nfs_vnops.c revision 1.182 1 /* $NetBSD: nfs_vnops.c,v 1.182 2003/10/25 08:48:11 christos Exp $ */
2
3 /*
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)nfs_vnops.c 8.19 (Berkeley) 7/31/95
35 */
36
37 /*
38 * vnode op calls for Sun NFS version 2 and 3
39 */
40
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: nfs_vnops.c,v 1.182 2003/10/25 08:48:11 christos Exp $");
43
44 #include "opt_nfs.h"
45 #include "opt_uvmhist.h"
46
47 #include <sys/param.h>
48 #include <sys/proc.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/resourcevar.h>
52 #include <sys/proc.h>
53 #include <sys/mount.h>
54 #include <sys/buf.h>
55 #include <sys/malloc.h>
56 #include <sys/mbuf.h>
57 #include <sys/namei.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/fcntl.h>
61 #include <sys/hash.h>
62 #include <sys/lockf.h>
63 #include <sys/stat.h>
64 #include <sys/unistd.h>
65
66 #include <uvm/uvm_extern.h>
67 #include <uvm/uvm.h>
68
69 #include <miscfs/fifofs/fifo.h>
70 #include <miscfs/genfs/genfs.h>
71 #include <miscfs/specfs/specdev.h>
72
73 #include <nfs/rpcv2.h>
74 #include <nfs/nfsproto.h>
75 #include <nfs/nfs.h>
76 #include <nfs/nfsnode.h>
77 #include <nfs/nfsmount.h>
78 #include <nfs/xdr_subs.h>
79 #include <nfs/nfsm_subs.h>
80 #include <nfs/nqnfs.h>
81 #include <nfs/nfs_var.h>
82
83 #include <net/if.h>
84 #include <netinet/in.h>
85 #include <netinet/in_var.h>
86
87 /*
88 * Global vfs data structures for nfs
89 */
90 int (**nfsv2_vnodeop_p) __P((void *));
91 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
92 { &vop_default_desc, vn_default_error },
93 { &vop_lookup_desc, nfs_lookup }, /* lookup */
94 { &vop_create_desc, nfs_create }, /* create */
95 { &vop_mknod_desc, nfs_mknod }, /* mknod */
96 { &vop_open_desc, nfs_open }, /* open */
97 { &vop_close_desc, nfs_close }, /* close */
98 { &vop_access_desc, nfs_access }, /* access */
99 { &vop_getattr_desc, nfs_getattr }, /* getattr */
100 { &vop_setattr_desc, nfs_setattr }, /* setattr */
101 { &vop_read_desc, nfs_read }, /* read */
102 { &vop_write_desc, nfs_write }, /* write */
103 { &vop_lease_desc, nfs_lease_check }, /* lease */
104 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
105 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */
106 { &vop_poll_desc, nfs_poll }, /* poll */
107 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */
108 { &vop_revoke_desc, nfs_revoke }, /* revoke */
109 { &vop_mmap_desc, nfs_mmap }, /* mmap */
110 { &vop_fsync_desc, nfs_fsync }, /* fsync */
111 { &vop_seek_desc, nfs_seek }, /* seek */
112 { &vop_remove_desc, nfs_remove }, /* remove */
113 { &vop_link_desc, nfs_link }, /* link */
114 { &vop_rename_desc, nfs_rename }, /* rename */
115 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */
116 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */
117 { &vop_symlink_desc, nfs_symlink }, /* symlink */
118 { &vop_readdir_desc, nfs_readdir }, /* readdir */
119 { &vop_readlink_desc, nfs_readlink }, /* readlink */
120 { &vop_abortop_desc, nfs_abortop }, /* abortop */
121 { &vop_inactive_desc, nfs_inactive }, /* inactive */
122 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
123 { &vop_lock_desc, nfs_lock }, /* lock */
124 { &vop_unlock_desc, nfs_unlock }, /* unlock */
125 { &vop_bmap_desc, nfs_bmap }, /* bmap */
126 { &vop_strategy_desc, nfs_strategy }, /* strategy */
127 { &vop_print_desc, nfs_print }, /* print */
128 { &vop_islocked_desc, nfs_islocked }, /* islocked */
129 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */
130 { &vop_advlock_desc, nfs_advlock }, /* advlock */
131 { &vop_blkatoff_desc, nfs_blkatoff }, /* blkatoff */
132 { &vop_valloc_desc, nfs_valloc }, /* valloc */
133 { &vop_reallocblks_desc, nfs_reallocblks }, /* reallocblks */
134 { &vop_vfree_desc, nfs_vfree }, /* vfree */
135 { &vop_truncate_desc, nfs_truncate }, /* truncate */
136 { &vop_update_desc, nfs_update }, /* update */
137 { &vop_bwrite_desc, nfs_bwrite }, /* bwrite */
138 { &vop_getpages_desc, nfs_getpages }, /* getpages */
139 { &vop_putpages_desc, genfs_putpages }, /* putpages */
140 { NULL, NULL }
141 };
142 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
143 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
144
145 /*
146 * Special device vnode ops
147 */
148 int (**spec_nfsv2nodeop_p) __P((void *));
149 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
150 { &vop_default_desc, vn_default_error },
151 { &vop_lookup_desc, spec_lookup }, /* lookup */
152 { &vop_create_desc, spec_create }, /* create */
153 { &vop_mknod_desc, spec_mknod }, /* mknod */
154 { &vop_open_desc, spec_open }, /* open */
155 { &vop_close_desc, nfsspec_close }, /* close */
156 { &vop_access_desc, nfsspec_access }, /* access */
157 { &vop_getattr_desc, nfs_getattr }, /* getattr */
158 { &vop_setattr_desc, nfs_setattr }, /* setattr */
159 { &vop_read_desc, nfsspec_read }, /* read */
160 { &vop_write_desc, nfsspec_write }, /* write */
161 { &vop_lease_desc, spec_lease_check }, /* lease */
162 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
163 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */
164 { &vop_poll_desc, spec_poll }, /* poll */
165 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */
166 { &vop_revoke_desc, spec_revoke }, /* revoke */
167 { &vop_mmap_desc, spec_mmap }, /* mmap */
168 { &vop_fsync_desc, spec_fsync }, /* fsync */
169 { &vop_seek_desc, spec_seek }, /* seek */
170 { &vop_remove_desc, spec_remove }, /* remove */
171 { &vop_link_desc, spec_link }, /* link */
172 { &vop_rename_desc, spec_rename }, /* rename */
173 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */
174 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */
175 { &vop_symlink_desc, spec_symlink }, /* symlink */
176 { &vop_readdir_desc, spec_readdir }, /* readdir */
177 { &vop_readlink_desc, spec_readlink }, /* readlink */
178 { &vop_abortop_desc, spec_abortop }, /* abortop */
179 { &vop_inactive_desc, nfs_inactive }, /* inactive */
180 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
181 { &vop_lock_desc, nfs_lock }, /* lock */
182 { &vop_unlock_desc, nfs_unlock }, /* unlock */
183 { &vop_bmap_desc, spec_bmap }, /* bmap */
184 { &vop_strategy_desc, spec_strategy }, /* strategy */
185 { &vop_print_desc, nfs_print }, /* print */
186 { &vop_islocked_desc, nfs_islocked }, /* islocked */
187 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */
188 { &vop_advlock_desc, spec_advlock }, /* advlock */
189 { &vop_blkatoff_desc, spec_blkatoff }, /* blkatoff */
190 { &vop_valloc_desc, spec_valloc }, /* valloc */
191 { &vop_reallocblks_desc, spec_reallocblks }, /* reallocblks */
192 { &vop_vfree_desc, spec_vfree }, /* vfree */
193 { &vop_truncate_desc, spec_truncate }, /* truncate */
194 { &vop_update_desc, nfs_update }, /* update */
195 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
196 { &vop_getpages_desc, spec_getpages }, /* getpages */
197 { &vop_putpages_desc, spec_putpages }, /* putpages */
198 { NULL, NULL }
199 };
200 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
201 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
202
203 int (**fifo_nfsv2nodeop_p) __P((void *));
204 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
205 { &vop_default_desc, vn_default_error },
206 { &vop_lookup_desc, fifo_lookup }, /* lookup */
207 { &vop_create_desc, fifo_create }, /* create */
208 { &vop_mknod_desc, fifo_mknod }, /* mknod */
209 { &vop_open_desc, fifo_open }, /* open */
210 { &vop_close_desc, nfsfifo_close }, /* close */
211 { &vop_access_desc, nfsspec_access }, /* access */
212 { &vop_getattr_desc, nfs_getattr }, /* getattr */
213 { &vop_setattr_desc, nfs_setattr }, /* setattr */
214 { &vop_read_desc, nfsfifo_read }, /* read */
215 { &vop_write_desc, nfsfifo_write }, /* write */
216 { &vop_lease_desc, fifo_lease_check }, /* lease */
217 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
218 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */
219 { &vop_poll_desc, fifo_poll }, /* poll */
220 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */
221 { &vop_revoke_desc, fifo_revoke }, /* revoke */
222 { &vop_mmap_desc, fifo_mmap }, /* mmap */
223 { &vop_fsync_desc, nfs_fsync }, /* fsync */
224 { &vop_seek_desc, fifo_seek }, /* seek */
225 { &vop_remove_desc, fifo_remove }, /* remove */
226 { &vop_link_desc, fifo_link }, /* link */
227 { &vop_rename_desc, fifo_rename }, /* rename */
228 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */
229 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */
230 { &vop_symlink_desc, fifo_symlink }, /* symlink */
231 { &vop_readdir_desc, fifo_readdir }, /* readdir */
232 { &vop_readlink_desc, fifo_readlink }, /* readlink */
233 { &vop_abortop_desc, fifo_abortop }, /* abortop */
234 { &vop_inactive_desc, nfs_inactive }, /* inactive */
235 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
236 { &vop_lock_desc, nfs_lock }, /* lock */
237 { &vop_unlock_desc, nfs_unlock }, /* unlock */
238 { &vop_bmap_desc, fifo_bmap }, /* bmap */
239 { &vop_strategy_desc, genfs_badop }, /* strategy */
240 { &vop_print_desc, nfs_print }, /* print */
241 { &vop_islocked_desc, nfs_islocked }, /* islocked */
242 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */
243 { &vop_advlock_desc, fifo_advlock }, /* advlock */
244 { &vop_blkatoff_desc, fifo_blkatoff }, /* blkatoff */
245 { &vop_valloc_desc, fifo_valloc }, /* valloc */
246 { &vop_reallocblks_desc, fifo_reallocblks }, /* reallocblks */
247 { &vop_vfree_desc, fifo_vfree }, /* vfree */
248 { &vop_truncate_desc, fifo_truncate }, /* truncate */
249 { &vop_update_desc, nfs_update }, /* update */
250 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
251 { &vop_putpages_desc, fifo_putpages }, /* putpages */
252 { NULL, NULL }
253 };
254 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
255 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
256
257 static void nfs_writerpc_extfree(struct mbuf *, caddr_t, size_t, void *);
258
259 /*
260 * Global variables
261 */
262 extern u_int32_t nfs_true, nfs_false;
263 extern u_int32_t nfs_xdrneg1;
264 extern const nfstype nfsv3_type[9];
265
266 int nfs_numasync = 0;
267 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
268
269 /*
270 * nfs null call from vfs.
271 */
272 int
273 nfs_null(vp, cred, procp)
274 struct vnode *vp;
275 struct ucred *cred;
276 struct proc *procp;
277 {
278 caddr_t bpos, dpos;
279 int error = 0;
280 struct mbuf *mreq, *mrep, *md, *mb;
281 struct nfsnode *np = VTONFS(vp);
282
283 nfsm_reqhead(np, NFSPROC_NULL, 0);
284 nfsm_request(np, NFSPROC_NULL, procp, cred);
285 nfsm_reqdone;
286 return (error);
287 }
288
289 /*
290 * nfs access vnode op.
291 * For nfs version 2, just return ok. File accesses may fail later.
292 * For nfs version 3, use the access rpc to check accessibility. If file modes
293 * are changed on the server, accesses might still fail later.
294 */
295 int
296 nfs_access(v)
297 void *v;
298 {
299 struct vop_access_args /* {
300 struct vnode *a_vp;
301 int a_mode;
302 struct ucred *a_cred;
303 struct proc *a_p;
304 } */ *ap = v;
305 struct vnode *vp = ap->a_vp;
306 u_int32_t *tl;
307 caddr_t cp;
308 int32_t t1, t2;
309 caddr_t bpos, dpos, cp2;
310 int error = 0, attrflag, cachevalid;
311 struct mbuf *mreq, *mrep, *md, *mb;
312 u_int32_t mode, rmode;
313 const int v3 = NFS_ISV3(vp);
314 struct nfsnode *np = VTONFS(vp);
315
316 cachevalid = (np->n_accstamp != -1 &&
317 (time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) &&
318 np->n_accuid == ap->a_cred->cr_uid);
319
320 /*
321 * Check access cache first. If this request has been made for this
322 * uid shortly before, use the cached result.
323 */
324 if (cachevalid) {
325 if (!np->n_accerror) {
326 if ((np->n_accmode & ap->a_mode) == ap->a_mode)
327 return np->n_accerror;
328 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
329 return np->n_accerror;
330 }
331
332 /*
333 * For nfs v3, do an access rpc, otherwise you are stuck emulating
334 * ufs_access() locally using the vattr. This may not be correct,
335 * since the server may apply other access criteria such as
336 * client uid-->server uid mapping that we do not know about, but
337 * this is better than just returning anything that is lying about
338 * in the cache.
339 */
340 if (v3) {
341 nfsstats.rpccnt[NFSPROC_ACCESS]++;
342 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
343 nfsm_fhtom(np, v3);
344 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
345 if (ap->a_mode & VREAD)
346 mode = NFSV3ACCESS_READ;
347 else
348 mode = 0;
349 if (vp->v_type != VDIR) {
350 if (ap->a_mode & VWRITE)
351 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
352 if (ap->a_mode & VEXEC)
353 mode |= NFSV3ACCESS_EXECUTE;
354 } else {
355 if (ap->a_mode & VWRITE)
356 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
357 NFSV3ACCESS_DELETE);
358 if (ap->a_mode & VEXEC)
359 mode |= NFSV3ACCESS_LOOKUP;
360 }
361 *tl = txdr_unsigned(mode);
362 nfsm_request(np, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
363 nfsm_postop_attr(vp, attrflag, 0);
364 if (!error) {
365 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
366 rmode = fxdr_unsigned(u_int32_t, *tl);
367 /*
368 * The NFS V3 spec does not clarify whether or not
369 * the returned access bits can be a superset of
370 * the ones requested, so...
371 */
372 if ((rmode & mode) != mode)
373 error = EACCES;
374 }
375 nfsm_reqdone;
376 } else
377 return (nfsspec_access(ap));
378 /*
379 * Disallow write attempts on filesystems mounted read-only;
380 * unless the file is a socket, fifo, or a block or character
381 * device resident on the filesystem.
382 */
383 if (!error && (ap->a_mode & VWRITE) &&
384 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
385 switch (vp->v_type) {
386 case VREG:
387 case VDIR:
388 case VLNK:
389 error = EROFS;
390 default:
391 break;
392 }
393 }
394
395 if (!error || error == EACCES) {
396 /*
397 * If we got the same result as for a previous,
398 * different request, OR it in. Don't update
399 * the timestamp in that case.
400 */
401 if (cachevalid && np->n_accstamp != -1 &&
402 error == np->n_accerror) {
403 if (!error)
404 np->n_accmode |= ap->a_mode;
405 else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
406 np->n_accmode = ap->a_mode;
407 } else {
408 np->n_accstamp = time.tv_sec;
409 np->n_accuid = ap->a_cred->cr_uid;
410 np->n_accmode = ap->a_mode;
411 np->n_accerror = error;
412 }
413 }
414
415 return (error);
416 }
417
418 /*
419 * nfs open vnode op
420 * Check to see if the type is ok
421 * and that deletion is not in progress.
422 * For paged in text files, you will need to flush the page cache
423 * if consistency is lost.
424 */
425 /* ARGSUSED */
426 int
427 nfs_open(v)
428 void *v;
429 {
430 struct vop_open_args /* {
431 struct vnode *a_vp;
432 int a_mode;
433 struct ucred *a_cred;
434 struct proc *a_p;
435 } */ *ap = v;
436 struct vnode *vp = ap->a_vp;
437 struct nfsnode *np = VTONFS(vp);
438 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
439 struct vattr vattr;
440 int error;
441
442 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
443 return (EACCES);
444 }
445
446 /*
447 * Initialize read and write creds here, for swapfiles
448 * and other paths that don't set the creds themselves.
449 */
450
451 if (ap->a_mode & FREAD) {
452 if (np->n_rcred) {
453 crfree(np->n_rcred);
454 }
455 np->n_rcred = ap->a_cred;
456 crhold(np->n_rcred);
457 }
458 if (ap->a_mode & FWRITE) {
459 if (np->n_wcred) {
460 crfree(np->n_wcred);
461 }
462 np->n_wcred = ap->a_cred;
463 crhold(np->n_wcred);
464 }
465
466 #ifndef NFS_V2_ONLY
467 /*
468 * Get a valid lease. If cached data is stale, flush it.
469 */
470 if (nmp->nm_flag & NFSMNT_NQNFS) {
471 if (NQNFS_CKINVALID(vp, np, ND_READ)) {
472 do {
473 error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
474 ap->a_p);
475 } while (error == NQNFS_EXPIRED);
476 if (error)
477 return (error);
478 if (np->n_lrev != np->n_brev ||
479 (np->n_flag & NQNFSNONCACHE)) {
480 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
481 ap->a_p, 1)) == EINTR)
482 return (error);
483 np->n_brev = np->n_lrev;
484 }
485 }
486 } else
487 #endif
488 {
489 if (np->n_flag & NMODIFIED) {
490 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
491 ap->a_p, 1)) == EINTR)
492 return (error);
493 np->n_attrstamp = 0;
494 if (vp->v_type == VDIR) {
495 nfs_invaldircache(vp, 0);
496 np->n_direofoffset = 0;
497 }
498 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
499 if (error)
500 return (error);
501 np->n_mtime = vattr.va_mtime;
502 } else {
503 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
504 if (error)
505 return (error);
506 if (timespeccmp(&np->n_mtime, &vattr.va_mtime, !=)) {
507 if (vp->v_type == VDIR) {
508 nfs_invaldircache(vp, 0);
509 np->n_direofoffset = 0;
510 }
511 if ((error = nfs_vinvalbuf(vp, V_SAVE,
512 ap->a_cred, ap->a_p, 1)) == EINTR)
513 return (error);
514 np->n_mtime = vattr.va_mtime;
515 }
516 }
517 }
518 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
519 np->n_attrstamp = 0; /* For Open/Close consistency */
520 return (0);
521 }
522
523 /*
524 * nfs close vnode op
525 * What an NFS client should do upon close after writing is a debatable issue.
526 * Most NFS clients push delayed writes to the server upon close, basically for
527 * two reasons:
528 * 1 - So that any write errors may be reported back to the client process
529 * doing the close system call. By far the two most likely errors are
530 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
531 * 2 - To put a worst case upper bound on cache inconsistency between
532 * multiple clients for the file.
533 * There is also a consistency problem for Version 2 of the protocol w.r.t.
534 * not being able to tell if other clients are writing a file concurrently,
535 * since there is no way of knowing if the changed modify time in the reply
536 * is only due to the write for this client.
537 * (NFS Version 3 provides weak cache consistency data in the reply that
538 * should be sufficient to detect and handle this case.)
539 *
540 * The current code does the following:
541 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
542 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
543 * or commit them (this satisfies 1 and 2 except for the
544 * case where the server crashes after this close but
545 * before the commit RPC, which is felt to be "good
546 * enough". Changing the last argument to nfs_flush() to
547 * a 1 would force a commit operation, if it is felt a
548 * commit is necessary now.
549 * for NQNFS - do nothing now, since 2 is dealt with via leases and
550 * 1 should be dealt with via an fsync() system call for
551 * cases where write errors are important.
552 */
553 /* ARGSUSED */
554 int
555 nfs_close(v)
556 void *v;
557 {
558 struct vop_close_args /* {
559 struct vnodeop_desc *a_desc;
560 struct vnode *a_vp;
561 int a_fflag;
562 struct ucred *a_cred;
563 struct proc *a_p;
564 } */ *ap = v;
565 struct vnode *vp = ap->a_vp;
566 struct nfsnode *np = VTONFS(vp);
567 int error = 0;
568 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
569
570 if (vp->v_type == VREG) {
571 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
572 (np->n_flag & NMODIFIED)) {
573 if (NFS_ISV3(vp)) {
574 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
575 np->n_flag &= ~NMODIFIED;
576 } else
577 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
578 np->n_attrstamp = 0;
579 }
580 if (np->n_flag & NWRITEERR) {
581 np->n_flag &= ~NWRITEERR;
582 error = np->n_error;
583 }
584 }
585 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
586 return (error);
587 }
588
589 /*
590 * nfs getattr call from vfs.
591 */
592 int
593 nfs_getattr(v)
594 void *v;
595 {
596 struct vop_getattr_args /* {
597 struct vnode *a_vp;
598 struct vattr *a_vap;
599 struct ucred *a_cred;
600 struct proc *a_p;
601 } */ *ap = v;
602 struct vnode *vp = ap->a_vp;
603 struct nfsnode *np = VTONFS(vp);
604 caddr_t cp;
605 u_int32_t *tl;
606 int32_t t1, t2;
607 caddr_t bpos, dpos;
608 int error = 0;
609 struct mbuf *mreq, *mrep, *md, *mb;
610 const int v3 = NFS_ISV3(vp);
611
612 /*
613 * Update local times for special files.
614 */
615 if (np->n_flag & (NACC | NUPD))
616 np->n_flag |= NCHG;
617
618 /*
619 * if we have delayed truncation, do it now.
620 */
621 nfs_delayedtruncate(vp);
622
623 /*
624 * First look in the cache.
625 */
626 if (nfs_getattrcache(vp, ap->a_vap) == 0)
627 return (0);
628 nfsstats.rpccnt[NFSPROC_GETATTR]++;
629 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
630 nfsm_fhtom(np, v3);
631 nfsm_request(np, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
632 if (!error) {
633 nfsm_loadattr(vp, ap->a_vap, 0);
634 if (vp->v_type == VDIR &&
635 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
636 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
637 }
638 nfsm_reqdone;
639 return (error);
640 }
641
642 /*
643 * nfs setattr call.
644 */
645 int
646 nfs_setattr(v)
647 void *v;
648 {
649 struct vop_setattr_args /* {
650 struct vnodeop_desc *a_desc;
651 struct vnode *a_vp;
652 struct vattr *a_vap;
653 struct ucred *a_cred;
654 struct proc *a_p;
655 } */ *ap = v;
656 struct vnode *vp = ap->a_vp;
657 struct nfsnode *np = VTONFS(vp);
658 struct vattr *vap = ap->a_vap;
659 int error = 0;
660 u_quad_t tsize = 0;
661
662 /*
663 * Setting of flags is not supported.
664 */
665 if (vap->va_flags != VNOVAL)
666 return (EOPNOTSUPP);
667
668 /*
669 * Disallow write attempts if the filesystem is mounted read-only.
670 */
671 if ((vap->va_uid != (uid_t)VNOVAL ||
672 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
673 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
674 (vp->v_mount->mnt_flag & MNT_RDONLY))
675 return (EROFS);
676 if (vap->va_size != VNOVAL) {
677 switch (vp->v_type) {
678 case VDIR:
679 return (EISDIR);
680 case VCHR:
681 case VBLK:
682 case VSOCK:
683 case VFIFO:
684 if (vap->va_mtime.tv_sec == VNOVAL &&
685 vap->va_atime.tv_sec == VNOVAL &&
686 vap->va_mode == (mode_t)VNOVAL &&
687 vap->va_uid == (uid_t)VNOVAL &&
688 vap->va_gid == (gid_t)VNOVAL)
689 return (0);
690 vap->va_size = VNOVAL;
691 break;
692 default:
693 /*
694 * Disallow write attempts if the filesystem is
695 * mounted read-only.
696 */
697 if (vp->v_mount->mnt_flag & MNT_RDONLY)
698 return (EROFS);
699 uvm_vnp_setsize(vp, vap->va_size);
700 tsize = np->n_size;
701 np->n_size = vap->va_size;
702 if (vap->va_size == 0)
703 error = nfs_vinvalbuf(vp, 0,
704 ap->a_cred, ap->a_p, 1);
705 else
706 error = nfs_vinvalbuf(vp, V_SAVE,
707 ap->a_cred, ap->a_p, 1);
708 if (error) {
709 uvm_vnp_setsize(vp, tsize);
710 return (error);
711 }
712 np->n_vattr->va_size = vap->va_size;
713 }
714 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
715 vap->va_atime.tv_sec != VNOVAL) &&
716 vp->v_type == VREG &&
717 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
718 ap->a_p, 1)) == EINTR)
719 return (error);
720 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
721 if (error && vap->va_size != VNOVAL) {
722 np->n_size = np->n_vattr->va_size = tsize;
723 uvm_vnp_setsize(vp, np->n_size);
724 }
725 VN_KNOTE(vp, NOTE_ATTRIB);
726 return (error);
727 }
728
729 /*
730 * Do an nfs setattr rpc.
731 */
732 int
733 nfs_setattrrpc(vp, vap, cred, procp)
734 struct vnode *vp;
735 struct vattr *vap;
736 struct ucred *cred;
737 struct proc *procp;
738 {
739 struct nfsv2_sattr *sp;
740 caddr_t cp;
741 int32_t t1, t2;
742 caddr_t bpos, dpos, cp2;
743 u_int32_t *tl;
744 int error = 0, wccflag = NFSV3_WCCRATTR;
745 struct mbuf *mreq, *mrep, *md, *mb;
746 const int v3 = NFS_ISV3(vp);
747 struct nfsnode *np = VTONFS(vp);
748
749 nfsstats.rpccnt[NFSPROC_SETATTR]++;
750 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
751 nfsm_fhtom(np, v3);
752 if (v3) {
753 nfsm_v3attrbuild(vap, TRUE);
754 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
755 *tl = nfs_false;
756 } else {
757 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
758 if (vap->va_mode == (mode_t)VNOVAL)
759 sp->sa_mode = nfs_xdrneg1;
760 else
761 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
762 if (vap->va_uid == (uid_t)VNOVAL)
763 sp->sa_uid = nfs_xdrneg1;
764 else
765 sp->sa_uid = txdr_unsigned(vap->va_uid);
766 if (vap->va_gid == (gid_t)VNOVAL)
767 sp->sa_gid = nfs_xdrneg1;
768 else
769 sp->sa_gid = txdr_unsigned(vap->va_gid);
770 sp->sa_size = txdr_unsigned(vap->va_size);
771 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
772 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
773 }
774 nfsm_request(np, NFSPROC_SETATTR, procp, cred);
775 if (v3) {
776 nfsm_wcc_data(vp, wccflag, 0);
777 } else
778 nfsm_loadattr(vp, (struct vattr *)0, 0);
779 nfsm_reqdone;
780 return (error);
781 }
782
783 /*
784 * nfs lookup call, one step at a time...
785 * First look in cache
786 * If not found, unlock the directory nfsnode and do the rpc
787 *
788 * This code is full of lock/unlock statements and checks, because
789 * we continue after cache_lookup has finished (we need to check
790 * with the attr cache and do an rpc if it has timed out). This means
791 * that the locking effects of cache_lookup have to be taken into
792 * account.
793 */
794 int
795 nfs_lookup(v)
796 void *v;
797 {
798 struct vop_lookup_args /* {
799 struct vnodeop_desc *a_desc;
800 struct vnode *a_dvp;
801 struct vnode **a_vpp;
802 struct componentname *a_cnp;
803 } */ *ap = v;
804 struct componentname *cnp = ap->a_cnp;
805 struct vnode *dvp = ap->a_dvp;
806 struct vnode **vpp = ap->a_vpp;
807 int flags;
808 struct vnode *newvp;
809 u_int32_t *tl;
810 caddr_t cp;
811 int32_t t1, t2;
812 caddr_t bpos, dpos, cp2;
813 struct mbuf *mreq, *mrep, *md, *mb;
814 long len;
815 nfsfh_t *fhp;
816 struct nfsnode *np;
817 int lockparent, wantparent, error = 0, attrflag, fhsize;
818 const int v3 = NFS_ISV3(dvp);
819
820 cnp->cn_flags &= ~PDIRUNLOCK;
821 flags = cnp->cn_flags;
822
823 *vpp = NULLVP;
824 newvp = NULLVP;
825 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
826 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
827 return (EROFS);
828 if (dvp->v_type != VDIR)
829 return (ENOTDIR);
830
831 lockparent = flags & LOCKPARENT;
832 wantparent = flags & (LOCKPARENT|WANTPARENT);
833 np = VTONFS(dvp);
834
835 /*
836 * Before tediously performing a linear scan of the directory,
837 * check the name cache to see if the directory/name pair
838 * we are looking for is known already.
839 * If the directory/name pair is found in the name cache,
840 * we have to ensure the directory has not changed from
841 * the time the cache entry has been created. If it has,
842 * the cache entry has to be ignored.
843 */
844 if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) {
845 struct vattr vattr;
846 int err2;
847
848 if (error && error != ENOENT) {
849 *vpp = NULLVP;
850 return error;
851 }
852
853 if (cnp->cn_flags & PDIRUNLOCK) {
854 err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
855 if (err2 != 0) {
856 *vpp = NULLVP;
857 return err2;
858 }
859 cnp->cn_flags &= ~PDIRUNLOCK;
860 }
861
862 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
863 if (err2 != 0) {
864 if (error == 0) {
865 if (*vpp != dvp)
866 vput(*vpp);
867 else
868 vrele(*vpp);
869 }
870 *vpp = NULLVP;
871 return err2;
872 }
873
874 if (error == ENOENT) {
875 if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
876 cnp->cn_proc) && timespeccmp(&vattr.va_mtime,
877 &VTONFS(dvp)->n_nctime, ==))
878 return ENOENT;
879 cache_purge(dvp);
880 timespecclear(&np->n_nctime);
881 goto dorpc;
882 }
883
884 newvp = *vpp;
885 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
886 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime)
887 {
888 nfsstats.lookupcache_hits++;
889 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
890 cnp->cn_flags |= SAVENAME;
891 if ((!lockparent || !(flags & ISLASTCN)) &&
892 newvp != dvp)
893 VOP_UNLOCK(dvp, 0);
894 return (0);
895 }
896 cache_purge(newvp);
897 if (newvp != dvp)
898 vput(newvp);
899 else
900 vrele(newvp);
901 *vpp = NULLVP;
902 }
903 dorpc:
904 error = 0;
905 newvp = NULLVP;
906 nfsstats.lookupcache_misses++;
907 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
908 len = cnp->cn_namelen;
909 nfsm_reqhead(np, NFSPROC_LOOKUP,
910 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
911 nfsm_fhtom(np, v3);
912 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
913 nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
914 if (error) {
915 nfsm_postop_attr(dvp, attrflag, 0);
916 m_freem(mrep);
917 goto nfsmout;
918 }
919 nfsm_getfh(fhp, fhsize, v3);
920
921 /*
922 * Handle RENAME case...
923 */
924 if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
925 if (NFS_CMPFH(np, fhp, fhsize)) {
926 m_freem(mrep);
927 return (EISDIR);
928 }
929 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
930 if (error) {
931 m_freem(mrep);
932 return error;
933 }
934 newvp = NFSTOV(np);
935 if (v3) {
936 nfsm_postop_attr(newvp, attrflag, 0);
937 nfsm_postop_attr(dvp, attrflag, 0);
938 } else
939 nfsm_loadattr(newvp, (struct vattr *)0, 0);
940 *vpp = newvp;
941 m_freem(mrep);
942 cnp->cn_flags |= SAVENAME;
943 if (!lockparent) {
944 VOP_UNLOCK(dvp, 0);
945 cnp->cn_flags |= PDIRUNLOCK;
946 }
947 return (0);
948 }
949
950 /*
951 * The postop attr handling is duplicated for each if case,
952 * because it should be done while dvp is locked (unlocking
953 * dvp is different for each case).
954 */
955
956 if (NFS_CMPFH(np, fhp, fhsize)) {
957 /*
958 * "." lookup
959 */
960 VREF(dvp);
961 newvp = dvp;
962 if (v3) {
963 nfsm_postop_attr(newvp, attrflag, 0);
964 nfsm_postop_attr(dvp, attrflag, 0);
965 } else
966 nfsm_loadattr(newvp, (struct vattr *)0, 0);
967 } else if (flags & ISDOTDOT) {
968 /*
969 * ".." lookup
970 */
971 VOP_UNLOCK(dvp, 0);
972 cnp->cn_flags |= PDIRUNLOCK;
973
974 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
975 if (error) {
976 if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
977 cnp->cn_flags &= ~PDIRUNLOCK;
978 m_freem(mrep);
979 return error;
980 }
981 newvp = NFSTOV(np);
982
983 if (v3) {
984 nfsm_postop_attr(newvp, attrflag, 0);
985 nfsm_postop_attr(dvp, attrflag, 0);
986 } else
987 nfsm_loadattr(newvp, (struct vattr *)0, 0);
988
989 if (lockparent && (flags & ISLASTCN)) {
990 if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
991 m_freem(mrep);
992 vput(newvp);
993 return error;
994 }
995 cnp->cn_flags &= ~PDIRUNLOCK;
996 }
997 } else {
998 /*
999 * Other lookups.
1000 */
1001 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1002 if (error) {
1003 m_freem(mrep);
1004 return error;
1005 }
1006 newvp = NFSTOV(np);
1007 if (v3) {
1008 nfsm_postop_attr(newvp, attrflag, 0);
1009 nfsm_postop_attr(dvp, attrflag, 0);
1010 } else
1011 nfsm_loadattr(newvp, (struct vattr *)0, 0);
1012 if (!lockparent || !(flags & ISLASTCN)) {
1013 VOP_UNLOCK(dvp, 0);
1014 cnp->cn_flags |= PDIRUNLOCK;
1015 }
1016 }
1017 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1018 cnp->cn_flags |= SAVENAME;
1019 if ((cnp->cn_flags & MAKEENTRY) &&
1020 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1021 np->n_ctime = np->n_vattr->va_ctime.tv_sec;
1022 cache_enter(dvp, newvp, cnp);
1023 }
1024 *vpp = newvp;
1025 nfsm_reqdone;
1026 if (error) {
1027 /*
1028 * We get here only because of errors returned by
1029 * the RPC. Otherwise we'll have returned above
1030 * (the nfsm_* macros will jump to nfsm_reqdone
1031 * on error).
1032 */
1033 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1034 cnp->cn_nameiop != CREATE) {
1035 if (!timespecisset(&VTONFS(dvp)->n_nctime))
1036 VTONFS(dvp)->n_nctime =
1037 VTONFS(dvp)->n_vattr->va_mtime;
1038
1039 cache_enter(dvp, NULL, cnp);
1040 }
1041 if (newvp != NULLVP) {
1042 vrele(newvp);
1043 if (newvp != dvp)
1044 VOP_UNLOCK(newvp, 0);
1045 }
1046 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1047 (flags & ISLASTCN) && error == ENOENT) {
1048 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1049 error = EROFS;
1050 else
1051 error = EJUSTRETURN;
1052 }
1053 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1054 cnp->cn_flags |= SAVENAME;
1055 *vpp = NULL;
1056 }
1057
1058 return error;
1059 }
1060
1061 /*
1062 * nfs read call.
1063 * Just call nfs_bioread() to do the work.
1064 */
1065 int
1066 nfs_read(v)
1067 void *v;
1068 {
1069 struct vop_read_args /* {
1070 struct vnode *a_vp;
1071 struct uio *a_uio;
1072 int a_ioflag;
1073 struct ucred *a_cred;
1074 } */ *ap = v;
1075 struct vnode *vp = ap->a_vp;
1076
1077 if (vp->v_type != VREG)
1078 return (EPERM);
1079 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1080 }
1081
1082 /*
1083 * nfs readlink call
1084 */
1085 int
1086 nfs_readlink(v)
1087 void *v;
1088 {
1089 struct vop_readlink_args /* {
1090 struct vnode *a_vp;
1091 struct uio *a_uio;
1092 struct ucred *a_cred;
1093 } */ *ap = v;
1094 struct vnode *vp = ap->a_vp;
1095
1096 if (vp->v_type != VLNK)
1097 return (EPERM);
1098 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1099 }
1100
1101 /*
1102 * Do a readlink rpc.
1103 * Called by nfs_doio() from below the buffer cache.
1104 */
1105 int
1106 nfs_readlinkrpc(vp, uiop, cred)
1107 struct vnode *vp;
1108 struct uio *uiop;
1109 struct ucred *cred;
1110 {
1111 u_int32_t *tl;
1112 caddr_t cp;
1113 int32_t t1, t2;
1114 caddr_t bpos, dpos, cp2;
1115 int error = 0, attrflag;
1116 uint32_t len;
1117 struct mbuf *mreq, *mrep, *md, *mb;
1118 const int v3 = NFS_ISV3(vp);
1119 struct nfsnode *np = VTONFS(vp);
1120
1121 nfsstats.rpccnt[NFSPROC_READLINK]++;
1122 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1123 nfsm_fhtom(np, v3);
1124 nfsm_request(np, NFSPROC_READLINK, uiop->uio_procp, cred);
1125 if (v3)
1126 nfsm_postop_attr(vp, attrflag, 0);
1127 if (!error) {
1128 if (v3) {
1129 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1130 len = fxdr_unsigned(uint32_t, *tl);
1131 if (len > MAXPATHLEN) {
1132 /*
1133 * this pathname is too long for us.
1134 */
1135 m_freem(mrep);
1136 /* Solaris returns EINVAL. should we follow? */
1137 error = ENAMETOOLONG;
1138 goto nfsmout;
1139 }
1140 } else {
1141 nfsm_strsiz(len, NFS_MAXPATHLEN);
1142 }
1143 nfsm_mtouio(uiop, len);
1144 }
1145 nfsm_reqdone;
1146 return (error);
1147 }
1148
1149 /*
1150 * nfs read rpc call
1151 * Ditto above
1152 */
1153 int
1154 nfs_readrpc(vp, uiop)
1155 struct vnode *vp;
1156 struct uio *uiop;
1157 {
1158 u_int32_t *tl;
1159 caddr_t cp;
1160 int32_t t1, t2;
1161 caddr_t bpos, dpos, cp2;
1162 struct mbuf *mreq, *mrep, *md, *mb;
1163 struct nfsmount *nmp;
1164 int error = 0, len, retlen, tsiz, eof, attrflag;
1165 const int v3 = NFS_ISV3(vp);
1166 struct nfsnode *np = VTONFS(vp);
1167
1168 #ifndef nolint
1169 eof = 0;
1170 #endif
1171 nmp = VFSTONFS(vp->v_mount);
1172 tsiz = uiop->uio_resid;
1173 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1174 return (EFBIG);
1175 while (tsiz > 0) {
1176 nfsstats.rpccnt[NFSPROC_READ]++;
1177 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1178 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1179 nfsm_fhtom(np, v3);
1180 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1181 if (v3) {
1182 txdr_hyper(uiop->uio_offset, tl);
1183 *(tl + 2) = txdr_unsigned(len);
1184 } else {
1185 *tl++ = txdr_unsigned(uiop->uio_offset);
1186 *tl++ = txdr_unsigned(len);
1187 *tl = 0;
1188 }
1189 nfsm_request(np, NFSPROC_READ, uiop->uio_procp,
1190 VTONFS(vp)->n_rcred);
1191 if (v3) {
1192 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1193 if (error) {
1194 m_freem(mrep);
1195 goto nfsmout;
1196 }
1197 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1198 eof = fxdr_unsigned(int, *(tl + 1));
1199 } else
1200 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1201 nfsm_strsiz(retlen, nmp->nm_rsize);
1202 nfsm_mtouio(uiop, retlen);
1203 m_freem(mrep);
1204 tsiz -= retlen;
1205 if (v3) {
1206 if (eof || retlen == 0)
1207 tsiz = 0;
1208 } else if (retlen < len)
1209 tsiz = 0;
1210 }
1211 nfsmout:
1212 return (error);
1213 }
1214
1215 struct nfs_writerpc_context {
1216 struct simplelock nwc_slock;
1217 volatile int nwc_mbufcount;
1218 };
1219
1220 /*
1221 * free mbuf used to refer protected pages while write rpc call.
1222 * called at splvm.
1223 */
1224 static void
1225 nfs_writerpc_extfree(struct mbuf *m, caddr_t buf, size_t size, void *arg)
1226 {
1227 struct nfs_writerpc_context *ctx = arg;
1228
1229 KASSERT(m != NULL);
1230 KASSERT(ctx != NULL);
1231 pool_cache_put(&mbpool_cache, m);
1232 simple_lock(&ctx->nwc_slock);
1233 if (--ctx->nwc_mbufcount == 0) {
1234 wakeup(ctx);
1235 }
1236 simple_unlock(&ctx->nwc_slock);
1237 }
1238
1239 /*
1240 * nfs write call
1241 */
1242 int
1243 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp)
1244 struct vnode *vp;
1245 struct uio *uiop;
1246 int *iomode;
1247 boolean_t pageprotected;
1248 boolean_t *stalewriteverfp;
1249 {
1250 u_int32_t *tl;
1251 caddr_t cp;
1252 int32_t t1, t2;
1253 caddr_t bpos, dpos, cp2;
1254 struct mbuf *mreq, *mrep, *md, *mb;
1255 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1256 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1257 const int v3 = NFS_ISV3(vp);
1258 int committed = NFSV3WRITE_FILESYNC;
1259 struct nfsnode *np = VTONFS(vp);
1260 struct nfs_writerpc_context ctx;
1261 int s;
1262 struct lwp *l = NULL;
1263 size_t origresid;
1264
1265 simple_lock_init(&ctx.nwc_slock);
1266 ctx.nwc_mbufcount = 1;
1267
1268 if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1269 panic("writerpc readonly vp %p", vp);
1270 }
1271
1272 #ifdef DIAGNOSTIC
1273 if (uiop->uio_iovcnt != 1)
1274 panic("nfs: writerpc iovcnt > 1");
1275 #endif
1276 tsiz = uiop->uio_resid;
1277 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1278 return (EFBIG);
1279 if (pageprotected) {
1280 l = curlwp;
1281 PHOLD(l);
1282 }
1283 retry:
1284 origresid = uiop->uio_resid;
1285 KASSERT(origresid == uiop->uio_iov->iov_len);
1286 while (tsiz > 0) {
1287 uint32_t datalen; /* data bytes need to be allocated in mbuf */
1288 uint32_t backup;
1289 boolean_t stalewriteverf = FALSE;
1290
1291 nfsstats.rpccnt[NFSPROC_WRITE]++;
1292 len = min(tsiz, nmp->nm_wsize);
1293 datalen = pageprotected ? 0 : nfsm_rndup(len);
1294 nfsm_reqhead(np, NFSPROC_WRITE,
1295 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1296 nfsm_fhtom(np, v3);
1297 if (v3) {
1298 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1299 txdr_hyper(uiop->uio_offset, tl);
1300 tl += 2;
1301 *tl++ = txdr_unsigned(len);
1302 *tl++ = txdr_unsigned(*iomode);
1303 *tl = txdr_unsigned(len);
1304 } else {
1305 u_int32_t x;
1306
1307 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1308 /* Set both "begin" and "current" to non-garbage. */
1309 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1310 *tl++ = x; /* "begin offset" */
1311 *tl++ = x; /* "current offset" */
1312 x = txdr_unsigned(len);
1313 *tl++ = x; /* total to this offset */
1314 *tl = x; /* size of this write */
1315
1316 }
1317 if (pageprotected) {
1318 /*
1319 * since we know pages can't be modified during i/o,
1320 * no need to copy them for us.
1321 */
1322 struct mbuf *m;
1323 struct iovec *iovp = uiop->uio_iov;
1324
1325 m = m_get(M_WAIT, MT_DATA);
1326 MCLAIM(m, &nfs_mowner);
1327 MEXTADD(m, iovp->iov_base, len, M_MBUF,
1328 nfs_writerpc_extfree, &ctx);
1329 m->m_flags |= M_EXT_ROMAP;
1330 m->m_len = len;
1331 mb->m_next = m;
1332 /*
1333 * no need to maintain mb and bpos here
1334 * because no one care them later.
1335 */
1336 #if 0
1337 mb = m;
1338 bpos = mtod(caddr_t, mb) + mb->m_len;
1339 #endif
1340 iovp->iov_base = (char *)iovp->iov_base + len;
1341 iovp->iov_len -= len;
1342 uiop->uio_offset += len;
1343 uiop->uio_resid -= len;
1344 s = splvm();
1345 simple_lock(&ctx.nwc_slock);
1346 ctx.nwc_mbufcount++;
1347 simple_unlock(&ctx.nwc_slock);
1348 splx(s);
1349 } else {
1350 nfsm_uiotom(uiop, len);
1351 }
1352 nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp,
1353 VTONFS(vp)->n_wcred);
1354 if (v3) {
1355 wccflag = NFSV3_WCCCHK;
1356 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC);
1357 if (!error) {
1358 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1359 + NFSX_V3WRITEVERF);
1360 rlen = fxdr_unsigned(int, *tl++);
1361 if (rlen == 0) {
1362 error = NFSERR_IO;
1363 m_freem(mrep);
1364 break;
1365 } else if (rlen < len) {
1366 backup = len - rlen;
1367 uiop->uio_iov->iov_base =
1368 (caddr_t)uiop->uio_iov->iov_base -
1369 backup;
1370 uiop->uio_iov->iov_len += backup;
1371 uiop->uio_offset -= backup;
1372 uiop->uio_resid += backup;
1373 len = rlen;
1374 }
1375 commit = fxdr_unsigned(int, *tl++);
1376
1377 /*
1378 * Return the lowest committment level
1379 * obtained by any of the RPCs.
1380 */
1381 if (committed == NFSV3WRITE_FILESYNC)
1382 committed = commit;
1383 else if (committed == NFSV3WRITE_DATASYNC &&
1384 commit == NFSV3WRITE_UNSTABLE)
1385 committed = commit;
1386 simple_lock(&nmp->nm_slock);
1387 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1388 memcpy(nmp->nm_writeverf, tl,
1389 NFSX_V3WRITEVERF);
1390 nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1391 } else if ((nmp->nm_iflag &
1392 NFSMNT_STALEWRITEVERF) ||
1393 memcmp(tl, nmp->nm_writeverf,
1394 NFSX_V3WRITEVERF)) {
1395 memcpy(nmp->nm_writeverf, tl,
1396 NFSX_V3WRITEVERF);
1397 /*
1398 * note NFSMNT_STALEWRITEVERF
1399 * if we're the first thread to
1400 * notice it.
1401 */
1402 if ((nmp->nm_iflag &
1403 NFSMNT_STALEWRITEVERF) == 0) {
1404 stalewriteverf = TRUE;
1405 nmp->nm_iflag |=
1406 NFSMNT_STALEWRITEVERF;
1407 }
1408 }
1409 simple_unlock(&nmp->nm_slock);
1410 }
1411 } else
1412 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1413 if (wccflag)
1414 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1415 m_freem(mrep);
1416 if (error)
1417 break;
1418 tsiz -= len;
1419 if (stalewriteverf) {
1420 *stalewriteverfp = TRUE;
1421 stalewriteverf = FALSE;
1422 if (committed == NFSV3WRITE_UNSTABLE &&
1423 len != origresid) {
1424 /*
1425 * if our write requests weren't atomic but
1426 * unstable, datas in previous iterations
1427 * might have already been lost now.
1428 * then, we should resend them to nfsd.
1429 */
1430 backup = origresid - tsiz;
1431 uiop->uio_iov->iov_base =
1432 (caddr_t)uiop->uio_iov->iov_base - backup;
1433 uiop->uio_iov->iov_len += backup;
1434 uiop->uio_offset -= backup;
1435 uiop->uio_resid += backup;
1436 tsiz = origresid;
1437 goto retry;
1438 }
1439 }
1440 }
1441 if (pageprotected) {
1442 /*
1443 * wait until mbufs go away.
1444 * retransmitted mbufs can survive longer than rpc requests
1445 * themselves.
1446 */
1447 s = splvm();
1448 simple_lock(&ctx.nwc_slock);
1449 ctx.nwc_mbufcount--;
1450 while (ctx.nwc_mbufcount > 0) {
1451 ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock);
1452 }
1453 simple_unlock(&ctx.nwc_slock);
1454 splx(s);
1455 PRELE(l);
1456 }
1457 nfsmout:
1458 *iomode = committed;
1459 if (error)
1460 uiop->uio_resid = tsiz;
1461 return (error);
1462 }
1463
1464 /*
1465 * nfs mknod rpc
1466 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1467 * mode set to specify the file type and the size field for rdev.
1468 */
1469 int
1470 nfs_mknodrpc(dvp, vpp, cnp, vap)
1471 struct vnode *dvp;
1472 struct vnode **vpp;
1473 struct componentname *cnp;
1474 struct vattr *vap;
1475 {
1476 struct nfsv2_sattr *sp;
1477 u_int32_t *tl;
1478 caddr_t cp;
1479 int32_t t1, t2;
1480 struct vnode *newvp = (struct vnode *)0;
1481 struct nfsnode *dnp, *np;
1482 char *cp2;
1483 caddr_t bpos, dpos;
1484 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1485 struct mbuf *mreq, *mrep, *md, *mb;
1486 u_int32_t rdev;
1487 const int v3 = NFS_ISV3(dvp);
1488
1489 if (vap->va_type == VCHR || vap->va_type == VBLK)
1490 rdev = txdr_unsigned(vap->va_rdev);
1491 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1492 rdev = nfs_xdrneg1;
1493 else {
1494 VOP_ABORTOP(dvp, cnp);
1495 vput(dvp);
1496 return (EOPNOTSUPP);
1497 }
1498 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1499 dnp = VTONFS(dvp);
1500 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1501 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1502 nfsm_fhtom(dnp, v3);
1503 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1504 if (v3) {
1505 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1506 *tl++ = vtonfsv3_type(vap->va_type);
1507 nfsm_v3attrbuild(vap, FALSE);
1508 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1509 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1510 *tl++ = txdr_unsigned(major(vap->va_rdev));
1511 *tl = txdr_unsigned(minor(vap->va_rdev));
1512 }
1513 } else {
1514 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1515 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1516 sp->sa_uid = nfs_xdrneg1;
1517 sp->sa_gid = nfs_xdrneg1;
1518 sp->sa_size = rdev;
1519 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1520 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1521 }
1522 nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1523 if (!error) {
1524 nfsm_mtofh(dvp, newvp, v3, gotvp);
1525 if (!gotvp) {
1526 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1527 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1528 if (!error)
1529 newvp = NFSTOV(np);
1530 }
1531 }
1532 if (v3)
1533 nfsm_wcc_data(dvp, wccflag, 0);
1534 nfsm_reqdone;
1535 if (error) {
1536 if (newvp)
1537 vput(newvp);
1538 } else {
1539 if (cnp->cn_flags & MAKEENTRY)
1540 cache_enter(dvp, newvp, cnp);
1541 *vpp = newvp;
1542 }
1543 PNBUF_PUT(cnp->cn_pnbuf);
1544 VTONFS(dvp)->n_flag |= NMODIFIED;
1545 if (!wccflag)
1546 VTONFS(dvp)->n_attrstamp = 0;
1547 vput(dvp);
1548 return (error);
1549 }
1550
1551 /*
1552 * nfs mknod vop
1553 * just call nfs_mknodrpc() to do the work.
1554 */
1555 /* ARGSUSED */
1556 int
1557 nfs_mknod(v)
1558 void *v;
1559 {
1560 struct vop_mknod_args /* {
1561 struct vnode *a_dvp;
1562 struct vnode **a_vpp;
1563 struct componentname *a_cnp;
1564 struct vattr *a_vap;
1565 } */ *ap = v;
1566 int error;
1567
1568 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1569 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1570 return (error);
1571 }
1572
1573 static u_long create_verf;
1574 /*
1575 * nfs file create call
1576 */
1577 int
1578 nfs_create(v)
1579 void *v;
1580 {
1581 struct vop_create_args /* {
1582 struct vnode *a_dvp;
1583 struct vnode **a_vpp;
1584 struct componentname *a_cnp;
1585 struct vattr *a_vap;
1586 } */ *ap = v;
1587 struct vnode *dvp = ap->a_dvp;
1588 struct vattr *vap = ap->a_vap;
1589 struct componentname *cnp = ap->a_cnp;
1590 struct nfsv2_sattr *sp;
1591 u_int32_t *tl;
1592 caddr_t cp;
1593 int32_t t1, t2;
1594 struct nfsnode *dnp, *np = (struct nfsnode *)0;
1595 struct vnode *newvp = (struct vnode *)0;
1596 caddr_t bpos, dpos, cp2;
1597 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1598 struct mbuf *mreq, *mrep, *md, *mb;
1599 const int v3 = NFS_ISV3(dvp);
1600
1601 /*
1602 * Oops, not for me..
1603 */
1604 if (vap->va_type == VSOCK)
1605 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1606
1607 #ifdef VA_EXCLUSIVE
1608 if (vap->va_vaflags & VA_EXCLUSIVE)
1609 fmode |= O_EXCL;
1610 #endif
1611 again:
1612 error = 0;
1613 nfsstats.rpccnt[NFSPROC_CREATE]++;
1614 dnp = VTONFS(dvp);
1615 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1616 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1617 nfsm_fhtom(dnp, v3);
1618 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1619 if (v3) {
1620 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1621 if (fmode & O_EXCL) {
1622 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1623 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1624 #ifdef INET
1625 if (TAILQ_FIRST(&in_ifaddr))
1626 *tl++ = TAILQ_FIRST(&in_ifaddr)->ia_addr.sin_addr.s_addr;
1627 else
1628 *tl++ = create_verf;
1629 #else
1630 *tl++ = create_verf;
1631 #endif
1632 *tl = ++create_verf;
1633 } else {
1634 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1635 nfsm_v3attrbuild(vap, FALSE);
1636 }
1637 } else {
1638 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1639 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1640 sp->sa_uid = nfs_xdrneg1;
1641 sp->sa_gid = nfs_xdrneg1;
1642 sp->sa_size = 0;
1643 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1644 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1645 }
1646 nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1647 if (!error) {
1648 nfsm_mtofh(dvp, newvp, v3, gotvp);
1649 if (!gotvp) {
1650 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1651 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1652 if (!error)
1653 newvp = NFSTOV(np);
1654 }
1655 }
1656 if (v3)
1657 nfsm_wcc_data(dvp, wccflag, 0);
1658 nfsm_reqdone;
1659 if (error) {
1660 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1661 fmode &= ~O_EXCL;
1662 goto again;
1663 }
1664 if (newvp)
1665 vput(newvp);
1666 } else if (v3 && (fmode & O_EXCL))
1667 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1668 if (!error) {
1669 if (cnp->cn_flags & MAKEENTRY)
1670 cache_enter(dvp, newvp, cnp);
1671 *ap->a_vpp = newvp;
1672 }
1673 PNBUF_PUT(cnp->cn_pnbuf);
1674 VTONFS(dvp)->n_flag |= NMODIFIED;
1675 if (!wccflag)
1676 VTONFS(dvp)->n_attrstamp = 0;
1677 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1678 vput(dvp);
1679 return (error);
1680 }
1681
1682 /*
1683 * nfs file remove call
1684 * To try and make nfs semantics closer to ufs semantics, a file that has
1685 * other processes using the vnode is renamed instead of removed and then
1686 * removed later on the last close.
1687 * - If v_usecount > 1
1688 * If a rename is not already in the works
1689 * call nfs_sillyrename() to set it up
1690 * else
1691 * do the remove rpc
1692 */
1693 int
1694 nfs_remove(v)
1695 void *v;
1696 {
1697 struct vop_remove_args /* {
1698 struct vnodeop_desc *a_desc;
1699 struct vnode * a_dvp;
1700 struct vnode * a_vp;
1701 struct componentname * a_cnp;
1702 } */ *ap = v;
1703 struct vnode *vp = ap->a_vp;
1704 struct vnode *dvp = ap->a_dvp;
1705 struct componentname *cnp = ap->a_cnp;
1706 struct nfsnode *np = VTONFS(vp);
1707 int error = 0;
1708 struct vattr vattr;
1709
1710 #ifndef DIAGNOSTIC
1711 if ((cnp->cn_flags & HASBUF) == 0)
1712 panic("nfs_remove: no name");
1713 if (vp->v_usecount < 1)
1714 panic("nfs_remove: bad v_usecount");
1715 #endif
1716 if (vp->v_type == VDIR)
1717 error = EPERM;
1718 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1719 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1720 vattr.va_nlink > 1)) {
1721 /*
1722 * Purge the name cache so that the chance of a lookup for
1723 * the name succeeding while the remove is in progress is
1724 * minimized. Without node locking it can still happen, such
1725 * that an I/O op returns ESTALE, but since you get this if
1726 * another host removes the file..
1727 */
1728 cache_purge(vp);
1729 /*
1730 * throw away biocache buffers, mainly to avoid
1731 * unnecessary delayed writes later.
1732 */
1733 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1734 /* Do the rpc */
1735 if (error != EINTR)
1736 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1737 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1738 /*
1739 * Kludge City: If the first reply to the remove rpc is lost..
1740 * the reply to the retransmitted request will be ENOENT
1741 * since the file was in fact removed
1742 * Therefore, we cheat and return success.
1743 */
1744 if (error == ENOENT)
1745 error = 0;
1746 } else if (!np->n_sillyrename)
1747 error = nfs_sillyrename(dvp, vp, cnp);
1748 PNBUF_PUT(cnp->cn_pnbuf);
1749 if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1750 vattr.va_nlink == 1) {
1751 np->n_flag |= NREMOVED;
1752 }
1753 np->n_attrstamp = 0;
1754 VN_KNOTE(vp, NOTE_DELETE);
1755 VN_KNOTE(dvp, NOTE_WRITE);
1756 if (dvp == vp)
1757 vrele(vp);
1758 else
1759 vput(vp);
1760 vput(dvp);
1761 return (error);
1762 }
1763
1764 /*
1765 * nfs file remove rpc called from nfs_inactive
1766 */
1767 int
1768 nfs_removeit(sp)
1769 struct sillyrename *sp;
1770 {
1771
1772 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1773 (struct proc *)0));
1774 }
1775
1776 /*
1777 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1778 */
1779 int
1780 nfs_removerpc(dvp, name, namelen, cred, proc)
1781 struct vnode *dvp;
1782 const char *name;
1783 int namelen;
1784 struct ucred *cred;
1785 struct proc *proc;
1786 {
1787 u_int32_t *tl;
1788 caddr_t cp;
1789 int32_t t1, t2;
1790 caddr_t bpos, dpos, cp2;
1791 int error = 0, wccflag = NFSV3_WCCRATTR;
1792 struct mbuf *mreq, *mrep, *md, *mb;
1793 const int v3 = NFS_ISV3(dvp);
1794 struct nfsnode *dnp = VTONFS(dvp);
1795
1796 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1797 nfsm_reqhead(dnp, NFSPROC_REMOVE,
1798 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1799 nfsm_fhtom(dnp, v3);
1800 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1801 nfsm_request(dnp, NFSPROC_REMOVE, proc, cred);
1802 if (v3)
1803 nfsm_wcc_data(dvp, wccflag, 0);
1804 nfsm_reqdone;
1805 VTONFS(dvp)->n_flag |= NMODIFIED;
1806 if (!wccflag)
1807 VTONFS(dvp)->n_attrstamp = 0;
1808 return (error);
1809 }
1810
1811 /*
1812 * nfs file rename call
1813 */
1814 int
1815 nfs_rename(v)
1816 void *v;
1817 {
1818 struct vop_rename_args /* {
1819 struct vnode *a_fdvp;
1820 struct vnode *a_fvp;
1821 struct componentname *a_fcnp;
1822 struct vnode *a_tdvp;
1823 struct vnode *a_tvp;
1824 struct componentname *a_tcnp;
1825 } */ *ap = v;
1826 struct vnode *fvp = ap->a_fvp;
1827 struct vnode *tvp = ap->a_tvp;
1828 struct vnode *fdvp = ap->a_fdvp;
1829 struct vnode *tdvp = ap->a_tdvp;
1830 struct componentname *tcnp = ap->a_tcnp;
1831 struct componentname *fcnp = ap->a_fcnp;
1832 int error;
1833
1834 #ifndef DIAGNOSTIC
1835 if ((tcnp->cn_flags & HASBUF) == 0 ||
1836 (fcnp->cn_flags & HASBUF) == 0)
1837 panic("nfs_rename: no name");
1838 #endif
1839 /* Check for cross-device rename */
1840 if ((fvp->v_mount != tdvp->v_mount) ||
1841 (tvp && (fvp->v_mount != tvp->v_mount))) {
1842 error = EXDEV;
1843 goto out;
1844 }
1845
1846 /*
1847 * If the tvp exists and is in use, sillyrename it before doing the
1848 * rename of the new file over it.
1849 */
1850 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1851 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1852 VN_KNOTE(tvp, NOTE_DELETE);
1853 vput(tvp);
1854 tvp = NULL;
1855 }
1856
1857 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1858 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1859 tcnp->cn_proc);
1860
1861 VN_KNOTE(fdvp, NOTE_WRITE);
1862 VN_KNOTE(tdvp, NOTE_WRITE);
1863 if (fvp->v_type == VDIR) {
1864 if (tvp != NULL && tvp->v_type == VDIR)
1865 cache_purge(tdvp);
1866 cache_purge(fdvp);
1867 }
1868 out:
1869 if (tdvp == tvp)
1870 vrele(tdvp);
1871 else
1872 vput(tdvp);
1873 if (tvp)
1874 vput(tvp);
1875 vrele(fdvp);
1876 vrele(fvp);
1877 /*
1878 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1879 */
1880 if (error == ENOENT)
1881 error = 0;
1882 return (error);
1883 }
1884
1885 /*
1886 * nfs file rename rpc called from nfs_remove() above
1887 */
1888 int
1889 nfs_renameit(sdvp, scnp, sp)
1890 struct vnode *sdvp;
1891 struct componentname *scnp;
1892 struct sillyrename *sp;
1893 {
1894 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1895 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1896 }
1897
1898 /*
1899 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1900 */
1901 int
1902 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1903 struct vnode *fdvp;
1904 const char *fnameptr;
1905 int fnamelen;
1906 struct vnode *tdvp;
1907 const char *tnameptr;
1908 int tnamelen;
1909 struct ucred *cred;
1910 struct proc *proc;
1911 {
1912 u_int32_t *tl;
1913 caddr_t cp;
1914 int32_t t1, t2;
1915 caddr_t bpos, dpos, cp2;
1916 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1917 struct mbuf *mreq, *mrep, *md, *mb;
1918 const int v3 = NFS_ISV3(fdvp);
1919 struct nfsnode *fdnp = VTONFS(fdvp);
1920
1921 nfsstats.rpccnt[NFSPROC_RENAME]++;
1922 nfsm_reqhead(fdnp, NFSPROC_RENAME,
1923 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1924 nfsm_rndup(tnamelen));
1925 nfsm_fhtom(fdnp, v3);
1926 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1927 nfsm_fhtom(VTONFS(tdvp), v3);
1928 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1929 nfsm_request(fdnp, NFSPROC_RENAME, proc, cred);
1930 if (v3) {
1931 nfsm_wcc_data(fdvp, fwccflag, 0);
1932 nfsm_wcc_data(tdvp, twccflag, 0);
1933 }
1934 nfsm_reqdone;
1935 VTONFS(fdvp)->n_flag |= NMODIFIED;
1936 VTONFS(tdvp)->n_flag |= NMODIFIED;
1937 if (!fwccflag)
1938 VTONFS(fdvp)->n_attrstamp = 0;
1939 if (!twccflag)
1940 VTONFS(tdvp)->n_attrstamp = 0;
1941 return (error);
1942 }
1943
1944 /*
1945 * nfs hard link create call
1946 */
1947 int
1948 nfs_link(v)
1949 void *v;
1950 {
1951 struct vop_link_args /* {
1952 struct vnode *a_dvp;
1953 struct vnode *a_vp;
1954 struct componentname *a_cnp;
1955 } */ *ap = v;
1956 struct vnode *vp = ap->a_vp;
1957 struct vnode *dvp = ap->a_dvp;
1958 struct componentname *cnp = ap->a_cnp;
1959 u_int32_t *tl;
1960 caddr_t cp;
1961 int32_t t1, t2;
1962 caddr_t bpos, dpos, cp2;
1963 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1964 struct mbuf *mreq, *mrep, *md, *mb;
1965 /* XXX Should be const and initialised? */
1966 int v3;
1967 struct nfsnode *np;
1968
1969 if (dvp->v_mount != vp->v_mount) {
1970 VOP_ABORTOP(dvp, cnp);
1971 vput(dvp);
1972 return (EXDEV);
1973 }
1974 if (dvp != vp) {
1975 error = vn_lock(vp, LK_EXCLUSIVE);
1976 if (error != 0) {
1977 VOP_ABORTOP(dvp, cnp);
1978 vput(dvp);
1979 return error;
1980 }
1981 }
1982
1983 /*
1984 * Push all writes to the server, so that the attribute cache
1985 * doesn't get "out of sync" with the server.
1986 * XXX There should be a better way!
1987 */
1988 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc);
1989
1990 v3 = NFS_ISV3(vp);
1991 nfsstats.rpccnt[NFSPROC_LINK]++;
1992 np = VTONFS(vp);
1993 nfsm_reqhead(np, NFSPROC_LINK,
1994 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1995 nfsm_fhtom(np, v3);
1996 nfsm_fhtom(VTONFS(dvp), v3);
1997 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1998 nfsm_request(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1999 if (v3) {
2000 nfsm_postop_attr(vp, attrflag, 0);
2001 nfsm_wcc_data(dvp, wccflag, 0);
2002 }
2003 nfsm_reqdone;
2004 PNBUF_PUT(cnp->cn_pnbuf);
2005 VTONFS(dvp)->n_flag |= NMODIFIED;
2006 if (!attrflag)
2007 VTONFS(vp)->n_attrstamp = 0;
2008 if (!wccflag)
2009 VTONFS(dvp)->n_attrstamp = 0;
2010 if (dvp != vp)
2011 VOP_UNLOCK(vp, 0);
2012 VN_KNOTE(vp, NOTE_LINK);
2013 VN_KNOTE(dvp, NOTE_WRITE);
2014 vput(dvp);
2015 /*
2016 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2017 */
2018 if (error == EEXIST)
2019 error = 0;
2020 return (error);
2021 }
2022
2023 /*
2024 * nfs symbolic link create call
2025 */
2026 int
2027 nfs_symlink(v)
2028 void *v;
2029 {
2030 struct vop_symlink_args /* {
2031 struct vnode *a_dvp;
2032 struct vnode **a_vpp;
2033 struct componentname *a_cnp;
2034 struct vattr *a_vap;
2035 char *a_target;
2036 } */ *ap = v;
2037 struct vnode *dvp = ap->a_dvp;
2038 struct vattr *vap = ap->a_vap;
2039 struct componentname *cnp = ap->a_cnp;
2040 struct nfsv2_sattr *sp;
2041 u_int32_t *tl;
2042 caddr_t cp;
2043 int32_t t1, t2;
2044 caddr_t bpos, dpos, cp2;
2045 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2046 struct mbuf *mreq, *mrep, *md, *mb;
2047 struct vnode *newvp = (struct vnode *)0;
2048 const int v3 = NFS_ISV3(dvp);
2049 struct nfsnode *dnp = VTONFS(dvp);
2050
2051 *ap->a_vpp = NULL;
2052 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2053 slen = strlen(ap->a_target);
2054 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2055 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2056 nfsm_fhtom(dnp, v3);
2057 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2058 if (v3)
2059 nfsm_v3attrbuild(vap, FALSE);
2060 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2061 if (!v3) {
2062 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2063 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2064 sp->sa_uid = nfs_xdrneg1;
2065 sp->sa_gid = nfs_xdrneg1;
2066 sp->sa_size = nfs_xdrneg1;
2067 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2068 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2069 }
2070 nfsm_request(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
2071 if (v3) {
2072 if (!error)
2073 nfsm_mtofh(dvp, newvp, v3, gotvp);
2074 nfsm_wcc_data(dvp, wccflag, 0);
2075 }
2076 nfsm_reqdone;
2077 /*
2078 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2079 */
2080 if (error == EEXIST)
2081 error = 0;
2082 if (error == 0 && newvp == NULL) {
2083 struct nfsnode *np = NULL;
2084
2085 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2086 cnp->cn_cred, cnp->cn_proc, &np);
2087 if (error == 0)
2088 newvp = NFSTOV(np);
2089 }
2090 if (error) {
2091 if (newvp != NULL)
2092 vput(newvp);
2093 } else {
2094 *ap->a_vpp = newvp;
2095 }
2096 PNBUF_PUT(cnp->cn_pnbuf);
2097 VTONFS(dvp)->n_flag |= NMODIFIED;
2098 if (!wccflag)
2099 VTONFS(dvp)->n_attrstamp = 0;
2100 VN_KNOTE(dvp, NOTE_WRITE);
2101 vput(dvp);
2102 return (error);
2103 }
2104
2105 /*
2106 * nfs make dir call
2107 */
2108 int
2109 nfs_mkdir(v)
2110 void *v;
2111 {
2112 struct vop_mkdir_args /* {
2113 struct vnode *a_dvp;
2114 struct vnode **a_vpp;
2115 struct componentname *a_cnp;
2116 struct vattr *a_vap;
2117 } */ *ap = v;
2118 struct vnode *dvp = ap->a_dvp;
2119 struct vattr *vap = ap->a_vap;
2120 struct componentname *cnp = ap->a_cnp;
2121 struct nfsv2_sattr *sp;
2122 u_int32_t *tl;
2123 caddr_t cp;
2124 int32_t t1, t2;
2125 int len;
2126 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2127 struct vnode *newvp = (struct vnode *)0;
2128 caddr_t bpos, dpos, cp2;
2129 int error = 0, wccflag = NFSV3_WCCRATTR;
2130 int gotvp = 0;
2131 struct mbuf *mreq, *mrep, *md, *mb;
2132 const int v3 = NFS_ISV3(dvp);
2133
2134 len = cnp->cn_namelen;
2135 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2136 nfsm_reqhead(dnp, NFSPROC_MKDIR,
2137 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2138 nfsm_fhtom(dnp, v3);
2139 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2140 if (v3) {
2141 nfsm_v3attrbuild(vap, FALSE);
2142 } else {
2143 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2144 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2145 sp->sa_uid = nfs_xdrneg1;
2146 sp->sa_gid = nfs_xdrneg1;
2147 sp->sa_size = nfs_xdrneg1;
2148 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2149 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2150 }
2151 nfsm_request(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
2152 if (!error)
2153 nfsm_mtofh(dvp, newvp, v3, gotvp);
2154 if (v3)
2155 nfsm_wcc_data(dvp, wccflag, 0);
2156 nfsm_reqdone;
2157 VTONFS(dvp)->n_flag |= NMODIFIED;
2158 if (!wccflag)
2159 VTONFS(dvp)->n_attrstamp = 0;
2160 /*
2161 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2162 * if we can succeed in looking up the directory.
2163 */
2164 if (error == EEXIST || (!error && !gotvp)) {
2165 if (newvp) {
2166 vput(newvp);
2167 newvp = (struct vnode *)0;
2168 }
2169 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2170 cnp->cn_proc, &np);
2171 if (!error) {
2172 newvp = NFSTOV(np);
2173 if (newvp->v_type != VDIR)
2174 error = EEXIST;
2175 }
2176 }
2177 if (error) {
2178 if (newvp)
2179 vput(newvp);
2180 } else {
2181 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2182 if (cnp->cn_flags & MAKEENTRY)
2183 cache_enter(dvp, newvp, cnp);
2184 *ap->a_vpp = newvp;
2185 }
2186 PNBUF_PUT(cnp->cn_pnbuf);
2187 vput(dvp);
2188 return (error);
2189 }
2190
2191 /*
2192 * nfs remove directory call
2193 */
2194 int
2195 nfs_rmdir(v)
2196 void *v;
2197 {
2198 struct vop_rmdir_args /* {
2199 struct vnode *a_dvp;
2200 struct vnode *a_vp;
2201 struct componentname *a_cnp;
2202 } */ *ap = v;
2203 struct vnode *vp = ap->a_vp;
2204 struct vnode *dvp = ap->a_dvp;
2205 struct componentname *cnp = ap->a_cnp;
2206 u_int32_t *tl;
2207 caddr_t cp;
2208 int32_t t1, t2;
2209 caddr_t bpos, dpos, cp2;
2210 int error = 0, wccflag = NFSV3_WCCRATTR;
2211 struct mbuf *mreq, *mrep, *md, *mb;
2212 const int v3 = NFS_ISV3(dvp);
2213 struct nfsnode *dnp;
2214
2215 if (dvp == vp) {
2216 vrele(dvp);
2217 vput(dvp);
2218 PNBUF_PUT(cnp->cn_pnbuf);
2219 return (EINVAL);
2220 }
2221 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2222 dnp = VTONFS(dvp);
2223 nfsm_reqhead(dnp, NFSPROC_RMDIR,
2224 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2225 nfsm_fhtom(dnp, v3);
2226 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2227 nfsm_request(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
2228 if (v3)
2229 nfsm_wcc_data(dvp, wccflag, 0);
2230 nfsm_reqdone;
2231 PNBUF_PUT(cnp->cn_pnbuf);
2232 VTONFS(dvp)->n_flag |= NMODIFIED;
2233 if (!wccflag)
2234 VTONFS(dvp)->n_attrstamp = 0;
2235 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2236 VN_KNOTE(vp, NOTE_DELETE);
2237 cache_purge(dvp);
2238 cache_purge(vp);
2239 vput(vp);
2240 vput(dvp);
2241 /*
2242 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2243 */
2244 if (error == ENOENT)
2245 error = 0;
2246 return (error);
2247 }
2248
2249 /*
2250 * nfs readdir call
2251 */
2252 int
2253 nfs_readdir(v)
2254 void *v;
2255 {
2256 struct vop_readdir_args /* {
2257 struct vnode *a_vp;
2258 struct uio *a_uio;
2259 struct ucred *a_cred;
2260 int *a_eofflag;
2261 off_t **a_cookies;
2262 int *a_ncookies;
2263 } */ *ap = v;
2264 struct vnode *vp = ap->a_vp;
2265 struct uio *uio = ap->a_uio;
2266 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2267 char *base = uio->uio_iov->iov_base;
2268 int tresid, error;
2269 size_t count, lost;
2270 struct dirent *dp;
2271 off_t *cookies = NULL;
2272 int ncookies = 0, nc;
2273
2274 if (vp->v_type != VDIR)
2275 return (EPERM);
2276
2277 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2278 count = uio->uio_resid - lost;
2279 if (count <= 0)
2280 return (EINVAL);
2281
2282 /*
2283 * Call nfs_bioread() to do the real work.
2284 */
2285 tresid = uio->uio_resid = count;
2286 error = nfs_bioread(vp, uio, 0, ap->a_cred,
2287 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2288
2289 if (!error && ap->a_cookies) {
2290 ncookies = count / 16;
2291 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2292 *ap->a_cookies = cookies;
2293 }
2294
2295 if (!error && uio->uio_resid == tresid) {
2296 uio->uio_resid += lost;
2297 nfsstats.direofcache_misses++;
2298 if (ap->a_cookies)
2299 *ap->a_ncookies = 0;
2300 *ap->a_eofflag = 1;
2301 return (0);
2302 }
2303
2304 if (!error && ap->a_cookies) {
2305 /*
2306 * Only the NFS server and emulations use cookies, and they
2307 * load the directory block into system space, so we can
2308 * just look at it directly.
2309 */
2310 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2311 panic("nfs_readdir: lost in space");
2312 for (nc = 0; ncookies-- &&
2313 base < (char *)uio->uio_iov->iov_base; nc++){
2314 dp = (struct dirent *) base;
2315 if (dp->d_reclen == 0)
2316 break;
2317 if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2318 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2319 else
2320 *(cookies++) = NFS_GETCOOKIE(dp);
2321 base += dp->d_reclen;
2322 }
2323 uio->uio_resid +=
2324 ((caddr_t)uio->uio_iov->iov_base - base);
2325 uio->uio_iov->iov_len +=
2326 ((caddr_t)uio->uio_iov->iov_base - base);
2327 uio->uio_iov->iov_base = base;
2328 *ap->a_ncookies = nc;
2329 }
2330
2331 uio->uio_resid += lost;
2332 *ap->a_eofflag = 0;
2333 return (error);
2334 }
2335
2336 /*
2337 * Readdir rpc call.
2338 * Called from below the buffer cache by nfs_doio().
2339 */
2340 int
2341 nfs_readdirrpc(vp, uiop, cred)
2342 struct vnode *vp;
2343 struct uio *uiop;
2344 struct ucred *cred;
2345 {
2346 int len, left;
2347 struct dirent *dp = NULL;
2348 u_int32_t *tl;
2349 caddr_t cp;
2350 int32_t t1, t2;
2351 caddr_t bpos, dpos, cp2;
2352 struct mbuf *mreq, *mrep, *md, *mb;
2353 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2354 struct nfsnode *dnp = VTONFS(vp);
2355 u_quad_t fileno;
2356 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2357 int attrflag, nrpcs = 0, reclen;
2358 const int v3 = NFS_ISV3(vp);
2359
2360 #ifdef DIAGNOSTIC
2361 /*
2362 * Should be called from buffer cache, so only amount of
2363 * NFS_DIRBLKSIZ will be requested.
2364 */
2365 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2366 panic("nfs readdirrpc bad uio");
2367 #endif
2368
2369 /*
2370 * Loop around doing readdir rpc's of size nm_readdirsize
2371 * truncated to a multiple of NFS_DIRFRAGSIZ.
2372 * The stopping criteria is EOF or buffer full.
2373 */
2374 while (more_dirs && bigenough) {
2375 /*
2376 * Heuristic: don't bother to do another RPC to further
2377 * fill up this block if there is not much room left. (< 50%
2378 * of the readdir RPC size). This wastes some buffer space
2379 * but can save up to 50% in RPC calls.
2380 */
2381 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2382 bigenough = 0;
2383 break;
2384 }
2385 nfsstats.rpccnt[NFSPROC_READDIR]++;
2386 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2387 NFSX_READDIR(v3));
2388 nfsm_fhtom(dnp, v3);
2389 if (v3) {
2390 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2391 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2392 txdr_swapcookie3(uiop->uio_offset, tl);
2393 } else {
2394 txdr_cookie3(uiop->uio_offset, tl);
2395 }
2396 tl += 2;
2397 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2398 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2399 } else {
2400 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2401 *tl++ = txdr_unsigned(uiop->uio_offset);
2402 }
2403 *tl = txdr_unsigned(nmp->nm_readdirsize);
2404 nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred);
2405 nrpcs++;
2406 if (v3) {
2407 nfsm_postop_attr(vp, attrflag, 0);
2408 if (!error) {
2409 nfsm_dissect(tl, u_int32_t *,
2410 2 * NFSX_UNSIGNED);
2411 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2412 dnp->n_cookieverf.nfsuquad[1] = *tl;
2413 } else {
2414 m_freem(mrep);
2415 goto nfsmout;
2416 }
2417 }
2418 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2419 more_dirs = fxdr_unsigned(int, *tl);
2420
2421 /* loop thru the dir entries, doctoring them to 4bsd form */
2422 while (more_dirs && bigenough) {
2423 if (v3) {
2424 nfsm_dissect(tl, u_int32_t *,
2425 3 * NFSX_UNSIGNED);
2426 fileno = fxdr_hyper(tl);
2427 len = fxdr_unsigned(int, *(tl + 2));
2428 } else {
2429 nfsm_dissect(tl, u_int32_t *,
2430 2 * NFSX_UNSIGNED);
2431 fileno = fxdr_unsigned(u_quad_t, *tl++);
2432 len = fxdr_unsigned(int, *tl);
2433 }
2434 if (len <= 0 || len > NFS_MAXNAMLEN) {
2435 error = EBADRPC;
2436 m_freem(mrep);
2437 goto nfsmout;
2438 }
2439 tlen = nfsm_rndup(len);
2440 if (tlen == len)
2441 tlen += 4; /* To ensure null termination */
2442 tlen += sizeof (off_t) + sizeof (int);
2443 reclen = ALIGN(tlen + DIRHDSIZ);
2444 tlen = reclen - DIRHDSIZ;
2445 left = NFS_DIRFRAGSIZ - blksiz;
2446 if (reclen > left) {
2447 dp->d_reclen += left;
2448 uiop->uio_iov->iov_base =
2449 (caddr_t)uiop->uio_iov->iov_base + left;
2450 uiop->uio_iov->iov_len -= left;
2451 uiop->uio_resid -= left;
2452 blksiz = 0;
2453 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2454 }
2455 if (reclen > uiop->uio_resid)
2456 bigenough = 0;
2457 if (bigenough) {
2458 dp = (struct dirent *)uiop->uio_iov->iov_base;
2459 dp->d_fileno = (int)fileno;
2460 dp->d_namlen = len;
2461 dp->d_reclen = reclen;
2462 dp->d_type = DT_UNKNOWN;
2463 blksiz += dp->d_reclen;
2464 if (blksiz == NFS_DIRFRAGSIZ)
2465 blksiz = 0;
2466 uiop->uio_resid -= DIRHDSIZ;
2467 uiop->uio_iov->iov_base =
2468 (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ;
2469 uiop->uio_iov->iov_len -= DIRHDSIZ;
2470 nfsm_mtouio(uiop, len);
2471 cp = uiop->uio_iov->iov_base;
2472 tlen -= len;
2473 *cp = '\0'; /* null terminate */
2474 uiop->uio_iov->iov_base =
2475 (caddr_t)uiop->uio_iov->iov_base + tlen;
2476 uiop->uio_iov->iov_len -= tlen;
2477 uiop->uio_resid -= tlen;
2478 } else
2479 nfsm_adv(nfsm_rndup(len));
2480 if (v3) {
2481 nfsm_dissect(tl, u_int32_t *,
2482 3 * NFSX_UNSIGNED);
2483 } else {
2484 nfsm_dissect(tl, u_int32_t *,
2485 2 * NFSX_UNSIGNED);
2486 }
2487 if (bigenough) {
2488 if (v3) {
2489 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2490 uiop->uio_offset =
2491 fxdr_swapcookie3(tl);
2492 else
2493 uiop->uio_offset =
2494 fxdr_cookie3(tl);
2495 }
2496 else {
2497 uiop->uio_offset =
2498 fxdr_unsigned(off_t, *tl);
2499 }
2500 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2501 }
2502 if (v3)
2503 tl += 2;
2504 else
2505 tl++;
2506 more_dirs = fxdr_unsigned(int, *tl);
2507 }
2508 /*
2509 * If at end of rpc data, get the eof boolean
2510 */
2511 if (!more_dirs) {
2512 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2513 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2514 }
2515 m_freem(mrep);
2516 }
2517 /*
2518 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2519 * by increasing d_reclen for the last record.
2520 */
2521 if (blksiz > 0) {
2522 left = NFS_DIRFRAGSIZ - blksiz;
2523 dp->d_reclen += left;
2524 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2525 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2526 left;
2527 uiop->uio_iov->iov_len -= left;
2528 uiop->uio_resid -= left;
2529 }
2530
2531 /*
2532 * We are now either at the end of the directory or have filled the
2533 * block.
2534 */
2535 if (bigenough)
2536 dnp->n_direofoffset = uiop->uio_offset;
2537 nfsmout:
2538 return (error);
2539 }
2540
2541 /*
2542 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2543 */
2544 int
2545 nfs_readdirplusrpc(vp, uiop, cred)
2546 struct vnode *vp;
2547 struct uio *uiop;
2548 struct ucred *cred;
2549 {
2550 int len, left;
2551 struct dirent *dp = NULL;
2552 u_int32_t *tl;
2553 caddr_t cp;
2554 int32_t t1, t2;
2555 struct vnode *newvp;
2556 caddr_t bpos, dpos, cp2;
2557 struct mbuf *mreq, *mrep, *md, *mb;
2558 struct nameidata nami, *ndp = &nami;
2559 struct componentname *cnp = &ndp->ni_cnd;
2560 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2561 struct nfsnode *dnp = VTONFS(vp), *np;
2562 nfsfh_t *fhp;
2563 u_quad_t fileno;
2564 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2565 int attrflag, fhsize, nrpcs = 0, reclen;
2566 struct nfs_fattr fattr, *fp;
2567
2568 #ifdef DIAGNOSTIC
2569 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2570 panic("nfs readdirplusrpc bad uio");
2571 #endif
2572 ndp->ni_dvp = vp;
2573 newvp = NULLVP;
2574
2575 /*
2576 * Loop around doing readdir rpc's of size nm_readdirsize
2577 * truncated to a multiple of NFS_DIRFRAGSIZ.
2578 * The stopping criteria is EOF or buffer full.
2579 */
2580 while (more_dirs && bigenough) {
2581 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2582 bigenough = 0;
2583 break;
2584 }
2585 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2586 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2587 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2588 nfsm_fhtom(dnp, 1);
2589 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2590 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2591 txdr_swapcookie3(uiop->uio_offset, tl);
2592 } else {
2593 txdr_cookie3(uiop->uio_offset, tl);
2594 }
2595 tl += 2;
2596 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2597 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2598 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2599 *tl = txdr_unsigned(nmp->nm_rsize);
2600 nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2601 nfsm_postop_attr(vp, attrflag, 0);
2602 if (error) {
2603 m_freem(mrep);
2604 goto nfsmout;
2605 }
2606 nrpcs++;
2607 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2608 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2609 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2610 more_dirs = fxdr_unsigned(int, *tl);
2611
2612 /* loop thru the dir entries, doctoring them to 4bsd form */
2613 while (more_dirs && bigenough) {
2614 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2615 fileno = fxdr_hyper(tl);
2616 len = fxdr_unsigned(int, *(tl + 2));
2617 if (len <= 0 || len > NFS_MAXNAMLEN) {
2618 error = EBADRPC;
2619 m_freem(mrep);
2620 goto nfsmout;
2621 }
2622 tlen = nfsm_rndup(len);
2623 if (tlen == len)
2624 tlen += 4; /* To ensure null termination*/
2625 tlen += sizeof (off_t) + sizeof (int);
2626 reclen = ALIGN(tlen + DIRHDSIZ);
2627 tlen = reclen - DIRHDSIZ;
2628 left = NFS_DIRFRAGSIZ - blksiz;
2629 if (reclen > left) {
2630 /*
2631 * DIRFRAGSIZ is aligned, no need to align
2632 * again here.
2633 */
2634 dp->d_reclen += left;
2635 uiop->uio_iov->iov_base =
2636 (caddr_t)uiop->uio_iov->iov_base + left;
2637 uiop->uio_iov->iov_len -= left;
2638 uiop->uio_resid -= left;
2639 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2640 blksiz = 0;
2641 }
2642 if (reclen > uiop->uio_resid)
2643 bigenough = 0;
2644 if (bigenough) {
2645 dp = (struct dirent *)uiop->uio_iov->iov_base;
2646 dp->d_fileno = (int)fileno;
2647 dp->d_namlen = len;
2648 dp->d_reclen = reclen;
2649 dp->d_type = DT_UNKNOWN;
2650 blksiz += dp->d_reclen;
2651 if (blksiz == NFS_DIRFRAGSIZ)
2652 blksiz = 0;
2653 uiop->uio_resid -= DIRHDSIZ;
2654 uiop->uio_iov->iov_base =
2655 (caddr_t)uiop->uio_iov->iov_base +
2656 DIRHDSIZ;
2657 uiop->uio_iov->iov_len -= DIRHDSIZ;
2658 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2659 cnp->cn_namelen = len;
2660 nfsm_mtouio(uiop, len);
2661 cp = uiop->uio_iov->iov_base;
2662 tlen -= len;
2663 *cp = '\0';
2664 uiop->uio_iov->iov_base =
2665 (caddr_t)uiop->uio_iov->iov_base + tlen;
2666 uiop->uio_iov->iov_len -= tlen;
2667 uiop->uio_resid -= tlen;
2668 } else
2669 nfsm_adv(nfsm_rndup(len));
2670 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2671 if (bigenough) {
2672 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2673 uiop->uio_offset =
2674 fxdr_swapcookie3(tl);
2675 else
2676 uiop->uio_offset =
2677 fxdr_cookie3(tl);
2678 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2679 }
2680 tl += 2;
2681
2682 /*
2683 * Since the attributes are before the file handle
2684 * (sigh), we must skip over the attributes and then
2685 * come back and get them.
2686 */
2687 attrflag = fxdr_unsigned(int, *tl);
2688 if (attrflag) {
2689 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2690 memcpy(&fattr, fp, NFSX_V3FATTR);
2691 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2692 doit = fxdr_unsigned(int, *tl);
2693 if (doit) {
2694 nfsm_getfh(fhp, fhsize, 1);
2695 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2696 VREF(vp);
2697 newvp = vp;
2698 np = dnp;
2699 } else {
2700 error = nfs_nget(vp->v_mount, fhp,
2701 fhsize, &np);
2702 if (!error)
2703 newvp = NFSTOV(np);
2704 }
2705 if (!error) {
2706 const char *cp;
2707
2708 nfs_loadattrcache(&newvp, &fattr, 0, 0);
2709 dp->d_type =
2710 IFTODT(VTTOIF(np->n_vattr->va_type));
2711 ndp->ni_vp = newvp;
2712 cp = cnp->cn_nameptr + cnp->cn_namelen;
2713 cnp->cn_hash =
2714 namei_hash(cnp->cn_nameptr, &cp);
2715 if (cnp->cn_namelen <= NCHNAMLEN)
2716 cache_enter(ndp->ni_dvp, ndp->ni_vp,
2717 cnp);
2718 }
2719 }
2720 } else {
2721 /* Just skip over the file handle */
2722 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2723 i = fxdr_unsigned(int, *tl);
2724 nfsm_adv(nfsm_rndup(i));
2725 }
2726 if (newvp != NULLVP) {
2727 if (newvp == vp)
2728 vrele(newvp);
2729 else
2730 vput(newvp);
2731 newvp = NULLVP;
2732 }
2733 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2734 more_dirs = fxdr_unsigned(int, *tl);
2735 }
2736 /*
2737 * If at end of rpc data, get the eof boolean
2738 */
2739 if (!more_dirs) {
2740 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2741 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2742 }
2743 m_freem(mrep);
2744 }
2745 /*
2746 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2747 * by increasing d_reclen for the last record.
2748 */
2749 if (blksiz > 0) {
2750 left = NFS_DIRFRAGSIZ - blksiz;
2751 dp->d_reclen += left;
2752 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2753 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2754 left;
2755 uiop->uio_iov->iov_len -= left;
2756 uiop->uio_resid -= left;
2757 }
2758
2759 /*
2760 * We are now either at the end of the directory or have filled the
2761 * block.
2762 */
2763 if (bigenough)
2764 dnp->n_direofoffset = uiop->uio_offset;
2765 nfsmout:
2766 if (newvp != NULLVP) {
2767 if(newvp == vp)
2768 vrele(newvp);
2769 else
2770 vput(newvp);
2771 }
2772 return (error);
2773 }
2774 static char hextoasc[] = "0123456789abcdef";
2775
2776 /*
2777 * Silly rename. To make the NFS filesystem that is stateless look a little
2778 * more like the "ufs" a remove of an active vnode is translated to a rename
2779 * to a funny looking filename that is removed by nfs_inactive on the
2780 * nfsnode. There is the potential for another process on a different client
2781 * to create the same funny name between the nfs_lookitup() fails and the
2782 * nfs_rename() completes, but...
2783 */
2784 int
2785 nfs_sillyrename(dvp, vp, cnp)
2786 struct vnode *dvp, *vp;
2787 struct componentname *cnp;
2788 {
2789 struct sillyrename *sp;
2790 struct nfsnode *np;
2791 int error;
2792 short pid;
2793
2794 cache_purge(dvp);
2795 np = VTONFS(vp);
2796 #ifndef DIAGNOSTIC
2797 if (vp->v_type == VDIR)
2798 panic("nfs: sillyrename dir");
2799 #endif
2800 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2801 M_NFSREQ, M_WAITOK);
2802 sp->s_cred = crdup(cnp->cn_cred);
2803 sp->s_dvp = dvp;
2804 VREF(dvp);
2805
2806 /* Fudge together a funny name */
2807 pid = cnp->cn_proc->p_pid;
2808 memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2809 sp->s_namlen = 12;
2810 sp->s_name[8] = hextoasc[pid & 0xf];
2811 sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
2812 sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
2813 sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
2814
2815 /* Try lookitups until we get one that isn't there */
2816 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2817 cnp->cn_proc, (struct nfsnode **)0) == 0) {
2818 sp->s_name[4]++;
2819 if (sp->s_name[4] > 'z') {
2820 error = EINVAL;
2821 goto bad;
2822 }
2823 }
2824 error = nfs_renameit(dvp, cnp, sp);
2825 if (error)
2826 goto bad;
2827 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2828 cnp->cn_proc, &np);
2829 np->n_sillyrename = sp;
2830 return (0);
2831 bad:
2832 vrele(sp->s_dvp);
2833 crfree(sp->s_cred);
2834 free((caddr_t)sp, M_NFSREQ);
2835 return (error);
2836 }
2837
2838 /*
2839 * Look up a file name and optionally either update the file handle or
2840 * allocate an nfsnode, depending on the value of npp.
2841 * npp == NULL --> just do the lookup
2842 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2843 * handled too
2844 * *npp != NULL --> update the file handle in the vnode
2845 */
2846 int
2847 nfs_lookitup(dvp, name, len, cred, procp, npp)
2848 struct vnode *dvp;
2849 const char *name;
2850 int len;
2851 struct ucred *cred;
2852 struct proc *procp;
2853 struct nfsnode **npp;
2854 {
2855 u_int32_t *tl;
2856 caddr_t cp;
2857 int32_t t1, t2;
2858 struct vnode *newvp = (struct vnode *)0;
2859 struct nfsnode *np, *dnp = VTONFS(dvp);
2860 caddr_t bpos, dpos, cp2;
2861 int error = 0, fhlen, attrflag;
2862 struct mbuf *mreq, *mrep, *md, *mb;
2863 nfsfh_t *nfhp;
2864 const int v3 = NFS_ISV3(dvp);
2865
2866 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2867 nfsm_reqhead(dnp, NFSPROC_LOOKUP,
2868 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2869 nfsm_fhtom(dnp, v3);
2870 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2871 nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred);
2872 if (npp && !error) {
2873 nfsm_getfh(nfhp, fhlen, v3);
2874 if (*npp) {
2875 np = *npp;
2876 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2877 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2878 np->n_fhp = &np->n_fh;
2879 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2880 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2881 memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
2882 np->n_fhsize = fhlen;
2883 newvp = NFSTOV(np);
2884 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2885 VREF(dvp);
2886 newvp = dvp;
2887 } else {
2888 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2889 if (error) {
2890 m_freem(mrep);
2891 return (error);
2892 }
2893 newvp = NFSTOV(np);
2894 }
2895 if (v3) {
2896 nfsm_postop_attr(newvp, attrflag, 0);
2897 if (!attrflag && *npp == NULL) {
2898 m_freem(mrep);
2899 vput(newvp);
2900 return (ENOENT);
2901 }
2902 } else
2903 nfsm_loadattr(newvp, (struct vattr *)0, 0);
2904 }
2905 nfsm_reqdone;
2906 if (npp && *npp == NULL) {
2907 if (error) {
2908 if (newvp)
2909 vput(newvp);
2910 } else
2911 *npp = np;
2912 }
2913 return (error);
2914 }
2915
2916 /*
2917 * Nfs Version 3 commit rpc
2918 */
2919 int
2920 nfs_commit(vp, offset, cnt, procp)
2921 struct vnode *vp;
2922 off_t offset;
2923 uint32_t cnt;
2924 struct proc *procp;
2925 {
2926 caddr_t cp;
2927 u_int32_t *tl;
2928 int32_t t1, t2;
2929 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2930 caddr_t bpos, dpos, cp2;
2931 int error = 0, wccflag = NFSV3_WCCRATTR;
2932 struct mbuf *mreq, *mrep, *md, *mb;
2933 struct nfsnode *np;
2934
2935 KASSERT(NFS_ISV3(vp));
2936
2937 #ifdef NFS_DEBUG_COMMIT
2938 printf("commit %lu - %lu\n", (unsigned long)offset,
2939 (unsigned long)(offset + cnt));
2940 #endif
2941
2942 simple_lock(&nmp->nm_slock);
2943 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
2944 simple_unlock(&nmp->nm_slock);
2945 return (0);
2946 }
2947 simple_unlock(&nmp->nm_slock);
2948 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2949 np = VTONFS(vp);
2950 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
2951 nfsm_fhtom(np, 1);
2952 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2953 txdr_hyper(offset, tl);
2954 tl += 2;
2955 *tl = txdr_unsigned(cnt);
2956 nfsm_request(np, NFSPROC_COMMIT, procp, VTONFS(vp)->n_wcred);
2957 nfsm_wcc_data(vp, wccflag, 0);
2958 if (!error) {
2959 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2960 simple_lock(&nmp->nm_slock);
2961 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
2962 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
2963 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
2964 error = NFSERR_STALEWRITEVERF;
2965 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
2966 }
2967 simple_unlock(&nmp->nm_slock);
2968 }
2969 nfsm_reqdone;
2970 return (error);
2971 }
2972
2973 /*
2974 * Kludge City..
2975 * - make nfs_bmap() essentially a no-op that does no translation
2976 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2977 * (Maybe I could use the process's page mapping, but I was concerned that
2978 * Kernel Write might not be enabled and also figured copyout() would do
2979 * a lot more work than memcpy() and also it currently happens in the
2980 * context of the swapper process (2).
2981 */
2982 int
2983 nfs_bmap(v)
2984 void *v;
2985 {
2986 struct vop_bmap_args /* {
2987 struct vnode *a_vp;
2988 daddr_t a_bn;
2989 struct vnode **a_vpp;
2990 daddr_t *a_bnp;
2991 int *a_runp;
2992 } */ *ap = v;
2993 struct vnode *vp = ap->a_vp;
2994 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
2995
2996 if (ap->a_vpp != NULL)
2997 *ap->a_vpp = vp;
2998 if (ap->a_bnp != NULL)
2999 *ap->a_bnp = ap->a_bn << bshift;
3000 if (ap->a_runp != NULL)
3001 *ap->a_runp = 1024 * 1024; /* XXX */
3002 return (0);
3003 }
3004
3005 /*
3006 * Strategy routine.
3007 * For async requests when nfsiod(s) are running, queue the request by
3008 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3009 * request.
3010 */
3011 int
3012 nfs_strategy(v)
3013 void *v;
3014 {
3015 struct vop_strategy_args *ap = v;
3016 struct buf *bp = ap->a_bp;
3017 struct proc *p;
3018 int error = 0;
3019
3020 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3021 panic("nfs physio/async");
3022 if (bp->b_flags & B_ASYNC)
3023 p = NULL;
3024 else
3025 p = curproc; /* XXX */
3026
3027 /*
3028 * If the op is asynchronous and an i/o daemon is waiting
3029 * queue the request, wake it up and wait for completion
3030 * otherwise just do it ourselves.
3031 */
3032
3033 if ((bp->b_flags & B_ASYNC) == 0 ||
3034 nfs_asyncio(bp))
3035 error = nfs_doio(bp, p);
3036 return (error);
3037 }
3038
3039 /*
3040 * fsync vnode op. Just call nfs_flush() with commit == 1.
3041 */
3042 /* ARGSUSED */
3043 int
3044 nfs_fsync(v)
3045 void *v;
3046 {
3047 struct vop_fsync_args /* {
3048 struct vnodeop_desc *a_desc;
3049 struct vnode * a_vp;
3050 struct ucred * a_cred;
3051 int a_flags;
3052 off_t offlo;
3053 off_t offhi;
3054 struct proc * a_p;
3055 } */ *ap = v;
3056
3057 return (nfs_flush(ap->a_vp, ap->a_cred,
3058 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1));
3059 }
3060
3061 /*
3062 * Flush all the data associated with a vnode.
3063 */
3064 int
3065 nfs_flush(vp, cred, waitfor, p, commit)
3066 struct vnode *vp;
3067 struct ucred *cred;
3068 int waitfor;
3069 struct proc *p;
3070 int commit;
3071 {
3072 struct nfsnode *np = VTONFS(vp);
3073 int error;
3074 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3075 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3076
3077 simple_lock(&vp->v_interlock);
3078 error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3079 if (np->n_flag & NWRITEERR) {
3080 error = np->n_error;
3081 np->n_flag &= ~NWRITEERR;
3082 }
3083 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3084 return (error);
3085 }
3086
3087 /*
3088 * Return POSIX pathconf information applicable to nfs.
3089 *
3090 * N.B. The NFS V2 protocol doesn't support this RPC.
3091 */
3092 /* ARGSUSED */
3093 int
3094 nfs_pathconf(v)
3095 void *v;
3096 {
3097 struct vop_pathconf_args /* {
3098 struct vnode *a_vp;
3099 int a_name;
3100 register_t *a_retval;
3101 } */ *ap = v;
3102 struct nfsv3_pathconf *pcp;
3103 struct vnode *vp = ap->a_vp;
3104 struct mbuf *mreq, *mrep, *md, *mb;
3105 int32_t t1, t2;
3106 u_int32_t *tl;
3107 caddr_t bpos, dpos, cp, cp2;
3108 int error = 0, attrflag;
3109 #ifndef NFS_V2_ONLY
3110 struct nfsmount *nmp;
3111 unsigned int l;
3112 u_int64_t maxsize;
3113 #endif
3114 const int v3 = NFS_ISV3(vp);
3115 struct nfsnode *np = VTONFS(vp);
3116
3117 switch (ap->a_name) {
3118 /* Names that can be resolved locally. */
3119 case _PC_PIPE_BUF:
3120 *ap->a_retval = PIPE_BUF;
3121 break;
3122 case _PC_SYNC_IO:
3123 *ap->a_retval = 1;
3124 break;
3125 /* Names that cannot be resolved locally; do an RPC, if possible. */
3126 case _PC_LINK_MAX:
3127 case _PC_NAME_MAX:
3128 case _PC_CHOWN_RESTRICTED:
3129 case _PC_NO_TRUNC:
3130 if (!v3) {
3131 error = EINVAL;
3132 break;
3133 }
3134 nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3135 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3136 nfsm_fhtom(np, 1);
3137 nfsm_request(np, NFSPROC_PATHCONF,
3138 curproc, curproc->p_ucred); /* XXX */
3139 nfsm_postop_attr(vp, attrflag, 0);
3140 if (!error) {
3141 nfsm_dissect(pcp, struct nfsv3_pathconf *,
3142 NFSX_V3PATHCONF);
3143 switch (ap->a_name) {
3144 case _PC_LINK_MAX:
3145 *ap->a_retval =
3146 fxdr_unsigned(register_t, pcp->pc_linkmax);
3147 break;
3148 case _PC_NAME_MAX:
3149 *ap->a_retval =
3150 fxdr_unsigned(register_t, pcp->pc_namemax);
3151 break;
3152 case _PC_CHOWN_RESTRICTED:
3153 *ap->a_retval =
3154 (pcp->pc_chownrestricted == nfs_true);
3155 break;
3156 case _PC_NO_TRUNC:
3157 *ap->a_retval =
3158 (pcp->pc_notrunc == nfs_true);
3159 break;
3160 }
3161 }
3162 nfsm_reqdone;
3163 break;
3164 case _PC_FILESIZEBITS:
3165 #ifndef NFS_V2_ONLY
3166 if (v3) {
3167 nmp = VFSTONFS(vp->v_mount);
3168 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3169 if ((error = nfs_fsinfo(nmp, vp,
3170 curproc->p_ucred, curproc)) != 0) /* XXX */
3171 break;
3172 for (l = 0, maxsize = nmp->nm_maxfilesize;
3173 (maxsize >> l) > 0; l++)
3174 ;
3175 *ap->a_retval = l + 1;
3176 } else
3177 #endif
3178 {
3179 *ap->a_retval = 32; /* NFS V2 limitation */
3180 }
3181 break;
3182 default:
3183 error = EINVAL;
3184 break;
3185 }
3186
3187 return (error);
3188 }
3189
3190 /*
3191 * NFS advisory byte-level locks.
3192 */
3193 int
3194 nfs_advlock(v)
3195 void *v;
3196 {
3197 struct vop_advlock_args /* {
3198 struct vnode *a_vp;
3199 caddr_t a_id;
3200 int a_op;
3201 struct flock *a_fl;
3202 int a_flags;
3203 } */ *ap = v;
3204 struct nfsnode *np = VTONFS(ap->a_vp);
3205
3206 return lf_advlock(ap, &np->n_lockf, np->n_size);
3207 }
3208
3209 /*
3210 * Print out the contents of an nfsnode.
3211 */
3212 int
3213 nfs_print(v)
3214 void *v;
3215 {
3216 struct vop_print_args /* {
3217 struct vnode *a_vp;
3218 } */ *ap = v;
3219 struct vnode *vp = ap->a_vp;
3220 struct nfsnode *np = VTONFS(vp);
3221
3222 printf("tag VT_NFS, fileid %ld fsid 0x%lx",
3223 np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3224 if (vp->v_type == VFIFO)
3225 fifo_printinfo(vp);
3226 printf("\n");
3227 return (0);
3228 }
3229
3230 /*
3231 * NFS file truncation.
3232 */
3233 int
3234 nfs_truncate(v)
3235 void *v;
3236 {
3237 #if 0
3238 struct vop_truncate_args /* {
3239 struct vnode *a_vp;
3240 off_t a_length;
3241 int a_flags;
3242 struct ucred *a_cred;
3243 struct proc *a_p;
3244 } */ *ap = v;
3245 #endif
3246
3247 /* Use nfs_setattr */
3248 return (EOPNOTSUPP);
3249 }
3250
3251 /*
3252 * NFS update.
3253 */
3254 int
3255 nfs_update(v)
3256 void *v;
3257 #if 0
3258 struct vop_update_args /* {
3259 struct vnode *a_vp;
3260 struct timespec *a_ta;
3261 struct timespec *a_tm;
3262 int a_waitfor;
3263 } */ *ap = v;
3264 #endif
3265 {
3266
3267 /* Use nfs_setattr */
3268 return (EOPNOTSUPP);
3269 }
3270
3271 /*
3272 * Just call bwrite().
3273 */
3274 int
3275 nfs_bwrite(v)
3276 void *v;
3277 {
3278 struct vop_bwrite_args /* {
3279 struct vnode *a_bp;
3280 } */ *ap = v;
3281
3282 return (bwrite(ap->a_bp));
3283 }
3284
3285 /*
3286 * nfs unlock wrapper.
3287 */
3288 int
3289 nfs_unlock(void *v)
3290 {
3291 struct vop_unlock_args /* {
3292 struct vnode *a_vp;
3293 int a_flags;
3294 } */ *ap = v;
3295 struct vnode *vp = ap->a_vp;
3296
3297 /*
3298 * VOP_UNLOCK can be called by nfs_loadattrcache
3299 * with v_data == 0.
3300 */
3301 if (VTONFS(vp)) {
3302 nfs_delayedtruncate(vp);
3303 }
3304
3305 return genfs_unlock(v);
3306 }
3307
3308 /*
3309 * nfs special file access vnode op.
3310 * Essentially just get vattr and then imitate iaccess() since the device is
3311 * local to the client.
3312 */
3313 int
3314 nfsspec_access(v)
3315 void *v;
3316 {
3317 struct vop_access_args /* {
3318 struct vnode *a_vp;
3319 int a_mode;
3320 struct ucred *a_cred;
3321 struct proc *a_p;
3322 } */ *ap = v;
3323 struct vattr va;
3324 struct vnode *vp = ap->a_vp;
3325 int error;
3326
3327 error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3328 if (error)
3329 return (error);
3330
3331 /*
3332 * Disallow write attempts on filesystems mounted read-only;
3333 * unless the file is a socket, fifo, or a block or character
3334 * device resident on the filesystem.
3335 */
3336 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3337 switch (vp->v_type) {
3338 case VREG:
3339 case VDIR:
3340 case VLNK:
3341 return (EROFS);
3342 default:
3343 break;
3344 }
3345 }
3346
3347 return (vaccess(va.va_type, va.va_mode,
3348 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3349 }
3350
3351 /*
3352 * Read wrapper for special devices.
3353 */
3354 int
3355 nfsspec_read(v)
3356 void *v;
3357 {
3358 struct vop_read_args /* {
3359 struct vnode *a_vp;
3360 struct uio *a_uio;
3361 int a_ioflag;
3362 struct ucred *a_cred;
3363 } */ *ap = v;
3364 struct nfsnode *np = VTONFS(ap->a_vp);
3365
3366 /*
3367 * Set access flag.
3368 */
3369 np->n_flag |= NACC;
3370 np->n_atim.tv_sec = time.tv_sec;
3371 np->n_atim.tv_nsec = time.tv_usec * 1000;
3372 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3373 }
3374
3375 /*
3376 * Write wrapper for special devices.
3377 */
3378 int
3379 nfsspec_write(v)
3380 void *v;
3381 {
3382 struct vop_write_args /* {
3383 struct vnode *a_vp;
3384 struct uio *a_uio;
3385 int a_ioflag;
3386 struct ucred *a_cred;
3387 } */ *ap = v;
3388 struct nfsnode *np = VTONFS(ap->a_vp);
3389
3390 /*
3391 * Set update flag.
3392 */
3393 np->n_flag |= NUPD;
3394 np->n_mtim.tv_sec = time.tv_sec;
3395 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3396 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3397 }
3398
3399 /*
3400 * Close wrapper for special devices.
3401 *
3402 * Update the times on the nfsnode then do device close.
3403 */
3404 int
3405 nfsspec_close(v)
3406 void *v;
3407 {
3408 struct vop_close_args /* {
3409 struct vnode *a_vp;
3410 int a_fflag;
3411 struct ucred *a_cred;
3412 struct proc *a_p;
3413 } */ *ap = v;
3414 struct vnode *vp = ap->a_vp;
3415 struct nfsnode *np = VTONFS(vp);
3416 struct vattr vattr;
3417
3418 if (np->n_flag & (NACC | NUPD)) {
3419 np->n_flag |= NCHG;
3420 if (vp->v_usecount == 1 &&
3421 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3422 VATTR_NULL(&vattr);
3423 if (np->n_flag & NACC)
3424 vattr.va_atime = np->n_atim;
3425 if (np->n_flag & NUPD)
3426 vattr.va_mtime = np->n_mtim;
3427 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3428 }
3429 }
3430 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3431 }
3432
3433 /*
3434 * Read wrapper for fifos.
3435 */
3436 int
3437 nfsfifo_read(v)
3438 void *v;
3439 {
3440 struct vop_read_args /* {
3441 struct vnode *a_vp;
3442 struct uio *a_uio;
3443 int a_ioflag;
3444 struct ucred *a_cred;
3445 } */ *ap = v;
3446 struct nfsnode *np = VTONFS(ap->a_vp);
3447
3448 /*
3449 * Set access flag.
3450 */
3451 np->n_flag |= NACC;
3452 np->n_atim.tv_sec = time.tv_sec;
3453 np->n_atim.tv_nsec = time.tv_usec * 1000;
3454 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3455 }
3456
3457 /*
3458 * Write wrapper for fifos.
3459 */
3460 int
3461 nfsfifo_write(v)
3462 void *v;
3463 {
3464 struct vop_write_args /* {
3465 struct vnode *a_vp;
3466 struct uio *a_uio;
3467 int a_ioflag;
3468 struct ucred *a_cred;
3469 } */ *ap = v;
3470 struct nfsnode *np = VTONFS(ap->a_vp);
3471
3472 /*
3473 * Set update flag.
3474 */
3475 np->n_flag |= NUPD;
3476 np->n_mtim.tv_sec = time.tv_sec;
3477 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3478 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3479 }
3480
3481 /*
3482 * Close wrapper for fifos.
3483 *
3484 * Update the times on the nfsnode then do fifo close.
3485 */
3486 int
3487 nfsfifo_close(v)
3488 void *v;
3489 {
3490 struct vop_close_args /* {
3491 struct vnode *a_vp;
3492 int a_fflag;
3493 struct ucred *a_cred;
3494 struct proc *a_p;
3495 } */ *ap = v;
3496 struct vnode *vp = ap->a_vp;
3497 struct nfsnode *np = VTONFS(vp);
3498 struct vattr vattr;
3499
3500 if (np->n_flag & (NACC | NUPD)) {
3501 if (np->n_flag & NACC) {
3502 np->n_atim.tv_sec = time.tv_sec;
3503 np->n_atim.tv_nsec = time.tv_usec * 1000;
3504 }
3505 if (np->n_flag & NUPD) {
3506 np->n_mtim.tv_sec = time.tv_sec;
3507 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3508 }
3509 np->n_flag |= NCHG;
3510 if (vp->v_usecount == 1 &&
3511 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3512 VATTR_NULL(&vattr);
3513 if (np->n_flag & NACC)
3514 vattr.va_atime = np->n_atim;
3515 if (np->n_flag & NUPD)
3516 vattr.va_mtime = np->n_mtim;
3517 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3518 }
3519 }
3520 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3521 }
3522