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);
502 struct inode *inode; /* path.dentry.d_inode */
512 } stack[MAX_NESTED_LINKS + 1];
516 * Path walking has 2 modes, rcu-walk and ref-walk (see
517 * Documentation/filesystems/path-lookup.txt). In situations when we can't
518 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
519 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
520 * mode. Refcounts are grabbed at the last known good point before rcu-walk
521 * got stuck, so ref-walk may continue from there. If this is not successful
522 * (eg. a seqcount has changed), then failure is returned and it's up to caller
523 * to restart the path walk from the beginning in ref-walk mode.
527 * unlazy_walk - try to switch to ref-walk mode.
528 * @nd: nameidata pathwalk data
529 * @dentry: child of nd->path.dentry or NULL
530 * Returns: 0 on success, -ECHILD on failure
532 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
533 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
534 * @nd or NULL. Must be called from rcu-walk context.
536 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
538 struct fs_struct *fs = current->fs;
539 struct dentry *parent = nd->path.dentry;
541 BUG_ON(!(nd->flags & LOOKUP_RCU));
544 * After legitimizing the bastards, terminate_walk()
545 * will do the right thing for non-RCU mode, and all our
546 * subsequent exit cases should rcu_read_unlock()
547 * before returning. Do vfsmount first; if dentry
548 * can't be legitimized, just set nd->path.dentry to NULL
549 * and rely on dput(NULL) being a no-op.
551 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
553 nd->flags &= ~LOOKUP_RCU;
555 if (!lockref_get_not_dead(&parent->d_lockref)) {
556 nd->path.dentry = NULL;
561 * For a negative lookup, the lookup sequence point is the parents
562 * sequence point, and it only needs to revalidate the parent dentry.
564 * For a positive lookup, we need to move both the parent and the
565 * dentry from the RCU domain to be properly refcounted. And the
566 * sequence number in the dentry validates *both* dentry counters,
567 * since we checked the sequence number of the parent after we got
568 * the child sequence number. So we know the parent must still
569 * be valid if the child sequence number is still valid.
572 if (read_seqcount_retry(&parent->d_seq, nd->seq))
574 BUG_ON(nd->inode != parent->d_inode);
576 if (!lockref_get_not_dead(&dentry->d_lockref))
578 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
583 * Sequence counts matched. Now make sure that the root is
584 * still valid and get it if required.
586 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
587 spin_lock(&fs->lock);
588 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
589 goto unlock_and_drop_dentry;
591 spin_unlock(&fs->lock);
597 unlock_and_drop_dentry:
598 spin_unlock(&fs->lock);
606 if (!(nd->flags & LOOKUP_ROOT))
611 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
613 return dentry->d_op->d_revalidate(dentry, flags);
617 * complete_walk - successful completion of path walk
618 * @nd: pointer nameidata
620 * If we had been in RCU mode, drop out of it and legitimize nd->path.
621 * Revalidate the final result, unless we'd already done that during
622 * the path walk or the filesystem doesn't ask for it. Return 0 on
623 * success, -error on failure. In case of failure caller does not
624 * need to drop nd->path.
626 static int complete_walk(struct nameidata *nd)
628 struct dentry *dentry = nd->path.dentry;
631 if (nd->flags & LOOKUP_RCU) {
632 nd->flags &= ~LOOKUP_RCU;
633 if (!(nd->flags & LOOKUP_ROOT))
636 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
640 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
642 mntput(nd->path.mnt);
645 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
648 mntput(nd->path.mnt);
654 if (likely(!(nd->flags & LOOKUP_JUMPED)))
657 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
660 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
671 static __always_inline void set_root(struct nameidata *nd)
673 get_fs_root(current->fs, &nd->root);
676 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
678 struct fs_struct *fs = current->fs;
682 seq = read_seqcount_begin(&fs->seq);
684 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
685 } while (read_seqcount_retry(&fs->seq, seq));
689 static void path_put_conditional(struct path *path, struct nameidata *nd)
692 if (path->mnt != nd->path.mnt)
696 static inline void path_to_nameidata(const struct path *path,
697 struct nameidata *nd)
699 if (!(nd->flags & LOOKUP_RCU)) {
700 dput(nd->path.dentry);
701 if (nd->path.mnt != path->mnt)
702 mntput(nd->path.mnt);
704 nd->path.mnt = path->mnt;
705 nd->path.dentry = path->dentry;
709 * Helper to directly jump to a known parsed path from ->follow_link,
710 * caller must have taken a reference to path beforehand.
712 void nd_jump_link(struct nameidata *nd, struct path *path)
717 nd->inode = nd->path.dentry->d_inode;
718 nd->flags |= LOOKUP_JUMPED;
721 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
723 struct inode *inode = link->dentry->d_inode;
724 if (cookie && inode->i_op->put_link)
725 inode->i_op->put_link(link->dentry, cookie);
729 int sysctl_protected_symlinks __read_mostly = 0;
730 int sysctl_protected_hardlinks __read_mostly = 0;
733 * may_follow_link - Check symlink following for unsafe situations
734 * @link: The path of the symlink
735 * @nd: nameidata pathwalk data
737 * In the case of the sysctl_protected_symlinks sysctl being enabled,
738 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
739 * in a sticky world-writable directory. This is to protect privileged
740 * processes from failing races against path names that may change out
741 * from under them by way of other users creating malicious symlinks.
742 * It will permit symlinks to be followed only when outside a sticky
743 * world-writable directory, or when the uid of the symlink and follower
744 * match, or when the directory owner matches the symlink's owner.
746 * Returns 0 if following the symlink is allowed, -ve on error.
748 static inline int may_follow_link(struct path *link, struct nameidata *nd)
750 const struct inode *inode;
751 const struct inode *parent;
753 if (!sysctl_protected_symlinks)
756 /* Allowed if owner and follower match. */
757 inode = link->dentry->d_inode;
758 if (uid_eq(current_cred()->fsuid, inode->i_uid))
761 /* Allowed if parent directory not sticky and world-writable. */
762 parent = nd->path.dentry->d_inode;
763 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
766 /* Allowed if parent directory and link owner match. */
767 if (uid_eq(parent->i_uid, inode->i_uid))
770 audit_log_link_denied("follow_link", link);
771 path_put_conditional(link, nd);
777 * safe_hardlink_source - Check for safe hardlink conditions
778 * @inode: the source inode to hardlink from
780 * Return false if at least one of the following conditions:
781 * - inode is not a regular file
783 * - inode is setgid and group-exec
784 * - access failure for read and write
786 * Otherwise returns true.
788 static bool safe_hardlink_source(struct inode *inode)
790 umode_t mode = inode->i_mode;
792 /* Special files should not get pinned to the filesystem. */
796 /* Setuid files should not get pinned to the filesystem. */
800 /* Executable setgid files should not get pinned to the filesystem. */
801 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
804 /* Hardlinking to unreadable or unwritable sources is dangerous. */
805 if (inode_permission(inode, MAY_READ | MAY_WRITE))
812 * may_linkat - Check permissions for creating a hardlink
813 * @link: the source to hardlink from
815 * Block hardlink when all of:
816 * - sysctl_protected_hardlinks enabled
817 * - fsuid does not match inode
818 * - hardlink source is unsafe (see safe_hardlink_source() above)
821 * Returns 0 if successful, -ve on error.
823 static int may_linkat(struct path *link)
825 const struct cred *cred;
828 if (!sysctl_protected_hardlinks)
831 cred = current_cred();
832 inode = link->dentry->d_inode;
834 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
835 * otherwise, it must be a safe source.
837 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
841 audit_log_link_denied("linkat", link);
845 static __always_inline const char *
846 get_link(struct path *link, struct nameidata *nd, void **p)
848 struct dentry *dentry = link->dentry;
849 struct inode *inode = dentry->d_inode;
853 BUG_ON(nd->flags & LOOKUP_RCU);
855 if (link->mnt == nd->path.mnt)
858 res = ERR_PTR(-ELOOP);
859 if (unlikely(current->total_link_count >= 40))
863 current->total_link_count++;
867 error = security_inode_follow_link(dentry);
868 res = ERR_PTR(error);
872 nd->last_type = LAST_BIND;
876 res = inode->i_op->follow_link(dentry, p, nd);
886 static int follow_up_rcu(struct path *path)
888 struct mount *mnt = real_mount(path->mnt);
889 struct mount *parent;
890 struct dentry *mountpoint;
892 parent = mnt->mnt_parent;
893 if (&parent->mnt == path->mnt)
895 mountpoint = mnt->mnt_mountpoint;
896 path->dentry = mountpoint;
897 path->mnt = &parent->mnt;
902 * follow_up - Find the mountpoint of path's vfsmount
904 * Given a path, find the mountpoint of its source file system.
905 * Replace @path with the path of the mountpoint in the parent mount.
908 * Return 1 if we went up a level and 0 if we were already at the
911 int follow_up(struct path *path)
913 struct mount *mnt = real_mount(path->mnt);
914 struct mount *parent;
915 struct dentry *mountpoint;
917 read_seqlock_excl(&mount_lock);
918 parent = mnt->mnt_parent;
920 read_sequnlock_excl(&mount_lock);
923 mntget(&parent->mnt);
924 mountpoint = dget(mnt->mnt_mountpoint);
925 read_sequnlock_excl(&mount_lock);
927 path->dentry = mountpoint;
929 path->mnt = &parent->mnt;
932 EXPORT_SYMBOL(follow_up);
935 * Perform an automount
936 * - return -EISDIR to tell follow_managed() to stop and return the path we
939 static int follow_automount(struct path *path, unsigned flags,
942 struct vfsmount *mnt;
945 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
948 /* We don't want to mount if someone's just doing a stat -
949 * unless they're stat'ing a directory and appended a '/' to
952 * We do, however, want to mount if someone wants to open or
953 * create a file of any type under the mountpoint, wants to
954 * traverse through the mountpoint or wants to open the
955 * mounted directory. Also, autofs may mark negative dentries
956 * as being automount points. These will need the attentions
957 * of the daemon to instantiate them before they can be used.
959 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
960 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
961 path->dentry->d_inode)
964 current->total_link_count++;
965 if (current->total_link_count >= 40)
968 mnt = path->dentry->d_op->d_automount(path);
971 * The filesystem is allowed to return -EISDIR here to indicate
972 * it doesn't want to automount. For instance, autofs would do
973 * this so that its userspace daemon can mount on this dentry.
975 * However, we can only permit this if it's a terminal point in
976 * the path being looked up; if it wasn't then the remainder of
977 * the path is inaccessible and we should say so.
979 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
984 if (!mnt) /* mount collision */
988 /* lock_mount() may release path->mnt on error */
992 err = finish_automount(mnt, path);
996 /* Someone else made a mount here whilst we were busy */
1001 path->dentry = dget(mnt->mnt_root);
1010 * Handle a dentry that is managed in some way.
1011 * - Flagged for transit management (autofs)
1012 * - Flagged as mountpoint
1013 * - Flagged as automount point
1015 * This may only be called in refwalk mode.
1017 * Serialization is taken care of in namespace.c
1019 static int follow_managed(struct path *path, unsigned flags)
1021 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1023 bool need_mntput = false;
1026 /* Given that we're not holding a lock here, we retain the value in a
1027 * local variable for each dentry as we look at it so that we don't see
1028 * the components of that value change under us */
1029 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1030 managed &= DCACHE_MANAGED_DENTRY,
1031 unlikely(managed != 0)) {
1032 /* Allow the filesystem to manage the transit without i_mutex
1034 if (managed & DCACHE_MANAGE_TRANSIT) {
1035 BUG_ON(!path->dentry->d_op);
1036 BUG_ON(!path->dentry->d_op->d_manage);
1037 ret = path->dentry->d_op->d_manage(path->dentry, false);
1042 /* Transit to a mounted filesystem. */
1043 if (managed & DCACHE_MOUNTED) {
1044 struct vfsmount *mounted = lookup_mnt(path);
1049 path->mnt = mounted;
1050 path->dentry = dget(mounted->mnt_root);
1055 /* Something is mounted on this dentry in another
1056 * namespace and/or whatever was mounted there in this
1057 * namespace got unmounted before lookup_mnt() could
1061 /* Handle an automount point */
1062 if (managed & DCACHE_NEED_AUTOMOUNT) {
1063 ret = follow_automount(path, flags, &need_mntput);
1069 /* We didn't change the current path point */
1073 if (need_mntput && path->mnt == mnt)
1077 return ret < 0 ? ret : need_mntput;
1080 int follow_down_one(struct path *path)
1082 struct vfsmount *mounted;
1084 mounted = lookup_mnt(path);
1088 path->mnt = mounted;
1089 path->dentry = dget(mounted->mnt_root);
1094 EXPORT_SYMBOL(follow_down_one);
1096 static inline int managed_dentry_rcu(struct dentry *dentry)
1098 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1099 dentry->d_op->d_manage(dentry, true) : 0;
1103 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1104 * we meet a managed dentry that would need blocking.
1106 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1107 struct inode **inode)
1110 struct mount *mounted;
1112 * Don't forget we might have a non-mountpoint managed dentry
1113 * that wants to block transit.
1115 switch (managed_dentry_rcu(path->dentry)) {
1125 if (!d_mountpoint(path->dentry))
1126 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1128 mounted = __lookup_mnt(path->mnt, path->dentry);
1131 path->mnt = &mounted->mnt;
1132 path->dentry = mounted->mnt.mnt_root;
1133 nd->flags |= LOOKUP_JUMPED;
1134 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1136 * Update the inode too. We don't need to re-check the
1137 * dentry sequence number here after this d_inode read,
1138 * because a mount-point is always pinned.
1140 *inode = path->dentry->d_inode;
1142 return !read_seqretry(&mount_lock, nd->m_seq) &&
1143 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1146 static int follow_dotdot_rcu(struct nameidata *nd)
1148 struct inode *inode = nd->inode;
1153 if (nd->path.dentry == nd->root.dentry &&
1154 nd->path.mnt == nd->root.mnt) {
1157 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1158 struct dentry *old = nd->path.dentry;
1159 struct dentry *parent = old->d_parent;
1162 inode = parent->d_inode;
1163 seq = read_seqcount_begin(&parent->d_seq);
1164 if (read_seqcount_retry(&old->d_seq, nd->seq))
1166 nd->path.dentry = parent;
1170 if (!follow_up_rcu(&nd->path))
1172 inode = nd->path.dentry->d_inode;
1173 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1175 while (d_mountpoint(nd->path.dentry)) {
1176 struct mount *mounted;
1177 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1180 nd->path.mnt = &mounted->mnt;
1181 nd->path.dentry = mounted->mnt.mnt_root;
1182 inode = nd->path.dentry->d_inode;
1183 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1184 if (read_seqretry(&mount_lock, nd->m_seq))
1191 nd->flags &= ~LOOKUP_RCU;
1192 if (!(nd->flags & LOOKUP_ROOT))
1193 nd->root.mnt = NULL;
1199 * Follow down to the covering mount currently visible to userspace. At each
1200 * point, the filesystem owning that dentry may be queried as to whether the
1201 * caller is permitted to proceed or not.
1203 int follow_down(struct path *path)
1208 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1209 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1210 /* Allow the filesystem to manage the transit without i_mutex
1213 * We indicate to the filesystem if someone is trying to mount
1214 * something here. This gives autofs the chance to deny anyone
1215 * other than its daemon the right to mount on its
1218 * The filesystem may sleep at this point.
1220 if (managed & DCACHE_MANAGE_TRANSIT) {
1221 BUG_ON(!path->dentry->d_op);
1222 BUG_ON(!path->dentry->d_op->d_manage);
1223 ret = path->dentry->d_op->d_manage(
1224 path->dentry, false);
1226 return ret == -EISDIR ? 0 : ret;
1229 /* Transit to a mounted filesystem. */
1230 if (managed & DCACHE_MOUNTED) {
1231 struct vfsmount *mounted = lookup_mnt(path);
1236 path->mnt = mounted;
1237 path->dentry = dget(mounted->mnt_root);
1241 /* Don't handle automount points here */
1246 EXPORT_SYMBOL(follow_down);
1249 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1251 static void follow_mount(struct path *path)
1253 while (d_mountpoint(path->dentry)) {
1254 struct vfsmount *mounted = lookup_mnt(path);
1259 path->mnt = mounted;
1260 path->dentry = dget(mounted->mnt_root);
1264 static void follow_dotdot(struct nameidata *nd)
1270 struct dentry *old = nd->path.dentry;
1272 if (nd->path.dentry == nd->root.dentry &&
1273 nd->path.mnt == nd->root.mnt) {
1276 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1277 /* rare case of legitimate dget_parent()... */
1278 nd->path.dentry = dget_parent(nd->path.dentry);
1282 if (!follow_up(&nd->path))
1285 follow_mount(&nd->path);
1286 nd->inode = nd->path.dentry->d_inode;
1290 * This looks up the name in dcache, possibly revalidates the old dentry and
1291 * allocates a new one if not found or not valid. In the need_lookup argument
1292 * returns whether i_op->lookup is necessary.
1294 * dir->d_inode->i_mutex must be held
1296 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1297 unsigned int flags, bool *need_lookup)
1299 struct dentry *dentry;
1302 *need_lookup = false;
1303 dentry = d_lookup(dir, name);
1305 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1306 error = d_revalidate(dentry, flags);
1307 if (unlikely(error <= 0)) {
1310 return ERR_PTR(error);
1312 d_invalidate(dentry);
1321 dentry = d_alloc(dir, name);
1322 if (unlikely(!dentry))
1323 return ERR_PTR(-ENOMEM);
1325 *need_lookup = true;
1331 * Call i_op->lookup on the dentry. The dentry must be negative and
1334 * dir->d_inode->i_mutex must be held
1336 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1341 /* Don't create child dentry for a dead directory. */
1342 if (unlikely(IS_DEADDIR(dir))) {
1344 return ERR_PTR(-ENOENT);
1347 old = dir->i_op->lookup(dir, dentry, flags);
1348 if (unlikely(old)) {
1355 static struct dentry *__lookup_hash(struct qstr *name,
1356 struct dentry *base, unsigned int flags)
1359 struct dentry *dentry;
1361 dentry = lookup_dcache(name, base, flags, &need_lookup);
1365 return lookup_real(base->d_inode, dentry, flags);
1369 * It's more convoluted than I'd like it to be, but... it's still fairly
1370 * small and for now I'd prefer to have fast path as straight as possible.
1371 * It _is_ time-critical.
1373 static int lookup_fast(struct nameidata *nd,
1374 struct path *path, struct inode **inode)
1376 struct vfsmount *mnt = nd->path.mnt;
1377 struct dentry *dentry, *parent = nd->path.dentry;
1383 * Rename seqlock is not required here because in the off chance
1384 * of a false negative due to a concurrent rename, we're going to
1385 * do the non-racy lookup, below.
1387 if (nd->flags & LOOKUP_RCU) {
1390 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1395 * This sequence count validates that the inode matches
1396 * the dentry name information from lookup.
1398 *inode = dentry->d_inode;
1399 negative = d_is_negative(dentry);
1400 if (read_seqcount_retry(&dentry->d_seq, seq))
1406 * This sequence count validates that the parent had no
1407 * changes while we did the lookup of the dentry above.
1409 * The memory barrier in read_seqcount_begin of child is
1410 * enough, we can use __read_seqcount_retry here.
1412 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1416 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1417 status = d_revalidate(dentry, nd->flags);
1418 if (unlikely(status <= 0)) {
1419 if (status != -ECHILD)
1425 path->dentry = dentry;
1426 if (likely(__follow_mount_rcu(nd, path, inode)))
1429 if (unlazy_walk(nd, dentry))
1432 dentry = __d_lookup(parent, &nd->last);
1435 if (unlikely(!dentry))
1438 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1439 status = d_revalidate(dentry, nd->flags);
1440 if (unlikely(status <= 0)) {
1445 d_invalidate(dentry);
1450 if (unlikely(d_is_negative(dentry))) {
1455 path->dentry = dentry;
1456 err = follow_managed(path, nd->flags);
1457 if (unlikely(err < 0)) {
1458 path_put_conditional(path, nd);
1462 nd->flags |= LOOKUP_JUMPED;
1463 *inode = path->dentry->d_inode;
1470 /* Fast lookup failed, do it the slow way */
1471 static int lookup_slow(struct nameidata *nd, struct path *path)
1473 struct dentry *dentry, *parent;
1476 parent = nd->path.dentry;
1477 BUG_ON(nd->inode != parent->d_inode);
1479 mutex_lock(&parent->d_inode->i_mutex);
1480 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1481 mutex_unlock(&parent->d_inode->i_mutex);
1483 return PTR_ERR(dentry);
1484 path->mnt = nd->path.mnt;
1485 path->dentry = dentry;
1486 err = follow_managed(path, nd->flags);
1487 if (unlikely(err < 0)) {
1488 path_put_conditional(path, nd);
1492 nd->flags |= LOOKUP_JUMPED;
1496 static inline int may_lookup(struct nameidata *nd)
1498 if (nd->flags & LOOKUP_RCU) {
1499 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1502 if (unlazy_walk(nd, NULL))
1505 return inode_permission(nd->inode, MAY_EXEC);
1508 static inline int handle_dots(struct nameidata *nd, int type)
1510 if (type == LAST_DOTDOT) {
1511 if (nd->flags & LOOKUP_RCU) {
1512 if (follow_dotdot_rcu(nd))
1520 static void terminate_walk(struct nameidata *nd)
1522 if (!(nd->flags & LOOKUP_RCU)) {
1523 path_put(&nd->path);
1525 nd->flags &= ~LOOKUP_RCU;
1526 if (!(nd->flags & LOOKUP_ROOT))
1527 nd->root.mnt = NULL;
1533 * Do we need to follow links? We _really_ want to be able
1534 * to do this check without having to look at inode->i_op,
1535 * so we keep a cache of "no, this doesn't need follow_link"
1536 * for the common case.
1538 static inline int should_follow_link(struct dentry *dentry, int follow)
1540 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1543 static int walk_component(struct nameidata *nd, int follow)
1546 struct inode *inode;
1549 * "." and ".." are special - ".." especially so because it has
1550 * to be able to know about the current root directory and
1551 * parent relationships.
1553 if (unlikely(nd->last_type != LAST_NORM))
1554 return handle_dots(nd, nd->last_type);
1555 err = lookup_fast(nd, &path, &inode);
1556 if (unlikely(err)) {
1560 err = lookup_slow(nd, &path);
1564 inode = path.dentry->d_inode;
1566 if (d_is_negative(path.dentry))
1570 if (should_follow_link(path.dentry, follow)) {
1571 if (nd->flags & LOOKUP_RCU) {
1572 if (unlikely(nd->path.mnt != path.mnt ||
1573 unlazy_walk(nd, path.dentry))) {
1578 BUG_ON(inode != path.dentry->d_inode);
1582 path_to_nameidata(&path, nd);
1587 path_to_nameidata(&path, nd);
1594 * We can do the critical dentry name comparison and hashing
1595 * operations one word at a time, but we are limited to:
1597 * - Architectures with fast unaligned word accesses. We could
1598 * do a "get_unaligned()" if this helps and is sufficiently
1601 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1602 * do not trap on the (extremely unlikely) case of a page
1603 * crossing operation.
1605 * - Furthermore, we need an efficient 64-bit compile for the
1606 * 64-bit case in order to generate the "number of bytes in
1607 * the final mask". Again, that could be replaced with a
1608 * efficient population count instruction or similar.
1610 #ifdef CONFIG_DCACHE_WORD_ACCESS
1612 #include <asm/word-at-a-time.h>
1616 static inline unsigned int fold_hash(unsigned long hash)
1618 return hash_64(hash, 32);
1621 #else /* 32-bit case */
1623 #define fold_hash(x) (x)
1627 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1629 unsigned long a, mask;
1630 unsigned long hash = 0;
1633 a = load_unaligned_zeropad(name);
1634 if (len < sizeof(unsigned long))
1638 name += sizeof(unsigned long);
1639 len -= sizeof(unsigned long);
1643 mask = bytemask_from_count(len);
1646 return fold_hash(hash);
1648 EXPORT_SYMBOL(full_name_hash);
1651 * Calculate the length and hash of the path component, and
1652 * return the "hash_len" as the result.
1654 static inline u64 hash_name(const char *name)
1656 unsigned long a, b, adata, bdata, mask, hash, len;
1657 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1660 len = -sizeof(unsigned long);
1662 hash = (hash + a) * 9;
1663 len += sizeof(unsigned long);
1664 a = load_unaligned_zeropad(name+len);
1665 b = a ^ REPEAT_BYTE('/');
1666 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1668 adata = prep_zero_mask(a, adata, &constants);
1669 bdata = prep_zero_mask(b, bdata, &constants);
1671 mask = create_zero_mask(adata | bdata);
1673 hash += a & zero_bytemask(mask);
1674 len += find_zero(mask);
1675 return hashlen_create(fold_hash(hash), len);
1680 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1682 unsigned long hash = init_name_hash();
1684 hash = partial_name_hash(*name++, hash);
1685 return end_name_hash(hash);
1687 EXPORT_SYMBOL(full_name_hash);
1690 * We know there's a real path component here of at least
1693 static inline u64 hash_name(const char *name)
1695 unsigned long hash = init_name_hash();
1696 unsigned long len = 0, c;
1698 c = (unsigned char)*name;
1701 hash = partial_name_hash(c, hash);
1702 c = (unsigned char)name[len];
1703 } while (c && c != '/');
1704 return hashlen_create(end_name_hash(hash), len);
1711 * This is the basic name resolution function, turning a pathname into
1712 * the final dentry. We expect 'base' to be positive and a directory.
1714 * Returns 0 and nd will have valid dentry and mnt on success.
1715 * Returns error and drops reference to input namei data on failure.
1717 static int link_path_walk(const char *name, struct nameidata *nd)
1719 struct saved *last = nd->stack;
1727 /* At this point we know we have a real path component. */
1732 err = may_lookup(nd);
1736 hash_len = hash_name(name);
1739 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1741 if (name[1] == '.') {
1743 nd->flags |= LOOKUP_JUMPED;
1749 if (likely(type == LAST_NORM)) {
1750 struct dentry *parent = nd->path.dentry;
1751 nd->flags &= ~LOOKUP_JUMPED;
1752 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1753 struct qstr this = { { .hash_len = hash_len }, .name = name };
1754 err = parent->d_op->d_hash(parent, &this);
1757 hash_len = this.hash_len;
1762 nd->last.hash_len = hash_len;
1763 nd->last.name = name;
1764 nd->last_type = type;
1766 name += hashlen_len(hash_len);
1770 * If it wasn't NUL, we know it was '/'. Skip that
1771 * slash, and continue until no more slashes.
1775 } while (unlikely(*name == '/'));
1779 err = walk_component(nd, LOOKUP_FOLLOW);
1787 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1788 path_put_conditional(&nd->link, nd);
1789 path_put(&nd->path);
1793 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1796 current->link_count++;
1799 last->link = nd->link;
1800 s = get_link(&last->link, nd, &last->cookie);
1802 if (unlikely(IS_ERR(s))) {
1804 current->link_count--;
1812 put_link(nd, &last->link, last->cookie);
1813 current->link_count--;
1820 path_put(&nd->path);
1821 nd->path = nd->root;
1822 path_get(&nd->root);
1823 nd->flags |= LOOKUP_JUMPED;
1824 while (unlikely(*++s == '/'))
1827 nd->inode = nd->path.dentry->d_inode;
1835 if (!d_can_lookup(nd->path.dentry)) {
1842 while (unlikely(nd->depth)) {
1843 put_link(nd, &last->link, last->cookie);
1844 current->link_count--;
1850 if (unlikely(nd->depth)) {
1852 err = walk_component(nd, LOOKUP_FOLLOW);
1853 put_link(nd, &last->link, last->cookie);
1854 current->link_count--;
1862 static int path_init(int dfd, const struct filename *name, unsigned int flags,
1863 struct nameidata *nd)
1866 const char *s = name->name;
1868 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1869 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1872 if (flags & LOOKUP_ROOT) {
1873 struct dentry *root = nd->root.dentry;
1874 struct inode *inode = root->d_inode;
1876 if (!d_can_lookup(root))
1878 retval = inode_permission(inode, MAY_EXEC);
1882 nd->path = nd->root;
1884 if (flags & LOOKUP_RCU) {
1886 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1887 nd->m_seq = read_seqbegin(&mount_lock);
1889 path_get(&nd->path);
1894 nd->root.mnt = NULL;
1896 nd->m_seq = read_seqbegin(&mount_lock);
1898 if (flags & LOOKUP_RCU) {
1900 nd->seq = set_root_rcu(nd);
1903 path_get(&nd->root);
1905 nd->path = nd->root;
1906 } else if (dfd == AT_FDCWD) {
1907 if (flags & LOOKUP_RCU) {
1908 struct fs_struct *fs = current->fs;
1914 seq = read_seqcount_begin(&fs->seq);
1916 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1917 } while (read_seqcount_retry(&fs->seq, seq));
1919 get_fs_pwd(current->fs, &nd->path);
1922 /* Caller must check execute permissions on the starting path component */
1923 struct fd f = fdget_raw(dfd);
1924 struct dentry *dentry;
1929 dentry = f.file->f_path.dentry;
1932 if (!d_can_lookup(dentry)) {
1938 nd->path = f.file->f_path;
1939 if (flags & LOOKUP_RCU) {
1940 if (f.flags & FDPUT_FPUT)
1942 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1945 path_get(&nd->path);
1950 nd->inode = nd->path.dentry->d_inode;
1951 if (!(flags & LOOKUP_RCU))
1953 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1955 if (!(nd->flags & LOOKUP_ROOT))
1956 nd->root.mnt = NULL;
1960 current->total_link_count = 0;
1961 return link_path_walk(s, nd);
1964 static void path_cleanup(struct nameidata *nd)
1966 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1967 path_put(&nd->root);
1968 nd->root.mnt = NULL;
1970 if (unlikely(nd->base))
1974 static int trailing_symlink(struct path *link, struct nameidata *nd, void **p)
1977 int error = may_follow_link(link, nd);
1978 if (unlikely(error))
1980 nd->flags |= LOOKUP_PARENT;
1981 s = get_link(link, nd, p);
1982 if (unlikely(IS_ERR(s)))
1989 path_put(&nd->path);
1990 nd->path = nd->root;
1991 path_get(&nd->root);
1992 nd->flags |= LOOKUP_JUMPED;
1994 nd->inode = nd->path.dentry->d_inode;
1995 error = link_path_walk(s, nd);
1996 if (unlikely(error))
1997 put_link(nd, link, *p);
2001 static inline int lookup_last(struct nameidata *nd)
2003 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2004 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2006 nd->flags &= ~LOOKUP_PARENT;
2007 return walk_component(nd, nd->flags & LOOKUP_FOLLOW);
2010 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2011 static int path_lookupat(int dfd, const struct filename *name,
2012 unsigned int flags, struct nameidata *nd)
2017 * Path walking is largely split up into 2 different synchronisation
2018 * schemes, rcu-walk and ref-walk (explained in
2019 * Documentation/filesystems/path-lookup.txt). These share much of the
2020 * path walk code, but some things particularly setup, cleanup, and
2021 * following mounts are sufficiently divergent that functions are
2022 * duplicated. Typically there is a function foo(), and its RCU
2023 * analogue, foo_rcu().
2025 * -ECHILD is the error number of choice (just to avoid clashes) that
2026 * is returned if some aspect of an rcu-walk fails. Such an error must
2027 * be handled by restarting a traditional ref-walk (which will always
2028 * be able to complete).
2030 err = path_init(dfd, name, flags, nd);
2031 if (!err && !(flags & LOOKUP_PARENT)) {
2032 err = lookup_last(nd);
2034 nd->stack[0].link = nd->link;
2035 err = trailing_symlink(&nd->stack[0].link,
2036 nd, &nd->stack[0].cookie);
2039 err = lookup_last(nd);
2040 put_link(nd, &nd->stack[0].link, nd->stack[0].cookie);
2045 err = complete_walk(nd);
2047 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2048 if (!d_can_lookup(nd->path.dentry)) {
2049 path_put(&nd->path);
2058 static int filename_lookup(int dfd, struct filename *name,
2059 unsigned int flags, struct nameidata *nd)
2061 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2062 if (unlikely(retval == -ECHILD))
2063 retval = path_lookupat(dfd, name, flags, nd);
2064 if (unlikely(retval == -ESTALE))
2065 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2067 if (likely(!retval))
2068 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2072 /* does lookup, returns the object with parent locked */
2073 struct dentry *kern_path_locked(const char *name, struct path *path)
2075 struct filename *filename = getname_kernel(name);
2076 struct nameidata nd;
2080 if (IS_ERR(filename))
2081 return ERR_CAST(filename);
2083 err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd);
2088 if (nd.last_type != LAST_NORM) {
2090 d = ERR_PTR(-EINVAL);
2093 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2094 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2096 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2106 int kern_path(const char *name, unsigned int flags, struct path *path)
2108 struct nameidata nd;
2109 struct filename *filename = getname_kernel(name);
2110 int res = PTR_ERR(filename);
2112 if (!IS_ERR(filename)) {
2113 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2120 EXPORT_SYMBOL(kern_path);
2123 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2124 * @dentry: pointer to dentry of the base directory
2125 * @mnt: pointer to vfs mount of the base directory
2126 * @name: pointer to file name
2127 * @flags: lookup flags
2128 * @path: pointer to struct path to fill
2130 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2131 const char *name, unsigned int flags,
2134 struct filename *filename = getname_kernel(name);
2135 int err = PTR_ERR(filename);
2137 BUG_ON(flags & LOOKUP_PARENT);
2139 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2140 if (!IS_ERR(filename)) {
2141 struct nameidata nd;
2142 nd.root.dentry = dentry;
2144 err = filename_lookup(AT_FDCWD, filename,
2145 flags | LOOKUP_ROOT, &nd);
2152 EXPORT_SYMBOL(vfs_path_lookup);
2155 * lookup_one_len - filesystem helper to lookup single pathname component
2156 * @name: pathname component to lookup
2157 * @base: base directory to lookup from
2158 * @len: maximum length @len should be interpreted to
2160 * Note that this routine is purely a helper for filesystem usage and should
2161 * not be called by generic code.
2163 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2169 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2173 this.hash = full_name_hash(name, len);
2175 return ERR_PTR(-EACCES);
2177 if (unlikely(name[0] == '.')) {
2178 if (len < 2 || (len == 2 && name[1] == '.'))
2179 return ERR_PTR(-EACCES);
2183 c = *(const unsigned char *)name++;
2184 if (c == '/' || c == '\0')
2185 return ERR_PTR(-EACCES);
2188 * See if the low-level filesystem might want
2189 * to use its own hash..
2191 if (base->d_flags & DCACHE_OP_HASH) {
2192 int err = base->d_op->d_hash(base, &this);
2194 return ERR_PTR(err);
2197 err = inode_permission(base->d_inode, MAY_EXEC);
2199 return ERR_PTR(err);
2201 return __lookup_hash(&this, base, 0);
2203 EXPORT_SYMBOL(lookup_one_len);
2205 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2206 struct path *path, int *empty)
2208 struct nameidata nd;
2209 struct filename *tmp = getname_flags(name, flags, empty);
2210 int err = PTR_ERR(tmp);
2213 BUG_ON(flags & LOOKUP_PARENT);
2215 err = filename_lookup(dfd, tmp, flags, &nd);
2223 int user_path_at(int dfd, const char __user *name, unsigned flags,
2226 return user_path_at_empty(dfd, name, flags, path, NULL);
2228 EXPORT_SYMBOL(user_path_at);
2231 * NB: most callers don't do anything directly with the reference to the
2232 * to struct filename, but the nd->last pointer points into the name string
2233 * allocated by getname. So we must hold the reference to it until all
2234 * path-walking is complete.
2236 static struct filename *
2237 user_path_parent(int dfd, const char __user *path,
2238 struct path *parent,
2243 struct nameidata nd;
2244 struct filename *s = getname(path);
2247 /* only LOOKUP_REVAL is allowed in extra flags */
2248 flags &= LOOKUP_REVAL;
2253 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, &nd);
2256 return ERR_PTR(error);
2260 *type = nd.last_type;
2266 * mountpoint_last - look up last component for umount
2267 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2268 * @path: pointer to container for result
2270 * This is a special lookup_last function just for umount. In this case, we
2271 * need to resolve the path without doing any revalidation.
2273 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2274 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2275 * in almost all cases, this lookup will be served out of the dcache. The only
2276 * cases where it won't are if nd->last refers to a symlink or the path is
2277 * bogus and it doesn't exist.
2280 * -error: if there was an error during lookup. This includes -ENOENT if the
2281 * lookup found a negative dentry. The nd->path reference will also be
2284 * 0: if we successfully resolved nd->path and found it to not to be a
2285 * symlink that needs to be followed. "path" will also be populated.
2286 * The nd->path reference will also be put.
2288 * 1: if we successfully resolved nd->last and found it to be a symlink
2289 * that needs to be followed. "path" will be populated with the path
2290 * to the link, and nd->path will *not* be put.
2293 mountpoint_last(struct nameidata *nd, struct path *path)
2296 struct dentry *dentry;
2297 struct dentry *dir = nd->path.dentry;
2299 /* If we're in rcuwalk, drop out of it to handle last component */
2300 if (nd->flags & LOOKUP_RCU) {
2301 if (unlazy_walk(nd, NULL)) {
2307 nd->flags &= ~LOOKUP_PARENT;
2309 if (unlikely(nd->last_type != LAST_NORM)) {
2310 error = handle_dots(nd, nd->last_type);
2313 dentry = dget(nd->path.dentry);
2317 mutex_lock(&dir->d_inode->i_mutex);
2318 dentry = d_lookup(dir, &nd->last);
2321 * No cached dentry. Mounted dentries are pinned in the cache,
2322 * so that means that this dentry is probably a symlink or the
2323 * path doesn't actually point to a mounted dentry.
2325 dentry = d_alloc(dir, &nd->last);
2328 mutex_unlock(&dir->d_inode->i_mutex);
2331 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2332 error = PTR_ERR(dentry);
2333 if (IS_ERR(dentry)) {
2334 mutex_unlock(&dir->d_inode->i_mutex);
2338 mutex_unlock(&dir->d_inode->i_mutex);
2341 if (d_is_negative(dentry)) {
2346 path->dentry = dentry;
2347 path->mnt = nd->path.mnt;
2348 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) {
2361 * path_mountpoint - look up a path to be umounted
2362 * @dfd: directory file descriptor to start walk from
2363 * @name: full pathname to walk
2364 * @path: pointer to container for result
2365 * @flags: lookup flags
2367 * Look up the given name, but don't attempt to revalidate the last component.
2368 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2371 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2372 struct nameidata *nd, unsigned int flags)
2374 int err = path_init(dfd, name, flags, nd);
2378 err = mountpoint_last(nd, path);
2380 nd->stack[0].link = nd->link;
2381 err = trailing_symlink(&nd->stack[0].link,
2382 nd, &nd->stack[0].cookie);
2385 err = mountpoint_last(nd, path);
2386 put_link(nd, &nd->stack[0].link, nd->stack[0].cookie);
2394 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2397 struct nameidata nd;
2400 return PTR_ERR(name);
2401 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2402 if (unlikely(error == -ECHILD))
2403 error = path_mountpoint(dfd, name, path, &nd, flags);
2404 if (unlikely(error == -ESTALE))
2405 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2407 audit_inode(name, path->dentry, 0);
2413 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2414 * @dfd: directory file descriptor
2415 * @name: pathname from userland
2416 * @flags: lookup flags
2417 * @path: pointer to container to hold result
2419 * A umount is a special case for path walking. We're not actually interested
2420 * in the inode in this situation, and ESTALE errors can be a problem. We
2421 * simply want track down the dentry and vfsmount attached at the mountpoint
2422 * and avoid revalidating the last component.
2424 * Returns 0 and populates "path" on success.
2427 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2430 return filename_mountpoint(dfd, getname(name), path, flags);
2434 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2437 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2439 EXPORT_SYMBOL(kern_path_mountpoint);
2441 int __check_sticky(struct inode *dir, struct inode *inode)
2443 kuid_t fsuid = current_fsuid();
2445 if (uid_eq(inode->i_uid, fsuid))
2447 if (uid_eq(dir->i_uid, fsuid))
2449 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2451 EXPORT_SYMBOL(__check_sticky);
2454 * Check whether we can remove a link victim from directory dir, check
2455 * whether the type of victim is right.
2456 * 1. We can't do it if dir is read-only (done in permission())
2457 * 2. We should have write and exec permissions on dir
2458 * 3. We can't remove anything from append-only dir
2459 * 4. We can't do anything with immutable dir (done in permission())
2460 * 5. If the sticky bit on dir is set we should either
2461 * a. be owner of dir, or
2462 * b. be owner of victim, or
2463 * c. have CAP_FOWNER capability
2464 * 6. If the victim is append-only or immutable we can't do antyhing with
2465 * links pointing to it.
2466 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2467 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2468 * 9. We can't remove a root or mountpoint.
2469 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2470 * nfs_async_unlink().
2472 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2474 struct inode *inode = victim->d_inode;
2477 if (d_is_negative(victim))
2481 BUG_ON(victim->d_parent->d_inode != dir);
2482 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2484 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2490 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2491 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2494 if (!d_is_dir(victim))
2496 if (IS_ROOT(victim))
2498 } else if (d_is_dir(victim))
2500 if (IS_DEADDIR(dir))
2502 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2507 /* Check whether we can create an object with dentry child in directory
2509 * 1. We can't do it if child already exists (open has special treatment for
2510 * this case, but since we are inlined it's OK)
2511 * 2. We can't do it if dir is read-only (done in permission())
2512 * 3. We should have write and exec permissions on dir
2513 * 4. We can't do it if dir is immutable (done in permission())
2515 static inline int may_create(struct inode *dir, struct dentry *child)
2517 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2520 if (IS_DEADDIR(dir))
2522 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2526 * p1 and p2 should be directories on the same fs.
2528 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2533 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2537 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2539 p = d_ancestor(p2, p1);
2541 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2542 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2546 p = d_ancestor(p1, p2);
2548 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2549 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2553 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2554 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2557 EXPORT_SYMBOL(lock_rename);
2559 void unlock_rename(struct dentry *p1, struct dentry *p2)
2561 mutex_unlock(&p1->d_inode->i_mutex);
2563 mutex_unlock(&p2->d_inode->i_mutex);
2564 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2567 EXPORT_SYMBOL(unlock_rename);
2569 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2572 int error = may_create(dir, dentry);
2576 if (!dir->i_op->create)
2577 return -EACCES; /* shouldn't it be ENOSYS? */
2580 error = security_inode_create(dir, dentry, mode);
2583 error = dir->i_op->create(dir, dentry, mode, want_excl);
2585 fsnotify_create(dir, dentry);
2588 EXPORT_SYMBOL(vfs_create);
2590 static int may_open(struct path *path, int acc_mode, int flag)
2592 struct dentry *dentry = path->dentry;
2593 struct inode *inode = dentry->d_inode;
2603 switch (inode->i_mode & S_IFMT) {
2607 if (acc_mode & MAY_WRITE)
2612 if (path->mnt->mnt_flags & MNT_NODEV)
2621 error = inode_permission(inode, acc_mode);
2626 * An append-only file must be opened in append mode for writing.
2628 if (IS_APPEND(inode)) {
2629 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2635 /* O_NOATIME can only be set by the owner or superuser */
2636 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2642 static int handle_truncate(struct file *filp)
2644 struct path *path = &filp->f_path;
2645 struct inode *inode = path->dentry->d_inode;
2646 int error = get_write_access(inode);
2650 * Refuse to truncate files with mandatory locks held on them.
2652 error = locks_verify_locked(filp);
2654 error = security_path_truncate(path);
2656 error = do_truncate(path->dentry, 0,
2657 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2660 put_write_access(inode);
2664 static inline int open_to_namei_flags(int flag)
2666 if ((flag & O_ACCMODE) == 3)
2671 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2673 int error = security_path_mknod(dir, dentry, mode, 0);
2677 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2681 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2685 * Attempt to atomically look up, create and open a file from a negative
2688 * Returns 0 if successful. The file will have been created and attached to
2689 * @file by the filesystem calling finish_open().
2691 * Returns 1 if the file was looked up only or didn't need creating. The
2692 * caller will need to perform the open themselves. @path will have been
2693 * updated to point to the new dentry. This may be negative.
2695 * Returns an error code otherwise.
2697 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2698 struct path *path, struct file *file,
2699 const struct open_flags *op,
2700 bool got_write, bool need_lookup,
2703 struct inode *dir = nd->path.dentry->d_inode;
2704 unsigned open_flag = open_to_namei_flags(op->open_flag);
2708 int create_error = 0;
2709 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2712 BUG_ON(dentry->d_inode);
2714 /* Don't create child dentry for a dead directory. */
2715 if (unlikely(IS_DEADDIR(dir))) {
2721 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2722 mode &= ~current_umask();
2724 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2726 open_flag &= ~O_TRUNC;
2729 * Checking write permission is tricky, bacuse we don't know if we are
2730 * going to actually need it: O_CREAT opens should work as long as the
2731 * file exists. But checking existence breaks atomicity. The trick is
2732 * to check access and if not granted clear O_CREAT from the flags.
2734 * Another problem is returing the "right" error value (e.g. for an
2735 * O_EXCL open we want to return EEXIST not EROFS).
2737 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2738 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2739 if (!(open_flag & O_CREAT)) {
2741 * No O_CREATE -> atomicity not a requirement -> fall
2742 * back to lookup + open
2745 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2746 /* Fall back and fail with the right error */
2747 create_error = -EROFS;
2750 /* No side effects, safe to clear O_CREAT */
2751 create_error = -EROFS;
2752 open_flag &= ~O_CREAT;
2756 if (open_flag & O_CREAT) {
2757 error = may_o_create(&nd->path, dentry, mode);
2759 create_error = error;
2760 if (open_flag & O_EXCL)
2762 open_flag &= ~O_CREAT;
2766 if (nd->flags & LOOKUP_DIRECTORY)
2767 open_flag |= O_DIRECTORY;
2769 file->f_path.dentry = DENTRY_NOT_SET;
2770 file->f_path.mnt = nd->path.mnt;
2771 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2774 if (create_error && error == -ENOENT)
2775 error = create_error;
2779 if (error) { /* returned 1, that is */
2780 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2784 if (file->f_path.dentry) {
2786 dentry = file->f_path.dentry;
2788 if (*opened & FILE_CREATED)
2789 fsnotify_create(dir, dentry);
2790 if (!dentry->d_inode) {
2791 WARN_ON(*opened & FILE_CREATED);
2793 error = create_error;
2797 if (excl && !(*opened & FILE_CREATED)) {
2806 * We didn't have the inode before the open, so check open permission
2809 acc_mode = op->acc_mode;
2810 if (*opened & FILE_CREATED) {
2811 WARN_ON(!(open_flag & O_CREAT));
2812 fsnotify_create(dir, dentry);
2813 acc_mode = MAY_OPEN;
2815 error = may_open(&file->f_path, acc_mode, open_flag);
2825 dentry = lookup_real(dir, dentry, nd->flags);
2827 return PTR_ERR(dentry);
2830 int open_flag = op->open_flag;
2832 error = create_error;
2833 if ((open_flag & O_EXCL)) {
2834 if (!dentry->d_inode)
2836 } else if (!dentry->d_inode) {
2838 } else if ((open_flag & O_TRUNC) &&
2842 /* will fail later, go on to get the right error */
2846 path->dentry = dentry;
2847 path->mnt = nd->path.mnt;
2852 * Look up and maybe create and open the last component.
2854 * Must be called with i_mutex held on parent.
2856 * Returns 0 if the file was successfully atomically created (if necessary) and
2857 * opened. In this case the file will be returned attached to @file.
2859 * Returns 1 if the file was not completely opened at this time, though lookups
2860 * and creations will have been performed and the dentry returned in @path will
2861 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2862 * specified then a negative dentry may be returned.
2864 * An error code is returned otherwise.
2866 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2867 * cleared otherwise prior to returning.
2869 static int lookup_open(struct nameidata *nd, struct path *path,
2871 const struct open_flags *op,
2872 bool got_write, int *opened)
2874 struct dentry *dir = nd->path.dentry;
2875 struct inode *dir_inode = dir->d_inode;
2876 struct dentry *dentry;
2880 *opened &= ~FILE_CREATED;
2881 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2883 return PTR_ERR(dentry);
2885 /* Cached positive dentry: will open in f_op->open */
2886 if (!need_lookup && dentry->d_inode)
2889 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2890 return atomic_open(nd, dentry, path, file, op, got_write,
2891 need_lookup, opened);
2895 BUG_ON(dentry->d_inode);
2897 dentry = lookup_real(dir_inode, dentry, nd->flags);
2899 return PTR_ERR(dentry);
2902 /* Negative dentry, just create the file */
2903 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2904 umode_t mode = op->mode;
2905 if (!IS_POSIXACL(dir->d_inode))
2906 mode &= ~current_umask();
2908 * This write is needed to ensure that a
2909 * rw->ro transition does not occur between
2910 * the time when the file is created and when
2911 * a permanent write count is taken through
2912 * the 'struct file' in finish_open().
2918 *opened |= FILE_CREATED;
2919 error = security_path_mknod(&nd->path, dentry, mode, 0);
2922 error = vfs_create(dir->d_inode, dentry, mode,
2923 nd->flags & LOOKUP_EXCL);
2928 path->dentry = dentry;
2929 path->mnt = nd->path.mnt;
2938 * Handle the last step of open()
2940 static int do_last(struct nameidata *nd,
2941 struct file *file, const struct open_flags *op,
2942 int *opened, struct filename *name)
2944 struct dentry *dir = nd->path.dentry;
2945 int open_flag = op->open_flag;
2946 bool will_truncate = (open_flag & O_TRUNC) != 0;
2947 bool got_write = false;
2948 int acc_mode = op->acc_mode;
2949 struct inode *inode;
2950 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2952 bool retried = false;
2955 nd->flags &= ~LOOKUP_PARENT;
2956 nd->flags |= op->intent;
2958 if (nd->last_type != LAST_NORM) {
2959 error = handle_dots(nd, nd->last_type);
2965 if (!(open_flag & O_CREAT)) {
2966 if (nd->last.name[nd->last.len])
2967 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2968 /* we _can_ be in RCU mode here */
2969 error = lookup_fast(nd, &path, &inode);
2976 BUG_ON(nd->inode != dir->d_inode);
2978 /* create side of things */
2980 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2981 * has been cleared when we got to the last component we are
2984 error = complete_walk(nd);
2988 audit_inode(name, dir, LOOKUP_PARENT);
2990 /* trailing slashes? */
2991 if (nd->last.name[nd->last.len])
2996 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2997 error = mnt_want_write(nd->path.mnt);
3001 * do _not_ fail yet - we might not need that or fail with
3002 * a different error; let lookup_open() decide; we'll be
3003 * dropping this one anyway.
3006 mutex_lock(&dir->d_inode->i_mutex);
3007 error = lookup_open(nd, &path, file, op, got_write, opened);
3008 mutex_unlock(&dir->d_inode->i_mutex);
3014 if ((*opened & FILE_CREATED) ||
3015 !S_ISREG(file_inode(file)->i_mode))
3016 will_truncate = false;
3018 audit_inode(name, file->f_path.dentry, 0);
3022 if (*opened & FILE_CREATED) {
3023 /* Don't check for write permission, don't truncate */
3024 open_flag &= ~O_TRUNC;
3025 will_truncate = false;
3026 acc_mode = MAY_OPEN;
3027 path_to_nameidata(&path, nd);
3028 goto finish_open_created;
3032 * create/update audit record if it already exists.
3034 if (d_is_positive(path.dentry))
3035 audit_inode(name, path.dentry, 0);
3038 * If atomic_open() acquired write access it is dropped now due to
3039 * possible mount and symlink following (this might be optimized away if
3043 mnt_drop_write(nd->path.mnt);
3048 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3051 error = follow_managed(&path, nd->flags);
3056 nd->flags |= LOOKUP_JUMPED;
3058 BUG_ON(nd->flags & LOOKUP_RCU);
3059 inode = path.dentry->d_inode;
3061 if (d_is_negative(path.dentry)) {
3062 path_to_nameidata(&path, nd);
3066 if (should_follow_link(path.dentry, nd->flags & LOOKUP_FOLLOW)) {
3067 if (nd->flags & LOOKUP_RCU) {
3068 if (unlikely(nd->path.mnt != path.mnt ||
3069 unlazy_walk(nd, path.dentry))) {
3074 BUG_ON(inode != path.dentry->d_inode);
3079 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3080 path_to_nameidata(&path, nd);
3085 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3086 path_to_nameidata(&path, nd);
3088 save_parent.dentry = nd->path.dentry;
3089 save_parent.mnt = mntget(path.mnt);
3090 nd->path.dentry = path.dentry;
3094 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3096 error = complete_walk(nd);
3098 path_put(&save_parent);
3101 audit_inode(name, nd->path.dentry, 0);
3103 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3106 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3108 if (!d_is_reg(nd->path.dentry))
3109 will_truncate = false;
3111 if (will_truncate) {
3112 error = mnt_want_write(nd->path.mnt);
3117 finish_open_created:
3118 error = may_open(&nd->path, acc_mode, open_flag);
3122 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3123 error = vfs_open(&nd->path, file, current_cred());
3125 *opened |= FILE_OPENED;
3127 if (error == -EOPENSTALE)
3132 error = open_check_o_direct(file);
3135 error = ima_file_check(file, op->acc_mode, *opened);
3139 if (will_truncate) {
3140 error = handle_truncate(file);
3146 mnt_drop_write(nd->path.mnt);
3147 path_put(&save_parent);
3152 path_put_conditional(&path, nd);
3159 /* If no saved parent or already retried then can't retry */
3160 if (!save_parent.dentry || retried)
3163 BUG_ON(save_parent.dentry != dir);
3164 path_put(&nd->path);
3165 nd->path = save_parent;
3166 nd->inode = dir->d_inode;
3167 save_parent.mnt = NULL;
3168 save_parent.dentry = NULL;
3170 mnt_drop_write(nd->path.mnt);
3177 static int do_tmpfile(int dfd, struct filename *pathname,
3178 struct nameidata *nd, int flags,
3179 const struct open_flags *op,
3180 struct file *file, int *opened)
3182 static const struct qstr name = QSTR_INIT("/", 1);
3183 struct dentry *dentry, *child;
3185 int error = path_lookupat(dfd, pathname,
3186 flags | LOOKUP_DIRECTORY, nd);
3187 if (unlikely(error))
3189 error = mnt_want_write(nd->path.mnt);
3190 if (unlikely(error))
3192 /* we want directory to be writable */
3193 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3196 dentry = nd->path.dentry;
3197 dir = dentry->d_inode;
3198 if (!dir->i_op->tmpfile) {
3199 error = -EOPNOTSUPP;
3202 child = d_alloc(dentry, &name);
3203 if (unlikely(!child)) {
3207 nd->flags &= ~LOOKUP_DIRECTORY;
3208 nd->flags |= op->intent;
3209 dput(nd->path.dentry);
3210 nd->path.dentry = child;
3211 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3214 audit_inode(pathname, nd->path.dentry, 0);
3215 /* Don't check for other permissions, the inode was just created */
3216 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3219 file->f_path.mnt = nd->path.mnt;
3220 error = finish_open(file, nd->path.dentry, NULL, opened);
3223 error = open_check_o_direct(file);
3226 } else if (!(op->open_flag & O_EXCL)) {
3227 struct inode *inode = file_inode(file);
3228 spin_lock(&inode->i_lock);
3229 inode->i_state |= I_LINKABLE;
3230 spin_unlock(&inode->i_lock);
3233 mnt_drop_write(nd->path.mnt);
3235 path_put(&nd->path);
3239 static struct file *path_openat(int dfd, struct filename *pathname,
3240 struct nameidata *nd, const struct open_flags *op, int flags)
3246 file = get_empty_filp();
3250 file->f_flags = op->open_flag;
3252 if (unlikely(file->f_flags & __O_TMPFILE)) {
3253 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3257 error = path_init(dfd, pathname, flags, nd);
3258 if (unlikely(error))
3261 error = do_last(nd, file, op, &opened, pathname);
3262 while (unlikely(error > 0)) { /* trailing symlink */
3263 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3264 nd->stack[0].link = nd->link;
3265 error= trailing_symlink(&nd->stack[0].link,
3266 nd, &nd->stack[0].cookie);
3267 if (unlikely(error))
3269 error = do_last(nd, file, op, &opened, pathname);
3270 put_link(nd, &nd->stack[0].link, nd->stack[0].cookie);
3275 if (!(opened & FILE_OPENED)) {
3279 if (unlikely(error)) {
3280 if (error == -EOPENSTALE) {
3281 if (flags & LOOKUP_RCU)
3286 file = ERR_PTR(error);
3291 struct file *do_filp_open(int dfd, struct filename *pathname,
3292 const struct open_flags *op)
3294 struct nameidata nd;
3295 int flags = op->lookup_flags;
3298 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3299 if (unlikely(filp == ERR_PTR(-ECHILD)))
3300 filp = path_openat(dfd, pathname, &nd, op, flags);
3301 if (unlikely(filp == ERR_PTR(-ESTALE)))
3302 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3306 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3307 const char *name, const struct open_flags *op)
3309 struct nameidata nd;
3311 struct filename *filename;
3312 int flags = op->lookup_flags | LOOKUP_ROOT;
3315 nd.root.dentry = dentry;
3317 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3318 return ERR_PTR(-ELOOP);
3320 filename = getname_kernel(name);
3321 if (unlikely(IS_ERR(filename)))
3322 return ERR_CAST(filename);
3324 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3325 if (unlikely(file == ERR_PTR(-ECHILD)))
3326 file = path_openat(-1, filename, &nd, op, flags);
3327 if (unlikely(file == ERR_PTR(-ESTALE)))
3328 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3333 static struct dentry *filename_create(int dfd, struct filename *name,
3334 struct path *path, unsigned int lookup_flags)
3336 struct dentry *dentry = ERR_PTR(-EEXIST);
3337 struct nameidata nd;
3340 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3343 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3344 * other flags passed in are ignored!
3346 lookup_flags &= LOOKUP_REVAL;
3348 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3350 return ERR_PTR(error);
3353 * Yucky last component or no last component at all?
3354 * (foo/., foo/.., /////)
3356 if (nd.last_type != LAST_NORM)
3358 nd.flags &= ~LOOKUP_PARENT;
3359 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3361 /* don't fail immediately if it's r/o, at least try to report other errors */
3362 err2 = mnt_want_write(nd.path.mnt);
3364 * Do the final lookup.
3366 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3367 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3372 if (d_is_positive(dentry))
3376 * Special case - lookup gave negative, but... we had foo/bar/
3377 * From the vfs_mknod() POV we just have a negative dentry -
3378 * all is fine. Let's be bastards - you had / on the end, you've
3379 * been asking for (non-existent) directory. -ENOENT for you.
3381 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3385 if (unlikely(err2)) {
3393 dentry = ERR_PTR(error);
3395 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3397 mnt_drop_write(nd.path.mnt);
3403 struct dentry *kern_path_create(int dfd, const char *pathname,
3404 struct path *path, unsigned int lookup_flags)
3406 struct filename *filename = getname_kernel(pathname);
3409 if (IS_ERR(filename))
3410 return ERR_CAST(filename);
3411 res = filename_create(dfd, filename, path, lookup_flags);
3415 EXPORT_SYMBOL(kern_path_create);
3417 void done_path_create(struct path *path, struct dentry *dentry)
3420 mutex_unlock(&path->dentry->d_inode->i_mutex);
3421 mnt_drop_write(path->mnt);
3424 EXPORT_SYMBOL(done_path_create);
3426 struct dentry *user_path_create(int dfd, const char __user *pathname,
3427 struct path *path, unsigned int lookup_flags)
3429 struct filename *tmp = getname(pathname);
3432 return ERR_CAST(tmp);
3433 res = filename_create(dfd, tmp, path, lookup_flags);
3437 EXPORT_SYMBOL(user_path_create);
3439 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3441 int error = may_create(dir, dentry);
3446 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3449 if (!dir->i_op->mknod)
3452 error = devcgroup_inode_mknod(mode, dev);
3456 error = security_inode_mknod(dir, dentry, mode, dev);
3460 error = dir->i_op->mknod(dir, dentry, mode, dev);
3462 fsnotify_create(dir, dentry);
3465 EXPORT_SYMBOL(vfs_mknod);
3467 static int may_mknod(umode_t mode)
3469 switch (mode & S_IFMT) {
3475 case 0: /* zero mode translates to S_IFREG */
3484 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3487 struct dentry *dentry;
3490 unsigned int lookup_flags = 0;
3492 error = may_mknod(mode);
3496 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3498 return PTR_ERR(dentry);
3500 if (!IS_POSIXACL(path.dentry->d_inode))
3501 mode &= ~current_umask();
3502 error = security_path_mknod(&path, dentry, mode, dev);
3505 switch (mode & S_IFMT) {
3506 case 0: case S_IFREG:
3507 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3509 case S_IFCHR: case S_IFBLK:
3510 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3511 new_decode_dev(dev));
3513 case S_IFIFO: case S_IFSOCK:
3514 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3518 done_path_create(&path, dentry);
3519 if (retry_estale(error, lookup_flags)) {
3520 lookup_flags |= LOOKUP_REVAL;
3526 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3528 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3531 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3533 int error = may_create(dir, dentry);
3534 unsigned max_links = dir->i_sb->s_max_links;
3539 if (!dir->i_op->mkdir)
3542 mode &= (S_IRWXUGO|S_ISVTX);
3543 error = security_inode_mkdir(dir, dentry, mode);
3547 if (max_links && dir->i_nlink >= max_links)
3550 error = dir->i_op->mkdir(dir, dentry, mode);
3552 fsnotify_mkdir(dir, dentry);
3555 EXPORT_SYMBOL(vfs_mkdir);
3557 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3559 struct dentry *dentry;
3562 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3565 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3567 return PTR_ERR(dentry);
3569 if (!IS_POSIXACL(path.dentry->d_inode))
3570 mode &= ~current_umask();
3571 error = security_path_mkdir(&path, dentry, mode);
3573 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3574 done_path_create(&path, dentry);
3575 if (retry_estale(error, lookup_flags)) {
3576 lookup_flags |= LOOKUP_REVAL;
3582 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3584 return sys_mkdirat(AT_FDCWD, pathname, mode);
3588 * The dentry_unhash() helper will try to drop the dentry early: we
3589 * should have a usage count of 1 if we're the only user of this
3590 * dentry, and if that is true (possibly after pruning the dcache),
3591 * then we drop the dentry now.
3593 * A low-level filesystem can, if it choses, legally
3596 * if (!d_unhashed(dentry))
3599 * if it cannot handle the case of removing a directory
3600 * that is still in use by something else..
3602 void dentry_unhash(struct dentry *dentry)
3604 shrink_dcache_parent(dentry);
3605 spin_lock(&dentry->d_lock);
3606 if (dentry->d_lockref.count == 1)
3608 spin_unlock(&dentry->d_lock);
3610 EXPORT_SYMBOL(dentry_unhash);
3612 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3614 int error = may_delete(dir, dentry, 1);
3619 if (!dir->i_op->rmdir)
3623 mutex_lock(&dentry->d_inode->i_mutex);
3626 if (is_local_mountpoint(dentry))
3629 error = security_inode_rmdir(dir, dentry);
3633 shrink_dcache_parent(dentry);
3634 error = dir->i_op->rmdir(dir, dentry);
3638 dentry->d_inode->i_flags |= S_DEAD;
3640 detach_mounts(dentry);
3643 mutex_unlock(&dentry->d_inode->i_mutex);
3649 EXPORT_SYMBOL(vfs_rmdir);
3651 static long do_rmdir(int dfd, const char __user *pathname)
3654 struct filename *name;
3655 struct dentry *dentry;
3659 unsigned int lookup_flags = 0;
3661 name = user_path_parent(dfd, pathname,
3662 &path, &last, &type, lookup_flags);
3664 return PTR_ERR(name);
3678 error = mnt_want_write(path.mnt);
3682 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3683 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3684 error = PTR_ERR(dentry);
3687 if (!dentry->d_inode) {
3691 error = security_path_rmdir(&path, dentry);
3694 error = vfs_rmdir(path.dentry->d_inode, dentry);
3698 mutex_unlock(&path.dentry->d_inode->i_mutex);
3699 mnt_drop_write(path.mnt);
3703 if (retry_estale(error, lookup_flags)) {
3704 lookup_flags |= LOOKUP_REVAL;
3710 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3712 return do_rmdir(AT_FDCWD, pathname);
3716 * vfs_unlink - unlink a filesystem object
3717 * @dir: parent directory
3719 * @delegated_inode: returns victim inode, if the inode is delegated.
3721 * The caller must hold dir->i_mutex.
3723 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3724 * return a reference to the inode in delegated_inode. The caller
3725 * should then break the delegation on that inode and retry. Because
3726 * breaking a delegation may take a long time, the caller should drop
3727 * dir->i_mutex before doing so.
3729 * Alternatively, a caller may pass NULL for delegated_inode. This may
3730 * be appropriate for callers that expect the underlying filesystem not
3731 * to be NFS exported.
3733 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3735 struct inode *target = dentry->d_inode;
3736 int error = may_delete(dir, dentry, 0);
3741 if (!dir->i_op->unlink)
3744 mutex_lock(&target->i_mutex);
3745 if (is_local_mountpoint(dentry))
3748 error = security_inode_unlink(dir, dentry);
3750 error = try_break_deleg(target, delegated_inode);
3753 error = dir->i_op->unlink(dir, dentry);
3756 detach_mounts(dentry);
3761 mutex_unlock(&target->i_mutex);
3763 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3764 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3765 fsnotify_link_count(target);
3771 EXPORT_SYMBOL(vfs_unlink);
3774 * Make sure that the actual truncation of the file will occur outside its
3775 * directory's i_mutex. Truncate can take a long time if there is a lot of
3776 * writeout happening, and we don't want to prevent access to the directory
3777 * while waiting on the I/O.
3779 static long do_unlinkat(int dfd, const char __user *pathname)
3782 struct filename *name;
3783 struct dentry *dentry;
3787 struct inode *inode = NULL;
3788 struct inode *delegated_inode = NULL;
3789 unsigned int lookup_flags = 0;
3791 name = user_path_parent(dfd, pathname,
3792 &path, &last, &type, lookup_flags);
3794 return PTR_ERR(name);
3797 if (type != LAST_NORM)
3800 error = mnt_want_write(path.mnt);
3804 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3805 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3806 error = PTR_ERR(dentry);
3807 if (!IS_ERR(dentry)) {
3808 /* Why not before? Because we want correct error value */
3809 if (last.name[last.len])
3811 inode = dentry->d_inode;
3812 if (d_is_negative(dentry))
3815 error = security_path_unlink(&path, dentry);
3818 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3822 mutex_unlock(&path.dentry->d_inode->i_mutex);
3824 iput(inode); /* truncate the inode here */
3826 if (delegated_inode) {
3827 error = break_deleg_wait(&delegated_inode);
3831 mnt_drop_write(path.mnt);
3835 if (retry_estale(error, lookup_flags)) {
3836 lookup_flags |= LOOKUP_REVAL;
3843 if (d_is_negative(dentry))
3845 else if (d_is_dir(dentry))
3852 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3854 if ((flag & ~AT_REMOVEDIR) != 0)
3857 if (flag & AT_REMOVEDIR)
3858 return do_rmdir(dfd, pathname);
3860 return do_unlinkat(dfd, pathname);
3863 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3865 return do_unlinkat(AT_FDCWD, pathname);
3868 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3870 int error = may_create(dir, dentry);
3875 if (!dir->i_op->symlink)
3878 error = security_inode_symlink(dir, dentry, oldname);
3882 error = dir->i_op->symlink(dir, dentry, oldname);
3884 fsnotify_create(dir, dentry);
3887 EXPORT_SYMBOL(vfs_symlink);
3889 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3890 int, newdfd, const char __user *, newname)
3893 struct filename *from;
3894 struct dentry *dentry;
3896 unsigned int lookup_flags = 0;
3898 from = getname(oldname);
3900 return PTR_ERR(from);
3902 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3903 error = PTR_ERR(dentry);
3907 error = security_path_symlink(&path, dentry, from->name);
3909 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3910 done_path_create(&path, dentry);
3911 if (retry_estale(error, lookup_flags)) {
3912 lookup_flags |= LOOKUP_REVAL;
3920 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3922 return sys_symlinkat(oldname, AT_FDCWD, newname);
3926 * vfs_link - create a new link
3927 * @old_dentry: object to be linked
3929 * @new_dentry: where to create the new link
3930 * @delegated_inode: returns inode needing a delegation break
3932 * The caller must hold dir->i_mutex
3934 * If vfs_link discovers a delegation on the to-be-linked file in need
3935 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3936 * inode in delegated_inode. The caller should then break the delegation
3937 * and retry. Because breaking a delegation may take a long time, the
3938 * caller should drop the i_mutex before doing so.
3940 * Alternatively, a caller may pass NULL for delegated_inode. This may
3941 * be appropriate for callers that expect the underlying filesystem not
3942 * to be NFS exported.
3944 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3946 struct inode *inode = old_dentry->d_inode;
3947 unsigned max_links = dir->i_sb->s_max_links;
3953 error = may_create(dir, new_dentry);
3957 if (dir->i_sb != inode->i_sb)
3961 * A link to an append-only or immutable file cannot be created.
3963 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3965 if (!dir->i_op->link)
3967 if (S_ISDIR(inode->i_mode))
3970 error = security_inode_link(old_dentry, dir, new_dentry);
3974 mutex_lock(&inode->i_mutex);
3975 /* Make sure we don't allow creating hardlink to an unlinked file */
3976 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3978 else if (max_links && inode->i_nlink >= max_links)
3981 error = try_break_deleg(inode, delegated_inode);
3983 error = dir->i_op->link(old_dentry, dir, new_dentry);
3986 if (!error && (inode->i_state & I_LINKABLE)) {
3987 spin_lock(&inode->i_lock);
3988 inode->i_state &= ~I_LINKABLE;
3989 spin_unlock(&inode->i_lock);
3991 mutex_unlock(&inode->i_mutex);
3993 fsnotify_link(dir, inode, new_dentry);
3996 EXPORT_SYMBOL(vfs_link);
3999 * Hardlinks are often used in delicate situations. We avoid
4000 * security-related surprises by not following symlinks on the
4003 * We don't follow them on the oldname either to be compatible
4004 * with linux 2.0, and to avoid hard-linking to directories
4005 * and other special files. --ADM
4007 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4008 int, newdfd, const char __user *, newname, int, flags)
4010 struct dentry *new_dentry;
4011 struct path old_path, new_path;
4012 struct inode *delegated_inode = NULL;
4016 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4019 * To use null names we require CAP_DAC_READ_SEARCH
4020 * This ensures that not everyone will be able to create
4021 * handlink using the passed filedescriptor.
4023 if (flags & AT_EMPTY_PATH) {
4024 if (!capable(CAP_DAC_READ_SEARCH))
4029 if (flags & AT_SYMLINK_FOLLOW)
4030 how |= LOOKUP_FOLLOW;
4032 error = user_path_at(olddfd, oldname, how, &old_path);
4036 new_dentry = user_path_create(newdfd, newname, &new_path,
4037 (how & LOOKUP_REVAL));
4038 error = PTR_ERR(new_dentry);
4039 if (IS_ERR(new_dentry))
4043 if (old_path.mnt != new_path.mnt)
4045 error = may_linkat(&old_path);
4046 if (unlikely(error))
4048 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4051 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4053 done_path_create(&new_path, new_dentry);
4054 if (delegated_inode) {
4055 error = break_deleg_wait(&delegated_inode);
4057 path_put(&old_path);
4061 if (retry_estale(error, how)) {
4062 path_put(&old_path);
4063 how |= LOOKUP_REVAL;
4067 path_put(&old_path);
4072 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4074 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4078 * vfs_rename - rename a filesystem object
4079 * @old_dir: parent of source
4080 * @old_dentry: source
4081 * @new_dir: parent of destination
4082 * @new_dentry: destination
4083 * @delegated_inode: returns an inode needing a delegation break
4084 * @flags: rename flags
4086 * The caller must hold multiple mutexes--see lock_rename()).
4088 * If vfs_rename discovers a delegation in need of breaking at either
4089 * the source or destination, it will return -EWOULDBLOCK and return a
4090 * reference to the inode in delegated_inode. The caller should then
4091 * break the delegation and retry. Because breaking a delegation may
4092 * take a long time, the caller should drop all locks before doing
4095 * Alternatively, a caller may pass NULL for delegated_inode. This may
4096 * be appropriate for callers that expect the underlying filesystem not
4097 * to be NFS exported.
4099 * The worst of all namespace operations - renaming directory. "Perverted"
4100 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4102 * a) we can get into loop creation.
4103 * b) race potential - two innocent renames can create a loop together.
4104 * That's where 4.4 screws up. Current fix: serialization on
4105 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4107 * c) we have to lock _four_ objects - parents and victim (if it exists),
4108 * and source (if it is not a directory).
4109 * And that - after we got ->i_mutex on parents (until then we don't know
4110 * whether the target exists). Solution: try to be smart with locking
4111 * order for inodes. We rely on the fact that tree topology may change
4112 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4113 * move will be locked. Thus we can rank directories by the tree
4114 * (ancestors first) and rank all non-directories after them.
4115 * That works since everybody except rename does "lock parent, lookup,
4116 * lock child" and rename is under ->s_vfs_rename_mutex.
4117 * HOWEVER, it relies on the assumption that any object with ->lookup()
4118 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4119 * we'd better make sure that there's no link(2) for them.
4120 * d) conversion from fhandle to dentry may come in the wrong moment - when
4121 * we are removing the target. Solution: we will have to grab ->i_mutex
4122 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4123 * ->i_mutex on parents, which works but leads to some truly excessive
4126 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4127 struct inode *new_dir, struct dentry *new_dentry,
4128 struct inode **delegated_inode, unsigned int flags)
4131 bool is_dir = d_is_dir(old_dentry);
4132 const unsigned char *old_name;
4133 struct inode *source = old_dentry->d_inode;
4134 struct inode *target = new_dentry->d_inode;
4135 bool new_is_dir = false;
4136 unsigned max_links = new_dir->i_sb->s_max_links;
4138 if (source == target)
4141 error = may_delete(old_dir, old_dentry, is_dir);
4146 error = may_create(new_dir, new_dentry);
4148 new_is_dir = d_is_dir(new_dentry);
4150 if (!(flags & RENAME_EXCHANGE))
4151 error = may_delete(new_dir, new_dentry, is_dir);
4153 error = may_delete(new_dir, new_dentry, new_is_dir);
4158 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4161 if (flags && !old_dir->i_op->rename2)
4165 * If we are going to change the parent - check write permissions,
4166 * we'll need to flip '..'.
4168 if (new_dir != old_dir) {
4170 error = inode_permission(source, MAY_WRITE);
4174 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4175 error = inode_permission(target, MAY_WRITE);
4181 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4186 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4188 if (!is_dir || (flags & RENAME_EXCHANGE))
4189 lock_two_nondirectories(source, target);
4191 mutex_lock(&target->i_mutex);
4194 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4197 if (max_links && new_dir != old_dir) {
4199 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4201 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4202 old_dir->i_nlink >= max_links)
4205 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4206 shrink_dcache_parent(new_dentry);
4208 error = try_break_deleg(source, delegated_inode);
4212 if (target && !new_is_dir) {
4213 error = try_break_deleg(target, delegated_inode);
4217 if (!old_dir->i_op->rename2) {
4218 error = old_dir->i_op->rename(old_dir, old_dentry,
4219 new_dir, new_dentry);
4221 WARN_ON(old_dir->i_op->rename != NULL);
4222 error = old_dir->i_op->rename2(old_dir, old_dentry,
4223 new_dir, new_dentry, flags);
4228 if (!(flags & RENAME_EXCHANGE) && target) {
4230 target->i_flags |= S_DEAD;
4231 dont_mount(new_dentry);
4232 detach_mounts(new_dentry);
4234 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4235 if (!(flags & RENAME_EXCHANGE))
4236 d_move(old_dentry, new_dentry);
4238 d_exchange(old_dentry, new_dentry);
4241 if (!is_dir || (flags & RENAME_EXCHANGE))
4242 unlock_two_nondirectories(source, target);
4244 mutex_unlock(&target->i_mutex);
4247 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4248 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4249 if (flags & RENAME_EXCHANGE) {
4250 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4251 new_is_dir, NULL, new_dentry);
4254 fsnotify_oldname_free(old_name);
4258 EXPORT_SYMBOL(vfs_rename);
4260 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4261 int, newdfd, const char __user *, newname, unsigned int, flags)
4263 struct dentry *old_dentry, *new_dentry;
4264 struct dentry *trap;
4265 struct path old_path, new_path;
4266 struct qstr old_last, new_last;
4267 int old_type, new_type;
4268 struct inode *delegated_inode = NULL;
4269 struct filename *from;
4270 struct filename *to;
4271 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4272 bool should_retry = false;
4275 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4278 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4279 (flags & RENAME_EXCHANGE))
4282 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4285 if (flags & RENAME_EXCHANGE)
4289 from = user_path_parent(olddfd, oldname,
4290 &old_path, &old_last, &old_type, lookup_flags);
4292 error = PTR_ERR(from);
4296 to = user_path_parent(newdfd, newname,
4297 &new_path, &new_last, &new_type, lookup_flags);
4299 error = PTR_ERR(to);
4304 if (old_path.mnt != new_path.mnt)
4308 if (old_type != LAST_NORM)
4311 if (flags & RENAME_NOREPLACE)
4313 if (new_type != LAST_NORM)
4316 error = mnt_want_write(old_path.mnt);
4321 trap = lock_rename(new_path.dentry, old_path.dentry);
4323 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4324 error = PTR_ERR(old_dentry);
4325 if (IS_ERR(old_dentry))
4327 /* source must exist */
4329 if (d_is_negative(old_dentry))
4331 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4332 error = PTR_ERR(new_dentry);
4333 if (IS_ERR(new_dentry))
4336 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4338 if (flags & RENAME_EXCHANGE) {
4340 if (d_is_negative(new_dentry))
4343 if (!d_is_dir(new_dentry)) {
4345 if (new_last.name[new_last.len])
4349 /* unless the source is a directory trailing slashes give -ENOTDIR */
4350 if (!d_is_dir(old_dentry)) {
4352 if (old_last.name[old_last.len])
4354 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4357 /* source should not be ancestor of target */
4359 if (old_dentry == trap)
4361 /* target should not be an ancestor of source */
4362 if (!(flags & RENAME_EXCHANGE))
4364 if (new_dentry == trap)
4367 error = security_path_rename(&old_path, old_dentry,
4368 &new_path, new_dentry, flags);
4371 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4372 new_path.dentry->d_inode, new_dentry,
4373 &delegated_inode, flags);
4379 unlock_rename(new_path.dentry, old_path.dentry);
4380 if (delegated_inode) {
4381 error = break_deleg_wait(&delegated_inode);
4385 mnt_drop_write(old_path.mnt);
4387 if (retry_estale(error, lookup_flags))
4388 should_retry = true;
4389 path_put(&new_path);
4392 path_put(&old_path);
4395 should_retry = false;
4396 lookup_flags |= LOOKUP_REVAL;
4403 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4404 int, newdfd, const char __user *, newname)
4406 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4409 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4411 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4414 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4416 int error = may_create(dir, dentry);
4420 if (!dir->i_op->mknod)
4423 return dir->i_op->mknod(dir, dentry,
4424 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4426 EXPORT_SYMBOL(vfs_whiteout);
4428 int readlink_copy(char __user *buffer, int buflen, const char *link)
4430 int len = PTR_ERR(link);
4435 if (len > (unsigned) buflen)
4437 if (copy_to_user(buffer, link, len))
4442 EXPORT_SYMBOL(readlink_copy);
4445 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4446 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4447 * using) it for any given inode is up to filesystem.
4449 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4452 const char *link = dentry->d_inode->i_link;
4456 link = dentry->d_inode->i_op->follow_link(dentry, &cookie, NULL);
4458 return PTR_ERR(link);
4460 res = readlink_copy(buffer, buflen, link);
4461 if (cookie && dentry->d_inode->i_op->put_link)
4462 dentry->d_inode->i_op->put_link(dentry, cookie);
4465 EXPORT_SYMBOL(generic_readlink);
4467 /* get the link contents into pagecache */
4468 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4472 struct address_space *mapping = dentry->d_inode->i_mapping;
4473 page = read_mapping_page(mapping, 0, NULL);
4478 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4482 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4484 struct page *page = NULL;
4485 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4488 page_cache_release(page);
4492 EXPORT_SYMBOL(page_readlink);
4494 const char *page_follow_link_light(struct dentry *dentry, void **cookie, struct nameidata *nd)
4496 struct page *page = NULL;
4497 char *res = page_getlink(dentry, &page);
4502 EXPORT_SYMBOL(page_follow_link_light);
4504 void page_put_link(struct dentry *dentry, void *cookie)
4506 struct page *page = cookie;
4508 page_cache_release(page);
4510 EXPORT_SYMBOL(page_put_link);
4513 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4515 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4517 struct address_space *mapping = inode->i_mapping;
4522 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4524 flags |= AOP_FLAG_NOFS;
4527 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4528 flags, &page, &fsdata);
4532 kaddr = kmap_atomic(page);
4533 memcpy(kaddr, symname, len-1);
4534 kunmap_atomic(kaddr);
4536 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4543 mark_inode_dirty(inode);
4548 EXPORT_SYMBOL(__page_symlink);
4550 int page_symlink(struct inode *inode, const char *symname, int len)
4552 return __page_symlink(inode, symname, len,
4553 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4555 EXPORT_SYMBOL(page_symlink);
4557 const struct inode_operations page_symlink_inode_operations = {
4558 .readlink = generic_readlink,
4559 .follow_link = page_follow_link_light,
4560 .put_link = page_put_link,
4562 EXPORT_SYMBOL(page_symlink_inode_operations);