Home | History | Annotate | Line # | Download | only in nfs
nfs_subs.c revision 1.218.2.1
      1 /*	$NetBSD: nfs_subs.c,v 1.218.2.1 2010/04/30 14:44:22 uebayasi 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 1.218.2.1 2010/04/30 14:44:22 uebayasi 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 
    102 #include <uvm/uvm_extern.h>
    103 
    104 #include <nfs/rpcv2.h>
    105 #include <nfs/nfsproto.h>
    106 #include <nfs/nfsnode.h>
    107 #include <nfs/nfs.h>
    108 #include <nfs/xdr_subs.h>
    109 #include <nfs/nfsm_subs.h>
    110 #include <nfs/nfsmount.h>
    111 #include <nfs/nfsrtt.h>
    112 #include <nfs/nfs_var.h>
    113 
    114 #include <miscfs/specfs/specdev.h>
    115 
    116 #include <netinet/in.h>
    117 
    118 static u_int32_t nfs_xid;
    119 
    120 int nuidhash_max = NFS_MAXUIDHASH;
    121 /*
    122  * Data items converted to xdr at startup, since they are constant
    123  * This is kinda hokey, but may save a little time doing byte swaps
    124  */
    125 u_int32_t nfs_xdrneg1;
    126 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
    127 	rpc_mismatch, rpc_auth_unix, rpc_msgaccepted,
    128 	rpc_auth_kerb;
    129 u_int32_t nfs_prog, nfs_true, nfs_false;
    130 
    131 /* And other global data */
    132 const nfstype nfsv2_type[9] =
    133 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON };
    134 const nfstype nfsv3_type[9] =
    135 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON };
    136 const enum vtype nv2tov_type[8] =
    137 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON };
    138 const enum vtype nv3tov_type[8] =
    139 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO };
    140 int nfs_ticks;
    141 
    142 /* NFS client/server stats. */
    143 struct nfsstats nfsstats;
    144 
    145 /*
    146  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
    147  */
    148 const int nfsv3_procid[NFS_NPROCS] = {
    149 	NFSPROC_NULL,
    150 	NFSPROC_GETATTR,
    151 	NFSPROC_SETATTR,
    152 	NFSPROC_NOOP,
    153 	NFSPROC_LOOKUP,
    154 	NFSPROC_READLINK,
    155 	NFSPROC_READ,
    156 	NFSPROC_NOOP,
    157 	NFSPROC_WRITE,
    158 	NFSPROC_CREATE,
    159 	NFSPROC_REMOVE,
    160 	NFSPROC_RENAME,
    161 	NFSPROC_LINK,
    162 	NFSPROC_SYMLINK,
    163 	NFSPROC_MKDIR,
    164 	NFSPROC_RMDIR,
    165 	NFSPROC_READDIR,
    166 	NFSPROC_FSSTAT,
    167 	NFSPROC_NOOP,
    168 	NFSPROC_NOOP,
    169 	NFSPROC_NOOP,
    170 	NFSPROC_NOOP,
    171 	NFSPROC_NOOP
    172 };
    173 
    174 /*
    175  * and the reverse mapping from generic to Version 2 procedure numbers
    176  */
    177 const int nfsv2_procid[NFS_NPROCS] = {
    178 	NFSV2PROC_NULL,
    179 	NFSV2PROC_GETATTR,
    180 	NFSV2PROC_SETATTR,
    181 	NFSV2PROC_LOOKUP,
    182 	NFSV2PROC_NOOP,
    183 	NFSV2PROC_READLINK,
    184 	NFSV2PROC_READ,
    185 	NFSV2PROC_WRITE,
    186 	NFSV2PROC_CREATE,
    187 	NFSV2PROC_MKDIR,
    188 	NFSV2PROC_SYMLINK,
    189 	NFSV2PROC_CREATE,
    190 	NFSV2PROC_REMOVE,
    191 	NFSV2PROC_RMDIR,
    192 	NFSV2PROC_RENAME,
    193 	NFSV2PROC_LINK,
    194 	NFSV2PROC_READDIR,
    195 	NFSV2PROC_NOOP,
    196 	NFSV2PROC_STATFS,
    197 	NFSV2PROC_NOOP,
    198 	NFSV2PROC_NOOP,
    199 	NFSV2PROC_NOOP,
    200 	NFSV2PROC_NOOP,
    201 };
    202 
    203 /*
    204  * Maps errno values to nfs error numbers.
    205  * Use NFSERR_IO as the catch all for ones not specifically defined in
    206  * RFC 1094.
    207  */
    208 static const u_char nfsrv_v2errmap[ELAST] = {
    209   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
    210   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
    211   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
    212   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
    213   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
    214   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
    215   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
    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_NAMETOL,	NFSERR_IO,	NFSERR_IO,
    222   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
    223   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
    224   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
    225   NFSERR_IO,	NFSERR_IO,
    226 };
    227 
    228 /*
    229  * Maps errno values to nfs error numbers.
    230  * Although it is not obvious whether or not NFS clients really care if
    231  * a returned error value is in the specified list for the procedure, the
    232  * safest thing to do is filter them appropriately. For Version 2, the
    233  * X/Open XNFS document is the only specification that defines error values
    234  * for each RPC (The RFC simply lists all possible error values for all RPCs),
    235  * so I have decided to not do this for Version 2.
    236  * The first entry is the default error return and the rest are the valid
    237  * errors for that RPC in increasing numeric order.
    238  */
    239 static const short nfsv3err_null[] = {
    240 	0,
    241 	0,
    242 };
    243 
    244 static const short nfsv3err_getattr[] = {
    245 	NFSERR_IO,
    246 	NFSERR_IO,
    247 	NFSERR_STALE,
    248 	NFSERR_BADHANDLE,
    249 	NFSERR_SERVERFAULT,
    250 	0,
    251 };
    252 
    253 static const short nfsv3err_setattr[] = {
    254 	NFSERR_IO,
    255 	NFSERR_PERM,
    256 	NFSERR_IO,
    257 	NFSERR_ACCES,
    258 	NFSERR_INVAL,
    259 	NFSERR_NOSPC,
    260 	NFSERR_ROFS,
    261 	NFSERR_DQUOT,
    262 	NFSERR_STALE,
    263 	NFSERR_BADHANDLE,
    264 	NFSERR_NOT_SYNC,
    265 	NFSERR_SERVERFAULT,
    266 	0,
    267 };
    268 
    269 static const short nfsv3err_lookup[] = {
    270 	NFSERR_IO,
    271 	NFSERR_NOENT,
    272 	NFSERR_IO,
    273 	NFSERR_ACCES,
    274 	NFSERR_NOTDIR,
    275 	NFSERR_NAMETOL,
    276 	NFSERR_STALE,
    277 	NFSERR_BADHANDLE,
    278 	NFSERR_SERVERFAULT,
    279 	0,
    280 };
    281 
    282 static const short nfsv3err_access[] = {
    283 	NFSERR_IO,
    284 	NFSERR_IO,
    285 	NFSERR_STALE,
    286 	NFSERR_BADHANDLE,
    287 	NFSERR_SERVERFAULT,
    288 	0,
    289 };
    290 
    291 static const short nfsv3err_readlink[] = {
    292 	NFSERR_IO,
    293 	NFSERR_IO,
    294 	NFSERR_ACCES,
    295 	NFSERR_INVAL,
    296 	NFSERR_STALE,
    297 	NFSERR_BADHANDLE,
    298 	NFSERR_NOTSUPP,
    299 	NFSERR_SERVERFAULT,
    300 	0,
    301 };
    302 
    303 static const short nfsv3err_read[] = {
    304 	NFSERR_IO,
    305 	NFSERR_IO,
    306 	NFSERR_NXIO,
    307 	NFSERR_ACCES,
    308 	NFSERR_INVAL,
    309 	NFSERR_STALE,
    310 	NFSERR_BADHANDLE,
    311 	NFSERR_SERVERFAULT,
    312 	NFSERR_JUKEBOX,
    313 	0,
    314 };
    315 
    316 static const short nfsv3err_write[] = {
    317 	NFSERR_IO,
    318 	NFSERR_IO,
    319 	NFSERR_ACCES,
    320 	NFSERR_INVAL,
    321 	NFSERR_FBIG,
    322 	NFSERR_NOSPC,
    323 	NFSERR_ROFS,
    324 	NFSERR_DQUOT,
    325 	NFSERR_STALE,
    326 	NFSERR_BADHANDLE,
    327 	NFSERR_SERVERFAULT,
    328 	NFSERR_JUKEBOX,
    329 	0,
    330 };
    331 
    332 static const short nfsv3err_create[] = {
    333 	NFSERR_IO,
    334 	NFSERR_IO,
    335 	NFSERR_ACCES,
    336 	NFSERR_EXIST,
    337 	NFSERR_NOTDIR,
    338 	NFSERR_NOSPC,
    339 	NFSERR_ROFS,
    340 	NFSERR_NAMETOL,
    341 	NFSERR_DQUOT,
    342 	NFSERR_STALE,
    343 	NFSERR_BADHANDLE,
    344 	NFSERR_NOTSUPP,
    345 	NFSERR_SERVERFAULT,
    346 	0,
    347 };
    348 
    349 static const short nfsv3err_mkdir[] = {
    350 	NFSERR_IO,
    351 	NFSERR_IO,
    352 	NFSERR_ACCES,
    353 	NFSERR_EXIST,
    354 	NFSERR_NOTDIR,
    355 	NFSERR_NOSPC,
    356 	NFSERR_ROFS,
    357 	NFSERR_NAMETOL,
    358 	NFSERR_DQUOT,
    359 	NFSERR_STALE,
    360 	NFSERR_BADHANDLE,
    361 	NFSERR_NOTSUPP,
    362 	NFSERR_SERVERFAULT,
    363 	0,
    364 };
    365 
    366 static const short nfsv3err_symlink[] = {
    367 	NFSERR_IO,
    368 	NFSERR_IO,
    369 	NFSERR_ACCES,
    370 	NFSERR_EXIST,
    371 	NFSERR_NOTDIR,
    372 	NFSERR_NOSPC,
    373 	NFSERR_ROFS,
    374 	NFSERR_NAMETOL,
    375 	NFSERR_DQUOT,
    376 	NFSERR_STALE,
    377 	NFSERR_BADHANDLE,
    378 	NFSERR_NOTSUPP,
    379 	NFSERR_SERVERFAULT,
    380 	0,
    381 };
    382 
    383 static const short nfsv3err_mknod[] = {
    384 	NFSERR_IO,
    385 	NFSERR_IO,
    386 	NFSERR_ACCES,
    387 	NFSERR_EXIST,
    388 	NFSERR_NOTDIR,
    389 	NFSERR_NOSPC,
    390 	NFSERR_ROFS,
    391 	NFSERR_NAMETOL,
    392 	NFSERR_DQUOT,
    393 	NFSERR_STALE,
    394 	NFSERR_BADHANDLE,
    395 	NFSERR_NOTSUPP,
    396 	NFSERR_SERVERFAULT,
    397 	NFSERR_BADTYPE,
    398 	0,
    399 };
    400 
    401 static const short nfsv3err_remove[] = {
    402 	NFSERR_IO,
    403 	NFSERR_NOENT,
    404 	NFSERR_IO,
    405 	NFSERR_ACCES,
    406 	NFSERR_NOTDIR,
    407 	NFSERR_ROFS,
    408 	NFSERR_NAMETOL,
    409 	NFSERR_STALE,
    410 	NFSERR_BADHANDLE,
    411 	NFSERR_SERVERFAULT,
    412 	0,
    413 };
    414 
    415 static const short nfsv3err_rmdir[] = {
    416 	NFSERR_IO,
    417 	NFSERR_NOENT,
    418 	NFSERR_IO,
    419 	NFSERR_ACCES,
    420 	NFSERR_EXIST,
    421 	NFSERR_NOTDIR,
    422 	NFSERR_INVAL,
    423 	NFSERR_ROFS,
    424 	NFSERR_NAMETOL,
    425 	NFSERR_NOTEMPTY,
    426 	NFSERR_STALE,
    427 	NFSERR_BADHANDLE,
    428 	NFSERR_NOTSUPP,
    429 	NFSERR_SERVERFAULT,
    430 	0,
    431 };
    432 
    433 static const short nfsv3err_rename[] = {
    434 	NFSERR_IO,
    435 	NFSERR_NOENT,
    436 	NFSERR_IO,
    437 	NFSERR_ACCES,
    438 	NFSERR_EXIST,
    439 	NFSERR_XDEV,
    440 	NFSERR_NOTDIR,
    441 	NFSERR_ISDIR,
    442 	NFSERR_INVAL,
    443 	NFSERR_NOSPC,
    444 	NFSERR_ROFS,
    445 	NFSERR_MLINK,
    446 	NFSERR_NAMETOL,
    447 	NFSERR_NOTEMPTY,
    448 	NFSERR_DQUOT,
    449 	NFSERR_STALE,
    450 	NFSERR_BADHANDLE,
    451 	NFSERR_NOTSUPP,
    452 	NFSERR_SERVERFAULT,
    453 	0,
    454 };
    455 
    456 static const short nfsv3err_link[] = {
    457 	NFSERR_IO,
    458 	NFSERR_IO,
    459 	NFSERR_ACCES,
    460 	NFSERR_EXIST,
    461 	NFSERR_XDEV,
    462 	NFSERR_NOTDIR,
    463 	NFSERR_INVAL,
    464 	NFSERR_NOSPC,
    465 	NFSERR_ROFS,
    466 	NFSERR_MLINK,
    467 	NFSERR_NAMETOL,
    468 	NFSERR_DQUOT,
    469 	NFSERR_STALE,
    470 	NFSERR_BADHANDLE,
    471 	NFSERR_NOTSUPP,
    472 	NFSERR_SERVERFAULT,
    473 	0,
    474 };
    475 
    476 static const short nfsv3err_readdir[] = {
    477 	NFSERR_IO,
    478 	NFSERR_IO,
    479 	NFSERR_ACCES,
    480 	NFSERR_NOTDIR,
    481 	NFSERR_STALE,
    482 	NFSERR_BADHANDLE,
    483 	NFSERR_BAD_COOKIE,
    484 	NFSERR_TOOSMALL,
    485 	NFSERR_SERVERFAULT,
    486 	0,
    487 };
    488 
    489 static const short nfsv3err_readdirplus[] = {
    490 	NFSERR_IO,
    491 	NFSERR_IO,
    492 	NFSERR_ACCES,
    493 	NFSERR_NOTDIR,
    494 	NFSERR_STALE,
    495 	NFSERR_BADHANDLE,
    496 	NFSERR_BAD_COOKIE,
    497 	NFSERR_NOTSUPP,
    498 	NFSERR_TOOSMALL,
    499 	NFSERR_SERVERFAULT,
    500 	0,
    501 };
    502 
    503 static const short nfsv3err_fsstat[] = {
    504 	NFSERR_IO,
    505 	NFSERR_IO,
    506 	NFSERR_STALE,
    507 	NFSERR_BADHANDLE,
    508 	NFSERR_SERVERFAULT,
    509 	0,
    510 };
    511 
    512 static const short nfsv3err_fsinfo[] = {
    513 	NFSERR_STALE,
    514 	NFSERR_STALE,
    515 	NFSERR_BADHANDLE,
    516 	NFSERR_SERVERFAULT,
    517 	0,
    518 };
    519 
    520 static const short nfsv3err_pathconf[] = {
    521 	NFSERR_STALE,
    522 	NFSERR_STALE,
    523 	NFSERR_BADHANDLE,
    524 	NFSERR_SERVERFAULT,
    525 	0,
    526 };
    527 
    528 static const short nfsv3err_commit[] = {
    529 	NFSERR_IO,
    530 	NFSERR_IO,
    531 	NFSERR_STALE,
    532 	NFSERR_BADHANDLE,
    533 	NFSERR_SERVERFAULT,
    534 	0,
    535 };
    536 
    537 static const short * const nfsrv_v3errmap[] = {
    538 	nfsv3err_null,
    539 	nfsv3err_getattr,
    540 	nfsv3err_setattr,
    541 	nfsv3err_lookup,
    542 	nfsv3err_access,
    543 	nfsv3err_readlink,
    544 	nfsv3err_read,
    545 	nfsv3err_write,
    546 	nfsv3err_create,
    547 	nfsv3err_mkdir,
    548 	nfsv3err_symlink,
    549 	nfsv3err_mknod,
    550 	nfsv3err_remove,
    551 	nfsv3err_rmdir,
    552 	nfsv3err_rename,
    553 	nfsv3err_link,
    554 	nfsv3err_readdir,
    555 	nfsv3err_readdirplus,
    556 	nfsv3err_fsstat,
    557 	nfsv3err_fsinfo,
    558 	nfsv3err_pathconf,
    559 	nfsv3err_commit,
    560 };
    561 
    562 extern struct nfsrtt nfsrtt;
    563 
    564 u_long nfsdirhashmask;
    565 
    566 int nfs_webnamei(struct nameidata *, struct vnode *, struct proc *);
    567 
    568 /*
    569  * Create the header for an rpc request packet
    570  * The hsiz is the size of the rest of the nfs request header.
    571  * (just used to decide if a cluster is a good idea)
    572  */
    573 struct mbuf *
    574 nfsm_reqh(struct nfsnode *np, u_long procid, int hsiz, char **bposp)
    575 {
    576 	struct mbuf *mb;
    577 	char *bpos;
    578 
    579 	mb = m_get(M_WAIT, MT_DATA);
    580 	MCLAIM(mb, &nfs_mowner);
    581 	if (hsiz >= MINCLSIZE)
    582 		m_clget(mb, M_WAIT);
    583 	mb->m_len = 0;
    584 	bpos = mtod(mb, void *);
    585 
    586 	/* Finally, return values */
    587 	*bposp = bpos;
    588 	return (mb);
    589 }
    590 
    591 /*
    592  * Build the RPC header and fill in the authorization info.
    593  * The authorization string argument is only used when the credentials
    594  * come from outside of the kernel.
    595  * Returns the head of the mbuf list.
    596  */
    597 struct mbuf *
    598 nfsm_rpchead(kauth_cred_t cr, int nmflag, int procid,
    599 	int auth_type, int auth_len, char *auth_str, int verf_len,
    600 	char *verf_str, struct mbuf *mrest, int mrest_len,
    601 	struct mbuf **mbp, uint32_t *xidp)
    602 {
    603 	struct mbuf *mb;
    604 	u_int32_t *tl;
    605 	char *bpos;
    606 	int i;
    607 	struct mbuf *mreq;
    608 	int siz, grpsiz, authsiz;
    609 
    610 	authsiz = nfsm_rndup(auth_len);
    611 	mb = m_gethdr(M_WAIT, MT_DATA);
    612 	MCLAIM(mb, &nfs_mowner);
    613 	if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
    614 		m_clget(mb, M_WAIT);
    615 	} else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
    616 		MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
    617 	} else {
    618 		MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
    619 	}
    620 	mb->m_len = 0;
    621 	mreq = mb;
    622 	bpos = mtod(mb, void *);
    623 
    624 	/*
    625 	 * First the RPC header.
    626 	 */
    627 	nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
    628 
    629 	*tl++ = *xidp = nfs_getxid();
    630 	*tl++ = rpc_call;
    631 	*tl++ = rpc_vers;
    632 	*tl++ = txdr_unsigned(NFS_PROG);
    633 	if (nmflag & NFSMNT_NFSV3)
    634 		*tl++ = txdr_unsigned(NFS_VER3);
    635 	else
    636 		*tl++ = txdr_unsigned(NFS_VER2);
    637 	if (nmflag & NFSMNT_NFSV3)
    638 		*tl++ = txdr_unsigned(procid);
    639 	else
    640 		*tl++ = txdr_unsigned(nfsv2_procid[procid]);
    641 
    642 	/*
    643 	 * And then the authorization cred.
    644 	 */
    645 	*tl++ = txdr_unsigned(auth_type);
    646 	*tl = txdr_unsigned(authsiz);
    647 	switch (auth_type) {
    648 	case RPCAUTH_UNIX:
    649 		nfsm_build(tl, u_int32_t *, auth_len);
    650 		*tl++ = 0;		/* stamp ?? */
    651 		*tl++ = 0;		/* NULL hostname */
    652 		*tl++ = txdr_unsigned(kauth_cred_geteuid(cr));
    653 		*tl++ = txdr_unsigned(kauth_cred_getegid(cr));
    654 		grpsiz = (auth_len >> 2) - 5;
    655 		*tl++ = txdr_unsigned(grpsiz);
    656 		for (i = 0; i < grpsiz; i++)
    657 			*tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */
    658 		break;
    659 	case RPCAUTH_KERB4:
    660 		siz = auth_len;
    661 		while (siz > 0) {
    662 			if (M_TRAILINGSPACE(mb) == 0) {
    663 				struct mbuf *mb2;
    664 				mb2 = m_get(M_WAIT, MT_DATA);
    665 				MCLAIM(mb2, &nfs_mowner);
    666 				if (siz >= MINCLSIZE)
    667 					m_clget(mb2, M_WAIT);
    668 				mb->m_next = mb2;
    669 				mb = mb2;
    670 				mb->m_len = 0;
    671 				bpos = mtod(mb, void *);
    672 			}
    673 			i = min(siz, M_TRAILINGSPACE(mb));
    674 			memcpy(bpos, auth_str, i);
    675 			mb->m_len += i;
    676 			auth_str += i;
    677 			bpos += i;
    678 			siz -= i;
    679 		}
    680 		if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
    681 			for (i = 0; i < siz; i++)
    682 				*bpos++ = '\0';
    683 			mb->m_len += siz;
    684 		}
    685 		break;
    686 	};
    687 
    688 	/*
    689 	 * And the verifier...
    690 	 */
    691 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
    692 	if (verf_str) {
    693 		*tl++ = txdr_unsigned(RPCAUTH_KERB4);
    694 		*tl = txdr_unsigned(verf_len);
    695 		siz = verf_len;
    696 		while (siz > 0) {
    697 			if (M_TRAILINGSPACE(mb) == 0) {
    698 				struct mbuf *mb2;
    699 				mb2 = m_get(M_WAIT, MT_DATA);
    700 				MCLAIM(mb2, &nfs_mowner);
    701 				if (siz >= MINCLSIZE)
    702 					m_clget(mb2, M_WAIT);
    703 				mb->m_next = mb2;
    704 				mb = mb2;
    705 				mb->m_len = 0;
    706 				bpos = mtod(mb, void *);
    707 			}
    708 			i = min(siz, M_TRAILINGSPACE(mb));
    709 			memcpy(bpos, verf_str, i);
    710 			mb->m_len += i;
    711 			verf_str += i;
    712 			bpos += i;
    713 			siz -= i;
    714 		}
    715 		if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
    716 			for (i = 0; i < siz; i++)
    717 				*bpos++ = '\0';
    718 			mb->m_len += siz;
    719 		}
    720 	} else {
    721 		*tl++ = txdr_unsigned(RPCAUTH_NULL);
    722 		*tl = 0;
    723 	}
    724 	mb->m_next = mrest;
    725 	mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
    726 	mreq->m_pkthdr.rcvif = (struct ifnet *)0;
    727 	*mbp = mb;
    728 	return (mreq);
    729 }
    730 
    731 /*
    732  * copies mbuf chain to the uio scatter/gather list
    733  */
    734 int
    735 nfsm_mbuftouio(struct mbuf **mrep, struct uio *uiop, int siz, char **dpos)
    736 {
    737 	char *mbufcp, *uiocp;
    738 	int xfer, left, len;
    739 	struct mbuf *mp;
    740 	long uiosiz, rem;
    741 	int error = 0;
    742 
    743 	mp = *mrep;
    744 	mbufcp = *dpos;
    745 	len = mtod(mp, char *) + mp->m_len - mbufcp;
    746 	rem = nfsm_rndup(siz)-siz;
    747 	while (siz > 0) {
    748 		if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
    749 			return (EFBIG);
    750 		left = uiop->uio_iov->iov_len;
    751 		uiocp = uiop->uio_iov->iov_base;
    752 		if (left > siz)
    753 			left = siz;
    754 		uiosiz = left;
    755 		while (left > 0) {
    756 			while (len == 0) {
    757 				mp = mp->m_next;
    758 				if (mp == NULL)
    759 					return (EBADRPC);
    760 				mbufcp = mtod(mp, void *);
    761 				len = mp->m_len;
    762 			}
    763 			xfer = (left > len) ? len : left;
    764 			error = copyout_vmspace(uiop->uio_vmspace, mbufcp,
    765 			    uiocp, xfer);
    766 			if (error) {
    767 				return error;
    768 			}
    769 			left -= xfer;
    770 			len -= xfer;
    771 			mbufcp += xfer;
    772 			uiocp += xfer;
    773 			uiop->uio_offset += xfer;
    774 			uiop->uio_resid -= xfer;
    775 		}
    776 		if (uiop->uio_iov->iov_len <= siz) {
    777 			uiop->uio_iovcnt--;
    778 			uiop->uio_iov++;
    779 		} else {
    780 			uiop->uio_iov->iov_base =
    781 			    (char *)uiop->uio_iov->iov_base + uiosiz;
    782 			uiop->uio_iov->iov_len -= uiosiz;
    783 		}
    784 		siz -= uiosiz;
    785 	}
    786 	*dpos = mbufcp;
    787 	*mrep = mp;
    788 	if (rem > 0) {
    789 		if (len < rem)
    790 			error = nfs_adv(mrep, dpos, rem, len);
    791 		else
    792 			*dpos += rem;
    793 	}
    794 	return (error);
    795 }
    796 
    797 /*
    798  * copies a uio scatter/gather list to an mbuf chain.
    799  * NOTE: can ony handle iovcnt == 1
    800  */
    801 int
    802 nfsm_uiotombuf(struct uio *uiop, struct mbuf **mq, int siz, char **bpos)
    803 {
    804 	char *uiocp;
    805 	struct mbuf *mp, *mp2;
    806 	int xfer, left, mlen;
    807 	int uiosiz, clflg, rem;
    808 	char *cp;
    809 	int error;
    810 
    811 #ifdef DIAGNOSTIC
    812 	if (uiop->uio_iovcnt != 1)
    813 		panic("nfsm_uiotombuf: iovcnt != 1");
    814 #endif
    815 
    816 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
    817 		clflg = 1;
    818 	else
    819 		clflg = 0;
    820 	rem = nfsm_rndup(siz)-siz;
    821 	mp = mp2 = *mq;
    822 	while (siz > 0) {
    823 		left = uiop->uio_iov->iov_len;
    824 		uiocp = uiop->uio_iov->iov_base;
    825 		if (left > siz)
    826 			left = siz;
    827 		uiosiz = left;
    828 		while (left > 0) {
    829 			mlen = M_TRAILINGSPACE(mp);
    830 			if (mlen == 0) {
    831 				mp = m_get(M_WAIT, MT_DATA);
    832 				MCLAIM(mp, &nfs_mowner);
    833 				if (clflg)
    834 					m_clget(mp, M_WAIT);
    835 				mp->m_len = 0;
    836 				mp2->m_next = mp;
    837 				mp2 = mp;
    838 				mlen = M_TRAILINGSPACE(mp);
    839 			}
    840 			xfer = (left > mlen) ? mlen : left;
    841 			cp = mtod(mp, char *) + mp->m_len;
    842 			error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp,
    843 			    xfer);
    844 			if (error) {
    845 				/* XXX */
    846 			}
    847 			mp->m_len += xfer;
    848 			left -= xfer;
    849 			uiocp += xfer;
    850 			uiop->uio_offset += xfer;
    851 			uiop->uio_resid -= xfer;
    852 		}
    853 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base +
    854 		    uiosiz;
    855 		uiop->uio_iov->iov_len -= uiosiz;
    856 		siz -= uiosiz;
    857 	}
    858 	if (rem > 0) {
    859 		if (rem > M_TRAILINGSPACE(mp)) {
    860 			mp = m_get(M_WAIT, MT_DATA);
    861 			MCLAIM(mp, &nfs_mowner);
    862 			mp->m_len = 0;
    863 			mp2->m_next = mp;
    864 		}
    865 		cp = mtod(mp, char *) + mp->m_len;
    866 		for (left = 0; left < rem; left++)
    867 			*cp++ = '\0';
    868 		mp->m_len += rem;
    869 		*bpos = cp;
    870 	} else
    871 		*bpos = mtod(mp, char *) + mp->m_len;
    872 	*mq = mp;
    873 	return (0);
    874 }
    875 
    876 /*
    877  * Get at least "siz" bytes of correctly aligned data.
    878  * When called the mbuf pointers are not necessarily correct,
    879  * dsosp points to what ought to be in m_data and left contains
    880  * what ought to be in m_len.
    881  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
    882  * cases. (The macros use the vars. dpos and dpos2)
    883  */
    884 int
    885 nfsm_disct(struct mbuf **mdp, char **dposp, int siz, int left, char **cp2)
    886 {
    887 	struct mbuf *m1, *m2;
    888 	struct mbuf *havebuf = NULL;
    889 	char *src = *dposp;
    890 	char *dst;
    891 	int len;
    892 
    893 #ifdef DEBUG
    894 	if (left < 0)
    895 		panic("nfsm_disct: left < 0");
    896 #endif
    897 	m1 = *mdp;
    898 	/*
    899 	 * Skip through the mbuf chain looking for an mbuf with
    900 	 * some data. If the first mbuf found has enough data
    901 	 * and it is correctly aligned return it.
    902 	 */
    903 	while (left == 0) {
    904 		havebuf = m1;
    905 		*mdp = m1 = m1->m_next;
    906 		if (m1 == NULL)
    907 			return (EBADRPC);
    908 		src = mtod(m1, void *);
    909 		left = m1->m_len;
    910 		/*
    911 		 * If we start a new mbuf and it is big enough
    912 		 * and correctly aligned just return it, don't
    913 		 * do any pull up.
    914 		 */
    915 		if (left >= siz && nfsm_aligned(src)) {
    916 			*cp2 = src;
    917 			*dposp = src + siz;
    918 			return (0);
    919 		}
    920 	}
    921 	if ((m1->m_flags & M_EXT) != 0) {
    922 		if (havebuf && M_TRAILINGSPACE(havebuf) >= siz &&
    923 		    nfsm_aligned(mtod(havebuf, char *) + havebuf->m_len)) {
    924 			/*
    925 			 * If the first mbuf with data has external data
    926 			 * and there is a previous mbuf with some trailing
    927 			 * space, use it to move the data into.
    928 			 */
    929 			m2 = m1;
    930 			*mdp = m1 = havebuf;
    931 			*cp2 = mtod(m1, char *) + m1->m_len;
    932 		} else if (havebuf) {
    933 			/*
    934 			 * If the first mbuf has a external data
    935 			 * and there is no previous empty mbuf
    936 			 * allocate a new mbuf and move the external
    937 			 * data to the new mbuf. Also make the first
    938 			 * mbuf look empty.
    939 			 */
    940 			m2 = m1;
    941 			*mdp = m1 = m_get(M_WAIT, MT_DATA);
    942 			MCLAIM(m1, m2->m_owner);
    943 			if ((m2->m_flags & M_PKTHDR) != 0) {
    944 				/* XXX MOVE */
    945 				M_COPY_PKTHDR(m1, m2);
    946 				m_tag_delete_chain(m2, NULL);
    947 				m2->m_flags &= ~M_PKTHDR;
    948 			}
    949 			if (havebuf) {
    950 				havebuf->m_next = m1;
    951 			}
    952 			m1->m_next = m2;
    953 			MRESETDATA(m1);
    954 			m1->m_len = 0;
    955 			m2->m_data = src;
    956 			m2->m_len = left;
    957 			*cp2 = mtod(m1, char *);
    958 		} else {
    959 			struct mbuf **nextp = &m1->m_next;
    960 
    961 			m1->m_len -= left;
    962 			do {
    963 				m2 = m_get(M_WAIT, MT_DATA);
    964 				MCLAIM(m2, m1->m_owner);
    965 				if (left >= MINCLSIZE) {
    966 					MCLGET(m2, M_WAIT);
    967 				}
    968 				m2->m_next = *nextp;
    969 				*nextp = m2;
    970 				nextp = &m2->m_next;
    971 				len = (m2->m_flags & M_EXT) != 0 ?
    972 				    MCLBYTES : MLEN;
    973 				if (len > left) {
    974 					len = left;
    975 				}
    976 				memcpy(mtod(m2, char *), src, len);
    977 				m2->m_len = len;
    978 				src += len;
    979 				left -= len;
    980 			} while (left > 0);
    981 			*mdp = m1 = m1->m_next;
    982 			m2 = m1->m_next;
    983 			*cp2 = mtod(m1, char *);
    984 		}
    985 	} else {
    986 		/*
    987 		 * If the first mbuf has no external data
    988 		 * move the data to the front of the mbuf.
    989 		 */
    990 		MRESETDATA(m1);
    991 		dst = mtod(m1, char *);
    992 		if (dst != src) {
    993 			memmove(dst, src, left);
    994 		}
    995 		m1->m_len = left;
    996 		m2 = m1->m_next;
    997 		*cp2 = m1->m_data;
    998 	}
    999 	*dposp = *cp2 + siz;
   1000 	/*
   1001 	 * Loop through mbufs pulling data up into first mbuf until
   1002 	 * the first mbuf is full or there is no more data to
   1003 	 * pullup.
   1004 	 */
   1005 	dst = mtod(m1, char *) + m1->m_len;
   1006 	while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) {
   1007 		if ((len = min(len, m2->m_len)) != 0) {
   1008 			memcpy(dst, mtod(m2, char *), len);
   1009 		}
   1010 		m1->m_len += len;
   1011 		dst += len;
   1012 		m2->m_data += len;
   1013 		m2->m_len -= len;
   1014 		m2 = m2->m_next;
   1015 	}
   1016 	if (m1->m_len < siz)
   1017 		return (EBADRPC);
   1018 	return (0);
   1019 }
   1020 
   1021 /*
   1022  * Advance the position in the mbuf chain.
   1023  */
   1024 int
   1025 nfs_adv(struct mbuf **mdp, char **dposp, int offs, int left)
   1026 {
   1027 	struct mbuf *m;
   1028 	int s;
   1029 
   1030 	m = *mdp;
   1031 	s = left;
   1032 	while (s < offs) {
   1033 		offs -= s;
   1034 		m = m->m_next;
   1035 		if (m == NULL)
   1036 			return (EBADRPC);
   1037 		s = m->m_len;
   1038 	}
   1039 	*mdp = m;
   1040 	*dposp = mtod(m, char *) + offs;
   1041 	return (0);
   1042 }
   1043 
   1044 /*
   1045  * Copy a string into mbufs for the hard cases...
   1046  */
   1047 int
   1048 nfsm_strtmbuf(struct mbuf **mb, char **bpos, const char *cp, long siz)
   1049 {
   1050 	struct mbuf *m1 = NULL, *m2;
   1051 	long left, xfer, len, tlen;
   1052 	u_int32_t *tl;
   1053 	int putsize;
   1054 
   1055 	putsize = 1;
   1056 	m2 = *mb;
   1057 	left = M_TRAILINGSPACE(m2);
   1058 	if (left > 0) {
   1059 		tl = ((u_int32_t *)(*bpos));
   1060 		*tl++ = txdr_unsigned(siz);
   1061 		putsize = 0;
   1062 		left -= NFSX_UNSIGNED;
   1063 		m2->m_len += NFSX_UNSIGNED;
   1064 		if (left > 0) {
   1065 			memcpy((void *) tl, cp, left);
   1066 			siz -= left;
   1067 			cp += left;
   1068 			m2->m_len += left;
   1069 			left = 0;
   1070 		}
   1071 	}
   1072 	/* Loop around adding mbufs */
   1073 	while (siz > 0) {
   1074 		m1 = m_get(M_WAIT, MT_DATA);
   1075 		MCLAIM(m1, &nfs_mowner);
   1076 		if (siz > MLEN)
   1077 			m_clget(m1, M_WAIT);
   1078 		m1->m_len = NFSMSIZ(m1);
   1079 		m2->m_next = m1;
   1080 		m2 = m1;
   1081 		tl = mtod(m1, u_int32_t *);
   1082 		tlen = 0;
   1083 		if (putsize) {
   1084 			*tl++ = txdr_unsigned(siz);
   1085 			m1->m_len -= NFSX_UNSIGNED;
   1086 			tlen = NFSX_UNSIGNED;
   1087 			putsize = 0;
   1088 		}
   1089 		if (siz < m1->m_len) {
   1090 			len = nfsm_rndup(siz);
   1091 			xfer = siz;
   1092 			if (xfer < len)
   1093 				*(tl+(xfer>>2)) = 0;
   1094 		} else {
   1095 			xfer = len = m1->m_len;
   1096 		}
   1097 		memcpy((void *) tl, cp, xfer);
   1098 		m1->m_len = len+tlen;
   1099 		siz -= xfer;
   1100 		cp += xfer;
   1101 	}
   1102 	*mb = m1;
   1103 	*bpos = mtod(m1, char *) + m1->m_len;
   1104 	return (0);
   1105 }
   1106 
   1107 /*
   1108  * Directory caching routines. They work as follows:
   1109  * - a cache is maintained per VDIR nfsnode.
   1110  * - for each offset cookie that is exported to userspace, and can
   1111  *   thus be thrown back at us as an offset to VOP_READDIR, store
   1112  *   information in the cache.
   1113  * - cached are:
   1114  *   - cookie itself
   1115  *   - blocknumber (essentially just a search key in the buffer cache)
   1116  *   - entry number in block.
   1117  *   - offset cookie of block in which this entry is stored
   1118  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
   1119  * - entries are looked up in a hash table
   1120  * - also maintained is an LRU list of entries, used to determine
   1121  *   which ones to delete if the cache grows too large.
   1122  * - if 32 <-> 64 translation mode is requested for a filesystem,
   1123  *   the cache also functions as a translation table
   1124  * - in the translation case, invalidating the cache does not mean
   1125  *   flushing it, but just marking entries as invalid, except for
   1126  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
   1127  *   still be able to use the cache as a translation table.
   1128  * - 32 bit cookies are uniquely created by combining the hash table
   1129  *   entry value, and one generation count per hash table entry,
   1130  *   incremented each time an entry is appended to the chain.
   1131  * - the cache is invalidated each time a direcory is modified
   1132  * - sanity checks are also done; if an entry in a block turns
   1133  *   out not to have a matching cookie, the cache is invalidated
   1134  *   and a new block starting from the wanted offset is fetched from
   1135  *   the server.
   1136  * - directory entries as read from the server are extended to contain
   1137  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
   1138  *   the cache and exporting them to userspace through the cookie
   1139  *   argument to VOP_READDIR.
   1140  */
   1141 
   1142 u_long
   1143 nfs_dirhash(off_t off)
   1144 {
   1145 	int i;
   1146 	char *cp = (char *)&off;
   1147 	u_long sum = 0L;
   1148 
   1149 	for (i = 0 ; i < sizeof (off); i++)
   1150 		sum += *cp++;
   1151 
   1152 	return sum;
   1153 }
   1154 
   1155 #define	_NFSDC_MTX(np)		(&NFSTOV(np)->v_interlock)
   1156 #define	NFSDC_LOCK(np)		mutex_enter(_NFSDC_MTX(np))
   1157 #define	NFSDC_UNLOCK(np)	mutex_exit(_NFSDC_MTX(np))
   1158 #define	NFSDC_ASSERT_LOCKED(np) KASSERT(mutex_owned(_NFSDC_MTX(np)))
   1159 
   1160 void
   1161 nfs_initdircache(struct vnode *vp)
   1162 {
   1163 	struct nfsnode *np = VTONFS(vp);
   1164 	struct nfsdirhashhead *dircache;
   1165 
   1166 	dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, true,
   1167 	    &nfsdirhashmask);
   1168 
   1169 	NFSDC_LOCK(np);
   1170 	if (np->n_dircache == NULL) {
   1171 		np->n_dircachesize = 0;
   1172 		np->n_dircache = dircache;
   1173 		dircache = NULL;
   1174 		TAILQ_INIT(&np->n_dirchain);
   1175 	}
   1176 	NFSDC_UNLOCK(np);
   1177 	if (dircache)
   1178 		hashdone(dircache, HASH_LIST, nfsdirhashmask);
   1179 }
   1180 
   1181 void
   1182 nfs_initdirxlatecookie(struct vnode *vp)
   1183 {
   1184 	struct nfsnode *np = VTONFS(vp);
   1185 	unsigned *dirgens;
   1186 
   1187 	KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE);
   1188 
   1189 	dirgens = kmem_zalloc(NFS_DIRHASHSIZ * sizeof(unsigned), KM_SLEEP);
   1190 	NFSDC_LOCK(np);
   1191 	if (np->n_dirgens == NULL) {
   1192 		np->n_dirgens = dirgens;
   1193 		dirgens = NULL;
   1194 	}
   1195 	NFSDC_UNLOCK(np);
   1196 	if (dirgens)
   1197 		kmem_free(dirgens, NFS_DIRHASHSIZ * sizeof(unsigned));
   1198 }
   1199 
   1200 static const struct nfsdircache dzero;
   1201 
   1202 static void nfs_unlinkdircache(struct nfsnode *np, struct nfsdircache *);
   1203 static void nfs_putdircache_unlocked(struct nfsnode *,
   1204     struct nfsdircache *);
   1205 
   1206 static void
   1207 nfs_unlinkdircache(struct nfsnode *np, struct nfsdircache *ndp)
   1208 {
   1209 
   1210 	NFSDC_ASSERT_LOCKED(np);
   1211 	KASSERT(ndp != &dzero);
   1212 
   1213 	if (LIST_NEXT(ndp, dc_hash) == (void *)-1)
   1214 		return;
   1215 
   1216 	TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
   1217 	LIST_REMOVE(ndp, dc_hash);
   1218 	LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */
   1219 
   1220 	nfs_putdircache_unlocked(np, ndp);
   1221 }
   1222 
   1223 void
   1224 nfs_putdircache(struct nfsnode *np, struct nfsdircache *ndp)
   1225 {
   1226 	int ref;
   1227 
   1228 	if (ndp == &dzero)
   1229 		return;
   1230 
   1231 	KASSERT(ndp->dc_refcnt > 0);
   1232 	NFSDC_LOCK(np);
   1233 	ref = --ndp->dc_refcnt;
   1234 	NFSDC_UNLOCK(np);
   1235 
   1236 	if (ref == 0)
   1237 		kmem_free(ndp, sizeof(*ndp));
   1238 }
   1239 
   1240 static void
   1241 nfs_putdircache_unlocked(struct nfsnode *np, struct nfsdircache *ndp)
   1242 {
   1243 	int ref;
   1244 
   1245 	NFSDC_ASSERT_LOCKED(np);
   1246 
   1247 	if (ndp == &dzero)
   1248 		return;
   1249 
   1250 	KASSERT(ndp->dc_refcnt > 0);
   1251 	ref = --ndp->dc_refcnt;
   1252 	if (ref == 0)
   1253 		kmem_free(ndp, sizeof(*ndp));
   1254 }
   1255 
   1256 struct nfsdircache *
   1257 nfs_searchdircache(struct vnode *vp, off_t off, int do32, int *hashent)
   1258 {
   1259 	struct nfsdirhashhead *ndhp;
   1260 	struct nfsdircache *ndp = NULL;
   1261 	struct nfsnode *np = VTONFS(vp);
   1262 	unsigned ent;
   1263 
   1264 	/*
   1265 	 * Zero is always a valid cookie.
   1266 	 */
   1267 	if (off == 0)
   1268 		/* XXXUNCONST */
   1269 		return (struct nfsdircache *)__UNCONST(&dzero);
   1270 
   1271 	if (!np->n_dircache)
   1272 		return NULL;
   1273 
   1274 	/*
   1275 	 * We use a 32bit cookie as search key, directly reconstruct
   1276 	 * the hashentry. Else use the hashfunction.
   1277 	 */
   1278 	if (do32) {
   1279 		ent = (u_int32_t)off >> 24;
   1280 		if (ent >= NFS_DIRHASHSIZ)
   1281 			return NULL;
   1282 		ndhp = &np->n_dircache[ent];
   1283 	} else {
   1284 		ndhp = NFSDIRHASH(np, off);
   1285 	}
   1286 
   1287 	if (hashent)
   1288 		*hashent = (int)(ndhp - np->n_dircache);
   1289 
   1290 	NFSDC_LOCK(np);
   1291 	if (do32) {
   1292 		LIST_FOREACH(ndp, ndhp, dc_hash) {
   1293 			if (ndp->dc_cookie32 == (u_int32_t)off) {
   1294 				/*
   1295 				 * An invalidated entry will become the
   1296 				 * start of a new block fetched from
   1297 				 * the server.
   1298 				 */
   1299 				if (ndp->dc_flags & NFSDC_INVALID) {
   1300 					ndp->dc_blkcookie = ndp->dc_cookie;
   1301 					ndp->dc_entry = 0;
   1302 					ndp->dc_flags &= ~NFSDC_INVALID;
   1303 				}
   1304 				break;
   1305 			}
   1306 		}
   1307 	} else {
   1308 		LIST_FOREACH(ndp, ndhp, dc_hash) {
   1309 			if (ndp->dc_cookie == off)
   1310 				break;
   1311 		}
   1312 	}
   1313 	if (ndp != NULL)
   1314 		ndp->dc_refcnt++;
   1315 	NFSDC_UNLOCK(np);
   1316 	return ndp;
   1317 }
   1318 
   1319 
   1320 struct nfsdircache *
   1321 nfs_enterdircache(struct vnode *vp, off_t off, off_t blkoff, int en,
   1322     daddr_t blkno)
   1323 {
   1324 	struct nfsnode *np = VTONFS(vp);
   1325 	struct nfsdirhashhead *ndhp;
   1326 	struct nfsdircache *ndp = NULL;
   1327 	struct nfsdircache *newndp = NULL;
   1328 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
   1329 	int hashent = 0, gen, overwrite;	/* XXX: GCC */
   1330 
   1331 	/*
   1332 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
   1333 	 * cookie 0 for the '.' entry, making this necessary. This
   1334 	 * isn't so bad, as 0 is a special case anyway.
   1335 	 */
   1336 	if (off == 0)
   1337 		/* XXXUNCONST */
   1338 		return (struct nfsdircache *)__UNCONST(&dzero);
   1339 
   1340 	if (!np->n_dircache)
   1341 		/*
   1342 		 * XXX would like to do this in nfs_nget but vtype
   1343 		 * isn't known at that time.
   1344 		 */
   1345 		nfs_initdircache(vp);
   1346 
   1347 	if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens)
   1348 		nfs_initdirxlatecookie(vp);
   1349 
   1350 retry:
   1351 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
   1352 
   1353 	NFSDC_LOCK(np);
   1354 	if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) {
   1355 		/*
   1356 		 * Overwriting an old entry. Check if it's the same.
   1357 		 * If so, just return. If not, remove the old entry.
   1358 		 */
   1359 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
   1360 			goto done;
   1361 		nfs_unlinkdircache(np, ndp);
   1362 		nfs_putdircache_unlocked(np, ndp);
   1363 		ndp = NULL;
   1364 	}
   1365 
   1366 	ndhp = &np->n_dircache[hashent];
   1367 
   1368 	if (!ndp) {
   1369 		if (newndp == NULL) {
   1370 			NFSDC_UNLOCK(np);
   1371 			newndp = kmem_alloc(sizeof(*newndp), KM_SLEEP);
   1372 			newndp->dc_refcnt = 1;
   1373 			LIST_NEXT(newndp, dc_hash) = (void *)-1;
   1374 			goto retry;
   1375 		}
   1376 		ndp = newndp;
   1377 		newndp = NULL;
   1378 		overwrite = 0;
   1379 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
   1380 			/*
   1381 			 * We're allocating a new entry, so bump the
   1382 			 * generation number.
   1383 			 */
   1384 			KASSERT(np->n_dirgens);
   1385 			gen = ++np->n_dirgens[hashent];
   1386 			if (gen == 0) {
   1387 				np->n_dirgens[hashent]++;
   1388 				gen++;
   1389 			}
   1390 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
   1391 		}
   1392 	} else
   1393 		overwrite = 1;
   1394 
   1395 	ndp->dc_cookie = off;
   1396 	ndp->dc_blkcookie = blkoff;
   1397 	ndp->dc_entry = en;
   1398 	ndp->dc_flags = 0;
   1399 
   1400 	if (overwrite)
   1401 		goto done;
   1402 
   1403 	/*
   1404 	 * If the maximum directory cookie cache size has been reached
   1405 	 * for this node, take one off the front. The idea is that
   1406 	 * directories are typically read front-to-back once, so that
   1407 	 * the oldest entries can be thrown away without much performance
   1408 	 * loss.
   1409 	 */
   1410 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
   1411 		nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain));
   1412 	} else
   1413 		np->n_dircachesize++;
   1414 
   1415 	KASSERT(ndp->dc_refcnt == 1);
   1416 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
   1417 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
   1418 	ndp->dc_refcnt++;
   1419 done:
   1420 	KASSERT(ndp->dc_refcnt > 0);
   1421 	NFSDC_UNLOCK(np);
   1422 	if (newndp)
   1423 		nfs_putdircache(np, newndp);
   1424 	return ndp;
   1425 }
   1426 
   1427 void
   1428 nfs_invaldircache(struct vnode *vp, int flags)
   1429 {
   1430 	struct nfsnode *np = VTONFS(vp);
   1431 	struct nfsdircache *ndp = NULL;
   1432 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
   1433 	const bool forcefree = flags & NFS_INVALDIRCACHE_FORCE;
   1434 
   1435 #ifdef DIAGNOSTIC
   1436 	if (vp->v_type != VDIR)
   1437 		panic("nfs: invaldircache: not dir");
   1438 #endif
   1439 
   1440 	if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0)
   1441 		np->n_flag &= ~NEOFVALID;
   1442 
   1443 	if (!np->n_dircache)
   1444 		return;
   1445 
   1446 	NFSDC_LOCK(np);
   1447 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
   1448 		while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) {
   1449 			KASSERT(!forcefree || ndp->dc_refcnt == 1);
   1450 			nfs_unlinkdircache(np, ndp);
   1451 		}
   1452 		np->n_dircachesize = 0;
   1453 		if (forcefree && np->n_dirgens) {
   1454 			kmem_free(np->n_dirgens,
   1455 			    NFS_DIRHASHSIZ * sizeof(unsigned));
   1456 			np->n_dirgens = NULL;
   1457 		}
   1458 	} else {
   1459 		TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain)
   1460 			ndp->dc_flags |= NFSDC_INVALID;
   1461 	}
   1462 
   1463 	NFSDC_UNLOCK(np);
   1464 }
   1465 
   1466 /*
   1467  * Called once before VFS init to initialize shared and
   1468  * server-specific data structures.
   1469  */
   1470 static int
   1471 nfs_init0(void)
   1472 {
   1473 
   1474 	nfsrtt.pos = 0;
   1475 	rpc_vers = txdr_unsigned(RPC_VER2);
   1476 	rpc_call = txdr_unsigned(RPC_CALL);
   1477 	rpc_reply = txdr_unsigned(RPC_REPLY);
   1478 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
   1479 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
   1480 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
   1481 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
   1482 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
   1483 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
   1484 	nfs_prog = txdr_unsigned(NFS_PROG);
   1485 	nfs_true = txdr_unsigned(true);
   1486 	nfs_false = txdr_unsigned(false);
   1487 	nfs_xdrneg1 = txdr_unsigned(-1);
   1488 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
   1489 	if (nfs_ticks < 1)
   1490 		nfs_ticks = 1;
   1491 	nfs_xid = arc4random();
   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 /*
   1505  * This is disgusting, but it must support both modular and monolothic
   1506  * configurations.  For monolithic builds NFSSERVER may not imply NFS.
   1507  *
   1508  * Yuck.
   1509  */
   1510 void
   1511 nfs_init(void)
   1512 {
   1513 	static ONCE_DECL(nfs_init_once);
   1514 
   1515 	RUN_ONCE(&nfs_init_once, nfs_init0);
   1516 }
   1517 
   1518 void
   1519 nfs_fini(void)
   1520 {
   1521 
   1522 	nfsdreq_fini();
   1523 	nfs_timer_fini();
   1524 	MOWNER_DETACH(&nfs_mowner);
   1525 }
   1526 
   1527 /*
   1528  * A fiddled version of m_adj() that ensures null fill to a 32-bit
   1529  * boundary and only trims off the back end
   1530  *
   1531  * 1. trim off 'len' bytes as m_adj(mp, -len).
   1532  * 2. add zero-padding 'nul' bytes at the end of the mbuf chain.
   1533  */
   1534 void
   1535 nfs_zeropad(struct mbuf *mp, int len, int nul)
   1536 {
   1537 	struct mbuf *m;
   1538 	int count;
   1539 
   1540 	/*
   1541 	 * Trim from tail.  Scan the mbuf chain,
   1542 	 * calculating its length and finding the last mbuf.
   1543 	 * If the adjustment only affects this mbuf, then just
   1544 	 * adjust and return.  Otherwise, rescan and truncate
   1545 	 * after the remaining size.
   1546 	 */
   1547 	count = 0;
   1548 	m = mp;
   1549 	for (;;) {
   1550 		count += m->m_len;
   1551 		if (m->m_next == NULL)
   1552 			break;
   1553 		m = m->m_next;
   1554 	}
   1555 
   1556 	KDASSERT(count >= len);
   1557 
   1558 	if (m->m_len >= len) {
   1559 		m->m_len -= len;
   1560 	} else {
   1561 		count -= len;
   1562 		/*
   1563 		 * Correct length for chain is "count".
   1564 		 * Find the mbuf with last data, adjust its length,
   1565 		 * and toss data from remaining mbufs on chain.
   1566 		 */
   1567 		for (m = mp; m; m = m->m_next) {
   1568 			if (m->m_len >= count) {
   1569 				m->m_len = count;
   1570 				break;
   1571 			}
   1572 			count -= m->m_len;
   1573 		}
   1574 		KASSERT(m && m->m_next);
   1575 		m_freem(m->m_next);
   1576 		m->m_next = NULL;
   1577 	}
   1578 
   1579 	KDASSERT(m->m_next == NULL);
   1580 
   1581 	/*
   1582 	 * zero-padding.
   1583 	 */
   1584 	if (nul > 0) {
   1585 		char *cp;
   1586 		int i;
   1587 
   1588 		if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) {
   1589 			struct mbuf *n;
   1590 
   1591 			KDASSERT(MLEN >= nul);
   1592 			n = m_get(M_WAIT, MT_DATA);
   1593 			MCLAIM(n, &nfs_mowner);
   1594 			n->m_len = nul;
   1595 			n->m_next = NULL;
   1596 			m->m_next = n;
   1597 			cp = mtod(n, void *);
   1598 		} else {
   1599 			cp = mtod(m, char *) + m->m_len;
   1600 			m->m_len += nul;
   1601 		}
   1602 		for (i = 0; i < nul; i++)
   1603 			*cp++ = '\0';
   1604 	}
   1605 	return;
   1606 }
   1607 
   1608 /*
   1609  * Make these functions instead of macros, so that the kernel text size
   1610  * doesn't get too big...
   1611  */
   1612 void
   1613 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)
   1614 {
   1615 	struct mbuf *mb = *mbp;
   1616 	char *bpos = *bposp;
   1617 	u_int32_t *tl;
   1618 
   1619 	if (before_ret) {
   1620 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
   1621 		*tl = nfs_false;
   1622 	} else {
   1623 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
   1624 		*tl++ = nfs_true;
   1625 		txdr_hyper(before_vap->va_size, tl);
   1626 		tl += 2;
   1627 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
   1628 		tl += 2;
   1629 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
   1630 	}
   1631 	*bposp = bpos;
   1632 	*mbp = mb;
   1633 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
   1634 }
   1635 
   1636 void
   1637 nfsm_srvpostopattr(struct nfsrv_descript *nfsd, int after_ret, struct vattr *after_vap, struct mbuf **mbp, char **bposp)
   1638 {
   1639 	struct mbuf *mb = *mbp;
   1640 	char *bpos = *bposp;
   1641 	u_int32_t *tl;
   1642 	struct nfs_fattr *fp;
   1643 
   1644 	if (after_ret) {
   1645 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
   1646 		*tl = nfs_false;
   1647 	} else {
   1648 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
   1649 		*tl++ = nfs_true;
   1650 		fp = (struct nfs_fattr *)tl;
   1651 		nfsm_srvfattr(nfsd, after_vap, fp);
   1652 	}
   1653 	*mbp = mb;
   1654 	*bposp = bpos;
   1655 }
   1656 
   1657 void
   1658 nfsm_srvfattr(struct nfsrv_descript *nfsd, struct vattr *vap, struct nfs_fattr *fp)
   1659 {
   1660 
   1661 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
   1662 	fp->fa_uid = txdr_unsigned(vap->va_uid);
   1663 	fp->fa_gid = txdr_unsigned(vap->va_gid);
   1664 	if (nfsd->nd_flag & ND_NFSV3) {
   1665 		fp->fa_type = vtonfsv3_type(vap->va_type);
   1666 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
   1667 		txdr_hyper(vap->va_size, &fp->fa3_size);
   1668 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
   1669 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
   1670 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
   1671 		fp->fa3_fsid.nfsuquad[0] = 0;
   1672 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
   1673 		txdr_hyper(vap->va_fileid, &fp->fa3_fileid);
   1674 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
   1675 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
   1676 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
   1677 	} else {
   1678 		fp->fa_type = vtonfsv2_type(vap->va_type);
   1679 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
   1680 		fp->fa2_size = txdr_unsigned(vap->va_size);
   1681 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
   1682 		if (vap->va_type == VFIFO)
   1683 			fp->fa2_rdev = 0xffffffff;
   1684 		else
   1685 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
   1686 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
   1687 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
   1688 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
   1689 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
   1690 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
   1691 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
   1692 	}
   1693 }
   1694 
   1695 /*
   1696  * This function compares two net addresses by family and returns true
   1697  * if they are the same host.
   1698  * If there is any doubt, return false.
   1699  * The AF_INET family is handled as a special case so that address mbufs
   1700  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
   1701  */
   1702 int
   1703 netaddr_match(int family, union nethostaddr *haddr, struct mbuf *nam)
   1704 {
   1705 	struct sockaddr_in *inetaddr;
   1706 
   1707 	switch (family) {
   1708 	case AF_INET:
   1709 		inetaddr = mtod(nam, struct sockaddr_in *);
   1710 		if (inetaddr->sin_family == AF_INET &&
   1711 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
   1712 			return (1);
   1713 		break;
   1714 	case AF_INET6:
   1715 	    {
   1716 		struct sockaddr_in6 *sin6_1, *sin6_2;
   1717 
   1718 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
   1719 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
   1720 		if (sin6_1->sin6_family == AF_INET6 &&
   1721 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
   1722 			return 1;
   1723 	    }
   1724 	default:
   1725 		break;
   1726 	};
   1727 	return (0);
   1728 }
   1729 
   1730 /*
   1731  * The write verifier has changed (probably due to a server reboot), so all
   1732  * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
   1733  * as dirty or are being written out just now, all this takes is clearing
   1734  * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
   1735  * the mount point.
   1736  */
   1737 void
   1738 nfs_clearcommit(struct mount *mp)
   1739 {
   1740 	struct vnode *vp;
   1741 	struct nfsnode *np;
   1742 	struct vm_page *pg;
   1743 	struct nfsmount *nmp = VFSTONFS(mp);
   1744 
   1745 	rw_enter(&nmp->nm_writeverflock, RW_WRITER);
   1746 	mutex_enter(&mntvnode_lock);
   1747 	TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
   1748 		KASSERT(vp->v_mount == mp);
   1749 		if (vp->v_type != VREG)
   1750 			continue;
   1751 		mutex_enter(&vp->v_interlock);
   1752 		if (vp->v_iflag & (VI_XLOCK | VI_CLEAN)) {
   1753 			mutex_exit(&vp->v_interlock);
   1754 			continue;
   1755 		}
   1756 		np = VTONFS(vp);
   1757 		np->n_pushlo = np->n_pushhi = np->n_pushedlo =
   1758 		    np->n_pushedhi = 0;
   1759 		np->n_commitflags &=
   1760 		    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
   1761 		TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq.queue) {
   1762 			pg->flags &= ~PG_NEEDCOMMIT;
   1763 		}
   1764 		mutex_exit(&vp->v_interlock);
   1765 	}
   1766 	mutex_exit(&mntvnode_lock);
   1767 	mutex_enter(&nmp->nm_lock);
   1768 	nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
   1769 	mutex_exit(&nmp->nm_lock);
   1770 	rw_exit(&nmp->nm_writeverflock);
   1771 }
   1772 
   1773 void
   1774 nfs_merge_commit_ranges(struct vnode *vp)
   1775 {
   1776 	struct nfsnode *np = VTONFS(vp);
   1777 
   1778 	KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
   1779 
   1780 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
   1781 		np->n_pushedlo = np->n_pushlo;
   1782 		np->n_pushedhi = np->n_pushhi;
   1783 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
   1784 	} else {
   1785 		if (np->n_pushlo < np->n_pushedlo)
   1786 			np->n_pushedlo = np->n_pushlo;
   1787 		if (np->n_pushhi > np->n_pushedhi)
   1788 			np->n_pushedhi = np->n_pushhi;
   1789 	}
   1790 
   1791 	np->n_pushlo = np->n_pushhi = 0;
   1792 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
   1793 
   1794 #ifdef NFS_DEBUG_COMMIT
   1795 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
   1796 	    (unsigned)np->n_pushedhi);
   1797 #endif
   1798 }
   1799 
   1800 int
   1801 nfs_in_committed_range(struct vnode *vp, off_t off, off_t len)
   1802 {
   1803 	struct nfsnode *np = VTONFS(vp);
   1804 	off_t lo, hi;
   1805 
   1806 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
   1807 		return 0;
   1808 	lo = off;
   1809 	hi = lo + len;
   1810 
   1811 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
   1812 }
   1813 
   1814 int
   1815 nfs_in_tobecommitted_range(struct vnode *vp, off_t off, off_t len)
   1816 {
   1817 	struct nfsnode *np = VTONFS(vp);
   1818 	off_t lo, hi;
   1819 
   1820 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
   1821 		return 0;
   1822 	lo = off;
   1823 	hi = lo + len;
   1824 
   1825 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
   1826 }
   1827 
   1828 void
   1829 nfs_add_committed_range(struct vnode *vp, off_t off, off_t len)
   1830 {
   1831 	struct nfsnode *np = VTONFS(vp);
   1832 	off_t lo, hi;
   1833 
   1834 	lo = off;
   1835 	hi = lo + len;
   1836 
   1837 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
   1838 		np->n_pushedlo = lo;
   1839 		np->n_pushedhi = hi;
   1840 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
   1841 	} else {
   1842 		if (hi > np->n_pushedhi)
   1843 			np->n_pushedhi = hi;
   1844 		if (lo < np->n_pushedlo)
   1845 			np->n_pushedlo = lo;
   1846 	}
   1847 #ifdef NFS_DEBUG_COMMIT
   1848 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
   1849 	    (unsigned)np->n_pushedhi);
   1850 #endif
   1851 }
   1852 
   1853 void
   1854 nfs_del_committed_range(struct vnode *vp, off_t off, off_t len)
   1855 {
   1856 	struct nfsnode *np = VTONFS(vp);
   1857 	off_t lo, hi;
   1858 
   1859 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
   1860 		return;
   1861 
   1862 	lo = off;
   1863 	hi = lo + len;
   1864 
   1865 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
   1866 		return;
   1867 	if (lo <= np->n_pushedlo)
   1868 		np->n_pushedlo = hi;
   1869 	else if (hi >= np->n_pushedhi)
   1870 		np->n_pushedhi = lo;
   1871 	else {
   1872 		/*
   1873 		 * XXX There's only one range. If the deleted range
   1874 		 * is in the middle, pick the largest of the
   1875 		 * contiguous ranges that it leaves.
   1876 		 */
   1877 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
   1878 			np->n_pushedhi = lo;
   1879 		else
   1880 			np->n_pushedlo = hi;
   1881 	}
   1882 #ifdef NFS_DEBUG_COMMIT
   1883 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
   1884 	    (unsigned)np->n_pushedhi);
   1885 #endif
   1886 }
   1887 
   1888 void
   1889 nfs_add_tobecommitted_range(struct vnode *vp, off_t off, off_t len)
   1890 {
   1891 	struct nfsnode *np = VTONFS(vp);
   1892 	off_t lo, hi;
   1893 
   1894 	lo = off;
   1895 	hi = lo + len;
   1896 
   1897 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
   1898 		np->n_pushlo = lo;
   1899 		np->n_pushhi = hi;
   1900 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
   1901 	} else {
   1902 		if (lo < np->n_pushlo)
   1903 			np->n_pushlo = lo;
   1904 		if (hi > np->n_pushhi)
   1905 			np->n_pushhi = hi;
   1906 	}
   1907 #ifdef NFS_DEBUG_COMMIT
   1908 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
   1909 	    (unsigned)np->n_pushhi);
   1910 #endif
   1911 }
   1912 
   1913 void
   1914 nfs_del_tobecommitted_range(struct vnode *vp, off_t off, off_t len)
   1915 {
   1916 	struct nfsnode *np = VTONFS(vp);
   1917 	off_t lo, hi;
   1918 
   1919 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
   1920 		return;
   1921 
   1922 	lo = off;
   1923 	hi = lo + len;
   1924 
   1925 	if (lo > np->n_pushhi || hi < np->n_pushlo)
   1926 		return;
   1927 
   1928 	if (lo <= np->n_pushlo)
   1929 		np->n_pushlo = hi;
   1930 	else if (hi >= np->n_pushhi)
   1931 		np->n_pushhi = lo;
   1932 	else {
   1933 		/*
   1934 		 * XXX There's only one range. If the deleted range
   1935 		 * is in the middle, pick the largest of the
   1936 		 * contiguous ranges that it leaves.
   1937 		 */
   1938 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
   1939 			np->n_pushhi = lo;
   1940 		else
   1941 			np->n_pushlo = hi;
   1942 	}
   1943 #ifdef NFS_DEBUG_COMMIT
   1944 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
   1945 	    (unsigned)np->n_pushhi);
   1946 #endif
   1947 }
   1948 
   1949 /*
   1950  * Map errnos to NFS error numbers. For Version 3 also filter out error
   1951  * numbers not specified for the associated procedure.
   1952  */
   1953 int
   1954 nfsrv_errmap(struct nfsrv_descript *nd, int err)
   1955 {
   1956 	const short *defaulterrp, *errp;
   1957 
   1958 	if (nd->nd_flag & ND_NFSV3) {
   1959 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
   1960 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
   1961 		while (*++errp) {
   1962 			if (*errp == err)
   1963 				return (err);
   1964 			else if (*errp > err)
   1965 				break;
   1966 		}
   1967 		return ((int)*defaulterrp);
   1968 	    } else
   1969 		return (err & 0xffff);
   1970 	}
   1971 	if (err <= ELAST)
   1972 		return ((int)nfsrv_v2errmap[err - 1]);
   1973 	return (NFSERR_IO);
   1974 }
   1975 
   1976 u_int32_t
   1977 nfs_getxid(void)
   1978 {
   1979 	u_int32_t newxid;
   1980 
   1981 	/* get next xid.  skip 0 */
   1982 	do {
   1983 		newxid = atomic_inc_32_nv(&nfs_xid);
   1984 	} while (__predict_false(newxid == 0));
   1985 
   1986 	return txdr_unsigned(newxid);
   1987 }
   1988 
   1989 /*
   1990  * assign a new xid for existing request.
   1991  * used for NFSERR_JUKEBOX handling.
   1992  */
   1993 void
   1994 nfs_renewxid(struct nfsreq *req)
   1995 {
   1996 	u_int32_t xid;
   1997 	int off;
   1998 
   1999 	xid = nfs_getxid();
   2000 	if (req->r_nmp->nm_sotype == SOCK_STREAM)
   2001 		off = sizeof(u_int32_t); /* RPC record mark */
   2002 	else
   2003 		off = 0;
   2004 
   2005 	m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid);
   2006 	req->r_xid = xid;
   2007 }
   2008