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. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
33 * $FreeBSD: src/sys/nfs/nfs_vnops.c,v 1.150.2.5 2001/12/20 19:56:28 dillon Exp $
34 */
35
36
37 /*
38 * vnode op calls for Sun NFS version 2 and 3
39 */
40
41 #include "opt_inet.h"
42
43 #include <sys/param.h>
44 #include <sys/kernel.h>
45 #include <sys/systm.h>
46 #include <sys/uio.h>
47 #include <sys/resourcevar.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/buf.h>
51 #include <sys/malloc.h>
52 #include <sys/mbuf.h>
53 #include <sys/namei.h>
54 #include <sys/nlookup.h>
55 #include <sys/socket.h>
56 #include <sys/vnode.h>
57 #include <sys/dirent.h>
58 #include <sys/fcntl.h>
59 #include <sys/lockf.h>
60 #include <sys/stat.h>
61 #include <sys/sysctl.h>
62 #include <sys/conf.h>
63
64 #include <vm/vm.h>
65 #include <vm/vm_extern.h>
66
67 #include <sys/buf2.h>
68
69 #include <vfs/fifofs/fifo.h>
70 #include <vfs/ufs/dir.h>
71
72 #undef DIRBLKSIZ
73
74 #include "rpcv2.h"
75 #include "nfsproto.h"
76 #include "nfs.h"
77 #include "nfsmount.h"
78 #include "nfsnode.h"
79 #include "xdr_subs.h"
80 #include "nfsm_subs.h"
81
82 #include <net/if.h>
83 #include <netinet/in.h>
84 #include <netinet/in_var.h>
85
86 /* Defs */
87 #define TRUE 1
88 #define FALSE 0
89
90 static int nfsfifo_read (struct vop_read_args *);
91 static int nfsfifo_write (struct vop_write_args *);
92 static int nfsfifo_close (struct vop_close_args *);
93 static int nfs_setattrrpc (struct vnode *,struct vattr *,struct ucred *,struct thread *);
94 static int nfs_lookup (struct vop_old_lookup_args *);
95 static int nfs_create (struct vop_old_create_args *);
96 static int nfs_mknod (struct vop_old_mknod_args *);
97 static int nfs_open (struct vop_open_args *);
98 static int nfs_close (struct vop_close_args *);
99 static int nfs_access (struct vop_access_args *);
100 static int nfs_getattr (struct vop_getattr_args *);
101 static int nfs_setattr (struct vop_setattr_args *);
102 static int nfs_read (struct vop_read_args *);
103 static int nfs_fsync (struct vop_fsync_args *);
104 static int nfs_remove (struct vop_old_remove_args *);
105 static int nfs_link (struct vop_old_link_args *);
106 static int nfs_rename (struct vop_old_rename_args *);
107 static int nfs_mkdir (struct vop_old_mkdir_args *);
108 static int nfs_rmdir (struct vop_old_rmdir_args *);
109 static int nfs_symlink (struct vop_old_symlink_args *);
110 static int nfs_readdir (struct vop_readdir_args *);
111 static int nfs_bmap (struct vop_bmap_args *);
112 static int nfs_strategy (struct vop_strategy_args *);
113 static int nfs_lookitup (struct vnode *, const char *, int,
114 struct ucred *, struct thread *, struct nfsnode **);
115 static int nfs_sillyrename (struct vnode *,struct vnode *,struct componentname *);
116 static int nfs_laccess (struct vop_access_args *);
117 static int nfs_readlink (struct vop_readlink_args *);
118 static int nfs_print (struct vop_print_args *);
119 static int nfs_advlock (struct vop_advlock_args *);
120 static int nfs_kqfilter (struct vop_kqfilter_args *ap);
121
122 static int nfs_nresolve (struct vop_nresolve_args *);
123 /*
124 * Global vfs data structures for nfs
125 */
126 struct vop_ops nfsv2_vnode_vops = {
127 .vop_default = vop_defaultop,
128 .vop_access = nfs_access,
129 .vop_advlock = nfs_advlock,
130 .vop_bmap = nfs_bmap,
131 .vop_close = nfs_close,
132 .vop_old_create = nfs_create,
133 .vop_fsync = nfs_fsync,
134 .vop_getattr = nfs_getattr,
135 .vop_getpages = vop_stdgetpages,
136 .vop_putpages = vop_stdputpages,
137 .vop_inactive = nfs_inactive,
138 .vop_old_link = nfs_link,
139 .vop_old_lookup = nfs_lookup,
140 .vop_old_mkdir = nfs_mkdir,
141 .vop_old_mknod = nfs_mknod,
142 .vop_open = nfs_open,
143 .vop_print = nfs_print,
144 .vop_read = nfs_read,
145 .vop_readdir = nfs_readdir,
146 .vop_readlink = nfs_readlink,
147 .vop_reclaim = nfs_reclaim,
148 .vop_old_remove = nfs_remove,
149 .vop_old_rename = nfs_rename,
150 .vop_old_rmdir = nfs_rmdir,
151 .vop_setattr = nfs_setattr,
152 .vop_strategy = nfs_strategy,
153 .vop_old_symlink = nfs_symlink,
154 .vop_write = nfs_write,
155 .vop_nresolve = nfs_nresolve,
156 .vop_kqfilter = nfs_kqfilter
157 };
158
159 /*
160 * Special device vnode ops
161 */
162 struct vop_ops nfsv2_spec_vops = {
163 .vop_default = vop_defaultop,
164 .vop_access = nfs_laccess,
165 .vop_close = nfs_close,
166 .vop_fsync = nfs_fsync,
167 .vop_getattr = nfs_getattr,
168 .vop_inactive = nfs_inactive,
169 .vop_print = nfs_print,
170 .vop_read = vop_stdnoread,
171 .vop_reclaim = nfs_reclaim,
172 .vop_setattr = nfs_setattr,
173 .vop_write = vop_stdnowrite
174 };
175
176 struct vop_ops nfsv2_fifo_vops = {
177 .vop_default = fifo_vnoperate,
178 .vop_access = nfs_laccess,
179 .vop_close = nfsfifo_close,
180 .vop_fsync = nfs_fsync,
181 .vop_getattr = nfs_getattr,
182 .vop_inactive = nfs_inactive,
183 .vop_print = nfs_print,
184 .vop_read = nfsfifo_read,
185 .vop_reclaim = nfs_reclaim,
186 .vop_setattr = nfs_setattr,
187 .vop_write = nfsfifo_write
188 };
189
190 static int nfs_mknodrpc (struct vnode *dvp, struct vnode **vpp,
191 struct componentname *cnp,
192 struct vattr *vap);
193 static int nfs_removerpc (struct vnode *dvp, const char *name,
194 int namelen,
195 struct ucred *cred, struct thread *td);
196 static int nfs_renamerpc (struct vnode *fdvp, const char *fnameptr,
197 int fnamelen, struct vnode *tdvp,
198 const char *tnameptr, int tnamelen,
199 struct ucred *cred, struct thread *td);
200 static int nfs_renameit (struct vnode *sdvp,
201 struct componentname *scnp,
202 struct sillyrename *sp);
203
204 SYSCTL_DECL(_vfs_nfs);
205
206 static int nfs_flush_on_rename = 1;
207 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_rename, CTLFLAG_RW,
208 &nfs_flush_on_rename, 0, "flush fvp prior to rename");
209 static int nfs_flush_on_hlink = 0;
210 SYSCTL_INT(_vfs_nfs, OID_AUTO, flush_on_hlink, CTLFLAG_RW,
211 &nfs_flush_on_hlink, 0, "flush fvp prior to hard link");
212
213 static int nfsaccess_cache_timeout = NFS_DEFATTRTIMO;
214 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
215 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
216
217 static int nfsneg_cache_timeout = NFS_MINATTRTIMO;
218 SYSCTL_INT(_vfs_nfs, OID_AUTO, neg_cache_timeout, CTLFLAG_RW,
219 &nfsneg_cache_timeout, 0, "NFS NEGATIVE NAMECACHE timeout");
220
221 static int nfspos_cache_timeout = NFS_MINATTRTIMO;
222 SYSCTL_INT(_vfs_nfs, OID_AUTO, pos_cache_timeout, CTLFLAG_RW,
223 &nfspos_cache_timeout, 0, "NFS POSITIVE NAMECACHE timeout");
224
225 static int nfsv3_commit_on_close = 0;
226 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
227 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
228 #if 0
229 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
230 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
231
232 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
233 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
234 #endif
235
236 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
237 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
238 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
239
240 static __inline
241 void
nfs_knote(struct vnode * vp,int flags)242 nfs_knote(struct vnode *vp, int flags)
243 {
244 if (flags)
245 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
246 }
247
248 /*
249 * Returns whether a name component is a degenerate '.' or '..'.
250 */
251 static __inline
252 int
nlcdegenerate(struct nlcomponent * nlc)253 nlcdegenerate(struct nlcomponent *nlc)
254 {
255 if (nlc->nlc_namelen == 1 && nlc->nlc_nameptr[0] == '.')
256 return(1);
257 if (nlc->nlc_namelen == 2 &&
258 nlc->nlc_nameptr[0] == '.' && nlc->nlc_nameptr[1] == '.')
259 return(1);
260 return(0);
261 }
262
263 static int
nfs3_access_otw(struct vnode * vp,int wmode,struct thread * td,struct ucred * cred)264 nfs3_access_otw(struct vnode *vp, int wmode,
265 struct thread *td, struct ucred *cred)
266 {
267 struct nfsnode *np = VTONFS(vp);
268 int attrflag;
269 int error = 0;
270 u_int32_t *tl;
271 u_int32_t rmode;
272 struct nfsm_info info;
273
274 info.mrep = NULL;
275 info.v3 = 1;
276
277 nfsstats.rpccnt[NFSPROC_ACCESS]++;
278 nfsm_reqhead(&info, vp, NFSPROC_ACCESS,
279 NFSX_FH(info.v3) + NFSX_UNSIGNED);
280 ERROROUT(nfsm_fhtom(&info, vp));
281 tl = nfsm_build(&info, NFSX_UNSIGNED);
282 *tl = txdr_unsigned(wmode);
283 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_ACCESS, td, cred, &error));
284 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag, NFS_LATTR_NOSHRINK));
285 if (error == 0) {
286 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
287 rmode = fxdr_unsigned(u_int32_t, *tl);
288 np->n_mode = rmode;
289 np->n_modeuid = cred->cr_uid;
290 np->n_modestamp = mycpu->gd_time_seconds;
291 }
292 m_freem(info.mrep);
293 info.mrep = NULL;
294 nfsmout:
295 return error;
296 }
297
298 /*
299 * nfs access vnode op.
300 * For nfs version 2, just return ok. File accesses may fail later.
301 * For nfs version 3, use the access rpc to check accessibility. If file modes
302 * are changed on the server, accesses might still fail later.
303 *
304 * nfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
305 */
306 static int
nfs_access(struct vop_access_args * ap)307 nfs_access(struct vop_access_args *ap)
308 {
309 struct ucred *cred;
310 struct ucred *ncred;
311 struct ucred *ocred;
312 struct vnode *vp = ap->a_vp;
313 thread_t td = curthread;
314 int error = 0;
315 u_int32_t mode, wmode;
316 struct nfsnode *np = VTONFS(vp);
317 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
318 int v3 = NFS_ISV3(vp);
319
320 lwkt_gettoken(&nmp->nm_token);
321
322 /*
323 * Disallow write attempts on filesystems mounted read-only;
324 * unless the file is a socket, fifo, or a block or character
325 * device resident on the filesystem.
326 */
327 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
328 switch (vp->v_type) {
329 case VREG:
330 case VDIR:
331 case VLNK:
332 lwkt_reltoken(&nmp->nm_token);
333 return (EROFS);
334 default:
335 break;
336 }
337 }
338
339 /*
340 * The NFS protocol passes only the effective uid/gid over the wire but
341 * we need to check access against real ids if AT_EACCESS not set.
342 * Handle this case by cloning the credentials and setting the
343 * effective ids to the real ones.
344 *
345 * The crdup() here can cause a lot of ucred structures to build-up
346 * (up to maxvnodes), so do our best to avoid it.
347 */
348 if (ap->a_flags & AT_EACCESS) {
349 cred = crhold(ap->a_cred);
350 } else {
351 cred = ap->a_cred;
352 if (cred->cr_uid == cred->cr_ruid &&
353 cred->cr_gid == cred->cr_rgid) {
354 cred = crhold(ap->a_cred);
355 } else {
356 cred = crdup(ap->a_cred);
357 cred->cr_uid = cred->cr_ruid;
358 cred->cr_gid = cred->cr_rgid;
359 }
360 }
361
362 /*
363 * For nfs v3, check to see if we have done this recently, and if
364 * so return our cached result instead of making an ACCESS call.
365 * If not, do an access rpc, otherwise you are stuck emulating
366 * ufs_access() locally using the vattr. This may not be correct,
367 * since the server may apply other access criteria such as
368 * client uid-->server uid mapping that we do not know about.
369 */
370 if (v3) {
371 if (ap->a_mode & VREAD)
372 mode = NFSV3ACCESS_READ;
373 else
374 mode = 0;
375 if (vp->v_type != VDIR) {
376 if (ap->a_mode & VWRITE)
377 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
378 if (ap->a_mode & VEXEC)
379 mode |= NFSV3ACCESS_EXECUTE;
380 } else {
381 if (ap->a_mode & VWRITE)
382 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
383 NFSV3ACCESS_DELETE);
384 if (ap->a_mode & VEXEC)
385 mode |= NFSV3ACCESS_LOOKUP;
386 }
387 /* XXX safety belt, only make blanket request if caching */
388 if (nfsaccess_cache_timeout > 0) {
389 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
390 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
391 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
392 } else {
393 wmode = mode;
394 }
395
396 /*
397 * Does our cached result allow us to give a definite yes to
398 * this request?
399 */
400 if (np->n_modestamp &&
401 (mycpu->gd_time_seconds < (np->n_modestamp + nfsaccess_cache_timeout)) &&
402 (cred->cr_uid == np->n_modeuid) &&
403 ((np->n_mode & mode) == mode)) {
404 nfsstats.accesscache_hits++;
405 } else {
406 /*
407 * Either a no, or a don't know. Go to the wire.
408 */
409 nfsstats.accesscache_misses++;
410 error = nfs3_access_otw(vp, wmode, td, cred);
411 if (!error) {
412 if ((np->n_mode & mode) != mode) {
413 error = EACCES;
414 }
415 }
416 }
417 } else {
418 if ((error = nfs_laccess(ap)) != 0) {
419 crfree(cred);
420 lwkt_reltoken(&nmp->nm_token);
421 return (error);
422 }
423
424 /*
425 * Attempt to prevent a mapped root from accessing a file
426 * which it shouldn't. We try to read a byte from the file
427 * if the user is root and the file is not zero length.
428 * After calling nfs_laccess, we should have the correct
429 * file size cached.
430 */
431 if (cred->cr_uid == 0 && (ap->a_mode & VREAD)
432 && VTONFS(vp)->n_size > 0) {
433 struct iovec aiov;
434 struct uio auio;
435 char buf[1];
436
437 aiov.iov_base = buf;
438 aiov.iov_len = 1;
439 auio.uio_iov = &aiov;
440 auio.uio_iovcnt = 1;
441 auio.uio_offset = 0;
442 auio.uio_resid = 1;
443 auio.uio_segflg = UIO_SYSSPACE;
444 auio.uio_rw = UIO_READ;
445 auio.uio_td = td;
446
447 if (vp->v_type == VREG) {
448 error = nfs_readrpc_uio(vp, &auio);
449 } else if (vp->v_type == VDIR) {
450 char* bp;
451 bp = kmalloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
452 aiov.iov_base = bp;
453 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
454 error = nfs_readdirrpc_uio(vp, &auio);
455 kfree(bp, M_TEMP);
456 } else if (vp->v_type == VLNK) {
457 error = nfs_readlinkrpc_uio(vp, &auio);
458 } else {
459 error = EACCES;
460 }
461 }
462 }
463 /*
464 * [re]record creds for reading and/or writing if access
465 * was granted. Assume the NFS server will grant read access
466 * for execute requests.
467 */
468 if (error == 0) {
469 if ((ap->a_mode & (VREAD|VEXEC)) &&
470 !nfs_crsame(cred, np->n_rucred)) {
471 ncred = nfs_crhold(cred);
472 ocred = np->n_rucred;
473 np->n_rucred = ncred;
474 if (ocred)
475 crfree(ocred);
476 }
477 if ((ap->a_mode & VWRITE) &&
478 !nfs_crsame(cred, np->n_wucred)) {
479 ncred = nfs_crhold(cred);
480 ocred = np->n_wucred;
481 np->n_wucred = ncred;
482 if (ocred)
483 crfree(ocred);
484 }
485 }
486 lwkt_reltoken(&nmp->nm_token);
487 crfree(cred);
488
489 return(error);
490 }
491
492 /*
493 * nfs open vnode op
494 * Check to see if the type is ok
495 * and that deletion is not in progress.
496 * For paged in text files, you will need to flush the page cache
497 * if consistency is lost.
498 *
499 * nfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
500 * struct file *a_fp)
501 */
502 /* ARGSUSED */
503 static int
nfs_open(struct vop_open_args * ap)504 nfs_open(struct vop_open_args *ap)
505 {
506 struct vnode *vp = ap->a_vp;
507 struct nfsnode *np = VTONFS(vp);
508 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
509 struct vattr vattr;
510 struct ucred *ncred;
511 struct ucred *ocred;
512 int error;
513
514 lwkt_gettoken(&nmp->nm_token);
515
516 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
517 #ifdef DIAGNOSTIC
518 kprintf("open eacces vtyp=%d\n",vp->v_type);
519 #endif
520 lwkt_reltoken(&nmp->nm_token);
521 return (EOPNOTSUPP);
522 }
523
524 /*
525 * Save valid creds for reading and writing for later RPCs.
526 */
527 if ((ap->a_mode & FREAD) && !nfs_crsame(ap->a_cred, np->n_rucred)) {
528 ncred = nfs_crhold(ap->a_cred);
529 ocred = np->n_rucred;
530 np->n_rucred = ncred;
531 if (ocred)
532 crfree(ocred);
533 }
534 if ((ap->a_mode & FWRITE) && !nfs_crsame(ap->a_cred, np->n_wucred)) {
535 ncred = nfs_crhold(ap->a_cred);
536 ocred = np->n_wucred;
537 np->n_wucred = ncred;
538 if (ocred)
539 crfree(ocred);
540 }
541
542 /*
543 * Clear the attribute cache only if opening with write access. It
544 * is unclear if we should do this at all here, but we certainly
545 * should not clear the cache unconditionally simply because a file
546 * is being opened.
547 */
548 if (ap->a_mode & FWRITE)
549 np->n_attrstamp = 0;
550
551 /*
552 * For normal NFS, reconcile changes made locally verses
553 * changes made remotely. Note that VOP_GETATTR only goes
554 * to the wire if the cached attribute has timed out or been
555 * cleared.
556 *
557 * If local modifications have been made clear the attribute
558 * cache to force an attribute and modified time check. If
559 * GETATTR detects that the file has been changed by someone
560 * other then us it will set NRMODIFIED.
561 *
562 * If we are opening a directory and local changes have been
563 * made we have to invalidate the cache in order to ensure
564 * that we get the most up-to-date information from the
565 * server. XXX
566 */
567 if (np->n_flag & NLMODIFIED) {
568 np->n_attrstamp = 0;
569 if (vp->v_type == VDIR) {
570 error = nfs_vinvalbuf(vp, V_SAVE, 1);
571 if (error == EINTR) {
572 lwkt_reltoken(&nmp->nm_token);
573 return (error);
574 }
575 nfs_invaldir(vp);
576 }
577 }
578 error = VOP_GETATTR(vp, &vattr);
579 if (error) {
580 lwkt_reltoken(&nmp->nm_token);
581 return (error);
582 }
583 if (np->n_flag & NRMODIFIED) {
584 if (vp->v_type == VDIR)
585 nfs_invaldir(vp);
586 error = nfs_vinvalbuf(vp, V_SAVE, 1);
587 if (error == EINTR) {
588 lwkt_reltoken(&nmp->nm_token);
589 return (error);
590 }
591 np->n_flag &= ~NRMODIFIED;
592 }
593 error = vop_stdopen(ap);
594 lwkt_reltoken(&nmp->nm_token);
595
596 return error;
597 }
598
599 /*
600 * nfs close vnode op
601 * What an NFS client should do upon close after writing is a debatable issue.
602 * Most NFS clients push delayed writes to the server upon close, basically for
603 * two reasons:
604 * 1 - So that any write errors may be reported back to the client process
605 * doing the close system call. By far the two most likely errors are
606 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
607 * 2 - To put a worst case upper bound on cache inconsistency between
608 * multiple clients for the file.
609 * There is also a consistency problem for Version 2 of the protocol w.r.t.
610 * not being able to tell if other clients are writing a file concurrently,
611 * since there is no way of knowing if the changed modify time in the reply
612 * is only due to the write for this client.
613 * (NFS Version 3 provides weak cache consistency data in the reply that
614 * should be sufficient to detect and handle this case.)
615 *
616 * The current code does the following:
617 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
618 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
619 * or commit them (this satisfies 1 and 2 except for the
620 * case where the server crashes after this close but
621 * before the commit RPC, which is felt to be "good
622 * enough". Changing the last argument to nfs_flush() to
623 * a 1 would force a commit operation, if it is felt a
624 * commit is necessary now.
625 * for NQNFS - do nothing now, since 2 is dealt with via leases and
626 * 1 should be dealt with via an fsync() system call for
627 * cases where write errors are important.
628 *
629 * nfs_close(struct vnode *a_vp, int a_fflag)
630 */
631 /* ARGSUSED */
632 static int
nfs_close(struct vop_close_args * ap)633 nfs_close(struct vop_close_args *ap)
634 {
635 struct vnode *vp = ap->a_vp;
636 struct nfsnode *np = VTONFS(vp);
637 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
638 int error = 0;
639 thread_t td = curthread;
640
641 vn_lock(vp, LK_UPGRADE | LK_RETRY); /* XXX */
642 lwkt_gettoken(&nmp->nm_token);
643
644 if (vp->v_type == VREG) {
645 if (np->n_flag & NLMODIFIED) {
646 if (NFS_ISV3(vp)) {
647 /*
648 * Under NFSv3 we have dirty buffers to dispose of. We
649 * must flush them to the NFS server. We have the option
650 * of waiting all the way through the commit rpc or just
651 * waiting for the initial write. The default is to only
652 * wait through the initial write so the data is in the
653 * server's cache, which is roughly similar to the state
654 * a standard disk subsystem leaves the file in on close().
655 *
656 * We cannot clear the NLMODIFIED bit in np->n_flag due to
657 * potential races with other processes, and certainly
658 * cannot clear it if we don't commit.
659 */
660 int cm = nfsv3_commit_on_close ? 1 : 0;
661 error = nfs_flush(vp, MNT_WAIT, td, cm);
662 /* np->n_flag &= ~NLMODIFIED; */
663 } else {
664 error = nfs_vinvalbuf(vp, V_SAVE, 1);
665 }
666 np->n_attrstamp = 0;
667 }
668 if (np->n_flag & NWRITEERR) {
669 np->n_flag &= ~NWRITEERR;
670 error = np->n_error;
671 }
672 }
673 vop_stdclose(ap);
674 lwkt_reltoken(&nmp->nm_token);
675
676 return (error);
677 }
678
679 /*
680 * nfs getattr call from vfs.
681 *
682 * nfs_getattr(struct vnode *a_vp, struct vattr *a_vap)
683 */
684 static int
nfs_getattr(struct vop_getattr_args * ap)685 nfs_getattr(struct vop_getattr_args *ap)
686 {
687 struct vnode *vp = ap->a_vp;
688 struct nfsnode *np = VTONFS(vp);
689 struct nfsmount *nmp;
690 int error = 0;
691 thread_t td = curthread;
692 struct nfsm_info info;
693
694 info.mrep = NULL;
695 info.v3 = NFS_ISV3(vp);
696 nmp = VFSTONFS(vp->v_mount);
697
698 lwkt_gettoken(&nmp->nm_token);
699
700 /*
701 * Update local times for special files.
702 */
703 if (np->n_flag & (NACC | NUPD))
704 np->n_flag |= NCHG;
705 /*
706 * First look in the cache.
707 */
708 if (nfs_getattrcache(vp, ap->a_vap) == 0)
709 goto done;
710
711 if (info.v3 && nfsaccess_cache_timeout > 0) {
712 nfsstats.accesscache_misses++;
713 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, nfs_vpcred(vp, ND_CHECK));
714 if (nfs_getattrcache(vp, ap->a_vap) == 0)
715 goto done;
716 }
717
718 nfsstats.rpccnt[NFSPROC_GETATTR]++;
719 nfsm_reqhead(&info, vp, NFSPROC_GETATTR, NFSX_FH(info.v3));
720 ERROROUT(nfsm_fhtom(&info, vp));
721 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_GETATTR, td,
722 nfs_vpcred(vp, ND_CHECK), &error));
723 if (error == 0) {
724 ERROROUT(nfsm_loadattr(&info, vp, ap->a_vap));
725 }
726 m_freem(info.mrep);
727 info.mrep = NULL;
728 done:
729 /*
730 * NFS doesn't support chflags flags. If the nfs mount was
731 * made -o cache set the UF_CACHE bit for swapcache.
732 */
733 if ((nmp->nm_flag & NFSMNT_CACHE) && (vp->v_flag & VROOT))
734 ap->a_vap->va_flags |= UF_CACHE;
735 nfsmout:
736 lwkt_reltoken(&nmp->nm_token);
737 return (error);
738 }
739
740 /*
741 * nfs setattr call.
742 *
743 * nfs_setattr(struct vnode *a_vp, struct vattr *a_vap, struct ucred *a_cred)
744 */
745 static int
nfs_setattr(struct vop_setattr_args * ap)746 nfs_setattr(struct vop_setattr_args *ap)
747 {
748 struct vnode *vp = ap->a_vp;
749 struct nfsnode *np = VTONFS(vp);
750 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
751 struct vattr *vap = ap->a_vap;
752 int error = 0;
753 int kflags = 0;
754 off_t tsize;
755 thread_t td = curthread;
756
757 #ifndef nolint
758 tsize = (off_t)0;
759 #endif
760 /*
761 * Setting of flags is not supported.
762 */
763 if (vap->va_flags != VNOVAL)
764 return (EOPNOTSUPP);
765
766 /*
767 * Disallow write attempts if the filesystem is mounted read-only.
768 */
769 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
770 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
771 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
772 (vp->v_mount->mnt_flag & MNT_RDONLY))
773 return (EROFS);
774
775 lwkt_gettoken(&nmp->nm_token);
776
777 /*
778 * Handle size changes
779 */
780 if (vap->va_size != VNOVAL) {
781 /*
782 * truncation requested
783 */
784 switch (vp->v_type) {
785 case VDIR:
786 lwkt_reltoken(&nmp->nm_token);
787 return (EISDIR);
788 case VCHR:
789 case VBLK:
790 case VSOCK:
791 case VFIFO:
792 if (vap->va_mtime.tv_sec == VNOVAL &&
793 vap->va_atime.tv_sec == VNOVAL &&
794 vap->va_mode == (mode_t)VNOVAL &&
795 vap->va_uid == (uid_t)VNOVAL &&
796 vap->va_gid == (gid_t)VNOVAL) {
797 lwkt_reltoken(&nmp->nm_token);
798 return (0);
799 }
800 vap->va_size = VNOVAL;
801 break;
802 default:
803 /*
804 * Disallow write attempts if the filesystem is
805 * mounted read-only.
806 */
807 if (vp->v_mount->mnt_flag & MNT_RDONLY) {
808 lwkt_reltoken(&nmp->nm_token);
809 return (EROFS);
810 }
811
812 tsize = np->n_size;
813 again:
814 error = nfs_meta_setsize(vp, td, vap->va_size, 0);
815
816 #if 0
817 if (np->n_flag & NLMODIFIED) {
818 if (vap->va_size == 0)
819 error = nfs_vinvalbuf(vp, 0, 1);
820 else
821 error = nfs_vinvalbuf(vp, V_SAVE, 1);
822 }
823 #endif
824 /*
825 * note: this loop case almost always happens at
826 * least once per truncation.
827 */
828 if (error == 0 && np->n_size != vap->va_size)
829 goto again;
830 np->n_vattr.va_size = vap->va_size;
831 kflags |= NOTE_WRITE;
832 if (tsize < vap->va_size)
833 kflags |= NOTE_EXTEND;
834 break;
835 }
836 }
837
838 /*
839 * If setting the mtime or if server/other-client modifications have
840 * been detected, we must fully flush any pending writes.
841 *
842 * This will slow down cp/cpdup/rdist/rsync and other operations which
843 * might call [l]utimes() to set the mtime after writing to a file,
844 * but honestly there is no way to properly defer the write flush
845 * and still get reasonably accurate/dependable synchronization of
846 * [l]utimes().
847 */
848 if ((np->n_flag & NLMODIFIED) && vp->v_type == VREG) {
849 if ((np->n_flag & NRMODIFIED) ||
850 (vap->va_mtime.tv_sec != VNOVAL)) {
851 error = nfs_vinvalbuf(vp, V_SAVE, 1);
852 if (error == EINTR) {
853 lwkt_reltoken(&nmp->nm_token);
854 return (error);
855 }
856 }
857 }
858
859 /*
860 * Get the blasted mtime to report properly.
861 */
862 if (vap->va_mtime.tv_sec != VNOVAL) {
863 np->n_mtime = vap->va_mtime.tv_sec;
864 np->n_flag &= ~NUPD;
865 np->n_vattr.va_mtime = vap->va_mtime;
866 }
867
868 /*
869 * Issue the setattr rpc, adjust our mtime and make sure NUPD
870 * has been cleared so it does not get overridden.
871 */
872 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
873 if (error == 0)
874 kflags |= NOTE_EXTEND;
875
876 /*
877 * Sanity check if a truncation was issued. This should only occur
878 * if multiple processes are racing on the same file.
879 */
880 if (error == 0 && vap->va_size != VNOVAL &&
881 np->n_size != vap->va_size) {
882 kprintf("NFS ftruncate: server disagrees on the file size: "
883 "%jd/%jd/%jd\n",
884 (intmax_t)tsize,
885 (intmax_t)vap->va_size,
886 (intmax_t)np->n_size);
887 goto again;
888 }
889 if (error && vap->va_size != VNOVAL) {
890 np->n_size = np->n_vattr.va_size = tsize;
891 nfs_meta_setsize(vp, td, np->n_size, 0);
892 }
893 lwkt_reltoken(&nmp->nm_token);
894 nfs_knote(vp, kflags);
895
896 return (error);
897 }
898
899 /*
900 * Do an nfs setattr rpc.
901 */
902 static int
nfs_setattrrpc(struct vnode * vp,struct vattr * vap,struct ucred * cred,struct thread * td)903 nfs_setattrrpc(struct vnode *vp, struct vattr *vap,
904 struct ucred *cred, struct thread *td)
905 {
906 struct nfsv2_sattr *sp;
907 struct nfsnode *np = VTONFS(vp);
908 u_int32_t *tl;
909 int error = 0, wccflag = NFSV3_WCCRATTR;
910 struct nfsm_info info;
911
912 info.mrep = NULL;
913 info.v3 = NFS_ISV3(vp);
914
915 nfsstats.rpccnt[NFSPROC_SETATTR]++;
916 nfsm_reqhead(&info, vp, NFSPROC_SETATTR,
917 NFSX_FH(info.v3) + NFSX_SATTR(info.v3));
918 ERROROUT(nfsm_fhtom(&info, vp));
919 if (info.v3) {
920 nfsm_v3attrbuild(&info, vap, TRUE);
921 tl = nfsm_build(&info, NFSX_UNSIGNED);
922 *tl = nfs_false;
923 } else {
924 sp = nfsm_build(&info, NFSX_V2SATTR);
925 if (vap->va_mode == (mode_t)VNOVAL)
926 sp->sa_mode = nfs_xdrneg1;
927 else
928 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
929 if (vap->va_uid == (uid_t)VNOVAL)
930 sp->sa_uid = nfs_xdrneg1;
931 else
932 sp->sa_uid = txdr_unsigned(vap->va_uid);
933 if (vap->va_gid == (gid_t)VNOVAL)
934 sp->sa_gid = nfs_xdrneg1;
935 else
936 sp->sa_gid = txdr_unsigned(vap->va_gid);
937 sp->sa_size = txdr_unsigned(vap->va_size);
938 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
939 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
940 }
941 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_SETATTR, td, cred, &error));
942 if (info.v3) {
943 np->n_modestamp = 0;
944 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
945 } else {
946 ERROROUT(nfsm_loadattr(&info, vp, NULL));
947 }
948 m_freem(info.mrep);
949 info.mrep = NULL;
950 nfsmout:
951 return (error);
952 }
953
954 static
955 void
nfs_cache_setvp(struct nchandle * nch,struct vnode * vp,int nctimeout)956 nfs_cache_setvp(struct nchandle *nch, struct vnode *vp, int nctimeout)
957 {
958 if (nctimeout == 0)
959 nctimeout = 1;
960 else
961 nctimeout *= hz;
962 cache_setvp(nch, vp);
963 cache_settimeout(nch, nctimeout);
964 }
965
966 /*
967 * NEW API CALL - replaces nfs_lookup(). However, we cannot remove
968 * nfs_lookup() until all remaining new api calls are implemented.
969 *
970 * Resolve a namecache entry. This function is passed a locked ncp and
971 * must call nfs_cache_setvp() on it as appropriate to resolve the entry.
972 */
973 static int
nfs_nresolve(struct vop_nresolve_args * ap)974 nfs_nresolve(struct vop_nresolve_args *ap)
975 {
976 struct thread *td = curthread;
977 struct namecache *ncp;
978 struct nfsmount *nmp;
979 struct nfsnode *np;
980 struct vnode *dvp;
981 struct vnode *nvp;
982 nfsfh_t *fhp;
983 int attrflag;
984 int fhsize;
985 int error;
986 int tmp_error;
987 int len;
988 struct nfsm_info info;
989
990 dvp = ap->a_dvp;
991 nmp = VFSTONFS(dvp->v_mount);
992
993 lwkt_gettoken(&nmp->nm_token);
994
995 if ((error = vget(dvp, LK_SHARED)) != 0) {
996 lwkt_reltoken(&nmp->nm_token);
997 return (error);
998 }
999
1000 info.mrep = NULL;
1001 info.v3 = NFS_ISV3(dvp);
1002
1003 nvp = NULL;
1004 nfsstats.lookupcache_misses++;
1005 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1006 ncp = ap->a_nch->ncp;
1007 len = ncp->nc_nlen;
1008 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1009 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1010 ERROROUT(nfsm_fhtom(&info, dvp));
1011 ERROROUT(nfsm_strtom(&info, ncp->nc_name, len, NFS_MAXNAMLEN));
1012 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td,
1013 ap->a_cred, &error));
1014 if (error) {
1015 /*
1016 * Cache negatve lookups to reduce NFS traffic, but use
1017 * a fast timeout. Otherwise use a timeout of 1 tick.
1018 * XXX we should add a namecache flag for no-caching
1019 * to uncache the negative hit as soon as possible, but
1020 * we cannot simply destroy the entry because it is used
1021 * as a placeholder by the caller.
1022 *
1023 * The refactored nfs code will overwrite a non-zero error
1024 * with 0 when we use ERROROUT(), so don't here.
1025 */
1026 if (error == ENOENT)
1027 nfs_cache_setvp(ap->a_nch, NULL, nfsneg_cache_timeout);
1028 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1029 NFS_LATTR_NOSHRINK);
1030 if (tmp_error) {
1031 error = tmp_error;
1032 goto nfsmout;
1033 }
1034 m_freem(info.mrep);
1035 info.mrep = NULL;
1036 goto nfsmout;
1037 }
1038
1039 /*
1040 * Success, get the file handle, do various checks, and load
1041 * post-operation data from the reply packet. Theoretically
1042 * we should never be looking up "." so, theoretically, we
1043 * should never get the same file handle as our directory. But
1044 * we check anyway. XXX
1045 *
1046 * Note that no timeout is set for the positive cache hit. We
1047 * assume, theoretically, that ESTALE returns will be dealt with
1048 * properly to handle NFS races and in anycase we cannot depend
1049 * on a timeout to deal with NFS open/create/excl issues so instead
1050 * of a bad hack here the rest of the NFS client code needs to do
1051 * the right thing.
1052 */
1053 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1054
1055 np = VTONFS(dvp);
1056 if (NFS_CMPFH(np, fhp, fhsize)) {
1057 vref(dvp);
1058 nvp = dvp;
1059 } else {
1060 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, NULL);
1061 if (error) {
1062 m_freem(info.mrep);
1063 info.mrep = NULL;
1064 vput(dvp);
1065 lwkt_reltoken(&nmp->nm_token);
1066 return (error);
1067 }
1068 nvp = NFSTOV(np);
1069 }
1070 if (info.v3) {
1071 ERROROUT(nfsm_postop_attr(&info, nvp, &attrflag,
1072 NFS_LATTR_NOSHRINK));
1073 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1074 NFS_LATTR_NOSHRINK));
1075 } else {
1076 ERROROUT(nfsm_loadattr(&info, nvp, NULL));
1077 }
1078 nfs_cache_setvp(ap->a_nch, nvp, nfspos_cache_timeout);
1079 m_freem(info.mrep);
1080 info.mrep = NULL;
1081 nfsmout:
1082 lwkt_reltoken(&nmp->nm_token);
1083 vput(dvp);
1084 if (nvp) {
1085 if (nvp == dvp)
1086 vrele(nvp);
1087 else
1088 vput(nvp);
1089 }
1090 return (error);
1091 }
1092
1093 /*
1094 * 'cached' nfs directory lookup
1095 *
1096 * NOTE: cannot be removed until NFS implements all the new n*() API calls.
1097 *
1098 * nfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
1099 * struct componentname *a_cnp)
1100 */
1101 static int
nfs_lookup(struct vop_old_lookup_args * ap)1102 nfs_lookup(struct vop_old_lookup_args *ap)
1103 {
1104 struct componentname *cnp = ap->a_cnp;
1105 struct vnode *dvp = ap->a_dvp;
1106 struct vnode **vpp = ap->a_vpp;
1107 int flags = cnp->cn_flags;
1108 struct vnode *newvp;
1109 struct vnode *notvp;
1110 struct nfsmount *nmp;
1111 long len;
1112 nfsfh_t *fhp;
1113 struct nfsnode *np;
1114 int lockparent, wantparent, attrflag, fhsize;
1115 int error;
1116 int tmp_error;
1117 struct nfsm_info info;
1118
1119 info.mrep = NULL;
1120 info.v3 = NFS_ISV3(dvp);
1121 error = 0;
1122
1123 notvp = (cnp->cn_flags & CNP_NOTVP) ? cnp->cn_notvp : NULL;
1124
1125 /*
1126 * Read-only mount check and directory check.
1127 */
1128 *vpp = NULLVP;
1129 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1130 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME))
1131 return (EROFS);
1132
1133 if (dvp->v_type != VDIR)
1134 return (ENOTDIR);
1135
1136 /*
1137 * Look it up in the cache. Note that ENOENT is only returned if we
1138 * previously entered a negative hit (see later on). The additional
1139 * nfsneg_cache_timeout check causes previously cached results to
1140 * be instantly ignored if the negative caching is turned off.
1141 */
1142 lockparent = flags & CNP_LOCKPARENT;
1143 wantparent = flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
1144 nmp = VFSTONFS(dvp->v_mount);
1145 np = VTONFS(dvp);
1146
1147 lwkt_gettoken(&nmp->nm_token);
1148
1149 /*
1150 * Go to the wire.
1151 */
1152 error = 0;
1153 newvp = NULLVP;
1154 nfsstats.lookupcache_misses++;
1155 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1156 len = cnp->cn_namelen;
1157 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
1158 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
1159 ERROROUT(nfsm_fhtom(&info, dvp));
1160 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
1161 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, cnp->cn_td,
1162 cnp->cn_cred, &error));
1163 if (error) {
1164 tmp_error = nfsm_postop_attr(&info, dvp, &attrflag,
1165 NFS_LATTR_NOSHRINK);
1166 if (tmp_error) {
1167 error = tmp_error;
1168 goto nfsmout;
1169 }
1170
1171 m_freem(info.mrep);
1172 info.mrep = NULL;
1173 goto nfsmout;
1174 }
1175 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
1176
1177 /*
1178 * Handle RENAME case...
1179 */
1180 if (cnp->cn_nameiop == NAMEI_RENAME && wantparent) {
1181 if (NFS_CMPFH(np, fhp, fhsize)) {
1182 m_freem(info.mrep);
1183 info.mrep = NULL;
1184 lwkt_reltoken(&nmp->nm_token);
1185 return (EISDIR);
1186 }
1187 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp);
1188 if (error) {
1189 m_freem(info.mrep);
1190 info.mrep = NULL;
1191 lwkt_reltoken(&nmp->nm_token);
1192 return (error);
1193 }
1194 newvp = NFSTOV(np);
1195 if (info.v3) {
1196 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1197 NFS_LATTR_NOSHRINK));
1198 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1199 NFS_LATTR_NOSHRINK));
1200 } else {
1201 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1202 }
1203 *vpp = newvp;
1204 m_freem(info.mrep);
1205 info.mrep = NULL;
1206 if (!lockparent) {
1207 vn_unlock(dvp);
1208 cnp->cn_flags |= CNP_PDIRUNLOCK;
1209 }
1210 lwkt_reltoken(&nmp->nm_token);
1211 return (0);
1212 }
1213
1214 if (flags & CNP_ISDOTDOT) {
1215 vn_unlock(dvp);
1216 cnp->cn_flags |= CNP_PDIRUNLOCK;
1217 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp);
1218 if (error) {
1219 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1220 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
1221 lwkt_reltoken(&nmp->nm_token);
1222 return (error); /* NOTE: return error from nget */
1223 }
1224 newvp = NFSTOV(np);
1225 if (lockparent) {
1226 error = vn_lock(dvp, LK_EXCLUSIVE | LK_FAILRECLAIM);
1227 if (error) {
1228 vput(newvp);
1229 lwkt_reltoken(&nmp->nm_token);
1230 return (error);
1231 }
1232 cnp->cn_flags |= CNP_PDIRUNLOCK;
1233 }
1234 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1235 vref(dvp);
1236 newvp = dvp;
1237 } else {
1238 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, notvp);
1239 if (error) {
1240 m_freem(info.mrep);
1241 info.mrep = NULL;
1242 lwkt_reltoken(&nmp->nm_token);
1243 return (error);
1244 }
1245 if (!lockparent) {
1246 vn_unlock(dvp);
1247 cnp->cn_flags |= CNP_PDIRUNLOCK;
1248 }
1249 newvp = NFSTOV(np);
1250 }
1251 if (info.v3) {
1252 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
1253 NFS_LATTR_NOSHRINK));
1254 ERROROUT(nfsm_postop_attr(&info, dvp, &attrflag,
1255 NFS_LATTR_NOSHRINK));
1256 } else {
1257 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
1258 }
1259 #if 0
1260 /* XXX MOVE TO nfs_nremove() */
1261 if ((cnp->cn_flags & CNP_MAKEENTRY) &&
1262 cnp->cn_nameiop != NAMEI_DELETE) {
1263 np->n_ctime = np->n_vattr.va_ctime.tv_sec; /* XXX */
1264 }
1265 #endif
1266 *vpp = newvp;
1267 m_freem(info.mrep);
1268 info.mrep = NULL;
1269 nfsmout:
1270 if (error) {
1271 if (newvp != NULLVP) {
1272 vrele(newvp);
1273 *vpp = NULLVP;
1274 }
1275 if ((cnp->cn_nameiop == NAMEI_CREATE ||
1276 cnp->cn_nameiop == NAMEI_RENAME) &&
1277 error == ENOENT) {
1278 if (!lockparent) {
1279 vn_unlock(dvp);
1280 cnp->cn_flags |= CNP_PDIRUNLOCK;
1281 }
1282 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1283 error = EROFS;
1284 else
1285 error = EJUSTRETURN;
1286 }
1287 }
1288 lwkt_reltoken(&nmp->nm_token);
1289 return (error);
1290 }
1291
1292 /*
1293 * nfs read call.
1294 * Just call nfs_bioread() to do the work.
1295 *
1296 * nfs_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1297 * struct ucred *a_cred)
1298 */
1299 static int
nfs_read(struct vop_read_args * ap)1300 nfs_read(struct vop_read_args *ap)
1301 {
1302 struct vnode *vp = ap->a_vp;
1303 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1304 int error;
1305
1306 lwkt_gettoken(&nmp->nm_token);
1307 error = nfs_bioread(vp, ap->a_uio, ap->a_ioflag);
1308 lwkt_reltoken(&nmp->nm_token);
1309
1310 return error;
1311 }
1312
1313 /*
1314 * nfs readlink call
1315 *
1316 * nfs_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
1317 */
1318 static int
nfs_readlink(struct vop_readlink_args * ap)1319 nfs_readlink(struct vop_readlink_args *ap)
1320 {
1321 struct vnode *vp = ap->a_vp;
1322 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1323 int error;
1324
1325 if (vp->v_type != VLNK)
1326 return (EINVAL);
1327
1328 lwkt_gettoken(&nmp->nm_token);
1329 error = nfs_bioread(vp, ap->a_uio, 0);
1330 lwkt_reltoken(&nmp->nm_token);
1331
1332 return error;
1333 }
1334
1335 /*
1336 * Do a readlink rpc.
1337 * Called by nfs_doio() from below the buffer cache.
1338 */
1339 int
nfs_readlinkrpc_uio(struct vnode * vp,struct uio * uiop)1340 nfs_readlinkrpc_uio(struct vnode *vp, struct uio *uiop)
1341 {
1342 int error = 0, len, attrflag;
1343 struct nfsm_info info;
1344
1345 info.mrep = NULL;
1346 info.v3 = NFS_ISV3(vp);
1347
1348 nfsstats.rpccnt[NFSPROC_READLINK]++;
1349 nfsm_reqhead(&info, vp, NFSPROC_READLINK, NFSX_FH(info.v3));
1350 ERROROUT(nfsm_fhtom(&info, vp));
1351 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READLINK, uiop->uio_td,
1352 nfs_vpcred(vp, ND_CHECK), &error));
1353 if (info.v3) {
1354 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1355 NFS_LATTR_NOSHRINK));
1356 }
1357 if (!error) {
1358 NEGATIVEOUT(len = nfsm_strsiz(&info, NFS_MAXPATHLEN));
1359 if (len == NFS_MAXPATHLEN) {
1360 struct nfsnode *np = VTONFS(vp);
1361 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1362 len = np->n_size;
1363 }
1364 ERROROUT(nfsm_mtouio(&info, uiop, len));
1365 }
1366 m_freem(info.mrep);
1367 info.mrep = NULL;
1368 nfsmout:
1369 return (error);
1370 }
1371
1372 /*
1373 * nfs synchronous read rpc using UIO
1374 */
1375 int
nfs_readrpc_uio(struct vnode * vp,struct uio * uiop)1376 nfs_readrpc_uio(struct vnode *vp, struct uio *uiop)
1377 {
1378 u_int32_t *tl;
1379 struct nfsmount *nmp;
1380 int error = 0, len, retlen, tsiz, eof, attrflag;
1381 struct nfsm_info info;
1382 off_t tmp_off;
1383
1384 info.mrep = NULL;
1385 info.v3 = NFS_ISV3(vp);
1386
1387 #ifndef nolint
1388 eof = 0;
1389 #endif
1390 nmp = VFSTONFS(vp->v_mount);
1391
1392 tsiz = uiop->uio_resid;
1393 tmp_off = uiop->uio_offset + tsiz;
1394 if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset)
1395 return (EFBIG);
1396 tmp_off = uiop->uio_offset;
1397 while (tsiz > 0) {
1398 nfsstats.rpccnt[NFSPROC_READ]++;
1399 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1400 nfsm_reqhead(&info, vp, NFSPROC_READ,
1401 NFSX_FH(info.v3) + NFSX_UNSIGNED * 3);
1402 ERROROUT(nfsm_fhtom(&info, vp));
1403 tl = nfsm_build(&info, NFSX_UNSIGNED * 3);
1404 if (info.v3) {
1405 txdr_hyper(uiop->uio_offset, tl);
1406 *(tl + 2) = txdr_unsigned(len);
1407 } else {
1408 *tl++ = txdr_unsigned(uiop->uio_offset);
1409 *tl++ = txdr_unsigned(len);
1410 *tl = 0;
1411 }
1412 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READ, uiop->uio_td,
1413 nfs_vpcred(vp, ND_READ), &error));
1414 if (info.v3) {
1415 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
1416 NFS_LATTR_NOSHRINK));
1417 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
1418 eof = fxdr_unsigned(int, *(tl + 1));
1419 } else {
1420 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1421 }
1422 NEGATIVEOUT(retlen = nfsm_strsiz(&info, len));
1423 ERROROUT(nfsm_mtouio(&info, uiop, retlen));
1424 m_freem(info.mrep);
1425 info.mrep = NULL;
1426
1427 /*
1428 * Handle short-read from server (NFSv3). If EOF is not
1429 * flagged (and no error occurred), but retlen is less
1430 * then the request size, we must zero-fill the remainder.
1431 */
1432 if (retlen < len && info.v3 && eof == 0) {
1433 ERROROUT(uiomovez(len - retlen, uiop));
1434 retlen = len;
1435 }
1436 tsiz -= retlen;
1437
1438 /*
1439 * Terminate loop on EOF or zero-length read.
1440 *
1441 * For NFSv2 a short-read indicates EOF, not zero-fill,
1442 * and also terminates the loop.
1443 */
1444 if (info.v3) {
1445 if (eof || retlen == 0)
1446 tsiz = 0;
1447 } else if (retlen < len) {
1448 tsiz = 0;
1449 }
1450 }
1451 nfsmout:
1452 return (error);
1453 }
1454
1455 /*
1456 * nfs write call
1457 */
1458 int
nfs_writerpc_uio(struct vnode * vp,struct uio * uiop,int * iomode,int * must_commit)1459 nfs_writerpc_uio(struct vnode *vp, struct uio *uiop,
1460 int *iomode, int *must_commit)
1461 {
1462 u_int32_t *tl;
1463 int32_t backup;
1464 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1465 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1466 int committed = NFSV3WRITE_FILESYNC;
1467 struct nfsm_info info;
1468
1469 info.mrep = NULL;
1470 info.v3 = NFS_ISV3(vp);
1471
1472 #ifndef DIAGNOSTIC
1473 if (uiop->uio_iovcnt != 1)
1474 panic("nfs: writerpc iovcnt > 1");
1475 #endif
1476 *must_commit = 0;
1477 tsiz = uiop->uio_resid;
1478 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1479 return (EFBIG);
1480 while (tsiz > 0) {
1481 nfsstats.rpccnt[NFSPROC_WRITE]++;
1482 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1483 nfsm_reqhead(&info, vp, NFSPROC_WRITE,
1484 NFSX_FH(info.v3) + 5 * NFSX_UNSIGNED +
1485 nfsm_rndup(len));
1486 ERROROUT(nfsm_fhtom(&info, vp));
1487 if (info.v3) {
1488 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
1489 txdr_hyper(uiop->uio_offset, tl);
1490 tl += 2;
1491 *tl++ = txdr_unsigned(len);
1492 *tl++ = txdr_unsigned(*iomode);
1493 *tl = txdr_unsigned(len);
1494 } else {
1495 u_int32_t x;
1496
1497 tl = nfsm_build(&info, 4 * NFSX_UNSIGNED);
1498 /* Set both "begin" and "current" to non-garbage. */
1499 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1500 *tl++ = x; /* "begin offset" */
1501 *tl++ = x; /* "current offset" */
1502 x = txdr_unsigned(len);
1503 *tl++ = x; /* total to this offset */
1504 *tl = x; /* size of this write */
1505 }
1506 ERROROUT(nfsm_uiotom(&info, uiop, len));
1507 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_WRITE, uiop->uio_td,
1508 nfs_vpcred(vp, ND_WRITE), &error));
1509 if (info.v3) {
1510 /*
1511 * The write RPC returns a before and after mtime. The
1512 * nfsm_wcc_data() macro checks the before n_mtime
1513 * against the before time and stores the after time
1514 * in the nfsnode's cached vattr and n_mtime field.
1515 * The NRMODIFIED bit will be set if the before
1516 * time did not match the original mtime.
1517 */
1518 wccflag = NFSV3_WCCCHK;
1519 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
1520 if (error == 0) {
1521 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED + NFSX_V3WRITEVERF));
1522 rlen = fxdr_unsigned(int, *tl++);
1523 if (rlen == 0) {
1524 error = NFSERR_IO;
1525 m_freem(info.mrep);
1526 info.mrep = NULL;
1527 break;
1528 } else if (rlen < len) {
1529 backup = len - rlen;
1530 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base - backup;
1531 uiop->uio_iov->iov_len += backup;
1532 uiop->uio_offset -= backup;
1533 uiop->uio_resid += backup;
1534 len = rlen;
1535 }
1536 commit = fxdr_unsigned(int, *tl++);
1537
1538 /*
1539 * Return the lowest committment level
1540 * obtained by any of the RPCs.
1541 */
1542 if (committed == NFSV3WRITE_FILESYNC)
1543 committed = commit;
1544 else if (committed == NFSV3WRITE_DATASYNC &&
1545 commit == NFSV3WRITE_UNSTABLE)
1546 committed = commit;
1547 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1548 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1549 NFSX_V3WRITEVERF);
1550 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1551 } else if (bcmp((caddr_t)tl,
1552 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1553 *must_commit = 1;
1554 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1555 NFSX_V3WRITEVERF);
1556 }
1557 }
1558 } else {
1559 ERROROUT(nfsm_loadattr(&info, vp, NULL));
1560 }
1561 m_freem(info.mrep);
1562 info.mrep = NULL;
1563 if (error)
1564 break;
1565 tsiz -= len;
1566 }
1567 nfsmout:
1568 if (vp->v_mount->mnt_flag & MNT_ASYNC)
1569 committed = NFSV3WRITE_FILESYNC;
1570 *iomode = committed;
1571 if (error)
1572 uiop->uio_resid = tsiz;
1573 return (error);
1574 }
1575
1576 /*
1577 * nfs mknod rpc
1578 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1579 * mode set to specify the file type and the size field for rdev.
1580 */
1581 static int
nfs_mknodrpc(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,struct vattr * vap)1582 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1583 struct vattr *vap)
1584 {
1585 struct nfsv2_sattr *sp;
1586 u_int32_t *tl;
1587 struct vnode *newvp = NULL;
1588 struct nfsnode *np = NULL;
1589 struct vattr vattr;
1590 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1591 int rmajor, rminor;
1592 struct nfsm_info info;
1593
1594 info.mrep = NULL;
1595 info.v3 = NFS_ISV3(dvp);
1596
1597 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1598 rmajor = txdr_unsigned(vap->va_rmajor);
1599 rminor = txdr_unsigned(vap->va_rminor);
1600 } else if (vap->va_type == VFIFO || vap->va_type == VSOCK) {
1601 rmajor = nfs_xdrneg1;
1602 rminor = nfs_xdrneg1;
1603 } else {
1604 return (EOPNOTSUPP);
1605 }
1606 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1607 return (error);
1608 }
1609 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1610 nfsm_reqhead(&info, dvp, NFSPROC_MKNOD,
1611 NFSX_FH(info.v3) + 4 * NFSX_UNSIGNED +
1612 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1613 ERROROUT(nfsm_fhtom(&info, dvp));
1614 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1615 NFS_MAXNAMLEN));
1616 if (info.v3) {
1617 tl = nfsm_build(&info, NFSX_UNSIGNED);
1618 *tl++ = vtonfsv3_type(vap->va_type);
1619 nfsm_v3attrbuild(&info, vap, FALSE);
1620 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1621 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
1622 *tl++ = txdr_unsigned(vap->va_rmajor);
1623 *tl = txdr_unsigned(vap->va_rminor);
1624 }
1625 } else {
1626 sp = nfsm_build(&info, NFSX_V2SATTR);
1627 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1628 sp->sa_uid = nfs_xdrneg1;
1629 sp->sa_gid = nfs_xdrneg1;
1630 sp->sa_size = makeudev(rmajor, rminor);
1631 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1632 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1633 }
1634 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKNOD, cnp->cn_td,
1635 cnp->cn_cred, &error));
1636 if (!error) {
1637 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1638 if (!gotvp) {
1639 if (newvp) {
1640 vput(newvp);
1641 newvp = NULL;
1642 }
1643 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1644 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1645 if (!error)
1646 newvp = NFSTOV(np);
1647 }
1648 }
1649 if (info.v3) {
1650 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1651 }
1652 m_freem(info.mrep);
1653 info.mrep = NULL;
1654 nfsmout:
1655 if (error) {
1656 if (newvp)
1657 vput(newvp);
1658 } else {
1659 *vpp = newvp;
1660 }
1661 VTONFS(dvp)->n_flag |= NLMODIFIED;
1662 if (!wccflag)
1663 VTONFS(dvp)->n_attrstamp = 0;
1664 return (error);
1665 }
1666
1667 /*
1668 * nfs mknod vop
1669 * just call nfs_mknodrpc() to do the work.
1670 *
1671 * nfs_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
1672 * struct componentname *a_cnp, struct vattr *a_vap)
1673 */
1674 /* ARGSUSED */
1675 static int
nfs_mknod(struct vop_old_mknod_args * ap)1676 nfs_mknod(struct vop_old_mknod_args *ap)
1677 {
1678 struct nfsmount *nmp = VFSTONFS(ap->a_dvp->v_mount);
1679 int error;
1680
1681 lwkt_gettoken(&nmp->nm_token);
1682 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1683 lwkt_reltoken(&nmp->nm_token);
1684 if (error == 0)
1685 nfs_knote(ap->a_dvp, NOTE_WRITE);
1686
1687 return error;
1688 }
1689
1690 static u_long create_verf;
1691 /*
1692 * nfs file create call
1693 *
1694 * nfs_create(struct vnode *a_dvp, struct vnode **a_vpp,
1695 * struct componentname *a_cnp, struct vattr *a_vap)
1696 */
1697 static int
nfs_create(struct vop_old_create_args * ap)1698 nfs_create(struct vop_old_create_args *ap)
1699 {
1700 struct vnode *dvp = ap->a_dvp;
1701 struct vattr *vap = ap->a_vap;
1702 struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1703 struct componentname *cnp = ap->a_cnp;
1704 struct nfsv2_sattr *sp;
1705 u_int32_t *tl;
1706 struct nfsnode *np = NULL;
1707 struct vnode *newvp = NULL;
1708 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1709 struct vattr vattr;
1710 struct nfsm_info info;
1711
1712 info.mrep = NULL;
1713 info.v3 = NFS_ISV3(dvp);
1714 lwkt_gettoken(&nmp->nm_token);
1715
1716 /*
1717 * Oops, not for me..
1718 */
1719 if (vap->va_type == VSOCK) {
1720 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1721 lwkt_reltoken(&nmp->nm_token);
1722 return error;
1723 }
1724
1725 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
1726 lwkt_reltoken(&nmp->nm_token);
1727 return (error);
1728 }
1729 if (vap->va_vaflags & VA_EXCLUSIVE)
1730 fmode |= O_EXCL;
1731 again:
1732 nfsstats.rpccnt[NFSPROC_CREATE]++;
1733 nfsm_reqhead(&info, dvp, NFSPROC_CREATE,
1734 NFSX_FH(info.v3) + 2 * NFSX_UNSIGNED +
1735 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(info.v3));
1736 ERROROUT(nfsm_fhtom(&info, dvp));
1737 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
1738 NFS_MAXNAMLEN));
1739 if (info.v3) {
1740 tl = nfsm_build(&info, NFSX_UNSIGNED);
1741 if (fmode & O_EXCL) {
1742 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1743 tl = nfsm_build(&info, NFSX_V3CREATEVERF);
1744 #ifdef INET
1745 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid]))
1746 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia)->sin_addr.s_addr;
1747 else
1748 #endif
1749 *tl++ = create_verf;
1750 *tl = ++create_verf;
1751 } else {
1752 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1753 nfsm_v3attrbuild(&info, vap, FALSE);
1754 }
1755 } else {
1756 sp = nfsm_build(&info, NFSX_V2SATTR);
1757 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1758 sp->sa_uid = nfs_xdrneg1;
1759 sp->sa_gid = nfs_xdrneg1;
1760 sp->sa_size = 0;
1761 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1762 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1763 }
1764 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_CREATE, cnp->cn_td,
1765 cnp->cn_cred, &error));
1766 if (error == 0) {
1767 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
1768 if (!gotvp) {
1769 if (newvp) {
1770 vput(newvp);
1771 newvp = NULL;
1772 }
1773 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1774 cnp->cn_namelen, cnp->cn_cred, cnp->cn_td, &np);
1775 if (!error)
1776 newvp = NFSTOV(np);
1777 }
1778 }
1779 if (info.v3) {
1780 if (error == 0)
1781 error = nfsm_wcc_data(&info, dvp, &wccflag);
1782 else
1783 (void)nfsm_wcc_data(&info, dvp, &wccflag);
1784 }
1785 m_freem(info.mrep);
1786 info.mrep = NULL;
1787 nfsmout:
1788 if (error) {
1789 if (info.v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1790 KKASSERT(newvp == NULL);
1791 fmode &= ~O_EXCL;
1792 goto again;
1793 }
1794 } else if (info.v3 && (fmode & O_EXCL)) {
1795 /*
1796 * We are normally called with only a partially initialized
1797 * VAP. Since the NFSv3 spec says that server may use the
1798 * file attributes to store the verifier, the spec requires
1799 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1800 * in atime, but we can't really assume that all servers will
1801 * so we ensure that our SETATTR sets both atime and mtime.
1802 */
1803 if (vap->va_mtime.tv_sec == VNOVAL)
1804 vfs_timestamp(&vap->va_mtime);
1805 if (vap->va_atime.tv_sec == VNOVAL)
1806 vap->va_atime = vap->va_mtime;
1807 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_td);
1808 }
1809 if (error == 0) {
1810 /*
1811 * The new np may have enough info for access
1812 * checks, make sure rucred and wucred are
1813 * initialized for read and write rpc's.
1814 */
1815 np = VTONFS(newvp);
1816 if (np->n_rucred == NULL)
1817 np->n_rucred = nfs_crhold(cnp->cn_cred);
1818 if (np->n_wucred == NULL)
1819 np->n_wucred = nfs_crhold(cnp->cn_cred);
1820 *ap->a_vpp = newvp;
1821 nfs_knote(dvp, NOTE_WRITE);
1822 } else if (newvp) {
1823 vput(newvp);
1824 }
1825 VTONFS(dvp)->n_flag |= NLMODIFIED;
1826 if (!wccflag)
1827 VTONFS(dvp)->n_attrstamp = 0;
1828 lwkt_reltoken(&nmp->nm_token);
1829 return (error);
1830 }
1831
1832 /*
1833 * nfs file remove call
1834 * To try and make nfs semantics closer to ufs semantics, a file that has
1835 * other processes using the vnode is renamed instead of removed and then
1836 * removed later on the last close.
1837 * - If v_refcnt > 1
1838 * If a rename is not already in the works
1839 * call nfs_sillyrename() to set it up
1840 * else
1841 * do the remove rpc
1842 *
1843 * nfs_remove(struct vnode *a_dvp, struct vnode *a_vp,
1844 * struct componentname *a_cnp)
1845 */
1846 static int
nfs_remove(struct vop_old_remove_args * ap)1847 nfs_remove(struct vop_old_remove_args *ap)
1848 {
1849 struct vnode *vp = ap->a_vp;
1850 struct vnode *dvp = ap->a_dvp;
1851 struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
1852 struct componentname *cnp = ap->a_cnp;
1853 struct nfsnode *np = VTONFS(vp);
1854 int error = 0;
1855 struct vattr vattr;
1856
1857 lwkt_gettoken(&nmp->nm_token);
1858 #ifndef DIAGNOSTIC
1859 if (VREFCNT(vp) < 1)
1860 panic("nfs_remove: bad v_refcnt");
1861 #endif
1862 if (vp->v_type == VDIR) {
1863 error = EPERM;
1864 } else if (VREFCNT(vp) == 1 || (np->n_sillyrename &&
1865 VOP_GETATTR(vp, &vattr) == 0 && vattr.va_nlink > 1)) {
1866 /*
1867 * Force finalization so the VOP_INACTIVE() call is not delayed.
1868 * This prevents cred structures from building up in nfsnodes
1869 * for deleted files.
1870 */
1871 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
1872 np->n_flag |= NREMOVED;
1873
1874 /*
1875 * Throw away biocache buffers, mainly to avoid
1876 * unnecessary delayed writes later.
1877 */
1878 error = nfs_vinvalbuf(vp, 0, 1);
1879 /* Do the rpc */
1880 if (error != EINTR) {
1881 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1882 cnp->cn_namelen,
1883 cnp->cn_cred, cnp->cn_td);
1884 }
1885
1886 /*
1887 * Kludge City: If the first reply to the remove rpc is lost..
1888 * the reply to the retransmitted request will be ENOENT
1889 * since the file was in fact removed
1890 * Therefore, we cheat and return success.
1891 */
1892 if (error == ENOENT)
1893 error = 0;
1894 } else if (!np->n_sillyrename) {
1895 error = nfs_sillyrename(dvp, vp, cnp);
1896 }
1897 np->n_attrstamp = 0;
1898 lwkt_reltoken(&nmp->nm_token);
1899 if (error == 0) {
1900 nfs_knote(vp, NOTE_DELETE);
1901 nfs_knote(dvp, NOTE_WRITE);
1902 }
1903
1904 return (error);
1905 }
1906
1907 /*
1908 * nfs file remove rpc called from nfs_inactive
1909 *
1910 * NOTE: s_dvp can be VBAD during a forced unmount.
1911 */
1912 int
nfs_removeit(struct sillyrename * sp)1913 nfs_removeit(struct sillyrename *sp)
1914 {
1915 if (sp->s_dvp->v_type == VBAD)
1916 return(0);
1917 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen,
1918 sp->s_cred, NULL));
1919 }
1920
1921 /*
1922 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1923 */
1924 static int
nfs_removerpc(struct vnode * dvp,const char * name,int namelen,struct ucred * cred,struct thread * td)1925 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1926 struct ucred *cred, struct thread *td)
1927 {
1928 int error = 0, wccflag = NFSV3_WCCRATTR;
1929 struct nfsm_info info;
1930
1931 info.mrep = NULL;
1932 info.v3 = NFS_ISV3(dvp);
1933
1934 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1935 nfsm_reqhead(&info, dvp, NFSPROC_REMOVE,
1936 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1937 ERROROUT(nfsm_fhtom(&info, dvp));
1938 ERROROUT(nfsm_strtom(&info, name, namelen, NFS_MAXNAMLEN));
1939 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_REMOVE, td, cred, &error));
1940 if (info.v3) {
1941 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
1942 }
1943 m_freem(info.mrep);
1944 info.mrep = NULL;
1945 nfsmout:
1946 VTONFS(dvp)->n_flag |= NLMODIFIED;
1947 if (!wccflag)
1948 VTONFS(dvp)->n_attrstamp = 0;
1949 return (error);
1950 }
1951
1952 /*
1953 * nfs file rename call
1954 *
1955 * nfs_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
1956 * struct componentname *a_fcnp, struct vnode *a_tdvp,
1957 * struct vnode *a_tvp, struct componentname *a_tcnp)
1958 */
1959 static int
nfs_rename(struct vop_old_rename_args * ap)1960 nfs_rename(struct vop_old_rename_args *ap)
1961 {
1962 struct vnode *fvp = ap->a_fvp;
1963 struct vnode *tvp = ap->a_tvp;
1964 struct vnode *fdvp = ap->a_fdvp;
1965 struct vnode *tdvp = ap->a_tdvp;
1966 struct componentname *tcnp = ap->a_tcnp;
1967 struct componentname *fcnp = ap->a_fcnp;
1968 struct nfsmount *nmp = VFSTONFS(fdvp->v_mount);
1969 int error;
1970
1971 lwkt_gettoken(&nmp->nm_token);
1972
1973 /*
1974 * Force finalization so the VOP_INACTIVE() call is not delayed.
1975 * This prevents cred structures from building up in nfsnodes
1976 * for deleted files.
1977 */
1978 if (tvp) {
1979 atomic_set_int(&tvp->v_refcnt, VREF_FINALIZE);
1980 if (VTONFS(tvp))
1981 VTONFS(tvp)->n_flag |= NREMOVED;
1982 }
1983
1984 /* Check for cross-device rename */
1985 if ((fvp->v_mount != tdvp->v_mount) ||
1986 (tvp && (fvp->v_mount != tvp->v_mount))) {
1987 error = EXDEV;
1988 goto out;
1989 }
1990
1991 /*
1992 * We shouldn't have to flush fvp on rename for most server-side
1993 * filesystems as the file handle should not change. Unfortunately
1994 * the inode for some filesystems (msdosfs) might be tied to the
1995 * file name or directory position so to be completely safe
1996 * vfs.nfs.flush_on_rename is set by default. Clear to improve
1997 * performance.
1998 *
1999 * We must flush tvp on rename because it might become stale on the
2000 * server after the rename.
2001 */
2002 if (nfs_flush_on_rename)
2003 VOP_FSYNC(fvp, MNT_WAIT, 0);
2004 if (tvp)
2005 VOP_FSYNC(tvp, MNT_WAIT, 0);
2006
2007 /*
2008 * If the tvp exists and is in use, sillyrename it before doing the
2009 * rename of the new file over it.
2010 *
2011 * XXX Can't sillyrename a directory.
2012 *
2013 * We do not attempt to do any namecache purges in this old API
2014 * routine. The new API compat functions have access to the actual
2015 * namecache structures and will do it for us.
2016 */
2017 if (tvp && VREFCNT(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
2018 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
2019 nfs_knote(tvp, NOTE_DELETE);
2020 vput(tvp);
2021 tvp = NULL;
2022 } else if (tvp) {
2023 nfs_knote(tvp, NOTE_DELETE);
2024 }
2025
2026 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
2027 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
2028 tcnp->cn_td);
2029
2030 out:
2031 if (error == 0) {
2032 nfs_knote(fdvp, NOTE_WRITE);
2033 nfs_knote(tdvp, NOTE_WRITE);
2034 nfs_knote(fvp, NOTE_RENAME);
2035 }
2036 lwkt_reltoken(&nmp->nm_token);
2037 if (tdvp == tvp)
2038 vrele(tdvp);
2039 else
2040 vput(tdvp);
2041 if (tvp)
2042 vput(tvp);
2043 vrele(fdvp);
2044 vrele(fvp);
2045 /*
2046 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2047 */
2048 if (error == ENOENT)
2049 error = 0;
2050 return (error);
2051 }
2052
2053 /*
2054 * nfs file rename rpc called from nfs_remove() above
2055 */
2056 static int
nfs_renameit(struct vnode * sdvp,struct componentname * scnp,struct sillyrename * sp)2057 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
2058 struct sillyrename *sp)
2059 {
2060 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
2061 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_td));
2062 }
2063
2064 /*
2065 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2066 */
2067 static int
nfs_renamerpc(struct vnode * fdvp,const char * fnameptr,int fnamelen,struct vnode * tdvp,const char * tnameptr,int tnamelen,struct ucred * cred,struct thread * td)2068 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
2069 struct vnode *tdvp, const char *tnameptr, int tnamelen,
2070 struct ucred *cred, struct thread *td)
2071 {
2072 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
2073 struct nfsm_info info;
2074
2075 info.mrep = NULL;
2076 info.v3 = NFS_ISV3(fdvp);
2077
2078 nfsstats.rpccnt[NFSPROC_RENAME]++;
2079 nfsm_reqhead(&info, fdvp, NFSPROC_RENAME,
2080 (NFSX_FH(info.v3) + NFSX_UNSIGNED)*2 +
2081 nfsm_rndup(fnamelen) + nfsm_rndup(tnamelen));
2082 ERROROUT(nfsm_fhtom(&info, fdvp));
2083 ERROROUT(nfsm_strtom(&info, fnameptr, fnamelen, NFS_MAXNAMLEN));
2084 ERROROUT(nfsm_fhtom(&info, tdvp));
2085 ERROROUT(nfsm_strtom(&info, tnameptr, tnamelen, NFS_MAXNAMLEN));
2086 NEGKEEPOUT(nfsm_request(&info, fdvp, NFSPROC_RENAME, td, cred, &error));
2087 if (info.v3) {
2088 ERROROUT(nfsm_wcc_data(&info, fdvp, &fwccflag));
2089 ERROROUT(nfsm_wcc_data(&info, tdvp, &twccflag));
2090 }
2091 m_freem(info.mrep);
2092 info.mrep = NULL;
2093 nfsmout:
2094 VTONFS(fdvp)->n_flag |= NLMODIFIED;
2095 VTONFS(tdvp)->n_flag |= NLMODIFIED;
2096 if (!fwccflag)
2097 VTONFS(fdvp)->n_attrstamp = 0;
2098 if (!twccflag)
2099 VTONFS(tdvp)->n_attrstamp = 0;
2100 return (error);
2101 }
2102
2103 /*
2104 * nfs hard link create call
2105 *
2106 * nfs_link(struct vnode *a_tdvp, struct vnode *a_vp,
2107 * struct componentname *a_cnp)
2108 */
2109 static int
nfs_link(struct vop_old_link_args * ap)2110 nfs_link(struct vop_old_link_args *ap)
2111 {
2112 struct vnode *vp = ap->a_vp;
2113 struct vnode *tdvp = ap->a_tdvp;
2114 struct nfsmount *nmp = VFSTONFS(tdvp->v_mount);
2115 struct componentname *cnp = ap->a_cnp;
2116 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2117 struct nfsm_info info;
2118
2119 if (vp->v_mount != tdvp->v_mount) {
2120 return (EXDEV);
2121 }
2122 lwkt_gettoken(&nmp->nm_token);
2123
2124 /*
2125 * The attribute cache may get out of sync with the server on link.
2126 * Pushing writes to the server before handle was inherited from
2127 * long long ago and it is unclear if we still need to do this.
2128 * Defaults to off.
2129 */
2130 if (nfs_flush_on_hlink)
2131 VOP_FSYNC(vp, MNT_WAIT, 0);
2132
2133 info.mrep = NULL;
2134 info.v3 = NFS_ISV3(vp);
2135
2136 nfsstats.rpccnt[NFSPROC_LINK]++;
2137 nfsm_reqhead(&info, vp, NFSPROC_LINK,
2138 NFSX_FH(info.v3) * 2 + NFSX_UNSIGNED +
2139 nfsm_rndup(cnp->cn_namelen));
2140 ERROROUT(nfsm_fhtom(&info, vp));
2141 ERROROUT(nfsm_fhtom(&info, tdvp));
2142 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2143 NFS_MAXNAMLEN));
2144 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_LINK, cnp->cn_td,
2145 cnp->cn_cred, &error));
2146 if (info.v3) {
2147 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2148 NFS_LATTR_NOSHRINK));
2149 ERROROUT(nfsm_wcc_data(&info, tdvp, &wccflag));
2150 }
2151 m_freem(info.mrep);
2152 info.mrep = NULL;
2153 nfsmout:
2154 VTONFS(tdvp)->n_flag |= NLMODIFIED;
2155 if (!attrflag)
2156 VTONFS(vp)->n_attrstamp = 0;
2157 if (!wccflag)
2158 VTONFS(tdvp)->n_attrstamp = 0;
2159 /*
2160 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2161 */
2162 if (error == EEXIST)
2163 error = 0;
2164 lwkt_reltoken(&nmp->nm_token);
2165 if (error == 0) {
2166 nfs_knote(vp, NOTE_LINK);
2167 nfs_knote(tdvp, NOTE_WRITE);
2168 }
2169
2170 return (error);
2171 }
2172
2173 /*
2174 * nfs symbolic link create call
2175 *
2176 * nfs_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
2177 * struct componentname *a_cnp, struct vattr *a_vap,
2178 * char *a_target)
2179 */
2180 static int
nfs_symlink(struct vop_old_symlink_args * ap)2181 nfs_symlink(struct vop_old_symlink_args *ap)
2182 {
2183 struct vnode *dvp = ap->a_dvp;
2184 struct vattr *vap = ap->a_vap;
2185 struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2186 struct componentname *cnp = ap->a_cnp;
2187 struct nfsv2_sattr *sp;
2188 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2189 struct vnode *newvp = NULL;
2190 struct nfsm_info info;
2191
2192 info.mrep = NULL;
2193 info.v3 = NFS_ISV3(dvp);
2194 lwkt_gettoken(&nmp->nm_token);
2195
2196 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2197 slen = strlen(ap->a_target);
2198 nfsm_reqhead(&info, dvp, NFSPROC_SYMLINK,
2199 NFSX_FH(info.v3) + 2*NFSX_UNSIGNED +
2200 nfsm_rndup(cnp->cn_namelen) +
2201 nfsm_rndup(slen) + NFSX_SATTR(info.v3));
2202 ERROROUT(nfsm_fhtom(&info, dvp));
2203 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2204 NFS_MAXNAMLEN));
2205 if (info.v3) {
2206 nfsm_v3attrbuild(&info, vap, FALSE);
2207 }
2208 ERROROUT(nfsm_strtom(&info, ap->a_target, slen, NFS_MAXPATHLEN));
2209 if (info.v3 == 0) {
2210 sp = nfsm_build(&info, NFSX_V2SATTR);
2211 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2212 sp->sa_uid = nfs_xdrneg1;
2213 sp->sa_gid = nfs_xdrneg1;
2214 sp->sa_size = nfs_xdrneg1;
2215 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2216 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2217 }
2218
2219 /*
2220 * Issue the NFS request and get the rpc response.
2221 *
2222 * Only NFSv3 responses returning an error of 0 actually return
2223 * a file handle that can be converted into newvp without having
2224 * to do an extra lookup rpc.
2225 */
2226 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_SYMLINK, cnp->cn_td,
2227 cnp->cn_cred, &error));
2228 if (info.v3) {
2229 if (error == 0) {
2230 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2231 }
2232 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2233 }
2234
2235 /*
2236 * out code jumps -> here, mrep is also freed.
2237 */
2238
2239 m_freem(info.mrep);
2240 info.mrep = NULL;
2241 nfsmout:
2242
2243 /*
2244 * If we get an EEXIST error, silently convert it to no-error
2245 * in case of an NFS retry.
2246 */
2247 if (error == EEXIST)
2248 error = 0;
2249
2250 /*
2251 * If we do not have (or no longer have) an error, and we could
2252 * not extract the newvp from the response due to the request being
2253 * NFSv2 or the error being EEXIST. We have to do a lookup in order
2254 * to obtain a newvp to return.
2255 */
2256 if (error == 0 && newvp == NULL) {
2257 struct nfsnode *np = NULL;
2258
2259 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2260 cnp->cn_cred, cnp->cn_td, &np);
2261 if (!error)
2262 newvp = NFSTOV(np);
2263 }
2264 if (error) {
2265 if (newvp)
2266 vput(newvp);
2267 } else {
2268 *ap->a_vpp = newvp;
2269 }
2270 VTONFS(dvp)->n_flag |= NLMODIFIED;
2271 if (!wccflag)
2272 VTONFS(dvp)->n_attrstamp = 0;
2273 if (error == 0 && *ap->a_vpp)
2274 nfs_knote(*ap->a_vpp, NOTE_WRITE);
2275 lwkt_reltoken(&nmp->nm_token);
2276
2277 return (error);
2278 }
2279
2280 /*
2281 * nfs make dir call
2282 *
2283 * nfs_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
2284 * struct componentname *a_cnp, struct vattr *a_vap)
2285 */
2286 static int
nfs_mkdir(struct vop_old_mkdir_args * ap)2287 nfs_mkdir(struct vop_old_mkdir_args *ap)
2288 {
2289 struct vnode *dvp = ap->a_dvp;
2290 struct vattr *vap = ap->a_vap;
2291 struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2292 struct componentname *cnp = ap->a_cnp;
2293 struct nfsv2_sattr *sp;
2294 struct nfsnode *np = NULL;
2295 struct vnode *newvp = NULL;
2296 struct vattr vattr;
2297 int error = 0, wccflag = NFSV3_WCCRATTR;
2298 int gotvp = 0;
2299 int len;
2300 struct nfsm_info info;
2301
2302 info.mrep = NULL;
2303 info.v3 = NFS_ISV3(dvp);
2304 lwkt_gettoken(&nmp->nm_token);
2305
2306 if ((error = VOP_GETATTR(dvp, &vattr)) != 0) {
2307 lwkt_reltoken(&nmp->nm_token);
2308 return (error);
2309 }
2310 len = cnp->cn_namelen;
2311 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2312 nfsm_reqhead(&info, dvp, NFSPROC_MKDIR,
2313 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2314 nfsm_rndup(len) + NFSX_SATTR(info.v3));
2315 ERROROUT(nfsm_fhtom(&info, dvp));
2316 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, len, NFS_MAXNAMLEN));
2317 if (info.v3) {
2318 nfsm_v3attrbuild(&info, vap, FALSE);
2319 } else {
2320 sp = nfsm_build(&info, NFSX_V2SATTR);
2321 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2322 sp->sa_uid = nfs_xdrneg1;
2323 sp->sa_gid = nfs_xdrneg1;
2324 sp->sa_size = nfs_xdrneg1;
2325 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2326 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2327 }
2328 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_MKDIR, cnp->cn_td,
2329 cnp->cn_cred, &error));
2330 if (error == 0) {
2331 ERROROUT(nfsm_mtofh(&info, dvp, &newvp, &gotvp));
2332 }
2333 if (info.v3) {
2334 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2335 }
2336 m_freem(info.mrep);
2337 info.mrep = NULL;
2338 nfsmout:
2339 VTONFS(dvp)->n_flag |= NLMODIFIED;
2340 if (!wccflag)
2341 VTONFS(dvp)->n_attrstamp = 0;
2342 /*
2343 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2344 * if we can succeed in looking up the directory.
2345 */
2346 if (error == EEXIST || (!error && !gotvp)) {
2347 if (newvp) {
2348 vrele(newvp);
2349 newvp = NULL;
2350 }
2351 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2352 cnp->cn_td, &np);
2353 if (!error) {
2354 newvp = NFSTOV(np);
2355 if (newvp->v_type != VDIR)
2356 error = EEXIST;
2357 }
2358 }
2359 if (error) {
2360 if (newvp)
2361 vrele(newvp);
2362 } else {
2363 nfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
2364 *ap->a_vpp = newvp;
2365 }
2366 lwkt_reltoken(&nmp->nm_token);
2367 return (error);
2368 }
2369
2370 /*
2371 * nfs remove directory call
2372 *
2373 * nfs_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
2374 * struct componentname *a_cnp)
2375 */
2376 static int
nfs_rmdir(struct vop_old_rmdir_args * ap)2377 nfs_rmdir(struct vop_old_rmdir_args *ap)
2378 {
2379 struct vnode *vp = ap->a_vp;
2380 struct vnode *dvp = ap->a_dvp;
2381 struct nfsmount *nmp = VFSTONFS(dvp->v_mount);
2382 struct componentname *cnp = ap->a_cnp;
2383 int error = 0, wccflag = NFSV3_WCCRATTR;
2384 struct nfsm_info info;
2385
2386 info.mrep = NULL;
2387 info.v3 = NFS_ISV3(dvp);
2388
2389 if (dvp == vp)
2390 return (EINVAL);
2391
2392 lwkt_gettoken(&nmp->nm_token);
2393
2394 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2395 nfsm_reqhead(&info, dvp, NFSPROC_RMDIR,
2396 NFSX_FH(info.v3) + NFSX_UNSIGNED +
2397 nfsm_rndup(cnp->cn_namelen));
2398 ERROROUT(nfsm_fhtom(&info, dvp));
2399 ERROROUT(nfsm_strtom(&info, cnp->cn_nameptr, cnp->cn_namelen,
2400 NFS_MAXNAMLEN));
2401 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_RMDIR, cnp->cn_td,
2402 cnp->cn_cred, &error));
2403 if (info.v3) {
2404 ERROROUT(nfsm_wcc_data(&info, dvp, &wccflag));
2405 }
2406 m_freem(info.mrep);
2407 info.mrep = NULL;
2408 nfsmout:
2409 VTONFS(dvp)->n_flag |= NLMODIFIED;
2410 if (!wccflag)
2411 VTONFS(dvp)->n_attrstamp = 0;
2412 /*
2413 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2414 */
2415 if (error == ENOENT)
2416 error = 0;
2417 else
2418 nfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
2419 lwkt_reltoken(&nmp->nm_token);
2420
2421 return (error);
2422 }
2423
2424 /*
2425 * nfs readdir call
2426 *
2427 * nfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
2428 */
2429 static int
nfs_readdir(struct vop_readdir_args * ap)2430 nfs_readdir(struct vop_readdir_args *ap)
2431 {
2432 struct vnode *vp = ap->a_vp;
2433 struct nfsnode *np = VTONFS(vp);
2434 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2435 struct uio *uio = ap->a_uio;
2436 int tresid, error;
2437 struct vattr vattr;
2438
2439 if (vp->v_type != VDIR)
2440 return (EPERM);
2441
2442 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
2443 if (error)
2444 return (error);
2445
2446 lwkt_gettoken(&nmp->nm_token);
2447
2448 /*
2449 * If we have a valid EOF offset cache we must call VOP_GETATTR()
2450 * and then check that is still valid, or if this is an NQNFS mount
2451 * we call NQNFS_CKCACHEABLE() instead of VOP_GETATTR(). Note that
2452 * VOP_GETATTR() does not necessarily go to the wire.
2453 */
2454 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2455 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0) {
2456 if (VOP_GETATTR(vp, &vattr) == 0 &&
2457 (np->n_flag & (NLMODIFIED|NRMODIFIED)) == 0
2458 ) {
2459 nfsstats.direofcache_hits++;
2460 goto done;
2461 }
2462 }
2463
2464 /*
2465 * Call nfs_bioread() to do the real work. nfs_bioread() does its
2466 * own cache coherency checks so we do not have to.
2467 */
2468 tresid = uio->uio_resid;
2469 error = nfs_bioread(vp, uio, 0);
2470
2471 if (!error && uio->uio_resid == tresid)
2472 nfsstats.direofcache_misses++;
2473 done:
2474 lwkt_reltoken(&nmp->nm_token);
2475 vn_unlock(vp);
2476
2477 return (error);
2478 }
2479
2480 /*
2481 * Readdir rpc call. nfs_bioread->nfs_doio->nfs_readdirrpc.
2482 *
2483 * Note that for directories, nfs_bioread maintains the underlying nfs-centric
2484 * offset/block and converts the nfs formatted directory entries for userland
2485 * consumption as well as deals with offsets into the middle of blocks.
2486 * nfs_doio only deals with logical blocks. In particular, uio_offset will
2487 * be block-bounded. It must convert to cookies for the actual RPC.
2488 */
2489 int
nfs_readdirrpc_uio(struct vnode * vp,struct uio * uiop)2490 nfs_readdirrpc_uio(struct vnode *vp, struct uio *uiop)
2491 {
2492 int len, left;
2493 struct nfs_dirent *dp = NULL;
2494 u_int32_t *tl;
2495 nfsuint64 *cookiep;
2496 caddr_t cp;
2497 nfsuint64 cookie;
2498 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2499 struct nfsnode *dnp = VTONFS(vp);
2500 u_quad_t fileno;
2501 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2502 int attrflag;
2503 struct nfsm_info info;
2504
2505 info.mrep = NULL;
2506 info.v3 = NFS_ISV3(vp);
2507
2508 #ifndef DIAGNOSTIC
2509 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2510 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2511 panic("nfs readdirrpc bad uio");
2512 #endif
2513
2514 /*
2515 * If there is no cookie, assume directory was stale.
2516 */
2517 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2518 if (cookiep)
2519 cookie = *cookiep;
2520 else
2521 return (NFSERR_BAD_COOKIE);
2522 /*
2523 * Loop around doing readdir rpc's of size nm_readdirsize
2524 * truncated to a multiple of DIRBLKSIZ.
2525 * The stopping criteria is EOF or buffer full.
2526 */
2527 while (more_dirs && bigenough) {
2528 nfsstats.rpccnt[NFSPROC_READDIR]++;
2529 nfsm_reqhead(&info, vp, NFSPROC_READDIR,
2530 NFSX_FH(info.v3) + NFSX_READDIR(info.v3));
2531 ERROROUT(nfsm_fhtom(&info, vp));
2532 if (info.v3) {
2533 tl = nfsm_build(&info, 5 * NFSX_UNSIGNED);
2534 *tl++ = cookie.nfsuquad[0];
2535 *tl++ = cookie.nfsuquad[1];
2536 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2537 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2538 } else {
2539 /*
2540 * WARNING! HAMMER DIRECTORIES WILL NOT WORK WELL
2541 * WITH NFSv2!!! There's nothing I can really do
2542 * about it other than to hope the server supports
2543 * rdirplus w/NFSv2.
2544 */
2545 tl = nfsm_build(&info, 2 * NFSX_UNSIGNED);
2546 *tl++ = cookie.nfsuquad[0];
2547 }
2548 *tl = txdr_unsigned(nmp->nm_readdirsize);
2549 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIR,
2550 uiop->uio_td,
2551 nfs_vpcred(vp, ND_READ), &error));
2552 if (info.v3) {
2553 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2554 NFS_LATTR_NOSHRINK));
2555 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2556 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2557 dnp->n_cookieverf.nfsuquad[1] = *tl;
2558 }
2559 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2560 more_dirs = fxdr_unsigned(int, *tl);
2561
2562 /* loop thru the dir entries, converting them to std form */
2563 while (more_dirs && bigenough) {
2564 if (info.v3) {
2565 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2566 fileno = fxdr_hyper(tl);
2567 len = fxdr_unsigned(int, *(tl + 2));
2568 } else {
2569 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2570 fileno = fxdr_unsigned(u_quad_t, *tl++);
2571 len = fxdr_unsigned(int, *tl);
2572 }
2573 if (len <= 0 || len > NFS_MAXNAMLEN) {
2574 error = EBADRPC;
2575 m_freem(info.mrep);
2576 info.mrep = NULL;
2577 goto nfsmout;
2578 }
2579
2580 /*
2581 * len is the number of bytes in the path element
2582 * name, not including the \0 termination.
2583 *
2584 * tlen is the number of bytes w have to reserve for
2585 * the path element name.
2586 */
2587 tlen = nfsm_rndup(len);
2588 if (tlen == len)
2589 tlen += 4; /* To ensure null termination */
2590
2591 /*
2592 * If the entry would cross a DIRBLKSIZ boundary,
2593 * extend the previous nfs_dirent to cover the
2594 * remaining space.
2595 */
2596 left = DIRBLKSIZ - blksiz;
2597 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2598 dp->nfs_reclen += left;
2599 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2600 uiop->uio_iov->iov_len -= left;
2601 uiop->uio_offset += left;
2602 uiop->uio_resid -= left;
2603 blksiz = 0;
2604 }
2605 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2606 bigenough = 0;
2607 if (bigenough) {
2608 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2609 dp->nfs_ino = fileno;
2610 dp->nfs_namlen = len;
2611 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2612 dp->nfs_type = DT_UNKNOWN;
2613 blksiz += dp->nfs_reclen;
2614 if (blksiz == DIRBLKSIZ)
2615 blksiz = 0;
2616 uiop->uio_offset += sizeof(struct nfs_dirent);
2617 uiop->uio_resid -= sizeof(struct nfs_dirent);
2618 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2619 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2620 ERROROUT(nfsm_mtouio(&info, uiop, len));
2621
2622 /*
2623 * The uiop has advanced by nfs_dirent + len
2624 * but really needs to advance by
2625 * nfs_dirent + tlen
2626 */
2627 cp = uiop->uio_iov->iov_base;
2628 tlen -= len;
2629 *cp = '\0'; /* null terminate */
2630 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2631 uiop->uio_iov->iov_len -= tlen;
2632 uiop->uio_offset += tlen;
2633 uiop->uio_resid -= tlen;
2634 } else {
2635 /*
2636 * NFS strings must be rounded up (nfsm_myouio
2637 * handled that in the bigenough case).
2638 */
2639 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2640 }
2641 if (info.v3) {
2642 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2643 } else {
2644 NULLOUT(tl = nfsm_dissect(&info, 2 * NFSX_UNSIGNED));
2645 }
2646
2647 /*
2648 * If we were able to accomodate the last entry,
2649 * get the cookie for the next one. Otherwise
2650 * hold-over the cookie for the one we were not
2651 * able to accomodate.
2652 */
2653 if (bigenough) {
2654 cookie.nfsuquad[0] = *tl++;
2655 if (info.v3)
2656 cookie.nfsuquad[1] = *tl++;
2657 } else if (info.v3) {
2658 tl += 2;
2659 } else {
2660 tl++;
2661 }
2662 more_dirs = fxdr_unsigned(int, *tl);
2663 }
2664 /*
2665 * If at end of rpc data, get the eof boolean
2666 */
2667 if (!more_dirs) {
2668 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2669 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2670 }
2671 m_freem(info.mrep);
2672 info.mrep = NULL;
2673 }
2674 /*
2675 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2676 * by increasing d_reclen for the last record.
2677 */
2678 if (blksiz > 0) {
2679 left = DIRBLKSIZ - blksiz;
2680 dp->nfs_reclen += left;
2681 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2682 uiop->uio_iov->iov_len -= left;
2683 uiop->uio_offset += left;
2684 uiop->uio_resid -= left;
2685 }
2686
2687 if (bigenough) {
2688 /*
2689 * We hit the end of the directory, update direofoffset.
2690 */
2691 dnp->n_direofoffset = uiop->uio_offset;
2692 } else {
2693 /*
2694 * There is more to go, insert the link cookie so the
2695 * next block can be read.
2696 */
2697 if (uiop->uio_resid > 0)
2698 kprintf("EEK! readdirrpc resid > 0\n");
2699 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2700 *cookiep = cookie;
2701 }
2702 nfsmout:
2703 return (error);
2704 }
2705
2706 /*
2707 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2708 */
2709 int
nfs_readdirplusrpc_uio(struct vnode * vp,struct uio * uiop)2710 nfs_readdirplusrpc_uio(struct vnode *vp, struct uio *uiop)
2711 {
2712 int len, left;
2713 struct nfs_dirent *dp;
2714 u_int32_t *tl;
2715 struct vnode *newvp;
2716 nfsuint64 *cookiep;
2717 caddr_t dpossav1, dpossav2;
2718 caddr_t cp;
2719 struct mbuf *mdsav1, *mdsav2;
2720 nfsuint64 cookie;
2721 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2722 struct nfsnode *dnp = VTONFS(vp), *np;
2723 nfsfh_t *fhp;
2724 u_quad_t fileno;
2725 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2726 int attrflag, fhsize;
2727 struct nchandle nch;
2728 struct nchandle dnch;
2729 struct nlcomponent nlc;
2730 struct nfsm_info info;
2731
2732 info.mrep = NULL;
2733 info.v3 = 1;
2734
2735 #ifndef nolint
2736 dp = NULL;
2737 #endif
2738 #ifndef DIAGNOSTIC
2739 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2740 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2741 panic("nfs readdirplusrpc bad uio");
2742 #endif
2743 /*
2744 * Obtain the namecache record for the directory so we have something
2745 * to use as a basis for creating the entries. This function will
2746 * return a held (but not locked) ncp. The ncp may be disconnected
2747 * from the tree and cannot be used for upward traversals, and the
2748 * ncp may be unnamed. Note that other unrelated operations may
2749 * cause the ncp to be named at any time.
2750 *
2751 * We have to lock the ncp to prevent a lock order reversal when
2752 * rdirplus does nlookups of the children, because the vnode is
2753 * locked and has to stay that way.
2754 */
2755 cache_fromdvp(vp, NULL, 0, &dnch);
2756 bzero(&nlc, sizeof(nlc));
2757 newvp = NULLVP;
2758
2759 /*
2760 * If there is no cookie, assume directory was stale.
2761 */
2762 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2763 if (cookiep) {
2764 cookie = *cookiep;
2765 } else {
2766 if (dnch.ncp)
2767 cache_drop(&dnch);
2768 return (NFSERR_BAD_COOKIE);
2769 }
2770
2771 /*
2772 * Loop around doing readdir rpc's of size nm_readdirsize
2773 * truncated to a multiple of DIRBLKSIZ.
2774 * The stopping criteria is EOF or buffer full.
2775 */
2776 while (more_dirs && bigenough) {
2777 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2778 nfsm_reqhead(&info, vp, NFSPROC_READDIRPLUS,
2779 NFSX_FH(info.v3) + 6 * NFSX_UNSIGNED);
2780 ERROROUT(nfsm_fhtom(&info, vp));
2781 tl = nfsm_build(&info, 6 * NFSX_UNSIGNED);
2782 *tl++ = cookie.nfsuquad[0];
2783 *tl++ = cookie.nfsuquad[1];
2784 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2785 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2786 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2787 *tl = txdr_unsigned(nmp->nm_rsize);
2788 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_READDIRPLUS,
2789 uiop->uio_td,
2790 nfs_vpcred(vp, ND_READ), &error));
2791 ERROROUT(nfsm_postop_attr(&info, vp, &attrflag,
2792 NFS_LATTR_NOSHRINK));
2793 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2794 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2795 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2796 more_dirs = fxdr_unsigned(int, *tl);
2797
2798 /* loop thru the dir entries, doctoring them to 4bsd form */
2799 while (more_dirs && bigenough) {
2800 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2801 fileno = fxdr_hyper(tl);
2802 len = fxdr_unsigned(int, *(tl + 2));
2803 if (len <= 0 || len > NFS_MAXNAMLEN) {
2804 error = EBADRPC;
2805 m_freem(info.mrep);
2806 info.mrep = NULL;
2807 goto nfsmout;
2808 }
2809 tlen = nfsm_rndup(len);
2810 if (tlen == len)
2811 tlen += 4; /* To ensure null termination*/
2812 left = DIRBLKSIZ - blksiz;
2813 if ((tlen + sizeof(struct nfs_dirent)) > left) {
2814 dp->nfs_reclen += left;
2815 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2816 uiop->uio_iov->iov_len -= left;
2817 uiop->uio_offset += left;
2818 uiop->uio_resid -= left;
2819 blksiz = 0;
2820 }
2821 if ((tlen + sizeof(struct nfs_dirent)) > uiop->uio_resid)
2822 bigenough = 0;
2823 if (bigenough) {
2824 dp = (struct nfs_dirent *)uiop->uio_iov->iov_base;
2825 dp->nfs_ino = fileno;
2826 dp->nfs_namlen = len;
2827 dp->nfs_reclen = tlen + sizeof(struct nfs_dirent);
2828 dp->nfs_type = DT_UNKNOWN;
2829 blksiz += dp->nfs_reclen;
2830 if (blksiz == DIRBLKSIZ)
2831 blksiz = 0;
2832 uiop->uio_offset += sizeof(struct nfs_dirent);
2833 uiop->uio_resid -= sizeof(struct nfs_dirent);
2834 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + sizeof(struct nfs_dirent);
2835 uiop->uio_iov->iov_len -= sizeof(struct nfs_dirent);
2836 nlc.nlc_nameptr = uiop->uio_iov->iov_base;
2837 nlc.nlc_namelen = len;
2838 ERROROUT(nfsm_mtouio(&info, uiop, len));
2839 cp = uiop->uio_iov->iov_base;
2840 tlen -= len;
2841 *cp = '\0';
2842 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + tlen;
2843 uiop->uio_iov->iov_len -= tlen;
2844 uiop->uio_offset += tlen;
2845 uiop->uio_resid -= tlen;
2846 } else {
2847 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
2848 }
2849 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
2850 if (bigenough) {
2851 cookie.nfsuquad[0] = *tl++;
2852 cookie.nfsuquad[1] = *tl++;
2853 } else {
2854 tl += 2;
2855 }
2856
2857 /*
2858 * Since the attributes are before the file handle
2859 * (sigh), we must skip over the attributes and then
2860 * come back and get them.
2861 */
2862 attrflag = fxdr_unsigned(int, *tl);
2863 if (attrflag) {
2864 dpossav1 = info.dpos;
2865 mdsav1 = info.md;
2866 ERROROUT(nfsm_adv(&info, NFSX_V3FATTR));
2867 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2868 doit = fxdr_unsigned(int, *tl);
2869 if (doit) {
2870 NEGATIVEOUT(fhsize = nfsm_getfh(&info, &fhp));
2871 }
2872 if (doit && bigenough && !nlcdegenerate(&nlc) &&
2873 !NFS_CMPFH(dnp, fhp, fhsize)
2874 ) {
2875 if (dnch.ncp) {
2876 #if 0
2877 kprintf("NFS/READDIRPLUS, ENTER %*.*s\n",
2878 nlc.nlc_namelen, nlc.nlc_namelen,
2879 nlc.nlc_nameptr);
2880 #endif
2881 /*
2882 * This is a bit hokey but there isn't
2883 * much we can do about it. We can't
2884 * hold the directory vp locked while
2885 * doing lookups and gets.
2886 */
2887 nch = cache_nlookup_nonblock(&dnch, &nlc);
2888 if (nch.ncp == NULL)
2889 goto rdfail;
2890 cache_setunresolved(&nch);
2891 error = nfs_nget_nonblock(vp->v_mount, fhp,
2892 fhsize, &np,
2893 NULL);
2894 if (error) {
2895 cache_put(&nch);
2896 goto rdfail;
2897 }
2898 newvp = NFSTOV(np);
2899 dpossav2 = info.dpos;
2900 info.dpos = dpossav1;
2901 mdsav2 = info.md;
2902 info.md = mdsav1;
2903 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
2904 info.dpos = dpossav2;
2905 info.md = mdsav2;
2906 dp->nfs_type =
2907 IFTODT(VTTOIF(np->n_vattr.va_type));
2908 nfs_cache_setvp(&nch, newvp,
2909 nfspos_cache_timeout);
2910 vput(newvp);
2911 newvp = NULLVP;
2912 cache_put(&nch);
2913 } else {
2914 rdfail:
2915 ;
2916 #if 0
2917 kprintf("Warning: NFS/rddirplus, "
2918 "UNABLE TO ENTER %*.*s\n",
2919 nlc.nlc_namelen, nlc.nlc_namelen,
2920 nlc.nlc_nameptr);
2921 #endif
2922 }
2923 }
2924 } else {
2925 /* Just skip over the file handle */
2926 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2927 i = fxdr_unsigned(int, *tl);
2928 ERROROUT(nfsm_adv(&info, nfsm_rndup(i)));
2929 }
2930 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2931 more_dirs = fxdr_unsigned(int, *tl);
2932 }
2933 /*
2934 * If at end of rpc data, get the eof boolean
2935 */
2936 if (!more_dirs) {
2937 NULLOUT(tl = nfsm_dissect(&info, NFSX_UNSIGNED));
2938 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2939 }
2940 m_freem(info.mrep);
2941 info.mrep = NULL;
2942 }
2943 /*
2944 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2945 * by increasing d_reclen for the last record.
2946 */
2947 if (blksiz > 0) {
2948 left = DIRBLKSIZ - blksiz;
2949 dp->nfs_reclen += left;
2950 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + left;
2951 uiop->uio_iov->iov_len -= left;
2952 uiop->uio_offset += left;
2953 uiop->uio_resid -= left;
2954 }
2955
2956 /*
2957 * We are now either at the end of the directory or have filled the
2958 * block.
2959 */
2960 if (bigenough) {
2961 dnp->n_direofoffset = uiop->uio_offset;
2962 } else {
2963 if (uiop->uio_resid > 0)
2964 kprintf("EEK! readdirplusrpc resid > 0\n");
2965 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2966 *cookiep = cookie;
2967 }
2968 nfsmout:
2969 if (newvp != NULLVP) {
2970 if (newvp == vp)
2971 vrele(newvp);
2972 else
2973 vput(newvp);
2974 newvp = NULLVP;
2975 }
2976 if (dnch.ncp)
2977 cache_drop(&dnch);
2978 return (error);
2979 }
2980
2981 /*
2982 * Silly rename. To make the NFS filesystem that is stateless look a little
2983 * more like the "ufs" a remove of an active vnode is translated to a rename
2984 * to a funny looking filename that is removed by nfs_inactive on the
2985 * nfsnode. There is the potential for another process on a different client
2986 * to create the same funny name between the nfs_lookitup() fails and the
2987 * nfs_rename() completes, but...
2988 */
2989 static int
nfs_sillyrename(struct vnode * dvp,struct vnode * vp,struct componentname * cnp)2990 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2991 {
2992 struct sillyrename *sp;
2993 struct nfsnode *np;
2994 int error;
2995
2996 /*
2997 * Force finalization so the VOP_INACTIVE() call is not delayed.
2998 * This prevents cred structures from building up in nfsnodes
2999 * for deleted files.
3000 */
3001 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
3002 np = VTONFS(vp);
3003 np->n_flag |= NREMOVED;
3004
3005 /*
3006 * We previously purged dvp instead of vp. I don't know why, it
3007 * completely destroys performance. We can't do it anyway with the
3008 * new VFS API since we would be breaking the namecache topology.
3009 */
3010 cache_purge(vp); /* XXX */
3011 #ifndef DIAGNOSTIC
3012 if (vp->v_type == VDIR)
3013 panic("nfs: sillyrename dir");
3014 #endif
3015 sp = kmalloc(sizeof(struct sillyrename), M_NFSREQ, M_WAITOK);
3016 sp->s_cred = crdup(cnp->cn_cred);
3017 sp->s_dvp = dvp;
3018 vref(dvp);
3019
3020 /* Fudge together a funny name */
3021 sp->s_namlen = ksprintf(sp->s_name, ".nfsA%08x4.4",
3022 (int)(intptr_t)cnp->cn_td);
3023
3024 /* Try lookitups until we get one that isn't there */
3025 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3026 cnp->cn_td, NULL) == 0) {
3027 sp->s_name[4]++;
3028 if (sp->s_name[4] > 'z') {
3029 error = EINVAL;
3030 goto bad;
3031 }
3032 }
3033 error = nfs_renameit(dvp, cnp, sp);
3034 if (error)
3035 goto bad;
3036 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3037 cnp->cn_td, &np);
3038 np->n_sillyrename = sp;
3039 return (0);
3040 bad:
3041 vrele(sp->s_dvp);
3042 crfree(sp->s_cred);
3043 kfree((caddr_t)sp, M_NFSREQ);
3044
3045 return (error);
3046 }
3047
3048 /*
3049 * Look up a file name and optionally either update the file handle or
3050 * allocate an nfsnode, depending on the value of npp.
3051 * npp == NULL --> just do the lookup
3052 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
3053 * handled too
3054 * *npp != NULL --> update the file handle in the vnode
3055 */
3056 static int
nfs_lookitup(struct vnode * dvp,const char * name,int len,struct ucred * cred,struct thread * td,struct nfsnode ** npp)3057 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
3058 struct thread *td, struct nfsnode **npp)
3059 {
3060 struct vnode *newvp = NULL;
3061 struct nfsnode *np, *dnp = VTONFS(dvp);
3062 int error = 0, fhlen, attrflag;
3063 nfsfh_t *nfhp;
3064 struct nfsm_info info;
3065
3066 info.mrep = NULL;
3067 info.v3 = NFS_ISV3(dvp);
3068
3069 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
3070 nfsm_reqhead(&info, dvp, NFSPROC_LOOKUP,
3071 NFSX_FH(info.v3) + NFSX_UNSIGNED + nfsm_rndup(len));
3072 ERROROUT(nfsm_fhtom(&info, dvp));
3073 ERROROUT(nfsm_strtom(&info, name, len, NFS_MAXNAMLEN));
3074 NEGKEEPOUT(nfsm_request(&info, dvp, NFSPROC_LOOKUP, td, cred, &error));
3075 if (npp && !error) {
3076 NEGATIVEOUT(fhlen = nfsm_getfh(&info, &nfhp));
3077 if (*npp) {
3078 np = *npp;
3079 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
3080 kfree((caddr_t)np->n_fhp, M_NFSBIGFH);
3081 np->n_fhp = &np->n_fh;
3082 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
3083 np->n_fhp =(nfsfh_t *)kmalloc(fhlen,M_NFSBIGFH,M_WAITOK);
3084 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
3085 np->n_fhsize = fhlen;
3086 newvp = NFSTOV(np);
3087 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
3088 vref(dvp);
3089 newvp = dvp;
3090 } else {
3091 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, NULL);
3092 if (error) {
3093 m_freem(info.mrep);
3094 info.mrep = NULL;
3095 return (error);
3096 }
3097 newvp = NFSTOV(np);
3098 }
3099 if (info.v3) {
3100 ERROROUT(nfsm_postop_attr(&info, newvp, &attrflag,
3101 NFS_LATTR_NOSHRINK));
3102 if (!attrflag && *npp == NULL) {
3103 m_freem(info.mrep);
3104 info.mrep = NULL;
3105 if (newvp == dvp)
3106 vrele(newvp);
3107 else
3108 vput(newvp);
3109 return (ENOENT);
3110 }
3111 } else {
3112 ERROROUT(nfsm_loadattr(&info, newvp, NULL));
3113 }
3114 }
3115 m_freem(info.mrep);
3116 info.mrep = NULL;
3117 nfsmout:
3118 if (npp && *npp == NULL) {
3119 if (error) {
3120 if (newvp) {
3121 if (newvp == dvp)
3122 vrele(newvp);
3123 else
3124 vput(newvp);
3125 }
3126 } else
3127 *npp = np;
3128 }
3129 return (error);
3130 }
3131
3132 /*
3133 * Nfs Version 3 commit rpc
3134 *
3135 * We call it 'uio' to distinguish it from 'bio' but there is no real uio
3136 * involved.
3137 */
3138 int
nfs_commitrpc_uio(struct vnode * vp,u_quad_t offset,int cnt,struct thread * td)3139 nfs_commitrpc_uio(struct vnode *vp, u_quad_t offset, int cnt, struct thread *td)
3140 {
3141 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3142 int error = 0, wccflag = NFSV3_WCCRATTR;
3143 struct nfsm_info info;
3144 u_int32_t *tl;
3145
3146 info.mrep = NULL;
3147 info.v3 = 1;
3148
3149 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
3150 return (0);
3151 nfsstats.rpccnt[NFSPROC_COMMIT]++;
3152 nfsm_reqhead(&info, vp, NFSPROC_COMMIT, NFSX_FH(1));
3153 ERROROUT(nfsm_fhtom(&info, vp));
3154 tl = nfsm_build(&info, 3 * NFSX_UNSIGNED);
3155 txdr_hyper(offset, tl);
3156 tl += 2;
3157 *tl = txdr_unsigned(cnt);
3158 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_COMMIT, td,
3159 nfs_vpcred(vp, ND_WRITE), &error));
3160 ERROROUT(nfsm_wcc_data(&info, vp, &wccflag));
3161 if (!error) {
3162 NULLOUT(tl = nfsm_dissect(&info, NFSX_V3WRITEVERF));
3163 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
3164 NFSX_V3WRITEVERF)) {
3165 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
3166 NFSX_V3WRITEVERF);
3167 error = NFSERR_STALEWRITEVERF;
3168 }
3169 }
3170 m_freem(info.mrep);
3171 info.mrep = NULL;
3172 nfsmout:
3173 return (error);
3174 }
3175
3176 /*
3177 * Kludge City..
3178 * - make nfs_bmap() essentially a no-op that does no translation
3179 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3180 * (Maybe I could use the process's page mapping, but I was concerned that
3181 * Kernel Write might not be enabled and also figured copyout() would do
3182 * a lot more work than bcopy() and also it currently happens in the
3183 * context of the swapper process (2).
3184 *
3185 * nfs_bmap(struct vnode *a_vp, off_t a_loffset,
3186 * off_t *a_doffsetp, int *a_runp, int *a_runb)
3187 */
3188 static int
nfs_bmap(struct vop_bmap_args * ap)3189 nfs_bmap(struct vop_bmap_args *ap)
3190 {
3191 /* no token lock required */
3192 if (ap->a_doffsetp != NULL)
3193 *ap->a_doffsetp = ap->a_loffset;
3194 if (ap->a_runp != NULL)
3195 *ap->a_runp = 0;
3196 if (ap->a_runb != NULL)
3197 *ap->a_runb = 0;
3198 return (0);
3199 }
3200
3201 /*
3202 * Strategy routine.
3203 */
3204 static int
nfs_strategy(struct vop_strategy_args * ap)3205 nfs_strategy(struct vop_strategy_args *ap)
3206 {
3207 struct bio *bio = ap->a_bio;
3208 struct bio *nbio;
3209 struct buf *bp __debugvar = bio->bio_buf;
3210 struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3211 struct thread *td;
3212 int error;
3213
3214 KASSERT(bp->b_cmd != BUF_CMD_DONE,
3215 ("nfs_strategy: buffer %p unexpectedly marked done", bp));
3216 KASSERT(BUF_LOCKINUSE(bp),
3217 ("nfs_strategy: buffer %p not locked", bp));
3218
3219 if (bio->bio_flags & BIO_SYNC)
3220 td = curthread; /* XXX */
3221 else
3222 td = NULL;
3223
3224 lwkt_gettoken(&nmp->nm_token);
3225
3226 /*
3227 * We probably don't need to push an nbio any more since no
3228 * block conversion is required due to the use of 64 bit byte
3229 * offsets, but do it anyway.
3230 *
3231 * NOTE: When NFS callers itself via this strategy routines and
3232 * sets up a synchronous I/O, it expects the I/O to run
3233 * synchronously (its bio_done routine just assumes it),
3234 * so for now we have to honor the bit.
3235 */
3236 nbio = push_bio(bio);
3237 nbio->bio_offset = bio->bio_offset;
3238 nbio->bio_flags = bio->bio_flags & BIO_SYNC;
3239
3240 /*
3241 * If the op is asynchronous and an i/o daemon is waiting
3242 * queue the request, wake it up and wait for completion
3243 * otherwise just do it ourselves.
3244 */
3245 if (bio->bio_flags & BIO_SYNC) {
3246 error = nfs_doio(ap->a_vp, nbio, td);
3247 } else {
3248 nfs_asyncio(ap->a_vp, nbio);
3249 error = 0;
3250 }
3251 lwkt_reltoken(&nmp->nm_token);
3252
3253 return (error);
3254 }
3255
3256 /*
3257 * fsync vnode op. Just call nfs_flush() with commit == 1.
3258 *
3259 * nfs_fsync(struct vnode *a_vp, int a_waitfor)
3260 */
3261 /* ARGSUSED */
3262 static int
nfs_fsync(struct vop_fsync_args * ap)3263 nfs_fsync(struct vop_fsync_args *ap)
3264 {
3265 struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3266 int error;
3267
3268 lwkt_gettoken(&nmp->nm_token);
3269
3270 /*
3271 * NOTE: Because attributes are set synchronously we currently
3272 * do not have to implement vsetisdirty()/vclrisdirty().
3273 */
3274 error = nfs_flush(ap->a_vp, ap->a_waitfor, curthread, 1);
3275
3276 lwkt_reltoken(&nmp->nm_token);
3277
3278 return error;
3279 }
3280
3281 /*
3282 * Flush all the blocks associated with a vnode. Dirty NFS buffers may be
3283 * in one of two states: If B_NEEDCOMMIT is clear then the buffer contains
3284 * new NFS data which needs to be written to the server. If B_NEEDCOMMIT is
3285 * set the buffer contains data that has already been written to the server
3286 * and which now needs a commit RPC.
3287 *
3288 * If commit is 0 we only take one pass and only flush buffers containing new
3289 * dirty data.
3290 *
3291 * If commit is 1 we take two passes, issuing a commit RPC in the second
3292 * pass.
3293 *
3294 * If waitfor is MNT_WAIT and commit is 1, we loop as many times as required
3295 * to completely flush all pending data.
3296 *
3297 * Note that the RB_SCAN code properly handles the case where the
3298 * callback might block and directly or indirectly (another thread) cause
3299 * the RB tree to change.
3300 */
3301
3302 #ifndef NFS_COMMITBVECSIZ
3303 #define NFS_COMMITBVECSIZ 16
3304 #endif
3305
3306 struct nfs_flush_info {
3307 enum { NFI_FLUSHNEW, NFI_COMMIT } mode;
3308 struct thread *td;
3309 struct vnode *vp;
3310 int waitfor;
3311 int slpflag;
3312 int slptimeo;
3313 int loops;
3314 struct buf *bvary[NFS_COMMITBVECSIZ];
3315 int bvsize;
3316 off_t beg_off;
3317 off_t end_off;
3318 };
3319
3320 static int nfs_flush_bp(struct buf *bp, void *data);
3321 static int nfs_flush_docommit(struct nfs_flush_info *info, int error);
3322
3323 int
nfs_flush(struct vnode * vp,int waitfor,struct thread * td,int commit)3324 nfs_flush(struct vnode *vp, int waitfor, struct thread *td, int commit)
3325 {
3326 struct nfsnode *np = VTONFS(vp);
3327 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3328 struct nfs_flush_info info;
3329 int error;
3330
3331 bzero(&info, sizeof(info));
3332 info.td = td;
3333 info.vp = vp;
3334 info.waitfor = waitfor;
3335 info.slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
3336 info.loops = 0;
3337 lwkt_gettoken(&vp->v_token);
3338
3339 do {
3340 /*
3341 * Flush mode
3342 */
3343 info.mode = NFI_FLUSHNEW;
3344 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3345 nfs_flush_bp, &info);
3346
3347 /*
3348 * Take a second pass if committing and no error occured.
3349 * Clean up any left over collection (whether an error
3350 * occurs or not).
3351 */
3352 if (commit && error == 0) {
3353 info.mode = NFI_COMMIT;
3354 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
3355 nfs_flush_bp, &info);
3356 if (info.bvsize)
3357 error = nfs_flush_docommit(&info, error);
3358 }
3359
3360 /*
3361 * Wait for pending I/O to complete before checking whether
3362 * any further dirty buffers exist.
3363 */
3364 while (waitfor == MNT_WAIT &&
3365 bio_track_active(&vp->v_track_write)) {
3366 error = bio_track_wait(&vp->v_track_write,
3367 info.slpflag, info.slptimeo);
3368 if (error) {
3369 /*
3370 * We have to be able to break out if this
3371 * is an 'intr' mount.
3372 */
3373 if (nfs_sigintr(nmp, NULL, td)) {
3374 error = -EINTR;
3375 break;
3376 }
3377
3378 /*
3379 * Since we do not process pending signals,
3380 * once we get a PCATCH our tsleep() will no
3381 * longer sleep, switch to a fixed timeout
3382 * instead.
3383 */
3384 if (info.slpflag == PCATCH) {
3385 info.slpflag = 0;
3386 info.slptimeo = 2 * hz;
3387 }
3388 error = 0;
3389 }
3390 }
3391 ++info.loops;
3392 /*
3393 * Loop if we are flushing synchronous as well as committing,
3394 * and dirty buffers are still present. Otherwise we might livelock.
3395 */
3396 } while (waitfor == MNT_WAIT && commit &&
3397 error == 0 && !RB_EMPTY(&vp->v_rbdirty_tree));
3398
3399 /*
3400 * The callbacks have to return a negative error to terminate the
3401 * RB scan.
3402 */
3403 if (error < 0)
3404 error = -error;
3405
3406 /*
3407 * Deal with any error collection
3408 */
3409 if (np->n_flag & NWRITEERR) {
3410 error = np->n_error;
3411 np->n_flag &= ~NWRITEERR;
3412 }
3413 lwkt_reltoken(&vp->v_token);
3414 return (error);
3415 }
3416
3417 static
3418 int
nfs_flush_bp(struct buf * bp,void * data)3419 nfs_flush_bp(struct buf *bp, void *data)
3420 {
3421 struct nfs_flush_info *info = data;
3422 int lkflags;
3423 int error;
3424 off_t toff;
3425
3426 error = 0;
3427 switch(info->mode) {
3428 case NFI_FLUSHNEW:
3429 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3430 if (error && info->loops && info->waitfor == MNT_WAIT) {
3431 error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT);
3432 if (error) {
3433 lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
3434 if (info->slpflag & PCATCH)
3435 lkflags |= LK_PCATCH;
3436 error = BUF_TIMELOCK(bp, lkflags, "nfsfsync",
3437 info->slptimeo);
3438 }
3439 }
3440
3441 /*
3442 * Ignore locking errors
3443 */
3444 if (error) {
3445 error = 0;
3446 break;
3447 }
3448
3449 /*
3450 * The buffer may have changed out from under us, even if
3451 * we did not block (MPSAFE). Check again now that it is
3452 * locked.
3453 */
3454 if (bp->b_vp == info->vp &&
3455 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == B_DELWRI) {
3456 bremfree(bp);
3457 bawrite(bp);
3458 } else {
3459 BUF_UNLOCK(bp);
3460 }
3461 break;
3462 case NFI_COMMIT:
3463 /*
3464 * Only process buffers in need of a commit which we can
3465 * immediately lock. This may prevent a buffer from being
3466 * committed, but the normal flush loop will block on the
3467 * same buffer so we shouldn't get into an endless loop.
3468 */
3469 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3470 (B_DELWRI | B_NEEDCOMMIT)) {
3471 break;
3472 }
3473 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
3474 break;
3475
3476 /*
3477 * We must recheck after successfully locking the buffer.
3478 */
3479 if (bp->b_vp != info->vp ||
3480 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3481 (B_DELWRI | B_NEEDCOMMIT)) {
3482 BUF_UNLOCK(bp);
3483 break;
3484 }
3485
3486 /*
3487 * NOTE: storing the bp in the bvary[] basically sets
3488 * it up for a commit operation.
3489 *
3490 * We must call vfs_busy_pages() now so the commit operation
3491 * is interlocked with user modifications to memory mapped
3492 * pages. The b_dirtyoff/b_dirtyend range is not correct
3493 * until after the pages have been busied.
3494 *
3495 * Note: to avoid loopback deadlocks, we do not
3496 * assign b_runningbufspace.
3497 */
3498 bremfree(bp);
3499 bp->b_cmd = BUF_CMD_WRITE;
3500 vfs_busy_pages(bp->b_vp, bp);
3501 info->bvary[info->bvsize] = bp;
3502 toff = bp->b_bio2.bio_offset + bp->b_dirtyoff;
3503 if (info->bvsize == 0 || toff < info->beg_off)
3504 info->beg_off = toff;
3505 toff += (off_t)(bp->b_dirtyend - bp->b_dirtyoff);
3506 if (info->bvsize == 0 || toff > info->end_off)
3507 info->end_off = toff;
3508 ++info->bvsize;
3509 if (info->bvsize == NFS_COMMITBVECSIZ) {
3510 error = nfs_flush_docommit(info, 0);
3511 KKASSERT(info->bvsize == 0);
3512 }
3513 }
3514 return (error);
3515 }
3516
3517 static
3518 int
nfs_flush_docommit(struct nfs_flush_info * info,int error)3519 nfs_flush_docommit(struct nfs_flush_info *info, int error)
3520 {
3521 struct vnode *vp;
3522 struct buf *bp;
3523 off_t bytes;
3524 int retv;
3525 int i;
3526
3527 vp = info->vp;
3528
3529 if (info->bvsize > 0) {
3530 /*
3531 * Commit data on the server, as required. Note that
3532 * nfs_commit will use the vnode's cred for the commit.
3533 * The NFSv3 commit RPC is limited to a 32 bit byte count.
3534 */
3535 bytes = info->end_off - info->beg_off;
3536 if (bytes > 0x40000000)
3537 bytes = 0x40000000;
3538 if (error) {
3539 retv = -error;
3540 } else {
3541 retv = nfs_commitrpc_uio(vp, info->beg_off,
3542 (int)bytes, info->td);
3543 if (retv == NFSERR_STALEWRITEVERF)
3544 nfs_clearcommit(vp->v_mount);
3545 }
3546
3547 /*
3548 * Now, either mark the blocks I/O done or mark the
3549 * blocks dirty, depending on whether the commit
3550 * succeeded.
3551 */
3552 for (i = 0; i < info->bvsize; ++i) {
3553 bp = info->bvary[i];
3554 if (retv || (bp->b_flags & B_NEEDCOMMIT) == 0) {
3555 /*
3556 * Either an error or the original
3557 * vfs_busy_pages() cleared B_NEEDCOMMIT
3558 * due to finding new dirty VM pages in
3559 * the buffer.
3560 *
3561 * Leave B_DELWRI intact.
3562 */
3563 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3564 vfs_unbusy_pages(bp);
3565 bp->b_cmd = BUF_CMD_DONE;
3566 bqrelse(bp);
3567 } else {
3568 /*
3569 * Success, remove B_DELWRI ( bundirty() ).
3570 *
3571 * b_dirtyoff/b_dirtyend seem to be NFS
3572 * specific. We should probably move that
3573 * into bundirty(). XXX
3574 *
3575 * We are faking an I/O write, we have to
3576 * start the transaction in order to
3577 * immediately biodone() it.
3578 */
3579 bundirty(bp);
3580 bp->b_flags &= ~B_ERROR;
3581 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3582 bp->b_dirtyoff = bp->b_dirtyend = 0;
3583 biodone(&bp->b_bio1);
3584 }
3585 }
3586 info->bvsize = 0;
3587 }
3588 return (error);
3589 }
3590
3591 /*
3592 * NFS advisory byte-level locks.
3593 * Currently unsupported.
3594 *
3595 * nfs_advlock(struct vnode *a_vp, caddr_t a_id, int a_op, struct flock *a_fl,
3596 * int a_flags)
3597 */
3598 static int
nfs_advlock(struct vop_advlock_args * ap)3599 nfs_advlock(struct vop_advlock_args *ap)
3600 {
3601 struct nfsnode *np = VTONFS(ap->a_vp);
3602
3603 /* no token lock currently required */
3604 /*
3605 * The following kludge is to allow diskless support to work
3606 * until a real NFS lockd is implemented. Basically, just pretend
3607 * that this is a local lock.
3608 */
3609 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
3610 }
3611
3612 /*
3613 * Print out the contents of an nfsnode.
3614 *
3615 * nfs_print(struct vnode *a_vp)
3616 */
3617 static int
nfs_print(struct vop_print_args * ap)3618 nfs_print(struct vop_print_args *ap)
3619 {
3620 struct vnode *vp = ap->a_vp;
3621 struct nfsnode *np = VTONFS(vp);
3622
3623 kprintf("tag VT_NFS, fileid %lld fsid 0x%x",
3624 (long long)np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3625 if (vp->v_type == VFIFO)
3626 fifo_printinfo(vp);
3627 kprintf("\n");
3628 return (0);
3629 }
3630
3631 /*
3632 * nfs special file access vnode op.
3633 *
3634 * nfs_laccess(struct vnode *a_vp, int a_mode, struct ucred *a_cred)
3635 */
3636 static int
nfs_laccess(struct vop_access_args * ap)3637 nfs_laccess(struct vop_access_args *ap)
3638 {
3639 struct nfsmount *nmp = VFSTONFS(ap->a_vp->v_mount);
3640 struct vattr vattr;
3641 int error;
3642
3643 lwkt_gettoken(&nmp->nm_token);
3644 error = VOP_GETATTR(ap->a_vp, &vattr);
3645 if (error == 0) {
3646 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid,
3647 vattr.va_mode, 0);
3648 }
3649 lwkt_reltoken(&nmp->nm_token);
3650
3651 return (error);
3652 }
3653
3654 /*
3655 * Read wrapper for fifos.
3656 *
3657 * nfsfifo_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3658 * struct ucred *a_cred)
3659 */
3660 static int
nfsfifo_read(struct vop_read_args * ap)3661 nfsfifo_read(struct vop_read_args *ap)
3662 {
3663 struct nfsnode *np = VTONFS(ap->a_vp);
3664
3665 /* no token access required */
3666 /*
3667 * Set access flag.
3668 */
3669 np->n_flag |= NACC;
3670 getnanotime(&np->n_atim);
3671 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3672 }
3673
3674 /*
3675 * Write wrapper for fifos.
3676 *
3677 * nfsfifo_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
3678 * struct ucred *a_cred)
3679 */
3680 static int
nfsfifo_write(struct vop_write_args * ap)3681 nfsfifo_write(struct vop_write_args *ap)
3682 {
3683 struct nfsnode *np = VTONFS(ap->a_vp);
3684
3685 /* no token access required */
3686 /*
3687 * Set update flag.
3688 */
3689 np->n_flag |= NUPD;
3690 getnanotime(&np->n_mtim);
3691 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3692 }
3693
3694 /*
3695 * Close wrapper for fifos.
3696 *
3697 * Update the times on the nfsnode then do fifo close.
3698 *
3699 * nfsfifo_close(struct vnode *a_vp, int a_fflag)
3700 */
3701 static int
nfsfifo_close(struct vop_close_args * ap)3702 nfsfifo_close(struct vop_close_args *ap)
3703 {
3704 struct vnode *vp = ap->a_vp;
3705 struct nfsnode *np = VTONFS(vp);
3706 struct vattr vattr;
3707 struct timespec ts;
3708
3709 /* no token access required */
3710
3711 vn_lock(vp, LK_UPGRADE | LK_RETRY); /* XXX */
3712 if (np->n_flag & (NACC | NUPD)) {
3713 getnanotime(&ts);
3714 if (np->n_flag & NACC)
3715 np->n_atim = ts;
3716 if (np->n_flag & NUPD)
3717 np->n_mtim = ts;
3718 np->n_flag |= NCHG;
3719 if (VREFCNT(vp) == 1 &&
3720 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3721 VATTR_NULL(&vattr);
3722 if (np->n_flag & NACC)
3723 vattr.va_atime = np->n_atim;
3724 if (np->n_flag & NUPD)
3725 vattr.va_mtime = np->n_mtim;
3726 (void)VOP_SETATTR(vp, &vattr, nfs_vpcred(vp, ND_WRITE));
3727 }
3728 }
3729 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
3730 }
3731
3732 /************************************************************************
3733 * KQFILTER OPS *
3734 ************************************************************************/
3735
3736 static void filt_nfsdetach(struct knote *kn);
3737 static int filt_nfsread(struct knote *kn, long hint);
3738 static int filt_nfswrite(struct knote *kn, long hint);
3739 static int filt_nfsvnode(struct knote *kn, long hint);
3740
3741 static struct filterops nfsread_filtops =
3742 { FILTEROP_ISFD | FILTEROP_MPSAFE,
3743 NULL, filt_nfsdetach, filt_nfsread };
3744 static struct filterops nfswrite_filtops =
3745 { FILTEROP_ISFD | FILTEROP_MPSAFE,
3746 NULL, filt_nfsdetach, filt_nfswrite };
3747 static struct filterops nfsvnode_filtops =
3748 { FILTEROP_ISFD | FILTEROP_MPSAFE,
3749 NULL, filt_nfsdetach, filt_nfsvnode };
3750
3751 static int
nfs_kqfilter(struct vop_kqfilter_args * ap)3752 nfs_kqfilter (struct vop_kqfilter_args *ap)
3753 {
3754 struct vnode *vp = ap->a_vp;
3755 struct knote *kn = ap->a_kn;
3756
3757 switch (kn->kn_filter) {
3758 case EVFILT_READ:
3759 kn->kn_fop = &nfsread_filtops;
3760 break;
3761 case EVFILT_WRITE:
3762 kn->kn_fop = &nfswrite_filtops;
3763 break;
3764 case EVFILT_VNODE:
3765 kn->kn_fop = &nfsvnode_filtops;
3766 break;
3767 default:
3768 return (EOPNOTSUPP);
3769 }
3770
3771 kn->kn_hook = (caddr_t)vp;
3772
3773 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
3774
3775 return(0);
3776 }
3777
3778 static void
filt_nfsdetach(struct knote * kn)3779 filt_nfsdetach(struct knote *kn)
3780 {
3781 struct vnode *vp = (void *)kn->kn_hook;
3782
3783 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
3784 }
3785
3786 static int
filt_nfsread(struct knote * kn,long hint)3787 filt_nfsread(struct knote *kn, long hint)
3788 {
3789 struct vnode *vp = (void *)kn->kn_hook;
3790 struct nfsnode *node = VTONFS(vp);
3791 off_t off;
3792
3793 if (hint == NOTE_REVOKE) {
3794 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
3795 return(1);
3796 }
3797
3798 /*
3799 * Interlock against MP races when performing this function. XXX
3800 */
3801 /* TMPFS_NODE_LOCK_SH(node); */
3802 off = node->n_size - kn->kn_fp->f_offset;
3803 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
3804 if (kn->kn_sfflags & NOTE_OLDAPI) {
3805 /* TMPFS_NODE_UNLOCK(node); */
3806 return(1);
3807 }
3808 if (kn->kn_data == 0) {
3809 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
3810 }
3811 /* TMPFS_NODE_UNLOCK(node); */
3812 return (kn->kn_data != 0);
3813 }
3814
3815 static int
filt_nfswrite(struct knote * kn,long hint)3816 filt_nfswrite(struct knote *kn, long hint)
3817 {
3818 if (hint == NOTE_REVOKE)
3819 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
3820 kn->kn_data = 0;
3821 return (1);
3822 }
3823
3824 static int
filt_nfsvnode(struct knote * kn,long hint)3825 filt_nfsvnode(struct knote *kn, long hint)
3826 {
3827 if (kn->kn_sfflags & hint)
3828 kn->kn_fflags |= hint;
3829 if (hint == NOTE_REVOKE) {
3830 kn->kn_flags |= (EV_EOF | EV_NODATA);
3831 return (1);
3832 }
3833 return (kn->kn_fflags != 0);
3834 }
3835