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/kernel.h> 48 #include <sys/capsicum.h> 49 #include <sys/fcntl.h> 50 #include <sys/jail.h> 51 #include <sys/lock.h> 52 #include <sys/mutex.h> 53 #include <sys/namei.h> 54 #include <sys/vnode.h> 55 #include <sys/mount.h> 56 #include <sys/filedesc.h> 57 #include <sys/proc.h> 58 #include <sys/sdt.h> 59 #include <sys/syscallsubr.h> 60 #include <sys/sysctl.h> 61 #ifdef KTRACE 62 #include <sys/ktrace.h> 63 #endif 64 #ifdef INVARIANTS 65 #include <machine/_inttypes.h> 66 #endif 67 68 #include <security/audit/audit.h> 69 #include <security/mac/mac_framework.h> 70 71 #include <vm/uma.h> 72 73 #define NAMEI_DIAGNOSTIC 1 74 #undef NAMEI_DIAGNOSTIC 75 76 SDT_PROVIDER_DEFINE(vfs); 77 SDT_PROBE_DEFINE4(vfs, namei, lookup, entry, "struct vnode *", "char *", 78 "unsigned long", "bool"); 79 SDT_PROBE_DEFINE3(vfs, namei, lookup, return, "int", "struct vnode *", "bool"); 80 81 /* Allocation zone for namei. */ 82 uma_zone_t namei_zone; 83 84 /* Placeholder vnode for mp traversal. */ 85 static struct vnode *vp_crossmp; 86 87 static int 88 crossmp_vop_islocked(struct vop_islocked_args *ap) 89 { 90 91 return (LK_SHARED); 92 } 93 94 static int 95 crossmp_vop_lock1(struct vop_lock1_args *ap) 96 { 97 struct vnode *vp; 98 struct lock *lk __unused; 99 const char *file __unused; 100 int flags, line __unused; 101 102 vp = ap->a_vp; 103 lk = vp->v_vnlock; 104 flags = ap->a_flags; 105 file = ap->a_file; 106 line = ap->a_line; 107 108 if ((flags & LK_SHARED) == 0) 109 panic("invalid lock request for crossmp"); 110 111 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, file, line, 112 flags & LK_INTERLOCK ? &VI_MTX(vp)->lock_object : NULL); 113 WITNESS_LOCK(&lk->lock_object, 0, file, line); 114 if ((flags & LK_INTERLOCK) != 0) 115 VI_UNLOCK(vp); 116 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, ap->a_file, line); 117 return (0); 118 } 119 120 static int 121 crossmp_vop_unlock(struct vop_unlock_args *ap) 122 { 123 struct vnode *vp; 124 struct lock *lk __unused; 125 126 vp = ap->a_vp; 127 lk = vp->v_vnlock; 128 129 WITNESS_UNLOCK(&lk->lock_object, 0, LOCK_FILE, LOCK_LINE); 130 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE, 131 LOCK_LINE); 132 return (0); 133 } 134 135 static struct vop_vector crossmp_vnodeops = { 136 .vop_default = &default_vnodeops, 137 .vop_islocked = crossmp_vop_islocked, 138 .vop_lock1 = crossmp_vop_lock1, 139 .vop_unlock = crossmp_vop_unlock, 140 }; 141 /* 142 * VFS_VOP_VECTOR_REGISTER(crossmp_vnodeops) is not used here since the vnode 143 * gets allocated early. See nameiinit for the direct call below. 144 */ 145 146 struct nameicap_tracker { 147 struct vnode *dp; 148 TAILQ_ENTRY(nameicap_tracker) nm_link; 149 }; 150 151 /* Zone for cap mode tracker elements used for dotdot capability checks. */ 152 static uma_zone_t nt_zone; 153 154 static void 155 nameiinit(void *dummy __unused) 156 { 157 158 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL, 159 UMA_ALIGN_PTR, 0); 160 nt_zone = uma_zcreate("rentr", sizeof(struct nameicap_tracker), 161 NULL, NULL, NULL, NULL, 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 if ((cnp->cn_flags & BENEATH) != 0 && 187 (ndp->ni_lcf & NI_LCF_BENEATH_LATCHED) == 0) { 188 MPASS((ndp->ni_lcf & NI_LCF_LATCH) != 0); 189 if (dp != ndp->ni_beneath_latch) 190 return; 191 ndp->ni_lcf |= NI_LCF_BENEATH_LATCHED; 192 } 193 nt = uma_zalloc(nt_zone, M_WAITOK); 194 vhold(dp); 195 nt->dp = dp; 196 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link); 197 } 198 199 static void 200 nameicap_cleanup(struct nameidata *ndp, bool clean_latch) 201 { 202 struct nameicap_tracker *nt, *nt1; 203 204 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) || 205 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative")); 206 TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) { 207 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link); 208 vdrop(nt->dp); 209 uma_zfree(nt_zone, nt); 210 } 211 if (clean_latch && (ndp->ni_lcf & NI_LCF_LATCH) != 0) { 212 ndp->ni_lcf &= ~NI_LCF_LATCH; 213 vrele(ndp->ni_beneath_latch); 214 } 215 } 216 217 /* 218 * For dotdot lookups in capability mode, only allow the component 219 * lookup to succeed if the resulting directory was already traversed 220 * during the operation. This catches situations where already 221 * traversed directory is moved to different parent, and then we walk 222 * over it with dotdots. 223 * 224 * Also allow to force failure of dotdot lookups for non-local 225 * filesystems, where external agents might assist local lookups to 226 * escape the compartment. 227 */ 228 static int 229 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp) 230 { 231 struct nameicap_tracker *nt; 232 struct mount *mp; 233 234 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp == NULL || 235 dp->v_type != VDIR) 236 return (0); 237 mp = dp->v_mount; 238 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL && 239 (mp->mnt_flag & MNT_LOCAL) == 0) 240 return (ENOTCAPABLE); 241 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head, 242 nm_link) { 243 if ((ndp->ni_lcf & NI_LCF_LATCH) != 0 && 244 ndp->ni_beneath_latch == nt->dp) { 245 ndp->ni_lcf &= ~NI_LCF_BENEATH_LATCHED; 246 nameicap_cleanup(ndp, false); 247 return (0); 248 } 249 if (dp == nt->dp) 250 return (0); 251 } 252 return (ENOTCAPABLE); 253 } 254 255 static void 256 namei_cleanup_cnp(struct componentname *cnp) 257 { 258 259 uma_zfree(namei_zone, cnp->cn_pnbuf); 260 #ifdef DIAGNOSTIC 261 cnp->cn_pnbuf = NULL; 262 cnp->cn_nameptr = NULL; 263 #endif 264 } 265 266 static int 267 namei_handle_root(struct nameidata *ndp, struct vnode **dpp) 268 { 269 struct componentname *cnp; 270 271 cnp = &ndp->ni_cnd; 272 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) { 273 #ifdef KTRACE 274 if (KTRPOINT(curthread, KTR_CAPFAIL)) 275 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 276 #endif 277 return (ENOTCAPABLE); 278 } 279 if ((cnp->cn_flags & BENEATH) != 0) { 280 ndp->ni_lcf |= NI_LCF_BENEATH_ABS; 281 ndp->ni_lcf &= ~NI_LCF_BENEATH_LATCHED; 282 nameicap_cleanup(ndp, false); 283 } 284 while (*(cnp->cn_nameptr) == '/') { 285 cnp->cn_nameptr++; 286 ndp->ni_pathlen--; 287 } 288 *dpp = ndp->ni_rootdir; 289 vrefact(*dpp); 290 return (0); 291 } 292 293 static int 294 namei_setup(struct nameidata *ndp, struct vnode **dpp, struct pwd **pwdp) 295 { 296 struct componentname *cnp; 297 struct file *dfp; 298 struct thread *td; 299 struct pwd *pwd; 300 cap_rights_t rights; 301 struct filecaps dirfd_caps; 302 int error; 303 bool startdir_used; 304 305 cnp = &ndp->ni_cnd; 306 td = cnp->cn_thread; 307 308 startdir_used = false; 309 *pwdp = NULL; 310 *dpp = NULL; 311 312 #ifdef CAPABILITY_MODE 313 /* 314 * In capability mode, lookups must be restricted to happen in 315 * the subtree with the root specified by the file descriptor: 316 * - The root must be real file descriptor, not the pseudo-descriptor 317 * AT_FDCWD. 318 * - The passed path must be relative and not absolute. 319 * - If lookup_cap_dotdot is disabled, path must not contain the 320 * '..' components. 321 * - If lookup_cap_dotdot is enabled, we verify that all '..' 322 * components lookups result in the directories which were 323 * previously walked by us, which prevents an escape from 324 * the relative root. 325 */ 326 if (IN_CAPABILITY_MODE(td) && (cnp->cn_flags & NOCAPCHECK) == 0) { 327 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE; 328 ndp->ni_resflags |= NIRES_STRICTREL; 329 if (ndp->ni_dirfd == AT_FDCWD) { 330 #ifdef KTRACE 331 if (KTRPOINT(td, KTR_CAPFAIL)) 332 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 333 #endif 334 return (ECAPMODE); 335 } 336 } 337 #endif 338 error = 0; 339 340 /* 341 * Get starting point for the translation. 342 */ 343 pwd = pwd_hold(td); 344 /* 345 * The reference on ni_rootdir is acquired in the block below to avoid 346 * back-to-back atomics for absolute lookups. 347 */ 348 ndp->ni_rootdir = pwd->pwd_rdir; 349 ndp->ni_topdir = pwd->pwd_jdir; 350 351 if (cnp->cn_pnbuf[0] == '/') { 352 ndp->ni_resflags |= NIRES_ABS; 353 error = namei_handle_root(ndp, dpp); 354 } else { 355 if (ndp->ni_startdir != NULL) { 356 *dpp = ndp->ni_startdir; 357 startdir_used = true; 358 } else if (ndp->ni_dirfd == AT_FDCWD) { 359 *dpp = pwd->pwd_cdir; 360 vrefact(*dpp); 361 } else { 362 rights = *ndp->ni_rightsneeded; 363 cap_rights_set_one(&rights, CAP_LOOKUP); 364 365 if (cnp->cn_flags & AUDITVNODE1) 366 AUDIT_ARG_ATFD1(ndp->ni_dirfd); 367 if (cnp->cn_flags & AUDITVNODE2) 368 AUDIT_ARG_ATFD2(ndp->ni_dirfd); 369 /* 370 * Effectively inlined fgetvp_rights, because we need to 371 * inspect the file as well as grabbing the vnode. 372 */ 373 error = fget_cap(td, ndp->ni_dirfd, &rights, 374 &dfp, &ndp->ni_filecaps); 375 if (error != 0) { 376 /* 377 * Preserve the error; it should either be EBADF 378 * or capability-related, both of which can be 379 * safely returned to the caller. 380 */ 381 } else { 382 if (dfp->f_ops == &badfileops) { 383 error = EBADF; 384 } else if (dfp->f_vnode == NULL) { 385 error = ENOTDIR; 386 } else { 387 *dpp = dfp->f_vnode; 388 vrefact(*dpp); 389 390 if ((dfp->f_flag & FSEARCH) != 0) 391 cnp->cn_flags |= NOEXECCHECK; 392 } 393 fdrop(dfp, td); 394 } 395 #ifdef CAPABILITIES 396 /* 397 * If file descriptor doesn't have all rights, 398 * all lookups relative to it must also be 399 * strictly relative. 400 */ 401 CAP_ALL(&rights); 402 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, 403 &rights) || 404 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || 405 ndp->ni_filecaps.fc_nioctls != -1) { 406 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE; 407 ndp->ni_resflags |= NIRES_STRICTREL; 408 } 409 #endif 410 } 411 if (error == 0 && (*dpp)->v_type != VDIR) 412 error = ENOTDIR; 413 } 414 if (error == 0 && (cnp->cn_flags & BENEATH) != 0) { 415 if (ndp->ni_dirfd == AT_FDCWD) { 416 ndp->ni_beneath_latch = pwd->pwd_cdir; 417 vrefact(ndp->ni_beneath_latch); 418 } else { 419 rights = *ndp->ni_rightsneeded; 420 cap_rights_set_one(&rights, CAP_LOOKUP); 421 error = fgetvp_rights(td, ndp->ni_dirfd, &rights, 422 &dirfd_caps, &ndp->ni_beneath_latch); 423 if (error == 0 && (*dpp)->v_type != VDIR) { 424 vrele(ndp->ni_beneath_latch); 425 error = ENOTDIR; 426 } 427 } 428 if (error == 0) 429 ndp->ni_lcf |= NI_LCF_LATCH; 430 } 431 if (error == 0 && (cnp->cn_flags & RBENEATH) != 0) { 432 if (cnp->cn_pnbuf[0] == '/' || 433 (ndp->ni_lcf & NI_LCF_BENEATH_ABS) != 0) { 434 error = EINVAL; 435 } else if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0) { 436 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE | 437 NI_LCF_CAP_DOTDOT; 438 } 439 } 440 441 /* 442 * If we are auditing the kernel pathname, save the user pathname. 443 */ 444 if (cnp->cn_flags & AUDITVNODE1) 445 AUDIT_ARG_UPATH1_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf); 446 if (cnp->cn_flags & AUDITVNODE2) 447 AUDIT_ARG_UPATH2_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf); 448 if (ndp->ni_startdir != NULL && !startdir_used) 449 vrele(ndp->ni_startdir); 450 if (error != 0) { 451 if (*dpp != NULL) 452 vrele(*dpp); 453 pwd_drop(pwd); 454 return (error); 455 } 456 MPASS((ndp->ni_lcf & (NI_LCF_BENEATH_ABS | NI_LCF_LATCH)) != 457 NI_LCF_BENEATH_ABS); 458 if (((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 && 459 lookup_cap_dotdot != 0) || 460 ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0 && 461 (cnp->cn_flags & BENEATH) != 0)) 462 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT; 463 SDT_PROBE4(vfs, namei, lookup, entry, *dpp, cnp->cn_pnbuf, 464 cnp->cn_flags, false); 465 *pwdp = pwd; 466 return (0); 467 } 468 469 /* 470 * Convert a pathname into a pointer to a locked vnode. 471 * 472 * The FOLLOW flag is set when symbolic links are to be followed 473 * when they occur at the end of the name translation process. 474 * Symbolic links are always followed for all other pathname 475 * components other than the last. 476 * 477 * The segflg defines whether the name is to be copied from user 478 * space or kernel space. 479 * 480 * Overall outline of namei: 481 * 482 * copy in name 483 * get starting directory 484 * while (!done && !error) { 485 * call lookup to search path. 486 * if symbolic link, massage name in buffer and continue 487 * } 488 */ 489 int 490 namei(struct nameidata *ndp) 491 { 492 char *cp; /* pointer into pathname argument */ 493 struct vnode *dp; /* the directory we are searching */ 494 struct iovec aiov; /* uio for reading symbolic links */ 495 struct componentname *cnp; 496 struct thread *td; 497 struct pwd *pwd; 498 struct uio auio; 499 int error, linklen; 500 enum cache_fpl_status status; 501 502 cnp = &ndp->ni_cnd; 503 td = cnp->cn_thread; 504 #ifdef INVARIANTS 505 /* 506 * For NDVALIDATE. 507 * 508 * While NDINIT may seem like a more natural place to do it, there are 509 * callers which directly modify flags past invoking init. 510 */ 511 cnp->cn_origflags = cnp->cn_flags; 512 #endif 513 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred; 514 KASSERT(cnp->cn_cred && td->td_proc, ("namei: bad cred/proc")); 515 KASSERT((cnp->cn_flags & NAMEI_INTERNAL_FLAGS) == 0, 516 ("namei: unexpected flags: %" PRIx64 "\n", 517 cnp->cn_flags & NAMEI_INTERNAL_FLAGS)); 518 if (cnp->cn_flags & NOCACHE) 519 KASSERT(cnp->cn_nameiop != LOOKUP, 520 ("%s: NOCACHE passed with LOOKUP", __func__)); 521 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR || 522 ndp->ni_startdir->v_type == VBAD); 523 524 ndp->ni_lcf = 0; 525 ndp->ni_vp = NULL; 526 527 /* 528 * Get a buffer for the name to be translated, and copy the 529 * name into the buffer. 530 */ 531 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 532 if (ndp->ni_segflg == UIO_SYSSPACE) 533 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN, 534 &ndp->ni_pathlen); 535 else 536 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN, 537 &ndp->ni_pathlen); 538 539 if (__predict_false(error != 0)) { 540 namei_cleanup_cnp(cnp); 541 return (error); 542 } 543 544 /* 545 * Don't allow empty pathnames. 546 */ 547 if (__predict_false(*cnp->cn_pnbuf == '\0')) { 548 namei_cleanup_cnp(cnp); 549 return (ENOENT); 550 } 551 552 #ifdef KTRACE 553 if (KTRPOINT(td, KTR_NAMEI)) { 554 KASSERT(cnp->cn_thread == curthread, 555 ("namei not using curthread")); 556 ktrnamei(cnp->cn_pnbuf); 557 } 558 #endif 559 560 cnp->cn_nameptr = cnp->cn_pnbuf; 561 562 /* 563 * First try looking up the target without locking any vnodes. 564 * 565 * We may need to start from scratch or pick up where it left off. 566 */ 567 error = cache_fplookup(ndp, &status, &pwd); 568 switch (status) { 569 case CACHE_FPL_STATUS_UNSET: 570 __assert_unreachable(); 571 break; 572 case CACHE_FPL_STATUS_HANDLED: 573 if (error == 0) 574 NDVALIDATE(ndp); 575 return (error); 576 case CACHE_FPL_STATUS_PARTIAL: 577 TAILQ_INIT(&ndp->ni_cap_tracker); 578 dp = ndp->ni_startdir; 579 break; 580 case CACHE_FPL_STATUS_ABORTED: 581 TAILQ_INIT(&ndp->ni_cap_tracker); 582 error = namei_setup(ndp, &dp, &pwd); 583 if (error != 0) { 584 namei_cleanup_cnp(cnp); 585 return (error); 586 } 587 break; 588 } 589 590 ndp->ni_loopcnt = 0; 591 592 /* 593 * Locked lookup. 594 */ 595 for (;;) { 596 ndp->ni_startdir = dp; 597 error = lookup(ndp); 598 if (error != 0) { 599 /* 600 * Override an error to not allow user to use 601 * BENEATH as an oracle. 602 */ 603 if ((ndp->ni_lcf & (NI_LCF_LATCH | 604 NI_LCF_BENEATH_LATCHED)) == NI_LCF_LATCH) 605 error = ENOTCAPABLE; 606 goto out; 607 } 608 609 /* 610 * If not a symbolic link, we're done. 611 */ 612 if ((cnp->cn_flags & ISSYMLINK) == 0) { 613 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) { 614 namei_cleanup_cnp(cnp); 615 } else 616 cnp->cn_flags |= HASBUF; 617 if ((ndp->ni_lcf & (NI_LCF_LATCH | 618 NI_LCF_BENEATH_LATCHED)) == NI_LCF_LATCH) { 619 NDFREE(ndp, 0); 620 error = ENOTCAPABLE; 621 } 622 nameicap_cleanup(ndp, true); 623 SDT_PROBE3(vfs, namei, lookup, return, error, 624 (error == 0 ? ndp->ni_vp : NULL), false); 625 pwd_drop(pwd); 626 if (error == 0) 627 NDVALIDATE(ndp); 628 return (error); 629 } 630 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 631 error = ELOOP; 632 break; 633 } 634 #ifdef MAC 635 if ((cnp->cn_flags & NOMACCHECK) == 0) { 636 error = mac_vnode_check_readlink(td->td_ucred, 637 ndp->ni_vp); 638 if (error != 0) 639 break; 640 } 641 #endif 642 if (ndp->ni_pathlen > 1) 643 cp = uma_zalloc(namei_zone, M_WAITOK); 644 else 645 cp = cnp->cn_pnbuf; 646 aiov.iov_base = cp; 647 aiov.iov_len = MAXPATHLEN; 648 auio.uio_iov = &aiov; 649 auio.uio_iovcnt = 1; 650 auio.uio_offset = 0; 651 auio.uio_rw = UIO_READ; 652 auio.uio_segflg = UIO_SYSSPACE; 653 auio.uio_td = td; 654 auio.uio_resid = MAXPATHLEN; 655 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 656 if (error != 0) { 657 if (ndp->ni_pathlen > 1) 658 uma_zfree(namei_zone, cp); 659 break; 660 } 661 linklen = MAXPATHLEN - auio.uio_resid; 662 if (linklen == 0) { 663 if (ndp->ni_pathlen > 1) 664 uma_zfree(namei_zone, cp); 665 error = ENOENT; 666 break; 667 } 668 if (linklen + ndp->ni_pathlen > MAXPATHLEN) { 669 if (ndp->ni_pathlen > 1) 670 uma_zfree(namei_zone, cp); 671 error = ENAMETOOLONG; 672 break; 673 } 674 if (ndp->ni_pathlen > 1) { 675 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen); 676 uma_zfree(namei_zone, cnp->cn_pnbuf); 677 cnp->cn_pnbuf = cp; 678 } else 679 cnp->cn_pnbuf[linklen] = '\0'; 680 ndp->ni_pathlen += linklen; 681 vput(ndp->ni_vp); 682 dp = ndp->ni_dvp; 683 /* 684 * Check if root directory should replace current directory. 685 */ 686 cnp->cn_nameptr = cnp->cn_pnbuf; 687 if (*(cnp->cn_nameptr) == '/') { 688 vrele(dp); 689 error = namei_handle_root(ndp, &dp); 690 if (error != 0) 691 goto out; 692 } 693 } 694 vput(ndp->ni_vp); 695 ndp->ni_vp = NULL; 696 vrele(ndp->ni_dvp); 697 out: 698 MPASS(error != 0); 699 namei_cleanup_cnp(cnp); 700 nameicap_cleanup(ndp, true); 701 SDT_PROBE3(vfs, namei, lookup, return, error, NULL, false); 702 pwd_drop(pwd); 703 return (error); 704 } 705 706 static int 707 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags) 708 { 709 710 if (mp == NULL || ((lkflags & LK_SHARED) && 711 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) || 712 ((cnflags & ISDOTDOT) && 713 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) { 714 lkflags &= ~LK_SHARED; 715 lkflags |= LK_EXCLUSIVE; 716 } 717 lkflags |= LK_NODDLKTREAT; 718 return (lkflags); 719 } 720 721 static __inline int 722 needs_exclusive_leaf(struct mount *mp, int flags) 723 { 724 725 /* 726 * Intermediate nodes can use shared locks, we only need to 727 * force an exclusive lock for leaf nodes. 728 */ 729 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF)) 730 return (0); 731 732 /* Always use exclusive locks if LOCKSHARED isn't set. */ 733 if (!(flags & LOCKSHARED)) 734 return (1); 735 736 /* 737 * For lookups during open(), if the mount point supports 738 * extended shared operations, then use a shared lock for the 739 * leaf node, otherwise use an exclusive lock. 740 */ 741 if ((flags & ISOPEN) != 0) 742 return (!MNT_EXTENDED_SHARED(mp)); 743 744 /* 745 * Lookup requests outside of open() that specify LOCKSHARED 746 * only need a shared lock on the leaf vnode. 747 */ 748 return (0); 749 } 750 751 /* 752 * Search a pathname. 753 * This is a very central and rather complicated routine. 754 * 755 * The pathname is pointed to by ni_ptr and is of length ni_pathlen. 756 * The starting directory is taken from ni_startdir. The pathname is 757 * descended until done, or a symbolic link is encountered. The variable 758 * ni_more is clear if the path is completed; it is set to one if a 759 * symbolic link needing interpretation is encountered. 760 * 761 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on 762 * whether the name is to be looked up, created, renamed, or deleted. 763 * When CREATE, RENAME, or DELETE is specified, information usable in 764 * creating, renaming, or deleting a directory entry may be calculated. 765 * If flag has LOCKPARENT or'ed into it, the parent directory is returned 766 * locked. If flag has WANTPARENT or'ed into it, the parent directory is 767 * returned unlocked. Otherwise the parent directory is not returned. If 768 * the target of the pathname exists and LOCKLEAF is or'ed into the flag 769 * the target is returned locked, otherwise it is returned unlocked. 770 * When creating or renaming and LOCKPARENT is specified, the target may not 771 * be ".". When deleting and LOCKPARENT is specified, the target may be ".". 772 * 773 * Overall outline of lookup: 774 * 775 * dirloop: 776 * identify next component of name at ndp->ni_ptr 777 * handle degenerate case where name is null string 778 * if .. and crossing mount points and on mounted filesys, find parent 779 * call VOP_LOOKUP routine for next component name 780 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set 781 * component vnode returned in ni_vp (if it exists), locked. 782 * if result vnode is mounted on and crossing mount points, 783 * find mounted on vnode 784 * if more components of name, do next level at dirloop 785 * return the answer in ni_vp, locked if LOCKLEAF set 786 * if LOCKPARENT set, return locked parent in ni_dvp 787 * if WANTPARENT set, return unlocked parent in ni_dvp 788 */ 789 int 790 lookup(struct nameidata *ndp) 791 { 792 char *cp; /* pointer into pathname argument */ 793 char *prev_ni_next; /* saved ndp->ni_next */ 794 struct vnode *dp = NULL; /* the directory we are searching */ 795 struct vnode *tdp; /* saved dp */ 796 struct mount *mp; /* mount table entry */ 797 struct prison *pr; 798 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */ 799 int docache; /* == 0 do not cache last component */ 800 int wantparent; /* 1 => wantparent or lockparent flag */ 801 int rdonly; /* lookup read-only flag bit */ 802 int error = 0; 803 int dpunlocked = 0; /* dp has already been unlocked */ 804 int relookup = 0; /* do not consume the path component */ 805 struct componentname *cnp = &ndp->ni_cnd; 806 int lkflags_save; 807 int ni_dvp_unlocked; 808 809 /* 810 * Setup: break out flag bits into variables. 811 */ 812 ni_dvp_unlocked = 0; 813 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT); 814 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent, 815 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT.")); 816 /* 817 * When set to zero, docache causes the last component of the 818 * pathname to be deleted from the cache and the full lookup 819 * of the name to be done (via VOP_CACHEDLOOKUP()). Often 820 * filesystems need some pre-computed values that are made 821 * during the full lookup, for instance UFS sets dp->i_offset. 822 * 823 * The docache variable is set to zero when requested by the 824 * NOCACHE flag and for all modifying operations except CREATE. 825 */ 826 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE; 827 if (cnp->cn_nameiop == DELETE || 828 (wantparent && cnp->cn_nameiop != CREATE && 829 cnp->cn_nameiop != LOOKUP)) 830 docache = 0; 831 rdonly = cnp->cn_flags & RDONLY; 832 cnp->cn_flags &= ~ISSYMLINK; 833 ndp->ni_dvp = NULL; 834 /* 835 * We use shared locks until we hit the parent of the last cn then 836 * we adjust based on the requesting flags. 837 */ 838 cnp->cn_lkflags = LK_SHARED; 839 dp = ndp->ni_startdir; 840 ndp->ni_startdir = NULLVP; 841 vn_lock(dp, 842 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY, 843 cnp->cn_flags)); 844 845 dirloop: 846 /* 847 * Search a new directory. 848 * 849 * The last component of the filename is left accessible via 850 * cnp->cn_nameptr for callers that need the name. Callers needing 851 * the name set the SAVENAME flag. When done, they assume 852 * responsibility for freeing the pathname buffer. 853 */ 854 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++) 855 continue; 856 cnp->cn_namelen = cp - cnp->cn_nameptr; 857 if (cnp->cn_namelen > NAME_MAX) { 858 error = ENAMETOOLONG; 859 goto bad; 860 } 861 #ifdef NAMEI_DIAGNOSTIC 862 { char c = *cp; 863 *cp = '\0'; 864 printf("{%s}: ", cnp->cn_nameptr); 865 *cp = c; } 866 #endif 867 prev_ni_pathlen = ndp->ni_pathlen; 868 ndp->ni_pathlen -= cnp->cn_namelen; 869 KASSERT(ndp->ni_pathlen <= PATH_MAX, 870 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen)); 871 prev_ni_next = ndp->ni_next; 872 ndp->ni_next = cp; 873 874 /* 875 * Replace multiple slashes by a single slash and trailing slashes 876 * by a null. This must be done before VOP_LOOKUP() because some 877 * fs's don't know about trailing slashes. Remember if there were 878 * trailing slashes to handle symlinks, existing non-directories 879 * and non-existing files that won't be directories specially later. 880 */ 881 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) { 882 cp++; 883 ndp->ni_pathlen--; 884 if (*cp == '\0') { 885 *ndp->ni_next = '\0'; 886 cnp->cn_flags |= TRAILINGSLASH; 887 } 888 } 889 ndp->ni_next = cp; 890 891 cnp->cn_flags |= MAKEENTRY; 892 if (*cp == '\0' && docache == 0) 893 cnp->cn_flags &= ~MAKEENTRY; 894 if (cnp->cn_namelen == 2 && 895 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.') 896 cnp->cn_flags |= ISDOTDOT; 897 else 898 cnp->cn_flags &= ~ISDOTDOT; 899 if (*ndp->ni_next == 0) 900 cnp->cn_flags |= ISLASTCN; 901 else 902 cnp->cn_flags &= ~ISLASTCN; 903 904 if ((cnp->cn_flags & ISLASTCN) != 0 && 905 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' && 906 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 907 error = EINVAL; 908 goto bad; 909 } 910 911 nameicap_tracker_add(ndp, dp); 912 913 /* 914 * Check for degenerate name (e.g. / or "") 915 * which is a way of talking about a directory, 916 * e.g. like "/." or ".". 917 */ 918 if (cnp->cn_nameptr[0] == '\0') { 919 if (dp->v_type != VDIR) { 920 error = ENOTDIR; 921 goto bad; 922 } 923 if (cnp->cn_nameiop != LOOKUP) { 924 error = EISDIR; 925 goto bad; 926 } 927 if (wantparent) { 928 ndp->ni_dvp = dp; 929 VREF(dp); 930 } 931 ndp->ni_vp = dp; 932 933 if (cnp->cn_flags & AUDITVNODE1) 934 AUDIT_ARG_VNODE1(dp); 935 else if (cnp->cn_flags & AUDITVNODE2) 936 AUDIT_ARG_VNODE2(dp); 937 938 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF))) 939 VOP_UNLOCK(dp); 940 /* XXX This should probably move to the top of function. */ 941 if (cnp->cn_flags & SAVESTART) 942 panic("lookup: SAVESTART"); 943 goto success; 944 } 945 946 /* 947 * Handle "..": five special cases. 948 * 0. If doing a capability lookup and lookup_cap_dotdot is 949 * disabled, return ENOTCAPABLE. 950 * 1. Return an error if this is the last component of 951 * the name and the operation is DELETE or RENAME. 952 * 2. If at root directory (e.g. after chroot) 953 * or at absolute root directory 954 * then ignore it so can't get out. 955 * 3. If this vnode is the root of a mounted 956 * filesystem, then replace it with the 957 * vnode which was mounted on so we take the 958 * .. in the other filesystem. 959 * 4. If the vnode is the top directory of 960 * the jail or chroot, don't let them out. 961 * 5. If doing a capability lookup and lookup_cap_dotdot is 962 * enabled, return ENOTCAPABLE if the lookup would escape 963 * from the initial file descriptor directory. Checks are 964 * done by ensuring that namei() already traversed the 965 * result of dotdot lookup. 966 */ 967 if (cnp->cn_flags & ISDOTDOT) { 968 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT)) 969 == NI_LCF_STRICTRELATIVE) { 970 #ifdef KTRACE 971 if (KTRPOINT(curthread, KTR_CAPFAIL)) 972 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 973 #endif 974 error = ENOTCAPABLE; 975 goto bad; 976 } 977 if ((cnp->cn_flags & ISLASTCN) != 0 && 978 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 979 error = EINVAL; 980 goto bad; 981 } 982 for (;;) { 983 for (pr = cnp->cn_cred->cr_prison; pr != NULL; 984 pr = pr->pr_parent) 985 if (dp == pr->pr_root) 986 break; 987 if (dp == ndp->ni_rootdir || 988 dp == ndp->ni_topdir || 989 dp == rootvnode || 990 pr != NULL || 991 ((dp->v_vflag & VV_ROOT) != 0 && 992 (cnp->cn_flags & NOCROSSMOUNT) != 0)) { 993 ndp->ni_dvp = dp; 994 ndp->ni_vp = dp; 995 VREF(dp); 996 goto nextname; 997 } 998 if ((dp->v_vflag & VV_ROOT) == 0) 999 break; 1000 if (VN_IS_DOOMED(dp)) { /* forced unmount */ 1001 error = ENOENT; 1002 goto bad; 1003 } 1004 tdp = dp; 1005 dp = dp->v_mount->mnt_vnodecovered; 1006 VREF(dp); 1007 vput(tdp); 1008 vn_lock(dp, 1009 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | 1010 LK_RETRY, ISDOTDOT)); 1011 error = nameicap_check_dotdot(ndp, dp); 1012 if (error != 0) { 1013 #ifdef KTRACE 1014 if (KTRPOINT(curthread, KTR_CAPFAIL)) 1015 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 1016 #endif 1017 goto bad; 1018 } 1019 } 1020 } 1021 1022 /* 1023 * We now have a segment name to search for, and a directory to search. 1024 */ 1025 unionlookup: 1026 #ifdef MAC 1027 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp, cnp); 1028 if (error) 1029 goto bad; 1030 #endif 1031 ndp->ni_dvp = dp; 1032 ndp->ni_vp = NULL; 1033 ASSERT_VOP_LOCKED(dp, "lookup"); 1034 /* 1035 * If we have a shared lock we may need to upgrade the lock for the 1036 * last operation. 1037 */ 1038 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) && 1039 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED) 1040 vn_lock(dp, LK_UPGRADE|LK_RETRY); 1041 if (VN_IS_DOOMED(dp)) { 1042 error = ENOENT; 1043 goto bad; 1044 } 1045 /* 1046 * If we're looking up the last component and we need an exclusive 1047 * lock, adjust our lkflags. 1048 */ 1049 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags)) 1050 cnp->cn_lkflags = LK_EXCLUSIVE; 1051 #ifdef NAMEI_DIAGNOSTIC 1052 vn_printf(dp, "lookup in "); 1053 #endif 1054 lkflags_save = cnp->cn_lkflags; 1055 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags, 1056 cnp->cn_flags); 1057 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp); 1058 cnp->cn_lkflags = lkflags_save; 1059 if (error != 0) { 1060 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty")); 1061 #ifdef NAMEI_DIAGNOSTIC 1062 printf("not found\n"); 1063 #endif 1064 if ((error == ENOENT) && 1065 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) && 1066 (dp->v_mount->mnt_flag & MNT_UNION)) { 1067 tdp = dp; 1068 dp = dp->v_mount->mnt_vnodecovered; 1069 VREF(dp); 1070 vput(tdp); 1071 vn_lock(dp, 1072 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | 1073 LK_RETRY, cnp->cn_flags)); 1074 nameicap_tracker_add(ndp, dp); 1075 goto unionlookup; 1076 } 1077 1078 if (error == ERELOOKUP) { 1079 vref(dp); 1080 ndp->ni_vp = dp; 1081 error = 0; 1082 relookup = 1; 1083 goto good; 1084 } 1085 1086 if (error != EJUSTRETURN) 1087 goto bad; 1088 /* 1089 * At this point, we know we're at the end of the 1090 * pathname. If creating / renaming, we can consider 1091 * allowing the file or directory to be created / renamed, 1092 * provided we're not on a read-only filesystem. 1093 */ 1094 if (rdonly) { 1095 error = EROFS; 1096 goto bad; 1097 } 1098 /* trailing slash only allowed for directories */ 1099 if ((cnp->cn_flags & TRAILINGSLASH) && 1100 !(cnp->cn_flags & WILLBEDIR)) { 1101 error = ENOENT; 1102 goto bad; 1103 } 1104 if ((cnp->cn_flags & LOCKPARENT) == 0) 1105 VOP_UNLOCK(dp); 1106 /* 1107 * We return with ni_vp NULL to indicate that the entry 1108 * doesn't currently exist, leaving a pointer to the 1109 * (possibly locked) directory vnode in ndp->ni_dvp. 1110 */ 1111 if (cnp->cn_flags & SAVESTART) { 1112 ndp->ni_startdir = ndp->ni_dvp; 1113 VREF(ndp->ni_startdir); 1114 } 1115 goto success; 1116 } 1117 1118 good: 1119 #ifdef NAMEI_DIAGNOSTIC 1120 printf("found\n"); 1121 #endif 1122 dp = ndp->ni_vp; 1123 1124 /* 1125 * Check to see if the vnode has been mounted on; 1126 * if so find the root of the mounted filesystem. 1127 */ 1128 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) && 1129 (cnp->cn_flags & NOCROSSMOUNT) == 0) { 1130 if (vfs_busy(mp, 0)) 1131 continue; 1132 vput(dp); 1133 if (dp != ndp->ni_dvp) 1134 vput(ndp->ni_dvp); 1135 else 1136 vrele(ndp->ni_dvp); 1137 vrefact(vp_crossmp); 1138 ndp->ni_dvp = vp_crossmp; 1139 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags, 1140 cnp->cn_flags), &tdp); 1141 vfs_unbusy(mp); 1142 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT)) 1143 panic("vp_crossmp exclusively locked or reclaimed"); 1144 if (error) { 1145 dpunlocked = 1; 1146 goto bad2; 1147 } 1148 ndp->ni_vp = dp = tdp; 1149 } 1150 1151 /* 1152 * Check for symbolic link 1153 */ 1154 if ((dp->v_type == VLNK) && 1155 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) || 1156 *ndp->ni_next == '/')) { 1157 cnp->cn_flags |= ISSYMLINK; 1158 if (VN_IS_DOOMED(dp)) { 1159 /* 1160 * We can't know whether the directory was mounted with 1161 * NOSYMFOLLOW, so we can't follow safely. 1162 */ 1163 error = ENOENT; 1164 goto bad2; 1165 } 1166 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) { 1167 error = EACCES; 1168 goto bad2; 1169 } 1170 /* 1171 * Symlink code always expects an unlocked dvp. 1172 */ 1173 if (ndp->ni_dvp != ndp->ni_vp) { 1174 VOP_UNLOCK(ndp->ni_dvp); 1175 ni_dvp_unlocked = 1; 1176 } 1177 goto success; 1178 } 1179 1180 nextname: 1181 /* 1182 * Not a symbolic link that we will follow. Continue with the 1183 * next component if there is any; otherwise, we're done. 1184 */ 1185 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/', 1186 ("lookup: invalid path state.")); 1187 if (relookup) { 1188 relookup = 0; 1189 ndp->ni_pathlen = prev_ni_pathlen; 1190 ndp->ni_next = prev_ni_next; 1191 if (ndp->ni_dvp != dp) 1192 vput(ndp->ni_dvp); 1193 else 1194 vrele(ndp->ni_dvp); 1195 goto dirloop; 1196 } 1197 if (cnp->cn_flags & ISDOTDOT) { 1198 error = nameicap_check_dotdot(ndp, ndp->ni_vp); 1199 if (error != 0) { 1200 #ifdef KTRACE 1201 if (KTRPOINT(curthread, KTR_CAPFAIL)) 1202 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 1203 #endif 1204 goto bad2; 1205 } 1206 } 1207 if (*ndp->ni_next == '/') { 1208 cnp->cn_nameptr = ndp->ni_next; 1209 while (*cnp->cn_nameptr == '/') { 1210 cnp->cn_nameptr++; 1211 ndp->ni_pathlen--; 1212 } 1213 if (ndp->ni_dvp != dp) 1214 vput(ndp->ni_dvp); 1215 else 1216 vrele(ndp->ni_dvp); 1217 goto dirloop; 1218 } 1219 /* 1220 * If we're processing a path with a trailing slash, 1221 * check that the end result is a directory. 1222 */ 1223 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) { 1224 error = ENOTDIR; 1225 goto bad2; 1226 } 1227 /* 1228 * Disallow directory write attempts on read-only filesystems. 1229 */ 1230 if (rdonly && 1231 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 1232 error = EROFS; 1233 goto bad2; 1234 } 1235 if (cnp->cn_flags & SAVESTART) { 1236 ndp->ni_startdir = ndp->ni_dvp; 1237 VREF(ndp->ni_startdir); 1238 } 1239 if (!wantparent) { 1240 ni_dvp_unlocked = 2; 1241 if (ndp->ni_dvp != dp) 1242 vput(ndp->ni_dvp); 1243 else 1244 vrele(ndp->ni_dvp); 1245 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) { 1246 VOP_UNLOCK(ndp->ni_dvp); 1247 ni_dvp_unlocked = 1; 1248 } 1249 1250 if (cnp->cn_flags & AUDITVNODE1) 1251 AUDIT_ARG_VNODE1(dp); 1252 else if (cnp->cn_flags & AUDITVNODE2) 1253 AUDIT_ARG_VNODE2(dp); 1254 1255 if ((cnp->cn_flags & LOCKLEAF) == 0) 1256 VOP_UNLOCK(dp); 1257 success: 1258 /* 1259 * FIXME: for lookups which only cross a mount point to fetch the 1260 * root vnode, ni_dvp will be set to vp_crossmp. This can be a problem 1261 * if either WANTPARENT or LOCKPARENT is set. 1262 */ 1263 /* 1264 * Because of shared lookup we may have the vnode shared locked, but 1265 * the caller may want it to be exclusively locked. 1266 */ 1267 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) && 1268 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) { 1269 vn_lock(dp, LK_UPGRADE | LK_RETRY); 1270 if (VN_IS_DOOMED(dp)) { 1271 error = ENOENT; 1272 goto bad2; 1273 } 1274 } 1275 if (ndp->ni_vp != NULL && ndp->ni_vp->v_type == VDIR) 1276 nameicap_tracker_add(ndp, ndp->ni_vp); 1277 return (0); 1278 1279 bad2: 1280 if (ni_dvp_unlocked != 2) { 1281 if (dp != ndp->ni_dvp && !ni_dvp_unlocked) 1282 vput(ndp->ni_dvp); 1283 else 1284 vrele(ndp->ni_dvp); 1285 } 1286 bad: 1287 if (!dpunlocked) 1288 vput(dp); 1289 ndp->ni_vp = NULL; 1290 return (error); 1291 } 1292 1293 /* 1294 * relookup - lookup a path name component 1295 * Used by lookup to re-acquire things. 1296 */ 1297 int 1298 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) 1299 { 1300 struct vnode *dp = NULL; /* the directory we are searching */ 1301 int wantparent; /* 1 => wantparent or lockparent flag */ 1302 int rdonly; /* lookup read-only flag bit */ 1303 int error = 0; 1304 1305 KASSERT(cnp->cn_flags & ISLASTCN, 1306 ("relookup: Not given last component.")); 1307 /* 1308 * Setup: break out flag bits into variables. 1309 */ 1310 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT); 1311 KASSERT(wantparent, ("relookup: parent not wanted.")); 1312 rdonly = cnp->cn_flags & RDONLY; 1313 cnp->cn_flags &= ~ISSYMLINK; 1314 dp = dvp; 1315 cnp->cn_lkflags = LK_EXCLUSIVE; 1316 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); 1317 1318 /* 1319 * Search a new directory. 1320 * 1321 * The last component of the filename is left accessible via 1322 * cnp->cn_nameptr for callers that need the name. Callers needing 1323 * the name set the SAVENAME flag. When done, they assume 1324 * responsibility for freeing the pathname buffer. 1325 */ 1326 #ifdef NAMEI_DIAGNOSTIC 1327 printf("{%s}: ", cnp->cn_nameptr); 1328 #endif 1329 1330 /* 1331 * Check for "" which represents the root directory after slash 1332 * removal. 1333 */ 1334 if (cnp->cn_nameptr[0] == '\0') { 1335 /* 1336 * Support only LOOKUP for "/" because lookup() 1337 * can't succeed for CREATE, DELETE and RENAME. 1338 */ 1339 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP")); 1340 KASSERT(dp->v_type == VDIR, ("dp is not a directory")); 1341 1342 if (!(cnp->cn_flags & LOCKLEAF)) 1343 VOP_UNLOCK(dp); 1344 *vpp = dp; 1345 /* XXX This should probably move to the top of function. */ 1346 if (cnp->cn_flags & SAVESTART) 1347 panic("lookup: SAVESTART"); 1348 return (0); 1349 } 1350 1351 if (cnp->cn_flags & ISDOTDOT) 1352 panic ("relookup: lookup on dot-dot"); 1353 1354 /* 1355 * We now have a segment name to search for, and a directory to search. 1356 */ 1357 #ifdef NAMEI_DIAGNOSTIC 1358 vn_printf(dp, "search in "); 1359 #endif 1360 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) { 1361 KASSERT(*vpp == NULL, ("leaf should be empty")); 1362 if (error != EJUSTRETURN) 1363 goto bad; 1364 /* 1365 * If creating and at end of pathname, then can consider 1366 * allowing file to be created. 1367 */ 1368 if (rdonly) { 1369 error = EROFS; 1370 goto bad; 1371 } 1372 /* ASSERT(dvp == ndp->ni_startdir) */ 1373 if (cnp->cn_flags & SAVESTART) 1374 VREF(dvp); 1375 if ((cnp->cn_flags & LOCKPARENT) == 0) 1376 VOP_UNLOCK(dp); 1377 /* 1378 * We return with ni_vp NULL to indicate that the entry 1379 * doesn't currently exist, leaving a pointer to the 1380 * (possibly locked) directory vnode in ndp->ni_dvp. 1381 */ 1382 return (0); 1383 } 1384 1385 dp = *vpp; 1386 1387 /* 1388 * Disallow directory write attempts on read-only filesystems. 1389 */ 1390 if (rdonly && 1391 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 1392 if (dvp == dp) 1393 vrele(dvp); 1394 else 1395 vput(dvp); 1396 error = EROFS; 1397 goto bad; 1398 } 1399 /* 1400 * Set the parent lock/ref state to the requested state. 1401 */ 1402 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) { 1403 if (wantparent) 1404 VOP_UNLOCK(dvp); 1405 else 1406 vput(dvp); 1407 } else if (!wantparent) 1408 vrele(dvp); 1409 /* 1410 * Check for symbolic link 1411 */ 1412 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW), 1413 ("relookup: symlink found.\n")); 1414 1415 /* ASSERT(dvp == ndp->ni_startdir) */ 1416 if (cnp->cn_flags & SAVESTART) 1417 VREF(dvp); 1418 1419 if ((cnp->cn_flags & LOCKLEAF) == 0) 1420 VOP_UNLOCK(dp); 1421 return (0); 1422 bad: 1423 vput(dp); 1424 *vpp = NULL; 1425 return (error); 1426 } 1427 1428 /* 1429 * Free data allocated by namei(); see namei(9) for details. 1430 */ 1431 void 1432 NDFREE_PNBUF(struct nameidata *ndp) 1433 { 1434 1435 if ((ndp->ni_cnd.cn_flags & HASBUF) != 0) { 1436 MPASS((ndp->ni_cnd.cn_flags & (SAVENAME | SAVESTART)) != 0); 1437 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 1438 ndp->ni_cnd.cn_flags &= ~HASBUF; 1439 } 1440 } 1441 1442 void 1443 (NDFREE)(struct nameidata *ndp, const u_int flags) 1444 { 1445 int unlock_dvp; 1446 int unlock_vp; 1447 1448 unlock_dvp = 0; 1449 unlock_vp = 0; 1450 1451 if (!(flags & NDF_NO_FREE_PNBUF)) { 1452 NDFREE_PNBUF(ndp); 1453 } 1454 if (!(flags & NDF_NO_VP_UNLOCK) && 1455 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) 1456 unlock_vp = 1; 1457 if (!(flags & NDF_NO_DVP_UNLOCK) && 1458 (ndp->ni_cnd.cn_flags & LOCKPARENT) && 1459 ndp->ni_dvp != ndp->ni_vp) 1460 unlock_dvp = 1; 1461 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) { 1462 if (unlock_vp) { 1463 vput(ndp->ni_vp); 1464 unlock_vp = 0; 1465 } else 1466 vrele(ndp->ni_vp); 1467 ndp->ni_vp = NULL; 1468 } 1469 if (unlock_vp) 1470 VOP_UNLOCK(ndp->ni_vp); 1471 if (!(flags & NDF_NO_DVP_RELE) && 1472 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { 1473 if (unlock_dvp) { 1474 vput(ndp->ni_dvp); 1475 unlock_dvp = 0; 1476 } else 1477 vrele(ndp->ni_dvp); 1478 ndp->ni_dvp = NULL; 1479 } 1480 if (unlock_dvp) 1481 VOP_UNLOCK(ndp->ni_dvp); 1482 if (!(flags & NDF_NO_STARTDIR_RELE) && 1483 (ndp->ni_cnd.cn_flags & SAVESTART)) { 1484 vrele(ndp->ni_startdir); 1485 ndp->ni_startdir = NULL; 1486 } 1487 } 1488 1489 #ifdef INVARIANTS 1490 /* 1491 * Validate the final state of ndp after the lookup. 1492 * 1493 * Historically filesystems were allowed to modify cn_flags. Most notably they 1494 * can add SAVENAME to the request, resulting in HASBUF and pushing subsequent 1495 * clean up to the consumer. In practice this seems to only concern != LOOKUP 1496 * operations. 1497 * 1498 * As a step towards stricter API contract this routine validates the state to 1499 * clean up. Note validation is a work in progress with the intent of becoming 1500 * stricter over time. 1501 */ 1502 #define NDMODIFYINGFLAGS (LOCKLEAF | LOCKPARENT | WANTPARENT | SAVENAME | SAVESTART | HASBUF) 1503 void 1504 NDVALIDATE(struct nameidata *ndp) 1505 { 1506 struct componentname *cnp; 1507 u_int64_t used, orig; 1508 1509 cnp = &ndp->ni_cnd; 1510 orig = cnp->cn_origflags; 1511 used = cnp->cn_flags; 1512 switch (cnp->cn_nameiop) { 1513 case LOOKUP: 1514 /* 1515 * For plain lookup we require strict conformance -- nothing 1516 * to clean up if it was not requested by the caller. 1517 */ 1518 orig &= NDMODIFYINGFLAGS; 1519 used &= NDMODIFYINGFLAGS; 1520 if ((orig & (SAVENAME | SAVESTART)) != 0) 1521 orig |= HASBUF; 1522 if (orig != used) { 1523 goto out_mismatch; 1524 } 1525 break; 1526 case CREATE: 1527 case DELETE: 1528 case RENAME: 1529 /* 1530 * Some filesystems set SAVENAME to provoke HASBUF, accomodate 1531 * for it until it gets fixed. 1532 */ 1533 orig &= NDMODIFYINGFLAGS; 1534 orig |= (SAVENAME | HASBUF); 1535 used &= NDMODIFYINGFLAGS; 1536 used |= (SAVENAME | HASBUF); 1537 if (orig != used) { 1538 goto out_mismatch; 1539 } 1540 break; 1541 } 1542 return; 1543 out_mismatch: 1544 panic("%s: mismatched flags for op %d: added %" PRIx64 ", " 1545 "removed %" PRIx64" (%" PRIx64" != %" PRIx64"; stored %" PRIx64" != %" PRIx64")", 1546 __func__, cnp->cn_nameiop, used & ~orig, orig &~ used, 1547 orig, used, cnp->cn_origflags, cnp->cn_flags); 1548 } 1549 #endif 1550 1551 /* 1552 * Determine if there is a suitable alternate filename under the specified 1553 * prefix for the specified path. If the create flag is set, then the 1554 * alternate prefix will be used so long as the parent directory exists. 1555 * This is used by the various compatibility ABIs so that Linux binaries prefer 1556 * files under /compat/linux for example. The chosen path (whether under 1557 * the prefix or under /) is returned in a kernel malloc'd buffer pointed 1558 * to by pathbuf. The caller is responsible for free'ing the buffer from 1559 * the M_TEMP bucket if one is returned. 1560 */ 1561 int 1562 kern_alternate_path(struct thread *td, const char *prefix, const char *path, 1563 enum uio_seg pathseg, char **pathbuf, int create, int dirfd) 1564 { 1565 struct nameidata nd, ndroot; 1566 char *ptr, *buf, *cp; 1567 size_t len, sz; 1568 int error; 1569 1570 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1571 *pathbuf = buf; 1572 1573 /* Copy the prefix into the new pathname as a starting point. */ 1574 len = strlcpy(buf, prefix, MAXPATHLEN); 1575 if (len >= MAXPATHLEN) { 1576 *pathbuf = NULL; 1577 free(buf, M_TEMP); 1578 return (EINVAL); 1579 } 1580 sz = MAXPATHLEN - len; 1581 ptr = buf + len; 1582 1583 /* Append the filename to the prefix. */ 1584 if (pathseg == UIO_SYSSPACE) 1585 error = copystr(path, ptr, sz, &len); 1586 else 1587 error = copyinstr(path, ptr, sz, &len); 1588 1589 if (error) { 1590 *pathbuf = NULL; 1591 free(buf, M_TEMP); 1592 return (error); 1593 } 1594 1595 /* Only use a prefix with absolute pathnames. */ 1596 if (*ptr != '/') { 1597 error = EINVAL; 1598 goto keeporig; 1599 } 1600 1601 if (dirfd != AT_FDCWD) { 1602 /* 1603 * We want the original because the "prefix" is 1604 * included in the already opened dirfd. 1605 */ 1606 bcopy(ptr, buf, len); 1607 return (0); 1608 } 1609 1610 /* 1611 * We know that there is a / somewhere in this pathname. 1612 * Search backwards for it, to find the file's parent dir 1613 * to see if it exists in the alternate tree. If it does, 1614 * and we want to create a file (cflag is set). We don't 1615 * need to worry about the root comparison in this case. 1616 */ 1617 1618 if (create) { 1619 for (cp = &ptr[len] - 1; *cp != '/'; cp--); 1620 *cp = '\0'; 1621 1622 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td); 1623 error = namei(&nd); 1624 *cp = '/'; 1625 if (error != 0) 1626 goto keeporig; 1627 } else { 1628 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td); 1629 1630 error = namei(&nd); 1631 if (error != 0) 1632 goto keeporig; 1633 1634 /* 1635 * We now compare the vnode of the prefix to the one 1636 * vnode asked. If they resolve to be the same, then we 1637 * ignore the match so that the real root gets used. 1638 * This avoids the problem of traversing "../.." to find the 1639 * root directory and never finding it, because "/" resolves 1640 * to the emulation root directory. This is expensive :-( 1641 */ 1642 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix, 1643 td); 1644 1645 /* We shouldn't ever get an error from this namei(). */ 1646 error = namei(&ndroot); 1647 if (error == 0) { 1648 if (nd.ni_vp == ndroot.ni_vp) 1649 error = ENOENT; 1650 1651 NDFREE(&ndroot, NDF_ONLY_PNBUF); 1652 vrele(ndroot.ni_vp); 1653 } 1654 } 1655 1656 NDFREE(&nd, NDF_ONLY_PNBUF); 1657 vrele(nd.ni_vp); 1658 1659 keeporig: 1660 /* If there was an error, use the original path name. */ 1661 if (error) 1662 bcopy(ptr, buf, len); 1663 return (error); 1664 } 1665