1 /* 2 * linux/fs/namei.c 3 * 4 * Copyright (C) 1991, 1992 Linus Torvalds 5 */ 6 7 /* 8 * Some corrections by tytso. 9 */ 10 11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 12 * lookup logic. 13 */ 14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 15 */ 16 17 #include <linux/init.h> 18 #include <linux/export.h> 19 #include <linux/kernel.h> 20 #include <linux/slab.h> 21 #include <linux/fs.h> 22 #include <linux/namei.h> 23 #include <linux/pagemap.h> 24 #include <linux/fsnotify.h> 25 #include <linux/personality.h> 26 #include <linux/security.h> 27 #include <linux/ima.h> 28 #include <linux/syscalls.h> 29 #include <linux/mount.h> 30 #include <linux/audit.h> 31 #include <linux/capability.h> 32 #include <linux/file.h> 33 #include <linux/fcntl.h> 34 #include <linux/device_cgroup.h> 35 #include <linux/fs_struct.h> 36 #include <linux/posix_acl.h> 37 #include <asm/uaccess.h> 38 39 #include "internal.h" 40 #include "mount.h" 41 42 /* [Feb-1997 T. Schoebel-Theuer] 43 * Fundamental changes in the pathname lookup mechanisms (namei) 44 * were necessary because of omirr. The reason is that omirr needs 45 * to know the _real_ pathname, not the user-supplied one, in case 46 * of symlinks (and also when transname replacements occur). 47 * 48 * The new code replaces the old recursive symlink resolution with 49 * an iterative one (in case of non-nested symlink chains). It does 50 * this with calls to <fs>_follow_link(). 51 * As a side effect, dir_namei(), _namei() and follow_link() are now 52 * replaced with a single function lookup_dentry() that can handle all 53 * the special cases of the former code. 54 * 55 * With the new dcache, the pathname is stored at each inode, at least as 56 * long as the refcount of the inode is positive. As a side effect, the 57 * size of the dcache depends on the inode cache and thus is dynamic. 58 * 59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 60 * resolution to correspond with current state of the code. 61 * 62 * Note that the symlink resolution is not *completely* iterative. 63 * There is still a significant amount of tail- and mid- recursion in 64 * the algorithm. Also, note that <fs>_readlink() is not used in 65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 66 * may return different results than <fs>_follow_link(). Many virtual 67 * filesystems (including /proc) exhibit this behavior. 68 */ 69 70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 72 * and the name already exists in form of a symlink, try to create the new 73 * name indicated by the symlink. The old code always complained that the 74 * name already exists, due to not following the symlink even if its target 75 * is nonexistent. The new semantics affects also mknod() and link() when 76 * the name is a symlink pointing to a non-existent name. 77 * 78 * I don't know which semantics is the right one, since I have no access 79 * to standards. But I found by trial that HP-UX 9.0 has the full "new" 80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 81 * "old" one. Personally, I think the new semantics is much more logical. 82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing 83 * file does succeed in both HP-UX and SunOs, but not in Solaris 84 * and in the old Linux semantics. 85 */ 86 87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 88 * semantics. See the comments in "open_namei" and "do_link" below. 89 * 90 * [10-Sep-98 Alan Modra] Another symlink change. 91 */ 92 93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 94 * inside the path - always follow. 95 * in the last component in creation/removal/renaming - never follow. 96 * if LOOKUP_FOLLOW passed - follow. 97 * if the pathname has trailing slashes - follow. 98 * otherwise - don't follow. 99 * (applied in that order). 100 * 101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 103 * During the 2.4 we need to fix the userland stuff depending on it - 104 * hopefully we will be able to get rid of that wart in 2.5. So far only 105 * XEmacs seems to be relying on it... 106 */ 107 /* 108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 110 * any extra contention... 111 */ 112 113 /* In order to reduce some races, while at the same time doing additional 114 * checking and hopefully speeding things up, we copy filenames to the 115 * kernel data space before using them.. 116 * 117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 118 * PATH_MAX includes the nul terminator --RR. 119 */ 120 void final_putname(struct filename *name) 121 { 122 if (name->separate) { 123 __putname(name->name); 124 kfree(name); 125 } else { 126 __putname(name); 127 } 128 } 129 130 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename)) 131 132 static struct filename * 133 getname_flags(const char __user *filename, int flags, int *empty) 134 { 135 struct filename *result, *err; 136 int len; 137 long max; 138 char *kname; 139 140 result = audit_reusename(filename); 141 if (result) 142 return result; 143 144 result = __getname(); 145 if (unlikely(!result)) 146 return ERR_PTR(-ENOMEM); 147 148 /* 149 * First, try to embed the struct filename inside the names_cache 150 * allocation 151 */ 152 kname = (char *)result + sizeof(*result); 153 result->name = kname; 154 result->separate = false; 155 max = EMBEDDED_NAME_MAX; 156 157 recopy: 158 len = strncpy_from_user(kname, filename, max); 159 if (unlikely(len < 0)) { 160 err = ERR_PTR(len); 161 goto error; 162 } 163 164 /* 165 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 166 * separate struct filename so we can dedicate the entire 167 * names_cache allocation for the pathname, and re-do the copy from 168 * userland. 169 */ 170 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) { 171 kname = (char *)result; 172 173 result = kzalloc(sizeof(*result), GFP_KERNEL); 174 if (!result) { 175 err = ERR_PTR(-ENOMEM); 176 result = (struct filename *)kname; 177 goto error; 178 } 179 result->name = kname; 180 result->separate = true; 181 max = PATH_MAX; 182 goto recopy; 183 } 184 185 /* The empty path is special. */ 186 if (unlikely(!len)) { 187 if (empty) 188 *empty = 1; 189 err = ERR_PTR(-ENOENT); 190 if (!(flags & LOOKUP_EMPTY)) 191 goto error; 192 } 193 194 err = ERR_PTR(-ENAMETOOLONG); 195 if (unlikely(len >= PATH_MAX)) 196 goto error; 197 198 result->uptr = filename; 199 audit_getname(result); 200 return result; 201 202 error: 203 final_putname(result); 204 return err; 205 } 206 207 struct filename * 208 getname(const char __user * filename) 209 { 210 return getname_flags(filename, 0, NULL); 211 } 212 EXPORT_SYMBOL(getname); 213 214 #ifdef CONFIG_AUDITSYSCALL 215 void putname(struct filename *name) 216 { 217 if (unlikely(!audit_dummy_context())) 218 return audit_putname(name); 219 final_putname(name); 220 } 221 #endif 222 223 static int check_acl(struct inode *inode, int mask) 224 { 225 #ifdef CONFIG_FS_POSIX_ACL 226 struct posix_acl *acl; 227 228 if (mask & MAY_NOT_BLOCK) { 229 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 230 if (!acl) 231 return -EAGAIN; 232 /* no ->get_acl() calls in RCU mode... */ 233 if (acl == ACL_NOT_CACHED) 234 return -ECHILD; 235 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 236 } 237 238 acl = get_cached_acl(inode, ACL_TYPE_ACCESS); 239 240 /* 241 * A filesystem can force a ACL callback by just never filling the 242 * ACL cache. But normally you'd fill the cache either at inode 243 * instantiation time, or on the first ->get_acl call. 244 * 245 * If the filesystem doesn't have a get_acl() function at all, we'll 246 * just create the negative cache entry. 247 */ 248 if (acl == ACL_NOT_CACHED) { 249 if (inode->i_op->get_acl) { 250 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); 251 if (IS_ERR(acl)) 252 return PTR_ERR(acl); 253 } else { 254 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); 255 return -EAGAIN; 256 } 257 } 258 259 if (acl) { 260 int error = posix_acl_permission(inode, acl, mask); 261 posix_acl_release(acl); 262 return error; 263 } 264 #endif 265 266 return -EAGAIN; 267 } 268 269 /* 270 * This does the basic permission checking 271 */ 272 static int acl_permission_check(struct inode *inode, int mask) 273 { 274 unsigned int mode = inode->i_mode; 275 276 if (likely(uid_eq(current_fsuid(), inode->i_uid))) 277 mode >>= 6; 278 else { 279 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 280 int error = check_acl(inode, mask); 281 if (error != -EAGAIN) 282 return error; 283 } 284 285 if (in_group_p(inode->i_gid)) 286 mode >>= 3; 287 } 288 289 /* 290 * If the DACs are ok we don't need any capability check. 291 */ 292 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 293 return 0; 294 return -EACCES; 295 } 296 297 /** 298 * generic_permission - check for access rights on a Posix-like filesystem 299 * @inode: inode to check access rights for 300 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 301 * 302 * Used to check for read/write/execute permissions on a file. 303 * We use "fsuid" for this, letting us set arbitrary permissions 304 * for filesystem access without changing the "normal" uids which 305 * are used for other things. 306 * 307 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 308 * request cannot be satisfied (eg. requires blocking or too much complexity). 309 * It would then be called again in ref-walk mode. 310 */ 311 int generic_permission(struct inode *inode, int mask) 312 { 313 int ret; 314 315 /* 316 * Do the basic permission checks. 317 */ 318 ret = acl_permission_check(inode, mask); 319 if (ret != -EACCES) 320 return ret; 321 322 if (S_ISDIR(inode->i_mode)) { 323 /* DACs are overridable for directories */ 324 if (inode_capable(inode, CAP_DAC_OVERRIDE)) 325 return 0; 326 if (!(mask & MAY_WRITE)) 327 if (inode_capable(inode, CAP_DAC_READ_SEARCH)) 328 return 0; 329 return -EACCES; 330 } 331 /* 332 * Read/write DACs are always overridable. 333 * Executable DACs are overridable when there is 334 * at least one exec bit set. 335 */ 336 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 337 if (inode_capable(inode, CAP_DAC_OVERRIDE)) 338 return 0; 339 340 /* 341 * Searching includes executable on directories, else just read. 342 */ 343 mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 344 if (mask == MAY_READ) 345 if (inode_capable(inode, CAP_DAC_READ_SEARCH)) 346 return 0; 347 348 return -EACCES; 349 } 350 351 /* 352 * We _really_ want to just do "generic_permission()" without 353 * even looking at the inode->i_op values. So we keep a cache 354 * flag in inode->i_opflags, that says "this has not special 355 * permission function, use the fast case". 356 */ 357 static inline int do_inode_permission(struct inode *inode, int mask) 358 { 359 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 360 if (likely(inode->i_op->permission)) 361 return inode->i_op->permission(inode, mask); 362 363 /* This gets set once for the inode lifetime */ 364 spin_lock(&inode->i_lock); 365 inode->i_opflags |= IOP_FASTPERM; 366 spin_unlock(&inode->i_lock); 367 } 368 return generic_permission(inode, mask); 369 } 370 371 /** 372 * __inode_permission - Check for access rights to a given inode 373 * @inode: Inode to check permission on 374 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 375 * 376 * Check for read/write/execute permissions on an inode. 377 * 378 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 379 * 380 * This does not check for a read-only file system. You probably want 381 * inode_permission(). 382 */ 383 int __inode_permission(struct inode *inode, int mask) 384 { 385 int retval; 386 387 if (unlikely(mask & MAY_WRITE)) { 388 /* 389 * Nobody gets write access to an immutable file. 390 */ 391 if (IS_IMMUTABLE(inode)) 392 return -EACCES; 393 } 394 395 retval = do_inode_permission(inode, mask); 396 if (retval) 397 return retval; 398 399 retval = devcgroup_inode_permission(inode, mask); 400 if (retval) 401 return retval; 402 403 return security_inode_permission(inode, mask); 404 } 405 406 /** 407 * sb_permission - Check superblock-level permissions 408 * @sb: Superblock of inode to check permission on 409 * @inode: Inode to check permission on 410 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 411 * 412 * Separate out file-system wide checks from inode-specific permission checks. 413 */ 414 static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 415 { 416 if (unlikely(mask & MAY_WRITE)) { 417 umode_t mode = inode->i_mode; 418 419 /* Nobody gets write access to a read-only fs. */ 420 if ((sb->s_flags & MS_RDONLY) && 421 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 422 return -EROFS; 423 } 424 return 0; 425 } 426 427 /** 428 * inode_permission - Check for access rights to a given inode 429 * @inode: Inode to check permission on 430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 431 * 432 * Check for read/write/execute permissions on an inode. We use fs[ug]id for 433 * this, letting us set arbitrary permissions for filesystem access without 434 * changing the "normal" UIDs which are used for other things. 435 * 436 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 437 */ 438 int inode_permission(struct inode *inode, int mask) 439 { 440 int retval; 441 442 retval = sb_permission(inode->i_sb, inode, mask); 443 if (retval) 444 return retval; 445 return __inode_permission(inode, mask); 446 } 447 448 /** 449 * path_get - get a reference to a path 450 * @path: path to get the reference to 451 * 452 * Given a path increment the reference count to the dentry and the vfsmount. 453 */ 454 void path_get(struct path *path) 455 { 456 mntget(path->mnt); 457 dget(path->dentry); 458 } 459 EXPORT_SYMBOL(path_get); 460 461 /** 462 * path_put - put a reference to a path 463 * @path: path to put the reference to 464 * 465 * Given a path decrement the reference count to the dentry and the vfsmount. 466 */ 467 void path_put(struct path *path) 468 { 469 dput(path->dentry); 470 mntput(path->mnt); 471 } 472 EXPORT_SYMBOL(path_put); 473 474 /* 475 * Path walking has 2 modes, rcu-walk and ref-walk (see 476 * Documentation/filesystems/path-lookup.txt). In situations when we can't 477 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 478 * normal reference counts on dentries and vfsmounts to transition to rcu-walk 479 * mode. Refcounts are grabbed at the last known good point before rcu-walk 480 * got stuck, so ref-walk may continue from there. If this is not successful 481 * (eg. a seqcount has changed), then failure is returned and it's up to caller 482 * to restart the path walk from the beginning in ref-walk mode. 483 */ 484 485 static inline void lock_rcu_walk(void) 486 { 487 br_read_lock(&vfsmount_lock); 488 rcu_read_lock(); 489 } 490 491 static inline void unlock_rcu_walk(void) 492 { 493 rcu_read_unlock(); 494 br_read_unlock(&vfsmount_lock); 495 } 496 497 /** 498 * unlazy_walk - try to switch to ref-walk mode. 499 * @nd: nameidata pathwalk data 500 * @dentry: child of nd->path.dentry or NULL 501 * Returns: 0 on success, -ECHILD on failure 502 * 503 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 504 * for ref-walk mode. @dentry must be a path found by a do_lookup call on 505 * @nd or NULL. Must be called from rcu-walk context. 506 */ 507 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 508 { 509 struct fs_struct *fs = current->fs; 510 struct dentry *parent = nd->path.dentry; 511 int want_root = 0; 512 513 BUG_ON(!(nd->flags & LOOKUP_RCU)); 514 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 515 want_root = 1; 516 spin_lock(&fs->lock); 517 if (nd->root.mnt != fs->root.mnt || 518 nd->root.dentry != fs->root.dentry) 519 goto err_root; 520 } 521 spin_lock(&parent->d_lock); 522 if (!dentry) { 523 if (!__d_rcu_to_refcount(parent, nd->seq)) 524 goto err_parent; 525 BUG_ON(nd->inode != parent->d_inode); 526 } else { 527 if (dentry->d_parent != parent) 528 goto err_parent; 529 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 530 if (!__d_rcu_to_refcount(dentry, nd->seq)) 531 goto err_child; 532 /* 533 * If the sequence check on the child dentry passed, then 534 * the child has not been removed from its parent. This 535 * means the parent dentry must be valid and able to take 536 * a reference at this point. 537 */ 538 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); 539 BUG_ON(!parent->d_count); 540 parent->d_count++; 541 spin_unlock(&dentry->d_lock); 542 } 543 spin_unlock(&parent->d_lock); 544 if (want_root) { 545 path_get(&nd->root); 546 spin_unlock(&fs->lock); 547 } 548 mntget(nd->path.mnt); 549 550 unlock_rcu_walk(); 551 nd->flags &= ~LOOKUP_RCU; 552 return 0; 553 554 err_child: 555 spin_unlock(&dentry->d_lock); 556 err_parent: 557 spin_unlock(&parent->d_lock); 558 err_root: 559 if (want_root) 560 spin_unlock(&fs->lock); 561 return -ECHILD; 562 } 563 564 static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 565 { 566 return dentry->d_op->d_revalidate(dentry, flags); 567 } 568 569 /** 570 * complete_walk - successful completion of path walk 571 * @nd: pointer nameidata 572 * 573 * If we had been in RCU mode, drop out of it and legitimize nd->path. 574 * Revalidate the final result, unless we'd already done that during 575 * the path walk or the filesystem doesn't ask for it. Return 0 on 576 * success, -error on failure. In case of failure caller does not 577 * need to drop nd->path. 578 */ 579 static int complete_walk(struct nameidata *nd) 580 { 581 struct dentry *dentry = nd->path.dentry; 582 int status; 583 584 if (nd->flags & LOOKUP_RCU) { 585 nd->flags &= ~LOOKUP_RCU; 586 if (!(nd->flags & LOOKUP_ROOT)) 587 nd->root.mnt = NULL; 588 spin_lock(&dentry->d_lock); 589 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) { 590 spin_unlock(&dentry->d_lock); 591 unlock_rcu_walk(); 592 return -ECHILD; 593 } 594 BUG_ON(nd->inode != dentry->d_inode); 595 spin_unlock(&dentry->d_lock); 596 mntget(nd->path.mnt); 597 unlock_rcu_walk(); 598 } 599 600 if (likely(!(nd->flags & LOOKUP_JUMPED))) 601 return 0; 602 603 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) 604 return 0; 605 606 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) 607 return 0; 608 609 /* Note: we do not d_invalidate() */ 610 status = d_revalidate(dentry, nd->flags); 611 if (status > 0) 612 return 0; 613 614 if (!status) 615 status = -ESTALE; 616 617 path_put(&nd->path); 618 return status; 619 } 620 621 static __always_inline void set_root(struct nameidata *nd) 622 { 623 if (!nd->root.mnt) 624 get_fs_root(current->fs, &nd->root); 625 } 626 627 static int link_path_walk(const char *, struct nameidata *); 628 629 static __always_inline void set_root_rcu(struct nameidata *nd) 630 { 631 if (!nd->root.mnt) { 632 struct fs_struct *fs = current->fs; 633 unsigned seq; 634 635 do { 636 seq = read_seqcount_begin(&fs->seq); 637 nd->root = fs->root; 638 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 639 } while (read_seqcount_retry(&fs->seq, seq)); 640 } 641 } 642 643 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 644 { 645 int ret; 646 647 if (IS_ERR(link)) 648 goto fail; 649 650 if (*link == '/') { 651 set_root(nd); 652 path_put(&nd->path); 653 nd->path = nd->root; 654 path_get(&nd->root); 655 nd->flags |= LOOKUP_JUMPED; 656 } 657 nd->inode = nd->path.dentry->d_inode; 658 659 ret = link_path_walk(link, nd); 660 return ret; 661 fail: 662 path_put(&nd->path); 663 return PTR_ERR(link); 664 } 665 666 static void path_put_conditional(struct path *path, struct nameidata *nd) 667 { 668 dput(path->dentry); 669 if (path->mnt != nd->path.mnt) 670 mntput(path->mnt); 671 } 672 673 static inline void path_to_nameidata(const struct path *path, 674 struct nameidata *nd) 675 { 676 if (!(nd->flags & LOOKUP_RCU)) { 677 dput(nd->path.dentry); 678 if (nd->path.mnt != path->mnt) 679 mntput(nd->path.mnt); 680 } 681 nd->path.mnt = path->mnt; 682 nd->path.dentry = path->dentry; 683 } 684 685 /* 686 * Helper to directly jump to a known parsed path from ->follow_link, 687 * caller must have taken a reference to path beforehand. 688 */ 689 void nd_jump_link(struct nameidata *nd, struct path *path) 690 { 691 path_put(&nd->path); 692 693 nd->path = *path; 694 nd->inode = nd->path.dentry->d_inode; 695 nd->flags |= LOOKUP_JUMPED; 696 697 BUG_ON(nd->inode->i_op->follow_link); 698 } 699 700 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 701 { 702 struct inode *inode = link->dentry->d_inode; 703 if (inode->i_op->put_link) 704 inode->i_op->put_link(link->dentry, nd, cookie); 705 path_put(link); 706 } 707 708 int sysctl_protected_symlinks __read_mostly = 0; 709 int sysctl_protected_hardlinks __read_mostly = 0; 710 711 /** 712 * may_follow_link - Check symlink following for unsafe situations 713 * @link: The path of the symlink 714 * @nd: nameidata pathwalk data 715 * 716 * In the case of the sysctl_protected_symlinks sysctl being enabled, 717 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 718 * in a sticky world-writable directory. This is to protect privileged 719 * processes from failing races against path names that may change out 720 * from under them by way of other users creating malicious symlinks. 721 * It will permit symlinks to be followed only when outside a sticky 722 * world-writable directory, or when the uid of the symlink and follower 723 * match, or when the directory owner matches the symlink's owner. 724 * 725 * Returns 0 if following the symlink is allowed, -ve on error. 726 */ 727 static inline int may_follow_link(struct path *link, struct nameidata *nd) 728 { 729 const struct inode *inode; 730 const struct inode *parent; 731 732 if (!sysctl_protected_symlinks) 733 return 0; 734 735 /* Allowed if owner and follower match. */ 736 inode = link->dentry->d_inode; 737 if (uid_eq(current_cred()->fsuid, inode->i_uid)) 738 return 0; 739 740 /* Allowed if parent directory not sticky and world-writable. */ 741 parent = nd->path.dentry->d_inode; 742 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 743 return 0; 744 745 /* Allowed if parent directory and link owner match. */ 746 if (uid_eq(parent->i_uid, inode->i_uid)) 747 return 0; 748 749 audit_log_link_denied("follow_link", link); 750 path_put_conditional(link, nd); 751 path_put(&nd->path); 752 return -EACCES; 753 } 754 755 /** 756 * safe_hardlink_source - Check for safe hardlink conditions 757 * @inode: the source inode to hardlink from 758 * 759 * Return false if at least one of the following conditions: 760 * - inode is not a regular file 761 * - inode is setuid 762 * - inode is setgid and group-exec 763 * - access failure for read and write 764 * 765 * Otherwise returns true. 766 */ 767 static bool safe_hardlink_source(struct inode *inode) 768 { 769 umode_t mode = inode->i_mode; 770 771 /* Special files should not get pinned to the filesystem. */ 772 if (!S_ISREG(mode)) 773 return false; 774 775 /* Setuid files should not get pinned to the filesystem. */ 776 if (mode & S_ISUID) 777 return false; 778 779 /* Executable setgid files should not get pinned to the filesystem. */ 780 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 781 return false; 782 783 /* Hardlinking to unreadable or unwritable sources is dangerous. */ 784 if (inode_permission(inode, MAY_READ | MAY_WRITE)) 785 return false; 786 787 return true; 788 } 789 790 /** 791 * may_linkat - Check permissions for creating a hardlink 792 * @link: the source to hardlink from 793 * 794 * Block hardlink when all of: 795 * - sysctl_protected_hardlinks enabled 796 * - fsuid does not match inode 797 * - hardlink source is unsafe (see safe_hardlink_source() above) 798 * - not CAP_FOWNER 799 * 800 * Returns 0 if successful, -ve on error. 801 */ 802 static int may_linkat(struct path *link) 803 { 804 const struct cred *cred; 805 struct inode *inode; 806 807 if (!sysctl_protected_hardlinks) 808 return 0; 809 810 cred = current_cred(); 811 inode = link->dentry->d_inode; 812 813 /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 814 * otherwise, it must be a safe source. 815 */ 816 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 817 capable(CAP_FOWNER)) 818 return 0; 819 820 audit_log_link_denied("linkat", link); 821 return -EPERM; 822 } 823 824 static __always_inline int 825 follow_link(struct path *link, struct nameidata *nd, void **p) 826 { 827 struct dentry *dentry = link->dentry; 828 int error; 829 char *s; 830 831 BUG_ON(nd->flags & LOOKUP_RCU); 832 833 if (link->mnt == nd->path.mnt) 834 mntget(link->mnt); 835 836 error = -ELOOP; 837 if (unlikely(current->total_link_count >= 40)) 838 goto out_put_nd_path; 839 840 cond_resched(); 841 current->total_link_count++; 842 843 touch_atime(link); 844 nd_set_link(nd, NULL); 845 846 error = security_inode_follow_link(link->dentry, nd); 847 if (error) 848 goto out_put_nd_path; 849 850 nd->last_type = LAST_BIND; 851 *p = dentry->d_inode->i_op->follow_link(dentry, nd); 852 error = PTR_ERR(*p); 853 if (IS_ERR(*p)) 854 goto out_put_nd_path; 855 856 error = 0; 857 s = nd_get_link(nd); 858 if (s) { 859 error = __vfs_follow_link(nd, s); 860 if (unlikely(error)) 861 put_link(nd, link, *p); 862 } 863 864 return error; 865 866 out_put_nd_path: 867 *p = NULL; 868 path_put(&nd->path); 869 path_put(link); 870 return error; 871 } 872 873 static int follow_up_rcu(struct path *path) 874 { 875 struct mount *mnt = real_mount(path->mnt); 876 struct mount *parent; 877 struct dentry *mountpoint; 878 879 parent = mnt->mnt_parent; 880 if (&parent->mnt == path->mnt) 881 return 0; 882 mountpoint = mnt->mnt_mountpoint; 883 path->dentry = mountpoint; 884 path->mnt = &parent->mnt; 885 return 1; 886 } 887 888 /* 889 * follow_up - Find the mountpoint of path's vfsmount 890 * 891 * Given a path, find the mountpoint of its source file system. 892 * Replace @path with the path of the mountpoint in the parent mount. 893 * Up is towards /. 894 * 895 * Return 1 if we went up a level and 0 if we were already at the 896 * root. 897 */ 898 int follow_up(struct path *path) 899 { 900 struct mount *mnt = real_mount(path->mnt); 901 struct mount *parent; 902 struct dentry *mountpoint; 903 904 br_read_lock(&vfsmount_lock); 905 parent = mnt->mnt_parent; 906 if (parent == mnt) { 907 br_read_unlock(&vfsmount_lock); 908 return 0; 909 } 910 mntget(&parent->mnt); 911 mountpoint = dget(mnt->mnt_mountpoint); 912 br_read_unlock(&vfsmount_lock); 913 dput(path->dentry); 914 path->dentry = mountpoint; 915 mntput(path->mnt); 916 path->mnt = &parent->mnt; 917 return 1; 918 } 919 920 /* 921 * Perform an automount 922 * - return -EISDIR to tell follow_managed() to stop and return the path we 923 * were called with. 924 */ 925 static int follow_automount(struct path *path, unsigned flags, 926 bool *need_mntput) 927 { 928 struct vfsmount *mnt; 929 int err; 930 931 if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 932 return -EREMOTE; 933 934 /* We don't want to mount if someone's just doing a stat - 935 * unless they're stat'ing a directory and appended a '/' to 936 * the name. 937 * 938 * We do, however, want to mount if someone wants to open or 939 * create a file of any type under the mountpoint, wants to 940 * traverse through the mountpoint or wants to open the 941 * mounted directory. Also, autofs may mark negative dentries 942 * as being automount points. These will need the attentions 943 * of the daemon to instantiate them before they can be used. 944 */ 945 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 946 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 947 path->dentry->d_inode) 948 return -EISDIR; 949 950 current->total_link_count++; 951 if (current->total_link_count >= 40) 952 return -ELOOP; 953 954 mnt = path->dentry->d_op->d_automount(path); 955 if (IS_ERR(mnt)) { 956 /* 957 * The filesystem is allowed to return -EISDIR here to indicate 958 * it doesn't want to automount. For instance, autofs would do 959 * this so that its userspace daemon can mount on this dentry. 960 * 961 * However, we can only permit this if it's a terminal point in 962 * the path being looked up; if it wasn't then the remainder of 963 * the path is inaccessible and we should say so. 964 */ 965 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 966 return -EREMOTE; 967 return PTR_ERR(mnt); 968 } 969 970 if (!mnt) /* mount collision */ 971 return 0; 972 973 if (!*need_mntput) { 974 /* lock_mount() may release path->mnt on error */ 975 mntget(path->mnt); 976 *need_mntput = true; 977 } 978 err = finish_automount(mnt, path); 979 980 switch (err) { 981 case -EBUSY: 982 /* Someone else made a mount here whilst we were busy */ 983 return 0; 984 case 0: 985 path_put(path); 986 path->mnt = mnt; 987 path->dentry = dget(mnt->mnt_root); 988 return 0; 989 default: 990 return err; 991 } 992 993 } 994 995 /* 996 * Handle a dentry that is managed in some way. 997 * - Flagged for transit management (autofs) 998 * - Flagged as mountpoint 999 * - Flagged as automount point 1000 * 1001 * This may only be called in refwalk mode. 1002 * 1003 * Serialization is taken care of in namespace.c 1004 */ 1005 static int follow_managed(struct path *path, unsigned flags) 1006 { 1007 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 1008 unsigned managed; 1009 bool need_mntput = false; 1010 int ret = 0; 1011 1012 /* Given that we're not holding a lock here, we retain the value in a 1013 * local variable for each dentry as we look at it so that we don't see 1014 * the components of that value change under us */ 1015 while (managed = ACCESS_ONCE(path->dentry->d_flags), 1016 managed &= DCACHE_MANAGED_DENTRY, 1017 unlikely(managed != 0)) { 1018 /* Allow the filesystem to manage the transit without i_mutex 1019 * being held. */ 1020 if (managed & DCACHE_MANAGE_TRANSIT) { 1021 BUG_ON(!path->dentry->d_op); 1022 BUG_ON(!path->dentry->d_op->d_manage); 1023 ret = path->dentry->d_op->d_manage(path->dentry, false); 1024 if (ret < 0) 1025 break; 1026 } 1027 1028 /* Transit to a mounted filesystem. */ 1029 if (managed & DCACHE_MOUNTED) { 1030 struct vfsmount *mounted = lookup_mnt(path); 1031 if (mounted) { 1032 dput(path->dentry); 1033 if (need_mntput) 1034 mntput(path->mnt); 1035 path->mnt = mounted; 1036 path->dentry = dget(mounted->mnt_root); 1037 need_mntput = true; 1038 continue; 1039 } 1040 1041 /* Something is mounted on this dentry in another 1042 * namespace and/or whatever was mounted there in this 1043 * namespace got unmounted before we managed to get the 1044 * vfsmount_lock */ 1045 } 1046 1047 /* Handle an automount point */ 1048 if (managed & DCACHE_NEED_AUTOMOUNT) { 1049 ret = follow_automount(path, flags, &need_mntput); 1050 if (ret < 0) 1051 break; 1052 continue; 1053 } 1054 1055 /* We didn't change the current path point */ 1056 break; 1057 } 1058 1059 if (need_mntput && path->mnt == mnt) 1060 mntput(path->mnt); 1061 if (ret == -EISDIR) 1062 ret = 0; 1063 return ret < 0 ? ret : need_mntput; 1064 } 1065 1066 int follow_down_one(struct path *path) 1067 { 1068 struct vfsmount *mounted; 1069 1070 mounted = lookup_mnt(path); 1071 if (mounted) { 1072 dput(path->dentry); 1073 mntput(path->mnt); 1074 path->mnt = mounted; 1075 path->dentry = dget(mounted->mnt_root); 1076 return 1; 1077 } 1078 return 0; 1079 } 1080 1081 static inline bool managed_dentry_might_block(struct dentry *dentry) 1082 { 1083 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && 1084 dentry->d_op->d_manage(dentry, true) < 0); 1085 } 1086 1087 /* 1088 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1089 * we meet a managed dentry that would need blocking. 1090 */ 1091 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1092 struct inode **inode) 1093 { 1094 for (;;) { 1095 struct mount *mounted; 1096 /* 1097 * Don't forget we might have a non-mountpoint managed dentry 1098 * that wants to block transit. 1099 */ 1100 if (unlikely(managed_dentry_might_block(path->dentry))) 1101 return false; 1102 1103 if (!d_mountpoint(path->dentry)) 1104 break; 1105 1106 mounted = __lookup_mnt(path->mnt, path->dentry, 1); 1107 if (!mounted) 1108 break; 1109 path->mnt = &mounted->mnt; 1110 path->dentry = mounted->mnt.mnt_root; 1111 nd->flags |= LOOKUP_JUMPED; 1112 nd->seq = read_seqcount_begin(&path->dentry->d_seq); 1113 /* 1114 * Update the inode too. We don't need to re-check the 1115 * dentry sequence number here after this d_inode read, 1116 * because a mount-point is always pinned. 1117 */ 1118 *inode = path->dentry->d_inode; 1119 } 1120 return true; 1121 } 1122 1123 static void follow_mount_rcu(struct nameidata *nd) 1124 { 1125 while (d_mountpoint(nd->path.dentry)) { 1126 struct mount *mounted; 1127 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); 1128 if (!mounted) 1129 break; 1130 nd->path.mnt = &mounted->mnt; 1131 nd->path.dentry = mounted->mnt.mnt_root; 1132 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1133 } 1134 } 1135 1136 static int follow_dotdot_rcu(struct nameidata *nd) 1137 { 1138 set_root_rcu(nd); 1139 1140 while (1) { 1141 if (nd->path.dentry == nd->root.dentry && 1142 nd->path.mnt == nd->root.mnt) { 1143 break; 1144 } 1145 if (nd->path.dentry != nd->path.mnt->mnt_root) { 1146 struct dentry *old = nd->path.dentry; 1147 struct dentry *parent = old->d_parent; 1148 unsigned seq; 1149 1150 seq = read_seqcount_begin(&parent->d_seq); 1151 if (read_seqcount_retry(&old->d_seq, nd->seq)) 1152 goto failed; 1153 nd->path.dentry = parent; 1154 nd->seq = seq; 1155 break; 1156 } 1157 if (!follow_up_rcu(&nd->path)) 1158 break; 1159 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1160 } 1161 follow_mount_rcu(nd); 1162 nd->inode = nd->path.dentry->d_inode; 1163 return 0; 1164 1165 failed: 1166 nd->flags &= ~LOOKUP_RCU; 1167 if (!(nd->flags & LOOKUP_ROOT)) 1168 nd->root.mnt = NULL; 1169 unlock_rcu_walk(); 1170 return -ECHILD; 1171 } 1172 1173 /* 1174 * Follow down to the covering mount currently visible to userspace. At each 1175 * point, the filesystem owning that dentry may be queried as to whether the 1176 * caller is permitted to proceed or not. 1177 */ 1178 int follow_down(struct path *path) 1179 { 1180 unsigned managed; 1181 int ret; 1182 1183 while (managed = ACCESS_ONCE(path->dentry->d_flags), 1184 unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1185 /* Allow the filesystem to manage the transit without i_mutex 1186 * being held. 1187 * 1188 * We indicate to the filesystem if someone is trying to mount 1189 * something here. This gives autofs the chance to deny anyone 1190 * other than its daemon the right to mount on its 1191 * superstructure. 1192 * 1193 * The filesystem may sleep at this point. 1194 */ 1195 if (managed & DCACHE_MANAGE_TRANSIT) { 1196 BUG_ON(!path->dentry->d_op); 1197 BUG_ON(!path->dentry->d_op->d_manage); 1198 ret = path->dentry->d_op->d_manage( 1199 path->dentry, false); 1200 if (ret < 0) 1201 return ret == -EISDIR ? 0 : ret; 1202 } 1203 1204 /* Transit to a mounted filesystem. */ 1205 if (managed & DCACHE_MOUNTED) { 1206 struct vfsmount *mounted = lookup_mnt(path); 1207 if (!mounted) 1208 break; 1209 dput(path->dentry); 1210 mntput(path->mnt); 1211 path->mnt = mounted; 1212 path->dentry = dget(mounted->mnt_root); 1213 continue; 1214 } 1215 1216 /* Don't handle automount points here */ 1217 break; 1218 } 1219 return 0; 1220 } 1221 1222 /* 1223 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 1224 */ 1225 static void follow_mount(struct path *path) 1226 { 1227 while (d_mountpoint(path->dentry)) { 1228 struct vfsmount *mounted = lookup_mnt(path); 1229 if (!mounted) 1230 break; 1231 dput(path->dentry); 1232 mntput(path->mnt); 1233 path->mnt = mounted; 1234 path->dentry = dget(mounted->mnt_root); 1235 } 1236 } 1237 1238 static void follow_dotdot(struct nameidata *nd) 1239 { 1240 set_root(nd); 1241 1242 while(1) { 1243 struct dentry *old = nd->path.dentry; 1244 1245 if (nd->path.dentry == nd->root.dentry && 1246 nd->path.mnt == nd->root.mnt) { 1247 break; 1248 } 1249 if (nd->path.dentry != nd->path.mnt->mnt_root) { 1250 /* rare case of legitimate dget_parent()... */ 1251 nd->path.dentry = dget_parent(nd->path.dentry); 1252 dput(old); 1253 break; 1254 } 1255 if (!follow_up(&nd->path)) 1256 break; 1257 } 1258 follow_mount(&nd->path); 1259 nd->inode = nd->path.dentry->d_inode; 1260 } 1261 1262 /* 1263 * This looks up the name in dcache, possibly revalidates the old dentry and 1264 * allocates a new one if not found or not valid. In the need_lookup argument 1265 * returns whether i_op->lookup is necessary. 1266 * 1267 * dir->d_inode->i_mutex must be held 1268 */ 1269 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1270 unsigned int flags, bool *need_lookup) 1271 { 1272 struct dentry *dentry; 1273 int error; 1274 1275 *need_lookup = false; 1276 dentry = d_lookup(dir, name); 1277 if (dentry) { 1278 if (d_need_lookup(dentry)) { 1279 *need_lookup = true; 1280 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1281 error = d_revalidate(dentry, flags); 1282 if (unlikely(error <= 0)) { 1283 if (error < 0) { 1284 dput(dentry); 1285 return ERR_PTR(error); 1286 } else if (!d_invalidate(dentry)) { 1287 dput(dentry); 1288 dentry = NULL; 1289 } 1290 } 1291 } 1292 } 1293 1294 if (!dentry) { 1295 dentry = d_alloc(dir, name); 1296 if (unlikely(!dentry)) 1297 return ERR_PTR(-ENOMEM); 1298 1299 *need_lookup = true; 1300 } 1301 return dentry; 1302 } 1303 1304 /* 1305 * Call i_op->lookup on the dentry. The dentry must be negative but may be 1306 * hashed if it was pouplated with DCACHE_NEED_LOOKUP. 1307 * 1308 * dir->d_inode->i_mutex must be held 1309 */ 1310 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 1311 unsigned int flags) 1312 { 1313 struct dentry *old; 1314 1315 /* Don't create child dentry for a dead directory. */ 1316 if (unlikely(IS_DEADDIR(dir))) { 1317 dput(dentry); 1318 return ERR_PTR(-ENOENT); 1319 } 1320 1321 old = dir->i_op->lookup(dir, dentry, flags); 1322 if (unlikely(old)) { 1323 dput(dentry); 1324 dentry = old; 1325 } 1326 return dentry; 1327 } 1328 1329 static struct dentry *__lookup_hash(struct qstr *name, 1330 struct dentry *base, unsigned int flags) 1331 { 1332 bool need_lookup; 1333 struct dentry *dentry; 1334 1335 dentry = lookup_dcache(name, base, flags, &need_lookup); 1336 if (!need_lookup) 1337 return dentry; 1338 1339 return lookup_real(base->d_inode, dentry, flags); 1340 } 1341 1342 /* 1343 * It's more convoluted than I'd like it to be, but... it's still fairly 1344 * small and for now I'd prefer to have fast path as straight as possible. 1345 * It _is_ time-critical. 1346 */ 1347 static int lookup_fast(struct nameidata *nd, struct qstr *name, 1348 struct path *path, struct inode **inode) 1349 { 1350 struct vfsmount *mnt = nd->path.mnt; 1351 struct dentry *dentry, *parent = nd->path.dentry; 1352 int need_reval = 1; 1353 int status = 1; 1354 int err; 1355 1356 /* 1357 * Rename seqlock is not required here because in the off chance 1358 * of a false negative due to a concurrent rename, we're going to 1359 * do the non-racy lookup, below. 1360 */ 1361 if (nd->flags & LOOKUP_RCU) { 1362 unsigned seq; 1363 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode); 1364 if (!dentry) 1365 goto unlazy; 1366 1367 /* 1368 * This sequence count validates that the inode matches 1369 * the dentry name information from lookup. 1370 */ 1371 *inode = dentry->d_inode; 1372 if (read_seqcount_retry(&dentry->d_seq, seq)) 1373 return -ECHILD; 1374 1375 /* 1376 * This sequence count validates that the parent had no 1377 * changes while we did the lookup of the dentry above. 1378 * 1379 * The memory barrier in read_seqcount_begin of child is 1380 * enough, we can use __read_seqcount_retry here. 1381 */ 1382 if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 1383 return -ECHILD; 1384 nd->seq = seq; 1385 1386 if (unlikely(d_need_lookup(dentry))) 1387 goto unlazy; 1388 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 1389 status = d_revalidate(dentry, nd->flags); 1390 if (unlikely(status <= 0)) { 1391 if (status != -ECHILD) 1392 need_reval = 0; 1393 goto unlazy; 1394 } 1395 } 1396 path->mnt = mnt; 1397 path->dentry = dentry; 1398 if (unlikely(!__follow_mount_rcu(nd, path, inode))) 1399 goto unlazy; 1400 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 1401 goto unlazy; 1402 return 0; 1403 unlazy: 1404 if (unlazy_walk(nd, dentry)) 1405 return -ECHILD; 1406 } else { 1407 dentry = __d_lookup(parent, name); 1408 } 1409 1410 if (unlikely(!dentry)) 1411 goto need_lookup; 1412 1413 if (unlikely(d_need_lookup(dentry))) { 1414 dput(dentry); 1415 goto need_lookup; 1416 } 1417 1418 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 1419 status = d_revalidate(dentry, nd->flags); 1420 if (unlikely(status <= 0)) { 1421 if (status < 0) { 1422 dput(dentry); 1423 return status; 1424 } 1425 if (!d_invalidate(dentry)) { 1426 dput(dentry); 1427 goto need_lookup; 1428 } 1429 } 1430 1431 path->mnt = mnt; 1432 path->dentry = dentry; 1433 err = follow_managed(path, nd->flags); 1434 if (unlikely(err < 0)) { 1435 path_put_conditional(path, nd); 1436 return err; 1437 } 1438 if (err) 1439 nd->flags |= LOOKUP_JUMPED; 1440 *inode = path->dentry->d_inode; 1441 return 0; 1442 1443 need_lookup: 1444 return 1; 1445 } 1446 1447 /* Fast lookup failed, do it the slow way */ 1448 static int lookup_slow(struct nameidata *nd, struct qstr *name, 1449 struct path *path) 1450 { 1451 struct dentry *dentry, *parent; 1452 int err; 1453 1454 parent = nd->path.dentry; 1455 BUG_ON(nd->inode != parent->d_inode); 1456 1457 mutex_lock(&parent->d_inode->i_mutex); 1458 dentry = __lookup_hash(name, parent, nd->flags); 1459 mutex_unlock(&parent->d_inode->i_mutex); 1460 if (IS_ERR(dentry)) 1461 return PTR_ERR(dentry); 1462 path->mnt = nd->path.mnt; 1463 path->dentry = dentry; 1464 err = follow_managed(path, nd->flags); 1465 if (unlikely(err < 0)) { 1466 path_put_conditional(path, nd); 1467 return err; 1468 } 1469 if (err) 1470 nd->flags |= LOOKUP_JUMPED; 1471 return 0; 1472 } 1473 1474 static inline int may_lookup(struct nameidata *nd) 1475 { 1476 if (nd->flags & LOOKUP_RCU) { 1477 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1478 if (err != -ECHILD) 1479 return err; 1480 if (unlazy_walk(nd, NULL)) 1481 return -ECHILD; 1482 } 1483 return inode_permission(nd->inode, MAY_EXEC); 1484 } 1485 1486 static inline int handle_dots(struct nameidata *nd, int type) 1487 { 1488 if (type == LAST_DOTDOT) { 1489 if (nd->flags & LOOKUP_RCU) { 1490 if (follow_dotdot_rcu(nd)) 1491 return -ECHILD; 1492 } else 1493 follow_dotdot(nd); 1494 } 1495 return 0; 1496 } 1497 1498 static void terminate_walk(struct nameidata *nd) 1499 { 1500 if (!(nd->flags & LOOKUP_RCU)) { 1501 path_put(&nd->path); 1502 } else { 1503 nd->flags &= ~LOOKUP_RCU; 1504 if (!(nd->flags & LOOKUP_ROOT)) 1505 nd->root.mnt = NULL; 1506 unlock_rcu_walk(); 1507 } 1508 } 1509 1510 /* 1511 * Do we need to follow links? We _really_ want to be able 1512 * to do this check without having to look at inode->i_op, 1513 * so we keep a cache of "no, this doesn't need follow_link" 1514 * for the common case. 1515 */ 1516 static inline int should_follow_link(struct inode *inode, int follow) 1517 { 1518 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { 1519 if (likely(inode->i_op->follow_link)) 1520 return follow; 1521 1522 /* This gets set once for the inode lifetime */ 1523 spin_lock(&inode->i_lock); 1524 inode->i_opflags |= IOP_NOFOLLOW; 1525 spin_unlock(&inode->i_lock); 1526 } 1527 return 0; 1528 } 1529 1530 static inline int walk_component(struct nameidata *nd, struct path *path, 1531 struct qstr *name, int type, int follow) 1532 { 1533 struct inode *inode; 1534 int err; 1535 /* 1536 * "." and ".." are special - ".." especially so because it has 1537 * to be able to know about the current root directory and 1538 * parent relationships. 1539 */ 1540 if (unlikely(type != LAST_NORM)) 1541 return handle_dots(nd, type); 1542 err = lookup_fast(nd, name, path, &inode); 1543 if (unlikely(err)) { 1544 if (err < 0) 1545 goto out_err; 1546 1547 err = lookup_slow(nd, name, path); 1548 if (err < 0) 1549 goto out_err; 1550 1551 inode = path->dentry->d_inode; 1552 } 1553 err = -ENOENT; 1554 if (!inode) 1555 goto out_path_put; 1556 1557 if (should_follow_link(inode, follow)) { 1558 if (nd->flags & LOOKUP_RCU) { 1559 if (unlikely(unlazy_walk(nd, path->dentry))) { 1560 err = -ECHILD; 1561 goto out_err; 1562 } 1563 } 1564 BUG_ON(inode != path->dentry->d_inode); 1565 return 1; 1566 } 1567 path_to_nameidata(path, nd); 1568 nd->inode = inode; 1569 return 0; 1570 1571 out_path_put: 1572 path_to_nameidata(path, nd); 1573 out_err: 1574 terminate_walk(nd); 1575 return err; 1576 } 1577 1578 /* 1579 * This limits recursive symlink follows to 8, while 1580 * limiting consecutive symlinks to 40. 1581 * 1582 * Without that kind of total limit, nasty chains of consecutive 1583 * symlinks can cause almost arbitrarily long lookups. 1584 */ 1585 static inline int nested_symlink(struct path *path, struct nameidata *nd) 1586 { 1587 int res; 1588 1589 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1590 path_put_conditional(path, nd); 1591 path_put(&nd->path); 1592 return -ELOOP; 1593 } 1594 BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1595 1596 nd->depth++; 1597 current->link_count++; 1598 1599 do { 1600 struct path link = *path; 1601 void *cookie; 1602 1603 res = follow_link(&link, nd, &cookie); 1604 if (res) 1605 break; 1606 res = walk_component(nd, path, &nd->last, 1607 nd->last_type, LOOKUP_FOLLOW); 1608 put_link(nd, &link, cookie); 1609 } while (res > 0); 1610 1611 current->link_count--; 1612 nd->depth--; 1613 return res; 1614 } 1615 1616 /* 1617 * We really don't want to look at inode->i_op->lookup 1618 * when we don't have to. So we keep a cache bit in 1619 * the inode ->i_opflags field that says "yes, we can 1620 * do lookup on this inode". 1621 */ 1622 static inline int can_lookup(struct inode *inode) 1623 { 1624 if (likely(inode->i_opflags & IOP_LOOKUP)) 1625 return 1; 1626 if (likely(!inode->i_op->lookup)) 1627 return 0; 1628 1629 /* We do this once for the lifetime of the inode */ 1630 spin_lock(&inode->i_lock); 1631 inode->i_opflags |= IOP_LOOKUP; 1632 spin_unlock(&inode->i_lock); 1633 return 1; 1634 } 1635 1636 /* 1637 * We can do the critical dentry name comparison and hashing 1638 * operations one word at a time, but we are limited to: 1639 * 1640 * - Architectures with fast unaligned word accesses. We could 1641 * do a "get_unaligned()" if this helps and is sufficiently 1642 * fast. 1643 * 1644 * - Little-endian machines (so that we can generate the mask 1645 * of low bytes efficiently). Again, we *could* do a byte 1646 * swapping load on big-endian architectures if that is not 1647 * expensive enough to make the optimization worthless. 1648 * 1649 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1650 * do not trap on the (extremely unlikely) case of a page 1651 * crossing operation. 1652 * 1653 * - Furthermore, we need an efficient 64-bit compile for the 1654 * 64-bit case in order to generate the "number of bytes in 1655 * the final mask". Again, that could be replaced with a 1656 * efficient population count instruction or similar. 1657 */ 1658 #ifdef CONFIG_DCACHE_WORD_ACCESS 1659 1660 #include <asm/word-at-a-time.h> 1661 1662 #ifdef CONFIG_64BIT 1663 1664 static inline unsigned int fold_hash(unsigned long hash) 1665 { 1666 hash += hash >> (8*sizeof(int)); 1667 return hash; 1668 } 1669 1670 #else /* 32-bit case */ 1671 1672 #define fold_hash(x) (x) 1673 1674 #endif 1675 1676 unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1677 { 1678 unsigned long a, mask; 1679 unsigned long hash = 0; 1680 1681 for (;;) { 1682 a = load_unaligned_zeropad(name); 1683 if (len < sizeof(unsigned long)) 1684 break; 1685 hash += a; 1686 hash *= 9; 1687 name += sizeof(unsigned long); 1688 len -= sizeof(unsigned long); 1689 if (!len) 1690 goto done; 1691 } 1692 mask = ~(~0ul << len*8); 1693 hash += mask & a; 1694 done: 1695 return fold_hash(hash); 1696 } 1697 EXPORT_SYMBOL(full_name_hash); 1698 1699 /* 1700 * Calculate the length and hash of the path component, and 1701 * return the length of the component; 1702 */ 1703 static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1704 { 1705 unsigned long a, b, adata, bdata, mask, hash, len; 1706 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1707 1708 hash = a = 0; 1709 len = -sizeof(unsigned long); 1710 do { 1711 hash = (hash + a) * 9; 1712 len += sizeof(unsigned long); 1713 a = load_unaligned_zeropad(name+len); 1714 b = a ^ REPEAT_BYTE('/'); 1715 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1716 1717 adata = prep_zero_mask(a, adata, &constants); 1718 bdata = prep_zero_mask(b, bdata, &constants); 1719 1720 mask = create_zero_mask(adata | bdata); 1721 1722 hash += a & zero_bytemask(mask); 1723 *hashp = fold_hash(hash); 1724 1725 return len + find_zero(mask); 1726 } 1727 1728 #else 1729 1730 unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1731 { 1732 unsigned long hash = init_name_hash(); 1733 while (len--) 1734 hash = partial_name_hash(*name++, hash); 1735 return end_name_hash(hash); 1736 } 1737 EXPORT_SYMBOL(full_name_hash); 1738 1739 /* 1740 * We know there's a real path component here of at least 1741 * one character. 1742 */ 1743 static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1744 { 1745 unsigned long hash = init_name_hash(); 1746 unsigned long len = 0, c; 1747 1748 c = (unsigned char)*name; 1749 do { 1750 len++; 1751 hash = partial_name_hash(c, hash); 1752 c = (unsigned char)name[len]; 1753 } while (c && c != '/'); 1754 *hashp = end_name_hash(hash); 1755 return len; 1756 } 1757 1758 #endif 1759 1760 /* 1761 * Name resolution. 1762 * This is the basic name resolution function, turning a pathname into 1763 * the final dentry. We expect 'base' to be positive and a directory. 1764 * 1765 * Returns 0 and nd will have valid dentry and mnt on success. 1766 * Returns error and drops reference to input namei data on failure. 1767 */ 1768 static int link_path_walk(const char *name, struct nameidata *nd) 1769 { 1770 struct path next; 1771 int err; 1772 1773 while (*name=='/') 1774 name++; 1775 if (!*name) 1776 return 0; 1777 1778 /* At this point we know we have a real path component. */ 1779 for(;;) { 1780 struct qstr this; 1781 long len; 1782 int type; 1783 1784 err = may_lookup(nd); 1785 if (err) 1786 break; 1787 1788 len = hash_name(name, &this.hash); 1789 this.name = name; 1790 this.len = len; 1791 1792 type = LAST_NORM; 1793 if (name[0] == '.') switch (len) { 1794 case 2: 1795 if (name[1] == '.') { 1796 type = LAST_DOTDOT; 1797 nd->flags |= LOOKUP_JUMPED; 1798 } 1799 break; 1800 case 1: 1801 type = LAST_DOT; 1802 } 1803 if (likely(type == LAST_NORM)) { 1804 struct dentry *parent = nd->path.dentry; 1805 nd->flags &= ~LOOKUP_JUMPED; 1806 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1807 err = parent->d_op->d_hash(parent, nd->inode, 1808 &this); 1809 if (err < 0) 1810 break; 1811 } 1812 } 1813 1814 if (!name[len]) 1815 goto last_component; 1816 /* 1817 * If it wasn't NUL, we know it was '/'. Skip that 1818 * slash, and continue until no more slashes. 1819 */ 1820 do { 1821 len++; 1822 } while (unlikely(name[len] == '/')); 1823 if (!name[len]) 1824 goto last_component; 1825 name += len; 1826 1827 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); 1828 if (err < 0) 1829 return err; 1830 1831 if (err) { 1832 err = nested_symlink(&next, nd); 1833 if (err) 1834 return err; 1835 } 1836 if (can_lookup(nd->inode)) 1837 continue; 1838 err = -ENOTDIR; 1839 break; 1840 /* here ends the main loop */ 1841 1842 last_component: 1843 nd->last = this; 1844 nd->last_type = type; 1845 return 0; 1846 } 1847 terminate_walk(nd); 1848 return err; 1849 } 1850 1851 static int path_init(int dfd, const char *name, unsigned int flags, 1852 struct nameidata *nd, struct file **fp) 1853 { 1854 int retval = 0; 1855 1856 nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1857 nd->flags = flags | LOOKUP_JUMPED; 1858 nd->depth = 0; 1859 if (flags & LOOKUP_ROOT) { 1860 struct inode *inode = nd->root.dentry->d_inode; 1861 if (*name) { 1862 if (!inode->i_op->lookup) 1863 return -ENOTDIR; 1864 retval = inode_permission(inode, MAY_EXEC); 1865 if (retval) 1866 return retval; 1867 } 1868 nd->path = nd->root; 1869 nd->inode = inode; 1870 if (flags & LOOKUP_RCU) { 1871 lock_rcu_walk(); 1872 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1873 } else { 1874 path_get(&nd->path); 1875 } 1876 return 0; 1877 } 1878 1879 nd->root.mnt = NULL; 1880 1881 if (*name=='/') { 1882 if (flags & LOOKUP_RCU) { 1883 lock_rcu_walk(); 1884 set_root_rcu(nd); 1885 } else { 1886 set_root(nd); 1887 path_get(&nd->root); 1888 } 1889 nd->path = nd->root; 1890 } else if (dfd == AT_FDCWD) { 1891 if (flags & LOOKUP_RCU) { 1892 struct fs_struct *fs = current->fs; 1893 unsigned seq; 1894 1895 lock_rcu_walk(); 1896 1897 do { 1898 seq = read_seqcount_begin(&fs->seq); 1899 nd->path = fs->pwd; 1900 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1901 } while (read_seqcount_retry(&fs->seq, seq)); 1902 } else { 1903 get_fs_pwd(current->fs, &nd->path); 1904 } 1905 } else { 1906 struct fd f = fdget_raw(dfd); 1907 struct dentry *dentry; 1908 1909 if (!f.file) 1910 return -EBADF; 1911 1912 dentry = f.file->f_path.dentry; 1913 1914 if (*name) { 1915 if (!S_ISDIR(dentry->d_inode->i_mode)) { 1916 fdput(f); 1917 return -ENOTDIR; 1918 } 1919 1920 retval = inode_permission(dentry->d_inode, MAY_EXEC); 1921 if (retval) { 1922 fdput(f); 1923 return retval; 1924 } 1925 } 1926 1927 nd->path = f.file->f_path; 1928 if (flags & LOOKUP_RCU) { 1929 if (f.need_put) 1930 *fp = f.file; 1931 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1932 lock_rcu_walk(); 1933 } else { 1934 path_get(&nd->path); 1935 fdput(f); 1936 } 1937 } 1938 1939 nd->inode = nd->path.dentry->d_inode; 1940 return 0; 1941 } 1942 1943 static inline int lookup_last(struct nameidata *nd, struct path *path) 1944 { 1945 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1946 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1947 1948 nd->flags &= ~LOOKUP_PARENT; 1949 return walk_component(nd, path, &nd->last, nd->last_type, 1950 nd->flags & LOOKUP_FOLLOW); 1951 } 1952 1953 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1954 static int path_lookupat(int dfd, const char *name, 1955 unsigned int flags, struct nameidata *nd) 1956 { 1957 struct file *base = NULL; 1958 struct path path; 1959 int err; 1960 1961 /* 1962 * Path walking is largely split up into 2 different synchronisation 1963 * schemes, rcu-walk and ref-walk (explained in 1964 * Documentation/filesystems/path-lookup.txt). These share much of the 1965 * path walk code, but some things particularly setup, cleanup, and 1966 * following mounts are sufficiently divergent that functions are 1967 * duplicated. Typically there is a function foo(), and its RCU 1968 * analogue, foo_rcu(). 1969 * 1970 * -ECHILD is the error number of choice (just to avoid clashes) that 1971 * is returned if some aspect of an rcu-walk fails. Such an error must 1972 * be handled by restarting a traditional ref-walk (which will always 1973 * be able to complete). 1974 */ 1975 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); 1976 1977 if (unlikely(err)) 1978 return err; 1979 1980 current->total_link_count = 0; 1981 err = link_path_walk(name, nd); 1982 1983 if (!err && !(flags & LOOKUP_PARENT)) { 1984 err = lookup_last(nd, &path); 1985 while (err > 0) { 1986 void *cookie; 1987 struct path link = path; 1988 err = may_follow_link(&link, nd); 1989 if (unlikely(err)) 1990 break; 1991 nd->flags |= LOOKUP_PARENT; 1992 err = follow_link(&link, nd, &cookie); 1993 if (err) 1994 break; 1995 err = lookup_last(nd, &path); 1996 put_link(nd, &link, cookie); 1997 } 1998 } 1999 2000 if (!err) 2001 err = complete_walk(nd); 2002 2003 if (!err && nd->flags & LOOKUP_DIRECTORY) { 2004 if (!nd->inode->i_op->lookup) { 2005 path_put(&nd->path); 2006 err = -ENOTDIR; 2007 } 2008 } 2009 2010 if (base) 2011 fput(base); 2012 2013 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 2014 path_put(&nd->root); 2015 nd->root.mnt = NULL; 2016 } 2017 return err; 2018 } 2019 2020 static int filename_lookup(int dfd, struct filename *name, 2021 unsigned int flags, struct nameidata *nd) 2022 { 2023 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); 2024 if (unlikely(retval == -ECHILD)) 2025 retval = path_lookupat(dfd, name->name, flags, nd); 2026 if (unlikely(retval == -ESTALE)) 2027 retval = path_lookupat(dfd, name->name, 2028 flags | LOOKUP_REVAL, nd); 2029 2030 if (likely(!retval)) 2031 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); 2032 return retval; 2033 } 2034 2035 static int do_path_lookup(int dfd, const char *name, 2036 unsigned int flags, struct nameidata *nd) 2037 { 2038 struct filename filename = { .name = name }; 2039 2040 return filename_lookup(dfd, &filename, flags, nd); 2041 } 2042 2043 /* does lookup, returns the object with parent locked */ 2044 struct dentry *kern_path_locked(const char *name, struct path *path) 2045 { 2046 struct nameidata nd; 2047 struct dentry *d; 2048 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); 2049 if (err) 2050 return ERR_PTR(err); 2051 if (nd.last_type != LAST_NORM) { 2052 path_put(&nd.path); 2053 return ERR_PTR(-EINVAL); 2054 } 2055 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2056 d = __lookup_hash(&nd.last, nd.path.dentry, 0); 2057 if (IS_ERR(d)) { 2058 mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 2059 path_put(&nd.path); 2060 return d; 2061 } 2062 *path = nd.path; 2063 return d; 2064 } 2065 2066 int kern_path(const char *name, unsigned int flags, struct path *path) 2067 { 2068 struct nameidata nd; 2069 int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 2070 if (!res) 2071 *path = nd.path; 2072 return res; 2073 } 2074 2075 /** 2076 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 2077 * @dentry: pointer to dentry of the base directory 2078 * @mnt: pointer to vfs mount of the base directory 2079 * @name: pointer to file name 2080 * @flags: lookup flags 2081 * @path: pointer to struct path to fill 2082 */ 2083 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 2084 const char *name, unsigned int flags, 2085 struct path *path) 2086 { 2087 struct nameidata nd; 2088 int err; 2089 nd.root.dentry = dentry; 2090 nd.root.mnt = mnt; 2091 BUG_ON(flags & LOOKUP_PARENT); 2092 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 2093 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); 2094 if (!err) 2095 *path = nd.path; 2096 return err; 2097 } 2098 2099 /* 2100 * Restricted form of lookup. Doesn't follow links, single-component only, 2101 * needs parent already locked. Doesn't follow mounts. 2102 * SMP-safe. 2103 */ 2104 static struct dentry *lookup_hash(struct nameidata *nd) 2105 { 2106 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); 2107 } 2108 2109 /** 2110 * lookup_one_len - filesystem helper to lookup single pathname component 2111 * @name: pathname component to lookup 2112 * @base: base directory to lookup from 2113 * @len: maximum length @len should be interpreted to 2114 * 2115 * Note that this routine is purely a helper for filesystem usage and should 2116 * not be called by generic code. Also note that by using this function the 2117 * nameidata argument is passed to the filesystem methods and a filesystem 2118 * using this helper needs to be prepared for that. 2119 */ 2120 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2121 { 2122 struct qstr this; 2123 unsigned int c; 2124 int err; 2125 2126 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 2127 2128 this.name = name; 2129 this.len = len; 2130 this.hash = full_name_hash(name, len); 2131 if (!len) 2132 return ERR_PTR(-EACCES); 2133 2134 while (len--) { 2135 c = *(const unsigned char *)name++; 2136 if (c == '/' || c == '\0') 2137 return ERR_PTR(-EACCES); 2138 } 2139 /* 2140 * See if the low-level filesystem might want 2141 * to use its own hash.. 2142 */ 2143 if (base->d_flags & DCACHE_OP_HASH) { 2144 int err = base->d_op->d_hash(base, base->d_inode, &this); 2145 if (err < 0) 2146 return ERR_PTR(err); 2147 } 2148 2149 err = inode_permission(base->d_inode, MAY_EXEC); 2150 if (err) 2151 return ERR_PTR(err); 2152 2153 return __lookup_hash(&this, base, 0); 2154 } 2155 2156 int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 2157 struct path *path, int *empty) 2158 { 2159 struct nameidata nd; 2160 struct filename *tmp = getname_flags(name, flags, empty); 2161 int err = PTR_ERR(tmp); 2162 if (!IS_ERR(tmp)) { 2163 2164 BUG_ON(flags & LOOKUP_PARENT); 2165 2166 err = filename_lookup(dfd, tmp, flags, &nd); 2167 putname(tmp); 2168 if (!err) 2169 *path = nd.path; 2170 } 2171 return err; 2172 } 2173 2174 int user_path_at(int dfd, const char __user *name, unsigned flags, 2175 struct path *path) 2176 { 2177 return user_path_at_empty(dfd, name, flags, path, NULL); 2178 } 2179 2180 /* 2181 * NB: most callers don't do anything directly with the reference to the 2182 * to struct filename, but the nd->last pointer points into the name string 2183 * allocated by getname. So we must hold the reference to it until all 2184 * path-walking is complete. 2185 */ 2186 static struct filename * 2187 user_path_parent(int dfd, const char __user *path, struct nameidata *nd) 2188 { 2189 struct filename *s = getname(path); 2190 int error; 2191 2192 if (IS_ERR(s)) 2193 return s; 2194 2195 error = filename_lookup(dfd, s, LOOKUP_PARENT, nd); 2196 if (error) { 2197 putname(s); 2198 return ERR_PTR(error); 2199 } 2200 2201 return s; 2202 } 2203 2204 /* 2205 * It's inline, so penalty for filesystems that don't use sticky bit is 2206 * minimal. 2207 */ 2208 static inline int check_sticky(struct inode *dir, struct inode *inode) 2209 { 2210 kuid_t fsuid = current_fsuid(); 2211 2212 if (!(dir->i_mode & S_ISVTX)) 2213 return 0; 2214 if (uid_eq(inode->i_uid, fsuid)) 2215 return 0; 2216 if (uid_eq(dir->i_uid, fsuid)) 2217 return 0; 2218 return !inode_capable(inode, CAP_FOWNER); 2219 } 2220 2221 /* 2222 * Check whether we can remove a link victim from directory dir, check 2223 * whether the type of victim is right. 2224 * 1. We can't do it if dir is read-only (done in permission()) 2225 * 2. We should have write and exec permissions on dir 2226 * 3. We can't remove anything from append-only dir 2227 * 4. We can't do anything with immutable dir (done in permission()) 2228 * 5. If the sticky bit on dir is set we should either 2229 * a. be owner of dir, or 2230 * b. be owner of victim, or 2231 * c. have CAP_FOWNER capability 2232 * 6. If the victim is append-only or immutable we can't do antyhing with 2233 * links pointing to it. 2234 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 2235 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 2236 * 9. We can't remove a root or mountpoint. 2237 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 2238 * nfs_async_unlink(). 2239 */ 2240 static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 2241 { 2242 int error; 2243 2244 if (!victim->d_inode) 2245 return -ENOENT; 2246 2247 BUG_ON(victim->d_parent->d_inode != dir); 2248 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 2249 2250 error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 2251 if (error) 2252 return error; 2253 if (IS_APPEND(dir)) 2254 return -EPERM; 2255 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 2256 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 2257 return -EPERM; 2258 if (isdir) { 2259 if (!S_ISDIR(victim->d_inode->i_mode)) 2260 return -ENOTDIR; 2261 if (IS_ROOT(victim)) 2262 return -EBUSY; 2263 } else if (S_ISDIR(victim->d_inode->i_mode)) 2264 return -EISDIR; 2265 if (IS_DEADDIR(dir)) 2266 return -ENOENT; 2267 if (victim->d_flags & DCACHE_NFSFS_RENAMED) 2268 return -EBUSY; 2269 return 0; 2270 } 2271 2272 /* Check whether we can create an object with dentry child in directory 2273 * dir. 2274 * 1. We can't do it if child already exists (open has special treatment for 2275 * this case, but since we are inlined it's OK) 2276 * 2. We can't do it if dir is read-only (done in permission()) 2277 * 3. We should have write and exec permissions on dir 2278 * 4. We can't do it if dir is immutable (done in permission()) 2279 */ 2280 static inline int may_create(struct inode *dir, struct dentry *child) 2281 { 2282 if (child->d_inode) 2283 return -EEXIST; 2284 if (IS_DEADDIR(dir)) 2285 return -ENOENT; 2286 return inode_permission(dir, MAY_WRITE | MAY_EXEC); 2287 } 2288 2289 /* 2290 * p1 and p2 should be directories on the same fs. 2291 */ 2292 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 2293 { 2294 struct dentry *p; 2295 2296 if (p1 == p2) { 2297 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2298 return NULL; 2299 } 2300 2301 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 2302 2303 p = d_ancestor(p2, p1); 2304 if (p) { 2305 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2306 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 2307 return p; 2308 } 2309 2310 p = d_ancestor(p1, p2); 2311 if (p) { 2312 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2313 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 2314 return p; 2315 } 2316 2317 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2318 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 2319 return NULL; 2320 } 2321 2322 void unlock_rename(struct dentry *p1, struct dentry *p2) 2323 { 2324 mutex_unlock(&p1->d_inode->i_mutex); 2325 if (p1 != p2) { 2326 mutex_unlock(&p2->d_inode->i_mutex); 2327 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 2328 } 2329 } 2330 2331 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2332 bool want_excl) 2333 { 2334 int error = may_create(dir, dentry); 2335 if (error) 2336 return error; 2337 2338 if (!dir->i_op->create) 2339 return -EACCES; /* shouldn't it be ENOSYS? */ 2340 mode &= S_IALLUGO; 2341 mode |= S_IFREG; 2342 error = security_inode_create(dir, dentry, mode); 2343 if (error) 2344 return error; 2345 error = dir->i_op->create(dir, dentry, mode, want_excl); 2346 if (!error) 2347 fsnotify_create(dir, dentry); 2348 return error; 2349 } 2350 2351 static int may_open(struct path *path, int acc_mode, int flag) 2352 { 2353 struct dentry *dentry = path->dentry; 2354 struct inode *inode = dentry->d_inode; 2355 int error; 2356 2357 /* O_PATH? */ 2358 if (!acc_mode) 2359 return 0; 2360 2361 if (!inode) 2362 return -ENOENT; 2363 2364 switch (inode->i_mode & S_IFMT) { 2365 case S_IFLNK: 2366 return -ELOOP; 2367 case S_IFDIR: 2368 if (acc_mode & MAY_WRITE) 2369 return -EISDIR; 2370 break; 2371 case S_IFBLK: 2372 case S_IFCHR: 2373 if (path->mnt->mnt_flags & MNT_NODEV) 2374 return -EACCES; 2375 /*FALLTHRU*/ 2376 case S_IFIFO: 2377 case S_IFSOCK: 2378 flag &= ~O_TRUNC; 2379 break; 2380 } 2381 2382 error = inode_permission(inode, acc_mode); 2383 if (error) 2384 return error; 2385 2386 /* 2387 * An append-only file must be opened in append mode for writing. 2388 */ 2389 if (IS_APPEND(inode)) { 2390 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 2391 return -EPERM; 2392 if (flag & O_TRUNC) 2393 return -EPERM; 2394 } 2395 2396 /* O_NOATIME can only be set by the owner or superuser */ 2397 if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 2398 return -EPERM; 2399 2400 return 0; 2401 } 2402 2403 static int handle_truncate(struct file *filp) 2404 { 2405 struct path *path = &filp->f_path; 2406 struct inode *inode = path->dentry->d_inode; 2407 int error = get_write_access(inode); 2408 if (error) 2409 return error; 2410 /* 2411 * Refuse to truncate files with mandatory locks held on them. 2412 */ 2413 error = locks_verify_locked(inode); 2414 if (!error) 2415 error = security_path_truncate(path); 2416 if (!error) { 2417 error = do_truncate(path->dentry, 0, 2418 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2419 filp); 2420 } 2421 put_write_access(inode); 2422 return error; 2423 } 2424 2425 static inline int open_to_namei_flags(int flag) 2426 { 2427 if ((flag & O_ACCMODE) == 3) 2428 flag--; 2429 return flag; 2430 } 2431 2432 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2433 { 2434 int error = security_path_mknod(dir, dentry, mode, 0); 2435 if (error) 2436 return error; 2437 2438 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2439 if (error) 2440 return error; 2441 2442 return security_inode_create(dir->dentry->d_inode, dentry, mode); 2443 } 2444 2445 /* 2446 * Attempt to atomically look up, create and open a file from a negative 2447 * dentry. 2448 * 2449 * Returns 0 if successful. The file will have been created and attached to 2450 * @file by the filesystem calling finish_open(). 2451 * 2452 * Returns 1 if the file was looked up only or didn't need creating. The 2453 * caller will need to perform the open themselves. @path will have been 2454 * updated to point to the new dentry. This may be negative. 2455 * 2456 * Returns an error code otherwise. 2457 */ 2458 static int atomic_open(struct nameidata *nd, struct dentry *dentry, 2459 struct path *path, struct file *file, 2460 const struct open_flags *op, 2461 bool got_write, bool need_lookup, 2462 int *opened) 2463 { 2464 struct inode *dir = nd->path.dentry->d_inode; 2465 unsigned open_flag = open_to_namei_flags(op->open_flag); 2466 umode_t mode; 2467 int error; 2468 int acc_mode; 2469 int create_error = 0; 2470 struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2471 2472 BUG_ON(dentry->d_inode); 2473 2474 /* Don't create child dentry for a dead directory. */ 2475 if (unlikely(IS_DEADDIR(dir))) { 2476 error = -ENOENT; 2477 goto out; 2478 } 2479 2480 mode = op->mode; 2481 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2482 mode &= ~current_umask(); 2483 2484 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) { 2485 open_flag &= ~O_TRUNC; 2486 *opened |= FILE_CREATED; 2487 } 2488 2489 /* 2490 * Checking write permission is tricky, bacuse we don't know if we are 2491 * going to actually need it: O_CREAT opens should work as long as the 2492 * file exists. But checking existence breaks atomicity. The trick is 2493 * to check access and if not granted clear O_CREAT from the flags. 2494 * 2495 * Another problem is returing the "right" error value (e.g. for an 2496 * O_EXCL open we want to return EEXIST not EROFS). 2497 */ 2498 if (((open_flag & (O_CREAT | O_TRUNC)) || 2499 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 2500 if (!(open_flag & O_CREAT)) { 2501 /* 2502 * No O_CREATE -> atomicity not a requirement -> fall 2503 * back to lookup + open 2504 */ 2505 goto no_open; 2506 } else if (open_flag & (O_EXCL | O_TRUNC)) { 2507 /* Fall back and fail with the right error */ 2508 create_error = -EROFS; 2509 goto no_open; 2510 } else { 2511 /* No side effects, safe to clear O_CREAT */ 2512 create_error = -EROFS; 2513 open_flag &= ~O_CREAT; 2514 } 2515 } 2516 2517 if (open_flag & O_CREAT) { 2518 error = may_o_create(&nd->path, dentry, mode); 2519 if (error) { 2520 create_error = error; 2521 if (open_flag & O_EXCL) 2522 goto no_open; 2523 open_flag &= ~O_CREAT; 2524 } 2525 } 2526 2527 if (nd->flags & LOOKUP_DIRECTORY) 2528 open_flag |= O_DIRECTORY; 2529 2530 file->f_path.dentry = DENTRY_NOT_SET; 2531 file->f_path.mnt = nd->path.mnt; 2532 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 2533 opened); 2534 if (error < 0) { 2535 if (create_error && error == -ENOENT) 2536 error = create_error; 2537 goto out; 2538 } 2539 2540 acc_mode = op->acc_mode; 2541 if (*opened & FILE_CREATED) { 2542 fsnotify_create(dir, dentry); 2543 acc_mode = MAY_OPEN; 2544 } 2545 2546 if (error) { /* returned 1, that is */ 2547 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 2548 error = -EIO; 2549 goto out; 2550 } 2551 if (file->f_path.dentry) { 2552 dput(dentry); 2553 dentry = file->f_path.dentry; 2554 } 2555 if (create_error && dentry->d_inode == NULL) { 2556 error = create_error; 2557 goto out; 2558 } 2559 goto looked_up; 2560 } 2561 2562 /* 2563 * We didn't have the inode before the open, so check open permission 2564 * here. 2565 */ 2566 error = may_open(&file->f_path, acc_mode, open_flag); 2567 if (error) 2568 fput(file); 2569 2570 out: 2571 dput(dentry); 2572 return error; 2573 2574 no_open: 2575 if (need_lookup) { 2576 dentry = lookup_real(dir, dentry, nd->flags); 2577 if (IS_ERR(dentry)) 2578 return PTR_ERR(dentry); 2579 2580 if (create_error) { 2581 int open_flag = op->open_flag; 2582 2583 error = create_error; 2584 if ((open_flag & O_EXCL)) { 2585 if (!dentry->d_inode) 2586 goto out; 2587 } else if (!dentry->d_inode) { 2588 goto out; 2589 } else if ((open_flag & O_TRUNC) && 2590 S_ISREG(dentry->d_inode->i_mode)) { 2591 goto out; 2592 } 2593 /* will fail later, go on to get the right error */ 2594 } 2595 } 2596 looked_up: 2597 path->dentry = dentry; 2598 path->mnt = nd->path.mnt; 2599 return 1; 2600 } 2601 2602 /* 2603 * Look up and maybe create and open the last component. 2604 * 2605 * Must be called with i_mutex held on parent. 2606 * 2607 * Returns 0 if the file was successfully atomically created (if necessary) and 2608 * opened. In this case the file will be returned attached to @file. 2609 * 2610 * Returns 1 if the file was not completely opened at this time, though lookups 2611 * and creations will have been performed and the dentry returned in @path will 2612 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 2613 * specified then a negative dentry may be returned. 2614 * 2615 * An error code is returned otherwise. 2616 * 2617 * FILE_CREATE will be set in @*opened if the dentry was created and will be 2618 * cleared otherwise prior to returning. 2619 */ 2620 static int lookup_open(struct nameidata *nd, struct path *path, 2621 struct file *file, 2622 const struct open_flags *op, 2623 bool got_write, int *opened) 2624 { 2625 struct dentry *dir = nd->path.dentry; 2626 struct inode *dir_inode = dir->d_inode; 2627 struct dentry *dentry; 2628 int error; 2629 bool need_lookup; 2630 2631 *opened &= ~FILE_CREATED; 2632 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2633 if (IS_ERR(dentry)) 2634 return PTR_ERR(dentry); 2635 2636 /* Cached positive dentry: will open in f_op->open */ 2637 if (!need_lookup && dentry->d_inode) 2638 goto out_no_open; 2639 2640 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 2641 return atomic_open(nd, dentry, path, file, op, got_write, 2642 need_lookup, opened); 2643 } 2644 2645 if (need_lookup) { 2646 BUG_ON(dentry->d_inode); 2647 2648 dentry = lookup_real(dir_inode, dentry, nd->flags); 2649 if (IS_ERR(dentry)) 2650 return PTR_ERR(dentry); 2651 } 2652 2653 /* Negative dentry, just create the file */ 2654 if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2655 umode_t mode = op->mode; 2656 if (!IS_POSIXACL(dir->d_inode)) 2657 mode &= ~current_umask(); 2658 /* 2659 * This write is needed to ensure that a 2660 * rw->ro transition does not occur between 2661 * the time when the file is created and when 2662 * a permanent write count is taken through 2663 * the 'struct file' in finish_open(). 2664 */ 2665 if (!got_write) { 2666 error = -EROFS; 2667 goto out_dput; 2668 } 2669 *opened |= FILE_CREATED; 2670 error = security_path_mknod(&nd->path, dentry, mode, 0); 2671 if (error) 2672 goto out_dput; 2673 error = vfs_create(dir->d_inode, dentry, mode, 2674 nd->flags & LOOKUP_EXCL); 2675 if (error) 2676 goto out_dput; 2677 } 2678 out_no_open: 2679 path->dentry = dentry; 2680 path->mnt = nd->path.mnt; 2681 return 1; 2682 2683 out_dput: 2684 dput(dentry); 2685 return error; 2686 } 2687 2688 /* 2689 * Handle the last step of open() 2690 */ 2691 static int do_last(struct nameidata *nd, struct path *path, 2692 struct file *file, const struct open_flags *op, 2693 int *opened, struct filename *name) 2694 { 2695 struct dentry *dir = nd->path.dentry; 2696 int open_flag = op->open_flag; 2697 bool will_truncate = (open_flag & O_TRUNC) != 0; 2698 bool got_write = false; 2699 int acc_mode = op->acc_mode; 2700 struct inode *inode; 2701 bool symlink_ok = false; 2702 struct path save_parent = { .dentry = NULL, .mnt = NULL }; 2703 bool retried = false; 2704 int error; 2705 2706 nd->flags &= ~LOOKUP_PARENT; 2707 nd->flags |= op->intent; 2708 2709 switch (nd->last_type) { 2710 case LAST_DOTDOT: 2711 case LAST_DOT: 2712 error = handle_dots(nd, nd->last_type); 2713 if (error) 2714 return error; 2715 /* fallthrough */ 2716 case LAST_ROOT: 2717 error = complete_walk(nd); 2718 if (error) 2719 return error; 2720 audit_inode(name, nd->path.dentry, 0); 2721 if (open_flag & O_CREAT) { 2722 error = -EISDIR; 2723 goto out; 2724 } 2725 goto finish_open; 2726 case LAST_BIND: 2727 error = complete_walk(nd); 2728 if (error) 2729 return error; 2730 audit_inode(name, dir, 0); 2731 goto finish_open; 2732 } 2733 2734 if (!(open_flag & O_CREAT)) { 2735 if (nd->last.name[nd->last.len]) 2736 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2737 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 2738 symlink_ok = true; 2739 /* we _can_ be in RCU mode here */ 2740 error = lookup_fast(nd, &nd->last, path, &inode); 2741 if (likely(!error)) 2742 goto finish_lookup; 2743 2744 if (error < 0) 2745 goto out; 2746 2747 BUG_ON(nd->inode != dir->d_inode); 2748 } else { 2749 /* create side of things */ 2750 /* 2751 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2752 * has been cleared when we got to the last component we are 2753 * about to look up 2754 */ 2755 error = complete_walk(nd); 2756 if (error) 2757 return error; 2758 2759 audit_inode(name, dir, 0); 2760 error = -EISDIR; 2761 /* trailing slashes? */ 2762 if (nd->last.name[nd->last.len]) 2763 goto out; 2764 } 2765 2766 retry_lookup: 2767 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 2768 error = mnt_want_write(nd->path.mnt); 2769 if (!error) 2770 got_write = true; 2771 /* 2772 * do _not_ fail yet - we might not need that or fail with 2773 * a different error; let lookup_open() decide; we'll be 2774 * dropping this one anyway. 2775 */ 2776 } 2777 mutex_lock(&dir->d_inode->i_mutex); 2778 error = lookup_open(nd, path, file, op, got_write, opened); 2779 mutex_unlock(&dir->d_inode->i_mutex); 2780 2781 if (error <= 0) { 2782 if (error) 2783 goto out; 2784 2785 if ((*opened & FILE_CREATED) || 2786 !S_ISREG(file->f_path.dentry->d_inode->i_mode)) 2787 will_truncate = false; 2788 2789 audit_inode(name, file->f_path.dentry, 0); 2790 goto opened; 2791 } 2792 2793 if (*opened & FILE_CREATED) { 2794 /* Don't check for write permission, don't truncate */ 2795 open_flag &= ~O_TRUNC; 2796 will_truncate = false; 2797 acc_mode = MAY_OPEN; 2798 path_to_nameidata(path, nd); 2799 goto finish_open_created; 2800 } 2801 2802 /* 2803 * create/update audit record if it already exists. 2804 */ 2805 if (path->dentry->d_inode) 2806 audit_inode(name, path->dentry, 0); 2807 2808 /* 2809 * If atomic_open() acquired write access it is dropped now due to 2810 * possible mount and symlink following (this might be optimized away if 2811 * necessary...) 2812 */ 2813 if (got_write) { 2814 mnt_drop_write(nd->path.mnt); 2815 got_write = false; 2816 } 2817 2818 error = -EEXIST; 2819 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) 2820 goto exit_dput; 2821 2822 error = follow_managed(path, nd->flags); 2823 if (error < 0) 2824 goto exit_dput; 2825 2826 if (error) 2827 nd->flags |= LOOKUP_JUMPED; 2828 2829 BUG_ON(nd->flags & LOOKUP_RCU); 2830 inode = path->dentry->d_inode; 2831 finish_lookup: 2832 /* we _can_ be in RCU mode here */ 2833 error = -ENOENT; 2834 if (!inode) { 2835 path_to_nameidata(path, nd); 2836 goto out; 2837 } 2838 2839 if (should_follow_link(inode, !symlink_ok)) { 2840 if (nd->flags & LOOKUP_RCU) { 2841 if (unlikely(unlazy_walk(nd, path->dentry))) { 2842 error = -ECHILD; 2843 goto out; 2844 } 2845 } 2846 BUG_ON(inode != path->dentry->d_inode); 2847 return 1; 2848 } 2849 2850 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { 2851 path_to_nameidata(path, nd); 2852 } else { 2853 save_parent.dentry = nd->path.dentry; 2854 save_parent.mnt = mntget(path->mnt); 2855 nd->path.dentry = path->dentry; 2856 2857 } 2858 nd->inode = inode; 2859 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 2860 error = complete_walk(nd); 2861 if (error) { 2862 path_put(&save_parent); 2863 return error; 2864 } 2865 error = -EISDIR; 2866 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode)) 2867 goto out; 2868 error = -ENOTDIR; 2869 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup) 2870 goto out; 2871 audit_inode(name, nd->path.dentry, 0); 2872 finish_open: 2873 if (!S_ISREG(nd->inode->i_mode)) 2874 will_truncate = false; 2875 2876 if (will_truncate) { 2877 error = mnt_want_write(nd->path.mnt); 2878 if (error) 2879 goto out; 2880 got_write = true; 2881 } 2882 finish_open_created: 2883 error = may_open(&nd->path, acc_mode, open_flag); 2884 if (error) 2885 goto out; 2886 file->f_path.mnt = nd->path.mnt; 2887 error = finish_open(file, nd->path.dentry, NULL, opened); 2888 if (error) { 2889 if (error == -EOPENSTALE) 2890 goto stale_open; 2891 goto out; 2892 } 2893 opened: 2894 error = open_check_o_direct(file); 2895 if (error) 2896 goto exit_fput; 2897 error = ima_file_check(file, op->acc_mode); 2898 if (error) 2899 goto exit_fput; 2900 2901 if (will_truncate) { 2902 error = handle_truncate(file); 2903 if (error) 2904 goto exit_fput; 2905 } 2906 out: 2907 if (got_write) 2908 mnt_drop_write(nd->path.mnt); 2909 path_put(&save_parent); 2910 terminate_walk(nd); 2911 return error; 2912 2913 exit_dput: 2914 path_put_conditional(path, nd); 2915 goto out; 2916 exit_fput: 2917 fput(file); 2918 goto out; 2919 2920 stale_open: 2921 /* If no saved parent or already retried then can't retry */ 2922 if (!save_parent.dentry || retried) 2923 goto out; 2924 2925 BUG_ON(save_parent.dentry != dir); 2926 path_put(&nd->path); 2927 nd->path = save_parent; 2928 nd->inode = dir->d_inode; 2929 save_parent.mnt = NULL; 2930 save_parent.dentry = NULL; 2931 if (got_write) { 2932 mnt_drop_write(nd->path.mnt); 2933 got_write = false; 2934 } 2935 retried = true; 2936 goto retry_lookup; 2937 } 2938 2939 static struct file *path_openat(int dfd, struct filename *pathname, 2940 struct nameidata *nd, const struct open_flags *op, int flags) 2941 { 2942 struct file *base = NULL; 2943 struct file *file; 2944 struct path path; 2945 int opened = 0; 2946 int error; 2947 2948 file = get_empty_filp(); 2949 if (!file) 2950 return ERR_PTR(-ENFILE); 2951 2952 file->f_flags = op->open_flag; 2953 2954 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); 2955 if (unlikely(error)) 2956 goto out; 2957 2958 current->total_link_count = 0; 2959 error = link_path_walk(pathname->name, nd); 2960 if (unlikely(error)) 2961 goto out; 2962 2963 error = do_last(nd, &path, file, op, &opened, pathname); 2964 while (unlikely(error > 0)) { /* trailing symlink */ 2965 struct path link = path; 2966 void *cookie; 2967 if (!(nd->flags & LOOKUP_FOLLOW)) { 2968 path_put_conditional(&path, nd); 2969 path_put(&nd->path); 2970 error = -ELOOP; 2971 break; 2972 } 2973 error = may_follow_link(&link, nd); 2974 if (unlikely(error)) 2975 break; 2976 nd->flags |= LOOKUP_PARENT; 2977 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 2978 error = follow_link(&link, nd, &cookie); 2979 if (unlikely(error)) 2980 break; 2981 error = do_last(nd, &path, file, op, &opened, pathname); 2982 put_link(nd, &link, cookie); 2983 } 2984 out: 2985 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) 2986 path_put(&nd->root); 2987 if (base) 2988 fput(base); 2989 if (!(opened & FILE_OPENED)) { 2990 BUG_ON(!error); 2991 put_filp(file); 2992 } 2993 if (unlikely(error)) { 2994 if (error == -EOPENSTALE) { 2995 if (flags & LOOKUP_RCU) 2996 error = -ECHILD; 2997 else 2998 error = -ESTALE; 2999 } 3000 file = ERR_PTR(error); 3001 } 3002 return file; 3003 } 3004 3005 struct file *do_filp_open(int dfd, struct filename *pathname, 3006 const struct open_flags *op, int flags) 3007 { 3008 struct nameidata nd; 3009 struct file *filp; 3010 3011 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 3012 if (unlikely(filp == ERR_PTR(-ECHILD))) 3013 filp = path_openat(dfd, pathname, &nd, op, flags); 3014 if (unlikely(filp == ERR_PTR(-ESTALE))) 3015 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 3016 return filp; 3017 } 3018 3019 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3020 const char *name, const struct open_flags *op, int flags) 3021 { 3022 struct nameidata nd; 3023 struct file *file; 3024 struct filename filename = { .name = name }; 3025 3026 nd.root.mnt = mnt; 3027 nd.root.dentry = dentry; 3028 3029 flags |= LOOKUP_ROOT; 3030 3031 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) 3032 return ERR_PTR(-ELOOP); 3033 3034 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); 3035 if (unlikely(file == ERR_PTR(-ECHILD))) 3036 file = path_openat(-1, &filename, &nd, op, flags); 3037 if (unlikely(file == ERR_PTR(-ESTALE))) 3038 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); 3039 return file; 3040 } 3041 3042 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir) 3043 { 3044 struct dentry *dentry = ERR_PTR(-EEXIST); 3045 struct nameidata nd; 3046 int err2; 3047 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd); 3048 if (error) 3049 return ERR_PTR(error); 3050 3051 /* 3052 * Yucky last component or no last component at all? 3053 * (foo/., foo/.., /////) 3054 */ 3055 if (nd.last_type != LAST_NORM) 3056 goto out; 3057 nd.flags &= ~LOOKUP_PARENT; 3058 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3059 3060 /* don't fail immediately if it's r/o, at least try to report other errors */ 3061 err2 = mnt_want_write(nd.path.mnt); 3062 /* 3063 * Do the final lookup. 3064 */ 3065 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3066 dentry = lookup_hash(&nd); 3067 if (IS_ERR(dentry)) 3068 goto unlock; 3069 3070 error = -EEXIST; 3071 if (dentry->d_inode) 3072 goto fail; 3073 /* 3074 * Special case - lookup gave negative, but... we had foo/bar/ 3075 * From the vfs_mknod() POV we just have a negative dentry - 3076 * all is fine. Let's be bastards - you had / on the end, you've 3077 * been asking for (non-existent) directory. -ENOENT for you. 3078 */ 3079 if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3080 error = -ENOENT; 3081 goto fail; 3082 } 3083 if (unlikely(err2)) { 3084 error = err2; 3085 goto fail; 3086 } 3087 *path = nd.path; 3088 return dentry; 3089 fail: 3090 dput(dentry); 3091 dentry = ERR_PTR(error); 3092 unlock: 3093 mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3094 if (!err2) 3095 mnt_drop_write(nd.path.mnt); 3096 out: 3097 path_put(&nd.path); 3098 return dentry; 3099 } 3100 EXPORT_SYMBOL(kern_path_create); 3101 3102 void done_path_create(struct path *path, struct dentry *dentry) 3103 { 3104 dput(dentry); 3105 mutex_unlock(&path->dentry->d_inode->i_mutex); 3106 mnt_drop_write(path->mnt); 3107 path_put(path); 3108 } 3109 EXPORT_SYMBOL(done_path_create); 3110 3111 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir) 3112 { 3113 struct filename *tmp = getname(pathname); 3114 struct dentry *res; 3115 if (IS_ERR(tmp)) 3116 return ERR_CAST(tmp); 3117 res = kern_path_create(dfd, tmp->name, path, is_dir); 3118 putname(tmp); 3119 return res; 3120 } 3121 EXPORT_SYMBOL(user_path_create); 3122 3123 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 3124 { 3125 int error = may_create(dir, dentry); 3126 3127 if (error) 3128 return error; 3129 3130 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 3131 return -EPERM; 3132 3133 if (!dir->i_op->mknod) 3134 return -EPERM; 3135 3136 error = devcgroup_inode_mknod(mode, dev); 3137 if (error) 3138 return error; 3139 3140 error = security_inode_mknod(dir, dentry, mode, dev); 3141 if (error) 3142 return error; 3143 3144 error = dir->i_op->mknod(dir, dentry, mode, dev); 3145 if (!error) 3146 fsnotify_create(dir, dentry); 3147 return error; 3148 } 3149 3150 static int may_mknod(umode_t mode) 3151 { 3152 switch (mode & S_IFMT) { 3153 case S_IFREG: 3154 case S_IFCHR: 3155 case S_IFBLK: 3156 case S_IFIFO: 3157 case S_IFSOCK: 3158 case 0: /* zero mode translates to S_IFREG */ 3159 return 0; 3160 case S_IFDIR: 3161 return -EPERM; 3162 default: 3163 return -EINVAL; 3164 } 3165 } 3166 3167 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 3168 unsigned, dev) 3169 { 3170 struct dentry *dentry; 3171 struct path path; 3172 int error; 3173 3174 error = may_mknod(mode); 3175 if (error) 3176 return error; 3177 3178 dentry = user_path_create(dfd, filename, &path, 0); 3179 if (IS_ERR(dentry)) 3180 return PTR_ERR(dentry); 3181 3182 if (!IS_POSIXACL(path.dentry->d_inode)) 3183 mode &= ~current_umask(); 3184 error = security_path_mknod(&path, dentry, mode, dev); 3185 if (error) 3186 goto out; 3187 switch (mode & S_IFMT) { 3188 case 0: case S_IFREG: 3189 error = vfs_create(path.dentry->d_inode,dentry,mode,true); 3190 break; 3191 case S_IFCHR: case S_IFBLK: 3192 error = vfs_mknod(path.dentry->d_inode,dentry,mode, 3193 new_decode_dev(dev)); 3194 break; 3195 case S_IFIFO: case S_IFSOCK: 3196 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 3197 break; 3198 } 3199 out: 3200 done_path_create(&path, dentry); 3201 return error; 3202 } 3203 3204 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 3205 { 3206 return sys_mknodat(AT_FDCWD, filename, mode, dev); 3207 } 3208 3209 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 3210 { 3211 int error = may_create(dir, dentry); 3212 unsigned max_links = dir->i_sb->s_max_links; 3213 3214 if (error) 3215 return error; 3216 3217 if (!dir->i_op->mkdir) 3218 return -EPERM; 3219 3220 mode &= (S_IRWXUGO|S_ISVTX); 3221 error = security_inode_mkdir(dir, dentry, mode); 3222 if (error) 3223 return error; 3224 3225 if (max_links && dir->i_nlink >= max_links) 3226 return -EMLINK; 3227 3228 error = dir->i_op->mkdir(dir, dentry, mode); 3229 if (!error) 3230 fsnotify_mkdir(dir, dentry); 3231 return error; 3232 } 3233 3234 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 3235 { 3236 struct dentry *dentry; 3237 struct path path; 3238 int error; 3239 3240 dentry = user_path_create(dfd, pathname, &path, 1); 3241 if (IS_ERR(dentry)) 3242 return PTR_ERR(dentry); 3243 3244 if (!IS_POSIXACL(path.dentry->d_inode)) 3245 mode &= ~current_umask(); 3246 error = security_path_mkdir(&path, dentry, mode); 3247 if (!error) 3248 error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3249 done_path_create(&path, dentry); 3250 return error; 3251 } 3252 3253 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 3254 { 3255 return sys_mkdirat(AT_FDCWD, pathname, mode); 3256 } 3257 3258 /* 3259 * The dentry_unhash() helper will try to drop the dentry early: we 3260 * should have a usage count of 1 if we're the only user of this 3261 * dentry, and if that is true (possibly after pruning the dcache), 3262 * then we drop the dentry now. 3263 * 3264 * A low-level filesystem can, if it choses, legally 3265 * do a 3266 * 3267 * if (!d_unhashed(dentry)) 3268 * return -EBUSY; 3269 * 3270 * if it cannot handle the case of removing a directory 3271 * that is still in use by something else.. 3272 */ 3273 void dentry_unhash(struct dentry *dentry) 3274 { 3275 shrink_dcache_parent(dentry); 3276 spin_lock(&dentry->d_lock); 3277 if (dentry->d_count == 1) 3278 __d_drop(dentry); 3279 spin_unlock(&dentry->d_lock); 3280 } 3281 3282 int vfs_rmdir(struct inode *dir, struct dentry *dentry) 3283 { 3284 int error = may_delete(dir, dentry, 1); 3285 3286 if (error) 3287 return error; 3288 3289 if (!dir->i_op->rmdir) 3290 return -EPERM; 3291 3292 dget(dentry); 3293 mutex_lock(&dentry->d_inode->i_mutex); 3294 3295 error = -EBUSY; 3296 if (d_mountpoint(dentry)) 3297 goto out; 3298 3299 error = security_inode_rmdir(dir, dentry); 3300 if (error) 3301 goto out; 3302 3303 shrink_dcache_parent(dentry); 3304 error = dir->i_op->rmdir(dir, dentry); 3305 if (error) 3306 goto out; 3307 3308 dentry->d_inode->i_flags |= S_DEAD; 3309 dont_mount(dentry); 3310 3311 out: 3312 mutex_unlock(&dentry->d_inode->i_mutex); 3313 dput(dentry); 3314 if (!error) 3315 d_delete(dentry); 3316 return error; 3317 } 3318 3319 static long do_rmdir(int dfd, const char __user *pathname) 3320 { 3321 int error = 0; 3322 struct filename *name; 3323 struct dentry *dentry; 3324 struct nameidata nd; 3325 3326 name = user_path_parent(dfd, pathname, &nd); 3327 if (IS_ERR(name)) 3328 return PTR_ERR(name); 3329 3330 switch(nd.last_type) { 3331 case LAST_DOTDOT: 3332 error = -ENOTEMPTY; 3333 goto exit1; 3334 case LAST_DOT: 3335 error = -EINVAL; 3336 goto exit1; 3337 case LAST_ROOT: 3338 error = -EBUSY; 3339 goto exit1; 3340 } 3341 3342 nd.flags &= ~LOOKUP_PARENT; 3343 error = mnt_want_write(nd.path.mnt); 3344 if (error) 3345 goto exit1; 3346 3347 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3348 dentry = lookup_hash(&nd); 3349 error = PTR_ERR(dentry); 3350 if (IS_ERR(dentry)) 3351 goto exit2; 3352 if (!dentry->d_inode) { 3353 error = -ENOENT; 3354 goto exit3; 3355 } 3356 error = security_path_rmdir(&nd.path, dentry); 3357 if (error) 3358 goto exit3; 3359 error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 3360 exit3: 3361 dput(dentry); 3362 exit2: 3363 mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3364 mnt_drop_write(nd.path.mnt); 3365 exit1: 3366 path_put(&nd.path); 3367 putname(name); 3368 return error; 3369 } 3370 3371 SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 3372 { 3373 return do_rmdir(AT_FDCWD, pathname); 3374 } 3375 3376 int vfs_unlink(struct inode *dir, struct dentry *dentry) 3377 { 3378 int error = may_delete(dir, dentry, 0); 3379 3380 if (error) 3381 return error; 3382 3383 if (!dir->i_op->unlink) 3384 return -EPERM; 3385 3386 mutex_lock(&dentry->d_inode->i_mutex); 3387 if (d_mountpoint(dentry)) 3388 error = -EBUSY; 3389 else { 3390 error = security_inode_unlink(dir, dentry); 3391 if (!error) { 3392 error = dir->i_op->unlink(dir, dentry); 3393 if (!error) 3394 dont_mount(dentry); 3395 } 3396 } 3397 mutex_unlock(&dentry->d_inode->i_mutex); 3398 3399 /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 3400 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 3401 fsnotify_link_count(dentry->d_inode); 3402 d_delete(dentry); 3403 } 3404 3405 return error; 3406 } 3407 3408 /* 3409 * Make sure that the actual truncation of the file will occur outside its 3410 * directory's i_mutex. Truncate can take a long time if there is a lot of 3411 * writeout happening, and we don't want to prevent access to the directory 3412 * while waiting on the I/O. 3413 */ 3414 static long do_unlinkat(int dfd, const char __user *pathname) 3415 { 3416 int error; 3417 struct filename *name; 3418 struct dentry *dentry; 3419 struct nameidata nd; 3420 struct inode *inode = NULL; 3421 3422 name = user_path_parent(dfd, pathname, &nd); 3423 if (IS_ERR(name)) 3424 return PTR_ERR(name); 3425 3426 error = -EISDIR; 3427 if (nd.last_type != LAST_NORM) 3428 goto exit1; 3429 3430 nd.flags &= ~LOOKUP_PARENT; 3431 error = mnt_want_write(nd.path.mnt); 3432 if (error) 3433 goto exit1; 3434 3435 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3436 dentry = lookup_hash(&nd); 3437 error = PTR_ERR(dentry); 3438 if (!IS_ERR(dentry)) { 3439 /* Why not before? Because we want correct error value */ 3440 if (nd.last.name[nd.last.len]) 3441 goto slashes; 3442 inode = dentry->d_inode; 3443 if (!inode) 3444 goto slashes; 3445 ihold(inode); 3446 error = security_path_unlink(&nd.path, dentry); 3447 if (error) 3448 goto exit2; 3449 error = vfs_unlink(nd.path.dentry->d_inode, dentry); 3450 exit2: 3451 dput(dentry); 3452 } 3453 mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3454 if (inode) 3455 iput(inode); /* truncate the inode here */ 3456 mnt_drop_write(nd.path.mnt); 3457 exit1: 3458 path_put(&nd.path); 3459 putname(name); 3460 return error; 3461 3462 slashes: 3463 error = !dentry->d_inode ? -ENOENT : 3464 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 3465 goto exit2; 3466 } 3467 3468 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 3469 { 3470 if ((flag & ~AT_REMOVEDIR) != 0) 3471 return -EINVAL; 3472 3473 if (flag & AT_REMOVEDIR) 3474 return do_rmdir(dfd, pathname); 3475 3476 return do_unlinkat(dfd, pathname); 3477 } 3478 3479 SYSCALL_DEFINE1(unlink, const char __user *, pathname) 3480 { 3481 return do_unlinkat(AT_FDCWD, pathname); 3482 } 3483 3484 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 3485 { 3486 int error = may_create(dir, dentry); 3487 3488 if (error) 3489 return error; 3490 3491 if (!dir->i_op->symlink) 3492 return -EPERM; 3493 3494 error = security_inode_symlink(dir, dentry, oldname); 3495 if (error) 3496 return error; 3497 3498 error = dir->i_op->symlink(dir, dentry, oldname); 3499 if (!error) 3500 fsnotify_create(dir, dentry); 3501 return error; 3502 } 3503 3504 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 3505 int, newdfd, const char __user *, newname) 3506 { 3507 int error; 3508 struct filename *from; 3509 struct dentry *dentry; 3510 struct path path; 3511 3512 from = getname(oldname); 3513 if (IS_ERR(from)) 3514 return PTR_ERR(from); 3515 3516 dentry = user_path_create(newdfd, newname, &path, 0); 3517 error = PTR_ERR(dentry); 3518 if (IS_ERR(dentry)) 3519 goto out_putname; 3520 3521 error = security_path_symlink(&path, dentry, from->name); 3522 if (!error) 3523 error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3524 done_path_create(&path, dentry); 3525 out_putname: 3526 putname(from); 3527 return error; 3528 } 3529 3530 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 3531 { 3532 return sys_symlinkat(oldname, AT_FDCWD, newname); 3533 } 3534 3535 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 3536 { 3537 struct inode *inode = old_dentry->d_inode; 3538 unsigned max_links = dir->i_sb->s_max_links; 3539 int error; 3540 3541 if (!inode) 3542 return -ENOENT; 3543 3544 error = may_create(dir, new_dentry); 3545 if (error) 3546 return error; 3547 3548 if (dir->i_sb != inode->i_sb) 3549 return -EXDEV; 3550 3551 /* 3552 * A link to an append-only or immutable file cannot be created. 3553 */ 3554 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 3555 return -EPERM; 3556 if (!dir->i_op->link) 3557 return -EPERM; 3558 if (S_ISDIR(inode->i_mode)) 3559 return -EPERM; 3560 3561 error = security_inode_link(old_dentry, dir, new_dentry); 3562 if (error) 3563 return error; 3564 3565 mutex_lock(&inode->i_mutex); 3566 /* Make sure we don't allow creating hardlink to an unlinked file */ 3567 if (inode->i_nlink == 0) 3568 error = -ENOENT; 3569 else if (max_links && inode->i_nlink >= max_links) 3570 error = -EMLINK; 3571 else 3572 error = dir->i_op->link(old_dentry, dir, new_dentry); 3573 mutex_unlock(&inode->i_mutex); 3574 if (!error) 3575 fsnotify_link(dir, inode, new_dentry); 3576 return error; 3577 } 3578 3579 /* 3580 * Hardlinks are often used in delicate situations. We avoid 3581 * security-related surprises by not following symlinks on the 3582 * newname. --KAB 3583 * 3584 * We don't follow them on the oldname either to be compatible 3585 * with linux 2.0, and to avoid hard-linking to directories 3586 * and other special files. --ADM 3587 */ 3588 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 3589 int, newdfd, const char __user *, newname, int, flags) 3590 { 3591 struct dentry *new_dentry; 3592 struct path old_path, new_path; 3593 int how = 0; 3594 int error; 3595 3596 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 3597 return -EINVAL; 3598 /* 3599 * To use null names we require CAP_DAC_READ_SEARCH 3600 * This ensures that not everyone will be able to create 3601 * handlink using the passed filedescriptor. 3602 */ 3603 if (flags & AT_EMPTY_PATH) { 3604 if (!capable(CAP_DAC_READ_SEARCH)) 3605 return -ENOENT; 3606 how = LOOKUP_EMPTY; 3607 } 3608 3609 if (flags & AT_SYMLINK_FOLLOW) 3610 how |= LOOKUP_FOLLOW; 3611 3612 error = user_path_at(olddfd, oldname, how, &old_path); 3613 if (error) 3614 return error; 3615 3616 new_dentry = user_path_create(newdfd, newname, &new_path, 0); 3617 error = PTR_ERR(new_dentry); 3618 if (IS_ERR(new_dentry)) 3619 goto out; 3620 3621 error = -EXDEV; 3622 if (old_path.mnt != new_path.mnt) 3623 goto out_dput; 3624 error = may_linkat(&old_path); 3625 if (unlikely(error)) 3626 goto out_dput; 3627 error = security_path_link(old_path.dentry, &new_path, new_dentry); 3628 if (error) 3629 goto out_dput; 3630 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); 3631 out_dput: 3632 done_path_create(&new_path, new_dentry); 3633 out: 3634 path_put(&old_path); 3635 3636 return error; 3637 } 3638 3639 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 3640 { 3641 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 3642 } 3643 3644 /* 3645 * The worst of all namespace operations - renaming directory. "Perverted" 3646 * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 3647 * Problems: 3648 * a) we can get into loop creation. Check is done in is_subdir(). 3649 * b) race potential - two innocent renames can create a loop together. 3650 * That's where 4.4 screws up. Current fix: serialization on 3651 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 3652 * story. 3653 * c) we have to lock _three_ objects - parents and victim (if it exists). 3654 * And that - after we got ->i_mutex on parents (until then we don't know 3655 * whether the target exists). Solution: try to be smart with locking 3656 * order for inodes. We rely on the fact that tree topology may change 3657 * only under ->s_vfs_rename_mutex _and_ that parent of the object we 3658 * move will be locked. Thus we can rank directories by the tree 3659 * (ancestors first) and rank all non-directories after them. 3660 * That works since everybody except rename does "lock parent, lookup, 3661 * lock child" and rename is under ->s_vfs_rename_mutex. 3662 * HOWEVER, it relies on the assumption that any object with ->lookup() 3663 * has no more than 1 dentry. If "hybrid" objects will ever appear, 3664 * we'd better make sure that there's no link(2) for them. 3665 * d) conversion from fhandle to dentry may come in the wrong moment - when 3666 * we are removing the target. Solution: we will have to grab ->i_mutex 3667 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 3668 * ->i_mutex on parents, which works but leads to some truly excessive 3669 * locking]. 3670 */ 3671 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 3672 struct inode *new_dir, struct dentry *new_dentry) 3673 { 3674 int error = 0; 3675 struct inode *target = new_dentry->d_inode; 3676 unsigned max_links = new_dir->i_sb->s_max_links; 3677 3678 /* 3679 * If we are going to change the parent - check write permissions, 3680 * we'll need to flip '..'. 3681 */ 3682 if (new_dir != old_dir) { 3683 error = inode_permission(old_dentry->d_inode, MAY_WRITE); 3684 if (error) 3685 return error; 3686 } 3687 3688 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 3689 if (error) 3690 return error; 3691 3692 dget(new_dentry); 3693 if (target) 3694 mutex_lock(&target->i_mutex); 3695 3696 error = -EBUSY; 3697 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) 3698 goto out; 3699 3700 error = -EMLINK; 3701 if (max_links && !target && new_dir != old_dir && 3702 new_dir->i_nlink >= max_links) 3703 goto out; 3704 3705 if (target) 3706 shrink_dcache_parent(new_dentry); 3707 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 3708 if (error) 3709 goto out; 3710 3711 if (target) { 3712 target->i_flags |= S_DEAD; 3713 dont_mount(new_dentry); 3714 } 3715 out: 3716 if (target) 3717 mutex_unlock(&target->i_mutex); 3718 dput(new_dentry); 3719 if (!error) 3720 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 3721 d_move(old_dentry,new_dentry); 3722 return error; 3723 } 3724 3725 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 3726 struct inode *new_dir, struct dentry *new_dentry) 3727 { 3728 struct inode *target = new_dentry->d_inode; 3729 int error; 3730 3731 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 3732 if (error) 3733 return error; 3734 3735 dget(new_dentry); 3736 if (target) 3737 mutex_lock(&target->i_mutex); 3738 3739 error = -EBUSY; 3740 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 3741 goto out; 3742 3743 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 3744 if (error) 3745 goto out; 3746 3747 if (target) 3748 dont_mount(new_dentry); 3749 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 3750 d_move(old_dentry, new_dentry); 3751 out: 3752 if (target) 3753 mutex_unlock(&target->i_mutex); 3754 dput(new_dentry); 3755 return error; 3756 } 3757 3758 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 3759 struct inode *new_dir, struct dentry *new_dentry) 3760 { 3761 int error; 3762 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 3763 const unsigned char *old_name; 3764 3765 if (old_dentry->d_inode == new_dentry->d_inode) 3766 return 0; 3767 3768 error = may_delete(old_dir, old_dentry, is_dir); 3769 if (error) 3770 return error; 3771 3772 if (!new_dentry->d_inode) 3773 error = may_create(new_dir, new_dentry); 3774 else 3775 error = may_delete(new_dir, new_dentry, is_dir); 3776 if (error) 3777 return error; 3778 3779 if (!old_dir->i_op->rename) 3780 return -EPERM; 3781 3782 old_name = fsnotify_oldname_init(old_dentry->d_name.name); 3783 3784 if (is_dir) 3785 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 3786 else 3787 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 3788 if (!error) 3789 fsnotify_move(old_dir, new_dir, old_name, is_dir, 3790 new_dentry->d_inode, old_dentry); 3791 fsnotify_oldname_free(old_name); 3792 3793 return error; 3794 } 3795 3796 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 3797 int, newdfd, const char __user *, newname) 3798 { 3799 struct dentry *old_dir, *new_dir; 3800 struct dentry *old_dentry, *new_dentry; 3801 struct dentry *trap; 3802 struct nameidata oldnd, newnd; 3803 struct filename *from; 3804 struct filename *to; 3805 int error; 3806 3807 from = user_path_parent(olddfd, oldname, &oldnd); 3808 if (IS_ERR(from)) { 3809 error = PTR_ERR(from); 3810 goto exit; 3811 } 3812 3813 to = user_path_parent(newdfd, newname, &newnd); 3814 if (IS_ERR(to)) { 3815 error = PTR_ERR(to); 3816 goto exit1; 3817 } 3818 3819 error = -EXDEV; 3820 if (oldnd.path.mnt != newnd.path.mnt) 3821 goto exit2; 3822 3823 old_dir = oldnd.path.dentry; 3824 error = -EBUSY; 3825 if (oldnd.last_type != LAST_NORM) 3826 goto exit2; 3827 3828 new_dir = newnd.path.dentry; 3829 if (newnd.last_type != LAST_NORM) 3830 goto exit2; 3831 3832 error = mnt_want_write(oldnd.path.mnt); 3833 if (error) 3834 goto exit2; 3835 3836 oldnd.flags &= ~LOOKUP_PARENT; 3837 newnd.flags &= ~LOOKUP_PARENT; 3838 newnd.flags |= LOOKUP_RENAME_TARGET; 3839 3840 trap = lock_rename(new_dir, old_dir); 3841 3842 old_dentry = lookup_hash(&oldnd); 3843 error = PTR_ERR(old_dentry); 3844 if (IS_ERR(old_dentry)) 3845 goto exit3; 3846 /* source must exist */ 3847 error = -ENOENT; 3848 if (!old_dentry->d_inode) 3849 goto exit4; 3850 /* unless the source is a directory trailing slashes give -ENOTDIR */ 3851 if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 3852 error = -ENOTDIR; 3853 if (oldnd.last.name[oldnd.last.len]) 3854 goto exit4; 3855 if (newnd.last.name[newnd.last.len]) 3856 goto exit4; 3857 } 3858 /* source should not be ancestor of target */ 3859 error = -EINVAL; 3860 if (old_dentry == trap) 3861 goto exit4; 3862 new_dentry = lookup_hash(&newnd); 3863 error = PTR_ERR(new_dentry); 3864 if (IS_ERR(new_dentry)) 3865 goto exit4; 3866 /* target should not be an ancestor of source */ 3867 error = -ENOTEMPTY; 3868 if (new_dentry == trap) 3869 goto exit5; 3870 3871 error = security_path_rename(&oldnd.path, old_dentry, 3872 &newnd.path, new_dentry); 3873 if (error) 3874 goto exit5; 3875 error = vfs_rename(old_dir->d_inode, old_dentry, 3876 new_dir->d_inode, new_dentry); 3877 exit5: 3878 dput(new_dentry); 3879 exit4: 3880 dput(old_dentry); 3881 exit3: 3882 unlock_rename(new_dir, old_dir); 3883 mnt_drop_write(oldnd.path.mnt); 3884 exit2: 3885 path_put(&newnd.path); 3886 putname(to); 3887 exit1: 3888 path_put(&oldnd.path); 3889 putname(from); 3890 exit: 3891 return error; 3892 } 3893 3894 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 3895 { 3896 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 3897 } 3898 3899 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 3900 { 3901 int len; 3902 3903 len = PTR_ERR(link); 3904 if (IS_ERR(link)) 3905 goto out; 3906 3907 len = strlen(link); 3908 if (len > (unsigned) buflen) 3909 len = buflen; 3910 if (copy_to_user(buffer, link, len)) 3911 len = -EFAULT; 3912 out: 3913 return len; 3914 } 3915 3916 /* 3917 * A helper for ->readlink(). This should be used *ONLY* for symlinks that 3918 * have ->follow_link() touching nd only in nd_set_link(). Using (or not 3919 * using) it for any given inode is up to filesystem. 3920 */ 3921 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 3922 { 3923 struct nameidata nd; 3924 void *cookie; 3925 int res; 3926 3927 nd.depth = 0; 3928 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 3929 if (IS_ERR(cookie)) 3930 return PTR_ERR(cookie); 3931 3932 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 3933 if (dentry->d_inode->i_op->put_link) 3934 dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 3935 return res; 3936 } 3937 3938 int vfs_follow_link(struct nameidata *nd, const char *link) 3939 { 3940 return __vfs_follow_link(nd, link); 3941 } 3942 3943 /* get the link contents into pagecache */ 3944 static char *page_getlink(struct dentry * dentry, struct page **ppage) 3945 { 3946 char *kaddr; 3947 struct page *page; 3948 struct address_space *mapping = dentry->d_inode->i_mapping; 3949 page = read_mapping_page(mapping, 0, NULL); 3950 if (IS_ERR(page)) 3951 return (char*)page; 3952 *ppage = page; 3953 kaddr = kmap(page); 3954 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 3955 return kaddr; 3956 } 3957 3958 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 3959 { 3960 struct page *page = NULL; 3961 char *s = page_getlink(dentry, &page); 3962 int res = vfs_readlink(dentry,buffer,buflen,s); 3963 if (page) { 3964 kunmap(page); 3965 page_cache_release(page); 3966 } 3967 return res; 3968 } 3969 3970 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 3971 { 3972 struct page *page = NULL; 3973 nd_set_link(nd, page_getlink(dentry, &page)); 3974 return page; 3975 } 3976 3977 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 3978 { 3979 struct page *page = cookie; 3980 3981 if (page) { 3982 kunmap(page); 3983 page_cache_release(page); 3984 } 3985 } 3986 3987 /* 3988 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 3989 */ 3990 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 3991 { 3992 struct address_space *mapping = inode->i_mapping; 3993 struct page *page; 3994 void *fsdata; 3995 int err; 3996 char *kaddr; 3997 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 3998 if (nofs) 3999 flags |= AOP_FLAG_NOFS; 4000 4001 retry: 4002 err = pagecache_write_begin(NULL, mapping, 0, len-1, 4003 flags, &page, &fsdata); 4004 if (err) 4005 goto fail; 4006 4007 kaddr = kmap_atomic(page); 4008 memcpy(kaddr, symname, len-1); 4009 kunmap_atomic(kaddr); 4010 4011 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4012 page, fsdata); 4013 if (err < 0) 4014 goto fail; 4015 if (err < len-1) 4016 goto retry; 4017 4018 mark_inode_dirty(inode); 4019 return 0; 4020 fail: 4021 return err; 4022 } 4023 4024 int page_symlink(struct inode *inode, const char *symname, int len) 4025 { 4026 return __page_symlink(inode, symname, len, 4027 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 4028 } 4029 4030 const struct inode_operations page_symlink_inode_operations = { 4031 .readlink = generic_readlink, 4032 .follow_link = page_follow_link_light, 4033 .put_link = page_put_link, 4034 }; 4035 4036 EXPORT_SYMBOL(user_path_at); 4037 EXPORT_SYMBOL(follow_down_one); 4038 EXPORT_SYMBOL(follow_down); 4039 EXPORT_SYMBOL(follow_up); 4040 EXPORT_SYMBOL(get_write_access); /* nfsd */ 4041 EXPORT_SYMBOL(lock_rename); 4042 EXPORT_SYMBOL(lookup_one_len); 4043 EXPORT_SYMBOL(page_follow_link_light); 4044 EXPORT_SYMBOL(page_put_link); 4045 EXPORT_SYMBOL(page_readlink); 4046 EXPORT_SYMBOL(__page_symlink); 4047 EXPORT_SYMBOL(page_symlink); 4048 EXPORT_SYMBOL(page_symlink_inode_operations); 4049 EXPORT_SYMBOL(kern_path); 4050 EXPORT_SYMBOL(vfs_path_lookup); 4051 EXPORT_SYMBOL(inode_permission); 4052 EXPORT_SYMBOL(unlock_rename); 4053 EXPORT_SYMBOL(vfs_create); 4054 EXPORT_SYMBOL(vfs_follow_link); 4055 EXPORT_SYMBOL(vfs_link); 4056 EXPORT_SYMBOL(vfs_mkdir); 4057 EXPORT_SYMBOL(vfs_mknod); 4058 EXPORT_SYMBOL(generic_permission); 4059 EXPORT_SYMBOL(vfs_readlink); 4060 EXPORT_SYMBOL(vfs_rename); 4061 EXPORT_SYMBOL(vfs_rmdir); 4062 EXPORT_SYMBOL(vfs_symlink); 4063 EXPORT_SYMBOL(vfs_unlink); 4064 EXPORT_SYMBOL(dentry_unhash); 4065 EXPORT_SYMBOL(generic_readlink); 4066