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 * 4. 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 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37
38 /*
39 * vnode op calls for Sun NFS version 2 and 3
40 */
41
42 #include "opt_inet.h"
43
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/resourcevar.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/namei.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/signalvar.h>
63
64 #include <vm/vm.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_extern.h>
67 #include <vm/vm_object.h>
68
69 #include <fs/fifofs/fifo.h>
70
71 #include <rpc/rpcclnt.h>
72
73 #include <nfs/rpcv2.h>
74 #include <nfs/nfsproto.h>
75 #include <nfsclient/nfs.h>
76 #include <nfsclient/nfsnode.h>
77 #include <nfsclient/nfsmount.h>
78 #include <nfsclient/nfs_lock.h>
79 #include <nfs/xdr_subs.h>
80 #include <nfsclient/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 /*
91 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
92 * calls are not in getblk() and brelse() so that they would not be necessary
93 * here.
94 */
95 #ifndef B_VMIO
96 #define vfs_busy_pages(bp, f)
97 #endif
98
99 static vop_read_t nfsfifo_read;
100 static vop_write_t nfsfifo_write;
101 static vop_close_t nfsfifo_close;
102 static int nfs_flush(struct vnode *, int, struct thread *,
103 int);
104 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
105 struct thread *);
106 static vop_lookup_t nfs_lookup;
107 static vop_create_t nfs_create;
108 static vop_mknod_t nfs_mknod;
109 static vop_open_t nfs_open;
110 static vop_close_t nfs_close;
111 static vop_access_t nfs_access;
112 static vop_getattr_t nfs_getattr;
113 static vop_setattr_t nfs_setattr;
114 static vop_read_t nfs_read;
115 static vop_fsync_t nfs_fsync;
116 static vop_remove_t nfs_remove;
117 static vop_link_t nfs_link;
118 static vop_rename_t nfs_rename;
119 static vop_mkdir_t nfs_mkdir;
120 static vop_rmdir_t nfs_rmdir;
121 static vop_symlink_t nfs_symlink;
122 static vop_readdir_t nfs_readdir;
123 static vop_strategy_t nfs_strategy;
124 static int nfs_lookitup(struct vnode *, const char *, int,
125 struct ucred *, struct thread *, struct nfsnode **);
126 static int nfs_sillyrename(struct vnode *, struct vnode *,
127 struct componentname *);
128 static vop_access_t nfsspec_access;
129 static vop_readlink_t nfs_readlink;
130 static vop_print_t nfs_print;
131 static vop_advlock_t nfs_advlock;
132
133 /*
134 * Global vfs data structures for nfs
135 */
136 struct vop_vector nfs_vnodeops = {
137 .vop_default = &default_vnodeops,
138 .vop_access = nfs_access,
139 .vop_advlock = nfs_advlock,
140 .vop_close = nfs_close,
141 .vop_create = nfs_create,
142 .vop_fsync = nfs_fsync,
143 .vop_getattr = nfs_getattr,
144 .vop_getpages = nfs_getpages,
145 .vop_putpages = nfs_putpages,
146 .vop_inactive = nfs_inactive,
147 .vop_lease = VOP_NULL,
148 .vop_link = nfs_link,
149 .vop_lookup = nfs_lookup,
150 .vop_mkdir = nfs_mkdir,
151 .vop_mknod = nfs_mknod,
152 .vop_open = nfs_open,
153 .vop_print = nfs_print,
154 .vop_read = nfs_read,
155 .vop_readdir = nfs_readdir,
156 .vop_readlink = nfs_readlink,
157 .vop_reclaim = nfs_reclaim,
158 .vop_remove = nfs_remove,
159 .vop_rename = nfs_rename,
160 .vop_rmdir = nfs_rmdir,
161 .vop_setattr = nfs_setattr,
162 .vop_strategy = nfs_strategy,
163 .vop_symlink = nfs_symlink,
164 .vop_write = nfs_write,
165 };
166
167 struct vop_vector nfs_fifoops = {
168 .vop_default = &fifo_specops,
169 .vop_access = nfsspec_access,
170 .vop_close = nfsfifo_close,
171 .vop_fsync = nfs_fsync,
172 .vop_getattr = nfs_getattr,
173 .vop_inactive = nfs_inactive,
174 .vop_print = nfs_print,
175 .vop_read = nfsfifo_read,
176 .vop_reclaim = nfs_reclaim,
177 .vop_setattr = nfs_setattr,
178 .vop_write = nfsfifo_write,
179 };
180
181 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
182 struct componentname *cnp, struct vattr *vap);
183 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
184 struct ucred *cred, struct thread *td);
185 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
186 int fnamelen, struct vnode *tdvp,
187 const char *tnameptr, int tnamelen,
188 struct ucred *cred, struct thread *td);
189 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
190 struct sillyrename *sp);
191
192 /*
193 * Global variables
194 */
195 struct mtx nfs_iod_mtx;
196 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
197 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
198 int nfs_numasync = 0;
199 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
200
201 SYSCTL_DECL(_vfs_nfs);
202
203 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
204 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
205 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
206
207 static int nfsv3_commit_on_close = 0;
208 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
209 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
210
211 static int nfs_clean_pages_on_close = 1;
212 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
213 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
214
215 int nfs_directio_enable = 0;
216 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
217 &nfs_directio_enable, 0, "Enable NFS directio");
218
219 /*
220 * This sysctl allows other processes to mmap a file that has been opened
221 * O_DIRECT by a process. In general, having processes mmap the file while
222 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
223 * this by default to prevent DoS attacks - to prevent a malicious user from
224 * opening up files O_DIRECT preventing other users from mmap'ing these
225 * files. "Protected" environments where stricter consistency guarantees are
226 * required can disable this knob. The process that opened the file O_DIRECT
227 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
228 * meaningful.
229 */
230 int nfs_directio_allow_mmap = 1;
231 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
232 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
233
234 #if 0
235 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
236 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
237
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
239 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
240 #endif
241
242 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
243 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
244 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
245
246 /*
247 * SMP Locking Note :
248 * The list of locks after the description of the lock is the ordering
249 * of other locks acquired with the lock held.
250 * np->n_mtx : Protects the fields in the nfsnode.
251 VM Object Lock
252 VI_MTX (acquired indirectly)
253 * nmp->nm_mtx : Protects the fields in the nfsmount.
254 rep->r_mtx
255 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
256 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
257 nmp->nm_mtx
258 rep->r_mtx
259 * rep->r_mtx : Protects the fields in an nfsreq.
260 */
261
262 static int
263 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
264 struct ucred *cred)
265 {
266 const int v3 = 1;
267 u_int32_t *tl;
268 int error = 0, attrflag;
269
270 struct mbuf *mreq, *mrep, *md, *mb;
271 caddr_t bpos, dpos;
272 u_int32_t rmode;
273 struct nfsnode *np = VTONFS(vp);
274
275 nfsstats.rpccnt[NFSPROC_ACCESS]++;
276 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
277 mb = mreq;
278 bpos = mtod(mb, caddr_t);
279 nfsm_fhtom(vp, v3);
280 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
281 *tl = txdr_unsigned(wmode);
282 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
283 nfsm_postop_attr(vp, attrflag);
284 if (!error) {
285 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
286 rmode = fxdr_unsigned(u_int32_t, *tl);
287 mtx_lock(&np->n_mtx);
288 np->n_mode = rmode;
289 np->n_modeuid = cred->cr_uid;
290 np->n_modestamp = time_second;
291 mtx_unlock(&np->n_mtx);
292 }
293 m_freem(mrep);
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 static int
305 nfs_access(struct vop_access_args *ap)
306 {
307 struct vnode *vp = ap->a_vp;
308 int error = 0;
309 u_int32_t mode, wmode;
310 int v3 = NFS_ISV3(vp);
311 struct nfsnode *np = VTONFS(vp);
312
313 /*
314 * Disallow write attempts on filesystems mounted read-only;
315 * unless the file is a socket, fifo, or a block or character
316 * device resident on the filesystem.
317 */
318 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
319 switch (vp->v_type) {
320 case VREG:
321 case VDIR:
322 case VLNK:
323 return (EROFS);
324 default:
325 break;
326 }
327 }
328 /*
329 * For nfs v3, check to see if we have done this recently, and if
330 * so return our cached result instead of making an ACCESS call.
331 * If not, do an access rpc, otherwise you are stuck emulating
332 * ufs_access() locally using the vattr. This may not be correct,
333 * since the server may apply other access criteria such as
334 * client uid-->server uid mapping that we do not know about.
335 */
336 if (v3) {
337 if (ap->a_mode & VREAD)
338 mode = NFSV3ACCESS_READ;
339 else
340 mode = 0;
341 if (vp->v_type != VDIR) {
342 if (ap->a_mode & VWRITE)
343 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
344 if (ap->a_mode & VEXEC)
345 mode |= NFSV3ACCESS_EXECUTE;
346 } else {
347 if (ap->a_mode & VWRITE)
348 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
349 NFSV3ACCESS_DELETE);
350 if (ap->a_mode & VEXEC)
351 mode |= NFSV3ACCESS_LOOKUP;
352 }
353 /* XXX safety belt, only make blanket request if caching */
354 if (nfsaccess_cache_timeout > 0) {
355 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
356 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
357 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
358 } else {
359 wmode = mode;
360 }
361
362 /*
363 * Does our cached result allow us to give a definite yes to
364 * this request?
365 */
366 mtx_lock(&np->n_mtx);
367 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
368 (ap->a_cred->cr_uid == np->n_modeuid) &&
369 ((np->n_mode & mode) == mode)) {
370 nfsstats.accesscache_hits++;
371 } else {
372 /*
373 * Either a no, or a don't know. Go to the wire.
374 */
375 nfsstats.accesscache_misses++;
376 mtx_unlock(&np->n_mtx);
377 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
378 mtx_lock(&np->n_mtx);
379 if (!error) {
380 if ((np->n_mode & mode) != mode) {
381 error = EACCES;
382 }
383 }
384 }
385 mtx_unlock(&np->n_mtx);
386 return (error);
387 } else {
388 if ((error = nfsspec_access(ap)) != 0) {
389 return (error);
390 }
391 /*
392 * Attempt to prevent a mapped root from accessing a file
393 * which it shouldn't. We try to read a byte from the file
394 * if the user is root and the file is not zero length.
395 * After calling nfsspec_access, we should have the correct
396 * file size cached.
397 */
398 mtx_lock(&np->n_mtx);
399 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
400 && VTONFS(vp)->n_size > 0) {
401 struct iovec aiov;
402 struct uio auio;
403 char buf[1];
404
405 mtx_unlock(&np->n_mtx);
406 aiov.iov_base = buf;
407 aiov.iov_len = 1;
408 auio.uio_iov = &aiov;
409 auio.uio_iovcnt = 1;
410 auio.uio_offset = 0;
411 auio.uio_resid = 1;
412 auio.uio_segflg = UIO_SYSSPACE;
413 auio.uio_rw = UIO_READ;
414 auio.uio_td = ap->a_td;
415
416 if (vp->v_type == VREG)
417 error = nfs_readrpc(vp, &auio, ap->a_cred);
418 else if (vp->v_type == VDIR) {
419 char* bp;
420 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
421 aiov.iov_base = bp;
422 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
423 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
424 free(bp, M_TEMP);
425 } else if (vp->v_type == VLNK)
426 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
427 else
428 error = EACCES;
429 } else
430 mtx_unlock(&np->n_mtx);
431 return (error);
432 }
433 }
434
435 int nfs_otw_getattr_avoid = 0;
436
437 /*
438 * nfs open vnode op
439 * Check to see if the type is ok
440 * and that deletion is not in progress.
441 * For paged in text files, you will need to flush the page cache
442 * if consistency is lost.
443 */
444 /* ARGSUSED */
445 static int
446 nfs_open(struct vop_open_args *ap)
447 {
448 struct vnode *vp = ap->a_vp;
449 struct nfsnode *np = VTONFS(vp);
450 struct vattr vattr;
451 int error;
452 int fmode = ap->a_mode;
453
454 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
455 return (EOPNOTSUPP);
456
457 /*
458 * Get a valid lease. If cached data is stale, flush it.
459 */
460 mtx_lock(&np->n_mtx);
461 if (np->n_flag & NMODIFIED) {
462 mtx_unlock(&np->n_mtx);
463 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
464 if (error == EINTR || error == EIO)
465 return (error);
466 np->n_attrstamp = 0;
467 if (vp->v_type == VDIR)
468 np->n_direofoffset = 0;
469 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
470 if (error)
471 return (error);
472 mtx_lock(&np->n_mtx);
473 np->n_mtime = vattr.va_mtime;
474 mtx_unlock(&np->n_mtx);
475 } else {
476 struct thread *td = curthread;
477
478 if (np->n_ac_ts_syscalls != td->td_syscalls ||
479 np->n_ac_ts_tid != td->td_tid ||
480 td->td_proc == NULL ||
481 np->n_ac_ts_pid != td->td_proc->p_pid) {
482 np->n_attrstamp = 0;
483 }
484 mtx_unlock(&np->n_mtx);
485 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
486 if (error)
487 return (error);
488 mtx_lock(&np->n_mtx);
489 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
490 if (vp->v_type == VDIR)
491 np->n_direofoffset = 0;
492 mtx_unlock(&np->n_mtx);
493 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
494 if (error == EINTR || error == EIO) {
495 return (error);
496 }
497 mtx_lock(&np->n_mtx);
498 np->n_mtime = vattr.va_mtime;
499 }
500 mtx_unlock(&np->n_mtx);
501 }
502 /*
503 * If the object has >= 1 O_DIRECT active opens, we disable caching.
504 */
505 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
506 if (np->n_directio_opens == 0) {
507 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
508 if (error)
509 return (error);
510 mtx_lock(&np->n_mtx);
511 np->n_flag |= NNONCACHE;
512 mtx_unlock(&np->n_mtx);
513 }
514 np->n_directio_opens++;
515 }
516 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
517 return (0);
518 }
519
520 /*
521 * nfs close vnode op
522 * What an NFS client should do upon close after writing is a debatable issue.
523 * Most NFS clients push delayed writes to the server upon close, basically for
524 * two reasons:
525 * 1 - So that any write errors may be reported back to the client process
526 * doing the close system call. By far the two most likely errors are
527 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
528 * 2 - To put a worst case upper bound on cache inconsistency between
529 * multiple clients for the file.
530 * There is also a consistency problem for Version 2 of the protocol w.r.t.
531 * not being able to tell if other clients are writing a file concurrently,
532 * since there is no way of knowing if the changed modify time in the reply
533 * is only due to the write for this client.
534 * (NFS Version 3 provides weak cache consistency data in the reply that
535 * should be sufficient to detect and handle this case.)
536 *
537 * The current code does the following:
538 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
539 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
540 * or commit them (this satisfies 1 and 2 except for the
541 * case where the server crashes after this close but
542 * before the commit RPC, which is felt to be "good
543 * enough". Changing the last argument to nfs_flush() to
544 * a 1 would force a commit operation, if it is felt a
545 * commit is necessary now.
546 */
547 /* ARGSUSED */
548 static int
549 nfs_close(struct vop_close_args *ap)
550 {
551 struct vnode *vp = ap->a_vp;
552 struct nfsnode *np = VTONFS(vp);
553 int error = 0;
554 int fmode = ap->a_fflag;
555
556 if (vp->v_type == VREG) {
557 /*
558 * Examine and clean dirty pages, regardless of NMODIFIED.
559 * This closes a major hole in close-to-open consistency.
560 * We want to push out all dirty pages (and buffers) on
561 * close, regardless of whether they were dirtied by
562 * mmap'ed writes or via write().
563 */
564 if (nfs_clean_pages_on_close && vp->v_object) {
565 VM_OBJECT_LOCK(vp->v_object);
566 vm_object_page_clean(vp->v_object, 0, 0, 0);
567 VM_OBJECT_UNLOCK(vp->v_object);
568 }
569 mtx_lock(&np->n_mtx);
570 if (np->n_flag & NMODIFIED) {
571 mtx_unlock(&np->n_mtx);
572 if (NFS_ISV3(vp)) {
573 /*
574 * Under NFSv3 we have dirty buffers to dispose of. We
575 * must flush them to the NFS server. We have the option
576 * of waiting all the way through the commit rpc or just
577 * waiting for the initial write. The default is to only
578 * wait through the initial write so the data is in the
579 * server's cache, which is roughly similar to the state
580 * a standard disk subsystem leaves the file in on close().
581 *
582 * We cannot clear the NMODIFIED bit in np->n_flag due to
583 * potential races with other processes, and certainly
584 * cannot clear it if we don't commit.
585 */
586 int cm = nfsv3_commit_on_close ? 1 : 0;
587 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
588 /* np->n_flag &= ~NMODIFIED; */
589 } else
590 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
591 mtx_lock(&np->n_mtx);
592 }
593 /*
594 * Invalidate the attribute cache in all cases.
595 * An open is going to fetch fresh attrs any way, other procs
596 * on this node that have file open will be forced to do an
597 * otw attr fetch, but this is safe.
598 */
599 np->n_attrstamp = 0;
600 if (np->n_flag & NWRITEERR) {
601 np->n_flag &= ~NWRITEERR;
602 error = np->n_error;
603 }
604 mtx_unlock(&np->n_mtx);
605 }
606 if (nfs_directio_enable)
607 KASSERT((np->n_directio_asyncwr == 0),
608 ("nfs_close: dirty unflushed (%d) directio buffers\n",
609 np->n_directio_asyncwr));
610 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
611 mtx_lock(&np->n_mtx);
612 KASSERT((np->n_directio_opens > 0),
613 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
614 np->n_directio_opens--;
615 if (np->n_directio_opens == 0)
616 np->n_flag &= ~NNONCACHE;
617 mtx_unlock(&np->n_mtx);
618 }
619 return (error);
620 }
621
622 /*
623 * nfs getattr call from vfs.
624 */
625 static int
626 nfs_getattr(struct vop_getattr_args *ap)
627 {
628 struct vnode *vp = ap->a_vp;
629 struct nfsnode *np = VTONFS(vp);
630 caddr_t bpos, dpos;
631 int error = 0;
632 struct mbuf *mreq, *mrep, *md, *mb;
633 int v3 = NFS_ISV3(vp);
634
635 /*
636 * Update local times for special files.
637 */
638 mtx_lock(&np->n_mtx);
639 if (np->n_flag & (NACC | NUPD))
640 np->n_flag |= NCHG;
641 mtx_unlock(&np->n_mtx);
642 /*
643 * First look in the cache.
644 */
645 if (nfs_getattrcache(vp, ap->a_vap) == 0)
646 goto nfsmout;
647 if (v3 && nfsaccess_cache_timeout > 0) {
648 nfsstats.accesscache_misses++;
649 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
650 if (nfs_getattrcache(vp, ap->a_vap) == 0)
651 goto nfsmout;
652 }
653 nfsstats.rpccnt[NFSPROC_GETATTR]++;
654 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
655 mb = mreq;
656 bpos = mtod(mb, caddr_t);
657 nfsm_fhtom(vp, v3);
658 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
659 if (!error) {
660 nfsm_loadattr(vp, ap->a_vap);
661 }
662 m_freem(mrep);
663 nfsmout:
664 return (error);
665 }
666
667 /*
668 * nfs setattr call.
669 */
670 static int
671 nfs_setattr(struct vop_setattr_args *ap)
672 {
673 struct vnode *vp = ap->a_vp;
674 struct nfsnode *np = VTONFS(vp);
675 struct vattr *vap = ap->a_vap;
676 int error = 0;
677 u_quad_t tsize;
678
679 #ifndef nolint
680 tsize = (u_quad_t)0;
681 #endif
682
683 /*
684 * Setting of flags and marking of atimes are not supported.
685 */
686 if (vap->va_flags != VNOVAL || (vap->va_vaflags & VA_MARK_ATIME))
687 return (EOPNOTSUPP);
688
689 /*
690 * Disallow write attempts if the filesystem is mounted read-only.
691 */
692 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
693 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
694 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
695 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
696 error = EROFS;
697 goto out;
698 }
699 if (vap->va_size != VNOVAL) {
700 switch (vp->v_type) {
701 case VDIR:
702 return (EISDIR);
703 case VCHR:
704 case VBLK:
705 case VSOCK:
706 case VFIFO:
707 if (vap->va_mtime.tv_sec == VNOVAL &&
708 vap->va_atime.tv_sec == VNOVAL &&
709 vap->va_mode == (mode_t)VNOVAL &&
710 vap->va_uid == (uid_t)VNOVAL &&
711 vap->va_gid == (gid_t)VNOVAL)
712 return (0);
713 vap->va_size = VNOVAL;
714 break;
715 default:
716 /*
717 * Disallow write attempts if the filesystem is
718 * mounted read-only.
719 */
720 if (vp->v_mount->mnt_flag & MNT_RDONLY)
721 return (EROFS);
722 /*
723 * We run vnode_pager_setsize() early (why?),
724 * we must set np->n_size now to avoid vinvalbuf
725 * V_SAVE races that might setsize a lower
726 * value.
727 */
728 mtx_lock(&np->n_mtx);
729 tsize = np->n_size;
730 mtx_unlock(&np->n_mtx);
731 error = nfs_meta_setsize(vp, ap->a_cred,
732 ap->a_td, vap->va_size);
733 mtx_lock(&np->n_mtx);
734 if (np->n_flag & NMODIFIED) {
735 tsize = np->n_size;
736 mtx_unlock(&np->n_mtx);
737 if (vap->va_size == 0)
738 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
739 else
740 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
741 if (error) {
742 vnode_pager_setsize(vp, tsize);
743 goto out;
744 }
745 } else
746 mtx_unlock(&np->n_mtx);
747 /*
748 * np->n_size has already been set to vap->va_size
749 * in nfs_meta_setsize(). We must set it again since
750 * nfs_loadattrcache() could be called through
751 * nfs_meta_setsize() and could modify np->n_size.
752 */
753 mtx_lock(&np->n_mtx);
754 np->n_vattr.va_size = np->n_size = vap->va_size;
755 mtx_unlock(&np->n_mtx);
756 };
757 } else {
758 mtx_lock(&np->n_mtx);
759 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
760 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
761 mtx_unlock(&np->n_mtx);
762 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) != 0 &&
763 (error == EINTR || error == EIO))
764 return error;
765 } else
766 mtx_unlock(&np->n_mtx);
767 }
768 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
769 if (error && vap->va_size != VNOVAL) {
770 mtx_lock(&np->n_mtx);
771 np->n_size = np->n_vattr.va_size = tsize;
772 vnode_pager_setsize(vp, tsize);
773 mtx_unlock(&np->n_mtx);
774 }
775 out:
776 return (error);
777 }
778
779 /*
780 * Do an nfs setattr rpc.
781 */
782 static int
783 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
784 struct thread *td)
785 {
786 struct nfsv2_sattr *sp;
787 struct nfsnode *np = VTONFS(vp);
788 caddr_t bpos, dpos;
789 u_int32_t *tl;
790 int error = 0, wccflag = NFSV3_WCCRATTR;
791 struct mbuf *mreq, *mrep, *md, *mb;
792 int v3 = NFS_ISV3(vp);
793
794 nfsstats.rpccnt[NFSPROC_SETATTR]++;
795 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
796 mb = mreq;
797 bpos = mtod(mb, caddr_t);
798 nfsm_fhtom(vp, v3);
799 if (v3) {
800 nfsm_v3attrbuild(vap, TRUE);
801 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
802 *tl = nfs_false;
803 } else {
804 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
805 if (vap->va_mode == (mode_t)VNOVAL)
806 sp->sa_mode = nfs_xdrneg1;
807 else
808 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
809 if (vap->va_uid == (uid_t)VNOVAL)
810 sp->sa_uid = nfs_xdrneg1;
811 else
812 sp->sa_uid = txdr_unsigned(vap->va_uid);
813 if (vap->va_gid == (gid_t)VNOVAL)
814 sp->sa_gid = nfs_xdrneg1;
815 else
816 sp->sa_gid = txdr_unsigned(vap->va_gid);
817 sp->sa_size = txdr_unsigned(vap->va_size);
818 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
819 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
820 }
821 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
822 if (v3) {
823 np->n_modestamp = 0;
824 nfsm_wcc_data(vp, wccflag);
825 } else
826 nfsm_loadattr(vp, NULL);
827 m_freem(mrep);
828 nfsmout:
829 return (error);
830 }
831
832 /*
833 * nfs lookup call, one step at a time...
834 * First look in cache
835 * If not found, unlock the directory nfsnode and do the rpc
836 */
837 static int
838 nfs_lookup(struct vop_lookup_args *ap)
839 {
840 struct componentname *cnp = ap->a_cnp;
841 struct vnode *dvp = ap->a_dvp;
842 struct vnode **vpp = ap->a_vpp;
843 int flags = cnp->cn_flags;
844 struct vnode *newvp;
845 struct nfsmount *nmp;
846 caddr_t bpos, dpos;
847 struct mbuf *mreq, *mrep, *md, *mb;
848 long len;
849 nfsfh_t *fhp;
850 struct nfsnode *np;
851 int error = 0, attrflag, fhsize;
852 int v3 = NFS_ISV3(dvp);
853 struct thread *td = cnp->cn_thread;
854
855 *vpp = NULLVP;
856 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
857 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
858 return (EROFS);
859 if (dvp->v_type != VDIR)
860 return (ENOTDIR);
861 nmp = VFSTONFS(dvp->v_mount);
862 np = VTONFS(dvp);
863 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
864 *vpp = NULLVP;
865 return (error);
866 }
867 if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
868 struct vattr vattr;
869
870 newvp = *vpp;
871 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
872 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
873 nfsstats.lookupcache_hits++;
874 if (cnp->cn_nameiop != LOOKUP &&
875 (flags & ISLASTCN))
876 cnp->cn_flags |= SAVENAME;
877 return (0);
878 }
879 cache_purge(newvp);
880 if (dvp != newvp)
881 vput(newvp);
882 else
883 vrele(newvp);
884 *vpp = NULLVP;
885 }
886 error = 0;
887 newvp = NULLVP;
888 nfsstats.lookupcache_misses++;
889 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
890 len = cnp->cn_namelen;
891 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
892 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
893 mb = mreq;
894 bpos = mtod(mb, caddr_t);
895 nfsm_fhtom(dvp, v3);
896 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
897 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
898 if (error) {
899 if (v3) {
900 nfsm_postop_attr(dvp, attrflag);
901 m_freem(mrep);
902 }
903 goto nfsmout;
904 }
905 nfsm_getfh(fhp, fhsize, v3);
906
907 /*
908 * Handle RENAME case...
909 */
910 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
911 if (NFS_CMPFH(np, fhp, fhsize)) {
912 m_freem(mrep);
913 return (EISDIR);
914 }
915 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, LK_EXCLUSIVE);
916 if (error) {
917 m_freem(mrep);
918 return (error);
919 }
920 newvp = NFSTOV(np);
921 if (v3) {
922 nfsm_postop_attr(newvp, attrflag);
923 nfsm_postop_attr(dvp, attrflag);
924 } else
925 nfsm_loadattr(newvp, NULL);
926 *vpp = newvp;
927 m_freem(mrep);
928 cnp->cn_flags |= SAVENAME;
929 return (0);
930 }
931
932 if (flags & ISDOTDOT) {
933 VOP_UNLOCK(dvp, 0, td);
934 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, cnp->cn_lkflags);
935 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
936 if (error)
937 return (error);
938 newvp = NFSTOV(np);
939 } else if (NFS_CMPFH(np, fhp, fhsize)) {
940 VREF(dvp);
941 newvp = dvp;
942 } else {
943 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, cnp->cn_lkflags);
944 if (error) {
945 m_freem(mrep);
946 return (error);
947 }
948 newvp = NFSTOV(np);
949 }
950 if (v3) {
951 nfsm_postop_attr(newvp, attrflag);
952 nfsm_postop_attr(dvp, attrflag);
953 } else
954 nfsm_loadattr(newvp, NULL);
955 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
956 cnp->cn_flags |= SAVENAME;
957 if ((cnp->cn_flags & MAKEENTRY) &&
958 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
959 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
960 cache_enter(dvp, newvp, cnp);
961 }
962 *vpp = newvp;
963 m_freem(mrep);
964 nfsmout:
965 if (error) {
966 if (newvp != NULLVP) {
967 vput(newvp);
968 *vpp = NULLVP;
969 }
970 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
971 (flags & ISLASTCN) && error == ENOENT) {
972 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
973 error = EROFS;
974 else
975 error = EJUSTRETURN;
976 }
977 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
978 cnp->cn_flags |= SAVENAME;
979 }
980 return (error);
981 }
982
983 /*
984 * nfs read call.
985 * Just call nfs_bioread() to do the work.
986 */
987 static int
988 nfs_read(struct vop_read_args *ap)
989 {
990 struct vnode *vp = ap->a_vp;
991
992 switch (vp->v_type) {
993 case VREG:
994 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
995 case VDIR:
996 return (EISDIR);
997 default:
998 return (EOPNOTSUPP);
999 }
1000 }
1001
1002 /*
1003 * nfs readlink call
1004 */
1005 static int
1006 nfs_readlink(struct vop_readlink_args *ap)
1007 {
1008 struct vnode *vp = ap->a_vp;
1009
1010 if (vp->v_type != VLNK)
1011 return (EINVAL);
1012 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1013 }
1014
1015 /*
1016 * Do a readlink rpc.
1017 * Called by nfs_doio() from below the buffer cache.
1018 */
1019 int
1020 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1021 {
1022 caddr_t bpos, dpos;
1023 int error = 0, len, attrflag;
1024 struct mbuf *mreq, *mrep, *md, *mb;
1025 int v3 = NFS_ISV3(vp);
1026
1027 nfsstats.rpccnt[NFSPROC_READLINK]++;
1028 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1029 mb = mreq;
1030 bpos = mtod(mb, caddr_t);
1031 nfsm_fhtom(vp, v3);
1032 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1033 if (v3)
1034 nfsm_postop_attr(vp, attrflag);
1035 if (!error) {
1036 nfsm_strsiz(len, NFS_MAXPATHLEN);
1037 if (len == NFS_MAXPATHLEN) {
1038 struct nfsnode *np = VTONFS(vp);
1039 mtx_lock(&np->n_mtx);
1040 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1041 len = np->n_size;
1042 mtx_unlock(&np->n_mtx);
1043 }
1044 nfsm_mtouio(uiop, len);
1045 }
1046 m_freem(mrep);
1047 nfsmout:
1048 return (error);
1049 }
1050
1051 /*
1052 * nfs read rpc call
1053 * Ditto above
1054 */
1055 int
1056 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1057 {
1058 u_int32_t *tl;
1059 caddr_t bpos, dpos;
1060 struct mbuf *mreq, *mrep, *md, *mb;
1061 struct nfsmount *nmp;
1062 int error = 0, len, retlen, tsiz, eof, attrflag;
1063 int v3 = NFS_ISV3(vp);
1064 int rsize;
1065
1066 #ifndef nolint
1067 eof = 0;
1068 #endif
1069 nmp = VFSTONFS(vp->v_mount);
1070 tsiz = uiop->uio_resid;
1071 mtx_lock(&nmp->nm_mtx);
1072 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1073 mtx_unlock(&nmp->nm_mtx);
1074 return (EFBIG);
1075 }
1076 rsize = nmp->nm_rsize;
1077 mtx_unlock(&nmp->nm_mtx);
1078 while (tsiz > 0) {
1079 nfsstats.rpccnt[NFSPROC_READ]++;
1080 len = (tsiz > rsize) ? rsize : tsiz;
1081 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1082 mb = mreq;
1083 bpos = mtod(mb, caddr_t);
1084 nfsm_fhtom(vp, v3);
1085 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1086 if (v3) {
1087 txdr_hyper(uiop->uio_offset, tl);
1088 *(tl + 2) = txdr_unsigned(len);
1089 } else {
1090 *tl++ = txdr_unsigned(uiop->uio_offset);
1091 *tl++ = txdr_unsigned(len);
1092 *tl = 0;
1093 }
1094 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1095 if (v3) {
1096 nfsm_postop_attr(vp, attrflag);
1097 if (error) {
1098 m_freem(mrep);
1099 goto nfsmout;
1100 }
1101 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1102 eof = fxdr_unsigned(int, *(tl + 1));
1103 } else {
1104 nfsm_loadattr(vp, NULL);
1105 }
1106 nfsm_strsiz(retlen, rsize);
1107 nfsm_mtouio(uiop, retlen);
1108 m_freem(mrep);
1109 tsiz -= retlen;
1110 if (v3) {
1111 if (eof || retlen == 0) {
1112 tsiz = 0;
1113 }
1114 } else if (retlen < len) {
1115 tsiz = 0;
1116 }
1117 }
1118 nfsmout:
1119 return (error);
1120 }
1121
1122 /*
1123 * nfs write call
1124 */
1125 int
1126 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1127 int *iomode, int *must_commit)
1128 {
1129 u_int32_t *tl;
1130 int32_t backup;
1131 caddr_t bpos, dpos;
1132 struct mbuf *mreq, *mrep, *md, *mb;
1133 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1134 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1135 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1136 int wsize;
1137
1138 #ifndef DIAGNOSTIC
1139 if (uiop->uio_iovcnt != 1)
1140 panic("nfs: writerpc iovcnt > 1");
1141 #endif
1142 *must_commit = 0;
1143 tsiz = uiop->uio_resid;
1144 mtx_lock(&nmp->nm_mtx);
1145 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1146 mtx_unlock(&nmp->nm_mtx);
1147 return (EFBIG);
1148 }
1149 wsize = nmp->nm_wsize;
1150 mtx_unlock(&nmp->nm_mtx);
1151 while (tsiz > 0) {
1152 nfsstats.rpccnt[NFSPROC_WRITE]++;
1153 len = (tsiz > wsize) ? wsize : tsiz;
1154 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1155 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1156 mb = mreq;
1157 bpos = mtod(mb, caddr_t);
1158 nfsm_fhtom(vp, v3);
1159 if (v3) {
1160 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1161 txdr_hyper(uiop->uio_offset, tl);
1162 tl += 2;
1163 *tl++ = txdr_unsigned(len);
1164 *tl++ = txdr_unsigned(*iomode);
1165 *tl = txdr_unsigned(len);
1166 } else {
1167 u_int32_t x;
1168
1169 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1170 /* Set both "begin" and "current" to non-garbage. */
1171 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1172 *tl++ = x; /* "begin offset" */
1173 *tl++ = x; /* "current offset" */
1174 x = txdr_unsigned(len);
1175 *tl++ = x; /* total to this offset */
1176 *tl = x; /* size of this write */
1177 }
1178 nfsm_uiotom(uiop, len);
1179 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1180 if (v3) {
1181 wccflag = NFSV3_WCCCHK;
1182 nfsm_wcc_data(vp, wccflag);
1183 if (!error) {
1184 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1185 + NFSX_V3WRITEVERF);
1186 rlen = fxdr_unsigned(int, *tl++);
1187 if (rlen == 0) {
1188 error = NFSERR_IO;
1189 m_freem(mrep);
1190 break;
1191 } else if (rlen < len) {
1192 backup = len - rlen;
1193 uiop->uio_iov->iov_base =
1194 (char *)uiop->uio_iov->iov_base -
1195 backup;
1196 uiop->uio_iov->iov_len += backup;
1197 uiop->uio_offset -= backup;
1198 uiop->uio_resid += backup;
1199 len = rlen;
1200 }
1201 commit = fxdr_unsigned(int, *tl++);
1202
1203 /*
1204 * Return the lowest committment level
1205 * obtained by any of the RPCs.
1206 */
1207 if (committed == NFSV3WRITE_FILESYNC)
1208 committed = commit;
1209 else if (committed == NFSV3WRITE_DATASYNC &&
1210 commit == NFSV3WRITE_UNSTABLE)
1211 committed = commit;
1212 mtx_lock(&nmp->nm_mtx);
1213 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1214 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1215 NFSX_V3WRITEVERF);
1216 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1217 } else if (bcmp((caddr_t)tl,
1218 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1219 *must_commit = 1;
1220 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1221 NFSX_V3WRITEVERF);
1222 }
1223 mtx_unlock(&nmp->nm_mtx);
1224 }
1225 } else {
1226 nfsm_loadattr(vp, NULL);
1227 }
1228 if (wccflag) {
1229 mtx_lock(&(VTONFS(vp))->n_mtx);
1230 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1231 mtx_unlock(&(VTONFS(vp))->n_mtx);
1232 }
1233 m_freem(mrep);
1234 if (error)
1235 break;
1236 tsiz -= len;
1237 }
1238 nfsmout:
1239 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1240 committed = NFSV3WRITE_FILESYNC;
1241 *iomode = committed;
1242 if (error)
1243 uiop->uio_resid = tsiz;
1244 return (error);
1245 }
1246
1247 /*
1248 * nfs mknod rpc
1249 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1250 * mode set to specify the file type and the size field for rdev.
1251 */
1252 static int
1253 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1254 struct vattr *vap)
1255 {
1256 struct nfsv2_sattr *sp;
1257 u_int32_t *tl;
1258 struct vnode *newvp = NULL;
1259 struct nfsnode *np = NULL;
1260 struct vattr vattr;
1261 caddr_t bpos, dpos;
1262 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1263 struct mbuf *mreq, *mrep, *md, *mb;
1264 u_int32_t rdev;
1265 int v3 = NFS_ISV3(dvp);
1266
1267 if (vap->va_type == VCHR || vap->va_type == VBLK)
1268 rdev = txdr_unsigned(vap->va_rdev);
1269 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1270 rdev = nfs_xdrneg1;
1271 else {
1272 return (EOPNOTSUPP);
1273 }
1274 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1275 return (error);
1276 }
1277 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1278 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1279 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1280 mb = mreq;
1281 bpos = mtod(mb, caddr_t);
1282 nfsm_fhtom(dvp, v3);
1283 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1284 if (v3) {
1285 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1286 *tl++ = vtonfsv3_type(vap->va_type);
1287 nfsm_v3attrbuild(vap, FALSE);
1288 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1289 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1290 *tl++ = txdr_unsigned(umajor(vap->va_rdev));
1291 *tl = txdr_unsigned(uminor(vap->va_rdev));
1292 }
1293 } else {
1294 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1295 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1296 sp->sa_uid = nfs_xdrneg1;
1297 sp->sa_gid = nfs_xdrneg1;
1298 sp->sa_size = rdev;
1299 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1300 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1301 }
1302 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1303 if (!error) {
1304 nfsm_mtofh(dvp, newvp, v3, gotvp);
1305 if (!gotvp) {
1306 if (newvp) {
1307 vput(newvp);
1308 newvp = NULL;
1309 }
1310 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1311 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1312 if (!error)
1313 newvp = NFSTOV(np);
1314 }
1315 }
1316 if (v3)
1317 nfsm_wcc_data(dvp, wccflag);
1318 m_freem(mrep);
1319 nfsmout:
1320 if (error) {
1321 if (newvp)
1322 vput(newvp);
1323 } else {
1324 if (cnp->cn_flags & MAKEENTRY)
1325 cache_enter(dvp, newvp, cnp);
1326 *vpp = newvp;
1327 }
1328 mtx_lock(&(VTONFS(dvp))->n_mtx);
1329 VTONFS(dvp)->n_flag |= NMODIFIED;
1330 if (!wccflag)
1331 VTONFS(dvp)->n_attrstamp = 0;
1332 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1333 return (error);
1334 }
1335
1336 /*
1337 * nfs mknod vop
1338 * just call nfs_mknodrpc() to do the work.
1339 */
1340 /* ARGSUSED */
1341 static int
1342 nfs_mknod(struct vop_mknod_args *ap)
1343 {
1344 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1345 }
1346
1347 static u_long create_verf;
1348 /*
1349 * nfs file create call
1350 */
1351 static int
1352 nfs_create(struct vop_create_args *ap)
1353 {
1354 struct vnode *dvp = ap->a_dvp;
1355 struct vattr *vap = ap->a_vap;
1356 struct componentname *cnp = ap->a_cnp;
1357 struct nfsv2_sattr *sp;
1358 u_int32_t *tl;
1359 struct nfsnode *np = NULL;
1360 struct vnode *newvp = NULL;
1361 caddr_t bpos, dpos;
1362 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1363 struct mbuf *mreq, *mrep, *md, *mb;
1364 struct vattr vattr;
1365 int v3 = NFS_ISV3(dvp);
1366
1367 /*
1368 * Oops, not for me..
1369 */
1370 if (vap->va_type == VSOCK)
1371 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1372
1373 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1374 return (error);
1375 }
1376 if (vap->va_vaflags & VA_EXCLUSIVE)
1377 fmode |= O_EXCL;
1378 again:
1379 nfsstats.rpccnt[NFSPROC_CREATE]++;
1380 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1381 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1382 mb = mreq;
1383 bpos = mtod(mb, caddr_t);
1384 nfsm_fhtom(dvp, v3);
1385 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1386 if (v3) {
1387 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1388 if (fmode & O_EXCL) {
1389 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1390 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1391 #ifdef INET
1392 if (!TAILQ_EMPTY(&in_ifaddrhead))
1393 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1394 else
1395 #endif
1396 *tl++ = create_verf;
1397 *tl = ++create_verf;
1398 } else {
1399 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1400 nfsm_v3attrbuild(vap, FALSE);
1401 }
1402 } else {
1403 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1404 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1405 sp->sa_uid = nfs_xdrneg1;
1406 sp->sa_gid = nfs_xdrneg1;
1407 sp->sa_size = 0;
1408 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1409 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1410 }
1411 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1412 if (!error) {
1413 nfsm_mtofh(dvp, newvp, v3, gotvp);
1414 if (!gotvp) {
1415 if (newvp) {
1416 vput(newvp);
1417 newvp = NULL;
1418 }
1419 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1420 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1421 if (!error)
1422 newvp = NFSTOV(np);
1423 }
1424 }
1425 if (v3)
1426 nfsm_wcc_data(dvp, wccflag);
1427 m_freem(mrep);
1428 nfsmout:
1429 if (error) {
1430 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1431 fmode &= ~O_EXCL;
1432 goto again;
1433 }
1434 if (newvp)
1435 vput(newvp);
1436 } else if (v3 && (fmode & O_EXCL)) {
1437 /*
1438 * We are normally called with only a partially initialized
1439 * VAP. Since the NFSv3 spec says that server may use the
1440 * file attributes to store the verifier, the spec requires
1441 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1442 * in atime, but we can't really assume that all servers will
1443 * so we ensure that our SETATTR sets both atime and mtime.
1444 */
1445 if (vap->va_mtime.tv_sec == VNOVAL)
1446 vfs_timestamp(&vap->va_mtime);
1447 if (vap->va_atime.tv_sec == VNOVAL)
1448 vap->va_atime = vap->va_mtime;
1449 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
1450 if (error)
1451 vput(newvp);
1452 }
1453 if (!error) {
1454 if (cnp->cn_flags & MAKEENTRY)
1455 cache_enter(dvp, newvp, cnp);
1456 *ap->a_vpp = newvp;
1457 }
1458 mtx_lock(&(VTONFS(dvp))->n_mtx);
1459 VTONFS(dvp)->n_flag |= NMODIFIED;
1460 if (!wccflag)
1461 VTONFS(dvp)->n_attrstamp = 0;
1462 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1463 return (error);
1464 }
1465
1466 /*
1467 * nfs file remove call
1468 * To try and make nfs semantics closer to ufs semantics, a file that has
1469 * other processes using the vnode is renamed instead of removed and then
1470 * removed later on the last close.
1471 * - If v_usecount > 1
1472 * If a rename is not already in the works
1473 * call nfs_sillyrename() to set it up
1474 * else
1475 * do the remove rpc
1476 */
1477 static int
1478 nfs_remove(struct vop_remove_args *ap)
1479 {
1480 struct vnode *vp = ap->a_vp;
1481 struct vnode *dvp = ap->a_dvp;
1482 struct componentname *cnp = ap->a_cnp;
1483 struct nfsnode *np = VTONFS(vp);
1484 int error = 0;
1485 struct vattr vattr;
1486
1487 #ifndef DIAGNOSTIC
1488 if ((cnp->cn_flags & HASBUF) == 0)
1489 panic("nfs_remove: no name");
1490 if (vrefcnt(vp) < 1)
1491 panic("nfs_remove: bad v_usecount");
1492 #endif
1493 if (vp->v_type == VDIR)
1494 error = EPERM;
1495 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1496 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
1497 vattr.va_nlink > 1)) {
1498 /*
1499 * Purge the name cache so that the chance of a lookup for
1500 * the name succeeding while the remove is in progress is
1501 * minimized. Without node locking it can still happen, such
1502 * that an I/O op returns ESTALE, but since you get this if
1503 * another host removes the file..
1504 */
1505 cache_purge(vp);
1506 /*
1507 * throw away biocache buffers, mainly to avoid
1508 * unnecessary delayed writes later.
1509 */
1510 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1511 /* Do the rpc */
1512 if (error != EINTR && error != EIO)
1513 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1514 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1515 /*
1516 * Kludge City: If the first reply to the remove rpc is lost..
1517 * the reply to the retransmitted request will be ENOENT
1518 * since the file was in fact removed
1519 * Therefore, we cheat and return success.
1520 */
1521 if (error == ENOENT)
1522 error = 0;
1523 } else if (!np->n_sillyrename)
1524 error = nfs_sillyrename(dvp, vp, cnp);
1525 np->n_attrstamp = 0;
1526 return (error);
1527 }
1528
1529 /*
1530 * nfs file remove rpc called from nfs_inactive
1531 */
1532 int
1533 nfs_removeit(struct sillyrename *sp)
1534 {
1535 /*
1536 * Make sure that the directory vnode is still valid.
1537 * XXX we should lock sp->s_dvp here.
1538 */
1539 if (sp->s_dvp->v_type == VBAD)
1540 return (0);
1541 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1542 NULL));
1543 }
1544
1545 /*
1546 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1547 */
1548 static int
1549 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1550 struct ucred *cred, struct thread *td)
1551 {
1552 caddr_t bpos, dpos;
1553 int error = 0, wccflag = NFSV3_WCCRATTR;
1554 struct mbuf *mreq, *mrep, *md, *mb;
1555 int v3 = NFS_ISV3(dvp);
1556
1557 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1558 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1559 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1560 mb = mreq;
1561 bpos = mtod(mb, caddr_t);
1562 nfsm_fhtom(dvp, v3);
1563 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1564 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1565 if (v3)
1566 nfsm_wcc_data(dvp, wccflag);
1567 m_freem(mrep);
1568 nfsmout:
1569 mtx_lock(&(VTONFS(dvp))->n_mtx);
1570 VTONFS(dvp)->n_flag |= NMODIFIED;
1571 if (!wccflag)
1572 VTONFS(dvp)->n_attrstamp = 0;
1573 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1574 return (error);
1575 }
1576
1577 /*
1578 * nfs file rename call
1579 */
1580 static int
1581 nfs_rename(struct vop_rename_args *ap)
1582 {
1583 struct vnode *fvp = ap->a_fvp;
1584 struct vnode *tvp = ap->a_tvp;
1585 struct vnode *fdvp = ap->a_fdvp;
1586 struct vnode *tdvp = ap->a_tdvp;
1587 struct componentname *tcnp = ap->a_tcnp;
1588 struct componentname *fcnp = ap->a_fcnp;
1589 int error;
1590
1591 #ifndef DIAGNOSTIC
1592 if ((tcnp->cn_flags & HASBUF) == 0 ||
1593 (fcnp->cn_flags & HASBUF) == 0)
1594 panic("nfs_rename: no name");
1595 #endif
1596 /* Check for cross-device rename */
1597 if ((fvp->v_mount != tdvp->v_mount) ||
1598 (tvp && (fvp->v_mount != tvp->v_mount))) {
1599 error = EXDEV;
1600 goto out;
1601 }
1602
1603 if (fvp == tvp) {
1604 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1605 error = 0;
1606 goto out;
1607 }
1608 if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0)
1609 goto out;
1610
1611 /*
1612 * We have to flush B_DELWRI data prior to renaming
1613 * the file. If we don't, the delayed-write buffers
1614 * can be flushed out later after the file has gone stale
1615 * under NFSV3. NFSV2 does not have this problem because
1616 * ( as far as I can tell ) it flushes dirty buffers more
1617 * often.
1618 *
1619 * Skip the rename operation if the fsync fails, this can happen
1620 * due to the server's volume being full, when we pushed out data
1621 * that was written back to our cache earlier. Not checking for
1622 * this condition can result in potential (silent) data loss.
1623 */
1624 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1625 VOP_UNLOCK(fvp, 0, fcnp->cn_thread);
1626 if (!error && tvp)
1627 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1628 if (error)
1629 goto out;
1630
1631 /*
1632 * If the tvp exists and is in use, sillyrename it before doing the
1633 * rename of the new file over it.
1634 * XXX Can't sillyrename a directory.
1635 */
1636 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1637 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1638 vput(tvp);
1639 tvp = NULL;
1640 }
1641
1642 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1643 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1644 tcnp->cn_thread);
1645
1646 if (fvp->v_type == VDIR) {
1647 if (tvp != NULL && tvp->v_type == VDIR)
1648 cache_purge(tdvp);
1649 cache_purge(fdvp);
1650 }
1651
1652 out:
1653 if (tdvp == tvp)
1654 vrele(tdvp);
1655 else
1656 vput(tdvp);
1657 if (tvp)
1658 vput(tvp);
1659 vrele(fdvp);
1660 vrele(fvp);
1661 /*
1662 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1663 */
1664 if (error == ENOENT)
1665 error = 0;
1666 return (error);
1667 }
1668
1669 /*
1670 * nfs file rename rpc called from nfs_remove() above
1671 */
1672 static int
1673 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1674 struct sillyrename *sp)
1675 {
1676
1677 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1678 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1679 }
1680
1681 /*
1682 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1683 */
1684 static int
1685 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1686 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1687 struct thread *td)
1688 {
1689 caddr_t bpos, dpos;
1690 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1691 struct mbuf *mreq, *mrep, *md, *mb;
1692 int v3 = NFS_ISV3(fdvp);
1693
1694 nfsstats.rpccnt[NFSPROC_RENAME]++;
1695 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1696 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1697 nfsm_rndup(tnamelen));
1698 mb = mreq;
1699 bpos = mtod(mb, caddr_t);
1700 nfsm_fhtom(fdvp, v3);
1701 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1702 nfsm_fhtom(tdvp, v3);
1703 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1704 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1705 if (v3) {
1706 nfsm_wcc_data(fdvp, fwccflag);
1707 nfsm_wcc_data(tdvp, twccflag);
1708 }
1709 m_freem(mrep);
1710 nfsmout:
1711 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1712 VTONFS(fdvp)->n_flag |= NMODIFIED;
1713 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1714 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1715 VTONFS(tdvp)->n_flag |= NMODIFIED;
1716 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1717 if (!fwccflag)
1718 VTONFS(fdvp)->n_attrstamp = 0;
1719 if (!twccflag)
1720 VTONFS(tdvp)->n_attrstamp = 0;
1721 return (error);
1722 }
1723
1724 /*
1725 * nfs hard link create call
1726 */
1727 static int
1728 nfs_link(struct vop_link_args *ap)
1729 {
1730 struct vnode *vp = ap->a_vp;
1731 struct vnode *tdvp = ap->a_tdvp;
1732 struct componentname *cnp = ap->a_cnp;
1733 caddr_t bpos, dpos;
1734 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1735 struct mbuf *mreq, *mrep, *md, *mb;
1736 int v3;
1737
1738 if (vp->v_mount != tdvp->v_mount) {
1739 return (EXDEV);
1740 }
1741
1742 /*
1743 * Push all writes to the server, so that the attribute cache
1744 * doesn't get "out of sync" with the server.
1745 * XXX There should be a better way!
1746 */
1747 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1748
1749 v3 = NFS_ISV3(vp);
1750 nfsstats.rpccnt[NFSPROC_LINK]++;
1751 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1752 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1753 mb = mreq;
1754 bpos = mtod(mb, caddr_t);
1755 nfsm_fhtom(vp, v3);
1756 nfsm_fhtom(tdvp, v3);
1757 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1758 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1759 if (v3) {
1760 nfsm_postop_attr(vp, attrflag);
1761 nfsm_wcc_data(tdvp, wccflag);
1762 }
1763 m_freem(mrep);
1764 nfsmout:
1765 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1766 VTONFS(tdvp)->n_flag |= NMODIFIED;
1767 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1768 if (!attrflag)
1769 VTONFS(vp)->n_attrstamp = 0;
1770 if (!wccflag)
1771 VTONFS(tdvp)->n_attrstamp = 0;
1772 return (error);
1773 }
1774
1775 /*
1776 * nfs symbolic link create call
1777 */
1778 static int
1779 nfs_symlink(struct vop_symlink_args *ap)
1780 {
1781 struct vnode *dvp = ap->a_dvp;
1782 struct vattr *vap = ap->a_vap;
1783 struct componentname *cnp = ap->a_cnp;
1784 struct nfsv2_sattr *sp;
1785 caddr_t bpos, dpos;
1786 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1787 struct mbuf *mreq, *mrep, *md, *mb;
1788 struct vnode *newvp = NULL;
1789 int v3 = NFS_ISV3(dvp);
1790
1791 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1792 slen = strlen(ap->a_target);
1793 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1794 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1795 mb = mreq;
1796 bpos = mtod(mb, caddr_t);
1797 nfsm_fhtom(dvp, v3);
1798 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1799 if (v3) {
1800 nfsm_v3attrbuild(vap, FALSE);
1801 }
1802 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1803 if (!v3) {
1804 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1805 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1806 sp->sa_uid = nfs_xdrneg1;
1807 sp->sa_gid = nfs_xdrneg1;
1808 sp->sa_size = nfs_xdrneg1;
1809 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1810 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1811 }
1812
1813 /*
1814 * Issue the NFS request and get the rpc response.
1815 *
1816 * Only NFSv3 responses returning an error of 0 actually return
1817 * a file handle that can be converted into newvp without having
1818 * to do an extra lookup rpc.
1819 */
1820 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
1821 if (v3) {
1822 if (error == 0)
1823 nfsm_mtofh(dvp, newvp, v3, gotvp);
1824 nfsm_wcc_data(dvp, wccflag);
1825 }
1826
1827 /*
1828 * out code jumps -> here, mrep is also freed.
1829 */
1830
1831 m_freem(mrep);
1832 nfsmout:
1833
1834 /*
1835 * If we do not have an error and we could not extract the newvp from
1836 * the response due to the request being NFSv2, we have to do a
1837 * lookup in order to obtain a newvp to return.
1838 */
1839 if (error == 0 && newvp == NULL) {
1840 struct nfsnode *np = NULL;
1841
1842 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1843 cnp->cn_cred, cnp->cn_thread, &np);
1844 if (!error)
1845 newvp = NFSTOV(np);
1846 }
1847 if (error) {
1848 if (newvp)
1849 vput(newvp);
1850 } else {
1851 *ap->a_vpp = newvp;
1852 }
1853 mtx_lock(&(VTONFS(dvp))->n_mtx);
1854 VTONFS(dvp)->n_flag |= NMODIFIED;
1855 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1856 if (!wccflag)
1857 VTONFS(dvp)->n_attrstamp = 0;
1858 return (error);
1859 }
1860
1861 /*
1862 * nfs make dir call
1863 */
1864 static int
1865 nfs_mkdir(struct vop_mkdir_args *ap)
1866 {
1867 struct vnode *dvp = ap->a_dvp;
1868 struct vattr *vap = ap->a_vap;
1869 struct componentname *cnp = ap->a_cnp;
1870 struct nfsv2_sattr *sp;
1871 int len;
1872 struct nfsnode *np = NULL;
1873 struct vnode *newvp = NULL;
1874 caddr_t bpos, dpos;
1875 int error = 0, wccflag = NFSV3_WCCRATTR;
1876 int gotvp = 0;
1877 struct mbuf *mreq, *mrep, *md, *mb;
1878 struct vattr vattr;
1879 int v3 = NFS_ISV3(dvp);
1880
1881 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1882 return (error);
1883 }
1884 len = cnp->cn_namelen;
1885 nfsstats.rpccnt[NFSPROC_MKDIR]++;
1886 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
1887 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1888 mb = mreq;
1889 bpos = mtod(mb, caddr_t);
1890 nfsm_fhtom(dvp, v3);
1891 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1892 if (v3) {
1893 nfsm_v3attrbuild(vap, FALSE);
1894 } else {
1895 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1896 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1897 sp->sa_uid = nfs_xdrneg1;
1898 sp->sa_gid = nfs_xdrneg1;
1899 sp->sa_size = nfs_xdrneg1;
1900 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1901 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1902 }
1903 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
1904 if (!error)
1905 nfsm_mtofh(dvp, newvp, v3, gotvp);
1906 if (v3)
1907 nfsm_wcc_data(dvp, wccflag);
1908 m_freem(mrep);
1909 nfsmout:
1910 mtx_lock(&(VTONFS(dvp))->n_mtx);
1911 VTONFS(dvp)->n_flag |= NMODIFIED;
1912 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1913 if (!wccflag)
1914 VTONFS(dvp)->n_attrstamp = 0;
1915 if (error == 0 && newvp == NULL) {
1916 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1917 cnp->cn_thread, &np);
1918 if (!error) {
1919 newvp = NFSTOV(np);
1920 if (newvp->v_type != VDIR)
1921 error = EEXIST;
1922 }
1923 }
1924 if (error) {
1925 if (newvp)
1926 vput(newvp);
1927 } else
1928 *ap->a_vpp = newvp;
1929 return (error);
1930 }
1931
1932 /*
1933 * nfs remove directory call
1934 */
1935 static int
1936 nfs_rmdir(struct vop_rmdir_args *ap)
1937 {
1938 struct vnode *vp = ap->a_vp;
1939 struct vnode *dvp = ap->a_dvp;
1940 struct componentname *cnp = ap->a_cnp;
1941 caddr_t bpos, dpos;
1942 int error = 0, wccflag = NFSV3_WCCRATTR;
1943 struct mbuf *mreq, *mrep, *md, *mb;
1944 int v3 = NFS_ISV3(dvp);
1945
1946 if (dvp == vp)
1947 return (EINVAL);
1948 nfsstats.rpccnt[NFSPROC_RMDIR]++;
1949 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
1950 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1951 mb = mreq;
1952 bpos = mtod(mb, caddr_t);
1953 nfsm_fhtom(dvp, v3);
1954 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1955 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
1956 if (v3)
1957 nfsm_wcc_data(dvp, wccflag);
1958 m_freem(mrep);
1959 nfsmout:
1960 mtx_lock(&(VTONFS(dvp))->n_mtx);
1961 VTONFS(dvp)->n_flag |= NMODIFIED;
1962 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1963 if (!wccflag)
1964 VTONFS(dvp)->n_attrstamp = 0;
1965 cache_purge(dvp);
1966 cache_purge(vp);
1967 /*
1968 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
1969 */
1970 if (error == ENOENT)
1971 error = 0;
1972 return (error);
1973 }
1974
1975 /*
1976 * nfs readdir call
1977 */
1978 static int
1979 nfs_readdir(struct vop_readdir_args *ap)
1980 {
1981 struct vnode *vp = ap->a_vp;
1982 struct nfsnode *np = VTONFS(vp);
1983 struct uio *uio = ap->a_uio;
1984 int tresid, error = 0;
1985 struct vattr vattr;
1986
1987 if (vp->v_type != VDIR)
1988 return(EPERM);
1989
1990 /*
1991 * First, check for hit on the EOF offset cache
1992 */
1993 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
1994 (np->n_flag & NMODIFIED) == 0) {
1995 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0) {
1996 mtx_lock(&np->n_mtx);
1997 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
1998 mtx_unlock(&np->n_mtx);
1999 nfsstats.direofcache_hits++;
2000 goto out;
2001 } else
2002 mtx_unlock(&np->n_mtx);
2003 }
2004 }
2005
2006 /*
2007 * Call nfs_bioread() to do the real work.
2008 */
2009 tresid = uio->uio_resid;
2010 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2011
2012 if (!error && uio->uio_resid == tresid) {
2013 nfsstats.direofcache_misses++;
2014 }
2015 out:
2016 return (error);
2017 }
2018
2019 /*
2020 * Readdir rpc call.
2021 * Called from below the buffer cache by nfs_doio().
2022 */
2023 int
2024 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2025 {
2026 int len, left;
2027 struct dirent *dp = NULL;
2028 u_int32_t *tl;
2029 caddr_t cp;
2030 nfsuint64 *cookiep;
2031 caddr_t bpos, dpos;
2032 struct mbuf *mreq, *mrep, *md, *mb;
2033 nfsuint64 cookie;
2034 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2035 struct nfsnode *dnp = VTONFS(vp);
2036 u_quad_t fileno;
2037 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2038 int attrflag;
2039 int v3 = NFS_ISV3(vp);
2040
2041 #ifndef DIAGNOSTIC
2042 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2043 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2044 panic("nfs readdirrpc bad uio");
2045 #endif
2046
2047 /*
2048 * If there is no cookie, assume directory was stale.
2049 */
2050 nfs_dircookie_lock(dnp);
2051 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2052 if (cookiep) {
2053 cookie = *cookiep;
2054 nfs_dircookie_unlock(dnp);
2055 } else {
2056 nfs_dircookie_unlock(dnp);
2057 return (NFSERR_BAD_COOKIE);
2058 }
2059
2060 /*
2061 * Loop around doing readdir rpc's of size nm_readdirsize
2062 * truncated to a multiple of DIRBLKSIZ.
2063 * The stopping criteria is EOF or buffer full.
2064 */
2065 while (more_dirs && bigenough) {
2066 nfsstats.rpccnt[NFSPROC_READDIR]++;
2067 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2068 NFSX_READDIR(v3));
2069 mb = mreq;
2070 bpos = mtod(mb, caddr_t);
2071 nfsm_fhtom(vp, v3);
2072 if (v3) {
2073 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2074 *tl++ = cookie.nfsuquad[0];
2075 *tl++ = cookie.nfsuquad[1];
2076 mtx_lock(&dnp->n_mtx);
2077 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2078 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2079 mtx_unlock(&dnp->n_mtx);
2080 } else {
2081 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2082 *tl++ = cookie.nfsuquad[0];
2083 }
2084 *tl = txdr_unsigned(nmp->nm_readdirsize);
2085 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2086 if (v3) {
2087 nfsm_postop_attr(vp, attrflag);
2088 if (!error) {
2089 tl = nfsm_dissect(u_int32_t *,
2090 2 * NFSX_UNSIGNED);
2091 mtx_lock(&dnp->n_mtx);
2092 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2093 dnp->n_cookieverf.nfsuquad[1] = *tl;
2094 mtx_unlock(&dnp->n_mtx);
2095 } else {
2096 m_freem(mrep);
2097 goto nfsmout;
2098 }
2099 }
2100 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2101 more_dirs = fxdr_unsigned(int, *tl);
2102
2103 /* loop thru the dir entries, doctoring them to 4bsd form */
2104 while (more_dirs && bigenough) {
2105 if (v3) {
2106 tl = nfsm_dissect(u_int32_t *,
2107 3 * NFSX_UNSIGNED);
2108 fileno = fxdr_hyper(tl);
2109 len = fxdr_unsigned(int, *(tl + 2));
2110 } else {
2111 tl = nfsm_dissect(u_int32_t *,
2112 2 * NFSX_UNSIGNED);
2113 fileno = fxdr_unsigned(u_quad_t, *tl++);
2114 len = fxdr_unsigned(int, *tl);
2115 }
2116 if (len <= 0 || len > NFS_MAXNAMLEN) {
2117 error = EBADRPC;
2118 m_freem(mrep);
2119 goto nfsmout;
2120 }
2121 tlen = nfsm_rndup(len);
2122 if (tlen == len)
2123 tlen += 4; /* To ensure null termination */
2124 left = DIRBLKSIZ - blksiz;
2125 if ((tlen + DIRHDSIZ) > left) {
2126 dp->d_reclen += left;
2127 uiop->uio_iov->iov_base =
2128 (char *)uiop->uio_iov->iov_base + left;
2129 uiop->uio_iov->iov_len -= left;
2130 uiop->uio_offset += left;
2131 uiop->uio_resid -= left;
2132 blksiz = 0;
2133 }
2134 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2135 bigenough = 0;
2136 if (bigenough) {
2137 dp = (struct dirent *)uiop->uio_iov->iov_base;
2138 dp->d_fileno = (int)fileno;
2139 dp->d_namlen = len;
2140 dp->d_reclen = tlen + DIRHDSIZ;
2141 dp->d_type = DT_UNKNOWN;
2142 blksiz += dp->d_reclen;
2143 if (blksiz == DIRBLKSIZ)
2144 blksiz = 0;
2145 uiop->uio_offset += DIRHDSIZ;
2146 uiop->uio_resid -= DIRHDSIZ;
2147 uiop->uio_iov->iov_base =
2148 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2149 uiop->uio_iov->iov_len -= DIRHDSIZ;
2150 nfsm_mtouio(uiop, len);
2151 cp = uiop->uio_iov->iov_base;
2152 tlen -= len;
2153 *cp = '\0'; /* null terminate */
2154 uiop->uio_iov->iov_base =
2155 (char *)uiop->uio_iov->iov_base + tlen;
2156 uiop->uio_iov->iov_len -= tlen;
2157 uiop->uio_offset += tlen;
2158 uiop->uio_resid -= tlen;
2159 } else
2160 nfsm_adv(nfsm_rndup(len));
2161 if (v3) {
2162 tl = nfsm_dissect(u_int32_t *,
2163 3 * NFSX_UNSIGNED);
2164 } else {
2165 tl = nfsm_dissect(u_int32_t *,
2166 2 * NFSX_UNSIGNED);
2167 }
2168 if (bigenough) {
2169 cookie.nfsuquad[0] = *tl++;
2170 if (v3)
2171 cookie.nfsuquad[1] = *tl++;
2172 } else if (v3)
2173 tl += 2;
2174 else
2175 tl++;
2176 more_dirs = fxdr_unsigned(int, *tl);
2177 }
2178 /*
2179 * If at end of rpc data, get the eof boolean
2180 */
2181 if (!more_dirs) {
2182 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2183 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2184 }
2185 m_freem(mrep);
2186 }
2187 /*
2188 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2189 * by increasing d_reclen for the last record.
2190 */
2191 if (blksiz > 0) {
2192 left = DIRBLKSIZ - blksiz;
2193 dp->d_reclen += left;
2194 uiop->uio_iov->iov_base =
2195 (char *)uiop->uio_iov->iov_base + left;
2196 uiop->uio_iov->iov_len -= left;
2197 uiop->uio_offset += left;
2198 uiop->uio_resid -= left;
2199 }
2200
2201 /*
2202 * We are now either at the end of the directory or have filled the
2203 * block.
2204 */
2205 if (bigenough)
2206 dnp->n_direofoffset = uiop->uio_offset;
2207 else {
2208 if (uiop->uio_resid > 0)
2209 nfs_printf("EEK! readdirrpc resid > 0\n");
2210 nfs_dircookie_lock(dnp);
2211 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2212 *cookiep = cookie;
2213 nfs_dircookie_unlock(dnp);
2214 }
2215 nfsmout:
2216 return (error);
2217 }
2218
2219 /*
2220 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2221 */
2222 int
2223 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2224 {
2225 int len, left;
2226 struct dirent *dp;
2227 u_int32_t *tl;
2228 caddr_t cp;
2229 struct vnode *newvp;
2230 nfsuint64 *cookiep;
2231 caddr_t bpos, dpos, dpossav1, dpossav2;
2232 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2233 struct nameidata nami, *ndp = &nami;
2234 struct componentname *cnp = &ndp->ni_cnd;
2235 nfsuint64 cookie;
2236 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2237 struct nfsnode *dnp = VTONFS(vp), *np;
2238 nfsfh_t *fhp;
2239 u_quad_t fileno;
2240 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2241 int attrflag, fhsize;
2242
2243 #ifndef nolint
2244 dp = NULL;
2245 #endif
2246 #ifndef DIAGNOSTIC
2247 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2248 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2249 panic("nfs readdirplusrpc bad uio");
2250 #endif
2251 ndp->ni_dvp = vp;
2252 newvp = NULLVP;
2253
2254 /*
2255 * If there is no cookie, assume directory was stale.
2256 */
2257 nfs_dircookie_lock(dnp);
2258 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2259 if (cookiep) {
2260 cookie = *cookiep;
2261 nfs_dircookie_unlock(dnp);
2262 } else {
2263 nfs_dircookie_unlock(dnp);
2264 return (NFSERR_BAD_COOKIE);
2265 }
2266 /*
2267 * Loop around doing readdir rpc's of size nm_readdirsize
2268 * truncated to a multiple of DIRBLKSIZ.
2269 * The stopping criteria is EOF or buffer full.
2270 */
2271 while (more_dirs && bigenough) {
2272 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2273 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2274 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2275 mb = mreq;
2276 bpos = mtod(mb, caddr_t);
2277 nfsm_fhtom(vp, 1);
2278 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2279 *tl++ = cookie.nfsuquad[0];
2280 *tl++ = cookie.nfsuquad[1];
2281 mtx_lock(&dnp->n_mtx);
2282 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2283 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2284 mtx_unlock(&dnp->n_mtx);
2285 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2286 *tl = txdr_unsigned(nmp->nm_rsize);
2287 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2288 nfsm_postop_attr(vp, attrflag);
2289 if (error) {
2290 m_freem(mrep);
2291 goto nfsmout;
2292 }
2293 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2294 mtx_lock(&dnp->n_mtx);
2295 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2296 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2297 mtx_unlock(&dnp->n_mtx);
2298 more_dirs = fxdr_unsigned(int, *tl);
2299
2300 /* loop thru the dir entries, doctoring them to 4bsd form */
2301 while (more_dirs && bigenough) {
2302 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2303 fileno = fxdr_hyper(tl);
2304 len = fxdr_unsigned(int, *(tl + 2));
2305 if (len <= 0 || len > NFS_MAXNAMLEN) {
2306 error = EBADRPC;
2307 m_freem(mrep);
2308 goto nfsmout;
2309 }
2310 tlen = nfsm_rndup(len);
2311 if (tlen == len)
2312 tlen += 4; /* To ensure null termination*/
2313 left = DIRBLKSIZ - blksiz;
2314 if ((tlen + DIRHDSIZ) > left) {
2315 dp->d_reclen += left;
2316 uiop->uio_iov->iov_base =
2317 (char *)uiop->uio_iov->iov_base + left;
2318 uiop->uio_iov->iov_len -= left;
2319 uiop->uio_offset += left;
2320 uiop->uio_resid -= left;
2321 blksiz = 0;
2322 }
2323 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2324 bigenough = 0;
2325 if (bigenough) {
2326 dp = (struct dirent *)uiop->uio_iov->iov_base;
2327 dp->d_fileno = (int)fileno;
2328 dp->d_namlen = len;
2329 dp->d_reclen = tlen + DIRHDSIZ;
2330 dp->d_type = DT_UNKNOWN;
2331 blksiz += dp->d_reclen;
2332 if (blksiz == DIRBLKSIZ)
2333 blksiz = 0;
2334 uiop->uio_offset += DIRHDSIZ;
2335 uiop->uio_resid -= DIRHDSIZ;
2336 uiop->uio_iov->iov_base =
2337 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2338 uiop->uio_iov->iov_len -= DIRHDSIZ;
2339 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2340 cnp->cn_namelen = len;
2341 nfsm_mtouio(uiop, len);
2342 cp = uiop->uio_iov->iov_base;
2343 tlen -= len;
2344 *cp = '\0';
2345 uiop->uio_iov->iov_base =
2346 (char *)uiop->uio_iov->iov_base + tlen;
2347 uiop->uio_iov->iov_len -= tlen;
2348 uiop->uio_offset += tlen;
2349 uiop->uio_resid -= tlen;
2350 } else
2351 nfsm_adv(nfsm_rndup(len));
2352 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2353 if (bigenough) {
2354 cookie.nfsuquad[0] = *tl++;
2355 cookie.nfsuquad[1] = *tl++;
2356 } else
2357 tl += 2;
2358
2359 /*
2360 * Since the attributes are before the file handle
2361 * (sigh), we must skip over the attributes and then
2362 * come back and get them.
2363 */
2364 attrflag = fxdr_unsigned(int, *tl);
2365 if (attrflag) {
2366 dpossav1 = dpos;
2367 mdsav1 = md;
2368 nfsm_adv(NFSX_V3FATTR);
2369 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2370 doit = fxdr_unsigned(int, *tl);
2371 /*
2372 * Skip loading the attrs for "..". There's a
2373 * race between loading the attrs here and
2374 * lookups that look for the directory currently
2375 * being read (in the parent). We try to acquire
2376 * the exclusive lock on ".." here, owning the
2377 * lock on the directory being read. Lookup will
2378 * hold the lock on ".." and try to acquire the
2379 * lock on the directory being read.
2380 *
2381 * There are other ways of fixing this, one would
2382 * be to do a trylock on the ".." vnode and skip
2383 * loading the attrs on ".." if it happens to be
2384 * locked by another process. But skipping the
2385 * attrload on ".." seems the easiest option.
2386 */
2387 if (strcmp(dp->d_name, "..") == 0) {
2388 doit = 0;
2389 /*
2390 * We've already skipped over the attrs,
2391 * skip over the filehandle. And store d_type
2392 * as VDIR.
2393 */
2394 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2395 i = fxdr_unsigned(int, *tl);
2396 nfsm_adv(nfsm_rndup(i));
2397 dp->d_type = IFTODT(VTTOIF(VDIR));
2398 }
2399 if (doit) {
2400 nfsm_getfh(fhp, fhsize, 1);
2401 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2402 VREF(vp);
2403 newvp = vp;
2404 np = dnp;
2405 } else {
2406 error = nfs_nget(vp->v_mount, fhp,
2407 fhsize, &np, LK_EXCLUSIVE);
2408 if (error)
2409 doit = 0;
2410 else
2411 newvp = NFSTOV(np);
2412 }
2413 }
2414 if (doit && bigenough) {
2415 dpossav2 = dpos;
2416 dpos = dpossav1;
2417 mdsav2 = md;
2418 md = mdsav1;
2419 nfsm_loadattr(newvp, NULL);
2420 dpos = dpossav2;
2421 md = mdsav2;
2422 dp->d_type =
2423 IFTODT(VTTOIF(np->n_vattr.va_type));
2424 ndp->ni_vp = newvp;
2425 /* Update n_ctime, so subsequent lookup doesn't purge entry */
2426 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
2427 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2428 }
2429 } else {
2430 /* Just skip over the file handle */
2431 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2432 i = fxdr_unsigned(int, *tl);
2433 if (i) {
2434 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2435 fhsize = fxdr_unsigned(int, *tl);
2436 nfsm_adv(nfsm_rndup(fhsize));
2437 }
2438 }
2439 if (newvp != NULLVP) {
2440 if (newvp == vp)
2441 vrele(newvp);
2442 else
2443 vput(newvp);
2444 newvp = NULLVP;
2445 }
2446 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2447 more_dirs = fxdr_unsigned(int, *tl);
2448 }
2449 /*
2450 * If at end of rpc data, get the eof boolean
2451 */
2452 if (!more_dirs) {
2453 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2454 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2455 }
2456 m_freem(mrep);
2457 }
2458 /*
2459 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2460 * by increasing d_reclen for the last record.
2461 */
2462 if (blksiz > 0) {
2463 left = DIRBLKSIZ - blksiz;
2464 dp->d_reclen += left;
2465 uiop->uio_iov->iov_base =
2466 (char *)uiop->uio_iov->iov_base + left;
2467 uiop->uio_iov->iov_len -= left;
2468 uiop->uio_offset += left;
2469 uiop->uio_resid -= left;
2470 }
2471
2472 /*
2473 * We are now either at the end of the directory or have filled the
2474 * block.
2475 */
2476 if (bigenough)
2477 dnp->n_direofoffset = uiop->uio_offset;
2478 else {
2479 if (uiop->uio_resid > 0)
2480 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2481 nfs_dircookie_lock(dnp);
2482 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2483 *cookiep = cookie;
2484 nfs_dircookie_unlock(dnp);
2485 }
2486 nfsmout:
2487 if (newvp != NULLVP) {
2488 if (newvp == vp)
2489 vrele(newvp);
2490 else
2491 vput(newvp);
2492 newvp = NULLVP;
2493 }
2494 return (error);
2495 }
2496
2497 /*
2498 * Silly rename. To make the NFS filesystem that is stateless look a little
2499 * more like the "ufs" a remove of an active vnode is translated to a rename
2500 * to a funny looking filename that is removed by nfs_inactive on the
2501 * nfsnode. There is the potential for another process on a different client
2502 * to create the same funny name between the nfs_lookitup() fails and the
2503 * nfs_rename() completes, but...
2504 */
2505 static int
2506 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2507 {
2508 struct sillyrename *sp;
2509 struct nfsnode *np;
2510 int error;
2511 short pid;
2512 unsigned int lticks;
2513
2514 cache_purge(dvp);
2515 np = VTONFS(vp);
2516 #ifndef DIAGNOSTIC
2517 if (vp->v_type == VDIR)
2518 panic("nfs: sillyrename dir");
2519 #endif
2520 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2521 M_NFSREQ, M_WAITOK);
2522 sp->s_cred = crhold(cnp->cn_cred);
2523 sp->s_dvp = dvp;
2524 sp->s_removeit = nfs_removeit;
2525 VREF(dvp);
2526
2527 /*
2528 * Fudge together a funny name.
2529 * Changing the format of the funny name to accomodate more
2530 * sillynames per directory.
2531 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2532 * CPU ticks since boot.
2533 */
2534 pid = cnp->cn_thread->td_proc->p_pid;
2535 lticks = (unsigned int)ticks;
2536 for ( ; ; ) {
2537 sp->s_namlen = sprintf(sp->s_name,
2538 ".nfs.%08x.%04x4.4", lticks,
2539 pid);
2540 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2541 cnp->cn_thread, NULL))
2542 break;
2543 lticks++;
2544 }
2545 error = nfs_renameit(dvp, cnp, sp);
2546 if (error)
2547 goto bad;
2548 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2549 cnp->cn_thread, &np);
2550 np->n_sillyrename = sp;
2551 return (0);
2552 bad:
2553 vrele(sp->s_dvp);
2554 crfree(sp->s_cred);
2555 free((caddr_t)sp, M_NFSREQ);
2556 return (error);
2557 }
2558
2559 /*
2560 * Look up a file name and optionally either update the file handle or
2561 * allocate an nfsnode, depending on the value of npp.
2562 * npp == NULL --> just do the lookup
2563 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2564 * handled too
2565 * *npp != NULL --> update the file handle in the vnode
2566 */
2567 static int
2568 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2569 struct thread *td, struct nfsnode **npp)
2570 {
2571 struct vnode *newvp = NULL;
2572 struct nfsnode *np, *dnp = VTONFS(dvp);
2573 caddr_t bpos, dpos;
2574 int error = 0, fhlen, attrflag;
2575 struct mbuf *mreq, *mrep, *md, *mb;
2576 nfsfh_t *nfhp;
2577 int v3 = NFS_ISV3(dvp);
2578
2579 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2580 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2581 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2582 mb = mreq;
2583 bpos = mtod(mb, caddr_t);
2584 nfsm_fhtom(dvp, v3);
2585 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2586 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2587 if (npp && !error) {
2588 nfsm_getfh(nfhp, fhlen, v3);
2589 if (*npp) {
2590 np = *npp;
2591 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2592 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2593 np->n_fhp = &np->n_fh;
2594 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2595 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2596 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2597 np->n_fhsize = fhlen;
2598 newvp = NFSTOV(np);
2599 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2600 VREF(dvp);
2601 newvp = dvp;
2602 } else {
2603 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2604 if (error) {
2605 m_freem(mrep);
2606 return (error);
2607 }
2608 newvp = NFSTOV(np);
2609 }
2610 if (v3) {
2611 nfsm_postop_attr(newvp, attrflag);
2612 if (!attrflag && *npp == NULL) {
2613 m_freem(mrep);
2614 if (newvp == dvp)
2615 vrele(newvp);
2616 else
2617 vput(newvp);
2618 return (ENOENT);
2619 }
2620 } else
2621 nfsm_loadattr(newvp, NULL);
2622 }
2623 m_freem(mrep);
2624 nfsmout:
2625 if (npp && *npp == NULL) {
2626 if (error) {
2627 if (newvp) {
2628 if (newvp == dvp)
2629 vrele(newvp);
2630 else
2631 vput(newvp);
2632 }
2633 } else
2634 *npp = np;
2635 }
2636 return (error);
2637 }
2638
2639 /*
2640 * Nfs Version 3 commit rpc
2641 */
2642 int
2643 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2644 struct thread *td)
2645 {
2646 u_int32_t *tl;
2647 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2648 caddr_t bpos, dpos;
2649 int error = 0, wccflag = NFSV3_WCCRATTR;
2650 struct mbuf *mreq, *mrep, *md, *mb;
2651
2652 mtx_lock(&nmp->nm_mtx);
2653 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2654 mtx_unlock(&nmp->nm_mtx);
2655 return (0);
2656 }
2657 mtx_unlock(&nmp->nm_mtx);
2658 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2659 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2660 mb = mreq;
2661 bpos = mtod(mb, caddr_t);
2662 nfsm_fhtom(vp, 1);
2663 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2664 txdr_hyper(offset, tl);
2665 tl += 2;
2666 *tl = txdr_unsigned(cnt);
2667 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2668 nfsm_wcc_data(vp, wccflag);
2669 if (!error) {
2670 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2671 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2672 NFSX_V3WRITEVERF)) {
2673 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2674 NFSX_V3WRITEVERF);
2675 error = NFSERR_STALEWRITEVERF;
2676 }
2677 }
2678 m_freem(mrep);
2679 nfsmout:
2680 return (error);
2681 }
2682
2683 /*
2684 * Strategy routine.
2685 * For async requests when nfsiod(s) are running, queue the request by
2686 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2687 * request.
2688 */
2689 static int
2690 nfs_strategy(struct vop_strategy_args *ap)
2691 {
2692 struct buf *bp = ap->a_bp;
2693 struct ucred *cr;
2694
2695 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2696 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
2697
2698 if (bp->b_iocmd == BIO_READ)
2699 cr = bp->b_rcred;
2700 else
2701 cr = bp->b_wcred;
2702
2703 /*
2704 * If the op is asynchronous and an i/o daemon is waiting
2705 * queue the request, wake it up and wait for completion
2706 * otherwise just do it ourselves.
2707 */
2708 if ((bp->b_flags & B_ASYNC) == 0 ||
2709 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2710 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2711 return (0);
2712 }
2713
2714 /*
2715 * fsync vnode op. Just call nfs_flush() with commit == 1.
2716 */
2717 /* ARGSUSED */
2718 static int
2719 nfs_fsync(struct vop_fsync_args *ap)
2720 {
2721 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2722 }
2723
2724 /*
2725 * Flush all the blocks associated with a vnode.
2726 * Walk through the buffer pool and push any dirty pages
2727 * associated with the vnode.
2728 */
2729 static int
2730 nfs_flush(struct vnode *vp, int waitfor, struct thread *td,
2731 int commit)
2732 {
2733 struct nfsnode *np = VTONFS(vp);
2734 struct buf *bp;
2735 int i;
2736 struct buf *nbp;
2737 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2738 int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2739 int passone = 1;
2740 u_quad_t off, endoff, toff;
2741 struct ucred* wcred = NULL;
2742 struct buf **bvec = NULL;
2743 #ifndef NFS_COMMITBVECSIZ
2744 #define NFS_COMMITBVECSIZ 20
2745 #endif
2746 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2747 int bvecsize = 0, bveccount;
2748
2749 if (nmp->nm_flag & NFSMNT_INT)
2750 slpflag = PCATCH;
2751 if (!commit)
2752 passone = 0;
2753 /*
2754 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2755 * server, but has not been committed to stable storage on the server
2756 * yet. On the first pass, the byte range is worked out and the commit
2757 * rpc is done. On the second pass, nfs_writebp() is called to do the
2758 * job.
2759 */
2760 again:
2761 off = (u_quad_t)-1;
2762 endoff = 0;
2763 bvecpos = 0;
2764 if (NFS_ISV3(vp) && commit) {
2765 s = splbio();
2766 if (bvec != NULL && bvec != bvec_on_stack)
2767 free(bvec, M_TEMP);
2768 /*
2769 * Count up how many buffers waiting for a commit.
2770 */
2771 bveccount = 0;
2772 VI_LOCK(vp);
2773 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2774 if (BUF_REFCNT(bp) == 0 &&
2775 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2776 == (B_DELWRI | B_NEEDCOMMIT))
2777 bveccount++;
2778 }
2779 /*
2780 * Allocate space to remember the list of bufs to commit. It is
2781 * important to use M_NOWAIT here to avoid a race with nfs_write.
2782 * If we can't get memory (for whatever reason), we will end up
2783 * committing the buffers one-by-one in the loop below.
2784 */
2785 if (bveccount > NFS_COMMITBVECSIZ) {
2786 /*
2787 * Release the vnode interlock to avoid a lock
2788 * order reversal.
2789 */
2790 VI_UNLOCK(vp);
2791 bvec = (struct buf **)
2792 malloc(bveccount * sizeof(struct buf *),
2793 M_TEMP, M_NOWAIT);
2794 VI_LOCK(vp);
2795 if (bvec == NULL) {
2796 bvec = bvec_on_stack;
2797 bvecsize = NFS_COMMITBVECSIZ;
2798 } else
2799 bvecsize = bveccount;
2800 } else {
2801 bvec = bvec_on_stack;
2802 bvecsize = NFS_COMMITBVECSIZ;
2803 }
2804 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2805 if (bvecpos >= bvecsize)
2806 break;
2807 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2808 nbp = TAILQ_NEXT(bp, b_bobufs);
2809 continue;
2810 }
2811 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2812 (B_DELWRI | B_NEEDCOMMIT)) {
2813 BUF_UNLOCK(bp);
2814 nbp = TAILQ_NEXT(bp, b_bobufs);
2815 continue;
2816 }
2817 VI_UNLOCK(vp);
2818 bremfree(bp);
2819 /*
2820 * Work out if all buffers are using the same cred
2821 * so we can deal with them all with one commit.
2822 *
2823 * NOTE: we are not clearing B_DONE here, so we have
2824 * to do it later on in this routine if we intend to
2825 * initiate I/O on the bp.
2826 *
2827 * Note: to avoid loopback deadlocks, we do not
2828 * assign b_runningbufspace.
2829 */
2830 if (wcred == NULL)
2831 wcred = bp->b_wcred;
2832 else if (wcred != bp->b_wcred)
2833 wcred = NOCRED;
2834 vfs_busy_pages(bp, 1);
2835
2836 VI_LOCK(vp);
2837 /*
2838 * bp is protected by being locked, but nbp is not
2839 * and vfs_busy_pages() may sleep. We have to
2840 * recalculate nbp.
2841 */
2842 nbp = TAILQ_NEXT(bp, b_bobufs);
2843
2844 /*
2845 * A list of these buffers is kept so that the
2846 * second loop knows which buffers have actually
2847 * been committed. This is necessary, since there
2848 * may be a race between the commit rpc and new
2849 * uncommitted writes on the file.
2850 */
2851 bvec[bvecpos++] = bp;
2852 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2853 bp->b_dirtyoff;
2854 if (toff < off)
2855 off = toff;
2856 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2857 if (toff > endoff)
2858 endoff = toff;
2859 }
2860 splx(s);
2861 VI_UNLOCK(vp);
2862 }
2863 if (bvecpos > 0) {
2864 /*
2865 * Commit data on the server, as required.
2866 * If all bufs are using the same wcred, then use that with
2867 * one call for all of them, otherwise commit each one
2868 * separately.
2869 */
2870 if (wcred != NOCRED)
2871 retv = nfs_commit(vp, off, (int)(endoff - off),
2872 wcred, td);
2873 else {
2874 retv = 0;
2875 for (i = 0; i < bvecpos; i++) {
2876 off_t off, size;
2877 bp = bvec[i];
2878 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2879 bp->b_dirtyoff;
2880 size = (u_quad_t)(bp->b_dirtyend
2881 - bp->b_dirtyoff);
2882 retv = nfs_commit(vp, off, (int)size,
2883 bp->b_wcred, td);
2884 if (retv) break;
2885 }
2886 }
2887
2888 if (retv == NFSERR_STALEWRITEVERF)
2889 nfs_clearcommit(vp->v_mount);
2890
2891 /*
2892 * Now, either mark the blocks I/O done or mark the
2893 * blocks dirty, depending on whether the commit
2894 * succeeded.
2895 */
2896 for (i = 0; i < bvecpos; i++) {
2897 bp = bvec[i];
2898 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2899 if (retv) {
2900 /*
2901 * Error, leave B_DELWRI intact
2902 */
2903 vfs_unbusy_pages(bp);
2904 brelse(bp);
2905 } else {
2906 /*
2907 * Success, remove B_DELWRI ( bundirty() ).
2908 *
2909 * b_dirtyoff/b_dirtyend seem to be NFS
2910 * specific. We should probably move that
2911 * into bundirty(). XXX
2912 */
2913 s = splbio();
2914 bufobj_wref(&vp->v_bufobj);
2915 bp->b_flags |= B_ASYNC;
2916 bundirty(bp);
2917 bp->b_flags &= ~B_DONE;
2918 bp->b_ioflags &= ~BIO_ERROR;
2919 bp->b_dirtyoff = bp->b_dirtyend = 0;
2920 splx(s);
2921 bufdone(bp);
2922 }
2923 }
2924 }
2925
2926 /*
2927 * Start/do any write(s) that are required.
2928 */
2929 loop:
2930 s = splbio();
2931 VI_LOCK(vp);
2932 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2933 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2934 if (waitfor != MNT_WAIT || passone)
2935 continue;
2936
2937 error = BUF_TIMELOCK(bp,
2938 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2939 VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
2940 splx(s);
2941 if (error == 0) {
2942 BUF_UNLOCK(bp);
2943 goto loop;
2944 }
2945 if (error == ENOLCK)
2946 goto loop;
2947 if (nfs_sigintr(nmp, NULL, td)) {
2948 error = EINTR;
2949 goto done;
2950 }
2951 if (slpflag == PCATCH) {
2952 slpflag = 0;
2953 slptimeo = 2 * hz;
2954 }
2955 goto loop;
2956 }
2957 if ((bp->b_flags & B_DELWRI) == 0)
2958 panic("nfs_fsync: not dirty");
2959 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2960 BUF_UNLOCK(bp);
2961 continue;
2962 }
2963 VI_UNLOCK(vp);
2964 bremfree(bp);
2965 if (passone || !commit)
2966 bp->b_flags |= B_ASYNC;
2967 else
2968 bp->b_flags |= B_ASYNC;
2969 splx(s);
2970 bwrite(bp);
2971 if (nfs_sigintr(nmp, NULL, td)) {
2972 error = EINTR;
2973 goto done;
2974 }
2975 goto loop;
2976 }
2977 splx(s);
2978 if (passone) {
2979 passone = 0;
2980 VI_UNLOCK(vp);
2981 goto again;
2982 }
2983 if (waitfor == MNT_WAIT) {
2984 while (vp->v_bufobj.bo_numoutput) {
2985 error = bufobj_wwait(&vp->v_bufobj, slpflag, slptimeo);
2986 if (error) {
2987 VI_UNLOCK(vp);
2988 error = nfs_sigintr(nmp, NULL, td);
2989 if (error)
2990 goto done;
2991 if (slpflag == PCATCH) {
2992 slpflag = 0;
2993 slptimeo = 2 * hz;
2994 }
2995 VI_LOCK(vp);
2996 }
2997 }
2998 if (vp->v_bufobj.bo_dirty.bv_cnt != 0 && commit) {
2999 VI_UNLOCK(vp);
3000 goto loop;
3001 }
3002 /*
3003 * Wait for all the async IO requests to drain
3004 */
3005 VI_UNLOCK(vp);
3006 mtx_lock(&np->n_mtx);
3007 while (np->n_directio_asyncwr > 0) {
3008 np->n_flag |= NFSYNCWAIT;
3009 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3010 &np->n_mtx, slpflag | (PRIBIO + 1),
3011 "nfsfsync", 0);
3012 if (error) {
3013 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
3014 mtx_unlock(&np->n_mtx);
3015 error = EINTR;
3016 goto done;
3017 }
3018 }
3019 }
3020 mtx_unlock(&np->n_mtx);
3021 } else
3022 VI_UNLOCK(vp);
3023 mtx_lock(&np->n_mtx);
3024 if (np->n_flag & NWRITEERR) {
3025 error = np->n_error;
3026 np->n_flag &= ~NWRITEERR;
3027 }
3028 if (commit && vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
3029 vp->v_bufobj.bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3030 np->n_flag &= ~NMODIFIED;
3031 mtx_unlock(&np->n_mtx);
3032 done:
3033 if (bvec != NULL && bvec != bvec_on_stack)
3034 free(bvec, M_TEMP);
3035 return (error);
3036 }
3037
3038 /*
3039 * NFS advisory byte-level locks.
3040 */
3041 static int
3042 nfs_advlock(struct vop_advlock_args *ap)
3043 {
3044 int error;
3045
3046 mtx_lock(&Giant);
3047 if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3048 struct nfsnode *np = VTONFS(ap->a_vp);
3049
3050 error = lf_advlock(ap, &(np->n_lockf), np->n_size);
3051 goto out;
3052 }
3053 error = nfs_dolock(ap);
3054 out:
3055 mtx_unlock(&Giant);
3056 return (error);
3057 }
3058
3059 /*
3060 * Print out the contents of an nfsnode.
3061 */
3062 static int
3063 nfs_print(struct vop_print_args *ap)
3064 {
3065 struct vnode *vp = ap->a_vp;
3066 struct nfsnode *np = VTONFS(vp);
3067
3068 nfs_printf("\tfileid %ld fsid 0x%x",
3069 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3070 if (vp->v_type == VFIFO)
3071 fifo_printinfo(vp);
3072 printf("\n");
3073 return (0);
3074 }
3075
3076 /*
3077 * This is the "real" nfs::bwrite(struct buf*).
3078 * We set B_CACHE if this is a VMIO buffer.
3079 */
3080 int
3081 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3082 {
3083 int s;
3084 int oldflags = bp->b_flags;
3085 #if 0
3086 int retv = 1;
3087 off_t off;
3088 #endif
3089
3090 if (BUF_REFCNT(bp) == 0)
3091 panic("bwrite: buffer is not locked???");
3092
3093 if (bp->b_flags & B_INVAL) {
3094 brelse(bp);
3095 return(0);
3096 }
3097
3098 bp->b_flags |= B_CACHE;
3099
3100 /*
3101 * Undirty the bp. We will redirty it later if the I/O fails.
3102 */
3103
3104 s = splbio();
3105 bundirty(bp);
3106 bp->b_flags &= ~B_DONE;
3107 bp->b_ioflags &= ~BIO_ERROR;
3108 bp->b_iocmd = BIO_WRITE;
3109
3110 bufobj_wref(bp->b_bufobj);
3111 curthread->td_ru.ru_oublock++;
3112 splx(s);
3113
3114 /*
3115 * Note: to avoid loopback deadlocks, we do not
3116 * assign b_runningbufspace.
3117 */
3118 vfs_busy_pages(bp, 1);
3119
3120 BUF_KERNPROC(bp);
3121 bp->b_iooffset = dbtob(bp->b_blkno);
3122 bstrategy(bp);
3123
3124 if( (oldflags & B_ASYNC) == 0) {
3125 int rtval = bufwait(bp);
3126
3127 if (oldflags & B_DELWRI) {
3128 s = splbio();
3129 reassignbuf(bp);
3130 splx(s);
3131 }
3132 brelse(bp);
3133 return (rtval);
3134 }
3135
3136 return (0);
3137 }
3138
3139 /*
3140 * nfs special file access vnode op.
3141 * Essentially just get vattr and then imitate iaccess() since the device is
3142 * local to the client.
3143 */
3144 static int
3145 nfsspec_access(struct vop_access_args *ap)
3146 {
3147 struct vattr *vap;
3148 struct ucred *cred = ap->a_cred;
3149 struct vnode *vp = ap->a_vp;
3150 mode_t mode = ap->a_mode;
3151 struct vattr vattr;
3152 int error;
3153
3154 /*
3155 * Disallow write attempts on filesystems mounted read-only;
3156 * unless the file is a socket, fifo, or a block or character
3157 * device resident on the filesystem.
3158 */
3159 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3160 switch (vp->v_type) {
3161 case VREG:
3162 case VDIR:
3163 case VLNK:
3164 return (EROFS);
3165 default:
3166 break;
3167 }
3168 }
3169 vap = &vattr;
3170 error = VOP_GETATTR(vp, vap, cred, ap->a_td);
3171 if (error)
3172 goto out;
3173 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3174 mode, cred, NULL);
3175 out:
3176 return error;
3177 }
3178
3179 /*
3180 * Read wrapper for fifos.
3181 */
3182 static int
3183 nfsfifo_read(struct vop_read_args *ap)
3184 {
3185 struct nfsnode *np = VTONFS(ap->a_vp);
3186 int error;
3187
3188 /*
3189 * Set access flag.
3190 */
3191 mtx_lock(&np->n_mtx);
3192 np->n_flag |= NACC;
3193 getnanotime(&np->n_atim);
3194 mtx_unlock(&np->n_mtx);
3195 error = fifo_specops.vop_read(ap);
3196 return error;
3197 }
3198
3199 /*
3200 * Write wrapper for fifos.
3201 */
3202 static int
3203 nfsfifo_write(struct vop_write_args *ap)
3204 {
3205 struct nfsnode *np = VTONFS(ap->a_vp);
3206
3207 /*
3208 * Set update flag.
3209 */
3210 mtx_lock(&np->n_mtx);
3211 np->n_flag |= NUPD;
3212 getnanotime(&np->n_mtim);
3213 mtx_unlock(&np->n_mtx);
3214 return(fifo_specops.vop_write(ap));
3215 }
3216
3217 /*
3218 * Close wrapper for fifos.
3219 *
3220 * Update the times on the nfsnode then do fifo close.
3221 */
3222 static int
3223 nfsfifo_close(struct vop_close_args *ap)
3224 {
3225 struct vnode *vp = ap->a_vp;
3226 struct nfsnode *np = VTONFS(vp);
3227 struct vattr vattr;
3228 struct timespec ts;
3229
3230 mtx_lock(&np->n_mtx);
3231 if (np->n_flag & (NACC | NUPD)) {
3232 getnanotime(&ts);
3233 if (np->n_flag & NACC)
3234 np->n_atim = ts;
3235 if (np->n_flag & NUPD)
3236 np->n_mtim = ts;
3237 np->n_flag |= NCHG;
3238 if (vrefcnt(vp) == 1 &&
3239 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3240 VATTR_NULL(&vattr);
3241 if (np->n_flag & NACC)
3242 vattr.va_atime = np->n_atim;
3243 if (np->n_flag & NUPD)
3244 vattr.va_mtime = np->n_mtim;
3245 mtx_unlock(&np->n_mtx);
3246 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
3247 goto out;
3248 }
3249 }
3250 mtx_unlock(&np->n_mtx);
3251 out:
3252 return (fifo_specops.vop_close(ap));
3253 }
3254
3255 /*
3256 * Just call nfs_writebp() with the force argument set to 1.
3257 *
3258 * NOTE: B_DONE may or may not be set in a_bp on call.
3259 */
3260 static int
3261 nfs_bwrite(struct buf *bp)
3262 {
3263
3264 return (nfs_writebp(bp, 1, curthread));
3265 }
3266
3267 struct buf_ops buf_ops_nfs = {
3268 .bop_name = "buf_ops_nfs",
3269 .bop_write = nfs_bwrite,
3270 .bop_strategy = bufstrategy,
3271 .bop_sync = bufsync,
3272 .bop_bdflush = bufbdflush,
3273 };
Cache object: 5b2a8618295572adf97e9f553c01c3e0
|