1 /*	$NetBSD: nfs_subs.c,v 1.228 2016/06/10 13:27:16 ozaki-r 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 1.228 2016/06/10 13:27:16 ozaki-r 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.h>
104 
105 #include <nfs/rpcv2.h>
106 #include <nfs/nfsproto.h>
107 #include <nfs/nfsnode.h>
108 #include <nfs/nfs.h>
109 #include <nfs/xdr_subs.h>
110 #include <nfs/nfsm_subs.h>
111 #include <nfs/nfsmount.h>
112 #include <nfs/nfsrtt.h>
113 #include <nfs/nfs_var.h>
114 
115 #include <miscfs/specfs/specdev.h>
116 
117 #include <netinet/in.h>
118 
119 static u_int32_t nfs_xid;
120 
121 int nuidhash_max = NFS_MAXUIDHASH;
122 /*
123  * Data items converted to xdr at startup, since they are constant
124  * This is kinda hokey, but may save a little time doing byte swaps
125  */
126 u_int32_t nfs_xdrneg1;
127 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
128 	rpc_mismatch, rpc_auth_unix, rpc_msgaccepted,
129 	rpc_auth_kerb;
130 u_int32_t nfs_prog, nfs_true, nfs_false;
131 
132 /* And other global data */
133 const nfstype nfsv2_type[9] =
134 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON };
135 const nfstype nfsv3_type[9] =
136 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON };
137 const enum vtype nv2tov_type[8] =
138 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON };
139 const enum vtype nv3tov_type[8] =
140 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO };
141 int nfs_ticks;
142 
143 /* NFS client/server stats. */
144 struct nfsstats nfsstats;
145 
146 /*
147  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
148  */
149 const int nfsv3_procid[NFS_NPROCS] = {
150 	NFSPROC_NULL,
151 	NFSPROC_GETATTR,
152 	NFSPROC_SETATTR,
153 	NFSPROC_NOOP,
154 	NFSPROC_LOOKUP,
155 	NFSPROC_READLINK,
156 	NFSPROC_READ,
157 	NFSPROC_NOOP,
158 	NFSPROC_WRITE,
159 	NFSPROC_CREATE,
160 	NFSPROC_REMOVE,
161 	NFSPROC_RENAME,
162 	NFSPROC_LINK,
163 	NFSPROC_SYMLINK,
164 	NFSPROC_MKDIR,
165 	NFSPROC_RMDIR,
166 	NFSPROC_READDIR,
167 	NFSPROC_FSSTAT,
168 	NFSPROC_NOOP,
169 	NFSPROC_NOOP,
170 	NFSPROC_NOOP,
171 	NFSPROC_NOOP,
172 	NFSPROC_NOOP
173 };
174 
175 /*
176  * and the reverse mapping from generic to Version 2 procedure numbers
177  */
178 const int nfsv2_procid[NFS_NPROCS] = {
179 	NFSV2PROC_NULL,
180 	NFSV2PROC_GETATTR,
181 	NFSV2PROC_SETATTR,
182 	NFSV2PROC_LOOKUP,
183 	NFSV2PROC_NOOP,
184 	NFSV2PROC_READLINK,
185 	NFSV2PROC_READ,
186 	NFSV2PROC_WRITE,
187 	NFSV2PROC_CREATE,
188 	NFSV2PROC_MKDIR,
189 	NFSV2PROC_SYMLINK,
190 	NFSV2PROC_CREATE,
191 	NFSV2PROC_REMOVE,
192 	NFSV2PROC_RMDIR,
193 	NFSV2PROC_RENAME,
194 	NFSV2PROC_LINK,
195 	NFSV2PROC_READDIR,
196 	NFSV2PROC_NOOP,
197 	NFSV2PROC_STATFS,
198 	NFSV2PROC_NOOP,
199 	NFSV2PROC_NOOP,
200 	NFSV2PROC_NOOP,
201 	NFSV2PROC_NOOP,
202 };
203 
204 /*
205  * Maps errno values to nfs error numbers.
206  * Use NFSERR_IO as the catch all for ones not specifically defined in
207  * RFC 1094.
208  */
209 static const u_char nfsrv_v2errmap[ELAST] = {
210   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
211   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
212   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
213   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
214   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
215   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
216   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
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_NAMETOL,	NFSERR_IO,	NFSERR_IO,
223   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
224   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
225   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
226   NFSERR_IO,	NFSERR_IO,
227 };
228 
229 /*
230  * Maps errno values to nfs error numbers.
231  * Although it is not obvious whether or not NFS clients really care if
232  * a returned error value is in the specified list for the procedure, the
233  * safest thing to do is filter them appropriately. For Version 2, the
234  * X/Open XNFS document is the only specification that defines error values
235  * for each RPC (The RFC simply lists all possible error values for all RPCs),
236  * so I have decided to not do this for Version 2.
237  * The first entry is the default error return and the rest are the valid
238  * errors for that RPC in increasing numeric order.
239  */
240 static const short nfsv3err_null[] = {
241 	0,
242 	0,
243 };
244 
245 static const short nfsv3err_getattr[] = {
246 	NFSERR_IO,
247 	NFSERR_IO,
248 	NFSERR_STALE,
249 	NFSERR_BADHANDLE,
250 	NFSERR_SERVERFAULT,
251 	0,
252 };
253 
254 static const short nfsv3err_setattr[] = {
255 	NFSERR_IO,
256 	NFSERR_PERM,
257 	NFSERR_IO,
258 	NFSERR_ACCES,
259 	NFSERR_INVAL,
260 	NFSERR_NOSPC,
261 	NFSERR_ROFS,
262 	NFSERR_DQUOT,
263 	NFSERR_STALE,
264 	NFSERR_BADHANDLE,
265 	NFSERR_NOT_SYNC,
266 	NFSERR_SERVERFAULT,
267 	0,
268 };
269 
270 static const short nfsv3err_lookup[] = {
271 	NFSERR_IO,
272 	NFSERR_NOENT,
273 	NFSERR_IO,
274 	NFSERR_ACCES,
275 	NFSERR_NOTDIR,
276 	NFSERR_NAMETOL,
277 	NFSERR_STALE,
278 	NFSERR_BADHANDLE,
279 	NFSERR_SERVERFAULT,
280 	0,
281 };
282 
283 static const short nfsv3err_access[] = {
284 	NFSERR_IO,
285 	NFSERR_IO,
286 	NFSERR_STALE,
287 	NFSERR_BADHANDLE,
288 	NFSERR_SERVERFAULT,
289 	0,
290 };
291 
292 static const short nfsv3err_readlink[] = {
293 	NFSERR_IO,
294 	NFSERR_IO,
295 	NFSERR_ACCES,
296 	NFSERR_INVAL,
297 	NFSERR_STALE,
298 	NFSERR_BADHANDLE,
299 	NFSERR_NOTSUPP,
300 	NFSERR_SERVERFAULT,
301 	0,
302 };
303 
304 static const short nfsv3err_read[] = {
305 	NFSERR_IO,
306 	NFSERR_IO,
307 	NFSERR_NXIO,
308 	NFSERR_ACCES,
309 	NFSERR_INVAL,
310 	NFSERR_STALE,
311 	NFSERR_BADHANDLE,
312 	NFSERR_SERVERFAULT,
313 	NFSERR_JUKEBOX,
314 	0,
315 };
316 
317 static const short nfsv3err_write[] = {
318 	NFSERR_IO,
319 	NFSERR_IO,
320 	NFSERR_ACCES,
321 	NFSERR_INVAL,
322 	NFSERR_FBIG,
323 	NFSERR_NOSPC,
324 	NFSERR_ROFS,
325 	NFSERR_DQUOT,
326 	NFSERR_STALE,
327 	NFSERR_BADHANDLE,
328 	NFSERR_SERVERFAULT,
329 	NFSERR_JUKEBOX,
330 	0,
331 };
332 
333 static const short nfsv3err_create[] = {
334 	NFSERR_IO,
335 	NFSERR_IO,
336 	NFSERR_ACCES,
337 	NFSERR_EXIST,
338 	NFSERR_NOTDIR,
339 	NFSERR_NOSPC,
340 	NFSERR_ROFS,
341 	NFSERR_NAMETOL,
342 	NFSERR_DQUOT,
343 	NFSERR_STALE,
344 	NFSERR_BADHANDLE,
345 	NFSERR_NOTSUPP,
346 	NFSERR_SERVERFAULT,
347 	0,
348 };
349 
350 static const short nfsv3err_mkdir[] = {
351 	NFSERR_IO,
352 	NFSERR_IO,
353 	NFSERR_ACCES,
354 	NFSERR_EXIST,
355 	NFSERR_NOTDIR,
356 	NFSERR_NOSPC,
357 	NFSERR_ROFS,
358 	NFSERR_NAMETOL,
359 	NFSERR_DQUOT,
360 	NFSERR_STALE,
361 	NFSERR_BADHANDLE,
362 	NFSERR_NOTSUPP,
363 	NFSERR_SERVERFAULT,
364 	0,
365 };
366 
367 static const short nfsv3err_symlink[] = {
368 	NFSERR_IO,
369 	NFSERR_IO,
370 	NFSERR_ACCES,
371 	NFSERR_EXIST,
372 	NFSERR_NOTDIR,
373 	NFSERR_NOSPC,
374 	NFSERR_ROFS,
375 	NFSERR_NAMETOL,
376 	NFSERR_DQUOT,
377 	NFSERR_STALE,
378 	NFSERR_BADHANDLE,
379 	NFSERR_NOTSUPP,
380 	NFSERR_SERVERFAULT,
381 	0,
382 };
383 
384 static const short nfsv3err_mknod[] = {
385 	NFSERR_IO,
386 	NFSERR_IO,
387 	NFSERR_ACCES,
388 	NFSERR_EXIST,
389 	NFSERR_NOTDIR,
390 	NFSERR_NOSPC,
391 	NFSERR_ROFS,
392 	NFSERR_NAMETOL,
393 	NFSERR_DQUOT,
394 	NFSERR_STALE,
395 	NFSERR_BADHANDLE,
396 	NFSERR_NOTSUPP,
397 	NFSERR_SERVERFAULT,
398 	NFSERR_BADTYPE,
399 	0,
400 };
401 
402 static const short nfsv3err_remove[] = {
403 	NFSERR_IO,
404 	NFSERR_NOENT,
405 	NFSERR_IO,
406 	NFSERR_ACCES,
407 	NFSERR_NOTDIR,
408 	NFSERR_ROFS,
409 	NFSERR_NAMETOL,
410 	NFSERR_STALE,
411 	NFSERR_BADHANDLE,
412 	NFSERR_SERVERFAULT,
413 	0,
414 };
415 
416 static const short nfsv3err_rmdir[] = {
417 	NFSERR_IO,
418 	NFSERR_NOENT,
419 	NFSERR_IO,
420 	NFSERR_ACCES,
421 	NFSERR_EXIST,
422 	NFSERR_NOTDIR,
423 	NFSERR_INVAL,
424 	NFSERR_ROFS,
425 	NFSERR_NAMETOL,
426 	NFSERR_NOTEMPTY,
427 	NFSERR_STALE,
428 	NFSERR_BADHANDLE,
429 	NFSERR_NOTSUPP,
430 	NFSERR_SERVERFAULT,
431 	0,
432 };
433 
434 static const short nfsv3err_rename[] = {
435 	NFSERR_IO,
436 	NFSERR_NOENT,
437 	NFSERR_IO,
438 	NFSERR_ACCES,
439 	NFSERR_EXIST,
440 	NFSERR_XDEV,
441 	NFSERR_NOTDIR,
442 	NFSERR_ISDIR,
443 	NFSERR_INVAL,
444 	NFSERR_NOSPC,
445 	NFSERR_ROFS,
446 	NFSERR_MLINK,
447 	NFSERR_NAMETOL,
448 	NFSERR_NOTEMPTY,
449 	NFSERR_DQUOT,
450 	NFSERR_STALE,
451 	NFSERR_BADHANDLE,
452 	NFSERR_NOTSUPP,
453 	NFSERR_SERVERFAULT,
454 	0,
455 };
456 
457 static const short nfsv3err_link[] = {
458 	NFSERR_IO,
459 	NFSERR_IO,
460 	NFSERR_ACCES,
461 	NFSERR_EXIST,
462 	NFSERR_XDEV,
463 	NFSERR_NOTDIR,
464 	NFSERR_INVAL,
465 	NFSERR_NOSPC,
466 	NFSERR_ROFS,
467 	NFSERR_MLINK,
468 	NFSERR_NAMETOL,
469 	NFSERR_DQUOT,
470 	NFSERR_STALE,
471 	NFSERR_BADHANDLE,
472 	NFSERR_NOTSUPP,
473 	NFSERR_SERVERFAULT,
474 	0,
475 };
476 
477 static const short nfsv3err_readdir[] = {
478 	NFSERR_IO,
479 	NFSERR_IO,
480 	NFSERR_ACCES,
481 	NFSERR_NOTDIR,
482 	NFSERR_STALE,
483 	NFSERR_BADHANDLE,
484 	NFSERR_BAD_COOKIE,
485 	NFSERR_TOOSMALL,
486 	NFSERR_SERVERFAULT,
487 	0,
488 };
489 
490 static const short nfsv3err_readdirplus[] = {
491 	NFSERR_IO,
492 	NFSERR_IO,
493 	NFSERR_ACCES,
494 	NFSERR_NOTDIR,
495 	NFSERR_STALE,
496 	NFSERR_BADHANDLE,
497 	NFSERR_BAD_COOKIE,
498 	NFSERR_NOTSUPP,
499 	NFSERR_TOOSMALL,
500 	NFSERR_SERVERFAULT,
501 	0,
502 };
503 
504 static const short nfsv3err_fsstat[] = {
505 	NFSERR_IO,
506 	NFSERR_IO,
507 	NFSERR_STALE,
508 	NFSERR_BADHANDLE,
509 	NFSERR_SERVERFAULT,
510 	0,
511 };
512 
513 static const short nfsv3err_fsinfo[] = {
514 	NFSERR_STALE,
515 	NFSERR_STALE,
516 	NFSERR_BADHANDLE,
517 	NFSERR_SERVERFAULT,
518 	0,
519 };
520 
521 static const short nfsv3err_pathconf[] = {
522 	NFSERR_STALE,
523 	NFSERR_STALE,
524 	NFSERR_BADHANDLE,
525 	NFSERR_SERVERFAULT,
526 	0,
527 };
528 
529 static const short nfsv3err_commit[] = {
530 	NFSERR_IO,
531 	NFSERR_IO,
532 	NFSERR_STALE,
533 	NFSERR_BADHANDLE,
534 	NFSERR_SERVERFAULT,
535 	0,
536 };
537 
538 static const short * const nfsrv_v3errmap[] = {
539 	nfsv3err_null,
540 	nfsv3err_getattr,
541 	nfsv3err_setattr,
542 	nfsv3err_lookup,
543 	nfsv3err_access,
544 	nfsv3err_readlink,
545 	nfsv3err_read,
546 	nfsv3err_write,
547 	nfsv3err_create,
548 	nfsv3err_mkdir,
549 	nfsv3err_symlink,
550 	nfsv3err_mknod,
551 	nfsv3err_remove,
552 	nfsv3err_rmdir,
553 	nfsv3err_rename,
554 	nfsv3err_link,
555 	nfsv3err_readdir,
556 	nfsv3err_readdirplus,
557 	nfsv3err_fsstat,
558 	nfsv3err_fsinfo,
559 	nfsv3err_pathconf,
560 	nfsv3err_commit,
561 };
562 
563 extern struct nfsrtt nfsrtt;
564 
565 u_long nfsdirhashmask;
566 
567 int nfs_webnamei(struct nameidata *, struct vnode *, struct proc *);
568 
569 /*
570  * Create the header for an rpc request packet
571  * The hsiz is the size of the rest of the nfs request header.
572  * (just used to decide if a cluster is a good idea)
573  */
574 struct mbuf *
nfsm_reqh(struct nfsnode * np,u_long procid,int hsiz,char ** bposp)575 nfsm_reqh(struct nfsnode *np, u_long procid, int hsiz, char **bposp)
576 {
577 	struct mbuf *mb;
578 	char *bpos;
579 
580 	mb = m_get(M_WAIT, MT_DATA);
581 	MCLAIM(mb, &nfs_mowner);
582 	if (hsiz >= MINCLSIZE)
583 		m_clget(mb, M_WAIT);
584 	mb->m_len = 0;
585 	bpos = mtod(mb, void *);
586 
587 	/* Finally, return values */
588 	*bposp = bpos;
589 	return (mb);
590 }
591 
592 /*
593  * Build the RPC header and fill in the authorization info.
594  * The authorization string argument is only used when the credentials
595  * come from outside of the kernel.
596  * Returns the head of the mbuf list.
597  */
598 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)599 nfsm_rpchead(kauth_cred_t cr, int nmflag, int procid,
600 	int auth_type, int auth_len, char *auth_str, int verf_len,
601 	char *verf_str, struct mbuf *mrest, int mrest_len,
602 	struct mbuf **mbp, uint32_t *xidp)
603 {
604 	struct mbuf *mb;
605 	u_int32_t *tl;
606 	char *bpos;
607 	int i;
608 	struct mbuf *mreq;
609 	int siz, grpsiz, authsiz;
610 
611 	authsiz = nfsm_rndup(auth_len);
612 	mb = m_gethdr(M_WAIT, MT_DATA);
613 	MCLAIM(mb, &nfs_mowner);
614 	if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
615 		m_clget(mb, M_WAIT);
616 	} else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
617 		MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
618 	} else {
619 		MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
620 	}
621 	mb->m_len = 0;
622 	mreq = mb;
623 	bpos = mtod(mb, void *);
624 
625 	/*
626 	 * First the RPC header.
627 	 */
628 	nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
629 
630 	*tl++ = *xidp = nfs_getxid();
631 	*tl++ = rpc_call;
632 	*tl++ = rpc_vers;
633 	*tl++ = txdr_unsigned(NFS_PROG);
634 	if (nmflag & NFSMNT_NFSV3)
635 		*tl++ = txdr_unsigned(NFS_VER3);
636 	else
637 		*tl++ = txdr_unsigned(NFS_VER2);
638 	if (nmflag & NFSMNT_NFSV3)
639 		*tl++ = txdr_unsigned(procid);
640 	else
641 		*tl++ = txdr_unsigned(nfsv2_procid[procid]);
642 
643 	/*
644 	 * And then the authorization cred.
645 	 */
646 	*tl++ = txdr_unsigned(auth_type);
647 	*tl = txdr_unsigned(authsiz);
648 	switch (auth_type) {
649 	case RPCAUTH_UNIX:
650 		nfsm_build(tl, u_int32_t *, auth_len);
651 		*tl++ = 0;		/* stamp ?? */
652 		*tl++ = 0;		/* NULL hostname */
653 		*tl++ = txdr_unsigned(kauth_cred_geteuid(cr));
654 		*tl++ = txdr_unsigned(kauth_cred_getegid(cr));
655 		grpsiz = (auth_len >> 2) - 5;
656 		*tl++ = txdr_unsigned(grpsiz);
657 		for (i = 0; i < grpsiz; i++)
658 			*tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */
659 		break;
660 	case RPCAUTH_KERB4:
661 		siz = auth_len;
662 		while (siz > 0) {
663 			if (M_TRAILINGSPACE(mb) == 0) {
664 				struct mbuf *mb2;
665 				mb2 = m_get(M_WAIT, MT_DATA);
666 				MCLAIM(mb2, &nfs_mowner);
667 				if (siz >= MINCLSIZE)
668 					m_clget(mb2, M_WAIT);
669 				mb->m_next = mb2;
670 				mb = mb2;
671 				mb->m_len = 0;
672 				bpos = mtod(mb, void *);
673 			}
674 			i = min(siz, M_TRAILINGSPACE(mb));
675 			memcpy(bpos, auth_str, i);
676 			mb->m_len += i;
677 			auth_str += i;
678 			bpos += i;
679 			siz -= i;
680 		}
681 		if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
682 			for (i = 0; i < siz; i++)
683 				*bpos++ = '\0';
684 			mb->m_len += siz;
685 		}
686 		break;
687 	};
688 
689 	/*
690 	 * And the verifier...
691 	 */
692 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
693 	if (verf_str) {
694 		*tl++ = txdr_unsigned(RPCAUTH_KERB4);
695 		*tl = txdr_unsigned(verf_len);
696 		siz = verf_len;
697 		while (siz > 0) {
698 			if (M_TRAILINGSPACE(mb) == 0) {
699 				struct mbuf *mb2;
700 				mb2 = m_get(M_WAIT, MT_DATA);
701 				MCLAIM(mb2, &nfs_mowner);
702 				if (siz >= MINCLSIZE)
703 					m_clget(mb2, M_WAIT);
704 				mb->m_next = mb2;
705 				mb = mb2;
706 				mb->m_len = 0;
707 				bpos = mtod(mb, void *);
708 			}
709 			i = min(siz, M_TRAILINGSPACE(mb));
710 			memcpy(bpos, verf_str, i);
711 			mb->m_len += i;
712 			verf_str += i;
713 			bpos += i;
714 			siz -= i;
715 		}
716 		if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
717 			for (i = 0; i < siz; i++)
718 				*bpos++ = '\0';
719 			mb->m_len += siz;
720 		}
721 	} else {
722 		*tl++ = txdr_unsigned(RPCAUTH_NULL);
723 		*tl = 0;
724 	}
725 	mb->m_next = mrest;
726 	mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
727 	m_reset_rcvif(mreq);
728 	*mbp = mb;
729 	return (mreq);
730 }
731 
732 /*
733  * copies mbuf chain to the uio scatter/gather list
734  */
735 int
nfsm_mbuftouio(struct mbuf ** mrep,struct uio * uiop,int siz,char ** dpos)736 nfsm_mbuftouio(struct mbuf **mrep, struct uio *uiop, int siz, char **dpos)
737 {
738 	char *mbufcp, *uiocp;
739 	int xfer, left, len;
740 	struct mbuf *mp;
741 	long uiosiz, rem;
742 	int error = 0;
743 
744 	mp = *mrep;
745 	mbufcp = *dpos;
746 	len = mtod(mp, char *) + mp->m_len - mbufcp;
747 	rem = nfsm_rndup(siz)-siz;
748 	while (siz > 0) {
749 		if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
750 			return (EFBIG);
751 		left = uiop->uio_iov->iov_len;
752 		uiocp = uiop->uio_iov->iov_base;
753 		if (left > siz)
754 			left = siz;
755 		uiosiz = left;
756 		while (left > 0) {
757 			while (len == 0) {
758 				mp = mp->m_next;
759 				if (mp == NULL)
760 					return (EBADRPC);
761 				mbufcp = mtod(mp, void *);
762 				len = mp->m_len;
763 			}
764 			xfer = (left > len) ? len : left;
765 			error = copyout_vmspace(uiop->uio_vmspace, mbufcp,
766 			    uiocp, xfer);
767 			if (error) {
768 				return error;
769 			}
770 			left -= xfer;
771 			len -= xfer;
772 			mbufcp += xfer;
773 			uiocp += xfer;
774 			uiop->uio_offset += xfer;
775 			uiop->uio_resid -= xfer;
776 		}
777 		if (uiop->uio_iov->iov_len <= siz) {
778 			uiop->uio_iovcnt--;
779 			uiop->uio_iov++;
780 		} else {
781 			uiop->uio_iov->iov_base =
782 			    (char *)uiop->uio_iov->iov_base + uiosiz;
783 			uiop->uio_iov->iov_len -= uiosiz;
784 		}
785 		siz -= uiosiz;
786 	}
787 	*dpos = mbufcp;
788 	*mrep = mp;
789 	if (rem > 0) {
790 		if (len < rem)
791 			error = nfs_adv(mrep, dpos, rem, len);
792 		else
793 			*dpos += rem;
794 	}
795 	return (error);
796 }
797 
798 /*
799  * copies a uio scatter/gather list to an mbuf chain.
800  * NOTE: can ony handle iovcnt == 1
801  */
802 int
nfsm_uiotombuf(struct uio * uiop,struct mbuf ** mq,int siz,char ** bpos)803 nfsm_uiotombuf(struct uio *uiop, struct mbuf **mq, int siz, char **bpos)
804 {
805 	char *uiocp;
806 	struct mbuf *mp, *mp2;
807 	int xfer, left, mlen;
808 	int uiosiz, clflg, rem;
809 	char *cp;
810 	int error;
811 
812 #ifdef DIAGNOSTIC
813 	if (uiop->uio_iovcnt != 1)
814 		panic("nfsm_uiotombuf: iovcnt != 1");
815 #endif
816 
817 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
818 		clflg = 1;
819 	else
820 		clflg = 0;
821 	rem = nfsm_rndup(siz)-siz;
822 	mp = mp2 = *mq;
823 	while (siz > 0) {
824 		left = uiop->uio_iov->iov_len;
825 		uiocp = uiop->uio_iov->iov_base;
826 		if (left > siz)
827 			left = siz;
828 		uiosiz = left;
829 		while (left > 0) {
830 			mlen = M_TRAILINGSPACE(mp);
831 			if (mlen == 0) {
832 				mp = m_get(M_WAIT, MT_DATA);
833 				MCLAIM(mp, &nfs_mowner);
834 				if (clflg)
835 					m_clget(mp, M_WAIT);
836 				mp->m_len = 0;
837 				mp2->m_next = mp;
838 				mp2 = mp;
839 				mlen = M_TRAILINGSPACE(mp);
840 			}
841 			xfer = (left > mlen) ? mlen : left;
842 			cp = mtod(mp, char *) + mp->m_len;
843 			error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp,
844 			    xfer);
845 			if (error) {
846 				/* XXX */
847 			}
848 			mp->m_len += xfer;
849 			left -= xfer;
850 			uiocp += xfer;
851 			uiop->uio_offset += xfer;
852 			uiop->uio_resid -= xfer;
853 		}
854 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base +
855 		    uiosiz;
856 		uiop->uio_iov->iov_len -= uiosiz;
857 		siz -= uiosiz;
858 	}
859 	if (rem > 0) {
860 		if (rem > M_TRAILINGSPACE(mp)) {
861 			mp = m_get(M_WAIT, MT_DATA);
862 			MCLAIM(mp, &nfs_mowner);
863 			mp->m_len = 0;
864 			mp2->m_next = mp;
865 		}
866 		cp = mtod(mp, char *) + mp->m_len;
867 		for (left = 0; left < rem; left++)
868 			*cp++ = '\0';
869 		mp->m_len += rem;
870 		*bpos = cp;
871 	} else
872 		*bpos = mtod(mp, char *) + mp->m_len;
873 	*mq = mp;
874 	return (0);
875 }
876 
877 /*
878  * Get at least "siz" bytes of correctly aligned data.
879  * When called the mbuf pointers are not necessarily correct,
880  * dsosp points to what ought to be in m_data and left contains
881  * what ought to be in m_len.
882  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
883  * cases. (The macros use the vars. dpos and dpos2)
884  */
885 int
nfsm_disct(struct mbuf ** mdp,char ** dposp,int siz,int left,char ** cp2)886 nfsm_disct(struct mbuf **mdp, char **dposp, int siz, int left, char **cp2)
887 {
888 	struct mbuf *m1, *m2;
889 	struct mbuf *havebuf = NULL;
890 	char *src = *dposp;
891 	char *dst;
892 	int len;
893 
894 #ifdef DEBUG
895 	if (left < 0)
896 		panic("nfsm_disct: left < 0");
897 #endif
898 	m1 = *mdp;
899 	/*
900 	 * Skip through the mbuf chain looking for an mbuf with
901 	 * some data. If the first mbuf found has enough data
902 	 * and it is correctly aligned return it.
903 	 */
904 	while (left == 0) {
905 		havebuf = m1;
906 		*mdp = m1 = m1->m_next;
907 		if (m1 == NULL)
908 			return (EBADRPC);
909 		src = mtod(m1, void *);
910 		left = m1->m_len;
911 		/*
912 		 * If we start a new mbuf and it is big enough
913 		 * and correctly aligned just return it, don't
914 		 * do any pull up.
915 		 */
916 		if (left >= siz && nfsm_aligned(src)) {
917 			*cp2 = src;
918 			*dposp = src + siz;
919 			return (0);
920 		}
921 	}
922 	if ((m1->m_flags & M_EXT) != 0) {
923 		if (havebuf && M_TRAILINGSPACE(havebuf) >= siz &&
924 		    nfsm_aligned(mtod(havebuf, char *) + havebuf->m_len)) {
925 			/*
926 			 * If the first mbuf with data has external data
927 			 * and there is a previous mbuf with some trailing
928 			 * space, use it to move the data into.
929 			 */
930 			m2 = m1;
931 			*mdp = m1 = havebuf;
932 			*cp2 = mtod(m1, char *) + m1->m_len;
933 		} else if (havebuf) {
934 			/*
935 			 * If the first mbuf has a external data
936 			 * and there is no previous empty mbuf
937 			 * allocate a new mbuf and move the external
938 			 * data to the new mbuf. Also make the first
939 			 * mbuf look empty.
940 			 */
941 			m2 = m1;
942 			*mdp = m1 = m_get(M_WAIT, MT_DATA);
943 			MCLAIM(m1, m2->m_owner);
944 			if ((m2->m_flags & M_PKTHDR) != 0) {
945 				/* XXX MOVE */
946 				M_COPY_PKTHDR(m1, m2);
947 				m_tag_delete_chain(m2, NULL);
948 				m2->m_flags &= ~M_PKTHDR;
949 			}
950 			if (havebuf) {
951 				havebuf->m_next = m1;
952 			}
953 			m1->m_next = m2;
954 			MRESETDATA(m1);
955 			m1->m_len = 0;
956 			m2->m_data = src;
957 			m2->m_len = left;
958 			*cp2 = mtod(m1, char *);
959 		} else {
960 			struct mbuf **nextp = &m1->m_next;
961 
962 			m1->m_len -= left;
963 			do {
964 				m2 = m_get(M_WAIT, MT_DATA);
965 				MCLAIM(m2, m1->m_owner);
966 				if (left >= MINCLSIZE) {
967 					MCLGET(m2, M_WAIT);
968 				}
969 				m2->m_next = *nextp;
970 				*nextp = m2;
971 				nextp = &m2->m_next;
972 				len = (m2->m_flags & M_EXT) != 0 ?
973 				    MCLBYTES : MLEN;
974 				if (len > left) {
975 					len = left;
976 				}
977 				memcpy(mtod(m2, char *), src, len);
978 				m2->m_len = len;
979 				src += len;
980 				left -= len;
981 			} while (left > 0);
982 			*mdp = m1 = m1->m_next;
983 			m2 = m1->m_next;
984 			*cp2 = mtod(m1, char *);
985 		}
986 	} else {
987 		/*
988 		 * If the first mbuf has no external data
989 		 * move the data to the front of the mbuf.
990 		 */
991 		MRESETDATA(m1);
992 		dst = mtod(m1, char *);
993 		if (dst != src) {
994 			memmove(dst, src, left);
995 		}
996 		m1->m_len = left;
997 		m2 = m1->m_next;
998 		*cp2 = m1->m_data;
999 	}
1000 	*dposp = *cp2 + siz;
1001 	/*
1002 	 * Loop through mbufs pulling data up into first mbuf until
1003 	 * the first mbuf is full or there is no more data to
1004 	 * pullup.
1005 	 */
1006 	dst = mtod(m1, char *) + m1->m_len;
1007 	while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) {
1008 		if ((len = min(len, m2->m_len)) != 0) {
1009 			memcpy(dst, mtod(m2, char *), len);
1010 		}
1011 		m1->m_len += len;
1012 		dst += len;
1013 		m2->m_data += len;
1014 		m2->m_len -= len;
1015 		m2 = m2->m_next;
1016 	}
1017 	if (m1->m_len < siz)
1018 		return (EBADRPC);
1019 	return (0);
1020 }
1021 
1022 /*
1023  * Advance the position in the mbuf chain.
1024  */
1025 int
nfs_adv(struct mbuf ** mdp,char ** dposp,int offs,int left)1026 nfs_adv(struct mbuf **mdp, char **dposp, int offs, int left)
1027 {
1028 	struct mbuf *m;
1029 	int s;
1030 
1031 	m = *mdp;
1032 	s = left;
1033 	while (s < offs) {
1034 		offs -= s;
1035 		m = m->m_next;
1036 		if (m == NULL)
1037 			return (EBADRPC);
1038 		s = m->m_len;
1039 	}
1040 	*mdp = m;
1041 	*dposp = mtod(m, char *) + offs;
1042 	return (0);
1043 }
1044 
1045 /*
1046  * Copy a string into mbufs for the hard cases...
1047  */
1048 int
nfsm_strtmbuf(struct mbuf ** mb,char ** bpos,const char * cp,long siz)1049 nfsm_strtmbuf(struct mbuf **mb, char **bpos, const char *cp, long siz)
1050 {
1051 	struct mbuf *m1 = NULL, *m2;
1052 	long left, xfer, len, tlen;
1053 	u_int32_t *tl;
1054 	int putsize;
1055 
1056 	putsize = 1;
1057 	m2 = *mb;
1058 	left = M_TRAILINGSPACE(m2);
1059 	if (left > 0) {
1060 		tl = ((u_int32_t *)(*bpos));
1061 		*tl++ = txdr_unsigned(siz);
1062 		putsize = 0;
1063 		left -= NFSX_UNSIGNED;
1064 		m2->m_len += NFSX_UNSIGNED;
1065 		if (left > 0) {
1066 			memcpy((void *) tl, cp, left);
1067 			siz -= left;
1068 			cp += left;
1069 			m2->m_len += left;
1070 			left = 0;
1071 		}
1072 	}
1073 	/* Loop around adding mbufs */
1074 	while (siz > 0) {
1075 		m1 = m_get(M_WAIT, MT_DATA);
1076 		MCLAIM(m1, &nfs_mowner);
1077 		if (siz > MLEN)
1078 			m_clget(m1, M_WAIT);
1079 		m1->m_len = NFSMSIZ(m1);
1080 		m2->m_next = m1;
1081 		m2 = m1;
1082 		tl = mtod(m1, u_int32_t *);
1083 		tlen = 0;
1084 		if (putsize) {
1085 			*tl++ = txdr_unsigned(siz);
1086 			m1->m_len -= NFSX_UNSIGNED;
1087 			tlen = NFSX_UNSIGNED;
1088 			putsize = 0;
1089 		}
1090 		if (siz < m1->m_len) {
1091 			len = nfsm_rndup(siz);
1092 			xfer = siz;
1093 			if (xfer < len)
1094 				*(tl+(xfer>>2)) = 0;
1095 		} else {
1096 			xfer = len = m1->m_len;
1097 		}
1098 		memcpy((void *) tl, cp, xfer);
1099 		m1->m_len = len+tlen;
1100 		siz -= xfer;
1101 		cp += xfer;
1102 	}
1103 	*mb = m1;
1104 	*bpos = mtod(m1, char *) + m1->m_len;
1105 	return (0);
1106 }
1107 
1108 /*
1109  * Directory caching routines. They work as follows:
1110  * - a cache is maintained per VDIR nfsnode.
1111  * - for each offset cookie that is exported to userspace, and can
1112  *   thus be thrown back at us as an offset to VOP_READDIR, store
1113  *   information in the cache.
1114  * - cached are:
1115  *   - cookie itself
1116  *   - blocknumber (essentially just a search key in the buffer cache)
1117  *   - entry number in block.
1118  *   - offset cookie of block in which this entry is stored
1119  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1120  * - entries are looked up in a hash table
1121  * - also maintained is an LRU list of entries, used to determine
1122  *   which ones to delete if the cache grows too large.
1123  * - if 32 <-> 64 translation mode is requested for a filesystem,
1124  *   the cache also functions as a translation table
1125  * - in the translation case, invalidating the cache does not mean
1126  *   flushing it, but just marking entries as invalid, except for
1127  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
1128  *   still be able to use the cache as a translation table.
1129  * - 32 bit cookies are uniquely created by combining the hash table
1130  *   entry value, and one generation count per hash table entry,
1131  *   incremented each time an entry is appended to the chain.
1132  * - the cache is invalidated each time a direcory is modified
1133  * - sanity checks are also done; if an entry in a block turns
1134  *   out not to have a matching cookie, the cache is invalidated
1135  *   and a new block starting from the wanted offset is fetched from
1136  *   the server.
1137  * - directory entries as read from the server are extended to contain
1138  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
1139  *   the cache and exporting them to userspace through the cookie
1140  *   argument to VOP_READDIR.
1141  */
1142 
1143 u_long
nfs_dirhash(off_t off)1144 nfs_dirhash(off_t off)
1145 {
1146 	int i;
1147 	char *cp = (char *)&off;
1148 	u_long sum = 0L;
1149 
1150 	for (i = 0 ; i < sizeof (off); i++)
1151 		sum += *cp++;
1152 
1153 	return sum;
1154 }
1155 
1156 #define	_NFSDC_MTX(np)		(NFSTOV(np)->v_interlock)
1157 #define	NFSDC_LOCK(np)		mutex_enter(_NFSDC_MTX(np))
1158 #define	NFSDC_UNLOCK(np)	mutex_exit(_NFSDC_MTX(np))
1159 #define	NFSDC_ASSERT_LOCKED(np) KASSERT(mutex_owned(_NFSDC_MTX(np)))
1160 
1161 void
nfs_initdircache(struct vnode * vp)1162 nfs_initdircache(struct vnode *vp)
1163 {
1164 	struct nfsnode *np = VTONFS(vp);
1165 	struct nfsdirhashhead *dircache;
1166 
1167 	dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, true,
1168 	    &nfsdirhashmask);
1169 
1170 	NFSDC_LOCK(np);
1171 	if (np->n_dircache == NULL) {
1172 		np->n_dircachesize = 0;
1173 		np->n_dircache = dircache;
1174 		dircache = NULL;
1175 		TAILQ_INIT(&np->n_dirchain);
1176 	}
1177 	NFSDC_UNLOCK(np);
1178 	if (dircache)
1179 		hashdone(dircache, HASH_LIST, nfsdirhashmask);
1180 }
1181 
1182 void
nfs_initdirxlatecookie(struct vnode * vp)1183 nfs_initdirxlatecookie(struct vnode *vp)
1184 {
1185 	struct nfsnode *np = VTONFS(vp);
1186 	unsigned *dirgens;
1187 
1188 	KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE);
1189 
1190 	dirgens = kmem_zalloc(NFS_DIRHASHSIZ * sizeof(unsigned), KM_SLEEP);
1191 	NFSDC_LOCK(np);
1192 	if (np->n_dirgens == NULL) {
1193 		np->n_dirgens = dirgens;
1194 		dirgens = NULL;
1195 	}
1196 	NFSDC_UNLOCK(np);
1197 	if (dirgens)
1198 		kmem_free(dirgens, NFS_DIRHASHSIZ * sizeof(unsigned));
1199 }
1200 
1201 static const struct nfsdircache dzero;
1202 
1203 static void nfs_unlinkdircache(struct nfsnode *np, struct nfsdircache *);
1204 static void nfs_putdircache_unlocked(struct nfsnode *,
1205     struct nfsdircache *);
1206 
1207 static void
nfs_unlinkdircache(struct nfsnode * np,struct nfsdircache * ndp)1208 nfs_unlinkdircache(struct nfsnode *np, struct nfsdircache *ndp)
1209 {
1210 
1211 	NFSDC_ASSERT_LOCKED(np);
1212 	KASSERT(ndp != &dzero);
1213 
1214 	if (LIST_NEXT(ndp, dc_hash) == (void *)-1)
1215 		return;
1216 
1217 	TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1218 	LIST_REMOVE(ndp, dc_hash);
1219 	LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */
1220 
1221 	nfs_putdircache_unlocked(np, ndp);
1222 }
1223 
1224 void
nfs_putdircache(struct nfsnode * np,struct nfsdircache * ndp)1225 nfs_putdircache(struct nfsnode *np, struct nfsdircache *ndp)
1226 {
1227 	int ref;
1228 
1229 	if (ndp == &dzero)
1230 		return;
1231 
1232 	KASSERT(ndp->dc_refcnt > 0);
1233 	NFSDC_LOCK(np);
1234 	ref = --ndp->dc_refcnt;
1235 	NFSDC_UNLOCK(np);
1236 
1237 	if (ref == 0)
1238 		kmem_free(ndp, sizeof(*ndp));
1239 }
1240 
1241 static void
nfs_putdircache_unlocked(struct nfsnode * np,struct nfsdircache * ndp)1242 nfs_putdircache_unlocked(struct nfsnode *np, struct nfsdircache *ndp)
1243 {
1244 	int ref;
1245 
1246 	NFSDC_ASSERT_LOCKED(np);
1247 
1248 	if (ndp == &dzero)
1249 		return;
1250 
1251 	KASSERT(ndp->dc_refcnt > 0);
1252 	ref = --ndp->dc_refcnt;
1253 	if (ref == 0)
1254 		kmem_free(ndp, sizeof(*ndp));
1255 }
1256 
1257 struct nfsdircache *
nfs_searchdircache(struct vnode * vp,off_t off,int do32,int * hashent)1258 nfs_searchdircache(struct vnode *vp, off_t off, int do32, int *hashent)
1259 {
1260 	struct nfsdirhashhead *ndhp;
1261 	struct nfsdircache *ndp = NULL;
1262 	struct nfsnode *np = VTONFS(vp);
1263 	unsigned ent;
1264 
1265 	/*
1266 	 * Zero is always a valid cookie.
1267 	 */
1268 	if (off == 0)
1269 		/* XXXUNCONST */
1270 		return (struct nfsdircache *)__UNCONST(&dzero);
1271 
1272 	if (!np->n_dircache)
1273 		return NULL;
1274 
1275 	/*
1276 	 * We use a 32bit cookie as search key, directly reconstruct
1277 	 * the hashentry. Else use the hashfunction.
1278 	 */
1279 	if (do32) {
1280 		ent = (u_int32_t)off >> 24;
1281 		if (ent >= NFS_DIRHASHSIZ)
1282 			return NULL;
1283 		ndhp = &np->n_dircache[ent];
1284 	} else {
1285 		ndhp = NFSDIRHASH(np, off);
1286 	}
1287 
1288 	if (hashent)
1289 		*hashent = (int)(ndhp - np->n_dircache);
1290 
1291 	NFSDC_LOCK(np);
1292 	if (do32) {
1293 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1294 			if (ndp->dc_cookie32 == (u_int32_t)off) {
1295 				/*
1296 				 * An invalidated entry will become the
1297 				 * start of a new block fetched from
1298 				 * the server.
1299 				 */
1300 				if (ndp->dc_flags & NFSDC_INVALID) {
1301 					ndp->dc_blkcookie = ndp->dc_cookie;
1302 					ndp->dc_entry = 0;
1303 					ndp->dc_flags &= ~NFSDC_INVALID;
1304 				}
1305 				break;
1306 			}
1307 		}
1308 	} else {
1309 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1310 			if (ndp->dc_cookie == off)
1311 				break;
1312 		}
1313 	}
1314 	if (ndp != NULL)
1315 		ndp->dc_refcnt++;
1316 	NFSDC_UNLOCK(np);
1317 	return ndp;
1318 }
1319 
1320 
1321 struct nfsdircache *
nfs_enterdircache(struct vnode * vp,off_t off,off_t blkoff,int en,daddr_t blkno)1322 nfs_enterdircache(struct vnode *vp, off_t off, off_t blkoff, int en,
1323     daddr_t blkno)
1324 {
1325 	struct nfsnode *np = VTONFS(vp);
1326 	struct nfsdirhashhead *ndhp;
1327 	struct nfsdircache *ndp = NULL;
1328 	struct nfsdircache *newndp = NULL;
1329 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1330 	int hashent = 0, gen, overwrite;	/* XXX: GCC */
1331 
1332 	/*
1333 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
1334 	 * cookie 0 for the '.' entry, making this necessary. This
1335 	 * isn't so bad, as 0 is a special case anyway.
1336 	 */
1337 	if (off == 0)
1338 		/* XXXUNCONST */
1339 		return (struct nfsdircache *)__UNCONST(&dzero);
1340 
1341 	if (!np->n_dircache)
1342 		/*
1343 		 * XXX would like to do this in nfs_nget but vtype
1344 		 * isn't known at that time.
1345 		 */
1346 		nfs_initdircache(vp);
1347 
1348 	if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens)
1349 		nfs_initdirxlatecookie(vp);
1350 
1351 retry:
1352 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
1353 
1354 	NFSDC_LOCK(np);
1355 	if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) {
1356 		/*
1357 		 * Overwriting an old entry. Check if it's the same.
1358 		 * If so, just return. If not, remove the old entry.
1359 		 */
1360 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1361 			goto done;
1362 		nfs_unlinkdircache(np, ndp);
1363 		nfs_putdircache_unlocked(np, ndp);
1364 		ndp = NULL;
1365 	}
1366 
1367 	ndhp = &np->n_dircache[hashent];
1368 
1369 	if (!ndp) {
1370 		if (newndp == NULL) {
1371 			NFSDC_UNLOCK(np);
1372 			newndp = kmem_alloc(sizeof(*newndp), KM_SLEEP);
1373 			newndp->dc_refcnt = 1;
1374 			LIST_NEXT(newndp, dc_hash) = (void *)-1;
1375 			goto retry;
1376 		}
1377 		ndp = newndp;
1378 		newndp = NULL;
1379 		overwrite = 0;
1380 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1381 			/*
1382 			 * We're allocating a new entry, so bump the
1383 			 * generation number.
1384 			 */
1385 			KASSERT(np->n_dirgens);
1386 			gen = ++np->n_dirgens[hashent];
1387 			if (gen == 0) {
1388 				np->n_dirgens[hashent]++;
1389 				gen++;
1390 			}
1391 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1392 		}
1393 	} else
1394 		overwrite = 1;
1395 
1396 	ndp->dc_cookie = off;
1397 	ndp->dc_blkcookie = blkoff;
1398 	ndp->dc_entry = en;
1399 	ndp->dc_flags = 0;
1400 
1401 	if (overwrite)
1402 		goto done;
1403 
1404 	/*
1405 	 * If the maximum directory cookie cache size has been reached
1406 	 * for this node, take one off the front. The idea is that
1407 	 * directories are typically read front-to-back once, so that
1408 	 * the oldest entries can be thrown away without much performance
1409 	 * loss.
1410 	 */
1411 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1412 		nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain));
1413 	} else
1414 		np->n_dircachesize++;
1415 
1416 	KASSERT(ndp->dc_refcnt == 1);
1417 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1418 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1419 	ndp->dc_refcnt++;
1420 done:
1421 	KASSERT(ndp->dc_refcnt > 0);
1422 	NFSDC_UNLOCK(np);
1423 	if (newndp)
1424 		nfs_putdircache(np, newndp);
1425 	return ndp;
1426 }
1427 
1428 void
nfs_invaldircache(struct vnode * vp,int flags)1429 nfs_invaldircache(struct vnode *vp, int flags)
1430 {
1431 	struct nfsnode *np = VTONFS(vp);
1432 	struct nfsdircache *ndp = NULL;
1433 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1434 	const bool forcefree = flags & NFS_INVALDIRCACHE_FORCE;
1435 
1436 #ifdef DIAGNOSTIC
1437 	if (vp->v_type != VDIR)
1438 		panic("nfs: invaldircache: not dir");
1439 #endif
1440 
1441 	if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0)
1442 		np->n_flag &= ~NEOFVALID;
1443 
1444 	if (!np->n_dircache)
1445 		return;
1446 
1447 	NFSDC_LOCK(np);
1448 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1449 		while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) {
1450 			KASSERT(!forcefree || ndp->dc_refcnt == 1);
1451 			nfs_unlinkdircache(np, ndp);
1452 		}
1453 		np->n_dircachesize = 0;
1454 		if (forcefree && np->n_dirgens) {
1455 			kmem_free(np->n_dirgens,
1456 			    NFS_DIRHASHSIZ * sizeof(unsigned));
1457 			np->n_dirgens = NULL;
1458 		}
1459 	} else {
1460 		TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain)
1461 			ndp->dc_flags |= NFSDC_INVALID;
1462 	}
1463 
1464 	NFSDC_UNLOCK(np);
1465 }
1466 
1467 /*
1468  * Called once before VFS init to initialize shared and
1469  * server-specific data structures.
1470  */
1471 static int
nfs_init0(void)1472 nfs_init0(void)
1473 {
1474 
1475 	nfsrtt.pos = 0;
1476 	rpc_vers = txdr_unsigned(RPC_VER2);
1477 	rpc_call = txdr_unsigned(RPC_CALL);
1478 	rpc_reply = txdr_unsigned(RPC_REPLY);
1479 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1480 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1481 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1482 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1483 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1484 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1485 	nfs_prog = txdr_unsigned(NFS_PROG);
1486 	nfs_true = txdr_unsigned(true);
1487 	nfs_false = txdr_unsigned(false);
1488 	nfs_xdrneg1 = txdr_unsigned(-1);
1489 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1490 	if (nfs_ticks < 1)
1491 		nfs_ticks = 1;
1492 	nfsdreq_init();
1493 
1494 	/*
1495 	 * Initialize reply list and start timer
1496 	 */
1497 	TAILQ_INIT(&nfs_reqq);
1498 	nfs_timer_init();
1499 	MOWNER_ATTACH(&nfs_mowner);
1500 
1501 	return 0;
1502 }
1503 
1504 static volatile uint32_t nfs_mutex;
1505 static uint32_t nfs_refcount;
1506 
1507 #define nfs_p()	while (atomic_cas_32(&nfs_mutex, 0, 1) == 0) continue;
1508 #define nfs_v()	while (atomic_cas_32(&nfs_mutex, 1, 0) == 1) continue;
1509 
1510 /*
1511  * This is disgusting, but it must support both modular and monolothic
1512  * configurations, plus the code is shared between server and client.
1513  * For monolithic builds NFSSERVER may not imply NFS. Unfortunately we
1514  * can't use regular mutexes here that would require static initialization
1515  * and we can get initialized from multiple places, so we improvise.
1516  *
1517  * Yuck.
1518  */
1519 void
nfs_init(void)1520 nfs_init(void)
1521 {
1522 
1523 	nfs_p();
1524 	if (nfs_refcount++ == 0)
1525 		nfs_init0();
1526 	nfs_v();
1527 }
1528 
1529 void
nfs_fini(void)1530 nfs_fini(void)
1531 {
1532 
1533 	nfs_p();
1534 	if (--nfs_refcount == 0) {
1535 		MOWNER_DETACH(&nfs_mowner);
1536 		nfs_timer_fini();
1537 		nfsdreq_fini();
1538 	}
1539 	nfs_v();
1540 }
1541 
1542 /*
1543  * A fiddled version of m_adj() that ensures null fill to a 32-bit
1544  * boundary and only trims off the back end
1545  *
1546  * 1. trim off 'len' bytes as m_adj(mp, -len).
1547  * 2. add zero-padding 'nul' bytes at the end of the mbuf chain.
1548  */
1549 void
nfs_zeropad(struct mbuf * mp,int len,int nul)1550 nfs_zeropad(struct mbuf *mp, int len, int nul)
1551 {
1552 	struct mbuf *m;
1553 	int count;
1554 
1555 	/*
1556 	 * Trim from tail.  Scan the mbuf chain,
1557 	 * calculating its length and finding the last mbuf.
1558 	 * If the adjustment only affects this mbuf, then just
1559 	 * adjust and return.  Otherwise, rescan and truncate
1560 	 * after the remaining size.
1561 	 */
1562 	count = 0;
1563 	m = mp;
1564 	for (;;) {
1565 		count += m->m_len;
1566 		if (m->m_next == NULL)
1567 			break;
1568 		m = m->m_next;
1569 	}
1570 
1571 	KDASSERT(count >= len);
1572 
1573 	if (m->m_len >= len) {
1574 		m->m_len -= len;
1575 	} else {
1576 		count -= len;
1577 		/*
1578 		 * Correct length for chain is "count".
1579 		 * Find the mbuf with last data, adjust its length,
1580 		 * and toss data from remaining mbufs on chain.
1581 		 */
1582 		for (m = mp; m; m = m->m_next) {
1583 			if (m->m_len >= count) {
1584 				m->m_len = count;
1585 				break;
1586 			}
1587 			count -= m->m_len;
1588 		}
1589 		KASSERT(m && m->m_next);
1590 		m_freem(m->m_next);
1591 		m->m_next = NULL;
1592 	}
1593 
1594 	KDASSERT(m->m_next == NULL);
1595 
1596 	/*
1597 	 * zero-padding.
1598 	 */
1599 	if (nul > 0) {
1600 		char *cp;
1601 		int i;
1602 
1603 		if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) {
1604 			struct mbuf *n;
1605 
1606 			KDASSERT(MLEN >= nul);
1607 			n = m_get(M_WAIT, MT_DATA);
1608 			MCLAIM(n, &nfs_mowner);
1609 			n->m_len = nul;
1610 			n->m_next = NULL;
1611 			m->m_next = n;
1612 			cp = mtod(n, void *);
1613 		} else {
1614 			cp = mtod(m, char *) + m->m_len;
1615 			m->m_len += nul;
1616 		}
1617 		for (i = 0; i < nul; i++)
1618 			*cp++ = '\0';
1619 	}
1620 	return;
1621 }
1622 
1623 /*
1624  * Make these functions instead of macros, so that the kernel text size
1625  * doesn't get too big...
1626  */
1627 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)1628 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)
1629 {
1630 	struct mbuf *mb = *mbp;
1631 	char *bpos = *bposp;
1632 	u_int32_t *tl;
1633 
1634 	if (before_ret) {
1635 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1636 		*tl = nfs_false;
1637 	} else {
1638 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
1639 		*tl++ = nfs_true;
1640 		txdr_hyper(before_vap->va_size, tl);
1641 		tl += 2;
1642 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
1643 		tl += 2;
1644 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
1645 	}
1646 	*bposp = bpos;
1647 	*mbp = mb;
1648 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
1649 }
1650 
1651 void
nfsm_srvpostopattr(struct nfsrv_descript * nfsd,int after_ret,struct vattr * after_vap,struct mbuf ** mbp,char ** bposp)1652 nfsm_srvpostopattr(struct nfsrv_descript *nfsd, int after_ret, struct vattr *after_vap, struct mbuf **mbp, char **bposp)
1653 {
1654 	struct mbuf *mb = *mbp;
1655 	char *bpos = *bposp;
1656 	u_int32_t *tl;
1657 	struct nfs_fattr *fp;
1658 
1659 	if (after_ret) {
1660 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
1661 		*tl = nfs_false;
1662 	} else {
1663 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
1664 		*tl++ = nfs_true;
1665 		fp = (struct nfs_fattr *)tl;
1666 		nfsm_srvfattr(nfsd, after_vap, fp);
1667 	}
1668 	*mbp = mb;
1669 	*bposp = bpos;
1670 }
1671 
1672 void
nfsm_srvfattr(struct nfsrv_descript * nfsd,struct vattr * vap,struct nfs_fattr * fp)1673 nfsm_srvfattr(struct nfsrv_descript *nfsd, struct vattr *vap, struct nfs_fattr *fp)
1674 {
1675 
1676 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
1677 	fp->fa_uid = txdr_unsigned(vap->va_uid);
1678 	fp->fa_gid = txdr_unsigned(vap->va_gid);
1679 	if (nfsd->nd_flag & ND_NFSV3) {
1680 		fp->fa_type = vtonfsv3_type(vap->va_type);
1681 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
1682 		txdr_hyper(vap->va_size, &fp->fa3_size);
1683 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
1684 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
1685 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
1686 		fp->fa3_fsid.nfsuquad[0] = 0;
1687 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
1688 		txdr_hyper(vap->va_fileid, &fp->fa3_fileid);
1689 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
1690 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
1691 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
1692 	} else {
1693 		fp->fa_type = vtonfsv2_type(vap->va_type);
1694 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1695 		fp->fa2_size = txdr_unsigned(vap->va_size);
1696 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
1697 		if (vap->va_type == VFIFO)
1698 			fp->fa2_rdev = 0xffffffff;
1699 		else
1700 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
1701 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
1702 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
1703 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
1704 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
1705 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
1706 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
1707 	}
1708 }
1709 
1710 /*
1711  * This function compares two net addresses by family and returns true
1712  * if they are the same host.
1713  * If there is any doubt, return false.
1714  * The AF_INET family is handled as a special case so that address mbufs
1715  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
1716  */
1717 int
netaddr_match(int family,union nethostaddr * haddr,struct mbuf * nam)1718 netaddr_match(int family, union nethostaddr *haddr, struct mbuf *nam)
1719 {
1720 	struct sockaddr_in *inetaddr;
1721 
1722 	switch (family) {
1723 	case AF_INET:
1724 		inetaddr = mtod(nam, struct sockaddr_in *);
1725 		if (inetaddr->sin_family == AF_INET &&
1726 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
1727 			return (1);
1728 		break;
1729 	case AF_INET6:
1730 	    {
1731 		struct sockaddr_in6 *sin6_1, *sin6_2;
1732 
1733 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
1734 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
1735 		if (sin6_1->sin6_family == AF_INET6 &&
1736 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
1737 			return 1;
1738 	    }
1739 	default:
1740 		break;
1741 	};
1742 	return (0);
1743 }
1744 
1745 struct nfs_clearcommit_ctx {
1746 	struct mount *mp;
1747 };
1748 
1749 static bool
nfs_clearcommit_selector(void * cl,struct vnode * vp)1750 nfs_clearcommit_selector(void *cl, struct vnode *vp)
1751 {
1752 	struct nfs_clearcommit_ctx *c = cl;
1753 	struct nfsnode *np;
1754 	struct vm_page *pg;
1755 
1756 	np = VTONFS(vp);
1757 	if (vp->v_type != VREG || vp->v_mount != c->mp || np == NULL)
1758 		return false;
1759 	np->n_pushlo = np->n_pushhi = np->n_pushedlo =
1760 	    np->n_pushedhi = 0;
1761 	np->n_commitflags &=
1762 	    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
1763 	TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq.queue) {
1764 		pg->flags &= ~PG_NEEDCOMMIT;
1765 	}
1766 	return false;
1767 }
1768 
1769 /*
1770  * The write verifier has changed (probably due to a server reboot), so all
1771  * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
1772  * as dirty or are being written out just now, all this takes is clearing
1773  * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
1774  * the mount point.
1775  */
1776 void
nfs_clearcommit(struct mount * mp)1777 nfs_clearcommit(struct mount *mp)
1778 {
1779 	struct vnode *vp __diagused;
1780 	struct vnode_iterator *marker;
1781 	struct nfsmount *nmp = VFSTONFS(mp);
1782 	struct nfs_clearcommit_ctx ctx;
1783 
1784 	rw_enter(&nmp->nm_writeverflock, RW_WRITER);
1785 	vfs_vnode_iterator_init(mp, &marker);
1786 	ctx.mp = mp;
1787 	vp = vfs_vnode_iterator_next(marker, nfs_clearcommit_selector, &ctx);
1788 	KASSERT(vp == NULL);
1789 	vfs_vnode_iterator_destroy(marker);
1790 	mutex_enter(&nmp->nm_lock);
1791 	nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
1792 	mutex_exit(&nmp->nm_lock);
1793 	rw_exit(&nmp->nm_writeverflock);
1794 }
1795 
1796 void
nfs_merge_commit_ranges(struct vnode * vp)1797 nfs_merge_commit_ranges(struct vnode *vp)
1798 {
1799 	struct nfsnode *np = VTONFS(vp);
1800 
1801 	KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
1802 
1803 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
1804 		np->n_pushedlo = np->n_pushlo;
1805 		np->n_pushedhi = np->n_pushhi;
1806 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
1807 	} else {
1808 		if (np->n_pushlo < np->n_pushedlo)
1809 			np->n_pushedlo = np->n_pushlo;
1810 		if (np->n_pushhi > np->n_pushedhi)
1811 			np->n_pushedhi = np->n_pushhi;
1812 	}
1813 
1814 	np->n_pushlo = np->n_pushhi = 0;
1815 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
1816 
1817 #ifdef NFS_DEBUG_COMMIT
1818 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
1819 	    (unsigned)np->n_pushedhi);
1820 #endif
1821 }
1822 
1823 int
nfs_in_committed_range(struct vnode * vp,off_t off,off_t len)1824 nfs_in_committed_range(struct vnode *vp, off_t off, off_t len)
1825 {
1826 	struct nfsnode *np = VTONFS(vp);
1827 	off_t lo, hi;
1828 
1829 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
1830 		return 0;
1831 	lo = off;
1832 	hi = lo + len;
1833 
1834 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
1835 }
1836 
1837 int
nfs_in_tobecommitted_range(struct vnode * vp,off_t off,off_t len)1838 nfs_in_tobecommitted_range(struct vnode *vp, off_t off, off_t len)
1839 {
1840 	struct nfsnode *np = VTONFS(vp);
1841 	off_t lo, hi;
1842 
1843 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
1844 		return 0;
1845 	lo = off;
1846 	hi = lo + len;
1847 
1848 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
1849 }
1850 
1851 void
nfs_add_committed_range(struct vnode * vp,off_t off,off_t len)1852 nfs_add_committed_range(struct vnode *vp, off_t off, off_t len)
1853 {
1854 	struct nfsnode *np = VTONFS(vp);
1855 	off_t lo, hi;
1856 
1857 	lo = off;
1858 	hi = lo + len;
1859 
1860 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
1861 		np->n_pushedlo = lo;
1862 		np->n_pushedhi = hi;
1863 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
1864 	} else {
1865 		if (hi > np->n_pushedhi)
1866 			np->n_pushedhi = hi;
1867 		if (lo < np->n_pushedlo)
1868 			np->n_pushedlo = lo;
1869 	}
1870 #ifdef NFS_DEBUG_COMMIT
1871 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
1872 	    (unsigned)np->n_pushedhi);
1873 #endif
1874 }
1875 
1876 void
nfs_del_committed_range(struct vnode * vp,off_t off,off_t len)1877 nfs_del_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 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
1883 		return;
1884 
1885 	lo = off;
1886 	hi = lo + len;
1887 
1888 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
1889 		return;
1890 	if (lo <= np->n_pushedlo)
1891 		np->n_pushedlo = hi;
1892 	else if (hi >= np->n_pushedhi)
1893 		np->n_pushedhi = lo;
1894 	else {
1895 		/*
1896 		 * XXX There's only one range. If the deleted range
1897 		 * is in the middle, pick the largest of the
1898 		 * contiguous ranges that it leaves.
1899 		 */
1900 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
1901 			np->n_pushedhi = lo;
1902 		else
1903 			np->n_pushedlo = hi;
1904 	}
1905 #ifdef NFS_DEBUG_COMMIT
1906 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
1907 	    (unsigned)np->n_pushedhi);
1908 #endif
1909 }
1910 
1911 void
nfs_add_tobecommitted_range(struct vnode * vp,off_t off,off_t len)1912 nfs_add_tobecommitted_range(struct vnode *vp, off_t off, off_t len)
1913 {
1914 	struct nfsnode *np = VTONFS(vp);
1915 	off_t lo, hi;
1916 
1917 	lo = off;
1918 	hi = lo + len;
1919 
1920 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
1921 		np->n_pushlo = lo;
1922 		np->n_pushhi = hi;
1923 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
1924 	} else {
1925 		if (lo < np->n_pushlo)
1926 			np->n_pushlo = lo;
1927 		if (hi > np->n_pushhi)
1928 			np->n_pushhi = hi;
1929 	}
1930 #ifdef NFS_DEBUG_COMMIT
1931 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
1932 	    (unsigned)np->n_pushhi);
1933 #endif
1934 }
1935 
1936 void
nfs_del_tobecommitted_range(struct vnode * vp,off_t off,off_t len)1937 nfs_del_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 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
1943 		return;
1944 
1945 	lo = off;
1946 	hi = lo + len;
1947 
1948 	if (lo > np->n_pushhi || hi < np->n_pushlo)
1949 		return;
1950 
1951 	if (lo <= np->n_pushlo)
1952 		np->n_pushlo = hi;
1953 	else if (hi >= np->n_pushhi)
1954 		np->n_pushhi = lo;
1955 	else {
1956 		/*
1957 		 * XXX There's only one range. If the deleted range
1958 		 * is in the middle, pick the largest of the
1959 		 * contiguous ranges that it leaves.
1960 		 */
1961 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
1962 			np->n_pushhi = lo;
1963 		else
1964 			np->n_pushlo = hi;
1965 	}
1966 #ifdef NFS_DEBUG_COMMIT
1967 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
1968 	    (unsigned)np->n_pushhi);
1969 #endif
1970 }
1971 
1972 /*
1973  * Map errnos to NFS error numbers. For Version 3 also filter out error
1974  * numbers not specified for the associated procedure.
1975  */
1976 int
nfsrv_errmap(struct nfsrv_descript * nd,int err)1977 nfsrv_errmap(struct nfsrv_descript *nd, int err)
1978 {
1979 	const short *defaulterrp, *errp;
1980 
1981 	if (nd->nd_flag & ND_NFSV3) {
1982 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
1983 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
1984 		while (*++errp) {
1985 			if (*errp == err)
1986 				return (err);
1987 			else if (*errp > err)
1988 				break;
1989 		}
1990 		return ((int)*defaulterrp);
1991 	    } else
1992 		return (err & 0xffff);
1993 	}
1994 	if (err <= ELAST)
1995 		return ((int)nfsrv_v2errmap[err - 1]);
1996 	return (NFSERR_IO);
1997 }
1998 
1999 u_int32_t
nfs_getxid(void)2000 nfs_getxid(void)
2001 {
2002 	u_int32_t newxid;
2003 
2004 	if (__predict_false(nfs_xid == 0)) {
2005 		nfs_xid = cprng_fast32();
2006 	}
2007 
2008 	/* get next xid.  skip 0 */
2009 	do {
2010 		newxid = atomic_inc_32_nv(&nfs_xid);
2011 	} while (__predict_false(newxid == 0));
2012 
2013 	return txdr_unsigned(newxid);
2014 }
2015 
2016 /*
2017  * assign a new xid for existing request.
2018  * used for NFSERR_JUKEBOX handling.
2019  */
2020 void
nfs_renewxid(struct nfsreq * req)2021 nfs_renewxid(struct nfsreq *req)
2022 {
2023 	u_int32_t xid;
2024 	int off;
2025 
2026 	xid = nfs_getxid();
2027 	if (req->r_nmp->nm_sotype == SOCK_STREAM)
2028 		off = sizeof(u_int32_t); /* RPC record mark */
2029 	else
2030 		off = 0;
2031 
2032 	m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid);
2033 	req->r_xid = xid;
2034 }
2035