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)
1121 return ret < 0 ? ret : need_mntput;
1124 int follow_down_one(struct path *path)
1126 struct vfsmount *mounted;
1128 mounted = lookup_mnt(path);
1132 path->mnt = mounted;
1133 path->dentry = dget(mounted->mnt_root);
1138 EXPORT_SYMBOL(follow_down_one);
1140 static inline int managed_dentry_rcu(struct dentry *dentry)
1142 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1143 dentry->d_op->d_manage(dentry, true) : 0;
1147 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1148 * we meet a managed dentry that would need blocking.
1150 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1151 struct inode **inode)
1154 struct mount *mounted;
1156 * Don't forget we might have a non-mountpoint managed dentry
1157 * that wants to block transit.
1159 switch (managed_dentry_rcu(path->dentry)) {
1169 if (!d_mountpoint(path->dentry))
1170 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1172 mounted = __lookup_mnt(path->mnt, path->dentry);
1175 path->mnt = &mounted->mnt;
1176 path->dentry = mounted->mnt.mnt_root;
1177 nd->flags |= LOOKUP_JUMPED;
1178 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1180 * Update the inode too. We don't need to re-check the
1181 * dentry sequence number here after this d_inode read,
1182 * because a mount-point is always pinned.
1184 *inode = path->dentry->d_inode;
1186 return !read_seqretry(&mount_lock, nd->m_seq) &&
1187 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1190 static int follow_dotdot_rcu(struct nameidata *nd)
1192 struct inode *inode = nd->inode;
1197 if (nd->path.dentry == nd->root.dentry &&
1198 nd->path.mnt == nd->root.mnt) {
1201 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1202 struct dentry *old = nd->path.dentry;
1203 struct dentry *parent = old->d_parent;
1206 inode = parent->d_inode;
1207 seq = read_seqcount_begin(&parent->d_seq);
1208 if (read_seqcount_retry(&old->d_seq, nd->seq))
1210 nd->path.dentry = parent;
1214 if (!follow_up_rcu(&nd->path))
1216 inode = nd->path.dentry->d_inode;
1217 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1219 while (d_mountpoint(nd->path.dentry)) {
1220 struct mount *mounted;
1221 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1224 nd->path.mnt = &mounted->mnt;
1225 nd->path.dentry = mounted->mnt.mnt_root;
1226 inode = nd->path.dentry->d_inode;
1227 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1228 if (read_seqretry(&mount_lock, nd->m_seq))
1239 * Follow down to the covering mount currently visible to userspace. At each
1240 * point, the filesystem owning that dentry may be queried as to whether the
1241 * caller is permitted to proceed or not.
1243 int follow_down(struct path *path)
1248 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1249 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1250 /* Allow the filesystem to manage the transit without i_mutex
1253 * We indicate to the filesystem if someone is trying to mount
1254 * something here. This gives autofs the chance to deny anyone
1255 * other than its daemon the right to mount on its
1258 * The filesystem may sleep at this point.
1260 if (managed & DCACHE_MANAGE_TRANSIT) {
1261 BUG_ON(!path->dentry->d_op);
1262 BUG_ON(!path->dentry->d_op->d_manage);
1263 ret = path->dentry->d_op->d_manage(
1264 path->dentry, false);
1266 return ret == -EISDIR ? 0 : ret;
1269 /* Transit to a mounted filesystem. */
1270 if (managed & DCACHE_MOUNTED) {
1271 struct vfsmount *mounted = lookup_mnt(path);
1276 path->mnt = mounted;
1277 path->dentry = dget(mounted->mnt_root);
1281 /* Don't handle automount points here */
1286 EXPORT_SYMBOL(follow_down);
1289 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1291 static void follow_mount(struct path *path)
1293 while (d_mountpoint(path->dentry)) {
1294 struct vfsmount *mounted = lookup_mnt(path);
1299 path->mnt = mounted;
1300 path->dentry = dget(mounted->mnt_root);
1304 static void follow_dotdot(struct nameidata *nd)
1310 struct dentry *old = nd->path.dentry;
1312 if (nd->path.dentry == nd->root.dentry &&
1313 nd->path.mnt == nd->root.mnt) {
1316 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1317 /* rare case of legitimate dget_parent()... */
1318 nd->path.dentry = dget_parent(nd->path.dentry);
1322 if (!follow_up(&nd->path))
1325 follow_mount(&nd->path);
1326 nd->inode = nd->path.dentry->d_inode;
1330 * This looks up the name in dcache, possibly revalidates the old dentry and
1331 * allocates a new one if not found or not valid. In the need_lookup argument
1332 * returns whether i_op->lookup is necessary.
1334 * dir->d_inode->i_mutex must be held
1336 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1337 unsigned int flags, bool *need_lookup)
1339 struct dentry *dentry;
1342 *need_lookup = false;
1343 dentry = d_lookup(dir, name);
1345 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1346 error = d_revalidate(dentry, flags);
1347 if (unlikely(error <= 0)) {
1350 return ERR_PTR(error);
1352 d_invalidate(dentry);
1361 dentry = d_alloc(dir, name);
1362 if (unlikely(!dentry))
1363 return ERR_PTR(-ENOMEM);
1365 *need_lookup = true;
1371 * Call i_op->lookup on the dentry. The dentry must be negative and
1374 * dir->d_inode->i_mutex must be held
1376 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1381 /* Don't create child dentry for a dead directory. */
1382 if (unlikely(IS_DEADDIR(dir))) {
1384 return ERR_PTR(-ENOENT);
1387 old = dir->i_op->lookup(dir, dentry, flags);
1388 if (unlikely(old)) {
1395 static struct dentry *__lookup_hash(struct qstr *name,
1396 struct dentry *base, unsigned int flags)
1399 struct dentry *dentry;
1401 dentry = lookup_dcache(name, base, flags, &need_lookup);
1405 return lookup_real(base->d_inode, dentry, flags);
1409 * It's more convoluted than I'd like it to be, but... it's still fairly
1410 * small and for now I'd prefer to have fast path as straight as possible.
1411 * It _is_ time-critical.
1413 static int lookup_fast(struct nameidata *nd,
1414 struct path *path, struct inode **inode)
1416 struct vfsmount *mnt = nd->path.mnt;
1417 struct dentry *dentry, *parent = nd->path.dentry;
1423 * Rename seqlock is not required here because in the off chance
1424 * of a false negative due to a concurrent rename, we're going to
1425 * do the non-racy lookup, below.
1427 if (nd->flags & LOOKUP_RCU) {
1430 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1435 * This sequence count validates that the inode matches
1436 * the dentry name information from lookup.
1438 *inode = dentry->d_inode;
1439 negative = d_is_negative(dentry);
1440 if (read_seqcount_retry(&dentry->d_seq, seq))
1446 * This sequence count validates that the parent had no
1447 * changes while we did the lookup of the dentry above.
1449 * The memory barrier in read_seqcount_begin of child is
1450 * enough, we can use __read_seqcount_retry here.
1452 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1456 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1457 status = d_revalidate(dentry, nd->flags);
1458 if (unlikely(status <= 0)) {
1459 if (status != -ECHILD)
1465 path->dentry = dentry;
1466 if (likely(__follow_mount_rcu(nd, path, inode)))
1469 if (unlazy_walk(nd, dentry))
1472 dentry = __d_lookup(parent, &nd->last);
1475 if (unlikely(!dentry))
1478 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1479 status = d_revalidate(dentry, nd->flags);
1480 if (unlikely(status <= 0)) {
1485 d_invalidate(dentry);
1490 if (unlikely(d_is_negative(dentry))) {
1495 path->dentry = dentry;
1496 err = follow_managed(path, nd);
1497 if (unlikely(err < 0)) {
1498 path_put_conditional(path, nd);
1502 nd->flags |= LOOKUP_JUMPED;
1503 *inode = path->dentry->d_inode;
1510 /* Fast lookup failed, do it the slow way */
1511 static int lookup_slow(struct nameidata *nd, struct path *path)
1513 struct dentry *dentry, *parent;
1516 parent = nd->path.dentry;
1517 BUG_ON(nd->inode != parent->d_inode);
1519 mutex_lock(&parent->d_inode->i_mutex);
1520 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1521 mutex_unlock(&parent->d_inode->i_mutex);
1523 return PTR_ERR(dentry);
1524 path->mnt = nd->path.mnt;
1525 path->dentry = dentry;
1526 err = follow_managed(path, nd);
1527 if (unlikely(err < 0)) {
1528 path_put_conditional(path, nd);
1532 nd->flags |= LOOKUP_JUMPED;
1536 static inline int may_lookup(struct nameidata *nd)
1538 if (nd->flags & LOOKUP_RCU) {
1539 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1542 if (unlazy_walk(nd, NULL))
1545 return inode_permission(nd->inode, MAY_EXEC);
1548 static inline int handle_dots(struct nameidata *nd, int type)
1550 if (type == LAST_DOTDOT) {
1551 if (nd->flags & LOOKUP_RCU) {
1552 return follow_dotdot_rcu(nd);
1559 static void terminate_walk(struct nameidata *nd)
1561 if (!(nd->flags & LOOKUP_RCU)) {
1562 path_put(&nd->path);
1564 nd->flags &= ~LOOKUP_RCU;
1565 if (!(nd->flags & LOOKUP_ROOT))
1566 nd->root.mnt = NULL;
1569 while (unlikely(nd->depth))
1573 static int pick_link(struct nameidata *nd, struct path *link)
1576 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1577 path_to_nameidata(link, nd);
1580 if (nd->flags & LOOKUP_RCU) {
1581 if (unlikely(nd->path.mnt != link->mnt ||
1582 unlazy_walk(nd, link->dentry))) {
1586 error = nd_alloc_stack(nd);
1587 if (unlikely(error)) {
1588 path_to_nameidata(link, nd);
1597 * Do we need to follow links? We _really_ want to be able
1598 * to do this check without having to look at inode->i_op,
1599 * so we keep a cache of "no, this doesn't need follow_link"
1600 * for the common case.
1602 static inline int should_follow_link(struct nameidata *nd, struct path *link, int follow)
1604 if (likely(!d_is_symlink(link->dentry)))
1608 return pick_link(nd, link);
1611 enum {WALK_GET = 1, WALK_PUT = 2};
1613 static int walk_component(struct nameidata *nd, int flags)
1616 struct inode *inode;
1619 * "." and ".." are special - ".." especially so because it has
1620 * to be able to know about the current root directory and
1621 * parent relationships.
1623 if (unlikely(nd->last_type != LAST_NORM)) {
1624 err = handle_dots(nd, nd->last_type);
1625 if (flags & WALK_PUT)
1629 err = lookup_fast(nd, &path, &inode);
1630 if (unlikely(err)) {
1634 err = lookup_slow(nd, &path);
1638 inode = path.dentry->d_inode;
1640 if (d_is_negative(path.dentry))
1644 if (flags & WALK_PUT)
1646 err = should_follow_link(nd, &path, flags & WALK_GET);
1649 path_to_nameidata(&path, nd);
1654 path_to_nameidata(&path, nd);
1659 * We can do the critical dentry name comparison and hashing
1660 * operations one word at a time, but we are limited to:
1662 * - Architectures with fast unaligned word accesses. We could
1663 * do a "get_unaligned()" if this helps and is sufficiently
1666 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1667 * do not trap on the (extremely unlikely) case of a page
1668 * crossing operation.
1670 * - Furthermore, we need an efficient 64-bit compile for the
1671 * 64-bit case in order to generate the "number of bytes in
1672 * the final mask". Again, that could be replaced with a
1673 * efficient population count instruction or similar.
1675 #ifdef CONFIG_DCACHE_WORD_ACCESS
1677 #include <asm/word-at-a-time.h>
1681 static inline unsigned int fold_hash(unsigned long hash)
1683 return hash_64(hash, 32);
1686 #else /* 32-bit case */
1688 #define fold_hash(x) (x)
1692 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1694 unsigned long a, mask;
1695 unsigned long hash = 0;
1698 a = load_unaligned_zeropad(name);
1699 if (len < sizeof(unsigned long))
1703 name += sizeof(unsigned long);
1704 len -= sizeof(unsigned long);
1708 mask = bytemask_from_count(len);
1711 return fold_hash(hash);
1713 EXPORT_SYMBOL(full_name_hash);
1716 * Calculate the length and hash of the path component, and
1717 * return the "hash_len" as the result.
1719 static inline u64 hash_name(const char *name)
1721 unsigned long a, b, adata, bdata, mask, hash, len;
1722 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1725 len = -sizeof(unsigned long);
1727 hash = (hash + a) * 9;
1728 len += sizeof(unsigned long);
1729 a = load_unaligned_zeropad(name+len);
1730 b = a ^ REPEAT_BYTE('/');
1731 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1733 adata = prep_zero_mask(a, adata, &constants);
1734 bdata = prep_zero_mask(b, bdata, &constants);
1736 mask = create_zero_mask(adata | bdata);
1738 hash += a & zero_bytemask(mask);
1739 len += find_zero(mask);
1740 return hashlen_create(fold_hash(hash), len);
1745 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1747 unsigned long hash = init_name_hash();
1749 hash = partial_name_hash(*name++, hash);
1750 return end_name_hash(hash);
1752 EXPORT_SYMBOL(full_name_hash);
1755 * We know there's a real path component here of at least
1758 static inline u64 hash_name(const char *name)
1760 unsigned long hash = init_name_hash();
1761 unsigned long len = 0, c;
1763 c = (unsigned char)*name;
1766 hash = partial_name_hash(c, hash);
1767 c = (unsigned char)name[len];
1768 } while (c && c != '/');
1769 return hashlen_create(end_name_hash(hash), len);
1776 * This is the basic name resolution function, turning a pathname into
1777 * the final dentry. We expect 'base' to be positive and a directory.
1779 * Returns 0 and nd will have valid dentry and mnt on success.
1780 * Returns error and drops reference to input namei data on failure.
1782 static int link_path_walk(const char *name, struct nameidata *nd)
1791 /* At this point we know we have a real path component. */
1796 err = may_lookup(nd);
1800 hash_len = hash_name(name);
1803 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1805 if (name[1] == '.') {
1807 nd->flags |= LOOKUP_JUMPED;
1813 if (likely(type == LAST_NORM)) {
1814 struct dentry *parent = nd->path.dentry;
1815 nd->flags &= ~LOOKUP_JUMPED;
1816 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1817 struct qstr this = { { .hash_len = hash_len }, .name = name };
1818 err = parent->d_op->d_hash(parent, &this);
1821 hash_len = this.hash_len;
1826 nd->last.hash_len = hash_len;
1827 nd->last.name = name;
1828 nd->last_type = type;
1830 name += hashlen_len(hash_len);
1834 * If it wasn't NUL, we know it was '/'. Skip that
1835 * slash, and continue until no more slashes.
1839 } while (unlikely(*name == '/'));
1840 if (unlikely(!*name)) {
1842 /* called from path_init(), done */
1845 name = nd->stack[nd->depth - 1].name;
1846 /* called from trailing_symlink(), done */
1849 /* last component of nested symlink */
1850 err = walk_component(nd, WALK_GET | WALK_PUT);
1852 err = walk_component(nd, WALK_GET);
1858 const char *s = get_link(nd);
1860 if (unlikely(IS_ERR(s))) {
1872 path_put(&nd->path);
1873 nd->path = nd->root;
1874 path_get(&nd->root);
1875 nd->flags |= LOOKUP_JUMPED;
1876 while (unlikely(*++s == '/'))
1879 nd->inode = nd->path.dentry->d_inode;
1880 nd->stack[nd->depth - 1].name = name;
1887 if (!d_can_lookup(nd->path.dentry)) {
1896 static int path_init(int dfd, const struct filename *name, unsigned int flags,
1897 struct nameidata *nd)
1900 const char *s = name->name;
1902 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1903 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1906 if (flags & LOOKUP_ROOT) {
1907 struct dentry *root = nd->root.dentry;
1908 struct inode *inode = root->d_inode;
1910 if (!d_can_lookup(root))
1912 retval = inode_permission(inode, MAY_EXEC);
1916 nd->path = nd->root;
1918 if (flags & LOOKUP_RCU) {
1920 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1921 nd->m_seq = read_seqbegin(&mount_lock);
1923 path_get(&nd->path);
1928 nd->root.mnt = NULL;
1930 nd->m_seq = read_seqbegin(&mount_lock);
1932 if (flags & LOOKUP_RCU) {
1934 nd->seq = set_root_rcu(nd);
1937 path_get(&nd->root);
1939 nd->path = nd->root;
1940 } else if (dfd == AT_FDCWD) {
1941 if (flags & LOOKUP_RCU) {
1942 struct fs_struct *fs = current->fs;
1948 seq = read_seqcount_begin(&fs->seq);
1950 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1951 } while (read_seqcount_retry(&fs->seq, seq));
1953 get_fs_pwd(current->fs, &nd->path);
1956 /* Caller must check execute permissions on the starting path component */
1957 struct fd f = fdget_raw(dfd);
1958 struct dentry *dentry;
1963 dentry = f.file->f_path.dentry;
1966 if (!d_can_lookup(dentry)) {
1972 nd->path = f.file->f_path;
1973 if (flags & LOOKUP_RCU) {
1974 if (f.flags & FDPUT_FPUT)
1976 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1979 path_get(&nd->path);
1984 nd->inode = nd->path.dentry->d_inode;
1985 if (!(flags & LOOKUP_RCU))
1987 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1989 if (!(nd->flags & LOOKUP_ROOT))
1990 nd->root.mnt = NULL;
1994 nd->total_link_count = 0;
1995 return link_path_walk(s, nd);
1998 static void path_cleanup(struct nameidata *nd)
2000 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2001 path_put(&nd->root);
2002 nd->root.mnt = NULL;
2004 if (unlikely(nd->base))
2008 static int trailing_symlink(struct nameidata *nd)
2011 int error = may_follow_link(&nd->link, nd);
2012 if (unlikely(error))
2014 nd->flags |= LOOKUP_PARENT;
2016 if (unlikely(IS_ERR(s))) {
2025 path_put(&nd->path);
2026 nd->path = nd->root;
2027 path_get(&nd->root);
2028 nd->flags |= LOOKUP_JUMPED;
2030 nd->inode = nd->path.dentry->d_inode;
2031 nd->stack[0].name = NULL;
2032 return link_path_walk(s, nd);
2035 static inline int lookup_last(struct nameidata *nd)
2038 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2039 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2041 nd->flags &= ~LOOKUP_PARENT;
2042 err = walk_component(nd,
2043 nd->flags & LOOKUP_FOLLOW
2045 ? WALK_PUT | WALK_GET
2053 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2054 static int path_lookupat(int dfd, const struct filename *name,
2055 unsigned int flags, struct nameidata *nd)
2060 * Path walking is largely split up into 2 different synchronisation
2061 * schemes, rcu-walk and ref-walk (explained in
2062 * Documentation/filesystems/path-lookup.txt). These share much of the
2063 * path walk code, but some things particularly setup, cleanup, and
2064 * following mounts are sufficiently divergent that functions are
2065 * duplicated. Typically there is a function foo(), and its RCU
2066 * analogue, foo_rcu().
2068 * -ECHILD is the error number of choice (just to avoid clashes) that
2069 * is returned if some aspect of an rcu-walk fails. Such an error must
2070 * be handled by restarting a traditional ref-walk (which will always
2071 * be able to complete).
2073 err = path_init(dfd, name, flags, nd);
2074 if (!err && !(flags & LOOKUP_PARENT)) {
2075 while ((err = lookup_last(nd)) > 0) {
2076 err = trailing_symlink(nd);
2083 err = complete_walk(nd);
2085 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2086 if (!d_can_lookup(nd->path.dentry)) {
2087 path_put(&nd->path);
2096 static int filename_lookup(int dfd, struct filename *name,
2097 unsigned int flags, struct nameidata *nd)
2100 struct nameidata *saved_nd = set_nameidata(nd);
2102 retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2103 if (unlikely(retval == -ECHILD))
2104 retval = path_lookupat(dfd, name, flags, nd);
2105 if (unlikely(retval == -ESTALE))
2106 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2108 if (likely(!retval))
2109 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2110 restore_nameidata(saved_nd);
2114 /* does lookup, returns the object with parent locked */
2115 struct dentry *kern_path_locked(const char *name, struct path *path)
2117 struct filename *filename = getname_kernel(name);
2118 struct nameidata nd;
2122 if (IS_ERR(filename))
2123 return ERR_CAST(filename);
2125 err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd);
2130 if (nd.last_type != LAST_NORM) {
2132 d = ERR_PTR(-EINVAL);
2135 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2136 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2138 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2148 int kern_path(const char *name, unsigned int flags, struct path *path)
2150 struct nameidata nd;
2151 struct filename *filename = getname_kernel(name);
2152 int res = PTR_ERR(filename);
2154 if (!IS_ERR(filename)) {
2155 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2162 EXPORT_SYMBOL(kern_path);
2165 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2166 * @dentry: pointer to dentry of the base directory
2167 * @mnt: pointer to vfs mount of the base directory
2168 * @name: pointer to file name
2169 * @flags: lookup flags
2170 * @path: pointer to struct path to fill
2172 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2173 const char *name, unsigned int flags,
2176 struct filename *filename = getname_kernel(name);
2177 int err = PTR_ERR(filename);
2179 BUG_ON(flags & LOOKUP_PARENT);
2181 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2182 if (!IS_ERR(filename)) {
2183 struct nameidata nd;
2184 nd.root.dentry = dentry;
2186 err = filename_lookup(AT_FDCWD, filename,
2187 flags | LOOKUP_ROOT, &nd);
2194 EXPORT_SYMBOL(vfs_path_lookup);
2197 * lookup_one_len - filesystem helper to lookup single pathname component
2198 * @name: pathname component to lookup
2199 * @base: base directory to lookup from
2200 * @len: maximum length @len should be interpreted to
2202 * Note that this routine is purely a helper for filesystem usage and should
2203 * not be called by generic code.
2205 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2211 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2215 this.hash = full_name_hash(name, len);
2217 return ERR_PTR(-EACCES);
2219 if (unlikely(name[0] == '.')) {
2220 if (len < 2 || (len == 2 && name[1] == '.'))
2221 return ERR_PTR(-EACCES);
2225 c = *(const unsigned char *)name++;
2226 if (c == '/' || c == '\0')
2227 return ERR_PTR(-EACCES);
2230 * See if the low-level filesystem might want
2231 * to use its own hash..
2233 if (base->d_flags & DCACHE_OP_HASH) {
2234 int err = base->d_op->d_hash(base, &this);
2236 return ERR_PTR(err);
2239 err = inode_permission(base->d_inode, MAY_EXEC);
2241 return ERR_PTR(err);
2243 return __lookup_hash(&this, base, 0);
2245 EXPORT_SYMBOL(lookup_one_len);
2247 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2248 struct path *path, int *empty)
2250 struct nameidata nd;
2251 struct filename *tmp = getname_flags(name, flags, empty);
2252 int err = PTR_ERR(tmp);
2255 BUG_ON(flags & LOOKUP_PARENT);
2257 err = filename_lookup(dfd, tmp, flags, &nd);
2265 int user_path_at(int dfd, const char __user *name, unsigned flags,
2268 return user_path_at_empty(dfd, name, flags, path, NULL);
2270 EXPORT_SYMBOL(user_path_at);
2273 * NB: most callers don't do anything directly with the reference to the
2274 * to struct filename, but the nd->last pointer points into the name string
2275 * allocated by getname. So we must hold the reference to it until all
2276 * path-walking is complete.
2278 static struct filename *
2279 user_path_parent(int dfd, const char __user *path,
2280 struct path *parent,
2285 struct nameidata nd;
2286 struct filename *s = getname(path);
2289 /* only LOOKUP_REVAL is allowed in extra flags */
2290 flags &= LOOKUP_REVAL;
2295 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, &nd);
2298 return ERR_PTR(error);
2302 *type = nd.last_type;
2308 * mountpoint_last - look up last component for umount
2309 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2310 * @path: pointer to container for result
2312 * This is a special lookup_last function just for umount. In this case, we
2313 * need to resolve the path without doing any revalidation.
2315 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2316 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2317 * in almost all cases, this lookup will be served out of the dcache. The only
2318 * cases where it won't are if nd->last refers to a symlink or the path is
2319 * bogus and it doesn't exist.
2322 * -error: if there was an error during lookup. This includes -ENOENT if the
2323 * lookup found a negative dentry. The nd->path reference will also be
2326 * 0: if we successfully resolved nd->path and found it to not to be a
2327 * symlink that needs to be followed. "path" will also be populated.
2328 * The nd->path reference will also be put.
2330 * 1: if we successfully resolved nd->last and found it to be a symlink
2331 * that needs to be followed. "path" will be populated with the path
2332 * to the link, and nd->path will *not* be put.
2335 mountpoint_last(struct nameidata *nd, struct path *path)
2338 struct dentry *dentry;
2339 struct dentry *dir = nd->path.dentry;
2341 /* If we're in rcuwalk, drop out of it to handle last component */
2342 if (nd->flags & LOOKUP_RCU) {
2343 if (unlazy_walk(nd, NULL)) {
2349 nd->flags &= ~LOOKUP_PARENT;
2351 if (unlikely(nd->last_type != LAST_NORM)) {
2352 error = handle_dots(nd, nd->last_type);
2355 dentry = dget(nd->path.dentry);
2359 mutex_lock(&dir->d_inode->i_mutex);
2360 dentry = d_lookup(dir, &nd->last);
2363 * No cached dentry. Mounted dentries are pinned in the cache,
2364 * so that means that this dentry is probably a symlink or the
2365 * path doesn't actually point to a mounted dentry.
2367 dentry = d_alloc(dir, &nd->last);
2370 mutex_unlock(&dir->d_inode->i_mutex);
2373 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2374 error = PTR_ERR(dentry);
2375 if (IS_ERR(dentry)) {
2376 mutex_unlock(&dir->d_inode->i_mutex);
2380 mutex_unlock(&dir->d_inode->i_mutex);
2383 if (d_is_negative(dentry)) {
2390 path->dentry = dentry;
2391 path->mnt = nd->path.mnt;
2392 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW);
2393 if (unlikely(error)) {
2407 * path_mountpoint - look up a path to be umounted
2408 * @dfd: directory file descriptor to start walk from
2409 * @name: full pathname to walk
2410 * @path: pointer to container for result
2411 * @flags: lookup flags
2413 * Look up the given name, but don't attempt to revalidate the last component.
2414 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2417 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2418 struct nameidata *nd, unsigned int flags)
2420 int err = path_init(dfd, name, flags, nd);
2424 while ((err = mountpoint_last(nd, path)) > 0) {
2425 err = trailing_symlink(nd);
2435 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2438 struct nameidata nd, *saved;
2441 return PTR_ERR(name);
2442 saved = set_nameidata(&nd);
2443 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2444 if (unlikely(error == -ECHILD))
2445 error = path_mountpoint(dfd, name, path, &nd, flags);
2446 if (unlikely(error == -ESTALE))
2447 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2449 audit_inode(name, path->dentry, 0);
2450 restore_nameidata(saved);
2456 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2457 * @dfd: directory file descriptor
2458 * @name: pathname from userland
2459 * @flags: lookup flags
2460 * @path: pointer to container to hold result
2462 * A umount is a special case for path walking. We're not actually interested
2463 * in the inode in this situation, and ESTALE errors can be a problem. We
2464 * simply want track down the dentry and vfsmount attached at the mountpoint
2465 * and avoid revalidating the last component.
2467 * Returns 0 and populates "path" on success.
2470 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2473 return filename_mountpoint(dfd, getname(name), path, flags);
2477 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2480 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2482 EXPORT_SYMBOL(kern_path_mountpoint);
2484 int __check_sticky(struct inode *dir, struct inode *inode)
2486 kuid_t fsuid = current_fsuid();
2488 if (uid_eq(inode->i_uid, fsuid))
2490 if (uid_eq(dir->i_uid, fsuid))
2492 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2494 EXPORT_SYMBOL(__check_sticky);
2497 * Check whether we can remove a link victim from directory dir, check
2498 * whether the type of victim is right.
2499 * 1. We can't do it if dir is read-only (done in permission())
2500 * 2. We should have write and exec permissions on dir
2501 * 3. We can't remove anything from append-only dir
2502 * 4. We can't do anything with immutable dir (done in permission())
2503 * 5. If the sticky bit on dir is set we should either
2504 * a. be owner of dir, or
2505 * b. be owner of victim, or
2506 * c. have CAP_FOWNER capability
2507 * 6. If the victim is append-only or immutable we can't do antyhing with
2508 * links pointing to it.
2509 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2510 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2511 * 9. We can't remove a root or mountpoint.
2512 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2513 * nfs_async_unlink().
2515 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2517 struct inode *inode = victim->d_inode;
2520 if (d_is_negative(victim))
2524 BUG_ON(victim->d_parent->d_inode != dir);
2525 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2527 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2533 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2534 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2537 if (!d_is_dir(victim))
2539 if (IS_ROOT(victim))
2541 } else if (d_is_dir(victim))
2543 if (IS_DEADDIR(dir))
2545 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2550 /* Check whether we can create an object with dentry child in directory
2552 * 1. We can't do it if child already exists (open has special treatment for
2553 * this case, but since we are inlined it's OK)
2554 * 2. We can't do it if dir is read-only (done in permission())
2555 * 3. We should have write and exec permissions on dir
2556 * 4. We can't do it if dir is immutable (done in permission())
2558 static inline int may_create(struct inode *dir, struct dentry *child)
2560 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2563 if (IS_DEADDIR(dir))
2565 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2569 * p1 and p2 should be directories on the same fs.
2571 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2576 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2580 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2582 p = d_ancestor(p2, p1);
2584 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2585 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2589 p = d_ancestor(p1, p2);
2591 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2592 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2596 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2597 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2600 EXPORT_SYMBOL(lock_rename);
2602 void unlock_rename(struct dentry *p1, struct dentry *p2)
2604 mutex_unlock(&p1->d_inode->i_mutex);
2606 mutex_unlock(&p2->d_inode->i_mutex);
2607 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2610 EXPORT_SYMBOL(unlock_rename);
2612 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2615 int error = may_create(dir, dentry);
2619 if (!dir->i_op->create)
2620 return -EACCES; /* shouldn't it be ENOSYS? */
2623 error = security_inode_create(dir, dentry, mode);
2626 error = dir->i_op->create(dir, dentry, mode, want_excl);
2628 fsnotify_create(dir, dentry);
2631 EXPORT_SYMBOL(vfs_create);
2633 static int may_open(struct path *path, int acc_mode, int flag)
2635 struct dentry *dentry = path->dentry;
2636 struct inode *inode = dentry->d_inode;
2646 switch (inode->i_mode & S_IFMT) {
2650 if (acc_mode & MAY_WRITE)
2655 if (path->mnt->mnt_flags & MNT_NODEV)
2664 error = inode_permission(inode, acc_mode);
2669 * An append-only file must be opened in append mode for writing.
2671 if (IS_APPEND(inode)) {
2672 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2678 /* O_NOATIME can only be set by the owner or superuser */
2679 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2685 static int handle_truncate(struct file *filp)
2687 struct path *path = &filp->f_path;
2688 struct inode *inode = path->dentry->d_inode;
2689 int error = get_write_access(inode);
2693 * Refuse to truncate files with mandatory locks held on them.
2695 error = locks_verify_locked(filp);
2697 error = security_path_truncate(path);
2699 error = do_truncate(path->dentry, 0,
2700 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2703 put_write_access(inode);
2707 static inline int open_to_namei_flags(int flag)
2709 if ((flag & O_ACCMODE) == 3)
2714 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2716 int error = security_path_mknod(dir, dentry, mode, 0);
2720 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2724 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2728 * Attempt to atomically look up, create and open a file from a negative
2731 * Returns 0 if successful. The file will have been created and attached to
2732 * @file by the filesystem calling finish_open().
2734 * Returns 1 if the file was looked up only or didn't need creating. The
2735 * caller will need to perform the open themselves. @path will have been
2736 * updated to point to the new dentry. This may be negative.
2738 * Returns an error code otherwise.
2740 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2741 struct path *path, struct file *file,
2742 const struct open_flags *op,
2743 bool got_write, bool need_lookup,
2746 struct inode *dir = nd->path.dentry->d_inode;
2747 unsigned open_flag = open_to_namei_flags(op->open_flag);
2751 int create_error = 0;
2752 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2755 BUG_ON(dentry->d_inode);
2757 /* Don't create child dentry for a dead directory. */
2758 if (unlikely(IS_DEADDIR(dir))) {
2764 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2765 mode &= ~current_umask();
2767 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2769 open_flag &= ~O_TRUNC;
2772 * Checking write permission is tricky, bacuse we don't know if we are
2773 * going to actually need it: O_CREAT opens should work as long as the
2774 * file exists. But checking existence breaks atomicity. The trick is
2775 * to check access and if not granted clear O_CREAT from the flags.
2777 * Another problem is returing the "right" error value (e.g. for an
2778 * O_EXCL open we want to return EEXIST not EROFS).
2780 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2781 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2782 if (!(open_flag & O_CREAT)) {
2784 * No O_CREATE -> atomicity not a requirement -> fall
2785 * back to lookup + open
2788 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2789 /* Fall back and fail with the right error */
2790 create_error = -EROFS;
2793 /* No side effects, safe to clear O_CREAT */
2794 create_error = -EROFS;
2795 open_flag &= ~O_CREAT;
2799 if (open_flag & O_CREAT) {
2800 error = may_o_create(&nd->path, dentry, mode);
2802 create_error = error;
2803 if (open_flag & O_EXCL)
2805 open_flag &= ~O_CREAT;
2809 if (nd->flags & LOOKUP_DIRECTORY)
2810 open_flag |= O_DIRECTORY;
2812 file->f_path.dentry = DENTRY_NOT_SET;
2813 file->f_path.mnt = nd->path.mnt;
2814 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2817 if (create_error && error == -ENOENT)
2818 error = create_error;
2822 if (error) { /* returned 1, that is */
2823 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2827 if (file->f_path.dentry) {
2829 dentry = file->f_path.dentry;
2831 if (*opened & FILE_CREATED)
2832 fsnotify_create(dir, dentry);
2833 if (!dentry->d_inode) {
2834 WARN_ON(*opened & FILE_CREATED);
2836 error = create_error;
2840 if (excl && !(*opened & FILE_CREATED)) {
2849 * We didn't have the inode before the open, so check open permission
2852 acc_mode = op->acc_mode;
2853 if (*opened & FILE_CREATED) {
2854 WARN_ON(!(open_flag & O_CREAT));
2855 fsnotify_create(dir, dentry);
2856 acc_mode = MAY_OPEN;
2858 error = may_open(&file->f_path, acc_mode, open_flag);
2868 dentry = lookup_real(dir, dentry, nd->flags);
2870 return PTR_ERR(dentry);
2873 int open_flag = op->open_flag;
2875 error = create_error;
2876 if ((open_flag & O_EXCL)) {
2877 if (!dentry->d_inode)
2879 } else if (!dentry->d_inode) {
2881 } else if ((open_flag & O_TRUNC) &&
2885 /* will fail later, go on to get the right error */
2889 path->dentry = dentry;
2890 path->mnt = nd->path.mnt;
2895 * Look up and maybe create and open the last component.
2897 * Must be called with i_mutex held on parent.
2899 * Returns 0 if the file was successfully atomically created (if necessary) and
2900 * opened. In this case the file will be returned attached to @file.
2902 * Returns 1 if the file was not completely opened at this time, though lookups
2903 * and creations will have been performed and the dentry returned in @path will
2904 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2905 * specified then a negative dentry may be returned.
2907 * An error code is returned otherwise.
2909 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2910 * cleared otherwise prior to returning.
2912 static int lookup_open(struct nameidata *nd, struct path *path,
2914 const struct open_flags *op,
2915 bool got_write, int *opened)
2917 struct dentry *dir = nd->path.dentry;
2918 struct inode *dir_inode = dir->d_inode;
2919 struct dentry *dentry;
2923 *opened &= ~FILE_CREATED;
2924 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2926 return PTR_ERR(dentry);
2928 /* Cached positive dentry: will open in f_op->open */
2929 if (!need_lookup && dentry->d_inode)
2932 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2933 return atomic_open(nd, dentry, path, file, op, got_write,
2934 need_lookup, opened);
2938 BUG_ON(dentry->d_inode);
2940 dentry = lookup_real(dir_inode, dentry, nd->flags);
2942 return PTR_ERR(dentry);
2945 /* Negative dentry, just create the file */
2946 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2947 umode_t mode = op->mode;
2948 if (!IS_POSIXACL(dir->d_inode))
2949 mode &= ~current_umask();
2951 * This write is needed to ensure that a
2952 * rw->ro transition does not occur between
2953 * the time when the file is created and when
2954 * a permanent write count is taken through
2955 * the 'struct file' in finish_open().
2961 *opened |= FILE_CREATED;
2962 error = security_path_mknod(&nd->path, dentry, mode, 0);
2965 error = vfs_create(dir->d_inode, dentry, mode,
2966 nd->flags & LOOKUP_EXCL);
2971 path->dentry = dentry;
2972 path->mnt = nd->path.mnt;
2981 * Handle the last step of open()
2983 static int do_last(struct nameidata *nd,
2984 struct file *file, const struct open_flags *op,
2985 int *opened, struct filename *name)
2987 struct dentry *dir = nd->path.dentry;
2988 int open_flag = op->open_flag;
2989 bool will_truncate = (open_flag & O_TRUNC) != 0;
2990 bool got_write = false;
2991 int acc_mode = op->acc_mode;
2992 struct inode *inode;
2993 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2995 bool retried = false;
2998 nd->flags &= ~LOOKUP_PARENT;
2999 nd->flags |= op->intent;
3001 if (nd->last_type != LAST_NORM) {
3002 error = handle_dots(nd, nd->last_type);
3003 if (unlikely(error)) {
3010 if (!(open_flag & O_CREAT)) {
3011 if (nd->last.name[nd->last.len])
3012 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3013 /* we _can_ be in RCU mode here */
3014 error = lookup_fast(nd, &path, &inode);
3021 BUG_ON(nd->inode != dir->d_inode);
3023 /* create side of things */
3025 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3026 * has been cleared when we got to the last component we are
3029 error = complete_walk(nd);
3036 audit_inode(name, dir, LOOKUP_PARENT);
3038 /* trailing slashes? */
3039 if (nd->last.name[nd->last.len])
3044 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3045 error = mnt_want_write(nd->path.mnt);
3049 * do _not_ fail yet - we might not need that or fail with
3050 * a different error; let lookup_open() decide; we'll be
3051 * dropping this one anyway.
3054 mutex_lock(&dir->d_inode->i_mutex);
3055 error = lookup_open(nd, &path, file, op, got_write, opened);
3056 mutex_unlock(&dir->d_inode->i_mutex);
3062 if ((*opened & FILE_CREATED) ||
3063 !S_ISREG(file_inode(file)->i_mode))
3064 will_truncate = false;
3066 audit_inode(name, file->f_path.dentry, 0);
3070 if (*opened & FILE_CREATED) {
3071 /* Don't check for write permission, don't truncate */
3072 open_flag &= ~O_TRUNC;
3073 will_truncate = false;
3074 acc_mode = MAY_OPEN;
3075 path_to_nameidata(&path, nd);
3076 goto finish_open_created;
3080 * create/update audit record if it already exists.
3082 if (d_is_positive(path.dentry))
3083 audit_inode(name, path.dentry, 0);
3086 * If atomic_open() acquired write access it is dropped now due to
3087 * possible mount and symlink following (this might be optimized away if
3091 mnt_drop_write(nd->path.mnt);
3096 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3099 error = follow_managed(&path, nd);
3104 nd->flags |= LOOKUP_JUMPED;
3106 BUG_ON(nd->flags & LOOKUP_RCU);
3107 inode = path.dentry->d_inode;
3109 if (d_is_negative(path.dentry)) {
3110 path_to_nameidata(&path, nd);
3116 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW);
3117 if (unlikely(error)) {
3123 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3124 path_to_nameidata(&path, nd);
3129 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3130 path_to_nameidata(&path, nd);
3132 save_parent.dentry = nd->path.dentry;
3133 save_parent.mnt = mntget(path.mnt);
3134 nd->path.dentry = path.dentry;
3138 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3140 error = complete_walk(nd);
3144 path_put(&save_parent);
3147 audit_inode(name, nd->path.dentry, 0);
3149 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3152 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3154 if (!d_is_reg(nd->path.dentry))
3155 will_truncate = false;
3157 if (will_truncate) {
3158 error = mnt_want_write(nd->path.mnt);
3163 finish_open_created:
3164 error = may_open(&nd->path, acc_mode, open_flag);
3168 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3169 error = vfs_open(&nd->path, file, current_cred());
3171 *opened |= FILE_OPENED;
3173 if (error == -EOPENSTALE)
3178 error = open_check_o_direct(file);
3181 error = ima_file_check(file, op->acc_mode, *opened);
3185 if (will_truncate) {
3186 error = handle_truncate(file);
3192 mnt_drop_write(nd->path.mnt);
3193 path_put(&save_parent);
3198 path_put_conditional(&path, nd);
3205 /* If no saved parent or already retried then can't retry */
3206 if (!save_parent.dentry || retried)
3209 BUG_ON(save_parent.dentry != dir);
3210 path_put(&nd->path);
3211 nd->path = save_parent;
3212 nd->inode = dir->d_inode;
3213 save_parent.mnt = NULL;
3214 save_parent.dentry = NULL;
3216 mnt_drop_write(nd->path.mnt);
3223 static int do_tmpfile(int dfd, struct filename *pathname,
3224 struct nameidata *nd, int flags,
3225 const struct open_flags *op,
3226 struct file *file, int *opened)
3228 static const struct qstr name = QSTR_INIT("/", 1);
3229 struct dentry *dentry, *child;
3231 int error = path_lookupat(dfd, pathname,
3232 flags | LOOKUP_DIRECTORY, nd);
3233 if (unlikely(error))
3235 error = mnt_want_write(nd->path.mnt);
3236 if (unlikely(error))
3238 /* we want directory to be writable */
3239 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3242 dentry = nd->path.dentry;
3243 dir = dentry->d_inode;
3244 if (!dir->i_op->tmpfile) {
3245 error = -EOPNOTSUPP;
3248 child = d_alloc(dentry, &name);
3249 if (unlikely(!child)) {
3253 nd->flags &= ~LOOKUP_DIRECTORY;
3254 nd->flags |= op->intent;
3255 dput(nd->path.dentry);
3256 nd->path.dentry = child;
3257 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3260 audit_inode(pathname, nd->path.dentry, 0);
3261 /* Don't check for other permissions, the inode was just created */
3262 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3265 file->f_path.mnt = nd->path.mnt;
3266 error = finish_open(file, nd->path.dentry, NULL, opened);
3269 error = open_check_o_direct(file);
3272 } else if (!(op->open_flag & O_EXCL)) {
3273 struct inode *inode = file_inode(file);
3274 spin_lock(&inode->i_lock);
3275 inode->i_state |= I_LINKABLE;
3276 spin_unlock(&inode->i_lock);
3279 mnt_drop_write(nd->path.mnt);
3281 path_put(&nd->path);
3285 static struct file *path_openat(int dfd, struct filename *pathname,
3286 struct nameidata *nd, const struct open_flags *op, int flags)
3292 file = get_empty_filp();
3296 file->f_flags = op->open_flag;
3298 if (unlikely(file->f_flags & __O_TMPFILE)) {
3299 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3303 error = path_init(dfd, pathname, flags, nd);
3304 if (unlikely(error))
3307 while ((error = do_last(nd, file, op, &opened, pathname)) > 0) {
3308 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3309 error = trailing_symlink(nd);
3310 if (unlikely(error))
3316 if (!(opened & FILE_OPENED)) {
3320 if (unlikely(error)) {
3321 if (error == -EOPENSTALE) {
3322 if (flags & LOOKUP_RCU)
3327 file = ERR_PTR(error);
3332 struct file *do_filp_open(int dfd, struct filename *pathname,
3333 const struct open_flags *op)
3335 struct nameidata nd, *saved_nd = set_nameidata(&nd);
3336 int flags = op->lookup_flags;
3339 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3340 if (unlikely(filp == ERR_PTR(-ECHILD)))
3341 filp = path_openat(dfd, pathname, &nd, op, flags);
3342 if (unlikely(filp == ERR_PTR(-ESTALE)))
3343 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3344 restore_nameidata(saved_nd);
3348 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3349 const char *name, const struct open_flags *op)
3351 struct nameidata nd, *saved_nd;
3353 struct filename *filename;
3354 int flags = op->lookup_flags | LOOKUP_ROOT;
3357 nd.root.dentry = dentry;
3359 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3360 return ERR_PTR(-ELOOP);
3362 filename = getname_kernel(name);
3363 if (unlikely(IS_ERR(filename)))
3364 return ERR_CAST(filename);
3366 saved_nd = set_nameidata(&nd);
3367 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3368 if (unlikely(file == ERR_PTR(-ECHILD)))
3369 file = path_openat(-1, filename, &nd, op, flags);
3370 if (unlikely(file == ERR_PTR(-ESTALE)))
3371 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3372 restore_nameidata(saved_nd);
3377 static struct dentry *filename_create(int dfd, struct filename *name,
3378 struct path *path, unsigned int lookup_flags)
3380 struct dentry *dentry = ERR_PTR(-EEXIST);
3381 struct nameidata nd;
3384 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3387 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3388 * other flags passed in are ignored!
3390 lookup_flags &= LOOKUP_REVAL;
3392 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3394 return ERR_PTR(error);
3397 * Yucky last component or no last component at all?
3398 * (foo/., foo/.., /////)
3400 if (nd.last_type != LAST_NORM)
3402 nd.flags &= ~LOOKUP_PARENT;
3403 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3405 /* don't fail immediately if it's r/o, at least try to report other errors */
3406 err2 = mnt_want_write(nd.path.mnt);
3408 * Do the final lookup.
3410 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3411 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3416 if (d_is_positive(dentry))
3420 * Special case - lookup gave negative, but... we had foo/bar/
3421 * From the vfs_mknod() POV we just have a negative dentry -
3422 * all is fine. Let's be bastards - you had / on the end, you've
3423 * been asking for (non-existent) directory. -ENOENT for you.
3425 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3429 if (unlikely(err2)) {
3437 dentry = ERR_PTR(error);
3439 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3441 mnt_drop_write(nd.path.mnt);
3447 struct dentry *kern_path_create(int dfd, const char *pathname,
3448 struct path *path, unsigned int lookup_flags)
3450 struct filename *filename = getname_kernel(pathname);
3453 if (IS_ERR(filename))
3454 return ERR_CAST(filename);
3455 res = filename_create(dfd, filename, path, lookup_flags);
3459 EXPORT_SYMBOL(kern_path_create);
3461 void done_path_create(struct path *path, struct dentry *dentry)
3464 mutex_unlock(&path->dentry->d_inode->i_mutex);
3465 mnt_drop_write(path->mnt);
3468 EXPORT_SYMBOL(done_path_create);
3470 struct dentry *user_path_create(int dfd, const char __user *pathname,
3471 struct path *path, unsigned int lookup_flags)
3473 struct filename *tmp = getname(pathname);
3476 return ERR_CAST(tmp);
3477 res = filename_create(dfd, tmp, path, lookup_flags);
3481 EXPORT_SYMBOL(user_path_create);
3483 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3485 int error = may_create(dir, dentry);
3490 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3493 if (!dir->i_op->mknod)
3496 error = devcgroup_inode_mknod(mode, dev);
3500 error = security_inode_mknod(dir, dentry, mode, dev);
3504 error = dir->i_op->mknod(dir, dentry, mode, dev);
3506 fsnotify_create(dir, dentry);
3509 EXPORT_SYMBOL(vfs_mknod);
3511 static int may_mknod(umode_t mode)
3513 switch (mode & S_IFMT) {
3519 case 0: /* zero mode translates to S_IFREG */
3528 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3531 struct dentry *dentry;
3534 unsigned int lookup_flags = 0;
3536 error = may_mknod(mode);
3540 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3542 return PTR_ERR(dentry);
3544 if (!IS_POSIXACL(path.dentry->d_inode))
3545 mode &= ~current_umask();
3546 error = security_path_mknod(&path, dentry, mode, dev);
3549 switch (mode & S_IFMT) {
3550 case 0: case S_IFREG:
3551 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3553 case S_IFCHR: case S_IFBLK:
3554 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3555 new_decode_dev(dev));
3557 case S_IFIFO: case S_IFSOCK:
3558 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3562 done_path_create(&path, dentry);
3563 if (retry_estale(error, lookup_flags)) {
3564 lookup_flags |= LOOKUP_REVAL;
3570 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3572 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3575 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3577 int error = may_create(dir, dentry);
3578 unsigned max_links = dir->i_sb->s_max_links;
3583 if (!dir->i_op->mkdir)
3586 mode &= (S_IRWXUGO|S_ISVTX);
3587 error = security_inode_mkdir(dir, dentry, mode);
3591 if (max_links && dir->i_nlink >= max_links)
3594 error = dir->i_op->mkdir(dir, dentry, mode);
3596 fsnotify_mkdir(dir, dentry);
3599 EXPORT_SYMBOL(vfs_mkdir);
3601 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3603 struct dentry *dentry;
3606 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3609 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3611 return PTR_ERR(dentry);
3613 if (!IS_POSIXACL(path.dentry->d_inode))
3614 mode &= ~current_umask();
3615 error = security_path_mkdir(&path, dentry, mode);
3617 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3618 done_path_create(&path, dentry);
3619 if (retry_estale(error, lookup_flags)) {
3620 lookup_flags |= LOOKUP_REVAL;
3626 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3628 return sys_mkdirat(AT_FDCWD, pathname, mode);
3632 * The dentry_unhash() helper will try to drop the dentry early: we
3633 * should have a usage count of 1 if we're the only user of this
3634 * dentry, and if that is true (possibly after pruning the dcache),
3635 * then we drop the dentry now.
3637 * A low-level filesystem can, if it choses, legally
3640 * if (!d_unhashed(dentry))
3643 * if it cannot handle the case of removing a directory
3644 * that is still in use by something else..
3646 void dentry_unhash(struct dentry *dentry)
3648 shrink_dcache_parent(dentry);
3649 spin_lock(&dentry->d_lock);
3650 if (dentry->d_lockref.count == 1)
3652 spin_unlock(&dentry->d_lock);
3654 EXPORT_SYMBOL(dentry_unhash);
3656 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3658 int error = may_delete(dir, dentry, 1);
3663 if (!dir->i_op->rmdir)
3667 mutex_lock(&dentry->d_inode->i_mutex);
3670 if (is_local_mountpoint(dentry))
3673 error = security_inode_rmdir(dir, dentry);
3677 shrink_dcache_parent(dentry);
3678 error = dir->i_op->rmdir(dir, dentry);
3682 dentry->d_inode->i_flags |= S_DEAD;
3684 detach_mounts(dentry);
3687 mutex_unlock(&dentry->d_inode->i_mutex);
3693 EXPORT_SYMBOL(vfs_rmdir);
3695 static long do_rmdir(int dfd, const char __user *pathname)
3698 struct filename *name;
3699 struct dentry *dentry;
3703 unsigned int lookup_flags = 0;
3705 name = user_path_parent(dfd, pathname,
3706 &path, &last, &type, lookup_flags);
3708 return PTR_ERR(name);
3722 error = mnt_want_write(path.mnt);
3726 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3727 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3728 error = PTR_ERR(dentry);
3731 if (!dentry->d_inode) {
3735 error = security_path_rmdir(&path, dentry);
3738 error = vfs_rmdir(path.dentry->d_inode, dentry);
3742 mutex_unlock(&path.dentry->d_inode->i_mutex);
3743 mnt_drop_write(path.mnt);
3747 if (retry_estale(error, lookup_flags)) {
3748 lookup_flags |= LOOKUP_REVAL;
3754 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3756 return do_rmdir(AT_FDCWD, pathname);
3760 * vfs_unlink - unlink a filesystem object
3761 * @dir: parent directory
3763 * @delegated_inode: returns victim inode, if the inode is delegated.
3765 * The caller must hold dir->i_mutex.
3767 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3768 * return a reference to the inode in delegated_inode. The caller
3769 * should then break the delegation on that inode and retry. Because
3770 * breaking a delegation may take a long time, the caller should drop
3771 * dir->i_mutex before doing so.
3773 * Alternatively, a caller may pass NULL for delegated_inode. This may
3774 * be appropriate for callers that expect the underlying filesystem not
3775 * to be NFS exported.
3777 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3779 struct inode *target = dentry->d_inode;
3780 int error = may_delete(dir, dentry, 0);
3785 if (!dir->i_op->unlink)
3788 mutex_lock(&target->i_mutex);
3789 if (is_local_mountpoint(dentry))
3792 error = security_inode_unlink(dir, dentry);
3794 error = try_break_deleg(target, delegated_inode);
3797 error = dir->i_op->unlink(dir, dentry);
3800 detach_mounts(dentry);
3805 mutex_unlock(&target->i_mutex);
3807 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3808 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3809 fsnotify_link_count(target);
3815 EXPORT_SYMBOL(vfs_unlink);
3818 * Make sure that the actual truncation of the file will occur outside its
3819 * directory's i_mutex. Truncate can take a long time if there is a lot of
3820 * writeout happening, and we don't want to prevent access to the directory
3821 * while waiting on the I/O.
3823 static long do_unlinkat(int dfd, const char __user *pathname)
3826 struct filename *name;
3827 struct dentry *dentry;
3831 struct inode *inode = NULL;
3832 struct inode *delegated_inode = NULL;
3833 unsigned int lookup_flags = 0;
3835 name = user_path_parent(dfd, pathname,
3836 &path, &last, &type, lookup_flags);
3838 return PTR_ERR(name);
3841 if (type != LAST_NORM)
3844 error = mnt_want_write(path.mnt);
3848 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3849 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3850 error = PTR_ERR(dentry);
3851 if (!IS_ERR(dentry)) {
3852 /* Why not before? Because we want correct error value */
3853 if (last.name[last.len])
3855 inode = dentry->d_inode;
3856 if (d_is_negative(dentry))
3859 error = security_path_unlink(&path, dentry);
3862 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3866 mutex_unlock(&path.dentry->d_inode->i_mutex);
3868 iput(inode); /* truncate the inode here */
3870 if (delegated_inode) {
3871 error = break_deleg_wait(&delegated_inode);
3875 mnt_drop_write(path.mnt);
3879 if (retry_estale(error, lookup_flags)) {
3880 lookup_flags |= LOOKUP_REVAL;
3887 if (d_is_negative(dentry))
3889 else if (d_is_dir(dentry))
3896 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3898 if ((flag & ~AT_REMOVEDIR) != 0)
3901 if (flag & AT_REMOVEDIR)
3902 return do_rmdir(dfd, pathname);
3904 return do_unlinkat(dfd, pathname);
3907 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3909 return do_unlinkat(AT_FDCWD, pathname);
3912 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3914 int error = may_create(dir, dentry);
3919 if (!dir->i_op->symlink)
3922 error = security_inode_symlink(dir, dentry, oldname);
3926 error = dir->i_op->symlink(dir, dentry, oldname);
3928 fsnotify_create(dir, dentry);
3931 EXPORT_SYMBOL(vfs_symlink);
3933 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3934 int, newdfd, const char __user *, newname)
3937 struct filename *from;
3938 struct dentry *dentry;
3940 unsigned int lookup_flags = 0;
3942 from = getname(oldname);
3944 return PTR_ERR(from);
3946 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3947 error = PTR_ERR(dentry);
3951 error = security_path_symlink(&path, dentry, from->name);
3953 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3954 done_path_create(&path, dentry);
3955 if (retry_estale(error, lookup_flags)) {
3956 lookup_flags |= LOOKUP_REVAL;
3964 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3966 return sys_symlinkat(oldname, AT_FDCWD, newname);
3970 * vfs_link - create a new link
3971 * @old_dentry: object to be linked
3973 * @new_dentry: where to create the new link
3974 * @delegated_inode: returns inode needing a delegation break
3976 * The caller must hold dir->i_mutex
3978 * If vfs_link discovers a delegation on the to-be-linked file in need
3979 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3980 * inode in delegated_inode. The caller should then break the delegation
3981 * and retry. Because breaking a delegation may take a long time, the
3982 * caller should drop the i_mutex before doing so.
3984 * Alternatively, a caller may pass NULL for delegated_inode. This may
3985 * be appropriate for callers that expect the underlying filesystem not
3986 * to be NFS exported.
3988 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3990 struct inode *inode = old_dentry->d_inode;
3991 unsigned max_links = dir->i_sb->s_max_links;
3997 error = may_create(dir, new_dentry);
4001 if (dir->i_sb != inode->i_sb)
4005 * A link to an append-only or immutable file cannot be created.
4007 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4009 if (!dir->i_op->link)
4011 if (S_ISDIR(inode->i_mode))
4014 error = security_inode_link(old_dentry, dir, new_dentry);
4018 mutex_lock(&inode->i_mutex);
4019 /* Make sure we don't allow creating hardlink to an unlinked file */
4020 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4022 else if (max_links && inode->i_nlink >= max_links)
4025 error = try_break_deleg(inode, delegated_inode);
4027 error = dir->i_op->link(old_dentry, dir, new_dentry);
4030 if (!error && (inode->i_state & I_LINKABLE)) {
4031 spin_lock(&inode->i_lock);
4032 inode->i_state &= ~I_LINKABLE;
4033 spin_unlock(&inode->i_lock);
4035 mutex_unlock(&inode->i_mutex);
4037 fsnotify_link(dir, inode, new_dentry);
4040 EXPORT_SYMBOL(vfs_link);
4043 * Hardlinks are often used in delicate situations. We avoid
4044 * security-related surprises by not following symlinks on the
4047 * We don't follow them on the oldname either to be compatible
4048 * with linux 2.0, and to avoid hard-linking to directories
4049 * and other special files. --ADM
4051 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4052 int, newdfd, const char __user *, newname, int, flags)
4054 struct dentry *new_dentry;
4055 struct path old_path, new_path;
4056 struct inode *delegated_inode = NULL;
4060 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4063 * To use null names we require CAP_DAC_READ_SEARCH
4064 * This ensures that not everyone will be able to create
4065 * handlink using the passed filedescriptor.
4067 if (flags & AT_EMPTY_PATH) {
4068 if (!capable(CAP_DAC_READ_SEARCH))
4073 if (flags & AT_SYMLINK_FOLLOW)
4074 how |= LOOKUP_FOLLOW;
4076 error = user_path_at(olddfd, oldname, how, &old_path);
4080 new_dentry = user_path_create(newdfd, newname, &new_path,
4081 (how & LOOKUP_REVAL));
4082 error = PTR_ERR(new_dentry);
4083 if (IS_ERR(new_dentry))
4087 if (old_path.mnt != new_path.mnt)
4089 error = may_linkat(&old_path);
4090 if (unlikely(error))
4092 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4095 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4097 done_path_create(&new_path, new_dentry);
4098 if (delegated_inode) {
4099 error = break_deleg_wait(&delegated_inode);
4101 path_put(&old_path);
4105 if (retry_estale(error, how)) {
4106 path_put(&old_path);
4107 how |= LOOKUP_REVAL;
4111 path_put(&old_path);
4116 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4118 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4122 * vfs_rename - rename a filesystem object
4123 * @old_dir: parent of source
4124 * @old_dentry: source
4125 * @new_dir: parent of destination
4126 * @new_dentry: destination
4127 * @delegated_inode: returns an inode needing a delegation break
4128 * @flags: rename flags
4130 * The caller must hold multiple mutexes--see lock_rename()).
4132 * If vfs_rename discovers a delegation in need of breaking at either
4133 * the source or destination, it will return -EWOULDBLOCK and return a
4134 * reference to the inode in delegated_inode. The caller should then
4135 * break the delegation and retry. Because breaking a delegation may
4136 * take a long time, the caller should drop all locks before doing
4139 * Alternatively, a caller may pass NULL for delegated_inode. This may
4140 * be appropriate for callers that expect the underlying filesystem not
4141 * to be NFS exported.
4143 * The worst of all namespace operations - renaming directory. "Perverted"
4144 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4146 * a) we can get into loop creation.
4147 * b) race potential - two innocent renames can create a loop together.
4148 * That's where 4.4 screws up. Current fix: serialization on
4149 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4151 * c) we have to lock _four_ objects - parents and victim (if it exists),
4152 * and source (if it is not a directory).
4153 * And that - after we got ->i_mutex on parents (until then we don't know
4154 * whether the target exists). Solution: try to be smart with locking
4155 * order for inodes. We rely on the fact that tree topology may change
4156 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4157 * move will be locked. Thus we can rank directories by the tree
4158 * (ancestors first) and rank all non-directories after them.
4159 * That works since everybody except rename does "lock parent, lookup,
4160 * lock child" and rename is under ->s_vfs_rename_mutex.
4161 * HOWEVER, it relies on the assumption that any object with ->lookup()
4162 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4163 * we'd better make sure that there's no link(2) for them.
4164 * d) conversion from fhandle to dentry may come in the wrong moment - when
4165 * we are removing the target. Solution: we will have to grab ->i_mutex
4166 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4167 * ->i_mutex on parents, which works but leads to some truly excessive
4170 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4171 struct inode *new_dir, struct dentry *new_dentry,
4172 struct inode **delegated_inode, unsigned int flags)
4175 bool is_dir = d_is_dir(old_dentry);
4176 const unsigned char *old_name;
4177 struct inode *source = old_dentry->d_inode;
4178 struct inode *target = new_dentry->d_inode;
4179 bool new_is_dir = false;
4180 unsigned max_links = new_dir->i_sb->s_max_links;
4182 if (source == target)
4185 error = may_delete(old_dir, old_dentry, is_dir);
4190 error = may_create(new_dir, new_dentry);
4192 new_is_dir = d_is_dir(new_dentry);
4194 if (!(flags & RENAME_EXCHANGE))
4195 error = may_delete(new_dir, new_dentry, is_dir);
4197 error = may_delete(new_dir, new_dentry, new_is_dir);
4202 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4205 if (flags && !old_dir->i_op->rename2)
4209 * If we are going to change the parent - check write permissions,
4210 * we'll need to flip '..'.
4212 if (new_dir != old_dir) {
4214 error = inode_permission(source, MAY_WRITE);
4218 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4219 error = inode_permission(target, MAY_WRITE);
4225 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4230 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4232 if (!is_dir || (flags & RENAME_EXCHANGE))
4233 lock_two_nondirectories(source, target);
4235 mutex_lock(&target->i_mutex);
4238 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4241 if (max_links && new_dir != old_dir) {
4243 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4245 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4246 old_dir->i_nlink >= max_links)
4249 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4250 shrink_dcache_parent(new_dentry);
4252 error = try_break_deleg(source, delegated_inode);
4256 if (target && !new_is_dir) {
4257 error = try_break_deleg(target, delegated_inode);
4261 if (!old_dir->i_op->rename2) {
4262 error = old_dir->i_op->rename(old_dir, old_dentry,
4263 new_dir, new_dentry);
4265 WARN_ON(old_dir->i_op->rename != NULL);
4266 error = old_dir->i_op->rename2(old_dir, old_dentry,
4267 new_dir, new_dentry, flags);
4272 if (!(flags & RENAME_EXCHANGE) && target) {
4274 target->i_flags |= S_DEAD;
4275 dont_mount(new_dentry);
4276 detach_mounts(new_dentry);
4278 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4279 if (!(flags & RENAME_EXCHANGE))
4280 d_move(old_dentry, new_dentry);
4282 d_exchange(old_dentry, new_dentry);
4285 if (!is_dir || (flags & RENAME_EXCHANGE))
4286 unlock_two_nondirectories(source, target);
4288 mutex_unlock(&target->i_mutex);
4291 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4292 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4293 if (flags & RENAME_EXCHANGE) {
4294 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4295 new_is_dir, NULL, new_dentry);
4298 fsnotify_oldname_free(old_name);
4302 EXPORT_SYMBOL(vfs_rename);
4304 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4305 int, newdfd, const char __user *, newname, unsigned int, flags)
4307 struct dentry *old_dentry, *new_dentry;
4308 struct dentry *trap;
4309 struct path old_path, new_path;
4310 struct qstr old_last, new_last;
4311 int old_type, new_type;
4312 struct inode *delegated_inode = NULL;
4313 struct filename *from;
4314 struct filename *to;
4315 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4316 bool should_retry = false;
4319 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4322 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4323 (flags & RENAME_EXCHANGE))
4326 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4329 if (flags & RENAME_EXCHANGE)
4333 from = user_path_parent(olddfd, oldname,
4334 &old_path, &old_last, &old_type, lookup_flags);
4336 error = PTR_ERR(from);
4340 to = user_path_parent(newdfd, newname,
4341 &new_path, &new_last, &new_type, lookup_flags);
4343 error = PTR_ERR(to);
4348 if (old_path.mnt != new_path.mnt)
4352 if (old_type != LAST_NORM)
4355 if (flags & RENAME_NOREPLACE)
4357 if (new_type != LAST_NORM)
4360 error = mnt_want_write(old_path.mnt);
4365 trap = lock_rename(new_path.dentry, old_path.dentry);
4367 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4368 error = PTR_ERR(old_dentry);
4369 if (IS_ERR(old_dentry))
4371 /* source must exist */
4373 if (d_is_negative(old_dentry))
4375 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4376 error = PTR_ERR(new_dentry);
4377 if (IS_ERR(new_dentry))
4380 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4382 if (flags & RENAME_EXCHANGE) {
4384 if (d_is_negative(new_dentry))
4387 if (!d_is_dir(new_dentry)) {
4389 if (new_last.name[new_last.len])
4393 /* unless the source is a directory trailing slashes give -ENOTDIR */
4394 if (!d_is_dir(old_dentry)) {
4396 if (old_last.name[old_last.len])
4398 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4401 /* source should not be ancestor of target */
4403 if (old_dentry == trap)
4405 /* target should not be an ancestor of source */
4406 if (!(flags & RENAME_EXCHANGE))
4408 if (new_dentry == trap)
4411 error = security_path_rename(&old_path, old_dentry,
4412 &new_path, new_dentry, flags);
4415 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4416 new_path.dentry->d_inode, new_dentry,
4417 &delegated_inode, flags);
4423 unlock_rename(new_path.dentry, old_path.dentry);
4424 if (delegated_inode) {
4425 error = break_deleg_wait(&delegated_inode);
4429 mnt_drop_write(old_path.mnt);
4431 if (retry_estale(error, lookup_flags))
4432 should_retry = true;
4433 path_put(&new_path);
4436 path_put(&old_path);
4439 should_retry = false;
4440 lookup_flags |= LOOKUP_REVAL;
4447 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4448 int, newdfd, const char __user *, newname)
4450 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4453 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4455 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4458 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4460 int error = may_create(dir, dentry);
4464 if (!dir->i_op->mknod)
4467 return dir->i_op->mknod(dir, dentry,
4468 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4470 EXPORT_SYMBOL(vfs_whiteout);
4472 int readlink_copy(char __user *buffer, int buflen, const char *link)
4474 int len = PTR_ERR(link);
4479 if (len > (unsigned) buflen)
4481 if (copy_to_user(buffer, link, len))
4486 EXPORT_SYMBOL(readlink_copy);
4489 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4490 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4491 * using) it for any given inode is up to filesystem.
4493 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4496 const char *link = dentry->d_inode->i_link;
4500 link = dentry->d_inode->i_op->follow_link(dentry, &cookie, NULL);
4502 return PTR_ERR(link);
4504 res = readlink_copy(buffer, buflen, link);
4505 if (cookie && dentry->d_inode->i_op->put_link)
4506 dentry->d_inode->i_op->put_link(dentry, cookie);
4509 EXPORT_SYMBOL(generic_readlink);
4511 /* get the link contents into pagecache */
4512 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4516 struct address_space *mapping = dentry->d_inode->i_mapping;
4517 page = read_mapping_page(mapping, 0, NULL);
4522 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4526 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4528 struct page *page = NULL;
4529 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4532 page_cache_release(page);
4536 EXPORT_SYMBOL(page_readlink);
4538 const char *page_follow_link_light(struct dentry *dentry, void **cookie, struct nameidata *nd)
4540 struct page *page = NULL;
4541 char *res = page_getlink(dentry, &page);
4546 EXPORT_SYMBOL(page_follow_link_light);
4548 void page_put_link(struct dentry *dentry, void *cookie)
4550 struct page *page = cookie;
4552 page_cache_release(page);
4554 EXPORT_SYMBOL(page_put_link);
4557 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4559 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4561 struct address_space *mapping = inode->i_mapping;
4566 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4568 flags |= AOP_FLAG_NOFS;
4571 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4572 flags, &page, &fsdata);
4576 kaddr = kmap_atomic(page);
4577 memcpy(kaddr, symname, len-1);
4578 kunmap_atomic(kaddr);
4580 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4587 mark_inode_dirty(inode);
4592 EXPORT_SYMBOL(__page_symlink);
4594 int page_symlink(struct inode *inode, const char *symname, int len)
4596 return __page_symlink(inode, symname, len,
4597 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4599 EXPORT_SYMBOL(page_symlink);
4601 const struct inode_operations page_symlink_inode_operations = {
4602 .readlink = generic_readlink,
4603 .follow_link = page_follow_link_light,
4604 .put_link = page_put_link,
4606 EXPORT_SYMBOL(page_symlink_inode_operations);