nfs_subs.c revision 1.149.2.2
1 /* $NetBSD: nfs_subs.c,v 1.149.2.2 2006/06/21 15:11:59 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 1.149.2.2 2006/06/21 15:11:59 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/nqnfs.h> 113 #include <nfs/nfsrtt.h> 114 #include <nfs/nfs_var.h> 115 116 #include <miscfs/specfs/specdev.h> 117 118 #include <netinet/in.h> 119 #ifdef ISO 120 #include <netiso/iso.h> 121 #endif 122 123 /* 124 * Data items converted to xdr at startup, since they are constant 125 * This is kinda hokey, but may save a little time doing byte swaps 126 */ 127 u_int32_t nfs_xdrneg1; 128 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr, 129 rpc_mismatch, rpc_auth_unix, rpc_msgaccepted, 130 rpc_auth_kerb; 131 u_int32_t nfs_prog, nqnfs_prog, nfs_true, nfs_false; 132 133 /* And other global data */ 134 const nfstype nfsv2_type[9] = 135 { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON }; 136 const nfstype nfsv3_type[9] = 137 { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON }; 138 const enum vtype nv2tov_type[8] = 139 { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON }; 140 const enum vtype nv3tov_type[8] = 141 { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO }; 142 int nfs_ticks; 143 int nfs_commitsize; 144 145 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies"); 146 147 /* NFS client/server stats. */ 148 struct nfsstats nfsstats; 149 150 /* 151 * Mapping of old NFS Version 2 RPC numbers to generic numbers. 152 */ 153 const int nfsv3_procid[NFS_NPROCS] = { 154 NFSPROC_NULL, 155 NFSPROC_GETATTR, 156 NFSPROC_SETATTR, 157 NFSPROC_NOOP, 158 NFSPROC_LOOKUP, 159 NFSPROC_READLINK, 160 NFSPROC_READ, 161 NFSPROC_NOOP, 162 NFSPROC_WRITE, 163 NFSPROC_CREATE, 164 NFSPROC_REMOVE, 165 NFSPROC_RENAME, 166 NFSPROC_LINK, 167 NFSPROC_SYMLINK, 168 NFSPROC_MKDIR, 169 NFSPROC_RMDIR, 170 NFSPROC_READDIR, 171 NFSPROC_FSSTAT, 172 NFSPROC_NOOP, 173 NFSPROC_NOOP, 174 NFSPROC_NOOP, 175 NFSPROC_NOOP, 176 NFSPROC_NOOP, 177 NFSPROC_NOOP, 178 NFSPROC_NOOP, 179 NFSPROC_NOOP 180 }; 181 182 /* 183 * and the reverse mapping from generic to Version 2 procedure numbers 184 */ 185 const int nfsv2_procid[NFS_NPROCS] = { 186 NFSV2PROC_NULL, 187 NFSV2PROC_GETATTR, 188 NFSV2PROC_SETATTR, 189 NFSV2PROC_LOOKUP, 190 NFSV2PROC_NOOP, 191 NFSV2PROC_READLINK, 192 NFSV2PROC_READ, 193 NFSV2PROC_WRITE, 194 NFSV2PROC_CREATE, 195 NFSV2PROC_MKDIR, 196 NFSV2PROC_SYMLINK, 197 NFSV2PROC_CREATE, 198 NFSV2PROC_REMOVE, 199 NFSV2PROC_RMDIR, 200 NFSV2PROC_RENAME, 201 NFSV2PROC_LINK, 202 NFSV2PROC_READDIR, 203 NFSV2PROC_NOOP, 204 NFSV2PROC_STATFS, 205 NFSV2PROC_NOOP, 206 NFSV2PROC_NOOP, 207 NFSV2PROC_NOOP, 208 NFSV2PROC_NOOP, 209 NFSV2PROC_NOOP, 210 NFSV2PROC_NOOP, 211 NFSV2PROC_NOOP, 212 }; 213 214 /* 215 * Maps errno values to nfs error numbers. 216 * Use NFSERR_IO as the catch all for ones not specifically defined in 217 * RFC 1094. 218 */ 219 static const u_char nfsrv_v2errmap[ELAST] = { 220 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 221 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 222 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 223 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 224 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 225 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 226 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 227 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 228 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 229 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 230 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 231 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 232 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO, 233 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 234 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 235 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 236 NFSERR_IO, NFSERR_IO, 237 }; 238 239 /* 240 * Maps errno values to nfs error numbers. 241 * Although it is not obvious whether or not NFS clients really care if 242 * a returned error value is in the specified list for the procedure, the 243 * safest thing to do is filter them appropriately. For Version 2, the 244 * X/Open XNFS document is the only specification that defines error values 245 * for each RPC (The RFC simply lists all possible error values for all RPCs), 246 * so I have decided to not do this for Version 2. 247 * The first entry is the default error return and the rest are the valid 248 * errors for that RPC in increasing numeric order. 249 */ 250 static const short nfsv3err_null[] = { 251 0, 252 0, 253 }; 254 255 static const short nfsv3err_getattr[] = { 256 NFSERR_IO, 257 NFSERR_IO, 258 NFSERR_STALE, 259 NFSERR_BADHANDLE, 260 NFSERR_SERVERFAULT, 261 0, 262 }; 263 264 static const short nfsv3err_setattr[] = { 265 NFSERR_IO, 266 NFSERR_PERM, 267 NFSERR_IO, 268 NFSERR_ACCES, 269 NFSERR_INVAL, 270 NFSERR_NOSPC, 271 NFSERR_ROFS, 272 NFSERR_DQUOT, 273 NFSERR_STALE, 274 NFSERR_BADHANDLE, 275 NFSERR_NOT_SYNC, 276 NFSERR_SERVERFAULT, 277 0, 278 }; 279 280 static const short nfsv3err_lookup[] = { 281 NFSERR_IO, 282 NFSERR_NOENT, 283 NFSERR_IO, 284 NFSERR_ACCES, 285 NFSERR_NOTDIR, 286 NFSERR_NAMETOL, 287 NFSERR_STALE, 288 NFSERR_BADHANDLE, 289 NFSERR_SERVERFAULT, 290 0, 291 }; 292 293 static const short nfsv3err_access[] = { 294 NFSERR_IO, 295 NFSERR_IO, 296 NFSERR_STALE, 297 NFSERR_BADHANDLE, 298 NFSERR_SERVERFAULT, 299 0, 300 }; 301 302 static const short nfsv3err_readlink[] = { 303 NFSERR_IO, 304 NFSERR_IO, 305 NFSERR_ACCES, 306 NFSERR_INVAL, 307 NFSERR_STALE, 308 NFSERR_BADHANDLE, 309 NFSERR_NOTSUPP, 310 NFSERR_SERVERFAULT, 311 0, 312 }; 313 314 static const short nfsv3err_read[] = { 315 NFSERR_IO, 316 NFSERR_IO, 317 NFSERR_NXIO, 318 NFSERR_ACCES, 319 NFSERR_INVAL, 320 NFSERR_STALE, 321 NFSERR_BADHANDLE, 322 NFSERR_SERVERFAULT, 323 NFSERR_JUKEBOX, 324 0, 325 }; 326 327 static const short nfsv3err_write[] = { 328 NFSERR_IO, 329 NFSERR_IO, 330 NFSERR_ACCES, 331 NFSERR_INVAL, 332 NFSERR_FBIG, 333 NFSERR_NOSPC, 334 NFSERR_ROFS, 335 NFSERR_DQUOT, 336 NFSERR_STALE, 337 NFSERR_BADHANDLE, 338 NFSERR_SERVERFAULT, 339 NFSERR_JUKEBOX, 340 0, 341 }; 342 343 static const short nfsv3err_create[] = { 344 NFSERR_IO, 345 NFSERR_IO, 346 NFSERR_ACCES, 347 NFSERR_EXIST, 348 NFSERR_NOTDIR, 349 NFSERR_NOSPC, 350 NFSERR_ROFS, 351 NFSERR_NAMETOL, 352 NFSERR_DQUOT, 353 NFSERR_STALE, 354 NFSERR_BADHANDLE, 355 NFSERR_NOTSUPP, 356 NFSERR_SERVERFAULT, 357 0, 358 }; 359 360 static const short nfsv3err_mkdir[] = { 361 NFSERR_IO, 362 NFSERR_IO, 363 NFSERR_ACCES, 364 NFSERR_EXIST, 365 NFSERR_NOTDIR, 366 NFSERR_NOSPC, 367 NFSERR_ROFS, 368 NFSERR_NAMETOL, 369 NFSERR_DQUOT, 370 NFSERR_STALE, 371 NFSERR_BADHANDLE, 372 NFSERR_NOTSUPP, 373 NFSERR_SERVERFAULT, 374 0, 375 }; 376 377 static const short nfsv3err_symlink[] = { 378 NFSERR_IO, 379 NFSERR_IO, 380 NFSERR_ACCES, 381 NFSERR_EXIST, 382 NFSERR_NOTDIR, 383 NFSERR_NOSPC, 384 NFSERR_ROFS, 385 NFSERR_NAMETOL, 386 NFSERR_DQUOT, 387 NFSERR_STALE, 388 NFSERR_BADHANDLE, 389 NFSERR_NOTSUPP, 390 NFSERR_SERVERFAULT, 391 0, 392 }; 393 394 static const short nfsv3err_mknod[] = { 395 NFSERR_IO, 396 NFSERR_IO, 397 NFSERR_ACCES, 398 NFSERR_EXIST, 399 NFSERR_NOTDIR, 400 NFSERR_NOSPC, 401 NFSERR_ROFS, 402 NFSERR_NAMETOL, 403 NFSERR_DQUOT, 404 NFSERR_STALE, 405 NFSERR_BADHANDLE, 406 NFSERR_NOTSUPP, 407 NFSERR_SERVERFAULT, 408 NFSERR_BADTYPE, 409 0, 410 }; 411 412 static const short nfsv3err_remove[] = { 413 NFSERR_IO, 414 NFSERR_NOENT, 415 NFSERR_IO, 416 NFSERR_ACCES, 417 NFSERR_NOTDIR, 418 NFSERR_ROFS, 419 NFSERR_NAMETOL, 420 NFSERR_STALE, 421 NFSERR_BADHANDLE, 422 NFSERR_SERVERFAULT, 423 0, 424 }; 425 426 static const short nfsv3err_rmdir[] = { 427 NFSERR_IO, 428 NFSERR_NOENT, 429 NFSERR_IO, 430 NFSERR_ACCES, 431 NFSERR_EXIST, 432 NFSERR_NOTDIR, 433 NFSERR_INVAL, 434 NFSERR_ROFS, 435 NFSERR_NAMETOL, 436 NFSERR_NOTEMPTY, 437 NFSERR_STALE, 438 NFSERR_BADHANDLE, 439 NFSERR_NOTSUPP, 440 NFSERR_SERVERFAULT, 441 0, 442 }; 443 444 static const short nfsv3err_rename[] = { 445 NFSERR_IO, 446 NFSERR_NOENT, 447 NFSERR_IO, 448 NFSERR_ACCES, 449 NFSERR_EXIST, 450 NFSERR_XDEV, 451 NFSERR_NOTDIR, 452 NFSERR_ISDIR, 453 NFSERR_INVAL, 454 NFSERR_NOSPC, 455 NFSERR_ROFS, 456 NFSERR_MLINK, 457 NFSERR_NAMETOL, 458 NFSERR_NOTEMPTY, 459 NFSERR_DQUOT, 460 NFSERR_STALE, 461 NFSERR_BADHANDLE, 462 NFSERR_NOTSUPP, 463 NFSERR_SERVERFAULT, 464 0, 465 }; 466 467 static const short nfsv3err_link[] = { 468 NFSERR_IO, 469 NFSERR_IO, 470 NFSERR_ACCES, 471 NFSERR_EXIST, 472 NFSERR_XDEV, 473 NFSERR_NOTDIR, 474 NFSERR_INVAL, 475 NFSERR_NOSPC, 476 NFSERR_ROFS, 477 NFSERR_MLINK, 478 NFSERR_NAMETOL, 479 NFSERR_DQUOT, 480 NFSERR_STALE, 481 NFSERR_BADHANDLE, 482 NFSERR_NOTSUPP, 483 NFSERR_SERVERFAULT, 484 0, 485 }; 486 487 static const short nfsv3err_readdir[] = { 488 NFSERR_IO, 489 NFSERR_IO, 490 NFSERR_ACCES, 491 NFSERR_NOTDIR, 492 NFSERR_STALE, 493 NFSERR_BADHANDLE, 494 NFSERR_BAD_COOKIE, 495 NFSERR_TOOSMALL, 496 NFSERR_SERVERFAULT, 497 0, 498 }; 499 500 static const short nfsv3err_readdirplus[] = { 501 NFSERR_IO, 502 NFSERR_IO, 503 NFSERR_ACCES, 504 NFSERR_NOTDIR, 505 NFSERR_STALE, 506 NFSERR_BADHANDLE, 507 NFSERR_BAD_COOKIE, 508 NFSERR_NOTSUPP, 509 NFSERR_TOOSMALL, 510 NFSERR_SERVERFAULT, 511 0, 512 }; 513 514 static const short nfsv3err_fsstat[] = { 515 NFSERR_IO, 516 NFSERR_IO, 517 NFSERR_STALE, 518 NFSERR_BADHANDLE, 519 NFSERR_SERVERFAULT, 520 0, 521 }; 522 523 static const short nfsv3err_fsinfo[] = { 524 NFSERR_STALE, 525 NFSERR_STALE, 526 NFSERR_BADHANDLE, 527 NFSERR_SERVERFAULT, 528 0, 529 }; 530 531 static const short nfsv3err_pathconf[] = { 532 NFSERR_STALE, 533 NFSERR_STALE, 534 NFSERR_BADHANDLE, 535 NFSERR_SERVERFAULT, 536 0, 537 }; 538 539 static const short nfsv3err_commit[] = { 540 NFSERR_IO, 541 NFSERR_IO, 542 NFSERR_STALE, 543 NFSERR_BADHANDLE, 544 NFSERR_SERVERFAULT, 545 0, 546 }; 547 548 static const short * const nfsrv_v3errmap[] = { 549 nfsv3err_null, 550 nfsv3err_getattr, 551 nfsv3err_setattr, 552 nfsv3err_lookup, 553 nfsv3err_access, 554 nfsv3err_readlink, 555 nfsv3err_read, 556 nfsv3err_write, 557 nfsv3err_create, 558 nfsv3err_mkdir, 559 nfsv3err_symlink, 560 nfsv3err_mknod, 561 nfsv3err_remove, 562 nfsv3err_rmdir, 563 nfsv3err_rename, 564 nfsv3err_link, 565 nfsv3err_readdir, 566 nfsv3err_readdirplus, 567 nfsv3err_fsstat, 568 nfsv3err_fsinfo, 569 nfsv3err_pathconf, 570 nfsv3err_commit, 571 }; 572 573 extern struct nfsrtt nfsrtt; 574 extern time_t nqnfsstarttime; 575 extern int nqsrv_clockskew; 576 extern int nqsrv_writeslack; 577 extern int nqsrv_maxlease; 578 extern const int nqnfs_piggy[NFS_NPROCS]; 579 extern struct nfsnodehashhead *nfsnodehashtbl; 580 extern u_long nfsnodehash; 581 582 u_long nfsdirhashmask; 583 584 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *)); 585 586 /* 587 * Create the header for an rpc request packet 588 * The hsiz is the size of the rest of the nfs request header. 589 * (just used to decide if a cluster is a good idea) 590 */ 591 struct mbuf * 592 nfsm_reqh(np, procid, hsiz, bposp) 593 struct nfsnode *np; 594 u_long procid; 595 int hsiz; 596 caddr_t *bposp; 597 { 598 struct mbuf *mb; 599 caddr_t bpos; 600 #ifndef NFS_V2_ONLY 601 struct nfsmount *nmp; 602 u_int32_t *tl; 603 int nqflag; 604 #endif 605 606 mb = m_get(M_WAIT, MT_DATA); 607 MCLAIM(mb, &nfs_mowner); 608 if (hsiz >= MINCLSIZE) 609 m_clget(mb, M_WAIT); 610 mb->m_len = 0; 611 bpos = mtod(mb, caddr_t); 612 613 #ifndef NFS_V2_ONLY 614 /* 615 * For NQNFS, add lease request. 616 */ 617 if (np) { 618 nmp = VFSTONFS(np->n_vnode->v_mount); 619 if (nmp->nm_flag & NFSMNT_NQNFS) { 620 nqflag = NQNFS_NEEDLEASE(np, procid); 621 if (nqflag) { 622 nfsm_build(tl, u_int32_t *, 2*NFSX_UNSIGNED); 623 *tl++ = txdr_unsigned(nqflag); 624 *tl = txdr_unsigned(nmp->nm_leaseterm); 625 } else { 626 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 627 *tl = 0; 628 } 629 } 630 } 631 #endif 632 /* Finally, return values */ 633 *bposp = bpos; 634 return (mb); 635 } 636 637 /* 638 * Build the RPC header and fill in the authorization info. 639 * The authorization string argument is only used when the credentials 640 * come from outside of the kernel. 641 * Returns the head of the mbuf list. 642 */ 643 struct mbuf * 644 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len, 645 verf_str, mrest, mrest_len, mbp, xidp) 646 kauth_cred_t cr; 647 int nmflag; 648 int procid; 649 int auth_type; 650 int auth_len; 651 char *auth_str; 652 int verf_len; 653 char *verf_str; 654 struct mbuf *mrest; 655 int mrest_len; 656 struct mbuf **mbp; 657 u_int32_t *xidp; 658 { 659 struct mbuf *mb; 660 u_int32_t *tl; 661 caddr_t bpos; 662 int i; 663 struct mbuf *mreq; 664 int siz, grpsiz, authsiz; 665 666 authsiz = nfsm_rndup(auth_len); 667 mb = m_gethdr(M_WAIT, MT_DATA); 668 MCLAIM(mb, &nfs_mowner); 669 if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) { 670 m_clget(mb, M_WAIT); 671 } else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) { 672 MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED); 673 } else { 674 MH_ALIGN(mb, 8 * NFSX_UNSIGNED); 675 } 676 mb->m_len = 0; 677 mreq = mb; 678 bpos = mtod(mb, caddr_t); 679 680 /* 681 * First the RPC header. 682 */ 683 nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 684 685 *tl++ = *xidp = nfs_getxid(); 686 *tl++ = rpc_call; 687 *tl++ = rpc_vers; 688 if (nmflag & NFSMNT_NQNFS) { 689 *tl++ = txdr_unsigned(NQNFS_PROG); 690 *tl++ = txdr_unsigned(NQNFS_VER3); 691 } else { 692 *tl++ = txdr_unsigned(NFS_PROG); 693 if (nmflag & NFSMNT_NFSV3) 694 *tl++ = txdr_unsigned(NFS_VER3); 695 else 696 *tl++ = txdr_unsigned(NFS_VER2); 697 } 698 if (nmflag & NFSMNT_NFSV3) 699 *tl++ = txdr_unsigned(procid); 700 else 701 *tl++ = txdr_unsigned(nfsv2_procid[procid]); 702 703 /* 704 * And then the authorization cred. 705 */ 706 *tl++ = txdr_unsigned(auth_type); 707 *tl = txdr_unsigned(authsiz); 708 switch (auth_type) { 709 case RPCAUTH_UNIX: 710 nfsm_build(tl, u_int32_t *, auth_len); 711 *tl++ = 0; /* stamp ?? */ 712 *tl++ = 0; /* NULL hostname */ 713 *tl++ = txdr_unsigned(kauth_cred_geteuid(cr)); 714 *tl++ = txdr_unsigned(kauth_cred_getegid(cr)); 715 grpsiz = (auth_len >> 2) - 5; 716 *tl++ = txdr_unsigned(grpsiz); 717 for (i = 0; i < grpsiz; i++) 718 *tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */ 719 break; 720 case RPCAUTH_KERB4: 721 siz = auth_len; 722 while (siz > 0) { 723 if (M_TRAILINGSPACE(mb) == 0) { 724 struct mbuf *mb2; 725 mb2 = m_get(M_WAIT, MT_DATA); 726 MCLAIM(mb2, &nfs_mowner); 727 if (siz >= MINCLSIZE) 728 m_clget(mb2, M_WAIT); 729 mb->m_next = mb2; 730 mb = mb2; 731 mb->m_len = 0; 732 bpos = mtod(mb, caddr_t); 733 } 734 i = min(siz, M_TRAILINGSPACE(mb)); 735 memcpy(bpos, auth_str, i); 736 mb->m_len += i; 737 auth_str += i; 738 bpos += i; 739 siz -= i; 740 } 741 if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) { 742 for (i = 0; i < siz; i++) 743 *bpos++ = '\0'; 744 mb->m_len += siz; 745 } 746 break; 747 }; 748 749 /* 750 * And the verifier... 751 */ 752 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 753 if (verf_str) { 754 *tl++ = txdr_unsigned(RPCAUTH_KERB4); 755 *tl = txdr_unsigned(verf_len); 756 siz = verf_len; 757 while (siz > 0) { 758 if (M_TRAILINGSPACE(mb) == 0) { 759 struct mbuf *mb2; 760 mb2 = m_get(M_WAIT, MT_DATA); 761 MCLAIM(mb2, &nfs_mowner); 762 if (siz >= MINCLSIZE) 763 m_clget(mb2, M_WAIT); 764 mb->m_next = mb2; 765 mb = mb2; 766 mb->m_len = 0; 767 bpos = mtod(mb, caddr_t); 768 } 769 i = min(siz, M_TRAILINGSPACE(mb)); 770 memcpy(bpos, verf_str, i); 771 mb->m_len += i; 772 verf_str += i; 773 bpos += i; 774 siz -= i; 775 } 776 if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) { 777 for (i = 0; i < siz; i++) 778 *bpos++ = '\0'; 779 mb->m_len += siz; 780 } 781 } else { 782 *tl++ = txdr_unsigned(RPCAUTH_NULL); 783 *tl = 0; 784 } 785 mb->m_next = mrest; 786 mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len; 787 mreq->m_pkthdr.rcvif = (struct ifnet *)0; 788 *mbp = mb; 789 return (mreq); 790 } 791 792 /* 793 * copies mbuf chain to the uio scatter/gather list 794 */ 795 int 796 nfsm_mbuftouio(mrep, uiop, siz, dpos) 797 struct mbuf **mrep; 798 struct uio *uiop; 799 int siz; 800 caddr_t *dpos; 801 { 802 char *mbufcp, *uiocp; 803 int xfer, left, len; 804 struct mbuf *mp; 805 long uiosiz, rem; 806 int error = 0; 807 808 mp = *mrep; 809 mbufcp = *dpos; 810 len = mtod(mp, caddr_t)+mp->m_len-mbufcp; 811 rem = nfsm_rndup(siz)-siz; 812 while (siz > 0) { 813 if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) 814 return (EFBIG); 815 left = uiop->uio_iov->iov_len; 816 uiocp = uiop->uio_iov->iov_base; 817 if (left > siz) 818 left = siz; 819 uiosiz = left; 820 while (left > 0) { 821 while (len == 0) { 822 mp = mp->m_next; 823 if (mp == NULL) 824 return (EBADRPC); 825 mbufcp = mtod(mp, caddr_t); 826 len = mp->m_len; 827 } 828 xfer = (left > len) ? len : left; 829 error = copyout_vmspace(uiop->uio_vmspace, mbufcp, 830 uiocp, xfer); 831 if (error) { 832 return error; 833 } 834 left -= xfer; 835 len -= xfer; 836 mbufcp += xfer; 837 uiocp += xfer; 838 uiop->uio_offset += xfer; 839 uiop->uio_resid -= xfer; 840 } 841 if (uiop->uio_iov->iov_len <= siz) { 842 uiop->uio_iovcnt--; 843 uiop->uio_iov++; 844 } else { 845 uiop->uio_iov->iov_base = 846 (caddr_t)uiop->uio_iov->iov_base + uiosiz; 847 uiop->uio_iov->iov_len -= uiosiz; 848 } 849 siz -= uiosiz; 850 } 851 *dpos = mbufcp; 852 *mrep = mp; 853 if (rem > 0) { 854 if (len < rem) 855 error = nfs_adv(mrep, dpos, rem, len); 856 else 857 *dpos += rem; 858 } 859 return (error); 860 } 861 862 /* 863 * copies a uio scatter/gather list to an mbuf chain. 864 * NOTE: can ony handle iovcnt == 1 865 */ 866 int 867 nfsm_uiotombuf(uiop, mq, siz, bpos) 868 struct uio *uiop; 869 struct mbuf **mq; 870 int siz; 871 caddr_t *bpos; 872 { 873 char *uiocp; 874 struct mbuf *mp, *mp2; 875 int xfer, left, mlen; 876 int uiosiz, clflg, rem; 877 char *cp; 878 int error; 879 880 #ifdef DIAGNOSTIC 881 if (uiop->uio_iovcnt != 1) 882 panic("nfsm_uiotombuf: iovcnt != 1"); 883 #endif 884 885 if (siz > MLEN) /* or should it >= MCLBYTES ?? */ 886 clflg = 1; 887 else 888 clflg = 0; 889 rem = nfsm_rndup(siz)-siz; 890 mp = mp2 = *mq; 891 while (siz > 0) { 892 left = uiop->uio_iov->iov_len; 893 uiocp = uiop->uio_iov->iov_base; 894 if (left > siz) 895 left = siz; 896 uiosiz = left; 897 while (left > 0) { 898 mlen = M_TRAILINGSPACE(mp); 899 if (mlen == 0) { 900 mp = m_get(M_WAIT, MT_DATA); 901 MCLAIM(mp, &nfs_mowner); 902 if (clflg) 903 m_clget(mp, M_WAIT); 904 mp->m_len = 0; 905 mp2->m_next = mp; 906 mp2 = mp; 907 mlen = M_TRAILINGSPACE(mp); 908 } 909 xfer = (left > mlen) ? mlen : left; 910 cp = mtod(mp, caddr_t) + mp->m_len; 911 error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp, 912 xfer); 913 if (error) { 914 /* XXX */ 915 } 916 mp->m_len += xfer; 917 left -= xfer; 918 uiocp += xfer; 919 uiop->uio_offset += xfer; 920 uiop->uio_resid -= xfer; 921 } 922 uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base + 923 uiosiz; 924 uiop->uio_iov->iov_len -= uiosiz; 925 siz -= uiosiz; 926 } 927 if (rem > 0) { 928 if (rem > M_TRAILINGSPACE(mp)) { 929 mp = m_get(M_WAIT, MT_DATA); 930 MCLAIM(mp, &nfs_mowner); 931 mp->m_len = 0; 932 mp2->m_next = mp; 933 } 934 cp = mtod(mp, caddr_t) + mp->m_len; 935 for (left = 0; left < rem; left++) 936 *cp++ = '\0'; 937 mp->m_len += rem; 938 *bpos = cp; 939 } else 940 *bpos = mtod(mp, caddr_t)+mp->m_len; 941 *mq = mp; 942 return (0); 943 } 944 945 /* 946 * Get at least "siz" bytes of correctly aligned data. 947 * When called the mbuf pointers are not necessarily correct, 948 * dsosp points to what ought to be in m_data and left contains 949 * what ought to be in m_len. 950 * This is used by the macros nfsm_dissect and nfsm_dissecton for tough 951 * cases. (The macros use the vars. dpos and dpos2) 952 */ 953 int 954 nfsm_disct(mdp, dposp, siz, left, cp2) 955 struct mbuf **mdp; 956 caddr_t *dposp; 957 int siz; 958 int left; 959 caddr_t *cp2; 960 { 961 struct mbuf *m1, *m2; 962 struct mbuf *havebuf = NULL; 963 caddr_t src = *dposp; 964 caddr_t dst; 965 int len; 966 967 #ifdef DEBUG 968 if (left < 0) 969 panic("nfsm_disct: left < 0"); 970 #endif 971 m1 = *mdp; 972 /* 973 * Skip through the mbuf chain looking for an mbuf with 974 * some data. If the first mbuf found has enough data 975 * and it is correctly aligned return it. 976 */ 977 while (left == 0) { 978 havebuf = m1; 979 *mdp = m1 = m1->m_next; 980 if (m1 == NULL) 981 return (EBADRPC); 982 src = mtod(m1, caddr_t); 983 left = m1->m_len; 984 /* 985 * If we start a new mbuf and it is big enough 986 * and correctly aligned just return it, don't 987 * do any pull up. 988 */ 989 if (left >= siz && nfsm_aligned(src)) { 990 *cp2 = src; 991 *dposp = src + siz; 992 return (0); 993 } 994 } 995 if ((m1->m_flags & M_EXT) != 0) { 996 if (havebuf && M_TRAILINGSPACE(havebuf) >= siz && 997 nfsm_aligned(mtod(havebuf, char *) + havebuf->m_len)) { 998 /* 999 * If the first mbuf with data has external data 1000 * and there is a previous mbuf with some trailing 1001 * space, use it to move the data into. 1002 */ 1003 m2 = m1; 1004 *mdp = m1 = havebuf; 1005 *cp2 = mtod(m1, char *) + m1->m_len; 1006 } else if (havebuf) { 1007 /* 1008 * If the first mbuf has a external data 1009 * and there is no previous empty mbuf 1010 * allocate a new mbuf and move the external 1011 * data to the new mbuf. Also make the first 1012 * mbuf look empty. 1013 */ 1014 m2 = m1; 1015 *mdp = m1 = m_get(M_WAIT, MT_DATA); 1016 MCLAIM(m1, m2->m_owner); 1017 if ((m2->m_flags & M_PKTHDR) != 0) { 1018 /* XXX MOVE */ 1019 M_COPY_PKTHDR(m1, m2); 1020 m_tag_delete_chain(m2, NULL); 1021 m2->m_flags &= ~M_PKTHDR; 1022 } 1023 if (havebuf) { 1024 havebuf->m_next = m1; 1025 } 1026 m1->m_next = m2; 1027 MRESETDATA(m1); 1028 m1->m_len = 0; 1029 m2->m_data = src; 1030 m2->m_len = left; 1031 *cp2 = mtod(m1, char *); 1032 } else { 1033 struct mbuf **nextp = &m1->m_next; 1034 1035 m1->m_len -= left; 1036 do { 1037 m2 = m_get(M_WAIT, MT_DATA); 1038 MCLAIM(m2, m1->m_owner); 1039 if (left >= MINCLSIZE) { 1040 MCLGET(m2, M_WAIT); 1041 } 1042 m2->m_next = *nextp; 1043 *nextp = m2; 1044 nextp = &m2->m_next; 1045 len = (m2->m_flags & M_EXT) != 0 ? 1046 MCLBYTES : MLEN; 1047 if (len > left) { 1048 len = left; 1049 } 1050 memcpy(mtod(m2, char *), src, len); 1051 m2->m_len = len; 1052 src += len; 1053 left -= len; 1054 } while (left > 0); 1055 *mdp = m1 = m1->m_next; 1056 m2 = m1->m_next; 1057 *cp2 = mtod(m1, char *); 1058 } 1059 } else { 1060 /* 1061 * If the first mbuf has no external data 1062 * move the data to the front of the mbuf. 1063 */ 1064 MRESETDATA(m1); 1065 dst = mtod(m1, char *); 1066 if (dst != src) { 1067 memmove(dst, src, left); 1068 } 1069 m1->m_len = left; 1070 m2 = m1->m_next; 1071 *cp2 = m1->m_data; 1072 } 1073 *dposp = *cp2 + siz; 1074 /* 1075 * Loop through mbufs pulling data up into first mbuf until 1076 * the first mbuf is full or there is no more data to 1077 * pullup. 1078 */ 1079 dst = mtod(m1, char *) + m1->m_len; 1080 while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) { 1081 if ((len = min(len, m2->m_len)) != 0) { 1082 memcpy(dst, mtod(m2, char *), len); 1083 } 1084 m1->m_len += len; 1085 dst += len; 1086 m2->m_data += len; 1087 m2->m_len -= len; 1088 m2 = m2->m_next; 1089 } 1090 if (m1->m_len < siz) 1091 return (EBADRPC); 1092 return (0); 1093 } 1094 1095 /* 1096 * Advance the position in the mbuf chain. 1097 */ 1098 int 1099 nfs_adv(mdp, dposp, offs, left) 1100 struct mbuf **mdp; 1101 caddr_t *dposp; 1102 int offs; 1103 int left; 1104 { 1105 struct mbuf *m; 1106 int s; 1107 1108 m = *mdp; 1109 s = left; 1110 while (s < offs) { 1111 offs -= s; 1112 m = m->m_next; 1113 if (m == NULL) 1114 return (EBADRPC); 1115 s = m->m_len; 1116 } 1117 *mdp = m; 1118 *dposp = mtod(m, caddr_t)+offs; 1119 return (0); 1120 } 1121 1122 /* 1123 * Copy a string into mbufs for the hard cases... 1124 */ 1125 int 1126 nfsm_strtmbuf(mb, bpos, cp, siz) 1127 struct mbuf **mb; 1128 char **bpos; 1129 const char *cp; 1130 long siz; 1131 { 1132 struct mbuf *m1 = NULL, *m2; 1133 long left, xfer, len, tlen; 1134 u_int32_t *tl; 1135 int putsize; 1136 1137 putsize = 1; 1138 m2 = *mb; 1139 left = M_TRAILINGSPACE(m2); 1140 if (left > 0) { 1141 tl = ((u_int32_t *)(*bpos)); 1142 *tl++ = txdr_unsigned(siz); 1143 putsize = 0; 1144 left -= NFSX_UNSIGNED; 1145 m2->m_len += NFSX_UNSIGNED; 1146 if (left > 0) { 1147 memcpy((caddr_t) tl, cp, left); 1148 siz -= left; 1149 cp += left; 1150 m2->m_len += left; 1151 left = 0; 1152 } 1153 } 1154 /* Loop around adding mbufs */ 1155 while (siz > 0) { 1156 m1 = m_get(M_WAIT, MT_DATA); 1157 MCLAIM(m1, &nfs_mowner); 1158 if (siz > MLEN) 1159 m_clget(m1, M_WAIT); 1160 m1->m_len = NFSMSIZ(m1); 1161 m2->m_next = m1; 1162 m2 = m1; 1163 tl = mtod(m1, u_int32_t *); 1164 tlen = 0; 1165 if (putsize) { 1166 *tl++ = txdr_unsigned(siz); 1167 m1->m_len -= NFSX_UNSIGNED; 1168 tlen = NFSX_UNSIGNED; 1169 putsize = 0; 1170 } 1171 if (siz < m1->m_len) { 1172 len = nfsm_rndup(siz); 1173 xfer = siz; 1174 if (xfer < len) 1175 *(tl+(xfer>>2)) = 0; 1176 } else { 1177 xfer = len = m1->m_len; 1178 } 1179 memcpy((caddr_t) tl, cp, xfer); 1180 m1->m_len = len+tlen; 1181 siz -= xfer; 1182 cp += xfer; 1183 } 1184 *mb = m1; 1185 *bpos = mtod(m1, caddr_t)+m1->m_len; 1186 return (0); 1187 } 1188 1189 /* 1190 * Directory caching routines. They work as follows: 1191 * - a cache is maintained per VDIR nfsnode. 1192 * - for each offset cookie that is exported to userspace, and can 1193 * thus be thrown back at us as an offset to VOP_READDIR, store 1194 * information in the cache. 1195 * - cached are: 1196 * - cookie itself 1197 * - blocknumber (essentially just a search key in the buffer cache) 1198 * - entry number in block. 1199 * - offset cookie of block in which this entry is stored 1200 * - 32 bit cookie if NFSMNT_XLATECOOKIE is used. 1201 * - entries are looked up in a hash table 1202 * - also maintained is an LRU list of entries, used to determine 1203 * which ones to delete if the cache grows too large. 1204 * - if 32 <-> 64 translation mode is requested for a filesystem, 1205 * the cache also functions as a translation table 1206 * - in the translation case, invalidating the cache does not mean 1207 * flushing it, but just marking entries as invalid, except for 1208 * the <64bit cookie, 32bitcookie> pair which is still valid, to 1209 * still be able to use the cache as a translation table. 1210 * - 32 bit cookies are uniquely created by combining the hash table 1211 * entry value, and one generation count per hash table entry, 1212 * incremented each time an entry is appended to the chain. 1213 * - the cache is invalidated each time a direcory is modified 1214 * - sanity checks are also done; if an entry in a block turns 1215 * out not to have a matching cookie, the cache is invalidated 1216 * and a new block starting from the wanted offset is fetched from 1217 * the server. 1218 * - directory entries as read from the server are extended to contain 1219 * the 64bit and, optionally, the 32bit cookies, for sanity checking 1220 * the cache and exporting them to userspace through the cookie 1221 * argument to VOP_READDIR. 1222 */ 1223 1224 u_long 1225 nfs_dirhash(off) 1226 off_t off; 1227 { 1228 int i; 1229 char *cp = (char *)&off; 1230 u_long sum = 0L; 1231 1232 for (i = 0 ; i < sizeof (off); i++) 1233 sum += *cp++; 1234 1235 return sum; 1236 } 1237 1238 #define _NFSDC_MTX(np) (&NFSTOV(np)->v_interlock) 1239 #define NFSDC_LOCK(np) simple_lock(_NFSDC_MTX(np)) 1240 #define NFSDC_UNLOCK(np) simple_unlock(_NFSDC_MTX(np)) 1241 #define NFSDC_ASSERT_LOCKED(np) LOCK_ASSERT(simple_lock_held(_NFSDC_MTX(np))) 1242 1243 void 1244 nfs_initdircache(vp) 1245 struct vnode *vp; 1246 { 1247 struct nfsnode *np = VTONFS(vp); 1248 struct nfsdirhashhead *dircache; 1249 1250 dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF, 1251 M_WAITOK, &nfsdirhashmask); 1252 1253 NFSDC_LOCK(np); 1254 if (np->n_dircache == NULL) { 1255 np->n_dircachesize = 0; 1256 np->n_dircache = dircache; 1257 dircache = NULL; 1258 TAILQ_INIT(&np->n_dirchain); 1259 } 1260 NFSDC_UNLOCK(np); 1261 if (dircache) 1262 hashdone(dircache, M_NFSDIROFF); 1263 } 1264 1265 void 1266 nfs_initdirxlatecookie(vp) 1267 struct vnode *vp; 1268 { 1269 struct nfsnode *np = VTONFS(vp); 1270 unsigned *dirgens; 1271 1272 KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE); 1273 1274 dirgens = malloc(NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF, 1275 M_WAITOK|M_ZERO); 1276 NFSDC_LOCK(np); 1277 if (np->n_dirgens == NULL) { 1278 np->n_dirgens = dirgens; 1279 dirgens = NULL; 1280 } 1281 NFSDC_UNLOCK(np); 1282 if (dirgens) 1283 free(dirgens, M_NFSDIROFF); 1284 } 1285 1286 static const struct nfsdircache dzero; 1287 1288 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *)); 1289 static void nfs_putdircache_unlocked __P((struct nfsnode *, 1290 struct nfsdircache *)); 1291 1292 static void 1293 nfs_unlinkdircache(np, ndp) 1294 struct nfsnode *np; 1295 struct nfsdircache *ndp; 1296 { 1297 1298 NFSDC_ASSERT_LOCKED(np); 1299 KASSERT(ndp != &dzero); 1300 1301 if (LIST_NEXT(ndp, dc_hash) == (void *)-1) 1302 return; 1303 1304 TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain); 1305 LIST_REMOVE(ndp, dc_hash); 1306 LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */ 1307 1308 nfs_putdircache_unlocked(np, ndp); 1309 } 1310 1311 void 1312 nfs_putdircache(np, ndp) 1313 struct nfsnode *np; 1314 struct nfsdircache *ndp; 1315 { 1316 int ref; 1317 1318 if (ndp == &dzero) 1319 return; 1320 1321 KASSERT(ndp->dc_refcnt > 0); 1322 NFSDC_LOCK(np); 1323 ref = --ndp->dc_refcnt; 1324 NFSDC_UNLOCK(np); 1325 1326 if (ref == 0) 1327 free(ndp, M_NFSDIROFF); 1328 } 1329 1330 static void 1331 nfs_putdircache_unlocked(np, ndp) 1332 struct nfsnode *np; 1333 struct nfsdircache *ndp; 1334 { 1335 int ref; 1336 1337 NFSDC_ASSERT_LOCKED(np); 1338 1339 if (ndp == &dzero) 1340 return; 1341 1342 KASSERT(ndp->dc_refcnt > 0); 1343 ref = --ndp->dc_refcnt; 1344 if (ref == 0) 1345 free(ndp, M_NFSDIROFF); 1346 } 1347 1348 struct nfsdircache * 1349 nfs_searchdircache(vp, off, do32, hashent) 1350 struct vnode *vp; 1351 off_t off; 1352 int do32; 1353 int *hashent; 1354 { 1355 struct nfsdirhashhead *ndhp; 1356 struct nfsdircache *ndp = NULL; 1357 struct nfsnode *np = VTONFS(vp); 1358 unsigned ent; 1359 1360 /* 1361 * Zero is always a valid cookie. 1362 */ 1363 if (off == 0) 1364 /* XXXUNCONST */ 1365 return (struct nfsdircache *)__UNCONST(&dzero); 1366 1367 if (!np->n_dircache) 1368 return NULL; 1369 1370 /* 1371 * We use a 32bit cookie as search key, directly reconstruct 1372 * the hashentry. Else use the hashfunction. 1373 */ 1374 if (do32) { 1375 ent = (u_int32_t)off >> 24; 1376 if (ent >= NFS_DIRHASHSIZ) 1377 return NULL; 1378 ndhp = &np->n_dircache[ent]; 1379 } else { 1380 ndhp = NFSDIRHASH(np, off); 1381 } 1382 1383 if (hashent) 1384 *hashent = (int)(ndhp - np->n_dircache); 1385 1386 NFSDC_LOCK(np); 1387 if (do32) { 1388 LIST_FOREACH(ndp, ndhp, dc_hash) { 1389 if (ndp->dc_cookie32 == (u_int32_t)off) { 1390 /* 1391 * An invalidated entry will become the 1392 * start of a new block fetched from 1393 * the server. 1394 */ 1395 if (ndp->dc_flags & NFSDC_INVALID) { 1396 ndp->dc_blkcookie = ndp->dc_cookie; 1397 ndp->dc_entry = 0; 1398 ndp->dc_flags &= ~NFSDC_INVALID; 1399 } 1400 break; 1401 } 1402 } 1403 } else { 1404 LIST_FOREACH(ndp, ndhp, dc_hash) { 1405 if (ndp->dc_cookie == off) 1406 break; 1407 } 1408 } 1409 if (ndp != NULL) 1410 ndp->dc_refcnt++; 1411 NFSDC_UNLOCK(np); 1412 return ndp; 1413 } 1414 1415 1416 struct nfsdircache * 1417 nfs_enterdircache(vp, off, blkoff, en, blkno) 1418 struct vnode *vp; 1419 off_t off, blkoff; 1420 int en; 1421 daddr_t blkno; 1422 { 1423 struct nfsnode *np = VTONFS(vp); 1424 struct nfsdirhashhead *ndhp; 1425 struct nfsdircache *ndp = NULL; 1426 struct nfsdircache *newndp = NULL; 1427 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1428 int hashent = 0, gen, overwrite; /* XXX: GCC */ 1429 1430 /* 1431 * XXX refuse entries for offset 0. amd(8) erroneously sets 1432 * cookie 0 for the '.' entry, making this necessary. This 1433 * isn't so bad, as 0 is a special case anyway. 1434 */ 1435 if (off == 0) 1436 /* XXXUNCONST */ 1437 return (struct nfsdircache *)__UNCONST(&dzero); 1438 1439 if (!np->n_dircache) 1440 /* 1441 * XXX would like to do this in nfs_nget but vtype 1442 * isn't known at that time. 1443 */ 1444 nfs_initdircache(vp); 1445 1446 if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens) 1447 nfs_initdirxlatecookie(vp); 1448 1449 retry: 1450 ndp = nfs_searchdircache(vp, off, 0, &hashent); 1451 1452 NFSDC_LOCK(np); 1453 if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) { 1454 /* 1455 * Overwriting an old entry. Check if it's the same. 1456 * If so, just return. If not, remove the old entry. 1457 */ 1458 if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en) 1459 goto done; 1460 nfs_unlinkdircache(np, ndp); 1461 nfs_putdircache_unlocked(np, ndp); 1462 ndp = NULL; 1463 } 1464 1465 ndhp = &np->n_dircache[hashent]; 1466 1467 if (!ndp) { 1468 if (newndp == NULL) { 1469 NFSDC_UNLOCK(np); 1470 newndp = malloc(sizeof(*ndp), M_NFSDIROFF, M_WAITOK); 1471 newndp->dc_refcnt = 1; 1472 LIST_NEXT(newndp, dc_hash) = (void *)-1; 1473 goto retry; 1474 } 1475 ndp = newndp; 1476 newndp = NULL; 1477 overwrite = 0; 1478 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 1479 /* 1480 * We're allocating a new entry, so bump the 1481 * generation number. 1482 */ 1483 KASSERT(np->n_dirgens); 1484 gen = ++np->n_dirgens[hashent]; 1485 if (gen == 0) { 1486 np->n_dirgens[hashent]++; 1487 gen++; 1488 } 1489 ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff); 1490 } 1491 } else 1492 overwrite = 1; 1493 1494 ndp->dc_cookie = off; 1495 ndp->dc_blkcookie = blkoff; 1496 ndp->dc_entry = en; 1497 ndp->dc_flags = 0; 1498 1499 if (overwrite) 1500 goto done; 1501 1502 /* 1503 * If the maximum directory cookie cache size has been reached 1504 * for this node, take one off the front. The idea is that 1505 * directories are typically read front-to-back once, so that 1506 * the oldest entries can be thrown away without much performance 1507 * loss. 1508 */ 1509 if (np->n_dircachesize == NFS_MAXDIRCACHE) { 1510 nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain)); 1511 } else 1512 np->n_dircachesize++; 1513 1514 KASSERT(ndp->dc_refcnt == 1); 1515 LIST_INSERT_HEAD(ndhp, ndp, dc_hash); 1516 TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain); 1517 ndp->dc_refcnt++; 1518 done: 1519 KASSERT(ndp->dc_refcnt > 0); 1520 NFSDC_UNLOCK(np); 1521 if (newndp) 1522 nfs_putdircache(np, newndp); 1523 return ndp; 1524 } 1525 1526 void 1527 nfs_invaldircache(vp, flags) 1528 struct vnode *vp; 1529 int flags; 1530 { 1531 struct nfsnode *np = VTONFS(vp); 1532 struct nfsdircache *ndp = NULL; 1533 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1534 const boolean_t forcefree = flags & NFS_INVALDIRCACHE_FORCE; 1535 1536 #ifdef DIAGNOSTIC 1537 if (vp->v_type != VDIR) 1538 panic("nfs: invaldircache: not dir"); 1539 #endif 1540 1541 if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0) 1542 np->n_flag &= ~NEOFVALID; 1543 1544 if (!np->n_dircache) 1545 return; 1546 1547 NFSDC_LOCK(np); 1548 if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) { 1549 while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) { 1550 KASSERT(!forcefree || ndp->dc_refcnt == 1); 1551 nfs_unlinkdircache(np, ndp); 1552 } 1553 np->n_dircachesize = 0; 1554 if (forcefree && np->n_dirgens) { 1555 FREE(np->n_dirgens, M_NFSDIROFF); 1556 np->n_dirgens = NULL; 1557 } 1558 } else { 1559 TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain) 1560 ndp->dc_flags |= NFSDC_INVALID; 1561 } 1562 1563 NFSDC_UNLOCK(np); 1564 } 1565 1566 /* 1567 * Called once before VFS init to initialize shared and 1568 * server-specific data structures. 1569 */ 1570 static int 1571 nfs_init0(void) 1572 { 1573 nfsrtt.pos = 0; 1574 rpc_vers = txdr_unsigned(RPC_VER2); 1575 rpc_call = txdr_unsigned(RPC_CALL); 1576 rpc_reply = txdr_unsigned(RPC_REPLY); 1577 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED); 1578 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED); 1579 rpc_mismatch = txdr_unsigned(RPC_MISMATCH); 1580 rpc_autherr = txdr_unsigned(RPC_AUTHERR); 1581 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX); 1582 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4); 1583 nfs_prog = txdr_unsigned(NFS_PROG); 1584 nqnfs_prog = txdr_unsigned(NQNFS_PROG); 1585 nfs_true = txdr_unsigned(TRUE); 1586 nfs_false = txdr_unsigned(FALSE); 1587 nfs_xdrneg1 = txdr_unsigned(-1); 1588 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; 1589 if (nfs_ticks < 1) 1590 nfs_ticks = 1; 1591 #ifdef NFSSERVER 1592 nfsrv_init(0); /* Init server data structures */ 1593 nfsrv_initcache(); /* Init the server request cache */ 1594 #endif /* NFSSERVER */ 1595 1596 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) 1597 nfsdreq_init(); 1598 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */ 1599 1600 #if defined(NFSSERVER) || !defined(NFS_V2_ONLY) 1601 /* 1602 * Initialize the nqnfs data structures. 1603 */ 1604 if (nqnfsstarttime == 0) { 1605 nqnfsstarttime = boottime.tv_sec + nqsrv_maxlease 1606 + nqsrv_clockskew + nqsrv_writeslack; 1607 NQLOADNOVRAM(nqnfsstarttime); 1608 CIRCLEQ_INIT(&nqtimerhead); 1609 nqfhhashtbl = hashinit(NQLCHSZ, HASH_LIST, M_NQLEASE, 1610 M_WAITOK, &nqfhhash); 1611 } 1612 #endif 1613 1614 exithook_establish(nfs_exit, NULL); 1615 1616 /* 1617 * Initialize reply list and start timer 1618 */ 1619 TAILQ_INIT(&nfs_reqq); 1620 nfs_timer(NULL); 1621 MOWNER_ATTACH(&nfs_mowner); 1622 1623 #ifdef NFS 1624 /* Initialize the kqueue structures */ 1625 nfs_kqinit(); 1626 /* Initialize the iod structures */ 1627 nfs_iodinit(); 1628 #endif 1629 return 0; 1630 } 1631 1632 void 1633 nfs_init(void) 1634 { 1635 static ONCE_DECL(nfs_init_once); 1636 1637 RUN_ONCE(&nfs_init_once, nfs_init0); 1638 } 1639 1640 #ifdef NFS 1641 /* 1642 * Called once at VFS init to initialize client-specific data structures. 1643 */ 1644 void 1645 nfs_vfs_init() 1646 { 1647 /* Initialize NFS server / client shared data. */ 1648 nfs_init(); 1649 1650 nfs_nhinit(); /* Init the nfsnode table */ 1651 nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4); 1652 } 1653 1654 void 1655 nfs_vfs_reinit() 1656 { 1657 nfs_nhreinit(); 1658 } 1659 1660 void 1661 nfs_vfs_done() 1662 { 1663 nfs_nhdone(); 1664 } 1665 1666 /* 1667 * Attribute cache routines. 1668 * nfs_loadattrcache() - loads or updates the cache contents from attributes 1669 * that are on the mbuf list 1670 * nfs_getattrcache() - returns valid attributes if found in cache, returns 1671 * error otherwise 1672 */ 1673 1674 /* 1675 * Load the attribute cache (that lives in the nfsnode entry) with 1676 * the values on the mbuf list and 1677 * Iff vap not NULL 1678 * copy the attributes to *vaper 1679 */ 1680 int 1681 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags) 1682 struct vnode **vpp; 1683 struct mbuf **mdp; 1684 caddr_t *dposp; 1685 struct vattr *vaper; 1686 int flags; 1687 { 1688 int32_t t1; 1689 caddr_t cp2; 1690 int error = 0; 1691 struct mbuf *md; 1692 int v3 = NFS_ISV3(*vpp); 1693 1694 md = *mdp; 1695 t1 = (mtod(md, caddr_t) + md->m_len) - *dposp; 1696 error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2); 1697 if (error) 1698 return (error); 1699 return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags); 1700 } 1701 1702 int 1703 nfs_loadattrcache(vpp, fp, vaper, flags) 1704 struct vnode **vpp; 1705 struct nfs_fattr *fp; 1706 struct vattr *vaper; 1707 int flags; 1708 { 1709 struct vnode *vp = *vpp; 1710 struct vattr *vap; 1711 int v3 = NFS_ISV3(vp); 1712 enum vtype vtyp; 1713 u_short vmode; 1714 struct timespec mtime; 1715 struct timespec ctime; 1716 struct vnode *nvp; 1717 int32_t rdev; 1718 struct nfsnode *np; 1719 extern int (**spec_nfsv2nodeop_p) __P((void *)); 1720 uid_t uid; 1721 gid_t gid; 1722 1723 if (v3) { 1724 vtyp = nfsv3tov_type(fp->fa_type); 1725 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1726 rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1), 1727 fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2)); 1728 fxdr_nfsv3time(&fp->fa3_mtime, &mtime); 1729 fxdr_nfsv3time(&fp->fa3_ctime, &ctime); 1730 } else { 1731 vtyp = nfsv2tov_type(fp->fa_type); 1732 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1733 if (vtyp == VNON || vtyp == VREG) 1734 vtyp = IFTOVT(vmode); 1735 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); 1736 fxdr_nfsv2time(&fp->fa2_mtime, &mtime); 1737 ctime.tv_sec = fxdr_unsigned(u_int32_t, 1738 fp->fa2_ctime.nfsv2_sec); 1739 ctime.tv_nsec = 0; 1740 1741 /* 1742 * Really ugly NFSv2 kludge. 1743 */ 1744 if (vtyp == VCHR && rdev == 0xffffffff) 1745 vtyp = VFIFO; 1746 } 1747 1748 vmode &= ALLPERMS; 1749 1750 /* 1751 * If v_type == VNON it is a new node, so fill in the v_type, 1752 * n_mtime fields. Check to see if it represents a special 1753 * device, and if so, check for a possible alias. Once the 1754 * correct vnode has been obtained, fill in the rest of the 1755 * information. 1756 */ 1757 np = VTONFS(vp); 1758 if (vp->v_type == VNON) { 1759 vp->v_type = vtyp; 1760 if (vp->v_type == VFIFO) { 1761 extern int (**fifo_nfsv2nodeop_p) __P((void *)); 1762 vp->v_op = fifo_nfsv2nodeop_p; 1763 } else if (vp->v_type == VREG) { 1764 lockinit(&np->n_commitlock, PINOD, "nfsclock", 0, 0); 1765 } else if (vp->v_type == VCHR || vp->v_type == VBLK) { 1766 vp->v_op = spec_nfsv2nodeop_p; 1767 nvp = checkalias(vp, (dev_t)rdev, vp->v_mount); 1768 if (nvp) { 1769 /* 1770 * Discard unneeded vnode, but save its nfsnode. 1771 * Since the nfsnode does not have a lock, its 1772 * vnode lock has to be carried over. 1773 */ 1774 /* 1775 * XXX is the old node sure to be locked here? 1776 */ 1777 KASSERT(lockstatus(&vp->v_lock) == 1778 LK_EXCLUSIVE); 1779 nvp->v_data = vp->v_data; 1780 vp->v_data = NULL; 1781 VOP_UNLOCK(vp, 0); 1782 vp->v_op = spec_vnodeop_p; 1783 vrele(vp); 1784 vgone(vp); 1785 lockmgr(&nvp->v_lock, LK_EXCLUSIVE, 1786 &nvp->v_interlock); 1787 /* 1788 * Reinitialize aliased node. 1789 */ 1790 np->n_vnode = nvp; 1791 *vpp = vp = nvp; 1792 } 1793 } 1794 np->n_mtime = mtime; 1795 } 1796 uid = fxdr_unsigned(uid_t, fp->fa_uid); 1797 gid = fxdr_unsigned(gid_t, fp->fa_gid); 1798 vap = np->n_vattr; 1799 1800 /* 1801 * Invalidate access cache if uid, gid, mode or ctime changed. 1802 */ 1803 if (np->n_accstamp != -1 && 1804 (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode 1805 || timespeccmp(&ctime, &vap->va_ctime, !=))) 1806 np->n_accstamp = -1; 1807 1808 vap->va_type = vtyp; 1809 vap->va_mode = vmode; 1810 vap->va_rdev = (dev_t)rdev; 1811 vap->va_mtime = mtime; 1812 vap->va_ctime = ctime; 1813 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 1814 switch (vtyp) { 1815 case VDIR: 1816 vap->va_blocksize = NFS_DIRFRAGSIZ; 1817 break; 1818 case VBLK: 1819 vap->va_blocksize = BLKDEV_IOSIZE; 1820 break; 1821 case VCHR: 1822 vap->va_blocksize = MAXBSIZE; 1823 break; 1824 default: 1825 vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize : 1826 fxdr_unsigned(int32_t, fp->fa2_blocksize); 1827 break; 1828 } 1829 if (v3) { 1830 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1831 vap->va_uid = uid; 1832 vap->va_gid = gid; 1833 vap->va_size = fxdr_hyper(&fp->fa3_size); 1834 vap->va_bytes = fxdr_hyper(&fp->fa3_used); 1835 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid); 1836 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); 1837 vap->va_flags = 0; 1838 vap->va_filerev = 0; 1839 } else { 1840 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1841 vap->va_uid = uid; 1842 vap->va_gid = gid; 1843 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); 1844 vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks) 1845 * NFS_FABLKSIZE; 1846 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); 1847 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); 1848 vap->va_flags = 0; 1849 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec); 1850 vap->va_filerev = 0; 1851 } 1852 if (vap->va_size != np->n_size) { 1853 if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) { 1854 vap->va_size = np->n_size; 1855 } else { 1856 np->n_size = vap->va_size; 1857 if (vap->va_type == VREG) { 1858 /* 1859 * we can't free pages if NAC_NOTRUNC because 1860 * the pages can be owned by ourselves. 1861 */ 1862 if (flags & NAC_NOTRUNC) { 1863 np->n_flag |= NTRUNCDELAYED; 1864 } else { 1865 simple_lock(&vp->v_interlock); 1866 (void)VOP_PUTPAGES(vp, 0, 1867 0, PGO_SYNCIO | PGO_CLEANIT | 1868 PGO_FREE | PGO_ALLPAGES); 1869 uvm_vnp_setsize(vp, np->n_size); 1870 } 1871 } 1872 } 1873 } 1874 np->n_attrstamp = time_second; 1875 if (vaper != NULL) { 1876 memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(*vap)); 1877 if (np->n_flag & NCHG) { 1878 if (np->n_flag & NACC) 1879 vaper->va_atime = np->n_atim; 1880 if (np->n_flag & NUPD) 1881 vaper->va_mtime = np->n_mtim; 1882 } 1883 } 1884 return (0); 1885 } 1886 1887 /* 1888 * Check the time stamp 1889 * If the cache is valid, copy contents to *vap and return 0 1890 * otherwise return an error 1891 */ 1892 int 1893 nfs_getattrcache(vp, vaper) 1894 struct vnode *vp; 1895 struct vattr *vaper; 1896 { 1897 struct nfsnode *np = VTONFS(vp); 1898 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1899 struct vattr *vap; 1900 1901 if (np->n_attrstamp == 0 || 1902 (time_second - np->n_attrstamp) >= NFS_ATTRTIMEO(nmp, np)) { 1903 nfsstats.attrcache_misses++; 1904 return (ENOENT); 1905 } 1906 nfsstats.attrcache_hits++; 1907 vap = np->n_vattr; 1908 if (vap->va_size != np->n_size) { 1909 if (vap->va_type == VREG) { 1910 if (np->n_flag & NMODIFIED) { 1911 if (vap->va_size < np->n_size) 1912 vap->va_size = np->n_size; 1913 else 1914 np->n_size = vap->va_size; 1915 } else 1916 np->n_size = vap->va_size; 1917 uvm_vnp_setsize(vp, np->n_size); 1918 } else 1919 np->n_size = vap->va_size; 1920 } 1921 memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(struct vattr)); 1922 if (np->n_flag & NCHG) { 1923 if (np->n_flag & NACC) 1924 vaper->va_atime = np->n_atim; 1925 if (np->n_flag & NUPD) 1926 vaper->va_mtime = np->n_mtim; 1927 } 1928 return (0); 1929 } 1930 1931 void 1932 nfs_delayedtruncate(vp) 1933 struct vnode *vp; 1934 { 1935 struct nfsnode *np = VTONFS(vp); 1936 1937 if (np->n_flag & NTRUNCDELAYED) { 1938 np->n_flag &= ~NTRUNCDELAYED; 1939 simple_lock(&vp->v_interlock); 1940 (void)VOP_PUTPAGES(vp, 0, 1941 0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES); 1942 uvm_vnp_setsize(vp, np->n_size); 1943 } 1944 } 1945 1946 #define NFS_WCCKLUDGE_TIMEOUT (24 * 60 * 60) /* 1 day */ 1947 #define NFS_WCCKLUDGE(nmp, now) \ 1948 (((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \ 1949 ((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0) 1950 1951 /* 1952 * nfs_check_wccdata: check inaccurate wcc_data 1953 * 1954 * => return non-zero if we shouldn't trust the wcc_data. 1955 * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed". 1956 */ 1957 1958 int 1959 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime, 1960 struct timespec *mtime, boolean_t docheck) 1961 { 1962 int error = 0; 1963 1964 #if !defined(NFS_V2_ONLY) 1965 1966 if (docheck) { 1967 struct vnode *vp = NFSTOV(np); 1968 struct nfsmount *nmp; 1969 long now = time_second; 1970 #if defined(DEBUG) 1971 const char *reason = NULL; /* XXX: gcc */ 1972 #endif 1973 1974 if (timespeccmp(&np->n_vattr->va_mtime, mtime, <=)) { 1975 #if defined(DEBUG) 1976 reason = "mtime"; 1977 #endif 1978 error = EINVAL; 1979 } 1980 1981 if (vp->v_type == VDIR && 1982 timespeccmp(&np->n_vattr->va_ctime, ctime, <=)) { 1983 #if defined(DEBUG) 1984 reason = "ctime"; 1985 #endif 1986 error = EINVAL; 1987 } 1988 1989 nmp = VFSTONFS(vp->v_mount); 1990 if (error) { 1991 1992 /* 1993 * despite of the fact that we've updated the file, 1994 * timestamps of the file were not updated as we 1995 * expected. 1996 * it means that the server has incompatible 1997 * semantics of timestamps or (more likely) 1998 * the server time is not precise enough to 1999 * track each modifications. 2000 * in that case, we disable wcc processing. 2001 * 2002 * yes, strictly speaking, we should disable all 2003 * caching. it's a compromise. 2004 */ 2005 2006 simple_lock(&nmp->nm_slock); 2007 #if defined(DEBUG) 2008 if (!NFS_WCCKLUDGE(nmp, now)) { 2009 printf("%s: inaccurate wcc data (%s) detected," 2010 " disabling wcc\n", 2011 vp->v_mount->mnt_stat.f_mntfromname, 2012 reason); 2013 } 2014 #endif 2015 nmp->nm_iflag |= NFSMNT_WCCKLUDGE; 2016 nmp->nm_wcckludgetime = now; 2017 simple_unlock(&nmp->nm_slock); 2018 } else if (NFS_WCCKLUDGE(nmp, now)) { 2019 error = EPERM; /* XXX */ 2020 } else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 2021 simple_lock(&nmp->nm_slock); 2022 if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 2023 #if defined(DEBUG) 2024 printf("%s: re-enabling wcc\n", 2025 vp->v_mount->mnt_stat.f_mntfromname); 2026 #endif 2027 nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE; 2028 } 2029 simple_unlock(&nmp->nm_slock); 2030 } 2031 } 2032 2033 #endif /* !defined(NFS_V2_ONLY) */ 2034 2035 return error; 2036 } 2037 2038 /* 2039 * Heuristic to see if the server XDR encodes directory cookies or not. 2040 * it is not supposed to, but a lot of servers may do this. Also, since 2041 * most/all servers will implement V2 as well, it is expected that they 2042 * may return just 32 bits worth of cookie information, so we need to 2043 * find out in which 32 bits this information is available. We do this 2044 * to avoid trouble with emulated binaries that can't handle 64 bit 2045 * directory offsets. 2046 */ 2047 2048 void 2049 nfs_cookieheuristic(vp, flagp, l, cred) 2050 struct vnode *vp; 2051 int *flagp; 2052 struct lwp *l; 2053 kauth_cred_t cred; 2054 { 2055 struct uio auio; 2056 struct iovec aiov; 2057 caddr_t tbuf, cp; 2058 struct dirent *dp; 2059 off_t *cookies = NULL, *cop; 2060 int error, eof, nc, len; 2061 2062 MALLOC(tbuf, caddr_t, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK); 2063 2064 aiov.iov_base = tbuf; 2065 aiov.iov_len = NFS_DIRFRAGSIZ; 2066 auio.uio_iov = &aiov; 2067 auio.uio_iovcnt = 1; 2068 auio.uio_rw = UIO_READ; 2069 auio.uio_resid = NFS_DIRFRAGSIZ; 2070 auio.uio_offset = 0; 2071 UIO_SETUP_SYSSPACE(&auio); 2072 2073 error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc); 2074 2075 len = NFS_DIRFRAGSIZ - auio.uio_resid; 2076 if (error || len == 0) { 2077 FREE(tbuf, M_TEMP); 2078 if (cookies) 2079 free(cookies, M_TEMP); 2080 return; 2081 } 2082 2083 /* 2084 * Find the first valid entry and look at its offset cookie. 2085 */ 2086 2087 cp = tbuf; 2088 for (cop = cookies; len > 0; len -= dp->d_reclen) { 2089 dp = (struct dirent *)cp; 2090 if (dp->d_fileno != 0 && len >= dp->d_reclen) { 2091 if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) { 2092 *flagp |= NFSMNT_SWAPCOOKIE; 2093 nfs_invaldircache(vp, 0); 2094 nfs_vinvalbuf(vp, 0, cred, l, 1); 2095 } 2096 break; 2097 } 2098 cop++; 2099 cp += dp->d_reclen; 2100 } 2101 2102 FREE(tbuf, M_TEMP); 2103 free(cookies, M_TEMP); 2104 } 2105 #endif /* NFS */ 2106 2107 #ifdef NFSSERVER 2108 /* 2109 * Set up nameidata for a lookup() call and do it. 2110 * 2111 * If pubflag is set, this call is done for a lookup operation on the 2112 * public filehandle. In that case we allow crossing mountpoints and 2113 * absolute pathnames. However, the caller is expected to check that 2114 * the lookup result is within the public fs, and deny access if 2115 * it is not. 2116 */ 2117 int 2118 nfs_namei(ndp, fhp, len, slp, nam, mdp, dposp, retdirp, l, kerbflag, pubflag) 2119 struct nameidata *ndp; 2120 fhandle_t *fhp; 2121 uint32_t len; 2122 struct nfssvc_sock *slp; 2123 struct mbuf *nam; 2124 struct mbuf **mdp; 2125 caddr_t *dposp; 2126 struct vnode **retdirp; 2127 struct lwp *l; 2128 int kerbflag, pubflag; 2129 { 2130 int i, rem; 2131 struct mbuf *md; 2132 char *fromcp, *tocp, *cp; 2133 struct iovec aiov; 2134 struct uio auio; 2135 struct vnode *dp; 2136 int error, rdonly, linklen; 2137 struct componentname *cnp = &ndp->ni_cnd; 2138 2139 *retdirp = (struct vnode *)0; 2140 2141 if ((len + 1) > MAXPATHLEN) 2142 return (ENAMETOOLONG); 2143 if (len == 0) 2144 return (EACCES); 2145 cnp->cn_pnbuf = PNBUF_GET(); 2146 2147 /* 2148 * Copy the name from the mbuf list to ndp->ni_pnbuf 2149 * and set the various ndp fields appropriately. 2150 */ 2151 fromcp = *dposp; 2152 tocp = cnp->cn_pnbuf; 2153 md = *mdp; 2154 rem = mtod(md, caddr_t) + md->m_len - fromcp; 2155 for (i = 0; i < len; i++) { 2156 while (rem == 0) { 2157 md = md->m_next; 2158 if (md == NULL) { 2159 error = EBADRPC; 2160 goto out; 2161 } 2162 fromcp = mtod(md, caddr_t); 2163 rem = md->m_len; 2164 } 2165 if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) { 2166 error = EACCES; 2167 goto out; 2168 } 2169 *tocp++ = *fromcp++; 2170 rem--; 2171 } 2172 *tocp = '\0'; 2173 *mdp = md; 2174 *dposp = fromcp; 2175 len = nfsm_rndup(len)-len; 2176 if (len > 0) { 2177 if (rem >= len) 2178 *dposp += len; 2179 else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0) 2180 goto out; 2181 } 2182 2183 /* 2184 * Extract and set starting directory. 2185 */ 2186 error = nfsrv_fhtovp(fhp, FALSE, &dp, ndp->ni_cnd.cn_cred, slp, 2187 nam, &rdonly, kerbflag, pubflag); 2188 if (error) 2189 goto out; 2190 if (dp->v_type != VDIR) { 2191 vrele(dp); 2192 error = ENOTDIR; 2193 goto out; 2194 } 2195 2196 if (rdonly) 2197 cnp->cn_flags |= RDONLY; 2198 2199 *retdirp = dp; 2200 2201 if (pubflag) { 2202 /* 2203 * Oh joy. For WebNFS, handle those pesky '%' escapes, 2204 * and the 'native path' indicator. 2205 */ 2206 cp = PNBUF_GET(); 2207 fromcp = cnp->cn_pnbuf; 2208 tocp = cp; 2209 if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) { 2210 switch ((unsigned char)*fromcp) { 2211 case WEBNFS_NATIVE_CHAR: 2212 /* 2213 * 'Native' path for us is the same 2214 * as a path according to the NFS spec, 2215 * just skip the escape char. 2216 */ 2217 fromcp++; 2218 break; 2219 /* 2220 * More may be added in the future, range 0x80-0xff 2221 */ 2222 default: 2223 error = EIO; 2224 PNBUF_PUT(cp); 2225 goto out; 2226 } 2227 } 2228 /* 2229 * Translate the '%' escapes, URL-style. 2230 */ 2231 while (*fromcp != '\0') { 2232 if (*fromcp == WEBNFS_ESC_CHAR) { 2233 if (fromcp[1] != '\0' && fromcp[2] != '\0') { 2234 fromcp++; 2235 *tocp++ = HEXSTRTOI(fromcp); 2236 fromcp += 2; 2237 continue; 2238 } else { 2239 error = ENOENT; 2240 PNBUF_PUT(cp); 2241 goto out; 2242 } 2243 } else 2244 *tocp++ = *fromcp++; 2245 } 2246 *tocp = '\0'; 2247 PNBUF_PUT(cnp->cn_pnbuf); 2248 cnp->cn_pnbuf = cp; 2249 } 2250 2251 ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1; 2252 ndp->ni_segflg = UIO_SYSSPACE; 2253 ndp->ni_rootdir = rootvnode; 2254 2255 if (pubflag) { 2256 ndp->ni_loopcnt = 0; 2257 if (cnp->cn_pnbuf[0] == '/') 2258 dp = rootvnode; 2259 } else { 2260 cnp->cn_flags |= NOCROSSMOUNT; 2261 } 2262 2263 cnp->cn_lwp = l; 2264 VREF(dp); 2265 2266 for (;;) { 2267 cnp->cn_nameptr = cnp->cn_pnbuf; 2268 ndp->ni_startdir = dp; 2269 /* 2270 * And call lookup() to do the real work 2271 */ 2272 error = lookup(ndp); 2273 if (error) { 2274 PNBUF_PUT(cnp->cn_pnbuf); 2275 return (error); 2276 } 2277 /* 2278 * Check for encountering a symbolic link 2279 */ 2280 if ((cnp->cn_flags & ISSYMLINK) == 0) { 2281 if (cnp->cn_flags & (SAVENAME | SAVESTART)) 2282 cnp->cn_flags |= HASBUF; 2283 else 2284 PNBUF_PUT(cnp->cn_pnbuf); 2285 return (0); 2286 } else { 2287 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN)) 2288 VOP_UNLOCK(ndp->ni_dvp, 0); 2289 if (!pubflag) { 2290 error = EINVAL; 2291 break; 2292 } 2293 2294 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 2295 error = ELOOP; 2296 break; 2297 } 2298 if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) { 2299 error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred, 2300 cnp->cn_lwp); 2301 if (error != 0) 2302 break; 2303 } 2304 if (ndp->ni_pathlen > 1) 2305 cp = PNBUF_GET(); 2306 else 2307 cp = cnp->cn_pnbuf; 2308 aiov.iov_base = cp; 2309 aiov.iov_len = MAXPATHLEN; 2310 auio.uio_iov = &aiov; 2311 auio.uio_iovcnt = 1; 2312 auio.uio_offset = 0; 2313 auio.uio_rw = UIO_READ; 2314 auio.uio_resid = MAXPATHLEN; 2315 UIO_SETUP_SYSSPACE(&auio); 2316 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 2317 if (error) { 2318 badlink: 2319 if (ndp->ni_pathlen > 1) 2320 PNBUF_PUT(cp); 2321 break; 2322 } 2323 linklen = MAXPATHLEN - auio.uio_resid; 2324 if (linklen == 0) { 2325 error = ENOENT; 2326 goto badlink; 2327 } 2328 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 2329 error = ENAMETOOLONG; 2330 goto badlink; 2331 } 2332 if (ndp->ni_pathlen > 1) { 2333 memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen); 2334 PNBUF_PUT(cnp->cn_pnbuf); 2335 cnp->cn_pnbuf = cp; 2336 } else 2337 cnp->cn_pnbuf[linklen] = '\0'; 2338 ndp->ni_pathlen += linklen; 2339 vput(ndp->ni_vp); 2340 dp = ndp->ni_dvp; 2341 /* 2342 * Check if root directory should replace current directory. 2343 */ 2344 if (cnp->cn_pnbuf[0] == '/') { 2345 vrele(dp); 2346 dp = ndp->ni_rootdir; 2347 VREF(dp); 2348 } 2349 } 2350 } 2351 vrele(ndp->ni_dvp); 2352 vput(ndp->ni_vp); 2353 ndp->ni_vp = NULL; 2354 out: 2355 PNBUF_PUT(cnp->cn_pnbuf); 2356 return (error); 2357 } 2358 #endif /* NFSSERVER */ 2359 2360 /* 2361 * A fiddled version of m_adj() that ensures null fill to a 32-bit 2362 * boundary and only trims off the back end 2363 * 2364 * 1. trim off 'len' bytes as m_adj(mp, -len). 2365 * 2. add zero-padding 'nul' bytes at the end of the mbuf chain. 2366 */ 2367 void 2368 nfs_zeropad(mp, len, nul) 2369 struct mbuf *mp; 2370 int len; 2371 int nul; 2372 { 2373 struct mbuf *m; 2374 int count; 2375 2376 /* 2377 * Trim from tail. Scan the mbuf chain, 2378 * calculating its length and finding the last mbuf. 2379 * If the adjustment only affects this mbuf, then just 2380 * adjust and return. Otherwise, rescan and truncate 2381 * after the remaining size. 2382 */ 2383 count = 0; 2384 m = mp; 2385 for (;;) { 2386 count += m->m_len; 2387 if (m->m_next == NULL) 2388 break; 2389 m = m->m_next; 2390 } 2391 2392 KDASSERT(count >= len); 2393 2394 if (m->m_len >= len) { 2395 m->m_len -= len; 2396 } else { 2397 count -= len; 2398 /* 2399 * Correct length for chain is "count". 2400 * Find the mbuf with last data, adjust its length, 2401 * and toss data from remaining mbufs on chain. 2402 */ 2403 for (m = mp; m; m = m->m_next) { 2404 if (m->m_len >= count) { 2405 m->m_len = count; 2406 break; 2407 } 2408 count -= m->m_len; 2409 } 2410 KASSERT(m && m->m_next); 2411 m_freem(m->m_next); 2412 m->m_next = NULL; 2413 } 2414 2415 KDASSERT(m->m_next == NULL); 2416 2417 /* 2418 * zero-padding. 2419 */ 2420 if (nul > 0) { 2421 char *cp; 2422 int i; 2423 2424 if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) { 2425 struct mbuf *n; 2426 2427 KDASSERT(MLEN >= nul); 2428 n = m_get(M_WAIT, MT_DATA); 2429 MCLAIM(n, &nfs_mowner); 2430 n->m_len = nul; 2431 n->m_next = NULL; 2432 m->m_next = n; 2433 cp = mtod(n, caddr_t); 2434 } else { 2435 cp = mtod(m, caddr_t) + m->m_len; 2436 m->m_len += nul; 2437 } 2438 for (i = 0; i < nul; i++) 2439 *cp++ = '\0'; 2440 } 2441 return; 2442 } 2443 2444 /* 2445 * Make these functions instead of macros, so that the kernel text size 2446 * doesn't get too big... 2447 */ 2448 void 2449 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp) 2450 struct nfsrv_descript *nfsd; 2451 int before_ret; 2452 struct vattr *before_vap; 2453 int after_ret; 2454 struct vattr *after_vap; 2455 struct mbuf **mbp; 2456 char **bposp; 2457 { 2458 struct mbuf *mb = *mbp; 2459 char *bpos = *bposp; 2460 u_int32_t *tl; 2461 2462 if (before_ret) { 2463 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2464 *tl = nfs_false; 2465 } else { 2466 nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED); 2467 *tl++ = nfs_true; 2468 txdr_hyper(before_vap->va_size, tl); 2469 tl += 2; 2470 txdr_nfsv3time(&(before_vap->va_mtime), tl); 2471 tl += 2; 2472 txdr_nfsv3time(&(before_vap->va_ctime), tl); 2473 } 2474 *bposp = bpos; 2475 *mbp = mb; 2476 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp); 2477 } 2478 2479 void 2480 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp) 2481 struct nfsrv_descript *nfsd; 2482 int after_ret; 2483 struct vattr *after_vap; 2484 struct mbuf **mbp; 2485 char **bposp; 2486 { 2487 struct mbuf *mb = *mbp; 2488 char *bpos = *bposp; 2489 u_int32_t *tl; 2490 struct nfs_fattr *fp; 2491 2492 if (after_ret) { 2493 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2494 *tl = nfs_false; 2495 } else { 2496 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR); 2497 *tl++ = nfs_true; 2498 fp = (struct nfs_fattr *)tl; 2499 nfsm_srvfattr(nfsd, after_vap, fp); 2500 } 2501 *mbp = mb; 2502 *bposp = bpos; 2503 } 2504 2505 void 2506 nfsm_srvfattr(nfsd, vap, fp) 2507 struct nfsrv_descript *nfsd; 2508 struct vattr *vap; 2509 struct nfs_fattr *fp; 2510 { 2511 2512 fp->fa_nlink = txdr_unsigned(vap->va_nlink); 2513 fp->fa_uid = txdr_unsigned(vap->va_uid); 2514 fp->fa_gid = txdr_unsigned(vap->va_gid); 2515 if (nfsd->nd_flag & ND_NFSV3) { 2516 fp->fa_type = vtonfsv3_type(vap->va_type); 2517 fp->fa_mode = vtonfsv3_mode(vap->va_mode); 2518 txdr_hyper(vap->va_size, &fp->fa3_size); 2519 txdr_hyper(vap->va_bytes, &fp->fa3_used); 2520 fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev)); 2521 fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev)); 2522 fp->fa3_fsid.nfsuquad[0] = 0; 2523 fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid); 2524 txdr_hyper(vap->va_fileid, &fp->fa3_fileid); 2525 txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime); 2526 txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime); 2527 txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime); 2528 } else { 2529 fp->fa_type = vtonfsv2_type(vap->va_type); 2530 fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 2531 fp->fa2_size = txdr_unsigned(vap->va_size); 2532 fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize); 2533 if (vap->va_type == VFIFO) 2534 fp->fa2_rdev = 0xffffffff; 2535 else 2536 fp->fa2_rdev = txdr_unsigned(vap->va_rdev); 2537 fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE); 2538 fp->fa2_fsid = txdr_unsigned(vap->va_fsid); 2539 fp->fa2_fileid = txdr_unsigned(vap->va_fileid); 2540 txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime); 2541 txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime); 2542 txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime); 2543 } 2544 } 2545 2546 #ifdef NFSSERVER 2547 /* 2548 * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked) 2549 * - look up fsid in mount list (if not found ret error) 2550 * - get vp and export rights by calling VFS_FHTOVP() 2551 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 2552 * - if not lockflag unlock it with VOP_UNLOCK() 2553 */ 2554 int 2555 nfsrv_fhtovp(fhp, lockflag, vpp, cred, slp, nam, rdonlyp, kerbflag, pubflag) 2556 fhandle_t *fhp; 2557 int lockflag; 2558 struct vnode **vpp; 2559 kauth_cred_t cred; 2560 struct nfssvc_sock *slp; 2561 struct mbuf *nam; 2562 int *rdonlyp; 2563 int kerbflag; 2564 { 2565 struct mount *mp; 2566 kauth_cred_t credanon; 2567 int error, exflags; 2568 struct sockaddr_in *saddr; 2569 2570 *vpp = (struct vnode *)0; 2571 2572 if (nfs_ispublicfh(fhp)) { 2573 if (!pubflag || !nfs_pub.np_valid) 2574 return (ESTALE); 2575 fhp = &nfs_pub.np_handle; 2576 } 2577 2578 error = netexport_check(&fhp->fh_fsid, nam, &mp, &exflags, &credanon); 2579 if (error) { 2580 return error; 2581 } 2582 2583 error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp); 2584 if (error) 2585 return (error); 2586 2587 if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) { 2588 saddr = mtod(nam, struct sockaddr_in *); 2589 if ((saddr->sin_family == AF_INET) && 2590 ntohs(saddr->sin_port) >= IPPORT_RESERVED) { 2591 vput(*vpp); 2592 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2593 } 2594 #ifdef INET6 2595 if ((saddr->sin_family == AF_INET6) && 2596 ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) { 2597 vput(*vpp); 2598 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2599 } 2600 #endif 2601 } 2602 /* 2603 * Check/setup credentials. 2604 */ 2605 if (exflags & MNT_EXKERB) { 2606 if (!kerbflag) { 2607 vput(*vpp); 2608 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2609 } 2610 } else if (kerbflag) { 2611 vput(*vpp); 2612 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2613 } else if (kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 2614 NULL) == 0 || (exflags & MNT_EXPORTANON)) { 2615 kauth_cred_clone(credanon, cred); 2616 } 2617 if (exflags & MNT_EXRDONLY) 2618 *rdonlyp = 1; 2619 else 2620 *rdonlyp = 0; 2621 if (!lockflag) 2622 VOP_UNLOCK(*vpp, 0); 2623 return (0); 2624 } 2625 2626 /* 2627 * WebNFS: check if a filehandle is a public filehandle. For v3, this 2628 * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has 2629 * transformed this to all zeroes in both cases, so check for it. 2630 */ 2631 int 2632 nfs_ispublicfh(fhp) 2633 fhandle_t *fhp; 2634 { 2635 char *cp = (char *)fhp; 2636 int i; 2637 2638 for (i = 0; i < NFSX_V3FH; i++) 2639 if (*cp++ != 0) 2640 return (FALSE); 2641 return (TRUE); 2642 } 2643 #endif /* NFSSERVER */ 2644 2645 /* 2646 * This function compares two net addresses by family and returns TRUE 2647 * if they are the same host. 2648 * If there is any doubt, return FALSE. 2649 * The AF_INET family is handled as a special case so that address mbufs 2650 * don't need to be saved to store "struct in_addr", which is only 4 bytes. 2651 */ 2652 int 2653 netaddr_match(family, haddr, nam) 2654 int family; 2655 union nethostaddr *haddr; 2656 struct mbuf *nam; 2657 { 2658 struct sockaddr_in *inetaddr; 2659 2660 switch (family) { 2661 case AF_INET: 2662 inetaddr = mtod(nam, struct sockaddr_in *); 2663 if (inetaddr->sin_family == AF_INET && 2664 inetaddr->sin_addr.s_addr == haddr->had_inetaddr) 2665 return (1); 2666 break; 2667 #ifdef INET6 2668 case AF_INET6: 2669 { 2670 struct sockaddr_in6 *sin6_1, *sin6_2; 2671 2672 sin6_1 = mtod(nam, struct sockaddr_in6 *); 2673 sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *); 2674 if (sin6_1->sin6_family == AF_INET6 && 2675 IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr)) 2676 return 1; 2677 } 2678 #endif 2679 #ifdef ISO 2680 case AF_ISO: 2681 { 2682 struct sockaddr_iso *isoaddr1, *isoaddr2; 2683 2684 isoaddr1 = mtod(nam, struct sockaddr_iso *); 2685 isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *); 2686 if (isoaddr1->siso_family == AF_ISO && 2687 isoaddr1->siso_nlen > 0 && 2688 isoaddr1->siso_nlen == isoaddr2->siso_nlen && 2689 SAME_ISOADDR(isoaddr1, isoaddr2)) 2690 return (1); 2691 break; 2692 } 2693 #endif /* ISO */ 2694 default: 2695 break; 2696 }; 2697 return (0); 2698 } 2699 2700 /* 2701 * The write verifier has changed (probably due to a server reboot), so all 2702 * PG_NEEDCOMMIT pages will have to be written again. Since they are marked 2703 * as dirty or are being written out just now, all this takes is clearing 2704 * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for 2705 * the mount point. 2706 */ 2707 void 2708 nfs_clearcommit(mp) 2709 struct mount *mp; 2710 { 2711 struct vnode *vp; 2712 struct nfsnode *np; 2713 struct vm_page *pg; 2714 struct nfsmount *nmp = VFSTONFS(mp); 2715 2716 lockmgr(&nmp->nm_writeverflock, LK_EXCLUSIVE, NULL); 2717 2718 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2719 KASSERT(vp->v_mount == mp); 2720 if (vp->v_type != VREG) 2721 continue; 2722 np = VTONFS(vp); 2723 np->n_pushlo = np->n_pushhi = np->n_pushedlo = 2724 np->n_pushedhi = 0; 2725 np->n_commitflags &= 2726 ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID); 2727 simple_lock(&vp->v_uobj.vmobjlock); 2728 TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) { 2729 pg->flags &= ~PG_NEEDCOMMIT; 2730 } 2731 simple_unlock(&vp->v_uobj.vmobjlock); 2732 } 2733 simple_lock(&nmp->nm_slock); 2734 nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF; 2735 simple_unlock(&nmp->nm_slock); 2736 lockmgr(&nmp->nm_writeverflock, LK_RELEASE, NULL); 2737 } 2738 2739 void 2740 nfs_merge_commit_ranges(vp) 2741 struct vnode *vp; 2742 { 2743 struct nfsnode *np = VTONFS(vp); 2744 2745 KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID); 2746 2747 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2748 np->n_pushedlo = np->n_pushlo; 2749 np->n_pushedhi = np->n_pushhi; 2750 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2751 } else { 2752 if (np->n_pushlo < np->n_pushedlo) 2753 np->n_pushedlo = np->n_pushlo; 2754 if (np->n_pushhi > np->n_pushedhi) 2755 np->n_pushedhi = np->n_pushhi; 2756 } 2757 2758 np->n_pushlo = np->n_pushhi = 0; 2759 np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID; 2760 2761 #ifdef NFS_DEBUG_COMMIT 2762 printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2763 (unsigned)np->n_pushedhi); 2764 #endif 2765 } 2766 2767 int 2768 nfs_in_committed_range(vp, off, len) 2769 struct vnode *vp; 2770 off_t off, len; 2771 { 2772 struct nfsnode *np = VTONFS(vp); 2773 off_t lo, hi; 2774 2775 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2776 return 0; 2777 lo = off; 2778 hi = lo + len; 2779 2780 return (lo >= np->n_pushedlo && hi <= np->n_pushedhi); 2781 } 2782 2783 int 2784 nfs_in_tobecommitted_range(vp, off, len) 2785 struct vnode *vp; 2786 off_t off, len; 2787 { 2788 struct nfsnode *np = VTONFS(vp); 2789 off_t lo, hi; 2790 2791 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2792 return 0; 2793 lo = off; 2794 hi = lo + len; 2795 2796 return (lo >= np->n_pushlo && hi <= np->n_pushhi); 2797 } 2798 2799 void 2800 nfs_add_committed_range(vp, off, len) 2801 struct vnode *vp; 2802 off_t off, len; 2803 { 2804 struct nfsnode *np = VTONFS(vp); 2805 off_t lo, hi; 2806 2807 lo = off; 2808 hi = lo + len; 2809 2810 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2811 np->n_pushedlo = lo; 2812 np->n_pushedhi = hi; 2813 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2814 } else { 2815 if (hi > np->n_pushedhi) 2816 np->n_pushedhi = hi; 2817 if (lo < np->n_pushedlo) 2818 np->n_pushedlo = lo; 2819 } 2820 #ifdef NFS_DEBUG_COMMIT 2821 printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2822 (unsigned)np->n_pushedhi); 2823 #endif 2824 } 2825 2826 void 2827 nfs_del_committed_range(vp, off, len) 2828 struct vnode *vp; 2829 off_t off, len; 2830 { 2831 struct nfsnode *np = VTONFS(vp); 2832 off_t lo, hi; 2833 2834 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2835 return; 2836 2837 lo = off; 2838 hi = lo + len; 2839 2840 if (lo > np->n_pushedhi || hi < np->n_pushedlo) 2841 return; 2842 if (lo <= np->n_pushedlo) 2843 np->n_pushedlo = hi; 2844 else if (hi >= np->n_pushedhi) 2845 np->n_pushedhi = 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_pushedlo - lo) > (hi - np->n_pushedhi)) 2853 np->n_pushedhi = lo; 2854 else 2855 np->n_pushedlo = hi; 2856 } 2857 #ifdef NFS_DEBUG_COMMIT 2858 printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2859 (unsigned)np->n_pushedhi); 2860 #endif 2861 } 2862 2863 void 2864 nfs_add_tobecommitted_range(vp, off, len) 2865 struct vnode *vp; 2866 off_t off, len; 2867 { 2868 struct nfsnode *np = VTONFS(vp); 2869 off_t lo, hi; 2870 2871 lo = off; 2872 hi = lo + len; 2873 2874 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) { 2875 np->n_pushlo = lo; 2876 np->n_pushhi = hi; 2877 np->n_commitflags |= NFS_COMMIT_PUSH_VALID; 2878 } else { 2879 if (lo < np->n_pushlo) 2880 np->n_pushlo = lo; 2881 if (hi > np->n_pushhi) 2882 np->n_pushhi = hi; 2883 } 2884 #ifdef NFS_DEBUG_COMMIT 2885 printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2886 (unsigned)np->n_pushhi); 2887 #endif 2888 } 2889 2890 void 2891 nfs_del_tobecommitted_range(vp, off, len) 2892 struct vnode *vp; 2893 off_t off, len; 2894 { 2895 struct nfsnode *np = VTONFS(vp); 2896 off_t lo, hi; 2897 2898 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2899 return; 2900 2901 lo = off; 2902 hi = lo + len; 2903 2904 if (lo > np->n_pushhi || hi < np->n_pushlo) 2905 return; 2906 2907 if (lo <= np->n_pushlo) 2908 np->n_pushlo = hi; 2909 else if (hi >= np->n_pushhi) 2910 np->n_pushhi = lo; 2911 else { 2912 /* 2913 * XXX There's only one range. If the deleted range 2914 * is in the middle, pick the largest of the 2915 * contiguous ranges that it leaves. 2916 */ 2917 if ((np->n_pushlo - lo) > (hi - np->n_pushhi)) 2918 np->n_pushhi = lo; 2919 else 2920 np->n_pushlo = hi; 2921 } 2922 #ifdef NFS_DEBUG_COMMIT 2923 printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2924 (unsigned)np->n_pushhi); 2925 #endif 2926 } 2927 2928 /* 2929 * Map errnos to NFS error numbers. For Version 3 also filter out error 2930 * numbers not specified for the associated procedure. 2931 */ 2932 int 2933 nfsrv_errmap(nd, err) 2934 struct nfsrv_descript *nd; 2935 int err; 2936 { 2937 const short *defaulterrp, *errp; 2938 2939 if (nd->nd_flag & ND_NFSV3) { 2940 if (nd->nd_procnum <= NFSPROC_COMMIT) { 2941 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum]; 2942 while (*++errp) { 2943 if (*errp == err) 2944 return (err); 2945 else if (*errp > err) 2946 break; 2947 } 2948 return ((int)*defaulterrp); 2949 } else 2950 return (err & 0xffff); 2951 } 2952 if (err <= ELAST) 2953 return ((int)nfsrv_v2errmap[err - 1]); 2954 return (NFSERR_IO); 2955 } 2956 2957 u_int32_t 2958 nfs_getxid() 2959 { 2960 static u_int32_t base; 2961 static u_int32_t nfs_xid = 0; 2962 static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER; 2963 u_int32_t newxid; 2964 2965 simple_lock(&nfs_xidlock); 2966 /* 2967 * derive initial xid from system time 2968 * XXX time is invalid if root not yet mounted 2969 */ 2970 if (__predict_false(!base && (rootvp))) { 2971 struct timeval tv; 2972 2973 microtime(&tv); 2974 base = tv.tv_sec << 12; 2975 nfs_xid = base; 2976 } 2977 2978 /* 2979 * Skip zero xid if it should ever happen. 2980 */ 2981 if (__predict_false(++nfs_xid == 0)) 2982 nfs_xid++; 2983 newxid = nfs_xid; 2984 simple_unlock(&nfs_xidlock); 2985 2986 return txdr_unsigned(newxid); 2987 } 2988 2989 /* 2990 * assign a new xid for existing request. 2991 * used for NFSERR_JUKEBOX handling. 2992 */ 2993 void 2994 nfs_renewxid(struct nfsreq *req) 2995 { 2996 u_int32_t xid; 2997 int off; 2998 2999 xid = nfs_getxid(); 3000 if (req->r_nmp->nm_sotype == SOCK_STREAM) 3001 off = sizeof(u_int32_t); /* RPC record mark */ 3002 else 3003 off = 0; 3004 3005 m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid); 3006 req->r_xid = xid; 3007 } 3008
Indexes created Thu Oct 02 14:10:14 GMT 2025