nfs_vnops.c revision 1.222.2.9 1 /* $NetBSD: nfs_vnops.c,v 1.222.2.9 2008/02/15 10:40:08 yamt Exp $ */
2
3 /*
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)nfs_vnops.c 8.19 (Berkeley) 7/31/95
35 */
36
37 /*
38 * vnode op calls for Sun NFS version 2 and 3
39 */
40
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: nfs_vnops.c,v 1.222.2.9 2008/02/15 10:40:08 yamt Exp $");
43
44 #include "opt_inet.h"
45 #include "opt_nfs.h"
46 #include "opt_uvmhist.h"
47
48 #include <sys/param.h>
49 #include <sys/proc.h>
50 #include <sys/kernel.h>
51 #include <sys/systm.h>
52 #include <sys/resourcevar.h>
53 #include <sys/mount.h>
54 #include <sys/buf.h>
55 #include <sys/condvar.h>
56 #include <sys/disk.h>
57 #include <sys/malloc.h>
58 #include <sys/kmem.h>
59 #include <sys/mbuf.h>
60 #include <sys/mutex.h>
61 #include <sys/namei.h>
62 #include <sys/vnode.h>
63 #include <sys/dirent.h>
64 #include <sys/fcntl.h>
65 #include <sys/hash.h>
66 #include <sys/lockf.h>
67 #include <sys/stat.h>
68 #include <sys/unistd.h>
69 #include <sys/kauth.h>
70
71 #include <uvm/uvm_extern.h>
72 #include <uvm/uvm.h>
73
74 #include <miscfs/fifofs/fifo.h>
75 #include <miscfs/genfs/genfs.h>
76 #include <miscfs/genfs/genfs_node.h>
77 #include <miscfs/specfs/specdev.h>
78
79 #include <nfs/rpcv2.h>
80 #include <nfs/nfsproto.h>
81 #include <nfs/nfs.h>
82 #include <nfs/nfsnode.h>
83 #include <nfs/nfsmount.h>
84 #include <nfs/xdr_subs.h>
85 #include <nfs/nfsm_subs.h>
86 #include <nfs/nfs_var.h>
87
88 #include <net/if.h>
89 #include <netinet/in.h>
90 #include <netinet/in_var.h>
91
92 /*
93 * Global vfs data structures for nfs
94 */
95 int (**nfsv2_vnodeop_p) __P((void *));
96 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
97 { &vop_default_desc, vn_default_error },
98 { &vop_lookup_desc, nfs_lookup }, /* lookup */
99 { &vop_create_desc, nfs_create }, /* create */
100 { &vop_mknod_desc, nfs_mknod }, /* mknod */
101 { &vop_open_desc, nfs_open }, /* open */
102 { &vop_close_desc, nfs_close }, /* close */
103 { &vop_access_desc, nfs_access }, /* access */
104 { &vop_getattr_desc, nfs_getattr }, /* getattr */
105 { &vop_setattr_desc, nfs_setattr }, /* setattr */
106 { &vop_read_desc, nfs_read }, /* read */
107 { &vop_write_desc, nfs_write }, /* write */
108 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
109 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */
110 { &vop_poll_desc, nfs_poll }, /* poll */
111 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */
112 { &vop_revoke_desc, nfs_revoke }, /* revoke */
113 { &vop_mmap_desc, nfs_mmap }, /* mmap */
114 { &vop_fsync_desc, nfs_fsync }, /* fsync */
115 { &vop_seek_desc, nfs_seek }, /* seek */
116 { &vop_remove_desc, nfs_remove }, /* remove */
117 { &vop_link_desc, nfs_link }, /* link */
118 { &vop_rename_desc, nfs_rename }, /* rename */
119 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */
120 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */
121 { &vop_symlink_desc, nfs_symlink }, /* symlink */
122 { &vop_readdir_desc, nfs_readdir }, /* readdir */
123 { &vop_readlink_desc, nfs_readlink }, /* readlink */
124 { &vop_abortop_desc, nfs_abortop }, /* abortop */
125 { &vop_inactive_desc, nfs_inactive }, /* inactive */
126 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
127 { &vop_lock_desc, nfs_lock }, /* lock */
128 { &vop_unlock_desc, nfs_unlock }, /* unlock */
129 { &vop_bmap_desc, nfs_bmap }, /* bmap */
130 { &vop_strategy_desc, nfs_strategy }, /* strategy */
131 { &vop_print_desc, nfs_print }, /* print */
132 { &vop_islocked_desc, nfs_islocked }, /* islocked */
133 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */
134 { &vop_advlock_desc, nfs_advlock }, /* advlock */
135 { &vop_bwrite_desc, genfs_badop }, /* bwrite */
136 { &vop_getpages_desc, nfs_getpages }, /* getpages */
137 { &vop_putpages_desc, genfs_putpages }, /* putpages */
138 { NULL, NULL }
139 };
140 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
141 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
142
143 /*
144 * Special device vnode ops
145 */
146 int (**spec_nfsv2nodeop_p) __P((void *));
147 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
148 { &vop_default_desc, vn_default_error },
149 { &vop_lookup_desc, spec_lookup }, /* lookup */
150 { &vop_create_desc, spec_create }, /* create */
151 { &vop_mknod_desc, spec_mknod }, /* mknod */
152 { &vop_open_desc, spec_open }, /* open */
153 { &vop_close_desc, nfsspec_close }, /* close */
154 { &vop_access_desc, nfsspec_access }, /* access */
155 { &vop_getattr_desc, nfs_getattr }, /* getattr */
156 { &vop_setattr_desc, nfs_setattr }, /* setattr */
157 { &vop_read_desc, nfsspec_read }, /* read */
158 { &vop_write_desc, nfsspec_write }, /* write */
159 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
160 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */
161 { &vop_poll_desc, spec_poll }, /* poll */
162 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */
163 { &vop_revoke_desc, spec_revoke }, /* revoke */
164 { &vop_mmap_desc, spec_mmap }, /* mmap */
165 { &vop_fsync_desc, spec_fsync }, /* fsync */
166 { &vop_seek_desc, spec_seek }, /* seek */
167 { &vop_remove_desc, spec_remove }, /* remove */
168 { &vop_link_desc, spec_link }, /* link */
169 { &vop_rename_desc, spec_rename }, /* rename */
170 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */
171 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */
172 { &vop_symlink_desc, spec_symlink }, /* symlink */
173 { &vop_readdir_desc, spec_readdir }, /* readdir */
174 { &vop_readlink_desc, spec_readlink }, /* readlink */
175 { &vop_abortop_desc, spec_abortop }, /* abortop */
176 { &vop_inactive_desc, nfs_inactive }, /* inactive */
177 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
178 { &vop_lock_desc, nfs_lock }, /* lock */
179 { &vop_unlock_desc, nfs_unlock }, /* unlock */
180 { &vop_bmap_desc, spec_bmap }, /* bmap */
181 { &vop_strategy_desc, spec_strategy }, /* strategy */
182 { &vop_print_desc, nfs_print }, /* print */
183 { &vop_islocked_desc, nfs_islocked }, /* islocked */
184 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */
185 { &vop_advlock_desc, spec_advlock }, /* advlock */
186 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */
187 { &vop_getpages_desc, spec_getpages }, /* getpages */
188 { &vop_putpages_desc, spec_putpages }, /* putpages */
189 { NULL, NULL }
190 };
191 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
192 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
193
194 int (**fifo_nfsv2nodeop_p) __P((void *));
195 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
196 { &vop_default_desc, vn_default_error },
197 { &vop_lookup_desc, fifo_lookup }, /* lookup */
198 { &vop_create_desc, fifo_create }, /* create */
199 { &vop_mknod_desc, fifo_mknod }, /* mknod */
200 { &vop_open_desc, fifo_open }, /* open */
201 { &vop_close_desc, nfsfifo_close }, /* close */
202 { &vop_access_desc, nfsspec_access }, /* access */
203 { &vop_getattr_desc, nfs_getattr }, /* getattr */
204 { &vop_setattr_desc, nfs_setattr }, /* setattr */
205 { &vop_read_desc, nfsfifo_read }, /* read */
206 { &vop_write_desc, nfsfifo_write }, /* write */
207 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
208 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */
209 { &vop_poll_desc, fifo_poll }, /* poll */
210 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */
211 { &vop_revoke_desc, fifo_revoke }, /* revoke */
212 { &vop_mmap_desc, fifo_mmap }, /* mmap */
213 { &vop_fsync_desc, nfs_fsync }, /* fsync */
214 { &vop_seek_desc, fifo_seek }, /* seek */
215 { &vop_remove_desc, fifo_remove }, /* remove */
216 { &vop_link_desc, fifo_link }, /* link */
217 { &vop_rename_desc, fifo_rename }, /* rename */
218 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */
219 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */
220 { &vop_symlink_desc, fifo_symlink }, /* symlink */
221 { &vop_readdir_desc, fifo_readdir }, /* readdir */
222 { &vop_readlink_desc, fifo_readlink }, /* readlink */
223 { &vop_abortop_desc, fifo_abortop }, /* abortop */
224 { &vop_inactive_desc, nfs_inactive }, /* inactive */
225 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
226 { &vop_lock_desc, nfs_lock }, /* lock */
227 { &vop_unlock_desc, nfs_unlock }, /* unlock */
228 { &vop_bmap_desc, fifo_bmap }, /* bmap */
229 { &vop_strategy_desc, genfs_badop }, /* strategy */
230 { &vop_print_desc, nfs_print }, /* print */
231 { &vop_islocked_desc, nfs_islocked }, /* islocked */
232 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */
233 { &vop_advlock_desc, fifo_advlock }, /* advlock */
234 { &vop_bwrite_desc, genfs_badop }, /* bwrite */
235 { &vop_putpages_desc, fifo_putpages }, /* putpages */
236 { NULL, NULL }
237 };
238 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
239 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
240
241 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *,
242 size_t, kauth_cred_t, struct lwp *);
243 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *);
244
245 /*
246 * Global variables
247 */
248 extern u_int32_t nfs_true, nfs_false;
249 extern u_int32_t nfs_xdrneg1;
250 extern const nfstype nfsv3_type[9];
251
252 int nfs_numasync = 0;
253 #define DIRHDSIZ _DIRENT_NAMEOFF(dp)
254 #define UIO_ADVANCE(uio, siz) \
255 (void)((uio)->uio_resid -= (siz), \
256 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \
257 (uio)->uio_iov->iov_len -= (siz))
258
259 static void nfs_cache_enter(struct vnode *, struct vnode *,
260 struct componentname *);
261
262 static void
263 nfs_cache_enter(struct vnode *dvp, struct vnode *vp,
264 struct componentname *cnp)
265 {
266 struct nfsnode *dnp = VTONFS(dvp);
267
268 if (vp != NULL) {
269 struct nfsnode *np = VTONFS(vp);
270
271 np->n_ctime = np->n_vattr->va_ctime.tv_sec;
272 }
273
274 if (!timespecisset(&dnp->n_nctime))
275 dnp->n_nctime = dnp->n_vattr->va_mtime;
276
277 cache_enter(dvp, vp, cnp);
278 }
279
280 /*
281 * nfs null call from vfs.
282 */
283 int
284 nfs_null(vp, cred, l)
285 struct vnode *vp;
286 kauth_cred_t cred;
287 struct lwp *l;
288 {
289 char *bpos, *dpos;
290 int error = 0;
291 struct mbuf *mreq, *mrep, *md, *mb;
292 struct nfsnode *np = VTONFS(vp);
293
294 nfsm_reqhead(np, NFSPROC_NULL, 0);
295 nfsm_request(np, NFSPROC_NULL, l, cred);
296 nfsm_reqdone;
297 return (error);
298 }
299
300 /*
301 * nfs access vnode op.
302 * For nfs version 2, just return ok. File accesses may fail later.
303 * For nfs version 3, use the access rpc to check accessibility. If file modes
304 * are changed on the server, accesses might still fail later.
305 */
306 int
307 nfs_access(v)
308 void *v;
309 {
310 struct vop_access_args /* {
311 struct vnode *a_vp;
312 int a_mode;
313 kauth_cred_t a_cred;
314 } */ *ap = v;
315 struct vnode *vp = ap->a_vp;
316 #ifndef NFS_V2_ONLY
317 u_int32_t *tl;
318 char *cp;
319 int32_t t1, t2;
320 char *bpos, *dpos, *cp2;
321 int error = 0, attrflag;
322 struct mbuf *mreq, *mrep, *md, *mb;
323 u_int32_t mode, rmode;
324 const int v3 = NFS_ISV3(vp);
325 #endif
326 int cachevalid;
327 struct nfsnode *np = VTONFS(vp);
328 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
329
330 cachevalid = (np->n_accstamp != -1 &&
331 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) &&
332 np->n_accuid == kauth_cred_geteuid(ap->a_cred));
333
334 /*
335 * Check access cache first. If this request has been made for this
336 * uid shortly before, use the cached result.
337 */
338 if (cachevalid) {
339 if (!np->n_accerror) {
340 if ((np->n_accmode & ap->a_mode) == ap->a_mode)
341 return np->n_accerror;
342 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
343 return np->n_accerror;
344 }
345
346 #ifndef NFS_V2_ONLY
347 /*
348 * For nfs v3, do an access rpc, otherwise you are stuck emulating
349 * ufs_access() locally using the vattr. This may not be correct,
350 * since the server may apply other access criteria such as
351 * client uid-->server uid mapping that we do not know about, but
352 * this is better than just returning anything that is lying about
353 * in the cache.
354 */
355 if (v3) {
356 nfsstats.rpccnt[NFSPROC_ACCESS]++;
357 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
358 nfsm_fhtom(np, v3);
359 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
360 if (ap->a_mode & VREAD)
361 mode = NFSV3ACCESS_READ;
362 else
363 mode = 0;
364 if (vp->v_type != VDIR) {
365 if (ap->a_mode & VWRITE)
366 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
367 if (ap->a_mode & VEXEC)
368 mode |= NFSV3ACCESS_EXECUTE;
369 } else {
370 if (ap->a_mode & VWRITE)
371 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
372 NFSV3ACCESS_DELETE);
373 if (ap->a_mode & VEXEC)
374 mode |= NFSV3ACCESS_LOOKUP;
375 }
376 *tl = txdr_unsigned(mode);
377 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred);
378 nfsm_postop_attr(vp, attrflag, 0);
379 if (!error) {
380 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
381 rmode = fxdr_unsigned(u_int32_t, *tl);
382 /*
383 * The NFS V3 spec does not clarify whether or not
384 * the returned access bits can be a superset of
385 * the ones requested, so...
386 */
387 if ((rmode & mode) != mode)
388 error = EACCES;
389 }
390 nfsm_reqdone;
391 } else
392 #endif
393 return (nfsspec_access(ap));
394 #ifndef NFS_V2_ONLY
395 /*
396 * Disallow write attempts on filesystems mounted read-only;
397 * unless the file is a socket, fifo, or a block or character
398 * device resident on the filesystem.
399 */
400 if (!error && (ap->a_mode & VWRITE) &&
401 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
402 switch (vp->v_type) {
403 case VREG:
404 case VDIR:
405 case VLNK:
406 error = EROFS;
407 default:
408 break;
409 }
410 }
411
412 if (!error || error == EACCES) {
413 /*
414 * If we got the same result as for a previous,
415 * different request, OR it in. Don't update
416 * the timestamp in that case.
417 */
418 if (cachevalid && np->n_accstamp != -1 &&
419 error == np->n_accerror) {
420 if (!error)
421 np->n_accmode |= ap->a_mode;
422 else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
423 np->n_accmode = ap->a_mode;
424 } else {
425 np->n_accstamp = time_uptime;
426 np->n_accuid = kauth_cred_geteuid(ap->a_cred);
427 np->n_accmode = ap->a_mode;
428 np->n_accerror = error;
429 }
430 }
431
432 return (error);
433 #endif
434 }
435
436 /*
437 * nfs open vnode op
438 * Check to see if the type is ok
439 * and that deletion is not in progress.
440 * For paged in text files, you will need to flush the page cache
441 * if consistency is lost.
442 */
443 /* ARGSUSED */
444 int
445 nfs_open(v)
446 void *v;
447 {
448 struct vop_open_args /* {
449 struct vnode *a_vp;
450 int a_mode;
451 kauth_cred_t a_cred;
452 } */ *ap = v;
453 struct vnode *vp = ap->a_vp;
454 struct nfsnode *np = VTONFS(vp);
455 int error;
456
457 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
458 return (EACCES);
459 }
460
461 if (ap->a_mode & FREAD) {
462 if (np->n_rcred != NULL)
463 kauth_cred_free(np->n_rcred);
464 np->n_rcred = ap->a_cred;
465 kauth_cred_hold(np->n_rcred);
466 }
467 if (ap->a_mode & FWRITE) {
468 if (np->n_wcred != NULL)
469 kauth_cred_free(np->n_wcred);
470 np->n_wcred = ap->a_cred;
471 kauth_cred_hold(np->n_wcred);
472 }
473
474 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0);
475 if (error)
476 return error;
477
478 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */
479
480 return (0);
481 }
482
483 /*
484 * nfs close vnode op
485 * What an NFS client should do upon close after writing is a debatable issue.
486 * Most NFS clients push delayed writes to the server upon close, basically for
487 * two reasons:
488 * 1 - So that any write errors may be reported back to the client process
489 * doing the close system call. By far the two most likely errors are
490 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
491 * 2 - To put a worst case upper bound on cache inconsistency between
492 * multiple clients for the file.
493 * There is also a consistency problem for Version 2 of the protocol w.r.t.
494 * not being able to tell if other clients are writing a file concurrently,
495 * since there is no way of knowing if the changed modify time in the reply
496 * is only due to the write for this client.
497 * (NFS Version 3 provides weak cache consistency data in the reply that
498 * should be sufficient to detect and handle this case.)
499 *
500 * The current code does the following:
501 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
502 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
503 * or commit them (this satisfies 1 and 2 except for the
504 * case where the server crashes after this close but
505 * before the commit RPC, which is felt to be "good
506 * enough". Changing the last argument to nfs_flush() to
507 * a 1 would force a commit operation, if it is felt a
508 * commit is necessary now.
509 */
510 /* ARGSUSED */
511 int
512 nfs_close(v)
513 void *v;
514 {
515 struct vop_close_args /* {
516 struct vnodeop_desc *a_desc;
517 struct vnode *a_vp;
518 int a_fflag;
519 kauth_cred_t a_cred;
520 } */ *ap = v;
521 struct vnode *vp = ap->a_vp;
522 struct nfsnode *np = VTONFS(vp);
523 int error = 0;
524 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
525
526 if (vp->v_type == VREG) {
527 if (np->n_flag & NMODIFIED) {
528 #ifndef NFS_V2_ONLY
529 if (NFS_ISV3(vp)) {
530 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0);
531 np->n_flag &= ~NMODIFIED;
532 } else
533 #endif
534 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1);
535 NFS_INVALIDATE_ATTRCACHE(np);
536 }
537 if (np->n_flag & NWRITEERR) {
538 np->n_flag &= ~NWRITEERR;
539 error = np->n_error;
540 }
541 }
542 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
543 return (error);
544 }
545
546 /*
547 * nfs getattr call from vfs.
548 */
549 int
550 nfs_getattr(v)
551 void *v;
552 {
553 struct vop_getattr_args /* {
554 struct vnode *a_vp;
555 struct vattr *a_vap;
556 kauth_cred_t a_cred;
557 } */ *ap = v;
558 struct vnode *vp = ap->a_vp;
559 struct nfsnode *np = VTONFS(vp);
560 char *cp;
561 u_int32_t *tl;
562 int32_t t1, t2;
563 char *bpos, *dpos;
564 int error = 0;
565 struct mbuf *mreq, *mrep, *md, *mb;
566 const int v3 = NFS_ISV3(vp);
567
568 /*
569 * Update local times for special files.
570 */
571 if (np->n_flag & (NACC | NUPD))
572 np->n_flag |= NCHG;
573
574 /*
575 * if we have delayed truncation, do it now.
576 */
577 nfs_delayedtruncate(vp);
578
579 /*
580 * First look in the cache.
581 */
582 if (nfs_getattrcache(vp, ap->a_vap) == 0)
583 return (0);
584 nfsstats.rpccnt[NFSPROC_GETATTR]++;
585 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
586 nfsm_fhtom(np, v3);
587 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred);
588 if (!error) {
589 nfsm_loadattr(vp, ap->a_vap, 0);
590 if (vp->v_type == VDIR &&
591 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
592 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
593 }
594 nfsm_reqdone;
595 return (error);
596 }
597
598 /*
599 * nfs setattr call.
600 */
601 int
602 nfs_setattr(v)
603 void *v;
604 {
605 struct vop_setattr_args /* {
606 struct vnodeop_desc *a_desc;
607 struct vnode *a_vp;
608 struct vattr *a_vap;
609 kauth_cred_t a_cred;
610 } */ *ap = v;
611 struct vnode *vp = ap->a_vp;
612 struct nfsnode *np = VTONFS(vp);
613 struct vattr *vap = ap->a_vap;
614 int error = 0;
615 u_quad_t tsize = 0;
616
617 /*
618 * Setting of flags is not supported.
619 */
620 if (vap->va_flags != VNOVAL)
621 return (EOPNOTSUPP);
622
623 /*
624 * Disallow write attempts if the filesystem is mounted read-only.
625 */
626 if ((vap->va_uid != (uid_t)VNOVAL ||
627 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
628 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
629 (vp->v_mount->mnt_flag & MNT_RDONLY))
630 return (EROFS);
631 if (vap->va_size != VNOVAL) {
632 switch (vp->v_type) {
633 case VDIR:
634 return (EISDIR);
635 case VCHR:
636 case VBLK:
637 case VSOCK:
638 case VFIFO:
639 if (vap->va_mtime.tv_sec == VNOVAL &&
640 vap->va_atime.tv_sec == VNOVAL &&
641 vap->va_mode == (mode_t)VNOVAL &&
642 vap->va_uid == (uid_t)VNOVAL &&
643 vap->va_gid == (gid_t)VNOVAL)
644 return (0);
645 vap->va_size = VNOVAL;
646 break;
647 default:
648 /*
649 * Disallow write attempts if the filesystem is
650 * mounted read-only.
651 */
652 if (vp->v_mount->mnt_flag & MNT_RDONLY)
653 return (EROFS);
654 genfs_node_wrlock(vp);
655 uvm_vnp_setsize(vp, vap->va_size);
656 tsize = np->n_size;
657 np->n_size = vap->va_size;
658 if (vap->va_size == 0)
659 error = nfs_vinvalbuf(vp, 0,
660 ap->a_cred, curlwp, 1);
661 else
662 error = nfs_vinvalbuf(vp, V_SAVE,
663 ap->a_cred, curlwp, 1);
664 if (error) {
665 uvm_vnp_setsize(vp, tsize);
666 genfs_node_unlock(vp);
667 return (error);
668 }
669 np->n_vattr->va_size = vap->va_size;
670 }
671 } else {
672 /*
673 * flush files before setattr because a later write of
674 * cached data might change timestamps or reset sugid bits
675 */
676 if ((vap->va_mtime.tv_sec != VNOVAL ||
677 vap->va_atime.tv_sec != VNOVAL ||
678 vap->va_mode != VNOVAL) &&
679 vp->v_type == VREG &&
680 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
681 curlwp, 1)) == EINTR)
682 return (error);
683 }
684 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp);
685 if (vap->va_size != VNOVAL) {
686 if (error) {
687 np->n_size = np->n_vattr->va_size = tsize;
688 uvm_vnp_setsize(vp, np->n_size);
689 }
690 genfs_node_unlock(vp);
691 }
692 VN_KNOTE(vp, NOTE_ATTRIB);
693 return (error);
694 }
695
696 /*
697 * Do an nfs setattr rpc.
698 */
699 int
700 nfs_setattrrpc(vp, vap, cred, l)
701 struct vnode *vp;
702 struct vattr *vap;
703 kauth_cred_t cred;
704 struct lwp *l;
705 {
706 struct nfsv2_sattr *sp;
707 char *cp;
708 int32_t t1, t2;
709 char *bpos, *dpos;
710 u_int32_t *tl;
711 int error = 0;
712 struct mbuf *mreq, *mrep, *md, *mb;
713 const int v3 = NFS_ISV3(vp);
714 struct nfsnode *np = VTONFS(vp);
715 #ifndef NFS_V2_ONLY
716 int wccflag = NFSV3_WCCRATTR;
717 char *cp2;
718 #endif
719
720 nfsstats.rpccnt[NFSPROC_SETATTR]++;
721 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
722 nfsm_fhtom(np, v3);
723 #ifndef NFS_V2_ONLY
724 if (v3) {
725 nfsm_v3attrbuild(vap, true);
726 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
727 *tl = nfs_false;
728 } else {
729 #endif
730 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
731 if (vap->va_mode == (mode_t)VNOVAL)
732 sp->sa_mode = nfs_xdrneg1;
733 else
734 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
735 if (vap->va_uid == (uid_t)VNOVAL)
736 sp->sa_uid = nfs_xdrneg1;
737 else
738 sp->sa_uid = txdr_unsigned(vap->va_uid);
739 if (vap->va_gid == (gid_t)VNOVAL)
740 sp->sa_gid = nfs_xdrneg1;
741 else
742 sp->sa_gid = txdr_unsigned(vap->va_gid);
743 sp->sa_size = txdr_unsigned(vap->va_size);
744 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
745 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
746 #ifndef NFS_V2_ONLY
747 }
748 #endif
749 nfsm_request(np, NFSPROC_SETATTR, l, cred);
750 #ifndef NFS_V2_ONLY
751 if (v3) {
752 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
753 } else
754 #endif
755 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
756 nfsm_reqdone;
757 return (error);
758 }
759
760 /*
761 * nfs lookup call, one step at a time...
762 * First look in cache
763 * If not found, unlock the directory nfsnode and do the rpc
764 *
765 * This code is full of lock/unlock statements and checks, because
766 * we continue after cache_lookup has finished (we need to check
767 * with the attr cache and do an rpc if it has timed out). This means
768 * that the locking effects of cache_lookup have to be taken into
769 * account.
770 */
771 int
772 nfs_lookup(v)
773 void *v;
774 {
775 struct vop_lookup_args /* {
776 struct vnodeop_desc *a_desc;
777 struct vnode *a_dvp;
778 struct vnode **a_vpp;
779 struct componentname *a_cnp;
780 } */ *ap = v;
781 struct componentname *cnp = ap->a_cnp;
782 struct vnode *dvp = ap->a_dvp;
783 struct vnode **vpp = ap->a_vpp;
784 int flags;
785 struct vnode *newvp;
786 u_int32_t *tl;
787 char *cp;
788 int32_t t1, t2;
789 char *bpos, *dpos, *cp2;
790 struct mbuf *mreq, *mrep, *md, *mb;
791 long len;
792 nfsfh_t *fhp;
793 struct nfsnode *np;
794 int error = 0, attrflag, fhsize;
795 const int v3 = NFS_ISV3(dvp);
796
797 flags = cnp->cn_flags;
798
799 *vpp = NULLVP;
800 newvp = NULLVP;
801 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
802 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
803 return (EROFS);
804 if (dvp->v_type != VDIR)
805 return (ENOTDIR);
806
807 /*
808 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves.
809 */
810 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
811 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
812 if (error)
813 return error;
814 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN))
815 return EISDIR;
816 VREF(dvp);
817 *vpp = dvp;
818 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
819 cnp->cn_flags |= SAVENAME;
820 return 0;
821 }
822
823 np = VTONFS(dvp);
824
825 /*
826 * Before tediously performing a linear scan of the directory,
827 * check the name cache to see if the directory/name pair
828 * we are looking for is known already.
829 * If the directory/name pair is found in the name cache,
830 * we have to ensure the directory has not changed from
831 * the time the cache entry has been created. If it has,
832 * the cache entry has to be ignored.
833 */
834 error = cache_lookup_raw(dvp, vpp, cnp);
835 KASSERT(dvp != *vpp);
836 if (error >= 0) {
837 struct vattr vattr;
838 int err2;
839
840 if (error && error != ENOENT) {
841 *vpp = NULLVP;
842 return error;
843 }
844
845 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
846 if (err2 != 0) {
847 if (error == 0)
848 vrele(*vpp);
849 *vpp = NULLVP;
850 return err2;
851 }
852
853 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred)
854 || timespeccmp(&vattr.va_mtime,
855 &VTONFS(dvp)->n_nctime, !=)) {
856 if (error == 0) {
857 vrele(*vpp);
858 *vpp = NULLVP;
859 }
860 cache_purge1(dvp, NULL, PURGE_CHILDREN);
861 timespecclear(&np->n_nctime);
862 goto dorpc;
863 }
864
865 if (error == ENOENT) {
866 goto noentry;
867 }
868
869 newvp = *vpp;
870 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
871 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
872 nfsstats.lookupcache_hits++;
873 if ((flags & ISDOTDOT) != 0) {
874 VOP_UNLOCK(dvp, 0);
875 }
876 error = vn_lock(newvp, LK_EXCLUSIVE);
877 if ((flags & ISDOTDOT) != 0) {
878 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
879 }
880 if (error) {
881 /* newvp has been revoked. */
882 vrele(newvp);
883 *vpp = NULL;
884 return error;
885 }
886 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
887 cnp->cn_flags |= SAVENAME;
888 KASSERT(newvp->v_type != VNON);
889 return (0);
890 }
891 cache_purge1(newvp, NULL, PURGE_PARENTS);
892 vrele(newvp);
893 *vpp = NULLVP;
894 }
895 dorpc:
896 #if 0
897 /*
898 * because nfsv3 has the same CREATE semantics as ours,
899 * we don't have to perform LOOKUPs beforehand.
900 *
901 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL.
902 * XXX although we have no way to know if O_EXCL is requested or not.
903 */
904
905 if (v3 && cnp->cn_nameiop == CREATE &&
906 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN &&
907 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
908 cnp->cn_flags |= SAVENAME;
909 return (EJUSTRETURN);
910 }
911 #endif /* 0 */
912
913 error = 0;
914 newvp = NULLVP;
915 nfsstats.lookupcache_misses++;
916 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
917 len = cnp->cn_namelen;
918 nfsm_reqhead(np, NFSPROC_LOOKUP,
919 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
920 nfsm_fhtom(np, v3);
921 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
922 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred);
923 if (error) {
924 nfsm_postop_attr(dvp, attrflag, 0);
925 m_freem(mrep);
926 goto nfsmout;
927 }
928 nfsm_getfh(fhp, fhsize, v3);
929
930 /*
931 * Handle RENAME case...
932 */
933 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
934 if (NFS_CMPFH(np, fhp, fhsize)) {
935 m_freem(mrep);
936 return (EISDIR);
937 }
938 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
939 if (error) {
940 m_freem(mrep);
941 return error;
942 }
943 newvp = NFSTOV(np);
944 #ifndef NFS_V2_ONLY
945 if (v3) {
946 nfsm_postop_attr(newvp, attrflag, 0);
947 nfsm_postop_attr(dvp, attrflag, 0);
948 } else
949 #endif
950 nfsm_loadattr(newvp, (struct vattr *)0, 0);
951 *vpp = newvp;
952 m_freem(mrep);
953 cnp->cn_flags |= SAVENAME;
954 goto validate;
955 }
956
957 /*
958 * The postop attr handling is duplicated for each if case,
959 * because it should be done while dvp is locked (unlocking
960 * dvp is different for each case).
961 */
962
963 if (NFS_CMPFH(np, fhp, fhsize)) {
964 /*
965 * "." lookup
966 */
967 VREF(dvp);
968 newvp = dvp;
969 #ifndef NFS_V2_ONLY
970 if (v3) {
971 nfsm_postop_attr(newvp, attrflag, 0);
972 nfsm_postop_attr(dvp, attrflag, 0);
973 } else
974 #endif
975 nfsm_loadattr(newvp, (struct vattr *)0, 0);
976 } else if (flags & ISDOTDOT) {
977 /*
978 * ".." lookup
979 */
980 VOP_UNLOCK(dvp, 0);
981 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
982 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
983 if (error) {
984 m_freem(mrep);
985 return error;
986 }
987 newvp = NFSTOV(np);
988
989 #ifndef NFS_V2_ONLY
990 if (v3) {
991 nfsm_postop_attr(newvp, attrflag, 0);
992 nfsm_postop_attr(dvp, attrflag, 0);
993 } else
994 #endif
995 nfsm_loadattr(newvp, (struct vattr *)0, 0);
996 } else {
997 /*
998 * Other lookups.
999 */
1000 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1001 if (error) {
1002 m_freem(mrep);
1003 return error;
1004 }
1005 newvp = NFSTOV(np);
1006 #ifndef NFS_V2_ONLY
1007 if (v3) {
1008 nfsm_postop_attr(newvp, attrflag, 0);
1009 nfsm_postop_attr(dvp, attrflag, 0);
1010 } else
1011 #endif
1012 nfsm_loadattr(newvp, (struct vattr *)0, 0);
1013 }
1014 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1015 cnp->cn_flags |= SAVENAME;
1016 if ((cnp->cn_flags & MAKEENTRY) &&
1017 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1018 nfs_cache_enter(dvp, newvp, cnp);
1019 }
1020 *vpp = newvp;
1021 nfsm_reqdone;
1022 if (error) {
1023 /*
1024 * We get here only because of errors returned by
1025 * the RPC. Otherwise we'll have returned above
1026 * (the nfsm_* macros will jump to nfsm_reqdone
1027 * on error).
1028 */
1029 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1030 cnp->cn_nameiop != CREATE) {
1031 nfs_cache_enter(dvp, NULL, cnp);
1032 }
1033 if (newvp != NULLVP) {
1034 if (newvp == dvp) {
1035 vrele(newvp);
1036 } else {
1037 vput(newvp);
1038 }
1039 }
1040 noentry:
1041 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1042 (flags & ISLASTCN) && error == ENOENT) {
1043 if (dvp->v_mount->mnt_flag & MNT_RDONLY) {
1044 error = EROFS;
1045 } else {
1046 error = EJUSTRETURN;
1047 cnp->cn_flags |= SAVENAME;
1048 }
1049 }
1050 *vpp = NULL;
1051 return error;
1052 }
1053
1054 validate:
1055 /*
1056 * make sure we have valid type and size.
1057 */
1058
1059 newvp = *vpp;
1060 if (newvp->v_type == VNON) {
1061 struct vattr vattr; /* dummy */
1062
1063 KASSERT(VTONFS(newvp)->n_attrstamp == 0);
1064 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred);
1065 if (error) {
1066 vput(newvp);
1067 *vpp = NULL;
1068 }
1069 }
1070
1071 return error;
1072 }
1073
1074 /*
1075 * nfs read call.
1076 * Just call nfs_bioread() to do the work.
1077 */
1078 int
1079 nfs_read(v)
1080 void *v;
1081 {
1082 struct vop_read_args /* {
1083 struct vnode *a_vp;
1084 struct uio *a_uio;
1085 int a_ioflag;
1086 kauth_cred_t a_cred;
1087 } */ *ap = v;
1088 struct vnode *vp = ap->a_vp;
1089
1090 if (vp->v_type != VREG)
1091 return EISDIR;
1092 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1093 }
1094
1095 /*
1096 * nfs readlink call
1097 */
1098 int
1099 nfs_readlink(v)
1100 void *v;
1101 {
1102 struct vop_readlink_args /* {
1103 struct vnode *a_vp;
1104 struct uio *a_uio;
1105 kauth_cred_t a_cred;
1106 } */ *ap = v;
1107 struct vnode *vp = ap->a_vp;
1108 struct nfsnode *np = VTONFS(vp);
1109
1110 if (vp->v_type != VLNK)
1111 return (EPERM);
1112
1113 if (np->n_rcred != NULL) {
1114 kauth_cred_free(np->n_rcred);
1115 }
1116 np->n_rcred = ap->a_cred;
1117 kauth_cred_hold(np->n_rcred);
1118
1119 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1120 }
1121
1122 /*
1123 * Do a readlink rpc.
1124 * Called by nfs_doio() from below the buffer cache.
1125 */
1126 int
1127 nfs_readlinkrpc(vp, uiop, cred)
1128 struct vnode *vp;
1129 struct uio *uiop;
1130 kauth_cred_t cred;
1131 {
1132 u_int32_t *tl;
1133 char *cp;
1134 int32_t t1, t2;
1135 char *bpos, *dpos, *cp2;
1136 int error = 0;
1137 uint32_t len;
1138 struct mbuf *mreq, *mrep, *md, *mb;
1139 const int v3 = NFS_ISV3(vp);
1140 struct nfsnode *np = VTONFS(vp);
1141 #ifndef NFS_V2_ONLY
1142 int attrflag;
1143 #endif
1144
1145 nfsstats.rpccnt[NFSPROC_READLINK]++;
1146 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1147 nfsm_fhtom(np, v3);
1148 nfsm_request(np, NFSPROC_READLINK, curlwp, cred);
1149 #ifndef NFS_V2_ONLY
1150 if (v3)
1151 nfsm_postop_attr(vp, attrflag, 0);
1152 #endif
1153 if (!error) {
1154 #ifndef NFS_V2_ONLY
1155 if (v3) {
1156 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1157 len = fxdr_unsigned(uint32_t, *tl);
1158 if (len > MAXPATHLEN) {
1159 /*
1160 * this pathname is too long for us.
1161 */
1162 m_freem(mrep);
1163 /* Solaris returns EINVAL. should we follow? */
1164 error = ENAMETOOLONG;
1165 goto nfsmout;
1166 }
1167 } else
1168 #endif
1169 {
1170 nfsm_strsiz(len, NFS_MAXPATHLEN);
1171 }
1172 nfsm_mtouio(uiop, len);
1173 }
1174 nfsm_reqdone;
1175 return (error);
1176 }
1177
1178 /*
1179 * nfs read rpc call
1180 * Ditto above
1181 */
1182 int
1183 nfs_readrpc(vp, uiop)
1184 struct vnode *vp;
1185 struct uio *uiop;
1186 {
1187 u_int32_t *tl;
1188 char *cp;
1189 int32_t t1, t2;
1190 char *bpos, *dpos, *cp2;
1191 struct mbuf *mreq, *mrep, *md, *mb;
1192 struct nfsmount *nmp;
1193 int error = 0, len, retlen, tsiz, eof, byte_count;
1194 const int v3 = NFS_ISV3(vp);
1195 struct nfsnode *np = VTONFS(vp);
1196 #ifndef NFS_V2_ONLY
1197 int attrflag;
1198 #endif
1199
1200 #ifndef nolint
1201 eof = 0;
1202 #endif
1203 nmp = VFSTONFS(vp->v_mount);
1204 tsiz = uiop->uio_resid;
1205 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1206 return (EFBIG);
1207 mutex_enter(&nmp->nm_lock);
1208 iostat_busy(nmp->nm_stats);
1209 mutex_exit(&nmp->nm_lock);
1210 byte_count = 0; /* count bytes actually transferred */
1211 while (tsiz > 0) {
1212 nfsstats.rpccnt[NFSPROC_READ]++;
1213 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1214 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1215 nfsm_fhtom(np, v3);
1216 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1217 #ifndef NFS_V2_ONLY
1218 if (v3) {
1219 txdr_hyper(uiop->uio_offset, tl);
1220 *(tl + 2) = txdr_unsigned(len);
1221 } else
1222 #endif
1223 {
1224 *tl++ = txdr_unsigned(uiop->uio_offset);
1225 *tl++ = txdr_unsigned(len);
1226 *tl = 0;
1227 }
1228 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred);
1229 #ifndef NFS_V2_ONLY
1230 if (v3) {
1231 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1232 if (error) {
1233 m_freem(mrep);
1234 goto nfsmout;
1235 }
1236 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1237 eof = fxdr_unsigned(int, *(tl + 1));
1238 } else
1239 #endif
1240 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1241 nfsm_strsiz(retlen, nmp->nm_rsize);
1242 nfsm_mtouio(uiop, retlen);
1243 m_freem(mrep);
1244 tsiz -= retlen;
1245 byte_count += retlen;
1246 #ifndef NFS_V2_ONLY
1247 if (v3) {
1248 if (eof || retlen == 0)
1249 tsiz = 0;
1250 } else
1251 #endif
1252 if (retlen < len)
1253 tsiz = 0;
1254 }
1255 nfsmout:
1256 mutex_enter(&nmp->nm_lock);
1257 iostat_unbusy(nmp->nm_stats, byte_count, 1);
1258 mutex_exit(&nmp->nm_lock);
1259 return (error);
1260 }
1261
1262 struct nfs_writerpc_context {
1263 kmutex_t nwc_lock;
1264 kcondvar_t nwc_cv;
1265 int nwc_mbufcount;
1266 };
1267
1268 /*
1269 * free mbuf used to refer protected pages while write rpc call.
1270 * called at splvm.
1271 */
1272 static void
1273 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg)
1274 {
1275 struct nfs_writerpc_context *ctx = arg;
1276
1277 KASSERT(m != NULL);
1278 KASSERT(ctx != NULL);
1279 pool_cache_put(mb_cache, m);
1280 mutex_enter(&ctx->nwc_lock);
1281 if (--ctx->nwc_mbufcount == 0) {
1282 cv_signal(&ctx->nwc_cv);
1283 }
1284 mutex_exit(&ctx->nwc_lock);
1285 }
1286
1287 /*
1288 * nfs write call
1289 */
1290 int
1291 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp)
1292 struct vnode *vp;
1293 struct uio *uiop;
1294 int *iomode;
1295 bool pageprotected;
1296 bool *stalewriteverfp;
1297 {
1298 u_int32_t *tl;
1299 char *cp;
1300 int32_t t1, t2;
1301 char *bpos, *dpos;
1302 struct mbuf *mreq, *mrep, *md, *mb;
1303 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1304 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR;
1305 const int v3 = NFS_ISV3(vp);
1306 int committed = NFSV3WRITE_FILESYNC;
1307 struct nfsnode *np = VTONFS(vp);
1308 struct nfs_writerpc_context ctx;
1309 int byte_count;
1310 struct lwp *l = NULL;
1311 size_t origresid;
1312 #ifndef NFS_V2_ONLY
1313 char *cp2;
1314 int rlen, commit;
1315 #endif
1316
1317 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM);
1318 cv_init(&ctx.nwc_cv, "nfsmblk");
1319 ctx.nwc_mbufcount = 1;
1320
1321 if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1322 panic("writerpc readonly vp %p", vp);
1323 }
1324
1325 #ifdef DIAGNOSTIC
1326 if (uiop->uio_iovcnt != 1)
1327 panic("nfs: writerpc iovcnt > 1");
1328 #endif
1329 tsiz = uiop->uio_resid;
1330 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1331 return (EFBIG);
1332 if (pageprotected) {
1333 l = curlwp;
1334 uvm_lwp_hold(l);
1335 }
1336 retry:
1337 origresid = uiop->uio_resid;
1338 KASSERT(origresid == uiop->uio_iov->iov_len);
1339 mutex_enter(&nmp->nm_lock);
1340 iostat_busy(nmp->nm_stats);
1341 mutex_exit(&nmp->nm_lock);
1342 byte_count = 0; /* count of bytes actually written */
1343 while (tsiz > 0) {
1344 uint32_t datalen; /* data bytes need to be allocated in mbuf */
1345 uint32_t backup;
1346 bool stalewriteverf = false;
1347
1348 nfsstats.rpccnt[NFSPROC_WRITE]++;
1349 len = min(tsiz, nmp->nm_wsize);
1350 datalen = pageprotected ? 0 : nfsm_rndup(len);
1351 nfsm_reqhead(np, NFSPROC_WRITE,
1352 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1353 nfsm_fhtom(np, v3);
1354 #ifndef NFS_V2_ONLY
1355 if (v3) {
1356 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1357 txdr_hyper(uiop->uio_offset, tl);
1358 tl += 2;
1359 *tl++ = txdr_unsigned(len);
1360 *tl++ = txdr_unsigned(*iomode);
1361 *tl = txdr_unsigned(len);
1362 } else
1363 #endif
1364 {
1365 u_int32_t x;
1366
1367 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1368 /* Set both "begin" and "current" to non-garbage. */
1369 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1370 *tl++ = x; /* "begin offset" */
1371 *tl++ = x; /* "current offset" */
1372 x = txdr_unsigned(len);
1373 *tl++ = x; /* total to this offset */
1374 *tl = x; /* size of this write */
1375
1376 }
1377 if (pageprotected) {
1378 /*
1379 * since we know pages can't be modified during i/o,
1380 * no need to copy them for us.
1381 */
1382 struct mbuf *m;
1383 struct iovec *iovp = uiop->uio_iov;
1384
1385 m = m_get(M_WAIT, MT_DATA);
1386 MCLAIM(m, &nfs_mowner);
1387 MEXTADD(m, iovp->iov_base, len, M_MBUF,
1388 nfs_writerpc_extfree, &ctx);
1389 m->m_flags |= M_EXT_ROMAP;
1390 m->m_len = len;
1391 mb->m_next = m;
1392 /*
1393 * no need to maintain mb and bpos here
1394 * because no one care them later.
1395 */
1396 #if 0
1397 mb = m;
1398 bpos = mtod(void *, mb) + mb->m_len;
1399 #endif
1400 UIO_ADVANCE(uiop, len);
1401 uiop->uio_offset += len;
1402 mutex_enter(&ctx.nwc_lock);
1403 ctx.nwc_mbufcount++;
1404 mutex_exit(&ctx.nwc_lock);
1405 nfs_zeropad(mb, 0, nfsm_padlen(len));
1406 } else {
1407 nfsm_uiotom(uiop, len);
1408 }
1409 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred);
1410 #ifndef NFS_V2_ONLY
1411 if (v3) {
1412 wccflag = NFSV3_WCCCHK;
1413 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error);
1414 if (!error) {
1415 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1416 + NFSX_V3WRITEVERF);
1417 rlen = fxdr_unsigned(int, *tl++);
1418 if (rlen == 0) {
1419 error = NFSERR_IO;
1420 m_freem(mrep);
1421 break;
1422 } else if (rlen < len) {
1423 backup = len - rlen;
1424 UIO_ADVANCE(uiop, -backup);
1425 uiop->uio_offset -= backup;
1426 len = rlen;
1427 }
1428 commit = fxdr_unsigned(int, *tl++);
1429
1430 /*
1431 * Return the lowest committment level
1432 * obtained by any of the RPCs.
1433 */
1434 if (committed == NFSV3WRITE_FILESYNC)
1435 committed = commit;
1436 else if (committed == NFSV3WRITE_DATASYNC &&
1437 commit == NFSV3WRITE_UNSTABLE)
1438 committed = commit;
1439 mutex_enter(&nmp->nm_lock);
1440 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1441 memcpy(nmp->nm_writeverf, tl,
1442 NFSX_V3WRITEVERF);
1443 nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1444 } else if ((nmp->nm_iflag &
1445 NFSMNT_STALEWRITEVERF) ||
1446 memcmp(tl, nmp->nm_writeverf,
1447 NFSX_V3WRITEVERF)) {
1448 memcpy(nmp->nm_writeverf, tl,
1449 NFSX_V3WRITEVERF);
1450 /*
1451 * note NFSMNT_STALEWRITEVERF
1452 * if we're the first thread to
1453 * notice it.
1454 */
1455 if ((nmp->nm_iflag &
1456 NFSMNT_STALEWRITEVERF) == 0) {
1457 stalewriteverf = true;
1458 nmp->nm_iflag |=
1459 NFSMNT_STALEWRITEVERF;
1460 }
1461 }
1462 mutex_exit(&nmp->nm_lock);
1463 }
1464 } else
1465 #endif
1466 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1467 if (wccflag)
1468 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1469 m_freem(mrep);
1470 if (error)
1471 break;
1472 tsiz -= len;
1473 byte_count += len;
1474 if (stalewriteverf) {
1475 *stalewriteverfp = true;
1476 stalewriteverf = false;
1477 if (committed == NFSV3WRITE_UNSTABLE &&
1478 len != origresid) {
1479 /*
1480 * if our write requests weren't atomic but
1481 * unstable, datas in previous iterations
1482 * might have already been lost now.
1483 * then, we should resend them to nfsd.
1484 */
1485 backup = origresid - tsiz;
1486 UIO_ADVANCE(uiop, -backup);
1487 uiop->uio_offset -= backup;
1488 tsiz = origresid;
1489 goto retry;
1490 }
1491 }
1492 }
1493 nfsmout:
1494 mutex_enter(&nmp->nm_lock);
1495 iostat_unbusy(nmp->nm_stats, byte_count, 0);
1496 mutex_exit(&nmp->nm_lock);
1497 if (pageprotected) {
1498 /*
1499 * wait until mbufs go away.
1500 * retransmitted mbufs can survive longer than rpc requests
1501 * themselves.
1502 */
1503 mutex_enter(&ctx.nwc_lock);
1504 ctx.nwc_mbufcount--;
1505 while (ctx.nwc_mbufcount > 0) {
1506 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock);
1507 }
1508 mutex_exit(&ctx.nwc_lock);
1509 uvm_lwp_rele(l);
1510 }
1511 mutex_destroy(&ctx.nwc_lock);
1512 cv_destroy(&ctx.nwc_cv);
1513 *iomode = committed;
1514 if (error)
1515 uiop->uio_resid = tsiz;
1516 return (error);
1517 }
1518
1519 /*
1520 * nfs mknod rpc
1521 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1522 * mode set to specify the file type and the size field for rdev.
1523 */
1524 int
1525 nfs_mknodrpc(dvp, vpp, cnp, vap)
1526 struct vnode *dvp;
1527 struct vnode **vpp;
1528 struct componentname *cnp;
1529 struct vattr *vap;
1530 {
1531 struct nfsv2_sattr *sp;
1532 u_int32_t *tl;
1533 char *cp;
1534 int32_t t1, t2;
1535 struct vnode *newvp = (struct vnode *)0;
1536 struct nfsnode *dnp, *np;
1537 char *cp2;
1538 char *bpos, *dpos;
1539 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1540 struct mbuf *mreq, *mrep, *md, *mb;
1541 u_int32_t rdev;
1542 const int v3 = NFS_ISV3(dvp);
1543
1544 if (vap->va_type == VCHR || vap->va_type == VBLK)
1545 rdev = txdr_unsigned(vap->va_rdev);
1546 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1547 rdev = nfs_xdrneg1;
1548 else {
1549 VOP_ABORTOP(dvp, cnp);
1550 vput(dvp);
1551 return (EOPNOTSUPP);
1552 }
1553 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1554 dnp = VTONFS(dvp);
1555 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1556 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1557 nfsm_fhtom(dnp, v3);
1558 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1559 #ifndef NFS_V2_ONLY
1560 if (v3) {
1561 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1562 *tl++ = vtonfsv3_type(vap->va_type);
1563 nfsm_v3attrbuild(vap, false);
1564 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1565 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1566 *tl++ = txdr_unsigned(major(vap->va_rdev));
1567 *tl = txdr_unsigned(minor(vap->va_rdev));
1568 }
1569 } else
1570 #endif
1571 {
1572 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1573 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1574 sp->sa_uid = nfs_xdrneg1;
1575 sp->sa_gid = nfs_xdrneg1;
1576 sp->sa_size = rdev;
1577 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1578 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1579 }
1580 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred);
1581 if (!error) {
1582 nfsm_mtofh(dvp, newvp, v3, gotvp);
1583 if (!gotvp) {
1584 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1585 cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1586 if (!error)
1587 newvp = NFSTOV(np);
1588 }
1589 }
1590 #ifndef NFS_V2_ONLY
1591 if (v3)
1592 nfsm_wcc_data(dvp, wccflag, 0, !error);
1593 #endif
1594 nfsm_reqdone;
1595 if (error) {
1596 if (newvp)
1597 vput(newvp);
1598 } else {
1599 if (cnp->cn_flags & MAKEENTRY)
1600 nfs_cache_enter(dvp, newvp, cnp);
1601 *vpp = newvp;
1602 }
1603 PNBUF_PUT(cnp->cn_pnbuf);
1604 VTONFS(dvp)->n_flag |= NMODIFIED;
1605 if (!wccflag)
1606 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1607 vput(dvp);
1608 return (error);
1609 }
1610
1611 /*
1612 * nfs mknod vop
1613 * just call nfs_mknodrpc() to do the work.
1614 */
1615 /* ARGSUSED */
1616 int
1617 nfs_mknod(v)
1618 void *v;
1619 {
1620 struct vop_mknod_args /* {
1621 struct vnode *a_dvp;
1622 struct vnode **a_vpp;
1623 struct componentname *a_cnp;
1624 struct vattr *a_vap;
1625 } */ *ap = v;
1626 struct vnode *dvp = ap->a_dvp;
1627 struct componentname *cnp = ap->a_cnp;
1628 int error;
1629
1630 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap);
1631 VN_KNOTE(dvp, NOTE_WRITE);
1632 if (error == 0 || error == EEXIST)
1633 cache_purge1(dvp, cnp, 0);
1634 return (error);
1635 }
1636
1637 #ifndef NFS_V2_ONLY
1638 static u_long create_verf;
1639 #endif
1640 /*
1641 * nfs file create call
1642 */
1643 int
1644 nfs_create(v)
1645 void *v;
1646 {
1647 struct vop_create_args /* {
1648 struct vnode *a_dvp;
1649 struct vnode **a_vpp;
1650 struct componentname *a_cnp;
1651 struct vattr *a_vap;
1652 } */ *ap = v;
1653 struct vnode *dvp = ap->a_dvp;
1654 struct vattr *vap = ap->a_vap;
1655 struct componentname *cnp = ap->a_cnp;
1656 struct nfsv2_sattr *sp;
1657 u_int32_t *tl;
1658 char *cp;
1659 int32_t t1, t2;
1660 struct nfsnode *dnp, *np = (struct nfsnode *)0;
1661 struct vnode *newvp = (struct vnode *)0;
1662 char *bpos, *dpos, *cp2;
1663 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1664 struct mbuf *mreq, *mrep, *md, *mb;
1665 const int v3 = NFS_ISV3(dvp);
1666 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED;
1667
1668 /*
1669 * Oops, not for me..
1670 */
1671 if (vap->va_type == VSOCK)
1672 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1673
1674 KASSERT(vap->va_type == VREG);
1675
1676 #ifdef VA_EXCLUSIVE
1677 if (vap->va_vaflags & VA_EXCLUSIVE) {
1678 excl_mode = NFSV3CREATE_EXCLUSIVE;
1679 }
1680 #endif
1681 again:
1682 error = 0;
1683 nfsstats.rpccnt[NFSPROC_CREATE]++;
1684 dnp = VTONFS(dvp);
1685 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1686 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1687 nfsm_fhtom(dnp, v3);
1688 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1689 #ifndef NFS_V2_ONLY
1690 if (v3) {
1691 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1692 if (excl_mode == NFSV3CREATE_EXCLUSIVE) {
1693 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1694 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1695 #ifdef INET
1696 if (TAILQ_FIRST(&in_ifaddrhead))
1697 *tl++ = TAILQ_FIRST(&in_ifaddrhead)->
1698 ia_addr.sin_addr.s_addr;
1699 else
1700 *tl++ = create_verf;
1701 #else
1702 *tl++ = create_verf;
1703 #endif
1704 *tl = ++create_verf;
1705 } else {
1706 *tl = txdr_unsigned(excl_mode);
1707 nfsm_v3attrbuild(vap, false);
1708 }
1709 } else
1710 #endif
1711 {
1712 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1713 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1714 sp->sa_uid = nfs_xdrneg1;
1715 sp->sa_gid = nfs_xdrneg1;
1716 sp->sa_size = 0;
1717 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1718 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1719 }
1720 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred);
1721 if (!error) {
1722 nfsm_mtofh(dvp, newvp, v3, gotvp);
1723 if (!gotvp) {
1724 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1725 cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1726 if (!error)
1727 newvp = NFSTOV(np);
1728 }
1729 }
1730 #ifndef NFS_V2_ONLY
1731 if (v3)
1732 nfsm_wcc_data(dvp, wccflag, 0, !error);
1733 #endif
1734 nfsm_reqdone;
1735 if (error) {
1736 /*
1737 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
1738 */
1739 if (v3 && error == ENOTSUP) {
1740 if (excl_mode == NFSV3CREATE_EXCLUSIVE) {
1741 excl_mode = NFSV3CREATE_GUARDED;
1742 goto again;
1743 } else if (excl_mode == NFSV3CREATE_GUARDED) {
1744 excl_mode = NFSV3CREATE_UNCHECKED;
1745 goto again;
1746 }
1747 }
1748 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) {
1749 struct timespec ts;
1750
1751 getnanotime(&ts);
1752
1753 /*
1754 * make sure that we'll update timestamps as
1755 * most server implementations use them to store
1756 * the create verifier.
1757 *
1758 * XXX it's better to use TOSERVER always.
1759 */
1760
1761 if (vap->va_atime.tv_sec == VNOVAL)
1762 vap->va_atime = ts;
1763 if (vap->va_mtime.tv_sec == VNOVAL)
1764 vap->va_mtime = ts;
1765
1766 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp);
1767 }
1768 if (error == 0) {
1769 if (cnp->cn_flags & MAKEENTRY)
1770 nfs_cache_enter(dvp, newvp, cnp);
1771 else
1772 cache_purge1(dvp, cnp, 0);
1773 *ap->a_vpp = newvp;
1774 } else {
1775 if (newvp)
1776 vput(newvp);
1777 if (error == EEXIST)
1778 cache_purge1(dvp, cnp, 0);
1779 }
1780 PNBUF_PUT(cnp->cn_pnbuf);
1781 VTONFS(dvp)->n_flag |= NMODIFIED;
1782 if (!wccflag)
1783 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1784 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1785 vput(dvp);
1786 return (error);
1787 }
1788
1789 /*
1790 * nfs file remove call
1791 * To try and make nfs semantics closer to ufs semantics, a file that has
1792 * other processes using the vnode is renamed instead of removed and then
1793 * removed later on the last close.
1794 * - If v_usecount > 1
1795 * If a rename is not already in the works
1796 * call nfs_sillyrename() to set it up
1797 * else
1798 * do the remove rpc
1799 */
1800 int
1801 nfs_remove(v)
1802 void *v;
1803 {
1804 struct vop_remove_args /* {
1805 struct vnodeop_desc *a_desc;
1806 struct vnode * a_dvp;
1807 struct vnode * a_vp;
1808 struct componentname * a_cnp;
1809 } */ *ap = v;
1810 struct vnode *vp = ap->a_vp;
1811 struct vnode *dvp = ap->a_dvp;
1812 struct componentname *cnp = ap->a_cnp;
1813 struct nfsnode *np = VTONFS(vp);
1814 int error = 0;
1815 struct vattr vattr;
1816
1817 #ifndef DIAGNOSTIC
1818 if ((cnp->cn_flags & HASBUF) == 0)
1819 panic("nfs_remove: no name");
1820 if (vp->v_usecount < 1)
1821 panic("nfs_remove: bad v_usecount");
1822 #endif
1823 if (vp->v_type == VDIR)
1824 error = EPERM;
1825 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1826 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1827 vattr.va_nlink > 1)) {
1828 /*
1829 * Purge the name cache so that the chance of a lookup for
1830 * the name succeeding while the remove is in progress is
1831 * minimized. Without node locking it can still happen, such
1832 * that an I/O op returns ESTALE, but since you get this if
1833 * another host removes the file..
1834 */
1835 cache_purge(vp);
1836 /*
1837 * throw away biocache buffers, mainly to avoid
1838 * unnecessary delayed writes later.
1839 */
1840 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1);
1841 /* Do the rpc */
1842 if (error != EINTR)
1843 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1844 cnp->cn_namelen, cnp->cn_cred, curlwp);
1845 } else if (!np->n_sillyrename)
1846 error = nfs_sillyrename(dvp, vp, cnp, false);
1847 PNBUF_PUT(cnp->cn_pnbuf);
1848 if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1849 vattr.va_nlink == 1) {
1850 np->n_flag |= NREMOVED;
1851 }
1852 NFS_INVALIDATE_ATTRCACHE(np);
1853 VN_KNOTE(vp, NOTE_DELETE);
1854 VN_KNOTE(dvp, NOTE_WRITE);
1855 if (dvp == vp)
1856 vrele(vp);
1857 else
1858 vput(vp);
1859 vput(dvp);
1860 return (error);
1861 }
1862
1863 /*
1864 * nfs file remove rpc called from nfs_inactive
1865 */
1866 int
1867 nfs_removeit(sp)
1868 struct sillyrename *sp;
1869 {
1870
1871 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1872 (struct lwp *)0));
1873 }
1874
1875 /*
1876 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1877 */
1878 int
1879 nfs_removerpc(dvp, name, namelen, cred, l)
1880 struct vnode *dvp;
1881 const char *name;
1882 int namelen;
1883 kauth_cred_t cred;
1884 struct lwp *l;
1885 {
1886 u_int32_t *tl;
1887 char *cp;
1888 #ifndef NFS_V2_ONLY
1889 int32_t t1;
1890 char *cp2;
1891 #endif
1892 int32_t t2;
1893 char *bpos, *dpos;
1894 int error = 0, wccflag = NFSV3_WCCRATTR;
1895 struct mbuf *mreq, *mrep, *md, *mb;
1896 const int v3 = NFS_ISV3(dvp);
1897 int rexmit = 0;
1898 struct nfsnode *dnp = VTONFS(dvp);
1899
1900 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1901 nfsm_reqhead(dnp, NFSPROC_REMOVE,
1902 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1903 nfsm_fhtom(dnp, v3);
1904 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1905 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit);
1906 #ifndef NFS_V2_ONLY
1907 if (v3)
1908 nfsm_wcc_data(dvp, wccflag, 0, !error);
1909 #endif
1910 nfsm_reqdone;
1911 VTONFS(dvp)->n_flag |= NMODIFIED;
1912 if (!wccflag)
1913 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1914 /*
1915 * Kludge City: If the first reply to the remove rpc is lost..
1916 * the reply to the retransmitted request will be ENOENT
1917 * since the file was in fact removed
1918 * Therefore, we cheat and return success.
1919 */
1920 if (rexmit && error == ENOENT)
1921 error = 0;
1922 return (error);
1923 }
1924
1925 /*
1926 * nfs file rename call
1927 */
1928 int
1929 nfs_rename(v)
1930 void *v;
1931 {
1932 struct vop_rename_args /* {
1933 struct vnode *a_fdvp;
1934 struct vnode *a_fvp;
1935 struct componentname *a_fcnp;
1936 struct vnode *a_tdvp;
1937 struct vnode *a_tvp;
1938 struct componentname *a_tcnp;
1939 } */ *ap = v;
1940 struct vnode *fvp = ap->a_fvp;
1941 struct vnode *tvp = ap->a_tvp;
1942 struct vnode *fdvp = ap->a_fdvp;
1943 struct vnode *tdvp = ap->a_tdvp;
1944 struct componentname *tcnp = ap->a_tcnp;
1945 struct componentname *fcnp = ap->a_fcnp;
1946 int error;
1947
1948 #ifndef DIAGNOSTIC
1949 if ((tcnp->cn_flags & HASBUF) == 0 ||
1950 (fcnp->cn_flags & HASBUF) == 0)
1951 panic("nfs_rename: no name");
1952 #endif
1953 /* Check for cross-device rename */
1954 if ((fvp->v_mount != tdvp->v_mount) ||
1955 (tvp && (fvp->v_mount != tvp->v_mount))) {
1956 error = EXDEV;
1957 goto out;
1958 }
1959
1960 /*
1961 * If the tvp exists and is in use, sillyrename it before doing the
1962 * rename of the new file over it.
1963 *
1964 * Have sillyrename use link instead of rename if possible,
1965 * so that we don't lose the file if the rename fails, and so
1966 * that there's no window when the "to" file doesn't exist.
1967 */
1968 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1969 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) {
1970 VN_KNOTE(tvp, NOTE_DELETE);
1971 vput(tvp);
1972 tvp = NULL;
1973 }
1974
1975 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1976 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1977 curlwp);
1978
1979 VN_KNOTE(fdvp, NOTE_WRITE);
1980 VN_KNOTE(tdvp, NOTE_WRITE);
1981 if (error == 0 || error == EEXIST) {
1982 if (fvp->v_type == VDIR)
1983 cache_purge(fvp);
1984 else
1985 cache_purge1(fdvp, fcnp, 0);
1986 if (tvp != NULL && tvp->v_type == VDIR)
1987 cache_purge(tvp);
1988 else
1989 cache_purge1(tdvp, tcnp, 0);
1990 }
1991 out:
1992 if (tdvp == tvp)
1993 vrele(tdvp);
1994 else
1995 vput(tdvp);
1996 if (tvp)
1997 vput(tvp);
1998 vrele(fdvp);
1999 vrele(fvp);
2000 return (error);
2001 }
2002
2003 /*
2004 * nfs file rename rpc called from nfs_remove() above
2005 */
2006 int
2007 nfs_renameit(sdvp, scnp, sp)
2008 struct vnode *sdvp;
2009 struct componentname *scnp;
2010 struct sillyrename *sp;
2011 {
2012 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
2013 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp));
2014 }
2015
2016 /*
2017 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2018 */
2019 int
2020 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, l)
2021 struct vnode *fdvp;
2022 const char *fnameptr;
2023 int fnamelen;
2024 struct vnode *tdvp;
2025 const char *tnameptr;
2026 int tnamelen;
2027 kauth_cred_t cred;
2028 struct lwp *l;
2029 {
2030 u_int32_t *tl;
2031 char *cp;
2032 #ifndef NFS_V2_ONLY
2033 int32_t t1;
2034 char *cp2;
2035 #endif
2036 int32_t t2;
2037 char *bpos, *dpos;
2038 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
2039 struct mbuf *mreq, *mrep, *md, *mb;
2040 const int v3 = NFS_ISV3(fdvp);
2041 int rexmit = 0;
2042 struct nfsnode *fdnp = VTONFS(fdvp);
2043
2044 nfsstats.rpccnt[NFSPROC_RENAME]++;
2045 nfsm_reqhead(fdnp, NFSPROC_RENAME,
2046 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
2047 nfsm_rndup(tnamelen));
2048 nfsm_fhtom(fdnp, v3);
2049 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
2050 nfsm_fhtom(VTONFS(tdvp), v3);
2051 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
2052 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit);
2053 #ifndef NFS_V2_ONLY
2054 if (v3) {
2055 nfsm_wcc_data(fdvp, fwccflag, 0, !error);
2056 nfsm_wcc_data(tdvp, twccflag, 0, !error);
2057 }
2058 #endif
2059 nfsm_reqdone;
2060 VTONFS(fdvp)->n_flag |= NMODIFIED;
2061 VTONFS(tdvp)->n_flag |= NMODIFIED;
2062 if (!fwccflag)
2063 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
2064 if (!twccflag)
2065 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
2066 /*
2067 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2068 */
2069 if (rexmit && error == ENOENT)
2070 error = 0;
2071 return (error);
2072 }
2073
2074 /*
2075 * NFS link RPC, called from nfs_link.
2076 * Assumes dvp and vp locked, and leaves them that way.
2077 */
2078
2079 static int
2080 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name,
2081 size_t namelen, kauth_cred_t cred, struct lwp *l)
2082 {
2083 u_int32_t *tl;
2084 char *cp;
2085 #ifndef NFS_V2_ONLY
2086 int32_t t1;
2087 char *cp2;
2088 #endif
2089 int32_t t2;
2090 char *bpos, *dpos;
2091 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2092 struct mbuf *mreq, *mrep, *md, *mb;
2093 const int v3 = NFS_ISV3(dvp);
2094 int rexmit = 0;
2095 struct nfsnode *np = VTONFS(vp);
2096
2097 nfsstats.rpccnt[NFSPROC_LINK]++;
2098 nfsm_reqhead(np, NFSPROC_LINK,
2099 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen));
2100 nfsm_fhtom(np, v3);
2101 nfsm_fhtom(VTONFS(dvp), v3);
2102 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
2103 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit);
2104 #ifndef NFS_V2_ONLY
2105 if (v3) {
2106 nfsm_postop_attr(vp, attrflag, 0);
2107 nfsm_wcc_data(dvp, wccflag, 0, !error);
2108 }
2109 #endif
2110 nfsm_reqdone;
2111
2112 VTONFS(dvp)->n_flag |= NMODIFIED;
2113 if (!attrflag)
2114 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
2115 if (!wccflag)
2116 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2117
2118 /*
2119 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2120 */
2121 if (rexmit && error == EEXIST)
2122 error = 0;
2123
2124 return error;
2125 }
2126
2127 /*
2128 * nfs hard link create call
2129 */
2130 int
2131 nfs_link(v)
2132 void *v;
2133 {
2134 struct vop_link_args /* {
2135 struct vnode *a_dvp;
2136 struct vnode *a_vp;
2137 struct componentname *a_cnp;
2138 } */ *ap = v;
2139 struct vnode *vp = ap->a_vp;
2140 struct vnode *dvp = ap->a_dvp;
2141 struct componentname *cnp = ap->a_cnp;
2142 int error = 0;
2143
2144 if (dvp->v_mount != vp->v_mount) {
2145 VOP_ABORTOP(dvp, cnp);
2146 vput(dvp);
2147 return (EXDEV);
2148 }
2149 if (dvp != vp) {
2150 error = vn_lock(vp, LK_EXCLUSIVE);
2151 if (error != 0) {
2152 VOP_ABORTOP(dvp, cnp);
2153 vput(dvp);
2154 return error;
2155 }
2156 }
2157
2158 /*
2159 * Push all writes to the server, so that the attribute cache
2160 * doesn't get "out of sync" with the server.
2161 * XXX There should be a better way!
2162 */
2163 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0);
2164
2165 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2166 cnp->cn_cred, curlwp);
2167
2168 if (error == 0)
2169 cache_purge1(dvp, cnp, 0);
2170 PNBUF_PUT(cnp->cn_pnbuf);
2171 if (dvp != vp)
2172 VOP_UNLOCK(vp, 0);
2173 VN_KNOTE(vp, NOTE_LINK);
2174 VN_KNOTE(dvp, NOTE_WRITE);
2175 vput(dvp);
2176 return (error);
2177 }
2178
2179 /*
2180 * nfs symbolic link create call
2181 */
2182 int
2183 nfs_symlink(v)
2184 void *v;
2185 {
2186 struct vop_symlink_args /* {
2187 struct vnode *a_dvp;
2188 struct vnode **a_vpp;
2189 struct componentname *a_cnp;
2190 struct vattr *a_vap;
2191 char *a_target;
2192 } */ *ap = v;
2193 struct vnode *dvp = ap->a_dvp;
2194 struct vattr *vap = ap->a_vap;
2195 struct componentname *cnp = ap->a_cnp;
2196 struct nfsv2_sattr *sp;
2197 u_int32_t *tl;
2198 char *cp;
2199 int32_t t1, t2;
2200 char *bpos, *dpos, *cp2;
2201 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2202 struct mbuf *mreq, *mrep, *md, *mb;
2203 struct vnode *newvp = (struct vnode *)0;
2204 const int v3 = NFS_ISV3(dvp);
2205 int rexmit = 0;
2206 struct nfsnode *dnp = VTONFS(dvp);
2207
2208 *ap->a_vpp = NULL;
2209 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2210 slen = strlen(ap->a_target);
2211 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2212 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2213 nfsm_fhtom(dnp, v3);
2214 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2215 #ifndef NFS_V2_ONlY
2216 if (v3)
2217 nfsm_v3attrbuild(vap, false);
2218 #endif
2219 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2220 #ifndef NFS_V2_ONlY
2221 if (!v3) {
2222 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2223 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2224 sp->sa_uid = nfs_xdrneg1;
2225 sp->sa_gid = nfs_xdrneg1;
2226 sp->sa_size = nfs_xdrneg1;
2227 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2228 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2229 }
2230 #endif
2231 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred,
2232 &rexmit);
2233 #ifndef NFS_V2_ONlY
2234 if (v3) {
2235 if (!error)
2236 nfsm_mtofh(dvp, newvp, v3, gotvp);
2237 nfsm_wcc_data(dvp, wccflag, 0, !error);
2238 }
2239 #endif
2240 nfsm_reqdone;
2241 /*
2242 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2243 */
2244 if (rexmit && error == EEXIST)
2245 error = 0;
2246 if (error == 0 || error == EEXIST)
2247 cache_purge1(dvp, cnp, 0);
2248 if (error == 0 && newvp == NULL) {
2249 struct nfsnode *np = NULL;
2250
2251 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2252 cnp->cn_cred, curlwp, &np);
2253 if (error == 0)
2254 newvp = NFSTOV(np);
2255 }
2256 if (error) {
2257 if (newvp != NULL)
2258 vput(newvp);
2259 } else {
2260 *ap->a_vpp = newvp;
2261 }
2262 PNBUF_PUT(cnp->cn_pnbuf);
2263 VTONFS(dvp)->n_flag |= NMODIFIED;
2264 if (!wccflag)
2265 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2266 VN_KNOTE(dvp, NOTE_WRITE);
2267 vput(dvp);
2268 return (error);
2269 }
2270
2271 /*
2272 * nfs make dir call
2273 */
2274 int
2275 nfs_mkdir(v)
2276 void *v;
2277 {
2278 struct vop_mkdir_args /* {
2279 struct vnode *a_dvp;
2280 struct vnode **a_vpp;
2281 struct componentname *a_cnp;
2282 struct vattr *a_vap;
2283 } */ *ap = v;
2284 struct vnode *dvp = ap->a_dvp;
2285 struct vattr *vap = ap->a_vap;
2286 struct componentname *cnp = ap->a_cnp;
2287 struct nfsv2_sattr *sp;
2288 u_int32_t *tl;
2289 char *cp;
2290 int32_t t1, t2;
2291 int len;
2292 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2293 struct vnode *newvp = (struct vnode *)0;
2294 char *bpos, *dpos, *cp2;
2295 int error = 0, wccflag = NFSV3_WCCRATTR;
2296 int gotvp = 0;
2297 int rexmit = 0;
2298 struct mbuf *mreq, *mrep, *md, *mb;
2299 const int v3 = NFS_ISV3(dvp);
2300
2301 len = cnp->cn_namelen;
2302 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2303 nfsm_reqhead(dnp, NFSPROC_MKDIR,
2304 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2305 nfsm_fhtom(dnp, v3);
2306 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2307 #ifndef NFS_V2_ONLY
2308 if (v3) {
2309 nfsm_v3attrbuild(vap, false);
2310 } else
2311 #endif
2312 {
2313 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2314 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2315 sp->sa_uid = nfs_xdrneg1;
2316 sp->sa_gid = nfs_xdrneg1;
2317 sp->sa_size = nfs_xdrneg1;
2318 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2319 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2320 }
2321 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit);
2322 if (!error)
2323 nfsm_mtofh(dvp, newvp, v3, gotvp);
2324 if (v3)
2325 nfsm_wcc_data(dvp, wccflag, 0, !error);
2326 nfsm_reqdone;
2327 VTONFS(dvp)->n_flag |= NMODIFIED;
2328 if (!wccflag)
2329 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2330 /*
2331 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2332 * if we can succeed in looking up the directory.
2333 */
2334 if ((rexmit && error == EEXIST) || (!error && !gotvp)) {
2335 if (newvp) {
2336 vput(newvp);
2337 newvp = (struct vnode *)0;
2338 }
2339 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2340 curlwp, &np);
2341 if (!error) {
2342 newvp = NFSTOV(np);
2343 if (newvp->v_type != VDIR || newvp == dvp)
2344 error = EEXIST;
2345 }
2346 }
2347 if (error) {
2348 if (newvp) {
2349 if (dvp != newvp)
2350 vput(newvp);
2351 else
2352 vrele(newvp);
2353 }
2354 } else {
2355 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2356 if (cnp->cn_flags & MAKEENTRY)
2357 nfs_cache_enter(dvp, newvp, cnp);
2358 *ap->a_vpp = newvp;
2359 }
2360 PNBUF_PUT(cnp->cn_pnbuf);
2361 vput(dvp);
2362 return (error);
2363 }
2364
2365 /*
2366 * nfs remove directory call
2367 */
2368 int
2369 nfs_rmdir(v)
2370 void *v;
2371 {
2372 struct vop_rmdir_args /* {
2373 struct vnode *a_dvp;
2374 struct vnode *a_vp;
2375 struct componentname *a_cnp;
2376 } */ *ap = v;
2377 struct vnode *vp = ap->a_vp;
2378 struct vnode *dvp = ap->a_dvp;
2379 struct componentname *cnp = ap->a_cnp;
2380 u_int32_t *tl;
2381 char *cp;
2382 #ifndef NFS_V2_ONLY
2383 int32_t t1;
2384 char *cp2;
2385 #endif
2386 int32_t t2;
2387 char *bpos, *dpos;
2388 int error = 0, wccflag = NFSV3_WCCRATTR;
2389 int rexmit = 0;
2390 struct mbuf *mreq, *mrep, *md, *mb;
2391 const int v3 = NFS_ISV3(dvp);
2392 struct nfsnode *dnp;
2393
2394 if (dvp == vp) {
2395 vrele(dvp);
2396 vput(dvp);
2397 PNBUF_PUT(cnp->cn_pnbuf);
2398 return (EINVAL);
2399 }
2400 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2401 dnp = VTONFS(dvp);
2402 nfsm_reqhead(dnp, NFSPROC_RMDIR,
2403 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2404 nfsm_fhtom(dnp, v3);
2405 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2406 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit);
2407 #ifndef NFS_V2_ONLY
2408 if (v3)
2409 nfsm_wcc_data(dvp, wccflag, 0, !error);
2410 #endif
2411 nfsm_reqdone;
2412 PNBUF_PUT(cnp->cn_pnbuf);
2413 VTONFS(dvp)->n_flag |= NMODIFIED;
2414 if (!wccflag)
2415 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2416 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2417 VN_KNOTE(vp, NOTE_DELETE);
2418 cache_purge(vp);
2419 vput(vp);
2420 vput(dvp);
2421 /*
2422 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2423 */
2424 if (rexmit && error == ENOENT)
2425 error = 0;
2426 return (error);
2427 }
2428
2429 /*
2430 * nfs readdir call
2431 */
2432 int
2433 nfs_readdir(v)
2434 void *v;
2435 {
2436 struct vop_readdir_args /* {
2437 struct vnode *a_vp;
2438 struct uio *a_uio;
2439 kauth_cred_t a_cred;
2440 int *a_eofflag;
2441 off_t **a_cookies;
2442 int *a_ncookies;
2443 } */ *ap = v;
2444 struct vnode *vp = ap->a_vp;
2445 struct uio *uio = ap->a_uio;
2446 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2447 char *base = uio->uio_iov->iov_base;
2448 int tresid, error;
2449 size_t count, lost;
2450 struct dirent *dp;
2451 off_t *cookies = NULL;
2452 int ncookies = 0, nc;
2453
2454 if (vp->v_type != VDIR)
2455 return (EPERM);
2456
2457 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2458 count = uio->uio_resid - lost;
2459 if (count <= 0)
2460 return (EINVAL);
2461
2462 /*
2463 * Call nfs_bioread() to do the real work.
2464 */
2465 tresid = uio->uio_resid = count;
2466 error = nfs_bioread(vp, uio, 0, ap->a_cred,
2467 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2468
2469 if (!error && ap->a_cookies) {
2470 ncookies = count / 16;
2471 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2472 *ap->a_cookies = cookies;
2473 }
2474
2475 if (!error && uio->uio_resid == tresid) {
2476 uio->uio_resid += lost;
2477 nfsstats.direofcache_misses++;
2478 if (ap->a_cookies)
2479 *ap->a_ncookies = 0;
2480 *ap->a_eofflag = 1;
2481 return (0);
2482 }
2483
2484 if (!error && ap->a_cookies) {
2485 /*
2486 * Only the NFS server and emulations use cookies, and they
2487 * load the directory block into system space, so we can
2488 * just look at it directly.
2489 */
2490 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) ||
2491 uio->uio_iovcnt != 1)
2492 panic("nfs_readdir: lost in space");
2493 for (nc = 0; ncookies-- &&
2494 base < (char *)uio->uio_iov->iov_base; nc++){
2495 dp = (struct dirent *) base;
2496 if (dp->d_reclen == 0)
2497 break;
2498 if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2499 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2500 else
2501 *(cookies++) = NFS_GETCOOKIE(dp);
2502 base += dp->d_reclen;
2503 }
2504 uio->uio_resid +=
2505 ((char *)uio->uio_iov->iov_base - base);
2506 uio->uio_iov->iov_len +=
2507 ((char *)uio->uio_iov->iov_base - base);
2508 uio->uio_iov->iov_base = base;
2509 *ap->a_ncookies = nc;
2510 }
2511
2512 uio->uio_resid += lost;
2513 *ap->a_eofflag = 0;
2514 return (error);
2515 }
2516
2517 /*
2518 * Readdir rpc call.
2519 * Called from below the buffer cache by nfs_doio().
2520 */
2521 int
2522 nfs_readdirrpc(vp, uiop, cred)
2523 struct vnode *vp;
2524 struct uio *uiop;
2525 kauth_cred_t cred;
2526 {
2527 int len, left;
2528 struct dirent *dp = NULL;
2529 u_int32_t *tl;
2530 char *cp;
2531 int32_t t1, t2;
2532 char *bpos, *dpos, *cp2;
2533 struct mbuf *mreq, *mrep, *md, *mb;
2534 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2535 struct nfsnode *dnp = VTONFS(vp);
2536 u_quad_t fileno;
2537 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1;
2538 #ifndef NFS_V2_ONLY
2539 int attrflag;
2540 #endif
2541 int nrpcs = 0, reclen;
2542 const int v3 = NFS_ISV3(vp);
2543
2544 #ifdef DIAGNOSTIC
2545 /*
2546 * Should be called from buffer cache, so only amount of
2547 * NFS_DIRBLKSIZ will be requested.
2548 */
2549 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2550 panic("nfs readdirrpc bad uio");
2551 #endif
2552
2553 /*
2554 * Loop around doing readdir rpc's of size nm_readdirsize
2555 * truncated to a multiple of NFS_DIRFRAGSIZ.
2556 * The stopping criteria is EOF or buffer full.
2557 */
2558 while (more_dirs && bigenough) {
2559 /*
2560 * Heuristic: don't bother to do another RPC to further
2561 * fill up this block if there is not much room left. (< 50%
2562 * of the readdir RPC size). This wastes some buffer space
2563 * but can save up to 50% in RPC calls.
2564 */
2565 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2566 bigenough = 0;
2567 break;
2568 }
2569 nfsstats.rpccnt[NFSPROC_READDIR]++;
2570 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2571 NFSX_READDIR(v3));
2572 nfsm_fhtom(dnp, v3);
2573 #ifndef NFS_V2_ONLY
2574 if (v3) {
2575 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2576 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2577 txdr_swapcookie3(uiop->uio_offset, tl);
2578 } else {
2579 txdr_cookie3(uiop->uio_offset, tl);
2580 }
2581 tl += 2;
2582 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2583 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2584 } else
2585 #endif
2586 {
2587 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2588 *tl++ = txdr_unsigned(uiop->uio_offset);
2589 }
2590 *tl = txdr_unsigned(nmp->nm_readdirsize);
2591 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred);
2592 nrpcs++;
2593 #ifndef NFS_V2_ONLY
2594 if (v3) {
2595 nfsm_postop_attr(vp, attrflag, 0);
2596 if (!error) {
2597 nfsm_dissect(tl, u_int32_t *,
2598 2 * NFSX_UNSIGNED);
2599 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2600 dnp->n_cookieverf.nfsuquad[1] = *tl;
2601 } else {
2602 m_freem(mrep);
2603 goto nfsmout;
2604 }
2605 }
2606 #endif
2607 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2608 more_dirs = fxdr_unsigned(int, *tl);
2609
2610 /* loop thru the dir entries, doctoring them to 4bsd form */
2611 while (more_dirs && bigenough) {
2612 #ifndef NFS_V2_ONLY
2613 if (v3) {
2614 nfsm_dissect(tl, u_int32_t *,
2615 3 * NFSX_UNSIGNED);
2616 fileno = fxdr_hyper(tl);
2617 len = fxdr_unsigned(int, *(tl + 2));
2618 } else
2619 #endif
2620 {
2621 nfsm_dissect(tl, u_int32_t *,
2622 2 * NFSX_UNSIGNED);
2623 fileno = fxdr_unsigned(u_quad_t, *tl++);
2624 len = fxdr_unsigned(int, *tl);
2625 }
2626 if (len <= 0 || len > NFS_MAXNAMLEN) {
2627 error = EBADRPC;
2628 m_freem(mrep);
2629 goto nfsmout;
2630 }
2631 /* for cookie stashing */
2632 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2633 left = NFS_DIRFRAGSIZ - blksiz;
2634 if (reclen > left) {
2635 memset(uiop->uio_iov->iov_base, 0, left);
2636 dp->d_reclen += left;
2637 UIO_ADVANCE(uiop, left);
2638 blksiz = 0;
2639 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2640 }
2641 if (reclen > uiop->uio_resid)
2642 bigenough = 0;
2643 if (bigenough) {
2644 int tlen;
2645
2646 dp = (struct dirent *)uiop->uio_iov->iov_base;
2647 dp->d_fileno = fileno;
2648 dp->d_namlen = len;
2649 dp->d_reclen = reclen;
2650 dp->d_type = DT_UNKNOWN;
2651 blksiz += reclen;
2652 if (blksiz == NFS_DIRFRAGSIZ)
2653 blksiz = 0;
2654 UIO_ADVANCE(uiop, DIRHDSIZ);
2655 nfsm_mtouio(uiop, len);
2656 tlen = reclen - (DIRHDSIZ + len);
2657 (void)memset(uiop->uio_iov->iov_base, 0, tlen);
2658 UIO_ADVANCE(uiop, tlen);
2659 } else
2660 nfsm_adv(nfsm_rndup(len));
2661 #ifndef NFS_V2_ONLY
2662 if (v3) {
2663 nfsm_dissect(tl, u_int32_t *,
2664 3 * NFSX_UNSIGNED);
2665 } else
2666 #endif
2667 {
2668 nfsm_dissect(tl, u_int32_t *,
2669 2 * NFSX_UNSIGNED);
2670 }
2671 if (bigenough) {
2672 #ifndef NFS_V2_ONLY
2673 if (v3) {
2674 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2675 uiop->uio_offset =
2676 fxdr_swapcookie3(tl);
2677 else
2678 uiop->uio_offset =
2679 fxdr_cookie3(tl);
2680 }
2681 else
2682 #endif
2683 {
2684 uiop->uio_offset =
2685 fxdr_unsigned(off_t, *tl);
2686 }
2687 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2688 }
2689 if (v3)
2690 tl += 2;
2691 else
2692 tl++;
2693 more_dirs = fxdr_unsigned(int, *tl);
2694 }
2695 /*
2696 * If at end of rpc data, get the eof boolean
2697 */
2698 if (!more_dirs) {
2699 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2700 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2701
2702 /*
2703 * kludge: if we got no entries, treat it as EOF.
2704 * some server sometimes send a reply without any
2705 * entries or EOF.
2706 * although it might mean the server has very long name,
2707 * we can't handle such entries anyway.
2708 */
2709
2710 if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2711 more_dirs = 0;
2712 }
2713 m_freem(mrep);
2714 }
2715 /*
2716 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2717 * by increasing d_reclen for the last record.
2718 */
2719 if (blksiz > 0) {
2720 left = NFS_DIRFRAGSIZ - blksiz;
2721 memset(uiop->uio_iov->iov_base, 0, left);
2722 dp->d_reclen += left;
2723 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2724 UIO_ADVANCE(uiop, left);
2725 }
2726
2727 /*
2728 * We are now either at the end of the directory or have filled the
2729 * block.
2730 */
2731 if (bigenough) {
2732 dnp->n_direofoffset = uiop->uio_offset;
2733 dnp->n_flag |= NEOFVALID;
2734 }
2735 nfsmout:
2736 return (error);
2737 }
2738
2739 #ifndef NFS_V2_ONLY
2740 /*
2741 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2742 */
2743 int
2744 nfs_readdirplusrpc(vp, uiop, cred)
2745 struct vnode *vp;
2746 struct uio *uiop;
2747 kauth_cred_t cred;
2748 {
2749 int len, left;
2750 struct dirent *dp = NULL;
2751 u_int32_t *tl;
2752 char *cp;
2753 int32_t t1, t2;
2754 struct vnode *newvp;
2755 char *bpos, *dpos, *cp2;
2756 struct mbuf *mreq, *mrep, *md, *mb;
2757 struct nameidata nami, *ndp = &nami;
2758 struct componentname *cnp = &ndp->ni_cnd;
2759 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2760 struct nfsnode *dnp = VTONFS(vp), *np;
2761 nfsfh_t *fhp;
2762 u_quad_t fileno;
2763 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2764 int attrflag, fhsize, nrpcs = 0, reclen;
2765 struct nfs_fattr fattr, *fp;
2766
2767 #ifdef DIAGNOSTIC
2768 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2769 panic("nfs readdirplusrpc bad uio");
2770 #endif
2771 ndp->ni_dvp = vp;
2772 newvp = NULLVP;
2773
2774 /*
2775 * Loop around doing readdir rpc's of size nm_readdirsize
2776 * truncated to a multiple of NFS_DIRFRAGSIZ.
2777 * The stopping criteria is EOF or buffer full.
2778 */
2779 while (more_dirs && bigenough) {
2780 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2781 bigenough = 0;
2782 break;
2783 }
2784 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2785 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2786 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2787 nfsm_fhtom(dnp, 1);
2788 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2789 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2790 txdr_swapcookie3(uiop->uio_offset, tl);
2791 } else {
2792 txdr_cookie3(uiop->uio_offset, tl);
2793 }
2794 tl += 2;
2795 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2796 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2797 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2798 *tl = txdr_unsigned(nmp->nm_rsize);
2799 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred);
2800 nfsm_postop_attr(vp, attrflag, 0);
2801 if (error) {
2802 m_freem(mrep);
2803 goto nfsmout;
2804 }
2805 nrpcs++;
2806 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2807 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2808 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2809 more_dirs = fxdr_unsigned(int, *tl);
2810
2811 /* loop thru the dir entries, doctoring them to 4bsd form */
2812 while (more_dirs && bigenough) {
2813 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2814 fileno = fxdr_hyper(tl);
2815 len = fxdr_unsigned(int, *(tl + 2));
2816 if (len <= 0 || len > NFS_MAXNAMLEN) {
2817 error = EBADRPC;
2818 m_freem(mrep);
2819 goto nfsmout;
2820 }
2821 /* for cookie stashing */
2822 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2823 left = NFS_DIRFRAGSIZ - blksiz;
2824 if (reclen > left) {
2825 /*
2826 * DIRFRAGSIZ is aligned, no need to align
2827 * again here.
2828 */
2829 memset(uiop->uio_iov->iov_base, 0, left);
2830 dp->d_reclen += left;
2831 UIO_ADVANCE(uiop, left);
2832 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2833 blksiz = 0;
2834 }
2835 if (reclen > uiop->uio_resid)
2836 bigenough = 0;
2837 if (bigenough) {
2838 int tlen;
2839
2840 dp = (struct dirent *)uiop->uio_iov->iov_base;
2841 dp->d_fileno = fileno;
2842 dp->d_namlen = len;
2843 dp->d_reclen = reclen;
2844 dp->d_type = DT_UNKNOWN;
2845 blksiz += reclen;
2846 if (blksiz == NFS_DIRFRAGSIZ)
2847 blksiz = 0;
2848 UIO_ADVANCE(uiop, DIRHDSIZ);
2849 nfsm_mtouio(uiop, len);
2850 tlen = reclen - (DIRHDSIZ + len);
2851 (void)memset(uiop->uio_iov->iov_base, 0, tlen);
2852 UIO_ADVANCE(uiop, tlen);
2853 cnp->cn_nameptr = dp->d_name;
2854 cnp->cn_namelen = dp->d_namlen;
2855 } else
2856 nfsm_adv(nfsm_rndup(len));
2857 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2858 if (bigenough) {
2859 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2860 uiop->uio_offset =
2861 fxdr_swapcookie3(tl);
2862 else
2863 uiop->uio_offset =
2864 fxdr_cookie3(tl);
2865 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2866 }
2867 tl += 2;
2868
2869 /*
2870 * Since the attributes are before the file handle
2871 * (sigh), we must skip over the attributes and then
2872 * come back and get them.
2873 */
2874 attrflag = fxdr_unsigned(int, *tl);
2875 if (attrflag) {
2876 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2877 memcpy(&fattr, fp, NFSX_V3FATTR);
2878 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2879 doit = fxdr_unsigned(int, *tl);
2880 if (doit) {
2881 nfsm_getfh(fhp, fhsize, 1);
2882 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2883 VREF(vp);
2884 newvp = vp;
2885 np = dnp;
2886 } else {
2887 error = nfs_nget1(vp->v_mount, fhp,
2888 fhsize, &np, LK_NOWAIT);
2889 if (!error)
2890 newvp = NFSTOV(np);
2891 }
2892 if (!error) {
2893 const char *xcp;
2894
2895 nfs_loadattrcache(&newvp, &fattr, 0, 0);
2896 if (bigenough) {
2897 dp->d_type =
2898 IFTODT(VTTOIF(np->n_vattr->va_type));
2899 if (cnp->cn_namelen <= NCHNAMLEN) {
2900 ndp->ni_vp = newvp;
2901 xcp = cnp->cn_nameptr +
2902 cnp->cn_namelen;
2903 cnp->cn_hash =
2904 namei_hash(cnp->cn_nameptr, &xcp);
2905 nfs_cache_enter(ndp->ni_dvp,
2906 ndp->ni_vp, cnp);
2907 }
2908 }
2909 }
2910 error = 0;
2911 }
2912 } else {
2913 /* Just skip over the file handle */
2914 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2915 i = fxdr_unsigned(int, *tl);
2916 nfsm_adv(nfsm_rndup(i));
2917 }
2918 if (newvp != NULLVP) {
2919 if (newvp == vp)
2920 vrele(newvp);
2921 else
2922 vput(newvp);
2923 newvp = NULLVP;
2924 }
2925 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2926 more_dirs = fxdr_unsigned(int, *tl);
2927 }
2928 /*
2929 * If at end of rpc data, get the eof boolean
2930 */
2931 if (!more_dirs) {
2932 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2933 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2934
2935 /*
2936 * kludge: see a comment in nfs_readdirrpc.
2937 */
2938
2939 if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2940 more_dirs = 0;
2941 }
2942 m_freem(mrep);
2943 }
2944 /*
2945 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2946 * by increasing d_reclen for the last record.
2947 */
2948 if (blksiz > 0) {
2949 left = NFS_DIRFRAGSIZ - blksiz;
2950 memset(uiop->uio_iov->iov_base, 0, left);
2951 dp->d_reclen += left;
2952 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2953 UIO_ADVANCE(uiop, left);
2954 }
2955
2956 /*
2957 * We are now either at the end of the directory or have filled the
2958 * block.
2959 */
2960 if (bigenough) {
2961 dnp->n_direofoffset = uiop->uio_offset;
2962 dnp->n_flag |= NEOFVALID;
2963 }
2964 nfsmout:
2965 if (newvp != NULLVP) {
2966 if(newvp == vp)
2967 vrele(newvp);
2968 else
2969 vput(newvp);
2970 }
2971 return (error);
2972 }
2973 #endif
2974
2975 /*
2976 * Silly rename. To make the NFS filesystem that is stateless look a little
2977 * more like the "ufs" a remove of an active vnode is translated to a rename
2978 * to a funny looking filename that is removed by nfs_inactive on the
2979 * nfsnode. There is the potential for another process on a different client
2980 * to create the same funny name between the nfs_lookitup() fails and the
2981 * nfs_rename() completes, but...
2982 */
2983 int
2984 nfs_sillyrename(dvp, vp, cnp, dolink)
2985 struct vnode *dvp, *vp;
2986 struct componentname *cnp;
2987 bool dolink;
2988 {
2989 struct sillyrename *sp;
2990 struct nfsnode *np;
2991 int error;
2992 short pid;
2993
2994 cache_purge(dvp);
2995 np = VTONFS(vp);
2996 #ifndef DIAGNOSTIC
2997 if (vp->v_type == VDIR)
2998 panic("nfs: sillyrename dir");
2999 #endif
3000 sp = kmem_alloc(sizeof(*sp), KM_SLEEP);
3001 sp->s_cred = kauth_cred_dup(cnp->cn_cred);
3002 sp->s_dvp = dvp;
3003 VREF(dvp);
3004
3005 /* Fudge together a funny name */
3006 pid = curlwp->l_proc->p_pid;
3007 memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
3008 sp->s_namlen = 12;
3009 sp->s_name[8] = hexdigits[pid & 0xf];
3010 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf];
3011 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf];
3012 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf];
3013
3014 /* Try lookitups until we get one that isn't there */
3015 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3016 curlwp, (struct nfsnode **)0) == 0) {
3017 sp->s_name[4]++;
3018 if (sp->s_name[4] > 'z') {
3019 error = EINVAL;
3020 goto bad;
3021 }
3022 }
3023 if (dolink) {
3024 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen,
3025 sp->s_cred, curlwp);
3026 /*
3027 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
3028 */
3029 if (error == ENOTSUP) {
3030 error = nfs_renameit(dvp, cnp, sp);
3031 }
3032 } else {
3033 error = nfs_renameit(dvp, cnp, sp);
3034 }
3035 if (error)
3036 goto bad;
3037 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3038 curlwp, &np);
3039 np->n_sillyrename = sp;
3040 return (0);
3041 bad:
3042 vrele(sp->s_dvp);
3043 kauth_cred_free(sp->s_cred);
3044 kmem_free(sp, sizeof(*sp));
3045 return (error);
3046 }
3047
3048 /*
3049 * Look up a file name and optionally either update the file handle or
3050 * allocate an nfsnode, depending on the value of npp.
3051 * npp == NULL --> just do the lookup
3052 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
3053 * handled too
3054 * *npp != NULL --> update the file handle in the vnode
3055 */
3056 int
3057 nfs_lookitup(dvp, name, len, cred, l, npp)
3058 struct vnode *dvp;
3059 const char *name;
3060 int len;
3061 kauth_cred_t cred;
3062 struct lwp *l;
3063 struct nfsnode **npp;
3064 {
3065 u_int32_t *tl;
3066 char *cp;
3067 int32_t t1, t2;
3068 struct vnode *newvp = (struct vnode *)0;
3069 struct nfsnode *np, *dnp = VTONFS(dvp);
3070 char *bpos, *dpos, *cp2;
3071 int error = 0, fhlen;
3072 #ifndef NFS_V2_ONLY
3073 int attrflag;
3074 #endif
3075 struct mbuf *mreq, *mrep, *md, *mb;
3076 nfsfh_t *nfhp;
3077 const int v3 = NFS_ISV3(dvp);
3078
3079 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
3080 nfsm_reqhead(dnp, NFSPROC_LOOKUP,
3081 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
3082 nfsm_fhtom(dnp, v3);
3083 nfsm_strtom(name, len, NFS_MAXNAMLEN);
3084 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred);
3085 if (npp && !error) {
3086 nfsm_getfh(nfhp, fhlen, v3);
3087 if (*npp) {
3088 np = *npp;
3089 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
3090 kmem_free(np->n_fhp, np->n_fhsize);
3091 np->n_fhp = &np->n_fh;
3092 }
3093 #if NFS_SMALLFH < NFSX_V3FHMAX
3094 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH)
3095 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP);
3096 #endif
3097 memcpy(np->n_fhp, nfhp, fhlen);
3098 np->n_fhsize = fhlen;
3099 newvp = NFSTOV(np);
3100 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
3101 VREF(dvp);
3102 newvp = dvp;
3103 np = dnp;
3104 } else {
3105 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
3106 if (error) {
3107 m_freem(mrep);
3108 return (error);
3109 }
3110 newvp = NFSTOV(np);
3111 }
3112 #ifndef NFS_V2_ONLY
3113 if (v3) {
3114 nfsm_postop_attr(newvp, attrflag, 0);
3115 if (!attrflag && *npp == NULL) {
3116 m_freem(mrep);
3117 vput(newvp);
3118 return (ENOENT);
3119 }
3120 } else
3121 #endif
3122 nfsm_loadattr(newvp, (struct vattr *)0, 0);
3123 }
3124 nfsm_reqdone;
3125 if (npp && *npp == NULL) {
3126 if (error) {
3127 if (newvp)
3128 vput(newvp);
3129 } else
3130 *npp = np;
3131 }
3132 return (error);
3133 }
3134
3135 #ifndef NFS_V2_ONLY
3136 /*
3137 * Nfs Version 3 commit rpc
3138 */
3139 int
3140 nfs_commit(vp, offset, cnt, l)
3141 struct vnode *vp;
3142 off_t offset;
3143 uint32_t cnt;
3144 struct lwp *l;
3145 {
3146 char *cp;
3147 u_int32_t *tl;
3148 int32_t t1, t2;
3149 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3150 char *bpos, *dpos, *cp2;
3151 int error = 0, wccflag = NFSV3_WCCRATTR;
3152 struct mbuf *mreq, *mrep, *md, *mb;
3153 struct nfsnode *np;
3154
3155 KASSERT(NFS_ISV3(vp));
3156
3157 #ifdef NFS_DEBUG_COMMIT
3158 printf("commit %lu - %lu\n", (unsigned long)offset,
3159 (unsigned long)(offset + cnt));
3160 #endif
3161
3162 mutex_enter(&nmp->nm_lock);
3163 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
3164 mutex_exit(&nmp->nm_lock);
3165 return (0);
3166 }
3167 mutex_exit(&nmp->nm_lock);
3168 nfsstats.rpccnt[NFSPROC_COMMIT]++;
3169 np = VTONFS(vp);
3170 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
3171 nfsm_fhtom(np, 1);
3172 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3173 txdr_hyper(offset, tl);
3174 tl += 2;
3175 *tl = txdr_unsigned(cnt);
3176 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred);
3177 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
3178 if (!error) {
3179 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
3180 mutex_enter(&nmp->nm_lock);
3181 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
3182 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
3183 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
3184 error = NFSERR_STALEWRITEVERF;
3185 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
3186 }
3187 mutex_exit(&nmp->nm_lock);
3188 }
3189 nfsm_reqdone;
3190 return (error);
3191 }
3192 #endif
3193
3194 /*
3195 * Kludge City..
3196 * - make nfs_bmap() essentially a no-op that does no translation
3197 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3198 * (Maybe I could use the process's page mapping, but I was concerned that
3199 * Kernel Write might not be enabled and also figured copyout() would do
3200 * a lot more work than memcpy() and also it currently happens in the
3201 * context of the swapper process (2).
3202 */
3203 int
3204 nfs_bmap(v)
3205 void *v;
3206 {
3207 struct vop_bmap_args /* {
3208 struct vnode *a_vp;
3209 daddr_t a_bn;
3210 struct vnode **a_vpp;
3211 daddr_t *a_bnp;
3212 int *a_runp;
3213 } */ *ap = v;
3214 struct vnode *vp = ap->a_vp;
3215 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
3216
3217 if (ap->a_vpp != NULL)
3218 *ap->a_vpp = vp;
3219 if (ap->a_bnp != NULL)
3220 *ap->a_bnp = ap->a_bn << bshift;
3221 if (ap->a_runp != NULL)
3222 *ap->a_runp = 1024 * 1024; /* XXX */
3223 return (0);
3224 }
3225
3226 /*
3227 * Strategy routine.
3228 * For async requests when nfsiod(s) are running, queue the request by
3229 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3230 * request.
3231 */
3232 int
3233 nfs_strategy(v)
3234 void *v;
3235 {
3236 struct vop_strategy_args *ap = v;
3237 struct buf *bp = ap->a_bp;
3238 int error = 0;
3239
3240 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3241 panic("nfs physio/async");
3242
3243 /*
3244 * If the op is asynchronous and an i/o daemon is waiting
3245 * queue the request, wake it up and wait for completion
3246 * otherwise just do it ourselves.
3247 */
3248 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp))
3249 error = nfs_doio(bp);
3250 return (error);
3251 }
3252
3253 /*
3254 * fsync vnode op. Just call nfs_flush() with commit == 1.
3255 */
3256 /* ARGSUSED */
3257 int
3258 nfs_fsync(v)
3259 void *v;
3260 {
3261 struct vop_fsync_args /* {
3262 struct vnodeop_desc *a_desc;
3263 struct vnode * a_vp;
3264 kauth_cred_t a_cred;
3265 int a_flags;
3266 off_t offlo;
3267 off_t offhi;
3268 struct lwp * a_l;
3269 } */ *ap = v;
3270
3271 struct vnode *vp = ap->a_vp;
3272
3273 if (vp->v_type != VREG)
3274 return 0;
3275
3276 return (nfs_flush(vp, ap->a_cred,
3277 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1));
3278 }
3279
3280 /*
3281 * Flush all the data associated with a vnode.
3282 */
3283 int
3284 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l,
3285 int commit)
3286 {
3287 struct nfsnode *np = VTONFS(vp);
3288 int error;
3289 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3290 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3291
3292 mutex_enter(&vp->v_interlock);
3293 error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3294 if (np->n_flag & NWRITEERR) {
3295 error = np->n_error;
3296 np->n_flag &= ~NWRITEERR;
3297 }
3298 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3299 return (error);
3300 }
3301
3302 /*
3303 * Return POSIX pathconf information applicable to nfs.
3304 *
3305 * N.B. The NFS V2 protocol doesn't support this RPC.
3306 */
3307 /* ARGSUSED */
3308 int
3309 nfs_pathconf(v)
3310 void *v;
3311 {
3312 struct vop_pathconf_args /* {
3313 struct vnode *a_vp;
3314 int a_name;
3315 register_t *a_retval;
3316 } */ *ap = v;
3317 struct nfsv3_pathconf *pcp;
3318 struct vnode *vp = ap->a_vp;
3319 struct mbuf *mreq, *mrep, *md, *mb;
3320 int32_t t1, t2;
3321 u_int32_t *tl;
3322 char *bpos, *dpos, *cp, *cp2;
3323 int error = 0, attrflag;
3324 #ifndef NFS_V2_ONLY
3325 struct nfsmount *nmp;
3326 unsigned int l;
3327 u_int64_t maxsize;
3328 #endif
3329 const int v3 = NFS_ISV3(vp);
3330 struct nfsnode *np = VTONFS(vp);
3331
3332 switch (ap->a_name) {
3333 /* Names that can be resolved locally. */
3334 case _PC_PIPE_BUF:
3335 *ap->a_retval = PIPE_BUF;
3336 break;
3337 case _PC_SYNC_IO:
3338 *ap->a_retval = 1;
3339 break;
3340 /* Names that cannot be resolved locally; do an RPC, if possible. */
3341 case _PC_LINK_MAX:
3342 case _PC_NAME_MAX:
3343 case _PC_CHOWN_RESTRICTED:
3344 case _PC_NO_TRUNC:
3345 if (!v3) {
3346 error = EINVAL;
3347 break;
3348 }
3349 nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3350 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3351 nfsm_fhtom(np, 1);
3352 nfsm_request(np, NFSPROC_PATHCONF,
3353 curlwp, curlwp->l_cred); /* XXX */
3354 nfsm_postop_attr(vp, attrflag, 0);
3355 if (!error) {
3356 nfsm_dissect(pcp, struct nfsv3_pathconf *,
3357 NFSX_V3PATHCONF);
3358 switch (ap->a_name) {
3359 case _PC_LINK_MAX:
3360 *ap->a_retval =
3361 fxdr_unsigned(register_t, pcp->pc_linkmax);
3362 break;
3363 case _PC_NAME_MAX:
3364 *ap->a_retval =
3365 fxdr_unsigned(register_t, pcp->pc_namemax);
3366 break;
3367 case _PC_CHOWN_RESTRICTED:
3368 *ap->a_retval =
3369 (pcp->pc_chownrestricted == nfs_true);
3370 break;
3371 case _PC_NO_TRUNC:
3372 *ap->a_retval =
3373 (pcp->pc_notrunc == nfs_true);
3374 break;
3375 }
3376 }
3377 nfsm_reqdone;
3378 break;
3379 case _PC_FILESIZEBITS:
3380 #ifndef NFS_V2_ONLY
3381 if (v3) {
3382 nmp = VFSTONFS(vp->v_mount);
3383 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3384 if ((error = nfs_fsinfo(nmp, vp,
3385 curlwp->l_cred, curlwp)) != 0) /* XXX */
3386 break;
3387 for (l = 0, maxsize = nmp->nm_maxfilesize;
3388 (maxsize >> l) > 0; l++)
3389 ;
3390 *ap->a_retval = l + 1;
3391 } else
3392 #endif
3393 {
3394 *ap->a_retval = 32; /* NFS V2 limitation */
3395 }
3396 break;
3397 default:
3398 error = EINVAL;
3399 break;
3400 }
3401
3402 return (error);
3403 }
3404
3405 /*
3406 * NFS advisory byte-level locks.
3407 */
3408 int
3409 nfs_advlock(v)
3410 void *v;
3411 {
3412 struct vop_advlock_args /* {
3413 struct vnode *a_vp;
3414 void *a_id;
3415 int a_op;
3416 struct flock *a_fl;
3417 int a_flags;
3418 } */ *ap = v;
3419 struct nfsnode *np = VTONFS(ap->a_vp);
3420
3421 return lf_advlock(ap, &np->n_lockf, np->n_size);
3422 }
3423
3424 /*
3425 * Print out the contents of an nfsnode.
3426 */
3427 int
3428 nfs_print(v)
3429 void *v;
3430 {
3431 struct vop_print_args /* {
3432 struct vnode *a_vp;
3433 } */ *ap = v;
3434 struct vnode *vp = ap->a_vp;
3435 struct nfsnode *np = VTONFS(vp);
3436
3437 printf("tag VT_NFS, fileid %lld fsid 0x%lx",
3438 (unsigned long long)np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3439 if (vp->v_type == VFIFO)
3440 fifo_printinfo(vp);
3441 printf("\n");
3442 return (0);
3443 }
3444
3445 /*
3446 * nfs unlock wrapper.
3447 */
3448 int
3449 nfs_unlock(void *v)
3450 {
3451 struct vop_unlock_args /* {
3452 struct vnode *a_vp;
3453 int a_flags;
3454 } */ *ap = v;
3455 struct vnode *vp = ap->a_vp;
3456
3457 /*
3458 * VOP_UNLOCK can be called by nfs_loadattrcache
3459 * with v_data == 0.
3460 */
3461 if (VTONFS(vp)) {
3462 nfs_delayedtruncate(vp);
3463 }
3464
3465 return genfs_unlock(v);
3466 }
3467
3468 /*
3469 * nfs special file access vnode op.
3470 * Essentially just get vattr and then imitate iaccess() since the device is
3471 * local to the client.
3472 */
3473 int
3474 nfsspec_access(v)
3475 void *v;
3476 {
3477 struct vop_access_args /* {
3478 struct vnode *a_vp;
3479 int a_mode;
3480 kauth_cred_t a_cred;
3481 struct lwp *a_l;
3482 } */ *ap = v;
3483 struct vattr va;
3484 struct vnode *vp = ap->a_vp;
3485 int error;
3486
3487 error = VOP_GETATTR(vp, &va, ap->a_cred);
3488 if (error)
3489 return (error);
3490
3491 /*
3492 * Disallow write attempts on filesystems mounted read-only;
3493 * unless the file is a socket, fifo, or a block or character
3494 * device resident on the filesystem.
3495 */
3496 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3497 switch (vp->v_type) {
3498 case VREG:
3499 case VDIR:
3500 case VLNK:
3501 return (EROFS);
3502 default:
3503 break;
3504 }
3505 }
3506
3507 return (vaccess(va.va_type, va.va_mode,
3508 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3509 }
3510
3511 /*
3512 * Read wrapper for special devices.
3513 */
3514 int
3515 nfsspec_read(v)
3516 void *v;
3517 {
3518 struct vop_read_args /* {
3519 struct vnode *a_vp;
3520 struct uio *a_uio;
3521 int a_ioflag;
3522 kauth_cred_t a_cred;
3523 } */ *ap = v;
3524 struct nfsnode *np = VTONFS(ap->a_vp);
3525
3526 /*
3527 * Set access flag.
3528 */
3529 np->n_flag |= NACC;
3530 getnanotime(&np->n_atim);
3531 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3532 }
3533
3534 /*
3535 * Write wrapper for special devices.
3536 */
3537 int
3538 nfsspec_write(v)
3539 void *v;
3540 {
3541 struct vop_write_args /* {
3542 struct vnode *a_vp;
3543 struct uio *a_uio;
3544 int a_ioflag;
3545 kauth_cred_t a_cred;
3546 } */ *ap = v;
3547 struct nfsnode *np = VTONFS(ap->a_vp);
3548
3549 /*
3550 * Set update flag.
3551 */
3552 np->n_flag |= NUPD;
3553 getnanotime(&np->n_mtim);
3554 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3555 }
3556
3557 /*
3558 * Close wrapper for special devices.
3559 *
3560 * Update the times on the nfsnode then do device close.
3561 */
3562 int
3563 nfsspec_close(v)
3564 void *v;
3565 {
3566 struct vop_close_args /* {
3567 struct vnode *a_vp;
3568 int a_fflag;
3569 kauth_cred_t a_cred;
3570 struct lwp *a_l;
3571 } */ *ap = v;
3572 struct vnode *vp = ap->a_vp;
3573 struct nfsnode *np = VTONFS(vp);
3574 struct vattr vattr;
3575
3576 if (np->n_flag & (NACC | NUPD)) {
3577 np->n_flag |= NCHG;
3578 if (vp->v_usecount == 1 &&
3579 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3580 VATTR_NULL(&vattr);
3581 if (np->n_flag & NACC)
3582 vattr.va_atime = np->n_atim;
3583 if (np->n_flag & NUPD)
3584 vattr.va_mtime = np->n_mtim;
3585 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3586 }
3587 }
3588 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3589 }
3590
3591 /*
3592 * Read wrapper for fifos.
3593 */
3594 int
3595 nfsfifo_read(v)
3596 void *v;
3597 {
3598 struct vop_read_args /* {
3599 struct vnode *a_vp;
3600 struct uio *a_uio;
3601 int a_ioflag;
3602 kauth_cred_t a_cred;
3603 } */ *ap = v;
3604 struct nfsnode *np = VTONFS(ap->a_vp);
3605
3606 /*
3607 * Set access flag.
3608 */
3609 np->n_flag |= NACC;
3610 getnanotime(&np->n_atim);
3611 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3612 }
3613
3614 /*
3615 * Write wrapper for fifos.
3616 */
3617 int
3618 nfsfifo_write(v)
3619 void *v;
3620 {
3621 struct vop_write_args /* {
3622 struct vnode *a_vp;
3623 struct uio *a_uio;
3624 int a_ioflag;
3625 kauth_cred_t a_cred;
3626 } */ *ap = v;
3627 struct nfsnode *np = VTONFS(ap->a_vp);
3628
3629 /*
3630 * Set update flag.
3631 */
3632 np->n_flag |= NUPD;
3633 getnanotime(&np->n_mtim);
3634 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3635 }
3636
3637 /*
3638 * Close wrapper for fifos.
3639 *
3640 * Update the times on the nfsnode then do fifo close.
3641 */
3642 int
3643 nfsfifo_close(v)
3644 void *v;
3645 {
3646 struct vop_close_args /* {
3647 struct vnode *a_vp;
3648 int a_fflag;
3649 kauth_cred_t a_cred;
3650 struct lwp *a_l;
3651 } */ *ap = v;
3652 struct vnode *vp = ap->a_vp;
3653 struct nfsnode *np = VTONFS(vp);
3654 struct vattr vattr;
3655
3656 if (np->n_flag & (NACC | NUPD)) {
3657 struct timespec ts;
3658
3659 getnanotime(&ts);
3660 if (np->n_flag & NACC)
3661 np->n_atim = ts;
3662 if (np->n_flag & NUPD)
3663 np->n_mtim = ts;
3664 np->n_flag |= NCHG;
3665 if (vp->v_usecount == 1 &&
3666 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3667 VATTR_NULL(&vattr);
3668 if (np->n_flag & NACC)
3669 vattr.va_atime = np->n_atim;
3670 if (np->n_flag & NUPD)
3671 vattr.va_mtime = np->n_mtim;
3672 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3673 }
3674 }
3675 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3676 }
3677