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 <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125 getname_flags(const char __user *filename, int flags, int *empty)
127 struct filename *result;
131 result = audit_reusename(filename);
135 result = __getname();
136 if (unlikely(!result))
137 return ERR_PTR(-ENOMEM);
140 * First, try to embed the struct filename inside the names_cache
143 kname = (char *)result->iname;
144 result->name = kname;
146 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
147 if (unlikely(len < 0)) {
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
158 if (unlikely(len == EMBEDDED_NAME_MAX)) {
159 const size_t size = offsetof(struct filename, iname[1]);
160 kname = (char *)result;
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
167 result = kzalloc(size, GFP_KERNEL);
168 if (unlikely(!result)) {
170 return ERR_PTR(-ENOMEM);
172 result->name = kname;
173 len = strncpy_from_user(kname, filename, PATH_MAX);
174 if (unlikely(len < 0)) {
179 if (unlikely(len == PATH_MAX)) {
182 return ERR_PTR(-ENAMETOOLONG);
187 /* The empty path is special. */
188 if (unlikely(!len)) {
191 if (!(flags & LOOKUP_EMPTY)) {
193 return ERR_PTR(-ENOENT);
197 result->uptr = filename;
198 result->aname = NULL;
199 audit_getname(result);
204 getname(const char __user * filename)
206 return getname_flags(filename, 0, NULL);
210 getname_kernel(const char * filename)
212 struct filename *result;
213 int len = strlen(filename) + 1;
215 result = __getname();
216 if (unlikely(!result))
217 return ERR_PTR(-ENOMEM);
219 if (len <= EMBEDDED_NAME_MAX) {
220 result->name = (char *)result->iname;
221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
227 return ERR_PTR(-ENOMEM);
229 tmp->name = (char *)result;
233 return ERR_PTR(-ENAMETOOLONG);
235 memcpy((char *)result->name, filename, len);
237 result->aname = NULL;
239 audit_getname(result);
244 void putname(struct filename *name)
246 BUG_ON(name->refcnt <= 0);
248 if (--name->refcnt > 0)
251 if (name->name != name->iname) {
252 __putname(name->name);
258 static int check_acl(struct inode *inode, int mask)
260 #ifdef CONFIG_FS_POSIX_ACL
261 struct posix_acl *acl;
263 if (mask & MAY_NOT_BLOCK) {
264 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 /* no ->get_acl() calls in RCU mode... */
268 if (acl == ACL_NOT_CACHED)
270 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273 acl = get_acl(inode, ACL_TYPE_ACCESS);
277 int error = posix_acl_permission(inode, acl, mask);
278 posix_acl_release(acl);
287 * This does the basic permission checking
289 static int acl_permission_check(struct inode *inode, int mask)
291 unsigned int mode = inode->i_mode;
293 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
297 int error = check_acl(inode, mask);
298 if (error != -EAGAIN)
302 if (in_group_p(inode->i_gid))
307 * If the DACs are ok we don't need any capability check.
309 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
315 * generic_permission - check for access rights on a Posix-like filesystem
316 * @inode: inode to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
322 * are used for other things.
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
328 int generic_permission(struct inode *inode, int mask)
333 * Do the basic permission checks.
335 ret = acl_permission_check(inode, mask);
339 if (S_ISDIR(inode->i_mode)) {
340 /* DACs are overridable for directories */
341 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
343 if (!(mask & MAY_WRITE))
344 if (capable_wrt_inode_uidgid(inode,
345 CAP_DAC_READ_SEARCH))
350 * Read/write DACs are always overridable.
351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
354 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
355 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
359 * Searching includes executable on directories, else just read.
361 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
362 if (mask == MAY_READ)
363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
368 EXPORT_SYMBOL(generic_permission);
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
376 static inline int do_inode_permission(struct inode *inode, int mask)
378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379 if (likely(inode->i_op->permission))
380 return inode->i_op->permission(inode, mask);
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode->i_lock);
384 inode->i_opflags |= IOP_FASTPERM;
385 spin_unlock(&inode->i_lock);
387 return generic_permission(inode, mask);
391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395 * Check for read/write/execute permissions on an inode.
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
402 int __inode_permission(struct inode *inode, int mask)
406 if (unlikely(mask & MAY_WRITE)) {
408 * Nobody gets write access to an immutable file.
410 if (IS_IMMUTABLE(inode))
414 retval = do_inode_permission(inode, mask);
418 retval = devcgroup_inode_permission(inode, mask);
422 return security_inode_permission(inode, mask);
424 EXPORT_SYMBOL(__inode_permission);
427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Separate out file-system wide checks from inode-specific permission checks.
434 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
436 if (unlikely(mask & MAY_WRITE)) {
437 umode_t mode = inode->i_mode;
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb->s_flags & MS_RDONLY) &&
441 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 int inode_permission(struct inode *inode, int mask)
462 retval = sb_permission(inode->i_sb, inode, mask);
465 return __inode_permission(inode, mask);
467 EXPORT_SYMBOL(inode_permission);
470 * path_get - get a reference to a path
471 * @path: path to get the reference to
473 * Given a path increment the reference count to the dentry and the vfsmount.
475 void path_get(const struct path *path)
480 EXPORT_SYMBOL(path_get);
483 * path_put - put a reference to a path
484 * @path: path to put the reference to
486 * Given a path decrement the reference count to the dentry and the vfsmount.
488 void path_put(const struct path *path)
493 EXPORT_SYMBOL(path_put);
495 #define EMBEDDED_LEVELS 2
503 struct inode *inode; /* path.dentry.d_inode */
508 int total_link_count;
514 } *stack, internal[EMBEDDED_LEVELS];
517 static struct nameidata *set_nameidata(struct nameidata *p)
519 struct nameidata *old = current->nameidata;
520 p->stack = p->internal;
521 p->total_link_count = old ? old->total_link_count : 0;
522 current->nameidata = p;
526 static void restore_nameidata(struct nameidata *old)
528 struct nameidata *now = current->nameidata;
530 current->nameidata = old;
532 old->total_link_count = now->total_link_count;
533 if (now->stack != now->internal) {
535 now->stack = now->internal;
539 static int __nd_alloc_stack(struct nameidata *nd)
541 struct saved *p = kmalloc(MAXSYMLINKS * sizeof(struct saved),
545 memcpy(p, nd->internal, sizeof(nd->internal));
550 static inline int nd_alloc_stack(struct nameidata *nd)
552 if (likely(nd->depth != EMBEDDED_LEVELS))
554 if (likely(nd->stack != nd->internal))
556 return __nd_alloc_stack(nd);
560 * Path walking has 2 modes, rcu-walk and ref-walk (see
561 * Documentation/filesystems/path-lookup.txt). In situations when we can't
562 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
563 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
564 * mode. Refcounts are grabbed at the last known good point before rcu-walk
565 * got stuck, so ref-walk may continue from there. If this is not successful
566 * (eg. a seqcount has changed), then failure is returned and it's up to caller
567 * to restart the path walk from the beginning in ref-walk mode.
571 * unlazy_walk - try to switch to ref-walk mode.
572 * @nd: nameidata pathwalk data
573 * @dentry: child of nd->path.dentry or NULL
574 * Returns: 0 on success, -ECHILD on failure
576 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
577 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
578 * @nd or NULL. Must be called from rcu-walk context.
580 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
582 struct fs_struct *fs = current->fs;
583 struct dentry *parent = nd->path.dentry;
585 BUG_ON(!(nd->flags & LOOKUP_RCU));
588 * After legitimizing the bastards, terminate_walk()
589 * will do the right thing for non-RCU mode, and all our
590 * subsequent exit cases should rcu_read_unlock()
591 * before returning. Do vfsmount first; if dentry
592 * can't be legitimized, just set nd->path.dentry to NULL
593 * and rely on dput(NULL) being a no-op.
595 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
597 nd->flags &= ~LOOKUP_RCU;
599 if (!lockref_get_not_dead(&parent->d_lockref)) {
600 nd->path.dentry = NULL;
605 * For a negative lookup, the lookup sequence point is the parents
606 * sequence point, and it only needs to revalidate the parent dentry.
608 * For a positive lookup, we need to move both the parent and the
609 * dentry from the RCU domain to be properly refcounted. And the
610 * sequence number in the dentry validates *both* dentry counters,
611 * since we checked the sequence number of the parent after we got
612 * the child sequence number. So we know the parent must still
613 * be valid if the child sequence number is still valid.
616 if (read_seqcount_retry(&parent->d_seq, nd->seq))
618 BUG_ON(nd->inode != parent->d_inode);
620 if (!lockref_get_not_dead(&dentry->d_lockref))
622 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
627 * Sequence counts matched. Now make sure that the root is
628 * still valid and get it if required.
630 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
631 spin_lock(&fs->lock);
632 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
633 goto unlock_and_drop_dentry;
635 spin_unlock(&fs->lock);
641 unlock_and_drop_dentry:
642 spin_unlock(&fs->lock);
650 if (!(nd->flags & LOOKUP_ROOT))
655 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
657 return dentry->d_op->d_revalidate(dentry, flags);
661 * complete_walk - successful completion of path walk
662 * @nd: pointer nameidata
664 * If we had been in RCU mode, drop out of it and legitimize nd->path.
665 * Revalidate the final result, unless we'd already done that during
666 * the path walk or the filesystem doesn't ask for it. Return 0 on
667 * success, -error on failure. In case of failure caller does not
668 * need to drop nd->path.
670 static int complete_walk(struct nameidata *nd)
672 struct dentry *dentry = nd->path.dentry;
675 if (nd->flags & LOOKUP_RCU) {
676 nd->flags &= ~LOOKUP_RCU;
677 if (!(nd->flags & LOOKUP_ROOT))
680 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
684 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
686 mntput(nd->path.mnt);
689 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
692 mntput(nd->path.mnt);
698 if (likely(!(nd->flags & LOOKUP_JUMPED)))
701 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
704 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
715 static __always_inline void set_root(struct nameidata *nd)
717 get_fs_root(current->fs, &nd->root);
720 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
722 struct fs_struct *fs = current->fs;
726 seq = read_seqcount_begin(&fs->seq);
728 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
729 } while (read_seqcount_retry(&fs->seq, seq));
733 static void path_put_conditional(struct path *path, struct nameidata *nd)
736 if (path->mnt != nd->path.mnt)
740 static inline void path_to_nameidata(const struct path *path,
741 struct nameidata *nd)
743 if (!(nd->flags & LOOKUP_RCU)) {
744 dput(nd->path.dentry);
745 if (nd->path.mnt != path->mnt)
746 mntput(nd->path.mnt);
748 nd->path.mnt = path->mnt;
749 nd->path.dentry = path->dentry;
753 * Helper to directly jump to a known parsed path from ->follow_link,
754 * caller must have taken a reference to path beforehand.
756 void nd_jump_link(struct nameidata *nd, struct path *path)
761 nd->inode = nd->path.dentry->d_inode;
762 nd->flags |= LOOKUP_JUMPED;
765 static inline void put_link(struct nameidata *nd)
767 struct saved *last = nd->stack + --nd->depth;
768 struct inode *inode = last->link.dentry->d_inode;
769 if (last->cookie && inode->i_op->put_link)
770 inode->i_op->put_link(last->link.dentry, last->cookie);
771 path_put(&last->link);
774 int sysctl_protected_symlinks __read_mostly = 0;
775 int sysctl_protected_hardlinks __read_mostly = 0;
778 * may_follow_link - Check symlink following for unsafe situations
779 * @link: The path of the symlink
780 * @nd: nameidata pathwalk data
782 * In the case of the sysctl_protected_symlinks sysctl being enabled,
783 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
784 * in a sticky world-writable directory. This is to protect privileged
785 * processes from failing races against path names that may change out
786 * from under them by way of other users creating malicious symlinks.
787 * It will permit symlinks to be followed only when outside a sticky
788 * world-writable directory, or when the uid of the symlink and follower
789 * match, or when the directory owner matches the symlink's owner.
791 * Returns 0 if following the symlink is allowed, -ve on error.
793 static inline int may_follow_link(struct path *link, struct nameidata *nd)
795 const struct inode *inode;
796 const struct inode *parent;
798 if (!sysctl_protected_symlinks)
801 /* Allowed if owner and follower match. */
802 inode = link->dentry->d_inode;
803 if (uid_eq(current_cred()->fsuid, inode->i_uid))
806 /* Allowed if parent directory not sticky and world-writable. */
807 parent = nd->path.dentry->d_inode;
808 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
811 /* Allowed if parent directory and link owner match. */
812 if (uid_eq(parent->i_uid, inode->i_uid))
815 audit_log_link_denied("follow_link", link);
816 path_put_conditional(link, nd);
822 * safe_hardlink_source - Check for safe hardlink conditions
823 * @inode: the source inode to hardlink from
825 * Return false if at least one of the following conditions:
826 * - inode is not a regular file
828 * - inode is setgid and group-exec
829 * - access failure for read and write
831 * Otherwise returns true.
833 static bool safe_hardlink_source(struct inode *inode)
835 umode_t mode = inode->i_mode;
837 /* Special files should not get pinned to the filesystem. */
841 /* Setuid files should not get pinned to the filesystem. */
845 /* Executable setgid files should not get pinned to the filesystem. */
846 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
849 /* Hardlinking to unreadable or unwritable sources is dangerous. */
850 if (inode_permission(inode, MAY_READ | MAY_WRITE))
857 * may_linkat - Check permissions for creating a hardlink
858 * @link: the source to hardlink from
860 * Block hardlink when all of:
861 * - sysctl_protected_hardlinks enabled
862 * - fsuid does not match inode
863 * - hardlink source is unsafe (see safe_hardlink_source() above)
866 * Returns 0 if successful, -ve on error.
868 static int may_linkat(struct path *link)
870 const struct cred *cred;
873 if (!sysctl_protected_hardlinks)
876 cred = current_cred();
877 inode = link->dentry->d_inode;
879 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
880 * otherwise, it must be a safe source.
882 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
886 audit_log_link_denied("linkat", link);
890 static __always_inline
891 const char *get_link(struct nameidata *nd)
893 struct saved *last = nd->stack + nd->depth;
894 struct dentry *dentry = nd->link.dentry;
895 struct inode *inode = dentry->d_inode;
899 BUG_ON(nd->flags & LOOKUP_RCU);
901 if (nd->link.mnt == nd->path.mnt)
902 mntget(nd->link.mnt);
904 last->link = nd->link;
909 touch_atime(&last->link);
911 error = security_inode_follow_link(dentry);
912 res = ERR_PTR(error);
916 nd->last_type = LAST_BIND;
919 res = inode->i_op->follow_link(dentry, &last->cookie, nd);
922 path_put(&last->link);
930 static int follow_up_rcu(struct path *path)
932 struct mount *mnt = real_mount(path->mnt);
933 struct mount *parent;
934 struct dentry *mountpoint;
936 parent = mnt->mnt_parent;
937 if (&parent->mnt == path->mnt)
939 mountpoint = mnt->mnt_mountpoint;
940 path->dentry = mountpoint;
941 path->mnt = &parent->mnt;
946 * follow_up - Find the mountpoint of path's vfsmount
948 * Given a path, find the mountpoint of its source file system.
949 * Replace @path with the path of the mountpoint in the parent mount.
952 * Return 1 if we went up a level and 0 if we were already at the
955 int follow_up(struct path *path)
957 struct mount *mnt = real_mount(path->mnt);
958 struct mount *parent;
959 struct dentry *mountpoint;
961 read_seqlock_excl(&mount_lock);
962 parent = mnt->mnt_parent;
964 read_sequnlock_excl(&mount_lock);
967 mntget(&parent->mnt);
968 mountpoint = dget(mnt->mnt_mountpoint);
969 read_sequnlock_excl(&mount_lock);
971 path->dentry = mountpoint;
973 path->mnt = &parent->mnt;
976 EXPORT_SYMBOL(follow_up);
979 * Perform an automount
980 * - return -EISDIR to tell follow_managed() to stop and return the path we
983 static int follow_automount(struct path *path, struct nameidata *nd,
986 struct vfsmount *mnt;
989 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
992 /* We don't want to mount if someone's just doing a stat -
993 * unless they're stat'ing a directory and appended a '/' to
996 * We do, however, want to mount if someone wants to open or
997 * create a file of any type under the mountpoint, wants to
998 * traverse through the mountpoint or wants to open the
999 * mounted directory. Also, autofs may mark negative dentries
1000 * as being automount points. These will need the attentions
1001 * of the daemon to instantiate them before they can be used.
1003 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1004 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1005 path->dentry->d_inode)
1008 nd->total_link_count++;
1009 if (nd->total_link_count >= 40)
1012 mnt = path->dentry->d_op->d_automount(path);
1015 * The filesystem is allowed to return -EISDIR here to indicate
1016 * it doesn't want to automount. For instance, autofs would do
1017 * this so that its userspace daemon can mount on this dentry.
1019 * However, we can only permit this if it's a terminal point in
1020 * the path being looked up; if it wasn't then the remainder of
1021 * the path is inaccessible and we should say so.
1023 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1025 return PTR_ERR(mnt);
1028 if (!mnt) /* mount collision */
1031 if (!*need_mntput) {
1032 /* lock_mount() may release path->mnt on error */
1034 *need_mntput = true;
1036 err = finish_automount(mnt, path);
1040 /* Someone else made a mount here whilst we were busy */
1045 path->dentry = dget(mnt->mnt_root);
1054 * Handle a dentry that is managed in some way.
1055 * - Flagged for transit management (autofs)
1056 * - Flagged as mountpoint
1057 * - Flagged as automount point
1059 * This may only be called in refwalk mode.
1061 * Serialization is taken care of in namespace.c
1063 static int follow_managed(struct path *path, struct nameidata *nd)
1065 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1067 bool need_mntput = false;
1070 /* Given that we're not holding a lock here, we retain the value in a
1071 * local variable for each dentry as we look at it so that we don't see
1072 * the components of that value change under us */
1073 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1074 managed &= DCACHE_MANAGED_DENTRY,
1075 unlikely(managed != 0)) {
1076 /* Allow the filesystem to manage the transit without i_mutex
1078 if (managed & DCACHE_MANAGE_TRANSIT) {
1079 BUG_ON(!path->dentry->d_op);
1080 BUG_ON(!path->dentry->d_op->d_manage);
1081 ret = path->dentry->d_op->d_manage(path->dentry, false);
1086 /* Transit to a mounted filesystem. */
1087 if (managed & DCACHE_MOUNTED) {
1088 struct vfsmount *mounted = lookup_mnt(path);
1093 path->mnt = mounted;
1094 path->dentry = dget(mounted->mnt_root);
1099 /* Something is mounted on this dentry in another
1100 * namespace and/or whatever was mounted there in this
1101 * namespace got unmounted before lookup_mnt() could
1105 /* Handle an automount point */
1106 if (managed & DCACHE_NEED_AUTOMOUNT) {
1107 ret = follow_automount(path, nd, &need_mntput);
1113 /* We didn't change the current path point */
1117 if (need_mntput && path->mnt == mnt)
1122 nd->flags |= LOOKUP_JUMPED;
1123 if (unlikely(ret < 0))
1124 path_put_conditional(path, nd);
1128 int follow_down_one(struct path *path)
1130 struct vfsmount *mounted;
1132 mounted = lookup_mnt(path);
1136 path->mnt = mounted;
1137 path->dentry = dget(mounted->mnt_root);
1142 EXPORT_SYMBOL(follow_down_one);
1144 static inline int managed_dentry_rcu(struct dentry *dentry)
1146 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1147 dentry->d_op->d_manage(dentry, true) : 0;
1151 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1152 * we meet a managed dentry that would need blocking.
1154 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1155 struct inode **inode)
1158 struct mount *mounted;
1160 * Don't forget we might have a non-mountpoint managed dentry
1161 * that wants to block transit.
1163 switch (managed_dentry_rcu(path->dentry)) {
1173 if (!d_mountpoint(path->dentry))
1174 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1176 mounted = __lookup_mnt(path->mnt, path->dentry);
1179 path->mnt = &mounted->mnt;
1180 path->dentry = mounted->mnt.mnt_root;
1181 nd->flags |= LOOKUP_JUMPED;
1182 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1184 * Update the inode too. We don't need to re-check the
1185 * dentry sequence number here after this d_inode read,
1186 * because a mount-point is always pinned.
1188 *inode = path->dentry->d_inode;
1190 return !read_seqretry(&mount_lock, nd->m_seq) &&
1191 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1194 static int follow_dotdot_rcu(struct nameidata *nd)
1196 struct inode *inode = nd->inode;
1201 if (nd->path.dentry == nd->root.dentry &&
1202 nd->path.mnt == nd->root.mnt) {
1205 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1206 struct dentry *old = nd->path.dentry;
1207 struct dentry *parent = old->d_parent;
1210 inode = parent->d_inode;
1211 seq = read_seqcount_begin(&parent->d_seq);
1212 if (read_seqcount_retry(&old->d_seq, nd->seq))
1214 nd->path.dentry = parent;
1218 if (!follow_up_rcu(&nd->path))
1220 inode = nd->path.dentry->d_inode;
1221 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1223 while (d_mountpoint(nd->path.dentry)) {
1224 struct mount *mounted;
1225 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1228 nd->path.mnt = &mounted->mnt;
1229 nd->path.dentry = mounted->mnt.mnt_root;
1230 inode = nd->path.dentry->d_inode;
1231 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1232 if (read_seqretry(&mount_lock, nd->m_seq))
1243 * Follow down to the covering mount currently visible to userspace. At each
1244 * point, the filesystem owning that dentry may be queried as to whether the
1245 * caller is permitted to proceed or not.
1247 int follow_down(struct path *path)
1252 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1253 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1254 /* Allow the filesystem to manage the transit without i_mutex
1257 * We indicate to the filesystem if someone is trying to mount
1258 * something here. This gives autofs the chance to deny anyone
1259 * other than its daemon the right to mount on its
1262 * The filesystem may sleep at this point.
1264 if (managed & DCACHE_MANAGE_TRANSIT) {
1265 BUG_ON(!path->dentry->d_op);
1266 BUG_ON(!path->dentry->d_op->d_manage);
1267 ret = path->dentry->d_op->d_manage(
1268 path->dentry, false);
1270 return ret == -EISDIR ? 0 : ret;
1273 /* Transit to a mounted filesystem. */
1274 if (managed & DCACHE_MOUNTED) {
1275 struct vfsmount *mounted = lookup_mnt(path);
1280 path->mnt = mounted;
1281 path->dentry = dget(mounted->mnt_root);
1285 /* Don't handle automount points here */
1290 EXPORT_SYMBOL(follow_down);
1293 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1295 static void follow_mount(struct path *path)
1297 while (d_mountpoint(path->dentry)) {
1298 struct vfsmount *mounted = lookup_mnt(path);
1303 path->mnt = mounted;
1304 path->dentry = dget(mounted->mnt_root);
1308 static void follow_dotdot(struct nameidata *nd)
1314 struct dentry *old = nd->path.dentry;
1316 if (nd->path.dentry == nd->root.dentry &&
1317 nd->path.mnt == nd->root.mnt) {
1320 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1321 /* rare case of legitimate dget_parent()... */
1322 nd->path.dentry = dget_parent(nd->path.dentry);
1326 if (!follow_up(&nd->path))
1329 follow_mount(&nd->path);
1330 nd->inode = nd->path.dentry->d_inode;
1334 * This looks up the name in dcache, possibly revalidates the old dentry and
1335 * allocates a new one if not found or not valid. In the need_lookup argument
1336 * returns whether i_op->lookup is necessary.
1338 * dir->d_inode->i_mutex must be held
1340 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1341 unsigned int flags, bool *need_lookup)
1343 struct dentry *dentry;
1346 *need_lookup = false;
1347 dentry = d_lookup(dir, name);
1349 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1350 error = d_revalidate(dentry, flags);
1351 if (unlikely(error <= 0)) {
1354 return ERR_PTR(error);
1356 d_invalidate(dentry);
1365 dentry = d_alloc(dir, name);
1366 if (unlikely(!dentry))
1367 return ERR_PTR(-ENOMEM);
1369 *need_lookup = true;
1375 * Call i_op->lookup on the dentry. The dentry must be negative and
1378 * dir->d_inode->i_mutex must be held
1380 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1385 /* Don't create child dentry for a dead directory. */
1386 if (unlikely(IS_DEADDIR(dir))) {
1388 return ERR_PTR(-ENOENT);
1391 old = dir->i_op->lookup(dir, dentry, flags);
1392 if (unlikely(old)) {
1399 static struct dentry *__lookup_hash(struct qstr *name,
1400 struct dentry *base, unsigned int flags)
1403 struct dentry *dentry;
1405 dentry = lookup_dcache(name, base, flags, &need_lookup);
1409 return lookup_real(base->d_inode, dentry, flags);
1413 * It's more convoluted than I'd like it to be, but... it's still fairly
1414 * small and for now I'd prefer to have fast path as straight as possible.
1415 * It _is_ time-critical.
1417 static int lookup_fast(struct nameidata *nd,
1418 struct path *path, struct inode **inode)
1420 struct vfsmount *mnt = nd->path.mnt;
1421 struct dentry *dentry, *parent = nd->path.dentry;
1427 * Rename seqlock is not required here because in the off chance
1428 * of a false negative due to a concurrent rename, we're going to
1429 * do the non-racy lookup, below.
1431 if (nd->flags & LOOKUP_RCU) {
1434 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1439 * This sequence count validates that the inode matches
1440 * the dentry name information from lookup.
1442 *inode = dentry->d_inode;
1443 negative = d_is_negative(dentry);
1444 if (read_seqcount_retry(&dentry->d_seq, seq))
1450 * This sequence count validates that the parent had no
1451 * changes while we did the lookup of the dentry above.
1453 * The memory barrier in read_seqcount_begin of child is
1454 * enough, we can use __read_seqcount_retry here.
1456 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1460 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1461 status = d_revalidate(dentry, nd->flags);
1462 if (unlikely(status <= 0)) {
1463 if (status != -ECHILD)
1469 path->dentry = dentry;
1470 if (likely(__follow_mount_rcu(nd, path, inode)))
1473 if (unlazy_walk(nd, dentry))
1476 dentry = __d_lookup(parent, &nd->last);
1479 if (unlikely(!dentry))
1482 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1483 status = d_revalidate(dentry, nd->flags);
1484 if (unlikely(status <= 0)) {
1489 d_invalidate(dentry);
1494 if (unlikely(d_is_negative(dentry))) {
1499 path->dentry = dentry;
1500 err = follow_managed(path, nd);
1502 *inode = path->dentry->d_inode;
1509 /* Fast lookup failed, do it the slow way */
1510 static int lookup_slow(struct nameidata *nd, struct path *path)
1512 struct dentry *dentry, *parent;
1514 parent = nd->path.dentry;
1515 BUG_ON(nd->inode != parent->d_inode);
1517 mutex_lock(&parent->d_inode->i_mutex);
1518 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1519 mutex_unlock(&parent->d_inode->i_mutex);
1521 return PTR_ERR(dentry);
1522 path->mnt = nd->path.mnt;
1523 path->dentry = dentry;
1524 return follow_managed(path, nd);
1527 static inline int may_lookup(struct nameidata *nd)
1529 if (nd->flags & LOOKUP_RCU) {
1530 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1533 if (unlazy_walk(nd, NULL))
1536 return inode_permission(nd->inode, MAY_EXEC);
1539 static inline int handle_dots(struct nameidata *nd, int type)
1541 if (type == LAST_DOTDOT) {
1542 if (nd->flags & LOOKUP_RCU) {
1543 return follow_dotdot_rcu(nd);
1550 static void terminate_walk(struct nameidata *nd)
1552 if (!(nd->flags & LOOKUP_RCU)) {
1553 path_put(&nd->path);
1555 nd->flags &= ~LOOKUP_RCU;
1556 if (!(nd->flags & LOOKUP_ROOT))
1557 nd->root.mnt = NULL;
1560 while (unlikely(nd->depth))
1564 static int pick_link(struct nameidata *nd, struct path *link)
1567 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1568 path_to_nameidata(link, nd);
1571 if (nd->flags & LOOKUP_RCU) {
1572 if (unlikely(nd->path.mnt != link->mnt ||
1573 unlazy_walk(nd, link->dentry))) {
1577 error = nd_alloc_stack(nd);
1578 if (unlikely(error)) {
1579 path_to_nameidata(link, nd);
1588 * Do we need to follow links? We _really_ want to be able
1589 * to do this check without having to look at inode->i_op,
1590 * so we keep a cache of "no, this doesn't need follow_link"
1591 * for the common case.
1593 static inline int should_follow_link(struct nameidata *nd, struct path *link, int follow)
1595 if (likely(!d_is_symlink(link->dentry)))
1599 return pick_link(nd, link);
1602 enum {WALK_GET = 1, WALK_PUT = 2};
1604 static int walk_component(struct nameidata *nd, int flags)
1607 struct inode *inode;
1610 * "." and ".." are special - ".." especially so because it has
1611 * to be able to know about the current root directory and
1612 * parent relationships.
1614 if (unlikely(nd->last_type != LAST_NORM)) {
1615 err = handle_dots(nd, nd->last_type);
1616 if (flags & WALK_PUT)
1620 err = lookup_fast(nd, &path, &inode);
1621 if (unlikely(err)) {
1625 err = lookup_slow(nd, &path);
1629 inode = path.dentry->d_inode;
1631 if (d_is_negative(path.dentry))
1635 if (flags & WALK_PUT)
1637 err = should_follow_link(nd, &path, flags & WALK_GET);
1640 path_to_nameidata(&path, nd);
1645 path_to_nameidata(&path, nd);
1650 * We can do the critical dentry name comparison and hashing
1651 * operations one word at a time, but we are limited to:
1653 * - Architectures with fast unaligned word accesses. We could
1654 * do a "get_unaligned()" if this helps and is sufficiently
1657 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1658 * do not trap on the (extremely unlikely) case of a page
1659 * crossing operation.
1661 * - Furthermore, we need an efficient 64-bit compile for the
1662 * 64-bit case in order to generate the "number of bytes in
1663 * the final mask". Again, that could be replaced with a
1664 * efficient population count instruction or similar.
1666 #ifdef CONFIG_DCACHE_WORD_ACCESS
1668 #include <asm/word-at-a-time.h>
1672 static inline unsigned int fold_hash(unsigned long hash)
1674 return hash_64(hash, 32);
1677 #else /* 32-bit case */
1679 #define fold_hash(x) (x)
1683 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1685 unsigned long a, mask;
1686 unsigned long hash = 0;
1689 a = load_unaligned_zeropad(name);
1690 if (len < sizeof(unsigned long))
1694 name += sizeof(unsigned long);
1695 len -= sizeof(unsigned long);
1699 mask = bytemask_from_count(len);
1702 return fold_hash(hash);
1704 EXPORT_SYMBOL(full_name_hash);
1707 * Calculate the length and hash of the path component, and
1708 * return the "hash_len" as the result.
1710 static inline u64 hash_name(const char *name)
1712 unsigned long a, b, adata, bdata, mask, hash, len;
1713 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1716 len = -sizeof(unsigned long);
1718 hash = (hash + a) * 9;
1719 len += sizeof(unsigned long);
1720 a = load_unaligned_zeropad(name+len);
1721 b = a ^ REPEAT_BYTE('/');
1722 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1724 adata = prep_zero_mask(a, adata, &constants);
1725 bdata = prep_zero_mask(b, bdata, &constants);
1727 mask = create_zero_mask(adata | bdata);
1729 hash += a & zero_bytemask(mask);
1730 len += find_zero(mask);
1731 return hashlen_create(fold_hash(hash), len);
1736 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1738 unsigned long hash = init_name_hash();
1740 hash = partial_name_hash(*name++, hash);
1741 return end_name_hash(hash);
1743 EXPORT_SYMBOL(full_name_hash);
1746 * We know there's a real path component here of at least
1749 static inline u64 hash_name(const char *name)
1751 unsigned long hash = init_name_hash();
1752 unsigned long len = 0, c;
1754 c = (unsigned char)*name;
1757 hash = partial_name_hash(c, hash);
1758 c = (unsigned char)name[len];
1759 } while (c && c != '/');
1760 return hashlen_create(end_name_hash(hash), len);
1767 * This is the basic name resolution function, turning a pathname into
1768 * the final dentry. We expect 'base' to be positive and a directory.
1770 * Returns 0 and nd will have valid dentry and mnt on success.
1771 * Returns error and drops reference to input namei data on failure.
1773 static int link_path_walk(const char *name, struct nameidata *nd)
1782 /* At this point we know we have a real path component. */
1787 err = may_lookup(nd);
1791 hash_len = hash_name(name);
1794 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1796 if (name[1] == '.') {
1798 nd->flags |= LOOKUP_JUMPED;
1804 if (likely(type == LAST_NORM)) {
1805 struct dentry *parent = nd->path.dentry;
1806 nd->flags &= ~LOOKUP_JUMPED;
1807 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1808 struct qstr this = { { .hash_len = hash_len }, .name = name };
1809 err = parent->d_op->d_hash(parent, &this);
1812 hash_len = this.hash_len;
1817 nd->last.hash_len = hash_len;
1818 nd->last.name = name;
1819 nd->last_type = type;
1821 name += hashlen_len(hash_len);
1825 * If it wasn't NUL, we know it was '/'. Skip that
1826 * slash, and continue until no more slashes.
1830 } while (unlikely(*name == '/'));
1831 if (unlikely(!*name)) {
1833 /* called from path_init(), done */
1836 name = nd->stack[nd->depth - 1].name;
1837 /* called from trailing_symlink(), done */
1840 /* last component of nested symlink */
1841 err = walk_component(nd, WALK_GET | WALK_PUT);
1843 err = walk_component(nd, WALK_GET);
1849 const char *s = get_link(nd);
1851 if (unlikely(IS_ERR(s))) {
1863 path_put(&nd->path);
1864 nd->path = nd->root;
1865 path_get(&nd->root);
1866 nd->flags |= LOOKUP_JUMPED;
1867 while (unlikely(*++s == '/'))
1870 nd->inode = nd->path.dentry->d_inode;
1871 nd->stack[nd->depth - 1].name = name;
1878 if (!d_can_lookup(nd->path.dentry)) {
1887 static int path_init(int dfd, const struct filename *name, unsigned int flags,
1888 struct nameidata *nd)
1891 const char *s = name->name;
1893 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1894 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1897 if (flags & LOOKUP_ROOT) {
1898 struct dentry *root = nd->root.dentry;
1899 struct inode *inode = root->d_inode;
1901 if (!d_can_lookup(root))
1903 retval = inode_permission(inode, MAY_EXEC);
1907 nd->path = nd->root;
1909 if (flags & LOOKUP_RCU) {
1911 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1912 nd->m_seq = read_seqbegin(&mount_lock);
1914 path_get(&nd->path);
1919 nd->root.mnt = NULL;
1921 nd->m_seq = read_seqbegin(&mount_lock);
1923 if (flags & LOOKUP_RCU) {
1925 nd->seq = set_root_rcu(nd);
1928 path_get(&nd->root);
1930 nd->path = nd->root;
1931 } else if (dfd == AT_FDCWD) {
1932 if (flags & LOOKUP_RCU) {
1933 struct fs_struct *fs = current->fs;
1939 seq = read_seqcount_begin(&fs->seq);
1941 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1942 } while (read_seqcount_retry(&fs->seq, seq));
1944 get_fs_pwd(current->fs, &nd->path);
1947 /* Caller must check execute permissions on the starting path component */
1948 struct fd f = fdget_raw(dfd);
1949 struct dentry *dentry;
1954 dentry = f.file->f_path.dentry;
1957 if (!d_can_lookup(dentry)) {
1963 nd->path = f.file->f_path;
1964 if (flags & LOOKUP_RCU) {
1965 if (f.flags & FDPUT_FPUT)
1967 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1970 path_get(&nd->path);
1975 nd->inode = nd->path.dentry->d_inode;
1976 if (!(flags & LOOKUP_RCU))
1978 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1980 if (!(nd->flags & LOOKUP_ROOT))
1981 nd->root.mnt = NULL;
1985 nd->total_link_count = 0;
1986 return link_path_walk(s, nd);
1989 static void path_cleanup(struct nameidata *nd)
1991 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1992 path_put(&nd->root);
1993 nd->root.mnt = NULL;
1995 if (unlikely(nd->base))
1999 static int trailing_symlink(struct nameidata *nd)
2002 int error = may_follow_link(&nd->link, nd);
2003 if (unlikely(error))
2005 nd->flags |= LOOKUP_PARENT;
2007 if (unlikely(IS_ERR(s))) {
2016 path_put(&nd->path);
2017 nd->path = nd->root;
2018 path_get(&nd->root);
2019 nd->flags |= LOOKUP_JUMPED;
2021 nd->inode = nd->path.dentry->d_inode;
2022 nd->stack[0].name = NULL;
2023 return link_path_walk(s, nd);
2026 static inline int lookup_last(struct nameidata *nd)
2029 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2030 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2032 nd->flags &= ~LOOKUP_PARENT;
2033 err = walk_component(nd,
2034 nd->flags & LOOKUP_FOLLOW
2036 ? WALK_PUT | WALK_GET
2044 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2045 static int path_lookupat(int dfd, const struct filename *name,
2046 unsigned int flags, struct nameidata *nd)
2051 * Path walking is largely split up into 2 different synchronisation
2052 * schemes, rcu-walk and ref-walk (explained in
2053 * Documentation/filesystems/path-lookup.txt). These share much of the
2054 * path walk code, but some things particularly setup, cleanup, and
2055 * following mounts are sufficiently divergent that functions are
2056 * duplicated. Typically there is a function foo(), and its RCU
2057 * analogue, foo_rcu().
2059 * -ECHILD is the error number of choice (just to avoid clashes) that
2060 * is returned if some aspect of an rcu-walk fails. Such an error must
2061 * be handled by restarting a traditional ref-walk (which will always
2062 * be able to complete).
2064 err = path_init(dfd, name, flags, nd);
2065 if (!err && !(flags & LOOKUP_PARENT)) {
2066 while ((err = lookup_last(nd)) > 0) {
2067 err = trailing_symlink(nd);
2074 err = complete_walk(nd);
2076 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2077 if (!d_can_lookup(nd->path.dentry)) {
2078 path_put(&nd->path);
2087 static int filename_lookup(int dfd, struct filename *name,
2088 unsigned int flags, struct nameidata *nd)
2091 struct nameidata *saved_nd = set_nameidata(nd);
2093 retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2094 if (unlikely(retval == -ECHILD))
2095 retval = path_lookupat(dfd, name, flags, nd);
2096 if (unlikely(retval == -ESTALE))
2097 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2099 if (likely(!retval))
2100 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2101 restore_nameidata(saved_nd);
2105 /* does lookup, returns the object with parent locked */
2106 struct dentry *kern_path_locked(const char *name, struct path *path)
2108 struct filename *filename = getname_kernel(name);
2109 struct nameidata nd;
2113 if (IS_ERR(filename))
2114 return ERR_CAST(filename);
2116 err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd);
2121 if (nd.last_type != LAST_NORM) {
2123 d = ERR_PTR(-EINVAL);
2126 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2127 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2129 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2139 int kern_path(const char *name, unsigned int flags, struct path *path)
2141 struct nameidata nd;
2142 struct filename *filename = getname_kernel(name);
2143 int res = PTR_ERR(filename);
2145 if (!IS_ERR(filename)) {
2146 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2153 EXPORT_SYMBOL(kern_path);
2156 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2157 * @dentry: pointer to dentry of the base directory
2158 * @mnt: pointer to vfs mount of the base directory
2159 * @name: pointer to file name
2160 * @flags: lookup flags
2161 * @path: pointer to struct path to fill
2163 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2164 const char *name, unsigned int flags,
2167 struct filename *filename = getname_kernel(name);
2168 int err = PTR_ERR(filename);
2170 BUG_ON(flags & LOOKUP_PARENT);
2172 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2173 if (!IS_ERR(filename)) {
2174 struct nameidata nd;
2175 nd.root.dentry = dentry;
2177 err = filename_lookup(AT_FDCWD, filename,
2178 flags | LOOKUP_ROOT, &nd);
2185 EXPORT_SYMBOL(vfs_path_lookup);
2188 * lookup_one_len - filesystem helper to lookup single pathname component
2189 * @name: pathname component to lookup
2190 * @base: base directory to lookup from
2191 * @len: maximum length @len should be interpreted to
2193 * Note that this routine is purely a helper for filesystem usage and should
2194 * not be called by generic code.
2196 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2202 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2206 this.hash = full_name_hash(name, len);
2208 return ERR_PTR(-EACCES);
2210 if (unlikely(name[0] == '.')) {
2211 if (len < 2 || (len == 2 && name[1] == '.'))
2212 return ERR_PTR(-EACCES);
2216 c = *(const unsigned char *)name++;
2217 if (c == '/' || c == '\0')
2218 return ERR_PTR(-EACCES);
2221 * See if the low-level filesystem might want
2222 * to use its own hash..
2224 if (base->d_flags & DCACHE_OP_HASH) {
2225 int err = base->d_op->d_hash(base, &this);
2227 return ERR_PTR(err);
2230 err = inode_permission(base->d_inode, MAY_EXEC);
2232 return ERR_PTR(err);
2234 return __lookup_hash(&this, base, 0);
2236 EXPORT_SYMBOL(lookup_one_len);
2238 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2239 struct path *path, int *empty)
2241 struct nameidata nd;
2242 struct filename *tmp = getname_flags(name, flags, empty);
2243 int err = PTR_ERR(tmp);
2246 BUG_ON(flags & LOOKUP_PARENT);
2248 err = filename_lookup(dfd, tmp, flags, &nd);
2256 int user_path_at(int dfd, const char __user *name, unsigned flags,
2259 return user_path_at_empty(dfd, name, flags, path, NULL);
2261 EXPORT_SYMBOL(user_path_at);
2264 * NB: most callers don't do anything directly with the reference to the
2265 * to struct filename, but the nd->last pointer points into the name string
2266 * allocated by getname. So we must hold the reference to it until all
2267 * path-walking is complete.
2269 static struct filename *
2270 user_path_parent(int dfd, const char __user *path,
2271 struct path *parent,
2276 struct nameidata nd;
2277 struct filename *s = getname(path);
2280 /* only LOOKUP_REVAL is allowed in extra flags */
2281 flags &= LOOKUP_REVAL;
2286 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, &nd);
2289 return ERR_PTR(error);
2293 *type = nd.last_type;
2299 * mountpoint_last - look up last component for umount
2300 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2301 * @path: pointer to container for result
2303 * This is a special lookup_last function just for umount. In this case, we
2304 * need to resolve the path without doing any revalidation.
2306 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2307 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2308 * in almost all cases, this lookup will be served out of the dcache. The only
2309 * cases where it won't are if nd->last refers to a symlink or the path is
2310 * bogus and it doesn't exist.
2313 * -error: if there was an error during lookup. This includes -ENOENT if the
2314 * lookup found a negative dentry. The nd->path reference will also be
2317 * 0: if we successfully resolved nd->path and found it to not to be a
2318 * symlink that needs to be followed. "path" will also be populated.
2319 * The nd->path reference will also be put.
2321 * 1: if we successfully resolved nd->last and found it to be a symlink
2322 * that needs to be followed. "path" will be populated with the path
2323 * to the link, and nd->path will *not* be put.
2326 mountpoint_last(struct nameidata *nd, struct path *path)
2329 struct dentry *dentry;
2330 struct dentry *dir = nd->path.dentry;
2332 /* If we're in rcuwalk, drop out of it to handle last component */
2333 if (nd->flags & LOOKUP_RCU) {
2334 if (unlazy_walk(nd, NULL)) {
2340 nd->flags &= ~LOOKUP_PARENT;
2342 if (unlikely(nd->last_type != LAST_NORM)) {
2343 error = handle_dots(nd, nd->last_type);
2346 dentry = dget(nd->path.dentry);
2350 mutex_lock(&dir->d_inode->i_mutex);
2351 dentry = d_lookup(dir, &nd->last);
2354 * No cached dentry. Mounted dentries are pinned in the cache,
2355 * so that means that this dentry is probably a symlink or the
2356 * path doesn't actually point to a mounted dentry.
2358 dentry = d_alloc(dir, &nd->last);
2361 mutex_unlock(&dir->d_inode->i_mutex);
2364 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2365 error = PTR_ERR(dentry);
2366 if (IS_ERR(dentry)) {
2367 mutex_unlock(&dir->d_inode->i_mutex);
2371 mutex_unlock(&dir->d_inode->i_mutex);
2374 if (d_is_negative(dentry)) {
2381 path->dentry = dentry;
2382 path->mnt = nd->path.mnt;
2383 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW);
2384 if (unlikely(error)) {
2398 * path_mountpoint - look up a path to be umounted
2399 * @dfd: directory file descriptor to start walk from
2400 * @name: full pathname to walk
2401 * @path: pointer to container for result
2402 * @flags: lookup flags
2404 * Look up the given name, but don't attempt to revalidate the last component.
2405 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2408 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2409 struct nameidata *nd, unsigned int flags)
2411 int err = path_init(dfd, name, flags, nd);
2415 while ((err = mountpoint_last(nd, path)) > 0) {
2416 err = trailing_symlink(nd);
2426 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2429 struct nameidata nd, *saved;
2432 return PTR_ERR(name);
2433 saved = set_nameidata(&nd);
2434 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2435 if (unlikely(error == -ECHILD))
2436 error = path_mountpoint(dfd, name, path, &nd, flags);
2437 if (unlikely(error == -ESTALE))
2438 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2440 audit_inode(name, path->dentry, 0);
2441 restore_nameidata(saved);
2447 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2448 * @dfd: directory file descriptor
2449 * @name: pathname from userland
2450 * @flags: lookup flags
2451 * @path: pointer to container to hold result
2453 * A umount is a special case for path walking. We're not actually interested
2454 * in the inode in this situation, and ESTALE errors can be a problem. We
2455 * simply want track down the dentry and vfsmount attached at the mountpoint
2456 * and avoid revalidating the last component.
2458 * Returns 0 and populates "path" on success.
2461 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2464 return filename_mountpoint(dfd, getname(name), path, flags);
2468 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2471 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2473 EXPORT_SYMBOL(kern_path_mountpoint);
2475 int __check_sticky(struct inode *dir, struct inode *inode)
2477 kuid_t fsuid = current_fsuid();
2479 if (uid_eq(inode->i_uid, fsuid))
2481 if (uid_eq(dir->i_uid, fsuid))
2483 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2485 EXPORT_SYMBOL(__check_sticky);
2488 * Check whether we can remove a link victim from directory dir, check
2489 * whether the type of victim is right.
2490 * 1. We can't do it if dir is read-only (done in permission())
2491 * 2. We should have write and exec permissions on dir
2492 * 3. We can't remove anything from append-only dir
2493 * 4. We can't do anything with immutable dir (done in permission())
2494 * 5. If the sticky bit on dir is set we should either
2495 * a. be owner of dir, or
2496 * b. be owner of victim, or
2497 * c. have CAP_FOWNER capability
2498 * 6. If the victim is append-only or immutable we can't do antyhing with
2499 * links pointing to it.
2500 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2501 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2502 * 9. We can't remove a root or mountpoint.
2503 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2504 * nfs_async_unlink().
2506 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2508 struct inode *inode = victim->d_inode;
2511 if (d_is_negative(victim))
2515 BUG_ON(victim->d_parent->d_inode != dir);
2516 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2518 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2524 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2525 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2528 if (!d_is_dir(victim))
2530 if (IS_ROOT(victim))
2532 } else if (d_is_dir(victim))
2534 if (IS_DEADDIR(dir))
2536 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2541 /* Check whether we can create an object with dentry child in directory
2543 * 1. We can't do it if child already exists (open has special treatment for
2544 * this case, but since we are inlined it's OK)
2545 * 2. We can't do it if dir is read-only (done in permission())
2546 * 3. We should have write and exec permissions on dir
2547 * 4. We can't do it if dir is immutable (done in permission())
2549 static inline int may_create(struct inode *dir, struct dentry *child)
2551 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2554 if (IS_DEADDIR(dir))
2556 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2560 * p1 and p2 should be directories on the same fs.
2562 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2567 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2571 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2573 p = d_ancestor(p2, p1);
2575 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2576 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2580 p = d_ancestor(p1, p2);
2582 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2583 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2587 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2588 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2591 EXPORT_SYMBOL(lock_rename);
2593 void unlock_rename(struct dentry *p1, struct dentry *p2)
2595 mutex_unlock(&p1->d_inode->i_mutex);
2597 mutex_unlock(&p2->d_inode->i_mutex);
2598 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2601 EXPORT_SYMBOL(unlock_rename);
2603 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2606 int error = may_create(dir, dentry);
2610 if (!dir->i_op->create)
2611 return -EACCES; /* shouldn't it be ENOSYS? */
2614 error = security_inode_create(dir, dentry, mode);
2617 error = dir->i_op->create(dir, dentry, mode, want_excl);
2619 fsnotify_create(dir, dentry);
2622 EXPORT_SYMBOL(vfs_create);
2624 static int may_open(struct path *path, int acc_mode, int flag)
2626 struct dentry *dentry = path->dentry;
2627 struct inode *inode = dentry->d_inode;
2637 switch (inode->i_mode & S_IFMT) {
2641 if (acc_mode & MAY_WRITE)
2646 if (path->mnt->mnt_flags & MNT_NODEV)
2655 error = inode_permission(inode, acc_mode);
2660 * An append-only file must be opened in append mode for writing.
2662 if (IS_APPEND(inode)) {
2663 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2669 /* O_NOATIME can only be set by the owner or superuser */
2670 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2676 static int handle_truncate(struct file *filp)
2678 struct path *path = &filp->f_path;
2679 struct inode *inode = path->dentry->d_inode;
2680 int error = get_write_access(inode);
2684 * Refuse to truncate files with mandatory locks held on them.
2686 error = locks_verify_locked(filp);
2688 error = security_path_truncate(path);
2690 error = do_truncate(path->dentry, 0,
2691 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2694 put_write_access(inode);
2698 static inline int open_to_namei_flags(int flag)
2700 if ((flag & O_ACCMODE) == 3)
2705 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2707 int error = security_path_mknod(dir, dentry, mode, 0);
2711 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2715 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2719 * Attempt to atomically look up, create and open a file from a negative
2722 * Returns 0 if successful. The file will have been created and attached to
2723 * @file by the filesystem calling finish_open().
2725 * Returns 1 if the file was looked up only or didn't need creating. The
2726 * caller will need to perform the open themselves. @path will have been
2727 * updated to point to the new dentry. This may be negative.
2729 * Returns an error code otherwise.
2731 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2732 struct path *path, struct file *file,
2733 const struct open_flags *op,
2734 bool got_write, bool need_lookup,
2737 struct inode *dir = nd->path.dentry->d_inode;
2738 unsigned open_flag = open_to_namei_flags(op->open_flag);
2742 int create_error = 0;
2743 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2746 BUG_ON(dentry->d_inode);
2748 /* Don't create child dentry for a dead directory. */
2749 if (unlikely(IS_DEADDIR(dir))) {
2755 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2756 mode &= ~current_umask();
2758 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2760 open_flag &= ~O_TRUNC;
2763 * Checking write permission is tricky, bacuse we don't know if we are
2764 * going to actually need it: O_CREAT opens should work as long as the
2765 * file exists. But checking existence breaks atomicity. The trick is
2766 * to check access and if not granted clear O_CREAT from the flags.
2768 * Another problem is returing the "right" error value (e.g. for an
2769 * O_EXCL open we want to return EEXIST not EROFS).
2771 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2772 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2773 if (!(open_flag & O_CREAT)) {
2775 * No O_CREATE -> atomicity not a requirement -> fall
2776 * back to lookup + open
2779 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2780 /* Fall back and fail with the right error */
2781 create_error = -EROFS;
2784 /* No side effects, safe to clear O_CREAT */
2785 create_error = -EROFS;
2786 open_flag &= ~O_CREAT;
2790 if (open_flag & O_CREAT) {
2791 error = may_o_create(&nd->path, dentry, mode);
2793 create_error = error;
2794 if (open_flag & O_EXCL)
2796 open_flag &= ~O_CREAT;
2800 if (nd->flags & LOOKUP_DIRECTORY)
2801 open_flag |= O_DIRECTORY;
2803 file->f_path.dentry = DENTRY_NOT_SET;
2804 file->f_path.mnt = nd->path.mnt;
2805 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2808 if (create_error && error == -ENOENT)
2809 error = create_error;
2813 if (error) { /* returned 1, that is */
2814 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2818 if (file->f_path.dentry) {
2820 dentry = file->f_path.dentry;
2822 if (*opened & FILE_CREATED)
2823 fsnotify_create(dir, dentry);
2824 if (!dentry->d_inode) {
2825 WARN_ON(*opened & FILE_CREATED);
2827 error = create_error;
2831 if (excl && !(*opened & FILE_CREATED)) {
2840 * We didn't have the inode before the open, so check open permission
2843 acc_mode = op->acc_mode;
2844 if (*opened & FILE_CREATED) {
2845 WARN_ON(!(open_flag & O_CREAT));
2846 fsnotify_create(dir, dentry);
2847 acc_mode = MAY_OPEN;
2849 error = may_open(&file->f_path, acc_mode, open_flag);
2859 dentry = lookup_real(dir, dentry, nd->flags);
2861 return PTR_ERR(dentry);
2864 int open_flag = op->open_flag;
2866 error = create_error;
2867 if ((open_flag & O_EXCL)) {
2868 if (!dentry->d_inode)
2870 } else if (!dentry->d_inode) {
2872 } else if ((open_flag & O_TRUNC) &&
2876 /* will fail later, go on to get the right error */
2880 path->dentry = dentry;
2881 path->mnt = nd->path.mnt;
2886 * Look up and maybe create and open the last component.
2888 * Must be called with i_mutex held on parent.
2890 * Returns 0 if the file was successfully atomically created (if necessary) and
2891 * opened. In this case the file will be returned attached to @file.
2893 * Returns 1 if the file was not completely opened at this time, though lookups
2894 * and creations will have been performed and the dentry returned in @path will
2895 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2896 * specified then a negative dentry may be returned.
2898 * An error code is returned otherwise.
2900 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2901 * cleared otherwise prior to returning.
2903 static int lookup_open(struct nameidata *nd, struct path *path,
2905 const struct open_flags *op,
2906 bool got_write, int *opened)
2908 struct dentry *dir = nd->path.dentry;
2909 struct inode *dir_inode = dir->d_inode;
2910 struct dentry *dentry;
2914 *opened &= ~FILE_CREATED;
2915 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2917 return PTR_ERR(dentry);
2919 /* Cached positive dentry: will open in f_op->open */
2920 if (!need_lookup && dentry->d_inode)
2923 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2924 return atomic_open(nd, dentry, path, file, op, got_write,
2925 need_lookup, opened);
2929 BUG_ON(dentry->d_inode);
2931 dentry = lookup_real(dir_inode, dentry, nd->flags);
2933 return PTR_ERR(dentry);
2936 /* Negative dentry, just create the file */
2937 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2938 umode_t mode = op->mode;
2939 if (!IS_POSIXACL(dir->d_inode))
2940 mode &= ~current_umask();
2942 * This write is needed to ensure that a
2943 * rw->ro transition does not occur between
2944 * the time when the file is created and when
2945 * a permanent write count is taken through
2946 * the 'struct file' in finish_open().
2952 *opened |= FILE_CREATED;
2953 error = security_path_mknod(&nd->path, dentry, mode, 0);
2956 error = vfs_create(dir->d_inode, dentry, mode,
2957 nd->flags & LOOKUP_EXCL);
2962 path->dentry = dentry;
2963 path->mnt = nd->path.mnt;
2972 * Handle the last step of open()
2974 static int do_last(struct nameidata *nd,
2975 struct file *file, const struct open_flags *op,
2976 int *opened, struct filename *name)
2978 struct dentry *dir = nd->path.dentry;
2979 int open_flag = op->open_flag;
2980 bool will_truncate = (open_flag & O_TRUNC) != 0;
2981 bool got_write = false;
2982 int acc_mode = op->acc_mode;
2983 struct inode *inode;
2984 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2986 bool retried = false;
2989 nd->flags &= ~LOOKUP_PARENT;
2990 nd->flags |= op->intent;
2992 if (nd->last_type != LAST_NORM) {
2993 error = handle_dots(nd, nd->last_type);
2994 if (unlikely(error)) {
3001 if (!(open_flag & O_CREAT)) {
3002 if (nd->last.name[nd->last.len])
3003 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3004 /* we _can_ be in RCU mode here */
3005 error = lookup_fast(nd, &path, &inode);
3012 BUG_ON(nd->inode != dir->d_inode);
3014 /* create side of things */
3016 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3017 * has been cleared when we got to the last component we are
3020 error = complete_walk(nd);
3027 audit_inode(name, dir, LOOKUP_PARENT);
3029 /* trailing slashes? */
3030 if (nd->last.name[nd->last.len])
3035 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3036 error = mnt_want_write(nd->path.mnt);
3040 * do _not_ fail yet - we might not need that or fail with
3041 * a different error; let lookup_open() decide; we'll be
3042 * dropping this one anyway.
3045 mutex_lock(&dir->d_inode->i_mutex);
3046 error = lookup_open(nd, &path, file, op, got_write, opened);
3047 mutex_unlock(&dir->d_inode->i_mutex);
3053 if ((*opened & FILE_CREATED) ||
3054 !S_ISREG(file_inode(file)->i_mode))
3055 will_truncate = false;
3057 audit_inode(name, file->f_path.dentry, 0);
3061 if (*opened & FILE_CREATED) {
3062 /* Don't check for write permission, don't truncate */
3063 open_flag &= ~O_TRUNC;
3064 will_truncate = false;
3065 acc_mode = MAY_OPEN;
3066 path_to_nameidata(&path, nd);
3067 goto finish_open_created;
3071 * create/update audit record if it already exists.
3073 if (d_is_positive(path.dentry))
3074 audit_inode(name, path.dentry, 0);
3077 * If atomic_open() acquired write access it is dropped now due to
3078 * possible mount and symlink following (this might be optimized away if
3082 mnt_drop_write(nd->path.mnt);
3087 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3090 error = follow_managed(&path, nd);
3094 BUG_ON(nd->flags & LOOKUP_RCU);
3095 inode = path.dentry->d_inode;
3097 if (d_is_negative(path.dentry)) {
3098 path_to_nameidata(&path, nd);
3104 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW);
3105 if (unlikely(error)) {
3111 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3112 path_to_nameidata(&path, nd);
3117 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3118 path_to_nameidata(&path, nd);
3120 save_parent.dentry = nd->path.dentry;
3121 save_parent.mnt = mntget(path.mnt);
3122 nd->path.dentry = path.dentry;
3126 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3128 error = complete_walk(nd);
3132 path_put(&save_parent);
3135 audit_inode(name, nd->path.dentry, 0);
3137 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3140 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3142 if (!d_is_reg(nd->path.dentry))
3143 will_truncate = false;
3145 if (will_truncate) {
3146 error = mnt_want_write(nd->path.mnt);
3151 finish_open_created:
3152 error = may_open(&nd->path, acc_mode, open_flag);
3156 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3157 error = vfs_open(&nd->path, file, current_cred());
3159 *opened |= FILE_OPENED;
3161 if (error == -EOPENSTALE)
3166 error = open_check_o_direct(file);
3169 error = ima_file_check(file, op->acc_mode, *opened);
3173 if (will_truncate) {
3174 error = handle_truncate(file);
3180 mnt_drop_write(nd->path.mnt);
3181 path_put(&save_parent);
3186 path_put_conditional(&path, nd);
3193 /* If no saved parent or already retried then can't retry */
3194 if (!save_parent.dentry || retried)
3197 BUG_ON(save_parent.dentry != dir);
3198 path_put(&nd->path);
3199 nd->path = save_parent;
3200 nd->inode = dir->d_inode;
3201 save_parent.mnt = NULL;
3202 save_parent.dentry = NULL;
3204 mnt_drop_write(nd->path.mnt);
3211 static int do_tmpfile(int dfd, struct filename *pathname,
3212 struct nameidata *nd, int flags,
3213 const struct open_flags *op,
3214 struct file *file, int *opened)
3216 static const struct qstr name = QSTR_INIT("/", 1);
3217 struct dentry *dentry, *child;
3219 int error = path_lookupat(dfd, pathname,
3220 flags | LOOKUP_DIRECTORY, nd);
3221 if (unlikely(error))
3223 error = mnt_want_write(nd->path.mnt);
3224 if (unlikely(error))
3226 /* we want directory to be writable */
3227 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3230 dentry = nd->path.dentry;
3231 dir = dentry->d_inode;
3232 if (!dir->i_op->tmpfile) {
3233 error = -EOPNOTSUPP;
3236 child = d_alloc(dentry, &name);
3237 if (unlikely(!child)) {
3241 nd->flags &= ~LOOKUP_DIRECTORY;
3242 nd->flags |= op->intent;
3243 dput(nd->path.dentry);
3244 nd->path.dentry = child;
3245 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3248 audit_inode(pathname, nd->path.dentry, 0);
3249 /* Don't check for other permissions, the inode was just created */
3250 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3253 file->f_path.mnt = nd->path.mnt;
3254 error = finish_open(file, nd->path.dentry, NULL, opened);
3257 error = open_check_o_direct(file);
3260 } else if (!(op->open_flag & O_EXCL)) {
3261 struct inode *inode = file_inode(file);
3262 spin_lock(&inode->i_lock);
3263 inode->i_state |= I_LINKABLE;
3264 spin_unlock(&inode->i_lock);
3267 mnt_drop_write(nd->path.mnt);
3269 path_put(&nd->path);
3273 static struct file *path_openat(int dfd, struct filename *pathname,
3274 struct nameidata *nd, const struct open_flags *op, int flags)
3280 file = get_empty_filp();
3284 file->f_flags = op->open_flag;
3286 if (unlikely(file->f_flags & __O_TMPFILE)) {
3287 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3291 error = path_init(dfd, pathname, flags, nd);
3292 if (unlikely(error))
3295 while ((error = do_last(nd, file, op, &opened, pathname)) > 0) {
3296 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3297 error = trailing_symlink(nd);
3298 if (unlikely(error))
3304 if (!(opened & FILE_OPENED)) {
3308 if (unlikely(error)) {
3309 if (error == -EOPENSTALE) {
3310 if (flags & LOOKUP_RCU)
3315 file = ERR_PTR(error);
3320 struct file *do_filp_open(int dfd, struct filename *pathname,
3321 const struct open_flags *op)
3323 struct nameidata nd, *saved_nd = set_nameidata(&nd);
3324 int flags = op->lookup_flags;
3327 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3328 if (unlikely(filp == ERR_PTR(-ECHILD)))
3329 filp = path_openat(dfd, pathname, &nd, op, flags);
3330 if (unlikely(filp == ERR_PTR(-ESTALE)))
3331 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3332 restore_nameidata(saved_nd);
3336 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3337 const char *name, const struct open_flags *op)
3339 struct nameidata nd, *saved_nd;
3341 struct filename *filename;
3342 int flags = op->lookup_flags | LOOKUP_ROOT;
3345 nd.root.dentry = dentry;
3347 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3348 return ERR_PTR(-ELOOP);
3350 filename = getname_kernel(name);
3351 if (unlikely(IS_ERR(filename)))
3352 return ERR_CAST(filename);
3354 saved_nd = set_nameidata(&nd);
3355 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3356 if (unlikely(file == ERR_PTR(-ECHILD)))
3357 file = path_openat(-1, filename, &nd, op, flags);
3358 if (unlikely(file == ERR_PTR(-ESTALE)))
3359 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3360 restore_nameidata(saved_nd);
3365 static struct dentry *filename_create(int dfd, struct filename *name,
3366 struct path *path, unsigned int lookup_flags)
3368 struct dentry *dentry = ERR_PTR(-EEXIST);
3369 struct nameidata nd;
3372 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3375 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3376 * other flags passed in are ignored!
3378 lookup_flags &= LOOKUP_REVAL;
3380 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3382 return ERR_PTR(error);
3385 * Yucky last component or no last component at all?
3386 * (foo/., foo/.., /////)
3388 if (nd.last_type != LAST_NORM)
3390 nd.flags &= ~LOOKUP_PARENT;
3391 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3393 /* don't fail immediately if it's r/o, at least try to report other errors */
3394 err2 = mnt_want_write(nd.path.mnt);
3396 * Do the final lookup.
3398 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3399 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3404 if (d_is_positive(dentry))
3408 * Special case - lookup gave negative, but... we had foo/bar/
3409 * From the vfs_mknod() POV we just have a negative dentry -
3410 * all is fine. Let's be bastards - you had / on the end, you've
3411 * been asking for (non-existent) directory. -ENOENT for you.
3413 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3417 if (unlikely(err2)) {
3425 dentry = ERR_PTR(error);
3427 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3429 mnt_drop_write(nd.path.mnt);
3435 struct dentry *kern_path_create(int dfd, const char *pathname,
3436 struct path *path, unsigned int lookup_flags)
3438 struct filename *filename = getname_kernel(pathname);
3441 if (IS_ERR(filename))
3442 return ERR_CAST(filename);
3443 res = filename_create(dfd, filename, path, lookup_flags);
3447 EXPORT_SYMBOL(kern_path_create);
3449 void done_path_create(struct path *path, struct dentry *dentry)
3452 mutex_unlock(&path->dentry->d_inode->i_mutex);
3453 mnt_drop_write(path->mnt);
3456 EXPORT_SYMBOL(done_path_create);
3458 struct dentry *user_path_create(int dfd, const char __user *pathname,
3459 struct path *path, unsigned int lookup_flags)
3461 struct filename *tmp = getname(pathname);
3464 return ERR_CAST(tmp);
3465 res = filename_create(dfd, tmp, path, lookup_flags);
3469 EXPORT_SYMBOL(user_path_create);
3471 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3473 int error = may_create(dir, dentry);
3478 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3481 if (!dir->i_op->mknod)
3484 error = devcgroup_inode_mknod(mode, dev);
3488 error = security_inode_mknod(dir, dentry, mode, dev);
3492 error = dir->i_op->mknod(dir, dentry, mode, dev);
3494 fsnotify_create(dir, dentry);
3497 EXPORT_SYMBOL(vfs_mknod);
3499 static int may_mknod(umode_t mode)
3501 switch (mode & S_IFMT) {
3507 case 0: /* zero mode translates to S_IFREG */
3516 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3519 struct dentry *dentry;
3522 unsigned int lookup_flags = 0;
3524 error = may_mknod(mode);
3528 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3530 return PTR_ERR(dentry);
3532 if (!IS_POSIXACL(path.dentry->d_inode))
3533 mode &= ~current_umask();
3534 error = security_path_mknod(&path, dentry, mode, dev);
3537 switch (mode & S_IFMT) {
3538 case 0: case S_IFREG:
3539 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3541 case S_IFCHR: case S_IFBLK:
3542 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3543 new_decode_dev(dev));
3545 case S_IFIFO: case S_IFSOCK:
3546 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3550 done_path_create(&path, dentry);
3551 if (retry_estale(error, lookup_flags)) {
3552 lookup_flags |= LOOKUP_REVAL;
3558 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3560 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3563 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3565 int error = may_create(dir, dentry);
3566 unsigned max_links = dir->i_sb->s_max_links;
3571 if (!dir->i_op->mkdir)
3574 mode &= (S_IRWXUGO|S_ISVTX);
3575 error = security_inode_mkdir(dir, dentry, mode);
3579 if (max_links && dir->i_nlink >= max_links)
3582 error = dir->i_op->mkdir(dir, dentry, mode);
3584 fsnotify_mkdir(dir, dentry);
3587 EXPORT_SYMBOL(vfs_mkdir);
3589 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3591 struct dentry *dentry;
3594 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3597 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3599 return PTR_ERR(dentry);
3601 if (!IS_POSIXACL(path.dentry->d_inode))
3602 mode &= ~current_umask();
3603 error = security_path_mkdir(&path, dentry, mode);
3605 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3606 done_path_create(&path, dentry);
3607 if (retry_estale(error, lookup_flags)) {
3608 lookup_flags |= LOOKUP_REVAL;
3614 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3616 return sys_mkdirat(AT_FDCWD, pathname, mode);
3620 * The dentry_unhash() helper will try to drop the dentry early: we
3621 * should have a usage count of 1 if we're the only user of this
3622 * dentry, and if that is true (possibly after pruning the dcache),
3623 * then we drop the dentry now.
3625 * A low-level filesystem can, if it choses, legally
3628 * if (!d_unhashed(dentry))
3631 * if it cannot handle the case of removing a directory
3632 * that is still in use by something else..
3634 void dentry_unhash(struct dentry *dentry)
3636 shrink_dcache_parent(dentry);
3637 spin_lock(&dentry->d_lock);
3638 if (dentry->d_lockref.count == 1)
3640 spin_unlock(&dentry->d_lock);
3642 EXPORT_SYMBOL(dentry_unhash);
3644 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3646 int error = may_delete(dir, dentry, 1);
3651 if (!dir->i_op->rmdir)
3655 mutex_lock(&dentry->d_inode->i_mutex);
3658 if (is_local_mountpoint(dentry))
3661 error = security_inode_rmdir(dir, dentry);
3665 shrink_dcache_parent(dentry);
3666 error = dir->i_op->rmdir(dir, dentry);
3670 dentry->d_inode->i_flags |= S_DEAD;
3672 detach_mounts(dentry);
3675 mutex_unlock(&dentry->d_inode->i_mutex);
3681 EXPORT_SYMBOL(vfs_rmdir);
3683 static long do_rmdir(int dfd, const char __user *pathname)
3686 struct filename *name;
3687 struct dentry *dentry;
3691 unsigned int lookup_flags = 0;
3693 name = user_path_parent(dfd, pathname,
3694 &path, &last, &type, lookup_flags);
3696 return PTR_ERR(name);
3710 error = mnt_want_write(path.mnt);
3714 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3715 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3716 error = PTR_ERR(dentry);
3719 if (!dentry->d_inode) {
3723 error = security_path_rmdir(&path, dentry);
3726 error = vfs_rmdir(path.dentry->d_inode, dentry);
3730 mutex_unlock(&path.dentry->d_inode->i_mutex);
3731 mnt_drop_write(path.mnt);
3735 if (retry_estale(error, lookup_flags)) {
3736 lookup_flags |= LOOKUP_REVAL;
3742 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3744 return do_rmdir(AT_FDCWD, pathname);
3748 * vfs_unlink - unlink a filesystem object
3749 * @dir: parent directory
3751 * @delegated_inode: returns victim inode, if the inode is delegated.
3753 * The caller must hold dir->i_mutex.
3755 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3756 * return a reference to the inode in delegated_inode. The caller
3757 * should then break the delegation on that inode and retry. Because
3758 * breaking a delegation may take a long time, the caller should drop
3759 * dir->i_mutex before doing so.
3761 * Alternatively, a caller may pass NULL for delegated_inode. This may
3762 * be appropriate for callers that expect the underlying filesystem not
3763 * to be NFS exported.
3765 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3767 struct inode *target = dentry->d_inode;
3768 int error = may_delete(dir, dentry, 0);
3773 if (!dir->i_op->unlink)
3776 mutex_lock(&target->i_mutex);
3777 if (is_local_mountpoint(dentry))
3780 error = security_inode_unlink(dir, dentry);
3782 error = try_break_deleg(target, delegated_inode);
3785 error = dir->i_op->unlink(dir, dentry);
3788 detach_mounts(dentry);
3793 mutex_unlock(&target->i_mutex);
3795 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3796 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3797 fsnotify_link_count(target);
3803 EXPORT_SYMBOL(vfs_unlink);
3806 * Make sure that the actual truncation of the file will occur outside its
3807 * directory's i_mutex. Truncate can take a long time if there is a lot of
3808 * writeout happening, and we don't want to prevent access to the directory
3809 * while waiting on the I/O.
3811 static long do_unlinkat(int dfd, const char __user *pathname)
3814 struct filename *name;
3815 struct dentry *dentry;
3819 struct inode *inode = NULL;
3820 struct inode *delegated_inode = NULL;
3821 unsigned int lookup_flags = 0;
3823 name = user_path_parent(dfd, pathname,
3824 &path, &last, &type, lookup_flags);
3826 return PTR_ERR(name);
3829 if (type != LAST_NORM)
3832 error = mnt_want_write(path.mnt);
3836 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3837 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3838 error = PTR_ERR(dentry);
3839 if (!IS_ERR(dentry)) {
3840 /* Why not before? Because we want correct error value */
3841 if (last.name[last.len])
3843 inode = dentry->d_inode;
3844 if (d_is_negative(dentry))
3847 error = security_path_unlink(&path, dentry);
3850 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3854 mutex_unlock(&path.dentry->d_inode->i_mutex);
3856 iput(inode); /* truncate the inode here */
3858 if (delegated_inode) {
3859 error = break_deleg_wait(&delegated_inode);
3863 mnt_drop_write(path.mnt);
3867 if (retry_estale(error, lookup_flags)) {
3868 lookup_flags |= LOOKUP_REVAL;
3875 if (d_is_negative(dentry))
3877 else if (d_is_dir(dentry))
3884 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3886 if ((flag & ~AT_REMOVEDIR) != 0)
3889 if (flag & AT_REMOVEDIR)
3890 return do_rmdir(dfd, pathname);
3892 return do_unlinkat(dfd, pathname);
3895 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3897 return do_unlinkat(AT_FDCWD, pathname);
3900 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3902 int error = may_create(dir, dentry);
3907 if (!dir->i_op->symlink)
3910 error = security_inode_symlink(dir, dentry, oldname);
3914 error = dir->i_op->symlink(dir, dentry, oldname);
3916 fsnotify_create(dir, dentry);
3919 EXPORT_SYMBOL(vfs_symlink);
3921 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3922 int, newdfd, const char __user *, newname)
3925 struct filename *from;
3926 struct dentry *dentry;
3928 unsigned int lookup_flags = 0;
3930 from = getname(oldname);
3932 return PTR_ERR(from);
3934 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3935 error = PTR_ERR(dentry);
3939 error = security_path_symlink(&path, dentry, from->name);
3941 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3942 done_path_create(&path, dentry);
3943 if (retry_estale(error, lookup_flags)) {
3944 lookup_flags |= LOOKUP_REVAL;
3952 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3954 return sys_symlinkat(oldname, AT_FDCWD, newname);
3958 * vfs_link - create a new link
3959 * @old_dentry: object to be linked
3961 * @new_dentry: where to create the new link
3962 * @delegated_inode: returns inode needing a delegation break
3964 * The caller must hold dir->i_mutex
3966 * If vfs_link discovers a delegation on the to-be-linked file in need
3967 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3968 * inode in delegated_inode. The caller should then break the delegation
3969 * and retry. Because breaking a delegation may take a long time, the
3970 * caller should drop the i_mutex before doing so.
3972 * Alternatively, a caller may pass NULL for delegated_inode. This may
3973 * be appropriate for callers that expect the underlying filesystem not
3974 * to be NFS exported.
3976 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3978 struct inode *inode = old_dentry->d_inode;
3979 unsigned max_links = dir->i_sb->s_max_links;
3985 error = may_create(dir, new_dentry);
3989 if (dir->i_sb != inode->i_sb)
3993 * A link to an append-only or immutable file cannot be created.
3995 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3997 if (!dir->i_op->link)
3999 if (S_ISDIR(inode->i_mode))
4002 error = security_inode_link(old_dentry, dir, new_dentry);
4006 mutex_lock(&inode->i_mutex);
4007 /* Make sure we don't allow creating hardlink to an unlinked file */
4008 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4010 else if (max_links && inode->i_nlink >= max_links)
4013 error = try_break_deleg(inode, delegated_inode);
4015 error = dir->i_op->link(old_dentry, dir, new_dentry);
4018 if (!error && (inode->i_state & I_LINKABLE)) {
4019 spin_lock(&inode->i_lock);
4020 inode->i_state &= ~I_LINKABLE;
4021 spin_unlock(&inode->i_lock);
4023 mutex_unlock(&inode->i_mutex);
4025 fsnotify_link(dir, inode, new_dentry);
4028 EXPORT_SYMBOL(vfs_link);
4031 * Hardlinks are often used in delicate situations. We avoid
4032 * security-related surprises by not following symlinks on the
4035 * We don't follow them on the oldname either to be compatible
4036 * with linux 2.0, and to avoid hard-linking to directories
4037 * and other special files. --ADM
4039 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4040 int, newdfd, const char __user *, newname, int, flags)
4042 struct dentry *new_dentry;
4043 struct path old_path, new_path;
4044 struct inode *delegated_inode = NULL;
4048 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4051 * To use null names we require CAP_DAC_READ_SEARCH
4052 * This ensures that not everyone will be able to create
4053 * handlink using the passed filedescriptor.
4055 if (flags & AT_EMPTY_PATH) {
4056 if (!capable(CAP_DAC_READ_SEARCH))
4061 if (flags & AT_SYMLINK_FOLLOW)
4062 how |= LOOKUP_FOLLOW;
4064 error = user_path_at(olddfd, oldname, how, &old_path);
4068 new_dentry = user_path_create(newdfd, newname, &new_path,
4069 (how & LOOKUP_REVAL));
4070 error = PTR_ERR(new_dentry);
4071 if (IS_ERR(new_dentry))
4075 if (old_path.mnt != new_path.mnt)
4077 error = may_linkat(&old_path);
4078 if (unlikely(error))
4080 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4083 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4085 done_path_create(&new_path, new_dentry);
4086 if (delegated_inode) {
4087 error = break_deleg_wait(&delegated_inode);
4089 path_put(&old_path);
4093 if (retry_estale(error, how)) {
4094 path_put(&old_path);
4095 how |= LOOKUP_REVAL;
4099 path_put(&old_path);
4104 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4106 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4110 * vfs_rename - rename a filesystem object
4111 * @old_dir: parent of source
4112 * @old_dentry: source
4113 * @new_dir: parent of destination
4114 * @new_dentry: destination
4115 * @delegated_inode: returns an inode needing a delegation break
4116 * @flags: rename flags
4118 * The caller must hold multiple mutexes--see lock_rename()).
4120 * If vfs_rename discovers a delegation in need of breaking at either
4121 * the source or destination, it will return -EWOULDBLOCK and return a
4122 * reference to the inode in delegated_inode. The caller should then
4123 * break the delegation and retry. Because breaking a delegation may
4124 * take a long time, the caller should drop all locks before doing
4127 * Alternatively, a caller may pass NULL for delegated_inode. This may
4128 * be appropriate for callers that expect the underlying filesystem not
4129 * to be NFS exported.
4131 * The worst of all namespace operations - renaming directory. "Perverted"
4132 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4134 * a) we can get into loop creation.
4135 * b) race potential - two innocent renames can create a loop together.
4136 * That's where 4.4 screws up. Current fix: serialization on
4137 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4139 * c) we have to lock _four_ objects - parents and victim (if it exists),
4140 * and source (if it is not a directory).
4141 * And that - after we got ->i_mutex on parents (until then we don't know
4142 * whether the target exists). Solution: try to be smart with locking
4143 * order for inodes. We rely on the fact that tree topology may change
4144 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4145 * move will be locked. Thus we can rank directories by the tree
4146 * (ancestors first) and rank all non-directories after them.
4147 * That works since everybody except rename does "lock parent, lookup,
4148 * lock child" and rename is under ->s_vfs_rename_mutex.
4149 * HOWEVER, it relies on the assumption that any object with ->lookup()
4150 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4151 * we'd better make sure that there's no link(2) for them.
4152 * d) conversion from fhandle to dentry may come in the wrong moment - when
4153 * we are removing the target. Solution: we will have to grab ->i_mutex
4154 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4155 * ->i_mutex on parents, which works but leads to some truly excessive
4158 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4159 struct inode *new_dir, struct dentry *new_dentry,
4160 struct inode **delegated_inode, unsigned int flags)
4163 bool is_dir = d_is_dir(old_dentry);
4164 const unsigned char *old_name;
4165 struct inode *source = old_dentry->d_inode;
4166 struct inode *target = new_dentry->d_inode;
4167 bool new_is_dir = false;
4168 unsigned max_links = new_dir->i_sb->s_max_links;
4170 if (source == target)
4173 error = may_delete(old_dir, old_dentry, is_dir);
4178 error = may_create(new_dir, new_dentry);
4180 new_is_dir = d_is_dir(new_dentry);
4182 if (!(flags & RENAME_EXCHANGE))
4183 error = may_delete(new_dir, new_dentry, is_dir);
4185 error = may_delete(new_dir, new_dentry, new_is_dir);
4190 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4193 if (flags && !old_dir->i_op->rename2)
4197 * If we are going to change the parent - check write permissions,
4198 * we'll need to flip '..'.
4200 if (new_dir != old_dir) {
4202 error = inode_permission(source, MAY_WRITE);
4206 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4207 error = inode_permission(target, MAY_WRITE);
4213 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4218 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4220 if (!is_dir || (flags & RENAME_EXCHANGE))
4221 lock_two_nondirectories(source, target);
4223 mutex_lock(&target->i_mutex);
4226 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4229 if (max_links && new_dir != old_dir) {
4231 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4233 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4234 old_dir->i_nlink >= max_links)
4237 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4238 shrink_dcache_parent(new_dentry);
4240 error = try_break_deleg(source, delegated_inode);
4244 if (target && !new_is_dir) {
4245 error = try_break_deleg(target, delegated_inode);
4249 if (!old_dir->i_op->rename2) {
4250 error = old_dir->i_op->rename(old_dir, old_dentry,
4251 new_dir, new_dentry);
4253 WARN_ON(old_dir->i_op->rename != NULL);
4254 error = old_dir->i_op->rename2(old_dir, old_dentry,
4255 new_dir, new_dentry, flags);
4260 if (!(flags & RENAME_EXCHANGE) && target) {
4262 target->i_flags |= S_DEAD;
4263 dont_mount(new_dentry);
4264 detach_mounts(new_dentry);
4266 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4267 if (!(flags & RENAME_EXCHANGE))
4268 d_move(old_dentry, new_dentry);
4270 d_exchange(old_dentry, new_dentry);
4273 if (!is_dir || (flags & RENAME_EXCHANGE))
4274 unlock_two_nondirectories(source, target);
4276 mutex_unlock(&target->i_mutex);
4279 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4280 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4281 if (flags & RENAME_EXCHANGE) {
4282 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4283 new_is_dir, NULL, new_dentry);
4286 fsnotify_oldname_free(old_name);
4290 EXPORT_SYMBOL(vfs_rename);
4292 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4293 int, newdfd, const char __user *, newname, unsigned int, flags)
4295 struct dentry *old_dentry, *new_dentry;
4296 struct dentry *trap;
4297 struct path old_path, new_path;
4298 struct qstr old_last, new_last;
4299 int old_type, new_type;
4300 struct inode *delegated_inode = NULL;
4301 struct filename *from;
4302 struct filename *to;
4303 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4304 bool should_retry = false;
4307 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4310 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4311 (flags & RENAME_EXCHANGE))
4314 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4317 if (flags & RENAME_EXCHANGE)
4321 from = user_path_parent(olddfd, oldname,
4322 &old_path, &old_last, &old_type, lookup_flags);
4324 error = PTR_ERR(from);
4328 to = user_path_parent(newdfd, newname,
4329 &new_path, &new_last, &new_type, lookup_flags);
4331 error = PTR_ERR(to);
4336 if (old_path.mnt != new_path.mnt)
4340 if (old_type != LAST_NORM)
4343 if (flags & RENAME_NOREPLACE)
4345 if (new_type != LAST_NORM)
4348 error = mnt_want_write(old_path.mnt);
4353 trap = lock_rename(new_path.dentry, old_path.dentry);
4355 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4356 error = PTR_ERR(old_dentry);
4357 if (IS_ERR(old_dentry))
4359 /* source must exist */
4361 if (d_is_negative(old_dentry))
4363 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4364 error = PTR_ERR(new_dentry);
4365 if (IS_ERR(new_dentry))
4368 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4370 if (flags & RENAME_EXCHANGE) {
4372 if (d_is_negative(new_dentry))
4375 if (!d_is_dir(new_dentry)) {
4377 if (new_last.name[new_last.len])
4381 /* unless the source is a directory trailing slashes give -ENOTDIR */
4382 if (!d_is_dir(old_dentry)) {
4384 if (old_last.name[old_last.len])
4386 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4389 /* source should not be ancestor of target */
4391 if (old_dentry == trap)
4393 /* target should not be an ancestor of source */
4394 if (!(flags & RENAME_EXCHANGE))
4396 if (new_dentry == trap)
4399 error = security_path_rename(&old_path, old_dentry,
4400 &new_path, new_dentry, flags);
4403 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4404 new_path.dentry->d_inode, new_dentry,
4405 &delegated_inode, flags);
4411 unlock_rename(new_path.dentry, old_path.dentry);
4412 if (delegated_inode) {
4413 error = break_deleg_wait(&delegated_inode);
4417 mnt_drop_write(old_path.mnt);
4419 if (retry_estale(error, lookup_flags))
4420 should_retry = true;
4421 path_put(&new_path);
4424 path_put(&old_path);
4427 should_retry = false;
4428 lookup_flags |= LOOKUP_REVAL;
4435 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4436 int, newdfd, const char __user *, newname)
4438 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4441 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4443 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4446 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4448 int error = may_create(dir, dentry);
4452 if (!dir->i_op->mknod)
4455 return dir->i_op->mknod(dir, dentry,
4456 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4458 EXPORT_SYMBOL(vfs_whiteout);
4460 int readlink_copy(char __user *buffer, int buflen, const char *link)
4462 int len = PTR_ERR(link);
4467 if (len > (unsigned) buflen)
4469 if (copy_to_user(buffer, link, len))
4474 EXPORT_SYMBOL(readlink_copy);
4477 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4478 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4479 * using) it for any given inode is up to filesystem.
4481 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4484 const char *link = dentry->d_inode->i_link;
4488 link = dentry->d_inode->i_op->follow_link(dentry, &cookie, NULL);
4490 return PTR_ERR(link);
4492 res = readlink_copy(buffer, buflen, link);
4493 if (cookie && dentry->d_inode->i_op->put_link)
4494 dentry->d_inode->i_op->put_link(dentry, cookie);
4497 EXPORT_SYMBOL(generic_readlink);
4499 /* get the link contents into pagecache */
4500 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4504 struct address_space *mapping = dentry->d_inode->i_mapping;
4505 page = read_mapping_page(mapping, 0, NULL);
4510 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4514 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4516 struct page *page = NULL;
4517 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4520 page_cache_release(page);
4524 EXPORT_SYMBOL(page_readlink);
4526 const char *page_follow_link_light(struct dentry *dentry, void **cookie, struct nameidata *nd)
4528 struct page *page = NULL;
4529 char *res = page_getlink(dentry, &page);
4534 EXPORT_SYMBOL(page_follow_link_light);
4536 void page_put_link(struct dentry *dentry, void *cookie)
4538 struct page *page = cookie;
4540 page_cache_release(page);
4542 EXPORT_SYMBOL(page_put_link);
4545 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4547 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4549 struct address_space *mapping = inode->i_mapping;
4554 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4556 flags |= AOP_FLAG_NOFS;
4559 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4560 flags, &page, &fsdata);
4564 kaddr = kmap_atomic(page);
4565 memcpy(kaddr, symname, len-1);
4566 kunmap_atomic(kaddr);
4568 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4575 mark_inode_dirty(inode);
4580 EXPORT_SYMBOL(__page_symlink);
4582 int page_symlink(struct inode *inode, const char *symname, int len)
4584 return __page_symlink(inode, symname, len,
4585 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4587 EXPORT_SYMBOL(page_symlink);
4589 const struct inode_operations page_symlink_inode_operations = {
4590 .readlink = generic_readlink,
4591 .follow_link = page_follow_link_light,
4592 .put_link = page_put_link,
4594 EXPORT_SYMBOL(page_symlink_inode_operations);