nfs_vnops.c revision 1.178 1 /* $NetBSD: nfs_vnops.c,v 1.178 2003/09/17 09:10:00 yamt Exp $ */
2
3 /*
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. 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 1.178 2003/09/17 09:10:00 yamt Exp $");
43
44 #include "opt_nfs.h"
45 #include "opt_uvmhist.h"
46
47 #include <sys/param.h>
48 #include <sys/proc.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/resourcevar.h>
52 #include <sys/proc.h>
53 #include <sys/mount.h>
54 #include <sys/buf.h>
55 #include <sys/malloc.h>
56 #include <sys/mbuf.h>
57 #include <sys/namei.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/fcntl.h>
61 #include <sys/hash.h>
62 #include <sys/lockf.h>
63 #include <sys/stat.h>
64 #include <sys/unistd.h>
65
66 #include <uvm/uvm_extern.h>
67 #include <uvm/uvm.h>
68
69 #include <miscfs/fifofs/fifo.h>
70 #include <miscfs/genfs/genfs.h>
71 #include <miscfs/specfs/specdev.h>
72
73 #include <nfs/rpcv2.h>
74 #include <nfs/nfsproto.h>
75 #include <nfs/nfs.h>
76 #include <nfs/nfsnode.h>
77 #include <nfs/nfsmount.h>
78 #include <nfs/xdr_subs.h>
79 #include <nfs/nfsm_subs.h>
80 #include <nfs/nqnfs.h>
81 #include <nfs/nfs_var.h>
82
83 #include <net/if.h>
84 #include <netinet/in.h>
85 #include <netinet/in_var.h>
86
87 /*
88 * Global vfs data structures for nfs
89 */
90 int (**nfsv2_vnodeop_p) __P((void *));
91 const struct vnodeopv_entry_desc nfsv2_vnodeop_entries[] = {
92 { &vop_default_desc, vn_default_error },
93 { &vop_lookup_desc, nfs_lookup }, /* lookup */
94 { &vop_create_desc, nfs_create }, /* create */
95 { &vop_mknod_desc, nfs_mknod }, /* mknod */
96 { &vop_open_desc, nfs_open }, /* open */
97 { &vop_close_desc, nfs_close }, /* close */
98 { &vop_access_desc, nfs_access }, /* access */
99 { &vop_getattr_desc, nfs_getattr }, /* getattr */
100 { &vop_setattr_desc, nfs_setattr }, /* setattr */
101 { &vop_read_desc, nfs_read }, /* read */
102 { &vop_write_desc, nfs_write }, /* write */
103 { &vop_lease_desc, nfs_lease_check }, /* lease */
104 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
105 { &vop_ioctl_desc, nfs_ioctl }, /* ioctl */
106 { &vop_poll_desc, nfs_poll }, /* poll */
107 { &vop_kqfilter_desc, nfs_kqfilter }, /* kqfilter */
108 { &vop_revoke_desc, nfs_revoke }, /* revoke */
109 { &vop_mmap_desc, nfs_mmap }, /* mmap */
110 { &vop_fsync_desc, nfs_fsync }, /* fsync */
111 { &vop_seek_desc, nfs_seek }, /* seek */
112 { &vop_remove_desc, nfs_remove }, /* remove */
113 { &vop_link_desc, nfs_link }, /* link */
114 { &vop_rename_desc, nfs_rename }, /* rename */
115 { &vop_mkdir_desc, nfs_mkdir }, /* mkdir */
116 { &vop_rmdir_desc, nfs_rmdir }, /* rmdir */
117 { &vop_symlink_desc, nfs_symlink }, /* symlink */
118 { &vop_readdir_desc, nfs_readdir }, /* readdir */
119 { &vop_readlink_desc, nfs_readlink }, /* readlink */
120 { &vop_abortop_desc, nfs_abortop }, /* abortop */
121 { &vop_inactive_desc, nfs_inactive }, /* inactive */
122 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
123 { &vop_lock_desc, nfs_lock }, /* lock */
124 { &vop_unlock_desc, nfs_unlock }, /* unlock */
125 { &vop_bmap_desc, nfs_bmap }, /* bmap */
126 { &vop_strategy_desc, nfs_strategy }, /* strategy */
127 { &vop_print_desc, nfs_print }, /* print */
128 { &vop_islocked_desc, nfs_islocked }, /* islocked */
129 { &vop_pathconf_desc, nfs_pathconf }, /* pathconf */
130 { &vop_advlock_desc, nfs_advlock }, /* advlock */
131 { &vop_blkatoff_desc, nfs_blkatoff }, /* blkatoff */
132 { &vop_valloc_desc, nfs_valloc }, /* valloc */
133 { &vop_reallocblks_desc, nfs_reallocblks }, /* reallocblks */
134 { &vop_vfree_desc, nfs_vfree }, /* vfree */
135 { &vop_truncate_desc, nfs_truncate }, /* truncate */
136 { &vop_update_desc, nfs_update }, /* update */
137 { &vop_bwrite_desc, nfs_bwrite }, /* bwrite */
138 { &vop_getpages_desc, nfs_getpages }, /* getpages */
139 { &vop_putpages_desc, genfs_putpages }, /* putpages */
140 { NULL, NULL }
141 };
142 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc =
143 { &nfsv2_vnodeop_p, nfsv2_vnodeop_entries };
144
145 /*
146 * Special device vnode ops
147 */
148 int (**spec_nfsv2nodeop_p) __P((void *));
149 const struct vnodeopv_entry_desc spec_nfsv2nodeop_entries[] = {
150 { &vop_default_desc, vn_default_error },
151 { &vop_lookup_desc, spec_lookup }, /* lookup */
152 { &vop_create_desc, spec_create }, /* create */
153 { &vop_mknod_desc, spec_mknod }, /* mknod */
154 { &vop_open_desc, spec_open }, /* open */
155 { &vop_close_desc, nfsspec_close }, /* close */
156 { &vop_access_desc, nfsspec_access }, /* access */
157 { &vop_getattr_desc, nfs_getattr }, /* getattr */
158 { &vop_setattr_desc, nfs_setattr }, /* setattr */
159 { &vop_read_desc, nfsspec_read }, /* read */
160 { &vop_write_desc, nfsspec_write }, /* write */
161 { &vop_lease_desc, spec_lease_check }, /* lease */
162 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
163 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */
164 { &vop_poll_desc, spec_poll }, /* poll */
165 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */
166 { &vop_revoke_desc, spec_revoke }, /* revoke */
167 { &vop_mmap_desc, spec_mmap }, /* mmap */
168 { &vop_fsync_desc, spec_fsync }, /* fsync */
169 { &vop_seek_desc, spec_seek }, /* seek */
170 { &vop_remove_desc, spec_remove }, /* remove */
171 { &vop_link_desc, spec_link }, /* link */
172 { &vop_rename_desc, spec_rename }, /* rename */
173 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */
174 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */
175 { &vop_symlink_desc, spec_symlink }, /* symlink */
176 { &vop_readdir_desc, spec_readdir }, /* readdir */
177 { &vop_readlink_desc, spec_readlink }, /* readlink */
178 { &vop_abortop_desc, spec_abortop }, /* abortop */
179 { &vop_inactive_desc, nfs_inactive }, /* inactive */
180 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
181 { &vop_lock_desc, nfs_lock }, /* lock */
182 { &vop_unlock_desc, nfs_unlock }, /* unlock */
183 { &vop_bmap_desc, spec_bmap }, /* bmap */
184 { &vop_strategy_desc, spec_strategy }, /* strategy */
185 { &vop_print_desc, nfs_print }, /* print */
186 { &vop_islocked_desc, nfs_islocked }, /* islocked */
187 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */
188 { &vop_advlock_desc, spec_advlock }, /* advlock */
189 { &vop_blkatoff_desc, spec_blkatoff }, /* blkatoff */
190 { &vop_valloc_desc, spec_valloc }, /* valloc */
191 { &vop_reallocblks_desc, spec_reallocblks }, /* reallocblks */
192 { &vop_vfree_desc, spec_vfree }, /* vfree */
193 { &vop_truncate_desc, spec_truncate }, /* truncate */
194 { &vop_update_desc, nfs_update }, /* update */
195 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
196 { &vop_getpages_desc, spec_getpages }, /* getpages */
197 { &vop_putpages_desc, spec_putpages }, /* putpages */
198 { NULL, NULL }
199 };
200 const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc =
201 { &spec_nfsv2nodeop_p, spec_nfsv2nodeop_entries };
202
203 int (**fifo_nfsv2nodeop_p) __P((void *));
204 const struct vnodeopv_entry_desc fifo_nfsv2nodeop_entries[] = {
205 { &vop_default_desc, vn_default_error },
206 { &vop_lookup_desc, fifo_lookup }, /* lookup */
207 { &vop_create_desc, fifo_create }, /* create */
208 { &vop_mknod_desc, fifo_mknod }, /* mknod */
209 { &vop_open_desc, fifo_open }, /* open */
210 { &vop_close_desc, nfsfifo_close }, /* close */
211 { &vop_access_desc, nfsspec_access }, /* access */
212 { &vop_getattr_desc, nfs_getattr }, /* getattr */
213 { &vop_setattr_desc, nfs_setattr }, /* setattr */
214 { &vop_read_desc, nfsfifo_read }, /* read */
215 { &vop_write_desc, nfsfifo_write }, /* write */
216 { &vop_lease_desc, fifo_lease_check }, /* lease */
217 { &vop_fcntl_desc, genfs_fcntl }, /* fcntl */
218 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */
219 { &vop_poll_desc, fifo_poll }, /* poll */
220 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */
221 { &vop_revoke_desc, fifo_revoke }, /* revoke */
222 { &vop_mmap_desc, fifo_mmap }, /* mmap */
223 { &vop_fsync_desc, nfs_fsync }, /* fsync */
224 { &vop_seek_desc, fifo_seek }, /* seek */
225 { &vop_remove_desc, fifo_remove }, /* remove */
226 { &vop_link_desc, fifo_link }, /* link */
227 { &vop_rename_desc, fifo_rename }, /* rename */
228 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */
229 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */
230 { &vop_symlink_desc, fifo_symlink }, /* symlink */
231 { &vop_readdir_desc, fifo_readdir }, /* readdir */
232 { &vop_readlink_desc, fifo_readlink }, /* readlink */
233 { &vop_abortop_desc, fifo_abortop }, /* abortop */
234 { &vop_inactive_desc, nfs_inactive }, /* inactive */
235 { &vop_reclaim_desc, nfs_reclaim }, /* reclaim */
236 { &vop_lock_desc, nfs_lock }, /* lock */
237 { &vop_unlock_desc, nfs_unlock }, /* unlock */
238 { &vop_bmap_desc, fifo_bmap }, /* bmap */
239 { &vop_strategy_desc, genfs_badop }, /* strategy */
240 { &vop_print_desc, nfs_print }, /* print */
241 { &vop_islocked_desc, nfs_islocked }, /* islocked */
242 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */
243 { &vop_advlock_desc, fifo_advlock }, /* advlock */
244 { &vop_blkatoff_desc, fifo_blkatoff }, /* blkatoff */
245 { &vop_valloc_desc, fifo_valloc }, /* valloc */
246 { &vop_reallocblks_desc, fifo_reallocblks }, /* reallocblks */
247 { &vop_vfree_desc, fifo_vfree }, /* vfree */
248 { &vop_truncate_desc, fifo_truncate }, /* truncate */
249 { &vop_update_desc, nfs_update }, /* update */
250 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */
251 { &vop_putpages_desc, fifo_putpages }, /* putpages */
252 { NULL, NULL }
253 };
254 const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc =
255 { &fifo_nfsv2nodeop_p, fifo_nfsv2nodeop_entries };
256
257 static void nfs_writerpc_extfree(struct mbuf *, caddr_t, size_t, void *);
258
259 /*
260 * Global variables
261 */
262 extern u_int32_t nfs_true, nfs_false;
263 extern u_int32_t nfs_xdrneg1;
264 extern const nfstype nfsv3_type[9];
265
266 int nfs_numasync = 0;
267 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
268
269 /*
270 * nfs null call from vfs.
271 */
272 int
273 nfs_null(vp, cred, procp)
274 struct vnode *vp;
275 struct ucred *cred;
276 struct proc *procp;
277 {
278 caddr_t bpos, dpos;
279 int error = 0;
280 struct mbuf *mreq, *mrep, *md, *mb;
281 struct nfsnode *np = VTONFS(vp);
282
283 nfsm_reqhead(np, NFSPROC_NULL, 0);
284 nfsm_request(np, NFSPROC_NULL, procp, cred);
285 nfsm_reqdone;
286 return (error);
287 }
288
289 /*
290 * nfs access vnode op.
291 * For nfs version 2, just return ok. File accesses may fail later.
292 * For nfs version 3, use the access rpc to check accessibility. If file modes
293 * are changed on the server, accesses might still fail later.
294 */
295 int
296 nfs_access(v)
297 void *v;
298 {
299 struct vop_access_args /* {
300 struct vnode *a_vp;
301 int a_mode;
302 struct ucred *a_cred;
303 struct proc *a_p;
304 } */ *ap = v;
305 struct vnode *vp = ap->a_vp;
306 u_int32_t *tl;
307 caddr_t cp;
308 int32_t t1, t2;
309 caddr_t bpos, dpos, cp2;
310 int error = 0, attrflag, cachevalid;
311 struct mbuf *mreq, *mrep, *md, *mb;
312 u_int32_t mode, rmode;
313 const int v3 = NFS_ISV3(vp);
314 struct nfsnode *np = VTONFS(vp);
315
316 cachevalid = (np->n_accstamp != -1 &&
317 (time.tv_sec - np->n_accstamp) < NFS_ATTRTIMEO(np) &&
318 np->n_accuid == ap->a_cred->cr_uid);
319
320 /*
321 * Check access cache first. If this request has been made for this
322 * uid shortly before, use the cached result.
323 */
324 if (cachevalid) {
325 if (!np->n_accerror) {
326 if ((np->n_accmode & ap->a_mode) == ap->a_mode)
327 return np->n_accerror;
328 } else if ((np->n_accmode & ap->a_mode) == np->n_accmode)
329 return np->n_accerror;
330 }
331
332 /*
333 * For nfs v3, do an access rpc, otherwise you are stuck emulating
334 * ufs_access() locally using the vattr. This may not be correct,
335 * since the server may apply other access criteria such as
336 * client uid-->server uid mapping that we do not know about, but
337 * this is better than just returning anything that is lying about
338 * in the cache.
339 */
340 if (v3) {
341 nfsstats.rpccnt[NFSPROC_ACCESS]++;
342 nfsm_reqhead(np, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
343 nfsm_fhtom(np, v3);
344 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
345 if (ap->a_mode & VREAD)
346 mode = NFSV3ACCESS_READ;
347 else
348 mode = 0;
349 if (vp->v_type != VDIR) {
350 if (ap->a_mode & VWRITE)
351 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
352 if (ap->a_mode & VEXEC)
353 mode |= NFSV3ACCESS_EXECUTE;
354 } else {
355 if (ap->a_mode & VWRITE)
356 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
357 NFSV3ACCESS_DELETE);
358 if (ap->a_mode & VEXEC)
359 mode |= NFSV3ACCESS_LOOKUP;
360 }
361 *tl = txdr_unsigned(mode);
362 nfsm_request(np, NFSPROC_ACCESS, ap->a_p, ap->a_cred);
363 nfsm_postop_attr(vp, attrflag, 0);
364 if (!error) {
365 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
366 rmode = fxdr_unsigned(u_int32_t, *tl);
367 /*
368 * The NFS V3 spec does not clarify whether or not
369 * the returned access bits can be a superset of
370 * the ones requested, so...
371 */
372 if ((rmode & mode) != mode)
373 error = EACCES;
374 }
375 nfsm_reqdone;
376 } else
377 return (nfsspec_access(ap));
378 /*
379 * Disallow write attempts on filesystems mounted read-only;
380 * unless the file is a socket, fifo, or a block or character
381 * device resident on the filesystem.
382 */
383 if (!error && (ap->a_mode & VWRITE) &&
384 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
385 switch (vp->v_type) {
386 case VREG:
387 case VDIR:
388 case VLNK:
389 error = EROFS;
390 default:
391 break;
392 }
393 }
394
395 if (!error || error == EACCES) {
396 /*
397 * If we got the same result as for a previous,
398 * different request, OR it in. Don't update
399 * the timestamp in that case.
400 */
401 if (cachevalid && np->n_accstamp != -1 &&
402 error == np->n_accerror) {
403 if (!error)
404 np->n_accmode |= ap->a_mode;
405 else if ((np->n_accmode & ap->a_mode) == ap->a_mode)
406 np->n_accmode = ap->a_mode;
407 } else {
408 np->n_accstamp = time.tv_sec;
409 np->n_accuid = ap->a_cred->cr_uid;
410 np->n_accmode = ap->a_mode;
411 np->n_accerror = error;
412 }
413 }
414
415 return (error);
416 }
417
418 /*
419 * nfs open vnode op
420 * Check to see if the type is ok
421 * and that deletion is not in progress.
422 * For paged in text files, you will need to flush the page cache
423 * if consistency is lost.
424 */
425 /* ARGSUSED */
426 int
427 nfs_open(v)
428 void *v;
429 {
430 struct vop_open_args /* {
431 struct vnode *a_vp;
432 int a_mode;
433 struct ucred *a_cred;
434 struct proc *a_p;
435 } */ *ap = v;
436 struct vnode *vp = ap->a_vp;
437 struct nfsnode *np = VTONFS(vp);
438 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
439 struct vattr vattr;
440 int error;
441
442 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) {
443 return (EACCES);
444 }
445
446 /*
447 * Initialize read and write creds here, for swapfiles
448 * and other paths that don't set the creds themselves.
449 */
450
451 if (ap->a_mode & FREAD) {
452 if (np->n_rcred) {
453 crfree(np->n_rcred);
454 }
455 np->n_rcred = ap->a_cred;
456 crhold(np->n_rcred);
457 }
458 if (ap->a_mode & FWRITE) {
459 if (np->n_wcred) {
460 crfree(np->n_wcred);
461 }
462 np->n_wcred = ap->a_cred;
463 crhold(np->n_wcred);
464 }
465
466 #ifndef NFS_V2_ONLY
467 /*
468 * Get a valid lease. If cached data is stale, flush it.
469 */
470 if (nmp->nm_flag & NFSMNT_NQNFS) {
471 if (NQNFS_CKINVALID(vp, np, ND_READ)) {
472 do {
473 error = nqnfs_getlease(vp, ND_READ, ap->a_cred,
474 ap->a_p);
475 } while (error == NQNFS_EXPIRED);
476 if (error)
477 return (error);
478 if (np->n_lrev != np->n_brev ||
479 (np->n_flag & NQNFSNONCACHE)) {
480 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
481 ap->a_p, 1)) == EINTR)
482 return (error);
483 np->n_brev = np->n_lrev;
484 }
485 }
486 } else
487 #endif
488 {
489 if (np->n_flag & NMODIFIED) {
490 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
491 ap->a_p, 1)) == EINTR)
492 return (error);
493 np->n_attrstamp = 0;
494 if (vp->v_type == VDIR) {
495 nfs_invaldircache(vp, 0);
496 np->n_direofoffset = 0;
497 }
498 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
499 if (error)
500 return (error);
501 np->n_mtime = vattr.va_mtime.tv_sec;
502 } else {
503 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_p);
504 if (error)
505 return (error);
506 if (np->n_mtime != vattr.va_mtime.tv_sec) {
507 if (vp->v_type == VDIR) {
508 nfs_invaldircache(vp, 0);
509 np->n_direofoffset = 0;
510 }
511 if ((error = nfs_vinvalbuf(vp, V_SAVE,
512 ap->a_cred, ap->a_p, 1)) == EINTR)
513 return (error);
514 np->n_mtime = vattr.va_mtime.tv_sec;
515 }
516 }
517 }
518 if ((nmp->nm_flag & NFSMNT_NQNFS) == 0)
519 np->n_attrstamp = 0; /* For Open/Close consistency */
520 return (0);
521 }
522
523 /*
524 * nfs close vnode op
525 * What an NFS client should do upon close after writing is a debatable issue.
526 * Most NFS clients push delayed writes to the server upon close, basically for
527 * two reasons:
528 * 1 - So that any write errors may be reported back to the client process
529 * doing the close system call. By far the two most likely errors are
530 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
531 * 2 - To put a worst case upper bound on cache inconsistency between
532 * multiple clients for the file.
533 * There is also a consistency problem for Version 2 of the protocol w.r.t.
534 * not being able to tell if other clients are writing a file concurrently,
535 * since there is no way of knowing if the changed modify time in the reply
536 * is only due to the write for this client.
537 * (NFS Version 3 provides weak cache consistency data in the reply that
538 * should be sufficient to detect and handle this case.)
539 *
540 * The current code does the following:
541 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
542 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
543 * or commit them (this satisfies 1 and 2 except for the
544 * case where the server crashes after this close but
545 * before the commit RPC, which is felt to be "good
546 * enough". Changing the last argument to nfs_flush() to
547 * a 1 would force a commit operation, if it is felt a
548 * commit is necessary now.
549 * for NQNFS - do nothing now, since 2 is dealt with via leases and
550 * 1 should be dealt with via an fsync() system call for
551 * cases where write errors are important.
552 */
553 /* ARGSUSED */
554 int
555 nfs_close(v)
556 void *v;
557 {
558 struct vop_close_args /* {
559 struct vnodeop_desc *a_desc;
560 struct vnode *a_vp;
561 int a_fflag;
562 struct ucred *a_cred;
563 struct proc *a_p;
564 } */ *ap = v;
565 struct vnode *vp = ap->a_vp;
566 struct nfsnode *np = VTONFS(vp);
567 int error = 0;
568 UVMHIST_FUNC("nfs_close"); UVMHIST_CALLED(ubchist);
569
570 if (vp->v_type == VREG) {
571 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS) == 0 &&
572 (np->n_flag & NMODIFIED)) {
573 if (NFS_ISV3(vp)) {
574 error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_p, 0);
575 np->n_flag &= ~NMODIFIED;
576 } else
577 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_p, 1);
578 np->n_attrstamp = 0;
579 }
580 if (np->n_flag & NWRITEERR) {
581 np->n_flag &= ~NWRITEERR;
582 error = np->n_error;
583 }
584 }
585 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
586 return (error);
587 }
588
589 /*
590 * nfs getattr call from vfs.
591 */
592 int
593 nfs_getattr(v)
594 void *v;
595 {
596 struct vop_getattr_args /* {
597 struct vnode *a_vp;
598 struct vattr *a_vap;
599 struct ucred *a_cred;
600 struct proc *a_p;
601 } */ *ap = v;
602 struct vnode *vp = ap->a_vp;
603 struct nfsnode *np = VTONFS(vp);
604 caddr_t cp;
605 u_int32_t *tl;
606 int32_t t1, t2;
607 caddr_t bpos, dpos;
608 int error = 0;
609 struct mbuf *mreq, *mrep, *md, *mb;
610 const int v3 = NFS_ISV3(vp);
611
612 /*
613 * Update local times for special files.
614 */
615 if (np->n_flag & (NACC | NUPD))
616 np->n_flag |= NCHG;
617 /*
618 * First look in the cache.
619 */
620 if (nfs_getattrcache(vp, ap->a_vap) == 0)
621 return (0);
622 nfsstats.rpccnt[NFSPROC_GETATTR]++;
623 nfsm_reqhead(np, NFSPROC_GETATTR, NFSX_FH(v3));
624 nfsm_fhtom(np, v3);
625 nfsm_request(np, NFSPROC_GETATTR, ap->a_p, ap->a_cred);
626 if (!error) {
627 nfsm_loadattr(vp, ap->a_vap, 0);
628 if (vp->v_type == VDIR &&
629 ap->a_vap->va_blocksize < NFS_DIRFRAGSIZ)
630 ap->a_vap->va_blocksize = NFS_DIRFRAGSIZ;
631 }
632 nfsm_reqdone;
633 return (error);
634 }
635
636 /*
637 * nfs setattr call.
638 */
639 int
640 nfs_setattr(v)
641 void *v;
642 {
643 struct vop_setattr_args /* {
644 struct vnodeop_desc *a_desc;
645 struct vnode *a_vp;
646 struct vattr *a_vap;
647 struct ucred *a_cred;
648 struct proc *a_p;
649 } */ *ap = v;
650 struct vnode *vp = ap->a_vp;
651 struct nfsnode *np = VTONFS(vp);
652 struct vattr *vap = ap->a_vap;
653 int error = 0;
654 u_quad_t tsize = 0;
655
656 /*
657 * Setting of flags is not supported.
658 */
659 if (vap->va_flags != VNOVAL)
660 return (EOPNOTSUPP);
661
662 /*
663 * Disallow write attempts if the filesystem is mounted read-only.
664 */
665 if ((vap->va_uid != (uid_t)VNOVAL ||
666 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
667 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
668 (vp->v_mount->mnt_flag & MNT_RDONLY))
669 return (EROFS);
670 if (vap->va_size != VNOVAL) {
671 switch (vp->v_type) {
672 case VDIR:
673 return (EISDIR);
674 case VCHR:
675 case VBLK:
676 case VSOCK:
677 case VFIFO:
678 if (vap->va_mtime.tv_sec == VNOVAL &&
679 vap->va_atime.tv_sec == VNOVAL &&
680 vap->va_mode == (mode_t)VNOVAL &&
681 vap->va_uid == (uid_t)VNOVAL &&
682 vap->va_gid == (gid_t)VNOVAL)
683 return (0);
684 vap->va_size = VNOVAL;
685 break;
686 default:
687 /*
688 * Disallow write attempts if the filesystem is
689 * mounted read-only.
690 */
691 if (vp->v_mount->mnt_flag & MNT_RDONLY)
692 return (EROFS);
693 uvm_vnp_setsize(vp, vap->va_size);
694 tsize = np->n_size;
695 np->n_size = vap->va_size;
696 if (vap->va_size == 0)
697 error = nfs_vinvalbuf(vp, 0,
698 ap->a_cred, ap->a_p, 1);
699 else
700 error = nfs_vinvalbuf(vp, V_SAVE,
701 ap->a_cred, ap->a_p, 1);
702 if (error) {
703 uvm_vnp_setsize(vp, tsize);
704 return (error);
705 }
706 np->n_vattr->va_size = vap->va_size;
707 }
708 } else if ((vap->va_mtime.tv_sec != VNOVAL ||
709 vap->va_atime.tv_sec != VNOVAL) &&
710 vp->v_type == VREG &&
711 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred,
712 ap->a_p, 1)) == EINTR)
713 return (error);
714 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_p);
715 if (error && vap->va_size != VNOVAL) {
716 np->n_size = np->n_vattr->va_size = tsize;
717 uvm_vnp_setsize(vp, np->n_size);
718 }
719 VN_KNOTE(vp, NOTE_ATTRIB);
720 return (error);
721 }
722
723 /*
724 * Do an nfs setattr rpc.
725 */
726 int
727 nfs_setattrrpc(vp, vap, cred, procp)
728 struct vnode *vp;
729 struct vattr *vap;
730 struct ucred *cred;
731 struct proc *procp;
732 {
733 struct nfsv2_sattr *sp;
734 caddr_t cp;
735 int32_t t1, t2;
736 caddr_t bpos, dpos, cp2;
737 u_int32_t *tl;
738 int error = 0, wccflag = NFSV3_WCCRATTR;
739 struct mbuf *mreq, *mrep, *md, *mb;
740 const int v3 = NFS_ISV3(vp);
741 struct nfsnode *np = VTONFS(vp);
742
743 nfsstats.rpccnt[NFSPROC_SETATTR]++;
744 nfsm_reqhead(np, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
745 nfsm_fhtom(np, v3);
746 if (v3) {
747 nfsm_v3attrbuild(vap, TRUE);
748 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
749 *tl = nfs_false;
750 } else {
751 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
752 if (vap->va_mode == (mode_t)VNOVAL)
753 sp->sa_mode = nfs_xdrneg1;
754 else
755 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
756 if (vap->va_uid == (uid_t)VNOVAL)
757 sp->sa_uid = nfs_xdrneg1;
758 else
759 sp->sa_uid = txdr_unsigned(vap->va_uid);
760 if (vap->va_gid == (gid_t)VNOVAL)
761 sp->sa_gid = nfs_xdrneg1;
762 else
763 sp->sa_gid = txdr_unsigned(vap->va_gid);
764 sp->sa_size = txdr_unsigned(vap->va_size);
765 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
766 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
767 }
768 nfsm_request(np, NFSPROC_SETATTR, procp, cred);
769 if (v3) {
770 nfsm_wcc_data(vp, wccflag, 0);
771 } else
772 nfsm_loadattr(vp, (struct vattr *)0, 0);
773 nfsm_reqdone;
774 return (error);
775 }
776
777 /*
778 * nfs lookup call, one step at a time...
779 * First look in cache
780 * If not found, unlock the directory nfsnode and do the rpc
781 *
782 * This code is full of lock/unlock statements and checks, because
783 * we continue after cache_lookup has finished (we need to check
784 * with the attr cache and do an rpc if it has timed out). This means
785 * that the locking effects of cache_lookup have to be taken into
786 * account.
787 */
788 int
789 nfs_lookup(v)
790 void *v;
791 {
792 struct vop_lookup_args /* {
793 struct vnodeop_desc *a_desc;
794 struct vnode *a_dvp;
795 struct vnode **a_vpp;
796 struct componentname *a_cnp;
797 } */ *ap = v;
798 struct componentname *cnp = ap->a_cnp;
799 struct vnode *dvp = ap->a_dvp;
800 struct vnode **vpp = ap->a_vpp;
801 int flags;
802 struct vnode *newvp;
803 u_int32_t *tl;
804 caddr_t cp;
805 int32_t t1, t2;
806 caddr_t bpos, dpos, cp2;
807 struct mbuf *mreq, *mrep, *md, *mb;
808 long len;
809 nfsfh_t *fhp;
810 struct nfsnode *np;
811 int lockparent, wantparent, error = 0, attrflag, fhsize;
812 const int v3 = NFS_ISV3(dvp);
813
814 cnp->cn_flags &= ~PDIRUNLOCK;
815 flags = cnp->cn_flags;
816
817 *vpp = NULLVP;
818 newvp = NULLVP;
819 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
820 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
821 return (EROFS);
822 if (dvp->v_type != VDIR)
823 return (ENOTDIR);
824
825 lockparent = flags & LOCKPARENT;
826 wantparent = flags & (LOCKPARENT|WANTPARENT);
827 np = VTONFS(dvp);
828
829 /*
830 * Before tediously performing a linear scan of the directory,
831 * check the name cache to see if the directory/name pair
832 * we are looking for is known already.
833 * If the directory/name pair is found in the name cache,
834 * we have to ensure the directory has not changed from
835 * the time the cache entry has been created. If it has,
836 * the cache entry has to be ignored.
837 */
838 if ((error = cache_lookup(dvp, vpp, cnp)) >= 0) {
839 struct vattr vattr;
840 int err2;
841
842 if (error && error != ENOENT) {
843 *vpp = NULLVP;
844 return error;
845 }
846
847 if (cnp->cn_flags & PDIRUNLOCK) {
848 err2 = vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
849 if (err2 != 0) {
850 *vpp = NULLVP;
851 return err2;
852 }
853 cnp->cn_flags &= ~PDIRUNLOCK;
854 }
855
856 err2 = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_proc);
857 if (err2 != 0) {
858 if (error == 0) {
859 if (*vpp != dvp)
860 vput(*vpp);
861 else
862 vrele(*vpp);
863 }
864 *vpp = NULLVP;
865 return err2;
866 }
867
868 if (error == ENOENT) {
869 if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred,
870 cnp->cn_proc) && timespeccmp(&vattr.va_mtime,
871 &VTONFS(dvp)->n_nctime, ==))
872 return ENOENT;
873 cache_purge(dvp);
874 timespecclear(&np->n_nctime);
875 goto dorpc;
876 }
877
878 newvp = *vpp;
879 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, cnp->cn_proc)
880 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime)
881 {
882 nfsstats.lookupcache_hits++;
883 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
884 cnp->cn_flags |= SAVENAME;
885 if ((!lockparent || !(flags & ISLASTCN)) &&
886 newvp != dvp)
887 VOP_UNLOCK(dvp, 0);
888 return (0);
889 }
890 cache_purge(newvp);
891 if (newvp != dvp)
892 vput(newvp);
893 else
894 vrele(newvp);
895 *vpp = NULLVP;
896 }
897 dorpc:
898 error = 0;
899 newvp = NULLVP;
900 nfsstats.lookupcache_misses++;
901 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
902 len = cnp->cn_namelen;
903 nfsm_reqhead(np, NFSPROC_LOOKUP,
904 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
905 nfsm_fhtom(np, v3);
906 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
907 nfsm_request(np, NFSPROC_LOOKUP, cnp->cn_proc, cnp->cn_cred);
908 if (error) {
909 nfsm_postop_attr(dvp, attrflag, 0);
910 m_freem(mrep);
911 goto nfsmout;
912 }
913 nfsm_getfh(fhp, fhsize, v3);
914
915 /*
916 * Handle RENAME case...
917 */
918 if (cnp->cn_nameiop == RENAME && wantparent && (flags & ISLASTCN)) {
919 if (NFS_CMPFH(np, fhp, fhsize)) {
920 m_freem(mrep);
921 return (EISDIR);
922 }
923 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
924 if (error) {
925 m_freem(mrep);
926 return error;
927 }
928 newvp = NFSTOV(np);
929 if (v3) {
930 nfsm_postop_attr(newvp, attrflag, 0);
931 nfsm_postop_attr(dvp, attrflag, 0);
932 } else
933 nfsm_loadattr(newvp, (struct vattr *)0, 0);
934 *vpp = newvp;
935 m_freem(mrep);
936 cnp->cn_flags |= SAVENAME;
937 if (!lockparent) {
938 VOP_UNLOCK(dvp, 0);
939 cnp->cn_flags |= PDIRUNLOCK;
940 }
941 return (0);
942 }
943
944 /*
945 * The postop attr handling is duplicated for each if case,
946 * because it should be done while dvp is locked (unlocking
947 * dvp is different for each case).
948 */
949
950 if (NFS_CMPFH(np, fhp, fhsize)) {
951 /*
952 * "." lookup
953 */
954 VREF(dvp);
955 newvp = dvp;
956 if (v3) {
957 nfsm_postop_attr(newvp, attrflag, 0);
958 nfsm_postop_attr(dvp, attrflag, 0);
959 } else
960 nfsm_loadattr(newvp, (struct vattr *)0, 0);
961 } else if (flags & ISDOTDOT) {
962 /*
963 * ".." lookup
964 */
965 VOP_UNLOCK(dvp, 0);
966 cnp->cn_flags |= PDIRUNLOCK;
967
968 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
969 if (error) {
970 if (vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY) == 0)
971 cnp->cn_flags &= ~PDIRUNLOCK;
972 m_freem(mrep);
973 return error;
974 }
975 newvp = NFSTOV(np);
976
977 if (v3) {
978 nfsm_postop_attr(newvp, attrflag, 0);
979 nfsm_postop_attr(dvp, attrflag, 0);
980 } else
981 nfsm_loadattr(newvp, (struct vattr *)0, 0);
982
983 if (lockparent && (flags & ISLASTCN)) {
984 if ((error = vn_lock(dvp, LK_EXCLUSIVE))) {
985 m_freem(mrep);
986 vput(newvp);
987 return error;
988 }
989 cnp->cn_flags &= ~PDIRUNLOCK;
990 }
991 } else {
992 /*
993 * Other lookups.
994 */
995 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
996 if (error) {
997 m_freem(mrep);
998 return error;
999 }
1000 newvp = NFSTOV(np);
1001 if (v3) {
1002 nfsm_postop_attr(newvp, attrflag, 0);
1003 nfsm_postop_attr(dvp, attrflag, 0);
1004 } else
1005 nfsm_loadattr(newvp, (struct vattr *)0, 0);
1006 if (!lockparent || !(flags & ISLASTCN)) {
1007 VOP_UNLOCK(dvp, 0);
1008 cnp->cn_flags |= PDIRUNLOCK;
1009 }
1010 }
1011 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1012 cnp->cn_flags |= SAVENAME;
1013 if ((cnp->cn_flags & MAKEENTRY) &&
1014 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1015 np->n_ctime = np->n_vattr->va_ctime.tv_sec;
1016 cache_enter(dvp, newvp, cnp);
1017 }
1018 *vpp = newvp;
1019 nfsm_reqdone;
1020 if (error) {
1021 /*
1022 * We get here only because of errors returned by
1023 * the RPC. Otherwise we'll have returned above
1024 * (the nfsm_* macros will jump to nfsm_reqdone
1025 * on error).
1026 */
1027 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1028 cnp->cn_nameiop != CREATE) {
1029 if (timespecisset(&VTONFS(dvp)->n_nctime))
1030 VTONFS(dvp)->n_nctime =
1031 VTONFS(dvp)->n_vattr->va_mtime;
1032
1033 cache_enter(dvp, NULL, cnp);
1034 }
1035 if (newvp != NULLVP) {
1036 vrele(newvp);
1037 if (newvp != dvp)
1038 VOP_UNLOCK(newvp, 0);
1039 }
1040 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1041 (flags & ISLASTCN) && error == ENOENT) {
1042 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1043 error = EROFS;
1044 else
1045 error = EJUSTRETURN;
1046 }
1047 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1048 cnp->cn_flags |= SAVENAME;
1049 *vpp = NULL;
1050 }
1051
1052 return error;
1053 }
1054
1055 /*
1056 * nfs read call.
1057 * Just call nfs_bioread() to do the work.
1058 */
1059 int
1060 nfs_read(v)
1061 void *v;
1062 {
1063 struct vop_read_args /* {
1064 struct vnode *a_vp;
1065 struct uio *a_uio;
1066 int a_ioflag;
1067 struct ucred *a_cred;
1068 } */ *ap = v;
1069 struct vnode *vp = ap->a_vp;
1070
1071 if (vp->v_type != VREG)
1072 return (EPERM);
1073 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred, 0));
1074 }
1075
1076 /*
1077 * nfs readlink call
1078 */
1079 int
1080 nfs_readlink(v)
1081 void *v;
1082 {
1083 struct vop_readlink_args /* {
1084 struct vnode *a_vp;
1085 struct uio *a_uio;
1086 struct ucred *a_cred;
1087 } */ *ap = v;
1088 struct vnode *vp = ap->a_vp;
1089
1090 if (vp->v_type != VLNK)
1091 return (EPERM);
1092 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred, 0));
1093 }
1094
1095 /*
1096 * Do a readlink rpc.
1097 * Called by nfs_doio() from below the buffer cache.
1098 */
1099 int
1100 nfs_readlinkrpc(vp, uiop, cred)
1101 struct vnode *vp;
1102 struct uio *uiop;
1103 struct ucred *cred;
1104 {
1105 u_int32_t *tl;
1106 caddr_t cp;
1107 int32_t t1, t2;
1108 caddr_t bpos, dpos, cp2;
1109 int error = 0, attrflag;
1110 uint32_t len;
1111 struct mbuf *mreq, *mrep, *md, *mb;
1112 const int v3 = NFS_ISV3(vp);
1113 struct nfsnode *np = VTONFS(vp);
1114
1115 nfsstats.rpccnt[NFSPROC_READLINK]++;
1116 nfsm_reqhead(np, NFSPROC_READLINK, NFSX_FH(v3));
1117 nfsm_fhtom(np, v3);
1118 nfsm_request(np, NFSPROC_READLINK, uiop->uio_procp, cred);
1119 if (v3)
1120 nfsm_postop_attr(vp, attrflag, 0);
1121 if (!error) {
1122 if (v3) {
1123 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1124 len = fxdr_unsigned(uint32_t, *tl);
1125 if (len > MAXPATHLEN) {
1126 /*
1127 * this pathname is too long for us.
1128 */
1129 m_freem(mrep);
1130 /* Solaris returns EINVAL. should we follow? */
1131 error = ENAMETOOLONG;
1132 goto nfsmout;
1133 }
1134 } else {
1135 nfsm_strsiz(len, NFS_MAXPATHLEN);
1136 }
1137 nfsm_mtouio(uiop, len);
1138 }
1139 nfsm_reqdone;
1140 return (error);
1141 }
1142
1143 /*
1144 * nfs read rpc call
1145 * Ditto above
1146 */
1147 int
1148 nfs_readrpc(vp, uiop)
1149 struct vnode *vp;
1150 struct uio *uiop;
1151 {
1152 u_int32_t *tl;
1153 caddr_t cp;
1154 int32_t t1, t2;
1155 caddr_t bpos, dpos, cp2;
1156 struct mbuf *mreq, *mrep, *md, *mb;
1157 struct nfsmount *nmp;
1158 int error = 0, len, retlen, tsiz, eof, attrflag;
1159 const int v3 = NFS_ISV3(vp);
1160 struct nfsnode *np = VTONFS(vp);
1161
1162 #ifndef nolint
1163 eof = 0;
1164 #endif
1165 nmp = VFSTONFS(vp->v_mount);
1166 tsiz = uiop->uio_resid;
1167 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1168 return (EFBIG);
1169 while (tsiz > 0) {
1170 nfsstats.rpccnt[NFSPROC_READ]++;
1171 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
1172 nfsm_reqhead(np, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1173 nfsm_fhtom(np, v3);
1174 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED * 3);
1175 if (v3) {
1176 txdr_hyper(uiop->uio_offset, tl);
1177 *(tl + 2) = txdr_unsigned(len);
1178 } else {
1179 *tl++ = txdr_unsigned(uiop->uio_offset);
1180 *tl++ = txdr_unsigned(len);
1181 *tl = 0;
1182 }
1183 nfsm_request(np, NFSPROC_READ, uiop->uio_procp,
1184 VTONFS(vp)->n_rcred);
1185 if (v3) {
1186 nfsm_postop_attr(vp, attrflag, NAC_NOTRUNC);
1187 if (error) {
1188 m_freem(mrep);
1189 goto nfsmout;
1190 }
1191 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1192 eof = fxdr_unsigned(int, *(tl + 1));
1193 } else
1194 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1195 nfsm_strsiz(retlen, nmp->nm_rsize);
1196 nfsm_mtouio(uiop, retlen);
1197 m_freem(mrep);
1198 tsiz -= retlen;
1199 if (v3) {
1200 if (eof || retlen == 0)
1201 tsiz = 0;
1202 } else if (retlen < len)
1203 tsiz = 0;
1204 }
1205 nfsmout:
1206 return (error);
1207 }
1208
1209 struct nfs_writerpc_context {
1210 struct simplelock nwc_slock;
1211 volatile int nwc_mbufcount;
1212 };
1213
1214 /*
1215 * free mbuf used to refer protected pages while write rpc call.
1216 * called at splvm.
1217 */
1218 static void
1219 nfs_writerpc_extfree(struct mbuf *m, caddr_t buf, size_t size, void *arg)
1220 {
1221 struct nfs_writerpc_context *ctx = arg;
1222
1223 KASSERT(m != NULL);
1224 KASSERT(ctx != NULL);
1225 pool_cache_put(&mbpool_cache, m);
1226 simple_lock(&ctx->nwc_slock);
1227 if (--ctx->nwc_mbufcount == 0) {
1228 wakeup(ctx);
1229 }
1230 simple_unlock(&ctx->nwc_slock);
1231 }
1232
1233 /*
1234 * nfs write call
1235 */
1236 int
1237 nfs_writerpc(vp, uiop, iomode, pageprotected, stalewriteverfp)
1238 struct vnode *vp;
1239 struct uio *uiop;
1240 int *iomode;
1241 boolean_t pageprotected;
1242 boolean_t *stalewriteverfp;
1243 {
1244 u_int32_t *tl;
1245 caddr_t cp;
1246 int32_t t1, t2;
1247 caddr_t bpos, dpos, cp2;
1248 struct mbuf *mreq, *mrep, *md, *mb;
1249 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1250 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1251 const int v3 = NFS_ISV3(vp);
1252 int committed = NFSV3WRITE_FILESYNC;
1253 struct nfsnode *np = VTONFS(vp);
1254 struct nfs_writerpc_context ctx;
1255 int s;
1256 struct lwp *l;
1257 size_t origresid;
1258
1259 simple_lock_init(&ctx.nwc_slock);
1260 ctx.nwc_mbufcount = 1;
1261
1262 if (vp->v_mount->mnt_flag & MNT_RDONLY) {
1263 panic("writerpc readonly vp %p", vp);
1264 }
1265
1266 #ifdef DIAGNOSTIC
1267 if (uiop->uio_iovcnt != 1)
1268 panic("nfs: writerpc iovcnt > 1");
1269 #endif
1270 tsiz = uiop->uio_resid;
1271 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
1272 return (EFBIG);
1273 if (pageprotected) {
1274 l = curlwp;
1275 PHOLD(l);
1276 }
1277 retry:
1278 origresid = uiop->uio_resid;
1279 KASSERT(origresid == uiop->uio_iov->iov_len);
1280 while (tsiz > 0) {
1281 uint32_t datalen; /* data bytes need to be allocated in mbuf */
1282 uint32_t backup;
1283 boolean_t stalewriteverf = FALSE;
1284
1285 nfsstats.rpccnt[NFSPROC_WRITE]++;
1286 len = min(tsiz, nmp->nm_wsize);
1287 datalen = pageprotected ? 0 : nfsm_rndup(len);
1288 nfsm_reqhead(np, NFSPROC_WRITE,
1289 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + datalen);
1290 nfsm_fhtom(np, v3);
1291 if (v3) {
1292 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1293 txdr_hyper(uiop->uio_offset, tl);
1294 tl += 2;
1295 *tl++ = txdr_unsigned(len);
1296 *tl++ = txdr_unsigned(*iomode);
1297 *tl = txdr_unsigned(len);
1298 } else {
1299 u_int32_t x;
1300
1301 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1302 /* Set both "begin" and "current" to non-garbage. */
1303 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1304 *tl++ = x; /* "begin offset" */
1305 *tl++ = x; /* "current offset" */
1306 x = txdr_unsigned(len);
1307 *tl++ = x; /* total to this offset */
1308 *tl = x; /* size of this write */
1309
1310 }
1311 if (pageprotected) {
1312 /*
1313 * since we know pages can't be modified during i/o,
1314 * no need to copy them for us.
1315 */
1316 struct mbuf *m;
1317 struct iovec *iovp = uiop->uio_iov;
1318
1319 m = m_get(M_WAIT, MT_DATA);
1320 MCLAIM(m, &nfs_mowner);
1321 MEXTADD(m, iovp->iov_base, len, M_MBUF,
1322 nfs_writerpc_extfree, &ctx);
1323 m->m_flags |= M_EXT_ROMAP;
1324 m->m_len = len;
1325 mb->m_next = m;
1326 /*
1327 * no need to maintain mb and bpos here
1328 * because no one care them later.
1329 */
1330 #if 0
1331 mb = m;
1332 bpos = mtod(caddr_t, mb) + mb->m_len;
1333 #endif
1334 iovp->iov_base = (char *)iovp->iov_base + len;
1335 iovp->iov_len -= len;
1336 uiop->uio_offset += len;
1337 uiop->uio_resid -= len;
1338 s = splvm();
1339 simple_lock(&ctx.nwc_slock);
1340 ctx.nwc_mbufcount++;
1341 simple_unlock(&ctx.nwc_slock);
1342 splx(s);
1343 } else {
1344 nfsm_uiotom(uiop, len);
1345 }
1346 nfsm_request(np, NFSPROC_WRITE, uiop->uio_procp,
1347 VTONFS(vp)->n_wcred);
1348 if (v3) {
1349 wccflag = NFSV3_WCCCHK;
1350 nfsm_wcc_data(vp, wccflag, NAC_NOTRUNC);
1351 if (!error) {
1352 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED
1353 + NFSX_V3WRITEVERF);
1354 rlen = fxdr_unsigned(int, *tl++);
1355 if (rlen == 0) {
1356 error = NFSERR_IO;
1357 m_freem(mrep);
1358 break;
1359 } else if (rlen < len) {
1360 backup = len - rlen;
1361 uiop->uio_iov->iov_base =
1362 (caddr_t)uiop->uio_iov->iov_base -
1363 backup;
1364 uiop->uio_iov->iov_len += backup;
1365 uiop->uio_offset -= backup;
1366 uiop->uio_resid += backup;
1367 len = rlen;
1368 }
1369 commit = fxdr_unsigned(int, *tl++);
1370
1371 /*
1372 * Return the lowest committment level
1373 * obtained by any of the RPCs.
1374 */
1375 if (committed == NFSV3WRITE_FILESYNC)
1376 committed = commit;
1377 else if (committed == NFSV3WRITE_DATASYNC &&
1378 commit == NFSV3WRITE_UNSTABLE)
1379 committed = commit;
1380 simple_lock(&nmp->nm_slock);
1381 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0){
1382 memcpy(nmp->nm_writeverf, tl,
1383 NFSX_V3WRITEVERF);
1384 nmp->nm_iflag |= NFSMNT_HASWRITEVERF;
1385 } else if ((nmp->nm_iflag &
1386 NFSMNT_STALEWRITEVERF) ||
1387 memcmp(tl, nmp->nm_writeverf,
1388 NFSX_V3WRITEVERF)) {
1389 memcpy(nmp->nm_writeverf, tl,
1390 NFSX_V3WRITEVERF);
1391 /*
1392 * note NFSMNT_STALEWRITEVERF
1393 * if we're the first thread to
1394 * notice it.
1395 */
1396 if ((nmp->nm_iflag &
1397 NFSMNT_STALEWRITEVERF) == 0) {
1398 stalewriteverf = TRUE;
1399 nmp->nm_iflag |=
1400 NFSMNT_STALEWRITEVERF;
1401 }
1402 }
1403 simple_unlock(&nmp->nm_slock);
1404 }
1405 } else
1406 nfsm_loadattr(vp, (struct vattr *)0, NAC_NOTRUNC);
1407 if (wccflag)
1408 VTONFS(vp)->n_mtime =
1409 VTONFS(vp)->n_vattr->va_mtime.tv_sec;
1410 m_freem(mrep);
1411 if (error)
1412 break;
1413 tsiz -= len;
1414 if (stalewriteverf) {
1415 *stalewriteverfp = TRUE;
1416 stalewriteverf = FALSE;
1417 if (committed == NFSV3WRITE_UNSTABLE &&
1418 len != origresid) {
1419 /*
1420 * if our write requests weren't atomic but
1421 * unstable, datas in previous iterations
1422 * might have already been lost now.
1423 * then, we should resend them to nfsd.
1424 */
1425 backup = origresid - tsiz;
1426 uiop->uio_iov->iov_base =
1427 (caddr_t)uiop->uio_iov->iov_base - backup;
1428 uiop->uio_iov->iov_len += backup;
1429 uiop->uio_offset -= backup;
1430 uiop->uio_resid += backup;
1431 tsiz = origresid;
1432 goto retry;
1433 }
1434 }
1435 }
1436 if (pageprotected) {
1437 /*
1438 * wait until mbufs go away.
1439 * retransmitted mbufs can survive longer than rpc requests
1440 * themselves.
1441 */
1442 s = splvm();
1443 simple_lock(&ctx.nwc_slock);
1444 ctx.nwc_mbufcount--;
1445 while (ctx.nwc_mbufcount > 0) {
1446 ltsleep(&ctx, PRIBIO, "nfsmblk", 0, &ctx.nwc_slock);
1447 }
1448 simple_unlock(&ctx.nwc_slock);
1449 splx(s);
1450 PRELE(l);
1451 }
1452 nfsmout:
1453 *iomode = committed;
1454 if (error)
1455 uiop->uio_resid = tsiz;
1456 return (error);
1457 }
1458
1459 /*
1460 * nfs mknod rpc
1461 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1462 * mode set to specify the file type and the size field for rdev.
1463 */
1464 int
1465 nfs_mknodrpc(dvp, vpp, cnp, vap)
1466 struct vnode *dvp;
1467 struct vnode **vpp;
1468 struct componentname *cnp;
1469 struct vattr *vap;
1470 {
1471 struct nfsv2_sattr *sp;
1472 u_int32_t *tl;
1473 caddr_t cp;
1474 int32_t t1, t2;
1475 struct vnode *newvp = (struct vnode *)0;
1476 struct nfsnode *dnp, *np;
1477 char *cp2;
1478 caddr_t bpos, dpos;
1479 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1480 struct mbuf *mreq, *mrep, *md, *mb;
1481 u_int32_t rdev;
1482 const int v3 = NFS_ISV3(dvp);
1483
1484 if (vap->va_type == VCHR || vap->va_type == VBLK)
1485 rdev = txdr_unsigned(vap->va_rdev);
1486 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1487 rdev = nfs_xdrneg1;
1488 else {
1489 VOP_ABORTOP(dvp, cnp);
1490 vput(dvp);
1491 return (EOPNOTSUPP);
1492 }
1493 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1494 dnp = VTONFS(dvp);
1495 nfsm_reqhead(dnp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1496 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1497 nfsm_fhtom(dnp, v3);
1498 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1499 if (v3) {
1500 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1501 *tl++ = vtonfsv3_type(vap->va_type);
1502 nfsm_v3attrbuild(vap, FALSE);
1503 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1504 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1505 *tl++ = txdr_unsigned(major(vap->va_rdev));
1506 *tl = txdr_unsigned(minor(vap->va_rdev));
1507 }
1508 } else {
1509 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1510 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1511 sp->sa_uid = nfs_xdrneg1;
1512 sp->sa_gid = nfs_xdrneg1;
1513 sp->sa_size = rdev;
1514 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1515 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1516 }
1517 nfsm_request(dnp, NFSPROC_MKNOD, cnp->cn_proc, cnp->cn_cred);
1518 if (!error) {
1519 nfsm_mtofh(dvp, newvp, v3, gotvp);
1520 if (!gotvp) {
1521 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1522 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1523 if (!error)
1524 newvp = NFSTOV(np);
1525 }
1526 }
1527 if (v3)
1528 nfsm_wcc_data(dvp, wccflag, 0);
1529 nfsm_reqdone;
1530 if (error) {
1531 if (newvp)
1532 vput(newvp);
1533 } else {
1534 if (cnp->cn_flags & MAKEENTRY)
1535 cache_enter(dvp, newvp, cnp);
1536 *vpp = newvp;
1537 }
1538 PNBUF_PUT(cnp->cn_pnbuf);
1539 VTONFS(dvp)->n_flag |= NMODIFIED;
1540 if (!wccflag)
1541 VTONFS(dvp)->n_attrstamp = 0;
1542 vput(dvp);
1543 return (error);
1544 }
1545
1546 /*
1547 * nfs mknod vop
1548 * just call nfs_mknodrpc() to do the work.
1549 */
1550 /* ARGSUSED */
1551 int
1552 nfs_mknod(v)
1553 void *v;
1554 {
1555 struct vop_mknod_args /* {
1556 struct vnode *a_dvp;
1557 struct vnode **a_vpp;
1558 struct componentname *a_cnp;
1559 struct vattr *a_vap;
1560 } */ *ap = v;
1561 int error;
1562
1563 error = nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap);
1564 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1565 return (error);
1566 }
1567
1568 static u_long create_verf;
1569 /*
1570 * nfs file create call
1571 */
1572 int
1573 nfs_create(v)
1574 void *v;
1575 {
1576 struct vop_create_args /* {
1577 struct vnode *a_dvp;
1578 struct vnode **a_vpp;
1579 struct componentname *a_cnp;
1580 struct vattr *a_vap;
1581 } */ *ap = v;
1582 struct vnode *dvp = ap->a_dvp;
1583 struct vattr *vap = ap->a_vap;
1584 struct componentname *cnp = ap->a_cnp;
1585 struct nfsv2_sattr *sp;
1586 u_int32_t *tl;
1587 caddr_t cp;
1588 int32_t t1, t2;
1589 struct nfsnode *dnp, *np = (struct nfsnode *)0;
1590 struct vnode *newvp = (struct vnode *)0;
1591 caddr_t bpos, dpos, cp2;
1592 int error, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1593 struct mbuf *mreq, *mrep, *md, *mb;
1594 const int v3 = NFS_ISV3(dvp);
1595
1596 /*
1597 * Oops, not for me..
1598 */
1599 if (vap->va_type == VSOCK)
1600 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1601
1602 #ifdef VA_EXCLUSIVE
1603 if (vap->va_vaflags & VA_EXCLUSIVE)
1604 fmode |= O_EXCL;
1605 #endif
1606 again:
1607 error = 0;
1608 nfsstats.rpccnt[NFSPROC_CREATE]++;
1609 dnp = VTONFS(dvp);
1610 nfsm_reqhead(dnp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1611 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1612 nfsm_fhtom(dnp, v3);
1613 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1614 if (v3) {
1615 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1616 if (fmode & O_EXCL) {
1617 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1618 nfsm_build(tl, u_int32_t *, NFSX_V3CREATEVERF);
1619 #ifdef INET
1620 if (TAILQ_FIRST(&in_ifaddr))
1621 *tl++ = TAILQ_FIRST(&in_ifaddr)->ia_addr.sin_addr.s_addr;
1622 else
1623 *tl++ = create_verf;
1624 #else
1625 *tl++ = create_verf;
1626 #endif
1627 *tl = ++create_verf;
1628 } else {
1629 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1630 nfsm_v3attrbuild(vap, FALSE);
1631 }
1632 } else {
1633 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1634 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1635 sp->sa_uid = nfs_xdrneg1;
1636 sp->sa_gid = nfs_xdrneg1;
1637 sp->sa_size = 0;
1638 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1639 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1640 }
1641 nfsm_request(dnp, NFSPROC_CREATE, cnp->cn_proc, cnp->cn_cred);
1642 if (!error) {
1643 nfsm_mtofh(dvp, newvp, v3, gotvp);
1644 if (!gotvp) {
1645 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1646 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc, &np);
1647 if (!error)
1648 newvp = NFSTOV(np);
1649 }
1650 }
1651 if (v3)
1652 nfsm_wcc_data(dvp, wccflag, 0);
1653 nfsm_reqdone;
1654 if (error) {
1655 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1656 fmode &= ~O_EXCL;
1657 goto again;
1658 }
1659 if (newvp)
1660 vput(newvp);
1661 } else if (v3 && (fmode & O_EXCL))
1662 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_proc);
1663 if (!error) {
1664 if (cnp->cn_flags & MAKEENTRY)
1665 cache_enter(dvp, newvp, cnp);
1666 *ap->a_vpp = newvp;
1667 }
1668 PNBUF_PUT(cnp->cn_pnbuf);
1669 VTONFS(dvp)->n_flag |= NMODIFIED;
1670 if (!wccflag)
1671 VTONFS(dvp)->n_attrstamp = 0;
1672 VN_KNOTE(ap->a_dvp, NOTE_WRITE);
1673 vput(dvp);
1674 return (error);
1675 }
1676
1677 /*
1678 * nfs file remove call
1679 * To try and make nfs semantics closer to ufs semantics, a file that has
1680 * other processes using the vnode is renamed instead of removed and then
1681 * removed later on the last close.
1682 * - If v_usecount > 1
1683 * If a rename is not already in the works
1684 * call nfs_sillyrename() to set it up
1685 * else
1686 * do the remove rpc
1687 */
1688 int
1689 nfs_remove(v)
1690 void *v;
1691 {
1692 struct vop_remove_args /* {
1693 struct vnodeop_desc *a_desc;
1694 struct vnode * a_dvp;
1695 struct vnode * a_vp;
1696 struct componentname * a_cnp;
1697 } */ *ap = v;
1698 struct vnode *vp = ap->a_vp;
1699 struct vnode *dvp = ap->a_dvp;
1700 struct componentname *cnp = ap->a_cnp;
1701 struct nfsnode *np = VTONFS(vp);
1702 int error = 0;
1703 struct vattr vattr;
1704
1705 #ifndef DIAGNOSTIC
1706 if ((cnp->cn_flags & HASBUF) == 0)
1707 panic("nfs_remove: no name");
1708 if (vp->v_usecount < 1)
1709 panic("nfs_remove: bad v_usecount");
1710 #endif
1711 if (vp->v_type == VDIR)
1712 error = EPERM;
1713 else if (vp->v_usecount == 1 || (np->n_sillyrename &&
1714 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_proc) == 0 &&
1715 vattr.va_nlink > 1)) {
1716 /*
1717 * Purge the name cache so that the chance of a lookup for
1718 * the name succeeding while the remove is in progress is
1719 * minimized. Without node locking it can still happen, such
1720 * that an I/O op returns ESTALE, but since you get this if
1721 * another host removes the file..
1722 */
1723 cache_purge(vp);
1724 /*
1725 * throw away biocache buffers, mainly to avoid
1726 * unnecessary delayed writes later.
1727 */
1728 error = nfs_vinvalbuf(vp, 0, cnp->cn_cred, cnp->cn_proc, 1);
1729 /* Do the rpc */
1730 if (error != EINTR)
1731 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1732 cnp->cn_namelen, cnp->cn_cred, cnp->cn_proc);
1733 /*
1734 * Kludge City: If the first reply to the remove rpc is lost..
1735 * the reply to the retransmitted request will be ENOENT
1736 * since the file was in fact removed
1737 * Therefore, we cheat and return success.
1738 */
1739 if (error == ENOENT)
1740 error = 0;
1741 } else if (!np->n_sillyrename)
1742 error = nfs_sillyrename(dvp, vp, cnp);
1743 PNBUF_PUT(cnp->cn_pnbuf);
1744 if (!error && nfs_getattrcache(vp, &vattr) == 0 &&
1745 vattr.va_nlink == 1) {
1746 np->n_flag |= NREMOVED;
1747 }
1748 np->n_attrstamp = 0;
1749 VN_KNOTE(vp, NOTE_DELETE);
1750 VN_KNOTE(dvp, NOTE_WRITE);
1751 if (dvp == vp)
1752 vrele(vp);
1753 else
1754 vput(vp);
1755 vput(dvp);
1756 return (error);
1757 }
1758
1759 /*
1760 * nfs file remove rpc called from nfs_inactive
1761 */
1762 int
1763 nfs_removeit(sp)
1764 struct sillyrename *sp;
1765 {
1766
1767 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1768 (struct proc *)0));
1769 }
1770
1771 /*
1772 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1773 */
1774 int
1775 nfs_removerpc(dvp, name, namelen, cred, proc)
1776 struct vnode *dvp;
1777 const char *name;
1778 int namelen;
1779 struct ucred *cred;
1780 struct proc *proc;
1781 {
1782 u_int32_t *tl;
1783 caddr_t cp;
1784 int32_t t1, t2;
1785 caddr_t bpos, dpos, cp2;
1786 int error = 0, wccflag = NFSV3_WCCRATTR;
1787 struct mbuf *mreq, *mrep, *md, *mb;
1788 const int v3 = NFS_ISV3(dvp);
1789 struct nfsnode *dnp = VTONFS(dvp);
1790
1791 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1792 nfsm_reqhead(dnp, NFSPROC_REMOVE,
1793 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1794 nfsm_fhtom(dnp, v3);
1795 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1796 nfsm_request(dnp, NFSPROC_REMOVE, proc, cred);
1797 if (v3)
1798 nfsm_wcc_data(dvp, wccflag, 0);
1799 nfsm_reqdone;
1800 VTONFS(dvp)->n_flag |= NMODIFIED;
1801 if (!wccflag)
1802 VTONFS(dvp)->n_attrstamp = 0;
1803 return (error);
1804 }
1805
1806 /*
1807 * nfs file rename call
1808 */
1809 int
1810 nfs_rename(v)
1811 void *v;
1812 {
1813 struct vop_rename_args /* {
1814 struct vnode *a_fdvp;
1815 struct vnode *a_fvp;
1816 struct componentname *a_fcnp;
1817 struct vnode *a_tdvp;
1818 struct vnode *a_tvp;
1819 struct componentname *a_tcnp;
1820 } */ *ap = v;
1821 struct vnode *fvp = ap->a_fvp;
1822 struct vnode *tvp = ap->a_tvp;
1823 struct vnode *fdvp = ap->a_fdvp;
1824 struct vnode *tdvp = ap->a_tdvp;
1825 struct componentname *tcnp = ap->a_tcnp;
1826 struct componentname *fcnp = ap->a_fcnp;
1827 int error;
1828
1829 #ifndef DIAGNOSTIC
1830 if ((tcnp->cn_flags & HASBUF) == 0 ||
1831 (fcnp->cn_flags & HASBUF) == 0)
1832 panic("nfs_rename: no name");
1833 #endif
1834 /* Check for cross-device rename */
1835 if ((fvp->v_mount != tdvp->v_mount) ||
1836 (tvp && (fvp->v_mount != tvp->v_mount))) {
1837 error = EXDEV;
1838 goto out;
1839 }
1840
1841 /*
1842 * If the tvp exists and is in use, sillyrename it before doing the
1843 * rename of the new file over it.
1844 */
1845 if (tvp && tvp->v_usecount > 1 && !VTONFS(tvp)->n_sillyrename &&
1846 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1847 VN_KNOTE(tvp, NOTE_DELETE);
1848 vput(tvp);
1849 tvp = NULL;
1850 }
1851
1852 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1853 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1854 tcnp->cn_proc);
1855
1856 VN_KNOTE(fdvp, NOTE_WRITE);
1857 VN_KNOTE(tdvp, NOTE_WRITE);
1858 if (fvp->v_type == VDIR) {
1859 if (tvp != NULL && tvp->v_type == VDIR)
1860 cache_purge(tdvp);
1861 cache_purge(fdvp);
1862 }
1863 out:
1864 if (tdvp == tvp)
1865 vrele(tdvp);
1866 else
1867 vput(tdvp);
1868 if (tvp)
1869 vput(tvp);
1870 vrele(fdvp);
1871 vrele(fvp);
1872 /*
1873 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1874 */
1875 if (error == ENOENT)
1876 error = 0;
1877 return (error);
1878 }
1879
1880 /*
1881 * nfs file rename rpc called from nfs_remove() above
1882 */
1883 int
1884 nfs_renameit(sdvp, scnp, sp)
1885 struct vnode *sdvp;
1886 struct componentname *scnp;
1887 struct sillyrename *sp;
1888 {
1889 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen,
1890 sdvp, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_proc));
1891 }
1892
1893 /*
1894 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1895 */
1896 int
1897 nfs_renamerpc(fdvp, fnameptr, fnamelen, tdvp, tnameptr, tnamelen, cred, proc)
1898 struct vnode *fdvp;
1899 const char *fnameptr;
1900 int fnamelen;
1901 struct vnode *tdvp;
1902 const char *tnameptr;
1903 int tnamelen;
1904 struct ucred *cred;
1905 struct proc *proc;
1906 {
1907 u_int32_t *tl;
1908 caddr_t cp;
1909 int32_t t1, t2;
1910 caddr_t bpos, dpos, cp2;
1911 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1912 struct mbuf *mreq, *mrep, *md, *mb;
1913 const int v3 = NFS_ISV3(fdvp);
1914 struct nfsnode *fdnp = VTONFS(fdvp);
1915
1916 nfsstats.rpccnt[NFSPROC_RENAME]++;
1917 nfsm_reqhead(fdnp, NFSPROC_RENAME,
1918 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1919 nfsm_rndup(tnamelen));
1920 nfsm_fhtom(fdnp, v3);
1921 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1922 nfsm_fhtom(VTONFS(tdvp), v3);
1923 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1924 nfsm_request(fdnp, NFSPROC_RENAME, proc, cred);
1925 if (v3) {
1926 nfsm_wcc_data(fdvp, fwccflag, 0);
1927 nfsm_wcc_data(tdvp, twccflag, 0);
1928 }
1929 nfsm_reqdone;
1930 VTONFS(fdvp)->n_flag |= NMODIFIED;
1931 VTONFS(tdvp)->n_flag |= NMODIFIED;
1932 if (!fwccflag)
1933 VTONFS(fdvp)->n_attrstamp = 0;
1934 if (!twccflag)
1935 VTONFS(tdvp)->n_attrstamp = 0;
1936 return (error);
1937 }
1938
1939 /*
1940 * nfs hard link create call
1941 */
1942 int
1943 nfs_link(v)
1944 void *v;
1945 {
1946 struct vop_link_args /* {
1947 struct vnode *a_dvp;
1948 struct vnode *a_vp;
1949 struct componentname *a_cnp;
1950 } */ *ap = v;
1951 struct vnode *vp = ap->a_vp;
1952 struct vnode *dvp = ap->a_dvp;
1953 struct componentname *cnp = ap->a_cnp;
1954 u_int32_t *tl;
1955 caddr_t cp;
1956 int32_t t1, t2;
1957 caddr_t bpos, dpos, cp2;
1958 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1959 struct mbuf *mreq, *mrep, *md, *mb;
1960 /* XXX Should be const and initialised? */
1961 int v3;
1962 struct nfsnode *np;
1963
1964 if (dvp->v_mount != vp->v_mount) {
1965 VOP_ABORTOP(dvp, cnp);
1966 vput(dvp);
1967 return (EXDEV);
1968 }
1969 if (dvp != vp) {
1970 error = vn_lock(vp, LK_EXCLUSIVE);
1971 if (error != 0) {
1972 VOP_ABORTOP(dvp, cnp);
1973 vput(dvp);
1974 return error;
1975 }
1976 }
1977
1978 /*
1979 * Push all writes to the server, so that the attribute cache
1980 * doesn't get "out of sync" with the server.
1981 * XXX There should be a better way!
1982 */
1983 VOP_FSYNC(vp, cnp->cn_cred, FSYNC_WAIT, 0, 0, cnp->cn_proc);
1984
1985 v3 = NFS_ISV3(vp);
1986 nfsstats.rpccnt[NFSPROC_LINK]++;
1987 np = VTONFS(vp);
1988 nfsm_reqhead(np, NFSPROC_LINK,
1989 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1990 nfsm_fhtom(np, v3);
1991 nfsm_fhtom(VTONFS(dvp), v3);
1992 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1993 nfsm_request(np, NFSPROC_LINK, cnp->cn_proc, cnp->cn_cred);
1994 if (v3) {
1995 nfsm_postop_attr(vp, attrflag, 0);
1996 nfsm_wcc_data(dvp, wccflag, 0);
1997 }
1998 nfsm_reqdone;
1999 PNBUF_PUT(cnp->cn_pnbuf);
2000 VTONFS(dvp)->n_flag |= NMODIFIED;
2001 if (!attrflag)
2002 VTONFS(vp)->n_attrstamp = 0;
2003 if (!wccflag)
2004 VTONFS(dvp)->n_attrstamp = 0;
2005 if (dvp != vp)
2006 VOP_UNLOCK(vp, 0);
2007 VN_KNOTE(vp, NOTE_LINK);
2008 VN_KNOTE(dvp, NOTE_WRITE);
2009 vput(dvp);
2010 /*
2011 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2012 */
2013 if (error == EEXIST)
2014 error = 0;
2015 return (error);
2016 }
2017
2018 /*
2019 * nfs symbolic link create call
2020 */
2021 int
2022 nfs_symlink(v)
2023 void *v;
2024 {
2025 struct vop_symlink_args /* {
2026 struct vnode *a_dvp;
2027 struct vnode **a_vpp;
2028 struct componentname *a_cnp;
2029 struct vattr *a_vap;
2030 char *a_target;
2031 } */ *ap = v;
2032 struct vnode *dvp = ap->a_dvp;
2033 struct vattr *vap = ap->a_vap;
2034 struct componentname *cnp = ap->a_cnp;
2035 struct nfsv2_sattr *sp;
2036 u_int32_t *tl;
2037 caddr_t cp;
2038 int32_t t1, t2;
2039 caddr_t bpos, dpos, cp2;
2040 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2041 struct mbuf *mreq, *mrep, *md, *mb;
2042 struct vnode *newvp = (struct vnode *)0;
2043 const int v3 = NFS_ISV3(dvp);
2044 struct nfsnode *dnp = VTONFS(dvp);
2045
2046 *ap->a_vpp = NULL;
2047 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2048 slen = strlen(ap->a_target);
2049 nfsm_reqhead(dnp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2050 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
2051 nfsm_fhtom(dnp, v3);
2052 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2053 if (v3)
2054 nfsm_v3attrbuild(vap, FALSE);
2055 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2056 if (!v3) {
2057 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2058 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2059 sp->sa_uid = nfs_xdrneg1;
2060 sp->sa_gid = nfs_xdrneg1;
2061 sp->sa_size = nfs_xdrneg1;
2062 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2063 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2064 }
2065 nfsm_request(dnp, NFSPROC_SYMLINK, cnp->cn_proc, cnp->cn_cred);
2066 if (v3) {
2067 if (!error)
2068 nfsm_mtofh(dvp, newvp, v3, gotvp);
2069 nfsm_wcc_data(dvp, wccflag, 0);
2070 }
2071 nfsm_reqdone;
2072 /*
2073 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
2074 */
2075 if (error == EEXIST)
2076 error = 0;
2077 if (error == 0 && newvp == NULL) {
2078 struct nfsnode *np = NULL;
2079
2080 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2081 cnp->cn_cred, cnp->cn_proc, &np);
2082 if (error == 0)
2083 newvp = NFSTOV(np);
2084 }
2085 if (error) {
2086 if (newvp != NULL)
2087 vput(newvp);
2088 } else {
2089 *ap->a_vpp = newvp;
2090 }
2091 PNBUF_PUT(cnp->cn_pnbuf);
2092 VTONFS(dvp)->n_flag |= NMODIFIED;
2093 if (!wccflag)
2094 VTONFS(dvp)->n_attrstamp = 0;
2095 VN_KNOTE(dvp, NOTE_WRITE);
2096 vput(dvp);
2097 return (error);
2098 }
2099
2100 /*
2101 * nfs make dir call
2102 */
2103 int
2104 nfs_mkdir(v)
2105 void *v;
2106 {
2107 struct vop_mkdir_args /* {
2108 struct vnode *a_dvp;
2109 struct vnode **a_vpp;
2110 struct componentname *a_cnp;
2111 struct vattr *a_vap;
2112 } */ *ap = v;
2113 struct vnode *dvp = ap->a_dvp;
2114 struct vattr *vap = ap->a_vap;
2115 struct componentname *cnp = ap->a_cnp;
2116 struct nfsv2_sattr *sp;
2117 u_int32_t *tl;
2118 caddr_t cp;
2119 int32_t t1, t2;
2120 int len;
2121 struct nfsnode *dnp = VTONFS(dvp), *np = (struct nfsnode *)0;
2122 struct vnode *newvp = (struct vnode *)0;
2123 caddr_t bpos, dpos, cp2;
2124 int error = 0, wccflag = NFSV3_WCCRATTR;
2125 int gotvp = 0;
2126 struct mbuf *mreq, *mrep, *md, *mb;
2127 const int v3 = NFS_ISV3(dvp);
2128
2129 len = cnp->cn_namelen;
2130 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2131 nfsm_reqhead(dnp, NFSPROC_MKDIR,
2132 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
2133 nfsm_fhtom(dnp, v3);
2134 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2135 if (v3) {
2136 nfsm_v3attrbuild(vap, FALSE);
2137 } else {
2138 nfsm_build(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2139 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2140 sp->sa_uid = nfs_xdrneg1;
2141 sp->sa_gid = nfs_xdrneg1;
2142 sp->sa_size = nfs_xdrneg1;
2143 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2144 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2145 }
2146 nfsm_request(dnp, NFSPROC_MKDIR, cnp->cn_proc, cnp->cn_cred);
2147 if (!error)
2148 nfsm_mtofh(dvp, newvp, v3, gotvp);
2149 if (v3)
2150 nfsm_wcc_data(dvp, wccflag, 0);
2151 nfsm_reqdone;
2152 VTONFS(dvp)->n_flag |= NMODIFIED;
2153 if (!wccflag)
2154 VTONFS(dvp)->n_attrstamp = 0;
2155 /*
2156 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
2157 * if we can succeed in looking up the directory.
2158 */
2159 if (error == EEXIST || (!error && !gotvp)) {
2160 if (newvp) {
2161 vput(newvp);
2162 newvp = (struct vnode *)0;
2163 }
2164 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2165 cnp->cn_proc, &np);
2166 if (!error) {
2167 newvp = NFSTOV(np);
2168 if (newvp->v_type != VDIR)
2169 error = EEXIST;
2170 }
2171 }
2172 if (error) {
2173 if (newvp)
2174 vput(newvp);
2175 } else {
2176 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2177 if (cnp->cn_flags & MAKEENTRY)
2178 cache_enter(dvp, newvp, cnp);
2179 *ap->a_vpp = newvp;
2180 }
2181 PNBUF_PUT(cnp->cn_pnbuf);
2182 vput(dvp);
2183 return (error);
2184 }
2185
2186 /*
2187 * nfs remove directory call
2188 */
2189 int
2190 nfs_rmdir(v)
2191 void *v;
2192 {
2193 struct vop_rmdir_args /* {
2194 struct vnode *a_dvp;
2195 struct vnode *a_vp;
2196 struct componentname *a_cnp;
2197 } */ *ap = v;
2198 struct vnode *vp = ap->a_vp;
2199 struct vnode *dvp = ap->a_dvp;
2200 struct componentname *cnp = ap->a_cnp;
2201 u_int32_t *tl;
2202 caddr_t cp;
2203 int32_t t1, t2;
2204 caddr_t bpos, dpos, cp2;
2205 int error = 0, wccflag = NFSV3_WCCRATTR;
2206 struct mbuf *mreq, *mrep, *md, *mb;
2207 const int v3 = NFS_ISV3(dvp);
2208 struct nfsnode *dnp;
2209
2210 if (dvp == vp) {
2211 vrele(dvp);
2212 vput(dvp);
2213 PNBUF_PUT(cnp->cn_pnbuf);
2214 return (EINVAL);
2215 }
2216 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2217 dnp = VTONFS(dvp);
2218 nfsm_reqhead(dnp, NFSPROC_RMDIR,
2219 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
2220 nfsm_fhtom(dnp, v3);
2221 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2222 nfsm_request(dnp, NFSPROC_RMDIR, cnp->cn_proc, cnp->cn_cred);
2223 if (v3)
2224 nfsm_wcc_data(dvp, wccflag, 0);
2225 nfsm_reqdone;
2226 PNBUF_PUT(cnp->cn_pnbuf);
2227 VTONFS(dvp)->n_flag |= NMODIFIED;
2228 if (!wccflag)
2229 VTONFS(dvp)->n_attrstamp = 0;
2230 VN_KNOTE(dvp, NOTE_WRITE | NOTE_LINK);
2231 VN_KNOTE(vp, NOTE_DELETE);
2232 cache_purge(dvp);
2233 cache_purge(vp);
2234 vput(vp);
2235 vput(dvp);
2236 /*
2237 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2238 */
2239 if (error == ENOENT)
2240 error = 0;
2241 return (error);
2242 }
2243
2244 /*
2245 * nfs readdir call
2246 */
2247 int
2248 nfs_readdir(v)
2249 void *v;
2250 {
2251 struct vop_readdir_args /* {
2252 struct vnode *a_vp;
2253 struct uio *a_uio;
2254 struct ucred *a_cred;
2255 int *a_eofflag;
2256 off_t **a_cookies;
2257 int *a_ncookies;
2258 } */ *ap = v;
2259 struct vnode *vp = ap->a_vp;
2260 struct uio *uio = ap->a_uio;
2261 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2262 char *base = uio->uio_iov->iov_base;
2263 int tresid, error;
2264 size_t count, lost;
2265 struct dirent *dp;
2266 off_t *cookies = NULL;
2267 int ncookies = 0, nc;
2268
2269 if (vp->v_type != VDIR)
2270 return (EPERM);
2271
2272 lost = uio->uio_resid & (NFS_DIRFRAGSIZ - 1);
2273 count = uio->uio_resid - lost;
2274 if (count <= 0)
2275 return (EINVAL);
2276
2277 /*
2278 * Call nfs_bioread() to do the real work.
2279 */
2280 tresid = uio->uio_resid = count;
2281 error = nfs_bioread(vp, uio, 0, ap->a_cred,
2282 ap->a_cookies ? NFSBIO_CACHECOOKIES : 0);
2283
2284 if (!error && ap->a_cookies) {
2285 ncookies = count / 16;
2286 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
2287 *ap->a_cookies = cookies;
2288 }
2289
2290 if (!error && uio->uio_resid == tresid) {
2291 uio->uio_resid += lost;
2292 nfsstats.direofcache_misses++;
2293 if (ap->a_cookies)
2294 *ap->a_ncookies = 0;
2295 *ap->a_eofflag = 1;
2296 return (0);
2297 }
2298
2299 if (!error && ap->a_cookies) {
2300 /*
2301 * Only the NFS server and emulations use cookies, and they
2302 * load the directory block into system space, so we can
2303 * just look at it directly.
2304 */
2305 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2306 panic("nfs_readdir: lost in space");
2307 for (nc = 0; ncookies-- &&
2308 base < (char *)uio->uio_iov->iov_base; nc++){
2309 dp = (struct dirent *) base;
2310 if (dp->d_reclen == 0)
2311 break;
2312 if (nmp->nm_flag & NFSMNT_XLATECOOKIE)
2313 *(cookies++) = (off_t)NFS_GETCOOKIE32(dp);
2314 else
2315 *(cookies++) = NFS_GETCOOKIE(dp);
2316 base += dp->d_reclen;
2317 }
2318 uio->uio_resid +=
2319 ((caddr_t)uio->uio_iov->iov_base - base);
2320 uio->uio_iov->iov_len +=
2321 ((caddr_t)uio->uio_iov->iov_base - base);
2322 uio->uio_iov->iov_base = base;
2323 *ap->a_ncookies = nc;
2324 }
2325
2326 uio->uio_resid += lost;
2327 *ap->a_eofflag = 0;
2328 return (error);
2329 }
2330
2331 /*
2332 * Readdir rpc call.
2333 * Called from below the buffer cache by nfs_doio().
2334 */
2335 int
2336 nfs_readdirrpc(vp, uiop, cred)
2337 struct vnode *vp;
2338 struct uio *uiop;
2339 struct ucred *cred;
2340 {
2341 int len, left;
2342 struct dirent *dp = NULL;
2343 u_int32_t *tl;
2344 caddr_t cp;
2345 int32_t t1, t2;
2346 caddr_t bpos, dpos, cp2;
2347 struct mbuf *mreq, *mrep, *md, *mb;
2348 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2349 struct nfsnode *dnp = VTONFS(vp);
2350 u_quad_t fileno;
2351 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2352 int attrflag, nrpcs = 0, reclen;
2353 const int v3 = NFS_ISV3(vp);
2354
2355 #ifdef DIAGNOSTIC
2356 /*
2357 * Should be called from buffer cache, so only amount of
2358 * NFS_DIRBLKSIZ will be requested.
2359 */
2360 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2361 panic("nfs readdirrpc bad uio");
2362 #endif
2363
2364 /*
2365 * Loop around doing readdir rpc's of size nm_readdirsize
2366 * truncated to a multiple of NFS_DIRFRAGSIZ.
2367 * The stopping criteria is EOF or buffer full.
2368 */
2369 while (more_dirs && bigenough) {
2370 /*
2371 * Heuristic: don't bother to do another RPC to further
2372 * fill up this block if there is not much room left. (< 50%
2373 * of the readdir RPC size). This wastes some buffer space
2374 * but can save up to 50% in RPC calls.
2375 */
2376 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2377 bigenough = 0;
2378 break;
2379 }
2380 nfsstats.rpccnt[NFSPROC_READDIR]++;
2381 nfsm_reqhead(dnp, NFSPROC_READDIR, NFSX_FH(v3) +
2382 NFSX_READDIR(v3));
2383 nfsm_fhtom(dnp, v3);
2384 if (v3) {
2385 nfsm_build(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2386 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2387 txdr_swapcookie3(uiop->uio_offset, tl);
2388 } else {
2389 txdr_cookie3(uiop->uio_offset, tl);
2390 }
2391 tl += 2;
2392 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2393 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2394 } else {
2395 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2396 *tl++ = txdr_unsigned(uiop->uio_offset);
2397 }
2398 *tl = txdr_unsigned(nmp->nm_readdirsize);
2399 nfsm_request(dnp, NFSPROC_READDIR, uiop->uio_procp, cred);
2400 nrpcs++;
2401 if (v3) {
2402 nfsm_postop_attr(vp, attrflag, 0);
2403 if (!error) {
2404 nfsm_dissect(tl, u_int32_t *,
2405 2 * NFSX_UNSIGNED);
2406 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2407 dnp->n_cookieverf.nfsuquad[1] = *tl;
2408 } else {
2409 m_freem(mrep);
2410 goto nfsmout;
2411 }
2412 }
2413 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2414 more_dirs = fxdr_unsigned(int, *tl);
2415
2416 /* loop thru the dir entries, doctoring them to 4bsd form */
2417 while (more_dirs && bigenough) {
2418 if (v3) {
2419 nfsm_dissect(tl, u_int32_t *,
2420 3 * NFSX_UNSIGNED);
2421 fileno = fxdr_hyper(tl);
2422 len = fxdr_unsigned(int, *(tl + 2));
2423 } else {
2424 nfsm_dissect(tl, u_int32_t *,
2425 2 * NFSX_UNSIGNED);
2426 fileno = fxdr_unsigned(u_quad_t, *tl++);
2427 len = fxdr_unsigned(int, *tl);
2428 }
2429 if (len <= 0 || len > NFS_MAXNAMLEN) {
2430 error = EBADRPC;
2431 m_freem(mrep);
2432 goto nfsmout;
2433 }
2434 tlen = nfsm_rndup(len);
2435 if (tlen == len)
2436 tlen += 4; /* To ensure null termination */
2437 tlen += sizeof (off_t) + sizeof (int);
2438 reclen = ALIGN(tlen + DIRHDSIZ);
2439 tlen = reclen - DIRHDSIZ;
2440 left = NFS_DIRFRAGSIZ - blksiz;
2441 if (reclen > left) {
2442 dp->d_reclen += left;
2443 uiop->uio_iov->iov_base =
2444 (caddr_t)uiop->uio_iov->iov_base + left;
2445 uiop->uio_iov->iov_len -= left;
2446 uiop->uio_resid -= left;
2447 blksiz = 0;
2448 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2449 }
2450 if (reclen > uiop->uio_resid)
2451 bigenough = 0;
2452 if (bigenough) {
2453 dp = (struct dirent *)uiop->uio_iov->iov_base;
2454 dp->d_fileno = (int)fileno;
2455 dp->d_namlen = len;
2456 dp->d_reclen = reclen;
2457 dp->d_type = DT_UNKNOWN;
2458 blksiz += dp->d_reclen;
2459 if (blksiz == NFS_DIRFRAGSIZ)
2460 blksiz = 0;
2461 uiop->uio_resid -= DIRHDSIZ;
2462 uiop->uio_iov->iov_base =
2463 (caddr_t)uiop->uio_iov->iov_base + DIRHDSIZ;
2464 uiop->uio_iov->iov_len -= DIRHDSIZ;
2465 nfsm_mtouio(uiop, len);
2466 cp = uiop->uio_iov->iov_base;
2467 tlen -= len;
2468 *cp = '\0'; /* null terminate */
2469 uiop->uio_iov->iov_base =
2470 (caddr_t)uiop->uio_iov->iov_base + tlen;
2471 uiop->uio_iov->iov_len -= tlen;
2472 uiop->uio_resid -= tlen;
2473 } else
2474 nfsm_adv(nfsm_rndup(len));
2475 if (v3) {
2476 nfsm_dissect(tl, u_int32_t *,
2477 3 * NFSX_UNSIGNED);
2478 } else {
2479 nfsm_dissect(tl, u_int32_t *,
2480 2 * NFSX_UNSIGNED);
2481 }
2482 if (bigenough) {
2483 if (v3) {
2484 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2485 uiop->uio_offset =
2486 fxdr_swapcookie3(tl);
2487 else
2488 uiop->uio_offset =
2489 fxdr_cookie3(tl);
2490 }
2491 else {
2492 uiop->uio_offset =
2493 fxdr_unsigned(off_t, *tl);
2494 }
2495 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2496 }
2497 if (v3)
2498 tl += 2;
2499 else
2500 tl++;
2501 more_dirs = fxdr_unsigned(int, *tl);
2502 }
2503 /*
2504 * If at end of rpc data, get the eof boolean
2505 */
2506 if (!more_dirs) {
2507 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2508 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2509 }
2510 m_freem(mrep);
2511 }
2512 /*
2513 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2514 * by increasing d_reclen for the last record.
2515 */
2516 if (blksiz > 0) {
2517 left = NFS_DIRFRAGSIZ - blksiz;
2518 dp->d_reclen += left;
2519 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2520 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2521 left;
2522 uiop->uio_iov->iov_len -= left;
2523 uiop->uio_resid -= left;
2524 }
2525
2526 /*
2527 * We are now either at the end of the directory or have filled the
2528 * block.
2529 */
2530 if (bigenough)
2531 dnp->n_direofoffset = uiop->uio_offset;
2532 nfsmout:
2533 return (error);
2534 }
2535
2536 /*
2537 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2538 */
2539 int
2540 nfs_readdirplusrpc(vp, uiop, cred)
2541 struct vnode *vp;
2542 struct uio *uiop;
2543 struct ucred *cred;
2544 {
2545 int len, left;
2546 struct dirent *dp = NULL;
2547 u_int32_t *tl;
2548 caddr_t cp;
2549 int32_t t1, t2;
2550 struct vnode *newvp;
2551 caddr_t bpos, dpos, cp2;
2552 struct mbuf *mreq, *mrep, *md, *mb;
2553 struct nameidata nami, *ndp = &nami;
2554 struct componentname *cnp = &ndp->ni_cnd;
2555 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2556 struct nfsnode *dnp = VTONFS(vp), *np;
2557 nfsfh_t *fhp;
2558 u_quad_t fileno;
2559 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2560 int attrflag, fhsize, nrpcs = 0, reclen;
2561 struct nfs_fattr fattr, *fp;
2562
2563 #ifdef DIAGNOSTIC
2564 if (uiop->uio_iovcnt != 1 || (uiop->uio_resid & (NFS_DIRBLKSIZ - 1)))
2565 panic("nfs readdirplusrpc bad uio");
2566 #endif
2567 ndp->ni_dvp = vp;
2568 newvp = NULLVP;
2569
2570 /*
2571 * Loop around doing readdir rpc's of size nm_readdirsize
2572 * truncated to a multiple of NFS_DIRFRAGSIZ.
2573 * The stopping criteria is EOF or buffer full.
2574 */
2575 while (more_dirs && bigenough) {
2576 if (nrpcs > 0 && uiop->uio_resid < (nmp->nm_readdirsize / 2)) {
2577 bigenough = 0;
2578 break;
2579 }
2580 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2581 nfsm_reqhead(dnp, NFSPROC_READDIRPLUS,
2582 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2583 nfsm_fhtom(dnp, 1);
2584 nfsm_build(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2585 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE) {
2586 txdr_swapcookie3(uiop->uio_offset, tl);
2587 } else {
2588 txdr_cookie3(uiop->uio_offset, tl);
2589 }
2590 tl += 2;
2591 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2592 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2593 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2594 *tl = txdr_unsigned(nmp->nm_rsize);
2595 nfsm_request(dnp, NFSPROC_READDIRPLUS, uiop->uio_procp, cred);
2596 nfsm_postop_attr(vp, attrflag, 0);
2597 if (error) {
2598 m_freem(mrep);
2599 goto nfsmout;
2600 }
2601 nrpcs++;
2602 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2603 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2604 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2605 more_dirs = fxdr_unsigned(int, *tl);
2606
2607 /* loop thru the dir entries, doctoring them to 4bsd form */
2608 while (more_dirs && bigenough) {
2609 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2610 fileno = fxdr_hyper(tl);
2611 len = fxdr_unsigned(int, *(tl + 2));
2612 if (len <= 0 || len > NFS_MAXNAMLEN) {
2613 error = EBADRPC;
2614 m_freem(mrep);
2615 goto nfsmout;
2616 }
2617 tlen = nfsm_rndup(len);
2618 if (tlen == len)
2619 tlen += 4; /* To ensure null termination*/
2620 tlen += sizeof (off_t) + sizeof (int);
2621 reclen = ALIGN(tlen + DIRHDSIZ);
2622 tlen = reclen - DIRHDSIZ;
2623 left = NFS_DIRFRAGSIZ - blksiz;
2624 if (reclen > left) {
2625 /*
2626 * DIRFRAGSIZ is aligned, no need to align
2627 * again here.
2628 */
2629 dp->d_reclen += left;
2630 uiop->uio_iov->iov_base =
2631 (caddr_t)uiop->uio_iov->iov_base + left;
2632 uiop->uio_iov->iov_len -= left;
2633 uiop->uio_resid -= left;
2634 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2635 blksiz = 0;
2636 }
2637 if (reclen > uiop->uio_resid)
2638 bigenough = 0;
2639 if (bigenough) {
2640 dp = (struct dirent *)uiop->uio_iov->iov_base;
2641 dp->d_fileno = (int)fileno;
2642 dp->d_namlen = len;
2643 dp->d_reclen = reclen;
2644 dp->d_type = DT_UNKNOWN;
2645 blksiz += dp->d_reclen;
2646 if (blksiz == NFS_DIRFRAGSIZ)
2647 blksiz = 0;
2648 uiop->uio_resid -= DIRHDSIZ;
2649 uiop->uio_iov->iov_base =
2650 (caddr_t)uiop->uio_iov->iov_base +
2651 DIRHDSIZ;
2652 uiop->uio_iov->iov_len -= DIRHDSIZ;
2653 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2654 cnp->cn_namelen = len;
2655 nfsm_mtouio(uiop, len);
2656 cp = uiop->uio_iov->iov_base;
2657 tlen -= len;
2658 *cp = '\0';
2659 uiop->uio_iov->iov_base =
2660 (caddr_t)uiop->uio_iov->iov_base + tlen;
2661 uiop->uio_iov->iov_len -= tlen;
2662 uiop->uio_resid -= tlen;
2663 } else
2664 nfsm_adv(nfsm_rndup(len));
2665 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2666 if (bigenough) {
2667 if (nmp->nm_iflag & NFSMNT_SWAPCOOKIE)
2668 uiop->uio_offset =
2669 fxdr_swapcookie3(tl);
2670 else
2671 uiop->uio_offset =
2672 fxdr_cookie3(tl);
2673 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2674 }
2675 tl += 2;
2676
2677 /*
2678 * Since the attributes are before the file handle
2679 * (sigh), we must skip over the attributes and then
2680 * come back and get them.
2681 */
2682 attrflag = fxdr_unsigned(int, *tl);
2683 if (attrflag) {
2684 nfsm_dissect(fp, struct nfs_fattr *, NFSX_V3FATTR);
2685 memcpy(&fattr, fp, NFSX_V3FATTR);
2686 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2687 doit = fxdr_unsigned(int, *tl);
2688 if (doit) {
2689 nfsm_getfh(fhp, fhsize, 1);
2690 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2691 VREF(vp);
2692 newvp = vp;
2693 np = dnp;
2694 } else {
2695 error = nfs_nget(vp->v_mount, fhp,
2696 fhsize, &np);
2697 if (!error)
2698 newvp = NFSTOV(np);
2699 }
2700 if (!error) {
2701 const char *cp;
2702
2703 nfs_loadattrcache(&newvp, &fattr, 0, 0);
2704 dp->d_type =
2705 IFTODT(VTTOIF(np->n_vattr->va_type));
2706 ndp->ni_vp = newvp;
2707 cp = cnp->cn_nameptr + cnp->cn_namelen;
2708 cnp->cn_hash =
2709 namei_hash(cnp->cn_nameptr, &cp);
2710 if (cnp->cn_namelen <= NCHNAMLEN)
2711 cache_enter(ndp->ni_dvp, ndp->ni_vp,
2712 cnp);
2713 }
2714 }
2715 } else {
2716 /* Just skip over the file handle */
2717 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2718 i = fxdr_unsigned(int, *tl);
2719 nfsm_adv(nfsm_rndup(i));
2720 }
2721 if (newvp != NULLVP) {
2722 if (newvp == vp)
2723 vrele(newvp);
2724 else
2725 vput(newvp);
2726 newvp = NULLVP;
2727 }
2728 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2729 more_dirs = fxdr_unsigned(int, *tl);
2730 }
2731 /*
2732 * If at end of rpc data, get the eof boolean
2733 */
2734 if (!more_dirs) {
2735 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2736 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2737 }
2738 m_freem(mrep);
2739 }
2740 /*
2741 * Fill last record, iff any, out to a multiple of NFS_DIRFRAGSIZ
2742 * by increasing d_reclen for the last record.
2743 */
2744 if (blksiz > 0) {
2745 left = NFS_DIRFRAGSIZ - blksiz;
2746 dp->d_reclen += left;
2747 NFS_STASHCOOKIE(dp, uiop->uio_offset);
2748 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
2749 left;
2750 uiop->uio_iov->iov_len -= left;
2751 uiop->uio_resid -= left;
2752 }
2753
2754 /*
2755 * We are now either at the end of the directory or have filled the
2756 * block.
2757 */
2758 if (bigenough)
2759 dnp->n_direofoffset = uiop->uio_offset;
2760 nfsmout:
2761 if (newvp != NULLVP) {
2762 if(newvp == vp)
2763 vrele(newvp);
2764 else
2765 vput(newvp);
2766 }
2767 return (error);
2768 }
2769 static char hextoasc[] = "0123456789abcdef";
2770
2771 /*
2772 * Silly rename. To make the NFS filesystem that is stateless look a little
2773 * more like the "ufs" a remove of an active vnode is translated to a rename
2774 * to a funny looking filename that is removed by nfs_inactive on the
2775 * nfsnode. There is the potential for another process on a different client
2776 * to create the same funny name between the nfs_lookitup() fails and the
2777 * nfs_rename() completes, but...
2778 */
2779 int
2780 nfs_sillyrename(dvp, vp, cnp)
2781 struct vnode *dvp, *vp;
2782 struct componentname *cnp;
2783 {
2784 struct sillyrename *sp;
2785 struct nfsnode *np;
2786 int error;
2787 short pid;
2788
2789 cache_purge(dvp);
2790 np = VTONFS(vp);
2791 #ifndef DIAGNOSTIC
2792 if (vp->v_type == VDIR)
2793 panic("nfs: sillyrename dir");
2794 #endif
2795 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2796 M_NFSREQ, M_WAITOK);
2797 sp->s_cred = crdup(cnp->cn_cred);
2798 sp->s_dvp = dvp;
2799 VREF(dvp);
2800
2801 /* Fudge together a funny name */
2802 pid = cnp->cn_proc->p_pid;
2803 memcpy(sp->s_name, ".nfsAxxxx4.4", 13);
2804 sp->s_namlen = 12;
2805 sp->s_name[8] = hextoasc[pid & 0xf];
2806 sp->s_name[7] = hextoasc[(pid >> 4) & 0xf];
2807 sp->s_name[6] = hextoasc[(pid >> 8) & 0xf];
2808 sp->s_name[5] = hextoasc[(pid >> 12) & 0xf];
2809
2810 /* Try lookitups until we get one that isn't there */
2811 while (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2812 cnp->cn_proc, (struct nfsnode **)0) == 0) {
2813 sp->s_name[4]++;
2814 if (sp->s_name[4] > 'z') {
2815 error = EINVAL;
2816 goto bad;
2817 }
2818 }
2819 error = nfs_renameit(dvp, cnp, sp);
2820 if (error)
2821 goto bad;
2822 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2823 cnp->cn_proc, &np);
2824 np->n_sillyrename = sp;
2825 return (0);
2826 bad:
2827 vrele(sp->s_dvp);
2828 crfree(sp->s_cred);
2829 free((caddr_t)sp, M_NFSREQ);
2830 return (error);
2831 }
2832
2833 /*
2834 * Look up a file name and optionally either update the file handle or
2835 * allocate an nfsnode, depending on the value of npp.
2836 * npp == NULL --> just do the lookup
2837 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2838 * handled too
2839 * *npp != NULL --> update the file handle in the vnode
2840 */
2841 int
2842 nfs_lookitup(dvp, name, len, cred, procp, npp)
2843 struct vnode *dvp;
2844 const char *name;
2845 int len;
2846 struct ucred *cred;
2847 struct proc *procp;
2848 struct nfsnode **npp;
2849 {
2850 u_int32_t *tl;
2851 caddr_t cp;
2852 int32_t t1, t2;
2853 struct vnode *newvp = (struct vnode *)0;
2854 struct nfsnode *np, *dnp = VTONFS(dvp);
2855 caddr_t bpos, dpos, cp2;
2856 int error = 0, fhlen, attrflag;
2857 struct mbuf *mreq, *mrep, *md, *mb;
2858 nfsfh_t *nfhp;
2859 const int v3 = NFS_ISV3(dvp);
2860
2861 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2862 nfsm_reqhead(dnp, NFSPROC_LOOKUP,
2863 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2864 nfsm_fhtom(dnp, v3);
2865 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2866 nfsm_request(dnp, NFSPROC_LOOKUP, procp, cred);
2867 if (npp && !error) {
2868 nfsm_getfh(nfhp, fhlen, v3);
2869 if (*npp) {
2870 np = *npp;
2871 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2872 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2873 np->n_fhp = &np->n_fh;
2874 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2875 np->n_fhp =(nfsfh_t *)malloc(fhlen,M_NFSBIGFH,M_WAITOK);
2876 memcpy((caddr_t)np->n_fhp, (caddr_t)nfhp, fhlen);
2877 np->n_fhsize = fhlen;
2878 newvp = NFSTOV(np);
2879 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2880 VREF(dvp);
2881 newvp = dvp;
2882 } else {
2883 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
2884 if (error) {
2885 m_freem(mrep);
2886 return (error);
2887 }
2888 newvp = NFSTOV(np);
2889 }
2890 if (v3) {
2891 nfsm_postop_attr(newvp, attrflag, 0);
2892 if (!attrflag && *npp == NULL) {
2893 m_freem(mrep);
2894 vput(newvp);
2895 return (ENOENT);
2896 }
2897 } else
2898 nfsm_loadattr(newvp, (struct vattr *)0, 0);
2899 }
2900 nfsm_reqdone;
2901 if (npp && *npp == NULL) {
2902 if (error) {
2903 if (newvp)
2904 vput(newvp);
2905 } else
2906 *npp = np;
2907 }
2908 return (error);
2909 }
2910
2911 /*
2912 * Nfs Version 3 commit rpc
2913 */
2914 int
2915 nfs_commit(vp, offset, cnt, procp)
2916 struct vnode *vp;
2917 off_t offset;
2918 uint32_t cnt;
2919 struct proc *procp;
2920 {
2921 caddr_t cp;
2922 u_int32_t *tl;
2923 int32_t t1, t2;
2924 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2925 caddr_t bpos, dpos, cp2;
2926 int error = 0, wccflag = NFSV3_WCCRATTR;
2927 struct mbuf *mreq, *mrep, *md, *mb;
2928 struct nfsnode *np;
2929
2930 KASSERT(NFS_ISV3(vp));
2931
2932 #ifdef NFS_DEBUG_COMMIT
2933 printf("commit %lu - %lu\n", (unsigned long)offset,
2934 (unsigned long)(offset + cnt));
2935 #endif
2936
2937 simple_lock(&nmp->nm_slock);
2938 if ((nmp->nm_iflag & NFSMNT_HASWRITEVERF) == 0) {
2939 simple_unlock(&nmp->nm_slock);
2940 return (0);
2941 }
2942 simple_unlock(&nmp->nm_slock);
2943 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2944 np = VTONFS(vp);
2945 nfsm_reqhead(np, NFSPROC_COMMIT, NFSX_FH(1));
2946 nfsm_fhtom(np, 1);
2947 nfsm_build(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2948 txdr_hyper(offset, tl);
2949 tl += 2;
2950 *tl = txdr_unsigned(cnt);
2951 nfsm_request(np, NFSPROC_COMMIT, procp, VTONFS(vp)->n_wcred);
2952 nfsm_wcc_data(vp, wccflag, 0);
2953 if (!error) {
2954 nfsm_dissect(tl, u_int32_t *, NFSX_V3WRITEVERF);
2955 simple_lock(&nmp->nm_slock);
2956 if ((nmp->nm_iflag & NFSMNT_STALEWRITEVERF) ||
2957 memcmp(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF)) {
2958 memcpy(nmp->nm_writeverf, tl, NFSX_V3WRITEVERF);
2959 error = NFSERR_STALEWRITEVERF;
2960 nmp->nm_iflag |= NFSMNT_STALEWRITEVERF;
2961 }
2962 simple_unlock(&nmp->nm_slock);
2963 }
2964 nfsm_reqdone;
2965 return (error);
2966 }
2967
2968 /*
2969 * Kludge City..
2970 * - make nfs_bmap() essentially a no-op that does no translation
2971 * - do nfs_strategy() by doing I/O with nfs_readrpc/nfs_writerpc
2972 * (Maybe I could use the process's page mapping, but I was concerned that
2973 * Kernel Write might not be enabled and also figured copyout() would do
2974 * a lot more work than memcpy() and also it currently happens in the
2975 * context of the swapper process (2).
2976 */
2977 int
2978 nfs_bmap(v)
2979 void *v;
2980 {
2981 struct vop_bmap_args /* {
2982 struct vnode *a_vp;
2983 daddr_t a_bn;
2984 struct vnode **a_vpp;
2985 daddr_t *a_bnp;
2986 int *a_runp;
2987 } */ *ap = v;
2988 struct vnode *vp = ap->a_vp;
2989 int bshift = vp->v_mount->mnt_fs_bshift - vp->v_mount->mnt_dev_bshift;
2990
2991 if (ap->a_vpp != NULL)
2992 *ap->a_vpp = vp;
2993 if (ap->a_bnp != NULL)
2994 *ap->a_bnp = ap->a_bn << bshift;
2995 if (ap->a_runp != NULL)
2996 *ap->a_runp = 1024 * 1024; /* XXX */
2997 return (0);
2998 }
2999
3000 /*
3001 * Strategy routine.
3002 * For async requests when nfsiod(s) are running, queue the request by
3003 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
3004 * request.
3005 */
3006 int
3007 nfs_strategy(v)
3008 void *v;
3009 {
3010 struct vop_strategy_args *ap = v;
3011 struct buf *bp = ap->a_bp;
3012 struct proc *p;
3013 int error = 0;
3014
3015 if ((bp->b_flags & (B_PHYS|B_ASYNC)) == (B_PHYS|B_ASYNC))
3016 panic("nfs physio/async");
3017 if (bp->b_flags & B_ASYNC)
3018 p = NULL;
3019 else
3020 p = curproc; /* XXX */
3021
3022 /*
3023 * If the op is asynchronous and an i/o daemon is waiting
3024 * queue the request, wake it up and wait for completion
3025 * otherwise just do it ourselves.
3026 */
3027
3028 if ((bp->b_flags & B_ASYNC) == 0 ||
3029 nfs_asyncio(bp))
3030 error = nfs_doio(bp, p);
3031 return (error);
3032 }
3033
3034 /*
3035 * fsync vnode op. Just call nfs_flush() with commit == 1.
3036 */
3037 /* ARGSUSED */
3038 int
3039 nfs_fsync(v)
3040 void *v;
3041 {
3042 struct vop_fsync_args /* {
3043 struct vnodeop_desc *a_desc;
3044 struct vnode * a_vp;
3045 struct ucred * a_cred;
3046 int a_flags;
3047 off_t offlo;
3048 off_t offhi;
3049 struct proc * a_p;
3050 } */ *ap = v;
3051
3052 return (nfs_flush(ap->a_vp, ap->a_cred,
3053 (ap->a_flags & FSYNC_WAIT) != 0 ? MNT_WAIT : 0, ap->a_p, 1));
3054 }
3055
3056 /*
3057 * Flush all the data associated with a vnode.
3058 */
3059 int
3060 nfs_flush(vp, cred, waitfor, p, commit)
3061 struct vnode *vp;
3062 struct ucred *cred;
3063 int waitfor;
3064 struct proc *p;
3065 int commit;
3066 {
3067 struct nfsnode *np = VTONFS(vp);
3068 int error;
3069 int flushflags = PGO_ALLPAGES|PGO_CLEANIT|PGO_SYNCIO;
3070 UVMHIST_FUNC("nfs_flush"); UVMHIST_CALLED(ubchist);
3071
3072 simple_lock(&vp->v_interlock);
3073 error = VOP_PUTPAGES(vp, 0, 0, flushflags);
3074 if (np->n_flag & NWRITEERR) {
3075 error = np->n_error;
3076 np->n_flag &= ~NWRITEERR;
3077 }
3078 UVMHIST_LOG(ubchist, "returning %d", error,0,0,0);
3079 return (error);
3080 }
3081
3082 /*
3083 * Return POSIX pathconf information applicable to nfs.
3084 *
3085 * N.B. The NFS V2 protocol doesn't support this RPC.
3086 */
3087 /* ARGSUSED */
3088 int
3089 nfs_pathconf(v)
3090 void *v;
3091 {
3092 struct vop_pathconf_args /* {
3093 struct vnode *a_vp;
3094 int a_name;
3095 register_t *a_retval;
3096 } */ *ap = v;
3097 struct nfsv3_pathconf *pcp;
3098 struct vnode *vp = ap->a_vp;
3099 struct mbuf *mreq, *mrep, *md, *mb;
3100 int32_t t1, t2;
3101 u_int32_t *tl;
3102 caddr_t bpos, dpos, cp, cp2;
3103 int error = 0, attrflag;
3104 #ifndef NFS_V2_ONLY
3105 struct nfsmount *nmp;
3106 unsigned int l;
3107 u_int64_t maxsize;
3108 #endif
3109 const int v3 = NFS_ISV3(vp);
3110 struct nfsnode *np = VTONFS(vp);
3111
3112 switch (ap->a_name) {
3113 /* Names that can be resolved locally. */
3114 case _PC_PIPE_BUF:
3115 *ap->a_retval = PIPE_BUF;
3116 break;
3117 case _PC_SYNC_IO:
3118 *ap->a_retval = 1;
3119 break;
3120 /* Names that cannot be resolved locally; do an RPC, if possible. */
3121 case _PC_LINK_MAX:
3122 case _PC_NAME_MAX:
3123 case _PC_CHOWN_RESTRICTED:
3124 case _PC_NO_TRUNC:
3125 if (!v3) {
3126 error = EINVAL;
3127 break;
3128 }
3129 nfsstats.rpccnt[NFSPROC_PATHCONF]++;
3130 nfsm_reqhead(np, NFSPROC_PATHCONF, NFSX_FH(1));
3131 nfsm_fhtom(np, 1);
3132 nfsm_request(np, NFSPROC_PATHCONF,
3133 curproc, curproc->p_ucred); /* XXX */
3134 nfsm_postop_attr(vp, attrflag, 0);
3135 if (!error) {
3136 nfsm_dissect(pcp, struct nfsv3_pathconf *,
3137 NFSX_V3PATHCONF);
3138 switch (ap->a_name) {
3139 case _PC_LINK_MAX:
3140 *ap->a_retval =
3141 fxdr_unsigned(register_t, pcp->pc_linkmax);
3142 break;
3143 case _PC_NAME_MAX:
3144 *ap->a_retval =
3145 fxdr_unsigned(register_t, pcp->pc_namemax);
3146 break;
3147 case _PC_CHOWN_RESTRICTED:
3148 *ap->a_retval =
3149 (pcp->pc_chownrestricted == nfs_true);
3150 break;
3151 case _PC_NO_TRUNC:
3152 *ap->a_retval =
3153 (pcp->pc_notrunc == nfs_true);
3154 break;
3155 }
3156 }
3157 nfsm_reqdone;
3158 break;
3159 case _PC_FILESIZEBITS:
3160 #ifndef NFS_V2_ONLY
3161 if (v3) {
3162 nmp = VFSTONFS(vp->v_mount);
3163 if ((nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
3164 if ((error = nfs_fsinfo(nmp, vp,
3165 curproc->p_ucred, curproc)) != 0) /* XXX */
3166 break;
3167 for (l = 0, maxsize = nmp->nm_maxfilesize;
3168 (maxsize >> l) > 0; l++)
3169 ;
3170 *ap->a_retval = l + 1;
3171 } else
3172 #endif
3173 {
3174 *ap->a_retval = 32; /* NFS V2 limitation */
3175 }
3176 break;
3177 default:
3178 error = EINVAL;
3179 break;
3180 }
3181
3182 return (error);
3183 }
3184
3185 /*
3186 * NFS advisory byte-level locks.
3187 */
3188 int
3189 nfs_advlock(v)
3190 void *v;
3191 {
3192 struct vop_advlock_args /* {
3193 struct vnode *a_vp;
3194 caddr_t a_id;
3195 int a_op;
3196 struct flock *a_fl;
3197 int a_flags;
3198 } */ *ap = v;
3199 struct nfsnode *np = VTONFS(ap->a_vp);
3200
3201 return lf_advlock(ap, &np->n_lockf, np->n_size);
3202 }
3203
3204 /*
3205 * Print out the contents of an nfsnode.
3206 */
3207 int
3208 nfs_print(v)
3209 void *v;
3210 {
3211 struct vop_print_args /* {
3212 struct vnode *a_vp;
3213 } */ *ap = v;
3214 struct vnode *vp = ap->a_vp;
3215 struct nfsnode *np = VTONFS(vp);
3216
3217 printf("tag VT_NFS, fileid %ld fsid 0x%lx",
3218 np->n_vattr->va_fileid, np->n_vattr->va_fsid);
3219 if (vp->v_type == VFIFO)
3220 fifo_printinfo(vp);
3221 printf("\n");
3222 return (0);
3223 }
3224
3225 /*
3226 * NFS file truncation.
3227 */
3228 int
3229 nfs_truncate(v)
3230 void *v;
3231 {
3232 #if 0
3233 struct vop_truncate_args /* {
3234 struct vnode *a_vp;
3235 off_t a_length;
3236 int a_flags;
3237 struct ucred *a_cred;
3238 struct proc *a_p;
3239 } */ *ap = v;
3240 #endif
3241
3242 /* Use nfs_setattr */
3243 return (EOPNOTSUPP);
3244 }
3245
3246 /*
3247 * NFS update.
3248 */
3249 int
3250 nfs_update(v)
3251 void *v;
3252 #if 0
3253 struct vop_update_args /* {
3254 struct vnode *a_vp;
3255 struct timespec *a_ta;
3256 struct timespec *a_tm;
3257 int a_waitfor;
3258 } */ *ap = v;
3259 #endif
3260 {
3261
3262 /* Use nfs_setattr */
3263 return (EOPNOTSUPP);
3264 }
3265
3266 /*
3267 * Just call bwrite().
3268 */
3269 int
3270 nfs_bwrite(v)
3271 void *v;
3272 {
3273 struct vop_bwrite_args /* {
3274 struct vnode *a_bp;
3275 } */ *ap = v;
3276
3277 return (bwrite(ap->a_bp));
3278 }
3279
3280 /*
3281 * nfs unlock wrapper.
3282 */
3283 int
3284 nfs_unlock(void *v)
3285 {
3286 struct vop_unlock_args /* {
3287 struct vnode *a_vp;
3288 int a_flags;
3289 } */ *ap = v;
3290 struct vnode *vp = ap->a_vp;
3291
3292 /*
3293 * VOP_UNLOCK can be called by nfs_loadattrcache
3294 * with v_data == 0.
3295 */
3296 if (VTONFS(vp)) {
3297 nfs_delayedtruncate(vp);
3298 }
3299
3300 return genfs_unlock(v);
3301 }
3302
3303 /*
3304 * nfs special file access vnode op.
3305 * Essentially just get vattr and then imitate iaccess() since the device is
3306 * local to the client.
3307 */
3308 int
3309 nfsspec_access(v)
3310 void *v;
3311 {
3312 struct vop_access_args /* {
3313 struct vnode *a_vp;
3314 int a_mode;
3315 struct ucred *a_cred;
3316 struct proc *a_p;
3317 } */ *ap = v;
3318 struct vattr va;
3319 struct vnode *vp = ap->a_vp;
3320 int error;
3321
3322 error = VOP_GETATTR(vp, &va, ap->a_cred, ap->a_p);
3323 if (error)
3324 return (error);
3325
3326 /*
3327 * Disallow write attempts on filesystems mounted read-only;
3328 * unless the file is a socket, fifo, or a block or character
3329 * device resident on the filesystem.
3330 */
3331 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3332 switch (vp->v_type) {
3333 case VREG:
3334 case VDIR:
3335 case VLNK:
3336 return (EROFS);
3337 default:
3338 break;
3339 }
3340 }
3341
3342 return (vaccess(va.va_type, va.va_mode,
3343 va.va_uid, va.va_gid, ap->a_mode, ap->a_cred));
3344 }
3345
3346 /*
3347 * Read wrapper for special devices.
3348 */
3349 int
3350 nfsspec_read(v)
3351 void *v;
3352 {
3353 struct vop_read_args /* {
3354 struct vnode *a_vp;
3355 struct uio *a_uio;
3356 int a_ioflag;
3357 struct ucred *a_cred;
3358 } */ *ap = v;
3359 struct nfsnode *np = VTONFS(ap->a_vp);
3360
3361 /*
3362 * Set access flag.
3363 */
3364 np->n_flag |= NACC;
3365 np->n_atim.tv_sec = time.tv_sec;
3366 np->n_atim.tv_nsec = time.tv_usec * 1000;
3367 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_read), ap));
3368 }
3369
3370 /*
3371 * Write wrapper for special devices.
3372 */
3373 int
3374 nfsspec_write(v)
3375 void *v;
3376 {
3377 struct vop_write_args /* {
3378 struct vnode *a_vp;
3379 struct uio *a_uio;
3380 int a_ioflag;
3381 struct ucred *a_cred;
3382 } */ *ap = v;
3383 struct nfsnode *np = VTONFS(ap->a_vp);
3384
3385 /*
3386 * Set update flag.
3387 */
3388 np->n_flag |= NUPD;
3389 np->n_mtim.tv_sec = time.tv_sec;
3390 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3391 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_write), ap));
3392 }
3393
3394 /*
3395 * Close wrapper for special devices.
3396 *
3397 * Update the times on the nfsnode then do device close.
3398 */
3399 int
3400 nfsspec_close(v)
3401 void *v;
3402 {
3403 struct vop_close_args /* {
3404 struct vnode *a_vp;
3405 int a_fflag;
3406 struct ucred *a_cred;
3407 struct proc *a_p;
3408 } */ *ap = v;
3409 struct vnode *vp = ap->a_vp;
3410 struct nfsnode *np = VTONFS(vp);
3411 struct vattr vattr;
3412
3413 if (np->n_flag & (NACC | NUPD)) {
3414 np->n_flag |= NCHG;
3415 if (vp->v_usecount == 1 &&
3416 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3417 VATTR_NULL(&vattr);
3418 if (np->n_flag & NACC)
3419 vattr.va_atime = np->n_atim;
3420 if (np->n_flag & NUPD)
3421 vattr.va_mtime = np->n_mtim;
3422 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3423 }
3424 }
3425 return (VOCALL(spec_vnodeop_p, VOFFSET(vop_close), ap));
3426 }
3427
3428 /*
3429 * Read wrapper for fifos.
3430 */
3431 int
3432 nfsfifo_read(v)
3433 void *v;
3434 {
3435 struct vop_read_args /* {
3436 struct vnode *a_vp;
3437 struct uio *a_uio;
3438 int a_ioflag;
3439 struct ucred *a_cred;
3440 } */ *ap = v;
3441 struct nfsnode *np = VTONFS(ap->a_vp);
3442
3443 /*
3444 * Set access flag.
3445 */
3446 np->n_flag |= NACC;
3447 np->n_atim.tv_sec = time.tv_sec;
3448 np->n_atim.tv_nsec = time.tv_usec * 1000;
3449 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), ap));
3450 }
3451
3452 /*
3453 * Write wrapper for fifos.
3454 */
3455 int
3456 nfsfifo_write(v)
3457 void *v;
3458 {
3459 struct vop_write_args /* {
3460 struct vnode *a_vp;
3461 struct uio *a_uio;
3462 int a_ioflag;
3463 struct ucred *a_cred;
3464 } */ *ap = v;
3465 struct nfsnode *np = VTONFS(ap->a_vp);
3466
3467 /*
3468 * Set update flag.
3469 */
3470 np->n_flag |= NUPD;
3471 np->n_mtim.tv_sec = time.tv_sec;
3472 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3473 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), ap));
3474 }
3475
3476 /*
3477 * Close wrapper for fifos.
3478 *
3479 * Update the times on the nfsnode then do fifo close.
3480 */
3481 int
3482 nfsfifo_close(v)
3483 void *v;
3484 {
3485 struct vop_close_args /* {
3486 struct vnode *a_vp;
3487 int a_fflag;
3488 struct ucred *a_cred;
3489 struct proc *a_p;
3490 } */ *ap = v;
3491 struct vnode *vp = ap->a_vp;
3492 struct nfsnode *np = VTONFS(vp);
3493 struct vattr vattr;
3494
3495 if (np->n_flag & (NACC | NUPD)) {
3496 if (np->n_flag & NACC) {
3497 np->n_atim.tv_sec = time.tv_sec;
3498 np->n_atim.tv_nsec = time.tv_usec * 1000;
3499 }
3500 if (np->n_flag & NUPD) {
3501 np->n_mtim.tv_sec = time.tv_sec;
3502 np->n_mtim.tv_nsec = time.tv_usec * 1000;
3503 }
3504 np->n_flag |= NCHG;
3505 if (vp->v_usecount == 1 &&
3506 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3507 VATTR_NULL(&vattr);
3508 if (np->n_flag & NACC)
3509 vattr.va_atime = np->n_atim;
3510 if (np->n_flag & NUPD)
3511 vattr.va_mtime = np->n_mtim;
3512 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_p);
3513 }
3514 }
3515 return (VOCALL(fifo_vnodeop_p, VOFFSET(vop_close), ap));
3516 }
3517