nfs_subs.c revision 1.149.2.1
1 /* $NetBSD: nfs_subs.c,v 1.149.2.1 2005/07/07 11:53:25 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 1.149.2.1 2005/07/07 11:53:25 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> 100 101 #include <uvm/uvm_extern.h> 102 103 #include <nfs/rpcv2.h> 104 #include <nfs/nfsproto.h> 105 #include <nfs/nfsnode.h> 106 #include <nfs/nfs.h> 107 #include <nfs/xdr_subs.h> 108 #include <nfs/nfsm_subs.h> 109 #include <nfs/nfsmount.h> 110 #include <nfs/nqnfs.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 #ifdef ISO 118 #include <netiso/iso.h> 119 #endif 120 121 /* 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, nqnfs_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 int nfs_commitsize; 142 143 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies"); 144 145 /* NFS client/server stats. */ 146 struct nfsstats nfsstats; 147 148 /* 149 * Mapping of old NFS Version 2 RPC numbers to generic numbers. 150 */ 151 const int nfsv3_procid[NFS_NPROCS] = { 152 NFSPROC_NULL, 153 NFSPROC_GETATTR, 154 NFSPROC_SETATTR, 155 NFSPROC_NOOP, 156 NFSPROC_LOOKUP, 157 NFSPROC_READLINK, 158 NFSPROC_READ, 159 NFSPROC_NOOP, 160 NFSPROC_WRITE, 161 NFSPROC_CREATE, 162 NFSPROC_REMOVE, 163 NFSPROC_RENAME, 164 NFSPROC_LINK, 165 NFSPROC_SYMLINK, 166 NFSPROC_MKDIR, 167 NFSPROC_RMDIR, 168 NFSPROC_READDIR, 169 NFSPROC_FSSTAT, 170 NFSPROC_NOOP, 171 NFSPROC_NOOP, 172 NFSPROC_NOOP, 173 NFSPROC_NOOP, 174 NFSPROC_NOOP, 175 NFSPROC_NOOP, 176 NFSPROC_NOOP, 177 NFSPROC_NOOP 178 }; 179 180 /* 181 * and the reverse mapping from generic to Version 2 procedure numbers 182 */ 183 const int nfsv2_procid[NFS_NPROCS] = { 184 NFSV2PROC_NULL, 185 NFSV2PROC_GETATTR, 186 NFSV2PROC_SETATTR, 187 NFSV2PROC_LOOKUP, 188 NFSV2PROC_NOOP, 189 NFSV2PROC_READLINK, 190 NFSV2PROC_READ, 191 NFSV2PROC_WRITE, 192 NFSV2PROC_CREATE, 193 NFSV2PROC_MKDIR, 194 NFSV2PROC_SYMLINK, 195 NFSV2PROC_CREATE, 196 NFSV2PROC_REMOVE, 197 NFSV2PROC_RMDIR, 198 NFSV2PROC_RENAME, 199 NFSV2PROC_LINK, 200 NFSV2PROC_READDIR, 201 NFSV2PROC_NOOP, 202 NFSV2PROC_STATFS, 203 NFSV2PROC_NOOP, 204 NFSV2PROC_NOOP, 205 NFSV2PROC_NOOP, 206 NFSV2PROC_NOOP, 207 NFSV2PROC_NOOP, 208 NFSV2PROC_NOOP, 209 NFSV2PROC_NOOP, 210 }; 211 212 /* 213 * Maps errno values to nfs error numbers. 214 * Use NFSERR_IO as the catch all for ones not specifically defined in 215 * RFC 1094. 216 */ 217 static const u_char nfsrv_v2errmap[ELAST] = { 218 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 219 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 220 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 221 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 222 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 223 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 224 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 225 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 226 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 227 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 228 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 229 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 230 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO, 231 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 232 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 233 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 234 NFSERR_IO, NFSERR_IO, 235 }; 236 237 /* 238 * Maps errno values to nfs error numbers. 239 * Although it is not obvious whether or not NFS clients really care if 240 * a returned error value is in the specified list for the procedure, the 241 * safest thing to do is filter them appropriately. For Version 2, the 242 * X/Open XNFS document is the only specification that defines error values 243 * for each RPC (The RFC simply lists all possible error values for all RPCs), 244 * so I have decided to not do this for Version 2. 245 * The first entry is the default error return and the rest are the valid 246 * errors for that RPC in increasing numeric order. 247 */ 248 static const short nfsv3err_null[] = { 249 0, 250 0, 251 }; 252 253 static const short nfsv3err_getattr[] = { 254 NFSERR_IO, 255 NFSERR_IO, 256 NFSERR_STALE, 257 NFSERR_BADHANDLE, 258 NFSERR_SERVERFAULT, 259 0, 260 }; 261 262 static const short nfsv3err_setattr[] = { 263 NFSERR_IO, 264 NFSERR_PERM, 265 NFSERR_IO, 266 NFSERR_ACCES, 267 NFSERR_INVAL, 268 NFSERR_NOSPC, 269 NFSERR_ROFS, 270 NFSERR_DQUOT, 271 NFSERR_STALE, 272 NFSERR_BADHANDLE, 273 NFSERR_NOT_SYNC, 274 NFSERR_SERVERFAULT, 275 0, 276 }; 277 278 static const short nfsv3err_lookup[] = { 279 NFSERR_IO, 280 NFSERR_NOENT, 281 NFSERR_IO, 282 NFSERR_ACCES, 283 NFSERR_NOTDIR, 284 NFSERR_NAMETOL, 285 NFSERR_STALE, 286 NFSERR_BADHANDLE, 287 NFSERR_SERVERFAULT, 288 0, 289 }; 290 291 static const short nfsv3err_access[] = { 292 NFSERR_IO, 293 NFSERR_IO, 294 NFSERR_STALE, 295 NFSERR_BADHANDLE, 296 NFSERR_SERVERFAULT, 297 0, 298 }; 299 300 static const short nfsv3err_readlink[] = { 301 NFSERR_IO, 302 NFSERR_IO, 303 NFSERR_ACCES, 304 NFSERR_INVAL, 305 NFSERR_STALE, 306 NFSERR_BADHANDLE, 307 NFSERR_NOTSUPP, 308 NFSERR_SERVERFAULT, 309 0, 310 }; 311 312 static const short nfsv3err_read[] = { 313 NFSERR_IO, 314 NFSERR_IO, 315 NFSERR_NXIO, 316 NFSERR_ACCES, 317 NFSERR_INVAL, 318 NFSERR_STALE, 319 NFSERR_BADHANDLE, 320 NFSERR_SERVERFAULT, 321 NFSERR_JUKEBOX, 322 0, 323 }; 324 325 static const short nfsv3err_write[] = { 326 NFSERR_IO, 327 NFSERR_IO, 328 NFSERR_ACCES, 329 NFSERR_INVAL, 330 NFSERR_FBIG, 331 NFSERR_NOSPC, 332 NFSERR_ROFS, 333 NFSERR_DQUOT, 334 NFSERR_STALE, 335 NFSERR_BADHANDLE, 336 NFSERR_SERVERFAULT, 337 NFSERR_JUKEBOX, 338 0, 339 }; 340 341 static const short nfsv3err_create[] = { 342 NFSERR_IO, 343 NFSERR_IO, 344 NFSERR_ACCES, 345 NFSERR_EXIST, 346 NFSERR_NOTDIR, 347 NFSERR_NOSPC, 348 NFSERR_ROFS, 349 NFSERR_NAMETOL, 350 NFSERR_DQUOT, 351 NFSERR_STALE, 352 NFSERR_BADHANDLE, 353 NFSERR_NOTSUPP, 354 NFSERR_SERVERFAULT, 355 0, 356 }; 357 358 static const short nfsv3err_mkdir[] = { 359 NFSERR_IO, 360 NFSERR_IO, 361 NFSERR_ACCES, 362 NFSERR_EXIST, 363 NFSERR_NOTDIR, 364 NFSERR_NOSPC, 365 NFSERR_ROFS, 366 NFSERR_NAMETOL, 367 NFSERR_DQUOT, 368 NFSERR_STALE, 369 NFSERR_BADHANDLE, 370 NFSERR_NOTSUPP, 371 NFSERR_SERVERFAULT, 372 0, 373 }; 374 375 static const short nfsv3err_symlink[] = { 376 NFSERR_IO, 377 NFSERR_IO, 378 NFSERR_ACCES, 379 NFSERR_EXIST, 380 NFSERR_NOTDIR, 381 NFSERR_NOSPC, 382 NFSERR_ROFS, 383 NFSERR_NAMETOL, 384 NFSERR_DQUOT, 385 NFSERR_STALE, 386 NFSERR_BADHANDLE, 387 NFSERR_NOTSUPP, 388 NFSERR_SERVERFAULT, 389 0, 390 }; 391 392 static const short nfsv3err_mknod[] = { 393 NFSERR_IO, 394 NFSERR_IO, 395 NFSERR_ACCES, 396 NFSERR_EXIST, 397 NFSERR_NOTDIR, 398 NFSERR_NOSPC, 399 NFSERR_ROFS, 400 NFSERR_NAMETOL, 401 NFSERR_DQUOT, 402 NFSERR_STALE, 403 NFSERR_BADHANDLE, 404 NFSERR_NOTSUPP, 405 NFSERR_SERVERFAULT, 406 NFSERR_BADTYPE, 407 0, 408 }; 409 410 static const short nfsv3err_remove[] = { 411 NFSERR_IO, 412 NFSERR_NOENT, 413 NFSERR_IO, 414 NFSERR_ACCES, 415 NFSERR_NOTDIR, 416 NFSERR_ROFS, 417 NFSERR_NAMETOL, 418 NFSERR_STALE, 419 NFSERR_BADHANDLE, 420 NFSERR_SERVERFAULT, 421 0, 422 }; 423 424 static const short nfsv3err_rmdir[] = { 425 NFSERR_IO, 426 NFSERR_NOENT, 427 NFSERR_IO, 428 NFSERR_ACCES, 429 NFSERR_EXIST, 430 NFSERR_NOTDIR, 431 NFSERR_INVAL, 432 NFSERR_ROFS, 433 NFSERR_NAMETOL, 434 NFSERR_NOTEMPTY, 435 NFSERR_STALE, 436 NFSERR_BADHANDLE, 437 NFSERR_NOTSUPP, 438 NFSERR_SERVERFAULT, 439 0, 440 }; 441 442 static const short nfsv3err_rename[] = { 443 NFSERR_IO, 444 NFSERR_NOENT, 445 NFSERR_IO, 446 NFSERR_ACCES, 447 NFSERR_EXIST, 448 NFSERR_XDEV, 449 NFSERR_NOTDIR, 450 NFSERR_ISDIR, 451 NFSERR_INVAL, 452 NFSERR_NOSPC, 453 NFSERR_ROFS, 454 NFSERR_MLINK, 455 NFSERR_NAMETOL, 456 NFSERR_NOTEMPTY, 457 NFSERR_DQUOT, 458 NFSERR_STALE, 459 NFSERR_BADHANDLE, 460 NFSERR_NOTSUPP, 461 NFSERR_SERVERFAULT, 462 0, 463 }; 464 465 static const short nfsv3err_link[] = { 466 NFSERR_IO, 467 NFSERR_IO, 468 NFSERR_ACCES, 469 NFSERR_EXIST, 470 NFSERR_XDEV, 471 NFSERR_NOTDIR, 472 NFSERR_INVAL, 473 NFSERR_NOSPC, 474 NFSERR_ROFS, 475 NFSERR_MLINK, 476 NFSERR_NAMETOL, 477 NFSERR_DQUOT, 478 NFSERR_STALE, 479 NFSERR_BADHANDLE, 480 NFSERR_NOTSUPP, 481 NFSERR_SERVERFAULT, 482 0, 483 }; 484 485 static const short nfsv3err_readdir[] = { 486 NFSERR_IO, 487 NFSERR_IO, 488 NFSERR_ACCES, 489 NFSERR_NOTDIR, 490 NFSERR_STALE, 491 NFSERR_BADHANDLE, 492 NFSERR_BAD_COOKIE, 493 NFSERR_TOOSMALL, 494 NFSERR_SERVERFAULT, 495 0, 496 }; 497 498 static const short nfsv3err_readdirplus[] = { 499 NFSERR_IO, 500 NFSERR_IO, 501 NFSERR_ACCES, 502 NFSERR_NOTDIR, 503 NFSERR_STALE, 504 NFSERR_BADHANDLE, 505 NFSERR_BAD_COOKIE, 506 NFSERR_NOTSUPP, 507 NFSERR_TOOSMALL, 508 NFSERR_SERVERFAULT, 509 0, 510 }; 511 512 static const short nfsv3err_fsstat[] = { 513 NFSERR_IO, 514 NFSERR_IO, 515 NFSERR_STALE, 516 NFSERR_BADHANDLE, 517 NFSERR_SERVERFAULT, 518 0, 519 }; 520 521 static const short nfsv3err_fsinfo[] = { 522 NFSERR_STALE, 523 NFSERR_STALE, 524 NFSERR_BADHANDLE, 525 NFSERR_SERVERFAULT, 526 0, 527 }; 528 529 static const short nfsv3err_pathconf[] = { 530 NFSERR_STALE, 531 NFSERR_STALE, 532 NFSERR_BADHANDLE, 533 NFSERR_SERVERFAULT, 534 0, 535 }; 536 537 static const short nfsv3err_commit[] = { 538 NFSERR_IO, 539 NFSERR_IO, 540 NFSERR_STALE, 541 NFSERR_BADHANDLE, 542 NFSERR_SERVERFAULT, 543 0, 544 }; 545 546 static const short * const nfsrv_v3errmap[] = { 547 nfsv3err_null, 548 nfsv3err_getattr, 549 nfsv3err_setattr, 550 nfsv3err_lookup, 551 nfsv3err_access, 552 nfsv3err_readlink, 553 nfsv3err_read, 554 nfsv3err_write, 555 nfsv3err_create, 556 nfsv3err_mkdir, 557 nfsv3err_symlink, 558 nfsv3err_mknod, 559 nfsv3err_remove, 560 nfsv3err_rmdir, 561 nfsv3err_rename, 562 nfsv3err_link, 563 nfsv3err_readdir, 564 nfsv3err_readdirplus, 565 nfsv3err_fsstat, 566 nfsv3err_fsinfo, 567 nfsv3err_pathconf, 568 nfsv3err_commit, 569 }; 570 571 extern struct nfsrtt nfsrtt; 572 extern time_t nqnfsstarttime; 573 extern int nqsrv_clockskew; 574 extern int nqsrv_writeslack; 575 extern int nqsrv_maxlease; 576 extern const int nqnfs_piggy[NFS_NPROCS]; 577 extern struct nfsnodehashhead *nfsnodehashtbl; 578 extern u_long nfsnodehash; 579 580 u_long nfsdirhashmask; 581 582 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *)); 583 584 /* 585 * Create the header for an rpc request packet 586 * The hsiz is the size of the rest of the nfs request header. 587 * (just used to decide if a cluster is a good idea) 588 */ 589 struct mbuf * 590 nfsm_reqh(np, procid, hsiz, bposp) 591 struct nfsnode *np; 592 u_long procid; 593 int hsiz; 594 caddr_t *bposp; 595 { 596 struct mbuf *mb; 597 caddr_t bpos; 598 #ifndef NFS_V2_ONLY 599 struct nfsmount *nmp; 600 u_int32_t *tl; 601 int nqflag; 602 #endif 603 604 mb = m_get(M_WAIT, MT_DATA); 605 MCLAIM(mb, &nfs_mowner); 606 if (hsiz >= MINCLSIZE) 607 m_clget(mb, M_WAIT); 608 mb->m_len = 0; 609 bpos = mtod(mb, caddr_t); 610 611 #ifndef NFS_V2_ONLY 612 /* 613 * For NQNFS, add lease request. 614 */ 615 if (np) { 616 nmp = VFSTONFS(np->n_vnode->v_mount); 617 if (nmp->nm_flag & NFSMNT_NQNFS) { 618 nqflag = NQNFS_NEEDLEASE(np, procid); 619 if (nqflag) { 620 nfsm_build(tl, u_int32_t *, 2*NFSX_UNSIGNED); 621 *tl++ = txdr_unsigned(nqflag); 622 *tl = txdr_unsigned(nmp->nm_leaseterm); 623 } else { 624 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 625 *tl = 0; 626 } 627 } 628 } 629 #endif 630 /* Finally, return values */ 631 *bposp = bpos; 632 return (mb); 633 } 634 635 /* 636 * Build the RPC header and fill in the authorization info. 637 * The authorization string argument is only used when the credentials 638 * come from outside of the kernel. 639 * Returns the head of the mbuf list. 640 */ 641 struct mbuf * 642 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len, 643 verf_str, mrest, mrest_len, mbp, xidp) 644 struct ucred *cr; 645 int nmflag; 646 int procid; 647 int auth_type; 648 int auth_len; 649 char *auth_str; 650 int verf_len; 651 char *verf_str; 652 struct mbuf *mrest; 653 int mrest_len; 654 struct mbuf **mbp; 655 u_int32_t *xidp; 656 { 657 struct mbuf *mb; 658 u_int32_t *tl; 659 caddr_t bpos; 660 int i; 661 struct mbuf *mreq; 662 int siz, grpsiz, authsiz; 663 664 authsiz = nfsm_rndup(auth_len); 665 mb = m_gethdr(M_WAIT, MT_DATA); 666 MCLAIM(mb, &nfs_mowner); 667 if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) { 668 m_clget(mb, M_WAIT); 669 } else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) { 670 MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED); 671 } else { 672 MH_ALIGN(mb, 8 * NFSX_UNSIGNED); 673 } 674 mb->m_len = 0; 675 mreq = mb; 676 bpos = mtod(mb, caddr_t); 677 678 /* 679 * First the RPC header. 680 */ 681 nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 682 683 *tl++ = *xidp = nfs_getxid(); 684 *tl++ = rpc_call; 685 *tl++ = rpc_vers; 686 if (nmflag & NFSMNT_NQNFS) { 687 *tl++ = txdr_unsigned(NQNFS_PROG); 688 *tl++ = txdr_unsigned(NQNFS_VER3); 689 } else { 690 *tl++ = txdr_unsigned(NFS_PROG); 691 if (nmflag & NFSMNT_NFSV3) 692 *tl++ = txdr_unsigned(NFS_VER3); 693 else 694 *tl++ = txdr_unsigned(NFS_VER2); 695 } 696 if (nmflag & NFSMNT_NFSV3) 697 *tl++ = txdr_unsigned(procid); 698 else 699 *tl++ = txdr_unsigned(nfsv2_procid[procid]); 700 701 /* 702 * And then the authorization cred. 703 */ 704 *tl++ = txdr_unsigned(auth_type); 705 *tl = txdr_unsigned(authsiz); 706 switch (auth_type) { 707 case RPCAUTH_UNIX: 708 nfsm_build(tl, u_int32_t *, auth_len); 709 *tl++ = 0; /* stamp ?? */ 710 *tl++ = 0; /* NULL hostname */ 711 *tl++ = txdr_unsigned(cr->cr_uid); 712 *tl++ = txdr_unsigned(cr->cr_gid); 713 grpsiz = (auth_len >> 2) - 5; 714 *tl++ = txdr_unsigned(grpsiz); 715 for (i = 0; i < grpsiz; i++) 716 *tl++ = txdr_unsigned(cr->cr_groups[i]); 717 break; 718 case RPCAUTH_KERB4: 719 siz = auth_len; 720 while (siz > 0) { 721 if (M_TRAILINGSPACE(mb) == 0) { 722 struct mbuf *mb2; 723 mb2 = m_get(M_WAIT, MT_DATA); 724 MCLAIM(mb2, &nfs_mowner); 725 if (siz >= MINCLSIZE) 726 m_clget(mb2, M_WAIT); 727 mb->m_next = mb2; 728 mb = mb2; 729 mb->m_len = 0; 730 bpos = mtod(mb, caddr_t); 731 } 732 i = min(siz, M_TRAILINGSPACE(mb)); 733 memcpy(bpos, auth_str, i); 734 mb->m_len += i; 735 auth_str += i; 736 bpos += i; 737 siz -= i; 738 } 739 if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) { 740 for (i = 0; i < siz; i++) 741 *bpos++ = '\0'; 742 mb->m_len += siz; 743 } 744 break; 745 }; 746 747 /* 748 * And the verifier... 749 */ 750 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 751 if (verf_str) { 752 *tl++ = txdr_unsigned(RPCAUTH_KERB4); 753 *tl = txdr_unsigned(verf_len); 754 siz = verf_len; 755 while (siz > 0) { 756 if (M_TRAILINGSPACE(mb) == 0) { 757 struct mbuf *mb2; 758 mb2 = m_get(M_WAIT, MT_DATA); 759 MCLAIM(mb2, &nfs_mowner); 760 if (siz >= MINCLSIZE) 761 m_clget(mb2, M_WAIT); 762 mb->m_next = mb2; 763 mb = mb2; 764 mb->m_len = 0; 765 bpos = mtod(mb, caddr_t); 766 } 767 i = min(siz, M_TRAILINGSPACE(mb)); 768 memcpy(bpos, verf_str, i); 769 mb->m_len += i; 770 verf_str += i; 771 bpos += i; 772 siz -= i; 773 } 774 if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) { 775 for (i = 0; i < siz; i++) 776 *bpos++ = '\0'; 777 mb->m_len += siz; 778 } 779 } else { 780 *tl++ = txdr_unsigned(RPCAUTH_NULL); 781 *tl = 0; 782 } 783 mb->m_next = mrest; 784 mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len; 785 mreq->m_pkthdr.rcvif = (struct ifnet *)0; 786 *mbp = mb; 787 return (mreq); 788 } 789 790 /* 791 * copies mbuf chain to the uio scatter/gather list 792 */ 793 int 794 nfsm_mbuftouio(mrep, uiop, siz, dpos) 795 struct mbuf **mrep; 796 struct uio *uiop; 797 int siz; 798 caddr_t *dpos; 799 { 800 char *mbufcp, *uiocp; 801 int xfer, left, len; 802 struct mbuf *mp; 803 long uiosiz, rem; 804 int error = 0; 805 806 mp = *mrep; 807 mbufcp = *dpos; 808 len = mtod(mp, caddr_t)+mp->m_len-mbufcp; 809 rem = nfsm_rndup(siz)-siz; 810 while (siz > 0) { 811 if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) 812 return (EFBIG); 813 left = uiop->uio_iov->iov_len; 814 uiocp = uiop->uio_iov->iov_base; 815 if (left > siz) 816 left = siz; 817 uiosiz = left; 818 while (left > 0) { 819 while (len == 0) { 820 mp = mp->m_next; 821 if (mp == NULL) 822 return (EBADRPC); 823 mbufcp = mtod(mp, caddr_t); 824 len = mp->m_len; 825 } 826 xfer = (left > len) ? len : left; 827 #ifdef notdef 828 /* Not Yet.. */ 829 if (uiop->uio_iov->iov_op != NULL) 830 (*(uiop->uio_iov->iov_op)) 831 (mbufcp, uiocp, xfer); 832 else 833 #endif 834 if (uiop->uio_segflg == UIO_SYSSPACE) 835 memcpy(uiocp, mbufcp, xfer); 836 else 837 copyout(mbufcp, uiocp, xfer); 838 left -= xfer; 839 len -= xfer; 840 mbufcp += xfer; 841 uiocp += xfer; 842 uiop->uio_offset += xfer; 843 uiop->uio_resid -= xfer; 844 } 845 if (uiop->uio_iov->iov_len <= siz) { 846 uiop->uio_iovcnt--; 847 uiop->uio_iov++; 848 } else { 849 uiop->uio_iov->iov_base = 850 (caddr_t)uiop->uio_iov->iov_base + uiosiz; 851 uiop->uio_iov->iov_len -= uiosiz; 852 } 853 siz -= uiosiz; 854 } 855 *dpos = mbufcp; 856 *mrep = mp; 857 if (rem > 0) { 858 if (len < rem) 859 error = nfs_adv(mrep, dpos, rem, len); 860 else 861 *dpos += rem; 862 } 863 return (error); 864 } 865 866 /* 867 * copies a uio scatter/gather list to an mbuf chain. 868 * NOTE: can ony handle iovcnt == 1 869 */ 870 int 871 nfsm_uiotombuf(uiop, mq, siz, bpos) 872 struct uio *uiop; 873 struct mbuf **mq; 874 int siz; 875 caddr_t *bpos; 876 { 877 char *uiocp; 878 struct mbuf *mp, *mp2; 879 int xfer, left, mlen; 880 int uiosiz, clflg, rem; 881 char *cp; 882 883 #ifdef DIAGNOSTIC 884 if (uiop->uio_iovcnt != 1) 885 panic("nfsm_uiotombuf: iovcnt != 1"); 886 #endif 887 888 if (siz > MLEN) /* or should it >= MCLBYTES ?? */ 889 clflg = 1; 890 else 891 clflg = 0; 892 rem = nfsm_rndup(siz)-siz; 893 mp = mp2 = *mq; 894 while (siz > 0) { 895 left = uiop->uio_iov->iov_len; 896 uiocp = uiop->uio_iov->iov_base; 897 if (left > siz) 898 left = siz; 899 uiosiz = left; 900 while (left > 0) { 901 mlen = M_TRAILINGSPACE(mp); 902 if (mlen == 0) { 903 mp = m_get(M_WAIT, MT_DATA); 904 MCLAIM(mp, &nfs_mowner); 905 if (clflg) 906 m_clget(mp, M_WAIT); 907 mp->m_len = 0; 908 mp2->m_next = mp; 909 mp2 = mp; 910 mlen = M_TRAILINGSPACE(mp); 911 } 912 xfer = (left > mlen) ? mlen : left; 913 #ifdef notdef 914 /* Not Yet.. */ 915 if (uiop->uio_iov->iov_op != NULL) 916 (*(uiop->uio_iov->iov_op)) 917 (uiocp, mtod(mp, caddr_t)+mp->m_len, xfer); 918 else 919 #endif 920 if (uiop->uio_segflg == UIO_SYSSPACE) 921 memcpy(mtod(mp, caddr_t)+mp->m_len, uiocp, xfer); 922 else 923 copyin(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer); 924 mp->m_len += xfer; 925 left -= xfer; 926 uiocp += xfer; 927 uiop->uio_offset += xfer; 928 uiop->uio_resid -= xfer; 929 } 930 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base + 931 uiosiz; 932 uiop->uio_iov->iov_len -= uiosiz; 933 siz -= uiosiz; 934 } 935 if (rem > 0) { 936 if (rem > M_TRAILINGSPACE(mp)) { 937 mp = m_get(M_WAIT, MT_DATA); 938 MCLAIM(mp, &nfs_mowner); 939 mp->m_len = 0; 940 mp2->m_next = mp; 941 } 942 cp = mtod(mp, caddr_t)+mp->m_len; 943 for (left = 0; left < rem; left++) 944 *cp++ = '\0'; 945 mp->m_len += rem; 946 *bpos = cp; 947 } else 948 *bpos = mtod(mp, caddr_t)+mp->m_len; 949 *mq = mp; 950 return (0); 951 } 952 953 /* 954 * Get at least "siz" bytes of correctly aligned data. 955 * When called the mbuf pointers are not necessarily correct, 956 * dsosp points to what ought to be in m_data and left contains 957 * what ought to be in m_len. 958 * This is used by the macros nfsm_dissect and nfsm_dissecton for tough 959 * cases. (The macros use the vars. dpos and dpos2) 960 */ 961 int 962 nfsm_disct(mdp, dposp, siz, left, cp2) 963 struct mbuf **mdp; 964 caddr_t *dposp; 965 int siz; 966 int left; 967 caddr_t *cp2; 968 { 969 struct mbuf *m1, *m2; 970 struct mbuf *havebuf = NULL; 971 caddr_t src = *dposp; 972 caddr_t dst; 973 int len; 974 975 #ifdef DEBUG 976 if (left < 0) 977 panic("nfsm_disct: left < 0"); 978 #endif 979 m1 = *mdp; 980 /* 981 * Skip through the mbuf chain looking for an mbuf with 982 * some data. If the first mbuf found has enough data 983 * and it is correctly aligned return it. 984 */ 985 while (left == 0) { 986 havebuf = m1; 987 *mdp = m1 = m1->m_next; 988 if (m1 == NULL) 989 return (EBADRPC); 990 src = mtod(m1, caddr_t); 991 left = m1->m_len; 992 /* 993 * If we start a new mbuf and it is big enough 994 * and correctly aligned just return it, don't 995 * do any pull up. 996 */ 997 if (left >= siz && nfsm_aligned(src)) { 998 *cp2 = src; 999 *dposp = src + siz; 1000 return (0); 1001 } 1002 } 1003 if ((m1->m_flags & M_EXT) != 0) { 1004 if (havebuf && M_TRAILINGSPACE(havebuf) >= siz && 1005 nfsm_aligned(mtod(havebuf, char *) + havebuf->m_len)) { 1006 /* 1007 * If the first mbuf with data has external data 1008 * and there is a previous mbuf with some trailing 1009 * space, use it to move the data into. 1010 */ 1011 m2 = m1; 1012 *mdp = m1 = havebuf; 1013 *cp2 = mtod(m1, char *) + m1->m_len; 1014 } else if (havebuf) { 1015 /* 1016 * If the first mbuf has a external data 1017 * and there is no previous empty mbuf 1018 * allocate a new mbuf and move the external 1019 * data to the new mbuf. Also make the first 1020 * mbuf look empty. 1021 */ 1022 m2 = m1; 1023 *mdp = m1 = m_get(M_WAIT, MT_DATA); 1024 MCLAIM(m1, m2->m_owner); 1025 if ((m2->m_flags & M_PKTHDR) != 0) { 1026 /* XXX MOVE */ 1027 M_COPY_PKTHDR(m1, m2); 1028 m_tag_delete_chain(m2, NULL); 1029 m2->m_flags &= ~M_PKTHDR; 1030 } 1031 if (havebuf) { 1032 havebuf->m_next = m1; 1033 } 1034 m1->m_next = m2; 1035 MRESETDATA(m1); 1036 m1->m_len = 0; 1037 m2->m_data = src; 1038 m2->m_len = left; 1039 *cp2 = mtod(m1, char *); 1040 } else { 1041 struct mbuf **nextp = &m1->m_next; 1042 1043 m1->m_len -= left; 1044 do { 1045 m2 = m_get(M_WAIT, MT_DATA); 1046 MCLAIM(m2, m1->m_owner); 1047 if (left >= MINCLSIZE) { 1048 MCLGET(m2, M_WAIT); 1049 } 1050 m2->m_next = *nextp; 1051 *nextp = m2; 1052 nextp = &m2->m_next; 1053 len = (m2->m_flags & M_EXT) != 0 ? 1054 MCLBYTES : MLEN; 1055 if (len > left) { 1056 len = left; 1057 } 1058 memcpy(mtod(m2, char *), src, len); 1059 m2->m_len = len; 1060 src += len; 1061 left -= len; 1062 } while (left > 0); 1063 *mdp = m1 = m1->m_next; 1064 m2 = m1->m_next; 1065 *cp2 = mtod(m1, char *); 1066 } 1067 } else { 1068 /* 1069 * If the first mbuf has no external data 1070 * move the data to the front of the mbuf. 1071 */ 1072 MRESETDATA(m1); 1073 dst = mtod(m1, char *); 1074 if (dst != src) { 1075 memmove(dst, src, left); 1076 } 1077 m1->m_len = left; 1078 m2 = m1->m_next; 1079 *cp2 = m1->m_data; 1080 } 1081 *dposp = *cp2 + siz; 1082 /* 1083 * Loop through mbufs pulling data up into first mbuf until 1084 * the first mbuf is full or there is no more data to 1085 * pullup. 1086 */ 1087 dst = mtod(m1, char *) + m1->m_len; 1088 while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) { 1089 if ((len = min(len, m2->m_len)) != 0) { 1090 memcpy(dst, mtod(m2, char *), len); 1091 } 1092 m1->m_len += len; 1093 dst += len; 1094 m2->m_data += len; 1095 m2->m_len -= len; 1096 m2 = m2->m_next; 1097 } 1098 if (m1->m_len < siz) 1099 return (EBADRPC); 1100 return (0); 1101 } 1102 1103 /* 1104 * Advance the position in the mbuf chain. 1105 */ 1106 int 1107 nfs_adv(mdp, dposp, offs, left) 1108 struct mbuf **mdp; 1109 caddr_t *dposp; 1110 int offs; 1111 int left; 1112 { 1113 struct mbuf *m; 1114 int s; 1115 1116 m = *mdp; 1117 s = left; 1118 while (s < offs) { 1119 offs -= s; 1120 m = m->m_next; 1121 if (m == NULL) 1122 return (EBADRPC); 1123 s = m->m_len; 1124 } 1125 *mdp = m; 1126 *dposp = mtod(m, caddr_t)+offs; 1127 return (0); 1128 } 1129 1130 /* 1131 * Copy a string into mbufs for the hard cases... 1132 */ 1133 int 1134 nfsm_strtmbuf(mb, bpos, cp, siz) 1135 struct mbuf **mb; 1136 char **bpos; 1137 const char *cp; 1138 long siz; 1139 { 1140 struct mbuf *m1 = NULL, *m2; 1141 long left, xfer, len, tlen; 1142 u_int32_t *tl; 1143 int putsize; 1144 1145 putsize = 1; 1146 m2 = *mb; 1147 left = M_TRAILINGSPACE(m2); 1148 if (left > 0) { 1149 tl = ((u_int32_t *)(*bpos)); 1150 *tl++ = txdr_unsigned(siz); 1151 putsize = 0; 1152 left -= NFSX_UNSIGNED; 1153 m2->m_len += NFSX_UNSIGNED; 1154 if (left > 0) { 1155 memcpy((caddr_t) tl, cp, left); 1156 siz -= left; 1157 cp += left; 1158 m2->m_len += left; 1159 left = 0; 1160 } 1161 } 1162 /* Loop around adding mbufs */ 1163 while (siz > 0) { 1164 m1 = m_get(M_WAIT, MT_DATA); 1165 MCLAIM(m1, &nfs_mowner); 1166 if (siz > MLEN) 1167 m_clget(m1, M_WAIT); 1168 m1->m_len = NFSMSIZ(m1); 1169 m2->m_next = m1; 1170 m2 = m1; 1171 tl = mtod(m1, u_int32_t *); 1172 tlen = 0; 1173 if (putsize) { 1174 *tl++ = txdr_unsigned(siz); 1175 m1->m_len -= NFSX_UNSIGNED; 1176 tlen = NFSX_UNSIGNED; 1177 putsize = 0; 1178 } 1179 if (siz < m1->m_len) { 1180 len = nfsm_rndup(siz); 1181 xfer = siz; 1182 if (xfer < len) 1183 *(tl+(xfer>>2)) = 0; 1184 } else { 1185 xfer = len = m1->m_len; 1186 } 1187 memcpy((caddr_t) tl, cp, xfer); 1188 m1->m_len = len+tlen; 1189 siz -= xfer; 1190 cp += xfer; 1191 } 1192 *mb = m1; 1193 *bpos = mtod(m1, caddr_t)+m1->m_len; 1194 return (0); 1195 } 1196 1197 /* 1198 * Directory caching routines. They work as follows: 1199 * - a cache is maintained per VDIR nfsnode. 1200 * - for each offset cookie that is exported to userspace, and can 1201 * thus be thrown back at us as an offset to VOP_READDIR, store 1202 * information in the cache. 1203 * - cached are: 1204 * - cookie itself 1205 * - blocknumber (essentially just a search key in the buffer cache) 1206 * - entry number in block. 1207 * - offset cookie of block in which this entry is stored 1208 * - 32 bit cookie if NFSMNT_XLATECOOKIE is used. 1209 * - entries are looked up in a hash table 1210 * - also maintained is an LRU list of entries, used to determine 1211 * which ones to delete if the cache grows too large. 1212 * - if 32 <-> 64 translation mode is requested for a filesystem, 1213 * the cache also functions as a translation table 1214 * - in the translation case, invalidating the cache does not mean 1215 * flushing it, but just marking entries as invalid, except for 1216 * the <64bit cookie, 32bitcookie> pair which is still valid, to 1217 * still be able to use the cache as a translation table. 1218 * - 32 bit cookies are uniquely created by combining the hash table 1219 * entry value, and one generation count per hash table entry, 1220 * incremented each time an entry is appended to the chain. 1221 * - the cache is invalidated each time a direcory is modified 1222 * - sanity checks are also done; if an entry in a block turns 1223 * out not to have a matching cookie, the cache is invalidated 1224 * and a new block starting from the wanted offset is fetched from 1225 * the server. 1226 * - directory entries as read from the server are extended to contain 1227 * the 64bit and, optionally, the 32bit cookies, for sanity checking 1228 * the cache and exporting them to userspace through the cookie 1229 * argument to VOP_READDIR. 1230 */ 1231 1232 u_long 1233 nfs_dirhash(off) 1234 off_t off; 1235 { 1236 int i; 1237 char *cp = (char *)&off; 1238 u_long sum = 0L; 1239 1240 for (i = 0 ; i < sizeof (off); i++) 1241 sum += *cp++; 1242 1243 return sum; 1244 } 1245 1246 #define _NFSDC_MTX(np) (&NFSTOV(np)->v_interlock) 1247 #define NFSDC_LOCK(np) simple_lock(_NFSDC_MTX(np)) 1248 #define NFSDC_UNLOCK(np) simple_unlock(_NFSDC_MTX(np)) 1249 #define NFSDC_ASSERT_LOCKED(np) LOCK_ASSERT(simple_lock_held(_NFSDC_MTX(np))) 1250 1251 void 1252 nfs_initdircache(vp) 1253 struct vnode *vp; 1254 { 1255 struct nfsnode *np = VTONFS(vp); 1256 struct nfsdirhashhead *dircache; 1257 1258 dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF, 1259 M_WAITOK, &nfsdirhashmask); 1260 1261 NFSDC_LOCK(np); 1262 if (np->n_dircache == NULL) { 1263 np->n_dircachesize = 0; 1264 np->n_dircache = dircache; 1265 dircache = NULL; 1266 TAILQ_INIT(&np->n_dirchain); 1267 } 1268 NFSDC_UNLOCK(np); 1269 if (dircache) 1270 hashdone(dircache, M_NFSDIROFF); 1271 } 1272 1273 void 1274 nfs_initdirxlatecookie(vp) 1275 struct vnode *vp; 1276 { 1277 struct nfsnode *np = VTONFS(vp); 1278 unsigned *dirgens; 1279 1280 KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE); 1281 1282 dirgens = malloc(NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF, 1283 M_WAITOK|M_ZERO); 1284 NFSDC_LOCK(np); 1285 if (np->n_dirgens == NULL) { 1286 np->n_dirgens = dirgens; 1287 dirgens = NULL; 1288 } 1289 NFSDC_UNLOCK(np); 1290 if (dirgens) 1291 free(dirgens, M_NFSDIROFF); 1292 } 1293 1294 static const struct nfsdircache dzero; 1295 1296 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *)); 1297 static void nfs_putdircache_unlocked __P((struct nfsnode *, 1298 struct nfsdircache *)); 1299 1300 static void 1301 nfs_unlinkdircache(np, ndp) 1302 struct nfsnode *np; 1303 struct nfsdircache *ndp; 1304 { 1305 1306 NFSDC_ASSERT_LOCKED(np); 1307 KASSERT(ndp != &dzero); 1308 1309 if (LIST_NEXT(ndp, dc_hash) == (void *)-1) 1310 return; 1311 1312 TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain); 1313 LIST_REMOVE(ndp, dc_hash); 1314 LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */ 1315 1316 nfs_putdircache_unlocked(np, ndp); 1317 } 1318 1319 void 1320 nfs_putdircache(np, ndp) 1321 struct nfsnode *np; 1322 struct nfsdircache *ndp; 1323 { 1324 int ref; 1325 1326 if (ndp == &dzero) 1327 return; 1328 1329 KASSERT(ndp->dc_refcnt > 0); 1330 NFSDC_LOCK(np); 1331 ref = --ndp->dc_refcnt; 1332 NFSDC_UNLOCK(np); 1333 1334 if (ref == 0) 1335 free(ndp, M_NFSDIROFF); 1336 } 1337 1338 static void 1339 nfs_putdircache_unlocked(np, ndp) 1340 struct nfsnode *np; 1341 struct nfsdircache *ndp; 1342 { 1343 int ref; 1344 1345 NFSDC_ASSERT_LOCKED(np); 1346 1347 if (ndp == &dzero) 1348 return; 1349 1350 KASSERT(ndp->dc_refcnt > 0); 1351 ref = --ndp->dc_refcnt; 1352 if (ref == 0) 1353 free(ndp, M_NFSDIROFF); 1354 } 1355 1356 struct nfsdircache * 1357 nfs_searchdircache(vp, off, do32, hashent) 1358 struct vnode *vp; 1359 off_t off; 1360 int do32; 1361 int *hashent; 1362 { 1363 struct nfsdirhashhead *ndhp; 1364 struct nfsdircache *ndp = NULL; 1365 struct nfsnode *np = VTONFS(vp); 1366 unsigned ent; 1367 1368 /* 1369 * Zero is always a valid cookie. 1370 */ 1371 if (off == 0) 1372 /* XXXUNCONST */ 1373 return (struct nfsdircache *)__UNCONST(&dzero); 1374 1375 if (!np->n_dircache) 1376 return NULL; 1377 1378 /* 1379 * We use a 32bit cookie as search key, directly reconstruct 1380 * the hashentry. Else use the hashfunction. 1381 */ 1382 if (do32) { 1383 ent = (u_int32_t)off >> 24; 1384 if (ent >= NFS_DIRHASHSIZ) 1385 return NULL; 1386 ndhp = &np->n_dircache[ent]; 1387 } else { 1388 ndhp = NFSDIRHASH(np, off); 1389 } 1390 1391 if (hashent) 1392 *hashent = (int)(ndhp - np->n_dircache); 1393 1394 NFSDC_LOCK(np); 1395 if (do32) { 1396 LIST_FOREACH(ndp, ndhp, dc_hash) { 1397 if (ndp->dc_cookie32 == (u_int32_t)off) { 1398 /* 1399 * An invalidated entry will become the 1400 * start of a new block fetched from 1401 * the server. 1402 */ 1403 if (ndp->dc_flags & NFSDC_INVALID) { 1404 ndp->dc_blkcookie = ndp->dc_cookie; 1405 ndp->dc_entry = 0; 1406 ndp->dc_flags &= ~NFSDC_INVALID; 1407 } 1408 break; 1409 } 1410 } 1411 } else { 1412 LIST_FOREACH(ndp, ndhp, dc_hash) { 1413 if (ndp->dc_cookie == off) 1414 break; 1415 } 1416 } 1417 if (ndp != NULL) 1418 ndp->dc_refcnt++; 1419 NFSDC_UNLOCK(np); 1420 return ndp; 1421 } 1422 1423 1424 struct nfsdircache * 1425 nfs_enterdircache(vp, off, blkoff, en, blkno) 1426 struct vnode *vp; 1427 off_t off, blkoff; 1428 int en; 1429 daddr_t blkno; 1430 { 1431 struct nfsnode *np = VTONFS(vp); 1432 struct nfsdirhashhead *ndhp; 1433 struct nfsdircache *ndp = NULL; 1434 struct nfsdircache *newndp = NULL; 1435 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1436 int hashent, gen, overwrite; 1437 1438 /* 1439 * XXX refuse entries for offset 0. amd(8) erroneously sets 1440 * cookie 0 for the '.' entry, making this necessary. This 1441 * isn't so bad, as 0 is a special case anyway. 1442 */ 1443 if (off == 0) 1444 /* XXXUNCONST */ 1445 return (struct nfsdircache *)__UNCONST(&dzero); 1446 1447 if (!np->n_dircache) 1448 /* 1449 * XXX would like to do this in nfs_nget but vtype 1450 * isn't known at that time. 1451 */ 1452 nfs_initdircache(vp); 1453 1454 if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens) 1455 nfs_initdirxlatecookie(vp); 1456 1457 retry: 1458 ndp = nfs_searchdircache(vp, off, 0, &hashent); 1459 1460 NFSDC_LOCK(np); 1461 if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) { 1462 /* 1463 * Overwriting an old entry. Check if it's the same. 1464 * If so, just return. If not, remove the old entry. 1465 */ 1466 if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en) 1467 goto done; 1468 nfs_unlinkdircache(np, ndp); 1469 nfs_putdircache_unlocked(np, ndp); 1470 ndp = NULL; 1471 } 1472 1473 ndhp = &np->n_dircache[hashent]; 1474 1475 if (!ndp) { 1476 if (newndp == NULL) { 1477 NFSDC_UNLOCK(np); 1478 newndp = malloc(sizeof(*ndp), M_NFSDIROFF, M_WAITOK); 1479 newndp->dc_refcnt = 1; 1480 LIST_NEXT(newndp, dc_hash) = (void *)-1; 1481 goto retry; 1482 } 1483 ndp = newndp; 1484 newndp = NULL; 1485 overwrite = 0; 1486 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 1487 /* 1488 * We're allocating a new entry, so bump the 1489 * generation number. 1490 */ 1491 gen = ++np->n_dirgens[hashent]; 1492 if (gen == 0) { 1493 np->n_dirgens[hashent]++; 1494 gen++; 1495 } 1496 ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff); 1497 } 1498 } else 1499 overwrite = 1; 1500 1501 ndp->dc_cookie = off; 1502 ndp->dc_blkcookie = blkoff; 1503 ndp->dc_entry = en; 1504 ndp->dc_flags = 0; 1505 1506 if (overwrite) 1507 goto done; 1508 1509 /* 1510 * If the maximum directory cookie cache size has been reached 1511 * for this node, take one off the front. The idea is that 1512 * directories are typically read front-to-back once, so that 1513 * the oldest entries can be thrown away without much performance 1514 * loss. 1515 */ 1516 if (np->n_dircachesize == NFS_MAXDIRCACHE) { 1517 nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain)); 1518 } else 1519 np->n_dircachesize++; 1520 1521 KASSERT(ndp->dc_refcnt == 1); 1522 LIST_INSERT_HEAD(ndhp, ndp, dc_hash); 1523 TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain); 1524 ndp->dc_refcnt++; 1525 done: 1526 KASSERT(ndp->dc_refcnt > 0); 1527 NFSDC_UNLOCK(np); 1528 if (newndp) 1529 nfs_putdircache(np, newndp); 1530 return ndp; 1531 } 1532 1533 void 1534 nfs_invaldircache(vp, flags) 1535 struct vnode *vp; 1536 int flags; 1537 { 1538 struct nfsnode *np = VTONFS(vp); 1539 struct nfsdircache *ndp = NULL; 1540 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1541 const boolean_t forcefree = flags & NFS_INVALDIRCACHE_FORCE; 1542 1543 #ifdef DIAGNOSTIC 1544 if (vp->v_type != VDIR) 1545 panic("nfs: invaldircache: not dir"); 1546 #endif 1547 1548 if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0) 1549 np->n_flag &= ~NEOFVALID; 1550 1551 if (!np->n_dircache) 1552 return; 1553 1554 NFSDC_LOCK(np); 1555 if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) { 1556 while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) { 1557 KASSERT(!forcefree || ndp->dc_refcnt == 1); 1558 nfs_unlinkdircache(np, ndp); 1559 } 1560 np->n_dircachesize = 0; 1561 if (forcefree && np->n_dirgens) { 1562 FREE(np->n_dirgens, M_NFSDIROFF); 1563 np->n_dirgens = NULL; 1564 } 1565 } else { 1566 TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain) 1567 ndp->dc_flags |= NFSDC_INVALID; 1568 } 1569 1570 NFSDC_UNLOCK(np); 1571 } 1572 1573 /* 1574 * Called once before VFS init to initialize shared and 1575 * server-specific data structures. 1576 */ 1577 void 1578 nfs_init() 1579 { 1580 nfsrtt.pos = 0; 1581 rpc_vers = txdr_unsigned(RPC_VER2); 1582 rpc_call = txdr_unsigned(RPC_CALL); 1583 rpc_reply = txdr_unsigned(RPC_REPLY); 1584 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED); 1585 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED); 1586 rpc_mismatch = txdr_unsigned(RPC_MISMATCH); 1587 rpc_autherr = txdr_unsigned(RPC_AUTHERR); 1588 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX); 1589 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4); 1590 nfs_prog = txdr_unsigned(NFS_PROG); 1591 nqnfs_prog = txdr_unsigned(NQNFS_PROG); 1592 nfs_true = txdr_unsigned(TRUE); 1593 nfs_false = txdr_unsigned(FALSE); 1594 nfs_xdrneg1 = txdr_unsigned(-1); 1595 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; 1596 if (nfs_ticks < 1) 1597 nfs_ticks = 1; 1598 #ifdef NFSSERVER 1599 nfsrv_init(0); /* Init server data structures */ 1600 nfsrv_initcache(); /* Init the server request cache */ 1601 pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript), 1602 0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr); 1603 #endif /* NFSSERVER */ 1604 1605 #if defined(NFSSERVER) || !defined(NFS_V2_ONLY) 1606 /* 1607 * Initialize the nqnfs data structures. 1608 */ 1609 if (nqnfsstarttime == 0) { 1610 nqnfsstarttime = boottime.tv_sec + nqsrv_maxlease 1611 + nqsrv_clockskew + nqsrv_writeslack; 1612 NQLOADNOVRAM(nqnfsstarttime); 1613 CIRCLEQ_INIT(&nqtimerhead); 1614 nqfhhashtbl = hashinit(NQLCHSZ, HASH_LIST, M_NQLEASE, 1615 M_WAITOK, &nqfhhash); 1616 } 1617 #endif 1618 1619 exithook_establish(nfs_exit, NULL); 1620 1621 /* 1622 * Initialize reply list and start timer 1623 */ 1624 TAILQ_INIT(&nfs_reqq); 1625 nfs_timer(NULL); 1626 MOWNER_ATTACH(&nfs_mowner); 1627 1628 #ifdef NFS 1629 /* Initialize the kqueue structures */ 1630 nfs_kqinit(); 1631 /* Initialize the iod structures */ 1632 nfs_iodinit(); 1633 #endif 1634 } 1635 1636 #ifdef NFS 1637 /* 1638 * Called once at VFS init to initialize client-specific data structures. 1639 */ 1640 void 1641 nfs_vfs_init() 1642 { 1643 nfs_nhinit(); /* Init the nfsnode table */ 1644 nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4); 1645 } 1646 1647 void 1648 nfs_vfs_reinit() 1649 { 1650 nfs_nhreinit(); 1651 } 1652 1653 void 1654 nfs_vfs_done() 1655 { 1656 nfs_nhdone(); 1657 } 1658 1659 /* 1660 * Attribute cache routines. 1661 * nfs_loadattrcache() - loads or updates the cache contents from attributes 1662 * that are on the mbuf list 1663 * nfs_getattrcache() - returns valid attributes if found in cache, returns 1664 * error otherwise 1665 */ 1666 1667 /* 1668 * Load the attribute cache (that lives in the nfsnode entry) with 1669 * the values on the mbuf list and 1670 * Iff vap not NULL 1671 * copy the attributes to *vaper 1672 */ 1673 int 1674 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags) 1675 struct vnode **vpp; 1676 struct mbuf **mdp; 1677 caddr_t *dposp; 1678 struct vattr *vaper; 1679 int flags; 1680 { 1681 int32_t t1; 1682 caddr_t cp2; 1683 int error = 0; 1684 struct mbuf *md; 1685 int v3 = NFS_ISV3(*vpp); 1686 1687 md = *mdp; 1688 t1 = (mtod(md, caddr_t) + md->m_len) - *dposp; 1689 error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2); 1690 if (error) 1691 return (error); 1692 return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags); 1693 } 1694 1695 int 1696 nfs_loadattrcache(vpp, fp, vaper, flags) 1697 struct vnode **vpp; 1698 struct nfs_fattr *fp; 1699 struct vattr *vaper; 1700 int flags; 1701 { 1702 struct vnode *vp = *vpp; 1703 struct vattr *vap; 1704 int v3 = NFS_ISV3(vp); 1705 enum vtype vtyp; 1706 u_short vmode; 1707 struct timespec mtime; 1708 struct timespec ctime; 1709 struct vnode *nvp; 1710 int32_t rdev; 1711 struct nfsnode *np; 1712 extern int (**spec_nfsv2nodeop_p) __P((void *)); 1713 uid_t uid; 1714 gid_t gid; 1715 1716 if (v3) { 1717 vtyp = nfsv3tov_type(fp->fa_type); 1718 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1719 rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1), 1720 fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2)); 1721 fxdr_nfsv3time(&fp->fa3_mtime, &mtime); 1722 fxdr_nfsv3time(&fp->fa3_ctime, &ctime); 1723 } else { 1724 vtyp = nfsv2tov_type(fp->fa_type); 1725 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1726 if (vtyp == VNON || vtyp == VREG) 1727 vtyp = IFTOVT(vmode); 1728 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); 1729 fxdr_nfsv2time(&fp->fa2_mtime, &mtime); 1730 ctime.tv_sec = fxdr_unsigned(u_int32_t, 1731 fp->fa2_ctime.nfsv2_sec); 1732 ctime.tv_nsec = 0; 1733 1734 /* 1735 * Really ugly NFSv2 kludge. 1736 */ 1737 if (vtyp == VCHR && rdev == 0xffffffff) 1738 vtyp = VFIFO; 1739 } 1740 1741 vmode &= ALLPERMS; 1742 1743 /* 1744 * If v_type == VNON it is a new node, so fill in the v_type, 1745 * n_mtime fields. Check to see if it represents a special 1746 * device, and if so, check for a possible alias. Once the 1747 * correct vnode has been obtained, fill in the rest of the 1748 * information. 1749 */ 1750 np = VTONFS(vp); 1751 if (vp->v_type == VNON) { 1752 vp->v_type = vtyp; 1753 if (vp->v_type == VFIFO) { 1754 extern int (**fifo_nfsv2nodeop_p) __P((void *)); 1755 vp->v_op = fifo_nfsv2nodeop_p; 1756 } else if (vp->v_type == VREG) { 1757 lockinit(&np->n_commitlock, PINOD, "nfsclock", 0, 0); 1758 } else if (vp->v_type == VCHR || vp->v_type == VBLK) { 1759 vp->v_op = spec_nfsv2nodeop_p; 1760 nvp = checkalias(vp, (dev_t)rdev, vp->v_mount); 1761 if (nvp) { 1762 /* 1763 * Discard unneeded vnode, but save its nfsnode. 1764 * Since the nfsnode does not have a lock, its 1765 * vnode lock has to be carried over. 1766 */ 1767 /* 1768 * XXX is the old node sure to be locked here? 1769 */ 1770 KASSERT(lockstatus(&vp->v_lock) == 1771 LK_EXCLUSIVE); 1772 nvp->v_data = vp->v_data; 1773 vp->v_data = NULL; 1774 VOP_UNLOCK(vp, 0); 1775 vp->v_op = spec_vnodeop_p; 1776 vrele(vp); 1777 vgone(vp); 1778 lockmgr(&nvp->v_lock, LK_EXCLUSIVE, 1779 &nvp->v_interlock); 1780 /* 1781 * Reinitialize aliased node. 1782 */ 1783 np->n_vnode = nvp; 1784 *vpp = vp = nvp; 1785 } 1786 } 1787 np->n_mtime = mtime; 1788 } 1789 uid = fxdr_unsigned(uid_t, fp->fa_uid); 1790 gid = fxdr_unsigned(gid_t, fp->fa_gid); 1791 vap = np->n_vattr; 1792 1793 /* 1794 * Invalidate access cache if uid, gid, mode or ctime changed. 1795 */ 1796 if (np->n_accstamp != -1 && 1797 (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode 1798 || timespeccmp(&ctime, &vap->va_ctime, !=))) 1799 np->n_accstamp = -1; 1800 1801 vap->va_type = vtyp; 1802 vap->va_mode = vmode; 1803 vap->va_rdev = (dev_t)rdev; 1804 vap->va_mtime = mtime; 1805 vap->va_ctime = ctime; 1806 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 1807 switch (vtyp) { 1808 case VDIR: 1809 vap->va_blocksize = NFS_DIRFRAGSIZ; 1810 break; 1811 case VBLK: 1812 vap->va_blocksize = BLKDEV_IOSIZE; 1813 break; 1814 case VCHR: 1815 vap->va_blocksize = MAXBSIZE; 1816 break; 1817 default: 1818 vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize : 1819 fxdr_unsigned(int32_t, fp->fa2_blocksize); 1820 break; 1821 } 1822 if (v3) { 1823 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1824 vap->va_uid = uid; 1825 vap->va_gid = gid; 1826 vap->va_size = fxdr_hyper(&fp->fa3_size); 1827 vap->va_bytes = fxdr_hyper(&fp->fa3_used); 1828 vap->va_fileid = fxdr_unsigned(int32_t, 1829 fp->fa3_fileid.nfsuquad[1]); 1830 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); 1831 vap->va_flags = 0; 1832 vap->va_filerev = 0; 1833 } else { 1834 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1835 vap->va_uid = uid; 1836 vap->va_gid = gid; 1837 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); 1838 vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks) 1839 * NFS_FABLKSIZE; 1840 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); 1841 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); 1842 vap->va_flags = 0; 1843 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec); 1844 vap->va_filerev = 0; 1845 } 1846 if (vap->va_size != np->n_size) { 1847 if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) { 1848 vap->va_size = np->n_size; 1849 } else { 1850 np->n_size = vap->va_size; 1851 if (vap->va_type == VREG) { 1852 /* 1853 * we can't free pages if NAC_NOTRUNC because 1854 * the pages can be owned by ourselves. 1855 */ 1856 if (flags & NAC_NOTRUNC) { 1857 np->n_flag |= NTRUNCDELAYED; 1858 } else { 1859 simple_lock(&vp->v_interlock); 1860 (void)VOP_PUTPAGES(vp, 0, 1861 0, PGO_SYNCIO | PGO_CLEANIT | 1862 PGO_FREE | PGO_ALLPAGES); 1863 uvm_vnp_setsize(vp, np->n_size); 1864 } 1865 } 1866 } 1867 } 1868 np->n_attrstamp = mono_time.tv_sec; 1869 if (vaper != NULL) { 1870 memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(*vap)); 1871 if (np->n_flag & NCHG) { 1872 if (np->n_flag & NACC) 1873 vaper->va_atime = np->n_atim; 1874 if (np->n_flag & NUPD) 1875 vaper->va_mtime = np->n_mtim; 1876 } 1877 } 1878 return (0); 1879 } 1880 1881 /* 1882 * Check the time stamp 1883 * If the cache is valid, copy contents to *vap and return 0 1884 * otherwise return an error 1885 */ 1886 int 1887 nfs_getattrcache(vp, vaper) 1888 struct vnode *vp; 1889 struct vattr *vaper; 1890 { 1891 struct nfsnode *np = VTONFS(vp); 1892 struct vattr *vap; 1893 1894 if (np->n_attrstamp == 0 || 1895 (mono_time.tv_sec - np->n_attrstamp) >= NFS_ATTRTIMEO(np)) { 1896 nfsstats.attrcache_misses++; 1897 return (ENOENT); 1898 } 1899 nfsstats.attrcache_hits++; 1900 vap = np->n_vattr; 1901 if (vap->va_size != np->n_size) { 1902 if (vap->va_type == VREG) { 1903 if (np->n_flag & NMODIFIED) { 1904 if (vap->va_size < np->n_size) 1905 vap->va_size = np->n_size; 1906 else 1907 np->n_size = vap->va_size; 1908 } else 1909 np->n_size = vap->va_size; 1910 uvm_vnp_setsize(vp, np->n_size); 1911 } else 1912 np->n_size = vap->va_size; 1913 } 1914 memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(struct vattr)); 1915 if (np->n_flag & NCHG) { 1916 if (np->n_flag & NACC) 1917 vaper->va_atime = np->n_atim; 1918 if (np->n_flag & NUPD) 1919 vaper->va_mtime = np->n_mtim; 1920 } 1921 return (0); 1922 } 1923 1924 void 1925 nfs_delayedtruncate(vp) 1926 struct vnode *vp; 1927 { 1928 struct nfsnode *np = VTONFS(vp); 1929 1930 if (np->n_flag & NTRUNCDELAYED) { 1931 np->n_flag &= ~NTRUNCDELAYED; 1932 simple_lock(&vp->v_interlock); 1933 (void)VOP_PUTPAGES(vp, 0, 1934 0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES); 1935 uvm_vnp_setsize(vp, np->n_size); 1936 } 1937 } 1938 1939 #define NFS_WCCKLUDGE_TIMEOUT (24 * 60 * 60) /* 1 day */ 1940 #define NFS_WCCKLUDGE(nmp, now) \ 1941 (((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \ 1942 ((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0) 1943 1944 /* 1945 * nfs_check_wccdata: check inaccurate wcc_data 1946 * 1947 * => return non-zero if we shouldn't trust the wcc_data. 1948 * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed". 1949 */ 1950 1951 int 1952 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime, 1953 struct timespec *mtime, boolean_t docheck) 1954 { 1955 int error = 0; 1956 1957 #if !defined(NFS_V2_ONLY) 1958 1959 if (docheck) { 1960 struct vnode *vp = NFSTOV(np); 1961 struct nfsmount *nmp; 1962 long now = mono_time.tv_sec; 1963 #if defined(DEBUG) 1964 const char *reason = NULL; /* XXX: gcc */ 1965 #endif 1966 1967 if (timespeccmp(&np->n_vattr->va_mtime, mtime, <=)) { 1968 #if defined(DEBUG) 1969 reason = "mtime"; 1970 #endif 1971 error = EINVAL; 1972 } 1973 1974 if (vp->v_type == VDIR && 1975 timespeccmp(&np->n_vattr->va_ctime, ctime, <=)) { 1976 #if defined(DEBUG) 1977 reason = "ctime"; 1978 #endif 1979 error = EINVAL; 1980 } 1981 1982 nmp = VFSTONFS(vp->v_mount); 1983 if (error) { 1984 1985 /* 1986 * despite of the fact that we've updated the file, 1987 * timestamps of the file were not updated as we 1988 * expected. 1989 * it means that the server has incompatible 1990 * semantics of timestamps or (more likely) 1991 * the server time is not precise enough to 1992 * track each modifications. 1993 * in that case, we disable wcc processing. 1994 * 1995 * yes, strictly speaking, we should disable all 1996 * caching. it's a compromise. 1997 */ 1998 1999 simple_lock(&nmp->nm_slock); 2000 #if defined(DEBUG) 2001 if (!NFS_WCCKLUDGE(nmp, now)) { 2002 printf("%s: inaccurate wcc data (%s) detected," 2003 " disabling wcc\n", 2004 vp->v_mount->mnt_stat.f_mntfromname, 2005 reason); 2006 } 2007 #endif 2008 nmp->nm_iflag |= NFSMNT_WCCKLUDGE; 2009 nmp->nm_wcckludgetime = now; 2010 simple_unlock(&nmp->nm_slock); 2011 } else if (NFS_WCCKLUDGE(nmp, now)) { 2012 error = EPERM; /* XXX */ 2013 } else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 2014 simple_lock(&nmp->nm_slock); 2015 if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 2016 #if defined(DEBUG) 2017 printf("%s: re-enabling wcc\n", 2018 vp->v_mount->mnt_stat.f_mntfromname); 2019 #endif 2020 nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE; 2021 } 2022 simple_unlock(&nmp->nm_slock); 2023 } 2024 } 2025 2026 #endif /* !defined(NFS_V2_ONLY) */ 2027 2028 return error; 2029 } 2030 2031 /* 2032 * Heuristic to see if the server XDR encodes directory cookies or not. 2033 * it is not supposed to, but a lot of servers may do this. Also, since 2034 * most/all servers will implement V2 as well, it is expected that they 2035 * may return just 32 bits worth of cookie information, so we need to 2036 * find out in which 32 bits this information is available. We do this 2037 * to avoid trouble with emulated binaries that can't handle 64 bit 2038 * directory offsets. 2039 */ 2040 2041 void 2042 nfs_cookieheuristic(vp, flagp, p, cred) 2043 struct vnode *vp; 2044 int *flagp; 2045 struct proc *p; 2046 struct ucred *cred; 2047 { 2048 struct uio auio; 2049 struct iovec aiov; 2050 caddr_t tbuf, cp; 2051 struct dirent *dp; 2052 off_t *cookies = NULL, *cop; 2053 int error, eof, nc, len; 2054 2055 MALLOC(tbuf, caddr_t, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK); 2056 2057 aiov.iov_base = tbuf; 2058 aiov.iov_len = NFS_DIRFRAGSIZ; 2059 auio.uio_iov = &aiov; 2060 auio.uio_iovcnt = 1; 2061 auio.uio_rw = UIO_READ; 2062 auio.uio_segflg = UIO_SYSSPACE; 2063 auio.uio_procp = NULL; 2064 auio.uio_resid = NFS_DIRFRAGSIZ; 2065 auio.uio_offset = 0; 2066 2067 error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc); 2068 2069 len = NFS_DIRFRAGSIZ - auio.uio_resid; 2070 if (error || len == 0) { 2071 FREE(tbuf, M_TEMP); 2072 if (cookies) 2073 free(cookies, M_TEMP); 2074 return; 2075 } 2076 2077 /* 2078 * Find the first valid entry and look at its offset cookie. 2079 */ 2080 2081 cp = tbuf; 2082 for (cop = cookies; len > 0; len -= dp->d_reclen) { 2083 dp = (struct dirent *)cp; 2084 if (dp->d_fileno != 0 && len >= dp->d_reclen) { 2085 if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) { 2086 *flagp |= NFSMNT_SWAPCOOKIE; 2087 nfs_invaldircache(vp, 0); 2088 nfs_vinvalbuf(vp, 0, cred, p, 1); 2089 } 2090 break; 2091 } 2092 cop++; 2093 cp += dp->d_reclen; 2094 } 2095 2096 FREE(tbuf, M_TEMP); 2097 free(cookies, M_TEMP); 2098 } 2099 #endif /* NFS */ 2100 2101 /* 2102 * Set up nameidata for a lookup() call and do it. 2103 * 2104 * If pubflag is set, this call is done for a lookup operation on the 2105 * public filehandle. In that case we allow crossing mountpoints and 2106 * absolute pathnames. However, the caller is expected to check that 2107 * the lookup result is within the public fs, and deny access if 2108 * it is not. 2109 */ 2110 int 2111 nfs_namei(ndp, fhp, len, slp, nam, mdp, dposp, retdirp, p, kerbflag, pubflag) 2112 struct nameidata *ndp; 2113 fhandle_t *fhp; 2114 uint32_t len; 2115 struct nfssvc_sock *slp; 2116 struct mbuf *nam; 2117 struct mbuf **mdp; 2118 caddr_t *dposp; 2119 struct vnode **retdirp; 2120 struct proc *p; 2121 int kerbflag, pubflag; 2122 { 2123 int i, rem; 2124 struct mbuf *md; 2125 char *fromcp, *tocp, *cp; 2126 struct iovec aiov; 2127 struct uio auio; 2128 struct vnode *dp; 2129 int error, rdonly, linklen; 2130 struct componentname *cnp = &ndp->ni_cnd; 2131 2132 *retdirp = (struct vnode *)0; 2133 2134 if ((len + 1) > MAXPATHLEN) 2135 return (ENAMETOOLONG); 2136 if (len == 0) 2137 return (EACCES); 2138 cnp->cn_pnbuf = PNBUF_GET(); 2139 2140 /* 2141 * Copy the name from the mbuf list to ndp->ni_pnbuf 2142 * and set the various ndp fields appropriately. 2143 */ 2144 fromcp = *dposp; 2145 tocp = cnp->cn_pnbuf; 2146 md = *mdp; 2147 rem = mtod(md, caddr_t) + md->m_len - fromcp; 2148 for (i = 0; i < len; i++) { 2149 while (rem == 0) { 2150 md = md->m_next; 2151 if (md == NULL) { 2152 error = EBADRPC; 2153 goto out; 2154 } 2155 fromcp = mtod(md, caddr_t); 2156 rem = md->m_len; 2157 } 2158 if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) { 2159 error = EACCES; 2160 goto out; 2161 } 2162 *tocp++ = *fromcp++; 2163 rem--; 2164 } 2165 *tocp = '\0'; 2166 *mdp = md; 2167 *dposp = fromcp; 2168 len = nfsm_rndup(len)-len; 2169 if (len > 0) { 2170 if (rem >= len) 2171 *dposp += len; 2172 else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0) 2173 goto out; 2174 } 2175 2176 /* 2177 * Extract and set starting directory. 2178 */ 2179 error = nfsrv_fhtovp(fhp, FALSE, &dp, ndp->ni_cnd.cn_cred, slp, 2180 nam, &rdonly, kerbflag, pubflag); 2181 if (error) 2182 goto out; 2183 if (dp->v_type != VDIR) { 2184 vrele(dp); 2185 error = ENOTDIR; 2186 goto out; 2187 } 2188 2189 if (rdonly) 2190 cnp->cn_flags |= RDONLY; 2191 2192 *retdirp = dp; 2193 2194 if (pubflag) { 2195 /* 2196 * Oh joy. For WebNFS, handle those pesky '%' escapes, 2197 * and the 'native path' indicator. 2198 */ 2199 cp = PNBUF_GET(); 2200 fromcp = cnp->cn_pnbuf; 2201 tocp = cp; 2202 if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) { 2203 switch ((unsigned char)*fromcp) { 2204 case WEBNFS_NATIVE_CHAR: 2205 /* 2206 * 'Native' path for us is the same 2207 * as a path according to the NFS spec, 2208 * just skip the escape char. 2209 */ 2210 fromcp++; 2211 break; 2212 /* 2213 * More may be added in the future, range 0x80-0xff 2214 */ 2215 default: 2216 error = EIO; 2217 PNBUF_PUT(cp); 2218 goto out; 2219 } 2220 } 2221 /* 2222 * Translate the '%' escapes, URL-style. 2223 */ 2224 while (*fromcp != '\0') { 2225 if (*fromcp == WEBNFS_ESC_CHAR) { 2226 if (fromcp[1] != '\0' && fromcp[2] != '\0') { 2227 fromcp++; 2228 *tocp++ = HEXSTRTOI(fromcp); 2229 fromcp += 2; 2230 continue; 2231 } else { 2232 error = ENOENT; 2233 PNBUF_PUT(cp); 2234 goto out; 2235 } 2236 } else 2237 *tocp++ = *fromcp++; 2238 } 2239 *tocp = '\0'; 2240 PNBUF_PUT(cnp->cn_pnbuf); 2241 cnp->cn_pnbuf = cp; 2242 } 2243 2244 ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1; 2245 ndp->ni_segflg = UIO_SYSSPACE; 2246 ndp->ni_rootdir = rootvnode; 2247 2248 if (pubflag) { 2249 ndp->ni_loopcnt = 0; 2250 if (cnp->cn_pnbuf[0] == '/') 2251 dp = rootvnode; 2252 } else { 2253 cnp->cn_flags |= NOCROSSMOUNT; 2254 } 2255 2256 cnp->cn_proc = p; 2257 VREF(dp); 2258 2259 for (;;) { 2260 cnp->cn_nameptr = cnp->cn_pnbuf; 2261 ndp->ni_startdir = dp; 2262 /* 2263 * And call lookup() to do the real work 2264 */ 2265 error = lookup(ndp); 2266 if (error) { 2267 PNBUF_PUT(cnp->cn_pnbuf); 2268 return (error); 2269 } 2270 /* 2271 * Check for encountering a symbolic link 2272 */ 2273 if ((cnp->cn_flags & ISSYMLINK) == 0) { 2274 if (cnp->cn_flags & (SAVENAME | SAVESTART)) 2275 cnp->cn_flags |= HASBUF; 2276 else 2277 PNBUF_PUT(cnp->cn_pnbuf); 2278 return (0); 2279 } else { 2280 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN)) 2281 VOP_UNLOCK(ndp->ni_dvp, 0); 2282 if (!pubflag) { 2283 error = EINVAL; 2284 break; 2285 } 2286 2287 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 2288 error = ELOOP; 2289 break; 2290 } 2291 if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) { 2292 error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred, 2293 cnp->cn_proc); 2294 if (error != 0) 2295 break; 2296 } 2297 if (ndp->ni_pathlen > 1) 2298 cp = PNBUF_GET(); 2299 else 2300 cp = cnp->cn_pnbuf; 2301 aiov.iov_base = cp; 2302 aiov.iov_len = MAXPATHLEN; 2303 auio.uio_iov = &aiov; 2304 auio.uio_iovcnt = 1; 2305 auio.uio_offset = 0; 2306 auio.uio_rw = UIO_READ; 2307 auio.uio_segflg = UIO_SYSSPACE; 2308 auio.uio_procp = NULL; 2309 auio.uio_resid = MAXPATHLEN; 2310 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 2311 if (error) { 2312 badlink: 2313 if (ndp->ni_pathlen > 1) 2314 PNBUF_PUT(cp); 2315 break; 2316 } 2317 linklen = MAXPATHLEN - auio.uio_resid; 2318 if (linklen == 0) { 2319 error = ENOENT; 2320 goto badlink; 2321 } 2322 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 2323 error = ENAMETOOLONG; 2324 goto badlink; 2325 } 2326 if (ndp->ni_pathlen > 1) { 2327 memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen); 2328 PNBUF_PUT(cnp->cn_pnbuf); 2329 cnp->cn_pnbuf = cp; 2330 } else 2331 cnp->cn_pnbuf[linklen] = '\0'; 2332 ndp->ni_pathlen += linklen; 2333 vput(ndp->ni_vp); 2334 dp = ndp->ni_dvp; 2335 /* 2336 * Check if root directory should replace current directory. 2337 */ 2338 if (cnp->cn_pnbuf[0] == '/') { 2339 vrele(dp); 2340 dp = ndp->ni_rootdir; 2341 VREF(dp); 2342 } 2343 } 2344 } 2345 vrele(ndp->ni_dvp); 2346 vput(ndp->ni_vp); 2347 ndp->ni_vp = NULL; 2348 out: 2349 PNBUF_PUT(cnp->cn_pnbuf); 2350 return (error); 2351 } 2352 2353 /* 2354 * A fiddled version of m_adj() that ensures null fill to a 32-bit 2355 * boundary and only trims off the back end 2356 * 2357 * 1. trim off 'len' bytes as m_adj(mp, -len). 2358 * 2. add zero-padding 'nul' bytes at the end of the mbuf chain. 2359 */ 2360 void 2361 nfs_zeropad(mp, len, nul) 2362 struct mbuf *mp; 2363 int len; 2364 int nul; 2365 { 2366 struct mbuf *m; 2367 int count; 2368 2369 /* 2370 * Trim from tail. Scan the mbuf chain, 2371 * calculating its length and finding the last mbuf. 2372 * If the adjustment only affects this mbuf, then just 2373 * adjust and return. Otherwise, rescan and truncate 2374 * after the remaining size. 2375 */ 2376 count = 0; 2377 m = mp; 2378 for (;;) { 2379 count += m->m_len; 2380 if (m->m_next == NULL) 2381 break; 2382 m = m->m_next; 2383 } 2384 2385 KDASSERT(count >= len); 2386 2387 if (m->m_len >= len) { 2388 m->m_len -= len; 2389 } else { 2390 count -= len; 2391 /* 2392 * Correct length for chain is "count". 2393 * Find the mbuf with last data, adjust its length, 2394 * and toss data from remaining mbufs on chain. 2395 */ 2396 for (m = mp; m; m = m->m_next) { 2397 if (m->m_len >= count) { 2398 m->m_len = count; 2399 break; 2400 } 2401 count -= m->m_len; 2402 } 2403 m_freem(m->m_next); 2404 m->m_next = NULL; 2405 } 2406 2407 KDASSERT(m->m_next == NULL); 2408 2409 /* 2410 * zero-padding. 2411 */ 2412 if (nul > 0) { 2413 char *cp; 2414 int i; 2415 2416 if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) { 2417 struct mbuf *n; 2418 2419 KDASSERT(MLEN >= nul); 2420 n = m_get(M_WAIT, MT_DATA); 2421 MCLAIM(n, &nfs_mowner); 2422 n->m_len = nul; 2423 n->m_next = NULL; 2424 m->m_next = n; 2425 cp = mtod(n, caddr_t); 2426 } else { 2427 cp = mtod(m, caddr_t) + m->m_len; 2428 m->m_len += nul; 2429 } 2430 for (i = 0; i < nul; i++) 2431 *cp++ = '\0'; 2432 } 2433 return; 2434 } 2435 2436 /* 2437 * Make these functions instead of macros, so that the kernel text size 2438 * doesn't get too big... 2439 */ 2440 void 2441 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp) 2442 struct nfsrv_descript *nfsd; 2443 int before_ret; 2444 struct vattr *before_vap; 2445 int after_ret; 2446 struct vattr *after_vap; 2447 struct mbuf **mbp; 2448 char **bposp; 2449 { 2450 struct mbuf *mb = *mbp; 2451 char *bpos = *bposp; 2452 u_int32_t *tl; 2453 2454 if (before_ret) { 2455 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2456 *tl = nfs_false; 2457 } else { 2458 nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED); 2459 *tl++ = nfs_true; 2460 txdr_hyper(before_vap->va_size, tl); 2461 tl += 2; 2462 txdr_nfsv3time(&(before_vap->va_mtime), tl); 2463 tl += 2; 2464 txdr_nfsv3time(&(before_vap->va_ctime), tl); 2465 } 2466 *bposp = bpos; 2467 *mbp = mb; 2468 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp); 2469 } 2470 2471 void 2472 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp) 2473 struct nfsrv_descript *nfsd; 2474 int after_ret; 2475 struct vattr *after_vap; 2476 struct mbuf **mbp; 2477 char **bposp; 2478 { 2479 struct mbuf *mb = *mbp; 2480 char *bpos = *bposp; 2481 u_int32_t *tl; 2482 struct nfs_fattr *fp; 2483 2484 if (after_ret) { 2485 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2486 *tl = nfs_false; 2487 } else { 2488 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR); 2489 *tl++ = nfs_true; 2490 fp = (struct nfs_fattr *)tl; 2491 nfsm_srvfattr(nfsd, after_vap, fp); 2492 } 2493 *mbp = mb; 2494 *bposp = bpos; 2495 } 2496 2497 void 2498 nfsm_srvfattr(nfsd, vap, fp) 2499 struct nfsrv_descript *nfsd; 2500 struct vattr *vap; 2501 struct nfs_fattr *fp; 2502 { 2503 2504 fp->fa_nlink = txdr_unsigned(vap->va_nlink); 2505 fp->fa_uid = txdr_unsigned(vap->va_uid); 2506 fp->fa_gid = txdr_unsigned(vap->va_gid); 2507 if (nfsd->nd_flag & ND_NFSV3) { 2508 fp->fa_type = vtonfsv3_type(vap->va_type); 2509 fp->fa_mode = vtonfsv3_mode(vap->va_mode); 2510 txdr_hyper(vap->va_size, &fp->fa3_size); 2511 txdr_hyper(vap->va_bytes, &fp->fa3_used); 2512 fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev)); 2513 fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev)); 2514 fp->fa3_fsid.nfsuquad[0] = 0; 2515 fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid); 2516 fp->fa3_fileid.nfsuquad[0] = 0; 2517 fp->fa3_fileid.nfsuquad[1] = txdr_unsigned(vap->va_fileid); 2518 txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime); 2519 txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime); 2520 txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime); 2521 } else { 2522 fp->fa_type = vtonfsv2_type(vap->va_type); 2523 fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 2524 fp->fa2_size = txdr_unsigned(vap->va_size); 2525 fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize); 2526 if (vap->va_type == VFIFO) 2527 fp->fa2_rdev = 0xffffffff; 2528 else 2529 fp->fa2_rdev = txdr_unsigned(vap->va_rdev); 2530 fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE); 2531 fp->fa2_fsid = txdr_unsigned(vap->va_fsid); 2532 fp->fa2_fileid = txdr_unsigned(vap->va_fileid); 2533 txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime); 2534 txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime); 2535 txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime); 2536 } 2537 } 2538 2539 /* 2540 * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked) 2541 * - look up fsid in mount list (if not found ret error) 2542 * - get vp and export rights by calling VFS_FHTOVP() 2543 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 2544 * - if not lockflag unlock it with VOP_UNLOCK() 2545 */ 2546 int 2547 nfsrv_fhtovp(fhp, lockflag, vpp, cred, slp, nam, rdonlyp, kerbflag, pubflag) 2548 fhandle_t *fhp; 2549 int lockflag; 2550 struct vnode **vpp; 2551 struct ucred *cred; 2552 struct nfssvc_sock *slp; 2553 struct mbuf *nam; 2554 int *rdonlyp; 2555 int kerbflag; 2556 { 2557 struct mount *mp; 2558 int i; 2559 struct ucred *credanon; 2560 int error, exflags; 2561 struct sockaddr_in *saddr; 2562 2563 *vpp = (struct vnode *)0; 2564 2565 if (nfs_ispublicfh(fhp)) { 2566 if (!pubflag || !nfs_pub.np_valid) 2567 return (ESTALE); 2568 fhp = &nfs_pub.np_handle; 2569 } 2570 2571 mp = vfs_getvfs(&fhp->fh_fsid); 2572 if (!mp) 2573 return (ESTALE); 2574 error = VFS_CHECKEXP(mp, nam, &exflags, &credanon); 2575 if (error) 2576 return (error); 2577 error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp); 2578 if (error) 2579 return (error); 2580 2581 if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) { 2582 saddr = mtod(nam, struct sockaddr_in *); 2583 if ((saddr->sin_family == AF_INET) && 2584 ntohs(saddr->sin_port) >= IPPORT_RESERVED) { 2585 vput(*vpp); 2586 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2587 } 2588 #ifdef INET6 2589 if ((saddr->sin_family == AF_INET6) && 2590 ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) { 2591 vput(*vpp); 2592 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2593 } 2594 #endif 2595 } 2596 /* 2597 * Check/setup credentials. 2598 */ 2599 if (exflags & MNT_EXKERB) { 2600 if (!kerbflag) { 2601 vput(*vpp); 2602 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2603 } 2604 } else if (kerbflag) { 2605 vput(*vpp); 2606 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2607 } else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) { 2608 cred->cr_uid = credanon->cr_uid; 2609 cred->cr_gid = credanon->cr_gid; 2610 for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++) 2611 cred->cr_groups[i] = credanon->cr_groups[i]; 2612 cred->cr_ngroups = i; 2613 } 2614 if (exflags & MNT_EXRDONLY) 2615 *rdonlyp = 1; 2616 else 2617 *rdonlyp = 0; 2618 if (!lockflag) 2619 VOP_UNLOCK(*vpp, 0); 2620 return (0); 2621 } 2622 2623 /* 2624 * WebNFS: check if a filehandle is a public filehandle. For v3, this 2625 * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has 2626 * transformed this to all zeroes in both cases, so check for it. 2627 */ 2628 int 2629 nfs_ispublicfh(fhp) 2630 fhandle_t *fhp; 2631 { 2632 char *cp = (char *)fhp; 2633 int i; 2634 2635 for (i = 0; i < NFSX_V3FH; i++) 2636 if (*cp++ != 0) 2637 return (FALSE); 2638 return (TRUE); 2639 } 2640 2641 /* 2642 * This function compares two net addresses by family and returns TRUE 2643 * if they are the same host. 2644 * If there is any doubt, return FALSE. 2645 * The AF_INET family is handled as a special case so that address mbufs 2646 * don't need to be saved to store "struct in_addr", which is only 4 bytes. 2647 */ 2648 int 2649 netaddr_match(family, haddr, nam) 2650 int family; 2651 union nethostaddr *haddr; 2652 struct mbuf *nam; 2653 { 2654 struct sockaddr_in *inetaddr; 2655 2656 switch (family) { 2657 case AF_INET: 2658 inetaddr = mtod(nam, struct sockaddr_in *); 2659 if (inetaddr->sin_family == AF_INET && 2660 inetaddr->sin_addr.s_addr == haddr->had_inetaddr) 2661 return (1); 2662 break; 2663 #ifdef INET6 2664 case AF_INET6: 2665 { 2666 struct sockaddr_in6 *sin6_1, *sin6_2; 2667 2668 sin6_1 = mtod(nam, struct sockaddr_in6 *); 2669 sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *); 2670 if (sin6_1->sin6_family == AF_INET6 && 2671 IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr)) 2672 return 1; 2673 } 2674 #endif 2675 #ifdef ISO 2676 case AF_ISO: 2677 { 2678 struct sockaddr_iso *isoaddr1, *isoaddr2; 2679 2680 isoaddr1 = mtod(nam, struct sockaddr_iso *); 2681 isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *); 2682 if (isoaddr1->siso_family == AF_ISO && 2683 isoaddr1->siso_nlen > 0 && 2684 isoaddr1->siso_nlen == isoaddr2->siso_nlen && 2685 SAME_ISOADDR(isoaddr1, isoaddr2)) 2686 return (1); 2687 break; 2688 } 2689 #endif /* ISO */ 2690 default: 2691 break; 2692 }; 2693 return (0); 2694 } 2695 2696 /* 2697 * The write verifier has changed (probably due to a server reboot), so all 2698 * PG_NEEDCOMMIT pages will have to be written again. Since they are marked 2699 * as dirty or are being written out just now, all this takes is clearing 2700 * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for 2701 * the mount point. 2702 */ 2703 void 2704 nfs_clearcommit(mp) 2705 struct mount *mp; 2706 { 2707 struct vnode *vp; 2708 struct nfsnode *np; 2709 struct vm_page *pg; 2710 struct nfsmount *nmp = VFSTONFS(mp); 2711 2712 lockmgr(&nmp->nm_writeverflock, LK_EXCLUSIVE, NULL); 2713 2714 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2715 KASSERT(vp->v_mount == mp); 2716 if (vp->v_type != VREG) 2717 continue; 2718 np = VTONFS(vp); 2719 np->n_pushlo = np->n_pushhi = np->n_pushedlo = 2720 np->n_pushedhi = 0; 2721 np->n_commitflags &= 2722 ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID); 2723 simple_lock(&vp->v_uobj.vmobjlock); 2724 TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) { 2725 pg->flags &= ~PG_NEEDCOMMIT; 2726 } 2727 simple_unlock(&vp->v_uobj.vmobjlock); 2728 } 2729 simple_lock(&nmp->nm_slock); 2730 nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF; 2731 simple_unlock(&nmp->nm_slock); 2732 lockmgr(&nmp->nm_writeverflock, LK_RELEASE, NULL); 2733 } 2734 2735 void 2736 nfs_merge_commit_ranges(vp) 2737 struct vnode *vp; 2738 { 2739 struct nfsnode *np = VTONFS(vp); 2740 2741 KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID); 2742 2743 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2744 np->n_pushedlo = np->n_pushlo; 2745 np->n_pushedhi = np->n_pushhi; 2746 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2747 } else { 2748 if (np->n_pushlo < np->n_pushedlo) 2749 np->n_pushedlo = np->n_pushlo; 2750 if (np->n_pushhi > np->n_pushedhi) 2751 np->n_pushedhi = np->n_pushhi; 2752 } 2753 2754 np->n_pushlo = np->n_pushhi = 0; 2755 np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID; 2756 2757 #ifdef NFS_DEBUG_COMMIT 2758 printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2759 (unsigned)np->n_pushedhi); 2760 #endif 2761 } 2762 2763 int 2764 nfs_in_committed_range(vp, off, len) 2765 struct vnode *vp; 2766 off_t off, len; 2767 { 2768 struct nfsnode *np = VTONFS(vp); 2769 off_t lo, hi; 2770 2771 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2772 return 0; 2773 lo = off; 2774 hi = lo + len; 2775 2776 return (lo >= np->n_pushedlo && hi <= np->n_pushedhi); 2777 } 2778 2779 int 2780 nfs_in_tobecommitted_range(vp, off, len) 2781 struct vnode *vp; 2782 off_t off, len; 2783 { 2784 struct nfsnode *np = VTONFS(vp); 2785 off_t lo, hi; 2786 2787 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2788 return 0; 2789 lo = off; 2790 hi = lo + len; 2791 2792 return (lo >= np->n_pushlo && hi <= np->n_pushhi); 2793 } 2794 2795 void 2796 nfs_add_committed_range(vp, off, len) 2797 struct vnode *vp; 2798 off_t off, len; 2799 { 2800 struct nfsnode *np = VTONFS(vp); 2801 off_t lo, hi; 2802 2803 lo = off; 2804 hi = lo + len; 2805 2806 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2807 np->n_pushedlo = lo; 2808 np->n_pushedhi = hi; 2809 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2810 } else { 2811 if (hi > np->n_pushedhi) 2812 np->n_pushedhi = hi; 2813 if (lo < np->n_pushedlo) 2814 np->n_pushedlo = lo; 2815 } 2816 #ifdef NFS_DEBUG_COMMIT 2817 printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2818 (unsigned)np->n_pushedhi); 2819 #endif 2820 } 2821 2822 void 2823 nfs_del_committed_range(vp, off, len) 2824 struct vnode *vp; 2825 off_t off, len; 2826 { 2827 struct nfsnode *np = VTONFS(vp); 2828 off_t lo, hi; 2829 2830 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2831 return; 2832 2833 lo = off; 2834 hi = lo + len; 2835 2836 if (lo > np->n_pushedhi || hi < np->n_pushedlo) 2837 return; 2838 if (lo <= np->n_pushedlo) 2839 np->n_pushedlo = hi; 2840 else if (hi >= np->n_pushedhi) 2841 np->n_pushedhi = lo; 2842 else { 2843 /* 2844 * XXX There's only one range. If the deleted range 2845 * is in the middle, pick the largest of the 2846 * contiguous ranges that it leaves. 2847 */ 2848 if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi)) 2849 np->n_pushedhi = lo; 2850 else 2851 np->n_pushedlo = hi; 2852 } 2853 #ifdef NFS_DEBUG_COMMIT 2854 printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2855 (unsigned)np->n_pushedhi); 2856 #endif 2857 } 2858 2859 void 2860 nfs_add_tobecommitted_range(vp, off, len) 2861 struct vnode *vp; 2862 off_t off, len; 2863 { 2864 struct nfsnode *np = VTONFS(vp); 2865 off_t lo, hi; 2866 2867 lo = off; 2868 hi = lo + len; 2869 2870 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) { 2871 np->n_pushlo = lo; 2872 np->n_pushhi = hi; 2873 np->n_commitflags |= NFS_COMMIT_PUSH_VALID; 2874 } else { 2875 if (lo < np->n_pushlo) 2876 np->n_pushlo = lo; 2877 if (hi > np->n_pushhi) 2878 np->n_pushhi = hi; 2879 } 2880 #ifdef NFS_DEBUG_COMMIT 2881 printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2882 (unsigned)np->n_pushhi); 2883 #endif 2884 } 2885 2886 void 2887 nfs_del_tobecommitted_range(vp, off, len) 2888 struct vnode *vp; 2889 off_t off, len; 2890 { 2891 struct nfsnode *np = VTONFS(vp); 2892 off_t lo, hi; 2893 2894 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2895 return; 2896 2897 lo = off; 2898 hi = lo + len; 2899 2900 if (lo > np->n_pushhi || hi < np->n_pushlo) 2901 return; 2902 2903 if (lo <= np->n_pushlo) 2904 np->n_pushlo = hi; 2905 else if (hi >= np->n_pushhi) 2906 np->n_pushhi = lo; 2907 else { 2908 /* 2909 * XXX There's only one range. If the deleted range 2910 * is in the middle, pick the largest of the 2911 * contiguous ranges that it leaves. 2912 */ 2913 if ((np->n_pushlo - lo) > (hi - np->n_pushhi)) 2914 np->n_pushhi = lo; 2915 else 2916 np->n_pushlo = hi; 2917 } 2918 #ifdef NFS_DEBUG_COMMIT 2919 printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2920 (unsigned)np->n_pushhi); 2921 #endif 2922 } 2923 2924 /* 2925 * Map errnos to NFS error numbers. For Version 3 also filter out error 2926 * numbers not specified for the associated procedure. 2927 */ 2928 int 2929 nfsrv_errmap(nd, err) 2930 struct nfsrv_descript *nd; 2931 int err; 2932 { 2933 const short *defaulterrp, *errp; 2934 2935 if (nd->nd_flag & ND_NFSV3) { 2936 if (nd->nd_procnum <= NFSPROC_COMMIT) { 2937 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum]; 2938 while (*++errp) { 2939 if (*errp == err) 2940 return (err); 2941 else if (*errp > err) 2942 break; 2943 } 2944 return ((int)*defaulterrp); 2945 } else 2946 return (err & 0xffff); 2947 } 2948 if (err <= ELAST) 2949 return ((int)nfsrv_v2errmap[err - 1]); 2950 return (NFSERR_IO); 2951 } 2952 2953 /* 2954 * Sort the group list in increasing numerical order. 2955 * (Insertion sort by Chris Torek, who was grossed out by the bubble sort 2956 * that used to be here.) 2957 */ 2958 void 2959 nfsrvw_sort(list, num) 2960 gid_t *list; 2961 int num; 2962 { 2963 int i, j; 2964 gid_t v; 2965 2966 /* Insertion sort. */ 2967 for (i = 1; i < num; i++) { 2968 v = list[i]; 2969 /* find correct slot for value v, moving others up */ 2970 for (j = i; --j >= 0 && v < list[j];) 2971 list[j + 1] = list[j]; 2972 list[j + 1] = v; 2973 } 2974 } 2975 2976 /* 2977 * copy credentials making sure that the result can be compared with memcmp(). 2978 */ 2979 void 2980 nfsrv_setcred(incred, outcred) 2981 struct ucred *incred, *outcred; 2982 { 2983 int i; 2984 2985 memset((caddr_t)outcred, 0, sizeof (struct ucred)); 2986 outcred->cr_ref = 1; 2987 outcred->cr_uid = incred->cr_uid; 2988 outcred->cr_gid = incred->cr_gid; 2989 outcred->cr_ngroups = incred->cr_ngroups; 2990 for (i = 0; i < incred->cr_ngroups; i++) 2991 outcred->cr_groups[i] = incred->cr_groups[i]; 2992 nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups); 2993 } 2994 2995 u_int32_t 2996 nfs_getxid() 2997 { 2998 static u_int32_t base; 2999 static u_int32_t nfs_xid = 0; 3000 static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER; 3001 u_int32_t newxid; 3002 3003 simple_lock(&nfs_xidlock); 3004 /* 3005 * derive initial xid from system time 3006 * XXX time is invalid if root not yet mounted 3007 */ 3008 if (__predict_false(!base && (rootvp))) { 3009 struct timeval tv; 3010 3011 microtime(&tv); 3012 base = tv.tv_sec << 12; 3013 nfs_xid = base; 3014 } 3015 3016 /* 3017 * Skip zero xid if it should ever happen. 3018 */ 3019 if (__predict_false(++nfs_xid == 0)) 3020 nfs_xid++; 3021 newxid = nfs_xid; 3022 simple_unlock(&nfs_xidlock); 3023 3024 return txdr_unsigned(newxid); 3025 } 3026 3027 /* 3028 * assign a new xid for existing request. 3029 * used for NFSERR_JUKEBOX handling. 3030 */ 3031 void 3032 nfs_renewxid(struct nfsreq *req) 3033 { 3034 u_int32_t xid; 3035 int off; 3036 3037 xid = nfs_getxid(); 3038 if (req->r_nmp->nm_sotype == SOCK_STREAM) 3039 off = sizeof(u_int32_t); /* RPC record mark */ 3040 else 3041 off = 0; 3042 3043 m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid); 3044 req->r_xid = xid; 3045 } 3046
Indexes created Thu Oct 02 14:10:14 GMT 2025