FreeBSD/Linux Kernel Cross Reference
sys/nfs/nfs_vnops.c
1 /* $NetBSD: nfs_vnops.c,v 1.186.2.10 2005/03/16 12:50:30 tron 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 1.186.2.10 2005/03/16 12:50:30 tron 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 static void nfs_cache_enter(struct vnode *, struct vnode *,
270 struct componentname *);
271
272 static void
273 nfs_cache_enter(struct vnode *dvp, struct vnode *vp,
274 struct componentname *cnp)
275 {
276 struct nfsnode *dnp = VTONFS(dvp);
277
278 if (vp != NULL) {
279 struct nfsnode *np = VTONFS(vp);
280
281 np->n_ctime = np->n_vattr->va_ctime.tv_sec;
282 }
283
284 if (!timespecisset(&dnp->n_nctime))
285 dnp->n_nctime = dnp->n_vattr->va_mtime;
286
287 cache_enter(dvp, vp, cnp);
288 }
289
290 /*
291 * nfs null call from vfs.
292 */
293 int
294 nfs_null(vp, cred, procp)
295 struct vnode *vp;
296 struct ucred *cred;
297 struct proc *procp;
298 {
299 caddr_t bpos, dpos;
300 int error = 0;
301 struct mbuf *mreq, *mrep, *md, *mb;
302 struct nfsnode *np = VTONFS(vp);
303
304 nfsm_reqhead(np, NFSPROC_NULL, 0);
305 nfsm_request(np, NFSPROC_NULL, procp, cred);
306 nfsm_reqdone;
307 return (error);
308 }
309
310 /*
311 * nfs access vnode op.
312 * For nfs version 2, just return ok. File accesses may fail later.
313 * For nfs version 3, use the access rpc to check accessibility. If file modes
314 * are changed on the server, accesses might still fail later.
315 */
316 int
317 nfs_access(v)
318 void *v;
319 {
320 struct vop_access_args /* {
321 struct vnode *a_vp;
322 int a_mode;
323 struct ucred *a_cred;
324 struct proc *a_p;
325 } */ *ap = v;
326 struct vnode *vp = ap->a_vp;
327 u_int32_t *tl;
328 caddr_t cp;
329 int32_t t1, t2;
330 caddr_t bpos, dpos, cp2;
331 int error = 0, attrflag, cachevalid;
332 struct mbuf *mreq, *mrep, *md, *mb;
333 u_int32_t mode, rmode;
334 const int v3 = NFS_ISV3(vp);
335 struct nfsnode *np = VTONFS(vp);
336
337 cachevalid = (np->n_accstamp != -1 &&
338 (time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) &&
339 np->n_accuid == ap->a_cred->cr_uid);
340
341 /*
342 * Check access cache first. If this request has been made for this
343 * uid shortly before, use the cached result.
344 */
345 if (cachevalid) {
346 if (!np->n_accerror) {
347 if ((np->n_accmode & ap->a_mode) == ap->a_mode)
348 return np->n_accerror;
349 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
350 return np->n_accerror;
351 }
352
353 /*
354 * For nfs v3, do an access rpc, otherwise you are stuck emulating
355 * ufs_access() locally using the vattr. This may not be correct,
356 * since the server may apply other access criteria such as
357 * client uid-->server uid mapping that we do not know about, but
358 * this is better than just returning anything that is lying about
359 * in the cache.
360 */
361 if (v3) {
362 nfsstats.rpccnt[NFSPROC_ACCESS]++;
363 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
364 nfsm_fhtom(np, v3);
365 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
366 if (ap->a_mode & VREAD)
367 mode = NFSV3ACCESS_READ;
368 else
369 mode = 0;
370 if (vp->v_type != VDIR) {
371 if (ap->a_mode & VWRITE)
372 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
373 if (ap->a_mode & VEXEC)
374 mode |= NFSV3ACCESS_EXECUTE;
375 } else {
376 if (ap->a_mode & VWRITE)
377 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
378 NFSV3ACCESS_DELETE);
379 if (ap->a_mode & VEXEC)
380 mode |= NFSV3ACCESS_LOOKUP;
381 }
382 *tl = txdr_unsigned(mode);
383 nfsm_request(np, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
384 nfsm_postop_attr(vp, attrflag, 0);
385 if (!error) {
386 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
387 rmode = fxdr_unsigned(u_int32_t, *tl);
388 /*
389 * The NFS V3 spec does not clarify whether or not
390 * the returned access bits can be a superset of
391 * the ones requested, so...
392 */
393 if ((rmode & mode) != mode)
394 error = EACCES;
395 }
396 nfsm_reqdone;
397 } else
398 return (nfsspec_access(ap));
399 /*
400 * Disallow write attempts on filesystems mounted read-only;
401 * unless the file is a socket, fifo, or a block or character
402 * device resident on the filesystem.
403 */
404 if (!error && (ap->a_mode & VWRITE) &&
405 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
406 switch (vp->v_type) {
407 case VREG:
408 case VDIR:
409 case VLNK:
410 error = EROFS;
411 default:
412 break;
413 }
414 }
415
416 if (!error || error == EACCES) {
417 /*
418 * If we got the same result as for a previous,
419 * different request, OR it in. Don't update
420 * the timestamp in that case.
421 */
422 if (cachevalid && np->n_accstamp != -1 &&
423 error == np->n_accerror) {
424 if (!error)
425 np->n_accmode |= ap->a_mode;
426 else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
427 np->n_accmode = ap->a_mode;
428 } else {
429 np->n_accstamp = time.tv_sec;
430 np->n_accuid = ap->a_cred->cr_uid;
431 np->n_accmode = ap->a_mode;
432 np->n_accerror = error;
433 }
434 }
435
436 return (error);
437 }
438
439 /*
440 * nfs open vnode op
441 * Check to see if the type is ok
442 * and that deletion is not in progress.
443 * For paged in text files, you will need to flush the page cache
444 * if consistency is lost.
445 */
446 /* ARGSUSED */
447 int
448 nfs_open(v)
449 void *v;
450 {
451 struct vop_open_args /* {
452 struct vnode *a_vp;
453 int a_mode;
454 struct ucred *a_cred;
455 struct proc *a_p;
456 } */ *ap = v;
457 struct vnode *vp = ap->a_vp;
458 struct nfsnode *np = VTONFS(vp);
459 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
460 struct vattr vattr;
461 int error;
462
463 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
464 return (EACCES);
465 }
466
467 if (ap->a_mode & FREAD) {
468 if (np->n_rcred != NULL)
469 crfree(np->n_rcred);
470 np->n_rcred = ap->a_cred;
471 crhold(np->n_rcred);
472 }
473 if (ap->a_mode & FWRITE) {
474 if (np->n_wcred != NULL)
475 crfree(np->n_wcred);
476 np->n_wcred = ap->a_cred;
477 crhold(np->n_wcred);
478 }
479
480 #ifndef NFS_V2_ONLY
481 /*
482 * Get a valid lease. If cached data is stale, flush it.
483 */
484 if (nmp->nm_flag & NFSMNT_NQNFS) {
485 if (NQNFS_CKINVALID(vp, np, ND_READ)) {
486 do {
487 error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
488 ap->a_p);
489 } while (error == NQNFS_EXPIRED);
490 if (error)
491 return (error);
492 if (np->n_lrev != np->n_brev ||
493 (np->n_flag & NQNFSNONCACHE)) {
494 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
495 ap->a_p, 1)) == EINTR)
496 return (error);
497 np->n_brev = np->n_lrev;
498 }
499 }
500 } else
501 #endif
502 {
503 if (np->n_flag & NMODIFIED) {
504 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
505 ap->a_p, 1)) == EINTR)
506 return (error);
507 NFS_INVALIDATE_ATTRCACHE(np);
508 if (vp->v_type == VDIR) {
509 nfs_invaldircache(vp, 0);
510 np->n_direofoffset = 0;
511 }
512 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
513 if (error)
514 return (error);
515 np->n_mtime = vattr.va_mtime;
516 } else {
517 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
518 if (error)
519 return (error);
520 if (timespeccmp(&np->n_mtime, &vattr.va_mtime, !=)) {
521 if (vp->v_type == VDIR) {
522 nfs_invaldircache(vp, 0);
523 np->n_direofoffset = 0;
524 }
525 if ((error = nfs_vinvalbuf(vp, V_SAVE,
526 ap->a_cred, ap->a_p, 1)) == EINTR)
527 return (error);
528 np->n_mtime = vattr.va_mtime;
529 }
530 }
531 }
532 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
533 NFS_INVALIDATE_ATTRCACHE(np); /* For Open/Close consistency */
534 return (0);
535 }
536
537 /*
538 * nfs close vnode op
539 * What an NFS client should do upon close after writing is a debatable issue.
540 * Most NFS clients push delayed writes to the server upon close, basically for
541 * two reasons:
542 * 1 - So that any write errors may be reported back to the client process
543 * doing the close system call. By far the two most likely errors are
544 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
545 * 2 - To put a worst case upper bound on cache inconsistency between
546 * multiple clients for the file.
547 * There is also a consistency problem for Version 2 of the protocol w.r.t.
548 * not being able to tell if other clients are writing a file concurrently,
549 * since there is no way of knowing if the changed modify time in the reply
550 * is only due to the write for this client.
551 * (NFS Version 3 provides weak cache consistency data in the reply that
552 * should be sufficient to detect and handle this case.)
553 *
554 * The current code does the following:
555 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
556 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
557 * or commit them (this satisfies 1 and 2 except for the
558 * case where the server crashes after this close but
559 * before the commit RPC, which is felt to be "good
560 * enough". Changing the last argument to nfs_flush() to
561 * a 1 would force a commit operation, if it is felt a
562 * commit is necessary now.
563 * for NQNFS - do nothing now, since 2 is dealt with via leases and
564 * 1 should be dealt with via an fsync() system call for
565 * cases where write errors are important.
566 */
567 /* ARGSUSED */
568 int
569 nfs_close(v)
570 void *v;
571 {
572 struct vop_close_args /* {
573 struct vnodeop_desc *a_desc;
574 struct vnode *a_vp;
575 int a_fflag;
576 struct ucred *a_cred;
577 struct proc *a_p;
578 } */ *ap = v;
579 struct vnode *vp = ap->a_vp;
580 struct nfsnode *np = VTONFS(vp);
581 int error = 0;
582 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
583
584 if (vp->v_type == VREG) {
585 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
586 (np->n_flag & NMODIFIED)) {
587 if (NFS_ISV3(vp)) {
588 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
589 np->n_flag &= ~NMODIFIED;
590 } else
591 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
592 NFS_INVALIDATE_ATTRCACHE(np);
593 }
594 if (np->n_flag & NWRITEERR) {
595 np->n_flag &= ~NWRITEERR;
596 error = np->n_error;
597 }
598 }
599 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
600 return (error);
601 }
602
603 /*
604 * nfs getattr call from vfs.
605 */
606 int
607 nfs_getattr(v)
608 void *v;
609 {
610 struct vop_getattr_args /* {
611 struct vnode *a_vp;
612 struct vattr *a_vap;
613 struct ucred *a_cred;
614 struct proc *a_p;
615 } */ *ap = v;
616 struct vnode *vp = ap->a_vp;
617 struct nfsnode *np = VTONFS(vp);
618 caddr_t cp;
619 u_int32_t *tl;
620 int32_t t1, t2;
621 caddr_t bpos, dpos;
622 int error = 0;
623 struct mbuf *mreq, *mrep, *md, *mb;
624 const int v3 = NFS_ISV3(vp);
625
626 /*
627 * Update local times for special files.
628 */
629 if (np->n_flag & (NACC | NUPD))
630 np->n_flag |= NCHG;
631
632 /*
633 * if we have delayed truncation, do it now.
634 */
635 nfs_delayedtruncate(vp);
636
637 /*
638 * First look in the cache.
639 */
640 if (nfs_getattrcache(vp, ap->a_vap) == 0)
641 return (0);
642 nfsstats.rpccnt[NFSPROC_GETATTR]++;
643 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
644 nfsm_fhtom(np, v3);
645 nfsm_request(np, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
646 if (!error) {
647 nfsm_loadattr(vp, ap->a_vap, 0);
648 if (vp->v_type == VDIR &&
649 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
650 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
651 }
652 nfsm_reqdone;
653 return (error);
654 }
655
656 /*
657 * nfs setattr call.
658 */
659 int
660 nfs_setattr(v)
661 void *v;
662 {
663 struct vop_setattr_args /* {
664 struct vnodeop_desc *a_desc;
665 struct vnode *a_vp;
666 struct vattr *a_vap;
667 struct ucred *a_cred;
668 struct proc *a_p;
669 } */ *ap = v;
670 struct vnode *vp = ap->a_vp;
671 struct nfsnode *np = VTONFS(vp);
672 struct vattr *vap = ap->a_vap;
673 int error = 0;
674 u_quad_t tsize = 0;
675
676 /*
677 * Setting of flags is not supported.
678 */
679 if (vap->va_flags != VNOVAL)
680 return (EOPNOTSUPP);
681
682 /*
683 * Disallow write attempts if the filesystem is mounted read-only.
684 */
685 if ((vap->va_uid != (uid_t)VNOVAL ||
686 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
687 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
688 (vp->v_mount->mnt_flag & MNT_RDONLY))
689 return (EROFS);
690 if (vap->va_size != VNOVAL) {
691 switch (vp->v_type) {
692 case VDIR:
693 return (EISDIR);
694 case VCHR:
695 case VBLK:
696 case VSOCK:
697 case VFIFO:
698 if (vap->va_mtime.tv_sec == VNOVAL &&
699 vap->va_atime.tv_sec == VNOVAL &&
700 vap->va_mode == (mode_t)VNOVAL &&
701 vap->va_uid == (uid_t)VNOVAL &&
702 vap->va_gid == (gid_t)VNOVAL)
703 return (0);
704 vap->va_size = VNOVAL;
705 break;
706 default:
707 /*
708 * Disallow write attempts if the filesystem is
709 * mounted read-only.
710 */
711 if (vp->v_mount->mnt_flag & MNT_RDONLY)
712 return (EROFS);
713 uvm_vnp_setsize(vp, vap->va_size);
714 tsize = np->n_size;
715 np->n_size = vap->va_size;
716 if (vap->va_size == 0)
717 error = nfs_vinvalbuf(vp, 0,
718 ap->a_cred, ap->a_p, 1);
719 else
720 error = nfs_vinvalbuf(vp, V_SAVE,
721 ap->a_cred, ap->a_p, 1);
722 if (error) {
723 uvm_vnp_setsize(vp, tsize);
724 return (error);
725 }
726 np->n_vattr->va_size = vap->va_size;
727 }
728 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
729 vap->va_atime.tv_sec != VNOVAL) &&
730 vp->v_type == VREG &&
731 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
732 ap->a_p, 1)) == EINTR)
733 return (error);
734 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
735 if (error && vap->va_size != VNOVAL) {
736 np->n_size = np->n_vattr->va_size = tsize;
737 uvm_vnp_setsize(vp, np->n_size);
738 }
739 VN_KNOTE(vp, NOTE_ATTRIB);
740 return (error);
741 }
742
743 /*
744 * Do an nfs setattr rpc.
745 */
746 int
747 nfs_setattrrpc(vp, vap, cred, procp)
748 struct vnode *vp;
749 struct vattr *vap;
750 struct ucred *cred;
751 struct proc *procp;
752 {
753 struct nfsv2_sattr *sp;
754 caddr_t cp;
755 int32_t t1, t2;
756 caddr_t bpos, dpos, cp2;
757 u_int32_t *tl;
758 int error = 0, wccflag = NFSV3_WCCRATTR;
759 struct mbuf *mreq, *mrep, *md, *mb;
760 const int v3 = NFS_ISV3(vp);
761 struct nfsnode *np = VTONFS(vp);
762
763 nfsstats.rpccnt[NFSPROC_SETATTR]++;
764 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
765 nfsm_fhtom(np, v3);
766 if (v3) {
767 nfsm_v3attrbuild(vap, TRUE);
768 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
769 *tl = nfs_false;
770 } else {
771 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
772 if (vap->va_mode == (mode_t)VNOVAL)
773 sp->sa_mode = nfs_xdrneg1;
774 else
775 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
776 if (vap->va_uid == (uid_t)VNOVAL)
777 sp->sa_uid = nfs_xdrneg1;
778 else
779 sp->sa_uid = txdr_unsigned(vap->va_uid);
780 if (vap->va_gid == (gid_t)VNOVAL)
781 sp->sa_gid = nfs_xdrneg1;
782 else
783 sp->sa_gid = txdr_unsigned(vap->va_gid);
784 sp->sa_size = txdr_unsigned(vap->va_size);
785 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
786 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
787 }
788 nfsm_request(np, NFSPROC_SETATTR, procp, cred);
789 if (v3) {
790 nfsm_wcc_data(vp, wccflag, 0);
791 } else
792 nfsm_loadattr(vp, (struct vattr *)0, 0);
793 nfsm_reqdone;
794 return (error);
795 }
796
797 /*
798 * nfs lookup call, one step at a time...
799 * First look in cache
800 * If not found, unlock the directory nfsnode and do the rpc
801 *
802 * This code is full of lock/unlock statements and checks, because
803 * we continue after cache_lookup has finished (we need to check
804 * with the attr cache and do an rpc if it has timed out). This means
805 * that the locking effects of cache_lookup have to be taken into
806 * account.
807 */
808 int
809 nfs_lookup(v)
810 void *v;
811 {
812 struct vop_lookup_args /* {
813 struct vnodeop_desc *a_desc;
814 struct vnode *a_dvp;
815 struct vnode **a_vpp;
816 struct componentname *a_cnp;
817 } */ *ap = v;
818 struct componentname *cnp = ap->a_cnp;
819 struct vnode *dvp = ap->a_dvp;
820 struct vnode **vpp = ap->a_vpp;
821 int flags;
822 struct vnode *newvp;
823 u_int32_t *tl;
824 caddr_t cp;
825 int32_t t1, t2;
826 caddr_t bpos, dpos, cp2;
827 struct mbuf *mreq, *mrep, *md, *mb;
828 long len;
829 nfsfh_t *fhp;
830 struct nfsnode *np;
831 int lockparent, wantparent, error = 0, attrflag, fhsize;
832 const int v3 = NFS_ISV3(dvp);
833
834 cnp->cn_flags &= ~PDIRUNLOCK;
835 flags = cnp->cn_flags;
836
837 *vpp = NULLVP;
838 newvp = NULLVP;
839 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
840 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
841 return (EROFS);
842 if (dvp->v_type != VDIR)
843 return (ENOTDIR);
844
845 lockparent = flags & LOCKPARENT;
846 wantparent = flags & (LOCKPARENT|WANTPARENT);
847 np = VTONFS(dvp);
848
849 /*
850 * Before tediously performing a linear scan of the directory,
851 * check the name cache to see if the directory/name pair
852 * we are looking for is known already.
853 * If the directory/name pair is found in the name cache,
854 * we have to ensure the directory has not changed from
855 * the time the cache entry has been created. If it has,
856 * the cache entry has to be ignored.
857 */
858 if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) {
859 struct vattr vattr;
860 int err2;
861
862 if (error && error != ENOENT) {
863 *vpp = NULLVP;
864 return error;
865 }
866
867 if (cnp->cn_flags & PDIRUNLOCK) {
868 err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
869 if (err2 != 0) {
870 *vpp = NULLVP;
871 return err2;
872 }
873 cnp->cn_flags &= ~PDIRUNLOCK;
874 }
875
876 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
877 if (err2 != 0) {
878 if (error == 0) {
879 if (*vpp != dvp)
880 vput(*vpp);
881 else
882 vrele(*vpp);
883 }
884 *vpp = NULLVP;
885 return err2;
886 }
887
888 if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
889 cnp->cn_proc) || timespeccmp(&vattr.va_mtime,
890 &VTONFS(dvp)->n_nctime, !=)) {
891 if (error == 0) {
892 if (*vpp != dvp)
893 vput(*vpp);
894 else
895 vrele(*vpp);
896 *vpp = NULLVP;
897 }
898 cache_purge1(dvp, NULL, PURGE_CHILDREN);
899 timespecclear(&np->n_nctime);
900 goto dorpc;
901 }
902
903 if (error == ENOENT) {
904 return ENOENT;
905 }
906
907 newvp = *vpp;
908 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
909 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime)
910 {
911 nfsstats.lookupcache_hits++;
912 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
913 cnp->cn_flags |= SAVENAME;
914 if ((!lockparent || !(flags & ISLASTCN)) &&
915 newvp != dvp)
916 VOP_UNLOCK(dvp, 0);
917 KASSERT(newvp->v_type != VNON);
918 return (0);
919 }
920 cache_purge1(newvp, NULL, PURGE_PARENTS);
921 if (newvp != dvp)
922 vput(newvp);
923 else
924 vrele(newvp);
925 *vpp = NULLVP;
926 }
927 dorpc:
928 error = 0;
929 newvp = NULLVP;
930 nfsstats.lookupcache_misses++;
931 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
932 len = cnp->cn_namelen;
933 nfsm_reqhead(np, NFSPROC_LOOKUP,
934 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
935 nfsm_fhtom(np, v3);
936 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
937 nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
938 if (error) {
939 nfsm_postop_attr(dvp, attrflag, 0);
940 m_freem(mrep);
941 goto nfsmout;
942 }
943 nfsm_getfh(fhp, fhsize, v3);
944
945 /*
946 * Handle RENAME case...
947 */
948 if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
949 if (NFS_CMPFH(np, fhp, fhsize)) {
950 m_freem(mrep);
951 return (EISDIR);
952 }
953 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
954 if (error) {
955 m_freem(mrep);
956 return error;
957 }
958 newvp = NFSTOV(np);
959 if (v3) {
960 nfsm_postop_attr(newvp, attrflag, 0);
961 nfsm_postop_attr(dvp, attrflag, 0);
962 } else
963 nfsm_loadattr(newvp, (struct vattr *)0, 0);
964 *vpp = newvp;
965 m_freem(mrep);
966 cnp->cn_flags |= SAVENAME;
967 if (!lockparent) {
968 VOP_UNLOCK(dvp, 0);
969 cnp->cn_flags |= PDIRUNLOCK;
970 }
971 goto validate;
972 }
973
974 /*
975 * The postop attr handling is duplicated for each if case,
976 * because it should be done while dvp is locked (unlocking
977 * dvp is different for each case).
978 */
979
980 if (NFS_CMPFH(np, fhp, fhsize)) {
981 /*
982 * "." lookup
983 */
984 VREF(dvp);
985 newvp = dvp;
986 if (v3) {
987 nfsm_postop_attr(newvp, attrflag, 0);
988 nfsm_postop_attr(dvp, attrflag, 0);
989 } else
990 nfsm_loadattr(newvp, (struct vattr *)0, 0);
991 } else if (flags & ISDOTDOT) {
992 /*
993 * ".." lookup
994 */
995 VOP_UNLOCK(dvp, 0);
996 cnp->cn_flags |= PDIRUNLOCK;
997
998 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
999 if (error) {
1000 if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
1001 cnp->cn_flags &= ~PDIRUNLOCK;
1002 m_freem(mrep);
1003 return error;
1004 }
1005 newvp = NFSTOV(np);
1006
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
1013 if (lockparent && (flags & ISLASTCN)) {
1014 if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
1015 m_freem(mrep);
1016 vput(newvp);
1017 return error;
1018 }
1019 cnp->cn_flags &= ~PDIRUNLOCK;
1020 }
1021 } else {
1022 /*
1023 * Other lookups.
1024 */
1025 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
1026 if (error) {
1027 m_freem(mrep);
1028 return error;
1029 }
1030 newvp = NFSTOV(np);
1031 if (v3) {
1032 nfsm_postop_attr(newvp, attrflag, 0);
1033 nfsm_postop_attr(dvp, attrflag, 0);
1034 } else
1035 nfsm_loadattr(newvp, (struct vattr *)0, 0);
1036 if (!lockparent || !(flags & ISLASTCN)) {
1037 VOP_UNLOCK(dvp, 0);
1038 cnp->cn_flags |= PDIRUNLOCK;
1039 }
1040 }
1041 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1042 cnp->cn_flags |= SAVENAME;
1043 if ((cnp->cn_flags & MAKEENTRY) &&
1044 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1045 nfs_cache_enter(dvp, newvp, cnp);
1046 }
1047 *vpp = newvp;
1048 nfsm_reqdone;
1049 if (error) {
1050 /*
1051 * We get here only because of errors returned by
1052 * the RPC. Otherwise we'll have returned above
1053 * (the nfsm_* macros will jump to nfsm_reqdone
1054 * on error).
1055 */
1056 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1057 cnp->cn_nameiop != CREATE) {
1058 nfs_cache_enter(dvp, NULL, cnp);
1059 }
1060 if (newvp != NULLVP) {
1061 vrele(newvp);
1062 if (newvp != dvp)
1063 VOP_UNLOCK(newvp, 0);
1064 }
1065 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1066 (flags & ISLASTCN) && error == ENOENT) {
1067 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1068 error = EROFS;
1069 else
1070 error = EJUSTRETURN;
1071 }
1072 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1073 cnp->cn_flags |= SAVENAME;
1074 *vpp = NULL;
1075 return error;
1076 }
1077
1078 validate:
1079 /*
1080 * make sure we have valid type and size.
1081 */
1082
1083 newvp = *vpp;
1084 if (newvp->v_type == VNON) {
1085 struct vattr vattr; /* dummy */
1086
1087 KASSERT(VTONFS(newvp)->n_attrstamp == 0);
1088 error = VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc);
1089 if (error) {
1090 vput(newvp);
1091 *vpp = NULL;
1092 }
1093 }
1094
1095 return error;
1096 }
1097
1098 /*
1099 * nfs read call.
1100 * Just call nfs_bioread() to do the work.
1101 */
1102 int
1103 nfs_read(v)
1104 void *v;
1105 {
1106 struct vop_read_args /* {
1107 struct vnode *a_vp;
1108 struct uio *a_uio;
1109 int a_ioflag;
1110 struct ucred *a_cred;
1111 } */ *ap = v;
1112 struct vnode *vp = ap->a_vp;
1113
1114 if (vp->v_type != VREG)
1115 return (EPERM);
1116 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1117 }
1118
1119 /*
1120 * nfs readlink call
1121 */
1122 int
1123 nfs_readlink(v)
1124 void *v;
1125 {
1126 struct vop_readlink_args /* {
1127 struct vnode *a_vp;
1128 struct uio *a_uio;
1129 struct ucred *a_cred;
1130 } */ *ap = v;
1131 struct vnode *vp = ap->a_vp;
1132
1133 if (vp->v_type != VLNK)
1134 return (EPERM);
1135 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1136 }
1137
1138 /*
1139 * Do a readlink rpc.
1140 * Called by nfs_doio() from below the buffer cache.
1141 */
1142 int
1143 nfs_readlinkrpc(vp, uiop, cred)
1144 struct vnode *vp;
1145 struct uio *uiop;
1146 struct ucred *cred;
1147 {
1148 u_int32_t *tl;
1149 caddr_t cp;
1150 int32_t t1, t2;
1151 caddr_t bpos, dpos, cp2;
1152 int error = 0, attrflag;
1153 uint32_t len;
1154 struct mbuf *mreq, *mrep, *md, *mb;
1155 const int v3 = NFS_ISV3(vp);
1156 struct nfsnode *np = VTONFS(vp);
1157
1158 nfsstats.rpccnt[NFSPROC_READLINK]++;
1159 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1160 nfsm_fhtom(np, v3);
1161 nfsm_request(np, NFSPROC_READLINK, uiop->uio_procp, cred);
1162 if (v3)
1163 nfsm_postop_attr(vp, attrflag, 0);
1164 if (!error) {
1165 if (v3) {
1166 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1167 len = fxdr_unsigned(uint32_t, *tl);
1168 if (len > MAXPATHLEN) {
1169 /*
1170 * this pathname is too long for us.
1171 */
1172 m_freem(mrep);
1173 /* Solaris returns EINVAL. should we follow? */
1174 error = ENAMETOOLONG;
1175 goto nfsmout;
1176 }
1177 } else {
1178 nfsm_strsiz(len, NFS_MAXPATHLEN);
1179 }
1180 nfsm_mtouio(uiop, len);
1181 }
1182 nfsm_reqdone;
1183 return (error);
1184 }
1185
1186 /*
1187 * nfs read rpc call
1188 * Ditto above
1189 */
1190 int
1191 nfs_readrpc(vp, uiop)
1192 struct vnode *vp;
1193 struct uio *uiop;
1194 {
1195 u_int32_t *tl;
1196 caddr_t cp;
1197 int32_t t1, t2;
1198 caddr_t bpos, dpos, cp2;
1199 struct mbuf *mreq, *mrep, *md, *mb;
1200 struct nfsmount *nmp;
1201 int error = 0, len, retlen, tsiz, eof, attrflag;
1202 const int v3 = NFS_ISV3(vp);
1203 struct nfsnode *np = VTONFS(vp);
1204
1205 #ifndef nolint
1206 eof = 0;
1207 #endif
1208 nmp = VFSTONFS(vp->v_mount);
1209 tsiz = uiop->uio_resid;
1210 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1211 return (EFBIG);
1212 while (tsiz > 0) {
1213 nfsstats.rpccnt[NFSPROC_READ]++;
1214 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1215 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1216 nfsm_fhtom(np, v3);
1217 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1218 if (v3) {
1219 txdr_hyper(uiop->uio_offset, tl);
1220 *(tl + 2) = txdr_unsigned(len);
1221 } else {
1222 *tl++ = txdr_unsigned(uiop->uio_offset);
1223 *tl++ = txdr_unsigned(len);
1224 *tl = 0;
1225 }
1226 nfsm_request(np, NFSPROC_READ, uiop->uio_procp, np->n_rcred);
1227 if (v3) {
1228 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1229 if (error) {
1230 m_freem(mrep);
1231 goto nfsmout;
1232 }
1233 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1234 eof = fxdr_unsigned(int, *(tl + 1));
1235 } else
1236 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1237 nfsm_strsiz(retlen, nmp->nm_rsize);
1238 nfsm_mtouio(uiop, retlen);
1239 m_freem(mrep);
1240 tsiz -= retlen;
1241 if (v3) {
1242 if (eof || retlen == 0)
1243 tsiz = 0;
1244 } else if (retlen < len)
1245 tsiz = 0;
1246 }
1247 nfsmout:
1248 return (error);
1249 }
1250
1251 struct nfs_writerpc_context {
1252 struct simplelock nwc_slock;
1253 volatile int nwc_mbufcount;
1254 };
1255
1256 /*
1257 * free mbuf used to refer protected pages while write rpc call.
1258 * called at splvm.
1259 */
1260 static void
1261 nfs_writerpc_extfree(struct mbuf *m, caddr_t buf, size_t size, void *arg)
1262 {
1263 struct nfs_writerpc_context *ctx = arg;
1264
1265 KASSERT(m != NULL);
1266 KASSERT(ctx != NULL);
1267 pool_cache_put(&mbpool_cache, m);
1268 simple_lock(&ctx->nwc_slock);
1269 if (--ctx->nwc_mbufcount == 0) {
1270 wakeup(ctx);
1271 }
1272 simple_unlock(&ctx->nwc_slock);
1273 }
1274
1275 /*
1276 * nfs write call
1277 */
1278 int
1279 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp)
1280 struct vnode *vp;
1281 struct uio *uiop;
1282 int *iomode;
1283 boolean_t pageprotected;
1284 boolean_t *stalewriteverfp;
1285 {
1286 u_int32_t *tl;
1287 caddr_t cp;
1288 int32_t t1, t2;
1289 caddr_t bpos, dpos, cp2;
1290 struct mbuf *mreq, *mrep, *md, *mb;
1291 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1292 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1293 const int v3 = NFS_ISV3(vp);
1294 int committed = NFSV3WRITE_FILESYNC;
1295 struct nfsnode *np = VTONFS(vp);
1296 struct nfs_writerpc_context ctx;
1297 int s;
1298 struct lwp *l = NULL;
1299 size_t origresid;
1300
1301 simple_lock_init(&ctx.nwc_slock);
1302 ctx.nwc_mbufcount = 1;
1303
1304 if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1305 panic("writerpc readonly vp %p", vp);
1306 }
1307
1308 #ifdef DIAGNOSTIC
1309 if (uiop->uio_iovcnt != 1)
1310 panic("nfs: writerpc iovcnt > 1");
1311 #endif
1312 tsiz = uiop->uio_resid;
1313 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1314 return (EFBIG);
1315 if (pageprotected) {
1316 l = curlwp;
1317 PHOLD(l);
1318 }
1319 retry:
1320 origresid = uiop->uio_resid;
1321 KASSERT(origresid == uiop->uio_iov->iov_len);
1322 while (tsiz > 0) {
1323 uint32_t datalen; /* data bytes need to be allocated in mbuf */
1324 uint32_t backup;
1325 boolean_t stalewriteverf = FALSE;
1326
1327 nfsstats.rpccnt[NFSPROC_WRITE]++;
1328 len = min(tsiz, nmp->nm_wsize);
1329 datalen = pageprotected ? 0 : nfsm_rndup(len);
1330 nfsm_reqhead(np, NFSPROC_WRITE,
1331 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1332 nfsm_fhtom(np, v3);
1333 if (v3) {
1334 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1335 txdr_hyper(uiop->uio_offset, tl);
1336 tl += 2;
1337 *tl++ = txdr_unsigned(len);
1338 *tl++ = txdr_unsigned(*iomode);
1339 *tl = txdr_unsigned(len);
1340 } else {
1341 u_int32_t x;
1342
1343 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1344 /* Set both "begin" and "current" to non-garbage. */
1345 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1346 *tl++ = x; /* "begin offset" */
1347 *tl++ = x; /* "current offset" */
1348 x = txdr_unsigned(len);
1349 *tl++ = x; /* total to this offset */
1350 *tl = x; /* size of this write */
1351
1352 }
1353 if (pageprotected) {
1354 /*
1355 * since we know pages can't be modified during i/o,
1356 * no need to copy them for us.
1357 */
1358 struct mbuf *m;
1359 struct iovec *iovp = uiop->uio_iov;
1360
1361 m = m_get(M_WAIT, MT_DATA);
1362 MCLAIM(m, &nfs_mowner);
1363 MEXTADD(m, iovp->iov_base, len, M_MBUF,
1364 nfs_writerpc_extfree, &ctx);
1365 m->m_flags |= M_EXT_ROMAP;
1366 m->m_len = len;
1367 mb->m_next = m;
1368 /*
1369 * no need to maintain mb and bpos here
1370 * because no one care them later.
1371 */
1372 #if 0
1373 mb = m;
1374 bpos = mtod(caddr_t, mb) + mb->m_len;
1375 #endif
1376 iovp->iov_base = (char *)iovp->iov_base + len;
1377 iovp->iov_len -= len;
1378 uiop->uio_offset += len;
1379 uiop->uio_resid -= len;
1380 s = splvm();
1381 simple_lock(&ctx.nwc_slock);
1382 ctx.nwc_mbufcount++;
1383 simple_unlock(&ctx.nwc_slock);
1384 splx(s);
1385 nfs_zeropad(mb, 0, nfsm_padlen(len));
1386 } else {
1387 nfsm_uiotom(uiop, len);
1388 }
1389 nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp, np->n_wcred);
1390 if (v3) {
1391 wccflag = NFSV3_WCCCHK;
1392 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC);
1393 if (!error) {
1394 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1395 + NFSX_V3WRITEVERF);
1396 rlen = fxdr_unsigned(int, *tl++);
1397 if (rlen == 0) {
1398 error = NFSERR_IO;
1399 m_freem(mrep);
1400 break;
1401 } else if (rlen < len) {
1402 backup = len - rlen;
1403 uiop->uio_iov->iov_base =
1404 (caddr_t)uiop->uio_iov->iov_base -
1405 backup;
1406 uiop->uio_iov->iov_len += backup;
1407 uiop->uio_offset -= backup;
1408 uiop->uio_resid += backup;
1409 len = rlen;
1410 }
1411 commit = fxdr_unsigned(int, *tl++);
1412
1413 /*
1414 * Return the lowest committment level
1415 * obtained by any of the RPCs.
1416 */
1417 if (committed == NFSV3WRITE_FILESYNC)
1418 committed = commit;
1419 else if (committed == NFSV3WRITE_DATASYNC &&
1420 commit == NFSV3WRITE_UNSTABLE)
1421 committed = commit;
1422 simple_lock(&nmp->nm_slock);
1423 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1424 memcpy(nmp->nm_writeverf, tl,
1425 NFSX_V3WRITEVERF);
1426 nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1427 } else if ((nmp->nm_iflag &
1428 NFSMNT_STALEWRITEVERF) ||
1429 memcmp(tl, nmp->nm_writeverf,
1430 NFSX_V3WRITEVERF)) {
1431 memcpy(nmp->nm_writeverf, tl,
1432 NFSX_V3WRITEVERF);
1433 /*
1434 * note NFSMNT_STALEWRITEVERF
1435 * if we're the first thread to
1436 * notice it.
1437 */
1438 if ((nmp->nm_iflag &
1439 NFSMNT_STALEWRITEVERF) == 0) {
1440 stalewriteverf = TRUE;
1441 nmp->nm_iflag |=
1442 NFSMNT_STALEWRITEVERF;
1443 }
1444 }
1445 simple_unlock(&nmp->nm_slock);
1446 }
1447 } else
1448 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1449 if (wccflag)
1450 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr->va_mtime;
1451 m_freem(mrep);
1452 if (error)
1453 break;
1454 tsiz -= len;
1455 if (stalewriteverf) {
1456 *stalewriteverfp = TRUE;
1457 stalewriteverf = FALSE;
1458 if (committed == NFSV3WRITE_UNSTABLE &&
1459 len != origresid) {
1460 /*
1461 * if our write requests weren't atomic but
1462 * unstable, datas in previous iterations
1463 * might have already been lost now.
1464 * then, we should resend them to nfsd.
1465 */
1466 backup = origresid - tsiz;
1467 uiop->uio_iov->iov_base =
1468 (caddr_t)uiop->uio_iov->iov_base - backup;
1469 uiop->uio_iov->iov_len += backup;
1470 uiop->uio_offset -= backup;
1471 uiop->uio_resid += backup;
1472 tsiz = origresid;
1473 goto retry;
1474 }
1475 }
1476 }
1477 if (pageprotected) {
1478 /*
1479 * wait until mbufs go away.
1480 * retransmitted mbufs can survive longer than rpc requests
1481 * themselves.
1482 */
1483 s = splvm();
1484 simple_lock(&ctx.nwc_slock);
1485 ctx.nwc_mbufcount--;
1486 while (ctx.nwc_mbufcount > 0) {
1487 ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock);
1488 }
1489 simple_unlock(&ctx.nwc_slock);
1490 splx(s);
1491 PRELE(l);
1492 }
1493 nfsmout:
1494 *iomode = committed;
1495 if (error)
1496 uiop->uio_resid = tsiz;
1497 return (error);
1498 }
1499
1500 /*
1501 * nfs mknod rpc
1502 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1503 * mode set to specify the file type and the size field for rdev.
1504 */
1505 int
1506 nfs_mknodrpc(dvp, vpp, cnp, vap)
1507 struct vnode *dvp;
1508 struct vnode **vpp;
1509 struct componentname *cnp;
1510 struct vattr *vap;
1511 {
1512 struct nfsv2_sattr *sp;
1513 u_int32_t *tl;
1514 caddr_t cp;
1515 int32_t t1, t2;
1516 struct vnode *newvp = (struct vnode *)0;
1517 struct nfsnode *dnp, *np;
1518 char *cp2;
1519 caddr_t bpos, dpos;
1520 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1521 struct mbuf *mreq, *mrep, *md, *mb;
1522 u_int32_t rdev;
1523 const int v3 = NFS_ISV3(dvp);
1524
1525 if (vap->va_type == VCHR || vap->va_type == VBLK)
1526 rdev = txdr_unsigned(vap->va_rdev);
1527 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1528 rdev = nfs_xdrneg1;
1529 else {
1530 VOP_ABORTOP(dvp, cnp);
1531 vput(dvp);
1532 return (EOPNOTSUPP);
1533 }
1534 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1535 dnp = VTONFS(dvp);
1536 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1537 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1538 nfsm_fhtom(dnp, v3);
1539 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1540 if (v3) {
1541 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1542 *tl++ = vtonfsv3_type(vap->va_type);
1543 nfsm_v3attrbuild(vap, FALSE);
1544 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1545 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1546 *tl++ = txdr_unsigned(major(vap->va_rdev));
1547 *tl = txdr_unsigned(minor(vap->va_rdev));
1548 }
1549 } else {
1550 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1551 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1552 sp->sa_uid = nfs_xdrneg1;
1553 sp->sa_gid = nfs_xdrneg1;
1554 sp->sa_size = rdev;
1555 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1556 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1557 }
1558 nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1559 if (!error) {
1560 nfsm_mtofh(dvp, newvp, v3, gotvp);
1561 if (!gotvp) {
1562 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1563 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1564 if (!error)
1565 newvp = NFSTOV(np);
1566 }
1567 }
1568 if (v3)
1569 nfsm_wcc_data(dvp, wccflag, 0);
1570 nfsm_reqdone;
1571 if (error) {
1572 if (newvp)
1573 vput(newvp);
1574 } else {
1575 if (cnp->cn_flags & MAKEENTRY)
1576 nfs_cache_enter(dvp, newvp, cnp);
1577 *vpp = newvp;
1578 }
1579 PNBUF_PUT(cnp->cn_pnbuf);
1580 VTONFS(dvp)->n_flag |= NMODIFIED;
1581 if (!wccflag)
1582 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1583 vput(dvp);
1584 return (error);
1585 }
1586
1587 /*
1588 * nfs mknod vop
1589 * just call nfs_mknodrpc() to do the work.
1590 */
1591 /* ARGSUSED */
1592 int
1593 nfs_mknod(v)
1594 void *v;
1595 {
1596 struct vop_mknod_args /* {
1597 struct vnode *a_dvp;
1598 struct vnode **a_vpp;
1599 struct componentname *a_cnp;
1600 struct vattr *a_vap;
1601 } */ *ap = v;
1602 int error;
1603
1604 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1605 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1606 return (error);
1607 }
1608
1609 static u_long create_verf;
1610 /*
1611 * nfs file create call
1612 */
1613 int
1614 nfs_create(v)
1615 void *v;
1616 {
1617 struct vop_create_args /* {
1618 struct vnode *a_dvp;
1619 struct vnode **a_vpp;
1620 struct componentname *a_cnp;
1621 struct vattr *a_vap;
1622 } */ *ap = v;
1623 struct vnode *dvp = ap->a_dvp;
1624 struct vattr *vap = ap->a_vap;
1625 struct componentname *cnp = ap->a_cnp;
1626 struct nfsv2_sattr *sp;
1627 u_int32_t *tl;
1628 caddr_t cp;
1629 int32_t t1, t2;
1630 struct nfsnode *dnp, *np = (struct nfsnode *)0;
1631 struct vnode *newvp = (struct vnode *)0;
1632 caddr_t bpos, dpos, cp2;
1633 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1634 struct mbuf *mreq, *mrep, *md, *mb;
1635 const int v3 = NFS_ISV3(dvp);
1636
1637 /*
1638 * Oops, not for me..
1639 */
1640 if (vap->va_type == VSOCK)
1641 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1642
1643 KASSERT(vap->va_type == VREG);
1644
1645 #ifdef VA_EXCLUSIVE
1646 if (vap->va_vaflags & VA_EXCLUSIVE)
1647 fmode |= O_EXCL;
1648 #endif
1649 again:
1650 error = 0;
1651 nfsstats.rpccnt[NFSPROC_CREATE]++;
1652 dnp = VTONFS(dvp);
1653 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1654 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1655 nfsm_fhtom(dnp, v3);
1656 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1657 if (v3) {
1658 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1659 if (fmode & O_EXCL) {
1660 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1661 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1662 #ifdef INET
1663 if (TAILQ_FIRST(&in_ifaddr))
1664 *tl++ = TAILQ_FIRST(&in_ifaddr)->ia_addr.sin_addr.s_addr;
1665 else
1666 *tl++ = create_verf;
1667 #else
1668 *tl++ = create_verf;
1669 #endif
1670 *tl = ++create_verf;
1671 } else {
1672 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1673 nfsm_v3attrbuild(vap, FALSE);
1674 }
1675 } else {
1676 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1677 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1678 sp->sa_uid = nfs_xdrneg1;
1679 sp->sa_gid = nfs_xdrneg1;
1680 sp->sa_size = 0;
1681 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1682 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1683 }
1684 nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1685 if (!error) {
1686 nfsm_mtofh(dvp, newvp, v3, gotvp);
1687 if (!gotvp) {
1688 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1689 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1690 if (!error)
1691 newvp = NFSTOV(np);
1692 }
1693 }
1694 if (v3)
1695 nfsm_wcc_data(dvp, wccflag, 0);
1696 nfsm_reqdone;
1697 if (error) {
1698 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1699 fmode &= ~O_EXCL;
1700 goto again;
1701 }
1702 if (newvp)
1703 vput(newvp);
1704 } else if (v3 && (fmode & O_EXCL)) {
1705 struct timeval tm = time;
1706
1707 /*
1708 * make sure that we'll update timestamps as
1709 * most server implementations use them to store
1710 * the create verifier.
1711 *
1712 * XXX it's better to use TOSERVER always.
1713 */
1714
1715 if (vap->va_atime.tv_sec == VNOVAL) {
1716 vap->va_atime.tv_sec = tm.tv_sec;
1717 vap->va_atime.tv_nsec = tm.tv_usec * 1000;
1718 }
1719 if (vap->va_mtime.tv_sec == VNOVAL) {
1720 vap->va_mtime.tv_sec = tm.tv_sec;
1721 vap->va_mtime.tv_nsec = tm.tv_usec * 1000;
1722 }
1723
1724 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1725 }
1726 if (!error) {
1727 if (cnp->cn_flags & MAKEENTRY)
1728 nfs_cache_enter(dvp, newvp, cnp);
1729 *ap->a_vpp = newvp;
1730 }
1731 PNBUF_PUT(cnp->cn_pnbuf);
1732 VTONFS(dvp)->n_flag |= NMODIFIED;
1733 if (!wccflag)
1734 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1735 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1736 vput(dvp);
1737 return (error);
1738 }
1739
1740 /*
1741 * nfs file remove call
1742 * To try and make nfs semantics closer to ufs semantics, a file that has
1743 * other processes using the vnode is renamed instead of removed and then
1744 * removed later on the last close.
1745 * - If v_usecount > 1
1746 * If a rename is not already in the works
1747 * call nfs_sillyrename() to set it up
1748 * else
1749 * do the remove rpc
1750 */
1751 int
1752 nfs_remove(v)
1753 void *v;
1754 {
1755 struct vop_remove_args /* {
1756 struct vnodeop_desc *a_desc;
1757 struct vnode * a_dvp;
1758 struct vnode * a_vp;
1759 struct componentname * a_cnp;
1760 } */ *ap = v;
1761 struct vnode *vp = ap->a_vp;
1762 struct vnode *dvp = ap->a_dvp;
1763 struct componentname *cnp = ap->a_cnp;
1764 struct nfsnode *np = VTONFS(vp);
1765 int error = 0;
1766 struct vattr vattr;
1767
1768 #ifndef DIAGNOSTIC
1769 if ((cnp->cn_flags & HASBUF) == 0)
1770 panic("nfs_remove: no name");
1771 if (vp->v_usecount < 1)
1772 panic("nfs_remove: bad v_usecount");
1773 #endif
1774 if (vp->v_type == VDIR)
1775 error = EPERM;
1776 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1777 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1778 vattr.va_nlink > 1)) {
1779 /*
1780 * Purge the name cache so that the chance of a lookup for
1781 * the name succeeding while the remove is in progress is
1782 * minimized. Without node locking it can still happen, such
1783 * that an I/O op returns ESTALE, but since you get this if
1784 * another host removes the file..
1785 */
1786 cache_purge(vp);
1787 /*
1788 * throw away biocache buffers, mainly to avoid
1789 * unnecessary delayed writes later.
1790 */
1791 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1792 /* Do the rpc */
1793 if (error != EINTR)
1794 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1795 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1796 } else if (!np->n_sillyrename)
1797 error = nfs_sillyrename(dvp, vp, cnp);
1798 PNBUF_PUT(cnp->cn_pnbuf);
1799 if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1800 vattr.va_nlink == 1) {
1801 np->n_flag |= NREMOVED;
1802 }
1803 NFS_INVALIDATE_ATTRCACHE(np);
1804 VN_KNOTE(vp, NOTE_DELETE);
1805 VN_KNOTE(dvp, NOTE_WRITE);
1806 if (dvp == vp)
1807 vrele(vp);
1808 else
1809 vput(vp);
1810 vput(dvp);
1811 return (error);
1812 }
1813
1814 /*
1815 * nfs file remove rpc called from nfs_inactive
1816 */
1817 int
1818 nfs_removeit(sp)
1819 struct sillyrename *sp;
1820 {
1821
1822 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1823 (struct proc *)0));
1824 }
1825
1826 /*
1827 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1828 */
1829 int
1830 nfs_removerpc(dvp, name, namelen, cred, proc)
1831 struct vnode *dvp;
1832 const char *name;
1833 int namelen;
1834 struct ucred *cred;
1835 struct proc *proc;
1836 {
1837 u_int32_t *tl;
1838 caddr_t cp;
1839 int32_t t1, t2;
1840 caddr_t bpos, dpos, cp2;
1841 int error = 0, wccflag = NFSV3_WCCRATTR;
1842 struct mbuf *mreq, *mrep, *md, *mb;
1843 const int v3 = NFS_ISV3(dvp);
1844 int rexmit;
1845 struct nfsnode *dnp = VTONFS(dvp);
1846
1847 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1848 nfsm_reqhead(dnp, NFSPROC_REMOVE,
1849 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1850 nfsm_fhtom(dnp, v3);
1851 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1852 nfsm_request1(dnp, NFSPROC_REMOVE, proc, cred, &rexmit);
1853 if (v3)
1854 nfsm_wcc_data(dvp, wccflag, 0);
1855 nfsm_reqdone;
1856 VTONFS(dvp)->n_flag |= NMODIFIED;
1857 if (!wccflag)
1858 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
1859 /*
1860 * Kludge City: If the first reply to the remove rpc is lost..
1861 * the reply to the retransmitted request will be ENOENT
1862 * since the file was in fact removed
1863 * Therefore, we cheat and return success.
1864 */
1865 if (rexmit && error == ENOENT)
1866 error = 0;
1867 return (error);
1868 }
1869
1870 /*
1871 * nfs file rename call
1872 */
1873 int
1874 nfs_rename(v)
1875 void *v;
1876 {
1877 struct vop_rename_args /* {
1878 struct vnode *a_fdvp;
1879 struct vnode *a_fvp;
1880 struct componentname *a_fcnp;
1881 struct vnode *a_tdvp;
1882 struct vnode *a_tvp;
1883 struct componentname *a_tcnp;
1884 } */ *ap = v;
1885 struct vnode *fvp = ap->a_fvp;
1886 struct vnode *tvp = ap->a_tvp;
1887 struct vnode *fdvp = ap->a_fdvp;
1888 struct vnode *tdvp = ap->a_tdvp;
1889 struct componentname *tcnp = ap->a_tcnp;
1890 struct componentname *fcnp = ap->a_fcnp;
1891 int error;
1892
1893 #ifndef DIAGNOSTIC
1894 if ((tcnp->cn_flags & HASBUF) == 0 ||
1895 (fcnp->cn_flags & HASBUF) == 0)
1896 panic("nfs_rename: no name");
1897 #endif
1898 /* Check for cross-device rename */
1899 if ((fvp->v_mount != tdvp->v_mount) ||
1900 (tvp && (fvp->v_mount != tvp->v_mount))) {
1901 error = EXDEV;
1902 goto out;
1903 }
1904
1905 /*
1906 * If the tvp exists and is in use, sillyrename it before doing the
1907 * rename of the new file over it.
1908 */
1909 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1910 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1911 VN_KNOTE(tvp, NOTE_DELETE);
1912 vput(tvp);
1913 tvp = NULL;
1914 }
1915
1916 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1917 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1918 tcnp->cn_proc);
1919
1920 VN_KNOTE(fdvp, NOTE_WRITE);
1921 VN_KNOTE(tdvp, NOTE_WRITE);
1922 if (fvp->v_type == VDIR) {
1923 if (tvp != NULL && tvp->v_type == VDIR)
1924 cache_purge(tdvp);
1925 cache_purge(fdvp);
1926 }
1927 out:
1928 if (tdvp == tvp)
1929 vrele(tdvp);
1930 else
1931 vput(tdvp);
1932 if (tvp)
1933 vput(tvp);
1934 vrele(fdvp);
1935 vrele(fvp);
1936 return (error);
1937 }
1938
1939 /*
1940 * nfs file rename rpc called from nfs_remove() above
1941 */
1942 int
1943 nfs_renameit(sdvp, scnp, sp)
1944 struct vnode *sdvp;
1945 struct componentname *scnp;
1946 struct sillyrename *sp;
1947 {
1948 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1949 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1950 }
1951
1952 /*
1953 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1954 */
1955 int
1956 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1957 struct vnode *fdvp;
1958 const char *fnameptr;
1959 int fnamelen;
1960 struct vnode *tdvp;
1961 const char *tnameptr;
1962 int tnamelen;
1963 struct ucred *cred;
1964 struct proc *proc;
1965 {
1966 u_int32_t *tl;
1967 caddr_t cp;
1968 int32_t t1, t2;
1969 caddr_t bpos, dpos, cp2;
1970 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1971 struct mbuf *mreq, *mrep, *md, *mb;
1972 const int v3 = NFS_ISV3(fdvp);
1973 int rexmit;
1974 struct nfsnode *fdnp = VTONFS(fdvp);
1975
1976 nfsstats.rpccnt[NFSPROC_RENAME]++;
1977 nfsm_reqhead(fdnp, NFSPROC_RENAME,
1978 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1979 nfsm_rndup(tnamelen));
1980 nfsm_fhtom(fdnp, v3);
1981 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1982 nfsm_fhtom(VTONFS(tdvp), v3);
1983 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1984 nfsm_request1(fdnp, NFSPROC_RENAME, proc, cred, &rexmit);
1985 if (v3) {
1986 nfsm_wcc_data(fdvp, fwccflag, 0);
1987 nfsm_wcc_data(tdvp, twccflag, 0);
1988 }
1989 nfsm_reqdone;
1990 VTONFS(fdvp)->n_flag |= NMODIFIED;
1991 VTONFS(tdvp)->n_flag |= NMODIFIED;
1992 if (!fwccflag)
1993 NFS_INVALIDATE_ATTRCACHE(VTONFS(fdvp));
1994 if (!twccflag)
1995 NFS_INVALIDATE_ATTRCACHE(VTONFS(tdvp));
1996 /*
1997 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1998 */
1999 if (rexmit && error == ENOENT)
2000 error = 0;
2001 return (error);
2002 }
2003
2004 /*
2005 * nfs hard link create call
2006 */
2007 int
2008 nfs_link(v)
2009 void *v;
2010 {
2011 struct vop_link_args /* {
2012 struct vnode *a_dvp;
2013 struct vnode *a_vp;
2014 struct componentname *a_cnp;
2015 } */ *ap = v;
2016 struct vnode *vp = ap->a_vp;
2017 struct vnode *dvp = ap->a_dvp;
2018 struct componentname *cnp = ap->a_cnp;
2019 u_int32_t *tl;
2020 caddr_t cp;
2021 int32_t t1, t2;
2022 caddr_t bpos, dpos, cp2;
2023 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
2024 struct mbuf *mreq, *mrep, *md, *mb;
2025 /* XXX Should be const and initialised? */
2026 int v3;
2027 int rexmit;
2028 struct nfsnode *np;
2029
2030 if (dvp->v_mount != vp->v_mount) {
2031 VOP_ABORTOP(dvp, cnp);
2032 vput(dvp);
2033 return (EXDEV);
2034 }
2035 if (dvp != vp) {
2036 error = vn_lock(vp, LK_EXCLUSIVE);
2037 if (error != 0) {
2038 VOP_ABORTOP(dvp, cnp);
2039 vput(dvp);
2040 return error;
2041 }
2042 }
2043
2044 /*
2045 * Push all writes to the server, so that the attribute cache
2046 * doesn't get "out of sync" with the server.
2047 * XXX There should be a better way!
2048 */
2049 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc);
2050
2051 v3 = NFS_ISV3(vp);
2052 nfsstats.rpccnt[NFSPROC_LINK]++;
2053 np = VTONFS(vp);
2054 nfsm_reqhead(np, NFSPROC_LINK,
2055 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2056 nfsm_fhtom(np, v3);
2057 nfsm_fhtom(VTONFS(dvp), v3);
2058 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2059 nfsm_request1(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred, &rexmit);
2060 if (v3) {
2061 nfsm_postop_attr(vp, attrflag, 0);
2062 nfsm_wcc_data(dvp, wccflag, 0);
2063 }
2064 nfsm_reqdone;
2065 PNBUF_PUT(cnp->cn_pnbuf);
2066 VTONFS(dvp)->n_flag |= NMODIFIED;
2067 if (!attrflag)
2068 NFS_INVALIDATE_ATTRCACHE(VTONFS(vp));
2069 if (!wccflag)
2070 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2071 if (dvp != vp)
2072 VOP_UNLOCK(vp, 0);
2073 VN_KNOTE(vp, NOTE_LINK);
2074 VN_KNOTE(dvp, NOTE_WRITE);
2075 vput(dvp);
2076 /*
2077 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2078 */
2079 if (rexmit && error == EEXIST)
2080 error = 0;
2081 return (error);
2082 }
2083
2084 /*
2085 * nfs symbolic link create call
2086 */
2087 int
2088 nfs_symlink(v)
2089 void *v;
2090 {
2091 struct vop_symlink_args /* {
2092 struct vnode *a_dvp;
2093 struct vnode **a_vpp;
2094 struct componentname *a_cnp;
2095 struct vattr *a_vap;
2096 char *a_target;
2097 } */ *ap = v;
2098 struct vnode *dvp = ap->a_dvp;
2099 struct vattr *vap = ap->a_vap;
2100 struct componentname *cnp = ap->a_cnp;
2101 struct nfsv2_sattr *sp;
2102 u_int32_t *tl;
2103 caddr_t cp;
2104 int32_t t1, t2;
2105 caddr_t bpos, dpos, cp2;
2106 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2107 struct mbuf *mreq, *mrep, *md, *mb;
2108 struct vnode *newvp = (struct vnode *)0;
2109 const int v3 = NFS_ISV3(dvp);
2110 int rexmit;
2111 struct nfsnode *dnp = VTONFS(dvp);
2112
2113 *ap->a_vpp = NULL;
2114 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2115 slen = strlen(ap->a_target);
2116 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2117 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2118 nfsm_fhtom(dnp, v3);
2119 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2120 if (v3)
2121 nfsm_v3attrbuild(vap, FALSE);
2122 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2123 if (!v3) {
2124 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2125 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2126 sp->sa_uid = nfs_xdrneg1;
2127 sp->sa_gid = nfs_xdrneg1;
2128 sp->sa_size = nfs_xdrneg1;
2129 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2130 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2131 }
2132 nfsm_request1(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred,
2133 &rexmit);
2134 if (v3) {
2135 if (!error)
2136 nfsm_mtofh(dvp, newvp, v3, gotvp);
2137 nfsm_wcc_data(dvp, wccflag, 0);
2138 }
2139 nfsm_reqdone;
2140 /*
2141 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2142 */
2143 if (rexmit && error == EEXIST)
2144 error = 0;
2145 if (error == 0 && newvp == NULL) {
2146 struct nfsnode *np = NULL;
2147
2148 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2149 cnp->cn_cred, cnp->cn_proc, &np);
2150 if (error == 0)
2151 newvp = NFSTOV(np);
2152 }
2153 if (error) {
2154 if (newvp != NULL)
2155 vput(newvp);
2156 } else {
2157 *ap->a_vpp = newvp;
2158 }
2159 PNBUF_PUT(cnp->cn_pnbuf);
2160 VTONFS(dvp)->n_flag |= NMODIFIED;
2161 if (!wccflag)
2162 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2163 VN_KNOTE(dvp, NOTE_WRITE);
2164 vput(dvp);
2165 return (error);
2166 }
2167
2168 /*
2169 * nfs make dir call
2170 */
2171 int
2172 nfs_mkdir(v)
2173 void *v;
2174 {
2175 struct vop_mkdir_args /* {
2176 struct vnode *a_dvp;
2177 struct vnode **a_vpp;
2178 struct componentname *a_cnp;
2179 struct vattr *a_vap;
2180 } */ *ap = v;
2181 struct vnode *dvp = ap->a_dvp;
2182 struct vattr *vap = ap->a_vap;
2183 struct componentname *cnp = ap->a_cnp;
2184 struct nfsv2_sattr *sp;
2185 u_int32_t *tl;
2186 caddr_t cp;
2187 int32_t t1, t2;
2188 int len;
2189 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2190 struct vnode *newvp = (struct vnode *)0;
2191 caddr_t bpos, dpos, cp2;
2192 int error = 0, wccflag = NFSV3_WCCRATTR;
2193 int gotvp = 0;
2194 int rexmit;
2195 struct mbuf *mreq, *mrep, *md, *mb;
2196 const int v3 = NFS_ISV3(dvp);
2197
2198 len = cnp->cn_namelen;
2199 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2200 nfsm_reqhead(dnp, NFSPROC_MKDIR,
2201 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2202 nfsm_fhtom(dnp, v3);
2203 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2204 if (v3) {
2205 nfsm_v3attrbuild(vap, FALSE);
2206 } else {
2207 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2208 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2209 sp->sa_uid = nfs_xdrneg1;
2210 sp->sa_gid = nfs_xdrneg1;
2211 sp->sa_size = nfs_xdrneg1;
2212 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2213 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2214 }
2215 nfsm_request1(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred, &rexmit);
2216 if (!error)
2217 nfsm_mtofh(dvp, newvp, v3, gotvp);
2218 if (v3)
2219 nfsm_wcc_data(dvp, wccflag, 0);
2220 nfsm_reqdone;
2221 VTONFS(dvp)->n_flag |= NMODIFIED;
2222 if (!wccflag)
2223 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2224 /*
2225 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2226 * if we can succeed in looking up the directory.
2227 */
2228 if ((rexmit && error == EEXIST) || (!error && !gotvp)) {
2229 if (newvp) {
2230 vput(newvp);
2231 newvp = (struct vnode *)0;
2232 }
2233 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2234 cnp->cn_proc, &np);
2235 if (!error) {
2236 newvp = NFSTOV(np);
2237 if (newvp->v_type != VDIR)
2238 error = EEXIST;
2239 }
2240 }
2241 if (error) {
2242 if (newvp)
2243 vput(newvp);
2244 } else {
2245 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2246 if (cnp->cn_flags & MAKEENTRY)
2247 nfs_cache_enter(dvp, newvp, cnp);
2248 *ap->a_vpp = newvp;
2249 }
2250 PNBUF_PUT(cnp->cn_pnbuf);
2251 vput(dvp);
2252 return (error);
2253 }
2254
2255 /*
2256 * nfs remove directory call
2257 */
2258 int
2259 nfs_rmdir(v)
2260 void *v;
2261 {
2262 struct vop_rmdir_args /* {
2263 struct vnode *a_dvp;
2264 struct vnode *a_vp;
2265 struct componentname *a_cnp;
2266 } */ *ap = v;
2267 struct vnode *vp = ap->a_vp;
2268 struct vnode *dvp = ap->a_dvp;
2269 struct componentname *cnp = ap->a_cnp;
2270 u_int32_t *tl;
2271 caddr_t cp;
2272 int32_t t1, t2;
2273 caddr_t bpos, dpos, cp2;
2274 int error = 0, wccflag = NFSV3_WCCRATTR;
2275 int rexmit;
2276 struct mbuf *mreq, *mrep, *md, *mb;
2277 const int v3 = NFS_ISV3(dvp);
2278 struct nfsnode *dnp;
2279
2280 if (dvp == vp) {
2281 vrele(dvp);
2282 vput(dvp);
2283 PNBUF_PUT(cnp->cn_pnbuf);
2284 return (EINVAL);
2285 }
2286 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2287 dnp = VTONFS(dvp);
2288 nfsm_reqhead(dnp, NFSPROC_RMDIR,
2289 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2290 nfsm_fhtom(dnp, v3);
2291 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2292 nfsm_request1(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred, &rexmit);
2293 if (v3)
2294 nfsm_wcc_data(dvp, wccflag, 0);
2295 nfsm_reqdone;
2296 PNBUF_PUT(cnp->cn_pnbuf);
2297 VTONFS(dvp)->n_flag |= NMODIFIED;
2298 if (!wccflag)
2299 NFS_INVALIDATE_ATTRCACHE(VTONFS(dvp));
2300 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2301 VN_KNOTE(vp, NOTE_DELETE);
2302 cache_purge(vp);
2303 vput(vp);
2304 vput(dvp);
2305 /*
2306 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2307 */
2308 if (rexmit && error == ENOENT)
2309 error = 0;
2310 return (error);
2311 }
2312
2313 /*
2314 * nfs readdir call
2315 */
2316 int
2317 nfs_readdir(v)
2318 void *v;
2319 {
2320 struct vop_readdir_args /* {
2321 struct vnode *a_vp;
2322 struct uio *a_uio;
2323 struct ucred *a_cred;
2324 int *a_eofflag;
2325 off_t **a_cookies;
2326 int *a_ncookies;
2327 } */ *ap = v;
2328 struct vnode *vp = ap->a_vp;
2329 struct uio *uio = ap->a_uio;
2330 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2331 char *base = uio->uio_iov->iov_base;
2332 int tresid, error;
2333 size_t count, lost;
2334 struct dirent *dp;
2335 off_t *cookies = NULL;
2336 int ncookies = 0, nc;
2337
2338 if (vp->v_type != VDIR)
2339 return (EPERM);
2340
2341 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2342 count = uio->uio_resid - lost;
2343 if (count <= 0)
2344 return (EINVAL);
2345
2346 /*
2347 * Call nfs_bioread() to do the real work.
2348 */
2349 tresid = uio->uio_resid = count;
2350 error = nfs_bioread(vp, uio, 0, ap->a_cred,
2351 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2352
2353 if (!error && ap->a_cookies) {
2354 ncookies = count / 16;
2355 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2356 *ap->a_cookies = cookies;
2357 }
2358
2359 if (!error && uio->uio_resid == tresid) {
2360 uio->uio_resid += lost;
2361 nfsstats.direofcache_misses++;
2362 if (ap->a_cookies)
2363 *ap->a_ncookies = 0;
2364 *ap->a_eofflag = 1;
2365 return (0);
2366 }
2367
2368 if (!error && ap->a_cookies) {
2369 /*
2370 * Only the NFS server and emulations use cookies, and they
2371 * load the directory block into system space, so we can
2372 * just look at it directly.
2373 */
2374 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2375 panic("nfs_readdir: lost in space");
2376 for (nc = 0; ncookies-- &&
2377 base < (char *)uio->uio_iov->iov_base; nc++){
2378 dp = (struct dirent *) base;
2379 if (dp->d_reclen == 0)
2380 break;
2381 if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2382 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2383 else
2384 *(cookies++) = NFS_GETCOOKIE(dp);
2385 base += dp->d_reclen;
2386 }
2387 uio->uio_resid +=
2388 ((caddr_t)uio->uio_iov->iov_base - base);
2389 uio->uio_iov->iov_len +=
2390 ((caddr_t)uio->uio_iov->iov_base - base);
2391 uio->uio_iov->iov_base = base;
2392 *ap->a_ncookies = nc;
2393 }
2394
2395 uio->uio_resid += lost;
2396 *ap->a_eofflag = 0;
2397 return (error);
2398 }
2399
2400 /*
2401 * Readdir rpc call.
2402 * Called from below the buffer cache by nfs_doio().
2403 */
2404 int
2405 nfs_readdirrpc(vp, uiop, cred)
2406 struct vnode *vp;
2407 struct uio *uiop;
2408 struct ucred *cred;
2409 {
2410 int len, left;
2411 struct dirent *dp = NULL;
2412 u_int32_t *tl;
2413 caddr_t cp;
2414 int32_t t1, t2;
2415 caddr_t bpos, dpos, cp2;
2416 struct mbuf *mreq, *mrep, *md, *mb;
2417 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2418 struct nfsnode *dnp = VTONFS(vp);
2419 u_quad_t fileno;
2420 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2421 int attrflag, nrpcs = 0, reclen;
2422 const int v3 = NFS_ISV3(vp);
2423
2424 #ifdef DIAGNOSTIC
2425 /*
2426 * Should be called from buffer cache, so only amount of
2427 * NFS_DIRBLKSIZ will be requested.
2428 */
2429 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2430 panic("nfs readdirrpc bad uio");
2431 #endif
2432
2433 /*
2434 * Loop around doing readdir rpc's of size nm_readdirsize
2435 * truncated to a multiple of NFS_DIRFRAGSIZ.
2436 * The stopping criteria is EOF or buffer full.
2437 */
2438 while (more_dirs && bigenough) {
2439 /*
2440 * Heuristic: don't bother to do another RPC to further
2441 * fill up this block if there is not much room left. (< 50%
2442 * of the readdir RPC size). This wastes some buffer space
2443 * but can save up to 50% in RPC calls.
2444 */
2445 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2446 bigenough = 0;
2447 break;
2448 }
2449 nfsstats.rpccnt[NFSPROC_READDIR]++;
2450 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2451 NFSX_READDIR(v3));
2452 nfsm_fhtom(dnp, v3);
2453 if (v3) {
2454 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2455 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2456 txdr_swapcookie3(uiop->uio_offset, tl);
2457 } else {
2458 txdr_cookie3(uiop->uio_offset, tl);
2459 }
2460 tl += 2;
2461 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2462 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2463 } else {
2464 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2465 *tl++ = txdr_unsigned(uiop->uio_offset);
2466 }
2467 *tl = txdr_unsigned(nmp->nm_readdirsize);
2468 nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred);
2469 nrpcs++;
2470 if (v3) {
2471 nfsm_postop_attr(vp, attrflag, 0);
2472 if (!error) {
2473 nfsm_dissect(tl, u_int32_t *,
2474 2 * NFSX_UNSIGNED);
2475 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2476 dnp->n_cookieverf.nfsuquad[1] = *tl;
2477 } else {
2478 m_freem(mrep);
2479 goto nfsmout;
2480 }
2481 }
2482 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2483 more_dirs = fxdr_unsigned(int, *tl);
2484
2485 /* loop thru the dir entries, doctoring them to 4bsd form */
2486 while (more_dirs && bigenough) {
2487 if (v3) {
2488 nfsm_dissect(tl, u_int32_t *,
2489 3 * NFSX_UNSIGNED);
2490 fileno = fxdr_hyper(tl);
2491 len = fxdr_unsigned(int, *(tl + 2));
2492 } else {
2493 nfsm_dissect(tl, u_int32_t *,
2494 2 * NFSX_UNSIGNED);
2495 fileno = fxdr_unsigned(u_quad_t, *tl++);
2496 len = fxdr_unsigned(int, *tl);
2497 }
2498 if (len <= 0 || len > NFS_MAXNAMLEN) {
2499 error = EBADRPC;
2500 m_freem(mrep);
2501 goto nfsmout;
2502 }
2503 tlen = nfsm_rndup(len);
2504 if (tlen == len)
2505 tlen += 4; /* To ensure null termination */
2506 tlen += sizeof (off_t) + sizeof (int);
2507 reclen = ALIGN(tlen + DIRHDSIZ);
2508 tlen = reclen - DIRHDSIZ;
2509 left = NFS_DIRFRAGSIZ - blksiz;
2510 if (reclen > left) {
2511 dp->d_reclen += left;
2512 uiop->uio_iov->iov_base =
2513 (caddr_t)uiop->uio_iov->iov_base + left;
2514 uiop->uio_iov->iov_len -= left;
2515 uiop->uio_resid -= left;
2516 blksiz = 0;
2517 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2518 }
2519 if (reclen > uiop->uio_resid)
2520 bigenough = 0;
2521 if (bigenough) {
2522 dp = (struct dirent *)uiop->uio_iov->iov_base;
2523 dp->d_fileno = (int)fileno;
2524 dp->d_namlen = len;
2525 dp->d_reclen = reclen;
2526 dp->d_type = DT_UNKNOWN;
2527 blksiz += dp->d_reclen;
2528 if (blksiz == NFS_DIRFRAGSIZ)
2529 blksiz = 0;
2530 uiop->uio_resid -= DIRHDSIZ;
2531 uiop->uio_iov->iov_base =
2532 (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ;
2533 uiop->uio_iov->iov_len -= DIRHDSIZ;
2534 nfsm_mtouio(uiop, len);
2535 cp = uiop->uio_iov->iov_base;
2536 tlen -= len;
2537 *cp = '\0'; /* null terminate */
2538 uiop->uio_iov->iov_base =
2539 (caddr_t)uiop->uio_iov->iov_base + tlen;
2540 uiop->uio_iov->iov_len -= tlen;
2541 uiop->uio_resid -= tlen;
2542 } else
2543 nfsm_adv(nfsm_rndup(len));
2544 if (v3) {
2545 nfsm_dissect(tl, u_int32_t *,
2546 3 * NFSX_UNSIGNED);
2547 } else {
2548 nfsm_dissect(tl, u_int32_t *,
2549 2 * NFSX_UNSIGNED);
2550 }
2551 if (bigenough) {
2552 if (v3) {
2553 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2554 uiop->uio_offset =
2555 fxdr_swapcookie3(tl);
2556 else
2557 uiop->uio_offset =
2558 fxdr_cookie3(tl);
2559 }
2560 else {
2561 uiop->uio_offset =
2562 fxdr_unsigned(off_t, *tl);
2563 }
2564 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2565 }
2566 if (v3)
2567 tl += 2;
2568 else
2569 tl++;
2570 more_dirs = fxdr_unsigned(int, *tl);
2571 }
2572 /*
2573 * If at end of rpc data, get the eof boolean
2574 */
2575 if (!more_dirs) {
2576 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2577 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2578
2579 /*
2580 * kludge: if we got no entries, treat it as EOF.
2581 * some server sometimes send a reply without any
2582 * entries or EOF.
2583 * although it might mean the server has very long name,
2584 * we can't handle such entries anyway.
2585 */
2586
2587 if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2588 more_dirs = 0;
2589 }
2590 m_freem(mrep);
2591 }
2592 /*
2593 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2594 * by increasing d_reclen for the last record.
2595 */
2596 if (blksiz > 0) {
2597 left = NFS_DIRFRAGSIZ - blksiz;
2598 dp->d_reclen += left;
2599 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2600 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2601 left;
2602 uiop->uio_iov->iov_len -= left;
2603 uiop->uio_resid -= left;
2604 }
2605
2606 /*
2607 * We are now either at the end of the directory or have filled the
2608 * block.
2609 */
2610 if (bigenough)
2611 dnp->n_direofoffset = uiop->uio_offset;
2612 nfsmout:
2613 return (error);
2614 }
2615
2616 /*
2617 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2618 */
2619 int
2620 nfs_readdirplusrpc(vp, uiop, cred)
2621 struct vnode *vp;
2622 struct uio *uiop;
2623 struct ucred *cred;
2624 {
2625 int len, left;
2626 struct dirent *dp = NULL;
2627 u_int32_t *tl;
2628 caddr_t cp;
2629 int32_t t1, t2;
2630 struct vnode *newvp;
2631 caddr_t bpos, dpos, cp2;
2632 struct mbuf *mreq, *mrep, *md, *mb;
2633 struct nameidata nami, *ndp = &nami;
2634 struct componentname *cnp = &ndp->ni_cnd;
2635 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2636 struct nfsnode *dnp = VTONFS(vp), *np;
2637 nfsfh_t *fhp;
2638 u_quad_t fileno;
2639 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2640 int attrflag, fhsize, nrpcs = 0, reclen;
2641 struct nfs_fattr fattr, *fp;
2642
2643 #ifdef DIAGNOSTIC
2644 if (uiop->uio_iovcnt != 1 || uiop->uio_resid != NFS_DIRBLKSIZ)
2645 panic("nfs readdirplusrpc bad uio");
2646 #endif
2647 ndp->ni_dvp = vp;
2648 newvp = NULLVP;
2649
2650 /*
2651 * Loop around doing readdir rpc's of size nm_readdirsize
2652 * truncated to a multiple of NFS_DIRFRAGSIZ.
2653 * The stopping criteria is EOF or buffer full.
2654 */
2655 while (more_dirs && bigenough) {
2656 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2657 bigenough = 0;
2658 break;
2659 }
2660 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2661 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2662 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2663 nfsm_fhtom(dnp, 1);
2664 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2665 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2666 txdr_swapcookie3(uiop->uio_offset, tl);
2667 } else {
2668 txdr_cookie3(uiop->uio_offset, tl);
2669 }
2670 tl += 2;
2671 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2672 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2673 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2674 *tl = txdr_unsigned(nmp->nm_rsize);
2675 nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2676 nfsm_postop_attr(vp, attrflag, 0);
2677 if (error) {
2678 m_freem(mrep);
2679 goto nfsmout;
2680 }
2681 nrpcs++;
2682 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2683 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2684 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2685 more_dirs = fxdr_unsigned(int, *tl);
2686
2687 /* loop thru the dir entries, doctoring them to 4bsd form */
2688 while (more_dirs && bigenough) {
2689 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2690 fileno = fxdr_hyper(tl);
2691 len = fxdr_unsigned(int, *(tl + 2));
2692 if (len <= 0 || len > NFS_MAXNAMLEN) {
2693 error = EBADRPC;
2694 m_freem(mrep);
2695 goto nfsmout;
2696 }
2697 tlen = nfsm_rndup(len);
2698 if (tlen == len)
2699 tlen += 4; /* To ensure null termination*/
2700 tlen += sizeof (off_t) + sizeof (int);
2701 reclen = ALIGN(tlen + DIRHDSIZ);
2702 tlen = reclen - DIRHDSIZ;
2703 left = NFS_DIRFRAGSIZ - blksiz;
2704 if (reclen > left) {
2705 /*
2706 * DIRFRAGSIZ is aligned, no need to align
2707 * again here.
2708 */
2709 dp->d_reclen += left;
2710 uiop->uio_iov->iov_base =
2711 (caddr_t)uiop->uio_iov->iov_base + left;
2712 uiop->uio_iov->iov_len -= left;
2713 uiop->uio_resid -= left;
2714 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2715 blksiz = 0;
2716 }
2717 if (reclen > uiop->uio_resid)
2718 bigenough = 0;
2719 if (bigenough) {
2720 dp = (struct dirent *)uiop->uio_iov->iov_base;
2721 dp->d_fileno = (int)fileno;
2722 dp->d_namlen = len;
2723 dp->d_reclen = reclen;
2724 dp->d_type = DT_UNKNOWN;
2725 blksiz += dp->d_reclen;
2726 if (blksiz == NFS_DIRFRAGSIZ)
2727 blksiz = 0;
2728 uiop->uio_resid -= DIRHDSIZ;
2729 uiop->uio_iov->iov_base =
2730 (caddr_t)uiop->uio_iov->iov_base +
2731 DIRHDSIZ;
2732 uiop->uio_iov->iov_len -= DIRHDSIZ;
2733 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2734 cnp->cn_namelen = len;
2735 nfsm_mtouio(uiop, len);
2736 cp = uiop->uio_iov->iov_base;
2737 tlen -= len;
2738 *cp = '\0';
2739 uiop->uio_iov->iov_base =
2740 (caddr_t)uiop->uio_iov->iov_base + tlen;
2741 uiop->uio_iov->iov_len -= tlen;
2742 uiop->uio_resid -= tlen;
2743 } else
2744 nfsm_adv(nfsm_rndup(len));
2745 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2746 if (bigenough) {
2747 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2748 uiop->uio_offset =
2749 fxdr_swapcookie3(tl);
2750 else
2751 uiop->uio_offset =
2752 fxdr_cookie3(tl);
2753 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2754 }
2755 tl += 2;
2756
2757 /*
2758 * Since the attributes are before the file handle
2759 * (sigh), we must skip over the attributes and then
2760 * come back and get them.
2761 */
2762 attrflag = fxdr_unsigned(int, *tl);
2763 if (attrflag) {
2764 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2765 memcpy(&fattr, fp, NFSX_V3FATTR);
2766 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2767 doit = fxdr_unsigned(int, *tl);
2768 if (doit) {
2769 nfsm_getfh(fhp, fhsize, 1);
2770 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2771 VREF(vp);
2772 newvp = vp;
2773 np = dnp;
2774 } else {
2775 error = nfs_nget1(vp->v_mount, fhp,
2776 fhsize, &np, LK_NOWAIT);
2777 if (!error)
2778 newvp = NFSTOV(np);
2779 }
2780 if (!error) {
2781 const char *cp;
2782
2783 nfs_loadattrcache(&newvp, &fattr, 0, 0);
2784 dp->d_type =
2785 IFTODT(VTTOIF(np->n_vattr->va_type));
2786 ndp->ni_vp = newvp;
2787 cp = cnp->cn_nameptr + cnp->cn_namelen;
2788 cnp->cn_hash =
2789 namei_hash(cnp->cn_nameptr, &cp);
2790 if (cnp->cn_namelen <= NCHNAMLEN) {
2791 cache_purge1(ndp->ni_dvp, cnp, 0);
2792 nfs_cache_enter(ndp->ni_dvp, ndp->ni_vp,
2793 cnp);
2794 }
2795 }
2796 }
2797 } else {
2798 /* Just skip over the file handle */
2799 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2800 i = fxdr_unsigned(int, *tl);
2801 nfsm_adv(nfsm_rndup(i));
2802 }
2803 if (newvp != NULLVP) {
2804 if (newvp == vp)
2805 vrele(newvp);
2806 else
2807 vput(newvp);
2808 newvp = NULLVP;
2809 }
2810 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2811 more_dirs = fxdr_unsigned(int, *tl);
2812 }
2813 /*
2814 * If at end of rpc data, get the eof boolean
2815 */
2816 if (!more_dirs) {
2817 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2818 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2819
2820 /*
2821 * kludge: see a comment in nfs_readdirrpc.
2822 */
2823
2824 if (uiop->uio_resid >= NFS_DIRBLKSIZ)
2825 more_dirs = 0;
2826 }
2827 m_freem(mrep);
2828 }
2829 /*
2830 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2831 * by increasing d_reclen for the last record.
2832 */
2833 if (blksiz > 0) {
2834 left = NFS_DIRFRAGSIZ - blksiz;
2835 dp->d_reclen += left;
2836 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2837 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2838 left;
2839 uiop->uio_iov->iov_len -= left;
2840 uiop->uio_resid -= left;
2841 }
2842
2843 /*
2844 * We are now either at the end of the directory or have filled the
2845 * block.
2846 */
2847 if (bigenough)
2848 dnp->n_direofoffset = uiop->uio_offset;
2849 nfsmout:
2850 if (newvp != NULLVP) {
2851 if(newvp == vp)
2852 vrele(newvp);
2853 else
2854 vput(newvp);
2855 }
2856 return (error);
2857 }
2858 static char hextoasc[] = "0123456789abcdef";
2859
2860 /*
2861 * Silly rename. To make the NFS filesystem that is stateless look a little
2862 * more like the "ufs" a remove of an active vnode is translated to a rename
2863 * to a funny looking filename that is removed by nfs_inactive on the
2864 * nfsnode. There is the potential for another process on a different client
2865 * to create the same funny name between the nfs_lookitup() fails and the
2866 * nfs_rename() completes, but...
2867 */
2868 int
2869 nfs_sillyrename(dvp, vp, cnp)
2870 struct vnode *dvp, *vp;
2871 struct componentname *cnp;
2872 {
2873 struct sillyrename *sp;
2874 struct nfsnode *np;
2875 int error;
2876 short pid;
2877
2878 cache_purge(dvp);
2879 np = VTONFS(vp);
2880 #ifndef DIAGNOSTIC
2881 if (vp->v_type == VDIR)
2882 panic("nfs: sillyrename dir");
2883 #endif
2884 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2885 M_NFSREQ, M_WAITOK);
2886 sp->s_cred = crdup(cnp->cn_cred);
2887 sp->s_dvp = dvp;
2888 VREF(dvp);
2889
2890 /* Fudge together a funny name */
2891 pid = cnp->cn_proc->p_pid;
2892 memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2893 sp->s_namlen = 12;
2894 sp->s_name[8] = hextoasc[pid & 0xf];
2895 sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
2896 sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
2897 sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
2898
2899 /* Try lookitups until we get one that isn't there */
2900 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2901 cnp->cn_proc, (struct nfsnode **)0) == 0) {
2902 sp->s_name[4]++;
2903 if (sp->s_name[4] > 'z') {
2904 error = EINVAL;
2905 goto bad;
2906 }
2907 }
2908 error = nfs_renameit(dvp, cnp, sp);
2909 if (error)
2910 goto bad;
2911 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2912 cnp->cn_proc, &np);
2913 np->n_sillyrename = sp;
2914 return (0);
2915 bad:
2916 vrele(sp->s_dvp);
2917 crfree(sp->s_cred);
2918 free((caddr_t)sp, M_NFSREQ);
2919 return (error);
2920 }
2921
2922 /*
2923 * Look up a file name and optionally either update the file handle or
2924 * allocate an nfsnode, depending on the value of npp.
2925 * npp == NULL --> just do the lookup
2926 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2927 * handled too
2928 * *npp != NULL --> update the file handle in the vnode
2929 */
2930 int
2931 nfs_lookitup(dvp, name, len, cred, procp, npp)
2932 struct vnode *dvp;
2933 const char *name;
2934 int len;
2935 struct ucred *cred;
2936 struct proc *procp;
2937 struct nfsnode **npp;
2938 {
2939 u_int32_t *tl;
2940 caddr_t cp;
2941 int32_t t1, t2;
2942 struct vnode *newvp = (struct vnode *)0;
2943 struct nfsnode *np, *dnp = VTONFS(dvp);
2944 caddr_t bpos, dpos, cp2;
2945 int error = 0, fhlen, attrflag;
2946 struct mbuf *mreq, *mrep, *md, *mb;
2947 nfsfh_t *nfhp;
2948 const int v3 = NFS_ISV3(dvp);
2949
2950 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2951 nfsm_reqhead(dnp, NFSPROC_LOOKUP,
2952 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2953 nfsm_fhtom(dnp, v3);
2954 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2955 nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred);
2956 if (npp && !error) {
2957 nfsm_getfh(nfhp, fhlen, v3);
2958 if (*npp) {
2959 np = *npp;
2960 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2961 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2962 np->n_fhp = &np->n_fh;
2963 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2964 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2965 memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
2966 np->n_fhsize = fhlen;
2967 newvp = NFSTOV(np);
2968 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2969 VREF(dvp);
2970 newvp = dvp;
2971 } else {
2972 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2973 if (error) {
2974 m_freem(mrep);
2975 return (error);
2976 }
2977 newvp = NFSTOV(np);
2978 }
2979 if (v3) {
2980 nfsm_postop_attr(newvp, attrflag, 0);
2981 if (!attrflag && *npp == NULL) {
2982 m_freem(mrep);
2983 vput(newvp);
2984 return (ENOENT);
2985 }
2986 } else
2987 nfsm_loadattr(newvp, (struct vattr *)0, 0);
2988 }
2989 nfsm_reqdone;
2990 if (npp && *npp == NULL) {
2991 if (error) {
2992 if (newvp)
2993 vput(newvp);
2994 } else
2995 *npp = np;
2996 }
2997 return (error);
2998 }
2999
3000 /*
3001 * Nfs Version 3 commit rpc
3002 */
3003 int
3004 nfs_commit(vp, offset, cnt, procp)
3005 struct vnode *vp;
3006 off_t offset;
3007 uint32_t cnt;
3008 struct proc *procp;
3009 {
3010 caddr_t cp;
3011 u_int32_t *tl;
3012 int32_t t1, t2;
3013 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3014 caddr_t bpos, dpos, cp2;
3015 int error = 0, wccflag = NFSV3_WCCRATTR;
3016 struct mbuf *mreq, *mrep, *md, *mb;
3017 struct nfsnode *np;
3018
3019 KASSERT(NFS_ISV3(vp));
3020
3021 #ifdef NFS_DEBUG_COMMIT
3022 printf("commit %lu - %lu\n", (unsigned long)offset,
3023 (unsigned long)(offset + cnt));
3024 #endif
3025
3026 simple_lock(&nmp->nm_slock);
3027 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
3028 simple_unlock(&nmp->nm_slock);
3029 return (0);
3030 }
3031 simple_unlock(&nmp->nm_slock);
3032 nfsstats.rpccnt[NFSPROC_COMMIT]++;
3033 np = VTONFS(vp);
3034 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
3035 nfsm_fhtom(np, 1);
3036 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3037 txdr_hyper(offset, tl);
3038 tl += 2;
3039 *tl = txdr_unsigned(cnt);
3040 nfsm_request(np, NFSPROC_COMMIT, procp, np->n_wcred);
3041 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC);
3042 if (!error) {
3043 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
3044 simple_lock(&nmp->nm_slock);
3045 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
3046 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
3047 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
3048 error = NFSERR_STALEWRITEVERF;
3049 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
3050 }
3051 simple_unlock(&nmp->nm_slock);
3052 }
3053 nfsm_reqdone;
3054 return (error);
3055 }
3056
3057 /*
3058 * Kludge City..
3059 * - make nfs_bmap() essentially a no-op that does no translation
3060 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
3061 * (Maybe I could use the process's page mapping, but I was concerned that
3062 * Kernel Write might not be enabled and also figured copyout() would do
3063 * a lot more work than memcpy() and also it currently happens in the
3064 * context of the swapper process (2).
3065 */
3066 int
3067 nfs_bmap(v)
3068 void *v;
3069 {
3070 struct vop_bmap_args /* {
3071 struct vnode *a_vp;
3072 daddr_t a_bn;
3073 struct vnode **a_vpp;
3074 daddr_t *a_bnp;
3075 int *a_runp;
3076 } */ *ap = v;
3077 struct vnode *vp = ap->a_vp;
3078 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
3079
3080 if (ap->a_vpp != NULL)
3081 *ap->a_vpp = vp;
3082 if (ap->a_bnp != NULL)
3083 *ap->a_bnp = ap->a_bn << bshift;
3084 if (ap->a_runp != NULL)
3085 *ap->a_runp = 1024 * 1024; /* XXX */
3086 return (0);
3087 }
3088
3089 /*
3090 * Strategy routine.
3091 * For async requests when nfsiod(s) are running, queue the request by
3092 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3093 * request.
3094 */
3095 int
3096 nfs_strategy(v)
3097 void *v;
3098 {
3099 struct vop_strategy_args *ap = v;
3100 struct buf *bp = ap->a_bp;
3101 struct proc *p;
3102 int error = 0;
3103
3104 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3105 panic("nfs physio/async");
3106 if (bp->b_flags & B_ASYNC)
3107 p = NULL;
3108 else
3109 p = curproc; /* XXX */
3110
3111 /*
3112 * If the op is asynchronous and an i/o daemon is waiting
3113 * queue the request, wake it up and wait for completion
3114 * otherwise just do it ourselves.
3115 */
3116
3117 if ((bp->b_flags & B_ASYNC) == 0 ||
3118 nfs_asyncio(bp))
3119 error = nfs_doio(bp, p);
3120 return (error);
3121 }
3122
3123 /*
3124 * fsync vnode op. Just call nfs_flush() with commit == 1.
3125 */
3126 /* ARGSUSED */
3127 int
3128 nfs_fsync(v)
3129 void *v;
3130 {
3131 struct vop_fsync_args /* {
3132 struct vnodeop_desc *a_desc;
3133 struct vnode * a_vp;
3134 struct ucred * a_cred;
3135 int a_flags;
3136 off_t offlo;
3137 off_t offhi;
3138 struct proc * a_p;
3139 } */ *ap = v;
3140
3141 struct vnode *vp = ap->a_vp;
3142
3143 if (vp->v_type != VREG)
3144 return 0;
3145
3146 return (nfs_flush(vp, ap->a_cred,
3147 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1));
3148 }
3149
3150 /*
3151 * Flush all the data associated with a vnode.
3152 */
3153 int
3154 nfs_flush(vp, cred, waitfor, p, commit)
3155 struct vnode *vp;
3156 struct ucred *cred;
3157 int waitfor;
3158 struct proc *p;
3159 int commit;
3160 {
3161 struct nfsnode *np = VTONFS(vp);
3162 int error;
3163 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3164 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3165
3166 simple_lock(&vp->v_interlock);
3167 error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3168 if (np->n_flag & NWRITEERR) {
3169 error = np->n_error;
3170 np->n_flag &= ~NWRITEERR;
3171 }
3172 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3173 return (error);
3174 }
3175
3176 /*
3177 * Return POSIX pathconf information applicable to nfs.
3178 *
3179 * N.B. The NFS V2 protocol doesn't support this RPC.
3180 */
3181 /* ARGSUSED */
3182 int
3183 nfs_pathconf(v)
3184 void *v;
3185 {
3186 struct vop_pathconf_args /* {
3187 struct vnode *a_vp;
3188 int a_name;
3189 register_t *a_retval;
3190 } */ *ap = v;
3191 struct nfsv3_pathconf *pcp;
3192 struct vnode *vp = ap->a_vp;
3193 struct mbuf *mreq, *mrep, *md, *mb;
3194 int32_t t1, t2;
3195 u_int32_t *tl;
3196 caddr_t bpos, dpos, cp, cp2;
3197 int error = 0, attrflag;
3198 #ifndef NFS_V2_ONLY
3199 struct nfsmount *nmp;
3200 unsigned int l;
3201 u_int64_t maxsize;
3202 #endif
3203 const int v3 = NFS_ISV3(vp);
3204 struct nfsnode *np = VTONFS(vp);
3205
3206 switch (ap->a_name) {
3207 /* Names that can be resolved locally. */
3208 case _PC_PIPE_BUF:
3209 *ap->a_retval = PIPE_BUF;
3210 break;
3211 case _PC_SYNC_IO:
3212 *ap->a_retval = 1;
3213 break;
3214 /* Names that cannot be resolved locally; do an RPC, if possible. */
3215 case _PC_LINK_MAX:
3216 case _PC_NAME_MAX:
3217 case _PC_CHOWN_RESTRICTED:
3218 case _PC_NO_TRUNC:
3219 if (!v3) {
3220 error = EINVAL;
3221 break;
3222 }
3223 nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3224 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3225 nfsm_fhtom(np, 1);
3226 nfsm_request(np, NFSPROC_PATHCONF,
3227 curproc, curproc->p_ucred); /* XXX */
3228 nfsm_postop_attr(vp, attrflag, 0);
3229 if (!error) {
3230 nfsm_dissect(pcp, struct nfsv3_pathconf *,
3231 NFSX_V3PATHCONF);
3232 switch (ap->a_name) {
3233 case _PC_LINK_MAX:
3234 *ap->a_retval =
3235 fxdr_unsigned(register_t, pcp->pc_linkmax);
3236 break;
3237 case _PC_NAME_MAX:
3238 *ap->a_retval =
3239 fxdr_unsigned(register_t, pcp->pc_namemax);
3240 break;
3241 case _PC_CHOWN_RESTRICTED:
3242 *ap->a_retval =
3243 (pcp->pc_chownrestricted == nfs_true);
3244 break;
3245 case _PC_NO_TRUNC:
3246 *ap->a_retval =
3247 (pcp->pc_notrunc == nfs_true);
3248 break;
3249 }
3250 }
3251 nfsm_reqdone;
3252 break;
3253 case _PC_FILESIZEBITS:
3254 #ifndef NFS_V2_ONLY
3255 if (v3) {
3256 nmp = VFSTONFS(vp->v_mount);
3257 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3258 if ((error = nfs_fsinfo(nmp, vp,
3259 curproc->p_ucred, curproc)) != 0) /* XXX */
3260 break;
3261 for (l = 0, maxsize = nmp->nm_maxfilesize;
3262 (maxsize >> l) > 0; l++)
3263 ;
3264 *ap->a_retval = l + 1;
3265 } else
3266 #endif
3267 {
3268 *ap->a_retval = 32; /* NFS V2 limitation */
3269 }
3270 break;
3271 default:
3272 error = EINVAL;
3273 break;
3274 }
3275
3276 return (error);
3277 }
3278
3279 /*
3280 * NFS advisory byte-level locks.
3281 */
3282 int
3283 nfs_advlock(v)
3284 void *v;
3285 {
3286 struct vop_advlock_args /* {
3287 struct vnode *a_vp;
3288 caddr_t a_id;
3289 int a_op;
3290 struct flock *a_fl;
3291 int a_flags;
3292 } */ *ap = v;
3293 struct nfsnode *np = VTONFS(ap->a_vp);
3294
3295 return lf_advlock(ap, &np->n_lockf, np->n_size);
3296 }
3297
3298 /*
3299 * Print out the contents of an nfsnode.
3300 */
3301 int
3302 nfs_print(v)
3303 void *v;
3304 {
3305 struct vop_print_args /* {
3306 struct vnode *a_vp;
3307 } */ *ap = v;
3308 struct vnode *vp = ap->a_vp;
3309 struct nfsnode *np = VTONFS(vp);
3310
3311 printf("tag VT_NFS, fileid %ld fsid 0x%lx",
3312 np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3313 if (vp->v_type == VFIFO)
3314 fifo_printinfo(vp);
3315 printf("\n");
3316 return (0);
3317 }
3318
3319 /*
3320 * NFS file truncation.
3321 */
3322 int
3323 nfs_truncate(v)
3324 void *v;
3325 {
3326 #if 0
3327 struct vop_truncate_args /* {
3328 struct vnode *a_vp;
3329 off_t a_length;
3330 int a_flags;
3331 struct ucred *a_cred;
3332 struct proc *a_p;
3333 } */ *ap = v;
3334 #endif
3335
3336 /* Use nfs_setattr */
3337 return (EOPNOTSUPP);
3338 }
3339
3340 /*
3341 * NFS update.
3342 */
3343 int
3344 nfs_update(v)
3345 void *v;
3346 #if 0
3347 struct vop_update_args /* {
3348 struct vnode *a_vp;
3349 struct timespec *a_ta;
3350 struct timespec *a_tm;
3351 int a_waitfor;
3352 } */ *ap = v;
3353 #endif
3354 {
3355
3356 /* Use nfs_setattr */
3357 return (EOPNOTSUPP);
3358 }
3359
3360 /*
3361 * Just call bwrite().
3362 */
3363 int
3364 nfs_bwrite(v)
3365 void *v;
3366 {
3367 struct vop_bwrite_args /* {
3368 struct vnode *a_bp;
3369 } */ *ap = v;
3370
3371 return (bwrite(ap->a_bp));
3372 }
3373
3374 /*
3375 * nfs unlock wrapper.
3376 */
3377 int
3378 nfs_unlock(void *v)
3379 {
3380 struct vop_unlock_args /* {
3381 struct vnode *a_vp;
3382 int a_flags;
3383 } */ *ap = v;
3384 struct vnode *vp = ap->a_vp;
3385
3386 /*
3387 * VOP_UNLOCK can be called by nfs_loadattrcache
3388 * with v_data == 0.
3389 */
3390 if (VTONFS(vp)) {
3391 nfs_delayedtruncate(vp);
3392 }
3393
3394 return genfs_unlock(v);
3395 }
3396
3397 /*
3398 * nfs special file access vnode op.
3399 * Essentially just get vattr and then imitate iaccess() since the device is
3400 * local to the client.
3401 */
3402 int
3403 nfsspec_access(v)
3404 void *v;
3405 {
3406 struct vop_access_args /* {
3407 struct vnode *a_vp;
3408 int a_mode;
3409 struct ucred *a_cred;
3410 struct proc *a_p;
3411 } */ *ap = v;
3412 struct vattr va;
3413 struct vnode *vp = ap->a_vp;
3414 int error;
3415
3416 error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3417 if (error)
3418 return (error);
3419
3420 /*
3421 * Disallow write attempts on filesystems mounted read-only;
3422 * unless the file is a socket, fifo, or a block or character
3423 * device resident on the filesystem.
3424 */
3425 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3426 switch (vp->v_type) {
3427 case VREG:
3428 case VDIR:
3429 case VLNK:
3430 return (EROFS);
3431 default:
3432 break;
3433 }
3434 }
3435
3436 return (vaccess(va.va_type, va.va_mode,
3437 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3438 }
3439
3440 /*
3441 * Read wrapper for special devices.
3442 */
3443 int
3444 nfsspec_read(v)
3445 void *v;
3446 {
3447 struct vop_read_args /* {
3448 struct vnode *a_vp;
3449 struct uio *a_uio;
3450 int a_ioflag;
3451 struct ucred *a_cred;
3452 } */ *ap = v;
3453 struct nfsnode *np = VTONFS(ap->a_vp);
3454
3455 /*
3456 * Set access flag.
3457 */
3458 np->n_flag |= NACC;
3459 np->n_atim.tv_sec = time.tv_sec;
3460 np->n_atim.tv_nsec = time.tv_usec * 1000;
3461 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3462 }
3463
3464 /*
3465 * Write wrapper for special devices.
3466 */
3467 int
3468 nfsspec_write(v)
3469 void *v;
3470 {
3471 struct vop_write_args /* {
3472 struct vnode *a_vp;
3473 struct uio *a_uio;
3474 int a_ioflag;
3475 struct ucred *a_cred;
3476 } */ *ap = v;
3477 struct nfsnode *np = VTONFS(ap->a_vp);
3478
3479 /*
3480 * Set update flag.
3481 */
3482 np->n_flag |= NUPD;
3483 np->n_mtim.tv_sec = time.tv_sec;
3484 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3485 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3486 }
3487
3488 /*
3489 * Close wrapper for special devices.
3490 *
3491 * Update the times on the nfsnode then do device close.
3492 */
3493 int
3494 nfsspec_close(v)
3495 void *v;
3496 {
3497 struct vop_close_args /* {
3498 struct vnode *a_vp;
3499 int a_fflag;
3500 struct ucred *a_cred;
3501 struct proc *a_p;
3502 } */ *ap = v;
3503 struct vnode *vp = ap->a_vp;
3504 struct nfsnode *np = VTONFS(vp);
3505 struct vattr vattr;
3506
3507 if (np->n_flag & (NACC | NUPD)) {
3508 np->n_flag |= NCHG;
3509 if (vp->v_usecount == 1 &&
3510 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3511 VATTR_NULL(&vattr);
3512 if (np->n_flag & NACC)
3513 vattr.va_atime = np->n_atim;
3514 if (np->n_flag & NUPD)
3515 vattr.va_mtime = np->n_mtim;
3516 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3517 }
3518 }
3519 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3520 }
3521
3522 /*
3523 * Read wrapper for fifos.
3524 */
3525 int
3526 nfsfifo_read(v)
3527 void *v;
3528 {
3529 struct vop_read_args /* {
3530 struct vnode *a_vp;
3531 struct uio *a_uio;
3532 int a_ioflag;
3533 struct ucred *a_cred;
3534 } */ *ap = v;
3535 struct nfsnode *np = VTONFS(ap->a_vp);
3536
3537 /*
3538 * Set access flag.
3539 */
3540 np->n_flag |= NACC;
3541 np->n_atim.tv_sec = time.tv_sec;
3542 np->n_atim.tv_nsec = time.tv_usec * 1000;
3543 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3544 }
3545
3546 /*
3547 * Write wrapper for fifos.
3548 */
3549 int
3550 nfsfifo_write(v)
3551 void *v;
3552 {
3553 struct vop_write_args /* {
3554 struct vnode *a_vp;
3555 struct uio *a_uio;
3556 int a_ioflag;
3557 struct ucred *a_cred;
3558 } */ *ap = v;
3559 struct nfsnode *np = VTONFS(ap->a_vp);
3560
3561 /*
3562 * Set update flag.
3563 */
3564 np->n_flag |= NUPD;
3565 np->n_mtim.tv_sec = time.tv_sec;
3566 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3567 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3568 }
3569
3570 /*
3571 * Close wrapper for fifos.
3572 *
3573 * Update the times on the nfsnode then do fifo close.
3574 */
3575 int
3576 nfsfifo_close(v)
3577 void *v;
3578 {
3579 struct vop_close_args /* {
3580 struct vnode *a_vp;
3581 int a_fflag;
3582 struct ucred *a_cred;
3583 struct proc *a_p;
3584 } */ *ap = v;
3585 struct vnode *vp = ap->a_vp;
3586 struct nfsnode *np = VTONFS(vp);
3587 struct vattr vattr;
3588
3589 if (np->n_flag & (NACC | NUPD)) {
3590 if (np->n_flag & NACC) {
3591 np->n_atim.tv_sec = time.tv_sec;
3592 np->n_atim.tv_nsec = time.tv_usec * 1000;
3593 }
3594 if (np->n_flag & NUPD) {
3595 np->n_mtim.tv_sec = time.tv_sec;
3596 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3597 }
3598 np->n_flag |= NCHG;
3599 if (vp->v_usecount == 1 &&
3600 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3601 VATTR_NULL(&vattr);
3602 if (np->n_flag & NACC)
3603 vattr.va_atime = np->n_atim;
3604 if (np->n_flag & NUPD)
3605 vattr.va_mtime = np->n_mtim;
3606 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3607 }
3608 }
3609 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3610 }
Cache object: 3159edf77dbd4a5608f8115c99e5e6c1
|