nfs_subs.c revision 1.184.2.5
1 /* $NetBSD: nfs_subs.c,v 1.184.2.5 2007/07/15 13:28:06 ad 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 1.184.2.5 2007/07/15 13:28:06 ad Exp $"); 74 75 #include "fs_nfs.h" 76 #include "opt_nfs.h" 77 #include "opt_nfsserver.h" 78 #include "opt_iso.h" 79 #include "opt_inet.h" 80 81 /* 82 * These functions support the macros and help fiddle mbuf chains for 83 * the nfs op functions. They do things like create the rpc header and 84 * copy data between mbuf chains and uio lists. 85 */ 86 #include <sys/param.h> 87 #include <sys/proc.h> 88 #include <sys/systm.h> 89 #include <sys/kernel.h> 90 #include <sys/mount.h> 91 #include <sys/vnode.h> 92 #include <sys/namei.h> 93 #include <sys/mbuf.h> 94 #include <sys/socket.h> 95 #include <sys/stat.h> 96 #include <sys/malloc.h> 97 #include <sys/filedesc.h> 98 #include <sys/time.h> 99 #include <sys/dirent.h> 100 #include <sys/once.h> 101 #include <sys/kauth.h> 102 103 #include <uvm/uvm_extern.h> 104 105 #include <nfs/rpcv2.h> 106 #include <nfs/nfsproto.h> 107 #include <nfs/nfsnode.h> 108 #include <nfs/nfs.h> 109 #include <nfs/xdr_subs.h> 110 #include <nfs/nfsm_subs.h> 111 #include <nfs/nfsmount.h> 112 #include <nfs/nfsrtt.h> 113 #include <nfs/nfs_var.h> 114 115 #include <miscfs/specfs/specdev.h> 116 117 #include <netinet/in.h> 118 #ifdef ISO 119 #include <netiso/iso.h> 120 #endif 121 122 /* 123 * Data items converted to xdr at startup, since they are constant 124 * This is kinda hokey, but may save a little time doing byte swaps 125 */ 126 u_int32_t nfs_xdrneg1; 127 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr, 128 rpc_mismatch, rpc_auth_unix, rpc_msgaccepted, 129 rpc_auth_kerb; 130 u_int32_t nfs_prog, nfs_true, nfs_false; 131 132 /* And other global data */ 133 const nfstype nfsv2_type[9] = 134 { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON }; 135 const nfstype nfsv3_type[9] = 136 { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON }; 137 const enum vtype nv2tov_type[8] = 138 { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON }; 139 const enum vtype nv3tov_type[8] = 140 { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO }; 141 int nfs_ticks; 142 int nfs_commitsize; 143 144 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies"); 145 146 /* NFS client/server stats. */ 147 struct nfsstats nfsstats; 148 149 /* 150 * Mapping of old NFS Version 2 RPC numbers to generic numbers. 151 */ 152 const int nfsv3_procid[NFS_NPROCS] = { 153 NFSPROC_NULL, 154 NFSPROC_GETATTR, 155 NFSPROC_SETATTR, 156 NFSPROC_NOOP, 157 NFSPROC_LOOKUP, 158 NFSPROC_READLINK, 159 NFSPROC_READ, 160 NFSPROC_NOOP, 161 NFSPROC_WRITE, 162 NFSPROC_CREATE, 163 NFSPROC_REMOVE, 164 NFSPROC_RENAME, 165 NFSPROC_LINK, 166 NFSPROC_SYMLINK, 167 NFSPROC_MKDIR, 168 NFSPROC_RMDIR, 169 NFSPROC_READDIR, 170 NFSPROC_FSSTAT, 171 NFSPROC_NOOP, 172 NFSPROC_NOOP, 173 NFSPROC_NOOP, 174 NFSPROC_NOOP, 175 NFSPROC_NOOP 176 }; 177 178 /* 179 * and the reverse mapping from generic to Version 2 procedure numbers 180 */ 181 const int nfsv2_procid[NFS_NPROCS] = { 182 NFSV2PROC_NULL, 183 NFSV2PROC_GETATTR, 184 NFSV2PROC_SETATTR, 185 NFSV2PROC_LOOKUP, 186 NFSV2PROC_NOOP, 187 NFSV2PROC_READLINK, 188 NFSV2PROC_READ, 189 NFSV2PROC_WRITE, 190 NFSV2PROC_CREATE, 191 NFSV2PROC_MKDIR, 192 NFSV2PROC_SYMLINK, 193 NFSV2PROC_CREATE, 194 NFSV2PROC_REMOVE, 195 NFSV2PROC_RMDIR, 196 NFSV2PROC_RENAME, 197 NFSV2PROC_LINK, 198 NFSV2PROC_READDIR, 199 NFSV2PROC_NOOP, 200 NFSV2PROC_STATFS, 201 NFSV2PROC_NOOP, 202 NFSV2PROC_NOOP, 203 NFSV2PROC_NOOP, 204 NFSV2PROC_NOOP, 205 }; 206 207 /* 208 * Maps errno values to nfs error numbers. 209 * Use NFSERR_IO as the catch all for ones not specifically defined in 210 * RFC 1094. 211 */ 212 static const u_char nfsrv_v2errmap[ELAST] = { 213 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 214 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 215 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 216 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 217 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 218 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 219 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 220 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 221 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 222 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 223 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 224 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 225 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO, 226 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 227 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 228 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 229 NFSERR_IO, NFSERR_IO, 230 }; 231 232 /* 233 * Maps errno values to nfs error numbers. 234 * Although it is not obvious whether or not NFS clients really care if 235 * a returned error value is in the specified list for the procedure, the 236 * safest thing to do is filter them appropriately. For Version 2, the 237 * X/Open XNFS document is the only specification that defines error values 238 * for each RPC (The RFC simply lists all possible error values for all RPCs), 239 * so I have decided to not do this for Version 2. 240 * The first entry is the default error return and the rest are the valid 241 * errors for that RPC in increasing numeric order. 242 */ 243 static const short nfsv3err_null[] = { 244 0, 245 0, 246 }; 247 248 static const short nfsv3err_getattr[] = { 249 NFSERR_IO, 250 NFSERR_IO, 251 NFSERR_STALE, 252 NFSERR_BADHANDLE, 253 NFSERR_SERVERFAULT, 254 0, 255 }; 256 257 static const short nfsv3err_setattr[] = { 258 NFSERR_IO, 259 NFSERR_PERM, 260 NFSERR_IO, 261 NFSERR_ACCES, 262 NFSERR_INVAL, 263 NFSERR_NOSPC, 264 NFSERR_ROFS, 265 NFSERR_DQUOT, 266 NFSERR_STALE, 267 NFSERR_BADHANDLE, 268 NFSERR_NOT_SYNC, 269 NFSERR_SERVERFAULT, 270 0, 271 }; 272 273 static const short nfsv3err_lookup[] = { 274 NFSERR_IO, 275 NFSERR_NOENT, 276 NFSERR_IO, 277 NFSERR_ACCES, 278 NFSERR_NOTDIR, 279 NFSERR_NAMETOL, 280 NFSERR_STALE, 281 NFSERR_BADHANDLE, 282 NFSERR_SERVERFAULT, 283 0, 284 }; 285 286 static const short nfsv3err_access[] = { 287 NFSERR_IO, 288 NFSERR_IO, 289 NFSERR_STALE, 290 NFSERR_BADHANDLE, 291 NFSERR_SERVERFAULT, 292 0, 293 }; 294 295 static const short nfsv3err_readlink[] = { 296 NFSERR_IO, 297 NFSERR_IO, 298 NFSERR_ACCES, 299 NFSERR_INVAL, 300 NFSERR_STALE, 301 NFSERR_BADHANDLE, 302 NFSERR_NOTSUPP, 303 NFSERR_SERVERFAULT, 304 0, 305 }; 306 307 static const short nfsv3err_read[] = { 308 NFSERR_IO, 309 NFSERR_IO, 310 NFSERR_NXIO, 311 NFSERR_ACCES, 312 NFSERR_INVAL, 313 NFSERR_STALE, 314 NFSERR_BADHANDLE, 315 NFSERR_SERVERFAULT, 316 NFSERR_JUKEBOX, 317 0, 318 }; 319 320 static const short nfsv3err_write[] = { 321 NFSERR_IO, 322 NFSERR_IO, 323 NFSERR_ACCES, 324 NFSERR_INVAL, 325 NFSERR_FBIG, 326 NFSERR_NOSPC, 327 NFSERR_ROFS, 328 NFSERR_DQUOT, 329 NFSERR_STALE, 330 NFSERR_BADHANDLE, 331 NFSERR_SERVERFAULT, 332 NFSERR_JUKEBOX, 333 0, 334 }; 335 336 static const short nfsv3err_create[] = { 337 NFSERR_IO, 338 NFSERR_IO, 339 NFSERR_ACCES, 340 NFSERR_EXIST, 341 NFSERR_NOTDIR, 342 NFSERR_NOSPC, 343 NFSERR_ROFS, 344 NFSERR_NAMETOL, 345 NFSERR_DQUOT, 346 NFSERR_STALE, 347 NFSERR_BADHANDLE, 348 NFSERR_NOTSUPP, 349 NFSERR_SERVERFAULT, 350 0, 351 }; 352 353 static const short nfsv3err_mkdir[] = { 354 NFSERR_IO, 355 NFSERR_IO, 356 NFSERR_ACCES, 357 NFSERR_EXIST, 358 NFSERR_NOTDIR, 359 NFSERR_NOSPC, 360 NFSERR_ROFS, 361 NFSERR_NAMETOL, 362 NFSERR_DQUOT, 363 NFSERR_STALE, 364 NFSERR_BADHANDLE, 365 NFSERR_NOTSUPP, 366 NFSERR_SERVERFAULT, 367 0, 368 }; 369 370 static const short nfsv3err_symlink[] = { 371 NFSERR_IO, 372 NFSERR_IO, 373 NFSERR_ACCES, 374 NFSERR_EXIST, 375 NFSERR_NOTDIR, 376 NFSERR_NOSPC, 377 NFSERR_ROFS, 378 NFSERR_NAMETOL, 379 NFSERR_DQUOT, 380 NFSERR_STALE, 381 NFSERR_BADHANDLE, 382 NFSERR_NOTSUPP, 383 NFSERR_SERVERFAULT, 384 0, 385 }; 386 387 static const short nfsv3err_mknod[] = { 388 NFSERR_IO, 389 NFSERR_IO, 390 NFSERR_ACCES, 391 NFSERR_EXIST, 392 NFSERR_NOTDIR, 393 NFSERR_NOSPC, 394 NFSERR_ROFS, 395 NFSERR_NAMETOL, 396 NFSERR_DQUOT, 397 NFSERR_STALE, 398 NFSERR_BADHANDLE, 399 NFSERR_NOTSUPP, 400 NFSERR_SERVERFAULT, 401 NFSERR_BADTYPE, 402 0, 403 }; 404 405 static const short nfsv3err_remove[] = { 406 NFSERR_IO, 407 NFSERR_NOENT, 408 NFSERR_IO, 409 NFSERR_ACCES, 410 NFSERR_NOTDIR, 411 NFSERR_ROFS, 412 NFSERR_NAMETOL, 413 NFSERR_STALE, 414 NFSERR_BADHANDLE, 415 NFSERR_SERVERFAULT, 416 0, 417 }; 418 419 static const short nfsv3err_rmdir[] = { 420 NFSERR_IO, 421 NFSERR_NOENT, 422 NFSERR_IO, 423 NFSERR_ACCES, 424 NFSERR_EXIST, 425 NFSERR_NOTDIR, 426 NFSERR_INVAL, 427 NFSERR_ROFS, 428 NFSERR_NAMETOL, 429 NFSERR_NOTEMPTY, 430 NFSERR_STALE, 431 NFSERR_BADHANDLE, 432 NFSERR_NOTSUPP, 433 NFSERR_SERVERFAULT, 434 0, 435 }; 436 437 static const short nfsv3err_rename[] = { 438 NFSERR_IO, 439 NFSERR_NOENT, 440 NFSERR_IO, 441 NFSERR_ACCES, 442 NFSERR_EXIST, 443 NFSERR_XDEV, 444 NFSERR_NOTDIR, 445 NFSERR_ISDIR, 446 NFSERR_INVAL, 447 NFSERR_NOSPC, 448 NFSERR_ROFS, 449 NFSERR_MLINK, 450 NFSERR_NAMETOL, 451 NFSERR_NOTEMPTY, 452 NFSERR_DQUOT, 453 NFSERR_STALE, 454 NFSERR_BADHANDLE, 455 NFSERR_NOTSUPP, 456 NFSERR_SERVERFAULT, 457 0, 458 }; 459 460 static const short nfsv3err_link[] = { 461 NFSERR_IO, 462 NFSERR_IO, 463 NFSERR_ACCES, 464 NFSERR_EXIST, 465 NFSERR_XDEV, 466 NFSERR_NOTDIR, 467 NFSERR_INVAL, 468 NFSERR_NOSPC, 469 NFSERR_ROFS, 470 NFSERR_MLINK, 471 NFSERR_NAMETOL, 472 NFSERR_DQUOT, 473 NFSERR_STALE, 474 NFSERR_BADHANDLE, 475 NFSERR_NOTSUPP, 476 NFSERR_SERVERFAULT, 477 0, 478 }; 479 480 static const short nfsv3err_readdir[] = { 481 NFSERR_IO, 482 NFSERR_IO, 483 NFSERR_ACCES, 484 NFSERR_NOTDIR, 485 NFSERR_STALE, 486 NFSERR_BADHANDLE, 487 NFSERR_BAD_COOKIE, 488 NFSERR_TOOSMALL, 489 NFSERR_SERVERFAULT, 490 0, 491 }; 492 493 static const short nfsv3err_readdirplus[] = { 494 NFSERR_IO, 495 NFSERR_IO, 496 NFSERR_ACCES, 497 NFSERR_NOTDIR, 498 NFSERR_STALE, 499 NFSERR_BADHANDLE, 500 NFSERR_BAD_COOKIE, 501 NFSERR_NOTSUPP, 502 NFSERR_TOOSMALL, 503 NFSERR_SERVERFAULT, 504 0, 505 }; 506 507 static const short nfsv3err_fsstat[] = { 508 NFSERR_IO, 509 NFSERR_IO, 510 NFSERR_STALE, 511 NFSERR_BADHANDLE, 512 NFSERR_SERVERFAULT, 513 0, 514 }; 515 516 static const short nfsv3err_fsinfo[] = { 517 NFSERR_STALE, 518 NFSERR_STALE, 519 NFSERR_BADHANDLE, 520 NFSERR_SERVERFAULT, 521 0, 522 }; 523 524 static const short nfsv3err_pathconf[] = { 525 NFSERR_STALE, 526 NFSERR_STALE, 527 NFSERR_BADHANDLE, 528 NFSERR_SERVERFAULT, 529 0, 530 }; 531 532 static const short nfsv3err_commit[] = { 533 NFSERR_IO, 534 NFSERR_IO, 535 NFSERR_STALE, 536 NFSERR_BADHANDLE, 537 NFSERR_SERVERFAULT, 538 0, 539 }; 540 541 static const short * const nfsrv_v3errmap[] = { 542 nfsv3err_null, 543 nfsv3err_getattr, 544 nfsv3err_setattr, 545 nfsv3err_lookup, 546 nfsv3err_access, 547 nfsv3err_readlink, 548 nfsv3err_read, 549 nfsv3err_write, 550 nfsv3err_create, 551 nfsv3err_mkdir, 552 nfsv3err_symlink, 553 nfsv3err_mknod, 554 nfsv3err_remove, 555 nfsv3err_rmdir, 556 nfsv3err_rename, 557 nfsv3err_link, 558 nfsv3err_readdir, 559 nfsv3err_readdirplus, 560 nfsv3err_fsstat, 561 nfsv3err_fsinfo, 562 nfsv3err_pathconf, 563 nfsv3err_commit, 564 }; 565 566 extern struct nfsrtt nfsrtt; 567 extern struct nfsnodehashhead *nfsnodehashtbl; 568 extern u_long nfsnodehash; 569 570 u_long nfsdirhashmask; 571 572 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *)); 573 574 /* 575 * Create the header for an rpc request packet 576 * The hsiz is the size of the rest of the nfs request header. 577 * (just used to decide if a cluster is a good idea) 578 */ 579 struct mbuf * 580 nfsm_reqh(struct nfsnode *np, u_long procid, int hsiz, char **bposp) 581 { 582 struct mbuf *mb; 583 char *bpos; 584 585 mb = m_get(M_WAIT, MT_DATA); 586 MCLAIM(mb, &nfs_mowner); 587 if (hsiz >= MINCLSIZE) 588 m_clget(mb, M_WAIT); 589 mb->m_len = 0; 590 bpos = mtod(mb, void *); 591 592 /* Finally, return values */ 593 *bposp = bpos; 594 return (mb); 595 } 596 597 /* 598 * Build the RPC header and fill in the authorization info. 599 * The authorization string argument is only used when the credentials 600 * come from outside of the kernel. 601 * Returns the head of the mbuf list. 602 */ 603 struct mbuf * 604 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len, 605 verf_str, mrest, mrest_len, mbp, xidp) 606 kauth_cred_t cr; 607 int nmflag; 608 int procid; 609 int auth_type; 610 int auth_len; 611 char *auth_str; 612 int verf_len; 613 char *verf_str; 614 struct mbuf *mrest; 615 int mrest_len; 616 struct mbuf **mbp; 617 u_int32_t *xidp; 618 { 619 struct mbuf *mb; 620 u_int32_t *tl; 621 char *bpos; 622 int i; 623 struct mbuf *mreq; 624 int siz, grpsiz, authsiz; 625 626 authsiz = nfsm_rndup(auth_len); 627 mb = m_gethdr(M_WAIT, MT_DATA); 628 MCLAIM(mb, &nfs_mowner); 629 if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) { 630 m_clget(mb, M_WAIT); 631 } else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) { 632 MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED); 633 } else { 634 MH_ALIGN(mb, 8 * NFSX_UNSIGNED); 635 } 636 mb->m_len = 0; 637 mreq = mb; 638 bpos = mtod(mb, void *); 639 640 /* 641 * First the RPC header. 642 */ 643 nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 644 645 *tl++ = *xidp = nfs_getxid(); 646 *tl++ = rpc_call; 647 *tl++ = rpc_vers; 648 *tl++ = txdr_unsigned(NFS_PROG); 649 if (nmflag & NFSMNT_NFSV3) 650 *tl++ = txdr_unsigned(NFS_VER3); 651 else 652 *tl++ = txdr_unsigned(NFS_VER2); 653 if (nmflag & NFSMNT_NFSV3) 654 *tl++ = txdr_unsigned(procid); 655 else 656 *tl++ = txdr_unsigned(nfsv2_procid[procid]); 657 658 /* 659 * And then the authorization cred. 660 */ 661 *tl++ = txdr_unsigned(auth_type); 662 *tl = txdr_unsigned(authsiz); 663 switch (auth_type) { 664 case RPCAUTH_UNIX: 665 nfsm_build(tl, u_int32_t *, auth_len); 666 *tl++ = 0; /* stamp ?? */ 667 *tl++ = 0; /* NULL hostname */ 668 *tl++ = txdr_unsigned(kauth_cred_geteuid(cr)); 669 *tl++ = txdr_unsigned(kauth_cred_getegid(cr)); 670 grpsiz = (auth_len >> 2) - 5; 671 *tl++ = txdr_unsigned(grpsiz); 672 for (i = 0; i < grpsiz; i++) 673 *tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */ 674 break; 675 case RPCAUTH_KERB4: 676 siz = auth_len; 677 while (siz > 0) { 678 if (M_TRAILINGSPACE(mb) == 0) { 679 struct mbuf *mb2; 680 mb2 = m_get(M_WAIT, MT_DATA); 681 MCLAIM(mb2, &nfs_mowner); 682 if (siz >= MINCLSIZE) 683 m_clget(mb2, M_WAIT); 684 mb->m_next = mb2; 685 mb = mb2; 686 mb->m_len = 0; 687 bpos = mtod(mb, void *); 688 } 689 i = min(siz, M_TRAILINGSPACE(mb)); 690 memcpy(bpos, auth_str, i); 691 mb->m_len += i; 692 auth_str += i; 693 bpos += i; 694 siz -= i; 695 } 696 if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) { 697 for (i = 0; i < siz; i++) 698 *bpos++ = '\0'; 699 mb->m_len += siz; 700 } 701 break; 702 }; 703 704 /* 705 * And the verifier... 706 */ 707 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 708 if (verf_str) { 709 *tl++ = txdr_unsigned(RPCAUTH_KERB4); 710 *tl = txdr_unsigned(verf_len); 711 siz = verf_len; 712 while (siz > 0) { 713 if (M_TRAILINGSPACE(mb) == 0) { 714 struct mbuf *mb2; 715 mb2 = m_get(M_WAIT, MT_DATA); 716 MCLAIM(mb2, &nfs_mowner); 717 if (siz >= MINCLSIZE) 718 m_clget(mb2, M_WAIT); 719 mb->m_next = mb2; 720 mb = mb2; 721 mb->m_len = 0; 722 bpos = mtod(mb, void *); 723 } 724 i = min(siz, M_TRAILINGSPACE(mb)); 725 memcpy(bpos, verf_str, i); 726 mb->m_len += i; 727 verf_str += i; 728 bpos += i; 729 siz -= i; 730 } 731 if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) { 732 for (i = 0; i < siz; i++) 733 *bpos++ = '\0'; 734 mb->m_len += siz; 735 } 736 } else { 737 *tl++ = txdr_unsigned(RPCAUTH_NULL); 738 *tl = 0; 739 } 740 mb->m_next = mrest; 741 mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len; 742 mreq->m_pkthdr.rcvif = (struct ifnet *)0; 743 *mbp = mb; 744 return (mreq); 745 } 746 747 /* 748 * copies mbuf chain to the uio scatter/gather list 749 */ 750 int 751 nfsm_mbuftouio(mrep, uiop, siz, dpos) 752 struct mbuf **mrep; 753 struct uio *uiop; 754 int siz; 755 char **dpos; 756 { 757 char *mbufcp, *uiocp; 758 int xfer, left, len; 759 struct mbuf *mp; 760 long uiosiz, rem; 761 int error = 0; 762 763 mp = *mrep; 764 mbufcp = *dpos; 765 len = mtod(mp, char *) + mp->m_len - mbufcp; 766 rem = nfsm_rndup(siz)-siz; 767 while (siz > 0) { 768 if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) 769 return (EFBIG); 770 left = uiop->uio_iov->iov_len; 771 uiocp = uiop->uio_iov->iov_base; 772 if (left > siz) 773 left = siz; 774 uiosiz = left; 775 while (left > 0) { 776 while (len == 0) { 777 mp = mp->m_next; 778 if (mp == NULL) 779 return (EBADRPC); 780 mbufcp = mtod(mp, void *); 781 len = mp->m_len; 782 } 783 xfer = (left > len) ? len : left; 784 error = copyout_vmspace(uiop->uio_vmspace, mbufcp, 785 uiocp, xfer); 786 if (error) { 787 return error; 788 } 789 left -= xfer; 790 len -= xfer; 791 mbufcp += xfer; 792 uiocp += xfer; 793 uiop->uio_offset += xfer; 794 uiop->uio_resid -= xfer; 795 } 796 if (uiop->uio_iov->iov_len <= siz) { 797 uiop->uio_iovcnt--; 798 uiop->uio_iov++; 799 } else { 800 uiop->uio_iov->iov_base = 801 (char *)uiop->uio_iov->iov_base + uiosiz; 802 uiop->uio_iov->iov_len -= uiosiz; 803 } 804 siz -= uiosiz; 805 } 806 *dpos = mbufcp; 807 *mrep = mp; 808 if (rem > 0) { 809 if (len < rem) 810 error = nfs_adv(mrep, dpos, rem, len); 811 else 812 *dpos += rem; 813 } 814 return (error); 815 } 816 817 /* 818 * copies a uio scatter/gather list to an mbuf chain. 819 * NOTE: can ony handle iovcnt == 1 820 */ 821 int 822 nfsm_uiotombuf(uiop, mq, siz, bpos) 823 struct uio *uiop; 824 struct mbuf **mq; 825 int siz; 826 char **bpos; 827 { 828 char *uiocp; 829 struct mbuf *mp, *mp2; 830 int xfer, left, mlen; 831 int uiosiz, clflg, rem; 832 char *cp; 833 int error; 834 835 #ifdef DIAGNOSTIC 836 if (uiop->uio_iovcnt != 1) 837 panic("nfsm_uiotombuf: iovcnt != 1"); 838 #endif 839 840 if (siz > MLEN) /* or should it >= MCLBYTES ?? */ 841 clflg = 1; 842 else 843 clflg = 0; 844 rem = nfsm_rndup(siz)-siz; 845 mp = mp2 = *mq; 846 while (siz > 0) { 847 left = uiop->uio_iov->iov_len; 848 uiocp = uiop->uio_iov->iov_base; 849 if (left > siz) 850 left = siz; 851 uiosiz = left; 852 while (left > 0) { 853 mlen = M_TRAILINGSPACE(mp); 854 if (mlen == 0) { 855 mp = m_get(M_WAIT, MT_DATA); 856 MCLAIM(mp, &nfs_mowner); 857 if (clflg) 858 m_clget(mp, M_WAIT); 859 mp->m_len = 0; 860 mp2->m_next = mp; 861 mp2 = mp; 862 mlen = M_TRAILINGSPACE(mp); 863 } 864 xfer = (left > mlen) ? mlen : left; 865 cp = mtod(mp, char *) + mp->m_len; 866 error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp, 867 xfer); 868 if (error) { 869 /* XXX */ 870 } 871 mp->m_len += xfer; 872 left -= xfer; 873 uiocp += xfer; 874 uiop->uio_offset += xfer; 875 uiop->uio_resid -= xfer; 876 } 877 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + 878 uiosiz; 879 uiop->uio_iov->iov_len -= uiosiz; 880 siz -= uiosiz; 881 } 882 if (rem > 0) { 883 if (rem > M_TRAILINGSPACE(mp)) { 884 mp = m_get(M_WAIT, MT_DATA); 885 MCLAIM(mp, &nfs_mowner); 886 mp->m_len = 0; 887 mp2->m_next = mp; 888 } 889 cp = mtod(mp, char *) + mp->m_len; 890 for (left = 0; left < rem; left++) 891 *cp++ = '\0'; 892 mp->m_len += rem; 893 *bpos = cp; 894 } else 895 *bpos = mtod(mp, char *) + mp->m_len; 896 *mq = mp; 897 return (0); 898 } 899 900 /* 901 * Get at least "siz" bytes of correctly aligned data. 902 * When called the mbuf pointers are not necessarily correct, 903 * dsosp points to what ought to be in m_data and left contains 904 * what ought to be in m_len. 905 * This is used by the macros nfsm_dissect and nfsm_dissecton for tough 906 * cases. (The macros use the vars. dpos and dpos2) 907 */ 908 int 909 nfsm_disct(mdp, dposp, siz, left, cp2) 910 struct mbuf **mdp; 911 char **dposp; 912 int siz; 913 int left; 914 char **cp2; 915 { 916 struct mbuf *m1, *m2; 917 struct mbuf *havebuf = NULL; 918 char *src = *dposp; 919 char *dst; 920 int len; 921 922 #ifdef DEBUG 923 if (left < 0) 924 panic("nfsm_disct: left < 0"); 925 #endif 926 m1 = *mdp; 927 /* 928 * Skip through the mbuf chain looking for an mbuf with 929 * some data. If the first mbuf found has enough data 930 * and it is correctly aligned return it. 931 */ 932 while (left == 0) { 933 havebuf = m1; 934 *mdp = m1 = m1->m_next; 935 if (m1 == NULL) 936 return (EBADRPC); 937 src = mtod(m1, void *); 938 left = m1->m_len; 939 /* 940 * If we start a new mbuf and it is big enough 941 * and correctly aligned just return it, don't 942 * do any pull up. 943 */ 944 if (left >= siz && nfsm_aligned(src)) { 945 *cp2 = src; 946 *dposp = src + siz; 947 return (0); 948 } 949 } 950 if (m1->m_flags & M_EXT) { 951 if (havebuf) { 952 /* If the first mbuf with data has external data 953 * and there is a previous empty mbuf use it 954 * to move the data into. 955 */ 956 m2 = m1; 957 *mdp = m1 = havebuf; 958 if (m1->m_flags & M_EXT) { 959 MEXTREMOVE(m1); 960 } 961 } else { 962 /* 963 * If the first mbuf has a external data 964 * and there is no previous empty mbuf 965 * allocate a new mbuf and move the external 966 * data to the new mbuf. Also make the first 967 * mbuf look empty. 968 */ 969 m2 = m_get(M_WAIT, MT_DATA); 970 m2->m_ext = m1->m_ext; 971 m2->m_data = src; 972 m2->m_len = left; 973 MCLADDREFERENCE(m1, m2); 974 MEXTREMOVE(m1); 975 m2->m_next = m1->m_next; 976 m1->m_next = m2; 977 } 978 m1->m_len = 0; 979 if (m1->m_flags & M_PKTHDR) 980 dst = m1->m_pktdat; 981 else 982 dst = m1->m_dat; 983 m1->m_data = dst; 984 } else { 985 /* 986 * If the first mbuf has no external data 987 * move the data to the front of the mbuf. 988 */ 989 if (m1->m_flags & M_PKTHDR) 990 dst = m1->m_pktdat; 991 else 992 dst = m1->m_dat; 993 m1->m_data = dst; 994 if (dst != src) 995 memmove(dst, src, left); 996 dst += left; 997 m1->m_len = left; 998 m2 = m1->m_next; 999 } 1000 *cp2 = m1->m_data; 1001 *dposp = mtod(m1, char *) + siz; 1002 /* 1003 * Loop through mbufs pulling data up into first mbuf until 1004 * the first mbuf is full or there is no more data to 1005 * pullup. 1006 */ 1007 while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) { 1008 if ((len = min(len, m2->m_len)) != 0) 1009 memcpy(dst, m2->m_data, len); 1010 m1->m_len += len; 1011 dst += len; 1012 m2->m_data += len; 1013 m2->m_len -= len; 1014 m2 = m2->m_next; 1015 } 1016 if (m1->m_len < siz) 1017 return (EBADRPC); 1018 return (0); 1019 } 1020 1021 /* 1022 * Advance the position in the mbuf chain. 1023 */ 1024 int 1025 nfs_adv(mdp, dposp, offs, left) 1026 struct mbuf **mdp; 1027 char **dposp; 1028 int offs; 1029 int left; 1030 { 1031 struct mbuf *m; 1032 int s; 1033 1034 m = *mdp; 1035 s = left; 1036 while (s < offs) { 1037 offs -= s; 1038 m = m->m_next; 1039 if (m == NULL) 1040 return (EBADRPC); 1041 s = m->m_len; 1042 } 1043 *mdp = m; 1044 *dposp = mtod(m, char *) + offs; 1045 return (0); 1046 } 1047 1048 /* 1049 * Copy a string into mbufs for the hard cases... 1050 */ 1051 int 1052 nfsm_strtmbuf(mb, bpos, cp, siz) 1053 struct mbuf **mb; 1054 char **bpos; 1055 const char *cp; 1056 long siz; 1057 { 1058 struct mbuf *m1 = NULL, *m2; 1059 long left, xfer, len, tlen; 1060 u_int32_t *tl; 1061 int putsize; 1062 1063 putsize = 1; 1064 m2 = *mb; 1065 left = M_TRAILINGSPACE(m2); 1066 if (left > 0) { 1067 tl = ((u_int32_t *)(*bpos)); 1068 *tl++ = txdr_unsigned(siz); 1069 putsize = 0; 1070 left -= NFSX_UNSIGNED; 1071 m2->m_len += NFSX_UNSIGNED; 1072 if (left > 0) { 1073 memcpy((void *) tl, cp, left); 1074 siz -= left; 1075 cp += left; 1076 m2->m_len += left; 1077 left = 0; 1078 } 1079 } 1080 /* Loop around adding mbufs */ 1081 while (siz > 0) { 1082 m1 = m_get(M_WAIT, MT_DATA); 1083 MCLAIM(m1, &nfs_mowner); 1084 if (siz > MLEN) 1085 m_clget(m1, M_WAIT); 1086 m1->m_len = NFSMSIZ(m1); 1087 m2->m_next = m1; 1088 m2 = m1; 1089 tl = mtod(m1, u_int32_t *); 1090 tlen = 0; 1091 if (putsize) { 1092 *tl++ = txdr_unsigned(siz); 1093 m1->m_len -= NFSX_UNSIGNED; 1094 tlen = NFSX_UNSIGNED; 1095 putsize = 0; 1096 } 1097 if (siz < m1->m_len) { 1098 len = nfsm_rndup(siz); 1099 xfer = siz; 1100 if (xfer < len) 1101 *(tl+(xfer>>2)) = 0; 1102 } else { 1103 xfer = len = m1->m_len; 1104 } 1105 memcpy((void *) tl, cp, xfer); 1106 m1->m_len = len+tlen; 1107 siz -= xfer; 1108 cp += xfer; 1109 } 1110 *mb = m1; 1111 *bpos = mtod(m1, char *) + m1->m_len; 1112 return (0); 1113 } 1114 1115 /* 1116 * Directory caching routines. They work as follows: 1117 * - a cache is maintained per VDIR nfsnode. 1118 * - for each offset cookie that is exported to userspace, and can 1119 * thus be thrown back at us as an offset to VOP_READDIR, store 1120 * information in the cache. 1121 * - cached are: 1122 * - cookie itself 1123 * - blocknumber (essentially just a search key in the buffer cache) 1124 * - entry number in block. 1125 * - offset cookie of block in which this entry is stored 1126 * - 32 bit cookie if NFSMNT_XLATECOOKIE is used. 1127 * - entries are looked up in a hash table 1128 * - also maintained is an LRU list of entries, used to determine 1129 * which ones to delete if the cache grows too large. 1130 * - if 32 <-> 64 translation mode is requested for a filesystem, 1131 * the cache also functions as a translation table 1132 * - in the translation case, invalidating the cache does not mean 1133 * flushing it, but just marking entries as invalid, except for 1134 * the <64bit cookie, 32bitcookie> pair which is still valid, to 1135 * still be able to use the cache as a translation table. 1136 * - 32 bit cookies are uniquely created by combining the hash table 1137 * entry value, and one generation count per hash table entry, 1138 * incremented each time an entry is appended to the chain. 1139 * - the cache is invalidated each time a direcory is modified 1140 * - sanity checks are also done; if an entry in a block turns 1141 * out not to have a matching cookie, the cache is invalidated 1142 * and a new block starting from the wanted offset is fetched from 1143 * the server. 1144 * - directory entries as read from the server are extended to contain 1145 * the 64bit and, optionally, the 32bit cookies, for sanity checking 1146 * the cache and exporting them to userspace through the cookie 1147 * argument to VOP_READDIR. 1148 */ 1149 1150 u_long 1151 nfs_dirhash(off) 1152 off_t off; 1153 { 1154 int i; 1155 char *cp = (char *)&off; 1156 u_long sum = 0L; 1157 1158 for (i = 0 ; i < sizeof (off); i++) 1159 sum += *cp++; 1160 1161 return sum; 1162 } 1163 1164 #define _NFSDC_MTX(np) (&NFSTOV(np)->v_interlock) 1165 #define NFSDC_LOCK(np) mutex_enter(_NFSDC_MTX(np)) 1166 #define NFSDC_UNLOCK(np) mutex_exit(_NFSDC_MTX(np)) 1167 #define NFSDC_ASSERT_LOCKED(np) KASSERT(mutex_owned(_NFSDC_MTX(np))) 1168 1169 void 1170 nfs_initdircache(vp) 1171 struct vnode *vp; 1172 { 1173 struct nfsnode *np = VTONFS(vp); 1174 struct nfsdirhashhead *dircache; 1175 1176 dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF, 1177 M_WAITOK, &nfsdirhashmask); 1178 1179 NFSDC_LOCK(np); 1180 if (np->n_dircache == NULL) { 1181 np->n_dircachesize = 0; 1182 np->n_dircache = dircache; 1183 dircache = NULL; 1184 TAILQ_INIT(&np->n_dirchain); 1185 } 1186 NFSDC_UNLOCK(np); 1187 if (dircache) 1188 hashdone(dircache, M_NFSDIROFF); 1189 } 1190 1191 void 1192 nfs_initdirxlatecookie(vp) 1193 struct vnode *vp; 1194 { 1195 struct nfsnode *np = VTONFS(vp); 1196 unsigned *dirgens; 1197 1198 KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE); 1199 1200 dirgens = malloc(NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF, 1201 M_WAITOK|M_ZERO); 1202 NFSDC_LOCK(np); 1203 if (np->n_dirgens == NULL) { 1204 np->n_dirgens = dirgens; 1205 dirgens = NULL; 1206 } 1207 NFSDC_UNLOCK(np); 1208 if (dirgens) 1209 free(dirgens, M_NFSDIROFF); 1210 } 1211 1212 static const struct nfsdircache dzero; 1213 1214 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *)); 1215 static void nfs_putdircache_unlocked __P((struct nfsnode *, 1216 struct nfsdircache *)); 1217 1218 static void 1219 nfs_unlinkdircache(np, ndp) 1220 struct nfsnode *np; 1221 struct nfsdircache *ndp; 1222 { 1223 1224 NFSDC_ASSERT_LOCKED(np); 1225 KASSERT(ndp != &dzero); 1226 1227 if (LIST_NEXT(ndp, dc_hash) == (void *)-1) 1228 return; 1229 1230 TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain); 1231 LIST_REMOVE(ndp, dc_hash); 1232 LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */ 1233 1234 nfs_putdircache_unlocked(np, ndp); 1235 } 1236 1237 void 1238 nfs_putdircache(np, ndp) 1239 struct nfsnode *np; 1240 struct nfsdircache *ndp; 1241 { 1242 int ref; 1243 1244 if (ndp == &dzero) 1245 return; 1246 1247 KASSERT(ndp->dc_refcnt > 0); 1248 NFSDC_LOCK(np); 1249 ref = --ndp->dc_refcnt; 1250 NFSDC_UNLOCK(np); 1251 1252 if (ref == 0) 1253 free(ndp, M_NFSDIROFF); 1254 } 1255 1256 static void 1257 nfs_putdircache_unlocked(struct nfsnode *np, struct nfsdircache *ndp) 1258 { 1259 int ref; 1260 1261 NFSDC_ASSERT_LOCKED(np); 1262 1263 if (ndp == &dzero) 1264 return; 1265 1266 KASSERT(ndp->dc_refcnt > 0); 1267 ref = --ndp->dc_refcnt; 1268 if (ref == 0) 1269 free(ndp, M_NFSDIROFF); 1270 } 1271 1272 struct nfsdircache * 1273 nfs_searchdircache(vp, off, do32, hashent) 1274 struct vnode *vp; 1275 off_t off; 1276 int do32; 1277 int *hashent; 1278 { 1279 struct nfsdirhashhead *ndhp; 1280 struct nfsdircache *ndp = NULL; 1281 struct nfsnode *np = VTONFS(vp); 1282 unsigned ent; 1283 1284 /* 1285 * Zero is always a valid cookie. 1286 */ 1287 if (off == 0) 1288 /* XXXUNCONST */ 1289 return (struct nfsdircache *)__UNCONST(&dzero); 1290 1291 if (!np->n_dircache) 1292 return NULL; 1293 1294 /* 1295 * We use a 32bit cookie as search key, directly reconstruct 1296 * the hashentry. Else use the hashfunction. 1297 */ 1298 if (do32) { 1299 ent = (u_int32_t)off >> 24; 1300 if (ent >= NFS_DIRHASHSIZ) 1301 return NULL; 1302 ndhp = &np->n_dircache[ent]; 1303 } else { 1304 ndhp = NFSDIRHASH(np, off); 1305 } 1306 1307 if (hashent) 1308 *hashent = (int)(ndhp - np->n_dircache); 1309 1310 NFSDC_LOCK(np); 1311 if (do32) { 1312 LIST_FOREACH(ndp, ndhp, dc_hash) { 1313 if (ndp->dc_cookie32 == (u_int32_t)off) { 1314 /* 1315 * An invalidated entry will become the 1316 * start of a new block fetched from 1317 * the server. 1318 */ 1319 if (ndp->dc_flags & NFSDC_INVALID) { 1320 ndp->dc_blkcookie = ndp->dc_cookie; 1321 ndp->dc_entry = 0; 1322 ndp->dc_flags &= ~NFSDC_INVALID; 1323 } 1324 break; 1325 } 1326 } 1327 } else { 1328 LIST_FOREACH(ndp, ndhp, dc_hash) { 1329 if (ndp->dc_cookie == off) 1330 break; 1331 } 1332 } 1333 if (ndp != NULL) 1334 ndp->dc_refcnt++; 1335 NFSDC_UNLOCK(np); 1336 return ndp; 1337 } 1338 1339 1340 struct nfsdircache * 1341 nfs_enterdircache(struct vnode *vp, off_t off, off_t blkoff, int en, 1342 daddr_t blkno) 1343 { 1344 struct nfsnode *np = VTONFS(vp); 1345 struct nfsdirhashhead *ndhp; 1346 struct nfsdircache *ndp = NULL; 1347 struct nfsdircache *newndp = NULL; 1348 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1349 int hashent = 0, gen, overwrite; /* XXX: GCC */ 1350 1351 /* 1352 * XXX refuse entries for offset 0. amd(8) erroneously sets 1353 * cookie 0 for the '.' entry, making this necessary. This 1354 * isn't so bad, as 0 is a special case anyway. 1355 */ 1356 if (off == 0) 1357 /* XXXUNCONST */ 1358 return (struct nfsdircache *)__UNCONST(&dzero); 1359 1360 if (!np->n_dircache) 1361 /* 1362 * XXX would like to do this in nfs_nget but vtype 1363 * isn't known at that time. 1364 */ 1365 nfs_initdircache(vp); 1366 1367 if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens) 1368 nfs_initdirxlatecookie(vp); 1369 1370 retry: 1371 ndp = nfs_searchdircache(vp, off, 0, &hashent); 1372 1373 NFSDC_LOCK(np); 1374 if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) { 1375 /* 1376 * Overwriting an old entry. Check if it's the same. 1377 * If so, just return. If not, remove the old entry. 1378 */ 1379 if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en) 1380 goto done; 1381 nfs_unlinkdircache(np, ndp); 1382 nfs_putdircache_unlocked(np, ndp); 1383 ndp = NULL; 1384 } 1385 1386 ndhp = &np->n_dircache[hashent]; 1387 1388 if (!ndp) { 1389 if (newndp == NULL) { 1390 NFSDC_UNLOCK(np); 1391 newndp = malloc(sizeof(*ndp), M_NFSDIROFF, M_WAITOK); 1392 newndp->dc_refcnt = 1; 1393 LIST_NEXT(newndp, dc_hash) = (void *)-1; 1394 goto retry; 1395 } 1396 ndp = newndp; 1397 newndp = NULL; 1398 overwrite = 0; 1399 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 1400 /* 1401 * We're allocating a new entry, so bump the 1402 * generation number. 1403 */ 1404 KASSERT(np->n_dirgens); 1405 gen = ++np->n_dirgens[hashent]; 1406 if (gen == 0) { 1407 np->n_dirgens[hashent]++; 1408 gen++; 1409 } 1410 ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff); 1411 } 1412 } else 1413 overwrite = 1; 1414 1415 ndp->dc_cookie = off; 1416 ndp->dc_blkcookie = blkoff; 1417 ndp->dc_entry = en; 1418 ndp->dc_flags = 0; 1419 1420 if (overwrite) 1421 goto done; 1422 1423 /* 1424 * If the maximum directory cookie cache size has been reached 1425 * for this node, take one off the front. The idea is that 1426 * directories are typically read front-to-back once, so that 1427 * the oldest entries can be thrown away without much performance 1428 * loss. 1429 */ 1430 if (np->n_dircachesize == NFS_MAXDIRCACHE) { 1431 nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain)); 1432 } else 1433 np->n_dircachesize++; 1434 1435 KASSERT(ndp->dc_refcnt == 1); 1436 LIST_INSERT_HEAD(ndhp, ndp, dc_hash); 1437 TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain); 1438 ndp->dc_refcnt++; 1439 done: 1440 KASSERT(ndp->dc_refcnt > 0); 1441 NFSDC_UNLOCK(np); 1442 if (newndp) 1443 nfs_putdircache(np, newndp); 1444 return ndp; 1445 } 1446 1447 void 1448 nfs_invaldircache(vp, flags) 1449 struct vnode *vp; 1450 int flags; 1451 { 1452 struct nfsnode *np = VTONFS(vp); 1453 struct nfsdircache *ndp = NULL; 1454 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1455 const bool forcefree = flags & NFS_INVALDIRCACHE_FORCE; 1456 1457 #ifdef DIAGNOSTIC 1458 if (vp->v_type != VDIR) 1459 panic("nfs: invaldircache: not dir"); 1460 #endif 1461 1462 if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0) 1463 np->n_flag &= ~NEOFVALID; 1464 1465 if (!np->n_dircache) 1466 return; 1467 1468 NFSDC_LOCK(np); 1469 if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) { 1470 while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) { 1471 KASSERT(!forcefree || ndp->dc_refcnt == 1); 1472 nfs_unlinkdircache(np, ndp); 1473 } 1474 np->n_dircachesize = 0; 1475 if (forcefree && np->n_dirgens) { 1476 FREE(np->n_dirgens, M_NFSDIROFF); 1477 np->n_dirgens = NULL; 1478 } 1479 } else { 1480 TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain) 1481 ndp->dc_flags |= NFSDC_INVALID; 1482 } 1483 1484 NFSDC_UNLOCK(np); 1485 } 1486 1487 /* 1488 * Called once before VFS init to initialize shared and 1489 * server-specific data structures. 1490 */ 1491 static int 1492 nfs_init0(void) 1493 { 1494 extern krwlock_t netexport_lock; /* XXX */ 1495 1496 rw_init(&netexport_lock); 1497 1498 nfsrtt.pos = 0; 1499 rpc_vers = txdr_unsigned(RPC_VER2); 1500 rpc_call = txdr_unsigned(RPC_CALL); 1501 rpc_reply = txdr_unsigned(RPC_REPLY); 1502 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED); 1503 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED); 1504 rpc_mismatch = txdr_unsigned(RPC_MISMATCH); 1505 rpc_autherr = txdr_unsigned(RPC_AUTHERR); 1506 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX); 1507 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4); 1508 nfs_prog = txdr_unsigned(NFS_PROG); 1509 nfs_true = txdr_unsigned(true); 1510 nfs_false = txdr_unsigned(false); 1511 nfs_xdrneg1 = txdr_unsigned(-1); 1512 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; 1513 if (nfs_ticks < 1) 1514 nfs_ticks = 1; 1515 #ifdef NFSSERVER 1516 nfsrv_init(0); /* Init server data structures */ 1517 nfsrv_initcache(); /* Init the server request cache */ 1518 { 1519 extern krwlock_t netexport_lock; /* XXX */ 1520 rw_init(&netexport_lock); 1521 } 1522 #endif /* NFSSERVER */ 1523 1524 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) 1525 nfsdreq_init(); 1526 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */ 1527 1528 /* 1529 * Initialize reply list and start timer 1530 */ 1531 TAILQ_INIT(&nfs_reqq); 1532 nfs_timer(nfs_timer); 1533 MOWNER_ATTACH(&nfs_mowner); 1534 1535 #ifdef NFS 1536 /* Initialize the kqueue structures */ 1537 nfs_kqinit(); 1538 /* Initialize the iod structures */ 1539 nfs_iodinit(); 1540 #endif 1541 return 0; 1542 } 1543 1544 void 1545 nfs_init(void) 1546 { 1547 static ONCE_DECL(nfs_init_once); 1548 1549 RUN_ONCE(&nfs_init_once, nfs_init0); 1550 } 1551 1552 #ifdef NFS 1553 /* 1554 * Called once at VFS init to initialize client-specific data structures. 1555 */ 1556 void 1557 nfs_vfs_init() 1558 { 1559 /* Initialize NFS server / client shared data. */ 1560 nfs_init(); 1561 1562 nfs_nhinit(); /* Init the nfsnode table */ 1563 nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4); 1564 } 1565 1566 void 1567 nfs_vfs_reinit() 1568 { 1569 nfs_nhreinit(); 1570 } 1571 1572 void 1573 nfs_vfs_done() 1574 { 1575 nfs_nhdone(); 1576 } 1577 1578 /* 1579 * Attribute cache routines. 1580 * nfs_loadattrcache() - loads or updates the cache contents from attributes 1581 * that are on the mbuf list 1582 * nfs_getattrcache() - returns valid attributes if found in cache, returns 1583 * error otherwise 1584 */ 1585 1586 /* 1587 * Load the attribute cache (that lives in the nfsnode entry) with 1588 * the values on the mbuf list and 1589 * Iff vap not NULL 1590 * copy the attributes to *vaper 1591 */ 1592 int 1593 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags) 1594 struct vnode **vpp; 1595 struct mbuf **mdp; 1596 char **dposp; 1597 struct vattr *vaper; 1598 int flags; 1599 { 1600 int32_t t1; 1601 char *cp2; 1602 int error = 0; 1603 struct mbuf *md; 1604 int v3 = NFS_ISV3(*vpp); 1605 1606 md = *mdp; 1607 t1 = (mtod(md, char *) + md->m_len) - *dposp; 1608 error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2); 1609 if (error) 1610 return (error); 1611 return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags); 1612 } 1613 1614 int 1615 nfs_loadattrcache(vpp, fp, vaper, flags) 1616 struct vnode **vpp; 1617 struct nfs_fattr *fp; 1618 struct vattr *vaper; 1619 int flags; 1620 { 1621 struct vnode *vp = *vpp; 1622 struct vattr *vap; 1623 int v3 = NFS_ISV3(vp); 1624 enum vtype vtyp; 1625 u_short vmode; 1626 struct timespec mtime; 1627 struct timespec ctime; 1628 struct vnode *nvp; 1629 int32_t rdev; 1630 struct nfsnode *np; 1631 extern int (**spec_nfsv2nodeop_p) __P((void *)); 1632 uid_t uid; 1633 gid_t gid; 1634 1635 if (v3) { 1636 vtyp = nfsv3tov_type(fp->fa_type); 1637 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1638 rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1), 1639 fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2)); 1640 fxdr_nfsv3time(&fp->fa3_mtime, &mtime); 1641 fxdr_nfsv3time(&fp->fa3_ctime, &ctime); 1642 } else { 1643 vtyp = nfsv2tov_type(fp->fa_type); 1644 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1645 if (vtyp == VNON || vtyp == VREG) 1646 vtyp = IFTOVT(vmode); 1647 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); 1648 fxdr_nfsv2time(&fp->fa2_mtime, &mtime); 1649 ctime.tv_sec = fxdr_unsigned(u_int32_t, 1650 fp->fa2_ctime.nfsv2_sec); 1651 ctime.tv_nsec = 0; 1652 1653 /* 1654 * Really ugly NFSv2 kludge. 1655 */ 1656 if (vtyp == VCHR && rdev == 0xffffffff) 1657 vtyp = VFIFO; 1658 } 1659 1660 vmode &= ALLPERMS; 1661 1662 /* 1663 * If v_type == VNON it is a new node, so fill in the v_type, 1664 * n_mtime fields. Check to see if it represents a special 1665 * device, and if so, check for a possible alias. Once the 1666 * correct vnode has been obtained, fill in the rest of the 1667 * information. 1668 */ 1669 np = VTONFS(vp); 1670 if (vp->v_type == VNON) { 1671 vp->v_type = vtyp; 1672 if (vp->v_type == VFIFO) { 1673 extern int (**fifo_nfsv2nodeop_p) __P((void *)); 1674 vp->v_op = fifo_nfsv2nodeop_p; 1675 } else if (vp->v_type == VREG) { 1676 mutex_init(&np->n_commitlock, MUTEX_DEFAULT, IPL_NONE); 1677 } else if (vp->v_type == VCHR || vp->v_type == VBLK) { 1678 vp->v_op = spec_nfsv2nodeop_p; 1679 nvp = checkalias(vp, (dev_t)rdev, vp->v_mount); 1680 if (nvp) { 1681 /* 1682 * Discard unneeded vnode, but save its nfsnode. 1683 * Since the nfsnode does not have a lock, its 1684 * vnode lock has to be carried over. 1685 */ 1686 /* 1687 * XXX is the old node sure to be locked here? 1688 */ 1689 KASSERT(lockstatus(&vp->v_lock) == 1690 LK_EXCLUSIVE); 1691 nvp->v_data = vp->v_data; 1692 vp->v_data = NULL; 1693 VOP_UNLOCK(vp, 0); 1694 vp->v_op = spec_vnodeop_p; 1695 vrele(vp); 1696 vgone(vp); 1697 lockmgr(&nvp->v_lock, LK_EXCLUSIVE, 1698 &nvp->v_interlock); 1699 /* 1700 * Reinitialize aliased node. 1701 */ 1702 np->n_vnode = nvp; 1703 *vpp = vp = nvp; 1704 } 1705 } 1706 np->n_mtime = mtime; 1707 } 1708 uid = fxdr_unsigned(uid_t, fp->fa_uid); 1709 gid = fxdr_unsigned(gid_t, fp->fa_gid); 1710 vap = np->n_vattr; 1711 1712 /* 1713 * Invalidate access cache if uid, gid, mode or ctime changed. 1714 */ 1715 if (np->n_accstamp != -1 && 1716 (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode 1717 || timespeccmp(&ctime, &vap->va_ctime, !=))) 1718 np->n_accstamp = -1; 1719 1720 vap->va_type = vtyp; 1721 vap->va_mode = vmode; 1722 vap->va_rdev = (dev_t)rdev; 1723 vap->va_mtime = mtime; 1724 vap->va_ctime = ctime; 1725 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 1726 switch (vtyp) { 1727 case VDIR: 1728 vap->va_blocksize = NFS_DIRFRAGSIZ; 1729 break; 1730 case VBLK: 1731 vap->va_blocksize = BLKDEV_IOSIZE; 1732 break; 1733 case VCHR: 1734 vap->va_blocksize = MAXBSIZE; 1735 break; 1736 default: 1737 vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize : 1738 fxdr_unsigned(int32_t, fp->fa2_blocksize); 1739 break; 1740 } 1741 if (v3) { 1742 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1743 vap->va_uid = uid; 1744 vap->va_gid = gid; 1745 vap->va_size = fxdr_hyper(&fp->fa3_size); 1746 vap->va_bytes = fxdr_hyper(&fp->fa3_used); 1747 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid); 1748 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); 1749 vap->va_flags = 0; 1750 vap->va_filerev = 0; 1751 } else { 1752 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1753 vap->va_uid = uid; 1754 vap->va_gid = gid; 1755 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); 1756 vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks) 1757 * NFS_FABLKSIZE; 1758 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); 1759 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); 1760 vap->va_flags = 0; 1761 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec); 1762 vap->va_filerev = 0; 1763 } 1764 if (vap->va_size != np->n_size) { 1765 if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) { 1766 vap->va_size = np->n_size; 1767 } else { 1768 np->n_size = vap->va_size; 1769 if (vap->va_type == VREG) { 1770 /* 1771 * we can't free pages if NAC_NOTRUNC because 1772 * the pages can be owned by ourselves. 1773 */ 1774 if (flags & NAC_NOTRUNC) { 1775 np->n_flag |= NTRUNCDELAYED; 1776 } else { 1777 genfs_node_wrlock(vp); 1778 mutex_enter(&vp->v_interlock); 1779 (void)VOP_PUTPAGES(vp, 0, 1780 0, PGO_SYNCIO | PGO_CLEANIT | 1781 PGO_FREE | PGO_ALLPAGES); 1782 uvm_vnp_setsize(vp, np->n_size); 1783 genfs_node_unlock(vp); 1784 } 1785 } 1786 } 1787 } 1788 np->n_attrstamp = time_second; 1789 if (vaper != NULL) { 1790 memcpy((void *)vaper, (void *)vap, sizeof(*vap)); 1791 if (np->n_flag & NCHG) { 1792 if (np->n_flag & NACC) 1793 vaper->va_atime = np->n_atim; 1794 if (np->n_flag & NUPD) 1795 vaper->va_mtime = np->n_mtim; 1796 } 1797 } 1798 return (0); 1799 } 1800 1801 /* 1802 * Check the time stamp 1803 * If the cache is valid, copy contents to *vap and return 0 1804 * otherwise return an error 1805 */ 1806 int 1807 nfs_getattrcache(vp, vaper) 1808 struct vnode *vp; 1809 struct vattr *vaper; 1810 { 1811 struct nfsnode *np = VTONFS(vp); 1812 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1813 struct vattr *vap; 1814 1815 if (np->n_attrstamp == 0 || 1816 (time_second - np->n_attrstamp) >= NFS_ATTRTIMEO(nmp, np)) { 1817 nfsstats.attrcache_misses++; 1818 return (ENOENT); 1819 } 1820 nfsstats.attrcache_hits++; 1821 vap = np->n_vattr; 1822 if (vap->va_size != np->n_size) { 1823 if (vap->va_type == VREG) { 1824 if ((np->n_flag & NMODIFIED) != 0 && 1825 vap->va_size < np->n_size) { 1826 vap->va_size = np->n_size; 1827 } else { 1828 np->n_size = vap->va_size; 1829 } 1830 genfs_node_wrlock(vp); 1831 uvm_vnp_setsize(vp, np->n_size); 1832 genfs_node_unlock(vp); 1833 } else 1834 np->n_size = vap->va_size; 1835 } 1836 memcpy((void *)vaper, (void *)vap, sizeof(struct vattr)); 1837 if (np->n_flag & NCHG) { 1838 if (np->n_flag & NACC) 1839 vaper->va_atime = np->n_atim; 1840 if (np->n_flag & NUPD) 1841 vaper->va_mtime = np->n_mtim; 1842 } 1843 return (0); 1844 } 1845 1846 void 1847 nfs_delayedtruncate(vp) 1848 struct vnode *vp; 1849 { 1850 struct nfsnode *np = VTONFS(vp); 1851 1852 if (np->n_flag & NTRUNCDELAYED) { 1853 np->n_flag &= ~NTRUNCDELAYED; 1854 genfs_node_wrlock(vp); 1855 mutex_enter(&vp->v_interlock); 1856 (void)VOP_PUTPAGES(vp, 0, 1857 0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES); 1858 uvm_vnp_setsize(vp, np->n_size); 1859 genfs_node_unlock(vp); 1860 } 1861 } 1862 1863 #define NFS_WCCKLUDGE_TIMEOUT (24 * 60 * 60) /* 1 day */ 1864 #define NFS_WCCKLUDGE(nmp, now) \ 1865 (((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \ 1866 ((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0) 1867 1868 /* 1869 * nfs_check_wccdata: check inaccurate wcc_data 1870 * 1871 * => return non-zero if we shouldn't trust the wcc_data. 1872 * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed". 1873 */ 1874 1875 int 1876 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime, 1877 struct timespec *mtime, bool docheck) 1878 { 1879 int error = 0; 1880 1881 #if !defined(NFS_V2_ONLY) 1882 1883 if (docheck) { 1884 struct vnode *vp = NFSTOV(np); 1885 struct nfsmount *nmp; 1886 long now = time_second; 1887 const struct timespec *omtime = &np->n_vattr->va_mtime; 1888 const struct timespec *octime = &np->n_vattr->va_ctime; 1889 #if defined(DEBUG) 1890 const char *reason = NULL; /* XXX: gcc */ 1891 #endif 1892 1893 if (timespeccmp(omtime, mtime, <=)) { 1894 #if defined(DEBUG) 1895 reason = "mtime"; 1896 #endif 1897 error = EINVAL; 1898 } 1899 1900 if (vp->v_type == VDIR && timespeccmp(octime, ctime, <=)) { 1901 #if defined(DEBUG) 1902 reason = "ctime"; 1903 #endif 1904 error = EINVAL; 1905 } 1906 1907 nmp = VFSTONFS(vp->v_mount); 1908 if (error) { 1909 1910 /* 1911 * despite of the fact that we've updated the file, 1912 * timestamps of the file were not updated as we 1913 * expected. 1914 * it means that the server has incompatible 1915 * semantics of timestamps or (more likely) 1916 * the server time is not precise enough to 1917 * track each modifications. 1918 * in that case, we disable wcc processing. 1919 * 1920 * yes, strictly speaking, we should disable all 1921 * caching. it's a compromise. 1922 */ 1923 1924 mutex_enter(&nmp->nm_lock); 1925 #if defined(DEBUG) 1926 if (!NFS_WCCKLUDGE(nmp, now)) { 1927 printf("%s: inaccurate wcc data (%s) detected," 1928 " disabling wcc" 1929 " (ctime %u.%09u %u.%09u," 1930 " mtime %u.%09u %u.%09u)\n", 1931 vp->v_mount->mnt_stat.f_mntfromname, 1932 reason, 1933 (unsigned int)octime->tv_sec, 1934 (unsigned int)octime->tv_nsec, 1935 (unsigned int)ctime->tv_sec, 1936 (unsigned int)ctime->tv_nsec, 1937 (unsigned int)omtime->tv_sec, 1938 (unsigned int)omtime->tv_nsec, 1939 (unsigned int)mtime->tv_sec, 1940 (unsigned int)mtime->tv_nsec); 1941 } 1942 #endif 1943 nmp->nm_iflag |= NFSMNT_WCCKLUDGE; 1944 nmp->nm_wcckludgetime = now; 1945 mutex_exit(&nmp->nm_lock); 1946 } else if (NFS_WCCKLUDGE(nmp, now)) { 1947 error = EPERM; /* XXX */ 1948 } else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1949 mutex_enter(&nmp->nm_lock); 1950 if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1951 #if defined(DEBUG) 1952 printf("%s: re-enabling wcc\n", 1953 vp->v_mount->mnt_stat.f_mntfromname); 1954 #endif 1955 nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE; 1956 } 1957 mutex_exit(&nmp->nm_lock); 1958 } 1959 } 1960 1961 #endif /* !defined(NFS_V2_ONLY) */ 1962 1963 return error; 1964 } 1965 1966 /* 1967 * Heuristic to see if the server XDR encodes directory cookies or not. 1968 * it is not supposed to, but a lot of servers may do this. Also, since 1969 * most/all servers will implement V2 as well, it is expected that they 1970 * may return just 32 bits worth of cookie information, so we need to 1971 * find out in which 32 bits this information is available. We do this 1972 * to avoid trouble with emulated binaries that can't handle 64 bit 1973 * directory offsets. 1974 */ 1975 1976 void 1977 nfs_cookieheuristic(vp, flagp, l, cred) 1978 struct vnode *vp; 1979 int *flagp; 1980 struct lwp *l; 1981 kauth_cred_t cred; 1982 { 1983 struct uio auio; 1984 struct iovec aiov; 1985 char *tbuf, *cp; 1986 struct dirent *dp; 1987 off_t *cookies = NULL, *cop; 1988 int error, eof, nc, len; 1989 1990 MALLOC(tbuf, void *, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK); 1991 1992 aiov.iov_base = tbuf; 1993 aiov.iov_len = NFS_DIRFRAGSIZ; 1994 auio.uio_iov = &aiov; 1995 auio.uio_iovcnt = 1; 1996 auio.uio_rw = UIO_READ; 1997 auio.uio_resid = NFS_DIRFRAGSIZ; 1998 auio.uio_offset = 0; 1999 UIO_SETUP_SYSSPACE(&auio); 2000 2001 error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc); 2002 2003 len = NFS_DIRFRAGSIZ - auio.uio_resid; 2004 if (error || len == 0) { 2005 FREE(tbuf, M_TEMP); 2006 if (cookies) 2007 free(cookies, M_TEMP); 2008 return; 2009 } 2010 2011 /* 2012 * Find the first valid entry and look at its offset cookie. 2013 */ 2014 2015 cp = tbuf; 2016 for (cop = cookies; len > 0; len -= dp->d_reclen) { 2017 dp = (struct dirent *)cp; 2018 if (dp->d_fileno != 0 && len >= dp->d_reclen) { 2019 if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) { 2020 *flagp |= NFSMNT_SWAPCOOKIE; 2021 nfs_invaldircache(vp, 0); 2022 nfs_vinvalbuf(vp, 0, cred, l, 1); 2023 } 2024 break; 2025 } 2026 cop++; 2027 cp += dp->d_reclen; 2028 } 2029 2030 FREE(tbuf, M_TEMP); 2031 free(cookies, M_TEMP); 2032 } 2033 #endif /* NFS */ 2034 2035 #ifdef NFSSERVER 2036 /* 2037 * Set up nameidata for a lookup() call and do it. 2038 * 2039 * If pubflag is set, this call is done for a lookup operation on the 2040 * public filehandle. In that case we allow crossing mountpoints and 2041 * absolute pathnames. However, the caller is expected to check that 2042 * the lookup result is within the public fs, and deny access if 2043 * it is not. 2044 */ 2045 int 2046 nfs_namei(ndp, nsfh, len, slp, nam, mdp, dposp, retdirp, l, kerbflag, pubflag) 2047 struct nameidata *ndp; 2048 nfsrvfh_t *nsfh; 2049 uint32_t len; 2050 struct nfssvc_sock *slp; 2051 struct mbuf *nam; 2052 struct mbuf **mdp; 2053 char **dposp; 2054 struct vnode **retdirp; 2055 struct lwp *l; 2056 int kerbflag, pubflag; 2057 { 2058 int i, rem; 2059 struct mbuf *md; 2060 char *fromcp, *tocp, *cp; 2061 struct iovec aiov; 2062 struct uio auio; 2063 struct vnode *dp; 2064 int error, rdonly, linklen; 2065 struct componentname *cnp = &ndp->ni_cnd; 2066 2067 *retdirp = NULL; 2068 2069 if ((len + 1) > MAXPATHLEN) 2070 return (ENAMETOOLONG); 2071 if (len == 0) 2072 return (EACCES); 2073 cnp->cn_pnbuf = PNBUF_GET(); 2074 2075 /* 2076 * Copy the name from the mbuf list to ndp->ni_pnbuf 2077 * and set the various ndp fields appropriately. 2078 */ 2079 fromcp = *dposp; 2080 tocp = cnp->cn_pnbuf; 2081 md = *mdp; 2082 rem = mtod(md, char *) + md->m_len - fromcp; 2083 for (i = 0; i < len; i++) { 2084 while (rem == 0) { 2085 md = md->m_next; 2086 if (md == NULL) { 2087 error = EBADRPC; 2088 goto out; 2089 } 2090 fromcp = mtod(md, void *); 2091 rem = md->m_len; 2092 } 2093 if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) { 2094 error = EACCES; 2095 goto out; 2096 } 2097 *tocp++ = *fromcp++; 2098 rem--; 2099 } 2100 *tocp = '\0'; 2101 *mdp = md; 2102 *dposp = fromcp; 2103 len = nfsm_rndup(len)-len; 2104 if (len > 0) { 2105 if (rem >= len) 2106 *dposp += len; 2107 else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0) 2108 goto out; 2109 } 2110 2111 /* 2112 * Extract and set starting directory. 2113 */ 2114 error = nfsrv_fhtovp(nsfh, false, &dp, ndp->ni_cnd.cn_cred, slp, 2115 nam, &rdonly, kerbflag, pubflag); 2116 if (error) 2117 goto out; 2118 if (dp->v_type != VDIR) { 2119 vrele(dp); 2120 error = ENOTDIR; 2121 goto out; 2122 } 2123 2124 if (rdonly) 2125 cnp->cn_flags |= RDONLY; 2126 2127 *retdirp = dp; 2128 2129 if (pubflag) { 2130 /* 2131 * Oh joy. For WebNFS, handle those pesky '%' escapes, 2132 * and the 'native path' indicator. 2133 */ 2134 cp = PNBUF_GET(); 2135 fromcp = cnp->cn_pnbuf; 2136 tocp = cp; 2137 if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) { 2138 switch ((unsigned char)*fromcp) { 2139 case WEBNFS_NATIVE_CHAR: 2140 /* 2141 * 'Native' path for us is the same 2142 * as a path according to the NFS spec, 2143 * just skip the escape char. 2144 */ 2145 fromcp++; 2146 break; 2147 /* 2148 * More may be added in the future, range 0x80-0xff 2149 */ 2150 default: 2151 error = EIO; 2152 vrele(dp); 2153 PNBUF_PUT(cp); 2154 goto out; 2155 } 2156 } 2157 /* 2158 * Translate the '%' escapes, URL-style. 2159 */ 2160 while (*fromcp != '\0') { 2161 if (*fromcp == WEBNFS_ESC_CHAR) { 2162 if (fromcp[1] != '\0' && fromcp[2] != '\0') { 2163 fromcp++; 2164 *tocp++ = HEXSTRTOI(fromcp); 2165 fromcp += 2; 2166 continue; 2167 } else { 2168 error = ENOENT; 2169 vrele(dp); 2170 PNBUF_PUT(cp); 2171 goto out; 2172 } 2173 } else 2174 *tocp++ = *fromcp++; 2175 } 2176 *tocp = '\0'; 2177 PNBUF_PUT(cnp->cn_pnbuf); 2178 cnp->cn_pnbuf = cp; 2179 } 2180 2181 ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1; 2182 ndp->ni_segflg = UIO_SYSSPACE; 2183 ndp->ni_rootdir = rootvnode; 2184 ndp->ni_erootdir = NULL; 2185 2186 if (pubflag) { 2187 ndp->ni_loopcnt = 0; 2188 if (cnp->cn_pnbuf[0] == '/') 2189 dp = rootvnode; 2190 } else { 2191 cnp->cn_flags |= NOCROSSMOUNT; 2192 } 2193 2194 cnp->cn_lwp = l; 2195 VREF(dp); 2196 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); 2197 2198 for (;;) { 2199 cnp->cn_nameptr = cnp->cn_pnbuf; 2200 ndp->ni_startdir = dp; 2201 2202 /* 2203 * And call lookup() to do the real work 2204 */ 2205 error = lookup(ndp); 2206 if (error) { 2207 if (ndp->ni_dvp) { 2208 vput(ndp->ni_dvp); 2209 } 2210 PNBUF_PUT(cnp->cn_pnbuf); 2211 return (error); 2212 } 2213 2214 /* 2215 * Check for encountering a symbolic link 2216 */ 2217 if ((cnp->cn_flags & ISSYMLINK) == 0) { 2218 if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp) { 2219 if (ndp->ni_dvp == ndp->ni_vp) { 2220 vrele(ndp->ni_dvp); 2221 } else { 2222 vput(ndp->ni_dvp); 2223 } 2224 } 2225 if (cnp->cn_flags & (SAVENAME | SAVESTART)) 2226 cnp->cn_flags |= HASBUF; 2227 else 2228 PNBUF_PUT(cnp->cn_pnbuf); 2229 return (0); 2230 } else { 2231 if (!pubflag) { 2232 error = EINVAL; 2233 break; 2234 } 2235 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 2236 error = ELOOP; 2237 break; 2238 } 2239 if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) { 2240 error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred, 2241 cnp->cn_lwp); 2242 if (error != 0) 2243 break; 2244 } 2245 if (ndp->ni_pathlen > 1) 2246 cp = PNBUF_GET(); 2247 else 2248 cp = cnp->cn_pnbuf; 2249 aiov.iov_base = cp; 2250 aiov.iov_len = MAXPATHLEN; 2251 auio.uio_iov = &aiov; 2252 auio.uio_iovcnt = 1; 2253 auio.uio_offset = 0; 2254 auio.uio_rw = UIO_READ; 2255 auio.uio_resid = MAXPATHLEN; 2256 UIO_SETUP_SYSSPACE(&auio); 2257 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 2258 if (error) { 2259 badlink: 2260 if (ndp->ni_pathlen > 1) 2261 PNBUF_PUT(cp); 2262 break; 2263 } 2264 linklen = MAXPATHLEN - auio.uio_resid; 2265 if (linklen == 0) { 2266 error = ENOENT; 2267 goto badlink; 2268 } 2269 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 2270 error = ENAMETOOLONG; 2271 goto badlink; 2272 } 2273 if (ndp->ni_pathlen > 1) { 2274 memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen); 2275 PNBUF_PUT(cnp->cn_pnbuf); 2276 cnp->cn_pnbuf = cp; 2277 } else 2278 cnp->cn_pnbuf[linklen] = '\0'; 2279 ndp->ni_pathlen += linklen; 2280 vput(ndp->ni_vp); 2281 dp = ndp->ni_dvp; 2282 2283 /* 2284 * Check if root directory should replace current directory. 2285 */ 2286 if (cnp->cn_pnbuf[0] == '/') { 2287 vput(dp); 2288 dp = ndp->ni_rootdir; 2289 VREF(dp); 2290 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); 2291 } 2292 } 2293 } 2294 vput(ndp->ni_dvp); 2295 vput(ndp->ni_vp); 2296 ndp->ni_vp = NULL; 2297 out: 2298 PNBUF_PUT(cnp->cn_pnbuf); 2299 return (error); 2300 } 2301 #endif /* NFSSERVER */ 2302 2303 /* 2304 * A fiddled version of m_adj() that ensures null fill to a 32-bit 2305 * boundary and only trims off the back end 2306 * 2307 * 1. trim off 'len' bytes as m_adj(mp, -len). 2308 * 2. add zero-padding 'nul' bytes at the end of the mbuf chain. 2309 */ 2310 void 2311 nfs_zeropad(mp, len, nul) 2312 struct mbuf *mp; 2313 int len; 2314 int nul; 2315 { 2316 struct mbuf *m; 2317 int count; 2318 2319 /* 2320 * Trim from tail. Scan the mbuf chain, 2321 * calculating its length and finding the last mbuf. 2322 * If the adjustment only affects this mbuf, then just 2323 * adjust and return. Otherwise, rescan and truncate 2324 * after the remaining size. 2325 */ 2326 count = 0; 2327 m = mp; 2328 for (;;) { 2329 count += m->m_len; 2330 if (m->m_next == NULL) 2331 break; 2332 m = m->m_next; 2333 } 2334 2335 KDASSERT(count >= len); 2336 2337 if (m->m_len >= len) { 2338 m->m_len -= len; 2339 } else { 2340 count -= len; 2341 /* 2342 * Correct length for chain is "count". 2343 * Find the mbuf with last data, adjust its length, 2344 * and toss data from remaining mbufs on chain. 2345 */ 2346 for (m = mp; m; m = m->m_next) { 2347 if (m->m_len >= count) { 2348 m->m_len = count; 2349 break; 2350 } 2351 count -= m->m_len; 2352 } 2353 KASSERT(m && m->m_next); 2354 m_freem(m->m_next); 2355 m->m_next = NULL; 2356 } 2357 2358 KDASSERT(m->m_next == NULL); 2359 2360 /* 2361 * zero-padding. 2362 */ 2363 if (nul > 0) { 2364 char *cp; 2365 int i; 2366 2367 if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) { 2368 struct mbuf *n; 2369 2370 KDASSERT(MLEN >= nul); 2371 n = m_get(M_WAIT, MT_DATA); 2372 MCLAIM(n, &nfs_mowner); 2373 n->m_len = nul; 2374 n->m_next = NULL; 2375 m->m_next = n; 2376 cp = mtod(n, void *); 2377 } else { 2378 cp = mtod(m, char *) + m->m_len; 2379 m->m_len += nul; 2380 } 2381 for (i = 0; i < nul; i++) 2382 *cp++ = '\0'; 2383 } 2384 return; 2385 } 2386 2387 /* 2388 * Make these functions instead of macros, so that the kernel text size 2389 * doesn't get too big... 2390 */ 2391 void 2392 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp) 2393 struct nfsrv_descript *nfsd; 2394 int before_ret; 2395 struct vattr *before_vap; 2396 int after_ret; 2397 struct vattr *after_vap; 2398 struct mbuf **mbp; 2399 char **bposp; 2400 { 2401 struct mbuf *mb = *mbp; 2402 char *bpos = *bposp; 2403 u_int32_t *tl; 2404 2405 if (before_ret) { 2406 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2407 *tl = nfs_false; 2408 } else { 2409 nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED); 2410 *tl++ = nfs_true; 2411 txdr_hyper(before_vap->va_size, tl); 2412 tl += 2; 2413 txdr_nfsv3time(&(before_vap->va_mtime), tl); 2414 tl += 2; 2415 txdr_nfsv3time(&(before_vap->va_ctime), tl); 2416 } 2417 *bposp = bpos; 2418 *mbp = mb; 2419 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp); 2420 } 2421 2422 void 2423 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp) 2424 struct nfsrv_descript *nfsd; 2425 int after_ret; 2426 struct vattr *after_vap; 2427 struct mbuf **mbp; 2428 char **bposp; 2429 { 2430 struct mbuf *mb = *mbp; 2431 char *bpos = *bposp; 2432 u_int32_t *tl; 2433 struct nfs_fattr *fp; 2434 2435 if (after_ret) { 2436 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2437 *tl = nfs_false; 2438 } else { 2439 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR); 2440 *tl++ = nfs_true; 2441 fp = (struct nfs_fattr *)tl; 2442 nfsm_srvfattr(nfsd, after_vap, fp); 2443 } 2444 *mbp = mb; 2445 *bposp = bpos; 2446 } 2447 2448 void 2449 nfsm_srvfattr(nfsd, vap, fp) 2450 struct nfsrv_descript *nfsd; 2451 struct vattr *vap; 2452 struct nfs_fattr *fp; 2453 { 2454 2455 fp->fa_nlink = txdr_unsigned(vap->va_nlink); 2456 fp->fa_uid = txdr_unsigned(vap->va_uid); 2457 fp->fa_gid = txdr_unsigned(vap->va_gid); 2458 if (nfsd->nd_flag & ND_NFSV3) { 2459 fp->fa_type = vtonfsv3_type(vap->va_type); 2460 fp->fa_mode = vtonfsv3_mode(vap->va_mode); 2461 txdr_hyper(vap->va_size, &fp->fa3_size); 2462 txdr_hyper(vap->va_bytes, &fp->fa3_used); 2463 fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev)); 2464 fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev)); 2465 fp->fa3_fsid.nfsuquad[0] = 0; 2466 fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid); 2467 txdr_hyper(vap->va_fileid, &fp->fa3_fileid); 2468 txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime); 2469 txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime); 2470 txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime); 2471 } else { 2472 fp->fa_type = vtonfsv2_type(vap->va_type); 2473 fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 2474 fp->fa2_size = txdr_unsigned(vap->va_size); 2475 fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize); 2476 if (vap->va_type == VFIFO) 2477 fp->fa2_rdev = 0xffffffff; 2478 else 2479 fp->fa2_rdev = txdr_unsigned(vap->va_rdev); 2480 fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE); 2481 fp->fa2_fsid = txdr_unsigned(vap->va_fsid); 2482 fp->fa2_fileid = txdr_unsigned(vap->va_fileid); 2483 txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime); 2484 txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime); 2485 txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime); 2486 } 2487 } 2488 2489 #ifdef NFSSERVER 2490 /* 2491 * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked) 2492 * - look up fsid in mount list (if not found ret error) 2493 * - get vp and export rights by calling VFS_FHTOVP() 2494 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 2495 * - if not lockflag unlock it with VOP_UNLOCK() 2496 */ 2497 int 2498 nfsrv_fhtovp(nfsrvfh_t *nsfh, int lockflag, struct vnode **vpp, 2499 kauth_cred_t cred, struct nfssvc_sock *slp, struct mbuf *nam, int *rdonlyp, 2500 int kerbflag, int pubflag) 2501 { 2502 struct mount *mp; 2503 kauth_cred_t credanon; 2504 int error, exflags; 2505 struct sockaddr_in *saddr; 2506 fhandle_t *fhp; 2507 2508 fhp = NFSRVFH_FHANDLE(nsfh); 2509 *vpp = (struct vnode *)0; 2510 2511 if (nfs_ispublicfh(nsfh)) { 2512 if (!pubflag || !nfs_pub.np_valid) 2513 return (ESTALE); 2514 fhp = nfs_pub.np_handle; 2515 } 2516 2517 error = netexport_check(&fhp->fh_fsid, nam, &mp, &exflags, &credanon); 2518 if (error) { 2519 return error; 2520 } 2521 2522 error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp); 2523 if (error) 2524 return (error); 2525 2526 if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) { 2527 saddr = mtod(nam, struct sockaddr_in *); 2528 if ((saddr->sin_family == AF_INET) && 2529 ntohs(saddr->sin_port) >= IPPORT_RESERVED) { 2530 vput(*vpp); 2531 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2532 } 2533 #ifdef INET6 2534 if ((saddr->sin_family == AF_INET6) && 2535 ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) { 2536 vput(*vpp); 2537 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2538 } 2539 #endif 2540 } 2541 /* 2542 * Check/setup credentials. 2543 */ 2544 if (exflags & MNT_EXKERB) { 2545 if (!kerbflag) { 2546 vput(*vpp); 2547 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2548 } 2549 } else if (kerbflag) { 2550 vput(*vpp); 2551 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2552 } else if (kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 2553 NULL) == 0 || (exflags & MNT_EXPORTANON)) { 2554 kauth_cred_clone(credanon, cred); 2555 } 2556 if (exflags & MNT_EXRDONLY) 2557 *rdonlyp = 1; 2558 else 2559 *rdonlyp = 0; 2560 if (!lockflag) 2561 VOP_UNLOCK(*vpp, 0); 2562 return (0); 2563 } 2564 2565 /* 2566 * WebNFS: check if a filehandle is a public filehandle. For v3, this 2567 * means a length of 0, for v2 it means all zeroes. 2568 */ 2569 int 2570 nfs_ispublicfh(const nfsrvfh_t *nsfh) 2571 { 2572 const char *cp = (const void *)(NFSRVFH_DATA(nsfh)); 2573 int i; 2574 2575 if (NFSRVFH_SIZE(nsfh) == 0) { 2576 return true; 2577 } 2578 if (NFSRVFH_SIZE(nsfh) != NFSX_V2FH) { 2579 return false; 2580 } 2581 for (i = 0; i < NFSX_V2FH; i++) 2582 if (*cp++ != 0) 2583 return false; 2584 return true; 2585 } 2586 #endif /* NFSSERVER */ 2587 2588 /* 2589 * This function compares two net addresses by family and returns true 2590 * if they are the same host. 2591 * If there is any doubt, return false. 2592 * The AF_INET family is handled as a special case so that address mbufs 2593 * don't need to be saved to store "struct in_addr", which is only 4 bytes. 2594 */ 2595 int 2596 netaddr_match(family, haddr, nam) 2597 int family; 2598 union nethostaddr *haddr; 2599 struct mbuf *nam; 2600 { 2601 struct sockaddr_in *inetaddr; 2602 2603 switch (family) { 2604 case AF_INET: 2605 inetaddr = mtod(nam, struct sockaddr_in *); 2606 if (inetaddr->sin_family == AF_INET && 2607 inetaddr->sin_addr.s_addr == haddr->had_inetaddr) 2608 return (1); 2609 break; 2610 #ifdef INET6 2611 case AF_INET6: 2612 { 2613 struct sockaddr_in6 *sin6_1, *sin6_2; 2614 2615 sin6_1 = mtod(nam, struct sockaddr_in6 *); 2616 sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *); 2617 if (sin6_1->sin6_family == AF_INET6 && 2618 IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr)) 2619 return 1; 2620 } 2621 #endif 2622 #ifdef ISO 2623 case AF_ISO: 2624 { 2625 struct sockaddr_iso *isoaddr1, *isoaddr2; 2626 2627 isoaddr1 = mtod(nam, struct sockaddr_iso *); 2628 isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *); 2629 if (isoaddr1->siso_family == AF_ISO && 2630 isoaddr1->siso_nlen > 0 && 2631 isoaddr1->siso_nlen == isoaddr2->siso_nlen && 2632 SAME_ISOADDR(isoaddr1, isoaddr2)) 2633 return (1); 2634 break; 2635 } 2636 #endif /* ISO */ 2637 default: 2638 break; 2639 }; 2640 return (0); 2641 } 2642 2643 /* 2644 * The write verifier has changed (probably due to a server reboot), so all 2645 * PG_NEEDCOMMIT pages will have to be written again. Since they are marked 2646 * as dirty or are being written out just now, all this takes is clearing 2647 * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for 2648 * the mount point. 2649 */ 2650 void 2651 nfs_clearcommit(mp) 2652 struct mount *mp; 2653 { 2654 struct vnode *vp; 2655 struct nfsnode *np; 2656 struct vm_page *pg; 2657 struct nfsmount *nmp = VFSTONFS(mp); 2658 2659 rw_enter(&nmp->nm_writeverflock, RW_WRITER); 2660 2661 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2662 KASSERT(vp->v_mount == mp); 2663 if (vp->v_type != VREG) 2664 continue; 2665 np = VTONFS(vp); 2666 np->n_pushlo = np->n_pushhi = np->n_pushedlo = 2667 np->n_pushedhi = 0; 2668 np->n_commitflags &= 2669 ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID); 2670 mutex_enter(&vp->v_uobj.vmobjlock); 2671 TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) { 2672 pg->flags &= ~PG_NEEDCOMMIT; 2673 } 2674 mutex_exit(&vp->v_uobj.vmobjlock); 2675 } 2676 mutex_enter(&nmp->nm_lock); 2677 nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF; 2678 mutex_exit(&nmp->nm_lock); 2679 rw_exit(&nmp->nm_writeverflock); 2680 } 2681 2682 void 2683 nfs_merge_commit_ranges(vp) 2684 struct vnode *vp; 2685 { 2686 struct nfsnode *np = VTONFS(vp); 2687 2688 KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID); 2689 2690 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2691 np->n_pushedlo = np->n_pushlo; 2692 np->n_pushedhi = np->n_pushhi; 2693 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2694 } else { 2695 if (np->n_pushlo < np->n_pushedlo) 2696 np->n_pushedlo = np->n_pushlo; 2697 if (np->n_pushhi > np->n_pushedhi) 2698 np->n_pushedhi = np->n_pushhi; 2699 } 2700 2701 np->n_pushlo = np->n_pushhi = 0; 2702 np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID; 2703 2704 #ifdef NFS_DEBUG_COMMIT 2705 printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2706 (unsigned)np->n_pushedhi); 2707 #endif 2708 } 2709 2710 int 2711 nfs_in_committed_range(vp, off, len) 2712 struct vnode *vp; 2713 off_t off, len; 2714 { 2715 struct nfsnode *np = VTONFS(vp); 2716 off_t lo, hi; 2717 2718 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2719 return 0; 2720 lo = off; 2721 hi = lo + len; 2722 2723 return (lo >= np->n_pushedlo && hi <= np->n_pushedhi); 2724 } 2725 2726 int 2727 nfs_in_tobecommitted_range(vp, off, len) 2728 struct vnode *vp; 2729 off_t off, len; 2730 { 2731 struct nfsnode *np = VTONFS(vp); 2732 off_t lo, hi; 2733 2734 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2735 return 0; 2736 lo = off; 2737 hi = lo + len; 2738 2739 return (lo >= np->n_pushlo && hi <= np->n_pushhi); 2740 } 2741 2742 void 2743 nfs_add_committed_range(vp, off, len) 2744 struct vnode *vp; 2745 off_t off, len; 2746 { 2747 struct nfsnode *np = VTONFS(vp); 2748 off_t lo, hi; 2749 2750 lo = off; 2751 hi = lo + len; 2752 2753 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2754 np->n_pushedlo = lo; 2755 np->n_pushedhi = hi; 2756 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2757 } else { 2758 if (hi > np->n_pushedhi) 2759 np->n_pushedhi = hi; 2760 if (lo < np->n_pushedlo) 2761 np->n_pushedlo = lo; 2762 } 2763 #ifdef NFS_DEBUG_COMMIT 2764 printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2765 (unsigned)np->n_pushedhi); 2766 #endif 2767 } 2768 2769 void 2770 nfs_del_committed_range(vp, off, len) 2771 struct vnode *vp; 2772 off_t off, len; 2773 { 2774 struct nfsnode *np = VTONFS(vp); 2775 off_t lo, hi; 2776 2777 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2778 return; 2779 2780 lo = off; 2781 hi = lo + len; 2782 2783 if (lo > np->n_pushedhi || hi < np->n_pushedlo) 2784 return; 2785 if (lo <= np->n_pushedlo) 2786 np->n_pushedlo = hi; 2787 else if (hi >= np->n_pushedhi) 2788 np->n_pushedhi = lo; 2789 else { 2790 /* 2791 * XXX There's only one range. If the deleted range 2792 * is in the middle, pick the largest of the 2793 * contiguous ranges that it leaves. 2794 */ 2795 if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi)) 2796 np->n_pushedhi = lo; 2797 else 2798 np->n_pushedlo = hi; 2799 } 2800 #ifdef NFS_DEBUG_COMMIT 2801 printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2802 (unsigned)np->n_pushedhi); 2803 #endif 2804 } 2805 2806 void 2807 nfs_add_tobecommitted_range(vp, off, len) 2808 struct vnode *vp; 2809 off_t off, len; 2810 { 2811 struct nfsnode *np = VTONFS(vp); 2812 off_t lo, hi; 2813 2814 lo = off; 2815 hi = lo + len; 2816 2817 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) { 2818 np->n_pushlo = lo; 2819 np->n_pushhi = hi; 2820 np->n_commitflags |= NFS_COMMIT_PUSH_VALID; 2821 } else { 2822 if (lo < np->n_pushlo) 2823 np->n_pushlo = lo; 2824 if (hi > np->n_pushhi) 2825 np->n_pushhi = hi; 2826 } 2827 #ifdef NFS_DEBUG_COMMIT 2828 printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2829 (unsigned)np->n_pushhi); 2830 #endif 2831 } 2832 2833 void 2834 nfs_del_tobecommitted_range(vp, off, len) 2835 struct vnode *vp; 2836 off_t off, len; 2837 { 2838 struct nfsnode *np = VTONFS(vp); 2839 off_t lo, hi; 2840 2841 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2842 return; 2843 2844 lo = off; 2845 hi = lo + len; 2846 2847 if (lo > np->n_pushhi || hi < np->n_pushlo) 2848 return; 2849 2850 if (lo <= np->n_pushlo) 2851 np->n_pushlo = hi; 2852 else if (hi >= np->n_pushhi) 2853 np->n_pushhi = lo; 2854 else { 2855 /* 2856 * XXX There's only one range. If the deleted range 2857 * is in the middle, pick the largest of the 2858 * contiguous ranges that it leaves. 2859 */ 2860 if ((np->n_pushlo - lo) > (hi - np->n_pushhi)) 2861 np->n_pushhi = lo; 2862 else 2863 np->n_pushlo = hi; 2864 } 2865 #ifdef NFS_DEBUG_COMMIT 2866 printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2867 (unsigned)np->n_pushhi); 2868 #endif 2869 } 2870 2871 /* 2872 * Map errnos to NFS error numbers. For Version 3 also filter out error 2873 * numbers not specified for the associated procedure. 2874 */ 2875 int 2876 nfsrv_errmap(nd, err) 2877 struct nfsrv_descript *nd; 2878 int err; 2879 { 2880 const short *defaulterrp, *errp; 2881 2882 if (nd->nd_flag & ND_NFSV3) { 2883 if (nd->nd_procnum <= NFSPROC_COMMIT) { 2884 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum]; 2885 while (*++errp) { 2886 if (*errp == err) 2887 return (err); 2888 else if (*errp > err) 2889 break; 2890 } 2891 return ((int)*defaulterrp); 2892 } else 2893 return (err & 0xffff); 2894 } 2895 if (err <= ELAST) 2896 return ((int)nfsrv_v2errmap[err - 1]); 2897 return (NFSERR_IO); 2898 } 2899 2900 u_int32_t 2901 nfs_getxid() 2902 { 2903 static u_int32_t base; 2904 static u_int32_t nfs_xid = 0; 2905 static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER; 2906 u_int32_t newxid; 2907 2908 simple_lock(&nfs_xidlock); 2909 /* 2910 * derive initial xid from system time 2911 * XXX time is invalid if root not yet mounted 2912 */ 2913 if (__predict_false(!base && (rootvp))) { 2914 struct timeval tv; 2915 2916 microtime(&tv); 2917 base = tv.tv_sec << 12; 2918 nfs_xid = base; 2919 } 2920 2921 /* 2922 * Skip zero xid if it should ever happen. 2923 */ 2924 if (__predict_false(++nfs_xid == 0)) 2925 nfs_xid++; 2926 newxid = nfs_xid; 2927 simple_unlock(&nfs_xidlock); 2928 2929 return txdr_unsigned(newxid); 2930 } 2931 2932 /* 2933 * assign a new xid for existing request. 2934 * used for NFSERR_JUKEBOX handling. 2935 */ 2936 void 2937 nfs_renewxid(struct nfsreq *req) 2938 { 2939 u_int32_t xid; 2940 int off; 2941 2942 xid = nfs_getxid(); 2943 if (req->r_nmp->nm_sotype == SOCK_STREAM) 2944 off = sizeof(u_int32_t); /* RPC record mark */ 2945 else 2946 off = 0; 2947 2948 m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid); 2949 req->r_xid = xid; 2950 } 2951 2952 #if defined(NFSSERVER) 2953 int 2954 nfsrv_composefh(struct vnode *vp, nfsrvfh_t *nsfh, bool v3) 2955 { 2956 int error; 2957 size_t fhsize; 2958 2959 fhsize = NFSD_MAXFHSIZE; 2960 error = vfs_composefh(vp, (void *)NFSRVFH_DATA(nsfh), &fhsize); 2961 if (NFSX_FHTOOBIG_P(fhsize, v3)) { 2962 error = EOPNOTSUPP; 2963 } 2964 if (error != 0) { 2965 return error; 2966 } 2967 if (!v3 && fhsize < NFSX_V2FH) { 2968 memset((char *)NFSRVFH_DATA(nsfh) + fhsize, 0, 2969 NFSX_V2FH - fhsize); 2970 fhsize = NFSX_V2FH; 2971 } 2972 if ((fhsize % NFSX_UNSIGNED) != 0) { 2973 return EOPNOTSUPP; 2974 } 2975 nsfh->nsfh_size = fhsize; 2976 return 0; 2977 } 2978 2979 int 2980 nfsrv_comparefh(const nfsrvfh_t *fh1, const nfsrvfh_t *fh2) 2981 { 2982 2983 if (NFSRVFH_SIZE(fh1) != NFSRVFH_SIZE(fh2)) { 2984 return NFSRVFH_SIZE(fh2) - NFSRVFH_SIZE(fh1); 2985 } 2986 return memcmp(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), NFSRVFH_SIZE(fh1)); 2987 } 2988 2989 void 2990 nfsrv_copyfh(nfsrvfh_t *fh1, const nfsrvfh_t *fh2) 2991 { 2992 size_t size; 2993 2994 fh1->nsfh_size = size = NFSRVFH_SIZE(fh2); 2995 memcpy(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), size); 2996 } 2997 #endif /* defined(NFSSERVER) */ 2998
Indexes created Thu Oct 02 14:10:14 GMT 2025