1 /* $NetBSD: nfs_subs.c,v 1.212 2008/12/17 20:51:38 cegger 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 1.212 2008/12/17 20:51:38 cegger 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 /* NFS client/server stats. */ 144 struct nfsstats nfsstats; 145 146 /* 147 * Mapping of old NFS Version 2 RPC numbers to generic numbers. 148 */ 149 const int nfsv3_procid[NFS_NPROCS] = { 150 NFSPROC_NULL, 151 NFSPROC_GETATTR, 152 NFSPROC_SETATTR, 153 NFSPROC_NOOP, 154 NFSPROC_LOOKUP, 155 NFSPROC_READLINK, 156 NFSPROC_READ, 157 NFSPROC_NOOP, 158 NFSPROC_WRITE, 159 NFSPROC_CREATE, 160 NFSPROC_REMOVE, 161 NFSPROC_RENAME, 162 NFSPROC_LINK, 163 NFSPROC_SYMLINK, 164 NFSPROC_MKDIR, 165 NFSPROC_RMDIR, 166 NFSPROC_READDIR, 167 NFSPROC_FSSTAT, 168 NFSPROC_NOOP, 169 NFSPROC_NOOP, 170 NFSPROC_NOOP, 171 NFSPROC_NOOP, 172 NFSPROC_NOOP 173 }; 174 175 /* 176 * and the reverse mapping from generic to Version 2 procedure numbers 177 */ 178 const int nfsv2_procid[NFS_NPROCS] = { 179 NFSV2PROC_NULL, 180 NFSV2PROC_GETATTR, 181 NFSV2PROC_SETATTR, 182 NFSV2PROC_LOOKUP, 183 NFSV2PROC_NOOP, 184 NFSV2PROC_READLINK, 185 NFSV2PROC_READ, 186 NFSV2PROC_WRITE, 187 NFSV2PROC_CREATE, 188 NFSV2PROC_MKDIR, 189 NFSV2PROC_SYMLINK, 190 NFSV2PROC_CREATE, 191 NFSV2PROC_REMOVE, 192 NFSV2PROC_RMDIR, 193 NFSV2PROC_RENAME, 194 NFSV2PROC_LINK, 195 NFSV2PROC_READDIR, 196 NFSV2PROC_NOOP, 197 NFSV2PROC_STATFS, 198 NFSV2PROC_NOOP, 199 NFSV2PROC_NOOP, 200 NFSV2PROC_NOOP, 201 NFSV2PROC_NOOP, 202 }; 203 204 /* 205 * Maps errno values to nfs error numbers. 206 * Use NFSERR_IO as the catch all for ones not specifically defined in 207 * RFC 1094. 208 */ 209 static const u_char nfsrv_v2errmap[ELAST] = { 210 NFSERR_PERM, NFSERR_NOENT, NFSERR_IO, NFSERR_IO, NFSERR_IO, 211 NFSERR_NXIO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 212 NFSERR_IO, NFSERR_IO, NFSERR_ACCES, NFSERR_IO, NFSERR_IO, 213 NFSERR_IO, NFSERR_EXIST, NFSERR_IO, NFSERR_NODEV, NFSERR_NOTDIR, 214 NFSERR_ISDIR, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 215 NFSERR_IO, NFSERR_FBIG, NFSERR_NOSPC, NFSERR_IO, NFSERR_ROFS, 216 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 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_NAMETOL, NFSERR_IO, NFSERR_IO, 223 NFSERR_NOTEMPTY, NFSERR_IO, NFSERR_IO, NFSERR_DQUOT, NFSERR_STALE, 224 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 225 NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, NFSERR_IO, 226 NFSERR_IO, NFSERR_IO, 227 }; 228 229 /* 230 * Maps errno values to nfs error numbers. 231 * Although it is not obvious whether or not NFS clients really care if 232 * a returned error value is in the specified list for the procedure, the 233 * safest thing to do is filter them appropriately. For Version 2, the 234 * X/Open XNFS document is the only specification that defines error values 235 * for each RPC (The RFC simply lists all possible error values for all RPCs), 236 * so I have decided to not do this for Version 2. 237 * The first entry is the default error return and the rest are the valid 238 * errors for that RPC in increasing numeric order. 239 */ 240 static const short nfsv3err_null[] = { 241 0, 242 0, 243 }; 244 245 static const short nfsv3err_getattr[] = { 246 NFSERR_IO, 247 NFSERR_IO, 248 NFSERR_STALE, 249 NFSERR_BADHANDLE, 250 NFSERR_SERVERFAULT, 251 0, 252 }; 253 254 static const short nfsv3err_setattr[] = { 255 NFSERR_IO, 256 NFSERR_PERM, 257 NFSERR_IO, 258 NFSERR_ACCES, 259 NFSERR_INVAL, 260 NFSERR_NOSPC, 261 NFSERR_ROFS, 262 NFSERR_DQUOT, 263 NFSERR_STALE, 264 NFSERR_BADHANDLE, 265 NFSERR_NOT_SYNC, 266 NFSERR_SERVERFAULT, 267 0, 268 }; 269 270 static const short nfsv3err_lookup[] = { 271 NFSERR_IO, 272 NFSERR_NOENT, 273 NFSERR_IO, 274 NFSERR_ACCES, 275 NFSERR_NOTDIR, 276 NFSERR_NAMETOL, 277 NFSERR_STALE, 278 NFSERR_BADHANDLE, 279 NFSERR_SERVERFAULT, 280 0, 281 }; 282 283 static const short nfsv3err_access[] = { 284 NFSERR_IO, 285 NFSERR_IO, 286 NFSERR_STALE, 287 NFSERR_BADHANDLE, 288 NFSERR_SERVERFAULT, 289 0, 290 }; 291 292 static const short nfsv3err_readlink[] = { 293 NFSERR_IO, 294 NFSERR_IO, 295 NFSERR_ACCES, 296 NFSERR_INVAL, 297 NFSERR_STALE, 298 NFSERR_BADHANDLE, 299 NFSERR_NOTSUPP, 300 NFSERR_SERVERFAULT, 301 0, 302 }; 303 304 static const short nfsv3err_read[] = { 305 NFSERR_IO, 306 NFSERR_IO, 307 NFSERR_NXIO, 308 NFSERR_ACCES, 309 NFSERR_INVAL, 310 NFSERR_STALE, 311 NFSERR_BADHANDLE, 312 NFSERR_SERVERFAULT, 313 NFSERR_JUKEBOX, 314 0, 315 }; 316 317 static const short nfsv3err_write[] = { 318 NFSERR_IO, 319 NFSERR_IO, 320 NFSERR_ACCES, 321 NFSERR_INVAL, 322 NFSERR_FBIG, 323 NFSERR_NOSPC, 324 NFSERR_ROFS, 325 NFSERR_DQUOT, 326 NFSERR_STALE, 327 NFSERR_BADHANDLE, 328 NFSERR_SERVERFAULT, 329 NFSERR_JUKEBOX, 330 0, 331 }; 332 333 static const short nfsv3err_create[] = { 334 NFSERR_IO, 335 NFSERR_IO, 336 NFSERR_ACCES, 337 NFSERR_EXIST, 338 NFSERR_NOTDIR, 339 NFSERR_NOSPC, 340 NFSERR_ROFS, 341 NFSERR_NAMETOL, 342 NFSERR_DQUOT, 343 NFSERR_STALE, 344 NFSERR_BADHANDLE, 345 NFSERR_NOTSUPP, 346 NFSERR_SERVERFAULT, 347 0, 348 }; 349 350 static const short nfsv3err_mkdir[] = { 351 NFSERR_IO, 352 NFSERR_IO, 353 NFSERR_ACCES, 354 NFSERR_EXIST, 355 NFSERR_NOTDIR, 356 NFSERR_NOSPC, 357 NFSERR_ROFS, 358 NFSERR_NAMETOL, 359 NFSERR_DQUOT, 360 NFSERR_STALE, 361 NFSERR_BADHANDLE, 362 NFSERR_NOTSUPP, 363 NFSERR_SERVERFAULT, 364 0, 365 }; 366 367 static const short nfsv3err_symlink[] = { 368 NFSERR_IO, 369 NFSERR_IO, 370 NFSERR_ACCES, 371 NFSERR_EXIST, 372 NFSERR_NOTDIR, 373 NFSERR_NOSPC, 374 NFSERR_ROFS, 375 NFSERR_NAMETOL, 376 NFSERR_DQUOT, 377 NFSERR_STALE, 378 NFSERR_BADHANDLE, 379 NFSERR_NOTSUPP, 380 NFSERR_SERVERFAULT, 381 0, 382 }; 383 384 static const short nfsv3err_mknod[] = { 385 NFSERR_IO, 386 NFSERR_IO, 387 NFSERR_ACCES, 388 NFSERR_EXIST, 389 NFSERR_NOTDIR, 390 NFSERR_NOSPC, 391 NFSERR_ROFS, 392 NFSERR_NAMETOL, 393 NFSERR_DQUOT, 394 NFSERR_STALE, 395 NFSERR_BADHANDLE, 396 NFSERR_NOTSUPP, 397 NFSERR_SERVERFAULT, 398 NFSERR_BADTYPE, 399 0, 400 }; 401 402 static const short nfsv3err_remove[] = { 403 NFSERR_IO, 404 NFSERR_NOENT, 405 NFSERR_IO, 406 NFSERR_ACCES, 407 NFSERR_NOTDIR, 408 NFSERR_ROFS, 409 NFSERR_NAMETOL, 410 NFSERR_STALE, 411 NFSERR_BADHANDLE, 412 NFSERR_SERVERFAULT, 413 0, 414 }; 415 416 static const short nfsv3err_rmdir[] = { 417 NFSERR_IO, 418 NFSERR_NOENT, 419 NFSERR_IO, 420 NFSERR_ACCES, 421 NFSERR_EXIST, 422 NFSERR_NOTDIR, 423 NFSERR_INVAL, 424 NFSERR_ROFS, 425 NFSERR_NAMETOL, 426 NFSERR_NOTEMPTY, 427 NFSERR_STALE, 428 NFSERR_BADHANDLE, 429 NFSERR_NOTSUPP, 430 NFSERR_SERVERFAULT, 431 0, 432 }; 433 434 static const short nfsv3err_rename[] = { 435 NFSERR_IO, 436 NFSERR_NOENT, 437 NFSERR_IO, 438 NFSERR_ACCES, 439 NFSERR_EXIST, 440 NFSERR_XDEV, 441 NFSERR_NOTDIR, 442 NFSERR_ISDIR, 443 NFSERR_INVAL, 444 NFSERR_NOSPC, 445 NFSERR_ROFS, 446 NFSERR_MLINK, 447 NFSERR_NAMETOL, 448 NFSERR_NOTEMPTY, 449 NFSERR_DQUOT, 450 NFSERR_STALE, 451 NFSERR_BADHANDLE, 452 NFSERR_NOTSUPP, 453 NFSERR_SERVERFAULT, 454 0, 455 }; 456 457 static const short nfsv3err_link[] = { 458 NFSERR_IO, 459 NFSERR_IO, 460 NFSERR_ACCES, 461 NFSERR_EXIST, 462 NFSERR_XDEV, 463 NFSERR_NOTDIR, 464 NFSERR_INVAL, 465 NFSERR_NOSPC, 466 NFSERR_ROFS, 467 NFSERR_MLINK, 468 NFSERR_NAMETOL, 469 NFSERR_DQUOT, 470 NFSERR_STALE, 471 NFSERR_BADHANDLE, 472 NFSERR_NOTSUPP, 473 NFSERR_SERVERFAULT, 474 0, 475 }; 476 477 static const short nfsv3err_readdir[] = { 478 NFSERR_IO, 479 NFSERR_IO, 480 NFSERR_ACCES, 481 NFSERR_NOTDIR, 482 NFSERR_STALE, 483 NFSERR_BADHANDLE, 484 NFSERR_BAD_COOKIE, 485 NFSERR_TOOSMALL, 486 NFSERR_SERVERFAULT, 487 0, 488 }; 489 490 static const short nfsv3err_readdirplus[] = { 491 NFSERR_IO, 492 NFSERR_IO, 493 NFSERR_ACCES, 494 NFSERR_NOTDIR, 495 NFSERR_STALE, 496 NFSERR_BADHANDLE, 497 NFSERR_BAD_COOKIE, 498 NFSERR_NOTSUPP, 499 NFSERR_TOOSMALL, 500 NFSERR_SERVERFAULT, 501 0, 502 }; 503 504 static const short nfsv3err_fsstat[] = { 505 NFSERR_IO, 506 NFSERR_IO, 507 NFSERR_STALE, 508 NFSERR_BADHANDLE, 509 NFSERR_SERVERFAULT, 510 0, 511 }; 512 513 static const short nfsv3err_fsinfo[] = { 514 NFSERR_STALE, 515 NFSERR_STALE, 516 NFSERR_BADHANDLE, 517 NFSERR_SERVERFAULT, 518 0, 519 }; 520 521 static const short nfsv3err_pathconf[] = { 522 NFSERR_STALE, 523 NFSERR_STALE, 524 NFSERR_BADHANDLE, 525 NFSERR_SERVERFAULT, 526 0, 527 }; 528 529 static const short nfsv3err_commit[] = { 530 NFSERR_IO, 531 NFSERR_IO, 532 NFSERR_STALE, 533 NFSERR_BADHANDLE, 534 NFSERR_SERVERFAULT, 535 0, 536 }; 537 538 static const short * const nfsrv_v3errmap[] = { 539 nfsv3err_null, 540 nfsv3err_getattr, 541 nfsv3err_setattr, 542 nfsv3err_lookup, 543 nfsv3err_access, 544 nfsv3err_readlink, 545 nfsv3err_read, 546 nfsv3err_write, 547 nfsv3err_create, 548 nfsv3err_mkdir, 549 nfsv3err_symlink, 550 nfsv3err_mknod, 551 nfsv3err_remove, 552 nfsv3err_rmdir, 553 nfsv3err_rename, 554 nfsv3err_link, 555 nfsv3err_readdir, 556 nfsv3err_readdirplus, 557 nfsv3err_fsstat, 558 nfsv3err_fsinfo, 559 nfsv3err_pathconf, 560 nfsv3err_commit, 561 }; 562 563 extern struct nfsrtt nfsrtt; 564 565 u_long nfsdirhashmask; 566 567 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *)); 568 569 /* 570 * Create the header for an rpc request packet 571 * The hsiz is the size of the rest of the nfs request header. 572 * (just used to decide if a cluster is a good idea) 573 */ 574 struct mbuf * 575 nfsm_reqh(struct nfsnode *np, u_long procid, int hsiz, char **bposp) 576 { 577 struct mbuf *mb; 578 char *bpos; 579 580 mb = m_get(M_WAIT, MT_DATA); 581 MCLAIM(mb, &nfs_mowner); 582 if (hsiz >= MINCLSIZE) 583 m_clget(mb, M_WAIT); 584 mb->m_len = 0; 585 bpos = mtod(mb, void *); 586 587 /* Finally, return values */ 588 *bposp = bpos; 589 return (mb); 590 } 591 592 /* 593 * Build the RPC header and fill in the authorization info. 594 * The authorization string argument is only used when the credentials 595 * come from outside of the kernel. 596 * Returns the head of the mbuf list. 597 */ 598 struct mbuf * 599 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len, 600 verf_str, mrest, mrest_len, mbp, xidp) 601 kauth_cred_t cr; 602 int nmflag; 603 int procid; 604 int auth_type; 605 int auth_len; 606 char *auth_str; 607 int verf_len; 608 char *verf_str; 609 struct mbuf *mrest; 610 int mrest_len; 611 struct mbuf **mbp; 612 u_int32_t *xidp; 613 { 614 struct mbuf *mb; 615 u_int32_t *tl; 616 char *bpos; 617 int i; 618 struct mbuf *mreq; 619 int siz, grpsiz, authsiz; 620 621 authsiz = nfsm_rndup(auth_len); 622 mb = m_gethdr(M_WAIT, MT_DATA); 623 MCLAIM(mb, &nfs_mowner); 624 if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) { 625 m_clget(mb, M_WAIT); 626 } else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) { 627 MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED); 628 } else { 629 MH_ALIGN(mb, 8 * NFSX_UNSIGNED); 630 } 631 mb->m_len = 0; 632 mreq = mb; 633 bpos = mtod(mb, void *); 634 635 /* 636 * First the RPC header. 637 */ 638 nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED); 639 640 *tl++ = *xidp = nfs_getxid(); 641 *tl++ = rpc_call; 642 *tl++ = rpc_vers; 643 *tl++ = txdr_unsigned(NFS_PROG); 644 if (nmflag & NFSMNT_NFSV3) 645 *tl++ = txdr_unsigned(NFS_VER3); 646 else 647 *tl++ = txdr_unsigned(NFS_VER2); 648 if (nmflag & NFSMNT_NFSV3) 649 *tl++ = txdr_unsigned(procid); 650 else 651 *tl++ = txdr_unsigned(nfsv2_procid[procid]); 652 653 /* 654 * And then the authorization cred. 655 */ 656 *tl++ = txdr_unsigned(auth_type); 657 *tl = txdr_unsigned(authsiz); 658 switch (auth_type) { 659 case RPCAUTH_UNIX: 660 nfsm_build(tl, u_int32_t *, auth_len); 661 *tl++ = 0; /* stamp ?? */ 662 *tl++ = 0; /* NULL hostname */ 663 *tl++ = txdr_unsigned(kauth_cred_geteuid(cr)); 664 *tl++ = txdr_unsigned(kauth_cred_getegid(cr)); 665 grpsiz = (auth_len >> 2) - 5; 666 *tl++ = txdr_unsigned(grpsiz); 667 for (i = 0; i < grpsiz; i++) 668 *tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */ 669 break; 670 case RPCAUTH_KERB4: 671 siz = auth_len; 672 while (siz > 0) { 673 if (M_TRAILINGSPACE(mb) == 0) { 674 struct mbuf *mb2; 675 mb2 = m_get(M_WAIT, MT_DATA); 676 MCLAIM(mb2, &nfs_mowner); 677 if (siz >= MINCLSIZE) 678 m_clget(mb2, M_WAIT); 679 mb->m_next = mb2; 680 mb = mb2; 681 mb->m_len = 0; 682 bpos = mtod(mb, void *); 683 } 684 i = min(siz, M_TRAILINGSPACE(mb)); 685 memcpy(bpos, auth_str, i); 686 mb->m_len += i; 687 auth_str += i; 688 bpos += i; 689 siz -= i; 690 } 691 if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) { 692 for (i = 0; i < siz; i++) 693 *bpos++ = '\0'; 694 mb->m_len += siz; 695 } 696 break; 697 }; 698 699 /* 700 * And the verifier... 701 */ 702 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED); 703 if (verf_str) { 704 *tl++ = txdr_unsigned(RPCAUTH_KERB4); 705 *tl = txdr_unsigned(verf_len); 706 siz = verf_len; 707 while (siz > 0) { 708 if (M_TRAILINGSPACE(mb) == 0) { 709 struct mbuf *mb2; 710 mb2 = m_get(M_WAIT, MT_DATA); 711 MCLAIM(mb2, &nfs_mowner); 712 if (siz >= MINCLSIZE) 713 m_clget(mb2, M_WAIT); 714 mb->m_next = mb2; 715 mb = mb2; 716 mb->m_len = 0; 717 bpos = mtod(mb, void *); 718 } 719 i = min(siz, M_TRAILINGSPACE(mb)); 720 memcpy(bpos, verf_str, i); 721 mb->m_len += i; 722 verf_str += i; 723 bpos += i; 724 siz -= i; 725 } 726 if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) { 727 for (i = 0; i < siz; i++) 728 *bpos++ = '\0'; 729 mb->m_len += siz; 730 } 731 } else { 732 *tl++ = txdr_unsigned(RPCAUTH_NULL); 733 *tl = 0; 734 } 735 mb->m_next = mrest; 736 mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len; 737 mreq->m_pkthdr.rcvif = (struct ifnet *)0; 738 *mbp = mb; 739 return (mreq); 740 } 741 742 /* 743 * copies mbuf chain to the uio scatter/gather list 744 */ 745 int 746 nfsm_mbuftouio(mrep, uiop, siz, dpos) 747 struct mbuf **mrep; 748 struct uio *uiop; 749 int siz; 750 char **dpos; 751 { 752 char *mbufcp, *uiocp; 753 int xfer, left, len; 754 struct mbuf *mp; 755 long uiosiz, rem; 756 int error = 0; 757 758 mp = *mrep; 759 mbufcp = *dpos; 760 len = mtod(mp, char *) + mp->m_len - mbufcp; 761 rem = nfsm_rndup(siz)-siz; 762 while (siz > 0) { 763 if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) 764 return (EFBIG); 765 left = uiop->uio_iov->iov_len; 766 uiocp = uiop->uio_iov->iov_base; 767 if (left > siz) 768 left = siz; 769 uiosiz = left; 770 while (left > 0) { 771 while (len == 0) { 772 mp = mp->m_next; 773 if (mp == NULL) 774 return (EBADRPC); 775 mbufcp = mtod(mp, void *); 776 len = mp->m_len; 777 } 778 xfer = (left > len) ? len : left; 779 error = copyout_vmspace(uiop->uio_vmspace, mbufcp, 780 uiocp, xfer); 781 if (error) { 782 return error; 783 } 784 left -= xfer; 785 len -= xfer; 786 mbufcp += xfer; 787 uiocp += xfer; 788 uiop->uio_offset += xfer; 789 uiop->uio_resid -= xfer; 790 } 791 if (uiop->uio_iov->iov_len <= siz) { 792 uiop->uio_iovcnt--; 793 uiop->uio_iov++; 794 } else { 795 uiop->uio_iov->iov_base = 796 (char *)uiop->uio_iov->iov_base + uiosiz; 797 uiop->uio_iov->iov_len -= uiosiz; 798 } 799 siz -= uiosiz; 800 } 801 *dpos = mbufcp; 802 *mrep = mp; 803 if (rem > 0) { 804 if (len < rem) 805 error = nfs_adv(mrep, dpos, rem, len); 806 else 807 *dpos += rem; 808 } 809 return (error); 810 } 811 812 /* 813 * copies a uio scatter/gather list to an mbuf chain. 814 * NOTE: can ony handle iovcnt == 1 815 */ 816 int 817 nfsm_uiotombuf(uiop, mq, siz, bpos) 818 struct uio *uiop; 819 struct mbuf **mq; 820 int siz; 821 char **bpos; 822 { 823 char *uiocp; 824 struct mbuf *mp, *mp2; 825 int xfer, left, mlen; 826 int uiosiz, clflg, rem; 827 char *cp; 828 int error; 829 830 #ifdef DIAGNOSTIC 831 if (uiop->uio_iovcnt != 1) 832 panic("nfsm_uiotombuf: iovcnt != 1"); 833 #endif 834 835 if (siz > MLEN) /* or should it >= MCLBYTES ?? */ 836 clflg = 1; 837 else 838 clflg = 0; 839 rem = nfsm_rndup(siz)-siz; 840 mp = mp2 = *mq; 841 while (siz > 0) { 842 left = uiop->uio_iov->iov_len; 843 uiocp = uiop->uio_iov->iov_base; 844 if (left > siz) 845 left = siz; 846 uiosiz = left; 847 while (left > 0) { 848 mlen = M_TRAILINGSPACE(mp); 849 if (mlen == 0) { 850 mp = m_get(M_WAIT, MT_DATA); 851 MCLAIM(mp, &nfs_mowner); 852 if (clflg) 853 m_clget(mp, M_WAIT); 854 mp->m_len = 0; 855 mp2->m_next = mp; 856 mp2 = mp; 857 mlen = M_TRAILINGSPACE(mp); 858 } 859 xfer = (left > mlen) ? mlen : left; 860 cp = mtod(mp, char *) + mp->m_len; 861 error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp, 862 xfer); 863 if (error) { 864 /* XXX */ 865 } 866 mp->m_len += xfer; 867 left -= xfer; 868 uiocp += xfer; 869 uiop->uio_offset += xfer; 870 uiop->uio_resid -= xfer; 871 } 872 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + 873 uiosiz; 874 uiop->uio_iov->iov_len -= uiosiz; 875 siz -= uiosiz; 876 } 877 if (rem > 0) { 878 if (rem > M_TRAILINGSPACE(mp)) { 879 mp = m_get(M_WAIT, MT_DATA); 880 MCLAIM(mp, &nfs_mowner); 881 mp->m_len = 0; 882 mp2->m_next = mp; 883 } 884 cp = mtod(mp, char *) + mp->m_len; 885 for (left = 0; left < rem; left++) 886 *cp++ = '\0'; 887 mp->m_len += rem; 888 *bpos = cp; 889 } else 890 *bpos = mtod(mp, char *) + mp->m_len; 891 *mq = mp; 892 return (0); 893 } 894 895 /* 896 * Get at least "siz" bytes of correctly aligned data. 897 * When called the mbuf pointers are not necessarily correct, 898 * dsosp points to what ought to be in m_data and left contains 899 * what ought to be in m_len. 900 * This is used by the macros nfsm_dissect and nfsm_dissecton for tough 901 * cases. (The macros use the vars. dpos and dpos2) 902 */ 903 int 904 nfsm_disct(mdp, dposp, siz, left, cp2) 905 struct mbuf **mdp; 906 char **dposp; 907 int siz; 908 int left; 909 char **cp2; 910 { 911 struct mbuf *m1, *m2; 912 struct mbuf *havebuf = NULL; 913 char *src = *dposp; 914 char *dst; 915 int len; 916 917 #ifdef DEBUG 918 if (left < 0) 919 panic("nfsm_disct: left < 0"); 920 #endif 921 m1 = *mdp; 922 /* 923 * Skip through the mbuf chain looking for an mbuf with 924 * some data. If the first mbuf found has enough data 925 * and it is correctly aligned return it. 926 */ 927 while (left == 0) { 928 havebuf = m1; 929 *mdp = m1 = m1->m_next; 930 if (m1 == NULL) 931 return (EBADRPC); 932 src = mtod(m1, void *); 933 left = m1->m_len; 934 /* 935 * If we start a new mbuf and it is big enough 936 * and correctly aligned just return it, don't 937 * do any pull up. 938 */ 939 if (left >= siz && nfsm_aligned(src)) { 940 *cp2 = src; 941 *dposp = src + siz; 942 return (0); 943 } 944 } 945 if ((m1->m_flags & M_EXT) != 0) { 946 if (havebuf && M_TRAILINGSPACE(havebuf) >= siz && 947 nfsm_aligned(mtod(havebuf, char *) + havebuf->m_len)) { 948 /* 949 * If the first mbuf with data has external data 950 * and there is a previous mbuf with some trailing 951 * space, use it to move the data into. 952 */ 953 m2 = m1; 954 *mdp = m1 = havebuf; 955 *cp2 = mtod(m1, char *) + m1->m_len; 956 } else if (havebuf) { 957 /* 958 * If the first mbuf has a external data 959 * and there is no previous empty mbuf 960 * allocate a new mbuf and move the external 961 * data to the new mbuf. Also make the first 962 * mbuf look empty. 963 */ 964 m2 = m1; 965 *mdp = m1 = m_get(M_WAIT, MT_DATA); 966 MCLAIM(m1, m2->m_owner); 967 if ((m2->m_flags & M_PKTHDR) != 0) { 968 /* XXX MOVE */ 969 M_COPY_PKTHDR(m1, m2); 970 m_tag_delete_chain(m2, NULL); 971 m2->m_flags &= ~M_PKTHDR; 972 } 973 if (havebuf) { 974 havebuf->m_next = m1; 975 } 976 m1->m_next = m2; 977 MRESETDATA(m1); 978 m1->m_len = 0; 979 m2->m_data = src; 980 m2->m_len = left; 981 *cp2 = mtod(m1, char *); 982 } else { 983 struct mbuf **nextp = &m1->m_next; 984 985 m1->m_len -= left; 986 do { 987 m2 = m_get(M_WAIT, MT_DATA); 988 MCLAIM(m2, m1->m_owner); 989 if (left >= MINCLSIZE) { 990 MCLGET(m2, M_WAIT); 991 } 992 m2->m_next = *nextp; 993 *nextp = m2; 994 nextp = &m2->m_next; 995 len = (m2->m_flags & M_EXT) != 0 ? 996 MCLBYTES : MLEN; 997 if (len > left) { 998 len = left; 999 } 1000 memcpy(mtod(m2, char *), src, len); 1001 m2->m_len = len; 1002 src += len; 1003 left -= len; 1004 } while (left > 0); 1005 *mdp = m1 = m1->m_next; 1006 m2 = m1->m_next; 1007 *cp2 = mtod(m1, char *); 1008 } 1009 } else { 1010 /* 1011 * If the first mbuf has no external data 1012 * move the data to the front of the mbuf. 1013 */ 1014 MRESETDATA(m1); 1015 dst = mtod(m1, char *); 1016 if (dst != src) { 1017 memmove(dst, src, left); 1018 } 1019 m1->m_len = left; 1020 m2 = m1->m_next; 1021 *cp2 = m1->m_data; 1022 } 1023 *dposp = *cp2 + siz; 1024 /* 1025 * Loop through mbufs pulling data up into first mbuf until 1026 * the first mbuf is full or there is no more data to 1027 * pullup. 1028 */ 1029 dst = mtod(m1, char *) + m1->m_len; 1030 while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) { 1031 if ((len = min(len, m2->m_len)) != 0) { 1032 memcpy(dst, mtod(m2, char *), len); 1033 } 1034 m1->m_len += len; 1035 dst += len; 1036 m2->m_data += len; 1037 m2->m_len -= len; 1038 m2 = m2->m_next; 1039 } 1040 if (m1->m_len < siz) 1041 return (EBADRPC); 1042 return (0); 1043 } 1044 1045 /* 1046 * Advance the position in the mbuf chain. 1047 */ 1048 int 1049 nfs_adv(mdp, dposp, offs, left) 1050 struct mbuf **mdp; 1051 char **dposp; 1052 int offs; 1053 int left; 1054 { 1055 struct mbuf *m; 1056 int s; 1057 1058 m = *mdp; 1059 s = left; 1060 while (s < offs) { 1061 offs -= s; 1062 m = m->m_next; 1063 if (m == NULL) 1064 return (EBADRPC); 1065 s = m->m_len; 1066 } 1067 *mdp = m; 1068 *dposp = mtod(m, char *) + offs; 1069 return (0); 1070 } 1071 1072 /* 1073 * Copy a string into mbufs for the hard cases... 1074 */ 1075 int 1076 nfsm_strtmbuf(mb, bpos, cp, siz) 1077 struct mbuf **mb; 1078 char **bpos; 1079 const char *cp; 1080 long siz; 1081 { 1082 struct mbuf *m1 = NULL, *m2; 1083 long left, xfer, len, tlen; 1084 u_int32_t *tl; 1085 int putsize; 1086 1087 putsize = 1; 1088 m2 = *mb; 1089 left = M_TRAILINGSPACE(m2); 1090 if (left > 0) { 1091 tl = ((u_int32_t *)(*bpos)); 1092 *tl++ = txdr_unsigned(siz); 1093 putsize = 0; 1094 left -= NFSX_UNSIGNED; 1095 m2->m_len += NFSX_UNSIGNED; 1096 if (left > 0) { 1097 memcpy((void *) tl, cp, left); 1098 siz -= left; 1099 cp += left; 1100 m2->m_len += left; 1101 left = 0; 1102 } 1103 } 1104 /* Loop around adding mbufs */ 1105 while (siz > 0) { 1106 m1 = m_get(M_WAIT, MT_DATA); 1107 MCLAIM(m1, &nfs_mowner); 1108 if (siz > MLEN) 1109 m_clget(m1, M_WAIT); 1110 m1->m_len = NFSMSIZ(m1); 1111 m2->m_next = m1; 1112 m2 = m1; 1113 tl = mtod(m1, u_int32_t *); 1114 tlen = 0; 1115 if (putsize) { 1116 *tl++ = txdr_unsigned(siz); 1117 m1->m_len -= NFSX_UNSIGNED; 1118 tlen = NFSX_UNSIGNED; 1119 putsize = 0; 1120 } 1121 if (siz < m1->m_len) { 1122 len = nfsm_rndup(siz); 1123 xfer = siz; 1124 if (xfer < len) 1125 *(tl+(xfer>>2)) = 0; 1126 } else { 1127 xfer = len = m1->m_len; 1128 } 1129 memcpy((void *) tl, cp, xfer); 1130 m1->m_len = len+tlen; 1131 siz -= xfer; 1132 cp += xfer; 1133 } 1134 *mb = m1; 1135 *bpos = mtod(m1, char *) + m1->m_len; 1136 return (0); 1137 } 1138 1139 /* 1140 * Directory caching routines. They work as follows: 1141 * - a cache is maintained per VDIR nfsnode. 1142 * - for each offset cookie that is exported to userspace, and can 1143 * thus be thrown back at us as an offset to VOP_READDIR, store 1144 * information in the cache. 1145 * - cached are: 1146 * - cookie itself 1147 * - blocknumber (essentially just a search key in the buffer cache) 1148 * - entry number in block. 1149 * - offset cookie of block in which this entry is stored 1150 * - 32 bit cookie if NFSMNT_XLATECOOKIE is used. 1151 * - entries are looked up in a hash table 1152 * - also maintained is an LRU list of entries, used to determine 1153 * which ones to delete if the cache grows too large. 1154 * - if 32 <-> 64 translation mode is requested for a filesystem, 1155 * the cache also functions as a translation table 1156 * - in the translation case, invalidating the cache does not mean 1157 * flushing it, but just marking entries as invalid, except for 1158 * the <64bit cookie, 32bitcookie> pair which is still valid, to 1159 * still be able to use the cache as a translation table. 1160 * - 32 bit cookies are uniquely created by combining the hash table 1161 * entry value, and one generation count per hash table entry, 1162 * incremented each time an entry is appended to the chain. 1163 * - the cache is invalidated each time a direcory is modified 1164 * - sanity checks are also done; if an entry in a block turns 1165 * out not to have a matching cookie, the cache is invalidated 1166 * and a new block starting from the wanted offset is fetched from 1167 * the server. 1168 * - directory entries as read from the server are extended to contain 1169 * the 64bit and, optionally, the 32bit cookies, for sanity checking 1170 * the cache and exporting them to userspace through the cookie 1171 * argument to VOP_READDIR. 1172 */ 1173 1174 u_long 1175 nfs_dirhash(off) 1176 off_t off; 1177 { 1178 int i; 1179 char *cp = (char *)&off; 1180 u_long sum = 0L; 1181 1182 for (i = 0 ; i < sizeof (off); i++) 1183 sum += *cp++; 1184 1185 return sum; 1186 } 1187 1188 #define _NFSDC_MTX(np) (&NFSTOV(np)->v_interlock) 1189 #define NFSDC_LOCK(np) mutex_enter(_NFSDC_MTX(np)) 1190 #define NFSDC_UNLOCK(np) mutex_exit(_NFSDC_MTX(np)) 1191 #define NFSDC_ASSERT_LOCKED(np) KASSERT(mutex_owned(_NFSDC_MTX(np))) 1192 1193 void 1194 nfs_initdircache(vp) 1195 struct vnode *vp; 1196 { 1197 struct nfsnode *np = VTONFS(vp); 1198 struct nfsdirhashhead *dircache; 1199 1200 dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, true, 1201 &nfsdirhashmask); 1202 1203 NFSDC_LOCK(np); 1204 if (np->n_dircache == NULL) { 1205 np->n_dircachesize = 0; 1206 np->n_dircache = dircache; 1207 dircache = NULL; 1208 TAILQ_INIT(&np->n_dirchain); 1209 } 1210 NFSDC_UNLOCK(np); 1211 if (dircache) 1212 hashdone(dircache, HASH_LIST, nfsdirhashmask); 1213 } 1214 1215 void 1216 nfs_initdirxlatecookie(vp) 1217 struct vnode *vp; 1218 { 1219 struct nfsnode *np = VTONFS(vp); 1220 unsigned *dirgens; 1221 1222 KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE); 1223 1224 dirgens = kmem_zalloc(NFS_DIRHASHSIZ * sizeof(unsigned), KM_SLEEP); 1225 NFSDC_LOCK(np); 1226 if (np->n_dirgens == NULL) { 1227 np->n_dirgens = dirgens; 1228 dirgens = NULL; 1229 } 1230 NFSDC_UNLOCK(np); 1231 if (dirgens) 1232 kmem_free(dirgens, NFS_DIRHASHSIZ * sizeof(unsigned)); 1233 } 1234 1235 static const struct nfsdircache dzero; 1236 1237 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *)); 1238 static void nfs_putdircache_unlocked __P((struct nfsnode *, 1239 struct nfsdircache *)); 1240 1241 static void 1242 nfs_unlinkdircache(np, ndp) 1243 struct nfsnode *np; 1244 struct nfsdircache *ndp; 1245 { 1246 1247 NFSDC_ASSERT_LOCKED(np); 1248 KASSERT(ndp != &dzero); 1249 1250 if (LIST_NEXT(ndp, dc_hash) == (void *)-1) 1251 return; 1252 1253 TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain); 1254 LIST_REMOVE(ndp, dc_hash); 1255 LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */ 1256 1257 nfs_putdircache_unlocked(np, ndp); 1258 } 1259 1260 void 1261 nfs_putdircache(np, ndp) 1262 struct nfsnode *np; 1263 struct nfsdircache *ndp; 1264 { 1265 int ref; 1266 1267 if (ndp == &dzero) 1268 return; 1269 1270 KASSERT(ndp->dc_refcnt > 0); 1271 NFSDC_LOCK(np); 1272 ref = --ndp->dc_refcnt; 1273 NFSDC_UNLOCK(np); 1274 1275 if (ref == 0) 1276 kmem_free(ndp, sizeof(*ndp)); 1277 } 1278 1279 static void 1280 nfs_putdircache_unlocked(struct nfsnode *np, struct nfsdircache *ndp) 1281 { 1282 int ref; 1283 1284 NFSDC_ASSERT_LOCKED(np); 1285 1286 if (ndp == &dzero) 1287 return; 1288 1289 KASSERT(ndp->dc_refcnt > 0); 1290 ref = --ndp->dc_refcnt; 1291 if (ref == 0) 1292 kmem_free(ndp, sizeof(*ndp)); 1293 } 1294 1295 struct nfsdircache * 1296 nfs_searchdircache(vp, off, do32, hashent) 1297 struct vnode *vp; 1298 off_t off; 1299 int do32; 1300 int *hashent; 1301 { 1302 struct nfsdirhashhead *ndhp; 1303 struct nfsdircache *ndp = NULL; 1304 struct nfsnode *np = VTONFS(vp); 1305 unsigned ent; 1306 1307 /* 1308 * Zero is always a valid cookie. 1309 */ 1310 if (off == 0) 1311 /* XXXUNCONST */ 1312 return (struct nfsdircache *)__UNCONST(&dzero); 1313 1314 if (!np->n_dircache) 1315 return NULL; 1316 1317 /* 1318 * We use a 32bit cookie as search key, directly reconstruct 1319 * the hashentry. Else use the hashfunction. 1320 */ 1321 if (do32) { 1322 ent = (u_int32_t)off >> 24; 1323 if (ent >= NFS_DIRHASHSIZ) 1324 return NULL; 1325 ndhp = &np->n_dircache[ent]; 1326 } else { 1327 ndhp = NFSDIRHASH(np, off); 1328 } 1329 1330 if (hashent) 1331 *hashent = (int)(ndhp - np->n_dircache); 1332 1333 NFSDC_LOCK(np); 1334 if (do32) { 1335 LIST_FOREACH(ndp, ndhp, dc_hash) { 1336 if (ndp->dc_cookie32 == (u_int32_t)off) { 1337 /* 1338 * An invalidated entry will become the 1339 * start of a new block fetched from 1340 * the server. 1341 */ 1342 if (ndp->dc_flags & NFSDC_INVALID) { 1343 ndp->dc_blkcookie = ndp->dc_cookie; 1344 ndp->dc_entry = 0; 1345 ndp->dc_flags &= ~NFSDC_INVALID; 1346 } 1347 break; 1348 } 1349 } 1350 } else { 1351 LIST_FOREACH(ndp, ndhp, dc_hash) { 1352 if (ndp->dc_cookie == off) 1353 break; 1354 } 1355 } 1356 if (ndp != NULL) 1357 ndp->dc_refcnt++; 1358 NFSDC_UNLOCK(np); 1359 return ndp; 1360 } 1361 1362 1363 struct nfsdircache * 1364 nfs_enterdircache(struct vnode *vp, off_t off, off_t blkoff, int en, 1365 daddr_t blkno) 1366 { 1367 struct nfsnode *np = VTONFS(vp); 1368 struct nfsdirhashhead *ndhp; 1369 struct nfsdircache *ndp = NULL; 1370 struct nfsdircache *newndp = NULL; 1371 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1372 int hashent = 0, gen, overwrite; /* XXX: GCC */ 1373 1374 /* 1375 * XXX refuse entries for offset 0. amd(8) erroneously sets 1376 * cookie 0 for the '.' entry, making this necessary. This 1377 * isn't so bad, as 0 is a special case anyway. 1378 */ 1379 if (off == 0) 1380 /* XXXUNCONST */ 1381 return (struct nfsdircache *)__UNCONST(&dzero); 1382 1383 if (!np->n_dircache) 1384 /* 1385 * XXX would like to do this in nfs_nget but vtype 1386 * isn't known at that time. 1387 */ 1388 nfs_initdircache(vp); 1389 1390 if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens) 1391 nfs_initdirxlatecookie(vp); 1392 1393 retry: 1394 ndp = nfs_searchdircache(vp, off, 0, &hashent); 1395 1396 NFSDC_LOCK(np); 1397 if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) { 1398 /* 1399 * Overwriting an old entry. Check if it's the same. 1400 * If so, just return. If not, remove the old entry. 1401 */ 1402 if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en) 1403 goto done; 1404 nfs_unlinkdircache(np, ndp); 1405 nfs_putdircache_unlocked(np, ndp); 1406 ndp = NULL; 1407 } 1408 1409 ndhp = &np->n_dircache[hashent]; 1410 1411 if (!ndp) { 1412 if (newndp == NULL) { 1413 NFSDC_UNLOCK(np); 1414 newndp = kmem_alloc(sizeof(*newndp), KM_SLEEP); 1415 newndp->dc_refcnt = 1; 1416 LIST_NEXT(newndp, dc_hash) = (void *)-1; 1417 goto retry; 1418 } 1419 ndp = newndp; 1420 newndp = NULL; 1421 overwrite = 0; 1422 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 1423 /* 1424 * We're allocating a new entry, so bump the 1425 * generation number. 1426 */ 1427 KASSERT(np->n_dirgens); 1428 gen = ++np->n_dirgens[hashent]; 1429 if (gen == 0) { 1430 np->n_dirgens[hashent]++; 1431 gen++; 1432 } 1433 ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff); 1434 } 1435 } else 1436 overwrite = 1; 1437 1438 ndp->dc_cookie = off; 1439 ndp->dc_blkcookie = blkoff; 1440 ndp->dc_entry = en; 1441 ndp->dc_flags = 0; 1442 1443 if (overwrite) 1444 goto done; 1445 1446 /* 1447 * If the maximum directory cookie cache size has been reached 1448 * for this node, take one off the front. The idea is that 1449 * directories are typically read front-to-back once, so that 1450 * the oldest entries can be thrown away without much performance 1451 * loss. 1452 */ 1453 if (np->n_dircachesize == NFS_MAXDIRCACHE) { 1454 nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain)); 1455 } else 1456 np->n_dircachesize++; 1457 1458 KASSERT(ndp->dc_refcnt == 1); 1459 LIST_INSERT_HEAD(ndhp, ndp, dc_hash); 1460 TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain); 1461 ndp->dc_refcnt++; 1462 done: 1463 KASSERT(ndp->dc_refcnt > 0); 1464 NFSDC_UNLOCK(np); 1465 if (newndp) 1466 nfs_putdircache(np, newndp); 1467 return ndp; 1468 } 1469 1470 void 1471 nfs_invaldircache(vp, flags) 1472 struct vnode *vp; 1473 int flags; 1474 { 1475 struct nfsnode *np = VTONFS(vp); 1476 struct nfsdircache *ndp = NULL; 1477 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1478 const bool forcefree = flags & NFS_INVALDIRCACHE_FORCE; 1479 1480 #ifdef DIAGNOSTIC 1481 if (vp->v_type != VDIR) 1482 panic("nfs: invaldircache: not dir"); 1483 #endif 1484 1485 if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0) 1486 np->n_flag &= ~NEOFVALID; 1487 1488 if (!np->n_dircache) 1489 return; 1490 1491 NFSDC_LOCK(np); 1492 if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) { 1493 while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) { 1494 KASSERT(!forcefree || ndp->dc_refcnt == 1); 1495 nfs_unlinkdircache(np, ndp); 1496 } 1497 np->n_dircachesize = 0; 1498 if (forcefree && np->n_dirgens) { 1499 kmem_free(np->n_dirgens, 1500 NFS_DIRHASHSIZ * sizeof(unsigned)); 1501 np->n_dirgens = NULL; 1502 } 1503 } else { 1504 TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain) 1505 ndp->dc_flags |= NFSDC_INVALID; 1506 } 1507 1508 NFSDC_UNLOCK(np); 1509 } 1510 1511 /* 1512 * Called once before VFS init to initialize shared and 1513 * server-specific data structures. 1514 */ 1515 static int 1516 nfs_init0(void) 1517 { 1518 1519 nfsrtt.pos = 0; 1520 rpc_vers = txdr_unsigned(RPC_VER2); 1521 rpc_call = txdr_unsigned(RPC_CALL); 1522 rpc_reply = txdr_unsigned(RPC_REPLY); 1523 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED); 1524 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED); 1525 rpc_mismatch = txdr_unsigned(RPC_MISMATCH); 1526 rpc_autherr = txdr_unsigned(RPC_AUTHERR); 1527 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX); 1528 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4); 1529 nfs_prog = txdr_unsigned(NFS_PROG); 1530 nfs_true = txdr_unsigned(true); 1531 nfs_false = txdr_unsigned(false); 1532 nfs_xdrneg1 = txdr_unsigned(-1); 1533 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; 1534 if (nfs_ticks < 1) 1535 nfs_ticks = 1; 1536 nfs_xid = arc4random(); 1537 nfsdreq_init(); 1538 1539 /* 1540 * Initialize reply list and start timer 1541 */ 1542 TAILQ_INIT(&nfs_reqq); 1543 nfs_timer_init(); 1544 MOWNER_ATTACH(&nfs_mowner); 1545 1546 #ifdef NFS 1547 /* Initialize the kqueue structures */ 1548 nfs_kqinit(); 1549 /* Initialize the iod structures */ 1550 nfs_iodinit(); 1551 #endif 1552 1553 return 0; 1554 } 1555 1556 /* 1557 * This is disgusting, but it must support both modular and monolothic 1558 * configurations. For monolithic builds NFSSERVER may not imply NFS. 1559 * 1560 * Yuck. 1561 */ 1562 void 1563 nfs_init(void) 1564 { 1565 static ONCE_DECL(nfs_init_once); 1566 1567 RUN_ONCE(&nfs_init_once, nfs_init0); 1568 } 1569 1570 void 1571 nfs_fini(void) 1572 { 1573 1574 #ifdef NFS 1575 nfs_kqfini(); 1576 nfs_iodfini(); 1577 #endif 1578 nfsdreq_fini(); 1579 nfs_timer_fini(); 1580 MOWNER_DETACH(&nfs_mowner); 1581 } 1582 1583 #ifdef NFS 1584 /* 1585 * Called once at VFS init to initialize client-specific data structures. 1586 */ 1587 void 1588 nfs_vfs_init() 1589 { 1590 1591 /* Initialize NFS server / client shared data. */ 1592 nfs_init(); 1593 nfs_node_init(); 1594 1595 nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4); 1596 } 1597 1598 void 1599 nfs_vfs_done() 1600 { 1601 1602 nfs_node_done(); 1603 } 1604 1605 /* 1606 * Attribute cache routines. 1607 * nfs_loadattrcache() - loads or updates the cache contents from attributes 1608 * that are on the mbuf list 1609 * nfs_getattrcache() - returns valid attributes if found in cache, returns 1610 * error otherwise 1611 */ 1612 1613 /* 1614 * Load the attribute cache (that lives in the nfsnode entry) with 1615 * the values on the mbuf list and 1616 * Iff vap not NULL 1617 * copy the attributes to *vaper 1618 */ 1619 int 1620 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags) 1621 struct vnode **vpp; 1622 struct mbuf **mdp; 1623 char **dposp; 1624 struct vattr *vaper; 1625 int flags; 1626 { 1627 int32_t t1; 1628 char *cp2; 1629 int error = 0; 1630 struct mbuf *md; 1631 int v3 = NFS_ISV3(*vpp); 1632 1633 md = *mdp; 1634 t1 = (mtod(md, char *) + md->m_len) - *dposp; 1635 error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2); 1636 if (error) 1637 return (error); 1638 return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags); 1639 } 1640 1641 int 1642 nfs_loadattrcache(vpp, fp, vaper, flags) 1643 struct vnode **vpp; 1644 struct nfs_fattr *fp; 1645 struct vattr *vaper; 1646 int flags; 1647 { 1648 struct vnode *vp = *vpp; 1649 struct vattr *vap; 1650 int v3 = NFS_ISV3(vp); 1651 enum vtype vtyp; 1652 u_short vmode; 1653 struct timespec mtime; 1654 struct timespec ctime; 1655 int32_t rdev; 1656 struct nfsnode *np; 1657 extern int (**spec_nfsv2nodeop_p) __P((void *)); 1658 uid_t uid; 1659 gid_t gid; 1660 1661 if (v3) { 1662 vtyp = nfsv3tov_type(fp->fa_type); 1663 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1664 rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1), 1665 fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2)); 1666 fxdr_nfsv3time(&fp->fa3_mtime, &mtime); 1667 fxdr_nfsv3time(&fp->fa3_ctime, &ctime); 1668 } else { 1669 vtyp = nfsv2tov_type(fp->fa_type); 1670 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1671 if (vtyp == VNON || vtyp == VREG) 1672 vtyp = IFTOVT(vmode); 1673 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); 1674 fxdr_nfsv2time(&fp->fa2_mtime, &mtime); 1675 ctime.tv_sec = fxdr_unsigned(u_int32_t, 1676 fp->fa2_ctime.nfsv2_sec); 1677 ctime.tv_nsec = 0; 1678 1679 /* 1680 * Really ugly NFSv2 kludge. 1681 */ 1682 if (vtyp == VCHR && rdev == 0xffffffff) 1683 vtyp = VFIFO; 1684 } 1685 1686 vmode &= ALLPERMS; 1687 1688 /* 1689 * If v_type == VNON it is a new node, so fill in the v_type, 1690 * n_mtime fields. Check to see if it represents a special 1691 * device, and if so, check for a possible alias. Once the 1692 * correct vnode has been obtained, fill in the rest of the 1693 * information. 1694 */ 1695 np = VTONFS(vp); 1696 if (vp->v_type == VNON) { 1697 vp->v_type = vtyp; 1698 if (vp->v_type == VFIFO) { 1699 extern int (**fifo_nfsv2nodeop_p) __P((void *)); 1700 vp->v_op = fifo_nfsv2nodeop_p; 1701 } else if (vp->v_type == VREG) { 1702 mutex_init(&np->n_commitlock, MUTEX_DEFAULT, IPL_NONE); 1703 } else if (vp->v_type == VCHR || vp->v_type == VBLK) { 1704 vp->v_op = spec_nfsv2nodeop_p; 1705 spec_node_init(vp, (dev_t)rdev); 1706 } 1707 np->n_mtime = mtime; 1708 } 1709 uid = fxdr_unsigned(uid_t, fp->fa_uid); 1710 gid = fxdr_unsigned(gid_t, fp->fa_gid); 1711 vap = np->n_vattr; 1712 1713 /* 1714 * Invalidate access cache if uid, gid, mode or ctime changed. 1715 */ 1716 if (np->n_accstamp != -1 && 1717 (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode 1718 || timespeccmp(&ctime, &vap->va_ctime, !=))) 1719 np->n_accstamp = -1; 1720 1721 vap->va_type = vtyp; 1722 vap->va_mode = vmode; 1723 vap->va_rdev = (dev_t)rdev; 1724 vap->va_mtime = mtime; 1725 vap->va_ctime = ctime; 1726 vap->va_birthtime.tv_sec = VNOVAL; 1727 vap->va_birthtime.tv_nsec = VNOVAL; 1728 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 1729 switch (vtyp) { 1730 case VDIR: 1731 vap->va_blocksize = NFS_DIRFRAGSIZ; 1732 break; 1733 case VBLK: 1734 vap->va_blocksize = BLKDEV_IOSIZE; 1735 break; 1736 case VCHR: 1737 vap->va_blocksize = MAXBSIZE; 1738 break; 1739 default: 1740 vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize : 1741 fxdr_unsigned(int32_t, fp->fa2_blocksize); 1742 break; 1743 } 1744 if (v3) { 1745 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1746 vap->va_uid = uid; 1747 vap->va_gid = gid; 1748 vap->va_size = fxdr_hyper(&fp->fa3_size); 1749 vap->va_bytes = fxdr_hyper(&fp->fa3_used); 1750 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid); 1751 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); 1752 vap->va_flags = 0; 1753 vap->va_filerev = 0; 1754 } else { 1755 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1756 vap->va_uid = uid; 1757 vap->va_gid = gid; 1758 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); 1759 vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks) 1760 * NFS_FABLKSIZE; 1761 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); 1762 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); 1763 vap->va_flags = 0; 1764 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec); 1765 vap->va_filerev = 0; 1766 } 1767 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 1768 return EFBIG; 1769 } 1770 if (vap->va_size != np->n_size) { 1771 if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) { 1772 vap->va_size = np->n_size; 1773 } else { 1774 np->n_size = vap->va_size; 1775 if (vap->va_type == VREG) { 1776 /* 1777 * we can't free pages if NAC_NOTRUNC because 1778 * the pages can be owned by ourselves. 1779 */ 1780 if (flags & NAC_NOTRUNC) { 1781 np->n_flag |= NTRUNCDELAYED; 1782 } else { 1783 genfs_node_wrlock(vp); 1784 mutex_enter(&vp->v_interlock); 1785 (void)VOP_PUTPAGES(vp, 0, 1786 0, PGO_SYNCIO | PGO_CLEANIT | 1787 PGO_FREE | PGO_ALLPAGES); 1788 uvm_vnp_setsize(vp, np->n_size); 1789 genfs_node_unlock(vp); 1790 } 1791 } 1792 } 1793 } 1794 np->n_attrstamp = time_second; 1795 if (vaper != NULL) { 1796 memcpy((void *)vaper, (void *)vap, sizeof(*vap)); 1797 if (np->n_flag & NCHG) { 1798 if (np->n_flag & NACC) 1799 vaper->va_atime = np->n_atim; 1800 if (np->n_flag & NUPD) 1801 vaper->va_mtime = np->n_mtim; 1802 } 1803 } 1804 return (0); 1805 } 1806 1807 /* 1808 * Check the time stamp 1809 * If the cache is valid, copy contents to *vap and return 0 1810 * otherwise return an error 1811 */ 1812 int 1813 nfs_getattrcache(vp, vaper) 1814 struct vnode *vp; 1815 struct vattr *vaper; 1816 { 1817 struct nfsnode *np = VTONFS(vp); 1818 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1819 struct vattr *vap; 1820 1821 if (np->n_attrstamp == 0 || 1822 (time_second - np->n_attrstamp) >= nfs_attrtimeo(nmp, np)) { 1823 nfsstats.attrcache_misses++; 1824 return (ENOENT); 1825 } 1826 nfsstats.attrcache_hits++; 1827 vap = np->n_vattr; 1828 if (vap->va_size != np->n_size) { 1829 if (vap->va_type == VREG) { 1830 if ((np->n_flag & NMODIFIED) != 0 && 1831 vap->va_size < np->n_size) { 1832 vap->va_size = np->n_size; 1833 } else { 1834 np->n_size = vap->va_size; 1835 } 1836 genfs_node_wrlock(vp); 1837 uvm_vnp_setsize(vp, np->n_size); 1838 genfs_node_unlock(vp); 1839 } else 1840 np->n_size = vap->va_size; 1841 } 1842 memcpy((void *)vaper, (void *)vap, sizeof(struct vattr)); 1843 if (np->n_flag & NCHG) { 1844 if (np->n_flag & NACC) 1845 vaper->va_atime = np->n_atim; 1846 if (np->n_flag & NUPD) 1847 vaper->va_mtime = np->n_mtim; 1848 } 1849 return (0); 1850 } 1851 1852 void 1853 nfs_delayedtruncate(vp) 1854 struct vnode *vp; 1855 { 1856 struct nfsnode *np = VTONFS(vp); 1857 1858 if (np->n_flag & NTRUNCDELAYED) { 1859 np->n_flag &= ~NTRUNCDELAYED; 1860 genfs_node_wrlock(vp); 1861 mutex_enter(&vp->v_interlock); 1862 (void)VOP_PUTPAGES(vp, 0, 1863 0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES); 1864 uvm_vnp_setsize(vp, np->n_size); 1865 genfs_node_unlock(vp); 1866 } 1867 } 1868 1869 #define NFS_WCCKLUDGE_TIMEOUT (24 * 60 * 60) /* 1 day */ 1870 #define NFS_WCCKLUDGE(nmp, now) \ 1871 (((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \ 1872 ((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0) 1873 1874 /* 1875 * nfs_check_wccdata: check inaccurate wcc_data 1876 * 1877 * => return non-zero if we shouldn't trust the wcc_data. 1878 * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed". 1879 */ 1880 1881 int 1882 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime, 1883 struct timespec *mtime, bool docheck) 1884 { 1885 int error = 0; 1886 1887 #if !defined(NFS_V2_ONLY) 1888 1889 if (docheck) { 1890 struct vnode *vp = NFSTOV(np); 1891 struct nfsmount *nmp; 1892 long now = time_second; 1893 const struct timespec *omtime = &np->n_vattr->va_mtime; 1894 const struct timespec *octime = &np->n_vattr->va_ctime; 1895 const char *reason = NULL; /* XXX: gcc */ 1896 1897 if (timespeccmp(omtime, mtime, <=)) { 1898 reason = "mtime"; 1899 error = EINVAL; 1900 } 1901 1902 if (vp->v_type == VDIR && timespeccmp(octime, ctime, <=)) { 1903 reason = "ctime"; 1904 error = EINVAL; 1905 } 1906 1907 nmp = VFSTONFS(vp->v_mount); 1908 if (error) { 1909 1910 /* 1911 * despite of the fact that we've updated the file, 1912 * timestamps of the file were not updated as we 1913 * expected. 1914 * it means that the server has incompatible 1915 * semantics of timestamps or (more likely) 1916 * the server time is not precise enough to 1917 * track each modifications. 1918 * in that case, we disable wcc processing. 1919 * 1920 * yes, strictly speaking, we should disable all 1921 * caching. it's a compromise. 1922 */ 1923 1924 mutex_enter(&nmp->nm_lock); 1925 if (!NFS_WCCKLUDGE(nmp, now)) { 1926 printf("%s: inaccurate wcc data (%s) detected," 1927 " disabling wcc" 1928 " (ctime %u.%09u %u.%09u," 1929 " mtime %u.%09u %u.%09u)\n", 1930 vp->v_mount->mnt_stat.f_mntfromname, 1931 reason, 1932 (unsigned int)octime->tv_sec, 1933 (unsigned int)octime->tv_nsec, 1934 (unsigned int)ctime->tv_sec, 1935 (unsigned int)ctime->tv_nsec, 1936 (unsigned int)omtime->tv_sec, 1937 (unsigned int)omtime->tv_nsec, 1938 (unsigned int)mtime->tv_sec, 1939 (unsigned int)mtime->tv_nsec); 1940 } 1941 nmp->nm_iflag |= NFSMNT_WCCKLUDGE; 1942 nmp->nm_wcckludgetime = now; 1943 mutex_exit(&nmp->nm_lock); 1944 } else if (NFS_WCCKLUDGE(nmp, now)) { 1945 error = EPERM; /* XXX */ 1946 } else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1947 mutex_enter(&nmp->nm_lock); 1948 if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1949 printf("%s: re-enabling wcc\n", 1950 vp->v_mount->mnt_stat.f_mntfromname); 1951 nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE; 1952 } 1953 mutex_exit(&nmp->nm_lock); 1954 } 1955 } 1956 1957 #endif /* !defined(NFS_V2_ONLY) */ 1958 1959 return error; 1960 } 1961 1962 /* 1963 * Heuristic to see if the server XDR encodes directory cookies or not. 1964 * it is not supposed to, but a lot of servers may do this. Also, since 1965 * most/all servers will implement V2 as well, it is expected that they 1966 * may return just 32 bits worth of cookie information, so we need to 1967 * find out in which 32 bits this information is available. We do this 1968 * to avoid trouble with emulated binaries that can't handle 64 bit 1969 * directory offsets. 1970 */ 1971 1972 void 1973 nfs_cookieheuristic(vp, flagp, l, cred) 1974 struct vnode *vp; 1975 int *flagp; 1976 struct lwp *l; 1977 kauth_cred_t cred; 1978 { 1979 struct uio auio; 1980 struct iovec aiov; 1981 char *tbuf, *cp; 1982 struct dirent *dp; 1983 off_t *cookies = NULL, *cop; 1984 int error, eof, nc, len; 1985 1986 tbuf = malloc(NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK); 1987 1988 aiov.iov_base = tbuf; 1989 aiov.iov_len = NFS_DIRFRAGSIZ; 1990 auio.uio_iov = &aiov; 1991 auio.uio_iovcnt = 1; 1992 auio.uio_rw = UIO_READ; 1993 auio.uio_resid = NFS_DIRFRAGSIZ; 1994 auio.uio_offset = 0; 1995 UIO_SETUP_SYSSPACE(&auio); 1996 1997 error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc); 1998 1999 len = NFS_DIRFRAGSIZ - auio.uio_resid; 2000 if (error || len == 0) { 2001 free(tbuf, M_TEMP); 2002 if (cookies) 2003 free(cookies, M_TEMP); 2004 return; 2005 } 2006 2007 /* 2008 * Find the first valid entry and look at its offset cookie. 2009 */ 2010 2011 cp = tbuf; 2012 for (cop = cookies; len > 0; len -= dp->d_reclen) { 2013 dp = (struct dirent *)cp; 2014 if (dp->d_fileno != 0 && len >= dp->d_reclen) { 2015 if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) { 2016 *flagp |= NFSMNT_SWAPCOOKIE; 2017 nfs_invaldircache(vp, 0); 2018 nfs_vinvalbuf(vp, 0, cred, l, 1); 2019 } 2020 break; 2021 } 2022 cop++; 2023 cp += dp->d_reclen; 2024 } 2025 2026 free(tbuf, M_TEMP); 2027 free(cookies, M_TEMP); 2028 } 2029 #endif /* NFS */ 2030 2031 /* 2032 * A fiddled version of m_adj() that ensures null fill to a 32-bit 2033 * boundary and only trims off the back end 2034 * 2035 * 1. trim off 'len' bytes as m_adj(mp, -len). 2036 * 2. add zero-padding 'nul' bytes at the end of the mbuf chain. 2037 */ 2038 void 2039 nfs_zeropad(mp, len, nul) 2040 struct mbuf *mp; 2041 int len; 2042 int nul; 2043 { 2044 struct mbuf *m; 2045 int count; 2046 2047 /* 2048 * Trim from tail. Scan the mbuf chain, 2049 * calculating its length and finding the last mbuf. 2050 * If the adjustment only affects this mbuf, then just 2051 * adjust and return. Otherwise, rescan and truncate 2052 * after the remaining size. 2053 */ 2054 count = 0; 2055 m = mp; 2056 for (;;) { 2057 count += m->m_len; 2058 if (m->m_next == NULL) 2059 break; 2060 m = m->m_next; 2061 } 2062 2063 KDASSERT(count >= len); 2064 2065 if (m->m_len >= len) { 2066 m->m_len -= len; 2067 } else { 2068 count -= len; 2069 /* 2070 * Correct length for chain is "count". 2071 * Find the mbuf with last data, adjust its length, 2072 * and toss data from remaining mbufs on chain. 2073 */ 2074 for (m = mp; m; m = m->m_next) { 2075 if (m->m_len >= count) { 2076 m->m_len = count; 2077 break; 2078 } 2079 count -= m->m_len; 2080 } 2081 KASSERT(m && m->m_next); 2082 m_freem(m->m_next); 2083 m->m_next = NULL; 2084 } 2085 2086 KDASSERT(m->m_next == NULL); 2087 2088 /* 2089 * zero-padding. 2090 */ 2091 if (nul > 0) { 2092 char *cp; 2093 int i; 2094 2095 if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) { 2096 struct mbuf *n; 2097 2098 KDASSERT(MLEN >= nul); 2099 n = m_get(M_WAIT, MT_DATA); 2100 MCLAIM(n, &nfs_mowner); 2101 n->m_len = nul; 2102 n->m_next = NULL; 2103 m->m_next = n; 2104 cp = mtod(n, void *); 2105 } else { 2106 cp = mtod(m, char *) + m->m_len; 2107 m->m_len += nul; 2108 } 2109 for (i = 0; i < nul; i++) 2110 *cp++ = '\0'; 2111 } 2112 return; 2113 } 2114 2115 /* 2116 * Make these functions instead of macros, so that the kernel text size 2117 * doesn't get too big... 2118 */ 2119 void 2120 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp) 2121 struct nfsrv_descript *nfsd; 2122 int before_ret; 2123 struct vattr *before_vap; 2124 int after_ret; 2125 struct vattr *after_vap; 2126 struct mbuf **mbp; 2127 char **bposp; 2128 { 2129 struct mbuf *mb = *mbp; 2130 char *bpos = *bposp; 2131 u_int32_t *tl; 2132 2133 if (before_ret) { 2134 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2135 *tl = nfs_false; 2136 } else { 2137 nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED); 2138 *tl++ = nfs_true; 2139 txdr_hyper(before_vap->va_size, tl); 2140 tl += 2; 2141 txdr_nfsv3time(&(before_vap->va_mtime), tl); 2142 tl += 2; 2143 txdr_nfsv3time(&(before_vap->va_ctime), tl); 2144 } 2145 *bposp = bpos; 2146 *mbp = mb; 2147 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp); 2148 } 2149 2150 void 2151 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp) 2152 struct nfsrv_descript *nfsd; 2153 int after_ret; 2154 struct vattr *after_vap; 2155 struct mbuf **mbp; 2156 char **bposp; 2157 { 2158 struct mbuf *mb = *mbp; 2159 char *bpos = *bposp; 2160 u_int32_t *tl; 2161 struct nfs_fattr *fp; 2162 2163 if (after_ret) { 2164 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2165 *tl = nfs_false; 2166 } else { 2167 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR); 2168 *tl++ = nfs_true; 2169 fp = (struct nfs_fattr *)tl; 2170 nfsm_srvfattr(nfsd, after_vap, fp); 2171 } 2172 *mbp = mb; 2173 *bposp = bpos; 2174 } 2175 2176 void 2177 nfsm_srvfattr(nfsd, vap, fp) 2178 struct nfsrv_descript *nfsd; 2179 struct vattr *vap; 2180 struct nfs_fattr *fp; 2181 { 2182 2183 fp->fa_nlink = txdr_unsigned(vap->va_nlink); 2184 fp->fa_uid = txdr_unsigned(vap->va_uid); 2185 fp->fa_gid = txdr_unsigned(vap->va_gid); 2186 if (nfsd->nd_flag & ND_NFSV3) { 2187 fp->fa_type = vtonfsv3_type(vap->va_type); 2188 fp->fa_mode = vtonfsv3_mode(vap->va_mode); 2189 txdr_hyper(vap->va_size, &fp->fa3_size); 2190 txdr_hyper(vap->va_bytes, &fp->fa3_used); 2191 fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev)); 2192 fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev)); 2193 fp->fa3_fsid.nfsuquad[0] = 0; 2194 fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid); 2195 txdr_hyper(vap->va_fileid, &fp->fa3_fileid); 2196 txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime); 2197 txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime); 2198 txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime); 2199 } else { 2200 fp->fa_type = vtonfsv2_type(vap->va_type); 2201 fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 2202 fp->fa2_size = txdr_unsigned(vap->va_size); 2203 fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize); 2204 if (vap->va_type == VFIFO) 2205 fp->fa2_rdev = 0xffffffff; 2206 else 2207 fp->fa2_rdev = txdr_unsigned(vap->va_rdev); 2208 fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE); 2209 fp->fa2_fsid = txdr_unsigned(vap->va_fsid); 2210 fp->fa2_fileid = txdr_unsigned(vap->va_fileid); 2211 txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime); 2212 txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime); 2213 txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime); 2214 } 2215 } 2216 2217 /* 2218 * This function compares two net addresses by family and returns true 2219 * if they are the same host. 2220 * If there is any doubt, return false. 2221 * The AF_INET family is handled as a special case so that address mbufs 2222 * don't need to be saved to store "struct in_addr", which is only 4 bytes. 2223 */ 2224 int 2225 netaddr_match(family, haddr, nam) 2226 int family; 2227 union nethostaddr *haddr; 2228 struct mbuf *nam; 2229 { 2230 struct sockaddr_in *inetaddr; 2231 2232 switch (family) { 2233 case AF_INET: 2234 inetaddr = mtod(nam, struct sockaddr_in *); 2235 if (inetaddr->sin_family == AF_INET && 2236 inetaddr->sin_addr.s_addr == haddr->had_inetaddr) 2237 return (1); 2238 break; 2239 case AF_INET6: 2240 { 2241 struct sockaddr_in6 *sin6_1, *sin6_2; 2242 2243 sin6_1 = mtod(nam, struct sockaddr_in6 *); 2244 sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *); 2245 if (sin6_1->sin6_family == AF_INET6 && 2246 IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr)) 2247 return 1; 2248 } 2249 default: 2250 break; 2251 }; 2252 return (0); 2253 } 2254 2255 /* 2256 * The write verifier has changed (probably due to a server reboot), so all 2257 * PG_NEEDCOMMIT pages will have to be written again. Since they are marked 2258 * as dirty or are being written out just now, all this takes is clearing 2259 * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for 2260 * the mount point. 2261 */ 2262 void 2263 nfs_clearcommit(mp) 2264 struct mount *mp; 2265 { 2266 struct vnode *vp; 2267 struct nfsnode *np; 2268 struct vm_page *pg; 2269 struct nfsmount *nmp = VFSTONFS(mp); 2270 2271 rw_enter(&nmp->nm_writeverflock, RW_WRITER); 2272 mutex_enter(&mntvnode_lock); 2273 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2274 KASSERT(vp->v_mount == mp); 2275 if (vp->v_type != VREG) 2276 continue; 2277 np = VTONFS(vp); 2278 np->n_pushlo = np->n_pushhi = np->n_pushedlo = 2279 np->n_pushedhi = 0; 2280 np->n_commitflags &= 2281 ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID); 2282 mutex_enter(&vp->v_uobj.vmobjlock); 2283 TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq.queue) { 2284 pg->flags &= ~PG_NEEDCOMMIT; 2285 } 2286 mutex_exit(&vp->v_uobj.vmobjlock); 2287 } 2288 mutex_exit(&mntvnode_lock); 2289 mutex_enter(&nmp->nm_lock); 2290 nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF; 2291 mutex_exit(&nmp->nm_lock); 2292 rw_exit(&nmp->nm_writeverflock); 2293 } 2294 2295 void 2296 nfs_merge_commit_ranges(vp) 2297 struct vnode *vp; 2298 { 2299 struct nfsnode *np = VTONFS(vp); 2300 2301 KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID); 2302 2303 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2304 np->n_pushedlo = np->n_pushlo; 2305 np->n_pushedhi = np->n_pushhi; 2306 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2307 } else { 2308 if (np->n_pushlo < np->n_pushedlo) 2309 np->n_pushedlo = np->n_pushlo; 2310 if (np->n_pushhi > np->n_pushedhi) 2311 np->n_pushedhi = np->n_pushhi; 2312 } 2313 2314 np->n_pushlo = np->n_pushhi = 0; 2315 np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID; 2316 2317 #ifdef NFS_DEBUG_COMMIT 2318 printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2319 (unsigned)np->n_pushedhi); 2320 #endif 2321 } 2322 2323 int 2324 nfs_in_committed_range(vp, off, len) 2325 struct vnode *vp; 2326 off_t off, len; 2327 { 2328 struct nfsnode *np = VTONFS(vp); 2329 off_t lo, hi; 2330 2331 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2332 return 0; 2333 lo = off; 2334 hi = lo + len; 2335 2336 return (lo >= np->n_pushedlo && hi <= np->n_pushedhi); 2337 } 2338 2339 int 2340 nfs_in_tobecommitted_range(vp, off, len) 2341 struct vnode *vp; 2342 off_t off, len; 2343 { 2344 struct nfsnode *np = VTONFS(vp); 2345 off_t lo, hi; 2346 2347 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2348 return 0; 2349 lo = off; 2350 hi = lo + len; 2351 2352 return (lo >= np->n_pushlo && hi <= np->n_pushhi); 2353 } 2354 2355 void 2356 nfs_add_committed_range(vp, off, len) 2357 struct vnode *vp; 2358 off_t off, len; 2359 { 2360 struct nfsnode *np = VTONFS(vp); 2361 off_t lo, hi; 2362 2363 lo = off; 2364 hi = lo + len; 2365 2366 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2367 np->n_pushedlo = lo; 2368 np->n_pushedhi = hi; 2369 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2370 } else { 2371 if (hi > np->n_pushedhi) 2372 np->n_pushedhi = hi; 2373 if (lo < np->n_pushedlo) 2374 np->n_pushedlo = lo; 2375 } 2376 #ifdef NFS_DEBUG_COMMIT 2377 printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2378 (unsigned)np->n_pushedhi); 2379 #endif 2380 } 2381 2382 void 2383 nfs_del_committed_range(vp, off, len) 2384 struct vnode *vp; 2385 off_t off, len; 2386 { 2387 struct nfsnode *np = VTONFS(vp); 2388 off_t lo, hi; 2389 2390 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2391 return; 2392 2393 lo = off; 2394 hi = lo + len; 2395 2396 if (lo > np->n_pushedhi || hi < np->n_pushedlo) 2397 return; 2398 if (lo <= np->n_pushedlo) 2399 np->n_pushedlo = hi; 2400 else if (hi >= np->n_pushedhi) 2401 np->n_pushedhi = lo; 2402 else { 2403 /* 2404 * XXX There's only one range. If the deleted range 2405 * is in the middle, pick the largest of the 2406 * contiguous ranges that it leaves. 2407 */ 2408 if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi)) 2409 np->n_pushedhi = lo; 2410 else 2411 np->n_pushedlo = hi; 2412 } 2413 #ifdef NFS_DEBUG_COMMIT 2414 printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2415 (unsigned)np->n_pushedhi); 2416 #endif 2417 } 2418 2419 void 2420 nfs_add_tobecommitted_range(vp, off, len) 2421 struct vnode *vp; 2422 off_t off, len; 2423 { 2424 struct nfsnode *np = VTONFS(vp); 2425 off_t lo, hi; 2426 2427 lo = off; 2428 hi = lo + len; 2429 2430 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) { 2431 np->n_pushlo = lo; 2432 np->n_pushhi = hi; 2433 np->n_commitflags |= NFS_COMMIT_PUSH_VALID; 2434 } else { 2435 if (lo < np->n_pushlo) 2436 np->n_pushlo = lo; 2437 if (hi > np->n_pushhi) 2438 np->n_pushhi = hi; 2439 } 2440 #ifdef NFS_DEBUG_COMMIT 2441 printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2442 (unsigned)np->n_pushhi); 2443 #endif 2444 } 2445 2446 void 2447 nfs_del_tobecommitted_range(vp, off, len) 2448 struct vnode *vp; 2449 off_t off, len; 2450 { 2451 struct nfsnode *np = VTONFS(vp); 2452 off_t lo, hi; 2453 2454 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2455 return; 2456 2457 lo = off; 2458 hi = lo + len; 2459 2460 if (lo > np->n_pushhi || hi < np->n_pushlo) 2461 return; 2462 2463 if (lo <= np->n_pushlo) 2464 np->n_pushlo = hi; 2465 else if (hi >= np->n_pushhi) 2466 np->n_pushhi = lo; 2467 else { 2468 /* 2469 * XXX There's only one range. If the deleted range 2470 * is in the middle, pick the largest of the 2471 * contiguous ranges that it leaves. 2472 */ 2473 if ((np->n_pushlo - lo) > (hi - np->n_pushhi)) 2474 np->n_pushhi = lo; 2475 else 2476 np->n_pushlo = hi; 2477 } 2478 #ifdef NFS_DEBUG_COMMIT 2479 printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2480 (unsigned)np->n_pushhi); 2481 #endif 2482 } 2483 2484 /* 2485 * Map errnos to NFS error numbers. For Version 3 also filter out error 2486 * numbers not specified for the associated procedure. 2487 */ 2488 int 2489 nfsrv_errmap(nd, err) 2490 struct nfsrv_descript *nd; 2491 int err; 2492 { 2493 const short *defaulterrp, *errp; 2494 2495 if (nd->nd_flag & ND_NFSV3) { 2496 if (nd->nd_procnum <= NFSPROC_COMMIT) { 2497 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum]; 2498 while (*++errp) { 2499 if (*errp == err) 2500 return (err); 2501 else if (*errp > err) 2502 break; 2503 } 2504 return ((int)*defaulterrp); 2505 } else 2506 return (err & 0xffff); 2507 } 2508 if (err <= ELAST) 2509 return ((int)nfsrv_v2errmap[err - 1]); 2510 return (NFSERR_IO); 2511 } 2512 2513 u_int32_t 2514 nfs_getxid() 2515 { 2516 u_int32_t newxid; 2517 2518 /* get next xid. skip 0 */ 2519 do { 2520 newxid = atomic_inc_32_nv(&nfs_xid); 2521 } while (__predict_false(newxid == 0)); 2522 2523 return txdr_unsigned(newxid); 2524 } 2525 2526 /* 2527 * assign a new xid for existing request. 2528 * used for NFSERR_JUKEBOX handling. 2529 */ 2530 void 2531 nfs_renewxid(struct nfsreq *req) 2532 { 2533 u_int32_t xid; 2534 int off; 2535 2536 xid = nfs_getxid(); 2537 if (req->r_nmp->nm_sotype == SOCK_STREAM) 2538 off = sizeof(u_int32_t); /* RPC record mark */ 2539 else 2540 off = 0; 2541 2542 m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid); 2543 req->r_xid = xid; 2544 } 2545 2546 #if defined(NFS) 2547 /* 2548 * Set the attribute timeout based on how recently the file has been modified. 2549 */ 2550 2551 time_t 2552 nfs_attrtimeo(struct nfsmount *nmp, struct nfsnode *np) 2553 { 2554 time_t timeo; 2555 2556 if ((nmp->nm_flag & NFSMNT_NOAC) != 0) 2557 return 0; 2558 2559 if (((np)->n_flag & NMODIFIED) != 0) 2560 return NFS_MINATTRTIMO; 2561 2562 timeo = (time_second - np->n_mtime.tv_sec) / 10; 2563 timeo = max(timeo, NFS_MINATTRTIMO); 2564 timeo = min(timeo, NFS_MAXATTRTIMO); 2565 return timeo; 2566 } 2567 #endif /* defined(NFS) */ 2568