1 /* $NetBSD: nfs_subs.c,v 1.242 2022/02/09 21:50:24 andvar 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_subs.c 8.8 (Berkeley) 5/22/95
35 */
36
37 /*
38 * Copyright 2000 Wasabi Systems, Inc.
39 * All rights reserved.
40 *
41 * Written by Frank van der Linden for Wasabi Systems, Inc.
42 *
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
45 * are met:
46 * 1. Redistributions of source code must retain the above copyright
47 * notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 * notice, this list of conditions and the following disclaimer in the
50 * documentation and/or other materials provided with the distribution.
51 * 3. All advertising materials mentioning features or use of this software
52 * must display the following acknowledgement:
53 * This product includes software developed for the NetBSD Project by
54 * Wasabi Systems, Inc.
55 * 4. The name of Wasabi Systems, Inc. may not be used to endorse
56 * or promote products derived from this software without specific prior
57 * written permission.
58 *
59 * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
61 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
62 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL WASABI SYSTEMS, INC
63 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
64 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
65 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
66 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
67 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
68 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
69 * POSSIBILITY OF SUCH DAMAGE.
70 */
71
72 #include <sys/cdefs.h>
73 __KERNEL_RCSID(0, "$NetBSD: nfs_subs.c,v 1.242 2022/02/09 21:50:24 andvar Exp $");
74
75 #ifdef _KERNEL_OPT
76 #include "opt_nfs.h"
77 #endif
78
79 /*
80 * These functions support the macros and help fiddle mbuf chains for
81 * the nfs op functions. They do things like create the rpc header and
82 * copy data between mbuf chains and uio lists.
83 */
84 #include <sys/param.h>
85 #include <sys/proc.h>
86 #include <sys/systm.h>
87 #include <sys/kernel.h>
88 #include <sys/kmem.h>
89 #include <sys/mount.h>
90 #include <sys/vnode.h>
91 #include <sys/namei.h>
92 #include <sys/mbuf.h>
93 #include <sys/socket.h>
94 #include <sys/stat.h>
95 #include <sys/filedesc.h>
96 #include <sys/time.h>
97 #include <sys/dirent.h>
98 #include <sys/once.h>
99 #include <sys/kauth.h>
100 #include <sys/atomic.h>
101 #include <sys/cprng.h>
102
103 #include <uvm/uvm_page.h>
104 #include <uvm/uvm_page_array.h>
105
106 #include <nfs/rpcv2.h>
107 #include <nfs/nfsproto.h>
108 #include <nfs/nfsnode.h>
109 #include <nfs/nfs.h>
110 #include <nfs/xdr_subs.h>
111 #include <nfs/nfsm_subs.h>
112 #include <nfs/nfsmount.h>
113 #include <nfs/nfsrtt.h>
114 #include <nfs/nfs_var.h>
115
116 #include <miscfs/specfs/specdev.h>
117
118 #include <netinet/in.h>
119
120 static u_int32_t nfs_xid;
121
122 int nuidhash_max = NFS_MAXUIDHASH;
123 /*
124 * Data items converted to xdr at startup, since they are constant
125 * This is kinda hokey, but may save a little time doing byte swaps
126 */
127 u_int32_t nfs_xdrneg1;
128 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
129 rpc_mismatch, rpc_auth_unix, rpc_msgaccepted,
130 rpc_auth_kerb;
131 u_int32_t nfs_prog, nfs_true, nfs_false;
132
133 /* And other global data */
134 const nfstype nfsv2_type[9] =
135 { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON };
136 const nfstype nfsv3_type[9] =
137 { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON };
138 const enum vtype nv2tov_type[8] =
139 { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON };
140 const enum vtype nv3tov_type[8] =
141 { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO };
142 int nfs_ticks;
143
144 /* NFS client/server stats. */
145 struct nfsstats nfsstats;
146
147 /*
148 * Mapping of old NFS Version 2 RPC numbers to generic numbers.
149 */
150 const int nfsv3_procid[NFS_NPROCS] = {
151 NFSPROC_NULL,
152 NFSPROC_GETATTR,
153 NFSPROC_SETATTR,
154 NFSPROC_NOOP,
155 NFSPROC_LOOKUP,
156 NFSPROC_READLINK,
157 NFSPROC_READ,
158 NFSPROC_NOOP,
159 NFSPROC_WRITE,
160 NFSPROC_CREATE,
161 NFSPROC_REMOVE,
162 NFSPROC_RENAME,
163 NFSPROC_LINK,
164 NFSPROC_SYMLINK,
165 NFSPROC_MKDIR,
166 NFSPROC_RMDIR,
167 NFSPROC_READDIR,
168 NFSPROC_FSSTAT,
169 NFSPROC_NOOP,
170 NFSPROC_NOOP,
171 NFSPROC_NOOP,
172 NFSPROC_NOOP,
173 NFSPROC_NOOP
174 };
175
176 /*
177 * and the reverse mapping from generic to Version 2 procedure numbers
178 */
179 const int nfsv2_procid[NFS_NPROCS] = {
180 NFSV2PROC_NULL,
181 NFSV2PROC_GETATTR,
182 NFSV2PROC_SETATTR,
183 NFSV2PROC_LOOKUP,
184 NFSV2PROC_NOOP,
185 NFSV2PROC_READLINK,
186 NFSV2PROC_READ,
187 NFSV2PROC_WRITE,
188 NFSV2PROC_CREATE,
189 NFSV2PROC_MKDIR,
190 NFSV2PROC_SYMLINK,
191 NFSV2PROC_CREATE,
192 NFSV2PROC_REMOVE,
193 NFSV2PROC_RMDIR,
194 NFSV2PROC_RENAME,
195 NFSV2PROC_LINK,
196 NFSV2PROC_READDIR,
197 NFSV2PROC_NOOP,
198 NFSV2PROC_STATFS,
199 NFSV2PROC_NOOP,
200 NFSV2PROC_NOOP,
201 NFSV2PROC_NOOP,
202 NFSV2PROC_NOOP,
203 };
204
205 /*
206 * Maps errno values to nfs error numbers.
207 * Use NFSERR_IO as the catch all for ones not specifically defined in
208 * RFC 1094.
209 */
210 static const u_char nfsrv_v2errmap[] = {
211 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO,
212 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
213 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO,
214 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR,
215 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
216 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS,
217 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
218 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
219 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
220 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
221 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
222 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
223 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO,
224 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE,
225 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
226 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
227 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
228 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
229 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO,
230 NFSERR_IO, NFSERR_IO, NFSERR_IO
231 };
232 __CTASSERT(__arraycount(nfsrv_v2errmap) == ELAST);
233
234 /*
235 * Maps errno values to nfs error numbers.
236 * Although it is not obvious whether or not NFS clients really care if
237 * a returned error value is in the specified list for the procedure, the
238 * safest thing to do is filter them appropriately. For Version 2, the
239 * X/Open XNFS document is the only specification that defines error values
240 * for each RPC (The RFC simply lists all possible error values for all RPCs),
241 * so I have decided to not do this for Version 2.
242 * The first entry is the default error return and the rest are the valid
243 * errors for that RPC in increasing numeric order.
244 */
245 static const short nfsv3err_null[] = {
246 0,
247 0,
248 };
249
250 static const short nfsv3err_getattr[] = {
251 NFSERR_IO,
252 NFSERR_IO,
253 NFSERR_STALE,
254 NFSERR_BADHANDLE,
255 NFSERR_SERVERFAULT,
256 0,
257 };
258
259 static const short nfsv3err_setattr[] = {
260 NFSERR_IO,
261 NFSERR_PERM,
262 NFSERR_IO,
263 NFSERR_ACCES,
264 NFSERR_INVAL,
265 NFSERR_NOSPC,
266 NFSERR_ROFS,
267 NFSERR_DQUOT,
268 NFSERR_STALE,
269 NFSERR_BADHANDLE,
270 NFSERR_NOT_SYNC,
271 NFSERR_SERVERFAULT,
272 0,
273 };
274
275 static const short nfsv3err_lookup[] = {
276 NFSERR_IO,
277 NFSERR_NOENT,
278 NFSERR_IO,
279 NFSERR_ACCES,
280 NFSERR_NOTDIR,
281 NFSERR_NAMETOL,
282 NFSERR_STALE,
283 NFSERR_BADHANDLE,
284 NFSERR_SERVERFAULT,
285 0,
286 };
287
288 static const short nfsv3err_access[] = {
289 NFSERR_IO,
290 NFSERR_IO,
291 NFSERR_STALE,
292 NFSERR_BADHANDLE,
293 NFSERR_SERVERFAULT,
294 0,
295 };
296
297 static const short nfsv3err_readlink[] = {
298 NFSERR_IO,
299 NFSERR_IO,
300 NFSERR_ACCES,
301 NFSERR_INVAL,
302 NFSERR_STALE,
303 NFSERR_BADHANDLE,
304 NFSERR_NOTSUPP,
305 NFSERR_SERVERFAULT,
306 0,
307 };
308
309 static const short nfsv3err_read[] = {
310 NFSERR_IO,
311 NFSERR_IO,
312 NFSERR_NXIO,
313 NFSERR_ACCES,
314 NFSERR_INVAL,
315 NFSERR_STALE,
316 NFSERR_BADHANDLE,
317 NFSERR_SERVERFAULT,
318 NFSERR_JUKEBOX,
319 0,
320 };
321
322 static const short nfsv3err_write[] = {
323 NFSERR_IO,
324 NFSERR_IO,
325 NFSERR_ACCES,
326 NFSERR_INVAL,
327 NFSERR_FBIG,
328 NFSERR_NOSPC,
329 NFSERR_ROFS,
330 NFSERR_DQUOT,
331 NFSERR_STALE,
332 NFSERR_BADHANDLE,
333 NFSERR_SERVERFAULT,
334 NFSERR_JUKEBOX,
335 0,
336 };
337
338 static const short nfsv3err_create[] = {
339 NFSERR_IO,
340 NFSERR_IO,
341 NFSERR_ACCES,
342 NFSERR_EXIST,
343 NFSERR_NOTDIR,
344 NFSERR_NOSPC,
345 NFSERR_ROFS,
346 NFSERR_NAMETOL,
347 NFSERR_DQUOT,
348 NFSERR_STALE,
349 NFSERR_BADHANDLE,
350 NFSERR_NOTSUPP,
351 NFSERR_SERVERFAULT,
352 0,
353 };
354
355 static const short nfsv3err_mkdir[] = {
356 NFSERR_IO,
357 NFSERR_IO,
358 NFSERR_ACCES,
359 NFSERR_EXIST,
360 NFSERR_NOTDIR,
361 NFSERR_NOSPC,
362 NFSERR_ROFS,
363 NFSERR_NAMETOL,
364 NFSERR_DQUOT,
365 NFSERR_STALE,
366 NFSERR_BADHANDLE,
367 NFSERR_NOTSUPP,
368 NFSERR_SERVERFAULT,
369 0,
370 };
371
372 static const short nfsv3err_symlink[] = {
373 NFSERR_IO,
374 NFSERR_IO,
375 NFSERR_ACCES,
376 NFSERR_EXIST,
377 NFSERR_NOTDIR,
378 NFSERR_NOSPC,
379 NFSERR_ROFS,
380 NFSERR_NAMETOL,
381 NFSERR_DQUOT,
382 NFSERR_STALE,
383 NFSERR_BADHANDLE,
384 NFSERR_NOTSUPP,
385 NFSERR_SERVERFAULT,
386 0,
387 };
388
389 static const short nfsv3err_mknod[] = {
390 NFSERR_IO,
391 NFSERR_IO,
392 NFSERR_ACCES,
393 NFSERR_EXIST,
394 NFSERR_NOTDIR,
395 NFSERR_NOSPC,
396 NFSERR_ROFS,
397 NFSERR_NAMETOL,
398 NFSERR_DQUOT,
399 NFSERR_STALE,
400 NFSERR_BADHANDLE,
401 NFSERR_NOTSUPP,
402 NFSERR_SERVERFAULT,
403 NFSERR_BADTYPE,
404 0,
405 };
406
407 static const short nfsv3err_remove[] = {
408 NFSERR_IO,
409 NFSERR_NOENT,
410 NFSERR_IO,
411 NFSERR_ACCES,
412 NFSERR_NOTDIR,
413 NFSERR_ROFS,
414 NFSERR_NAMETOL,
415 NFSERR_STALE,
416 NFSERR_BADHANDLE,
417 NFSERR_SERVERFAULT,
418 0,
419 };
420
421 static const short nfsv3err_rmdir[] = {
422 NFSERR_IO,
423 NFSERR_NOENT,
424 NFSERR_IO,
425 NFSERR_ACCES,
426 NFSERR_EXIST,
427 NFSERR_NOTDIR,
428 NFSERR_INVAL,
429 NFSERR_ROFS,
430 NFSERR_NAMETOL,
431 NFSERR_NOTEMPTY,
432 NFSERR_STALE,
433 NFSERR_BADHANDLE,
434 NFSERR_NOTSUPP,
435 NFSERR_SERVERFAULT,
436 0,
437 };
438
439 static const short nfsv3err_rename[] = {
440 NFSERR_IO,
441 NFSERR_NOENT,
442 NFSERR_IO,
443 NFSERR_ACCES,
444 NFSERR_EXIST,
445 NFSERR_XDEV,
446 NFSERR_NOTDIR,
447 NFSERR_ISDIR,
448 NFSERR_INVAL,
449 NFSERR_NOSPC,
450 NFSERR_ROFS,
451 NFSERR_MLINK,
452 NFSERR_NAMETOL,
453 NFSERR_NOTEMPTY,
454 NFSERR_DQUOT,
455 NFSERR_STALE,
456 NFSERR_BADHANDLE,
457 NFSERR_NOTSUPP,
458 NFSERR_SERVERFAULT,
459 0,
460 };
461
462 static const short nfsv3err_link[] = {
463 NFSERR_IO,
464 NFSERR_IO,
465 NFSERR_ACCES,
466 NFSERR_EXIST,
467 NFSERR_XDEV,
468 NFSERR_NOTDIR,
469 NFSERR_INVAL,
470 NFSERR_NOSPC,
471 NFSERR_ROFS,
472 NFSERR_MLINK,
473 NFSERR_NAMETOL,
474 NFSERR_DQUOT,
475 NFSERR_STALE,
476 NFSERR_BADHANDLE,
477 NFSERR_NOTSUPP,
478 NFSERR_SERVERFAULT,
479 0,
480 };
481
482 static const short nfsv3err_readdir[] = {
483 NFSERR_IO,
484 NFSERR_IO,
485 NFSERR_ACCES,
486 NFSERR_NOTDIR,
487 NFSERR_STALE,
488 NFSERR_BADHANDLE,
489 NFSERR_BAD_COOKIE,
490 NFSERR_TOOSMALL,
491 NFSERR_SERVERFAULT,
492 0,
493 };
494
495 static const short nfsv3err_readdirplus[] = {
496 NFSERR_IO,
497 NFSERR_IO,
498 NFSERR_ACCES,
499 NFSERR_NOTDIR,
500 NFSERR_STALE,
501 NFSERR_BADHANDLE,
502 NFSERR_BAD_COOKIE,
503 NFSERR_NOTSUPP,
504 NFSERR_TOOSMALL,
505 NFSERR_SERVERFAULT,
506 0,
507 };
508
509 static const short nfsv3err_fsstat[] = {
510 NFSERR_IO,
511 NFSERR_IO,
512 NFSERR_STALE,
513 NFSERR_BADHANDLE,
514 NFSERR_SERVERFAULT,
515 0,
516 };
517
518 static const short nfsv3err_fsinfo[] = {
519 NFSERR_STALE,
520 NFSERR_STALE,
521 NFSERR_BADHANDLE,
522 NFSERR_SERVERFAULT,
523 0,
524 };
525
526 static const short nfsv3err_pathconf[] = {
527 NFSERR_STALE,
528 NFSERR_STALE,
529 NFSERR_BADHANDLE,
530 NFSERR_SERVERFAULT,
531 0,
532 };
533
534 static const short nfsv3err_commit[] = {
535 NFSERR_IO,
536 NFSERR_IO,
537 NFSERR_STALE,
538 NFSERR_BADHANDLE,
539 NFSERR_SERVERFAULT,
540 0,
541 };
542
543 static const short * const nfsrv_v3errmap[] = {
544 nfsv3err_null,
545 nfsv3err_getattr,
546 nfsv3err_setattr,
547 nfsv3err_lookup,
548 nfsv3err_access,
549 nfsv3err_readlink,
550 nfsv3err_read,
551 nfsv3err_write,
552 nfsv3err_create,
553 nfsv3err_mkdir,
554 nfsv3err_symlink,
555 nfsv3err_mknod,
556 nfsv3err_remove,
557 nfsv3err_rmdir,
558 nfsv3err_rename,
559 nfsv3err_link,
560 nfsv3err_readdir,
561 nfsv3err_readdirplus,
562 nfsv3err_fsstat,
563 nfsv3err_fsinfo,
564 nfsv3err_pathconf,
565 nfsv3err_commit,
566 };
567
568 extern struct nfsrtt nfsrtt;
569
570 u_long nfsdirhashmask;
571
572 int nfs_webnamei(struct nameidata *, struct vnode *, struct proc *);
573
574 /*
575 * Create the header for an rpc request packet
576 * The hsiz is the size of the rest of the nfs request header.
577 * (just used to decide if a cluster is a good idea)
578 */
579 struct mbuf *
nfsm_reqh(struct nfsnode * np,u_long procid,int hsiz,char ** bposp)580 nfsm_reqh(struct nfsnode *np, u_long procid, int hsiz, char **bposp)
581 {
582 struct mbuf *mb;
583 char *bpos;
584
585 mb = m_get(M_WAIT, MT_DATA);
586 MCLAIM(mb, &nfs_mowner);
587 if (hsiz >= MINCLSIZE)
588 m_clget(mb, M_WAIT);
589 mb->m_len = 0;
590 bpos = mtod(mb, void *);
591
592 /* Finally, return values */
593 *bposp = bpos;
594 return (mb);
595 }
596
597 /*
598 * Build the RPC header and fill in the authorization info.
599 * The authorization string argument is only used when the credentials
600 * come from outside of the kernel.
601 * Returns the head of the mbuf list.
602 */
603 struct mbuf *
nfsm_rpchead(kauth_cred_t cr,int nmflag,int procid,int auth_type,int auth_len,char * auth_str,int verf_len,char * verf_str,struct mbuf * mrest,int mrest_len,struct mbuf ** mbp,uint32_t * xidp)604 nfsm_rpchead(kauth_cred_t cr, int nmflag, int procid,
605 int auth_type, int auth_len, char *auth_str, int verf_len,
606 char *verf_str, struct mbuf *mrest, int mrest_len,
607 struct mbuf **mbp, uint32_t *xidp)
608 {
609 struct mbuf *mb;
610 u_int32_t *tl;
611 char *bpos;
612 int i;
613 struct mbuf *mreq;
614 int siz, grpsiz, authsiz;
615
616 authsiz = nfsm_rndup(auth_len);
617 mb = m_gethdr(M_WAIT, MT_DATA);
618 MCLAIM(mb, &nfs_mowner);
619 if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
620 m_clget(mb, M_WAIT);
621 } else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
622 m_align(mb, authsiz + 10 * NFSX_UNSIGNED);
623 } else {
624 m_align(mb, 8 * NFSX_UNSIGNED);
625 }
626 mb->m_len = 0;
627 mreq = mb;
628 bpos = mtod(mb, void *);
629
630 /*
631 * First the RPC header.
632 */
633 nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
634
635 *tl++ = *xidp = nfs_getxid();
636 *tl++ = rpc_call;
637 *tl++ = rpc_vers;
638 *tl++ = txdr_unsigned(NFS_PROG);
639 if (nmflag & NFSMNT_NFSV3)
640 *tl++ = txdr_unsigned(NFS_VER3);
641 else
642 *tl++ = txdr_unsigned(NFS_VER2);
643 if (nmflag & NFSMNT_NFSV3)
644 *tl++ = txdr_unsigned(procid);
645 else
646 *tl++ = txdr_unsigned(nfsv2_procid[procid]);
647
648 /*
649 * And then the authorization cred.
650 */
651 *tl++ = txdr_unsigned(auth_type);
652 *tl = txdr_unsigned(authsiz);
653 switch (auth_type) {
654 case RPCAUTH_UNIX:
655 nfsm_build(tl, u_int32_t *, auth_len);
656 *tl++ = 0; /* stamp ?? */
657 *tl++ = 0; /* NULL hostname */
658 *tl++ = txdr_unsigned(kauth_cred_geteuid(cr));
659 *tl++ = txdr_unsigned(kauth_cred_getegid(cr));
660 grpsiz = (auth_len >> 2) - 5;
661 *tl++ = txdr_unsigned(grpsiz);
662 for (i = 0; i < grpsiz; i++)
663 *tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */
664 break;
665 case RPCAUTH_KERB4:
666 siz = auth_len;
667 while (siz > 0) {
668 if (M_TRAILINGSPACE(mb) == 0) {
669 struct mbuf *mb2;
670 mb2 = m_get(M_WAIT, MT_DATA);
671 MCLAIM(mb2, &nfs_mowner);
672 if (siz >= MINCLSIZE)
673 m_clget(mb2, M_WAIT);
674 mb->m_next = mb2;
675 mb = mb2;
676 mb->m_len = 0;
677 bpos = mtod(mb, void *);
678 }
679 i = uimin(siz, M_TRAILINGSPACE(mb));
680 memcpy(bpos, auth_str, i);
681 mb->m_len += i;
682 auth_str += i;
683 bpos += i;
684 siz -= i;
685 }
686 if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
687 for (i = 0; i < siz; i++)
688 *bpos++ = '\0';
689 mb->m_len += siz;
690 }
691 break;
692 };
693
694 /*
695 * And the verifier...
696 */
697 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
698 if (verf_str) {
699 *tl++ = txdr_unsigned(RPCAUTH_KERB4);
700 *tl = txdr_unsigned(verf_len);
701 siz = verf_len;
702 while (siz > 0) {
703 if (M_TRAILINGSPACE(mb) == 0) {
704 struct mbuf *mb2;
705 mb2 = m_get(M_WAIT, MT_DATA);
706 MCLAIM(mb2, &nfs_mowner);
707 if (siz >= MINCLSIZE)
708 m_clget(mb2, M_WAIT);
709 mb->m_next = mb2;
710 mb = mb2;
711 mb->m_len = 0;
712 bpos = mtod(mb, void *);
713 }
714 i = uimin(siz, M_TRAILINGSPACE(mb));
715 memcpy(bpos, verf_str, i);
716 mb->m_len += i;
717 verf_str += i;
718 bpos += i;
719 siz -= i;
720 }
721 if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
722 for (i = 0; i < siz; i++)
723 *bpos++ = '\0';
724 mb->m_len += siz;
725 }
726 } else {
727 *tl++ = txdr_unsigned(RPCAUTH_NULL);
728 *tl = 0;
729 }
730 mb->m_next = mrest;
731 mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
732 m_reset_rcvif(mreq);
733 *mbp = mb;
734 return (mreq);
735 }
736
737 /*
738 * copies mbuf chain to the uio scatter/gather list
739 */
740 int
nfsm_mbuftouio(struct mbuf ** mrep,struct uio * uiop,int siz,char ** dpos)741 nfsm_mbuftouio(struct mbuf **mrep, struct uio *uiop, int siz, char **dpos)
742 {
743 char *mbufcp, *uiocp;
744 int xfer, left, len;
745 struct mbuf *mp;
746 long uiosiz, rem;
747 int error = 0;
748
749 mp = *mrep;
750 mbufcp = *dpos;
751 len = mtod(mp, char *) + mp->m_len - mbufcp;
752 rem = nfsm_rndup(siz)-siz;
753 while (siz > 0) {
754 if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
755 return (EFBIG);
756 left = uiop->uio_iov->iov_len;
757 uiocp = uiop->uio_iov->iov_base;
758 if (left > siz)
759 left = siz;
760 uiosiz = left;
761 while (left > 0) {
762 while (len == 0) {
763 mp = mp->m_next;
764 if (mp == NULL)
765 return (EBADRPC);
766 mbufcp = mtod(mp, void *);
767 len = mp->m_len;
768 }
769 xfer = (left > len) ? len : left;
770 error = copyout_vmspace(uiop->uio_vmspace, mbufcp,
771 uiocp, xfer);
772 if (error) {
773 return error;
774 }
775 left -= xfer;
776 len -= xfer;
777 mbufcp += xfer;
778 uiocp += xfer;
779 uiop->uio_offset += xfer;
780 uiop->uio_resid -= xfer;
781 }
782 if (uiop->uio_iov->iov_len <= siz) {
783 uiop->uio_iovcnt--;
784 uiop->uio_iov++;
785 } else {
786 uiop->uio_iov->iov_base =
787 (char *)uiop->uio_iov->iov_base + uiosiz;
788 uiop->uio_iov->iov_len -= uiosiz;
789 }
790 siz -= uiosiz;
791 }
792 *dpos = mbufcp;
793 *mrep = mp;
794 if (rem > 0) {
795 if (len < rem)
796 error = nfs_adv(mrep, dpos, rem, len);
797 else
798 *dpos += rem;
799 }
800 return (error);
801 }
802
803 /*
804 * copies a uio scatter/gather list to an mbuf chain.
805 * NOTE: can only handle iovcnt == 1
806 */
807 int
nfsm_uiotombuf(struct uio * uiop,struct mbuf ** mq,int siz,char ** bpos)808 nfsm_uiotombuf(struct uio *uiop, struct mbuf **mq, int siz, char **bpos)
809 {
810 char *uiocp;
811 struct mbuf *mp, *mp2;
812 int xfer, left, mlen;
813 int uiosiz, clflg, rem;
814 char *cp;
815 int error;
816
817 #ifdef DIAGNOSTIC
818 if (uiop->uio_iovcnt != 1)
819 panic("nfsm_uiotombuf: iovcnt != 1");
820 #endif
821
822 if (siz > MLEN) /* or should it >= MCLBYTES ?? */
823 clflg = 1;
824 else
825 clflg = 0;
826 rem = nfsm_rndup(siz)-siz;
827 mp = mp2 = *mq;
828 while (siz > 0) {
829 left = uiop->uio_iov->iov_len;
830 uiocp = uiop->uio_iov->iov_base;
831 if (left > siz)
832 left = siz;
833 uiosiz = left;
834 while (left > 0) {
835 mlen = M_TRAILINGSPACE(mp);
836 if (mlen == 0) {
837 mp = m_get(M_WAIT, MT_DATA);
838 MCLAIM(mp, &nfs_mowner);
839 if (clflg)
840 m_clget(mp, M_WAIT);
841 mp->m_len = 0;
842 mp2->m_next = mp;
843 mp2 = mp;
844 mlen = M_TRAILINGSPACE(mp);
845 }
846 xfer = (left > mlen) ? mlen : left;
847 cp = mtod(mp, char *) + mp->m_len;
848 error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp,
849 xfer);
850 if (error) {
851 /* XXX */
852 }
853 mp->m_len += xfer;
854 left -= xfer;
855 uiocp += xfer;
856 uiop->uio_offset += xfer;
857 uiop->uio_resid -= xfer;
858 }
859 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base +
860 uiosiz;
861 uiop->uio_iov->iov_len -= uiosiz;
862 siz -= uiosiz;
863 }
864 if (rem > 0) {
865 if (rem > M_TRAILINGSPACE(mp)) {
866 mp = m_get(M_WAIT, MT_DATA);
867 MCLAIM(mp, &nfs_mowner);
868 mp->m_len = 0;
869 mp2->m_next = mp;
870 }
871 cp = mtod(mp, char *) + mp->m_len;
872 for (left = 0; left < rem; left++)
873 *cp++ = '\0';
874 mp->m_len += rem;
875 *bpos = cp;
876 } else
877 *bpos = mtod(mp, char *) + mp->m_len;
878 *mq = mp;
879 return (0);
880 }
881
882 /*
883 * Get at least "siz" bytes of correctly aligned data.
884 * When called the mbuf pointers are not necessarily correct,
885 * dsosp points to what ought to be in m_data and left contains
886 * what ought to be in m_len.
887 * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
888 * cases. (The macros use the vars. dpos and dpos2)
889 */
890 int
nfsm_disct(struct mbuf ** mdp,char ** dposp,int siz,int left,char ** cp2)891 nfsm_disct(struct mbuf **mdp, char **dposp, int siz, int left, char **cp2)
892 {
893 struct mbuf *m1, *m2;
894 struct mbuf *havebuf = NULL;
895 char *src = *dposp;
896 char *dst;
897 int len;
898
899 #ifdef DEBUG
900 if (left < 0)
901 panic("nfsm_disct: left < 0");
902 #endif
903 m1 = *mdp;
904 /*
905 * Skip through the mbuf chain looking for an mbuf with
906 * some data. If the first mbuf found has enough data
907 * and it is correctly aligned return it.
908 */
909 while (left == 0) {
910 havebuf = m1;
911 *mdp = m1 = m1->m_next;
912 if (m1 == NULL)
913 return (EBADRPC);
914 src = mtod(m1, void *);
915 left = m1->m_len;
916 /*
917 * If we start a new mbuf and it is big enough
918 * and correctly aligned just return it, don't
919 * do any pull up.
920 */
921 if (left >= siz && nfsm_aligned(src)) {
922 *cp2 = src;
923 *dposp = src + siz;
924 return (0);
925 }
926 }
927 if ((m1->m_flags & M_EXT) != 0) {
928 if (havebuf && M_TRAILINGSPACE(havebuf) >= siz &&
929 nfsm_aligned(mtod(havebuf, char *) + havebuf->m_len)) {
930 /*
931 * If the first mbuf with data has external data
932 * and there is a previous mbuf with some trailing
933 * space, use it to move the data into.
934 */
935 m2 = m1;
936 *mdp = m1 = havebuf;
937 *cp2 = mtod(m1, char *) + m1->m_len;
938 } else if (havebuf) {
939 /*
940 * If the first mbuf has a external data
941 * and there is no previous empty mbuf
942 * allocate a new mbuf and move the external
943 * data to the new mbuf. Also make the first
944 * mbuf look empty.
945 */
946 m2 = m1;
947 *mdp = m1 = m_get(M_WAIT, MT_DATA);
948 MCLAIM(m1, m2->m_owner);
949 if ((m2->m_flags & M_PKTHDR) != 0) {
950 m_move_pkthdr(m1, m2);
951 }
952 if (havebuf) {
953 havebuf->m_next = m1;
954 }
955 m1->m_next = m2;
956 MRESETDATA(m1);
957 m1->m_len = 0;
958 m2->m_data = src;
959 m2->m_len = left;
960 *cp2 = mtod(m1, char *);
961 } else {
962 struct mbuf **nextp = &m1->m_next;
963
964 m1->m_len -= left;
965 do {
966 m2 = m_get(M_WAIT, MT_DATA);
967 MCLAIM(m2, m1->m_owner);
968 if (left >= MINCLSIZE) {
969 MCLGET(m2, M_WAIT);
970 }
971 m2->m_next = *nextp;
972 *nextp = m2;
973 nextp = &m2->m_next;
974 len = (m2->m_flags & M_EXT) != 0 ?
975 MCLBYTES : MLEN;
976 if (len > left) {
977 len = left;
978 }
979 memcpy(mtod(m2, char *), src, len);
980 m2->m_len = len;
981 src += len;
982 left -= len;
983 } while (left > 0);
984 *mdp = m1 = m1->m_next;
985 m2 = m1->m_next;
986 *cp2 = mtod(m1, char *);
987 }
988 } else {
989 /*
990 * If the first mbuf has no external data
991 * move the data to the front of the mbuf.
992 */
993 MRESETDATA(m1);
994 dst = mtod(m1, char *);
995 if (dst != src) {
996 memmove(dst, src, left);
997 }
998 m1->m_len = left;
999 m2 = m1->m_next;
1000 *cp2 = m1->m_data;
1001 }
1002 *dposp = *cp2 + siz;
1003 /*
1004 * Loop through mbufs pulling data up into first mbuf until
1005 * the first mbuf is full or there is no more data to
1006 * pullup.
1007 */
1008 dst = mtod(m1, char *) + m1->m_len;
1009 while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) {
1010 if ((len = uimin(len, m2->m_len)) != 0) {
1011 memcpy(dst, mtod(m2, char *), len);
1012 }
1013 m1->m_len += len;
1014 dst += len;
1015 m2->m_data += len;
1016 m2->m_len -= len;
1017 m2 = m2->m_next;
1018 }
1019 if (m1->m_len < siz)
1020 return (EBADRPC);
1021 return (0);
1022 }
1023
1024 /*
1025 * Advance the position in the mbuf chain.
1026 */
1027 int
nfs_adv(struct mbuf ** mdp,char ** dposp,int offs,int left)1028 nfs_adv(struct mbuf **mdp, char **dposp, int offs, int left)
1029 {
1030 struct mbuf *m;
1031 int s;
1032
1033 m = *mdp;
1034 s = left;
1035 while (s < offs) {
1036 offs -= s;
1037 m = m->m_next;
1038 if (m == NULL)
1039 return (EBADRPC);
1040 s = m->m_len;
1041 }
1042 *mdp = m;
1043 *dposp = mtod(m, char *) + offs;
1044 return (0);
1045 }
1046
1047 /*
1048 * Copy a string into mbufs for the hard cases...
1049 */
1050 int
nfsm_strtmbuf(struct mbuf ** mb,char ** bpos,const char * cp,long siz)1051 nfsm_strtmbuf(struct mbuf **mb, char **bpos, const char *cp, long siz)
1052 {
1053 struct mbuf *m1 = NULL, *m2;
1054 long left, xfer, len, tlen;
1055 u_int32_t *tl;
1056 int putsize;
1057
1058 putsize = 1;
1059 m2 = *mb;
1060 left = M_TRAILINGSPACE(m2);
1061 if (left > 0) {
1062 tl = ((u_int32_t *)(*bpos));
1063 *tl++ = txdr_unsigned(siz);
1064 putsize = 0;
1065 left -= NFSX_UNSIGNED;
1066 m2->m_len += NFSX_UNSIGNED;
1067 if (left > 0) {
1068 memcpy((void *) tl, cp, left);
1069 siz -= left;
1070 cp += left;
1071 m2->m_len += left;
1072 left = 0;
1073 }
1074 }
1075 /* Loop around adding mbufs */
1076 while (siz > 0) {
1077 m1 = m_get(M_WAIT, MT_DATA);
1078 MCLAIM(m1, &nfs_mowner);
1079 if (siz > MLEN)
1080 m_clget(m1, M_WAIT);
1081 m1->m_len = NFSMSIZ(m1);
1082 m2->m_next = m1;
1083 m2 = m1;
1084 tl = mtod(m1, u_int32_t *);
1085 tlen = 0;
1086 if (putsize) {
1087 *tl++ = txdr_unsigned(siz);
1088 m1->m_len -= NFSX_UNSIGNED;
1089 tlen = NFSX_UNSIGNED;
1090 putsize = 0;
1091 }
1092 if (siz < m1->m_len) {
1093 len = nfsm_rndup(siz);
1094 xfer = siz;
1095 if (xfer < len)
1096 *(tl+(xfer>>2)) = 0;
1097 } else {
1098 xfer = len = m1->m_len;
1099 }
1100 memcpy((void *) tl, cp, xfer);
1101 m1->m_len = len+tlen;
1102 siz -= xfer;
1103 cp += xfer;
1104 }
1105 *mb = m1;
1106 *bpos = mtod(m1, char *) + m1->m_len;
1107 return (0);
1108 }
1109
1110 /*
1111 * Directory caching routines. They work as follows:
1112 * - a cache is maintained per VDIR nfsnode.
1113 * - for each offset cookie that is exported to userspace, and can
1114 * thus be thrown back at us as an offset to VOP_READDIR, store
1115 * information in the cache.
1116 * - cached are:
1117 * - cookie itself
1118 * - blocknumber (essentially just a search key in the buffer cache)
1119 * - entry number in block.
1120 * - offset cookie of block in which this entry is stored
1121 * - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1122 * - entries are looked up in a hash table
1123 * - also maintained is an LRU list of entries, used to determine
1124 * which ones to delete if the cache grows too large.
1125 * - if 32 <-> 64 translation mode is requested for a filesystem,
1126 * the cache also functions as a translation table
1127 * - in the translation case, invalidating the cache does not mean
1128 * flushing it, but just marking entries as invalid, except for
1129 * the <64bit cookie, 32bitcookie> pair which is still valid, to
1130 * still be able to use the cache as a translation table.
1131 * - 32 bit cookies are uniquely created by combining the hash table
1132 * entry value, and one generation count per hash table entry,
1133 * incremented each time an entry is appended to the chain.
1134 * - the cache is invalidated each time a direcory is modified
1135 * - sanity checks are also done; if an entry in a block turns
1136 * out not to have a matching cookie, the cache is invalidated
1137 * and a new block starting from the wanted offset is fetched from
1138 * the server.
1139 * - directory entries as read from the server are extended to contain
1140 * the 64bit and, optionally, the 32bit cookies, for sanity checking
1141 * the cache and exporting them to userspace through the cookie
1142 * argument to VOP_READDIR.
1143 */
1144
1145 u_long
nfs_dirhash(off_t off)1146 nfs_dirhash(off_t off)
1147 {
1148 int i;
1149 char *cp = (char *)&off;
1150 u_long sum = 0L;
1151
1152 for (i = 0 ; i < sizeof (off); i++)
1153 sum += *cp++;
1154
1155 return sum;
1156 }
1157
1158 #define _NFSDC_MTX(np) (NFSTOV(np)->v_interlock)
1159 #define NFSDC_LOCK(np) mutex_enter(_NFSDC_MTX(np))
1160 #define NFSDC_UNLOCK(np) mutex_exit(_NFSDC_MTX(np))
1161 #define NFSDC_ASSERT_LOCKED(np) KASSERT(mutex_owned(_NFSDC_MTX(np)))
1162
1163 void
nfs_initdircache(struct vnode * vp)1164 nfs_initdircache(struct vnode *vp)
1165 {
1166 struct nfsnode *np = VTONFS(vp);
1167 struct nfsdirhashhead *dircache;
1168
1169 dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, true,
1170 &nfsdirhashmask);
1171
1172 NFSDC_LOCK(np);
1173 if (np->n_dircache == NULL) {
1174 np->n_dircachesize = 0;
1175 np->n_dircache = dircache;
1176 dircache = NULL;
1177 TAILQ_INIT(&np->n_dirchain);
1178 }
1179 NFSDC_UNLOCK(np);
1180 if (dircache)
1181 hashdone(dircache, HASH_LIST, nfsdirhashmask);
1182 }
1183
1184 void
nfs_initdirxlatecookie(struct vnode * vp)1185 nfs_initdirxlatecookie(struct vnode *vp)
1186 {
1187 struct nfsnode *np = VTONFS(vp);
1188 unsigned *dirgens;
1189
1190 KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE);
1191
1192 dirgens = kmem_zalloc(NFS_DIRHASHSIZ * sizeof(unsigned), KM_SLEEP);
1193 NFSDC_LOCK(np);
1194 if (np->n_dirgens == NULL) {
1195 np->n_dirgens = dirgens;
1196 dirgens = NULL;
1197 }
1198 NFSDC_UNLOCK(np);
1199 if (dirgens)
1200 kmem_free(dirgens, NFS_DIRHASHSIZ * sizeof(unsigned));
1201 }
1202
1203 static const struct nfsdircache dzero;
1204
1205 static void nfs_unlinkdircache(struct nfsnode *np, struct nfsdircache *);
1206 static void nfs_putdircache_unlocked(struct nfsnode *,
1207 struct nfsdircache *);
1208
1209 static void
nfs_unlinkdircache(struct nfsnode * np,struct nfsdircache * ndp)1210 nfs_unlinkdircache(struct nfsnode *np, struct nfsdircache *ndp)
1211 {
1212
1213 NFSDC_ASSERT_LOCKED(np);
1214 KASSERT(ndp != &dzero);
1215
1216 if (LIST_NEXT(ndp, dc_hash) == (void *)-1)
1217 return;
1218
1219 TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1220 LIST_REMOVE(ndp, dc_hash);
1221 LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */
1222
1223 nfs_putdircache_unlocked(np, ndp);
1224 }
1225
1226 void
nfs_putdircache(struct nfsnode * np,struct nfsdircache * ndp)1227 nfs_putdircache(struct nfsnode *np, struct nfsdircache *ndp)
1228 {
1229 int ref;
1230
1231 if (ndp == &dzero)
1232 return;
1233
1234 KASSERT(ndp->dc_refcnt > 0);
1235 NFSDC_LOCK(np);
1236 ref = --ndp->dc_refcnt;
1237 NFSDC_UNLOCK(np);
1238
1239 if (ref == 0)
1240 kmem_free(ndp, sizeof(*ndp));
1241 }
1242
1243 static void
nfs_putdircache_unlocked(struct nfsnode * np,struct nfsdircache * ndp)1244 nfs_putdircache_unlocked(struct nfsnode *np, struct nfsdircache *ndp)
1245 {
1246 int ref;
1247
1248 NFSDC_ASSERT_LOCKED(np);
1249
1250 if (ndp == &dzero)
1251 return;
1252
1253 KASSERT(ndp->dc_refcnt > 0);
1254 ref = --ndp->dc_refcnt;
1255 if (ref == 0)
1256 kmem_free(ndp, sizeof(*ndp));
1257 }
1258
1259 struct nfsdircache *
nfs_searchdircache(struct vnode * vp,off_t off,int do32,int * hashent)1260 nfs_searchdircache(struct vnode *vp, off_t off, int do32, int *hashent)
1261 {
1262 struct nfsdirhashhead *ndhp;
1263 struct nfsdircache *ndp = NULL;
1264 struct nfsnode *np = VTONFS(vp);
1265 unsigned ent;
1266
1267 /*
1268 * Zero is always a valid cookie.
1269 */
1270 if (off == 0)
1271 /* XXXUNCONST */
1272 return (struct nfsdircache *)__UNCONST(&dzero);
1273
1274 if (!np->n_dircache)
1275 return NULL;
1276
1277 /*
1278 * We use a 32bit cookie as search key, directly reconstruct
1279 * the hashentry. Else use the hashfunction.
1280 */
1281 if (do32) {
1282 ent = (u_int32_t)off >> 24;
1283 if (ent >= NFS_DIRHASHSIZ)
1284 return NULL;
1285 ndhp = &np->n_dircache[ent];
1286 } else {
1287 ndhp = NFSDIRHASH(np, off);
1288 }
1289
1290 if (hashent)
1291 *hashent = (int)(ndhp - np->n_dircache);
1292
1293 NFSDC_LOCK(np);
1294 if (do32) {
1295 LIST_FOREACH(ndp, ndhp, dc_hash) {
1296 if (ndp->dc_cookie32 == (u_int32_t)off) {
1297 /*
1298 * An invalidated entry will become the
1299 * start of a new block fetched from
1300 * the server.
1301 */
1302 if (ndp->dc_flags & NFSDC_INVALID) {
1303 ndp->dc_blkcookie = ndp->dc_cookie;
1304 ndp->dc_entry = 0;
1305 ndp->dc_flags &= ~NFSDC_INVALID;
1306 }
1307 break;
1308 }
1309 }
1310 } else {
1311 LIST_FOREACH(ndp, ndhp, dc_hash) {
1312 if (ndp->dc_cookie == off)
1313 break;
1314 }
1315 }
1316 if (ndp != NULL)
1317 ndp->dc_refcnt++;
1318 NFSDC_UNLOCK(np);
1319 return ndp;
1320 }
1321
1322
1323 struct nfsdircache *
nfs_enterdircache(struct vnode * vp,off_t off,off_t blkoff,int en,daddr_t blkno)1324 nfs_enterdircache(struct vnode *vp, off_t off, off_t blkoff, int en,
1325 daddr_t blkno)
1326 {
1327 struct nfsnode *np = VTONFS(vp);
1328 struct nfsdirhashhead *ndhp;
1329 struct nfsdircache *ndp = NULL;
1330 struct nfsdircache *newndp = NULL;
1331 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1332 int hashent = 0, gen, overwrite; /* XXX: GCC */
1333
1334 /*
1335 * XXX refuse entries for offset 0. amd(8) erroneously sets
1336 * cookie 0 for the '.' entry, making this necessary. This
1337 * isn't so bad, as 0 is a special case anyway.
1338 */
1339 if (off == 0)
1340 /* XXXUNCONST */
1341 return (struct nfsdircache *)__UNCONST(&dzero);
1342
1343 if (!np->n_dircache)
1344 /*
1345 * XXX would like to do this in nfs_nget but vtype
1346 * isn't known at that time.
1347 */
1348 nfs_initdircache(vp);
1349
1350 if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens)
1351 nfs_initdirxlatecookie(vp);
1352
1353 retry:
1354 ndp = nfs_searchdircache(vp, off, 0, &hashent);
1355
1356 NFSDC_LOCK(np);
1357 if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) {
1358 /*
1359 * Overwriting an old entry. Check if it's the same.
1360 * If so, just return. If not, remove the old entry.
1361 */
1362 if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1363 goto done;
1364 nfs_unlinkdircache(np, ndp);
1365 nfs_putdircache_unlocked(np, ndp);
1366 ndp = NULL;
1367 }
1368
1369 ndhp = &np->n_dircache[hashent];
1370
1371 if (!ndp) {
1372 if (newndp == NULL) {
1373 NFSDC_UNLOCK(np);
1374 newndp = kmem_alloc(sizeof(*newndp), KM_SLEEP);
1375 newndp->dc_refcnt = 1;
1376 LIST_NEXT(newndp, dc_hash) = (void *)-1;
1377 goto retry;
1378 }
1379 ndp = newndp;
1380 newndp = NULL;
1381 overwrite = 0;
1382 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1383 /*
1384 * We're allocating a new entry, so bump the
1385 * generation number.
1386 */
1387 KASSERT(np->n_dirgens);
1388 gen = ++np->n_dirgens[hashent];
1389 if (gen == 0) {
1390 np->n_dirgens[hashent]++;
1391 gen++;
1392 }
1393 ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1394 }
1395 } else
1396 overwrite = 1;
1397
1398 ndp->dc_cookie = off;
1399 ndp->dc_blkcookie = blkoff;
1400 ndp->dc_entry = en;
1401 ndp->dc_flags = 0;
1402
1403 if (overwrite)
1404 goto done;
1405
1406 /*
1407 * If the maximum directory cookie cache size has been reached
1408 * for this node, take one off the front. The idea is that
1409 * directories are typically read front-to-back once, so that
1410 * the oldest entries can be thrown away without much performance
1411 * loss.
1412 */
1413 if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1414 nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain));
1415 } else
1416 np->n_dircachesize++;
1417
1418 KASSERT(ndp->dc_refcnt == 1);
1419 LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1420 TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1421 ndp->dc_refcnt++;
1422 done:
1423 KASSERT(ndp->dc_refcnt > 0);
1424 NFSDC_UNLOCK(np);
1425 if (newndp)
1426 nfs_putdircache(np, newndp);
1427 return ndp;
1428 }
1429
1430 void
nfs_invaldircache(struct vnode * vp,int flags)1431 nfs_invaldircache(struct vnode *vp, int flags)
1432 {
1433 struct nfsnode *np = VTONFS(vp);
1434 struct nfsdircache *ndp = NULL;
1435 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1436 const bool forcefree = flags & NFS_INVALDIRCACHE_FORCE;
1437
1438 #ifdef DIAGNOSTIC
1439 if (vp->v_type != VDIR)
1440 panic("nfs: invaldircache: not dir");
1441 #endif
1442
1443 if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0)
1444 np->n_flag &= ~NEOFVALID;
1445
1446 if (!np->n_dircache)
1447 return;
1448
1449 NFSDC_LOCK(np);
1450 if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1451 while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) {
1452 KASSERT(!forcefree || ndp->dc_refcnt == 1);
1453 nfs_unlinkdircache(np, ndp);
1454 }
1455 np->n_dircachesize = 0;
1456 if (forcefree && np->n_dirgens) {
1457 kmem_free(np->n_dirgens,
1458 NFS_DIRHASHSIZ * sizeof(unsigned));
1459 np->n_dirgens = NULL;
1460 }
1461 } else {
1462 TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain)
1463 ndp->dc_flags |= NFSDC_INVALID;
1464 }
1465
1466 NFSDC_UNLOCK(np);
1467 }
1468
1469 /*
1470 * Called once before VFS init to initialize shared and
1471 * server-specific data structures.
1472 */
1473 static int
nfs_init0(void)1474 nfs_init0(void)
1475 {
1476
1477 nfsrtt.pos = 0;
1478 rpc_vers = txdr_unsigned(RPC_VER2);
1479 rpc_call = txdr_unsigned(RPC_CALL);
1480 rpc_reply = txdr_unsigned(RPC_REPLY);
1481 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1482 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1483 rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1484 rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1485 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1486 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1487 nfs_prog = txdr_unsigned(NFS_PROG);
1488 nfs_true = txdr_unsigned(true);
1489 nfs_false = txdr_unsigned(false);
1490 nfs_xdrneg1 = txdr_unsigned(-1);
1491 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1492 if (nfs_ticks < 1)
1493 nfs_ticks = 1;
1494 nfsdreq_init();
1495
1496 /*
1497 * Initialize reply list and start timer
1498 */
1499 TAILQ_INIT(&nfs_reqq);
1500 mutex_init(&nfs_reqq_lock, MUTEX_DEFAULT, IPL_NONE);
1501 nfs_timer_init();
1502 MOWNER_ATTACH(&nfs_mowner);
1503
1504 return 0;
1505 }
1506
1507 static volatile uint32_t nfs_mutex;
1508 static uint32_t nfs_refcount;
1509
1510 #define nfs_p() while (atomic_cas_32(&nfs_mutex, 0, 1) == 0) continue;
1511 #define nfs_v() while (atomic_cas_32(&nfs_mutex, 1, 0) == 1) continue;
1512
1513 /*
1514 * This is disgusting, but it must support both modular and monolothic
1515 * configurations, plus the code is shared between server and client.
1516 * For monolithic builds NFSSERVER may not imply NFS. Unfortunately we
1517 * can't use regular mutexes here that would require static initialization
1518 * and we can get initialized from multiple places, so we improvise.
1519 *
1520 * Yuck.
1521 */
1522 void
nfs_init(void)1523 nfs_init(void)
1524 {
1525
1526 nfs_p();
1527 if (nfs_refcount++ == 0)
1528 nfs_init0();
1529 nfs_v();
1530 }
1531
1532 void
nfs_fini(void)1533 nfs_fini(void)
1534 {
1535
1536 nfs_p();
1537 if (--nfs_refcount == 0) {
1538 MOWNER_DETACH(&nfs_mowner);
1539 nfs_timer_fini();
1540 mutex_destroy(&nfs_reqq_lock);
1541 nfsdreq_fini();
1542 }
1543 nfs_v();
1544 }
1545
1546 /*
1547 * A fiddled version of m_adj() that ensures null fill to a 32-bit
1548 * boundary and only trims off the back end
1549 *
1550 * 1. trim off 'len' bytes as m_adj(mp, -len).
1551 * 2. add zero-padding 'nul' bytes at the end of the mbuf chain.
1552 */
1553 void
nfs_zeropad(struct mbuf * mp,int len,int nul)1554 nfs_zeropad(struct mbuf *mp, int len, int nul)
1555 {
1556 struct mbuf *m;
1557 int count;
1558
1559 /*
1560 * Trim from tail. Scan the mbuf chain,
1561 * calculating its length and finding the last mbuf.
1562 * If the adjustment only affects this mbuf, then just
1563 * adjust and return. Otherwise, rescan and truncate
1564 * after the remaining size.
1565 */
1566 count = 0;
1567 m = mp;
1568 for (;;) {
1569 count += m->m_len;
1570 if (m->m_next == NULL)
1571 break;
1572 m = m->m_next;
1573 }
1574
1575 KDASSERT(count >= len);
1576
1577 if (m->m_len >= len) {
1578 m->m_len -= len;
1579 } else {
1580 count -= len;
1581 /*
1582 * Correct length for chain is "count".
1583 * Find the mbuf with last data, adjust its length,
1584 * and toss data from remaining mbufs on chain.
1585 */
1586 for (m = mp; m; m = m->m_next) {
1587 if (m->m_len >= count) {
1588 m->m_len = count;
1589 break;
1590 }
1591 count -= m->m_len;
1592 }
1593 KASSERT(m && m->m_next);
1594 m_freem(m->m_next);
1595 m->m_next = NULL;
1596 }
1597
1598 KDASSERT(m->m_next == NULL);
1599
1600 /*
1601 * zero-padding.
1602 */
1603 if (nul > 0) {
1604 char *cp;
1605 int i;
1606
1607 if (M_READONLY(m) || M_TRAILINGSPACE(m) < nul) {
1608 struct mbuf *n;
1609
1610 KDASSERT(MLEN >= nul);
1611 n = m_get(M_WAIT, MT_DATA);
1612 MCLAIM(n, &nfs_mowner);
1613 n->m_len = nul;
1614 n->m_next = NULL;
1615 m->m_next = n;
1616 cp = mtod(n, void *);
1617 } else {
1618 cp = mtod(m, char *) + m->m_len;
1619 m->m_len += nul;
1620 }
1621 for (i = 0; i < nul; i++)
1622 *cp++ = '\0';
1623 }
1624 return;
1625 }
1626
1627 /*
1628 * Make these functions instead of macros, so that the kernel text size
1629 * doesn't get too big...
1630 */
1631 void
nfsm_srvwcc(struct nfsrv_descript * nfsd,int before_ret,struct vattr * before_vap,int after_ret,struct vattr * after_vap,struct mbuf ** mbp,char ** bposp)1632 nfsm_srvwcc(struct nfsrv_descript *nfsd, int before_ret, struct vattr *before_vap, int after_ret, struct vattr *after_vap, struct mbuf **mbp, char **bposp)
1633 {
1634 struct mbuf *mb = *mbp;
1635 char *bpos = *bposp;
1636 u_int32_t *tl;
1637
1638 if (before_ret) {
1639 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1640 *tl = nfs_false;
1641 } else {
1642 nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
1643 *tl++ = nfs_true;
1644 txdr_hyper(before_vap->va_size, tl);
1645 tl += 2;
1646 txdr_nfsv3time(&(before_vap->va_mtime), tl);
1647 tl += 2;
1648 txdr_nfsv3time(&(before_vap->va_ctime), tl);
1649 }
1650 *bposp = bpos;
1651 *mbp = mb;
1652 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
1653 }
1654
1655 void
nfsm_srvpostopattr(struct nfsrv_descript * nfsd,int after_ret,struct vattr * after_vap,struct mbuf ** mbp,char ** bposp)1656 nfsm_srvpostopattr(struct nfsrv_descript *nfsd, int after_ret, struct vattr *after_vap, struct mbuf **mbp, char **bposp)
1657 {
1658 struct mbuf *mb = *mbp;
1659 char *bpos = *bposp;
1660 u_int32_t *tl;
1661 struct nfs_fattr *fp;
1662
1663 if (after_ret) {
1664 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1665 *tl = nfs_false;
1666 } else {
1667 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
1668 *tl++ = nfs_true;
1669 fp = (struct nfs_fattr *)tl;
1670 nfsm_srvfattr(nfsd, after_vap, fp);
1671 }
1672 *mbp = mb;
1673 *bposp = bpos;
1674 }
1675
1676 void
nfsm_srvfattr(struct nfsrv_descript * nfsd,struct vattr * vap,struct nfs_fattr * fp)1677 nfsm_srvfattr(struct nfsrv_descript *nfsd, struct vattr *vap, struct nfs_fattr *fp)
1678 {
1679
1680 fp->fa_nlink = txdr_unsigned(vap->va_nlink);
1681 fp->fa_uid = txdr_unsigned(vap->va_uid);
1682 fp->fa_gid = txdr_unsigned(vap->va_gid);
1683 if (nfsd->nd_flag & ND_NFSV3) {
1684 fp->fa_type = vtonfsv3_type(vap->va_type);
1685 fp->fa_mode = vtonfsv3_mode(vap->va_mode);
1686 txdr_hyper(vap->va_size, &fp->fa3_size);
1687 txdr_hyper(vap->va_bytes, &fp->fa3_used);
1688 fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
1689 fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
1690 fp->fa3_fsid.nfsuquad[0] = 0;
1691 fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
1692 txdr_hyper(vap->va_fileid, &fp->fa3_fileid);
1693 txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
1694 txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
1695 txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
1696 } else {
1697 fp->fa_type = vtonfsv2_type(vap->va_type);
1698 fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1699 fp->fa2_size = txdr_unsigned(NFS_V2CLAMP32(vap->va_size));
1700 fp->fa2_blocksize = txdr_unsigned(NFS_V2CLAMP16(vap->va_blocksize));
1701 if (vap->va_type == VFIFO)
1702 fp->fa2_rdev = 0xffffffff;
1703 else
1704 fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
1705 fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
1706 fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
1707 fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
1708 txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
1709 txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
1710 txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
1711 }
1712 }
1713
1714 /*
1715 * This function compares two net addresses by family and returns true
1716 * if they are the same host.
1717 * If there is any doubt, return false.
1718 * The AF_INET family is handled as a special case so that address mbufs
1719 * don't need to be saved to store "struct in_addr", which is only 4 bytes.
1720 */
1721 int
netaddr_match(int family,union nethostaddr * haddr,struct mbuf * nam)1722 netaddr_match(int family, union nethostaddr *haddr, struct mbuf *nam)
1723 {
1724 struct sockaddr_in *inetaddr;
1725
1726 switch (family) {
1727 case AF_INET:
1728 inetaddr = mtod(nam, struct sockaddr_in *);
1729 if (inetaddr->sin_family == AF_INET &&
1730 inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
1731 return (1);
1732 break;
1733 case AF_INET6:
1734 {
1735 struct sockaddr_in6 *sin6_1, *sin6_2;
1736
1737 sin6_1 = mtod(nam, struct sockaddr_in6 *);
1738 sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
1739 if (sin6_1->sin6_family == AF_INET6 &&
1740 IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
1741 return 1;
1742 }
1743 default:
1744 break;
1745 };
1746 return (0);
1747 }
1748
1749 struct nfs_clearcommit_ctx {
1750 struct mount *mp;
1751 };
1752
1753 static bool
nfs_clearcommit_selector(void * cl,struct vnode * vp)1754 nfs_clearcommit_selector(void *cl, struct vnode *vp)
1755 {
1756 struct nfs_clearcommit_ctx *c = cl;
1757 struct nfsnode *np;
1758
1759 KASSERT(mutex_owned(vp->v_interlock));
1760
1761 /* XXXAD mountpoint check looks like nonsense to me */
1762 np = VTONFS(vp);
1763 if (vp->v_type != VREG || vp->v_mount != c->mp || np == NULL)
1764 return false;
1765 return false;
1766 }
1767
1768 /*
1769 * The write verifier has changed (probably due to a server reboot), so all
1770 * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
1771 * as dirty or are being written out just now, all this takes is clearing
1772 * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
1773 * the mount point.
1774 */
1775 void
nfs_clearcommit(struct mount * mp)1776 nfs_clearcommit(struct mount *mp)
1777 {
1778 struct vnode *vp;
1779 struct vnode_iterator *marker;
1780 struct nfsmount *nmp = VFSTONFS(mp);
1781 struct nfs_clearcommit_ctx ctx;
1782 struct nfsnode *np;
1783 struct vm_page *pg;
1784 struct uvm_page_array a;
1785 voff_t off;
1786
1787 rw_enter(&nmp->nm_writeverflock, RW_WRITER);
1788 vfs_vnode_iterator_init(mp, &marker);
1789 ctx.mp = mp;
1790 for (;;) {
1791 vp = vfs_vnode_iterator_next(marker, nfs_clearcommit_selector,
1792 &ctx);
1793 if (vp == NULL)
1794 break;
1795 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
1796 np = VTONFS(vp);
1797 np->n_pushlo = np->n_pushhi = np->n_pushedlo =
1798 np->n_pushedhi = 0;
1799 np->n_commitflags &=
1800 ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
1801 uvm_page_array_init(&a, &vp->v_uobj, 0);
1802 off = 0;
1803 while ((pg = uvm_page_array_fill_and_peek(&a, off, 0)) !=
1804 NULL) {
1805 pg->flags &= ~PG_NEEDCOMMIT;
1806 uvm_page_array_advance(&a);
1807 off = pg->offset + PAGE_SIZE;
1808 }
1809 uvm_page_array_fini(&a);
1810 rw_exit(vp->v_uobj.vmobjlock);
1811 vrele(vp);
1812 }
1813 KASSERT(vp == NULL);
1814 vfs_vnode_iterator_destroy(marker);
1815 mutex_enter(&nmp->nm_lock);
1816 nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
1817 mutex_exit(&nmp->nm_lock);
1818 rw_exit(&nmp->nm_writeverflock);
1819 }
1820
1821 void
nfs_merge_commit_ranges(struct vnode * vp)1822 nfs_merge_commit_ranges(struct vnode *vp)
1823 {
1824 struct nfsnode *np = VTONFS(vp);
1825
1826 KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
1827
1828 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
1829 np->n_pushedlo = np->n_pushlo;
1830 np->n_pushedhi = np->n_pushhi;
1831 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
1832 } else {
1833 if (np->n_pushlo < np->n_pushedlo)
1834 np->n_pushedlo = np->n_pushlo;
1835 if (np->n_pushhi > np->n_pushedhi)
1836 np->n_pushedhi = np->n_pushhi;
1837 }
1838
1839 np->n_pushlo = np->n_pushhi = 0;
1840 np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
1841
1842 #ifdef NFS_DEBUG_COMMIT
1843 printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
1844 (unsigned)np->n_pushedhi);
1845 #endif
1846 }
1847
1848 int
nfs_in_committed_range(struct vnode * vp,off_t off,off_t len)1849 nfs_in_committed_range(struct vnode *vp, off_t off, off_t len)
1850 {
1851 struct nfsnode *np = VTONFS(vp);
1852 off_t lo, hi;
1853
1854 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
1855 return 0;
1856 lo = off;
1857 hi = lo + len;
1858
1859 return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
1860 }
1861
1862 int
nfs_in_tobecommitted_range(struct vnode * vp,off_t off,off_t len)1863 nfs_in_tobecommitted_range(struct vnode *vp, off_t off, off_t len)
1864 {
1865 struct nfsnode *np = VTONFS(vp);
1866 off_t lo, hi;
1867
1868 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
1869 return 0;
1870 lo = off;
1871 hi = lo + len;
1872
1873 return (lo >= np->n_pushlo && hi <= np->n_pushhi);
1874 }
1875
1876 void
nfs_add_committed_range(struct vnode * vp,off_t off,off_t len)1877 nfs_add_committed_range(struct vnode *vp, off_t off, off_t len)
1878 {
1879 struct nfsnode *np = VTONFS(vp);
1880 off_t lo, hi;
1881
1882 lo = off;
1883 hi = lo + len;
1884
1885 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
1886 np->n_pushedlo = lo;
1887 np->n_pushedhi = hi;
1888 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
1889 } else {
1890 if (hi > np->n_pushedhi)
1891 np->n_pushedhi = hi;
1892 if (lo < np->n_pushedlo)
1893 np->n_pushedlo = lo;
1894 }
1895 #ifdef NFS_DEBUG_COMMIT
1896 printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
1897 (unsigned)np->n_pushedhi);
1898 #endif
1899 }
1900
1901 void
nfs_del_committed_range(struct vnode * vp,off_t off,off_t len)1902 nfs_del_committed_range(struct vnode *vp, off_t off, off_t len)
1903 {
1904 struct nfsnode *np = VTONFS(vp);
1905 off_t lo, hi;
1906
1907 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
1908 return;
1909
1910 lo = off;
1911 hi = lo + len;
1912
1913 if (lo > np->n_pushedhi || hi < np->n_pushedlo)
1914 return;
1915 if (lo <= np->n_pushedlo)
1916 np->n_pushedlo = hi;
1917 else if (hi >= np->n_pushedhi)
1918 np->n_pushedhi = lo;
1919 else {
1920 /*
1921 * XXX There's only one range. If the deleted range
1922 * is in the middle, pick the largest of the
1923 * contiguous ranges that it leaves.
1924 */
1925 if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
1926 np->n_pushedhi = lo;
1927 else
1928 np->n_pushedlo = hi;
1929 }
1930 #ifdef NFS_DEBUG_COMMIT
1931 printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
1932 (unsigned)np->n_pushedhi);
1933 #endif
1934 }
1935
1936 void
nfs_add_tobecommitted_range(struct vnode * vp,off_t off,off_t len)1937 nfs_add_tobecommitted_range(struct vnode *vp, off_t off, off_t len)
1938 {
1939 struct nfsnode *np = VTONFS(vp);
1940 off_t lo, hi;
1941
1942 lo = off;
1943 hi = lo + len;
1944
1945 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
1946 np->n_pushlo = lo;
1947 np->n_pushhi = hi;
1948 np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
1949 } else {
1950 if (lo < np->n_pushlo)
1951 np->n_pushlo = lo;
1952 if (hi > np->n_pushhi)
1953 np->n_pushhi = hi;
1954 }
1955 #ifdef NFS_DEBUG_COMMIT
1956 printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
1957 (unsigned)np->n_pushhi);
1958 #endif
1959 }
1960
1961 void
nfs_del_tobecommitted_range(struct vnode * vp,off_t off,off_t len)1962 nfs_del_tobecommitted_range(struct vnode *vp, off_t off, off_t len)
1963 {
1964 struct nfsnode *np = VTONFS(vp);
1965 off_t lo, hi;
1966
1967 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
1968 return;
1969
1970 lo = off;
1971 hi = lo + len;
1972
1973 if (lo > np->n_pushhi || hi < np->n_pushlo)
1974 return;
1975
1976 if (lo <= np->n_pushlo)
1977 np->n_pushlo = hi;
1978 else if (hi >= np->n_pushhi)
1979 np->n_pushhi = lo;
1980 else {
1981 /*
1982 * XXX There's only one range. If the deleted range
1983 * is in the middle, pick the largest of the
1984 * contiguous ranges that it leaves.
1985 */
1986 if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
1987 np->n_pushhi = lo;
1988 else
1989 np->n_pushlo = hi;
1990 }
1991 #ifdef NFS_DEBUG_COMMIT
1992 printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
1993 (unsigned)np->n_pushhi);
1994 #endif
1995 }
1996
1997 /*
1998 * Map errnos to NFS error numbers. For Version 3 also filter out error
1999 * numbers not specified for the associated procedure.
2000 */
2001 int
nfsrv_errmap(struct nfsrv_descript * nd,int err)2002 nfsrv_errmap(struct nfsrv_descript *nd, int err)
2003 {
2004 const short *defaulterrp, *errp;
2005
2006 if (nd->nd_flag & ND_NFSV3) {
2007 if (nd->nd_procnum <= NFSPROC_COMMIT) {
2008 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2009 while (*++errp) {
2010 if (*errp == err)
2011 return (err);
2012 else if (*errp > err)
2013 break;
2014 }
2015 return ((int)*defaulterrp);
2016 } else
2017 return (err & 0xffff);
2018 }
2019 if (err <= ELAST)
2020 return ((int)nfsrv_v2errmap[err - 1]);
2021 return (NFSERR_IO);
2022 }
2023
2024 u_int32_t
nfs_getxid(void)2025 nfs_getxid(void)
2026 {
2027 u_int32_t newxid;
2028
2029 if (__predict_false(nfs_xid == 0)) {
2030 nfs_xid = cprng_fast32();
2031 }
2032
2033 /* get next xid. skip 0 */
2034 do {
2035 newxid = atomic_inc_32_nv(&nfs_xid);
2036 } while (__predict_false(newxid == 0));
2037
2038 return txdr_unsigned(newxid);
2039 }
2040
2041 /*
2042 * assign a new xid for existing request.
2043 * used for NFSERR_JUKEBOX handling.
2044 */
2045 void
nfs_renewxid(struct nfsreq * req)2046 nfs_renewxid(struct nfsreq *req)
2047 {
2048 u_int32_t xid;
2049 int off;
2050
2051 xid = nfs_getxid();
2052 if (req->r_nmp->nm_sotype == SOCK_STREAM)
2053 off = sizeof(u_int32_t); /* RPC record mark */
2054 else
2055 off = 0;
2056
2057 m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid);
2058 req->r_xid = xid;
2059 }
2060