vfs_lookup.c revision 273411
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)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 273411 2014-10-21 20:20:07Z delphij $"); 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/capability.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 return (lkflags); 396} 397 398static __inline int 399needs_exclusive_leaf(struct mount *mp, int flags) 400{ 401 402 /* 403 * Intermediate nodes can use shared locks, we only need to 404 * force an exclusive lock for leaf nodes. 405 */ 406 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF)) 407 return (0); 408 409 /* Always use exclusive locks if LOCKSHARED isn't set. */ 410 if (!(flags & LOCKSHARED)) 411 return (1); 412 413 /* 414 * For lookups during open(), if the mount point supports 415 * extended shared operations, then use a shared lock for the 416 * leaf node, otherwise use an exclusive lock. 417 */ 418 if ((flags & ISOPEN) != 0) 419 return (!MNT_EXTENDED_SHARED(mp)); 420 421 /* 422 * Lookup requests outside of open() that specify LOCKSHARED 423 * only need a shared lock on the leaf vnode. 424 */ 425 return (0); 426} 427 428/* 429 * Search a pathname. 430 * This is a very central and rather complicated routine. 431 * 432 * The pathname is pointed to by ni_ptr and is of length ni_pathlen. 433 * The starting directory is taken from ni_startdir. The pathname is 434 * descended until done, or a symbolic link is encountered. The variable 435 * ni_more is clear if the path is completed; it is set to one if a 436 * symbolic link needing interpretation is encountered. 437 * 438 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on 439 * whether the name is to be looked up, created, renamed, or deleted. 440 * When CREATE, RENAME, or DELETE is specified, information usable in 441 * creating, renaming, or deleting a directory entry may be calculated. 442 * If flag has LOCKPARENT or'ed into it, the parent directory is returned 443 * locked. If flag has WANTPARENT or'ed into it, the parent directory is 444 * returned unlocked. Otherwise the parent directory is not returned. If 445 * the target of the pathname exists and LOCKLEAF is or'ed into the flag 446 * the target is returned locked, otherwise it is returned unlocked. 447 * When creating or renaming and LOCKPARENT is specified, the target may not 448 * be ".". When deleting and LOCKPARENT is specified, the target may be ".". 449 * 450 * Overall outline of lookup: 451 * 452 * dirloop: 453 * identify next component of name at ndp->ni_ptr 454 * handle degenerate case where name is null string 455 * if .. and crossing mount points and on mounted filesys, find parent 456 * call VOP_LOOKUP routine for next component name 457 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set 458 * component vnode returned in ni_vp (if it exists), locked. 459 * if result vnode is mounted on and crossing mount points, 460 * find mounted on vnode 461 * if more components of name, do next level at dirloop 462 * return the answer in ni_vp, locked if LOCKLEAF set 463 * if LOCKPARENT set, return locked parent in ni_dvp 464 * if WANTPARENT set, return unlocked parent in ni_dvp 465 */ 466int 467lookup(struct nameidata *ndp) 468{ 469 char *cp; /* pointer into pathname argument */ 470 struct vnode *dp = 0; /* the directory we are searching */ 471 struct vnode *tdp; /* saved dp */ 472 struct mount *mp; /* mount table entry */ 473 struct prison *pr; 474 int docache; /* == 0 do not cache last component */ 475 int wantparent; /* 1 => wantparent or lockparent flag */ 476 int rdonly; /* lookup read-only flag bit */ 477 int error = 0; 478 int dpunlocked = 0; /* dp has already been unlocked */ 479 struct componentname *cnp = &ndp->ni_cnd; 480 int lkflags_save; 481 int ni_dvp_unlocked; 482 483 /* 484 * Setup: break out flag bits into variables. 485 */ 486 ni_dvp_unlocked = 0; 487 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT); 488 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent, 489 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT.")); 490 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE; 491 if (cnp->cn_nameiop == DELETE || 492 (wantparent && cnp->cn_nameiop != CREATE && 493 cnp->cn_nameiop != LOOKUP)) 494 docache = 0; 495 rdonly = cnp->cn_flags & RDONLY; 496 cnp->cn_flags &= ~ISSYMLINK; 497 ndp->ni_dvp = NULL; 498 /* 499 * We use shared locks until we hit the parent of the last cn then 500 * we adjust based on the requesting flags. 501 */ 502 if (lookup_shared) 503 cnp->cn_lkflags = LK_SHARED; 504 else 505 cnp->cn_lkflags = LK_EXCLUSIVE; 506 dp = ndp->ni_startdir; 507 ndp->ni_startdir = NULLVP; 508 vn_lock(dp, 509 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY, 510 cnp->cn_flags)); 511 512dirloop: 513 /* 514 * Search a new directory. 515 * 516 * The last component of the filename is left accessible via 517 * cnp->cn_nameptr for callers that need the name. Callers needing 518 * the name set the SAVENAME flag. When done, they assume 519 * responsibility for freeing the pathname buffer. 520 */ 521 cnp->cn_consume = 0; 522 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++) 523 continue; 524 cnp->cn_namelen = cp - cnp->cn_nameptr; 525 if (cnp->cn_namelen > NAME_MAX) { 526 error = ENAMETOOLONG; 527 goto bad; 528 } 529#ifdef NAMEI_DIAGNOSTIC 530 { char c = *cp; 531 *cp = '\0'; 532 printf("{%s}: ", cnp->cn_nameptr); 533 *cp = c; } 534#endif 535 ndp->ni_pathlen -= cnp->cn_namelen; 536 ndp->ni_next = cp; 537 538 /* 539 * Replace multiple slashes by a single slash and trailing slashes 540 * by a null. This must be done before VOP_LOOKUP() because some 541 * fs's don't know about trailing slashes. Remember if there were 542 * trailing slashes to handle symlinks, existing non-directories 543 * and non-existing files that won't be directories specially later. 544 */ 545 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) { 546 cp++; 547 ndp->ni_pathlen--; 548 if (*cp == '\0') { 549 *ndp->ni_next = '\0'; 550 cnp->cn_flags |= TRAILINGSLASH; 551 } 552 } 553 ndp->ni_next = cp; 554 555 cnp->cn_flags |= MAKEENTRY; 556 if (*cp == '\0' && docache == 0) 557 cnp->cn_flags &= ~MAKEENTRY; 558 if (cnp->cn_namelen == 2 && 559 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.') 560 cnp->cn_flags |= ISDOTDOT; 561 else 562 cnp->cn_flags &= ~ISDOTDOT; 563 if (*ndp->ni_next == 0) 564 cnp->cn_flags |= ISLASTCN; 565 else 566 cnp->cn_flags &= ~ISLASTCN; 567 568 if ((cnp->cn_flags & ISLASTCN) != 0 && 569 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' && 570 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 571 error = EINVAL; 572 goto bad; 573 } 574 575 /* 576 * Check for degenerate name (e.g. / or "") 577 * which is a way of talking about a directory, 578 * e.g. like "/." or ".". 579 */ 580 if (cnp->cn_nameptr[0] == '\0') { 581 if (dp->v_type != VDIR) { 582 error = ENOTDIR; 583 goto bad; 584 } 585 if (cnp->cn_nameiop != LOOKUP) { 586 error = EISDIR; 587 goto bad; 588 } 589 if (wantparent) { 590 ndp->ni_dvp = dp; 591 VREF(dp); 592 } 593 ndp->ni_vp = dp; 594 595 if (cnp->cn_flags & AUDITVNODE1) 596 AUDIT_ARG_VNODE1(dp); 597 else if (cnp->cn_flags & AUDITVNODE2) 598 AUDIT_ARG_VNODE2(dp); 599 600 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF))) 601 VOP_UNLOCK(dp, 0); 602 /* XXX This should probably move to the top of function. */ 603 if (cnp->cn_flags & SAVESTART) 604 panic("lookup: SAVESTART"); 605 goto success; 606 } 607 608 /* 609 * Handle "..": five special cases. 610 * 0. If doing a capability lookup, return ENOTCAPABLE (this is a 611 * fairly conservative design choice, but it's the only one that we 612 * are satisfied guarantees the property we're looking for). 613 * 1. Return an error if this is the last component of 614 * the name and the operation is DELETE or RENAME. 615 * 2. If at root directory (e.g. after chroot) 616 * or at absolute root directory 617 * then ignore it so can't get out. 618 * 3. If this vnode is the root of a mounted 619 * filesystem, then replace it with the 620 * vnode which was mounted on so we take the 621 * .. in the other filesystem. 622 * 4. If the vnode is the top directory of 623 * the jail or chroot, don't let them out. 624 */ 625 if (cnp->cn_flags & ISDOTDOT) { 626 if (ndp->ni_strictrelative != 0) { 627#ifdef KTRACE 628 if (KTRPOINT(curthread, KTR_CAPFAIL)) 629 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL); 630#endif 631 error = ENOTCAPABLE; 632 goto bad; 633 } 634 if ((cnp->cn_flags & ISLASTCN) != 0 && 635 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 636 error = EINVAL; 637 goto bad; 638 } 639 for (;;) { 640 for (pr = cnp->cn_cred->cr_prison; pr != NULL; 641 pr = pr->pr_parent) 642 if (dp == pr->pr_root) 643 break; 644 if (dp == ndp->ni_rootdir || 645 dp == ndp->ni_topdir || 646 dp == rootvnode || 647 pr != NULL || 648 ((dp->v_vflag & VV_ROOT) != 0 && 649 (cnp->cn_flags & NOCROSSMOUNT) != 0)) { 650 ndp->ni_dvp = dp; 651 ndp->ni_vp = dp; 652 VREF(dp); 653 goto nextname; 654 } 655 if ((dp->v_vflag & VV_ROOT) == 0) 656 break; 657 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */ 658 error = ENOENT; 659 goto bad; 660 } 661 tdp = dp; 662 dp = dp->v_mount->mnt_vnodecovered; 663 VREF(dp); 664 vput(tdp); 665 vn_lock(dp, 666 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | 667 LK_RETRY, ISDOTDOT)); 668 } 669 } 670 671 /* 672 * We now have a segment name to search for, and a directory to search. 673 */ 674unionlookup: 675#ifdef MAC 676 if ((cnp->cn_flags & NOMACCHECK) == 0) { 677 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp, 678 cnp); 679 if (error) 680 goto bad; 681 } 682#endif 683 ndp->ni_dvp = dp; 684 ndp->ni_vp = NULL; 685 ASSERT_VOP_LOCKED(dp, "lookup"); 686 /* 687 * If we have a shared lock we may need to upgrade the lock for the 688 * last operation. 689 */ 690 if (dp != vp_crossmp && 691 VOP_ISLOCKED(dp) == LK_SHARED && 692 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT)) 693 vn_lock(dp, LK_UPGRADE|LK_RETRY); 694 if ((dp->v_iflag & VI_DOOMED) != 0) { 695 error = ENOENT; 696 goto bad; 697 } 698 /* 699 * If we're looking up the last component and we need an exclusive 700 * lock, adjust our lkflags. 701 */ 702 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags)) 703 cnp->cn_lkflags = LK_EXCLUSIVE; 704#ifdef NAMEI_DIAGNOSTIC 705 vprint("lookup in", dp); 706#endif 707 lkflags_save = cnp->cn_lkflags; 708 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags, 709 cnp->cn_flags); 710 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) { 711 cnp->cn_lkflags = lkflags_save; 712 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty")); 713#ifdef NAMEI_DIAGNOSTIC 714 printf("not found\n"); 715#endif 716 if ((error == ENOENT) && 717 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) && 718 (dp->v_mount->mnt_flag & MNT_UNION)) { 719 tdp = dp; 720 dp = dp->v_mount->mnt_vnodecovered; 721 VREF(dp); 722 vput(tdp); 723 vn_lock(dp, 724 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | 725 LK_RETRY, cnp->cn_flags)); 726 goto unionlookup; 727 } 728 729 if (error != EJUSTRETURN) 730 goto bad; 731 /* 732 * At this point, we know we're at the end of the 733 * pathname. If creating / renaming, we can consider 734 * allowing the file or directory to be created / renamed, 735 * provided we're not on a read-only filesystem. 736 */ 737 if (rdonly) { 738 error = EROFS; 739 goto bad; 740 } 741 /* trailing slash only allowed for directories */ 742 if ((cnp->cn_flags & TRAILINGSLASH) && 743 !(cnp->cn_flags & WILLBEDIR)) { 744 error = ENOENT; 745 goto bad; 746 } 747 if ((cnp->cn_flags & LOCKPARENT) == 0) 748 VOP_UNLOCK(dp, 0); 749 /* 750 * We return with ni_vp NULL to indicate that the entry 751 * doesn't currently exist, leaving a pointer to the 752 * (possibly locked) directory vnode in ndp->ni_dvp. 753 */ 754 if (cnp->cn_flags & SAVESTART) { 755 ndp->ni_startdir = ndp->ni_dvp; 756 VREF(ndp->ni_startdir); 757 } 758 goto success; 759 } else 760 cnp->cn_lkflags = lkflags_save; 761#ifdef NAMEI_DIAGNOSTIC 762 printf("found\n"); 763#endif 764 /* 765 * Take into account any additional components consumed by 766 * the underlying filesystem. 767 */ 768 if (cnp->cn_consume > 0) { 769 cnp->cn_nameptr += cnp->cn_consume; 770 ndp->ni_next += cnp->cn_consume; 771 ndp->ni_pathlen -= cnp->cn_consume; 772 cnp->cn_consume = 0; 773 } 774 775 dp = ndp->ni_vp; 776 777 /* 778 * Check to see if the vnode has been mounted on; 779 * if so find the root of the mounted filesystem. 780 */ 781 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) && 782 (cnp->cn_flags & NOCROSSMOUNT) == 0) { 783 if (vfs_busy(mp, 0)) 784 continue; 785 vput(dp); 786 if (dp != ndp->ni_dvp) 787 vput(ndp->ni_dvp); 788 else 789 vrele(ndp->ni_dvp); 790 vref(vp_crossmp); 791 ndp->ni_dvp = vp_crossmp; 792 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags, 793 cnp->cn_flags), &tdp); 794 vfs_unbusy(mp); 795 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT)) 796 panic("vp_crossmp exclusively locked or reclaimed"); 797 if (error) { 798 dpunlocked = 1; 799 goto bad2; 800 } 801 ndp->ni_vp = dp = tdp; 802 } 803 804 /* 805 * Check for symbolic link 806 */ 807 if ((dp->v_type == VLNK) && 808 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) || 809 *ndp->ni_next == '/')) { 810 cnp->cn_flags |= ISSYMLINK; 811 if (dp->v_iflag & VI_DOOMED) { 812 /* 813 * We can't know whether the directory was mounted with 814 * NOSYMFOLLOW, so we can't follow safely. 815 */ 816 error = ENOENT; 817 goto bad2; 818 } 819 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) { 820 error = EACCES; 821 goto bad2; 822 } 823 /* 824 * Symlink code always expects an unlocked dvp. 825 */ 826 if (ndp->ni_dvp != ndp->ni_vp) { 827 VOP_UNLOCK(ndp->ni_dvp, 0); 828 ni_dvp_unlocked = 1; 829 } 830 goto success; 831 } 832 833nextname: 834 /* 835 * Not a symbolic link that we will follow. Continue with the 836 * next component if there is any; otherwise, we're done. 837 */ 838 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/', 839 ("lookup: invalid path state.")); 840 if (*ndp->ni_next == '/') { 841 cnp->cn_nameptr = ndp->ni_next; 842 while (*cnp->cn_nameptr == '/') { 843 cnp->cn_nameptr++; 844 ndp->ni_pathlen--; 845 } 846 if (ndp->ni_dvp != dp) 847 vput(ndp->ni_dvp); 848 else 849 vrele(ndp->ni_dvp); 850 goto dirloop; 851 } 852 /* 853 * If we're processing a path with a trailing slash, 854 * check that the end result is a directory. 855 */ 856 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) { 857 error = ENOTDIR; 858 goto bad2; 859 } 860 /* 861 * Disallow directory write attempts on read-only filesystems. 862 */ 863 if (rdonly && 864 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 865 error = EROFS; 866 goto bad2; 867 } 868 if (cnp->cn_flags & SAVESTART) { 869 ndp->ni_startdir = ndp->ni_dvp; 870 VREF(ndp->ni_startdir); 871 } 872 if (!wantparent) { 873 ni_dvp_unlocked = 2; 874 if (ndp->ni_dvp != dp) 875 vput(ndp->ni_dvp); 876 else 877 vrele(ndp->ni_dvp); 878 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) { 879 VOP_UNLOCK(ndp->ni_dvp, 0); 880 ni_dvp_unlocked = 1; 881 } 882 883 if (cnp->cn_flags & AUDITVNODE1) 884 AUDIT_ARG_VNODE1(dp); 885 else if (cnp->cn_flags & AUDITVNODE2) 886 AUDIT_ARG_VNODE2(dp); 887 888 if ((cnp->cn_flags & LOCKLEAF) == 0) 889 VOP_UNLOCK(dp, 0); 890success: 891 /* 892 * Because of lookup_shared we may have the vnode shared locked, but 893 * the caller may want it to be exclusively locked. 894 */ 895 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) && 896 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) { 897 vn_lock(dp, LK_UPGRADE | LK_RETRY); 898 if (dp->v_iflag & VI_DOOMED) { 899 error = ENOENT; 900 goto bad2; 901 } 902 } 903 return (0); 904 905bad2: 906 if (ni_dvp_unlocked != 2) { 907 if (dp != ndp->ni_dvp && !ni_dvp_unlocked) 908 vput(ndp->ni_dvp); 909 else 910 vrele(ndp->ni_dvp); 911 } 912bad: 913 if (!dpunlocked) 914 vput(dp); 915 ndp->ni_vp = NULL; 916 return (error); 917} 918 919/* 920 * relookup - lookup a path name component 921 * Used by lookup to re-acquire things. 922 */ 923int 924relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) 925{ 926 struct vnode *dp = 0; /* the directory we are searching */ 927 int wantparent; /* 1 => wantparent or lockparent flag */ 928 int rdonly; /* lookup read-only flag bit */ 929 int error = 0; 930 931 KASSERT(cnp->cn_flags & ISLASTCN, 932 ("relookup: Not given last component.")); 933 /* 934 * Setup: break out flag bits into variables. 935 */ 936 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT); 937 KASSERT(wantparent, ("relookup: parent not wanted.")); 938 rdonly = cnp->cn_flags & RDONLY; 939 cnp->cn_flags &= ~ISSYMLINK; 940 dp = dvp; 941 cnp->cn_lkflags = LK_EXCLUSIVE; 942 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY); 943 944 /* 945 * Search a new directory. 946 * 947 * The last component of the filename is left accessible via 948 * cnp->cn_nameptr for callers that need the name. Callers needing 949 * the name set the SAVENAME flag. When done, they assume 950 * responsibility for freeing the pathname buffer. 951 */ 952#ifdef NAMEI_DIAGNOSTIC 953 printf("{%s}: ", cnp->cn_nameptr); 954#endif 955 956 /* 957 * Check for "" which represents the root directory after slash 958 * removal. 959 */ 960 if (cnp->cn_nameptr[0] == '\0') { 961 /* 962 * Support only LOOKUP for "/" because lookup() 963 * can't succeed for CREATE, DELETE and RENAME. 964 */ 965 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP")); 966 KASSERT(dp->v_type == VDIR, ("dp is not a directory")); 967 968 if (!(cnp->cn_flags & LOCKLEAF)) 969 VOP_UNLOCK(dp, 0); 970 *vpp = dp; 971 /* XXX This should probably move to the top of function. */ 972 if (cnp->cn_flags & SAVESTART) 973 panic("lookup: SAVESTART"); 974 return (0); 975 } 976 977 if (cnp->cn_flags & ISDOTDOT) 978 panic ("relookup: lookup on dot-dot"); 979 980 /* 981 * We now have a segment name to search for, and a directory to search. 982 */ 983#ifdef NAMEI_DIAGNOSTIC 984 vprint("search in:", dp); 985#endif 986 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) { 987 KASSERT(*vpp == NULL, ("leaf should be empty")); 988 if (error != EJUSTRETURN) 989 goto bad; 990 /* 991 * If creating and at end of pathname, then can consider 992 * allowing file to be created. 993 */ 994 if (rdonly) { 995 error = EROFS; 996 goto bad; 997 } 998 /* ASSERT(dvp == ndp->ni_startdir) */ 999 if (cnp->cn_flags & SAVESTART) 1000 VREF(dvp); 1001 if ((cnp->cn_flags & LOCKPARENT) == 0) 1002 VOP_UNLOCK(dp, 0); 1003 /* 1004 * We return with ni_vp NULL to indicate that the entry 1005 * doesn't currently exist, leaving a pointer to the 1006 * (possibly locked) directory vnode in ndp->ni_dvp. 1007 */ 1008 return (0); 1009 } 1010 1011 dp = *vpp; 1012 1013 /* 1014 * Disallow directory write attempts on read-only filesystems. 1015 */ 1016 if (rdonly && 1017 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) { 1018 if (dvp == dp) 1019 vrele(dvp); 1020 else 1021 vput(dvp); 1022 error = EROFS; 1023 goto bad; 1024 } 1025 /* 1026 * Set the parent lock/ref state to the requested state. 1027 */ 1028 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) { 1029 if (wantparent) 1030 VOP_UNLOCK(dvp, 0); 1031 else 1032 vput(dvp); 1033 } else if (!wantparent) 1034 vrele(dvp); 1035 /* 1036 * Check for symbolic link 1037 */ 1038 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW), 1039 ("relookup: symlink found.\n")); 1040 1041 /* ASSERT(dvp == ndp->ni_startdir) */ 1042 if (cnp->cn_flags & SAVESTART) 1043 VREF(dvp); 1044 1045 if ((cnp->cn_flags & LOCKLEAF) == 0) 1046 VOP_UNLOCK(dp, 0); 1047 return (0); 1048bad: 1049 vput(dp); 1050 *vpp = NULL; 1051 return (error); 1052} 1053 1054void 1055NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg, 1056 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp, 1057 struct thread *td) 1058{ 1059 1060 ndp->ni_cnd.cn_nameiop = op; 1061 ndp->ni_cnd.cn_flags = flags; 1062 ndp->ni_segflg = segflg; 1063 ndp->ni_dirp = namep; 1064 ndp->ni_dirfd = dirfd; 1065 ndp->ni_startdir = startdir; 1066 ndp->ni_strictrelative = 0; 1067 if (rightsp != NULL) 1068 ndp->ni_rightsneeded = *rightsp; 1069 else 1070 cap_rights_init(&ndp->ni_rightsneeded); 1071 filecaps_init(&ndp->ni_filecaps); 1072 ndp->ni_cnd.cn_thread = td; 1073} 1074 1075/* 1076 * Free data allocated by namei(); see namei(9) for details. 1077 */ 1078void 1079NDFREE(struct nameidata *ndp, const u_int flags) 1080{ 1081 int unlock_dvp; 1082 int unlock_vp; 1083 1084 unlock_dvp = 0; 1085 unlock_vp = 0; 1086 1087 if (!(flags & NDF_NO_FREE_PNBUF) && 1088 (ndp->ni_cnd.cn_flags & HASBUF)) { 1089 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 1090 ndp->ni_cnd.cn_flags &= ~HASBUF; 1091 } 1092 if (!(flags & NDF_NO_VP_UNLOCK) && 1093 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) 1094 unlock_vp = 1; 1095 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) { 1096 if (unlock_vp) { 1097 vput(ndp->ni_vp); 1098 unlock_vp = 0; 1099 } else 1100 vrele(ndp->ni_vp); 1101 ndp->ni_vp = NULL; 1102 } 1103 if (unlock_vp) 1104 VOP_UNLOCK(ndp->ni_vp, 0); 1105 if (!(flags & NDF_NO_DVP_UNLOCK) && 1106 (ndp->ni_cnd.cn_flags & LOCKPARENT) && 1107 ndp->ni_dvp != ndp->ni_vp) 1108 unlock_dvp = 1; 1109 if (!(flags & NDF_NO_DVP_RELE) && 1110 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { 1111 if (unlock_dvp) { 1112 vput(ndp->ni_dvp); 1113 unlock_dvp = 0; 1114 } else 1115 vrele(ndp->ni_dvp); 1116 ndp->ni_dvp = NULL; 1117 } 1118 if (unlock_dvp) 1119 VOP_UNLOCK(ndp->ni_dvp, 0); 1120 if (!(flags & NDF_NO_STARTDIR_RELE) && 1121 (ndp->ni_cnd.cn_flags & SAVESTART)) { 1122 vrele(ndp->ni_startdir); 1123 ndp->ni_startdir = NULL; 1124 } 1125} 1126 1127/* 1128 * Determine if there is a suitable alternate filename under the specified 1129 * prefix for the specified path. If the create flag is set, then the 1130 * alternate prefix will be used so long as the parent directory exists. 1131 * This is used by the various compatiblity ABIs so that Linux binaries prefer 1132 * files under /compat/linux for example. The chosen path (whether under 1133 * the prefix or under /) is returned in a kernel malloc'd buffer pointed 1134 * to by pathbuf. The caller is responsible for free'ing the buffer from 1135 * the M_TEMP bucket if one is returned. 1136 */ 1137int 1138kern_alternate_path(struct thread *td, const char *prefix, const char *path, 1139 enum uio_seg pathseg, char **pathbuf, int create, int dirfd) 1140{ 1141 struct nameidata nd, ndroot; 1142 char *ptr, *buf, *cp; 1143 size_t len, sz; 1144 int error; 1145 1146 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 1147 *pathbuf = buf; 1148 1149 /* Copy the prefix into the new pathname as a starting point. */ 1150 len = strlcpy(buf, prefix, MAXPATHLEN); 1151 if (len >= MAXPATHLEN) { 1152 *pathbuf = NULL; 1153 free(buf, M_TEMP); 1154 return (EINVAL); 1155 } 1156 sz = MAXPATHLEN - len; 1157 ptr = buf + len; 1158 1159 /* Append the filename to the prefix. */ 1160 if (pathseg == UIO_SYSSPACE) 1161 error = copystr(path, ptr, sz, &len); 1162 else 1163 error = copyinstr(path, ptr, sz, &len); 1164 1165 if (error) { 1166 *pathbuf = NULL; 1167 free(buf, M_TEMP); 1168 return (error); 1169 } 1170 1171 /* Only use a prefix with absolute pathnames. */ 1172 if (*ptr != '/') { 1173 error = EINVAL; 1174 goto keeporig; 1175 } 1176 1177 if (dirfd != AT_FDCWD) { 1178 /* 1179 * We want the original because the "prefix" is 1180 * included in the already opened dirfd. 1181 */ 1182 bcopy(ptr, buf, len); 1183 return (0); 1184 } 1185 1186 /* 1187 * We know that there is a / somewhere in this pathname. 1188 * Search backwards for it, to find the file's parent dir 1189 * to see if it exists in the alternate tree. If it does, 1190 * and we want to create a file (cflag is set). We don't 1191 * need to worry about the root comparison in this case. 1192 */ 1193 1194 if (create) { 1195 for (cp = &ptr[len] - 1; *cp != '/'; cp--); 1196 *cp = '\0'; 1197 1198 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td); 1199 error = namei(&nd); 1200 *cp = '/'; 1201 if (error != 0) 1202 goto keeporig; 1203 } else { 1204 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td); 1205 1206 error = namei(&nd); 1207 if (error != 0) 1208 goto keeporig; 1209 1210 /* 1211 * We now compare the vnode of the prefix to the one 1212 * vnode asked. If they resolve to be the same, then we 1213 * ignore the match so that the real root gets used. 1214 * This avoids the problem of traversing "../.." to find the 1215 * root directory and never finding it, because "/" resolves 1216 * to the emulation root directory. This is expensive :-( 1217 */ 1218 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix, 1219 td); 1220 1221 /* We shouldn't ever get an error from this namei(). */ 1222 error = namei(&ndroot); 1223 if (error == 0) { 1224 if (nd.ni_vp == ndroot.ni_vp) 1225 error = ENOENT; 1226 1227 NDFREE(&ndroot, NDF_ONLY_PNBUF); 1228 vrele(ndroot.ni_vp); 1229 } 1230 } 1231 1232 NDFREE(&nd, NDF_ONLY_PNBUF); 1233 vrele(nd.ni_vp); 1234 1235keeporig: 1236 /* If there was an error, use the original path name. */ 1237 if (error) 1238 bcopy(ptr, buf, len); 1239 return (error); 1240} 1241