libstand is used by the boot loader, make sure the stack protector is
[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 1.38 2005/03/17 17:28:46 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
76#include "rpcv2.h"
77#include "nfsproto.h"
78#include "nfs.h"
79#include "nfsmount.h"
80#include "nfsnode.h"
81#include "xdr_subs.h"
82#include "nfsm_subs.h"
83#include "nqnfs.h"
84
85#include <net/if.h>
86#include <netinet/in.h>
87#include <netinet/in_var.h>
88
89/* Defs */
90#define TRUE 1
91#define FALSE 0
92
93/*
94 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
95 * calls are not in getblk() and brelse() so that they would not be necessary
96 * here.
97 */
98#ifndef B_VMIO
99#define vfs_busy_pages(bp, f)
100#endif
101
102static int nfsspec_read (struct vop_read_args *);
103static int nfsspec_write (struct vop_write_args *);
104static int nfsfifo_read (struct vop_read_args *);
105static int nfsfifo_write (struct vop_write_args *);
106static int nfsspec_close (struct vop_close_args *);
107static int nfsfifo_close (struct vop_close_args *);
108#define nfs_poll vop_nopoll
109static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
110static int nfs_lookup (struct vop_lookup_args *);
111static int nfs_create (struct vop_create_args *);
112static int nfs_mknod (struct vop_mknod_args *);
113static int nfs_open (struct vop_open_args *);
114static int nfs_close (struct vop_close_args *);
115static int nfs_access (struct vop_access_args *);
116static int nfs_getattr (struct vop_getattr_args *);
117static int nfs_setattr (struct vop_setattr_args *);
118static int nfs_read (struct vop_read_args *);
119static int nfs_mmap (struct vop_mmap_args *);
120static int nfs_fsync (struct vop_fsync_args *);
121static int nfs_remove (struct vop_remove_args *);
122static int nfs_link (struct vop_link_args *);
123static int nfs_rename (struct vop_rename_args *);
124static int nfs_mkdir (struct vop_mkdir_args *);
125static int nfs_rmdir (struct vop_rmdir_args *);
126static int nfs_symlink (struct vop_symlink_args *);
127static int nfs_readdir (struct vop_readdir_args *);
128static int nfs_bmap (struct vop_bmap_args *);
129static int nfs_strategy (struct vop_strategy_args *);
130static int nfs_lookitup (struct vnode *, const char *, int,
131 struct ucred *, struct thread *, struct nfsnode **);
132static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
133static int nfsspec_access (struct vop_access_args *);
134static int nfs_readlink (struct vop_readlink_args *);
135static int nfs_print (struct vop_print_args *);
136static int nfs_advlock (struct vop_advlock_args *);
137static int nfs_bwrite (struct vop_bwrite_args *);
138
139static int nfs_nresolve (struct vop_nresolve_args *);
140/*
141 * Global vfs data structures for nfs
142 */
143struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
144 { &vop_default_desc, vop_defaultop },
145 { &vop_access_desc, (vnodeopv_entry_t) nfs_access },
146 { &vop_advlock_desc, (vnodeopv_entry_t) nfs_advlock },
147 { &vop_bmap_desc, (vnodeopv_entry_t) nfs_bmap },
148 { &vop_bwrite_desc, (vnodeopv_entry_t) nfs_bwrite },
149 { &vop_close_desc, (vnodeopv_entry_t) nfs_close },
150 { &vop_create_desc, (vnodeopv_entry_t) nfs_create },
151 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync },
152 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr },
153 { &vop_getpages_desc, (vnodeopv_entry_t) nfs_getpages },
154 { &vop_putpages_desc, (vnodeopv_entry_t) nfs_putpages },
155 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive },
156 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked },
157 { &vop_lease_desc, vop_null },
158 { &vop_link_desc, (vnodeopv_entry_t) nfs_link },
159 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock },
160 { &vop_lookup_desc, (vnodeopv_entry_t) nfs_lookup },
161 { &vop_mkdir_desc, (vnodeopv_entry_t) nfs_mkdir },
162 { &vop_mknod_desc, (vnodeopv_entry_t) nfs_mknod },
163 { &vop_mmap_desc, (vnodeopv_entry_t) nfs_mmap },
164 { &vop_open_desc, (vnodeopv_entry_t) nfs_open },
165 { &vop_poll_desc, (vnodeopv_entry_t) nfs_poll },
166 { &vop_print_desc, (vnodeopv_entry_t) nfs_print },
167 { &vop_read_desc, (vnodeopv_entry_t) nfs_read },
168 { &vop_readdir_desc, (vnodeopv_entry_t) nfs_readdir },
169 { &vop_readlink_desc, (vnodeopv_entry_t) nfs_readlink },
170 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim },
171 { &vop_remove_desc, (vnodeopv_entry_t) nfs_remove },
172 { &vop_rename_desc, (vnodeopv_entry_t) nfs_rename },
173 { &vop_rmdir_desc, (vnodeopv_entry_t) nfs_rmdir },
174 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr },
175 { &vop_strategy_desc, (vnodeopv_entry_t) nfs_strategy },
176 { &vop_symlink_desc, (vnodeopv_entry_t) nfs_symlink },
177 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock },
178 { &vop_write_desc, (vnodeopv_entry_t) nfs_write },
179
180 { &vop_nresolve_desc, (vnodeopv_entry_t) nfs_nresolve },
181 { NULL, NULL }
182};
183
184/*
185 * Special device vnode ops
186 */
187struct vnodeopv_entry_desc nfsv2_specop_entries[] = {
188 { &vop_default_desc, (vnodeopv_entry_t) spec_vnoperate },
189 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access },
190 { &vop_close_desc, (vnodeopv_entry_t) nfsspec_close },
191 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync },
192 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr },
193 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive },
194 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked },
195 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock },
196 { &vop_print_desc, (vnodeopv_entry_t) nfs_print },
197 { &vop_read_desc, (vnodeopv_entry_t) nfsspec_read },
198 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim },
199 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr },
200 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock },
201 { &vop_write_desc, (vnodeopv_entry_t) nfsspec_write },
202 { NULL, NULL }
203};
204
205struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = {
206 { &vop_default_desc, (vnodeopv_entry_t) fifo_vnoperate },
207 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access },
208 { &vop_close_desc, (vnodeopv_entry_t) nfsfifo_close },
209 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync },
210 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr },
211 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive },
212 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked },
213 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock },
214 { &vop_print_desc, (vnodeopv_entry_t) nfs_print },
215 { &vop_read_desc, (vnodeopv_entry_t) nfsfifo_read },
216 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim },
217 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr },
218 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock },
219 { &vop_write_desc, (vnodeopv_entry_t) nfsfifo_write },
220 { NULL, NULL }
221};
222
223static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
224 struct componentname *cnp,
225 struct vattr *vap);
226static int nfs_removerpc (struct vnode *dvp, const char *name,
227 int namelen,
228 struct ucred *cred, struct thread *td);
229static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
230 int fnamelen, struct vnode *tdvp,
231 const char *tnameptr, int tnamelen,
232 struct ucred *cred, struct thread *td);
233static int nfs_renameit (struct vnode *sdvp,
234 struct componentname *scnp,
235 struct sillyrename *sp);
236
237/*
238 * Global variables
239 */
240extern u_int32_t nfs_true, nfs_false;
241extern u_int32_t nfs_xdrneg1;
242extern struct nfsstats nfsstats;
243extern nfstype nfsv3_type[9];
244struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON];
245struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
246int nfs_numasync = 0;
247#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
248
249SYSCTL_DECL(_vfs_nfs);
250
251static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
252SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
253 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
254
255static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
256SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
257 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache timeout");
258
259static int nfsv3_commit_on_close = 0;
260SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
261 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
262#if 0
263SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
264 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
265
266SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
267 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
268#endif
269
270#define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
271 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
272 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
273static int
274nfs3_access_otw(struct vnode *vp, int wmode,
275 struct thread *td, struct ucred *cred)
276{
277 const int v3 = 1;
278 u_int32_t *tl;
279 int error = 0, attrflag;
280
281 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
282 caddr_t bpos, dpos, cp2;
283 int32_t t1, t2;
284 caddr_t cp;
285 u_int32_t rmode;
286 struct nfsnode *np = VTONFS(vp);
287
288 nfsstats.rpccnt[NFSPROC_ACCESS]++;
289 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
290 nfsm_fhtom(vp, v3);
291 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
292 *tl = txdr_unsigned(wmode);
293 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
294 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
295 if (!error) {
296 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
297 rmode = fxdr_unsigned(u_int32_t, *tl);
298 np->n_mode = rmode;
299 np->n_modeuid = cred->cr_uid;
300 np->n_modestamp = mycpu->gd_time_seconds;
301 }
302 m_freem(mrep);
303nfsmout:
304 return error;
305}
306
307/*
308 * nfs access vnode op.
309 * For nfs version 2, just return ok. File accesses may fail later.
310 * For nfs version 3, use the access rpc to check accessibility. If file modes
311 * are changed on the server, accesses might still fail later.
312 *
313 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
314 * struct thread *a_td)
315 */
316static int
317nfs_access(struct vop_access_args *ap)
318{
319 struct vnode *vp = ap->a_vp;
320 int error = 0;
321 u_int32_t mode, wmode;
322 int v3 = NFS_ISV3(vp);
323 struct nfsnode *np = VTONFS(vp);
324
325 /*
326 * Disallow write attempts on filesystems mounted read-only;
327 * unless the file is a socket, fifo, or a block or character
328 * device resident on the filesystem.
329 */
330 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
331 switch (vp->v_type) {
332 case VREG:
333 case VDIR:
334 case VLNK:
335 return (EROFS);
336 default:
337 break;
338 }
339 }
340 /*
341 * For nfs v3, check to see if we have done this recently, and if
342 * so return our cached result instead of making an ACCESS call.
343 * If not, do an access rpc, otherwise you are stuck emulating
344 * ufs_access() locally using the vattr. This may not be correct,
345 * since the server may apply other access criteria such as
346 * client uid-->server uid mapping that we do not know about.
347 */
348 if (v3) {
349 if (ap->a_mode & VREAD)
350 mode = NFSV3ACCESS_READ;
351 else
352 mode = 0;
353 if (vp->v_type != VDIR) {
354 if (ap->a_mode & VWRITE)
355 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
356 if (ap->a_mode & VEXEC)
357 mode |= NFSV3ACCESS_EXECUTE;
358 } else {
359 if (ap->a_mode & VWRITE)
360 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
361 NFSV3ACCESS_DELETE);
362 if (ap->a_mode & VEXEC)
363 mode |= NFSV3ACCESS_LOOKUP;
364 }
365 /* XXX safety belt, only make blanket request if caching */
366 if (nfsaccess_cache_timeout > 0) {
367 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
368 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
369 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
370 } else {
371 wmode = mode;
372 }
373
374 /*
375 * Does our cached result allow us to give a definite yes to
376 * this request?
377 */
378 if (np->n_modestamp &&
379 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
380 (ap->a_cred->cr_uid == np->n_modeuid) &&
381 ((np->n_mode & mode) == mode)) {
382 nfsstats.accesscache_hits++;
383 } else {
384 /*
385 * Either a no, or a don't know. Go to the wire.
386 */
387 nfsstats.accesscache_misses++;
388 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
389 if (!error) {
390 if ((np->n_mode & mode) != mode) {
391 error = EACCES;
392 }
393 }
394 }
395 } else {
396 if ((error = nfsspec_access(ap)) != 0)
397 return (error);
398
399 /*
400 * Attempt to prevent a mapped root from accessing a file
401 * which it shouldn't. We try to read a byte from the file
402 * if the user is root and the file is not zero length.
403 * After calling nfsspec_access, we should have the correct
404 * file size cached.
405 */
406 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
407 && VTONFS(vp)->n_size > 0) {
408 struct iovec aiov;
409 struct uio auio;
410 char buf[1];
411
412 aiov.iov_base = buf;
413 aiov.iov_len = 1;
414 auio.uio_iov = &aiov;
415 auio.uio_iovcnt = 1;
416 auio.uio_offset = 0;
417 auio.uio_resid = 1;
418 auio.uio_segflg = UIO_SYSSPACE;
419 auio.uio_rw = UIO_READ;
420 auio.uio_td = ap->a_td;
421
422 if (vp->v_type == VREG) {
423 error = nfs_readrpc(vp, &auio);
424 } else if (vp->v_type == VDIR) {
425 char* bp;
426 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
427 aiov.iov_base = bp;
428 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
429 error = nfs_readdirrpc(vp, &auio);
430 free(bp, M_TEMP);
431 } else if (vp->v_type == VLNK) {
432 error = nfs_readlinkrpc(vp, &auio);
433 } else {
434 error = EACCES;
435 }
436 }
437 }
438 /*
439 * [re]record creds for reading and/or writing if access
440 * was granted. Assume the NFS server will grant read access
441 * for execute requests.
442 */
443 if (error == 0) {
444 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) {
445 crhold(ap->a_cred);
446 if (np->n_rucred)
447 crfree(np->n_rucred);
448 np->n_rucred = ap->a_cred;
449 }
450 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
451 crhold(ap->a_cred);
452 if (np->n_wucred)
453 crfree(np->n_wucred);
454 np->n_wucred = ap->a_cred;
455 }
456 }
457 return(error);
458}
459
460/*
461 * nfs open vnode op
462 * Check to see if the type is ok
463 * and that deletion is not in progress.
464 * For paged in text files, you will need to flush the page cache
465 * if consistency is lost.
466 *
467 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
468 * struct thread *a_td)
469 */
470/* ARGSUSED */
471static int
472nfs_open(struct vop_open_args *ap)
473{
474 struct vnode *vp = ap->a_vp;
475 struct nfsnode *np = VTONFS(vp);
476 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
477 struct vattr vattr;
478 int error;
479
480 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
481#ifdef DIAGNOSTIC
482 printf("open eacces vtyp=%d\n",vp->v_type);
483#endif
484 return (EOPNOTSUPP);
485 }
486
487 /*
488 * Clear the attribute cache only if opening with write access. It
489 * is unclear if we should do this at all here, but we certainly
490 * should not clear the cache unconditionally simply because a file
491 * is being opened.
492 */
493 if (ap->a_mode & FWRITE)
494 np->n_attrstamp = 0;
495
496 if (nmp->nm_flag & NFSMNT_NQNFS) {
497 /*
498 * If NQNFS is active, get a valid lease
499 */
500 if (NQNFS_CKINVALID(vp, np, ND_READ)) {
501 do {
502 error = nqnfs_getlease(vp, ND_READ, ap->a_td);
503 } while (error == NQNFS_EXPIRED);
504 if (error)
505 return (error);
506 if (np->n_lrev != np->n_brev ||
507 (np->n_flag & NQNFSNONCACHE)) {
508 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1))
509 == EINTR) {
510 return (error);
511 }
512 np->n_brev = np->n_lrev;
513 }
514 }
515 } else {
516 /*
517 * For normal NFS, reconcile changes made locally verses
518 * changes made remotely. Note that VOP_GETATTR only goes
519 * to the wire if the cached attribute has timed out or been
520 * cleared.
521 *
522 * If local modifications have been made clear the attribute
523 * cache to force an attribute and modified time check. If
524 * GETATTR detects that the file has been changed by someone
525 * other then us it will set NRMODIFIED.
526 *
527 * If we are opening a directory and local changes have been
528 * made we have to invalidate the cache in order to ensure
529 * that we get the most up-to-date information from the
530 * server. XXX
531 */
532 if (np->n_flag & NLMODIFIED) {
533 np->n_attrstamp = 0;
534 if (vp->v_type == VDIR) {
535 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
536 if (error == EINTR)
537 return (error);
538 nfs_invaldir(vp);
539 }
540 }
541 error = VOP_GETATTR(vp, &vattr, ap->a_td);
542 if (error)
543 return (error);
544 if (np->n_flag & NRMODIFIED) {
545 if (vp->v_type == VDIR)
546 nfs_invaldir(vp);
547 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
548 if (error == EINTR)
549 return (error);
550 np->n_flag &= ~NRMODIFIED;
551 }
552 }
553
554 return (0);
555}
556
557/*
558 * nfs close vnode op
559 * What an NFS client should do upon close after writing is a debatable issue.
560 * Most NFS clients push delayed writes to the server upon close, basically for
561 * two reasons:
562 * 1 - So that any write errors may be reported back to the client process
563 * doing the close system call. By far the two most likely errors are
564 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
565 * 2 - To put a worst case upper bound on cache inconsistency between
566 * multiple clients for the file.
567 * There is also a consistency problem for Version 2 of the protocol w.r.t.
568 * not being able to tell if other clients are writing a file concurrently,
569 * since there is no way of knowing if the changed modify time in the reply
570 * is only due to the write for this client.
571 * (NFS Version 3 provides weak cache consistency data in the reply that
572 * should be sufficient to detect and handle this case.)
573 *
574 * The current code does the following:
575 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
576 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
577 * or commit them (this satisfies 1 and 2 except for the
578 * case where the server crashes after this close but
579 * before the commit RPC, which is felt to be "good
580 * enough". Changing the last argument to nfs_flush() to
581 * a 1 would force a commit operation, if it is felt a
582 * commit is necessary now.
583 * for NQNFS - do nothing now, since 2 is dealt with via leases and
584 * 1 should be dealt with via an fsync() system call for
585 * cases where write errors are important.
586 *
587 * nfs_close(struct vnodeop_desc *a_desc, struct vnode *a_vp, int a_fflag,
588 * struct ucred *a_cred, struct thread *a_td)
589 */
590/* ARGSUSED */
591static int
592nfs_close(struct vop_close_args *ap)
593{
594 struct vnode *vp = ap->a_vp;
595 struct nfsnode *np = VTONFS(vp);
596 int error = 0;
597
598 if (vp->v_type == VREG) {
599 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
600 (np->n_flag & NLMODIFIED)) {
601 if (NFS_ISV3(vp)) {
602 /*
603 * Under NFSv3 we have dirty buffers to dispose of. We
604 * must flush them to the NFS server. We have the option
605 * of waiting all the way through the commit rpc or just
606 * waiting for the initial write. The default is to only
607 * wait through the initial write so the data is in the
608 * server's cache, which is roughly similar to the state
609 * a standard disk subsystem leaves the file in on close().
610 *
611 * We cannot clear the NLMODIFIED bit in np->n_flag due to
612 * potential races with other processes, and certainly
613 * cannot clear it if we don't commit.
614 */
615 int cm = nfsv3_commit_on_close ? 1 : 0;
616 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
617 /* np->n_flag &= ~NLMODIFIED; */
618 } else {
619 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
620 }
621 np->n_attrstamp = 0;
622 }
623 if (np->n_flag & NWRITEERR) {
624 np->n_flag &= ~NWRITEERR;
625 error = np->n_error;
626 }
627 }
628 return (error);
629}
630
631/*
632 * nfs getattr call from vfs.
633 *
634 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred,
635 * struct thread *a_td)
636 */
637static int
638nfs_getattr(struct vop_getattr_args *ap)
639{
640 struct vnode *vp = ap->a_vp;
641 struct nfsnode *np = VTONFS(vp);
642 caddr_t cp;
643 u_int32_t *tl;
644 int32_t t1, t2;
645 caddr_t bpos, dpos;
646 int error = 0;
647 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
648 int v3 = NFS_ISV3(vp);
649
650 /*
651 * Update local times for special files.
652 */
653 if (np->n_flag & (NACC | NUPD))
654 np->n_flag |= NCHG;
655 /*
656 * First look in the cache.
657 */
658 if (nfs_getattrcache(vp, ap->a_vap) == 0)
659 return (0);
660
661 if (v3 && nfsaccess_cache_timeout > 0) {
662 nfsstats.accesscache_misses++;
663 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK));
664 if (nfs_getattrcache(vp, ap->a_vap) == 0)
665 return (0);
666 }
667
668 nfsstats.rpccnt[NFSPROC_GETATTR]++;
669 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
670 nfsm_fhtom(vp, v3);
671 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK));
672 if (!error) {
673 nfsm_loadattr(vp, ap->a_vap);
674 }
675 m_freem(mrep);
676nfsmout:
677 return (error);
678}
679
680/*
681 * nfs setattr call.
682 *
683 * nfs_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp,
684 * struct vattr *a_vap, struct ucred *a_cred,
685 * struct thread *a_td)
686 */
687static int
688nfs_setattr(struct vop_setattr_args *ap)
689{
690 struct vnode *vp = ap->a_vp;
691 struct nfsnode *np = VTONFS(vp);
692 struct vattr *vap = ap->a_vap;
693 int error = 0;
694 u_quad_t tsize;
695
696#ifndef nolint
697 tsize = (u_quad_t)0;
698#endif
699
700 /*
701 * Setting of flags is not supported.
702 */
703 if (vap->va_flags != VNOVAL)
704 return (EOPNOTSUPP);
705
706 /*
707 * Disallow write attempts if the filesystem is mounted read-only.
708 */
709 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
710 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
711 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
712 (vp->v_mount->mnt_flag & MNT_RDONLY))
713 return (EROFS);
714 if (vap->va_size != VNOVAL) {
715 switch (vp->v_type) {
716 case VDIR:
717 return (EISDIR);
718 case VCHR:
719 case VBLK:
720 case VSOCK:
721 case VFIFO:
722 if (vap->va_mtime.tv_sec == VNOVAL &&
723 vap->va_atime.tv_sec == VNOVAL &&
724 vap->va_mode == (mode_t)VNOVAL &&
725 vap->va_uid == (uid_t)VNOVAL &&
726 vap->va_gid == (gid_t)VNOVAL)
727 return (0);
728 vap->va_size = VNOVAL;
729 break;
730 default:
731 /*
732 * Disallow write attempts if the filesystem is
733 * mounted read-only.
734 */
735 if (vp->v_mount->mnt_flag & MNT_RDONLY)
736 return (EROFS);
737
738 /*
739 * We run vnode_pager_setsize() early (why?),
740 * we must set np->n_size now to avoid vinvalbuf
741 * V_SAVE races that might setsize a lower
742 * value.
743 */
744
745 tsize = np->n_size;
746 error = nfs_meta_setsize(vp, ap->a_td, vap->va_size);
747
748 if (np->n_flag & NLMODIFIED) {
749 if (vap->va_size == 0)
750 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
751 else
752 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
753 if (error) {
754 np->n_size = tsize;
755 vnode_pager_setsize(vp, np->n_size);
756 return (error);
757 }
758 }
759 /*
760 * np->n_size has already been set to vap->va_size
761 * in nfs_meta_setsize(). We must set it again since
762 * nfs_loadattrcache() could be called through
763 * nfs_meta_setsize() and could modify np->n_size.
764 *
765 * (note that nfs_loadattrcache() will have called
766 * vnode_pager_setsize() for us in that case).
767 */
768 np->n_vattr.va_size = np->n_size = vap->va_size;
769 break;
770 }
771 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
772 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) &&
773 vp->v_type == VREG &&
774 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR)
775 return (error);
776 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
777 if (error && vap->va_size != VNOVAL) {
778 np->n_size = np->n_vattr.va_size = tsize;
779 vnode_pager_setsize(vp, np->n_size);
780 }
781 return (error);
782}
783
784/*
785 * Do an nfs setattr rpc.
786 */
787static int
788nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
789 struct ucred *cred, struct thread *td)
790{
791 struct nfsv2_sattr *sp;
792 struct nfsnode *np = VTONFS(vp);
793 caddr_t cp;
794 int32_t t1, t2;
795 caddr_t bpos, dpos, cp2;
796 u_int32_t *tl;
797 int error = 0, wccflag = NFSV3_WCCRATTR;
798 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
799 int v3 = NFS_ISV3(vp);
800
801 nfsstats.rpccnt[NFSPROC_SETATTR]++;
802 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
803 nfsm_fhtom(vp, v3);
804 if (v3) {
805 nfsm_v3attrbuild(vap, TRUE);
806 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
807 *tl = nfs_false;
808 } else {
809 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
810 if (vap->va_mode == (mode_t)VNOVAL)
811 sp->sa_mode = nfs_xdrneg1;
812 else
813 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
814 if (vap->va_uid == (uid_t)VNOVAL)
815 sp->sa_uid = nfs_xdrneg1;
816 else
817 sp->sa_uid = txdr_unsigned(vap->va_uid);
818 if (vap->va_gid == (gid_t)VNOVAL)
819 sp->sa_gid = nfs_xdrneg1;
820 else
821 sp->sa_gid = txdr_unsigned(vap->va_gid);
822 sp->sa_size = txdr_unsigned(vap->va_size);
823 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
824 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
825 }
826 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
827 if (v3) {
828 np->n_modestamp = 0;
829 nfsm_wcc_data(vp, wccflag);
830 } else
831 nfsm_loadattr(vp, (struct vattr *)0);
832 m_freem(mrep);
833nfsmout:
834 return (error);
835}
836
837/*
838 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
839 * nfs_lookup() until all remaining new api calls are implemented.
840 *
841 * Resolve a namecache entry. This function is passed a locked ncp and
842 * must call cache_setvp() on it as appropriate to resolve the entry.
843 */
844static int
845nfs_nresolve(struct vop_nresolve_args *ap)
846{
847 struct thread *td = curthread;
848 struct namecache *ncp;
849 struct ucred *cred;
850 struct nfsnode *np;
851 struct vnode *dvp;
852 struct vnode *nvp;
853 nfsfh_t *fhp;
854 int attrflag;
855 int fhsize;
856 int error;
857 int len;
858 int v3;
859 /******NFSM MACROS********/
860 struct mbuf *mb, *mrep, *mreq, *mb2, *md;
861 caddr_t bpos, dpos, cp, cp2;
862 u_int32_t *tl;
863 int32_t t1, t2;
864
865 cred = ap->a_cred;
866 ncp = ap->a_ncp;
867
868 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp);
869 dvp = ncp->nc_parent->nc_vp;
870 if ((error = vget(dvp, LK_SHARED, td)) != 0)
871 return (error);
872
873 nvp = NULL;
874 v3 = NFS_ISV3(dvp);
875 nfsstats.lookupcache_misses++;
876 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
877 len = ncp->nc_nlen;
878 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
879 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
880 nfsm_fhtom(dvp, v3);
881 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN);
882 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred);
883 if (error) {
884 /*
885 * Cache negatve lookups to reduce NFS traffic, but use
886 * a fast timeout. Otherwise use a timeout of 1 tick.
887 * XXX we should add a namecache flag for no-caching
888 * to uncache the negative hit as soon as possible, but
889 * we cannot simply destroy the entry because it is used
890 * as a placeholder by the caller.
891 */
892 if (error == ENOENT) {
893 int nticks;
894
895 if (nfsneg_cache_timeout)
896 nticks = nfsneg_cache_timeout * hz;
897 else
898 nticks = 1;
899 cache_setvp(ncp, NULL);
900 cache_settimeout(ncp, nticks);
901 }
902 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
903 m_freem(mrep);
904 goto nfsmout;
905 }
906
907 /*
908 * Success, get the file handle, do various checks, and load
909 * post-operation data from the reply packet. Theoretically
910 * we should never be looking up "." so, theoretically, we
911 * should never get the same file handle as our directory. But
912 * we check anyway. XXX
913 *
914 * Note that no timeout is set for the positive cache hit. We
915 * assume, theoretically, that ESTALE returns will be dealt with
916 * properly to handle NFS races and in anycase we cannot depend
917 * on a timeout to deal with NFS open/create/excl issues so instead
918 * of a bad hack here the rest of the NFS client code needs to do
919 * the right thing.
920 */
921 nfsm_getfh(fhp, fhsize, v3);
922
923 np = VTONFS(dvp);
924 if (NFS_CMPFH(np, fhp, fhsize)) {
925 vref(dvp);
926 nvp = dvp;
927 } else {
928 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
929 if (error) {
930 m_freem(mrep);
931 vput(dvp);
932 return (error);
933 }
934 nvp = NFSTOV(np);
935 }
936 if (v3) {
937 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK);
938 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
939 } else {
940 nfsm_loadattr(nvp, NULL);
941 }
942 cache_setvp(ncp, nvp);
943 m_freem(mrep);
944nfsmout:
945 vput(dvp);
946 if (nvp) {
947 if (nvp == dvp)
948 vrele(nvp);
949 else
950 vput(nvp);
951 }
952 return (error);
953}
954
955/*
956 * 'cached' nfs directory lookup
957 *
958 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
959 *
960 * nfs_lookup(struct vnodeop_desc *a_desc, struct vnode *a_dvp,
961 * struct vnode **a_vpp, struct componentname *a_cnp)
962 */
963static int
964nfs_lookup(struct vop_lookup_args *ap)
965{
966 struct componentname *cnp = ap->a_cnp;
967 struct vnode *dvp = ap->a_dvp;
968 struct vnode **vpp = ap->a_vpp;
969 int flags = cnp->cn_flags;
970 struct vnode *newvp;
971 u_int32_t *tl;
972 caddr_t cp;
973 int32_t t1, t2;
974 struct nfsmount *nmp;
975 caddr_t bpos, dpos, cp2;
976 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
977 long len;
978 nfsfh_t *fhp;
979 struct nfsnode *np;
980 int lockparent, wantparent, error = 0, attrflag, fhsize;
981 int v3 = NFS_ISV3(dvp);
982 struct thread *td = cnp->cn_td;
983
984 /*
985 * Read-only mount check and directory check.
986 */
987 *vpp = NULLVP;
988 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
989 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
990 return (EROFS);
991
992 if (dvp->v_type != VDIR)
993 return (ENOTDIR);
994
995 /*
996 * Look it up in the cache. Note that ENOENT is only returned if we
997 * previously entered a negative hit (see later on). The additional
998 * nfsneg_cache_timeout check causes previously cached results to
999 * be instantly ignored if the negative caching is turned off.
1000 */
1001 lockparent = flags & CNP_LOCKPARENT;
1002 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1003 nmp = VFSTONFS(dvp->v_mount);
1004 np = VTONFS(dvp);
1005
1006 /*
1007 * Go to the wire.
1008 */
1009 error = 0;
1010 newvp = NULLVP;
1011 nfsstats.lookupcache_misses++;
1012 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1013 len = cnp->cn_namelen;
1014 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1015 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1016 nfsm_fhtom(dvp, v3);
1017 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1018 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred);
1019 if (error) {
1020 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1021 m_freem(mrep);
1022 goto nfsmout;
1023 }
1024 nfsm_getfh(fhp, fhsize, v3);
1025
1026 /*
1027 * Handle RENAME case...
1028 */
1029 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1030 if (NFS_CMPFH(np, fhp, fhsize)) {
1031 m_freem(mrep);
1032 return (EISDIR);
1033 }
1034 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1035 if (error) {
1036 m_freem(mrep);
1037 return (error);
1038 }
1039 newvp = NFSTOV(np);
1040 if (v3) {
1041 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1042 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1043 } else
1044 nfsm_loadattr(newvp, (struct vattr *)0);
1045 *vpp = newvp;
1046 m_freem(mrep);
1047 if (!lockparent) {
1048 VOP_UNLOCK(dvp, 0, td);
1049 cnp->cn_flags |= CNP_PDIRUNLOCK;
1050 }
1051 return (0);
1052 }
1053
1054 if (flags & CNP_ISDOTDOT) {
1055 VOP_UNLOCK(dvp, 0, td);
1056 cnp->cn_flags |= CNP_PDIRUNLOCK;
1057 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1058 if (error) {
1059 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
1060 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1061 return (error); /* NOTE: return error from nget */
1062 }
1063 newvp = NFSTOV(np);
1064 if (lockparent) {
1065 error = vn_lock(dvp, LK_EXCLUSIVE, td);
1066 if (error) {
1067 vput(newvp);
1068 return (error);
1069 }
1070 cnp->cn_flags |= CNP_PDIRUNLOCK;
1071 }
1072 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1073 vref(dvp);
1074 newvp = dvp;
1075 } else {
1076 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1077 if (error) {
1078 m_freem(mrep);
1079 return (error);
1080 }
1081 if (!lockparent) {
1082 VOP_UNLOCK(dvp, 0, td);
1083 cnp->cn_flags |= CNP_PDIRUNLOCK;
1084 }
1085 newvp = NFSTOV(np);
1086 }
1087 if (v3) {
1088 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1089 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1090 } else
1091 nfsm_loadattr(newvp, (struct vattr *)0);
1092#if 0
1093 /* XXX MOVE TO nfs_nremove() */
1094 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1095 cnp->cn_nameiop != NAMEI_DELETE) {
1096 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1097 }
1098#endif
1099 *vpp = newvp;
1100 m_freem(mrep);
1101nfsmout:
1102 if (error) {
1103 if (newvp != NULLVP) {
1104 vrele(newvp);
1105 *vpp = NULLVP;
1106 }
1107 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1108 cnp->cn_nameiop == NAMEI_RENAME) &&
1109 error == ENOENT) {
1110 if (!lockparent) {
1111 VOP_UNLOCK(dvp, 0, td);
1112 cnp->cn_flags |= CNP_PDIRUNLOCK;
1113 }
1114 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1115 error = EROFS;
1116 else
1117 error = EJUSTRETURN;
1118 }
1119 }
1120 return (error);
1121}
1122
1123/*
1124 * nfs read call.
1125 * Just call nfs_bioread() to do the work.
1126 *
1127 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1128 * struct ucred *a_cred)
1129 */
1130static int
1131nfs_read(struct vop_read_args *ap)
1132{
1133 struct vnode *vp = ap->a_vp;
1134
1135 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1136 switch (vp->v_type) {
1137 case VREG:
1138 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1139 case VDIR:
1140 return (EISDIR);
1141 default:
1142 return EOPNOTSUPP;
1143 }
1144}
1145
1146/*
1147 * nfs readlink call
1148 *
1149 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1150 */
1151static int
1152nfs_readlink(struct vop_readlink_args *ap)
1153{
1154 struct vnode *vp = ap->a_vp;
1155
1156 if (vp->v_type != VLNK)
1157 return (EINVAL);
1158 return (nfs_bioread(vp, ap->a_uio, 0));
1159}
1160
1161/*
1162 * Do a readlink rpc.
1163 * Called by nfs_doio() from below the buffer cache.
1164 */
1165int
1166nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1167{
1168 u_int32_t *tl;
1169 caddr_t cp;
1170 int32_t t1, t2;
1171 caddr_t bpos, dpos, cp2;
1172 int error = 0, len, attrflag;
1173 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1174 int v3 = NFS_ISV3(vp);
1175
1176 nfsstats.rpccnt[NFSPROC_READLINK]++;
1177 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1178 nfsm_fhtom(vp, v3);
1179 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK));
1180 if (v3)
1181 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1182 if (!error) {
1183 nfsm_strsiz(len, NFS_MAXPATHLEN);
1184 if (len == NFS_MAXPATHLEN) {
1185 struct nfsnode *np = VTONFS(vp);
1186 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1187 len = np->n_size;
1188 }
1189 nfsm_mtouio(uiop, len);
1190 }
1191 m_freem(mrep);
1192nfsmout:
1193 return (error);
1194}
1195
1196/*
1197 * nfs read rpc call
1198 * Ditto above
1199 */
1200int
1201nfs_readrpc(struct vnode *vp, struct uio *uiop)
1202{
1203 u_int32_t *tl;
1204 caddr_t cp;
1205 int32_t t1, t2;
1206 caddr_t bpos, dpos, cp2;
1207 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1208 struct nfsmount *nmp;
1209 int error = 0, len, retlen, tsiz, eof, attrflag;
1210 int v3 = NFS_ISV3(vp);
1211
1212#ifndef nolint
1213 eof = 0;
1214#endif
1215 nmp = VFSTONFS(vp->v_mount);
1216 tsiz = uiop->uio_resid;
1217 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1218 return (EFBIG);
1219 while (tsiz > 0) {
1220 nfsstats.rpccnt[NFSPROC_READ]++;
1221 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1222 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1223 nfsm_fhtom(vp, v3);
1224 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1225 if (v3) {
1226 txdr_hyper(uiop->uio_offset, tl);
1227 *(tl + 2) = txdr_unsigned(len);
1228 } else {
1229 *tl++ = txdr_unsigned(uiop->uio_offset);
1230 *tl++ = txdr_unsigned(len);
1231 *tl = 0;
1232 }
1233 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ));
1234 if (v3) {
1235 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1236 if (error) {
1237 m_freem(mrep);
1238 goto nfsmout;
1239 }
1240 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1241 eof = fxdr_unsigned(int, *(tl + 1));
1242 } else
1243 nfsm_loadattr(vp, (struct vattr *)0);
1244 nfsm_strsiz(retlen, nmp->nm_rsize);
1245 nfsm_mtouio(uiop, retlen);
1246 m_freem(mrep);
1247 tsiz -= retlen;
1248 if (v3) {
1249 if (eof || retlen == 0) {
1250 tsiz = 0;
1251 }
1252 } else if (retlen < len) {
1253 tsiz = 0;
1254 }
1255 }
1256nfsmout:
1257 return (error);
1258}
1259
1260/*
1261 * nfs write call
1262 */
1263int
1264nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1265{
1266 u_int32_t *tl;
1267 caddr_t cp;
1268 int32_t t1, t2, backup;
1269 caddr_t bpos, dpos, cp2;
1270 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1271 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1272 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1273 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1274
1275#ifndef DIAGNOSTIC
1276 if (uiop->uio_iovcnt != 1)
1277 panic("nfs: writerpc iovcnt > 1");
1278#endif
1279 *must_commit = 0;
1280 tsiz = uiop->uio_resid;
1281 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1282 return (EFBIG);
1283 while (tsiz > 0) {
1284 nfsstats.rpccnt[NFSPROC_WRITE]++;
1285 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1286 nfsm_reqhead(vp, NFSPROC_WRITE,
1287 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1288 nfsm_fhtom(vp, v3);
1289 if (v3) {
1290 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1291 txdr_hyper(uiop->uio_offset, tl);
1292 tl += 2;
1293 *tl++ = txdr_unsigned(len);
1294 *tl++ = txdr_unsigned(*iomode);
1295 *tl = txdr_unsigned(len);
1296 } else {
1297 u_int32_t x;
1298
1299 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1300 /* Set both "begin" and "current" to non-garbage. */
1301 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1302 *tl++ = x; /* "begin offset" */
1303 *tl++ = x; /* "current offset" */
1304 x = txdr_unsigned(len);
1305 *tl++ = x; /* total to this offset */
1306 *tl = x; /* size of this write */
1307 }
1308 nfsm_uiotom(uiop, len);
1309 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE));
1310 if (v3) {
1311 /*
1312 * The write RPC returns a before and after mtime. The
1313 * nfsm_wcc_data() macro checks the before n_mtime
1314 * against the before time and stores the after time
1315 * in the nfsnode's cached vattr and n_mtime field.
1316 * The NRMODIFIED bit will be set if the before
1317 * time did not match the original mtime.
1318 */
1319 wccflag = NFSV3_WCCCHK;
1320 nfsm_wcc_data(vp, wccflag);
1321 if (!error) {
1322 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1323 + NFSX_V3WRITEVERF);
1324 rlen = fxdr_unsigned(int, *tl++);
1325 if (rlen == 0) {
1326 error = NFSERR_IO;
1327 m_freem(mrep);
1328 break;
1329 } else if (rlen < len) {
1330 backup = len - rlen;
1331 uiop->uio_iov->iov_base -= backup;
1332 uiop->uio_iov->iov_len += backup;
1333 uiop->uio_offset -= backup;
1334 uiop->uio_resid += backup;
1335 len = rlen;
1336 }
1337 commit = fxdr_unsigned(int, *tl++);
1338
1339 /*
1340 * Return the lowest committment level
1341 * obtained by any of the RPCs.
1342 */
1343 if (committed == NFSV3WRITE_FILESYNC)
1344 committed = commit;
1345 else if (committed == NFSV3WRITE_DATASYNC &&
1346 commit == NFSV3WRITE_UNSTABLE)
1347 committed = commit;
1348 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1349 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1350 NFSX_V3WRITEVERF);
1351 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1352 } else if (bcmp((caddr_t)tl,
1353 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1354 *must_commit = 1;
1355 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1356 NFSX_V3WRITEVERF);
1357 }
1358 }
1359 } else {
1360 nfsm_loadattr(vp, (struct vattr *)0);
1361 }
1362 m_freem(mrep);
1363 if (error)
1364 break;
1365 tsiz -= len;
1366 }
1367nfsmout:
1368 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1369 committed = NFSV3WRITE_FILESYNC;
1370 *iomode = committed;
1371 if (error)
1372 uiop->uio_resid = tsiz;
1373 return (error);
1374}
1375
1376/*
1377 * nfs mknod rpc
1378 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1379 * mode set to specify the file type and the size field for rdev.
1380 */
1381static int
1382nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1383 struct vattr *vap)
1384{
1385 struct nfsv2_sattr *sp;
1386 u_int32_t *tl;
1387 caddr_t cp;
1388 int32_t t1, t2;
1389 struct vnode *newvp = (struct vnode *)0;
1390 struct nfsnode *np = (struct nfsnode *)0;
1391 struct vattr vattr;
1392 char *cp2;
1393 caddr_t bpos, dpos;
1394 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1395 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1396 u_int32_t rdev;
1397 int v3 = NFS_ISV3(dvp);
1398
1399 if (vap->va_type == VCHR || vap->va_type == VBLK)
1400 rdev = txdr_unsigned(vap->va_rdev);
1401 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1402 rdev = nfs_xdrneg1;
1403 else {
1404 return (EOPNOTSUPP);
1405 }
1406 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1407 return (error);
1408 }
1409 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1410 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1411 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1412 nfsm_fhtom(dvp, v3);
1413 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1414 if (v3) {
1415 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1416 *tl++ = vtonfsv3_type(vap->va_type);
1417 nfsm_v3attrbuild(vap, FALSE);
1418 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1419 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1420 *tl++ = txdr_unsigned(umajor(vap->va_rdev));
1421 *tl = txdr_unsigned(uminor(vap->va_rdev));
1422 }
1423 } else {
1424 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1425 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1426 sp->sa_uid = nfs_xdrneg1;
1427 sp->sa_gid = nfs_xdrneg1;
1428 sp->sa_size = rdev;
1429 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1430 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1431 }
1432 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred);
1433 if (!error) {
1434 nfsm_mtofh(dvp, newvp, v3, gotvp);
1435 if (!gotvp) {
1436 if (newvp) {
1437 vput(newvp);
1438 newvp = (struct vnode *)0;
1439 }
1440 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1441 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1442 if (!error)
1443 newvp = NFSTOV(np);
1444 }
1445 }
1446 if (v3)
1447 nfsm_wcc_data(dvp, wccflag);
1448 m_freem(mrep);
1449nfsmout:
1450 if (error) {
1451 if (newvp)
1452 vput(newvp);
1453 } else {
1454 *vpp = newvp;
1455 }
1456 VTONFS(dvp)->n_flag |= NLMODIFIED;
1457 if (!wccflag)
1458 VTONFS(dvp)->n_attrstamp = 0;
1459 return (error);
1460}
1461
1462/*
1463 * nfs mknod vop
1464 * just call nfs_mknodrpc() to do the work.
1465 *
1466 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1467 * struct componentname *a_cnp, struct vattr *a_vap)
1468 */
1469/* ARGSUSED */
1470static int
1471nfs_mknod(struct vop_mknod_args *ap)
1472{
1473 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1474}
1475
1476static u_long create_verf;
1477/*
1478 * nfs file create call
1479 *
1480 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1481 * struct componentname *a_cnp, struct vattr *a_vap)
1482 */
1483static int
1484nfs_create(struct vop_create_args *ap)
1485{
1486 struct vnode *dvp = ap->a_dvp;
1487 struct vattr *vap = ap->a_vap;
1488 struct componentname *cnp = ap->a_cnp;
1489 struct nfsv2_sattr *sp;
1490 u_int32_t *tl;
1491 caddr_t cp;
1492 int32_t t1, t2;
1493 struct nfsnode *np = (struct nfsnode *)0;
1494 struct vnode *newvp = (struct vnode *)0;
1495 caddr_t bpos, dpos, cp2;
1496 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1497 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1498 struct vattr vattr;
1499 int v3 = NFS_ISV3(dvp);
1500
1501 /*
1502 * Oops, not for me..
1503 */
1504 if (vap->va_type == VSOCK)
1505 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1506
1507 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1508 return (error);
1509 }
1510 if (vap->va_vaflags & VA_EXCLUSIVE)
1511 fmode |= O_EXCL;
1512again:
1513 nfsstats.rpccnt[NFSPROC_CREATE]++;
1514 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1515 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1516 nfsm_fhtom(dvp, v3);
1517 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1518 if (v3) {
1519 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1520 if (fmode & O_EXCL) {
1521 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1522 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1523#ifdef INET
1524 if (!TAILQ_EMPTY(&in_ifaddrhead))
1525 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1526 else
1527#endif
1528 *tl++ = create_verf;
1529 *tl = ++create_verf;
1530 } else {
1531 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1532 nfsm_v3attrbuild(vap, FALSE);
1533 }
1534 } else {
1535 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1536 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1537 sp->sa_uid = nfs_xdrneg1;
1538 sp->sa_gid = nfs_xdrneg1;
1539 sp->sa_size = 0;
1540 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1541 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1542 }
1543 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred);
1544 if (!error) {
1545 nfsm_mtofh(dvp, newvp, v3, gotvp);
1546 if (!gotvp) {
1547 if (newvp) {
1548 vput(newvp);
1549 newvp = (struct vnode *)0;
1550 }
1551 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1552 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1553 if (!error)
1554 newvp = NFSTOV(np);
1555 }
1556 }
1557 if (v3)
1558 nfsm_wcc_data(dvp, wccflag);
1559 m_freem(mrep);
1560nfsmout:
1561 if (error) {
1562 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1563 fmode &= ~O_EXCL;
1564 goto again;
1565 }
1566 if (newvp)
1567 vput(newvp);
1568 } else if (v3 && (fmode & O_EXCL)) {
1569 /*
1570 * We are normally called with only a partially initialized
1571 * VAP. Since the NFSv3 spec says that server may use the
1572 * file attributes to store the verifier, the spec requires
1573 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1574 * in atime, but we can't really assume that all servers will
1575 * so we ensure that our SETATTR sets both atime and mtime.
1576 */
1577 if (vap->va_mtime.tv_sec == VNOVAL)
1578 vfs_timestamp(&vap->va_mtime);
1579 if (vap->va_atime.tv_sec == VNOVAL)
1580 vap->va_atime = vap->va_mtime;
1581 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1582 }
1583 if (!error) {
1584 /*
1585 * The new np may have enough info for access
1586 * checks, make sure rucred and wucred are
1587 * initialized for read and write rpc's.
1588 */
1589 np = VTONFS(newvp);
1590 if (np->n_rucred == NULL)
1591 np->n_rucred = crhold(cnp->cn_cred);
1592 if (np->n_wucred == NULL)
1593 np->n_wucred = crhold(cnp->cn_cred);
1594 *ap->a_vpp = newvp;
1595 }
1596 VTONFS(dvp)->n_flag |= NLMODIFIED;
1597 if (!wccflag)
1598 VTONFS(dvp)->n_attrstamp = 0;
1599 return (error);
1600}
1601
1602/*
1603 * nfs file remove call
1604 * To try and make nfs semantics closer to ufs semantics, a file that has
1605 * other processes using the vnode is renamed instead of removed and then
1606 * removed later on the last close.
1607 * - If v_usecount > 1
1608 * If a rename is not already in the works
1609 * call nfs_sillyrename() to set it up
1610 * else
1611 * do the remove rpc
1612 *
1613 * nfs_remove(struct vnodeop_desc *a_desc, struct vnode *a_dvp,
1614 * struct vnode *a_vp, struct componentname *a_cnp)
1615 */
1616static int
1617nfs_remove(struct vop_remove_args *ap)
1618{
1619 struct vnode *vp = ap->a_vp;
1620 struct vnode *dvp = ap->a_dvp;
1621 struct componentname *cnp = ap->a_cnp;
1622 struct nfsnode *np = VTONFS(vp);
1623 int error = 0;
1624 struct vattr vattr;
1625
1626#ifndef DIAGNOSTIC
1627 if (vp->v_usecount < 1)
1628 panic("nfs_remove: bad v_usecount");
1629#endif
1630 if (vp->v_type == VDIR)
1631 error = EPERM;
1632 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1633 VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 &&
1634 vattr.va_nlink > 1)) {
1635 /*
1636 * throw away biocache buffers, mainly to avoid
1637 * unnecessary delayed writes later.
1638 */
1639 error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1);
1640 /* Do the rpc */
1641 if (error != EINTR)
1642 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1643 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1644 /*
1645 * Kludge City: If the first reply to the remove rpc is lost..
1646 * the reply to the retransmitted request will be ENOENT
1647 * since the file was in fact removed
1648 * Therefore, we cheat and return success.
1649 */
1650 if (error == ENOENT)
1651 error = 0;
1652 } else if (!np->n_sillyrename) {
1653 error = nfs_sillyrename(dvp, vp, cnp);
1654 }
1655 np->n_attrstamp = 0;
1656 return (error);
1657}
1658
1659/*
1660 * nfs file remove rpc called from nfs_inactive
1661 */
1662int
1663nfs_removeit(struct sillyrename *sp)
1664{
1665 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1666 sp->s_cred, NULL));
1667}
1668
1669/*
1670 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1671 */
1672static int
1673nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1674 struct ucred *cred, struct thread *td)
1675{
1676 u_int32_t *tl;
1677 caddr_t cp;
1678 int32_t t1, t2;
1679 caddr_t bpos, dpos, cp2;
1680 int error = 0, wccflag = NFSV3_WCCRATTR;
1681 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1682 int v3 = NFS_ISV3(dvp);
1683
1684 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1685 nfsm_reqhead(dvp, NFSPROC_REMOVE,
1686 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1687 nfsm_fhtom(dvp, v3);
1688 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1689 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1690 if (v3)
1691 nfsm_wcc_data(dvp, wccflag);
1692 m_freem(mrep);
1693nfsmout:
1694 VTONFS(dvp)->n_flag |= NLMODIFIED;
1695 if (!wccflag)
1696 VTONFS(dvp)->n_attrstamp = 0;
1697 return (error);
1698}
1699
1700/*
1701 * nfs file rename call
1702 *
1703 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1704 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1705 * struct vnode *a_tvp, struct componentname *a_tcnp)
1706 */
1707static int
1708nfs_rename(struct vop_rename_args *ap)
1709{
1710 struct vnode *fvp = ap->a_fvp;
1711 struct vnode *tvp = ap->a_tvp;
1712 struct vnode *fdvp = ap->a_fdvp;
1713 struct vnode *tdvp = ap->a_tdvp;
1714 struct componentname *tcnp = ap->a_tcnp;
1715 struct componentname *fcnp = ap->a_fcnp;
1716 int error;
1717
1718 /* Check for cross-device rename */
1719 if ((fvp->v_mount != tdvp->v_mount) ||
1720 (tvp && (fvp->v_mount != tvp->v_mount))) {
1721 error = EXDEV;
1722 goto out;
1723 }
1724
1725 /*
1726 * We have to flush B_DELWRI data prior to renaming
1727 * the file. If we don't, the delayed-write buffers
1728 * can be flushed out later after the file has gone stale
1729 * under NFSV3. NFSV2 does not have this problem because
1730 * ( as far as I can tell ) it flushes dirty buffers more
1731 * often.
1732 */
1733
1734 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td);
1735 if (tvp)
1736 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td);
1737
1738 /*
1739 * If the tvp exists and is in use, sillyrename it before doing the
1740 * rename of the new file over it.
1741 *
1742 * XXX Can't sillyrename a directory.
1743 *
1744 * We do not attempt to do any namecache purges in this old API
1745 * routine. The new API compat functions have access to the actual
1746 * namecache structures and will do it for us.
1747 */
1748 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1749 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1750 vput(tvp);
1751 tvp = NULL;
1752 } else if (tvp) {
1753 ;
1754 }
1755
1756 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1757 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1758 tcnp->cn_td);
1759
1760out:
1761 if (tdvp == tvp)
1762 vrele(tdvp);
1763 else
1764 vput(tdvp);
1765 if (tvp)
1766 vput(tvp);
1767 vrele(fdvp);
1768 vrele(fvp);
1769 /*
1770 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1771 */
1772 if (error == ENOENT)
1773 error = 0;
1774 return (error);
1775}
1776
1777/*
1778 * nfs file rename rpc called from nfs_remove() above
1779 */
1780static int
1781nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1782 struct sillyrename *sp)
1783{
1784 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1785 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1786}
1787
1788/*
1789 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1790 */
1791static int
1792nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1793 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1794 struct ucred *cred, struct thread *td)
1795{
1796 u_int32_t *tl;
1797 caddr_t cp;
1798 int32_t t1, t2;
1799 caddr_t bpos, dpos, cp2;
1800 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1801 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1802 int v3 = NFS_ISV3(fdvp);
1803
1804 nfsstats.rpccnt[NFSPROC_RENAME]++;
1805 nfsm_reqhead(fdvp, NFSPROC_RENAME,
1806 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1807 nfsm_rndup(tnamelen));
1808 nfsm_fhtom(fdvp, v3);
1809 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1810 nfsm_fhtom(tdvp, v3);
1811 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1812 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1813 if (v3) {
1814 nfsm_wcc_data(fdvp, fwccflag);
1815 nfsm_wcc_data(tdvp, twccflag);
1816 }
1817 m_freem(mrep);
1818nfsmout:
1819 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1820 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1821 if (!fwccflag)
1822 VTONFS(fdvp)->n_attrstamp = 0;
1823 if (!twccflag)
1824 VTONFS(tdvp)->n_attrstamp = 0;
1825 return (error);
1826}
1827
1828/*
1829 * nfs hard link create call
1830 *
1831 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1832 * struct componentname *a_cnp)
1833 */
1834static int
1835nfs_link(struct vop_link_args *ap)
1836{
1837 struct vnode *vp = ap->a_vp;
1838 struct vnode *tdvp = ap->a_tdvp;
1839 struct componentname *cnp = ap->a_cnp;
1840 u_int32_t *tl;
1841 caddr_t cp;
1842 int32_t t1, t2;
1843 caddr_t bpos, dpos, cp2;
1844 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1845 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1846 int v3;
1847
1848 if (vp->v_mount != tdvp->v_mount) {
1849 return (EXDEV);
1850 }
1851
1852 /*
1853 * Push all writes to the server, so that the attribute cache
1854 * doesn't get "out of sync" with the server.
1855 * XXX There should be a better way!
1856 */
1857 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td);
1858
1859 v3 = NFS_ISV3(vp);
1860 nfsstats.rpccnt[NFSPROC_LINK]++;
1861 nfsm_reqhead(vp, NFSPROC_LINK,
1862 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1863 nfsm_fhtom(vp, v3);
1864 nfsm_fhtom(tdvp, v3);
1865 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1866 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred);
1867 if (v3) {
1868 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1869 nfsm_wcc_data(tdvp, wccflag);
1870 }
1871 m_freem(mrep);
1872nfsmout:
1873 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1874 if (!attrflag)
1875 VTONFS(vp)->n_attrstamp = 0;
1876 if (!wccflag)
1877 VTONFS(tdvp)->n_attrstamp = 0;
1878 /*
1879 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1880 */
1881 if (error == EEXIST)
1882 error = 0;
1883 return (error);
1884}
1885
1886/*
1887 * nfs symbolic link create call
1888 *
1889 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1890 * struct componentname *a_cnp, struct vattr *a_vap,
1891 * char *a_target)
1892 */
1893static int
1894nfs_symlink(struct vop_symlink_args *ap)
1895{
1896 struct vnode *dvp = ap->a_dvp;
1897 struct vattr *vap = ap->a_vap;
1898 struct componentname *cnp = ap->a_cnp;
1899 struct nfsv2_sattr *sp;
1900 u_int32_t *tl;
1901 caddr_t cp;
1902 int32_t t1, t2;
1903 caddr_t bpos, dpos, cp2;
1904 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1905 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1906 struct vnode *newvp = (struct vnode *)0;
1907 int v3 = NFS_ISV3(dvp);
1908
1909 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1910 slen = strlen(ap->a_target);
1911 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1912 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1913 nfsm_fhtom(dvp, v3);
1914 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1915 if (v3) {
1916 nfsm_v3attrbuild(vap, FALSE);
1917 }
1918 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1919 if (!v3) {
1920 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1921 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1922 sp->sa_uid = nfs_xdrneg1;
1923 sp->sa_gid = nfs_xdrneg1;
1924 sp->sa_size = nfs_xdrneg1;
1925 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1926 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1927 }
1928
1929 /*
1930 * Issue the NFS request and get the rpc response.
1931 *
1932 * Only NFSv3 responses returning an error of 0 actually return
1933 * a file handle that can be converted into newvp without having
1934 * to do an extra lookup rpc.
1935 */
1936 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred);
1937 if (v3) {
1938 if (error == 0)
1939 nfsm_mtofh(dvp, newvp, v3, gotvp);
1940 nfsm_wcc_data(dvp, wccflag);
1941 }
1942
1943 /*
1944 * out code jumps -> here, mrep is also freed.
1945 */
1946
1947 m_freem(mrep);
1948nfsmout:
1949
1950 /*
1951 * If we get an EEXIST error, silently convert it to no-error
1952 * in case of an NFS retry.
1953 */
1954 if (error == EEXIST)
1955 error = 0;
1956
1957 /*
1958 * If we do not have (or no longer have) an error, and we could
1959 * not extract the newvp from the response due to the request being
1960 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1961 * to obtain a newvp to return.
1962 */
1963 if (error == 0 && newvp == NULL) {
1964 struct nfsnode *np = NULL;
1965
1966 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1967 cnp->cn_cred, cnp->cn_td, &np);
1968 if (!error)
1969 newvp = NFSTOV(np);
1970 }
1971 if (error) {
1972 if (newvp)
1973 vput(newvp);
1974 } else {
1975 *ap->a_vpp = newvp;
1976 }
1977 VTONFS(dvp)->n_flag |= NLMODIFIED;
1978 if (!wccflag)
1979 VTONFS(dvp)->n_attrstamp = 0;
1980 return (error);
1981}
1982
1983/*
1984 * nfs make dir call
1985 *
1986 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
1987 * struct componentname *a_cnp, struct vattr *a_vap)
1988 */
1989static int
1990nfs_mkdir(struct vop_mkdir_args *ap)
1991{
1992 struct vnode *dvp = ap->a_dvp;
1993 struct vattr *vap = ap->a_vap;
1994 struct componentname *cnp = ap->a_cnp;
1995 struct nfsv2_sattr *sp;
1996 u_int32_t *tl;
1997 caddr_t cp;
1998 int32_t t1, t2;
1999 int len;
2000 struct nfsnode *np = (struct nfsnode *)0;
2001 struct vnode *newvp = (struct vnode *)0;
2002 caddr_t bpos, dpos, cp2;
2003 int error = 0, wccflag = NFSV3_WCCRATTR;
2004 int gotvp = 0;
2005 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2006 struct vattr vattr;
2007 int v3 = NFS_ISV3(dvp);
2008
2009 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
2010 return (error);
2011 }
2012 len = cnp->cn_namelen;
2013 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2014 nfsm_reqhead(dvp, NFSPROC_MKDIR,
2015 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2016 nfsm_fhtom(dvp, v3);
2017 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2018 if (v3) {
2019 nfsm_v3attrbuild(vap, FALSE);
2020 } else {
2021 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2022 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2023 sp->sa_uid = nfs_xdrneg1;
2024 sp->sa_gid = nfs_xdrneg1;
2025 sp->sa_size = nfs_xdrneg1;
2026 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2027 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2028 }
2029 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred);
2030 if (!error)
2031 nfsm_mtofh(dvp, newvp, v3, gotvp);
2032 if (v3)
2033 nfsm_wcc_data(dvp, wccflag);
2034 m_freem(mrep);
2035nfsmout:
2036 VTONFS(dvp)->n_flag |= NLMODIFIED;
2037 if (!wccflag)
2038 VTONFS(dvp)->n_attrstamp = 0;
2039 /*
2040 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2041 * if we can succeed in looking up the directory.
2042 */
2043 if (error == EEXIST || (!error && !gotvp)) {
2044 if (newvp) {
2045 vrele(newvp);
2046 newvp = (struct vnode *)0;
2047 }
2048 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2049 cnp->cn_td, &np);
2050 if (!error) {
2051 newvp = NFSTOV(np);
2052 if (newvp->v_type != VDIR)
2053 error = EEXIST;
2054 }
2055 }
2056 if (error) {
2057 if (newvp)
2058 vrele(newvp);
2059 } else
2060 *ap->a_vpp = newvp;
2061 return (error);
2062}
2063
2064/*
2065 * nfs remove directory call
2066 *
2067 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2068 * struct componentname *a_cnp)
2069 */
2070static int
2071nfs_rmdir(struct vop_rmdir_args *ap)
2072{
2073 struct vnode *vp = ap->a_vp;
2074 struct vnode *dvp = ap->a_dvp;
2075 struct componentname *cnp = ap->a_cnp;
2076 u_int32_t *tl;
2077 caddr_t cp;
2078 int32_t t1, t2;
2079 caddr_t bpos, dpos, cp2;
2080 int error = 0, wccflag = NFSV3_WCCRATTR;
2081 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2082 int v3 = NFS_ISV3(dvp);
2083
2084 if (dvp == vp)
2085 return (EINVAL);
2086 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2087 nfsm_reqhead(dvp, NFSPROC_RMDIR,
2088 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2089 nfsm_fhtom(dvp, v3);
2090 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2091 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred);
2092 if (v3)
2093 nfsm_wcc_data(dvp, wccflag);
2094 m_freem(mrep);
2095nfsmout:
2096 VTONFS(dvp)->n_flag |= NLMODIFIED;
2097 if (!wccflag)
2098 VTONFS(dvp)->n_attrstamp = 0;
2099 /*
2100 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2101 */
2102 if (error == ENOENT)
2103 error = 0;
2104 return (error);
2105}
2106
2107/*
2108 * nfs readdir call
2109 *
2110 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2111 */
2112static int
2113nfs_readdir(struct vop_readdir_args *ap)
2114{
2115 struct vnode *vp = ap->a_vp;
2116 struct nfsnode *np = VTONFS(vp);
2117 struct uio *uio = ap->a_uio;
2118 int tresid, error;
2119 struct vattr vattr;
2120
2121 if (vp->v_type != VDIR)
2122 return (EPERM);
2123
2124 /*
2125 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2126 * and then check that is still valid, or if this is an NQNFS mount
2127 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2128 * VOP_GETATTR() does not necessarily go to the wire.
2129 */
2130 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2131 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2132 if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) {
2133 if (NQNFS_CKCACHABLE(vp, ND_READ)) {
2134 nfsstats.direofcache_hits++;
2135 return (0);
2136 }
2137 } else if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 &&
2138 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2139 ) {
2140 nfsstats.direofcache_hits++;
2141 return (0);
2142 }
2143 }
2144
2145 /*
2146 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2147 * own cache coherency checks so we do not have to.
2148 */
2149 tresid = uio->uio_resid;
2150 error = nfs_bioread(vp, uio, 0);
2151
2152 if (!error && uio->uio_resid == tresid)
2153 nfsstats.direofcache_misses++;
2154 return (error);
2155}
2156
2157/*
2158 * Readdir rpc call.
2159 * Called from below the buffer cache by nfs_doio().
2160 */
2161int
2162nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2163{
2164 int len, left;
2165 struct dirent *dp = NULL;
2166 u_int32_t *tl;
2167 caddr_t cp;
2168 int32_t t1, t2;
2169 nfsuint64 *cookiep;
2170 caddr_t bpos, dpos, cp2;
2171 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2172 nfsuint64 cookie;
2173 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2174 struct nfsnode *dnp = VTONFS(vp);
2175 u_quad_t fileno;
2176 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2177 int attrflag;
2178 int v3 = NFS_ISV3(vp);
2179
2180#ifndef DIAGNOSTIC
2181 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2182 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2183 panic("nfs readdirrpc bad uio");
2184#endif
2185
2186 /*
2187 * If there is no cookie, assume directory was stale.
2188 */
2189 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2190 if (cookiep)
2191 cookie = *cookiep;
2192 else
2193 return (NFSERR_BAD_COOKIE);
2194 /*
2195 * Loop around doing readdir rpc's of size nm_readdirsize
2196 * truncated to a multiple of DIRBLKSIZ.
2197 * The stopping criteria is EOF or buffer full.
2198 */
2199 while (more_dirs && bigenough) {
2200 nfsstats.rpccnt[NFSPROC_READDIR]++;
2201 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2202 NFSX_READDIR(v3));
2203 nfsm_fhtom(vp, v3);
2204 if (v3) {
2205 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2206 *tl++ = cookie.nfsuquad[0];
2207 *tl++ = cookie.nfsuquad[1];
2208 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2209 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2210 } else {
2211 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2212 *tl++ = cookie.nfsuquad[0];
2213 }
2214 *tl = txdr_unsigned(nmp->nm_readdirsize);
2215 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2216 if (v3) {
2217 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2218 if (!error) {
2219 nfsm_dissect(tl, u_int32_t *,
2220 2 * NFSX_UNSIGNED);
2221 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2222 dnp->n_cookieverf.nfsuquad[1] = *tl;
2223 } else {
2224 m_freem(mrep);
2225 goto nfsmout;
2226 }
2227 }
2228 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2229 more_dirs = fxdr_unsigned(int, *tl);
2230
2231 /* loop thru the dir entries, doctoring them to 4bsd form */
2232 while (more_dirs && bigenough) {
2233 if (v3) {
2234 nfsm_dissect(tl, u_int32_t *,
2235 3 * NFSX_UNSIGNED);
2236 fileno = fxdr_hyper(tl);
2237 len = fxdr_unsigned(int, *(tl + 2));
2238 } else {
2239 nfsm_dissect(tl, u_int32_t *,
2240 2 * NFSX_UNSIGNED);
2241 fileno = fxdr_unsigned(u_quad_t, *tl++);
2242 len = fxdr_unsigned(int, *tl);
2243 }
2244 if (len <= 0 || len > NFS_MAXNAMLEN) {
2245 error = EBADRPC;
2246 m_freem(mrep);
2247 goto nfsmout;
2248 }
2249 tlen = nfsm_rndup(len);
2250 if (tlen == len)
2251 tlen += 4; /* To ensure null termination */
2252 left = DIRBLKSIZ - blksiz;
2253 if ((tlen + DIRHDSIZ) > left) {
2254 dp->d_reclen += left;
2255 uiop->uio_iov->iov_base += left;
2256 uiop->uio_iov->iov_len -= left;
2257 uiop->uio_offset += left;
2258 uiop->uio_resid -= left;
2259 blksiz = 0;
2260 }
2261 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2262 bigenough = 0;
2263 if (bigenough) {
2264 dp = (struct dirent *)uiop->uio_iov->iov_base;
2265 dp->d_fileno = (int)fileno;
2266 dp->d_namlen = len;
2267 dp->d_reclen = tlen + DIRHDSIZ;
2268 dp->d_type = DT_UNKNOWN;
2269 blksiz += dp->d_reclen;
2270 if (blksiz == DIRBLKSIZ)
2271 blksiz = 0;
2272 uiop->uio_offset += DIRHDSIZ;
2273 uiop->uio_resid -= DIRHDSIZ;
2274 uiop->uio_iov->iov_base += DIRHDSIZ;
2275 uiop->uio_iov->iov_len -= DIRHDSIZ;
2276 nfsm_mtouio(uiop, len);
2277 cp = uiop->uio_iov->iov_base;
2278 tlen -= len;
2279 *cp = '\0'; /* null terminate */
2280 uiop->uio_iov->iov_base += tlen;
2281 uiop->uio_iov->iov_len -= tlen;
2282 uiop->uio_offset += tlen;
2283 uiop->uio_resid -= tlen;
2284 } else
2285 nfsm_adv(nfsm_rndup(len));
2286 if (v3) {
2287 nfsm_dissect(tl, u_int32_t *,
2288 3 * NFSX_UNSIGNED);
2289 } else {
2290 nfsm_dissect(tl, u_int32_t *,
2291 2 * NFSX_UNSIGNED);
2292 }
2293 if (bigenough) {
2294 cookie.nfsuquad[0] = *tl++;
2295 if (v3)
2296 cookie.nfsuquad[1] = *tl++;
2297 } else if (v3)
2298 tl += 2;
2299 else
2300 tl++;
2301 more_dirs = fxdr_unsigned(int, *tl);
2302 }
2303 /*
2304 * If at end of rpc data, get the eof boolean
2305 */
2306 if (!more_dirs) {
2307 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2308 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2309 }
2310 m_freem(mrep);
2311 }
2312 /*
2313 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2314 * by increasing d_reclen for the last record.
2315 */
2316 if (blksiz > 0) {
2317 left = DIRBLKSIZ - blksiz;
2318 dp->d_reclen += left;
2319 uiop->uio_iov->iov_base += left;
2320 uiop->uio_iov->iov_len -= left;
2321 uiop->uio_offset += left;
2322 uiop->uio_resid -= left;
2323 }
2324
2325 /*
2326 * We are now either at the end of the directory or have filled the
2327 * block.
2328 */
2329 if (bigenough)
2330 dnp->n_direofoffset = uiop->uio_offset;
2331 else {
2332 if (uiop->uio_resid > 0)
2333 printf("EEK! readdirrpc resid > 0\n");
2334 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2335 *cookiep = cookie;
2336 }
2337nfsmout:
2338 return (error);
2339}
2340
2341/*
2342 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2343 */
2344int
2345nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2346{
2347 int len, left;
2348 struct dirent *dp;
2349 u_int32_t *tl;
2350 caddr_t cp;
2351 int32_t t1, t2;
2352 struct vnode *newvp;
2353 nfsuint64 *cookiep;
2354 caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2355 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2356 nfsuint64 cookie;
2357 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2358 struct nfsnode *dnp = VTONFS(vp), *np;
2359 nfsfh_t *fhp;
2360 u_quad_t fileno;
2361 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2362 int attrflag, fhsize;
2363 struct namecache *ncp;
2364 struct namecache *dncp;
2365 struct nlcomponent nlc;
2366
2367#ifndef nolint
2368 dp = (struct dirent *)0;
2369#endif
2370#ifndef DIAGNOSTIC
2371 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2372 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2373 panic("nfs readdirplusrpc bad uio");
2374#endif
2375 /*
2376 * Obtain the namecache record for the directory so we have something
2377 * to use as a basis for creating the entries. This function will
2378 * return a held (but not locked) ncp. The ncp may be disconnected
2379 * from the tree and cannot be used for upward traversals, and the
2380 * ncp may be unnamed. Note that other unrelated operations may
2381 * cause the ncp to be named at any time.
2382 */
2383 dncp = cache_fromdvp(vp, NULL, 0);
2384 bzero(&nlc, sizeof(nlc));
2385 newvp = NULLVP;
2386
2387 /*
2388 * If there is no cookie, assume directory was stale.
2389 */
2390 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2391 if (cookiep)
2392 cookie = *cookiep;
2393 else
2394 return (NFSERR_BAD_COOKIE);
2395 /*
2396 * Loop around doing readdir rpc's of size nm_readdirsize
2397 * truncated to a multiple of DIRBLKSIZ.
2398 * The stopping criteria is EOF or buffer full.
2399 */
2400 while (more_dirs && bigenough) {
2401 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2402 nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2403 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2404 nfsm_fhtom(vp, 1);
2405 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2406 *tl++ = cookie.nfsuquad[0];
2407 *tl++ = cookie.nfsuquad[1];
2408 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2409 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2410 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2411 *tl = txdr_unsigned(nmp->nm_rsize);
2412 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2413 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2414 if (error) {
2415 m_freem(mrep);
2416 goto nfsmout;
2417 }
2418 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2419 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2420 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2421 more_dirs = fxdr_unsigned(int, *tl);
2422
2423 /* loop thru the dir entries, doctoring them to 4bsd form */
2424 while (more_dirs && bigenough) {
2425 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2426 fileno = fxdr_hyper(tl);
2427 len = fxdr_unsigned(int, *(tl + 2));
2428 if (len <= 0 || len > NFS_MAXNAMLEN) {
2429 error = EBADRPC;
2430 m_freem(mrep);
2431 goto nfsmout;
2432 }
2433 tlen = nfsm_rndup(len);
2434 if (tlen == len)
2435 tlen += 4; /* To ensure null termination*/
2436 left = DIRBLKSIZ - blksiz;
2437 if ((tlen + DIRHDSIZ) > left) {
2438 dp->d_reclen += left;
2439 uiop->uio_iov->iov_base += left;
2440 uiop->uio_iov->iov_len -= left;
2441 uiop->uio_offset += left;
2442 uiop->uio_resid -= left;
2443 blksiz = 0;
2444 }
2445 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2446 bigenough = 0;
2447 if (bigenough) {
2448 dp = (struct dirent *)uiop->uio_iov->iov_base;
2449 dp->d_fileno = (int)fileno;
2450 dp->d_namlen = len;
2451 dp->d_reclen = tlen + DIRHDSIZ;
2452 dp->d_type = DT_UNKNOWN;
2453 blksiz += dp->d_reclen;
2454 if (blksiz == DIRBLKSIZ)
2455 blksiz = 0;
2456 uiop->uio_offset += DIRHDSIZ;
2457 uiop->uio_resid -= DIRHDSIZ;
2458 uiop->uio_iov->iov_base += DIRHDSIZ;
2459 uiop->uio_iov->iov_len -= DIRHDSIZ;
2460 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2461 nlc.nlc_namelen = len;
2462 nfsm_mtouio(uiop, len);
2463 cp = uiop->uio_iov->iov_base;
2464 tlen -= len;
2465 *cp = '\0';
2466 uiop->uio_iov->iov_base += tlen;
2467 uiop->uio_iov->iov_len -= tlen;
2468 uiop->uio_offset += tlen;
2469 uiop->uio_resid -= tlen;
2470 } else
2471 nfsm_adv(nfsm_rndup(len));
2472 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2473 if (bigenough) {
2474 cookie.nfsuquad[0] = *tl++;
2475 cookie.nfsuquad[1] = *tl++;
2476 } else
2477 tl += 2;
2478
2479 /*
2480 * Since the attributes are before the file handle
2481 * (sigh), we must skip over the attributes and then
2482 * come back and get them.
2483 */
2484 attrflag = fxdr_unsigned(int, *tl);
2485 if (attrflag) {
2486 dpossav1 = dpos;
2487 mdsav1 = md;
2488 nfsm_adv(NFSX_V3FATTR);
2489 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2490 doit = fxdr_unsigned(int, *tl);
2491 if (doit) {
2492 nfsm_getfh(fhp, fhsize, 1);
2493 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2494 vref(vp);
2495 newvp = vp;
2496 np = dnp;
2497 } else {
2498 error = nfs_nget(vp->v_mount, fhp,
2499 fhsize, &np);
2500 if (error)
2501 doit = 0;
2502 else
2503 newvp = NFSTOV(np);
2504 }
2505 }
2506 if (doit && bigenough) {
2507 dpossav2 = dpos;
2508 dpos = dpossav1;
2509 mdsav2 = md;
2510 md = mdsav1;
2511 nfsm_loadattr(newvp, (struct vattr *)0);
2512 dpos = dpossav2;
2513 md = mdsav2;
2514 dp->d_type =
2515 IFTODT(VTTOIF(np->n_vattr.va_type));
2516 if (dncp) {
2517 printf("NFS/READDIRPLUS, ENTER %*.*s\n",
2518 nlc.nlc_namelen, nlc.nlc_namelen,
2519 nlc.nlc_nameptr);
2520 ncp = cache_nlookup(dncp, &nlc);
2521 cache_setunresolved(ncp);
2522 cache_setvp(ncp, newvp);
2523 cache_put(ncp);
2524 } else {
2525 printf("NFS/READDIRPLUS, UNABLE TO ENTER"
2526 " %*.*s\n",
2527 nlc.nlc_namelen, nlc.nlc_namelen,
2528 nlc.nlc_nameptr);
2529 }
2530 }
2531 } else {
2532 /* Just skip over the file handle */
2533 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2534 i = fxdr_unsigned(int, *tl);
2535 nfsm_adv(nfsm_rndup(i));
2536 }
2537 if (newvp != NULLVP) {
2538 if (newvp == vp)
2539 vrele(newvp);
2540 else
2541 vput(newvp);
2542 newvp = NULLVP;
2543 }
2544 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2545 more_dirs = fxdr_unsigned(int, *tl);
2546 }
2547 /*
2548 * If at end of rpc data, get the eof boolean
2549 */
2550 if (!more_dirs) {
2551 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2552 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2553 }
2554 m_freem(mrep);
2555 }
2556 /*
2557 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2558 * by increasing d_reclen for the last record.
2559 */
2560 if (blksiz > 0) {
2561 left = DIRBLKSIZ - blksiz;
2562 dp->d_reclen += left;
2563 uiop->uio_iov->iov_base += left;
2564 uiop->uio_iov->iov_len -= left;
2565 uiop->uio_offset += left;
2566 uiop->uio_resid -= left;
2567 }
2568
2569 /*
2570 * We are now either at the end of the directory or have filled the
2571 * block.
2572 */
2573 if (bigenough)
2574 dnp->n_direofoffset = uiop->uio_offset;
2575 else {
2576 if (uiop->uio_resid > 0)
2577 printf("EEK! readdirplusrpc resid > 0\n");
2578 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2579 *cookiep = cookie;
2580 }
2581nfsmout:
2582 if (newvp != NULLVP) {
2583 if (newvp == vp)
2584 vrele(newvp);
2585 else
2586 vput(newvp);
2587 newvp = NULLVP;
2588 }
2589 if (dncp)
2590 cache_drop(dncp);
2591 return (error);
2592}
2593
2594/*
2595 * Silly rename. To make the NFS filesystem that is stateless look a little
2596 * more like the "ufs" a remove of an active vnode is translated to a rename
2597 * to a funny looking filename that is removed by nfs_inactive on the
2598 * nfsnode. There is the potential for another process on a different client
2599 * to create the same funny name between the nfs_lookitup() fails and the
2600 * nfs_rename() completes, but...
2601 */
2602static int
2603nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2604{
2605 struct sillyrename *sp;
2606 struct nfsnode *np;
2607 int error;
2608
2609 /*
2610 * We previously purged dvp instead of vp. I don't know why, it
2611 * completely destroys performance. We can't do it anyway with the
2612 * new VFS API since we would be breaking the namecache topology.
2613 */
2614 cache_purge(vp); /* XXX */
2615 np = VTONFS(vp);
2616#ifndef DIAGNOSTIC
2617 if (vp->v_type == VDIR)
2618 panic("nfs: sillyrename dir");
2619#endif
2620 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2621 M_NFSREQ, M_WAITOK);
2622 sp->s_cred = crdup(cnp->cn_cred);
2623 sp->s_dvp = dvp;
2624 vref(dvp);
2625
2626 /* Fudge together a funny name */
2627 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td);
2628
2629 /* Try lookitups until we get one that isn't there */
2630 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2631 cnp->cn_td, (struct nfsnode **)0) == 0) {
2632 sp->s_name[4]++;
2633 if (sp->s_name[4] > 'z') {
2634 error = EINVAL;
2635 goto bad;
2636 }
2637 }
2638 error = nfs_renameit(dvp, cnp, sp);
2639 if (error)
2640 goto bad;
2641 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2642 cnp->cn_td, &np);
2643 np->n_sillyrename = sp;
2644 return (0);
2645bad:
2646 vrele(sp->s_dvp);
2647 crfree(sp->s_cred);
2648 free((caddr_t)sp, M_NFSREQ);
2649 return (error);
2650}
2651
2652/*
2653 * Look up a file name and optionally either update the file handle or
2654 * allocate an nfsnode, depending on the value of npp.
2655 * npp == NULL --> just do the lookup
2656 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2657 * handled too
2658 * *npp != NULL --> update the file handle in the vnode
2659 */
2660static int
2661nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2662 struct thread *td, struct nfsnode **npp)
2663{
2664 u_int32_t *tl;
2665 caddr_t cp;
2666 int32_t t1, t2;
2667 struct vnode *newvp = (struct vnode *)0;
2668 struct nfsnode *np, *dnp = VTONFS(dvp);
2669 caddr_t bpos, dpos, cp2;
2670 int error = 0, fhlen, attrflag;
2671 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2672 nfsfh_t *nfhp;
2673 int v3 = NFS_ISV3(dvp);
2674
2675 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2676 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2677 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2678 nfsm_fhtom(dvp, v3);
2679 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2680 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2681 if (npp && !error) {
2682 nfsm_getfh(nfhp, fhlen, v3);
2683 if (*npp) {
2684 np = *npp;
2685 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2686 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2687 np->n_fhp = &np->n_fh;
2688 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2689 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2690 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2691 np->n_fhsize = fhlen;
2692 newvp = NFSTOV(np);
2693 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2694 vref(dvp);
2695 newvp = dvp;
2696 } else {
2697 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2698 if (error) {
2699 m_freem(mrep);
2700 return (error);
2701 }
2702 newvp = NFSTOV(np);
2703 }
2704 if (v3) {
2705 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
2706 if (!attrflag && *npp == NULL) {
2707 m_freem(mrep);
2708 if (newvp == dvp)
2709 vrele(newvp);
2710 else
2711 vput(newvp);
2712 return (ENOENT);
2713 }
2714 } else
2715 nfsm_loadattr(newvp, (struct vattr *)0);
2716 }
2717 m_freem(mrep);
2718nfsmout:
2719 if (npp && *npp == NULL) {
2720 if (error) {
2721 if (newvp) {
2722 if (newvp == dvp)
2723 vrele(newvp);
2724 else
2725 vput(newvp);
2726 }
2727 } else
2728 *npp = np;
2729 }
2730 return (error);
2731}
2732
2733/*
2734 * Nfs Version 3 commit rpc
2735 */
2736int
2737nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2738{
2739 caddr_t cp;
2740 u_int32_t *tl;
2741 int32_t t1, t2;
2742 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2743 caddr_t bpos, dpos, cp2;
2744 int error = 0, wccflag = NFSV3_WCCRATTR;
2745 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2746
2747 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2748 return (0);
2749 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2750 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2751 nfsm_fhtom(vp, 1);
2752 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2753 txdr_hyper(offset, tl);
2754 tl += 2;
2755 *tl = txdr_unsigned(cnt);
2756 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE));
2757 nfsm_wcc_data(vp, wccflag);
2758 if (!error) {
2759 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2760 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2761 NFSX_V3WRITEVERF)) {
2762 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2763 NFSX_V3WRITEVERF);
2764 error = NFSERR_STALEWRITEVERF;
2765 }
2766 }
2767 m_freem(mrep);
2768nfsmout:
2769 return (error);
2770}
2771
2772/*
2773 * Kludge City..
2774 * - make nfs_bmap() essentially a no-op that does no translation
2775 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2776 * (Maybe I could use the process's page mapping, but I was concerned that
2777 * Kernel Write might not be enabled and also figured copyout() would do
2778 * a lot more work than bcopy() and also it currently happens in the
2779 * context of the swapper process (2).
2780 *
2781 * nfs_bmap(struct vnode *a_vp, daddr_t a_bn, struct vnode **a_vpp,
2782 * daddr_t *a_bnp, int *a_runp, int *a_runb)
2783 */
2784static int
2785nfs_bmap(struct vop_bmap_args *ap)
2786{
2787 struct vnode *vp = ap->a_vp;
2788
2789 if (ap->a_vpp != NULL)
2790 *ap->a_vpp = vp;
2791 if (ap->a_bnp != NULL)
2792 *ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize);
2793 if (ap->a_runp != NULL)
2794 *ap->a_runp = 0;
2795 if (ap->a_runb != NULL)
2796 *ap->a_runb = 0;
2797 return (0);
2798}
2799
2800/*
2801 * Strategy routine.
2802 * For async requests when nfsiod(s) are running, queue the request by
2803 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2804 * request.
2805 */
2806static int
2807nfs_strategy(struct vop_strategy_args *ap)
2808{
2809 struct buf *bp = ap->a_bp;
2810 struct thread *td;
2811 int error = 0;
2812
2813 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2814 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
2815
2816 if (bp->b_flags & B_PHYS)
2817 panic("nfs physio");
2818
2819 if (bp->b_flags & B_ASYNC)
2820 td = NULL;
2821 else
2822 td = curthread; /* XXX */
2823
2824 /*
2825 * If the op is asynchronous and an i/o daemon is waiting
2826 * queue the request, wake it up and wait for completion
2827 * otherwise just do it ourselves.
2828 */
2829 if ((bp->b_flags & B_ASYNC) == 0 ||
2830 nfs_asyncio(bp, td))
2831 error = nfs_doio(bp, td);
2832 return (error);
2833}
2834
2835/*
2836 * Mmap a file
2837 *
2838 * NB Currently unsupported.
2839 *
2840 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred,
2841 * struct thread *a_td)
2842 */
2843/* ARGSUSED */
2844static int
2845nfs_mmap(struct vop_mmap_args *ap)
2846{
2847 return (EINVAL);
2848}
2849
2850/*
2851 * fsync vnode op. Just call nfs_flush() with commit == 1.
2852 *
2853 * nfs_fsync(struct vnodeop_desc *a_desc, struct vnode *a_vp,
2854 * struct ucred * a_cred, int a_waitfor, struct thread *a_td)
2855 */
2856/* ARGSUSED */
2857static int
2858nfs_fsync(struct vop_fsync_args *ap)
2859{
2860 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2861}
2862
2863/*
2864 * Flush all the blocks associated with a vnode.
2865 * Walk through the buffer pool and push any dirty pages
2866 * associated with the vnode.
2867 */
2868int
2869nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
2870{
2871 struct nfsnode *np = VTONFS(vp);
2872 struct buf *bp;
2873 int i;
2874 struct buf *nbp;
2875 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2876 int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2877 int passone = 1;
2878 u_quad_t off, endoff, toff;
2879 struct buf **bvec = NULL;
2880#ifndef NFS_COMMITBVECSIZ
2881#define NFS_COMMITBVECSIZ 20
2882#endif
2883 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2884 int bvecsize = 0, bveccount;
2885
2886 if (nmp->nm_flag & NFSMNT_INT)
2887 slpflag = PCATCH;
2888 if (!commit)
2889 passone = 0;
2890 /*
2891 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2892 * server, but nas not been committed to stable storage on the server
2893 * yet. On the first pass, the byte range is worked out and the commit
2894 * rpc is done. On the second pass, nfs_writebp() is called to do the
2895 * job.
2896 */
2897again:
2898 off = (u_quad_t)-1;
2899 endoff = 0;
2900 bvecpos = 0;
2901 if (NFS_ISV3(vp) && commit) {
2902 s = splbio();
2903 /*
2904 * Count up how many buffers waiting for a commit.
2905 */
2906 bveccount = 0;
2907 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2908 nbp = TAILQ_NEXT(bp, b_vnbufs);
2909 if (BUF_REFCNT(bp) == 0 &&
2910 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2911 == (B_DELWRI | B_NEEDCOMMIT))
2912 bveccount++;
2913 }
2914 /*
2915 * Allocate space to remember the list of bufs to commit. It is
2916 * important to use M_NOWAIT here to avoid a race with nfs_write.
2917 * If we can't get memory (for whatever reason), we will end up
2918 * committing the buffers one-by-one in the loop below.
2919 */
2920 if (bvec != NULL && bvec != bvec_on_stack)
2921 free(bvec, M_TEMP);
2922 if (bveccount > NFS_COMMITBVECSIZ) {
2923 bvec = (struct buf **)
2924 malloc(bveccount * sizeof(struct buf *),
2925 M_TEMP, M_NOWAIT);
2926 if (bvec == NULL) {
2927 bvec = bvec_on_stack;
2928 bvecsize = NFS_COMMITBVECSIZ;
2929 } else
2930 bvecsize = bveccount;
2931 } else {
2932 bvec = bvec_on_stack;
2933 bvecsize = NFS_COMMITBVECSIZ;
2934 }
2935 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
2936 nbp = TAILQ_NEXT(bp, b_vnbufs);
2937 if (bvecpos >= bvecsize)
2938 break;
2939 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2940 (B_DELWRI | B_NEEDCOMMIT) ||
2941 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
2942 continue;
2943 bremfree(bp);
2944 /*
2945 * NOTE: we are not clearing B_DONE here, so we have
2946 * to do it later on in this routine if we intend to
2947 * initiate I/O on the bp.
2948 *
2949 * Note: to avoid loopback deadlocks, we do not
2950 * assign b_runningbufspace.
2951 */
2952 bp->b_flags |= B_WRITEINPROG;
2953 vfs_busy_pages(bp, 1);
2954
2955 /*
2956 * bp is protected by being locked, but nbp is not
2957 * and vfs_busy_pages() may sleep. We have to
2958 * recalculate nbp.
2959 */
2960 nbp = TAILQ_NEXT(bp, b_vnbufs);
2961
2962 /*
2963 * A list of these buffers is kept so that the
2964 * second loop knows which buffers have actually
2965 * been committed. This is necessary, since there
2966 * may be a race between the commit rpc and new
2967 * uncommitted writes on the file.
2968 */
2969 bvec[bvecpos++] = bp;
2970 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2971 bp->b_dirtyoff;
2972 if (toff < off)
2973 off = toff;
2974 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2975 if (toff > endoff)
2976 endoff = toff;
2977 }
2978 splx(s);
2979 }
2980 if (bvecpos > 0) {
2981 /*
2982 * Commit data on the server, as required. Note that
2983 * nfs_commit will use the vnode's cred for the commit.
2984 */
2985 retv = nfs_commit(vp, off, (int)(endoff - off), td);
2986
2987 if (retv == NFSERR_STALEWRITEVERF)
2988 nfs_clearcommit(vp->v_mount);
2989
2990 /*
2991 * Now, either mark the blocks I/O done or mark the
2992 * blocks dirty, depending on whether the commit
2993 * succeeded.
2994 */
2995 for (i = 0; i < bvecpos; i++) {
2996 bp = bvec[i];
2997 bp->b_flags &= ~(B_NEEDCOMMIT | B_WRITEINPROG | B_CLUSTEROK);
2998 if (retv) {
2999 /*
3000 * Error, leave B_DELWRI intact
3001 */
3002 vfs_unbusy_pages(bp);
3003 brelse(bp);
3004 } else {
3005 /*
3006 * Success, remove B_DELWRI ( bundirty() ).
3007 *
3008 * b_dirtyoff/b_dirtyend seem to be NFS
3009 * specific. We should probably move that
3010 * into bundirty(). XXX
3011 */
3012 s = splbio();
3013 vp->v_numoutput++;
3014 bp->b_flags |= B_ASYNC;
3015 bundirty(bp);
3016 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3017 bp->b_dirtyoff = bp->b_dirtyend = 0;
3018 splx(s);
3019 biodone(bp);
3020 }
3021 }
3022 }
3023
3024 /*
3025 * Start/do any write(s) that are required.
3026 */
3027loop:
3028 s = splbio();
3029 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
3030 nbp = TAILQ_NEXT(bp, b_vnbufs);
3031 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
3032 if (waitfor != MNT_WAIT || passone)
3033 continue;
3034 error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL,
3035 "nfsfsync", slpflag, slptimeo);
3036 splx(s);
3037 if (error == 0)
3038 panic("nfs_fsync: inconsistent lock");
3039 if (error == ENOLCK)
3040 goto loop;
3041 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
3042 error = EINTR;
3043 goto done;
3044 }
3045 if (slpflag == PCATCH) {
3046 slpflag = 0;
3047 slptimeo = 2 * hz;
3048 }
3049 goto loop;
3050 }
3051 if ((bp->b_flags & B_DELWRI) == 0)
3052 panic("nfs_fsync: not dirty");
3053 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3054 BUF_UNLOCK(bp);
3055 continue;
3056 }
3057 bremfree(bp);
3058 if (passone || !commit)
3059 bp->b_flags |= B_ASYNC;
3060 else
3061 bp->b_flags |= B_ASYNC | B_WRITEINPROG;
3062 splx(s);
3063 VOP_BWRITE(bp->b_vp, bp);
3064 goto loop;
3065 }
3066 splx(s);
3067 if (passone) {
3068 passone = 0;
3069 goto again;
3070 }
3071 if (waitfor == MNT_WAIT) {
3072 while (vp->v_numoutput) {
3073 vp->v_flag |= VBWAIT;
3074 error = tsleep((caddr_t)&vp->v_numoutput,
3075 slpflag, "nfsfsync", slptimeo);
3076 if (error) {
3077 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
3078 error = EINTR;
3079 goto done;
3080 }
3081 if (slpflag == PCATCH) {
3082 slpflag = 0;
3083 slptimeo = 2 * hz;
3084 }
3085 }
3086 }
3087 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) && commit) {
3088 goto loop;
3089 }
3090 }
3091 if (np->n_flag & NWRITEERR) {
3092 error = np->n_error;
3093 np->n_flag &= ~NWRITEERR;
3094 }
3095done:
3096 if (bvec != NULL && bvec != bvec_on_stack)
3097 free(bvec, M_TEMP);
3098 return (error);
3099}
3100
3101/*
3102 * NFS advisory byte-level locks.
3103 * Currently unsupported.
3104 *
3105 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3106 * int a_flags)
3107 */
3108static int
3109nfs_advlock(struct vop_advlock_args *ap)
3110{
3111 struct nfsnode *np = VTONFS(ap->a_vp);
3112
3113 /*
3114 * The following kludge is to allow diskless support to work
3115 * until a real NFS lockd is implemented. Basically, just pretend
3116 * that this is a local lock.
3117 */
3118 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3119}
3120
3121/*
3122 * Print out the contents of an nfsnode.
3123 *
3124 * nfs_print(struct vnode *a_vp)
3125 */
3126static int
3127nfs_print(struct vop_print_args *ap)
3128{
3129 struct vnode *vp = ap->a_vp;
3130 struct nfsnode *np = VTONFS(vp);
3131
3132 printf("tag VT_NFS, fileid %ld fsid 0x%x",
3133 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3134 if (vp->v_type == VFIFO)
3135 fifo_printinfo(vp);
3136 printf("\n");
3137 return (0);
3138}
3139
3140/*
3141 * Just call nfs_writebp() with the force argument set to 1.
3142 *
3143 * NOTE: B_DONE may or may not be set in a_bp on call.
3144 *
3145 * nfs_bwrite(struct vnode *a_bp)
3146 */
3147static int
3148nfs_bwrite(struct vop_bwrite_args *ap)
3149{
3150 return (nfs_writebp(ap->a_bp, 1, curthread));
3151}
3152
3153/*
3154 * This is a clone of vn_bwrite(), except that B_WRITEINPROG isn't set unless
3155 * the force flag is one and it also handles the B_NEEDCOMMIT flag. We set
3156 * B_CACHE if this is a VMIO buffer.
3157 */
3158int
3159nfs_writebp(struct buf *bp, int force, struct thread *td)
3160{
3161 int s;
3162 int oldflags = bp->b_flags;
3163#if 0
3164 int retv = 1;
3165 off_t off;
3166#endif
3167
3168 if (BUF_REFCNT(bp) == 0)
3169 panic("bwrite: buffer is not locked???");
3170
3171 if (bp->b_flags & B_INVAL) {
3172 brelse(bp);
3173 return(0);
3174 }
3175
3176 bp->b_flags |= B_CACHE;
3177
3178 /*
3179 * Undirty the bp. We will redirty it later if the I/O fails.
3180 */
3181
3182 s = splbio();
3183 bundirty(bp);
3184 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3185
3186 bp->b_vp->v_numoutput++;
3187 splx(s);
3188
3189 /*
3190 * Note: to avoid loopback deadlocks, we do not
3191 * assign b_runningbufspace.
3192 */
3193 vfs_busy_pages(bp, 1);
3194
3195 if (force)
3196 bp->b_flags |= B_WRITEINPROG;
3197 BUF_KERNPROC(bp);
3198 VOP_STRATEGY(bp->b_vp, bp);
3199
3200 if( (oldflags & B_ASYNC) == 0) {
3201 int rtval = biowait(bp);
3202
3203 if (oldflags & B_DELWRI) {
3204 s = splbio();
3205 reassignbuf(bp, bp->b_vp);
3206 splx(s);
3207 }
3208
3209 brelse(bp);
3210 return (rtval);
3211 }
3212
3213 return (0);
3214}
3215
3216/*
3217 * nfs special file access vnode op.
3218 * Essentially just get vattr and then imitate iaccess() since the device is
3219 * local to the client.
3220 *
3221 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
3222 * struct thread *a_td)
3223 */
3224static int
3225nfsspec_access(struct vop_access_args *ap)
3226{
3227 struct vattr *vap;
3228 gid_t *gp;
3229 struct ucred *cred = ap->a_cred;
3230 struct vnode *vp = ap->a_vp;
3231 mode_t mode = ap->a_mode;
3232 struct vattr vattr;
3233 int i;
3234 int error;
3235
3236 /*
3237 * Disallow write attempts on filesystems mounted read-only;
3238 * unless the file is a socket, fifo, or a block or character
3239 * device resident on the filesystem.
3240 */
3241 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3242 switch (vp->v_type) {
3243 case VREG:
3244 case VDIR:
3245 case VLNK:
3246 return (EROFS);
3247 default:
3248 break;
3249 }
3250 }
3251 /*
3252 * If you're the super-user,
3253 * you always get access.
3254 */
3255 if (cred->cr_uid == 0)
3256 return (0);
3257 vap = &vattr;
3258 error = VOP_GETATTR(vp, vap, ap->a_td);
3259 if (error)
3260 return (error);
3261 /*
3262 * Access check is based on only one of owner, group, public.
3263 * If not owner, then check group. If not a member of the
3264 * group, then check public access.
3265 */
3266 if (cred->cr_uid != vap->va_uid) {
3267 mode >>= 3;
3268 gp = cred->cr_groups;
3269 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3270 if (vap->va_gid == *gp)
3271 goto found;
3272 mode >>= 3;
3273found:
3274 ;
3275 }
3276 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3277 return (error);
3278}
3279
3280/*
3281 * Read wrapper for special devices.
3282 *
3283 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3284 * struct ucred *a_cred)
3285 */
3286static int
3287nfsspec_read(struct vop_read_args *ap)
3288{
3289 struct nfsnode *np = VTONFS(ap->a_vp);
3290
3291 /*
3292 * Set access flag.
3293 */
3294 np->n_flag |= NACC;
3295 getnanotime(&np->n_atim);
3296 return (VOCALL(spec_vnode_vops, &ap->a_head));
3297}
3298
3299/*
3300 * Write wrapper for special devices.
3301 *
3302 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3303 * struct ucred *a_cred)
3304 */
3305static int
3306nfsspec_write(struct vop_write_args *ap)
3307{
3308 struct nfsnode *np = VTONFS(ap->a_vp);
3309
3310 /*
3311 * Set update flag.
3312 */
3313 np->n_flag |= NUPD;
3314 getnanotime(&np->n_mtim);
3315 return (VOCALL(spec_vnode_vops, &ap->a_head));
3316}
3317
3318/*
3319 * Close wrapper for special devices.
3320 *
3321 * Update the times on the nfsnode then do device close.
3322 *
3323 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
3324 * struct thread *a_td)
3325 */
3326static int
3327nfsspec_close(struct vop_close_args *ap)
3328{
3329 struct vnode *vp = ap->a_vp;
3330 struct nfsnode *np = VTONFS(vp);
3331 struct vattr vattr;
3332
3333 if (np->n_flag & (NACC | NUPD)) {
3334 np->n_flag |= NCHG;
3335 if (vp->v_usecount == 1 &&
3336 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3337 VATTR_NULL(&vattr);
3338 if (np->n_flag & NACC)
3339 vattr.va_atime = np->n_atim;
3340 if (np->n_flag & NUPD)
3341 vattr.va_mtime = np->n_mtim;
3342 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
3343 }
3344 }
3345 return (VOCALL(spec_vnode_vops, &ap->a_head));
3346}
3347
3348/*
3349 * Read wrapper for fifos.
3350 *
3351 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3352 * struct ucred *a_cred)
3353 */
3354static int
3355nfsfifo_read(struct vop_read_args *ap)
3356{
3357 struct nfsnode *np = VTONFS(ap->a_vp);
3358
3359 /*
3360 * Set access flag.
3361 */
3362 np->n_flag |= NACC;
3363 getnanotime(&np->n_atim);
3364 return (VOCALL(fifo_vnode_vops, &ap->a_head));
3365}
3366
3367/*
3368 * Write wrapper for fifos.
3369 *
3370 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3371 * struct ucred *a_cred)
3372 */
3373static int
3374nfsfifo_write(struct vop_write_args *ap)
3375{
3376 struct nfsnode *np = VTONFS(ap->a_vp);
3377
3378 /*
3379 * Set update flag.
3380 */
3381 np->n_flag |= NUPD;
3382 getnanotime(&np->n_mtim);
3383 return (VOCALL(fifo_vnode_vops, &ap->a_head));
3384}
3385
3386/*
3387 * Close wrapper for fifos.
3388 *
3389 * Update the times on the nfsnode then do fifo close.
3390 *
3391 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td)
3392 */
3393static int
3394nfsfifo_close(struct vop_close_args *ap)
3395{
3396 struct vnode *vp = ap->a_vp;
3397 struct nfsnode *np = VTONFS(vp);
3398 struct vattr vattr;
3399 struct timespec ts;
3400
3401 if (np->n_flag & (NACC | NUPD)) {
3402 getnanotime(&ts);
3403 if (np->n_flag & NACC)
3404 np->n_atim = ts;
3405 if (np->n_flag & NUPD)
3406 np->n_mtim = ts;
3407 np->n_flag |= NCHG;
3408 if (vp->v_usecount == 1 &&
3409 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3410 VATTR_NULL(&vattr);
3411 if (np->n_flag & NACC)
3412 vattr.va_atime = np->n_atim;
3413 if (np->n_flag & NUPD)
3414 vattr.va_mtime = np->n_mtim;
3415 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
3416 }
3417 }
3418 return (VOCALL(fifo_vnode_vops, &ap->a_head));
3419}
3420