nfs_subs.c revision 1.215
1 /* $NetBSD: nfs_subs.c,v 1.215 2009/03/14 21:04:25 dsl 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 1.215 2009/03/14 21:04:25 dsl 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(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(struct mbuf **mrep, struct uio *uiop, int siz, char **dpos) 747 { 748 char *mbufcp, *uiocp; 749 int xfer, left, len; 750 struct mbuf *mp; 751 long uiosiz, rem; 752 int error = 0; 753 754 mp = *mrep; 755 mbufcp = *dpos; 756 len = mtod(mp, char *) + mp->m_len - mbufcp; 757 rem = nfsm_rndup(siz)-siz; 758 while (siz > 0) { 759 if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) 760 return (EFBIG); 761 left = uiop->uio_iov->iov_len; 762 uiocp = uiop->uio_iov->iov_base; 763 if (left > siz) 764 left = siz; 765 uiosiz = left; 766 while (left > 0) { 767 while (len == 0) { 768 mp = mp->m_next; 769 if (mp == NULL) 770 return (EBADRPC); 771 mbufcp = mtod(mp, void *); 772 len = mp->m_len; 773 } 774 xfer = (left > len) ? len : left; 775 error = copyout_vmspace(uiop->uio_vmspace, mbufcp, 776 uiocp, xfer); 777 if (error) { 778 return error; 779 } 780 left -= xfer; 781 len -= xfer; 782 mbufcp += xfer; 783 uiocp += xfer; 784 uiop->uio_offset += xfer; 785 uiop->uio_resid -= xfer; 786 } 787 if (uiop->uio_iov->iov_len <= siz) { 788 uiop->uio_iovcnt--; 789 uiop->uio_iov++; 790 } else { 791 uiop->uio_iov->iov_base = 792 (char *)uiop->uio_iov->iov_base + uiosiz; 793 uiop->uio_iov->iov_len -= uiosiz; 794 } 795 siz -= uiosiz; 796 } 797 *dpos = mbufcp; 798 *mrep = mp; 799 if (rem > 0) { 800 if (len < rem) 801 error = nfs_adv(mrep, dpos, rem, len); 802 else 803 *dpos += rem; 804 } 805 return (error); 806 } 807 808 /* 809 * copies a uio scatter/gather list to an mbuf chain. 810 * NOTE: can ony handle iovcnt == 1 811 */ 812 int 813 nfsm_uiotombuf(struct uio *uiop, struct mbuf **mq, int siz, char **bpos) 814 { 815 char *uiocp; 816 struct mbuf *mp, *mp2; 817 int xfer, left, mlen; 818 int uiosiz, clflg, rem; 819 char *cp; 820 int error; 821 822 #ifdef DIAGNOSTIC 823 if (uiop->uio_iovcnt != 1) 824 panic("nfsm_uiotombuf: iovcnt != 1"); 825 #endif 826 827 if (siz > MLEN) /* or should it >= MCLBYTES ?? */ 828 clflg = 1; 829 else 830 clflg = 0; 831 rem = nfsm_rndup(siz)-siz; 832 mp = mp2 = *mq; 833 while (siz > 0) { 834 left = uiop->uio_iov->iov_len; 835 uiocp = uiop->uio_iov->iov_base; 836 if (left > siz) 837 left = siz; 838 uiosiz = left; 839 while (left > 0) { 840 mlen = M_TRAILINGSPACE(mp); 841 if (mlen == 0) { 842 mp = m_get(M_WAIT, MT_DATA); 843 MCLAIM(mp, &nfs_mowner); 844 if (clflg) 845 m_clget(mp, M_WAIT); 846 mp->m_len = 0; 847 mp2->m_next = mp; 848 mp2 = mp; 849 mlen = M_TRAILINGSPACE(mp); 850 } 851 xfer = (left > mlen) ? mlen : left; 852 cp = mtod(mp, char *) + mp->m_len; 853 error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp, 854 xfer); 855 if (error) { 856 /* XXX */ 857 } 858 mp->m_len += xfer; 859 left -= xfer; 860 uiocp += xfer; 861 uiop->uio_offset += xfer; 862 uiop->uio_resid -= xfer; 863 } 864 uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + 865 uiosiz; 866 uiop->uio_iov->iov_len -= uiosiz; 867 siz -= uiosiz; 868 } 869 if (rem > 0) { 870 if (rem > M_TRAILINGSPACE(mp)) { 871 mp = m_get(M_WAIT, MT_DATA); 872 MCLAIM(mp, &nfs_mowner); 873 mp->m_len = 0; 874 mp2->m_next = mp; 875 } 876 cp = mtod(mp, char *) + mp->m_len; 877 for (left = 0; left < rem; left++) 878 *cp++ = '\0'; 879 mp->m_len += rem; 880 *bpos = cp; 881 } else 882 *bpos = mtod(mp, char *) + mp->m_len; 883 *mq = mp; 884 return (0); 885 } 886 887 /* 888 * Get at least "siz" bytes of correctly aligned data. 889 * When called the mbuf pointers are not necessarily correct, 890 * dsosp points to what ought to be in m_data and left contains 891 * what ought to be in m_len. 892 * This is used by the macros nfsm_dissect and nfsm_dissecton for tough 893 * cases. (The macros use the vars. dpos and dpos2) 894 */ 895 int 896 nfsm_disct(struct mbuf **mdp, char **dposp, int siz, int left, char **cp2) 897 { 898 struct mbuf *m1, *m2; 899 struct mbuf *havebuf = NULL; 900 char *src = *dposp; 901 char *dst; 902 int len; 903 904 #ifdef DEBUG 905 if (left < 0) 906 panic("nfsm_disct: left < 0"); 907 #endif 908 m1 = *mdp; 909 /* 910 * Skip through the mbuf chain looking for an mbuf with 911 * some data. If the first mbuf found has enough data 912 * and it is correctly aligned return it. 913 */ 914 while (left == 0) { 915 havebuf = m1; 916 *mdp = m1 = m1->m_next; 917 if (m1 == NULL) 918 return (EBADRPC); 919 src = mtod(m1, void *); 920 left = m1->m_len; 921 /* 922 * If we start a new mbuf and it is big enough 923 * and correctly aligned just return it, don't 924 * do any pull up. 925 */ 926 if (left >= siz && nfsm_aligned(src)) { 927 *cp2 = src; 928 *dposp = src + siz; 929 return (0); 930 } 931 } 932 if ((m1->m_flags & M_EXT) != 0) { 933 if (havebuf && M_TRAILINGSPACE(havebuf) >= siz && 934 nfsm_aligned(mtod(havebuf, char *) + havebuf->m_len)) { 935 /* 936 * If the first mbuf with data has external data 937 * and there is a previous mbuf with some trailing 938 * space, use it to move the data into. 939 */ 940 m2 = m1; 941 *mdp = m1 = havebuf; 942 *cp2 = mtod(m1, char *) + m1->m_len; 943 } else if (havebuf) { 944 /* 945 * If the first mbuf has a external data 946 * and there is no previous empty mbuf 947 * allocate a new mbuf and move the external 948 * data to the new mbuf. Also make the first 949 * mbuf look empty. 950 */ 951 m2 = m1; 952 *mdp = m1 = m_get(M_WAIT, MT_DATA); 953 MCLAIM(m1, m2->m_owner); 954 if ((m2->m_flags & M_PKTHDR) != 0) { 955 /* XXX MOVE */ 956 M_COPY_PKTHDR(m1, m2); 957 m_tag_delete_chain(m2, NULL); 958 m2->m_flags &= ~M_PKTHDR; 959 } 960 if (havebuf) { 961 havebuf->m_next = m1; 962 } 963 m1->m_next = m2; 964 MRESETDATA(m1); 965 m1->m_len = 0; 966 m2->m_data = src; 967 m2->m_len = left; 968 *cp2 = mtod(m1, char *); 969 } else { 970 struct mbuf **nextp = &m1->m_next; 971 972 m1->m_len -= left; 973 do { 974 m2 = m_get(M_WAIT, MT_DATA); 975 MCLAIM(m2, m1->m_owner); 976 if (left >= MINCLSIZE) { 977 MCLGET(m2, M_WAIT); 978 } 979 m2->m_next = *nextp; 980 *nextp = m2; 981 nextp = &m2->m_next; 982 len = (m2->m_flags & M_EXT) != 0 ? 983 MCLBYTES : MLEN; 984 if (len > left) { 985 len = left; 986 } 987 memcpy(mtod(m2, char *), src, len); 988 m2->m_len = len; 989 src += len; 990 left -= len; 991 } while (left > 0); 992 *mdp = m1 = m1->m_next; 993 m2 = m1->m_next; 994 *cp2 = mtod(m1, char *); 995 } 996 } else { 997 /* 998 * If the first mbuf has no external data 999 * move the data to the front of the mbuf. 1000 */ 1001 MRESETDATA(m1); 1002 dst = mtod(m1, char *); 1003 if (dst != src) { 1004 memmove(dst, src, left); 1005 } 1006 m1->m_len = left; 1007 m2 = m1->m_next; 1008 *cp2 = m1->m_data; 1009 } 1010 *dposp = *cp2 + siz; 1011 /* 1012 * Loop through mbufs pulling data up into first mbuf until 1013 * the first mbuf is full or there is no more data to 1014 * pullup. 1015 */ 1016 dst = mtod(m1, char *) + m1->m_len; 1017 while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) { 1018 if ((len = min(len, m2->m_len)) != 0) { 1019 memcpy(dst, mtod(m2, char *), len); 1020 } 1021 m1->m_len += len; 1022 dst += len; 1023 m2->m_data += len; 1024 m2->m_len -= len; 1025 m2 = m2->m_next; 1026 } 1027 if (m1->m_len < siz) 1028 return (EBADRPC); 1029 return (0); 1030 } 1031 1032 /* 1033 * Advance the position in the mbuf chain. 1034 */ 1035 int 1036 nfs_adv(struct mbuf **mdp, char **dposp, int offs, int left) 1037 { 1038 struct mbuf *m; 1039 int s; 1040 1041 m = *mdp; 1042 s = left; 1043 while (s < offs) { 1044 offs -= s; 1045 m = m->m_next; 1046 if (m == NULL) 1047 return (EBADRPC); 1048 s = m->m_len; 1049 } 1050 *mdp = m; 1051 *dposp = mtod(m, char *) + offs; 1052 return (0); 1053 } 1054 1055 /* 1056 * Copy a string into mbufs for the hard cases... 1057 */ 1058 int 1059 nfsm_strtmbuf(struct mbuf **mb, char **bpos, const char *cp, long siz) 1060 { 1061 struct mbuf *m1 = NULL, *m2; 1062 long left, xfer, len, tlen; 1063 u_int32_t *tl; 1064 int putsize; 1065 1066 putsize = 1; 1067 m2 = *mb; 1068 left = M_TRAILINGSPACE(m2); 1069 if (left > 0) { 1070 tl = ((u_int32_t *)(*bpos)); 1071 *tl++ = txdr_unsigned(siz); 1072 putsize = 0; 1073 left -= NFSX_UNSIGNED; 1074 m2->m_len += NFSX_UNSIGNED; 1075 if (left > 0) { 1076 memcpy((void *) tl, cp, left); 1077 siz -= left; 1078 cp += left; 1079 m2->m_len += left; 1080 left = 0; 1081 } 1082 } 1083 /* Loop around adding mbufs */ 1084 while (siz > 0) { 1085 m1 = m_get(M_WAIT, MT_DATA); 1086 MCLAIM(m1, &nfs_mowner); 1087 if (siz > MLEN) 1088 m_clget(m1, M_WAIT); 1089 m1->m_len = NFSMSIZ(m1); 1090 m2->m_next = m1; 1091 m2 = m1; 1092 tl = mtod(m1, u_int32_t *); 1093 tlen = 0; 1094 if (putsize) { 1095 *tl++ = txdr_unsigned(siz); 1096 m1->m_len -= NFSX_UNSIGNED; 1097 tlen = NFSX_UNSIGNED; 1098 putsize = 0; 1099 } 1100 if (siz < m1->m_len) { 1101 len = nfsm_rndup(siz); 1102 xfer = siz; 1103 if (xfer < len) 1104 *(tl+(xfer>>2)) = 0; 1105 } else { 1106 xfer = len = m1->m_len; 1107 } 1108 memcpy((void *) tl, cp, xfer); 1109 m1->m_len = len+tlen; 1110 siz -= xfer; 1111 cp += xfer; 1112 } 1113 *mb = m1; 1114 *bpos = mtod(m1, char *) + m1->m_len; 1115 return (0); 1116 } 1117 1118 /* 1119 * Directory caching routines. They work as follows: 1120 * - a cache is maintained per VDIR nfsnode. 1121 * - for each offset cookie that is exported to userspace, and can 1122 * thus be thrown back at us as an offset to VOP_READDIR, store 1123 * information in the cache. 1124 * - cached are: 1125 * - cookie itself 1126 * - blocknumber (essentially just a search key in the buffer cache) 1127 * - entry number in block. 1128 * - offset cookie of block in which this entry is stored 1129 * - 32 bit cookie if NFSMNT_XLATECOOKIE is used. 1130 * - entries are looked up in a hash table 1131 * - also maintained is an LRU list of entries, used to determine 1132 * which ones to delete if the cache grows too large. 1133 * - if 32 <-> 64 translation mode is requested for a filesystem, 1134 * the cache also functions as a translation table 1135 * - in the translation case, invalidating the cache does not mean 1136 * flushing it, but just marking entries as invalid, except for 1137 * the <64bit cookie, 32bitcookie> pair which is still valid, to 1138 * still be able to use the cache as a translation table. 1139 * - 32 bit cookies are uniquely created by combining the hash table 1140 * entry value, and one generation count per hash table entry, 1141 * incremented each time an entry is appended to the chain. 1142 * - the cache is invalidated each time a direcory is modified 1143 * - sanity checks are also done; if an entry in a block turns 1144 * out not to have a matching cookie, the cache is invalidated 1145 * and a new block starting from the wanted offset is fetched from 1146 * the server. 1147 * - directory entries as read from the server are extended to contain 1148 * the 64bit and, optionally, the 32bit cookies, for sanity checking 1149 * the cache and exporting them to userspace through the cookie 1150 * argument to VOP_READDIR. 1151 */ 1152 1153 u_long 1154 nfs_dirhash(off_t off) 1155 { 1156 int i; 1157 char *cp = (char *)&off; 1158 u_long sum = 0L; 1159 1160 for (i = 0 ; i < sizeof (off); i++) 1161 sum += *cp++; 1162 1163 return sum; 1164 } 1165 1166 #define _NFSDC_MTX(np) (&NFSTOV(np)->v_interlock) 1167 #define NFSDC_LOCK(np) mutex_enter(_NFSDC_MTX(np)) 1168 #define NFSDC_UNLOCK(np) mutex_exit(_NFSDC_MTX(np)) 1169 #define NFSDC_ASSERT_LOCKED(np) KASSERT(mutex_owned(_NFSDC_MTX(np))) 1170 1171 void 1172 nfs_initdircache(struct vnode *vp) 1173 { 1174 struct nfsnode *np = VTONFS(vp); 1175 struct nfsdirhashhead *dircache; 1176 1177 dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, true, 1178 &nfsdirhashmask); 1179 1180 NFSDC_LOCK(np); 1181 if (np->n_dircache == NULL) { 1182 np->n_dircachesize = 0; 1183 np->n_dircache = dircache; 1184 dircache = NULL; 1185 TAILQ_INIT(&np->n_dirchain); 1186 } 1187 NFSDC_UNLOCK(np); 1188 if (dircache) 1189 hashdone(dircache, HASH_LIST, nfsdirhashmask); 1190 } 1191 1192 void 1193 nfs_initdirxlatecookie(struct vnode *vp) 1194 { 1195 struct nfsnode *np = VTONFS(vp); 1196 unsigned *dirgens; 1197 1198 KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE); 1199 1200 dirgens = kmem_zalloc(NFS_DIRHASHSIZ * sizeof(unsigned), KM_SLEEP); 1201 NFSDC_LOCK(np); 1202 if (np->n_dirgens == NULL) { 1203 np->n_dirgens = dirgens; 1204 dirgens = NULL; 1205 } 1206 NFSDC_UNLOCK(np); 1207 if (dirgens) 1208 kmem_free(dirgens, NFS_DIRHASHSIZ * sizeof(unsigned)); 1209 } 1210 1211 static const struct nfsdircache dzero; 1212 1213 static void nfs_unlinkdircache(struct nfsnode *np, struct nfsdircache *); 1214 static void nfs_putdircache_unlocked(struct nfsnode *, 1215 struct nfsdircache *); 1216 1217 static void 1218 nfs_unlinkdircache(struct nfsnode *np, struct nfsdircache *ndp) 1219 { 1220 1221 NFSDC_ASSERT_LOCKED(np); 1222 KASSERT(ndp != &dzero); 1223 1224 if (LIST_NEXT(ndp, dc_hash) == (void *)-1) 1225 return; 1226 1227 TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain); 1228 LIST_REMOVE(ndp, dc_hash); 1229 LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */ 1230 1231 nfs_putdircache_unlocked(np, ndp); 1232 } 1233 1234 void 1235 nfs_putdircache(struct nfsnode *np, struct nfsdircache *ndp) 1236 { 1237 int ref; 1238 1239 if (ndp == &dzero) 1240 return; 1241 1242 KASSERT(ndp->dc_refcnt > 0); 1243 NFSDC_LOCK(np); 1244 ref = --ndp->dc_refcnt; 1245 NFSDC_UNLOCK(np); 1246 1247 if (ref == 0) 1248 kmem_free(ndp, sizeof(*ndp)); 1249 } 1250 1251 static void 1252 nfs_putdircache_unlocked(struct nfsnode *np, struct nfsdircache *ndp) 1253 { 1254 int ref; 1255 1256 NFSDC_ASSERT_LOCKED(np); 1257 1258 if (ndp == &dzero) 1259 return; 1260 1261 KASSERT(ndp->dc_refcnt > 0); 1262 ref = --ndp->dc_refcnt; 1263 if (ref == 0) 1264 kmem_free(ndp, sizeof(*ndp)); 1265 } 1266 1267 struct nfsdircache * 1268 nfs_searchdircache(struct vnode *vp, off_t off, int do32, int *hashent) 1269 { 1270 struct nfsdirhashhead *ndhp; 1271 struct nfsdircache *ndp = NULL; 1272 struct nfsnode *np = VTONFS(vp); 1273 unsigned ent; 1274 1275 /* 1276 * Zero is always a valid cookie. 1277 */ 1278 if (off == 0) 1279 /* XXXUNCONST */ 1280 return (struct nfsdircache *)__UNCONST(&dzero); 1281 1282 if (!np->n_dircache) 1283 return NULL; 1284 1285 /* 1286 * We use a 32bit cookie as search key, directly reconstruct 1287 * the hashentry. Else use the hashfunction. 1288 */ 1289 if (do32) { 1290 ent = (u_int32_t)off >> 24; 1291 if (ent >= NFS_DIRHASHSIZ) 1292 return NULL; 1293 ndhp = &np->n_dircache[ent]; 1294 } else { 1295 ndhp = NFSDIRHASH(np, off); 1296 } 1297 1298 if (hashent) 1299 *hashent = (int)(ndhp - np->n_dircache); 1300 1301 NFSDC_LOCK(np); 1302 if (do32) { 1303 LIST_FOREACH(ndp, ndhp, dc_hash) { 1304 if (ndp->dc_cookie32 == (u_int32_t)off) { 1305 /* 1306 * An invalidated entry will become the 1307 * start of a new block fetched from 1308 * the server. 1309 */ 1310 if (ndp->dc_flags & NFSDC_INVALID) { 1311 ndp->dc_blkcookie = ndp->dc_cookie; 1312 ndp->dc_entry = 0; 1313 ndp->dc_flags &= ~NFSDC_INVALID; 1314 } 1315 break; 1316 } 1317 } 1318 } else { 1319 LIST_FOREACH(ndp, ndhp, dc_hash) { 1320 if (ndp->dc_cookie == off) 1321 break; 1322 } 1323 } 1324 if (ndp != NULL) 1325 ndp->dc_refcnt++; 1326 NFSDC_UNLOCK(np); 1327 return ndp; 1328 } 1329 1330 1331 struct nfsdircache * 1332 nfs_enterdircache(struct vnode *vp, off_t off, off_t blkoff, int en, 1333 daddr_t blkno) 1334 { 1335 struct nfsnode *np = VTONFS(vp); 1336 struct nfsdirhashhead *ndhp; 1337 struct nfsdircache *ndp = NULL; 1338 struct nfsdircache *newndp = NULL; 1339 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1340 int hashent = 0, gen, overwrite; /* XXX: GCC */ 1341 1342 /* 1343 * XXX refuse entries for offset 0. amd(8) erroneously sets 1344 * cookie 0 for the '.' entry, making this necessary. This 1345 * isn't so bad, as 0 is a special case anyway. 1346 */ 1347 if (off == 0) 1348 /* XXXUNCONST */ 1349 return (struct nfsdircache *)__UNCONST(&dzero); 1350 1351 if (!np->n_dircache) 1352 /* 1353 * XXX would like to do this in nfs_nget but vtype 1354 * isn't known at that time. 1355 */ 1356 nfs_initdircache(vp); 1357 1358 if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens) 1359 nfs_initdirxlatecookie(vp); 1360 1361 retry: 1362 ndp = nfs_searchdircache(vp, off, 0, &hashent); 1363 1364 NFSDC_LOCK(np); 1365 if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) { 1366 /* 1367 * Overwriting an old entry. Check if it's the same. 1368 * If so, just return. If not, remove the old entry. 1369 */ 1370 if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en) 1371 goto done; 1372 nfs_unlinkdircache(np, ndp); 1373 nfs_putdircache_unlocked(np, ndp); 1374 ndp = NULL; 1375 } 1376 1377 ndhp = &np->n_dircache[hashent]; 1378 1379 if (!ndp) { 1380 if (newndp == NULL) { 1381 NFSDC_UNLOCK(np); 1382 newndp = kmem_alloc(sizeof(*newndp), KM_SLEEP); 1383 newndp->dc_refcnt = 1; 1384 LIST_NEXT(newndp, dc_hash) = (void *)-1; 1385 goto retry; 1386 } 1387 ndp = newndp; 1388 newndp = NULL; 1389 overwrite = 0; 1390 if (nmp->nm_flag & NFSMNT_XLATECOOKIE) { 1391 /* 1392 * We're allocating a new entry, so bump the 1393 * generation number. 1394 */ 1395 KASSERT(np->n_dirgens); 1396 gen = ++np->n_dirgens[hashent]; 1397 if (gen == 0) { 1398 np->n_dirgens[hashent]++; 1399 gen++; 1400 } 1401 ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff); 1402 } 1403 } else 1404 overwrite = 1; 1405 1406 ndp->dc_cookie = off; 1407 ndp->dc_blkcookie = blkoff; 1408 ndp->dc_entry = en; 1409 ndp->dc_flags = 0; 1410 1411 if (overwrite) 1412 goto done; 1413 1414 /* 1415 * If the maximum directory cookie cache size has been reached 1416 * for this node, take one off the front. The idea is that 1417 * directories are typically read front-to-back once, so that 1418 * the oldest entries can be thrown away without much performance 1419 * loss. 1420 */ 1421 if (np->n_dircachesize == NFS_MAXDIRCACHE) { 1422 nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain)); 1423 } else 1424 np->n_dircachesize++; 1425 1426 KASSERT(ndp->dc_refcnt == 1); 1427 LIST_INSERT_HEAD(ndhp, ndp, dc_hash); 1428 TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain); 1429 ndp->dc_refcnt++; 1430 done: 1431 KASSERT(ndp->dc_refcnt > 0); 1432 NFSDC_UNLOCK(np); 1433 if (newndp) 1434 nfs_putdircache(np, newndp); 1435 return ndp; 1436 } 1437 1438 void 1439 nfs_invaldircache(struct vnode *vp, int flags) 1440 { 1441 struct nfsnode *np = VTONFS(vp); 1442 struct nfsdircache *ndp = NULL; 1443 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1444 const bool forcefree = flags & NFS_INVALDIRCACHE_FORCE; 1445 1446 #ifdef DIAGNOSTIC 1447 if (vp->v_type != VDIR) 1448 panic("nfs: invaldircache: not dir"); 1449 #endif 1450 1451 if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0) 1452 np->n_flag &= ~NEOFVALID; 1453 1454 if (!np->n_dircache) 1455 return; 1456 1457 NFSDC_LOCK(np); 1458 if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) { 1459 while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) { 1460 KASSERT(!forcefree || ndp->dc_refcnt == 1); 1461 nfs_unlinkdircache(np, ndp); 1462 } 1463 np->n_dircachesize = 0; 1464 if (forcefree && np->n_dirgens) { 1465 kmem_free(np->n_dirgens, 1466 NFS_DIRHASHSIZ * sizeof(unsigned)); 1467 np->n_dirgens = NULL; 1468 } 1469 } else { 1470 TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain) 1471 ndp->dc_flags |= NFSDC_INVALID; 1472 } 1473 1474 NFSDC_UNLOCK(np); 1475 } 1476 1477 /* 1478 * Called once before VFS init to initialize shared and 1479 * server-specific data structures. 1480 */ 1481 static int 1482 nfs_init0(void) 1483 { 1484 1485 nfsrtt.pos = 0; 1486 rpc_vers = txdr_unsigned(RPC_VER2); 1487 rpc_call = txdr_unsigned(RPC_CALL); 1488 rpc_reply = txdr_unsigned(RPC_REPLY); 1489 rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED); 1490 rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED); 1491 rpc_mismatch = txdr_unsigned(RPC_MISMATCH); 1492 rpc_autherr = txdr_unsigned(RPC_AUTHERR); 1493 rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX); 1494 rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4); 1495 nfs_prog = txdr_unsigned(NFS_PROG); 1496 nfs_true = txdr_unsigned(true); 1497 nfs_false = txdr_unsigned(false); 1498 nfs_xdrneg1 = txdr_unsigned(-1); 1499 nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000; 1500 if (nfs_ticks < 1) 1501 nfs_ticks = 1; 1502 nfs_xid = arc4random(); 1503 nfsdreq_init(); 1504 1505 /* 1506 * Initialize reply list and start timer 1507 */ 1508 TAILQ_INIT(&nfs_reqq); 1509 nfs_timer_init(); 1510 MOWNER_ATTACH(&nfs_mowner); 1511 1512 #ifdef NFS 1513 /* Initialize the kqueue structures */ 1514 nfs_kqinit(); 1515 /* Initialize the iod structures */ 1516 nfs_iodinit(); 1517 #endif 1518 1519 return 0; 1520 } 1521 1522 /* 1523 * This is disgusting, but it must support both modular and monolothic 1524 * configurations. For monolithic builds NFSSERVER may not imply NFS. 1525 * 1526 * Yuck. 1527 */ 1528 void 1529 nfs_init(void) 1530 { 1531 static ONCE_DECL(nfs_init_once); 1532 1533 RUN_ONCE(&nfs_init_once, nfs_init0); 1534 } 1535 1536 void 1537 nfs_fini(void) 1538 { 1539 1540 #ifdef NFS 1541 nfs_kqfini(); 1542 nfs_iodfini(); 1543 #endif 1544 nfsdreq_fini(); 1545 nfs_timer_fini(); 1546 MOWNER_DETACH(&nfs_mowner); 1547 } 1548 1549 #ifdef NFS 1550 /* 1551 * Called once at VFS init to initialize client-specific data structures. 1552 */ 1553 void 1554 nfs_vfs_init() 1555 { 1556 1557 /* Initialize NFS server / client shared data. */ 1558 nfs_init(); 1559 nfs_node_init(); 1560 1561 nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4); 1562 } 1563 1564 void 1565 nfs_vfs_done() 1566 { 1567 1568 nfs_node_done(); 1569 } 1570 1571 /* 1572 * Attribute cache routines. 1573 * nfs_loadattrcache() - loads or updates the cache contents from attributes 1574 * that are on the mbuf list 1575 * nfs_getattrcache() - returns valid attributes if found in cache, returns 1576 * error otherwise 1577 */ 1578 1579 /* 1580 * Load the attribute cache (that lives in the nfsnode entry) with 1581 * the values on the mbuf list and 1582 * Iff vap not NULL 1583 * copy the attributes to *vaper 1584 */ 1585 int 1586 nfsm_loadattrcache(struct vnode **vpp, struct mbuf **mdp, char **dposp, struct vattr *vaper, int flags) 1587 { 1588 int32_t t1; 1589 char *cp2; 1590 int error = 0; 1591 struct mbuf *md; 1592 int v3 = NFS_ISV3(*vpp); 1593 1594 md = *mdp; 1595 t1 = (mtod(md, char *) + md->m_len) - *dposp; 1596 error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2); 1597 if (error) 1598 return (error); 1599 return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags); 1600 } 1601 1602 int 1603 nfs_loadattrcache(struct vnode **vpp, struct nfs_fattr *fp, struct vattr *vaper, int flags) 1604 { 1605 struct vnode *vp = *vpp; 1606 struct vattr *vap; 1607 int v3 = NFS_ISV3(vp); 1608 enum vtype vtyp; 1609 u_short vmode; 1610 struct timespec mtime; 1611 struct timespec ctime; 1612 int32_t rdev; 1613 struct nfsnode *np; 1614 extern int (**spec_nfsv2nodeop_p)(void *); 1615 uid_t uid; 1616 gid_t gid; 1617 1618 if (v3) { 1619 vtyp = nfsv3tov_type(fp->fa_type); 1620 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1621 rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1), 1622 fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2)); 1623 fxdr_nfsv3time(&fp->fa3_mtime, &mtime); 1624 fxdr_nfsv3time(&fp->fa3_ctime, &ctime); 1625 } else { 1626 vtyp = nfsv2tov_type(fp->fa_type); 1627 vmode = fxdr_unsigned(u_short, fp->fa_mode); 1628 if (vtyp == VNON || vtyp == VREG) 1629 vtyp = IFTOVT(vmode); 1630 rdev = fxdr_unsigned(int32_t, fp->fa2_rdev); 1631 fxdr_nfsv2time(&fp->fa2_mtime, &mtime); 1632 ctime.tv_sec = fxdr_unsigned(u_int32_t, 1633 fp->fa2_ctime.nfsv2_sec); 1634 ctime.tv_nsec = 0; 1635 1636 /* 1637 * Really ugly NFSv2 kludge. 1638 */ 1639 if (vtyp == VCHR && rdev == 0xffffffff) 1640 vtyp = VFIFO; 1641 } 1642 1643 vmode &= ALLPERMS; 1644 1645 /* 1646 * If v_type == VNON it is a new node, so fill in the v_type, 1647 * n_mtime fields. Check to see if it represents a special 1648 * device, and if so, check for a possible alias. Once the 1649 * correct vnode has been obtained, fill in the rest of the 1650 * information. 1651 */ 1652 np = VTONFS(vp); 1653 if (vp->v_type == VNON) { 1654 vp->v_type = vtyp; 1655 if (vp->v_type == VFIFO) { 1656 extern int (**fifo_nfsv2nodeop_p)(void *); 1657 vp->v_op = fifo_nfsv2nodeop_p; 1658 } else if (vp->v_type == VREG) { 1659 mutex_init(&np->n_commitlock, MUTEX_DEFAULT, IPL_NONE); 1660 } else if (vp->v_type == VCHR || vp->v_type == VBLK) { 1661 vp->v_op = spec_nfsv2nodeop_p; 1662 spec_node_init(vp, (dev_t)rdev); 1663 } 1664 np->n_mtime = mtime; 1665 } 1666 uid = fxdr_unsigned(uid_t, fp->fa_uid); 1667 gid = fxdr_unsigned(gid_t, fp->fa_gid); 1668 vap = np->n_vattr; 1669 1670 /* 1671 * Invalidate access cache if uid, gid, mode or ctime changed. 1672 */ 1673 if (np->n_accstamp != -1 && 1674 (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode 1675 || timespeccmp(&ctime, &vap->va_ctime, !=))) 1676 np->n_accstamp = -1; 1677 1678 vap->va_type = vtyp; 1679 vap->va_mode = vmode; 1680 vap->va_rdev = (dev_t)rdev; 1681 vap->va_mtime = mtime; 1682 vap->va_ctime = ctime; 1683 vap->va_birthtime.tv_sec = VNOVAL; 1684 vap->va_birthtime.tv_nsec = VNOVAL; 1685 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 1686 switch (vtyp) { 1687 case VDIR: 1688 vap->va_blocksize = NFS_DIRFRAGSIZ; 1689 break; 1690 case VBLK: 1691 vap->va_blocksize = BLKDEV_IOSIZE; 1692 break; 1693 case VCHR: 1694 vap->va_blocksize = MAXBSIZE; 1695 break; 1696 default: 1697 vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize : 1698 fxdr_unsigned(int32_t, fp->fa2_blocksize); 1699 break; 1700 } 1701 if (v3) { 1702 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1703 vap->va_uid = uid; 1704 vap->va_gid = gid; 1705 vap->va_size = fxdr_hyper(&fp->fa3_size); 1706 vap->va_bytes = fxdr_hyper(&fp->fa3_used); 1707 vap->va_fileid = fxdr_hyper(&fp->fa3_fileid); 1708 fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime); 1709 vap->va_flags = 0; 1710 vap->va_filerev = 0; 1711 } else { 1712 vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink); 1713 vap->va_uid = uid; 1714 vap->va_gid = gid; 1715 vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size); 1716 vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks) 1717 * NFS_FABLKSIZE; 1718 vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid); 1719 fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime); 1720 vap->va_flags = 0; 1721 vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec); 1722 vap->va_filerev = 0; 1723 } 1724 if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) { 1725 return EFBIG; 1726 } 1727 if (vap->va_size != np->n_size) { 1728 if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) { 1729 vap->va_size = np->n_size; 1730 } else { 1731 np->n_size = vap->va_size; 1732 if (vap->va_type == VREG) { 1733 /* 1734 * we can't free pages if NAC_NOTRUNC because 1735 * the pages can be owned by ourselves. 1736 */ 1737 if (flags & NAC_NOTRUNC) { 1738 np->n_flag |= NTRUNCDELAYED; 1739 } else { 1740 genfs_node_wrlock(vp); 1741 mutex_enter(&vp->v_interlock); 1742 (void)VOP_PUTPAGES(vp, 0, 1743 0, PGO_SYNCIO | PGO_CLEANIT | 1744 PGO_FREE | PGO_ALLPAGES); 1745 uvm_vnp_setsize(vp, np->n_size); 1746 genfs_node_unlock(vp); 1747 } 1748 } 1749 } 1750 } 1751 np->n_attrstamp = time_second; 1752 if (vaper != NULL) { 1753 memcpy((void *)vaper, (void *)vap, sizeof(*vap)); 1754 if (np->n_flag & NCHG) { 1755 if (np->n_flag & NACC) 1756 vaper->va_atime = np->n_atim; 1757 if (np->n_flag & NUPD) 1758 vaper->va_mtime = np->n_mtim; 1759 } 1760 } 1761 return (0); 1762 } 1763 1764 /* 1765 * Check the time stamp 1766 * If the cache is valid, copy contents to *vap and return 0 1767 * otherwise return an error 1768 */ 1769 int 1770 nfs_getattrcache(struct vnode *vp, struct vattr *vaper) 1771 { 1772 struct nfsnode *np = VTONFS(vp); 1773 struct nfsmount *nmp = VFSTONFS(vp->v_mount); 1774 struct vattr *vap; 1775 1776 if (np->n_attrstamp == 0 || 1777 (time_second - np->n_attrstamp) >= nfs_attrtimeo(nmp, np)) { 1778 nfsstats.attrcache_misses++; 1779 return (ENOENT); 1780 } 1781 nfsstats.attrcache_hits++; 1782 vap = np->n_vattr; 1783 if (vap->va_size != np->n_size) { 1784 if (vap->va_type == VREG) { 1785 if ((np->n_flag & NMODIFIED) != 0 && 1786 vap->va_size < np->n_size) { 1787 vap->va_size = np->n_size; 1788 } else { 1789 np->n_size = vap->va_size; 1790 } 1791 genfs_node_wrlock(vp); 1792 uvm_vnp_setsize(vp, np->n_size); 1793 genfs_node_unlock(vp); 1794 } else 1795 np->n_size = vap->va_size; 1796 } 1797 memcpy((void *)vaper, (void *)vap, sizeof(struct vattr)); 1798 if (np->n_flag & NCHG) { 1799 if (np->n_flag & NACC) 1800 vaper->va_atime = np->n_atim; 1801 if (np->n_flag & NUPD) 1802 vaper->va_mtime = np->n_mtim; 1803 } 1804 return (0); 1805 } 1806 1807 void 1808 nfs_delayedtruncate(struct vnode *vp) 1809 { 1810 struct nfsnode *np = VTONFS(vp); 1811 1812 if (np->n_flag & NTRUNCDELAYED) { 1813 np->n_flag &= ~NTRUNCDELAYED; 1814 genfs_node_wrlock(vp); 1815 mutex_enter(&vp->v_interlock); 1816 (void)VOP_PUTPAGES(vp, 0, 1817 0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES); 1818 uvm_vnp_setsize(vp, np->n_size); 1819 genfs_node_unlock(vp); 1820 } 1821 } 1822 1823 #define NFS_WCCKLUDGE_TIMEOUT (24 * 60 * 60) /* 1 day */ 1824 #define NFS_WCCKLUDGE(nmp, now) \ 1825 (((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \ 1826 ((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0) 1827 1828 /* 1829 * nfs_check_wccdata: check inaccurate wcc_data 1830 * 1831 * => return non-zero if we shouldn't trust the wcc_data. 1832 * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed". 1833 */ 1834 1835 int 1836 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime, 1837 struct timespec *mtime, bool docheck) 1838 { 1839 int error = 0; 1840 1841 #if !defined(NFS_V2_ONLY) 1842 1843 if (docheck) { 1844 struct vnode *vp = NFSTOV(np); 1845 struct nfsmount *nmp; 1846 long now = time_second; 1847 const struct timespec *omtime = &np->n_vattr->va_mtime; 1848 const struct timespec *octime = &np->n_vattr->va_ctime; 1849 const char *reason = NULL; /* XXX: gcc */ 1850 1851 if (timespeccmp(omtime, mtime, <=)) { 1852 reason = "mtime"; 1853 error = EINVAL; 1854 } 1855 1856 if (vp->v_type == VDIR && timespeccmp(octime, ctime, <=)) { 1857 reason = "ctime"; 1858 error = EINVAL; 1859 } 1860 1861 nmp = VFSTONFS(vp->v_mount); 1862 if (error) { 1863 1864 /* 1865 * despite of the fact that we've updated the file, 1866 * timestamps of the file were not updated as we 1867 * expected. 1868 * it means that the server has incompatible 1869 * semantics of timestamps or (more likely) 1870 * the server time is not precise enough to 1871 * track each modifications. 1872 * in that case, we disable wcc processing. 1873 * 1874 * yes, strictly speaking, we should disable all 1875 * caching. it's a compromise. 1876 */ 1877 1878 mutex_enter(&nmp->nm_lock); 1879 if (!NFS_WCCKLUDGE(nmp, now)) { 1880 printf("%s: inaccurate wcc data (%s) detected," 1881 " disabling wcc" 1882 " (ctime %u.%09u %u.%09u," 1883 " mtime %u.%09u %u.%09u)\n", 1884 vp->v_mount->mnt_stat.f_mntfromname, 1885 reason, 1886 (unsigned int)octime->tv_sec, 1887 (unsigned int)octime->tv_nsec, 1888 (unsigned int)ctime->tv_sec, 1889 (unsigned int)ctime->tv_nsec, 1890 (unsigned int)omtime->tv_sec, 1891 (unsigned int)omtime->tv_nsec, 1892 (unsigned int)mtime->tv_sec, 1893 (unsigned int)mtime->tv_nsec); 1894 } 1895 nmp->nm_iflag |= NFSMNT_WCCKLUDGE; 1896 nmp->nm_wcckludgetime = now; 1897 mutex_exit(&nmp->nm_lock); 1898 } else if (NFS_WCCKLUDGE(nmp, now)) { 1899 error = EPERM; /* XXX */ 1900 } else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1901 mutex_enter(&nmp->nm_lock); 1902 if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) { 1903 printf("%s: re-enabling wcc\n", 1904 vp->v_mount->mnt_stat.f_mntfromname); 1905 nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE; 1906 } 1907 mutex_exit(&nmp->nm_lock); 1908 } 1909 } 1910 1911 #endif /* !defined(NFS_V2_ONLY) */ 1912 1913 return error; 1914 } 1915 1916 /* 1917 * Heuristic to see if the server XDR encodes directory cookies or not. 1918 * it is not supposed to, but a lot of servers may do this. Also, since 1919 * most/all servers will implement V2 as well, it is expected that they 1920 * may return just 32 bits worth of cookie information, so we need to 1921 * find out in which 32 bits this information is available. We do this 1922 * to avoid trouble with emulated binaries that can't handle 64 bit 1923 * directory offsets. 1924 */ 1925 1926 void 1927 nfs_cookieheuristic(struct vnode *vp, int *flagp, struct lwp *l, kauth_cred_t cred) 1928 { 1929 struct uio auio; 1930 struct iovec aiov; 1931 char *tbuf, *cp; 1932 struct dirent *dp; 1933 off_t *cookies = NULL, *cop; 1934 int error, eof, nc, len; 1935 1936 tbuf = malloc(NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK); 1937 1938 aiov.iov_base = tbuf; 1939 aiov.iov_len = NFS_DIRFRAGSIZ; 1940 auio.uio_iov = &aiov; 1941 auio.uio_iovcnt = 1; 1942 auio.uio_rw = UIO_READ; 1943 auio.uio_resid = NFS_DIRFRAGSIZ; 1944 auio.uio_offset = 0; 1945 UIO_SETUP_SYSSPACE(&auio); 1946 1947 error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc); 1948 1949 len = NFS_DIRFRAGSIZ - auio.uio_resid; 1950 if (error || len == 0) { 1951 free(tbuf, M_TEMP); 1952 if (cookies) 1953 free(cookies, M_TEMP); 1954 return; 1955 } 1956 1957 /* 1958 * Find the first valid entry and look at its offset cookie. 1959 */ 1960 1961 cp = tbuf; 1962 for (cop = cookies; len > 0; len -= dp->d_reclen) { 1963 dp = (struct dirent *)cp; 1964 if (dp->d_fileno != 0 && len >= dp->d_reclen) { 1965 if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) { 1966 *flagp |= NFSMNT_SWAPCOOKIE; 1967 nfs_invaldircache(vp, 0); 1968 nfs_vinvalbuf(vp, 0, cred, l, 1); 1969 } 1970 break; 1971 } 1972 cop++; 1973 cp += dp->d_reclen; 1974 } 1975 1976 free(tbuf, M_TEMP); 1977 free(cookies, M_TEMP); 1978 } 1979 #endif /* NFS */ 1980 1981 /* 1982 * A fiddled version of m_adj() that ensures null fill to a 32-bit 1983 * boundary and only trims off the back end 1984 * 1985 * 1. trim off 'len' bytes as m_adj(mp, -len). 1986 * 2. add zero-padding 'nul' bytes at the end of the mbuf chain. 1987 */ 1988 void 1989 nfs_zeropad(struct mbuf *mp, int len, int nul) 1990 { 1991 struct mbuf *m; 1992 int count; 1993 1994 /* 1995 * Trim from tail. Scan the mbuf chain, 1996 * calculating its length and finding the last mbuf. 1997 * If the adjustment only affects this mbuf, then just 1998 * adjust and return. Otherwise, rescan and truncate 1999 * after the remaining size. 2000 */ 2001 count = 0; 2002 m = mp; 2003 for (;;) { 2004 count += m->m_len; 2005 if (m->m_next == NULL) 2006 break; 2007 m = m->m_next; 2008 } 2009 2010 KDASSERT(count >= len); 2011 2012 if (m->m_len >= len) { 2013 m->m_len -= len; 2014 } else { 2015 count -= len; 2016 /* 2017 * Correct length for chain is "count". 2018 * Find the mbuf with last data, adjust its length, 2019 * and toss data from remaining mbufs on chain. 2020 */ 2021 for (m = mp; m; m = m->m_next) { 2022 if (m->m_len >= count) { 2023 m->m_len = count; 2024 break; 2025 } 2026 count -= m->m_len; 2027 } 2028 KASSERT(m && m->m_next); 2029 m_freem(m->m_next); 2030 m->m_next = NULL; 2031 } 2032 2033 KDASSERT(m->m_next == NULL); 2034 2035 /* 2036 * zero-padding. 2037 */ 2038 if (nul > 0) { 2039 char *cp; 2040 int i; 2041 2042 if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) { 2043 struct mbuf *n; 2044 2045 KDASSERT(MLEN >= nul); 2046 n = m_get(M_WAIT, MT_DATA); 2047 MCLAIM(n, &nfs_mowner); 2048 n->m_len = nul; 2049 n->m_next = NULL; 2050 m->m_next = n; 2051 cp = mtod(n, void *); 2052 } else { 2053 cp = mtod(m, char *) + m->m_len; 2054 m->m_len += nul; 2055 } 2056 for (i = 0; i < nul; i++) 2057 *cp++ = '\0'; 2058 } 2059 return; 2060 } 2061 2062 /* 2063 * Make these functions instead of macros, so that the kernel text size 2064 * doesn't get too big... 2065 */ 2066 void 2067 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) 2068 { 2069 struct mbuf *mb = *mbp; 2070 char *bpos = *bposp; 2071 u_int32_t *tl; 2072 2073 if (before_ret) { 2074 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2075 *tl = nfs_false; 2076 } else { 2077 nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED); 2078 *tl++ = nfs_true; 2079 txdr_hyper(before_vap->va_size, tl); 2080 tl += 2; 2081 txdr_nfsv3time(&(before_vap->va_mtime), tl); 2082 tl += 2; 2083 txdr_nfsv3time(&(before_vap->va_ctime), tl); 2084 } 2085 *bposp = bpos; 2086 *mbp = mb; 2087 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp); 2088 } 2089 2090 void 2091 nfsm_srvpostopattr(struct nfsrv_descript *nfsd, int after_ret, struct vattr *after_vap, struct mbuf **mbp, char **bposp) 2092 { 2093 struct mbuf *mb = *mbp; 2094 char *bpos = *bposp; 2095 u_int32_t *tl; 2096 struct nfs_fattr *fp; 2097 2098 if (after_ret) { 2099 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED); 2100 *tl = nfs_false; 2101 } else { 2102 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR); 2103 *tl++ = nfs_true; 2104 fp = (struct nfs_fattr *)tl; 2105 nfsm_srvfattr(nfsd, after_vap, fp); 2106 } 2107 *mbp = mb; 2108 *bposp = bpos; 2109 } 2110 2111 void 2112 nfsm_srvfattr(struct nfsrv_descript *nfsd, struct vattr *vap, struct nfs_fattr *fp) 2113 { 2114 2115 fp->fa_nlink = txdr_unsigned(vap->va_nlink); 2116 fp->fa_uid = txdr_unsigned(vap->va_uid); 2117 fp->fa_gid = txdr_unsigned(vap->va_gid); 2118 if (nfsd->nd_flag & ND_NFSV3) { 2119 fp->fa_type = vtonfsv3_type(vap->va_type); 2120 fp->fa_mode = vtonfsv3_mode(vap->va_mode); 2121 txdr_hyper(vap->va_size, &fp->fa3_size); 2122 txdr_hyper(vap->va_bytes, &fp->fa3_used); 2123 fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev)); 2124 fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev)); 2125 fp->fa3_fsid.nfsuquad[0] = 0; 2126 fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid); 2127 txdr_hyper(vap->va_fileid, &fp->fa3_fileid); 2128 txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime); 2129 txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime); 2130 txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime); 2131 } else { 2132 fp->fa_type = vtonfsv2_type(vap->va_type); 2133 fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); 2134 fp->fa2_size = txdr_unsigned(vap->va_size); 2135 fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize); 2136 if (vap->va_type == VFIFO) 2137 fp->fa2_rdev = 0xffffffff; 2138 else 2139 fp->fa2_rdev = txdr_unsigned(vap->va_rdev); 2140 fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE); 2141 fp->fa2_fsid = txdr_unsigned(vap->va_fsid); 2142 fp->fa2_fileid = txdr_unsigned(vap->va_fileid); 2143 txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime); 2144 txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime); 2145 txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime); 2146 } 2147 } 2148 2149 /* 2150 * This function compares two net addresses by family and returns true 2151 * if they are the same host. 2152 * If there is any doubt, return false. 2153 * The AF_INET family is handled as a special case so that address mbufs 2154 * don't need to be saved to store "struct in_addr", which is only 4 bytes. 2155 */ 2156 int 2157 netaddr_match(int family, union nethostaddr *haddr, struct mbuf *nam) 2158 { 2159 struct sockaddr_in *inetaddr; 2160 2161 switch (family) { 2162 case AF_INET: 2163 inetaddr = mtod(nam, struct sockaddr_in *); 2164 if (inetaddr->sin_family == AF_INET && 2165 inetaddr->sin_addr.s_addr == haddr->had_inetaddr) 2166 return (1); 2167 break; 2168 case AF_INET6: 2169 { 2170 struct sockaddr_in6 *sin6_1, *sin6_2; 2171 2172 sin6_1 = mtod(nam, struct sockaddr_in6 *); 2173 sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *); 2174 if (sin6_1->sin6_family == AF_INET6 && 2175 IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr)) 2176 return 1; 2177 } 2178 default: 2179 break; 2180 }; 2181 return (0); 2182 } 2183 2184 /* 2185 * The write verifier has changed (probably due to a server reboot), so all 2186 * PG_NEEDCOMMIT pages will have to be written again. Since they are marked 2187 * as dirty or are being written out just now, all this takes is clearing 2188 * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for 2189 * the mount point. 2190 */ 2191 void 2192 nfs_clearcommit(struct mount *mp) 2193 { 2194 struct vnode *vp; 2195 struct nfsnode *np; 2196 struct vm_page *pg; 2197 struct nfsmount *nmp = VFSTONFS(mp); 2198 2199 rw_enter(&nmp->nm_writeverflock, RW_WRITER); 2200 mutex_enter(&mntvnode_lock); 2201 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2202 KASSERT(vp->v_mount == mp); 2203 if (vp->v_type != VREG) 2204 continue; 2205 np = VTONFS(vp); 2206 np->n_pushlo = np->n_pushhi = np->n_pushedlo = 2207 np->n_pushedhi = 0; 2208 np->n_commitflags &= 2209 ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID); 2210 mutex_enter(&vp->v_uobj.vmobjlock); 2211 TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq.queue) { 2212 pg->flags &= ~PG_NEEDCOMMIT; 2213 } 2214 mutex_exit(&vp->v_uobj.vmobjlock); 2215 } 2216 mutex_exit(&mntvnode_lock); 2217 mutex_enter(&nmp->nm_lock); 2218 nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF; 2219 mutex_exit(&nmp->nm_lock); 2220 rw_exit(&nmp->nm_writeverflock); 2221 } 2222 2223 void 2224 nfs_merge_commit_ranges(struct vnode *vp) 2225 { 2226 struct nfsnode *np = VTONFS(vp); 2227 2228 KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID); 2229 2230 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2231 np->n_pushedlo = np->n_pushlo; 2232 np->n_pushedhi = np->n_pushhi; 2233 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2234 } else { 2235 if (np->n_pushlo < np->n_pushedlo) 2236 np->n_pushedlo = np->n_pushlo; 2237 if (np->n_pushhi > np->n_pushedhi) 2238 np->n_pushedhi = np->n_pushhi; 2239 } 2240 2241 np->n_pushlo = np->n_pushhi = 0; 2242 np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID; 2243 2244 #ifdef NFS_DEBUG_COMMIT 2245 printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2246 (unsigned)np->n_pushedhi); 2247 #endif 2248 } 2249 2250 int 2251 nfs_in_committed_range(struct vnode *vp, off_t off, off_t len) 2252 { 2253 struct nfsnode *np = VTONFS(vp); 2254 off_t lo, hi; 2255 2256 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2257 return 0; 2258 lo = off; 2259 hi = lo + len; 2260 2261 return (lo >= np->n_pushedlo && hi <= np->n_pushedhi); 2262 } 2263 2264 int 2265 nfs_in_tobecommitted_range(struct vnode *vp, off_t off, off_t len) 2266 { 2267 struct nfsnode *np = VTONFS(vp); 2268 off_t lo, hi; 2269 2270 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2271 return 0; 2272 lo = off; 2273 hi = lo + len; 2274 2275 return (lo >= np->n_pushlo && hi <= np->n_pushhi); 2276 } 2277 2278 void 2279 nfs_add_committed_range(struct vnode *vp, off_t off, off_t len) 2280 { 2281 struct nfsnode *np = VTONFS(vp); 2282 off_t lo, hi; 2283 2284 lo = off; 2285 hi = lo + len; 2286 2287 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) { 2288 np->n_pushedlo = lo; 2289 np->n_pushedhi = hi; 2290 np->n_commitflags |= NFS_COMMIT_PUSHED_VALID; 2291 } else { 2292 if (hi > np->n_pushedhi) 2293 np->n_pushedhi = hi; 2294 if (lo < np->n_pushedlo) 2295 np->n_pushedlo = lo; 2296 } 2297 #ifdef NFS_DEBUG_COMMIT 2298 printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2299 (unsigned)np->n_pushedhi); 2300 #endif 2301 } 2302 2303 void 2304 nfs_del_committed_range(struct vnode *vp, off_t off, off_t len) 2305 { 2306 struct nfsnode *np = VTONFS(vp); 2307 off_t lo, hi; 2308 2309 if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) 2310 return; 2311 2312 lo = off; 2313 hi = lo + len; 2314 2315 if (lo > np->n_pushedhi || hi < np->n_pushedlo) 2316 return; 2317 if (lo <= np->n_pushedlo) 2318 np->n_pushedlo = hi; 2319 else if (hi >= np->n_pushedhi) 2320 np->n_pushedhi = lo; 2321 else { 2322 /* 2323 * XXX There's only one range. If the deleted range 2324 * is in the middle, pick the largest of the 2325 * contiguous ranges that it leaves. 2326 */ 2327 if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi)) 2328 np->n_pushedhi = lo; 2329 else 2330 np->n_pushedlo = hi; 2331 } 2332 #ifdef NFS_DEBUG_COMMIT 2333 printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo, 2334 (unsigned)np->n_pushedhi); 2335 #endif 2336 } 2337 2338 void 2339 nfs_add_tobecommitted_range(struct vnode *vp, off_t off, off_t len) 2340 { 2341 struct nfsnode *np = VTONFS(vp); 2342 off_t lo, hi; 2343 2344 lo = off; 2345 hi = lo + len; 2346 2347 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) { 2348 np->n_pushlo = lo; 2349 np->n_pushhi = hi; 2350 np->n_commitflags |= NFS_COMMIT_PUSH_VALID; 2351 } else { 2352 if (lo < np->n_pushlo) 2353 np->n_pushlo = lo; 2354 if (hi > np->n_pushhi) 2355 np->n_pushhi = hi; 2356 } 2357 #ifdef NFS_DEBUG_COMMIT 2358 printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2359 (unsigned)np->n_pushhi); 2360 #endif 2361 } 2362 2363 void 2364 nfs_del_tobecommitted_range(struct vnode *vp, off_t off, off_t len) 2365 { 2366 struct nfsnode *np = VTONFS(vp); 2367 off_t lo, hi; 2368 2369 if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) 2370 return; 2371 2372 lo = off; 2373 hi = lo + len; 2374 2375 if (lo > np->n_pushhi || hi < np->n_pushlo) 2376 return; 2377 2378 if (lo <= np->n_pushlo) 2379 np->n_pushlo = hi; 2380 else if (hi >= np->n_pushhi) 2381 np->n_pushhi = lo; 2382 else { 2383 /* 2384 * XXX There's only one range. If the deleted range 2385 * is in the middle, pick the largest of the 2386 * contiguous ranges that it leaves. 2387 */ 2388 if ((np->n_pushlo - lo) > (hi - np->n_pushhi)) 2389 np->n_pushhi = lo; 2390 else 2391 np->n_pushlo = hi; 2392 } 2393 #ifdef NFS_DEBUG_COMMIT 2394 printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo, 2395 (unsigned)np->n_pushhi); 2396 #endif 2397 } 2398 2399 /* 2400 * Map errnos to NFS error numbers. For Version 3 also filter out error 2401 * numbers not specified for the associated procedure. 2402 */ 2403 int 2404 nfsrv_errmap(struct nfsrv_descript *nd, int err) 2405 { 2406 const short *defaulterrp, *errp; 2407 2408 if (nd->nd_flag & ND_NFSV3) { 2409 if (nd->nd_procnum <= NFSPROC_COMMIT) { 2410 errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum]; 2411 while (*++errp) { 2412 if (*errp == err) 2413 return (err); 2414 else if (*errp > err) 2415 break; 2416 } 2417 return ((int)*defaulterrp); 2418 } else 2419 return (err & 0xffff); 2420 } 2421 if (err <= ELAST) 2422 return ((int)nfsrv_v2errmap[err - 1]); 2423 return (NFSERR_IO); 2424 } 2425 2426 u_int32_t 2427 nfs_getxid() 2428 { 2429 u_int32_t newxid; 2430 2431 /* get next xid. skip 0 */ 2432 do { 2433 newxid = atomic_inc_32_nv(&nfs_xid); 2434 } while (__predict_false(newxid == 0)); 2435 2436 return txdr_unsigned(newxid); 2437 } 2438 2439 /* 2440 * assign a new xid for existing request. 2441 * used for NFSERR_JUKEBOX handling. 2442 */ 2443 void 2444 nfs_renewxid(struct nfsreq *req) 2445 { 2446 u_int32_t xid; 2447 int off; 2448 2449 xid = nfs_getxid(); 2450 if (req->r_nmp->nm_sotype == SOCK_STREAM) 2451 off = sizeof(u_int32_t); /* RPC record mark */ 2452 else 2453 off = 0; 2454 2455 m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid); 2456 req->r_xid = xid; 2457 } 2458 2459 #if defined(NFS) 2460 /* 2461 * Set the attribute timeout based on how recently the file has been modified. 2462 */ 2463 2464 time_t 2465 nfs_attrtimeo(struct nfsmount *nmp, struct nfsnode *np) 2466 { 2467 time_t timeo; 2468 2469 if ((nmp->nm_flag & NFSMNT_NOAC) != 0) 2470 return 0; 2471 2472 if (((np)->n_flag & NMODIFIED) != 0) 2473 return NFS_MINATTRTIMO; 2474 2475 timeo = (time_second - np->n_mtime.tv_sec) / 10; 2476 timeo = max(timeo, NFS_MINATTRTIMO); 2477 timeo = min(timeo, NFS_MAXATTRTIMO); 2478 return timeo; 2479 } 2480 #endif /* defined(NFS) */ 2481
Indexes created Wed Sep 24 05:09:52 GMT 2025