1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)vfs_lookup.c 8.4 (Berkeley) 2/16/94 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_capsicum.h" 43 #include "opt_ktrace.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/dirent.h> 48 #include <sys/kernel.h> 49 #include <sys/capsicum.h> 50 #include <sys/fcntl.h> 51 #include <sys/jail.h> 52 #include <sys/lock.h> 53 #include <sys/mutex.h> 54 #include <sys/namei.h> 55 #include <sys/vnode.h> 56 #include <sys/mount.h> 57 #include <sys/filedesc.h> 58 #include <sys/proc.h> 59 #include <sys/sdt.h> 60 #include <sys/syscallsubr.h> 61 #include <sys/sysctl.h> 62 #ifdef KTRACE 63 #include <sys/ktrace.h> 64 #endif 65 #ifdef INVARIANTS 66 #include <machine/_inttypes.h> 67 #endif 68 69 #include <security/audit/audit.h> 70 #include <security/mac/mac_framework.h> 71 72 #include <vm/uma.h> 73 74 #define NAMEI_DIAGNOSTIC 1 75 #undef NAMEI_DIAGNOSTIC 76 77 SDT_PROVIDER_DEFINE(vfs); 78 SDT_PROBE_DEFINE4(vfs, namei, lookup, entry, "struct vnode *", "char *", 79 "unsigned long", "bool"); 80 SDT_PROBE_DEFINE4(vfs, namei, lookup, return, "int", "struct vnode *", "bool", 81 "struct nameidata"); 82 83 /* Allocation zone for namei. */ 84 uma_zone_t namei_zone; 85 86 /* Placeholder vnode for mp traversal. */ 87 static struct vnode *vp_crossmp; 88 89 static int 90 crossmp_vop_islocked(struct vop_islocked_args *ap) 91 { 92 93 return (LK_SHARED); 94 } 95 96 static int 97 crossmp_vop_lock1(struct vop_lock1_args *ap) 98 { 99 struct vnode *vp; 100 struct lock *lk __unused; 101 const char *file __unused; 102 int flags, line __unused; 103 104 vp = ap->a_vp; 105 lk = vp->v_vnlock; 106 flags = ap->a_flags; 107 file = ap->a_file; 108 line = ap->a_line; 109 110 if ((flags & LK_SHARED) == 0) 111 panic("invalid lock request for crossmp"); 112 113 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, file, line, 114 flags & LK_INTERLOCK ? &VI_MTX(vp)->lock_object : NULL); 115 WITNESS_LOCK(&lk->lock_object, 0, file, line); 116 if ((flags & LK_INTERLOCK) != 0) 117 VI_UNLOCK(vp); 118 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, ap->a_file, line); 119 return (0); 120 } 121 122 static int 123 crossmp_vop_unlock(struct vop_unlock_args *ap) 124 { 125 struct vnode *vp; 126 struct lock *lk __unused; 127 128 vp = ap->a_vp; 129 lk = vp->v_vnlock; 130 131 WITNESS_UNLOCK(&lk->lock_object, 0, LOCK_FILE, LOCK_LINE); 132 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE, 133 LOCK_LINE); 134 return (0); 135 } 136 137 static struct vop_vector crossmp_vnodeops = { 138 .vop_default = &default_vnodeops, 139 .vop_islocked = crossmp_vop_islocked, 140 .vop_lock1 = crossmp_vop_lock1, 141 .vop_unlock = crossmp_vop_unlock, 142 }; 143 /* 144 * VFS_VOP_VECTOR_REGISTER(crossmp_vnodeops) is not used here since the vnode 145 * gets allocated early. See nameiinit for the direct call below. 146 */ 147 148 struct nameicap_tracker { 149 struct vnode *dp; 150 TAILQ_ENTRY(nameicap_tracker) nm_link; 151 }; 152 153 /* Zone for cap mode tracker elements used for dotdot capability checks. */ 154 MALLOC_DEFINE(M_NAMEITRACKER, "namei_tracker", "namei tracking for dotdot"); 155 156 static void 157 nameiinit(void *dummy __unused) 158 { 159 160 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL, 161 UMA_ALIGN_PTR, 0); 162 vfs_vector_op_register(&crossmp_vnodeops); 163 getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp); 164 } 165 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL); 166 167 static int lookup_cap_dotdot = 1; 168 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN, 169 &lookup_cap_dotdot, 0, 170 "enables \"..\" components in path lookup in capability mode"); 171 static int lookup_cap_dotdot_nonlocal = 1; 172 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN, 173 &lookup_cap_dotdot_nonlocal, 0, 174 "enables \"..\" components in path lookup in capability mode " 175 "on non-local mount"); 176 177 static void 178 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp) 179 { 180 struct nameicap_tracker *nt; 181 struct componentname *cnp; 182 183 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR) 184 return; 185 cnp = &ndp->ni_cnd; 186 nt = TAILQ_LAST(&ndp->ni_cap_tracker, nameicap_tracker_head); 187 if (nt != NULL && nt->dp == dp) 188 return; 189 nt = malloc(sizeof(*nt), M_NAMEITRACKER, M_WAITOK); 190 vhold(dp); 191 nt->dp = dp; 192 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link); 193 } 194 195 static void 196 nameicap_cleanup_from(struct nameidata *ndp, struct nameicap_tracker *first) 197 { 198 struct nameicap_tracker *nt, *nt1; 199 200 nt = first; 201 TAILQ_FOREACH_FROM_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) { 202 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link); 203 vdrop(nt->dp); 204 free(nt, M_NAMEITRACKER); 205 } 206 } 207 208 static void 209 nameicap_cleanup(struct nameidata *ndp) 210 { 211 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) || 212 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative")); 213 nameicap_cleanup_from(ndp, NULL); 214 } 215 216 /* 217 * For dotdot lookups in capability mode, only allow the component 218 * lookup to succeed if the resulting directory was already traversed 219 * during the operation. This catches situations where already 220 * traversed directory is moved to different parent, and then we walk 221 * over it with dotdots. 222 * 223 * Also allow to force failure of dotdot lookups for non-local 224 * filesystems, where external agents might assist local lookups to 225 * escape the compartment. 226 */ 227 static int 228 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp) 229 { 230 struct nameicap_tracker *nt; 231 struct mount *mp; 232 233 if (dp == NULL || dp->v_type != VDIR || (ndp->ni_lcf & 234 NI_LCF_STRICTRELATIVE) == 0) 235 return (0); 236 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0) 237 return (ENOTCAPABLE); 238 mp = dp->v_mount; 239 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL && 240 (mp->mnt_flag & MNT_LOCAL) == 0) 241 return (ENOTCAPABLE); 242 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head, 243 nm_link) { 244 if (dp == nt->dp) { 245 nt = TAILQ_NEXT(nt, nm_link); 246 if (nt != NULL) 247 nameicap_cleanup_from(ndp, nt); 248 return (0); 249 } 250 } 251 return (ENOTCAPABLE); 252 } 253 254 static void 255 namei_cleanup_cnp(struct componentname *cnp) 256 { 257 258 uma_zfree(namei_zone, cnp->cn_pnbuf); 259 #ifdef DIAGNOSTIC 260 cnp->cn_pnbuf = NULL; 261 cnp->cn_nameptr = NULL; 262 #endif 263 } 264 265 static int 266 namei_handle_root(struct nameidata *ndp, struct vnode **dpp) 267 { 268 struct componentname *cnp; 269 270 cnp = &ndp->ni_cnd; 271 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) { 272 #ifdef KTRACE 273 if (KTRPOINT(curthread, KTR_CAPFAIL)) 274 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 275 #endif 276 return (ENOTCAPABLE); 277 } 278 while (*(cnp->cn_nameptr) == '/') { 279 cnp->cn_nameptr++; 280 ndp->ni_pathlen--; 281 } 282 *dpp = ndp->ni_rootdir; 283 vrefact(*dpp); 284 return (0); 285 } 286 287 static int 288 namei_setup(struct nameidata *ndp, struct vnode **dpp, struct pwd **pwdp) 289 { 290 struct componentname *cnp; 291 struct file *dfp; 292 struct thread *td; 293 struct pwd *pwd; 294 cap_rights_t rights; 295 int error; 296 bool startdir_used; 297 298 cnp = &ndp->ni_cnd; 299 td = cnp->cn_thread; 300 301 startdir_used = false; 302 *pwdp = NULL; 303 *dpp = NULL; 304 305 #ifdef CAPABILITY_MODE 306 /* 307 * In capability mode, lookups must be restricted to happen in 308 * the subtree with the root specified by the file descriptor: 309 * - The root must be real file descriptor, not the pseudo-descriptor 310 * AT_FDCWD. 311 * - The passed path must be relative and not absolute. 312 * - If lookup_cap_dotdot is disabled, path must not contain the 313 * '..' components. 314 * - If lookup_cap_dotdot is enabled, we verify that all '..' 315 * components lookups result in the directories which were 316 * previously walked by us, which prevents an escape from 317 * the relative root. 318 */ 319 if (IN_CAPABILITY_MODE(td) && (cnp->cn_flags & NOCAPCHECK) == 0) { 320 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE; 321 ndp->ni_resflags |= NIRES_STRICTREL; 322 if (ndp->ni_dirfd == AT_FDCWD) { 323 #ifdef KTRACE 324 if (KTRPOINT(td, KTR_CAPFAIL)) 325 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 326 #endif 327 return (ECAPMODE); 328 } 329 } 330 #endif 331 error = 0; 332 333 /* 334 * Get starting point for the translation. 335 */ 336 pwd = pwd_hold(td); 337 /* 338 * The reference on ni_rootdir is acquired in the block below to avoid 339 * back-to-back atomics for absolute lookups. 340 */ 341 ndp->ni_rootdir = pwd->pwd_rdir; 342 ndp->ni_topdir = pwd->pwd_jdir; 343 344 if (cnp->cn_pnbuf[0] == '/') { 345 ndp->ni_resflags |= NIRES_ABS; 346 error = namei_handle_root(ndp, dpp); 347 } else { 348 if (ndp->ni_startdir != NULL) { 349 *dpp = ndp->ni_startdir; 350 startdir_used = true; 351 } else if (ndp->ni_dirfd == AT_FDCWD) { 352 *dpp = pwd->pwd_cdir; 353 vrefact(*dpp); 354 } else { 355 rights = *ndp->ni_rightsneeded; 356 cap_rights_set_one(&rights, CAP_LOOKUP); 357 358 if (cnp->cn_flags & AUDITVNODE1) 359 AUDIT_ARG_ATFD1(ndp->ni_dirfd); 360 if (cnp->cn_flags & AUDITVNODE2) 361 AUDIT_ARG_ATFD2(ndp->ni_dirfd); 362 /* 363 * Effectively inlined fgetvp_rights, because 364 * we need to inspect the file as well as 365 * grabbing the vnode. No check for O_PATH, 366 * files to implement its semantic. 367 */ 368 error = fget_cap(td, ndp->ni_dirfd, &rights, 369 &dfp, &ndp->ni_filecaps); 370 if (error != 0) { 371 /* 372 * Preserve the error; it should either be EBADF 373 * or capability-related, both of which can be 374 * safely returned to the caller. 375 */ 376 } else { 377 if (dfp->f_ops == &badfileops) { 378 error = EBADF; 379 } else if (dfp->f_vnode == NULL) { 380 error = ENOTDIR; 381 } else { 382 *dpp = dfp->f_vnode; 383 vref(*dpp); 384 385 if ((dfp->f_flag & FSEARCH) != 0) 386 cnp->cn_flags |= NOEXECCHECK; 387 } 388 fdrop(dfp, td); 389 } 390 #ifdef CAPABILITIES 391 /* 392 * If file descriptor doesn't have all rights, 393 * all lookups relative to it must also be 394 * strictly relative. 395 */ 396 CAP_ALL(&rights); 397 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, 398 &rights) || 399 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || 400 ndp->ni_filecaps.fc_nioctls != -1) { 401 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE; 402 ndp->ni_resflags |= NIRES_STRICTREL; 403 } 404 #endif 405 } 406 if (error == 0 && (*dpp)->v_type != VDIR && 407 (cnp->cn_pnbuf[0] != '\0' || 408 (cnp->cn_flags & EMPTYPATH) == 0)) 409 error = ENOTDIR; 410 } 411 if (error == 0 && (cnp->cn_flags & RBENEATH) != 0) { 412 if (cnp->cn_pnbuf[0] == '/') { 413 error = ENOTCAPABLE; 414 } else if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0) { 415 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE | 416 NI_LCF_CAP_DOTDOT; 417 } 418 } 419 420 /* 421 * If we are auditing the kernel pathname, save the user pathname. 422 */ 423 if (cnp->cn_flags & AUDITVNODE1) 424 AUDIT_ARG_UPATH1_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf); 425 if (cnp->cn_flags & AUDITVNODE2) 426 AUDIT_ARG_UPATH2_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf); 427 if (ndp->ni_startdir != NULL && !startdir_used) 428 vrele(ndp->ni_startdir); 429 if (error != 0) { 430 if (*dpp != NULL) 431 vrele(*dpp); 432 pwd_drop(pwd); 433 return (error); 434 } 435 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 && 436 lookup_cap_dotdot != 0) 437 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT; 438 SDT_PROBE4(vfs, namei, lookup, entry, *dpp, cnp->cn_pnbuf, 439 cnp->cn_flags, false); 440 *pwdp = pwd; 441 return (0); 442 } 443 444 static int 445 namei_getpath(struct nameidata *ndp) 446 { 447 struct componentname *cnp; 448 int error; 449 450 cnp = &ndp->ni_cnd; 451 452 /* 453 * Get a buffer for the name to be translated, and copy the 454 * name into the buffer. 455 */ 456 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 457 if (ndp->ni_segflg == UIO_SYSSPACE) { 458 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN, 459 &ndp->ni_pathlen); 460 } else { 461 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN, 462 &ndp->ni_pathlen); 463 } 464 465 if (__predict_false(error != 0)) 466 return (error); 467 468 /* 469 * Don't allow empty pathnames unless EMPTYPATH is specified. 470 * Caller checks for ENOENT as an indication for the empty path. 471 */ 472 if (__predict_false(*cnp->cn_pnbuf == '\0')) 473 return (ENOENT); 474 475 cnp->cn_nameptr = cnp->cn_pnbuf; 476 return (0); 477 } 478 479 static int 480 namei_emptypath(struct nameidata *ndp) 481 { 482 struct componentname *cnp; 483 struct pwd *pwd; 484 struct vnode *dp; 485 int error; 486 487 cnp = &ndp->ni_cnd; 488 MPASS(*cnp->cn_pnbuf == '\0'); 489 MPASS((cnp->cn_flags & EMPTYPATH) != 0); 490 MPASS((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) == 0); 491 492 error = namei_setup(ndp, &dp, &pwd); 493 if (error != 0) { 494 namei_cleanup_cnp(cnp); 495 goto errout; 496 } 497 498 /* 499 * Usecount on dp already provided by namei_setup. 500 */ 501 ndp->ni_vp = dp; 502 namei_cleanup_cnp(cnp); 503 pwd_drop(pwd); 504 ndp->ni_resflags |= NIRES_EMPTYPATH; 505 NDVALIDATE(ndp); 506 if ((cnp->cn_flags & LOCKLEAF) != 0) { 507 VOP_LOCK(dp, (cnp->cn_flags & LOCKSHARED) != 0 ? 508 LK_SHARED : LK_EXCLUSIVE); 509 if (VN_IS_DOOMED(dp)) { 510 vput(dp); 511 error = ENOENT; 512 goto errout; 513 } 514 } 515 SDT_PROBE4(vfs, namei, lookup, return, 0, ndp->ni_vp, false, ndp); 516 return (0); 517 518 errout: 519 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp); 520 return (error); 521 } 522 523 /* 524 * Convert a pathname into a pointer to a locked vnode. 525 * 526 * The FOLLOW flag is set when symbolic links are to be followed 527 * when they occur at the end of the name translation process. 528 * Symbolic links are always followed for all other pathname 529 * components other than the last. 530 * 531 * The segflg defines whether the name is to be copied from user 532 * space or kernel space. 533 * 534 * Overall outline of namei: 535 * 536 * copy in name 537 * get starting directory 538 * while (!done && !error) { 539 * call lookup to search path. 540 * if symbolic link, massage name in buffer and continue 541 * } 542 */ 543 int 544 namei(struct nameidata *ndp) 545 { 546 char *cp; /* pointer into pathname argument */ 547 struct vnode *dp; /* the directory we are searching */ 548 struct iovec aiov; /* uio for reading symbolic links */ 549 struct componentname *cnp; 550 struct thread *td; 551 struct pwd *pwd; 552 struct uio auio; 553 int error, linklen; 554 enum cache_fpl_status status; 555 556 cnp = &ndp->ni_cnd; 557 td = cnp->cn_thread; 558 #ifdef INVARIANTS 559 KASSERT((ndp->ni_debugflags & NAMEI_DBG_CALLED) == 0, 560 ("%s: repeated call to namei without NDREINIT", __func__)); 561 KASSERT(ndp->ni_debugflags == NAMEI_DBG_INITED, 562 ("%s: bad debugflags %d", __func__, ndp->ni_debugflags)); 563 ndp->ni_debugflags |= NAMEI_DBG_CALLED; 564 if (ndp->ni_startdir != NULL) 565 ndp->ni_debugflags |= NAMEI_DBG_HADSTARTDIR; 566 if (cnp->cn_flags & FAILIFEXISTS) { 567 KASSERT(cnp->cn_nameiop == CREATE, 568 ("%s: FAILIFEXISTS passed for op %d", __func__, cnp->cn_nameiop)); 569 /* 570 * The limitation below is to restrict hairy corner cases. 571 */ 572 KASSERT((cnp->cn_flags & (LOCKPARENT | LOCKLEAF)) == LOCKPARENT, 573 ("%s: FAILIFEXISTS must be passed with LOCKPARENT and without LOCKLEAF", 574 __func__)); 575 } 576 /* 577 * For NDVALIDATE. 578 * 579 * While NDINIT may seem like a more natural place to do it, there are 580 * callers which directly modify flags past invoking init. 581 */ 582 cnp->cn_origflags = cnp->cn_flags; 583 #endif 584 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred; 585 KASSERT(ndp->ni_resflags == 0, ("%s: garbage in ni_resflags: %x\n", 586 __func__, ndp->ni_resflags)); 587 KASSERT(cnp->cn_cred && td->td_proc, ("namei: bad cred/proc")); 588 KASSERT((cnp->cn_flags & NAMEI_INTERNAL_FLAGS) == 0, 589 ("namei: unexpected flags: %" PRIx64 "\n", 590 cnp->cn_flags & NAMEI_INTERNAL_FLAGS)); 591 if (cnp->cn_flags & NOCACHE) 592 KASSERT(cnp->cn_nameiop != LOOKUP, 593 ("%s: NOCACHE passed with LOOKUP", __func__)); 594 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR || 595 ndp->ni_startdir->v_type == VBAD); 596 597 ndp->ni_lcf = 0; 598 ndp->ni_loopcnt = 0; 599 ndp->ni_vp = NULL; 600 601 error = namei_getpath(ndp); 602 if (__predict_false(error != 0)) { 603 if (error == ENOENT && (cnp->cn_flags & EMPTYPATH) != 0) 604 return (namei_emptypath(ndp)); 605 namei_cleanup_cnp(cnp); 606 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, 607 false, ndp); 608 return (error); 609 } 610 611 #ifdef KTRACE 612 if (KTRPOINT(td, KTR_NAMEI)) { 613 KASSERT(cnp->cn_thread == curthread, 614 ("namei not using curthread")); 615 ktrnamei(cnp->cn_pnbuf); 616 } 617 #endif 618 619 /* 620 * First try looking up the target without locking any vnodes. 621 * 622 * We may need to start from scratch or pick up where it left off. 623 */ 624 error = cache_fplookup(ndp, &status, &pwd); 625 switch (status) { 626 case CACHE_FPL_STATUS_UNSET: 627 __assert_unreachable(); 628 break; 629 case CACHE_FPL_STATUS_HANDLED: 630 if (error == 0) 631 NDVALIDATE(ndp); 632 return (error); 633 case CACHE_FPL_STATUS_PARTIAL: 634 TAILQ_INIT(&ndp->ni_cap_tracker); 635 dp = ndp->ni_startdir; 636 break; 637 case CACHE_FPL_STATUS_DESTROYED: 638 ndp->ni_loopcnt = 0; 639 error = namei_getpath(ndp); 640 if (__predict_false(error != 0)) { 641 namei_cleanup_cnp(cnp); 642 return (error); 643 } 644 /* FALLTHROUGH */ 645 case CACHE_FPL_STATUS_ABORTED: 646 TAILQ_INIT(&ndp->ni_cap_tracker); 647 MPASS(ndp->ni_lcf == 0); 648 error = namei_setup(ndp, &dp, &pwd); 649 if (error != 0) { 650 namei_cleanup_cnp(cnp); 651 return (error); 652 } 653 break; 654 } 655 656 /* 657 * Locked lookup. 658 */ 659 for (;;) { 660 ndp->ni_startdir = dp; 661 error = lookup(ndp); 662 if (error != 0) 663 goto out; 664 665 /* 666 * If not a symbolic link, we're done. 667 */ 668 if ((cnp->cn_flags & ISSYMLINK) == 0) { 669 SDT_PROBE4(vfs, namei, lookup, return, error, 670 (error == 0 ? ndp->ni_vp : NULL), false, ndp); 671 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) { 672 namei_cleanup_cnp(cnp); 673 } else 674 cnp->cn_flags |= HASBUF; 675 nameicap_cleanup(ndp); 676 pwd_drop(pwd); 677 if (error == 0) 678 NDVALIDATE(ndp); 679 return (error); 680 } 681 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 682 error = ELOOP; 683 break; 684 } 685 #ifdef MAC 686 if ((cnp->cn_flags & NOMACCHECK) == 0) { 687 error = mac_vnode_check_readlink(td->td_ucred, 688 ndp->ni_vp); 689 if (error != 0) 690 break; 691 } 692 #endif 693 if (ndp->ni_pathlen > 1) 694 cp = uma_zalloc(namei_zone, M_WAITOK); 695 else 696 cp = cnp->cn_pnbuf; 697 aiov.iov_base = cp; 698 aiov.iov_len = MAXPATHLEN; 699 auio.uio_iov = &aiov; 700 auio.uio_iovcnt = 1; 701 auio.uio_offset = 0; 702 auio.uio_rw = UIO_READ; 703 auio.uio_segflg = UIO_SYSSPACE; 704 auio.uio_td = td; 705 auio.uio_resid = MAXPATHLEN; 706 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 707 if (error != 0) { 708 if (ndp->ni_pathlen > 1) 709 uma_zfree(namei_zone, cp); 710 break; 711 } 712 linklen = MAXPATHLEN - auio.uio_resid; 713 if (linklen == 0) { 714 if (ndp->ni_pathlen > 1) 715 uma_zfree(namei_zone, cp); 716 error = ENOENT; 717 break; 718 } 719 if (linklen + ndp->ni_pathlen > MAXPATHLEN) { 720 if (ndp->ni_pathlen > 1) 721 uma_zfree(namei_zone, cp); 722 error = ENAMETOOLONG; 723 break; 724 } 725 if (ndp->ni_pathlen > 1) { 726 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen); 727 uma_zfree(namei_zone, cnp->cn_pnbuf); 728 cnp->cn_pnbuf = cp; 729 } else 730 cnp->cn_pnbuf[linklen] = '\0'; 731 ndp->ni_pathlen += linklen; 732 vput(ndp->ni_vp); 733 dp = ndp->ni_dvp; 734 /* 735 * Check if root directory should replace current directory. 736 */ 737 cnp->cn_nameptr = cnp->cn_pnbuf; 738 if (*(cnp->cn_nameptr) == '/') { 739 vrele(dp); 740 error = namei_handle_root(ndp, &dp); 741 if (error != 0) 742 goto out; 743 } 744 } 745 vput(ndp->ni_vp); 746 ndp->ni_vp = NULL; 747 vrele(ndp->ni_dvp); 748 out: 749 MPASS(error != 0); 750 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp); 751 namei_cleanup_cnp(cnp); 752 nameicap_cleanup(ndp); 753 pwd_drop(pwd); 754 return (error); 755 } 756 757 static int 758 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags) 759 { 760 761 if (mp == NULL || ((lkflags & LK_SHARED) && 762 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) || 763 ((cnflags & ISDOTDOT) && 764 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) { 765 lkflags &= ~LK_SHARED; 766 lkflags |= LK_EXCLUSIVE; 767 } 768 lkflags |= LK_NODDLKTREAT; 769 return (lkflags); 770 } 771 772 static __inline int 773 needs_exclusive_leaf(struct mount *mp, int flags) 774 { 775 776 /* 777 * Intermediate nodes can use shared locks, we only need to 778 * force an exclusive lock for leaf nodes. 779 */ 780 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF)) 781 return (0); 782 783 /* Always use exclusive locks if LOCKSHARED isn't set. */ 784 if (!(flags & LOCKSHARED)) 785 return (1); 786 787 /* 788 * For lookups during open(), if the mount point supports 789 * extended shared operations, then use a shared lock for the 790 * leaf node, otherwise use an exclusive lock. 791 */ 792 if ((flags & ISOPEN) != 0) 793 return (!MNT_EXTENDED_SHARED(mp)); 794 795 /* 796 * Lookup requests outside of open() that specify LOCKSHARED 797 * only need a shared lock on the leaf vnode. 798 */ 799 return (0); 800 } 801 802 /* 803 * Various filesystems expect to be able to copy a name component with length 804 * bounded by NAME_MAX into a directory entry buffer of size MAXNAMLEN. Make 805 * sure that these are the same size. 806 */ 807 _Static_assert(MAXNAMLEN == NAME_MAX, 808 "MAXNAMLEN and NAME_MAX have different values"); 809 810 /* 811 * Search a pathname. 812 * This is a very central and rather complicated routine. 813 * 814 * The pathname is pointed to by ni_ptr and is of length ni_pathlen. 815 * The starting directory is taken from ni_startdir. The pathname is 816 * descended until done, or a symbolic link is encountered. The variable 817 * ni_more is clear if the path is completed; it is set to one if a 818 * symbolic link needing interpretation is encountered. 819 * 820 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on 821 * whether the name is to be looked up, created, renamed, or deleted. 822 * When CREATE, RENAME, or DELETE is specified, information usable in 823 * creating, renaming, or deleting a directory entry may be calculated. 824 * If flag has LOCKPARENT or'ed into it, the parent directory is returned 825 * locked. If flag has WANTPARENT or'ed into it, the parent directory is 826 * returned unlocked. Otherwise the parent directory is not returned. If 827 * the target of the pathname exists and LOCKLEAF is or'ed into the flag 828 * the target is returned locked, otherwise it is returned unlocked. 829 * When creating or renaming and LOCKPARENT is specified, the target may not 830 * be ".". When deleting and LOCKPARENT is specified, the target may be ".". 831 * 832 * Overall outline of lookup: 833 * 834 * dirloop: 835 * identify next component of name at ndp->ni_ptr 836 * handle degenerate case where name is null string 837 * if .. and crossing mount points and on mounted filesys, find parent 838 * call VOP_LOOKUP routine for next component name 839 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set 840 * component vnode returned in ni_vp (if it exists), locked. 841 * if result vnode is mounted on and crossing mount points, 842 * find mounted on vnode 843 * if more components of name, do next level at dirloop 844 * return the answer in ni_vp, locked if LOCKLEAF set 845 * if LOCKPARENT set, return locked parent in ni_dvp 846 * if WANTPARENT set, return unlocked parent in ni_dvp 847 */ 848 int 849 lookup(struct nameidata *ndp) 850 { 851 char *cp; /* pointer into pathname argument */ 852 char *prev_ni_next; /* saved ndp->ni_next */ 853 char *nulchar; /* location of '\0' in cn_pnbuf */ 854 struct vnode *dp = NULL; /* the directory we are searching */ 855 struct vnode *tdp; /* saved dp */ 856 struct mount *mp; /* mount table entry */ 857 struct prison *pr; 858 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */ 859 int docache; /* == 0 do not cache last component */ 860 int wantparent; /* 1 => wantparent or lockparent flag */ 861 int rdonly; /* lookup read-only flag bit */ 862 int error = 0; 863 int dpunlocked = 0; /* dp has already been unlocked */ 864 int relookup = 0; /* do not consume the path component */ 865 struct componentname *cnp = &ndp->ni_cnd; 866 int lkflags_save; 867 int ni_dvp_unlocked; 868 869 /* 870 * Setup: break out flag bits into variables. 871 */ 872 ni_dvp_unlocked = 0; 873 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT); 874 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent, 875 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT.")); 876 /* 877 * When set to zero, docache causes the last component of the 878 * pathname to be deleted from the cache and the full lookup 879 * of the name to be done (via VOP_CACHEDLOOKUP()). Often 880 * filesystems need some pre-computed values that are made 881 * during the full lookup, for instance UFS sets dp->i_offset. 882 * 883 * The docache variable is set to zero when requested by the 884 * NOCACHE flag and for all modifying operations except CREATE. 885 */ 886 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE; 887 if (cnp->cn_nameiop == DELETE || 888 (wantparent && cnp->cn_nameiop != CREATE && 889 cnp->cn_nameiop != LOOKUP)) 890 docache = 0; 891 rdonly = cnp->cn_flags & RDONLY; 892 cnp->cn_flags &= ~ISSYMLINK; 893 ndp->ni_dvp = NULL; 894 /* 895 * We use shared locks until we hit the parent of the last cn then 896 * we adjust based on the requesting flags. 897 */ 898 cnp->cn_lkflags = LK_SHARED; 899 dp = ndp->ni_startdir; 900 ndp->ni_startdir = NULLVP; 901 vn_lock(dp, 902 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY, 903 cnp->cn_flags)); 904 905 dirloop: 906 /* 907 * Search a new directory. 908 * 909 * The last component of the filename is left accessible via 910 * cnp->cn_nameptr for callers that need the name. Callers needing 911 * the name set the SAVENAME flag. When done, they assume 912 * responsibility for freeing the pathname buffer. 913 * 914 * Store / as a temporary sentinel so that we only have one character 915 * to test for. Pathnames tend to be short so this should not be 916 * resulting in cache misses. 917 */ 918 nulchar = &cnp->cn_nameptr[ndp->ni_pathlen - 1]; 919 KASSERT(*nulchar == '\0', 920 ("%s: expected nul at %p; string [%s]\n", __func__, nulchar, 921 cnp->cn_pnbuf)); 922 *nulchar = '/'; 923 for (cp = cnp->cn_nameptr; *cp != '/'; cp++) { 924 KASSERT(*cp != '\0', 925 ("%s: encountered unexpected nul; string [%s]\n", __func__, 926 cnp->cn_nameptr)); 927 continue; 928 } 929 *nulchar = '\0'; 930 cnp->cn_namelen = cp - cnp->cn_nameptr; 931 if (cnp->cn_namelen > NAME_MAX) { 932 error = ENAMETOOLONG; 933 goto bad; 934 } 935 #ifdef NAMEI_DIAGNOSTIC 936 { char c = *cp; 937 *cp = '\0'; 938 printf("{%s}: ", cnp->cn_nameptr); 939 *cp = c; } 940 #endif 941 prev_ni_pathlen = ndp->ni_pathlen; 942 ndp->ni_pathlen -= cnp->cn_namelen; 943 KASSERT(ndp->ni_pathlen <= PATH_MAX, 944 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen)); 945 prev_ni_next = ndp->ni_next; 946 ndp->ni_next = cp; 947 948 /* 949 * Replace multiple slashes by a single slash and trailing slashes 950 * by a null. This must be done before VOP_LOOKUP() because some 951 * fs's don't know about trailing slashes. Remember if there were 952 * trailing slashes to handle symlinks, existing non-directories 953 * and non-existing files that won't be directories specially later. 954 */ 955 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) { 956 cp++; 957 ndp->ni_pathlen--; 958 if (*cp == '\0') { 959 *ndp->ni_next = '\0'; 960 cnp->cn_flags |= TRAILINGSLASH; 961 } 962 } 963 ndp->ni_next = cp; 964 965 cnp->cn_flags |= MAKEENTRY; 966 if (*cp == '\0' && docache == 0) 967 cnp->cn_flags &= ~MAKEENTRY; 968 if (cnp->cn_namelen == 2 && 969 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.') 970 cnp->cn_flags |= ISDOTDOT; 971 else 972 cnp->cn_flags &= ~ISDOTDOT; 973 if (*ndp->ni_next == 0) 974 cnp->cn_flags |= ISLASTCN; 975 else 976 cnp->cn_flags &= ~ISLASTCN; 977 978 if ((cnp->cn_flags & ISLASTCN) != 0 && 979 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' && 980 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 981 error = EINVAL; 982 goto bad; 983 } 984 985 nameicap_tracker_add(ndp, dp); 986 987 /* 988 * Check for degenerate name (e.g. / or "") 989 * which is a way of talking about a directory, 990 * e.g. like "/." or ".". 991 */ 992 if (cnp->cn_nameptr[0] == '\0') { 993 if (dp->v_type != VDIR) { 994 error = ENOTDIR; 995 goto bad; 996 } 997 if (cnp->cn_nameiop != LOOKUP) { 998 error = EISDIR; 999 goto bad; 1000 } 1001 if (wantparent) { 1002 ndp->ni_dvp = dp; 1003 VREF(dp); 1004 } 1005 ndp->ni_vp = dp; 1006 1007 if (cnp->cn_flags & AUDITVNODE1) 1008 AUDIT_ARG_VNODE1(dp); 1009 else if (cnp->cn_flags & AUDITVNODE2) 1010 AUDIT_ARG_VNODE2(dp); 1011 1012 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF))) 1013 VOP_UNLOCK(dp); 1014 /* XXX This should probably move to the top of function. */ 1015 if (cnp->cn_flags & SAVESTART) 1016 panic("lookup: SAVESTART"); 1017 goto success; 1018 } 1019 1020 /* 1021 * Handle "..": five special cases. 1022 * 0. If doing a capability lookup and lookup_cap_dotdot is 1023 * disabled, return ENOTCAPABLE. 1024 * 1. Return an error if this is the last component of 1025 * the name and the operation is DELETE or RENAME. 1026 * 2. If at root directory (e.g. after chroot) 1027 * or at absolute root directory 1028 * then ignore it so can't get out. 1029 * 3. If this vnode is the root of a mounted 1030 * filesystem, then replace it with the 1031 * vnode which was mounted on so we take the 1032 * .. in the other filesystem. 1033 * 4. If the vnode is the top directory of 1034 * the jail or chroot, don't let them out. 1035 * 5. If doing a capability lookup and lookup_cap_dotdot is 1036 * enabled, return ENOTCAPABLE if the lookup would escape 1037 * from the initial file descriptor directory. Checks are 1038 * done by ensuring that namei() already traversed the 1039 * result of dotdot lookup. 1040 */ 1041 if (cnp->cn_flags & ISDOTDOT) { 1042 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT)) 1043 == NI_LCF_STRICTRELATIVE) { 1044 #ifdef KTRACE 1045 if (KTRPOINT(curthread, KTR_CAPFAIL)) 1046 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 1047 #endif 1048 error = ENOTCAPABLE; 1049 goto bad; 1050 } 1051 if ((cnp->cn_flags & ISLASTCN) != 0 && 1052 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 1053 error = EINVAL; 1054 goto bad; 1055 } 1056 for (;;) { 1057 for (pr = cnp->cn_cred->cr_prison; pr != NULL; 1058 pr = pr->pr_parent) 1059 if (dp == pr->pr_root) 1060 break; 1061 if (dp == ndp->ni_rootdir || 1062 dp == ndp->ni_topdir || 1063 dp == rootvnode || 1064 pr != NULL || 1065 ((dp->v_vflag & VV_ROOT) != 0 && 1066 (cnp->cn_flags & NOCROSSMOUNT) != 0)) { 1067 ndp->ni_dvp = dp; 1068 ndp->ni_vp = dp; 1069 VREF(dp); 1070 goto nextname; 1071 } 1072 if ((dp->v_vflag & VV_ROOT) == 0) 1073 break; 1074 if (VN_IS_DOOMED(dp)) { /* forced unmount */ 1075 error = ENOENT; 1076 goto bad; 1077 } 1078 tdp = dp; 1079 dp = dp->v_mount->mnt_vnodecovered; 1080 VREF(dp); 1081 vput(tdp); 1082 vn_lock(dp, 1083 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | 1084 LK_RETRY, ISDOTDOT)); 1085 error = nameicap_check_dotdot(ndp, dp); 1086 if (error != 0) { 1087 #ifdef KTRACE 1088 if (KTRPOINT(curthread, KTR_CAPFAIL)) 1089 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 1090 #endif 1091 goto bad; 1092 } 1093 } 1094 } 1095 1096 /* 1097 * We now have a segment name to search for, and a directory to search. 1098 */ 1099 unionlookup: 1100 #ifdef MAC 1101 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp, cnp); 1102 if (error) 1103 goto bad; 1104 #endif 1105 ndp->ni_dvp = dp; 1106 ndp->ni_vp = NULL; 1107 ASSERT_VOP_LOCKED(dp, "lookup"); 1108 /* 1109 * If we have a shared lock we may need to upgrade the lock for the 1110 * last operation. 1111 */ 1112 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) && 1113 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED) 1114 vn_lock(dp, LK_UPGRADE|LK_RETRY); 1115 if (VN_IS_DOOMED(dp)) { 1116 error = ENOENT; 1117 goto bad; 1118 } 1119 /* 1120 * If we're looking up the last component and we need an exclusive 1121 * lock, adjust our lkflags. 1122 */ 1123 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags)) 1124 cnp->cn_lkflags = LK_EXCLUSIVE; 1125 #ifdef NAMEI_DIAGNOSTIC 1126 vn_printf(dp, "lookup in "); 1127 #endif 1128 lkflags_save = cnp->cn_lkflags; 1129 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags, 1130 cnp->cn_flags); 1131 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp); 1132 cnp->cn_lkflags = lkflags_save; 1133 if (error != 0) { 1134 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty")); 1135 #ifdef NAMEI_DIAGNOSTIC 1136 printf("not found\n"); 1137 #endif 1138 if ((error == ENOENT) && 1139 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) && 1140 (dp->v_mount->mnt_flag & MNT_UNION)) { 1141 tdp = dp; 1142 dp = dp->v_mount->mnt_vnodecovered; 1143 VREF(dp); 1144 vput(tdp); 1145 vn_lock(dp, 1146 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | 1147 LK_RETRY, cnp->cn_flags)); 1148 nameicap_tracker_add(ndp, dp); 1149 goto unionlookup; 1150 } 1151 1152 if (error == ERELOOKUP) { 1153 vref(dp); 1154 ndp->ni_vp = dp; 1155 error = 0; 1156 relookup = 1; 1157 goto good; 1158 } 1159 1160 if (error != EJUSTRETURN) 1161 goto bad; 1162 /* 1163 * At this point, we know we're at the end of the 1164 * pathname. If creating / renaming, we can consider 1165 * allowing the file or directory to be created / renamed, 1166 * provided we're not on a read-only filesystem. 1167 */ 1168 if (rdonly) { 1169 error = EROFS; 1170 goto bad; 1171 } 1172 /* trailing slash only allowed for directories */ 1173 if ((cnp->cn_flags & TRAILINGSLASH) && 1174 !(cnp->cn_flags & WILLBEDIR)) { 1175 error = ENOENT; 1176 goto bad; 1177 } 1178 if ((cnp->cn_flags & LOCKPARENT) == 0) 1179 VOP_UNLOCK(dp); 1180 /* 1181 * We return with ni_vp NULL to indicate that the entry 1182 * doesn't currently exist, leaving a pointer to the 1183 * (possibly locked) directory vnode in ndp->ni_dvp. 1184 */ 1185 if (cnp->cn_flags & SAVESTART) { 1186 ndp->ni_startdir = ndp->ni_dvp; 1187 VREF(ndp->ni_startdir); 1188 } 1189 goto success; 1190 } 1191 1192 good: 1193 #ifdef NAMEI_DIAGNOSTIC 1194 printf("found\n"); 1195 #endif 1196 dp = ndp->ni_vp; 1197 1198 /* 1199 * Check to see if the vnode has been mounted on; 1200 * if so find the root of the mounted filesystem. 1201 */ 1202 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) && 1203 (cnp->cn_flags & NOCROSSMOUNT) == 0) { 1204 if (vfs_busy(mp, 0)) 1205 continue; 1206 vput(dp); 1207 if (dp != ndp->ni_dvp) 1208 vput(ndp->ni_dvp); 1209 else 1210 vrele(ndp->ni_dvp); 1211 vrefact(vp_crossmp); 1212 ndp->ni_dvp = vp_crossmp; 1213 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags, 1214 cnp->cn_flags), &tdp); 1215 vfs_unbusy(mp); 1216 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT)) 1217 panic("vp_crossmp exclusively locked or reclaimed"); 1218 if (error) { 1219 dpunlocked = 1; 1220 goto bad2; 1221 } 1222 ndp->ni_vp = dp = tdp; 1223 } 1224 1225 /* 1226 * Check for symbolic link 1227 */ 1228 if ((dp->v_type == VLNK) && 1229 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) || 1230 *ndp->ni_next == '/')) { 1231 cnp->cn_flags |= ISSYMLINK; 1232 if (VN_IS_DOOMED(dp)) { 1233 /* 1234 * We can't know whether the directory was mounted with 1235 * NOSYMFOLLOW, so we can't follow safely. 1236 */ 1237 error = ENOENT; 1238 goto bad2; 1239 } 1240 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) { 1241 error = EACCES; 1242 goto bad2; 1243 } 1244 /* 1245 * Symlink code always expects an unlocked dvp. 1246 */ 1247 if (ndp->ni_dvp != ndp->ni_vp) { 1248 VOP_UNLOCK(ndp->ni_dvp); 1249 ni_dvp_unlocked = 1; 1250 } 1251 goto success; 1252 } 1253 1254 nextname: 1255 /* 1256 * Not a symbolic link that we will follow. Continue with the 1257 * next component if there is any; otherwise, we're done. 1258 */ 1259 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/', 1260 ("lookup: invalid path state.")); 1261 if (relookup) { 1262 relookup = 0; 1263 ndp->ni_pathlen = prev_ni_pathlen; 1264 ndp->ni_next = prev_ni_next; 1265 if (ndp->ni_dvp != dp) 1266 vput(ndp->ni_dvp); 1267 else 1268 vrele(ndp->ni_dvp); 1269 goto dirloop; 1270 } 1271 if (cnp->cn_flags & ISDOTDOT) { 1272 error = nameicap_check_dotdot(ndp, ndp->ni_vp); 1273 if (error != 0) { 1274 #ifdef KTRACE 1275 if (KTRPOINT(curthread, KTR_CAPFAIL)) 1276 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 1277 #endif 1278 goto bad2; 1279 } 1280 } 1281 if (*ndp->ni_next == '/') { 1282 cnp->cn_nameptr = ndp->ni_next; 1283 while (*cnp->cn_nameptr == '/') { 1284 cnp->cn_nameptr++; 1285 ndp->ni_pathlen--; 1286 } 1287 if (ndp->ni_dvp != dp) 1288 vput(ndp->ni_dvp); 1289 else 1290 vrele(ndp->ni_dvp); 1291 goto dirloop; 1292 } 1293 /* 1294 * If we're processing a path with a trailing slash, 1295 * check that the end result is a directory. 1296 */ 1297 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) { 1298 error = ENOTDIR; 1299 goto bad2; 1300 } 1301 /* 1302 * Disallow directory write attempts on read-only filesystems. 1303 */ 1304 if (rdonly && 1305 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 1306 error = EROFS; 1307 goto bad2; 1308 } 1309 if (cnp->cn_flags & SAVESTART) { 1310 ndp->ni_startdir = ndp->ni_dvp; 1311 VREF(ndp->ni_startdir); 1312 } 1313 if (!wantparent) { 1314 ni_dvp_unlocked = 2; 1315 if (ndp->ni_dvp != dp) 1316 vput(ndp->ni_dvp); 1317 else 1318 vrele(ndp->ni_dvp); 1319 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) { 1320 VOP_UNLOCK(ndp->ni_dvp); 1321 ni_dvp_unlocked = 1; 1322 } 1323 1324 if (cnp->cn_flags & AUDITVNODE1) 1325 AUDIT_ARG_VNODE1(dp); 1326 else if (cnp->cn_flags & AUDITVNODE2) 1327 AUDIT_ARG_VNODE2(dp); 1328 1329 if ((cnp->cn_flags & LOCKLEAF) == 0) 1330 VOP_UNLOCK(dp); 1331 success: 1332 /* 1333 * FIXME: for lookups which only cross a mount point to fetch the 1334 * root vnode, ni_dvp will be set to vp_crossmp. This can be a problem 1335 * if either WANTPARENT or LOCKPARENT is set. 1336 */ 1337 /* 1338 * Because of shared lookup we may have the vnode shared locked, but 1339 * the caller may want it to be exclusively locked. 1340 */ 1341 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) && 1342 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) { 1343 vn_lock(dp, LK_UPGRADE | LK_RETRY); 1344 if (VN_IS_DOOMED(dp)) { 1345 error = ENOENT; 1346 goto bad2; 1347 } 1348 } 1349 if (ndp->ni_vp != NULL) { 1350 if ((cnp->cn_flags & ISDOTDOT) == 0) 1351 nameicap_tracker_add(ndp, ndp->ni_vp); 1352 if ((cnp->cn_flags & (FAILIFEXISTS | ISSYMLINK)) == FAILIFEXISTS) 1353 goto bad_eexist; 1354 } 1355 return (0); 1356 1357 bad2: 1358 if (ni_dvp_unlocked != 2) { 1359 if (dp != ndp->ni_dvp && !ni_dvp_unlocked) 1360 vput(ndp->ni_dvp); 1361 else 1362 vrele(ndp->ni_dvp); 1363 } 1364 bad: 1365 if (!dpunlocked) 1366 vput(dp); 1367 ndp->ni_vp = NULL; 1368 return (error); 1369 bad_eexist: 1370 /* 1371 * FAILIFEXISTS handling. 1372 * 1373 * XXX namei called with LOCKPARENT but not LOCKLEAF has the strange 1374 * behaviour of leaving the vnode unlocked if the target is the same 1375 * vnode as the parent. 1376 */ 1377 MPASS((cnp->cn_flags & ISSYMLINK) == 0); 1378 if (ndp->ni_vp == ndp->ni_dvp) 1379 vrele(ndp->ni_dvp); 1380 else 1381 vput(ndp->ni_dvp); 1382 vrele(ndp->ni_vp); 1383 ndp->ni_dvp = NULL; 1384 ndp->ni_vp = NULL; 1385 NDFREE(ndp, NDF_ONLY_PNBUF); 1386 return (EEXIST); 1387 } 1388 1389 /* 1390 * relookup - lookup a path name component 1391 * Used by lookup to re-acquire things. 1392 */ 1393 int 1394 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) 1395 { 1396 struct vnode *dp = NULL; /* the directory we are searching */ 1397 int rdonly; /* lookup read-only flag bit */ 1398 int error = 0; 1399 1400 KASSERT(cnp->cn_flags & ISLASTCN, 1401 ("relookup: Not given last component.")); 1402 /* 1403 * Setup: break out flag bits into variables. 1404 */ 1405 KASSERT((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) != 0, 1406 ("relookup: parent not wanted")); 1407 rdonly = cnp->cn_flags & RDONLY; 1408 cnp->cn_flags &= ~ISSYMLINK; 1409 dp = dvp; 1410 cnp->cn_lkflags = LK_EXCLUSIVE; 1411 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); 1412 1413 /* 1414 * Search a new directory. 1415 * 1416 * The last component of the filename is left accessible via 1417 * cnp->cn_nameptr for callers that need the name. Callers needing 1418 * the name set the SAVENAME flag. When done, they assume 1419 * responsibility for freeing the pathname buffer. 1420 */ 1421 #ifdef NAMEI_DIAGNOSTIC 1422 printf("{%s}: ", cnp->cn_nameptr); 1423 #endif 1424 1425 /* 1426 * Check for "" which represents the root directory after slash 1427 * removal. 1428 */ 1429 if (cnp->cn_nameptr[0] == '\0') { 1430 /* 1431 * Support only LOOKUP for "/" because lookup() 1432 * can't succeed for CREATE, DELETE and RENAME. 1433 */ 1434 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP")); 1435 KASSERT(dp->v_type == VDIR, ("dp is not a directory")); 1436 1437 if (!(cnp->cn_flags & LOCKLEAF)) 1438 VOP_UNLOCK(dp); 1439 *vpp = dp; 1440 /* XXX This should probably move to the top of function. */ 1441 if (cnp->cn_flags & SAVESTART) 1442 panic("lookup: SAVESTART"); 1443 return (0); 1444 } 1445 1446 if (cnp->cn_flags & ISDOTDOT) 1447 panic ("relookup: lookup on dot-dot"); 1448 1449 /* 1450 * We now have a segment name to search for, and a directory to search. 1451 */ 1452 #ifdef NAMEI_DIAGNOSTIC 1453 vn_printf(dp, "search in "); 1454 #endif 1455 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) { 1456 KASSERT(*vpp == NULL, ("leaf should be empty")); 1457 if (error != EJUSTRETURN) 1458 goto bad; 1459 /* 1460 * If creating and at end of pathname, then can consider 1461 * allowing file to be created. 1462 */ 1463 if (rdonly) { 1464 error = EROFS; 1465 goto bad; 1466 } 1467 /* ASSERT(dvp == ndp->ni_startdir) */ 1468 if (cnp->cn_flags & SAVESTART) 1469 VREF(dvp); 1470 if ((cnp->cn_flags & LOCKPARENT) == 0) 1471 VOP_UNLOCK(dp); 1472 /* 1473 * We return with ni_vp NULL to indicate that the entry 1474 * doesn't currently exist, leaving a pointer to the 1475 * (possibly locked) directory vnode in ndp->ni_dvp. 1476 */ 1477 return (0); 1478 } 1479 1480 dp = *vpp; 1481 1482 /* 1483 * Disallow directory write attempts on read-only filesystems. 1484 */ 1485 if (rdonly && 1486 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 1487 if (dvp == dp) 1488 vrele(dvp); 1489 else 1490 vput(dvp); 1491 error = EROFS; 1492 goto bad; 1493 } 1494 /* 1495 * Set the parent lock/ref state to the requested state. 1496 */ 1497 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) 1498 VOP_UNLOCK(dvp); 1499 /* 1500 * Check for symbolic link 1501 */ 1502 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW), 1503 ("relookup: symlink found.\n")); 1504 1505 /* ASSERT(dvp == ndp->ni_startdir) */ 1506 if (cnp->cn_flags & SAVESTART) 1507 VREF(dvp); 1508 1509 if ((cnp->cn_flags & LOCKLEAF) == 0) 1510 VOP_UNLOCK(dp); 1511 return (0); 1512 bad: 1513 vput(dp); 1514 *vpp = NULL; 1515 return (error); 1516 } 1517 1518 /* 1519 * Free data allocated by namei(); see namei(9) for details. 1520 */ 1521 void 1522 NDFREE_PNBUF(struct nameidata *ndp) 1523 { 1524 1525 if ((ndp->ni_cnd.cn_flags & HASBUF) != 0) { 1526 MPASS((ndp->ni_cnd.cn_flags & (SAVENAME | SAVESTART)) != 0); 1527 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 1528 ndp->ni_cnd.cn_flags &= ~HASBUF; 1529 } 1530 } 1531 1532 /* 1533 * NDFREE_PNBUF replacement for callers that know there is no buffer. 1534 * 1535 * This is a hack. Preferably the VFS layer would not produce anything more 1536 * than it was asked to do. Unfortunately several non-LOOKUP cases can add the 1537 * HASBUF flag to the result. Even then an interface could be implemented where 1538 * the caller specifies what they expected to see in the result and what they 1539 * are going to take care of. 1540 * 1541 * In the meantime provide this kludge as a trivial replacement for NDFREE_PNBUF 1542 * calls scattered throughout the kernel where we know for a fact the flag must not 1543 * be seen. 1544 */ 1545 #ifdef INVARIANTS 1546 void 1547 NDFREE_NOTHING(struct nameidata *ndp) 1548 { 1549 struct componentname *cnp; 1550 1551 cnp = &ndp->ni_cnd; 1552 KASSERT(cnp->cn_nameiop == LOOKUP, ("%s: got non-LOOKUP op %d\n", 1553 __func__, cnp->cn_nameiop)); 1554 KASSERT((cnp->cn_flags & (SAVENAME | HASBUF)) == 0, 1555 ("%s: bad flags \%" PRIx64 "\n", __func__, cnp->cn_flags)); 1556 } 1557 #endif 1558 1559 void 1560 (NDFREE)(struct nameidata *ndp, const u_int flags) 1561 { 1562 int unlock_dvp; 1563 int unlock_vp; 1564 1565 unlock_dvp = 0; 1566 unlock_vp = 0; 1567 1568 if (!(flags & NDF_NO_FREE_PNBUF)) { 1569 NDFREE_PNBUF(ndp); 1570 } 1571 if (!(flags & NDF_NO_VP_UNLOCK) && 1572 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) 1573 unlock_vp = 1; 1574 if (!(flags & NDF_NO_DVP_UNLOCK) && 1575 (ndp->ni_cnd.cn_flags & LOCKPARENT) && 1576 ndp->ni_dvp != ndp->ni_vp) 1577 unlock_dvp = 1; 1578 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) { 1579 if (unlock_vp) { 1580 vput(ndp->ni_vp); 1581 unlock_vp = 0; 1582 } else 1583 vrele(ndp->ni_vp); 1584 ndp->ni_vp = NULL; 1585 } 1586 if (unlock_vp) 1587 VOP_UNLOCK(ndp->ni_vp); 1588 if (!(flags & NDF_NO_DVP_RELE) && 1589 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { 1590 if (unlock_dvp) { 1591 vput(ndp->ni_dvp); 1592 unlock_dvp = 0; 1593 } else 1594 vrele(ndp->ni_dvp); 1595 ndp->ni_dvp = NULL; 1596 } 1597 if (unlock_dvp) 1598 VOP_UNLOCK(ndp->ni_dvp); 1599 if (!(flags & NDF_NO_STARTDIR_RELE) && 1600 (ndp->ni_cnd.cn_flags & SAVESTART)) { 1601 vrele(ndp->ni_startdir); 1602 ndp->ni_startdir = NULL; 1603 } 1604 } 1605 1606 #ifdef INVARIANTS 1607 /* 1608 * Validate the final state of ndp after the lookup. 1609 * 1610 * Historically filesystems were allowed to modify cn_flags. Most notably they 1611 * can add SAVENAME to the request, resulting in HASBUF and pushing subsequent 1612 * clean up to the consumer. In practice this seems to only concern != LOOKUP 1613 * operations. 1614 * 1615 * As a step towards stricter API contract this routine validates the state to 1616 * clean up. Note validation is a work in progress with the intent of becoming 1617 * stricter over time. 1618 */ 1619 #define NDMODIFYINGFLAGS (LOCKLEAF | LOCKPARENT | WANTPARENT | SAVENAME | SAVESTART | HASBUF) 1620 void 1621 NDVALIDATE(struct nameidata *ndp) 1622 { 1623 struct componentname *cnp; 1624 u_int64_t used, orig; 1625 1626 cnp = &ndp->ni_cnd; 1627 orig = cnp->cn_origflags; 1628 used = cnp->cn_flags; 1629 switch (cnp->cn_nameiop) { 1630 case LOOKUP: 1631 /* 1632 * For plain lookup we require strict conformance -- nothing 1633 * to clean up if it was not requested by the caller. 1634 */ 1635 orig &= NDMODIFYINGFLAGS; 1636 used &= NDMODIFYINGFLAGS; 1637 if ((orig & (SAVENAME | SAVESTART)) != 0) 1638 orig |= HASBUF; 1639 if (orig != used) { 1640 goto out_mismatch; 1641 } 1642 break; 1643 case CREATE: 1644 case DELETE: 1645 case RENAME: 1646 /* 1647 * Some filesystems set SAVENAME to provoke HASBUF, accomodate 1648 * for it until it gets fixed. 1649 */ 1650 orig &= NDMODIFYINGFLAGS; 1651 orig |= (SAVENAME | HASBUF); 1652 used &= NDMODIFYINGFLAGS; 1653 used |= (SAVENAME | HASBUF); 1654 if (orig != used) { 1655 goto out_mismatch; 1656 } 1657 break; 1658 } 1659 return; 1660 out_mismatch: 1661 panic("%s: mismatched flags for op %d: added %" PRIx64 ", " 1662 "removed %" PRIx64" (%" PRIx64" != %" PRIx64"; stored %" PRIx64" != %" PRIx64")", 1663 __func__, cnp->cn_nameiop, used & ~orig, orig &~ used, 1664 orig, used, cnp->cn_origflags, cnp->cn_flags); 1665 } 1666 #endif 1667 1668 /* 1669 * Determine if there is a suitable alternate filename under the specified 1670 * prefix for the specified path. If the create flag is set, then the 1671 * alternate prefix will be used so long as the parent directory exists. 1672 * This is used by the various compatibility ABIs so that Linux binaries prefer 1673 * files under /compat/linux for example. The chosen path (whether under 1674 * the prefix or under /) is returned in a kernel malloc'd buffer pointed 1675 * to by pathbuf. The caller is responsible for free'ing the buffer from 1676 * the M_TEMP bucket if one is returned. 1677 */ 1678 int 1679 kern_alternate_path(struct thread *td, const char *prefix, const char *path, 1680 enum uio_seg pathseg, char **pathbuf, int create, int dirfd) 1681 { 1682 struct nameidata nd, ndroot; 1683 char *ptr, *buf, *cp; 1684 size_t len, sz; 1685 int error; 1686 1687 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1688 *pathbuf = buf; 1689 1690 /* Copy the prefix into the new pathname as a starting point. */ 1691 len = strlcpy(buf, prefix, MAXPATHLEN); 1692 if (len >= MAXPATHLEN) { 1693 *pathbuf = NULL; 1694 free(buf, M_TEMP); 1695 return (EINVAL); 1696 } 1697 sz = MAXPATHLEN - len; 1698 ptr = buf + len; 1699 1700 /* Append the filename to the prefix. */ 1701 if (pathseg == UIO_SYSSPACE) 1702 error = copystr(path, ptr, sz, &len); 1703 else 1704 error = copyinstr(path, ptr, sz, &len); 1705 1706 if (error) { 1707 *pathbuf = NULL; 1708 free(buf, M_TEMP); 1709 return (error); 1710 } 1711 1712 /* Only use a prefix with absolute pathnames. */ 1713 if (*ptr != '/') { 1714 error = EINVAL; 1715 goto keeporig; 1716 } 1717 1718 if (dirfd != AT_FDCWD) { 1719 /* 1720 * We want the original because the "prefix" is 1721 * included in the already opened dirfd. 1722 */ 1723 bcopy(ptr, buf, len); 1724 return (0); 1725 } 1726 1727 /* 1728 * We know that there is a / somewhere in this pathname. 1729 * Search backwards for it, to find the file's parent dir 1730 * to see if it exists in the alternate tree. If it does, 1731 * and we want to create a file (cflag is set). We don't 1732 * need to worry about the root comparison in this case. 1733 */ 1734 1735 if (create) { 1736 for (cp = &ptr[len] - 1; *cp != '/'; cp--); 1737 *cp = '\0'; 1738 1739 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td); 1740 error = namei(&nd); 1741 *cp = '/'; 1742 if (error != 0) 1743 goto keeporig; 1744 } else { 1745 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td); 1746 1747 error = namei(&nd); 1748 if (error != 0) 1749 goto keeporig; 1750 1751 /* 1752 * We now compare the vnode of the prefix to the one 1753 * vnode asked. If they resolve to be the same, then we 1754 * ignore the match so that the real root gets used. 1755 * This avoids the problem of traversing "../.." to find the 1756 * root directory and never finding it, because "/" resolves 1757 * to the emulation root directory. This is expensive :-( 1758 */ 1759 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix, 1760 td); 1761 1762 /* We shouldn't ever get an error from this namei(). */ 1763 error = namei(&ndroot); 1764 if (error == 0) { 1765 if (nd.ni_vp == ndroot.ni_vp) 1766 error = ENOENT; 1767 1768 NDFREE(&ndroot, NDF_ONLY_PNBUF); 1769 vrele(ndroot.ni_vp); 1770 } 1771 } 1772 1773 NDFREE(&nd, NDF_ONLY_PNBUF); 1774 vrele(nd.ni_vp); 1775 1776 keeporig: 1777 /* If there was an error, use the original path name. */ 1778 if (error) 1779 bcopy(ptr, buf, len); 1780 return (error); 1781 } 1782