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 result->aname = NULL;
200 audit_getname(result);
204 final_putname(result);
209 getname(const char __user * filename)
211 return getname_flags(filename, 0, NULL);
215 * The "getname_kernel()" interface doesn't do pathnames longer
216 * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user.
219 getname_kernel(const char * filename)
221 struct filename *result;
225 len = strlen(filename);
226 if (len >= EMBEDDED_NAME_MAX)
227 return ERR_PTR(-ENAMETOOLONG);
229 result = __getname();
230 if (unlikely(!result))
231 return ERR_PTR(-ENOMEM);
233 kname = (char *)result + sizeof(*result);
234 result->name = kname;
236 result->aname = NULL;
237 result->separate = false;
239 strlcpy(kname, filename, EMBEDDED_NAME_MAX);
243 #ifdef CONFIG_AUDITSYSCALL
244 void putname(struct filename *name)
246 if (unlikely(!audit_dummy_context()))
247 return audit_putname(name);
252 static int check_acl(struct inode *inode, int mask)
254 #ifdef CONFIG_FS_POSIX_ACL
255 struct posix_acl *acl;
257 if (mask & MAY_NOT_BLOCK) {
258 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
261 /* no ->get_acl() calls in RCU mode... */
262 if (acl == ACL_NOT_CACHED)
264 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
267 acl = get_acl(inode, ACL_TYPE_ACCESS);
271 int error = posix_acl_permission(inode, acl, mask);
272 posix_acl_release(acl);
281 * This does the basic permission checking
283 static int acl_permission_check(struct inode *inode, int mask)
285 unsigned int mode = inode->i_mode;
287 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
290 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
291 int error = check_acl(inode, mask);
292 if (error != -EAGAIN)
296 if (in_group_p(inode->i_gid))
301 * If the DACs are ok we don't need any capability check.
303 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
309 * generic_permission - check for access rights on a Posix-like filesystem
310 * @inode: inode to check access rights for
311 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
313 * Used to check for read/write/execute permissions on a file.
314 * We use "fsuid" for this, letting us set arbitrary permissions
315 * for filesystem access without changing the "normal" uids which
316 * are used for other things.
318 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
319 * request cannot be satisfied (eg. requires blocking or too much complexity).
320 * It would then be called again in ref-walk mode.
322 int generic_permission(struct inode *inode, int mask)
327 * Do the basic permission checks.
329 ret = acl_permission_check(inode, mask);
333 if (S_ISDIR(inode->i_mode)) {
334 /* DACs are overridable for directories */
335 if (inode_capable(inode, CAP_DAC_OVERRIDE))
337 if (!(mask & MAY_WRITE))
338 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
343 * Read/write DACs are always overridable.
344 * Executable DACs are overridable when there is
345 * at least one exec bit set.
347 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
348 if (inode_capable(inode, CAP_DAC_OVERRIDE))
352 * Searching includes executable on directories, else just read.
354 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
355 if (mask == MAY_READ)
356 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
363 * We _really_ want to just do "generic_permission()" without
364 * even looking at the inode->i_op values. So we keep a cache
365 * flag in inode->i_opflags, that says "this has not special
366 * permission function, use the fast case".
368 static inline int do_inode_permission(struct inode *inode, int mask)
370 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
371 if (likely(inode->i_op->permission))
372 return inode->i_op->permission(inode, mask);
374 /* This gets set once for the inode lifetime */
375 spin_lock(&inode->i_lock);
376 inode->i_opflags |= IOP_FASTPERM;
377 spin_unlock(&inode->i_lock);
379 return generic_permission(inode, mask);
383 * __inode_permission - Check for access rights to a given inode
384 * @inode: Inode to check permission on
385 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
387 * Check for read/write/execute permissions on an inode.
389 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
391 * This does not check for a read-only file system. You probably want
392 * inode_permission().
394 int __inode_permission(struct inode *inode, int mask)
398 if (unlikely(mask & MAY_WRITE)) {
400 * Nobody gets write access to an immutable file.
402 if (IS_IMMUTABLE(inode))
406 retval = do_inode_permission(inode, mask);
410 retval = devcgroup_inode_permission(inode, mask);
414 return security_inode_permission(inode, mask);
418 * sb_permission - Check superblock-level permissions
419 * @sb: Superblock of inode to check permission on
420 * @inode: Inode to check permission on
421 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
423 * Separate out file-system wide checks from inode-specific permission checks.
425 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
427 if (unlikely(mask & MAY_WRITE)) {
428 umode_t mode = inode->i_mode;
430 /* Nobody gets write access to a read-only fs. */
431 if ((sb->s_flags & MS_RDONLY) &&
432 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
439 * inode_permission - Check for access rights to a given inode
440 * @inode: Inode to check permission on
441 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
443 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
444 * this, letting us set arbitrary permissions for filesystem access without
445 * changing the "normal" UIDs which are used for other things.
447 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
449 int inode_permission(struct inode *inode, int mask)
453 retval = sb_permission(inode->i_sb, inode, mask);
456 return __inode_permission(inode, mask);
460 * path_get - get a reference to a path
461 * @path: path to get the reference to
463 * Given a path increment the reference count to the dentry and the vfsmount.
465 void path_get(const struct path *path)
470 EXPORT_SYMBOL(path_get);
473 * path_put - put a reference to a path
474 * @path: path to put the reference to
476 * Given a path decrement the reference count to the dentry and the vfsmount.
478 void path_put(const struct path *path)
483 EXPORT_SYMBOL(path_put);
486 * Path walking has 2 modes, rcu-walk and ref-walk (see
487 * Documentation/filesystems/path-lookup.txt). In situations when we can't
488 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
489 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
490 * mode. Refcounts are grabbed at the last known good point before rcu-walk
491 * got stuck, so ref-walk may continue from there. If this is not successful
492 * (eg. a seqcount has changed), then failure is returned and it's up to caller
493 * to restart the path walk from the beginning in ref-walk mode.
497 * unlazy_walk - try to switch to ref-walk mode.
498 * @nd: nameidata pathwalk data
499 * @dentry: child of nd->path.dentry or NULL
500 * Returns: 0 on success, -ECHILD on failure
502 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
503 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
504 * @nd or NULL. Must be called from rcu-walk context.
506 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
508 struct fs_struct *fs = current->fs;
509 struct dentry *parent = nd->path.dentry;
511 BUG_ON(!(nd->flags & LOOKUP_RCU));
514 * After legitimizing the bastards, terminate_walk()
515 * will do the right thing for non-RCU mode, and all our
516 * subsequent exit cases should rcu_read_unlock()
517 * before returning. Do vfsmount first; if dentry
518 * can't be legitimized, just set nd->path.dentry to NULL
519 * and rely on dput(NULL) being a no-op.
521 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
523 nd->flags &= ~LOOKUP_RCU;
525 if (!lockref_get_not_dead(&parent->d_lockref)) {
526 nd->path.dentry = NULL;
531 * For a negative lookup, the lookup sequence point is the parents
532 * sequence point, and it only needs to revalidate the parent dentry.
534 * For a positive lookup, we need to move both the parent and the
535 * dentry from the RCU domain to be properly refcounted. And the
536 * sequence number in the dentry validates *both* dentry counters,
537 * since we checked the sequence number of the parent after we got
538 * the child sequence number. So we know the parent must still
539 * be valid if the child sequence number is still valid.
542 if (read_seqcount_retry(&parent->d_seq, nd->seq))
544 BUG_ON(nd->inode != parent->d_inode);
546 if (!lockref_get_not_dead(&dentry->d_lockref))
548 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
553 * Sequence counts matched. Now make sure that the root is
554 * still valid and get it if required.
556 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
557 spin_lock(&fs->lock);
558 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
559 goto unlock_and_drop_dentry;
561 spin_unlock(&fs->lock);
567 unlock_and_drop_dentry:
568 spin_unlock(&fs->lock);
576 if (!(nd->flags & LOOKUP_ROOT))
581 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
583 return dentry->d_op->d_revalidate(dentry, flags);
587 * complete_walk - successful completion of path walk
588 * @nd: pointer nameidata
590 * If we had been in RCU mode, drop out of it and legitimize nd->path.
591 * Revalidate the final result, unless we'd already done that during
592 * the path walk or the filesystem doesn't ask for it. Return 0 on
593 * success, -error on failure. In case of failure caller does not
594 * need to drop nd->path.
596 static int complete_walk(struct nameidata *nd)
598 struct dentry *dentry = nd->path.dentry;
601 if (nd->flags & LOOKUP_RCU) {
602 nd->flags &= ~LOOKUP_RCU;
603 if (!(nd->flags & LOOKUP_ROOT))
606 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
610 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
612 mntput(nd->path.mnt);
615 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
618 mntput(nd->path.mnt);
624 if (likely(!(nd->flags & LOOKUP_JUMPED)))
627 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
630 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
641 static __always_inline void set_root(struct nameidata *nd)
644 get_fs_root(current->fs, &nd->root);
647 static int link_path_walk(const char *, struct nameidata *);
649 static __always_inline void set_root_rcu(struct nameidata *nd)
652 struct fs_struct *fs = current->fs;
656 seq = read_seqcount_begin(&fs->seq);
658 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
659 } while (read_seqcount_retry(&fs->seq, seq));
663 static void path_put_conditional(struct path *path, struct nameidata *nd)
666 if (path->mnt != nd->path.mnt)
670 static inline void path_to_nameidata(const struct path *path,
671 struct nameidata *nd)
673 if (!(nd->flags & LOOKUP_RCU)) {
674 dput(nd->path.dentry);
675 if (nd->path.mnt != path->mnt)
676 mntput(nd->path.mnt);
678 nd->path.mnt = path->mnt;
679 nd->path.dentry = path->dentry;
683 * Helper to directly jump to a known parsed path from ->follow_link,
684 * caller must have taken a reference to path beforehand.
686 void nd_jump_link(struct nameidata *nd, struct path *path)
691 nd->inode = nd->path.dentry->d_inode;
692 nd->flags |= LOOKUP_JUMPED;
695 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
697 struct inode *inode = link->dentry->d_inode;
698 if (inode->i_op->put_link)
699 inode->i_op->put_link(link->dentry, nd, cookie);
703 int sysctl_protected_symlinks __read_mostly = 0;
704 int sysctl_protected_hardlinks __read_mostly = 0;
707 * may_follow_link - Check symlink following for unsafe situations
708 * @link: The path of the symlink
709 * @nd: nameidata pathwalk data
711 * In the case of the sysctl_protected_symlinks sysctl being enabled,
712 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
713 * in a sticky world-writable directory. This is to protect privileged
714 * processes from failing races against path names that may change out
715 * from under them by way of other users creating malicious symlinks.
716 * It will permit symlinks to be followed only when outside a sticky
717 * world-writable directory, or when the uid of the symlink and follower
718 * match, or when the directory owner matches the symlink's owner.
720 * Returns 0 if following the symlink is allowed, -ve on error.
722 static inline int may_follow_link(struct path *link, struct nameidata *nd)
724 const struct inode *inode;
725 const struct inode *parent;
727 if (!sysctl_protected_symlinks)
730 /* Allowed if owner and follower match. */
731 inode = link->dentry->d_inode;
732 if (uid_eq(current_cred()->fsuid, inode->i_uid))
735 /* Allowed if parent directory not sticky and world-writable. */
736 parent = nd->path.dentry->d_inode;
737 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
740 /* Allowed if parent directory and link owner match. */
741 if (uid_eq(parent->i_uid, inode->i_uid))
744 audit_log_link_denied("follow_link", link);
745 path_put_conditional(link, nd);
751 * safe_hardlink_source - Check for safe hardlink conditions
752 * @inode: the source inode to hardlink from
754 * Return false if at least one of the following conditions:
755 * - inode is not a regular file
757 * - inode is setgid and group-exec
758 * - access failure for read and write
760 * Otherwise returns true.
762 static bool safe_hardlink_source(struct inode *inode)
764 umode_t mode = inode->i_mode;
766 /* Special files should not get pinned to the filesystem. */
770 /* Setuid files should not get pinned to the filesystem. */
774 /* Executable setgid files should not get pinned to the filesystem. */
775 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
778 /* Hardlinking to unreadable or unwritable sources is dangerous. */
779 if (inode_permission(inode, MAY_READ | MAY_WRITE))
786 * may_linkat - Check permissions for creating a hardlink
787 * @link: the source to hardlink from
789 * Block hardlink when all of:
790 * - sysctl_protected_hardlinks enabled
791 * - fsuid does not match inode
792 * - hardlink source is unsafe (see safe_hardlink_source() above)
795 * Returns 0 if successful, -ve on error.
797 static int may_linkat(struct path *link)
799 const struct cred *cred;
802 if (!sysctl_protected_hardlinks)
805 cred = current_cred();
806 inode = link->dentry->d_inode;
808 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
809 * otherwise, it must be a safe source.
811 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
815 audit_log_link_denied("linkat", link);
819 static __always_inline int
820 follow_link(struct path *link, struct nameidata *nd, void **p)
822 struct dentry *dentry = link->dentry;
826 BUG_ON(nd->flags & LOOKUP_RCU);
828 if (link->mnt == nd->path.mnt)
832 if (unlikely(current->total_link_count >= 40))
833 goto out_put_nd_path;
836 current->total_link_count++;
839 nd_set_link(nd, NULL);
841 error = security_inode_follow_link(link->dentry, nd);
843 goto out_put_nd_path;
845 nd->last_type = LAST_BIND;
846 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
849 goto out_put_nd_path;
854 if (unlikely(IS_ERR(s))) {
856 put_link(nd, link, *p);
864 nd->flags |= LOOKUP_JUMPED;
866 nd->inode = nd->path.dentry->d_inode;
867 error = link_path_walk(s, nd);
869 put_link(nd, link, *p);
881 static int follow_up_rcu(struct path *path)
883 struct mount *mnt = real_mount(path->mnt);
884 struct mount *parent;
885 struct dentry *mountpoint;
887 parent = mnt->mnt_parent;
888 if (&parent->mnt == path->mnt)
890 mountpoint = mnt->mnt_mountpoint;
891 path->dentry = mountpoint;
892 path->mnt = &parent->mnt;
897 * follow_up - Find the mountpoint of path's vfsmount
899 * Given a path, find the mountpoint of its source file system.
900 * Replace @path with the path of the mountpoint in the parent mount.
903 * Return 1 if we went up a level and 0 if we were already at the
906 int follow_up(struct path *path)
908 struct mount *mnt = real_mount(path->mnt);
909 struct mount *parent;
910 struct dentry *mountpoint;
912 read_seqlock_excl(&mount_lock);
913 parent = mnt->mnt_parent;
915 read_sequnlock_excl(&mount_lock);
918 mntget(&parent->mnt);
919 mountpoint = dget(mnt->mnt_mountpoint);
920 read_sequnlock_excl(&mount_lock);
922 path->dentry = mountpoint;
924 path->mnt = &parent->mnt;
929 * Perform an automount
930 * - return -EISDIR to tell follow_managed() to stop and return the path we
933 static int follow_automount(struct path *path, unsigned flags,
936 struct vfsmount *mnt;
939 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
942 /* We don't want to mount if someone's just doing a stat -
943 * unless they're stat'ing a directory and appended a '/' to
946 * We do, however, want to mount if someone wants to open or
947 * create a file of any type under the mountpoint, wants to
948 * traverse through the mountpoint or wants to open the
949 * mounted directory. Also, autofs may mark negative dentries
950 * as being automount points. These will need the attentions
951 * of the daemon to instantiate them before they can be used.
953 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
954 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
955 path->dentry->d_inode)
958 current->total_link_count++;
959 if (current->total_link_count >= 40)
962 mnt = path->dentry->d_op->d_automount(path);
965 * The filesystem is allowed to return -EISDIR here to indicate
966 * it doesn't want to automount. For instance, autofs would do
967 * this so that its userspace daemon can mount on this dentry.
969 * However, we can only permit this if it's a terminal point in
970 * the path being looked up; if it wasn't then the remainder of
971 * the path is inaccessible and we should say so.
973 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
978 if (!mnt) /* mount collision */
982 /* lock_mount() may release path->mnt on error */
986 err = finish_automount(mnt, path);
990 /* Someone else made a mount here whilst we were busy */
995 path->dentry = dget(mnt->mnt_root);
1004 * Handle a dentry that is managed in some way.
1005 * - Flagged for transit management (autofs)
1006 * - Flagged as mountpoint
1007 * - Flagged as automount point
1009 * This may only be called in refwalk mode.
1011 * Serialization is taken care of in namespace.c
1013 static int follow_managed(struct path *path, unsigned flags)
1015 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1017 bool need_mntput = false;
1020 /* Given that we're not holding a lock here, we retain the value in a
1021 * local variable for each dentry as we look at it so that we don't see
1022 * the components of that value change under us */
1023 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1024 managed &= DCACHE_MANAGED_DENTRY,
1025 unlikely(managed != 0)) {
1026 /* Allow the filesystem to manage the transit without i_mutex
1028 if (managed & DCACHE_MANAGE_TRANSIT) {
1029 BUG_ON(!path->dentry->d_op);
1030 BUG_ON(!path->dentry->d_op->d_manage);
1031 ret = path->dentry->d_op->d_manage(path->dentry, false);
1036 /* Transit to a mounted filesystem. */
1037 if (managed & DCACHE_MOUNTED) {
1038 struct vfsmount *mounted = lookup_mnt(path);
1043 path->mnt = mounted;
1044 path->dentry = dget(mounted->mnt_root);
1049 /* Something is mounted on this dentry in another
1050 * namespace and/or whatever was mounted there in this
1051 * namespace got unmounted before lookup_mnt() could
1055 /* Handle an automount point */
1056 if (managed & DCACHE_NEED_AUTOMOUNT) {
1057 ret = follow_automount(path, flags, &need_mntput);
1063 /* We didn't change the current path point */
1067 if (need_mntput && path->mnt == mnt)
1071 return ret < 0 ? ret : need_mntput;
1074 int follow_down_one(struct path *path)
1076 struct vfsmount *mounted;
1078 mounted = lookup_mnt(path);
1082 path->mnt = mounted;
1083 path->dentry = dget(mounted->mnt_root);
1089 static inline bool managed_dentry_might_block(struct dentry *dentry)
1091 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1092 dentry->d_op->d_manage(dentry, true) < 0);
1096 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1097 * we meet a managed dentry that would need blocking.
1099 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1100 struct inode **inode)
1103 struct mount *mounted;
1105 * Don't forget we might have a non-mountpoint managed dentry
1106 * that wants to block transit.
1108 if (unlikely(managed_dentry_might_block(path->dentry)))
1111 if (!d_mountpoint(path->dentry))
1114 mounted = __lookup_mnt(path->mnt, path->dentry);
1117 path->mnt = &mounted->mnt;
1118 path->dentry = mounted->mnt.mnt_root;
1119 nd->flags |= LOOKUP_JUMPED;
1120 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1122 * Update the inode too. We don't need to re-check the
1123 * dentry sequence number here after this d_inode read,
1124 * because a mount-point is always pinned.
1126 *inode = path->dentry->d_inode;
1128 return read_seqretry(&mount_lock, nd->m_seq);
1131 static int follow_dotdot_rcu(struct nameidata *nd)
1136 if (nd->path.dentry == nd->root.dentry &&
1137 nd->path.mnt == nd->root.mnt) {
1140 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1141 struct dentry *old = nd->path.dentry;
1142 struct dentry *parent = old->d_parent;
1145 seq = read_seqcount_begin(&parent->d_seq);
1146 if (read_seqcount_retry(&old->d_seq, nd->seq))
1148 nd->path.dentry = parent;
1152 if (!follow_up_rcu(&nd->path))
1154 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1156 while (d_mountpoint(nd->path.dentry)) {
1157 struct mount *mounted;
1158 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1161 nd->path.mnt = &mounted->mnt;
1162 nd->path.dentry = mounted->mnt.mnt_root;
1163 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1164 if (!read_seqretry(&mount_lock, nd->m_seq))
1167 nd->inode = nd->path.dentry->d_inode;
1171 nd->flags &= ~LOOKUP_RCU;
1172 if (!(nd->flags & LOOKUP_ROOT))
1173 nd->root.mnt = NULL;
1179 * Follow down to the covering mount currently visible to userspace. At each
1180 * point, the filesystem owning that dentry may be queried as to whether the
1181 * caller is permitted to proceed or not.
1183 int follow_down(struct path *path)
1188 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1189 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1190 /* Allow the filesystem to manage the transit without i_mutex
1193 * We indicate to the filesystem if someone is trying to mount
1194 * something here. This gives autofs the chance to deny anyone
1195 * other than its daemon the right to mount on its
1198 * The filesystem may sleep at this point.
1200 if (managed & DCACHE_MANAGE_TRANSIT) {
1201 BUG_ON(!path->dentry->d_op);
1202 BUG_ON(!path->dentry->d_op->d_manage);
1203 ret = path->dentry->d_op->d_manage(
1204 path->dentry, false);
1206 return ret == -EISDIR ? 0 : ret;
1209 /* Transit to a mounted filesystem. */
1210 if (managed & DCACHE_MOUNTED) {
1211 struct vfsmount *mounted = lookup_mnt(path);
1216 path->mnt = mounted;
1217 path->dentry = dget(mounted->mnt_root);
1221 /* Don't handle automount points here */
1228 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1230 static void follow_mount(struct path *path)
1232 while (d_mountpoint(path->dentry)) {
1233 struct vfsmount *mounted = lookup_mnt(path);
1238 path->mnt = mounted;
1239 path->dentry = dget(mounted->mnt_root);
1243 static void follow_dotdot(struct nameidata *nd)
1248 struct dentry *old = nd->path.dentry;
1250 if (nd->path.dentry == nd->root.dentry &&
1251 nd->path.mnt == nd->root.mnt) {
1254 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1255 /* rare case of legitimate dget_parent()... */
1256 nd->path.dentry = dget_parent(nd->path.dentry);
1260 if (!follow_up(&nd->path))
1263 follow_mount(&nd->path);
1264 nd->inode = nd->path.dentry->d_inode;
1268 * This looks up the name in dcache, possibly revalidates the old dentry and
1269 * allocates a new one if not found or not valid. In the need_lookup argument
1270 * returns whether i_op->lookup is necessary.
1272 * dir->d_inode->i_mutex must be held
1274 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1275 unsigned int flags, bool *need_lookup)
1277 struct dentry *dentry;
1280 *need_lookup = false;
1281 dentry = d_lookup(dir, name);
1283 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1284 error = d_revalidate(dentry, flags);
1285 if (unlikely(error <= 0)) {
1288 return ERR_PTR(error);
1289 } else if (!d_invalidate(dentry)) {
1298 dentry = d_alloc(dir, name);
1299 if (unlikely(!dentry))
1300 return ERR_PTR(-ENOMEM);
1302 *need_lookup = true;
1308 * Call i_op->lookup on the dentry. The dentry must be negative and
1311 * dir->d_inode->i_mutex must be held
1313 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1318 /* Don't create child dentry for a dead directory. */
1319 if (unlikely(IS_DEADDIR(dir))) {
1321 return ERR_PTR(-ENOENT);
1324 old = dir->i_op->lookup(dir, dentry, flags);
1325 if (unlikely(old)) {
1332 static struct dentry *__lookup_hash(struct qstr *name,
1333 struct dentry *base, unsigned int flags)
1336 struct dentry *dentry;
1338 dentry = lookup_dcache(name, base, flags, &need_lookup);
1342 return lookup_real(base->d_inode, dentry, flags);
1346 * It's more convoluted than I'd like it to be, but... it's still fairly
1347 * small and for now I'd prefer to have fast path as straight as possible.
1348 * It _is_ time-critical.
1350 static int lookup_fast(struct nameidata *nd,
1351 struct path *path, struct inode **inode)
1353 struct vfsmount *mnt = nd->path.mnt;
1354 struct dentry *dentry, *parent = nd->path.dentry;
1360 * Rename seqlock is not required here because in the off chance
1361 * of a false negative due to a concurrent rename, we're going to
1362 * do the non-racy lookup, below.
1364 if (nd->flags & LOOKUP_RCU) {
1366 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1371 * This sequence count validates that the inode matches
1372 * the dentry name information from lookup.
1374 *inode = dentry->d_inode;
1375 if (read_seqcount_retry(&dentry->d_seq, seq))
1379 * This sequence count validates that the parent had no
1380 * changes while we did the lookup of the dentry above.
1382 * The memory barrier in read_seqcount_begin of child is
1383 * enough, we can use __read_seqcount_retry here.
1385 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1389 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1390 status = d_revalidate(dentry, nd->flags);
1391 if (unlikely(status <= 0)) {
1392 if (status != -ECHILD)
1398 path->dentry = dentry;
1399 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1401 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1405 if (unlazy_walk(nd, dentry))
1408 dentry = __d_lookup(parent, &nd->last);
1411 if (unlikely(!dentry))
1414 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1415 status = d_revalidate(dentry, nd->flags);
1416 if (unlikely(status <= 0)) {
1421 if (!d_invalidate(dentry)) {
1428 path->dentry = dentry;
1429 err = follow_managed(path, nd->flags);
1430 if (unlikely(err < 0)) {
1431 path_put_conditional(path, nd);
1435 nd->flags |= LOOKUP_JUMPED;
1436 *inode = path->dentry->d_inode;
1443 /* Fast lookup failed, do it the slow way */
1444 static int lookup_slow(struct nameidata *nd, struct path *path)
1446 struct dentry *dentry, *parent;
1449 parent = nd->path.dentry;
1450 BUG_ON(nd->inode != parent->d_inode);
1452 mutex_lock(&parent->d_inode->i_mutex);
1453 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1454 mutex_unlock(&parent->d_inode->i_mutex);
1456 return PTR_ERR(dentry);
1457 path->mnt = nd->path.mnt;
1458 path->dentry = dentry;
1459 err = follow_managed(path, nd->flags);
1460 if (unlikely(err < 0)) {
1461 path_put_conditional(path, nd);
1465 nd->flags |= LOOKUP_JUMPED;
1469 static inline int may_lookup(struct nameidata *nd)
1471 if (nd->flags & LOOKUP_RCU) {
1472 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1475 if (unlazy_walk(nd, NULL))
1478 return inode_permission(nd->inode, MAY_EXEC);
1481 static inline int handle_dots(struct nameidata *nd, int type)
1483 if (type == LAST_DOTDOT) {
1484 if (nd->flags & LOOKUP_RCU) {
1485 if (follow_dotdot_rcu(nd))
1493 static void terminate_walk(struct nameidata *nd)
1495 if (!(nd->flags & LOOKUP_RCU)) {
1496 path_put(&nd->path);
1498 nd->flags &= ~LOOKUP_RCU;
1499 if (!(nd->flags & LOOKUP_ROOT))
1500 nd->root.mnt = NULL;
1506 * Do we need to follow links? We _really_ want to be able
1507 * to do this check without having to look at inode->i_op,
1508 * so we keep a cache of "no, this doesn't need follow_link"
1509 * for the common case.
1511 static inline int should_follow_link(struct dentry *dentry, int follow)
1513 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1516 static inline int walk_component(struct nameidata *nd, struct path *path,
1519 struct inode *inode;
1522 * "." and ".." are special - ".." especially so because it has
1523 * to be able to know about the current root directory and
1524 * parent relationships.
1526 if (unlikely(nd->last_type != LAST_NORM))
1527 return handle_dots(nd, nd->last_type);
1528 err = lookup_fast(nd, path, &inode);
1529 if (unlikely(err)) {
1533 err = lookup_slow(nd, path);
1537 inode = path->dentry->d_inode;
1543 if (should_follow_link(path->dentry, follow)) {
1544 if (nd->flags & LOOKUP_RCU) {
1545 if (unlikely(unlazy_walk(nd, path->dentry))) {
1550 BUG_ON(inode != path->dentry->d_inode);
1553 path_to_nameidata(path, nd);
1558 path_to_nameidata(path, nd);
1565 * This limits recursive symlink follows to 8, while
1566 * limiting consecutive symlinks to 40.
1568 * Without that kind of total limit, nasty chains of consecutive
1569 * symlinks can cause almost arbitrarily long lookups.
1571 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1575 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1576 path_put_conditional(path, nd);
1577 path_put(&nd->path);
1580 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1583 current->link_count++;
1586 struct path link = *path;
1589 res = follow_link(&link, nd, &cookie);
1592 res = walk_component(nd, path, LOOKUP_FOLLOW);
1593 put_link(nd, &link, cookie);
1596 current->link_count--;
1602 * We can do the critical dentry name comparison and hashing
1603 * operations one word at a time, but we are limited to:
1605 * - Architectures with fast unaligned word accesses. We could
1606 * do a "get_unaligned()" if this helps and is sufficiently
1609 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1610 * do not trap on the (extremely unlikely) case of a page
1611 * crossing operation.
1613 * - Furthermore, we need an efficient 64-bit compile for the
1614 * 64-bit case in order to generate the "number of bytes in
1615 * the final mask". Again, that could be replaced with a
1616 * efficient population count instruction or similar.
1618 #ifdef CONFIG_DCACHE_WORD_ACCESS
1620 #include <asm/word-at-a-time.h>
1624 static inline unsigned int fold_hash(unsigned long hash)
1626 hash += hash >> (8*sizeof(int));
1630 #else /* 32-bit case */
1632 #define fold_hash(x) (x)
1636 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1638 unsigned long a, mask;
1639 unsigned long hash = 0;
1642 a = load_unaligned_zeropad(name);
1643 if (len < sizeof(unsigned long))
1647 name += sizeof(unsigned long);
1648 len -= sizeof(unsigned long);
1652 mask = bytemask_from_count(len);
1655 return fold_hash(hash);
1657 EXPORT_SYMBOL(full_name_hash);
1660 * Calculate the length and hash of the path component, and
1661 * return the length of the component;
1663 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1665 unsigned long a, b, adata, bdata, mask, hash, len;
1666 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1669 len = -sizeof(unsigned long);
1671 hash = (hash + a) * 9;
1672 len += sizeof(unsigned long);
1673 a = load_unaligned_zeropad(name+len);
1674 b = a ^ REPEAT_BYTE('/');
1675 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1677 adata = prep_zero_mask(a, adata, &constants);
1678 bdata = prep_zero_mask(b, bdata, &constants);
1680 mask = create_zero_mask(adata | bdata);
1682 hash += a & zero_bytemask(mask);
1683 *hashp = fold_hash(hash);
1685 return len + find_zero(mask);
1690 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1692 unsigned long hash = init_name_hash();
1694 hash = partial_name_hash(*name++, hash);
1695 return end_name_hash(hash);
1697 EXPORT_SYMBOL(full_name_hash);
1700 * We know there's a real path component here of at least
1703 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1705 unsigned long hash = init_name_hash();
1706 unsigned long len = 0, c;
1708 c = (unsigned char)*name;
1711 hash = partial_name_hash(c, hash);
1712 c = (unsigned char)name[len];
1713 } while (c && c != '/');
1714 *hashp = end_name_hash(hash);
1722 * This is the basic name resolution function, turning a pathname into
1723 * the final dentry. We expect 'base' to be positive and a directory.
1725 * Returns 0 and nd will have valid dentry and mnt on success.
1726 * Returns error and drops reference to input namei data on failure.
1728 static int link_path_walk(const char *name, struct nameidata *nd)
1738 /* At this point we know we have a real path component. */
1744 err = may_lookup(nd);
1748 len = hash_name(name, &this.hash);
1753 if (name[0] == '.') switch (len) {
1755 if (name[1] == '.') {
1757 nd->flags |= LOOKUP_JUMPED;
1763 if (likely(type == LAST_NORM)) {
1764 struct dentry *parent = nd->path.dentry;
1765 nd->flags &= ~LOOKUP_JUMPED;
1766 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1767 err = parent->d_op->d_hash(parent, &this);
1774 nd->last_type = type;
1779 * If it wasn't NUL, we know it was '/'. Skip that
1780 * slash, and continue until no more slashes.
1784 } while (unlikely(name[len] == '/'));
1790 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1795 err = nested_symlink(&next, nd);
1799 if (!d_is_directory(nd->path.dentry)) {
1808 static int path_init(int dfd, const char *name, unsigned int flags,
1809 struct nameidata *nd, struct file **fp)
1813 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1814 nd->flags = flags | LOOKUP_JUMPED;
1816 if (flags & LOOKUP_ROOT) {
1817 struct dentry *root = nd->root.dentry;
1818 struct inode *inode = root->d_inode;
1820 if (!d_is_directory(root))
1822 retval = inode_permission(inode, MAY_EXEC);
1826 nd->path = nd->root;
1828 if (flags & LOOKUP_RCU) {
1830 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1831 nd->m_seq = read_seqbegin(&mount_lock);
1833 path_get(&nd->path);
1838 nd->root.mnt = NULL;
1840 nd->m_seq = read_seqbegin(&mount_lock);
1842 if (flags & LOOKUP_RCU) {
1847 path_get(&nd->root);
1849 nd->path = nd->root;
1850 } else if (dfd == AT_FDCWD) {
1851 if (flags & LOOKUP_RCU) {
1852 struct fs_struct *fs = current->fs;
1858 seq = read_seqcount_begin(&fs->seq);
1860 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1861 } while (read_seqcount_retry(&fs->seq, seq));
1863 get_fs_pwd(current->fs, &nd->path);
1866 /* Caller must check execute permissions on the starting path component */
1867 struct fd f = fdget_raw(dfd);
1868 struct dentry *dentry;
1873 dentry = f.file->f_path.dentry;
1876 if (!d_is_directory(dentry)) {
1882 nd->path = f.file->f_path;
1883 if (flags & LOOKUP_RCU) {
1884 if (f.flags & FDPUT_FPUT)
1886 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1889 path_get(&nd->path);
1894 nd->inode = nd->path.dentry->d_inode;
1898 static inline int lookup_last(struct nameidata *nd, struct path *path)
1900 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1901 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1903 nd->flags &= ~LOOKUP_PARENT;
1904 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1907 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1908 static int path_lookupat(int dfd, const char *name,
1909 unsigned int flags, struct nameidata *nd)
1911 struct file *base = NULL;
1916 * Path walking is largely split up into 2 different synchronisation
1917 * schemes, rcu-walk and ref-walk (explained in
1918 * Documentation/filesystems/path-lookup.txt). These share much of the
1919 * path walk code, but some things particularly setup, cleanup, and
1920 * following mounts are sufficiently divergent that functions are
1921 * duplicated. Typically there is a function foo(), and its RCU
1922 * analogue, foo_rcu().
1924 * -ECHILD is the error number of choice (just to avoid clashes) that
1925 * is returned if some aspect of an rcu-walk fails. Such an error must
1926 * be handled by restarting a traditional ref-walk (which will always
1927 * be able to complete).
1929 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1934 current->total_link_count = 0;
1935 err = link_path_walk(name, nd);
1937 if (!err && !(flags & LOOKUP_PARENT)) {
1938 err = lookup_last(nd, &path);
1941 struct path link = path;
1942 err = may_follow_link(&link, nd);
1945 nd->flags |= LOOKUP_PARENT;
1946 err = follow_link(&link, nd, &cookie);
1949 err = lookup_last(nd, &path);
1950 put_link(nd, &link, cookie);
1955 err = complete_walk(nd);
1957 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1958 if (!d_is_directory(nd->path.dentry)) {
1959 path_put(&nd->path);
1967 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1968 path_put(&nd->root);
1969 nd->root.mnt = NULL;
1974 static int filename_lookup(int dfd, struct filename *name,
1975 unsigned int flags, struct nameidata *nd)
1977 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
1978 if (unlikely(retval == -ECHILD))
1979 retval = path_lookupat(dfd, name->name, flags, nd);
1980 if (unlikely(retval == -ESTALE))
1981 retval = path_lookupat(dfd, name->name,
1982 flags | LOOKUP_REVAL, nd);
1984 if (likely(!retval))
1985 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
1989 static int do_path_lookup(int dfd, const char *name,
1990 unsigned int flags, struct nameidata *nd)
1992 struct filename filename = { .name = name };
1994 return filename_lookup(dfd, &filename, flags, nd);
1997 /* does lookup, returns the object with parent locked */
1998 struct dentry *kern_path_locked(const char *name, struct path *path)
2000 struct nameidata nd;
2002 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2004 return ERR_PTR(err);
2005 if (nd.last_type != LAST_NORM) {
2007 return ERR_PTR(-EINVAL);
2009 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2010 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2012 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2020 int kern_path(const char *name, unsigned int flags, struct path *path)
2022 struct nameidata nd;
2023 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2030 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2031 * @dentry: pointer to dentry of the base directory
2032 * @mnt: pointer to vfs mount of the base directory
2033 * @name: pointer to file name
2034 * @flags: lookup flags
2035 * @path: pointer to struct path to fill
2037 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2038 const char *name, unsigned int flags,
2041 struct nameidata nd;
2043 nd.root.dentry = dentry;
2045 BUG_ON(flags & LOOKUP_PARENT);
2046 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2047 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2054 * Restricted form of lookup. Doesn't follow links, single-component only,
2055 * needs parent already locked. Doesn't follow mounts.
2058 static struct dentry *lookup_hash(struct nameidata *nd)
2060 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2064 * lookup_one_len - filesystem helper to lookup single pathname component
2065 * @name: pathname component to lookup
2066 * @base: base directory to lookup from
2067 * @len: maximum length @len should be interpreted to
2069 * Note that this routine is purely a helper for filesystem usage and should
2070 * not be called by generic code. Also note that by using this function the
2071 * nameidata argument is passed to the filesystem methods and a filesystem
2072 * using this helper needs to be prepared for that.
2074 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2080 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2084 this.hash = full_name_hash(name, len);
2086 return ERR_PTR(-EACCES);
2088 if (unlikely(name[0] == '.')) {
2089 if (len < 2 || (len == 2 && name[1] == '.'))
2090 return ERR_PTR(-EACCES);
2094 c = *(const unsigned char *)name++;
2095 if (c == '/' || c == '\0')
2096 return ERR_PTR(-EACCES);
2099 * See if the low-level filesystem might want
2100 * to use its own hash..
2102 if (base->d_flags & DCACHE_OP_HASH) {
2103 int err = base->d_op->d_hash(base, &this);
2105 return ERR_PTR(err);
2108 err = inode_permission(base->d_inode, MAY_EXEC);
2110 return ERR_PTR(err);
2112 return __lookup_hash(&this, base, 0);
2115 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2116 struct path *path, int *empty)
2118 struct nameidata nd;
2119 struct filename *tmp = getname_flags(name, flags, empty);
2120 int err = PTR_ERR(tmp);
2123 BUG_ON(flags & LOOKUP_PARENT);
2125 err = filename_lookup(dfd, tmp, flags, &nd);
2133 int user_path_at(int dfd, const char __user *name, unsigned flags,
2136 return user_path_at_empty(dfd, name, flags, path, NULL);
2140 * NB: most callers don't do anything directly with the reference to the
2141 * to struct filename, but the nd->last pointer points into the name string
2142 * allocated by getname. So we must hold the reference to it until all
2143 * path-walking is complete.
2145 static struct filename *
2146 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2149 struct filename *s = getname(path);
2152 /* only LOOKUP_REVAL is allowed in extra flags */
2153 flags &= LOOKUP_REVAL;
2158 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2161 return ERR_PTR(error);
2168 * mountpoint_last - look up last component for umount
2169 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2170 * @path: pointer to container for result
2172 * This is a special lookup_last function just for umount. In this case, we
2173 * need to resolve the path without doing any revalidation.
2175 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2176 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2177 * in almost all cases, this lookup will be served out of the dcache. The only
2178 * cases where it won't are if nd->last refers to a symlink or the path is
2179 * bogus and it doesn't exist.
2182 * -error: if there was an error during lookup. This includes -ENOENT if the
2183 * lookup found a negative dentry. The nd->path reference will also be
2186 * 0: if we successfully resolved nd->path and found it to not to be a
2187 * symlink that needs to be followed. "path" will also be populated.
2188 * The nd->path reference will also be put.
2190 * 1: if we successfully resolved nd->last and found it to be a symlink
2191 * that needs to be followed. "path" will be populated with the path
2192 * to the link, and nd->path will *not* be put.
2195 mountpoint_last(struct nameidata *nd, struct path *path)
2198 struct dentry *dentry;
2199 struct dentry *dir = nd->path.dentry;
2201 /* If we're in rcuwalk, drop out of it to handle last component */
2202 if (nd->flags & LOOKUP_RCU) {
2203 if (unlazy_walk(nd, NULL)) {
2209 nd->flags &= ~LOOKUP_PARENT;
2211 if (unlikely(nd->last_type != LAST_NORM)) {
2212 error = handle_dots(nd, nd->last_type);
2215 dentry = dget(nd->path.dentry);
2219 mutex_lock(&dir->d_inode->i_mutex);
2220 dentry = d_lookup(dir, &nd->last);
2223 * No cached dentry. Mounted dentries are pinned in the cache,
2224 * so that means that this dentry is probably a symlink or the
2225 * path doesn't actually point to a mounted dentry.
2227 dentry = d_alloc(dir, &nd->last);
2230 mutex_unlock(&dir->d_inode->i_mutex);
2233 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2234 error = PTR_ERR(dentry);
2235 if (IS_ERR(dentry)) {
2236 mutex_unlock(&dir->d_inode->i_mutex);
2240 mutex_unlock(&dir->d_inode->i_mutex);
2243 if (!dentry->d_inode) {
2248 path->dentry = dentry;
2249 path->mnt = mntget(nd->path.mnt);
2250 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2260 * path_mountpoint - look up a path to be umounted
2261 * @dfd: directory file descriptor to start walk from
2262 * @name: full pathname to walk
2263 * @path: pointer to container for result
2264 * @flags: lookup flags
2266 * Look up the given name, but don't attempt to revalidate the last component.
2267 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2270 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2272 struct file *base = NULL;
2273 struct nameidata nd;
2276 err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2280 current->total_link_count = 0;
2281 err = link_path_walk(name, &nd);
2285 err = mountpoint_last(&nd, path);
2288 struct path link = *path;
2289 err = may_follow_link(&link, &nd);
2292 nd.flags |= LOOKUP_PARENT;
2293 err = follow_link(&link, &nd, &cookie);
2296 err = mountpoint_last(&nd, path);
2297 put_link(&nd, &link, cookie);
2303 if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2310 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2313 int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2314 if (unlikely(error == -ECHILD))
2315 error = path_mountpoint(dfd, s->name, path, flags);
2316 if (unlikely(error == -ESTALE))
2317 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2319 audit_inode(s, path->dentry, 0);
2324 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2325 * @dfd: directory file descriptor
2326 * @name: pathname from userland
2327 * @flags: lookup flags
2328 * @path: pointer to container to hold result
2330 * A umount is a special case for path walking. We're not actually interested
2331 * in the inode in this situation, and ESTALE errors can be a problem. We
2332 * simply want track down the dentry and vfsmount attached at the mountpoint
2333 * and avoid revalidating the last component.
2335 * Returns 0 and populates "path" on success.
2338 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2341 struct filename *s = getname(name);
2345 error = filename_mountpoint(dfd, s, path, flags);
2351 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2354 struct filename s = {.name = name};
2355 return filename_mountpoint(dfd, &s, path, flags);
2357 EXPORT_SYMBOL(kern_path_mountpoint);
2360 * It's inline, so penalty for filesystems that don't use sticky bit is
2363 static inline int check_sticky(struct inode *dir, struct inode *inode)
2365 kuid_t fsuid = current_fsuid();
2367 if (!(dir->i_mode & S_ISVTX))
2369 if (uid_eq(inode->i_uid, fsuid))
2371 if (uid_eq(dir->i_uid, fsuid))
2373 return !inode_capable(inode, CAP_FOWNER);
2377 * Check whether we can remove a link victim from directory dir, check
2378 * whether the type of victim is right.
2379 * 1. We can't do it if dir is read-only (done in permission())
2380 * 2. We should have write and exec permissions on dir
2381 * 3. We can't remove anything from append-only dir
2382 * 4. We can't do anything with immutable dir (done in permission())
2383 * 5. If the sticky bit on dir is set we should either
2384 * a. be owner of dir, or
2385 * b. be owner of victim, or
2386 * c. have CAP_FOWNER capability
2387 * 6. If the victim is append-only or immutable we can't do antyhing with
2388 * links pointing to it.
2389 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2390 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2391 * 9. We can't remove a root or mountpoint.
2392 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2393 * nfs_async_unlink().
2395 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2397 struct inode *inode = victim->d_inode;
2400 if (d_is_negative(victim))
2404 BUG_ON(victim->d_parent->d_inode != dir);
2405 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2407 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2413 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2414 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2417 if (!d_is_directory(victim) && !d_is_autodir(victim))
2419 if (IS_ROOT(victim))
2421 } else if (d_is_directory(victim) || d_is_autodir(victim))
2423 if (IS_DEADDIR(dir))
2425 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2430 /* Check whether we can create an object with dentry child in directory
2432 * 1. We can't do it if child already exists (open has special treatment for
2433 * this case, but since we are inlined it's OK)
2434 * 2. We can't do it if dir is read-only (done in permission())
2435 * 3. We should have write and exec permissions on dir
2436 * 4. We can't do it if dir is immutable (done in permission())
2438 static inline int may_create(struct inode *dir, struct dentry *child)
2440 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2443 if (IS_DEADDIR(dir))
2445 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2449 * p1 and p2 should be directories on the same fs.
2451 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2456 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2460 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2462 p = d_ancestor(p2, p1);
2464 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2465 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2469 p = d_ancestor(p1, p2);
2471 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2472 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2476 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2477 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2481 void unlock_rename(struct dentry *p1, struct dentry *p2)
2483 mutex_unlock(&p1->d_inode->i_mutex);
2485 mutex_unlock(&p2->d_inode->i_mutex);
2486 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2490 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2493 int error = may_create(dir, dentry);
2497 if (!dir->i_op->create)
2498 return -EACCES; /* shouldn't it be ENOSYS? */
2501 error = security_inode_create(dir, dentry, mode);
2504 error = dir->i_op->create(dir, dentry, mode, want_excl);
2506 fsnotify_create(dir, dentry);
2510 static int may_open(struct path *path, int acc_mode, int flag)
2512 struct dentry *dentry = path->dentry;
2513 struct inode *inode = dentry->d_inode;
2523 switch (inode->i_mode & S_IFMT) {
2527 if (acc_mode & MAY_WRITE)
2532 if (path->mnt->mnt_flags & MNT_NODEV)
2541 error = inode_permission(inode, acc_mode);
2546 * An append-only file must be opened in append mode for writing.
2548 if (IS_APPEND(inode)) {
2549 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2555 /* O_NOATIME can only be set by the owner or superuser */
2556 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2562 static int handle_truncate(struct file *filp)
2564 struct path *path = &filp->f_path;
2565 struct inode *inode = path->dentry->d_inode;
2566 int error = get_write_access(inode);
2570 * Refuse to truncate files with mandatory locks held on them.
2572 error = locks_verify_locked(inode);
2574 error = security_path_truncate(path);
2576 error = do_truncate(path->dentry, 0,
2577 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2580 put_write_access(inode);
2584 static inline int open_to_namei_flags(int flag)
2586 if ((flag & O_ACCMODE) == 3)
2591 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2593 int error = security_path_mknod(dir, dentry, mode, 0);
2597 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2601 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2605 * Attempt to atomically look up, create and open a file from a negative
2608 * Returns 0 if successful. The file will have been created and attached to
2609 * @file by the filesystem calling finish_open().
2611 * Returns 1 if the file was looked up only or didn't need creating. The
2612 * caller will need to perform the open themselves. @path will have been
2613 * updated to point to the new dentry. This may be negative.
2615 * Returns an error code otherwise.
2617 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2618 struct path *path, struct file *file,
2619 const struct open_flags *op,
2620 bool got_write, bool need_lookup,
2623 struct inode *dir = nd->path.dentry->d_inode;
2624 unsigned open_flag = open_to_namei_flags(op->open_flag);
2628 int create_error = 0;
2629 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2632 BUG_ON(dentry->d_inode);
2634 /* Don't create child dentry for a dead directory. */
2635 if (unlikely(IS_DEADDIR(dir))) {
2641 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2642 mode &= ~current_umask();
2644 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2646 open_flag &= ~O_TRUNC;
2649 * Checking write permission is tricky, bacuse we don't know if we are
2650 * going to actually need it: O_CREAT opens should work as long as the
2651 * file exists. But checking existence breaks atomicity. The trick is
2652 * to check access and if not granted clear O_CREAT from the flags.
2654 * Another problem is returing the "right" error value (e.g. for an
2655 * O_EXCL open we want to return EEXIST not EROFS).
2657 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2658 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2659 if (!(open_flag & O_CREAT)) {
2661 * No O_CREATE -> atomicity not a requirement -> fall
2662 * back to lookup + open
2665 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2666 /* Fall back and fail with the right error */
2667 create_error = -EROFS;
2670 /* No side effects, safe to clear O_CREAT */
2671 create_error = -EROFS;
2672 open_flag &= ~O_CREAT;
2676 if (open_flag & O_CREAT) {
2677 error = may_o_create(&nd->path, dentry, mode);
2679 create_error = error;
2680 if (open_flag & O_EXCL)
2682 open_flag &= ~O_CREAT;
2686 if (nd->flags & LOOKUP_DIRECTORY)
2687 open_flag |= O_DIRECTORY;
2689 file->f_path.dentry = DENTRY_NOT_SET;
2690 file->f_path.mnt = nd->path.mnt;
2691 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2694 if (create_error && error == -ENOENT)
2695 error = create_error;
2699 if (error) { /* returned 1, that is */
2700 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2704 if (file->f_path.dentry) {
2706 dentry = file->f_path.dentry;
2708 if (*opened & FILE_CREATED)
2709 fsnotify_create(dir, dentry);
2710 if (!dentry->d_inode) {
2711 WARN_ON(*opened & FILE_CREATED);
2713 error = create_error;
2717 if (excl && !(*opened & FILE_CREATED)) {
2726 * We didn't have the inode before the open, so check open permission
2729 acc_mode = op->acc_mode;
2730 if (*opened & FILE_CREATED) {
2731 WARN_ON(!(open_flag & O_CREAT));
2732 fsnotify_create(dir, dentry);
2733 acc_mode = MAY_OPEN;
2735 error = may_open(&file->f_path, acc_mode, open_flag);
2745 dentry = lookup_real(dir, dentry, nd->flags);
2747 return PTR_ERR(dentry);
2750 int open_flag = op->open_flag;
2752 error = create_error;
2753 if ((open_flag & O_EXCL)) {
2754 if (!dentry->d_inode)
2756 } else if (!dentry->d_inode) {
2758 } else if ((open_flag & O_TRUNC) &&
2759 S_ISREG(dentry->d_inode->i_mode)) {
2762 /* will fail later, go on to get the right error */
2766 path->dentry = dentry;
2767 path->mnt = nd->path.mnt;
2772 * Look up and maybe create and open the last component.
2774 * Must be called with i_mutex held on parent.
2776 * Returns 0 if the file was successfully atomically created (if necessary) and
2777 * opened. In this case the file will be returned attached to @file.
2779 * Returns 1 if the file was not completely opened at this time, though lookups
2780 * and creations will have been performed and the dentry returned in @path will
2781 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2782 * specified then a negative dentry may be returned.
2784 * An error code is returned otherwise.
2786 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2787 * cleared otherwise prior to returning.
2789 static int lookup_open(struct nameidata *nd, struct path *path,
2791 const struct open_flags *op,
2792 bool got_write, int *opened)
2794 struct dentry *dir = nd->path.dentry;
2795 struct inode *dir_inode = dir->d_inode;
2796 struct dentry *dentry;
2800 *opened &= ~FILE_CREATED;
2801 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2803 return PTR_ERR(dentry);
2805 /* Cached positive dentry: will open in f_op->open */
2806 if (!need_lookup && dentry->d_inode)
2809 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2810 return atomic_open(nd, dentry, path, file, op, got_write,
2811 need_lookup, opened);
2815 BUG_ON(dentry->d_inode);
2817 dentry = lookup_real(dir_inode, dentry, nd->flags);
2819 return PTR_ERR(dentry);
2822 /* Negative dentry, just create the file */
2823 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2824 umode_t mode = op->mode;
2825 if (!IS_POSIXACL(dir->d_inode))
2826 mode &= ~current_umask();
2828 * This write is needed to ensure that a
2829 * rw->ro transition does not occur between
2830 * the time when the file is created and when
2831 * a permanent write count is taken through
2832 * the 'struct file' in finish_open().
2838 *opened |= FILE_CREATED;
2839 error = security_path_mknod(&nd->path, dentry, mode, 0);
2842 error = vfs_create(dir->d_inode, dentry, mode,
2843 nd->flags & LOOKUP_EXCL);
2848 path->dentry = dentry;
2849 path->mnt = nd->path.mnt;
2858 * Handle the last step of open()
2860 static int do_last(struct nameidata *nd, struct path *path,
2861 struct file *file, const struct open_flags *op,
2862 int *opened, struct filename *name)
2864 struct dentry *dir = nd->path.dentry;
2865 int open_flag = op->open_flag;
2866 bool will_truncate = (open_flag & O_TRUNC) != 0;
2867 bool got_write = false;
2868 int acc_mode = op->acc_mode;
2869 struct inode *inode;
2870 bool symlink_ok = false;
2871 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2872 bool retried = false;
2875 nd->flags &= ~LOOKUP_PARENT;
2876 nd->flags |= op->intent;
2878 if (nd->last_type != LAST_NORM) {
2879 error = handle_dots(nd, nd->last_type);
2885 if (!(open_flag & O_CREAT)) {
2886 if (nd->last.name[nd->last.len])
2887 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2888 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2890 /* we _can_ be in RCU mode here */
2891 error = lookup_fast(nd, path, &inode);
2898 BUG_ON(nd->inode != dir->d_inode);
2900 /* create side of things */
2902 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2903 * has been cleared when we got to the last component we are
2906 error = complete_walk(nd);
2910 audit_inode(name, dir, LOOKUP_PARENT);
2912 /* trailing slashes? */
2913 if (nd->last.name[nd->last.len])
2918 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2919 error = mnt_want_write(nd->path.mnt);
2923 * do _not_ fail yet - we might not need that or fail with
2924 * a different error; let lookup_open() decide; we'll be
2925 * dropping this one anyway.
2928 mutex_lock(&dir->d_inode->i_mutex);
2929 error = lookup_open(nd, path, file, op, got_write, opened);
2930 mutex_unlock(&dir->d_inode->i_mutex);
2936 if ((*opened & FILE_CREATED) ||
2937 !S_ISREG(file_inode(file)->i_mode))
2938 will_truncate = false;
2940 audit_inode(name, file->f_path.dentry, 0);
2944 if (*opened & FILE_CREATED) {
2945 /* Don't check for write permission, don't truncate */
2946 open_flag &= ~O_TRUNC;
2947 will_truncate = false;
2948 acc_mode = MAY_OPEN;
2949 path_to_nameidata(path, nd);
2950 goto finish_open_created;
2954 * create/update audit record if it already exists.
2956 if (d_is_positive(path->dentry))
2957 audit_inode(name, path->dentry, 0);
2960 * If atomic_open() acquired write access it is dropped now due to
2961 * possible mount and symlink following (this might be optimized away if
2965 mnt_drop_write(nd->path.mnt);
2970 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
2973 error = follow_managed(path, nd->flags);
2978 nd->flags |= LOOKUP_JUMPED;
2980 BUG_ON(nd->flags & LOOKUP_RCU);
2981 inode = path->dentry->d_inode;
2983 /* we _can_ be in RCU mode here */
2985 if (d_is_negative(path->dentry)) {
2986 path_to_nameidata(path, nd);
2990 if (should_follow_link(path->dentry, !symlink_ok)) {
2991 if (nd->flags & LOOKUP_RCU) {
2992 if (unlikely(unlazy_walk(nd, path->dentry))) {
2997 BUG_ON(inode != path->dentry->d_inode);
3001 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3002 path_to_nameidata(path, nd);
3004 save_parent.dentry = nd->path.dentry;
3005 save_parent.mnt = mntget(path->mnt);
3006 nd->path.dentry = path->dentry;
3010 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3012 error = complete_walk(nd);
3014 path_put(&save_parent);
3017 audit_inode(name, nd->path.dentry, 0);
3019 if ((open_flag & O_CREAT) &&
3020 (d_is_directory(nd->path.dentry) || d_is_autodir(nd->path.dentry)))
3023 if ((nd->flags & LOOKUP_DIRECTORY) && !d_is_directory(nd->path.dentry))
3025 if (!S_ISREG(nd->inode->i_mode))
3026 will_truncate = false;
3028 if (will_truncate) {
3029 error = mnt_want_write(nd->path.mnt);
3034 finish_open_created:
3035 error = may_open(&nd->path, acc_mode, open_flag);
3038 file->f_path.mnt = nd->path.mnt;
3039 error = finish_open(file, nd->path.dentry, NULL, opened);
3041 if (error == -EOPENSTALE)
3046 error = open_check_o_direct(file);
3049 error = ima_file_check(file, op->acc_mode);
3053 if (will_truncate) {
3054 error = handle_truncate(file);
3060 mnt_drop_write(nd->path.mnt);
3061 path_put(&save_parent);
3066 path_put_conditional(path, nd);
3073 /* If no saved parent or already retried then can't retry */
3074 if (!save_parent.dentry || retried)
3077 BUG_ON(save_parent.dentry != dir);
3078 path_put(&nd->path);
3079 nd->path = save_parent;
3080 nd->inode = dir->d_inode;
3081 save_parent.mnt = NULL;
3082 save_parent.dentry = NULL;
3084 mnt_drop_write(nd->path.mnt);
3091 static int do_tmpfile(int dfd, struct filename *pathname,
3092 struct nameidata *nd, int flags,
3093 const struct open_flags *op,
3094 struct file *file, int *opened)
3096 static const struct qstr name = QSTR_INIT("/", 1);
3097 struct dentry *dentry, *child;
3099 int error = path_lookupat(dfd, pathname->name,
3100 flags | LOOKUP_DIRECTORY, nd);
3101 if (unlikely(error))
3103 error = mnt_want_write(nd->path.mnt);
3104 if (unlikely(error))
3106 /* we want directory to be writable */
3107 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3110 dentry = nd->path.dentry;
3111 dir = dentry->d_inode;
3112 if (!dir->i_op->tmpfile) {
3113 error = -EOPNOTSUPP;
3116 child = d_alloc(dentry, &name);
3117 if (unlikely(!child)) {
3121 nd->flags &= ~LOOKUP_DIRECTORY;
3122 nd->flags |= op->intent;
3123 dput(nd->path.dentry);
3124 nd->path.dentry = child;
3125 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3128 audit_inode(pathname, nd->path.dentry, 0);
3129 error = may_open(&nd->path, op->acc_mode, op->open_flag);
3132 file->f_path.mnt = nd->path.mnt;
3133 error = finish_open(file, nd->path.dentry, NULL, opened);
3136 error = open_check_o_direct(file);
3139 } else if (!(op->open_flag & O_EXCL)) {
3140 struct inode *inode = file_inode(file);
3141 spin_lock(&inode->i_lock);
3142 inode->i_state |= I_LINKABLE;
3143 spin_unlock(&inode->i_lock);
3146 mnt_drop_write(nd->path.mnt);
3148 path_put(&nd->path);
3152 static struct file *path_openat(int dfd, struct filename *pathname,
3153 struct nameidata *nd, const struct open_flags *op, int flags)
3155 struct file *base = NULL;
3161 file = get_empty_filp();
3165 file->f_flags = op->open_flag;
3167 if (unlikely(file->f_flags & __O_TMPFILE)) {
3168 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3172 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
3173 if (unlikely(error))
3176 current->total_link_count = 0;
3177 error = link_path_walk(pathname->name, nd);
3178 if (unlikely(error))
3181 error = do_last(nd, &path, file, op, &opened, pathname);
3182 while (unlikely(error > 0)) { /* trailing symlink */
3183 struct path link = path;
3185 if (!(nd->flags & LOOKUP_FOLLOW)) {
3186 path_put_conditional(&path, nd);
3187 path_put(&nd->path);
3191 error = may_follow_link(&link, nd);
3192 if (unlikely(error))
3194 nd->flags |= LOOKUP_PARENT;
3195 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3196 error = follow_link(&link, nd, &cookie);
3197 if (unlikely(error))
3199 error = do_last(nd, &path, file, op, &opened, pathname);
3200 put_link(nd, &link, cookie);
3203 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3204 path_put(&nd->root);
3207 if (!(opened & FILE_OPENED)) {
3211 if (unlikely(error)) {
3212 if (error == -EOPENSTALE) {
3213 if (flags & LOOKUP_RCU)
3218 file = ERR_PTR(error);
3223 struct file *do_filp_open(int dfd, struct filename *pathname,
3224 const struct open_flags *op)
3226 struct nameidata nd;
3227 int flags = op->lookup_flags;
3230 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3231 if (unlikely(filp == ERR_PTR(-ECHILD)))
3232 filp = path_openat(dfd, pathname, &nd, op, flags);
3233 if (unlikely(filp == ERR_PTR(-ESTALE)))
3234 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3238 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3239 const char *name, const struct open_flags *op)
3241 struct nameidata nd;
3243 struct filename filename = { .name = name };
3244 int flags = op->lookup_flags | LOOKUP_ROOT;
3247 nd.root.dentry = dentry;
3249 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3250 return ERR_PTR(-ELOOP);
3252 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3253 if (unlikely(file == ERR_PTR(-ECHILD)))
3254 file = path_openat(-1, &filename, &nd, op, flags);
3255 if (unlikely(file == ERR_PTR(-ESTALE)))
3256 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3260 struct dentry *kern_path_create(int dfd, const char *pathname,
3261 struct path *path, unsigned int lookup_flags)
3263 struct dentry *dentry = ERR_PTR(-EEXIST);
3264 struct nameidata nd;
3267 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3270 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3271 * other flags passed in are ignored!
3273 lookup_flags &= LOOKUP_REVAL;
3275 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3277 return ERR_PTR(error);
3280 * Yucky last component or no last component at all?
3281 * (foo/., foo/.., /////)
3283 if (nd.last_type != LAST_NORM)
3285 nd.flags &= ~LOOKUP_PARENT;
3286 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3288 /* don't fail immediately if it's r/o, at least try to report other errors */
3289 err2 = mnt_want_write(nd.path.mnt);
3291 * Do the final lookup.
3293 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3294 dentry = lookup_hash(&nd);
3299 if (d_is_positive(dentry))
3303 * Special case - lookup gave negative, but... we had foo/bar/
3304 * From the vfs_mknod() POV we just have a negative dentry -
3305 * all is fine. Let's be bastards - you had / on the end, you've
3306 * been asking for (non-existent) directory. -ENOENT for you.
3308 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3312 if (unlikely(err2)) {
3320 dentry = ERR_PTR(error);
3322 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3324 mnt_drop_write(nd.path.mnt);
3329 EXPORT_SYMBOL(kern_path_create);
3331 void done_path_create(struct path *path, struct dentry *dentry)
3334 mutex_unlock(&path->dentry->d_inode->i_mutex);
3335 mnt_drop_write(path->mnt);
3338 EXPORT_SYMBOL(done_path_create);
3340 struct dentry *user_path_create(int dfd, const char __user *pathname,
3341 struct path *path, unsigned int lookup_flags)
3343 struct filename *tmp = getname(pathname);
3346 return ERR_CAST(tmp);
3347 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3351 EXPORT_SYMBOL(user_path_create);
3353 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3355 int error = may_create(dir, dentry);
3360 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3363 if (!dir->i_op->mknod)
3366 error = devcgroup_inode_mknod(mode, dev);
3370 error = security_inode_mknod(dir, dentry, mode, dev);
3374 error = dir->i_op->mknod(dir, dentry, mode, dev);
3376 fsnotify_create(dir, dentry);
3380 static int may_mknod(umode_t mode)
3382 switch (mode & S_IFMT) {
3388 case 0: /* zero mode translates to S_IFREG */
3397 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3400 struct dentry *dentry;
3403 unsigned int lookup_flags = 0;
3405 error = may_mknod(mode);
3409 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3411 return PTR_ERR(dentry);
3413 if (!IS_POSIXACL(path.dentry->d_inode))
3414 mode &= ~current_umask();
3415 error = security_path_mknod(&path, dentry, mode, dev);
3418 switch (mode & S_IFMT) {
3419 case 0: case S_IFREG:
3420 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3422 case S_IFCHR: case S_IFBLK:
3423 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3424 new_decode_dev(dev));
3426 case S_IFIFO: case S_IFSOCK:
3427 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3431 done_path_create(&path, dentry);
3432 if (retry_estale(error, lookup_flags)) {
3433 lookup_flags |= LOOKUP_REVAL;
3439 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3441 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3444 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3446 int error = may_create(dir, dentry);
3447 unsigned max_links = dir->i_sb->s_max_links;
3452 if (!dir->i_op->mkdir)
3455 mode &= (S_IRWXUGO|S_ISVTX);
3456 error = security_inode_mkdir(dir, dentry, mode);
3460 if (max_links && dir->i_nlink >= max_links)
3463 error = dir->i_op->mkdir(dir, dentry, mode);
3465 fsnotify_mkdir(dir, dentry);
3469 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3471 struct dentry *dentry;
3474 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3477 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3479 return PTR_ERR(dentry);
3481 if (!IS_POSIXACL(path.dentry->d_inode))
3482 mode &= ~current_umask();
3483 error = security_path_mkdir(&path, dentry, mode);
3485 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3486 done_path_create(&path, dentry);
3487 if (retry_estale(error, lookup_flags)) {
3488 lookup_flags |= LOOKUP_REVAL;
3494 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3496 return sys_mkdirat(AT_FDCWD, pathname, mode);
3500 * The dentry_unhash() helper will try to drop the dentry early: we
3501 * should have a usage count of 1 if we're the only user of this
3502 * dentry, and if that is true (possibly after pruning the dcache),
3503 * then we drop the dentry now.
3505 * A low-level filesystem can, if it choses, legally
3508 * if (!d_unhashed(dentry))
3511 * if it cannot handle the case of removing a directory
3512 * that is still in use by something else..
3514 void dentry_unhash(struct dentry *dentry)
3516 shrink_dcache_parent(dentry);
3517 spin_lock(&dentry->d_lock);
3518 if (dentry->d_lockref.count == 1)
3520 spin_unlock(&dentry->d_lock);
3523 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3525 int error = may_delete(dir, dentry, 1);
3530 if (!dir->i_op->rmdir)
3534 mutex_lock(&dentry->d_inode->i_mutex);
3537 if (d_mountpoint(dentry))
3540 error = security_inode_rmdir(dir, dentry);
3544 shrink_dcache_parent(dentry);
3545 error = dir->i_op->rmdir(dir, dentry);
3549 dentry->d_inode->i_flags |= S_DEAD;
3553 mutex_unlock(&dentry->d_inode->i_mutex);
3560 static long do_rmdir(int dfd, const char __user *pathname)
3563 struct filename *name;
3564 struct dentry *dentry;
3565 struct nameidata nd;
3566 unsigned int lookup_flags = 0;
3568 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3570 return PTR_ERR(name);
3572 switch(nd.last_type) {
3584 nd.flags &= ~LOOKUP_PARENT;
3585 error = mnt_want_write(nd.path.mnt);
3589 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3590 dentry = lookup_hash(&nd);
3591 error = PTR_ERR(dentry);
3594 if (!dentry->d_inode) {
3598 error = security_path_rmdir(&nd.path, dentry);
3601 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3605 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3606 mnt_drop_write(nd.path.mnt);
3610 if (retry_estale(error, lookup_flags)) {
3611 lookup_flags |= LOOKUP_REVAL;
3617 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3619 return do_rmdir(AT_FDCWD, pathname);
3623 * vfs_unlink - unlink a filesystem object
3624 * @dir: parent directory
3626 * @delegated_inode: returns victim inode, if the inode is delegated.
3628 * The caller must hold dir->i_mutex.
3630 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3631 * return a reference to the inode in delegated_inode. The caller
3632 * should then break the delegation on that inode and retry. Because
3633 * breaking a delegation may take a long time, the caller should drop
3634 * dir->i_mutex before doing so.
3636 * Alternatively, a caller may pass NULL for delegated_inode. This may
3637 * be appropriate for callers that expect the underlying filesystem not
3638 * to be NFS exported.
3640 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3642 struct inode *target = dentry->d_inode;
3643 int error = may_delete(dir, dentry, 0);
3648 if (!dir->i_op->unlink)
3651 mutex_lock(&target->i_mutex);
3652 if (d_mountpoint(dentry))
3655 error = security_inode_unlink(dir, dentry);
3657 error = try_break_deleg(target, delegated_inode);
3660 error = dir->i_op->unlink(dir, dentry);
3666 mutex_unlock(&target->i_mutex);
3668 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3669 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3670 fsnotify_link_count(target);
3678 * Make sure that the actual truncation of the file will occur outside its
3679 * directory's i_mutex. Truncate can take a long time if there is a lot of
3680 * writeout happening, and we don't want to prevent access to the directory
3681 * while waiting on the I/O.
3683 static long do_unlinkat(int dfd, const char __user *pathname)
3686 struct filename *name;
3687 struct dentry *dentry;
3688 struct nameidata nd;
3689 struct inode *inode = NULL;
3690 struct inode *delegated_inode = NULL;
3691 unsigned int lookup_flags = 0;
3693 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3695 return PTR_ERR(name);
3698 if (nd.last_type != LAST_NORM)
3701 nd.flags &= ~LOOKUP_PARENT;
3702 error = mnt_want_write(nd.path.mnt);
3706 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3707 dentry = lookup_hash(&nd);
3708 error = PTR_ERR(dentry);
3709 if (!IS_ERR(dentry)) {
3710 /* Why not before? Because we want correct error value */
3711 if (nd.last.name[nd.last.len])
3713 inode = dentry->d_inode;
3714 if (d_is_negative(dentry))
3717 error = security_path_unlink(&nd.path, dentry);
3720 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3724 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3726 iput(inode); /* truncate the inode here */
3728 if (delegated_inode) {
3729 error = break_deleg_wait(&delegated_inode);
3733 mnt_drop_write(nd.path.mnt);
3737 if (retry_estale(error, lookup_flags)) {
3738 lookup_flags |= LOOKUP_REVAL;
3745 if (d_is_negative(dentry))
3747 else if (d_is_directory(dentry) || d_is_autodir(dentry))
3754 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3756 if ((flag & ~AT_REMOVEDIR) != 0)
3759 if (flag & AT_REMOVEDIR)
3760 return do_rmdir(dfd, pathname);
3762 return do_unlinkat(dfd, pathname);
3765 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3767 return do_unlinkat(AT_FDCWD, pathname);
3770 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3772 int error = may_create(dir, dentry);
3777 if (!dir->i_op->symlink)
3780 error = security_inode_symlink(dir, dentry, oldname);
3784 error = dir->i_op->symlink(dir, dentry, oldname);
3786 fsnotify_create(dir, dentry);
3790 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3791 int, newdfd, const char __user *, newname)
3794 struct filename *from;
3795 struct dentry *dentry;
3797 unsigned int lookup_flags = 0;
3799 from = getname(oldname);
3801 return PTR_ERR(from);
3803 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3804 error = PTR_ERR(dentry);
3808 error = security_path_symlink(&path, dentry, from->name);
3810 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3811 done_path_create(&path, dentry);
3812 if (retry_estale(error, lookup_flags)) {
3813 lookup_flags |= LOOKUP_REVAL;
3821 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3823 return sys_symlinkat(oldname, AT_FDCWD, newname);
3827 * vfs_link - create a new link
3828 * @old_dentry: object to be linked
3830 * @new_dentry: where to create the new link
3831 * @delegated_inode: returns inode needing a delegation break
3833 * The caller must hold dir->i_mutex
3835 * If vfs_link discovers a delegation on the to-be-linked file in need
3836 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3837 * inode in delegated_inode. The caller should then break the delegation
3838 * and retry. Because breaking a delegation may take a long time, the
3839 * caller should drop the i_mutex before doing so.
3841 * Alternatively, a caller may pass NULL for delegated_inode. This may
3842 * be appropriate for callers that expect the underlying filesystem not
3843 * to be NFS exported.
3845 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3847 struct inode *inode = old_dentry->d_inode;
3848 unsigned max_links = dir->i_sb->s_max_links;
3854 error = may_create(dir, new_dentry);
3858 if (dir->i_sb != inode->i_sb)
3862 * A link to an append-only or immutable file cannot be created.
3864 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3866 if (!dir->i_op->link)
3868 if (S_ISDIR(inode->i_mode))
3871 error = security_inode_link(old_dentry, dir, new_dentry);
3875 mutex_lock(&inode->i_mutex);
3876 /* Make sure we don't allow creating hardlink to an unlinked file */
3877 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3879 else if (max_links && inode->i_nlink >= max_links)
3882 error = try_break_deleg(inode, delegated_inode);
3884 error = dir->i_op->link(old_dentry, dir, new_dentry);
3887 if (!error && (inode->i_state & I_LINKABLE)) {
3888 spin_lock(&inode->i_lock);
3889 inode->i_state &= ~I_LINKABLE;
3890 spin_unlock(&inode->i_lock);
3892 mutex_unlock(&inode->i_mutex);
3894 fsnotify_link(dir, inode, new_dentry);
3899 * Hardlinks are often used in delicate situations. We avoid
3900 * security-related surprises by not following symlinks on the
3903 * We don't follow them on the oldname either to be compatible
3904 * with linux 2.0, and to avoid hard-linking to directories
3905 * and other special files. --ADM
3907 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3908 int, newdfd, const char __user *, newname, int, flags)
3910 struct dentry *new_dentry;
3911 struct path old_path, new_path;
3912 struct inode *delegated_inode = NULL;
3916 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3919 * To use null names we require CAP_DAC_READ_SEARCH
3920 * This ensures that not everyone will be able to create
3921 * handlink using the passed filedescriptor.
3923 if (flags & AT_EMPTY_PATH) {
3924 if (!capable(CAP_DAC_READ_SEARCH))
3929 if (flags & AT_SYMLINK_FOLLOW)
3930 how |= LOOKUP_FOLLOW;
3932 error = user_path_at(olddfd, oldname, how, &old_path);
3936 new_dentry = user_path_create(newdfd, newname, &new_path,
3937 (how & LOOKUP_REVAL));
3938 error = PTR_ERR(new_dentry);
3939 if (IS_ERR(new_dentry))
3943 if (old_path.mnt != new_path.mnt)
3945 error = may_linkat(&old_path);
3946 if (unlikely(error))
3948 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3951 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
3953 done_path_create(&new_path, new_dentry);
3954 if (delegated_inode) {
3955 error = break_deleg_wait(&delegated_inode);
3957 path_put(&old_path);
3961 if (retry_estale(error, how)) {
3962 path_put(&old_path);
3963 how |= LOOKUP_REVAL;
3967 path_put(&old_path);
3972 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3974 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3978 * The worst of all namespace operations - renaming directory. "Perverted"
3979 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3981 * a) we can get into loop creation. Check is done in is_subdir().
3982 * b) race potential - two innocent renames can create a loop together.
3983 * That's where 4.4 screws up. Current fix: serialization on
3984 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3986 * c) we have to lock _four_ objects - parents and victim (if it exists),
3987 * and source (if it is not a directory).
3988 * And that - after we got ->i_mutex on parents (until then we don't know
3989 * whether the target exists). Solution: try to be smart with locking
3990 * order for inodes. We rely on the fact that tree topology may change
3991 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3992 * move will be locked. Thus we can rank directories by the tree
3993 * (ancestors first) and rank all non-directories after them.
3994 * That works since everybody except rename does "lock parent, lookup,
3995 * lock child" and rename is under ->s_vfs_rename_mutex.
3996 * HOWEVER, it relies on the assumption that any object with ->lookup()
3997 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3998 * we'd better make sure that there's no link(2) for them.
3999 * d) conversion from fhandle to dentry may come in the wrong moment - when
4000 * we are removing the target. Solution: we will have to grab ->i_mutex
4001 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4002 * ->i_mutex on parents, which works but leads to some truly excessive
4005 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
4006 struct inode *new_dir, struct dentry *new_dentry)
4009 struct inode *target = new_dentry->d_inode;
4010 unsigned max_links = new_dir->i_sb->s_max_links;
4013 * If we are going to change the parent - check write permissions,
4014 * we'll need to flip '..'.
4016 if (new_dir != old_dir) {
4017 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
4022 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4028 mutex_lock(&target->i_mutex);
4031 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
4035 if (max_links && !target && new_dir != old_dir &&
4036 new_dir->i_nlink >= max_links)
4040 shrink_dcache_parent(new_dentry);
4041 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4046 target->i_flags |= S_DEAD;
4047 dont_mount(new_dentry);
4051 mutex_unlock(&target->i_mutex);
4054 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4055 d_move(old_dentry,new_dentry);
4059 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
4060 struct inode *new_dir, struct dentry *new_dentry,
4061 struct inode **delegated_inode)
4063 struct inode *target = new_dentry->d_inode;
4064 struct inode *source = old_dentry->d_inode;
4067 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4072 lock_two_nondirectories(source, target);
4075 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
4078 error = try_break_deleg(source, delegated_inode);
4082 error = try_break_deleg(target, delegated_inode);
4086 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4091 dont_mount(new_dentry);
4092 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4093 d_move(old_dentry, new_dentry);
4095 unlock_two_nondirectories(source, target);
4101 * vfs_rename - rename a filesystem object
4102 * @old_dir: parent of source
4103 * @old_dentry: source
4104 * @new_dir: parent of destination
4105 * @new_dentry: destination
4106 * @delegated_inode: returns an inode needing a delegation break
4108 * The caller must hold multiple mutexes--see lock_rename()).
4110 * If vfs_rename discovers a delegation in need of breaking at either
4111 * the source or destination, it will return -EWOULDBLOCK and return a
4112 * reference to the inode in delegated_inode. The caller should then
4113 * break the delegation and retry. Because breaking a delegation may
4114 * take a long time, the caller should drop all locks before doing
4117 * Alternatively, a caller may pass NULL for delegated_inode. This may
4118 * be appropriate for callers that expect the underlying filesystem not
4119 * to be NFS exported.
4121 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4122 struct inode *new_dir, struct dentry *new_dentry,
4123 struct inode **delegated_inode)
4126 int is_dir = d_is_directory(old_dentry) || d_is_autodir(old_dentry);
4127 const unsigned char *old_name;
4129 if (old_dentry->d_inode == new_dentry->d_inode)
4132 error = may_delete(old_dir, old_dentry, is_dir);
4136 if (!new_dentry->d_inode)
4137 error = may_create(new_dir, new_dentry);
4139 error = may_delete(new_dir, new_dentry, is_dir);
4143 if (!old_dir->i_op->rename)
4146 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4149 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
4151 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry,delegated_inode);
4153 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4154 new_dentry->d_inode, old_dentry);
4155 fsnotify_oldname_free(old_name);
4160 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4161 int, newdfd, const char __user *, newname)
4163 struct dentry *old_dir, *new_dir;
4164 struct dentry *old_dentry, *new_dentry;
4165 struct dentry *trap;
4166 struct nameidata oldnd, newnd;
4167 struct inode *delegated_inode = NULL;
4168 struct filename *from;
4169 struct filename *to;
4170 unsigned int lookup_flags = 0;
4171 bool should_retry = false;
4174 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4176 error = PTR_ERR(from);
4180 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4182 error = PTR_ERR(to);
4187 if (oldnd.path.mnt != newnd.path.mnt)
4190 old_dir = oldnd.path.dentry;
4192 if (oldnd.last_type != LAST_NORM)
4195 new_dir = newnd.path.dentry;
4196 if (newnd.last_type != LAST_NORM)
4199 error = mnt_want_write(oldnd.path.mnt);
4203 oldnd.flags &= ~LOOKUP_PARENT;
4204 newnd.flags &= ~LOOKUP_PARENT;
4205 newnd.flags |= LOOKUP_RENAME_TARGET;
4208 trap = lock_rename(new_dir, old_dir);
4210 old_dentry = lookup_hash(&oldnd);
4211 error = PTR_ERR(old_dentry);
4212 if (IS_ERR(old_dentry))
4214 /* source must exist */
4216 if (d_is_negative(old_dentry))
4218 /* unless the source is a directory trailing slashes give -ENOTDIR */
4219 if (!d_is_directory(old_dentry) && !d_is_autodir(old_dentry)) {
4221 if (oldnd.last.name[oldnd.last.len])
4223 if (newnd.last.name[newnd.last.len])
4226 /* source should not be ancestor of target */
4228 if (old_dentry == trap)
4230 new_dentry = lookup_hash(&newnd);
4231 error = PTR_ERR(new_dentry);
4232 if (IS_ERR(new_dentry))
4234 /* target should not be an ancestor of source */
4236 if (new_dentry == trap)
4239 error = security_path_rename(&oldnd.path, old_dentry,
4240 &newnd.path, new_dentry);
4243 error = vfs_rename(old_dir->d_inode, old_dentry,
4244 new_dir->d_inode, new_dentry,
4251 unlock_rename(new_dir, old_dir);
4252 if (delegated_inode) {
4253 error = break_deleg_wait(&delegated_inode);
4257 mnt_drop_write(oldnd.path.mnt);
4259 if (retry_estale(error, lookup_flags))
4260 should_retry = true;
4261 path_put(&newnd.path);
4264 path_put(&oldnd.path);
4267 should_retry = false;
4268 lookup_flags |= LOOKUP_REVAL;
4275 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4277 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
4280 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
4284 len = PTR_ERR(link);
4289 if (len > (unsigned) buflen)
4291 if (copy_to_user(buffer, link, len))
4298 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4299 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4300 * using) it for any given inode is up to filesystem.
4302 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4304 struct nameidata nd;
4309 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4311 return PTR_ERR(cookie);
4313 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
4314 if (dentry->d_inode->i_op->put_link)
4315 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4319 /* get the link contents into pagecache */
4320 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4324 struct address_space *mapping = dentry->d_inode->i_mapping;
4325 page = read_mapping_page(mapping, 0, NULL);
4330 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4334 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4336 struct page *page = NULL;
4337 char *s = page_getlink(dentry, &page);
4338 int res = vfs_readlink(dentry,buffer,buflen,s);
4341 page_cache_release(page);
4346 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4348 struct page *page = NULL;
4349 nd_set_link(nd, page_getlink(dentry, &page));
4353 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4355 struct page *page = cookie;
4359 page_cache_release(page);
4364 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4366 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4368 struct address_space *mapping = inode->i_mapping;
4373 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4375 flags |= AOP_FLAG_NOFS;
4378 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4379 flags, &page, &fsdata);
4383 kaddr = kmap_atomic(page);
4384 memcpy(kaddr, symname, len-1);
4385 kunmap_atomic(kaddr);
4387 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4394 mark_inode_dirty(inode);
4400 int page_symlink(struct inode *inode, const char *symname, int len)
4402 return __page_symlink(inode, symname, len,
4403 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4406 const struct inode_operations page_symlink_inode_operations = {
4407 .readlink = generic_readlink,
4408 .follow_link = page_follow_link_light,
4409 .put_link = page_put_link,
4412 EXPORT_SYMBOL(user_path_at);
4413 EXPORT_SYMBOL(follow_down_one);
4414 EXPORT_SYMBOL(follow_down);
4415 EXPORT_SYMBOL(follow_up);
4416 EXPORT_SYMBOL(get_write_access); /* nfsd */
4417 EXPORT_SYMBOL(lock_rename);
4418 EXPORT_SYMBOL(lookup_one_len);
4419 EXPORT_SYMBOL(page_follow_link_light);
4420 EXPORT_SYMBOL(page_put_link);
4421 EXPORT_SYMBOL(page_readlink);
4422 EXPORT_SYMBOL(__page_symlink);
4423 EXPORT_SYMBOL(page_symlink);
4424 EXPORT_SYMBOL(page_symlink_inode_operations);
4425 EXPORT_SYMBOL(kern_path);
4426 EXPORT_SYMBOL(vfs_path_lookup);
4427 EXPORT_SYMBOL(inode_permission);
4428 EXPORT_SYMBOL(unlock_rename);
4429 EXPORT_SYMBOL(vfs_create);
4430 EXPORT_SYMBOL(vfs_link);
4431 EXPORT_SYMBOL(vfs_mkdir);
4432 EXPORT_SYMBOL(vfs_mknod);
4433 EXPORT_SYMBOL(generic_permission);
4434 EXPORT_SYMBOL(vfs_readlink);
4435 EXPORT_SYMBOL(vfs_rename);
4436 EXPORT_SYMBOL(vfs_rmdir);
4437 EXPORT_SYMBOL(vfs_symlink);
4438 EXPORT_SYMBOL(vfs_unlink);
4439 EXPORT_SYMBOL(dentry_unhash);
4440 EXPORT_SYMBOL(generic_readlink);