nfs_subs.c revision 1.188
1 /* $NetBSD: nfs_subs.c,v 1.188 2007/06/06 09:23:55 yamt Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_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 1.188 2007/06/06 09:23:55 yamt Exp $"); 74 75 #include "fs_nfs.h" 76 #include "opt_nfs.h" 77 #include "opt_nfsserver.h" 78 #include "opt_iso.h" 79 #include "opt_inet.h" 80 81 /* 82 * These functions support the macros and help fiddle mbuf chains for 83 * the nfs op functions. They do things like create the rpc header and 84 * copy data between mbuf chains and uio lists. 85 */ 86 #include <sys/param.h> 87 #include <sys/proc.h> 88 #include <sys/systm.h> 89 #include <sys/kernel.h> 90 #include <sys/mount.h> 91 #include <sys/vnode.h> 92 #include <sys/namei.h> 93 #include <sys/mbuf.h> 94 #include <sys/socket.h> 95 #include <sys/stat.h> 96 #include <sys/malloc.h> 97 #include <sys/filedesc.h> 98 #include <sys/time.h> 99 #include <sys/dirent.h> 100 #include <sys/once.h> 101 #include <sys/kauth.h> 102 103 #include <uvm/uvm_extern.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 #ifdef ISO 119 #include <netiso/iso.h> 120 #endif 121 122 /* 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 int nfs_commitsize; 143 144 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies"); 145 146 /* NFS client/server stats. */ 147 struct nfsstats nfsstats; 148 149 /* 150 * Mapping of old NFS Version 2 RPC numbers to generic numbers. 151 */ 152 const int nfsv3_procid[NFS_NPROCS] = { 153 NFSPROC_NULL, 154 NFSPROC_GETATTR, 155 NFSPROC_SETATTR, 156 NFSPROC_NOOP, 157 NFSPROC_LOOKUP, 158 NFSPROC_READLINK, 159 NFSPROC_READ, 160 NFSPROC_NOOP, 161 NFSPROC_WRITE, 162 NFSPROC_CREATE, 163 NFSPROC_REMOVE, 164 NFSPROC_RENAME, 165 NFSPROC_LINK, 166 NFSPROC_SYMLINK, 167 NFSPROC_MKDIR, 168 NFSPROC_RMDIR, 169 NFSPROC_READDIR, 170 NFSPROC_FSSTAT, 171 NFSPROC_NOOP, 172 NFSPROC_NOOP, 173 NFSPROC_NOOP, 174 NFSPROC_NOOP, 175 NFSPROC_NOOP 176 }; 177 178 /* 179 * and the reverse mapping from generic to Version 2 procedure numbers 180 */ 181 const int nfsv2_procid[NFS_NPROCS] = { 182 NFSV2PROC_NULL, 183 NFSV2PROC_GETATTR, 184 NFSV2PROC_SETATTR, 185 NFSV2PROC_LOOKUP, 186 NFSV2PROC_NOOP, 187 NFSV2PROC_READLINK, 188 NFSV2PROC_READ, 189 NFSV2PROC_WRITE, 190 NFSV2PROC_CREATE, 191 NFSV2PROC_MKDIR, 192 NFSV2PROC_SYMLINK, 193 NFSV2PROC_CREATE, 194 NFSV2PROC_REMOVE, 195 NFSV2PROC_RMDIR, 196 NFSV2PROC_RENAME, 197 NFSV2PROC_LINK, 198 NFSV2PROC_READDIR, 199 NFSV2PROC_NOOP, 200 NFSV2PROC_STATFS, 201 NFSV2PROC_NOOP, 202 NFSV2PROC_NOOP, 203 NFSV2PROC_NOOP, 204 NFSV2PROC_NOOP, 205 }; 206 207 /* 208 * Maps errno values to nfs error numbers. 209 * Use NFSERR_IO as the catch all for ones not specifically defined in 210 * RFC 1094. 211 */ 212 static const u_char nfsrv_v2errmap[ELAST] = { 213 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 214 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 215 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 216 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 217 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 218 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 219 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 220 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 221 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 222 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 223 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 224 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 225 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO, 226 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 227 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 228 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 229 NFSERR_IO, NFSERR_IO, 230 }; 231 232 /* 233 * Maps errno values to nfs error numbers. 234 * Although it is not obvious whether or not NFS clients really care if 235 * a returned error value is in the specified list for the procedure, the 236 * safest thing to do is filter them appropriately. For Version 2, the 237 * X/Open XNFS document is the only specification that defines error values 238 * for each RPC (The RFC simply lists all possible error values for all RPCs), 239 * so I have decided to not do this for Version 2. 240 * The first entry is the default error return and the rest are the valid 241 * errors for that RPC in increasing numeric order. 242 */ 243 static const short nfsv3err_null[] = { 244 0, 245 0, 246 }; 247 248 static const short nfsv3err_getattr[] = { 249 NFSERR_IO, 250 NFSERR_IO, 251 NFSERR_STALE, 252 NFSERR_BADHANDLE, 253 NFSERR_SERVERFAULT, 254 0, 255 }; 256 257 static const short nfsv3err_setattr[] = { 258 NFSERR_IO, 259 NFSERR_PERM, 260 NFSERR_IO, 261 NFSERR_ACCES, 262 NFSERR_INVAL, 263 NFSERR_NOSPC, 264 NFSERR_ROFS, 265 NFSERR_DQUOT, 266 NFSERR_STALE, 267 NFSERR_BADHANDLE, 268 NFSERR_NOT_SYNC, 269 NFSERR_SERVERFAULT, 270 0, 271 }; 272 273 static const short nfsv3err_lookup[] = { 274 NFSERR_IO, 275 NFSERR_NOENT, 276 NFSERR_IO, 277 NFSERR_ACCES, 278 NFSERR_NOTDIR, 279 NFSERR_NAMETOL, 280 NFSERR_STALE, 281 NFSERR_BADHANDLE, 282 NFSERR_SERVERFAULT, 283 0, 284 }; 285 286 static const short nfsv3err_access[] = { 287 NFSERR_IO, 288 NFSERR_IO, 289 NFSERR_STALE, 290 NFSERR_BADHANDLE, 291 NFSERR_SERVERFAULT, 292 0, 293 }; 294 295 static const short nfsv3err_readlink[] = { 296 NFSERR_IO, 297 NFSERR_IO, 298 NFSERR_ACCES, 299 NFSERR_INVAL, 300 NFSERR_STALE, 301 NFSERR_BADHANDLE, 302 NFSERR_NOTSUPP, 303 NFSERR_SERVERFAULT, 304 0, 305 }; 306 307 static const short nfsv3err_read[] = { 308 NFSERR_IO, 309 NFSERR_IO, 310 NFSERR_NXIO, 311 NFSERR_ACCES, 312 NFSERR_INVAL, 313 NFSERR_STALE, 314 NFSERR_BADHANDLE, 315 NFSERR_SERVERFAULT, 316 NFSERR_JUKEBOX, 317 0, 318 }; 319 320 static const short nfsv3err_write[] = { 321 NFSERR_IO, 322 NFSERR_IO, 323 NFSERR_ACCES, 324 NFSERR_INVAL, 325 NFSERR_FBIG, 326 NFSERR_NOSPC, 327 NFSERR_ROFS, 328 NFSERR_DQUOT, 329 NFSERR_STALE, 330 NFSERR_BADHANDLE, 331 NFSERR_SERVERFAULT, 332 NFSERR_JUKEBOX, 333 0, 334 }; 335 336 static const short nfsv3err_create[] = { 337 NFSERR_IO, 338 NFSERR_IO, 339 NFSERR_ACCES, 340 NFSERR_EXIST, 341 NFSERR_NOTDIR, 342 NFSERR_NOSPC, 343 NFSERR_ROFS, 344 NFSERR_NAMETOL, 345 NFSERR_DQUOT, 346 NFSERR_STALE, 347 NFSERR_BADHANDLE, 348 NFSERR_NOTSUPP, 349 NFSERR_SERVERFAULT, 350 0, 351 }; 352 353 static const short nfsv3err_mkdir[] = { 354 NFSERR_IO, 355 NFSERR_IO, 356 NFSERR_ACCES, 357 NFSERR_EXIST, 358 NFSERR_NOTDIR, 359 NFSERR_NOSPC, 360 NFSERR_ROFS, 361 NFSERR_NAMETOL, 362 NFSERR_DQUOT, 363 NFSERR_STALE, 364 NFSERR_BADHANDLE, 365 NFSERR_NOTSUPP, 366 NFSERR_SERVERFAULT, 367 0, 368 }; 369 370 static const short nfsv3err_symlink[] = { 371 NFSERR_IO, 372 NFSERR_IO, 373 NFSERR_ACCES, 374 NFSERR_EXIST, 375 NFSERR_NOTDIR, 376 NFSERR_NOSPC, 377 NFSERR_ROFS, 378 NFSERR_NAMETOL, 379 NFSERR_DQUOT, 380 NFSERR_STALE, 381 NFSERR_BADHANDLE, 382 NFSERR_NOTSUPP, 383 NFSERR_SERVERFAULT, 384 0, 385 }; 386 387 static const short nfsv3err_mknod[] = { 388 NFSERR_IO, 389 NFSERR_IO, 390 NFSERR_ACCES, 391 NFSERR_EXIST, 392 NFSERR_NOTDIR, 393 NFSERR_NOSPC, 394 NFSERR_ROFS, 395 NFSERR_NAMETOL, 396 NFSERR_DQUOT, 397 NFSERR_STALE, 398 NFSERR_BADHANDLE, 399 NFSERR_NOTSUPP, 400 NFSERR_SERVERFAULT, 401 NFSERR_BADTYPE, 402 0, 403 }; 404 405 static const short nfsv3err_remove[] = { 406 NFSERR_IO, 407 NFSERR_NOENT, 408 NFSERR_IO, 409 NFSERR_ACCES, 410 NFSERR_NOTDIR, 411 NFSERR_ROFS, 412 NFSERR_NAMETOL, 413 NFSERR_STALE, 414 NFSERR_BADHANDLE, 415 NFSERR_SERVERFAULT, 416 0, 417 }; 418 419 static const short nfsv3err_rmdir[] = { 420 NFSERR_IO, 421 NFSERR_NOENT, 422 NFSERR_IO, 423 NFSERR_ACCES, 424 NFSERR_EXIST, 425 NFSERR_NOTDIR, 426 NFSERR_INVAL, 427 NFSERR_ROFS, 428 NFSERR_NAMETOL, 429 NFSERR_NOTEMPTY, 430 NFSERR_STALE, 431 NFSERR_BADHANDLE, 432 NFSERR_NOTSUPP, 433 NFSERR_SERVERFAULT, 434 0, 435 }; 436 437 static const short nfsv3err_rename[] = { 438 NFSERR_IO, 439 NFSERR_NOENT, 440 NFSERR_IO, 441 NFSERR_ACCES, 442 NFSERR_EXIST, 443 NFSERR_XDEV, 444 NFSERR_NOTDIR, 445 NFSERR_ISDIR, 446 NFSERR_INVAL, 447 NFSERR_NOSPC, 448 NFSERR_ROFS, 449 NFSERR_MLINK, 450 NFSERR_NAMETOL, 451 NFSERR_NOTEMPTY, 452 NFSERR_DQUOT, 453 NFSERR_STALE, 454 NFSERR_BADHANDLE, 455 NFSERR_NOTSUPP, 456 NFSERR_SERVERFAULT, 457 0, 458 }; 459 460 static const short nfsv3err_link[] = { 461 NFSERR_IO, 462 NFSERR_IO, 463 NFSERR_ACCES, 464 NFSERR_EXIST, 465 NFSERR_XDEV, 466 NFSERR_NOTDIR, 467 NFSERR_INVAL, 468 NFSERR_NOSPC, 469 NFSERR_ROFS, 470 NFSERR_MLINK, 471 NFSERR_NAMETOL, 472 NFSERR_DQUOT, 473 NFSERR_STALE, 474 NFSERR_BADHANDLE, 475 NFSERR_NOTSUPP, 476 NFSERR_SERVERFAULT, 477 0, 478 }; 479 480 static const short nfsv3err_readdir[] = { 481 NFSERR_IO, 482 NFSERR_IO, 483 NFSERR_ACCES, 484 NFSERR_NOTDIR, 485 NFSERR_STALE, 486 NFSERR_BADHANDLE, 487 NFSERR_BAD_COOKIE, 488 NFSERR_TOOSMALL, 489 NFSERR_SERVERFAULT, 490 0, 491 }; 492 493 static const short nfsv3err_readdirplus[] = { 494 NFSERR_IO, 495 NFSERR_IO, 496 NFSERR_ACCES, 497 NFSERR_NOTDIR, 498 NFSERR_STALE, 499 NFSERR_BADHANDLE, 500 NFSERR_BAD_COOKIE, 501 NFSERR_NOTSUPP, 502 NFSERR_TOOSMALL, 503 NFSERR_SERVERFAULT, 504 0, 505 }; 506 507 static const short nfsv3err_fsstat[] = { 508 NFSERR_IO, 509 NFSERR_IO, 510 NFSERR_STALE, 511 NFSERR_BADHANDLE, 512 NFSERR_SERVERFAULT, 513 0, 514 }; 515 516 static const short nfsv3err_fsinfo[] = { 517 NFSERR_STALE, 518 NFSERR_STALE, 519 NFSERR_BADHANDLE, 520 NFSERR_SERVERFAULT, 521 0, 522 }; 523 524 static const short nfsv3err_pathconf[] = { 525 NFSERR_STALE, 526 NFSERR_STALE, 527 NFSERR_BADHANDLE, 528 NFSERR_SERVERFAULT, 529 0, 530 }; 531 532 static const short nfsv3err_commit[] = { 533 NFSERR_IO, 534 NFSERR_IO, 535 NFSERR_STALE, 536 NFSERR_BADHANDLE, 537 NFSERR_SERVERFAULT, 538 0, 539 }; 540 541 static const short * const nfsrv_v3errmap[] = { 542 nfsv3err_null, 543 nfsv3err_getattr, 544 nfsv3err_setattr, 545 nfsv3err_lookup, 546 nfsv3err_access, 547 nfsv3err_readlink, 548 nfsv3err_read, 549 nfsv3err_write, 550 nfsv3err_create, 551 nfsv3err_mkdir, 552 nfsv3err_symlink, 553 nfsv3err_mknod, 554 nfsv3err_remove, 555 nfsv3err_rmdir, 556 nfsv3err_rename, 557 nfsv3err_link, 558 nfsv3err_readdir, 559 nfsv3err_readdirplus, 560 nfsv3err_fsstat, 561 nfsv3err_fsinfo, 562 nfsv3err_pathconf, 563 nfsv3err_commit, 564 }; 565 566 extern struct nfsrtt nfsrtt; 567 extern struct nfsnodehashhead *nfsnodehashtbl; 568 extern u_long nfsnodehash; 569 570 u_long nfsdirhashmask; 571 572 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *)); 573 574 /* 575 * Create the header for an rpc request packet 576 * The hsiz is the size of the rest of the nfs request header. 577 * (just used to decide if a cluster is a good idea) 578 */ 579 struct mbuf * 580 nfsm_reqh(struct nfsnode *np, u_long procid, int hsiz, char **bposp) 581 { 582 struct mbuf *mb; 583 char *bpos; 584 585 mb = m_get(M_WAIT, MT_DATA); 586 MCLAIM(mb, &nfs_mowner); 587 if (hsiz >= MINCLSIZE) 588 m_clget(mb, M_WAIT); 589 mb->m_len = 0; 590 bpos = mtod(mb, void *); 591 592 /* Finally, return values */ 593 *bposp = bpos; 594 return (mb); 595 } 596 597 /* 598 * Build the RPC header and fill in the authorization info. 599 * The authorization string argument is only used when the credentials 600 * come from outside of the kernel. 601 * Returns the head of the mbuf list. 602 */ 603 struct mbuf * 604 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len, 605 verf_str, mrest, mrest_len, mbp, xidp) 606 kauth_cred_t cr; 607 int nmflag; 608 int procid; 609 int auth_type; 610 int auth_len; 611 char *auth_str; 612 int verf_len; 613 char *verf_str; 614 struct mbuf *mrest; 615 int mrest_len; 616 struct mbuf **mbp; 617 u_int32_t *xidp; 618 { 619 struct mbuf *mb; 620 u_int32_t *tl; 621 char *bpos; 622 int i; 623 struct mbuf *mreq; 624 int siz, grpsiz, authsiz; 625 626 authsiz = nfsm_rndup(auth_len); 627 mb = m_gethdr(M_WAIT, MT_DATA); 628 MCLAIM(mb, &nfs_mowner); 629 if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) { 630 m_clget(mb, M_WAIT); 631 } else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) { 632 MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED); 633 } else { 634 MH_ALIGN(mb, 8 * NFSX_UNSIGNED); 635 } 636 mb->m_len = 0; 637 mreq = mb; 638 bpos = mtod(mb, void *); 639 640 /* 641 * First the RPC header. 642 */ 643 nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 644 645 *tl++ = *xidp = nfs_getxid(); 646 *tl++ = rpc_call; 647 *tl++ = rpc_vers; 648 *tl++ = txdr_unsigned(NFS_PROG); 649 if (nmflag & NFSMNT_NFSV3) 650 *tl++ = txdr_unsigned(NFS_VER3); 651 else 652 *tl++ = txdr_unsigned(NFS_VER2); 653 if (nmflag & NFSMNT_NFSV3) 654 *tl++ = txdr_unsigned(procid); 655 else 656 *tl++ = txdr_unsigned(nfsv2_procid[procid]); 657 658 /* 659 * And then the authorization cred. 660 */ 661 *tl++ = txdr_unsigned(auth_type); 662 *tl = txdr_unsigned(authsiz); 663 switch (auth_type) { 664 case RPCAUTH_UNIX: 665 nfsm_build(tl, u_int32_t *, auth_len); 666 *tl++ = 0; /* stamp ?? */ 667 *tl++ = 0; /* NULL hostname */ 668 *tl++ = txdr_unsigned(kauth_cred_geteuid(cr)); 669 *tl++ = txdr_unsigned(kauth_cred_getegid(cr)); 670 grpsiz = (auth_len >> 2) - 5; 671 *tl++ = txdr_unsigned(grpsiz); 672 for (i = 0; i < grpsiz; i++) 673 *tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */ 674 break; 675 case RPCAUTH_KERB4: 676 siz = auth_len; 677 while (siz > 0) { 678 if (M_TRAILINGSPACE(mb) == 0) { 679 struct mbuf *mb2; 680 mb2 = m_get(M_WAIT, MT_DATA); 681 MCLAIM(mb2, &nfs_mowner); 682 if (siz >= MINCLSIZE) 683 m_clget(mb2, M_WAIT); 684 mb->m_next = mb2; 685 mb = mb2; 686 mb->m_len = 0; 687 bpos = mtod(mb, void *); 688 } 689 i = min(siz, M_TRAILINGSPACE(mb)); 690 memcpy(bpos, auth_str, i); 691 mb->m_len += i; 692 auth_str += i; 693 bpos += i; 694 siz -= i; 695 } 696 if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) { 697 for (i = 0; i < siz; i++) 698 *bpos++ = '\0'; 699 mb->m_len += siz; 700 } 701 break; 702 }; 703 704 /* 705 * And the verifier... 706 */ 707 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 708 if (verf_str) { 709 *tl++ = txdr_unsigned(RPCAUTH_KERB4); 710 *tl = txdr_unsigned(verf_len); 711 siz = verf_len; 712 while (siz > 0) { 713 if (M_TRAILINGSPACE(mb) == 0) { 714 struct mbuf *mb2; 715 mb2 = m_get(M_WAIT, MT_DATA); 716 MCLAIM(mb2, &nfs_mowner); 717 if (siz >= MINCLSIZE) 718 m_clget(mb2, M_WAIT); 719 mb->m_next = mb2; 720 mb = mb2; 721 mb->m_len = 0; 722 bpos = mtod(mb, void *); 723 } 724 i = min(siz, M_TRAILINGSPACE(mb)); 725 memcpy(bpos, verf_str, i); 726 mb->m_len += i; 727 verf_str += i; 728 bpos += i; 729 siz -= i; 730 } 731 if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) { 732 for (i = 0; i < siz; i++) 733 *bpos++ = '\0'; 734 mb->m_len += siz; 735 } 736 } else { 737 *tl++ = txdr_unsigned(RPCAUTH_NULL); 738 *tl = 0; 739 } 740 mb->m_next = mrest; 741 mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len; 742 mreq->m_pkthdr.rcvif = (struct ifnet *)0; 743 *mbp = mb; 744 return (mreq); 745 } 746 747 /* 748 * copies mbuf chain to the uio scatter/gather list 749 */ 750 int 751 nfsm_mbuftouio(mrep, uiop, siz, dpos) 752 struct mbuf **mrep; 753 struct uio *uiop; 754 int siz; 755 char **dpos; 756 { 757 char *mbufcp, *uiocp; 758 int xfer, left, len; 759 struct mbuf *mp; 760 long uiosiz, rem; 761 int error = 0; 762 763 mp = *mrep; 764 mbufcp = *dpos; 765 len = mtod(mp, char *) + mp->m_len - mbufcp; 766 rem = nfsm_rndup(siz)-siz; 767 while (siz > 0) { 768 if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) 769 return (EFBIG); 770 left = uiop->uio_iov->iov_len; 771 uiocp = uiop->uio_iov->iov_base; 772 if (left > siz) 773 left = siz; 774 uiosiz = left; 775 while (left > 0) { 776 while (len == 0) { 777 mp = mp->m_next; 778 if (mp == NULL) 779 return (EBADRPC); 780 mbufcp = mtod(mp, void *); 781 len = mp->m_len; 782 } 783 xfer = (left > len) ? len : left; 784 error = copyout_vmspace(uiop->uio_vmspace, mbufcp, 785 uiocp, xfer); 786 if (error) { 787 return error; 788 } 789 left -= xfer; 790 len -= xfer; 791 mbufcp += xfer; 792 uiocp += xfer; 793 uiop->uio_offset += xfer; 794 uiop->uio_resid -= xfer; 795 } 796 if (uiop->uio_iov->iov_len <= siz) { 797 uiop->uio_iovcnt--; 798 uiop->uio_iov++; 799 } else { 800 uiop->uio_iov->iov_base = 801 (char *)uiop->uio_iov->iov_base + uiosiz; 802 uiop->uio_iov->iov_len -= uiosiz; 803 } 804 siz -= uiosiz; 805 } 806 *dpos = mbufcp; 807 *mrep = mp; 808 if (rem > 0) { 809 if (len < rem) 810 error = nfs_adv(mrep, dpos, rem, len); 811 else 812 *dpos += rem; 813 } 814 return (error); 815 } 816 817 /* 818 * copies a uio scatter/gather list to an mbuf chain. 819 * NOTE: can ony handle iovcnt == 1 820 */ 821 int 822 nfsm_uiotombuf(uiop, mq, siz, bpos) 823 struct uio *uiop; 824 struct mbuf **mq; 825 int siz; 826 char **bpos; 827 { 828 char *uiocp; 829 struct mbuf *mp, *mp2; 830 int xfer, left, mlen; 831 int uiosiz, clflg, rem; 832 char *cp; 833 int error; 834 835 #ifdef DIAGNOSTIC 836 if (uiop->uio_iovcnt != 1) 837 panic("nfsm_uiotombuf: iovcnt != 1"); 838 #endif 839 840 if (siz > MLEN) /* or should it >= MCLBYTES ?? */ 841 clflg = 1; 842 else 843 clflg = 0; 844 rem = nfsm_rndup(siz)-siz; 845 mp = mp2 = *mq; 846 while (siz > 0) { 847 left = uiop->uio_iov->iov_len; 848 uiocp = uiop->uio_iov->iov_base; 849 if (left > siz) 850 left = siz; 851 uiosiz = left; 852 while (left > 0) { 853 mlen = M_TRAILINGSPACE(mp); 854 if (mlen == 0) { 855 mp = m_get(M_WAIT, MT_DATA); 856 MCLAIM(mp, &nfs_mowner); 857 if (clflg) 858 m_clget(mp, M_WAIT); 859 mp->m_len = 0; 860 mp2->m_next = mp; 861 mp2 = mp; 862 mlen = M_TRAILINGSPACE(mp); 863 } 864 xfer = (left > mlen) ? mlen : left; 865 cp = mtod(mp, char *) + mp->m_len; 866 error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp, 867 xfer); 868 if (error) { 869 /* XXX */ 870 } 871 mp->m_len += xfer; 872 left -= xfer; 873 uiocp += xfer; 874 uiop->uio_offset += xfer; 875 uiop->uio_resid -= xfer; 876 } 877 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + 878 uiosiz; 879 uiop->uio_iov->iov_len -= uiosiz; 880 siz -= uiosiz; 881 } 882 if (rem > 0) { 883 if (rem > M_TRAILINGSPACE(mp)) { 884 mp = m_get(M_WAIT, MT_DATA); 885 MCLAIM(mp, &nfs_mowner); 886 mp->m_len = 0; 887 mp2->m_next = mp; 888 } 889 cp = mtod(mp, char *) + mp->m_len; 890 for (left = 0; left < rem; left++) 891 *cp++ = '\0'; 892 mp->m_len += rem; 893 *bpos = cp; 894 } else 895 *bpos = mtod(mp, char *) + mp->m_len; 896 *mq = mp; 897 return (0); 898 } 899 900 /* 901 * Get at least "siz" bytes of correctly aligned data. 902 * When called the mbuf pointers are not necessarily correct, 903 * dsosp points to what ought to be in m_data and left contains 904 * what ought to be in m_len. 905 * This is used by the macros nfsm_dissect and nfsm_dissecton for tough 906 * cases. (The macros use the vars. dpos and dpos2) 907 */ 908 int 909 nfsm_disct(mdp, dposp, siz, left, cp2) 910 struct mbuf **mdp; 911 char **dposp; 912 int siz; 913 int left; 914 char **cp2; 915 { 916 struct mbuf *m1, *m2; 917 struct mbuf *havebuf = NULL; 918 char *src = *dposp; 919 char *dst; 920 int len; 921 922 #ifdef DEBUG 923 if (left < 0) 924 panic("nfsm_disct: left < 0"); 925 #endif 926 m1 = *mdp; 927 /* 928 * Skip through the mbuf chain looking for an mbuf with 929 * some data. If the first mbuf found has enough data 930 * and it is correctly aligned return it. 931 */ 932 while (left == 0) { 933 havebuf = m1; 934 *mdp = m1 = m1->m_next; 935 if (m1 == NULL) 936 return (EBADRPC); 937 src = mtod(m1, void *); 938 left = m1->m_len; 939 /* 940 * If we start a new mbuf and it is big enough 941 * and correctly aligned just return it, don't 942 * do any pull up. 943 */ 944 if (left >= siz && nfsm_aligned(src)) { 945 *cp2 = src; 946 *dposp = src + siz; 947 return (0); 948 } 949 } 950 if (m1->m_flags & M_EXT) { 951 if (havebuf) { 952 /* If the first mbuf with data has external data 953 * and there is a previous empty mbuf use it 954 * to move the data into. 955 */ 956 m2 = m1; 957 *mdp = m1 = havebuf; 958 if (m1->m_flags & M_EXT) { 959 MEXTREMOVE(m1); 960 } 961 } else { 962 /* 963 * If the first mbuf has a external data 964 * and there is no previous empty mbuf 965 * allocate a new mbuf and move the external 966 * data to the new mbuf. Also make the first 967 * mbuf look empty. 968 */ 969 m2 = m_get(M_WAIT, MT_DATA); 970 m2->m_ext = m1->m_ext; 971 m2->m_data = src; 972 m2->m_len = left; 973 MCLADDREFERENCE(m1, m2); 974 MEXTREMOVE(m1); 975 m2->m_next = m1->m_next; 976 m1->m_next = m2; 977 } 978 m1->m_len = 0; 979 if (m1->m_flags & M_PKTHDR) 980 dst = m1->m_pktdat; 981 else 982 dst = m1->m_dat; 983 m1->m_data = dst; 984 } else { 985 /* 986 * If the first mbuf has no external data 987 * move the data to the front of the mbuf. 988 */ 989 if (m1->m_flags & M_PKTHDR) 990 dst = m1->m_pktdat; 991 else 992 dst = m1->m_dat; 993 m1->m_data = dst; 994 if (dst != src) 995 memmove(dst, src, left); 996 dst += left; 997 m1->m_len = left; 998 m2 = m1->m_next; 999 } 1000 *cp2 = m1->m_data; 1001 *dposp = mtod(m1, char *) + siz; 1002 /* 1003 * Loop through mbufs pulling data up into first mbuf until 1004 * the first mbuf is full or there is no more data to 1005 * pullup. 1006 */ 1007 while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) { 1008 if ((len = min(len, m2->m_len)) != 0) 1009 memcpy(dst, m2->m_data, len); 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(mdp, dposp, offs, left) 1026 struct mbuf **mdp; 1027 char **dposp; 1028 int offs; 1029 int left; 1030 { 1031 struct mbuf *m; 1032 int s; 1033 1034 m = *mdp; 1035 s = left; 1036 while (s < offs) { 1037 offs -= s; 1038 m = m->m_next; 1039 if (m == NULL) 1040 return (EBADRPC); 1041 s = m->m_len; 1042 } 1043 *mdp = m; 1044 *dposp = mtod(m, char *) + offs; 1045 return (0); 1046 } 1047 1048 /* 1049 * Copy a string into mbufs for the hard cases... 1050 */ 1051 int 1052 nfsm_strtmbuf(mb, bpos, cp, siz) 1053 struct mbuf **mb; 1054 char **bpos; 1055 const char *cp; 1056 long siz; 1057 { 1058 struct mbuf *m1 = NULL, *m2; 1059 long left, xfer, len, tlen; 1060 u_int32_t *tl; 1061 int putsize; 1062 1063 putsize = 1; 1064 m2 = *mb; 1065 left = M_TRAILINGSPACE(m2); 1066 if (left > 0) { 1067 tl = ((u_int32_t *)(*bpos)); 1068 *tl++ = txdr_unsigned(siz); 1069 putsize = 0; 1070 left -= NFSX_UNSIGNED; 1071 m2->m_len += NFSX_UNSIGNED; 1072 if (left > 0) { 1073 memcpy((void *) tl, cp, left); 1074 siz -= left; 1075 cp += left; 1076 m2->m_len += left; 1077 left = 0; 1078 } 1079 } 1080 /* Loop around adding mbufs */ 1081 while (siz > 0) { 1082 m1 = m_get(M_WAIT, MT_DATA); 1083 MCLAIM(m1, &nfs_mowner); 1084 if (siz > MLEN) 1085 m_clget(m1, M_WAIT); 1086 m1->m_len = NFSMSIZ(m1); 1087 m2->m_next = m1; 1088 m2 = m1; 1089 tl = mtod(m1, u_int32_t *); 1090 tlen = 0; 1091 if (putsize) { 1092 *tl++ = txdr_unsigned(siz); 1093 m1->m_len -= NFSX_UNSIGNED; 1094 tlen = NFSX_UNSIGNED; 1095 putsize = 0; 1096 } 1097 if (siz < m1->m_len) { 1098 len = nfsm_rndup(siz); 1099 xfer = siz; 1100 if (xfer < len) 1101 *(tl+(xfer>>2)) = 0; 1102 } else { 1103 xfer = len = m1->m_len; 1104 } 1105 memcpy((void *) tl, cp, xfer); 1106 m1->m_len = len+tlen; 1107 siz -= xfer; 1108 cp += xfer; 1109 } 1110 *mb = m1; 1111 *bpos = mtod(m1, char *) + m1->m_len; 1112 return (0); 1113 } 1114 1115 /* 1116 * Directory caching routines. They work as follows: 1117 * - a cache is maintained per VDIR nfsnode. 1118 * - for each offset cookie that is exported to userspace, and can 1119 * thus be thrown back at us as an offset to VOP_READDIR, store 1120 * information in the cache. 1121 * - cached are: 1122 * - cookie itself 1123 * - blocknumber (essentially just a search key in the buffer cache) 1124 * - entry number in block. 1125 * - offset cookie of block in which this entry is stored 1126 * - 32 bit cookie if NFSMNT_XLATECOOKIE is used. 1127 * - entries are looked up in a hash table 1128 * - also maintained is an LRU list of entries, used to determine 1129 * which ones to delete if the cache grows too large. 1130 * - if 32 <-> 64 translation mode is requested for a filesystem, 1131 * the cache also functions as a translation table 1132 * - in the translation case, invalidating the cache does not mean 1133 * flushing it, but just marking entries as invalid, except for 1134 * the <64bit cookie, 32bitcookie> pair which is still valid, to 1135 * still be able to use the cache as a translation table. 1136 * - 32 bit cookies are uniquely created by combining the hash table 1137 * entry value, and one generation count per hash table entry, 1138 * incremented each time an entry is appended to the chain. 1139 * - the cache is invalidated each time a direcory is modified 1140 * - sanity checks are also done; if an entry in a block turns 1141 * out not to have a matching cookie, the cache is invalidated 1142 * and a new block starting from the wanted offset is fetched from 1143 * the server. 1144 * - directory entries as read from the server are extended to contain 1145 * the 64bit and, optionally, the 32bit cookies, for sanity checking 1146 * the cache and exporting them to userspace through the cookie 1147 * argument to VOP_READDIR. 1148 */ 1149 1150 u_long 1151 nfs_dirhash(off) 1152 off_t off; 1153 { 1154 int i; 1155 char *cp = (char *)&off; 1156 u_long sum = 0L; 1157 1158 for (i = 0 ; i < sizeof (off); i++) 1159 sum += *cp++; 1160 1161 return sum; 1162 } 1163 1164 #define _NFSDC_MTX(np) (&NFSTOV(np)->v_interlock) 1165 #define NFSDC_LOCK(np) simple_lock(_NFSDC_MTX(np)) 1166 #define NFSDC_UNLOCK(np) simple_unlock(_NFSDC_MTX(np)) 1167 #define NFSDC_ASSERT_LOCKED(np) LOCK_ASSERT(simple_lock_held(_NFSDC_MTX(np))) 1168 1169 void 1170 nfs_initdircache(vp) 1171 struct vnode *vp; 1172 { 1173 struct nfsnode *np = VTONFS(vp); 1174 struct nfsdirhashhead *dircache; 1175 1176 dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF, 1177 M_WAITOK, &nfsdirhashmask); 1178 1179 NFSDC_LOCK(np); 1180 if (np->n_dircache == NULL) { 1181 np->n_dircachesize = 0; 1182 np->n_dircache = dircache; 1183 dircache = NULL; 1184 TAILQ_INIT(&np->n_dirchain); 1185 } 1186 NFSDC_UNLOCK(np); 1187 if (dircache) 1188 hashdone(dircache, M_NFSDIROFF); 1189 } 1190 1191 void 1192 nfs_initdirxlatecookie(vp) 1193 struct vnode *vp; 1194 { 1195 struct nfsnode *np = VTONFS(vp); 1196 unsigned *dirgens; 1197 1198 KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE); 1199 1200 dirgens = malloc(NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF, 1201 M_WAITOK|M_ZERO); 1202 NFSDC_LOCK(np); 1203 if (np->n_dirgens == NULL) { 1204 np->n_dirgens = dirgens; 1205 dirgens = NULL; 1206 } 1207 NFSDC_UNLOCK(np); 1208 if (dirgens) 1209 free(dirgens, M_NFSDIROFF); 1210 } 1211 1212 static const struct nfsdircache dzero; 1213 1214 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *)); 1215 static void nfs_putdircache_unlocked __P((struct nfsnode *, 1216 struct nfsdircache *)); 1217 1218 static void 1219 nfs_unlinkdircache(np, ndp) 1220 struct nfsnode *np; 1221 struct nfsdircache *ndp; 1222 { 1223 1224 NFSDC_ASSERT_LOCKED(np); 1225 KASSERT(ndp != &dzero); 1226 1227 if (LIST_NEXT(ndp, dc_hash) == (void *)-1) 1228 return; 1229 1230 TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain); 1231 LIST_REMOVE(ndp, dc_hash); 1232 LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */ 1233 1234 nfs_putdircache_unlocked(np, ndp); 1235 } 1236 1237 void 1238 nfs_putdircache(np, ndp) 1239 struct nfsnode *np; 1240 struct nfsdircache *ndp; 1241 { 1242 int ref; 1243 1244 if (ndp == &dzero) 1245 return; 1246 1247 KASSERT(ndp->dc_refcnt > 0); 1248 NFSDC_LOCK(np); 1249 ref = --ndp->dc_refcnt; 1250 NFSDC_UNLOCK(np); 1251 1252 if (ref == 0) 1253 free(ndp, M_NFSDIROFF); 1254 } 1255 1256 static void 1257 nfs_putdircache_unlocked(struct nfsnode *np, struct nfsdircache *ndp) 1258 { 1259 int ref; 1260 1261 NFSDC_ASSERT_LOCKED(np); 1262 1263 if (ndp == &dzero) 1264 return; 1265 1266 KASSERT(ndp->dc_refcnt > 0); 1267 ref = --ndp->dc_refcnt; 1268 if (ref == 0) 1269 free(ndp, M_NFSDIROFF); 1270 } 1271 1272 struct nfsdircache * 1273 nfs_searchdircache(vp, off, do32, hashent) 1274 struct vnode *vp; 1275 off_t off; 1276 int do32; 1277 int *hashent; 1278 { 1279 struct nfsdirhashhead *ndhp; 1280 struct nfsdircache *ndp = NULL; 1281 struct nfsnode *np = VTONFS(vp); 1282 unsigned ent; 1283 1284 /* 1285 * Zero is always a valid cookie. 1286 */ 1287 if (off == 0) 1288 /* XXXUNCONST */ 1289 return (struct nfsdircache *)__UNCONST(&dzero); 1290 1291 if (!np->n_dircache) 1292 return NULL; 1293 1294 /* 1295 * We use a 32bit cookie as search key, directly reconstruct 1296 * the hashentry. Else use the hashfunction. 1297 */ 1298 if (do32) { 1299 ent = (u_int32_t)off >> 24; 1300 if (ent >= NFS_DIRHASHSIZ) 1301 return NULL; 1302 ndhp = &np->n_dircache[ent]; 1303 } else { 1304 ndhp = NFSDIRHASH(np, off); 1305 } 1306 1307 if (hashent) 1308 *hashent = (int)(ndhp - np->n_dircache); 1309 1310 NFSDC_LOCK(np); 1311 if (do32) { 1312 LIST_FOREACH(ndp, ndhp, dc_hash) { 1313 if (ndp->dc_cookie32 == (u_int32_t)off) { 1314 /* 1315 * An invalidated entry will become the 1316 * start of a new block fetched from 1317 * the server. 1318 */ 1319 if (ndp->dc_flags & NFSDC_INVALID) { 1320 ndp->dc_blkcookie = ndp->dc_cookie; 1321 ndp->dc_entry = 0; 1322 ndp->dc_flags &= ~NFSDC_INVALID; 1323 } 1324 break; 1325 } 1326 } 1327 } else { 1328 LIST_FOREACH(ndp, ndhp, dc_hash) { 1329 if (ndp->dc_cookie == off) 1330 break; 1331 } 1332 } 1333 if (ndp != NULL) 1334 ndp->dc_refcnt++; 1335 NFSDC_UNLOCK(np); 1336 return ndp; 1337 } 1338 1339 1340 struct nfsdircache * 1341 nfs_enterdircache(struct vnode *vp, off_t off, off_t blkoff, int en, 1342 daddr_t blkno) 1343 { 1344 struct nfsnode *np = VTONFS(vp); 1345 struct nfsdirhashhead *ndhp; 1346 struct nfsdircache *ndp = NULL; 1347 struct nfsdircache *newndp = NULL; 1348 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1349 int hashent = 0, gen, overwrite; /* XXX: GCC */ 1350 1351 /* 1352 * XXX refuse entries for offset 0. amd(8) erroneously sets 1353 * cookie 0 for the '.' entry, making this necessary. This 1354 * isn't so bad, as 0 is a special case anyway. 1355 */ 1356 if (off == 0) 1357 /* XXXUNCONST */ 1358 return (struct nfsdircache *)__UNCONST(&dzero); 1359 1360 if (!np->n_dircache) 1361 /* 1362 * XXX would like to do this in nfs_nget but vtype 1363 * isn't known at that time. 1364 */ 1365 nfs_initdircache(vp); 1366 1367 if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens) 1368 nfs_initdirxlatecookie(vp); 1369 1370 retry: 1371 ndp = nfs_searchdircache(vp, off, 0, &hashent); 1372 1373 NFSDC_LOCK(np); 1374 if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) { 1375 /* 1376 * Overwriting an old entry. Check if it's the same. 1377 * If so, just return. If not, remove the old entry. 1378 */ 1379 if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en) 1380 goto done; 1381 nfs_unlinkdircache(np, ndp); 1382 nfs_putdircache_unlocked(np, ndp); 1383 ndp = NULL; 1384 } 1385 1386 ndhp = &np->n_dircache[hashent]; 1387 1388 if (!ndp) { 1389 if (newndp == NULL) { 1390 NFSDC_UNLOCK(np); 1391 newndp = malloc(sizeof(*ndp), M_NFSDIROFF, M_WAITOK); 1392 newndp->dc_refcnt = 1; 1393 LIST_NEXT(newndp, dc_hash) = (void *)-1; 1394 goto retry; 1395 } 1396 ndp = newndp; 1397 newndp = NULL; 1398 overwrite = 0; 1399 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 1400 /* 1401 * We're allocating a new entry, so bump the 1402 * generation number. 1403 */ 1404 KASSERT(np->n_dirgens); 1405 gen = ++np->n_dirgens[hashent]; 1406 if (gen == 0) { 1407 np->n_dirgens[hashent]++; 1408 gen++; 1409 } 1410 ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff); 1411 } 1412 } else 1413 overwrite = 1; 1414 1415 ndp->dc_cookie = off; 1416 ndp->dc_blkcookie = blkoff; 1417 ndp->dc_entry = en; 1418 ndp->dc_flags = 0; 1419 1420 if (overwrite) 1421 goto done; 1422 1423 /* 1424 * If the maximum directory cookie cache size has been reached 1425 * for this node, take one off the front. The idea is that 1426 * directories are typically read front-to-back once, so that 1427 * the oldest entries can be thrown away without much performance 1428 * loss. 1429 */ 1430 if (np->n_dircachesize == NFS_MAXDIRCACHE) { 1431 nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain)); 1432 } else 1433 np->n_dircachesize++; 1434 1435 KASSERT(ndp->dc_refcnt == 1); 1436 LIST_INSERT_HEAD(ndhp, ndp, dc_hash); 1437 TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain); 1438 ndp->dc_refcnt++; 1439 done: 1440 KASSERT(ndp->dc_refcnt > 0); 1441 NFSDC_UNLOCK(np); 1442 if (newndp) 1443 nfs_putdircache(np, newndp); 1444 return ndp; 1445 } 1446 1447 void 1448 nfs_invaldircache(vp, flags) 1449 struct vnode *vp; 1450 int flags; 1451 { 1452 struct nfsnode *np = VTONFS(vp); 1453 struct nfsdircache *ndp = NULL; 1454 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1455 const bool forcefree = flags & NFS_INVALDIRCACHE_FORCE; 1456 1457 #ifdef DIAGNOSTIC 1458 if (vp->v_type != VDIR) 1459 panic("nfs: invaldircache: not dir"); 1460 #endif 1461 1462 if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0) 1463 np->n_flag &= ~NEOFVALID; 1464 1465 if (!np->n_dircache) 1466 return; 1467 1468 NFSDC_LOCK(np); 1469 if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) { 1470 while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) { 1471 KASSERT(!forcefree || ndp->dc_refcnt == 1); 1472 nfs_unlinkdircache(np, ndp); 1473 } 1474 np->n_dircachesize = 0; 1475 if (forcefree && np->n_dirgens) { 1476 FREE(np->n_dirgens, M_NFSDIROFF); 1477 np->n_dirgens = NULL; 1478 } 1479 } else { 1480 TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain) 1481 ndp->dc_flags |= NFSDC_INVALID; 1482 } 1483 1484 NFSDC_UNLOCK(np); 1485 } 1486 1487 /* 1488 * Called once before VFS init to initialize shared and 1489 * server-specific data structures. 1490 */ 1491 static int 1492 nfs_init0(void) 1493 { 1494 nfsrtt.pos = 0; 1495 rpc_vers = txdr_unsigned(RPC_VER2); 1496 rpc_call = txdr_unsigned(RPC_CALL); 1497 rpc_reply = txdr_unsigned(RPC_REPLY); 1498 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED); 1499 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED); 1500 rpc_mismatch = txdr_unsigned(RPC_MISMATCH); 1501 rpc_autherr = txdr_unsigned(RPC_AUTHERR); 1502 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX); 1503 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4); 1504 nfs_prog = txdr_unsigned(NFS_PROG); 1505 nfs_true = txdr_unsigned(true); 1506 nfs_false = txdr_unsigned(false); 1507 nfs_xdrneg1 = txdr_unsigned(-1); 1508 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; 1509 if (nfs_ticks < 1) 1510 nfs_ticks = 1; 1511 #ifdef NFSSERVER 1512 nfsrv_init(0); /* Init server data structures */ 1513 nfsrv_initcache(); /* Init the server request cache */ 1514 #endif /* NFSSERVER */ 1515 1516 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) 1517 nfsdreq_init(); 1518 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */ 1519 1520 /* 1521 * Initialize reply list and start timer 1522 */ 1523 TAILQ_INIT(&nfs_reqq); 1524 nfs_timer(NULL); 1525 MOWNER_ATTACH(&nfs_mowner); 1526 1527 #ifdef NFS 1528 /* Initialize the kqueue structures */ 1529 nfs_kqinit(); 1530 /* Initialize the iod structures */ 1531 nfs_iodinit(); 1532 #endif 1533 return 0; 1534 } 1535 1536 void 1537 nfs_init(void) 1538 { 1539 static ONCE_DECL(nfs_init_once); 1540 1541 RUN_ONCE(&nfs_init_once, nfs_init0); 1542 } 1543 1544 #ifdef NFS 1545 /* 1546 * Called once at VFS init to initialize client-specific data structures. 1547 */ 1548 void 1549 nfs_vfs_init() 1550 { 1551 /* Initialize NFS server / client shared data. */ 1552 nfs_init(); 1553 1554 nfs_nhinit(); /* Init the nfsnode table */ 1555 nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4); 1556 } 1557 1558 void 1559 nfs_vfs_reinit() 1560 { 1561 nfs_nhreinit(); 1562 } 1563 1564 void 1565 nfs_vfs_done() 1566 { 1567 nfs_nhdone(); 1568 } 1569 1570 /* 1571 * Attribute cache routines. 1572 * nfs_loadattrcache() - loads or updates the cache contents from attributes 1573 * that are on the mbuf list 1574 * nfs_getattrcache() - returns valid attributes if found in cache, returns 1575 * error otherwise 1576 */ 1577 1578 /* 1579 * Load the attribute cache (that lives in the nfsnode entry) with 1580 * the values on the mbuf list and 1581 * Iff vap not NULL 1582 * copy the attributes to *vaper 1583 */ 1584 int 1585 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags) 1586 struct vnode **vpp; 1587 struct mbuf **mdp; 1588 char **dposp; 1589 struct vattr *vaper; 1590 int flags; 1591 { 1592 int32_t t1; 1593 char *cp2; 1594 int error = 0; 1595 struct mbuf *md; 1596 int v3 = NFS_ISV3(*vpp); 1597 1598 md = *mdp; 1599 t1 = (mtod(md, char *) + md->m_len) - *dposp; 1600 error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2); 1601 if (error) 1602 return (error); 1603 return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags); 1604 } 1605 1606 int 1607 nfs_loadattrcache(vpp, fp, vaper, flags) 1608 struct vnode **vpp; 1609 struct nfs_fattr *fp; 1610 struct vattr *vaper; 1611 int flags; 1612 { 1613 struct vnode *vp = *vpp; 1614 struct vattr *vap; 1615 int v3 = NFS_ISV3(vp); 1616 enum vtype vtyp; 1617 u_short vmode; 1618 struct timespec mtime; 1619 struct timespec ctime; 1620 struct vnode *nvp; 1621 int32_t rdev; 1622 struct nfsnode *np; 1623 extern int (**spec_nfsv2nodeop_p) __P((void *)); 1624 uid_t uid; 1625 gid_t gid; 1626 1627 if (v3) { 1628 vtyp = nfsv3tov_type(fp->fa_type); 1629 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1630 rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1), 1631 fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2)); 1632 fxdr_nfsv3time(&fp->fa3_mtime, &mtime); 1633 fxdr_nfsv3time(&fp->fa3_ctime, &ctime); 1634 } else { 1635 vtyp = nfsv2tov_type(fp->fa_type); 1636 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1637 if (vtyp == VNON || vtyp == VREG) 1638 vtyp = IFTOVT(vmode); 1639 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); 1640 fxdr_nfsv2time(&fp->fa2_mtime, &mtime); 1641 ctime.tv_sec = fxdr_unsigned(u_int32_t, 1642 fp->fa2_ctime.nfsv2_sec); 1643 ctime.tv_nsec = 0; 1644 1645 /* 1646 * Really ugly NFSv2 kludge. 1647 */ 1648 if (vtyp == VCHR && rdev == 0xffffffff) 1649 vtyp = VFIFO; 1650 } 1651 1652 vmode &= ALLPERMS; 1653 1654 /* 1655 * If v_type == VNON it is a new node, so fill in the v_type, 1656 * n_mtime fields. Check to see if it represents a special 1657 * device, and if so, check for a possible alias. Once the 1658 * correct vnode has been obtained, fill in the rest of the 1659 * information. 1660 */ 1661 np = VTONFS(vp); 1662 if (vp->v_type == VNON) { 1663 vp->v_type = vtyp; 1664 if (vp->v_type == VFIFO) { 1665 extern int (**fifo_nfsv2nodeop_p) __P((void *)); 1666 vp->v_op = fifo_nfsv2nodeop_p; 1667 } else if (vp->v_type == VREG) { 1668 mutex_init(&np->n_commitlock, MUTEX_DEFAULT, IPL_NONE); 1669 } else if (vp->v_type == VCHR || vp->v_type == VBLK) { 1670 vp->v_op = spec_nfsv2nodeop_p; 1671 nvp = checkalias(vp, (dev_t)rdev, vp->v_mount); 1672 if (nvp) { 1673 /* 1674 * Discard unneeded vnode, but save its nfsnode. 1675 * Since the nfsnode does not have a lock, its 1676 * vnode lock has to be carried over. 1677 */ 1678 /* 1679 * XXX is the old node sure to be locked here? 1680 */ 1681 KASSERT(lockstatus(&vp->v_lock) == 1682 LK_EXCLUSIVE); 1683 nvp->v_data = vp->v_data; 1684 vp->v_data = NULL; 1685 VOP_UNLOCK(vp, 0); 1686 vp->v_op = spec_vnodeop_p; 1687 vrele(vp); 1688 vgone(vp); 1689 lockmgr(&nvp->v_lock, LK_EXCLUSIVE, 1690 &nvp->v_interlock); 1691 /* 1692 * Reinitialize aliased node. 1693 */ 1694 np->n_vnode = nvp; 1695 *vpp = vp = nvp; 1696 } 1697 } 1698 np->n_mtime = mtime; 1699 } 1700 uid = fxdr_unsigned(uid_t, fp->fa_uid); 1701 gid = fxdr_unsigned(gid_t, fp->fa_gid); 1702 vap = np->n_vattr; 1703 1704 /* 1705 * Invalidate access cache if uid, gid, mode or ctime changed. 1706 */ 1707 if (np->n_accstamp != -1 && 1708 (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode 1709 || timespeccmp(&ctime, &vap->va_ctime, !=))) 1710 np->n_accstamp = -1; 1711 1712 vap->va_type = vtyp; 1713 vap->va_mode = vmode; 1714 vap->va_rdev = (dev_t)rdev; 1715 vap->va_mtime = mtime; 1716 vap->va_ctime = ctime; 1717 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 1718 switch (vtyp) { 1719 case VDIR: 1720 vap->va_blocksize = NFS_DIRFRAGSIZ; 1721 break; 1722 case VBLK: 1723 vap->va_blocksize = BLKDEV_IOSIZE; 1724 break; 1725 case VCHR: 1726 vap->va_blocksize = MAXBSIZE; 1727 break; 1728 default: 1729 vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize : 1730 fxdr_unsigned(int32_t, fp->fa2_blocksize); 1731 break; 1732 } 1733 if (v3) { 1734 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1735 vap->va_uid = uid; 1736 vap->va_gid = gid; 1737 vap->va_size = fxdr_hyper(&fp->fa3_size); 1738 vap->va_bytes = fxdr_hyper(&fp->fa3_used); 1739 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid); 1740 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); 1741 vap->va_flags = 0; 1742 vap->va_filerev = 0; 1743 } else { 1744 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1745 vap->va_uid = uid; 1746 vap->va_gid = gid; 1747 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); 1748 vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks) 1749 * NFS_FABLKSIZE; 1750 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); 1751 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); 1752 vap->va_flags = 0; 1753 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec); 1754 vap->va_filerev = 0; 1755 } 1756 if (vap->va_size != np->n_size) { 1757 if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) { 1758 vap->va_size = np->n_size; 1759 } else { 1760 np->n_size = vap->va_size; 1761 if (vap->va_type == VREG) { 1762 /* 1763 * we can't free pages if NAC_NOTRUNC because 1764 * the pages can be owned by ourselves. 1765 */ 1766 if (flags & NAC_NOTRUNC) { 1767 np->n_flag |= NTRUNCDELAYED; 1768 } else { 1769 genfs_node_wrlock(vp); 1770 simple_lock(&vp->v_interlock); 1771 (void)VOP_PUTPAGES(vp, 0, 1772 0, PGO_SYNCIO | PGO_CLEANIT | 1773 PGO_FREE | PGO_ALLPAGES); 1774 uvm_vnp_setsize(vp, np->n_size); 1775 genfs_node_unlock(vp); 1776 } 1777 } 1778 } 1779 } 1780 np->n_attrstamp = time_second; 1781 if (vaper != NULL) { 1782 memcpy((void *)vaper, (void *)vap, sizeof(*vap)); 1783 if (np->n_flag & NCHG) { 1784 if (np->n_flag & NACC) 1785 vaper->va_atime = np->n_atim; 1786 if (np->n_flag & NUPD) 1787 vaper->va_mtime = np->n_mtim; 1788 } 1789 } 1790 return (0); 1791 } 1792 1793 /* 1794 * Check the time stamp 1795 * If the cache is valid, copy contents to *vap and return 0 1796 * otherwise return an error 1797 */ 1798 int 1799 nfs_getattrcache(vp, vaper) 1800 struct vnode *vp; 1801 struct vattr *vaper; 1802 { 1803 struct nfsnode *np = VTONFS(vp); 1804 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1805 struct vattr *vap; 1806 1807 if (np->n_attrstamp == 0 || 1808 (time_second - np->n_attrstamp) >= NFS_ATTRTIMEO(nmp, np)) { 1809 nfsstats.attrcache_misses++; 1810 return (ENOENT); 1811 } 1812 nfsstats.attrcache_hits++; 1813 vap = np->n_vattr; 1814 if (vap->va_size != np->n_size) { 1815 if (vap->va_type == VREG) { 1816 if ((np->n_flag & NMODIFIED) != 0 && 1817 vap->va_size < np->n_size) { 1818 vap->va_size = np->n_size; 1819 } else { 1820 np->n_size = vap->va_size; 1821 } 1822 genfs_node_wrlock(vp); 1823 uvm_vnp_setsize(vp, np->n_size); 1824 genfs_node_unlock(vp); 1825 } else 1826 np->n_size = vap->va_size; 1827 } 1828 memcpy((void *)vaper, (void *)vap, sizeof(struct vattr)); 1829 if (np->n_flag & NCHG) { 1830 if (np->n_flag & NACC) 1831 vaper->va_atime = np->n_atim; 1832 if (np->n_flag & NUPD) 1833 vaper->va_mtime = np->n_mtim; 1834 } 1835 return (0); 1836 } 1837 1838 void 1839 nfs_delayedtruncate(vp) 1840 struct vnode *vp; 1841 { 1842 struct nfsnode *np = VTONFS(vp); 1843 1844 if (np->n_flag & NTRUNCDELAYED) { 1845 np->n_flag &= ~NTRUNCDELAYED; 1846 genfs_node_wrlock(vp); 1847 simple_lock(&vp->v_interlock); 1848 (void)VOP_PUTPAGES(vp, 0, 1849 0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES); 1850 uvm_vnp_setsize(vp, np->n_size); 1851 genfs_node_unlock(vp); 1852 } 1853 } 1854 1855 #define NFS_WCCKLUDGE_TIMEOUT (24 * 60 * 60) /* 1 day */ 1856 #define NFS_WCCKLUDGE(nmp, now) \ 1857 (((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \ 1858 ((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0) 1859 1860 /* 1861 * nfs_check_wccdata: check inaccurate wcc_data 1862 * 1863 * => return non-zero if we shouldn't trust the wcc_data. 1864 * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed". 1865 */ 1866 1867 int 1868 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime, 1869 struct timespec *mtime, bool docheck) 1870 { 1871 int error = 0; 1872 1873 #if !defined(NFS_V2_ONLY) 1874 1875 if (docheck) { 1876 struct vnode *vp = NFSTOV(np); 1877 struct nfsmount *nmp; 1878 long now = time_second; 1879 const struct timespec *omtime = &np->n_vattr->va_mtime; 1880 const struct timespec *octime = &np->n_vattr->va_ctime; 1881 #if defined(DEBUG) 1882 const char *reason = NULL; /* XXX: gcc */ 1883 #endif 1884 1885 if (timespeccmp(omtime, mtime, <=)) { 1886 #if defined(DEBUG) 1887 reason = "mtime"; 1888 #endif 1889 error = EINVAL; 1890 } 1891 1892 if (vp->v_type == VDIR && timespeccmp(octime, ctime, <=)) { 1893 #if defined(DEBUG) 1894 reason = "ctime"; 1895 #endif 1896 error = EINVAL; 1897 } 1898 1899 nmp = VFSTONFS(vp->v_mount); 1900 if (error) { 1901 1902 /* 1903 * despite of the fact that we've updated the file, 1904 * timestamps of the file were not updated as we 1905 * expected. 1906 * it means that the server has incompatible 1907 * semantics of timestamps or (more likely) 1908 * the server time is not precise enough to 1909 * track each modifications. 1910 * in that case, we disable wcc processing. 1911 * 1912 * yes, strictly speaking, we should disable all 1913 * caching. it's a compromise. 1914 */ 1915 1916 mutex_enter(&nmp->nm_lock); 1917 #if defined(DEBUG) 1918 if (!NFS_WCCKLUDGE(nmp, now)) { 1919 printf("%s: inaccurate wcc data (%s) detected," 1920 " disabling wcc" 1921 " (ctime %u.%09u %u.%09u," 1922 " mtime %u.%09u %u.%09u)\n", 1923 vp->v_mount->mnt_stat.f_mntfromname, 1924 reason, 1925 (unsigned int)octime->tv_sec, 1926 (unsigned int)octime->tv_nsec, 1927 (unsigned int)ctime->tv_sec, 1928 (unsigned int)ctime->tv_nsec, 1929 (unsigned int)omtime->tv_sec, 1930 (unsigned int)omtime->tv_nsec, 1931 (unsigned int)mtime->tv_sec, 1932 (unsigned int)mtime->tv_nsec); 1933 } 1934 #endif 1935 nmp->nm_iflag |= NFSMNT_WCCKLUDGE; 1936 nmp->nm_wcckludgetime = now; 1937 mutex_exit(&nmp->nm_lock); 1938 } else if (NFS_WCCKLUDGE(nmp, now)) { 1939 error = EPERM; /* XXX */ 1940 } else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1941 mutex_enter(&nmp->nm_lock); 1942 if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1943 #if defined(DEBUG) 1944 printf("%s: re-enabling wcc\n", 1945 vp->v_mount->mnt_stat.f_mntfromname); 1946 #endif 1947 nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE; 1948 } 1949 mutex_exit(&nmp->nm_lock); 1950 } 1951 } 1952 1953 #endif /* !defined(NFS_V2_ONLY) */ 1954 1955 return error; 1956 } 1957 1958 /* 1959 * Heuristic to see if the server XDR encodes directory cookies or not. 1960 * it is not supposed to, but a lot of servers may do this. Also, since 1961 * most/all servers will implement V2 as well, it is expected that they 1962 * may return just 32 bits worth of cookie information, so we need to 1963 * find out in which 32 bits this information is available. We do this 1964 * to avoid trouble with emulated binaries that can't handle 64 bit 1965 * directory offsets. 1966 */ 1967 1968 void 1969 nfs_cookieheuristic(vp, flagp, l, cred) 1970 struct vnode *vp; 1971 int *flagp; 1972 struct lwp *l; 1973 kauth_cred_t cred; 1974 { 1975 struct uio auio; 1976 struct iovec aiov; 1977 char *tbuf, *cp; 1978 struct dirent *dp; 1979 off_t *cookies = NULL, *cop; 1980 int error, eof, nc, len; 1981 1982 MALLOC(tbuf, void *, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK); 1983 1984 aiov.iov_base = tbuf; 1985 aiov.iov_len = NFS_DIRFRAGSIZ; 1986 auio.uio_iov = &aiov; 1987 auio.uio_iovcnt = 1; 1988 auio.uio_rw = UIO_READ; 1989 auio.uio_resid = NFS_DIRFRAGSIZ; 1990 auio.uio_offset = 0; 1991 UIO_SETUP_SYSSPACE(&auio); 1992 1993 error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc); 1994 1995 len = NFS_DIRFRAGSIZ - auio.uio_resid; 1996 if (error || len == 0) { 1997 FREE(tbuf, M_TEMP); 1998 if (cookies) 1999 free(cookies, M_TEMP); 2000 return; 2001 } 2002 2003 /* 2004 * Find the first valid entry and look at its offset cookie. 2005 */ 2006 2007 cp = tbuf; 2008 for (cop = cookies; len > 0; len -= dp->d_reclen) { 2009 dp = (struct dirent *)cp; 2010 if (dp->d_fileno != 0 && len >= dp->d_reclen) { 2011 if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) { 2012 *flagp |= NFSMNT_SWAPCOOKIE; 2013 nfs_invaldircache(vp, 0); 2014 nfs_vinvalbuf(vp, 0, cred, l, 1); 2015 } 2016 break; 2017 } 2018 cop++; 2019 cp += dp->d_reclen; 2020 } 2021 2022 FREE(tbuf, M_TEMP); 2023 free(cookies, M_TEMP); 2024 } 2025 #endif /* NFS */ 2026 2027 #ifdef NFSSERVER 2028 /* 2029 * Set up nameidata for a lookup() call and do it. 2030 * 2031 * If pubflag is set, this call is done for a lookup operation on the 2032 * public filehandle. In that case we allow crossing mountpoints and 2033 * absolute pathnames. However, the caller is expected to check that 2034 * the lookup result is within the public fs, and deny access if 2035 * it is not. 2036 */ 2037 int 2038 nfs_namei(ndp, nsfh, len, slp, nam, mdp, dposp, retdirp, l, kerbflag, pubflag) 2039 struct nameidata *ndp; 2040 nfsrvfh_t *nsfh; 2041 uint32_t len; 2042 struct nfssvc_sock *slp; 2043 struct mbuf *nam; 2044 struct mbuf **mdp; 2045 char **dposp; 2046 struct vnode **retdirp; 2047 struct lwp *l; 2048 int kerbflag, pubflag; 2049 { 2050 int i, rem; 2051 struct mbuf *md; 2052 char *fromcp, *tocp, *cp; 2053 struct iovec aiov; 2054 struct uio auio; 2055 struct vnode *dp; 2056 int error, rdonly, linklen; 2057 struct componentname *cnp = &ndp->ni_cnd; 2058 2059 *retdirp = NULL; 2060 2061 if ((len + 1) > MAXPATHLEN) 2062 return (ENAMETOOLONG); 2063 if (len == 0) 2064 return (EACCES); 2065 cnp->cn_pnbuf = PNBUF_GET(); 2066 2067 /* 2068 * Copy the name from the mbuf list to ndp->ni_pnbuf 2069 * and set the various ndp fields appropriately. 2070 */ 2071 fromcp = *dposp; 2072 tocp = cnp->cn_pnbuf; 2073 md = *mdp; 2074 rem = mtod(md, char *) + md->m_len - fromcp; 2075 for (i = 0; i < len; i++) { 2076 while (rem == 0) { 2077 md = md->m_next; 2078 if (md == NULL) { 2079 error = EBADRPC; 2080 goto out; 2081 } 2082 fromcp = mtod(md, void *); 2083 rem = md->m_len; 2084 } 2085 if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) { 2086 error = EACCES; 2087 goto out; 2088 } 2089 *tocp++ = *fromcp++; 2090 rem--; 2091 } 2092 *tocp = '\0'; 2093 *mdp = md; 2094 *dposp = fromcp; 2095 len = nfsm_rndup(len)-len; 2096 if (len > 0) { 2097 if (rem >= len) 2098 *dposp += len; 2099 else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0) 2100 goto out; 2101 } 2102 2103 /* 2104 * Extract and set starting directory. 2105 */ 2106 error = nfsrv_fhtovp(nsfh, false, &dp, ndp->ni_cnd.cn_cred, slp, 2107 nam, &rdonly, kerbflag, pubflag); 2108 if (error) 2109 goto out; 2110 if (dp->v_type != VDIR) { 2111 vrele(dp); 2112 error = ENOTDIR; 2113 goto out; 2114 } 2115 2116 if (rdonly) 2117 cnp->cn_flags |= RDONLY; 2118 2119 *retdirp = dp; 2120 2121 if (pubflag) { 2122 /* 2123 * Oh joy. For WebNFS, handle those pesky '%' escapes, 2124 * and the 'native path' indicator. 2125 */ 2126 cp = PNBUF_GET(); 2127 fromcp = cnp->cn_pnbuf; 2128 tocp = cp; 2129 if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) { 2130 switch ((unsigned char)*fromcp) { 2131 case WEBNFS_NATIVE_CHAR: 2132 /* 2133 * 'Native' path for us is the same 2134 * as a path according to the NFS spec, 2135 * just skip the escape char. 2136 */ 2137 fromcp++; 2138 break; 2139 /* 2140 * More may be added in the future, range 0x80-0xff 2141 */ 2142 default: 2143 error = EIO; 2144 vrele(dp); 2145 PNBUF_PUT(cp); 2146 goto out; 2147 } 2148 } 2149 /* 2150 * Translate the '%' escapes, URL-style. 2151 */ 2152 while (*fromcp != '\0') { 2153 if (*fromcp == WEBNFS_ESC_CHAR) { 2154 if (fromcp[1] != '\0' && fromcp[2] != '\0') { 2155 fromcp++; 2156 *tocp++ = HEXSTRTOI(fromcp); 2157 fromcp += 2; 2158 continue; 2159 } else { 2160 error = ENOENT; 2161 vrele(dp); 2162 PNBUF_PUT(cp); 2163 goto out; 2164 } 2165 } else 2166 *tocp++ = *fromcp++; 2167 } 2168 *tocp = '\0'; 2169 PNBUF_PUT(cnp->cn_pnbuf); 2170 cnp->cn_pnbuf = cp; 2171 } 2172 2173 ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1; 2174 ndp->ni_segflg = UIO_SYSSPACE; 2175 ndp->ni_rootdir = rootvnode; 2176 ndp->ni_erootdir = NULL; 2177 2178 if (pubflag) { 2179 ndp->ni_loopcnt = 0; 2180 if (cnp->cn_pnbuf[0] == '/') 2181 dp = rootvnode; 2182 } else { 2183 cnp->cn_flags |= NOCROSSMOUNT; 2184 } 2185 2186 cnp->cn_lwp = l; 2187 VREF(dp); 2188 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); 2189 2190 for (;;) { 2191 cnp->cn_nameptr = cnp->cn_pnbuf; 2192 ndp->ni_startdir = dp; 2193 2194 /* 2195 * And call lookup() to do the real work 2196 */ 2197 error = lookup(ndp); 2198 if (error) { 2199 if (ndp->ni_dvp) { 2200 vput(ndp->ni_dvp); 2201 } 2202 PNBUF_PUT(cnp->cn_pnbuf); 2203 return (error); 2204 } 2205 2206 /* 2207 * Check for encountering a symbolic link 2208 */ 2209 if ((cnp->cn_flags & ISSYMLINK) == 0) { 2210 if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp) { 2211 if (ndp->ni_dvp == ndp->ni_vp) { 2212 vrele(ndp->ni_dvp); 2213 } else { 2214 vput(ndp->ni_dvp); 2215 } 2216 } 2217 if (cnp->cn_flags & (SAVENAME | SAVESTART)) 2218 cnp->cn_flags |= HASBUF; 2219 else 2220 PNBUF_PUT(cnp->cn_pnbuf); 2221 return (0); 2222 } else { 2223 if (!pubflag) { 2224 error = EINVAL; 2225 break; 2226 } 2227 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 2228 error = ELOOP; 2229 break; 2230 } 2231 if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) { 2232 error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred, 2233 cnp->cn_lwp); 2234 if (error != 0) 2235 break; 2236 } 2237 if (ndp->ni_pathlen > 1) 2238 cp = PNBUF_GET(); 2239 else 2240 cp = cnp->cn_pnbuf; 2241 aiov.iov_base = cp; 2242 aiov.iov_len = MAXPATHLEN; 2243 auio.uio_iov = &aiov; 2244 auio.uio_iovcnt = 1; 2245 auio.uio_offset = 0; 2246 auio.uio_rw = UIO_READ; 2247 auio.uio_resid = MAXPATHLEN; 2248 UIO_SETUP_SYSSPACE(&auio); 2249 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 2250 if (error) { 2251 badlink: 2252 if (ndp->ni_pathlen > 1) 2253 PNBUF_PUT(cp); 2254 break; 2255 } 2256 linklen = MAXPATHLEN - auio.uio_resid; 2257 if (linklen == 0) { 2258 error = ENOENT; 2259 goto badlink; 2260 } 2261 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 2262 error = ENAMETOOLONG; 2263 goto badlink; 2264 } 2265 if (ndp->ni_pathlen > 1) { 2266 memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen); 2267 PNBUF_PUT(cnp->cn_pnbuf); 2268 cnp->cn_pnbuf = cp; 2269 } else 2270 cnp->cn_pnbuf[linklen] = '\0'; 2271 ndp->ni_pathlen += linklen; 2272 vput(ndp->ni_vp); 2273 dp = ndp->ni_dvp; 2274 2275 /* 2276 * Check if root directory should replace current directory. 2277 */ 2278 if (cnp->cn_pnbuf[0] == '/') { 2279 vput(dp); 2280 dp = ndp->ni_rootdir; 2281 VREF(dp); 2282 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); 2283 } 2284 } 2285 } 2286 vput(ndp->ni_dvp); 2287 vput(ndp->ni_vp); 2288 ndp->ni_vp = NULL; 2289 out: 2290 PNBUF_PUT(cnp->cn_pnbuf); 2291 return (error); 2292 } 2293 #endif /* NFSSERVER */ 2294 2295 /* 2296 * A fiddled version of m_adj() that ensures null fill to a 32-bit 2297 * boundary and only trims off the back end 2298 * 2299 * 1. trim off 'len' bytes as m_adj(mp, -len). 2300 * 2. add zero-padding 'nul' bytes at the end of the mbuf chain. 2301 */ 2302 void 2303 nfs_zeropad(mp, len, nul) 2304 struct mbuf *mp; 2305 int len; 2306 int nul; 2307 { 2308 struct mbuf *m; 2309 int count; 2310 2311 /* 2312 * Trim from tail. Scan the mbuf chain, 2313 * calculating its length and finding the last mbuf. 2314 * If the adjustment only affects this mbuf, then just 2315 * adjust and return. Otherwise, rescan and truncate 2316 * after the remaining size. 2317 */ 2318 count = 0; 2319 m = mp; 2320 for (;;) { 2321 count += m->m_len; 2322 if (m->m_next == NULL) 2323 break; 2324 m = m->m_next; 2325 } 2326 2327 KDASSERT(count >= len); 2328 2329 if (m->m_len >= len) { 2330 m->m_len -= len; 2331 } else { 2332 count -= len; 2333 /* 2334 * Correct length for chain is "count". 2335 * Find the mbuf with last data, adjust its length, 2336 * and toss data from remaining mbufs on chain. 2337 */ 2338 for (m = mp; m; m = m->m_next) { 2339 if (m->m_len >= count) { 2340 m->m_len = count; 2341 break; 2342 } 2343 count -= m->m_len; 2344 } 2345 KASSERT(m && m->m_next); 2346 m_freem(m->m_next); 2347 m->m_next = NULL; 2348 } 2349 2350 KDASSERT(m->m_next == NULL); 2351 2352 /* 2353 * zero-padding. 2354 */ 2355 if (nul > 0) { 2356 char *cp; 2357 int i; 2358 2359 if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) { 2360 struct mbuf *n; 2361 2362 KDASSERT(MLEN >= nul); 2363 n = m_get(M_WAIT, MT_DATA); 2364 MCLAIM(n, &nfs_mowner); 2365 n->m_len = nul; 2366 n->m_next = NULL; 2367 m->m_next = n; 2368 cp = mtod(n, void *); 2369 } else { 2370 cp = mtod(m, char *) + m->m_len; 2371 m->m_len += nul; 2372 } 2373 for (i = 0; i < nul; i++) 2374 *cp++ = '\0'; 2375 } 2376 return; 2377 } 2378 2379 /* 2380 * Make these functions instead of macros, so that the kernel text size 2381 * doesn't get too big... 2382 */ 2383 void 2384 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp) 2385 struct nfsrv_descript *nfsd; 2386 int before_ret; 2387 struct vattr *before_vap; 2388 int after_ret; 2389 struct vattr *after_vap; 2390 struct mbuf **mbp; 2391 char **bposp; 2392 { 2393 struct mbuf *mb = *mbp; 2394 char *bpos = *bposp; 2395 u_int32_t *tl; 2396 2397 if (before_ret) { 2398 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2399 *tl = nfs_false; 2400 } else { 2401 nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED); 2402 *tl++ = nfs_true; 2403 txdr_hyper(before_vap->va_size, tl); 2404 tl += 2; 2405 txdr_nfsv3time(&(before_vap->va_mtime), tl); 2406 tl += 2; 2407 txdr_nfsv3time(&(before_vap->va_ctime), tl); 2408 } 2409 *bposp = bpos; 2410 *mbp = mb; 2411 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp); 2412 } 2413 2414 void 2415 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp) 2416 struct nfsrv_descript *nfsd; 2417 int after_ret; 2418 struct vattr *after_vap; 2419 struct mbuf **mbp; 2420 char **bposp; 2421 { 2422 struct mbuf *mb = *mbp; 2423 char *bpos = *bposp; 2424 u_int32_t *tl; 2425 struct nfs_fattr *fp; 2426 2427 if (after_ret) { 2428 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2429 *tl = nfs_false; 2430 } else { 2431 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR); 2432 *tl++ = nfs_true; 2433 fp = (struct nfs_fattr *)tl; 2434 nfsm_srvfattr(nfsd, after_vap, fp); 2435 } 2436 *mbp = mb; 2437 *bposp = bpos; 2438 } 2439 2440 void 2441 nfsm_srvfattr(nfsd, vap, fp) 2442 struct nfsrv_descript *nfsd; 2443 struct vattr *vap; 2444 struct nfs_fattr *fp; 2445 { 2446 2447 fp->fa_nlink = txdr_unsigned(vap->va_nlink); 2448 fp->fa_uid = txdr_unsigned(vap->va_uid); 2449 fp->fa_gid = txdr_unsigned(vap->va_gid); 2450 if (nfsd->nd_flag & ND_NFSV3) { 2451 fp->fa_type = vtonfsv3_type(vap->va_type); 2452 fp->fa_mode = vtonfsv3_mode(vap->va_mode); 2453 txdr_hyper(vap->va_size, &fp->fa3_size); 2454 txdr_hyper(vap->va_bytes, &fp->fa3_used); 2455 fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev)); 2456 fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev)); 2457 fp->fa3_fsid.nfsuquad[0] = 0; 2458 fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid); 2459 txdr_hyper(vap->va_fileid, &fp->fa3_fileid); 2460 txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime); 2461 txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime); 2462 txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime); 2463 } else { 2464 fp->fa_type = vtonfsv2_type(vap->va_type); 2465 fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 2466 fp->fa2_size = txdr_unsigned(vap->va_size); 2467 fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize); 2468 if (vap->va_type == VFIFO) 2469 fp->fa2_rdev = 0xffffffff; 2470 else 2471 fp->fa2_rdev = txdr_unsigned(vap->va_rdev); 2472 fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE); 2473 fp->fa2_fsid = txdr_unsigned(vap->va_fsid); 2474 fp->fa2_fileid = txdr_unsigned(vap->va_fileid); 2475 txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime); 2476 txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime); 2477 txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime); 2478 } 2479 } 2480 2481 #ifdef NFSSERVER 2482 /* 2483 * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked) 2484 * - look up fsid in mount list (if not found ret error) 2485 * - get vp and export rights by calling VFS_FHTOVP() 2486 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 2487 * - if not lockflag unlock it with VOP_UNLOCK() 2488 */ 2489 int 2490 nfsrv_fhtovp(nfsrvfh_t *nsfh, int lockflag, struct vnode **vpp, 2491 kauth_cred_t cred, struct nfssvc_sock *slp, struct mbuf *nam, int *rdonlyp, 2492 int kerbflag, int pubflag) 2493 { 2494 struct mount *mp; 2495 kauth_cred_t credanon; 2496 int error, exflags; 2497 struct sockaddr_in *saddr; 2498 fhandle_t *fhp; 2499 2500 fhp = NFSRVFH_FHANDLE(nsfh); 2501 *vpp = (struct vnode *)0; 2502 2503 if (nfs_ispublicfh(nsfh)) { 2504 if (!pubflag || !nfs_pub.np_valid) 2505 return (ESTALE); 2506 fhp = nfs_pub.np_handle; 2507 } 2508 2509 error = netexport_check(&fhp->fh_fsid, nam, &mp, &exflags, &credanon); 2510 if (error) { 2511 return error; 2512 } 2513 2514 error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp); 2515 if (error) 2516 return (error); 2517 2518 if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) { 2519 saddr = mtod(nam, struct sockaddr_in *); 2520 if ((saddr->sin_family == AF_INET) && 2521 ntohs(saddr->sin_port) >= IPPORT_RESERVED) { 2522 vput(*vpp); 2523 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2524 } 2525 #ifdef INET6 2526 if ((saddr->sin_family == AF_INET6) && 2527 ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) { 2528 vput(*vpp); 2529 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2530 } 2531 #endif 2532 } 2533 /* 2534 * Check/setup credentials. 2535 */ 2536 if (exflags & MNT_EXKERB) { 2537 if (!kerbflag) { 2538 vput(*vpp); 2539 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2540 } 2541 } else if (kerbflag) { 2542 vput(*vpp); 2543 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2544 } else if (kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 2545 NULL) == 0 || (exflags & MNT_EXPORTANON)) { 2546 kauth_cred_clone(credanon, cred); 2547 } 2548 if (exflags & MNT_EXRDONLY) 2549 *rdonlyp = 1; 2550 else 2551 *rdonlyp = 0; 2552 if (!lockflag) 2553 VOP_UNLOCK(*vpp, 0); 2554 return (0); 2555 } 2556 2557 /* 2558 * WebNFS: check if a filehandle is a public filehandle. For v3, this 2559 * means a length of 0, for v2 it means all zeroes. 2560 */ 2561 int 2562 nfs_ispublicfh(const nfsrvfh_t *nsfh) 2563 { 2564 const char *cp = (const void *)(NFSRVFH_DATA(nsfh)); 2565 int i; 2566 2567 if (NFSRVFH_SIZE(nsfh) == 0) { 2568 return true; 2569 } 2570 if (NFSRVFH_SIZE(nsfh) != NFSX_V2FH) { 2571 return false; 2572 } 2573 for (i = 0; i < NFSX_V2FH; i++) 2574 if (*cp++ != 0) 2575 return false; 2576 return true; 2577 } 2578 #endif /* NFSSERVER */ 2579 2580 /* 2581 * This function compares two net addresses by family and returns true 2582 * if they are the same host. 2583 * If there is any doubt, return false. 2584 * The AF_INET family is handled as a special case so that address mbufs 2585 * don't need to be saved to store "struct in_addr", which is only 4 bytes. 2586 */ 2587 int 2588 netaddr_match(family, haddr, nam) 2589 int family; 2590 union nethostaddr *haddr; 2591 struct mbuf *nam; 2592 { 2593 struct sockaddr_in *inetaddr; 2594 2595 switch (family) { 2596 case AF_INET: 2597 inetaddr = mtod(nam, struct sockaddr_in *); 2598 if (inetaddr->sin_family == AF_INET && 2599 inetaddr->sin_addr.s_addr == haddr->had_inetaddr) 2600 return (1); 2601 break; 2602 #ifdef INET6 2603 case AF_INET6: 2604 { 2605 struct sockaddr_in6 *sin6_1, *sin6_2; 2606 2607 sin6_1 = mtod(nam, struct sockaddr_in6 *); 2608 sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *); 2609 if (sin6_1->sin6_family == AF_INET6 && 2610 IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr)) 2611 return 1; 2612 } 2613 #endif 2614 #ifdef ISO 2615 case AF_ISO: 2616 { 2617 struct sockaddr_iso *isoaddr1, *isoaddr2; 2618 2619 isoaddr1 = mtod(nam, struct sockaddr_iso *); 2620 isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *); 2621 if (isoaddr1->siso_family == AF_ISO && 2622 isoaddr1->siso_nlen > 0 && 2623 isoaddr1->siso_nlen == isoaddr2->siso_nlen && 2624 SAME_ISOADDR(isoaddr1, isoaddr2)) 2625 return (1); 2626 break; 2627 } 2628 #endif /* ISO */ 2629 default: 2630 break; 2631 }; 2632 return (0); 2633 } 2634 2635 /* 2636 * The write verifier has changed (probably due to a server reboot), so all 2637 * PG_NEEDCOMMIT pages will have to be written again. Since they are marked 2638 * as dirty or are being written out just now, all this takes is clearing 2639 * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for 2640 * the mount point. 2641 */ 2642 void 2643 nfs_clearcommit(mp) 2644 struct mount *mp; 2645 { 2646 struct vnode *vp; 2647 struct nfsnode *np; 2648 struct vm_page *pg; 2649 struct nfsmount *nmp = VFSTONFS(mp); 2650 2651 rw_enter(&nmp->nm_writeverflock, RW_WRITER); 2652 2653 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2654 KASSERT(vp->v_mount == mp); 2655 if (vp->v_type != VREG) 2656 continue; 2657 np = VTONFS(vp); 2658 np->n_pushlo = np->n_pushhi = np->n_pushedlo = 2659 np->n_pushedhi = 0; 2660 np->n_commitflags &= 2661 ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID); 2662 simple_lock(&vp->v_uobj.vmobjlock); 2663 TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) { 2664 pg->flags &= ~PG_NEEDCOMMIT; 2665 } 2666 simple_unlock(&vp->v_uobj.vmobjlock); 2667 } 2668 mutex_enter(&nmp->nm_lock); 2669 nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF; 2670 mutex_exit(&nmp->nm_lock); 2671 rw_exit(&nmp->nm_writeverflock); 2672 } 2673 2674 void 2675 nfs_merge_commit_ranges(vp) 2676 struct vnode *vp; 2677 { 2678 struct nfsnode *np = VTONFS(vp); 2679 2680 KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID); 2681 2682 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2683 np->n_pushedlo = np->n_pushlo; 2684 np->n_pushedhi = np->n_pushhi; 2685 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2686 } else { 2687 if (np->n_pushlo < np->n_pushedlo) 2688 np->n_pushedlo = np->n_pushlo; 2689 if (np->n_pushhi > np->n_pushedhi) 2690 np->n_pushedhi = np->n_pushhi; 2691 } 2692 2693 np->n_pushlo = np->n_pushhi = 0; 2694 np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID; 2695 2696 #ifdef NFS_DEBUG_COMMIT 2697 printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2698 (unsigned)np->n_pushedhi); 2699 #endif 2700 } 2701 2702 int 2703 nfs_in_committed_range(vp, off, len) 2704 struct vnode *vp; 2705 off_t off, len; 2706 { 2707 struct nfsnode *np = VTONFS(vp); 2708 off_t lo, hi; 2709 2710 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2711 return 0; 2712 lo = off; 2713 hi = lo + len; 2714 2715 return (lo >= np->n_pushedlo && hi <= np->n_pushedhi); 2716 } 2717 2718 int 2719 nfs_in_tobecommitted_range(vp, off, len) 2720 struct vnode *vp; 2721 off_t off, len; 2722 { 2723 struct nfsnode *np = VTONFS(vp); 2724 off_t lo, hi; 2725 2726 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2727 return 0; 2728 lo = off; 2729 hi = lo + len; 2730 2731 return (lo >= np->n_pushlo && hi <= np->n_pushhi); 2732 } 2733 2734 void 2735 nfs_add_committed_range(vp, off, len) 2736 struct vnode *vp; 2737 off_t off, len; 2738 { 2739 struct nfsnode *np = VTONFS(vp); 2740 off_t lo, hi; 2741 2742 lo = off; 2743 hi = lo + len; 2744 2745 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2746 np->n_pushedlo = lo; 2747 np->n_pushedhi = hi; 2748 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2749 } else { 2750 if (hi > np->n_pushedhi) 2751 np->n_pushedhi = hi; 2752 if (lo < np->n_pushedlo) 2753 np->n_pushedlo = lo; 2754 } 2755 #ifdef NFS_DEBUG_COMMIT 2756 printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2757 (unsigned)np->n_pushedhi); 2758 #endif 2759 } 2760 2761 void 2762 nfs_del_committed_range(vp, off, len) 2763 struct vnode *vp; 2764 off_t off, len; 2765 { 2766 struct nfsnode *np = VTONFS(vp); 2767 off_t lo, hi; 2768 2769 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2770 return; 2771 2772 lo = off; 2773 hi = lo + len; 2774 2775 if (lo > np->n_pushedhi || hi < np->n_pushedlo) 2776 return; 2777 if (lo <= np->n_pushedlo) 2778 np->n_pushedlo = hi; 2779 else if (hi >= np->n_pushedhi) 2780 np->n_pushedhi = lo; 2781 else { 2782 /* 2783 * XXX There's only one range. If the deleted range 2784 * is in the middle, pick the largest of the 2785 * contiguous ranges that it leaves. 2786 */ 2787 if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi)) 2788 np->n_pushedhi = lo; 2789 else 2790 np->n_pushedlo = hi; 2791 } 2792 #ifdef NFS_DEBUG_COMMIT 2793 printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2794 (unsigned)np->n_pushedhi); 2795 #endif 2796 } 2797 2798 void 2799 nfs_add_tobecommitted_range(vp, off, len) 2800 struct vnode *vp; 2801 off_t off, len; 2802 { 2803 struct nfsnode *np = VTONFS(vp); 2804 off_t lo, hi; 2805 2806 lo = off; 2807 hi = lo + len; 2808 2809 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) { 2810 np->n_pushlo = lo; 2811 np->n_pushhi = hi; 2812 np->n_commitflags |= NFS_COMMIT_PUSH_VALID; 2813 } else { 2814 if (lo < np->n_pushlo) 2815 np->n_pushlo = lo; 2816 if (hi > np->n_pushhi) 2817 np->n_pushhi = hi; 2818 } 2819 #ifdef NFS_DEBUG_COMMIT 2820 printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2821 (unsigned)np->n_pushhi); 2822 #endif 2823 } 2824 2825 void 2826 nfs_del_tobecommitted_range(vp, off, len) 2827 struct vnode *vp; 2828 off_t off, len; 2829 { 2830 struct nfsnode *np = VTONFS(vp); 2831 off_t lo, hi; 2832 2833 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2834 return; 2835 2836 lo = off; 2837 hi = lo + len; 2838 2839 if (lo > np->n_pushhi || hi < np->n_pushlo) 2840 return; 2841 2842 if (lo <= np->n_pushlo) 2843 np->n_pushlo = hi; 2844 else if (hi >= np->n_pushhi) 2845 np->n_pushhi = lo; 2846 else { 2847 /* 2848 * XXX There's only one range. If the deleted range 2849 * is in the middle, pick the largest of the 2850 * contiguous ranges that it leaves. 2851 */ 2852 if ((np->n_pushlo - lo) > (hi - np->n_pushhi)) 2853 np->n_pushhi = lo; 2854 else 2855 np->n_pushlo = hi; 2856 } 2857 #ifdef NFS_DEBUG_COMMIT 2858 printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2859 (unsigned)np->n_pushhi); 2860 #endif 2861 } 2862 2863 /* 2864 * Map errnos to NFS error numbers. For Version 3 also filter out error 2865 * numbers not specified for the associated procedure. 2866 */ 2867 int 2868 nfsrv_errmap(nd, err) 2869 struct nfsrv_descript *nd; 2870 int err; 2871 { 2872 const short *defaulterrp, *errp; 2873 2874 if (nd->nd_flag & ND_NFSV3) { 2875 if (nd->nd_procnum <= NFSPROC_COMMIT) { 2876 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum]; 2877 while (*++errp) { 2878 if (*errp == err) 2879 return (err); 2880 else if (*errp > err) 2881 break; 2882 } 2883 return ((int)*defaulterrp); 2884 } else 2885 return (err & 0xffff); 2886 } 2887 if (err <= ELAST) 2888 return ((int)nfsrv_v2errmap[err - 1]); 2889 return (NFSERR_IO); 2890 } 2891 2892 u_int32_t 2893 nfs_getxid() 2894 { 2895 static u_int32_t base; 2896 static u_int32_t nfs_xid = 0; 2897 static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER; 2898 u_int32_t newxid; 2899 2900 simple_lock(&nfs_xidlock); 2901 /* 2902 * derive initial xid from system time 2903 * XXX time is invalid if root not yet mounted 2904 */ 2905 if (__predict_false(!base && (rootvp))) { 2906 struct timeval tv; 2907 2908 microtime(&tv); 2909 base = tv.tv_sec << 12; 2910 nfs_xid = base; 2911 } 2912 2913 /* 2914 * Skip zero xid if it should ever happen. 2915 */ 2916 if (__predict_false(++nfs_xid == 0)) 2917 nfs_xid++; 2918 newxid = nfs_xid; 2919 simple_unlock(&nfs_xidlock); 2920 2921 return txdr_unsigned(newxid); 2922 } 2923 2924 /* 2925 * assign a new xid for existing request. 2926 * used for NFSERR_JUKEBOX handling. 2927 */ 2928 void 2929 nfs_renewxid(struct nfsreq *req) 2930 { 2931 u_int32_t xid; 2932 int off; 2933 2934 xid = nfs_getxid(); 2935 if (req->r_nmp->nm_sotype == SOCK_STREAM) 2936 off = sizeof(u_int32_t); /* RPC record mark */ 2937 else 2938 off = 0; 2939 2940 m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid); 2941 req->r_xid = xid; 2942 } 2943 2944 #if defined(NFSSERVER) 2945 int 2946 nfsrv_composefh(struct vnode *vp, nfsrvfh_t *nsfh, bool v3) 2947 { 2948 int error; 2949 size_t fhsize; 2950 2951 fhsize = NFSD_MAXFHSIZE; 2952 error = vfs_composefh(vp, (void *)NFSRVFH_DATA(nsfh), &fhsize); 2953 if (NFSX_FHTOOBIG_P(fhsize, v3)) { 2954 error = EOPNOTSUPP; 2955 } 2956 if (error != 0) { 2957 return error; 2958 } 2959 if (!v3 && fhsize < NFSX_V2FH) { 2960 memset((char *)NFSRVFH_DATA(nsfh) + fhsize, 0, 2961 NFSX_V2FH - fhsize); 2962 fhsize = NFSX_V2FH; 2963 } 2964 if ((fhsize % NFSX_UNSIGNED) != 0) { 2965 return EOPNOTSUPP; 2966 } 2967 nsfh->nsfh_size = fhsize; 2968 return 0; 2969 } 2970 2971 int 2972 nfsrv_comparefh(const nfsrvfh_t *fh1, const nfsrvfh_t *fh2) 2973 { 2974 2975 if (NFSRVFH_SIZE(fh1) != NFSRVFH_SIZE(fh2)) { 2976 return NFSRVFH_SIZE(fh2) - NFSRVFH_SIZE(fh1); 2977 } 2978 return memcmp(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), NFSRVFH_SIZE(fh1)); 2979 } 2980 2981 void 2982 nfsrv_copyfh(nfsrvfh_t *fh1, const nfsrvfh_t *fh2) 2983 { 2984 size_t size; 2985 2986 fh1->nsfh_size = size = NFSRVFH_SIZE(fh2); 2987 memcpy(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), size); 2988 } 2989 #endif /* defined(NFSSERVER) */ 2990
Indexes created Thu Sep 25 16:09:42 GMT 2025