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/module.h> 19 #include <linux/slab.h> 20 #include <linux/fs.h> 21 #include <linux/namei.h> 22 #include <linux/quotaops.h> 23 #include <linux/pagemap.h> 24 #include <linux/fsnotify.h> 25 #include <linux/personality.h> 26 #include <linux/security.h> 27 #include <linux/syscalls.h> 28 #include <linux/mount.h> 29 #include <linux/audit.h> 30 #include <linux/capability.h> 31 #include <linux/file.h> 32 #include <linux/fcntl.h> 33 #include <asm/namei.h> 34 #include <asm/uaccess.h> 35 36 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE]) 37 38 /* [Feb-1997 T. Schoebel-Theuer] 39 * Fundamental changes in the pathname lookup mechanisms (namei) 40 * were necessary because of omirr. The reason is that omirr needs 41 * to know the _real_ pathname, not the user-supplied one, in case 42 * of symlinks (and also when transname replacements occur). 43 * 44 * The new code replaces the old recursive symlink resolution with 45 * an iterative one (in case of non-nested symlink chains). It does 46 * this with calls to <fs>_follow_link(). 47 * As a side effect, dir_namei(), _namei() and follow_link() are now 48 * replaced with a single function lookup_dentry() that can handle all 49 * the special cases of the former code. 50 * 51 * With the new dcache, the pathname is stored at each inode, at least as 52 * long as the refcount of the inode is positive. As a side effect, the 53 * size of the dcache depends on the inode cache and thus is dynamic. 54 * 55 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 56 * resolution to correspond with current state of the code. 57 * 58 * Note that the symlink resolution is not *completely* iterative. 59 * There is still a significant amount of tail- and mid- recursion in 60 * the algorithm. Also, note that <fs>_readlink() is not used in 61 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 62 * may return different results than <fs>_follow_link(). Many virtual 63 * filesystems (including /proc) exhibit this behavior. 64 */ 65 66 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 67 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 68 * and the name already exists in form of a symlink, try to create the new 69 * name indicated by the symlink. The old code always complained that the 70 * name already exists, due to not following the symlink even if its target 71 * is nonexistent. The new semantics affects also mknod() and link() when 72 * the name is a symlink pointing to a non-existant name. 73 * 74 * I don't know which semantics is the right one, since I have no access 75 * to standards. But I found by trial that HP-UX 9.0 has the full "new" 76 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 77 * "old" one. Personally, I think the new semantics is much more logical. 78 * Note that "ln old new" where "new" is a symlink pointing to a non-existing 79 * file does succeed in both HP-UX and SunOs, but not in Solaris 80 * and in the old Linux semantics. 81 */ 82 83 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 84 * semantics. See the comments in "open_namei" and "do_link" below. 85 * 86 * [10-Sep-98 Alan Modra] Another symlink change. 87 */ 88 89 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 90 * inside the path - always follow. 91 * in the last component in creation/removal/renaming - never follow. 92 * if LOOKUP_FOLLOW passed - follow. 93 * if the pathname has trailing slashes - follow. 94 * otherwise - don't follow. 95 * (applied in that order). 96 * 97 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 98 * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 99 * During the 2.4 we need to fix the userland stuff depending on it - 100 * hopefully we will be able to get rid of that wart in 2.5. So far only 101 * XEmacs seems to be relying on it... 102 */ 103 /* 104 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 105 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 106 * any extra contention... 107 */ 108 109 static int fastcall link_path_walk(const char *name, struct nameidata *nd); 110 111 /* In order to reduce some races, while at the same time doing additional 112 * checking and hopefully speeding things up, we copy filenames to the 113 * kernel data space before using them.. 114 * 115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 116 * PATH_MAX includes the nul terminator --RR. 117 */ 118 static int do_getname(const char __user *filename, char *page) 119 { 120 int retval; 121 unsigned long len = PATH_MAX; 122 123 if (!segment_eq(get_fs(), KERNEL_DS)) { 124 if ((unsigned long) filename >= TASK_SIZE) 125 return -EFAULT; 126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX) 127 len = TASK_SIZE - (unsigned long) filename; 128 } 129 130 retval = strncpy_from_user(page, filename, len); 131 if (retval > 0) { 132 if (retval < len) 133 return 0; 134 return -ENAMETOOLONG; 135 } else if (!retval) 136 retval = -ENOENT; 137 return retval; 138 } 139 140 char * getname(const char __user * filename) 141 { 142 char *tmp, *result; 143 144 result = ERR_PTR(-ENOMEM); 145 tmp = __getname(); 146 if (tmp) { 147 int retval = do_getname(filename, tmp); 148 149 result = tmp; 150 if (retval < 0) { 151 __putname(tmp); 152 result = ERR_PTR(retval); 153 } 154 } 155 audit_getname(result); 156 return result; 157 } 158 159 #ifdef CONFIG_AUDITSYSCALL 160 void putname(const char *name) 161 { 162 if (unlikely(!audit_dummy_context())) 163 audit_putname(name); 164 else 165 __putname(name); 166 } 167 EXPORT_SYMBOL(putname); 168 #endif 169 170 171 /** 172 * generic_permission - check for access rights on a Posix-like filesystem 173 * @inode: inode to check access rights for 174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 175 * @check_acl: optional callback to check for Posix ACLs 176 * 177 * Used to check for read/write/execute permissions on a file. 178 * We use "fsuid" for this, letting us set arbitrary permissions 179 * for filesystem access without changing the "normal" uids which 180 * are used for other things.. 181 */ 182 int generic_permission(struct inode *inode, int mask, 183 int (*check_acl)(struct inode *inode, int mask)) 184 { 185 umode_t mode = inode->i_mode; 186 187 if (current->fsuid == inode->i_uid) 188 mode >>= 6; 189 else { 190 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) { 191 int error = check_acl(inode, mask); 192 if (error == -EACCES) 193 goto check_capabilities; 194 else if (error != -EAGAIN) 195 return error; 196 } 197 198 if (in_group_p(inode->i_gid)) 199 mode >>= 3; 200 } 201 202 /* 203 * If the DACs are ok we don't need any capability check. 204 */ 205 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask)) 206 return 0; 207 208 check_capabilities: 209 /* 210 * Read/write DACs are always overridable. 211 * Executable DACs are overridable if at least one exec bit is set. 212 */ 213 if (!(mask & MAY_EXEC) || 214 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode)) 215 if (capable(CAP_DAC_OVERRIDE)) 216 return 0; 217 218 /* 219 * Searching includes executable on directories, else just read. 220 */ 221 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))) 222 if (capable(CAP_DAC_READ_SEARCH)) 223 return 0; 224 225 return -EACCES; 226 } 227 228 int permission(struct inode *inode, int mask, struct nameidata *nd) 229 { 230 int retval, submask; 231 struct vfsmount *mnt = NULL; 232 233 if (nd) 234 mnt = nd->mnt; 235 236 if (mask & MAY_WRITE) { 237 umode_t mode = inode->i_mode; 238 239 /* 240 * Nobody gets write access to a read-only fs. 241 */ 242 if (IS_RDONLY(inode) && 243 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 244 return -EROFS; 245 246 /* 247 * Nobody gets write access to an immutable file. 248 */ 249 if (IS_IMMUTABLE(inode)) 250 return -EACCES; 251 } 252 253 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) { 254 /* 255 * MAY_EXEC on regular files is denied if the fs is mounted 256 * with the "noexec" flag. 257 */ 258 if (mnt && (mnt->mnt_flags & MNT_NOEXEC)) 259 return -EACCES; 260 } 261 262 /* Ordinary permission routines do not understand MAY_APPEND. */ 263 submask = mask & ~MAY_APPEND; 264 if (inode->i_op && inode->i_op->permission) { 265 retval = inode->i_op->permission(inode, submask, nd); 266 if (!retval) { 267 /* 268 * Exec permission on a regular file is denied if none 269 * of the execute bits are set. 270 * 271 * This check should be done by the ->permission() 272 * method. 273 */ 274 if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode) && 275 !(inode->i_mode & S_IXUGO)) 276 return -EACCES; 277 } 278 } else { 279 retval = generic_permission(inode, submask, NULL); 280 } 281 if (retval) 282 return retval; 283 284 return security_inode_permission(inode, mask, nd); 285 } 286 287 /** 288 * vfs_permission - check for access rights to a given path 289 * @nd: lookup result that describes the path 290 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 291 * 292 * Used to check for read/write/execute permissions on a path. 293 * We use "fsuid" for this, letting us set arbitrary permissions 294 * for filesystem access without changing the "normal" uids which 295 * are used for other things. 296 */ 297 int vfs_permission(struct nameidata *nd, int mask) 298 { 299 return permission(nd->dentry->d_inode, mask, nd); 300 } 301 302 /** 303 * file_permission - check for additional access rights to a given file 304 * @file: file to check access rights for 305 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 306 * 307 * Used to check for read/write/execute permissions on an already opened 308 * file. 309 * 310 * Note: 311 * Do not use this function in new code. All access checks should 312 * be done using vfs_permission(). 313 */ 314 int file_permission(struct file *file, int mask) 315 { 316 return permission(file->f_path.dentry->d_inode, mask, NULL); 317 } 318 319 /* 320 * get_write_access() gets write permission for a file. 321 * put_write_access() releases this write permission. 322 * This is used for regular files. 323 * We cannot support write (and maybe mmap read-write shared) accesses and 324 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode 325 * can have the following values: 326 * 0: no writers, no VM_DENYWRITE mappings 327 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist 328 * > 0: (i_writecount) users are writing to the file. 329 * 330 * Normally we operate on that counter with atomic_{inc,dec} and it's safe 331 * except for the cases where we don't hold i_writecount yet. Then we need to 332 * use {get,deny}_write_access() - these functions check the sign and refuse 333 * to do the change if sign is wrong. Exclusion between them is provided by 334 * the inode->i_lock spinlock. 335 */ 336 337 int get_write_access(struct inode * inode) 338 { 339 spin_lock(&inode->i_lock); 340 if (atomic_read(&inode->i_writecount) < 0) { 341 spin_unlock(&inode->i_lock); 342 return -ETXTBSY; 343 } 344 atomic_inc(&inode->i_writecount); 345 spin_unlock(&inode->i_lock); 346 347 return 0; 348 } 349 350 int deny_write_access(struct file * file) 351 { 352 struct inode *inode = file->f_path.dentry->d_inode; 353 354 spin_lock(&inode->i_lock); 355 if (atomic_read(&inode->i_writecount) > 0) { 356 spin_unlock(&inode->i_lock); 357 return -ETXTBSY; 358 } 359 atomic_dec(&inode->i_writecount); 360 spin_unlock(&inode->i_lock); 361 362 return 0; 363 } 364 365 void path_release(struct nameidata *nd) 366 { 367 dput(nd->dentry); 368 mntput(nd->mnt); 369 } 370 371 /* 372 * umount() mustn't call path_release()/mntput() as that would clear 373 * mnt_expiry_mark 374 */ 375 void path_release_on_umount(struct nameidata *nd) 376 { 377 dput(nd->dentry); 378 mntput_no_expire(nd->mnt); 379 } 380 381 /** 382 * release_open_intent - free up open intent resources 383 * @nd: pointer to nameidata 384 */ 385 void release_open_intent(struct nameidata *nd) 386 { 387 if (nd->intent.open.file->f_path.dentry == NULL) 388 put_filp(nd->intent.open.file); 389 else 390 fput(nd->intent.open.file); 391 } 392 393 static inline struct dentry * 394 do_revalidate(struct dentry *dentry, struct nameidata *nd) 395 { 396 int status = dentry->d_op->d_revalidate(dentry, nd); 397 if (unlikely(status <= 0)) { 398 /* 399 * The dentry failed validation. 400 * If d_revalidate returned 0 attempt to invalidate 401 * the dentry otherwise d_revalidate is asking us 402 * to return a fail status. 403 */ 404 if (!status) { 405 if (!d_invalidate(dentry)) { 406 dput(dentry); 407 dentry = NULL; 408 } 409 } else { 410 dput(dentry); 411 dentry = ERR_PTR(status); 412 } 413 } 414 return dentry; 415 } 416 417 /* 418 * Internal lookup() using the new generic dcache. 419 * SMP-safe 420 */ 421 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd) 422 { 423 struct dentry * dentry = __d_lookup(parent, name); 424 425 /* lockess __d_lookup may fail due to concurrent d_move() 426 * in some unrelated directory, so try with d_lookup 427 */ 428 if (!dentry) 429 dentry = d_lookup(parent, name); 430 431 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) 432 dentry = do_revalidate(dentry, nd); 433 434 return dentry; 435 } 436 437 /* 438 * Short-cut version of permission(), for calling by 439 * path_walk(), when dcache lock is held. Combines parts 440 * of permission() and generic_permission(), and tests ONLY for 441 * MAY_EXEC permission. 442 * 443 * If appropriate, check DAC only. If not appropriate, or 444 * short-cut DAC fails, then call permission() to do more 445 * complete permission check. 446 */ 447 static int exec_permission_lite(struct inode *inode, 448 struct nameidata *nd) 449 { 450 umode_t mode = inode->i_mode; 451 452 if (inode->i_op && inode->i_op->permission) 453 return -EAGAIN; 454 455 if (current->fsuid == inode->i_uid) 456 mode >>= 6; 457 else if (in_group_p(inode->i_gid)) 458 mode >>= 3; 459 460 if (mode & MAY_EXEC) 461 goto ok; 462 463 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE)) 464 goto ok; 465 466 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE)) 467 goto ok; 468 469 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH)) 470 goto ok; 471 472 return -EACCES; 473 ok: 474 return security_inode_permission(inode, MAY_EXEC, nd); 475 } 476 477 /* 478 * This is called when everything else fails, and we actually have 479 * to go to the low-level filesystem to find out what we should do.. 480 * 481 * We get the directory semaphore, and after getting that we also 482 * make sure that nobody added the entry to the dcache in the meantime.. 483 * SMP-safe 484 */ 485 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd) 486 { 487 struct dentry * result; 488 struct inode *dir = parent->d_inode; 489 490 mutex_lock(&dir->i_mutex); 491 /* 492 * First re-do the cached lookup just in case it was created 493 * while we waited for the directory semaphore.. 494 * 495 * FIXME! This could use version numbering or similar to 496 * avoid unnecessary cache lookups. 497 * 498 * The "dcache_lock" is purely to protect the RCU list walker 499 * from concurrent renames at this point (we mustn't get false 500 * negatives from the RCU list walk here, unlike the optimistic 501 * fast walk). 502 * 503 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup 504 */ 505 result = d_lookup(parent, name); 506 if (!result) { 507 struct dentry * dentry = d_alloc(parent, name); 508 result = ERR_PTR(-ENOMEM); 509 if (dentry) { 510 result = dir->i_op->lookup(dir, dentry, nd); 511 if (result) 512 dput(dentry); 513 else 514 result = dentry; 515 } 516 mutex_unlock(&dir->i_mutex); 517 return result; 518 } 519 520 /* 521 * Uhhuh! Nasty case: the cache was re-populated while 522 * we waited on the semaphore. Need to revalidate. 523 */ 524 mutex_unlock(&dir->i_mutex); 525 if (result->d_op && result->d_op->d_revalidate) { 526 result = do_revalidate(result, nd); 527 if (!result) 528 result = ERR_PTR(-ENOENT); 529 } 530 return result; 531 } 532 533 static int __emul_lookup_dentry(const char *, struct nameidata *); 534 535 /* SMP-safe */ 536 static __always_inline int 537 walk_init_root(const char *name, struct nameidata *nd) 538 { 539 struct fs_struct *fs = current->fs; 540 541 read_lock(&fs->lock); 542 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) { 543 nd->mnt = mntget(fs->altrootmnt); 544 nd->dentry = dget(fs->altroot); 545 read_unlock(&fs->lock); 546 if (__emul_lookup_dentry(name,nd)) 547 return 0; 548 read_lock(&fs->lock); 549 } 550 nd->mnt = mntget(fs->rootmnt); 551 nd->dentry = dget(fs->root); 552 read_unlock(&fs->lock); 553 return 1; 554 } 555 556 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 557 { 558 int res = 0; 559 char *name; 560 if (IS_ERR(link)) 561 goto fail; 562 563 if (*link == '/') { 564 path_release(nd); 565 if (!walk_init_root(link, nd)) 566 /* weird __emul_prefix() stuff did it */ 567 goto out; 568 } 569 res = link_path_walk(link, nd); 570 out: 571 if (nd->depth || res || nd->last_type!=LAST_NORM) 572 return res; 573 /* 574 * If it is an iterative symlinks resolution in open_namei() we 575 * have to copy the last component. And all that crap because of 576 * bloody create() on broken symlinks. Furrfu... 577 */ 578 name = __getname(); 579 if (unlikely(!name)) { 580 path_release(nd); 581 return -ENOMEM; 582 } 583 strcpy(name, nd->last.name); 584 nd->last.name = name; 585 return 0; 586 fail: 587 path_release(nd); 588 return PTR_ERR(link); 589 } 590 591 static inline void dput_path(struct path *path, struct nameidata *nd) 592 { 593 dput(path->dentry); 594 if (path->mnt != nd->mnt) 595 mntput(path->mnt); 596 } 597 598 static inline void path_to_nameidata(struct path *path, struct nameidata *nd) 599 { 600 dput(nd->dentry); 601 if (nd->mnt != path->mnt) 602 mntput(nd->mnt); 603 nd->mnt = path->mnt; 604 nd->dentry = path->dentry; 605 } 606 607 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd) 608 { 609 int error; 610 void *cookie; 611 struct dentry *dentry = path->dentry; 612 613 touch_atime(path->mnt, dentry); 614 nd_set_link(nd, NULL); 615 616 if (path->mnt != nd->mnt) { 617 path_to_nameidata(path, nd); 618 dget(dentry); 619 } 620 mntget(path->mnt); 621 cookie = dentry->d_inode->i_op->follow_link(dentry, nd); 622 error = PTR_ERR(cookie); 623 if (!IS_ERR(cookie)) { 624 char *s = nd_get_link(nd); 625 error = 0; 626 if (s) 627 error = __vfs_follow_link(nd, s); 628 if (dentry->d_inode->i_op->put_link) 629 dentry->d_inode->i_op->put_link(dentry, nd, cookie); 630 } 631 dput(dentry); 632 mntput(path->mnt); 633 634 return error; 635 } 636 637 /* 638 * This limits recursive symlink follows to 8, while 639 * limiting consecutive symlinks to 40. 640 * 641 * Without that kind of total limit, nasty chains of consecutive 642 * symlinks can cause almost arbitrarily long lookups. 643 */ 644 static inline int do_follow_link(struct path *path, struct nameidata *nd) 645 { 646 int err = -ELOOP; 647 if (current->link_count >= MAX_NESTED_LINKS) 648 goto loop; 649 if (current->total_link_count >= 40) 650 goto loop; 651 BUG_ON(nd->depth >= MAX_NESTED_LINKS); 652 cond_resched(); 653 err = security_inode_follow_link(path->dentry, nd); 654 if (err) 655 goto loop; 656 current->link_count++; 657 current->total_link_count++; 658 nd->depth++; 659 err = __do_follow_link(path, nd); 660 current->link_count--; 661 nd->depth--; 662 return err; 663 loop: 664 dput_path(path, nd); 665 path_release(nd); 666 return err; 667 } 668 669 int follow_up(struct vfsmount **mnt, struct dentry **dentry) 670 { 671 struct vfsmount *parent; 672 struct dentry *mountpoint; 673 spin_lock(&vfsmount_lock); 674 parent=(*mnt)->mnt_parent; 675 if (parent == *mnt) { 676 spin_unlock(&vfsmount_lock); 677 return 0; 678 } 679 mntget(parent); 680 mountpoint=dget((*mnt)->mnt_mountpoint); 681 spin_unlock(&vfsmount_lock); 682 dput(*dentry); 683 *dentry = mountpoint; 684 mntput(*mnt); 685 *mnt = parent; 686 return 1; 687 } 688 689 /* no need for dcache_lock, as serialization is taken care in 690 * namespace.c 691 */ 692 static int __follow_mount(struct path *path) 693 { 694 int res = 0; 695 while (d_mountpoint(path->dentry)) { 696 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry); 697 if (!mounted) 698 break; 699 dput(path->dentry); 700 if (res) 701 mntput(path->mnt); 702 path->mnt = mounted; 703 path->dentry = dget(mounted->mnt_root); 704 res = 1; 705 } 706 return res; 707 } 708 709 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry) 710 { 711 while (d_mountpoint(*dentry)) { 712 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry); 713 if (!mounted) 714 break; 715 dput(*dentry); 716 mntput(*mnt); 717 *mnt = mounted; 718 *dentry = dget(mounted->mnt_root); 719 } 720 } 721 722 /* no need for dcache_lock, as serialization is taken care in 723 * namespace.c 724 */ 725 int follow_down(struct vfsmount **mnt, struct dentry **dentry) 726 { 727 struct vfsmount *mounted; 728 729 mounted = lookup_mnt(*mnt, *dentry); 730 if (mounted) { 731 dput(*dentry); 732 mntput(*mnt); 733 *mnt = mounted; 734 *dentry = dget(mounted->mnt_root); 735 return 1; 736 } 737 return 0; 738 } 739 740 static __always_inline void follow_dotdot(struct nameidata *nd) 741 { 742 struct fs_struct *fs = current->fs; 743 744 while(1) { 745 struct vfsmount *parent; 746 struct dentry *old = nd->dentry; 747 748 read_lock(&fs->lock); 749 if (nd->dentry == fs->root && 750 nd->mnt == fs->rootmnt) { 751 read_unlock(&fs->lock); 752 break; 753 } 754 read_unlock(&fs->lock); 755 spin_lock(&dcache_lock); 756 if (nd->dentry != nd->mnt->mnt_root) { 757 nd->dentry = dget(nd->dentry->d_parent); 758 spin_unlock(&dcache_lock); 759 dput(old); 760 break; 761 } 762 spin_unlock(&dcache_lock); 763 spin_lock(&vfsmount_lock); 764 parent = nd->mnt->mnt_parent; 765 if (parent == nd->mnt) { 766 spin_unlock(&vfsmount_lock); 767 break; 768 } 769 mntget(parent); 770 nd->dentry = dget(nd->mnt->mnt_mountpoint); 771 spin_unlock(&vfsmount_lock); 772 dput(old); 773 mntput(nd->mnt); 774 nd->mnt = parent; 775 } 776 follow_mount(&nd->mnt, &nd->dentry); 777 } 778 779 /* 780 * It's more convoluted than I'd like it to be, but... it's still fairly 781 * small and for now I'd prefer to have fast path as straight as possible. 782 * It _is_ time-critical. 783 */ 784 static int do_lookup(struct nameidata *nd, struct qstr *name, 785 struct path *path) 786 { 787 struct vfsmount *mnt = nd->mnt; 788 struct dentry *dentry = __d_lookup(nd->dentry, name); 789 790 if (!dentry) 791 goto need_lookup; 792 if (dentry->d_op && dentry->d_op->d_revalidate) 793 goto need_revalidate; 794 done: 795 path->mnt = mnt; 796 path->dentry = dentry; 797 __follow_mount(path); 798 return 0; 799 800 need_lookup: 801 dentry = real_lookup(nd->dentry, name, nd); 802 if (IS_ERR(dentry)) 803 goto fail; 804 goto done; 805 806 need_revalidate: 807 dentry = do_revalidate(dentry, nd); 808 if (!dentry) 809 goto need_lookup; 810 if (IS_ERR(dentry)) 811 goto fail; 812 goto done; 813 814 fail: 815 return PTR_ERR(dentry); 816 } 817 818 /* 819 * Name resolution. 820 * This is the basic name resolution function, turning a pathname into 821 * the final dentry. We expect 'base' to be positive and a directory. 822 * 823 * Returns 0 and nd will have valid dentry and mnt on success. 824 * Returns error and drops reference to input namei data on failure. 825 */ 826 static fastcall int __link_path_walk(const char * name, struct nameidata *nd) 827 { 828 struct path next; 829 struct inode *inode; 830 int err; 831 unsigned int lookup_flags = nd->flags; 832 833 while (*name=='/') 834 name++; 835 if (!*name) 836 goto return_reval; 837 838 inode = nd->dentry->d_inode; 839 if (nd->depth) 840 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE); 841 842 /* At this point we know we have a real path component. */ 843 for(;;) { 844 unsigned long hash; 845 struct qstr this; 846 unsigned int c; 847 848 nd->flags |= LOOKUP_CONTINUE; 849 err = exec_permission_lite(inode, nd); 850 if (err == -EAGAIN) 851 err = vfs_permission(nd, MAY_EXEC); 852 if (err) 853 break; 854 855 this.name = name; 856 c = *(const unsigned char *)name; 857 858 hash = init_name_hash(); 859 do { 860 name++; 861 hash = partial_name_hash(c, hash); 862 c = *(const unsigned char *)name; 863 } while (c && (c != '/')); 864 this.len = name - (const char *) this.name; 865 this.hash = end_name_hash(hash); 866 867 /* remove trailing slashes? */ 868 if (!c) 869 goto last_component; 870 while (*++name == '/'); 871 if (!*name) 872 goto last_with_slashes; 873 874 /* 875 * "." and ".." are special - ".." especially so because it has 876 * to be able to know about the current root directory and 877 * parent relationships. 878 */ 879 if (this.name[0] == '.') switch (this.len) { 880 default: 881 break; 882 case 2: 883 if (this.name[1] != '.') 884 break; 885 follow_dotdot(nd); 886 inode = nd->dentry->d_inode; 887 /* fallthrough */ 888 case 1: 889 continue; 890 } 891 /* 892 * See if the low-level filesystem might want 893 * to use its own hash.. 894 */ 895 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) { 896 err = nd->dentry->d_op->d_hash(nd->dentry, &this); 897 if (err < 0) 898 break; 899 } 900 /* This does the actual lookups.. */ 901 err = do_lookup(nd, &this, &next); 902 if (err) 903 break; 904 905 err = -ENOENT; 906 inode = next.dentry->d_inode; 907 if (!inode) 908 goto out_dput; 909 err = -ENOTDIR; 910 if (!inode->i_op) 911 goto out_dput; 912 913 if (inode->i_op->follow_link) { 914 err = do_follow_link(&next, nd); 915 if (err) 916 goto return_err; 917 err = -ENOENT; 918 inode = nd->dentry->d_inode; 919 if (!inode) 920 break; 921 err = -ENOTDIR; 922 if (!inode->i_op) 923 break; 924 } else 925 path_to_nameidata(&next, nd); 926 err = -ENOTDIR; 927 if (!inode->i_op->lookup) 928 break; 929 continue; 930 /* here ends the main loop */ 931 932 last_with_slashes: 933 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 934 last_component: 935 /* Clear LOOKUP_CONTINUE iff it was previously unset */ 936 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; 937 if (lookup_flags & LOOKUP_PARENT) 938 goto lookup_parent; 939 if (this.name[0] == '.') switch (this.len) { 940 default: 941 break; 942 case 2: 943 if (this.name[1] != '.') 944 break; 945 follow_dotdot(nd); 946 inode = nd->dentry->d_inode; 947 /* fallthrough */ 948 case 1: 949 goto return_reval; 950 } 951 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) { 952 err = nd->dentry->d_op->d_hash(nd->dentry, &this); 953 if (err < 0) 954 break; 955 } 956 err = do_lookup(nd, &this, &next); 957 if (err) 958 break; 959 inode = next.dentry->d_inode; 960 if ((lookup_flags & LOOKUP_FOLLOW) 961 && inode && inode->i_op && inode->i_op->follow_link) { 962 err = do_follow_link(&next, nd); 963 if (err) 964 goto return_err; 965 inode = nd->dentry->d_inode; 966 } else 967 path_to_nameidata(&next, nd); 968 err = -ENOENT; 969 if (!inode) 970 break; 971 if (lookup_flags & LOOKUP_DIRECTORY) { 972 err = -ENOTDIR; 973 if (!inode->i_op || !inode->i_op->lookup) 974 break; 975 } 976 goto return_base; 977 lookup_parent: 978 nd->last = this; 979 nd->last_type = LAST_NORM; 980 if (this.name[0] != '.') 981 goto return_base; 982 if (this.len == 1) 983 nd->last_type = LAST_DOT; 984 else if (this.len == 2 && this.name[1] == '.') 985 nd->last_type = LAST_DOTDOT; 986 else 987 goto return_base; 988 return_reval: 989 /* 990 * We bypassed the ordinary revalidation routines. 991 * We may need to check the cached dentry for staleness. 992 */ 993 if (nd->dentry && nd->dentry->d_sb && 994 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) { 995 err = -ESTALE; 996 /* Note: we do not d_invalidate() */ 997 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd)) 998 break; 999 } 1000 return_base: 1001 return 0; 1002 out_dput: 1003 dput_path(&next, nd); 1004 break; 1005 } 1006 path_release(nd); 1007 return_err: 1008 return err; 1009 } 1010 1011 /* 1012 * Wrapper to retry pathname resolution whenever the underlying 1013 * file system returns an ESTALE. 1014 * 1015 * Retry the whole path once, forcing real lookup requests 1016 * instead of relying on the dcache. 1017 */ 1018 static int fastcall link_path_walk(const char *name, struct nameidata *nd) 1019 { 1020 struct nameidata save = *nd; 1021 int result; 1022 1023 /* make sure the stuff we saved doesn't go away */ 1024 dget(save.dentry); 1025 mntget(save.mnt); 1026 1027 result = __link_path_walk(name, nd); 1028 if (result == -ESTALE) { 1029 *nd = save; 1030 dget(nd->dentry); 1031 mntget(nd->mnt); 1032 nd->flags |= LOOKUP_REVAL; 1033 result = __link_path_walk(name, nd); 1034 } 1035 1036 dput(save.dentry); 1037 mntput(save.mnt); 1038 1039 return result; 1040 } 1041 1042 static int fastcall path_walk(const char * name, struct nameidata *nd) 1043 { 1044 current->total_link_count = 0; 1045 return link_path_walk(name, nd); 1046 } 1047 1048 /* 1049 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if 1050 * everything is done. Returns 0 and drops input nd, if lookup failed; 1051 */ 1052 static int __emul_lookup_dentry(const char *name, struct nameidata *nd) 1053 { 1054 if (path_walk(name, nd)) 1055 return 0; /* something went wrong... */ 1056 1057 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) { 1058 struct dentry *old_dentry = nd->dentry; 1059 struct vfsmount *old_mnt = nd->mnt; 1060 struct qstr last = nd->last; 1061 int last_type = nd->last_type; 1062 struct fs_struct *fs = current->fs; 1063 1064 /* 1065 * NAME was not found in alternate root or it's a directory. 1066 * Try to find it in the normal root: 1067 */ 1068 nd->last_type = LAST_ROOT; 1069 read_lock(&fs->lock); 1070 nd->mnt = mntget(fs->rootmnt); 1071 nd->dentry = dget(fs->root); 1072 read_unlock(&fs->lock); 1073 if (path_walk(name, nd) == 0) { 1074 if (nd->dentry->d_inode) { 1075 dput(old_dentry); 1076 mntput(old_mnt); 1077 return 1; 1078 } 1079 path_release(nd); 1080 } 1081 nd->dentry = old_dentry; 1082 nd->mnt = old_mnt; 1083 nd->last = last; 1084 nd->last_type = last_type; 1085 } 1086 return 1; 1087 } 1088 1089 void set_fs_altroot(void) 1090 { 1091 char *emul = __emul_prefix(); 1092 struct nameidata nd; 1093 struct vfsmount *mnt = NULL, *oldmnt; 1094 struct dentry *dentry = NULL, *olddentry; 1095 int err; 1096 struct fs_struct *fs = current->fs; 1097 1098 if (!emul) 1099 goto set_it; 1100 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd); 1101 if (!err) { 1102 mnt = nd.mnt; 1103 dentry = nd.dentry; 1104 } 1105 set_it: 1106 write_lock(&fs->lock); 1107 oldmnt = fs->altrootmnt; 1108 olddentry = fs->altroot; 1109 fs->altrootmnt = mnt; 1110 fs->altroot = dentry; 1111 write_unlock(&fs->lock); 1112 if (olddentry) { 1113 dput(olddentry); 1114 mntput(oldmnt); 1115 } 1116 } 1117 1118 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1119 static int fastcall do_path_lookup(int dfd, const char *name, 1120 unsigned int flags, struct nameidata *nd) 1121 { 1122 int retval = 0; 1123 int fput_needed; 1124 struct file *file; 1125 struct fs_struct *fs = current->fs; 1126 1127 nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1128 nd->flags = flags; 1129 nd->depth = 0; 1130 1131 if (*name=='/') { 1132 read_lock(&fs->lock); 1133 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) { 1134 nd->mnt = mntget(fs->altrootmnt); 1135 nd->dentry = dget(fs->altroot); 1136 read_unlock(&fs->lock); 1137 if (__emul_lookup_dentry(name,nd)) 1138 goto out; /* found in altroot */ 1139 read_lock(&fs->lock); 1140 } 1141 nd->mnt = mntget(fs->rootmnt); 1142 nd->dentry = dget(fs->root); 1143 read_unlock(&fs->lock); 1144 } else if (dfd == AT_FDCWD) { 1145 read_lock(&fs->lock); 1146 nd->mnt = mntget(fs->pwdmnt); 1147 nd->dentry = dget(fs->pwd); 1148 read_unlock(&fs->lock); 1149 } else { 1150 struct dentry *dentry; 1151 1152 file = fget_light(dfd, &fput_needed); 1153 retval = -EBADF; 1154 if (!file) 1155 goto out_fail; 1156 1157 dentry = file->f_path.dentry; 1158 1159 retval = -ENOTDIR; 1160 if (!S_ISDIR(dentry->d_inode->i_mode)) 1161 goto fput_fail; 1162 1163 retval = file_permission(file, MAY_EXEC); 1164 if (retval) 1165 goto fput_fail; 1166 1167 nd->mnt = mntget(file->f_path.mnt); 1168 nd->dentry = dget(dentry); 1169 1170 fput_light(file, fput_needed); 1171 } 1172 1173 retval = path_walk(name, nd); 1174 out: 1175 if (unlikely(!retval && !audit_dummy_context() && nd->dentry && 1176 nd->dentry->d_inode)) 1177 audit_inode(name, nd->dentry); 1178 out_fail: 1179 return retval; 1180 1181 fput_fail: 1182 fput_light(file, fput_needed); 1183 goto out_fail; 1184 } 1185 1186 int fastcall path_lookup(const char *name, unsigned int flags, 1187 struct nameidata *nd) 1188 { 1189 return do_path_lookup(AT_FDCWD, name, flags, nd); 1190 } 1191 1192 /** 1193 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 1194 * @dentry: pointer to dentry of the base directory 1195 * @mnt: pointer to vfs mount of the base directory 1196 * @name: pointer to file name 1197 * @flags: lookup flags 1198 * @nd: pointer to nameidata 1199 */ 1200 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 1201 const char *name, unsigned int flags, 1202 struct nameidata *nd) 1203 { 1204 int retval; 1205 1206 /* same as do_path_lookup */ 1207 nd->last_type = LAST_ROOT; 1208 nd->flags = flags; 1209 nd->depth = 0; 1210 1211 nd->mnt = mntget(mnt); 1212 nd->dentry = dget(dentry); 1213 1214 retval = path_walk(name, nd); 1215 if (unlikely(!retval && !audit_dummy_context() && nd->dentry && 1216 nd->dentry->d_inode)) 1217 audit_inode(name, nd->dentry); 1218 1219 return retval; 1220 1221 } 1222 1223 static int __path_lookup_intent_open(int dfd, const char *name, 1224 unsigned int lookup_flags, struct nameidata *nd, 1225 int open_flags, int create_mode) 1226 { 1227 struct file *filp = get_empty_filp(); 1228 int err; 1229 1230 if (filp == NULL) 1231 return -ENFILE; 1232 nd->intent.open.file = filp; 1233 nd->intent.open.flags = open_flags; 1234 nd->intent.open.create_mode = create_mode; 1235 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd); 1236 if (IS_ERR(nd->intent.open.file)) { 1237 if (err == 0) { 1238 err = PTR_ERR(nd->intent.open.file); 1239 path_release(nd); 1240 } 1241 } else if (err != 0) 1242 release_open_intent(nd); 1243 return err; 1244 } 1245 1246 /** 1247 * path_lookup_open - lookup a file path with open intent 1248 * @dfd: the directory to use as base, or AT_FDCWD 1249 * @name: pointer to file name 1250 * @lookup_flags: lookup intent flags 1251 * @nd: pointer to nameidata 1252 * @open_flags: open intent flags 1253 */ 1254 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags, 1255 struct nameidata *nd, int open_flags) 1256 { 1257 return __path_lookup_intent_open(dfd, name, lookup_flags, nd, 1258 open_flags, 0); 1259 } 1260 1261 /** 1262 * path_lookup_create - lookup a file path with open + create intent 1263 * @dfd: the directory to use as base, or AT_FDCWD 1264 * @name: pointer to file name 1265 * @lookup_flags: lookup intent flags 1266 * @nd: pointer to nameidata 1267 * @open_flags: open intent flags 1268 * @create_mode: create intent flags 1269 */ 1270 static int path_lookup_create(int dfd, const char *name, 1271 unsigned int lookup_flags, struct nameidata *nd, 1272 int open_flags, int create_mode) 1273 { 1274 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE, 1275 nd, open_flags, create_mode); 1276 } 1277 1278 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags, 1279 struct nameidata *nd, int open_flags) 1280 { 1281 char *tmp = getname(name); 1282 int err = PTR_ERR(tmp); 1283 1284 if (!IS_ERR(tmp)) { 1285 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0); 1286 putname(tmp); 1287 } 1288 return err; 1289 } 1290 1291 static struct dentry *__lookup_hash(struct qstr *name, 1292 struct dentry *base, struct nameidata *nd) 1293 { 1294 struct dentry *dentry; 1295 struct inode *inode; 1296 int err; 1297 1298 inode = base->d_inode; 1299 1300 /* 1301 * See if the low-level filesystem might want 1302 * to use its own hash.. 1303 */ 1304 if (base->d_op && base->d_op->d_hash) { 1305 err = base->d_op->d_hash(base, name); 1306 dentry = ERR_PTR(err); 1307 if (err < 0) 1308 goto out; 1309 } 1310 1311 dentry = cached_lookup(base, name, nd); 1312 if (!dentry) { 1313 struct dentry *new = d_alloc(base, name); 1314 dentry = ERR_PTR(-ENOMEM); 1315 if (!new) 1316 goto out; 1317 dentry = inode->i_op->lookup(inode, new, nd); 1318 if (!dentry) 1319 dentry = new; 1320 else 1321 dput(new); 1322 } 1323 out: 1324 return dentry; 1325 } 1326 1327 /* 1328 * Restricted form of lookup. Doesn't follow links, single-component only, 1329 * needs parent already locked. Doesn't follow mounts. 1330 * SMP-safe. 1331 */ 1332 static struct dentry *lookup_hash(struct nameidata *nd) 1333 { 1334 int err; 1335 1336 err = permission(nd->dentry->d_inode, MAY_EXEC, nd); 1337 if (err) 1338 return ERR_PTR(err); 1339 return __lookup_hash(&nd->last, nd->dentry, nd); 1340 } 1341 1342 static int __lookup_one_len(const char *name, struct qstr *this, 1343 struct dentry *base, int len) 1344 { 1345 unsigned long hash; 1346 unsigned int c; 1347 1348 this->name = name; 1349 this->len = len; 1350 if (!len) 1351 return -EACCES; 1352 1353 hash = init_name_hash(); 1354 while (len--) { 1355 c = *(const unsigned char *)name++; 1356 if (c == '/' || c == '\0') 1357 return -EACCES; 1358 hash = partial_name_hash(c, hash); 1359 } 1360 this->hash = end_name_hash(hash); 1361 return 0; 1362 } 1363 1364 /** 1365 * lookup_one_len: filesystem helper to lookup single pathname component 1366 * @name: pathname component to lookup 1367 * @base: base directory to lookup from 1368 * @len: maximum length @len should be interpreted to 1369 * 1370 * Note that this routine is purely a helper for filesystem useage and should 1371 * not be called by generic code. Also note that by using this function to 1372 * nameidata argument is passed to the filesystem methods and a filesystem 1373 * using this helper needs to be prepared for that. 1374 */ 1375 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 1376 { 1377 int err; 1378 struct qstr this; 1379 1380 err = __lookup_one_len(name, &this, base, len); 1381 if (err) 1382 return ERR_PTR(err); 1383 1384 err = permission(base->d_inode, MAY_EXEC, NULL); 1385 if (err) 1386 return ERR_PTR(err); 1387 return __lookup_hash(&this, base, NULL); 1388 } 1389 1390 /** 1391 * lookup_one_noperm - bad hack for sysfs 1392 * @name: pathname component to lookup 1393 * @base: base directory to lookup from 1394 * 1395 * This is a variant of lookup_one_len that doesn't perform any permission 1396 * checks. It's a horrible hack to work around the braindead sysfs 1397 * architecture and should not be used anywhere else. 1398 * 1399 * DON'T USE THIS FUNCTION EVER, thanks. 1400 */ 1401 struct dentry *lookup_one_noperm(const char *name, struct dentry *base) 1402 { 1403 int err; 1404 struct qstr this; 1405 1406 err = __lookup_one_len(name, &this, base, strlen(name)); 1407 if (err) 1408 return ERR_PTR(err); 1409 return __lookup_hash(&this, base, NULL); 1410 } 1411 1412 int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags, 1413 struct nameidata *nd) 1414 { 1415 char *tmp = getname(name); 1416 int err = PTR_ERR(tmp); 1417 1418 if (!IS_ERR(tmp)) { 1419 err = do_path_lookup(dfd, tmp, flags, nd); 1420 putname(tmp); 1421 } 1422 return err; 1423 } 1424 1425 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd) 1426 { 1427 return __user_walk_fd(AT_FDCWD, name, flags, nd); 1428 } 1429 1430 /* 1431 * It's inline, so penalty for filesystems that don't use sticky bit is 1432 * minimal. 1433 */ 1434 static inline int check_sticky(struct inode *dir, struct inode *inode) 1435 { 1436 if (!(dir->i_mode & S_ISVTX)) 1437 return 0; 1438 if (inode->i_uid == current->fsuid) 1439 return 0; 1440 if (dir->i_uid == current->fsuid) 1441 return 0; 1442 return !capable(CAP_FOWNER); 1443 } 1444 1445 /* 1446 * Check whether we can remove a link victim from directory dir, check 1447 * whether the type of victim is right. 1448 * 1. We can't do it if dir is read-only (done in permission()) 1449 * 2. We should have write and exec permissions on dir 1450 * 3. We can't remove anything from append-only dir 1451 * 4. We can't do anything with immutable dir (done in permission()) 1452 * 5. If the sticky bit on dir is set we should either 1453 * a. be owner of dir, or 1454 * b. be owner of victim, or 1455 * c. have CAP_FOWNER capability 1456 * 6. If the victim is append-only or immutable we can't do antyhing with 1457 * links pointing to it. 1458 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 1459 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 1460 * 9. We can't remove a root or mountpoint. 1461 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 1462 * nfs_async_unlink(). 1463 */ 1464 static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 1465 { 1466 int error; 1467 1468 if (!victim->d_inode) 1469 return -ENOENT; 1470 1471 BUG_ON(victim->d_parent->d_inode != dir); 1472 audit_inode_child(victim->d_name.name, victim, dir); 1473 1474 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL); 1475 if (error) 1476 return error; 1477 if (IS_APPEND(dir)) 1478 return -EPERM; 1479 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 1480 IS_IMMUTABLE(victim->d_inode)) 1481 return -EPERM; 1482 if (isdir) { 1483 if (!S_ISDIR(victim->d_inode->i_mode)) 1484 return -ENOTDIR; 1485 if (IS_ROOT(victim)) 1486 return -EBUSY; 1487 } else if (S_ISDIR(victim->d_inode->i_mode)) 1488 return -EISDIR; 1489 if (IS_DEADDIR(dir)) 1490 return -ENOENT; 1491 if (victim->d_flags & DCACHE_NFSFS_RENAMED) 1492 return -EBUSY; 1493 return 0; 1494 } 1495 1496 /* Check whether we can create an object with dentry child in directory 1497 * dir. 1498 * 1. We can't do it if child already exists (open has special treatment for 1499 * this case, but since we are inlined it's OK) 1500 * 2. We can't do it if dir is read-only (done in permission()) 1501 * 3. We should have write and exec permissions on dir 1502 * 4. We can't do it if dir is immutable (done in permission()) 1503 */ 1504 static inline int may_create(struct inode *dir, struct dentry *child, 1505 struct nameidata *nd) 1506 { 1507 if (child->d_inode) 1508 return -EEXIST; 1509 if (IS_DEADDIR(dir)) 1510 return -ENOENT; 1511 return permission(dir,MAY_WRITE | MAY_EXEC, nd); 1512 } 1513 1514 /* 1515 * O_DIRECTORY translates into forcing a directory lookup. 1516 */ 1517 static inline int lookup_flags(unsigned int f) 1518 { 1519 unsigned long retval = LOOKUP_FOLLOW; 1520 1521 if (f & O_NOFOLLOW) 1522 retval &= ~LOOKUP_FOLLOW; 1523 1524 if (f & O_DIRECTORY) 1525 retval |= LOOKUP_DIRECTORY; 1526 1527 return retval; 1528 } 1529 1530 /* 1531 * p1 and p2 should be directories on the same fs. 1532 */ 1533 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 1534 { 1535 struct dentry *p; 1536 1537 if (p1 == p2) { 1538 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1539 return NULL; 1540 } 1541 1542 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 1543 1544 for (p = p1; p->d_parent != p; p = p->d_parent) { 1545 if (p->d_parent == p2) { 1546 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 1547 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 1548 return p; 1549 } 1550 } 1551 1552 for (p = p2; p->d_parent != p; p = p->d_parent) { 1553 if (p->d_parent == p1) { 1554 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1555 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 1556 return p; 1557 } 1558 } 1559 1560 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1561 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 1562 return NULL; 1563 } 1564 1565 void unlock_rename(struct dentry *p1, struct dentry *p2) 1566 { 1567 mutex_unlock(&p1->d_inode->i_mutex); 1568 if (p1 != p2) { 1569 mutex_unlock(&p2->d_inode->i_mutex); 1570 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 1571 } 1572 } 1573 1574 int vfs_create(struct inode *dir, struct dentry *dentry, int mode, 1575 struct nameidata *nd) 1576 { 1577 int error = may_create(dir, dentry, nd); 1578 1579 if (error) 1580 return error; 1581 1582 if (!dir->i_op || !dir->i_op->create) 1583 return -EACCES; /* shouldn't it be ENOSYS? */ 1584 mode &= S_IALLUGO; 1585 mode |= S_IFREG; 1586 error = security_inode_create(dir, dentry, mode); 1587 if (error) 1588 return error; 1589 DQUOT_INIT(dir); 1590 error = dir->i_op->create(dir, dentry, mode, nd); 1591 if (!error) 1592 fsnotify_create(dir, dentry); 1593 return error; 1594 } 1595 1596 int may_open(struct nameidata *nd, int acc_mode, int flag) 1597 { 1598 struct dentry *dentry = nd->dentry; 1599 struct inode *inode = dentry->d_inode; 1600 int error; 1601 1602 if (!inode) 1603 return -ENOENT; 1604 1605 if (S_ISLNK(inode->i_mode)) 1606 return -ELOOP; 1607 1608 if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE)) 1609 return -EISDIR; 1610 1611 /* 1612 * FIFO's, sockets and device files are special: they don't 1613 * actually live on the filesystem itself, and as such you 1614 * can write to them even if the filesystem is read-only. 1615 */ 1616 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { 1617 flag &= ~O_TRUNC; 1618 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) { 1619 if (nd->mnt->mnt_flags & MNT_NODEV) 1620 return -EACCES; 1621 1622 flag &= ~O_TRUNC; 1623 } else if (IS_RDONLY(inode) && (acc_mode & MAY_WRITE)) 1624 return -EROFS; 1625 1626 error = vfs_permission(nd, acc_mode); 1627 if (error) 1628 return error; 1629 /* 1630 * An append-only file must be opened in append mode for writing. 1631 */ 1632 if (IS_APPEND(inode)) { 1633 if ((flag & FMODE_WRITE) && !(flag & O_APPEND)) 1634 return -EPERM; 1635 if (flag & O_TRUNC) 1636 return -EPERM; 1637 } 1638 1639 /* O_NOATIME can only be set by the owner or superuser */ 1640 if (flag & O_NOATIME) 1641 if (!is_owner_or_cap(inode)) 1642 return -EPERM; 1643 1644 /* 1645 * Ensure there are no outstanding leases on the file. 1646 */ 1647 error = break_lease(inode, flag); 1648 if (error) 1649 return error; 1650 1651 if (flag & O_TRUNC) { 1652 error = get_write_access(inode); 1653 if (error) 1654 return error; 1655 1656 /* 1657 * Refuse to truncate files with mandatory locks held on them. 1658 */ 1659 error = locks_verify_locked(inode); 1660 if (!error) { 1661 DQUOT_INIT(inode); 1662 1663 error = do_truncate(dentry, 0, 1664 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 1665 NULL); 1666 } 1667 put_write_access(inode); 1668 if (error) 1669 return error; 1670 } else 1671 if (flag & FMODE_WRITE) 1672 DQUOT_INIT(inode); 1673 1674 return 0; 1675 } 1676 1677 static int open_namei_create(struct nameidata *nd, struct path *path, 1678 int flag, int mode) 1679 { 1680 int error; 1681 struct dentry *dir = nd->dentry; 1682 1683 if (!IS_POSIXACL(dir->d_inode)) 1684 mode &= ~current->fs->umask; 1685 error = vfs_create(dir->d_inode, path->dentry, mode, nd); 1686 mutex_unlock(&dir->d_inode->i_mutex); 1687 dput(nd->dentry); 1688 nd->dentry = path->dentry; 1689 if (error) 1690 return error; 1691 /* Don't check for write permission, don't truncate */ 1692 return may_open(nd, 0, flag & ~O_TRUNC); 1693 } 1694 1695 /* 1696 * open_namei() 1697 * 1698 * namei for open - this is in fact almost the whole open-routine. 1699 * 1700 * Note that the low bits of "flag" aren't the same as in the open 1701 * system call - they are 00 - no permissions needed 1702 * 01 - read permission needed 1703 * 10 - write permission needed 1704 * 11 - read/write permissions needed 1705 * which is a lot more logical, and also allows the "no perm" needed 1706 * for symlinks (where the permissions are checked later). 1707 * SMP-safe 1708 */ 1709 int open_namei(int dfd, const char *pathname, int flag, 1710 int mode, struct nameidata *nd) 1711 { 1712 int acc_mode, error; 1713 struct path path; 1714 struct dentry *dir; 1715 int count = 0; 1716 1717 acc_mode = ACC_MODE(flag); 1718 1719 /* O_TRUNC implies we need access checks for write permissions */ 1720 if (flag & O_TRUNC) 1721 acc_mode |= MAY_WRITE; 1722 1723 /* Allow the LSM permission hook to distinguish append 1724 access from general write access. */ 1725 if (flag & O_APPEND) 1726 acc_mode |= MAY_APPEND; 1727 1728 /* 1729 * The simplest case - just a plain lookup. 1730 */ 1731 if (!(flag & O_CREAT)) { 1732 error = path_lookup_open(dfd, pathname, lookup_flags(flag), 1733 nd, flag); 1734 if (error) 1735 return error; 1736 goto ok; 1737 } 1738 1739 /* 1740 * Create - we need to know the parent. 1741 */ 1742 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode); 1743 if (error) 1744 return error; 1745 1746 /* 1747 * We have the parent and last component. First of all, check 1748 * that we are not asked to creat(2) an obvious directory - that 1749 * will not do. 1750 */ 1751 error = -EISDIR; 1752 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len]) 1753 goto exit; 1754 1755 dir = nd->dentry; 1756 nd->flags &= ~LOOKUP_PARENT; 1757 mutex_lock(&dir->d_inode->i_mutex); 1758 path.dentry = lookup_hash(nd); 1759 path.mnt = nd->mnt; 1760 1761 do_last: 1762 error = PTR_ERR(path.dentry); 1763 if (IS_ERR(path.dentry)) { 1764 mutex_unlock(&dir->d_inode->i_mutex); 1765 goto exit; 1766 } 1767 1768 if (IS_ERR(nd->intent.open.file)) { 1769 mutex_unlock(&dir->d_inode->i_mutex); 1770 error = PTR_ERR(nd->intent.open.file); 1771 goto exit_dput; 1772 } 1773 1774 /* Negative dentry, just create the file */ 1775 if (!path.dentry->d_inode) { 1776 error = open_namei_create(nd, &path, flag, mode); 1777 if (error) 1778 goto exit; 1779 return 0; 1780 } 1781 1782 /* 1783 * It already exists. 1784 */ 1785 mutex_unlock(&dir->d_inode->i_mutex); 1786 audit_inode(pathname, path.dentry); 1787 1788 error = -EEXIST; 1789 if (flag & O_EXCL) 1790 goto exit_dput; 1791 1792 if (__follow_mount(&path)) { 1793 error = -ELOOP; 1794 if (flag & O_NOFOLLOW) 1795 goto exit_dput; 1796 } 1797 1798 error = -ENOENT; 1799 if (!path.dentry->d_inode) 1800 goto exit_dput; 1801 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link) 1802 goto do_link; 1803 1804 path_to_nameidata(&path, nd); 1805 error = -EISDIR; 1806 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode)) 1807 goto exit; 1808 ok: 1809 error = may_open(nd, acc_mode, flag); 1810 if (error) 1811 goto exit; 1812 return 0; 1813 1814 exit_dput: 1815 dput_path(&path, nd); 1816 exit: 1817 if (!IS_ERR(nd->intent.open.file)) 1818 release_open_intent(nd); 1819 path_release(nd); 1820 return error; 1821 1822 do_link: 1823 error = -ELOOP; 1824 if (flag & O_NOFOLLOW) 1825 goto exit_dput; 1826 /* 1827 * This is subtle. Instead of calling do_follow_link() we do the 1828 * thing by hands. The reason is that this way we have zero link_count 1829 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT. 1830 * After that we have the parent and last component, i.e. 1831 * we are in the same situation as after the first path_walk(). 1832 * Well, almost - if the last component is normal we get its copy 1833 * stored in nd->last.name and we will have to putname() it when we 1834 * are done. Procfs-like symlinks just set LAST_BIND. 1835 */ 1836 nd->flags |= LOOKUP_PARENT; 1837 error = security_inode_follow_link(path.dentry, nd); 1838 if (error) 1839 goto exit_dput; 1840 error = __do_follow_link(&path, nd); 1841 if (error) { 1842 /* Does someone understand code flow here? Or it is only 1843 * me so stupid? Anathema to whoever designed this non-sense 1844 * with "intent.open". 1845 */ 1846 release_open_intent(nd); 1847 return error; 1848 } 1849 nd->flags &= ~LOOKUP_PARENT; 1850 if (nd->last_type == LAST_BIND) 1851 goto ok; 1852 error = -EISDIR; 1853 if (nd->last_type != LAST_NORM) 1854 goto exit; 1855 if (nd->last.name[nd->last.len]) { 1856 __putname(nd->last.name); 1857 goto exit; 1858 } 1859 error = -ELOOP; 1860 if (count++==32) { 1861 __putname(nd->last.name); 1862 goto exit; 1863 } 1864 dir = nd->dentry; 1865 mutex_lock(&dir->d_inode->i_mutex); 1866 path.dentry = lookup_hash(nd); 1867 path.mnt = nd->mnt; 1868 __putname(nd->last.name); 1869 goto do_last; 1870 } 1871 1872 /** 1873 * lookup_create - lookup a dentry, creating it if it doesn't exist 1874 * @nd: nameidata info 1875 * @is_dir: directory flag 1876 * 1877 * Simple function to lookup and return a dentry and create it 1878 * if it doesn't exist. Is SMP-safe. 1879 * 1880 * Returns with nd->dentry->d_inode->i_mutex locked. 1881 */ 1882 struct dentry *lookup_create(struct nameidata *nd, int is_dir) 1883 { 1884 struct dentry *dentry = ERR_PTR(-EEXIST); 1885 1886 mutex_lock_nested(&nd->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 1887 /* 1888 * Yucky last component or no last component at all? 1889 * (foo/., foo/.., /////) 1890 */ 1891 if (nd->last_type != LAST_NORM) 1892 goto fail; 1893 nd->flags &= ~LOOKUP_PARENT; 1894 nd->flags |= LOOKUP_CREATE; 1895 nd->intent.open.flags = O_EXCL; 1896 1897 /* 1898 * Do the final lookup. 1899 */ 1900 dentry = lookup_hash(nd); 1901 if (IS_ERR(dentry)) 1902 goto fail; 1903 1904 /* 1905 * Special case - lookup gave negative, but... we had foo/bar/ 1906 * From the vfs_mknod() POV we just have a negative dentry - 1907 * all is fine. Let's be bastards - you had / on the end, you've 1908 * been asking for (non-existent) directory. -ENOENT for you. 1909 */ 1910 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode) 1911 goto enoent; 1912 return dentry; 1913 enoent: 1914 dput(dentry); 1915 dentry = ERR_PTR(-ENOENT); 1916 fail: 1917 return dentry; 1918 } 1919 EXPORT_SYMBOL_GPL(lookup_create); 1920 1921 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1922 { 1923 int error = may_create(dir, dentry, NULL); 1924 1925 if (error) 1926 return error; 1927 1928 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 1929 return -EPERM; 1930 1931 if (!dir->i_op || !dir->i_op->mknod) 1932 return -EPERM; 1933 1934 error = security_inode_mknod(dir, dentry, mode, dev); 1935 if (error) 1936 return error; 1937 1938 DQUOT_INIT(dir); 1939 error = dir->i_op->mknod(dir, dentry, mode, dev); 1940 if (!error) 1941 fsnotify_create(dir, dentry); 1942 return error; 1943 } 1944 1945 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode, 1946 unsigned dev) 1947 { 1948 int error = 0; 1949 char * tmp; 1950 struct dentry * dentry; 1951 struct nameidata nd; 1952 1953 if (S_ISDIR(mode)) 1954 return -EPERM; 1955 tmp = getname(filename); 1956 if (IS_ERR(tmp)) 1957 return PTR_ERR(tmp); 1958 1959 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd); 1960 if (error) 1961 goto out; 1962 dentry = lookup_create(&nd, 0); 1963 error = PTR_ERR(dentry); 1964 1965 if (!IS_POSIXACL(nd.dentry->d_inode)) 1966 mode &= ~current->fs->umask; 1967 if (!IS_ERR(dentry)) { 1968 switch (mode & S_IFMT) { 1969 case 0: case S_IFREG: 1970 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd); 1971 break; 1972 case S_IFCHR: case S_IFBLK: 1973 error = vfs_mknod(nd.dentry->d_inode,dentry,mode, 1974 new_decode_dev(dev)); 1975 break; 1976 case S_IFIFO: case S_IFSOCK: 1977 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0); 1978 break; 1979 case S_IFDIR: 1980 error = -EPERM; 1981 break; 1982 default: 1983 error = -EINVAL; 1984 } 1985 dput(dentry); 1986 } 1987 mutex_unlock(&nd.dentry->d_inode->i_mutex); 1988 path_release(&nd); 1989 out: 1990 putname(tmp); 1991 1992 return error; 1993 } 1994 1995 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev) 1996 { 1997 return sys_mknodat(AT_FDCWD, filename, mode, dev); 1998 } 1999 2000 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 2001 { 2002 int error = may_create(dir, dentry, NULL); 2003 2004 if (error) 2005 return error; 2006 2007 if (!dir->i_op || !dir->i_op->mkdir) 2008 return -EPERM; 2009 2010 mode &= (S_IRWXUGO|S_ISVTX); 2011 error = security_inode_mkdir(dir, dentry, mode); 2012 if (error) 2013 return error; 2014 2015 DQUOT_INIT(dir); 2016 error = dir->i_op->mkdir(dir, dentry, mode); 2017 if (!error) 2018 fsnotify_mkdir(dir, dentry); 2019 return error; 2020 } 2021 2022 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode) 2023 { 2024 int error = 0; 2025 char * tmp; 2026 struct dentry *dentry; 2027 struct nameidata nd; 2028 2029 tmp = getname(pathname); 2030 error = PTR_ERR(tmp); 2031 if (IS_ERR(tmp)) 2032 goto out_err; 2033 2034 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd); 2035 if (error) 2036 goto out; 2037 dentry = lookup_create(&nd, 1); 2038 error = PTR_ERR(dentry); 2039 if (IS_ERR(dentry)) 2040 goto out_unlock; 2041 2042 if (!IS_POSIXACL(nd.dentry->d_inode)) 2043 mode &= ~current->fs->umask; 2044 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode); 2045 dput(dentry); 2046 out_unlock: 2047 mutex_unlock(&nd.dentry->d_inode->i_mutex); 2048 path_release(&nd); 2049 out: 2050 putname(tmp); 2051 out_err: 2052 return error; 2053 } 2054 2055 asmlinkage long sys_mkdir(const char __user *pathname, int mode) 2056 { 2057 return sys_mkdirat(AT_FDCWD, pathname, mode); 2058 } 2059 2060 /* 2061 * We try to drop the dentry early: we should have 2062 * a usage count of 2 if we're the only user of this 2063 * dentry, and if that is true (possibly after pruning 2064 * the dcache), then we drop the dentry now. 2065 * 2066 * A low-level filesystem can, if it choses, legally 2067 * do a 2068 * 2069 * if (!d_unhashed(dentry)) 2070 * return -EBUSY; 2071 * 2072 * if it cannot handle the case of removing a directory 2073 * that is still in use by something else.. 2074 */ 2075 void dentry_unhash(struct dentry *dentry) 2076 { 2077 dget(dentry); 2078 shrink_dcache_parent(dentry); 2079 spin_lock(&dcache_lock); 2080 spin_lock(&dentry->d_lock); 2081 if (atomic_read(&dentry->d_count) == 2) 2082 __d_drop(dentry); 2083 spin_unlock(&dentry->d_lock); 2084 spin_unlock(&dcache_lock); 2085 } 2086 2087 int vfs_rmdir(struct inode *dir, struct dentry *dentry) 2088 { 2089 int error = may_delete(dir, dentry, 1); 2090 2091 if (error) 2092 return error; 2093 2094 if (!dir->i_op || !dir->i_op->rmdir) 2095 return -EPERM; 2096 2097 DQUOT_INIT(dir); 2098 2099 mutex_lock(&dentry->d_inode->i_mutex); 2100 dentry_unhash(dentry); 2101 if (d_mountpoint(dentry)) 2102 error = -EBUSY; 2103 else { 2104 error = security_inode_rmdir(dir, dentry); 2105 if (!error) { 2106 error = dir->i_op->rmdir(dir, dentry); 2107 if (!error) 2108 dentry->d_inode->i_flags |= S_DEAD; 2109 } 2110 } 2111 mutex_unlock(&dentry->d_inode->i_mutex); 2112 if (!error) { 2113 d_delete(dentry); 2114 } 2115 dput(dentry); 2116 2117 return error; 2118 } 2119 2120 static long do_rmdir(int dfd, const char __user *pathname) 2121 { 2122 int error = 0; 2123 char * name; 2124 struct dentry *dentry; 2125 struct nameidata nd; 2126 2127 name = getname(pathname); 2128 if(IS_ERR(name)) 2129 return PTR_ERR(name); 2130 2131 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd); 2132 if (error) 2133 goto exit; 2134 2135 switch(nd.last_type) { 2136 case LAST_DOTDOT: 2137 error = -ENOTEMPTY; 2138 goto exit1; 2139 case LAST_DOT: 2140 error = -EINVAL; 2141 goto exit1; 2142 case LAST_ROOT: 2143 error = -EBUSY; 2144 goto exit1; 2145 } 2146 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2147 dentry = lookup_hash(&nd); 2148 error = PTR_ERR(dentry); 2149 if (IS_ERR(dentry)) 2150 goto exit2; 2151 error = vfs_rmdir(nd.dentry->d_inode, dentry); 2152 dput(dentry); 2153 exit2: 2154 mutex_unlock(&nd.dentry->d_inode->i_mutex); 2155 exit1: 2156 path_release(&nd); 2157 exit: 2158 putname(name); 2159 return error; 2160 } 2161 2162 asmlinkage long sys_rmdir(const char __user *pathname) 2163 { 2164 return do_rmdir(AT_FDCWD, pathname); 2165 } 2166 2167 int vfs_unlink(struct inode *dir, struct dentry *dentry) 2168 { 2169 int error = may_delete(dir, dentry, 0); 2170 2171 if (error) 2172 return error; 2173 2174 if (!dir->i_op || !dir->i_op->unlink) 2175 return -EPERM; 2176 2177 DQUOT_INIT(dir); 2178 2179 mutex_lock(&dentry->d_inode->i_mutex); 2180 if (d_mountpoint(dentry)) 2181 error = -EBUSY; 2182 else { 2183 error = security_inode_unlink(dir, dentry); 2184 if (!error) 2185 error = dir->i_op->unlink(dir, dentry); 2186 } 2187 mutex_unlock(&dentry->d_inode->i_mutex); 2188 2189 /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 2190 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 2191 d_delete(dentry); 2192 } 2193 2194 return error; 2195 } 2196 2197 /* 2198 * Make sure that the actual truncation of the file will occur outside its 2199 * directory's i_mutex. Truncate can take a long time if there is a lot of 2200 * writeout happening, and we don't want to prevent access to the directory 2201 * while waiting on the I/O. 2202 */ 2203 static long do_unlinkat(int dfd, const char __user *pathname) 2204 { 2205 int error = 0; 2206 char * name; 2207 struct dentry *dentry; 2208 struct nameidata nd; 2209 struct inode *inode = NULL; 2210 2211 name = getname(pathname); 2212 if(IS_ERR(name)) 2213 return PTR_ERR(name); 2214 2215 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd); 2216 if (error) 2217 goto exit; 2218 error = -EISDIR; 2219 if (nd.last_type != LAST_NORM) 2220 goto exit1; 2221 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2222 dentry = lookup_hash(&nd); 2223 error = PTR_ERR(dentry); 2224 if (!IS_ERR(dentry)) { 2225 /* Why not before? Because we want correct error value */ 2226 if (nd.last.name[nd.last.len]) 2227 goto slashes; 2228 inode = dentry->d_inode; 2229 if (inode) 2230 atomic_inc(&inode->i_count); 2231 error = vfs_unlink(nd.dentry->d_inode, dentry); 2232 exit2: 2233 dput(dentry); 2234 } 2235 mutex_unlock(&nd.dentry->d_inode->i_mutex); 2236 if (inode) 2237 iput(inode); /* truncate the inode here */ 2238 exit1: 2239 path_release(&nd); 2240 exit: 2241 putname(name); 2242 return error; 2243 2244 slashes: 2245 error = !dentry->d_inode ? -ENOENT : 2246 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 2247 goto exit2; 2248 } 2249 2250 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag) 2251 { 2252 if ((flag & ~AT_REMOVEDIR) != 0) 2253 return -EINVAL; 2254 2255 if (flag & AT_REMOVEDIR) 2256 return do_rmdir(dfd, pathname); 2257 2258 return do_unlinkat(dfd, pathname); 2259 } 2260 2261 asmlinkage long sys_unlink(const char __user *pathname) 2262 { 2263 return do_unlinkat(AT_FDCWD, pathname); 2264 } 2265 2266 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode) 2267 { 2268 int error = may_create(dir, dentry, NULL); 2269 2270 if (error) 2271 return error; 2272 2273 if (!dir->i_op || !dir->i_op->symlink) 2274 return -EPERM; 2275 2276 error = security_inode_symlink(dir, dentry, oldname); 2277 if (error) 2278 return error; 2279 2280 DQUOT_INIT(dir); 2281 error = dir->i_op->symlink(dir, dentry, oldname); 2282 if (!error) 2283 fsnotify_create(dir, dentry); 2284 return error; 2285 } 2286 2287 asmlinkage long sys_symlinkat(const char __user *oldname, 2288 int newdfd, const char __user *newname) 2289 { 2290 int error = 0; 2291 char * from; 2292 char * to; 2293 struct dentry *dentry; 2294 struct nameidata nd; 2295 2296 from = getname(oldname); 2297 if(IS_ERR(from)) 2298 return PTR_ERR(from); 2299 to = getname(newname); 2300 error = PTR_ERR(to); 2301 if (IS_ERR(to)) 2302 goto out_putname; 2303 2304 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd); 2305 if (error) 2306 goto out; 2307 dentry = lookup_create(&nd, 0); 2308 error = PTR_ERR(dentry); 2309 if (IS_ERR(dentry)) 2310 goto out_unlock; 2311 2312 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO); 2313 dput(dentry); 2314 out_unlock: 2315 mutex_unlock(&nd.dentry->d_inode->i_mutex); 2316 path_release(&nd); 2317 out: 2318 putname(to); 2319 out_putname: 2320 putname(from); 2321 return error; 2322 } 2323 2324 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname) 2325 { 2326 return sys_symlinkat(oldname, AT_FDCWD, newname); 2327 } 2328 2329 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 2330 { 2331 struct inode *inode = old_dentry->d_inode; 2332 int error; 2333 2334 if (!inode) 2335 return -ENOENT; 2336 2337 error = may_create(dir, new_dentry, NULL); 2338 if (error) 2339 return error; 2340 2341 if (dir->i_sb != inode->i_sb) 2342 return -EXDEV; 2343 2344 /* 2345 * A link to an append-only or immutable file cannot be created. 2346 */ 2347 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 2348 return -EPERM; 2349 if (!dir->i_op || !dir->i_op->link) 2350 return -EPERM; 2351 if (S_ISDIR(old_dentry->d_inode->i_mode)) 2352 return -EPERM; 2353 2354 error = security_inode_link(old_dentry, dir, new_dentry); 2355 if (error) 2356 return error; 2357 2358 mutex_lock(&old_dentry->d_inode->i_mutex); 2359 DQUOT_INIT(dir); 2360 error = dir->i_op->link(old_dentry, dir, new_dentry); 2361 mutex_unlock(&old_dentry->d_inode->i_mutex); 2362 if (!error) 2363 fsnotify_create(dir, new_dentry); 2364 return error; 2365 } 2366 2367 /* 2368 * Hardlinks are often used in delicate situations. We avoid 2369 * security-related surprises by not following symlinks on the 2370 * newname. --KAB 2371 * 2372 * We don't follow them on the oldname either to be compatible 2373 * with linux 2.0, and to avoid hard-linking to directories 2374 * and other special files. --ADM 2375 */ 2376 asmlinkage long sys_linkat(int olddfd, const char __user *oldname, 2377 int newdfd, const char __user *newname, 2378 int flags) 2379 { 2380 struct dentry *new_dentry; 2381 struct nameidata nd, old_nd; 2382 int error; 2383 char * to; 2384 2385 if ((flags & ~AT_SYMLINK_FOLLOW) != 0) 2386 return -EINVAL; 2387 2388 to = getname(newname); 2389 if (IS_ERR(to)) 2390 return PTR_ERR(to); 2391 2392 error = __user_walk_fd(olddfd, oldname, 2393 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0, 2394 &old_nd); 2395 if (error) 2396 goto exit; 2397 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd); 2398 if (error) 2399 goto out; 2400 error = -EXDEV; 2401 if (old_nd.mnt != nd.mnt) 2402 goto out_release; 2403 new_dentry = lookup_create(&nd, 0); 2404 error = PTR_ERR(new_dentry); 2405 if (IS_ERR(new_dentry)) 2406 goto out_unlock; 2407 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry); 2408 dput(new_dentry); 2409 out_unlock: 2410 mutex_unlock(&nd.dentry->d_inode->i_mutex); 2411 out_release: 2412 path_release(&nd); 2413 out: 2414 path_release(&old_nd); 2415 exit: 2416 putname(to); 2417 2418 return error; 2419 } 2420 2421 asmlinkage long sys_link(const char __user *oldname, const char __user *newname) 2422 { 2423 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 2424 } 2425 2426 /* 2427 * The worst of all namespace operations - renaming directory. "Perverted" 2428 * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 2429 * Problems: 2430 * a) we can get into loop creation. Check is done in is_subdir(). 2431 * b) race potential - two innocent renames can create a loop together. 2432 * That's where 4.4 screws up. Current fix: serialization on 2433 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 2434 * story. 2435 * c) we have to lock _three_ objects - parents and victim (if it exists). 2436 * And that - after we got ->i_mutex on parents (until then we don't know 2437 * whether the target exists). Solution: try to be smart with locking 2438 * order for inodes. We rely on the fact that tree topology may change 2439 * only under ->s_vfs_rename_mutex _and_ that parent of the object we 2440 * move will be locked. Thus we can rank directories by the tree 2441 * (ancestors first) and rank all non-directories after them. 2442 * That works since everybody except rename does "lock parent, lookup, 2443 * lock child" and rename is under ->s_vfs_rename_mutex. 2444 * HOWEVER, it relies on the assumption that any object with ->lookup() 2445 * has no more than 1 dentry. If "hybrid" objects will ever appear, 2446 * we'd better make sure that there's no link(2) for them. 2447 * d) some filesystems don't support opened-but-unlinked directories, 2448 * either because of layout or because they are not ready to deal with 2449 * all cases correctly. The latter will be fixed (taking this sort of 2450 * stuff into VFS), but the former is not going away. Solution: the same 2451 * trick as in rmdir(). 2452 * e) conversion from fhandle to dentry may come in the wrong moment - when 2453 * we are removing the target. Solution: we will have to grab ->i_mutex 2454 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 2455 * ->i_mutex on parents, which works but leads to some truely excessive 2456 * locking]. 2457 */ 2458 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 2459 struct inode *new_dir, struct dentry *new_dentry) 2460 { 2461 int error = 0; 2462 struct inode *target; 2463 2464 /* 2465 * If we are going to change the parent - check write permissions, 2466 * we'll need to flip '..'. 2467 */ 2468 if (new_dir != old_dir) { 2469 error = permission(old_dentry->d_inode, MAY_WRITE, NULL); 2470 if (error) 2471 return error; 2472 } 2473 2474 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 2475 if (error) 2476 return error; 2477 2478 target = new_dentry->d_inode; 2479 if (target) { 2480 mutex_lock(&target->i_mutex); 2481 dentry_unhash(new_dentry); 2482 } 2483 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 2484 error = -EBUSY; 2485 else 2486 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 2487 if (target) { 2488 if (!error) 2489 target->i_flags |= S_DEAD; 2490 mutex_unlock(&target->i_mutex); 2491 if (d_unhashed(new_dentry)) 2492 d_rehash(new_dentry); 2493 dput(new_dentry); 2494 } 2495 if (!error) 2496 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 2497 d_move(old_dentry,new_dentry); 2498 return error; 2499 } 2500 2501 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 2502 struct inode *new_dir, struct dentry *new_dentry) 2503 { 2504 struct inode *target; 2505 int error; 2506 2507 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 2508 if (error) 2509 return error; 2510 2511 dget(new_dentry); 2512 target = new_dentry->d_inode; 2513 if (target) 2514 mutex_lock(&target->i_mutex); 2515 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 2516 error = -EBUSY; 2517 else 2518 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 2519 if (!error) { 2520 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 2521 d_move(old_dentry, new_dentry); 2522 } 2523 if (target) 2524 mutex_unlock(&target->i_mutex); 2525 dput(new_dentry); 2526 return error; 2527 } 2528 2529 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 2530 struct inode *new_dir, struct dentry *new_dentry) 2531 { 2532 int error; 2533 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 2534 const char *old_name; 2535 2536 if (old_dentry->d_inode == new_dentry->d_inode) 2537 return 0; 2538 2539 error = may_delete(old_dir, old_dentry, is_dir); 2540 if (error) 2541 return error; 2542 2543 if (!new_dentry->d_inode) 2544 error = may_create(new_dir, new_dentry, NULL); 2545 else 2546 error = may_delete(new_dir, new_dentry, is_dir); 2547 if (error) 2548 return error; 2549 2550 if (!old_dir->i_op || !old_dir->i_op->rename) 2551 return -EPERM; 2552 2553 DQUOT_INIT(old_dir); 2554 DQUOT_INIT(new_dir); 2555 2556 old_name = fsnotify_oldname_init(old_dentry->d_name.name); 2557 2558 if (is_dir) 2559 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 2560 else 2561 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 2562 if (!error) { 2563 const char *new_name = old_dentry->d_name.name; 2564 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir, 2565 new_dentry->d_inode, old_dentry); 2566 } 2567 fsnotify_oldname_free(old_name); 2568 2569 return error; 2570 } 2571 2572 static int do_rename(int olddfd, const char *oldname, 2573 int newdfd, const char *newname) 2574 { 2575 int error = 0; 2576 struct dentry * old_dir, * new_dir; 2577 struct dentry * old_dentry, *new_dentry; 2578 struct dentry * trap; 2579 struct nameidata oldnd, newnd; 2580 2581 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd); 2582 if (error) 2583 goto exit; 2584 2585 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd); 2586 if (error) 2587 goto exit1; 2588 2589 error = -EXDEV; 2590 if (oldnd.mnt != newnd.mnt) 2591 goto exit2; 2592 2593 old_dir = oldnd.dentry; 2594 error = -EBUSY; 2595 if (oldnd.last_type != LAST_NORM) 2596 goto exit2; 2597 2598 new_dir = newnd.dentry; 2599 if (newnd.last_type != LAST_NORM) 2600 goto exit2; 2601 2602 trap = lock_rename(new_dir, old_dir); 2603 2604 old_dentry = lookup_hash(&oldnd); 2605 error = PTR_ERR(old_dentry); 2606 if (IS_ERR(old_dentry)) 2607 goto exit3; 2608 /* source must exist */ 2609 error = -ENOENT; 2610 if (!old_dentry->d_inode) 2611 goto exit4; 2612 /* unless the source is a directory trailing slashes give -ENOTDIR */ 2613 if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 2614 error = -ENOTDIR; 2615 if (oldnd.last.name[oldnd.last.len]) 2616 goto exit4; 2617 if (newnd.last.name[newnd.last.len]) 2618 goto exit4; 2619 } 2620 /* source should not be ancestor of target */ 2621 error = -EINVAL; 2622 if (old_dentry == trap) 2623 goto exit4; 2624 new_dentry = lookup_hash(&newnd); 2625 error = PTR_ERR(new_dentry); 2626 if (IS_ERR(new_dentry)) 2627 goto exit4; 2628 /* target should not be an ancestor of source */ 2629 error = -ENOTEMPTY; 2630 if (new_dentry == trap) 2631 goto exit5; 2632 2633 error = vfs_rename(old_dir->d_inode, old_dentry, 2634 new_dir->d_inode, new_dentry); 2635 exit5: 2636 dput(new_dentry); 2637 exit4: 2638 dput(old_dentry); 2639 exit3: 2640 unlock_rename(new_dir, old_dir); 2641 exit2: 2642 path_release(&newnd); 2643 exit1: 2644 path_release(&oldnd); 2645 exit: 2646 return error; 2647 } 2648 2649 asmlinkage long sys_renameat(int olddfd, const char __user *oldname, 2650 int newdfd, const char __user *newname) 2651 { 2652 int error; 2653 char * from; 2654 char * to; 2655 2656 from = getname(oldname); 2657 if(IS_ERR(from)) 2658 return PTR_ERR(from); 2659 to = getname(newname); 2660 error = PTR_ERR(to); 2661 if (!IS_ERR(to)) { 2662 error = do_rename(olddfd, from, newdfd, to); 2663 putname(to); 2664 } 2665 putname(from); 2666 return error; 2667 } 2668 2669 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname) 2670 { 2671 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 2672 } 2673 2674 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 2675 { 2676 int len; 2677 2678 len = PTR_ERR(link); 2679 if (IS_ERR(link)) 2680 goto out; 2681 2682 len = strlen(link); 2683 if (len > (unsigned) buflen) 2684 len = buflen; 2685 if (copy_to_user(buffer, link, len)) 2686 len = -EFAULT; 2687 out: 2688 return len; 2689 } 2690 2691 /* 2692 * A helper for ->readlink(). This should be used *ONLY* for symlinks that 2693 * have ->follow_link() touching nd only in nd_set_link(). Using (or not 2694 * using) it for any given inode is up to filesystem. 2695 */ 2696 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 2697 { 2698 struct nameidata nd; 2699 void *cookie; 2700 2701 nd.depth = 0; 2702 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 2703 if (!IS_ERR(cookie)) { 2704 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 2705 if (dentry->d_inode->i_op->put_link) 2706 dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 2707 cookie = ERR_PTR(res); 2708 } 2709 return PTR_ERR(cookie); 2710 } 2711 2712 int vfs_follow_link(struct nameidata *nd, const char *link) 2713 { 2714 return __vfs_follow_link(nd, link); 2715 } 2716 2717 /* get the link contents into pagecache */ 2718 static char *page_getlink(struct dentry * dentry, struct page **ppage) 2719 { 2720 struct page * page; 2721 struct address_space *mapping = dentry->d_inode->i_mapping; 2722 page = read_mapping_page(mapping, 0, NULL); 2723 if (IS_ERR(page)) 2724 return (char*)page; 2725 *ppage = page; 2726 return kmap(page); 2727 } 2728 2729 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 2730 { 2731 struct page *page = NULL; 2732 char *s = page_getlink(dentry, &page); 2733 int res = vfs_readlink(dentry,buffer,buflen,s); 2734 if (page) { 2735 kunmap(page); 2736 page_cache_release(page); 2737 } 2738 return res; 2739 } 2740 2741 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 2742 { 2743 struct page *page = NULL; 2744 nd_set_link(nd, page_getlink(dentry, &page)); 2745 return page; 2746 } 2747 2748 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 2749 { 2750 struct page *page = cookie; 2751 2752 if (page) { 2753 kunmap(page); 2754 page_cache_release(page); 2755 } 2756 } 2757 2758 int __page_symlink(struct inode *inode, const char *symname, int len, 2759 gfp_t gfp_mask) 2760 { 2761 struct address_space *mapping = inode->i_mapping; 2762 struct page *page; 2763 void *fsdata; 2764 int err; 2765 char *kaddr; 2766 2767 retry: 2768 err = pagecache_write_begin(NULL, mapping, 0, len-1, 2769 AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata); 2770 if (err) 2771 goto fail; 2772 2773 kaddr = kmap_atomic(page, KM_USER0); 2774 memcpy(kaddr, symname, len-1); 2775 kunmap_atomic(kaddr, KM_USER0); 2776 2777 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 2778 page, fsdata); 2779 if (err < 0) 2780 goto fail; 2781 if (err < len-1) 2782 goto retry; 2783 2784 mark_inode_dirty(inode); 2785 return 0; 2786 fail: 2787 return err; 2788 } 2789 2790 int page_symlink(struct inode *inode, const char *symname, int len) 2791 { 2792 return __page_symlink(inode, symname, len, 2793 mapping_gfp_mask(inode->i_mapping)); 2794 } 2795 2796 const struct inode_operations page_symlink_inode_operations = { 2797 .readlink = generic_readlink, 2798 .follow_link = page_follow_link_light, 2799 .put_link = page_put_link, 2800 }; 2801 2802 EXPORT_SYMBOL(__user_walk); 2803 EXPORT_SYMBOL(__user_walk_fd); 2804 EXPORT_SYMBOL(follow_down); 2805 EXPORT_SYMBOL(follow_up); 2806 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ 2807 EXPORT_SYMBOL(getname); 2808 EXPORT_SYMBOL(lock_rename); 2809 EXPORT_SYMBOL(lookup_one_len); 2810 EXPORT_SYMBOL(page_follow_link_light); 2811 EXPORT_SYMBOL(page_put_link); 2812 EXPORT_SYMBOL(page_readlink); 2813 EXPORT_SYMBOL(__page_symlink); 2814 EXPORT_SYMBOL(page_symlink); 2815 EXPORT_SYMBOL(page_symlink_inode_operations); 2816 EXPORT_SYMBOL(path_lookup); 2817 EXPORT_SYMBOL(vfs_path_lookup); 2818 EXPORT_SYMBOL(path_release); 2819 EXPORT_SYMBOL(permission); 2820 EXPORT_SYMBOL(vfs_permission); 2821 EXPORT_SYMBOL(file_permission); 2822 EXPORT_SYMBOL(unlock_rename); 2823 EXPORT_SYMBOL(vfs_create); 2824 EXPORT_SYMBOL(vfs_follow_link); 2825 EXPORT_SYMBOL(vfs_link); 2826 EXPORT_SYMBOL(vfs_mkdir); 2827 EXPORT_SYMBOL(vfs_mknod); 2828 EXPORT_SYMBOL(generic_permission); 2829 EXPORT_SYMBOL(vfs_readlink); 2830 EXPORT_SYMBOL(vfs_rename); 2831 EXPORT_SYMBOL(vfs_rmdir); 2832 EXPORT_SYMBOL(vfs_symlink); 2833 EXPORT_SYMBOL(vfs_unlink); 2834 EXPORT_SYMBOL(dentry_unhash); 2835 EXPORT_SYMBOL(generic_readlink); 2836