nfs_vnops.c revision 1.284 1 /* $NetBSD: nfs_vnops.c,v 1.284 2010/06/24 13:03:17 hannken 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 1.284 2010/06/24 13:03:17 hannken Exp $");
43
44 #ifdef _KERNEL_OPT
45 #include "opt_nfs.h"
46 #include "opt_uvmhist.h"
47 #endif
48
49 #include <sys/param.h>
50 #include <sys/proc.h>
51 #include <sys/kernel.h>
52 #include <sys/systm.h>
53 #include <sys/resourcevar.h>
54 #include <sys/mount.h>
55 #include <sys/buf.h>
56 #include <sys/condvar.h>
57 #include <sys/disk.h>
58 #include <sys/malloc.h>
59 #include <sys/kmem.h>
60 #include <sys/mbuf.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/vnode.h>
64 #include <sys/dirent.h>
65 #include <sys/fcntl.h>
66 #include <sys/hash.h>
67 #include <sys/lockf.h>
68 #include <sys/stat.h>
69 #include <sys/unistd.h>
70 #include <sys/kauth.h>
71
72 #include <uvm/uvm_extern.h>
73 #include <uvm/uvm.h>
74
75 #include <miscfs/fifofs/fifo.h>
76 #include <miscfs/genfs/genfs.h>
77 #include <miscfs/genfs/genfs_node.h>
78 #include <miscfs/specfs/specdev.h>
79
80 #include <nfs/rpcv2.h>
81 #include <nfs/nfsproto.h>
82 #include <nfs/nfs.h>
83 #include <nfs/nfsnode.h>
84 #include <nfs/nfsmount.h>
85 #include <nfs/xdr_subs.h>
86 #include <nfs/nfsm_subs.h>
87 #include <nfs/nfs_var.h>
88
89 #include <net/if.h>
90 #include <netinet/in.h>
91 #include <netinet/in_var.h>
92
93 /*
94 * Global vfs data structures for nfs
95 */
96 int (**nfsv2_vnodeop_p)(void *);
97 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
98 { &vop_default_desc, vn_default_error },
99 { &vop_lookup_desc, nfs_lookup }, /* lookup */
100 { &vop_create_desc, nfs_create }, /* create */
101 { &vop_mknod_desc, nfs_mknod }, /* mknod */
102 { &vop_open_desc, nfs_open }, /* open */
103 { &vop_close_desc, nfs_close }, /* close */
104 { &vop_access_desc, nfs_access }, /* access */
105 { &vop_getattr_desc, nfs_getattr }, /* getattr */
106 { &vop_setattr_desc, nfs_setattr }, /* setattr */
107 { &vop_read_desc, nfs_read }, /* read */
108 { &vop_write_desc, nfs_write }, /* write */
109 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
110 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */
111 { &vop_poll_desc, nfs_poll }, /* poll */
112 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */
113 { &vop_revoke_desc, nfs_revoke }, /* revoke */
114 { &vop_mmap_desc, nfs_mmap }, /* mmap */
115 { &vop_fsync_desc, nfs_fsync }, /* fsync */
116 { &vop_seek_desc, nfs_seek }, /* seek */
117 { &vop_remove_desc, nfs_remove }, /* remove */
118 { &vop_link_desc, nfs_link }, /* link */
119 { &vop_rename_desc, nfs_rename }, /* rename */
120 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */
121 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */
122 { &vop_symlink_desc, nfs_symlink }, /* symlink */
123 { &vop_readdir_desc, nfs_readdir }, /* readdir */
124 { &vop_readlink_desc, nfs_readlink }, /* readlink */
125 { &vop_abortop_desc, nfs_abortop }, /* abortop */
126 { &vop_inactive_desc, nfs_inactive }, /* inactive */
127 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
128 { &vop_lock_desc, nfs_lock }, /* lock */
129 { &vop_unlock_desc, nfs_unlock }, /* unlock */
130 { &vop_bmap_desc, nfs_bmap }, /* bmap */
131 { &vop_strategy_desc, nfs_strategy }, /* strategy */
132 { &vop_print_desc, nfs_print }, /* print */
133 { &vop_islocked_desc, nfs_islocked }, /* islocked */
134 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */
135 { &vop_advlock_desc, nfs_advlock }, /* advlock */
136 { &vop_bwrite_desc, genfs_badop }, /* bwrite */
137 { &vop_getpages_desc, nfs_getpages }, /* getpages */
138 { &vop_putpages_desc, genfs_putpages }, /* putpages */
139 { NULL, NULL }
140 };
141 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
142 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
143
144 /*
145 * Special device vnode ops
146 */
147 int (**spec_nfsv2nodeop_p)(void *);
148 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
149 { &vop_default_desc, vn_default_error },
150 { &vop_lookup_desc, spec_lookup }, /* lookup */
151 { &vop_create_desc, spec_create }, /* create */
152 { &vop_mknod_desc, spec_mknod }, /* mknod */
153 { &vop_open_desc, spec_open }, /* open */
154 { &vop_close_desc, nfsspec_close }, /* close */
155 { &vop_access_desc, nfsspec_access }, /* access */
156 { &vop_getattr_desc, nfs_getattr }, /* getattr */
157 { &vop_setattr_desc, nfs_setattr }, /* setattr */
158 { &vop_read_desc, nfsspec_read }, /* read */
159 { &vop_write_desc, nfsspec_write }, /* write */
160 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
161 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */
162 { &vop_poll_desc, spec_poll }, /* poll */
163 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */
164 { &vop_revoke_desc, spec_revoke }, /* revoke */
165 { &vop_mmap_desc, spec_mmap }, /* mmap */
166 { &vop_fsync_desc, spec_fsync }, /* fsync */
167 { &vop_seek_desc, spec_seek }, /* seek */
168 { &vop_remove_desc, spec_remove }, /* remove */
169 { &vop_link_desc, spec_link }, /* link */
170 { &vop_rename_desc, spec_rename }, /* rename */
171 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */
172 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */
173 { &vop_symlink_desc, spec_symlink }, /* symlink */
174 { &vop_readdir_desc, spec_readdir }, /* readdir */
175 { &vop_readlink_desc, spec_readlink }, /* readlink */
176 { &vop_abortop_desc, spec_abortop }, /* abortop */
177 { &vop_inactive_desc, nfs_inactive }, /* inactive */
178 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
179 { &vop_lock_desc, nfs_lock }, /* lock */
180 { &vop_unlock_desc, nfs_unlock }, /* unlock */
181 { &vop_bmap_desc, spec_bmap }, /* bmap */
182 { &vop_strategy_desc, spec_strategy }, /* strategy */
183 { &vop_print_desc, nfs_print }, /* print */
184 { &vop_islocked_desc, nfs_islocked }, /* islocked */
185 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */
186 { &vop_advlock_desc, spec_advlock }, /* advlock */
187 { &vop_bwrite_desc, spec_bwrite }, /* bwrite */
188 { &vop_getpages_desc, spec_getpages }, /* getpages */
189 { &vop_putpages_desc, spec_putpages }, /* putpages */
190 { NULL, NULL }
191 };
192 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
193 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
194
195 int (**fifo_nfsv2nodeop_p)(void *);
196 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
197 { &vop_default_desc, vn_default_error },
198 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup */
199 { &vop_create_desc, vn_fifo_bypass }, /* create */
200 { &vop_mknod_desc, vn_fifo_bypass }, /* mknod */
201 { &vop_open_desc, vn_fifo_bypass }, /* open */
202 { &vop_close_desc, nfsfifo_close }, /* close */
203 { &vop_access_desc, nfsspec_access }, /* access */
204 { &vop_getattr_desc, nfs_getattr }, /* getattr */
205 { &vop_setattr_desc, nfs_setattr }, /* setattr */
206 { &vop_read_desc, nfsfifo_read }, /* read */
207 { &vop_write_desc, nfsfifo_write }, /* write */
208 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
209 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */
210 { &vop_poll_desc, vn_fifo_bypass }, /* poll */
211 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */
212 { &vop_revoke_desc, vn_fifo_bypass }, /* revoke */
213 { &vop_mmap_desc, vn_fifo_bypass }, /* mmap */
214 { &vop_fsync_desc, nfs_fsync }, /* fsync */
215 { &vop_seek_desc, vn_fifo_bypass }, /* seek */
216 { &vop_remove_desc, vn_fifo_bypass }, /* remove */
217 { &vop_link_desc, vn_fifo_bypass }, /* link */
218 { &vop_rename_desc, vn_fifo_bypass }, /* rename */
219 { &vop_mkdir_desc, vn_fifo_bypass }, /* mkdir */
220 { &vop_rmdir_desc, vn_fifo_bypass }, /* rmdir */
221 { &vop_symlink_desc, vn_fifo_bypass }, /* symlink */
222 { &vop_readdir_desc, vn_fifo_bypass }, /* readdir */
223 { &vop_readlink_desc, vn_fifo_bypass }, /* readlink */
224 { &vop_abortop_desc, vn_fifo_bypass }, /* abortop */
225 { &vop_inactive_desc, nfs_inactive }, /* inactive */
226 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
227 { &vop_lock_desc, nfs_lock }, /* lock */
228 { &vop_unlock_desc, nfs_unlock }, /* unlock */
229 { &vop_bmap_desc, vn_fifo_bypass }, /* bmap */
230 { &vop_strategy_desc, genfs_badop }, /* strategy */
231 { &vop_print_desc, nfs_print }, /* print */
232 { &vop_islocked_desc, nfs_islocked }, /* islocked */
233 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */
234 { &vop_advlock_desc, vn_fifo_bypass }, /* advlock */
235 { &vop_bwrite_desc, genfs_badop }, /* bwrite */
236 { &vop_putpages_desc, vn_fifo_bypass }, /* putpages */
237 { NULL, NULL }
238 };
239 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
240 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
241
242 static int nfs_linkrpc(struct vnode *, struct vnode *, const char *,
243 size_t, kauth_cred_t, struct lwp *);
244 static void nfs_writerpc_extfree(struct mbuf *, void *, size_t, void *);
245
246 /*
247 * Global variables
248 */
249 extern u_int32_t nfs_true, nfs_false;
250 extern u_int32_t nfs_xdrneg1;
251 extern const nfstype nfsv3_type[9];
252
253 int nfs_numasync = 0;
254 #define DIRHDSIZ _DIRENT_NAMEOFF(dp)
255 #define UIO_ADVANCE(uio, siz) \
256 (void)((uio)->uio_resid -= (siz), \
257 (uio)->uio_iov->iov_base = (char *)(uio)->uio_iov->iov_base + (siz), \
258 (uio)->uio_iov->iov_len -= (siz))
259
260 static void nfs_cache_enter(struct vnode *, struct vnode *,
261 struct componentname *);
262
263 static void
264 nfs_cache_enter(struct vnode *dvp, struct vnode *vp,
265 struct componentname *cnp)
266 {
267 struct nfsnode *dnp = VTONFS(dvp);
268
269 if (vp != NULL) {
270 struct nfsnode *np = VTONFS(vp);
271
272 np->n_ctime = np->n_vattr->va_ctime.tv_sec;
273 }
274
275 if (!timespecisset(&dnp->n_nctime))
276 dnp->n_nctime = dnp->n_vattr->va_mtime;
277
278 cache_enter(dvp, vp, cnp);
279 }
280
281 /*
282 * nfs null call from vfs.
283 */
284 int
285 nfs_null(struct vnode *vp, kauth_cred_t cred, struct lwp *l)
286 {
287 char *bpos, *dpos;
288 int error = 0;
289 struct mbuf *mreq, *mrep, *md, *mb;
290 struct nfsnode *np = VTONFS(vp);
291
292 nfsm_reqhead(np, NFSPROC_NULL, 0);
293 nfsm_request(np, NFSPROC_NULL, l, cred);
294 nfsm_reqdone;
295 return (error);
296 }
297
298 /*
299 * nfs access vnode op.
300 * For nfs version 2, just return ok. File accesses may fail later.
301 * For nfs version 3, use the access rpc to check accessibility. If file modes
302 * are changed on the server, accesses might still fail later.
303 */
304 int
305 nfs_access(void *v)
306 {
307 struct vop_access_args /* {
308 struct vnode *a_vp;
309 int a_mode;
310 kauth_cred_t a_cred;
311 } */ *ap = v;
312 struct vnode *vp = ap->a_vp;
313 #ifndef NFS_V2_ONLY
314 u_int32_t *tl;
315 char *cp;
316 int32_t t1, t2;
317 char *bpos, *dpos, *cp2;
318 int error = 0, attrflag;
319 struct mbuf *mreq, *mrep, *md, *mb;
320 u_int32_t mode, rmode;
321 const int v3 = NFS_ISV3(vp);
322 #endif
323 int cachevalid;
324 struct nfsnode *np = VTONFS(vp);
325 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
326
327 cachevalid = (np->n_accstamp != -1 &&
328 (time_uptime - np->n_accstamp) < nfs_attrtimeo(nmp, np) &&
329 np->n_accuid == kauth_cred_geteuid(ap->a_cred));
330
331 /*
332 * Check access cache first. If this request has been made for this
333 * uid shortly before, use the cached result.
334 */
335 if (cachevalid) {
336 if (!np->n_accerror) {
337 if ((np->n_accmode & ap->a_mode) == ap->a_mode)
338 return np->n_accerror;
339 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
340 return np->n_accerror;
341 }
342
343 #ifndef NFS_V2_ONLY
344 /*
345 * For nfs v3, do an access rpc, otherwise you are stuck emulating
346 * ufs_access() locally using the vattr. This may not be correct,
347 * since the server may apply other access criteria such as
348 * client uid-->server uid mapping that we do not know about, but
349 * this is better than just returning anything that is lying about
350 * in the cache.
351 */
352 if (v3) {
353 nfsstats.rpccnt[NFSPROC_ACCESS]++;
354 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
355 nfsm_fhtom(np, v3);
356 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
357 if (ap->a_mode & VREAD)
358 mode = NFSV3ACCESS_READ;
359 else
360 mode = 0;
361 if (vp->v_type != VDIR) {
362 if (ap->a_mode & VWRITE)
363 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
364 if (ap->a_mode & VEXEC)
365 mode |= NFSV3ACCESS_EXECUTE;
366 } else {
367 if (ap->a_mode & VWRITE)
368 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
369 NFSV3ACCESS_DELETE);
370 if (ap->a_mode & VEXEC)
371 mode |= NFSV3ACCESS_LOOKUP;
372 }
373 *tl = txdr_unsigned(mode);
374 nfsm_request(np, NFSPROC_ACCESS, curlwp, ap->a_cred);
375 nfsm_postop_attr(vp, attrflag, 0);
376 if (!error) {
377 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
378 rmode = fxdr_unsigned(u_int32_t, *tl);
379 /*
380 * The NFS V3 spec does not clarify whether or not
381 * the returned access bits can be a superset of
382 * the ones requested, so...
383 */
384 if ((rmode & mode) != mode)
385 error = EACCES;
386 }
387 nfsm_reqdone;
388 } else
389 #endif
390 return (nfsspec_access(ap));
391 #ifndef NFS_V2_ONLY
392 /*
393 * Disallow write attempts on filesystems mounted read-only;
394 * unless the file is a socket, fifo, or a block or character
395 * device resident on the filesystem.
396 */
397 if (!error && (ap->a_mode & VWRITE) &&
398 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
399 switch (vp->v_type) {
400 case VREG:
401 case VDIR:
402 case VLNK:
403 error = EROFS;
404 default:
405 break;
406 }
407 }
408
409 if (!error || error == EACCES) {
410 /*
411 * If we got the same result as for a previous,
412 * different request, OR it in. Don't update
413 * the timestamp in that case.
414 */
415 if (cachevalid && np->n_accstamp != -1 &&
416 error == np->n_accerror) {
417 if (!error)
418 np->n_accmode |= ap->a_mode;
419 else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
420 np->n_accmode = ap->a_mode;
421 } else {
422 np->n_accstamp = time_uptime;
423 np->n_accuid = kauth_cred_geteuid(ap->a_cred);
424 np->n_accmode = ap->a_mode;
425 np->n_accerror = error;
426 }
427 }
428
429 return (error);
430 #endif
431 }
432
433 /*
434 * nfs open vnode op
435 * Check to see if the type is ok
436 * and that deletion is not in progress.
437 * For paged in text files, you will need to flush the page cache
438 * if consistency is lost.
439 */
440 /* ARGSUSED */
441 int
442 nfs_open(void *v)
443 {
444 struct vop_open_args /* {
445 struct vnode *a_vp;
446 int a_mode;
447 kauth_cred_t a_cred;
448 } */ *ap = v;
449 struct vnode *vp = ap->a_vp;
450 struct nfsnode *np = VTONFS(vp);
451 int error;
452
453 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
454 return (EACCES);
455 }
456
457 if (ap->a_mode & FREAD) {
458 if (np->n_rcred != NULL)
459 kauth_cred_free(np->n_rcred);
460 np->n_rcred = ap->a_cred;
461 kauth_cred_hold(np->n_rcred);
462 }
463 if (ap->a_mode & FWRITE) {
464 if (np->n_wcred != NULL)
465 kauth_cred_free(np->n_wcred);
466 np->n_wcred = ap->a_cred;
467 kauth_cred_hold(np->n_wcred);
468 }
469
470 error = nfs_flushstalebuf(vp, ap->a_cred, curlwp, 0);
471 if (error)
472 return error;
473
474 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */
475
476 return (0);
477 }
478
479 /*
480 * nfs close vnode op
481 * What an NFS client should do upon close after writing is a debatable issue.
482 * Most NFS clients push delayed writes to the server upon close, basically for
483 * two reasons:
484 * 1 - So that any write errors may be reported back to the client process
485 * doing the close system call. By far the two most likely errors are
486 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
487 * 2 - To put a worst case upper bound on cache inconsistency between
488 * multiple clients for the file.
489 * There is also a consistency problem for Version 2 of the protocol w.r.t.
490 * not being able to tell if other clients are writing a file concurrently,
491 * since there is no way of knowing if the changed modify time in the reply
492 * is only due to the write for this client.
493 * (NFS Version 3 provides weak cache consistency data in the reply that
494 * should be sufficient to detect and handle this case.)
495 *
496 * The current code does the following:
497 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
498 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
499 * or commit them (this satisfies 1 and 2 except for the
500 * case where the server crashes after this close but
501 * before the commit RPC, which is felt to be "good
502 * enough". Changing the last argument to nfs_flush() to
503 * a 1 would force a commit operation, if it is felt a
504 * commit is necessary now.
505 */
506 /* ARGSUSED */
507 int
508 nfs_close(void *v)
509 {
510 struct vop_close_args /* {
511 struct vnodeop_desc *a_desc;
512 struct vnode *a_vp;
513 int a_fflag;
514 kauth_cred_t a_cred;
515 } */ *ap = v;
516 struct vnode *vp = ap->a_vp;
517 struct nfsnode *np = VTONFS(vp);
518 int error = 0;
519 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
520
521 if (vp->v_type == VREG) {
522 if (np->n_flag & NMODIFIED) {
523 #ifndef NFS_V2_ONLY
524 if (NFS_ISV3(vp)) {
525 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, curlwp, 0);
526 np->n_flag &= ~NMODIFIED;
527 } else
528 #endif
529 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, curlwp, 1);
530 NFS_INVALIDATE_ATTRCACHE(np);
531 }
532 if (np->n_flag & NWRITEERR) {
533 np->n_flag &= ~NWRITEERR;
534 error = np->n_error;
535 }
536 }
537 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
538 return (error);
539 }
540
541 /*
542 * nfs getattr call from vfs.
543 */
544 int
545 nfs_getattr(void *v)
546 {
547 struct vop_getattr_args /* {
548 struct vnode *a_vp;
549 struct vattr *a_vap;
550 kauth_cred_t a_cred;
551 } */ *ap = v;
552 struct vnode *vp = ap->a_vp;
553 struct nfsnode *np = VTONFS(vp);
554 char *cp;
555 u_int32_t *tl;
556 int32_t t1, t2;
557 char *bpos, *dpos;
558 int error = 0;
559 struct mbuf *mreq, *mrep, *md, *mb;
560 const int v3 = NFS_ISV3(vp);
561
562 /*
563 * Update local times for special files.
564 */
565 if (np->n_flag & (NACC | NUPD))
566 np->n_flag |= NCHG;
567
568 /*
569 * if we have delayed truncation, do it now.
570 */
571 nfs_delayedtruncate(vp);
572
573 /*
574 * First look in the cache.
575 */
576 if (nfs_getattrcache(vp, ap->a_vap) == 0)
577 return (0);
578 nfsstats.rpccnt[NFSPROC_GETATTR]++;
579 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
580 nfsm_fhtom(np, v3);
581 nfsm_request(np, NFSPROC_GETATTR, curlwp, ap->a_cred);
582 if (!error) {
583 nfsm_loadattr(vp, ap->a_vap, 0);
584 if (vp->v_type == VDIR &&
585 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
586 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
587 }
588 nfsm_reqdone;
589 return (error);
590 }
591
592 /*
593 * nfs setattr call.
594 */
595 int
596 nfs_setattr(void *v)
597 {
598 struct vop_setattr_args /* {
599 struct vnodeop_desc *a_desc;
600 struct vnode *a_vp;
601 struct vattr *a_vap;
602 kauth_cred_t a_cred;
603 } */ *ap = v;
604 struct vnode *vp = ap->a_vp;
605 struct nfsnode *np = VTONFS(vp);
606 struct vattr *vap = ap->a_vap;
607 int error = 0;
608 u_quad_t tsize = 0;
609
610 /*
611 * Setting of flags is not supported.
612 */
613 if (vap->va_flags != VNOVAL)
614 return (EOPNOTSUPP);
615
616 /*
617 * Disallow write attempts if the filesystem is mounted read-only.
618 */
619 if ((vap->va_uid != (uid_t)VNOVAL ||
620 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
621 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
622 (vp->v_mount->mnt_flag & MNT_RDONLY))
623 return (EROFS);
624 if (vap->va_size != VNOVAL) {
625 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) {
626 return EFBIG;
627 }
628 switch (vp->v_type) {
629 case VDIR:
630 return (EISDIR);
631 case VCHR:
632 case VBLK:
633 case VSOCK:
634 case VFIFO:
635 if (vap->va_mtime.tv_sec == VNOVAL &&
636 vap->va_atime.tv_sec == VNOVAL &&
637 vap->va_mode == (mode_t)VNOVAL &&
638 vap->va_uid == (uid_t)VNOVAL &&
639 vap->va_gid == (gid_t)VNOVAL)
640 return (0);
641 vap->va_size = VNOVAL;
642 break;
643 default:
644 /*
645 * Disallow write attempts if the filesystem is
646 * mounted read-only.
647 */
648 if (vp->v_mount->mnt_flag & MNT_RDONLY)
649 return (EROFS);
650 genfs_node_wrlock(vp);
651 uvm_vnp_setsize(vp, vap->va_size);
652 tsize = np->n_size;
653 np->n_size = vap->va_size;
654 if (vap->va_size == 0)
655 error = nfs_vinvalbuf(vp, 0,
656 ap->a_cred, curlwp, 1);
657 else
658 error = nfs_vinvalbuf(vp, V_SAVE,
659 ap->a_cred, curlwp, 1);
660 if (error) {
661 uvm_vnp_setsize(vp, tsize);
662 genfs_node_unlock(vp);
663 return (error);
664 }
665 np->n_vattr->va_size = vap->va_size;
666 }
667 } else {
668 /*
669 * flush files before setattr because a later write of
670 * cached data might change timestamps or reset sugid bits
671 */
672 if ((vap->va_mtime.tv_sec != VNOVAL ||
673 vap->va_atime.tv_sec != VNOVAL ||
674 vap->va_mode != VNOVAL) &&
675 vp->v_type == VREG &&
676 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
677 curlwp, 1)) == EINTR)
678 return (error);
679 }
680 error = nfs_setattrrpc(vp, vap, ap->a_cred, curlwp);
681 if (vap->va_size != VNOVAL) {
682 if (error) {
683 np->n_size = np->n_vattr->va_size = tsize;
684 uvm_vnp_setsize(vp, np->n_size);
685 }
686 genfs_node_unlock(vp);
687 }
688 VN_KNOTE(vp, NOTE_ATTRIB);
689 return (error);
690 }
691
692 /*
693 * Do an nfs setattr rpc.
694 */
695 int
696 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, struct lwp *l)
697 {
698 struct nfsv2_sattr *sp;
699 char *cp;
700 int32_t t1, t2;
701 char *bpos, *dpos;
702 u_int32_t *tl;
703 int error = 0;
704 struct mbuf *mreq, *mrep, *md, *mb;
705 const int v3 = NFS_ISV3(vp);
706 struct nfsnode *np = VTONFS(vp);
707 #ifndef NFS_V2_ONLY
708 int wccflag = NFSV3_WCCRATTR;
709 char *cp2;
710 #endif
711
712 nfsstats.rpccnt[NFSPROC_SETATTR]++;
713 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
714 nfsm_fhtom(np, v3);
715 #ifndef NFS_V2_ONLY
716 if (v3) {
717 nfsm_v3attrbuild(vap, true);
718 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
719 *tl = nfs_false;
720 } else {
721 #endif
722 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
723 if (vap->va_mode == (mode_t)VNOVAL)
724 sp->sa_mode = nfs_xdrneg1;
725 else
726 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
727 if (vap->va_uid == (uid_t)VNOVAL)
728 sp->sa_uid = nfs_xdrneg1;
729 else
730 sp->sa_uid = txdr_unsigned(vap->va_uid);
731 if (vap->va_gid == (gid_t)VNOVAL)
732 sp->sa_gid = nfs_xdrneg1;
733 else
734 sp->sa_gid = txdr_unsigned(vap->va_gid);
735 sp->sa_size = txdr_unsigned(vap->va_size);
736 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
737 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
738 #ifndef NFS_V2_ONLY
739 }
740 #endif
741 nfsm_request(np, NFSPROC_SETATTR, l, cred);
742 #ifndef NFS_V2_ONLY
743 if (v3) {
744 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
745 } else
746 #endif
747 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
748 nfsm_reqdone;
749 return (error);
750 }
751
752 /*
753 * nfs lookup call, one step at a time...
754 * First look in cache
755 * If not found, do the rpc.
756 */
757 int
758 nfs_lookup(void *v)
759 {
760 struct vop_lookup_args /* {
761 struct vnodeop_desc *a_desc;
762 struct vnode *a_dvp;
763 struct vnode **a_vpp;
764 struct componentname *a_cnp;
765 } */ *ap = v;
766 struct componentname *cnp = ap->a_cnp;
767 struct vnode *dvp = ap->a_dvp;
768 struct vnode **vpp = ap->a_vpp;
769 int flags;
770 struct vnode *newvp;
771 u_int32_t *tl;
772 char *cp;
773 int32_t t1, t2;
774 char *bpos, *dpos, *cp2;
775 struct mbuf *mreq, *mrep, *md, *mb;
776 long len;
777 nfsfh_t *fhp;
778 struct nfsnode *np;
779 int error = 0, attrflag, fhsize;
780 const int v3 = NFS_ISV3(dvp);
781
782 flags = cnp->cn_flags;
783
784 *vpp = NULLVP;
785 newvp = NULLVP;
786 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
787 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
788 return (EROFS);
789 if (dvp->v_type != VDIR)
790 return (ENOTDIR);
791
792 /*
793 * RFC1813(nfsv3) 3.2 says clients should handle "." by themselves.
794 */
795 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
796 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
797 if (error)
798 return error;
799 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN))
800 return EISDIR;
801 vref(dvp);
802 *vpp = dvp;
803 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
804 cnp->cn_flags |= SAVENAME;
805 return 0;
806 }
807
808 np = VTONFS(dvp);
809
810 /*
811 * Before performing an RPC, check the name cache to see if
812 * the directory/name pair we are looking for is known already.
813 * If the directory/name pair is found in the name cache,
814 * we have to ensure the directory has not changed from
815 * the time the cache entry has been created. If it has,
816 * the cache entry has to be ignored.
817 */
818 error = cache_lookup_raw(dvp, vpp, cnp);
819 KASSERT(dvp != *vpp);
820 KASSERT((cnp->cn_flags & ISWHITEOUT) == 0);
821 if (error >= 0) {
822 struct vattr vattr;
823 int err2;
824
825 if (error && error != ENOENT) {
826 *vpp = NULLVP;
827 return error;
828 }
829
830 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred);
831 if (err2 != 0) {
832 if (error == 0)
833 vrele(*vpp);
834 *vpp = NULLVP;
835 return err2;
836 }
837
838 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred)
839 || timespeccmp(&vattr.va_mtime,
840 &VTONFS(dvp)->n_nctime, !=)) {
841 if (error == 0) {
842 vrele(*vpp);
843 *vpp = NULLVP;
844 }
845 cache_purge1(dvp, NULL, PURGE_CHILDREN);
846 timespecclear(&np->n_nctime);
847 goto dorpc;
848 }
849
850 if (error == ENOENT) {
851 goto noentry;
852 }
853
854 /*
855 * investigate the vnode returned by cache_lookup_raw.
856 * if it isn't appropriate, do an rpc.
857 */
858 newvp = *vpp;
859 if ((flags & ISDOTDOT) != 0) {
860 VOP_UNLOCK(dvp);
861 }
862 error = vn_lock(newvp, LK_EXCLUSIVE);
863 if ((flags & ISDOTDOT) != 0) {
864 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
865 }
866 if (error != 0) {
867 /* newvp has been reclaimed. */
868 vrele(newvp);
869 *vpp = NULLVP;
870 goto dorpc;
871 }
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 (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
876 cnp->cn_flags |= SAVENAME;
877 KASSERT(newvp->v_type != VNON);
878 return (0);
879 }
880 cache_purge1(newvp, NULL, PURGE_PARENTS);
881 vput(newvp);
882 *vpp = NULLVP;
883 }
884 dorpc:
885 #if 0
886 /*
887 * because nfsv3 has the same CREATE semantics as ours,
888 * we don't have to perform LOOKUPs beforehand.
889 *
890 * XXX ideally we can do the same for nfsv2 in the case of !O_EXCL.
891 * XXX although we have no way to know if O_EXCL is requested or not.
892 */
893
894 if (v3 && cnp->cn_nameiop == CREATE &&
895 (flags & (ISLASTCN|ISDOTDOT)) == ISLASTCN &&
896 (dvp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
897 cnp->cn_flags |= SAVENAME;
898 return (EJUSTRETURN);
899 }
900 #endif /* 0 */
901
902 error = 0;
903 newvp = NULLVP;
904 nfsstats.lookupcache_misses++;
905 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
906 len = cnp->cn_namelen;
907 nfsm_reqhead(np, NFSPROC_LOOKUP,
908 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
909 nfsm_fhtom(np, v3);
910 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
911 nfsm_request(np, NFSPROC_LOOKUP, curlwp, cnp->cn_cred);
912 if (error) {
913 nfsm_postop_attr(dvp, attrflag, 0);
914 m_freem(mrep);
915 goto nfsmout;
916 }
917 nfsm_getfh(fhp, fhsize, v3);
918
919 /*
920 * Handle RENAME case...
921 */
922 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
923 if (NFS_CMPFH(np, fhp, fhsize)) {
924 m_freem(mrep);
925 return (EISDIR);
926 }
927 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
928 if (error) {
929 m_freem(mrep);
930 return error;
931 }
932 newvp = NFSTOV(np);
933 #ifndef NFS_V2_ONLY
934 if (v3) {
935 nfsm_postop_attr(newvp, attrflag, 0);
936 nfsm_postop_attr(dvp, attrflag, 0);
937 } else
938 #endif
939 nfsm_loadattr(newvp, (struct vattr *)0, 0);
940 *vpp = newvp;
941 m_freem(mrep);
942 cnp->cn_flags |= SAVENAME;
943 goto validate;
944 }
945
946 /*
947 * The postop attr handling is duplicated for each if case,
948 * because it should be done while dvp is locked (unlocking
949 * dvp is different for each case).
950 */
951
952 if (NFS_CMPFH(np, fhp, fhsize)) {
953 /*
954 * as we handle "." lookup locally, this should be
955 * a broken server.
956 */
957 vref(dvp);
958 newvp = dvp;
959 #ifndef NFS_V2_ONLY
960 if (v3) {
961 nfsm_postop_attr(newvp, attrflag, 0);
962 nfsm_postop_attr(dvp, attrflag, 0);
963 } else
964 #endif
965 nfsm_loadattr(newvp, (struct vattr *)0, 0);
966 } else if (flags & ISDOTDOT) {
967 /*
968 * ".." lookup
969 */
970 VOP_UNLOCK(dvp);
971 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
972 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
973 if (error) {
974 m_freem(mrep);
975 return error;
976 }
977 newvp = NFSTOV(np);
978
979 #ifndef NFS_V2_ONLY
980 if (v3) {
981 nfsm_postop_attr(newvp, attrflag, 0);
982 nfsm_postop_attr(dvp, attrflag, 0);
983 } else
984 #endif
985 nfsm_loadattr(newvp, (struct vattr *)0, 0);
986 } else {
987 /*
988 * Other lookups.
989 */
990 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
991 if (error) {
992 m_freem(mrep);
993 return error;
994 }
995 newvp = NFSTOV(np);
996 #ifndef NFS_V2_ONLY
997 if (v3) {
998 nfsm_postop_attr(newvp, attrflag, 0);
999 nfsm_postop_attr(dvp, attrflag, 0);
1000 } else
1001 #endif
1002 nfsm_loadattr(newvp, (struct vattr *)0, 0);
1003 }
1004 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1005 cnp->cn_flags |= SAVENAME;
1006 if ((cnp->cn_flags & MAKEENTRY) &&
1007 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1008 nfs_cache_enter(dvp, newvp, cnp);
1009 }
1010 *vpp = newvp;
1011 nfsm_reqdone;
1012 if (error) {
1013 /*
1014 * We get here only because of errors returned by
1015 * the RPC. Otherwise we'll have returned above
1016 * (the nfsm_* macros will jump to nfsm_reqdone
1017 * on error).
1018 */
1019 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1020 cnp->cn_nameiop != CREATE) {
1021 nfs_cache_enter(dvp, NULL, cnp);
1022 }
1023 if (newvp != NULLVP) {
1024 if (newvp == dvp) {
1025 vrele(newvp);
1026 } else {
1027 vput(newvp);
1028 }
1029 }
1030 noentry:
1031 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1032 (flags & ISLASTCN) && error == ENOENT) {
1033 if (dvp->v_mount->mnt_flag & MNT_RDONLY) {
1034 error = EROFS;
1035 } else {
1036 error = EJUSTRETURN;
1037 cnp->cn_flags |= SAVENAME;
1038 }
1039 }
1040 *vpp = NULL;
1041 return error;
1042 }
1043
1044 validate:
1045 /*
1046 * make sure we have valid type and size.
1047 */
1048
1049 newvp = *vpp;
1050 if (newvp->v_type == VNON) {
1051 struct vattr vattr; /* dummy */
1052
1053 KASSERT(VTONFS(newvp)->n_attrstamp == 0);
1054 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred);
1055 if (error) {
1056 vput(newvp);
1057 *vpp = NULL;
1058 }
1059 }
1060
1061 return error;
1062 }
1063
1064 /*
1065 * nfs read call.
1066 * Just call nfs_bioread() to do the work.
1067 */
1068 int
1069 nfs_read(void *v)
1070 {
1071 struct vop_read_args /* {
1072 struct vnode *a_vp;
1073 struct uio *a_uio;
1074 int a_ioflag;
1075 kauth_cred_t a_cred;
1076 } */ *ap = v;
1077 struct vnode *vp = ap->a_vp;
1078
1079 if (vp->v_type != VREG)
1080 return EISDIR;
1081 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1082 }
1083
1084 /*
1085 * nfs readlink call
1086 */
1087 int
1088 nfs_readlink(void *v)
1089 {
1090 struct vop_readlink_args /* {
1091 struct vnode *a_vp;
1092 struct uio *a_uio;
1093 kauth_cred_t a_cred;
1094 } */ *ap = v;
1095 struct vnode *vp = ap->a_vp;
1096 struct nfsnode *np = VTONFS(vp);
1097
1098 if (vp->v_type != VLNK)
1099 return (EPERM);
1100
1101 if (np->n_rcred != NULL) {
1102 kauth_cred_free(np->n_rcred);
1103 }
1104 np->n_rcred = ap->a_cred;
1105 kauth_cred_hold(np->n_rcred);
1106
1107 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1108 }
1109
1110 /*
1111 * Do a readlink rpc.
1112 * Called by nfs_doio() from below the buffer cache.
1113 */
1114 int
1115 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred)
1116 {
1117 u_int32_t *tl;
1118 char *cp;
1119 int32_t t1, t2;
1120 char *bpos, *dpos, *cp2;
1121 int error = 0;
1122 uint32_t len;
1123 struct mbuf *mreq, *mrep, *md, *mb;
1124 const int v3 = NFS_ISV3(vp);
1125 struct nfsnode *np = VTONFS(vp);
1126 #ifndef NFS_V2_ONLY
1127 int attrflag;
1128 #endif
1129
1130 nfsstats.rpccnt[NFSPROC_READLINK]++;
1131 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1132 nfsm_fhtom(np, v3);
1133 nfsm_request(np, NFSPROC_READLINK, curlwp, cred);
1134 #ifndef NFS_V2_ONLY
1135 if (v3)
1136 nfsm_postop_attr(vp, attrflag, 0);
1137 #endif
1138 if (!error) {
1139 #ifndef NFS_V2_ONLY
1140 if (v3) {
1141 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1142 len = fxdr_unsigned(uint32_t, *tl);
1143 if (len > MAXPATHLEN) {
1144 /*
1145 * this pathname is too long for us.
1146 */
1147 m_freem(mrep);
1148 /* Solaris returns EINVAL. should we follow? */
1149 error = ENAMETOOLONG;
1150 goto nfsmout;
1151 }
1152 } else
1153 #endif
1154 {
1155 nfsm_strsiz(len, NFS_MAXPATHLEN);
1156 }
1157 nfsm_mtouio(uiop, len);
1158 }
1159 nfsm_reqdone;
1160 return (error);
1161 }
1162
1163 /*
1164 * nfs read rpc call
1165 * Ditto above
1166 */
1167 int
1168 nfs_readrpc(struct vnode *vp, struct uio *uiop)
1169 {
1170 u_int32_t *tl;
1171 char *cp;
1172 int32_t t1, t2;
1173 char *bpos, *dpos, *cp2;
1174 struct mbuf *mreq, *mrep, *md, *mb;
1175 struct nfsmount *nmp;
1176 int error = 0, len, retlen, tsiz, eof, byte_count;
1177 const int v3 = NFS_ISV3(vp);
1178 struct nfsnode *np = VTONFS(vp);
1179 #ifndef NFS_V2_ONLY
1180 int attrflag;
1181 #endif
1182
1183 #ifndef nolint
1184 eof = 0;
1185 #endif
1186 nmp = VFSTONFS(vp->v_mount);
1187 tsiz = uiop->uio_resid;
1188 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1189 return (EFBIG);
1190 iostat_busy(nmp->nm_stats);
1191 byte_count = 0; /* count bytes actually transferred */
1192 while (tsiz > 0) {
1193 nfsstats.rpccnt[NFSPROC_READ]++;
1194 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1195 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1196 nfsm_fhtom(np, v3);
1197 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1198 #ifndef NFS_V2_ONLY
1199 if (v3) {
1200 txdr_hyper(uiop->uio_offset, tl);
1201 *(tl + 2) = txdr_unsigned(len);
1202 } else
1203 #endif
1204 {
1205 *tl++ = txdr_unsigned(uiop->uio_offset);
1206 *tl++ = txdr_unsigned(len);
1207 *tl = 0;
1208 }
1209 nfsm_request(np, NFSPROC_READ, curlwp, np->n_rcred);
1210 #ifndef NFS_V2_ONLY
1211 if (v3) {
1212 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1213 if (error) {
1214 m_freem(mrep);
1215 goto nfsmout;
1216 }
1217 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1218 eof = fxdr_unsigned(int, *(tl + 1));
1219 } else
1220 #endif
1221 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1222 nfsm_strsiz(retlen, nmp->nm_rsize);
1223 nfsm_mtouio(uiop, retlen);
1224 m_freem(mrep);
1225 tsiz -= retlen;
1226 byte_count += retlen;
1227 #ifndef NFS_V2_ONLY
1228 if (v3) {
1229 if (eof || retlen == 0)
1230 tsiz = 0;
1231 } else
1232 #endif
1233 if (retlen < len)
1234 tsiz = 0;
1235 }
1236 nfsmout:
1237 iostat_unbusy(nmp->nm_stats, byte_count, 1);
1238 return (error);
1239 }
1240
1241 struct nfs_writerpc_context {
1242 kmutex_t nwc_lock;
1243 kcondvar_t nwc_cv;
1244 int nwc_mbufcount;
1245 };
1246
1247 /*
1248 * free mbuf used to refer protected pages while write rpc call.
1249 * called at splvm.
1250 */
1251 static void
1252 nfs_writerpc_extfree(struct mbuf *m, void *tbuf, size_t size, void *arg)
1253 {
1254 struct nfs_writerpc_context *ctx = arg;
1255
1256 KASSERT(m != NULL);
1257 KASSERT(ctx != NULL);
1258 pool_cache_put(mb_cache, m);
1259 mutex_enter(&ctx->nwc_lock);
1260 if (--ctx->nwc_mbufcount == 0) {
1261 cv_signal(&ctx->nwc_cv);
1262 }
1263 mutex_exit(&ctx->nwc_lock);
1264 }
1265
1266 /*
1267 * nfs write call
1268 */
1269 int
1270 nfs_writerpc(struct vnode *vp, struct uio *uiop, int *iomode, bool pageprotected, bool *stalewriteverfp)
1271 {
1272 u_int32_t *tl;
1273 char *cp;
1274 int32_t t1, t2;
1275 char *bpos, *dpos;
1276 struct mbuf *mreq, *mrep, *md, *mb;
1277 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1278 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR;
1279 const int v3 = NFS_ISV3(vp);
1280 int committed = NFSV3WRITE_FILESYNC;
1281 struct nfsnode *np = VTONFS(vp);
1282 struct nfs_writerpc_context ctx;
1283 int byte_count;
1284 size_t origresid;
1285 #ifndef NFS_V2_ONLY
1286 char *cp2;
1287 int rlen, commit;
1288 #endif
1289
1290 mutex_init(&ctx.nwc_lock, MUTEX_DRIVER, IPL_VM);
1291 cv_init(&ctx.nwc_cv, "nfsmblk");
1292 ctx.nwc_mbufcount = 1;
1293
1294 if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1295 panic("writerpc readonly vp %p", vp);
1296 }
1297
1298 #ifdef DIAGNOSTIC
1299 if (uiop->uio_iovcnt != 1)
1300 panic("nfs: writerpc iovcnt > 1");
1301 #endif
1302 tsiz = uiop->uio_resid;
1303 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1304 return (EFBIG);
1305 retry:
1306 origresid = uiop->uio_resid;
1307 KASSERT(origresid == uiop->uio_iov->iov_len);
1308 iostat_busy(nmp->nm_stats);
1309 byte_count = 0; /* count of bytes actually written */
1310 while (tsiz > 0) {
1311 uint32_t datalen; /* data bytes need to be allocated in mbuf */
1312 uint32_t backup;
1313 bool stalewriteverf = false;
1314
1315 nfsstats.rpccnt[NFSPROC_WRITE]++;
1316 len = min(tsiz, nmp->nm_wsize);
1317 datalen = pageprotected ? 0 : nfsm_rndup(len);
1318 nfsm_reqhead(np, NFSPROC_WRITE,
1319 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1320 nfsm_fhtom(np, v3);
1321 #ifndef NFS_V2_ONLY
1322 if (v3) {
1323 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1324 txdr_hyper(uiop->uio_offset, tl);
1325 tl += 2;
1326 *tl++ = txdr_unsigned(len);
1327 *tl++ = txdr_unsigned(*iomode);
1328 *tl = txdr_unsigned(len);
1329 } else
1330 #endif
1331 {
1332 u_int32_t x;
1333
1334 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1335 /* Set both "begin" and "current" to non-garbage. */
1336 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1337 *tl++ = x; /* "begin offset" */
1338 *tl++ = x; /* "current offset" */
1339 x = txdr_unsigned(len);
1340 *tl++ = x; /* total to this offset */
1341 *tl = x; /* size of this write */
1342
1343 }
1344 if (pageprotected) {
1345 /*
1346 * since we know pages can't be modified during i/o,
1347 * no need to copy them for us.
1348 */
1349 struct mbuf *m;
1350 struct iovec *iovp = uiop->uio_iov;
1351
1352 m = m_get(M_WAIT, MT_DATA);
1353 MCLAIM(m, &nfs_mowner);
1354 MEXTADD(m, iovp->iov_base, len, M_MBUF,
1355 nfs_writerpc_extfree, &ctx);
1356 m->m_flags |= M_EXT_ROMAP;
1357 m->m_len = len;
1358 mb->m_next = m;
1359 /*
1360 * no need to maintain mb and bpos here
1361 * because no one care them later.
1362 */
1363 #if 0
1364 mb = m;
1365 bpos = mtod(void *, mb) + mb->m_len;
1366 #endif
1367 UIO_ADVANCE(uiop, len);
1368 uiop->uio_offset += len;
1369 mutex_enter(&ctx.nwc_lock);
1370 ctx.nwc_mbufcount++;
1371 mutex_exit(&ctx.nwc_lock);
1372 nfs_zeropad(mb, 0, nfsm_padlen(len));
1373 } else {
1374 nfsm_uiotom(uiop, len);
1375 }
1376 nfsm_request(np, NFSPROC_WRITE, curlwp, np->n_wcred);
1377 #ifndef NFS_V2_ONLY
1378 if (v3) {
1379 wccflag = NFSV3_WCCCHK;
1380 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, !error);
1381 if (!error) {
1382 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1383 + NFSX_V3WRITEVERF);
1384 rlen = fxdr_unsigned(int, *tl++);
1385 if (rlen == 0) {
1386 error = NFSERR_IO;
1387 m_freem(mrep);
1388 break;
1389 } else if (rlen < len) {
1390 backup = len - rlen;
1391 UIO_ADVANCE(uiop, -backup);
1392 uiop->uio_offset -= backup;
1393 len = rlen;
1394 }
1395 commit = fxdr_unsigned(int, *tl++);
1396
1397 /*
1398 * Return the lowest committment level
1399 * obtained by any of the RPCs.
1400 */
1401 if (committed == NFSV3WRITE_FILESYNC)
1402 committed = commit;
1403 else if (committed == NFSV3WRITE_DATASYNC &&
1404 commit == NFSV3WRITE_UNSTABLE)
1405 committed = commit;
1406 mutex_enter(&nmp->nm_lock);
1407 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1408 memcpy(nmp->nm_writeverf, tl,
1409 NFSX_V3WRITEVERF);
1410 nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1411 } else if ((nmp->nm_iflag &
1412 NFSMNT_STALEWRITEVERF) ||
1413 memcmp(tl, nmp->nm_writeverf,
1414 NFSX_V3WRITEVERF)) {
1415 memcpy(nmp->nm_writeverf, tl,
1416 NFSX_V3WRITEVERF);
1417 /*
1418 * note NFSMNT_STALEWRITEVERF
1419 * if we're the first thread to
1420 * notice it.
1421 */
1422 if ((nmp->nm_iflag &
1423 NFSMNT_STALEWRITEVERF) == 0) {
1424 stalewriteverf = true;
1425 nmp->nm_iflag |=
1426 NFSMNT_STALEWRITEVERF;
1427 }
1428 }
1429 mutex_exit(&nmp->nm_lock);
1430 }
1431 } else
1432 #endif
1433 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1434 if (wccflag)
1435 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1436 m_freem(mrep);
1437 if (error)
1438 break;
1439 tsiz -= len;
1440 byte_count += len;
1441 if (stalewriteverf) {
1442 *stalewriteverfp = true;
1443 stalewriteverf = false;
1444 if (committed == NFSV3WRITE_UNSTABLE &&
1445 len != origresid) {
1446 /*
1447 * if our write requests weren't atomic but
1448 * unstable, datas in previous iterations
1449 * might have already been lost now.
1450 * then, we should resend them to nfsd.
1451 */
1452 backup = origresid - tsiz;
1453 UIO_ADVANCE(uiop, -backup);
1454 uiop->uio_offset -= backup;
1455 tsiz = origresid;
1456 goto retry;
1457 }
1458 }
1459 }
1460 nfsmout:
1461 iostat_unbusy(nmp->nm_stats, byte_count, 0);
1462 if (pageprotected) {
1463 /*
1464 * wait until mbufs go away.
1465 * retransmitted mbufs can survive longer than rpc requests
1466 * themselves.
1467 */
1468 mutex_enter(&ctx.nwc_lock);
1469 ctx.nwc_mbufcount--;
1470 while (ctx.nwc_mbufcount > 0) {
1471 cv_wait(&ctx.nwc_cv, &ctx.nwc_lock);
1472 }
1473 mutex_exit(&ctx.nwc_lock);
1474 }
1475 mutex_destroy(&ctx.nwc_lock);
1476 cv_destroy(&ctx.nwc_cv);
1477 *iomode = committed;
1478 if (error)
1479 uiop->uio_resid = tsiz;
1480 return (error);
1481 }
1482
1483 /*
1484 * nfs mknod rpc
1485 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1486 * mode set to specify the file type and the size field for rdev.
1487 */
1488 int
1489 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap)
1490 {
1491 struct nfsv2_sattr *sp;
1492 u_int32_t *tl;
1493 char *cp;
1494 int32_t t1, t2;
1495 struct vnode *newvp = (struct vnode *)0;
1496 struct nfsnode *dnp, *np;
1497 char *cp2;
1498 char *bpos, *dpos;
1499 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1500 struct mbuf *mreq, *mrep, *md, *mb;
1501 u_int32_t rdev;
1502 const int v3 = NFS_ISV3(dvp);
1503
1504 if (vap->va_type == VCHR || vap->va_type == VBLK)
1505 rdev = txdr_unsigned(vap->va_rdev);
1506 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1507 rdev = nfs_xdrneg1;
1508 else {
1509 VOP_ABORTOP(dvp, cnp);
1510 vput(dvp);
1511 return (EOPNOTSUPP);
1512 }
1513 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1514 dnp = VTONFS(dvp);
1515 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1516 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1517 nfsm_fhtom(dnp, v3);
1518 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1519 #ifndef NFS_V2_ONLY
1520 if (v3) {
1521 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1522 *tl++ = vtonfsv3_type(vap->va_type);
1523 nfsm_v3attrbuild(vap, false);
1524 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1525 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1526 *tl++ = txdr_unsigned(major(vap->va_rdev));
1527 *tl = txdr_unsigned(minor(vap->va_rdev));
1528 }
1529 } else
1530 #endif
1531 {
1532 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1533 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1534 sp->sa_uid = nfs_xdrneg1;
1535 sp->sa_gid = nfs_xdrneg1;
1536 sp->sa_size = rdev;
1537 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1538 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1539 }
1540 nfsm_request(dnp, NFSPROC_MKNOD, curlwp, cnp->cn_cred);
1541 if (!error) {
1542 nfsm_mtofh(dvp, newvp, v3, gotvp);
1543 if (!gotvp) {
1544 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1545 cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1546 if (!error)
1547 newvp = NFSTOV(np);
1548 }
1549 }
1550 #ifndef NFS_V2_ONLY
1551 if (v3)
1552 nfsm_wcc_data(dvp, wccflag, 0, !error);
1553 #endif
1554 nfsm_reqdone;
1555 if (error) {
1556 if (newvp)
1557 vput(newvp);
1558 } else {
1559 if (cnp->cn_flags & MAKEENTRY)
1560 nfs_cache_enter(dvp, newvp, cnp);
1561 *vpp = newvp;
1562 }
1563 PNBUF_PUT(cnp->cn_pnbuf);
1564 VTONFS(dvp)->n_flag |= NMODIFIED;
1565 if (!wccflag)
1566 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1567 vput(dvp);
1568 return (error);
1569 }
1570
1571 /*
1572 * nfs mknod vop
1573 * just call nfs_mknodrpc() to do the work.
1574 */
1575 /* ARGSUSED */
1576 int
1577 nfs_mknod(void *v)
1578 {
1579 struct vop_mknod_args /* {
1580 struct vnode *a_dvp;
1581 struct vnode **a_vpp;
1582 struct componentname *a_cnp;
1583 struct vattr *a_vap;
1584 } */ *ap = v;
1585 struct vnode *dvp = ap->a_dvp;
1586 struct componentname *cnp = ap->a_cnp;
1587 int error;
1588
1589 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, ap->a_vap);
1590 VN_KNOTE(dvp, NOTE_WRITE);
1591 if (error == 0 || error == EEXIST)
1592 cache_purge1(dvp, cnp, 0);
1593 return (error);
1594 }
1595
1596 /*
1597 * nfs file create call
1598 */
1599 int
1600 nfs_create(void *v)
1601 {
1602 struct vop_create_args /* {
1603 struct vnode *a_dvp;
1604 struct vnode **a_vpp;
1605 struct componentname *a_cnp;
1606 struct vattr *a_vap;
1607 } */ *ap = v;
1608 struct vnode *dvp = ap->a_dvp;
1609 struct vattr *vap = ap->a_vap;
1610 struct componentname *cnp = ap->a_cnp;
1611 struct nfsv2_sattr *sp;
1612 u_int32_t *tl;
1613 char *cp;
1614 int32_t t1, t2;
1615 struct nfsnode *dnp, *np = (struct nfsnode *)0;
1616 struct vnode *newvp = (struct vnode *)0;
1617 char *bpos, *dpos, *cp2;
1618 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1619 struct mbuf *mreq, *mrep, *md, *mb;
1620 const int v3 = NFS_ISV3(dvp);
1621 u_int32_t excl_mode = NFSV3CREATE_UNCHECKED;
1622
1623 /*
1624 * Oops, not for me..
1625 */
1626 if (vap->va_type == VSOCK)
1627 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1628
1629 KASSERT(vap->va_type == VREG);
1630
1631 #ifdef VA_EXCLUSIVE
1632 if (vap->va_vaflags & VA_EXCLUSIVE) {
1633 excl_mode = NFSV3CREATE_EXCLUSIVE;
1634 }
1635 #endif
1636 again:
1637 error = 0;
1638 nfsstats.rpccnt[NFSPROC_CREATE]++;
1639 dnp = VTONFS(dvp);
1640 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1641 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1642 nfsm_fhtom(dnp, v3);
1643 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1644 #ifndef NFS_V2_ONLY
1645 if (v3) {
1646 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1647 if (excl_mode == NFSV3CREATE_EXCLUSIVE) {
1648 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1649 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1650 *tl++ = arc4random();
1651 *tl = arc4random();
1652 } else {
1653 *tl = txdr_unsigned(excl_mode);
1654 nfsm_v3attrbuild(vap, false);
1655 }
1656 } else
1657 #endif
1658 {
1659 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1660 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1661 sp->sa_uid = nfs_xdrneg1;
1662 sp->sa_gid = nfs_xdrneg1;
1663 sp->sa_size = 0;
1664 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1665 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1666 }
1667 nfsm_request(dnp, NFSPROC_CREATE, curlwp, cnp->cn_cred);
1668 if (!error) {
1669 nfsm_mtofh(dvp, newvp, v3, gotvp);
1670 if (!gotvp) {
1671 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1672 cnp->cn_namelen, cnp->cn_cred, curlwp, &np);
1673 if (!error)
1674 newvp = NFSTOV(np);
1675 }
1676 }
1677 #ifndef NFS_V2_ONLY
1678 if (v3)
1679 nfsm_wcc_data(dvp, wccflag, 0, !error);
1680 #endif
1681 nfsm_reqdone;
1682 if (error) {
1683 /*
1684 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
1685 */
1686 if (v3 && error == ENOTSUP) {
1687 if (excl_mode == NFSV3CREATE_EXCLUSIVE) {
1688 excl_mode = NFSV3CREATE_GUARDED;
1689 goto again;
1690 } else if (excl_mode == NFSV3CREATE_GUARDED) {
1691 excl_mode = NFSV3CREATE_UNCHECKED;
1692 goto again;
1693 }
1694 }
1695 } else if (v3 && (excl_mode == NFSV3CREATE_EXCLUSIVE)) {
1696 struct timespec ts;
1697
1698 getnanotime(&ts);
1699
1700 /*
1701 * make sure that we'll update timestamps as
1702 * most server implementations use them to store
1703 * the create verifier.
1704 *
1705 * XXX it's better to use TOSERVER always.
1706 */
1707
1708 if (vap->va_atime.tv_sec == VNOVAL)
1709 vap->va_atime = ts;
1710 if (vap->va_mtime.tv_sec == VNOVAL)
1711 vap->va_mtime = ts;
1712
1713 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, curlwp);
1714 }
1715 if (error == 0) {
1716 if (cnp->cn_flags & MAKEENTRY)
1717 nfs_cache_enter(dvp, newvp, cnp);
1718 else
1719 cache_purge1(dvp, cnp, 0);
1720 *ap->a_vpp = newvp;
1721 } else {
1722 if (newvp)
1723 vput(newvp);
1724 if (error == EEXIST)
1725 cache_purge1(dvp, cnp, 0);
1726 }
1727 PNBUF_PUT(cnp->cn_pnbuf);
1728 VTONFS(dvp)->n_flag |= NMODIFIED;
1729 if (!wccflag)
1730 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1731 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1732 vput(dvp);
1733 return (error);
1734 }
1735
1736 /*
1737 * nfs file remove call
1738 * To try and make nfs semantics closer to ufs semantics, a file that has
1739 * other processes using the vnode is renamed instead of removed and then
1740 * removed later on the last close.
1741 * - If v_usecount > 1
1742 * If a rename is not already in the works
1743 * call nfs_sillyrename() to set it up
1744 * else
1745 * do the remove rpc
1746 */
1747 int
1748 nfs_remove(void *v)
1749 {
1750 struct vop_remove_args /* {
1751 struct vnodeop_desc *a_desc;
1752 struct vnode * a_dvp;
1753 struct vnode * a_vp;
1754 struct componentname * a_cnp;
1755 } */ *ap = v;
1756 struct vnode *vp = ap->a_vp;
1757 struct vnode *dvp = ap->a_dvp;
1758 struct componentname *cnp = ap->a_cnp;
1759 struct nfsnode *np = VTONFS(vp);
1760 int error = 0;
1761 struct vattr vattr;
1762
1763 #ifndef DIAGNOSTIC
1764 if ((cnp->cn_flags & HASBUF) == 0)
1765 panic("nfs_remove: no name");
1766 if (vp->v_usecount < 1)
1767 panic("nfs_remove: bad v_usecount");
1768 #endif
1769 if (vp->v_type == VDIR)
1770 error = EPERM;
1771 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1772 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1773 vattr.va_nlink > 1)) {
1774 /*
1775 * Purge the name cache so that the chance of a lookup for
1776 * the name succeeding while the remove is in progress is
1777 * minimized. Without node locking it can still happen, such
1778 * that an I/O op returns ESTALE, but since you get this if
1779 * another host removes the file..
1780 */
1781 cache_purge(vp);
1782 /*
1783 * throw away biocache buffers, mainly to avoid
1784 * unnecessary delayed writes later.
1785 */
1786 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, curlwp, 1);
1787 /* Do the rpc */
1788 if (error != EINTR)
1789 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1790 cnp->cn_namelen, cnp->cn_cred, curlwp);
1791 } else if (!np->n_sillyrename)
1792 error = nfs_sillyrename(dvp, vp, cnp, false);
1793 PNBUF_PUT(cnp->cn_pnbuf);
1794 if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1795 vattr.va_nlink == 1) {
1796 np->n_flag |= NREMOVED;
1797 }
1798 NFS_INVALIDATE_ATTRCACHE(np);
1799 VN_KNOTE(vp, NOTE_DELETE);
1800 VN_KNOTE(dvp, NOTE_WRITE);
1801 if (dvp == vp)
1802 vrele(vp);
1803 else
1804 vput(vp);
1805 vput(dvp);
1806 return (error);
1807 }
1808
1809 /*
1810 * nfs file remove rpc called from nfs_inactive
1811 */
1812 int
1813 nfs_removeit(struct sillyrename *sp)
1814 {
1815
1816 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1817 (struct lwp *)0));
1818 }
1819
1820 /*
1821 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1822 */
1823 int
1824 nfs_removerpc(struct vnode *dvp, const char *name, int namelen, kauth_cred_t cred, struct lwp *l)
1825 {
1826 u_int32_t *tl;
1827 char *cp;
1828 #ifndef NFS_V2_ONLY
1829 int32_t t1;
1830 char *cp2;
1831 #endif
1832 int32_t t2;
1833 char *bpos, *dpos;
1834 int error = 0, wccflag = NFSV3_WCCRATTR;
1835 struct mbuf *mreq, *mrep, *md, *mb;
1836 const int v3 = NFS_ISV3(dvp);
1837 int rexmit = 0;
1838 struct nfsnode *dnp = VTONFS(dvp);
1839
1840 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1841 nfsm_reqhead(dnp, NFSPROC_REMOVE,
1842 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1843 nfsm_fhtom(dnp, v3);
1844 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1845 nfsm_request1(dnp, NFSPROC_REMOVE, l, cred, &rexmit);
1846 #ifndef NFS_V2_ONLY
1847 if (v3)
1848 nfsm_wcc_data(dvp, wccflag, 0, !error);
1849 #endif
1850 nfsm_reqdone;
1851 VTONFS(dvp)->n_flag |= NMODIFIED;
1852 if (!wccflag)
1853 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1854 /*
1855 * Kludge City: If the first reply to the remove rpc is lost..
1856 * the reply to the retransmitted request will be ENOENT
1857 * since the file was in fact removed
1858 * Therefore, we cheat and return success.
1859 */
1860 if (rexmit && error == ENOENT)
1861 error = 0;
1862 return (error);
1863 }
1864
1865 /*
1866 * nfs file rename call
1867 */
1868 int
1869 nfs_rename(void *v)
1870 {
1871 struct vop_rename_args /* {
1872 struct vnode *a_fdvp;
1873 struct vnode *a_fvp;
1874 struct componentname *a_fcnp;
1875 struct vnode *a_tdvp;
1876 struct vnode *a_tvp;
1877 struct componentname *a_tcnp;
1878 } */ *ap = v;
1879 struct vnode *fvp = ap->a_fvp;
1880 struct vnode *tvp = ap->a_tvp;
1881 struct vnode *fdvp = ap->a_fdvp;
1882 struct vnode *tdvp = ap->a_tdvp;
1883 struct componentname *tcnp = ap->a_tcnp;
1884 struct componentname *fcnp = ap->a_fcnp;
1885 int error;
1886
1887 #ifndef DIAGNOSTIC
1888 if ((tcnp->cn_flags & HASBUF) == 0 ||
1889 (fcnp->cn_flags & HASBUF) == 0)
1890 panic("nfs_rename: no name");
1891 #endif
1892 /* Check for cross-device rename */
1893 if ((fvp->v_mount != tdvp->v_mount) ||
1894 (tvp && (fvp->v_mount != tvp->v_mount))) {
1895 error = EXDEV;
1896 goto out;
1897 }
1898
1899 /*
1900 * If the tvp exists and is in use, sillyrename it before doing the
1901 * rename of the new file over it.
1902 *
1903 * Have sillyrename use link instead of rename if possible,
1904 * so that we don't lose the file if the rename fails, and so
1905 * that there's no window when the "to" file doesn't exist.
1906 */
1907 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1908 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp, true)) {
1909 VN_KNOTE(tvp, NOTE_DELETE);
1910 vput(tvp);
1911 tvp = NULL;
1912 }
1913
1914 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1915 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1916 curlwp);
1917
1918 VN_KNOTE(fdvp, NOTE_WRITE);
1919 VN_KNOTE(tdvp, NOTE_WRITE);
1920 if (error == 0 || error == EEXIST) {
1921 if (fvp->v_type == VDIR)
1922 cache_purge(fvp);
1923 else
1924 cache_purge1(fdvp, fcnp, 0);
1925 if (tvp != NULL && tvp->v_type == VDIR)
1926 cache_purge(tvp);
1927 else
1928 cache_purge1(tdvp, tcnp, 0);
1929 }
1930 out:
1931 if (tdvp == tvp)
1932 vrele(tdvp);
1933 else
1934 vput(tdvp);
1935 if (tvp)
1936 vput(tvp);
1937 vrele(fdvp);
1938 vrele(fvp);
1939 return (error);
1940 }
1941
1942 /*
1943 * nfs file rename rpc called from nfs_remove() above
1944 */
1945 int
1946 nfs_renameit(struct vnode *sdvp, struct componentname *scnp, struct sillyrename *sp)
1947 {
1948 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1949 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, curlwp));
1950 }
1951
1952 /*
1953 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1954 */
1955 int
1956 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen, struct vnode *tdvp, const char *tnameptr, int tnamelen, kauth_cred_t cred, struct lwp *l)
1957 {
1958 u_int32_t *tl;
1959 char *cp;
1960 #ifndef NFS_V2_ONLY
1961 int32_t t1;
1962 char *cp2;
1963 #endif
1964 int32_t t2;
1965 char *bpos, *dpos;
1966 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1967 struct mbuf *mreq, *mrep, *md, *mb;
1968 const int v3 = NFS_ISV3(fdvp);
1969 int rexmit = 0;
1970 struct nfsnode *fdnp = VTONFS(fdvp);
1971
1972 nfsstats.rpccnt[NFSPROC_RENAME]++;
1973 nfsm_reqhead(fdnp, NFSPROC_RENAME,
1974 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1975 nfsm_rndup(tnamelen));
1976 nfsm_fhtom(fdnp, v3);
1977 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1978 nfsm_fhtom(VTONFS(tdvp), v3);
1979 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1980 nfsm_request1(fdnp, NFSPROC_RENAME, l, cred, &rexmit);
1981 #ifndef NFS_V2_ONLY
1982 if (v3) {
1983 nfsm_wcc_data(fdvp, fwccflag, 0, !error);
1984 nfsm_wcc_data(tdvp, twccflag, 0, !error);
1985 }
1986 #endif
1987 nfsm_reqdone;
1988 VTONFS(fdvp)->n_flag |= NMODIFIED;
1989 VTONFS(tdvp)->n_flag |= NMODIFIED;
1990 if (!fwccflag)
1991 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
1992 if (!twccflag)
1993 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
1994 /*
1995 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1996 */
1997 if (rexmit && error == ENOENT)
1998 error = 0;
1999 return (error);
2000 }
2001
2002 /*
2003 * NFS link RPC, called from nfs_link.
2004 * Assumes dvp and vp locked, and leaves them that way.
2005 */
2006
2007 static int
2008 nfs_linkrpc(struct vnode *dvp, struct vnode *vp, const char *name,
2009 size_t namelen, kauth_cred_t cred, struct lwp *l)
2010 {
2011 u_int32_t *tl;
2012 char *cp;
2013 #ifndef NFS_V2_ONLY
2014 int32_t t1;
2015 char *cp2;
2016 #endif
2017 int32_t t2;
2018 char *bpos, *dpos;
2019 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2020 struct mbuf *mreq, *mrep, *md, *mb;
2021 const int v3 = NFS_ISV3(dvp);
2022 int rexmit = 0;
2023 struct nfsnode *np = VTONFS(vp);
2024
2025 nfsstats.rpccnt[NFSPROC_LINK]++;
2026 nfsm_reqhead(np, NFSPROC_LINK,
2027 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(namelen));
2028 nfsm_fhtom(np, v3);
2029 nfsm_fhtom(VTONFS(dvp), v3);
2030 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
2031 nfsm_request1(np, NFSPROC_LINK, l, cred, &rexmit);
2032 #ifndef NFS_V2_ONLY
2033 if (v3) {
2034 nfsm_postop_attr(vp, attrflag, 0);
2035 nfsm_wcc_data(dvp, wccflag, 0, !error);
2036 }
2037 #endif
2038 nfsm_reqdone;
2039
2040 VTONFS(dvp)->n_flag |= NMODIFIED;
2041 if (!attrflag)
2042 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
2043 if (!wccflag)
2044 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2045
2046 /*
2047 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2048 */
2049 if (rexmit && error == EEXIST)
2050 error = 0;
2051
2052 return error;
2053 }
2054
2055 /*
2056 * nfs hard link create call
2057 */
2058 int
2059 nfs_link(void *v)
2060 {
2061 struct vop_link_args /* {
2062 struct vnode *a_dvp;
2063 struct vnode *a_vp;
2064 struct componentname *a_cnp;
2065 } */ *ap = v;
2066 struct vnode *vp = ap->a_vp;
2067 struct vnode *dvp = ap->a_dvp;
2068 struct componentname *cnp = ap->a_cnp;
2069 int error = 0;
2070
2071 if (dvp->v_mount != vp->v_mount) {
2072 VOP_ABORTOP(dvp, cnp);
2073 vput(dvp);
2074 return (EXDEV);
2075 }
2076 if (dvp != vp) {
2077 error = vn_lock(vp, LK_EXCLUSIVE);
2078 if (error != 0) {
2079 VOP_ABORTOP(dvp, cnp);
2080 vput(dvp);
2081 return error;
2082 }
2083 }
2084
2085 /*
2086 * Push all writes to the server, so that the attribute cache
2087 * doesn't get "out of sync" with the server.
2088 * XXX There should be a better way!
2089 */
2090 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0);
2091
2092 error = nfs_linkrpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2093 cnp->cn_cred, curlwp);
2094
2095 if (error == 0)
2096 cache_purge1(dvp, cnp, 0);
2097 PNBUF_PUT(cnp->cn_pnbuf);
2098 if (dvp != vp)
2099 VOP_UNLOCK(vp);
2100 VN_KNOTE(vp, NOTE_LINK);
2101 VN_KNOTE(dvp, NOTE_WRITE);
2102 vput(dvp);
2103 return (error);
2104 }
2105
2106 /*
2107 * nfs symbolic link create call
2108 */
2109 int
2110 nfs_symlink(void *v)
2111 {
2112 struct vop_symlink_args /* {
2113 struct vnode *a_dvp;
2114 struct vnode **a_vpp;
2115 struct componentname *a_cnp;
2116 struct vattr *a_vap;
2117 char *a_target;
2118 } */ *ap = v;
2119 struct vnode *dvp = ap->a_dvp;
2120 struct vattr *vap = ap->a_vap;
2121 struct componentname *cnp = ap->a_cnp;
2122 struct nfsv2_sattr *sp;
2123 u_int32_t *tl;
2124 char *cp;
2125 int32_t t1, t2;
2126 char *bpos, *dpos, *cp2;
2127 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2128 struct mbuf *mreq, *mrep, *md, *mb;
2129 struct vnode *newvp = (struct vnode *)0;
2130 const int v3 = NFS_ISV3(dvp);
2131 int rexmit = 0;
2132 struct nfsnode *dnp = VTONFS(dvp);
2133
2134 *ap->a_vpp = NULL;
2135 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2136 slen = strlen(ap->a_target);
2137 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2138 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2139 nfsm_fhtom(dnp, v3);
2140 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2141 #ifndef NFS_V2_ONlY
2142 if (v3)
2143 nfsm_v3attrbuild(vap, false);
2144 #endif
2145 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2146 #ifndef NFS_V2_ONlY
2147 if (!v3) {
2148 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2149 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2150 sp->sa_uid = nfs_xdrneg1;
2151 sp->sa_gid = nfs_xdrneg1;
2152 sp->sa_size = nfs_xdrneg1;
2153 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2154 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2155 }
2156 #endif
2157 nfsm_request1(dnp, NFSPROC_SYMLINK, curlwp, cnp->cn_cred,
2158 &rexmit);
2159 #ifndef NFS_V2_ONlY
2160 if (v3) {
2161 if (!error)
2162 nfsm_mtofh(dvp, newvp, v3, gotvp);
2163 nfsm_wcc_data(dvp, wccflag, 0, !error);
2164 }
2165 #endif
2166 nfsm_reqdone;
2167 /*
2168 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2169 */
2170 if (rexmit && error == EEXIST)
2171 error = 0;
2172 if (error == 0 || error == EEXIST)
2173 cache_purge1(dvp, cnp, 0);
2174 if (error == 0 && newvp == NULL) {
2175 struct nfsnode *np = NULL;
2176
2177 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2178 cnp->cn_cred, curlwp, &np);
2179 if (error == 0)
2180 newvp = NFSTOV(np);
2181 }
2182 if (error) {
2183 if (newvp != NULL)
2184 vput(newvp);
2185 } else {
2186 *ap->a_vpp = newvp;
2187 }
2188 PNBUF_PUT(cnp->cn_pnbuf);
2189 VTONFS(dvp)->n_flag |= NMODIFIED;
2190 if (!wccflag)
2191 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2192 VN_KNOTE(dvp, NOTE_WRITE);
2193 vput(dvp);
2194 return (error);
2195 }
2196
2197 /*
2198 * nfs make dir call
2199 */
2200 int
2201 nfs_mkdir(void *v)
2202 {
2203 struct vop_mkdir_args /* {
2204 struct vnode *a_dvp;
2205 struct vnode **a_vpp;
2206 struct componentname *a_cnp;
2207 struct vattr *a_vap;
2208 } */ *ap = v;
2209 struct vnode *dvp = ap->a_dvp;
2210 struct vattr *vap = ap->a_vap;
2211 struct componentname *cnp = ap->a_cnp;
2212 struct nfsv2_sattr *sp;
2213 u_int32_t *tl;
2214 char *cp;
2215 int32_t t1, t2;
2216 int len;
2217 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2218 struct vnode *newvp = (struct vnode *)0;
2219 char *bpos, *dpos, *cp2;
2220 int error = 0, wccflag = NFSV3_WCCRATTR;
2221 int gotvp = 0;
2222 int rexmit = 0;
2223 struct mbuf *mreq, *mrep, *md, *mb;
2224 const int v3 = NFS_ISV3(dvp);
2225
2226 len = cnp->cn_namelen;
2227 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2228 nfsm_reqhead(dnp, NFSPROC_MKDIR,
2229 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2230 nfsm_fhtom(dnp, v3);
2231 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2232 #ifndef NFS_V2_ONLY
2233 if (v3) {
2234 nfsm_v3attrbuild(vap, false);
2235 } else
2236 #endif
2237 {
2238 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2239 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2240 sp->sa_uid = nfs_xdrneg1;
2241 sp->sa_gid = nfs_xdrneg1;
2242 sp->sa_size = nfs_xdrneg1;
2243 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2244 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2245 }
2246 nfsm_request1(dnp, NFSPROC_MKDIR, curlwp, cnp->cn_cred, &rexmit);
2247 if (!error)
2248 nfsm_mtofh(dvp, newvp, v3, gotvp);
2249 if (v3)
2250 nfsm_wcc_data(dvp, wccflag, 0, !error);
2251 nfsm_reqdone;
2252 VTONFS(dvp)->n_flag |= NMODIFIED;
2253 if (!wccflag)
2254 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2255 /*
2256 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2257 * if we can succeed in looking up the directory.
2258 */
2259 if ((rexmit && error == EEXIST) || (!error && !gotvp)) {
2260 if (newvp) {
2261 vput(newvp);
2262 newvp = (struct vnode *)0;
2263 }
2264 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2265 curlwp, &np);
2266 if (!error) {
2267 newvp = NFSTOV(np);
2268 if (newvp->v_type != VDIR || newvp == dvp)
2269 error = EEXIST;
2270 }
2271 }
2272 if (error) {
2273 if (newvp) {
2274 if (dvp != newvp)
2275 vput(newvp);
2276 else
2277 vrele(newvp);
2278 }
2279 } else {
2280 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2281 if (cnp->cn_flags & MAKEENTRY)
2282 nfs_cache_enter(dvp, newvp, cnp);
2283 *ap->a_vpp = newvp;
2284 }
2285 PNBUF_PUT(cnp->cn_pnbuf);
2286 vput(dvp);
2287 return (error);
2288 }
2289
2290 /*
2291 * nfs remove directory call
2292 */
2293 int
2294 nfs_rmdir(void *v)
2295 {
2296 struct vop_rmdir_args /* {
2297 struct vnode *a_dvp;
2298 struct vnode *a_vp;
2299 struct componentname *a_cnp;
2300 } */ *ap = v;
2301 struct vnode *vp = ap->a_vp;
2302 struct vnode *dvp = ap->a_dvp;
2303 struct componentname *cnp = ap->a_cnp;
2304 u_int32_t *tl;
2305 char *cp;
2306 #ifndef NFS_V2_ONLY
2307 int32_t t1;
2308 char *cp2;
2309 #endif
2310 int32_t t2;
2311 char *bpos, *dpos;
2312 int error = 0, wccflag = NFSV3_WCCRATTR;
2313 int rexmit = 0;
2314 struct mbuf *mreq, *mrep, *md, *mb;
2315 const int v3 = NFS_ISV3(dvp);
2316 struct nfsnode *dnp;
2317
2318 if (dvp == vp) {
2319 vrele(dvp);
2320 vput(dvp);
2321 PNBUF_PUT(cnp->cn_pnbuf);
2322 return (EINVAL);
2323 }
2324 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2325 dnp = VTONFS(dvp);
2326 nfsm_reqhead(dnp, NFSPROC_RMDIR,
2327 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2328 nfsm_fhtom(dnp, v3);
2329 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2330 nfsm_request1(dnp, NFSPROC_RMDIR, curlwp, cnp->cn_cred, &rexmit);
2331 #ifndef NFS_V2_ONLY
2332 if (v3)
2333 nfsm_wcc_data(dvp, wccflag, 0, !error);
2334 #endif
2335 nfsm_reqdone;
2336 PNBUF_PUT(cnp->cn_pnbuf);
2337 VTONFS(dvp)->n_flag |= NMODIFIED;
2338 if (!wccflag)
2339 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2340 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2341 VN_KNOTE(vp, NOTE_DELETE);
2342 cache_purge(vp);
2343 vput(vp);
2344 vput(dvp);
2345 /*
2346 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2347 */
2348 if (rexmit && error == ENOENT)
2349 error = 0;
2350 return (error);
2351 }
2352
2353 /*
2354 * nfs readdir call
2355 */
2356 int
2357 nfs_readdir(void *v)
2358 {
2359 struct vop_readdir_args /* {
2360 struct vnode *a_vp;
2361 struct uio *a_uio;
2362 kauth_cred_t a_cred;
2363 int *a_eofflag;
2364 off_t **a_cookies;
2365 int *a_ncookies;
2366 } */ *ap = v;
2367 struct vnode *vp = ap->a_vp;
2368 struct uio *uio = ap->a_uio;
2369 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2370 char *base = uio->uio_iov->iov_base;
2371 int tresid, error;
2372 size_t count, lost;
2373 struct dirent *dp;
2374 off_t *cookies = NULL;
2375 int ncookies = 0, nc;
2376
2377 if (vp->v_type != VDIR)
2378 return (EPERM);
2379
2380 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2381 count = uio->uio_resid - lost;
2382 if (count <= 0)
2383 return (EINVAL);
2384
2385 /*
2386 * Call nfs_bioread() to do the real work.
2387 */
2388 tresid = uio->uio_resid = count;
2389 error = nfs_bioread(vp, uio, 0, ap->a_cred,
2390 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2391
2392 if (!error && ap->a_cookies) {
2393 ncookies = count / 16;
2394 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2395 *ap->a_cookies = cookies;
2396 }
2397
2398 if (!error && uio->uio_resid == tresid) {
2399 uio->uio_resid += lost;
2400 nfsstats.direofcache_misses++;
2401 if (ap->a_cookies)
2402 *ap->a_ncookies = 0;
2403 *ap->a_eofflag = 1;
2404 return (0);
2405 }
2406
2407 if (!error && ap->a_cookies) {
2408 /*
2409 * Only the NFS server and emulations use cookies, and they
2410 * load the directory block into system space, so we can
2411 * just look at it directly.
2412 */
2413 if (!VMSPACE_IS_KERNEL_P(uio->uio_vmspace) ||
2414 uio->uio_iovcnt != 1)
2415 panic("nfs_readdir: lost in space");
2416 for (nc = 0; ncookies-- &&
2417 base < (char *)uio->uio_iov->iov_base; nc++){
2418 dp = (struct dirent *) base;
2419 if (dp->d_reclen == 0)
2420 break;
2421 if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2422 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2423 else
2424 *(cookies++) = NFS_GETCOOKIE(dp);
2425 base += dp->d_reclen;
2426 }
2427 uio->uio_resid +=
2428 ((char *)uio->uio_iov->iov_base - base);
2429 uio->uio_iov->iov_len +=
2430 ((char *)uio->uio_iov->iov_base - base);
2431 uio->uio_iov->iov_base = base;
2432 *ap->a_ncookies = nc;
2433 }
2434
2435 uio->uio_resid += lost;
2436 *ap->a_eofflag = 0;
2437 return (error);
2438 }
2439
2440 /*
2441 * Readdir rpc call.
2442 * Called from below the buffer cache by nfs_doio().
2443 */
2444 int
2445 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred)
2446 {
2447 int len, left;
2448 struct dirent *dp = NULL;
2449 u_int32_t *tl;
2450 char *cp;
2451 int32_t t1, t2;
2452 char *bpos, *dpos, *cp2;
2453 struct mbuf *mreq, *mrep, *md, *mb;
2454 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2455 struct nfsnode *dnp = VTONFS(vp);
2456 u_quad_t fileno;
2457 int error = 0, more_dirs = 1, blksiz = 0, bigenough = 1;
2458 #ifndef NFS_V2_ONLY
2459 int attrflag;
2460 #endif
2461 int nrpcs = 0, reclen;
2462 const int v3 = NFS_ISV3(vp);
2463
2464 #ifdef DIAGNOSTIC
2465 /*
2466 * Should be called from buffer cache, so only amount of
2467 * NFS_DIRBLKSIZ will be requested.
2468 */
2469 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2470 panic("nfs readdirrpc bad uio");
2471 #endif
2472
2473 /*
2474 * Loop around doing readdir rpc's of size nm_readdirsize
2475 * truncated to a multiple of NFS_DIRFRAGSIZ.
2476 * The stopping criteria is EOF or buffer full.
2477 */
2478 while (more_dirs && bigenough) {
2479 /*
2480 * Heuristic: don't bother to do another RPC to further
2481 * fill up this block if there is not much room left. (< 50%
2482 * of the readdir RPC size). This wastes some buffer space
2483 * but can save up to 50% in RPC calls.
2484 */
2485 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2486 bigenough = 0;
2487 break;
2488 }
2489 nfsstats.rpccnt[NFSPROC_READDIR]++;
2490 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2491 NFSX_READDIR(v3));
2492 nfsm_fhtom(dnp, v3);
2493 #ifndef NFS_V2_ONLY
2494 if (v3) {
2495 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2496 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2497 txdr_swapcookie3(uiop->uio_offset, tl);
2498 } else {
2499 txdr_cookie3(uiop->uio_offset, tl);
2500 }
2501 tl += 2;
2502 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2503 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2504 } else
2505 #endif
2506 {
2507 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2508 *tl++ = txdr_unsigned(uiop->uio_offset);
2509 }
2510 *tl = txdr_unsigned(nmp->nm_readdirsize);
2511 nfsm_request(dnp, NFSPROC_READDIR, curlwp, cred);
2512 nrpcs++;
2513 #ifndef NFS_V2_ONLY
2514 if (v3) {
2515 nfsm_postop_attr(vp, attrflag, 0);
2516 if (!error) {
2517 nfsm_dissect(tl, u_int32_t *,
2518 2 * NFSX_UNSIGNED);
2519 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2520 dnp->n_cookieverf.nfsuquad[1] = *tl;
2521 } else {
2522 m_freem(mrep);
2523 goto nfsmout;
2524 }
2525 }
2526 #endif
2527 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2528 more_dirs = fxdr_unsigned(int, *tl);
2529
2530 /* loop thru the dir entries, doctoring them to 4bsd form */
2531 while (more_dirs && bigenough) {
2532 #ifndef NFS_V2_ONLY
2533 if (v3) {
2534 nfsm_dissect(tl, u_int32_t *,
2535 3 * NFSX_UNSIGNED);
2536 fileno = fxdr_hyper(tl);
2537 len = fxdr_unsigned(int, *(tl + 2));
2538 } else
2539 #endif
2540 {
2541 nfsm_dissect(tl, u_int32_t *,
2542 2 * NFSX_UNSIGNED);
2543 fileno = fxdr_unsigned(u_quad_t, *tl++);
2544 len = fxdr_unsigned(int, *tl);
2545 }
2546 if (len <= 0 || len > NFS_MAXNAMLEN) {
2547 error = EBADRPC;
2548 m_freem(mrep);
2549 goto nfsmout;
2550 }
2551 /* for cookie stashing */
2552 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2553 left = NFS_DIRFRAGSIZ - blksiz;
2554 if (reclen > left) {
2555 memset(uiop->uio_iov->iov_base, 0, left);
2556 dp->d_reclen += left;
2557 UIO_ADVANCE(uiop, left);
2558 blksiz = 0;
2559 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2560 }
2561 if (reclen > uiop->uio_resid)
2562 bigenough = 0;
2563 if (bigenough) {
2564 int tlen;
2565
2566 dp = (struct dirent *)uiop->uio_iov->iov_base;
2567 dp->d_fileno = fileno;
2568 dp->d_namlen = len;
2569 dp->d_reclen = reclen;
2570 dp->d_type = DT_UNKNOWN;
2571 blksiz += reclen;
2572 if (blksiz == NFS_DIRFRAGSIZ)
2573 blksiz = 0;
2574 UIO_ADVANCE(uiop, DIRHDSIZ);
2575 nfsm_mtouio(uiop, len);
2576 tlen = reclen - (DIRHDSIZ + len);
2577 (void)memset(uiop->uio_iov->iov_base, 0, tlen);
2578 UIO_ADVANCE(uiop, tlen);
2579 } else
2580 nfsm_adv(nfsm_rndup(len));
2581 #ifndef NFS_V2_ONLY
2582 if (v3) {
2583 nfsm_dissect(tl, u_int32_t *,
2584 3 * NFSX_UNSIGNED);
2585 } else
2586 #endif
2587 {
2588 nfsm_dissect(tl, u_int32_t *,
2589 2 * NFSX_UNSIGNED);
2590 }
2591 if (bigenough) {
2592 #ifndef NFS_V2_ONLY
2593 if (v3) {
2594 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2595 uiop->uio_offset =
2596 fxdr_swapcookie3(tl);
2597 else
2598 uiop->uio_offset =
2599 fxdr_cookie3(tl);
2600 }
2601 else
2602 #endif
2603 {
2604 uiop->uio_offset =
2605 fxdr_unsigned(off_t, *tl);
2606 }
2607 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2608 }
2609 if (v3)
2610 tl += 2;
2611 else
2612 tl++;
2613 more_dirs = fxdr_unsigned(int, *tl);
2614 }
2615 /*
2616 * If at end of rpc data, get the eof boolean
2617 */
2618 if (!more_dirs) {
2619 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2620 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2621
2622 /*
2623 * kludge: if we got no entries, treat it as EOF.
2624 * some server sometimes send a reply without any
2625 * entries or EOF.
2626 * although it might mean the server has very long name,
2627 * we can't handle such entries anyway.
2628 */
2629
2630 if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2631 more_dirs = 0;
2632 }
2633 m_freem(mrep);
2634 }
2635 /*
2636 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2637 * by increasing d_reclen for the last record.
2638 */
2639 if (blksiz > 0) {
2640 left = NFS_DIRFRAGSIZ - blksiz;
2641 memset(uiop->uio_iov->iov_base, 0, left);
2642 dp->d_reclen += left;
2643 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2644 UIO_ADVANCE(uiop, left);
2645 }
2646
2647 /*
2648 * We are now either at the end of the directory or have filled the
2649 * block.
2650 */
2651 if (bigenough) {
2652 dnp->n_direofoffset = uiop->uio_offset;
2653 dnp->n_flag |= NEOFVALID;
2654 }
2655 nfsmout:
2656 return (error);
2657 }
2658
2659 #ifndef NFS_V2_ONLY
2660 /*
2661 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2662 */
2663 int
2664 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, kauth_cred_t cred)
2665 {
2666 int len, left;
2667 struct dirent *dp = NULL;
2668 u_int32_t *tl;
2669 char *cp;
2670 int32_t t1, t2;
2671 struct vnode *newvp;
2672 char *bpos, *dpos, *cp2;
2673 struct mbuf *mreq, *mrep, *md, *mb;
2674 struct nameidata nami, *ndp = &nami;
2675 struct componentname *cnp = &ndp->ni_cnd;
2676 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2677 struct nfsnode *dnp = VTONFS(vp), *np;
2678 nfsfh_t *fhp;
2679 u_quad_t fileno;
2680 int error = 0, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2681 int attrflag, fhsize, nrpcs = 0, reclen;
2682 struct nfs_fattr fattr, *fp;
2683
2684 #ifdef DIAGNOSTIC
2685 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2686 panic("nfs readdirplusrpc bad uio");
2687 #endif
2688 ndp->ni_dvp = vp;
2689 newvp = NULLVP;
2690
2691 /*
2692 * Loop around doing readdir rpc's of size nm_readdirsize
2693 * truncated to a multiple of NFS_DIRFRAGSIZ.
2694 * The stopping criteria is EOF or buffer full.
2695 */
2696 while (more_dirs && bigenough) {
2697 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2698 bigenough = 0;
2699 break;
2700 }
2701 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2702 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2703 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2704 nfsm_fhtom(dnp, 1);
2705 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2706 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2707 txdr_swapcookie3(uiop->uio_offset, tl);
2708 } else {
2709 txdr_cookie3(uiop->uio_offset, tl);
2710 }
2711 tl += 2;
2712 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2713 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2714 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2715 *tl = txdr_unsigned(nmp->nm_rsize);
2716 nfsm_request(dnp, NFSPROC_READDIRPLUS, curlwp, cred);
2717 nfsm_postop_attr(vp, attrflag, 0);
2718 if (error) {
2719 m_freem(mrep);
2720 goto nfsmout;
2721 }
2722 nrpcs++;
2723 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2724 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2725 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2726 more_dirs = fxdr_unsigned(int, *tl);
2727
2728 /* loop thru the dir entries, doctoring them to 4bsd form */
2729 while (more_dirs && bigenough) {
2730 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2731 fileno = fxdr_hyper(tl);
2732 len = fxdr_unsigned(int, *(tl + 2));
2733 if (len <= 0 || len > NFS_MAXNAMLEN) {
2734 error = EBADRPC;
2735 m_freem(mrep);
2736 goto nfsmout;
2737 }
2738 /* for cookie stashing */
2739 reclen = _DIRENT_RECLEN(dp, len) + 2 * sizeof(off_t);
2740 left = NFS_DIRFRAGSIZ - blksiz;
2741 if (reclen > left) {
2742 /*
2743 * DIRFRAGSIZ is aligned, no need to align
2744 * again here.
2745 */
2746 memset(uiop->uio_iov->iov_base, 0, left);
2747 dp->d_reclen += left;
2748 UIO_ADVANCE(uiop, left);
2749 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2750 blksiz = 0;
2751 }
2752 if (reclen > uiop->uio_resid)
2753 bigenough = 0;
2754 if (bigenough) {
2755 int tlen;
2756
2757 dp = (struct dirent *)uiop->uio_iov->iov_base;
2758 dp->d_fileno = fileno;
2759 dp->d_namlen = len;
2760 dp->d_reclen = reclen;
2761 dp->d_type = DT_UNKNOWN;
2762 blksiz += reclen;
2763 if (blksiz == NFS_DIRFRAGSIZ)
2764 blksiz = 0;
2765 UIO_ADVANCE(uiop, DIRHDSIZ);
2766 nfsm_mtouio(uiop, len);
2767 tlen = reclen - (DIRHDSIZ + len);
2768 (void)memset(uiop->uio_iov->iov_base, 0, tlen);
2769 UIO_ADVANCE(uiop, tlen);
2770 cnp->cn_nameptr = dp->d_name;
2771 cnp->cn_namelen = dp->d_namlen;
2772 } else
2773 nfsm_adv(nfsm_rndup(len));
2774 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2775 if (bigenough) {
2776 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2777 uiop->uio_offset =
2778 fxdr_swapcookie3(tl);
2779 else
2780 uiop->uio_offset =
2781 fxdr_cookie3(tl);
2782 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2783 }
2784 tl += 2;
2785
2786 /*
2787 * Since the attributes are before the file handle
2788 * (sigh), we must skip over the attributes and then
2789 * come back and get them.
2790 */
2791 attrflag = fxdr_unsigned(int, *tl);
2792 if (attrflag) {
2793 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2794 memcpy(&fattr, fp, NFSX_V3FATTR);
2795 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2796 doit = fxdr_unsigned(int, *tl);
2797 if (doit) {
2798 nfsm_getfh(fhp, fhsize, 1);
2799 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2800 vref(vp);
2801 newvp = vp;
2802 np = dnp;
2803 } else {
2804 error = nfs_nget1(vp->v_mount, fhp,
2805 fhsize, &np, LK_NOWAIT);
2806 if (!error)
2807 newvp = NFSTOV(np);
2808 }
2809 if (!error) {
2810 const char *xcp;
2811
2812 nfs_loadattrcache(&newvp, &fattr, 0, 0);
2813 if (bigenough) {
2814 dp->d_type =
2815 IFTODT(VTTOIF(np->n_vattr->va_type));
2816 if (cnp->cn_namelen <= NCHNAMLEN) {
2817 ndp->ni_vp = newvp;
2818 xcp = cnp->cn_nameptr +
2819 cnp->cn_namelen;
2820 cnp->cn_hash =
2821 namei_hash(cnp->cn_nameptr, &xcp);
2822 nfs_cache_enter(ndp->ni_dvp,
2823 ndp->ni_vp, cnp);
2824 }
2825 }
2826 }
2827 error = 0;
2828 }
2829 } else {
2830 /* Just skip over the file handle */
2831 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2832 i = fxdr_unsigned(int, *tl);
2833 nfsm_adv(nfsm_rndup(i));
2834 }
2835 if (newvp != NULLVP) {
2836 if (newvp == vp)
2837 vrele(newvp);
2838 else
2839 vput(newvp);
2840 newvp = NULLVP;
2841 }
2842 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2843 more_dirs = fxdr_unsigned(int, *tl);
2844 }
2845 /*
2846 * If at end of rpc data, get the eof boolean
2847 */
2848 if (!more_dirs) {
2849 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2850 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2851
2852 /*
2853 * kludge: see a comment in nfs_readdirrpc.
2854 */
2855
2856 if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2857 more_dirs = 0;
2858 }
2859 m_freem(mrep);
2860 }
2861 /*
2862 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2863 * by increasing d_reclen for the last record.
2864 */
2865 if (blksiz > 0) {
2866 left = NFS_DIRFRAGSIZ - blksiz;
2867 memset(uiop->uio_iov->iov_base, 0, left);
2868 dp->d_reclen += left;
2869 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2870 UIO_ADVANCE(uiop, left);
2871 }
2872
2873 /*
2874 * We are now either at the end of the directory or have filled the
2875 * block.
2876 */
2877 if (bigenough) {
2878 dnp->n_direofoffset = uiop->uio_offset;
2879 dnp->n_flag |= NEOFVALID;
2880 }
2881 nfsmout:
2882 if (newvp != NULLVP) {
2883 if(newvp == vp)
2884 vrele(newvp);
2885 else
2886 vput(newvp);
2887 }
2888 return (error);
2889 }
2890 #endif
2891
2892 /*
2893 * Silly rename. To make the NFS filesystem that is stateless look a little
2894 * more like the "ufs" a remove of an active vnode is translated to a rename
2895 * to a funny looking filename that is removed by nfs_inactive on the
2896 * nfsnode. There is the potential for another process on a different client
2897 * to create the same funny name between the nfs_lookitup() fails and the
2898 * nfs_rename() completes, but...
2899 */
2900 int
2901 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, bool dolink)
2902 {
2903 struct sillyrename *sp;
2904 struct nfsnode *np;
2905 int error;
2906 pid_t pid;
2907
2908 cache_purge(dvp);
2909 np = VTONFS(vp);
2910 #ifndef DIAGNOSTIC
2911 if (vp->v_type == VDIR)
2912 panic("nfs: sillyrename dir");
2913 #endif
2914 sp = kmem_alloc(sizeof(*sp), KM_SLEEP);
2915 sp->s_cred = kauth_cred_dup(cnp->cn_cred);
2916 sp->s_dvp = dvp;
2917 vref(dvp);
2918
2919 /* Fudge together a funny name */
2920 pid = curlwp->l_proc->p_pid;
2921 memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2922 sp->s_namlen = 12;
2923 sp->s_name[8] = hexdigits[pid & 0xf];
2924 sp->s_name[7] = hexdigits[(pid >> 4) & 0xf];
2925 sp->s_name[6] = hexdigits[(pid >> 8) & 0xf];
2926 sp->s_name[5] = hexdigits[(pid >> 12) & 0xf];
2927
2928 /* Try lookitups until we get one that isn't there */
2929 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2930 curlwp, (struct nfsnode **)0) == 0) {
2931 sp->s_name[4]++;
2932 if (sp->s_name[4] > 'z') {
2933 error = EINVAL;
2934 goto bad;
2935 }
2936 }
2937 if (dolink) {
2938 error = nfs_linkrpc(dvp, vp, sp->s_name, sp->s_namlen,
2939 sp->s_cred, curlwp);
2940 /*
2941 * nfs_request maps NFSERR_NOTSUPP to ENOTSUP.
2942 */
2943 if (error == ENOTSUP) {
2944 error = nfs_renameit(dvp, cnp, sp);
2945 }
2946 } else {
2947 error = nfs_renameit(dvp, cnp, sp);
2948 }
2949 if (error)
2950 goto bad;
2951 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2952 curlwp, &np);
2953 np->n_sillyrename = sp;
2954 return (0);
2955 bad:
2956 vrele(sp->s_dvp);
2957 kauth_cred_free(sp->s_cred);
2958 kmem_free(sp, sizeof(*sp));
2959 return (error);
2960 }
2961
2962 /*
2963 * Look up a file name and optionally either update the file handle or
2964 * allocate an nfsnode, depending on the value of npp.
2965 * npp == NULL --> just do the lookup
2966 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2967 * handled too
2968 * *npp != NULL --> update the file handle in the vnode
2969 */
2970 int
2971 nfs_lookitup(struct vnode *dvp, const char *name, int len, kauth_cred_t cred, struct lwp *l, struct nfsnode **npp)
2972 {
2973 u_int32_t *tl;
2974 char *cp;
2975 int32_t t1, t2;
2976 struct vnode *newvp = (struct vnode *)0;
2977 struct nfsnode *np, *dnp = VTONFS(dvp);
2978 char *bpos, *dpos, *cp2;
2979 int error = 0, fhlen;
2980 #ifndef NFS_V2_ONLY
2981 int attrflag;
2982 #endif
2983 struct mbuf *mreq, *mrep, *md, *mb;
2984 nfsfh_t *nfhp;
2985 const int v3 = NFS_ISV3(dvp);
2986
2987 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2988 nfsm_reqhead(dnp, NFSPROC_LOOKUP,
2989 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2990 nfsm_fhtom(dnp, v3);
2991 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2992 nfsm_request(dnp, NFSPROC_LOOKUP, l, cred);
2993 if (npp && !error) {
2994 nfsm_getfh(nfhp, fhlen, v3);
2995 if (*npp) {
2996 np = *npp;
2997 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2998 kmem_free(np->n_fhp, np->n_fhsize);
2999 np->n_fhp = &np->n_fh;
3000 }
3001 #if NFS_SMALLFH < NFSX_V3FHMAX
3002 else if (np->n_fhsize <= NFS_SMALLFH && fhlen > NFS_SMALLFH)
3003 np->n_fhp = kmem_alloc(fhlen, KM_SLEEP);
3004 #endif
3005 memcpy(np->n_fhp, nfhp, fhlen);
3006 np->n_fhsize = fhlen;
3007 newvp = NFSTOV(np);
3008 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
3009 vref(dvp);
3010 newvp = dvp;
3011 np = dnp;
3012 } else {
3013 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
3014 if (error) {
3015 m_freem(mrep);
3016 return (error);
3017 }
3018 newvp = NFSTOV(np);
3019 }
3020 #ifndef NFS_V2_ONLY
3021 if (v3) {
3022 nfsm_postop_attr(newvp, attrflag, 0);
3023 if (!attrflag && *npp == NULL) {
3024 m_freem(mrep);
3025 vput(newvp);
3026 return (ENOENT);
3027 }
3028 } else
3029 #endif
3030 nfsm_loadattr(newvp, (struct vattr *)0, 0);
3031 }
3032 nfsm_reqdone;
3033 if (npp && *npp == NULL) {
3034 if (error) {
3035 if (newvp)
3036 vput(newvp);
3037 } else
3038 *npp = np;
3039 }
3040 return (error);
3041 }
3042
3043 #ifndef NFS_V2_ONLY
3044 /*
3045 * Nfs Version 3 commit rpc
3046 */
3047 int
3048 nfs_commit(struct vnode *vp, off_t offset, uint32_t cnt, struct lwp *l)
3049 {
3050 char *cp;
3051 u_int32_t *tl;
3052 int32_t t1, t2;
3053 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3054 char *bpos, *dpos, *cp2;
3055 int error = 0, wccflag = NFSV3_WCCRATTR;
3056 struct mbuf *mreq, *mrep, *md, *mb;
3057 struct nfsnode *np;
3058
3059 KASSERT(NFS_ISV3(vp));
3060
3061 #ifdef NFS_DEBUG_COMMIT
3062 printf("commit %lu - %lu\n", (unsigned long)offset,
3063 (unsigned long)(offset + cnt));
3064 #endif
3065
3066 mutex_enter(&nmp->nm_lock);
3067 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
3068 mutex_exit(&nmp->nm_lock);
3069 return (0);
3070 }
3071 mutex_exit(&nmp->nm_lock);
3072 nfsstats.rpccnt[NFSPROC_COMMIT]++;
3073 np = VTONFS(vp);
3074 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
3075 nfsm_fhtom(np, 1);
3076 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3077 txdr_hyper(offset, tl);
3078 tl += 2;
3079 *tl = txdr_unsigned(cnt);
3080 nfsm_request(np, NFSPROC_COMMIT, l, np->n_wcred);
3081 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC, false);
3082 if (!error) {
3083 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
3084 mutex_enter(&nmp->nm_lock);
3085 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
3086 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
3087 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
3088 error = NFSERR_STALEWRITEVERF;
3089 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
3090 }
3091 mutex_exit(&nmp->nm_lock);
3092 }
3093 nfsm_reqdone;
3094 return (error);
3095 }
3096 #endif
3097
3098 /*
3099 * Kludge City..
3100 * - make nfs_bmap() essentially a no-op that does no translation
3101 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3102 * (Maybe I could use the process's page mapping, but I was concerned that
3103 * Kernel Write might not be enabled and also figured copyout() would do
3104 * a lot more work than memcpy() and also it currently happens in the
3105 * context of the swapper process (2).
3106 */
3107 int
3108 nfs_bmap(void *v)
3109 {
3110 struct vop_bmap_args /* {
3111 struct vnode *a_vp;
3112 daddr_t a_bn;
3113 struct vnode **a_vpp;
3114 daddr_t *a_bnp;
3115 int *a_runp;
3116 } */ *ap = v;
3117 struct vnode *vp = ap->a_vp;
3118 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
3119
3120 if (ap->a_vpp != NULL)
3121 *ap->a_vpp = vp;
3122 if (ap->a_bnp != NULL)
3123 *ap->a_bnp = ap->a_bn << bshift;
3124 if (ap->a_runp != NULL)
3125 *ap->a_runp = 1024 * 1024; /* XXX */
3126 return (0);
3127 }
3128
3129 /*
3130 * Strategy routine.
3131 * For async requests when nfsiod(s) are running, queue the request by
3132 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3133 * request.
3134 */
3135 int
3136 nfs_strategy(void *v)
3137 {
3138 struct vop_strategy_args *ap = v;
3139 struct buf *bp = ap->a_bp;
3140 int error = 0;
3141
3142 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3143 panic("nfs physio/async");
3144
3145 /*
3146 * If the op is asynchronous and an i/o daemon is waiting
3147 * queue the request, wake it up and wait for completion
3148 * otherwise just do it ourselves.
3149 */
3150 if ((bp->b_flags & B_ASYNC) == 0 || nfs_asyncio(bp))
3151 error = nfs_doio(bp);
3152 return (error);
3153 }
3154
3155 /*
3156 * fsync vnode op. Just call nfs_flush() with commit == 1.
3157 */
3158 /* ARGSUSED */
3159 int
3160 nfs_fsync(void *v)
3161 {
3162 struct vop_fsync_args /* {
3163 struct vnodeop_desc *a_desc;
3164 struct vnode * a_vp;
3165 kauth_cred_t a_cred;
3166 int a_flags;
3167 off_t offlo;
3168 off_t offhi;
3169 struct lwp * a_l;
3170 } */ *ap = v;
3171
3172 struct vnode *vp = ap->a_vp;
3173
3174 if (vp->v_type != VREG)
3175 return 0;
3176
3177 return (nfs_flush(vp, ap->a_cred,
3178 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, curlwp, 1));
3179 }
3180
3181 /*
3182 * Flush all the data associated with a vnode.
3183 */
3184 int
3185 nfs_flush(struct vnode *vp, kauth_cred_t cred, int waitfor, struct lwp *l,
3186 int commit)
3187 {
3188 struct nfsnode *np = VTONFS(vp);
3189 int error;
3190 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3191 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3192
3193 mutex_enter(&vp->v_interlock);
3194 error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3195 if (np->n_flag & NWRITEERR) {
3196 error = np->n_error;
3197 np->n_flag &= ~NWRITEERR;
3198 }
3199 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3200 return (error);
3201 }
3202
3203 /*
3204 * Return POSIX pathconf information applicable to nfs.
3205 *
3206 * N.B. The NFS V2 protocol doesn't support this RPC.
3207 */
3208 /* ARGSUSED */
3209 int
3210 nfs_pathconf(void *v)
3211 {
3212 struct vop_pathconf_args /* {
3213 struct vnode *a_vp;
3214 int a_name;
3215 register_t *a_retval;
3216 } */ *ap = v;
3217 struct nfsv3_pathconf *pcp;
3218 struct vnode *vp = ap->a_vp;
3219 struct mbuf *mreq, *mrep, *md, *mb;
3220 int32_t t1, t2;
3221 u_int32_t *tl;
3222 char *bpos, *dpos, *cp, *cp2;
3223 int error = 0, attrflag;
3224 #ifndef NFS_V2_ONLY
3225 struct nfsmount *nmp;
3226 unsigned int l;
3227 u_int64_t maxsize;
3228 #endif
3229 const int v3 = NFS_ISV3(vp);
3230 struct nfsnode *np = VTONFS(vp);
3231
3232 switch (ap->a_name) {
3233 /* Names that can be resolved locally. */
3234 case _PC_PIPE_BUF:
3235 *ap->a_retval = PIPE_BUF;
3236 break;
3237 case _PC_SYNC_IO:
3238 *ap->a_retval = 1;
3239 break;
3240 /* Names that cannot be resolved locally; do an RPC, if possible. */
3241 case _PC_LINK_MAX:
3242 case _PC_NAME_MAX:
3243 case _PC_CHOWN_RESTRICTED:
3244 case _PC_NO_TRUNC:
3245 if (!v3) {
3246 error = EINVAL;
3247 break;
3248 }
3249 nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3250 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3251 nfsm_fhtom(np, 1);
3252 nfsm_request(np, NFSPROC_PATHCONF,
3253 curlwp, curlwp->l_cred); /* XXX */
3254 nfsm_postop_attr(vp, attrflag, 0);
3255 if (!error) {
3256 nfsm_dissect(pcp, struct nfsv3_pathconf *,
3257 NFSX_V3PATHCONF);
3258 switch (ap->a_name) {
3259 case _PC_LINK_MAX:
3260 *ap->a_retval =
3261 fxdr_unsigned(register_t, pcp->pc_linkmax);
3262 break;
3263 case _PC_NAME_MAX:
3264 *ap->a_retval =
3265 fxdr_unsigned(register_t, pcp->pc_namemax);
3266 break;
3267 case _PC_CHOWN_RESTRICTED:
3268 *ap->a_retval =
3269 (pcp->pc_chownrestricted == nfs_true);
3270 break;
3271 case _PC_NO_TRUNC:
3272 *ap->a_retval =
3273 (pcp->pc_notrunc == nfs_true);
3274 break;
3275 }
3276 }
3277 nfsm_reqdone;
3278 break;
3279 case _PC_FILESIZEBITS:
3280 #ifndef NFS_V2_ONLY
3281 if (v3) {
3282 nmp = VFSTONFS(vp->v_mount);
3283 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3284 if ((error = nfs_fsinfo(nmp, vp,
3285 curlwp->l_cred, curlwp)) != 0) /* XXX */
3286 break;
3287 for (l = 0, maxsize = nmp->nm_maxfilesize;
3288 (maxsize >> l) > 0; l++)
3289 ;
3290 *ap->a_retval = l + 1;
3291 } else
3292 #endif
3293 {
3294 *ap->a_retval = 32; /* NFS V2 limitation */
3295 }
3296 break;
3297 default:
3298 error = EINVAL;
3299 break;
3300 }
3301
3302 return (error);
3303 }
3304
3305 /*
3306 * NFS advisory byte-level locks.
3307 */
3308 int
3309 nfs_advlock(void *v)
3310 {
3311 struct vop_advlock_args /* {
3312 struct vnode *a_vp;
3313 void *a_id;
3314 int a_op;
3315 struct flock *a_fl;
3316 int a_flags;
3317 } */ *ap = v;
3318 struct nfsnode *np = VTONFS(ap->a_vp);
3319
3320 return lf_advlock(ap, &np->n_lockf, np->n_size);
3321 }
3322
3323 /*
3324 * Print out the contents of an nfsnode.
3325 */
3326 int
3327 nfs_print(void *v)
3328 {
3329 struct vop_print_args /* {
3330 struct vnode *a_vp;
3331 } */ *ap = v;
3332 struct vnode *vp = ap->a_vp;
3333 struct nfsnode *np = VTONFS(vp);
3334
3335 printf("tag VT_NFS, fileid %lld fsid 0x%llx",
3336 (unsigned long long)np->n_vattr->va_fileid,
3337 (unsigned long long)np->n_vattr->va_fsid);
3338 if (vp->v_type == VFIFO)
3339 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v);
3340 printf("\n");
3341 return (0);
3342 }
3343
3344 /*
3345 * nfs unlock wrapper.
3346 */
3347 int
3348 nfs_unlock(void *v)
3349 {
3350 struct vop_unlock_args /* {
3351 struct vnode *a_vp;
3352 int a_flags;
3353 } */ *ap = v;
3354 struct vnode *vp = ap->a_vp;
3355
3356 /*
3357 * VOP_UNLOCK can be called by nfs_loadattrcache
3358 * with v_data == 0.
3359 */
3360 if (VTONFS(vp)) {
3361 nfs_delayedtruncate(vp);
3362 }
3363
3364 return genfs_unlock(v);
3365 }
3366
3367 /*
3368 * nfs special file access vnode op.
3369 * Essentially just get vattr and then imitate iaccess() since the device is
3370 * local to the client.
3371 */
3372 int
3373 nfsspec_access(void *v)
3374 {
3375 struct vop_access_args /* {
3376 struct vnode *a_vp;
3377 int a_mode;
3378 kauth_cred_t a_cred;
3379 struct lwp *a_l;
3380 } */ *ap = v;
3381 struct vattr va;
3382 struct vnode *vp = ap->a_vp;
3383 int error;
3384
3385 error = VOP_GETATTR(vp, &va, ap->a_cred);
3386 if (error)
3387 return (error);
3388
3389 /*
3390 * Disallow write attempts on filesystems mounted read-only;
3391 * unless the file is a socket, fifo, or a block or character
3392 * device resident on the filesystem.
3393 */
3394 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3395 switch (vp->v_type) {
3396 case VREG:
3397 case VDIR:
3398 case VLNK:
3399 return (EROFS);
3400 default:
3401 break;
3402 }
3403 }
3404
3405 return (genfs_can_access(va.va_type, va.va_mode,
3406 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3407 }
3408
3409 /*
3410 * Read wrapper for special devices.
3411 */
3412 int
3413 nfsspec_read(void *v)
3414 {
3415 struct vop_read_args /* {
3416 struct vnode *a_vp;
3417 struct uio *a_uio;
3418 int a_ioflag;
3419 kauth_cred_t a_cred;
3420 } */ *ap = v;
3421 struct nfsnode *np = VTONFS(ap->a_vp);
3422
3423 /*
3424 * Set access flag.
3425 */
3426 np->n_flag |= NACC;
3427 getnanotime(&np->n_atim);
3428 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3429 }
3430
3431 /*
3432 * Write wrapper for special devices.
3433 */
3434 int
3435 nfsspec_write(void *v)
3436 {
3437 struct vop_write_args /* {
3438 struct vnode *a_vp;
3439 struct uio *a_uio;
3440 int a_ioflag;
3441 kauth_cred_t a_cred;
3442 } */ *ap = v;
3443 struct nfsnode *np = VTONFS(ap->a_vp);
3444
3445 /*
3446 * Set update flag.
3447 */
3448 np->n_flag |= NUPD;
3449 getnanotime(&np->n_mtim);
3450 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3451 }
3452
3453 /*
3454 * Close wrapper for special devices.
3455 *
3456 * Update the times on the nfsnode then do device close.
3457 */
3458 int
3459 nfsspec_close(void *v)
3460 {
3461 struct vop_close_args /* {
3462 struct vnode *a_vp;
3463 int a_fflag;
3464 kauth_cred_t a_cred;
3465 struct lwp *a_l;
3466 } */ *ap = v;
3467 struct vnode *vp = ap->a_vp;
3468 struct nfsnode *np = VTONFS(vp);
3469 struct vattr vattr;
3470
3471 if (np->n_flag & (NACC | NUPD)) {
3472 np->n_flag |= NCHG;
3473 if (vp->v_usecount == 1 &&
3474 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3475 vattr_null(&vattr);
3476 if (np->n_flag & NACC)
3477 vattr.va_atime = np->n_atim;
3478 if (np->n_flag & NUPD)
3479 vattr.va_mtime = np->n_mtim;
3480 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3481 }
3482 }
3483 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3484 }
3485
3486 /*
3487 * Read wrapper for fifos.
3488 */
3489 int
3490 nfsfifo_read(void *v)
3491 {
3492 struct vop_read_args /* {
3493 struct vnode *a_vp;
3494 struct uio *a_uio;
3495 int a_ioflag;
3496 kauth_cred_t a_cred;
3497 } */ *ap = v;
3498 struct nfsnode *np = VTONFS(ap->a_vp);
3499
3500 /*
3501 * Set access flag.
3502 */
3503 np->n_flag |= NACC;
3504 getnanotime(&np->n_atim);
3505 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3506 }
3507
3508 /*
3509 * Write wrapper for fifos.
3510 */
3511 int
3512 nfsfifo_write(void *v)
3513 {
3514 struct vop_write_args /* {
3515 struct vnode *a_vp;
3516 struct uio *a_uio;
3517 int a_ioflag;
3518 kauth_cred_t a_cred;
3519 } */ *ap = v;
3520 struct nfsnode *np = VTONFS(ap->a_vp);
3521
3522 /*
3523 * Set update flag.
3524 */
3525 np->n_flag |= NUPD;
3526 getnanotime(&np->n_mtim);
3527 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3528 }
3529
3530 /*
3531 * Close wrapper for fifos.
3532 *
3533 * Update the times on the nfsnode then do fifo close.
3534 */
3535 int
3536 nfsfifo_close(void *v)
3537 {
3538 struct vop_close_args /* {
3539 struct vnode *a_vp;
3540 int a_fflag;
3541 kauth_cred_t a_cred;
3542 struct lwp *a_l;
3543 } */ *ap = v;
3544 struct vnode *vp = ap->a_vp;
3545 struct nfsnode *np = VTONFS(vp);
3546 struct vattr vattr;
3547
3548 if (np->n_flag & (NACC | NUPD)) {
3549 struct timespec ts;
3550
3551 getnanotime(&ts);
3552 if (np->n_flag & NACC)
3553 np->n_atim = ts;
3554 if (np->n_flag & NUPD)
3555 np->n_mtim = ts;
3556 np->n_flag |= NCHG;
3557 if (vp->v_usecount == 1 &&
3558 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3559 vattr_null(&vattr);
3560 if (np->n_flag & NACC)
3561 vattr.va_atime = np->n_atim;
3562 if (np->n_flag & NUPD)
3563 vattr.va_mtime = np->n_mtim;
3564 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3565 }
3566 }
3567 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3568 }
3569