Synchronise with NetBSD:
[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 1.41 2005/06/06 15:09:38 drhodus 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#include <sys/thread2.h>
90
91/* Defs */
92#define TRUE 1
93#define FALSE 0
94
95/*
96 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
97 * calls are not in getblk() and brelse() so that they would not be necessary
98 * here.
99 */
100#ifndef B_VMIO
101#define vfs_busy_pages(bp, f)
102#endif
103
104static int nfsspec_read (struct vop_read_args *);
105static int nfsspec_write (struct vop_write_args *);
106static int nfsfifo_read (struct vop_read_args *);
107static int nfsfifo_write (struct vop_write_args *);
108static int nfsspec_close (struct vop_close_args *);
109static int nfsfifo_close (struct vop_close_args *);
110#define nfs_poll vop_nopoll
111static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
112static int nfs_lookup (struct vop_lookup_args *);
113static int nfs_create (struct vop_create_args *);
114static int nfs_mknod (struct vop_mknod_args *);
115static int nfs_open (struct vop_open_args *);
116static int nfs_close (struct vop_close_args *);
117static int nfs_access (struct vop_access_args *);
118static int nfs_getattr (struct vop_getattr_args *);
119static int nfs_setattr (struct vop_setattr_args *);
120static int nfs_read (struct vop_read_args *);
121static int nfs_mmap (struct vop_mmap_args *);
122static int nfs_fsync (struct vop_fsync_args *);
123static int nfs_remove (struct vop_remove_args *);
124static int nfs_link (struct vop_link_args *);
125static int nfs_rename (struct vop_rename_args *);
126static int nfs_mkdir (struct vop_mkdir_args *);
127static int nfs_rmdir (struct vop_rmdir_args *);
128static int nfs_symlink (struct vop_symlink_args *);
129static int nfs_readdir (struct vop_readdir_args *);
130static int nfs_bmap (struct vop_bmap_args *);
131static int nfs_strategy (struct vop_strategy_args *);
132static int nfs_lookitup (struct vnode *, const char *, int,
133 struct ucred *, struct thread *, struct nfsnode **);
134static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
135static int nfsspec_access (struct vop_access_args *);
136static int nfs_readlink (struct vop_readlink_args *);
137static int nfs_print (struct vop_print_args *);
138static int nfs_advlock (struct vop_advlock_args *);
139static int nfs_bwrite (struct vop_bwrite_args *);
140
141static int nfs_nresolve (struct vop_nresolve_args *);
142/*
143 * Global vfs data structures for nfs
144 */
145struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
146 { &vop_default_desc, vop_defaultop },
147 { &vop_access_desc, (vnodeopv_entry_t) nfs_access },
148 { &vop_advlock_desc, (vnodeopv_entry_t) nfs_advlock },
149 { &vop_bmap_desc, (vnodeopv_entry_t) nfs_bmap },
150 { &vop_bwrite_desc, (vnodeopv_entry_t) nfs_bwrite },
151 { &vop_close_desc, (vnodeopv_entry_t) nfs_close },
152 { &vop_create_desc, (vnodeopv_entry_t) nfs_create },
153 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync },
154 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr },
155 { &vop_getpages_desc, (vnodeopv_entry_t) nfs_getpages },
156 { &vop_putpages_desc, (vnodeopv_entry_t) nfs_putpages },
157 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive },
158 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked },
159 { &vop_lease_desc, vop_null },
160 { &vop_link_desc, (vnodeopv_entry_t) nfs_link },
161 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock },
162 { &vop_lookup_desc, (vnodeopv_entry_t) nfs_lookup },
163 { &vop_mkdir_desc, (vnodeopv_entry_t) nfs_mkdir },
164 { &vop_mknod_desc, (vnodeopv_entry_t) nfs_mknod },
165 { &vop_mmap_desc, (vnodeopv_entry_t) nfs_mmap },
166 { &vop_open_desc, (vnodeopv_entry_t) nfs_open },
167 { &vop_poll_desc, (vnodeopv_entry_t) nfs_poll },
168 { &vop_print_desc, (vnodeopv_entry_t) nfs_print },
169 { &vop_read_desc, (vnodeopv_entry_t) nfs_read },
170 { &vop_readdir_desc, (vnodeopv_entry_t) nfs_readdir },
171 { &vop_readlink_desc, (vnodeopv_entry_t) nfs_readlink },
172 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim },
173 { &vop_remove_desc, (vnodeopv_entry_t) nfs_remove },
174 { &vop_rename_desc, (vnodeopv_entry_t) nfs_rename },
175 { &vop_rmdir_desc, (vnodeopv_entry_t) nfs_rmdir },
176 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr },
177 { &vop_strategy_desc, (vnodeopv_entry_t) nfs_strategy },
178 { &vop_symlink_desc, (vnodeopv_entry_t) nfs_symlink },
179 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock },
180 { &vop_write_desc, (vnodeopv_entry_t) nfs_write },
181
182 { &vop_nresolve_desc, (vnodeopv_entry_t) nfs_nresolve },
183 { NULL, NULL }
184};
185
186/*
187 * Special device vnode ops
188 */
189struct vnodeopv_entry_desc nfsv2_specop_entries[] = {
190 { &vop_default_desc, (vnodeopv_entry_t) spec_vnoperate },
191 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access },
192 { &vop_close_desc, (vnodeopv_entry_t) nfsspec_close },
193 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync },
194 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr },
195 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive },
196 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked },
197 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock },
198 { &vop_print_desc, (vnodeopv_entry_t) nfs_print },
199 { &vop_read_desc, (vnodeopv_entry_t) nfsspec_read },
200 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim },
201 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr },
202 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock },
203 { &vop_write_desc, (vnodeopv_entry_t) nfsspec_write },
204 { NULL, NULL }
205};
206
207struct vnodeopv_entry_desc nfsv2_fifoop_entries[] = {
208 { &vop_default_desc, (vnodeopv_entry_t) fifo_vnoperate },
209 { &vop_access_desc, (vnodeopv_entry_t) nfsspec_access },
210 { &vop_close_desc, (vnodeopv_entry_t) nfsfifo_close },
211 { &vop_fsync_desc, (vnodeopv_entry_t) nfs_fsync },
212 { &vop_getattr_desc, (vnodeopv_entry_t) nfs_getattr },
213 { &vop_inactive_desc, (vnodeopv_entry_t) nfs_inactive },
214 { &vop_islocked_desc, (vnodeopv_entry_t) vop_stdislocked },
215 { &vop_lock_desc, (vnodeopv_entry_t) vop_stdlock },
216 { &vop_print_desc, (vnodeopv_entry_t) nfs_print },
217 { &vop_read_desc, (vnodeopv_entry_t) nfsfifo_read },
218 { &vop_reclaim_desc, (vnodeopv_entry_t) nfs_reclaim },
219 { &vop_setattr_desc, (vnodeopv_entry_t) nfs_setattr },
220 { &vop_unlock_desc, (vnodeopv_entry_t) vop_stdunlock },
221 { &vop_write_desc, (vnodeopv_entry_t) nfsfifo_write },
222 { NULL, NULL }
223};
224
225static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
226 struct componentname *cnp,
227 struct vattr *vap);
228static int nfs_removerpc (struct vnode *dvp, const char *name,
229 int namelen,
230 struct ucred *cred, struct thread *td);
231static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
232 int fnamelen, struct vnode *tdvp,
233 const char *tnameptr, int tnamelen,
234 struct ucred *cred, struct thread *td);
235static int nfs_renameit (struct vnode *sdvp,
236 struct componentname *scnp,
237 struct sillyrename *sp);
238
239/*
240 * Global variables
241 */
242extern u_int32_t nfs_true, nfs_false;
243extern u_int32_t nfs_xdrneg1;
244extern struct nfsstats nfsstats;
245extern nfstype nfsv3_type[9];
246struct thread *nfs_iodwant[NFS_MAXASYNCDAEMON];
247struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
248int nfs_numasync = 0;
249#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
250
251SYSCTL_DECL(_vfs_nfs);
252
253static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
254SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
255 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
256
257static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
258SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
259 &nfsneg_cache_timeout, 0, "NFS NEGATIVE ACCESS cache timeout");
260
261static int nfsv3_commit_on_close = 0;
262SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
263 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
264#if 0
265SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
266 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
267
268SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
269 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
270#endif
271
272#define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
273 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
274 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
275static int
276nfs3_access_otw(struct vnode *vp, int wmode,
277 struct thread *td, struct ucred *cred)
278{
279 const int v3 = 1;
280 u_int32_t *tl;
281 int error = 0, attrflag;
282
283 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
284 caddr_t bpos, dpos, cp2;
285 int32_t t1, t2;
286 caddr_t cp;
287 u_int32_t rmode;
288 struct nfsnode *np = VTONFS(vp);
289
290 nfsstats.rpccnt[NFSPROC_ACCESS]++;
291 nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
292 nfsm_fhtom(vp, v3);
293 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
294 *tl = txdr_unsigned(wmode);
295 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
296 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
297 if (!error) {
298 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
299 rmode = fxdr_unsigned(u_int32_t, *tl);
300 np->n_mode = rmode;
301 np->n_modeuid = cred->cr_uid;
302 np->n_modestamp = mycpu->gd_time_seconds;
303 }
304 m_freem(mrep);
305nfsmout:
306 return error;
307}
308
309/*
310 * nfs access vnode op.
311 * For nfs version 2, just return ok. File accesses may fail later.
312 * For nfs version 3, use the access rpc to check accessibility. If file modes
313 * are changed on the server, accesses might still fail later.
314 *
315 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
316 * struct thread *a_td)
317 */
318static int
319nfs_access(struct vop_access_args *ap)
320{
321 struct vnode *vp = ap->a_vp;
322 int error = 0;
323 u_int32_t mode, wmode;
324 int v3 = NFS_ISV3(vp);
325 struct nfsnode *np = VTONFS(vp);
326
327 /*
328 * Disallow write attempts on filesystems mounted read-only;
329 * unless the file is a socket, fifo, or a block or character
330 * device resident on the filesystem.
331 */
332 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
333 switch (vp->v_type) {
334 case VREG:
335 case VDIR:
336 case VLNK:
337 return (EROFS);
338 default:
339 break;
340 }
341 }
342 /*
343 * For nfs v3, check to see if we have done this recently, and if
344 * so return our cached result instead of making an ACCESS call.
345 * If not, do an access rpc, otherwise you are stuck emulating
346 * ufs_access() locally using the vattr. This may not be correct,
347 * since the server may apply other access criteria such as
348 * client uid-->server uid mapping that we do not know about.
349 */
350 if (v3) {
351 if (ap->a_mode & VREAD)
352 mode = NFSV3ACCESS_READ;
353 else
354 mode = 0;
355 if (vp->v_type != VDIR) {
356 if (ap->a_mode & VWRITE)
357 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
358 if (ap->a_mode & VEXEC)
359 mode |= NFSV3ACCESS_EXECUTE;
360 } else {
361 if (ap->a_mode & VWRITE)
362 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
363 NFSV3ACCESS_DELETE);
364 if (ap->a_mode & VEXEC)
365 mode |= NFSV3ACCESS_LOOKUP;
366 }
367 /* XXX safety belt, only make blanket request if caching */
368 if (nfsaccess_cache_timeout > 0) {
369 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
370 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
371 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
372 } else {
373 wmode = mode;
374 }
375
376 /*
377 * Does our cached result allow us to give a definite yes to
378 * this request?
379 */
380 if (np->n_modestamp &&
381 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
382 (ap->a_cred->cr_uid == np->n_modeuid) &&
383 ((np->n_mode & mode) == mode)) {
384 nfsstats.accesscache_hits++;
385 } else {
386 /*
387 * Either a no, or a don't know. Go to the wire.
388 */
389 nfsstats.accesscache_misses++;
390 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
391 if (!error) {
392 if ((np->n_mode & mode) != mode) {
393 error = EACCES;
394 }
395 }
396 }
397 } else {
398 if ((error = nfsspec_access(ap)) != 0)
399 return (error);
400
401 /*
402 * Attempt to prevent a mapped root from accessing a file
403 * which it shouldn't. We try to read a byte from the file
404 * if the user is root and the file is not zero length.
405 * After calling nfsspec_access, we should have the correct
406 * file size cached.
407 */
408 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
409 && VTONFS(vp)->n_size > 0) {
410 struct iovec aiov;
411 struct uio auio;
412 char buf[1];
413
414 aiov.iov_base = buf;
415 aiov.iov_len = 1;
416 auio.uio_iov = &aiov;
417 auio.uio_iovcnt = 1;
418 auio.uio_offset = 0;
419 auio.uio_resid = 1;
420 auio.uio_segflg = UIO_SYSSPACE;
421 auio.uio_rw = UIO_READ;
422 auio.uio_td = ap->a_td;
423
424 if (vp->v_type == VREG) {
425 error = nfs_readrpc(vp, &auio);
426 } else if (vp->v_type == VDIR) {
427 char* bp;
428 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
429 aiov.iov_base = bp;
430 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
431 error = nfs_readdirrpc(vp, &auio);
432 free(bp, M_TEMP);
433 } else if (vp->v_type == VLNK) {
434 error = nfs_readlinkrpc(vp, &auio);
435 } else {
436 error = EACCES;
437 }
438 }
439 }
440 /*
441 * [re]record creds for reading and/or writing if access
442 * was granted. Assume the NFS server will grant read access
443 * for execute requests.
444 */
445 if (error == 0) {
446 if ((ap->a_mode & (VREAD|VEXEC)) && ap->a_cred != np->n_rucred) {
447 crhold(ap->a_cred);
448 if (np->n_rucred)
449 crfree(np->n_rucred);
450 np->n_rucred = ap->a_cred;
451 }
452 if ((ap->a_mode & VWRITE) && ap->a_cred != np->n_wucred) {
453 crhold(ap->a_cred);
454 if (np->n_wucred)
455 crfree(np->n_wucred);
456 np->n_wucred = ap->a_cred;
457 }
458 }
459 return(error);
460}
461
462/*
463 * nfs open vnode op
464 * Check to see if the type is ok
465 * and that deletion is not in progress.
466 * For paged in text files, you will need to flush the page cache
467 * if consistency is lost.
468 *
469 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
470 * struct thread *a_td)
471 */
472/* ARGSUSED */
473static int
474nfs_open(struct vop_open_args *ap)
475{
476 struct vnode *vp = ap->a_vp;
477 struct nfsnode *np = VTONFS(vp);
478 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
479 struct vattr vattr;
480 int error;
481
482 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
483#ifdef DIAGNOSTIC
484 printf("open eacces vtyp=%d\n",vp->v_type);
485#endif
486 return (EOPNOTSUPP);
487 }
488
489 /*
490 * Clear the attribute cache only if opening with write access. It
491 * is unclear if we should do this at all here, but we certainly
492 * should not clear the cache unconditionally simply because a file
493 * is being opened.
494 */
495 if (ap->a_mode & FWRITE)
496 np->n_attrstamp = 0;
497
498 if (nmp->nm_flag & NFSMNT_NQNFS) {
499 /*
500 * If NQNFS is active, get a valid lease
501 */
502 if (NQNFS_CKINVALID(vp, np, ND_READ)) {
503 do {
504 error = nqnfs_getlease(vp, ND_READ, ap->a_td);
505 } while (error == NQNFS_EXPIRED);
506 if (error)
507 return (error);
508 if (np->n_lrev != np->n_brev ||
509 (np->n_flag & NQNFSNONCACHE)) {
510 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1))
511 == EINTR) {
512 return (error);
513 }
514 np->n_brev = np->n_lrev;
515 }
516 }
517 } else {
518 /*
519 * For normal NFS, reconcile changes made locally verses
520 * changes made remotely. Note that VOP_GETATTR only goes
521 * to the wire if the cached attribute has timed out or been
522 * cleared.
523 *
524 * If local modifications have been made clear the attribute
525 * cache to force an attribute and modified time check. If
526 * GETATTR detects that the file has been changed by someone
527 * other then us it will set NRMODIFIED.
528 *
529 * If we are opening a directory and local changes have been
530 * made we have to invalidate the cache in order to ensure
531 * that we get the most up-to-date information from the
532 * server. XXX
533 */
534 if (np->n_flag & NLMODIFIED) {
535 np->n_attrstamp = 0;
536 if (vp->v_type == VDIR) {
537 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
538 if (error == EINTR)
539 return (error);
540 nfs_invaldir(vp);
541 }
542 }
543 error = VOP_GETATTR(vp, &vattr, ap->a_td);
544 if (error)
545 return (error);
546 if (np->n_flag & NRMODIFIED) {
547 if (vp->v_type == VDIR)
548 nfs_invaldir(vp);
549 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
550 if (error == EINTR)
551 return (error);
552 np->n_flag &= ~NRMODIFIED;
553 }
554 }
555
556 return (0);
557}
558
559/*
560 * nfs close vnode op
561 * What an NFS client should do upon close after writing is a debatable issue.
562 * Most NFS clients push delayed writes to the server upon close, basically for
563 * two reasons:
564 * 1 - So that any write errors may be reported back to the client process
565 * doing the close system call. By far the two most likely errors are
566 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
567 * 2 - To put a worst case upper bound on cache inconsistency between
568 * multiple clients for the file.
569 * There is also a consistency problem for Version 2 of the protocol w.r.t.
570 * not being able to tell if other clients are writing a file concurrently,
571 * since there is no way of knowing if the changed modify time in the reply
572 * is only due to the write for this client.
573 * (NFS Version 3 provides weak cache consistency data in the reply that
574 * should be sufficient to detect and handle this case.)
575 *
576 * The current code does the following:
577 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
578 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
579 * or commit them (this satisfies 1 and 2 except for the
580 * case where the server crashes after this close but
581 * before the commit RPC, which is felt to be "good
582 * enough". Changing the last argument to nfs_flush() to
583 * a 1 would force a commit operation, if it is felt a
584 * commit is necessary now.
585 * for NQNFS - do nothing now, since 2 is dealt with via leases and
586 * 1 should be dealt with via an fsync() system call for
587 * cases where write errors are important.
588 *
589 * nfs_close(struct vnodeop_desc *a_desc, struct vnode *a_vp, int a_fflag,
590 * struct ucred *a_cred, struct thread *a_td)
591 */
592/* ARGSUSED */
593static int
594nfs_close(struct vop_close_args *ap)
595{
596 struct vnode *vp = ap->a_vp;
597 struct nfsnode *np = VTONFS(vp);
598 int error = 0;
599
600 if (vp->v_type == VREG) {
601 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
602 (np->n_flag & NLMODIFIED)) {
603 if (NFS_ISV3(vp)) {
604 /*
605 * Under NFSv3 we have dirty buffers to dispose of. We
606 * must flush them to the NFS server. We have the option
607 * of waiting all the way through the commit rpc or just
608 * waiting for the initial write. The default is to only
609 * wait through the initial write so the data is in the
610 * server's cache, which is roughly similar to the state
611 * a standard disk subsystem leaves the file in on close().
612 *
613 * We cannot clear the NLMODIFIED bit in np->n_flag due to
614 * potential races with other processes, and certainly
615 * cannot clear it if we don't commit.
616 */
617 int cm = nfsv3_commit_on_close ? 1 : 0;
618 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
619 /* np->n_flag &= ~NLMODIFIED; */
620 } else {
621 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
622 }
623 np->n_attrstamp = 0;
624 }
625 if (np->n_flag & NWRITEERR) {
626 np->n_flag &= ~NWRITEERR;
627 error = np->n_error;
628 }
629 }
630 return (error);
631}
632
633/*
634 * nfs getattr call from vfs.
635 *
636 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred,
637 * struct thread *a_td)
638 */
639static int
640nfs_getattr(struct vop_getattr_args *ap)
641{
642 struct vnode *vp = ap->a_vp;
643 struct nfsnode *np = VTONFS(vp);
644 caddr_t cp;
645 u_int32_t *tl;
646 int32_t t1, t2;
647 caddr_t bpos, dpos;
648 int error = 0;
649 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
650 int v3 = NFS_ISV3(vp);
651
652 /*
653 * Update local times for special files.
654 */
655 if (np->n_flag & (NACC | NUPD))
656 np->n_flag |= NCHG;
657 /*
658 * First look in the cache.
659 */
660 if (nfs_getattrcache(vp, ap->a_vap) == 0)
661 return (0);
662
663 if (v3 && nfsaccess_cache_timeout > 0) {
664 nfsstats.accesscache_misses++;
665 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, nfs_vpcred(vp, ND_CHECK));
666 if (nfs_getattrcache(vp, ap->a_vap) == 0)
667 return (0);
668 }
669
670 nfsstats.rpccnt[NFSPROC_GETATTR]++;
671 nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
672 nfsm_fhtom(vp, v3);
673 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, nfs_vpcred(vp, ND_CHECK));
674 if (!error) {
675 nfsm_loadattr(vp, ap->a_vap);
676 }
677 m_freem(mrep);
678nfsmout:
679 return (error);
680}
681
682/*
683 * nfs setattr call.
684 *
685 * nfs_setattr(struct vnodeop_desc *a_desc, struct vnode *a_vp,
686 * struct vattr *a_vap, struct ucred *a_cred,
687 * struct thread *a_td)
688 */
689static int
690nfs_setattr(struct vop_setattr_args *ap)
691{
692 struct vnode *vp = ap->a_vp;
693 struct nfsnode *np = VTONFS(vp);
694 struct vattr *vap = ap->a_vap;
695 int error = 0;
696 u_quad_t tsize;
697
698#ifndef nolint
699 tsize = (u_quad_t)0;
700#endif
701
702 /*
703 * Setting of flags is not supported.
704 */
705 if (vap->va_flags != VNOVAL)
706 return (EOPNOTSUPP);
707
708 /*
709 * Disallow write attempts if the filesystem is mounted read-only.
710 */
711 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
712 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
713 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
714 (vp->v_mount->mnt_flag & MNT_RDONLY))
715 return (EROFS);
716 if (vap->va_size != VNOVAL) {
717 switch (vp->v_type) {
718 case VDIR:
719 return (EISDIR);
720 case VCHR:
721 case VBLK:
722 case VSOCK:
723 case VFIFO:
724 if (vap->va_mtime.tv_sec == VNOVAL &&
725 vap->va_atime.tv_sec == VNOVAL &&
726 vap->va_mode == (mode_t)VNOVAL &&
727 vap->va_uid == (uid_t)VNOVAL &&
728 vap->va_gid == (gid_t)VNOVAL)
729 return (0);
730 vap->va_size = VNOVAL;
731 break;
732 default:
733 /*
734 * Disallow write attempts if the filesystem is
735 * mounted read-only.
736 */
737 if (vp->v_mount->mnt_flag & MNT_RDONLY)
738 return (EROFS);
739
740 /*
741 * This is nasty. The RPCs we send to flush pending
742 * data often return attribute information which is
743 * cached via a callback to nfs_loadattrcache(), which
744 * has the effect of changing our notion of the file
745 * size. Due to flushed appends and other operations
746 * the file size can be set to virtually anything,
747 * including values that do not match either the old
748 * or intended file size.
749 *
750 * When this condition is detected we must loop to
751 * try the operation again. Hopefully no more
752 * flushing is required on the loop so it works the
753 * second time around. THIS CASE ALMOST ALWAYS
754 * HAPPENS!
755 */
756 tsize = np->n_size;
757again:
758 error = nfs_meta_setsize(vp, ap->a_td, vap->va_size);
759
760 if (np->n_flag & NLMODIFIED) {
761 if (vap->va_size == 0)
762 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
763 else
764 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
765 }
766 /*
767 * note: this loop case almost always happens at
768 * least once per truncation.
769 */
770 if (error == 0 && np->n_size != vap->va_size)
771 goto again;
772 np->n_vattr.va_size = vap->va_size;
773 break;
774 }
775 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
776 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NLMODIFIED) &&
777 vp->v_type == VREG &&
778 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR
779 ) {
780 return (error);
781 }
782 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
783
784 /*
785 * Sanity check if a truncation was issued. This should only occur
786 * if multiple processes are racing on the same file.
787 */
788 if (error == 0 && vap->va_size != VNOVAL &&
789 np->n_size != vap->va_size) {
790 printf("NFS ftruncate: server disagrees on the file size: %lld/%lld/%lld\n", tsize, vap->va_size, np->n_size);
791 goto again;
792 }
793 if (error && vap->va_size != VNOVAL) {
794 np->n_size = np->n_vattr.va_size = tsize;
795 vnode_pager_setsize(vp, np->n_size);
796 }
797 return (error);
798}
799
800/*
801 * Do an nfs setattr rpc.
802 */
803static int
804nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
805 struct ucred *cred, struct thread *td)
806{
807 struct nfsv2_sattr *sp;
808 struct nfsnode *np = VTONFS(vp);
809 caddr_t cp;
810 int32_t t1, t2;
811 caddr_t bpos, dpos, cp2;
812 u_int32_t *tl;
813 int error = 0, wccflag = NFSV3_WCCRATTR;
814 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
815 int v3 = NFS_ISV3(vp);
816
817 nfsstats.rpccnt[NFSPROC_SETATTR]++;
818 nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
819 nfsm_fhtom(vp, v3);
820 if (v3) {
821 nfsm_v3attrbuild(vap, TRUE);
822 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
823 *tl = nfs_false;
824 } else {
825 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
826 if (vap->va_mode == (mode_t)VNOVAL)
827 sp->sa_mode = nfs_xdrneg1;
828 else
829 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
830 if (vap->va_uid == (uid_t)VNOVAL)
831 sp->sa_uid = nfs_xdrneg1;
832 else
833 sp->sa_uid = txdr_unsigned(vap->va_uid);
834 if (vap->va_gid == (gid_t)VNOVAL)
835 sp->sa_gid = nfs_xdrneg1;
836 else
837 sp->sa_gid = txdr_unsigned(vap->va_gid);
838 sp->sa_size = txdr_unsigned(vap->va_size);
839 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
840 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
841 }
842 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
843 if (v3) {
844 np->n_modestamp = 0;
845 nfsm_wcc_data(vp, wccflag);
846 } else
847 nfsm_loadattr(vp, (struct vattr *)0);
848 m_freem(mrep);
849nfsmout:
850 return (error);
851}
852
853/*
854 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
855 * nfs_lookup() until all remaining new api calls are implemented.
856 *
857 * Resolve a namecache entry. This function is passed a locked ncp and
858 * must call cache_setvp() on it as appropriate to resolve the entry.
859 */
860static int
861nfs_nresolve(struct vop_nresolve_args *ap)
862{
863 struct thread *td = curthread;
864 struct namecache *ncp;
865 struct ucred *cred;
866 struct nfsnode *np;
867 struct vnode *dvp;
868 struct vnode *nvp;
869 nfsfh_t *fhp;
870 int attrflag;
871 int fhsize;
872 int error;
873 int len;
874 int v3;
875 /******NFSM MACROS********/
876 struct mbuf *mb, *mrep, *mreq, *mb2, *md;
877 caddr_t bpos, dpos, cp, cp2;
878 u_int32_t *tl;
879 int32_t t1, t2;
880
881 cred = ap->a_cred;
882 ncp = ap->a_ncp;
883
884 KKASSERT(ncp->nc_parent && ncp->nc_parent->nc_vp);
885 dvp = ncp->nc_parent->nc_vp;
886 if ((error = vget(dvp, LK_SHARED, td)) != 0)
887 return (error);
888
889 nvp = NULL;
890 v3 = NFS_ISV3(dvp);
891 nfsstats.lookupcache_misses++;
892 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
893 len = ncp->nc_nlen;
894 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
895 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
896 nfsm_fhtom(dvp, v3);
897 nfsm_strtom(ncp->nc_name, len, NFS_MAXNAMLEN);
898 nfsm_request(dvp, NFSPROC_LOOKUP, td, ap->a_cred);
899 if (error) {
900 /*
901 * Cache negatve lookups to reduce NFS traffic, but use
902 * a fast timeout. Otherwise use a timeout of 1 tick.
903 * XXX we should add a namecache flag for no-caching
904 * to uncache the negative hit as soon as possible, but
905 * we cannot simply destroy the entry because it is used
906 * as a placeholder by the caller.
907 */
908 if (error == ENOENT) {
909 int nticks;
910
911 if (nfsneg_cache_timeout)
912 nticks = nfsneg_cache_timeout * hz;
913 else
914 nticks = 1;
915 cache_setvp(ncp, NULL);
916 cache_settimeout(ncp, nticks);
917 }
918 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
919 m_freem(mrep);
920 goto nfsmout;
921 }
922
923 /*
924 * Success, get the file handle, do various checks, and load
925 * post-operation data from the reply packet. Theoretically
926 * we should never be looking up "." so, theoretically, we
927 * should never get the same file handle as our directory. But
928 * we check anyway. XXX
929 *
930 * Note that no timeout is set for the positive cache hit. We
931 * assume, theoretically, that ESTALE returns will be dealt with
932 * properly to handle NFS races and in anycase we cannot depend
933 * on a timeout to deal with NFS open/create/excl issues so instead
934 * of a bad hack here the rest of the NFS client code needs to do
935 * the right thing.
936 */
937 nfsm_getfh(fhp, fhsize, v3);
938
939 np = VTONFS(dvp);
940 if (NFS_CMPFH(np, fhp, fhsize)) {
941 vref(dvp);
942 nvp = dvp;
943 } else {
944 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
945 if (error) {
946 m_freem(mrep);
947 vput(dvp);
948 return (error);
949 }
950 nvp = NFSTOV(np);
951 }
952 if (v3) {
953 nfsm_postop_attr(nvp, attrflag, NFS_LATTR_NOSHRINK);
954 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
955 } else {
956 nfsm_loadattr(nvp, NULL);
957 }
958 cache_setvp(ncp, nvp);
959 m_freem(mrep);
960nfsmout:
961 vput(dvp);
962 if (nvp) {
963 if (nvp == dvp)
964 vrele(nvp);
965 else
966 vput(nvp);
967 }
968 return (error);
969}
970
971/*
972 * 'cached' nfs directory lookup
973 *
974 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
975 *
976 * nfs_lookup(struct vnodeop_desc *a_desc, struct vnode *a_dvp,
977 * struct vnode **a_vpp, struct componentname *a_cnp)
978 */
979static int
980nfs_lookup(struct vop_lookup_args *ap)
981{
982 struct componentname *cnp = ap->a_cnp;
983 struct vnode *dvp = ap->a_dvp;
984 struct vnode **vpp = ap->a_vpp;
985 int flags = cnp->cn_flags;
986 struct vnode *newvp;
987 u_int32_t *tl;
988 caddr_t cp;
989 int32_t t1, t2;
990 struct nfsmount *nmp;
991 caddr_t bpos, dpos, cp2;
992 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
993 long len;
994 nfsfh_t *fhp;
995 struct nfsnode *np;
996 int lockparent, wantparent, error = 0, attrflag, fhsize;
997 int v3 = NFS_ISV3(dvp);
998 struct thread *td = cnp->cn_td;
999
1000 /*
1001 * Read-only mount check and directory check.
1002 */
1003 *vpp = NULLVP;
1004 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1005 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1006 return (EROFS);
1007
1008 if (dvp->v_type != VDIR)
1009 return (ENOTDIR);
1010
1011 /*
1012 * Look it up in the cache. Note that ENOENT is only returned if we
1013 * previously entered a negative hit (see later on). The additional
1014 * nfsneg_cache_timeout check causes previously cached results to
1015 * be instantly ignored if the negative caching is turned off.
1016 */
1017 lockparent = flags & CNP_LOCKPARENT;
1018 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1019 nmp = VFSTONFS(dvp->v_mount);
1020 np = VTONFS(dvp);
1021
1022 /*
1023 * Go to the wire.
1024 */
1025 error = 0;
1026 newvp = NULLVP;
1027 nfsstats.lookupcache_misses++;
1028 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1029 len = cnp->cn_namelen;
1030 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
1031 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1032 nfsm_fhtom(dvp, v3);
1033 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1034 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_td, cnp->cn_cred);
1035 if (error) {
1036 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1037 m_freem(mrep);
1038 goto nfsmout;
1039 }
1040 nfsm_getfh(fhp, fhsize, v3);
1041
1042 /*
1043 * Handle RENAME case...
1044 */
1045 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1046 if (NFS_CMPFH(np, fhp, fhsize)) {
1047 m_freem(mrep);
1048 return (EISDIR);
1049 }
1050 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1051 if (error) {
1052 m_freem(mrep);
1053 return (error);
1054 }
1055 newvp = NFSTOV(np);
1056 if (v3) {
1057 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1058 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1059 } else
1060 nfsm_loadattr(newvp, (struct vattr *)0);
1061 *vpp = newvp;
1062 m_freem(mrep);
1063 if (!lockparent) {
1064 VOP_UNLOCK(dvp, 0, td);
1065 cnp->cn_flags |= CNP_PDIRUNLOCK;
1066 }
1067 return (0);
1068 }
1069
1070 if (flags & CNP_ISDOTDOT) {
1071 VOP_UNLOCK(dvp, 0, td);
1072 cnp->cn_flags |= CNP_PDIRUNLOCK;
1073 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1074 if (error) {
1075 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
1076 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1077 return (error); /* NOTE: return error from nget */
1078 }
1079 newvp = NFSTOV(np);
1080 if (lockparent) {
1081 error = vn_lock(dvp, LK_EXCLUSIVE, td);
1082 if (error) {
1083 vput(newvp);
1084 return (error);
1085 }
1086 cnp->cn_flags |= CNP_PDIRUNLOCK;
1087 }
1088 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1089 vref(dvp);
1090 newvp = dvp;
1091 } else {
1092 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1093 if (error) {
1094 m_freem(mrep);
1095 return (error);
1096 }
1097 if (!lockparent) {
1098 VOP_UNLOCK(dvp, 0, td);
1099 cnp->cn_flags |= CNP_PDIRUNLOCK;
1100 }
1101 newvp = NFSTOV(np);
1102 }
1103 if (v3) {
1104 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
1105 nfsm_postop_attr(dvp, attrflag, NFS_LATTR_NOSHRINK);
1106 } else
1107 nfsm_loadattr(newvp, (struct vattr *)0);
1108#if 0
1109 /* XXX MOVE TO nfs_nremove() */
1110 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1111 cnp->cn_nameiop != NAMEI_DELETE) {
1112 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1113 }
1114#endif
1115 *vpp = newvp;
1116 m_freem(mrep);
1117nfsmout:
1118 if (error) {
1119 if (newvp != NULLVP) {
1120 vrele(newvp);
1121 *vpp = NULLVP;
1122 }
1123 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1124 cnp->cn_nameiop == NAMEI_RENAME) &&
1125 error == ENOENT) {
1126 if (!lockparent) {
1127 VOP_UNLOCK(dvp, 0, td);
1128 cnp->cn_flags |= CNP_PDIRUNLOCK;
1129 }
1130 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1131 error = EROFS;
1132 else
1133 error = EJUSTRETURN;
1134 }
1135 }
1136 return (error);
1137}
1138
1139/*
1140 * nfs read call.
1141 * Just call nfs_bioread() to do the work.
1142 *
1143 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1144 * struct ucred *a_cred)
1145 */
1146static int
1147nfs_read(struct vop_read_args *ap)
1148{
1149 struct vnode *vp = ap->a_vp;
1150
1151 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1152 switch (vp->v_type) {
1153 case VREG:
1154 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag));
1155 case VDIR:
1156 return (EISDIR);
1157 default:
1158 return EOPNOTSUPP;
1159 }
1160}
1161
1162/*
1163 * nfs readlink call
1164 *
1165 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1166 */
1167static int
1168nfs_readlink(struct vop_readlink_args *ap)
1169{
1170 struct vnode *vp = ap->a_vp;
1171
1172 if (vp->v_type != VLNK)
1173 return (EINVAL);
1174 return (nfs_bioread(vp, ap->a_uio, 0));
1175}
1176
1177/*
1178 * Do a readlink rpc.
1179 * Called by nfs_doio() from below the buffer cache.
1180 */
1181int
1182nfs_readlinkrpc(struct vnode *vp, struct uio *uiop)
1183{
1184 u_int32_t *tl;
1185 caddr_t cp;
1186 int32_t t1, t2;
1187 caddr_t bpos, dpos, cp2;
1188 int error = 0, len, attrflag;
1189 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1190 int v3 = NFS_ISV3(vp);
1191
1192 nfsstats.rpccnt[NFSPROC_READLINK]++;
1193 nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1194 nfsm_fhtom(vp, v3);
1195 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, nfs_vpcred(vp, ND_CHECK));
1196 if (v3)
1197 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1198 if (!error) {
1199 nfsm_strsiz(len, NFS_MAXPATHLEN);
1200 if (len == NFS_MAXPATHLEN) {
1201 struct nfsnode *np = VTONFS(vp);
1202 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1203 len = np->n_size;
1204 }
1205 nfsm_mtouio(uiop, len);
1206 }
1207 m_freem(mrep);
1208nfsmout:
1209 return (error);
1210}
1211
1212/*
1213 * nfs read rpc call
1214 * Ditto above
1215 */
1216int
1217nfs_readrpc(struct vnode *vp, struct uio *uiop)
1218{
1219 u_int32_t *tl;
1220 caddr_t cp;
1221 int32_t t1, t2;
1222 caddr_t bpos, dpos, cp2;
1223 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1224 struct nfsmount *nmp;
1225 int error = 0, len, retlen, tsiz, eof, attrflag;
1226 int v3 = NFS_ISV3(vp);
1227
1228#ifndef nolint
1229 eof = 0;
1230#endif
1231 nmp = VFSTONFS(vp->v_mount);
1232 tsiz = uiop->uio_resid;
1233 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1234 return (EFBIG);
1235 while (tsiz > 0) {
1236 nfsstats.rpccnt[NFSPROC_READ]++;
1237 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1238 nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1239 nfsm_fhtom(vp, v3);
1240 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1241 if (v3) {
1242 txdr_hyper(uiop->uio_offset, tl);
1243 *(tl + 2) = txdr_unsigned(len);
1244 } else {
1245 *tl++ = txdr_unsigned(uiop->uio_offset);
1246 *tl++ = txdr_unsigned(len);
1247 *tl = 0;
1248 }
1249 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, nfs_vpcred(vp, ND_READ));
1250 if (v3) {
1251 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1252 if (error) {
1253 m_freem(mrep);
1254 goto nfsmout;
1255 }
1256 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1257 eof = fxdr_unsigned(int, *(tl + 1));
1258 } else
1259 nfsm_loadattr(vp, (struct vattr *)0);
1260 nfsm_strsiz(retlen, nmp->nm_rsize);
1261 nfsm_mtouio(uiop, retlen);
1262 m_freem(mrep);
1263 tsiz -= retlen;
1264 if (v3) {
1265 if (eof || retlen == 0) {
1266 tsiz = 0;
1267 }
1268 } else if (retlen < len) {
1269 tsiz = 0;
1270 }
1271 }
1272nfsmout:
1273 return (error);
1274}
1275
1276/*
1277 * nfs write call
1278 */
1279int
1280nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, int *must_commit)
1281{
1282 u_int32_t *tl;
1283 caddr_t cp;
1284 int32_t t1, t2, backup;
1285 caddr_t bpos, dpos, cp2;
1286 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1287 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1288 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1289 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1290
1291#ifndef DIAGNOSTIC
1292 if (uiop->uio_iovcnt != 1)
1293 panic("nfs: writerpc iovcnt > 1");
1294#endif
1295 *must_commit = 0;
1296 tsiz = uiop->uio_resid;
1297 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1298 return (EFBIG);
1299 while (tsiz > 0) {
1300 nfsstats.rpccnt[NFSPROC_WRITE]++;
1301 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1302 nfsm_reqhead(vp, NFSPROC_WRITE,
1303 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1304 nfsm_fhtom(vp, v3);
1305 if (v3) {
1306 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1307 txdr_hyper(uiop->uio_offset, tl);
1308 tl += 2;
1309 *tl++ = txdr_unsigned(len);
1310 *tl++ = txdr_unsigned(*iomode);
1311 *tl = txdr_unsigned(len);
1312 } else {
1313 u_int32_t x;
1314
1315 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1316 /* Set both "begin" and "current" to non-garbage. */
1317 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1318 *tl++ = x; /* "begin offset" */
1319 *tl++ = x; /* "current offset" */
1320 x = txdr_unsigned(len);
1321 *tl++ = x; /* total to this offset */
1322 *tl = x; /* size of this write */
1323 }
1324 nfsm_uiotom(uiop, len);
1325 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, nfs_vpcred(vp, ND_WRITE));
1326 if (v3) {
1327 /*
1328 * The write RPC returns a before and after mtime. The
1329 * nfsm_wcc_data() macro checks the before n_mtime
1330 * against the before time and stores the after time
1331 * in the nfsnode's cached vattr and n_mtime field.
1332 * The NRMODIFIED bit will be set if the before
1333 * time did not match the original mtime.
1334 */
1335 wccflag = NFSV3_WCCCHK;
1336 nfsm_wcc_data(vp, wccflag);
1337 if (!error) {
1338 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1339 + NFSX_V3WRITEVERF);
1340 rlen = fxdr_unsigned(int, *tl++);
1341 if (rlen == 0) {
1342 error = NFSERR_IO;
1343 m_freem(mrep);
1344 break;
1345 } else if (rlen < len) {
1346 backup = len - rlen;
1347 uiop->uio_iov->iov_base -= backup;
1348 uiop->uio_iov->iov_len += backup;
1349 uiop->uio_offset -= backup;
1350 uiop->uio_resid += backup;
1351 len = rlen;
1352 }
1353 commit = fxdr_unsigned(int, *tl++);
1354
1355 /*
1356 * Return the lowest committment level
1357 * obtained by any of the RPCs.
1358 */
1359 if (committed == NFSV3WRITE_FILESYNC)
1360 committed = commit;
1361 else if (committed == NFSV3WRITE_DATASYNC &&
1362 commit == NFSV3WRITE_UNSTABLE)
1363 committed = commit;
1364 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1365 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1366 NFSX_V3WRITEVERF);
1367 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1368 } else if (bcmp((caddr_t)tl,
1369 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1370 *must_commit = 1;
1371 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1372 NFSX_V3WRITEVERF);
1373 }
1374 }
1375 } else {
1376 nfsm_loadattr(vp, (struct vattr *)0);
1377 }
1378 m_freem(mrep);
1379 if (error)
1380 break;
1381 tsiz -= len;
1382 }
1383nfsmout:
1384 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1385 committed = NFSV3WRITE_FILESYNC;
1386 *iomode = committed;
1387 if (error)
1388 uiop->uio_resid = tsiz;
1389 return (error);
1390}
1391
1392/*
1393 * nfs mknod rpc
1394 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1395 * mode set to specify the file type and the size field for rdev.
1396 */
1397static int
1398nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1399 struct vattr *vap)
1400{
1401 struct nfsv2_sattr *sp;
1402 u_int32_t *tl;
1403 caddr_t cp;
1404 int32_t t1, t2;
1405 struct vnode *newvp = (struct vnode *)0;
1406 struct nfsnode *np = (struct nfsnode *)0;
1407 struct vattr vattr;
1408 char *cp2;
1409 caddr_t bpos, dpos;
1410 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1411 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1412 u_int32_t rdev;
1413 int v3 = NFS_ISV3(dvp);
1414
1415 if (vap->va_type == VCHR || vap->va_type == VBLK)
1416 rdev = txdr_unsigned(vap->va_rdev);
1417 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1418 rdev = nfs_xdrneg1;
1419 else {
1420 return (EOPNOTSUPP);
1421 }
1422 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1423 return (error);
1424 }
1425 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1426 nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1427 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1428 nfsm_fhtom(dvp, v3);
1429 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1430 if (v3) {
1431 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1432 *tl++ = vtonfsv3_type(vap->va_type);
1433 nfsm_v3attrbuild(vap, FALSE);
1434 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1435 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1436 *tl++ = txdr_unsigned(umajor(vap->va_rdev));
1437 *tl = txdr_unsigned(uminor(vap->va_rdev));
1438 }
1439 } else {
1440 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1441 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1442 sp->sa_uid = nfs_xdrneg1;
1443 sp->sa_gid = nfs_xdrneg1;
1444 sp->sa_size = rdev;
1445 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1446 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1447 }
1448 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_td, cnp->cn_cred);
1449 if (!error) {
1450 nfsm_mtofh(dvp, newvp, v3, gotvp);
1451 if (!gotvp) {
1452 if (newvp) {
1453 vput(newvp);
1454 newvp = (struct vnode *)0;
1455 }
1456 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1457 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1458 if (!error)
1459 newvp = NFSTOV(np);
1460 }
1461 }
1462 if (v3)
1463 nfsm_wcc_data(dvp, wccflag);
1464 m_freem(mrep);
1465nfsmout:
1466 if (error) {
1467 if (newvp)
1468 vput(newvp);
1469 } else {
1470 *vpp = newvp;
1471 }
1472 VTONFS(dvp)->n_flag |= NLMODIFIED;
1473 if (!wccflag)
1474 VTONFS(dvp)->n_attrstamp = 0;
1475 return (error);
1476}
1477
1478/*
1479 * nfs mknod vop
1480 * just call nfs_mknodrpc() to do the work.
1481 *
1482 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1483 * struct componentname *a_cnp, struct vattr *a_vap)
1484 */
1485/* ARGSUSED */
1486static int
1487nfs_mknod(struct vop_mknod_args *ap)
1488{
1489 return nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1490}
1491
1492static u_long create_verf;
1493/*
1494 * nfs file create call
1495 *
1496 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1497 * struct componentname *a_cnp, struct vattr *a_vap)
1498 */
1499static int
1500nfs_create(struct vop_create_args *ap)
1501{
1502 struct vnode *dvp = ap->a_dvp;
1503 struct vattr *vap = ap->a_vap;
1504 struct componentname *cnp = ap->a_cnp;
1505 struct nfsv2_sattr *sp;
1506 u_int32_t *tl;
1507 caddr_t cp;
1508 int32_t t1, t2;
1509 struct nfsnode *np = (struct nfsnode *)0;
1510 struct vnode *newvp = (struct vnode *)0;
1511 caddr_t bpos, dpos, cp2;
1512 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1513 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1514 struct vattr vattr;
1515 int v3 = NFS_ISV3(dvp);
1516
1517 /*
1518 * Oops, not for me..
1519 */
1520 if (vap->va_type == VSOCK)
1521 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1522
1523 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
1524 return (error);
1525 }
1526 if (vap->va_vaflags & VA_EXCLUSIVE)
1527 fmode |= O_EXCL;
1528again:
1529 nfsstats.rpccnt[NFSPROC_CREATE]++;
1530 nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1531 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1532 nfsm_fhtom(dvp, v3);
1533 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1534 if (v3) {
1535 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1536 if (fmode & O_EXCL) {
1537 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1538 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1539#ifdef INET
1540 if (!TAILQ_EMPTY(&in_ifaddrhead))
1541 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1542 else
1543#endif
1544 *tl++ = create_verf;
1545 *tl = ++create_verf;
1546 } else {
1547 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1548 nfsm_v3attrbuild(vap, FALSE);
1549 }
1550 } else {
1551 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1552 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1553 sp->sa_uid = nfs_xdrneg1;
1554 sp->sa_gid = nfs_xdrneg1;
1555 sp->sa_size = 0;
1556 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1557 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1558 }
1559 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_td, cnp->cn_cred);
1560 if (!error) {
1561 nfsm_mtofh(dvp, newvp, v3, gotvp);
1562 if (!gotvp) {
1563 if (newvp) {
1564 vput(newvp);
1565 newvp = (struct vnode *)0;
1566 }
1567 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1568 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1569 if (!error)
1570 newvp = NFSTOV(np);
1571 }
1572 }
1573 if (v3)
1574 nfsm_wcc_data(dvp, wccflag);
1575 m_freem(mrep);
1576nfsmout:
1577 if (error) {
1578 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1579 fmode &= ~O_EXCL;
1580 goto again;
1581 }
1582 if (newvp)
1583 vput(newvp);
1584 } else if (v3 && (fmode & O_EXCL)) {
1585 /*
1586 * We are normally called with only a partially initialized
1587 * VAP. Since the NFSv3 spec says that server may use the
1588 * file attributes to store the verifier, the spec requires
1589 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1590 * in atime, but we can't really assume that all servers will
1591 * so we ensure that our SETATTR sets both atime and mtime.
1592 */
1593 if (vap->va_mtime.tv_sec == VNOVAL)
1594 vfs_timestamp(&vap->va_mtime);
1595 if (vap->va_atime.tv_sec == VNOVAL)
1596 vap->va_atime = vap->va_mtime;
1597 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1598 }
1599 if (!error) {
1600 /*
1601 * The new np may have enough info for access
1602 * checks, make sure rucred and wucred are
1603 * initialized for read and write rpc's.
1604 */
1605 np = VTONFS(newvp);
1606 if (np->n_rucred == NULL)
1607 np->n_rucred = crhold(cnp->cn_cred);
1608 if (np->n_wucred == NULL)
1609 np->n_wucred = crhold(cnp->cn_cred);
1610 *ap->a_vpp = newvp;
1611 }
1612 VTONFS(dvp)->n_flag |= NLMODIFIED;
1613 if (!wccflag)
1614 VTONFS(dvp)->n_attrstamp = 0;
1615 return (error);
1616}
1617
1618/*
1619 * nfs file remove call
1620 * To try and make nfs semantics closer to ufs semantics, a file that has
1621 * other processes using the vnode is renamed instead of removed and then
1622 * removed later on the last close.
1623 * - If v_usecount > 1
1624 * If a rename is not already in the works
1625 * call nfs_sillyrename() to set it up
1626 * else
1627 * do the remove rpc
1628 *
1629 * nfs_remove(struct vnodeop_desc *a_desc, struct vnode *a_dvp,
1630 * struct vnode *a_vp, struct componentname *a_cnp)
1631 */
1632static int
1633nfs_remove(struct vop_remove_args *ap)
1634{
1635 struct vnode *vp = ap->a_vp;
1636 struct vnode *dvp = ap->a_dvp;
1637 struct componentname *cnp = ap->a_cnp;
1638 struct nfsnode *np = VTONFS(vp);
1639 int error = 0;
1640 struct vattr vattr;
1641
1642#ifndef DIAGNOSTIC
1643 if (vp->v_usecount < 1)
1644 panic("nfs_remove: bad v_usecount");
1645#endif
1646 if (vp->v_type == VDIR)
1647 error = EPERM;
1648 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1649 VOP_GETATTR(vp, &vattr, cnp->cn_td) == 0 &&
1650 vattr.va_nlink > 1)) {
1651 /*
1652 * throw away biocache buffers, mainly to avoid
1653 * unnecessary delayed writes later.
1654 */
1655 error = nfs_vinvalbuf(vp, 0, cnp->cn_td, 1);
1656 /* Do the rpc */
1657 if (error != EINTR)
1658 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1659 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td);
1660 /*
1661 * Kludge City: If the first reply to the remove rpc is lost..
1662 * the reply to the retransmitted request will be ENOENT
1663 * since the file was in fact removed
1664 * Therefore, we cheat and return success.
1665 */
1666 if (error == ENOENT)
1667 error = 0;
1668 } else if (!np->n_sillyrename) {
1669 error = nfs_sillyrename(dvp, vp, cnp);
1670 }
1671 np->n_attrstamp = 0;
1672 return (error);
1673}
1674
1675/*
1676 * nfs file remove rpc called from nfs_inactive
1677 */
1678int
1679nfs_removeit(struct sillyrename *sp)
1680{
1681 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1682 sp->s_cred, NULL));
1683}
1684
1685/*
1686 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1687 */
1688static int
1689nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1690 struct ucred *cred, struct thread *td)
1691{
1692 u_int32_t *tl;
1693 caddr_t cp;
1694 int32_t t1, t2;
1695 caddr_t bpos, dpos, cp2;
1696 int error = 0, wccflag = NFSV3_WCCRATTR;
1697 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1698 int v3 = NFS_ISV3(dvp);
1699
1700 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1701 nfsm_reqhead(dvp, NFSPROC_REMOVE,
1702 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1703 nfsm_fhtom(dvp, v3);
1704 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1705 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1706 if (v3)
1707 nfsm_wcc_data(dvp, wccflag);
1708 m_freem(mrep);
1709nfsmout:
1710 VTONFS(dvp)->n_flag |= NLMODIFIED;
1711 if (!wccflag)
1712 VTONFS(dvp)->n_attrstamp = 0;
1713 return (error);
1714}
1715
1716/*
1717 * nfs file rename call
1718 *
1719 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1720 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1721 * struct vnode *a_tvp, struct componentname *a_tcnp)
1722 */
1723static int
1724nfs_rename(struct vop_rename_args *ap)
1725{
1726 struct vnode *fvp = ap->a_fvp;
1727 struct vnode *tvp = ap->a_tvp;
1728 struct vnode *fdvp = ap->a_fdvp;
1729 struct vnode *tdvp = ap->a_tdvp;
1730 struct componentname *tcnp = ap->a_tcnp;
1731 struct componentname *fcnp = ap->a_fcnp;
1732 int error;
1733
1734 /* Check for cross-device rename */
1735 if ((fvp->v_mount != tdvp->v_mount) ||
1736 (tvp && (fvp->v_mount != tvp->v_mount))) {
1737 error = EXDEV;
1738 goto out;
1739 }
1740
1741 /*
1742 * We have to flush B_DELWRI data prior to renaming
1743 * the file. If we don't, the delayed-write buffers
1744 * can be flushed out later after the file has gone stale
1745 * under NFSV3. NFSV2 does not have this problem because
1746 * ( as far as I can tell ) it flushes dirty buffers more
1747 * often.
1748 */
1749
1750 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_td);
1751 if (tvp)
1752 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_td);
1753
1754 /*
1755 * If the tvp exists and is in use, sillyrename it before doing the
1756 * rename of the new file over it.
1757 *
1758 * XXX Can't sillyrename a directory.
1759 *
1760 * We do not attempt to do any namecache purges in this old API
1761 * routine. The new API compat functions have access to the actual
1762 * namecache structures and will do it for us.
1763 */
1764 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1765 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1766 vput(tvp);
1767 tvp = NULL;
1768 } else if (tvp) {
1769 ;
1770 }
1771
1772 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1773 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1774 tcnp->cn_td);
1775
1776out:
1777 if (tdvp == tvp)
1778 vrele(tdvp);
1779 else
1780 vput(tdvp);
1781 if (tvp)
1782 vput(tvp);
1783 vrele(fdvp);
1784 vrele(fvp);
1785 /*
1786 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1787 */
1788 if (error == ENOENT)
1789 error = 0;
1790 return (error);
1791}
1792
1793/*
1794 * nfs file rename rpc called from nfs_remove() above
1795 */
1796static int
1797nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1798 struct sillyrename *sp)
1799{
1800 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1801 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
1802}
1803
1804/*
1805 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1806 */
1807static int
1808nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1809 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1810 struct ucred *cred, struct thread *td)
1811{
1812 u_int32_t *tl;
1813 caddr_t cp;
1814 int32_t t1, t2;
1815 caddr_t bpos, dpos, cp2;
1816 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1817 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1818 int v3 = NFS_ISV3(fdvp);
1819
1820 nfsstats.rpccnt[NFSPROC_RENAME]++;
1821 nfsm_reqhead(fdvp, NFSPROC_RENAME,
1822 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1823 nfsm_rndup(tnamelen));
1824 nfsm_fhtom(fdvp, v3);
1825 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1826 nfsm_fhtom(tdvp, v3);
1827 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1828 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1829 if (v3) {
1830 nfsm_wcc_data(fdvp, fwccflag);
1831 nfsm_wcc_data(tdvp, twccflag);
1832 }
1833 m_freem(mrep);
1834nfsmout:
1835 VTONFS(fdvp)->n_flag |= NLMODIFIED;
1836 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1837 if (!fwccflag)
1838 VTONFS(fdvp)->n_attrstamp = 0;
1839 if (!twccflag)
1840 VTONFS(tdvp)->n_attrstamp = 0;
1841 return (error);
1842}
1843
1844/*
1845 * nfs hard link create call
1846 *
1847 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
1848 * struct componentname *a_cnp)
1849 */
1850static int
1851nfs_link(struct vop_link_args *ap)
1852{
1853 struct vnode *vp = ap->a_vp;
1854 struct vnode *tdvp = ap->a_tdvp;
1855 struct componentname *cnp = ap->a_cnp;
1856 u_int32_t *tl;
1857 caddr_t cp;
1858 int32_t t1, t2;
1859 caddr_t bpos, dpos, cp2;
1860 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1861 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1862 int v3;
1863
1864 if (vp->v_mount != tdvp->v_mount) {
1865 return (EXDEV);
1866 }
1867
1868 /*
1869 * Push all writes to the server, so that the attribute cache
1870 * doesn't get "out of sync" with the server.
1871 * XXX There should be a better way!
1872 */
1873 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_td);
1874
1875 v3 = NFS_ISV3(vp);
1876 nfsstats.rpccnt[NFSPROC_LINK]++;
1877 nfsm_reqhead(vp, NFSPROC_LINK,
1878 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1879 nfsm_fhtom(vp, v3);
1880 nfsm_fhtom(tdvp, v3);
1881 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1882 nfsm_request(vp, NFSPROC_LINK, cnp->cn_td, cnp->cn_cred);
1883 if (v3) {
1884 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
1885 nfsm_wcc_data(tdvp, wccflag);
1886 }
1887 m_freem(mrep);
1888nfsmout:
1889 VTONFS(tdvp)->n_flag |= NLMODIFIED;
1890 if (!attrflag)
1891 VTONFS(vp)->n_attrstamp = 0;
1892 if (!wccflag)
1893 VTONFS(tdvp)->n_attrstamp = 0;
1894 /*
1895 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1896 */
1897 if (error == EEXIST)
1898 error = 0;
1899 return (error);
1900}
1901
1902/*
1903 * nfs symbolic link create call
1904 *
1905 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
1906 * struct componentname *a_cnp, struct vattr *a_vap,
1907 * char *a_target)
1908 */
1909static int
1910nfs_symlink(struct vop_symlink_args *ap)
1911{
1912 struct vnode *dvp = ap->a_dvp;
1913 struct vattr *vap = ap->a_vap;
1914 struct componentname *cnp = ap->a_cnp;
1915 struct nfsv2_sattr *sp;
1916 u_int32_t *tl;
1917 caddr_t cp;
1918 int32_t t1, t2;
1919 caddr_t bpos, dpos, cp2;
1920 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1921 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
1922 struct vnode *newvp = (struct vnode *)0;
1923 int v3 = NFS_ISV3(dvp);
1924
1925 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1926 slen = strlen(ap->a_target);
1927 nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1928 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1929 nfsm_fhtom(dvp, v3);
1930 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1931 if (v3) {
1932 nfsm_v3attrbuild(vap, FALSE);
1933 }
1934 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1935 if (!v3) {
1936 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1937 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1938 sp->sa_uid = nfs_xdrneg1;
1939 sp->sa_gid = nfs_xdrneg1;
1940 sp->sa_size = nfs_xdrneg1;
1941 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1942 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1943 }
1944
1945 /*
1946 * Issue the NFS request and get the rpc response.
1947 *
1948 * Only NFSv3 responses returning an error of 0 actually return
1949 * a file handle that can be converted into newvp without having
1950 * to do an extra lookup rpc.
1951 */
1952 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_td, cnp->cn_cred);
1953 if (v3) {
1954 if (error == 0)
1955 nfsm_mtofh(dvp, newvp, v3, gotvp);
1956 nfsm_wcc_data(dvp, wccflag);
1957 }
1958
1959 /*
1960 * out code jumps -> here, mrep is also freed.
1961 */
1962
1963 m_freem(mrep);
1964nfsmout:
1965
1966 /*
1967 * If we get an EEXIST error, silently convert it to no-error
1968 * in case of an NFS retry.
1969 */
1970 if (error == EEXIST)
1971 error = 0;
1972
1973 /*
1974 * If we do not have (or no longer have) an error, and we could
1975 * not extract the newvp from the response due to the request being
1976 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1977 * to obtain a newvp to return.
1978 */
1979 if (error == 0 && newvp == NULL) {
1980 struct nfsnode *np = NULL;
1981
1982 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1983 cnp->cn_cred, cnp->cn_td, &np);
1984 if (!error)
1985 newvp = NFSTOV(np);
1986 }
1987 if (error) {
1988 if (newvp)
1989 vput(newvp);
1990 } else {
1991 *ap->a_vpp = newvp;
1992 }
1993 VTONFS(dvp)->n_flag |= NLMODIFIED;
1994 if (!wccflag)
1995 VTONFS(dvp)->n_attrstamp = 0;
1996 return (error);
1997}
1998
1999/*
2000 * nfs make dir call
2001 *
2002 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2003 * struct componentname *a_cnp, struct vattr *a_vap)
2004 */
2005static int
2006nfs_mkdir(struct vop_mkdir_args *ap)
2007{
2008 struct vnode *dvp = ap->a_dvp;
2009 struct vattr *vap = ap->a_vap;
2010 struct componentname *cnp = ap->a_cnp;
2011 struct nfsv2_sattr *sp;
2012 u_int32_t *tl;
2013 caddr_t cp;
2014 int32_t t1, t2;
2015 int len;
2016 struct nfsnode *np = (struct nfsnode *)0;
2017 struct vnode *newvp = (struct vnode *)0;
2018 caddr_t bpos, dpos, cp2;
2019 int error = 0, wccflag = NFSV3_WCCRATTR;
2020 int gotvp = 0;
2021 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2022 struct vattr vattr;
2023 int v3 = NFS_ISV3(dvp);
2024
2025 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_td)) != 0) {
2026 return (error);
2027 }
2028 len = cnp->cn_namelen;
2029 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2030 nfsm_reqhead(dvp, NFSPROC_MKDIR,
2031 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2032 nfsm_fhtom(dvp, v3);
2033 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2034 if (v3) {
2035 nfsm_v3attrbuild(vap, FALSE);
2036 } else {
2037 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2038 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2039 sp->sa_uid = nfs_xdrneg1;
2040 sp->sa_gid = nfs_xdrneg1;
2041 sp->sa_size = nfs_xdrneg1;
2042 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2043 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2044 }
2045 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_td, cnp->cn_cred);
2046 if (!error)
2047 nfsm_mtofh(dvp, newvp, v3, gotvp);
2048 if (v3)
2049 nfsm_wcc_data(dvp, wccflag);
2050 m_freem(mrep);
2051nfsmout:
2052 VTONFS(dvp)->n_flag |= NLMODIFIED;
2053 if (!wccflag)
2054 VTONFS(dvp)->n_attrstamp = 0;
2055 /*
2056 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2057 * if we can succeed in looking up the directory.
2058 */
2059 if (error == EEXIST || (!error && !gotvp)) {
2060 if (newvp) {
2061 vrele(newvp);
2062 newvp = (struct vnode *)0;
2063 }
2064 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2065 cnp->cn_td, &np);
2066 if (!error) {
2067 newvp = NFSTOV(np);
2068 if (newvp->v_type != VDIR)
2069 error = EEXIST;
2070 }
2071 }
2072 if (error) {
2073 if (newvp)
2074 vrele(newvp);
2075 } else
2076 *ap->a_vpp = newvp;
2077 return (error);
2078}
2079
2080/*
2081 * nfs remove directory call
2082 *
2083 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2084 * struct componentname *a_cnp)
2085 */
2086static int
2087nfs_rmdir(struct vop_rmdir_args *ap)
2088{
2089 struct vnode *vp = ap->a_vp;
2090 struct vnode *dvp = ap->a_dvp;
2091 struct componentname *cnp = ap->a_cnp;
2092 u_int32_t *tl;
2093 caddr_t cp;
2094 int32_t t1, t2;
2095 caddr_t bpos, dpos, cp2;
2096 int error = 0, wccflag = NFSV3_WCCRATTR;
2097 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2098 int v3 = NFS_ISV3(dvp);
2099
2100 if (dvp == vp)
2101 return (EINVAL);
2102 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2103 nfsm_reqhead(dvp, NFSPROC_RMDIR,
2104 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2105 nfsm_fhtom(dvp, v3);
2106 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2107 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_td, cnp->cn_cred);
2108 if (v3)
2109 nfsm_wcc_data(dvp, wccflag);
2110 m_freem(mrep);
2111nfsmout:
2112 VTONFS(dvp)->n_flag |= NLMODIFIED;
2113 if (!wccflag)
2114 VTONFS(dvp)->n_attrstamp = 0;
2115 /*
2116 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2117 */
2118 if (error == ENOENT)
2119 error = 0;
2120 return (error);
2121}
2122
2123/*
2124 * nfs readdir call
2125 *
2126 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2127 */
2128static int
2129nfs_readdir(struct vop_readdir_args *ap)
2130{
2131 struct vnode *vp = ap->a_vp;
2132 struct nfsnode *np = VTONFS(vp);
2133 struct uio *uio = ap->a_uio;
2134 int tresid, error;
2135 struct vattr vattr;
2136
2137 if (vp->v_type != VDIR)
2138 return (EPERM);
2139
2140 /*
2141 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2142 * and then check that is still valid, or if this is an NQNFS mount
2143 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2144 * VOP_GETATTR() does not necessarily go to the wire.
2145 */
2146 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2147 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2148 if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) {
2149 if (NQNFS_CKCACHABLE(vp, ND_READ)) {
2150 nfsstats.direofcache_hits++;
2151 return (0);
2152 }
2153 } else if (VOP_GETATTR(vp, &vattr, uio->uio_td) == 0 &&
2154 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2155 ) {
2156 nfsstats.direofcache_hits++;
2157 return (0);
2158 }
2159 }
2160
2161 /*
2162 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2163 * own cache coherency checks so we do not have to.
2164 */
2165 tresid = uio->uio_resid;
2166 error = nfs_bioread(vp, uio, 0);
2167
2168 if (!error && uio->uio_resid == tresid)
2169 nfsstats.direofcache_misses++;
2170 return (error);
2171}
2172
2173/*
2174 * Readdir rpc call.
2175 * Called from below the buffer cache by nfs_doio().
2176 */
2177int
2178nfs_readdirrpc(struct vnode *vp, struct uio *uiop)
2179{
2180 int len, left;
2181 struct dirent *dp = NULL;
2182 u_int32_t *tl;
2183 caddr_t cp;
2184 int32_t t1, t2;
2185 nfsuint64 *cookiep;
2186 caddr_t bpos, dpos, cp2;
2187 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2188 nfsuint64 cookie;
2189 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2190 struct nfsnode *dnp = VTONFS(vp);
2191 u_quad_t fileno;
2192 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2193 int attrflag;
2194 int v3 = NFS_ISV3(vp);
2195
2196#ifndef DIAGNOSTIC
2197 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2198 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2199 panic("nfs readdirrpc bad uio");
2200#endif
2201
2202 /*
2203 * If there is no cookie, assume directory was stale.
2204 */
2205 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2206 if (cookiep)
2207 cookie = *cookiep;
2208 else
2209 return (NFSERR_BAD_COOKIE);
2210 /*
2211 * Loop around doing readdir rpc's of size nm_readdirsize
2212 * truncated to a multiple of DIRBLKSIZ.
2213 * The stopping criteria is EOF or buffer full.
2214 */
2215 while (more_dirs && bigenough) {
2216 nfsstats.rpccnt[NFSPROC_READDIR]++;
2217 nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2218 NFSX_READDIR(v3));
2219 nfsm_fhtom(vp, v3);
2220 if (v3) {
2221 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2222 *tl++ = cookie.nfsuquad[0];
2223 *tl++ = cookie.nfsuquad[1];
2224 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2225 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2226 } else {
2227 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2228 *tl++ = cookie.nfsuquad[0];
2229 }
2230 *tl = txdr_unsigned(nmp->nm_readdirsize);
2231 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2232 if (v3) {
2233 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2234 if (!error) {
2235 nfsm_dissect(tl, u_int32_t *,
2236 2 * NFSX_UNSIGNED);
2237 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2238 dnp->n_cookieverf.nfsuquad[1] = *tl;
2239 } else {
2240 m_freem(mrep);
2241 goto nfsmout;
2242 }
2243 }
2244 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2245 more_dirs = fxdr_unsigned(int, *tl);
2246
2247 /* loop thru the dir entries, doctoring them to 4bsd form */
2248 while (more_dirs && bigenough) {
2249 if (v3) {
2250 nfsm_dissect(tl, u_int32_t *,
2251 3 * NFSX_UNSIGNED);
2252 fileno = fxdr_hyper(tl);
2253 len = fxdr_unsigned(int, *(tl + 2));
2254 } else {
2255 nfsm_dissect(tl, u_int32_t *,
2256 2 * NFSX_UNSIGNED);
2257 fileno = fxdr_unsigned(u_quad_t, *tl++);
2258 len = fxdr_unsigned(int, *tl);
2259 }
2260 if (len <= 0 || len > NFS_MAXNAMLEN) {
2261 error = EBADRPC;
2262 m_freem(mrep);
2263 goto nfsmout;
2264 }
2265 tlen = nfsm_rndup(len);
2266 if (tlen == len)
2267 tlen += 4; /* To ensure null termination */
2268 left = DIRBLKSIZ - blksiz;
2269 if ((tlen + DIRHDSIZ) > left) {
2270 dp->d_reclen += left;
2271 uiop->uio_iov->iov_base += left;
2272 uiop->uio_iov->iov_len -= left;
2273 uiop->uio_offset += left;
2274 uiop->uio_resid -= left;
2275 blksiz = 0;
2276 }
2277 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2278 bigenough = 0;
2279 if (bigenough) {
2280 dp = (struct dirent *)uiop->uio_iov->iov_base;
2281 dp->d_fileno = (int)fileno;
2282 dp->d_namlen = len;
2283 dp->d_reclen = tlen + DIRHDSIZ;
2284 dp->d_type = DT_UNKNOWN;
2285 blksiz += dp->d_reclen;
2286 if (blksiz == DIRBLKSIZ)
2287 blksiz = 0;
2288 uiop->uio_offset += DIRHDSIZ;
2289 uiop->uio_resid -= DIRHDSIZ;
2290 uiop->uio_iov->iov_base += DIRHDSIZ;
2291 uiop->uio_iov->iov_len -= DIRHDSIZ;
2292 nfsm_mtouio(uiop, len);
2293 cp = uiop->uio_iov->iov_base;
2294 tlen -= len;
2295 *cp = '\0'; /* null terminate */
2296 uiop->uio_iov->iov_base += tlen;
2297 uiop->uio_iov->iov_len -= tlen;
2298 uiop->uio_offset += tlen;
2299 uiop->uio_resid -= tlen;
2300 } else
2301 nfsm_adv(nfsm_rndup(len));
2302 if (v3) {
2303 nfsm_dissect(tl, u_int32_t *,
2304 3 * NFSX_UNSIGNED);
2305 } else {
2306 nfsm_dissect(tl, u_int32_t *,
2307 2 * NFSX_UNSIGNED);
2308 }
2309 if (bigenough) {
2310 cookie.nfsuquad[0] = *tl++;
2311 if (v3)
2312 cookie.nfsuquad[1] = *tl++;
2313 } else if (v3)
2314 tl += 2;
2315 else
2316 tl++;
2317 more_dirs = fxdr_unsigned(int, *tl);
2318 }
2319 /*
2320 * If at end of rpc data, get the eof boolean
2321 */
2322 if (!more_dirs) {
2323 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2324 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2325 }
2326 m_freem(mrep);
2327 }
2328 /*
2329 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2330 * by increasing d_reclen for the last record.
2331 */
2332 if (blksiz > 0) {
2333 left = DIRBLKSIZ - blksiz;
2334 dp->d_reclen += left;
2335 uiop->uio_iov->iov_base += left;
2336 uiop->uio_iov->iov_len -= left;
2337 uiop->uio_offset += left;
2338 uiop->uio_resid -= left;
2339 }
2340
2341 /*
2342 * We are now either at the end of the directory or have filled the
2343 * block.
2344 */
2345 if (bigenough)
2346 dnp->n_direofoffset = uiop->uio_offset;
2347 else {
2348 if (uiop->uio_resid > 0)
2349 printf("EEK! readdirrpc resid > 0\n");
2350 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2351 *cookiep = cookie;
2352 }
2353nfsmout:
2354 return (error);
2355}
2356
2357/*
2358 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2359 */
2360int
2361nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop)
2362{
2363 int len, left;
2364 struct dirent *dp;
2365 u_int32_t *tl;
2366 caddr_t cp;
2367 int32_t t1, t2;
2368 struct vnode *newvp;
2369 nfsuint64 *cookiep;
2370 caddr_t bpos, dpos, cp2, dpossav1, dpossav2;
2371 struct mbuf *mreq, *mrep, *md, *mb, *mb2, *mdsav1, *mdsav2;
2372 nfsuint64 cookie;
2373 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2374 struct nfsnode *dnp = VTONFS(vp), *np;
2375 nfsfh_t *fhp;
2376 u_quad_t fileno;
2377 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2378 int attrflag, fhsize;
2379 struct namecache *ncp;
2380 struct namecache *dncp;
2381 struct nlcomponent nlc;
2382
2383#ifndef nolint
2384 dp = (struct dirent *)0;
2385#endif
2386#ifndef DIAGNOSTIC
2387 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2388 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2389 panic("nfs readdirplusrpc bad uio");
2390#endif
2391 /*
2392 * Obtain the namecache record for the directory so we have something
2393 * to use as a basis for creating the entries. This function will
2394 * return a held (but not locked) ncp. The ncp may be disconnected
2395 * from the tree and cannot be used for upward traversals, and the
2396 * ncp may be unnamed. Note that other unrelated operations may
2397 * cause the ncp to be named at any time.
2398 */
2399 dncp = cache_fromdvp(vp, NULL, 0);
2400 bzero(&nlc, sizeof(nlc));
2401 newvp = NULLVP;
2402
2403 /*
2404 * If there is no cookie, assume directory was stale.
2405 */
2406 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2407 if (cookiep)
2408 cookie = *cookiep;
2409 else
2410 return (NFSERR_BAD_COOKIE);
2411 /*
2412 * Loop around doing readdir rpc's of size nm_readdirsize
2413 * truncated to a multiple of DIRBLKSIZ.
2414 * The stopping criteria is EOF or buffer full.
2415 */
2416 while (more_dirs && bigenough) {
2417 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2418 nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2419 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2420 nfsm_fhtom(vp, 1);
2421 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2422 *tl++ = cookie.nfsuquad[0];
2423 *tl++ = cookie.nfsuquad[1];
2424 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2425 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2426 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2427 *tl = txdr_unsigned(nmp->nm_rsize);
2428 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, nfs_vpcred(vp, ND_READ));
2429 nfsm_postop_attr(vp, attrflag, NFS_LATTR_NOSHRINK);
2430 if (error) {
2431 m_freem(mrep);
2432 goto nfsmout;
2433 }
2434 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2435 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2436 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2437 more_dirs = fxdr_unsigned(int, *tl);
2438
2439 /* loop thru the dir entries, doctoring them to 4bsd form */
2440 while (more_dirs && bigenough) {
2441 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2442 fileno = fxdr_hyper(tl);
2443 len = fxdr_unsigned(int, *(tl + 2));
2444 if (len <= 0 || len > NFS_MAXNAMLEN) {
2445 error = EBADRPC;
2446 m_freem(mrep);
2447 goto nfsmout;
2448 }
2449 tlen = nfsm_rndup(len);
2450 if (tlen == len)
2451 tlen += 4; /* To ensure null termination*/
2452 left = DIRBLKSIZ - blksiz;
2453 if ((tlen + DIRHDSIZ) > left) {
2454 dp->d_reclen += left;
2455 uiop->uio_iov->iov_base += left;
2456 uiop->uio_iov->iov_len -= left;
2457 uiop->uio_offset += left;
2458 uiop->uio_resid -= left;
2459 blksiz = 0;
2460 }
2461 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2462 bigenough = 0;
2463 if (bigenough) {
2464 dp = (struct dirent *)uiop->uio_iov->iov_base;
2465 dp->d_fileno = (int)fileno;
2466 dp->d_namlen = len;
2467 dp->d_reclen = tlen + DIRHDSIZ;
2468 dp->d_type = DT_UNKNOWN;
2469 blksiz += dp->d_reclen;
2470 if (blksiz == DIRBLKSIZ)
2471 blksiz = 0;
2472 uiop->uio_offset += DIRHDSIZ;
2473 uiop->uio_resid -= DIRHDSIZ;
2474 uiop->uio_iov->iov_base += DIRHDSIZ;
2475 uiop->uio_iov->iov_len -= DIRHDSIZ;
2476 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2477 nlc.nlc_namelen = len;
2478 nfsm_mtouio(uiop, len);
2479 cp = uiop->uio_iov->iov_base;
2480 tlen -= len;
2481 *cp = '\0';
2482 uiop->uio_iov->iov_base += tlen;
2483 uiop->uio_iov->iov_len -= tlen;
2484 uiop->uio_offset += tlen;
2485 uiop->uio_resid -= tlen;
2486 } else
2487 nfsm_adv(nfsm_rndup(len));
2488 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2489 if (bigenough) {
2490 cookie.nfsuquad[0] = *tl++;
2491 cookie.nfsuquad[1] = *tl++;
2492 } else
2493 tl += 2;
2494
2495 /*
2496 * Since the attributes are before the file handle
2497 * (sigh), we must skip over the attributes and then
2498 * come back and get them.
2499 */
2500 attrflag = fxdr_unsigned(int, *tl);
2501 if (attrflag) {
2502 dpossav1 = dpos;
2503 mdsav1 = md;
2504 nfsm_adv(NFSX_V3FATTR);
2505 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2506 doit = fxdr_unsigned(int, *tl);
2507 if (doit) {
2508 nfsm_getfh(fhp, fhsize, 1);
2509 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2510 vref(vp);
2511 newvp = vp;
2512 np = dnp;
2513 } else {
2514 error = nfs_nget(vp->v_mount, fhp,
2515 fhsize, &np);
2516 if (error)
2517 doit = 0;
2518 else
2519 newvp = NFSTOV(np);
2520 }
2521 }
2522 if (doit && bigenough) {
2523 dpossav2 = dpos;
2524 dpos = dpossav1;
2525 mdsav2 = md;
2526 md = mdsav1;
2527 nfsm_loadattr(newvp, (struct vattr *)0);
2528 dpos = dpossav2;
2529 md = mdsav2;
2530 dp->d_type =
2531 IFTODT(VTTOIF(np->n_vattr.va_type));
2532 if (dncp) {
2533 printf("NFS/READDIRPLUS, ENTER %*.*s\n",
2534 nlc.nlc_namelen, nlc.nlc_namelen,
2535 nlc.nlc_nameptr);
2536 ncp = cache_nlookup(dncp, &nlc);
2537 cache_setunresolved(ncp);
2538 cache_setvp(ncp, newvp);
2539 cache_put(ncp);
2540 } else {
2541 printf("NFS/READDIRPLUS, UNABLE TO ENTER"
2542 " %*.*s\n",
2543 nlc.nlc_namelen, nlc.nlc_namelen,
2544 nlc.nlc_nameptr);
2545 }
2546 }
2547 } else {
2548 /* Just skip over the file handle */
2549 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2550 i = fxdr_unsigned(int, *tl);
2551 nfsm_adv(nfsm_rndup(i));
2552 }
2553 if (newvp != NULLVP) {
2554 if (newvp == vp)
2555 vrele(newvp);
2556 else
2557 vput(newvp);
2558 newvp = NULLVP;
2559 }
2560 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2561 more_dirs = fxdr_unsigned(int, *tl);
2562 }
2563 /*
2564 * If at end of rpc data, get the eof boolean
2565 */
2566 if (!more_dirs) {
2567 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2568 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2569 }
2570 m_freem(mrep);
2571 }
2572 /*
2573 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2574 * by increasing d_reclen for the last record.
2575 */
2576 if (blksiz > 0) {
2577 left = DIRBLKSIZ - blksiz;
2578 dp->d_reclen += left;
2579 uiop->uio_iov->iov_base += left;
2580 uiop->uio_iov->iov_len -= left;
2581 uiop->uio_offset += left;
2582 uiop->uio_resid -= left;
2583 }
2584
2585 /*
2586 * We are now either at the end of the directory or have filled the
2587 * block.
2588 */
2589 if (bigenough)
2590 dnp->n_direofoffset = uiop->uio_offset;
2591 else {
2592 if (uiop->uio_resid > 0)
2593 printf("EEK! readdirplusrpc resid > 0\n");
2594 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2595 *cookiep = cookie;
2596 }
2597nfsmout:
2598 if (newvp != NULLVP) {
2599 if (newvp == vp)
2600 vrele(newvp);
2601 else
2602 vput(newvp);
2603 newvp = NULLVP;
2604 }
2605 if (dncp)
2606 cache_drop(dncp);
2607 return (error);
2608}
2609
2610/*
2611 * Silly rename. To make the NFS filesystem that is stateless look a little
2612 * more like the "ufs" a remove of an active vnode is translated to a rename
2613 * to a funny looking filename that is removed by nfs_inactive on the
2614 * nfsnode. There is the potential for another process on a different client
2615 * to create the same funny name between the nfs_lookitup() fails and the
2616 * nfs_rename() completes, but...
2617 */
2618static int
2619nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2620{
2621 struct sillyrename *sp;
2622 struct nfsnode *np;
2623 int error;
2624
2625 /*
2626 * We previously purged dvp instead of vp. I don't know why, it
2627 * completely destroys performance. We can't do it anyway with the
2628 * new VFS API since we would be breaking the namecache topology.
2629 */
2630 cache_purge(vp); /* XXX */
2631 np = VTONFS(vp);
2632#ifndef DIAGNOSTIC
2633 if (vp->v_type == VDIR)
2634 panic("nfs: sillyrename dir");
2635#endif
2636 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2637 M_NFSREQ, M_WAITOK);
2638 sp->s_cred = crdup(cnp->cn_cred);
2639 sp->s_dvp = dvp;
2640 vref(dvp);
2641
2642 /* Fudge together a funny name */
2643 sp->s_namlen = sprintf(sp->s_name, ".nfsA%08x4.4", (int)cnp->cn_td);
2644
2645 /* Try lookitups until we get one that isn't there */
2646 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2647 cnp->cn_td, (struct nfsnode **)0) == 0) {
2648 sp->s_name[4]++;
2649 if (sp->s_name[4] > 'z') {
2650 error = EINVAL;
2651 goto bad;
2652 }
2653 }
2654 error = nfs_renameit(dvp, cnp, sp);
2655 if (error)
2656 goto bad;
2657 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2658 cnp->cn_td, &np);
2659 np->n_sillyrename = sp;
2660 return (0);
2661bad:
2662 vrele(sp->s_dvp);
2663 crfree(sp->s_cred);
2664 free((caddr_t)sp, M_NFSREQ);
2665 return (error);
2666}
2667
2668/*
2669 * Look up a file name and optionally either update the file handle or
2670 * allocate an nfsnode, depending on the value of npp.
2671 * npp == NULL --> just do the lookup
2672 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2673 * handled too
2674 * *npp != NULL --> update the file handle in the vnode
2675 */
2676static int
2677nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2678 struct thread *td, struct nfsnode **npp)
2679{
2680 u_int32_t *tl;
2681 caddr_t cp;
2682 int32_t t1, t2;
2683 struct vnode *newvp = (struct vnode *)0;
2684 struct nfsnode *np, *dnp = VTONFS(dvp);
2685 caddr_t bpos, dpos, cp2;
2686 int error = 0, fhlen, attrflag;
2687 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2688 nfsfh_t *nfhp;
2689 int v3 = NFS_ISV3(dvp);
2690
2691 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2692 nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2693 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2694 nfsm_fhtom(dvp, v3);
2695 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2696 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2697 if (npp && !error) {
2698 nfsm_getfh(nfhp, fhlen, v3);
2699 if (*npp) {
2700 np = *npp;
2701 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2702 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2703 np->n_fhp = &np->n_fh;
2704 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2705 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2706 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2707 np->n_fhsize = fhlen;
2708 newvp = NFSTOV(np);
2709 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2710 vref(dvp);
2711 newvp = dvp;
2712 } else {
2713 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2714 if (error) {
2715 m_freem(mrep);
2716 return (error);
2717 }
2718 newvp = NFSTOV(np);
2719 }
2720 if (v3) {
2721 nfsm_postop_attr(newvp, attrflag, NFS_LATTR_NOSHRINK);
2722 if (!attrflag && *npp == NULL) {
2723 m_freem(mrep);
2724 if (newvp == dvp)
2725 vrele(newvp);
2726 else
2727 vput(newvp);
2728 return (ENOENT);
2729 }
2730 } else
2731 nfsm_loadattr(newvp, (struct vattr *)0);
2732 }
2733 m_freem(mrep);
2734nfsmout:
2735 if (npp && *npp == NULL) {
2736 if (error) {
2737 if (newvp) {
2738 if (newvp == dvp)
2739 vrele(newvp);
2740 else
2741 vput(newvp);
2742 }
2743 } else
2744 *npp = np;
2745 }
2746 return (error);
2747}
2748
2749/*
2750 * Nfs Version 3 commit rpc
2751 */
2752int
2753nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
2754{
2755 caddr_t cp;
2756 u_int32_t *tl;
2757 int32_t t1, t2;
2758 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2759 caddr_t bpos, dpos, cp2;
2760 int error = 0, wccflag = NFSV3_WCCRATTR;
2761 struct mbuf *mreq, *mrep, *md, *mb, *mb2;
2762
2763 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2764 return (0);
2765 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2766 nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2767 nfsm_fhtom(vp, 1);
2768 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2769 txdr_hyper(offset, tl);
2770 tl += 2;
2771 *tl = txdr_unsigned(cnt);
2772 nfsm_request(vp, NFSPROC_COMMIT, td, nfs_vpcred(vp, ND_WRITE));
2773 nfsm_wcc_data(vp, wccflag);
2774 if (!error) {
2775 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2776 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2777 NFSX_V3WRITEVERF)) {
2778 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2779 NFSX_V3WRITEVERF);
2780 error = NFSERR_STALEWRITEVERF;
2781 }
2782 }
2783 m_freem(mrep);
2784nfsmout:
2785 return (error);
2786}
2787
2788/*
2789 * Kludge City..
2790 * - make nfs_bmap() essentially a no-op that does no translation
2791 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2792 * (Maybe I could use the process's page mapping, but I was concerned that
2793 * Kernel Write might not be enabled and also figured copyout() would do
2794 * a lot more work than bcopy() and also it currently happens in the
2795 * context of the swapper process (2).
2796 *
2797 * nfs_bmap(struct vnode *a_vp, daddr_t a_bn, struct vnode **a_vpp,
2798 * daddr_t *a_bnp, int *a_runp, int *a_runb)
2799 */
2800static int
2801nfs_bmap(struct vop_bmap_args *ap)
2802{
2803 struct vnode *vp = ap->a_vp;
2804
2805 if (ap->a_vpp != NULL)
2806 *ap->a_vpp = vp;
2807 if (ap->a_bnp != NULL)
2808 *ap->a_bnp = ap->a_bn * btodb(vp->v_mount->mnt_stat.f_iosize);
2809 if (ap->a_runp != NULL)
2810 *ap->a_runp = 0;
2811 if (ap->a_runb != NULL)
2812 *ap->a_runb = 0;
2813 return (0);
2814}
2815
2816/*
2817 * Strategy routine.
2818 * For async requests when nfsiod(s) are running, queue the request by
2819 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2820 * request.
2821 */
2822static int
2823nfs_strategy(struct vop_strategy_args *ap)
2824{
2825 struct buf *bp = ap->a_bp;
2826 struct thread *td;
2827 int error = 0;
2828
2829 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2830 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
2831
2832 if (bp->b_flags & B_PHYS)
2833 panic("nfs physio");
2834
2835 if (bp->b_flags & B_ASYNC)
2836 td = NULL;
2837 else
2838 td = curthread; /* XXX */
2839
2840 /*
2841 * If the op is asynchronous and an i/o daemon is waiting
2842 * queue the request, wake it up and wait for completion
2843 * otherwise just do it ourselves.
2844 */
2845 if ((bp->b_flags & B_ASYNC) == 0 ||
2846 nfs_asyncio(bp, td))
2847 error = nfs_doio(bp, td);
2848 return (error);
2849}
2850
2851/*
2852 * Mmap a file
2853 *
2854 * NB Currently unsupported.
2855 *
2856 * nfs_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred,
2857 * struct thread *a_td)
2858 */
2859/* ARGSUSED */
2860static int
2861nfs_mmap(struct vop_mmap_args *ap)
2862{
2863 return (EINVAL);
2864}
2865
2866/*
2867 * fsync vnode op. Just call nfs_flush() with commit == 1.
2868 *
2869 * nfs_fsync(struct vnodeop_desc *a_desc, struct vnode *a_vp,
2870 * struct ucred * a_cred, int a_waitfor, struct thread *a_td)
2871 */
2872/* ARGSUSED */
2873static int
2874nfs_fsync(struct vop_fsync_args *ap)
2875{
2876 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2877}
2878
2879/*
2880 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
2881 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
2882 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
2883 * set the buffer contains data that has already been written to the server
2884 * and which now needs a commit RPC.
2885 *
2886 * If commit is 0 we only take one pass and only flush buffers containing new
2887 * dirty data.
2888 *
2889 * If commit is 1 we take two passes, issuing a commit RPC in the second
2890 * pass.
2891 *
2892 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
2893 * to completely flush all pending data.
2894 *
2895 * Note that the RB_SCAN code properly handles the case where the
2896 * callback might block and directly or indirectly (another thread) cause
2897 * the RB tree to change.
2898 */
2899
2900#ifndef NFS_COMMITBVECSIZ
2901#define NFS_COMMITBVECSIZ 16
2902#endif
2903
2904struct nfs_flush_info {
2905 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
2906 struct thread *td;
2907 struct vnode *vp;
2908 int waitfor;
2909 int slpflag;
2910 int slptimeo;
2911 int loops;
2912 struct buf *bvary[NFS_COMMITBVECSIZ];
2913 int bvsize;
2914 off_t beg_off;
2915 off_t end_off;
2916};
2917
2918static int nfs_flush_bp(struct buf *bp, void *data);
2919static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
2920
2921int
2922nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
2923{
2924 struct nfsnode *np = VTONFS(vp);
2925 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2926 struct nfs_flush_info info;
2927 int error;
2928
2929 bzero(&info, sizeof(info));
2930 info.td = td;
2931 info.vp = vp;
2932 info.waitfor = waitfor;
2933 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
2934 info.loops = 0;
2935
2936 do {
2937 /*
2938 * Flush mode
2939 */
2940 info.mode = NFI_FLUSHNEW;
2941 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2942 nfs_flush_bp, &info);
2943
2944 /*
2945 * Take a second pass if committing and no error occured.
2946 * Clean up any left over collection (whether an error
2947 * occurs or not).
2948 */
2949 if (commit && error == 0) {
2950 info.mode = NFI_COMMIT;
2951 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
2952 nfs_flush_bp, &info);
2953 if (info.bvsize)
2954 error = nfs_flush_docommit(&info, error);
2955 }
2956
2957 /*
2958 * Wait for pending I/O to complete before checking whether
2959 * any further dirty buffers exist.
2960 */
2961 while (waitfor == MNT_WAIT && vp->v_numoutput) {
2962 vp->v_flag |= VBWAIT;
2963 error = tsleep((caddr_t)&vp->v_numoutput,
2964 info.slpflag, "nfsfsync", info.slptimeo);
2965 if (error) {
2966 /*
2967 * We have to be able to break out if this
2968 * is an 'intr' mount.
2969 */
2970 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
2971 error = -EINTR;
2972 break;
2973 }
2974
2975 /*
2976 * Since we do not process pending signals,
2977 * once we get a PCATCH our tsleep() will no
2978 * longer sleep, switch to a fixed timeout
2979 * instead.
2980 */
2981 if (info.slpflag == PCATCH) {
2982 info.slpflag = 0;
2983 info.slptimeo = 2 * hz;
2984 }
2985 error = 0;
2986 }
2987 }
2988 ++info.loops;
2989 /*
2990 * Loop if we are flushing synchronous as well as committing,
2991 * and dirty buffers are still present. Otherwise we might livelock.
2992 */
2993 } while (waitfor == MNT_WAIT && commit &&
2994 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
2995
2996 /*
2997 * The callbacks have to return a negative error to terminate the
2998 * RB scan.
2999 */
3000 if (error < 0)
3001 error = -error;
3002
3003 /*
3004 * Deal with any error collection
3005 */
3006 if (np->n_flag & NWRITEERR) {
3007 error = np->n_error;
3008 np->n_flag &= ~NWRITEERR;
3009 }
3010 return (error);
3011}
3012
3013
3014static
3015int
3016nfs_flush_bp(struct buf *bp, void *data)
3017{
3018 struct nfs_flush_info *info = data;
3019 off_t toff;
3020 int error;
3021
3022 error = 0;
3023 switch(info->mode) {
3024 case NFI_FLUSHNEW:
3025 crit_enter();
3026 if (info->loops && info->waitfor == MNT_WAIT) {
3027 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3028 if (error) {
3029 error = BUF_TIMELOCK(bp,
3030 LK_EXCLUSIVE | LK_SLEEPFAIL,
3031 "nfsfsync",
3032 info->slpflag, info->slptimeo);
3033 }
3034 } else {
3035 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3036 }
3037 if (error == 0) {
3038 if ((bp->b_flags & B_DELWRI) == 0)
3039 panic("nfs_fsync: not dirty");
3040 if (bp->b_flags & B_NEEDCOMMIT) {
3041 BUF_UNLOCK(bp);
3042 crit_exit();
3043 break;
3044 }
3045 bremfree(bp);
3046
3047 bp->b_flags |= B_ASYNC;
3048 crit_exit();
3049 VOP_BWRITE(bp->b_vp, bp);
3050 } else {
3051 crit_exit();
3052 error = 0;
3053 }
3054 break;
3055 case NFI_COMMIT:
3056 /*
3057 * Only process buffers in need of a commit which we can
3058 * immediately lock. This may prevent a buffer from being
3059 * committed, but the normal flush loop will block on the
3060 * same buffer so we shouldn't get into an endless loop.
3061 */
3062 crit_enter();
3063 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3064 (B_DELWRI | B_NEEDCOMMIT) ||
3065 BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
3066 crit_exit();
3067 break;
3068 }
3069
3070 bremfree(bp);
3071
3072 /*
3073 * NOTE: we are not clearing B_DONE here, so we have
3074 * to do it later on in this routine if we intend to
3075 * initiate I/O on the bp.
3076 *
3077 * Note: to avoid loopback deadlocks, we do not
3078 * assign b_runningbufspace.
3079 */
3080 vfs_busy_pages(bp, 1);
3081
3082 info->bvary[info->bvsize] = bp;
3083 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3084 bp->b_dirtyoff;
3085 if (info->bvsize == 0 || toff < info->beg_off)
3086 info->beg_off = toff;
3087 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3088 if (info->bvsize == 0 || toff > info->end_off)
3089 info->end_off = toff;
3090 ++info->bvsize;
3091 if (info->bvsize == NFS_COMMITBVECSIZ) {
3092 error = nfs_flush_docommit(info, 0);
3093 KKASSERT(info->bvsize == 0);
3094 }
3095 crit_exit();
3096 }
3097 return (error);
3098}
3099
3100static
3101int
3102nfs_flush_docommit(struct nfs_flush_info *info, int error)
3103{
3104 struct vnode *vp;
3105 struct buf *bp;
3106 off_t bytes;
3107 int retv;
3108 int i;
3109
3110 vp = info->vp;
3111
3112 if (info->bvsize > 0) {
3113 /*
3114 * Commit data on the server, as required. Note that
3115 * nfs_commit will use the vnode's cred for the commit.
3116 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3117 */
3118 bytes = info->end_off - info->beg_off;
3119 if (bytes > 0x40000000)
3120 bytes = 0x40000000;
3121 if (error) {
3122 retv = -error;
3123 } else {
3124 retv = nfs_commit(vp, info->beg_off,
3125 (int)bytes, info->td);
3126 if (retv == NFSERR_STALEWRITEVERF)
3127 nfs_clearcommit(vp->v_mount);
3128 }
3129
3130 /*
3131 * Now, either mark the blocks I/O done or mark the
3132 * blocks dirty, depending on whether the commit
3133 * succeeded.
3134 */
3135 for (i = 0; i < info->bvsize; ++i) {
3136 bp = info->bvary[i];
3137 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3138 if (retv) {
3139 /*
3140 * Error, leave B_DELWRI intact
3141 */
3142 vfs_unbusy_pages(bp);
3143 brelse(bp);
3144 } else {
3145 /*
3146 * Success, remove B_DELWRI ( bundirty() ).
3147 *
3148 * b_dirtyoff/b_dirtyend seem to be NFS
3149 * specific. We should probably move that
3150 * into bundirty(). XXX
3151 */
3152 crit_enter();
3153 vp->v_numoutput++;
3154 bp->b_flags |= B_ASYNC;
3155 bundirty(bp);
3156 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3157 bp->b_dirtyoff = bp->b_dirtyend = 0;
3158 crit_exit();
3159 biodone(bp);
3160 }
3161 }
3162 info->bvsize = 0;
3163 }
3164 return (error);
3165}
3166
3167/*
3168 * NFS advisory byte-level locks.
3169 * Currently unsupported.
3170 *
3171 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3172 * int a_flags)
3173 */
3174static int
3175nfs_advlock(struct vop_advlock_args *ap)
3176{
3177 struct nfsnode *np = VTONFS(ap->a_vp);
3178
3179 /*
3180 * The following kludge is to allow diskless support to work
3181 * until a real NFS lockd is implemented. Basically, just pretend
3182 * that this is a local lock.
3183 */
3184 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3185}
3186
3187/*
3188 * Print out the contents of an nfsnode.
3189 *
3190 * nfs_print(struct vnode *a_vp)
3191 */
3192static int
3193nfs_print(struct vop_print_args *ap)
3194{
3195 struct vnode *vp = ap->a_vp;
3196 struct nfsnode *np = VTONFS(vp);
3197
3198 printf("tag VT_NFS, fileid %ld fsid 0x%x",
3199 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3200 if (vp->v_type == VFIFO)
3201 fifo_printinfo(vp);
3202 printf("\n");
3203 return (0);
3204}
3205
3206/*
3207 * Just call nfs_writebp() with the force argument set to 1.
3208 *
3209 * NOTE: B_DONE may or may not be set in a_bp on call.
3210 *
3211 * nfs_bwrite(struct vnode *a_bp)
3212 */
3213static int
3214nfs_bwrite(struct vop_bwrite_args *ap)
3215{
3216 return (nfs_writebp(ap->a_bp, 1, curthread));
3217}
3218
3219/*
3220 * This is a clone of vn_bwrite(), except that it also handles the
3221 * B_NEEDCOMMIT flag. We set B_CACHE if this is a VMIO buffer.
3222 */
3223int
3224nfs_writebp(struct buf *bp, int force, struct thread *td)
3225{
3226 int oldflags = bp->b_flags;
3227#if 0
3228 int retv = 1;
3229 off_t off;
3230#endif
3231
3232 if (BUF_REFCNT(bp) == 0)
3233 panic("bwrite: buffer is not locked???");
3234
3235 if (bp->b_flags & B_INVAL) {
3236 brelse(bp);
3237 return(0);
3238 }
3239
3240 bp->b_flags |= B_CACHE;
3241
3242 /*
3243 * Undirty the bp. We will redirty it later if the I/O fails.
3244 */
3245
3246 crit_enter();
3247 bundirty(bp);
3248 bp->b_flags &= ~(B_READ|B_DONE|B_ERROR);
3249
3250 bp->b_vp->v_numoutput++;
3251 crit_exit();
3252
3253 /*
3254 * Note: to avoid loopback deadlocks, we do not
3255 * assign b_runningbufspace.
3256 */
3257 vfs_busy_pages(bp, 1);
3258
3259 BUF_KERNPROC(bp);
3260 VOP_STRATEGY(bp->b_vp, bp);
3261
3262 if( (oldflags & B_ASYNC) == 0) {
3263 int rtval = biowait(bp);
3264
3265 if (oldflags & B_DELWRI) {
3266 crit_enter();
3267 reassignbuf(bp, bp->b_vp);
3268 crit_exit();
3269 }
3270
3271 brelse(bp);
3272 return (rtval);
3273 }
3274
3275 return (0);
3276}
3277
3278/*
3279 * nfs special file access vnode op.
3280 * Essentially just get vattr and then imitate iaccess() since the device is
3281 * local to the client.
3282 *
3283 * nfsspec_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
3284 * struct thread *a_td)
3285 */
3286static int
3287nfsspec_access(struct vop_access_args *ap)
3288{
3289 struct vattr *vap;
3290 gid_t *gp;
3291 struct ucred *cred = ap->a_cred;
3292 struct vnode *vp = ap->a_vp;
3293 mode_t mode = ap->a_mode;
3294 struct vattr vattr;
3295 int i;
3296 int error;
3297
3298 /*
3299 * Disallow write attempts on filesystems mounted read-only;
3300 * unless the file is a socket, fifo, or a block or character
3301 * device resident on the filesystem.
3302 */
3303 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3304 switch (vp->v_type) {
3305 case VREG:
3306 case VDIR:
3307 case VLNK:
3308 return (EROFS);
3309 default:
3310 break;
3311 }
3312 }
3313 /*
3314 * If you're the super-user,
3315 * you always get access.
3316 */
3317 if (cred->cr_uid == 0)
3318 return (0);
3319 vap = &vattr;
3320 error = VOP_GETATTR(vp, vap, ap->a_td);
3321 if (error)
3322 return (error);
3323 /*
3324 * Access check is based on only one of owner, group, public.
3325 * If not owner, then check group. If not a member of the
3326 * group, then check public access.
3327 */
3328 if (cred->cr_uid != vap->va_uid) {
3329 mode >>= 3;
3330 gp = cred->cr_groups;
3331 for (i = 0; i < cred->cr_ngroups; i++, gp++)
3332 if (vap->va_gid == *gp)
3333 goto found;
3334 mode >>= 3;
3335found:
3336 ;
3337 }
3338 error = (vap->va_mode & mode) == mode ? 0 : EACCES;
3339 return (error);
3340}
3341
3342/*
3343 * Read wrapper for special devices.
3344 *
3345 * nfsspec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3346 * struct ucred *a_cred)
3347 */
3348static int
3349nfsspec_read(struct vop_read_args *ap)
3350{
3351 struct nfsnode *np = VTONFS(ap->a_vp);
3352
3353 /*
3354 * Set access flag.
3355 */
3356 np->n_flag |= NACC;
3357 getnanotime(&np->n_atim);
3358 return (VOCALL(spec_vnode_vops, &ap->a_head));
3359}
3360
3361/*
3362 * Write wrapper for special devices.
3363 *
3364 * nfsspec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3365 * struct ucred *a_cred)
3366 */
3367static int
3368nfsspec_write(struct vop_write_args *ap)
3369{
3370 struct nfsnode *np = VTONFS(ap->a_vp);
3371
3372 /*
3373 * Set update flag.
3374 */
3375 np->n_flag |= NUPD;
3376 getnanotime(&np->n_mtim);
3377 return (VOCALL(spec_vnode_vops, &ap->a_head));
3378}
3379
3380/*
3381 * Close wrapper for special devices.
3382 *
3383 * Update the times on the nfsnode then do device close.
3384 *
3385 * nfsspec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
3386 * struct thread *a_td)
3387 */
3388static int
3389nfsspec_close(struct vop_close_args *ap)
3390{
3391 struct vnode *vp = ap->a_vp;
3392 struct nfsnode *np = VTONFS(vp);
3393 struct vattr vattr;
3394
3395 if (np->n_flag & (NACC | NUPD)) {
3396 np->n_flag |= NCHG;
3397 if (vp->v_usecount == 1 &&
3398 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3399 VATTR_NULL(&vattr);
3400 if (np->n_flag & NACC)
3401 vattr.va_atime = np->n_atim;
3402 if (np->n_flag & NUPD)
3403 vattr.va_mtime = np->n_mtim;
3404 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
3405 }
3406 }
3407 return (VOCALL(spec_vnode_vops, &ap->a_head));
3408}
3409
3410/*
3411 * Read wrapper for fifos.
3412 *
3413 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3414 * struct ucred *a_cred)
3415 */
3416static int
3417nfsfifo_read(struct vop_read_args *ap)
3418{
3419 struct nfsnode *np = VTONFS(ap->a_vp);
3420
3421 /*
3422 * Set access flag.
3423 */
3424 np->n_flag |= NACC;
3425 getnanotime(&np->n_atim);
3426 return (VOCALL(fifo_vnode_vops, &ap->a_head));
3427}
3428
3429/*
3430 * Write wrapper for fifos.
3431 *
3432 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3433 * struct ucred *a_cred)
3434 */
3435static int
3436nfsfifo_write(struct vop_write_args *ap)
3437{
3438 struct nfsnode *np = VTONFS(ap->a_vp);
3439
3440 /*
3441 * Set update flag.
3442 */
3443 np->n_flag |= NUPD;
3444 getnanotime(&np->n_mtim);
3445 return (VOCALL(fifo_vnode_vops, &ap->a_head));
3446}
3447
3448/*
3449 * Close wrapper for fifos.
3450 *
3451 * Update the times on the nfsnode then do fifo close.
3452 *
3453 * nfsfifo_close(struct vnode *a_vp, int a_fflag, struct thread *a_td)
3454 */
3455static int
3456nfsfifo_close(struct vop_close_args *ap)
3457{
3458 struct vnode *vp = ap->a_vp;
3459 struct nfsnode *np = VTONFS(vp);
3460 struct vattr vattr;
3461 struct timespec ts;
3462
3463 if (np->n_flag & (NACC | NUPD)) {
3464 getnanotime(&ts);
3465 if (np->n_flag & NACC)
3466 np->n_atim = ts;
3467 if (np->n_flag & NUPD)
3468 np->n_mtim = ts;
3469 np->n_flag |= NCHG;
3470 if (vp->v_usecount == 1 &&
3471 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3472 VATTR_NULL(&vattr);
3473 if (np->n_flag & NACC)
3474 vattr.va_atime = np->n_atim;
3475 if (np->n_flag & NUPD)
3476 vattr.va_mtime = np->n_mtim;
3477 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE), ap->a_td);
3478 }
3479 }
3480 return (VOCALL(fifo_vnode_vops, &ap->a_head));
3481}
3482