nfs_subs.c revision 1.196
1 /* $NetBSD: nfs_subs.c,v 1.196 2008/01/02 19:26:46 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 1.196 2008/01/02 19:26:46 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/kmem.h> 91 #include <sys/mount.h> 92 #include <sys/vnode.h> 93 #include <sys/namei.h> 94 #include <sys/mbuf.h> 95 #include <sys/socket.h> 96 #include <sys/stat.h> 97 #include <sys/malloc.h> 98 #include <sys/filedesc.h> 99 #include <sys/time.h> 100 #include <sys/dirent.h> 101 #include <sys/once.h> 102 #include <sys/kauth.h> 103 104 #include <uvm/uvm_extern.h> 105 106 #include <nfs/rpcv2.h> 107 #include <nfs/nfsproto.h> 108 #include <nfs/nfsnode.h> 109 #include <nfs/nfs.h> 110 #include <nfs/xdr_subs.h> 111 #include <nfs/nfsm_subs.h> 112 #include <nfs/nfsmount.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, 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 }; 178 179 /* 180 * and the reverse mapping from generic to Version 2 procedure numbers 181 */ 182 const int nfsv2_procid[NFS_NPROCS] = { 183 NFSV2PROC_NULL, 184 NFSV2PROC_GETATTR, 185 NFSV2PROC_SETATTR, 186 NFSV2PROC_LOOKUP, 187 NFSV2PROC_NOOP, 188 NFSV2PROC_READLINK, 189 NFSV2PROC_READ, 190 NFSV2PROC_WRITE, 191 NFSV2PROC_CREATE, 192 NFSV2PROC_MKDIR, 193 NFSV2PROC_SYMLINK, 194 NFSV2PROC_CREATE, 195 NFSV2PROC_REMOVE, 196 NFSV2PROC_RMDIR, 197 NFSV2PROC_RENAME, 198 NFSV2PROC_LINK, 199 NFSV2PROC_READDIR, 200 NFSV2PROC_NOOP, 201 NFSV2PROC_STATFS, 202 NFSV2PROC_NOOP, 203 NFSV2PROC_NOOP, 204 NFSV2PROC_NOOP, 205 NFSV2PROC_NOOP, 206 }; 207 208 /* 209 * Maps errno values to nfs error numbers. 210 * Use NFSERR_IO as the catch all for ones not specifically defined in 211 * RFC 1094. 212 */ 213 static const u_char nfsrv_v2errmap[ELAST] = { 214 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 215 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 216 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 217 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 218 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 219 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 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_IO, NFSERR_IO, NFSERR_IO, 226 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO, 227 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 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, 231 }; 232 233 /* 234 * Maps errno values to nfs error numbers. 235 * Although it is not obvious whether or not NFS clients really care if 236 * a returned error value is in the specified list for the procedure, the 237 * safest thing to do is filter them appropriately. For Version 2, the 238 * X/Open XNFS document is the only specification that defines error values 239 * for each RPC (The RFC simply lists all possible error values for all RPCs), 240 * so I have decided to not do this for Version 2. 241 * The first entry is the default error return and the rest are the valid 242 * errors for that RPC in increasing numeric order. 243 */ 244 static const short nfsv3err_null[] = { 245 0, 246 0, 247 }; 248 249 static const short nfsv3err_getattr[] = { 250 NFSERR_IO, 251 NFSERR_IO, 252 NFSERR_STALE, 253 NFSERR_BADHANDLE, 254 NFSERR_SERVERFAULT, 255 0, 256 }; 257 258 static const short nfsv3err_setattr[] = { 259 NFSERR_IO, 260 NFSERR_PERM, 261 NFSERR_IO, 262 NFSERR_ACCES, 263 NFSERR_INVAL, 264 NFSERR_NOSPC, 265 NFSERR_ROFS, 266 NFSERR_DQUOT, 267 NFSERR_STALE, 268 NFSERR_BADHANDLE, 269 NFSERR_NOT_SYNC, 270 NFSERR_SERVERFAULT, 271 0, 272 }; 273 274 static const short nfsv3err_lookup[] = { 275 NFSERR_IO, 276 NFSERR_NOENT, 277 NFSERR_IO, 278 NFSERR_ACCES, 279 NFSERR_NOTDIR, 280 NFSERR_NAMETOL, 281 NFSERR_STALE, 282 NFSERR_BADHANDLE, 283 NFSERR_SERVERFAULT, 284 0, 285 }; 286 287 static const short nfsv3err_access[] = { 288 NFSERR_IO, 289 NFSERR_IO, 290 NFSERR_STALE, 291 NFSERR_BADHANDLE, 292 NFSERR_SERVERFAULT, 293 0, 294 }; 295 296 static const short nfsv3err_readlink[] = { 297 NFSERR_IO, 298 NFSERR_IO, 299 NFSERR_ACCES, 300 NFSERR_INVAL, 301 NFSERR_STALE, 302 NFSERR_BADHANDLE, 303 NFSERR_NOTSUPP, 304 NFSERR_SERVERFAULT, 305 0, 306 }; 307 308 static const short nfsv3err_read[] = { 309 NFSERR_IO, 310 NFSERR_IO, 311 NFSERR_NXIO, 312 NFSERR_ACCES, 313 NFSERR_INVAL, 314 NFSERR_STALE, 315 NFSERR_BADHANDLE, 316 NFSERR_SERVERFAULT, 317 NFSERR_JUKEBOX, 318 0, 319 }; 320 321 static const short nfsv3err_write[] = { 322 NFSERR_IO, 323 NFSERR_IO, 324 NFSERR_ACCES, 325 NFSERR_INVAL, 326 NFSERR_FBIG, 327 NFSERR_NOSPC, 328 NFSERR_ROFS, 329 NFSERR_DQUOT, 330 NFSERR_STALE, 331 NFSERR_BADHANDLE, 332 NFSERR_SERVERFAULT, 333 NFSERR_JUKEBOX, 334 0, 335 }; 336 337 static const short nfsv3err_create[] = { 338 NFSERR_IO, 339 NFSERR_IO, 340 NFSERR_ACCES, 341 NFSERR_EXIST, 342 NFSERR_NOTDIR, 343 NFSERR_NOSPC, 344 NFSERR_ROFS, 345 NFSERR_NAMETOL, 346 NFSERR_DQUOT, 347 NFSERR_STALE, 348 NFSERR_BADHANDLE, 349 NFSERR_NOTSUPP, 350 NFSERR_SERVERFAULT, 351 0, 352 }; 353 354 static const short nfsv3err_mkdir[] = { 355 NFSERR_IO, 356 NFSERR_IO, 357 NFSERR_ACCES, 358 NFSERR_EXIST, 359 NFSERR_NOTDIR, 360 NFSERR_NOSPC, 361 NFSERR_ROFS, 362 NFSERR_NAMETOL, 363 NFSERR_DQUOT, 364 NFSERR_STALE, 365 NFSERR_BADHANDLE, 366 NFSERR_NOTSUPP, 367 NFSERR_SERVERFAULT, 368 0, 369 }; 370 371 static const short nfsv3err_symlink[] = { 372 NFSERR_IO, 373 NFSERR_IO, 374 NFSERR_ACCES, 375 NFSERR_EXIST, 376 NFSERR_NOTDIR, 377 NFSERR_NOSPC, 378 NFSERR_ROFS, 379 NFSERR_NAMETOL, 380 NFSERR_DQUOT, 381 NFSERR_STALE, 382 NFSERR_BADHANDLE, 383 NFSERR_NOTSUPP, 384 NFSERR_SERVERFAULT, 385 0, 386 }; 387 388 static const short nfsv3err_mknod[] = { 389 NFSERR_IO, 390 NFSERR_IO, 391 NFSERR_ACCES, 392 NFSERR_EXIST, 393 NFSERR_NOTDIR, 394 NFSERR_NOSPC, 395 NFSERR_ROFS, 396 NFSERR_NAMETOL, 397 NFSERR_DQUOT, 398 NFSERR_STALE, 399 NFSERR_BADHANDLE, 400 NFSERR_NOTSUPP, 401 NFSERR_SERVERFAULT, 402 NFSERR_BADTYPE, 403 0, 404 }; 405 406 static const short nfsv3err_remove[] = { 407 NFSERR_IO, 408 NFSERR_NOENT, 409 NFSERR_IO, 410 NFSERR_ACCES, 411 NFSERR_NOTDIR, 412 NFSERR_ROFS, 413 NFSERR_NAMETOL, 414 NFSERR_STALE, 415 NFSERR_BADHANDLE, 416 NFSERR_SERVERFAULT, 417 0, 418 }; 419 420 static const short nfsv3err_rmdir[] = { 421 NFSERR_IO, 422 NFSERR_NOENT, 423 NFSERR_IO, 424 NFSERR_ACCES, 425 NFSERR_EXIST, 426 NFSERR_NOTDIR, 427 NFSERR_INVAL, 428 NFSERR_ROFS, 429 NFSERR_NAMETOL, 430 NFSERR_NOTEMPTY, 431 NFSERR_STALE, 432 NFSERR_BADHANDLE, 433 NFSERR_NOTSUPP, 434 NFSERR_SERVERFAULT, 435 0, 436 }; 437 438 static const short nfsv3err_rename[] = { 439 NFSERR_IO, 440 NFSERR_NOENT, 441 NFSERR_IO, 442 NFSERR_ACCES, 443 NFSERR_EXIST, 444 NFSERR_XDEV, 445 NFSERR_NOTDIR, 446 NFSERR_ISDIR, 447 NFSERR_INVAL, 448 NFSERR_NOSPC, 449 NFSERR_ROFS, 450 NFSERR_MLINK, 451 NFSERR_NAMETOL, 452 NFSERR_NOTEMPTY, 453 NFSERR_DQUOT, 454 NFSERR_STALE, 455 NFSERR_BADHANDLE, 456 NFSERR_NOTSUPP, 457 NFSERR_SERVERFAULT, 458 0, 459 }; 460 461 static const short nfsv3err_link[] = { 462 NFSERR_IO, 463 NFSERR_IO, 464 NFSERR_ACCES, 465 NFSERR_EXIST, 466 NFSERR_XDEV, 467 NFSERR_NOTDIR, 468 NFSERR_INVAL, 469 NFSERR_NOSPC, 470 NFSERR_ROFS, 471 NFSERR_MLINK, 472 NFSERR_NAMETOL, 473 NFSERR_DQUOT, 474 NFSERR_STALE, 475 NFSERR_BADHANDLE, 476 NFSERR_NOTSUPP, 477 NFSERR_SERVERFAULT, 478 0, 479 }; 480 481 static const short nfsv3err_readdir[] = { 482 NFSERR_IO, 483 NFSERR_IO, 484 NFSERR_ACCES, 485 NFSERR_NOTDIR, 486 NFSERR_STALE, 487 NFSERR_BADHANDLE, 488 NFSERR_BAD_COOKIE, 489 NFSERR_TOOSMALL, 490 NFSERR_SERVERFAULT, 491 0, 492 }; 493 494 static const short nfsv3err_readdirplus[] = { 495 NFSERR_IO, 496 NFSERR_IO, 497 NFSERR_ACCES, 498 NFSERR_NOTDIR, 499 NFSERR_STALE, 500 NFSERR_BADHANDLE, 501 NFSERR_BAD_COOKIE, 502 NFSERR_NOTSUPP, 503 NFSERR_TOOSMALL, 504 NFSERR_SERVERFAULT, 505 0, 506 }; 507 508 static const short nfsv3err_fsstat[] = { 509 NFSERR_IO, 510 NFSERR_IO, 511 NFSERR_STALE, 512 NFSERR_BADHANDLE, 513 NFSERR_SERVERFAULT, 514 0, 515 }; 516 517 static const short nfsv3err_fsinfo[] = { 518 NFSERR_STALE, 519 NFSERR_STALE, 520 NFSERR_BADHANDLE, 521 NFSERR_SERVERFAULT, 522 0, 523 }; 524 525 static const short nfsv3err_pathconf[] = { 526 NFSERR_STALE, 527 NFSERR_STALE, 528 NFSERR_BADHANDLE, 529 NFSERR_SERVERFAULT, 530 0, 531 }; 532 533 static const short nfsv3err_commit[] = { 534 NFSERR_IO, 535 NFSERR_IO, 536 NFSERR_STALE, 537 NFSERR_BADHANDLE, 538 NFSERR_SERVERFAULT, 539 0, 540 }; 541 542 static const short * const nfsrv_v3errmap[] = { 543 nfsv3err_null, 544 nfsv3err_getattr, 545 nfsv3err_setattr, 546 nfsv3err_lookup, 547 nfsv3err_access, 548 nfsv3err_readlink, 549 nfsv3err_read, 550 nfsv3err_write, 551 nfsv3err_create, 552 nfsv3err_mkdir, 553 nfsv3err_symlink, 554 nfsv3err_mknod, 555 nfsv3err_remove, 556 nfsv3err_rmdir, 557 nfsv3err_rename, 558 nfsv3err_link, 559 nfsv3err_readdir, 560 nfsv3err_readdirplus, 561 nfsv3err_fsstat, 562 nfsv3err_fsinfo, 563 nfsv3err_pathconf, 564 nfsv3err_commit, 565 }; 566 567 extern struct nfsrtt nfsrtt; 568 extern struct nfsnodehashhead *nfsnodehashtbl; 569 extern u_long nfsnodehash; 570 571 u_long nfsdirhashmask; 572 573 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *)); 574 575 /* 576 * Create the header for an rpc request packet 577 * The hsiz is the size of the rest of the nfs request header. 578 * (just used to decide if a cluster is a good idea) 579 */ 580 struct mbuf * 581 nfsm_reqh(struct nfsnode *np, u_long procid, int hsiz, char **bposp) 582 { 583 struct mbuf *mb; 584 char *bpos; 585 586 mb = m_get(M_WAIT, MT_DATA); 587 MCLAIM(mb, &nfs_mowner); 588 if (hsiz >= MINCLSIZE) 589 m_clget(mb, M_WAIT); 590 mb->m_len = 0; 591 bpos = mtod(mb, void *); 592 593 /* Finally, return values */ 594 *bposp = bpos; 595 return (mb); 596 } 597 598 /* 599 * Build the RPC header and fill in the authorization info. 600 * The authorization string argument is only used when the credentials 601 * come from outside of the kernel. 602 * Returns the head of the mbuf list. 603 */ 604 struct mbuf * 605 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len, 606 verf_str, mrest, mrest_len, mbp, xidp) 607 kauth_cred_t cr; 608 int nmflag; 609 int procid; 610 int auth_type; 611 int auth_len; 612 char *auth_str; 613 int verf_len; 614 char *verf_str; 615 struct mbuf *mrest; 616 int mrest_len; 617 struct mbuf **mbp; 618 u_int32_t *xidp; 619 { 620 struct mbuf *mb; 621 u_int32_t *tl; 622 char *bpos; 623 int i; 624 struct mbuf *mreq; 625 int siz, grpsiz, authsiz; 626 627 authsiz = nfsm_rndup(auth_len); 628 mb = m_gethdr(M_WAIT, MT_DATA); 629 MCLAIM(mb, &nfs_mowner); 630 if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) { 631 m_clget(mb, M_WAIT); 632 } else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) { 633 MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED); 634 } else { 635 MH_ALIGN(mb, 8 * NFSX_UNSIGNED); 636 } 637 mb->m_len = 0; 638 mreq = mb; 639 bpos = mtod(mb, void *); 640 641 /* 642 * First the RPC header. 643 */ 644 nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 645 646 *tl++ = *xidp = nfs_getxid(); 647 *tl++ = rpc_call; 648 *tl++ = rpc_vers; 649 *tl++ = txdr_unsigned(NFS_PROG); 650 if (nmflag & NFSMNT_NFSV3) 651 *tl++ = txdr_unsigned(NFS_VER3); 652 else 653 *tl++ = txdr_unsigned(NFS_VER2); 654 if (nmflag & NFSMNT_NFSV3) 655 *tl++ = txdr_unsigned(procid); 656 else 657 *tl++ = txdr_unsigned(nfsv2_procid[procid]); 658 659 /* 660 * And then the authorization cred. 661 */ 662 *tl++ = txdr_unsigned(auth_type); 663 *tl = txdr_unsigned(authsiz); 664 switch (auth_type) { 665 case RPCAUTH_UNIX: 666 nfsm_build(tl, u_int32_t *, auth_len); 667 *tl++ = 0; /* stamp ?? */ 668 *tl++ = 0; /* NULL hostname */ 669 *tl++ = txdr_unsigned(kauth_cred_geteuid(cr)); 670 *tl++ = txdr_unsigned(kauth_cred_getegid(cr)); 671 grpsiz = (auth_len >> 2) - 5; 672 *tl++ = txdr_unsigned(grpsiz); 673 for (i = 0; i < grpsiz; i++) 674 *tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */ 675 break; 676 case RPCAUTH_KERB4: 677 siz = auth_len; 678 while (siz > 0) { 679 if (M_TRAILINGSPACE(mb) == 0) { 680 struct mbuf *mb2; 681 mb2 = m_get(M_WAIT, MT_DATA); 682 MCLAIM(mb2, &nfs_mowner); 683 if (siz >= MINCLSIZE) 684 m_clget(mb2, M_WAIT); 685 mb->m_next = mb2; 686 mb = mb2; 687 mb->m_len = 0; 688 bpos = mtod(mb, void *); 689 } 690 i = min(siz, M_TRAILINGSPACE(mb)); 691 memcpy(bpos, auth_str, i); 692 mb->m_len += i; 693 auth_str += i; 694 bpos += i; 695 siz -= i; 696 } 697 if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) { 698 for (i = 0; i < siz; i++) 699 *bpos++ = '\0'; 700 mb->m_len += siz; 701 } 702 break; 703 }; 704 705 /* 706 * And the verifier... 707 */ 708 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 709 if (verf_str) { 710 *tl++ = txdr_unsigned(RPCAUTH_KERB4); 711 *tl = txdr_unsigned(verf_len); 712 siz = verf_len; 713 while (siz > 0) { 714 if (M_TRAILINGSPACE(mb) == 0) { 715 struct mbuf *mb2; 716 mb2 = m_get(M_WAIT, MT_DATA); 717 MCLAIM(mb2, &nfs_mowner); 718 if (siz >= MINCLSIZE) 719 m_clget(mb2, M_WAIT); 720 mb->m_next = mb2; 721 mb = mb2; 722 mb->m_len = 0; 723 bpos = mtod(mb, void *); 724 } 725 i = min(siz, M_TRAILINGSPACE(mb)); 726 memcpy(bpos, verf_str, i); 727 mb->m_len += i; 728 verf_str += i; 729 bpos += i; 730 siz -= i; 731 } 732 if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) { 733 for (i = 0; i < siz; i++) 734 *bpos++ = '\0'; 735 mb->m_len += siz; 736 } 737 } else { 738 *tl++ = txdr_unsigned(RPCAUTH_NULL); 739 *tl = 0; 740 } 741 mb->m_next = mrest; 742 mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len; 743 mreq->m_pkthdr.rcvif = (struct ifnet *)0; 744 *mbp = mb; 745 return (mreq); 746 } 747 748 /* 749 * copies mbuf chain to the uio scatter/gather list 750 */ 751 int 752 nfsm_mbuftouio(mrep, uiop, siz, dpos) 753 struct mbuf **mrep; 754 struct uio *uiop; 755 int siz; 756 char **dpos; 757 { 758 char *mbufcp, *uiocp; 759 int xfer, left, len; 760 struct mbuf *mp; 761 long uiosiz, rem; 762 int error = 0; 763 764 mp = *mrep; 765 mbufcp = *dpos; 766 len = mtod(mp, char *) + mp->m_len - mbufcp; 767 rem = nfsm_rndup(siz)-siz; 768 while (siz > 0) { 769 if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) 770 return (EFBIG); 771 left = uiop->uio_iov->iov_len; 772 uiocp = uiop->uio_iov->iov_base; 773 if (left > siz) 774 left = siz; 775 uiosiz = left; 776 while (left > 0) { 777 while (len == 0) { 778 mp = mp->m_next; 779 if (mp == NULL) 780 return (EBADRPC); 781 mbufcp = mtod(mp, void *); 782 len = mp->m_len; 783 } 784 xfer = (left > len) ? len : left; 785 error = copyout_vmspace(uiop->uio_vmspace, mbufcp, 786 uiocp, xfer); 787 if (error) { 788 return error; 789 } 790 left -= xfer; 791 len -= xfer; 792 mbufcp += xfer; 793 uiocp += xfer; 794 uiop->uio_offset += xfer; 795 uiop->uio_resid -= xfer; 796 } 797 if (uiop->uio_iov->iov_len <= siz) { 798 uiop->uio_iovcnt--; 799 uiop->uio_iov++; 800 } else { 801 uiop->uio_iov->iov_base = 802 (char *)uiop->uio_iov->iov_base + uiosiz; 803 uiop->uio_iov->iov_len -= uiosiz; 804 } 805 siz -= uiosiz; 806 } 807 *dpos = mbufcp; 808 *mrep = mp; 809 if (rem > 0) { 810 if (len < rem) 811 error = nfs_adv(mrep, dpos, rem, len); 812 else 813 *dpos += rem; 814 } 815 return (error); 816 } 817 818 /* 819 * copies a uio scatter/gather list to an mbuf chain. 820 * NOTE: can ony handle iovcnt == 1 821 */ 822 int 823 nfsm_uiotombuf(uiop, mq, siz, bpos) 824 struct uio *uiop; 825 struct mbuf **mq; 826 int siz; 827 char **bpos; 828 { 829 char *uiocp; 830 struct mbuf *mp, *mp2; 831 int xfer, left, mlen; 832 int uiosiz, clflg, rem; 833 char *cp; 834 int error; 835 836 #ifdef DIAGNOSTIC 837 if (uiop->uio_iovcnt != 1) 838 panic("nfsm_uiotombuf: iovcnt != 1"); 839 #endif 840 841 if (siz > MLEN) /* or should it >= MCLBYTES ?? */ 842 clflg = 1; 843 else 844 clflg = 0; 845 rem = nfsm_rndup(siz)-siz; 846 mp = mp2 = *mq; 847 while (siz > 0) { 848 left = uiop->uio_iov->iov_len; 849 uiocp = uiop->uio_iov->iov_base; 850 if (left > siz) 851 left = siz; 852 uiosiz = left; 853 while (left > 0) { 854 mlen = M_TRAILINGSPACE(mp); 855 if (mlen == 0) { 856 mp = m_get(M_WAIT, MT_DATA); 857 MCLAIM(mp, &nfs_mowner); 858 if (clflg) 859 m_clget(mp, M_WAIT); 860 mp->m_len = 0; 861 mp2->m_next = mp; 862 mp2 = mp; 863 mlen = M_TRAILINGSPACE(mp); 864 } 865 xfer = (left > mlen) ? mlen : left; 866 cp = mtod(mp, char *) + mp->m_len; 867 error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp, 868 xfer); 869 if (error) { 870 /* XXX */ 871 } 872 mp->m_len += xfer; 873 left -= xfer; 874 uiocp += xfer; 875 uiop->uio_offset += xfer; 876 uiop->uio_resid -= xfer; 877 } 878 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + 879 uiosiz; 880 uiop->uio_iov->iov_len -= uiosiz; 881 siz -= uiosiz; 882 } 883 if (rem > 0) { 884 if (rem > M_TRAILINGSPACE(mp)) { 885 mp = m_get(M_WAIT, MT_DATA); 886 MCLAIM(mp, &nfs_mowner); 887 mp->m_len = 0; 888 mp2->m_next = mp; 889 } 890 cp = mtod(mp, char *) + mp->m_len; 891 for (left = 0; left < rem; left++) 892 *cp++ = '\0'; 893 mp->m_len += rem; 894 *bpos = cp; 895 } else 896 *bpos = mtod(mp, char *) + mp->m_len; 897 *mq = mp; 898 return (0); 899 } 900 901 /* 902 * Get at least "siz" bytes of correctly aligned data. 903 * When called the mbuf pointers are not necessarily correct, 904 * dsosp points to what ought to be in m_data and left contains 905 * what ought to be in m_len. 906 * This is used by the macros nfsm_dissect and nfsm_dissecton for tough 907 * cases. (The macros use the vars. dpos and dpos2) 908 */ 909 int 910 nfsm_disct(mdp, dposp, siz, left, cp2) 911 struct mbuf **mdp; 912 char **dposp; 913 int siz; 914 int left; 915 char **cp2; 916 { 917 struct mbuf *m1, *m2; 918 struct mbuf *havebuf = NULL; 919 char *src = *dposp; 920 char *dst; 921 int len; 922 923 #ifdef DEBUG 924 if (left < 0) 925 panic("nfsm_disct: left < 0"); 926 #endif 927 m1 = *mdp; 928 /* 929 * Skip through the mbuf chain looking for an mbuf with 930 * some data. If the first mbuf found has enough data 931 * and it is correctly aligned return it. 932 */ 933 while (left == 0) { 934 havebuf = m1; 935 *mdp = m1 = m1->m_next; 936 if (m1 == NULL) 937 return (EBADRPC); 938 src = mtod(m1, void *); 939 left = m1->m_len; 940 /* 941 * If we start a new mbuf and it is big enough 942 * and correctly aligned just return it, don't 943 * do any pull up. 944 */ 945 if (left >= siz && nfsm_aligned(src)) { 946 *cp2 = src; 947 *dposp = src + siz; 948 return (0); 949 } 950 } 951 if (m1->m_flags & M_EXT) { 952 if (havebuf) { 953 /* If the first mbuf with data has external data 954 * and there is a previous empty mbuf use it 955 * to move the data into. 956 */ 957 m2 = m1; 958 *mdp = m1 = havebuf; 959 if (m1->m_flags & M_EXT) { 960 MEXTREMOVE(m1); 961 } 962 } else { 963 /* 964 * If the first mbuf has a external data 965 * and there is no previous empty mbuf 966 * allocate a new mbuf and move the external 967 * data to the new mbuf. Also make the first 968 * mbuf look empty. 969 */ 970 m2 = m_get(M_WAIT, MT_DATA); 971 m2->m_ext = m1->m_ext; 972 m2->m_data = src; 973 m2->m_len = left; 974 MCLADDREFERENCE(m1, m2); 975 MEXTREMOVE(m1); 976 m2->m_next = m1->m_next; 977 m1->m_next = m2; 978 } 979 m1->m_len = 0; 980 if (m1->m_flags & M_PKTHDR) 981 dst = m1->m_pktdat; 982 else 983 dst = m1->m_dat; 984 m1->m_data = dst; 985 } else { 986 /* 987 * If the first mbuf has no external data 988 * move the data to the front of the mbuf. 989 */ 990 if (m1->m_flags & M_PKTHDR) 991 dst = m1->m_pktdat; 992 else 993 dst = m1->m_dat; 994 m1->m_data = dst; 995 if (dst != src) 996 memmove(dst, src, left); 997 dst += left; 998 m1->m_len = left; 999 m2 = m1->m_next; 1000 } 1001 *cp2 = m1->m_data; 1002 *dposp = mtod(m1, char *) + siz; 1003 /* 1004 * Loop through mbufs pulling data up into first mbuf until 1005 * the first mbuf is full or there is no more data to 1006 * pullup. 1007 */ 1008 while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) { 1009 if ((len = min(len, m2->m_len)) != 0) 1010 memcpy(dst, m2->m_data, len); 1011 m1->m_len += len; 1012 dst += len; 1013 m2->m_data += len; 1014 m2->m_len -= len; 1015 m2 = m2->m_next; 1016 } 1017 if (m1->m_len < siz) 1018 return (EBADRPC); 1019 return (0); 1020 } 1021 1022 /* 1023 * Advance the position in the mbuf chain. 1024 */ 1025 int 1026 nfs_adv(mdp, dposp, offs, left) 1027 struct mbuf **mdp; 1028 char **dposp; 1029 int offs; 1030 int left; 1031 { 1032 struct mbuf *m; 1033 int s; 1034 1035 m = *mdp; 1036 s = left; 1037 while (s < offs) { 1038 offs -= s; 1039 m = m->m_next; 1040 if (m == NULL) 1041 return (EBADRPC); 1042 s = m->m_len; 1043 } 1044 *mdp = m; 1045 *dposp = mtod(m, char *) + offs; 1046 return (0); 1047 } 1048 1049 /* 1050 * Copy a string into mbufs for the hard cases... 1051 */ 1052 int 1053 nfsm_strtmbuf(mb, bpos, cp, siz) 1054 struct mbuf **mb; 1055 char **bpos; 1056 const char *cp; 1057 long siz; 1058 { 1059 struct mbuf *m1 = NULL, *m2; 1060 long left, xfer, len, tlen; 1061 u_int32_t *tl; 1062 int putsize; 1063 1064 putsize = 1; 1065 m2 = *mb; 1066 left = M_TRAILINGSPACE(m2); 1067 if (left > 0) { 1068 tl = ((u_int32_t *)(*bpos)); 1069 *tl++ = txdr_unsigned(siz); 1070 putsize = 0; 1071 left -= NFSX_UNSIGNED; 1072 m2->m_len += NFSX_UNSIGNED; 1073 if (left > 0) { 1074 memcpy((void *) tl, cp, left); 1075 siz -= left; 1076 cp += left; 1077 m2->m_len += left; 1078 left = 0; 1079 } 1080 } 1081 /* Loop around adding mbufs */ 1082 while (siz > 0) { 1083 m1 = m_get(M_WAIT, MT_DATA); 1084 MCLAIM(m1, &nfs_mowner); 1085 if (siz > MLEN) 1086 m_clget(m1, M_WAIT); 1087 m1->m_len = NFSMSIZ(m1); 1088 m2->m_next = m1; 1089 m2 = m1; 1090 tl = mtod(m1, u_int32_t *); 1091 tlen = 0; 1092 if (putsize) { 1093 *tl++ = txdr_unsigned(siz); 1094 m1->m_len -= NFSX_UNSIGNED; 1095 tlen = NFSX_UNSIGNED; 1096 putsize = 0; 1097 } 1098 if (siz < m1->m_len) { 1099 len = nfsm_rndup(siz); 1100 xfer = siz; 1101 if (xfer < len) 1102 *(tl+(xfer>>2)) = 0; 1103 } else { 1104 xfer = len = m1->m_len; 1105 } 1106 memcpy((void *) tl, cp, xfer); 1107 m1->m_len = len+tlen; 1108 siz -= xfer; 1109 cp += xfer; 1110 } 1111 *mb = m1; 1112 *bpos = mtod(m1, char *) + m1->m_len; 1113 return (0); 1114 } 1115 1116 /* 1117 * Directory caching routines. They work as follows: 1118 * - a cache is maintained per VDIR nfsnode. 1119 * - for each offset cookie that is exported to userspace, and can 1120 * thus be thrown back at us as an offset to VOP_READDIR, store 1121 * information in the cache. 1122 * - cached are: 1123 * - cookie itself 1124 * - blocknumber (essentially just a search key in the buffer cache) 1125 * - entry number in block. 1126 * - offset cookie of block in which this entry is stored 1127 * - 32 bit cookie if NFSMNT_XLATECOOKIE is used. 1128 * - entries are looked up in a hash table 1129 * - also maintained is an LRU list of entries, used to determine 1130 * which ones to delete if the cache grows too large. 1131 * - if 32 <-> 64 translation mode is requested for a filesystem, 1132 * the cache also functions as a translation table 1133 * - in the translation case, invalidating the cache does not mean 1134 * flushing it, but just marking entries as invalid, except for 1135 * the <64bit cookie, 32bitcookie> pair which is still valid, to 1136 * still be able to use the cache as a translation table. 1137 * - 32 bit cookies are uniquely created by combining the hash table 1138 * entry value, and one generation count per hash table entry, 1139 * incremented each time an entry is appended to the chain. 1140 * - the cache is invalidated each time a direcory is modified 1141 * - sanity checks are also done; if an entry in a block turns 1142 * out not to have a matching cookie, the cache is invalidated 1143 * and a new block starting from the wanted offset is fetched from 1144 * the server. 1145 * - directory entries as read from the server are extended to contain 1146 * the 64bit and, optionally, the 32bit cookies, for sanity checking 1147 * the cache and exporting them to userspace through the cookie 1148 * argument to VOP_READDIR. 1149 */ 1150 1151 u_long 1152 nfs_dirhash(off) 1153 off_t off; 1154 { 1155 int i; 1156 char *cp = (char *)&off; 1157 u_long sum = 0L; 1158 1159 for (i = 0 ; i < sizeof (off); i++) 1160 sum += *cp++; 1161 1162 return sum; 1163 } 1164 1165 #define _NFSDC_MTX(np) (&NFSTOV(np)->v_interlock) 1166 #define NFSDC_LOCK(np) mutex_enter(_NFSDC_MTX(np)) 1167 #define NFSDC_UNLOCK(np) mutex_exit(_NFSDC_MTX(np)) 1168 #define NFSDC_ASSERT_LOCKED(np) KASSERT(mutex_owned(_NFSDC_MTX(np))) 1169 1170 void 1171 nfs_initdircache(vp) 1172 struct vnode *vp; 1173 { 1174 struct nfsnode *np = VTONFS(vp); 1175 struct nfsdirhashhead *dircache; 1176 1177 dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF, 1178 M_WAITOK, &nfsdirhashmask); 1179 1180 NFSDC_LOCK(np); 1181 if (np->n_dircache == NULL) { 1182 np->n_dircachesize = 0; 1183 np->n_dircache = dircache; 1184 dircache = NULL; 1185 TAILQ_INIT(&np->n_dirchain); 1186 } 1187 NFSDC_UNLOCK(np); 1188 if (dircache) 1189 hashdone(dircache, M_NFSDIROFF); 1190 } 1191 1192 void 1193 nfs_initdirxlatecookie(vp) 1194 struct vnode *vp; 1195 { 1196 struct nfsnode *np = VTONFS(vp); 1197 unsigned *dirgens; 1198 1199 KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE); 1200 1201 dirgens = kmem_zalloc(NFS_DIRHASHSIZ * sizeof(unsigned), KM_SLEEP); 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 kmem_free(dirgens, NFS_DIRHASHSIZ * sizeof(unsigned)); 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 kmem_free(ndp, sizeof(*ndp)); 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 kmem_free(ndp, sizeof(*ndp)); 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 = kmem_alloc(sizeof(*newndp), KM_SLEEP); 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 kmem_free(np->n_dirgens, 1477 NFS_DIRHASHSIZ * sizeof(unsigned)); 1478 np->n_dirgens = NULL; 1479 } 1480 } else { 1481 TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain) 1482 ndp->dc_flags |= NFSDC_INVALID; 1483 } 1484 1485 NFSDC_UNLOCK(np); 1486 } 1487 1488 /* 1489 * Called once before VFS init to initialize shared and 1490 * server-specific data structures. 1491 */ 1492 static int 1493 nfs_init0(void) 1494 { 1495 1496 nfsrtt.pos = 0; 1497 rpc_vers = txdr_unsigned(RPC_VER2); 1498 rpc_call = txdr_unsigned(RPC_CALL); 1499 rpc_reply = txdr_unsigned(RPC_REPLY); 1500 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED); 1501 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED); 1502 rpc_mismatch = txdr_unsigned(RPC_MISMATCH); 1503 rpc_autherr = txdr_unsigned(RPC_AUTHERR); 1504 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX); 1505 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4); 1506 nfs_prog = txdr_unsigned(NFS_PROG); 1507 nfs_true = txdr_unsigned(true); 1508 nfs_false = txdr_unsigned(false); 1509 nfs_xdrneg1 = txdr_unsigned(-1); 1510 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; 1511 if (nfs_ticks < 1) 1512 nfs_ticks = 1; 1513 #ifdef NFSSERVER 1514 nfsrv_init(0); /* Init server data structures */ 1515 nfsrv_initcache(); /* Init the server request cache */ 1516 { 1517 extern krwlock_t netexport_lock; /* XXX */ 1518 rw_init(&netexport_lock); 1519 } 1520 #endif /* NFSSERVER */ 1521 1522 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) 1523 nfsdreq_init(); 1524 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */ 1525 1526 /* 1527 * Initialize reply list and start timer 1528 */ 1529 TAILQ_INIT(&nfs_reqq); 1530 nfs_timer_init(); 1531 MOWNER_ATTACH(&nfs_mowner); 1532 1533 #ifdef NFS 1534 /* Initialize the kqueue structures */ 1535 nfs_kqinit(); 1536 /* Initialize the iod structures */ 1537 nfs_iodinit(); 1538 #endif 1539 return 0; 1540 } 1541 1542 void 1543 nfs_init(void) 1544 { 1545 static ONCE_DECL(nfs_init_once); 1546 1547 RUN_ONCE(&nfs_init_once, nfs_init0); 1548 } 1549 1550 #ifdef NFS 1551 /* 1552 * Called once at VFS init to initialize client-specific data structures. 1553 */ 1554 void 1555 nfs_vfs_init() 1556 { 1557 /* Initialize NFS server / client shared data. */ 1558 nfs_init(); 1559 1560 nfs_nhinit(); /* Init the nfsnode table */ 1561 nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4); 1562 } 1563 1564 void 1565 nfs_vfs_reinit() 1566 { 1567 nfs_nhreinit(); 1568 } 1569 1570 void 1571 nfs_vfs_done() 1572 { 1573 nfs_nhdone(); 1574 } 1575 1576 /* 1577 * Attribute cache routines. 1578 * nfs_loadattrcache() - loads or updates the cache contents from attributes 1579 * that are on the mbuf list 1580 * nfs_getattrcache() - returns valid attributes if found in cache, returns 1581 * error otherwise 1582 */ 1583 1584 /* 1585 * Load the attribute cache (that lives in the nfsnode entry) with 1586 * the values on the mbuf list and 1587 * Iff vap not NULL 1588 * copy the attributes to *vaper 1589 */ 1590 int 1591 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags) 1592 struct vnode **vpp; 1593 struct mbuf **mdp; 1594 char **dposp; 1595 struct vattr *vaper; 1596 int flags; 1597 { 1598 int32_t t1; 1599 char *cp2; 1600 int error = 0; 1601 struct mbuf *md; 1602 int v3 = NFS_ISV3(*vpp); 1603 1604 md = *mdp; 1605 t1 = (mtod(md, char *) + md->m_len) - *dposp; 1606 error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2); 1607 if (error) 1608 return (error); 1609 return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags); 1610 } 1611 1612 int 1613 nfs_loadattrcache(vpp, fp, vaper, flags) 1614 struct vnode **vpp; 1615 struct nfs_fattr *fp; 1616 struct vattr *vaper; 1617 int flags; 1618 { 1619 struct vnode *vp = *vpp; 1620 struct vattr *vap; 1621 int v3 = NFS_ISV3(vp); 1622 enum vtype vtyp; 1623 u_short vmode; 1624 struct timespec mtime; 1625 struct timespec ctime; 1626 struct vnode *nvp; 1627 int32_t rdev; 1628 struct nfsnode *np; 1629 extern int (**spec_nfsv2nodeop_p) __P((void *)); 1630 uid_t uid; 1631 gid_t gid; 1632 1633 if (v3) { 1634 vtyp = nfsv3tov_type(fp->fa_type); 1635 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1636 rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1), 1637 fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2)); 1638 fxdr_nfsv3time(&fp->fa3_mtime, &mtime); 1639 fxdr_nfsv3time(&fp->fa3_ctime, &ctime); 1640 } else { 1641 vtyp = nfsv2tov_type(fp->fa_type); 1642 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1643 if (vtyp == VNON || vtyp == VREG) 1644 vtyp = IFTOVT(vmode); 1645 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); 1646 fxdr_nfsv2time(&fp->fa2_mtime, &mtime); 1647 ctime.tv_sec = fxdr_unsigned(u_int32_t, 1648 fp->fa2_ctime.nfsv2_sec); 1649 ctime.tv_nsec = 0; 1650 1651 /* 1652 * Really ugly NFSv2 kludge. 1653 */ 1654 if (vtyp == VCHR && rdev == 0xffffffff) 1655 vtyp = VFIFO; 1656 } 1657 1658 vmode &= ALLPERMS; 1659 1660 /* 1661 * If v_type == VNON it is a new node, so fill in the v_type, 1662 * n_mtime fields. Check to see if it represents a special 1663 * device, and if so, check for a possible alias. Once the 1664 * correct vnode has been obtained, fill in the rest of the 1665 * information. 1666 */ 1667 np = VTONFS(vp); 1668 if (vp->v_type == VNON) { 1669 vp->v_type = vtyp; 1670 if (vp->v_type == VFIFO) { 1671 extern int (**fifo_nfsv2nodeop_p) __P((void *)); 1672 vp->v_op = fifo_nfsv2nodeop_p; 1673 } else if (vp->v_type == VREG) { 1674 mutex_init(&np->n_commitlock, MUTEX_DEFAULT, IPL_NONE); 1675 } else if (vp->v_type == VCHR || vp->v_type == VBLK) { 1676 vp->v_op = spec_nfsv2nodeop_p; 1677 nvp = checkalias(vp, (dev_t)rdev, vp->v_mount); 1678 if (nvp) { 1679 /* 1680 * Discard unneeded vnode, but save its nfsnode. 1681 * Since the nfsnode does not have a lock, its 1682 * vnode lock has to be carried over. 1683 */ 1684 /* 1685 * XXX is the old node sure to be locked here? 1686 */ 1687 KASSERT(lockstatus(&vp->v_lock) == 1688 LK_EXCLUSIVE); 1689 nvp->v_data = vp->v_data; 1690 vp->v_data = NULL; 1691 VOP_UNLOCK(vp, 0); 1692 vp->v_op = spec_vnodeop_p; 1693 vgone(vp); 1694 lockmgr(&nvp->v_lock, LK_EXCLUSIVE, 1695 &nvp->v_interlock); 1696 /* 1697 * Reinitialize aliased node. 1698 */ 1699 np->n_vnode = nvp; 1700 *vpp = vp = nvp; 1701 } 1702 } 1703 np->n_mtime = mtime; 1704 } 1705 uid = fxdr_unsigned(uid_t, fp->fa_uid); 1706 gid = fxdr_unsigned(gid_t, fp->fa_gid); 1707 vap = np->n_vattr; 1708 1709 /* 1710 * Invalidate access cache if uid, gid, mode or ctime changed. 1711 */ 1712 if (np->n_accstamp != -1 && 1713 (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode 1714 || timespeccmp(&ctime, &vap->va_ctime, !=))) 1715 np->n_accstamp = -1; 1716 1717 vap->va_type = vtyp; 1718 vap->va_mode = vmode; 1719 vap->va_rdev = (dev_t)rdev; 1720 vap->va_mtime = mtime; 1721 vap->va_ctime = ctime; 1722 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 1723 switch (vtyp) { 1724 case VDIR: 1725 vap->va_blocksize = NFS_DIRFRAGSIZ; 1726 break; 1727 case VBLK: 1728 vap->va_blocksize = BLKDEV_IOSIZE; 1729 break; 1730 case VCHR: 1731 vap->va_blocksize = MAXBSIZE; 1732 break; 1733 default: 1734 vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize : 1735 fxdr_unsigned(int32_t, fp->fa2_blocksize); 1736 break; 1737 } 1738 if (v3) { 1739 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1740 vap->va_uid = uid; 1741 vap->va_gid = gid; 1742 vap->va_size = fxdr_hyper(&fp->fa3_size); 1743 vap->va_bytes = fxdr_hyper(&fp->fa3_used); 1744 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid); 1745 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); 1746 vap->va_flags = 0; 1747 vap->va_filerev = 0; 1748 } else { 1749 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1750 vap->va_uid = uid; 1751 vap->va_gid = gid; 1752 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); 1753 vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks) 1754 * NFS_FABLKSIZE; 1755 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); 1756 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); 1757 vap->va_flags = 0; 1758 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec); 1759 vap->va_filerev = 0; 1760 } 1761 if (vap->va_size != np->n_size) { 1762 if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) { 1763 vap->va_size = np->n_size; 1764 } else { 1765 np->n_size = vap->va_size; 1766 if (vap->va_type == VREG) { 1767 /* 1768 * we can't free pages if NAC_NOTRUNC because 1769 * the pages can be owned by ourselves. 1770 */ 1771 if (flags & NAC_NOTRUNC) { 1772 np->n_flag |= NTRUNCDELAYED; 1773 } else { 1774 genfs_node_wrlock(vp); 1775 mutex_enter(&vp->v_interlock); 1776 (void)VOP_PUTPAGES(vp, 0, 1777 0, PGO_SYNCIO | PGO_CLEANIT | 1778 PGO_FREE | PGO_ALLPAGES); 1779 uvm_vnp_setsize(vp, np->n_size); 1780 genfs_node_unlock(vp); 1781 } 1782 } 1783 } 1784 } 1785 np->n_attrstamp = time_second; 1786 if (vaper != NULL) { 1787 memcpy((void *)vaper, (void *)vap, sizeof(*vap)); 1788 if (np->n_flag & NCHG) { 1789 if (np->n_flag & NACC) 1790 vaper->va_atime = np->n_atim; 1791 if (np->n_flag & NUPD) 1792 vaper->va_mtime = np->n_mtim; 1793 } 1794 } 1795 return (0); 1796 } 1797 1798 /* 1799 * Check the time stamp 1800 * If the cache is valid, copy contents to *vap and return 0 1801 * otherwise return an error 1802 */ 1803 int 1804 nfs_getattrcache(vp, vaper) 1805 struct vnode *vp; 1806 struct vattr *vaper; 1807 { 1808 struct nfsnode *np = VTONFS(vp); 1809 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1810 struct vattr *vap; 1811 1812 if (np->n_attrstamp == 0 || 1813 (time_second - np->n_attrstamp) >= nfs_attrtimeo(nmp, np)) { 1814 nfsstats.attrcache_misses++; 1815 return (ENOENT); 1816 } 1817 nfsstats.attrcache_hits++; 1818 vap = np->n_vattr; 1819 if (vap->va_size != np->n_size) { 1820 if (vap->va_type == VREG) { 1821 if ((np->n_flag & NMODIFIED) != 0 && 1822 vap->va_size < np->n_size) { 1823 vap->va_size = np->n_size; 1824 } else { 1825 np->n_size = vap->va_size; 1826 } 1827 genfs_node_wrlock(vp); 1828 uvm_vnp_setsize(vp, np->n_size); 1829 genfs_node_unlock(vp); 1830 } else 1831 np->n_size = vap->va_size; 1832 } 1833 memcpy((void *)vaper, (void *)vap, sizeof(struct vattr)); 1834 if (np->n_flag & NCHG) { 1835 if (np->n_flag & NACC) 1836 vaper->va_atime = np->n_atim; 1837 if (np->n_flag & NUPD) 1838 vaper->va_mtime = np->n_mtim; 1839 } 1840 return (0); 1841 } 1842 1843 void 1844 nfs_delayedtruncate(vp) 1845 struct vnode *vp; 1846 { 1847 struct nfsnode *np = VTONFS(vp); 1848 1849 if (np->n_flag & NTRUNCDELAYED) { 1850 np->n_flag &= ~NTRUNCDELAYED; 1851 genfs_node_wrlock(vp); 1852 mutex_enter(&vp->v_interlock); 1853 (void)VOP_PUTPAGES(vp, 0, 1854 0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES); 1855 uvm_vnp_setsize(vp, np->n_size); 1856 genfs_node_unlock(vp); 1857 } 1858 } 1859 1860 #define NFS_WCCKLUDGE_TIMEOUT (24 * 60 * 60) /* 1 day */ 1861 #define NFS_WCCKLUDGE(nmp, now) \ 1862 (((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \ 1863 ((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0) 1864 1865 /* 1866 * nfs_check_wccdata: check inaccurate wcc_data 1867 * 1868 * => return non-zero if we shouldn't trust the wcc_data. 1869 * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed". 1870 */ 1871 1872 int 1873 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime, 1874 struct timespec *mtime, bool docheck) 1875 { 1876 int error = 0; 1877 1878 #if !defined(NFS_V2_ONLY) 1879 1880 if (docheck) { 1881 struct vnode *vp = NFSTOV(np); 1882 struct nfsmount *nmp; 1883 long now = time_second; 1884 const struct timespec *omtime = &np->n_vattr->va_mtime; 1885 const struct timespec *octime = &np->n_vattr->va_ctime; 1886 #if defined(DEBUG) 1887 const char *reason = NULL; /* XXX: gcc */ 1888 #endif 1889 1890 if (timespeccmp(omtime, mtime, <=)) { 1891 #if defined(DEBUG) 1892 reason = "mtime"; 1893 #endif 1894 error = EINVAL; 1895 } 1896 1897 if (vp->v_type == VDIR && timespeccmp(octime, ctime, <=)) { 1898 #if defined(DEBUG) 1899 reason = "ctime"; 1900 #endif 1901 error = EINVAL; 1902 } 1903 1904 nmp = VFSTONFS(vp->v_mount); 1905 if (error) { 1906 1907 /* 1908 * despite of the fact that we've updated the file, 1909 * timestamps of the file were not updated as we 1910 * expected. 1911 * it means that the server has incompatible 1912 * semantics of timestamps or (more likely) 1913 * the server time is not precise enough to 1914 * track each modifications. 1915 * in that case, we disable wcc processing. 1916 * 1917 * yes, strictly speaking, we should disable all 1918 * caching. it's a compromise. 1919 */ 1920 1921 mutex_enter(&nmp->nm_lock); 1922 #if defined(DEBUG) 1923 if (!NFS_WCCKLUDGE(nmp, now)) { 1924 printf("%s: inaccurate wcc data (%s) detected," 1925 " disabling wcc" 1926 " (ctime %u.%09u %u.%09u," 1927 " mtime %u.%09u %u.%09u)\n", 1928 vp->v_mount->mnt_stat.f_mntfromname, 1929 reason, 1930 (unsigned int)octime->tv_sec, 1931 (unsigned int)octime->tv_nsec, 1932 (unsigned int)ctime->tv_sec, 1933 (unsigned int)ctime->tv_nsec, 1934 (unsigned int)omtime->tv_sec, 1935 (unsigned int)omtime->tv_nsec, 1936 (unsigned int)mtime->tv_sec, 1937 (unsigned int)mtime->tv_nsec); 1938 } 1939 #endif 1940 nmp->nm_iflag |= NFSMNT_WCCKLUDGE; 1941 nmp->nm_wcckludgetime = now; 1942 mutex_exit(&nmp->nm_lock); 1943 } else if (NFS_WCCKLUDGE(nmp, now)) { 1944 error = EPERM; /* XXX */ 1945 } else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1946 mutex_enter(&nmp->nm_lock); 1947 if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1948 #if defined(DEBUG) 1949 printf("%s: re-enabling wcc\n", 1950 vp->v_mount->mnt_stat.f_mntfromname); 1951 #endif 1952 nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE; 1953 } 1954 mutex_exit(&nmp->nm_lock); 1955 } 1956 } 1957 1958 #endif /* !defined(NFS_V2_ONLY) */ 1959 1960 return error; 1961 } 1962 1963 /* 1964 * Heuristic to see if the server XDR encodes directory cookies or not. 1965 * it is not supposed to, but a lot of servers may do this. Also, since 1966 * most/all servers will implement V2 as well, it is expected that they 1967 * may return just 32 bits worth of cookie information, so we need to 1968 * find out in which 32 bits this information is available. We do this 1969 * to avoid trouble with emulated binaries that can't handle 64 bit 1970 * directory offsets. 1971 */ 1972 1973 void 1974 nfs_cookieheuristic(vp, flagp, l, cred) 1975 struct vnode *vp; 1976 int *flagp; 1977 struct lwp *l; 1978 kauth_cred_t cred; 1979 { 1980 struct uio auio; 1981 struct iovec aiov; 1982 char *tbuf, *cp; 1983 struct dirent *dp; 1984 off_t *cookies = NULL, *cop; 1985 int error, eof, nc, len; 1986 1987 MALLOC(tbuf, void *, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK); 1988 1989 aiov.iov_base = tbuf; 1990 aiov.iov_len = NFS_DIRFRAGSIZ; 1991 auio.uio_iov = &aiov; 1992 auio.uio_iovcnt = 1; 1993 auio.uio_rw = UIO_READ; 1994 auio.uio_resid = NFS_DIRFRAGSIZ; 1995 auio.uio_offset = 0; 1996 UIO_SETUP_SYSSPACE(&auio); 1997 1998 error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc); 1999 2000 len = NFS_DIRFRAGSIZ - auio.uio_resid; 2001 if (error || len == 0) { 2002 FREE(tbuf, M_TEMP); 2003 if (cookies) 2004 free(cookies, M_TEMP); 2005 return; 2006 } 2007 2008 /* 2009 * Find the first valid entry and look at its offset cookie. 2010 */ 2011 2012 cp = tbuf; 2013 for (cop = cookies; len > 0; len -= dp->d_reclen) { 2014 dp = (struct dirent *)cp; 2015 if (dp->d_fileno != 0 && len >= dp->d_reclen) { 2016 if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) { 2017 *flagp |= NFSMNT_SWAPCOOKIE; 2018 nfs_invaldircache(vp, 0); 2019 nfs_vinvalbuf(vp, 0, cred, l, 1); 2020 } 2021 break; 2022 } 2023 cop++; 2024 cp += dp->d_reclen; 2025 } 2026 2027 FREE(tbuf, M_TEMP); 2028 free(cookies, M_TEMP); 2029 } 2030 #endif /* NFS */ 2031 2032 #ifdef NFSSERVER 2033 /* 2034 * Set up nameidata for a lookup() call and do it. 2035 * 2036 * If pubflag is set, this call is done for a lookup operation on the 2037 * public filehandle. In that case we allow crossing mountpoints and 2038 * absolute pathnames. However, the caller is expected to check that 2039 * the lookup result is within the public fs, and deny access if 2040 * it is not. 2041 */ 2042 int 2043 nfs_namei(ndp, nsfh, len, slp, nam, mdp, dposp, retdirp, l, kerbflag, pubflag) 2044 struct nameidata *ndp; 2045 nfsrvfh_t *nsfh; 2046 uint32_t len; 2047 struct nfssvc_sock *slp; 2048 struct mbuf *nam; 2049 struct mbuf **mdp; 2050 char **dposp; 2051 struct vnode **retdirp; 2052 struct lwp *l; 2053 int kerbflag, pubflag; 2054 { 2055 int i, rem; 2056 struct mbuf *md; 2057 char *fromcp, *tocp, *cp; 2058 struct iovec aiov; 2059 struct uio auio; 2060 struct vnode *dp; 2061 int error, rdonly, linklen; 2062 struct componentname *cnp = &ndp->ni_cnd; 2063 2064 *retdirp = NULL; 2065 2066 if ((len + 1) > MAXPATHLEN) 2067 return (ENAMETOOLONG); 2068 if (len == 0) 2069 return (EACCES); 2070 cnp->cn_pnbuf = PNBUF_GET(); 2071 2072 /* 2073 * Copy the name from the mbuf list to ndp->ni_pnbuf 2074 * and set the various ndp fields appropriately. 2075 */ 2076 fromcp = *dposp; 2077 tocp = cnp->cn_pnbuf; 2078 md = *mdp; 2079 rem = mtod(md, char *) + md->m_len - fromcp; 2080 for (i = 0; i < len; i++) { 2081 while (rem == 0) { 2082 md = md->m_next; 2083 if (md == NULL) { 2084 error = EBADRPC; 2085 goto out; 2086 } 2087 fromcp = mtod(md, void *); 2088 rem = md->m_len; 2089 } 2090 if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) { 2091 error = EACCES; 2092 goto out; 2093 } 2094 *tocp++ = *fromcp++; 2095 rem--; 2096 } 2097 *tocp = '\0'; 2098 *mdp = md; 2099 *dposp = fromcp; 2100 len = nfsm_rndup(len)-len; 2101 if (len > 0) { 2102 if (rem >= len) 2103 *dposp += len; 2104 else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0) 2105 goto out; 2106 } 2107 2108 /* 2109 * Extract and set starting directory. 2110 */ 2111 error = nfsrv_fhtovp(nsfh, false, &dp, ndp->ni_cnd.cn_cred, slp, 2112 nam, &rdonly, kerbflag, pubflag); 2113 if (error) 2114 goto out; 2115 if (dp->v_type != VDIR) { 2116 vrele(dp); 2117 error = ENOTDIR; 2118 goto out; 2119 } 2120 2121 if (rdonly) 2122 cnp->cn_flags |= RDONLY; 2123 2124 *retdirp = dp; 2125 2126 if (pubflag) { 2127 /* 2128 * Oh joy. For WebNFS, handle those pesky '%' escapes, 2129 * and the 'native path' indicator. 2130 */ 2131 cp = PNBUF_GET(); 2132 fromcp = cnp->cn_pnbuf; 2133 tocp = cp; 2134 if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) { 2135 switch ((unsigned char)*fromcp) { 2136 case WEBNFS_NATIVE_CHAR: 2137 /* 2138 * 'Native' path for us is the same 2139 * as a path according to the NFS spec, 2140 * just skip the escape char. 2141 */ 2142 fromcp++; 2143 break; 2144 /* 2145 * More may be added in the future, range 0x80-0xff 2146 */ 2147 default: 2148 error = EIO; 2149 vrele(dp); 2150 PNBUF_PUT(cp); 2151 goto out; 2152 } 2153 } 2154 /* 2155 * Translate the '%' escapes, URL-style. 2156 */ 2157 while (*fromcp != '\0') { 2158 if (*fromcp == WEBNFS_ESC_CHAR) { 2159 if (fromcp[1] != '\0' && fromcp[2] != '\0') { 2160 fromcp++; 2161 *tocp++ = HEXSTRTOI(fromcp); 2162 fromcp += 2; 2163 continue; 2164 } else { 2165 error = ENOENT; 2166 vrele(dp); 2167 PNBUF_PUT(cp); 2168 goto out; 2169 } 2170 } else 2171 *tocp++ = *fromcp++; 2172 } 2173 *tocp = '\0'; 2174 PNBUF_PUT(cnp->cn_pnbuf); 2175 cnp->cn_pnbuf = cp; 2176 } 2177 2178 ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1; 2179 ndp->ni_segflg = UIO_SYSSPACE; 2180 ndp->ni_rootdir = rootvnode; 2181 ndp->ni_erootdir = NULL; 2182 2183 if (pubflag) { 2184 ndp->ni_loopcnt = 0; 2185 if (cnp->cn_pnbuf[0] == '/') 2186 dp = rootvnode; 2187 } else { 2188 cnp->cn_flags |= NOCROSSMOUNT; 2189 } 2190 2191 VREF(dp); 2192 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); 2193 2194 for (;;) { 2195 cnp->cn_nameptr = cnp->cn_pnbuf; 2196 ndp->ni_startdir = dp; 2197 2198 /* 2199 * And call lookup() to do the real work 2200 */ 2201 error = lookup(ndp); 2202 if (error) { 2203 if (ndp->ni_dvp) { 2204 vput(ndp->ni_dvp); 2205 } 2206 PNBUF_PUT(cnp->cn_pnbuf); 2207 return (error); 2208 } 2209 2210 /* 2211 * Check for encountering a symbolic link 2212 */ 2213 if ((cnp->cn_flags & ISSYMLINK) == 0) { 2214 if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp) { 2215 if (ndp->ni_dvp == ndp->ni_vp) { 2216 vrele(ndp->ni_dvp); 2217 } else { 2218 vput(ndp->ni_dvp); 2219 } 2220 } 2221 if (cnp->cn_flags & (SAVENAME | SAVESTART)) 2222 cnp->cn_flags |= HASBUF; 2223 else 2224 PNBUF_PUT(cnp->cn_pnbuf); 2225 return (0); 2226 } else { 2227 if (!pubflag) { 2228 error = EINVAL; 2229 break; 2230 } 2231 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 2232 error = ELOOP; 2233 break; 2234 } 2235 if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) { 2236 error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred); 2237 if (error != 0) 2238 break; 2239 } 2240 if (ndp->ni_pathlen > 1) 2241 cp = PNBUF_GET(); 2242 else 2243 cp = cnp->cn_pnbuf; 2244 aiov.iov_base = cp; 2245 aiov.iov_len = MAXPATHLEN; 2246 auio.uio_iov = &aiov; 2247 auio.uio_iovcnt = 1; 2248 auio.uio_offset = 0; 2249 auio.uio_rw = UIO_READ; 2250 auio.uio_resid = MAXPATHLEN; 2251 UIO_SETUP_SYSSPACE(&auio); 2252 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 2253 if (error) { 2254 badlink: 2255 if (ndp->ni_pathlen > 1) 2256 PNBUF_PUT(cp); 2257 break; 2258 } 2259 linklen = MAXPATHLEN - auio.uio_resid; 2260 if (linklen == 0) { 2261 error = ENOENT; 2262 goto badlink; 2263 } 2264 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 2265 error = ENAMETOOLONG; 2266 goto badlink; 2267 } 2268 if (ndp->ni_pathlen > 1) { 2269 memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen); 2270 PNBUF_PUT(cnp->cn_pnbuf); 2271 cnp->cn_pnbuf = cp; 2272 } else 2273 cnp->cn_pnbuf[linklen] = '\0'; 2274 ndp->ni_pathlen += linklen; 2275 vput(ndp->ni_vp); 2276 dp = ndp->ni_dvp; 2277 2278 /* 2279 * Check if root directory should replace current directory. 2280 */ 2281 if (cnp->cn_pnbuf[0] == '/') { 2282 vput(dp); 2283 dp = ndp->ni_rootdir; 2284 VREF(dp); 2285 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); 2286 } 2287 } 2288 } 2289 vput(ndp->ni_dvp); 2290 vput(ndp->ni_vp); 2291 ndp->ni_vp = NULL; 2292 out: 2293 PNBUF_PUT(cnp->cn_pnbuf); 2294 return (error); 2295 } 2296 #endif /* NFSSERVER */ 2297 2298 /* 2299 * A fiddled version of m_adj() that ensures null fill to a 32-bit 2300 * boundary and only trims off the back end 2301 * 2302 * 1. trim off 'len' bytes as m_adj(mp, -len). 2303 * 2. add zero-padding 'nul' bytes at the end of the mbuf chain. 2304 */ 2305 void 2306 nfs_zeropad(mp, len, nul) 2307 struct mbuf *mp; 2308 int len; 2309 int nul; 2310 { 2311 struct mbuf *m; 2312 int count; 2313 2314 /* 2315 * Trim from tail. Scan the mbuf chain, 2316 * calculating its length and finding the last mbuf. 2317 * If the adjustment only affects this mbuf, then just 2318 * adjust and return. Otherwise, rescan and truncate 2319 * after the remaining size. 2320 */ 2321 count = 0; 2322 m = mp; 2323 for (;;) { 2324 count += m->m_len; 2325 if (m->m_next == NULL) 2326 break; 2327 m = m->m_next; 2328 } 2329 2330 KDASSERT(count >= len); 2331 2332 if (m->m_len >= len) { 2333 m->m_len -= len; 2334 } else { 2335 count -= len; 2336 /* 2337 * Correct length for chain is "count". 2338 * Find the mbuf with last data, adjust its length, 2339 * and toss data from remaining mbufs on chain. 2340 */ 2341 for (m = mp; m; m = m->m_next) { 2342 if (m->m_len >= count) { 2343 m->m_len = count; 2344 break; 2345 } 2346 count -= m->m_len; 2347 } 2348 KASSERT(m && m->m_next); 2349 m_freem(m->m_next); 2350 m->m_next = NULL; 2351 } 2352 2353 KDASSERT(m->m_next == NULL); 2354 2355 /* 2356 * zero-padding. 2357 */ 2358 if (nul > 0) { 2359 char *cp; 2360 int i; 2361 2362 if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) { 2363 struct mbuf *n; 2364 2365 KDASSERT(MLEN >= nul); 2366 n = m_get(M_WAIT, MT_DATA); 2367 MCLAIM(n, &nfs_mowner); 2368 n->m_len = nul; 2369 n->m_next = NULL; 2370 m->m_next = n; 2371 cp = mtod(n, void *); 2372 } else { 2373 cp = mtod(m, char *) + m->m_len; 2374 m->m_len += nul; 2375 } 2376 for (i = 0; i < nul; i++) 2377 *cp++ = '\0'; 2378 } 2379 return; 2380 } 2381 2382 /* 2383 * Make these functions instead of macros, so that the kernel text size 2384 * doesn't get too big... 2385 */ 2386 void 2387 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp) 2388 struct nfsrv_descript *nfsd; 2389 int before_ret; 2390 struct vattr *before_vap; 2391 int after_ret; 2392 struct vattr *after_vap; 2393 struct mbuf **mbp; 2394 char **bposp; 2395 { 2396 struct mbuf *mb = *mbp; 2397 char *bpos = *bposp; 2398 u_int32_t *tl; 2399 2400 if (before_ret) { 2401 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2402 *tl = nfs_false; 2403 } else { 2404 nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED); 2405 *tl++ = nfs_true; 2406 txdr_hyper(before_vap->va_size, tl); 2407 tl += 2; 2408 txdr_nfsv3time(&(before_vap->va_mtime), tl); 2409 tl += 2; 2410 txdr_nfsv3time(&(before_vap->va_ctime), tl); 2411 } 2412 *bposp = bpos; 2413 *mbp = mb; 2414 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp); 2415 } 2416 2417 void 2418 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp) 2419 struct nfsrv_descript *nfsd; 2420 int after_ret; 2421 struct vattr *after_vap; 2422 struct mbuf **mbp; 2423 char **bposp; 2424 { 2425 struct mbuf *mb = *mbp; 2426 char *bpos = *bposp; 2427 u_int32_t *tl; 2428 struct nfs_fattr *fp; 2429 2430 if (after_ret) { 2431 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2432 *tl = nfs_false; 2433 } else { 2434 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR); 2435 *tl++ = nfs_true; 2436 fp = (struct nfs_fattr *)tl; 2437 nfsm_srvfattr(nfsd, after_vap, fp); 2438 } 2439 *mbp = mb; 2440 *bposp = bpos; 2441 } 2442 2443 void 2444 nfsm_srvfattr(nfsd, vap, fp) 2445 struct nfsrv_descript *nfsd; 2446 struct vattr *vap; 2447 struct nfs_fattr *fp; 2448 { 2449 2450 fp->fa_nlink = txdr_unsigned(vap->va_nlink); 2451 fp->fa_uid = txdr_unsigned(vap->va_uid); 2452 fp->fa_gid = txdr_unsigned(vap->va_gid); 2453 if (nfsd->nd_flag & ND_NFSV3) { 2454 fp->fa_type = vtonfsv3_type(vap->va_type); 2455 fp->fa_mode = vtonfsv3_mode(vap->va_mode); 2456 txdr_hyper(vap->va_size, &fp->fa3_size); 2457 txdr_hyper(vap->va_bytes, &fp->fa3_used); 2458 fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev)); 2459 fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev)); 2460 fp->fa3_fsid.nfsuquad[0] = 0; 2461 fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid); 2462 txdr_hyper(vap->va_fileid, &fp->fa3_fileid); 2463 txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime); 2464 txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime); 2465 txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime); 2466 } else { 2467 fp->fa_type = vtonfsv2_type(vap->va_type); 2468 fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 2469 fp->fa2_size = txdr_unsigned(vap->va_size); 2470 fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize); 2471 if (vap->va_type == VFIFO) 2472 fp->fa2_rdev = 0xffffffff; 2473 else 2474 fp->fa2_rdev = txdr_unsigned(vap->va_rdev); 2475 fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE); 2476 fp->fa2_fsid = txdr_unsigned(vap->va_fsid); 2477 fp->fa2_fileid = txdr_unsigned(vap->va_fileid); 2478 txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime); 2479 txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime); 2480 txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime); 2481 } 2482 } 2483 2484 #ifdef NFSSERVER 2485 /* 2486 * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked) 2487 * - look up fsid in mount list (if not found ret error) 2488 * - get vp and export rights by calling VFS_FHTOVP() 2489 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 2490 * - if not lockflag unlock it with VOP_UNLOCK() 2491 */ 2492 int 2493 nfsrv_fhtovp(nfsrvfh_t *nsfh, int lockflag, struct vnode **vpp, 2494 kauth_cred_t cred, struct nfssvc_sock *slp, struct mbuf *nam, int *rdonlyp, 2495 int kerbflag, int pubflag) 2496 { 2497 struct mount *mp; 2498 kauth_cred_t credanon; 2499 int error, exflags; 2500 struct sockaddr_in *saddr; 2501 fhandle_t *fhp; 2502 2503 fhp = NFSRVFH_FHANDLE(nsfh); 2504 *vpp = (struct vnode *)0; 2505 2506 if (nfs_ispublicfh(nsfh)) { 2507 if (!pubflag || !nfs_pub.np_valid) 2508 return (ESTALE); 2509 fhp = nfs_pub.np_handle; 2510 } 2511 2512 error = netexport_check(&fhp->fh_fsid, nam, &mp, &exflags, &credanon); 2513 if (error) { 2514 return error; 2515 } 2516 2517 error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp); 2518 if (error) 2519 return (error); 2520 2521 if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) { 2522 saddr = mtod(nam, struct sockaddr_in *); 2523 if ((saddr->sin_family == AF_INET) && 2524 ntohs(saddr->sin_port) >= IPPORT_RESERVED) { 2525 vput(*vpp); 2526 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2527 } 2528 #ifdef INET6 2529 if ((saddr->sin_family == AF_INET6) && 2530 ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) { 2531 vput(*vpp); 2532 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2533 } 2534 #endif 2535 } 2536 /* 2537 * Check/setup credentials. 2538 */ 2539 if (exflags & MNT_EXKERB) { 2540 if (!kerbflag) { 2541 vput(*vpp); 2542 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2543 } 2544 } else if (kerbflag) { 2545 vput(*vpp); 2546 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2547 } else if (kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 2548 NULL) == 0 || (exflags & MNT_EXPORTANON)) { 2549 kauth_cred_clone(credanon, cred); 2550 } 2551 if (exflags & MNT_EXRDONLY) 2552 *rdonlyp = 1; 2553 else 2554 *rdonlyp = 0; 2555 if (!lockflag) 2556 VOP_UNLOCK(*vpp, 0); 2557 return (0); 2558 } 2559 2560 /* 2561 * WebNFS: check if a filehandle is a public filehandle. For v3, this 2562 * means a length of 0, for v2 it means all zeroes. 2563 */ 2564 int 2565 nfs_ispublicfh(const nfsrvfh_t *nsfh) 2566 { 2567 const char *cp = (const void *)(NFSRVFH_DATA(nsfh)); 2568 int i; 2569 2570 if (NFSRVFH_SIZE(nsfh) == 0) { 2571 return true; 2572 } 2573 if (NFSRVFH_SIZE(nsfh) != NFSX_V2FH) { 2574 return false; 2575 } 2576 for (i = 0; i < NFSX_V2FH; i++) 2577 if (*cp++ != 0) 2578 return false; 2579 return true; 2580 } 2581 #endif /* NFSSERVER */ 2582 2583 /* 2584 * This function compares two net addresses by family and returns true 2585 * if they are the same host. 2586 * If there is any doubt, return false. 2587 * The AF_INET family is handled as a special case so that address mbufs 2588 * don't need to be saved to store "struct in_addr", which is only 4 bytes. 2589 */ 2590 int 2591 netaddr_match(family, haddr, nam) 2592 int family; 2593 union nethostaddr *haddr; 2594 struct mbuf *nam; 2595 { 2596 struct sockaddr_in *inetaddr; 2597 2598 switch (family) { 2599 case AF_INET: 2600 inetaddr = mtod(nam, struct sockaddr_in *); 2601 if (inetaddr->sin_family == AF_INET && 2602 inetaddr->sin_addr.s_addr == haddr->had_inetaddr) 2603 return (1); 2604 break; 2605 #ifdef INET6 2606 case AF_INET6: 2607 { 2608 struct sockaddr_in6 *sin6_1, *sin6_2; 2609 2610 sin6_1 = mtod(nam, struct sockaddr_in6 *); 2611 sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *); 2612 if (sin6_1->sin6_family == AF_INET6 && 2613 IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr)) 2614 return 1; 2615 } 2616 #endif 2617 #ifdef ISO 2618 case AF_ISO: 2619 { 2620 struct sockaddr_iso *isoaddr1, *isoaddr2; 2621 2622 isoaddr1 = mtod(nam, struct sockaddr_iso *); 2623 isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *); 2624 if (isoaddr1->siso_family == AF_ISO && 2625 isoaddr1->siso_nlen > 0 && 2626 isoaddr1->siso_nlen == isoaddr2->siso_nlen && 2627 SAME_ISOADDR(isoaddr1, isoaddr2)) 2628 return (1); 2629 break; 2630 } 2631 #endif /* ISO */ 2632 default: 2633 break; 2634 }; 2635 return (0); 2636 } 2637 2638 /* 2639 * The write verifier has changed (probably due to a server reboot), so all 2640 * PG_NEEDCOMMIT pages will have to be written again. Since they are marked 2641 * as dirty or are being written out just now, all this takes is clearing 2642 * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for 2643 * the mount point. 2644 */ 2645 void 2646 nfs_clearcommit(mp) 2647 struct mount *mp; 2648 { 2649 struct vnode *vp; 2650 struct nfsnode *np; 2651 struct vm_page *pg; 2652 struct nfsmount *nmp = VFSTONFS(mp); 2653 2654 rw_enter(&nmp->nm_writeverflock, RW_WRITER); 2655 mutex_enter(&mntvnode_lock); 2656 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2657 KASSERT(vp->v_mount == mp); 2658 if (vp->v_type != VREG) 2659 continue; 2660 np = VTONFS(vp); 2661 np->n_pushlo = np->n_pushhi = np->n_pushedlo = 2662 np->n_pushedhi = 0; 2663 np->n_commitflags &= 2664 ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID); 2665 mutex_enter(&vp->v_uobj.vmobjlock); 2666 TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) { 2667 pg->flags &= ~PG_NEEDCOMMIT; 2668 } 2669 mutex_exit(&vp->v_uobj.vmobjlock); 2670 } 2671 mutex_exit(&mntvnode_lock); 2672 mutex_enter(&nmp->nm_lock); 2673 nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF; 2674 mutex_exit(&nmp->nm_lock); 2675 rw_exit(&nmp->nm_writeverflock); 2676 } 2677 2678 void 2679 nfs_merge_commit_ranges(vp) 2680 struct vnode *vp; 2681 { 2682 struct nfsnode *np = VTONFS(vp); 2683 2684 KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID); 2685 2686 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2687 np->n_pushedlo = np->n_pushlo; 2688 np->n_pushedhi = np->n_pushhi; 2689 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2690 } else { 2691 if (np->n_pushlo < np->n_pushedlo) 2692 np->n_pushedlo = np->n_pushlo; 2693 if (np->n_pushhi > np->n_pushedhi) 2694 np->n_pushedhi = np->n_pushhi; 2695 } 2696 2697 np->n_pushlo = np->n_pushhi = 0; 2698 np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID; 2699 2700 #ifdef NFS_DEBUG_COMMIT 2701 printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2702 (unsigned)np->n_pushedhi); 2703 #endif 2704 } 2705 2706 int 2707 nfs_in_committed_range(vp, off, len) 2708 struct vnode *vp; 2709 off_t off, len; 2710 { 2711 struct nfsnode *np = VTONFS(vp); 2712 off_t lo, hi; 2713 2714 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2715 return 0; 2716 lo = off; 2717 hi = lo + len; 2718 2719 return (lo >= np->n_pushedlo && hi <= np->n_pushedhi); 2720 } 2721 2722 int 2723 nfs_in_tobecommitted_range(vp, off, len) 2724 struct vnode *vp; 2725 off_t off, len; 2726 { 2727 struct nfsnode *np = VTONFS(vp); 2728 off_t lo, hi; 2729 2730 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2731 return 0; 2732 lo = off; 2733 hi = lo + len; 2734 2735 return (lo >= np->n_pushlo && hi <= np->n_pushhi); 2736 } 2737 2738 void 2739 nfs_add_committed_range(vp, off, len) 2740 struct vnode *vp; 2741 off_t off, len; 2742 { 2743 struct nfsnode *np = VTONFS(vp); 2744 off_t lo, hi; 2745 2746 lo = off; 2747 hi = lo + len; 2748 2749 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2750 np->n_pushedlo = lo; 2751 np->n_pushedhi = hi; 2752 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2753 } else { 2754 if (hi > np->n_pushedhi) 2755 np->n_pushedhi = hi; 2756 if (lo < np->n_pushedlo) 2757 np->n_pushedlo = lo; 2758 } 2759 #ifdef NFS_DEBUG_COMMIT 2760 printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2761 (unsigned)np->n_pushedhi); 2762 #endif 2763 } 2764 2765 void 2766 nfs_del_committed_range(vp, off, len) 2767 struct vnode *vp; 2768 off_t off, len; 2769 { 2770 struct nfsnode *np = VTONFS(vp); 2771 off_t lo, hi; 2772 2773 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2774 return; 2775 2776 lo = off; 2777 hi = lo + len; 2778 2779 if (lo > np->n_pushedhi || hi < np->n_pushedlo) 2780 return; 2781 if (lo <= np->n_pushedlo) 2782 np->n_pushedlo = hi; 2783 else if (hi >= np->n_pushedhi) 2784 np->n_pushedhi = lo; 2785 else { 2786 /* 2787 * XXX There's only one range. If the deleted range 2788 * is in the middle, pick the largest of the 2789 * contiguous ranges that it leaves. 2790 */ 2791 if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi)) 2792 np->n_pushedhi = lo; 2793 else 2794 np->n_pushedlo = hi; 2795 } 2796 #ifdef NFS_DEBUG_COMMIT 2797 printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2798 (unsigned)np->n_pushedhi); 2799 #endif 2800 } 2801 2802 void 2803 nfs_add_tobecommitted_range(vp, off, len) 2804 struct vnode *vp; 2805 off_t off, len; 2806 { 2807 struct nfsnode *np = VTONFS(vp); 2808 off_t lo, hi; 2809 2810 lo = off; 2811 hi = lo + len; 2812 2813 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) { 2814 np->n_pushlo = lo; 2815 np->n_pushhi = hi; 2816 np->n_commitflags |= NFS_COMMIT_PUSH_VALID; 2817 } else { 2818 if (lo < np->n_pushlo) 2819 np->n_pushlo = lo; 2820 if (hi > np->n_pushhi) 2821 np->n_pushhi = hi; 2822 } 2823 #ifdef NFS_DEBUG_COMMIT 2824 printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2825 (unsigned)np->n_pushhi); 2826 #endif 2827 } 2828 2829 void 2830 nfs_del_tobecommitted_range(vp, off, len) 2831 struct vnode *vp; 2832 off_t off, len; 2833 { 2834 struct nfsnode *np = VTONFS(vp); 2835 off_t lo, hi; 2836 2837 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2838 return; 2839 2840 lo = off; 2841 hi = lo + len; 2842 2843 if (lo > np->n_pushhi || hi < np->n_pushlo) 2844 return; 2845 2846 if (lo <= np->n_pushlo) 2847 np->n_pushlo = hi; 2848 else if (hi >= np->n_pushhi) 2849 np->n_pushhi = lo; 2850 else { 2851 /* 2852 * XXX There's only one range. If the deleted range 2853 * is in the middle, pick the largest of the 2854 * contiguous ranges that it leaves. 2855 */ 2856 if ((np->n_pushlo - lo) > (hi - np->n_pushhi)) 2857 np->n_pushhi = lo; 2858 else 2859 np->n_pushlo = hi; 2860 } 2861 #ifdef NFS_DEBUG_COMMIT 2862 printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2863 (unsigned)np->n_pushhi); 2864 #endif 2865 } 2866 2867 /* 2868 * Map errnos to NFS error numbers. For Version 3 also filter out error 2869 * numbers not specified for the associated procedure. 2870 */ 2871 int 2872 nfsrv_errmap(nd, err) 2873 struct nfsrv_descript *nd; 2874 int err; 2875 { 2876 const short *defaulterrp, *errp; 2877 2878 if (nd->nd_flag & ND_NFSV3) { 2879 if (nd->nd_procnum <= NFSPROC_COMMIT) { 2880 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum]; 2881 while (*++errp) { 2882 if (*errp == err) 2883 return (err); 2884 else if (*errp > err) 2885 break; 2886 } 2887 return ((int)*defaulterrp); 2888 } else 2889 return (err & 0xffff); 2890 } 2891 if (err <= ELAST) 2892 return ((int)nfsrv_v2errmap[err - 1]); 2893 return (NFSERR_IO); 2894 } 2895 2896 u_int32_t 2897 nfs_getxid() 2898 { 2899 static u_int32_t base; 2900 static u_int32_t nfs_xid = 0; 2901 static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER; 2902 u_int32_t newxid; 2903 2904 simple_lock(&nfs_xidlock); 2905 /* 2906 * derive initial xid from system time 2907 * XXX time is invalid if root not yet mounted 2908 */ 2909 if (__predict_false(!base && (rootvp))) { 2910 struct timeval tv; 2911 2912 microtime(&tv); 2913 base = tv.tv_sec << 12; 2914 nfs_xid = base; 2915 } 2916 2917 /* 2918 * Skip zero xid if it should ever happen. 2919 */ 2920 if (__predict_false(++nfs_xid == 0)) 2921 nfs_xid++; 2922 newxid = nfs_xid; 2923 simple_unlock(&nfs_xidlock); 2924 2925 return txdr_unsigned(newxid); 2926 } 2927 2928 /* 2929 * assign a new xid for existing request. 2930 * used for NFSERR_JUKEBOX handling. 2931 */ 2932 void 2933 nfs_renewxid(struct nfsreq *req) 2934 { 2935 u_int32_t xid; 2936 int off; 2937 2938 xid = nfs_getxid(); 2939 if (req->r_nmp->nm_sotype == SOCK_STREAM) 2940 off = sizeof(u_int32_t); /* RPC record mark */ 2941 else 2942 off = 0; 2943 2944 m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid); 2945 req->r_xid = xid; 2946 } 2947 2948 #if defined(NFSSERVER) 2949 int 2950 nfsrv_composefh(struct vnode *vp, nfsrvfh_t *nsfh, bool v3) 2951 { 2952 int error; 2953 size_t fhsize; 2954 2955 fhsize = NFSD_MAXFHSIZE; 2956 error = vfs_composefh(vp, (void *)NFSRVFH_DATA(nsfh), &fhsize); 2957 if (NFSX_FHTOOBIG_P(fhsize, v3)) { 2958 error = EOPNOTSUPP; 2959 } 2960 if (error != 0) { 2961 return error; 2962 } 2963 if (!v3 && fhsize < NFSX_V2FH) { 2964 memset((char *)NFSRVFH_DATA(nsfh) + fhsize, 0, 2965 NFSX_V2FH - fhsize); 2966 fhsize = NFSX_V2FH; 2967 } 2968 if ((fhsize % NFSX_UNSIGNED) != 0) { 2969 return EOPNOTSUPP; 2970 } 2971 nsfh->nsfh_size = fhsize; 2972 return 0; 2973 } 2974 2975 int 2976 nfsrv_comparefh(const nfsrvfh_t *fh1, const nfsrvfh_t *fh2) 2977 { 2978 2979 if (NFSRVFH_SIZE(fh1) != NFSRVFH_SIZE(fh2)) { 2980 return NFSRVFH_SIZE(fh2) - NFSRVFH_SIZE(fh1); 2981 } 2982 return memcmp(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), NFSRVFH_SIZE(fh1)); 2983 } 2984 2985 void 2986 nfsrv_copyfh(nfsrvfh_t *fh1, const nfsrvfh_t *fh2) 2987 { 2988 size_t size; 2989 2990 fh1->nsfh_size = size = NFSRVFH_SIZE(fh2); 2991 memcpy(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), size); 2992 } 2993 #endif /* defined(NFSSERVER) */ 2994 2995 #if defined(NFS) 2996 /* 2997 * Set the attribute timeout based on how recently the file has been modified. 2998 */ 2999 3000 time_t 3001 nfs_attrtimeo(struct nfsmount *nmp, struct nfsnode *np) 3002 { 3003 time_t timeo; 3004 3005 if ((nmp->nm_flag & NFSMNT_NOAC) != 0) 3006 return 0; 3007 3008 if (((np)->n_flag & NMODIFIED) != 0) 3009 return NFS_MINATTRTIMO; 3010 3011 timeo = (time_second - np->n_mtime.tv_sec) / 10; 3012 timeo = max(timeo, NFS_MINATTRTIMO); 3013 timeo = min(timeo, NFS_MAXATTRTIMO); 3014 return timeo; 3015 } 3016 #endif /* defined(NFS) */ 3017
Indexes created Mon Sep 22 05:09:51 GMT 2025