1 /* $Id: nfs_vnops.c,v 1.45 2003/11/05 14:59:02 rees Exp $ */
2
3 /*-
4 * copyright (c) 2003
5 * the regents of the university of michigan
6 * all rights reserved
7 *
8 * permission is granted to use, copy, create derivative works and redistribute
9 * this software and such derivative works for any purpose, so long as the name
10 * of the university of michigan is not used in any advertising or publicity
11 * pertaining to the use or distribution of this software without specific,
12 * written prior authorization. if the above copyright notice or any other
13 * identification of the university of michigan is included in any copy of any
14 * portion of this software, then the disclaimer below must also be included.
15 *
16 * this software is provided as is, without representation from the university
17 * of michigan as to its fitness for any purpose, and without warranty by the
18 * university of michigan of any kind, either express or implied, including
19 * without limitation the implied warranties of merchantability and fitness for
20 * a particular purpose. the regents of the university of michigan shall not be
21 * liable for any damages, including special, indirect, incidental, or
22 * consequential damages, with respect to any claim arising out of or in
23 * connection with the use of the software, even if it has been or is hereafter
24 * advised of the possibility of such damages.
25 */
26
27 /*
28 * Copyright (c) 1989, 1993
29 * The Regents of the University of California. All rights reserved.
30 *
31 * This code is derived from software contributed to Berkeley by
32 * Rick Macklem at The University of Guelph.
33 *
34 * Redistribution and use in source and binary forms, with or without
35 * modification, are permitted provided that the following conditions
36 * are met:
37 * 1. Redistributions of source code must retain the above copyright
38 * notice, this list of conditions and the following disclaimer.
39 * 2. Redistributions in binary form must reproduce the above copyright
40 * notice, this list of conditions and the following disclaimer in the
41 * documentation and/or other materials provided with the distribution.
42 * 4. Neither the name of the University nor the names of its contributors
43 * may be used to endorse or promote products derived from this software
44 * without specific prior written permission.
45 *
46 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
47 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
48 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
49 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
50 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
51 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
52 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
53 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
54 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
55 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
56 * SUCH DAMAGE.
57 *
58 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
59 */
60
61 #include <sys/cdefs.h>
62 __FBSDID("$FreeBSD$");
63
64 /*
65 * vnode op calls for Sun NFS version 2 and 3
66 */
67
68 #include "opt_inet.h"
69
70 #include <sys/param.h>
71 #include <sys/kernel.h>
72 #include <sys/systm.h>
73 #include <sys/resourcevar.h>
74 #include <sys/proc.h>
75 #include <sys/mount.h>
76 #include <sys/bio.h>
77 #include <sys/buf.h>
78 #include <sys/malloc.h>
79 #include <sys/mbuf.h>
80 #include <sys/namei.h>
81 #include <sys/socket.h>
82 #include <sys/vnode.h>
83 #include <sys/dirent.h>
84 #include <sys/fcntl.h>
85 #include <sys/lockf.h>
86 #include <sys/stat.h>
87 #include <sys/sysctl.h>
88 #include <sys/lockmgr.h>
89 #include <sys/signalvar.h>
90
91 #include <vm/vm.h>
92 #include <vm/vm_extern.h>
93
94 #include <fs/fifofs/fifo.h>
95
96 #include <rpc/rpcclnt.h>
97
98 #include <nfs/rpcv2.h>
99 #include <nfs/nfsproto.h>
100 #include <nfsclient/nfs.h>
101 #include <nfs4client/nfs4.h>
102 #include <nfsclient/nfsnode.h>
103 #include <nfsclient/nfsmount.h>
104 #include <nfsclient/nfs_lock.h>
105 #include <nfs/xdr_subs.h>
106 #include <nfsclient/nfsm_subs.h>
107
108 #include <net/if.h>
109 #include <netinet/in.h>
110 #include <netinet/in_var.h>
111
112 /* NFSv4 */
113 #include <nfs4client/nfs4m_subs.h>
114 #include <nfs4client/nfs4_vn.h>
115
116 /* Defs */
117 #define TRUE 1
118 #define FALSE 0
119
120 /*
121 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
122 * calls are not in getblk() and brelse() so that they would not be necessary
123 * here.
124 */
125 #ifndef B_VMIO
126 #define vfs_busy_pages(bp, f)
127 #endif
128
129 static int nfs4_flush(struct vnode *, int, struct thread *,
130 int);
131 static int nfs4_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
132 struct thread *);
133 static int nfs4_closerpc(struct vnode *, struct ucred *, struct thread *, int);
134
135 static vop_lookup_t nfs4_lookup;
136 static vop_create_t nfs4_create;
137 static vop_mknod_t nfs4_mknod;
138 static vop_open_t nfs4_open;
139 static vop_close_t nfs4_close;
140 static vop_access_t nfs4_access;
141 static vop_getattr_t nfs4_getattr;
142 static vop_setattr_t nfs4_setattr;
143 static vop_read_t nfs4_read;
144 static vop_fsync_t nfs4_fsync;
145 static vop_remove_t nfs4_remove;
146 static vop_link_t nfs4_link;
147 static vop_rename_t nfs4_rename;
148 static vop_mkdir_t nfs4_mkdir;
149 static vop_rmdir_t nfs4_rmdir;
150 static vop_symlink_t nfs4_symlink;
151 static vop_readdir_t nfs4_readdir;
152 static vop_strategy_t nfs4_strategy;
153 static int nfs4_lookitup(struct vnode *, const char *, int,
154 struct ucred *, struct thread *, struct nfsnode **);
155 static int nfs4_sillyrename(struct vnode *, struct vnode *,
156 struct componentname *);
157 static vop_readlink_t nfs4_readlink;
158 static vop_print_t nfs4_print;
159 static vop_advlock_t nfs4_advlock;
160
161 /*
162 * Global vfs data structures for nfs
163 */
164 struct vop_vector nfs4_vnodeops = {
165 .vop_default = &default_vnodeops,
166 .vop_access = nfs4_access,
167 .vop_advlock = nfs4_advlock,
168 .vop_close = nfs4_close,
169 .vop_create = nfs4_create,
170 .vop_fsync = nfs4_fsync,
171 .vop_getattr = nfs4_getattr,
172 .vop_getpages = nfs_getpages,
173 .vop_putpages = nfs_putpages,
174 .vop_inactive = nfs_inactive,
175 .vop_lease = VOP_NULL,
176 .vop_link = nfs4_link,
177 .vop_lookup = nfs4_lookup,
178 .vop_mkdir = nfs4_mkdir,
179 .vop_mknod = nfs4_mknod,
180 .vop_open = nfs4_open,
181 .vop_print = nfs4_print,
182 .vop_read = nfs4_read,
183 .vop_readdir = nfs4_readdir,
184 .vop_readlink = nfs4_readlink,
185 .vop_reclaim = nfs_reclaim,
186 .vop_remove = nfs4_remove,
187 .vop_rename = nfs4_rename,
188 .vop_rmdir = nfs4_rmdir,
189 .vop_setattr = nfs4_setattr,
190 .vop_strategy = nfs4_strategy,
191 .vop_symlink = nfs4_symlink,
192 .vop_write = nfs_write,
193 };
194
195 static int nfs4_removerpc(struct vnode *dvp, const char *name, int namelen,
196 struct ucred *cred, struct thread *td);
197 static int nfs4_renamerpc(struct vnode *fdvp, const char *fnameptr,
198 int fnamelen, struct vnode *tdvp,
199 const char *tnameptr, int tnamelen,
200 struct ucred *cred, struct thread *td);
201 static int nfs4_renameit(struct vnode *sdvp, struct componentname *scnp,
202 struct sillyrename *sp);
203 static int nfs4_openrpc(struct vnode *, struct vnode **,
204 struct componentname *, int, struct vattr *);
205 static int nfs4_open_confirm(struct vnode *vp, struct nfs4_compound *cpp,
206 struct nfs4_oparg_open *openap,
207 struct nfs4_oparg_getfh *gfh,
208 struct ucred *cred, struct thread *td);
209 static int nfs4_createrpc(struct vnode *, struct vnode **,
210 struct componentname *, nfstype,
211 struct vattr *, char *);
212
213 /*
214 * Global variables
215 */
216 struct nfs4_lowner nfs4_masterlowner;
217
218 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
219
220 SYSCTL_DECL(_vfs_nfs4);
221
222 static int nfs4_access_cache_timeout = NFS_MAXATTRTIMO;
223 SYSCTL_INT(_vfs_nfs4, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
224 &nfs4_access_cache_timeout, 0, "NFS ACCESS cache timeout");
225
226 #if 0
227 static int nfsv3_commit_on_close = 0;
228 SYSCTL_INT(_vfs_nfs4, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
229 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
230
231 SYSCTL_INT(_vfs_nfs4, OID_AUTO, access_cache_hits, CTLFLAG_RD,
232 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
233
234 SYSCTL_INT(_vfs_nfs4, OID_AUTO, access_cache_misses, CTLFLAG_RD,
235 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
236 #endif
237
238 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
239 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
240 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
241 static int
242 nfs4_v3_access_otw(struct vnode *vp, int wmode, struct thread *td,
243 struct ucred *cred)
244 {
245 const int v3 = 1;
246 u_int32_t *tl;
247 int error = 0, attrflag;
248
249 return (0);
250
251 struct mbuf *mreq, *mrep = NULL, *md, *mb;
252 caddr_t bpos, dpos;
253 u_int32_t rmode;
254 struct nfsnode *np = VTONFS(vp);
255
256 nfsstats.rpccnt[NFSPROC_ACCESS]++;
257 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
258 mb = mreq;
259 bpos = mtod(mb, caddr_t);
260 nfsm_fhtom(vp, v3);
261 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
262 *tl = txdr_unsigned(wmode);
263 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
264 nfsm_postop_attr(vp, attrflag);
265 if (!error) {
266 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
267 rmode = fxdr_unsigned(u_int32_t, *tl);
268 np->n_mode = rmode;
269 np->n_modeuid = cred->cr_uid;
270 np->n_modestamp = time_second;
271 }
272 m_freem(mrep);
273 nfsmout:
274 return error;
275 }
276
277 /*
278 * nfs access vnode op.
279 * For nfs version 2, just return ok. File accesses may fail later.
280 * For nfs version 3, use the access rpc to check accessibility. If file modes
281 * are changed on the server, accesses might still fail later.
282 */
283 static int
284 nfs4_access(struct vop_access_args *ap)
285 {
286 struct vnode *vp = ap->a_vp;
287 int error = 0;
288 u_int32_t mode, wmode;
289 int v3 = NFS_ISV3(vp); /* v3 \in v4 */
290 struct nfsnode *np = VTONFS(vp);
291 caddr_t bpos, dpos;
292 struct mbuf *mreq, *mrep = NULL, *md, *mb;
293 struct nfs4_compound cp;
294 struct nfs4_oparg_access acc;
295 struct thread *td = ap->a_td;
296 struct ucred *cred = ap->a_cred;
297
298 /*
299 * Disallow write attempts on filesystems mounted read-only;
300 * unless the file is a socket, fifo, or a block or character
301 * device resident on the filesystem.
302 */
303 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
304 switch (vp->v_type) {
305 case VREG:
306 case VDIR:
307 case VLNK:
308 return (EROFS);
309 default:
310 break;
311 }
312 }
313 /*
314 * For nfs v3, check to see if we have done this recently, and if
315 * so return our cached result instead of making an ACCESS call.
316 * If not, do an access rpc, otherwise you are stuck emulating
317 * ufs_access() locally using the vattr. This may not be correct,
318 * since the server may apply other access criteria such as
319 * client uid-->server uid mapping that we do not know about.
320 */
321 /* XXX Disable this for now; needs fixing of _access_otw() */
322 if (0 && v3) {
323 if (ap->a_mode & VREAD)
324 mode = NFSV3ACCESS_READ;
325 else
326 mode = 0;
327 if (vp->v_type != VDIR) {
328 if (ap->a_mode & VWRITE)
329 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
330 if (ap->a_mode & VEXEC)
331 mode |= NFSV3ACCESS_EXECUTE;
332 } else {
333 if (ap->a_mode & VWRITE)
334 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
335 NFSV3ACCESS_DELETE);
336 if (ap->a_mode & VEXEC)
337 mode |= NFSV3ACCESS_LOOKUP;
338 }
339 /* XXX safety belt, only make blanket request if caching */
340 if (nfs4_access_cache_timeout > 0) {
341 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
342 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
343 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
344 } else {
345 wmode = mode;
346 }
347
348 /*
349 * Does our cached result allow us to give a definite yes to
350 * this request?
351 */
352 if (time_second < np->n_modestamp + nfs4_access_cache_timeout &&
353 ap->a_cred->cr_uid == np->n_modeuid &&
354 (np->n_mode & mode) == mode) {
355 nfsstats.accesscache_hits++;
356 } else {
357 /*
358 * Either a no, or a don't know. Go to the wire.
359 */
360 nfsstats.accesscache_misses++;
361 error = nfs4_v3_access_otw(vp, wmode, ap->a_td,
362 ap->a_cred);
363 if (error == 0) {
364 if ((np->n_mode & mode) != mode)
365 error = EACCES;
366 }
367 }
368 return (error);
369 }
370
371 /* XXX use generic access code here? */
372 mode = ap->a_mode & VREAD ? NFSV4ACCESS_READ : 0;
373 if (vp->v_type == VDIR) {
374 if (ap->a_mode & VWRITE)
375 mode |= NFSV4ACCESS_MODIFY | NFSV4ACCESS_EXTEND | NFSV4ACCESS_DELETE;
376 if (ap->a_mode & VEXEC)
377 mode |= NFSV4ACCESS_LOOKUP;
378 } else {
379 if (ap->a_mode & VWRITE)
380 mode |= NFSV4ACCESS_MODIFY | NFSV4ACCESS_EXTEND;
381 if (ap->a_mode & VEXEC)
382 mode |= NFSV4ACCESS_EXECUTE;
383 }
384
385 nfs_v4initcompound(&cp);
386 acc.mode = mode;
387
388 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
389 mb = mreq;
390 bpos = mtod(mb, caddr_t);
391
392 nfsm_v4build_compound(&cp, "nfs4_access()");
393 nfsm_v4build_putfh(&cp, vp);
394 nfsm_v4build_access(&cp, &acc);
395 nfsm_v4build_finalize(&cp);
396
397 nfsm_request(vp, NFSV4PROC_COMPOUND, td, cred);
398 if (error != 0)
399 goto nfsmout;
400
401 nfsm_v4dissect_compound(&cp);
402 nfsm_v4dissect_putfh(&cp);
403 nfsm_v4dissect_access(&cp, &acc);
404
405 if ((acc.rmode & mode) != mode)
406 error = EACCES;
407
408 nfsmout:
409 error = nfs_v4postop(&cp, error);
410
411 if (mrep != NULL)
412 m_freem(mrep);
413
414 return (error);
415 }
416
417 static int
418 nfs4_openrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
419 int flags, struct vattr *vap)
420 {
421 struct vnode *vp = *vpp;
422 struct nfs4_oparg_getattr getattr;
423 struct nfs4_oparg_getfh getfh;
424 struct nfs4_oparg_open opena;
425 struct nfs4_compound cp;
426 caddr_t bpos, dpos;
427 int error = 0;
428 struct mbuf *mreq, *mrep = NULL, *md, *mb;
429 struct ucred *cred = cnp->cn_cred;
430 struct thread *td = cnp->cn_thread;
431 struct nfs4_fctx xfc, *fcp;
432 struct nfsnode *np;
433
434 if (vp == NULL) {
435 /* Create a new file */
436 np = NULL;
437 fcp = &xfc;
438 bzero(fcp, sizeof(*fcp));
439 } else {
440 np = VTONFS(vp);
441 fcp = flags & FWRITE ? &np->n_wfc : &np->n_rfc;
442 }
443
444 /*
445 * Since we are currently only one lockowner; we only open the
446 * file once each for reading and writing.
447 */
448 if (fcp->refcnt++ != 0) {
449 *vpp = vp;
450 /*printf("not opening %s\n", np->n_name != NULL ? np->n_name : "");*/
451 return (0);
452 }
453
454 fcp->lop = &nfs4_masterlowner;
455 fcp->np = np;
456
457 nfs_v4initcompound(&cp);
458 cp.nmp = VFSTONFS(dvp->v_mount);
459
460 opena.ctype = NCLNULL;
461 opena.flags = flags;
462 opena.vap = vap;
463 opena.fcp = fcp; /* For lockowner */
464 opena.cnp = cnp;
465
466 getattr.bm = &nfsv4_getattrbm;
467
468 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
469 mb = mreq;
470 bpos = mtod(mb, caddr_t);
471
472 nfsm_v4build_compound(&cp, "nfs4_openrpc()");
473 nfsm_v4build_putfh(&cp, dvp);
474 nfsm_v4build_open(&cp, &opena);
475 nfsm_v4build_getattr(&cp, &getattr);
476 nfsm_v4build_getfh(&cp, &getfh);
477 nfsm_v4build_finalize(&cp);
478
479 nfsm_request(vp != NULL ? vp : dvp, NFSV4PROC_COMPOUND, td, cred);
480 if (error != 0)
481 goto nfsmout;
482
483 nfsm_v4dissect_compound(&cp);
484 nfsm_v4dissect_putfh(&cp);
485 nfsm_v4dissect_open(&cp, &opena);
486 nfsm_v4dissect_getattr(&cp, &getattr);
487 nfsm_v4dissect_getfh(&cp, &getfh);
488
489 error = nfs_v4postop(&cp, error);
490
491 if (opena.rflags & NFSV4OPENRES_CONFIRM) {
492 error = nfs4_open_confirm(vp ? vp : dvp, &cp, &opena, &getfh, cred, td);
493 if (error != 0)
494 goto nfsmout;
495 }
496
497 if (vp == NULL) {
498 /* New file */
499 error = nfs_nget(dvp->v_mount, &getfh.fh_val,
500 getfh.fh_len, &np, LK_EXCLUSIVE);
501 if (error != 0)
502 goto nfsmout;
503
504 vp = NFSTOV(np);
505 np->n_dvp = dvp;
506 np->n_namelen = cnp->cn_namelen; /* XXX memory leaks on these; track! */
507 if (np->n_name != NULL)
508 FREE(np->n_name, M_NFSREQ);
509 MALLOC(np->n_name, u_char *, np->n_namelen + 1, M_NFSREQ, M_WAITOK);
510 bcopy(cnp->cn_nameptr, np->n_name, np->n_namelen);
511 np->n_name[np->n_namelen] = '\0';
512 if (flags & FWRITE)
513 np->n_wfc = *fcp;
514 else
515 np->n_rfc = *fcp;
516
517 /*printf("opened new file %s\n", np->n_name);*/
518
519 nfs4_vnop_loadattrcache(vp, &getattr.fa, NULL);
520 *vpp = vp;
521 } else {
522 /*printf("openend \"old\" %s\n", np->n_name != NULL ? np->n_name : "");*/
523
524 if (flags & O_TRUNC && np->n_size != 0) {
525 struct vattr va;
526
527 VATTR_NULL(&va);
528 va.va_size = 0;
529 error = nfs4_setattrrpc(vp, &va,
530 cnp->cn_cred, cnp->cn_thread);
531 }
532 np->n_attrstamp = 0;
533 }
534
535 nfsmout:
536 if (mrep != NULL)
537 m_freem(mrep);
538
539 return (error);
540 }
541
542 static int
543 nfs4_open_confirm(struct vnode *vp, struct nfs4_compound *cpp,
544 struct nfs4_oparg_open *openap, struct nfs4_oparg_getfh *gfh,
545 struct ucred *cred, struct thread *td)
546 {
547 caddr_t bpos, dpos;
548 int error = 0;
549 struct mbuf *mreq, *mrep = NULL, *md, *mb;
550
551 nfs_v4initcompound(cpp);
552 cpp->nmp = VFSTONFS(vp->v_mount);
553
554 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
555 mb = mreq;
556 bpos = mtod(mb, caddr_t);
557
558 nfsm_v4build_compound(cpp, "nfs4_open_confirm()");
559 nfsm_v4build_putfh_nv(cpp, gfh);
560 nfsm_v4build_open_confirm(cpp, openap);
561 nfsm_v4build_finalize(cpp);
562
563 nfsm_request(vp, NFSV4PROC_COMPOUND, td, cred);
564 if (error != 0)
565 goto nfsmout;
566
567 nfsm_v4dissect_compound(cpp);
568 nfsm_v4dissect_putfh(cpp);
569 nfsm_v4dissect_open_confirm(cpp, openap);
570
571 nfsmout:
572 error = nfs_v4postop(cpp, error);
573
574 if (mrep != NULL)
575 m_freem(mrep);
576
577 return (error);
578 }
579
580
581 /*
582 * nfs open vnode op
583 * Check to see if the type is ok
584 * and that deletion is not in progress.
585 * For paged in text files, you will need to flush the page cache
586 * if consistency is lost.
587 */
588 /* ARGSUSED */
589 static int
590 nfs4_open(struct vop_open_args *ap)
591 {
592 struct vnode *vp = ap->a_vp;
593 struct nfsnode *np = VTONFS(vp);
594 enum vtype vtype = vp->v_type;
595 int mode = ap->a_mode;
596 struct componentname cn;
597
598 if (vtype != VREG) {
599 if (vtype != VDIR && vtype != VLNK) {
600 #ifdef DIAGNOSTIC
601 printf("open eacces vtyp=%d\n", vp->v_type);
602 #endif
603 return (EACCES);
604 } else
605 return (0);
606 }
607
608 if (np->n_flag & NCREATED) {
609 np->n_flag &= ~NCREATED;
610 return (0);
611 }
612
613 cn.cn_nameptr = np->n_name;
614 cn.cn_namelen = np->n_namelen;
615 cn.cn_cred = ap->a_cred;
616 cn.cn_thread = ap->a_td;
617
618 return (nfs4_openrpc(np->n_dvp, &vp, &cn, mode, NULL));
619 }
620
621 static int
622 nfs4_closerpc(struct vnode *vp, struct ucred *cred, struct thread *td, int flags)
623 {
624 caddr_t bpos, dpos;
625 int error = 0;
626 struct mbuf *mreq, *mrep = NULL, *md, *mb;
627 struct nfs4_fctx *fcp;
628 struct nfs4_compound cp;
629 struct nfsnode *np = VTONFS(vp);
630
631 fcp = flags & FWRITE ? &np->n_wfc : &np->n_rfc;
632
633 nfs_v4initcompound(&cp);
634
635 if (--fcp->refcnt != 0)
636 return (0);
637
638 /*printf("closing %s\n", np->n_name != NULL ? np->n_name : "");*/
639
640 cp.fcp = fcp;
641
642 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
643 mb = mreq;
644 bpos = mtod(mb, caddr_t);
645
646 nfsm_v4build_compound(&cp, "nfs4_closerpc()");
647 nfsm_v4build_putfh(&cp, vp);
648 nfsm_v4build_close(&cp, fcp);
649 nfsm_v4build_finalize(&cp);
650
651 nfsm_request(vp, NFSV4PROC_COMPOUND, td, cred);
652 if (error != 0)
653 goto nfsmout;
654
655 nfsm_v4dissect_compound(&cp);
656 nfsm_v4dissect_putfh(&cp);
657 nfsm_v4dissect_close(&cp, fcp);
658
659 nfsmout:
660 error = nfs_v4postop(&cp, error);
661
662 if (mrep != NULL)
663 m_freem(mrep);
664
665 return (error);
666 }
667
668 /*
669 * nfs close vnode op
670 * play it safe for now (see comments in v2/v3 nfs_close regarding dirty buffers)
671 */
672 /* ARGSUSED */
673 static int
674 nfs4_close(struct vop_close_args *ap)
675 {
676 struct vnode *vp = ap->a_vp;
677 struct nfsnode *np = VTONFS(vp);
678 int error = 0;
679
680 if (vp->v_type != VREG)
681 return (0);
682
683 if (np->n_flag & NMODIFIED) {
684 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
685 np->n_attrstamp = 0;
686 }
687
688 error = nfs4_closerpc(vp, ap->a_cred, ap->a_td, ap->a_fflag);
689
690 if (!error && np->n_flag & NWRITEERR) {
691 np->n_flag &= ~NWRITEERR;
692 error = np->n_error;
693 }
694 return (error);
695 }
696
697 /*
698 * nfs getattr call from vfs.
699 */
700 static int
701 nfs4_getattr(struct vop_getattr_args *ap)
702 {
703 struct vnode *vp = ap->a_vp;
704 struct nfsnode *np = VTONFS(vp);
705 caddr_t bpos, dpos;
706 int error = 0;
707 struct mbuf *mreq, *mrep = NULL, *md, *mb;
708 struct nfs4_oparg_getattr ga;
709 struct nfs4_compound cp;
710
711 /*
712 * Update local times for special files.
713 */
714 if (np->n_flag & (NACC | NUPD))
715 np->n_flag |= NCHG;
716 /*
717 * First look in the cache.
718 */
719 if (nfs_getattrcache(vp, ap->a_vap) == 0)
720 return (0);
721
722 nfsstats.rpccnt[NFSPROC_GETATTR]++;
723
724 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, NFSX_FH(1));
725 mb = mreq;
726 bpos = mtod(mb, caddr_t);
727
728 ga.bm = &nfsv4_getattrbm;
729 nfs_v4initcompound(&cp);
730
731 nfsm_v4build_compound(&cp, "nfs4_getattr()");
732 nfsm_v4build_putfh(&cp, vp);
733 nfsm_v4build_getattr(&cp, &ga);
734 nfsm_v4build_finalize(&cp);
735
736 nfsm_request(vp, NFSV4PROC_COMPOUND, ap->a_td, ap->a_cred);
737 if (error != 0)
738 goto nfsmout;
739
740 nfsm_v4dissect_compound(&cp);
741 nfsm_v4dissect_putfh(&cp);
742 nfsm_v4dissect_getattr(&cp, &ga);
743
744 nfs4_vnop_loadattrcache(vp, &ga.fa, ap->a_vap);
745
746 nfsmout:
747 error = nfs_v4postop(&cp, error);
748
749 if (mrep != NULL)
750 m_freem(mrep);
751 return (error);
752 }
753
754 /*
755 * nfs setattr call.
756 */
757 static int
758 nfs4_setattr(struct vop_setattr_args *ap)
759 {
760 struct vnode *vp = ap->a_vp;
761 struct nfsnode *np = VTONFS(vp);
762 struct vattr *vap = ap->a_vap;
763 int error = 0;
764 u_quad_t tsize;
765
766 #ifndef nolint
767 tsize = (u_quad_t)0;
768 #endif
769
770 /*
771 * Setting of flags is not supported.
772 */
773 if (vap->va_flags != VNOVAL)
774 return (EOPNOTSUPP);
775
776 /*
777 * Disallow write attempts if the filesystem is mounted read-only.
778 */
779 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
780 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
781 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
782 (vp->v_mount->mnt_flag & MNT_RDONLY))
783 return (EROFS);
784 if (vap->va_size != VNOVAL) {
785 switch (vp->v_type) {
786 case VDIR:
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 return (0);
798 vap->va_size = VNOVAL;
799 break;
800 default:
801 /*
802 * Disallow write attempts if the filesystem is
803 * mounted read-only.
804 */
805 if (vp->v_mount->mnt_flag & MNT_RDONLY)
806 return (EROFS);
807
808 /*
809 * We run vnode_pager_setsize() early (why?),
810 * we must set np->n_size now to avoid vinvalbuf
811 * V_SAVE races that might setsize a lower
812 * value.
813 */
814
815 tsize = np->n_size;
816 error = nfs_meta_setsize(vp, ap->a_cred,
817 ap->a_td, vap->va_size);
818
819 if (np->n_flag & NMODIFIED) {
820 if (vap->va_size == 0)
821 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
822 else
823 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
824 if (error) {
825 vnode_pager_setsize(vp, np->n_size);
826 return (error);
827 }
828 }
829 /*
830 * np->n_size has already been set to vap->va_size
831 * in nfs_meta_setsize(). We must set it again since
832 * nfs_loadattrcache() could be called through
833 * nfs_meta_setsize() and could modify np->n_size.
834 */
835 np->n_vattr.va_size = np->n_size = vap->va_size;
836 };
837 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
838 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
839 vp->v_type == VREG &&
840 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) == EINTR)
841 return (error);
842
843 if (vap->va_size != VNOVAL && np->n_wfc.refcnt == 0) {
844 /* Have to open the file before we can truncate it */
845 struct componentname cn;
846
847 cn.cn_nameptr = np->n_name;
848 cn.cn_namelen = np->n_namelen;
849 cn.cn_cred = ap->a_cred;
850 cn.cn_thread = ap->a_td;
851 error = nfs4_openrpc(np->n_dvp, &vp, &cn, FWRITE, NULL);
852 if (error)
853 return error;
854 np->n_flag |= NTRUNCATE;
855 }
856
857 error = nfs4_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
858 if (error && vap->va_size != VNOVAL) {
859 np->n_size = np->n_vattr.va_size = tsize;
860 vnode_pager_setsize(vp, np->n_size);
861 }
862 return (error);
863 }
864
865 /*
866 * Do an nfs setattr rpc.
867 */
868 static int
869 nfs4_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
870 struct thread *td)
871 {
872 caddr_t bpos, dpos;
873 int error = 0;
874 struct mbuf *mreq, *mrep = NULL, *md, *mb;
875 struct nfs4_compound cp;
876 struct nfs4_oparg_getattr ga;
877 struct nfsnode *np = VTONFS(vp);
878 struct nfs4_fctx *fcp;
879
880 nfsstats.rpccnt[NFSPROC_SETATTR]++;
881 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
882 mb = mreq;
883 bpos = mtod(mb, caddr_t);
884
885 ga.bm = &nfsv4_getattrbm;
886 fcp = (vap->va_size != VNOVAL) ? &np->n_wfc : NULL;
887 nfs_v4initcompound(&cp);
888
889 nfsm_v4build_compound(&cp, "nfs4_setattrrpc");
890 nfsm_v4build_putfh(&cp, vp);
891 nfsm_v4build_setattr(&cp, vap, fcp);
892 nfsm_v4build_getattr(&cp, &ga);
893 nfsm_v4build_finalize(&cp);
894
895 nfsm_request(vp, NFSV4PROC_COMPOUND, td, cred);
896 if (error != 0)
897 goto nfsmout;
898
899 nfsm_v4dissect_compound(&cp);
900 nfsm_v4dissect_putfh(&cp);
901 nfsm_v4dissect_setattr(&cp);
902 nfsm_v4dissect_getattr(&cp, &ga);
903
904 nfs4_vnop_loadattrcache(vp, &ga.fa, NULL);
905
906 /* TODO: do the settatr and close in a single compound rpc */
907 if (np->n_flag & NTRUNCATE) {
908 error = nfs4_closerpc(vp, cred, td, FWRITE);
909 np->n_flag &= ~NTRUNCATE;
910 }
911
912 nfsmout:
913 error = nfs_v4postop(&cp, error);
914
915 if (mrep != NULL)
916 m_freem(mrep);
917
918 return (error);
919 }
920
921 /*
922 * nfs lookup call, one step at a time...
923 * First look in cache
924 * If not found, unlock the directory nfsnode and do the rpc
925 */
926 static int
927 nfs4_lookup(struct vop_lookup_args *ap)
928 {
929 struct componentname *cnp = ap->a_cnp;
930 struct vnode *dvp = ap->a_dvp;
931 struct vnode **vpp = ap->a_vpp;
932 int isdot, flags = cnp->cn_flags;
933 struct vnode *newvp;
934 struct nfsmount *nmp;
935 caddr_t bpos, dpos;
936 struct mbuf *mreq, *mrep = NULL, *md, *mb;
937 long len;
938 nfsfh_t *fhp;
939 struct nfsnode *np;
940 int error = 0, fhsize;
941 struct thread *td = cnp->cn_thread;
942 struct nfs4_compound cp;
943 struct nfs4_oparg_getattr ga, dga;
944 struct nfs4_oparg_lookup l;
945 struct nfs4_oparg_getfh gfh;
946
947 *vpp = NULLVP;
948 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
949 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
950 return (EROFS);
951 if (dvp->v_type != VDIR)
952 return (ENOTDIR);
953 nmp = VFSTONFS(dvp->v_mount);
954 np = VTONFS(dvp);
955
956 isdot = cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.';
957
958 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
959 *vpp = NULLVP;
960 return (error);
961 }
962 if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
963 struct vattr vattr;
964
965 newvp = *vpp;
966 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
967 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
968 nfsstats.lookupcache_hits++;
969 if (cnp->cn_nameiop != LOOKUP &&
970 (flags & ISLASTCN))
971 cnp->cn_flags |= SAVENAME;
972 return (0);
973 }
974 cache_purge(newvp);
975 if (newvp != dvp)
976 vput(newvp);
977 else
978 vrele(newvp);
979 }
980
981 error = 0;
982 newvp = NULLVP;
983 nfsstats.lookupcache_misses++;
984 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
985
986 len = cnp->cn_namelen;
987 mreq = nfsm_reqhead(NULL, NFSV4PROC_COMPOUND, 0);
988 mb = mreq;
989 bpos = mtod(mb, caddr_t);
990
991 ga.bm = &nfsv4_getattrbm;
992 dga.bm = &nfsv4_getattrbm;
993 nfs_v4initcompound(&cp);
994
995 nfsm_v4build_compound(&cp, "nfs4_lookup()");
996 nfsm_v4build_putfh(&cp, dvp);
997 nfsm_v4build_getattr(&cp, &dga);
998 if (flags & ISDOTDOT)
999 nfsm_v4build_lookupp(&cp);
1000 else if (!isdot) {
1001 l.name = cnp->cn_nameptr;
1002 l.namelen = len;
1003 nfsm_v4build_lookup(&cp, &l);
1004 }
1005 nfsm_v4build_getattr(&cp, &ga);
1006 nfsm_v4build_getfh(&cp, &gfh);
1007 nfsm_v4build_finalize(&cp);
1008
1009 nfsm_request(dvp, NFSV4PROC_COMPOUND, cnp->cn_thread, cnp->cn_cred);
1010 if (error != 0)
1011 goto nfsmout;
1012
1013 nfsm_v4dissect_compound(&cp);
1014 nfsm_v4dissect_putfh(&cp);
1015 nfsm_v4dissect_getattr(&cp, &dga);
1016 if (flags & ISDOTDOT)
1017 nfsm_v4dissect_lookupp(&cp);
1018 else if (!isdot)
1019 nfsm_v4dissect_lookup(&cp);
1020 nfsm_v4dissect_getattr(&cp, &ga);
1021 nfsm_v4dissect_getfh(&cp, &gfh);
1022
1023 nfs4_vnop_loadattrcache(dvp, &dga.fa, NULL);
1024 fhp = &gfh.fh_val;
1025 fhsize = gfh.fh_len;
1026
1027 /*
1028 * Handle RENAME case...
1029 */
1030 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1031 if (NFS_CMPFH(np, fhp, fhsize))
1032 return (EISDIR);
1033
1034 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, LK_EXCLUSIVE);
1035 if (error)
1036 return (error);
1037
1038 newvp = NFSTOV(np);
1039
1040 nfs4_vnop_loadattrcache(newvp, &ga.fa, NULL);
1041
1042 *vpp = newvp;
1043 cnp->cn_flags |= SAVENAME;
1044 return (0);
1045 }
1046
1047 if (flags & ISDOTDOT) {
1048 VOP_UNLOCK(dvp, 0, td);
1049
1050 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, LK_EXCLUSIVE);
1051 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
1052 if (error)
1053 return (error);
1054 newvp = NFSTOV(np);
1055
1056 nfs4_vnop_loadattrcache(newvp, &ga.fa, NULL);
1057 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1058 VREF(dvp);
1059 newvp = dvp;
1060 } else {
1061 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, LK_EXCLUSIVE);
1062 if (error)
1063 return (error);
1064 newvp = NFSTOV(np);
1065
1066 /* Fill in np used by open. */
1067 np->n_dvp = dvp;
1068 np->n_namelen = cnp->cn_namelen;
1069 if (np->n_name != NULL)
1070 FREE(np->n_name, M_NFSREQ);
1071 MALLOC(np->n_name, u_char *, np->n_namelen + 1, M_NFSREQ, M_WAITOK);
1072 bcopy(cnp->cn_nameptr, np->n_name, np->n_namelen);
1073 np->n_name[np->n_namelen] = '\0';
1074
1075 nfs4_vnop_loadattrcache(newvp, &ga.fa, NULL);
1076 }
1077
1078 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1079 cnp->cn_flags |= SAVENAME;
1080 if ((cnp->cn_flags & MAKEENTRY) &&
1081 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1082 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
1083 cache_enter(dvp, newvp, cnp);
1084 }
1085 *vpp = newvp;
1086 m_freem(mrep);
1087 nfsmout:
1088 error = nfs_v4postop(&cp, error);
1089
1090 if (error) {
1091 if (newvp != NULLVP) {
1092 vrele(newvp);
1093 *vpp = NULLVP;
1094 }
1095 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1096 (flags & ISLASTCN) && error == ENOENT) {
1097 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1098 error = EROFS;
1099 else
1100 error = EJUSTRETURN;
1101 }
1102 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1103 cnp->cn_flags |= SAVENAME;
1104 }
1105
1106 return (error);
1107 }
1108
1109 /*
1110 * nfs read call.
1111 * Just call nfs_bioread() to do the work.
1112 */
1113 static int
1114 nfs4_read(struct vop_read_args *ap)
1115 {
1116 struct vnode *vp = ap->a_vp;
1117
1118 switch (vp->v_type) {
1119 case VREG:
1120 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1121 case VDIR:
1122 return (EISDIR);
1123 default:
1124 return (EOPNOTSUPP);
1125 }
1126 }
1127
1128 /*
1129 * nfs readlink call
1130 */
1131 static int
1132 nfs4_readlink(struct vop_readlink_args *ap)
1133 {
1134 struct vnode *vp = ap->a_vp;
1135
1136 if (vp->v_type != VLNK)
1137 return (EINVAL);
1138 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1139 }
1140
1141 /*
1142 * Do a readlink rpc.
1143 * Called by nfs_doio() from below the buffer cache.
1144 */
1145 int
1146 nfs4_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1147 {
1148 caddr_t bpos, dpos;
1149 int error = 0;
1150 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1151 struct nfs4_compound cp;
1152
1153 nfsstats.rpccnt[NFSPROC_READLINK]++;
1154
1155 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
1156 mb = mreq;
1157 bpos = mtod(mb, caddr_t);
1158
1159 nfs_v4initcompound(&cp);
1160
1161 nfsm_v4build_compound(&cp, "nfs4_readlinkrpc()");
1162 nfsm_v4build_putfh(&cp, vp);
1163 nfsm_v4build_readlink(&cp);
1164 nfsm_v4build_finalize(&cp);
1165
1166 nfsm_request(vp, NFSV4PROC_COMPOUND, uiop->uio_td, cred);
1167 if (error != 0)
1168 goto nfsmout;
1169
1170 nfsm_v4dissect_compound(&cp);
1171 nfsm_v4dissect_putfh(&cp);
1172 nfsm_v4dissect_readlink(&cp, uiop);
1173
1174 nfsmout:
1175 error = nfs_v4postop(&cp, error);
1176
1177 if (mrep != NULL)
1178 m_freem(mrep);
1179 return (error);
1180 }
1181
1182 /*
1183 * nfs read rpc call
1184 * Ditto above
1185 */
1186 int
1187 nfs4_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1188 {
1189 caddr_t bpos, dpos;
1190 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1191 struct nfsmount *nmp;
1192 int error = 0, len, tsiz;
1193 struct nfs4_compound cp;
1194 struct nfs4_oparg_read read;
1195 struct nfsnode *np = VTONFS(vp);
1196
1197 nmp = VFSTONFS(vp->v_mount);
1198 tsiz = uiop->uio_resid;
1199 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1200 return (EFBIG);
1201
1202 if (tsiz == 0)
1203 return (0);
1204
1205 read.uiop = uiop;
1206 read.fcp = np->n_rfc.refcnt > 0 ? &np->n_rfc : &np->n_wfc;
1207
1208 while (tsiz > 0) {
1209 nfsstats.rpccnt[NFSPROC_READ]++;
1210 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1211
1212 read.off = uiop->uio_offset;
1213 read.maxcnt = len;
1214 nfs_v4initcompound(&cp);
1215
1216 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
1217 mb = mreq;
1218 bpos = mtod(mb, caddr_t);
1219
1220 nfsm_v4build_compound(&cp, "nfs4_readrpc()");
1221 nfsm_v4build_putfh(&cp, vp);
1222 nfsm_v4build_read(&cp, &read);
1223 nfsm_v4build_finalize(&cp);
1224
1225 nfsm_request(vp, NFSV4PROC_COMPOUND, uiop->uio_td, cred);
1226 if (error != 0) {
1227 error = nfs_v4postop(&cp, error);
1228 goto nfsmout;
1229 }
1230
1231 nfsm_v4dissect_compound(&cp);
1232 nfsm_v4dissect_putfh(&cp);
1233 nfsm_v4dissect_read(&cp, &read);
1234
1235 if (read.eof || read.retlen == 0)
1236 tsiz = 0;
1237 else
1238 tsiz -= read.retlen;
1239
1240 error = nfs_v4postop(&cp, error);
1241
1242 m_freem(mrep);
1243 mrep = NULL;
1244 }
1245 nfsmout:
1246 if (mrep != NULL)
1247 m_freem(mrep);
1248
1249 return (error);
1250 }
1251
1252 /*
1253 * nfs write call
1254 */
1255 int
1256 nfs4_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1257 int *iomode, int *must_commit)
1258 {
1259 int32_t backup;
1260 caddr_t bpos, dpos;
1261 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1262 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1263 int error = 0, len, tsiz, wccflag = 1, rlen;
1264 struct nfs4_compound cp;
1265 struct nfs4_oparg_write write;
1266 nfsv4stablehow commit, committed = NSHFILESYNC;
1267 caddr_t verf;
1268 struct nfsnode *np = VTONFS(vp);
1269
1270 #ifndef DIAGNOSTIC
1271 if (uiop->uio_iovcnt != 1)
1272 panic("nfs: writerpc iovcnt > 1");
1273 #endif
1274 *must_commit = 0;
1275 tsiz = uiop->uio_resid;
1276 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1277 return (EFBIG);
1278
1279 if (tsiz == 0)
1280 return (0);
1281
1282 write.stable = (nfsv4stablehow)*iomode;
1283 write.uiop = uiop;
1284 write.fcp = &np->n_wfc;
1285
1286 while (tsiz > 0) {
1287 nfsstats.rpccnt[NFSPROC_WRITE]++;
1288 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
1289
1290 write.off = uiop->uio_offset;
1291 write.cnt = len;
1292 nfs_v4initcompound(&cp);
1293
1294 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
1295 mb = mreq;
1296 bpos = mtod(mb, caddr_t);
1297
1298 nfsm_v4build_compound(&cp, "nfs4_writerpc()");
1299 nfsm_v4build_putfh(&cp, vp);
1300 nfsm_v4build_write(&cp, &write);
1301 nfsm_v4build_finalize(&cp);
1302
1303 nfsm_request(vp, NFSV4PROC_COMPOUND, uiop->uio_td, cred);
1304 if (error != 0) {
1305 error = nfs_v4postop(&cp, error);
1306 goto nfsmout;
1307 }
1308
1309 nfsm_v4dissect_compound(&cp);
1310 nfsm_v4dissect_putfh(&cp);
1311 nfsm_v4dissect_write(&cp, &write);
1312
1313 rlen = write.retlen;
1314 if (rlen == 0) {
1315 error = NFSERR_IO;
1316 break;
1317 } else if (rlen < len) {
1318 backup = len - rlen;
1319 uiop->uio_iov->iov_base =
1320 (char *)uiop->uio_iov->iov_base - backup;
1321 uiop->uio_iov->iov_len += backup;
1322 uiop->uio_offset -= backup;
1323 uiop->uio_resid += backup;
1324 len = rlen;
1325 }
1326
1327 commit = write.committed;
1328
1329 if (committed == NSHFILESYNC ||
1330 (committed = NSHDATASYNC && commit == NSHUNSTABLE))
1331 committed = commit;
1332
1333 verf = (caddr_t)write.wverf;
1334
1335 if ((nmp->nm_flag & NFSSTA_HASWRITEVERF) == 0) {
1336 bcopy(verf, nmp->nm_verf, NFSX_V4VERF);
1337 nmp->nm_flag |= NFSMNT_HASWRITEVERF;
1338 } else if (bcmp(verf, nmp->nm_verf, NFSX_V4VERF)) {
1339 *must_commit = 1;
1340 bcopy(verf, nmp->nm_verf, NFSX_V4VERF);
1341 }
1342
1343 /* XXX wccflag */
1344 if (wccflag)
1345 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1346
1347 error = nfs_v4postop(&cp, error);
1348
1349 m_freem(mrep);
1350 mrep = NULL;
1351 if (error)
1352 break;
1353 tsiz -= len;
1354 }
1355 nfsmout:
1356 if (mrep != NULL)
1357 m_freem(mrep);
1358 *iomode = committed;
1359 if (error)
1360 uiop->uio_resid = tsiz;
1361 return (error);
1362 }
1363
1364 /* ARGSUSED */
1365 static int
1366 nfs4_mknod(struct vop_mknod_args *ap)
1367 {
1368 struct vattr *vap = ap->a_vap;
1369 struct vnode *newvp = NULL;
1370 int error;
1371
1372 error = nfs4_createrpc(ap->a_dvp, &newvp,
1373 ap->a_cnp, (nfstype)vap->va_type, vap, NULL);
1374
1375 /* XXX - is this actually referenced here? */
1376 if (error == 0) {
1377 *ap->a_vpp = newvp;
1378 vrele(newvp);
1379 }
1380
1381 return (error);
1382 }
1383
1384 static int
1385 nfs4_createrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1386 nfstype ftype, struct vattr *vap, char *linktarget)
1387 {
1388 struct nfsnode *dnp = VTONFS(dvp);
1389 struct nfsnode *np = NULL;
1390 struct vnode *newvp = NULL;
1391 struct nfs4_compound cp;
1392 struct nfs4_oparg_create c;
1393 struct nfs4_oparg_getattr ga;
1394 struct nfs4_oparg_getfh gfh;
1395 caddr_t bpos, dpos;
1396 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1397 int error = 0;
1398
1399 nfsstats.rpccnt[NFSPROC_CREATE]++;
1400
1401 mreq = nfsm_reqhead(dvp, NFSV4PROC_COMPOUND, 0);
1402 mb = mreq;
1403 bpos = mtod(mb, caddr_t);
1404
1405 bzero(&c, sizeof(c));
1406 bzero(&ga, sizeof(ga));
1407
1408 c.type = ftype;
1409 c.vap = vap;
1410 c.linktext = linktarget;
1411 c.name = cnp->cn_nameptr;
1412 c.namelen = cnp->cn_namelen;
1413
1414 ga.bm = &nfsv4_getattrbm;
1415 nfs_v4initcompound(&cp);
1416
1417 nfsm_v4build_compound(&cp, "nfs4_createrpc()");
1418 nfsm_v4build_putfh(&cp, dvp);
1419 nfsm_v4build_create(&cp, &c);
1420 nfsm_v4build_getattr(&cp, &ga);
1421 nfsm_v4build_getfh(&cp, &gfh);
1422 nfsm_v4build_finalize(&cp);
1423
1424 nfsm_request(dvp, NFSV4PROC_COMPOUND, cnp->cn_thread, cnp->cn_cred);
1425 if (error != 0)
1426 goto nfsmout;
1427
1428 nfsm_v4dissect_compound(&cp);
1429 nfsm_v4dissect_putfh(&cp);
1430 nfsm_v4dissect_create(&cp, &c);
1431 nfsm_v4dissect_getattr(&cp, &ga);
1432 nfsm_v4dissect_getfh(&cp, &gfh);
1433
1434 error = nfs_nget(dvp->v_mount, &gfh.fh_val, gfh.fh_len, &np, LK_EXCLUSIVE);
1435 if (error != 0)
1436 goto nfsmout;
1437
1438 newvp = NFSTOV(np);
1439 nfs4_vnop_loadattrcache(newvp, &ga.fa, NULL);
1440
1441 if (cnp->cn_flags & MAKEENTRY)
1442 cache_enter(dvp, newvp, cnp);
1443
1444 dnp->n_flag |= NMODIFIED;
1445 dnp->n_attrstamp = 0;
1446
1447 nfsmout:
1448 error = nfs_v4postop(&cp, error);
1449
1450 if (mrep != NULL)
1451 m_freem(mrep);
1452
1453 /* XXX */
1454 /*FREE(cnp->cn_pnbuf, M_NAMEI);*/
1455 if (error != 0 && newvp != NULL)
1456 vput(newvp);
1457 else if (error == 0)
1458 *vpp = newvp;
1459
1460 return (error);
1461 }
1462
1463 static int
1464 nfs4_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1465 struct vnode *tdvp, const char *tnameptr, int tnamelen,
1466 struct ucred *cred, struct thread *td)
1467 {
1468
1469 struct nfsnode *fnp = VTONFS(fdvp), *tnp = VTONFS(tdvp);
1470 caddr_t bpos, dpos;
1471 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1472 struct nfs4_compound cp;
1473 struct nfs4_oparg_rename r;
1474 int error = 0;
1475
1476 nfsstats.rpccnt[NFSPROC_RENAME]++;
1477
1478 r.fname = fnameptr;
1479 r.fnamelen = fnamelen;
1480 r.tname = tnameptr;
1481 r.tnamelen = tnamelen;
1482 nfs_v4initcompound(&cp);
1483
1484 mreq = nfsm_reqhead(fdvp, NFSV4PROC_COMPOUND, 0);
1485 mb = mreq;
1486 bpos = mtod(mb, caddr_t);
1487
1488 nfsm_v4build_compound(&cp, "nfs4_renamerpc()");
1489 nfsm_v4build_putfh(&cp, fdvp);
1490 nfsm_v4build_savefh(&cp);
1491 nfsm_v4build_putfh(&cp, tdvp);
1492 nfsm_v4build_rename(&cp, &r);
1493 nfsm_v4build_finalize(&cp);
1494
1495 nfsm_request(fdvp, NFSV4PROC_COMPOUND, td, cred);
1496 if (error != 0)
1497 goto nfsmout;
1498
1499 nfsm_v4dissect_compound(&cp);
1500 nfsm_v4dissect_putfh(&cp);
1501 nfsm_v4dissect_savefh(&cp);
1502 nfsm_v4dissect_putfh(&cp);
1503 nfsm_v4dissect_rename(&cp);
1504
1505 /* XXX should this always be performed? */
1506 fnp->n_flag |= NMODIFIED;
1507 tnp->n_flag |= NMODIFIED;
1508 fnp->n_attrstamp = tnp->n_attrstamp = 0;
1509
1510 nfsmout:
1511 error = nfs_v4postop(&cp, error);
1512
1513 if (mrep != NULL)
1514 m_freem(mrep);
1515
1516 return (error);
1517 }
1518
1519 /*
1520 * nfs file create call
1521 */
1522 static int
1523 nfs4_create(struct vop_create_args *ap)
1524 {
1525 struct vnode *dvp = ap->a_dvp;
1526 struct vattr *vap = ap->a_vap;
1527 struct nfsnode *dnp = VTONFS(dvp);
1528 struct componentname *cnp = ap->a_cnp;
1529 struct vnode *newvp = NULL;
1530 int error = 0, fmode = (O_CREAT | FREAD | FWRITE);
1531 struct vattr vattr;
1532
1533 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0)
1534 return (error);
1535
1536 if (vap->va_vaflags & VA_EXCLUSIVE)
1537 fmode |= O_EXCL;
1538
1539 error = nfs4_openrpc(dvp, &newvp, cnp, fmode, vap);
1540 if (error != 0)
1541 goto out;
1542
1543 VTONFS(newvp)->n_flag |= NCREATED;
1544
1545 if (cnp->cn_flags & MAKEENTRY)
1546 cache_enter(dvp, newvp, cnp);
1547
1548 *ap->a_vpp = newvp;
1549
1550 dnp->n_flag |= NMODIFIED;
1551 dnp->n_attrstamp = 0; /* XXX; wccflag */
1552
1553 out:
1554 return (error);
1555 }
1556
1557 /*
1558 * nfs file remove call
1559 * To try and make nfs semantics closer to ufs semantics, a file that has
1560 * other processes using the vnode is renamed instead of removed and then
1561 * removed later on the last close.
1562 * - If v_usecount > 1
1563 * If a rename is not already in the works
1564 * call nfs4_sillyrename() to set it up
1565 * else
1566 * do the remove rpc
1567 */
1568 static int
1569 nfs4_remove(struct vop_remove_args *ap)
1570 {
1571 struct vnode *vp = ap->a_vp;
1572 struct vnode *dvp = ap->a_dvp;
1573 struct componentname *cnp = ap->a_cnp;
1574 struct nfsnode *np = VTONFS(vp);
1575 int error = 0;
1576 struct vattr vattr;
1577
1578 #ifndef DIAGNOSTIC
1579 if ((cnp->cn_flags & HASBUF) == 0)
1580 panic("nfs4_remove: no name");
1581 if (vrefcnt(vp) < 1)
1582 panic("nfs4_remove: bad v_usecount");
1583 #endif
1584 if (vp->v_type == VDIR)
1585 error = EPERM;
1586 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1587 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
1588 vattr.va_nlink > 1)) {
1589 /*
1590 * Purge the name cache so that the chance of a lookup for
1591 * the name succeeding while the remove is in progress is
1592 * minimized. Without node locking it can still happen, such
1593 * that an I/O op returns ESTALE, but since you get this if
1594 * another host removes the file..
1595 */
1596 cache_purge(vp);
1597 /*
1598 * throw away biocache buffers, mainly to avoid
1599 * unnecessary delayed writes later.
1600 */
1601 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1602 /* Do the rpc */
1603 if (error != EINTR)
1604 error = nfs4_removerpc(dvp, cnp->cn_nameptr,
1605 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1606 /*
1607 * Kludge City: If the first reply to the remove rpc is lost..
1608 * the reply to the retransmitted request will be ENOENT
1609 * since the file was in fact removed
1610 * Therefore, we cheat and return success.
1611 */
1612 if (error == ENOENT)
1613 error = 0;
1614 } else if (!np->n_sillyrename)
1615 error = nfs4_sillyrename(dvp, vp, cnp);
1616 np->n_attrstamp = 0;
1617 return (error);
1618 }
1619
1620 /*
1621 * nfs file remove rpc called from nfs_inactive
1622 */
1623 int
1624 nfs4_removeit(struct sillyrename *sp)
1625 {
1626 /*
1627 * Make sure that the directory vnode is still valid.
1628 * XXX we should lock sp->s_dvp here.
1629 */
1630 if (sp->s_dvp->v_type == VBAD)
1631 return (0);
1632 return (nfs4_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1633 NULL));
1634 }
1635
1636 /*
1637 * Nfs remove rpc, called from nfs4_remove() and nfs4_removeit().
1638 */
1639 static int
1640 nfs4_removerpc(struct vnode *dvp, const char *name, int namelen,
1641 struct ucred *cred, struct thread *td)
1642 {
1643 caddr_t bpos, dpos;
1644 int error = 0;
1645 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1646 struct nfs4_compound cp;
1647
1648 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1649
1650 mreq = nfsm_reqhead(dvp, NFSV4PROC_COMPOUND, 0);
1651 mb = mreq;
1652 bpos = mtod(mb, caddr_t);
1653
1654 nfs_v4initcompound(&cp);
1655
1656 nfsm_v4build_compound(&cp, "nfs4_removerpc()");
1657 nfsm_v4build_putfh(&cp, dvp);
1658 nfsm_v4build_remove(&cp, name, namelen);
1659 nfsm_v4build_finalize(&cp);
1660
1661 nfsm_request(dvp, NFSV4PROC_COMPOUND, td, cred);
1662 if (error != 0)
1663 goto nfsmout;
1664
1665 nfsm_v4dissect_compound(&cp);
1666 nfsm_v4dissect_putfh(&cp);
1667 nfsm_v4dissect_remove(&cp);
1668
1669 nfsmout:
1670 error = nfs_v4postop(&cp, error);
1671
1672 if (mrep != NULL)
1673 m_freem(mrep);
1674
1675 VTONFS(dvp)->n_flag |= NMODIFIED;
1676 VTONFS(dvp)->n_attrstamp = 0; /* XXX wccflag */
1677
1678 return (error);
1679 }
1680
1681 /*
1682 * nfs file rename call
1683 */
1684 static int
1685 nfs4_rename(struct vop_rename_args *ap)
1686 {
1687 struct vnode *fvp = ap->a_fvp;
1688 struct vnode *tvp = ap->a_tvp;
1689 struct vnode *fdvp = ap->a_fdvp;
1690 struct vnode *tdvp = ap->a_tdvp;
1691 struct componentname *tcnp = ap->a_tcnp;
1692 struct componentname *fcnp = ap->a_fcnp;
1693 int error;
1694
1695 #ifndef DIAGNOSTIC
1696 if ((tcnp->cn_flags & HASBUF) == 0 ||
1697 (fcnp->cn_flags & HASBUF) == 0)
1698 panic("nfs4_rename: no name");
1699 #endif
1700 /* Check for cross-device rename */
1701 if ((fvp->v_mount != tdvp->v_mount) ||
1702 (tvp && (fvp->v_mount != tvp->v_mount))) {
1703 error = EXDEV;
1704 goto out;
1705 }
1706
1707 if (fvp == tvp) {
1708 printf("nfs4_rename: fvp == tvp (can't happen)\n");
1709 error = 0;
1710 goto out;
1711 }
1712 if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0)
1713 goto out;
1714
1715 /*
1716 * We have to flush B_DELWRI data prior to renaming
1717 * the file. If we don't, the delayed-write buffers
1718 * can be flushed out later after the file has gone stale
1719 * under NFSV3. NFSV2 does not have this problem because
1720 * ( as far as I can tell ) it flushes dirty buffers more
1721 * often.
1722 */
1723 VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1724 VOP_UNLOCK(fvp, 0, fcnp->cn_thread);
1725 if (tvp)
1726 VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1727
1728 /*
1729 * If the tvp exists and is in use, sillyrename it before doing the
1730 * rename of the new file over it.
1731 * XXX Can't sillyrename a directory.
1732 */
1733 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1734 tvp->v_type != VDIR && !nfs4_sillyrename(tdvp, tvp, tcnp)) {
1735 vput(tvp);
1736 tvp = NULL;
1737 }
1738
1739 error = nfs4_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1740 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1741 tcnp->cn_thread);
1742
1743 if (fvp->v_type == VDIR) {
1744 if (tvp != NULL && tvp->v_type == VDIR)
1745 cache_purge(tdvp);
1746 cache_purge(fdvp);
1747 }
1748
1749 out:
1750 if (tdvp == tvp)
1751 vrele(tdvp);
1752 else
1753 vput(tdvp);
1754 if (tvp)
1755 vput(tvp);
1756 vrele(fdvp);
1757 vrele(fvp);
1758 /*
1759 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1760 */
1761 if (error == ENOENT)
1762 error = 0;
1763 return (error);
1764 }
1765
1766 /*
1767 * nfs file rename rpc called from nfs4_remove() above
1768 */
1769 static int
1770 nfs4_renameit(struct vnode *sdvp, struct componentname *scnp,
1771 struct sillyrename *sp)
1772 {
1773 return (nfs4_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1774 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1775 }
1776
1777 /*
1778 * nfs hard link create call
1779 */
1780 static int
1781 nfs4_link(struct vop_link_args *ap)
1782 {
1783 struct vnode *vp = ap->a_vp;
1784 struct vnode *tdvp = ap->a_tdvp;
1785 struct componentname *cnp = ap->a_cnp;
1786 caddr_t bpos, dpos;
1787 int error = 0;
1788 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1789 struct nfs4_compound cp;
1790 struct nfs4_oparg_link l;
1791
1792 if (vp->v_mount != tdvp->v_mount) {
1793 return (EXDEV);
1794 }
1795
1796 /*
1797 * Push all writes to the server, so that the attribute cache
1798 * doesn't get "out of sync" with the server.
1799 * XXX There should be a better way!
1800 */
1801 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1802
1803 nfsstats.rpccnt[NFSPROC_LINK]++;
1804
1805 l.name = cnp->cn_nameptr;
1806 l.namelen = cnp->cn_namelen;
1807 nfs_v4initcompound(&cp);
1808
1809 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
1810 mb = mreq;
1811 bpos = mtod(mb, caddr_t);
1812
1813 nfsm_v4build_compound(&cp, "nfs4_link()");
1814 nfsm_v4build_putfh(&cp, vp);
1815 nfsm_v4build_savefh(&cp);
1816 nfsm_v4build_putfh(&cp, tdvp);
1817 nfsm_v4build_link(&cp, &l);
1818 nfsm_v4build_finalize(&cp);
1819
1820 nfsm_request(vp, NFSV4PROC_COMPOUND, cnp->cn_thread, cnp->cn_cred);
1821 if (error != 0)
1822 goto nfsmout;
1823
1824 nfsm_v4dissect_compound(&cp);
1825 nfsm_v4dissect_putfh(&cp);
1826 nfsm_v4dissect_savefh(&cp);
1827 nfsm_v4dissect_putfh(&cp);
1828 nfsm_v4dissect_link(&cp);
1829
1830 VTONFS(tdvp)->n_flag |= NMODIFIED;
1831 VTONFS(vp)->n_attrstamp = 0;
1832 VTONFS(tdvp)->n_attrstamp = 0;
1833
1834 nfsmout:
1835 error = nfs_v4postop(&cp, error);
1836
1837 if (mrep != NULL)
1838 m_freem(mrep);
1839
1840 return (error);
1841 }
1842
1843 /*
1844 * nfs symbolic link create call
1845 */
1846 static int
1847 nfs4_symlink(struct vop_symlink_args *ap)
1848 {
1849 struct vnode *dvp = ap->a_dvp;
1850 int error = 0;
1851 struct vnode *newvp = NULL;
1852
1853 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1854
1855 error = nfs4_createrpc(ap->a_dvp, &newvp, ap->a_cnp, NFLNK,
1856 ap->a_vap, ap->a_target);
1857
1858 if (error != 0 && newvp != NULL)
1859 vput(newvp);
1860 else if (error == 0)
1861 *ap->a_vpp = newvp;
1862
1863 VTONFS(dvp)->n_flag |= NMODIFIED;
1864 VTONFS(dvp)->n_attrstamp = 0; /* XXX wccflags */
1865
1866 return (error);
1867 }
1868
1869 /*
1870 * nfs make dir call
1871 */
1872 static int
1873 nfs4_mkdir(struct vop_mkdir_args *ap)
1874 {
1875 return (nfs4_createrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, NFDIR,
1876 ap->a_vap, NULL));
1877 }
1878
1879 /*
1880 * nfs remove directory call
1881 */
1882 static int
1883 nfs4_rmdir(struct vop_rmdir_args *ap)
1884 {
1885 struct vnode *vp = ap->a_vp;
1886 struct vnode *dvp = ap->a_dvp;
1887 struct nfsnode *dnp = VTONFS(dvp);
1888 struct componentname *cnp = ap->a_cnp;
1889 int error = 0;
1890
1891 if (dvp == vp)
1892 return (EINVAL);
1893
1894 error = (nfs4_removerpc(dvp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_cred,
1895 NULL));
1896 if (error)
1897 return (error);
1898
1899 dnp->n_flag |= NMODIFIED;
1900 dnp->n_attrstamp = 0;
1901 cache_purge(dvp);
1902 cache_purge(vp);
1903
1904 return (error);
1905 }
1906
1907 /*
1908 * nfs readdir call
1909 */
1910 static int
1911 nfs4_readdir(struct vop_readdir_args *ap)
1912 {
1913 struct vnode *vp = ap->a_vp;
1914 struct nfsnode *np = VTONFS(vp);
1915 struct uio *uio = ap->a_uio;
1916 int tresid, error;
1917 struct vattr vattr;
1918
1919 if (vp->v_type != VDIR)
1920 return (EPERM);
1921 /*
1922 * First, check for hit on the EOF offset cache
1923 */
1924 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
1925 (np->n_flag & NMODIFIED) == 0) {
1926 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 &&
1927 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
1928 nfsstats.direofcache_hits++;
1929 return (0);
1930 }
1931 }
1932
1933 /*
1934 * Call nfs_bioread() to do the real work.
1935 */
1936 tresid = uio->uio_resid;
1937 error = nfs_bioread(vp, uio, 0, ap->a_cred);
1938
1939 if (!error && uio->uio_resid == tresid)
1940 nfsstats.direofcache_misses++;
1941 return (error);
1942 }
1943
1944 static u_char fty_to_dty[] = {
1945 DT_UNKNOWN, /* NFNON */
1946 DT_REG, /* NFREG */
1947 DT_DIR, /* NFDIR */
1948 DT_BLK, /* NFBLK */
1949 DT_CHR, /* NFCHR */
1950 DT_LNK, /* NFLNK */
1951 DT_SOCK, /* NFSOCK */
1952 DT_FIFO, /* NFFIFO */
1953 DT_UNKNOWN, /* NFATTRDIT */
1954 DT_UNKNOWN, /* NFNAMEDATTR */
1955 DT_UNKNOWN, /* NFBAD */
1956 };
1957
1958 /*
1959 * Readdir rpc call.
1960 * Called from below the buffer cache by nfs_doio().
1961 */
1962 int
1963 nfs4_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1964 {
1965 int len, left;
1966 struct dirent *dp = NULL;
1967 u_int32_t *tl;
1968 caddr_t p;
1969 uint64_t *cookiep;
1970 caddr_t bpos, dpos;
1971 struct mbuf *mreq, *mrep = NULL, *md, *mb;
1972 uint64_t cookie;
1973 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1974 struct nfsnode *dnp = VTONFS(vp);
1975 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
1976 struct nfs4_compound cp;
1977 struct nfs4_oparg_readdir readdir;
1978 struct nfsv4_fattr fattr;
1979 u_int fty;
1980
1981 #ifndef DIAGNOSTIC
1982 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
1983 (uiop->uio_resid & (DIRBLKSIZ - 1)))
1984 panic("nfs readdirrpc bad uio");
1985 #endif
1986
1987 /*
1988 * If there is no cookie, assume directory was stale.
1989 */
1990 cookiep = nfs4_getcookie(dnp, uiop->uio_offset, 0);
1991 if (cookiep)
1992 cookie = *cookiep;
1993 else
1994 return (NFSERR_BAD_COOKIE);
1995
1996 /* Generate fake entries for "." and ".." */
1997 while (cookie < 2 && bigenough) {
1998 cookie++;
1999 len = 4 + DIRHDSIZ;
2000
2001 if (len > uiop->uio_resid) {
2002 bigenough = 0;
2003 break;
2004 }
2005 dp = (struct dirent *)uiop->uio_iov->iov_base;
2006
2007 dp->d_namlen = cookie;
2008 dp->d_reclen = len;
2009 dp->d_type = DT_DIR;
2010 if (cookie == 1)
2011 dp->d_fileno = dnp->n_vattr.va_fileid; /* XXX has problems with pynfs virtualhandles */
2012 else
2013 dp->d_fileno = dnp->n_dvp != NULL ?
2014 VTONFS(dnp->n_dvp)->n_vattr.va_fileid : cookie;
2015
2016 p = dp->d_name;
2017 *p++ = '.';
2018 if (cookie == 2)
2019 *p++ = '.';
2020 *p = '\0';
2021
2022 blksiz += len;
2023 if (blksiz == DIRBLKSIZ)
2024 blksiz = 0;
2025 uiop->uio_offset += len;
2026 uiop->uio_resid -= len;
2027 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + len;
2028 uiop->uio_iov->iov_len -= len;
2029 }
2030
2031 if (cookie == 2)
2032 cookie = 0;
2033
2034 /* This is sort of ugly, to prevent v4postop() from acting weird */
2035 bzero(&cp, sizeof(cp));
2036
2037 /*
2038 * Loop around doing readdir rpc's of size nm_readdirsize
2039 * truncated to a multiple of DIRBLKSIZ.
2040 * The stopping criteria is EOF or buffer full.
2041 */
2042 /*
2043 * XXX this is sort of ugly for nfsv4; we don't maintain the
2044 * strict abstraction, but do the decoding inline. that's ok.
2045 */
2046 while (more_dirs && bigenough) {
2047 nfsstats.rpccnt[NFSPROC_READDIR]++;
2048
2049 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
2050 mb = mreq;
2051 bpos = mtod(mb, caddr_t);
2052
2053 readdir.cnt = nmp->nm_readdirsize;
2054 readdir.cookie = cookie;
2055 readdir.bm = &nfsv4_readdirbm;
2056 if (cookie == 0)
2057 bzero(&readdir.verf, sizeof(readdir.verf));
2058 else
2059 bcopy(&dnp->n_cookieverf, &readdir.verf,
2060 sizeof(readdir.verf));
2061
2062 nfs_v4initcompound(&cp);
2063
2064 nfsm_v4build_compound(&cp, "nfs4_readdirrpc()");
2065 nfsm_v4build_putfh(&cp, vp);
2066 nfsm_v4build_readdir(&cp, &readdir);
2067 nfsm_v4build_finalize(&cp);
2068
2069 nfsm_request(vp, NFSV4PROC_COMPOUND, uiop->uio_td, cred);
2070 if (error != 0)
2071 goto nfsmout;
2072
2073 nfsm_v4dissect_compound(&cp);
2074 nfsm_v4dissect_putfh(&cp);
2075
2076 /*
2077 * XXX - Readdir gets handled inline like in
2078 * NFSv{2,3}. This is a nasty inconsistency and
2079 * should be fixed.
2080 */
2081
2082 tl = nfsm_dissect(uint32_t *, 5 * NFSX_UNSIGNED);
2083 if (fxdr_unsigned(uint32_t, *tl++) != NFSV4OP_READDIR) {
2084 error = EBADRPC;
2085 goto nfsmout;
2086 }
2087 if (fxdr_unsigned(uint32_t, *tl++) != 0) {
2088 error = EBADRPC;
2089 goto nfsmout;
2090 }
2091
2092 bcopy(tl, &dnp->n_cookieverf, NFSX_V4VERF);
2093 tl += 2;
2094 more_dirs = fxdr_unsigned(int, *tl++);
2095
2096 /* loop thru the dir entries, doctoring them to 4bsd form */
2097 while (more_dirs && bigenough) {
2098 tl = nfsm_dissect(uint32_t *, 3 * NFSX_UNSIGNED);
2099 cookie = fxdr_hyper(tl);
2100 tl += 2;
2101 /* XXX cookie sanity check */
2102 len = fxdr_unsigned(int, *tl++);
2103 if (len <= 0 || len > NFS_MAXNAMLEN) {
2104 error = EBADRPC;
2105 goto nfsmout;
2106 }
2107 tlen = nfsm_rndup(len);
2108 if (tlen == len)
2109 tlen += 4; /* To ensure null termination */
2110 left = DIRBLKSIZ - blksiz;
2111 if ((tlen + DIRHDSIZ) > left) {
2112 dp->d_reclen += left;
2113 uiop->uio_iov->iov_base =
2114 (char *)uiop->uio_iov->iov_base + left;
2115 uiop->uio_iov->iov_len -= left;
2116 uiop->uio_offset += left;
2117 uiop->uio_resid -= left;
2118 blksiz = 0;
2119 }
2120 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2121 bigenough = 0;
2122 if (bigenough) {
2123 dp = (struct dirent *)uiop->uio_iov->iov_base;
2124
2125 dp->d_namlen = len;
2126 dp->d_reclen = tlen + DIRHDSIZ;
2127
2128 blksiz += dp->d_reclen;
2129 if (blksiz == DIRBLKSIZ)
2130 blksiz = 0;
2131 uiop->uio_offset += DIRHDSIZ;
2132 uiop->uio_resid -= DIRHDSIZ;
2133 uiop->uio_iov->iov_base =
2134 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2135 uiop->uio_iov->iov_len -= DIRHDSIZ;
2136
2137 /* Copy name */
2138 nfsm_mtouio(uiop, len);
2139 p = uiop->uio_iov->iov_base;
2140 tlen -= len;
2141 *p = '\0'; /* null terminate */
2142 /* printf("nfs4_readdirrpc: name: \"%s\" cookie %d\n",
2143 p - len, (int) cookie);*/
2144 uiop->uio_iov->iov_base =
2145 (char *)uiop->uio_iov->iov_base + tlen;
2146 uiop->uio_iov->iov_len -= tlen;
2147 uiop->uio_offset += tlen;
2148 uiop->uio_resid -= tlen;
2149
2150 /* Copy attributes */
2151 nfsm_v4dissect_attrs(&fattr);
2152
2153 dp->d_fileno = nfs_v4fileid4_to_fileid(
2154 fattr.fa4_valid & FA4V_FILEID &&
2155 fattr.fa4_fileid ?
2156 fattr.fa4_fileid : cookie);
2157
2158 fty = (u_int)fattr.fa4_type;
2159 dp->d_type = fattr.fa4_valid & FA4V_TYPE &&
2160 (fty < sizeof(fty_to_dty)) ?
2161 fty_to_dty[fty] : DT_UNKNOWN;
2162 } else
2163 nfsm_adv(nfsm_rndup(len));
2164
2165 tl = nfsm_dissect(uint32_t *, NFSX_UNSIGNED);
2166 more_dirs = fxdr_unsigned(int, *tl++);
2167 }
2168 /*
2169 * If at end of rpc data, get the eof boolean
2170 */
2171 if (!more_dirs) {
2172 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2173 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2174 }
2175
2176 error = nfs_v4postop(&cp, error);
2177
2178 m_freem(mrep);
2179 mrep = NULL;
2180 }
2181 /*
2182 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2183 * by increasing d_reclen for the last record.
2184 */
2185 if (blksiz > 0) {
2186 left = DIRBLKSIZ - blksiz;
2187 dp->d_reclen += left;
2188 uiop->uio_iov->iov_base =
2189 (char *)uiop->uio_iov->iov_base + left;
2190 uiop->uio_iov->iov_len -= left;
2191 uiop->uio_offset += left;
2192 uiop->uio_resid -= left;
2193 }
2194
2195 /*
2196 * We are now either at the end of the directory or have filled the
2197 * block.
2198 */
2199 if (bigenough)
2200 dnp->n_direofoffset = uiop->uio_offset;
2201 else {
2202 if (uiop->uio_resid > 0)
2203 printf("EEK! readdirrpc resid > 0\n");
2204 cookiep = nfs4_getcookie(dnp, uiop->uio_offset, 1);
2205 *cookiep = cookie;
2206 }
2207 nfsmout:
2208 if (mrep != NULL)
2209 m_freem(mrep);
2210 return (error);
2211 }
2212
2213 /*
2214 * Silly rename. To make the NFS filesystem that is stateless look a little
2215 * more like the "ufs" a remove of an active vnode is translated to a rename
2216 * to a funny looking filename that is removed by nfs_inactive on the
2217 * nfsnode. There is the potential for another process on a different client
2218 * to create the same funny name between the nfs_lookitup() fails and the
2219 * nfs_rename() completes, but...
2220 */
2221 static int
2222 nfs4_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2223 {
2224 struct sillyrename *sp;
2225 struct nfsnode *np;
2226 int error;
2227 short pid;
2228
2229 cache_purge(dvp);
2230 np = VTONFS(vp);
2231 #ifndef DIAGNOSTIC
2232 if (vp->v_type == VDIR)
2233 panic("nfs: sillyrename dir");
2234 #endif
2235 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2236 M_NFSREQ, M_WAITOK);
2237 sp->s_cred = crhold(cnp->cn_cred);
2238 sp->s_dvp = dvp;
2239 sp->s_removeit = nfs4_removeit;
2240 VREF(dvp);
2241
2242 /* Fudge together a funny name */
2243 pid = cnp->cn_thread->td_proc->p_pid;
2244 sp->s_namlen = sprintf(sp->s_name, ".nfsA%04x4.4", pid);
2245
2246 /* Try lookitups until we get one that isn't there */
2247 while (nfs4_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2248 cnp->cn_thread, NULL) == 0) {
2249 sp->s_name[4]++;
2250 if (sp->s_name[4] > 'z') {
2251 error = EINVAL;
2252 goto bad;
2253 }
2254 }
2255 error = nfs4_renameit(dvp, cnp, sp);
2256 if (error)
2257 goto bad;
2258 error = nfs4_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2259 cnp->cn_thread, &np);
2260 np->n_sillyrename = sp;
2261 return (0);
2262 bad:
2263 vrele(sp->s_dvp);
2264 crfree(sp->s_cred);
2265 free((caddr_t)sp, M_NFSREQ);
2266 return (error);
2267 }
2268
2269 /*
2270 * Look up a file name and optionally either update the file handle or
2271 * allocate an nfsnode, depending on the value of npp.
2272 * npp == NULL --> just do the lookup
2273 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2274 * handled too
2275 * *npp != NULL --> update the file handle in the vnode
2276 */
2277 static int
2278 nfs4_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2279 struct thread *td, struct nfsnode **npp)
2280 {
2281 struct vnode *newvp = NULL;
2282 struct nfsnode *np, *dnp = VTONFS(dvp);
2283 caddr_t bpos, dpos;
2284 int error = 0, fhlen;
2285 struct mbuf *mreq, *mrep = NULL, *md, *mb;
2286 nfsfh_t *nfhp;
2287 struct nfs4_compound cp;
2288 struct nfs4_oparg_lookup l;
2289 struct nfs4_oparg_getfh gfh;
2290 struct nfs4_oparg_getattr ga;
2291
2292 nfsstats.rpccnt[NFSPROC_RENAME]++;
2293
2294 mreq = nfsm_reqhead(dvp, NFSV4PROC_COMPOUND, 0);
2295 mb = mreq;
2296 bpos = mtod(mb, caddr_t);
2297
2298 l.name = name;
2299 l.namelen = len;
2300
2301 nfs_v4initcompound(&cp);
2302
2303 ga.bm = &nfsv4_getattrbm;
2304
2305 nfsm_v4build_compound(&cp, "nfs4_renamerpc()");
2306 nfsm_v4build_putfh(&cp, dvp);
2307 nfsm_v4build_lookup(&cp, &l);
2308 nfsm_v4build_getfh(&cp, &gfh);
2309 nfsm_v4build_getattr(&cp, &ga);
2310
2311 nfsm_request(dvp, NFSV4PROC_COMPOUND, td, cred);
2312 if (error != 0)
2313 goto nfsmout;
2314
2315 nfsm_v4dissect_compound(&cp);
2316 nfsm_v4dissect_putfh(&cp);
2317 nfsm_v4dissect_lookup(&cp);
2318 nfsm_v4dissect_getfh(&cp, &gfh);
2319 nfsm_v4dissect_getattr(&cp, &ga);
2320
2321 if (npp != NULL && error == 0) {
2322 nfhp = &gfh.fh_val;
2323 fhlen = gfh.fh_len;
2324
2325 if (*npp != NULL) {
2326 np = *npp;
2327 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2328 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2329 np->n_fhp = &np->n_fh;
2330 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2331 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2332 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2333 np->n_fhsize = fhlen;
2334 newvp = NFSTOV(np);
2335 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2336 VREF(dvp);
2337 newvp = dvp;
2338 } else {
2339 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2340 if (error) {
2341 m_freem(mrep);
2342 return (error);
2343 }
2344 newvp = NFSTOV(np);
2345 }
2346
2347 if (newvp != dvp) {
2348 np->n_dvp = dvp;
2349 np->n_namelen = len;
2350 if (np->n_name != NULL)
2351 FREE(np->n_name, M_NFSREQ);
2352 MALLOC(np->n_name, u_char *,
2353 np->n_namelen + 1, M_NFSREQ, M_WAITOK);
2354 memcpy(np->n_name, name, len);
2355 np->n_name[len] = '\0';
2356 }
2357 nfs4_vnop_loadattrcache(newvp, &ga.fa, NULL);
2358 }
2359
2360 nfsmout:
2361 error = nfs_v4postop(&cp, error);
2362
2363 if (mrep != NULL)
2364 m_freem(mrep);
2365 if (npp && *npp == NULL) {
2366 if (error) {
2367 if (newvp) {
2368 if (newvp == dvp)
2369 vrele(newvp);
2370 else
2371 vput(newvp);
2372 }
2373 } else
2374 *npp = np;
2375 }
2376
2377
2378 return (error);
2379 }
2380
2381 /*
2382 * Nfs Version 3 commit rpc
2383 */
2384 int
2385 nfs4_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2386 struct thread *td)
2387 {
2388 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2389 caddr_t bpos, dpos;
2390 int error = 0;
2391 struct mbuf *mreq, *mrep = NULL, *md, *mb;
2392 struct nfs4_compound cp;
2393 struct nfs4_oparg_commit commit;
2394
2395 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
2396 return (0);
2397 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2398
2399 mreq = nfsm_reqhead(vp, NFSV4PROC_COMPOUND, 0);
2400 mb = mreq;
2401 bpos = mtod(mb, caddr_t);
2402
2403 commit.start = offset;
2404 commit.len = cnt;
2405
2406 nfs_v4initcompound(&cp);
2407
2408 nfsm_v4build_compound(&cp, "nfs4_commit()");
2409 nfsm_v4build_putfh(&cp, vp);
2410 nfsm_v4build_commit(&cp, &commit);
2411 nfsm_v4build_finalize(&cp);
2412
2413 nfsm_request(vp, NFSV4PROC_COMPOUND, td, cred);
2414 if (error != 0)
2415 goto nfsmout;
2416
2417 nfsm_v4dissect_compound(&cp);
2418 nfsm_v4dissect_putfh(&cp);
2419 nfsm_v4dissect_commit(&cp, &commit);
2420
2421 /* XXX */
2422 /* nfsm_wcc_data(vp, wccflag);*/
2423 if (bcmp(nmp->nm_verf, commit.verf, NFSX_V4VERF)) {
2424 bcopy(commit.verf, nmp->nm_verf, NFSX_V4VERF);
2425 error = NFSERR_STALEWRITEVERF;
2426 }
2427
2428 nfsmout:
2429 error = nfs_v4postop(&cp, error);
2430
2431 if (mrep == NULL)
2432 m_freem(mrep);
2433 return (error);
2434 }
2435
2436 /*
2437 * Strategy routine.
2438 * For async requests when nfsiod(s) are running, queue the request by
2439 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2440 * request.
2441 */
2442 static int
2443 nfs4_strategy(struct vop_strategy_args *ap)
2444 {
2445 struct buf *bp = ap->a_bp;
2446 struct ucred *cr;
2447 int error = 0;
2448
2449 KASSERT(!(bp->b_flags & B_DONE), ("nfs4_strategy: buffer %p unexpectedly marked B_DONE", bp));
2450 KASSERT(BUF_REFCNT(bp) > 0, ("nfs4_strategy: buffer %p not locked", bp));
2451
2452 if (bp->b_iocmd == BIO_READ)
2453 cr = bp->b_rcred;
2454 else
2455 cr = bp->b_wcred;
2456
2457 /*
2458 * If the op is asynchronous and an i/o daemon is waiting
2459 * queue the request, wake it up and wait for completion
2460 * otherwise just do it ourselves.
2461 */
2462 if ((bp->b_flags & B_ASYNC) == 0 ||
2463 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2464 error = nfs_doio(ap->a_vp, bp, cr, curthread);
2465 return (error);
2466 }
2467
2468 /*
2469 * fsync vnode op. Just call nfs4_flush() with commit == 1.
2470 */
2471 /* ARGSUSED */
2472 static int
2473 nfs4_fsync(struct vop_fsync_args *ap)
2474 {
2475 return (nfs4_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2476 }
2477
2478 /*
2479 * Flush all the blocks associated with a vnode.
2480 * Walk through the buffer pool and push any dirty pages
2481 * associated with the vnode.
2482 */
2483 static int
2484 nfs4_flush(struct vnode *vp, int waitfor, struct thread *td,
2485 int commit)
2486 {
2487 struct nfsnode *np = VTONFS(vp);
2488 struct buf *bp;
2489 int i;
2490 struct buf *nbp;
2491 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2492 int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2493 int passone = 1;
2494 u_quad_t off, endoff, toff;
2495 struct ucred* wcred = NULL;
2496 struct buf **bvec = NULL;
2497 #ifndef NFS_COMMITBVECSIZ
2498 #define NFS_COMMITBVECSIZ 20
2499 #endif
2500 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2501 int bvecsize = 0, bveccount;
2502
2503 if (nmp->nm_flag & NFSMNT_INT)
2504 slpflag = PCATCH;
2505 if (!commit)
2506 passone = 0;
2507 /*
2508 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2509 * server, but nas not been committed to stable storage on the server
2510 * yet. On the first pass, the byte range is worked out and the commit
2511 * rpc is done. On the second pass, nfs_writebp() is called to do the
2512 * job.
2513 */
2514 again:
2515 off = (u_quad_t)-1;
2516 endoff = 0;
2517 bvecpos = 0;
2518 if (NFS_ISV3(vp) && commit) {
2519 s = splbio();
2520 if (bvec != NULL && bvec != bvec_on_stack)
2521 free(bvec, M_TEMP);
2522 /*
2523 * Count up how many buffers waiting for a commit.
2524 */
2525 bveccount = 0;
2526 VI_LOCK(vp);
2527 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2528 if (BUF_REFCNT(bp) == 0 &&
2529 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2530 == (B_DELWRI | B_NEEDCOMMIT))
2531 bveccount++;
2532 }
2533 /*
2534 * Allocate space to remember the list of bufs to commit. It is
2535 * important to use M_NOWAIT here to avoid a race with nfs4_write.
2536 * If we can't get memory (for whatever reason), we will end up
2537 * committing the buffers one-by-one in the loop below.
2538 */
2539 if (bveccount > NFS_COMMITBVECSIZ) {
2540 /*
2541 * Release the vnode interlock to avoid a lock
2542 * order reversal.
2543 */
2544 VI_UNLOCK(vp);
2545 bvec = (struct buf **)
2546 malloc(bveccount * sizeof(struct buf *),
2547 M_TEMP, M_NOWAIT);
2548 VI_LOCK(vp);
2549 if (bvec == NULL) {
2550 bvec = bvec_on_stack;
2551 bvecsize = NFS_COMMITBVECSIZ;
2552 } else
2553 bvecsize = bveccount;
2554 } else {
2555 bvec = bvec_on_stack;
2556 bvecsize = NFS_COMMITBVECSIZ;
2557 }
2558 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2559 if (bvecpos >= bvecsize)
2560 break;
2561 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2562 nbp = TAILQ_NEXT(bp, b_bobufs);
2563 continue;
2564 }
2565 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2566 (B_DELWRI | B_NEEDCOMMIT)) {
2567 BUF_UNLOCK(bp);
2568 nbp = TAILQ_NEXT(bp, b_bobufs);
2569 continue;
2570 }
2571 VI_UNLOCK(vp);
2572 bremfree(bp);
2573 /*
2574 * Work out if all buffers are using the same cred
2575 * so we can deal with them all with one commit.
2576 *
2577 * NOTE: we are not clearing B_DONE here, so we have
2578 * to do it later on in this routine if we intend to
2579 * initiate I/O on the bp.
2580 *
2581 * Note: to avoid loopback deadlocks, we do not
2582 * assign b_runningbufspace.
2583 */
2584 if (wcred == NULL)
2585 wcred = bp->b_wcred;
2586 else if (wcred != bp->b_wcred)
2587 wcred = NOCRED;
2588 vfs_busy_pages(bp, 1);
2589
2590 VI_LOCK(vp);
2591 /*
2592 * bp is protected by being locked, but nbp is not
2593 * and vfs_busy_pages() may sleep. We have to
2594 * recalculate nbp.
2595 */
2596 nbp = TAILQ_NEXT(bp, b_bobufs);
2597
2598 /*
2599 * A list of these buffers is kept so that the
2600 * second loop knows which buffers have actually
2601 * been committed. This is necessary, since there
2602 * may be a race between the commit rpc and new
2603 * uncommitted writes on the file.
2604 */
2605 bvec[bvecpos++] = bp;
2606 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2607 bp->b_dirtyoff;
2608 if (toff < off)
2609 off = toff;
2610 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2611 if (toff > endoff)
2612 endoff = toff;
2613 }
2614 splx(s);
2615 VI_UNLOCK(vp);
2616 }
2617 if (bvecpos > 0) {
2618 /*
2619 * Commit data on the server, as required.
2620 * If all bufs are using the same wcred, then use that with
2621 * one call for all of them, otherwise commit each one
2622 * separately.
2623 */
2624 if (wcred != NOCRED)
2625 retv = nfs4_commit(vp, off, (int)(endoff - off),
2626 wcred, td);
2627 else {
2628 retv = 0;
2629 for (i = 0; i < bvecpos; i++) {
2630 off_t off, size;
2631 bp = bvec[i];
2632 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2633 bp->b_dirtyoff;
2634 size = (u_quad_t)(bp->b_dirtyend
2635 - bp->b_dirtyoff);
2636 retv = nfs4_commit(vp, off, (int)size,
2637 bp->b_wcred, td);
2638 if (retv) break;
2639 }
2640 }
2641
2642 if (retv == NFSERR_STALEWRITEVERF)
2643 nfs_clearcommit(vp->v_mount);
2644
2645 /*
2646 * Now, either mark the blocks I/O done or mark the
2647 * blocks dirty, depending on whether the commit
2648 * succeeded.
2649 */
2650 for (i = 0; i < bvecpos; i++) {
2651 bp = bvec[i];
2652 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2653 if (retv) {
2654 /*
2655 * Error, leave B_DELWRI intact
2656 */
2657 vfs_unbusy_pages(bp);
2658 brelse(bp);
2659 } else {
2660 /*
2661 * Success, remove B_DELWRI ( bundirty() ).
2662 *
2663 * b_dirtyoff/b_dirtyend seem to be NFS
2664 * specific. We should probably move that
2665 * into bundirty(). XXX
2666 */
2667 s = splbio();
2668 bufobj_wref(&vp->v_bufobj);
2669 bp->b_flags |= B_ASYNC;
2670 bundirty(bp);
2671 bp->b_flags &= ~B_DONE;
2672 bp->b_ioflags &= ~BIO_ERROR;
2673 bp->b_dirtyoff = bp->b_dirtyend = 0;
2674 splx(s);
2675 bufdone(bp);
2676 }
2677 }
2678 }
2679
2680 /*
2681 * Start/do any write(s) that are required.
2682 */
2683 loop:
2684 s = splbio();
2685 VI_LOCK(vp);
2686 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2687 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2688 if (waitfor != MNT_WAIT || passone)
2689 continue;
2690
2691 error = BUF_TIMELOCK(bp,
2692 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2693 VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
2694 splx(s);
2695 if (error == 0)
2696 panic("nfs4_fsync: inconsistent lock");
2697 if (error == ENOLCK)
2698 goto loop;
2699 if (nfs4_sigintr(nmp, NULL, td)) {
2700 error = EINTR;
2701 goto done;
2702 }
2703 if (slpflag == PCATCH) {
2704 slpflag = 0;
2705 slptimeo = 2 * hz;
2706 }
2707 goto loop;
2708 }
2709 if ((bp->b_flags & B_DELWRI) == 0)
2710 panic("nfs4_fsync: not dirty");
2711 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2712 BUF_UNLOCK(bp);
2713 continue;
2714 }
2715 VI_UNLOCK(vp);
2716 bremfree(bp);
2717 if (passone || !commit)
2718 bp->b_flags |= B_ASYNC;
2719 else
2720 bp->b_flags |= B_ASYNC;
2721 splx(s);
2722 bwrite(bp);
2723 goto loop;
2724 }
2725 splx(s);
2726 if (passone) {
2727 passone = 0;
2728 VI_UNLOCK(vp);
2729 goto again;
2730 }
2731 if (waitfor == MNT_WAIT) {
2732 while (vp->v_bufobj.bo_numoutput) {
2733 error = bufobj_wwait(&vp->v_bufobj, slpflag, slptimeo);
2734 if (error) {
2735 VI_UNLOCK(vp);
2736 if (nfs4_sigintr(nmp, NULL, td)) {
2737 error = EINTR;
2738 goto done;
2739 }
2740 if (slpflag == PCATCH) {
2741 slpflag = 0;
2742 slptimeo = 2 * hz;
2743 }
2744 VI_LOCK(vp);
2745 }
2746 }
2747 if (vp->v_bufobj.bo_dirty.bv_cnt > 0 && commit) {
2748 VI_UNLOCK(vp);
2749 goto loop;
2750 }
2751 }
2752 VI_UNLOCK(vp);
2753 if (np->n_flag & NWRITEERR) {
2754 error = np->n_error;
2755 np->n_flag &= ~NWRITEERR;
2756 }
2757 done:
2758 if (bvec != NULL && bvec != bvec_on_stack)
2759 free(bvec, M_TEMP);
2760 return (error);
2761 }
2762
2763 /*
2764 * NFS advisory byte-level locks.
2765 */
2766 static int
2767 nfs4_advlock(struct vop_advlock_args *ap)
2768 {
2769 return (EPERM);
2770
2771 if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2772 struct nfsnode *np = VTONFS(ap->a_vp);
2773
2774 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
2775 }
2776 return (nfs_dolock(ap));
2777 }
2778
2779 /*
2780 * Print out the contents of an nfsnode.
2781 */
2782 static int
2783 nfs4_print(struct vop_print_args *ap)
2784 {
2785 struct vnode *vp = ap->a_vp;
2786 struct nfsnode *np = VTONFS(vp);
2787
2788 printf("\tfileid %ld fsid 0x%x",
2789 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
2790 if (vp->v_type == VFIFO)
2791 fifo_printinfo(vp);
2792 printf("\n");
2793 return (0);
2794 }
2795
2796 /*
2797 * This is the "real" nfs::bwrite(struct buf*).
2798 * We set B_CACHE if this is a VMIO buffer.
2799 */
2800 int
2801 nfs4_writebp(struct buf *bp, int force __unused, struct thread *td)
2802 {
2803 int s;
2804 int oldflags = bp->b_flags;
2805 #if 0
2806 int retv = 1;
2807 off_t off;
2808 #endif
2809
2810 if (BUF_REFCNT(bp) == 0)
2811 panic("bwrite: buffer is not locked???");
2812
2813 if (bp->b_flags & B_INVAL) {
2814 brelse(bp);
2815 return(0);
2816 }
2817
2818 bp->b_flags |= B_CACHE;
2819
2820 /*
2821 * Undirty the bp. We will redirty it later if the I/O fails.
2822 */
2823
2824 s = splbio();
2825 bundirty(bp);
2826 bp->b_flags &= ~B_DONE;
2827 bp->b_ioflags &= ~BIO_ERROR;
2828 bp->b_iocmd = BIO_WRITE;
2829
2830 bufobj_wref(bp->b_bufobj);
2831 curthread->td_ru.ru_oublock++;
2832 splx(s);
2833
2834 /*
2835 * Note: to avoid loopback deadlocks, we do not
2836 * assign b_runningbufspace.
2837 */
2838 vfs_busy_pages(bp, 1);
2839
2840 BUF_KERNPROC(bp);
2841 bp->b_iooffset = dbtob(bp->b_blkno);
2842 bstrategy(bp);
2843
2844 if( (oldflags & B_ASYNC) == 0) {
2845 int rtval = bufwait(bp);
2846
2847 if (oldflags & B_DELWRI) {
2848 s = splbio();
2849 reassignbuf(bp);
2850 splx(s);
2851 }
2852
2853 brelse(bp);
2854 return (rtval);
2855 }
2856
2857 return (0);
2858 }
2859
2860 /*
2861 * Just call nfs_writebp() with the force argument set to 1.
2862 *
2863 * NOTE: B_DONE may or may not be set in a_bp on call.
2864 */
2865 static int
2866 nfs4_bwrite(struct buf *bp)
2867 {
2868
2869 return (nfs4_writebp(bp, 1, curthread));
2870 }
2871
2872 struct buf_ops buf_ops_nfs4 = {
2873 .bop_name = "buf_ops_nfs4",
2874 .bop_write = nfs4_bwrite,
2875 .bop_strategy = bufstrategy,
2876 .bop_sync = bufsync,
2877 .bop_bdflush = bufbdflush,
2878 };
Cache object: 8345c08af5f29cc94616df2186862067
|