nfs_subs.c revision 1.210
1 /* $NetBSD: nfs_subs.c,v 1.210 2008/11/19 18:36:09 ad Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_subs.c 8.8 (Berkeley) 5/22/95 35 */ 36 37 /* 38 * Copyright 2000 Wasabi Systems, Inc. 39 * All rights reserved. 40 * 41 * Written by Frank van der Linden for Wasabi Systems, Inc. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. All advertising materials mentioning features or use of this software 52 * must display the following acknowledgement: 53 * This product includes software developed for the NetBSD Project by 54 * Wasabi Systems, Inc. 55 * 4. The name of Wasabi Systems, Inc. may not be used to endorse 56 * or promote products derived from this software without specific prior 57 * written permission. 58 * 59 * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND 60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 61 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 62 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL WASABI SYSTEMS, INC 63 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 64 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 65 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 66 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 67 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 68 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 69 * POSSIBILITY OF SUCH DAMAGE. 70 */ 71 72 #include <sys/cdefs.h> 73 __KERNEL_RCSID(0, "$NetBSD: nfs_subs.c,v 1.210 2008/11/19 18:36:09 ad Exp $"); 74 75 #ifdef _KERNEL_OPT 76 #include "fs_nfs.h" 77 #include "opt_nfs.h" 78 #endif 79 80 /* 81 * These functions support the macros and help fiddle mbuf chains for 82 * the nfs op functions. They do things like create the rpc header and 83 * copy data between mbuf chains and uio lists. 84 */ 85 #include <sys/param.h> 86 #include <sys/proc.h> 87 #include <sys/systm.h> 88 #include <sys/kernel.h> 89 #include <sys/kmem.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/filedesc.h> 97 #include <sys/time.h> 98 #include <sys/dirent.h> 99 #include <sys/once.h> 100 #include <sys/kauth.h> 101 #include <sys/atomic.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> 118 119 static u_int32_t nfs_xid; 120 121 /* 122 * Data items converted to xdr at startup, since they are constant 123 * This is kinda hokey, but may save a little time doing byte swaps 124 */ 125 u_int32_t nfs_xdrneg1; 126 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr, 127 rpc_mismatch, rpc_auth_unix, rpc_msgaccepted, 128 rpc_auth_kerb; 129 u_int32_t nfs_prog, nfs_true, nfs_false; 130 131 /* And other global data */ 132 const nfstype nfsv2_type[9] = 133 { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON }; 134 const nfstype nfsv3_type[9] = 135 { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON }; 136 const enum vtype nv2tov_type[8] = 137 { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON }; 138 const enum vtype nv3tov_type[8] = 139 { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO }; 140 int nfs_ticks; 141 int nfs_commitsize; 142 143 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies"); 144 145 /* NFS client/server stats. */ 146 struct nfsstats nfsstats; 147 148 /* 149 * Mapping of old NFS Version 2 RPC numbers to generic numbers. 150 */ 151 const int nfsv3_procid[NFS_NPROCS] = { 152 NFSPROC_NULL, 153 NFSPROC_GETATTR, 154 NFSPROC_SETATTR, 155 NFSPROC_NOOP, 156 NFSPROC_LOOKUP, 157 NFSPROC_READLINK, 158 NFSPROC_READ, 159 NFSPROC_NOOP, 160 NFSPROC_WRITE, 161 NFSPROC_CREATE, 162 NFSPROC_REMOVE, 163 NFSPROC_RENAME, 164 NFSPROC_LINK, 165 NFSPROC_SYMLINK, 166 NFSPROC_MKDIR, 167 NFSPROC_RMDIR, 168 NFSPROC_READDIR, 169 NFSPROC_FSSTAT, 170 NFSPROC_NOOP, 171 NFSPROC_NOOP, 172 NFSPROC_NOOP, 173 NFSPROC_NOOP, 174 NFSPROC_NOOP 175 }; 176 177 /* 178 * and the reverse mapping from generic to Version 2 procedure numbers 179 */ 180 const int nfsv2_procid[NFS_NPROCS] = { 181 NFSV2PROC_NULL, 182 NFSV2PROC_GETATTR, 183 NFSV2PROC_SETATTR, 184 NFSV2PROC_LOOKUP, 185 NFSV2PROC_NOOP, 186 NFSV2PROC_READLINK, 187 NFSV2PROC_READ, 188 NFSV2PROC_WRITE, 189 NFSV2PROC_CREATE, 190 NFSV2PROC_MKDIR, 191 NFSV2PROC_SYMLINK, 192 NFSV2PROC_CREATE, 193 NFSV2PROC_REMOVE, 194 NFSV2PROC_RMDIR, 195 NFSV2PROC_RENAME, 196 NFSV2PROC_LINK, 197 NFSV2PROC_READDIR, 198 NFSV2PROC_NOOP, 199 NFSV2PROC_STATFS, 200 NFSV2PROC_NOOP, 201 NFSV2PROC_NOOP, 202 NFSV2PROC_NOOP, 203 NFSV2PROC_NOOP, 204 }; 205 206 /* 207 * Maps errno values to nfs error numbers. 208 * Use NFSERR_IO as the catch all for ones not specifically defined in 209 * RFC 1094. 210 */ 211 static const u_char nfsrv_v2errmap[ELAST] = { 212 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 213 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 214 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 215 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 216 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 217 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 218 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 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_NAMETOL, NFSERR_IO, NFSERR_IO, 225 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 226 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 227 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 228 NFSERR_IO, NFSERR_IO, 229 }; 230 231 /* 232 * Maps errno values to nfs error numbers. 233 * Although it is not obvious whether or not NFS clients really care if 234 * a returned error value is in the specified list for the procedure, the 235 * safest thing to do is filter them appropriately. For Version 2, the 236 * X/Open XNFS document is the only specification that defines error values 237 * for each RPC (The RFC simply lists all possible error values for all RPCs), 238 * so I have decided to not do this for Version 2. 239 * The first entry is the default error return and the rest are the valid 240 * errors for that RPC in increasing numeric order. 241 */ 242 static const short nfsv3err_null[] = { 243 0, 244 0, 245 }; 246 247 static const short nfsv3err_getattr[] = { 248 NFSERR_IO, 249 NFSERR_IO, 250 NFSERR_STALE, 251 NFSERR_BADHANDLE, 252 NFSERR_SERVERFAULT, 253 0, 254 }; 255 256 static const short nfsv3err_setattr[] = { 257 NFSERR_IO, 258 NFSERR_PERM, 259 NFSERR_IO, 260 NFSERR_ACCES, 261 NFSERR_INVAL, 262 NFSERR_NOSPC, 263 NFSERR_ROFS, 264 NFSERR_DQUOT, 265 NFSERR_STALE, 266 NFSERR_BADHANDLE, 267 NFSERR_NOT_SYNC, 268 NFSERR_SERVERFAULT, 269 0, 270 }; 271 272 static const short nfsv3err_lookup[] = { 273 NFSERR_IO, 274 NFSERR_NOENT, 275 NFSERR_IO, 276 NFSERR_ACCES, 277 NFSERR_NOTDIR, 278 NFSERR_NAMETOL, 279 NFSERR_STALE, 280 NFSERR_BADHANDLE, 281 NFSERR_SERVERFAULT, 282 0, 283 }; 284 285 static const short nfsv3err_access[] = { 286 NFSERR_IO, 287 NFSERR_IO, 288 NFSERR_STALE, 289 NFSERR_BADHANDLE, 290 NFSERR_SERVERFAULT, 291 0, 292 }; 293 294 static const short nfsv3err_readlink[] = { 295 NFSERR_IO, 296 NFSERR_IO, 297 NFSERR_ACCES, 298 NFSERR_INVAL, 299 NFSERR_STALE, 300 NFSERR_BADHANDLE, 301 NFSERR_NOTSUPP, 302 NFSERR_SERVERFAULT, 303 0, 304 }; 305 306 static const short nfsv3err_read[] = { 307 NFSERR_IO, 308 NFSERR_IO, 309 NFSERR_NXIO, 310 NFSERR_ACCES, 311 NFSERR_INVAL, 312 NFSERR_STALE, 313 NFSERR_BADHANDLE, 314 NFSERR_SERVERFAULT, 315 NFSERR_JUKEBOX, 316 0, 317 }; 318 319 static const short nfsv3err_write[] = { 320 NFSERR_IO, 321 NFSERR_IO, 322 NFSERR_ACCES, 323 NFSERR_INVAL, 324 NFSERR_FBIG, 325 NFSERR_NOSPC, 326 NFSERR_ROFS, 327 NFSERR_DQUOT, 328 NFSERR_STALE, 329 NFSERR_BADHANDLE, 330 NFSERR_SERVERFAULT, 331 NFSERR_JUKEBOX, 332 0, 333 }; 334 335 static const short nfsv3err_create[] = { 336 NFSERR_IO, 337 NFSERR_IO, 338 NFSERR_ACCES, 339 NFSERR_EXIST, 340 NFSERR_NOTDIR, 341 NFSERR_NOSPC, 342 NFSERR_ROFS, 343 NFSERR_NAMETOL, 344 NFSERR_DQUOT, 345 NFSERR_STALE, 346 NFSERR_BADHANDLE, 347 NFSERR_NOTSUPP, 348 NFSERR_SERVERFAULT, 349 0, 350 }; 351 352 static const short nfsv3err_mkdir[] = { 353 NFSERR_IO, 354 NFSERR_IO, 355 NFSERR_ACCES, 356 NFSERR_EXIST, 357 NFSERR_NOTDIR, 358 NFSERR_NOSPC, 359 NFSERR_ROFS, 360 NFSERR_NAMETOL, 361 NFSERR_DQUOT, 362 NFSERR_STALE, 363 NFSERR_BADHANDLE, 364 NFSERR_NOTSUPP, 365 NFSERR_SERVERFAULT, 366 0, 367 }; 368 369 static const short nfsv3err_symlink[] = { 370 NFSERR_IO, 371 NFSERR_IO, 372 NFSERR_ACCES, 373 NFSERR_EXIST, 374 NFSERR_NOTDIR, 375 NFSERR_NOSPC, 376 NFSERR_ROFS, 377 NFSERR_NAMETOL, 378 NFSERR_DQUOT, 379 NFSERR_STALE, 380 NFSERR_BADHANDLE, 381 NFSERR_NOTSUPP, 382 NFSERR_SERVERFAULT, 383 0, 384 }; 385 386 static const short nfsv3err_mknod[] = { 387 NFSERR_IO, 388 NFSERR_IO, 389 NFSERR_ACCES, 390 NFSERR_EXIST, 391 NFSERR_NOTDIR, 392 NFSERR_NOSPC, 393 NFSERR_ROFS, 394 NFSERR_NAMETOL, 395 NFSERR_DQUOT, 396 NFSERR_STALE, 397 NFSERR_BADHANDLE, 398 NFSERR_NOTSUPP, 399 NFSERR_SERVERFAULT, 400 NFSERR_BADTYPE, 401 0, 402 }; 403 404 static const short nfsv3err_remove[] = { 405 NFSERR_IO, 406 NFSERR_NOENT, 407 NFSERR_IO, 408 NFSERR_ACCES, 409 NFSERR_NOTDIR, 410 NFSERR_ROFS, 411 NFSERR_NAMETOL, 412 NFSERR_STALE, 413 NFSERR_BADHANDLE, 414 NFSERR_SERVERFAULT, 415 0, 416 }; 417 418 static const short nfsv3err_rmdir[] = { 419 NFSERR_IO, 420 NFSERR_NOENT, 421 NFSERR_IO, 422 NFSERR_ACCES, 423 NFSERR_EXIST, 424 NFSERR_NOTDIR, 425 NFSERR_INVAL, 426 NFSERR_ROFS, 427 NFSERR_NAMETOL, 428 NFSERR_NOTEMPTY, 429 NFSERR_STALE, 430 NFSERR_BADHANDLE, 431 NFSERR_NOTSUPP, 432 NFSERR_SERVERFAULT, 433 0, 434 }; 435 436 static const short nfsv3err_rename[] = { 437 NFSERR_IO, 438 NFSERR_NOENT, 439 NFSERR_IO, 440 NFSERR_ACCES, 441 NFSERR_EXIST, 442 NFSERR_XDEV, 443 NFSERR_NOTDIR, 444 NFSERR_ISDIR, 445 NFSERR_INVAL, 446 NFSERR_NOSPC, 447 NFSERR_ROFS, 448 NFSERR_MLINK, 449 NFSERR_NAMETOL, 450 NFSERR_NOTEMPTY, 451 NFSERR_DQUOT, 452 NFSERR_STALE, 453 NFSERR_BADHANDLE, 454 NFSERR_NOTSUPP, 455 NFSERR_SERVERFAULT, 456 0, 457 }; 458 459 static const short nfsv3err_link[] = { 460 NFSERR_IO, 461 NFSERR_IO, 462 NFSERR_ACCES, 463 NFSERR_EXIST, 464 NFSERR_XDEV, 465 NFSERR_NOTDIR, 466 NFSERR_INVAL, 467 NFSERR_NOSPC, 468 NFSERR_ROFS, 469 NFSERR_MLINK, 470 NFSERR_NAMETOL, 471 NFSERR_DQUOT, 472 NFSERR_STALE, 473 NFSERR_BADHANDLE, 474 NFSERR_NOTSUPP, 475 NFSERR_SERVERFAULT, 476 0, 477 }; 478 479 static const short nfsv3err_readdir[] = { 480 NFSERR_IO, 481 NFSERR_IO, 482 NFSERR_ACCES, 483 NFSERR_NOTDIR, 484 NFSERR_STALE, 485 NFSERR_BADHANDLE, 486 NFSERR_BAD_COOKIE, 487 NFSERR_TOOSMALL, 488 NFSERR_SERVERFAULT, 489 0, 490 }; 491 492 static const short nfsv3err_readdirplus[] = { 493 NFSERR_IO, 494 NFSERR_IO, 495 NFSERR_ACCES, 496 NFSERR_NOTDIR, 497 NFSERR_STALE, 498 NFSERR_BADHANDLE, 499 NFSERR_BAD_COOKIE, 500 NFSERR_NOTSUPP, 501 NFSERR_TOOSMALL, 502 NFSERR_SERVERFAULT, 503 0, 504 }; 505 506 static const short nfsv3err_fsstat[] = { 507 NFSERR_IO, 508 NFSERR_IO, 509 NFSERR_STALE, 510 NFSERR_BADHANDLE, 511 NFSERR_SERVERFAULT, 512 0, 513 }; 514 515 static const short nfsv3err_fsinfo[] = { 516 NFSERR_STALE, 517 NFSERR_STALE, 518 NFSERR_BADHANDLE, 519 NFSERR_SERVERFAULT, 520 0, 521 }; 522 523 static const short nfsv3err_pathconf[] = { 524 NFSERR_STALE, 525 NFSERR_STALE, 526 NFSERR_BADHANDLE, 527 NFSERR_SERVERFAULT, 528 0, 529 }; 530 531 static const short nfsv3err_commit[] = { 532 NFSERR_IO, 533 NFSERR_IO, 534 NFSERR_STALE, 535 NFSERR_BADHANDLE, 536 NFSERR_SERVERFAULT, 537 0, 538 }; 539 540 static const short * const nfsrv_v3errmap[] = { 541 nfsv3err_null, 542 nfsv3err_getattr, 543 nfsv3err_setattr, 544 nfsv3err_lookup, 545 nfsv3err_access, 546 nfsv3err_readlink, 547 nfsv3err_read, 548 nfsv3err_write, 549 nfsv3err_create, 550 nfsv3err_mkdir, 551 nfsv3err_symlink, 552 nfsv3err_mknod, 553 nfsv3err_remove, 554 nfsv3err_rmdir, 555 nfsv3err_rename, 556 nfsv3err_link, 557 nfsv3err_readdir, 558 nfsv3err_readdirplus, 559 nfsv3err_fsstat, 560 nfsv3err_fsinfo, 561 nfsv3err_pathconf, 562 nfsv3err_commit, 563 }; 564 565 extern struct nfsrtt nfsrtt; 566 567 u_long nfsdirhashmask; 568 569 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *)); 570 571 /* 572 * Create the header for an rpc request packet 573 * The hsiz is the size of the rest of the nfs request header. 574 * (just used to decide if a cluster is a good idea) 575 */ 576 struct mbuf * 577 nfsm_reqh(struct nfsnode *np, u_long procid, int hsiz, char **bposp) 578 { 579 struct mbuf *mb; 580 char *bpos; 581 582 mb = m_get(M_WAIT, MT_DATA); 583 MCLAIM(mb, &nfs_mowner); 584 if (hsiz >= MINCLSIZE) 585 m_clget(mb, M_WAIT); 586 mb->m_len = 0; 587 bpos = mtod(mb, void *); 588 589 /* Finally, return values */ 590 *bposp = bpos; 591 return (mb); 592 } 593 594 /* 595 * Build the RPC header and fill in the authorization info. 596 * The authorization string argument is only used when the credentials 597 * come from outside of the kernel. 598 * Returns the head of the mbuf list. 599 */ 600 struct mbuf * 601 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len, 602 verf_str, mrest, mrest_len, mbp, xidp) 603 kauth_cred_t cr; 604 int nmflag; 605 int procid; 606 int auth_type; 607 int auth_len; 608 char *auth_str; 609 int verf_len; 610 char *verf_str; 611 struct mbuf *mrest; 612 int mrest_len; 613 struct mbuf **mbp; 614 u_int32_t *xidp; 615 { 616 struct mbuf *mb; 617 u_int32_t *tl; 618 char *bpos; 619 int i; 620 struct mbuf *mreq; 621 int siz, grpsiz, authsiz; 622 623 authsiz = nfsm_rndup(auth_len); 624 mb = m_gethdr(M_WAIT, MT_DATA); 625 MCLAIM(mb, &nfs_mowner); 626 if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) { 627 m_clget(mb, M_WAIT); 628 } else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) { 629 MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED); 630 } else { 631 MH_ALIGN(mb, 8 * NFSX_UNSIGNED); 632 } 633 mb->m_len = 0; 634 mreq = mb; 635 bpos = mtod(mb, void *); 636 637 /* 638 * First the RPC header. 639 */ 640 nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 641 642 *tl++ = *xidp = nfs_getxid(); 643 *tl++ = rpc_call; 644 *tl++ = rpc_vers; 645 *tl++ = txdr_unsigned(NFS_PROG); 646 if (nmflag & NFSMNT_NFSV3) 647 *tl++ = txdr_unsigned(NFS_VER3); 648 else 649 *tl++ = txdr_unsigned(NFS_VER2); 650 if (nmflag & NFSMNT_NFSV3) 651 *tl++ = txdr_unsigned(procid); 652 else 653 *tl++ = txdr_unsigned(nfsv2_procid[procid]); 654 655 /* 656 * And then the authorization cred. 657 */ 658 *tl++ = txdr_unsigned(auth_type); 659 *tl = txdr_unsigned(authsiz); 660 switch (auth_type) { 661 case RPCAUTH_UNIX: 662 nfsm_build(tl, u_int32_t *, auth_len); 663 *tl++ = 0; /* stamp ?? */ 664 *tl++ = 0; /* NULL hostname */ 665 *tl++ = txdr_unsigned(kauth_cred_geteuid(cr)); 666 *tl++ = txdr_unsigned(kauth_cred_getegid(cr)); 667 grpsiz = (auth_len >> 2) - 5; 668 *tl++ = txdr_unsigned(grpsiz); 669 for (i = 0; i < grpsiz; i++) 670 *tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */ 671 break; 672 case RPCAUTH_KERB4: 673 siz = auth_len; 674 while (siz > 0) { 675 if (M_TRAILINGSPACE(mb) == 0) { 676 struct mbuf *mb2; 677 mb2 = m_get(M_WAIT, MT_DATA); 678 MCLAIM(mb2, &nfs_mowner); 679 if (siz >= MINCLSIZE) 680 m_clget(mb2, M_WAIT); 681 mb->m_next = mb2; 682 mb = mb2; 683 mb->m_len = 0; 684 bpos = mtod(mb, void *); 685 } 686 i = min(siz, M_TRAILINGSPACE(mb)); 687 memcpy(bpos, auth_str, i); 688 mb->m_len += i; 689 auth_str += i; 690 bpos += i; 691 siz -= i; 692 } 693 if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) { 694 for (i = 0; i < siz; i++) 695 *bpos++ = '\0'; 696 mb->m_len += siz; 697 } 698 break; 699 }; 700 701 /* 702 * And the verifier... 703 */ 704 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 705 if (verf_str) { 706 *tl++ = txdr_unsigned(RPCAUTH_KERB4); 707 *tl = txdr_unsigned(verf_len); 708 siz = verf_len; 709 while (siz > 0) { 710 if (M_TRAILINGSPACE(mb) == 0) { 711 struct mbuf *mb2; 712 mb2 = m_get(M_WAIT, MT_DATA); 713 MCLAIM(mb2, &nfs_mowner); 714 if (siz >= MINCLSIZE) 715 m_clget(mb2, M_WAIT); 716 mb->m_next = mb2; 717 mb = mb2; 718 mb->m_len = 0; 719 bpos = mtod(mb, void *); 720 } 721 i = min(siz, M_TRAILINGSPACE(mb)); 722 memcpy(bpos, verf_str, i); 723 mb->m_len += i; 724 verf_str += i; 725 bpos += i; 726 siz -= i; 727 } 728 if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) { 729 for (i = 0; i < siz; i++) 730 *bpos++ = '\0'; 731 mb->m_len += siz; 732 } 733 } else { 734 *tl++ = txdr_unsigned(RPCAUTH_NULL); 735 *tl = 0; 736 } 737 mb->m_next = mrest; 738 mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len; 739 mreq->m_pkthdr.rcvif = (struct ifnet *)0; 740 *mbp = mb; 741 return (mreq); 742 } 743 744 /* 745 * copies mbuf chain to the uio scatter/gather list 746 */ 747 int 748 nfsm_mbuftouio(mrep, uiop, siz, dpos) 749 struct mbuf **mrep; 750 struct uio *uiop; 751 int siz; 752 char **dpos; 753 { 754 char *mbufcp, *uiocp; 755 int xfer, left, len; 756 struct mbuf *mp; 757 long uiosiz, rem; 758 int error = 0; 759 760 mp = *mrep; 761 mbufcp = *dpos; 762 len = mtod(mp, char *) + mp->m_len - mbufcp; 763 rem = nfsm_rndup(siz)-siz; 764 while (siz > 0) { 765 if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) 766 return (EFBIG); 767 left = uiop->uio_iov->iov_len; 768 uiocp = uiop->uio_iov->iov_base; 769 if (left > siz) 770 left = siz; 771 uiosiz = left; 772 while (left > 0) { 773 while (len == 0) { 774 mp = mp->m_next; 775 if (mp == NULL) 776 return (EBADRPC); 777 mbufcp = mtod(mp, void *); 778 len = mp->m_len; 779 } 780 xfer = (left > len) ? len : left; 781 error = copyout_vmspace(uiop->uio_vmspace, mbufcp, 782 uiocp, xfer); 783 if (error) { 784 return error; 785 } 786 left -= xfer; 787 len -= xfer; 788 mbufcp += xfer; 789 uiocp += xfer; 790 uiop->uio_offset += xfer; 791 uiop->uio_resid -= xfer; 792 } 793 if (uiop->uio_iov->iov_len <= siz) { 794 uiop->uio_iovcnt--; 795 uiop->uio_iov++; 796 } else { 797 uiop->uio_iov->iov_base = 798 (char *)uiop->uio_iov->iov_base + uiosiz; 799 uiop->uio_iov->iov_len -= uiosiz; 800 } 801 siz -= uiosiz; 802 } 803 *dpos = mbufcp; 804 *mrep = mp; 805 if (rem > 0) { 806 if (len < rem) 807 error = nfs_adv(mrep, dpos, rem, len); 808 else 809 *dpos += rem; 810 } 811 return (error); 812 } 813 814 /* 815 * copies a uio scatter/gather list to an mbuf chain. 816 * NOTE: can ony handle iovcnt == 1 817 */ 818 int 819 nfsm_uiotombuf(uiop, mq, siz, bpos) 820 struct uio *uiop; 821 struct mbuf **mq; 822 int siz; 823 char **bpos; 824 { 825 char *uiocp; 826 struct mbuf *mp, *mp2; 827 int xfer, left, mlen; 828 int uiosiz, clflg, rem; 829 char *cp; 830 int error; 831 832 #ifdef DIAGNOSTIC 833 if (uiop->uio_iovcnt != 1) 834 panic("nfsm_uiotombuf: iovcnt != 1"); 835 #endif 836 837 if (siz > MLEN) /* or should it >= MCLBYTES ?? */ 838 clflg = 1; 839 else 840 clflg = 0; 841 rem = nfsm_rndup(siz)-siz; 842 mp = mp2 = *mq; 843 while (siz > 0) { 844 left = uiop->uio_iov->iov_len; 845 uiocp = uiop->uio_iov->iov_base; 846 if (left > siz) 847 left = siz; 848 uiosiz = left; 849 while (left > 0) { 850 mlen = M_TRAILINGSPACE(mp); 851 if (mlen == 0) { 852 mp = m_get(M_WAIT, MT_DATA); 853 MCLAIM(mp, &nfs_mowner); 854 if (clflg) 855 m_clget(mp, M_WAIT); 856 mp->m_len = 0; 857 mp2->m_next = mp; 858 mp2 = mp; 859 mlen = M_TRAILINGSPACE(mp); 860 } 861 xfer = (left > mlen) ? mlen : left; 862 cp = mtod(mp, char *) + mp->m_len; 863 error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp, 864 xfer); 865 if (error) { 866 /* XXX */ 867 } 868 mp->m_len += xfer; 869 left -= xfer; 870 uiocp += xfer; 871 uiop->uio_offset += xfer; 872 uiop->uio_resid -= xfer; 873 } 874 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + 875 uiosiz; 876 uiop->uio_iov->iov_len -= uiosiz; 877 siz -= uiosiz; 878 } 879 if (rem > 0) { 880 if (rem > M_TRAILINGSPACE(mp)) { 881 mp = m_get(M_WAIT, MT_DATA); 882 MCLAIM(mp, &nfs_mowner); 883 mp->m_len = 0; 884 mp2->m_next = mp; 885 } 886 cp = mtod(mp, char *) + mp->m_len; 887 for (left = 0; left < rem; left++) 888 *cp++ = '\0'; 889 mp->m_len += rem; 890 *bpos = cp; 891 } else 892 *bpos = mtod(mp, char *) + mp->m_len; 893 *mq = mp; 894 return (0); 895 } 896 897 /* 898 * Get at least "siz" bytes of correctly aligned data. 899 * When called the mbuf pointers are not necessarily correct, 900 * dsosp points to what ought to be in m_data and left contains 901 * what ought to be in m_len. 902 * This is used by the macros nfsm_dissect and nfsm_dissecton for tough 903 * cases. (The macros use the vars. dpos and dpos2) 904 */ 905 int 906 nfsm_disct(mdp, dposp, siz, left, cp2) 907 struct mbuf **mdp; 908 char **dposp; 909 int siz; 910 int left; 911 char **cp2; 912 { 913 struct mbuf *m1, *m2; 914 struct mbuf *havebuf = NULL; 915 char *src = *dposp; 916 char *dst; 917 int len; 918 919 #ifdef DEBUG 920 if (left < 0) 921 panic("nfsm_disct: left < 0"); 922 #endif 923 m1 = *mdp; 924 /* 925 * Skip through the mbuf chain looking for an mbuf with 926 * some data. If the first mbuf found has enough data 927 * and it is correctly aligned return it. 928 */ 929 while (left == 0) { 930 havebuf = m1; 931 *mdp = m1 = m1->m_next; 932 if (m1 == NULL) 933 return (EBADRPC); 934 src = mtod(m1, void *); 935 left = m1->m_len; 936 /* 937 * If we start a new mbuf and it is big enough 938 * and correctly aligned just return it, don't 939 * do any pull up. 940 */ 941 if (left >= siz && nfsm_aligned(src)) { 942 *cp2 = src; 943 *dposp = src + siz; 944 return (0); 945 } 946 } 947 if ((m1->m_flags & M_EXT) != 0) { 948 if (havebuf && M_TRAILINGSPACE(havebuf) >= siz && 949 nfsm_aligned(mtod(havebuf, char *) + havebuf->m_len)) { 950 /* 951 * If the first mbuf with data has external data 952 * and there is a previous mbuf with some trailing 953 * space, use it to move the data into. 954 */ 955 m2 = m1; 956 *mdp = m1 = havebuf; 957 *cp2 = mtod(m1, char *) + m1->m_len; 958 } else if (havebuf) { 959 /* 960 * If the first mbuf has a external data 961 * and there is no previous empty mbuf 962 * allocate a new mbuf and move the external 963 * data to the new mbuf. Also make the first 964 * mbuf look empty. 965 */ 966 m2 = m1; 967 *mdp = m1 = m_get(M_WAIT, MT_DATA); 968 MCLAIM(m1, m2->m_owner); 969 if ((m2->m_flags & M_PKTHDR) != 0) { 970 /* XXX MOVE */ 971 M_COPY_PKTHDR(m1, m2); 972 m_tag_delete_chain(m2, NULL); 973 m2->m_flags &= ~M_PKTHDR; 974 } 975 if (havebuf) { 976 havebuf->m_next = m1; 977 } 978 m1->m_next = m2; 979 MRESETDATA(m1); 980 m1->m_len = 0; 981 m2->m_data = src; 982 m2->m_len = left; 983 *cp2 = mtod(m1, char *); 984 } else { 985 struct mbuf **nextp = &m1->m_next; 986 987 m1->m_len -= left; 988 do { 989 m2 = m_get(M_WAIT, MT_DATA); 990 MCLAIM(m2, m1->m_owner); 991 if (left >= MINCLSIZE) { 992 MCLGET(m2, M_WAIT); 993 } 994 m2->m_next = *nextp; 995 *nextp = m2; 996 nextp = &m2->m_next; 997 len = (m2->m_flags & M_EXT) != 0 ? 998 MCLBYTES : MLEN; 999 if (len > left) { 1000 len = left; 1001 } 1002 memcpy(mtod(m2, char *), src, len); 1003 m2->m_len = len; 1004 src += len; 1005 left -= len; 1006 } while (left > 0); 1007 *mdp = m1 = m1->m_next; 1008 m2 = m1->m_next; 1009 *cp2 = mtod(m1, char *); 1010 } 1011 } else { 1012 /* 1013 * If the first mbuf has no external data 1014 * move the data to the front of the mbuf. 1015 */ 1016 MRESETDATA(m1); 1017 dst = mtod(m1, char *); 1018 if (dst != src) { 1019 memmove(dst, src, left); 1020 } 1021 m1->m_len = left; 1022 m2 = m1->m_next; 1023 *cp2 = m1->m_data; 1024 } 1025 *dposp = *cp2 + siz; 1026 /* 1027 * Loop through mbufs pulling data up into first mbuf until 1028 * the first mbuf is full or there is no more data to 1029 * pullup. 1030 */ 1031 dst = mtod(m1, char *) + m1->m_len; 1032 while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) { 1033 if ((len = min(len, m2->m_len)) != 0) { 1034 memcpy(dst, mtod(m2, char *), len); 1035 } 1036 m1->m_len += len; 1037 dst += len; 1038 m2->m_data += len; 1039 m2->m_len -= len; 1040 m2 = m2->m_next; 1041 } 1042 if (m1->m_len < siz) 1043 return (EBADRPC); 1044 return (0); 1045 } 1046 1047 /* 1048 * Advance the position in the mbuf chain. 1049 */ 1050 int 1051 nfs_adv(mdp, dposp, offs, left) 1052 struct mbuf **mdp; 1053 char **dposp; 1054 int offs; 1055 int left; 1056 { 1057 struct mbuf *m; 1058 int s; 1059 1060 m = *mdp; 1061 s = left; 1062 while (s < offs) { 1063 offs -= s; 1064 m = m->m_next; 1065 if (m == NULL) 1066 return (EBADRPC); 1067 s = m->m_len; 1068 } 1069 *mdp = m; 1070 *dposp = mtod(m, char *) + offs; 1071 return (0); 1072 } 1073 1074 /* 1075 * Copy a string into mbufs for the hard cases... 1076 */ 1077 int 1078 nfsm_strtmbuf(mb, bpos, cp, siz) 1079 struct mbuf **mb; 1080 char **bpos; 1081 const char *cp; 1082 long siz; 1083 { 1084 struct mbuf *m1 = NULL, *m2; 1085 long left, xfer, len, tlen; 1086 u_int32_t *tl; 1087 int putsize; 1088 1089 putsize = 1; 1090 m2 = *mb; 1091 left = M_TRAILINGSPACE(m2); 1092 if (left > 0) { 1093 tl = ((u_int32_t *)(*bpos)); 1094 *tl++ = txdr_unsigned(siz); 1095 putsize = 0; 1096 left -= NFSX_UNSIGNED; 1097 m2->m_len += NFSX_UNSIGNED; 1098 if (left > 0) { 1099 memcpy((void *) tl, cp, left); 1100 siz -= left; 1101 cp += left; 1102 m2->m_len += left; 1103 left = 0; 1104 } 1105 } 1106 /* Loop around adding mbufs */ 1107 while (siz > 0) { 1108 m1 = m_get(M_WAIT, MT_DATA); 1109 MCLAIM(m1, &nfs_mowner); 1110 if (siz > MLEN) 1111 m_clget(m1, M_WAIT); 1112 m1->m_len = NFSMSIZ(m1); 1113 m2->m_next = m1; 1114 m2 = m1; 1115 tl = mtod(m1, u_int32_t *); 1116 tlen = 0; 1117 if (putsize) { 1118 *tl++ = txdr_unsigned(siz); 1119 m1->m_len -= NFSX_UNSIGNED; 1120 tlen = NFSX_UNSIGNED; 1121 putsize = 0; 1122 } 1123 if (siz < m1->m_len) { 1124 len = nfsm_rndup(siz); 1125 xfer = siz; 1126 if (xfer < len) 1127 *(tl+(xfer>>2)) = 0; 1128 } else { 1129 xfer = len = m1->m_len; 1130 } 1131 memcpy((void *) tl, cp, xfer); 1132 m1->m_len = len+tlen; 1133 siz -= xfer; 1134 cp += xfer; 1135 } 1136 *mb = m1; 1137 *bpos = mtod(m1, char *) + m1->m_len; 1138 return (0); 1139 } 1140 1141 /* 1142 * Directory caching routines. They work as follows: 1143 * - a cache is maintained per VDIR nfsnode. 1144 * - for each offset cookie that is exported to userspace, and can 1145 * thus be thrown back at us as an offset to VOP_READDIR, store 1146 * information in the cache. 1147 * - cached are: 1148 * - cookie itself 1149 * - blocknumber (essentially just a search key in the buffer cache) 1150 * - entry number in block. 1151 * - offset cookie of block in which this entry is stored 1152 * - 32 bit cookie if NFSMNT_XLATECOOKIE is used. 1153 * - entries are looked up in a hash table 1154 * - also maintained is an LRU list of entries, used to determine 1155 * which ones to delete if the cache grows too large. 1156 * - if 32 <-> 64 translation mode is requested for a filesystem, 1157 * the cache also functions as a translation table 1158 * - in the translation case, invalidating the cache does not mean 1159 * flushing it, but just marking entries as invalid, except for 1160 * the <64bit cookie, 32bitcookie> pair which is still valid, to 1161 * still be able to use the cache as a translation table. 1162 * - 32 bit cookies are uniquely created by combining the hash table 1163 * entry value, and one generation count per hash table entry, 1164 * incremented each time an entry is appended to the chain. 1165 * - the cache is invalidated each time a direcory is modified 1166 * - sanity checks are also done; if an entry in a block turns 1167 * out not to have a matching cookie, the cache is invalidated 1168 * and a new block starting from the wanted offset is fetched from 1169 * the server. 1170 * - directory entries as read from the server are extended to contain 1171 * the 64bit and, optionally, the 32bit cookies, for sanity checking 1172 * the cache and exporting them to userspace through the cookie 1173 * argument to VOP_READDIR. 1174 */ 1175 1176 u_long 1177 nfs_dirhash(off) 1178 off_t off; 1179 { 1180 int i; 1181 char *cp = (char *)&off; 1182 u_long sum = 0L; 1183 1184 for (i = 0 ; i < sizeof (off); i++) 1185 sum += *cp++; 1186 1187 return sum; 1188 } 1189 1190 #define _NFSDC_MTX(np) (&NFSTOV(np)->v_interlock) 1191 #define NFSDC_LOCK(np) mutex_enter(_NFSDC_MTX(np)) 1192 #define NFSDC_UNLOCK(np) mutex_exit(_NFSDC_MTX(np)) 1193 #define NFSDC_ASSERT_LOCKED(np) KASSERT(mutex_owned(_NFSDC_MTX(np))) 1194 1195 void 1196 nfs_initdircache(vp) 1197 struct vnode *vp; 1198 { 1199 struct nfsnode *np = VTONFS(vp); 1200 struct nfsdirhashhead *dircache; 1201 1202 dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, true, 1203 &nfsdirhashmask); 1204 1205 NFSDC_LOCK(np); 1206 if (np->n_dircache == NULL) { 1207 np->n_dircachesize = 0; 1208 np->n_dircache = dircache; 1209 dircache = NULL; 1210 TAILQ_INIT(&np->n_dirchain); 1211 } 1212 NFSDC_UNLOCK(np); 1213 if (dircache) 1214 hashdone(dircache, HASH_LIST, nfsdirhashmask); 1215 } 1216 1217 void 1218 nfs_initdirxlatecookie(vp) 1219 struct vnode *vp; 1220 { 1221 struct nfsnode *np = VTONFS(vp); 1222 unsigned *dirgens; 1223 1224 KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE); 1225 1226 dirgens = kmem_zalloc(NFS_DIRHASHSIZ * sizeof(unsigned), KM_SLEEP); 1227 NFSDC_LOCK(np); 1228 if (np->n_dirgens == NULL) { 1229 np->n_dirgens = dirgens; 1230 dirgens = NULL; 1231 } 1232 NFSDC_UNLOCK(np); 1233 if (dirgens) 1234 kmem_free(dirgens, NFS_DIRHASHSIZ * sizeof(unsigned)); 1235 } 1236 1237 static const struct nfsdircache dzero; 1238 1239 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *)); 1240 static void nfs_putdircache_unlocked __P((struct nfsnode *, 1241 struct nfsdircache *)); 1242 1243 static void 1244 nfs_unlinkdircache(np, ndp) 1245 struct nfsnode *np; 1246 struct nfsdircache *ndp; 1247 { 1248 1249 NFSDC_ASSERT_LOCKED(np); 1250 KASSERT(ndp != &dzero); 1251 1252 if (LIST_NEXT(ndp, dc_hash) == (void *)-1) 1253 return; 1254 1255 TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain); 1256 LIST_REMOVE(ndp, dc_hash); 1257 LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */ 1258 1259 nfs_putdircache_unlocked(np, ndp); 1260 } 1261 1262 void 1263 nfs_putdircache(np, ndp) 1264 struct nfsnode *np; 1265 struct nfsdircache *ndp; 1266 { 1267 int ref; 1268 1269 if (ndp == &dzero) 1270 return; 1271 1272 KASSERT(ndp->dc_refcnt > 0); 1273 NFSDC_LOCK(np); 1274 ref = --ndp->dc_refcnt; 1275 NFSDC_UNLOCK(np); 1276 1277 if (ref == 0) 1278 kmem_free(ndp, sizeof(*ndp)); 1279 } 1280 1281 static void 1282 nfs_putdircache_unlocked(struct nfsnode *np, struct nfsdircache *ndp) 1283 { 1284 int ref; 1285 1286 NFSDC_ASSERT_LOCKED(np); 1287 1288 if (ndp == &dzero) 1289 return; 1290 1291 KASSERT(ndp->dc_refcnt > 0); 1292 ref = --ndp->dc_refcnt; 1293 if (ref == 0) 1294 kmem_free(ndp, sizeof(*ndp)); 1295 } 1296 1297 struct nfsdircache * 1298 nfs_searchdircache(vp, off, do32, hashent) 1299 struct vnode *vp; 1300 off_t off; 1301 int do32; 1302 int *hashent; 1303 { 1304 struct nfsdirhashhead *ndhp; 1305 struct nfsdircache *ndp = NULL; 1306 struct nfsnode *np = VTONFS(vp); 1307 unsigned ent; 1308 1309 /* 1310 * Zero is always a valid cookie. 1311 */ 1312 if (off == 0) 1313 /* XXXUNCONST */ 1314 return (struct nfsdircache *)__UNCONST(&dzero); 1315 1316 if (!np->n_dircache) 1317 return NULL; 1318 1319 /* 1320 * We use a 32bit cookie as search key, directly reconstruct 1321 * the hashentry. Else use the hashfunction. 1322 */ 1323 if (do32) { 1324 ent = (u_int32_t)off >> 24; 1325 if (ent >= NFS_DIRHASHSIZ) 1326 return NULL; 1327 ndhp = &np->n_dircache[ent]; 1328 } else { 1329 ndhp = NFSDIRHASH(np, off); 1330 } 1331 1332 if (hashent) 1333 *hashent = (int)(ndhp - np->n_dircache); 1334 1335 NFSDC_LOCK(np); 1336 if (do32) { 1337 LIST_FOREACH(ndp, ndhp, dc_hash) { 1338 if (ndp->dc_cookie32 == (u_int32_t)off) { 1339 /* 1340 * An invalidated entry will become the 1341 * start of a new block fetched from 1342 * the server. 1343 */ 1344 if (ndp->dc_flags & NFSDC_INVALID) { 1345 ndp->dc_blkcookie = ndp->dc_cookie; 1346 ndp->dc_entry = 0; 1347 ndp->dc_flags &= ~NFSDC_INVALID; 1348 } 1349 break; 1350 } 1351 } 1352 } else { 1353 LIST_FOREACH(ndp, ndhp, dc_hash) { 1354 if (ndp->dc_cookie == off) 1355 break; 1356 } 1357 } 1358 if (ndp != NULL) 1359 ndp->dc_refcnt++; 1360 NFSDC_UNLOCK(np); 1361 return ndp; 1362 } 1363 1364 1365 struct nfsdircache * 1366 nfs_enterdircache(struct vnode *vp, off_t off, off_t blkoff, int en, 1367 daddr_t blkno) 1368 { 1369 struct nfsnode *np = VTONFS(vp); 1370 struct nfsdirhashhead *ndhp; 1371 struct nfsdircache *ndp = NULL; 1372 struct nfsdircache *newndp = NULL; 1373 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1374 int hashent = 0, gen, overwrite; /* XXX: GCC */ 1375 1376 /* 1377 * XXX refuse entries for offset 0. amd(8) erroneously sets 1378 * cookie 0 for the '.' entry, making this necessary. This 1379 * isn't so bad, as 0 is a special case anyway. 1380 */ 1381 if (off == 0) 1382 /* XXXUNCONST */ 1383 return (struct nfsdircache *)__UNCONST(&dzero); 1384 1385 if (!np->n_dircache) 1386 /* 1387 * XXX would like to do this in nfs_nget but vtype 1388 * isn't known at that time. 1389 */ 1390 nfs_initdircache(vp); 1391 1392 if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens) 1393 nfs_initdirxlatecookie(vp); 1394 1395 retry: 1396 ndp = nfs_searchdircache(vp, off, 0, &hashent); 1397 1398 NFSDC_LOCK(np); 1399 if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) { 1400 /* 1401 * Overwriting an old entry. Check if it's the same. 1402 * If so, just return. If not, remove the old entry. 1403 */ 1404 if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en) 1405 goto done; 1406 nfs_unlinkdircache(np, ndp); 1407 nfs_putdircache_unlocked(np, ndp); 1408 ndp = NULL; 1409 } 1410 1411 ndhp = &np->n_dircache[hashent]; 1412 1413 if (!ndp) { 1414 if (newndp == NULL) { 1415 NFSDC_UNLOCK(np); 1416 newndp = kmem_alloc(sizeof(*newndp), KM_SLEEP); 1417 newndp->dc_refcnt = 1; 1418 LIST_NEXT(newndp, dc_hash) = (void *)-1; 1419 goto retry; 1420 } 1421 ndp = newndp; 1422 newndp = NULL; 1423 overwrite = 0; 1424 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 1425 /* 1426 * We're allocating a new entry, so bump the 1427 * generation number. 1428 */ 1429 KASSERT(np->n_dirgens); 1430 gen = ++np->n_dirgens[hashent]; 1431 if (gen == 0) { 1432 np->n_dirgens[hashent]++; 1433 gen++; 1434 } 1435 ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff); 1436 } 1437 } else 1438 overwrite = 1; 1439 1440 ndp->dc_cookie = off; 1441 ndp->dc_blkcookie = blkoff; 1442 ndp->dc_entry = en; 1443 ndp->dc_flags = 0; 1444 1445 if (overwrite) 1446 goto done; 1447 1448 /* 1449 * If the maximum directory cookie cache size has been reached 1450 * for this node, take one off the front. The idea is that 1451 * directories are typically read front-to-back once, so that 1452 * the oldest entries can be thrown away without much performance 1453 * loss. 1454 */ 1455 if (np->n_dircachesize == NFS_MAXDIRCACHE) { 1456 nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain)); 1457 } else 1458 np->n_dircachesize++; 1459 1460 KASSERT(ndp->dc_refcnt == 1); 1461 LIST_INSERT_HEAD(ndhp, ndp, dc_hash); 1462 TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain); 1463 ndp->dc_refcnt++; 1464 done: 1465 KASSERT(ndp->dc_refcnt > 0); 1466 NFSDC_UNLOCK(np); 1467 if (newndp) 1468 nfs_putdircache(np, newndp); 1469 return ndp; 1470 } 1471 1472 void 1473 nfs_invaldircache(vp, flags) 1474 struct vnode *vp; 1475 int flags; 1476 { 1477 struct nfsnode *np = VTONFS(vp); 1478 struct nfsdircache *ndp = NULL; 1479 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1480 const bool forcefree = flags & NFS_INVALDIRCACHE_FORCE; 1481 1482 #ifdef DIAGNOSTIC 1483 if (vp->v_type != VDIR) 1484 panic("nfs: invaldircache: not dir"); 1485 #endif 1486 1487 if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0) 1488 np->n_flag &= ~NEOFVALID; 1489 1490 if (!np->n_dircache) 1491 return; 1492 1493 NFSDC_LOCK(np); 1494 if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) { 1495 while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) { 1496 KASSERT(!forcefree || ndp->dc_refcnt == 1); 1497 nfs_unlinkdircache(np, ndp); 1498 } 1499 np->n_dircachesize = 0; 1500 if (forcefree && np->n_dirgens) { 1501 kmem_free(np->n_dirgens, 1502 NFS_DIRHASHSIZ * sizeof(unsigned)); 1503 np->n_dirgens = NULL; 1504 } 1505 } else { 1506 TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain) 1507 ndp->dc_flags |= NFSDC_INVALID; 1508 } 1509 1510 NFSDC_UNLOCK(np); 1511 } 1512 1513 /* 1514 * Called once before VFS init to initialize shared and 1515 * server-specific data structures. 1516 */ 1517 static int 1518 nfs_init0(void) 1519 { 1520 1521 nfsrtt.pos = 0; 1522 rpc_vers = txdr_unsigned(RPC_VER2); 1523 rpc_call = txdr_unsigned(RPC_CALL); 1524 rpc_reply = txdr_unsigned(RPC_REPLY); 1525 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED); 1526 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED); 1527 rpc_mismatch = txdr_unsigned(RPC_MISMATCH); 1528 rpc_autherr = txdr_unsigned(RPC_AUTHERR); 1529 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX); 1530 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4); 1531 nfs_prog = txdr_unsigned(NFS_PROG); 1532 nfs_true = txdr_unsigned(true); 1533 nfs_false = txdr_unsigned(false); 1534 nfs_xdrneg1 = txdr_unsigned(-1); 1535 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; 1536 if (nfs_ticks < 1) 1537 nfs_ticks = 1; 1538 nfs_xid = arc4random(); 1539 nfsdreq_init(); 1540 1541 /* 1542 * Initialize reply list and start timer 1543 */ 1544 TAILQ_INIT(&nfs_reqq); 1545 nfs_timer_init(); 1546 MOWNER_ATTACH(&nfs_mowner); 1547 1548 #ifdef NFS 1549 /* Initialize the kqueue structures */ 1550 nfs_kqinit(); 1551 /* Initialize the iod structures */ 1552 nfs_iodinit(); 1553 #endif 1554 1555 return 0; 1556 } 1557 1558 /* 1559 * This is disgusting, but it must support both modular and monolothic 1560 * configurations. For monolithic builds NFSSERVER may not imply NFS. 1561 * 1562 * Yuck. 1563 */ 1564 void 1565 nfs_init(void) 1566 { 1567 static ONCE_DECL(nfs_init_once); 1568 1569 RUN_ONCE(&nfs_init_once, nfs_init0); 1570 } 1571 1572 void 1573 nfs_fini(void) 1574 { 1575 1576 #ifdef NFS 1577 nfs_kqfini(); 1578 nfs_iodfini(); 1579 #endif 1580 nfsdreq_fini(); 1581 nfs_timer_fini(); 1582 MOWNER_DETACH(&nfs_mowner); 1583 } 1584 1585 #ifdef NFS 1586 /* 1587 * Called once at VFS init to initialize client-specific data structures. 1588 */ 1589 void 1590 nfs_vfs_init() 1591 { 1592 1593 /* Initialize NFS server / client shared data. */ 1594 nfs_init(); 1595 nfs_node_init(); 1596 1597 nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4); 1598 } 1599 1600 void 1601 nfs_vfs_done() 1602 { 1603 1604 nfs_node_done(); 1605 } 1606 1607 /* 1608 * Attribute cache routines. 1609 * nfs_loadattrcache() - loads or updates the cache contents from attributes 1610 * that are on the mbuf list 1611 * nfs_getattrcache() - returns valid attributes if found in cache, returns 1612 * error otherwise 1613 */ 1614 1615 /* 1616 * Load the attribute cache (that lives in the nfsnode entry) with 1617 * the values on the mbuf list and 1618 * Iff vap not NULL 1619 * copy the attributes to *vaper 1620 */ 1621 int 1622 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags) 1623 struct vnode **vpp; 1624 struct mbuf **mdp; 1625 char **dposp; 1626 struct vattr *vaper; 1627 int flags; 1628 { 1629 int32_t t1; 1630 char *cp2; 1631 int error = 0; 1632 struct mbuf *md; 1633 int v3 = NFS_ISV3(*vpp); 1634 1635 md = *mdp; 1636 t1 = (mtod(md, char *) + md->m_len) - *dposp; 1637 error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2); 1638 if (error) 1639 return (error); 1640 return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags); 1641 } 1642 1643 int 1644 nfs_loadattrcache(vpp, fp, vaper, flags) 1645 struct vnode **vpp; 1646 struct nfs_fattr *fp; 1647 struct vattr *vaper; 1648 int flags; 1649 { 1650 struct vnode *vp = *vpp; 1651 struct vattr *vap; 1652 int v3 = NFS_ISV3(vp); 1653 enum vtype vtyp; 1654 u_short vmode; 1655 struct timespec mtime; 1656 struct timespec ctime; 1657 int32_t rdev; 1658 struct nfsnode *np; 1659 extern int (**spec_nfsv2nodeop_p) __P((void *)); 1660 uid_t uid; 1661 gid_t gid; 1662 1663 if (v3) { 1664 vtyp = nfsv3tov_type(fp->fa_type); 1665 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1666 rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1), 1667 fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2)); 1668 fxdr_nfsv3time(&fp->fa3_mtime, &mtime); 1669 fxdr_nfsv3time(&fp->fa3_ctime, &ctime); 1670 } else { 1671 vtyp = nfsv2tov_type(fp->fa_type); 1672 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1673 if (vtyp == VNON || vtyp == VREG) 1674 vtyp = IFTOVT(vmode); 1675 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); 1676 fxdr_nfsv2time(&fp->fa2_mtime, &mtime); 1677 ctime.tv_sec = fxdr_unsigned(u_int32_t, 1678 fp->fa2_ctime.nfsv2_sec); 1679 ctime.tv_nsec = 0; 1680 1681 /* 1682 * Really ugly NFSv2 kludge. 1683 */ 1684 if (vtyp == VCHR && rdev == 0xffffffff) 1685 vtyp = VFIFO; 1686 } 1687 1688 vmode &= ALLPERMS; 1689 1690 /* 1691 * If v_type == VNON it is a new node, so fill in the v_type, 1692 * n_mtime fields. Check to see if it represents a special 1693 * device, and if so, check for a possible alias. Once the 1694 * correct vnode has been obtained, fill in the rest of the 1695 * information. 1696 */ 1697 np = VTONFS(vp); 1698 if (vp->v_type == VNON) { 1699 vp->v_type = vtyp; 1700 if (vp->v_type == VFIFO) { 1701 extern int (**fifo_nfsv2nodeop_p) __P((void *)); 1702 vp->v_op = fifo_nfsv2nodeop_p; 1703 } else if (vp->v_type == VREG) { 1704 mutex_init(&np->n_commitlock, MUTEX_DEFAULT, IPL_NONE); 1705 } else if (vp->v_type == VCHR || vp->v_type == VBLK) { 1706 vp->v_op = spec_nfsv2nodeop_p; 1707 spec_node_init(vp, (dev_t)rdev); 1708 } 1709 np->n_mtime = mtime; 1710 } 1711 uid = fxdr_unsigned(uid_t, fp->fa_uid); 1712 gid = fxdr_unsigned(gid_t, fp->fa_gid); 1713 vap = np->n_vattr; 1714 1715 /* 1716 * Invalidate access cache if uid, gid, mode or ctime changed. 1717 */ 1718 if (np->n_accstamp != -1 && 1719 (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode 1720 || timespeccmp(&ctime, &vap->va_ctime, !=))) 1721 np->n_accstamp = -1; 1722 1723 vap->va_type = vtyp; 1724 vap->va_mode = vmode; 1725 vap->va_rdev = (dev_t)rdev; 1726 vap->va_mtime = mtime; 1727 vap->va_ctime = ctime; 1728 vap->va_birthtime.tv_sec = VNOVAL; 1729 vap->va_birthtime.tv_nsec = VNOVAL; 1730 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 1731 switch (vtyp) { 1732 case VDIR: 1733 vap->va_blocksize = NFS_DIRFRAGSIZ; 1734 break; 1735 case VBLK: 1736 vap->va_blocksize = BLKDEV_IOSIZE; 1737 break; 1738 case VCHR: 1739 vap->va_blocksize = MAXBSIZE; 1740 break; 1741 default: 1742 vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize : 1743 fxdr_unsigned(int32_t, fp->fa2_blocksize); 1744 break; 1745 } 1746 if (v3) { 1747 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1748 vap->va_uid = uid; 1749 vap->va_gid = gid; 1750 vap->va_size = fxdr_hyper(&fp->fa3_size); 1751 vap->va_bytes = fxdr_hyper(&fp->fa3_used); 1752 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid); 1753 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); 1754 vap->va_flags = 0; 1755 vap->va_filerev = 0; 1756 } else { 1757 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1758 vap->va_uid = uid; 1759 vap->va_gid = gid; 1760 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); 1761 vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks) 1762 * NFS_FABLKSIZE; 1763 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); 1764 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); 1765 vap->va_flags = 0; 1766 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec); 1767 vap->va_filerev = 0; 1768 } 1769 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 1770 return EFBIG; 1771 } 1772 if (vap->va_size != np->n_size) { 1773 if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) { 1774 vap->va_size = np->n_size; 1775 } else { 1776 np->n_size = vap->va_size; 1777 if (vap->va_type == VREG) { 1778 /* 1779 * we can't free pages if NAC_NOTRUNC because 1780 * the pages can be owned by ourselves. 1781 */ 1782 if (flags & NAC_NOTRUNC) { 1783 np->n_flag |= NTRUNCDELAYED; 1784 } else { 1785 genfs_node_wrlock(vp); 1786 mutex_enter(&vp->v_interlock); 1787 (void)VOP_PUTPAGES(vp, 0, 1788 0, PGO_SYNCIO | PGO_CLEANIT | 1789 PGO_FREE | PGO_ALLPAGES); 1790 uvm_vnp_setsize(vp, np->n_size); 1791 genfs_node_unlock(vp); 1792 } 1793 } 1794 } 1795 } 1796 np->n_attrstamp = time_second; 1797 if (vaper != NULL) { 1798 memcpy((void *)vaper, (void *)vap, sizeof(*vap)); 1799 if (np->n_flag & NCHG) { 1800 if (np->n_flag & NACC) 1801 vaper->va_atime = np->n_atim; 1802 if (np->n_flag & NUPD) 1803 vaper->va_mtime = np->n_mtim; 1804 } 1805 } 1806 return (0); 1807 } 1808 1809 /* 1810 * Check the time stamp 1811 * If the cache is valid, copy contents to *vap and return 0 1812 * otherwise return an error 1813 */ 1814 int 1815 nfs_getattrcache(vp, vaper) 1816 struct vnode *vp; 1817 struct vattr *vaper; 1818 { 1819 struct nfsnode *np = VTONFS(vp); 1820 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1821 struct vattr *vap; 1822 1823 if (np->n_attrstamp == 0 || 1824 (time_second - np->n_attrstamp) >= nfs_attrtimeo(nmp, np)) { 1825 nfsstats.attrcache_misses++; 1826 return (ENOENT); 1827 } 1828 nfsstats.attrcache_hits++; 1829 vap = np->n_vattr; 1830 if (vap->va_size != np->n_size) { 1831 if (vap->va_type == VREG) { 1832 if ((np->n_flag & NMODIFIED) != 0 && 1833 vap->va_size < np->n_size) { 1834 vap->va_size = np->n_size; 1835 } else { 1836 np->n_size = vap->va_size; 1837 } 1838 genfs_node_wrlock(vp); 1839 uvm_vnp_setsize(vp, np->n_size); 1840 genfs_node_unlock(vp); 1841 } else 1842 np->n_size = vap->va_size; 1843 } 1844 memcpy((void *)vaper, (void *)vap, sizeof(struct vattr)); 1845 if (np->n_flag & NCHG) { 1846 if (np->n_flag & NACC) 1847 vaper->va_atime = np->n_atim; 1848 if (np->n_flag & NUPD) 1849 vaper->va_mtime = np->n_mtim; 1850 } 1851 return (0); 1852 } 1853 1854 void 1855 nfs_delayedtruncate(vp) 1856 struct vnode *vp; 1857 { 1858 struct nfsnode *np = VTONFS(vp); 1859 1860 if (np->n_flag & NTRUNCDELAYED) { 1861 np->n_flag &= ~NTRUNCDELAYED; 1862 genfs_node_wrlock(vp); 1863 mutex_enter(&vp->v_interlock); 1864 (void)VOP_PUTPAGES(vp, 0, 1865 0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES); 1866 uvm_vnp_setsize(vp, np->n_size); 1867 genfs_node_unlock(vp); 1868 } 1869 } 1870 1871 #define NFS_WCCKLUDGE_TIMEOUT (24 * 60 * 60) /* 1 day */ 1872 #define NFS_WCCKLUDGE(nmp, now) \ 1873 (((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \ 1874 ((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0) 1875 1876 /* 1877 * nfs_check_wccdata: check inaccurate wcc_data 1878 * 1879 * => return non-zero if we shouldn't trust the wcc_data. 1880 * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed". 1881 */ 1882 1883 int 1884 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime, 1885 struct timespec *mtime, bool docheck) 1886 { 1887 int error = 0; 1888 1889 #if !defined(NFS_V2_ONLY) 1890 1891 if (docheck) { 1892 struct vnode *vp = NFSTOV(np); 1893 struct nfsmount *nmp; 1894 long now = time_second; 1895 const struct timespec *omtime = &np->n_vattr->va_mtime; 1896 const struct timespec *octime = &np->n_vattr->va_ctime; 1897 const char *reason = NULL; /* XXX: gcc */ 1898 1899 if (timespeccmp(omtime, mtime, <=)) { 1900 reason = "mtime"; 1901 error = EINVAL; 1902 } 1903 1904 if (vp->v_type == VDIR && timespeccmp(octime, ctime, <=)) { 1905 reason = "ctime"; 1906 error = EINVAL; 1907 } 1908 1909 nmp = VFSTONFS(vp->v_mount); 1910 if (error) { 1911 1912 /* 1913 * despite of the fact that we've updated the file, 1914 * timestamps of the file were not updated as we 1915 * expected. 1916 * it means that the server has incompatible 1917 * semantics of timestamps or (more likely) 1918 * the server time is not precise enough to 1919 * track each modifications. 1920 * in that case, we disable wcc processing. 1921 * 1922 * yes, strictly speaking, we should disable all 1923 * caching. it's a compromise. 1924 */ 1925 1926 mutex_enter(&nmp->nm_lock); 1927 if (!NFS_WCCKLUDGE(nmp, now)) { 1928 printf("%s: inaccurate wcc data (%s) detected," 1929 " disabling wcc" 1930 " (ctime %u.%09u %u.%09u," 1931 " mtime %u.%09u %u.%09u)\n", 1932 vp->v_mount->mnt_stat.f_mntfromname, 1933 reason, 1934 (unsigned int)octime->tv_sec, 1935 (unsigned int)octime->tv_nsec, 1936 (unsigned int)ctime->tv_sec, 1937 (unsigned int)ctime->tv_nsec, 1938 (unsigned int)omtime->tv_sec, 1939 (unsigned int)omtime->tv_nsec, 1940 (unsigned int)mtime->tv_sec, 1941 (unsigned int)mtime->tv_nsec); 1942 } 1943 nmp->nm_iflag |= NFSMNT_WCCKLUDGE; 1944 nmp->nm_wcckludgetime = now; 1945 mutex_exit(&nmp->nm_lock); 1946 } else if (NFS_WCCKLUDGE(nmp, now)) { 1947 error = EPERM; /* XXX */ 1948 } else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1949 mutex_enter(&nmp->nm_lock); 1950 if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1951 printf("%s: re-enabling wcc\n", 1952 vp->v_mount->mnt_stat.f_mntfromname); 1953 nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE; 1954 } 1955 mutex_exit(&nmp->nm_lock); 1956 } 1957 } 1958 1959 #endif /* !defined(NFS_V2_ONLY) */ 1960 1961 return error; 1962 } 1963 1964 /* 1965 * Heuristic to see if the server XDR encodes directory cookies or not. 1966 * it is not supposed to, but a lot of servers may do this. Also, since 1967 * most/all servers will implement V2 as well, it is expected that they 1968 * may return just 32 bits worth of cookie information, so we need to 1969 * find out in which 32 bits this information is available. We do this 1970 * to avoid trouble with emulated binaries that can't handle 64 bit 1971 * directory offsets. 1972 */ 1973 1974 void 1975 nfs_cookieheuristic(vp, flagp, l, cred) 1976 struct vnode *vp; 1977 int *flagp; 1978 struct lwp *l; 1979 kauth_cred_t cred; 1980 { 1981 struct uio auio; 1982 struct iovec aiov; 1983 char *tbuf, *cp; 1984 struct dirent *dp; 1985 off_t *cookies = NULL, *cop; 1986 int error, eof, nc, len; 1987 1988 MALLOC(tbuf, void *, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK); 1989 1990 aiov.iov_base = tbuf; 1991 aiov.iov_len = NFS_DIRFRAGSIZ; 1992 auio.uio_iov = &aiov; 1993 auio.uio_iovcnt = 1; 1994 auio.uio_rw = UIO_READ; 1995 auio.uio_resid = NFS_DIRFRAGSIZ; 1996 auio.uio_offset = 0; 1997 UIO_SETUP_SYSSPACE(&auio); 1998 1999 error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc); 2000 2001 len = NFS_DIRFRAGSIZ - auio.uio_resid; 2002 if (error || len == 0) { 2003 FREE(tbuf, M_TEMP); 2004 if (cookies) 2005 free(cookies, M_TEMP); 2006 return; 2007 } 2008 2009 /* 2010 * Find the first valid entry and look at its offset cookie. 2011 */ 2012 2013 cp = tbuf; 2014 for (cop = cookies; len > 0; len -= dp->d_reclen) { 2015 dp = (struct dirent *)cp; 2016 if (dp->d_fileno != 0 && len >= dp->d_reclen) { 2017 if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) { 2018 *flagp |= NFSMNT_SWAPCOOKIE; 2019 nfs_invaldircache(vp, 0); 2020 nfs_vinvalbuf(vp, 0, cred, l, 1); 2021 } 2022 break; 2023 } 2024 cop++; 2025 cp += dp->d_reclen; 2026 } 2027 2028 FREE(tbuf, M_TEMP); 2029 free(cookies, M_TEMP); 2030 } 2031 #endif /* NFS */ 2032 2033 /* 2034 * A fiddled version of m_adj() that ensures null fill to a 32-bit 2035 * boundary and only trims off the back end 2036 * 2037 * 1. trim off 'len' bytes as m_adj(mp, -len). 2038 * 2. add zero-padding 'nul' bytes at the end of the mbuf chain. 2039 */ 2040 void 2041 nfs_zeropad(mp, len, nul) 2042 struct mbuf *mp; 2043 int len; 2044 int nul; 2045 { 2046 struct mbuf *m; 2047 int count; 2048 2049 /* 2050 * Trim from tail. Scan the mbuf chain, 2051 * calculating its length and finding the last mbuf. 2052 * If the adjustment only affects this mbuf, then just 2053 * adjust and return. Otherwise, rescan and truncate 2054 * after the remaining size. 2055 */ 2056 count = 0; 2057 m = mp; 2058 for (;;) { 2059 count += m->m_len; 2060 if (m->m_next == NULL) 2061 break; 2062 m = m->m_next; 2063 } 2064 2065 KDASSERT(count >= len); 2066 2067 if (m->m_len >= len) { 2068 m->m_len -= len; 2069 } else { 2070 count -= len; 2071 /* 2072 * Correct length for chain is "count". 2073 * Find the mbuf with last data, adjust its length, 2074 * and toss data from remaining mbufs on chain. 2075 */ 2076 for (m = mp; m; m = m->m_next) { 2077 if (m->m_len >= count) { 2078 m->m_len = count; 2079 break; 2080 } 2081 count -= m->m_len; 2082 } 2083 KASSERT(m && m->m_next); 2084 m_freem(m->m_next); 2085 m->m_next = NULL; 2086 } 2087 2088 KDASSERT(m->m_next == NULL); 2089 2090 /* 2091 * zero-padding. 2092 */ 2093 if (nul > 0) { 2094 char *cp; 2095 int i; 2096 2097 if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) { 2098 struct mbuf *n; 2099 2100 KDASSERT(MLEN >= nul); 2101 n = m_get(M_WAIT, MT_DATA); 2102 MCLAIM(n, &nfs_mowner); 2103 n->m_len = nul; 2104 n->m_next = NULL; 2105 m->m_next = n; 2106 cp = mtod(n, void *); 2107 } else { 2108 cp = mtod(m, char *) + m->m_len; 2109 m->m_len += nul; 2110 } 2111 for (i = 0; i < nul; i++) 2112 *cp++ = '\0'; 2113 } 2114 return; 2115 } 2116 2117 /* 2118 * Make these functions instead of macros, so that the kernel text size 2119 * doesn't get too big... 2120 */ 2121 void 2122 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp) 2123 struct nfsrv_descript *nfsd; 2124 int before_ret; 2125 struct vattr *before_vap; 2126 int after_ret; 2127 struct vattr *after_vap; 2128 struct mbuf **mbp; 2129 char **bposp; 2130 { 2131 struct mbuf *mb = *mbp; 2132 char *bpos = *bposp; 2133 u_int32_t *tl; 2134 2135 if (before_ret) { 2136 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2137 *tl = nfs_false; 2138 } else { 2139 nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED); 2140 *tl++ = nfs_true; 2141 txdr_hyper(before_vap->va_size, tl); 2142 tl += 2; 2143 txdr_nfsv3time(&(before_vap->va_mtime), tl); 2144 tl += 2; 2145 txdr_nfsv3time(&(before_vap->va_ctime), tl); 2146 } 2147 *bposp = bpos; 2148 *mbp = mb; 2149 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp); 2150 } 2151 2152 void 2153 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp) 2154 struct nfsrv_descript *nfsd; 2155 int after_ret; 2156 struct vattr *after_vap; 2157 struct mbuf **mbp; 2158 char **bposp; 2159 { 2160 struct mbuf *mb = *mbp; 2161 char *bpos = *bposp; 2162 u_int32_t *tl; 2163 struct nfs_fattr *fp; 2164 2165 if (after_ret) { 2166 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2167 *tl = nfs_false; 2168 } else { 2169 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR); 2170 *tl++ = nfs_true; 2171 fp = (struct nfs_fattr *)tl; 2172 nfsm_srvfattr(nfsd, after_vap, fp); 2173 } 2174 *mbp = mb; 2175 *bposp = bpos; 2176 } 2177 2178 void 2179 nfsm_srvfattr(nfsd, vap, fp) 2180 struct nfsrv_descript *nfsd; 2181 struct vattr *vap; 2182 struct nfs_fattr *fp; 2183 { 2184 2185 fp->fa_nlink = txdr_unsigned(vap->va_nlink); 2186 fp->fa_uid = txdr_unsigned(vap->va_uid); 2187 fp->fa_gid = txdr_unsigned(vap->va_gid); 2188 if (nfsd->nd_flag & ND_NFSV3) { 2189 fp->fa_type = vtonfsv3_type(vap->va_type); 2190 fp->fa_mode = vtonfsv3_mode(vap->va_mode); 2191 txdr_hyper(vap->va_size, &fp->fa3_size); 2192 txdr_hyper(vap->va_bytes, &fp->fa3_used); 2193 fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev)); 2194 fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev)); 2195 fp->fa3_fsid.nfsuquad[0] = 0; 2196 fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid); 2197 txdr_hyper(vap->va_fileid, &fp->fa3_fileid); 2198 txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime); 2199 txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime); 2200 txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime); 2201 } else { 2202 fp->fa_type = vtonfsv2_type(vap->va_type); 2203 fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 2204 fp->fa2_size = txdr_unsigned(vap->va_size); 2205 fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize); 2206 if (vap->va_type == VFIFO) 2207 fp->fa2_rdev = 0xffffffff; 2208 else 2209 fp->fa2_rdev = txdr_unsigned(vap->va_rdev); 2210 fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE); 2211 fp->fa2_fsid = txdr_unsigned(vap->va_fsid); 2212 fp->fa2_fileid = txdr_unsigned(vap->va_fileid); 2213 txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime); 2214 txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime); 2215 txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime); 2216 } 2217 } 2218 2219 /* 2220 * This function compares two net addresses by family and returns true 2221 * if they are the same host. 2222 * If there is any doubt, return false. 2223 * The AF_INET family is handled as a special case so that address mbufs 2224 * don't need to be saved to store "struct in_addr", which is only 4 bytes. 2225 */ 2226 int 2227 netaddr_match(family, haddr, nam) 2228 int family; 2229 union nethostaddr *haddr; 2230 struct mbuf *nam; 2231 { 2232 struct sockaddr_in *inetaddr; 2233 2234 switch (family) { 2235 case AF_INET: 2236 inetaddr = mtod(nam, struct sockaddr_in *); 2237 if (inetaddr->sin_family == AF_INET && 2238 inetaddr->sin_addr.s_addr == haddr->had_inetaddr) 2239 return (1); 2240 break; 2241 case AF_INET6: 2242 { 2243 struct sockaddr_in6 *sin6_1, *sin6_2; 2244 2245 sin6_1 = mtod(nam, struct sockaddr_in6 *); 2246 sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *); 2247 if (sin6_1->sin6_family == AF_INET6 && 2248 IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr)) 2249 return 1; 2250 } 2251 default: 2252 break; 2253 }; 2254 return (0); 2255 } 2256 2257 /* 2258 * The write verifier has changed (probably due to a server reboot), so all 2259 * PG_NEEDCOMMIT pages will have to be written again. Since they are marked 2260 * as dirty or are being written out just now, all this takes is clearing 2261 * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for 2262 * the mount point. 2263 */ 2264 void 2265 nfs_clearcommit(mp) 2266 struct mount *mp; 2267 { 2268 struct vnode *vp; 2269 struct nfsnode *np; 2270 struct vm_page *pg; 2271 struct nfsmount *nmp = VFSTONFS(mp); 2272 2273 rw_enter(&nmp->nm_writeverflock, RW_WRITER); 2274 mutex_enter(&mntvnode_lock); 2275 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2276 KASSERT(vp->v_mount == mp); 2277 if (vp->v_type != VREG) 2278 continue; 2279 np = VTONFS(vp); 2280 np->n_pushlo = np->n_pushhi = np->n_pushedlo = 2281 np->n_pushedhi = 0; 2282 np->n_commitflags &= 2283 ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID); 2284 mutex_enter(&vp->v_uobj.vmobjlock); 2285 TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq.queue) { 2286 pg->flags &= ~PG_NEEDCOMMIT; 2287 } 2288 mutex_exit(&vp->v_uobj.vmobjlock); 2289 } 2290 mutex_exit(&mntvnode_lock); 2291 mutex_enter(&nmp->nm_lock); 2292 nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF; 2293 mutex_exit(&nmp->nm_lock); 2294 rw_exit(&nmp->nm_writeverflock); 2295 } 2296 2297 void 2298 nfs_merge_commit_ranges(vp) 2299 struct vnode *vp; 2300 { 2301 struct nfsnode *np = VTONFS(vp); 2302 2303 KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID); 2304 2305 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2306 np->n_pushedlo = np->n_pushlo; 2307 np->n_pushedhi = np->n_pushhi; 2308 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2309 } else { 2310 if (np->n_pushlo < np->n_pushedlo) 2311 np->n_pushedlo = np->n_pushlo; 2312 if (np->n_pushhi > np->n_pushedhi) 2313 np->n_pushedhi = np->n_pushhi; 2314 } 2315 2316 np->n_pushlo = np->n_pushhi = 0; 2317 np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID; 2318 2319 #ifdef NFS_DEBUG_COMMIT 2320 printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2321 (unsigned)np->n_pushedhi); 2322 #endif 2323 } 2324 2325 int 2326 nfs_in_committed_range(vp, off, len) 2327 struct vnode *vp; 2328 off_t off, len; 2329 { 2330 struct nfsnode *np = VTONFS(vp); 2331 off_t lo, hi; 2332 2333 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2334 return 0; 2335 lo = off; 2336 hi = lo + len; 2337 2338 return (lo >= np->n_pushedlo && hi <= np->n_pushedhi); 2339 } 2340 2341 int 2342 nfs_in_tobecommitted_range(vp, off, len) 2343 struct vnode *vp; 2344 off_t off, len; 2345 { 2346 struct nfsnode *np = VTONFS(vp); 2347 off_t lo, hi; 2348 2349 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2350 return 0; 2351 lo = off; 2352 hi = lo + len; 2353 2354 return (lo >= np->n_pushlo && hi <= np->n_pushhi); 2355 } 2356 2357 void 2358 nfs_add_committed_range(vp, off, len) 2359 struct vnode *vp; 2360 off_t off, len; 2361 { 2362 struct nfsnode *np = VTONFS(vp); 2363 off_t lo, hi; 2364 2365 lo = off; 2366 hi = lo + len; 2367 2368 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2369 np->n_pushedlo = lo; 2370 np->n_pushedhi = hi; 2371 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2372 } else { 2373 if (hi > np->n_pushedhi) 2374 np->n_pushedhi = hi; 2375 if (lo < np->n_pushedlo) 2376 np->n_pushedlo = lo; 2377 } 2378 #ifdef NFS_DEBUG_COMMIT 2379 printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2380 (unsigned)np->n_pushedhi); 2381 #endif 2382 } 2383 2384 void 2385 nfs_del_committed_range(vp, off, len) 2386 struct vnode *vp; 2387 off_t off, len; 2388 { 2389 struct nfsnode *np = VTONFS(vp); 2390 off_t lo, hi; 2391 2392 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2393 return; 2394 2395 lo = off; 2396 hi = lo + len; 2397 2398 if (lo > np->n_pushedhi || hi < np->n_pushedlo) 2399 return; 2400 if (lo <= np->n_pushedlo) 2401 np->n_pushedlo = hi; 2402 else if (hi >= np->n_pushedhi) 2403 np->n_pushedhi = lo; 2404 else { 2405 /* 2406 * XXX There's only one range. If the deleted range 2407 * is in the middle, pick the largest of the 2408 * contiguous ranges that it leaves. 2409 */ 2410 if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi)) 2411 np->n_pushedhi = lo; 2412 else 2413 np->n_pushedlo = hi; 2414 } 2415 #ifdef NFS_DEBUG_COMMIT 2416 printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2417 (unsigned)np->n_pushedhi); 2418 #endif 2419 } 2420 2421 void 2422 nfs_add_tobecommitted_range(vp, off, len) 2423 struct vnode *vp; 2424 off_t off, len; 2425 { 2426 struct nfsnode *np = VTONFS(vp); 2427 off_t lo, hi; 2428 2429 lo = off; 2430 hi = lo + len; 2431 2432 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) { 2433 np->n_pushlo = lo; 2434 np->n_pushhi = hi; 2435 np->n_commitflags |= NFS_COMMIT_PUSH_VALID; 2436 } else { 2437 if (lo < np->n_pushlo) 2438 np->n_pushlo = lo; 2439 if (hi > np->n_pushhi) 2440 np->n_pushhi = hi; 2441 } 2442 #ifdef NFS_DEBUG_COMMIT 2443 printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2444 (unsigned)np->n_pushhi); 2445 #endif 2446 } 2447 2448 void 2449 nfs_del_tobecommitted_range(vp, off, len) 2450 struct vnode *vp; 2451 off_t off, len; 2452 { 2453 struct nfsnode *np = VTONFS(vp); 2454 off_t lo, hi; 2455 2456 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2457 return; 2458 2459 lo = off; 2460 hi = lo + len; 2461 2462 if (lo > np->n_pushhi || hi < np->n_pushlo) 2463 return; 2464 2465 if (lo <= np->n_pushlo) 2466 np->n_pushlo = hi; 2467 else if (hi >= np->n_pushhi) 2468 np->n_pushhi = lo; 2469 else { 2470 /* 2471 * XXX There's only one range. If the deleted range 2472 * is in the middle, pick the largest of the 2473 * contiguous ranges that it leaves. 2474 */ 2475 if ((np->n_pushlo - lo) > (hi - np->n_pushhi)) 2476 np->n_pushhi = lo; 2477 else 2478 np->n_pushlo = hi; 2479 } 2480 #ifdef NFS_DEBUG_COMMIT 2481 printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2482 (unsigned)np->n_pushhi); 2483 #endif 2484 } 2485 2486 /* 2487 * Map errnos to NFS error numbers. For Version 3 also filter out error 2488 * numbers not specified for the associated procedure. 2489 */ 2490 int 2491 nfsrv_errmap(nd, err) 2492 struct nfsrv_descript *nd; 2493 int err; 2494 { 2495 const short *defaulterrp, *errp; 2496 2497 if (nd->nd_flag & ND_NFSV3) { 2498 if (nd->nd_procnum <= NFSPROC_COMMIT) { 2499 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum]; 2500 while (*++errp) { 2501 if (*errp == err) 2502 return (err); 2503 else if (*errp > err) 2504 break; 2505 } 2506 return ((int)*defaulterrp); 2507 } else 2508 return (err & 0xffff); 2509 } 2510 if (err <= ELAST) 2511 return ((int)nfsrv_v2errmap[err - 1]); 2512 return (NFSERR_IO); 2513 } 2514 2515 u_int32_t 2516 nfs_getxid() 2517 { 2518 u_int32_t newxid; 2519 2520 /* get next xid. skip 0 */ 2521 do { 2522 newxid = atomic_inc_32_nv(&nfs_xid); 2523 } while (__predict_false(newxid == 0)); 2524 2525 return txdr_unsigned(newxid); 2526 } 2527 2528 /* 2529 * assign a new xid for existing request. 2530 * used for NFSERR_JUKEBOX handling. 2531 */ 2532 void 2533 nfs_renewxid(struct nfsreq *req) 2534 { 2535 u_int32_t xid; 2536 int off; 2537 2538 xid = nfs_getxid(); 2539 if (req->r_nmp->nm_sotype == SOCK_STREAM) 2540 off = sizeof(u_int32_t); /* RPC record mark */ 2541 else 2542 off = 0; 2543 2544 m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid); 2545 req->r_xid = xid; 2546 } 2547 2548 #if defined(NFS) 2549 /* 2550 * Set the attribute timeout based on how recently the file has been modified. 2551 */ 2552 2553 time_t 2554 nfs_attrtimeo(struct nfsmount *nmp, struct nfsnode *np) 2555 { 2556 time_t timeo; 2557 2558 if ((nmp->nm_flag & NFSMNT_NOAC) != 0) 2559 return 0; 2560 2561 if (((np)->n_flag & NMODIFIED) != 0) 2562 return NFS_MINATTRTIMO; 2563 2564 timeo = (time_second - np->n_mtime.tv_sec) / 10; 2565 timeo = max(timeo, NFS_MINATTRTIMO); 2566 timeo = min(timeo, NFS_MAXATTRTIMO); 2567 return timeo; 2568 } 2569 #endif /* defined(NFS) */ 2570
Indexes created Fri Sep 26 20:09:58 GMT 2025