1 /* $NetBSD: nfs_subs.c,v 1.218 2009/12/31 20:01:33 christos 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 1.218 2009/12/31 20:01:33 christos 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 int nuidhash_max = NFS_MAXUIDHASH; 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 /* NFS client/server stats. */ 145 struct nfsstats nfsstats; 146 147 /* 148 * Mapping of old NFS Version 2 RPC numbers to generic numbers. 149 */ 150 const int nfsv3_procid[NFS_NPROCS] = { 151 NFSPROC_NULL, 152 NFSPROC_GETATTR, 153 NFSPROC_SETATTR, 154 NFSPROC_NOOP, 155 NFSPROC_LOOKUP, 156 NFSPROC_READLINK, 157 NFSPROC_READ, 158 NFSPROC_NOOP, 159 NFSPROC_WRITE, 160 NFSPROC_CREATE, 161 NFSPROC_REMOVE, 162 NFSPROC_RENAME, 163 NFSPROC_LINK, 164 NFSPROC_SYMLINK, 165 NFSPROC_MKDIR, 166 NFSPROC_RMDIR, 167 NFSPROC_READDIR, 168 NFSPROC_FSSTAT, 169 NFSPROC_NOOP, 170 NFSPROC_NOOP, 171 NFSPROC_NOOP, 172 NFSPROC_NOOP, 173 NFSPROC_NOOP 174 }; 175 176 /* 177 * and the reverse mapping from generic to Version 2 procedure numbers 178 */ 179 const int nfsv2_procid[NFS_NPROCS] = { 180 NFSV2PROC_NULL, 181 NFSV2PROC_GETATTR, 182 NFSV2PROC_SETATTR, 183 NFSV2PROC_LOOKUP, 184 NFSV2PROC_NOOP, 185 NFSV2PROC_READLINK, 186 NFSV2PROC_READ, 187 NFSV2PROC_WRITE, 188 NFSV2PROC_CREATE, 189 NFSV2PROC_MKDIR, 190 NFSV2PROC_SYMLINK, 191 NFSV2PROC_CREATE, 192 NFSV2PROC_REMOVE, 193 NFSV2PROC_RMDIR, 194 NFSV2PROC_RENAME, 195 NFSV2PROC_LINK, 196 NFSV2PROC_READDIR, 197 NFSV2PROC_NOOP, 198 NFSV2PROC_STATFS, 199 NFSV2PROC_NOOP, 200 NFSV2PROC_NOOP, 201 NFSV2PROC_NOOP, 202 NFSV2PROC_NOOP, 203 }; 204 205 /* 206 * Maps errno values to nfs error numbers. 207 * Use NFSERR_IO as the catch all for ones not specifically defined in 208 * RFC 1094. 209 */ 210 static const u_char nfsrv_v2errmap[ELAST] = { 211 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 212 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 213 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 214 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 215 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 216 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 217 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 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_NAMETOL, NFSERR_IO, NFSERR_IO, 224 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 225 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 226 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 227 NFSERR_IO, NFSERR_IO, 228 }; 229 230 /* 231 * Maps errno values to nfs error numbers. 232 * Although it is not obvious whether or not NFS clients really care if 233 * a returned error value is in the specified list for the procedure, the 234 * safest thing to do is filter them appropriately. For Version 2, the 235 * X/Open XNFS document is the only specification that defines error values 236 * for each RPC (The RFC simply lists all possible error values for all RPCs), 237 * so I have decided to not do this for Version 2. 238 * The first entry is the default error return and the rest are the valid 239 * errors for that RPC in increasing numeric order. 240 */ 241 static const short nfsv3err_null[] = { 242 0, 243 0, 244 }; 245 246 static const short nfsv3err_getattr[] = { 247 NFSERR_IO, 248 NFSERR_IO, 249 NFSERR_STALE, 250 NFSERR_BADHANDLE, 251 NFSERR_SERVERFAULT, 252 0, 253 }; 254 255 static const short nfsv3err_setattr[] = { 256 NFSERR_IO, 257 NFSERR_PERM, 258 NFSERR_IO, 259 NFSERR_ACCES, 260 NFSERR_INVAL, 261 NFSERR_NOSPC, 262 NFSERR_ROFS, 263 NFSERR_DQUOT, 264 NFSERR_STALE, 265 NFSERR_BADHANDLE, 266 NFSERR_NOT_SYNC, 267 NFSERR_SERVERFAULT, 268 0, 269 }; 270 271 static const short nfsv3err_lookup[] = { 272 NFSERR_IO, 273 NFSERR_NOENT, 274 NFSERR_IO, 275 NFSERR_ACCES, 276 NFSERR_NOTDIR, 277 NFSERR_NAMETOL, 278 NFSERR_STALE, 279 NFSERR_BADHANDLE, 280 NFSERR_SERVERFAULT, 281 0, 282 }; 283 284 static const short nfsv3err_access[] = { 285 NFSERR_IO, 286 NFSERR_IO, 287 NFSERR_STALE, 288 NFSERR_BADHANDLE, 289 NFSERR_SERVERFAULT, 290 0, 291 }; 292 293 static const short nfsv3err_readlink[] = { 294 NFSERR_IO, 295 NFSERR_IO, 296 NFSERR_ACCES, 297 NFSERR_INVAL, 298 NFSERR_STALE, 299 NFSERR_BADHANDLE, 300 NFSERR_NOTSUPP, 301 NFSERR_SERVERFAULT, 302 0, 303 }; 304 305 static const short nfsv3err_read[] = { 306 NFSERR_IO, 307 NFSERR_IO, 308 NFSERR_NXIO, 309 NFSERR_ACCES, 310 NFSERR_INVAL, 311 NFSERR_STALE, 312 NFSERR_BADHANDLE, 313 NFSERR_SERVERFAULT, 314 NFSERR_JUKEBOX, 315 0, 316 }; 317 318 static const short nfsv3err_write[] = { 319 NFSERR_IO, 320 NFSERR_IO, 321 NFSERR_ACCES, 322 NFSERR_INVAL, 323 NFSERR_FBIG, 324 NFSERR_NOSPC, 325 NFSERR_ROFS, 326 NFSERR_DQUOT, 327 NFSERR_STALE, 328 NFSERR_BADHANDLE, 329 NFSERR_SERVERFAULT, 330 NFSERR_JUKEBOX, 331 0, 332 }; 333 334 static const short nfsv3err_create[] = { 335 NFSERR_IO, 336 NFSERR_IO, 337 NFSERR_ACCES, 338 NFSERR_EXIST, 339 NFSERR_NOTDIR, 340 NFSERR_NOSPC, 341 NFSERR_ROFS, 342 NFSERR_NAMETOL, 343 NFSERR_DQUOT, 344 NFSERR_STALE, 345 NFSERR_BADHANDLE, 346 NFSERR_NOTSUPP, 347 NFSERR_SERVERFAULT, 348 0, 349 }; 350 351 static const short nfsv3err_mkdir[] = { 352 NFSERR_IO, 353 NFSERR_IO, 354 NFSERR_ACCES, 355 NFSERR_EXIST, 356 NFSERR_NOTDIR, 357 NFSERR_NOSPC, 358 NFSERR_ROFS, 359 NFSERR_NAMETOL, 360 NFSERR_DQUOT, 361 NFSERR_STALE, 362 NFSERR_BADHANDLE, 363 NFSERR_NOTSUPP, 364 NFSERR_SERVERFAULT, 365 0, 366 }; 367 368 static const short nfsv3err_symlink[] = { 369 NFSERR_IO, 370 NFSERR_IO, 371 NFSERR_ACCES, 372 NFSERR_EXIST, 373 NFSERR_NOTDIR, 374 NFSERR_NOSPC, 375 NFSERR_ROFS, 376 NFSERR_NAMETOL, 377 NFSERR_DQUOT, 378 NFSERR_STALE, 379 NFSERR_BADHANDLE, 380 NFSERR_NOTSUPP, 381 NFSERR_SERVERFAULT, 382 0, 383 }; 384 385 static const short nfsv3err_mknod[] = { 386 NFSERR_IO, 387 NFSERR_IO, 388 NFSERR_ACCES, 389 NFSERR_EXIST, 390 NFSERR_NOTDIR, 391 NFSERR_NOSPC, 392 NFSERR_ROFS, 393 NFSERR_NAMETOL, 394 NFSERR_DQUOT, 395 NFSERR_STALE, 396 NFSERR_BADHANDLE, 397 NFSERR_NOTSUPP, 398 NFSERR_SERVERFAULT, 399 NFSERR_BADTYPE, 400 0, 401 }; 402 403 static const short nfsv3err_remove[] = { 404 NFSERR_IO, 405 NFSERR_NOENT, 406 NFSERR_IO, 407 NFSERR_ACCES, 408 NFSERR_NOTDIR, 409 NFSERR_ROFS, 410 NFSERR_NAMETOL, 411 NFSERR_STALE, 412 NFSERR_BADHANDLE, 413 NFSERR_SERVERFAULT, 414 0, 415 }; 416 417 static const short nfsv3err_rmdir[] = { 418 NFSERR_IO, 419 NFSERR_NOENT, 420 NFSERR_IO, 421 NFSERR_ACCES, 422 NFSERR_EXIST, 423 NFSERR_NOTDIR, 424 NFSERR_INVAL, 425 NFSERR_ROFS, 426 NFSERR_NAMETOL, 427 NFSERR_NOTEMPTY, 428 NFSERR_STALE, 429 NFSERR_BADHANDLE, 430 NFSERR_NOTSUPP, 431 NFSERR_SERVERFAULT, 432 0, 433 }; 434 435 static const short nfsv3err_rename[] = { 436 NFSERR_IO, 437 NFSERR_NOENT, 438 NFSERR_IO, 439 NFSERR_ACCES, 440 NFSERR_EXIST, 441 NFSERR_XDEV, 442 NFSERR_NOTDIR, 443 NFSERR_ISDIR, 444 NFSERR_INVAL, 445 NFSERR_NOSPC, 446 NFSERR_ROFS, 447 NFSERR_MLINK, 448 NFSERR_NAMETOL, 449 NFSERR_NOTEMPTY, 450 NFSERR_DQUOT, 451 NFSERR_STALE, 452 NFSERR_BADHANDLE, 453 NFSERR_NOTSUPP, 454 NFSERR_SERVERFAULT, 455 0, 456 }; 457 458 static const short nfsv3err_link[] = { 459 NFSERR_IO, 460 NFSERR_IO, 461 NFSERR_ACCES, 462 NFSERR_EXIST, 463 NFSERR_XDEV, 464 NFSERR_NOTDIR, 465 NFSERR_INVAL, 466 NFSERR_NOSPC, 467 NFSERR_ROFS, 468 NFSERR_MLINK, 469 NFSERR_NAMETOL, 470 NFSERR_DQUOT, 471 NFSERR_STALE, 472 NFSERR_BADHANDLE, 473 NFSERR_NOTSUPP, 474 NFSERR_SERVERFAULT, 475 0, 476 }; 477 478 static const short nfsv3err_readdir[] = { 479 NFSERR_IO, 480 NFSERR_IO, 481 NFSERR_ACCES, 482 NFSERR_NOTDIR, 483 NFSERR_STALE, 484 NFSERR_BADHANDLE, 485 NFSERR_BAD_COOKIE, 486 NFSERR_TOOSMALL, 487 NFSERR_SERVERFAULT, 488 0, 489 }; 490 491 static const short nfsv3err_readdirplus[] = { 492 NFSERR_IO, 493 NFSERR_IO, 494 NFSERR_ACCES, 495 NFSERR_NOTDIR, 496 NFSERR_STALE, 497 NFSERR_BADHANDLE, 498 NFSERR_BAD_COOKIE, 499 NFSERR_NOTSUPP, 500 NFSERR_TOOSMALL, 501 NFSERR_SERVERFAULT, 502 0, 503 }; 504 505 static const short nfsv3err_fsstat[] = { 506 NFSERR_IO, 507 NFSERR_IO, 508 NFSERR_STALE, 509 NFSERR_BADHANDLE, 510 NFSERR_SERVERFAULT, 511 0, 512 }; 513 514 static const short nfsv3err_fsinfo[] = { 515 NFSERR_STALE, 516 NFSERR_STALE, 517 NFSERR_BADHANDLE, 518 NFSERR_SERVERFAULT, 519 0, 520 }; 521 522 static const short nfsv3err_pathconf[] = { 523 NFSERR_STALE, 524 NFSERR_STALE, 525 NFSERR_BADHANDLE, 526 NFSERR_SERVERFAULT, 527 0, 528 }; 529 530 static const short nfsv3err_commit[] = { 531 NFSERR_IO, 532 NFSERR_IO, 533 NFSERR_STALE, 534 NFSERR_BADHANDLE, 535 NFSERR_SERVERFAULT, 536 0, 537 }; 538 539 static const short * const nfsrv_v3errmap[] = { 540 nfsv3err_null, 541 nfsv3err_getattr, 542 nfsv3err_setattr, 543 nfsv3err_lookup, 544 nfsv3err_access, 545 nfsv3err_readlink, 546 nfsv3err_read, 547 nfsv3err_write, 548 nfsv3err_create, 549 nfsv3err_mkdir, 550 nfsv3err_symlink, 551 nfsv3err_mknod, 552 nfsv3err_remove, 553 nfsv3err_rmdir, 554 nfsv3err_rename, 555 nfsv3err_link, 556 nfsv3err_readdir, 557 nfsv3err_readdirplus, 558 nfsv3err_fsstat, 559 nfsv3err_fsinfo, 560 nfsv3err_pathconf, 561 nfsv3err_commit, 562 }; 563 564 extern struct nfsrtt nfsrtt; 565 566 u_long nfsdirhashmask; 567 568 int nfs_webnamei(struct nameidata *, struct vnode *, struct proc *); 569 570 /* 571 * Create the header for an rpc request packet 572 * The hsiz is the size of the rest of the nfs request header. 573 * (just used to decide if a cluster is a good idea) 574 */ 575 struct mbuf * 576 nfsm_reqh(struct nfsnode *np, u_long procid, int hsiz, char **bposp) 577 { 578 struct mbuf *mb; 579 char *bpos; 580 581 mb = m_get(M_WAIT, MT_DATA); 582 MCLAIM(mb, &nfs_mowner); 583 if (hsiz >= MINCLSIZE) 584 m_clget(mb, M_WAIT); 585 mb->m_len = 0; 586 bpos = mtod(mb, void *); 587 588 /* Finally, return values */ 589 *bposp = bpos; 590 return (mb); 591 } 592 593 /* 594 * Build the RPC header and fill in the authorization info. 595 * The authorization string argument is only used when the credentials 596 * come from outside of the kernel. 597 * Returns the head of the mbuf list. 598 */ 599 struct mbuf * 600 nfsm_rpchead(kauth_cred_t cr, int nmflag, int procid, 601 int auth_type, int auth_len, char *auth_str, int verf_len, 602 char *verf_str, struct mbuf *mrest, int mrest_len, 603 struct mbuf **mbp, uint32_t *xidp) 604 { 605 struct mbuf *mb; 606 u_int32_t *tl; 607 char *bpos; 608 int i; 609 struct mbuf *mreq; 610 int siz, grpsiz, authsiz; 611 612 authsiz = nfsm_rndup(auth_len); 613 mb = m_gethdr(M_WAIT, MT_DATA); 614 MCLAIM(mb, &nfs_mowner); 615 if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) { 616 m_clget(mb, M_WAIT); 617 } else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) { 618 MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED); 619 } else { 620 MH_ALIGN(mb, 8 * NFSX_UNSIGNED); 621 } 622 mb->m_len = 0; 623 mreq = mb; 624 bpos = mtod(mb, void *); 625 626 /* 627 * First the RPC header. 628 */ 629 nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 630 631 *tl++ = *xidp = nfs_getxid(); 632 *tl++ = rpc_call; 633 *tl++ = rpc_vers; 634 *tl++ = txdr_unsigned(NFS_PROG); 635 if (nmflag & NFSMNT_NFSV3) 636 *tl++ = txdr_unsigned(NFS_VER3); 637 else 638 *tl++ = txdr_unsigned(NFS_VER2); 639 if (nmflag & NFSMNT_NFSV3) 640 *tl++ = txdr_unsigned(procid); 641 else 642 *tl++ = txdr_unsigned(nfsv2_procid[procid]); 643 644 /* 645 * And then the authorization cred. 646 */ 647 *tl++ = txdr_unsigned(auth_type); 648 *tl = txdr_unsigned(authsiz); 649 switch (auth_type) { 650 case RPCAUTH_UNIX: 651 nfsm_build(tl, u_int32_t *, auth_len); 652 *tl++ = 0; /* stamp ?? */ 653 *tl++ = 0; /* NULL hostname */ 654 *tl++ = txdr_unsigned(kauth_cred_geteuid(cr)); 655 *tl++ = txdr_unsigned(kauth_cred_getegid(cr)); 656 grpsiz = (auth_len >> 2) - 5; 657 *tl++ = txdr_unsigned(grpsiz); 658 for (i = 0; i < grpsiz; i++) 659 *tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */ 660 break; 661 case RPCAUTH_KERB4: 662 siz = auth_len; 663 while (siz > 0) { 664 if (M_TRAILINGSPACE(mb) == 0) { 665 struct mbuf *mb2; 666 mb2 = m_get(M_WAIT, MT_DATA); 667 MCLAIM(mb2, &nfs_mowner); 668 if (siz >= MINCLSIZE) 669 m_clget(mb2, M_WAIT); 670 mb->m_next = mb2; 671 mb = mb2; 672 mb->m_len = 0; 673 bpos = mtod(mb, void *); 674 } 675 i = min(siz, M_TRAILINGSPACE(mb)); 676 memcpy(bpos, auth_str, i); 677 mb->m_len += i; 678 auth_str += i; 679 bpos += i; 680 siz -= i; 681 } 682 if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) { 683 for (i = 0; i < siz; i++) 684 *bpos++ = '\0'; 685 mb->m_len += siz; 686 } 687 break; 688 }; 689 690 /* 691 * And the verifier... 692 */ 693 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 694 if (verf_str) { 695 *tl++ = txdr_unsigned(RPCAUTH_KERB4); 696 *tl = txdr_unsigned(verf_len); 697 siz = verf_len; 698 while (siz > 0) { 699 if (M_TRAILINGSPACE(mb) == 0) { 700 struct mbuf *mb2; 701 mb2 = m_get(M_WAIT, MT_DATA); 702 MCLAIM(mb2, &nfs_mowner); 703 if (siz >= MINCLSIZE) 704 m_clget(mb2, M_WAIT); 705 mb->m_next = mb2; 706 mb = mb2; 707 mb->m_len = 0; 708 bpos = mtod(mb, void *); 709 } 710 i = min(siz, M_TRAILINGSPACE(mb)); 711 memcpy(bpos, verf_str, i); 712 mb->m_len += i; 713 verf_str += i; 714 bpos += i; 715 siz -= i; 716 } 717 if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) { 718 for (i = 0; i < siz; i++) 719 *bpos++ = '\0'; 720 mb->m_len += siz; 721 } 722 } else { 723 *tl++ = txdr_unsigned(RPCAUTH_NULL); 724 *tl = 0; 725 } 726 mb->m_next = mrest; 727 mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len; 728 mreq->m_pkthdr.rcvif = (struct ifnet *)0; 729 *mbp = mb; 730 return (mreq); 731 } 732 733 /* 734 * copies mbuf chain to the uio scatter/gather list 735 */ 736 int 737 nfsm_mbuftouio(struct mbuf **mrep, struct uio *uiop, int siz, char **dpos) 738 { 739 char *mbufcp, *uiocp; 740 int xfer, left, len; 741 struct mbuf *mp; 742 long uiosiz, rem; 743 int error = 0; 744 745 mp = *mrep; 746 mbufcp = *dpos; 747 len = mtod(mp, char *) + mp->m_len - mbufcp; 748 rem = nfsm_rndup(siz)-siz; 749 while (siz > 0) { 750 if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) 751 return (EFBIG); 752 left = uiop->uio_iov->iov_len; 753 uiocp = uiop->uio_iov->iov_base; 754 if (left > siz) 755 left = siz; 756 uiosiz = left; 757 while (left > 0) { 758 while (len == 0) { 759 mp = mp->m_next; 760 if (mp == NULL) 761 return (EBADRPC); 762 mbufcp = mtod(mp, void *); 763 len = mp->m_len; 764 } 765 xfer = (left > len) ? len : left; 766 error = copyout_vmspace(uiop->uio_vmspace, mbufcp, 767 uiocp, xfer); 768 if (error) { 769 return error; 770 } 771 left -= xfer; 772 len -= xfer; 773 mbufcp += xfer; 774 uiocp += xfer; 775 uiop->uio_offset += xfer; 776 uiop->uio_resid -= xfer; 777 } 778 if (uiop->uio_iov->iov_len <= siz) { 779 uiop->uio_iovcnt--; 780 uiop->uio_iov++; 781 } else { 782 uiop->uio_iov->iov_base = 783 (char *)uiop->uio_iov->iov_base + uiosiz; 784 uiop->uio_iov->iov_len -= uiosiz; 785 } 786 siz -= uiosiz; 787 } 788 *dpos = mbufcp; 789 *mrep = mp; 790 if (rem > 0) { 791 if (len < rem) 792 error = nfs_adv(mrep, dpos, rem, len); 793 else 794 *dpos += rem; 795 } 796 return (error); 797 } 798 799 /* 800 * copies a uio scatter/gather list to an mbuf chain. 801 * NOTE: can ony handle iovcnt == 1 802 */ 803 int 804 nfsm_uiotombuf(struct uio *uiop, struct mbuf **mq, int siz, char **bpos) 805 { 806 char *uiocp; 807 struct mbuf *mp, *mp2; 808 int xfer, left, mlen; 809 int uiosiz, clflg, rem; 810 char *cp; 811 int error; 812 813 #ifdef DIAGNOSTIC 814 if (uiop->uio_iovcnt != 1) 815 panic("nfsm_uiotombuf: iovcnt != 1"); 816 #endif 817 818 if (siz > MLEN) /* or should it >= MCLBYTES ?? */ 819 clflg = 1; 820 else 821 clflg = 0; 822 rem = nfsm_rndup(siz)-siz; 823 mp = mp2 = *mq; 824 while (siz > 0) { 825 left = uiop->uio_iov->iov_len; 826 uiocp = uiop->uio_iov->iov_base; 827 if (left > siz) 828 left = siz; 829 uiosiz = left; 830 while (left > 0) { 831 mlen = M_TRAILINGSPACE(mp); 832 if (mlen == 0) { 833 mp = m_get(M_WAIT, MT_DATA); 834 MCLAIM(mp, &nfs_mowner); 835 if (clflg) 836 m_clget(mp, M_WAIT); 837 mp->m_len = 0; 838 mp2->m_next = mp; 839 mp2 = mp; 840 mlen = M_TRAILINGSPACE(mp); 841 } 842 xfer = (left > mlen) ? mlen : left; 843 cp = mtod(mp, char *) + mp->m_len; 844 error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp, 845 xfer); 846 if (error) { 847 /* XXX */ 848 } 849 mp->m_len += xfer; 850 left -= xfer; 851 uiocp += xfer; 852 uiop->uio_offset += xfer; 853 uiop->uio_resid -= xfer; 854 } 855 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + 856 uiosiz; 857 uiop->uio_iov->iov_len -= uiosiz; 858 siz -= uiosiz; 859 } 860 if (rem > 0) { 861 if (rem > M_TRAILINGSPACE(mp)) { 862 mp = m_get(M_WAIT, MT_DATA); 863 MCLAIM(mp, &nfs_mowner); 864 mp->m_len = 0; 865 mp2->m_next = mp; 866 } 867 cp = mtod(mp, char *) + mp->m_len; 868 for (left = 0; left < rem; left++) 869 *cp++ = '\0'; 870 mp->m_len += rem; 871 *bpos = cp; 872 } else 873 *bpos = mtod(mp, char *) + mp->m_len; 874 *mq = mp; 875 return (0); 876 } 877 878 /* 879 * Get at least "siz" bytes of correctly aligned data. 880 * When called the mbuf pointers are not necessarily correct, 881 * dsosp points to what ought to be in m_data and left contains 882 * what ought to be in m_len. 883 * This is used by the macros nfsm_dissect and nfsm_dissecton for tough 884 * cases. (The macros use the vars. dpos and dpos2) 885 */ 886 int 887 nfsm_disct(struct mbuf **mdp, char **dposp, int siz, int left, char **cp2) 888 { 889 struct mbuf *m1, *m2; 890 struct mbuf *havebuf = NULL; 891 char *src = *dposp; 892 char *dst; 893 int len; 894 895 #ifdef DEBUG 896 if (left < 0) 897 panic("nfsm_disct: left < 0"); 898 #endif 899 m1 = *mdp; 900 /* 901 * Skip through the mbuf chain looking for an mbuf with 902 * some data. If the first mbuf found has enough data 903 * and it is correctly aligned return it. 904 */ 905 while (left == 0) { 906 havebuf = m1; 907 *mdp = m1 = m1->m_next; 908 if (m1 == NULL) 909 return (EBADRPC); 910 src = mtod(m1, void *); 911 left = m1->m_len; 912 /* 913 * If we start a new mbuf and it is big enough 914 * and correctly aligned just return it, don't 915 * do any pull up. 916 */ 917 if (left >= siz && nfsm_aligned(src)) { 918 *cp2 = src; 919 *dposp = src + siz; 920 return (0); 921 } 922 } 923 if ((m1->m_flags & M_EXT) != 0) { 924 if (havebuf && M_TRAILINGSPACE(havebuf) >= siz && 925 nfsm_aligned(mtod(havebuf, char *) + havebuf->m_len)) { 926 /* 927 * If the first mbuf with data has external data 928 * and there is a previous mbuf with some trailing 929 * space, use it to move the data into. 930 */ 931 m2 = m1; 932 *mdp = m1 = havebuf; 933 *cp2 = mtod(m1, char *) + m1->m_len; 934 } else if (havebuf) { 935 /* 936 * If the first mbuf has a external data 937 * and there is no previous empty mbuf 938 * allocate a new mbuf and move the external 939 * data to the new mbuf. Also make the first 940 * mbuf look empty. 941 */ 942 m2 = m1; 943 *mdp = m1 = m_get(M_WAIT, MT_DATA); 944 MCLAIM(m1, m2->m_owner); 945 if ((m2->m_flags & M_PKTHDR) != 0) { 946 /* XXX MOVE */ 947 M_COPY_PKTHDR(m1, m2); 948 m_tag_delete_chain(m2, NULL); 949 m2->m_flags &= ~M_PKTHDR; 950 } 951 if (havebuf) { 952 havebuf->m_next = m1; 953 } 954 m1->m_next = m2; 955 MRESETDATA(m1); 956 m1->m_len = 0; 957 m2->m_data = src; 958 m2->m_len = left; 959 *cp2 = mtod(m1, char *); 960 } else { 961 struct mbuf **nextp = &m1->m_next; 962 963 m1->m_len -= left; 964 do { 965 m2 = m_get(M_WAIT, MT_DATA); 966 MCLAIM(m2, m1->m_owner); 967 if (left >= MINCLSIZE) { 968 MCLGET(m2, M_WAIT); 969 } 970 m2->m_next = *nextp; 971 *nextp = m2; 972 nextp = &m2->m_next; 973 len = (m2->m_flags & M_EXT) != 0 ? 974 MCLBYTES : MLEN; 975 if (len > left) { 976 len = left; 977 } 978 memcpy(mtod(m2, char *), src, len); 979 m2->m_len = len; 980 src += len; 981 left -= len; 982 } while (left > 0); 983 *mdp = m1 = m1->m_next; 984 m2 = m1->m_next; 985 *cp2 = mtod(m1, char *); 986 } 987 } else { 988 /* 989 * If the first mbuf has no external data 990 * move the data to the front of the mbuf. 991 */ 992 MRESETDATA(m1); 993 dst = mtod(m1, char *); 994 if (dst != src) { 995 memmove(dst, src, left); 996 } 997 m1->m_len = left; 998 m2 = m1->m_next; 999 *cp2 = m1->m_data; 1000 } 1001 *dposp = *cp2 + siz; 1002 /* 1003 * Loop through mbufs pulling data up into first mbuf until 1004 * the first mbuf is full or there is no more data to 1005 * pullup. 1006 */ 1007 dst = mtod(m1, char *) + m1->m_len; 1008 while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) { 1009 if ((len = min(len, m2->m_len)) != 0) { 1010 memcpy(dst, mtod(m2, char *), len); 1011 } 1012 m1->m_len += len; 1013 dst += len; 1014 m2->m_data += len; 1015 m2->m_len -= len; 1016 m2 = m2->m_next; 1017 } 1018 if (m1->m_len < siz) 1019 return (EBADRPC); 1020 return (0); 1021 } 1022 1023 /* 1024 * Advance the position in the mbuf chain. 1025 */ 1026 int 1027 nfs_adv(struct mbuf **mdp, char **dposp, int offs, int left) 1028 { 1029 struct mbuf *m; 1030 int s; 1031 1032 m = *mdp; 1033 s = left; 1034 while (s < offs) { 1035 offs -= s; 1036 m = m->m_next; 1037 if (m == NULL) 1038 return (EBADRPC); 1039 s = m->m_len; 1040 } 1041 *mdp = m; 1042 *dposp = mtod(m, char *) + offs; 1043 return (0); 1044 } 1045 1046 /* 1047 * Copy a string into mbufs for the hard cases... 1048 */ 1049 int 1050 nfsm_strtmbuf(struct mbuf **mb, char **bpos, const char *cp, long siz) 1051 { 1052 struct mbuf *m1 = NULL, *m2; 1053 long left, xfer, len, tlen; 1054 u_int32_t *tl; 1055 int putsize; 1056 1057 putsize = 1; 1058 m2 = *mb; 1059 left = M_TRAILINGSPACE(m2); 1060 if (left > 0) { 1061 tl = ((u_int32_t *)(*bpos)); 1062 *tl++ = txdr_unsigned(siz); 1063 putsize = 0; 1064 left -= NFSX_UNSIGNED; 1065 m2->m_len += NFSX_UNSIGNED; 1066 if (left > 0) { 1067 memcpy((void *) tl, cp, left); 1068 siz -= left; 1069 cp += left; 1070 m2->m_len += left; 1071 left = 0; 1072 } 1073 } 1074 /* Loop around adding mbufs */ 1075 while (siz > 0) { 1076 m1 = m_get(M_WAIT, MT_DATA); 1077 MCLAIM(m1, &nfs_mowner); 1078 if (siz > MLEN) 1079 m_clget(m1, M_WAIT); 1080 m1->m_len = NFSMSIZ(m1); 1081 m2->m_next = m1; 1082 m2 = m1; 1083 tl = mtod(m1, u_int32_t *); 1084 tlen = 0; 1085 if (putsize) { 1086 *tl++ = txdr_unsigned(siz); 1087 m1->m_len -= NFSX_UNSIGNED; 1088 tlen = NFSX_UNSIGNED; 1089 putsize = 0; 1090 } 1091 if (siz < m1->m_len) { 1092 len = nfsm_rndup(siz); 1093 xfer = siz; 1094 if (xfer < len) 1095 *(tl+(xfer>>2)) = 0; 1096 } else { 1097 xfer = len = m1->m_len; 1098 } 1099 memcpy((void *) tl, cp, xfer); 1100 m1->m_len = len+tlen; 1101 siz -= xfer; 1102 cp += xfer; 1103 } 1104 *mb = m1; 1105 *bpos = mtod(m1, char *) + m1->m_len; 1106 return (0); 1107 } 1108 1109 /* 1110 * Directory caching routines. They work as follows: 1111 * - a cache is maintained per VDIR nfsnode. 1112 * - for each offset cookie that is exported to userspace, and can 1113 * thus be thrown back at us as an offset to VOP_READDIR, store 1114 * information in the cache. 1115 * - cached are: 1116 * - cookie itself 1117 * - blocknumber (essentially just a search key in the buffer cache) 1118 * - entry number in block. 1119 * - offset cookie of block in which this entry is stored 1120 * - 32 bit cookie if NFSMNT_XLATECOOKIE is used. 1121 * - entries are looked up in a hash table 1122 * - also maintained is an LRU list of entries, used to determine 1123 * which ones to delete if the cache grows too large. 1124 * - if 32 <-> 64 translation mode is requested for a filesystem, 1125 * the cache also functions as a translation table 1126 * - in the translation case, invalidating the cache does not mean 1127 * flushing it, but just marking entries as invalid, except for 1128 * the <64bit cookie, 32bitcookie> pair which is still valid, to 1129 * still be able to use the cache as a translation table. 1130 * - 32 bit cookies are uniquely created by combining the hash table 1131 * entry value, and one generation count per hash table entry, 1132 * incremented each time an entry is appended to the chain. 1133 * - the cache is invalidated each time a direcory is modified 1134 * - sanity checks are also done; if an entry in a block turns 1135 * out not to have a matching cookie, the cache is invalidated 1136 * and a new block starting from the wanted offset is fetched from 1137 * the server. 1138 * - directory entries as read from the server are extended to contain 1139 * the 64bit and, optionally, the 32bit cookies, for sanity checking 1140 * the cache and exporting them to userspace through the cookie 1141 * argument to VOP_READDIR. 1142 */ 1143 1144 u_long 1145 nfs_dirhash(off_t off) 1146 { 1147 int i; 1148 char *cp = (char *)&off; 1149 u_long sum = 0L; 1150 1151 for (i = 0 ; i < sizeof (off); i++) 1152 sum += *cp++; 1153 1154 return sum; 1155 } 1156 1157 #define _NFSDC_MTX(np) (&NFSTOV(np)->v_interlock) 1158 #define NFSDC_LOCK(np) mutex_enter(_NFSDC_MTX(np)) 1159 #define NFSDC_UNLOCK(np) mutex_exit(_NFSDC_MTX(np)) 1160 #define NFSDC_ASSERT_LOCKED(np) KASSERT(mutex_owned(_NFSDC_MTX(np))) 1161 1162 void 1163 nfs_initdircache(struct vnode *vp) 1164 { 1165 struct nfsnode *np = VTONFS(vp); 1166 struct nfsdirhashhead *dircache; 1167 1168 dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, true, 1169 &nfsdirhashmask); 1170 1171 NFSDC_LOCK(np); 1172 if (np->n_dircache == NULL) { 1173 np->n_dircachesize = 0; 1174 np->n_dircache = dircache; 1175 dircache = NULL; 1176 TAILQ_INIT(&np->n_dirchain); 1177 } 1178 NFSDC_UNLOCK(np); 1179 if (dircache) 1180 hashdone(dircache, HASH_LIST, nfsdirhashmask); 1181 } 1182 1183 void 1184 nfs_initdirxlatecookie(struct vnode *vp) 1185 { 1186 struct nfsnode *np = VTONFS(vp); 1187 unsigned *dirgens; 1188 1189 KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE); 1190 1191 dirgens = kmem_zalloc(NFS_DIRHASHSIZ * sizeof(unsigned), KM_SLEEP); 1192 NFSDC_LOCK(np); 1193 if (np->n_dirgens == NULL) { 1194 np->n_dirgens = dirgens; 1195 dirgens = NULL; 1196 } 1197 NFSDC_UNLOCK(np); 1198 if (dirgens) 1199 kmem_free(dirgens, NFS_DIRHASHSIZ * sizeof(unsigned)); 1200 } 1201 1202 static const struct nfsdircache dzero; 1203 1204 static void nfs_unlinkdircache(struct nfsnode *np, struct nfsdircache *); 1205 static void nfs_putdircache_unlocked(struct nfsnode *, 1206 struct nfsdircache *); 1207 1208 static void 1209 nfs_unlinkdircache(struct nfsnode *np, struct nfsdircache *ndp) 1210 { 1211 1212 NFSDC_ASSERT_LOCKED(np); 1213 KASSERT(ndp != &dzero); 1214 1215 if (LIST_NEXT(ndp, dc_hash) == (void *)-1) 1216 return; 1217 1218 TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain); 1219 LIST_REMOVE(ndp, dc_hash); 1220 LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */ 1221 1222 nfs_putdircache_unlocked(np, ndp); 1223 } 1224 1225 void 1226 nfs_putdircache(struct nfsnode *np, struct nfsdircache *ndp) 1227 { 1228 int ref; 1229 1230 if (ndp == &dzero) 1231 return; 1232 1233 KASSERT(ndp->dc_refcnt > 0); 1234 NFSDC_LOCK(np); 1235 ref = --ndp->dc_refcnt; 1236 NFSDC_UNLOCK(np); 1237 1238 if (ref == 0) 1239 kmem_free(ndp, sizeof(*ndp)); 1240 } 1241 1242 static void 1243 nfs_putdircache_unlocked(struct nfsnode *np, struct nfsdircache *ndp) 1244 { 1245 int ref; 1246 1247 NFSDC_ASSERT_LOCKED(np); 1248 1249 if (ndp == &dzero) 1250 return; 1251 1252 KASSERT(ndp->dc_refcnt > 0); 1253 ref = --ndp->dc_refcnt; 1254 if (ref == 0) 1255 kmem_free(ndp, sizeof(*ndp)); 1256 } 1257 1258 struct nfsdircache * 1259 nfs_searchdircache(struct vnode *vp, off_t off, int do32, int *hashent) 1260 { 1261 struct nfsdirhashhead *ndhp; 1262 struct nfsdircache *ndp = NULL; 1263 struct nfsnode *np = VTONFS(vp); 1264 unsigned ent; 1265 1266 /* 1267 * Zero is always a valid cookie. 1268 */ 1269 if (off == 0) 1270 /* XXXUNCONST */ 1271 return (struct nfsdircache *)__UNCONST(&dzero); 1272 1273 if (!np->n_dircache) 1274 return NULL; 1275 1276 /* 1277 * We use a 32bit cookie as search key, directly reconstruct 1278 * the hashentry. Else use the hashfunction. 1279 */ 1280 if (do32) { 1281 ent = (u_int32_t)off >> 24; 1282 if (ent >= NFS_DIRHASHSIZ) 1283 return NULL; 1284 ndhp = &np->n_dircache[ent]; 1285 } else { 1286 ndhp = NFSDIRHASH(np, off); 1287 } 1288 1289 if (hashent) 1290 *hashent = (int)(ndhp - np->n_dircache); 1291 1292 NFSDC_LOCK(np); 1293 if (do32) { 1294 LIST_FOREACH(ndp, ndhp, dc_hash) { 1295 if (ndp->dc_cookie32 == (u_int32_t)off) { 1296 /* 1297 * An invalidated entry will become the 1298 * start of a new block fetched from 1299 * the server. 1300 */ 1301 if (ndp->dc_flags & NFSDC_INVALID) { 1302 ndp->dc_blkcookie = ndp->dc_cookie; 1303 ndp->dc_entry = 0; 1304 ndp->dc_flags &= ~NFSDC_INVALID; 1305 } 1306 break; 1307 } 1308 } 1309 } else { 1310 LIST_FOREACH(ndp, ndhp, dc_hash) { 1311 if (ndp->dc_cookie == off) 1312 break; 1313 } 1314 } 1315 if (ndp != NULL) 1316 ndp->dc_refcnt++; 1317 NFSDC_UNLOCK(np); 1318 return ndp; 1319 } 1320 1321 1322 struct nfsdircache * 1323 nfs_enterdircache(struct vnode *vp, off_t off, off_t blkoff, int en, 1324 daddr_t blkno) 1325 { 1326 struct nfsnode *np = VTONFS(vp); 1327 struct nfsdirhashhead *ndhp; 1328 struct nfsdircache *ndp = NULL; 1329 struct nfsdircache *newndp = NULL; 1330 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1331 int hashent = 0, gen, overwrite; /* XXX: GCC */ 1332 1333 /* 1334 * XXX refuse entries for offset 0. amd(8) erroneously sets 1335 * cookie 0 for the '.' entry, making this necessary. This 1336 * isn't so bad, as 0 is a special case anyway. 1337 */ 1338 if (off == 0) 1339 /* XXXUNCONST */ 1340 return (struct nfsdircache *)__UNCONST(&dzero); 1341 1342 if (!np->n_dircache) 1343 /* 1344 * XXX would like to do this in nfs_nget but vtype 1345 * isn't known at that time. 1346 */ 1347 nfs_initdircache(vp); 1348 1349 if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens) 1350 nfs_initdirxlatecookie(vp); 1351 1352 retry: 1353 ndp = nfs_searchdircache(vp, off, 0, &hashent); 1354 1355 NFSDC_LOCK(np); 1356 if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) { 1357 /* 1358 * Overwriting an old entry. Check if it's the same. 1359 * If so, just return. If not, remove the old entry. 1360 */ 1361 if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en) 1362 goto done; 1363 nfs_unlinkdircache(np, ndp); 1364 nfs_putdircache_unlocked(np, ndp); 1365 ndp = NULL; 1366 } 1367 1368 ndhp = &np->n_dircache[hashent]; 1369 1370 if (!ndp) { 1371 if (newndp == NULL) { 1372 NFSDC_UNLOCK(np); 1373 newndp = kmem_alloc(sizeof(*newndp), KM_SLEEP); 1374 newndp->dc_refcnt = 1; 1375 LIST_NEXT(newndp, dc_hash) = (void *)-1; 1376 goto retry; 1377 } 1378 ndp = newndp; 1379 newndp = NULL; 1380 overwrite = 0; 1381 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 1382 /* 1383 * We're allocating a new entry, so bump the 1384 * generation number. 1385 */ 1386 KASSERT(np->n_dirgens); 1387 gen = ++np->n_dirgens[hashent]; 1388 if (gen == 0) { 1389 np->n_dirgens[hashent]++; 1390 gen++; 1391 } 1392 ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff); 1393 } 1394 } else 1395 overwrite = 1; 1396 1397 ndp->dc_cookie = off; 1398 ndp->dc_blkcookie = blkoff; 1399 ndp->dc_entry = en; 1400 ndp->dc_flags = 0; 1401 1402 if (overwrite) 1403 goto done; 1404 1405 /* 1406 * If the maximum directory cookie cache size has been reached 1407 * for this node, take one off the front. The idea is that 1408 * directories are typically read front-to-back once, so that 1409 * the oldest entries can be thrown away without much performance 1410 * loss. 1411 */ 1412 if (np->n_dircachesize == NFS_MAXDIRCACHE) { 1413 nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain)); 1414 } else 1415 np->n_dircachesize++; 1416 1417 KASSERT(ndp->dc_refcnt == 1); 1418 LIST_INSERT_HEAD(ndhp, ndp, dc_hash); 1419 TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain); 1420 ndp->dc_refcnt++; 1421 done: 1422 KASSERT(ndp->dc_refcnt > 0); 1423 NFSDC_UNLOCK(np); 1424 if (newndp) 1425 nfs_putdircache(np, newndp); 1426 return ndp; 1427 } 1428 1429 void 1430 nfs_invaldircache(struct vnode *vp, int flags) 1431 { 1432 struct nfsnode *np = VTONFS(vp); 1433 struct nfsdircache *ndp = NULL; 1434 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1435 const bool forcefree = flags & NFS_INVALDIRCACHE_FORCE; 1436 1437 #ifdef DIAGNOSTIC 1438 if (vp->v_type != VDIR) 1439 panic("nfs: invaldircache: not dir"); 1440 #endif 1441 1442 if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0) 1443 np->n_flag &= ~NEOFVALID; 1444 1445 if (!np->n_dircache) 1446 return; 1447 1448 NFSDC_LOCK(np); 1449 if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) { 1450 while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) { 1451 KASSERT(!forcefree || ndp->dc_refcnt == 1); 1452 nfs_unlinkdircache(np, ndp); 1453 } 1454 np->n_dircachesize = 0; 1455 if (forcefree && np->n_dirgens) { 1456 kmem_free(np->n_dirgens, 1457 NFS_DIRHASHSIZ * sizeof(unsigned)); 1458 np->n_dirgens = NULL; 1459 } 1460 } else { 1461 TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain) 1462 ndp->dc_flags |= NFSDC_INVALID; 1463 } 1464 1465 NFSDC_UNLOCK(np); 1466 } 1467 1468 /* 1469 * Called once before VFS init to initialize shared and 1470 * server-specific data structures. 1471 */ 1472 static int 1473 nfs_init0(void) 1474 { 1475 1476 nfsrtt.pos = 0; 1477 rpc_vers = txdr_unsigned(RPC_VER2); 1478 rpc_call = txdr_unsigned(RPC_CALL); 1479 rpc_reply = txdr_unsigned(RPC_REPLY); 1480 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED); 1481 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED); 1482 rpc_mismatch = txdr_unsigned(RPC_MISMATCH); 1483 rpc_autherr = txdr_unsigned(RPC_AUTHERR); 1484 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX); 1485 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4); 1486 nfs_prog = txdr_unsigned(NFS_PROG); 1487 nfs_true = txdr_unsigned(true); 1488 nfs_false = txdr_unsigned(false); 1489 nfs_xdrneg1 = txdr_unsigned(-1); 1490 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; 1491 if (nfs_ticks < 1) 1492 nfs_ticks = 1; 1493 nfs_xid = arc4random(); 1494 nfsdreq_init(); 1495 1496 /* 1497 * Initialize reply list and start timer 1498 */ 1499 TAILQ_INIT(&nfs_reqq); 1500 nfs_timer_init(); 1501 MOWNER_ATTACH(&nfs_mowner); 1502 1503 #ifdef NFS 1504 /* Initialize the kqueue structures */ 1505 nfs_kqinit(); 1506 /* Initialize the iod structures */ 1507 nfs_iodinit(); 1508 #endif 1509 1510 return 0; 1511 } 1512 1513 /* 1514 * This is disgusting, but it must support both modular and monolothic 1515 * configurations. For monolithic builds NFSSERVER may not imply NFS. 1516 * 1517 * Yuck. 1518 */ 1519 void 1520 nfs_init(void) 1521 { 1522 static ONCE_DECL(nfs_init_once); 1523 1524 RUN_ONCE(&nfs_init_once, nfs_init0); 1525 } 1526 1527 void 1528 nfs_fini(void) 1529 { 1530 1531 #ifdef NFS 1532 nfs_kqfini(); 1533 nfs_iodfini(); 1534 #endif 1535 nfsdreq_fini(); 1536 nfs_timer_fini(); 1537 MOWNER_DETACH(&nfs_mowner); 1538 } 1539 1540 #ifdef NFS 1541 /* 1542 * Called once at VFS init to initialize client-specific data structures. 1543 */ 1544 void 1545 nfs_vfs_init(void) 1546 { 1547 1548 /* Initialize NFS server / client shared data. */ 1549 nfs_init(); 1550 nfs_node_init(); 1551 1552 nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4); 1553 } 1554 1555 void 1556 nfs_vfs_done(void) 1557 { 1558 1559 nfs_node_done(); 1560 } 1561 1562 /* 1563 * Attribute cache routines. 1564 * nfs_loadattrcache() - loads or updates the cache contents from attributes 1565 * that are on the mbuf list 1566 * nfs_getattrcache() - returns valid attributes if found in cache, returns 1567 * error otherwise 1568 */ 1569 1570 /* 1571 * Load the attribute cache (that lives in the nfsnode entry) with 1572 * the values on the mbuf list and 1573 * Iff vap not NULL 1574 * copy the attributes to *vaper 1575 */ 1576 int 1577 nfsm_loadattrcache(struct vnode **vpp, struct mbuf **mdp, char **dposp, struct vattr *vaper, int flags) 1578 { 1579 int32_t t1; 1580 char *cp2; 1581 int error = 0; 1582 struct mbuf *md; 1583 int v3 = NFS_ISV3(*vpp); 1584 1585 md = *mdp; 1586 t1 = (mtod(md, char *) + md->m_len) - *dposp; 1587 error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2); 1588 if (error) 1589 return (error); 1590 return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags); 1591 } 1592 1593 int 1594 nfs_loadattrcache(struct vnode **vpp, struct nfs_fattr *fp, struct vattr *vaper, int flags) 1595 { 1596 struct vnode *vp = *vpp; 1597 struct vattr *vap; 1598 int v3 = NFS_ISV3(vp); 1599 enum vtype vtyp; 1600 u_short vmode; 1601 struct timespec mtime; 1602 struct timespec ctime; 1603 int32_t rdev; 1604 struct nfsnode *np; 1605 extern int (**spec_nfsv2nodeop_p)(void *); 1606 uid_t uid; 1607 gid_t gid; 1608 1609 if (v3) { 1610 vtyp = nfsv3tov_type(fp->fa_type); 1611 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1612 rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1), 1613 fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2)); 1614 fxdr_nfsv3time(&fp->fa3_mtime, &mtime); 1615 fxdr_nfsv3time(&fp->fa3_ctime, &ctime); 1616 } else { 1617 vtyp = nfsv2tov_type(fp->fa_type); 1618 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1619 if (vtyp == VNON || vtyp == VREG) 1620 vtyp = IFTOVT(vmode); 1621 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); 1622 fxdr_nfsv2time(&fp->fa2_mtime, &mtime); 1623 ctime.tv_sec = fxdr_unsigned(u_int32_t, 1624 fp->fa2_ctime.nfsv2_sec); 1625 ctime.tv_nsec = 0; 1626 1627 /* 1628 * Really ugly NFSv2 kludge. 1629 */ 1630 if (vtyp == VCHR && rdev == 0xffffffff) 1631 vtyp = VFIFO; 1632 } 1633 1634 vmode &= ALLPERMS; 1635 1636 /* 1637 * If v_type == VNON it is a new node, so fill in the v_type, 1638 * n_mtime fields. Check to see if it represents a special 1639 * device, and if so, check for a possible alias. Once the 1640 * correct vnode has been obtained, fill in the rest of the 1641 * information. 1642 */ 1643 np = VTONFS(vp); 1644 if (vp->v_type == VNON) { 1645 vp->v_type = vtyp; 1646 if (vp->v_type == VFIFO) { 1647 extern int (**fifo_nfsv2nodeop_p)(void *); 1648 vp->v_op = fifo_nfsv2nodeop_p; 1649 } else if (vp->v_type == VREG) { 1650 mutex_init(&np->n_commitlock, MUTEX_DEFAULT, IPL_NONE); 1651 } else if (vp->v_type == VCHR || vp->v_type == VBLK) { 1652 vp->v_op = spec_nfsv2nodeop_p; 1653 spec_node_init(vp, (dev_t)rdev); 1654 } 1655 np->n_mtime = mtime; 1656 } 1657 uid = fxdr_unsigned(uid_t, fp->fa_uid); 1658 gid = fxdr_unsigned(gid_t, fp->fa_gid); 1659 vap = np->n_vattr; 1660 1661 /* 1662 * Invalidate access cache if uid, gid, mode or ctime changed. 1663 */ 1664 if (np->n_accstamp != -1 && 1665 (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode 1666 || timespeccmp(&ctime, &vap->va_ctime, !=))) 1667 np->n_accstamp = -1; 1668 1669 vap->va_type = vtyp; 1670 vap->va_mode = vmode; 1671 vap->va_rdev = (dev_t)rdev; 1672 vap->va_mtime = mtime; 1673 vap->va_ctime = ctime; 1674 vap->va_birthtime.tv_sec = VNOVAL; 1675 vap->va_birthtime.tv_nsec = VNOVAL; 1676 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 1677 switch (vtyp) { 1678 case VDIR: 1679 vap->va_blocksize = NFS_DIRFRAGSIZ; 1680 break; 1681 case VBLK: 1682 vap->va_blocksize = BLKDEV_IOSIZE; 1683 break; 1684 case VCHR: 1685 vap->va_blocksize = MAXBSIZE; 1686 break; 1687 default: 1688 vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize : 1689 fxdr_unsigned(int32_t, fp->fa2_blocksize); 1690 break; 1691 } 1692 if (v3) { 1693 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1694 vap->va_uid = uid; 1695 vap->va_gid = gid; 1696 vap->va_size = fxdr_hyper(&fp->fa3_size); 1697 vap->va_bytes = fxdr_hyper(&fp->fa3_used); 1698 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid); 1699 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); 1700 vap->va_flags = 0; 1701 vap->va_filerev = 0; 1702 } else { 1703 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1704 vap->va_uid = uid; 1705 vap->va_gid = gid; 1706 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); 1707 vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks) 1708 * NFS_FABLKSIZE; 1709 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); 1710 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); 1711 vap->va_flags = 0; 1712 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec); 1713 vap->va_filerev = 0; 1714 } 1715 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 1716 return EFBIG; 1717 } 1718 if (vap->va_size != np->n_size) { 1719 if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) { 1720 vap->va_size = np->n_size; 1721 } else { 1722 np->n_size = vap->va_size; 1723 if (vap->va_type == VREG) { 1724 /* 1725 * we can't free pages if NAC_NOTRUNC because 1726 * the pages can be owned by ourselves. 1727 */ 1728 if (flags & NAC_NOTRUNC) { 1729 np->n_flag |= NTRUNCDELAYED; 1730 } else { 1731 genfs_node_wrlock(vp); 1732 mutex_enter(&vp->v_interlock); 1733 (void)VOP_PUTPAGES(vp, 0, 1734 0, PGO_SYNCIO | PGO_CLEANIT | 1735 PGO_FREE | PGO_ALLPAGES); 1736 uvm_vnp_setsize(vp, np->n_size); 1737 genfs_node_unlock(vp); 1738 } 1739 } 1740 } 1741 } 1742 np->n_attrstamp = time_second; 1743 if (vaper != NULL) { 1744 memcpy((void *)vaper, (void *)vap, sizeof(*vap)); 1745 if (np->n_flag & NCHG) { 1746 if (np->n_flag & NACC) 1747 vaper->va_atime = np->n_atim; 1748 if (np->n_flag & NUPD) 1749 vaper->va_mtime = np->n_mtim; 1750 } 1751 } 1752 return (0); 1753 } 1754 1755 /* 1756 * Check the time stamp 1757 * If the cache is valid, copy contents to *vap and return 0 1758 * otherwise return an error 1759 */ 1760 int 1761 nfs_getattrcache(struct vnode *vp, struct vattr *vaper) 1762 { 1763 struct nfsnode *np = VTONFS(vp); 1764 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1765 struct vattr *vap; 1766 1767 if (np->n_attrstamp == 0 || 1768 (time_second - np->n_attrstamp) >= nfs_attrtimeo(nmp, np)) { 1769 nfsstats.attrcache_misses++; 1770 return (ENOENT); 1771 } 1772 nfsstats.attrcache_hits++; 1773 vap = np->n_vattr; 1774 if (vap->va_size != np->n_size) { 1775 if (vap->va_type == VREG) { 1776 if ((np->n_flag & NMODIFIED) != 0 && 1777 vap->va_size < np->n_size) { 1778 vap->va_size = np->n_size; 1779 } else { 1780 np->n_size = vap->va_size; 1781 } 1782 genfs_node_wrlock(vp); 1783 uvm_vnp_setsize(vp, np->n_size); 1784 genfs_node_unlock(vp); 1785 } else 1786 np->n_size = vap->va_size; 1787 } 1788 memcpy((void *)vaper, (void *)vap, sizeof(struct vattr)); 1789 if (np->n_flag & NCHG) { 1790 if (np->n_flag & NACC) 1791 vaper->va_atime = np->n_atim; 1792 if (np->n_flag & NUPD) 1793 vaper->va_mtime = np->n_mtim; 1794 } 1795 return (0); 1796 } 1797 1798 void 1799 nfs_delayedtruncate(struct vnode *vp) 1800 { 1801 struct nfsnode *np = VTONFS(vp); 1802 1803 if (np->n_flag & NTRUNCDELAYED) { 1804 np->n_flag &= ~NTRUNCDELAYED; 1805 genfs_node_wrlock(vp); 1806 mutex_enter(&vp->v_interlock); 1807 (void)VOP_PUTPAGES(vp, 0, 1808 0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES); 1809 uvm_vnp_setsize(vp, np->n_size); 1810 genfs_node_unlock(vp); 1811 } 1812 } 1813 1814 #define NFS_WCCKLUDGE_TIMEOUT (24 * 60 * 60) /* 1 day */ 1815 #define NFS_WCCKLUDGE(nmp, now) \ 1816 (((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \ 1817 ((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0) 1818 1819 /* 1820 * nfs_check_wccdata: check inaccurate wcc_data 1821 * 1822 * => return non-zero if we shouldn't trust the wcc_data. 1823 * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed". 1824 */ 1825 1826 int 1827 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime, 1828 struct timespec *mtime, bool docheck) 1829 { 1830 int error = 0; 1831 1832 #if !defined(NFS_V2_ONLY) 1833 1834 if (docheck) { 1835 struct vnode *vp = NFSTOV(np); 1836 struct nfsmount *nmp; 1837 long now = time_second; 1838 const struct timespec *omtime = &np->n_vattr->va_mtime; 1839 const struct timespec *octime = &np->n_vattr->va_ctime; 1840 const char *reason = NULL; /* XXX: gcc */ 1841 1842 if (timespeccmp(omtime, mtime, <=)) { 1843 reason = "mtime"; 1844 error = EINVAL; 1845 } 1846 1847 if (vp->v_type == VDIR && timespeccmp(octime, ctime, <=)) { 1848 reason = "ctime"; 1849 error = EINVAL; 1850 } 1851 1852 nmp = VFSTONFS(vp->v_mount); 1853 if (error) { 1854 1855 /* 1856 * despite of the fact that we've updated the file, 1857 * timestamps of the file were not updated as we 1858 * expected. 1859 * it means that the server has incompatible 1860 * semantics of timestamps or (more likely) 1861 * the server time is not precise enough to 1862 * track each modifications. 1863 * in that case, we disable wcc processing. 1864 * 1865 * yes, strictly speaking, we should disable all 1866 * caching. it's a compromise. 1867 */ 1868 1869 mutex_enter(&nmp->nm_lock); 1870 if (!NFS_WCCKLUDGE(nmp, now)) { 1871 printf("%s: inaccurate wcc data (%s) detected," 1872 " disabling wcc" 1873 " (ctime %u.%09u %u.%09u," 1874 " mtime %u.%09u %u.%09u)\n", 1875 vp->v_mount->mnt_stat.f_mntfromname, 1876 reason, 1877 (unsigned int)octime->tv_sec, 1878 (unsigned int)octime->tv_nsec, 1879 (unsigned int)ctime->tv_sec, 1880 (unsigned int)ctime->tv_nsec, 1881 (unsigned int)omtime->tv_sec, 1882 (unsigned int)omtime->tv_nsec, 1883 (unsigned int)mtime->tv_sec, 1884 (unsigned int)mtime->tv_nsec); 1885 } 1886 nmp->nm_iflag |= NFSMNT_WCCKLUDGE; 1887 nmp->nm_wcckludgetime = now; 1888 mutex_exit(&nmp->nm_lock); 1889 } else if (NFS_WCCKLUDGE(nmp, now)) { 1890 error = EPERM; /* XXX */ 1891 } else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1892 mutex_enter(&nmp->nm_lock); 1893 if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1894 printf("%s: re-enabling wcc\n", 1895 vp->v_mount->mnt_stat.f_mntfromname); 1896 nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE; 1897 } 1898 mutex_exit(&nmp->nm_lock); 1899 } 1900 } 1901 1902 #endif /* !defined(NFS_V2_ONLY) */ 1903 1904 return error; 1905 } 1906 1907 /* 1908 * Heuristic to see if the server XDR encodes directory cookies or not. 1909 * it is not supposed to, but a lot of servers may do this. Also, since 1910 * most/all servers will implement V2 as well, it is expected that they 1911 * may return just 32 bits worth of cookie information, so we need to 1912 * find out in which 32 bits this information is available. We do this 1913 * to avoid trouble with emulated binaries that can't handle 64 bit 1914 * directory offsets. 1915 */ 1916 1917 void 1918 nfs_cookieheuristic(struct vnode *vp, int *flagp, struct lwp *l, kauth_cred_t cred) 1919 { 1920 struct uio auio; 1921 struct iovec aiov; 1922 char *tbuf, *cp; 1923 struct dirent *dp; 1924 off_t *cookies = NULL, *cop; 1925 int error, eof, nc, len; 1926 1927 tbuf = malloc(NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK); 1928 1929 aiov.iov_base = tbuf; 1930 aiov.iov_len = NFS_DIRFRAGSIZ; 1931 auio.uio_iov = &aiov; 1932 auio.uio_iovcnt = 1; 1933 auio.uio_rw = UIO_READ; 1934 auio.uio_resid = NFS_DIRFRAGSIZ; 1935 auio.uio_offset = 0; 1936 UIO_SETUP_SYSSPACE(&auio); 1937 1938 error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc); 1939 1940 len = NFS_DIRFRAGSIZ - auio.uio_resid; 1941 if (error || len == 0) { 1942 free(tbuf, M_TEMP); 1943 if (cookies) 1944 free(cookies, M_TEMP); 1945 return; 1946 } 1947 1948 /* 1949 * Find the first valid entry and look at its offset cookie. 1950 */ 1951 1952 cp = tbuf; 1953 for (cop = cookies; len > 0; len -= dp->d_reclen) { 1954 dp = (struct dirent *)cp; 1955 if (dp->d_fileno != 0 && len >= dp->d_reclen) { 1956 if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) { 1957 *flagp |= NFSMNT_SWAPCOOKIE; 1958 nfs_invaldircache(vp, 0); 1959 nfs_vinvalbuf(vp, 0, cred, l, 1); 1960 } 1961 break; 1962 } 1963 cop++; 1964 cp += dp->d_reclen; 1965 } 1966 1967 free(tbuf, M_TEMP); 1968 free(cookies, M_TEMP); 1969 } 1970 #endif /* NFS */ 1971 1972 /* 1973 * A fiddled version of m_adj() that ensures null fill to a 32-bit 1974 * boundary and only trims off the back end 1975 * 1976 * 1. trim off 'len' bytes as m_adj(mp, -len). 1977 * 2. add zero-padding 'nul' bytes at the end of the mbuf chain. 1978 */ 1979 void 1980 nfs_zeropad(struct mbuf *mp, int len, int nul) 1981 { 1982 struct mbuf *m; 1983 int count; 1984 1985 /* 1986 * Trim from tail. Scan the mbuf chain, 1987 * calculating its length and finding the last mbuf. 1988 * If the adjustment only affects this mbuf, then just 1989 * adjust and return. Otherwise, rescan and truncate 1990 * after the remaining size. 1991 */ 1992 count = 0; 1993 m = mp; 1994 for (;;) { 1995 count += m->m_len; 1996 if (m->m_next == NULL) 1997 break; 1998 m = m->m_next; 1999 } 2000 2001 KDASSERT(count >= len); 2002 2003 if (m->m_len >= len) { 2004 m->m_len -= len; 2005 } else { 2006 count -= len; 2007 /* 2008 * Correct length for chain is "count". 2009 * Find the mbuf with last data, adjust its length, 2010 * and toss data from remaining mbufs on chain. 2011 */ 2012 for (m = mp; m; m = m->m_next) { 2013 if (m->m_len >= count) { 2014 m->m_len = count; 2015 break; 2016 } 2017 count -= m->m_len; 2018 } 2019 KASSERT(m && m->m_next); 2020 m_freem(m->m_next); 2021 m->m_next = NULL; 2022 } 2023 2024 KDASSERT(m->m_next == NULL); 2025 2026 /* 2027 * zero-padding. 2028 */ 2029 if (nul > 0) { 2030 char *cp; 2031 int i; 2032 2033 if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) { 2034 struct mbuf *n; 2035 2036 KDASSERT(MLEN >= nul); 2037 n = m_get(M_WAIT, MT_DATA); 2038 MCLAIM(n, &nfs_mowner); 2039 n->m_len = nul; 2040 n->m_next = NULL; 2041 m->m_next = n; 2042 cp = mtod(n, void *); 2043 } else { 2044 cp = mtod(m, char *) + m->m_len; 2045 m->m_len += nul; 2046 } 2047 for (i = 0; i < nul; i++) 2048 *cp++ = '\0'; 2049 } 2050 return; 2051 } 2052 2053 /* 2054 * Make these functions instead of macros, so that the kernel text size 2055 * doesn't get too big... 2056 */ 2057 void 2058 nfsm_srvwcc(struct nfsrv_descript *nfsd, int before_ret, struct vattr *before_vap, int after_ret, struct vattr *after_vap, struct mbuf **mbp, char **bposp) 2059 { 2060 struct mbuf *mb = *mbp; 2061 char *bpos = *bposp; 2062 u_int32_t *tl; 2063 2064 if (before_ret) { 2065 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2066 *tl = nfs_false; 2067 } else { 2068 nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED); 2069 *tl++ = nfs_true; 2070 txdr_hyper(before_vap->va_size, tl); 2071 tl += 2; 2072 txdr_nfsv3time(&(before_vap->va_mtime), tl); 2073 tl += 2; 2074 txdr_nfsv3time(&(before_vap->va_ctime), tl); 2075 } 2076 *bposp = bpos; 2077 *mbp = mb; 2078 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp); 2079 } 2080 2081 void 2082 nfsm_srvpostopattr(struct nfsrv_descript *nfsd, int after_ret, struct vattr *after_vap, struct mbuf **mbp, char **bposp) 2083 { 2084 struct mbuf *mb = *mbp; 2085 char *bpos = *bposp; 2086 u_int32_t *tl; 2087 struct nfs_fattr *fp; 2088 2089 if (after_ret) { 2090 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2091 *tl = nfs_false; 2092 } else { 2093 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR); 2094 *tl++ = nfs_true; 2095 fp = (struct nfs_fattr *)tl; 2096 nfsm_srvfattr(nfsd, after_vap, fp); 2097 } 2098 *mbp = mb; 2099 *bposp = bpos; 2100 } 2101 2102 void 2103 nfsm_srvfattr(struct nfsrv_descript *nfsd, struct vattr *vap, struct nfs_fattr *fp) 2104 { 2105 2106 fp->fa_nlink = txdr_unsigned(vap->va_nlink); 2107 fp->fa_uid = txdr_unsigned(vap->va_uid); 2108 fp->fa_gid = txdr_unsigned(vap->va_gid); 2109 if (nfsd->nd_flag & ND_NFSV3) { 2110 fp->fa_type = vtonfsv3_type(vap->va_type); 2111 fp->fa_mode = vtonfsv3_mode(vap->va_mode); 2112 txdr_hyper(vap->va_size, &fp->fa3_size); 2113 txdr_hyper(vap->va_bytes, &fp->fa3_used); 2114 fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev)); 2115 fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev)); 2116 fp->fa3_fsid.nfsuquad[0] = 0; 2117 fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid); 2118 txdr_hyper(vap->va_fileid, &fp->fa3_fileid); 2119 txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime); 2120 txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime); 2121 txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime); 2122 } else { 2123 fp->fa_type = vtonfsv2_type(vap->va_type); 2124 fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 2125 fp->fa2_size = txdr_unsigned(vap->va_size); 2126 fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize); 2127 if (vap->va_type == VFIFO) 2128 fp->fa2_rdev = 0xffffffff; 2129 else 2130 fp->fa2_rdev = txdr_unsigned(vap->va_rdev); 2131 fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE); 2132 fp->fa2_fsid = txdr_unsigned(vap->va_fsid); 2133 fp->fa2_fileid = txdr_unsigned(vap->va_fileid); 2134 txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime); 2135 txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime); 2136 txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime); 2137 } 2138 } 2139 2140 /* 2141 * This function compares two net addresses by family and returns true 2142 * if they are the same host. 2143 * If there is any doubt, return false. 2144 * The AF_INET family is handled as a special case so that address mbufs 2145 * don't need to be saved to store "struct in_addr", which is only 4 bytes. 2146 */ 2147 int 2148 netaddr_match(int family, union nethostaddr *haddr, struct mbuf *nam) 2149 { 2150 struct sockaddr_in *inetaddr; 2151 2152 switch (family) { 2153 case AF_INET: 2154 inetaddr = mtod(nam, struct sockaddr_in *); 2155 if (inetaddr->sin_family == AF_INET && 2156 inetaddr->sin_addr.s_addr == haddr->had_inetaddr) 2157 return (1); 2158 break; 2159 case AF_INET6: 2160 { 2161 struct sockaddr_in6 *sin6_1, *sin6_2; 2162 2163 sin6_1 = mtod(nam, struct sockaddr_in6 *); 2164 sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *); 2165 if (sin6_1->sin6_family == AF_INET6 && 2166 IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr)) 2167 return 1; 2168 } 2169 default: 2170 break; 2171 }; 2172 return (0); 2173 } 2174 2175 /* 2176 * The write verifier has changed (probably due to a server reboot), so all 2177 * PG_NEEDCOMMIT pages will have to be written again. Since they are marked 2178 * as dirty or are being written out just now, all this takes is clearing 2179 * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for 2180 * the mount point. 2181 */ 2182 void 2183 nfs_clearcommit(struct mount *mp) 2184 { 2185 struct vnode *vp; 2186 struct nfsnode *np; 2187 struct vm_page *pg; 2188 struct nfsmount *nmp = VFSTONFS(mp); 2189 2190 rw_enter(&nmp->nm_writeverflock, RW_WRITER); 2191 mutex_enter(&mntvnode_lock); 2192 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2193 KASSERT(vp->v_mount == mp); 2194 if (vp->v_type != VREG) 2195 continue; 2196 mutex_enter(&vp->v_interlock); 2197 if (vp->v_iflag & (VI_XLOCK | VI_CLEAN)) { 2198 mutex_exit(&vp->v_interlock); 2199 continue; 2200 } 2201 np = VTONFS(vp); 2202 np->n_pushlo = np->n_pushhi = np->n_pushedlo = 2203 np->n_pushedhi = 0; 2204 np->n_commitflags &= 2205 ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID); 2206 TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq.queue) { 2207 pg->flags &= ~PG_NEEDCOMMIT; 2208 } 2209 mutex_exit(&vp->v_interlock); 2210 } 2211 mutex_exit(&mntvnode_lock); 2212 mutex_enter(&nmp->nm_lock); 2213 nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF; 2214 mutex_exit(&nmp->nm_lock); 2215 rw_exit(&nmp->nm_writeverflock); 2216 } 2217 2218 void 2219 nfs_merge_commit_ranges(struct vnode *vp) 2220 { 2221 struct nfsnode *np = VTONFS(vp); 2222 2223 KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID); 2224 2225 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2226 np->n_pushedlo = np->n_pushlo; 2227 np->n_pushedhi = np->n_pushhi; 2228 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2229 } else { 2230 if (np->n_pushlo < np->n_pushedlo) 2231 np->n_pushedlo = np->n_pushlo; 2232 if (np->n_pushhi > np->n_pushedhi) 2233 np->n_pushedhi = np->n_pushhi; 2234 } 2235 2236 np->n_pushlo = np->n_pushhi = 0; 2237 np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID; 2238 2239 #ifdef NFS_DEBUG_COMMIT 2240 printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2241 (unsigned)np->n_pushedhi); 2242 #endif 2243 } 2244 2245 int 2246 nfs_in_committed_range(struct vnode *vp, off_t off, off_t len) 2247 { 2248 struct nfsnode *np = VTONFS(vp); 2249 off_t lo, hi; 2250 2251 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2252 return 0; 2253 lo = off; 2254 hi = lo + len; 2255 2256 return (lo >= np->n_pushedlo && hi <= np->n_pushedhi); 2257 } 2258 2259 int 2260 nfs_in_tobecommitted_range(struct vnode *vp, off_t off, off_t len) 2261 { 2262 struct nfsnode *np = VTONFS(vp); 2263 off_t lo, hi; 2264 2265 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2266 return 0; 2267 lo = off; 2268 hi = lo + len; 2269 2270 return (lo >= np->n_pushlo && hi <= np->n_pushhi); 2271 } 2272 2273 void 2274 nfs_add_committed_range(struct vnode *vp, off_t off, off_t len) 2275 { 2276 struct nfsnode *np = VTONFS(vp); 2277 off_t lo, hi; 2278 2279 lo = off; 2280 hi = lo + len; 2281 2282 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2283 np->n_pushedlo = lo; 2284 np->n_pushedhi = hi; 2285 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2286 } else { 2287 if (hi > np->n_pushedhi) 2288 np->n_pushedhi = hi; 2289 if (lo < np->n_pushedlo) 2290 np->n_pushedlo = lo; 2291 } 2292 #ifdef NFS_DEBUG_COMMIT 2293 printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2294 (unsigned)np->n_pushedhi); 2295 #endif 2296 } 2297 2298 void 2299 nfs_del_committed_range(struct vnode *vp, off_t off, off_t len) 2300 { 2301 struct nfsnode *np = VTONFS(vp); 2302 off_t lo, hi; 2303 2304 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2305 return; 2306 2307 lo = off; 2308 hi = lo + len; 2309 2310 if (lo > np->n_pushedhi || hi < np->n_pushedlo) 2311 return; 2312 if (lo <= np->n_pushedlo) 2313 np->n_pushedlo = hi; 2314 else if (hi >= np->n_pushedhi) 2315 np->n_pushedhi = lo; 2316 else { 2317 /* 2318 * XXX There's only one range. If the deleted range 2319 * is in the middle, pick the largest of the 2320 * contiguous ranges that it leaves. 2321 */ 2322 if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi)) 2323 np->n_pushedhi = lo; 2324 else 2325 np->n_pushedlo = hi; 2326 } 2327 #ifdef NFS_DEBUG_COMMIT 2328 printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2329 (unsigned)np->n_pushedhi); 2330 #endif 2331 } 2332 2333 void 2334 nfs_add_tobecommitted_range(struct vnode *vp, off_t off, off_t len) 2335 { 2336 struct nfsnode *np = VTONFS(vp); 2337 off_t lo, hi; 2338 2339 lo = off; 2340 hi = lo + len; 2341 2342 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) { 2343 np->n_pushlo = lo; 2344 np->n_pushhi = hi; 2345 np->n_commitflags |= NFS_COMMIT_PUSH_VALID; 2346 } else { 2347 if (lo < np->n_pushlo) 2348 np->n_pushlo = lo; 2349 if (hi > np->n_pushhi) 2350 np->n_pushhi = hi; 2351 } 2352 #ifdef NFS_DEBUG_COMMIT 2353 printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2354 (unsigned)np->n_pushhi); 2355 #endif 2356 } 2357 2358 void 2359 nfs_del_tobecommitted_range(struct vnode *vp, off_t off, off_t len) 2360 { 2361 struct nfsnode *np = VTONFS(vp); 2362 off_t lo, hi; 2363 2364 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2365 return; 2366 2367 lo = off; 2368 hi = lo + len; 2369 2370 if (lo > np->n_pushhi || hi < np->n_pushlo) 2371 return; 2372 2373 if (lo <= np->n_pushlo) 2374 np->n_pushlo = hi; 2375 else if (hi >= np->n_pushhi) 2376 np->n_pushhi = lo; 2377 else { 2378 /* 2379 * XXX There's only one range. If the deleted range 2380 * is in the middle, pick the largest of the 2381 * contiguous ranges that it leaves. 2382 */ 2383 if ((np->n_pushlo - lo) > (hi - np->n_pushhi)) 2384 np->n_pushhi = lo; 2385 else 2386 np->n_pushlo = hi; 2387 } 2388 #ifdef NFS_DEBUG_COMMIT 2389 printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2390 (unsigned)np->n_pushhi); 2391 #endif 2392 } 2393 2394 /* 2395 * Map errnos to NFS error numbers. For Version 3 also filter out error 2396 * numbers not specified for the associated procedure. 2397 */ 2398 int 2399 nfsrv_errmap(struct nfsrv_descript *nd, int err) 2400 { 2401 const short *defaulterrp, *errp; 2402 2403 if (nd->nd_flag & ND_NFSV3) { 2404 if (nd->nd_procnum <= NFSPROC_COMMIT) { 2405 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum]; 2406 while (*++errp) { 2407 if (*errp == err) 2408 return (err); 2409 else if (*errp > err) 2410 break; 2411 } 2412 return ((int)*defaulterrp); 2413 } else 2414 return (err & 0xffff); 2415 } 2416 if (err <= ELAST) 2417 return ((int)nfsrv_v2errmap[err - 1]); 2418 return (NFSERR_IO); 2419 } 2420 2421 u_int32_t 2422 nfs_getxid(void) 2423 { 2424 u_int32_t newxid; 2425 2426 /* get next xid. skip 0 */ 2427 do { 2428 newxid = atomic_inc_32_nv(&nfs_xid); 2429 } while (__predict_false(newxid == 0)); 2430 2431 return txdr_unsigned(newxid); 2432 } 2433 2434 /* 2435 * assign a new xid for existing request. 2436 * used for NFSERR_JUKEBOX handling. 2437 */ 2438 void 2439 nfs_renewxid(struct nfsreq *req) 2440 { 2441 u_int32_t xid; 2442 int off; 2443 2444 xid = nfs_getxid(); 2445 if (req->r_nmp->nm_sotype == SOCK_STREAM) 2446 off = sizeof(u_int32_t); /* RPC record mark */ 2447 else 2448 off = 0; 2449 2450 m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid); 2451 req->r_xid = xid; 2452 } 2453 2454 #if defined(NFS) 2455 /* 2456 * Set the attribute timeout based on how recently the file has been modified. 2457 */ 2458 2459 time_t 2460 nfs_attrtimeo(struct nfsmount *nmp, struct nfsnode *np) 2461 { 2462 time_t timeo; 2463 2464 if ((nmp->nm_flag & NFSMNT_NOAC) != 0) 2465 return 0; 2466 2467 if (((np)->n_flag & NMODIFIED) != 0) 2468 return NFS_MINATTRTIMO; 2469 2470 timeo = (time_second - np->n_mtime.tv_sec) / 10; 2471 timeo = max(timeo, NFS_MINATTRTIMO); 2472 timeo = min(timeo, NFS_MAXATTRTIMO); 2473 return timeo; 2474 } 2475 #endif /* defined(NFS) */ 2476