bio_offset;
2892
2893 /*
2894 * If the op is asynchronous and an i/o daemon is waiting
2895 * queue the request, wake it up and wait for completion
2896 * otherwise just do it ourselves.
2897 */
2898 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(ap->a_vp, nbio, td))
2899 error = nfs_doio(ap->a_vp, nbio, td);
2900 return (error);
2901}
2902
2903/*
2904 * Mmap a file
2905 *
2906 * NB Currently unsupported.
2907 *
2908 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred)
2909 */
2910/* ARGSUSED */
2911static int
2912nfs_mmap(struct vop_mmap_args *ap)
2913{
2914 return (EINVAL);
2915}
2916
2917/*
2918 * fsync vnode op. Just call nfs_flush() with commit == 1.
2919 *
2920 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
2921 */
2922/* ARGSUSED */
2923static int
2924nfs_fsync(struct vop_fsync_args *ap)
2925{
2926 return (nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1));
2927}
2928
2929/*
2930 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2931 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2932 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2933 * set the buffer contains data that has already been written to the server
2934 * and which now needs a commit RPC.
2935 *
2936 * If commit is 0 we only take one pass and only flush buffers containing new
2937 * dirty data.
2938 *
2939 * If commit is 1 we take two passes, issuing a commit RPC in the second
2940 * pass.
2941 *
2942 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
2943 * to completely flush all pending data.
2944 *
2945 * Note that the RB_SCAN code properly handles the case where the
2946 * callback might block and directly or indirectly (another thread) cause
2947 * the RB tree to change.
2948 */
2949
2950#ifndef NFS_COMMITBVECSIZ
2951#define NFS_COMMITBVECSIZ 16
2952#endif
2953
2954struct nfs_flush_info {
2955 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
2956 struct thread *td;
2957 struct vnode *vp;
2958 int waitfor;
2959 int slpflag;
2960 int slptimeo;
2961 int loops;
2962 struct buf *bvary[NFS_COMMITBVECSIZ];
2963 int bvsize;
2964 off_t beg_off;
2965 off_t end_off;
2966};
2967
2968static int nfs_flush_bp(struct buf *bp, void *data);
2969static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
2970
2971int
2972nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
2973{
2974 struct nfsnode *np = VTONFS(vp);
2975 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2976 struct nfs_flush_info info;
2977 int error;
2978
2979 bzero(&info, sizeof(info));
2980 info.td = td;
2981 info.vp = vp;
2982 info.waitfor = waitfor;
2983 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
2984 info.loops = 0;
2985
2986 do {
2987 /*
2988 * Flush mode
2989 */
2990 info.mode = NFI_FLUSHNEW;
2991 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2992 nfs_flush_bp, &info);
2993
2994 /*
2995 * Take a second pass if committing and no error occured.
2996 * Clean up any left over collection (whether an error
2997 * occurs or not).
2998 */
2999 if (commit && error == 0) {
3000 info.mode = NFI_COMMIT;
3001 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3002 nfs_flush_bp, &info);
3003 if (info.bvsize)
3004 error = nfs_flush_docommit(&info, error);
3005 }
3006
3007 /*
3008 * Wait for pending I/O to complete before checking whether
3009 * any further dirty buffers exist.
3010 */
3011 while (waitfor == MNT_WAIT && vp->v_track_write.bk_active) {
3012 vp->v_track_write.bk_waitflag = 1;
3013 error = tsleep(&vp->v_track_write,
3014 info.slpflag, "nfsfsync", info.slptimeo);
3015 if (error) {
3016 /*
3017 * We have to be able to break out if this
3018 * is an 'intr' mount.
3019 */
3020 if (nfs_sigintr(nmp, NULL, td)) {
3021 error = -EINTR;
3022 break;
3023 }
3024
3025 /*
3026 * Since we do not process pending signals,
3027 * once we get a PCATCH our tsleep() will no
3028 * longer sleep, switch to a fixed timeout
3029 * instead.
3030 */
3031 if (info.slpflag == PCATCH) {
3032 info.slpflag = 0;
3033 info.slptimeo = 2 * hz;
3034 }
3035 error = 0;
3036 }
3037 }
3038 ++info.loops;
3039 /*
3040 * Loop if we are flushing synchronous as well as committing,
3041 * and dirty buffers are still present. Otherwise we might livelock.
3042 */
3043 } while (waitfor == MNT_WAIT && commit &&
3044 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3045
3046 /*
3047 * The callbacks have to return a negative error to terminate the
3048 * RB scan.
3049 */
3050 if (error < 0)
3051 error = -error;
3052
3053 /*
3054 * Deal with any error collection
3055 */
3056 if (np->n_flag & NWRITEERR) {
3057 error = np->n_error;
3058 np->n_flag &= ~NWRITEERR;
3059 }
3060 return (error);
3061}
3062
3063
3064static
3065int
3066nfs_flush_bp(struct buf *bp, void *data)
3067{
3068 struct nfs_flush_info *info = data;
3069 off_t toff;
3070 int error;
3071
3072 error = 0;
3073 switch(info->mode) {
3074 case NFI_FLUSHNEW:
3075 crit_enter();
3076 if (info->loops && info->waitfor == MNT_WAIT) {
3077 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3078 if (error) {
3079 int lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3080 if (info->slpflag & PCATCH)
3081 lkflags |= LK_PCATCH;
3082 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3083 info->slptimeo);
3084 }
3085 } else {
3086 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3087 }
3088 if (error == 0) {
3089 KKASSERT(bp->b_vp == info->vp);
3090
3091 if ((bp->b_flags & B_DELWRI) == 0)
3092 panic("nfs_fsync: not dirty");
3093 if (bp->b_flags & B_NEEDCOMMIT) {
3094 BUF_UNLOCK(bp);
3095 crit_exit();
3096 break;
3097 }
3098 bremfree(bp);
3099
3100 crit_exit();
3101 bawrite(bp);
3102 } else {
3103 crit_exit();
3104 error = 0;
3105 }
3106 break;
3107 case NFI_COMMIT:
3108 /*
3109 * Only process buffers in need of a commit which we can
3110 * immediately lock. This may prevent a buffer from being
3111 * committed, but the normal flush loop will block on the
3112 * same buffer so we shouldn't get into an endless loop.
3113 */
3114 crit_enter();
3115 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3116 (B_DELWRI | B_NEEDCOMMIT) ||
3117 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
3118 crit_exit();
3119 break;
3120 }
3121
3122 KKASSERT(bp->b_vp == info->vp);
3123 bremfree(bp);
3124
3125 /*
3126 * NOTE: storing the bp in the bvary[] basically sets
3127 * it up for a commit operation.
3128 *
3129 * We must call vfs_busy_pages() now so the commit operation
3130 * is interlocked with user modifications to memory mapped
3131 * pages.
3132 *
3133 * Note: to avoid loopback deadlocks, we do not
3134 * assign b_runningbufspace.
3135 */
3136 bp->b_cmd = BUF_CMD_WRITE;
3137 vfs_busy_pages(bp->b_vp, bp);
3138 info->bvary[info->bvsize] = bp;
3139 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3140 if (info->bvsize == 0 || toff < info->beg_off)
3141 info->beg_off = toff;
3142 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3143 if (info->bvsize == 0 || toff > info->end_off)
3144 info->end_off = toff;
3145 ++info->bvsize;
3146 if (info->bvsize == NFS_COMMITBVECSIZ) {
3147 error = nfs_flush_docommit(info, 0);
3148 KKASSERT(info->bvsize == 0);
3149 }
3150 crit_exit();
3151 }
3152 return (error);
3153}
3154
3155static
3156int
3157nfs_flush_docommit(struct nfs_flush_info *info, int error)
3158{
3159 struct vnode *vp;
3160 struct buf *bp;
3161 off_t bytes;
3162 int retv;
3163 int i;
3164
3165 vp = info->vp;
3166
3167 if (info->bvsize > 0) {
3168 /*
3169 * Commit data on the server, as required. Note that
3170 * nfs_commit will use the vnode's cred for the commit.
3171 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3172 */
3173 bytes = info->end_off - info->beg_off;
3174 if (bytes > 0x40000000)
3175 bytes = 0x40000000;
3176 if (error) {
3177 retv = -error;
3178 } else {
3179 retv = nfs_commit(vp, info->beg_off,
3180 (int)bytes, info->td);
3181 if (retv == NFSERR_STALEWRITEVERF)
3182 nfs_clearcommit(vp->v_mount);
3183 }
3184
3185 /*
3186 * Now, either mark the blocks I/O done or mark the
3187 * blocks dirty, depending on whether the commit
3188 * succeeded.
3189 */
3190 for (i = 0; i < info->bvsize; ++i) {
3191 bp = info->bvary[i];
3192 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3193 if (retv) {
3194 /*
3195 * Error, leave B_DELWRI intact
3196 */
3197 vfs_unbusy_pages(bp);
3198 bp->b_cmd = BUF_CMD_DONE;
3199 brelse(bp);
3200 } else {
3201 /*
3202 * Success, remove B_DELWRI ( bundirty() ).
3203 *
3204 * b_dirtyoff/b_dirtyend seem to be NFS
3205 * specific. We should probably move that
3206 * into bundirty(). XXX
3207 *
3208 * We are faking an I/O write, we have to
3209 * start the transaction in order to
3210 * immediately biodone() it.
3211 */
3212 crit_enter();
3213 bp->b_flags |= B_ASYNC;
3214 bundirty(bp);
3215 bp->b_flags &= ~B_ERROR;
3216 bp->b_dirtyoff = bp->b_dirtyend = 0;
3217 crit_exit();
3218 biodone(&bp->b_bio1);
3219 }
3220 }
3221 info->bvsize = 0;
3222 }
3223 return (error);
3224}
3225
3226/*
3227 * NFS advisory byte-level locks.
3228 * Currently unsupported.
3229 *
3230 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3231 * int a_flags)
3232 */
3233static int
3234nfs_advlock(struct vop_advlock_args *ap)
3235{
3236 struct nfsnode *np = VTONFS(ap->a_vp);
3237
3238 /*
3239 * The following kludge is to allow diskless support to work
3240 * until a real NFS lockd is implemented. Basically, just pretend
3241 * that this is a local lock.
3242 */
3243 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3244}
3245
3246/*
3247 * Print out the contents of an nfsnode.
3248 *
3249 * nfs_print(struct vnode *a_vp)
3250 */
3251static int
3252nfs_print(struct vop_print_args *ap)
3253{
3254 struct vnode *vp = ap->a_vp;
3255 struct nfsnode *np = VTONFS(vp);
3256
3257 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3258 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3259 if (vp->v_type == VFIFO)
3260 fifo_printinfo(vp);
3261 kprintf("\n");
3262 return (0);
3263}
3264
3265/*
3266 * nfs special file access vnode op.
3267 * Essentially just get vattr and then imitate iaccess() since the device is
3268 * local to the client.
3269 *
3270 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3271 */
3272static int
3273nfsspec_access(struct vop_access_args *ap)
3274{
3275 struct vattr *vap;
3276 gid_t *gp;
3277 struct ucred *cred = ap->a_cred;
3278 struct vnode *vp = ap->a_vp;
3279 mode_t mode = ap->a_mode;
3280 struct vattr vattr;
3281 int i;
3282 int error;
3283
3284 /*
3285 * Disallow write attempts on filesystems mounted read-only;
3286 * unless the file is a socket, fifo, or a block or character
3287 * device resident on the filesystem.
3288 */
3289 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3290 switch (vp->v_type) {
3291 case VREG:
3292 case VDIR:
3293 case VLNK:
3294 return (EROFS);
3295 default:
3296 break;
3297 }
3298 }
3299 /*
3300 * If you're the super-user,
3301 * you always get access.
3302 */
3303 if (cred->cr_uid == 0)
3304 return (0);
3305 vap = &vattr;
3306 error = VOP_GETATTR(vp, vap);
3307 if (error)
3308 return (error);
3309 /*
3310 * Access check is based on only one of owner, group, public.
3311 * If not owner, then check group. If not a member of the
3312 * group, then check public access.
3313 */
3314 if (cred->cr_uid != vap->va_uid) {
3315 mode >>= 3;
3316 gp = cred->cr_groups;
3317 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3318 if (vap->va_gid == *gp)
3319 goto found;
3320 mode >>= 3;
3321found:
3322 ;
3323 }
3324 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3325 return (error);
3326}
3327
3328/*
3329 * Read wrapper for special devices.
3330 *
3331 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3332 * struct ucred *a_cred)
3333 */
3334static int
3335nfsspec_read(struct vop_read_args *ap)
3336{
3337 struct nfsnode *np = VTONFS(ap->a_vp);
3338
3339 /*
3340 * Set access flag.
3341 */
3342 np->n_flag |= NACC;
3343 getnanotime(&np->n_atim);
3344 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3345}
3346
3347/*
3348 * Write wrapper for special devices.
3349 *
3350 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3351 * struct ucred *a_cred)
3352 */
3353static int
3354nfsspec_write(struct vop_write_args *ap)
3355{
3356 struct nfsnode *np = VTONFS(ap->a_vp);
3357
3358 /*
3359 * Set update flag.
3360 */
3361 np->n_flag |= NUPD;
3362 getnanotime(&np->n_mtim);
3363 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3364}
3365
3366/*
3367 * Close wrapper for special devices.
3368 *
3369 * Update the times on the nfsnode then do device close.
3370 *
3371 * nfsspec_close(struct vnode *a_vp, int a_fflag)
3372 */
3373static int
3374nfsspec_close(struct vop_close_args *ap)
3375{
3376 struct vnode *vp = ap->a_vp;
3377 struct nfsnode *np = VTONFS(vp);
3378 struct vattr vattr;
3379
3380 if (np->n_flag & (NACC | NUPD)) {
3381 np->n_flag |= NCHG;
3382 if (vp->v_sysref.refcnt == 1 &&
3383 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3384 VATTR_NULL(&vattr);
3385 if (np->n_flag & NACC)
3386 vattr.va_atime = np->n_atim;
3387 if (np->n_flag & NUPD)
3388 vattr.va_mtime = np->n_mtim;
3389 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3390 }
3391 }
3392 return (VOCALL(&spec_vnode_vops, &ap->a_head));
3393}
3394
3395/*
3396 * Read wrapper for fifos.
3397 *
3398 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3399 * struct ucred *a_cred)
3400 */
3401static int
3402nfsfifo_read(struct vop_read_args *ap)
3403{
3404 struct nfsnode *np = VTONFS(ap->a_vp);
3405
3406 /*
3407 * Set access flag.
3408 */
3409 np->n_flag |= NACC;
3410 getnanotime(&np->n_atim);
3411 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3412}
3413
3414/*
3415 * Write wrapper for fifos.
3416 *
3417 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3418 * struct ucred *a_cred)
3419 */
3420static int
3421nfsfifo_write(struct vop_write_args *ap)
3422{
3423 struct nfsnode *np = VTONFS(ap->a_vp);
3424
3425 /*
3426 * Set update flag.
3427 */
3428 np->n_flag |= NUPD;
3429 getnanotime(&np->n_mtim);
3430 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3431}
3432
3433/*
3434 * Close wrapper for fifos.
3435 *
3436 * Update the times on the nfsnode then do fifo close.
3437 *
3438 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3439 */
3440static int
3441nfsfifo_close(struct vop_close_args *ap)
3442{
3443 struct vnode *vp = ap->a_vp;
3444 struct nfsnode *np = VTONFS(vp);
3445 struct vattr vattr;
3446 struct timespec ts;
3447
3448 if (np->n_flag & (NACC | NUPD)) {
3449 getnanotime(&ts);
3450 if (np->n_flag & NACC)
3451 np->n_atim = ts;
3452 if (np->n_flag & NUPD)
3453 np->n_mtim = ts;
3454 np->n_flag |= NCHG;
3455 if (vp->v_sysref.refcnt == 1 &&
3456 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3457 VATTR_NULL(&vattr);
3458 if (np->n_flag & NACC)
3459 vattr.va_atime = np->n_atim;
3460 if (np->n_flag & NUPD)
3461 vattr.va_mtime = np->n_mtim;
3462 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3463 }
3464 }
3465 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3466}
3467