nfs_subs.c revision 1.149.2.7
1 /* $NetBSD: nfs_subs.c,v 1.149.2.7 2007/12/07 17:34:44 yamt Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_subs.c 8.8 (Berkeley) 5/22/95 35 */ 36 37 /* 38 * Copyright 2000 Wasabi Systems, Inc. 39 * All rights reserved. 40 * 41 * Written by Frank van der Linden for Wasabi Systems, Inc. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. All advertising materials mentioning features or use of this software 52 * must display the following acknowledgement: 53 * This product includes software developed for the NetBSD Project by 54 * Wasabi Systems, Inc. 55 * 4. The name of Wasabi Systems, Inc. may not be used to endorse 56 * or promote products derived from this software without specific prior 57 * written permission. 58 * 59 * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND 60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 61 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 62 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL WASABI SYSTEMS, INC 63 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 64 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 65 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 66 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 67 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 68 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 69 * POSSIBILITY OF SUCH DAMAGE. 70 */ 71 72 #include <sys/cdefs.h> 73 __KERNEL_RCSID(0, "$NetBSD: nfs_subs.c,v 1.149.2.7 2007/12/07 17:34:44 yamt Exp $"); 74 75 #include "fs_nfs.h" 76 #include "opt_nfs.h" 77 #include "opt_nfsserver.h" 78 #include "opt_iso.h" 79 #include "opt_inet.h" 80 81 /* 82 * These functions support the macros and help fiddle mbuf chains for 83 * the nfs op functions. They do things like create the rpc header and 84 * copy data between mbuf chains and uio lists. 85 */ 86 #include <sys/param.h> 87 #include <sys/proc.h> 88 #include <sys/systm.h> 89 #include <sys/kernel.h> 90 #include <sys/mount.h> 91 #include <sys/vnode.h> 92 #include <sys/namei.h> 93 #include <sys/mbuf.h> 94 #include <sys/socket.h> 95 #include <sys/stat.h> 96 #include <sys/malloc.h> 97 #include <sys/filedesc.h> 98 #include <sys/time.h> 99 #include <sys/dirent.h> 100 #include <sys/once.h> 101 #include <sys/kauth.h> 102 103 #include <uvm/uvm_extern.h> 104 105 #include <nfs/rpcv2.h> 106 #include <nfs/nfsproto.h> 107 #include <nfs/nfsnode.h> 108 #include <nfs/nfs.h> 109 #include <nfs/xdr_subs.h> 110 #include <nfs/nfsm_subs.h> 111 #include <nfs/nfsmount.h> 112 #include <nfs/nfsrtt.h> 113 #include <nfs/nfs_var.h> 114 115 #include <miscfs/specfs/specdev.h> 116 117 #include <netinet/in.h> 118 #ifdef ISO 119 #include <netiso/iso.h> 120 #endif 121 122 /* 123 * Data items converted to xdr at startup, since they are constant 124 * This is kinda hokey, but may save a little time doing byte swaps 125 */ 126 u_int32_t nfs_xdrneg1; 127 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr, 128 rpc_mismatch, rpc_auth_unix, rpc_msgaccepted, 129 rpc_auth_kerb; 130 u_int32_t nfs_prog, nfs_true, nfs_false; 131 132 /* And other global data */ 133 const nfstype nfsv2_type[9] = 134 { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON }; 135 const nfstype nfsv3_type[9] = 136 { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON }; 137 const enum vtype nv2tov_type[8] = 138 { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON }; 139 const enum vtype nv3tov_type[8] = 140 { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO }; 141 int nfs_ticks; 142 int nfs_commitsize; 143 144 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies"); 145 146 /* NFS client/server stats. */ 147 struct nfsstats nfsstats; 148 149 /* 150 * Mapping of old NFS Version 2 RPC numbers to generic numbers. 151 */ 152 const int nfsv3_procid[NFS_NPROCS] = { 153 NFSPROC_NULL, 154 NFSPROC_GETATTR, 155 NFSPROC_SETATTR, 156 NFSPROC_NOOP, 157 NFSPROC_LOOKUP, 158 NFSPROC_READLINK, 159 NFSPROC_READ, 160 NFSPROC_NOOP, 161 NFSPROC_WRITE, 162 NFSPROC_CREATE, 163 NFSPROC_REMOVE, 164 NFSPROC_RENAME, 165 NFSPROC_LINK, 166 NFSPROC_SYMLINK, 167 NFSPROC_MKDIR, 168 NFSPROC_RMDIR, 169 NFSPROC_READDIR, 170 NFSPROC_FSSTAT, 171 NFSPROC_NOOP, 172 NFSPROC_NOOP, 173 NFSPROC_NOOP, 174 NFSPROC_NOOP, 175 NFSPROC_NOOP 176 }; 177 178 /* 179 * and the reverse mapping from generic to Version 2 procedure numbers 180 */ 181 const int nfsv2_procid[NFS_NPROCS] = { 182 NFSV2PROC_NULL, 183 NFSV2PROC_GETATTR, 184 NFSV2PROC_SETATTR, 185 NFSV2PROC_LOOKUP, 186 NFSV2PROC_NOOP, 187 NFSV2PROC_READLINK, 188 NFSV2PROC_READ, 189 NFSV2PROC_WRITE, 190 NFSV2PROC_CREATE, 191 NFSV2PROC_MKDIR, 192 NFSV2PROC_SYMLINK, 193 NFSV2PROC_CREATE, 194 NFSV2PROC_REMOVE, 195 NFSV2PROC_RMDIR, 196 NFSV2PROC_RENAME, 197 NFSV2PROC_LINK, 198 NFSV2PROC_READDIR, 199 NFSV2PROC_NOOP, 200 NFSV2PROC_STATFS, 201 NFSV2PROC_NOOP, 202 NFSV2PROC_NOOP, 203 NFSV2PROC_NOOP, 204 NFSV2PROC_NOOP, 205 }; 206 207 /* 208 * Maps errno values to nfs error numbers. 209 * Use NFSERR_IO as the catch all for ones not specifically defined in 210 * RFC 1094. 211 */ 212 static const u_char nfsrv_v2errmap[ELAST] = { 213 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 214 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 215 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 216 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 217 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 218 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 219 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 220 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 221 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 222 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 223 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 224 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 225 NFSERR_IO, NFSERR_IO, NFSERR_NAMETOL, NFSERR_IO, NFSERR_IO, 226 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 227 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 228 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 229 NFSERR_IO, NFSERR_IO, 230 }; 231 232 /* 233 * Maps errno values to nfs error numbers. 234 * Although it is not obvious whether or not NFS clients really care if 235 * a returned error value is in the specified list for the procedure, the 236 * safest thing to do is filter them appropriately. For Version 2, the 237 * X/Open XNFS document is the only specification that defines error values 238 * for each RPC (The RFC simply lists all possible error values for all RPCs), 239 * so I have decided to not do this for Version 2. 240 * The first entry is the default error return and the rest are the valid 241 * errors for that RPC in increasing numeric order. 242 */ 243 static const short nfsv3err_null[] = { 244 0, 245 0, 246 }; 247 248 static const short nfsv3err_getattr[] = { 249 NFSERR_IO, 250 NFSERR_IO, 251 NFSERR_STALE, 252 NFSERR_BADHANDLE, 253 NFSERR_SERVERFAULT, 254 0, 255 }; 256 257 static const short nfsv3err_setattr[] = { 258 NFSERR_IO, 259 NFSERR_PERM, 260 NFSERR_IO, 261 NFSERR_ACCES, 262 NFSERR_INVAL, 263 NFSERR_NOSPC, 264 NFSERR_ROFS, 265 NFSERR_DQUOT, 266 NFSERR_STALE, 267 NFSERR_BADHANDLE, 268 NFSERR_NOT_SYNC, 269 NFSERR_SERVERFAULT, 270 0, 271 }; 272 273 static const short nfsv3err_lookup[] = { 274 NFSERR_IO, 275 NFSERR_NOENT, 276 NFSERR_IO, 277 NFSERR_ACCES, 278 NFSERR_NOTDIR, 279 NFSERR_NAMETOL, 280 NFSERR_STALE, 281 NFSERR_BADHANDLE, 282 NFSERR_SERVERFAULT, 283 0, 284 }; 285 286 static const short nfsv3err_access[] = { 287 NFSERR_IO, 288 NFSERR_IO, 289 NFSERR_STALE, 290 NFSERR_BADHANDLE, 291 NFSERR_SERVERFAULT, 292 0, 293 }; 294 295 static const short nfsv3err_readlink[] = { 296 NFSERR_IO, 297 NFSERR_IO, 298 NFSERR_ACCES, 299 NFSERR_INVAL, 300 NFSERR_STALE, 301 NFSERR_BADHANDLE, 302 NFSERR_NOTSUPP, 303 NFSERR_SERVERFAULT, 304 0, 305 }; 306 307 static const short nfsv3err_read[] = { 308 NFSERR_IO, 309 NFSERR_IO, 310 NFSERR_NXIO, 311 NFSERR_ACCES, 312 NFSERR_INVAL, 313 NFSERR_STALE, 314 NFSERR_BADHANDLE, 315 NFSERR_SERVERFAULT, 316 NFSERR_JUKEBOX, 317 0, 318 }; 319 320 static const short nfsv3err_write[] = { 321 NFSERR_IO, 322 NFSERR_IO, 323 NFSERR_ACCES, 324 NFSERR_INVAL, 325 NFSERR_FBIG, 326 NFSERR_NOSPC, 327 NFSERR_ROFS, 328 NFSERR_DQUOT, 329 NFSERR_STALE, 330 NFSERR_BADHANDLE, 331 NFSERR_SERVERFAULT, 332 NFSERR_JUKEBOX, 333 0, 334 }; 335 336 static const short nfsv3err_create[] = { 337 NFSERR_IO, 338 NFSERR_IO, 339 NFSERR_ACCES, 340 NFSERR_EXIST, 341 NFSERR_NOTDIR, 342 NFSERR_NOSPC, 343 NFSERR_ROFS, 344 NFSERR_NAMETOL, 345 NFSERR_DQUOT, 346 NFSERR_STALE, 347 NFSERR_BADHANDLE, 348 NFSERR_NOTSUPP, 349 NFSERR_SERVERFAULT, 350 0, 351 }; 352 353 static const short nfsv3err_mkdir[] = { 354 NFSERR_IO, 355 NFSERR_IO, 356 NFSERR_ACCES, 357 NFSERR_EXIST, 358 NFSERR_NOTDIR, 359 NFSERR_NOSPC, 360 NFSERR_ROFS, 361 NFSERR_NAMETOL, 362 NFSERR_DQUOT, 363 NFSERR_STALE, 364 NFSERR_BADHANDLE, 365 NFSERR_NOTSUPP, 366 NFSERR_SERVERFAULT, 367 0, 368 }; 369 370 static const short nfsv3err_symlink[] = { 371 NFSERR_IO, 372 NFSERR_IO, 373 NFSERR_ACCES, 374 NFSERR_EXIST, 375 NFSERR_NOTDIR, 376 NFSERR_NOSPC, 377 NFSERR_ROFS, 378 NFSERR_NAMETOL, 379 NFSERR_DQUOT, 380 NFSERR_STALE, 381 NFSERR_BADHANDLE, 382 NFSERR_NOTSUPP, 383 NFSERR_SERVERFAULT, 384 0, 385 }; 386 387 static const short nfsv3err_mknod[] = { 388 NFSERR_IO, 389 NFSERR_IO, 390 NFSERR_ACCES, 391 NFSERR_EXIST, 392 NFSERR_NOTDIR, 393 NFSERR_NOSPC, 394 NFSERR_ROFS, 395 NFSERR_NAMETOL, 396 NFSERR_DQUOT, 397 NFSERR_STALE, 398 NFSERR_BADHANDLE, 399 NFSERR_NOTSUPP, 400 NFSERR_SERVERFAULT, 401 NFSERR_BADTYPE, 402 0, 403 }; 404 405 static const short nfsv3err_remove[] = { 406 NFSERR_IO, 407 NFSERR_NOENT, 408 NFSERR_IO, 409 NFSERR_ACCES, 410 NFSERR_NOTDIR, 411 NFSERR_ROFS, 412 NFSERR_NAMETOL, 413 NFSERR_STALE, 414 NFSERR_BADHANDLE, 415 NFSERR_SERVERFAULT, 416 0, 417 }; 418 419 static const short nfsv3err_rmdir[] = { 420 NFSERR_IO, 421 NFSERR_NOENT, 422 NFSERR_IO, 423 NFSERR_ACCES, 424 NFSERR_EXIST, 425 NFSERR_NOTDIR, 426 NFSERR_INVAL, 427 NFSERR_ROFS, 428 NFSERR_NAMETOL, 429 NFSERR_NOTEMPTY, 430 NFSERR_STALE, 431 NFSERR_BADHANDLE, 432 NFSERR_NOTSUPP, 433 NFSERR_SERVERFAULT, 434 0, 435 }; 436 437 static const short nfsv3err_rename[] = { 438 NFSERR_IO, 439 NFSERR_NOENT, 440 NFSERR_IO, 441 NFSERR_ACCES, 442 NFSERR_EXIST, 443 NFSERR_XDEV, 444 NFSERR_NOTDIR, 445 NFSERR_ISDIR, 446 NFSERR_INVAL, 447 NFSERR_NOSPC, 448 NFSERR_ROFS, 449 NFSERR_MLINK, 450 NFSERR_NAMETOL, 451 NFSERR_NOTEMPTY, 452 NFSERR_DQUOT, 453 NFSERR_STALE, 454 NFSERR_BADHANDLE, 455 NFSERR_NOTSUPP, 456 NFSERR_SERVERFAULT, 457 0, 458 }; 459 460 static const short nfsv3err_link[] = { 461 NFSERR_IO, 462 NFSERR_IO, 463 NFSERR_ACCES, 464 NFSERR_EXIST, 465 NFSERR_XDEV, 466 NFSERR_NOTDIR, 467 NFSERR_INVAL, 468 NFSERR_NOSPC, 469 NFSERR_ROFS, 470 NFSERR_MLINK, 471 NFSERR_NAMETOL, 472 NFSERR_DQUOT, 473 NFSERR_STALE, 474 NFSERR_BADHANDLE, 475 NFSERR_NOTSUPP, 476 NFSERR_SERVERFAULT, 477 0, 478 }; 479 480 static const short nfsv3err_readdir[] = { 481 NFSERR_IO, 482 NFSERR_IO, 483 NFSERR_ACCES, 484 NFSERR_NOTDIR, 485 NFSERR_STALE, 486 NFSERR_BADHANDLE, 487 NFSERR_BAD_COOKIE, 488 NFSERR_TOOSMALL, 489 NFSERR_SERVERFAULT, 490 0, 491 }; 492 493 static const short nfsv3err_readdirplus[] = { 494 NFSERR_IO, 495 NFSERR_IO, 496 NFSERR_ACCES, 497 NFSERR_NOTDIR, 498 NFSERR_STALE, 499 NFSERR_BADHANDLE, 500 NFSERR_BAD_COOKIE, 501 NFSERR_NOTSUPP, 502 NFSERR_TOOSMALL, 503 NFSERR_SERVERFAULT, 504 0, 505 }; 506 507 static const short nfsv3err_fsstat[] = { 508 NFSERR_IO, 509 NFSERR_IO, 510 NFSERR_STALE, 511 NFSERR_BADHANDLE, 512 NFSERR_SERVERFAULT, 513 0, 514 }; 515 516 static const short nfsv3err_fsinfo[] = { 517 NFSERR_STALE, 518 NFSERR_STALE, 519 NFSERR_BADHANDLE, 520 NFSERR_SERVERFAULT, 521 0, 522 }; 523 524 static const short nfsv3err_pathconf[] = { 525 NFSERR_STALE, 526 NFSERR_STALE, 527 NFSERR_BADHANDLE, 528 NFSERR_SERVERFAULT, 529 0, 530 }; 531 532 static const short nfsv3err_commit[] = { 533 NFSERR_IO, 534 NFSERR_IO, 535 NFSERR_STALE, 536 NFSERR_BADHANDLE, 537 NFSERR_SERVERFAULT, 538 0, 539 }; 540 541 static const short * const nfsrv_v3errmap[] = { 542 nfsv3err_null, 543 nfsv3err_getattr, 544 nfsv3err_setattr, 545 nfsv3err_lookup, 546 nfsv3err_access, 547 nfsv3err_readlink, 548 nfsv3err_read, 549 nfsv3err_write, 550 nfsv3err_create, 551 nfsv3err_mkdir, 552 nfsv3err_symlink, 553 nfsv3err_mknod, 554 nfsv3err_remove, 555 nfsv3err_rmdir, 556 nfsv3err_rename, 557 nfsv3err_link, 558 nfsv3err_readdir, 559 nfsv3err_readdirplus, 560 nfsv3err_fsstat, 561 nfsv3err_fsinfo, 562 nfsv3err_pathconf, 563 nfsv3err_commit, 564 }; 565 566 extern struct nfsrtt nfsrtt; 567 extern struct nfsnodehashhead *nfsnodehashtbl; 568 extern u_long nfsnodehash; 569 570 u_long nfsdirhashmask; 571 572 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *)); 573 574 /* 575 * Create the header for an rpc request packet 576 * The hsiz is the size of the rest of the nfs request header. 577 * (just used to decide if a cluster is a good idea) 578 */ 579 struct mbuf * 580 nfsm_reqh(struct nfsnode *np, u_long procid, int hsiz, char **bposp) 581 { 582 struct mbuf *mb; 583 char *bpos; 584 585 mb = m_get(M_WAIT, MT_DATA); 586 MCLAIM(mb, &nfs_mowner); 587 if (hsiz >= MINCLSIZE) 588 m_clget(mb, M_WAIT); 589 mb->m_len = 0; 590 bpos = mtod(mb, void *); 591 592 /* Finally, return values */ 593 *bposp = bpos; 594 return (mb); 595 } 596 597 /* 598 * Build the RPC header and fill in the authorization info. 599 * The authorization string argument is only used when the credentials 600 * come from outside of the kernel. 601 * Returns the head of the mbuf list. 602 */ 603 struct mbuf * 604 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len, 605 verf_str, mrest, mrest_len, mbp, xidp) 606 kauth_cred_t cr; 607 int nmflag; 608 int procid; 609 int auth_type; 610 int auth_len; 611 char *auth_str; 612 int verf_len; 613 char *verf_str; 614 struct mbuf *mrest; 615 int mrest_len; 616 struct mbuf **mbp; 617 u_int32_t *xidp; 618 { 619 struct mbuf *mb; 620 u_int32_t *tl; 621 char *bpos; 622 int i; 623 struct mbuf *mreq; 624 int siz, grpsiz, authsiz; 625 626 authsiz = nfsm_rndup(auth_len); 627 mb = m_gethdr(M_WAIT, MT_DATA); 628 MCLAIM(mb, &nfs_mowner); 629 if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) { 630 m_clget(mb, M_WAIT); 631 } else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) { 632 MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED); 633 } else { 634 MH_ALIGN(mb, 8 * NFSX_UNSIGNED); 635 } 636 mb->m_len = 0; 637 mreq = mb; 638 bpos = mtod(mb, void *); 639 640 /* 641 * First the RPC header. 642 */ 643 nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 644 645 *tl++ = *xidp = nfs_getxid(); 646 *tl++ = rpc_call; 647 *tl++ = rpc_vers; 648 *tl++ = txdr_unsigned(NFS_PROG); 649 if (nmflag & NFSMNT_NFSV3) 650 *tl++ = txdr_unsigned(NFS_VER3); 651 else 652 *tl++ = txdr_unsigned(NFS_VER2); 653 if (nmflag & NFSMNT_NFSV3) 654 *tl++ = txdr_unsigned(procid); 655 else 656 *tl++ = txdr_unsigned(nfsv2_procid[procid]); 657 658 /* 659 * And then the authorization cred. 660 */ 661 *tl++ = txdr_unsigned(auth_type); 662 *tl = txdr_unsigned(authsiz); 663 switch (auth_type) { 664 case RPCAUTH_UNIX: 665 nfsm_build(tl, u_int32_t *, auth_len); 666 *tl++ = 0; /* stamp ?? */ 667 *tl++ = 0; /* NULL hostname */ 668 *tl++ = txdr_unsigned(kauth_cred_geteuid(cr)); 669 *tl++ = txdr_unsigned(kauth_cred_getegid(cr)); 670 grpsiz = (auth_len >> 2) - 5; 671 *tl++ = txdr_unsigned(grpsiz); 672 for (i = 0; i < grpsiz; i++) 673 *tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */ 674 break; 675 case RPCAUTH_KERB4: 676 siz = auth_len; 677 while (siz > 0) { 678 if (M_TRAILINGSPACE(mb) == 0) { 679 struct mbuf *mb2; 680 mb2 = m_get(M_WAIT, MT_DATA); 681 MCLAIM(mb2, &nfs_mowner); 682 if (siz >= MINCLSIZE) 683 m_clget(mb2, M_WAIT); 684 mb->m_next = mb2; 685 mb = mb2; 686 mb->m_len = 0; 687 bpos = mtod(mb, void *); 688 } 689 i = min(siz, M_TRAILINGSPACE(mb)); 690 memcpy(bpos, auth_str, i); 691 mb->m_len += i; 692 auth_str += i; 693 bpos += i; 694 siz -= i; 695 } 696 if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) { 697 for (i = 0; i < siz; i++) 698 *bpos++ = '\0'; 699 mb->m_len += siz; 700 } 701 break; 702 }; 703 704 /* 705 * And the verifier... 706 */ 707 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 708 if (verf_str) { 709 *tl++ = txdr_unsigned(RPCAUTH_KERB4); 710 *tl = txdr_unsigned(verf_len); 711 siz = verf_len; 712 while (siz > 0) { 713 if (M_TRAILINGSPACE(mb) == 0) { 714 struct mbuf *mb2; 715 mb2 = m_get(M_WAIT, MT_DATA); 716 MCLAIM(mb2, &nfs_mowner); 717 if (siz >= MINCLSIZE) 718 m_clget(mb2, M_WAIT); 719 mb->m_next = mb2; 720 mb = mb2; 721 mb->m_len = 0; 722 bpos = mtod(mb, void *); 723 } 724 i = min(siz, M_TRAILINGSPACE(mb)); 725 memcpy(bpos, verf_str, i); 726 mb->m_len += i; 727 verf_str += i; 728 bpos += i; 729 siz -= i; 730 } 731 if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) { 732 for (i = 0; i < siz; i++) 733 *bpos++ = '\0'; 734 mb->m_len += siz; 735 } 736 } else { 737 *tl++ = txdr_unsigned(RPCAUTH_NULL); 738 *tl = 0; 739 } 740 mb->m_next = mrest; 741 mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len; 742 mreq->m_pkthdr.rcvif = (struct ifnet *)0; 743 *mbp = mb; 744 return (mreq); 745 } 746 747 /* 748 * copies mbuf chain to the uio scatter/gather list 749 */ 750 int 751 nfsm_mbuftouio(mrep, uiop, siz, dpos) 752 struct mbuf **mrep; 753 struct uio *uiop; 754 int siz; 755 char **dpos; 756 { 757 char *mbufcp, *uiocp; 758 int xfer, left, len; 759 struct mbuf *mp; 760 long uiosiz, rem; 761 int error = 0; 762 763 mp = *mrep; 764 mbufcp = *dpos; 765 len = mtod(mp, char *) + mp->m_len - mbufcp; 766 rem = nfsm_rndup(siz)-siz; 767 while (siz > 0) { 768 if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) 769 return (EFBIG); 770 left = uiop->uio_iov->iov_len; 771 uiocp = uiop->uio_iov->iov_base; 772 if (left > siz) 773 left = siz; 774 uiosiz = left; 775 while (left > 0) { 776 while (len == 0) { 777 mp = mp->m_next; 778 if (mp == NULL) 779 return (EBADRPC); 780 mbufcp = mtod(mp, void *); 781 len = mp->m_len; 782 } 783 xfer = (left > len) ? len : left; 784 error = copyout_vmspace(uiop->uio_vmspace, mbufcp, 785 uiocp, xfer); 786 if (error) { 787 return error; 788 } 789 left -= xfer; 790 len -= xfer; 791 mbufcp += xfer; 792 uiocp += xfer; 793 uiop->uio_offset += xfer; 794 uiop->uio_resid -= xfer; 795 } 796 if (uiop->uio_iov->iov_len <= siz) { 797 uiop->uio_iovcnt--; 798 uiop->uio_iov++; 799 } else { 800 uiop->uio_iov->iov_base = 801 (char *)uiop->uio_iov->iov_base + uiosiz; 802 uiop->uio_iov->iov_len -= uiosiz; 803 } 804 siz -= uiosiz; 805 } 806 *dpos = mbufcp; 807 *mrep = mp; 808 if (rem > 0) { 809 if (len < rem) 810 error = nfs_adv(mrep, dpos, rem, len); 811 else 812 *dpos += rem; 813 } 814 return (error); 815 } 816 817 /* 818 * copies a uio scatter/gather list to an mbuf chain. 819 * NOTE: can ony handle iovcnt == 1 820 */ 821 int 822 nfsm_uiotombuf(uiop, mq, siz, bpos) 823 struct uio *uiop; 824 struct mbuf **mq; 825 int siz; 826 char **bpos; 827 { 828 char *uiocp; 829 struct mbuf *mp, *mp2; 830 int xfer, left, mlen; 831 int uiosiz, clflg, rem; 832 char *cp; 833 int error; 834 835 #ifdef DIAGNOSTIC 836 if (uiop->uio_iovcnt != 1) 837 panic("nfsm_uiotombuf: iovcnt != 1"); 838 #endif 839 840 if (siz > MLEN) /* or should it >= MCLBYTES ?? */ 841 clflg = 1; 842 else 843 clflg = 0; 844 rem = nfsm_rndup(siz)-siz; 845 mp = mp2 = *mq; 846 while (siz > 0) { 847 left = uiop->uio_iov->iov_len; 848 uiocp = uiop->uio_iov->iov_base; 849 if (left > siz) 850 left = siz; 851 uiosiz = left; 852 while (left > 0) { 853 mlen = M_TRAILINGSPACE(mp); 854 if (mlen == 0) { 855 mp = m_get(M_WAIT, MT_DATA); 856 MCLAIM(mp, &nfs_mowner); 857 if (clflg) 858 m_clget(mp, M_WAIT); 859 mp->m_len = 0; 860 mp2->m_next = mp; 861 mp2 = mp; 862 mlen = M_TRAILINGSPACE(mp); 863 } 864 xfer = (left > mlen) ? mlen : left; 865 cp = mtod(mp, char *) + mp->m_len; 866 error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp, 867 xfer); 868 if (error) { 869 /* XXX */ 870 } 871 mp->m_len += xfer; 872 left -= xfer; 873 uiocp += xfer; 874 uiop->uio_offset += xfer; 875 uiop->uio_resid -= xfer; 876 } 877 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + 878 uiosiz; 879 uiop->uio_iov->iov_len -= uiosiz; 880 siz -= uiosiz; 881 } 882 if (rem > 0) { 883 if (rem > M_TRAILINGSPACE(mp)) { 884 mp = m_get(M_WAIT, MT_DATA); 885 MCLAIM(mp, &nfs_mowner); 886 mp->m_len = 0; 887 mp2->m_next = mp; 888 } 889 cp = mtod(mp, char *) + mp->m_len; 890 for (left = 0; left < rem; left++) 891 *cp++ = '\0'; 892 mp->m_len += rem; 893 *bpos = cp; 894 } else 895 *bpos = mtod(mp, char *) + mp->m_len; 896 *mq = mp; 897 return (0); 898 } 899 900 /* 901 * Get at least "siz" bytes of correctly aligned data. 902 * When called the mbuf pointers are not necessarily correct, 903 * dsosp points to what ought to be in m_data and left contains 904 * what ought to be in m_len. 905 * This is used by the macros nfsm_dissect and nfsm_dissecton for tough 906 * cases. (The macros use the vars. dpos and dpos2) 907 */ 908 int 909 nfsm_disct(mdp, dposp, siz, left, cp2) 910 struct mbuf **mdp; 911 char **dposp; 912 int siz; 913 int left; 914 char **cp2; 915 { 916 struct mbuf *m1, *m2; 917 struct mbuf *havebuf = NULL; 918 char *src = *dposp; 919 char *dst; 920 int len; 921 922 #ifdef DEBUG 923 if (left < 0) 924 panic("nfsm_disct: left < 0"); 925 #endif 926 m1 = *mdp; 927 /* 928 * Skip through the mbuf chain looking for an mbuf with 929 * some data. If the first mbuf found has enough data 930 * and it is correctly aligned return it. 931 */ 932 while (left == 0) { 933 havebuf = m1; 934 *mdp = m1 = m1->m_next; 935 if (m1 == NULL) 936 return (EBADRPC); 937 src = mtod(m1, void *); 938 left = m1->m_len; 939 /* 940 * If we start a new mbuf and it is big enough 941 * and correctly aligned just return it, don't 942 * do any pull up. 943 */ 944 if (left >= siz && nfsm_aligned(src)) { 945 *cp2 = src; 946 *dposp = src + siz; 947 return (0); 948 } 949 } 950 if ((m1->m_flags & M_EXT) != 0) { 951 if (havebuf && M_TRAILINGSPACE(havebuf) >= siz && 952 nfsm_aligned(mtod(havebuf, char *) + havebuf->m_len)) { 953 /* 954 * If the first mbuf with data has external data 955 * and there is a previous mbuf with some trailing 956 * space, use it to move the data into. 957 */ 958 m2 = m1; 959 *mdp = m1 = havebuf; 960 *cp2 = mtod(m1, char *) + m1->m_len; 961 } else if (havebuf) { 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 = m1; 970 *mdp = m1 = m_get(M_WAIT, MT_DATA); 971 MCLAIM(m1, m2->m_owner); 972 if ((m2->m_flags & M_PKTHDR) != 0) { 973 /* XXX MOVE */ 974 M_COPY_PKTHDR(m1, m2); 975 m_tag_delete_chain(m2, NULL); 976 m2->m_flags &= ~M_PKTHDR; 977 } 978 if (havebuf) { 979 havebuf->m_next = m1; 980 } 981 m1->m_next = m2; 982 MRESETDATA(m1); 983 m1->m_len = 0; 984 m2->m_data = src; 985 m2->m_len = left; 986 *cp2 = mtod(m1, char *); 987 } else { 988 struct mbuf **nextp = &m1->m_next; 989 990 m1->m_len -= left; 991 do { 992 m2 = m_get(M_WAIT, MT_DATA); 993 MCLAIM(m2, m1->m_owner); 994 if (left >= MINCLSIZE) { 995 MCLGET(m2, M_WAIT); 996 } 997 m2->m_next = *nextp; 998 *nextp = m2; 999 nextp = &m2->m_next; 1000 len = (m2->m_flags & M_EXT) != 0 ? 1001 MCLBYTES : MLEN; 1002 if (len > left) { 1003 len = left; 1004 } 1005 memcpy(mtod(m2, char *), src, len); 1006 m2->m_len = len; 1007 src += len; 1008 left -= len; 1009 } while (left > 0); 1010 *mdp = m1 = m1->m_next; 1011 m2 = m1->m_next; 1012 *cp2 = mtod(m1, char *); 1013 } 1014 } else { 1015 /* 1016 * If the first mbuf has no external data 1017 * move the data to the front of the mbuf. 1018 */ 1019 MRESETDATA(m1); 1020 dst = mtod(m1, char *); 1021 if (dst != src) { 1022 memmove(dst, src, left); 1023 } 1024 m1->m_len = left; 1025 m2 = m1->m_next; 1026 *cp2 = m1->m_data; 1027 } 1028 *dposp = *cp2 + siz; 1029 /* 1030 * Loop through mbufs pulling data up into first mbuf until 1031 * the first mbuf is full or there is no more data to 1032 * pullup. 1033 */ 1034 dst = mtod(m1, char *) + m1->m_len; 1035 while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) { 1036 if ((len = min(len, m2->m_len)) != 0) { 1037 memcpy(dst, mtod(m2, char *), len); 1038 } 1039 m1->m_len += len; 1040 dst += len; 1041 m2->m_data += len; 1042 m2->m_len -= len; 1043 m2 = m2->m_next; 1044 } 1045 if (m1->m_len < siz) 1046 return (EBADRPC); 1047 return (0); 1048 } 1049 1050 /* 1051 * Advance the position in the mbuf chain. 1052 */ 1053 int 1054 nfs_adv(mdp, dposp, offs, left) 1055 struct mbuf **mdp; 1056 char **dposp; 1057 int offs; 1058 int left; 1059 { 1060 struct mbuf *m; 1061 int s; 1062 1063 m = *mdp; 1064 s = left; 1065 while (s < offs) { 1066 offs -= s; 1067 m = m->m_next; 1068 if (m == NULL) 1069 return (EBADRPC); 1070 s = m->m_len; 1071 } 1072 *mdp = m; 1073 *dposp = mtod(m, char *) + offs; 1074 return (0); 1075 } 1076 1077 /* 1078 * Copy a string into mbufs for the hard cases... 1079 */ 1080 int 1081 nfsm_strtmbuf(mb, bpos, cp, siz) 1082 struct mbuf **mb; 1083 char **bpos; 1084 const char *cp; 1085 long siz; 1086 { 1087 struct mbuf *m1 = NULL, *m2; 1088 long left, xfer, len, tlen; 1089 u_int32_t *tl; 1090 int putsize; 1091 1092 putsize = 1; 1093 m2 = *mb; 1094 left = M_TRAILINGSPACE(m2); 1095 if (left > 0) { 1096 tl = ((u_int32_t *)(*bpos)); 1097 *tl++ = txdr_unsigned(siz); 1098 putsize = 0; 1099 left -= NFSX_UNSIGNED; 1100 m2->m_len += NFSX_UNSIGNED; 1101 if (left > 0) { 1102 memcpy((void *) tl, cp, left); 1103 siz -= left; 1104 cp += left; 1105 m2->m_len += left; 1106 left = 0; 1107 } 1108 } 1109 /* Loop around adding mbufs */ 1110 while (siz > 0) { 1111 m1 = m_get(M_WAIT, MT_DATA); 1112 MCLAIM(m1, &nfs_mowner); 1113 if (siz > MLEN) 1114 m_clget(m1, M_WAIT); 1115 m1->m_len = NFSMSIZ(m1); 1116 m2->m_next = m1; 1117 m2 = m1; 1118 tl = mtod(m1, u_int32_t *); 1119 tlen = 0; 1120 if (putsize) { 1121 *tl++ = txdr_unsigned(siz); 1122 m1->m_len -= NFSX_UNSIGNED; 1123 tlen = NFSX_UNSIGNED; 1124 putsize = 0; 1125 } 1126 if (siz < m1->m_len) { 1127 len = nfsm_rndup(siz); 1128 xfer = siz; 1129 if (xfer < len) 1130 *(tl+(xfer>>2)) = 0; 1131 } else { 1132 xfer = len = m1->m_len; 1133 } 1134 memcpy((void *) tl, cp, xfer); 1135 m1->m_len = len+tlen; 1136 siz -= xfer; 1137 cp += xfer; 1138 } 1139 *mb = m1; 1140 *bpos = mtod(m1, char *) + m1->m_len; 1141 return (0); 1142 } 1143 1144 /* 1145 * Directory caching routines. They work as follows: 1146 * - a cache is maintained per VDIR nfsnode. 1147 * - for each offset cookie that is exported to userspace, and can 1148 * thus be thrown back at us as an offset to VOP_READDIR, store 1149 * information in the cache. 1150 * - cached are: 1151 * - cookie itself 1152 * - blocknumber (essentially just a search key in the buffer cache) 1153 * - entry number in block. 1154 * - offset cookie of block in which this entry is stored 1155 * - 32 bit cookie if NFSMNT_XLATECOOKIE is used. 1156 * - entries are looked up in a hash table 1157 * - also maintained is an LRU list of entries, used to determine 1158 * which ones to delete if the cache grows too large. 1159 * - if 32 <-> 64 translation mode is requested for a filesystem, 1160 * the cache also functions as a translation table 1161 * - in the translation case, invalidating the cache does not mean 1162 * flushing it, but just marking entries as invalid, except for 1163 * the <64bit cookie, 32bitcookie> pair which is still valid, to 1164 * still be able to use the cache as a translation table. 1165 * - 32 bit cookies are uniquely created by combining the hash table 1166 * entry value, and one generation count per hash table entry, 1167 * incremented each time an entry is appended to the chain. 1168 * - the cache is invalidated each time a direcory is modified 1169 * - sanity checks are also done; if an entry in a block turns 1170 * out not to have a matching cookie, the cache is invalidated 1171 * and a new block starting from the wanted offset is fetched from 1172 * the server. 1173 * - directory entries as read from the server are extended to contain 1174 * the 64bit and, optionally, the 32bit cookies, for sanity checking 1175 * the cache and exporting them to userspace through the cookie 1176 * argument to VOP_READDIR. 1177 */ 1178 1179 u_long 1180 nfs_dirhash(off) 1181 off_t off; 1182 { 1183 int i; 1184 char *cp = (char *)&off; 1185 u_long sum = 0L; 1186 1187 for (i = 0 ; i < sizeof (off); i++) 1188 sum += *cp++; 1189 1190 return sum; 1191 } 1192 1193 #define _NFSDC_MTX(np) (&NFSTOV(np)->v_interlock) 1194 #define NFSDC_LOCK(np) simple_lock(_NFSDC_MTX(np)) 1195 #define NFSDC_UNLOCK(np) simple_unlock(_NFSDC_MTX(np)) 1196 #define NFSDC_ASSERT_LOCKED(np) LOCK_ASSERT(simple_lock_held(_NFSDC_MTX(np))) 1197 1198 void 1199 nfs_initdircache(vp) 1200 struct vnode *vp; 1201 { 1202 struct nfsnode *np = VTONFS(vp); 1203 struct nfsdirhashhead *dircache; 1204 1205 dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF, 1206 M_WAITOK, &nfsdirhashmask); 1207 1208 NFSDC_LOCK(np); 1209 if (np->n_dircache == NULL) { 1210 np->n_dircachesize = 0; 1211 np->n_dircache = dircache; 1212 dircache = NULL; 1213 TAILQ_INIT(&np->n_dirchain); 1214 } 1215 NFSDC_UNLOCK(np); 1216 if (dircache) 1217 hashdone(dircache, M_NFSDIROFF); 1218 } 1219 1220 void 1221 nfs_initdirxlatecookie(vp) 1222 struct vnode *vp; 1223 { 1224 struct nfsnode *np = VTONFS(vp); 1225 unsigned *dirgens; 1226 1227 KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE); 1228 1229 dirgens = malloc(NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF, 1230 M_WAITOK|M_ZERO); 1231 NFSDC_LOCK(np); 1232 if (np->n_dirgens == NULL) { 1233 np->n_dirgens = dirgens; 1234 dirgens = NULL; 1235 } 1236 NFSDC_UNLOCK(np); 1237 if (dirgens) 1238 free(dirgens, M_NFSDIROFF); 1239 } 1240 1241 static const struct nfsdircache dzero; 1242 1243 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *)); 1244 static void nfs_putdircache_unlocked __P((struct nfsnode *, 1245 struct nfsdircache *)); 1246 1247 static void 1248 nfs_unlinkdircache(np, ndp) 1249 struct nfsnode *np; 1250 struct nfsdircache *ndp; 1251 { 1252 1253 NFSDC_ASSERT_LOCKED(np); 1254 KASSERT(ndp != &dzero); 1255 1256 if (LIST_NEXT(ndp, dc_hash) == (void *)-1) 1257 return; 1258 1259 TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain); 1260 LIST_REMOVE(ndp, dc_hash); 1261 LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */ 1262 1263 nfs_putdircache_unlocked(np, ndp); 1264 } 1265 1266 void 1267 nfs_putdircache(np, ndp) 1268 struct nfsnode *np; 1269 struct nfsdircache *ndp; 1270 { 1271 int ref; 1272 1273 if (ndp == &dzero) 1274 return; 1275 1276 KASSERT(ndp->dc_refcnt > 0); 1277 NFSDC_LOCK(np); 1278 ref = --ndp->dc_refcnt; 1279 NFSDC_UNLOCK(np); 1280 1281 if (ref == 0) 1282 free(ndp, M_NFSDIROFF); 1283 } 1284 1285 static void 1286 nfs_putdircache_unlocked(struct nfsnode *np, struct nfsdircache *ndp) 1287 { 1288 int ref; 1289 1290 NFSDC_ASSERT_LOCKED(np); 1291 1292 if (ndp == &dzero) 1293 return; 1294 1295 KASSERT(ndp->dc_refcnt > 0); 1296 ref = --ndp->dc_refcnt; 1297 if (ref == 0) 1298 free(ndp, M_NFSDIROFF); 1299 } 1300 1301 struct nfsdircache * 1302 nfs_searchdircache(vp, off, do32, hashent) 1303 struct vnode *vp; 1304 off_t off; 1305 int do32; 1306 int *hashent; 1307 { 1308 struct nfsdirhashhead *ndhp; 1309 struct nfsdircache *ndp = NULL; 1310 struct nfsnode *np = VTONFS(vp); 1311 unsigned ent; 1312 1313 /* 1314 * Zero is always a valid cookie. 1315 */ 1316 if (off == 0) 1317 /* XXXUNCONST */ 1318 return (struct nfsdircache *)__UNCONST(&dzero); 1319 1320 if (!np->n_dircache) 1321 return NULL; 1322 1323 /* 1324 * We use a 32bit cookie as search key, directly reconstruct 1325 * the hashentry. Else use the hashfunction. 1326 */ 1327 if (do32) { 1328 ent = (u_int32_t)off >> 24; 1329 if (ent >= NFS_DIRHASHSIZ) 1330 return NULL; 1331 ndhp = &np->n_dircache[ent]; 1332 } else { 1333 ndhp = NFSDIRHASH(np, off); 1334 } 1335 1336 if (hashent) 1337 *hashent = (int)(ndhp - np->n_dircache); 1338 1339 NFSDC_LOCK(np); 1340 if (do32) { 1341 LIST_FOREACH(ndp, ndhp, dc_hash) { 1342 if (ndp->dc_cookie32 == (u_int32_t)off) { 1343 /* 1344 * An invalidated entry will become the 1345 * start of a new block fetched from 1346 * the server. 1347 */ 1348 if (ndp->dc_flags & NFSDC_INVALID) { 1349 ndp->dc_blkcookie = ndp->dc_cookie; 1350 ndp->dc_entry = 0; 1351 ndp->dc_flags &= ~NFSDC_INVALID; 1352 } 1353 break; 1354 } 1355 } 1356 } else { 1357 LIST_FOREACH(ndp, ndhp, dc_hash) { 1358 if (ndp->dc_cookie == off) 1359 break; 1360 } 1361 } 1362 if (ndp != NULL) 1363 ndp->dc_refcnt++; 1364 NFSDC_UNLOCK(np); 1365 return ndp; 1366 } 1367 1368 1369 struct nfsdircache * 1370 nfs_enterdircache(struct vnode *vp, off_t off, off_t blkoff, int en, 1371 daddr_t blkno) 1372 { 1373 struct nfsnode *np = VTONFS(vp); 1374 struct nfsdirhashhead *ndhp; 1375 struct nfsdircache *ndp = NULL; 1376 struct nfsdircache *newndp = NULL; 1377 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1378 int hashent = 0, gen, overwrite; /* XXX: GCC */ 1379 1380 /* 1381 * XXX refuse entries for offset 0. amd(8) erroneously sets 1382 * cookie 0 for the '.' entry, making this necessary. This 1383 * isn't so bad, as 0 is a special case anyway. 1384 */ 1385 if (off == 0) 1386 /* XXXUNCONST */ 1387 return (struct nfsdircache *)__UNCONST(&dzero); 1388 1389 if (!np->n_dircache) 1390 /* 1391 * XXX would like to do this in nfs_nget but vtype 1392 * isn't known at that time. 1393 */ 1394 nfs_initdircache(vp); 1395 1396 if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens) 1397 nfs_initdirxlatecookie(vp); 1398 1399 retry: 1400 ndp = nfs_searchdircache(vp, off, 0, &hashent); 1401 1402 NFSDC_LOCK(np); 1403 if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) { 1404 /* 1405 * Overwriting an old entry. Check if it's the same. 1406 * If so, just return. If not, remove the old entry. 1407 */ 1408 if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en) 1409 goto done; 1410 nfs_unlinkdircache(np, ndp); 1411 nfs_putdircache_unlocked(np, ndp); 1412 ndp = NULL; 1413 } 1414 1415 ndhp = &np->n_dircache[hashent]; 1416 1417 if (!ndp) { 1418 if (newndp == NULL) { 1419 NFSDC_UNLOCK(np); 1420 newndp = malloc(sizeof(*ndp), M_NFSDIROFF, M_WAITOK); 1421 newndp->dc_refcnt = 1; 1422 LIST_NEXT(newndp, dc_hash) = (void *)-1; 1423 goto retry; 1424 } 1425 ndp = newndp; 1426 newndp = NULL; 1427 overwrite = 0; 1428 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 1429 /* 1430 * We're allocating a new entry, so bump the 1431 * generation number. 1432 */ 1433 KASSERT(np->n_dirgens); 1434 gen = ++np->n_dirgens[hashent]; 1435 if (gen == 0) { 1436 np->n_dirgens[hashent]++; 1437 gen++; 1438 } 1439 ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff); 1440 } 1441 } else 1442 overwrite = 1; 1443 1444 ndp->dc_cookie = off; 1445 ndp->dc_blkcookie = blkoff; 1446 ndp->dc_entry = en; 1447 ndp->dc_flags = 0; 1448 1449 if (overwrite) 1450 goto done; 1451 1452 /* 1453 * If the maximum directory cookie cache size has been reached 1454 * for this node, take one off the front. The idea is that 1455 * directories are typically read front-to-back once, so that 1456 * the oldest entries can be thrown away without much performance 1457 * loss. 1458 */ 1459 if (np->n_dircachesize == NFS_MAXDIRCACHE) { 1460 nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain)); 1461 } else 1462 np->n_dircachesize++; 1463 1464 KASSERT(ndp->dc_refcnt == 1); 1465 LIST_INSERT_HEAD(ndhp, ndp, dc_hash); 1466 TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain); 1467 ndp->dc_refcnt++; 1468 done: 1469 KASSERT(ndp->dc_refcnt > 0); 1470 NFSDC_UNLOCK(np); 1471 if (newndp) 1472 nfs_putdircache(np, newndp); 1473 return ndp; 1474 } 1475 1476 void 1477 nfs_invaldircache(vp, flags) 1478 struct vnode *vp; 1479 int flags; 1480 { 1481 struct nfsnode *np = VTONFS(vp); 1482 struct nfsdircache *ndp = NULL; 1483 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1484 const bool forcefree = flags & NFS_INVALDIRCACHE_FORCE; 1485 1486 #ifdef DIAGNOSTIC 1487 if (vp->v_type != VDIR) 1488 panic("nfs: invaldircache: not dir"); 1489 #endif 1490 1491 if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0) 1492 np->n_flag &= ~NEOFVALID; 1493 1494 if (!np->n_dircache) 1495 return; 1496 1497 NFSDC_LOCK(np); 1498 if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) { 1499 while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) { 1500 KASSERT(!forcefree || ndp->dc_refcnt == 1); 1501 nfs_unlinkdircache(np, ndp); 1502 } 1503 np->n_dircachesize = 0; 1504 if (forcefree && np->n_dirgens) { 1505 FREE(np->n_dirgens, M_NFSDIROFF); 1506 np->n_dirgens = NULL; 1507 } 1508 } else { 1509 TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain) 1510 ndp->dc_flags |= NFSDC_INVALID; 1511 } 1512 1513 NFSDC_UNLOCK(np); 1514 } 1515 1516 /* 1517 * Called once before VFS init to initialize shared and 1518 * server-specific data structures. 1519 */ 1520 static int 1521 nfs_init0(void) 1522 { 1523 1524 nfsrtt.pos = 0; 1525 rpc_vers = txdr_unsigned(RPC_VER2); 1526 rpc_call = txdr_unsigned(RPC_CALL); 1527 rpc_reply = txdr_unsigned(RPC_REPLY); 1528 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED); 1529 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED); 1530 rpc_mismatch = txdr_unsigned(RPC_MISMATCH); 1531 rpc_autherr = txdr_unsigned(RPC_AUTHERR); 1532 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX); 1533 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4); 1534 nfs_prog = txdr_unsigned(NFS_PROG); 1535 nfs_true = txdr_unsigned(true); 1536 nfs_false = txdr_unsigned(false); 1537 nfs_xdrneg1 = txdr_unsigned(-1); 1538 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; 1539 if (nfs_ticks < 1) 1540 nfs_ticks = 1; 1541 #ifdef NFSSERVER 1542 nfsrv_init(0); /* Init server data structures */ 1543 nfsrv_initcache(); /* Init the server request cache */ 1544 { 1545 extern krwlock_t netexport_lock; /* XXX */ 1546 rw_init(&netexport_lock); 1547 } 1548 #endif /* NFSSERVER */ 1549 1550 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) 1551 nfsdreq_init(); 1552 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */ 1553 1554 /* 1555 * Initialize reply list and start timer 1556 */ 1557 TAILQ_INIT(&nfs_reqq); 1558 nfs_timer_init(); 1559 MOWNER_ATTACH(&nfs_mowner); 1560 1561 #ifdef NFS 1562 /* Initialize the kqueue structures */ 1563 nfs_kqinit(); 1564 /* Initialize the iod structures */ 1565 nfs_iodinit(); 1566 #endif 1567 return 0; 1568 } 1569 1570 void 1571 nfs_init(void) 1572 { 1573 static ONCE_DECL(nfs_init_once); 1574 1575 RUN_ONCE(&nfs_init_once, nfs_init0); 1576 } 1577 1578 #ifdef NFS 1579 /* 1580 * Called once at VFS init to initialize client-specific data structures. 1581 */ 1582 void 1583 nfs_vfs_init() 1584 { 1585 /* Initialize NFS server / client shared data. */ 1586 nfs_init(); 1587 1588 nfs_nhinit(); /* Init the nfsnode table */ 1589 nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4); 1590 } 1591 1592 void 1593 nfs_vfs_reinit() 1594 { 1595 nfs_nhreinit(); 1596 } 1597 1598 void 1599 nfs_vfs_done() 1600 { 1601 nfs_nhdone(); 1602 } 1603 1604 /* 1605 * Attribute cache routines. 1606 * nfs_loadattrcache() - loads or updates the cache contents from attributes 1607 * that are on the mbuf list 1608 * nfs_getattrcache() - returns valid attributes if found in cache, returns 1609 * error otherwise 1610 */ 1611 1612 /* 1613 * Load the attribute cache (that lives in the nfsnode entry) with 1614 * the values on the mbuf list and 1615 * Iff vap not NULL 1616 * copy the attributes to *vaper 1617 */ 1618 int 1619 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags) 1620 struct vnode **vpp; 1621 struct mbuf **mdp; 1622 char **dposp; 1623 struct vattr *vaper; 1624 int flags; 1625 { 1626 int32_t t1; 1627 char *cp2; 1628 int error = 0; 1629 struct mbuf *md; 1630 int v3 = NFS_ISV3(*vpp); 1631 1632 md = *mdp; 1633 t1 = (mtod(md, char *) + md->m_len) - *dposp; 1634 error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2); 1635 if (error) 1636 return (error); 1637 return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags); 1638 } 1639 1640 int 1641 nfs_loadattrcache(vpp, fp, vaper, flags) 1642 struct vnode **vpp; 1643 struct nfs_fattr *fp; 1644 struct vattr *vaper; 1645 int flags; 1646 { 1647 struct vnode *vp = *vpp; 1648 struct vattr *vap; 1649 int v3 = NFS_ISV3(vp); 1650 enum vtype vtyp; 1651 u_short vmode; 1652 struct timespec mtime; 1653 struct timespec ctime; 1654 struct vnode *nvp; 1655 int32_t rdev; 1656 struct nfsnode *np; 1657 extern int (**spec_nfsv2nodeop_p) __P((void *)); 1658 uid_t uid; 1659 gid_t gid; 1660 1661 if (v3) { 1662 vtyp = nfsv3tov_type(fp->fa_type); 1663 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1664 rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1), 1665 fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2)); 1666 fxdr_nfsv3time(&fp->fa3_mtime, &mtime); 1667 fxdr_nfsv3time(&fp->fa3_ctime, &ctime); 1668 } else { 1669 vtyp = nfsv2tov_type(fp->fa_type); 1670 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1671 if (vtyp == VNON || vtyp == VREG) 1672 vtyp = IFTOVT(vmode); 1673 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); 1674 fxdr_nfsv2time(&fp->fa2_mtime, &mtime); 1675 ctime.tv_sec = fxdr_unsigned(u_int32_t, 1676 fp->fa2_ctime.nfsv2_sec); 1677 ctime.tv_nsec = 0; 1678 1679 /* 1680 * Really ugly NFSv2 kludge. 1681 */ 1682 if (vtyp == VCHR && rdev == 0xffffffff) 1683 vtyp = VFIFO; 1684 } 1685 1686 vmode &= ALLPERMS; 1687 1688 /* 1689 * If v_type == VNON it is a new node, so fill in the v_type, 1690 * n_mtime fields. Check to see if it represents a special 1691 * device, and if so, check for a possible alias. Once the 1692 * correct vnode has been obtained, fill in the rest of the 1693 * information. 1694 */ 1695 np = VTONFS(vp); 1696 if (vp->v_type == VNON) { 1697 vp->v_type = vtyp; 1698 if (vp->v_type == VFIFO) { 1699 extern int (**fifo_nfsv2nodeop_p) __P((void *)); 1700 vp->v_op = fifo_nfsv2nodeop_p; 1701 } else if (vp->v_type == VREG) { 1702 mutex_init(&np->n_commitlock, MUTEX_DEFAULT, IPL_NONE); 1703 } else if (vp->v_type == VCHR || vp->v_type == VBLK) { 1704 vp->v_op = spec_nfsv2nodeop_p; 1705 nvp = checkalias(vp, (dev_t)rdev, vp->v_mount); 1706 if (nvp) { 1707 /* 1708 * Discard unneeded vnode, but save its nfsnode. 1709 * Since the nfsnode does not have a lock, its 1710 * vnode lock has to be carried over. 1711 */ 1712 /* 1713 * XXX is the old node sure to be locked here? 1714 */ 1715 KASSERT(lockstatus(&vp->v_lock) == 1716 LK_EXCLUSIVE); 1717 nvp->v_data = vp->v_data; 1718 vp->v_data = NULL; 1719 VOP_UNLOCK(vp, 0); 1720 vp->v_op = spec_vnodeop_p; 1721 vrele(vp); 1722 vgone(vp); 1723 lockmgr(&nvp->v_lock, LK_EXCLUSIVE, 1724 &nvp->v_interlock); 1725 /* 1726 * Reinitialize aliased node. 1727 */ 1728 np->n_vnode = nvp; 1729 *vpp = vp = nvp; 1730 } 1731 } 1732 np->n_mtime = mtime; 1733 } 1734 uid = fxdr_unsigned(uid_t, fp->fa_uid); 1735 gid = fxdr_unsigned(gid_t, fp->fa_gid); 1736 vap = np->n_vattr; 1737 1738 /* 1739 * Invalidate access cache if uid, gid, mode or ctime changed. 1740 */ 1741 if (np->n_accstamp != -1 && 1742 (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode 1743 || timespeccmp(&ctime, &vap->va_ctime, !=))) 1744 np->n_accstamp = -1; 1745 1746 vap->va_type = vtyp; 1747 vap->va_mode = vmode; 1748 vap->va_rdev = (dev_t)rdev; 1749 vap->va_mtime = mtime; 1750 vap->va_ctime = ctime; 1751 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 1752 switch (vtyp) { 1753 case VDIR: 1754 vap->va_blocksize = NFS_DIRFRAGSIZ; 1755 break; 1756 case VBLK: 1757 vap->va_blocksize = BLKDEV_IOSIZE; 1758 break; 1759 case VCHR: 1760 vap->va_blocksize = MAXBSIZE; 1761 break; 1762 default: 1763 vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize : 1764 fxdr_unsigned(int32_t, fp->fa2_blocksize); 1765 break; 1766 } 1767 if (v3) { 1768 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1769 vap->va_uid = uid; 1770 vap->va_gid = gid; 1771 vap->va_size = fxdr_hyper(&fp->fa3_size); 1772 vap->va_bytes = fxdr_hyper(&fp->fa3_used); 1773 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid); 1774 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); 1775 vap->va_flags = 0; 1776 vap->va_filerev = 0; 1777 } else { 1778 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1779 vap->va_uid = uid; 1780 vap->va_gid = gid; 1781 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); 1782 vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks) 1783 * NFS_FABLKSIZE; 1784 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); 1785 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); 1786 vap->va_flags = 0; 1787 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec); 1788 vap->va_filerev = 0; 1789 } 1790 if (vap->va_size != np->n_size) { 1791 if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) { 1792 vap->va_size = np->n_size; 1793 } else { 1794 np->n_size = vap->va_size; 1795 if (vap->va_type == VREG) { 1796 /* 1797 * we can't free pages if NAC_NOTRUNC because 1798 * the pages can be owned by ourselves. 1799 */ 1800 if (flags & NAC_NOTRUNC) { 1801 np->n_flag |= NTRUNCDELAYED; 1802 } else { 1803 genfs_node_wrlock(vp); 1804 simple_lock(&vp->v_interlock); 1805 (void)VOP_PUTPAGES(vp, 0, 1806 0, PGO_SYNCIO | PGO_CLEANIT | 1807 PGO_FREE | PGO_ALLPAGES); 1808 uvm_vnp_setsize(vp, np->n_size); 1809 genfs_node_unlock(vp); 1810 } 1811 } 1812 } 1813 } 1814 np->n_attrstamp = time_second; 1815 if (vaper != NULL) { 1816 memcpy((void *)vaper, (void *)vap, sizeof(*vap)); 1817 if (np->n_flag & NCHG) { 1818 if (np->n_flag & NACC) 1819 vaper->va_atime = np->n_atim; 1820 if (np->n_flag & NUPD) 1821 vaper->va_mtime = np->n_mtim; 1822 } 1823 } 1824 return (0); 1825 } 1826 1827 /* 1828 * Check the time stamp 1829 * If the cache is valid, copy contents to *vap and return 0 1830 * otherwise return an error 1831 */ 1832 int 1833 nfs_getattrcache(vp, vaper) 1834 struct vnode *vp; 1835 struct vattr *vaper; 1836 { 1837 struct nfsnode *np = VTONFS(vp); 1838 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1839 struct vattr *vap; 1840 1841 if (np->n_attrstamp == 0 || 1842 (time_second - np->n_attrstamp) >= nfs_attrtimeo(nmp, np)) { 1843 nfsstats.attrcache_misses++; 1844 return (ENOENT); 1845 } 1846 nfsstats.attrcache_hits++; 1847 vap = np->n_vattr; 1848 if (vap->va_size != np->n_size) { 1849 if (vap->va_type == VREG) { 1850 if ((np->n_flag & NMODIFIED) != 0 && 1851 vap->va_size < np->n_size) { 1852 vap->va_size = np->n_size; 1853 } else { 1854 np->n_size = vap->va_size; 1855 } 1856 genfs_node_wrlock(vp); 1857 uvm_vnp_setsize(vp, np->n_size); 1858 genfs_node_unlock(vp); 1859 } else 1860 np->n_size = vap->va_size; 1861 } 1862 memcpy((void *)vaper, (void *)vap, sizeof(struct vattr)); 1863 if (np->n_flag & NCHG) { 1864 if (np->n_flag & NACC) 1865 vaper->va_atime = np->n_atim; 1866 if (np->n_flag & NUPD) 1867 vaper->va_mtime = np->n_mtim; 1868 } 1869 return (0); 1870 } 1871 1872 void 1873 nfs_delayedtruncate(vp) 1874 struct vnode *vp; 1875 { 1876 struct nfsnode *np = VTONFS(vp); 1877 1878 if (np->n_flag & NTRUNCDELAYED) { 1879 np->n_flag &= ~NTRUNCDELAYED; 1880 genfs_node_wrlock(vp); 1881 simple_lock(&vp->v_interlock); 1882 (void)VOP_PUTPAGES(vp, 0, 1883 0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES); 1884 uvm_vnp_setsize(vp, np->n_size); 1885 genfs_node_unlock(vp); 1886 } 1887 } 1888 1889 #define NFS_WCCKLUDGE_TIMEOUT (24 * 60 * 60) /* 1 day */ 1890 #define NFS_WCCKLUDGE(nmp, now) \ 1891 (((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \ 1892 ((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0) 1893 1894 /* 1895 * nfs_check_wccdata: check inaccurate wcc_data 1896 * 1897 * => return non-zero if we shouldn't trust the wcc_data. 1898 * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed". 1899 */ 1900 1901 int 1902 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime, 1903 struct timespec *mtime, bool docheck) 1904 { 1905 int error = 0; 1906 1907 #if !defined(NFS_V2_ONLY) 1908 1909 if (docheck) { 1910 struct vnode *vp = NFSTOV(np); 1911 struct nfsmount *nmp; 1912 long now = time_second; 1913 const struct timespec *omtime = &np->n_vattr->va_mtime; 1914 const struct timespec *octime = &np->n_vattr->va_ctime; 1915 #if defined(DEBUG) 1916 const char *reason = NULL; /* XXX: gcc */ 1917 #endif 1918 1919 if (timespeccmp(omtime, mtime, <=)) { 1920 #if defined(DEBUG) 1921 reason = "mtime"; 1922 #endif 1923 error = EINVAL; 1924 } 1925 1926 if (vp->v_type == VDIR && timespeccmp(octime, ctime, <=)) { 1927 #if defined(DEBUG) 1928 reason = "ctime"; 1929 #endif 1930 error = EINVAL; 1931 } 1932 1933 nmp = VFSTONFS(vp->v_mount); 1934 if (error) { 1935 1936 /* 1937 * despite of the fact that we've updated the file, 1938 * timestamps of the file were not updated as we 1939 * expected. 1940 * it means that the server has incompatible 1941 * semantics of timestamps or (more likely) 1942 * the server time is not precise enough to 1943 * track each modifications. 1944 * in that case, we disable wcc processing. 1945 * 1946 * yes, strictly speaking, we should disable all 1947 * caching. it's a compromise. 1948 */ 1949 1950 mutex_enter(&nmp->nm_lock); 1951 #if defined(DEBUG) 1952 if (!NFS_WCCKLUDGE(nmp, now)) { 1953 printf("%s: inaccurate wcc data (%s) detected," 1954 " disabling wcc" 1955 " (ctime %u.%09u %u.%09u," 1956 " mtime %u.%09u %u.%09u)\n", 1957 vp->v_mount->mnt_stat.f_mntfromname, 1958 reason, 1959 (unsigned int)octime->tv_sec, 1960 (unsigned int)octime->tv_nsec, 1961 (unsigned int)ctime->tv_sec, 1962 (unsigned int)ctime->tv_nsec, 1963 (unsigned int)omtime->tv_sec, 1964 (unsigned int)omtime->tv_nsec, 1965 (unsigned int)mtime->tv_sec, 1966 (unsigned int)mtime->tv_nsec); 1967 } 1968 #endif 1969 nmp->nm_iflag |= NFSMNT_WCCKLUDGE; 1970 nmp->nm_wcckludgetime = now; 1971 mutex_exit(&nmp->nm_lock); 1972 } else if (NFS_WCCKLUDGE(nmp, now)) { 1973 error = EPERM; /* XXX */ 1974 } else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1975 mutex_enter(&nmp->nm_lock); 1976 if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1977 #if defined(DEBUG) 1978 printf("%s: re-enabling wcc\n", 1979 vp->v_mount->mnt_stat.f_mntfromname); 1980 #endif 1981 nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE; 1982 } 1983 mutex_exit(&nmp->nm_lock); 1984 } 1985 } 1986 1987 #endif /* !defined(NFS_V2_ONLY) */ 1988 1989 return error; 1990 } 1991 1992 /* 1993 * Heuristic to see if the server XDR encodes directory cookies or not. 1994 * it is not supposed to, but a lot of servers may do this. Also, since 1995 * most/all servers will implement V2 as well, it is expected that they 1996 * may return just 32 bits worth of cookie information, so we need to 1997 * find out in which 32 bits this information is available. We do this 1998 * to avoid trouble with emulated binaries that can't handle 64 bit 1999 * directory offsets. 2000 */ 2001 2002 void 2003 nfs_cookieheuristic(vp, flagp, l, cred) 2004 struct vnode *vp; 2005 int *flagp; 2006 struct lwp *l; 2007 kauth_cred_t cred; 2008 { 2009 struct uio auio; 2010 struct iovec aiov; 2011 char *tbuf, *cp; 2012 struct dirent *dp; 2013 off_t *cookies = NULL, *cop; 2014 int error, eof, nc, len; 2015 2016 MALLOC(tbuf, void *, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK); 2017 2018 aiov.iov_base = tbuf; 2019 aiov.iov_len = NFS_DIRFRAGSIZ; 2020 auio.uio_iov = &aiov; 2021 auio.uio_iovcnt = 1; 2022 auio.uio_rw = UIO_READ; 2023 auio.uio_resid = NFS_DIRFRAGSIZ; 2024 auio.uio_offset = 0; 2025 UIO_SETUP_SYSSPACE(&auio); 2026 2027 error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc); 2028 2029 len = NFS_DIRFRAGSIZ - auio.uio_resid; 2030 if (error || len == 0) { 2031 FREE(tbuf, M_TEMP); 2032 if (cookies) 2033 free(cookies, M_TEMP); 2034 return; 2035 } 2036 2037 /* 2038 * Find the first valid entry and look at its offset cookie. 2039 */ 2040 2041 cp = tbuf; 2042 for (cop = cookies; len > 0; len -= dp->d_reclen) { 2043 dp = (struct dirent *)cp; 2044 if (dp->d_fileno != 0 && len >= dp->d_reclen) { 2045 if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) { 2046 *flagp |= NFSMNT_SWAPCOOKIE; 2047 nfs_invaldircache(vp, 0); 2048 nfs_vinvalbuf(vp, 0, cred, l, 1); 2049 } 2050 break; 2051 } 2052 cop++; 2053 cp += dp->d_reclen; 2054 } 2055 2056 FREE(tbuf, M_TEMP); 2057 free(cookies, M_TEMP); 2058 } 2059 #endif /* NFS */ 2060 2061 #ifdef NFSSERVER 2062 /* 2063 * Set up nameidata for a lookup() call and do it. 2064 * 2065 * If pubflag is set, this call is done for a lookup operation on the 2066 * public filehandle. In that case we allow crossing mountpoints and 2067 * absolute pathnames. However, the caller is expected to check that 2068 * the lookup result is within the public fs, and deny access if 2069 * it is not. 2070 */ 2071 int 2072 nfs_namei(ndp, nsfh, len, slp, nam, mdp, dposp, retdirp, l, kerbflag, pubflag) 2073 struct nameidata *ndp; 2074 nfsrvfh_t *nsfh; 2075 uint32_t len; 2076 struct nfssvc_sock *slp; 2077 struct mbuf *nam; 2078 struct mbuf **mdp; 2079 char **dposp; 2080 struct vnode **retdirp; 2081 struct lwp *l; 2082 int kerbflag, pubflag; 2083 { 2084 int i, rem; 2085 struct mbuf *md; 2086 char *fromcp, *tocp, *cp; 2087 struct iovec aiov; 2088 struct uio auio; 2089 struct vnode *dp; 2090 int error, rdonly, linklen; 2091 struct componentname *cnp = &ndp->ni_cnd; 2092 2093 *retdirp = NULL; 2094 2095 if ((len + 1) > MAXPATHLEN) 2096 return (ENAMETOOLONG); 2097 if (len == 0) 2098 return (EACCES); 2099 cnp->cn_pnbuf = PNBUF_GET(); 2100 2101 /* 2102 * Copy the name from the mbuf list to ndp->ni_pnbuf 2103 * and set the various ndp fields appropriately. 2104 */ 2105 fromcp = *dposp; 2106 tocp = cnp->cn_pnbuf; 2107 md = *mdp; 2108 rem = mtod(md, char *) + md->m_len - fromcp; 2109 for (i = 0; i < len; i++) { 2110 while (rem == 0) { 2111 md = md->m_next; 2112 if (md == NULL) { 2113 error = EBADRPC; 2114 goto out; 2115 } 2116 fromcp = mtod(md, void *); 2117 rem = md->m_len; 2118 } 2119 if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) { 2120 error = EACCES; 2121 goto out; 2122 } 2123 *tocp++ = *fromcp++; 2124 rem--; 2125 } 2126 *tocp = '\0'; 2127 *mdp = md; 2128 *dposp = fromcp; 2129 len = nfsm_rndup(len)-len; 2130 if (len > 0) { 2131 if (rem >= len) 2132 *dposp += len; 2133 else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0) 2134 goto out; 2135 } 2136 2137 /* 2138 * Extract and set starting directory. 2139 */ 2140 error = nfsrv_fhtovp(nsfh, false, &dp, ndp->ni_cnd.cn_cred, slp, 2141 nam, &rdonly, kerbflag, pubflag); 2142 if (error) 2143 goto out; 2144 if (dp->v_type != VDIR) { 2145 vrele(dp); 2146 error = ENOTDIR; 2147 goto out; 2148 } 2149 2150 if (rdonly) 2151 cnp->cn_flags |= RDONLY; 2152 2153 *retdirp = dp; 2154 2155 if (pubflag) { 2156 /* 2157 * Oh joy. For WebNFS, handle those pesky '%' escapes, 2158 * and the 'native path' indicator. 2159 */ 2160 cp = PNBUF_GET(); 2161 fromcp = cnp->cn_pnbuf; 2162 tocp = cp; 2163 if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) { 2164 switch ((unsigned char)*fromcp) { 2165 case WEBNFS_NATIVE_CHAR: 2166 /* 2167 * 'Native' path for us is the same 2168 * as a path according to the NFS spec, 2169 * just skip the escape char. 2170 */ 2171 fromcp++; 2172 break; 2173 /* 2174 * More may be added in the future, range 0x80-0xff 2175 */ 2176 default: 2177 error = EIO; 2178 vrele(dp); 2179 PNBUF_PUT(cp); 2180 goto out; 2181 } 2182 } 2183 /* 2184 * Translate the '%' escapes, URL-style. 2185 */ 2186 while (*fromcp != '\0') { 2187 if (*fromcp == WEBNFS_ESC_CHAR) { 2188 if (fromcp[1] != '\0' && fromcp[2] != '\0') { 2189 fromcp++; 2190 *tocp++ = HEXSTRTOI(fromcp); 2191 fromcp += 2; 2192 continue; 2193 } else { 2194 error = ENOENT; 2195 vrele(dp); 2196 PNBUF_PUT(cp); 2197 goto out; 2198 } 2199 } else 2200 *tocp++ = *fromcp++; 2201 } 2202 *tocp = '\0'; 2203 PNBUF_PUT(cnp->cn_pnbuf); 2204 cnp->cn_pnbuf = cp; 2205 } 2206 2207 ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1; 2208 ndp->ni_segflg = UIO_SYSSPACE; 2209 ndp->ni_rootdir = rootvnode; 2210 ndp->ni_erootdir = NULL; 2211 2212 if (pubflag) { 2213 ndp->ni_loopcnt = 0; 2214 if (cnp->cn_pnbuf[0] == '/') 2215 dp = rootvnode; 2216 } else { 2217 cnp->cn_flags |= NOCROSSMOUNT; 2218 } 2219 2220 cnp->cn_lwp = l; 2221 VREF(dp); 2222 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); 2223 2224 for (;;) { 2225 cnp->cn_nameptr = cnp->cn_pnbuf; 2226 ndp->ni_startdir = dp; 2227 2228 /* 2229 * And call lookup() to do the real work 2230 */ 2231 error = lookup(ndp); 2232 if (error) { 2233 if (ndp->ni_dvp) { 2234 vput(ndp->ni_dvp); 2235 } 2236 PNBUF_PUT(cnp->cn_pnbuf); 2237 return (error); 2238 } 2239 2240 /* 2241 * Check for encountering a symbolic link 2242 */ 2243 if ((cnp->cn_flags & ISSYMLINK) == 0) { 2244 if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp) { 2245 if (ndp->ni_dvp == ndp->ni_vp) { 2246 vrele(ndp->ni_dvp); 2247 } else { 2248 vput(ndp->ni_dvp); 2249 } 2250 } 2251 if (cnp->cn_flags & (SAVENAME | SAVESTART)) 2252 cnp->cn_flags |= HASBUF; 2253 else 2254 PNBUF_PUT(cnp->cn_pnbuf); 2255 return (0); 2256 } else { 2257 if (!pubflag) { 2258 error = EINVAL; 2259 break; 2260 } 2261 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 2262 error = ELOOP; 2263 break; 2264 } 2265 if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) { 2266 error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred); 2267 if (error != 0) 2268 break; 2269 } 2270 if (ndp->ni_pathlen > 1) 2271 cp = PNBUF_GET(); 2272 else 2273 cp = cnp->cn_pnbuf; 2274 aiov.iov_base = cp; 2275 aiov.iov_len = MAXPATHLEN; 2276 auio.uio_iov = &aiov; 2277 auio.uio_iovcnt = 1; 2278 auio.uio_offset = 0; 2279 auio.uio_rw = UIO_READ; 2280 auio.uio_resid = MAXPATHLEN; 2281 UIO_SETUP_SYSSPACE(&auio); 2282 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 2283 if (error) { 2284 badlink: 2285 if (ndp->ni_pathlen > 1) 2286 PNBUF_PUT(cp); 2287 break; 2288 } 2289 linklen = MAXPATHLEN - auio.uio_resid; 2290 if (linklen == 0) { 2291 error = ENOENT; 2292 goto badlink; 2293 } 2294 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 2295 error = ENAMETOOLONG; 2296 goto badlink; 2297 } 2298 if (ndp->ni_pathlen > 1) { 2299 memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen); 2300 PNBUF_PUT(cnp->cn_pnbuf); 2301 cnp->cn_pnbuf = cp; 2302 } else 2303 cnp->cn_pnbuf[linklen] = '\0'; 2304 ndp->ni_pathlen += linklen; 2305 vput(ndp->ni_vp); 2306 dp = ndp->ni_dvp; 2307 2308 /* 2309 * Check if root directory should replace current directory. 2310 */ 2311 if (cnp->cn_pnbuf[0] == '/') { 2312 vput(dp); 2313 dp = ndp->ni_rootdir; 2314 VREF(dp); 2315 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); 2316 } 2317 } 2318 } 2319 vput(ndp->ni_dvp); 2320 vput(ndp->ni_vp); 2321 ndp->ni_vp = NULL; 2322 out: 2323 PNBUF_PUT(cnp->cn_pnbuf); 2324 return (error); 2325 } 2326 #endif /* NFSSERVER */ 2327 2328 /* 2329 * A fiddled version of m_adj() that ensures null fill to a 32-bit 2330 * boundary and only trims off the back end 2331 * 2332 * 1. trim off 'len' bytes as m_adj(mp, -len). 2333 * 2. add zero-padding 'nul' bytes at the end of the mbuf chain. 2334 */ 2335 void 2336 nfs_zeropad(mp, len, nul) 2337 struct mbuf *mp; 2338 int len; 2339 int nul; 2340 { 2341 struct mbuf *m; 2342 int count; 2343 2344 /* 2345 * Trim from tail. Scan the mbuf chain, 2346 * calculating its length and finding the last mbuf. 2347 * If the adjustment only affects this mbuf, then just 2348 * adjust and return. Otherwise, rescan and truncate 2349 * after the remaining size. 2350 */ 2351 count = 0; 2352 m = mp; 2353 for (;;) { 2354 count += m->m_len; 2355 if (m->m_next == NULL) 2356 break; 2357 m = m->m_next; 2358 } 2359 2360 KDASSERT(count >= len); 2361 2362 if (m->m_len >= len) { 2363 m->m_len -= len; 2364 } else { 2365 count -= len; 2366 /* 2367 * Correct length for chain is "count". 2368 * Find the mbuf with last data, adjust its length, 2369 * and toss data from remaining mbufs on chain. 2370 */ 2371 for (m = mp; m; m = m->m_next) { 2372 if (m->m_len >= count) { 2373 m->m_len = count; 2374 break; 2375 } 2376 count -= m->m_len; 2377 } 2378 KASSERT(m && m->m_next); 2379 m_freem(m->m_next); 2380 m->m_next = NULL; 2381 } 2382 2383 KDASSERT(m->m_next == NULL); 2384 2385 /* 2386 * zero-padding. 2387 */ 2388 if (nul > 0) { 2389 char *cp; 2390 int i; 2391 2392 if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) { 2393 struct mbuf *n; 2394 2395 KDASSERT(MLEN >= nul); 2396 n = m_get(M_WAIT, MT_DATA); 2397 MCLAIM(n, &nfs_mowner); 2398 n->m_len = nul; 2399 n->m_next = NULL; 2400 m->m_next = n; 2401 cp = mtod(n, void *); 2402 } else { 2403 cp = mtod(m, char *) + m->m_len; 2404 m->m_len += nul; 2405 } 2406 for (i = 0; i < nul; i++) 2407 *cp++ = '\0'; 2408 } 2409 return; 2410 } 2411 2412 /* 2413 * Make these functions instead of macros, so that the kernel text size 2414 * doesn't get too big... 2415 */ 2416 void 2417 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp) 2418 struct nfsrv_descript *nfsd; 2419 int before_ret; 2420 struct vattr *before_vap; 2421 int after_ret; 2422 struct vattr *after_vap; 2423 struct mbuf **mbp; 2424 char **bposp; 2425 { 2426 struct mbuf *mb = *mbp; 2427 char *bpos = *bposp; 2428 u_int32_t *tl; 2429 2430 if (before_ret) { 2431 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2432 *tl = nfs_false; 2433 } else { 2434 nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED); 2435 *tl++ = nfs_true; 2436 txdr_hyper(before_vap->va_size, tl); 2437 tl += 2; 2438 txdr_nfsv3time(&(before_vap->va_mtime), tl); 2439 tl += 2; 2440 txdr_nfsv3time(&(before_vap->va_ctime), tl); 2441 } 2442 *bposp = bpos; 2443 *mbp = mb; 2444 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp); 2445 } 2446 2447 void 2448 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp) 2449 struct nfsrv_descript *nfsd; 2450 int after_ret; 2451 struct vattr *after_vap; 2452 struct mbuf **mbp; 2453 char **bposp; 2454 { 2455 struct mbuf *mb = *mbp; 2456 char *bpos = *bposp; 2457 u_int32_t *tl; 2458 struct nfs_fattr *fp; 2459 2460 if (after_ret) { 2461 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2462 *tl = nfs_false; 2463 } else { 2464 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR); 2465 *tl++ = nfs_true; 2466 fp = (struct nfs_fattr *)tl; 2467 nfsm_srvfattr(nfsd, after_vap, fp); 2468 } 2469 *mbp = mb; 2470 *bposp = bpos; 2471 } 2472 2473 void 2474 nfsm_srvfattr(nfsd, vap, fp) 2475 struct nfsrv_descript *nfsd; 2476 struct vattr *vap; 2477 struct nfs_fattr *fp; 2478 { 2479 2480 fp->fa_nlink = txdr_unsigned(vap->va_nlink); 2481 fp->fa_uid = txdr_unsigned(vap->va_uid); 2482 fp->fa_gid = txdr_unsigned(vap->va_gid); 2483 if (nfsd->nd_flag & ND_NFSV3) { 2484 fp->fa_type = vtonfsv3_type(vap->va_type); 2485 fp->fa_mode = vtonfsv3_mode(vap->va_mode); 2486 txdr_hyper(vap->va_size, &fp->fa3_size); 2487 txdr_hyper(vap->va_bytes, &fp->fa3_used); 2488 fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev)); 2489 fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev)); 2490 fp->fa3_fsid.nfsuquad[0] = 0; 2491 fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid); 2492 txdr_hyper(vap->va_fileid, &fp->fa3_fileid); 2493 txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime); 2494 txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime); 2495 txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime); 2496 } else { 2497 fp->fa_type = vtonfsv2_type(vap->va_type); 2498 fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 2499 fp->fa2_size = txdr_unsigned(vap->va_size); 2500 fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize); 2501 if (vap->va_type == VFIFO) 2502 fp->fa2_rdev = 0xffffffff; 2503 else 2504 fp->fa2_rdev = txdr_unsigned(vap->va_rdev); 2505 fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE); 2506 fp->fa2_fsid = txdr_unsigned(vap->va_fsid); 2507 fp->fa2_fileid = txdr_unsigned(vap->va_fileid); 2508 txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime); 2509 txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime); 2510 txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime); 2511 } 2512 } 2513 2514 #ifdef NFSSERVER 2515 /* 2516 * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked) 2517 * - look up fsid in mount list (if not found ret error) 2518 * - get vp and export rights by calling VFS_FHTOVP() 2519 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon 2520 * - if not lockflag unlock it with VOP_UNLOCK() 2521 */ 2522 int 2523 nfsrv_fhtovp(nfsrvfh_t *nsfh, int lockflag, struct vnode **vpp, 2524 kauth_cred_t cred, struct nfssvc_sock *slp, struct mbuf *nam, int *rdonlyp, 2525 int kerbflag, int pubflag) 2526 { 2527 struct mount *mp; 2528 kauth_cred_t credanon; 2529 int error, exflags; 2530 struct sockaddr_in *saddr; 2531 fhandle_t *fhp; 2532 2533 fhp = NFSRVFH_FHANDLE(nsfh); 2534 *vpp = (struct vnode *)0; 2535 2536 if (nfs_ispublicfh(nsfh)) { 2537 if (!pubflag || !nfs_pub.np_valid) 2538 return (ESTALE); 2539 fhp = nfs_pub.np_handle; 2540 } 2541 2542 error = netexport_check(&fhp->fh_fsid, nam, &mp, &exflags, &credanon); 2543 if (error) { 2544 return error; 2545 } 2546 2547 error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp); 2548 if (error) 2549 return (error); 2550 2551 if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) { 2552 saddr = mtod(nam, struct sockaddr_in *); 2553 if ((saddr->sin_family == AF_INET) && 2554 ntohs(saddr->sin_port) >= IPPORT_RESERVED) { 2555 vput(*vpp); 2556 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2557 } 2558 #ifdef INET6 2559 if ((saddr->sin_family == AF_INET6) && 2560 ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) { 2561 vput(*vpp); 2562 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2563 } 2564 #endif 2565 } 2566 /* 2567 * Check/setup credentials. 2568 */ 2569 if (exflags & MNT_EXKERB) { 2570 if (!kerbflag) { 2571 vput(*vpp); 2572 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2573 } 2574 } else if (kerbflag) { 2575 vput(*vpp); 2576 return (NFSERR_AUTHERR | AUTH_TOOWEAK); 2577 } else if (kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 2578 NULL) == 0 || (exflags & MNT_EXPORTANON)) { 2579 kauth_cred_clone(credanon, cred); 2580 } 2581 if (exflags & MNT_EXRDONLY) 2582 *rdonlyp = 1; 2583 else 2584 *rdonlyp = 0; 2585 if (!lockflag) 2586 VOP_UNLOCK(*vpp, 0); 2587 return (0); 2588 } 2589 2590 /* 2591 * WebNFS: check if a filehandle is a public filehandle. For v3, this 2592 * means a length of 0, for v2 it means all zeroes. 2593 */ 2594 int 2595 nfs_ispublicfh(const nfsrvfh_t *nsfh) 2596 { 2597 const char *cp = (const void *)(NFSRVFH_DATA(nsfh)); 2598 int i; 2599 2600 if (NFSRVFH_SIZE(nsfh) == 0) { 2601 return true; 2602 } 2603 if (NFSRVFH_SIZE(nsfh) != NFSX_V2FH) { 2604 return false; 2605 } 2606 for (i = 0; i < NFSX_V2FH; i++) 2607 if (*cp++ != 0) 2608 return false; 2609 return true; 2610 } 2611 #endif /* NFSSERVER */ 2612 2613 /* 2614 * This function compares two net addresses by family and returns true 2615 * if they are the same host. 2616 * If there is any doubt, return false. 2617 * The AF_INET family is handled as a special case so that address mbufs 2618 * don't need to be saved to store "struct in_addr", which is only 4 bytes. 2619 */ 2620 int 2621 netaddr_match(family, haddr, nam) 2622 int family; 2623 union nethostaddr *haddr; 2624 struct mbuf *nam; 2625 { 2626 struct sockaddr_in *inetaddr; 2627 2628 switch (family) { 2629 case AF_INET: 2630 inetaddr = mtod(nam, struct sockaddr_in *); 2631 if (inetaddr->sin_family == AF_INET && 2632 inetaddr->sin_addr.s_addr == haddr->had_inetaddr) 2633 return (1); 2634 break; 2635 #ifdef INET6 2636 case AF_INET6: 2637 { 2638 struct sockaddr_in6 *sin6_1, *sin6_2; 2639 2640 sin6_1 = mtod(nam, struct sockaddr_in6 *); 2641 sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *); 2642 if (sin6_1->sin6_family == AF_INET6 && 2643 IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr)) 2644 return 1; 2645 } 2646 #endif 2647 #ifdef ISO 2648 case AF_ISO: 2649 { 2650 struct sockaddr_iso *isoaddr1, *isoaddr2; 2651 2652 isoaddr1 = mtod(nam, struct sockaddr_iso *); 2653 isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *); 2654 if (isoaddr1->siso_family == AF_ISO && 2655 isoaddr1->siso_nlen > 0 && 2656 isoaddr1->siso_nlen == isoaddr2->siso_nlen && 2657 SAME_ISOADDR(isoaddr1, isoaddr2)) 2658 return (1); 2659 break; 2660 } 2661 #endif /* ISO */ 2662 default: 2663 break; 2664 }; 2665 return (0); 2666 } 2667 2668 /* 2669 * The write verifier has changed (probably due to a server reboot), so all 2670 * PG_NEEDCOMMIT pages will have to be written again. Since they are marked 2671 * as dirty or are being written out just now, all this takes is clearing 2672 * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for 2673 * the mount point. 2674 */ 2675 void 2676 nfs_clearcommit(mp) 2677 struct mount *mp; 2678 { 2679 struct vnode *vp; 2680 struct nfsnode *np; 2681 struct vm_page *pg; 2682 struct nfsmount *nmp = VFSTONFS(mp); 2683 2684 rw_enter(&nmp->nm_writeverflock, RW_WRITER); 2685 2686 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2687 KASSERT(vp->v_mount == mp); 2688 if (vp->v_type != VREG) 2689 continue; 2690 np = VTONFS(vp); 2691 np->n_pushlo = np->n_pushhi = np->n_pushedlo = 2692 np->n_pushedhi = 0; 2693 np->n_commitflags &= 2694 ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID); 2695 simple_lock(&vp->v_uobj.vmobjlock); 2696 TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) { 2697 pg->flags &= ~PG_NEEDCOMMIT; 2698 } 2699 simple_unlock(&vp->v_uobj.vmobjlock); 2700 } 2701 mutex_enter(&nmp->nm_lock); 2702 nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF; 2703 mutex_exit(&nmp->nm_lock); 2704 rw_exit(&nmp->nm_writeverflock); 2705 } 2706 2707 void 2708 nfs_merge_commit_ranges(vp) 2709 struct vnode *vp; 2710 { 2711 struct nfsnode *np = VTONFS(vp); 2712 2713 KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID); 2714 2715 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2716 np->n_pushedlo = np->n_pushlo; 2717 np->n_pushedhi = np->n_pushhi; 2718 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2719 } else { 2720 if (np->n_pushlo < np->n_pushedlo) 2721 np->n_pushedlo = np->n_pushlo; 2722 if (np->n_pushhi > np->n_pushedhi) 2723 np->n_pushedhi = np->n_pushhi; 2724 } 2725 2726 np->n_pushlo = np->n_pushhi = 0; 2727 np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID; 2728 2729 #ifdef NFS_DEBUG_COMMIT 2730 printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2731 (unsigned)np->n_pushedhi); 2732 #endif 2733 } 2734 2735 int 2736 nfs_in_committed_range(vp, off, len) 2737 struct vnode *vp; 2738 off_t off, len; 2739 { 2740 struct nfsnode *np = VTONFS(vp); 2741 off_t lo, hi; 2742 2743 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2744 return 0; 2745 lo = off; 2746 hi = lo + len; 2747 2748 return (lo >= np->n_pushedlo && hi <= np->n_pushedhi); 2749 } 2750 2751 int 2752 nfs_in_tobecommitted_range(vp, off, len) 2753 struct vnode *vp; 2754 off_t off, len; 2755 { 2756 struct nfsnode *np = VTONFS(vp); 2757 off_t lo, hi; 2758 2759 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2760 return 0; 2761 lo = off; 2762 hi = lo + len; 2763 2764 return (lo >= np->n_pushlo && hi <= np->n_pushhi); 2765 } 2766 2767 void 2768 nfs_add_committed_range(vp, off, len) 2769 struct vnode *vp; 2770 off_t off, len; 2771 { 2772 struct nfsnode *np = VTONFS(vp); 2773 off_t lo, hi; 2774 2775 lo = off; 2776 hi = lo + len; 2777 2778 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2779 np->n_pushedlo = lo; 2780 np->n_pushedhi = hi; 2781 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2782 } else { 2783 if (hi > np->n_pushedhi) 2784 np->n_pushedhi = hi; 2785 if (lo < np->n_pushedlo) 2786 np->n_pushedlo = lo; 2787 } 2788 #ifdef NFS_DEBUG_COMMIT 2789 printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2790 (unsigned)np->n_pushedhi); 2791 #endif 2792 } 2793 2794 void 2795 nfs_del_committed_range(vp, off, len) 2796 struct vnode *vp; 2797 off_t off, len; 2798 { 2799 struct nfsnode *np = VTONFS(vp); 2800 off_t lo, hi; 2801 2802 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2803 return; 2804 2805 lo = off; 2806 hi = lo + len; 2807 2808 if (lo > np->n_pushedhi || hi < np->n_pushedlo) 2809 return; 2810 if (lo <= np->n_pushedlo) 2811 np->n_pushedlo = hi; 2812 else if (hi >= np->n_pushedhi) 2813 np->n_pushedhi = lo; 2814 else { 2815 /* 2816 * XXX There's only one range. If the deleted range 2817 * is in the middle, pick the largest of the 2818 * contiguous ranges that it leaves. 2819 */ 2820 if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi)) 2821 np->n_pushedhi = lo; 2822 else 2823 np->n_pushedlo = hi; 2824 } 2825 #ifdef NFS_DEBUG_COMMIT 2826 printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2827 (unsigned)np->n_pushedhi); 2828 #endif 2829 } 2830 2831 void 2832 nfs_add_tobecommitted_range(vp, off, len) 2833 struct vnode *vp; 2834 off_t off, len; 2835 { 2836 struct nfsnode *np = VTONFS(vp); 2837 off_t lo, hi; 2838 2839 lo = off; 2840 hi = lo + len; 2841 2842 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) { 2843 np->n_pushlo = lo; 2844 np->n_pushhi = hi; 2845 np->n_commitflags |= NFS_COMMIT_PUSH_VALID; 2846 } else { 2847 if (lo < np->n_pushlo) 2848 np->n_pushlo = lo; 2849 if (hi > np->n_pushhi) 2850 np->n_pushhi = hi; 2851 } 2852 #ifdef NFS_DEBUG_COMMIT 2853 printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2854 (unsigned)np->n_pushhi); 2855 #endif 2856 } 2857 2858 void 2859 nfs_del_tobecommitted_range(vp, off, len) 2860 struct vnode *vp; 2861 off_t off, len; 2862 { 2863 struct nfsnode *np = VTONFS(vp); 2864 off_t lo, hi; 2865 2866 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2867 return; 2868 2869 lo = off; 2870 hi = lo + len; 2871 2872 if (lo > np->n_pushhi || hi < np->n_pushlo) 2873 return; 2874 2875 if (lo <= np->n_pushlo) 2876 np->n_pushlo = hi; 2877 else if (hi >= np->n_pushhi) 2878 np->n_pushhi = lo; 2879 else { 2880 /* 2881 * XXX There's only one range. If the deleted range 2882 * is in the middle, pick the largest of the 2883 * contiguous ranges that it leaves. 2884 */ 2885 if ((np->n_pushlo - lo) > (hi - np->n_pushhi)) 2886 np->n_pushhi = lo; 2887 else 2888 np->n_pushlo = hi; 2889 } 2890 #ifdef NFS_DEBUG_COMMIT 2891 printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2892 (unsigned)np->n_pushhi); 2893 #endif 2894 } 2895 2896 /* 2897 * Map errnos to NFS error numbers. For Version 3 also filter out error 2898 * numbers not specified for the associated procedure. 2899 */ 2900 int 2901 nfsrv_errmap(nd, err) 2902 struct nfsrv_descript *nd; 2903 int err; 2904 { 2905 const short *defaulterrp, *errp; 2906 2907 if (nd->nd_flag & ND_NFSV3) { 2908 if (nd->nd_procnum <= NFSPROC_COMMIT) { 2909 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum]; 2910 while (*++errp) { 2911 if (*errp == err) 2912 return (err); 2913 else if (*errp > err) 2914 break; 2915 } 2916 return ((int)*defaulterrp); 2917 } else 2918 return (err & 0xffff); 2919 } 2920 if (err <= ELAST) 2921 return ((int)nfsrv_v2errmap[err - 1]); 2922 return (NFSERR_IO); 2923 } 2924 2925 u_int32_t 2926 nfs_getxid() 2927 { 2928 static u_int32_t base; 2929 static u_int32_t nfs_xid = 0; 2930 static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER; 2931 u_int32_t newxid; 2932 2933 simple_lock(&nfs_xidlock); 2934 /* 2935 * derive initial xid from system time 2936 * XXX time is invalid if root not yet mounted 2937 */ 2938 if (__predict_false(!base && (rootvp))) { 2939 struct timeval tv; 2940 2941 microtime(&tv); 2942 base = tv.tv_sec << 12; 2943 nfs_xid = base; 2944 } 2945 2946 /* 2947 * Skip zero xid if it should ever happen. 2948 */ 2949 if (__predict_false(++nfs_xid == 0)) 2950 nfs_xid++; 2951 newxid = nfs_xid; 2952 simple_unlock(&nfs_xidlock); 2953 2954 return txdr_unsigned(newxid); 2955 } 2956 2957 /* 2958 * assign a new xid for existing request. 2959 * used for NFSERR_JUKEBOX handling. 2960 */ 2961 void 2962 nfs_renewxid(struct nfsreq *req) 2963 { 2964 u_int32_t xid; 2965 int off; 2966 2967 xid = nfs_getxid(); 2968 if (req->r_nmp->nm_sotype == SOCK_STREAM) 2969 off = sizeof(u_int32_t); /* RPC record mark */ 2970 else 2971 off = 0; 2972 2973 m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid); 2974 req->r_xid = xid; 2975 } 2976 2977 #if defined(NFSSERVER) 2978 int 2979 nfsrv_composefh(struct vnode *vp, nfsrvfh_t *nsfh, bool v3) 2980 { 2981 int error; 2982 size_t fhsize; 2983 2984 fhsize = NFSD_MAXFHSIZE; 2985 error = vfs_composefh(vp, (void *)NFSRVFH_DATA(nsfh), &fhsize); 2986 if (NFSX_FHTOOBIG_P(fhsize, v3)) { 2987 error = EOPNOTSUPP; 2988 } 2989 if (error != 0) { 2990 return error; 2991 } 2992 if (!v3 && fhsize < NFSX_V2FH) { 2993 memset((char *)NFSRVFH_DATA(nsfh) + fhsize, 0, 2994 NFSX_V2FH - fhsize); 2995 fhsize = NFSX_V2FH; 2996 } 2997 if ((fhsize % NFSX_UNSIGNED) != 0) { 2998 return EOPNOTSUPP; 2999 } 3000 nsfh->nsfh_size = fhsize; 3001 return 0; 3002 } 3003 3004 int 3005 nfsrv_comparefh(const nfsrvfh_t *fh1, const nfsrvfh_t *fh2) 3006 { 3007 3008 if (NFSRVFH_SIZE(fh1) != NFSRVFH_SIZE(fh2)) { 3009 return NFSRVFH_SIZE(fh2) - NFSRVFH_SIZE(fh1); 3010 } 3011 return memcmp(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), NFSRVFH_SIZE(fh1)); 3012 } 3013 3014 void 3015 nfsrv_copyfh(nfsrvfh_t *fh1, const nfsrvfh_t *fh2) 3016 { 3017 size_t size; 3018 3019 fh1->nsfh_size = size = NFSRVFH_SIZE(fh2); 3020 memcpy(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), size); 3021 } 3022 #endif /* defined(NFSSERVER) */ 3023 3024 #if defined(NFS) 3025 /* 3026 * Set the attribute timeout based on how recently the file has been modified. 3027 */ 3028 3029 time_t 3030 nfs_attrtimeo(struct nfsmount *nmp, struct nfsnode *np) 3031 { 3032 time_t timeo; 3033 3034 if ((nmp->nm_flag & NFSMNT_NOAC) != 0) 3035 return 0; 3036 3037 if (((np)->n_flag & NMODIFIED) != 0) 3038 return NFS_MINATTRTIMO; 3039 3040 timeo = (time_second - np->n_mtime.tv_sec) / 10; 3041 timeo = max(timeo, NFS_MINATTRTIMO); 3042 timeo = min(timeo, NFS_MAXATTRTIMO); 3043 return timeo; 3044 } 3045 #endif /* defined(NFS) */ 3046
Indexes created Thu Oct 02 14:10:14 GMT 2025