nfs_vnops.c revision 1.176 1 /* $NetBSD: nfs_vnops.c,v 1.176 2003/07/30 12:25:39 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 1.176 2003/07/30 12:25:39 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, stalewriteverfp)
1241 struct vnode *vp;
1242 struct uio *uiop;
1243 int *iomode;
1244 boolean_t pageprotected;
1245 boolean_t *stalewriteverfp;
1246 {
1247 u_int32_t *tl;
1248 caddr_t cp;
1249 int32_t t1, t2;
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 size_t origresid;
1261
1262 simple_lock_init(&ctx.nwc_slock);
1263 ctx.nwc_mbufcount = 1;
1264
1265 if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1266 panic("writerpc readonly vp %p", vp);
1267 }
1268
1269 #ifdef DIAGNOSTIC
1270 if (uiop->uio_iovcnt != 1)
1271 panic("nfs: writerpc iovcnt > 1");
1272 #endif
1273 tsiz = uiop->uio_resid;
1274 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1275 return (EFBIG);
1276 if (pageprotected) {
1277 l = curlwp;
1278 PHOLD(l);
1279 }
1280 retry:
1281 origresid = uiop->uio_resid;
1282 KASSERT(origresid == uiop->uio_iov->iov_len);
1283 while (tsiz > 0) {
1284 uint32_t datalen; /* data bytes need to be allocated in mbuf */
1285 uint32_t backup;
1286 boolean_t stalewriteverf = FALSE;
1287
1288 nfsstats.rpccnt[NFSPROC_WRITE]++;
1289 len = min(tsiz, nmp->nm_wsize);
1290 datalen = pageprotected ? 0 : nfsm_rndup(len);
1291 nfsm_reqhead(np, NFSPROC_WRITE,
1292 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1293 nfsm_fhtom(np, v3);
1294 if (v3) {
1295 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1296 txdr_hyper(uiop->uio_offset, tl);
1297 tl += 2;
1298 *tl++ = txdr_unsigned(len);
1299 *tl++ = txdr_unsigned(*iomode);
1300 *tl = txdr_unsigned(len);
1301 } else {
1302 u_int32_t x;
1303
1304 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1305 /* Set both "begin" and "current" to non-garbage. */
1306 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1307 *tl++ = x; /* "begin offset" */
1308 *tl++ = x; /* "current offset" */
1309 x = txdr_unsigned(len);
1310 *tl++ = x; /* total to this offset */
1311 *tl = x; /* size of this write */
1312
1313 }
1314 if (pageprotected) {
1315 /*
1316 * since we know pages can't be modified during i/o,
1317 * no need to copy them for us.
1318 */
1319 struct mbuf *m;
1320 struct iovec *iovp = uiop->uio_iov;
1321
1322 m = m_get(M_WAIT, MT_DATA);
1323 MCLAIM(m, &nfs_mowner);
1324 MEXTADD(m, iovp->iov_base, len, M_MBUF,
1325 nfs_writerpc_extfree, &ctx);
1326 m->m_flags |= M_EXT_ROMAP;
1327 m->m_len = len;
1328 mb->m_next = m;
1329 /*
1330 * no need to maintain mb and bpos here
1331 * because no one care them later.
1332 */
1333 #if 0
1334 mb = m;
1335 bpos = mtod(caddr_t, mb) + mb->m_len;
1336 #endif
1337 iovp->iov_base = (char *)iovp->iov_base + len;
1338 iovp->iov_len -= len;
1339 uiop->uio_offset += len;
1340 uiop->uio_resid -= len;
1341 s = splvm();
1342 simple_lock(&ctx.nwc_slock);
1343 ctx.nwc_mbufcount++;
1344 simple_unlock(&ctx.nwc_slock);
1345 splx(s);
1346 } else {
1347 nfsm_uiotom(uiop, len);
1348 }
1349 nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp,
1350 VTONFS(vp)->n_wcred);
1351 if (v3) {
1352 wccflag = NFSV3_WCCCHK;
1353 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC);
1354 if (!error) {
1355 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1356 + NFSX_V3WRITEVERF);
1357 rlen = fxdr_unsigned(int, *tl++);
1358 if (rlen == 0) {
1359 error = NFSERR_IO;
1360 m_freem(mrep);
1361 break;
1362 } else if (rlen < len) {
1363 backup = len - rlen;
1364 uiop->uio_iov->iov_base =
1365 (caddr_t)uiop->uio_iov->iov_base -
1366 backup;
1367 uiop->uio_iov->iov_len += backup;
1368 uiop->uio_offset -= backup;
1369 uiop->uio_resid += backup;
1370 len = rlen;
1371 }
1372 commit = fxdr_unsigned(int, *tl++);
1373
1374 /*
1375 * Return the lowest committment level
1376 * obtained by any of the RPCs.
1377 */
1378 if (committed == NFSV3WRITE_FILESYNC)
1379 committed = commit;
1380 else if (committed == NFSV3WRITE_DATASYNC &&
1381 commit == NFSV3WRITE_UNSTABLE)
1382 committed = commit;
1383 simple_lock(&nmp->nm_slock);
1384 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1385 memcpy(nmp->nm_writeverf, tl,
1386 NFSX_V3WRITEVERF);
1387 nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1388 } else if ((nmp->nm_iflag &
1389 NFSMNT_STALEWRITEVERF) ||
1390 memcmp(tl, nmp->nm_writeverf,
1391 NFSX_V3WRITEVERF)) {
1392 memcpy(nmp->nm_writeverf, tl,
1393 NFSX_V3WRITEVERF);
1394 /*
1395 * note NFSMNT_STALEWRITEVERF
1396 * if we're the first thread to
1397 * notice it.
1398 */
1399 if ((nmp->nm_iflag &
1400 NFSMNT_STALEWRITEVERF) == 0) {
1401 stalewriteverf = TRUE;
1402 nmp->nm_iflag |=
1403 NFSMNT_STALEWRITEVERF;
1404 }
1405 }
1406 simple_unlock(&nmp->nm_slock);
1407 }
1408 } else
1409 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1410 if (wccflag)
1411 VTONFS(vp)->n_mtime =
1412 VTONFS(vp)->n_vattr->va_mtime.tv_sec;
1413 m_freem(mrep);
1414 if (error)
1415 break;
1416 tsiz -= len;
1417 if (stalewriteverf) {
1418 *stalewriteverfp = TRUE;
1419 stalewriteverf = FALSE;
1420 if (committed == NFSV3WRITE_UNSTABLE &&
1421 len != origresid) {
1422 /*
1423 * if our write requests weren't atomic but
1424 * unstable, datas in previous iterations
1425 * might have already been lost now.
1426 * then, we should resend them to nfsd.
1427 */
1428 backup = origresid - tsiz;
1429 uiop->uio_iov->iov_base =
1430 (caddr_t)uiop->uio_iov->iov_base - backup;
1431 uiop->uio_iov->iov_len += backup;
1432 uiop->uio_offset -= backup;
1433 uiop->uio_resid += backup;
1434 tsiz = origresid;
1435 goto retry;
1436 }
1437 }
1438 }
1439 if (pageprotected) {
1440 /*
1441 * wait until mbufs go away.
1442 * retransmitted mbufs can survive longer than rpc requests
1443 * themselves.
1444 */
1445 s = splvm();
1446 simple_lock(&ctx.nwc_slock);
1447 ctx.nwc_mbufcount--;
1448 while (ctx.nwc_mbufcount > 0) {
1449 ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock);
1450 }
1451 simple_unlock(&ctx.nwc_slock);
1452 splx(s);
1453 PRELE(l);
1454 }
1455 nfsmout:
1456 *iomode = committed;
1457 if (error)
1458 uiop->uio_resid = tsiz;
1459 return (error);
1460 }
1461
1462 /*
1463 * nfs mknod rpc
1464 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1465 * mode set to specify the file type and the size field for rdev.
1466 */
1467 int
1468 nfs_mknodrpc(dvp, vpp, cnp, vap)
1469 struct vnode *dvp;
1470 struct vnode **vpp;
1471 struct componentname *cnp;
1472 struct vattr *vap;
1473 {
1474 struct nfsv2_sattr *sp;
1475 u_int32_t *tl;
1476 caddr_t cp;
1477 int32_t t1, t2;
1478 struct vnode *newvp = (struct vnode *)0;
1479 struct nfsnode *dnp, *np;
1480 char *cp2;
1481 caddr_t bpos, dpos;
1482 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1483 struct mbuf *mreq, *mrep, *md, *mb;
1484 u_int32_t rdev;
1485 const int v3 = NFS_ISV3(dvp);
1486
1487 if (vap->va_type == VCHR || vap->va_type == VBLK)
1488 rdev = txdr_unsigned(vap->va_rdev);
1489 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1490 rdev = nfs_xdrneg1;
1491 else {
1492 VOP_ABORTOP(dvp, cnp);
1493 vput(dvp);
1494 return (EOPNOTSUPP);
1495 }
1496 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1497 dnp = VTONFS(dvp);
1498 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1499 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1500 nfsm_fhtom(dnp, v3);
1501 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1502 if (v3) {
1503 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1504 *tl++ = vtonfsv3_type(vap->va_type);
1505 nfsm_v3attrbuild(vap, FALSE);
1506 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1507 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1508 *tl++ = txdr_unsigned(major(vap->va_rdev));
1509 *tl = txdr_unsigned(minor(vap->va_rdev));
1510 }
1511 } else {
1512 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1513 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1514 sp->sa_uid = nfs_xdrneg1;
1515 sp->sa_gid = nfs_xdrneg1;
1516 sp->sa_size = rdev;
1517 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1518 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1519 }
1520 nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1521 if (!error) {
1522 nfsm_mtofh(dvp, newvp, v3, gotvp);
1523 if (!gotvp) {
1524 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1525 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1526 if (!error)
1527 newvp = NFSTOV(np);
1528 }
1529 }
1530 if (v3)
1531 nfsm_wcc_data(dvp, wccflag, 0);
1532 nfsm_reqdone;
1533 if (error) {
1534 if (newvp)
1535 vput(newvp);
1536 } else {
1537 if (cnp->cn_flags & MAKEENTRY)
1538 cache_enter(dvp, newvp, cnp);
1539 *vpp = newvp;
1540 }
1541 PNBUF_PUT(cnp->cn_pnbuf);
1542 VTONFS(dvp)->n_flag |= NMODIFIED;
1543 if (!wccflag)
1544 VTONFS(dvp)->n_attrstamp = 0;
1545 vput(dvp);
1546 return (error);
1547 }
1548
1549 /*
1550 * nfs mknod vop
1551 * just call nfs_mknodrpc() to do the work.
1552 */
1553 /* ARGSUSED */
1554 int
1555 nfs_mknod(v)
1556 void *v;
1557 {
1558 struct vop_mknod_args /* {
1559 struct vnode *a_dvp;
1560 struct vnode **a_vpp;
1561 struct componentname *a_cnp;
1562 struct vattr *a_vap;
1563 } */ *ap = v;
1564 int error;
1565
1566 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1567 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1568 return (error);
1569 }
1570
1571 static u_long create_verf;
1572 /*
1573 * nfs file create call
1574 */
1575 int
1576 nfs_create(v)
1577 void *v;
1578 {
1579 struct vop_create_args /* {
1580 struct vnode *a_dvp;
1581 struct vnode **a_vpp;
1582 struct componentname *a_cnp;
1583 struct vattr *a_vap;
1584 } */ *ap = v;
1585 struct vnode *dvp = ap->a_dvp;
1586 struct vattr *vap = ap->a_vap;
1587 struct componentname *cnp = ap->a_cnp;
1588 struct nfsv2_sattr *sp;
1589 u_int32_t *tl;
1590 caddr_t cp;
1591 int32_t t1, t2;
1592 struct nfsnode *dnp, *np = (struct nfsnode *)0;
1593 struct vnode *newvp = (struct vnode *)0;
1594 caddr_t bpos, dpos, cp2;
1595 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1596 struct mbuf *mreq, *mrep, *md, *mb;
1597 const int v3 = NFS_ISV3(dvp);
1598
1599 /*
1600 * Oops, not for me..
1601 */
1602 if (vap->va_type == VSOCK)
1603 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1604
1605 #ifdef VA_EXCLUSIVE
1606 if (vap->va_vaflags & VA_EXCLUSIVE)
1607 fmode |= O_EXCL;
1608 #endif
1609 again:
1610 error = 0;
1611 nfsstats.rpccnt[NFSPROC_CREATE]++;
1612 dnp = VTONFS(dvp);
1613 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1614 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1615 nfsm_fhtom(dnp, v3);
1616 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1617 if (v3) {
1618 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1619 if (fmode & O_EXCL) {
1620 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1621 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1622 #ifdef INET
1623 if (TAILQ_FIRST(&in_ifaddr))
1624 *tl++ = TAILQ_FIRST(&in_ifaddr)->ia_addr.sin_addr.s_addr;
1625 else
1626 *tl++ = create_verf;
1627 #else
1628 *tl++ = create_verf;
1629 #endif
1630 *tl = ++create_verf;
1631 } else {
1632 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1633 nfsm_v3attrbuild(vap, FALSE);
1634 }
1635 } else {
1636 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1637 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1638 sp->sa_uid = nfs_xdrneg1;
1639 sp->sa_gid = nfs_xdrneg1;
1640 sp->sa_size = 0;
1641 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1642 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1643 }
1644 nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1645 if (!error) {
1646 nfsm_mtofh(dvp, newvp, v3, gotvp);
1647 if (!gotvp) {
1648 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1649 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1650 if (!error)
1651 newvp = NFSTOV(np);
1652 }
1653 }
1654 if (v3)
1655 nfsm_wcc_data(dvp, wccflag, 0);
1656 nfsm_reqdone;
1657 if (error) {
1658 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1659 fmode &= ~O_EXCL;
1660 goto again;
1661 }
1662 if (newvp)
1663 vput(newvp);
1664 } else if (v3 && (fmode & O_EXCL))
1665 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1666 if (!error) {
1667 if (cnp->cn_flags & MAKEENTRY)
1668 cache_enter(dvp, newvp, cnp);
1669 *ap->a_vpp = newvp;
1670 }
1671 PNBUF_PUT(cnp->cn_pnbuf);
1672 VTONFS(dvp)->n_flag |= NMODIFIED;
1673 if (!wccflag)
1674 VTONFS(dvp)->n_attrstamp = 0;
1675 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1676 vput(dvp);
1677 return (error);
1678 }
1679
1680 /*
1681 * nfs file remove call
1682 * To try and make nfs semantics closer to ufs semantics, a file that has
1683 * other processes using the vnode is renamed instead of removed and then
1684 * removed later on the last close.
1685 * - If v_usecount > 1
1686 * If a rename is not already in the works
1687 * call nfs_sillyrename() to set it up
1688 * else
1689 * do the remove rpc
1690 */
1691 int
1692 nfs_remove(v)
1693 void *v;
1694 {
1695 struct vop_remove_args /* {
1696 struct vnodeop_desc *a_desc;
1697 struct vnode * a_dvp;
1698 struct vnode * a_vp;
1699 struct componentname * a_cnp;
1700 } */ *ap = v;
1701 struct vnode *vp = ap->a_vp;
1702 struct vnode *dvp = ap->a_dvp;
1703 struct componentname *cnp = ap->a_cnp;
1704 struct nfsnode *np = VTONFS(vp);
1705 int error = 0;
1706 struct vattr vattr;
1707
1708 #ifndef DIAGNOSTIC
1709 if ((cnp->cn_flags & HASBUF) == 0)
1710 panic("nfs_remove: no name");
1711 if (vp->v_usecount < 1)
1712 panic("nfs_remove: bad v_usecount");
1713 #endif
1714 if (vp->v_type == VDIR)
1715 error = EPERM;
1716 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1717 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1718 vattr.va_nlink > 1)) {
1719 /*
1720 * Purge the name cache so that the chance of a lookup for
1721 * the name succeeding while the remove is in progress is
1722 * minimized. Without node locking it can still happen, such
1723 * that an I/O op returns ESTALE, but since you get this if
1724 * another host removes the file..
1725 */
1726 cache_purge(vp);
1727 /*
1728 * throw away biocache buffers, mainly to avoid
1729 * unnecessary delayed writes later.
1730 */
1731 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1732 /* Do the rpc */
1733 if (error != EINTR)
1734 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1735 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1736 /*
1737 * Kludge City: If the first reply to the remove rpc is lost..
1738 * the reply to the retransmitted request will be ENOENT
1739 * since the file was in fact removed
1740 * Therefore, we cheat and return success.
1741 */
1742 if (error == ENOENT)
1743 error = 0;
1744 } else if (!np->n_sillyrename)
1745 error = nfs_sillyrename(dvp, vp, cnp);
1746 PNBUF_PUT(cnp->cn_pnbuf);
1747 if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1748 vattr.va_nlink == 1) {
1749 np->n_flag |= NREMOVED;
1750 }
1751 np->n_attrstamp = 0;
1752 VN_KNOTE(vp, NOTE_DELETE);
1753 VN_KNOTE(dvp, NOTE_WRITE);
1754 if (dvp == vp)
1755 vrele(vp);
1756 else
1757 vput(vp);
1758 vput(dvp);
1759 return (error);
1760 }
1761
1762 /*
1763 * nfs file remove rpc called from nfs_inactive
1764 */
1765 int
1766 nfs_removeit(sp)
1767 struct sillyrename *sp;
1768 {
1769
1770 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1771 (struct proc *)0));
1772 }
1773
1774 /*
1775 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1776 */
1777 int
1778 nfs_removerpc(dvp, name, namelen, cred, proc)
1779 struct vnode *dvp;
1780 const char *name;
1781 int namelen;
1782 struct ucred *cred;
1783 struct proc *proc;
1784 {
1785 u_int32_t *tl;
1786 caddr_t cp;
1787 int32_t t1, t2;
1788 caddr_t bpos, dpos, cp2;
1789 int error = 0, wccflag = NFSV3_WCCRATTR;
1790 struct mbuf *mreq, *mrep, *md, *mb;
1791 const int v3 = NFS_ISV3(dvp);
1792 struct nfsnode *dnp = VTONFS(dvp);
1793
1794 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1795 nfsm_reqhead(dnp, NFSPROC_REMOVE,
1796 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1797 nfsm_fhtom(dnp, v3);
1798 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1799 nfsm_request(dnp, NFSPROC_REMOVE, proc, cred);
1800 if (v3)
1801 nfsm_wcc_data(dvp, wccflag, 0);
1802 nfsm_reqdone;
1803 VTONFS(dvp)->n_flag |= NMODIFIED;
1804 if (!wccflag)
1805 VTONFS(dvp)->n_attrstamp = 0;
1806 return (error);
1807 }
1808
1809 /*
1810 * nfs file rename call
1811 */
1812 int
1813 nfs_rename(v)
1814 void *v;
1815 {
1816 struct vop_rename_args /* {
1817 struct vnode *a_fdvp;
1818 struct vnode *a_fvp;
1819 struct componentname *a_fcnp;
1820 struct vnode *a_tdvp;
1821 struct vnode *a_tvp;
1822 struct componentname *a_tcnp;
1823 } */ *ap = v;
1824 struct vnode *fvp = ap->a_fvp;
1825 struct vnode *tvp = ap->a_tvp;
1826 struct vnode *fdvp = ap->a_fdvp;
1827 struct vnode *tdvp = ap->a_tdvp;
1828 struct componentname *tcnp = ap->a_tcnp;
1829 struct componentname *fcnp = ap->a_fcnp;
1830 int error;
1831
1832 #ifndef DIAGNOSTIC
1833 if ((tcnp->cn_flags & HASBUF) == 0 ||
1834 (fcnp->cn_flags & HASBUF) == 0)
1835 panic("nfs_rename: no name");
1836 #endif
1837 /* Check for cross-device rename */
1838 if ((fvp->v_mount != tdvp->v_mount) ||
1839 (tvp && (fvp->v_mount != tvp->v_mount))) {
1840 error = EXDEV;
1841 goto out;
1842 }
1843
1844 /*
1845 * If the tvp exists and is in use, sillyrename it before doing the
1846 * rename of the new file over it.
1847 */
1848 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1849 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1850 VN_KNOTE(tvp, NOTE_DELETE);
1851 vput(tvp);
1852 tvp = NULL;
1853 }
1854
1855 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1856 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1857 tcnp->cn_proc);
1858
1859 VN_KNOTE(fdvp, NOTE_WRITE);
1860 VN_KNOTE(tdvp, NOTE_WRITE);
1861 if (fvp->v_type == VDIR) {
1862 if (tvp != NULL && tvp->v_type == VDIR)
1863 cache_purge(tdvp);
1864 cache_purge(fdvp);
1865 }
1866 out:
1867 if (tdvp == tvp)
1868 vrele(tdvp);
1869 else
1870 vput(tdvp);
1871 if (tvp)
1872 vput(tvp);
1873 vrele(fdvp);
1874 vrele(fvp);
1875 /*
1876 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1877 */
1878 if (error == ENOENT)
1879 error = 0;
1880 return (error);
1881 }
1882
1883 /*
1884 * nfs file rename rpc called from nfs_remove() above
1885 */
1886 int
1887 nfs_renameit(sdvp, scnp, sp)
1888 struct vnode *sdvp;
1889 struct componentname *scnp;
1890 struct sillyrename *sp;
1891 {
1892 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1893 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1894 }
1895
1896 /*
1897 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1898 */
1899 int
1900 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1901 struct vnode *fdvp;
1902 const char *fnameptr;
1903 int fnamelen;
1904 struct vnode *tdvp;
1905 const char *tnameptr;
1906 int tnamelen;
1907 struct ucred *cred;
1908 struct proc *proc;
1909 {
1910 u_int32_t *tl;
1911 caddr_t cp;
1912 int32_t t1, t2;
1913 caddr_t bpos, dpos, cp2;
1914 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1915 struct mbuf *mreq, *mrep, *md, *mb;
1916 const int v3 = NFS_ISV3(fdvp);
1917 struct nfsnode *fdnp = VTONFS(fdvp);
1918
1919 nfsstats.rpccnt[NFSPROC_RENAME]++;
1920 nfsm_reqhead(fdnp, NFSPROC_RENAME,
1921 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1922 nfsm_rndup(tnamelen));
1923 nfsm_fhtom(fdnp, v3);
1924 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1925 nfsm_fhtom(VTONFS(tdvp), v3);
1926 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1927 nfsm_request(fdnp, NFSPROC_RENAME, proc, cred);
1928 if (v3) {
1929 nfsm_wcc_data(fdvp, fwccflag, 0);
1930 nfsm_wcc_data(tdvp, twccflag, 0);
1931 }
1932 nfsm_reqdone;
1933 VTONFS(fdvp)->n_flag |= NMODIFIED;
1934 VTONFS(tdvp)->n_flag |= NMODIFIED;
1935 if (!fwccflag)
1936 VTONFS(fdvp)->n_attrstamp = 0;
1937 if (!twccflag)
1938 VTONFS(tdvp)->n_attrstamp = 0;
1939 return (error);
1940 }
1941
1942 /*
1943 * nfs hard link create call
1944 */
1945 int
1946 nfs_link(v)
1947 void *v;
1948 {
1949 struct vop_link_args /* {
1950 struct vnode *a_dvp;
1951 struct vnode *a_vp;
1952 struct componentname *a_cnp;
1953 } */ *ap = v;
1954 struct vnode *vp = ap->a_vp;
1955 struct vnode *dvp = ap->a_dvp;
1956 struct componentname *cnp = ap->a_cnp;
1957 u_int32_t *tl;
1958 caddr_t cp;
1959 int32_t t1, t2;
1960 caddr_t bpos, dpos, cp2;
1961 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1962 struct mbuf *mreq, *mrep, *md, *mb;
1963 /* XXX Should be const and initialised? */
1964 int v3;
1965 struct nfsnode *np;
1966
1967 if (dvp->v_mount != vp->v_mount) {
1968 VOP_ABORTOP(dvp, cnp);
1969 vput(dvp);
1970 return (EXDEV);
1971 }
1972 if (dvp != vp) {
1973 error = vn_lock(vp, LK_EXCLUSIVE);
1974 if (error != 0) {
1975 VOP_ABORTOP(dvp, cnp);
1976 vput(dvp);
1977 return error;
1978 }
1979 }
1980
1981 /*
1982 * Push all writes to the server, so that the attribute cache
1983 * doesn't get "out of sync" with the server.
1984 * XXX There should be a better way!
1985 */
1986 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc);
1987
1988 v3 = NFS_ISV3(vp);
1989 nfsstats.rpccnt[NFSPROC_LINK]++;
1990 np = VTONFS(vp);
1991 nfsm_reqhead(np, NFSPROC_LINK,
1992 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1993 nfsm_fhtom(np, v3);
1994 nfsm_fhtom(VTONFS(dvp), v3);
1995 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1996 nfsm_request(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1997 if (v3) {
1998 nfsm_postop_attr(vp, attrflag, 0);
1999 nfsm_wcc_data(dvp, wccflag, 0);
2000 }
2001 nfsm_reqdone;
2002 PNBUF_PUT(cnp->cn_pnbuf);
2003 VTONFS(dvp)->n_flag |= NMODIFIED;
2004 if (!attrflag)
2005 VTONFS(vp)->n_attrstamp = 0;
2006 if (!wccflag)
2007 VTONFS(dvp)->n_attrstamp = 0;
2008 if (dvp != vp)
2009 VOP_UNLOCK(vp, 0);
2010 VN_KNOTE(vp, NOTE_LINK);
2011 VN_KNOTE(dvp, NOTE_WRITE);
2012 vput(dvp);
2013 /*
2014 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2015 */
2016 if (error == EEXIST)
2017 error = 0;
2018 return (error);
2019 }
2020
2021 /*
2022 * nfs symbolic link create call
2023 */
2024 int
2025 nfs_symlink(v)
2026 void *v;
2027 {
2028 struct vop_symlink_args /* {
2029 struct vnode *a_dvp;
2030 struct vnode **a_vpp;
2031 struct componentname *a_cnp;
2032 struct vattr *a_vap;
2033 char *a_target;
2034 } */ *ap = v;
2035 struct vnode *dvp = ap->a_dvp;
2036 struct vattr *vap = ap->a_vap;
2037 struct componentname *cnp = ap->a_cnp;
2038 struct nfsv2_sattr *sp;
2039 u_int32_t *tl;
2040 caddr_t cp;
2041 int32_t t1, t2;
2042 caddr_t bpos, dpos, cp2;
2043 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2044 struct mbuf *mreq, *mrep, *md, *mb;
2045 struct vnode *newvp = (struct vnode *)0;
2046 const int v3 = NFS_ISV3(dvp);
2047 struct nfsnode *dnp = VTONFS(dvp);
2048
2049 *ap->a_vpp = NULL;
2050 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2051 slen = strlen(ap->a_target);
2052 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2053 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2054 nfsm_fhtom(dnp, v3);
2055 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2056 if (v3)
2057 nfsm_v3attrbuild(vap, FALSE);
2058 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2059 if (!v3) {
2060 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2061 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2062 sp->sa_uid = nfs_xdrneg1;
2063 sp->sa_gid = nfs_xdrneg1;
2064 sp->sa_size = nfs_xdrneg1;
2065 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2066 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2067 }
2068 nfsm_request(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
2069 if (v3) {
2070 if (!error)
2071 nfsm_mtofh(dvp, newvp, v3, gotvp);
2072 nfsm_wcc_data(dvp, wccflag, 0);
2073 }
2074 nfsm_reqdone;
2075 /*
2076 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2077 */
2078 if (error == EEXIST)
2079 error = 0;
2080 if (error == 0 && newvp == NULL) {
2081 struct nfsnode *np = NULL;
2082
2083 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2084 cnp->cn_cred, cnp->cn_proc, &np);
2085 if (error == 0)
2086 newvp = NFSTOV(np);
2087 }
2088 if (error) {
2089 if (newvp != NULL)
2090 vput(newvp);
2091 } else {
2092 *ap->a_vpp = newvp;
2093 }
2094 PNBUF_PUT(cnp->cn_pnbuf);
2095 VTONFS(dvp)->n_flag |= NMODIFIED;
2096 if (!wccflag)
2097 VTONFS(dvp)->n_attrstamp = 0;
2098 VN_KNOTE(dvp, NOTE_WRITE);
2099 vput(dvp);
2100 return (error);
2101 }
2102
2103 /*
2104 * nfs make dir call
2105 */
2106 int
2107 nfs_mkdir(v)
2108 void *v;
2109 {
2110 struct vop_mkdir_args /* {
2111 struct vnode *a_dvp;
2112 struct vnode **a_vpp;
2113 struct componentname *a_cnp;
2114 struct vattr *a_vap;
2115 } */ *ap = v;
2116 struct vnode *dvp = ap->a_dvp;
2117 struct vattr *vap = ap->a_vap;
2118 struct componentname *cnp = ap->a_cnp;
2119 struct nfsv2_sattr *sp;
2120 u_int32_t *tl;
2121 caddr_t cp;
2122 int32_t t1, t2;
2123 int len;
2124 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2125 struct vnode *newvp = (struct vnode *)0;
2126 caddr_t bpos, dpos, cp2;
2127 int error = 0, wccflag = NFSV3_WCCRATTR;
2128 int gotvp = 0;
2129 struct mbuf *mreq, *mrep, *md, *mb;
2130 const int v3 = NFS_ISV3(dvp);
2131
2132 len = cnp->cn_namelen;
2133 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2134 nfsm_reqhead(dnp, NFSPROC_MKDIR,
2135 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2136 nfsm_fhtom(dnp, v3);
2137 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2138 if (v3) {
2139 nfsm_v3attrbuild(vap, FALSE);
2140 } else {
2141 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2142 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2143 sp->sa_uid = nfs_xdrneg1;
2144 sp->sa_gid = nfs_xdrneg1;
2145 sp->sa_size = nfs_xdrneg1;
2146 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2147 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2148 }
2149 nfsm_request(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
2150 if (!error)
2151 nfsm_mtofh(dvp, newvp, v3, gotvp);
2152 if (v3)
2153 nfsm_wcc_data(dvp, wccflag, 0);
2154 nfsm_reqdone;
2155 VTONFS(dvp)->n_flag |= NMODIFIED;
2156 if (!wccflag)
2157 VTONFS(dvp)->n_attrstamp = 0;
2158 /*
2159 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2160 * if we can succeed in looking up the directory.
2161 */
2162 if (error == EEXIST || (!error && !gotvp)) {
2163 if (newvp) {
2164 vput(newvp);
2165 newvp = (struct vnode *)0;
2166 }
2167 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2168 cnp->cn_proc, &np);
2169 if (!error) {
2170 newvp = NFSTOV(np);
2171 if (newvp->v_type != VDIR)
2172 error = EEXIST;
2173 }
2174 }
2175 if (error) {
2176 if (newvp)
2177 vput(newvp);
2178 } else {
2179 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2180 if (cnp->cn_flags & MAKEENTRY)
2181 cache_enter(dvp, newvp, cnp);
2182 *ap->a_vpp = newvp;
2183 }
2184 PNBUF_PUT(cnp->cn_pnbuf);
2185 vput(dvp);
2186 return (error);
2187 }
2188
2189 /*
2190 * nfs remove directory call
2191 */
2192 int
2193 nfs_rmdir(v)
2194 void *v;
2195 {
2196 struct vop_rmdir_args /* {
2197 struct vnode *a_dvp;
2198 struct vnode *a_vp;
2199 struct componentname *a_cnp;
2200 } */ *ap = v;
2201 struct vnode *vp = ap->a_vp;
2202 struct vnode *dvp = ap->a_dvp;
2203 struct componentname *cnp = ap->a_cnp;
2204 u_int32_t *tl;
2205 caddr_t cp;
2206 int32_t t1, t2;
2207 caddr_t bpos, dpos, cp2;
2208 int error = 0, wccflag = NFSV3_WCCRATTR;
2209 struct mbuf *mreq, *mrep, *md, *mb;
2210 const int v3 = NFS_ISV3(dvp);
2211 struct nfsnode *dnp;
2212
2213 if (dvp == vp) {
2214 vrele(dvp);
2215 vput(dvp);
2216 PNBUF_PUT(cnp->cn_pnbuf);
2217 return (EINVAL);
2218 }
2219 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2220 dnp = VTONFS(dvp);
2221 nfsm_reqhead(dnp, NFSPROC_RMDIR,
2222 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2223 nfsm_fhtom(dnp, v3);
2224 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2225 nfsm_request(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
2226 if (v3)
2227 nfsm_wcc_data(dvp, wccflag, 0);
2228 nfsm_reqdone;
2229 PNBUF_PUT(cnp->cn_pnbuf);
2230 VTONFS(dvp)->n_flag |= NMODIFIED;
2231 if (!wccflag)
2232 VTONFS(dvp)->n_attrstamp = 0;
2233 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2234 VN_KNOTE(vp, NOTE_DELETE);
2235 cache_purge(dvp);
2236 cache_purge(vp);
2237 vput(vp);
2238 vput(dvp);
2239 /*
2240 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2241 */
2242 if (error == ENOENT)
2243 error = 0;
2244 return (error);
2245 }
2246
2247 /*
2248 * nfs readdir call
2249 */
2250 int
2251 nfs_readdir(v)
2252 void *v;
2253 {
2254 struct vop_readdir_args /* {
2255 struct vnode *a_vp;
2256 struct uio *a_uio;
2257 struct ucred *a_cred;
2258 int *a_eofflag;
2259 off_t **a_cookies;
2260 int *a_ncookies;
2261 } */ *ap = v;
2262 struct vnode *vp = ap->a_vp;
2263 struct uio *uio = ap->a_uio;
2264 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2265 char *base = uio->uio_iov->iov_base;
2266 int tresid, error;
2267 size_t count, lost;
2268 struct dirent *dp;
2269 off_t *cookies = NULL;
2270 int ncookies = 0, nc;
2271
2272 if (vp->v_type != VDIR)
2273 return (EPERM);
2274
2275 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2276 count = uio->uio_resid - lost;
2277 if (count <= 0)
2278 return (EINVAL);
2279
2280 /*
2281 * Call nfs_bioread() to do the real work.
2282 */
2283 tresid = uio->uio_resid = count;
2284 error = nfs_bioread(vp, uio, 0, ap->a_cred,
2285 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2286
2287 if (!error && ap->a_cookies) {
2288 ncookies = count / 16;
2289 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2290 *ap->a_cookies = cookies;
2291 }
2292
2293 if (!error && uio->uio_resid == tresid) {
2294 uio->uio_resid += lost;
2295 nfsstats.direofcache_misses++;
2296 if (ap->a_cookies)
2297 *ap->a_ncookies = 0;
2298 *ap->a_eofflag = 1;
2299 return (0);
2300 }
2301
2302 if (!error && ap->a_cookies) {
2303 /*
2304 * Only the NFS server and emulations use cookies, and they
2305 * load the directory block into system space, so we can
2306 * just look at it directly.
2307 */
2308 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2309 panic("nfs_readdir: lost in space");
2310 for (nc = 0; ncookies-- &&
2311 base < (char *)uio->uio_iov->iov_base; nc++){
2312 dp = (struct dirent *) base;
2313 if (dp->d_reclen == 0)
2314 break;
2315 if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2316 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2317 else
2318 *(cookies++) = NFS_GETCOOKIE(dp);
2319 base += dp->d_reclen;
2320 }
2321 uio->uio_resid +=
2322 ((caddr_t)uio->uio_iov->iov_base - base);
2323 uio->uio_iov->iov_len +=
2324 ((caddr_t)uio->uio_iov->iov_base - base);
2325 uio->uio_iov->iov_base = base;
2326 *ap->a_ncookies = nc;
2327 }
2328
2329 uio->uio_resid += lost;
2330 *ap->a_eofflag = 0;
2331 return (error);
2332 }
2333
2334 /*
2335 * Readdir rpc call.
2336 * Called from below the buffer cache by nfs_doio().
2337 */
2338 int
2339 nfs_readdirrpc(vp, uiop, cred)
2340 struct vnode *vp;
2341 struct uio *uiop;
2342 struct ucred *cred;
2343 {
2344 int len, left;
2345 struct dirent *dp = NULL;
2346 u_int32_t *tl;
2347 caddr_t cp;
2348 int32_t t1, t2;
2349 caddr_t bpos, dpos, cp2;
2350 struct mbuf *mreq, *mrep, *md, *mb;
2351 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2352 struct nfsnode *dnp = VTONFS(vp);
2353 u_quad_t fileno;
2354 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2355 int attrflag, nrpcs = 0, reclen;
2356 const int v3 = NFS_ISV3(vp);
2357
2358 #ifdef DIAGNOSTIC
2359 /*
2360 * Should be called from buffer cache, so only amount of
2361 * NFS_DIRBLKSIZ will be requested.
2362 */
2363 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2364 panic("nfs readdirrpc bad uio");
2365 #endif
2366
2367 /*
2368 * Loop around doing readdir rpc's of size nm_readdirsize
2369 * truncated to a multiple of NFS_DIRFRAGSIZ.
2370 * The stopping criteria is EOF or buffer full.
2371 */
2372 while (more_dirs && bigenough) {
2373 /*
2374 * Heuristic: don't bother to do another RPC to further
2375 * fill up this block if there is not much room left. (< 50%
2376 * of the readdir RPC size). This wastes some buffer space
2377 * but can save up to 50% in RPC calls.
2378 */
2379 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2380 bigenough = 0;
2381 break;
2382 }
2383 nfsstats.rpccnt[NFSPROC_READDIR]++;
2384 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2385 NFSX_READDIR(v3));
2386 nfsm_fhtom(dnp, v3);
2387 if (v3) {
2388 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2389 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2390 txdr_swapcookie3(uiop->uio_offset, tl);
2391 } else {
2392 txdr_cookie3(uiop->uio_offset, tl);
2393 }
2394 tl += 2;
2395 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2396 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2397 } else {
2398 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2399 *tl++ = txdr_unsigned(uiop->uio_offset);
2400 }
2401 *tl = txdr_unsigned(nmp->nm_readdirsize);
2402 nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred);
2403 nrpcs++;
2404 if (v3) {
2405 nfsm_postop_attr(vp, attrflag, 0);
2406 if (!error) {
2407 nfsm_dissect(tl, u_int32_t *,
2408 2 * NFSX_UNSIGNED);
2409 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2410 dnp->n_cookieverf.nfsuquad[1] = *tl;
2411 } else {
2412 m_freem(mrep);
2413 goto nfsmout;
2414 }
2415 }
2416 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2417 more_dirs = fxdr_unsigned(int, *tl);
2418
2419 /* loop thru the dir entries, doctoring them to 4bsd form */
2420 while (more_dirs && bigenough) {
2421 if (v3) {
2422 nfsm_dissect(tl, u_int32_t *,
2423 3 * NFSX_UNSIGNED);
2424 fileno = fxdr_hyper(tl);
2425 len = fxdr_unsigned(int, *(tl + 2));
2426 } else {
2427 nfsm_dissect(tl, u_int32_t *,
2428 2 * NFSX_UNSIGNED);
2429 fileno = fxdr_unsigned(u_quad_t, *tl++);
2430 len = fxdr_unsigned(int, *tl);
2431 }
2432 if (len <= 0 || len > NFS_MAXNAMLEN) {
2433 error = EBADRPC;
2434 m_freem(mrep);
2435 goto nfsmout;
2436 }
2437 tlen = nfsm_rndup(len);
2438 if (tlen == len)
2439 tlen += 4; /* To ensure null termination */
2440 tlen += sizeof (off_t) + sizeof (int);
2441 reclen = ALIGN(tlen + DIRHDSIZ);
2442 tlen = reclen - DIRHDSIZ;
2443 left = NFS_DIRFRAGSIZ - blksiz;
2444 if (reclen > left) {
2445 dp->d_reclen += left;
2446 uiop->uio_iov->iov_base =
2447 (caddr_t)uiop->uio_iov->iov_base + left;
2448 uiop->uio_iov->iov_len -= left;
2449 uiop->uio_resid -= left;
2450 blksiz = 0;
2451 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2452 }
2453 if (reclen > uiop->uio_resid)
2454 bigenough = 0;
2455 if (bigenough) {
2456 dp = (struct dirent *)uiop->uio_iov->iov_base;
2457 dp->d_fileno = (int)fileno;
2458 dp->d_namlen = len;
2459 dp->d_reclen = reclen;
2460 dp->d_type = DT_UNKNOWN;
2461 blksiz += dp->d_reclen;
2462 if (blksiz == NFS_DIRFRAGSIZ)
2463 blksiz = 0;
2464 uiop->uio_resid -= DIRHDSIZ;
2465 uiop->uio_iov->iov_base =
2466 (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ;
2467 uiop->uio_iov->iov_len -= DIRHDSIZ;
2468 nfsm_mtouio(uiop, len);
2469 cp = uiop->uio_iov->iov_base;
2470 tlen -= len;
2471 *cp = '\0'; /* null terminate */
2472 uiop->uio_iov->iov_base =
2473 (caddr_t)uiop->uio_iov->iov_base + tlen;
2474 uiop->uio_iov->iov_len -= tlen;
2475 uiop->uio_resid -= tlen;
2476 } else
2477 nfsm_adv(nfsm_rndup(len));
2478 if (v3) {
2479 nfsm_dissect(tl, u_int32_t *,
2480 3 * NFSX_UNSIGNED);
2481 } else {
2482 nfsm_dissect(tl, u_int32_t *,
2483 2 * NFSX_UNSIGNED);
2484 }
2485 if (bigenough) {
2486 if (v3) {
2487 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2488 uiop->uio_offset =
2489 fxdr_swapcookie3(tl);
2490 else
2491 uiop->uio_offset =
2492 fxdr_cookie3(tl);
2493 }
2494 else {
2495 uiop->uio_offset =
2496 fxdr_unsigned(off_t, *tl);
2497 }
2498 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2499 }
2500 if (v3)
2501 tl += 2;
2502 else
2503 tl++;
2504 more_dirs = fxdr_unsigned(int, *tl);
2505 }
2506 /*
2507 * If at end of rpc data, get the eof boolean
2508 */
2509 if (!more_dirs) {
2510 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2511 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2512 }
2513 m_freem(mrep);
2514 }
2515 /*
2516 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2517 * by increasing d_reclen for the last record.
2518 */
2519 if (blksiz > 0) {
2520 left = NFS_DIRFRAGSIZ - blksiz;
2521 dp->d_reclen += left;
2522 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2523 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2524 left;
2525 uiop->uio_iov->iov_len -= left;
2526 uiop->uio_resid -= left;
2527 }
2528
2529 /*
2530 * We are now either at the end of the directory or have filled the
2531 * block.
2532 */
2533 if (bigenough)
2534 dnp->n_direofoffset = uiop->uio_offset;
2535 nfsmout:
2536 return (error);
2537 }
2538
2539 /*
2540 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2541 */
2542 int
2543 nfs_readdirplusrpc(vp, uiop, cred)
2544 struct vnode *vp;
2545 struct uio *uiop;
2546 struct ucred *cred;
2547 {
2548 int len, left;
2549 struct dirent *dp = NULL;
2550 u_int32_t *tl;
2551 caddr_t cp;
2552 int32_t t1, t2;
2553 struct vnode *newvp;
2554 caddr_t bpos, dpos, cp2;
2555 struct mbuf *mreq, *mrep, *md, *mb;
2556 struct nameidata nami, *ndp = &nami;
2557 struct componentname *cnp = &ndp->ni_cnd;
2558 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2559 struct nfsnode *dnp = VTONFS(vp), *np;
2560 nfsfh_t *fhp;
2561 u_quad_t fileno;
2562 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2563 int attrflag, fhsize, nrpcs = 0, reclen;
2564 struct nfs_fattr fattr, *fp;
2565
2566 #ifdef DIAGNOSTIC
2567 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2568 panic("nfs readdirplusrpc bad uio");
2569 #endif
2570 ndp->ni_dvp = vp;
2571 newvp = NULLVP;
2572
2573 /*
2574 * Loop around doing readdir rpc's of size nm_readdirsize
2575 * truncated to a multiple of NFS_DIRFRAGSIZ.
2576 * The stopping criteria is EOF or buffer full.
2577 */
2578 while (more_dirs && bigenough) {
2579 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2580 bigenough = 0;
2581 break;
2582 }
2583 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2584 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2585 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2586 nfsm_fhtom(dnp, 1);
2587 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2588 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2589 txdr_swapcookie3(uiop->uio_offset, tl);
2590 } else {
2591 txdr_cookie3(uiop->uio_offset, tl);
2592 }
2593 tl += 2;
2594 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2595 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2596 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2597 *tl = txdr_unsigned(nmp->nm_rsize);
2598 nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2599 nfsm_postop_attr(vp, attrflag, 0);
2600 if (error) {
2601 m_freem(mrep);
2602 goto nfsmout;
2603 }
2604 nrpcs++;
2605 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2606 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2607 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2608 more_dirs = fxdr_unsigned(int, *tl);
2609
2610 /* loop thru the dir entries, doctoring them to 4bsd form */
2611 while (more_dirs && bigenough) {
2612 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2613 fileno = fxdr_hyper(tl);
2614 len = fxdr_unsigned(int, *(tl + 2));
2615 if (len <= 0 || len > NFS_MAXNAMLEN) {
2616 error = EBADRPC;
2617 m_freem(mrep);
2618 goto nfsmout;
2619 }
2620 tlen = nfsm_rndup(len);
2621 if (tlen == len)
2622 tlen += 4; /* To ensure null termination*/
2623 tlen += sizeof (off_t) + sizeof (int);
2624 reclen = ALIGN(tlen + DIRHDSIZ);
2625 tlen = reclen - DIRHDSIZ;
2626 left = NFS_DIRFRAGSIZ - blksiz;
2627 if (reclen > left) {
2628 /*
2629 * DIRFRAGSIZ is aligned, no need to align
2630 * again here.
2631 */
2632 dp->d_reclen += left;
2633 uiop->uio_iov->iov_base =
2634 (caddr_t)uiop->uio_iov->iov_base + left;
2635 uiop->uio_iov->iov_len -= left;
2636 uiop->uio_resid -= left;
2637 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2638 blksiz = 0;
2639 }
2640 if (reclen > uiop->uio_resid)
2641 bigenough = 0;
2642 if (bigenough) {
2643 dp = (struct dirent *)uiop->uio_iov->iov_base;
2644 dp->d_fileno = (int)fileno;
2645 dp->d_namlen = len;
2646 dp->d_reclen = reclen;
2647 dp->d_type = DT_UNKNOWN;
2648 blksiz += dp->d_reclen;
2649 if (blksiz == NFS_DIRFRAGSIZ)
2650 blksiz = 0;
2651 uiop->uio_resid -= DIRHDSIZ;
2652 uiop->uio_iov->iov_base =
2653 (caddr_t)uiop->uio_iov->iov_base +
2654 DIRHDSIZ;
2655 uiop->uio_iov->iov_len -= DIRHDSIZ;
2656 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2657 cnp->cn_namelen = len;
2658 nfsm_mtouio(uiop, len);
2659 cp = uiop->uio_iov->iov_base;
2660 tlen -= len;
2661 *cp = '\0';
2662 uiop->uio_iov->iov_base =
2663 (caddr_t)uiop->uio_iov->iov_base + tlen;
2664 uiop->uio_iov->iov_len -= tlen;
2665 uiop->uio_resid -= tlen;
2666 } else
2667 nfsm_adv(nfsm_rndup(len));
2668 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2669 if (bigenough) {
2670 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2671 uiop->uio_offset =
2672 fxdr_swapcookie3(tl);
2673 else
2674 uiop->uio_offset =
2675 fxdr_cookie3(tl);
2676 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2677 }
2678 tl += 2;
2679
2680 /*
2681 * Since the attributes are before the file handle
2682 * (sigh), we must skip over the attributes and then
2683 * come back and get them.
2684 */
2685 attrflag = fxdr_unsigned(int, *tl);
2686 if (attrflag) {
2687 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2688 memcpy(&fattr, fp, NFSX_V3FATTR);
2689 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2690 doit = fxdr_unsigned(int, *tl);
2691 if (doit) {
2692 nfsm_getfh(fhp, fhsize, 1);
2693 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2694 VREF(vp);
2695 newvp = vp;
2696 np = dnp;
2697 } else {
2698 error = nfs_nget(vp->v_mount, fhp,
2699 fhsize, &np);
2700 if (!error)
2701 newvp = NFSTOV(np);
2702 }
2703 if (!error) {
2704 const char *cp;
2705
2706 nfs_loadattrcache(&newvp, &fattr, 0, 0);
2707 dp->d_type =
2708 IFTODT(VTTOIF(np->n_vattr->va_type));
2709 ndp->ni_vp = newvp;
2710 cp = cnp->cn_nameptr + cnp->cn_namelen;
2711 cnp->cn_hash =
2712 namei_hash(cnp->cn_nameptr, &cp);
2713 if (cnp->cn_namelen <= NCHNAMLEN)
2714 cache_enter(ndp->ni_dvp, ndp->ni_vp,
2715 cnp);
2716 }
2717 }
2718 } else {
2719 /* Just skip over the file handle */
2720 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2721 i = fxdr_unsigned(int, *tl);
2722 nfsm_adv(nfsm_rndup(i));
2723 }
2724 if (newvp != NULLVP) {
2725 if (newvp == vp)
2726 vrele(newvp);
2727 else
2728 vput(newvp);
2729 newvp = NULLVP;
2730 }
2731 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2732 more_dirs = fxdr_unsigned(int, *tl);
2733 }
2734 /*
2735 * If at end of rpc data, get the eof boolean
2736 */
2737 if (!more_dirs) {
2738 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2739 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2740 }
2741 m_freem(mrep);
2742 }
2743 /*
2744 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2745 * by increasing d_reclen for the last record.
2746 */
2747 if (blksiz > 0) {
2748 left = NFS_DIRFRAGSIZ - blksiz;
2749 dp->d_reclen += left;
2750 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2751 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2752 left;
2753 uiop->uio_iov->iov_len -= left;
2754 uiop->uio_resid -= left;
2755 }
2756
2757 /*
2758 * We are now either at the end of the directory or have filled the
2759 * block.
2760 */
2761 if (bigenough)
2762 dnp->n_direofoffset = uiop->uio_offset;
2763 nfsmout:
2764 if (newvp != NULLVP) {
2765 if(newvp == vp)
2766 vrele(newvp);
2767 else
2768 vput(newvp);
2769 }
2770 return (error);
2771 }
2772 static char hextoasc[] = "0123456789abcdef";
2773
2774 /*
2775 * Silly rename. To make the NFS filesystem that is stateless look a little
2776 * more like the "ufs" a remove of an active vnode is translated to a rename
2777 * to a funny looking filename that is removed by nfs_inactive on the
2778 * nfsnode. There is the potential for another process on a different client
2779 * to create the same funny name between the nfs_lookitup() fails and the
2780 * nfs_rename() completes, but...
2781 */
2782 int
2783 nfs_sillyrename(dvp, vp, cnp)
2784 struct vnode *dvp, *vp;
2785 struct componentname *cnp;
2786 {
2787 struct sillyrename *sp;
2788 struct nfsnode *np;
2789 int error;
2790 short pid;
2791
2792 cache_purge(dvp);
2793 np = VTONFS(vp);
2794 #ifndef DIAGNOSTIC
2795 if (vp->v_type == VDIR)
2796 panic("nfs: sillyrename dir");
2797 #endif
2798 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2799 M_NFSREQ, M_WAITOK);
2800 sp->s_cred = crdup(cnp->cn_cred);
2801 sp->s_dvp = dvp;
2802 VREF(dvp);
2803
2804 /* Fudge together a funny name */
2805 pid = cnp->cn_proc->p_pid;
2806 memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2807 sp->s_namlen = 12;
2808 sp->s_name[8] = hextoasc[pid & 0xf];
2809 sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
2810 sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
2811 sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
2812
2813 /* Try lookitups until we get one that isn't there */
2814 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2815 cnp->cn_proc, (struct nfsnode **)0) == 0) {
2816 sp->s_name[4]++;
2817 if (sp->s_name[4] > 'z') {
2818 error = EINVAL;
2819 goto bad;
2820 }
2821 }
2822 error = nfs_renameit(dvp, cnp, sp);
2823 if (error)
2824 goto bad;
2825 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2826 cnp->cn_proc, &np);
2827 np->n_sillyrename = sp;
2828 return (0);
2829 bad:
2830 vrele(sp->s_dvp);
2831 crfree(sp->s_cred);
2832 free((caddr_t)sp, M_NFSREQ);
2833 return (error);
2834 }
2835
2836 /*
2837 * Look up a file name and optionally either update the file handle or
2838 * allocate an nfsnode, depending on the value of npp.
2839 * npp == NULL --> just do the lookup
2840 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2841 * handled too
2842 * *npp != NULL --> update the file handle in the vnode
2843 */
2844 int
2845 nfs_lookitup(dvp, name, len, cred, procp, npp)
2846 struct vnode *dvp;
2847 const char *name;
2848 int len;
2849 struct ucred *cred;
2850 struct proc *procp;
2851 struct nfsnode **npp;
2852 {
2853 u_int32_t *tl;
2854 caddr_t cp;
2855 int32_t t1, t2;
2856 struct vnode *newvp = (struct vnode *)0;
2857 struct nfsnode *np, *dnp = VTONFS(dvp);
2858 caddr_t bpos, dpos, cp2;
2859 int error = 0, fhlen, attrflag;
2860 struct mbuf *mreq, *mrep, *md, *mb;
2861 nfsfh_t *nfhp;
2862 const int v3 = NFS_ISV3(dvp);
2863
2864 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2865 nfsm_reqhead(dnp, NFSPROC_LOOKUP,
2866 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2867 nfsm_fhtom(dnp, v3);
2868 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2869 nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred);
2870 if (npp && !error) {
2871 nfsm_getfh(nfhp, fhlen, v3);
2872 if (*npp) {
2873 np = *npp;
2874 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2875 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2876 np->n_fhp = &np->n_fh;
2877 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2878 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2879 memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
2880 np->n_fhsize = fhlen;
2881 newvp = NFSTOV(np);
2882 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2883 VREF(dvp);
2884 newvp = dvp;
2885 } else {
2886 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2887 if (error) {
2888 m_freem(mrep);
2889 return (error);
2890 }
2891 newvp = NFSTOV(np);
2892 }
2893 if (v3) {
2894 nfsm_postop_attr(newvp, attrflag, 0);
2895 if (!attrflag && *npp == NULL) {
2896 m_freem(mrep);
2897 vput(newvp);
2898 return (ENOENT);
2899 }
2900 } else
2901 nfsm_loadattr(newvp, (struct vattr *)0, 0);
2902 }
2903 nfsm_reqdone;
2904 if (npp && *npp == NULL) {
2905 if (error) {
2906 if (newvp)
2907 vput(newvp);
2908 } else
2909 *npp = np;
2910 }
2911 return (error);
2912 }
2913
2914 /*
2915 * Nfs Version 3 commit rpc
2916 */
2917 int
2918 nfs_commit(vp, offset, cnt, procp)
2919 struct vnode *vp;
2920 off_t offset;
2921 uint32_t cnt;
2922 struct proc *procp;
2923 {
2924 caddr_t cp;
2925 u_int32_t *tl;
2926 int32_t t1, t2;
2927 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2928 caddr_t bpos, dpos, cp2;
2929 int error = 0, wccflag = NFSV3_WCCRATTR;
2930 struct mbuf *mreq, *mrep, *md, *mb;
2931 struct nfsnode *np;
2932
2933 KASSERT(NFS_ISV3(vp));
2934
2935 #ifdef NFS_DEBUG_COMMIT
2936 printf("commit %lu - %lu\n", (unsigned long)offset,
2937 (unsigned long)(offset + cnt));
2938 #endif
2939
2940 simple_lock(&nmp->nm_slock);
2941 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
2942 simple_unlock(&nmp->nm_slock);
2943 return (0);
2944 }
2945 simple_unlock(&nmp->nm_slock);
2946 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2947 np = VTONFS(vp);
2948 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
2949 nfsm_fhtom(np, 1);
2950 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2951 txdr_hyper(offset, tl);
2952 tl += 2;
2953 *tl = txdr_unsigned(cnt);
2954 nfsm_request(np, NFSPROC_COMMIT, procp, VTONFS(vp)->n_wcred);
2955 nfsm_wcc_data(vp, wccflag, 0);
2956 if (!error) {
2957 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2958 simple_lock(&nmp->nm_slock);
2959 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
2960 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
2961 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
2962 error = NFSERR_STALEWRITEVERF;
2963 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
2964 }
2965 simple_unlock(&nmp->nm_slock);
2966 }
2967 nfsm_reqdone;
2968 return (error);
2969 }
2970
2971 /*
2972 * Kludge City..
2973 * - make nfs_bmap() essentially a no-op that does no translation
2974 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2975 * (Maybe I could use the process's page mapping, but I was concerned that
2976 * Kernel Write might not be enabled and also figured copyout() would do
2977 * a lot more work than memcpy() and also it currently happens in the
2978 * context of the swapper process (2).
2979 */
2980 int
2981 nfs_bmap(v)
2982 void *v;
2983 {
2984 struct vop_bmap_args /* {
2985 struct vnode *a_vp;
2986 daddr_t a_bn;
2987 struct vnode **a_vpp;
2988 daddr_t *a_bnp;
2989 int *a_runp;
2990 } */ *ap = v;
2991 struct vnode *vp = ap->a_vp;
2992 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
2993
2994 if (ap->a_vpp != NULL)
2995 *ap->a_vpp = vp;
2996 if (ap->a_bnp != NULL)
2997 *ap->a_bnp = ap->a_bn << bshift;
2998 if (ap->a_runp != NULL)
2999 *ap->a_runp = 1024 * 1024; /* XXX */
3000 return (0);
3001 }
3002
3003 /*
3004 * Strategy routine.
3005 * For async requests when nfsiod(s) are running, queue the request by
3006 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3007 * request.
3008 */
3009 int
3010 nfs_strategy(v)
3011 void *v;
3012 {
3013 struct vop_strategy_args *ap = v;
3014 struct buf *bp = ap->a_bp;
3015 struct proc *p;
3016 int error = 0;
3017
3018 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3019 panic("nfs physio/async");
3020 if (bp->b_flags & B_ASYNC)
3021 p = NULL;
3022 else
3023 p = curproc; /* XXX */
3024
3025 /*
3026 * If the op is asynchronous and an i/o daemon is waiting
3027 * queue the request, wake it up and wait for completion
3028 * otherwise just do it ourselves.
3029 */
3030
3031 if ((bp->b_flags & B_ASYNC) == 0 ||
3032 nfs_asyncio(bp))
3033 error = nfs_doio(bp, p);
3034 return (error);
3035 }
3036
3037 /*
3038 * fsync vnode op. Just call nfs_flush() with commit == 1.
3039 */
3040 /* ARGSUSED */
3041 int
3042 nfs_fsync(v)
3043 void *v;
3044 {
3045 struct vop_fsync_args /* {
3046 struct vnodeop_desc *a_desc;
3047 struct vnode * a_vp;
3048 struct ucred * a_cred;
3049 int a_flags;
3050 off_t offlo;
3051 off_t offhi;
3052 struct proc * a_p;
3053 } */ *ap = v;
3054
3055 return (nfs_flush(ap->a_vp, ap->a_cred,
3056 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1));
3057 }
3058
3059 /*
3060 * Flush all the data associated with a vnode.
3061 */
3062 int
3063 nfs_flush(vp, cred, waitfor, p, commit)
3064 struct vnode *vp;
3065 struct ucred *cred;
3066 int waitfor;
3067 struct proc *p;
3068 int commit;
3069 {
3070 struct nfsnode *np = VTONFS(vp);
3071 int error;
3072 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3073 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3074
3075 simple_lock(&vp->v_interlock);
3076 error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3077 if (np->n_flag & NWRITEERR) {
3078 error = np->n_error;
3079 np->n_flag &= ~NWRITEERR;
3080 }
3081 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3082 return (error);
3083 }
3084
3085 /*
3086 * Return POSIX pathconf information applicable to nfs.
3087 *
3088 * N.B. The NFS V2 protocol doesn't support this RPC.
3089 */
3090 /* ARGSUSED */
3091 int
3092 nfs_pathconf(v)
3093 void *v;
3094 {
3095 struct vop_pathconf_args /* {
3096 struct vnode *a_vp;
3097 int a_name;
3098 register_t *a_retval;
3099 } */ *ap = v;
3100 struct nfsv3_pathconf *pcp;
3101 struct vnode *vp = ap->a_vp;
3102 struct mbuf *mreq, *mrep, *md, *mb;
3103 int32_t t1, t2;
3104 u_int32_t *tl;
3105 caddr_t bpos, dpos, cp, cp2;
3106 int error = 0, attrflag;
3107 #ifndef NFS_V2_ONLY
3108 struct nfsmount *nmp;
3109 unsigned int l;
3110 u_int64_t maxsize;
3111 #endif
3112 const int v3 = NFS_ISV3(vp);
3113 struct nfsnode *np = VTONFS(vp);
3114
3115 switch (ap->a_name) {
3116 /* Names that can be resolved locally. */
3117 case _PC_PIPE_BUF:
3118 *ap->a_retval = PIPE_BUF;
3119 break;
3120 case _PC_SYNC_IO:
3121 *ap->a_retval = 1;
3122 break;
3123 /* Names that cannot be resolved locally; do an RPC, if possible. */
3124 case _PC_LINK_MAX:
3125 case _PC_NAME_MAX:
3126 case _PC_CHOWN_RESTRICTED:
3127 case _PC_NO_TRUNC:
3128 if (!v3) {
3129 error = EINVAL;
3130 break;
3131 }
3132 nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3133 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3134 nfsm_fhtom(np, 1);
3135 nfsm_request(np, NFSPROC_PATHCONF,
3136 curproc, curproc->p_ucred); /* XXX */
3137 nfsm_postop_attr(vp, attrflag, 0);
3138 if (!error) {
3139 nfsm_dissect(pcp, struct nfsv3_pathconf *,
3140 NFSX_V3PATHCONF);
3141 switch (ap->a_name) {
3142 case _PC_LINK_MAX:
3143 *ap->a_retval =
3144 fxdr_unsigned(register_t, pcp->pc_linkmax);
3145 break;
3146 case _PC_NAME_MAX:
3147 *ap->a_retval =
3148 fxdr_unsigned(register_t, pcp->pc_namemax);
3149 break;
3150 case _PC_CHOWN_RESTRICTED:
3151 *ap->a_retval =
3152 (pcp->pc_chownrestricted == nfs_true);
3153 break;
3154 case _PC_NO_TRUNC:
3155 *ap->a_retval =
3156 (pcp->pc_notrunc == nfs_true);
3157 break;
3158 }
3159 }
3160 nfsm_reqdone;
3161 break;
3162 case _PC_FILESIZEBITS:
3163 #ifndef NFS_V2_ONLY
3164 if (v3) {
3165 nmp = VFSTONFS(vp->v_mount);
3166 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3167 if ((error = nfs_fsinfo(nmp, vp,
3168 curproc->p_ucred, curproc)) != 0) /* XXX */
3169 break;
3170 for (l = 0, maxsize = nmp->nm_maxfilesize;
3171 (maxsize >> l) > 0; l++)
3172 ;
3173 *ap->a_retval = l + 1;
3174 } else
3175 #endif
3176 {
3177 *ap->a_retval = 32; /* NFS V2 limitation */
3178 }
3179 break;
3180 default:
3181 error = EINVAL;
3182 break;
3183 }
3184
3185 return (error);
3186 }
3187
3188 /*
3189 * NFS advisory byte-level locks.
3190 */
3191 int
3192 nfs_advlock(v)
3193 void *v;
3194 {
3195 struct vop_advlock_args /* {
3196 struct vnode *a_vp;
3197 caddr_t a_id;
3198 int a_op;
3199 struct flock *a_fl;
3200 int a_flags;
3201 } */ *ap = v;
3202 struct nfsnode *np = VTONFS(ap->a_vp);
3203
3204 return lf_advlock(ap, &np->n_lockf, np->n_size);
3205 }
3206
3207 /*
3208 * Print out the contents of an nfsnode.
3209 */
3210 int
3211 nfs_print(v)
3212 void *v;
3213 {
3214 struct vop_print_args /* {
3215 struct vnode *a_vp;
3216 } */ *ap = v;
3217 struct vnode *vp = ap->a_vp;
3218 struct nfsnode *np = VTONFS(vp);
3219
3220 printf("tag VT_NFS, fileid %ld fsid 0x%lx",
3221 np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3222 if (vp->v_type == VFIFO)
3223 fifo_printinfo(vp);
3224 printf("\n");
3225 return (0);
3226 }
3227
3228 /*
3229 * NFS file truncation.
3230 */
3231 int
3232 nfs_truncate(v)
3233 void *v;
3234 {
3235 #if 0
3236 struct vop_truncate_args /* {
3237 struct vnode *a_vp;
3238 off_t a_length;
3239 int a_flags;
3240 struct ucred *a_cred;
3241 struct proc *a_p;
3242 } */ *ap = v;
3243 #endif
3244
3245 /* Use nfs_setattr */
3246 return (EOPNOTSUPP);
3247 }
3248
3249 /*
3250 * NFS update.
3251 */
3252 int
3253 nfs_update(v)
3254 void *v;
3255 #if 0
3256 struct vop_update_args /* {
3257 struct vnode *a_vp;
3258 struct timespec *a_ta;
3259 struct timespec *a_tm;
3260 int a_waitfor;
3261 } */ *ap = v;
3262 #endif
3263 {
3264
3265 /* Use nfs_setattr */
3266 return (EOPNOTSUPP);
3267 }
3268
3269 /*
3270 * Just call bwrite().
3271 */
3272 int
3273 nfs_bwrite(v)
3274 void *v;
3275 {
3276 struct vop_bwrite_args /* {
3277 struct vnode *a_bp;
3278 } */ *ap = v;
3279
3280 return (bwrite(ap->a_bp));
3281 }
3282
3283 /*
3284 * nfs unlock wrapper.
3285 */
3286 int
3287 nfs_unlock(void *v)
3288 {
3289 struct vop_unlock_args /* {
3290 struct vnode *a_vp;
3291 int a_flags;
3292 } */ *ap = v;
3293 struct vnode *vp = ap->a_vp;
3294
3295 /*
3296 * VOP_UNLOCK can be called by nfs_loadattrcache
3297 * with v_data == 0.
3298 */
3299 if (VTONFS(vp)) {
3300 nfs_delayedtruncate(vp);
3301 }
3302
3303 return genfs_unlock(v);
3304 }
3305
3306 /*
3307 * nfs special file access vnode op.
3308 * Essentially just get vattr and then imitate iaccess() since the device is
3309 * local to the client.
3310 */
3311 int
3312 nfsspec_access(v)
3313 void *v;
3314 {
3315 struct vop_access_args /* {
3316 struct vnode *a_vp;
3317 int a_mode;
3318 struct ucred *a_cred;
3319 struct proc *a_p;
3320 } */ *ap = v;
3321 struct vattr va;
3322 struct vnode *vp = ap->a_vp;
3323 int error;
3324
3325 error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3326 if (error)
3327 return (error);
3328
3329 /*
3330 * Disallow write attempts on filesystems mounted read-only;
3331 * unless the file is a socket, fifo, or a block or character
3332 * device resident on the filesystem.
3333 */
3334 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3335 switch (vp->v_type) {
3336 case VREG:
3337 case VDIR:
3338 case VLNK:
3339 return (EROFS);
3340 default:
3341 break;
3342 }
3343 }
3344
3345 return (vaccess(va.va_type, va.va_mode,
3346 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3347 }
3348
3349 /*
3350 * Read wrapper for special devices.
3351 */
3352 int
3353 nfsspec_read(v)
3354 void *v;
3355 {
3356 struct vop_read_args /* {
3357 struct vnode *a_vp;
3358 struct uio *a_uio;
3359 int a_ioflag;
3360 struct ucred *a_cred;
3361 } */ *ap = v;
3362 struct nfsnode *np = VTONFS(ap->a_vp);
3363
3364 /*
3365 * Set access flag.
3366 */
3367 np->n_flag |= NACC;
3368 np->n_atim.tv_sec = time.tv_sec;
3369 np->n_atim.tv_nsec = time.tv_usec * 1000;
3370 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3371 }
3372
3373 /*
3374 * Write wrapper for special devices.
3375 */
3376 int
3377 nfsspec_write(v)
3378 void *v;
3379 {
3380 struct vop_write_args /* {
3381 struct vnode *a_vp;
3382 struct uio *a_uio;
3383 int a_ioflag;
3384 struct ucred *a_cred;
3385 } */ *ap = v;
3386 struct nfsnode *np = VTONFS(ap->a_vp);
3387
3388 /*
3389 * Set update flag.
3390 */
3391 np->n_flag |= NUPD;
3392 np->n_mtim.tv_sec = time.tv_sec;
3393 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3394 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3395 }
3396
3397 /*
3398 * Close wrapper for special devices.
3399 *
3400 * Update the times on the nfsnode then do device close.
3401 */
3402 int
3403 nfsspec_close(v)
3404 void *v;
3405 {
3406 struct vop_close_args /* {
3407 struct vnode *a_vp;
3408 int a_fflag;
3409 struct ucred *a_cred;
3410 struct proc *a_p;
3411 } */ *ap = v;
3412 struct vnode *vp = ap->a_vp;
3413 struct nfsnode *np = VTONFS(vp);
3414 struct vattr vattr;
3415
3416 if (np->n_flag & (NACC | NUPD)) {
3417 np->n_flag |= NCHG;
3418 if (vp->v_usecount == 1 &&
3419 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3420 VATTR_NULL(&vattr);
3421 if (np->n_flag & NACC)
3422 vattr.va_atime = np->n_atim;
3423 if (np->n_flag & NUPD)
3424 vattr.va_mtime = np->n_mtim;
3425 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3426 }
3427 }
3428 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3429 }
3430
3431 /*
3432 * Read wrapper for fifos.
3433 */
3434 int
3435 nfsfifo_read(v)
3436 void *v;
3437 {
3438 struct vop_read_args /* {
3439 struct vnode *a_vp;
3440 struct uio *a_uio;
3441 int a_ioflag;
3442 struct ucred *a_cred;
3443 } */ *ap = v;
3444 struct nfsnode *np = VTONFS(ap->a_vp);
3445
3446 /*
3447 * Set access flag.
3448 */
3449 np->n_flag |= NACC;
3450 np->n_atim.tv_sec = time.tv_sec;
3451 np->n_atim.tv_nsec = time.tv_usec * 1000;
3452 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3453 }
3454
3455 /*
3456 * Write wrapper for fifos.
3457 */
3458 int
3459 nfsfifo_write(v)
3460 void *v;
3461 {
3462 struct vop_write_args /* {
3463 struct vnode *a_vp;
3464 struct uio *a_uio;
3465 int a_ioflag;
3466 struct ucred *a_cred;
3467 } */ *ap = v;
3468 struct nfsnode *np = VTONFS(ap->a_vp);
3469
3470 /*
3471 * Set update flag.
3472 */
3473 np->n_flag |= NUPD;
3474 np->n_mtim.tv_sec = time.tv_sec;
3475 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3476 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3477 }
3478
3479 /*
3480 * Close wrapper for fifos.
3481 *
3482 * Update the times on the nfsnode then do fifo close.
3483 */
3484 int
3485 nfsfifo_close(v)
3486 void *v;
3487 {
3488 struct vop_close_args /* {
3489 struct vnode *a_vp;
3490 int a_fflag;
3491 struct ucred *a_cred;
3492 struct proc *a_p;
3493 } */ *ap = v;
3494 struct vnode *vp = ap->a_vp;
3495 struct nfsnode *np = VTONFS(vp);
3496 struct vattr vattr;
3497
3498 if (np->n_flag & (NACC | NUPD)) {
3499 if (np->n_flag & NACC) {
3500 np->n_atim.tv_sec = time.tv_sec;
3501 np->n_atim.tv_nsec = time.tv_usec * 1000;
3502 }
3503 if (np->n_flag & NUPD) {
3504 np->n_mtim.tv_sec = time.tv_sec;
3505 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3506 }
3507 np->n_flag |= NCHG;
3508 if (vp->v_usecount == 1 &&
3509 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3510 VATTR_NULL(&vattr);
3511 if (np->n_flag & NACC)
3512 vattr.va_atime = np->n_atim;
3513 if (np->n_flag & NUPD)
3514 vattr.va_mtime = np->n_mtim;
3515 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3516 }
3517 }
3518 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3519 }
3520