4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.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>
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).
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.
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.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
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.
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.
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.
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.
90 * [10-Sep-98 Alan Modra] Another symlink change.
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).
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...
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...
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..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 void final_putname(struct filename *name)
122 if (name->separate) {
123 __putname(name->name);
130 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
132 static struct filename *
133 getname_flags(const char __user *filename, int flags, int *empty)
135 struct filename *result, *err;
140 result = audit_reusename(filename);
144 result = __getname();
145 if (unlikely(!result))
146 return ERR_PTR(-ENOMEM);
149 * First, try to embed the struct filename inside the names_cache
152 kname = (char *)result + sizeof(*result);
153 result->name = kname;
154 result->separate = false;
155 max = EMBEDDED_NAME_MAX;
158 len = strncpy_from_user(kname, filename, max);
159 if (unlikely(len < 0)) {
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
170 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
171 kname = (char *)result;
173 result = kzalloc(sizeof(*result), GFP_KERNEL);
175 err = ERR_PTR(-ENOMEM);
176 result = (struct filename *)kname;
179 result->name = kname;
180 result->separate = true;
185 /* The empty path is special. */
186 if (unlikely(!len)) {
189 err = ERR_PTR(-ENOENT);
190 if (!(flags & LOOKUP_EMPTY))
194 err = ERR_PTR(-ENAMETOOLONG);
195 if (unlikely(len >= PATH_MAX))
198 result->uptr = filename;
199 audit_getname(result);
203 final_putname(result);
208 getname(const char __user * filename)
210 return getname_flags(filename, 0, NULL);
212 EXPORT_SYMBOL(getname);
214 #ifdef CONFIG_AUDITSYSCALL
215 void putname(struct filename *name)
217 if (unlikely(!audit_dummy_context()))
218 return audit_putname(name);
223 static int check_acl(struct inode *inode, int mask)
225 #ifdef CONFIG_FS_POSIX_ACL
226 struct posix_acl *acl;
228 if (mask & MAY_NOT_BLOCK) {
229 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
232 /* no ->get_acl() calls in RCU mode... */
233 if (acl == ACL_NOT_CACHED)
235 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
238 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
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.
245 * If the filesystem doesn't have a get_acl() function at all, we'll
246 * just create the negative cache entry.
248 if (acl == ACL_NOT_CACHED) {
249 if (inode->i_op->get_acl) {
250 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
254 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
260 int error = posix_acl_permission(inode, acl, mask);
261 posix_acl_release(acl);
270 * This does the basic permission checking
272 static int acl_permission_check(struct inode *inode, int mask)
274 unsigned int mode = inode->i_mode;
276 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
279 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
280 int error = check_acl(inode, mask);
281 if (error != -EAGAIN)
285 if (in_group_p(inode->i_gid))
290 * If the DACs are ok we don't need any capability check.
292 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
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, ...)
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.
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.
311 int generic_permission(struct inode *inode, int mask)
316 * Do the basic permission checks.
318 ret = acl_permission_check(inode, mask);
322 if (S_ISDIR(inode->i_mode)) {
323 /* DACs are overridable for directories */
324 if (inode_capable(inode, CAP_DAC_OVERRIDE))
326 if (!(mask & MAY_WRITE))
327 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
332 * Read/write DACs are always overridable.
333 * Executable DACs are overridable when there is
334 * at least one exec bit set.
336 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
337 if (inode_capable(inode, CAP_DAC_OVERRIDE))
341 * Searching includes executable on directories, else just read.
343 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
344 if (mask == MAY_READ)
345 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
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".
357 static inline int do_inode_permission(struct inode *inode, int mask)
359 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
360 if (likely(inode->i_op->permission))
361 return inode->i_op->permission(inode, mask);
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);
368 return generic_permission(inode, mask);
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)
376 * Check for read/write/execute permissions on an inode.
378 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
380 * This does not check for a read-only file system. You probably want
381 * inode_permission().
383 int __inode_permission(struct inode *inode, int mask)
387 if (unlikely(mask & MAY_WRITE)) {
389 * Nobody gets write access to an immutable file.
391 if (IS_IMMUTABLE(inode))
395 retval = do_inode_permission(inode, mask);
399 retval = devcgroup_inode_permission(inode, mask);
403 return security_inode_permission(inode, mask);
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)
412 * Separate out file-system wide checks from inode-specific permission checks.
414 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
416 if (unlikely(mask & MAY_WRITE)) {
417 umode_t mode = inode->i_mode;
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)))
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)
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.
436 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
438 int inode_permission(struct inode *inode, int mask)
442 retval = sb_permission(inode->i_sb, inode, mask);
445 return __inode_permission(inode, mask);
449 * path_get - get a reference to a path
450 * @path: path to get the reference to
452 * Given a path increment the reference count to the dentry and the vfsmount.
454 void path_get(const struct path *path)
459 EXPORT_SYMBOL(path_get);
462 * path_put - put a reference to a path
463 * @path: path to put the reference to
465 * Given a path decrement the reference count to the dentry and the vfsmount.
467 void path_put(const struct path *path)
472 EXPORT_SYMBOL(path_put);
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.
485 static inline void lock_rcu_walk(void)
487 br_read_lock(&vfsmount_lock);
491 static inline void unlock_rcu_walk(void)
494 br_read_unlock(&vfsmount_lock);
498 * When we move over from the RCU domain to properly refcounted
499 * long-lived dentries, we need to check the sequence numbers
500 * we got before lookup very carefully.
502 * We cannot blindly increment a dentry refcount - even if it
503 * is not locked - if it is zero, because it may have gone
504 * through the final d_kill() logic already.
506 * So for a zero refcount, we need to get the spinlock (which is
507 * safe even for a dead dentry because the de-allocation is
508 * RCU-delayed), and check the sequence count under the lock.
510 * Once we have checked the sequence count, we know it is live,
511 * and since we hold the spinlock it cannot die from under us.
513 * In contrast, if the reference count wasn't zero, we can just
514 * increment the lockref without having to take the spinlock.
515 * Even if the sequence number ends up being stale, we haven't
516 * gone through the final dput() and killed the dentry yet.
518 static inline int d_rcu_to_refcount(struct dentry *dentry, seqcount_t *validate, unsigned seq)
522 gotref = lockref_get_or_lock(&dentry->d_lockref);
524 /* Does the sequence number still match? */
525 if (read_seqcount_retry(validate, seq)) {
529 spin_unlock(&dentry->d_lock);
533 /* Get the ref now, if we couldn't get it originally */
535 dentry->d_lockref.count++;
536 spin_unlock(&dentry->d_lock);
542 * unlazy_walk - try to switch to ref-walk mode.
543 * @nd: nameidata pathwalk data
544 * @dentry: child of nd->path.dentry or NULL
545 * Returns: 0 on success, -ECHILD on failure
547 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
548 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
549 * @nd or NULL. Must be called from rcu-walk context.
551 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
553 struct fs_struct *fs = current->fs;
554 struct dentry *parent = nd->path.dentry;
557 BUG_ON(!(nd->flags & LOOKUP_RCU));
558 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
560 spin_lock(&fs->lock);
561 if (nd->root.mnt != fs->root.mnt ||
562 nd->root.dentry != fs->root.dentry)
567 * For a negative lookup, the lookup sequence point is the parents
568 * sequence point, and it only needs to revalidate the parent dentry.
570 * For a positive lookup, we need to move both the parent and the
571 * dentry from the RCU domain to be properly refcounted. And the
572 * sequence number in the dentry validates *both* dentry counters,
573 * since we checked the sequence number of the parent after we got
574 * the child sequence number. So we know the parent must still
575 * be valid if the child sequence number is still valid.
578 if (d_rcu_to_refcount(parent, &parent->d_seq, nd->seq) < 0)
580 BUG_ON(nd->inode != parent->d_inode);
582 if (d_rcu_to_refcount(dentry, &dentry->d_seq, nd->seq) < 0)
584 if (d_rcu_to_refcount(parent, &dentry->d_seq, nd->seq) < 0)
589 spin_unlock(&fs->lock);
591 mntget(nd->path.mnt);
594 nd->flags &= ~LOOKUP_RCU;
601 spin_unlock(&fs->lock);
605 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
607 return dentry->d_op->d_revalidate(dentry, flags);
611 * complete_walk - successful completion of path walk
612 * @nd: pointer nameidata
614 * If we had been in RCU mode, drop out of it and legitimize nd->path.
615 * Revalidate the final result, unless we'd already done that during
616 * the path walk or the filesystem doesn't ask for it. Return 0 on
617 * success, -error on failure. In case of failure caller does not
618 * need to drop nd->path.
620 static int complete_walk(struct nameidata *nd)
622 struct dentry *dentry = nd->path.dentry;
625 if (nd->flags & LOOKUP_RCU) {
626 nd->flags &= ~LOOKUP_RCU;
627 if (!(nd->flags & LOOKUP_ROOT))
630 if (d_rcu_to_refcount(dentry, &dentry->d_seq, nd->seq) < 0) {
634 mntget(nd->path.mnt);
638 if (likely(!(nd->flags & LOOKUP_JUMPED)))
641 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
644 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
655 static __always_inline void set_root(struct nameidata *nd)
658 get_fs_root(current->fs, &nd->root);
661 static int link_path_walk(const char *, struct nameidata *);
663 static __always_inline void set_root_rcu(struct nameidata *nd)
666 struct fs_struct *fs = current->fs;
670 seq = read_seqcount_begin(&fs->seq);
672 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
673 } while (read_seqcount_retry(&fs->seq, seq));
677 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
689 nd->flags |= LOOKUP_JUMPED;
691 nd->inode = nd->path.dentry->d_inode;
693 ret = link_path_walk(link, nd);
697 return PTR_ERR(link);
700 static void path_put_conditional(struct path *path, struct nameidata *nd)
703 if (path->mnt != nd->path.mnt)
707 static inline void path_to_nameidata(const struct path *path,
708 struct nameidata *nd)
710 if (!(nd->flags & LOOKUP_RCU)) {
711 dput(nd->path.dentry);
712 if (nd->path.mnt != path->mnt)
713 mntput(nd->path.mnt);
715 nd->path.mnt = path->mnt;
716 nd->path.dentry = path->dentry;
720 * Helper to directly jump to a known parsed path from ->follow_link,
721 * caller must have taken a reference to path beforehand.
723 void nd_jump_link(struct nameidata *nd, struct path *path)
728 nd->inode = nd->path.dentry->d_inode;
729 nd->flags |= LOOKUP_JUMPED;
732 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
734 struct inode *inode = link->dentry->d_inode;
735 if (inode->i_op->put_link)
736 inode->i_op->put_link(link->dentry, nd, cookie);
740 int sysctl_protected_symlinks __read_mostly = 0;
741 int sysctl_protected_hardlinks __read_mostly = 0;
744 * may_follow_link - Check symlink following for unsafe situations
745 * @link: The path of the symlink
746 * @nd: nameidata pathwalk data
748 * In the case of the sysctl_protected_symlinks sysctl being enabled,
749 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
750 * in a sticky world-writable directory. This is to protect privileged
751 * processes from failing races against path names that may change out
752 * from under them by way of other users creating malicious symlinks.
753 * It will permit symlinks to be followed only when outside a sticky
754 * world-writable directory, or when the uid of the symlink and follower
755 * match, or when the directory owner matches the symlink's owner.
757 * Returns 0 if following the symlink is allowed, -ve on error.
759 static inline int may_follow_link(struct path *link, struct nameidata *nd)
761 const struct inode *inode;
762 const struct inode *parent;
764 if (!sysctl_protected_symlinks)
767 /* Allowed if owner and follower match. */
768 inode = link->dentry->d_inode;
769 if (uid_eq(current_cred()->fsuid, inode->i_uid))
772 /* Allowed if parent directory not sticky and world-writable. */
773 parent = nd->path.dentry->d_inode;
774 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
777 /* Allowed if parent directory and link owner match. */
778 if (uid_eq(parent->i_uid, inode->i_uid))
781 audit_log_link_denied("follow_link", link);
782 path_put_conditional(link, nd);
788 * safe_hardlink_source - Check for safe hardlink conditions
789 * @inode: the source inode to hardlink from
791 * Return false if at least one of the following conditions:
792 * - inode is not a regular file
794 * - inode is setgid and group-exec
795 * - access failure for read and write
797 * Otherwise returns true.
799 static bool safe_hardlink_source(struct inode *inode)
801 umode_t mode = inode->i_mode;
803 /* Special files should not get pinned to the filesystem. */
807 /* Setuid files should not get pinned to the filesystem. */
811 /* Executable setgid files should not get pinned to the filesystem. */
812 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
815 /* Hardlinking to unreadable or unwritable sources is dangerous. */
816 if (inode_permission(inode, MAY_READ | MAY_WRITE))
823 * may_linkat - Check permissions for creating a hardlink
824 * @link: the source to hardlink from
826 * Block hardlink when all of:
827 * - sysctl_protected_hardlinks enabled
828 * - fsuid does not match inode
829 * - hardlink source is unsafe (see safe_hardlink_source() above)
832 * Returns 0 if successful, -ve on error.
834 static int may_linkat(struct path *link)
836 const struct cred *cred;
839 if (!sysctl_protected_hardlinks)
842 cred = current_cred();
843 inode = link->dentry->d_inode;
845 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
846 * otherwise, it must be a safe source.
848 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
852 audit_log_link_denied("linkat", link);
856 static __always_inline int
857 follow_link(struct path *link, struct nameidata *nd, void **p)
859 struct dentry *dentry = link->dentry;
863 BUG_ON(nd->flags & LOOKUP_RCU);
865 if (link->mnt == nd->path.mnt)
869 if (unlikely(current->total_link_count >= 40))
870 goto out_put_nd_path;
873 current->total_link_count++;
876 nd_set_link(nd, NULL);
878 error = security_inode_follow_link(link->dentry, nd);
880 goto out_put_nd_path;
882 nd->last_type = LAST_BIND;
883 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
886 goto out_put_nd_path;
891 error = __vfs_follow_link(nd, s);
893 put_link(nd, link, *p);
905 static int follow_up_rcu(struct path *path)
907 struct mount *mnt = real_mount(path->mnt);
908 struct mount *parent;
909 struct dentry *mountpoint;
911 parent = mnt->mnt_parent;
912 if (&parent->mnt == path->mnt)
914 mountpoint = mnt->mnt_mountpoint;
915 path->dentry = mountpoint;
916 path->mnt = &parent->mnt;
921 * follow_up - Find the mountpoint of path's vfsmount
923 * Given a path, find the mountpoint of its source file system.
924 * Replace @path with the path of the mountpoint in the parent mount.
927 * Return 1 if we went up a level and 0 if we were already at the
930 int follow_up(struct path *path)
932 struct mount *mnt = real_mount(path->mnt);
933 struct mount *parent;
934 struct dentry *mountpoint;
936 br_read_lock(&vfsmount_lock);
937 parent = mnt->mnt_parent;
939 br_read_unlock(&vfsmount_lock);
942 mntget(&parent->mnt);
943 mountpoint = dget(mnt->mnt_mountpoint);
944 br_read_unlock(&vfsmount_lock);
946 path->dentry = mountpoint;
948 path->mnt = &parent->mnt;
953 * Perform an automount
954 * - return -EISDIR to tell follow_managed() to stop and return the path we
957 static int follow_automount(struct path *path, unsigned flags,
960 struct vfsmount *mnt;
963 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
966 /* We don't want to mount if someone's just doing a stat -
967 * unless they're stat'ing a directory and appended a '/' to
970 * We do, however, want to mount if someone wants to open or
971 * create a file of any type under the mountpoint, wants to
972 * traverse through the mountpoint or wants to open the
973 * mounted directory. Also, autofs may mark negative dentries
974 * as being automount points. These will need the attentions
975 * of the daemon to instantiate them before they can be used.
977 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
978 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
979 path->dentry->d_inode)
982 current->total_link_count++;
983 if (current->total_link_count >= 40)
986 mnt = path->dentry->d_op->d_automount(path);
989 * The filesystem is allowed to return -EISDIR here to indicate
990 * it doesn't want to automount. For instance, autofs would do
991 * this so that its userspace daemon can mount on this dentry.
993 * However, we can only permit this if it's a terminal point in
994 * the path being looked up; if it wasn't then the remainder of
995 * the path is inaccessible and we should say so.
997 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
1002 if (!mnt) /* mount collision */
1005 if (!*need_mntput) {
1006 /* lock_mount() may release path->mnt on error */
1008 *need_mntput = true;
1010 err = finish_automount(mnt, path);
1014 /* Someone else made a mount here whilst we were busy */
1019 path->dentry = dget(mnt->mnt_root);
1028 * Handle a dentry that is managed in some way.
1029 * - Flagged for transit management (autofs)
1030 * - Flagged as mountpoint
1031 * - Flagged as automount point
1033 * This may only be called in refwalk mode.
1035 * Serialization is taken care of in namespace.c
1037 static int follow_managed(struct path *path, unsigned flags)
1039 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1041 bool need_mntput = false;
1044 /* Given that we're not holding a lock here, we retain the value in a
1045 * local variable for each dentry as we look at it so that we don't see
1046 * the components of that value change under us */
1047 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1048 managed &= DCACHE_MANAGED_DENTRY,
1049 unlikely(managed != 0)) {
1050 /* Allow the filesystem to manage the transit without i_mutex
1052 if (managed & DCACHE_MANAGE_TRANSIT) {
1053 BUG_ON(!path->dentry->d_op);
1054 BUG_ON(!path->dentry->d_op->d_manage);
1055 ret = path->dentry->d_op->d_manage(path->dentry, false);
1060 /* Transit to a mounted filesystem. */
1061 if (managed & DCACHE_MOUNTED) {
1062 struct vfsmount *mounted = lookup_mnt(path);
1067 path->mnt = mounted;
1068 path->dentry = dget(mounted->mnt_root);
1073 /* Something is mounted on this dentry in another
1074 * namespace and/or whatever was mounted there in this
1075 * namespace got unmounted before we managed to get the
1079 /* Handle an automount point */
1080 if (managed & DCACHE_NEED_AUTOMOUNT) {
1081 ret = follow_automount(path, flags, &need_mntput);
1087 /* We didn't change the current path point */
1091 if (need_mntput && path->mnt == mnt)
1095 return ret < 0 ? ret : need_mntput;
1098 int follow_down_one(struct path *path)
1100 struct vfsmount *mounted;
1102 mounted = lookup_mnt(path);
1106 path->mnt = mounted;
1107 path->dentry = dget(mounted->mnt_root);
1113 static inline bool managed_dentry_might_block(struct dentry *dentry)
1115 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1116 dentry->d_op->d_manage(dentry, true) < 0);
1120 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1121 * we meet a managed dentry that would need blocking.
1123 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1124 struct inode **inode)
1127 struct mount *mounted;
1129 * Don't forget we might have a non-mountpoint managed dentry
1130 * that wants to block transit.
1132 if (unlikely(managed_dentry_might_block(path->dentry)))
1135 if (!d_mountpoint(path->dentry))
1138 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1141 path->mnt = &mounted->mnt;
1142 path->dentry = mounted->mnt.mnt_root;
1143 nd->flags |= LOOKUP_JUMPED;
1144 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1146 * Update the inode too. We don't need to re-check the
1147 * dentry sequence number here after this d_inode read,
1148 * because a mount-point is always pinned.
1150 *inode = path->dentry->d_inode;
1155 static void follow_mount_rcu(struct nameidata *nd)
1157 while (d_mountpoint(nd->path.dentry)) {
1158 struct mount *mounted;
1159 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
1162 nd->path.mnt = &mounted->mnt;
1163 nd->path.dentry = mounted->mnt.mnt_root;
1164 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1168 static int follow_dotdot_rcu(struct nameidata *nd)
1173 if (nd->path.dentry == nd->root.dentry &&
1174 nd->path.mnt == nd->root.mnt) {
1177 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1178 struct dentry *old = nd->path.dentry;
1179 struct dentry *parent = old->d_parent;
1182 seq = read_seqcount_begin(&parent->d_seq);
1183 if (read_seqcount_retry(&old->d_seq, nd->seq))
1185 nd->path.dentry = parent;
1189 if (!follow_up_rcu(&nd->path))
1191 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1193 follow_mount_rcu(nd);
1194 nd->inode = nd->path.dentry->d_inode;
1198 nd->flags &= ~LOOKUP_RCU;
1199 if (!(nd->flags & LOOKUP_ROOT))
1200 nd->root.mnt = NULL;
1206 * Follow down to the covering mount currently visible to userspace. At each
1207 * point, the filesystem owning that dentry may be queried as to whether the
1208 * caller is permitted to proceed or not.
1210 int follow_down(struct path *path)
1215 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1216 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1217 /* Allow the filesystem to manage the transit without i_mutex
1220 * We indicate to the filesystem if someone is trying to mount
1221 * something here. This gives autofs the chance to deny anyone
1222 * other than its daemon the right to mount on its
1225 * The filesystem may sleep at this point.
1227 if (managed & DCACHE_MANAGE_TRANSIT) {
1228 BUG_ON(!path->dentry->d_op);
1229 BUG_ON(!path->dentry->d_op->d_manage);
1230 ret = path->dentry->d_op->d_manage(
1231 path->dentry, false);
1233 return ret == -EISDIR ? 0 : ret;
1236 /* Transit to a mounted filesystem. */
1237 if (managed & DCACHE_MOUNTED) {
1238 struct vfsmount *mounted = lookup_mnt(path);
1243 path->mnt = mounted;
1244 path->dentry = dget(mounted->mnt_root);
1248 /* Don't handle automount points here */
1255 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1257 static void follow_mount(struct path *path)
1259 while (d_mountpoint(path->dentry)) {
1260 struct vfsmount *mounted = lookup_mnt(path);
1265 path->mnt = mounted;
1266 path->dentry = dget(mounted->mnt_root);
1270 static void follow_dotdot(struct nameidata *nd)
1275 struct dentry *old = nd->path.dentry;
1277 if (nd->path.dentry == nd->root.dentry &&
1278 nd->path.mnt == nd->root.mnt) {
1281 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1282 /* rare case of legitimate dget_parent()... */
1283 nd->path.dentry = dget_parent(nd->path.dentry);
1287 if (!follow_up(&nd->path))
1290 follow_mount(&nd->path);
1291 nd->inode = nd->path.dentry->d_inode;
1295 * This looks up the name in dcache, possibly revalidates the old dentry and
1296 * allocates a new one if not found or not valid. In the need_lookup argument
1297 * returns whether i_op->lookup is necessary.
1299 * dir->d_inode->i_mutex must be held
1301 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1302 unsigned int flags, bool *need_lookup)
1304 struct dentry *dentry;
1307 *need_lookup = false;
1308 dentry = d_lookup(dir, name);
1310 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1311 error = d_revalidate(dentry, flags);
1312 if (unlikely(error <= 0)) {
1315 return ERR_PTR(error);
1316 } else if (!d_invalidate(dentry)) {
1325 dentry = d_alloc(dir, name);
1326 if (unlikely(!dentry))
1327 return ERR_PTR(-ENOMEM);
1329 *need_lookup = true;
1335 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1336 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1338 * dir->d_inode->i_mutex must be held
1340 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1345 /* Don't create child dentry for a dead directory. */
1346 if (unlikely(IS_DEADDIR(dir))) {
1348 return ERR_PTR(-ENOENT);
1351 old = dir->i_op->lookup(dir, dentry, flags);
1352 if (unlikely(old)) {
1359 static struct dentry *__lookup_hash(struct qstr *name,
1360 struct dentry *base, unsigned int flags)
1363 struct dentry *dentry;
1365 dentry = lookup_dcache(name, base, flags, &need_lookup);
1369 return lookup_real(base->d_inode, dentry, flags);
1373 * It's more convoluted than I'd like it to be, but... it's still fairly
1374 * small and for now I'd prefer to have fast path as straight as possible.
1375 * It _is_ time-critical.
1377 static int lookup_fast(struct nameidata *nd,
1378 struct path *path, struct inode **inode)
1380 struct vfsmount *mnt = nd->path.mnt;
1381 struct dentry *dentry, *parent = nd->path.dentry;
1387 * Rename seqlock is not required here because in the off chance
1388 * of a false negative due to a concurrent rename, we're going to
1389 * do the non-racy lookup, below.
1391 if (nd->flags & LOOKUP_RCU) {
1393 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1398 * This sequence count validates that the inode matches
1399 * the dentry name information from lookup.
1401 *inode = dentry->d_inode;
1402 if (read_seqcount_retry(&dentry->d_seq, seq))
1406 * This sequence count validates that the parent had no
1407 * changes while we did the lookup of the dentry above.
1409 * The memory barrier in read_seqcount_begin of child is
1410 * enough, we can use __read_seqcount_retry here.
1412 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1416 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1417 status = d_revalidate(dentry, nd->flags);
1418 if (unlikely(status <= 0)) {
1419 if (status != -ECHILD)
1425 path->dentry = dentry;
1426 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1428 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1432 if (unlazy_walk(nd, dentry))
1435 dentry = __d_lookup(parent, &nd->last);
1438 if (unlikely(!dentry))
1441 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1442 status = d_revalidate(dentry, nd->flags);
1443 if (unlikely(status <= 0)) {
1448 if (!d_invalidate(dentry)) {
1455 path->dentry = dentry;
1456 err = follow_managed(path, nd->flags);
1457 if (unlikely(err < 0)) {
1458 path_put_conditional(path, nd);
1462 nd->flags |= LOOKUP_JUMPED;
1463 *inode = path->dentry->d_inode;
1470 /* Fast lookup failed, do it the slow way */
1471 static int lookup_slow(struct nameidata *nd, struct path *path)
1473 struct dentry *dentry, *parent;
1476 parent = nd->path.dentry;
1477 BUG_ON(nd->inode != parent->d_inode);
1479 mutex_lock(&parent->d_inode->i_mutex);
1480 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1481 mutex_unlock(&parent->d_inode->i_mutex);
1483 return PTR_ERR(dentry);
1484 path->mnt = nd->path.mnt;
1485 path->dentry = dentry;
1486 err = follow_managed(path, nd->flags);
1487 if (unlikely(err < 0)) {
1488 path_put_conditional(path, nd);
1492 nd->flags |= LOOKUP_JUMPED;
1496 static inline int may_lookup(struct nameidata *nd)
1498 if (nd->flags & LOOKUP_RCU) {
1499 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1502 if (unlazy_walk(nd, NULL))
1505 return inode_permission(nd->inode, MAY_EXEC);
1508 static inline int handle_dots(struct nameidata *nd, int type)
1510 if (type == LAST_DOTDOT) {
1511 if (nd->flags & LOOKUP_RCU) {
1512 if (follow_dotdot_rcu(nd))
1520 static void terminate_walk(struct nameidata *nd)
1522 if (!(nd->flags & LOOKUP_RCU)) {
1523 path_put(&nd->path);
1525 nd->flags &= ~LOOKUP_RCU;
1526 if (!(nd->flags & LOOKUP_ROOT))
1527 nd->root.mnt = NULL;
1533 * Do we need to follow links? We _really_ want to be able
1534 * to do this check without having to look at inode->i_op,
1535 * so we keep a cache of "no, this doesn't need follow_link"
1536 * for the common case.
1538 static inline int should_follow_link(struct inode *inode, int follow)
1540 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1541 if (likely(inode->i_op->follow_link))
1544 /* This gets set once for the inode lifetime */
1545 spin_lock(&inode->i_lock);
1546 inode->i_opflags |= IOP_NOFOLLOW;
1547 spin_unlock(&inode->i_lock);
1552 static inline int walk_component(struct nameidata *nd, struct path *path,
1555 struct inode *inode;
1558 * "." and ".." are special - ".." especially so because it has
1559 * to be able to know about the current root directory and
1560 * parent relationships.
1562 if (unlikely(nd->last_type != LAST_NORM))
1563 return handle_dots(nd, nd->last_type);
1564 err = lookup_fast(nd, path, &inode);
1565 if (unlikely(err)) {
1569 err = lookup_slow(nd, path);
1573 inode = path->dentry->d_inode;
1579 if (should_follow_link(inode, follow)) {
1580 if (nd->flags & LOOKUP_RCU) {
1581 if (unlikely(unlazy_walk(nd, path->dentry))) {
1586 BUG_ON(inode != path->dentry->d_inode);
1589 path_to_nameidata(path, nd);
1594 path_to_nameidata(path, nd);
1601 * This limits recursive symlink follows to 8, while
1602 * limiting consecutive symlinks to 40.
1604 * Without that kind of total limit, nasty chains of consecutive
1605 * symlinks can cause almost arbitrarily long lookups.
1607 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1611 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1612 path_put_conditional(path, nd);
1613 path_put(&nd->path);
1616 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1619 current->link_count++;
1622 struct path link = *path;
1625 res = follow_link(&link, nd, &cookie);
1628 res = walk_component(nd, path, LOOKUP_FOLLOW);
1629 put_link(nd, &link, cookie);
1632 current->link_count--;
1638 * We really don't want to look at inode->i_op->lookup
1639 * when we don't have to. So we keep a cache bit in
1640 * the inode ->i_opflags field that says "yes, we can
1641 * do lookup on this inode".
1643 static inline int can_lookup(struct inode *inode)
1645 if (likely(inode->i_opflags & IOP_LOOKUP))
1647 if (likely(!inode->i_op->lookup))
1650 /* We do this once for the lifetime of the inode */
1651 spin_lock(&inode->i_lock);
1652 inode->i_opflags |= IOP_LOOKUP;
1653 spin_unlock(&inode->i_lock);
1658 * We can do the critical dentry name comparison and hashing
1659 * operations one word at a time, but we are limited to:
1661 * - Architectures with fast unaligned word accesses. We could
1662 * do a "get_unaligned()" if this helps and is sufficiently
1665 * - Little-endian machines (so that we can generate the mask
1666 * of low bytes efficiently). Again, we *could* do a byte
1667 * swapping load on big-endian architectures if that is not
1668 * expensive enough to make the optimization worthless.
1670 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1671 * do not trap on the (extremely unlikely) case of a page
1672 * crossing operation.
1674 * - Furthermore, we need an efficient 64-bit compile for the
1675 * 64-bit case in order to generate the "number of bytes in
1676 * the final mask". Again, that could be replaced with a
1677 * efficient population count instruction or similar.
1679 #ifdef CONFIG_DCACHE_WORD_ACCESS
1681 #include <asm/word-at-a-time.h>
1685 static inline unsigned int fold_hash(unsigned long hash)
1687 hash += hash >> (8*sizeof(int));
1691 #else /* 32-bit case */
1693 #define fold_hash(x) (x)
1697 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1699 unsigned long a, mask;
1700 unsigned long hash = 0;
1703 a = load_unaligned_zeropad(name);
1704 if (len < sizeof(unsigned long))
1708 name += sizeof(unsigned long);
1709 len -= sizeof(unsigned long);
1713 mask = ~(~0ul << len*8);
1716 return fold_hash(hash);
1718 EXPORT_SYMBOL(full_name_hash);
1721 * Calculate the length and hash of the path component, and
1722 * return the length of the component;
1724 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1726 unsigned long a, b, adata, bdata, mask, hash, len;
1727 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1730 len = -sizeof(unsigned long);
1732 hash = (hash + a) * 9;
1733 len += sizeof(unsigned long);
1734 a = load_unaligned_zeropad(name+len);
1735 b = a ^ REPEAT_BYTE('/');
1736 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1738 adata = prep_zero_mask(a, adata, &constants);
1739 bdata = prep_zero_mask(b, bdata, &constants);
1741 mask = create_zero_mask(adata | bdata);
1743 hash += a & zero_bytemask(mask);
1744 *hashp = fold_hash(hash);
1746 return len + find_zero(mask);
1751 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1753 unsigned long hash = init_name_hash();
1755 hash = partial_name_hash(*name++, hash);
1756 return end_name_hash(hash);
1758 EXPORT_SYMBOL(full_name_hash);
1761 * We know there's a real path component here of at least
1764 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1766 unsigned long hash = init_name_hash();
1767 unsigned long len = 0, c;
1769 c = (unsigned char)*name;
1772 hash = partial_name_hash(c, hash);
1773 c = (unsigned char)name[len];
1774 } while (c && c != '/');
1775 *hashp = end_name_hash(hash);
1783 * This is the basic name resolution function, turning a pathname into
1784 * the final dentry. We expect 'base' to be positive and a directory.
1786 * Returns 0 and nd will have valid dentry and mnt on success.
1787 * Returns error and drops reference to input namei data on failure.
1789 static int link_path_walk(const char *name, struct nameidata *nd)
1799 /* At this point we know we have a real path component. */
1805 err = may_lookup(nd);
1809 len = hash_name(name, &this.hash);
1814 if (name[0] == '.') switch (len) {
1816 if (name[1] == '.') {
1818 nd->flags |= LOOKUP_JUMPED;
1824 if (likely(type == LAST_NORM)) {
1825 struct dentry *parent = nd->path.dentry;
1826 nd->flags &= ~LOOKUP_JUMPED;
1827 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1828 err = parent->d_op->d_hash(parent, &this);
1835 nd->last_type = type;
1840 * If it wasn't NUL, we know it was '/'. Skip that
1841 * slash, and continue until no more slashes.
1845 } while (unlikely(name[len] == '/'));
1851 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1856 err = nested_symlink(&next, nd);
1860 if (!can_lookup(nd->inode)) {
1869 static int path_init(int dfd, const char *name, unsigned int flags,
1870 struct nameidata *nd, struct file **fp)
1874 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1875 nd->flags = flags | LOOKUP_JUMPED;
1877 if (flags & LOOKUP_ROOT) {
1878 struct inode *inode = nd->root.dentry->d_inode;
1880 if (!can_lookup(inode))
1882 retval = inode_permission(inode, MAY_EXEC);
1886 nd->path = nd->root;
1888 if (flags & LOOKUP_RCU) {
1890 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1892 path_get(&nd->path);
1897 nd->root.mnt = NULL;
1900 if (flags & LOOKUP_RCU) {
1905 path_get(&nd->root);
1907 nd->path = nd->root;
1908 } else if (dfd == AT_FDCWD) {
1909 if (flags & LOOKUP_RCU) {
1910 struct fs_struct *fs = current->fs;
1916 seq = read_seqcount_begin(&fs->seq);
1918 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1919 } while (read_seqcount_retry(&fs->seq, seq));
1921 get_fs_pwd(current->fs, &nd->path);
1924 /* Caller must check execute permissions on the starting path component */
1925 struct fd f = fdget_raw(dfd);
1926 struct dentry *dentry;
1931 dentry = f.file->f_path.dentry;
1934 if (!can_lookup(dentry->d_inode)) {
1940 nd->path = f.file->f_path;
1941 if (flags & LOOKUP_RCU) {
1944 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1947 path_get(&nd->path);
1952 nd->inode = nd->path.dentry->d_inode;
1956 static inline int lookup_last(struct nameidata *nd, struct path *path)
1958 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1959 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1961 nd->flags &= ~LOOKUP_PARENT;
1962 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1965 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1966 static int path_lookupat(int dfd, const char *name,
1967 unsigned int flags, struct nameidata *nd)
1969 struct file *base = NULL;
1974 * Path walking is largely split up into 2 different synchronisation
1975 * schemes, rcu-walk and ref-walk (explained in
1976 * Documentation/filesystems/path-lookup.txt). These share much of the
1977 * path walk code, but some things particularly setup, cleanup, and
1978 * following mounts are sufficiently divergent that functions are
1979 * duplicated. Typically there is a function foo(), and its RCU
1980 * analogue, foo_rcu().
1982 * -ECHILD is the error number of choice (just to avoid clashes) that
1983 * is returned if some aspect of an rcu-walk fails. Such an error must
1984 * be handled by restarting a traditional ref-walk (which will always
1985 * be able to complete).
1987 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1992 current->total_link_count = 0;
1993 err = link_path_walk(name, nd);
1995 if (!err && !(flags & LOOKUP_PARENT)) {
1996 err = lookup_last(nd, &path);
1999 struct path link = path;
2000 err = may_follow_link(&link, nd);
2003 nd->flags |= LOOKUP_PARENT;
2004 err = follow_link(&link, nd, &cookie);
2007 err = lookup_last(nd, &path);
2008 put_link(nd, &link, cookie);
2013 err = complete_walk(nd);
2015 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2016 if (!can_lookup(nd->inode)) {
2017 path_put(&nd->path);
2025 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2026 path_put(&nd->root);
2027 nd->root.mnt = NULL;
2032 static int filename_lookup(int dfd, struct filename *name,
2033 unsigned int flags, struct nameidata *nd)
2035 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2036 if (unlikely(retval == -ECHILD))
2037 retval = path_lookupat(dfd, name->name, flags, nd);
2038 if (unlikely(retval == -ESTALE))
2039 retval = path_lookupat(dfd, name->name,
2040 flags | LOOKUP_REVAL, nd);
2042 if (likely(!retval))
2043 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2047 static int do_path_lookup(int dfd, const char *name,
2048 unsigned int flags, struct nameidata *nd)
2050 struct filename filename = { .name = name };
2052 return filename_lookup(dfd, &filename, flags, nd);
2055 /* does lookup, returns the object with parent locked */
2056 struct dentry *kern_path_locked(const char *name, struct path *path)
2058 struct nameidata nd;
2060 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2062 return ERR_PTR(err);
2063 if (nd.last_type != LAST_NORM) {
2065 return ERR_PTR(-EINVAL);
2067 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2068 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2070 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2078 int kern_path(const char *name, unsigned int flags, struct path *path)
2080 struct nameidata nd;
2081 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2088 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2089 * @dentry: pointer to dentry of the base directory
2090 * @mnt: pointer to vfs mount of the base directory
2091 * @name: pointer to file name
2092 * @flags: lookup flags
2093 * @path: pointer to struct path to fill
2095 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2096 const char *name, unsigned int flags,
2099 struct nameidata nd;
2101 nd.root.dentry = dentry;
2103 BUG_ON(flags & LOOKUP_PARENT);
2104 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2105 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2112 * Restricted form of lookup. Doesn't follow links, single-component only,
2113 * needs parent already locked. Doesn't follow mounts.
2116 static struct dentry *lookup_hash(struct nameidata *nd)
2118 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2122 * lookup_one_len - filesystem helper to lookup single pathname component
2123 * @name: pathname component to lookup
2124 * @base: base directory to lookup from
2125 * @len: maximum length @len should be interpreted to
2127 * Note that this routine is purely a helper for filesystem usage and should
2128 * not be called by generic code. Also note that by using this function the
2129 * nameidata argument is passed to the filesystem methods and a filesystem
2130 * using this helper needs to be prepared for that.
2132 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2138 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2142 this.hash = full_name_hash(name, len);
2144 return ERR_PTR(-EACCES);
2146 if (unlikely(name[0] == '.')) {
2147 if (len < 2 || (len == 2 && name[1] == '.'))
2148 return ERR_PTR(-EACCES);
2152 c = *(const unsigned char *)name++;
2153 if (c == '/' || c == '\0')
2154 return ERR_PTR(-EACCES);
2157 * See if the low-level filesystem might want
2158 * to use its own hash..
2160 if (base->d_flags & DCACHE_OP_HASH) {
2161 int err = base->d_op->d_hash(base, &this);
2163 return ERR_PTR(err);
2166 err = inode_permission(base->d_inode, MAY_EXEC);
2168 return ERR_PTR(err);
2170 return __lookup_hash(&this, base, 0);
2173 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2174 struct path *path, int *empty)
2176 struct nameidata nd;
2177 struct filename *tmp = getname_flags(name, flags, empty);
2178 int err = PTR_ERR(tmp);
2181 BUG_ON(flags & LOOKUP_PARENT);
2183 err = filename_lookup(dfd, tmp, flags, &nd);
2191 int user_path_at(int dfd, const char __user *name, unsigned flags,
2194 return user_path_at_empty(dfd, name, flags, path, NULL);
2198 * NB: most callers don't do anything directly with the reference to the
2199 * to struct filename, but the nd->last pointer points into the name string
2200 * allocated by getname. So we must hold the reference to it until all
2201 * path-walking is complete.
2203 static struct filename *
2204 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2207 struct filename *s = getname(path);
2210 /* only LOOKUP_REVAL is allowed in extra flags */
2211 flags &= LOOKUP_REVAL;
2216 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2219 return ERR_PTR(error);
2226 * umount_lookup_last - look up last component for umount
2227 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2228 * @path: pointer to container for result
2230 * This is a special lookup_last function just for umount. In this case, we
2231 * need to resolve the path without doing any revalidation.
2233 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2234 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2235 * in almost all cases, this lookup will be served out of the dcache. The only
2236 * cases where it won't are if nd->last refers to a symlink or the path is
2237 * bogus and it doesn't exist.
2240 * -error: if there was an error during lookup. This includes -ENOENT if the
2241 * lookup found a negative dentry. The nd->path reference will also be
2244 * 0: if we successfully resolved nd->path and found it to not to be a
2245 * symlink that needs to be followed. "path" will also be populated.
2246 * The nd->path reference will also be put.
2248 * 1: if we successfully resolved nd->last and found it to be a symlink
2249 * that needs to be followed. "path" will be populated with the path
2250 * to the link, and nd->path will *not* be put.
2253 umount_lookup_last(struct nameidata *nd, struct path *path)
2256 struct dentry *dentry;
2257 struct dentry *dir = nd->path.dentry;
2259 /* If we're in rcuwalk, drop out of it to handle last component */
2260 if (nd->flags & LOOKUP_RCU) {
2261 if (unlazy_walk(nd, NULL)) {
2267 nd->flags &= ~LOOKUP_PARENT;
2269 if (unlikely(nd->last_type != LAST_NORM)) {
2270 error = handle_dots(nd, nd->last_type);
2273 dentry = dget(nd->path.dentry);
2277 mutex_lock(&dir->d_inode->i_mutex);
2278 dentry = d_lookup(dir, &nd->last);
2281 * No cached dentry. Mounted dentries are pinned in the cache,
2282 * so that means that this dentry is probably a symlink or the
2283 * path doesn't actually point to a mounted dentry.
2285 dentry = d_alloc(dir, &nd->last);
2290 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2291 error = PTR_ERR(dentry);
2295 mutex_unlock(&dir->d_inode->i_mutex);
2298 if (!dentry->d_inode) {
2303 path->dentry = dentry;
2304 path->mnt = mntget(nd->path.mnt);
2305 if (should_follow_link(dentry->d_inode, nd->flags & LOOKUP_FOLLOW))
2315 * path_umountat - look up a path to be umounted
2316 * @dfd: directory file descriptor to start walk from
2317 * @name: full pathname to walk
2318 * @flags: lookup flags
2319 * @nd: pathwalk nameidata
2321 * Look up the given name, but don't attempt to revalidate the last component.
2322 * Returns 0 and "path" will be valid on success; Retuns error otherwise.
2325 path_umountat(int dfd, const char *name, struct path *path, unsigned int flags)
2327 struct file *base = NULL;
2328 struct nameidata nd;
2331 err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2335 current->total_link_count = 0;
2336 err = link_path_walk(name, &nd);
2340 err = umount_lookup_last(&nd, path);
2343 struct path link = *path;
2344 err = may_follow_link(&link, &nd);
2347 nd.flags |= LOOKUP_PARENT;
2348 err = follow_link(&link, &nd, &cookie);
2351 err = umount_lookup_last(&nd, path);
2352 put_link(&nd, &link, cookie);
2358 if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2365 * user_path_umountat - lookup a path from userland in order to umount it
2366 * @dfd: directory file descriptor
2367 * @name: pathname from userland
2368 * @flags: lookup flags
2369 * @path: pointer to container to hold result
2371 * A umount is a special case for path walking. We're not actually interested
2372 * in the inode in this situation, and ESTALE errors can be a problem. We
2373 * simply want track down the dentry and vfsmount attached at the mountpoint
2374 * and avoid revalidating the last component.
2376 * Returns 0 and populates "path" on success.
2379 user_path_umountat(int dfd, const char __user *name, unsigned int flags,
2382 struct filename *s = getname(name);
2388 error = path_umountat(dfd, s->name, path, flags | LOOKUP_RCU);
2389 if (unlikely(error == -ECHILD))
2390 error = path_umountat(dfd, s->name, path, flags);
2391 if (unlikely(error == -ESTALE))
2392 error = path_umountat(dfd, s->name, path, flags | LOOKUP_REVAL);
2395 audit_inode(s, path->dentry, 0);
2402 * It's inline, so penalty for filesystems that don't use sticky bit is
2405 static inline int check_sticky(struct inode *dir, struct inode *inode)
2407 kuid_t fsuid = current_fsuid();
2409 if (!(dir->i_mode & S_ISVTX))
2411 if (uid_eq(inode->i_uid, fsuid))
2413 if (uid_eq(dir->i_uid, fsuid))
2415 return !inode_capable(inode, CAP_FOWNER);
2419 * Check whether we can remove a link victim from directory dir, check
2420 * whether the type of victim is right.
2421 * 1. We can't do it if dir is read-only (done in permission())
2422 * 2. We should have write and exec permissions on dir
2423 * 3. We can't remove anything from append-only dir
2424 * 4. We can't do anything with immutable dir (done in permission())
2425 * 5. If the sticky bit on dir is set we should either
2426 * a. be owner of dir, or
2427 * b. be owner of victim, or
2428 * c. have CAP_FOWNER capability
2429 * 6. If the victim is append-only or immutable we can't do antyhing with
2430 * links pointing to it.
2431 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2432 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2433 * 9. We can't remove a root or mountpoint.
2434 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2435 * nfs_async_unlink().
2437 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2441 if (!victim->d_inode)
2444 BUG_ON(victim->d_parent->d_inode != dir);
2445 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2447 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2452 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2453 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2456 if (!S_ISDIR(victim->d_inode->i_mode))
2458 if (IS_ROOT(victim))
2460 } else if (S_ISDIR(victim->d_inode->i_mode))
2462 if (IS_DEADDIR(dir))
2464 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2469 /* Check whether we can create an object with dentry child in directory
2471 * 1. We can't do it if child already exists (open has special treatment for
2472 * this case, but since we are inlined it's OK)
2473 * 2. We can't do it if dir is read-only (done in permission())
2474 * 3. We should have write and exec permissions on dir
2475 * 4. We can't do it if dir is immutable (done in permission())
2477 static inline int may_create(struct inode *dir, struct dentry *child)
2481 if (IS_DEADDIR(dir))
2483 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2487 * p1 and p2 should be directories on the same fs.
2489 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2494 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2498 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2500 p = d_ancestor(p2, p1);
2502 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2503 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2507 p = d_ancestor(p1, p2);
2509 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2510 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2514 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2515 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2519 void unlock_rename(struct dentry *p1, struct dentry *p2)
2521 mutex_unlock(&p1->d_inode->i_mutex);
2523 mutex_unlock(&p2->d_inode->i_mutex);
2524 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2528 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2531 int error = may_create(dir, dentry);
2535 if (!dir->i_op->create)
2536 return -EACCES; /* shouldn't it be ENOSYS? */
2539 error = security_inode_create(dir, dentry, mode);
2542 error = dir->i_op->create(dir, dentry, mode, want_excl);
2544 fsnotify_create(dir, dentry);
2548 static int may_open(struct path *path, int acc_mode, int flag)
2550 struct dentry *dentry = path->dentry;
2551 struct inode *inode = dentry->d_inode;
2561 switch (inode->i_mode & S_IFMT) {
2565 if (acc_mode & MAY_WRITE)
2570 if (path->mnt->mnt_flags & MNT_NODEV)
2579 error = inode_permission(inode, acc_mode);
2584 * An append-only file must be opened in append mode for writing.
2586 if (IS_APPEND(inode)) {
2587 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2593 /* O_NOATIME can only be set by the owner or superuser */
2594 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2600 static int handle_truncate(struct file *filp)
2602 struct path *path = &filp->f_path;
2603 struct inode *inode = path->dentry->d_inode;
2604 int error = get_write_access(inode);
2608 * Refuse to truncate files with mandatory locks held on them.
2610 error = locks_verify_locked(inode);
2612 error = security_path_truncate(path);
2614 error = do_truncate(path->dentry, 0,
2615 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2618 put_write_access(inode);
2622 static inline int open_to_namei_flags(int flag)
2624 if ((flag & O_ACCMODE) == 3)
2629 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2631 int error = security_path_mknod(dir, dentry, mode, 0);
2635 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2639 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2643 * Attempt to atomically look up, create and open a file from a negative
2646 * Returns 0 if successful. The file will have been created and attached to
2647 * @file by the filesystem calling finish_open().
2649 * Returns 1 if the file was looked up only or didn't need creating. The
2650 * caller will need to perform the open themselves. @path will have been
2651 * updated to point to the new dentry. This may be negative.
2653 * Returns an error code otherwise.
2655 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2656 struct path *path, struct file *file,
2657 const struct open_flags *op,
2658 bool got_write, bool need_lookup,
2661 struct inode *dir = nd->path.dentry->d_inode;
2662 unsigned open_flag = open_to_namei_flags(op->open_flag);
2666 int create_error = 0;
2667 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2669 BUG_ON(dentry->d_inode);
2671 /* Don't create child dentry for a dead directory. */
2672 if (unlikely(IS_DEADDIR(dir))) {
2678 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2679 mode &= ~current_umask();
2681 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) {
2682 open_flag &= ~O_TRUNC;
2683 *opened |= FILE_CREATED;
2687 * Checking write permission is tricky, bacuse we don't know if we are
2688 * going to actually need it: O_CREAT opens should work as long as the
2689 * file exists. But checking existence breaks atomicity. The trick is
2690 * to check access and if not granted clear O_CREAT from the flags.
2692 * Another problem is returing the "right" error value (e.g. for an
2693 * O_EXCL open we want to return EEXIST not EROFS).
2695 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2696 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2697 if (!(open_flag & O_CREAT)) {
2699 * No O_CREATE -> atomicity not a requirement -> fall
2700 * back to lookup + open
2703 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2704 /* Fall back and fail with the right error */
2705 create_error = -EROFS;
2708 /* No side effects, safe to clear O_CREAT */
2709 create_error = -EROFS;
2710 open_flag &= ~O_CREAT;
2714 if (open_flag & O_CREAT) {
2715 error = may_o_create(&nd->path, dentry, mode);
2717 create_error = error;
2718 if (open_flag & O_EXCL)
2720 open_flag &= ~O_CREAT;
2724 if (nd->flags & LOOKUP_DIRECTORY)
2725 open_flag |= O_DIRECTORY;
2727 file->f_path.dentry = DENTRY_NOT_SET;
2728 file->f_path.mnt = nd->path.mnt;
2729 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2732 if (create_error && error == -ENOENT)
2733 error = create_error;
2737 acc_mode = op->acc_mode;
2738 if (*opened & FILE_CREATED) {
2739 fsnotify_create(dir, dentry);
2740 acc_mode = MAY_OPEN;
2743 if (error) { /* returned 1, that is */
2744 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2748 if (file->f_path.dentry) {
2750 dentry = file->f_path.dentry;
2752 if (create_error && dentry->d_inode == NULL) {
2753 error = create_error;
2760 * We didn't have the inode before the open, so check open permission
2763 error = may_open(&file->f_path, acc_mode, open_flag);
2773 dentry = lookup_real(dir, dentry, nd->flags);
2775 return PTR_ERR(dentry);
2778 int open_flag = op->open_flag;
2780 error = create_error;
2781 if ((open_flag & O_EXCL)) {
2782 if (!dentry->d_inode)
2784 } else if (!dentry->d_inode) {
2786 } else if ((open_flag & O_TRUNC) &&
2787 S_ISREG(dentry->d_inode->i_mode)) {
2790 /* will fail later, go on to get the right error */
2794 path->dentry = dentry;
2795 path->mnt = nd->path.mnt;
2800 * Look up and maybe create and open the last component.
2802 * Must be called with i_mutex held on parent.
2804 * Returns 0 if the file was successfully atomically created (if necessary) and
2805 * opened. In this case the file will be returned attached to @file.
2807 * Returns 1 if the file was not completely opened at this time, though lookups
2808 * and creations will have been performed and the dentry returned in @path will
2809 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2810 * specified then a negative dentry may be returned.
2812 * An error code is returned otherwise.
2814 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2815 * cleared otherwise prior to returning.
2817 static int lookup_open(struct nameidata *nd, struct path *path,
2819 const struct open_flags *op,
2820 bool got_write, int *opened)
2822 struct dentry *dir = nd->path.dentry;
2823 struct inode *dir_inode = dir->d_inode;
2824 struct dentry *dentry;
2828 *opened &= ~FILE_CREATED;
2829 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2831 return PTR_ERR(dentry);
2833 /* Cached positive dentry: will open in f_op->open */
2834 if (!need_lookup && dentry->d_inode)
2837 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2838 return atomic_open(nd, dentry, path, file, op, got_write,
2839 need_lookup, opened);
2843 BUG_ON(dentry->d_inode);
2845 dentry = lookup_real(dir_inode, dentry, nd->flags);
2847 return PTR_ERR(dentry);
2850 /* Negative dentry, just create the file */
2851 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2852 umode_t mode = op->mode;
2853 if (!IS_POSIXACL(dir->d_inode))
2854 mode &= ~current_umask();
2856 * This write is needed to ensure that a
2857 * rw->ro transition does not occur between
2858 * the time when the file is created and when
2859 * a permanent write count is taken through
2860 * the 'struct file' in finish_open().
2866 *opened |= FILE_CREATED;
2867 error = security_path_mknod(&nd->path, dentry, mode, 0);
2870 error = vfs_create(dir->d_inode, dentry, mode,
2871 nd->flags & LOOKUP_EXCL);
2876 path->dentry = dentry;
2877 path->mnt = nd->path.mnt;
2886 * Handle the last step of open()
2888 static int do_last(struct nameidata *nd, struct path *path,
2889 struct file *file, const struct open_flags *op,
2890 int *opened, struct filename *name)
2892 struct dentry *dir = nd->path.dentry;
2893 int open_flag = op->open_flag;
2894 bool will_truncate = (open_flag & O_TRUNC) != 0;
2895 bool got_write = false;
2896 int acc_mode = op->acc_mode;
2897 struct inode *inode;
2898 bool symlink_ok = false;
2899 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2900 bool retried = false;
2903 nd->flags &= ~LOOKUP_PARENT;
2904 nd->flags |= op->intent;
2906 if (nd->last_type != LAST_NORM) {
2907 error = handle_dots(nd, nd->last_type);
2913 if (!(open_flag & O_CREAT)) {
2914 if (nd->last.name[nd->last.len])
2915 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2916 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2918 /* we _can_ be in RCU mode here */
2919 error = lookup_fast(nd, path, &inode);
2926 BUG_ON(nd->inode != dir->d_inode);
2928 /* create side of things */
2930 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2931 * has been cleared when we got to the last component we are
2934 error = complete_walk(nd);
2938 audit_inode(name, dir, LOOKUP_PARENT);
2940 /* trailing slashes? */
2941 if (nd->last.name[nd->last.len])
2946 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2947 error = mnt_want_write(nd->path.mnt);
2951 * do _not_ fail yet - we might not need that or fail with
2952 * a different error; let lookup_open() decide; we'll be
2953 * dropping this one anyway.
2956 mutex_lock(&dir->d_inode->i_mutex);
2957 error = lookup_open(nd, path, file, op, got_write, opened);
2958 mutex_unlock(&dir->d_inode->i_mutex);
2964 if ((*opened & FILE_CREATED) ||
2965 !S_ISREG(file_inode(file)->i_mode))
2966 will_truncate = false;
2968 audit_inode(name, file->f_path.dentry, 0);
2972 if (*opened & FILE_CREATED) {
2973 /* Don't check for write permission, don't truncate */
2974 open_flag &= ~O_TRUNC;
2975 will_truncate = false;
2976 acc_mode = MAY_OPEN;
2977 path_to_nameidata(path, nd);
2978 goto finish_open_created;
2982 * create/update audit record if it already exists.
2984 if (path->dentry->d_inode)
2985 audit_inode(name, path->dentry, 0);
2988 * If atomic_open() acquired write access it is dropped now due to
2989 * possible mount and symlink following (this might be optimized away if
2993 mnt_drop_write(nd->path.mnt);
2998 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3001 error = follow_managed(path, nd->flags);
3006 nd->flags |= LOOKUP_JUMPED;
3008 BUG_ON(nd->flags & LOOKUP_RCU);
3009 inode = path->dentry->d_inode;
3011 /* we _can_ be in RCU mode here */
3014 path_to_nameidata(path, nd);
3018 if (should_follow_link(inode, !symlink_ok)) {
3019 if (nd->flags & LOOKUP_RCU) {
3020 if (unlikely(unlazy_walk(nd, path->dentry))) {
3025 BUG_ON(inode != path->dentry->d_inode);
3029 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3030 path_to_nameidata(path, nd);
3032 save_parent.dentry = nd->path.dentry;
3033 save_parent.mnt = mntget(path->mnt);
3034 nd->path.dentry = path->dentry;
3038 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3040 error = complete_walk(nd);
3042 path_put(&save_parent);
3045 audit_inode(name, nd->path.dentry, 0);
3047 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
3050 if ((nd->flags & LOOKUP_DIRECTORY) && !can_lookup(nd->inode))
3052 if (!S_ISREG(nd->inode->i_mode))
3053 will_truncate = false;
3055 if (will_truncate) {
3056 error = mnt_want_write(nd->path.mnt);
3061 finish_open_created:
3062 error = may_open(&nd->path, acc_mode, open_flag);
3065 file->f_path.mnt = nd->path.mnt;
3066 error = finish_open(file, nd->path.dentry, NULL, opened);
3068 if (error == -EOPENSTALE)
3073 error = open_check_o_direct(file);
3076 error = ima_file_check(file, op->acc_mode);
3080 if (will_truncate) {
3081 error = handle_truncate(file);
3087 mnt_drop_write(nd->path.mnt);
3088 path_put(&save_parent);
3093 path_put_conditional(path, nd);
3100 /* If no saved parent or already retried then can't retry */
3101 if (!save_parent.dentry || retried)
3104 BUG_ON(save_parent.dentry != dir);
3105 path_put(&nd->path);
3106 nd->path = save_parent;
3107 nd->inode = dir->d_inode;
3108 save_parent.mnt = NULL;
3109 save_parent.dentry = NULL;
3111 mnt_drop_write(nd->path.mnt);
3118 static int do_tmpfile(int dfd, struct filename *pathname,
3119 struct nameidata *nd, int flags,
3120 const struct open_flags *op,
3121 struct file *file, int *opened)
3123 static const struct qstr name = QSTR_INIT("/", 1);
3124 struct dentry *dentry, *child;
3126 int error = path_lookupat(dfd, pathname->name,
3127 flags | LOOKUP_DIRECTORY, nd);
3128 if (unlikely(error))
3130 error = mnt_want_write(nd->path.mnt);
3131 if (unlikely(error))
3133 /* we want directory to be writable */
3134 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3137 dentry = nd->path.dentry;
3138 dir = dentry->d_inode;
3139 if (!dir->i_op->tmpfile) {
3140 error = -EOPNOTSUPP;
3143 child = d_alloc(dentry, &name);
3144 if (unlikely(!child)) {
3148 nd->flags &= ~LOOKUP_DIRECTORY;
3149 nd->flags |= op->intent;
3150 dput(nd->path.dentry);
3151 nd->path.dentry = child;
3152 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3155 audit_inode(pathname, nd->path.dentry, 0);
3156 error = may_open(&nd->path, op->acc_mode, op->open_flag);
3159 file->f_path.mnt = nd->path.mnt;
3160 error = finish_open(file, nd->path.dentry, NULL, opened);
3163 error = open_check_o_direct(file);
3166 } else if (!(op->open_flag & O_EXCL)) {
3167 struct inode *inode = file_inode(file);
3168 spin_lock(&inode->i_lock);
3169 inode->i_state |= I_LINKABLE;
3170 spin_unlock(&inode->i_lock);
3173 mnt_drop_write(nd->path.mnt);
3175 path_put(&nd->path);
3179 static struct file *path_openat(int dfd, struct filename *pathname,
3180 struct nameidata *nd, const struct open_flags *op, int flags)
3182 struct file *base = NULL;
3188 file = get_empty_filp();
3192 file->f_flags = op->open_flag;
3194 if (unlikely(file->f_flags & __O_TMPFILE)) {
3195 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3199 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
3200 if (unlikely(error))
3203 current->total_link_count = 0;
3204 error = link_path_walk(pathname->name, nd);
3205 if (unlikely(error))
3208 error = do_last(nd, &path, file, op, &opened, pathname);
3209 while (unlikely(error > 0)) { /* trailing symlink */
3210 struct path link = path;
3212 if (!(nd->flags & LOOKUP_FOLLOW)) {
3213 path_put_conditional(&path, nd);
3214 path_put(&nd->path);
3218 error = may_follow_link(&link, nd);
3219 if (unlikely(error))
3221 nd->flags |= LOOKUP_PARENT;
3222 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3223 error = follow_link(&link, nd, &cookie);
3224 if (unlikely(error))
3226 error = do_last(nd, &path, file, op, &opened, pathname);
3227 put_link(nd, &link, cookie);
3230 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3231 path_put(&nd->root);
3234 if (!(opened & FILE_OPENED)) {
3238 if (unlikely(error)) {
3239 if (error == -EOPENSTALE) {
3240 if (flags & LOOKUP_RCU)
3245 file = ERR_PTR(error);
3250 struct file *do_filp_open(int dfd, struct filename *pathname,
3251 const struct open_flags *op)
3253 struct nameidata nd;
3254 int flags = op->lookup_flags;
3257 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3258 if (unlikely(filp == ERR_PTR(-ECHILD)))
3259 filp = path_openat(dfd, pathname, &nd, op, flags);
3260 if (unlikely(filp == ERR_PTR(-ESTALE)))
3261 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3265 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3266 const char *name, const struct open_flags *op)
3268 struct nameidata nd;
3270 struct filename filename = { .name = name };
3271 int flags = op->lookup_flags | LOOKUP_ROOT;
3274 nd.root.dentry = dentry;
3276 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
3277 return ERR_PTR(-ELOOP);
3279 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3280 if (unlikely(file == ERR_PTR(-ECHILD)))
3281 file = path_openat(-1, &filename, &nd, op, flags);
3282 if (unlikely(file == ERR_PTR(-ESTALE)))
3283 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3287 struct dentry *kern_path_create(int dfd, const char *pathname,
3288 struct path *path, unsigned int lookup_flags)
3290 struct dentry *dentry = ERR_PTR(-EEXIST);
3291 struct nameidata nd;
3294 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3297 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3298 * other flags passed in are ignored!
3300 lookup_flags &= LOOKUP_REVAL;
3302 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3304 return ERR_PTR(error);
3307 * Yucky last component or no last component at all?
3308 * (foo/., foo/.., /////)
3310 if (nd.last_type != LAST_NORM)
3312 nd.flags &= ~LOOKUP_PARENT;
3313 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3315 /* don't fail immediately if it's r/o, at least try to report other errors */
3316 err2 = mnt_want_write(nd.path.mnt);
3318 * Do the final lookup.
3320 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3321 dentry = lookup_hash(&nd);
3326 if (dentry->d_inode)
3329 * Special case - lookup gave negative, but... we had foo/bar/
3330 * From the vfs_mknod() POV we just have a negative dentry -
3331 * all is fine. Let's be bastards - you had / on the end, you've
3332 * been asking for (non-existent) directory. -ENOENT for you.
3334 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3338 if (unlikely(err2)) {
3346 dentry = ERR_PTR(error);
3348 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3350 mnt_drop_write(nd.path.mnt);
3355 EXPORT_SYMBOL(kern_path_create);
3357 void done_path_create(struct path *path, struct dentry *dentry)
3360 mutex_unlock(&path->dentry->d_inode->i_mutex);
3361 mnt_drop_write(path->mnt);
3364 EXPORT_SYMBOL(done_path_create);
3366 struct dentry *user_path_create(int dfd, const char __user *pathname,
3367 struct path *path, unsigned int lookup_flags)
3369 struct filename *tmp = getname(pathname);
3372 return ERR_CAST(tmp);
3373 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3377 EXPORT_SYMBOL(user_path_create);
3379 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3381 int error = may_create(dir, dentry);
3386 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3389 if (!dir->i_op->mknod)
3392 error = devcgroup_inode_mknod(mode, dev);
3396 error = security_inode_mknod(dir, dentry, mode, dev);
3400 error = dir->i_op->mknod(dir, dentry, mode, dev);
3402 fsnotify_create(dir, dentry);
3406 static int may_mknod(umode_t mode)
3408 switch (mode & S_IFMT) {
3414 case 0: /* zero mode translates to S_IFREG */
3423 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3426 struct dentry *dentry;
3429 unsigned int lookup_flags = 0;
3431 error = may_mknod(mode);
3435 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3437 return PTR_ERR(dentry);
3439 if (!IS_POSIXACL(path.dentry->d_inode))
3440 mode &= ~current_umask();
3441 error = security_path_mknod(&path, dentry, mode, dev);
3444 switch (mode & S_IFMT) {
3445 case 0: case S_IFREG:
3446 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3448 case S_IFCHR: case S_IFBLK:
3449 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3450 new_decode_dev(dev));
3452 case S_IFIFO: case S_IFSOCK:
3453 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3457 done_path_create(&path, dentry);
3458 if (retry_estale(error, lookup_flags)) {
3459 lookup_flags |= LOOKUP_REVAL;
3465 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3467 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3470 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3472 int error = may_create(dir, dentry);
3473 unsigned max_links = dir->i_sb->s_max_links;
3478 if (!dir->i_op->mkdir)
3481 mode &= (S_IRWXUGO|S_ISVTX);
3482 error = security_inode_mkdir(dir, dentry, mode);
3486 if (max_links && dir->i_nlink >= max_links)
3489 error = dir->i_op->mkdir(dir, dentry, mode);
3491 fsnotify_mkdir(dir, dentry);
3495 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3497 struct dentry *dentry;
3500 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3503 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3505 return PTR_ERR(dentry);
3507 if (!IS_POSIXACL(path.dentry->d_inode))
3508 mode &= ~current_umask();
3509 error = security_path_mkdir(&path, dentry, mode);
3511 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3512 done_path_create(&path, dentry);
3513 if (retry_estale(error, lookup_flags)) {
3514 lookup_flags |= LOOKUP_REVAL;
3520 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3522 return sys_mkdirat(AT_FDCWD, pathname, mode);
3526 * The dentry_unhash() helper will try to drop the dentry early: we
3527 * should have a usage count of 1 if we're the only user of this
3528 * dentry, and if that is true (possibly after pruning the dcache),
3529 * then we drop the dentry now.
3531 * A low-level filesystem can, if it choses, legally
3534 * if (!d_unhashed(dentry))
3537 * if it cannot handle the case of removing a directory
3538 * that is still in use by something else..
3540 void dentry_unhash(struct dentry *dentry)
3542 shrink_dcache_parent(dentry);
3543 spin_lock(&dentry->d_lock);
3544 if (dentry->d_lockref.count == 1)
3546 spin_unlock(&dentry->d_lock);
3549 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3551 int error = may_delete(dir, dentry, 1);
3556 if (!dir->i_op->rmdir)
3560 mutex_lock(&dentry->d_inode->i_mutex);
3563 if (d_mountpoint(dentry))
3566 error = security_inode_rmdir(dir, dentry);
3570 shrink_dcache_parent(dentry);
3571 error = dir->i_op->rmdir(dir, dentry);
3575 dentry->d_inode->i_flags |= S_DEAD;
3579 mutex_unlock(&dentry->d_inode->i_mutex);
3586 static long do_rmdir(int dfd, const char __user *pathname)
3589 struct filename *name;
3590 struct dentry *dentry;
3591 struct nameidata nd;
3592 unsigned int lookup_flags = 0;
3594 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3596 return PTR_ERR(name);
3598 switch(nd.last_type) {
3610 nd.flags &= ~LOOKUP_PARENT;
3611 error = mnt_want_write(nd.path.mnt);
3615 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3616 dentry = lookup_hash(&nd);
3617 error = PTR_ERR(dentry);
3620 if (!dentry->d_inode) {
3624 error = security_path_rmdir(&nd.path, dentry);
3627 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3631 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3632 mnt_drop_write(nd.path.mnt);
3636 if (retry_estale(error, lookup_flags)) {
3637 lookup_flags |= LOOKUP_REVAL;
3643 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3645 return do_rmdir(AT_FDCWD, pathname);
3648 int vfs_unlink(struct inode *dir, struct dentry *dentry)
3650 int error = may_delete(dir, dentry, 0);
3655 if (!dir->i_op->unlink)
3658 mutex_lock(&dentry->d_inode->i_mutex);
3659 if (d_mountpoint(dentry))
3662 error = security_inode_unlink(dir, dentry);
3664 error = dir->i_op->unlink(dir, dentry);
3669 mutex_unlock(&dentry->d_inode->i_mutex);
3671 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3672 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3673 fsnotify_link_count(dentry->d_inode);
3681 * Make sure that the actual truncation of the file will occur outside its
3682 * directory's i_mutex. Truncate can take a long time if there is a lot of
3683 * writeout happening, and we don't want to prevent access to the directory
3684 * while waiting on the I/O.
3686 static long do_unlinkat(int dfd, const char __user *pathname)
3689 struct filename *name;
3690 struct dentry *dentry;
3691 struct nameidata nd;
3692 struct inode *inode = NULL;
3693 unsigned int lookup_flags = 0;
3695 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3697 return PTR_ERR(name);
3700 if (nd.last_type != LAST_NORM)
3703 nd.flags &= ~LOOKUP_PARENT;
3704 error = mnt_want_write(nd.path.mnt);
3708 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3709 dentry = lookup_hash(&nd);
3710 error = PTR_ERR(dentry);
3711 if (!IS_ERR(dentry)) {
3712 /* Why not before? Because we want correct error value */
3713 if (nd.last.name[nd.last.len])
3715 inode = dentry->d_inode;
3719 error = security_path_unlink(&nd.path, dentry);
3722 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3726 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3728 iput(inode); /* truncate the inode here */
3729 mnt_drop_write(nd.path.mnt);
3733 if (retry_estale(error, lookup_flags)) {
3734 lookup_flags |= LOOKUP_REVAL;
3741 error = !dentry->d_inode ? -ENOENT :
3742 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3746 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3748 if ((flag & ~AT_REMOVEDIR) != 0)
3751 if (flag & AT_REMOVEDIR)
3752 return do_rmdir(dfd, pathname);
3754 return do_unlinkat(dfd, pathname);
3757 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3759 return do_unlinkat(AT_FDCWD, pathname);
3762 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3764 int error = may_create(dir, dentry);
3769 if (!dir->i_op->symlink)
3772 error = security_inode_symlink(dir, dentry, oldname);
3776 error = dir->i_op->symlink(dir, dentry, oldname);
3778 fsnotify_create(dir, dentry);
3782 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3783 int, newdfd, const char __user *, newname)
3786 struct filename *from;
3787 struct dentry *dentry;
3789 unsigned int lookup_flags = 0;
3791 from = getname(oldname);
3793 return PTR_ERR(from);
3795 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3796 error = PTR_ERR(dentry);
3800 error = security_path_symlink(&path, dentry, from->name);
3802 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3803 done_path_create(&path, dentry);
3804 if (retry_estale(error, lookup_flags)) {
3805 lookup_flags |= LOOKUP_REVAL;
3813 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3815 return sys_symlinkat(oldname, AT_FDCWD, newname);
3818 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3820 struct inode *inode = old_dentry->d_inode;
3821 unsigned max_links = dir->i_sb->s_max_links;
3827 error = may_create(dir, new_dentry);
3831 if (dir->i_sb != inode->i_sb)
3835 * A link to an append-only or immutable file cannot be created.
3837 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3839 if (!dir->i_op->link)
3841 if (S_ISDIR(inode->i_mode))
3844 error = security_inode_link(old_dentry, dir, new_dentry);
3848 mutex_lock(&inode->i_mutex);
3849 /* Make sure we don't allow creating hardlink to an unlinked file */
3850 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3852 else if (max_links && inode->i_nlink >= max_links)
3855 error = dir->i_op->link(old_dentry, dir, new_dentry);
3857 if (!error && (inode->i_state & I_LINKABLE)) {
3858 spin_lock(&inode->i_lock);
3859 inode->i_state &= ~I_LINKABLE;
3860 spin_unlock(&inode->i_lock);
3862 mutex_unlock(&inode->i_mutex);
3864 fsnotify_link(dir, inode, new_dentry);
3869 * Hardlinks are often used in delicate situations. We avoid
3870 * security-related surprises by not following symlinks on the
3873 * We don't follow them on the oldname either to be compatible
3874 * with linux 2.0, and to avoid hard-linking to directories
3875 * and other special files. --ADM
3877 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3878 int, newdfd, const char __user *, newname, int, flags)
3880 struct dentry *new_dentry;
3881 struct path old_path, new_path;
3885 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3888 * To use null names we require CAP_DAC_READ_SEARCH
3889 * This ensures that not everyone will be able to create
3890 * handlink using the passed filedescriptor.
3892 if (flags & AT_EMPTY_PATH) {
3893 if (!capable(CAP_DAC_READ_SEARCH))
3898 if (flags & AT_SYMLINK_FOLLOW)
3899 how |= LOOKUP_FOLLOW;
3901 error = user_path_at(olddfd, oldname, how, &old_path);
3905 new_dentry = user_path_create(newdfd, newname, &new_path,
3906 (how & LOOKUP_REVAL));
3907 error = PTR_ERR(new_dentry);
3908 if (IS_ERR(new_dentry))
3912 if (old_path.mnt != new_path.mnt)
3914 error = may_linkat(&old_path);
3915 if (unlikely(error))
3917 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3920 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3922 done_path_create(&new_path, new_dentry);
3923 if (retry_estale(error, how)) {
3924 how |= LOOKUP_REVAL;
3928 path_put(&old_path);
3933 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3935 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3939 * The worst of all namespace operations - renaming directory. "Perverted"
3940 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3942 * a) we can get into loop creation. Check is done in is_subdir().
3943 * b) race potential - two innocent renames can create a loop together.
3944 * That's where 4.4 screws up. Current fix: serialization on
3945 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3947 * c) we have to lock _three_ objects - parents and victim (if it exists).
3948 * And that - after we got ->i_mutex on parents (until then we don't know
3949 * whether the target exists). Solution: try to be smart with locking
3950 * order for inodes. We rely on the fact that tree topology may change
3951 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3952 * move will be locked. Thus we can rank directories by the tree
3953 * (ancestors first) and rank all non-directories after them.
3954 * That works since everybody except rename does "lock parent, lookup,
3955 * lock child" and rename is under ->s_vfs_rename_mutex.
3956 * HOWEVER, it relies on the assumption that any object with ->lookup()
3957 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3958 * we'd better make sure that there's no link(2) for them.
3959 * d) conversion from fhandle to dentry may come in the wrong moment - when
3960 * we are removing the target. Solution: we will have to grab ->i_mutex
3961 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3962 * ->i_mutex on parents, which works but leads to some truly excessive
3965 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3966 struct inode *new_dir, struct dentry *new_dentry)
3969 struct inode *target = new_dentry->d_inode;
3970 unsigned max_links = new_dir->i_sb->s_max_links;
3973 * If we are going to change the parent - check write permissions,
3974 * we'll need to flip '..'.
3976 if (new_dir != old_dir) {
3977 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3982 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3988 mutex_lock(&target->i_mutex);
3991 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3995 if (max_links && !target && new_dir != old_dir &&
3996 new_dir->i_nlink >= max_links)
4000 shrink_dcache_parent(new_dentry);
4001 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4006 target->i_flags |= S_DEAD;
4007 dont_mount(new_dentry);
4011 mutex_unlock(&target->i_mutex);
4014 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4015 d_move(old_dentry,new_dentry);
4019 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
4020 struct inode *new_dir, struct dentry *new_dentry)
4022 struct inode *target = new_dentry->d_inode;
4025 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4031 mutex_lock(&target->i_mutex);
4034 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
4037 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4042 dont_mount(new_dentry);
4043 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4044 d_move(old_dentry, new_dentry);
4047 mutex_unlock(&target->i_mutex);
4052 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4053 struct inode *new_dir, struct dentry *new_dentry)
4056 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
4057 const unsigned char *old_name;
4059 if (old_dentry->d_inode == new_dentry->d_inode)
4062 error = may_delete(old_dir, old_dentry, is_dir);
4066 if (!new_dentry->d_inode)
4067 error = may_create(new_dir, new_dentry);
4069 error = may_delete(new_dir, new_dentry, is_dir);
4073 if (!old_dir->i_op->rename)
4076 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4079 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
4081 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
4083 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4084 new_dentry->d_inode, old_dentry);
4085 fsnotify_oldname_free(old_name);
4090 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4091 int, newdfd, const char __user *, newname)
4093 struct dentry *old_dir, *new_dir;
4094 struct dentry *old_dentry, *new_dentry;
4095 struct dentry *trap;
4096 struct nameidata oldnd, newnd;
4097 struct filename *from;
4098 struct filename *to;
4099 unsigned int lookup_flags = 0;
4100 bool should_retry = false;
4103 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4105 error = PTR_ERR(from);
4109 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4111 error = PTR_ERR(to);
4116 if (oldnd.path.mnt != newnd.path.mnt)
4119 old_dir = oldnd.path.dentry;
4121 if (oldnd.last_type != LAST_NORM)
4124 new_dir = newnd.path.dentry;
4125 if (newnd.last_type != LAST_NORM)
4128 error = mnt_want_write(oldnd.path.mnt);
4132 oldnd.flags &= ~LOOKUP_PARENT;
4133 newnd.flags &= ~LOOKUP_PARENT;
4134 newnd.flags |= LOOKUP_RENAME_TARGET;
4136 trap = lock_rename(new_dir, old_dir);
4138 old_dentry = lookup_hash(&oldnd);
4139 error = PTR_ERR(old_dentry);
4140 if (IS_ERR(old_dentry))
4142 /* source must exist */
4144 if (!old_dentry->d_inode)
4146 /* unless the source is a directory trailing slashes give -ENOTDIR */
4147 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
4149 if (oldnd.last.name[oldnd.last.len])
4151 if (newnd.last.name[newnd.last.len])
4154 /* source should not be ancestor of target */
4156 if (old_dentry == trap)
4158 new_dentry = lookup_hash(&newnd);
4159 error = PTR_ERR(new_dentry);
4160 if (IS_ERR(new_dentry))
4162 /* target should not be an ancestor of source */
4164 if (new_dentry == trap)
4167 error = security_path_rename(&oldnd.path, old_dentry,
4168 &newnd.path, new_dentry);
4171 error = vfs_rename(old_dir->d_inode, old_dentry,
4172 new_dir->d_inode, new_dentry);
4178 unlock_rename(new_dir, old_dir);
4179 mnt_drop_write(oldnd.path.mnt);
4181 if (retry_estale(error, lookup_flags))
4182 should_retry = true;
4183 path_put(&newnd.path);
4186 path_put(&oldnd.path);
4189 should_retry = false;
4190 lookup_flags |= LOOKUP_REVAL;
4197 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4199 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
4202 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
4206 len = PTR_ERR(link);
4211 if (len > (unsigned) buflen)
4213 if (copy_to_user(buffer, link, len))
4220 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4221 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4222 * using) it for any given inode is up to filesystem.
4224 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4226 struct nameidata nd;
4231 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4233 return PTR_ERR(cookie);
4235 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
4236 if (dentry->d_inode->i_op->put_link)
4237 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4241 int vfs_follow_link(struct nameidata *nd, const char *link)
4243 return __vfs_follow_link(nd, link);
4246 /* get the link contents into pagecache */
4247 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4251 struct address_space *mapping = dentry->d_inode->i_mapping;
4252 page = read_mapping_page(mapping, 0, NULL);
4257 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4261 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4263 struct page *page = NULL;
4264 char *s = page_getlink(dentry, &page);
4265 int res = vfs_readlink(dentry,buffer,buflen,s);
4268 page_cache_release(page);
4273 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4275 struct page *page = NULL;
4276 nd_set_link(nd, page_getlink(dentry, &page));
4280 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4282 struct page *page = cookie;
4286 page_cache_release(page);
4291 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4293 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4295 struct address_space *mapping = inode->i_mapping;
4300 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4302 flags |= AOP_FLAG_NOFS;
4305 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4306 flags, &page, &fsdata);
4310 kaddr = kmap_atomic(page);
4311 memcpy(kaddr, symname, len-1);
4312 kunmap_atomic(kaddr);
4314 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4321 mark_inode_dirty(inode);
4327 int page_symlink(struct inode *inode, const char *symname, int len)
4329 return __page_symlink(inode, symname, len,
4330 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4333 const struct inode_operations page_symlink_inode_operations = {
4334 .readlink = generic_readlink,
4335 .follow_link = page_follow_link_light,
4336 .put_link = page_put_link,
4339 EXPORT_SYMBOL(user_path_at);
4340 EXPORT_SYMBOL(follow_down_one);
4341 EXPORT_SYMBOL(follow_down);
4342 EXPORT_SYMBOL(follow_up);
4343 EXPORT_SYMBOL(get_write_access); /* nfsd */
4344 EXPORT_SYMBOL(lock_rename);
4345 EXPORT_SYMBOL(lookup_one_len);
4346 EXPORT_SYMBOL(page_follow_link_light);
4347 EXPORT_SYMBOL(page_put_link);
4348 EXPORT_SYMBOL(page_readlink);
4349 EXPORT_SYMBOL(__page_symlink);
4350 EXPORT_SYMBOL(page_symlink);
4351 EXPORT_SYMBOL(page_symlink_inode_operations);
4352 EXPORT_SYMBOL(kern_path);
4353 EXPORT_SYMBOL(vfs_path_lookup);
4354 EXPORT_SYMBOL(inode_permission);
4355 EXPORT_SYMBOL(unlock_rename);
4356 EXPORT_SYMBOL(vfs_create);
4357 EXPORT_SYMBOL(vfs_follow_link);
4358 EXPORT_SYMBOL(vfs_link);
4359 EXPORT_SYMBOL(vfs_mkdir);
4360 EXPORT_SYMBOL(vfs_mknod);
4361 EXPORT_SYMBOL(generic_permission);
4362 EXPORT_SYMBOL(vfs_readlink);
4363 EXPORT_SYMBOL(vfs_rename);
4364 EXPORT_SYMBOL(vfs_rmdir);
4365 EXPORT_SYMBOL(vfs_symlink);
4366 EXPORT_SYMBOL(vfs_unlink);
4367 EXPORT_SYMBOL(dentry_unhash);
4368 EXPORT_SYMBOL(generic_readlink);