FreeBSD/Linux Kernel Cross Reference
sys/nfs/nfs_vnops.c
1 /* $NetBSD: nfs_vnops.c,v 1.267 2008/10/15 14:21:35 pooka 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 1.267 2008/10/15 14:21:35 pooka Exp $");
43
44 #include "opt_nfs.h"
45 #include "opt_uvmhist.h"
46
47 #include <sys/param.h>
48 #include <sys/proc.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/resourcevar.h>
52 #include <sys/mount.h>
53 #include <sys/buf.h>
54 #include <sys/condvar.h>
55 #include <sys/disk.h>
56 #include <sys/malloc.h>
57 #include <sys/kmem.h>
58 #include <sys/mbuf.h>
59 #include <sys/mutex.h>
60 #include <sys/namei.h>
61 #include <sys/vnode.h>
62 #include <sys/dirent.h>
63 #include <sys/fcntl.h>
64 #include <sys/hash.h>
65 #include <sys/lockf.h>
66 #include <sys/stat.h>
67 #include <sys/unistd.h>
68 #include <sys/kauth.h>
69
70 #include <uvm/uvm_extern.h>
71 #include <uvm/uvm.h>
72
73 #include <miscfs/fifofs/fifo.h>
74 #include <miscfs/genfs/genfs.h>
75 #include <miscfs/genfs/genfs_node.h>
76 #include <miscfs/specfs/specdev.h>
77
78 #include <nfs/rpcv2.h>
79 #include <nfs/nfsproto.h>
80 #include <nfs/nfs.h>
81 #include <nfs/nfsnode.h>
82 #include <nfs/nfsmount.h>
83 #include <nfs/xdr_subs.h>
84 #include <nfs/nfsm_subs.h>
85 #include <nfs/nfs_var.h>
86
87 #include <net/if.h>
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
90
91 /*
92 * Global vfs data structures for nfs
93 */
94 int (**nfsv2_vnodeop_p) __P((void *));
95 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
96 { &vop_default_desc, vn_default_error },
97 { &vop_lookup_desc, nfs_lookup }, /* lookup */
98 { &vop_create_desc, nfs_create }, /* create */
99 { &vop_mknod_desc, nfs_mknod }, /* mknod */
100 { &vop_open_desc, nfs_open }, /* open */
101 { &vop_close_desc, nfs_close }, /* close */
102 { &vop_access_desc, nfs_access }, /* access */
103 { &vop_getattr_desc, nfs_getattr }, /* getattr */
104 { &vop_setattr_desc, nfs_setattr }, /* setattr */
105 { &vop_read_desc, nfs_read }, /* read */
106 { &vop_write_desc, nfs_write }, /* write */
107 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
108 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */
109 { &vop_poll_desc, nfs_poll }, /* poll */
110 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */
111 { &vop_revoke_desc, nfs_revoke }, /* revoke */
112 { &vop_mmap_desc, nfs_mmap }, /* mmap */
113 { &vop_fsync_desc, nfs_fsync }, /* fsync */
114 { &vop_seek_desc, nfs_seek }, /* seek */
115 { &vop_remove_desc, nfs_remove }, /* remove */
116 { &vop_link_desc, nfs_link }, /* link */
117 { &vop_rename_desc, nfs_rename }, /* rename */
118 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */
119 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */
120 { &vop_symlink_desc, nfs_symlink }, /* symlink */
121 { &vop_readdir_desc, nfs_readdir }, /* readdir */
122 { &vop_readlink_desc, nfs_readlink }, /* readlink */
123 { &vop_abortop_desc, nfs_abortop }, /* abortop */
124 { &vop_inactive_desc, nfs_inactive }, /* inactive */
125 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
126 { &vop_lock_desc, nfs_lock }, /* lock */
127 { &vop_unlock_desc, nfs_unlock }, /* unlock */
128 { &vop_bmap_desc, nfs_bmap }, /* bmap */
129 { &vop_strategy_desc, nfs_strategy }, /* strategy */
130 { &vop_print_desc, nfs_print }, /* print */
131 { &vop_islocked_desc, nfs_islocked }, /* islocked */
132 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */
133 { &vop_advlock_desc, nfs_advlock }, /* advlock */
134 { &vop_bwrite_desc, genfs_badop }, /* bwrite */
135 { &vop_getpages_desc, nfs_getpages }, /* getpages */
136 { &vop_putpages_desc, genfs_putpages }, /* putpages */
137 { NULL, NULL }
138 };
139 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
140 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
141
142 /*
143 * Special device vnode ops
144 */
145 int (**spec_nfsv2nodeop_p) __P((void *));
146 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
147 { &vop_default_desc, vn_default_error },
148 { &vop_lookup_desc, spec_lookup }, /* lookup */
149 { &vop_create_desc, spec_create }, /* create */
150 { &vop_mknod_desc, spec_mknod }, /* mknod */
151 { &vop_open_desc, spec_open }, /* open */
152 { &vop_close_desc, nfsspec_close }, /* close */
153 { &vop_access_desc, nfsspec_access }, /* access */
154 { &vop_getattr_desc, nfs_getattr }, /* getattr */
155 { &vop_setattr_desc, nfs_setattr }, /* setattr */
156 { &vop_read_desc, nfsspec_read }, /* read */
157 { &vop_write_desc, nfsspec_write }, /* write */
158 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
159 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */
160 { &vop_poll_desc, spec_poll }, /* poll */
161 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */
162 { &vop_revoke_desc, spec_revoke }, /* revoke */
163 { &vop_mmap_desc, spec_mmap }, /* mmap */
164 { &vop_fsync_desc, spec_fsync }, /* fsync */
165 { &vop_seek_desc, spec_seek }, /* seek */
166 { &vop_remove_desc, spec_remove }, /* remove */
167 { &vop_link_desc, spec_link }, /* link */
168 { &vop_rename_desc, spec_rename }, /* rename */
169 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */
170 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */
171 { &vop_symlink_desc, spec_symlink }, /* symlink */
172 { &vop_readdir_desc, spec_readdir }, /* readdir */
173 { &vop_readlink_desc, spec_readlink }, /* readlink */
174 { &vop_abortop_desc, spec_abortop }, /* abortop */
175 { &vop_inactive_desc, nfs_inactive }, /* inactive */
176 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
177 { &vop_lock_desc, nfs_lock }, /* lock */
178 { &vop_unlock_desc, nfs_unlock }, /* unlock */
179 { &vop_bmap_desc, spec_bmap }, /* bmap */
180 { &vop_strategy_desc, spec_strategy }, /* strategy */
181 { &vop_print_desc, nfs_print }, /* print */
182 { &vop_islocked_desc, nfs_islocked }, /* islocked */
183 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */
184 { &vop_advlock_desc, spec_advlock }, /* advlock */
185 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */
186 { &vop_getpages_desc, spec_getpages }, /* getpages */
187 { &vop_putpages_desc, spec_putpages }, /* putpages */
188 { NULL, NULL }
189 };
190 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
191 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
192
193 int (**fifo_nfsv2nodeop_p) __P((void *));
194 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
195 { &vop_default_desc, vn_default_error },
196 { &vop_lookup_desc, fifo_lookup }, /* lookup */
197 { &vop_create_desc, fifo_create }, /* create */
198 { &vop_mknod_desc, fifo_mknod }, /* mknod */
199 { &vop_open_desc, fifo_open }, /* open */
200 { &vop_close_desc, nfsfifo_close }, /* close */
201 { &vop_access_desc, nfsspec_access }, /* access */
202 { &vop_getattr_desc, nfs_getattr }, /* getattr */
203 { &vop_setattr_desc, nfs_setattr }, /* setattr */
204 { &vop_read_desc, nfsfifo_read }, /* read */
205 { &vop_write_desc, nfsfifo_write }, /* write */
206 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
207 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */
208 { &vop_poll_desc, fifo_poll }, /* poll */
209 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */
210 { &vop_revoke_desc, fifo_revoke }, /* revoke */
211 { &vop_mmap_desc, fifo_mmap }, /* mmap */
212 { &vop_fsync_desc, nfs_fsync }, /* fsync */
213 { &vop_seek_desc, fifo_seek }, /* seek */
214 { &vop_remove_desc, fifo_remove }, /* remove */
215 { &vop_link_desc, fifo_link }, /* link */
216 { &vop_rename_desc, fifo_rename }, /* rename */
217 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */
218 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */
219 { &vop_symlink_desc, fifo_symlink }, /* symlink */
220 { &vop_readdir_desc, fifo_readdir }, /* readdir */
221 { &vop_readlink_desc, fifo_readlink }, /* readlink */
222 { &vop_abortop_desc, fifo_abortop }, /* abortop */
223 { &vop_inactive_desc, nfs_inactive }, /* inactive */
224 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
225 { &vop_lock_desc, nfs_lock }, /* lock */
226 { &vop_unlock_desc, nfs_unlock }, /* unlock */
227 { &vop_bmap_desc, fifo_bmap }, /* bmap */
228 { &vop_strategy_desc, genfs_badop }, /* strategy */
229 { &vop_print_desc, nfs_print }, /* print */
230 { &vop_islocked_desc, nfs_islocked }, /* islocked */
231 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */
232 { &vop_advlock_desc, fifo_advlock }, /* advlock */
233 { &vop_bwrite_desc, genfs_badop }, /* bwrite */
234 { &vop_putpages_desc, fifo_putpages }, /* putpages */
235 { NULL, NULL }
236 };
237 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
238 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
239
240 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *,
241 size_t, kauth_cred_t, struct lwp *);
242 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *);
243
244 /*
245 * Global variables
246 */
247 extern u_int32_t nfs_true, nfs_false;
248 extern u_int32_t nfs_xdrneg1;
249 extern const nfstype nfsv3_type[9];
250
251 int nfs_numasync = 0;
252 #define DIRHDSIZ _DIRENT_NAMEOFF(dp)
253 #define UIO_ADVANCE(uio, siz) \
254 (void)((uio)->uio_resid -= (siz), \
255 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \
256 (uio)->uio_iov->iov_len -= (siz))
257
258 static void nfs_cache_enter(struct vnode *, struct vnode *,
259 struct componentname *);
260
261 static void
262 nfs_cache_enter(struct vnode *dvp, struct vnode *vp,
263 struct componentname *cnp)
264 {
265 struct nfsnode *dnp = VTONFS(dvp);
266
267 if (vp != NULL) {
268 struct nfsnode *np = VTONFS(vp);
269
270 np->n_ctime = np->n_vattr->va_ctime.tv_sec;
271 }
272
273 if (!timespecisset(&dnp->n_nctime))
274 dnp->n_nctime = dnp->n_vattr->va_mtime;
275
276 cache_enter(dvp, vp, cnp);
277 }
278
279 /*
280 * nfs null call from vfs.
281 */
282 int
283 nfs_null(vp, cred, l)
284 struct vnode *vp;
285 kauth_cred_t cred;
286 struct lwp *l;
287 {
288 char *bpos, *dpos;
289 int error = 0;
290 struct mbuf *mreq, *mrep, *md, *mb;
291 struct nfsnode *np = VTONFS(vp);
292
293 nfsm_reqhead(np, NFSPROC_NULL, 0);
294 nfsm_request(np, NFSPROC_NULL, l, cred);
295 nfsm_reqdone;
296 return (error);
297 }
298
299 /*
300 * nfs access vnode op.
301 * For nfs version 2, just return ok. File accesses may fail later.
302 * For nfs version 3, use the access rpc to check accessibility. If file modes
303 * are changed on the server, accesses might still fail later.
304 */
305 int
306 nfs_access(v)
307 void *v;
308 {
309 struct vop_access_args /* {
310 struct vnode *a_vp;
311 int a_mode;
312 kauth_cred_t a_cred;
313 } */ *ap = v;
314 struct vnode *vp = ap->a_vp;
315 #ifndef NFS_V2_ONLY
316 u_int32_t *tl;
317 char *cp;
318 int32_t t1, t2;
319 char *bpos, *dpos, *cp2;
320 int error = 0, attrflag;
321 struct mbuf *mreq, *mrep, *md, *mb;
322 u_int32_t mode, rmode;
323 const int v3 = NFS_ISV3(vp);
324 #endif
325 int cachevalid;
326 struct nfsnode *np = VTONFS(vp);
327 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
328
329 cachevalid = (np->n_accstamp != -1 &&
330 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) &&
331 np->n_accuid == kauth_cred_geteuid(ap->a_cred));
332
333 /*
334 * Check access cache first. If this request has been made for this
335 * uid shortly before, use the cached result.
336 */
337 if (cachevalid) {
338 if (!np->n_accerror) {
339 if ((np->n_accmode & ap->a_mode) == ap->a_mode)
340 return np->n_accerror;
341 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
342 return np->n_accerror;
343 }
344
345 #ifndef NFS_V2_ONLY
346 /*
347 * For nfs v3, do an access rpc, otherwise you are stuck emulating
348 * ufs_access() locally using the vattr. This may not be correct,
349 * since the server may apply other access criteria such as
350 * client uid-->server uid mapping that we do not know about, but
351 * this is better than just returning anything that is lying about
352 * in the cache.
353 */
354 if (v3) {
355 nfsstats.rpccnt[NFSPROC_ACCESS]++;
356 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
357 nfsm_fhtom(np, v3);
358 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
359 if (ap->a_mode & VREAD)
360 mode = NFSV3ACCESS_READ;
361 else
362 mode = 0;
363 if (vp->v_type != VDIR) {
364 if (ap->a_mode & VWRITE)
365 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
366 if (ap->a_mode & VEXEC)
367 mode |= NFSV3ACCESS_EXECUTE;
368 } else {
369 if (ap->a_mode & VWRITE)
370 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
371 NFSV3ACCESS_DELETE);
372 if (ap->a_mode & VEXEC)
373 mode |= NFSV3ACCESS_LOOKUP;
374 }
375 *tl = txdr_unsigned(mode);
376 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred);
377 nfsm_postop_attr(vp, attrflag, 0);
378 if (!error) {
379 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
380 rmode = fxdr_unsigned(u_int32_t, *tl);
381 /*
382 * The NFS V3 spec does not clarify whether or not
383 * the returned access bits can be a superset of
384 * the ones requested, so...
385 */
386 if ((rmode & mode) != mode)
387 error = EACCES;
388 }
389 nfsm_reqdone;
390 } else
391 #endif
392 return (nfsspec_access(ap));
393 #ifndef NFS_V2_ONLY
394 /*
395 * Disallow write attempts on filesystems mounted read-only;
396 * unless the file is a socket, fifo, or a block or character
397 * device resident on the filesystem.
398 */
399 if (!error && (ap->a_mode & VWRITE) &&
400 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
401 switch (vp->v_type) {
402 case VREG:
403 case VDIR:
404 case VLNK:
405 error = EROFS;
406 default:
407 break;
408 }
409 }
410
411 if (!error || error == EACCES) {
412 /*
413 * If we got the same result as for a previous,
414 * different request, OR it in. Don't update
415 * the timestamp in that case.
416 */
417 if (cachevalid && np->n_accstamp != -1 &&
418 error == np->n_accerror) {
419 if (!error)
420 np->n_accmode |= ap->a_mode;
421 else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
422 np->n_accmode = ap->a_mode;
423 } else {
424 np->n_accstamp = time_uptime;
425 np->n_accuid = kauth_cred_geteuid(ap->a_cred);
426 np->n_accmode = ap->a_mode;
427 np->n_accerror = error;
428 }
429 }
430
431 return (error);
432 #endif
433 }
434
435 /*
436 * nfs open vnode op
437 * Check to see if the type is ok
438 * and that deletion is not in progress.
439 * For paged in text files, you will need to flush the page cache
440 * if consistency is lost.
441 */
442 /* ARGSUSED */
443 int
444 nfs_open(v)
445 void *v;
446 {
447 struct vop_open_args /* {
448 struct vnode *a_vp;
449 int a_mode;
450 kauth_cred_t a_cred;
451 } */ *ap = v;
452 struct vnode *vp = ap->a_vp;
453 struct nfsnode *np = VTONFS(vp);
454 int error;
455
456 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
457 return (EACCES);
458 }
459
460 if (ap->a_mode & FREAD) {
461 if (np->n_rcred != NULL)
462 kauth_cred_free(np->n_rcred);
463 np->n_rcred = ap->a_cred;
464 kauth_cred_hold(np->n_rcred);
465 }
466 if (ap->a_mode & FWRITE) {
467 if (np->n_wcred != NULL)
468 kauth_cred_free(np->n_wcred);
469 np->n_wcred = ap->a_cred;
470 kauth_cred_hold(np->n_wcred);
471 }
472
473 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0);
474 if (error)
475 return error;
476
477 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */
478
479 return (0);
480 }
481
482 /*
483 * nfs close vnode op
484 * What an NFS client should do upon close after writing is a debatable issue.
485 * Most NFS clients push delayed writes to the server upon close, basically for
486 * two reasons:
487 * 1 - So that any write errors may be reported back to the client process
488 * doing the close system call. By far the two most likely errors are
489 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
490 * 2 - To put a worst case upper bound on cache inconsistency between
491 * multiple clients for the file.
492 * There is also a consistency problem for Version 2 of the protocol w.r.t.
493 * not being able to tell if other clients are writing a file concurrently,
494 * since there is no way of knowing if the changed modify time in the reply
495 * is only due to the write for this client.
496 * (NFS Version 3 provides weak cache consistency data in the reply that
497 * should be sufficient to detect and handle this case.)
498 *
499 * The current code does the following:
500 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
501 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
502 * or commit them (this satisfies 1 and 2 except for the
503 * case where the server crashes after this close but
504 * before the commit RPC, which is felt to be "good
505 * enough". Changing the last argument to nfs_flush() to
506 * a 1 would force a commit operation, if it is felt a
507 * commit is necessary now.
508 */
509 /* ARGSUSED */
510 int
511 nfs_close(v)
512 void *v;
513 {
514 struct vop_close_args /* {
515 struct vnodeop_desc *a_desc;
516 struct vnode *a_vp;
517 int a_fflag;
518 kauth_cred_t a_cred;
519 } */ *ap = v;
520 struct vnode *vp = ap->a_vp;
521 struct nfsnode *np = VTONFS(vp);
522 int error = 0;
523 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
524
525 if (vp->v_type == VREG) {
526 if (np->n_flag & NMODIFIED) {
527 #ifndef NFS_V2_ONLY
528 if (NFS_ISV3(vp)) {
529 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0);
530 np->n_flag &= ~NMODIFIED;
531 } else
532 #endif
533 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1);
534 NFS_INVALIDATE_ATTRCACHE(np);
535 }
536 if (np->n_flag & NWRITEERR) {
537 np->n_flag &= ~NWRITEERR;
538 error = np->n_error;
539 }
540 }
541 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
542 return (error);
543 }
544
545 /*
546 * nfs getattr call from vfs.
547 */
548 int
549 nfs_getattr(v)
550 void *v;
551 {
552 struct vop_getattr_args /* {
553 struct vnode *a_vp;
554 struct vattr *a_vap;
555 kauth_cred_t a_cred;
556 } */ *ap = v;
557 struct vnode *vp = ap->a_vp;
558 struct nfsnode *np = VTONFS(vp);
559 char *cp;
560 u_int32_t *tl;
561 int32_t t1, t2;
562 char *bpos, *dpos;
563 int error = 0;
564 struct mbuf *mreq, *mrep, *md, *mb;
565 const int v3 = NFS_ISV3(vp);
566
567 /*
568 * Update local times for special files.
569 */
570 if (np->n_flag & (NACC | NUPD))
571 np->n_flag |= NCHG;
572
573 /*
574 * if we have delayed truncation, do it now.
575 */
576 nfs_delayedtruncate(vp);
577
578 /*
579 * First look in the cache.
580 */
581 if (nfs_getattrcache(vp, ap->a_vap) == 0)
582 return (0);
583 nfsstats.rpccnt[NFSPROC_GETATTR]++;
584 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
585 nfsm_fhtom(np, v3);
586 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred);
587 if (!error) {
588 nfsm_loadattr(vp, ap->a_vap, 0);
589 if (vp->v_type == VDIR &&
590 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
591 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
592 }
593 nfsm_reqdone;
594 return (error);
595 }
596
597 /*
598 * nfs setattr call.
599 */
600 int
601 nfs_setattr(v)
602 void *v;
603 {
604 struct vop_setattr_args /* {
605 struct vnodeop_desc *a_desc;
606 struct vnode *a_vp;
607 struct vattr *a_vap;
608 kauth_cred_t a_cred;
609 } */ *ap = v;
610 struct vnode *vp = ap->a_vp;
611 struct nfsnode *np = VTONFS(vp);
612 struct vattr *vap = ap->a_vap;
613 int error = 0;
614 u_quad_t tsize = 0;
615
616 /*
617 * Setting of flags is not supported.
618 */
619 if (vap->va_flags != VNOVAL)
620 return (EOPNOTSUPP);
621
622 /*
623 * Disallow write attempts if the filesystem is mounted read-only.
624 */
625 if ((vap->va_uid != (uid_t)VNOVAL ||
626 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
627 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
628 (vp->v_mount->mnt_flag & MNT_RDONLY))
629 return (EROFS);
630 if (vap->va_size != VNOVAL) {
631 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) {
632 return EFBIG;
633 }
634 switch (vp->v_type) {
635 case VDIR:
636 return (EISDIR);
637 case VCHR:
638 case VBLK:
639 case VSOCK:
640 case VFIFO:
641 if (vap->va_mtime.tv_sec == VNOVAL &&
642 vap->va_atime.tv_sec == VNOVAL &&
643 vap->va_mode == (mode_t)VNOVAL &&
644 vap->va_uid == (uid_t)VNOVAL &&
645 vap->va_gid == (gid_t)VNOVAL)
646 return (0);
647 vap->va_size = VNOVAL;
648 break;
649 default:
650 /*
651 * Disallow write attempts if the filesystem is
652 * mounted read-only.
653 */
654 if (vp->v_mount->mnt_flag & MNT_RDONLY)
655 return (EROFS);
656 genfs_node_wrlock(vp);
657 uvm_vnp_setsize(vp, vap->va_size);
658 tsize = np->n_size;
659 np->n_size = vap->va_size;
660 if (vap->va_size == 0)
661 error = nfs_vinvalbuf(vp, 0,
662 ap->a_cred, curlwp, 1);
663 else
664 error = nfs_vinvalbuf(vp, V_SAVE,
665 ap->a_cred, curlwp, 1);
666 if (error) {
667 uvm_vnp_setsize(vp, tsize);
668 genfs_node_unlock(vp);
669 return (error);
670 }
671 np->n_vattr->va_size = vap->va_size;
672 }
673 } else {
674 /*
675 * flush files before setattr because a later write of
676 * cached data might change timestamps or reset sugid bits
677 */
678 if ((vap->va_mtime.tv_sec != VNOVAL ||
679 vap->va_atime.tv_sec != VNOVAL ||
680 vap->va_mode != VNOVAL) &&
681 vp->v_type == VREG &&
682 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
683 curlwp, 1)) == EINTR)
684 return (error);
685 }
686 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp);
687 if (vap->va_size != VNOVAL) {
688 if (error) {
689 np->n_size = np->n_vattr->va_size = tsize;
690 uvm_vnp_setsize(vp, np->n_size);
691 }
692 genfs_node_unlock(vp);
693 }
694 VN_KNOTE(vp, NOTE_ATTRIB);
695 return (error);
696 }
697
698 /*
699 * Do an nfs setattr rpc.
700 */
701 int
702 nfs_setattrrpc(vp, vap, cred, l)
703 struct vnode *vp;
704 struct vattr *vap;
705 kauth_cred_t cred;
706 struct lwp *l;
707 {
708 struct nfsv2_sattr *sp;
709 char *cp;
710 int32_t t1, t2;
711 char *bpos, *dpos;
712 u_int32_t *tl;
713 int error = 0;
714 struct mbuf *mreq, *mrep, *md, *mb;
715 const int v3 = NFS_ISV3(vp);
716 struct nfsnode *np = VTONFS(vp);
717 #ifndef NFS_V2_ONLY
718 int wccflag = NFSV3_WCCRATTR;
719 char *cp2;
720 #endif
721
722 nfsstats.rpccnt[NFSPROC_SETATTR]++;
723 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
724 nfsm_fhtom(np, v3);
725 #ifndef NFS_V2_ONLY
726 if (v3) {
727 nfsm_v3attrbuild(vap, true);
728 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
729 *tl = nfs_false;
730 } else {
731 #endif
732 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
733 if (vap->va_mode == (mode_t)VNOVAL)
734 sp->sa_mode = nfs_xdrneg1;
735 else
736 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
737 if (vap->va_uid == (uid_t)VNOVAL)
738 sp->sa_uid = nfs_xdrneg1;
739 else
740 sp->sa_uid = txdr_unsigned(vap->va_uid);
741 if (vap->va_gid == (gid_t)VNOVAL)
742 sp->sa_gid = nfs_xdrneg1;
743 else
744 sp->sa_gid = txdr_unsigned(vap->va_gid);
745 sp->sa_size = txdr_unsigned(vap->va_size);
746 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
747 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
748 #ifndef NFS_V2_ONLY
749 }
750 #endif
751 nfsm_request(np, NFSPROC_SETATTR, l, cred);
752 #ifndef NFS_V2_ONLY
753 if (v3) {
754 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
755 } else
756 #endif
757 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
758 nfsm_reqdone;
759 return (error);
760 }
761
762 /*
763 * nfs lookup call, one step at a time...
764 * First look in cache
765 * If not found, unlock the directory nfsnode and do the rpc
766 *
767 * This code is full of lock/unlock statements and checks, because
768 * we continue after cache_lookup has finished (we need to check
769 * with the attr cache and do an rpc if it has timed out). This means
770 * that the locking effects of cache_lookup have to be taken into
771 * account.
772 */
773 int
774 nfs_lookup(v)
775 void *v;
776 {
777 struct vop_lookup_args /* {
778 struct vnodeop_desc *a_desc;
779 struct vnode *a_dvp;
780 struct vnode **a_vpp;
781 struct componentname *a_cnp;
782 } */ *ap = v;
783 struct componentname *cnp = ap->a_cnp;
784 struct vnode *dvp = ap->a_dvp;
785 struct vnode **vpp = ap->a_vpp;
786 int flags;
787 struct vnode *newvp;
788 u_int32_t *tl;
789 char *cp;
790 int32_t t1, t2;
791 char *bpos, *dpos, *cp2;
792 struct mbuf *mreq, *mrep, *md, *mb;
793 long len;
794 nfsfh_t *fhp;
795 struct nfsnode *np;
796 int error = 0, attrflag, fhsize;
797 const int v3 = NFS_ISV3(dvp);
798
799 flags = cnp->cn_flags;
800
801 *vpp = NULLVP;
802 newvp = NULLVP;
803 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
804 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
805 return (EROFS);
806 if (dvp->v_type != VDIR)
807 return (ENOTDIR);
808
809 /*
810 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves.
811 */
812 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
813 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
814 if (error)
815 return error;
816 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN))
817 return EISDIR;
818 VREF(dvp);
819 *vpp = dvp;
820 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
821 cnp->cn_flags |= SAVENAME;
822 return 0;
823 }
824
825 np = VTONFS(dvp);
826
827 /*
828 * Before tediously performing a linear scan of the directory,
829 * check the name cache to see if the directory/name pair
830 * we are looking for is known already.
831 * If the directory/name pair is found in the name cache,
832 * we have to ensure the directory has not changed from
833 * the time the cache entry has been created. If it has,
834 * the cache entry has to be ignored.
835 */
836 error = cache_lookup_raw(dvp, vpp, cnp);
837 KASSERT(dvp != *vpp);
838 if (error >= 0) {
839 struct vattr vattr;
840 int err2;
841
842 if (error && error != ENOENT) {
843 *vpp = NULLVP;
844 return error;
845 }
846
847 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
848 if (err2 != 0) {
849 if (error == 0)
850 vrele(*vpp);
851 *vpp = NULLVP;
852 return err2;
853 }
854
855 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred)
856 || timespeccmp(&vattr.va_mtime,
857 &VTONFS(dvp)->n_nctime, !=)) {
858 if (error == 0) {
859 vrele(*vpp);
860 *vpp = NULLVP;
861 }
862 cache_purge1(dvp, NULL, PURGE_CHILDREN);
863 timespecclear(&np->n_nctime);
864 goto dorpc;
865 }
866
867 if (error == ENOENT) {
868 goto noentry;
869 }
870
871 newvp = *vpp;
872 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
873 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
874 nfsstats.lookupcache_hits++;
875 if ((flags & ISDOTDOT) != 0) {
876 VOP_UNLOCK(dvp, 0);
877 }
878 error = vn_lock(newvp, LK_EXCLUSIVE);
879 if ((flags & ISDOTDOT) != 0) {
880 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
881 }
882 if (error) {
883 /* newvp has been revoked. */
884 vrele(newvp);
885 *vpp = NULL;
886 return error;
887 }
888 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
889 cnp->cn_flags |= SAVENAME;
890 KASSERT(newvp->v_type != VNON);
891 return (0);
892 }
893 cache_purge1(newvp, NULL, PURGE_PARENTS);
894 vrele(newvp);
895 *vpp = NULLVP;
896 }
897 dorpc:
898 #if 0
899 /*
900 * because nfsv3 has the same CREATE semantics as ours,
901 * we don't have to perform LOOKUPs beforehand.
902 *
903 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL.
904 * XXX although we have no way to know if O_EXCL is requested or not.
905 */
906
907 if (v3 && cnp->cn_nameiop == CREATE &&
908 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN &&
909 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
910 cnp->cn_flags |= SAVENAME;
911 return (EJUSTRETURN);
912 }
913 #endif /* 0 */
914
915 error = 0;
916 newvp = NULLVP;
917 nfsstats.lookupcache_misses++;
918 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
919 len = cnp->cn_namelen;
920 nfsm_reqhead(np, NFSPROC_LOOKUP,
921 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
922 nfsm_fhtom(np, v3);
923 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
924 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred);
925 if (error) {
926 nfsm_postop_attr(dvp, attrflag, 0);
927 m_freem(mrep);
928 goto nfsmout;
929 }
930 nfsm_getfh(fhp, fhsize, v3);
931
932 /*
933 * Handle RENAME case...
934 */
935 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
936 if (NFS_CMPFH(np, fhp, fhsize)) {
937 m_freem(mrep);
938 return (EISDIR);
939 }
940 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
941 if (error) {
942 m_freem(mrep);
943 return error;
944 }
945 newvp = NFSTOV(np);
946 #ifndef NFS_V2_ONLY
947 if (v3) {
948 nfsm_postop_attr(newvp, attrflag, 0);
949 nfsm_postop_attr(dvp, attrflag, 0);
950 } else
951 #endif
952 nfsm_loadattr(newvp, (struct vattr *)0, 0);
953 *vpp = newvp;
954 m_freem(mrep);
955 cnp->cn_flags |= SAVENAME;
956 goto validate;
957 }
958
959 /*
960 * The postop attr handling is duplicated for each if case,
961 * because it should be done while dvp is locked (unlocking
962 * dvp is different for each case).
963 */
964
965 if (NFS_CMPFH(np, fhp, fhsize)) {
966 /*
967 * "." lookup
968 */
969 VREF(dvp);
970 newvp = dvp;
971 #ifndef NFS_V2_ONLY
972 if (v3) {
973 nfsm_postop_attr(newvp, attrflag, 0);
974 nfsm_postop_attr(dvp, attrflag, 0);
975 } else
976 #endif
977 nfsm_loadattr(newvp, (struct vattr *)0, 0);
978 } else if (flags & ISDOTDOT) {
979 /*
980 * ".." lookup
981 */
982 VOP_UNLOCK(dvp, 0);
983 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
984 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
985 if (error) {
986 m_freem(mrep);
987 return error;
988 }
989 newvp = NFSTOV(np);
990
991 #ifndef NFS_V2_ONLY
992 if (v3) {
993 nfsm_postop_attr(newvp, attrflag, 0);
994 nfsm_postop_attr(dvp, attrflag, 0);
995 } else
996 #endif
997 nfsm_loadattr(newvp, (struct vattr *)0, 0);
998 } else {
999 /*
1000 * Other lookups.
1001 */
1002 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1003 if (error) {
1004 m_freem(mrep);
1005 return error;
1006 }
1007 newvp = NFSTOV(np);
1008 #ifndef NFS_V2_ONLY
1009 if (v3) {
1010 nfsm_postop_attr(newvp, attrflag, 0);
1011 nfsm_postop_attr(dvp, attrflag, 0);
1012 } else
1013 #endif
1014 nfsm_loadattr(newvp, (struct vattr *)0, 0);
1015 }
1016 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1017 cnp->cn_flags |= SAVENAME;
1018 if ((cnp->cn_flags & MAKEENTRY) &&
1019 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1020 nfs_cache_enter(dvp, newvp, cnp);
1021 }
1022 *vpp = newvp;
1023 nfsm_reqdone;
1024 if (error) {
1025 /*
1026 * We get here only because of errors returned by
1027 * the RPC. Otherwise we'll have returned above
1028 * (the nfsm_* macros will jump to nfsm_reqdone
1029 * on error).
1030 */
1031 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1032 cnp->cn_nameiop != CREATE) {
1033 nfs_cache_enter(dvp, NULL, cnp);
1034 }
1035 if (newvp != NULLVP) {
1036 if (newvp == dvp) {
1037 vrele(newvp);
1038 } else {
1039 vput(newvp);
1040 }
1041 }
1042 noentry:
1043 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1044 (flags & ISLASTCN) && error == ENOENT) {
1045 if (dvp->v_mount->mnt_flag & MNT_RDONLY) {
1046 error = EROFS;
1047 } else {
1048 error = EJUSTRETURN;
1049 cnp->cn_flags |= SAVENAME;
1050 }
1051 }
1052 *vpp = NULL;
1053 return error;
1054 }
1055
1056 validate:
1057 /*
1058 * make sure we have valid type and size.
1059 */
1060
1061 newvp = *vpp;
1062 if (newvp->v_type == VNON) {
1063 struct vattr vattr; /* dummy */
1064
1065 KASSERT(VTONFS(newvp)->n_attrstamp == 0);
1066 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred);
1067 if (error) {
1068 vput(newvp);
1069 *vpp = NULL;
1070 }
1071 }
1072
1073 return error;
1074 }
1075
1076 /*
1077 * nfs read call.
1078 * Just call nfs_bioread() to do the work.
1079 */
1080 int
1081 nfs_read(v)
1082 void *v;
1083 {
1084 struct vop_read_args /* {
1085 struct vnode *a_vp;
1086 struct uio *a_uio;
1087 int a_ioflag;
1088 kauth_cred_t a_cred;
1089 } */ *ap = v;
1090 struct vnode *vp = ap->a_vp;
1091
1092 if (vp->v_type != VREG)
1093 return EISDIR;
1094 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1095 }
1096
1097 /*
1098 * nfs readlink call
1099 */
1100 int
1101 nfs_readlink(v)
1102 void *v;
1103 {
1104 struct vop_readlink_args /* {
1105 struct vnode *a_vp;
1106 struct uio *a_uio;
1107 kauth_cred_t a_cred;
1108 } */ *ap = v;
1109 struct vnode *vp = ap->a_vp;
1110 struct nfsnode *np = VTONFS(vp);
1111
1112 if (vp->v_type != VLNK)
1113 return (EPERM);
1114
1115 if (np->n_rcred != NULL) {
1116 kauth_cred_free(np->n_rcred);
1117 }
1118 np->n_rcred = ap->a_cred;
1119 kauth_cred_hold(np->n_rcred);
1120
1121 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1122 }
1123
1124 /*
1125 * Do a readlink rpc.
1126 * Called by nfs_doio() from below the buffer cache.
1127 */
1128 int
1129 nfs_readlinkrpc(vp, uiop, cred)
1130 struct vnode *vp;
1131 struct uio *uiop;
1132 kauth_cred_t cred;
1133 {
1134 u_int32_t *tl;
1135 char *cp;
1136 int32_t t1, t2;
1137 char *bpos, *dpos, *cp2;
1138 int error = 0;
1139 uint32_t len;
1140 struct mbuf *mreq, *mrep, *md, *mb;
1141 const int v3 = NFS_ISV3(vp);
1142 struct nfsnode *np = VTONFS(vp);
1143 #ifndef NFS_V2_ONLY
1144 int attrflag;
1145 #endif
1146
1147 nfsstats.rpccnt[NFSPROC_READLINK]++;
1148 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1149 nfsm_fhtom(np, v3);
1150 nfsm_request(np, NFSPROC_READLINK, curlwp, cred);
1151 #ifndef NFS_V2_ONLY
1152 if (v3)
1153 nfsm_postop_attr(vp, attrflag, 0);
1154 #endif
1155 if (!error) {
1156 #ifndef NFS_V2_ONLY
1157 if (v3) {
1158 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1159 len = fxdr_unsigned(uint32_t, *tl);
1160 if (len > MAXPATHLEN) {
1161 /*
1162 * this pathname is too long for us.
1163 */
1164 m_freem(mrep);
1165 /* Solaris returns EINVAL. should we follow? */
1166 error = ENAMETOOLONG;
1167 goto nfsmout;
1168 }
1169 } else
1170 #endif
1171 {
1172 nfsm_strsiz(len, NFS_MAXPATHLEN);
1173 }
1174 nfsm_mtouio(uiop, len);
1175 }
1176 nfsm_reqdone;
1177 return (error);
1178 }
1179
1180 /*
1181 * nfs read rpc call
1182 * Ditto above
1183 */
1184 int
1185 nfs_readrpc(vp, uiop)
1186 struct vnode *vp;
1187 struct uio *uiop;
1188 {
1189 u_int32_t *tl;
1190 char *cp;
1191 int32_t t1, t2;
1192 char *bpos, *dpos, *cp2;
1193 struct mbuf *mreq, *mrep, *md, *mb;
1194 struct nfsmount *nmp;
1195 int error = 0, len, retlen, tsiz, eof, byte_count;
1196 const int v3 = NFS_ISV3(vp);
1197 struct nfsnode *np = VTONFS(vp);
1198 #ifndef NFS_V2_ONLY
1199 int attrflag;
1200 #endif
1201
1202 #ifndef nolint
1203 eof = 0;
1204 #endif
1205 nmp = VFSTONFS(vp->v_mount);
1206 tsiz = uiop->uio_resid;
1207 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1208 return (EFBIG);
1209 iostat_busy(nmp->nm_stats);
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 iostat_unbusy(nmp->nm_stats, byte_count, 1);
1257 return (error);
1258 }
1259
1260 struct nfs_writerpc_context {
1261 kmutex_t nwc_lock;
1262 kcondvar_t nwc_cv;
1263 int nwc_mbufcount;
1264 };
1265
1266 /*
1267 * free mbuf used to refer protected pages while write rpc call.
1268 * called at splvm.
1269 */
1270 static void
1271 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg)
1272 {
1273 struct nfs_writerpc_context *ctx = arg;
1274
1275 KASSERT(m != NULL);
1276 KASSERT(ctx != NULL);
1277 pool_cache_put(mb_cache, m);
1278 mutex_enter(&ctx->nwc_lock);
1279 if (--ctx->nwc_mbufcount == 0) {
1280 cv_signal(&ctx->nwc_cv);
1281 }
1282 mutex_exit(&ctx->nwc_lock);
1283 }
1284
1285 /*
1286 * nfs write call
1287 */
1288 int
1289 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp)
1290 struct vnode *vp;
1291 struct uio *uiop;
1292 int *iomode;
1293 bool pageprotected;
1294 bool *stalewriteverfp;
1295 {
1296 u_int32_t *tl;
1297 char *cp;
1298 int32_t t1, t2;
1299 char *bpos, *dpos;
1300 struct mbuf *mreq, *mrep, *md, *mb;
1301 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1302 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR;
1303 const int v3 = NFS_ISV3(vp);
1304 int committed = NFSV3WRITE_FILESYNC;
1305 struct nfsnode *np = VTONFS(vp);
1306 struct nfs_writerpc_context ctx;
1307 int byte_count;
1308 struct lwp *l = NULL;
1309 size_t origresid;
1310 #ifndef NFS_V2_ONLY
1311 char *cp2;
1312 int rlen, commit;
1313 #endif
1314
1315 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM);
1316 cv_init(&ctx.nwc_cv, "nfsmblk");
1317 ctx.nwc_mbufcount = 1;
1318
1319 if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1320 panic("writerpc readonly vp %p", vp);
1321 }
1322
1323 #ifdef DIAGNOSTIC
1324 if (uiop->uio_iovcnt != 1)
1325 panic("nfs: writerpc iovcnt > 1");
1326 #endif
1327 tsiz = uiop->uio_resid;
1328 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1329 return (EFBIG);
1330 if (pageprotected) {
1331 l = curlwp;
1332 uvm_lwp_hold(l);
1333 }
1334 retry:
1335 origresid = uiop->uio_resid;
1336 KASSERT(origresid == uiop->uio_iov->iov_len);
1337 iostat_busy(nmp->nm_stats);
1338 byte_count = 0; /* count of bytes actually written */
1339 while (tsiz > 0) {
1340 uint32_t datalen; /* data bytes need to be allocated in mbuf */
1341 uint32_t backup;
1342 bool stalewriteverf = false;
1343
1344 nfsstats.rpccnt[NFSPROC_WRITE]++;
1345 len = min(tsiz, nmp->nm_wsize);
1346 datalen = pageprotected ? 0 : nfsm_rndup(len);
1347 nfsm_reqhead(np, NFSPROC_WRITE,
1348 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1349 nfsm_fhtom(np, v3);
1350 #ifndef NFS_V2_ONLY
1351 if (v3) {
1352 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1353 txdr_hyper(uiop->uio_offset, tl);
1354 tl += 2;
1355 *tl++ = txdr_unsigned(len);
1356 *tl++ = txdr_unsigned(*iomode);
1357 *tl = txdr_unsigned(len);
1358 } else
1359 #endif
1360 {
1361 u_int32_t x;
1362
1363 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1364 /* Set both "begin" and "current" to non-garbage. */
1365 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1366 *tl++ = x; /* "begin offset" */
1367 *tl++ = x; /* "current offset" */
1368 x = txdr_unsigned(len);
1369 *tl++ = x; /* total to this offset */
1370 *tl = x; /* size of this write */
1371
1372 }
1373 if (pageprotected) {
1374 /*
1375 * since we know pages can't be modified during i/o,
1376 * no need to copy them for us.
1377 */
1378 struct mbuf *m;
1379 struct iovec *iovp = uiop->uio_iov;
1380
1381 m = m_get(M_WAIT, MT_DATA);
1382 MCLAIM(m, &nfs_mowner);
1383 MEXTADD(m, iovp->iov_base, len, M_MBUF,
1384 nfs_writerpc_extfree, &ctx);
1385 m->m_flags |= M_EXT_ROMAP;
1386 m->m_len = len;
1387 mb->m_next = m;
1388 /*
1389 * no need to maintain mb and bpos here
1390 * because no one care them later.
1391 */
1392 #if 0
1393 mb = m;
1394 bpos = mtod(void *, mb) + mb->m_len;
1395 #endif
1396 UIO_ADVANCE(uiop, len);
1397 uiop->uio_offset += len;
1398 mutex_enter(&ctx.nwc_lock);
1399 ctx.nwc_mbufcount++;
1400 mutex_exit(&ctx.nwc_lock);
1401 nfs_zeropad(mb, 0, nfsm_padlen(len));
1402 } else {
1403 nfsm_uiotom(uiop, len);
1404 }
1405 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred);
1406 #ifndef NFS_V2_ONLY
1407 if (v3) {
1408 wccflag = NFSV3_WCCCHK;
1409 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error);
1410 if (!error) {
1411 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1412 + NFSX_V3WRITEVERF);
1413 rlen = fxdr_unsigned(int, *tl++);
1414 if (rlen == 0) {
1415 error = NFSERR_IO;
1416 m_freem(mrep);
1417 break;
1418 } else if (rlen < len) {
1419 backup = len - rlen;
1420 UIO_ADVANCE(uiop, -backup);
1421 uiop->uio_offset -= backup;
1422 len = rlen;
1423 }
1424 commit = fxdr_unsigned(int, *tl++);
1425
1426 /*
1427 * Return the lowest committment level
1428 * obtained by any of the RPCs.
1429 */
1430 if (committed == NFSV3WRITE_FILESYNC)
1431 committed = commit;
1432 else if (committed == NFSV3WRITE_DATASYNC &&
1433 commit == NFSV3WRITE_UNSTABLE)
1434 committed = commit;
1435 mutex_enter(&nmp->nm_lock);
1436 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1437 memcpy(nmp->nm_writeverf, tl,
1438 NFSX_V3WRITEVERF);
1439 nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1440 } else if ((nmp->nm_iflag &
1441 NFSMNT_STALEWRITEVERF) ||
1442 memcmp(tl, nmp->nm_writeverf,
1443 NFSX_V3WRITEVERF)) {
1444 memcpy(nmp->nm_writeverf, tl,
1445 NFSX_V3WRITEVERF);
1446 /*
1447 * note NFSMNT_STALEWRITEVERF
1448 * if we're the first thread to
1449 * notice it.
1450 */
1451 if ((nmp->nm_iflag &
1452 NFSMNT_STALEWRITEVERF) == 0) {
1453 stalewriteverf = true;
1454 nmp->nm_iflag |=
1455 NFSMNT_STALEWRITEVERF;
1456 }
1457 }
1458 mutex_exit(&nmp->nm_lock);
1459 }
1460 } else
1461 #endif
1462 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1463 if (wccflag)
1464 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1465 m_freem(mrep);
1466 if (error)
1467 break;
1468 tsiz -= len;
1469 byte_count += len;
1470 if (stalewriteverf) {
1471 *stalewriteverfp = true;
1472 stalewriteverf = false;
1473 if (committed == NFSV3WRITE_UNSTABLE &&
1474 len != origresid) {
1475 /*
1476 * if our write requests weren't atomic but
1477 * unstable, datas in previous iterations
1478 * might have already been lost now.
1479 * then, we should resend them to nfsd.
1480 */
1481 backup = origresid - tsiz;
1482 UIO_ADVANCE(uiop, -backup);
1483 uiop->uio_offset -= backup;
1484 tsiz = origresid;
1485 goto retry;
1486 }
1487 }
1488 }
1489 nfsmout:
1490 iostat_unbusy(nmp->nm_stats, byte_count, 0);
1491 if (pageprotected) {
1492 /*
1493 * wait until mbufs go away.
1494 * retransmitted mbufs can survive longer than rpc requests
1495 * themselves.
1496 */
1497 mutex_enter(&ctx.nwc_lock);
1498 ctx.nwc_mbufcount--;
1499 while (ctx.nwc_mbufcount > 0) {
1500 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock);
1501 }
1502 mutex_exit(&ctx.nwc_lock);
1503 uvm_lwp_rele(l);
1504 }
1505 mutex_destroy(&ctx.nwc_lock);
1506 cv_destroy(&ctx.nwc_cv);
1507 *iomode = committed;
1508 if (error)
1509 uiop->uio_resid = tsiz;
1510 return (error);
1511 }
1512
1513 /*
1514 * nfs mknod rpc
1515 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1516 * mode set to specify the file type and the size field for rdev.
1517 */
1518 int
1519 nfs_mknodrpc(dvp, vpp, cnp, vap)
1520 struct vnode *dvp;
1521 struct vnode **vpp;
1522 struct componentname *cnp;
1523 struct vattr *vap;
1524 {
1525 struct nfsv2_sattr *sp;
1526 u_int32_t *tl;
1527 char *cp;
1528 int32_t t1, t2;
1529 struct vnode *newvp = (struct vnode *)0;
1530 struct nfsnode *dnp, *np;
1531 char *cp2;
1532 char *bpos, *dpos;
1533 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1534 struct mbuf *mreq, *mrep, *md, *mb;
1535 u_int32_t rdev;
1536 const int v3 = NFS_ISV3(dvp);
1537
1538 if (vap->va_type == VCHR || vap->va_type == VBLK)
1539 rdev = txdr_unsigned(vap->va_rdev);
1540 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1541 rdev = nfs_xdrneg1;
1542 else {
1543 VOP_ABORTOP(dvp, cnp);
1544 vput(dvp);
1545 return (EOPNOTSUPP);
1546 }
1547 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1548 dnp = VTONFS(dvp);
1549 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1550 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1551 nfsm_fhtom(dnp, v3);
1552 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1553 #ifndef NFS_V2_ONLY
1554 if (v3) {
1555 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1556 *tl++ = vtonfsv3_type(vap->va_type);
1557 nfsm_v3attrbuild(vap, false);
1558 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1559 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1560 *tl++ = txdr_unsigned(major(vap->va_rdev));
1561 *tl = txdr_unsigned(minor(vap->va_rdev));
1562 }
1563 } else
1564 #endif
1565 {
1566 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1567 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1568 sp->sa_uid = nfs_xdrneg1;
1569 sp->sa_gid = nfs_xdrneg1;
1570 sp->sa_size = rdev;
1571 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1572 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1573 }
1574 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred);
1575 if (!error) {
1576 nfsm_mtofh(dvp, newvp, v3, gotvp);
1577 if (!gotvp) {
1578 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1579 cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1580 if (!error)
1581 newvp = NFSTOV(np);
1582 }
1583 }
1584 #ifndef NFS_V2_ONLY
1585 if (v3)
1586 nfsm_wcc_data(dvp, wccflag, 0, !error);
1587 #endif
1588 nfsm_reqdone;
1589 if (error) {
1590 if (newvp)
1591 vput(newvp);
1592 } else {
1593 if (cnp->cn_flags & MAKEENTRY)
1594 nfs_cache_enter(dvp, newvp, cnp);
1595 *vpp = newvp;
1596 }
1597 PNBUF_PUT(cnp->cn_pnbuf);
1598 VTONFS(dvp)->n_flag |= NMODIFIED;
1599 if (!wccflag)
1600 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1601 vput(dvp);
1602 return (error);
1603 }
1604
1605 /*
1606 * nfs mknod vop
1607 * just call nfs_mknodrpc() to do the work.
1608 */
1609 /* ARGSUSED */
1610 int
1611 nfs_mknod(v)
1612 void *v;
1613 {
1614 struct vop_mknod_args /* {
1615 struct vnode *a_dvp;
1616 struct vnode **a_vpp;
1617 struct componentname *a_cnp;
1618 struct vattr *a_vap;
1619 } */ *ap = v;
1620 struct vnode *dvp = ap->a_dvp;
1621 struct componentname *cnp = ap->a_cnp;
1622 int error;
1623
1624 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap);
1625 VN_KNOTE(dvp, NOTE_WRITE);
1626 if (error == 0 || error == EEXIST)
1627 cache_purge1(dvp, cnp, 0);
1628 return (error);
1629 }
1630
1631 /*
1632 * nfs file create call
1633 */
1634 int
1635 nfs_create(v)
1636 void *v;
1637 {
1638 struct vop_create_args /* {
1639 struct vnode *a_dvp;
1640 struct vnode **a_vpp;
1641 struct componentname *a_cnp;
1642 struct vattr *a_vap;
1643 } */ *ap = v;
1644 struct vnode *dvp = ap->a_dvp;
1645 struct vattr *vap = ap->a_vap;
1646 struct componentname *cnp = ap->a_cnp;
1647 struct nfsv2_sattr *sp;
1648 u_int32_t *tl;
1649 char *cp;
1650 int32_t t1, t2;
1651 struct nfsnode *dnp, *np = (struct nfsnode *)0;
1652 struct vnode *newvp = (struct vnode *)0;
1653 char *bpos, *dpos, *cp2;
1654 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1655 struct mbuf *mreq, *mrep, *md, *mb;
1656 const int v3 = NFS_ISV3(dvp);
1657 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED;
1658
1659 /*
1660 * Oops, not for me..
1661 */
1662 if (vap->va_type == VSOCK)
1663 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1664
1665 KASSERT(vap->va_type == VREG);
1666
1667 #ifdef VA_EXCLUSIVE
1668 if (vap->va_vaflags & VA_EXCLUSIVE) {
1669 excl_mode = NFSV3CREATE_EXCLUSIVE;
1670 }
1671 #endif
1672 again:
1673 error = 0;
1674 nfsstats.rpccnt[NFSPROC_CREATE]++;
1675 dnp = VTONFS(dvp);
1676 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1677 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1678 nfsm_fhtom(dnp, v3);
1679 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1680 #ifndef NFS_V2_ONLY
1681 if (v3) {
1682 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1683 if (excl_mode == NFSV3CREATE_EXCLUSIVE) {
1684 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1685 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1686 *tl++ = arc4random();
1687 *tl = arc4random();
1688 } else {
1689 *tl = txdr_unsigned(excl_mode);
1690 nfsm_v3attrbuild(vap, false);
1691 }
1692 } else
1693 #endif
1694 {
1695 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1696 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1697 sp->sa_uid = nfs_xdrneg1;
1698 sp->sa_gid = nfs_xdrneg1;
1699 sp->sa_size = 0;
1700 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1701 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1702 }
1703 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred);
1704 if (!error) {
1705 nfsm_mtofh(dvp, newvp, v3, gotvp);
1706 if (!gotvp) {
1707 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1708 cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1709 if (!error)
1710 newvp = NFSTOV(np);
1711 }
1712 }
1713 #ifndef NFS_V2_ONLY
1714 if (v3)
1715 nfsm_wcc_data(dvp, wccflag, 0, !error);
1716 #endif
1717 nfsm_reqdone;
1718 if (error) {
1719 /*
1720 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
1721 */
1722 if (v3 && error == ENOTSUP) {
1723 if (excl_mode == NFSV3CREATE_EXCLUSIVE) {
1724 excl_mode = NFSV3CREATE_GUARDED;
1725 goto again;
1726 } else if (excl_mode == NFSV3CREATE_GUARDED) {
1727 excl_mode = NFSV3CREATE_UNCHECKED;
1728 goto again;
1729 }
1730 }
1731 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) {
1732 struct timespec ts;
1733
1734 getnanotime(&ts);
1735
1736 /*
1737 * make sure that we'll update timestamps as
1738 * most server implementations use them to store
1739 * the create verifier.
1740 *
1741 * XXX it's better to use TOSERVER always.
1742 */
1743
1744 if (vap->va_atime.tv_sec == VNOVAL)
1745 vap->va_atime = ts;
1746 if (vap->va_mtime.tv_sec == VNOVAL)
1747 vap->va_mtime = ts;
1748
1749 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp);
1750 }
1751 if (error == 0) {
1752 if (cnp->cn_flags & MAKEENTRY)
1753 nfs_cache_enter(dvp, newvp, cnp);
1754 else
1755 cache_purge1(dvp, cnp, 0);
1756 *ap->a_vpp = newvp;
1757 } else {
1758 if (newvp)
1759 vput(newvp);
1760 if (error == EEXIST)
1761 cache_purge1(dvp, cnp, 0);
1762 }
1763 PNBUF_PUT(cnp->cn_pnbuf);
1764 VTONFS(dvp)->n_flag |= NMODIFIED;
1765 if (!wccflag)
1766 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1767 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1768 vput(dvp);
1769 return (error);
1770 }
1771
1772 /*
1773 * nfs file remove call
1774 * To try and make nfs semantics closer to ufs semantics, a file that has
1775 * other processes using the vnode is renamed instead of removed and then
1776 * removed later on the last close.
1777 * - If v_usecount > 1
1778 * If a rename is not already in the works
1779 * call nfs_sillyrename() to set it up
1780 * else
1781 * do the remove rpc
1782 */
1783 int
1784 nfs_remove(v)
1785 void *v;
1786 {
1787 struct vop_remove_args /* {
1788 struct vnodeop_desc *a_desc;
1789 struct vnode * a_dvp;
1790 struct vnode * a_vp;
1791 struct componentname * a_cnp;
1792 } */ *ap = v;
1793 struct vnode *vp = ap->a_vp;
1794 struct vnode *dvp = ap->a_dvp;
1795 struct componentname *cnp = ap->a_cnp;
1796 struct nfsnode *np = VTONFS(vp);
1797 int error = 0;
1798 struct vattr vattr;
1799
1800 #ifndef DIAGNOSTIC
1801 if ((cnp->cn_flags & HASBUF) == 0)
1802 panic("nfs_remove: no name");
1803 if (vp->v_usecount < 1)
1804 panic("nfs_remove: bad v_usecount");
1805 #endif
1806 if (vp->v_type == VDIR)
1807 error = EPERM;
1808 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1809 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1810 vattr.va_nlink > 1)) {
1811 /*
1812 * Purge the name cache so that the chance of a lookup for
1813 * the name succeeding while the remove is in progress is
1814 * minimized. Without node locking it can still happen, such
1815 * that an I/O op returns ESTALE, but since you get this if
1816 * another host removes the file..
1817 */
1818 cache_purge(vp);
1819 /*
1820 * throw away biocache buffers, mainly to avoid
1821 * unnecessary delayed writes later.
1822 */
1823 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1);
1824 /* Do the rpc */
1825 if (error != EINTR)
1826 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1827 cnp->cn_namelen, cnp->cn_cred, curlwp);
1828 } else if (!np->n_sillyrename)
1829 error = nfs_sillyrename(dvp, vp, cnp, false);
1830 PNBUF_PUT(cnp->cn_pnbuf);
1831 if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1832 vattr.va_nlink == 1) {
1833 np->n_flag |= NREMOVED;
1834 }
1835 NFS_INVALIDATE_ATTRCACHE(np);
1836 VN_KNOTE(vp, NOTE_DELETE);
1837 VN_KNOTE(dvp, NOTE_WRITE);
1838 if (dvp == vp)
1839 vrele(vp);
1840 else
1841 vput(vp);
1842 vput(dvp);
1843 return (error);
1844 }
1845
1846 /*
1847 * nfs file remove rpc called from nfs_inactive
1848 */
1849 int
1850 nfs_removeit(sp)
1851 struct sillyrename *sp;
1852 {
1853
1854 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1855 (struct lwp *)0));
1856 }
1857
1858 /*
1859 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1860 */
1861 int
1862 nfs_removerpc(dvp, name, namelen, cred, l)
1863 struct vnode *dvp;
1864 const char *name;
1865 int namelen;
1866 kauth_cred_t cred;
1867 struct lwp *l;
1868 {
1869 u_int32_t *tl;
1870 char *cp;
1871 #ifndef NFS_V2_ONLY
1872 int32_t t1;
1873 char *cp2;
1874 #endif
1875 int32_t t2;
1876 char *bpos, *dpos;
1877 int error = 0, wccflag = NFSV3_WCCRATTR;
1878 struct mbuf *mreq, *mrep, *md, *mb;
1879 const int v3 = NFS_ISV3(dvp);
1880 int rexmit = 0;
1881 struct nfsnode *dnp = VTONFS(dvp);
1882
1883 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1884 nfsm_reqhead(dnp, NFSPROC_REMOVE,
1885 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1886 nfsm_fhtom(dnp, v3);
1887 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1888 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit);
1889 #ifndef NFS_V2_ONLY
1890 if (v3)
1891 nfsm_wcc_data(dvp, wccflag, 0, !error);
1892 #endif
1893 nfsm_reqdone;
1894 VTONFS(dvp)->n_flag |= NMODIFIED;
1895 if (!wccflag)
1896 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1897 /*
1898 * Kludge City: If the first reply to the remove rpc is lost..
1899 * the reply to the retransmitted request will be ENOENT
1900 * since the file was in fact removed
1901 * Therefore, we cheat and return success.
1902 */
1903 if (rexmit && error == ENOENT)
1904 error = 0;
1905 return (error);
1906 }
1907
1908 /*
1909 * nfs file rename call
1910 */
1911 int
1912 nfs_rename(v)
1913 void *v;
1914 {
1915 struct vop_rename_args /* {
1916 struct vnode *a_fdvp;
1917 struct vnode *a_fvp;
1918 struct componentname *a_fcnp;
1919 struct vnode *a_tdvp;
1920 struct vnode *a_tvp;
1921 struct componentname *a_tcnp;
1922 } */ *ap = v;
1923 struct vnode *fvp = ap->a_fvp;
1924 struct vnode *tvp = ap->a_tvp;
1925 struct vnode *fdvp = ap->a_fdvp;
1926 struct vnode *tdvp = ap->a_tdvp;
1927 struct componentname *tcnp = ap->a_tcnp;
1928 struct componentname *fcnp = ap->a_fcnp;
1929 int error;
1930
1931 #ifndef DIAGNOSTIC
1932 if ((tcnp->cn_flags & HASBUF) == 0 ||
1933 (fcnp->cn_flags & HASBUF) == 0)
1934 panic("nfs_rename: no name");
1935 #endif
1936 /* Check for cross-device rename */
1937 if ((fvp->v_mount != tdvp->v_mount) ||
1938 (tvp && (fvp->v_mount != tvp->v_mount))) {
1939 error = EXDEV;
1940 goto out;
1941 }
1942
1943 /*
1944 * If the tvp exists and is in use, sillyrename it before doing the
1945 * rename of the new file over it.
1946 *
1947 * Have sillyrename use link instead of rename if possible,
1948 * so that we don't lose the file if the rename fails, and so
1949 * that there's no window when the "to" file doesn't exist.
1950 */
1951 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1952 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) {
1953 VN_KNOTE(tvp, NOTE_DELETE);
1954 vput(tvp);
1955 tvp = NULL;
1956 }
1957
1958 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1959 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1960 curlwp);
1961
1962 VN_KNOTE(fdvp, NOTE_WRITE);
1963 VN_KNOTE(tdvp, NOTE_WRITE);
1964 if (error == 0 || error == EEXIST) {
1965 if (fvp->v_type == VDIR)
1966 cache_purge(fvp);
1967 else
1968 cache_purge1(fdvp, fcnp, 0);
1969 if (tvp != NULL && tvp->v_type == VDIR)
1970 cache_purge(tvp);
1971 else
1972 cache_purge1(tdvp, tcnp, 0);
1973 }
1974 out:
1975 if (tdvp == tvp)
1976 vrele(tdvp);
1977 else
1978 vput(tdvp);
1979 if (tvp)
1980 vput(tvp);
1981 vrele(fdvp);
1982 vrele(fvp);
1983 return (error);
1984 }
1985
1986 /*
1987 * nfs file rename rpc called from nfs_remove() above
1988 */
1989 int
1990 nfs_renameit(sdvp, scnp, sp)
1991 struct vnode *sdvp;
1992 struct componentname *scnp;
1993 struct sillyrename *sp;
1994 {
1995 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1996 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp));
1997 }
1998
1999 /*
2000 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2001 */
2002 int
2003 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, l)
2004 struct vnode *fdvp;
2005 const char *fnameptr;
2006 int fnamelen;
2007 struct vnode *tdvp;
2008 const char *tnameptr;
2009 int tnamelen;
2010 kauth_cred_t cred;
2011 struct lwp *l;
2012 {
2013 u_int32_t *tl;
2014 char *cp;
2015 #ifndef NFS_V2_ONLY
2016 int32_t t1;
2017 char *cp2;
2018 #endif
2019 int32_t t2;
2020 char *bpos, *dpos;
2021 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
2022 struct mbuf *mreq, *mrep, *md, *mb;
2023 const int v3 = NFS_ISV3(fdvp);
2024 int rexmit = 0;
2025 struct nfsnode *fdnp = VTONFS(fdvp);
2026
2027 nfsstats.rpccnt[NFSPROC_RENAME]++;
2028 nfsm_reqhead(fdnp, NFSPROC_RENAME,
2029 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
2030 nfsm_rndup(tnamelen));
2031 nfsm_fhtom(fdnp, v3);
2032 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
2033 nfsm_fhtom(VTONFS(tdvp), v3);
2034 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
2035 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit);
2036 #ifndef NFS_V2_ONLY
2037 if (v3) {
2038 nfsm_wcc_data(fdvp, fwccflag, 0, !error);
2039 nfsm_wcc_data(tdvp, twccflag, 0, !error);
2040 }
2041 #endif
2042 nfsm_reqdone;
2043 VTONFS(fdvp)->n_flag |= NMODIFIED;
2044 VTONFS(tdvp)->n_flag |= NMODIFIED;
2045 if (!fwccflag)
2046 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
2047 if (!twccflag)
2048 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
2049 /*
2050 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2051 */
2052 if (rexmit && error == ENOENT)
2053 error = 0;
2054 return (error);
2055 }
2056
2057 /*
2058 * NFS link RPC, called from nfs_link.
2059 * Assumes dvp and vp locked, and leaves them that way.
2060 */
2061
2062 static int
2063 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name,
2064 size_t namelen, kauth_cred_t cred, struct lwp *l)
2065 {
2066 u_int32_t *tl;
2067 char *cp;
2068 #ifndef NFS_V2_ONLY
2069 int32_t t1;
2070 char *cp2;
2071 #endif
2072 int32_t t2;
2073 char *bpos, *dpos;
2074 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2075 struct mbuf *mreq, *mrep, *md, *mb;
2076 const int v3 = NFS_ISV3(dvp);
2077 int rexmit = 0;
2078 struct nfsnode *np = VTONFS(vp);
2079
2080 nfsstats.rpccnt[NFSPROC_LINK]++;
2081 nfsm_reqhead(np, NFSPROC_LINK,
2082 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen));
2083 nfsm_fhtom(np, v3);
2084 nfsm_fhtom(VTONFS(dvp), v3);
2085 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
2086 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit);
2087 #ifndef NFS_V2_ONLY
2088 if (v3) {
2089 nfsm_postop_attr(vp, attrflag, 0);
2090 nfsm_wcc_data(dvp, wccflag, 0, !error);
2091 }
2092 #endif
2093 nfsm_reqdone;
2094
2095 VTONFS(dvp)->n_flag |= NMODIFIED;
2096 if (!attrflag)
2097 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
2098 if (!wccflag)
2099 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2100
2101 /*
2102 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2103 */
2104 if (rexmit && error == EEXIST)
2105 error = 0;
2106
2107 return error;
2108 }
2109
2110 /*
2111 * nfs hard link create call
2112 */
2113 int
2114 nfs_link(v)
2115 void *v;
2116 {
2117 struct vop_link_args /* {
2118 struct vnode *a_dvp;
2119 struct vnode *a_vp;
2120 struct componentname *a_cnp;
2121 } */ *ap = v;
2122 struct vnode *vp = ap->a_vp;
2123 struct vnode *dvp = ap->a_dvp;
2124 struct componentname *cnp = ap->a_cnp;
2125 int error = 0;
2126
2127 if (dvp->v_mount != vp->v_mount) {
2128 VOP_ABORTOP(dvp, cnp);
2129 vput(dvp);
2130 return (EXDEV);
2131 }
2132 if (dvp != vp) {
2133 error = vn_lock(vp, LK_EXCLUSIVE);
2134 if (error != 0) {
2135 VOP_ABORTOP(dvp, cnp);
2136 vput(dvp);
2137 return error;
2138 }
2139 }
2140
2141 /*
2142 * Push all writes to the server, so that the attribute cache
2143 * doesn't get "out of sync" with the server.
2144 * XXX There should be a better way!
2145 */
2146 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0);
2147
2148 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2149 cnp->cn_cred, curlwp);
2150
2151 if (error == 0)
2152 cache_purge1(dvp, cnp, 0);
2153 PNBUF_PUT(cnp->cn_pnbuf);
2154 if (dvp != vp)
2155 VOP_UNLOCK(vp, 0);
2156 VN_KNOTE(vp, NOTE_LINK);
2157 VN_KNOTE(dvp, NOTE_WRITE);
2158 vput(dvp);
2159 return (error);
2160 }
2161
2162 /*
2163 * nfs symbolic link create call
2164 */
2165 int
2166 nfs_symlink(v)
2167 void *v;
2168 {
2169 struct vop_symlink_args /* {
2170 struct vnode *a_dvp;
2171 struct vnode **a_vpp;
2172 struct componentname *a_cnp;
2173 struct vattr *a_vap;
2174 char *a_target;
2175 } */ *ap = v;
2176 struct vnode *dvp = ap->a_dvp;
2177 struct vattr *vap = ap->a_vap;
2178 struct componentname *cnp = ap->a_cnp;
2179 struct nfsv2_sattr *sp;
2180 u_int32_t *tl;
2181 char *cp;
2182 int32_t t1, t2;
2183 char *bpos, *dpos, *cp2;
2184 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2185 struct mbuf *mreq, *mrep, *md, *mb;
2186 struct vnode *newvp = (struct vnode *)0;
2187 const int v3 = NFS_ISV3(dvp);
2188 int rexmit = 0;
2189 struct nfsnode *dnp = VTONFS(dvp);
2190
2191 *ap->a_vpp = NULL;
2192 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2193 slen = strlen(ap->a_target);
2194 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2195 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2196 nfsm_fhtom(dnp, v3);
2197 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2198 #ifndef NFS_V2_ONlY
2199 if (v3)
2200 nfsm_v3attrbuild(vap, false);
2201 #endif
2202 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2203 #ifndef NFS_V2_ONlY
2204 if (!v3) {
2205 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2206 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2207 sp->sa_uid = nfs_xdrneg1;
2208 sp->sa_gid = nfs_xdrneg1;
2209 sp->sa_size = nfs_xdrneg1;
2210 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2211 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2212 }
2213 #endif
2214 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred,
2215 &rexmit);
2216 #ifndef NFS_V2_ONlY
2217 if (v3) {
2218 if (!error)
2219 nfsm_mtofh(dvp, newvp, v3, gotvp);
2220 nfsm_wcc_data(dvp, wccflag, 0, !error);
2221 }
2222 #endif
2223 nfsm_reqdone;
2224 /*
2225 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2226 */
2227 if (rexmit && error == EEXIST)
2228 error = 0;
2229 if (error == 0 || error == EEXIST)
2230 cache_purge1(dvp, cnp, 0);
2231 if (error == 0 && newvp == NULL) {
2232 struct nfsnode *np = NULL;
2233
2234 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2235 cnp->cn_cred, curlwp, &np);
2236 if (error == 0)
2237 newvp = NFSTOV(np);
2238 }
2239 if (error) {
2240 if (newvp != NULL)
2241 vput(newvp);
2242 } else {
2243 *ap->a_vpp = newvp;
2244 }
2245 PNBUF_PUT(cnp->cn_pnbuf);
2246 VTONFS(dvp)->n_flag |= NMODIFIED;
2247 if (!wccflag)
2248 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2249 VN_KNOTE(dvp, NOTE_WRITE);
2250 vput(dvp);
2251 return (error);
2252 }
2253
2254 /*
2255 * nfs make dir call
2256 */
2257 int
2258 nfs_mkdir(v)
2259 void *v;
2260 {
2261 struct vop_mkdir_args /* {
2262 struct vnode *a_dvp;
2263 struct vnode **a_vpp;
2264 struct componentname *a_cnp;
2265 struct vattr *a_vap;
2266 } */ *ap = v;
2267 struct vnode *dvp = ap->a_dvp;
2268 struct vattr *vap = ap->a_vap;
2269 struct componentname *cnp = ap->a_cnp;
2270 struct nfsv2_sattr *sp;
2271 u_int32_t *tl;
2272 char *cp;
2273 int32_t t1, t2;
2274 int len;
2275 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2276 struct vnode *newvp = (struct vnode *)0;
2277 char *bpos, *dpos, *cp2;
2278 int error = 0, wccflag = NFSV3_WCCRATTR;
2279 int gotvp = 0;
2280 int rexmit = 0;
2281 struct mbuf *mreq, *mrep, *md, *mb;
2282 const int v3 = NFS_ISV3(dvp);
2283
2284 len = cnp->cn_namelen;
2285 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2286 nfsm_reqhead(dnp, NFSPROC_MKDIR,
2287 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2288 nfsm_fhtom(dnp, v3);
2289 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2290 #ifndef NFS_V2_ONLY
2291 if (v3) {
2292 nfsm_v3attrbuild(vap, false);
2293 } else
2294 #endif
2295 {
2296 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2297 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2298 sp->sa_uid = nfs_xdrneg1;
2299 sp->sa_gid = nfs_xdrneg1;
2300 sp->sa_size = nfs_xdrneg1;
2301 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2302 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2303 }
2304 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit);
2305 if (!error)
2306 nfsm_mtofh(dvp, newvp, v3, gotvp);
2307 if (v3)
2308 nfsm_wcc_data(dvp, wccflag, 0, !error);
2309 nfsm_reqdone;
2310 VTONFS(dvp)->n_flag |= NMODIFIED;
2311 if (!wccflag)
2312 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2313 /*
2314 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2315 * if we can succeed in looking up the directory.
2316 */
2317 if ((rexmit && error == EEXIST) || (!error && !gotvp)) {
2318 if (newvp) {
2319 vput(newvp);
2320 newvp = (struct vnode *)0;
2321 }
2322 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2323 curlwp, &np);
2324 if (!error) {
2325 newvp = NFSTOV(np);
2326 if (newvp->v_type != VDIR || newvp == dvp)
2327 error = EEXIST;
2328 }
2329 }
2330 if (error) {
2331 if (newvp) {
2332 if (dvp != newvp)
2333 vput(newvp);
2334 else
2335 vrele(newvp);
2336 }
2337 } else {
2338 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2339 if (cnp->cn_flags & MAKEENTRY)
2340 nfs_cache_enter(dvp, newvp, cnp);
2341 *ap->a_vpp = newvp;
2342 }
2343 PNBUF_PUT(cnp->cn_pnbuf);
2344 vput(dvp);
2345 return (error);
2346 }
2347
2348 /*
2349 * nfs remove directory call
2350 */
2351 int
2352 nfs_rmdir(v)
2353 void *v;
2354 {
2355 struct vop_rmdir_args /* {
2356 struct vnode *a_dvp;
2357 struct vnode *a_vp;
2358 struct componentname *a_cnp;
2359 } */ *ap = v;
2360 struct vnode *vp = ap->a_vp;
2361 struct vnode *dvp = ap->a_dvp;
2362 struct componentname *cnp = ap->a_cnp;
2363 u_int32_t *tl;
2364 char *cp;
2365 #ifndef NFS_V2_ONLY
2366 int32_t t1;
2367 char *cp2;
2368 #endif
2369 int32_t t2;
2370 char *bpos, *dpos;
2371 int error = 0, wccflag = NFSV3_WCCRATTR;
2372 int rexmit = 0;
2373 struct mbuf *mreq, *mrep, *md, *mb;
2374 const int v3 = NFS_ISV3(dvp);
2375 struct nfsnode *dnp;
2376
2377 if (dvp == vp) {
2378 vrele(dvp);
2379 vput(dvp);
2380 PNBUF_PUT(cnp->cn_pnbuf);
2381 return (EINVAL);
2382 }
2383 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2384 dnp = VTONFS(dvp);
2385 nfsm_reqhead(dnp, NFSPROC_RMDIR,
2386 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2387 nfsm_fhtom(dnp, v3);
2388 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2389 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit);
2390 #ifndef NFS_V2_ONLY
2391 if (v3)
2392 nfsm_wcc_data(dvp, wccflag, 0, !error);
2393 #endif
2394 nfsm_reqdone;
2395 PNBUF_PUT(cnp->cn_pnbuf);
2396 VTONFS(dvp)->n_flag |= NMODIFIED;
2397 if (!wccflag)
2398 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2399 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2400 VN_KNOTE(vp, NOTE_DELETE);
2401 cache_purge(vp);
2402 vput(vp);
2403 vput(dvp);
2404 /*
2405 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2406 */
2407 if (rexmit && error == ENOENT)
2408 error = 0;
2409 return (error);
2410 }
2411
2412 /*
2413 * nfs readdir call
2414 */
2415 int
2416 nfs_readdir(v)
2417 void *v;
2418 {
2419 struct vop_readdir_args /* {
2420 struct vnode *a_vp;
2421 struct uio *a_uio;
2422 kauth_cred_t a_cred;
2423 int *a_eofflag;
2424 off_t **a_cookies;
2425 int *a_ncookies;
2426 } */ *ap = v;
2427 struct vnode *vp = ap->a_vp;
2428 struct uio *uio = ap->a_uio;
2429 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2430 char *base = uio->uio_iov->iov_base;
2431 int tresid, error;
2432 size_t count, lost;
2433 struct dirent *dp;
2434 off_t *cookies = NULL;
2435 int ncookies = 0, nc;
2436
2437 if (vp->v_type != VDIR)
2438 return (EPERM);
2439
2440 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2441 count = uio->uio_resid - lost;
2442 if (count <= 0)
2443 return (EINVAL);
2444
2445 /*
2446 * Call nfs_bioread() to do the real work.
2447 */
2448 tresid = uio->uio_resid = count;
2449 error = nfs_bioread(vp, uio, 0, ap->a_cred,
2450 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2451
2452 if (!error && ap->a_cookies) {
2453 ncookies = count / 16;
2454 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2455 *ap->a_cookies = cookies;
2456 }
2457
2458 if (!error && uio->uio_resid == tresid) {
2459 uio->uio_resid += lost;
2460 nfsstats.direofcache_misses++;
2461 if (ap->a_cookies)
2462 *ap->a_ncookies = 0;
2463 *ap->a_eofflag = 1;
2464 return (0);
2465 }
2466
2467 if (!error && ap->a_cookies) {
2468 /*
2469 * Only the NFS server and emulations use cookies, and they
2470 * load the directory block into system space, so we can
2471 * just look at it directly.
2472 */
2473 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) ||
2474 uio->uio_iovcnt != 1)
2475 panic("nfs_readdir: lost in space");
2476 for (nc = 0; ncookies-- &&
2477 base < (char *)uio->uio_iov->iov_base; nc++){
2478 dp = (struct dirent *) base;
2479 if (dp->d_reclen == 0)
2480 break;
2481 if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2482 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2483 else
2484 *(cookies++) = NFS_GETCOOKIE(dp);
2485 base += dp->d_reclen;
2486 }
2487 uio->uio_resid +=
2488 ((char *)uio->uio_iov->iov_base - base);
2489 uio->uio_iov->iov_len +=
2490 ((char *)uio->uio_iov->iov_base - base);
2491 uio->uio_iov->iov_base = base;
2492 *ap->a_ncookies = nc;
2493 }
2494
2495 uio->uio_resid += lost;
2496 *ap->a_eofflag = 0;
2497 return (error);
2498 }
2499
2500 /*
2501 * Readdir rpc call.
2502 * Called from below the buffer cache by nfs_doio().
2503 */
2504 int
2505 nfs_readdirrpc(vp, uiop, cred)
2506 struct vnode *vp;
2507 struct uio *uiop;
2508 kauth_cred_t cred;
2509 {
2510 int len, left;
2511 struct dirent *dp = NULL;
2512 u_int32_t *tl;
2513 char *cp;
2514 int32_t t1, t2;
2515 char *bpos, *dpos, *cp2;
2516 struct mbuf *mreq, *mrep, *md, *mb;
2517 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2518 struct nfsnode *dnp = VTONFS(vp);
2519 u_quad_t fileno;
2520 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1;
2521 #ifndef NFS_V2_ONLY
2522 int attrflag;
2523 #endif
2524 int nrpcs = 0, reclen;
2525 const int v3 = NFS_ISV3(vp);
2526
2527 #ifdef DIAGNOSTIC
2528 /*
2529 * Should be called from buffer cache, so only amount of
2530 * NFS_DIRBLKSIZ will be requested.
2531 */
2532 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2533 panic("nfs readdirrpc bad uio");
2534 #endif
2535
2536 /*
2537 * Loop around doing readdir rpc's of size nm_readdirsize
2538 * truncated to a multiple of NFS_DIRFRAGSIZ.
2539 * The stopping criteria is EOF or buffer full.
2540 */
2541 while (more_dirs && bigenough) {
2542 /*
2543 * Heuristic: don't bother to do another RPC to further
2544 * fill up this block if there is not much room left. (< 50%
2545 * of the readdir RPC size). This wastes some buffer space
2546 * but can save up to 50% in RPC calls.
2547 */
2548 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2549 bigenough = 0;
2550 break;
2551 }
2552 nfsstats.rpccnt[NFSPROC_READDIR]++;
2553 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2554 NFSX_READDIR(v3));
2555 nfsm_fhtom(dnp, v3);
2556 #ifndef NFS_V2_ONLY
2557 if (v3) {
2558 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2559 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2560 txdr_swapcookie3(uiop->uio_offset, tl);
2561 } else {
2562 txdr_cookie3(uiop->uio_offset, tl);
2563 }
2564 tl += 2;
2565 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2566 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2567 } else
2568 #endif
2569 {
2570 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2571 *tl++ = txdr_unsigned(uiop->uio_offset);
2572 }
2573 *tl = txdr_unsigned(nmp->nm_readdirsize);
2574 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred);
2575 nrpcs++;
2576 #ifndef NFS_V2_ONLY
2577 if (v3) {
2578 nfsm_postop_attr(vp, attrflag, 0);
2579 if (!error) {
2580 nfsm_dissect(tl, u_int32_t *,
2581 2 * NFSX_UNSIGNED);
2582 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2583 dnp->n_cookieverf.nfsuquad[1] = *tl;
2584 } else {
2585 m_freem(mrep);
2586 goto nfsmout;
2587 }
2588 }
2589 #endif
2590 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2591 more_dirs = fxdr_unsigned(int, *tl);
2592
2593 /* loop thru the dir entries, doctoring them to 4bsd form */
2594 while (more_dirs && bigenough) {
2595 #ifndef NFS_V2_ONLY
2596 if (v3) {
2597 nfsm_dissect(tl, u_int32_t *,
2598 3 * NFSX_UNSIGNED);
2599 fileno = fxdr_hyper(tl);
2600 len = fxdr_unsigned(int, *(tl + 2));
2601 } else
2602 #endif
2603 {
2604 nfsm_dissect(tl, u_int32_t *,
2605 2 * NFSX_UNSIGNED);
2606 fileno = fxdr_unsigned(u_quad_t, *tl++);
2607 len = fxdr_unsigned(int, *tl);
2608 }
2609 if (len <= 0 || len > NFS_MAXNAMLEN) {
2610 error = EBADRPC;
2611 m_freem(mrep);
2612 goto nfsmout;
2613 }
2614 /* for cookie stashing */
2615 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2616 left = NFS_DIRFRAGSIZ - blksiz;
2617 if (reclen > left) {
2618 memset(uiop->uio_iov->iov_base, 0, left);
2619 dp->d_reclen += left;
2620 UIO_ADVANCE(uiop, left);
2621 blksiz = 0;
2622 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2623 }
2624 if (reclen > uiop->uio_resid)
2625 bigenough = 0;
2626 if (bigenough) {
2627 int tlen;
2628
2629 dp = (struct dirent *)uiop->uio_iov->iov_base;
2630 dp->d_fileno = fileno;
2631 dp->d_namlen = len;
2632 dp->d_reclen = reclen;
2633 dp->d_type = DT_UNKNOWN;
2634 blksiz += reclen;
2635 if (blksiz == NFS_DIRFRAGSIZ)
2636 blksiz = 0;
2637 UIO_ADVANCE(uiop, DIRHDSIZ);
2638 nfsm_mtouio(uiop, len);
2639 tlen = reclen - (DIRHDSIZ + len);
2640 (void)memset(uiop->uio_iov->iov_base, 0, tlen);
2641 UIO_ADVANCE(uiop, tlen);
2642 } else
2643 nfsm_adv(nfsm_rndup(len));
2644 #ifndef NFS_V2_ONLY
2645 if (v3) {
2646 nfsm_dissect(tl, u_int32_t *,
2647 3 * NFSX_UNSIGNED);
2648 } else
2649 #endif
2650 {
2651 nfsm_dissect(tl, u_int32_t *,
2652 2 * NFSX_UNSIGNED);
2653 }
2654 if (bigenough) {
2655 #ifndef NFS_V2_ONLY
2656 if (v3) {
2657 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2658 uiop->uio_offset =
2659 fxdr_swapcookie3(tl);
2660 else
2661 uiop->uio_offset =
2662 fxdr_cookie3(tl);
2663 }
2664 else
2665 #endif
2666 {
2667 uiop->uio_offset =
2668 fxdr_unsigned(off_t, *tl);
2669 }
2670 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2671 }
2672 if (v3)
2673 tl += 2;
2674 else
2675 tl++;
2676 more_dirs = fxdr_unsigned(int, *tl);
2677 }
2678 /*
2679 * If at end of rpc data, get the eof boolean
2680 */
2681 if (!more_dirs) {
2682 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2683 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2684
2685 /*
2686 * kludge: if we got no entries, treat it as EOF.
2687 * some server sometimes send a reply without any
2688 * entries or EOF.
2689 * although it might mean the server has very long name,
2690 * we can't handle such entries anyway.
2691 */
2692
2693 if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2694 more_dirs = 0;
2695 }
2696 m_freem(mrep);
2697 }
2698 /*
2699 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2700 * by increasing d_reclen for the last record.
2701 */
2702 if (blksiz > 0) {
2703 left = NFS_DIRFRAGSIZ - blksiz;
2704 memset(uiop->uio_iov->iov_base, 0, left);
2705 dp->d_reclen += left;
2706 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2707 UIO_ADVANCE(uiop, left);
2708 }
2709
2710 /*
2711 * We are now either at the end of the directory or have filled the
2712 * block.
2713 */
2714 if (bigenough) {
2715 dnp->n_direofoffset = uiop->uio_offset;
2716 dnp->n_flag |= NEOFVALID;
2717 }
2718 nfsmout:
2719 return (error);
2720 }
2721
2722 #ifndef NFS_V2_ONLY
2723 /*
2724 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2725 */
2726 int
2727 nfs_readdirplusrpc(vp, uiop, cred)
2728 struct vnode *vp;
2729 struct uio *uiop;
2730 kauth_cred_t cred;
2731 {
2732 int len, left;
2733 struct dirent *dp = NULL;
2734 u_int32_t *tl;
2735 char *cp;
2736 int32_t t1, t2;
2737 struct vnode *newvp;
2738 char *bpos, *dpos, *cp2;
2739 struct mbuf *mreq, *mrep, *md, *mb;
2740 struct nameidata nami, *ndp = &nami;
2741 struct componentname *cnp = &ndp->ni_cnd;
2742 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2743 struct nfsnode *dnp = VTONFS(vp), *np;
2744 nfsfh_t *fhp;
2745 u_quad_t fileno;
2746 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2747 int attrflag, fhsize, nrpcs = 0, reclen;
2748 struct nfs_fattr fattr, *fp;
2749
2750 #ifdef DIAGNOSTIC
2751 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2752 panic("nfs readdirplusrpc bad uio");
2753 #endif
2754 ndp->ni_dvp = vp;
2755 newvp = NULLVP;
2756
2757 /*
2758 * Loop around doing readdir rpc's of size nm_readdirsize
2759 * truncated to a multiple of NFS_DIRFRAGSIZ.
2760 * The stopping criteria is EOF or buffer full.
2761 */
2762 while (more_dirs && bigenough) {
2763 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2764 bigenough = 0;
2765 break;
2766 }
2767 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2768 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2769 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2770 nfsm_fhtom(dnp, 1);
2771 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2772 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2773 txdr_swapcookie3(uiop->uio_offset, tl);
2774 } else {
2775 txdr_cookie3(uiop->uio_offset, tl);
2776 }
2777 tl += 2;
2778 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2779 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2780 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2781 *tl = txdr_unsigned(nmp->nm_rsize);
2782 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred);
2783 nfsm_postop_attr(vp, attrflag, 0);
2784 if (error) {
2785 m_freem(mrep);
2786 goto nfsmout;
2787 }
2788 nrpcs++;
2789 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2790 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2791 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2792 more_dirs = fxdr_unsigned(int, *tl);
2793
2794 /* loop thru the dir entries, doctoring them to 4bsd form */
2795 while (more_dirs && bigenough) {
2796 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2797 fileno = fxdr_hyper(tl);
2798 len = fxdr_unsigned(int, *(tl + 2));
2799 if (len <= 0 || len > NFS_MAXNAMLEN) {
2800 error = EBADRPC;
2801 m_freem(mrep);
2802 goto nfsmout;
2803 }
2804 /* for cookie stashing */
2805 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2806 left = NFS_DIRFRAGSIZ - blksiz;
2807 if (reclen > left) {
2808 /*
2809 * DIRFRAGSIZ is aligned, no need to align
2810 * again here.
2811 */
2812 memset(uiop->uio_iov->iov_base, 0, left);
2813 dp->d_reclen += left;
2814 UIO_ADVANCE(uiop, left);
2815 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2816 blksiz = 0;
2817 }
2818 if (reclen > uiop->uio_resid)
2819 bigenough = 0;
2820 if (bigenough) {
2821 int tlen;
2822
2823 dp = (struct dirent *)uiop->uio_iov->iov_base;
2824 dp->d_fileno = fileno;
2825 dp->d_namlen = len;
2826 dp->d_reclen = reclen;
2827 dp->d_type = DT_UNKNOWN;
2828 blksiz += reclen;
2829 if (blksiz == NFS_DIRFRAGSIZ)
2830 blksiz = 0;
2831 UIO_ADVANCE(uiop, DIRHDSIZ);
2832 nfsm_mtouio(uiop, len);
2833 tlen = reclen - (DIRHDSIZ + len);
2834 (void)memset(uiop->uio_iov->iov_base, 0, tlen);
2835 UIO_ADVANCE(uiop, tlen);
2836 cnp->cn_nameptr = dp->d_name;
2837 cnp->cn_namelen = dp->d_namlen;
2838 } else
2839 nfsm_adv(nfsm_rndup(len));
2840 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2841 if (bigenough) {
2842 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2843 uiop->uio_offset =
2844 fxdr_swapcookie3(tl);
2845 else
2846 uiop->uio_offset =
2847 fxdr_cookie3(tl);
2848 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2849 }
2850 tl += 2;
2851
2852 /*
2853 * Since the attributes are before the file handle
2854 * (sigh), we must skip over the attributes and then
2855 * come back and get them.
2856 */
2857 attrflag = fxdr_unsigned(int, *tl);
2858 if (attrflag) {
2859 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2860 memcpy(&fattr, fp, NFSX_V3FATTR);
2861 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2862 doit = fxdr_unsigned(int, *tl);
2863 if (doit) {
2864 nfsm_getfh(fhp, fhsize, 1);
2865 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2866 VREF(vp);
2867 newvp = vp;
2868 np = dnp;
2869 } else {
2870 error = nfs_nget1(vp->v_mount, fhp,
2871 fhsize, &np, LK_NOWAIT);
2872 if (!error)
2873 newvp = NFSTOV(np);
2874 }
2875 if (!error) {
2876 const char *xcp;
2877
2878 nfs_loadattrcache(&newvp, &fattr, 0, 0);
2879 if (bigenough) {
2880 dp->d_type =
2881 IFTODT(VTTOIF(np->n_vattr->va_type));
2882 if (cnp->cn_namelen <= NCHNAMLEN) {
2883 ndp->ni_vp = newvp;
2884 xcp = cnp->cn_nameptr +
2885 cnp->cn_namelen;
2886 cnp->cn_hash =
2887 namei_hash(cnp->cn_nameptr, &xcp);
2888 nfs_cache_enter(ndp->ni_dvp,
2889 ndp->ni_vp, cnp);
2890 }
2891 }
2892 }
2893 error = 0;
2894 }
2895 } else {
2896 /* Just skip over the file handle */
2897 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2898 i = fxdr_unsigned(int, *tl);
2899 nfsm_adv(nfsm_rndup(i));
2900 }
2901 if (newvp != NULLVP) {
2902 if (newvp == vp)
2903 vrele(newvp);
2904 else
2905 vput(newvp);
2906 newvp = NULLVP;
2907 }
2908 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2909 more_dirs = fxdr_unsigned(int, *tl);
2910 }
2911 /*
2912 * If at end of rpc data, get the eof boolean
2913 */
2914 if (!more_dirs) {
2915 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2916 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2917
2918 /*
2919 * kludge: see a comment in nfs_readdirrpc.
2920 */
2921
2922 if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2923 more_dirs = 0;
2924 }
2925 m_freem(mrep);
2926 }
2927 /*
2928 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2929 * by increasing d_reclen for the last record.
2930 */
2931 if (blksiz > 0) {
2932 left = NFS_DIRFRAGSIZ - blksiz;
2933 memset(uiop->uio_iov->iov_base, 0, left);
2934 dp->d_reclen += left;
2935 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2936 UIO_ADVANCE(uiop, left);
2937 }
2938
2939 /*
2940 * We are now either at the end of the directory or have filled the
2941 * block.
2942 */
2943 if (bigenough) {
2944 dnp->n_direofoffset = uiop->uio_offset;
2945 dnp->n_flag |= NEOFVALID;
2946 }
2947 nfsmout:
2948 if (newvp != NULLVP) {
2949 if(newvp == vp)
2950 vrele(newvp);
2951 else
2952 vput(newvp);
2953 }
2954 return (error);
2955 }
2956 #endif
2957
2958 /*
2959 * Silly rename. To make the NFS filesystem that is stateless look a little
2960 * more like the "ufs" a remove of an active vnode is translated to a rename
2961 * to a funny looking filename that is removed by nfs_inactive on the
2962 * nfsnode. There is the potential for another process on a different client
2963 * to create the same funny name between the nfs_lookitup() fails and the
2964 * nfs_rename() completes, but...
2965 */
2966 int
2967 nfs_sillyrename(dvp, vp, cnp, dolink)
2968 struct vnode *dvp, *vp;
2969 struct componentname *cnp;
2970 bool dolink;
2971 {
2972 struct sillyrename *sp;
2973 struct nfsnode *np;
2974 int error;
2975 short pid;
2976
2977 cache_purge(dvp);
2978 np = VTONFS(vp);
2979 #ifndef DIAGNOSTIC
2980 if (vp->v_type == VDIR)
2981 panic("nfs: sillyrename dir");
2982 #endif
2983 sp = kmem_alloc(sizeof(*sp), KM_SLEEP);
2984 sp->s_cred = kauth_cred_dup(cnp->cn_cred);
2985 sp->s_dvp = dvp;
2986 VREF(dvp);
2987
2988 /* Fudge together a funny name */
2989 pid = curlwp->l_proc->p_pid;
2990 memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2991 sp->s_namlen = 12;
2992 sp->s_name[8] = hexdigits[pid & 0xf];
2993 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf];
2994 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf];
2995 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf];
2996
2997 /* Try lookitups until we get one that isn't there */
2998 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2999 curlwp, (struct nfsnode **)0) == 0) {
3000 sp->s_name[4]++;
3001 if (sp->s_name[4] > 'z') {
3002 error = EINVAL;
3003 goto bad;
3004 }
3005 }
3006 if (dolink) {
3007 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen,
3008 sp->s_cred, curlwp);
3009 /*
3010 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
3011 */
3012 if (error == ENOTSUP) {
3013 error = nfs_renameit(dvp, cnp, sp);
3014 }
3015 } else {
3016 error = nfs_renameit(dvp, cnp, sp);
3017 }
3018 if (error)
3019 goto bad;
3020 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
3021 curlwp, &np);
3022 np->n_sillyrename = sp;
3023 return (0);
3024 bad:
3025 vrele(sp->s_dvp);
3026 kauth_cred_free(sp->s_cred);
3027 kmem_free(sp, sizeof(*sp));
3028 return (error);
3029 }
3030
3031 /*
3032 * Look up a file name and optionally either update the file handle or
3033 * allocate an nfsnode, depending on the value of npp.
3034 * npp == NULL --> just do the lookup
3035 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
3036 * handled too
3037 * *npp != NULL --> update the file handle in the vnode
3038 */
3039 int
3040 nfs_lookitup(dvp, name, len, cred, l, npp)
3041 struct vnode *dvp;
3042 const char *name;
3043 int len;
3044 kauth_cred_t cred;
3045 struct lwp *l;
3046 struct nfsnode **npp;
3047 {
3048 u_int32_t *tl;
3049 char *cp;
3050 int32_t t1, t2;
3051 struct vnode *newvp = (struct vnode *)0;
3052 struct nfsnode *np, *dnp = VTONFS(dvp);
3053 char *bpos, *dpos, *cp2;
3054 int error = 0, fhlen;
3055 #ifndef NFS_V2_ONLY
3056 int attrflag;
3057 #endif
3058 struct mbuf *mreq, *mrep, *md, *mb;
3059 nfsfh_t *nfhp;
3060 const int v3 = NFS_ISV3(dvp);
3061
3062 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
3063 nfsm_reqhead(dnp, NFSPROC_LOOKUP,
3064 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
3065 nfsm_fhtom(dnp, v3);
3066 nfsm_strtom(name, len, NFS_MAXNAMLEN);
3067 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred);
3068 if (npp && !error) {
3069 nfsm_getfh(nfhp, fhlen, v3);
3070 if (*npp) {
3071 np = *npp;
3072 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
3073 kmem_free(np->n_fhp, np->n_fhsize);
3074 np->n_fhp = &np->n_fh;
3075 }
3076 #if NFS_SMALLFH < NFSX_V3FHMAX
3077 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH)
3078 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP);
3079 #endif
3080 memcpy(np->n_fhp, nfhp, fhlen);
3081 np->n_fhsize = fhlen;
3082 newvp = NFSTOV(np);
3083 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
3084 VREF(dvp);
3085 newvp = dvp;
3086 np = dnp;
3087 } else {
3088 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
3089 if (error) {
3090 m_freem(mrep);
3091 return (error);
3092 }
3093 newvp = NFSTOV(np);
3094 }
3095 #ifndef NFS_V2_ONLY
3096 if (v3) {
3097 nfsm_postop_attr(newvp, attrflag, 0);
3098 if (!attrflag && *npp == NULL) {
3099 m_freem(mrep);
3100 vput(newvp);
3101 return (ENOENT);
3102 }
3103 } else
3104 #endif
3105 nfsm_loadattr(newvp, (struct vattr *)0, 0);
3106 }
3107 nfsm_reqdone;
3108 if (npp && *npp == NULL) {
3109 if (error) {
3110 if (newvp)
3111 vput(newvp);
3112 } else
3113 *npp = np;
3114 }
3115 return (error);
3116 }
3117
3118 #ifndef NFS_V2_ONLY
3119 /*
3120 * Nfs Version 3 commit rpc
3121 */
3122 int
3123 nfs_commit(vp, offset, cnt, l)
3124 struct vnode *vp;
3125 off_t offset;
3126 uint32_t cnt;
3127 struct lwp *l;
3128 {
3129 char *cp;
3130 u_int32_t *tl;
3131 int32_t t1, t2;
3132 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3133 char *bpos, *dpos, *cp2;
3134 int error = 0, wccflag = NFSV3_WCCRATTR;
3135 struct mbuf *mreq, *mrep, *md, *mb;
3136 struct nfsnode *np;
3137
3138 KASSERT(NFS_ISV3(vp));
3139
3140 #ifdef NFS_DEBUG_COMMIT
3141 printf("commit %lu - %lu\n", (unsigned long)offset,
3142 (unsigned long)(offset + cnt));
3143 #endif
3144
3145 mutex_enter(&nmp->nm_lock);
3146 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
3147 mutex_exit(&nmp->nm_lock);
3148 return (0);
3149 }
3150 mutex_exit(&nmp->nm_lock);
3151 nfsstats.rpccnt[NFSPROC_COMMIT]++;
3152 np = VTONFS(vp);
3153 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
3154 nfsm_fhtom(np, 1);
3155 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3156 txdr_hyper(offset, tl);
3157 tl += 2;
3158 *tl = txdr_unsigned(cnt);
3159 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred);
3160 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
3161 if (!error) {
3162 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
3163 mutex_enter(&nmp->nm_lock);
3164 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
3165 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
3166 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
3167 error = NFSERR_STALEWRITEVERF;
3168 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
3169 }
3170 mutex_exit(&nmp->nm_lock);
3171 }
3172 nfsm_reqdone;
3173 return (error);
3174 }
3175 #endif
3176
3177 /*
3178 * Kludge City..
3179 * - make nfs_bmap() essentially a no-op that does no translation
3180 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3181 * (Maybe I could use the process's page mapping, but I was concerned that
3182 * Kernel Write might not be enabled and also figured copyout() would do
3183 * a lot more work than memcpy() and also it currently happens in the
3184 * context of the swapper process (2).
3185 */
3186 int
3187 nfs_bmap(v)
3188 void *v;
3189 {
3190 struct vop_bmap_args /* {
3191 struct vnode *a_vp;
3192 daddr_t a_bn;
3193 struct vnode **a_vpp;
3194 daddr_t *a_bnp;
3195 int *a_runp;
3196 } */ *ap = v;
3197 struct vnode *vp = ap->a_vp;
3198 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
3199
3200 if (ap->a_vpp != NULL)
3201 *ap->a_vpp = vp;
3202 if (ap->a_bnp != NULL)
3203 *ap->a_bnp = ap->a_bn << bshift;
3204 if (ap->a_runp != NULL)
3205 *ap->a_runp = 1024 * 1024; /* XXX */
3206 return (0);
3207 }
3208
3209 /*
3210 * Strategy routine.
3211 * For async requests when nfsiod(s) are running, queue the request by
3212 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3213 * request.
3214 */
3215 int
3216 nfs_strategy(v)
3217 void *v;
3218 {
3219 struct vop_strategy_args *ap = v;
3220 struct buf *bp = ap->a_bp;
3221 int error = 0;
3222
3223 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3224 panic("nfs physio/async");
3225
3226 /*
3227 * If the op is asynchronous and an i/o daemon is waiting
3228 * queue the request, wake it up and wait for completion
3229 * otherwise just do it ourselves.
3230 */
3231 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp))
3232 error = nfs_doio(bp);
3233 return (error);
3234 }
3235
3236 /*
3237 * fsync vnode op. Just call nfs_flush() with commit == 1.
3238 */
3239 /* ARGSUSED */
3240 int
3241 nfs_fsync(v)
3242 void *v;
3243 {
3244 struct vop_fsync_args /* {
3245 struct vnodeop_desc *a_desc;
3246 struct vnode * a_vp;
3247 kauth_cred_t a_cred;
3248 int a_flags;
3249 off_t offlo;
3250 off_t offhi;
3251 struct lwp * a_l;
3252 } */ *ap = v;
3253
3254 struct vnode *vp = ap->a_vp;
3255
3256 if (vp->v_type != VREG)
3257 return 0;
3258
3259 return (nfs_flush(vp, ap->a_cred,
3260 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1));
3261 }
3262
3263 /*
3264 * Flush all the data associated with a vnode.
3265 */
3266 int
3267 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l,
3268 int commit)
3269 {
3270 struct nfsnode *np = VTONFS(vp);
3271 int error;
3272 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3273 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3274
3275 mutex_enter(&vp->v_interlock);
3276 error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3277 if (np->n_flag & NWRITEERR) {
3278 error = np->n_error;
3279 np->n_flag &= ~NWRITEERR;
3280 }
3281 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3282 return (error);
3283 }
3284
3285 /*
3286 * Return POSIX pathconf information applicable to nfs.
3287 *
3288 * N.B. The NFS V2 protocol doesn't support this RPC.
3289 */
3290 /* ARGSUSED */
3291 int
3292 nfs_pathconf(v)
3293 void *v;
3294 {
3295 struct vop_pathconf_args /* {
3296 struct vnode *a_vp;
3297 int a_name;
3298 register_t *a_retval;
3299 } */ *ap = v;
3300 struct nfsv3_pathconf *pcp;
3301 struct vnode *vp = ap->a_vp;
3302 struct mbuf *mreq, *mrep, *md, *mb;
3303 int32_t t1, t2;
3304 u_int32_t *tl;
3305 char *bpos, *dpos, *cp, *cp2;
3306 int error = 0, attrflag;
3307 #ifndef NFS_V2_ONLY
3308 struct nfsmount *nmp;
3309 unsigned int l;
3310 u_int64_t maxsize;
3311 #endif
3312 const int v3 = NFS_ISV3(vp);
3313 struct nfsnode *np = VTONFS(vp);
3314
3315 switch (ap->a_name) {
3316 /* Names that can be resolved locally. */
3317 case _PC_PIPE_BUF:
3318 *ap->a_retval = PIPE_BUF;
3319 break;
3320 case _PC_SYNC_IO:
3321 *ap->a_retval = 1;
3322 break;
3323 /* Names that cannot be resolved locally; do an RPC, if possible. */
3324 case _PC_LINK_MAX:
3325 case _PC_NAME_MAX:
3326 case _PC_CHOWN_RESTRICTED:
3327 case _PC_NO_TRUNC:
3328 if (!v3) {
3329 error = EINVAL;
3330 break;
3331 }
3332 nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3333 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3334 nfsm_fhtom(np, 1);
3335 nfsm_request(np, NFSPROC_PATHCONF,
3336 curlwp, curlwp->l_cred); /* XXX */
3337 nfsm_postop_attr(vp, attrflag, 0);
3338 if (!error) {
3339 nfsm_dissect(pcp, struct nfsv3_pathconf *,
3340 NFSX_V3PATHCONF);
3341 switch (ap->a_name) {
3342 case _PC_LINK_MAX:
3343 *ap->a_retval =
3344 fxdr_unsigned(register_t, pcp->pc_linkmax);
3345 break;
3346 case _PC_NAME_MAX:
3347 *ap->a_retval =
3348 fxdr_unsigned(register_t, pcp->pc_namemax);
3349 break;
3350 case _PC_CHOWN_RESTRICTED:
3351 *ap->a_retval =
3352 (pcp->pc_chownrestricted == nfs_true);
3353 break;
3354 case _PC_NO_TRUNC:
3355 *ap->a_retval =
3356 (pcp->pc_notrunc == nfs_true);
3357 break;
3358 }
3359 }
3360 nfsm_reqdone;
3361 break;
3362 case _PC_FILESIZEBITS:
3363 #ifndef NFS_V2_ONLY
3364 if (v3) {
3365 nmp = VFSTONFS(vp->v_mount);
3366 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3367 if ((error = nfs_fsinfo(nmp, vp,
3368 curlwp->l_cred, curlwp)) != 0) /* XXX */
3369 break;
3370 for (l = 0, maxsize = nmp->nm_maxfilesize;
3371 (maxsize >> l) > 0; l++)
3372 ;
3373 *ap->a_retval = l + 1;
3374 } else
3375 #endif
3376 {
3377 *ap->a_retval = 32; /* NFS V2 limitation */
3378 }
3379 break;
3380 default:
3381 error = EINVAL;
3382 break;
3383 }
3384
3385 return (error);
3386 }
3387
3388 /*
3389 * NFS advisory byte-level locks.
3390 */
3391 int
3392 nfs_advlock(v)
3393 void *v;
3394 {
3395 struct vop_advlock_args /* {
3396 struct vnode *a_vp;
3397 void *a_id;
3398 int a_op;
3399 struct flock *a_fl;
3400 int a_flags;
3401 } */ *ap = v;
3402 struct nfsnode *np = VTONFS(ap->a_vp);
3403
3404 return lf_advlock(ap, &np->n_lockf, np->n_size);
3405 }
3406
3407 /*
3408 * Print out the contents of an nfsnode.
3409 */
3410 int
3411 nfs_print(v)
3412 void *v;
3413 {
3414 struct vop_print_args /* {
3415 struct vnode *a_vp;
3416 } */ *ap = v;
3417 struct vnode *vp = ap->a_vp;
3418 struct nfsnode *np = VTONFS(vp);
3419
3420 printf("tag VT_NFS, fileid %lld fsid 0x%lx",
3421 (unsigned long long)np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3422 if (vp->v_type == VFIFO)
3423 fifo_printinfo(vp);
3424 printf("\n");
3425 return (0);
3426 }
3427
3428 /*
3429 * nfs unlock wrapper.
3430 */
3431 int
3432 nfs_unlock(void *v)
3433 {
3434 struct vop_unlock_args /* {
3435 struct vnode *a_vp;
3436 int a_flags;
3437 } */ *ap = v;
3438 struct vnode *vp = ap->a_vp;
3439
3440 /*
3441 * VOP_UNLOCK can be called by nfs_loadattrcache
3442 * with v_data == 0.
3443 */
3444 if (VTONFS(vp)) {
3445 nfs_delayedtruncate(vp);
3446 }
3447
3448 return genfs_unlock(v);
3449 }
3450
3451 /*
3452 * nfs special file access vnode op.
3453 * Essentially just get vattr and then imitate iaccess() since the device is
3454 * local to the client.
3455 */
3456 int
3457 nfsspec_access(v)
3458 void *v;
3459 {
3460 struct vop_access_args /* {
3461 struct vnode *a_vp;
3462 int a_mode;
3463 kauth_cred_t a_cred;
3464 struct lwp *a_l;
3465 } */ *ap = v;
3466 struct vattr va;
3467 struct vnode *vp = ap->a_vp;
3468 int error;
3469
3470 error = VOP_GETATTR(vp, &va, ap->a_cred);
3471 if (error)
3472 return (error);
3473
3474 /*
3475 * Disallow write attempts on filesystems mounted read-only;
3476 * unless the file is a socket, fifo, or a block or character
3477 * device resident on the filesystem.
3478 */
3479 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3480 switch (vp->v_type) {
3481 case VREG:
3482 case VDIR:
3483 case VLNK:
3484 return (EROFS);
3485 default:
3486 break;
3487 }
3488 }
3489
3490 return (vaccess(va.va_type, va.va_mode,
3491 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3492 }
3493
3494 /*
3495 * Read wrapper for special devices.
3496 */
3497 int
3498 nfsspec_read(v)
3499 void *v;
3500 {
3501 struct vop_read_args /* {
3502 struct vnode *a_vp;
3503 struct uio *a_uio;
3504 int a_ioflag;
3505 kauth_cred_t a_cred;
3506 } */ *ap = v;
3507 struct nfsnode *np = VTONFS(ap->a_vp);
3508
3509 /*
3510 * Set access flag.
3511 */
3512 np->n_flag |= NACC;
3513 getnanotime(&np->n_atim);
3514 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3515 }
3516
3517 /*
3518 * Write wrapper for special devices.
3519 */
3520 int
3521 nfsspec_write(v)
3522 void *v;
3523 {
3524 struct vop_write_args /* {
3525 struct vnode *a_vp;
3526 struct uio *a_uio;
3527 int a_ioflag;
3528 kauth_cred_t a_cred;
3529 } */ *ap = v;
3530 struct nfsnode *np = VTONFS(ap->a_vp);
3531
3532 /*
3533 * Set update flag.
3534 */
3535 np->n_flag |= NUPD;
3536 getnanotime(&np->n_mtim);
3537 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3538 }
3539
3540 /*
3541 * Close wrapper for special devices.
3542 *
3543 * Update the times on the nfsnode then do device close.
3544 */
3545 int
3546 nfsspec_close(v)
3547 void *v;
3548 {
3549 struct vop_close_args /* {
3550 struct vnode *a_vp;
3551 int a_fflag;
3552 kauth_cred_t a_cred;
3553 struct lwp *a_l;
3554 } */ *ap = v;
3555 struct vnode *vp = ap->a_vp;
3556 struct nfsnode *np = VTONFS(vp);
3557 struct vattr vattr;
3558
3559 if (np->n_flag & (NACC | NUPD)) {
3560 np->n_flag |= NCHG;
3561 if (vp->v_usecount == 1 &&
3562 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3563 VATTR_NULL(&vattr);
3564 if (np->n_flag & NACC)
3565 vattr.va_atime = np->n_atim;
3566 if (np->n_flag & NUPD)
3567 vattr.va_mtime = np->n_mtim;
3568 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3569 }
3570 }
3571 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3572 }
3573
3574 /*
3575 * Read wrapper for fifos.
3576 */
3577 int
3578 nfsfifo_read(v)
3579 void *v;
3580 {
3581 struct vop_read_args /* {
3582 struct vnode *a_vp;
3583 struct uio *a_uio;
3584 int a_ioflag;
3585 kauth_cred_t a_cred;
3586 } */ *ap = v;
3587 struct nfsnode *np = VTONFS(ap->a_vp);
3588
3589 /*
3590 * Set access flag.
3591 */
3592 np->n_flag |= NACC;
3593 getnanotime(&np->n_atim);
3594 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3595 }
3596
3597 /*
3598 * Write wrapper for fifos.
3599 */
3600 int
3601 nfsfifo_write(v)
3602 void *v;
3603 {
3604 struct vop_write_args /* {
3605 struct vnode *a_vp;
3606 struct uio *a_uio;
3607 int a_ioflag;
3608 kauth_cred_t a_cred;
3609 } */ *ap = v;
3610 struct nfsnode *np = VTONFS(ap->a_vp);
3611
3612 /*
3613 * Set update flag.
3614 */
3615 np->n_flag |= NUPD;
3616 getnanotime(&np->n_mtim);
3617 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3618 }
3619
3620 /*
3621 * Close wrapper for fifos.
3622 *
3623 * Update the times on the nfsnode then do fifo close.
3624 */
3625 int
3626 nfsfifo_close(v)
3627 void *v;
3628 {
3629 struct vop_close_args /* {
3630 struct vnode *a_vp;
3631 int a_fflag;
3632 kauth_cred_t a_cred;
3633 struct lwp *a_l;
3634 } */ *ap = v;
3635 struct vnode *vp = ap->a_vp;
3636 struct nfsnode *np = VTONFS(vp);
3637 struct vattr vattr;
3638
3639 if (np->n_flag & (NACC | NUPD)) {
3640 struct timespec ts;
3641
3642 getnanotime(&ts);
3643 if (np->n_flag & NACC)
3644 np->n_atim = ts;
3645 if (np->n_flag & NUPD)
3646 np->n_mtim = ts;
3647 np->n_flag |= NCHG;
3648 if (vp->v_usecount == 1 &&
3649 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3650 VATTR_NULL(&vattr);
3651 if (np->n_flag & NACC)
3652 vattr.va_atime = np->n_atim;
3653 if (np->n_flag & NUPD)
3654 vattr.va_mtime = np->n_mtim;
3655 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3656 }
3657 }
3658 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3659 }
Cache object: 97a69d927cf5bb5a08c71a473352159a
|