vfs_lookup.c revision 280258
1187322Ssam/*- 2187322Ssam * Copyright (c) 1982, 1986, 1989, 1993 3187322Ssam * The Regents of the University of California. All rights reserved. 4187322Ssam * (c) UNIX System Laboratories, Inc. 5187322Ssam * All or some portions of this file are derived from material licensed 6187322Ssam * to the University of California by American Telephone and Telegraph 7187322Ssam * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8187322Ssam * the permission of UNIX System Laboratories, Inc. 9187322Ssam * 10187322Ssam * Redistribution and use in source and binary forms, with or without 11187322Ssam * modification, are permitted provided that the following conditions 12187322Ssam * are met: 13187322Ssam * 1. Redistributions of source code must retain the above copyright 14187322Ssam * notice, this list of conditions and the following disclaimer. 15187322Ssam * 2. Redistributions in binary form must reproduce the above copyright 16187322Ssam * notice, this list of conditions and the following disclaimer in the 17187322Ssam * documentation and/or other materials provided with the distribution. 18187322Ssam * 4. Neither the name of the University nor the names of its contributors 19187322Ssam * may be used to endorse or promote products derived from this software 20187322Ssam * without specific prior written permission. 21187322Ssam * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)vfs_lookup.c 8.4 (Berkeley) 2/16/94 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: stable/10/sys/kern/vfs_lookup.c 280258 2015-03-19 13:37:36Z rwatson $"); 39 40#include "opt_capsicum.h" 41#include "opt_kdtrace.h" 42#include "opt_ktrace.h" 43 44#include <sys/param.h> 45#include <sys/systm.h> 46#include <sys/kernel.h> 47#include <sys/capsicum.h> 48#include <sys/fcntl.h> 49#include <sys/jail.h> 50#include <sys/lock.h> 51#include <sys/mutex.h> 52#include <sys/namei.h> 53#include <sys/vnode.h> 54#include <sys/mount.h> 55#include <sys/filedesc.h> 56#include <sys/proc.h> 57#include <sys/sdt.h> 58#include <sys/syscallsubr.h> 59#include <sys/sysctl.h> 60#ifdef KTRACE 61#include <sys/ktrace.h> 62#endif 63 64#include <security/audit/audit.h> 65#include <security/mac/mac_framework.h> 66 67#include <vm/uma.h> 68 69#define NAMEI_DIAGNOSTIC 1 70#undef NAMEI_DIAGNOSTIC 71 72SDT_PROVIDER_DECLARE(vfs); 73SDT_PROBE_DEFINE3(vfs, namei, lookup, entry, "struct vnode *", "char *", 74 "unsigned long"); 75SDT_PROBE_DEFINE2(vfs, namei, lookup, return, "int", "struct vnode *"); 76 77/* 78 * Allocation zone for namei 79 */ 80uma_zone_t namei_zone; 81/* 82 * Placeholder vnode for mp traversal 83 */ 84static struct vnode *vp_crossmp; 85 86static void 87nameiinit(void *dummy __unused) 88{ 89 90 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL, 91 UMA_ALIGN_PTR, 0); 92 getnewvnode("crossmp", NULL, &dead_vnodeops, &vp_crossmp); 93 vn_lock(vp_crossmp, LK_EXCLUSIVE); 94 VN_LOCK_ASHARE(vp_crossmp); 95 VOP_UNLOCK(vp_crossmp, 0); 96} 97SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL); 98 99static int lookup_shared = 1; 100SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RW, &lookup_shared, 0, 101 "Enables/Disables shared locks for path name translation"); 102TUNABLE_INT("vfs.lookup_shared", &lookup_shared); 103 104/* 105 * Convert a pathname into a pointer to a locked vnode. 106 * 107 * The FOLLOW flag is set when symbolic links are to be followed 108 * when they occur at the end of the name translation process. 109 * Symbolic links are always followed for all other pathname 110 * components other than the last. 111 * 112 * The segflg defines whether the name is to be copied from user 113 * space or kernel space. 114 * 115 * Overall outline of namei: 116 * 117 * copy in name 118 * get starting directory 119 * while (!done && !error) { 120 * call lookup to search path. 121 * if symbolic link, massage name in buffer and continue 122 * } 123 */ 124static void 125namei_cleanup_cnp(struct componentname *cnp) 126{ 127 uma_zfree(namei_zone, cnp->cn_pnbuf); 128#ifdef DIAGNOSTIC 129 cnp->cn_pnbuf = NULL; 130 cnp->cn_nameptr = NULL; 131#endif 132} 133 134int 135namei(struct nameidata *ndp) 136{ 137 struct filedesc *fdp; /* pointer to file descriptor state */ 138 char *cp; /* pointer into pathname argument */ 139 struct vnode *dp; /* the directory we are searching */ 140 struct iovec aiov; /* uio for reading symbolic links */ 141 struct uio auio; 142 int error, linklen; 143 struct componentname *cnp = &ndp->ni_cnd; 144 struct thread *td = cnp->cn_thread; 145 struct proc *p = td->td_proc; 146 147 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred; 148 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc")); 149 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0, 150 ("namei: nameiop contaminated with flags")); 151 KASSERT((cnp->cn_flags & OPMASK) == 0, 152 ("namei: flags contaminated with nameiops")); 153 if (!lookup_shared) 154 cnp->cn_flags &= ~LOCKSHARED; 155 fdp = p->p_fd; 156 157 /* We will set this ourselves if we need it. */ 158 cnp->cn_flags &= ~TRAILINGSLASH; 159 160 /* 161 * Get a buffer for the name to be translated, and copy the 162 * name into the buffer. 163 */ 164 if ((cnp->cn_flags & HASBUF) == 0) 165 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK); 166 if (ndp->ni_segflg == UIO_SYSSPACE) 167 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, 168 MAXPATHLEN, (size_t *)&ndp->ni_pathlen); 169 else 170 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, 171 MAXPATHLEN, (size_t *)&ndp->ni_pathlen); 172 173 /* 174 * Don't allow empty pathnames. 175 */ 176 if (!error && *cnp->cn_pnbuf == '\0') 177 error = ENOENT; 178 179#ifdef CAPABILITY_MODE 180 /* 181 * In capability mode, lookups must be "strictly relative" (i.e. 182 * not an absolute path, and not containing '..' components) to 183 * a real file descriptor, not the pseudo-descriptor AT_FDCWD. 184 */ 185 if (error == 0 && IN_CAPABILITY_MODE(td) && 186 (cnp->cn_flags & NOCAPCHECK) == 0) { 187 ndp->ni_strictrelative = 1; 188 if (ndp->ni_dirfd == AT_FDCWD) { 189#ifdef KTRACE 190 if (KTRPOINT(td, KTR_CAPFAIL)) 191 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 192#endif 193 error = ECAPMODE; 194 } 195 } 196#endif 197 if (error) { 198 namei_cleanup_cnp(cnp); 199 ndp->ni_vp = NULL; 200 return (error); 201 } 202 ndp->ni_loopcnt = 0; 203#ifdef KTRACE 204 if (KTRPOINT(td, KTR_NAMEI)) { 205 KASSERT(cnp->cn_thread == curthread, 206 ("namei not using curthread")); 207 ktrnamei(cnp->cn_pnbuf); 208 } 209#endif 210 /* 211 * Get starting point for the translation. 212 */ 213 FILEDESC_SLOCK(fdp); 214 ndp->ni_rootdir = fdp->fd_rdir; 215 ndp->ni_topdir = fdp->fd_jdir; 216 217 /* 218 * If we are auditing the kernel pathname, save the user pathname. 219 */ 220 if (cnp->cn_flags & AUDITVNODE1) 221 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf); 222 if (cnp->cn_flags & AUDITVNODE2) 223 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf); 224 225 dp = NULL; 226 if (cnp->cn_pnbuf[0] != '/') { 227 if (ndp->ni_startdir != NULL) { 228 dp = ndp->ni_startdir; 229 error = 0; 230 } else if (ndp->ni_dirfd != AT_FDCWD) { 231 cap_rights_t rights; 232 233 rights = ndp->ni_rightsneeded; 234 cap_rights_set(&rights, CAP_LOOKUP); 235 236 if (cnp->cn_flags & AUDITVNODE1) 237 AUDIT_ARG_ATFD1(ndp->ni_dirfd); 238 if (cnp->cn_flags & AUDITVNODE2) 239 AUDIT_ARG_ATFD2(ndp->ni_dirfd); 240 error = fgetvp_rights(td, ndp->ni_dirfd, 241 &rights, &ndp->ni_filecaps, &dp); 242#ifdef CAPABILITIES 243 /* 244 * If file descriptor doesn't have all rights, 245 * all lookups relative to it must also be 246 * strictly relative. 247 */ 248 CAP_ALL(&rights); 249 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, 250 &rights) || 251 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || 252 ndp->ni_filecaps.fc_nioctls != -1) { 253 ndp->ni_strictrelative = 1; 254 } 255#endif 256 } 257 if (error != 0 || dp != NULL) { 258 FILEDESC_SUNLOCK(fdp); 259 if (error == 0 && dp->v_type != VDIR) { 260 vrele(dp); 261 error = ENOTDIR; 262 } 263 } 264 if (error) { 265 namei_cleanup_cnp(cnp); 266 return (error); 267 } 268 } 269 if (dp == NULL) { 270 dp = fdp->fd_cdir; 271 VREF(dp); 272 FILEDESC_SUNLOCK(fdp); 273 if (ndp->ni_startdir != NULL) 274 vrele(ndp->ni_startdir); 275 } 276 SDT_PROBE(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf, 277 cnp->cn_flags, 0, 0); 278 for (;;) { 279 /* 280 * Check if root directory should replace current directory. 281 * Done at start of translation and after symbolic link. 282 */ 283 cnp->cn_nameptr = cnp->cn_pnbuf; 284 if (*(cnp->cn_nameptr) == '/') { 285 vrele(dp); 286 if (ndp->ni_strictrelative != 0) { 287#ifdef KTRACE 288 if (KTRPOINT(curthread, KTR_CAPFAIL)) 289 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 290#endif 291 namei_cleanup_cnp(cnp); 292 return (ENOTCAPABLE); 293 } 294 while (*(cnp->cn_nameptr) == '/') { 295 cnp->cn_nameptr++; 296 ndp->ni_pathlen--; 297 } 298 dp = ndp->ni_rootdir; 299 VREF(dp); 300 } 301 ndp->ni_startdir = dp; 302 error = lookup(ndp); 303 if (error) { 304 namei_cleanup_cnp(cnp); 305 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0, 306 0, 0); 307 return (error); 308 } 309 /* 310 * If not a symbolic link, we're done. 311 */ 312 if ((cnp->cn_flags & ISSYMLINK) == 0) { 313 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) { 314 namei_cleanup_cnp(cnp); 315 } else 316 cnp->cn_flags |= HASBUF; 317 318 SDT_PROBE(vfs, namei, lookup, return, 0, ndp->ni_vp, 319 0, 0, 0); 320 return (0); 321 } 322 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) { 323 error = ELOOP; 324 break; 325 } 326#ifdef MAC 327 if ((cnp->cn_flags & NOMACCHECK) == 0) { 328 error = mac_vnode_check_readlink(td->td_ucred, 329 ndp->ni_vp); 330 if (error) 331 break; 332 } 333#endif 334 if (ndp->ni_pathlen > 1) 335 cp = uma_zalloc(namei_zone, M_WAITOK); 336 else 337 cp = cnp->cn_pnbuf; 338 aiov.iov_base = cp; 339 aiov.iov_len = MAXPATHLEN; 340 auio.uio_iov = &aiov; 341 auio.uio_iovcnt = 1; 342 auio.uio_offset = 0; 343 auio.uio_rw = UIO_READ; 344 auio.uio_segflg = UIO_SYSSPACE; 345 auio.uio_td = td; 346 auio.uio_resid = MAXPATHLEN; 347 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred); 348 if (error) { 349 if (ndp->ni_pathlen > 1) 350 uma_zfree(namei_zone, cp); 351 break; 352 } 353 linklen = MAXPATHLEN - auio.uio_resid; 354 if (linklen == 0) { 355 if (ndp->ni_pathlen > 1) 356 uma_zfree(namei_zone, cp); 357 error = ENOENT; 358 break; 359 } 360 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) { 361 if (ndp->ni_pathlen > 1) 362 uma_zfree(namei_zone, cp); 363 error = ENAMETOOLONG; 364 break; 365 } 366 if (ndp->ni_pathlen > 1) { 367 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen); 368 uma_zfree(namei_zone, cnp->cn_pnbuf); 369 cnp->cn_pnbuf = cp; 370 } else 371 cnp->cn_pnbuf[linklen] = '\0'; 372 ndp->ni_pathlen += linklen; 373 vput(ndp->ni_vp); 374 dp = ndp->ni_dvp; 375 } 376 namei_cleanup_cnp(cnp); 377 vput(ndp->ni_vp); 378 ndp->ni_vp = NULL; 379 vrele(ndp->ni_dvp); 380 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0, 0, 0); 381 return (error); 382} 383 384static int 385compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags) 386{ 387 388 if (mp == NULL || ((lkflags & LK_SHARED) && 389 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) || 390 ((cnflags & ISDOTDOT) && 391 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) { 392 lkflags &= ~LK_SHARED; 393 lkflags |= LK_EXCLUSIVE; 394 } 395 lkflags |= LK_NODDLKTREAT; 396 return (lkflags); 397} 398 399static __inline int 400needs_exclusive_leaf(struct mount *mp, int flags) 401{ 402 403 /* 404 * Intermediate nodes can use shared locks, we only need to 405 * force an exclusive lock for leaf nodes. 406 */ 407 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF)) 408 return (0); 409 410 /* Always use exclusive locks if LOCKSHARED isn't set. */ 411 if (!(flags & LOCKSHARED)) 412 return (1); 413 414 /* 415 * For lookups during open(), if the mount point supports 416 * extended shared operations, then use a shared lock for the 417 * leaf node, otherwise use an exclusive lock. 418 */ 419 if ((flags & ISOPEN) != 0) 420 return (!MNT_EXTENDED_SHARED(mp)); 421 422 /* 423 * Lookup requests outside of open() that specify LOCKSHARED 424 * only need a shared lock on the leaf vnode. 425 */ 426 return (0); 427} 428 429/* 430 * Search a pathname. 431 * This is a very central and rather complicated routine. 432 * 433 * The pathname is pointed to by ni_ptr and is of length ni_pathlen. 434 * The starting directory is taken from ni_startdir. The pathname is 435 * descended until done, or a symbolic link is encountered. The variable 436 * ni_more is clear if the path is completed; it is set to one if a 437 * symbolic link needing interpretation is encountered. 438 * 439 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on 440 * whether the name is to be looked up, created, renamed, or deleted. 441 * When CREATE, RENAME, or DELETE is specified, information usable in 442 * creating, renaming, or deleting a directory entry may be calculated. 443 * If flag has LOCKPARENT or'ed into it, the parent directory is returned 444 * locked. If flag has WANTPARENT or'ed into it, the parent directory is 445 * returned unlocked. Otherwise the parent directory is not returned. If 446 * the target of the pathname exists and LOCKLEAF is or'ed into the flag 447 * the target is returned locked, otherwise it is returned unlocked. 448 * When creating or renaming and LOCKPARENT is specified, the target may not 449 * be ".". When deleting and LOCKPARENT is specified, the target may be ".". 450 * 451 * Overall outline of lookup: 452 * 453 * dirloop: 454 * identify next component of name at ndp->ni_ptr 455 * handle degenerate case where name is null string 456 * if .. and crossing mount points and on mounted filesys, find parent 457 * call VOP_LOOKUP routine for next component name 458 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set 459 * component vnode returned in ni_vp (if it exists), locked. 460 * if result vnode is mounted on and crossing mount points, 461 * find mounted on vnode 462 * if more components of name, do next level at dirloop 463 * return the answer in ni_vp, locked if LOCKLEAF set 464 * if LOCKPARENT set, return locked parent in ni_dvp 465 * if WANTPARENT set, return unlocked parent in ni_dvp 466 */ 467int 468lookup(struct nameidata *ndp) 469{ 470 char *cp; /* pointer into pathname argument */ 471 struct vnode *dp = 0; /* the directory we are searching */ 472 struct vnode *tdp; /* saved dp */ 473 struct mount *mp; /* mount table entry */ 474 struct prison *pr; 475 int docache; /* == 0 do not cache last component */ 476 int wantparent; /* 1 => wantparent or lockparent flag */ 477 int rdonly; /* lookup read-only flag bit */ 478 int error = 0; 479 int dpunlocked = 0; /* dp has already been unlocked */ 480 struct componentname *cnp = &ndp->ni_cnd; 481 int lkflags_save; 482 int ni_dvp_unlocked; 483 484 /* 485 * Setup: break out flag bits into variables. 486 */ 487 ni_dvp_unlocked = 0; 488 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT); 489 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent, 490 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT.")); 491 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE; 492 if (cnp->cn_nameiop == DELETE || 493 (wantparent && cnp->cn_nameiop != CREATE && 494 cnp->cn_nameiop != LOOKUP)) 495 docache = 0; 496 rdonly = cnp->cn_flags & RDONLY; 497 cnp->cn_flags &= ~ISSYMLINK; 498 ndp->ni_dvp = NULL; 499 /* 500 * We use shared locks until we hit the parent of the last cn then 501 * we adjust based on the requesting flags. 502 */ 503 if (lookup_shared) 504 cnp->cn_lkflags = LK_SHARED; 505 else 506 cnp->cn_lkflags = LK_EXCLUSIVE; 507 dp = ndp->ni_startdir; 508 ndp->ni_startdir = NULLVP; 509 vn_lock(dp, 510 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY, 511 cnp->cn_flags)); 512 513dirloop: 514 /* 515 * Search a new directory. 516 * 517 * The last component of the filename is left accessible via 518 * cnp->cn_nameptr for callers that need the name. Callers needing 519 * the name set the SAVENAME flag. When done, they assume 520 * responsibility for freeing the pathname buffer. 521 */ 522 cnp->cn_consume = 0; 523 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++) 524 continue; 525 cnp->cn_namelen = cp - cnp->cn_nameptr; 526 if (cnp->cn_namelen > NAME_MAX) { 527 error = ENAMETOOLONG; 528 goto bad; 529 } 530#ifdef NAMEI_DIAGNOSTIC 531 { char c = *cp; 532 *cp = '\0'; 533 printf("{%s}: ", cnp->cn_nameptr); 534 *cp = c; } 535#endif 536 ndp->ni_pathlen -= cnp->cn_namelen; 537 ndp->ni_next = cp; 538 539 /* 540 * Replace multiple slashes by a single slash and trailing slashes 541 * by a null. This must be done before VOP_LOOKUP() because some 542 * fs's don't know about trailing slashes. Remember if there were 543 * trailing slashes to handle symlinks, existing non-directories 544 * and non-existing files that won't be directories specially later. 545 */ 546 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) { 547 cp++; 548 ndp->ni_pathlen--; 549 if (*cp == '\0') { 550 *ndp->ni_next = '\0'; 551 cnp->cn_flags |= TRAILINGSLASH; 552 } 553 } 554 ndp->ni_next = cp; 555 556 cnp->cn_flags |= MAKEENTRY; 557 if (*cp == '\0' && docache == 0) 558 cnp->cn_flags &= ~MAKEENTRY; 559 if (cnp->cn_namelen == 2 && 560 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.') 561 cnp->cn_flags |= ISDOTDOT; 562 else 563 cnp->cn_flags &= ~ISDOTDOT; 564 if (*ndp->ni_next == 0) 565 cnp->cn_flags |= ISLASTCN; 566 else 567 cnp->cn_flags &= ~ISLASTCN; 568 569 if ((cnp->cn_flags & ISLASTCN) != 0 && 570 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' && 571 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 572 error = EINVAL; 573 goto bad; 574 } 575 576 /* 577 * Check for degenerate name (e.g. / or "") 578 * which is a way of talking about a directory, 579 * e.g. like "/." or ".". 580 */ 581 if (cnp->cn_nameptr[0] == '\0') { 582 if (dp->v_type != VDIR) { 583 error = ENOTDIR; 584 goto bad; 585 } 586 if (cnp->cn_nameiop != LOOKUP) { 587 error = EISDIR; 588 goto bad; 589 } 590 if (wantparent) { 591 ndp->ni_dvp = dp; 592 VREF(dp); 593 } 594 ndp->ni_vp = dp; 595 596 if (cnp->cn_flags & AUDITVNODE1) 597 AUDIT_ARG_VNODE1(dp); 598 else if (cnp->cn_flags & AUDITVNODE2) 599 AUDIT_ARG_VNODE2(dp); 600 601 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF))) 602 VOP_UNLOCK(dp, 0); 603 /* XXX This should probably move to the top of function. */ 604 if (cnp->cn_flags & SAVESTART) 605 panic("lookup: SAVESTART"); 606 goto success; 607 } 608 609 /* 610 * Handle "..": five special cases. 611 * 0. If doing a capability lookup, return ENOTCAPABLE (this is a 612 * fairly conservative design choice, but it's the only one that we 613 * are satisfied guarantees the property we're looking for). 614 * 1. Return an error if this is the last component of 615 * the name and the operation is DELETE or RENAME. 616 * 2. If at root directory (e.g. after chroot) 617 * or at absolute root directory 618 * then ignore it so can't get out. 619 * 3. If this vnode is the root of a mounted 620 * filesystem, then replace it with the 621 * vnode which was mounted on so we take the 622 * .. in the other filesystem. 623 * 4. If the vnode is the top directory of 624 * the jail or chroot, don't let them out. 625 */ 626 if (cnp->cn_flags & ISDOTDOT) { 627 if (ndp->ni_strictrelative != 0) { 628#ifdef KTRACE 629 if (KTRPOINT(curthread, KTR_CAPFAIL)) 630 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 631#endif 632 error = ENOTCAPABLE; 633 goto bad; 634 } 635 if ((cnp->cn_flags & ISLASTCN) != 0 && 636 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 637 error = EINVAL; 638 goto bad; 639 } 640 for (;;) { 641 for (pr = cnp->cn_cred->cr_prison; pr != NULL; 642 pr = pr->pr_parent) 643 if (dp == pr->pr_root) 644 break; 645 if (dp == ndp->ni_rootdir || 646 dp == ndp->ni_topdir || 647 dp == rootvnode || 648 pr != NULL || 649 ((dp->v_vflag & VV_ROOT) != 0 && 650 (cnp->cn_flags & NOCROSSMOUNT) != 0)) { 651 ndp->ni_dvp = dp; 652 ndp->ni_vp = dp; 653 VREF(dp); 654 goto nextname; 655 } 656 if ((dp->v_vflag & VV_ROOT) == 0) 657 break; 658 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */ 659 error = ENOENT; 660 goto bad; 661 } 662 tdp = dp; 663 dp = dp->v_mount->mnt_vnodecovered; 664 VREF(dp); 665 vput(tdp); 666 vn_lock(dp, 667 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | 668 LK_RETRY, ISDOTDOT)); 669 } 670 } 671 672 /* 673 * We now have a segment name to search for, and a directory to search. 674 */ 675unionlookup: 676#ifdef MAC 677 if ((cnp->cn_flags & NOMACCHECK) == 0) { 678 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp, 679 cnp); 680 if (error) 681 goto bad; 682 } 683#endif 684 ndp->ni_dvp = dp; 685 ndp->ni_vp = NULL; 686 ASSERT_VOP_LOCKED(dp, "lookup"); 687 /* 688 * If we have a shared lock we may need to upgrade the lock for the 689 * last operation. 690 */ 691 if (dp != vp_crossmp && 692 VOP_ISLOCKED(dp) == LK_SHARED && 693 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT)) 694 vn_lock(dp, LK_UPGRADE|LK_RETRY); 695 if ((dp->v_iflag & VI_DOOMED) != 0) { 696 error = ENOENT; 697 goto bad; 698 } 699 /* 700 * If we're looking up the last component and we need an exclusive 701 * lock, adjust our lkflags. 702 */ 703 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags)) 704 cnp->cn_lkflags = LK_EXCLUSIVE; 705#ifdef NAMEI_DIAGNOSTIC 706 vprint("lookup in", dp); 707#endif 708 lkflags_save = cnp->cn_lkflags; 709 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags, 710 cnp->cn_flags); 711 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) { 712 cnp->cn_lkflags = lkflags_save; 713 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty")); 714#ifdef NAMEI_DIAGNOSTIC 715 printf("not found\n"); 716#endif 717 if ((error == ENOENT) && 718 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) && 719 (dp->v_mount->mnt_flag & MNT_UNION)) { 720 tdp = dp; 721 dp = dp->v_mount->mnt_vnodecovered; 722 VREF(dp); 723 vput(tdp); 724 vn_lock(dp, 725 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | 726 LK_RETRY, cnp->cn_flags)); 727 goto unionlookup; 728 } 729 730 if (error != EJUSTRETURN) 731 goto bad; 732 /* 733 * At this point, we know we're at the end of the 734 * pathname. If creating / renaming, we can consider 735 * allowing the file or directory to be created / renamed, 736 * provided we're not on a read-only filesystem. 737 */ 738 if (rdonly) { 739 error = EROFS; 740 goto bad; 741 } 742 /* trailing slash only allowed for directories */ 743 if ((cnp->cn_flags & TRAILINGSLASH) && 744 !(cnp->cn_flags & WILLBEDIR)) { 745 error = ENOENT; 746 goto bad; 747 } 748 if ((cnp->cn_flags & LOCKPARENT) == 0) 749 VOP_UNLOCK(dp, 0); 750 /* 751 * We return with ni_vp NULL to indicate that the entry 752 * doesn't currently exist, leaving a pointer to the 753 * (possibly locked) directory vnode in ndp->ni_dvp. 754 */ 755 if (cnp->cn_flags & SAVESTART) { 756 ndp->ni_startdir = ndp->ni_dvp; 757 VREF(ndp->ni_startdir); 758 } 759 goto success; 760 } else 761 cnp->cn_lkflags = lkflags_save; 762#ifdef NAMEI_DIAGNOSTIC 763 printf("found\n"); 764#endif 765 /* 766 * Take into account any additional components consumed by 767 * the underlying filesystem. 768 */ 769 if (cnp->cn_consume > 0) { 770 cnp->cn_nameptr += cnp->cn_consume; 771 ndp->ni_next += cnp->cn_consume; 772 ndp->ni_pathlen -= cnp->cn_consume; 773 cnp->cn_consume = 0; 774 } 775 776 dp = ndp->ni_vp; 777 778 /* 779 * Check to see if the vnode has been mounted on; 780 * if so find the root of the mounted filesystem. 781 */ 782 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) && 783 (cnp->cn_flags & NOCROSSMOUNT) == 0) { 784 if (vfs_busy(mp, 0)) 785 continue; 786 vput(dp); 787 if (dp != ndp->ni_dvp) 788 vput(ndp->ni_dvp); 789 else 790 vrele(ndp->ni_dvp); 791 vref(vp_crossmp); 792 ndp->ni_dvp = vp_crossmp; 793 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags, 794 cnp->cn_flags), &tdp); 795 vfs_unbusy(mp); 796 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT)) 797 panic("vp_crossmp exclusively locked or reclaimed"); 798 if (error) { 799 dpunlocked = 1; 800 goto bad2; 801 } 802 ndp->ni_vp = dp = tdp; 803 } 804 805 /* 806 * Check for symbolic link 807 */ 808 if ((dp->v_type == VLNK) && 809 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) || 810 *ndp->ni_next == '/')) { 811 cnp->cn_flags |= ISSYMLINK; 812 if (dp->v_iflag & VI_DOOMED) { 813 /* 814 * We can't know whether the directory was mounted with 815 * NOSYMFOLLOW, so we can't follow safely. 816 */ 817 error = ENOENT; 818 goto bad2; 819 } 820 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) { 821 error = EACCES; 822 goto bad2; 823 } 824 /* 825 * Symlink code always expects an unlocked dvp. 826 */ 827 if (ndp->ni_dvp != ndp->ni_vp) { 828 VOP_UNLOCK(ndp->ni_dvp, 0); 829 ni_dvp_unlocked = 1; 830 } 831 goto success; 832 } 833 834nextname: 835 /* 836 * Not a symbolic link that we will follow. Continue with the 837 * next component if there is any; otherwise, we're done. 838 */ 839 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/', 840 ("lookup: invalid path state.")); 841 if (*ndp->ni_next == '/') { 842 cnp->cn_nameptr = ndp->ni_next; 843 while (*cnp->cn_nameptr == '/') { 844 cnp->cn_nameptr++; 845 ndp->ni_pathlen--; 846 } 847 if (ndp->ni_dvp != dp) 848 vput(ndp->ni_dvp); 849 else 850 vrele(ndp->ni_dvp); 851 goto dirloop; 852 } 853 /* 854 * If we're processing a path with a trailing slash, 855 * check that the end result is a directory. 856 */ 857 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) { 858 error = ENOTDIR; 859 goto bad2; 860 } 861 /* 862 * Disallow directory write attempts on read-only filesystems. 863 */ 864 if (rdonly && 865 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 866 error = EROFS; 867 goto bad2; 868 } 869 if (cnp->cn_flags & SAVESTART) { 870 ndp->ni_startdir = ndp->ni_dvp; 871 VREF(ndp->ni_startdir); 872 } 873 if (!wantparent) { 874 ni_dvp_unlocked = 2; 875 if (ndp->ni_dvp != dp) 876 vput(ndp->ni_dvp); 877 else 878 vrele(ndp->ni_dvp); 879 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) { 880 VOP_UNLOCK(ndp->ni_dvp, 0); 881 ni_dvp_unlocked = 1; 882 } 883 884 if (cnp->cn_flags & AUDITVNODE1) 885 AUDIT_ARG_VNODE1(dp); 886 else if (cnp->cn_flags & AUDITVNODE2) 887 AUDIT_ARG_VNODE2(dp); 888 889 if ((cnp->cn_flags & LOCKLEAF) == 0) 890 VOP_UNLOCK(dp, 0); 891success: 892 /* 893 * Because of lookup_shared we may have the vnode shared locked, but 894 * the caller may want it to be exclusively locked. 895 */ 896 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) && 897 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) { 898 vn_lock(dp, LK_UPGRADE | LK_RETRY); 899 if (dp->v_iflag & VI_DOOMED) { 900 error = ENOENT; 901 goto bad2; 902 } 903 } 904 return (0); 905 906bad2: 907 if (ni_dvp_unlocked != 2) { 908 if (dp != ndp->ni_dvp && !ni_dvp_unlocked) 909 vput(ndp->ni_dvp); 910 else 911 vrele(ndp->ni_dvp); 912 } 913bad: 914 if (!dpunlocked) 915 vput(dp); 916 ndp->ni_vp = NULL; 917 return (error); 918} 919 920/* 921 * relookup - lookup a path name component 922 * Used by lookup to re-acquire things. 923 */ 924int 925relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) 926{ 927 struct vnode *dp = 0; /* the directory we are searching */ 928 int wantparent; /* 1 => wantparent or lockparent flag */ 929 int rdonly; /* lookup read-only flag bit */ 930 int error = 0; 931 932 KASSERT(cnp->cn_flags & ISLASTCN, 933 ("relookup: Not given last component.")); 934 /* 935 * Setup: break out flag bits into variables. 936 */ 937 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT); 938 KASSERT(wantparent, ("relookup: parent not wanted.")); 939 rdonly = cnp->cn_flags & RDONLY; 940 cnp->cn_flags &= ~ISSYMLINK; 941 dp = dvp; 942 cnp->cn_lkflags = LK_EXCLUSIVE; 943 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); 944 945 /* 946 * Search a new directory. 947 * 948 * The last component of the filename is left accessible via 949 * cnp->cn_nameptr for callers that need the name. Callers needing 950 * the name set the SAVENAME flag. When done, they assume 951 * responsibility for freeing the pathname buffer. 952 */ 953#ifdef NAMEI_DIAGNOSTIC 954 printf("{%s}: ", cnp->cn_nameptr); 955#endif 956 957 /* 958 * Check for "" which represents the root directory after slash 959 * removal. 960 */ 961 if (cnp->cn_nameptr[0] == '\0') { 962 /* 963 * Support only LOOKUP for "/" because lookup() 964 * can't succeed for CREATE, DELETE and RENAME. 965 */ 966 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP")); 967 KASSERT(dp->v_type == VDIR, ("dp is not a directory")); 968 969 if (!(cnp->cn_flags & LOCKLEAF)) 970 VOP_UNLOCK(dp, 0); 971 *vpp = dp; 972 /* XXX This should probably move to the top of function. */ 973 if (cnp->cn_flags & SAVESTART) 974 panic("lookup: SAVESTART"); 975 return (0); 976 } 977 978 if (cnp->cn_flags & ISDOTDOT) 979 panic ("relookup: lookup on dot-dot"); 980 981 /* 982 * We now have a segment name to search for, and a directory to search. 983 */ 984#ifdef NAMEI_DIAGNOSTIC 985 vprint("search in:", dp); 986#endif 987 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) { 988 KASSERT(*vpp == NULL, ("leaf should be empty")); 989 if (error != EJUSTRETURN) 990 goto bad; 991 /* 992 * If creating and at end of pathname, then can consider 993 * allowing file to be created. 994 */ 995 if (rdonly) { 996 error = EROFS; 997 goto bad; 998 } 999 /* ASSERT(dvp == ndp->ni_startdir) */ 1000 if (cnp->cn_flags & SAVESTART) 1001 VREF(dvp); 1002 if ((cnp->cn_flags & LOCKPARENT) == 0) 1003 VOP_UNLOCK(dp, 0); 1004 /* 1005 * We return with ni_vp NULL to indicate that the entry 1006 * doesn't currently exist, leaving a pointer to the 1007 * (possibly locked) directory vnode in ndp->ni_dvp. 1008 */ 1009 return (0); 1010 } 1011 1012 dp = *vpp; 1013 1014 /* 1015 * Disallow directory write attempts on read-only filesystems. 1016 */ 1017 if (rdonly && 1018 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 1019 if (dvp == dp) 1020 vrele(dvp); 1021 else 1022 vput(dvp); 1023 error = EROFS; 1024 goto bad; 1025 } 1026 /* 1027 * Set the parent lock/ref state to the requested state. 1028 */ 1029 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) { 1030 if (wantparent) 1031 VOP_UNLOCK(dvp, 0); 1032 else 1033 vput(dvp); 1034 } else if (!wantparent) 1035 vrele(dvp); 1036 /* 1037 * Check for symbolic link 1038 */ 1039 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW), 1040 ("relookup: symlink found.\n")); 1041 1042 /* ASSERT(dvp == ndp->ni_startdir) */ 1043 if (cnp->cn_flags & SAVESTART) 1044 VREF(dvp); 1045 1046 if ((cnp->cn_flags & LOCKLEAF) == 0) 1047 VOP_UNLOCK(dp, 0); 1048 return (0); 1049bad: 1050 vput(dp); 1051 *vpp = NULL; 1052 return (error); 1053} 1054 1055void 1056NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg, 1057 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp, 1058 struct thread *td) 1059{ 1060 1061 ndp->ni_cnd.cn_nameiop = op; 1062 ndp->ni_cnd.cn_flags = flags; 1063 ndp->ni_segflg = segflg; 1064 ndp->ni_dirp = namep; 1065 ndp->ni_dirfd = dirfd; 1066 ndp->ni_startdir = startdir; 1067 ndp->ni_strictrelative = 0; 1068 if (rightsp != NULL) 1069 ndp->ni_rightsneeded = *rightsp; 1070 else 1071 cap_rights_init(&ndp->ni_rightsneeded); 1072 filecaps_init(&ndp->ni_filecaps); 1073 ndp->ni_cnd.cn_thread = td; 1074} 1075 1076/* 1077 * Free data allocated by namei(); see namei(9) for details. 1078 */ 1079void 1080NDFREE(struct nameidata *ndp, const u_int flags) 1081{ 1082 int unlock_dvp; 1083 int unlock_vp; 1084 1085 unlock_dvp = 0; 1086 unlock_vp = 0; 1087 1088 if (!(flags & NDF_NO_FREE_PNBUF) && 1089 (ndp->ni_cnd.cn_flags & HASBUF)) { 1090 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 1091 ndp->ni_cnd.cn_flags &= ~HASBUF; 1092 } 1093 if (!(flags & NDF_NO_VP_UNLOCK) && 1094 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) 1095 unlock_vp = 1; 1096 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) { 1097 if (unlock_vp) { 1098 vput(ndp->ni_vp); 1099 unlock_vp = 0; 1100 } else 1101 vrele(ndp->ni_vp); 1102 ndp->ni_vp = NULL; 1103 } 1104 if (unlock_vp) 1105 VOP_UNLOCK(ndp->ni_vp, 0); 1106 if (!(flags & NDF_NO_DVP_UNLOCK) && 1107 (ndp->ni_cnd.cn_flags & LOCKPARENT) && 1108 ndp->ni_dvp != ndp->ni_vp) 1109 unlock_dvp = 1; 1110 if (!(flags & NDF_NO_DVP_RELE) && 1111 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { 1112 if (unlock_dvp) { 1113 vput(ndp->ni_dvp); 1114 unlock_dvp = 0; 1115 } else 1116 vrele(ndp->ni_dvp); 1117 ndp->ni_dvp = NULL; 1118 } 1119 if (unlock_dvp) 1120 VOP_UNLOCK(ndp->ni_dvp, 0); 1121 if (!(flags & NDF_NO_STARTDIR_RELE) && 1122 (ndp->ni_cnd.cn_flags & SAVESTART)) { 1123 vrele(ndp->ni_startdir); 1124 ndp->ni_startdir = NULL; 1125 } 1126} 1127 1128/* 1129 * Determine if there is a suitable alternate filename under the specified 1130 * prefix for the specified path. If the create flag is set, then the 1131 * alternate prefix will be used so long as the parent directory exists. 1132 * This is used by the various compatiblity ABIs so that Linux binaries prefer 1133 * files under /compat/linux for example. The chosen path (whether under 1134 * the prefix or under /) is returned in a kernel malloc'd buffer pointed 1135 * to by pathbuf. The caller is responsible for free'ing the buffer from 1136 * the M_TEMP bucket if one is returned. 1137 */ 1138int 1139kern_alternate_path(struct thread *td, const char *prefix, const char *path, 1140 enum uio_seg pathseg, char **pathbuf, int create, int dirfd) 1141{ 1142 struct nameidata nd, ndroot; 1143 char *ptr, *buf, *cp; 1144 size_t len, sz; 1145 int error; 1146 1147 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1148 *pathbuf = buf; 1149 1150 /* Copy the prefix into the new pathname as a starting point. */ 1151 len = strlcpy(buf, prefix, MAXPATHLEN); 1152 if (len >= MAXPATHLEN) { 1153 *pathbuf = NULL; 1154 free(buf, M_TEMP); 1155 return (EINVAL); 1156 } 1157 sz = MAXPATHLEN - len; 1158 ptr = buf + len; 1159 1160 /* Append the filename to the prefix. */ 1161 if (pathseg == UIO_SYSSPACE) 1162 error = copystr(path, ptr, sz, &len); 1163 else 1164 error = copyinstr(path, ptr, sz, &len); 1165 1166 if (error) { 1167 *pathbuf = NULL; 1168 free(buf, M_TEMP); 1169 return (error); 1170 } 1171 1172 /* Only use a prefix with absolute pathnames. */ 1173 if (*ptr != '/') { 1174 error = EINVAL; 1175 goto keeporig; 1176 } 1177 1178 if (dirfd != AT_FDCWD) { 1179 /* 1180 * We want the original because the "prefix" is 1181 * included in the already opened dirfd. 1182 */ 1183 bcopy(ptr, buf, len); 1184 return (0); 1185 } 1186 1187 /* 1188 * We know that there is a / somewhere in this pathname. 1189 * Search backwards for it, to find the file's parent dir 1190 * to see if it exists in the alternate tree. If it does, 1191 * and we want to create a file (cflag is set). We don't 1192 * need to worry about the root comparison in this case. 1193 */ 1194 1195 if (create) { 1196 for (cp = &ptr[len] - 1; *cp != '/'; cp--); 1197 *cp = '\0'; 1198 1199 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td); 1200 error = namei(&nd); 1201 *cp = '/'; 1202 if (error != 0) 1203 goto keeporig; 1204 } else { 1205 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td); 1206 1207 error = namei(&nd); 1208 if (error != 0) 1209 goto keeporig; 1210 1211 /* 1212 * We now compare the vnode of the prefix to the one 1213 * vnode asked. If they resolve to be the same, then we 1214 * ignore the match so that the real root gets used. 1215 * This avoids the problem of traversing "../.." to find the 1216 * root directory and never finding it, because "/" resolves 1217 * to the emulation root directory. This is expensive :-( 1218 */ 1219 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix, 1220 td); 1221 1222 /* We shouldn't ever get an error from this namei(). */ 1223 error = namei(&ndroot); 1224 if (error == 0) { 1225 if (nd.ni_vp == ndroot.ni_vp) 1226 error = ENOENT; 1227 1228 NDFREE(&ndroot, NDF_ONLY_PNBUF); 1229 vrele(ndroot.ni_vp); 1230 } 1231 } 1232 1233 NDFREE(&nd, NDF_ONLY_PNBUF); 1234 vrele(nd.ni_vp); 1235 1236keeporig: 1237 /* If there was an error, use the original path name. */ 1238 if (error) 1239 bcopy(ptr, buf, len); 1240 return (error); 1241} 1242