nfs_vnops.c revision 1.171 1 /* $NetBSD: nfs_vnops.c,v 1.171 2003/06/03 14:27:48 yamt Exp $ */
2
3 /*
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * @(#)nfs_vnops.c 8.19 (Berkeley) 7/31/95
39 */
40
41 /*
42 * vnode op calls for Sun NFS version 2 and 3
43 */
44
45 #include <sys/cdefs.h>
46 __KERNEL_RCSID(0, "$NetBSD: nfs_vnops.c,v 1.171 2003/06/03 14:27:48 yamt Exp $");
47
48 #include "opt_nfs.h"
49 #include "opt_uvmhist.h"
50
51 #include <sys/param.h>
52 #include <sys/proc.h>
53 #include <sys/kernel.h>
54 #include <sys/systm.h>
55 #include <sys/resourcevar.h>
56 #include <sys/proc.h>
57 #include <sys/mount.h>
58 #include <sys/buf.h>
59 #include <sys/malloc.h>
60 #include <sys/mbuf.h>
61 #include <sys/namei.h>
62 #include <sys/vnode.h>
63 #include <sys/dirent.h>
64 #include <sys/fcntl.h>
65 #include <sys/hash.h>
66 #include <sys/lockf.h>
67 #include <sys/stat.h>
68 #include <sys/unistd.h>
69
70 #include <uvm/uvm_extern.h>
71 #include <uvm/uvm.h>
72
73 #include <miscfs/fifofs/fifo.h>
74 #include <miscfs/genfs/genfs.h>
75 #include <miscfs/specfs/specdev.h>
76
77 #include <nfs/rpcv2.h>
78 #include <nfs/nfsproto.h>
79 #include <nfs/nfs.h>
80 #include <nfs/nfsnode.h>
81 #include <nfs/nfsmount.h>
82 #include <nfs/xdr_subs.h>
83 #include <nfs/nfsm_subs.h>
84 #include <nfs/nqnfs.h>
85 #include <nfs/nfs_var.h>
86
87 #include <net/if.h>
88 #include <netinet/in.h>
89 #include <netinet/in_var.h>
90
91 /*
92 * Global vfs data structures for nfs
93 */
94 int (**nfsv2_vnodeop_p) __P((void *));
95 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
96 { &vop_default_desc, vn_default_error },
97 { &vop_lookup_desc, nfs_lookup }, /* lookup */
98 { &vop_create_desc, nfs_create }, /* create */
99 { &vop_mknod_desc, nfs_mknod }, /* mknod */
100 { &vop_open_desc, nfs_open }, /* open */
101 { &vop_close_desc, nfs_close }, /* close */
102 { &vop_access_desc, nfs_access }, /* access */
103 { &vop_getattr_desc, nfs_getattr }, /* getattr */
104 { &vop_setattr_desc, nfs_setattr }, /* setattr */
105 { &vop_read_desc, nfs_read }, /* read */
106 { &vop_write_desc, nfs_write }, /* write */
107 { &vop_lease_desc, nfs_lease_check }, /* lease */
108 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
109 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */
110 { &vop_poll_desc, nfs_poll }, /* poll */
111 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */
112 { &vop_revoke_desc, nfs_revoke }, /* revoke */
113 { &vop_mmap_desc, nfs_mmap }, /* mmap */
114 { &vop_fsync_desc, nfs_fsync }, /* fsync */
115 { &vop_seek_desc, nfs_seek }, /* seek */
116 { &vop_remove_desc, nfs_remove }, /* remove */
117 { &vop_link_desc, nfs_link }, /* link */
118 { &vop_rename_desc, nfs_rename }, /* rename */
119 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */
120 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */
121 { &vop_symlink_desc, nfs_symlink }, /* symlink */
122 { &vop_readdir_desc, nfs_readdir }, /* readdir */
123 { &vop_readlink_desc, nfs_readlink }, /* readlink */
124 { &vop_abortop_desc, nfs_abortop }, /* abortop */
125 { &vop_inactive_desc, nfs_inactive }, /* inactive */
126 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
127 { &vop_lock_desc, nfs_lock }, /* lock */
128 { &vop_unlock_desc, nfs_unlock }, /* unlock */
129 { &vop_bmap_desc, nfs_bmap }, /* bmap */
130 { &vop_strategy_desc, nfs_strategy }, /* strategy */
131 { &vop_print_desc, nfs_print }, /* print */
132 { &vop_islocked_desc, nfs_islocked }, /* islocked */
133 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */
134 { &vop_advlock_desc, nfs_advlock }, /* advlock */
135 { &vop_blkatoff_desc, nfs_blkatoff }, /* blkatoff */
136 { &vop_valloc_desc, nfs_valloc }, /* valloc */
137 { &vop_reallocblks_desc, nfs_reallocblks }, /* reallocblks */
138 { &vop_vfree_desc, nfs_vfree }, /* vfree */
139 { &vop_truncate_desc, nfs_truncate }, /* truncate */
140 { &vop_update_desc, nfs_update }, /* update */
141 { &vop_bwrite_desc, nfs_bwrite }, /* bwrite */
142 { &vop_getpages_desc, nfs_getpages }, /* getpages */
143 { &vop_putpages_desc, genfs_putpages }, /* putpages */
144 { NULL, NULL }
145 };
146 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
147 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
148
149 /*
150 * Special device vnode ops
151 */
152 int (**spec_nfsv2nodeop_p) __P((void *));
153 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
154 { &vop_default_desc, vn_default_error },
155 { &vop_lookup_desc, spec_lookup }, /* lookup */
156 { &vop_create_desc, spec_create }, /* create */
157 { &vop_mknod_desc, spec_mknod }, /* mknod */
158 { &vop_open_desc, spec_open }, /* open */
159 { &vop_close_desc, nfsspec_close }, /* close */
160 { &vop_access_desc, nfsspec_access }, /* access */
161 { &vop_getattr_desc, nfs_getattr }, /* getattr */
162 { &vop_setattr_desc, nfs_setattr }, /* setattr */
163 { &vop_read_desc, nfsspec_read }, /* read */
164 { &vop_write_desc, nfsspec_write }, /* write */
165 { &vop_lease_desc, spec_lease_check }, /* lease */
166 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
167 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */
168 { &vop_poll_desc, spec_poll }, /* poll */
169 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */
170 { &vop_revoke_desc, spec_revoke }, /* revoke */
171 { &vop_mmap_desc, spec_mmap }, /* mmap */
172 { &vop_fsync_desc, spec_fsync }, /* fsync */
173 { &vop_seek_desc, spec_seek }, /* seek */
174 { &vop_remove_desc, spec_remove }, /* remove */
175 { &vop_link_desc, spec_link }, /* link */
176 { &vop_rename_desc, spec_rename }, /* rename */
177 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */
178 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */
179 { &vop_symlink_desc, spec_symlink }, /* symlink */
180 { &vop_readdir_desc, spec_readdir }, /* readdir */
181 { &vop_readlink_desc, spec_readlink }, /* readlink */
182 { &vop_abortop_desc, spec_abortop }, /* abortop */
183 { &vop_inactive_desc, nfs_inactive }, /* inactive */
184 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
185 { &vop_lock_desc, nfs_lock }, /* lock */
186 { &vop_unlock_desc, nfs_unlock }, /* unlock */
187 { &vop_bmap_desc, spec_bmap }, /* bmap */
188 { &vop_strategy_desc, spec_strategy }, /* strategy */
189 { &vop_print_desc, nfs_print }, /* print */
190 { &vop_islocked_desc, nfs_islocked }, /* islocked */
191 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */
192 { &vop_advlock_desc, spec_advlock }, /* advlock */
193 { &vop_blkatoff_desc, spec_blkatoff }, /* blkatoff */
194 { &vop_valloc_desc, spec_valloc }, /* valloc */
195 { &vop_reallocblks_desc, spec_reallocblks }, /* reallocblks */
196 { &vop_vfree_desc, spec_vfree }, /* vfree */
197 { &vop_truncate_desc, spec_truncate }, /* truncate */
198 { &vop_update_desc, nfs_update }, /* update */
199 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
200 { &vop_getpages_desc, spec_getpages }, /* getpages */
201 { &vop_putpages_desc, spec_putpages }, /* putpages */
202 { NULL, NULL }
203 };
204 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
205 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
206
207 int (**fifo_nfsv2nodeop_p) __P((void *));
208 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
209 { &vop_default_desc, vn_default_error },
210 { &vop_lookup_desc, fifo_lookup }, /* lookup */
211 { &vop_create_desc, fifo_create }, /* create */
212 { &vop_mknod_desc, fifo_mknod }, /* mknod */
213 { &vop_open_desc, fifo_open }, /* open */
214 { &vop_close_desc, nfsfifo_close }, /* close */
215 { &vop_access_desc, nfsspec_access }, /* access */
216 { &vop_getattr_desc, nfs_getattr }, /* getattr */
217 { &vop_setattr_desc, nfs_setattr }, /* setattr */
218 { &vop_read_desc, nfsfifo_read }, /* read */
219 { &vop_write_desc, nfsfifo_write }, /* write */
220 { &vop_lease_desc, fifo_lease_check }, /* lease */
221 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
222 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */
223 { &vop_poll_desc, fifo_poll }, /* poll */
224 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */
225 { &vop_revoke_desc, fifo_revoke }, /* revoke */
226 { &vop_mmap_desc, fifo_mmap }, /* mmap */
227 { &vop_fsync_desc, nfs_fsync }, /* fsync */
228 { &vop_seek_desc, fifo_seek }, /* seek */
229 { &vop_remove_desc, fifo_remove }, /* remove */
230 { &vop_link_desc, fifo_link }, /* link */
231 { &vop_rename_desc, fifo_rename }, /* rename */
232 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */
233 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */
234 { &vop_symlink_desc, fifo_symlink }, /* symlink */
235 { &vop_readdir_desc, fifo_readdir }, /* readdir */
236 { &vop_readlink_desc, fifo_readlink }, /* readlink */
237 { &vop_abortop_desc, fifo_abortop }, /* abortop */
238 { &vop_inactive_desc, nfs_inactive }, /* inactive */
239 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
240 { &vop_lock_desc, nfs_lock }, /* lock */
241 { &vop_unlock_desc, nfs_unlock }, /* unlock */
242 { &vop_bmap_desc, fifo_bmap }, /* bmap */
243 { &vop_strategy_desc, genfs_badop }, /* strategy */
244 { &vop_print_desc, nfs_print }, /* print */
245 { &vop_islocked_desc, nfs_islocked }, /* islocked */
246 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */
247 { &vop_advlock_desc, fifo_advlock }, /* advlock */
248 { &vop_blkatoff_desc, fifo_blkatoff }, /* blkatoff */
249 { &vop_valloc_desc, fifo_valloc }, /* valloc */
250 { &vop_reallocblks_desc, fifo_reallocblks }, /* reallocblks */
251 { &vop_vfree_desc, fifo_vfree }, /* vfree */
252 { &vop_truncate_desc, fifo_truncate }, /* truncate */
253 { &vop_update_desc, nfs_update }, /* update */
254 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
255 { &vop_putpages_desc, fifo_putpages }, /* putpages */
256 { NULL, NULL }
257 };
258 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
259 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
260
261 static void nfs_writerpc_extfree(struct mbuf *, caddr_t, size_t, void *);
262
263 /*
264 * Global variables
265 */
266 extern u_int32_t nfs_true, nfs_false;
267 extern u_int32_t nfs_xdrneg1;
268 extern const nfstype nfsv3_type[9];
269
270 int nfs_numasync = 0;
271 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
272
273 /*
274 * nfs null call from vfs.
275 */
276 int
277 nfs_null(vp, cred, procp)
278 struct vnode *vp;
279 struct ucred *cred;
280 struct proc *procp;
281 {
282 caddr_t bpos, dpos;
283 int error = 0;
284 struct mbuf *mreq, *mrep, *md, *mb;
285 struct nfsnode *np = VTONFS(vp);
286
287 nfsm_reqhead(np, NFSPROC_NULL, 0);
288 nfsm_request(np, NFSPROC_NULL, procp, cred);
289 nfsm_reqdone;
290 return (error);
291 }
292
293 /*
294 * nfs access vnode op.
295 * For nfs version 2, just return ok. File accesses may fail later.
296 * For nfs version 3, use the access rpc to check accessibility. If file modes
297 * are changed on the server, accesses might still fail later.
298 */
299 int
300 nfs_access(v)
301 void *v;
302 {
303 struct vop_access_args /* {
304 struct vnode *a_vp;
305 int a_mode;
306 struct ucred *a_cred;
307 struct proc *a_p;
308 } */ *ap = v;
309 struct vnode *vp = ap->a_vp;
310 u_int32_t *tl;
311 caddr_t cp;
312 int32_t t1, t2;
313 caddr_t bpos, dpos, cp2;
314 int error = 0, attrflag, cachevalid;
315 struct mbuf *mreq, *mrep, *md, *mb;
316 u_int32_t mode, rmode;
317 const int v3 = NFS_ISV3(vp);
318 struct nfsnode *np = VTONFS(vp);
319
320 cachevalid = (np->n_accstamp != -1 &&
321 (time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) &&
322 np->n_accuid == ap->a_cred->cr_uid);
323
324 /*
325 * Check access cache first. If this request has been made for this
326 * uid shortly before, use the cached result.
327 */
328 if (cachevalid) {
329 if (!np->n_accerror) {
330 if ((np->n_accmode & ap->a_mode) == ap->a_mode)
331 return np->n_accerror;
332 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
333 return np->n_accerror;
334 }
335
336 /*
337 * For nfs v3, do an access rpc, otherwise you are stuck emulating
338 * ufs_access() locally using the vattr. This may not be correct,
339 * since the server may apply other access criteria such as
340 * client uid-->server uid mapping that we do not know about, but
341 * this is better than just returning anything that is lying about
342 * in the cache.
343 */
344 if (v3) {
345 nfsstats.rpccnt[NFSPROC_ACCESS]++;
346 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
347 nfsm_fhtom(np, v3);
348 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
349 if (ap->a_mode & VREAD)
350 mode = NFSV3ACCESS_READ;
351 else
352 mode = 0;
353 if (vp->v_type != VDIR) {
354 if (ap->a_mode & VWRITE)
355 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
356 if (ap->a_mode & VEXEC)
357 mode |= NFSV3ACCESS_EXECUTE;
358 } else {
359 if (ap->a_mode & VWRITE)
360 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
361 NFSV3ACCESS_DELETE);
362 if (ap->a_mode & VEXEC)
363 mode |= NFSV3ACCESS_LOOKUP;
364 }
365 *tl = txdr_unsigned(mode);
366 nfsm_request(np, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
367 nfsm_postop_attr(vp, attrflag, 0);
368 if (!error) {
369 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
370 rmode = fxdr_unsigned(u_int32_t, *tl);
371 /*
372 * The NFS V3 spec does not clarify whether or not
373 * the returned access bits can be a superset of
374 * the ones requested, so...
375 */
376 if ((rmode & mode) != mode)
377 error = EACCES;
378 }
379 nfsm_reqdone;
380 } else
381 return (nfsspec_access(ap));
382 /*
383 * Disallow write attempts on filesystems mounted read-only;
384 * unless the file is a socket, fifo, or a block or character
385 * device resident on the filesystem.
386 */
387 if (!error && (ap->a_mode & VWRITE) &&
388 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
389 switch (vp->v_type) {
390 case VREG:
391 case VDIR:
392 case VLNK:
393 error = EROFS;
394 default:
395 break;
396 }
397 }
398
399 if (!error || error == EACCES) {
400 /*
401 * If we got the same result as for a previous,
402 * different request, OR it in. Don't update
403 * the timestamp in that case.
404 */
405 if (cachevalid && np->n_accstamp != -1 &&
406 error == np->n_accerror) {
407 if (!error)
408 np->n_accmode |= ap->a_mode;
409 else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
410 np->n_accmode = ap->a_mode;
411 } else {
412 np->n_accstamp = time.tv_sec;
413 np->n_accuid = ap->a_cred->cr_uid;
414 np->n_accmode = ap->a_mode;
415 np->n_accerror = error;
416 }
417 }
418
419 return (error);
420 }
421
422 /*
423 * nfs open vnode op
424 * Check to see if the type is ok
425 * and that deletion is not in progress.
426 * For paged in text files, you will need to flush the page cache
427 * if consistency is lost.
428 */
429 /* ARGSUSED */
430 int
431 nfs_open(v)
432 void *v;
433 {
434 struct vop_open_args /* {
435 struct vnode *a_vp;
436 int a_mode;
437 struct ucred *a_cred;
438 struct proc *a_p;
439 } */ *ap = v;
440 struct vnode *vp = ap->a_vp;
441 struct nfsnode *np = VTONFS(vp);
442 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
443 struct vattr vattr;
444 int error;
445
446 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
447 return (EACCES);
448 }
449
450 /*
451 * Initialize read and write creds here, for swapfiles
452 * and other paths that don't set the creds themselves.
453 */
454
455 if (ap->a_mode & FREAD) {
456 if (np->n_rcred) {
457 crfree(np->n_rcred);
458 }
459 np->n_rcred = ap->a_cred;
460 crhold(np->n_rcred);
461 }
462 if (ap->a_mode & FWRITE) {
463 if (np->n_wcred) {
464 crfree(np->n_wcred);
465 }
466 np->n_wcred = ap->a_cred;
467 crhold(np->n_wcred);
468 }
469
470 #ifndef NFS_V2_ONLY
471 /*
472 * Get a valid lease. If cached data is stale, flush it.
473 */
474 if (nmp->nm_flag & NFSMNT_NQNFS) {
475 if (NQNFS_CKINVALID(vp, np, ND_READ)) {
476 do {
477 error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
478 ap->a_p);
479 } while (error == NQNFS_EXPIRED);
480 if (error)
481 return (error);
482 if (np->n_lrev != np->n_brev ||
483 (np->n_flag & NQNFSNONCACHE)) {
484 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
485 ap->a_p, 1)) == EINTR)
486 return (error);
487 np->n_brev = np->n_lrev;
488 }
489 }
490 } else
491 #endif
492 {
493 if (np->n_flag & NMODIFIED) {
494 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
495 ap->a_p, 1)) == EINTR)
496 return (error);
497 np->n_attrstamp = 0;
498 if (vp->v_type == VDIR) {
499 nfs_invaldircache(vp, 0);
500 np->n_direofoffset = 0;
501 }
502 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
503 if (error)
504 return (error);
505 np->n_mtime = vattr.va_mtime.tv_sec;
506 } else {
507 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
508 if (error)
509 return (error);
510 if (np->n_mtime != vattr.va_mtime.tv_sec) {
511 if (vp->v_type == VDIR) {
512 nfs_invaldircache(vp, 0);
513 np->n_direofoffset = 0;
514 }
515 if ((error = nfs_vinvalbuf(vp, V_SAVE,
516 ap->a_cred, ap->a_p, 1)) == EINTR)
517 return (error);
518 np->n_mtime = vattr.va_mtime.tv_sec;
519 }
520 }
521 }
522 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
523 np->n_attrstamp = 0; /* For Open/Close consistency */
524 return (0);
525 }
526
527 /*
528 * nfs close vnode op
529 * What an NFS client should do upon close after writing is a debatable issue.
530 * Most NFS clients push delayed writes to the server upon close, basically for
531 * two reasons:
532 * 1 - So that any write errors may be reported back to the client process
533 * doing the close system call. By far the two most likely errors are
534 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
535 * 2 - To put a worst case upper bound on cache inconsistency between
536 * multiple clients for the file.
537 * There is also a consistency problem for Version 2 of the protocol w.r.t.
538 * not being able to tell if other clients are writing a file concurrently,
539 * since there is no way of knowing if the changed modify time in the reply
540 * is only due to the write for this client.
541 * (NFS Version 3 provides weak cache consistency data in the reply that
542 * should be sufficient to detect and handle this case.)
543 *
544 * The current code does the following:
545 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
546 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
547 * or commit them (this satisfies 1 and 2 except for the
548 * case where the server crashes after this close but
549 * before the commit RPC, which is felt to be "good
550 * enough". Changing the last argument to nfs_flush() to
551 * a 1 would force a commit operation, if it is felt a
552 * commit is necessary now.
553 * for NQNFS - do nothing now, since 2 is dealt with via leases and
554 * 1 should be dealt with via an fsync() system call for
555 * cases where write errors are important.
556 */
557 /* ARGSUSED */
558 int
559 nfs_close(v)
560 void *v;
561 {
562 struct vop_close_args /* {
563 struct vnodeop_desc *a_desc;
564 struct vnode *a_vp;
565 int a_fflag;
566 struct ucred *a_cred;
567 struct proc *a_p;
568 } */ *ap = v;
569 struct vnode *vp = ap->a_vp;
570 struct nfsnode *np = VTONFS(vp);
571 int error = 0;
572 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
573
574 if (vp->v_type == VREG) {
575 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
576 (np->n_flag & NMODIFIED)) {
577 if (NFS_ISV3(vp)) {
578 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
579 np->n_flag &= ~NMODIFIED;
580 } else
581 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
582 np->n_attrstamp = 0;
583 }
584 if (np->n_flag & NWRITEERR) {
585 np->n_flag &= ~NWRITEERR;
586 error = np->n_error;
587 }
588 }
589 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
590 return (error);
591 }
592
593 /*
594 * nfs getattr call from vfs.
595 */
596 int
597 nfs_getattr(v)
598 void *v;
599 {
600 struct vop_getattr_args /* {
601 struct vnode *a_vp;
602 struct vattr *a_vap;
603 struct ucred *a_cred;
604 struct proc *a_p;
605 } */ *ap = v;
606 struct vnode *vp = ap->a_vp;
607 struct nfsnode *np = VTONFS(vp);
608 caddr_t cp;
609 u_int32_t *tl;
610 int32_t t1, t2;
611 caddr_t bpos, dpos;
612 int error = 0;
613 struct mbuf *mreq, *mrep, *md, *mb;
614 const int v3 = NFS_ISV3(vp);
615
616 /*
617 * Update local times for special files.
618 */
619 if (np->n_flag & (NACC | NUPD))
620 np->n_flag |= NCHG;
621 /*
622 * First look in the cache.
623 */
624 if (nfs_getattrcache(vp, ap->a_vap) == 0)
625 return (0);
626 nfsstats.rpccnt[NFSPROC_GETATTR]++;
627 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
628 nfsm_fhtom(np, v3);
629 nfsm_request(np, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
630 if (!error) {
631 nfsm_loadattr(vp, ap->a_vap, 0);
632 if (vp->v_type == VDIR &&
633 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
634 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
635 }
636 nfsm_reqdone;
637 return (error);
638 }
639
640 /*
641 * nfs setattr call.
642 */
643 int
644 nfs_setattr(v)
645 void *v;
646 {
647 struct vop_setattr_args /* {
648 struct vnodeop_desc *a_desc;
649 struct vnode *a_vp;
650 struct vattr *a_vap;
651 struct ucred *a_cred;
652 struct proc *a_p;
653 } */ *ap = v;
654 struct vnode *vp = ap->a_vp;
655 struct nfsnode *np = VTONFS(vp);
656 struct vattr *vap = ap->a_vap;
657 int error = 0;
658 u_quad_t tsize = 0;
659
660 /*
661 * Setting of flags is not supported.
662 */
663 if (vap->va_flags != VNOVAL)
664 return (EOPNOTSUPP);
665
666 /*
667 * Disallow write attempts if the filesystem is mounted read-only.
668 */
669 if ((vap->va_uid != (uid_t)VNOVAL ||
670 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
671 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
672 (vp->v_mount->mnt_flag & MNT_RDONLY))
673 return (EROFS);
674 if (vap->va_size != VNOVAL) {
675 switch (vp->v_type) {
676 case VDIR:
677 return (EISDIR);
678 case VCHR:
679 case VBLK:
680 case VSOCK:
681 case VFIFO:
682 if (vap->va_mtime.tv_sec == VNOVAL &&
683 vap->va_atime.tv_sec == VNOVAL &&
684 vap->va_mode == (mode_t)VNOVAL &&
685 vap->va_uid == (uid_t)VNOVAL &&
686 vap->va_gid == (gid_t)VNOVAL)
687 return (0);
688 vap->va_size = VNOVAL;
689 break;
690 default:
691 /*
692 * Disallow write attempts if the filesystem is
693 * mounted read-only.
694 */
695 if (vp->v_mount->mnt_flag & MNT_RDONLY)
696 return (EROFS);
697 uvm_vnp_setsize(vp, vap->va_size);
698 tsize = np->n_size;
699 np->n_size = vap->va_size;
700 if (vap->va_size == 0)
701 error = nfs_vinvalbuf(vp, 0,
702 ap->a_cred, ap->a_p, 1);
703 else
704 error = nfs_vinvalbuf(vp, V_SAVE,
705 ap->a_cred, ap->a_p, 1);
706 if (error) {
707 uvm_vnp_setsize(vp, tsize);
708 return (error);
709 }
710 np->n_vattr->va_size = vap->va_size;
711 }
712 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
713 vap->va_atime.tv_sec != VNOVAL) &&
714 vp->v_type == VREG &&
715 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
716 ap->a_p, 1)) == EINTR)
717 return (error);
718 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
719 if (error && vap->va_size != VNOVAL) {
720 np->n_size = np->n_vattr->va_size = tsize;
721 uvm_vnp_setsize(vp, np->n_size);
722 }
723 VN_KNOTE(vp, NOTE_ATTRIB);
724 return (error);
725 }
726
727 /*
728 * Do an nfs setattr rpc.
729 */
730 int
731 nfs_setattrrpc(vp, vap, cred, procp)
732 struct vnode *vp;
733 struct vattr *vap;
734 struct ucred *cred;
735 struct proc *procp;
736 {
737 struct nfsv2_sattr *sp;
738 caddr_t cp;
739 int32_t t1, t2;
740 caddr_t bpos, dpos, cp2;
741 u_int32_t *tl;
742 int error = 0, wccflag = NFSV3_WCCRATTR;
743 struct mbuf *mreq, *mrep, *md, *mb;
744 const int v3 = NFS_ISV3(vp);
745 struct nfsnode *np = VTONFS(vp);
746
747 nfsstats.rpccnt[NFSPROC_SETATTR]++;
748 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
749 nfsm_fhtom(np, v3);
750 if (v3) {
751 nfsm_v3attrbuild(vap, TRUE);
752 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
753 *tl = nfs_false;
754 } else {
755 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
756 if (vap->va_mode == (mode_t)VNOVAL)
757 sp->sa_mode = nfs_xdrneg1;
758 else
759 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
760 if (vap->va_uid == (uid_t)VNOVAL)
761 sp->sa_uid = nfs_xdrneg1;
762 else
763 sp->sa_uid = txdr_unsigned(vap->va_uid);
764 if (vap->va_gid == (gid_t)VNOVAL)
765 sp->sa_gid = nfs_xdrneg1;
766 else
767 sp->sa_gid = txdr_unsigned(vap->va_gid);
768 sp->sa_size = txdr_unsigned(vap->va_size);
769 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
770 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
771 }
772 nfsm_request(np, NFSPROC_SETATTR, procp, cred);
773 if (v3) {
774 nfsm_wcc_data(vp, wccflag, 0);
775 } else
776 nfsm_loadattr(vp, (struct vattr *)0, 0);
777 nfsm_reqdone;
778 return (error);
779 }
780
781 /*
782 * nfs lookup call, one step at a time...
783 * First look in cache
784 * If not found, unlock the directory nfsnode and do the rpc
785 *
786 * This code is full of lock/unlock statements and checks, because
787 * we continue after cache_lookup has finished (we need to check
788 * with the attr cache and do an rpc if it has timed out). This means
789 * that the locking effects of cache_lookup have to be taken into
790 * account.
791 */
792 int
793 nfs_lookup(v)
794 void *v;
795 {
796 struct vop_lookup_args /* {
797 struct vnodeop_desc *a_desc;
798 struct vnode *a_dvp;
799 struct vnode **a_vpp;
800 struct componentname *a_cnp;
801 } */ *ap = v;
802 struct componentname *cnp = ap->a_cnp;
803 struct vnode *dvp = ap->a_dvp;
804 struct vnode **vpp = ap->a_vpp;
805 int flags;
806 struct vnode *newvp;
807 u_int32_t *tl;
808 caddr_t cp;
809 int32_t t1, t2;
810 caddr_t bpos, dpos, cp2;
811 struct mbuf *mreq, *mrep, *md, *mb;
812 long len;
813 nfsfh_t *fhp;
814 struct nfsnode *np;
815 int lockparent, wantparent, error = 0, attrflag, fhsize;
816 const int v3 = NFS_ISV3(dvp);
817
818 cnp->cn_flags &= ~PDIRUNLOCK;
819 flags = cnp->cn_flags;
820
821 *vpp = NULLVP;
822 newvp = NULLVP;
823 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
824 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
825 return (EROFS);
826 if (dvp->v_type != VDIR)
827 return (ENOTDIR);
828
829 lockparent = flags & LOCKPARENT;
830 wantparent = flags & (LOCKPARENT|WANTPARENT);
831 np = VTONFS(dvp);
832
833 /*
834 * Before tediously performing a linear scan of the directory,
835 * check the name cache to see if the directory/name pair
836 * we are looking for is known already.
837 * If the directory/name pair is found in the name cache,
838 * we have to ensure the directory has not changed from
839 * the time the cache entry has been created. If it has,
840 * the cache entry has to be ignored.
841 */
842 if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) {
843 struct vattr vattr;
844 int err2;
845
846 if (error && error != ENOENT) {
847 *vpp = NULLVP;
848 return error;
849 }
850
851 if (cnp->cn_flags & PDIRUNLOCK) {
852 err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
853 if (err2 != 0) {
854 *vpp = NULLVP;
855 return err2;
856 }
857 cnp->cn_flags &= ~PDIRUNLOCK;
858 }
859
860 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
861 if (err2 != 0) {
862 if (error == 0) {
863 if (*vpp != dvp)
864 vput(*vpp);
865 else
866 vrele(*vpp);
867 }
868 *vpp = NULLVP;
869 return err2;
870 }
871
872 if (error == ENOENT) {
873 if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
874 cnp->cn_proc) && vattr.va_mtime.tv_sec ==
875 VTONFS(dvp)->n_nctime)
876 return ENOENT;
877 cache_purge(dvp);
878 np->n_nctime = 0;
879 goto dorpc;
880 }
881
882 newvp = *vpp;
883 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
884 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime)
885 {
886 nfsstats.lookupcache_hits++;
887 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
888 cnp->cn_flags |= SAVENAME;
889 if ((!lockparent || !(flags & ISLASTCN)) &&
890 newvp != dvp)
891 VOP_UNLOCK(dvp, 0);
892 return (0);
893 }
894 cache_purge(newvp);
895 if (newvp != dvp)
896 vput(newvp);
897 else
898 vrele(newvp);
899 *vpp = NULLVP;
900 }
901 dorpc:
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, cnp->cn_proc, 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 && wantparent && (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 if (v3) {
934 nfsm_postop_attr(newvp, attrflag, 0);
935 nfsm_postop_attr(dvp, attrflag, 0);
936 } else
937 nfsm_loadattr(newvp, (struct vattr *)0, 0);
938 *vpp = newvp;
939 m_freem(mrep);
940 cnp->cn_flags |= SAVENAME;
941 if (!lockparent) {
942 VOP_UNLOCK(dvp, 0);
943 cnp->cn_flags |= PDIRUNLOCK;
944 }
945 return (0);
946 }
947
948 /*
949 * The postop attr handling is duplicated for each if case,
950 * because it should be done while dvp is locked (unlocking
951 * dvp is different for each case).
952 */
953
954 if (NFS_CMPFH(np, fhp, fhsize)) {
955 /*
956 * "." lookup
957 */
958 VREF(dvp);
959 newvp = dvp;
960 if (v3) {
961 nfsm_postop_attr(newvp, attrflag, 0);
962 nfsm_postop_attr(dvp, attrflag, 0);
963 } else
964 nfsm_loadattr(newvp, (struct vattr *)0, 0);
965 } else if (flags & ISDOTDOT) {
966 /*
967 * ".." lookup
968 */
969 VOP_UNLOCK(dvp, 0);
970 cnp->cn_flags |= PDIRUNLOCK;
971
972 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
973 if (error) {
974 if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
975 cnp->cn_flags &= ~PDIRUNLOCK;
976 m_freem(mrep);
977 return error;
978 }
979 newvp = NFSTOV(np);
980
981 if (v3) {
982 nfsm_postop_attr(newvp, attrflag, 0);
983 nfsm_postop_attr(dvp, attrflag, 0);
984 } else
985 nfsm_loadattr(newvp, (struct vattr *)0, 0);
986
987 if (lockparent && (flags & ISLASTCN)) {
988 if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
989 m_freem(mrep);
990 vput(newvp);
991 return error;
992 }
993 cnp->cn_flags &= ~PDIRUNLOCK;
994 }
995 } else {
996 /*
997 * Other lookups.
998 */
999 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1000 if (error) {
1001 m_freem(mrep);
1002 return error;
1003 }
1004 newvp = NFSTOV(np);
1005 if (v3) {
1006 nfsm_postop_attr(newvp, attrflag, 0);
1007 nfsm_postop_attr(dvp, attrflag, 0);
1008 } else
1009 nfsm_loadattr(newvp, (struct vattr *)0, 0);
1010 if (!lockparent || !(flags & ISLASTCN)) {
1011 VOP_UNLOCK(dvp, 0);
1012 cnp->cn_flags |= PDIRUNLOCK;
1013 }
1014 }
1015 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1016 cnp->cn_flags |= SAVENAME;
1017 if ((cnp->cn_flags & MAKEENTRY) &&
1018 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1019 np->n_ctime = np->n_vattr->va_ctime.tv_sec;
1020 cache_enter(dvp, newvp, cnp);
1021 }
1022 *vpp = newvp;
1023 nfsm_reqdone;
1024 if (error) {
1025 /*
1026 * We get here only because of errors returned by
1027 * the RPC. Otherwise we'll have returned above
1028 * (the nfsm_* macros will jump to nfsm_reqdone
1029 * on error).
1030 */
1031 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1032 cnp->cn_nameiop != CREATE) {
1033 if (VTONFS(dvp)->n_nctime == 0)
1034 VTONFS(dvp)->n_nctime =
1035 VTONFS(dvp)->n_vattr->va_mtime.tv_sec;
1036 cache_enter(dvp, NULL, cnp);
1037 }
1038 if (newvp != NULLVP) {
1039 vrele(newvp);
1040 if (newvp != dvp)
1041 VOP_UNLOCK(newvp, 0);
1042 }
1043 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1044 (flags & ISLASTCN) && error == ENOENT) {
1045 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1046 error = EROFS;
1047 else
1048 error = EJUSTRETURN;
1049 }
1050 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1051 cnp->cn_flags |= SAVENAME;
1052 *vpp = NULL;
1053 }
1054
1055 return error;
1056 }
1057
1058 /*
1059 * nfs read call.
1060 * Just call nfs_bioread() to do the work.
1061 */
1062 int
1063 nfs_read(v)
1064 void *v;
1065 {
1066 struct vop_read_args /* {
1067 struct vnode *a_vp;
1068 struct uio *a_uio;
1069 int a_ioflag;
1070 struct ucred *a_cred;
1071 } */ *ap = v;
1072 struct vnode *vp = ap->a_vp;
1073
1074 if (vp->v_type != VREG)
1075 return (EPERM);
1076 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1077 }
1078
1079 /*
1080 * nfs readlink call
1081 */
1082 int
1083 nfs_readlink(v)
1084 void *v;
1085 {
1086 struct vop_readlink_args /* {
1087 struct vnode *a_vp;
1088 struct uio *a_uio;
1089 struct ucred *a_cred;
1090 } */ *ap = v;
1091 struct vnode *vp = ap->a_vp;
1092
1093 if (vp->v_type != VLNK)
1094 return (EPERM);
1095 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1096 }
1097
1098 /*
1099 * Do a readlink rpc.
1100 * Called by nfs_doio() from below the buffer cache.
1101 */
1102 int
1103 nfs_readlinkrpc(vp, uiop, cred)
1104 struct vnode *vp;
1105 struct uio *uiop;
1106 struct ucred *cred;
1107 {
1108 u_int32_t *tl;
1109 caddr_t cp;
1110 int32_t t1, t2;
1111 caddr_t bpos, dpos, cp2;
1112 int error = 0, attrflag;
1113 uint32_t len;
1114 struct mbuf *mreq, *mrep, *md, *mb;
1115 const int v3 = NFS_ISV3(vp);
1116 struct nfsnode *np = VTONFS(vp);
1117
1118 nfsstats.rpccnt[NFSPROC_READLINK]++;
1119 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1120 nfsm_fhtom(np, v3);
1121 nfsm_request(np, NFSPROC_READLINK, uiop->uio_procp, cred);
1122 if (v3)
1123 nfsm_postop_attr(vp, attrflag, 0);
1124 if (!error) {
1125 if (v3) {
1126 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1127 len = fxdr_unsigned(uint32_t, *tl);
1128 if (len > MAXPATHLEN) {
1129 /*
1130 * this pathname is too long for us.
1131 */
1132 m_freem(mrep);
1133 /* Solaris returns EINVAL. should we follow? */
1134 error = ENAMETOOLONG;
1135 goto nfsmout;
1136 }
1137 } else {
1138 nfsm_strsiz(len, NFS_MAXPATHLEN);
1139 }
1140 nfsm_mtouio(uiop, len);
1141 }
1142 nfsm_reqdone;
1143 return (error);
1144 }
1145
1146 /*
1147 * nfs read rpc call
1148 * Ditto above
1149 */
1150 int
1151 nfs_readrpc(vp, uiop)
1152 struct vnode *vp;
1153 struct uio *uiop;
1154 {
1155 u_int32_t *tl;
1156 caddr_t cp;
1157 int32_t t1, t2;
1158 caddr_t bpos, dpos, cp2;
1159 struct mbuf *mreq, *mrep, *md, *mb;
1160 struct nfsmount *nmp;
1161 int error = 0, len, retlen, tsiz, eof, attrflag;
1162 const int v3 = NFS_ISV3(vp);
1163 struct nfsnode *np = VTONFS(vp);
1164
1165 #ifndef nolint
1166 eof = 0;
1167 #endif
1168 nmp = VFSTONFS(vp->v_mount);
1169 tsiz = uiop->uio_resid;
1170 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1171 return (EFBIG);
1172 while (tsiz > 0) {
1173 nfsstats.rpccnt[NFSPROC_READ]++;
1174 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1175 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1176 nfsm_fhtom(np, v3);
1177 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1178 if (v3) {
1179 txdr_hyper(uiop->uio_offset, tl);
1180 *(tl + 2) = txdr_unsigned(len);
1181 } else {
1182 *tl++ = txdr_unsigned(uiop->uio_offset);
1183 *tl++ = txdr_unsigned(len);
1184 *tl = 0;
1185 }
1186 nfsm_request(np, NFSPROC_READ, uiop->uio_procp,
1187 VTONFS(vp)->n_rcred);
1188 if (v3) {
1189 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1190 if (error) {
1191 m_freem(mrep);
1192 goto nfsmout;
1193 }
1194 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1195 eof = fxdr_unsigned(int, *(tl + 1));
1196 } else
1197 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1198 nfsm_strsiz(retlen, nmp->nm_rsize);
1199 nfsm_mtouio(uiop, retlen);
1200 m_freem(mrep);
1201 tsiz -= retlen;
1202 if (v3) {
1203 if (eof || retlen == 0)
1204 tsiz = 0;
1205 } else if (retlen < len)
1206 tsiz = 0;
1207 }
1208 nfsmout:
1209 return (error);
1210 }
1211
1212 struct nfs_writerpc_context {
1213 struct simplelock nwc_slock;
1214 volatile int nwc_mbufcount;
1215 };
1216
1217 /*
1218 * free mbuf used to refer protected pages while write rpc call.
1219 * called at splvm.
1220 */
1221 static void
1222 nfs_writerpc_extfree(struct mbuf *m, caddr_t buf, size_t size, void *arg)
1223 {
1224 struct nfs_writerpc_context *ctx = arg;
1225
1226 KASSERT(m != NULL);
1227 KASSERT(ctx != NULL);
1228 pool_cache_put(&mbpool_cache, m);
1229 simple_lock(&ctx->nwc_slock);
1230 if (--ctx->nwc_mbufcount == 0) {
1231 wakeup(ctx);
1232 }
1233 simple_unlock(&ctx->nwc_slock);
1234 }
1235
1236 /*
1237 * nfs write call
1238 */
1239 int
1240 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverf)
1241 struct vnode *vp;
1242 struct uio *uiop;
1243 int *iomode;
1244 boolean_t pageprotected;
1245 boolean_t *stalewriteverf;
1246 {
1247 u_int32_t *tl;
1248 caddr_t cp;
1249 int32_t t1, t2, backup;
1250 caddr_t bpos, dpos, cp2;
1251 struct mbuf *mreq, *mrep, *md, *mb;
1252 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1253 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1254 const int v3 = NFS_ISV3(vp);
1255 int committed = NFSV3WRITE_FILESYNC;
1256 struct nfsnode *np = VTONFS(vp);
1257 struct nfs_writerpc_context ctx;
1258 int s;
1259 struct lwp *l;
1260
1261 simple_lock_init(&ctx.nwc_slock);
1262 ctx.nwc_mbufcount = 1;
1263
1264 if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1265 panic("writerpc readonly vp %p", vp);
1266 }
1267
1268 #ifdef DIAGNOSTIC
1269 if (uiop->uio_iovcnt != 1)
1270 panic("nfs: writerpc iovcnt > 1");
1271 #endif
1272 tsiz = uiop->uio_resid;
1273 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1274 return (EFBIG);
1275 if (pageprotected) {
1276 l = curlwp;
1277 PHOLD(l);
1278 }
1279 while (tsiz > 0) {
1280 int datalen;
1281
1282 nfsstats.rpccnt[NFSPROC_WRITE]++;
1283 len = min(tsiz, nmp->nm_wsize);
1284 datalen = pageprotected ? 0 : nfsm_rndup(len);
1285 nfsm_reqhead(np, NFSPROC_WRITE,
1286 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1287 nfsm_fhtom(np, v3);
1288 if (v3) {
1289 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1290 txdr_hyper(uiop->uio_offset, tl);
1291 tl += 2;
1292 *tl++ = txdr_unsigned(len);
1293 *tl++ = txdr_unsigned(*iomode);
1294 *tl = txdr_unsigned(len);
1295 } else {
1296 u_int32_t x;
1297
1298 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1299 /* Set both "begin" and "current" to non-garbage. */
1300 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1301 *tl++ = x; /* "begin offset" */
1302 *tl++ = x; /* "current offset" */
1303 x = txdr_unsigned(len);
1304 *tl++ = x; /* total to this offset */
1305 *tl = x; /* size of this write */
1306
1307 }
1308 if (pageprotected) {
1309 /*
1310 * since we know pages can't be modified during i/o,
1311 * no need to copy them for us.
1312 */
1313 struct mbuf *m;
1314 struct iovec *iovp = uiop->uio_iov;
1315
1316 m = m_get(M_WAIT, MT_DATA);
1317 MCLAIM(m, &nfs_mowner);
1318 MEXTADD(m, iovp->iov_base, len, M_MBUF,
1319 nfs_writerpc_extfree, &ctx);
1320 m->m_flags |= M_EXT_ROMAP;
1321 m->m_len = len;
1322 mb->m_next = m;
1323 /*
1324 * no need to maintain mb and bpos here
1325 * because no one care them later.
1326 */
1327 #if 0
1328 mb = m;
1329 bpos = mtod(caddr_t, mb) + mb->m_len;
1330 #endif
1331 iovp->iov_base = (char *)iovp->iov_base + len;
1332 iovp->iov_len -= len;
1333 uiop->uio_offset += len;
1334 uiop->uio_resid -= len;
1335 s = splvm();
1336 simple_lock(&ctx.nwc_slock);
1337 ctx.nwc_mbufcount++;
1338 simple_unlock(&ctx.nwc_slock);
1339 splx(s);
1340 } else {
1341 nfsm_uiotom(uiop, len);
1342 }
1343 nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp,
1344 VTONFS(vp)->n_wcred);
1345 if (v3) {
1346 wccflag = NFSV3_WCCCHK;
1347 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC);
1348 if (!error) {
1349 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1350 + NFSX_V3WRITEVERF);
1351 rlen = fxdr_unsigned(int, *tl++);
1352 if (rlen == 0) {
1353 error = NFSERR_IO;
1354 m_freem(mrep);
1355 break;
1356 } else if (rlen < len) {
1357 backup = len - rlen;
1358 uiop->uio_iov->iov_base =
1359 (caddr_t)uiop->uio_iov->iov_base -
1360 backup;
1361 uiop->uio_iov->iov_len += backup;
1362 uiop->uio_offset -= backup;
1363 uiop->uio_resid += backup;
1364 len = rlen;
1365 }
1366 commit = fxdr_unsigned(int, *tl++);
1367
1368 /*
1369 * Return the lowest committment level
1370 * obtained by any of the RPCs.
1371 */
1372 if (committed == NFSV3WRITE_FILESYNC)
1373 committed = commit;
1374 else if (committed == NFSV3WRITE_DATASYNC &&
1375 commit == NFSV3WRITE_UNSTABLE)
1376 committed = commit;
1377 simple_lock(&nmp->nm_slock);
1378 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1379 memcpy(nmp->nm_writeverf, tl,
1380 NFSX_V3WRITEVERF);
1381 nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1382 } else if ((nmp->nm_iflag &
1383 NFSMNT_STALEWRITEVERF) ||
1384 memcmp(tl, nmp->nm_writeverf,
1385 NFSX_V3WRITEVERF)) {
1386 *stalewriteverf = TRUE;
1387 memcpy(nmp->nm_writeverf, tl,
1388 NFSX_V3WRITEVERF);
1389 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
1390 }
1391 simple_unlock(&nmp->nm_slock);
1392 }
1393 } else
1394 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1395 if (wccflag)
1396 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime.tv_sec;
1397 m_freem(mrep);
1398 if (error)
1399 break;
1400 tsiz -= len;
1401 }
1402 if (pageprotected) {
1403 /*
1404 * wait until mbufs go away.
1405 * retransmitted mbufs can survive longer than rpc requests
1406 * themselves.
1407 */
1408 s = splvm();
1409 simple_lock(&ctx.nwc_slock);
1410 ctx.nwc_mbufcount--;
1411 while (ctx.nwc_mbufcount > 0) {
1412 ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock);
1413 }
1414 simple_unlock(&ctx.nwc_slock);
1415 splx(s);
1416 PRELE(l);
1417 }
1418 nfsmout:
1419 *iomode = committed;
1420 if (error)
1421 uiop->uio_resid = tsiz;
1422 return (error);
1423 }
1424
1425 /*
1426 * nfs mknod rpc
1427 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1428 * mode set to specify the file type and the size field for rdev.
1429 */
1430 int
1431 nfs_mknodrpc(dvp, vpp, cnp, vap)
1432 struct vnode *dvp;
1433 struct vnode **vpp;
1434 struct componentname *cnp;
1435 struct vattr *vap;
1436 {
1437 struct nfsv2_sattr *sp;
1438 u_int32_t *tl;
1439 caddr_t cp;
1440 int32_t t1, t2;
1441 struct vnode *newvp = (struct vnode *)0;
1442 struct nfsnode *dnp, *np;
1443 char *cp2;
1444 caddr_t bpos, dpos;
1445 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1446 struct mbuf *mreq, *mrep, *md, *mb;
1447 u_int32_t rdev;
1448 const int v3 = NFS_ISV3(dvp);
1449
1450 if (vap->va_type == VCHR || vap->va_type == VBLK)
1451 rdev = txdr_unsigned(vap->va_rdev);
1452 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1453 rdev = nfs_xdrneg1;
1454 else {
1455 VOP_ABORTOP(dvp, cnp);
1456 vput(dvp);
1457 return (EOPNOTSUPP);
1458 }
1459 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1460 dnp = VTONFS(dvp);
1461 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1462 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1463 nfsm_fhtom(dnp, v3);
1464 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1465 if (v3) {
1466 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1467 *tl++ = vtonfsv3_type(vap->va_type);
1468 nfsm_v3attrbuild(vap, FALSE);
1469 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1470 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1471 *tl++ = txdr_unsigned(major(vap->va_rdev));
1472 *tl = txdr_unsigned(minor(vap->va_rdev));
1473 }
1474 } else {
1475 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1476 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1477 sp->sa_uid = nfs_xdrneg1;
1478 sp->sa_gid = nfs_xdrneg1;
1479 sp->sa_size = rdev;
1480 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1481 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1482 }
1483 nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1484 if (!error) {
1485 nfsm_mtofh(dvp, newvp, v3, gotvp);
1486 if (!gotvp) {
1487 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1488 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1489 if (!error)
1490 newvp = NFSTOV(np);
1491 }
1492 }
1493 if (v3)
1494 nfsm_wcc_data(dvp, wccflag, 0);
1495 nfsm_reqdone;
1496 if (error) {
1497 if (newvp)
1498 vput(newvp);
1499 } else {
1500 if (cnp->cn_flags & MAKEENTRY)
1501 cache_enter(dvp, newvp, cnp);
1502 *vpp = newvp;
1503 }
1504 PNBUF_PUT(cnp->cn_pnbuf);
1505 VTONFS(dvp)->n_flag |= NMODIFIED;
1506 if (!wccflag)
1507 VTONFS(dvp)->n_attrstamp = 0;
1508 vput(dvp);
1509 return (error);
1510 }
1511
1512 /*
1513 * nfs mknod vop
1514 * just call nfs_mknodrpc() to do the work.
1515 */
1516 /* ARGSUSED */
1517 int
1518 nfs_mknod(v)
1519 void *v;
1520 {
1521 struct vop_mknod_args /* {
1522 struct vnode *a_dvp;
1523 struct vnode **a_vpp;
1524 struct componentname *a_cnp;
1525 struct vattr *a_vap;
1526 } */ *ap = v;
1527 int error;
1528
1529 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1530 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1531 return (error);
1532 }
1533
1534 static u_long create_verf;
1535 /*
1536 * nfs file create call
1537 */
1538 int
1539 nfs_create(v)
1540 void *v;
1541 {
1542 struct vop_create_args /* {
1543 struct vnode *a_dvp;
1544 struct vnode **a_vpp;
1545 struct componentname *a_cnp;
1546 struct vattr *a_vap;
1547 } */ *ap = v;
1548 struct vnode *dvp = ap->a_dvp;
1549 struct vattr *vap = ap->a_vap;
1550 struct componentname *cnp = ap->a_cnp;
1551 struct nfsv2_sattr *sp;
1552 u_int32_t *tl;
1553 caddr_t cp;
1554 int32_t t1, t2;
1555 struct nfsnode *dnp, *np = (struct nfsnode *)0;
1556 struct vnode *newvp = (struct vnode *)0;
1557 caddr_t bpos, dpos, cp2;
1558 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1559 struct mbuf *mreq, *mrep, *md, *mb;
1560 const int v3 = NFS_ISV3(dvp);
1561
1562 /*
1563 * Oops, not for me..
1564 */
1565 if (vap->va_type == VSOCK)
1566 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1567
1568 #ifdef VA_EXCLUSIVE
1569 if (vap->va_vaflags & VA_EXCLUSIVE)
1570 fmode |= O_EXCL;
1571 #endif
1572 again:
1573 error = 0;
1574 nfsstats.rpccnt[NFSPROC_CREATE]++;
1575 dnp = VTONFS(dvp);
1576 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1577 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1578 nfsm_fhtom(dnp, v3);
1579 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1580 if (v3) {
1581 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1582 if (fmode & O_EXCL) {
1583 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1584 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1585 #ifdef INET
1586 if (TAILQ_FIRST(&in_ifaddr))
1587 *tl++ = TAILQ_FIRST(&in_ifaddr)->ia_addr.sin_addr.s_addr;
1588 else
1589 *tl++ = create_verf;
1590 #else
1591 *tl++ = create_verf;
1592 #endif
1593 *tl = ++create_verf;
1594 } else {
1595 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1596 nfsm_v3attrbuild(vap, FALSE);
1597 }
1598 } else {
1599 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1600 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1601 sp->sa_uid = nfs_xdrneg1;
1602 sp->sa_gid = nfs_xdrneg1;
1603 sp->sa_size = 0;
1604 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1605 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1606 }
1607 nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1608 if (!error) {
1609 nfsm_mtofh(dvp, newvp, v3, gotvp);
1610 if (!gotvp) {
1611 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1612 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1613 if (!error)
1614 newvp = NFSTOV(np);
1615 }
1616 }
1617 if (v3)
1618 nfsm_wcc_data(dvp, wccflag, 0);
1619 nfsm_reqdone;
1620 if (error) {
1621 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1622 fmode &= ~O_EXCL;
1623 goto again;
1624 }
1625 if (newvp)
1626 vput(newvp);
1627 } else if (v3 && (fmode & O_EXCL))
1628 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1629 if (!error) {
1630 if (cnp->cn_flags & MAKEENTRY)
1631 cache_enter(dvp, newvp, cnp);
1632 *ap->a_vpp = newvp;
1633 }
1634 PNBUF_PUT(cnp->cn_pnbuf);
1635 VTONFS(dvp)->n_flag |= NMODIFIED;
1636 if (!wccflag)
1637 VTONFS(dvp)->n_attrstamp = 0;
1638 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1639 vput(dvp);
1640 return (error);
1641 }
1642
1643 /*
1644 * nfs file remove call
1645 * To try and make nfs semantics closer to ufs semantics, a file that has
1646 * other processes using the vnode is renamed instead of removed and then
1647 * removed later on the last close.
1648 * - If v_usecount > 1
1649 * If a rename is not already in the works
1650 * call nfs_sillyrename() to set it up
1651 * else
1652 * do the remove rpc
1653 */
1654 int
1655 nfs_remove(v)
1656 void *v;
1657 {
1658 struct vop_remove_args /* {
1659 struct vnodeop_desc *a_desc;
1660 struct vnode * a_dvp;
1661 struct vnode * a_vp;
1662 struct componentname * a_cnp;
1663 } */ *ap = v;
1664 struct vnode *vp = ap->a_vp;
1665 struct vnode *dvp = ap->a_dvp;
1666 struct componentname *cnp = ap->a_cnp;
1667 struct nfsnode *np = VTONFS(vp);
1668 int error = 0;
1669 struct vattr vattr;
1670
1671 #ifndef DIAGNOSTIC
1672 if ((cnp->cn_flags & HASBUF) == 0)
1673 panic("nfs_remove: no name");
1674 if (vp->v_usecount < 1)
1675 panic("nfs_remove: bad v_usecount");
1676 #endif
1677 if (vp->v_type == VDIR)
1678 error = EPERM;
1679 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1680 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1681 vattr.va_nlink > 1)) {
1682 /*
1683 * Purge the name cache so that the chance of a lookup for
1684 * the name succeeding while the remove is in progress is
1685 * minimized. Without node locking it can still happen, such
1686 * that an I/O op returns ESTALE, but since you get this if
1687 * another host removes the file..
1688 */
1689 cache_purge(vp);
1690 /*
1691 * throw away biocache buffers, mainly to avoid
1692 * unnecessary delayed writes later.
1693 */
1694 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1695 /* Do the rpc */
1696 if (error != EINTR)
1697 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1698 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1699 /*
1700 * Kludge City: If the first reply to the remove rpc is lost..
1701 * the reply to the retransmitted request will be ENOENT
1702 * since the file was in fact removed
1703 * Therefore, we cheat and return success.
1704 */
1705 if (error == ENOENT)
1706 error = 0;
1707 } else if (!np->n_sillyrename)
1708 error = nfs_sillyrename(dvp, vp, cnp);
1709 PNBUF_PUT(cnp->cn_pnbuf);
1710 np->n_attrstamp = 0;
1711 VN_KNOTE(vp, NOTE_DELETE);
1712 VN_KNOTE(dvp, NOTE_WRITE);
1713 if (dvp == vp)
1714 vrele(vp);
1715 else
1716 vput(vp);
1717 vput(dvp);
1718 return (error);
1719 }
1720
1721 /*
1722 * nfs file remove rpc called from nfs_inactive
1723 */
1724 int
1725 nfs_removeit(sp)
1726 struct sillyrename *sp;
1727 {
1728
1729 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1730 (struct proc *)0));
1731 }
1732
1733 /*
1734 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1735 */
1736 int
1737 nfs_removerpc(dvp, name, namelen, cred, proc)
1738 struct vnode *dvp;
1739 const char *name;
1740 int namelen;
1741 struct ucred *cred;
1742 struct proc *proc;
1743 {
1744 u_int32_t *tl;
1745 caddr_t cp;
1746 int32_t t1, t2;
1747 caddr_t bpos, dpos, cp2;
1748 int error = 0, wccflag = NFSV3_WCCRATTR;
1749 struct mbuf *mreq, *mrep, *md, *mb;
1750 const int v3 = NFS_ISV3(dvp);
1751 struct nfsnode *dnp = VTONFS(dvp);
1752
1753 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1754 nfsm_reqhead(dnp, NFSPROC_REMOVE,
1755 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1756 nfsm_fhtom(dnp, v3);
1757 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1758 nfsm_request(dnp, NFSPROC_REMOVE, proc, cred);
1759 if (v3)
1760 nfsm_wcc_data(dvp, wccflag, 0);
1761 nfsm_reqdone;
1762 VTONFS(dvp)->n_flag |= NMODIFIED;
1763 if (!wccflag)
1764 VTONFS(dvp)->n_attrstamp = 0;
1765 return (error);
1766 }
1767
1768 /*
1769 * nfs file rename call
1770 */
1771 int
1772 nfs_rename(v)
1773 void *v;
1774 {
1775 struct vop_rename_args /* {
1776 struct vnode *a_fdvp;
1777 struct vnode *a_fvp;
1778 struct componentname *a_fcnp;
1779 struct vnode *a_tdvp;
1780 struct vnode *a_tvp;
1781 struct componentname *a_tcnp;
1782 } */ *ap = v;
1783 struct vnode *fvp = ap->a_fvp;
1784 struct vnode *tvp = ap->a_tvp;
1785 struct vnode *fdvp = ap->a_fdvp;
1786 struct vnode *tdvp = ap->a_tdvp;
1787 struct componentname *tcnp = ap->a_tcnp;
1788 struct componentname *fcnp = ap->a_fcnp;
1789 int error;
1790
1791 #ifndef DIAGNOSTIC
1792 if ((tcnp->cn_flags & HASBUF) == 0 ||
1793 (fcnp->cn_flags & HASBUF) == 0)
1794 panic("nfs_rename: no name");
1795 #endif
1796 /* Check for cross-device rename */
1797 if ((fvp->v_mount != tdvp->v_mount) ||
1798 (tvp && (fvp->v_mount != tvp->v_mount))) {
1799 error = EXDEV;
1800 goto out;
1801 }
1802
1803 /*
1804 * If the tvp exists and is in use, sillyrename it before doing the
1805 * rename of the new file over it.
1806 */
1807 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1808 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1809 VN_KNOTE(tvp, NOTE_DELETE);
1810 vput(tvp);
1811 tvp = NULL;
1812 }
1813
1814 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1815 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1816 tcnp->cn_proc);
1817
1818 VN_KNOTE(fdvp, NOTE_WRITE);
1819 VN_KNOTE(tdvp, NOTE_WRITE);
1820 if (fvp->v_type == VDIR) {
1821 if (tvp != NULL && tvp->v_type == VDIR)
1822 cache_purge(tdvp);
1823 cache_purge(fdvp);
1824 }
1825 out:
1826 if (tdvp == tvp)
1827 vrele(tdvp);
1828 else
1829 vput(tdvp);
1830 if (tvp)
1831 vput(tvp);
1832 vrele(fdvp);
1833 vrele(fvp);
1834 /*
1835 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1836 */
1837 if (error == ENOENT)
1838 error = 0;
1839 return (error);
1840 }
1841
1842 /*
1843 * nfs file rename rpc called from nfs_remove() above
1844 */
1845 int
1846 nfs_renameit(sdvp, scnp, sp)
1847 struct vnode *sdvp;
1848 struct componentname *scnp;
1849 struct sillyrename *sp;
1850 {
1851 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1852 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1853 }
1854
1855 /*
1856 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1857 */
1858 int
1859 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1860 struct vnode *fdvp;
1861 const char *fnameptr;
1862 int fnamelen;
1863 struct vnode *tdvp;
1864 const char *tnameptr;
1865 int tnamelen;
1866 struct ucred *cred;
1867 struct proc *proc;
1868 {
1869 u_int32_t *tl;
1870 caddr_t cp;
1871 int32_t t1, t2;
1872 caddr_t bpos, dpos, cp2;
1873 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1874 struct mbuf *mreq, *mrep, *md, *mb;
1875 const int v3 = NFS_ISV3(fdvp);
1876 struct nfsnode *fdnp = VTONFS(fdvp);
1877
1878 nfsstats.rpccnt[NFSPROC_RENAME]++;
1879 nfsm_reqhead(fdnp, NFSPROC_RENAME,
1880 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1881 nfsm_rndup(tnamelen));
1882 nfsm_fhtom(fdnp, v3);
1883 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1884 nfsm_fhtom(VTONFS(tdvp), v3);
1885 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1886 nfsm_request(fdnp, NFSPROC_RENAME, proc, cred);
1887 if (v3) {
1888 nfsm_wcc_data(fdvp, fwccflag, 0);
1889 nfsm_wcc_data(tdvp, twccflag, 0);
1890 }
1891 nfsm_reqdone;
1892 VTONFS(fdvp)->n_flag |= NMODIFIED;
1893 VTONFS(tdvp)->n_flag |= NMODIFIED;
1894 if (!fwccflag)
1895 VTONFS(fdvp)->n_attrstamp = 0;
1896 if (!twccflag)
1897 VTONFS(tdvp)->n_attrstamp = 0;
1898 return (error);
1899 }
1900
1901 /*
1902 * nfs hard link create call
1903 */
1904 int
1905 nfs_link(v)
1906 void *v;
1907 {
1908 struct vop_link_args /* {
1909 struct vnode *a_dvp;
1910 struct vnode *a_vp;
1911 struct componentname *a_cnp;
1912 } */ *ap = v;
1913 struct vnode *vp = ap->a_vp;
1914 struct vnode *dvp = ap->a_dvp;
1915 struct componentname *cnp = ap->a_cnp;
1916 u_int32_t *tl;
1917 caddr_t cp;
1918 int32_t t1, t2;
1919 caddr_t bpos, dpos, cp2;
1920 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1921 struct mbuf *mreq, *mrep, *md, *mb;
1922 /* XXX Should be const and initialised? */
1923 int v3;
1924 struct nfsnode *np;
1925
1926 if (dvp->v_mount != vp->v_mount) {
1927 VOP_ABORTOP(dvp, cnp);
1928 vput(dvp);
1929 return (EXDEV);
1930 }
1931 if (dvp != vp) {
1932 error = vn_lock(vp, LK_EXCLUSIVE);
1933 if (error != 0) {
1934 VOP_ABORTOP(dvp, cnp);
1935 vput(dvp);
1936 return error;
1937 }
1938 }
1939
1940 /*
1941 * Push all writes to the server, so that the attribute cache
1942 * doesn't get "out of sync" with the server.
1943 * XXX There should be a better way!
1944 */
1945 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc);
1946
1947 v3 = NFS_ISV3(vp);
1948 nfsstats.rpccnt[NFSPROC_LINK]++;
1949 np = VTONFS(vp);
1950 nfsm_reqhead(np, NFSPROC_LINK,
1951 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1952 nfsm_fhtom(np, v3);
1953 nfsm_fhtom(VTONFS(dvp), v3);
1954 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1955 nfsm_request(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1956 if (v3) {
1957 nfsm_postop_attr(vp, attrflag, 0);
1958 nfsm_wcc_data(dvp, wccflag, 0);
1959 }
1960 nfsm_reqdone;
1961 PNBUF_PUT(cnp->cn_pnbuf);
1962 VTONFS(dvp)->n_flag |= NMODIFIED;
1963 if (!attrflag)
1964 VTONFS(vp)->n_attrstamp = 0;
1965 if (!wccflag)
1966 VTONFS(dvp)->n_attrstamp = 0;
1967 if (dvp != vp)
1968 VOP_UNLOCK(vp, 0);
1969 VN_KNOTE(vp, NOTE_LINK);
1970 VN_KNOTE(dvp, NOTE_WRITE);
1971 vput(dvp);
1972 /*
1973 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1974 */
1975 if (error == EEXIST)
1976 error = 0;
1977 return (error);
1978 }
1979
1980 /*
1981 * nfs symbolic link create call
1982 */
1983 int
1984 nfs_symlink(v)
1985 void *v;
1986 {
1987 struct vop_symlink_args /* {
1988 struct vnode *a_dvp;
1989 struct vnode **a_vpp;
1990 struct componentname *a_cnp;
1991 struct vattr *a_vap;
1992 char *a_target;
1993 } */ *ap = v;
1994 struct vnode *dvp = ap->a_dvp;
1995 struct vattr *vap = ap->a_vap;
1996 struct componentname *cnp = ap->a_cnp;
1997 struct nfsv2_sattr *sp;
1998 u_int32_t *tl;
1999 caddr_t cp;
2000 int32_t t1, t2;
2001 caddr_t bpos, dpos, cp2;
2002 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2003 struct mbuf *mreq, *mrep, *md, *mb;
2004 struct vnode *newvp = (struct vnode *)0;
2005 const int v3 = NFS_ISV3(dvp);
2006 struct nfsnode *dnp = VTONFS(dvp);
2007
2008 *ap->a_vpp = NULL;
2009 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2010 slen = strlen(ap->a_target);
2011 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2012 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2013 nfsm_fhtom(dnp, v3);
2014 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2015 if (v3)
2016 nfsm_v3attrbuild(vap, FALSE);
2017 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2018 if (!v3) {
2019 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2020 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2021 sp->sa_uid = nfs_xdrneg1;
2022 sp->sa_gid = nfs_xdrneg1;
2023 sp->sa_size = nfs_xdrneg1;
2024 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2025 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2026 }
2027 nfsm_request(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
2028 if (v3) {
2029 if (!error)
2030 nfsm_mtofh(dvp, newvp, v3, gotvp);
2031 nfsm_wcc_data(dvp, wccflag, 0);
2032 }
2033 nfsm_reqdone;
2034 /*
2035 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2036 */
2037 if (error == EEXIST)
2038 error = 0;
2039 if (error == 0 && newvp == NULL) {
2040 struct nfsnode *np = NULL;
2041
2042 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2043 cnp->cn_cred, cnp->cn_proc, &np);
2044 if (error == 0)
2045 newvp = NFSTOV(np);
2046 }
2047 if (error) {
2048 if (newvp != NULL)
2049 vput(newvp);
2050 } else {
2051 *ap->a_vpp = newvp;
2052 }
2053 PNBUF_PUT(cnp->cn_pnbuf);
2054 VTONFS(dvp)->n_flag |= NMODIFIED;
2055 if (!wccflag)
2056 VTONFS(dvp)->n_attrstamp = 0;
2057 VN_KNOTE(dvp, NOTE_WRITE);
2058 vput(dvp);
2059 return (error);
2060 }
2061
2062 /*
2063 * nfs make dir call
2064 */
2065 int
2066 nfs_mkdir(v)
2067 void *v;
2068 {
2069 struct vop_mkdir_args /* {
2070 struct vnode *a_dvp;
2071 struct vnode **a_vpp;
2072 struct componentname *a_cnp;
2073 struct vattr *a_vap;
2074 } */ *ap = v;
2075 struct vnode *dvp = ap->a_dvp;
2076 struct vattr *vap = ap->a_vap;
2077 struct componentname *cnp = ap->a_cnp;
2078 struct nfsv2_sattr *sp;
2079 u_int32_t *tl;
2080 caddr_t cp;
2081 int32_t t1, t2;
2082 int len;
2083 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2084 struct vnode *newvp = (struct vnode *)0;
2085 caddr_t bpos, dpos, cp2;
2086 int error = 0, wccflag = NFSV3_WCCRATTR;
2087 int gotvp = 0;
2088 struct mbuf *mreq, *mrep, *md, *mb;
2089 const int v3 = NFS_ISV3(dvp);
2090
2091 len = cnp->cn_namelen;
2092 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2093 nfsm_reqhead(dnp, NFSPROC_MKDIR,
2094 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2095 nfsm_fhtom(dnp, v3);
2096 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2097 if (v3) {
2098 nfsm_v3attrbuild(vap, FALSE);
2099 } else {
2100 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2101 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2102 sp->sa_uid = nfs_xdrneg1;
2103 sp->sa_gid = nfs_xdrneg1;
2104 sp->sa_size = nfs_xdrneg1;
2105 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2106 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2107 }
2108 nfsm_request(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
2109 if (!error)
2110 nfsm_mtofh(dvp, newvp, v3, gotvp);
2111 if (v3)
2112 nfsm_wcc_data(dvp, wccflag, 0);
2113 nfsm_reqdone;
2114 VTONFS(dvp)->n_flag |= NMODIFIED;
2115 if (!wccflag)
2116 VTONFS(dvp)->n_attrstamp = 0;
2117 /*
2118 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2119 * if we can succeed in looking up the directory.
2120 */
2121 if (error == EEXIST || (!error && !gotvp)) {
2122 if (newvp) {
2123 vput(newvp);
2124 newvp = (struct vnode *)0;
2125 }
2126 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2127 cnp->cn_proc, &np);
2128 if (!error) {
2129 newvp = NFSTOV(np);
2130 if (newvp->v_type != VDIR)
2131 error = EEXIST;
2132 }
2133 }
2134 if (error) {
2135 if (newvp)
2136 vput(newvp);
2137 } else {
2138 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2139 if (cnp->cn_flags & MAKEENTRY)
2140 cache_enter(dvp, newvp, cnp);
2141 *ap->a_vpp = newvp;
2142 }
2143 PNBUF_PUT(cnp->cn_pnbuf);
2144 vput(dvp);
2145 return (error);
2146 }
2147
2148 /*
2149 * nfs remove directory call
2150 */
2151 int
2152 nfs_rmdir(v)
2153 void *v;
2154 {
2155 struct vop_rmdir_args /* {
2156 struct vnode *a_dvp;
2157 struct vnode *a_vp;
2158 struct componentname *a_cnp;
2159 } */ *ap = v;
2160 struct vnode *vp = ap->a_vp;
2161 struct vnode *dvp = ap->a_dvp;
2162 struct componentname *cnp = ap->a_cnp;
2163 u_int32_t *tl;
2164 caddr_t cp;
2165 int32_t t1, t2;
2166 caddr_t bpos, dpos, cp2;
2167 int error = 0, wccflag = NFSV3_WCCRATTR;
2168 struct mbuf *mreq, *mrep, *md, *mb;
2169 const int v3 = NFS_ISV3(dvp);
2170 struct nfsnode *dnp;
2171
2172 if (dvp == vp) {
2173 vrele(dvp);
2174 vput(dvp);
2175 PNBUF_PUT(cnp->cn_pnbuf);
2176 return (EINVAL);
2177 }
2178 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2179 dnp = VTONFS(dvp);
2180 nfsm_reqhead(dnp, NFSPROC_RMDIR,
2181 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2182 nfsm_fhtom(dnp, v3);
2183 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2184 nfsm_request(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
2185 if (v3)
2186 nfsm_wcc_data(dvp, wccflag, 0);
2187 nfsm_reqdone;
2188 PNBUF_PUT(cnp->cn_pnbuf);
2189 VTONFS(dvp)->n_flag |= NMODIFIED;
2190 if (!wccflag)
2191 VTONFS(dvp)->n_attrstamp = 0;
2192 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2193 VN_KNOTE(vp, NOTE_DELETE);
2194 cache_purge(dvp);
2195 cache_purge(vp);
2196 vput(vp);
2197 vput(dvp);
2198 /*
2199 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2200 */
2201 if (error == ENOENT)
2202 error = 0;
2203 return (error);
2204 }
2205
2206 /*
2207 * nfs readdir call
2208 */
2209 int
2210 nfs_readdir(v)
2211 void *v;
2212 {
2213 struct vop_readdir_args /* {
2214 struct vnode *a_vp;
2215 struct uio *a_uio;
2216 struct ucred *a_cred;
2217 int *a_eofflag;
2218 off_t **a_cookies;
2219 int *a_ncookies;
2220 } */ *ap = v;
2221 struct vnode *vp = ap->a_vp;
2222 struct uio *uio = ap->a_uio;
2223 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2224 char *base = uio->uio_iov->iov_base;
2225 int tresid, error;
2226 size_t count, lost;
2227 struct dirent *dp;
2228 off_t *cookies = NULL;
2229 int ncookies = 0, nc;
2230
2231 if (vp->v_type != VDIR)
2232 return (EPERM);
2233
2234 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2235 count = uio->uio_resid - lost;
2236 if (count <= 0)
2237 return (EINVAL);
2238
2239 /*
2240 * Call nfs_bioread() to do the real work.
2241 */
2242 tresid = uio->uio_resid = count;
2243 error = nfs_bioread(vp, uio, 0, ap->a_cred,
2244 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2245
2246 if (!error && ap->a_cookies) {
2247 ncookies = count / 16;
2248 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2249 *ap->a_cookies = cookies;
2250 }
2251
2252 if (!error && uio->uio_resid == tresid) {
2253 uio->uio_resid += lost;
2254 nfsstats.direofcache_misses++;
2255 if (ap->a_cookies)
2256 *ap->a_ncookies = 0;
2257 *ap->a_eofflag = 1;
2258 return (0);
2259 }
2260
2261 if (!error && ap->a_cookies) {
2262 /*
2263 * Only the NFS server and emulations use cookies, and they
2264 * load the directory block into system space, so we can
2265 * just look at it directly.
2266 */
2267 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2268 panic("nfs_readdir: lost in space");
2269 for (nc = 0; ncookies-- &&
2270 base < (char *)uio->uio_iov->iov_base; nc++){
2271 dp = (struct dirent *) base;
2272 if (dp->d_reclen == 0)
2273 break;
2274 if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2275 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2276 else
2277 *(cookies++) = NFS_GETCOOKIE(dp);
2278 base += dp->d_reclen;
2279 }
2280 uio->uio_resid +=
2281 ((caddr_t)uio->uio_iov->iov_base - base);
2282 uio->uio_iov->iov_len +=
2283 ((caddr_t)uio->uio_iov->iov_base - base);
2284 uio->uio_iov->iov_base = base;
2285 *ap->a_ncookies = nc;
2286 }
2287
2288 uio->uio_resid += lost;
2289 *ap->a_eofflag = 0;
2290 return (error);
2291 }
2292
2293 /*
2294 * Readdir rpc call.
2295 * Called from below the buffer cache by nfs_doio().
2296 */
2297 int
2298 nfs_readdirrpc(vp, uiop, cred)
2299 struct vnode *vp;
2300 struct uio *uiop;
2301 struct ucred *cred;
2302 {
2303 int len, left;
2304 struct dirent *dp = NULL;
2305 u_int32_t *tl;
2306 caddr_t cp;
2307 int32_t t1, t2;
2308 caddr_t bpos, dpos, cp2;
2309 struct mbuf *mreq, *mrep, *md, *mb;
2310 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2311 struct nfsnode *dnp = VTONFS(vp);
2312 u_quad_t fileno;
2313 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2314 int attrflag, nrpcs = 0, reclen;
2315 const int v3 = NFS_ISV3(vp);
2316
2317 #ifdef DIAGNOSTIC
2318 /*
2319 * Should be called from buffer cache, so only amount of
2320 * NFS_DIRBLKSIZ will be requested.
2321 */
2322 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2323 panic("nfs readdirrpc bad uio");
2324 #endif
2325
2326 /*
2327 * Loop around doing readdir rpc's of size nm_readdirsize
2328 * truncated to a multiple of NFS_DIRFRAGSIZ.
2329 * The stopping criteria is EOF or buffer full.
2330 */
2331 while (more_dirs && bigenough) {
2332 /*
2333 * Heuristic: don't bother to do another RPC to further
2334 * fill up this block if there is not much room left. (< 50%
2335 * of the readdir RPC size). This wastes some buffer space
2336 * but can save up to 50% in RPC calls.
2337 */
2338 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2339 bigenough = 0;
2340 break;
2341 }
2342 nfsstats.rpccnt[NFSPROC_READDIR]++;
2343 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2344 NFSX_READDIR(v3));
2345 nfsm_fhtom(dnp, v3);
2346 if (v3) {
2347 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2348 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2349 txdr_swapcookie3(uiop->uio_offset, tl);
2350 } else {
2351 txdr_cookie3(uiop->uio_offset, tl);
2352 }
2353 tl += 2;
2354 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2355 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2356 } else {
2357 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2358 *tl++ = txdr_unsigned(uiop->uio_offset);
2359 }
2360 *tl = txdr_unsigned(nmp->nm_readdirsize);
2361 nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred);
2362 nrpcs++;
2363 if (v3) {
2364 nfsm_postop_attr(vp, attrflag, 0);
2365 if (!error) {
2366 nfsm_dissect(tl, u_int32_t *,
2367 2 * NFSX_UNSIGNED);
2368 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2369 dnp->n_cookieverf.nfsuquad[1] = *tl;
2370 } else {
2371 m_freem(mrep);
2372 goto nfsmout;
2373 }
2374 }
2375 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2376 more_dirs = fxdr_unsigned(int, *tl);
2377
2378 /* loop thru the dir entries, doctoring them to 4bsd form */
2379 while (more_dirs && bigenough) {
2380 if (v3) {
2381 nfsm_dissect(tl, u_int32_t *,
2382 3 * NFSX_UNSIGNED);
2383 fileno = fxdr_hyper(tl);
2384 len = fxdr_unsigned(int, *(tl + 2));
2385 } else {
2386 nfsm_dissect(tl, u_int32_t *,
2387 2 * NFSX_UNSIGNED);
2388 fileno = fxdr_unsigned(u_quad_t, *tl++);
2389 len = fxdr_unsigned(int, *tl);
2390 }
2391 if (len <= 0 || len > NFS_MAXNAMLEN) {
2392 error = EBADRPC;
2393 m_freem(mrep);
2394 goto nfsmout;
2395 }
2396 tlen = nfsm_rndup(len);
2397 if (tlen == len)
2398 tlen += 4; /* To ensure null termination */
2399 tlen += sizeof (off_t) + sizeof (int);
2400 reclen = ALIGN(tlen + DIRHDSIZ);
2401 tlen = reclen - DIRHDSIZ;
2402 left = NFS_DIRFRAGSIZ - blksiz;
2403 if (reclen > left) {
2404 dp->d_reclen += left;
2405 uiop->uio_iov->iov_base =
2406 (caddr_t)uiop->uio_iov->iov_base + left;
2407 uiop->uio_iov->iov_len -= left;
2408 uiop->uio_resid -= left;
2409 blksiz = 0;
2410 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2411 }
2412 if (reclen > uiop->uio_resid)
2413 bigenough = 0;
2414 if (bigenough) {
2415 dp = (struct dirent *)uiop->uio_iov->iov_base;
2416 dp->d_fileno = (int)fileno;
2417 dp->d_namlen = len;
2418 dp->d_reclen = reclen;
2419 dp->d_type = DT_UNKNOWN;
2420 blksiz += dp->d_reclen;
2421 if (blksiz == NFS_DIRFRAGSIZ)
2422 blksiz = 0;
2423 uiop->uio_resid -= DIRHDSIZ;
2424 uiop->uio_iov->iov_base =
2425 (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ;
2426 uiop->uio_iov->iov_len -= DIRHDSIZ;
2427 nfsm_mtouio(uiop, len);
2428 cp = uiop->uio_iov->iov_base;
2429 tlen -= len;
2430 *cp = '\0'; /* null terminate */
2431 uiop->uio_iov->iov_base =
2432 (caddr_t)uiop->uio_iov->iov_base + tlen;
2433 uiop->uio_iov->iov_len -= tlen;
2434 uiop->uio_resid -= tlen;
2435 } else
2436 nfsm_adv(nfsm_rndup(len));
2437 if (v3) {
2438 nfsm_dissect(tl, u_int32_t *,
2439 3 * NFSX_UNSIGNED);
2440 } else {
2441 nfsm_dissect(tl, u_int32_t *,
2442 2 * NFSX_UNSIGNED);
2443 }
2444 if (bigenough) {
2445 if (v3) {
2446 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2447 uiop->uio_offset =
2448 fxdr_swapcookie3(tl);
2449 else
2450 uiop->uio_offset =
2451 fxdr_cookie3(tl);
2452 }
2453 else {
2454 uiop->uio_offset =
2455 fxdr_unsigned(off_t, *tl);
2456 }
2457 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2458 }
2459 if (v3)
2460 tl += 2;
2461 else
2462 tl++;
2463 more_dirs = fxdr_unsigned(int, *tl);
2464 }
2465 /*
2466 * If at end of rpc data, get the eof boolean
2467 */
2468 if (!more_dirs) {
2469 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2470 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2471 }
2472 m_freem(mrep);
2473 }
2474 /*
2475 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2476 * by increasing d_reclen for the last record.
2477 */
2478 if (blksiz > 0) {
2479 left = NFS_DIRFRAGSIZ - blksiz;
2480 dp->d_reclen += left;
2481 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2482 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2483 left;
2484 uiop->uio_iov->iov_len -= left;
2485 uiop->uio_resid -= left;
2486 }
2487
2488 /*
2489 * We are now either at the end of the directory or have filled the
2490 * block.
2491 */
2492 if (bigenough)
2493 dnp->n_direofoffset = uiop->uio_offset;
2494 nfsmout:
2495 return (error);
2496 }
2497
2498 /*
2499 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2500 */
2501 int
2502 nfs_readdirplusrpc(vp, uiop, cred)
2503 struct vnode *vp;
2504 struct uio *uiop;
2505 struct ucred *cred;
2506 {
2507 int len, left;
2508 struct dirent *dp = NULL;
2509 u_int32_t *tl;
2510 caddr_t cp;
2511 int32_t t1, t2;
2512 struct vnode *newvp;
2513 caddr_t bpos, dpos, cp2;
2514 struct mbuf *mreq, *mrep, *md, *mb;
2515 struct nameidata nami, *ndp = &nami;
2516 struct componentname *cnp = &ndp->ni_cnd;
2517 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2518 struct nfsnode *dnp = VTONFS(vp), *np;
2519 nfsfh_t *fhp;
2520 u_quad_t fileno;
2521 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2522 int attrflag, fhsize, nrpcs = 0, reclen;
2523 struct nfs_fattr fattr, *fp;
2524
2525 #ifdef DIAGNOSTIC
2526 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2527 panic("nfs readdirplusrpc bad uio");
2528 #endif
2529 ndp->ni_dvp = vp;
2530 newvp = NULLVP;
2531
2532 /*
2533 * Loop around doing readdir rpc's of size nm_readdirsize
2534 * truncated to a multiple of NFS_DIRFRAGSIZ.
2535 * The stopping criteria is EOF or buffer full.
2536 */
2537 while (more_dirs && bigenough) {
2538 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2539 bigenough = 0;
2540 break;
2541 }
2542 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2543 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2544 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2545 nfsm_fhtom(dnp, 1);
2546 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2547 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2548 txdr_swapcookie3(uiop->uio_offset, tl);
2549 } else {
2550 txdr_cookie3(uiop->uio_offset, tl);
2551 }
2552 tl += 2;
2553 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2554 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2555 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2556 *tl = txdr_unsigned(nmp->nm_rsize);
2557 nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2558 nfsm_postop_attr(vp, attrflag, 0);
2559 if (error) {
2560 m_freem(mrep);
2561 goto nfsmout;
2562 }
2563 nrpcs++;
2564 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2565 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2566 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2567 more_dirs = fxdr_unsigned(int, *tl);
2568
2569 /* loop thru the dir entries, doctoring them to 4bsd form */
2570 while (more_dirs && bigenough) {
2571 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2572 fileno = fxdr_hyper(tl);
2573 len = fxdr_unsigned(int, *(tl + 2));
2574 if (len <= 0 || len > NFS_MAXNAMLEN) {
2575 error = EBADRPC;
2576 m_freem(mrep);
2577 goto nfsmout;
2578 }
2579 tlen = nfsm_rndup(len);
2580 if (tlen == len)
2581 tlen += 4; /* To ensure null termination*/
2582 tlen += sizeof (off_t) + sizeof (int);
2583 reclen = ALIGN(tlen + DIRHDSIZ);
2584 tlen = reclen - DIRHDSIZ;
2585 left = NFS_DIRFRAGSIZ - blksiz;
2586 if (reclen > left) {
2587 /*
2588 * DIRFRAGSIZ is aligned, no need to align
2589 * again here.
2590 */
2591 dp->d_reclen += left;
2592 uiop->uio_iov->iov_base =
2593 (caddr_t)uiop->uio_iov->iov_base + left;
2594 uiop->uio_iov->iov_len -= left;
2595 uiop->uio_resid -= left;
2596 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2597 blksiz = 0;
2598 }
2599 if (reclen > uiop->uio_resid)
2600 bigenough = 0;
2601 if (bigenough) {
2602 dp = (struct dirent *)uiop->uio_iov->iov_base;
2603 dp->d_fileno = (int)fileno;
2604 dp->d_namlen = len;
2605 dp->d_reclen = reclen;
2606 dp->d_type = DT_UNKNOWN;
2607 blksiz += dp->d_reclen;
2608 if (blksiz == NFS_DIRFRAGSIZ)
2609 blksiz = 0;
2610 uiop->uio_resid -= DIRHDSIZ;
2611 uiop->uio_iov->iov_base =
2612 (caddr_t)uiop->uio_iov->iov_base +
2613 DIRHDSIZ;
2614 uiop->uio_iov->iov_len -= DIRHDSIZ;
2615 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2616 cnp->cn_namelen = len;
2617 nfsm_mtouio(uiop, len);
2618 cp = uiop->uio_iov->iov_base;
2619 tlen -= len;
2620 *cp = '\0';
2621 uiop->uio_iov->iov_base =
2622 (caddr_t)uiop->uio_iov->iov_base + tlen;
2623 uiop->uio_iov->iov_len -= tlen;
2624 uiop->uio_resid -= tlen;
2625 } else
2626 nfsm_adv(nfsm_rndup(len));
2627 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2628 if (bigenough) {
2629 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2630 uiop->uio_offset =
2631 fxdr_swapcookie3(tl);
2632 else
2633 uiop->uio_offset =
2634 fxdr_cookie3(tl);
2635 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2636 }
2637 tl += 2;
2638
2639 /*
2640 * Since the attributes are before the file handle
2641 * (sigh), we must skip over the attributes and then
2642 * come back and get them.
2643 */
2644 attrflag = fxdr_unsigned(int, *tl);
2645 if (attrflag) {
2646 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2647 memcpy(&fattr, fp, NFSX_V3FATTR);
2648 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2649 doit = fxdr_unsigned(int, *tl);
2650 if (doit) {
2651 nfsm_getfh(fhp, fhsize, 1);
2652 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2653 VREF(vp);
2654 newvp = vp;
2655 np = dnp;
2656 } else {
2657 error = nfs_nget(vp->v_mount, fhp,
2658 fhsize, &np);
2659 if (!error)
2660 newvp = NFSTOV(np);
2661 }
2662 if (!error) {
2663 const char *cp;
2664
2665 nfs_loadattrcache(&newvp, &fattr, 0, 0);
2666 dp->d_type =
2667 IFTODT(VTTOIF(np->n_vattr->va_type));
2668 ndp->ni_vp = newvp;
2669 cp = cnp->cn_nameptr + cnp->cn_namelen;
2670 cnp->cn_hash =
2671 namei_hash(cnp->cn_nameptr, &cp);
2672 if (cnp->cn_namelen <= NCHNAMLEN)
2673 cache_enter(ndp->ni_dvp, ndp->ni_vp,
2674 cnp);
2675 }
2676 }
2677 } else {
2678 /* Just skip over the file handle */
2679 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2680 i = fxdr_unsigned(int, *tl);
2681 nfsm_adv(nfsm_rndup(i));
2682 }
2683 if (newvp != NULLVP) {
2684 if (newvp == vp)
2685 vrele(newvp);
2686 else
2687 vput(newvp);
2688 newvp = NULLVP;
2689 }
2690 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2691 more_dirs = fxdr_unsigned(int, *tl);
2692 }
2693 /*
2694 * If at end of rpc data, get the eof boolean
2695 */
2696 if (!more_dirs) {
2697 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2698 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2699 }
2700 m_freem(mrep);
2701 }
2702 /*
2703 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2704 * by increasing d_reclen for the last record.
2705 */
2706 if (blksiz > 0) {
2707 left = NFS_DIRFRAGSIZ - blksiz;
2708 dp->d_reclen += left;
2709 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2710 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2711 left;
2712 uiop->uio_iov->iov_len -= left;
2713 uiop->uio_resid -= left;
2714 }
2715
2716 /*
2717 * We are now either at the end of the directory or have filled the
2718 * block.
2719 */
2720 if (bigenough)
2721 dnp->n_direofoffset = uiop->uio_offset;
2722 nfsmout:
2723 if (newvp != NULLVP) {
2724 if(newvp == vp)
2725 vrele(newvp);
2726 else
2727 vput(newvp);
2728 }
2729 return (error);
2730 }
2731 static char hextoasc[] = "0123456789abcdef";
2732
2733 /*
2734 * Silly rename. To make the NFS filesystem that is stateless look a little
2735 * more like the "ufs" a remove of an active vnode is translated to a rename
2736 * to a funny looking filename that is removed by nfs_inactive on the
2737 * nfsnode. There is the potential for another process on a different client
2738 * to create the same funny name between the nfs_lookitup() fails and the
2739 * nfs_rename() completes, but...
2740 */
2741 int
2742 nfs_sillyrename(dvp, vp, cnp)
2743 struct vnode *dvp, *vp;
2744 struct componentname *cnp;
2745 {
2746 struct sillyrename *sp;
2747 struct nfsnode *np;
2748 int error;
2749 short pid;
2750
2751 cache_purge(dvp);
2752 np = VTONFS(vp);
2753 #ifndef DIAGNOSTIC
2754 if (vp->v_type == VDIR)
2755 panic("nfs: sillyrename dir");
2756 #endif
2757 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2758 M_NFSREQ, M_WAITOK);
2759 sp->s_cred = crdup(cnp->cn_cred);
2760 sp->s_dvp = dvp;
2761 VREF(dvp);
2762
2763 /* Fudge together a funny name */
2764 pid = cnp->cn_proc->p_pid;
2765 memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2766 sp->s_namlen = 12;
2767 sp->s_name[8] = hextoasc[pid & 0xf];
2768 sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
2769 sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
2770 sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
2771
2772 /* Try lookitups until we get one that isn't there */
2773 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2774 cnp->cn_proc, (struct nfsnode **)0) == 0) {
2775 sp->s_name[4]++;
2776 if (sp->s_name[4] > 'z') {
2777 error = EINVAL;
2778 goto bad;
2779 }
2780 }
2781 error = nfs_renameit(dvp, cnp, sp);
2782 if (error)
2783 goto bad;
2784 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2785 cnp->cn_proc, &np);
2786 np->n_sillyrename = sp;
2787 return (0);
2788 bad:
2789 vrele(sp->s_dvp);
2790 crfree(sp->s_cred);
2791 free((caddr_t)sp, M_NFSREQ);
2792 return (error);
2793 }
2794
2795 /*
2796 * Look up a file name and optionally either update the file handle or
2797 * allocate an nfsnode, depending on the value of npp.
2798 * npp == NULL --> just do the lookup
2799 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2800 * handled too
2801 * *npp != NULL --> update the file handle in the vnode
2802 */
2803 int
2804 nfs_lookitup(dvp, name, len, cred, procp, npp)
2805 struct vnode *dvp;
2806 const char *name;
2807 int len;
2808 struct ucred *cred;
2809 struct proc *procp;
2810 struct nfsnode **npp;
2811 {
2812 u_int32_t *tl;
2813 caddr_t cp;
2814 int32_t t1, t2;
2815 struct vnode *newvp = (struct vnode *)0;
2816 struct nfsnode *np, *dnp = VTONFS(dvp);
2817 caddr_t bpos, dpos, cp2;
2818 int error = 0, fhlen, attrflag;
2819 struct mbuf *mreq, *mrep, *md, *mb;
2820 nfsfh_t *nfhp;
2821 const int v3 = NFS_ISV3(dvp);
2822
2823 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2824 nfsm_reqhead(dnp, NFSPROC_LOOKUP,
2825 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2826 nfsm_fhtom(dnp, v3);
2827 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2828 nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred);
2829 if (npp && !error) {
2830 nfsm_getfh(nfhp, fhlen, v3);
2831 if (*npp) {
2832 np = *npp;
2833 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2834 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2835 np->n_fhp = &np->n_fh;
2836 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2837 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2838 memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
2839 np->n_fhsize = fhlen;
2840 newvp = NFSTOV(np);
2841 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2842 VREF(dvp);
2843 newvp = dvp;
2844 } else {
2845 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2846 if (error) {
2847 m_freem(mrep);
2848 return (error);
2849 }
2850 newvp = NFSTOV(np);
2851 }
2852 if (v3) {
2853 nfsm_postop_attr(newvp, attrflag, 0);
2854 if (!attrflag && *npp == NULL) {
2855 m_freem(mrep);
2856 vput(newvp);
2857 return (ENOENT);
2858 }
2859 } else
2860 nfsm_loadattr(newvp, (struct vattr *)0, 0);
2861 }
2862 nfsm_reqdone;
2863 if (npp && *npp == NULL) {
2864 if (error) {
2865 if (newvp)
2866 vput(newvp);
2867 } else
2868 *npp = np;
2869 }
2870 return (error);
2871 }
2872
2873 /*
2874 * Nfs Version 3 commit rpc
2875 */
2876 int
2877 nfs_commit(vp, offset, cnt, procp)
2878 struct vnode *vp;
2879 off_t offset;
2880 uint32_t cnt;
2881 struct proc *procp;
2882 {
2883 caddr_t cp;
2884 u_int32_t *tl;
2885 int32_t t1, t2;
2886 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2887 caddr_t bpos, dpos, cp2;
2888 int error = 0, wccflag = NFSV3_WCCRATTR;
2889 struct mbuf *mreq, *mrep, *md, *mb;
2890 struct nfsnode *np;
2891
2892 KASSERT(NFS_ISV3(vp));
2893
2894 #ifdef NFS_DEBUG_COMMIT
2895 printf("commit %lu - %lu\n", (unsigned long)offset,
2896 (unsigned long)(offset + cnt));
2897 #endif
2898
2899 simple_lock(&nmp->nm_slock);
2900 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
2901 simple_unlock(&nmp->nm_slock);
2902 return (0);
2903 }
2904 simple_unlock(&nmp->nm_slock);
2905 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2906 np = VTONFS(vp);
2907 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
2908 nfsm_fhtom(np, 1);
2909 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2910 txdr_hyper(offset, tl);
2911 tl += 2;
2912 *tl = txdr_unsigned(cnt);
2913 nfsm_request(np, NFSPROC_COMMIT, procp, VTONFS(vp)->n_wcred);
2914 nfsm_wcc_data(vp, wccflag, 0);
2915 if (!error) {
2916 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2917 simple_lock(&nmp->nm_slock);
2918 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
2919 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
2920 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
2921 error = NFSERR_STALEWRITEVERF;
2922 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
2923 }
2924 simple_unlock(&nmp->nm_slock);
2925 }
2926 nfsm_reqdone;
2927 return (error);
2928 }
2929
2930 /*
2931 * Kludge City..
2932 * - make nfs_bmap() essentially a no-op that does no translation
2933 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2934 * (Maybe I could use the process's page mapping, but I was concerned that
2935 * Kernel Write might not be enabled and also figured copyout() would do
2936 * a lot more work than memcpy() and also it currently happens in the
2937 * context of the swapper process (2).
2938 */
2939 int
2940 nfs_bmap(v)
2941 void *v;
2942 {
2943 struct vop_bmap_args /* {
2944 struct vnode *a_vp;
2945 daddr_t a_bn;
2946 struct vnode **a_vpp;
2947 daddr_t *a_bnp;
2948 int *a_runp;
2949 } */ *ap = v;
2950 struct vnode *vp = ap->a_vp;
2951 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
2952
2953 if (ap->a_vpp != NULL)
2954 *ap->a_vpp = vp;
2955 if (ap->a_bnp != NULL)
2956 *ap->a_bnp = ap->a_bn << bshift;
2957 if (ap->a_runp != NULL)
2958 *ap->a_runp = 1024 * 1024; /* XXX */
2959 return (0);
2960 }
2961
2962 /*
2963 * Strategy routine.
2964 * For async requests when nfsiod(s) are running, queue the request by
2965 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2966 * request.
2967 */
2968 int
2969 nfs_strategy(v)
2970 void *v;
2971 {
2972 struct vop_strategy_args *ap = v;
2973 struct buf *bp = ap->a_bp;
2974 struct proc *p;
2975 int error = 0;
2976
2977 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
2978 panic("nfs physio/async");
2979 if (bp->b_flags & B_ASYNC)
2980 p = NULL;
2981 else
2982 p = curproc; /* XXX */
2983
2984 /*
2985 * If the op is asynchronous and an i/o daemon is waiting
2986 * queue the request, wake it up and wait for completion
2987 * otherwise just do it ourselves.
2988 */
2989
2990 if ((bp->b_flags & B_ASYNC) == 0 ||
2991 nfs_asyncio(bp))
2992 error = nfs_doio(bp, p);
2993 return (error);
2994 }
2995
2996 /*
2997 * fsync vnode op. Just call nfs_flush() with commit == 1.
2998 */
2999 /* ARGSUSED */
3000 int
3001 nfs_fsync(v)
3002 void *v;
3003 {
3004 struct vop_fsync_args /* {
3005 struct vnodeop_desc *a_desc;
3006 struct vnode * a_vp;
3007 struct ucred * a_cred;
3008 int a_flags;
3009 off_t offlo;
3010 off_t offhi;
3011 struct proc * a_p;
3012 } */ *ap = v;
3013
3014 return (nfs_flush(ap->a_vp, ap->a_cred,
3015 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1));
3016 }
3017
3018 /*
3019 * Flush all the data associated with a vnode.
3020 */
3021 int
3022 nfs_flush(vp, cred, waitfor, p, commit)
3023 struct vnode *vp;
3024 struct ucred *cred;
3025 int waitfor;
3026 struct proc *p;
3027 int commit;
3028 {
3029 struct nfsnode *np = VTONFS(vp);
3030 int error;
3031 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3032 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3033
3034 simple_lock(&vp->v_interlock);
3035 error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3036 if (np->n_flag & NWRITEERR) {
3037 error = np->n_error;
3038 np->n_flag &= ~NWRITEERR;
3039 }
3040 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3041 return (error);
3042 }
3043
3044 /*
3045 * Return POSIX pathconf information applicable to nfs.
3046 *
3047 * N.B. The NFS V2 protocol doesn't support this RPC.
3048 */
3049 /* ARGSUSED */
3050 int
3051 nfs_pathconf(v)
3052 void *v;
3053 {
3054 struct vop_pathconf_args /* {
3055 struct vnode *a_vp;
3056 int a_name;
3057 register_t *a_retval;
3058 } */ *ap = v;
3059 struct nfsv3_pathconf *pcp;
3060 struct vnode *vp = ap->a_vp;
3061 struct mbuf *mreq, *mrep, *md, *mb;
3062 int32_t t1, t2;
3063 u_int32_t *tl;
3064 caddr_t bpos, dpos, cp, cp2;
3065 int error = 0, attrflag;
3066 #ifndef NFS_V2_ONLY
3067 struct nfsmount *nmp;
3068 unsigned int l;
3069 u_int64_t maxsize;
3070 #endif
3071 const int v3 = NFS_ISV3(vp);
3072 struct nfsnode *np = VTONFS(vp);
3073
3074 switch (ap->a_name) {
3075 /* Names that can be resolved locally. */
3076 case _PC_PIPE_BUF:
3077 *ap->a_retval = PIPE_BUF;
3078 break;
3079 case _PC_SYNC_IO:
3080 *ap->a_retval = 1;
3081 break;
3082 /* Names that cannot be resolved locally; do an RPC, if possible. */
3083 case _PC_LINK_MAX:
3084 case _PC_NAME_MAX:
3085 case _PC_CHOWN_RESTRICTED:
3086 case _PC_NO_TRUNC:
3087 if (!v3) {
3088 error = EINVAL;
3089 break;
3090 }
3091 nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3092 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3093 nfsm_fhtom(np, 1);
3094 nfsm_request(np, NFSPROC_PATHCONF,
3095 curproc, curproc->p_ucred); /* XXX */
3096 nfsm_postop_attr(vp, attrflag, 0);
3097 if (!error) {
3098 nfsm_dissect(pcp, struct nfsv3_pathconf *,
3099 NFSX_V3PATHCONF);
3100 switch (ap->a_name) {
3101 case _PC_LINK_MAX:
3102 *ap->a_retval =
3103 fxdr_unsigned(register_t, pcp->pc_linkmax);
3104 break;
3105 case _PC_NAME_MAX:
3106 *ap->a_retval =
3107 fxdr_unsigned(register_t, pcp->pc_namemax);
3108 break;
3109 case _PC_CHOWN_RESTRICTED:
3110 *ap->a_retval =
3111 (pcp->pc_chownrestricted == nfs_true);
3112 break;
3113 case _PC_NO_TRUNC:
3114 *ap->a_retval =
3115 (pcp->pc_notrunc == nfs_true);
3116 break;
3117 }
3118 }
3119 nfsm_reqdone;
3120 break;
3121 case _PC_FILESIZEBITS:
3122 #ifndef NFS_V2_ONLY
3123 if (v3) {
3124 nmp = VFSTONFS(vp->v_mount);
3125 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3126 if ((error = nfs_fsinfo(nmp, vp,
3127 curproc->p_ucred, curproc)) != 0) /* XXX */
3128 break;
3129 for (l = 0, maxsize = nmp->nm_maxfilesize;
3130 (maxsize >> l) > 0; l++)
3131 ;
3132 *ap->a_retval = l + 1;
3133 } else
3134 #endif
3135 {
3136 *ap->a_retval = 32; /* NFS V2 limitation */
3137 }
3138 break;
3139 default:
3140 error = EINVAL;
3141 break;
3142 }
3143
3144 return (error);
3145 }
3146
3147 /*
3148 * NFS advisory byte-level locks.
3149 */
3150 int
3151 nfs_advlock(v)
3152 void *v;
3153 {
3154 struct vop_advlock_args /* {
3155 struct vnode *a_vp;
3156 caddr_t a_id;
3157 int a_op;
3158 struct flock *a_fl;
3159 int a_flags;
3160 } */ *ap = v;
3161 struct nfsnode *np = VTONFS(ap->a_vp);
3162
3163 return lf_advlock(ap, &np->n_lockf, np->n_size);
3164 }
3165
3166 /*
3167 * Print out the contents of an nfsnode.
3168 */
3169 int
3170 nfs_print(v)
3171 void *v;
3172 {
3173 struct vop_print_args /* {
3174 struct vnode *a_vp;
3175 } */ *ap = v;
3176 struct vnode *vp = ap->a_vp;
3177 struct nfsnode *np = VTONFS(vp);
3178
3179 printf("tag VT_NFS, fileid %ld fsid 0x%lx",
3180 np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3181 if (vp->v_type == VFIFO)
3182 fifo_printinfo(vp);
3183 printf("\n");
3184 return (0);
3185 }
3186
3187 /*
3188 * NFS file truncation.
3189 */
3190 int
3191 nfs_truncate(v)
3192 void *v;
3193 {
3194 #if 0
3195 struct vop_truncate_args /* {
3196 struct vnode *a_vp;
3197 off_t a_length;
3198 int a_flags;
3199 struct ucred *a_cred;
3200 struct proc *a_p;
3201 } */ *ap = v;
3202 #endif
3203
3204 /* Use nfs_setattr */
3205 return (EOPNOTSUPP);
3206 }
3207
3208 /*
3209 * NFS update.
3210 */
3211 int
3212 nfs_update(v)
3213 void *v;
3214 #if 0
3215 struct vop_update_args /* {
3216 struct vnode *a_vp;
3217 struct timespec *a_ta;
3218 struct timespec *a_tm;
3219 int a_waitfor;
3220 } */ *ap = v;
3221 #endif
3222 {
3223
3224 /* Use nfs_setattr */
3225 return (EOPNOTSUPP);
3226 }
3227
3228 /*
3229 * Just call bwrite().
3230 */
3231 int
3232 nfs_bwrite(v)
3233 void *v;
3234 {
3235 struct vop_bwrite_args /* {
3236 struct vnode *a_bp;
3237 } */ *ap = v;
3238
3239 return (bwrite(ap->a_bp));
3240 }
3241
3242 /*
3243 * nfs unlock wrapper.
3244 */
3245 int
3246 nfs_unlock(void *v)
3247 {
3248 struct vop_unlock_args /* {
3249 struct vnode *a_vp;
3250 int a_flags;
3251 } */ *ap = v;
3252 struct vnode *vp = ap->a_vp;
3253
3254 /*
3255 * VOP_UNLOCK can be called by nfs_loadattrcache
3256 * with v_data == 0.
3257 */
3258 if (VTONFS(vp)) {
3259 nfs_delayedtruncate(vp);
3260 }
3261
3262 return genfs_unlock(v);
3263 }
3264
3265 /*
3266 * nfs special file access vnode op.
3267 * Essentially just get vattr and then imitate iaccess() since the device is
3268 * local to the client.
3269 */
3270 int
3271 nfsspec_access(v)
3272 void *v;
3273 {
3274 struct vop_access_args /* {
3275 struct vnode *a_vp;
3276 int a_mode;
3277 struct ucred *a_cred;
3278 struct proc *a_p;
3279 } */ *ap = v;
3280 struct vattr va;
3281 struct vnode *vp = ap->a_vp;
3282 int error;
3283
3284 error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3285 if (error)
3286 return (error);
3287
3288 /*
3289 * Disallow write attempts on filesystems mounted read-only;
3290 * unless the file is a socket, fifo, or a block or character
3291 * device resident on the filesystem.
3292 */
3293 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3294 switch (vp->v_type) {
3295 case VREG:
3296 case VDIR:
3297 case VLNK:
3298 return (EROFS);
3299 default:
3300 break;
3301 }
3302 }
3303
3304 return (vaccess(va.va_type, va.va_mode,
3305 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3306 }
3307
3308 /*
3309 * Read wrapper for special devices.
3310 */
3311 int
3312 nfsspec_read(v)
3313 void *v;
3314 {
3315 struct vop_read_args /* {
3316 struct vnode *a_vp;
3317 struct uio *a_uio;
3318 int a_ioflag;
3319 struct ucred *a_cred;
3320 } */ *ap = v;
3321 struct nfsnode *np = VTONFS(ap->a_vp);
3322
3323 /*
3324 * Set access flag.
3325 */
3326 np->n_flag |= NACC;
3327 np->n_atim.tv_sec = time.tv_sec;
3328 np->n_atim.tv_nsec = time.tv_usec * 1000;
3329 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3330 }
3331
3332 /*
3333 * Write wrapper for special devices.
3334 */
3335 int
3336 nfsspec_write(v)
3337 void *v;
3338 {
3339 struct vop_write_args /* {
3340 struct vnode *a_vp;
3341 struct uio *a_uio;
3342 int a_ioflag;
3343 struct ucred *a_cred;
3344 } */ *ap = v;
3345 struct nfsnode *np = VTONFS(ap->a_vp);
3346
3347 /*
3348 * Set update flag.
3349 */
3350 np->n_flag |= NUPD;
3351 np->n_mtim.tv_sec = time.tv_sec;
3352 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3353 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3354 }
3355
3356 /*
3357 * Close wrapper for special devices.
3358 *
3359 * Update the times on the nfsnode then do device close.
3360 */
3361 int
3362 nfsspec_close(v)
3363 void *v;
3364 {
3365 struct vop_close_args /* {
3366 struct vnode *a_vp;
3367 int a_fflag;
3368 struct ucred *a_cred;
3369 struct proc *a_p;
3370 } */ *ap = v;
3371 struct vnode *vp = ap->a_vp;
3372 struct nfsnode *np = VTONFS(vp);
3373 struct vattr vattr;
3374
3375 if (np->n_flag & (NACC | NUPD)) {
3376 np->n_flag |= NCHG;
3377 if (vp->v_usecount == 1 &&
3378 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3379 VATTR_NULL(&vattr);
3380 if (np->n_flag & NACC)
3381 vattr.va_atime = np->n_atim;
3382 if (np->n_flag & NUPD)
3383 vattr.va_mtime = np->n_mtim;
3384 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3385 }
3386 }
3387 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3388 }
3389
3390 /*
3391 * Read wrapper for fifos.
3392 */
3393 int
3394 nfsfifo_read(v)
3395 void *v;
3396 {
3397 struct vop_read_args /* {
3398 struct vnode *a_vp;
3399 struct uio *a_uio;
3400 int a_ioflag;
3401 struct ucred *a_cred;
3402 } */ *ap = v;
3403 struct nfsnode *np = VTONFS(ap->a_vp);
3404
3405 /*
3406 * Set access flag.
3407 */
3408 np->n_flag |= NACC;
3409 np->n_atim.tv_sec = time.tv_sec;
3410 np->n_atim.tv_nsec = time.tv_usec * 1000;
3411 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3412 }
3413
3414 /*
3415 * Write wrapper for fifos.
3416 */
3417 int
3418 nfsfifo_write(v)
3419 void *v;
3420 {
3421 struct vop_write_args /* {
3422 struct vnode *a_vp;
3423 struct uio *a_uio;
3424 int a_ioflag;
3425 struct ucred *a_cred;
3426 } */ *ap = v;
3427 struct nfsnode *np = VTONFS(ap->a_vp);
3428
3429 /*
3430 * Set update flag.
3431 */
3432 np->n_flag |= NUPD;
3433 np->n_mtim.tv_sec = time.tv_sec;
3434 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3435 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3436 }
3437
3438 /*
3439 * Close wrapper for fifos.
3440 *
3441 * Update the times on the nfsnode then do fifo close.
3442 */
3443 int
3444 nfsfifo_close(v)
3445 void *v;
3446 {
3447 struct vop_close_args /* {
3448 struct vnode *a_vp;
3449 int a_fflag;
3450 struct ucred *a_cred;
3451 struct proc *a_p;
3452 } */ *ap = v;
3453 struct vnode *vp = ap->a_vp;
3454 struct nfsnode *np = VTONFS(vp);
3455 struct vattr vattr;
3456
3457 if (np->n_flag & (NACC | NUPD)) {
3458 if (np->n_flag & NACC) {
3459 np->n_atim.tv_sec = time.tv_sec;
3460 np->n_atim.tv_nsec = time.tv_usec * 1000;
3461 }
3462 if (np->n_flag & NUPD) {
3463 np->n_mtim.tv_sec = time.tv_sec;
3464 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3465 }
3466 np->n_flag |= NCHG;
3467 if (vp->v_usecount == 1 &&
3468 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3469 VATTR_NULL(&vattr);
3470 if (np->n_flag & NACC)
3471 vattr.va_atime = np->n_atim;
3472 if (np->n_flag & NUPD)
3473 vattr.va_mtime = np->n_mtim;
3474 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3475 }
3476 }
3477 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3478 }
3479