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.
121 void final_putname(struct filename *name)
123 if (name->separate) {
124 __putname(name->name);
131 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
133 static struct filename *
134 getname_flags(const char __user *filename, int flags, int *empty)
136 struct filename *result, *err;
141 result = audit_reusename(filename);
145 result = __getname();
146 if (unlikely(!result))
147 return ERR_PTR(-ENOMEM);
150 * First, try to embed the struct filename inside the names_cache
153 kname = (char *)result + sizeof(*result);
154 result->name = kname;
155 result->separate = false;
156 max = EMBEDDED_NAME_MAX;
159 len = strncpy_from_user(kname, filename, max);
160 if (unlikely(len < 0)) {
166 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
167 * separate struct filename so we can dedicate the entire
168 * names_cache allocation for the pathname, and re-do the copy from
171 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
172 kname = (char *)result;
174 result = kzalloc(sizeof(*result), GFP_KERNEL);
176 err = ERR_PTR(-ENOMEM);
177 result = (struct filename *)kname;
180 result->name = kname;
181 result->separate = true;
186 /* The empty path is special. */
187 if (unlikely(!len)) {
190 err = ERR_PTR(-ENOENT);
191 if (!(flags & LOOKUP_EMPTY))
195 err = ERR_PTR(-ENAMETOOLONG);
196 if (unlikely(len >= PATH_MAX))
199 result->uptr = filename;
200 result->aname = NULL;
201 audit_getname(result);
205 final_putname(result);
210 getname(const char __user * filename)
212 return getname_flags(filename, 0, NULL);
216 * The "getname_kernel()" interface doesn't do pathnames longer
217 * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user.
220 getname_kernel(const char * filename)
222 struct filename *result;
226 len = strlen(filename);
227 if (len >= EMBEDDED_NAME_MAX)
228 return ERR_PTR(-ENAMETOOLONG);
230 result = __getname();
231 if (unlikely(!result))
232 return ERR_PTR(-ENOMEM);
234 kname = (char *)result + sizeof(*result);
235 result->name = kname;
237 result->aname = NULL;
238 result->separate = false;
240 strlcpy(kname, filename, EMBEDDED_NAME_MAX);
244 #ifdef CONFIG_AUDITSYSCALL
245 void putname(struct filename *name)
247 if (unlikely(!audit_dummy_context()))
248 return audit_putname(name);
253 static int check_acl(struct inode *inode, int mask)
255 #ifdef CONFIG_FS_POSIX_ACL
256 struct posix_acl *acl;
258 if (mask & MAY_NOT_BLOCK) {
259 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
262 /* no ->get_acl() calls in RCU mode... */
263 if (acl == ACL_NOT_CACHED)
265 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
268 acl = get_acl(inode, ACL_TYPE_ACCESS);
272 int error = posix_acl_permission(inode, acl, mask);
273 posix_acl_release(acl);
282 * This does the basic permission checking
284 static int acl_permission_check(struct inode *inode, int mask)
286 unsigned int mode = inode->i_mode;
288 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
291 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
292 int error = check_acl(inode, mask);
293 if (error != -EAGAIN)
297 if (in_group_p(inode->i_gid))
302 * If the DACs are ok we don't need any capability check.
304 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
310 * generic_permission - check for access rights on a Posix-like filesystem
311 * @inode: inode to check access rights for
312 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
314 * Used to check for read/write/execute permissions on a file.
315 * We use "fsuid" for this, letting us set arbitrary permissions
316 * for filesystem access without changing the "normal" uids which
317 * are used for other things.
319 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
320 * request cannot be satisfied (eg. requires blocking or too much complexity).
321 * It would then be called again in ref-walk mode.
323 int generic_permission(struct inode *inode, int mask)
328 * Do the basic permission checks.
330 ret = acl_permission_check(inode, mask);
334 if (S_ISDIR(inode->i_mode)) {
335 /* DACs are overridable for directories */
336 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
338 if (!(mask & MAY_WRITE))
339 if (capable_wrt_inode_uidgid(inode,
340 CAP_DAC_READ_SEARCH))
345 * Read/write DACs are always overridable.
346 * Executable DACs are overridable when there is
347 * at least one exec bit set.
349 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
350 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
354 * Searching includes executable on directories, else just read.
356 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
357 if (mask == MAY_READ)
358 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
363 EXPORT_SYMBOL(generic_permission);
366 * We _really_ want to just do "generic_permission()" without
367 * even looking at the inode->i_op values. So we keep a cache
368 * flag in inode->i_opflags, that says "this has not special
369 * permission function, use the fast case".
371 static inline int do_inode_permission(struct inode *inode, int mask)
373 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
374 if (likely(inode->i_op->permission))
375 return inode->i_op->permission(inode, mask);
377 /* This gets set once for the inode lifetime */
378 spin_lock(&inode->i_lock);
379 inode->i_opflags |= IOP_FASTPERM;
380 spin_unlock(&inode->i_lock);
382 return generic_permission(inode, mask);
386 * __inode_permission - Check for access rights to a given inode
387 * @inode: Inode to check permission on
388 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
390 * Check for read/write/execute permissions on an inode.
392 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
394 * This does not check for a read-only file system. You probably want
395 * inode_permission().
397 int __inode_permission(struct inode *inode, int mask)
401 if (unlikely(mask & MAY_WRITE)) {
403 * Nobody gets write access to an immutable file.
405 if (IS_IMMUTABLE(inode))
409 retval = do_inode_permission(inode, mask);
413 retval = devcgroup_inode_permission(inode, mask);
417 return security_inode_permission(inode, mask);
421 * sb_permission - Check superblock-level permissions
422 * @sb: Superblock of inode to check permission on
423 * @inode: Inode to check permission on
424 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
426 * Separate out file-system wide checks from inode-specific permission checks.
428 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
430 if (unlikely(mask & MAY_WRITE)) {
431 umode_t mode = inode->i_mode;
433 /* Nobody gets write access to a read-only fs. */
434 if ((sb->s_flags & MS_RDONLY) &&
435 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
442 * inode_permission - Check for access rights to a given inode
443 * @inode: Inode to check permission on
444 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
446 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
447 * this, letting us set arbitrary permissions for filesystem access without
448 * changing the "normal" UIDs which are used for other things.
450 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
452 int inode_permission(struct inode *inode, int mask)
456 retval = sb_permission(inode->i_sb, inode, mask);
459 return __inode_permission(inode, mask);
461 EXPORT_SYMBOL(inode_permission);
464 * path_get - get a reference to a path
465 * @path: path to get the reference to
467 * Given a path increment the reference count to the dentry and the vfsmount.
469 void path_get(const struct path *path)
474 EXPORT_SYMBOL(path_get);
477 * path_put - put a reference to a path
478 * @path: path to put the reference to
480 * Given a path decrement the reference count to the dentry and the vfsmount.
482 void path_put(const struct path *path)
487 EXPORT_SYMBOL(path_put);
490 * Path walking has 2 modes, rcu-walk and ref-walk (see
491 * Documentation/filesystems/path-lookup.txt). In situations when we can't
492 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
493 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
494 * mode. Refcounts are grabbed at the last known good point before rcu-walk
495 * got stuck, so ref-walk may continue from there. If this is not successful
496 * (eg. a seqcount has changed), then failure is returned and it's up to caller
497 * to restart the path walk from the beginning in ref-walk mode.
501 * unlazy_walk - try to switch to ref-walk mode.
502 * @nd: nameidata pathwalk data
503 * @dentry: child of nd->path.dentry or NULL
504 * Returns: 0 on success, -ECHILD on failure
506 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
507 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
508 * @nd or NULL. Must be called from rcu-walk context.
510 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
512 struct fs_struct *fs = current->fs;
513 struct dentry *parent = nd->path.dentry;
515 BUG_ON(!(nd->flags & LOOKUP_RCU));
518 * After legitimizing the bastards, terminate_walk()
519 * will do the right thing for non-RCU mode, and all our
520 * subsequent exit cases should rcu_read_unlock()
521 * before returning. Do vfsmount first; if dentry
522 * can't be legitimized, just set nd->path.dentry to NULL
523 * and rely on dput(NULL) being a no-op.
525 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
527 nd->flags &= ~LOOKUP_RCU;
529 if (!lockref_get_not_dead(&parent->d_lockref)) {
530 nd->path.dentry = NULL;
535 * For a negative lookup, the lookup sequence point is the parents
536 * sequence point, and it only needs to revalidate the parent dentry.
538 * For a positive lookup, we need to move both the parent and the
539 * dentry from the RCU domain to be properly refcounted. And the
540 * sequence number in the dentry validates *both* dentry counters,
541 * since we checked the sequence number of the parent after we got
542 * the child sequence number. So we know the parent must still
543 * be valid if the child sequence number is still valid.
546 if (read_seqcount_retry(&parent->d_seq, nd->seq))
548 BUG_ON(nd->inode != parent->d_inode);
550 if (!lockref_get_not_dead(&dentry->d_lockref))
552 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
557 * Sequence counts matched. Now make sure that the root is
558 * still valid and get it if required.
560 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
561 spin_lock(&fs->lock);
562 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
563 goto unlock_and_drop_dentry;
565 spin_unlock(&fs->lock);
571 unlock_and_drop_dentry:
572 spin_unlock(&fs->lock);
580 if (!(nd->flags & LOOKUP_ROOT))
585 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
587 return dentry->d_op->d_revalidate(dentry, flags);
591 * complete_walk - successful completion of path walk
592 * @nd: pointer nameidata
594 * If we had been in RCU mode, drop out of it and legitimize nd->path.
595 * Revalidate the final result, unless we'd already done that during
596 * the path walk or the filesystem doesn't ask for it. Return 0 on
597 * success, -error on failure. In case of failure caller does not
598 * need to drop nd->path.
600 static int complete_walk(struct nameidata *nd)
602 struct dentry *dentry = nd->path.dentry;
605 if (nd->flags & LOOKUP_RCU) {
606 nd->flags &= ~LOOKUP_RCU;
607 if (!(nd->flags & LOOKUP_ROOT))
610 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
614 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
616 mntput(nd->path.mnt);
619 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
622 mntput(nd->path.mnt);
628 if (likely(!(nd->flags & LOOKUP_JUMPED)))
631 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
634 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
645 static __always_inline void set_root(struct nameidata *nd)
647 get_fs_root(current->fs, &nd->root);
650 static int link_path_walk(const char *, struct nameidata *);
652 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
654 struct fs_struct *fs = current->fs;
658 seq = read_seqcount_begin(&fs->seq);
660 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
661 } while (read_seqcount_retry(&fs->seq, seq));
665 static void path_put_conditional(struct path *path, struct nameidata *nd)
668 if (path->mnt != nd->path.mnt)
672 static inline void path_to_nameidata(const struct path *path,
673 struct nameidata *nd)
675 if (!(nd->flags & LOOKUP_RCU)) {
676 dput(nd->path.dentry);
677 if (nd->path.mnt != path->mnt)
678 mntput(nd->path.mnt);
680 nd->path.mnt = path->mnt;
681 nd->path.dentry = path->dentry;
685 * Helper to directly jump to a known parsed path from ->follow_link,
686 * caller must have taken a reference to path beforehand.
688 void nd_jump_link(struct nameidata *nd, struct path *path)
693 nd->inode = nd->path.dentry->d_inode;
694 nd->flags |= LOOKUP_JUMPED;
697 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
699 struct inode *inode = link->dentry->d_inode;
700 if (inode->i_op->put_link)
701 inode->i_op->put_link(link->dentry, nd, cookie);
705 int sysctl_protected_symlinks __read_mostly = 0;
706 int sysctl_protected_hardlinks __read_mostly = 0;
709 * may_follow_link - Check symlink following for unsafe situations
710 * @link: The path of the symlink
711 * @nd: nameidata pathwalk data
713 * In the case of the sysctl_protected_symlinks sysctl being enabled,
714 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
715 * in a sticky world-writable directory. This is to protect privileged
716 * processes from failing races against path names that may change out
717 * from under them by way of other users creating malicious symlinks.
718 * It will permit symlinks to be followed only when outside a sticky
719 * world-writable directory, or when the uid of the symlink and follower
720 * match, or when the directory owner matches the symlink's owner.
722 * Returns 0 if following the symlink is allowed, -ve on error.
724 static inline int may_follow_link(struct path *link, struct nameidata *nd)
726 const struct inode *inode;
727 const struct inode *parent;
729 if (!sysctl_protected_symlinks)
732 /* Allowed if owner and follower match. */
733 inode = link->dentry->d_inode;
734 if (uid_eq(current_cred()->fsuid, inode->i_uid))
737 /* Allowed if parent directory not sticky and world-writable. */
738 parent = nd->path.dentry->d_inode;
739 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
742 /* Allowed if parent directory and link owner match. */
743 if (uid_eq(parent->i_uid, inode->i_uid))
746 audit_log_link_denied("follow_link", link);
747 path_put_conditional(link, nd);
753 * safe_hardlink_source - Check for safe hardlink conditions
754 * @inode: the source inode to hardlink from
756 * Return false if at least one of the following conditions:
757 * - inode is not a regular file
759 * - inode is setgid and group-exec
760 * - access failure for read and write
762 * Otherwise returns true.
764 static bool safe_hardlink_source(struct inode *inode)
766 umode_t mode = inode->i_mode;
768 /* Special files should not get pinned to the filesystem. */
772 /* Setuid files should not get pinned to the filesystem. */
776 /* Executable setgid files should not get pinned to the filesystem. */
777 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
780 /* Hardlinking to unreadable or unwritable sources is dangerous. */
781 if (inode_permission(inode, MAY_READ | MAY_WRITE))
788 * may_linkat - Check permissions for creating a hardlink
789 * @link: the source to hardlink from
791 * Block hardlink when all of:
792 * - sysctl_protected_hardlinks enabled
793 * - fsuid does not match inode
794 * - hardlink source is unsafe (see safe_hardlink_source() above)
797 * Returns 0 if successful, -ve on error.
799 static int may_linkat(struct path *link)
801 const struct cred *cred;
804 if (!sysctl_protected_hardlinks)
807 cred = current_cred();
808 inode = link->dentry->d_inode;
810 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
811 * otherwise, it must be a safe source.
813 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
817 audit_log_link_denied("linkat", link);
821 static __always_inline int
822 follow_link(struct path *link, struct nameidata *nd, void **p)
824 struct dentry *dentry = link->dentry;
828 BUG_ON(nd->flags & LOOKUP_RCU);
830 if (link->mnt == nd->path.mnt)
834 if (unlikely(current->total_link_count >= 40))
835 goto out_put_nd_path;
838 current->total_link_count++;
841 nd_set_link(nd, NULL);
843 error = security_inode_follow_link(link->dentry, nd);
845 goto out_put_nd_path;
847 nd->last_type = LAST_BIND;
848 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
851 goto out_put_nd_path;
856 if (unlikely(IS_ERR(s))) {
858 put_link(nd, link, *p);
867 nd->flags |= LOOKUP_JUMPED;
869 nd->inode = nd->path.dentry->d_inode;
870 error = link_path_walk(s, nd);
872 put_link(nd, link, *p);
884 static int follow_up_rcu(struct path *path)
886 struct mount *mnt = real_mount(path->mnt);
887 struct mount *parent;
888 struct dentry *mountpoint;
890 parent = mnt->mnt_parent;
891 if (&parent->mnt == path->mnt)
893 mountpoint = mnt->mnt_mountpoint;
894 path->dentry = mountpoint;
895 path->mnt = &parent->mnt;
900 * follow_up - Find the mountpoint of path's vfsmount
902 * Given a path, find the mountpoint of its source file system.
903 * Replace @path with the path of the mountpoint in the parent mount.
906 * Return 1 if we went up a level and 0 if we were already at the
909 int follow_up(struct path *path)
911 struct mount *mnt = real_mount(path->mnt);
912 struct mount *parent;
913 struct dentry *mountpoint;
915 read_seqlock_excl(&mount_lock);
916 parent = mnt->mnt_parent;
918 read_sequnlock_excl(&mount_lock);
921 mntget(&parent->mnt);
922 mountpoint = dget(mnt->mnt_mountpoint);
923 read_sequnlock_excl(&mount_lock);
925 path->dentry = mountpoint;
927 path->mnt = &parent->mnt;
930 EXPORT_SYMBOL(follow_up);
933 * Perform an automount
934 * - return -EISDIR to tell follow_managed() to stop and return the path we
937 static int follow_automount(struct path *path, unsigned flags,
940 struct vfsmount *mnt;
943 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
946 /* We don't want to mount if someone's just doing a stat -
947 * unless they're stat'ing a directory and appended a '/' to
950 * We do, however, want to mount if someone wants to open or
951 * create a file of any type under the mountpoint, wants to
952 * traverse through the mountpoint or wants to open the
953 * mounted directory. Also, autofs may mark negative dentries
954 * as being automount points. These will need the attentions
955 * of the daemon to instantiate them before they can be used.
957 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
958 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
959 path->dentry->d_inode)
962 current->total_link_count++;
963 if (current->total_link_count >= 40)
966 mnt = path->dentry->d_op->d_automount(path);
969 * The filesystem is allowed to return -EISDIR here to indicate
970 * it doesn't want to automount. For instance, autofs would do
971 * this so that its userspace daemon can mount on this dentry.
973 * However, we can only permit this if it's a terminal point in
974 * the path being looked up; if it wasn't then the remainder of
975 * the path is inaccessible and we should say so.
977 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
982 if (!mnt) /* mount collision */
986 /* lock_mount() may release path->mnt on error */
990 err = finish_automount(mnt, path);
994 /* Someone else made a mount here whilst we were busy */
999 path->dentry = dget(mnt->mnt_root);
1008 * Handle a dentry that is managed in some way.
1009 * - Flagged for transit management (autofs)
1010 * - Flagged as mountpoint
1011 * - Flagged as automount point
1013 * This may only be called in refwalk mode.
1015 * Serialization is taken care of in namespace.c
1017 static int follow_managed(struct path *path, unsigned flags)
1019 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1021 bool need_mntput = false;
1024 /* Given that we're not holding a lock here, we retain the value in a
1025 * local variable for each dentry as we look at it so that we don't see
1026 * the components of that value change under us */
1027 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1028 managed &= DCACHE_MANAGED_DENTRY,
1029 unlikely(managed != 0)) {
1030 /* Allow the filesystem to manage the transit without i_mutex
1032 if (managed & DCACHE_MANAGE_TRANSIT) {
1033 BUG_ON(!path->dentry->d_op);
1034 BUG_ON(!path->dentry->d_op->d_manage);
1035 ret = path->dentry->d_op->d_manage(path->dentry, false);
1040 /* Transit to a mounted filesystem. */
1041 if (managed & DCACHE_MOUNTED) {
1042 struct vfsmount *mounted = lookup_mnt(path);
1047 path->mnt = mounted;
1048 path->dentry = dget(mounted->mnt_root);
1053 /* Something is mounted on this dentry in another
1054 * namespace and/or whatever was mounted there in this
1055 * namespace got unmounted before lookup_mnt() could
1059 /* Handle an automount point */
1060 if (managed & DCACHE_NEED_AUTOMOUNT) {
1061 ret = follow_automount(path, flags, &need_mntput);
1067 /* We didn't change the current path point */
1071 if (need_mntput && path->mnt == mnt)
1075 return ret < 0 ? ret : need_mntput;
1078 int follow_down_one(struct path *path)
1080 struct vfsmount *mounted;
1082 mounted = lookup_mnt(path);
1086 path->mnt = mounted;
1087 path->dentry = dget(mounted->mnt_root);
1092 EXPORT_SYMBOL(follow_down_one);
1094 static inline int managed_dentry_rcu(struct dentry *dentry)
1096 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1097 dentry->d_op->d_manage(dentry, true) : 0;
1101 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1102 * we meet a managed dentry that would need blocking.
1104 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1105 struct inode **inode)
1108 struct mount *mounted;
1110 * Don't forget we might have a non-mountpoint managed dentry
1111 * that wants to block transit.
1113 switch (managed_dentry_rcu(path->dentry)) {
1123 if (!d_mountpoint(path->dentry))
1124 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1126 mounted = __lookup_mnt(path->mnt, path->dentry);
1129 path->mnt = &mounted->mnt;
1130 path->dentry = mounted->mnt.mnt_root;
1131 nd->flags |= LOOKUP_JUMPED;
1132 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1134 * Update the inode too. We don't need to re-check the
1135 * dentry sequence number here after this d_inode read,
1136 * because a mount-point is always pinned.
1138 *inode = path->dentry->d_inode;
1140 return !read_seqretry(&mount_lock, nd->m_seq) &&
1141 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1144 static int follow_dotdot_rcu(struct nameidata *nd)
1146 struct inode *inode = nd->inode;
1151 if (nd->path.dentry == nd->root.dentry &&
1152 nd->path.mnt == nd->root.mnt) {
1155 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1156 struct dentry *old = nd->path.dentry;
1157 struct dentry *parent = old->d_parent;
1160 inode = parent->d_inode;
1161 seq = read_seqcount_begin(&parent->d_seq);
1162 if (read_seqcount_retry(&old->d_seq, nd->seq))
1164 nd->path.dentry = parent;
1168 if (!follow_up_rcu(&nd->path))
1170 inode = nd->path.dentry->d_inode;
1171 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1173 while (d_mountpoint(nd->path.dentry)) {
1174 struct mount *mounted;
1175 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1178 nd->path.mnt = &mounted->mnt;
1179 nd->path.dentry = mounted->mnt.mnt_root;
1180 inode = nd->path.dentry->d_inode;
1181 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1182 if (read_seqretry(&mount_lock, nd->m_seq))
1189 nd->flags &= ~LOOKUP_RCU;
1190 if (!(nd->flags & LOOKUP_ROOT))
1191 nd->root.mnt = NULL;
1197 * Follow down to the covering mount currently visible to userspace. At each
1198 * point, the filesystem owning that dentry may be queried as to whether the
1199 * caller is permitted to proceed or not.
1201 int follow_down(struct path *path)
1206 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1207 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1208 /* Allow the filesystem to manage the transit without i_mutex
1211 * We indicate to the filesystem if someone is trying to mount
1212 * something here. This gives autofs the chance to deny anyone
1213 * other than its daemon the right to mount on its
1216 * The filesystem may sleep at this point.
1218 if (managed & DCACHE_MANAGE_TRANSIT) {
1219 BUG_ON(!path->dentry->d_op);
1220 BUG_ON(!path->dentry->d_op->d_manage);
1221 ret = path->dentry->d_op->d_manage(
1222 path->dentry, false);
1224 return ret == -EISDIR ? 0 : ret;
1227 /* Transit to a mounted filesystem. */
1228 if (managed & DCACHE_MOUNTED) {
1229 struct vfsmount *mounted = lookup_mnt(path);
1234 path->mnt = mounted;
1235 path->dentry = dget(mounted->mnt_root);
1239 /* Don't handle automount points here */
1244 EXPORT_SYMBOL(follow_down);
1247 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1249 static void follow_mount(struct path *path)
1251 while (d_mountpoint(path->dentry)) {
1252 struct vfsmount *mounted = lookup_mnt(path);
1257 path->mnt = mounted;
1258 path->dentry = dget(mounted->mnt_root);
1262 static void follow_dotdot(struct nameidata *nd)
1268 struct dentry *old = nd->path.dentry;
1270 if (nd->path.dentry == nd->root.dentry &&
1271 nd->path.mnt == nd->root.mnt) {
1274 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1275 /* rare case of legitimate dget_parent()... */
1276 nd->path.dentry = dget_parent(nd->path.dentry);
1280 if (!follow_up(&nd->path))
1283 follow_mount(&nd->path);
1284 nd->inode = nd->path.dentry->d_inode;
1288 * This looks up the name in dcache, possibly revalidates the old dentry and
1289 * allocates a new one if not found or not valid. In the need_lookup argument
1290 * returns whether i_op->lookup is necessary.
1292 * dir->d_inode->i_mutex must be held
1294 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1295 unsigned int flags, bool *need_lookup)
1297 struct dentry *dentry;
1300 *need_lookup = false;
1301 dentry = d_lookup(dir, name);
1303 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1304 error = d_revalidate(dentry, flags);
1305 if (unlikely(error <= 0)) {
1308 return ERR_PTR(error);
1310 d_invalidate(dentry);
1319 dentry = d_alloc(dir, name);
1320 if (unlikely(!dentry))
1321 return ERR_PTR(-ENOMEM);
1323 *need_lookup = true;
1329 * Call i_op->lookup on the dentry. The dentry must be negative and
1332 * dir->d_inode->i_mutex must be held
1334 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1339 /* Don't create child dentry for a dead directory. */
1340 if (unlikely(IS_DEADDIR(dir))) {
1342 return ERR_PTR(-ENOENT);
1345 old = dir->i_op->lookup(dir, dentry, flags);
1346 if (unlikely(old)) {
1353 static struct dentry *__lookup_hash(struct qstr *name,
1354 struct dentry *base, unsigned int flags)
1357 struct dentry *dentry;
1359 dentry = lookup_dcache(name, base, flags, &need_lookup);
1363 return lookup_real(base->d_inode, dentry, flags);
1367 * It's more convoluted than I'd like it to be, but... it's still fairly
1368 * small and for now I'd prefer to have fast path as straight as possible.
1369 * It _is_ time-critical.
1371 static int lookup_fast(struct nameidata *nd,
1372 struct path *path, struct inode **inode)
1374 struct vfsmount *mnt = nd->path.mnt;
1375 struct dentry *dentry, *parent = nd->path.dentry;
1381 * Rename seqlock is not required here because in the off chance
1382 * of a false negative due to a concurrent rename, we're going to
1383 * do the non-racy lookup, below.
1385 if (nd->flags & LOOKUP_RCU) {
1387 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1392 * This sequence count validates that the inode matches
1393 * the dentry name information from lookup.
1395 *inode = dentry->d_inode;
1396 if (read_seqcount_retry(&dentry->d_seq, seq))
1400 * This sequence count validates that the parent had no
1401 * changes while we did the lookup of the dentry above.
1403 * The memory barrier in read_seqcount_begin of child is
1404 * enough, we can use __read_seqcount_retry here.
1406 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1410 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1411 status = d_revalidate(dentry, nd->flags);
1412 if (unlikely(status <= 0)) {
1413 if (status != -ECHILD)
1419 path->dentry = dentry;
1420 if (likely(__follow_mount_rcu(nd, path, inode)))
1423 if (unlazy_walk(nd, dentry))
1426 dentry = __d_lookup(parent, &nd->last);
1429 if (unlikely(!dentry))
1432 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1433 status = d_revalidate(dentry, nd->flags);
1434 if (unlikely(status <= 0)) {
1439 d_invalidate(dentry);
1445 path->dentry = dentry;
1446 err = follow_managed(path, nd->flags);
1447 if (unlikely(err < 0)) {
1448 path_put_conditional(path, nd);
1452 nd->flags |= LOOKUP_JUMPED;
1453 *inode = path->dentry->d_inode;
1460 /* Fast lookup failed, do it the slow way */
1461 static int lookup_slow(struct nameidata *nd, struct path *path)
1463 struct dentry *dentry, *parent;
1466 parent = nd->path.dentry;
1467 BUG_ON(nd->inode != parent->d_inode);
1469 mutex_lock(&parent->d_inode->i_mutex);
1470 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1471 mutex_unlock(&parent->d_inode->i_mutex);
1473 return PTR_ERR(dentry);
1474 path->mnt = nd->path.mnt;
1475 path->dentry = dentry;
1476 err = follow_managed(path, nd->flags);
1477 if (unlikely(err < 0)) {
1478 path_put_conditional(path, nd);
1482 nd->flags |= LOOKUP_JUMPED;
1486 static inline int may_lookup(struct nameidata *nd)
1488 if (nd->flags & LOOKUP_RCU) {
1489 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1492 if (unlazy_walk(nd, NULL))
1495 return inode_permission(nd->inode, MAY_EXEC);
1498 static inline int handle_dots(struct nameidata *nd, int type)
1500 if (type == LAST_DOTDOT) {
1501 if (nd->flags & LOOKUP_RCU) {
1502 if (follow_dotdot_rcu(nd))
1510 static void terminate_walk(struct nameidata *nd)
1512 if (!(nd->flags & LOOKUP_RCU)) {
1513 path_put(&nd->path);
1515 nd->flags &= ~LOOKUP_RCU;
1516 if (!(nd->flags & LOOKUP_ROOT))
1517 nd->root.mnt = NULL;
1523 * Do we need to follow links? We _really_ want to be able
1524 * to do this check without having to look at inode->i_op,
1525 * so we keep a cache of "no, this doesn't need follow_link"
1526 * for the common case.
1528 static inline int should_follow_link(struct dentry *dentry, int follow)
1530 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1533 static inline int walk_component(struct nameidata *nd, struct path *path,
1536 struct inode *inode;
1539 * "." and ".." are special - ".." especially so because it has
1540 * to be able to know about the current root directory and
1541 * parent relationships.
1543 if (unlikely(nd->last_type != LAST_NORM))
1544 return handle_dots(nd, nd->last_type);
1545 err = lookup_fast(nd, path, &inode);
1546 if (unlikely(err)) {
1550 err = lookup_slow(nd, path);
1554 inode = path->dentry->d_inode;
1557 if (!inode || d_is_negative(path->dentry))
1560 if (should_follow_link(path->dentry, follow)) {
1561 if (nd->flags & LOOKUP_RCU) {
1562 if (unlikely(unlazy_walk(nd, path->dentry))) {
1567 BUG_ON(inode != path->dentry->d_inode);
1570 path_to_nameidata(path, nd);
1575 path_to_nameidata(path, nd);
1582 * This limits recursive symlink follows to 8, while
1583 * limiting consecutive symlinks to 40.
1585 * Without that kind of total limit, nasty chains of consecutive
1586 * symlinks can cause almost arbitrarily long lookups.
1588 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1592 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1593 path_put_conditional(path, nd);
1594 path_put(&nd->path);
1597 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1600 current->link_count++;
1603 struct path link = *path;
1606 res = follow_link(&link, nd, &cookie);
1609 res = walk_component(nd, path, LOOKUP_FOLLOW);
1610 put_link(nd, &link, cookie);
1613 current->link_count--;
1619 * We can do the critical dentry name comparison and hashing
1620 * operations one word at a time, but we are limited to:
1622 * - Architectures with fast unaligned word accesses. We could
1623 * do a "get_unaligned()" if this helps and is sufficiently
1626 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1627 * do not trap on the (extremely unlikely) case of a page
1628 * crossing operation.
1630 * - Furthermore, we need an efficient 64-bit compile for the
1631 * 64-bit case in order to generate the "number of bytes in
1632 * the final mask". Again, that could be replaced with a
1633 * efficient population count instruction or similar.
1635 #ifdef CONFIG_DCACHE_WORD_ACCESS
1637 #include <asm/word-at-a-time.h>
1641 static inline unsigned int fold_hash(unsigned long hash)
1643 return hash_64(hash, 32);
1646 #else /* 32-bit case */
1648 #define fold_hash(x) (x)
1652 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1654 unsigned long a, mask;
1655 unsigned long hash = 0;
1658 a = load_unaligned_zeropad(name);
1659 if (len < sizeof(unsigned long))
1663 name += sizeof(unsigned long);
1664 len -= sizeof(unsigned long);
1668 mask = bytemask_from_count(len);
1671 return fold_hash(hash);
1673 EXPORT_SYMBOL(full_name_hash);
1676 * Calculate the length and hash of the path component, and
1677 * return the "hash_len" as the result.
1679 static inline u64 hash_name(const char *name)
1681 unsigned long a, b, adata, bdata, mask, hash, len;
1682 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1685 len = -sizeof(unsigned long);
1687 hash = (hash + a) * 9;
1688 len += sizeof(unsigned long);
1689 a = load_unaligned_zeropad(name+len);
1690 b = a ^ REPEAT_BYTE('/');
1691 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1693 adata = prep_zero_mask(a, adata, &constants);
1694 bdata = prep_zero_mask(b, bdata, &constants);
1696 mask = create_zero_mask(adata | bdata);
1698 hash += a & zero_bytemask(mask);
1699 len += find_zero(mask);
1700 return hashlen_create(fold_hash(hash), len);
1705 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1707 unsigned long hash = init_name_hash();
1709 hash = partial_name_hash(*name++, hash);
1710 return end_name_hash(hash);
1712 EXPORT_SYMBOL(full_name_hash);
1715 * We know there's a real path component here of at least
1718 static inline u64 hash_name(const char *name)
1720 unsigned long hash = init_name_hash();
1721 unsigned long len = 0, c;
1723 c = (unsigned char)*name;
1726 hash = partial_name_hash(c, hash);
1727 c = (unsigned char)name[len];
1728 } while (c && c != '/');
1729 return hashlen_create(end_name_hash(hash), len);
1736 * This is the basic name resolution function, turning a pathname into
1737 * the final dentry. We expect 'base' to be positive and a directory.
1739 * Returns 0 and nd will have valid dentry and mnt on success.
1740 * Returns error and drops reference to input namei data on failure.
1742 static int link_path_walk(const char *name, struct nameidata *nd)
1752 /* At this point we know we have a real path component. */
1757 err = may_lookup(nd);
1761 hash_len = hash_name(name);
1764 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1766 if (name[1] == '.') {
1768 nd->flags |= LOOKUP_JUMPED;
1774 if (likely(type == LAST_NORM)) {
1775 struct dentry *parent = nd->path.dentry;
1776 nd->flags &= ~LOOKUP_JUMPED;
1777 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1778 struct qstr this = { { .hash_len = hash_len }, .name = name };
1779 err = parent->d_op->d_hash(parent, &this);
1782 hash_len = this.hash_len;
1787 nd->last.hash_len = hash_len;
1788 nd->last.name = name;
1789 nd->last_type = type;
1791 name += hashlen_len(hash_len);
1795 * If it wasn't NUL, we know it was '/'. Skip that
1796 * slash, and continue until no more slashes.
1800 } while (unlikely(*name == '/'));
1804 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1809 err = nested_symlink(&next, nd);
1813 if (!d_can_lookup(nd->path.dentry)) {
1822 static int path_init(int dfd, const char *name, unsigned int flags,
1823 struct nameidata *nd, struct file **fp)
1827 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1828 nd->flags = flags | LOOKUP_JUMPED;
1830 if (flags & LOOKUP_ROOT) {
1831 struct dentry *root = nd->root.dentry;
1832 struct inode *inode = root->d_inode;
1834 if (!d_can_lookup(root))
1836 retval = inode_permission(inode, MAY_EXEC);
1840 nd->path = nd->root;
1842 if (flags & LOOKUP_RCU) {
1844 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1845 nd->m_seq = read_seqbegin(&mount_lock);
1847 path_get(&nd->path);
1852 nd->root.mnt = NULL;
1854 nd->m_seq = read_seqbegin(&mount_lock);
1856 if (flags & LOOKUP_RCU) {
1858 nd->seq = set_root_rcu(nd);
1861 path_get(&nd->root);
1863 nd->path = nd->root;
1864 } else if (dfd == AT_FDCWD) {
1865 if (flags & LOOKUP_RCU) {
1866 struct fs_struct *fs = current->fs;
1872 seq = read_seqcount_begin(&fs->seq);
1874 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1875 } while (read_seqcount_retry(&fs->seq, seq));
1877 get_fs_pwd(current->fs, &nd->path);
1880 /* Caller must check execute permissions on the starting path component */
1881 struct fd f = fdget_raw(dfd);
1882 struct dentry *dentry;
1887 dentry = f.file->f_path.dentry;
1890 if (!d_can_lookup(dentry)) {
1896 nd->path = f.file->f_path;
1897 if (flags & LOOKUP_RCU) {
1898 if (f.flags & FDPUT_FPUT)
1900 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1903 path_get(&nd->path);
1908 nd->inode = nd->path.dentry->d_inode;
1909 if (!(flags & LOOKUP_RCU))
1911 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1913 if (!(nd->flags & LOOKUP_ROOT))
1914 nd->root.mnt = NULL;
1919 static inline int lookup_last(struct nameidata *nd, struct path *path)
1921 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1922 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1924 nd->flags &= ~LOOKUP_PARENT;
1925 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1928 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1929 static int path_lookupat(int dfd, const char *name,
1930 unsigned int flags, struct nameidata *nd)
1932 struct file *base = NULL;
1937 * Path walking is largely split up into 2 different synchronisation
1938 * schemes, rcu-walk and ref-walk (explained in
1939 * Documentation/filesystems/path-lookup.txt). These share much of the
1940 * path walk code, but some things particularly setup, cleanup, and
1941 * following mounts are sufficiently divergent that functions are
1942 * duplicated. Typically there is a function foo(), and its RCU
1943 * analogue, foo_rcu().
1945 * -ECHILD is the error number of choice (just to avoid clashes) that
1946 * is returned if some aspect of an rcu-walk fails. Such an error must
1947 * be handled by restarting a traditional ref-walk (which will always
1948 * be able to complete).
1950 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1955 current->total_link_count = 0;
1956 err = link_path_walk(name, nd);
1958 if (!err && !(flags & LOOKUP_PARENT)) {
1959 err = lookup_last(nd, &path);
1962 struct path link = path;
1963 err = may_follow_link(&link, nd);
1966 nd->flags |= LOOKUP_PARENT;
1967 err = follow_link(&link, nd, &cookie);
1970 err = lookup_last(nd, &path);
1971 put_link(nd, &link, cookie);
1976 err = complete_walk(nd);
1978 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1979 if (!d_can_lookup(nd->path.dentry)) {
1980 path_put(&nd->path);
1989 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1990 path_put(&nd->root);
1991 nd->root.mnt = NULL;
1996 static int filename_lookup(int dfd, struct filename *name,
1997 unsigned int flags, struct nameidata *nd)
1999 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2000 if (unlikely(retval == -ECHILD))
2001 retval = path_lookupat(dfd, name->name, flags, nd);
2002 if (unlikely(retval == -ESTALE))
2003 retval = path_lookupat(dfd, name->name,
2004 flags | LOOKUP_REVAL, nd);
2006 if (likely(!retval))
2007 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2011 static int do_path_lookup(int dfd, const char *name,
2012 unsigned int flags, struct nameidata *nd)
2014 struct filename filename = { .name = name };
2016 return filename_lookup(dfd, &filename, flags, nd);
2019 /* does lookup, returns the object with parent locked */
2020 struct dentry *kern_path_locked(const char *name, struct path *path)
2022 struct nameidata nd;
2024 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2026 return ERR_PTR(err);
2027 if (nd.last_type != LAST_NORM) {
2029 return ERR_PTR(-EINVAL);
2031 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2032 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2034 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2042 int kern_path(const char *name, unsigned int flags, struct path *path)
2044 struct nameidata nd;
2045 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2050 EXPORT_SYMBOL(kern_path);
2053 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2054 * @dentry: pointer to dentry of the base directory
2055 * @mnt: pointer to vfs mount of the base directory
2056 * @name: pointer to file name
2057 * @flags: lookup flags
2058 * @path: pointer to struct path to fill
2060 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2061 const char *name, unsigned int flags,
2064 struct nameidata nd;
2066 nd.root.dentry = dentry;
2068 BUG_ON(flags & LOOKUP_PARENT);
2069 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2070 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2075 EXPORT_SYMBOL(vfs_path_lookup);
2078 * Restricted form of lookup. Doesn't follow links, single-component only,
2079 * needs parent already locked. Doesn't follow mounts.
2082 static struct dentry *lookup_hash(struct nameidata *nd)
2084 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2088 * lookup_one_len - filesystem helper to lookup single pathname component
2089 * @name: pathname component to lookup
2090 * @base: base directory to lookup from
2091 * @len: maximum length @len should be interpreted to
2093 * Note that this routine is purely a helper for filesystem usage and should
2094 * not be called by generic code. Also note that by using this function the
2095 * nameidata argument is passed to the filesystem methods and a filesystem
2096 * using this helper needs to be prepared for that.
2098 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2104 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2108 this.hash = full_name_hash(name, len);
2110 return ERR_PTR(-EACCES);
2112 if (unlikely(name[0] == '.')) {
2113 if (len < 2 || (len == 2 && name[1] == '.'))
2114 return ERR_PTR(-EACCES);
2118 c = *(const unsigned char *)name++;
2119 if (c == '/' || c == '\0')
2120 return ERR_PTR(-EACCES);
2123 * See if the low-level filesystem might want
2124 * to use its own hash..
2126 if (base->d_flags & DCACHE_OP_HASH) {
2127 int err = base->d_op->d_hash(base, &this);
2129 return ERR_PTR(err);
2132 err = inode_permission(base->d_inode, MAY_EXEC);
2134 return ERR_PTR(err);
2136 return __lookup_hash(&this, base, 0);
2138 EXPORT_SYMBOL(lookup_one_len);
2140 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2141 struct path *path, int *empty)
2143 struct nameidata nd;
2144 struct filename *tmp = getname_flags(name, flags, empty);
2145 int err = PTR_ERR(tmp);
2148 BUG_ON(flags & LOOKUP_PARENT);
2150 err = filename_lookup(dfd, tmp, flags, &nd);
2158 int user_path_at(int dfd, const char __user *name, unsigned flags,
2161 return user_path_at_empty(dfd, name, flags, path, NULL);
2163 EXPORT_SYMBOL(user_path_at);
2166 * NB: most callers don't do anything directly with the reference to the
2167 * to struct filename, but the nd->last pointer points into the name string
2168 * allocated by getname. So we must hold the reference to it until all
2169 * path-walking is complete.
2171 static struct filename *
2172 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2175 struct filename *s = getname(path);
2178 /* only LOOKUP_REVAL is allowed in extra flags */
2179 flags &= LOOKUP_REVAL;
2184 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2187 return ERR_PTR(error);
2194 * mountpoint_last - look up last component for umount
2195 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2196 * @path: pointer to container for result
2198 * This is a special lookup_last function just for umount. In this case, we
2199 * need to resolve the path without doing any revalidation.
2201 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2202 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2203 * in almost all cases, this lookup will be served out of the dcache. The only
2204 * cases where it won't are if nd->last refers to a symlink or the path is
2205 * bogus and it doesn't exist.
2208 * -error: if there was an error during lookup. This includes -ENOENT if the
2209 * lookup found a negative dentry. The nd->path reference will also be
2212 * 0: if we successfully resolved nd->path and found it to not to be a
2213 * symlink that needs to be followed. "path" will also be populated.
2214 * The nd->path reference will also be put.
2216 * 1: if we successfully resolved nd->last and found it to be a symlink
2217 * that needs to be followed. "path" will be populated with the path
2218 * to the link, and nd->path will *not* be put.
2221 mountpoint_last(struct nameidata *nd, struct path *path)
2224 struct dentry *dentry;
2225 struct dentry *dir = nd->path.dentry;
2227 /* If we're in rcuwalk, drop out of it to handle last component */
2228 if (nd->flags & LOOKUP_RCU) {
2229 if (unlazy_walk(nd, NULL)) {
2235 nd->flags &= ~LOOKUP_PARENT;
2237 if (unlikely(nd->last_type != LAST_NORM)) {
2238 error = handle_dots(nd, nd->last_type);
2241 dentry = dget(nd->path.dentry);
2245 mutex_lock(&dir->d_inode->i_mutex);
2246 dentry = d_lookup(dir, &nd->last);
2249 * No cached dentry. Mounted dentries are pinned in the cache,
2250 * so that means that this dentry is probably a symlink or the
2251 * path doesn't actually point to a mounted dentry.
2253 dentry = d_alloc(dir, &nd->last);
2256 mutex_unlock(&dir->d_inode->i_mutex);
2259 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2260 error = PTR_ERR(dentry);
2261 if (IS_ERR(dentry)) {
2262 mutex_unlock(&dir->d_inode->i_mutex);
2266 mutex_unlock(&dir->d_inode->i_mutex);
2269 if (!dentry->d_inode || d_is_negative(dentry)) {
2274 path->dentry = dentry;
2275 path->mnt = nd->path.mnt;
2276 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2287 * path_mountpoint - look up a path to be umounted
2288 * @dfd: directory file descriptor to start walk from
2289 * @name: full pathname to walk
2290 * @path: pointer to container for result
2291 * @flags: lookup flags
2293 * Look up the given name, but don't attempt to revalidate the last component.
2294 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2297 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2299 struct file *base = NULL;
2300 struct nameidata nd;
2303 err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2307 current->total_link_count = 0;
2308 err = link_path_walk(name, &nd);
2312 err = mountpoint_last(&nd, path);
2315 struct path link = *path;
2316 err = may_follow_link(&link, &nd);
2319 nd.flags |= LOOKUP_PARENT;
2320 err = follow_link(&link, &nd, &cookie);
2323 err = mountpoint_last(&nd, path);
2324 put_link(&nd, &link, cookie);
2330 if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2337 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2340 int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2341 if (unlikely(error == -ECHILD))
2342 error = path_mountpoint(dfd, s->name, path, flags);
2343 if (unlikely(error == -ESTALE))
2344 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2346 audit_inode(s, path->dentry, 0);
2351 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2352 * @dfd: directory file descriptor
2353 * @name: pathname from userland
2354 * @flags: lookup flags
2355 * @path: pointer to container to hold result
2357 * A umount is a special case for path walking. We're not actually interested
2358 * in the inode in this situation, and ESTALE errors can be a problem. We
2359 * simply want track down the dentry and vfsmount attached at the mountpoint
2360 * and avoid revalidating the last component.
2362 * Returns 0 and populates "path" on success.
2365 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2368 struct filename *s = getname(name);
2372 error = filename_mountpoint(dfd, s, path, flags);
2378 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2381 struct filename s = {.name = name};
2382 return filename_mountpoint(dfd, &s, path, flags);
2384 EXPORT_SYMBOL(kern_path_mountpoint);
2387 * It's inline, so penalty for filesystems that don't use sticky bit is
2390 static inline int check_sticky(struct inode *dir, struct inode *inode)
2392 kuid_t fsuid = current_fsuid();
2394 if (!(dir->i_mode & S_ISVTX))
2396 if (uid_eq(inode->i_uid, fsuid))
2398 if (uid_eq(dir->i_uid, fsuid))
2400 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2404 * Check whether we can remove a link victim from directory dir, check
2405 * whether the type of victim is right.
2406 * 1. We can't do it if dir is read-only (done in permission())
2407 * 2. We should have write and exec permissions on dir
2408 * 3. We can't remove anything from append-only dir
2409 * 4. We can't do anything with immutable dir (done in permission())
2410 * 5. If the sticky bit on dir is set we should either
2411 * a. be owner of dir, or
2412 * b. be owner of victim, or
2413 * c. have CAP_FOWNER capability
2414 * 6. If the victim is append-only or immutable we can't do antyhing with
2415 * links pointing to it.
2416 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2417 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2418 * 9. We can't remove a root or mountpoint.
2419 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2420 * nfs_async_unlink().
2422 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2424 struct inode *inode = victim->d_inode;
2427 if (d_is_negative(victim))
2431 BUG_ON(victim->d_parent->d_inode != dir);
2432 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2434 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2440 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2441 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2444 if (!d_is_dir(victim))
2446 if (IS_ROOT(victim))
2448 } else if (d_is_dir(victim))
2450 if (IS_DEADDIR(dir))
2452 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2457 /* Check whether we can create an object with dentry child in directory
2459 * 1. We can't do it if child already exists (open has special treatment for
2460 * this case, but since we are inlined it's OK)
2461 * 2. We can't do it if dir is read-only (done in permission())
2462 * 3. We should have write and exec permissions on dir
2463 * 4. We can't do it if dir is immutable (done in permission())
2465 static inline int may_create(struct inode *dir, struct dentry *child)
2467 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2470 if (IS_DEADDIR(dir))
2472 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2476 * p1 and p2 should be directories on the same fs.
2478 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2483 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2487 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2489 p = d_ancestor(p2, p1);
2491 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2492 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2496 p = d_ancestor(p1, p2);
2498 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2499 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2503 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2504 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2507 EXPORT_SYMBOL(lock_rename);
2509 void unlock_rename(struct dentry *p1, struct dentry *p2)
2511 mutex_unlock(&p1->d_inode->i_mutex);
2513 mutex_unlock(&p2->d_inode->i_mutex);
2514 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2517 EXPORT_SYMBOL(unlock_rename);
2519 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2522 int error = may_create(dir, dentry);
2526 if (!dir->i_op->create)
2527 return -EACCES; /* shouldn't it be ENOSYS? */
2530 error = security_inode_create(dir, dentry, mode);
2533 error = dir->i_op->create(dir, dentry, mode, want_excl);
2535 fsnotify_create(dir, dentry);
2538 EXPORT_SYMBOL(vfs_create);
2540 static int may_open(struct path *path, int acc_mode, int flag)
2542 struct dentry *dentry = path->dentry;
2543 struct inode *inode = dentry->d_inode;
2553 switch (inode->i_mode & S_IFMT) {
2557 if (acc_mode & MAY_WRITE)
2562 if (path->mnt->mnt_flags & MNT_NODEV)
2571 error = inode_permission(inode, acc_mode);
2576 * An append-only file must be opened in append mode for writing.
2578 if (IS_APPEND(inode)) {
2579 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2585 /* O_NOATIME can only be set by the owner or superuser */
2586 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2592 static int handle_truncate(struct file *filp)
2594 struct path *path = &filp->f_path;
2595 struct inode *inode = path->dentry->d_inode;
2596 int error = get_write_access(inode);
2600 * Refuse to truncate files with mandatory locks held on them.
2602 error = locks_verify_locked(filp);
2604 error = security_path_truncate(path);
2606 error = do_truncate(path->dentry, 0,
2607 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2610 put_write_access(inode);
2614 static inline int open_to_namei_flags(int flag)
2616 if ((flag & O_ACCMODE) == 3)
2621 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2623 int error = security_path_mknod(dir, dentry, mode, 0);
2627 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2631 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2635 * Attempt to atomically look up, create and open a file from a negative
2638 * Returns 0 if successful. The file will have been created and attached to
2639 * @file by the filesystem calling finish_open().
2641 * Returns 1 if the file was looked up only or didn't need creating. The
2642 * caller will need to perform the open themselves. @path will have been
2643 * updated to point to the new dentry. This may be negative.
2645 * Returns an error code otherwise.
2647 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2648 struct path *path, struct file *file,
2649 const struct open_flags *op,
2650 bool got_write, bool need_lookup,
2653 struct inode *dir = nd->path.dentry->d_inode;
2654 unsigned open_flag = open_to_namei_flags(op->open_flag);
2658 int create_error = 0;
2659 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2662 BUG_ON(dentry->d_inode);
2664 /* Don't create child dentry for a dead directory. */
2665 if (unlikely(IS_DEADDIR(dir))) {
2671 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2672 mode &= ~current_umask();
2674 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2676 open_flag &= ~O_TRUNC;
2679 * Checking write permission is tricky, bacuse we don't know if we are
2680 * going to actually need it: O_CREAT opens should work as long as the
2681 * file exists. But checking existence breaks atomicity. The trick is
2682 * to check access and if not granted clear O_CREAT from the flags.
2684 * Another problem is returing the "right" error value (e.g. for an
2685 * O_EXCL open we want to return EEXIST not EROFS).
2687 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2688 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2689 if (!(open_flag & O_CREAT)) {
2691 * No O_CREATE -> atomicity not a requirement -> fall
2692 * back to lookup + open
2695 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2696 /* Fall back and fail with the right error */
2697 create_error = -EROFS;
2700 /* No side effects, safe to clear O_CREAT */
2701 create_error = -EROFS;
2702 open_flag &= ~O_CREAT;
2706 if (open_flag & O_CREAT) {
2707 error = may_o_create(&nd->path, dentry, mode);
2709 create_error = error;
2710 if (open_flag & O_EXCL)
2712 open_flag &= ~O_CREAT;
2716 if (nd->flags & LOOKUP_DIRECTORY)
2717 open_flag |= O_DIRECTORY;
2719 file->f_path.dentry = DENTRY_NOT_SET;
2720 file->f_path.mnt = nd->path.mnt;
2721 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2724 if (create_error && error == -ENOENT)
2725 error = create_error;
2729 if (error) { /* returned 1, that is */
2730 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2734 if (file->f_path.dentry) {
2736 dentry = file->f_path.dentry;
2738 if (*opened & FILE_CREATED)
2739 fsnotify_create(dir, dentry);
2740 if (!dentry->d_inode) {
2741 WARN_ON(*opened & FILE_CREATED);
2743 error = create_error;
2747 if (excl && !(*opened & FILE_CREATED)) {
2756 * We didn't have the inode before the open, so check open permission
2759 acc_mode = op->acc_mode;
2760 if (*opened & FILE_CREATED) {
2761 WARN_ON(!(open_flag & O_CREAT));
2762 fsnotify_create(dir, dentry);
2763 acc_mode = MAY_OPEN;
2765 error = may_open(&file->f_path, acc_mode, open_flag);
2775 dentry = lookup_real(dir, dentry, nd->flags);
2777 return PTR_ERR(dentry);
2780 int open_flag = op->open_flag;
2782 error = create_error;
2783 if ((open_flag & O_EXCL)) {
2784 if (!dentry->d_inode)
2786 } else if (!dentry->d_inode) {
2788 } else if ((open_flag & O_TRUNC) &&
2789 S_ISREG(dentry->d_inode->i_mode)) {
2792 /* will fail later, go on to get the right error */
2796 path->dentry = dentry;
2797 path->mnt = nd->path.mnt;
2802 * Look up and maybe create and open the last component.
2804 * Must be called with i_mutex held on parent.
2806 * Returns 0 if the file was successfully atomically created (if necessary) and
2807 * opened. In this case the file will be returned attached to @file.
2809 * Returns 1 if the file was not completely opened at this time, though lookups
2810 * and creations will have been performed and the dentry returned in @path will
2811 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2812 * specified then a negative dentry may be returned.
2814 * An error code is returned otherwise.
2816 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2817 * cleared otherwise prior to returning.
2819 static int lookup_open(struct nameidata *nd, struct path *path,
2821 const struct open_flags *op,
2822 bool got_write, int *opened)
2824 struct dentry *dir = nd->path.dentry;
2825 struct inode *dir_inode = dir->d_inode;
2826 struct dentry *dentry;
2830 *opened &= ~FILE_CREATED;
2831 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2833 return PTR_ERR(dentry);
2835 /* Cached positive dentry: will open in f_op->open */
2836 if (!need_lookup && dentry->d_inode)
2839 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2840 return atomic_open(nd, dentry, path, file, op, got_write,
2841 need_lookup, opened);
2845 BUG_ON(dentry->d_inode);
2847 dentry = lookup_real(dir_inode, dentry, nd->flags);
2849 return PTR_ERR(dentry);
2852 /* Negative dentry, just create the file */
2853 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2854 umode_t mode = op->mode;
2855 if (!IS_POSIXACL(dir->d_inode))
2856 mode &= ~current_umask();
2858 * This write is needed to ensure that a
2859 * rw->ro transition does not occur between
2860 * the time when the file is created and when
2861 * a permanent write count is taken through
2862 * the 'struct file' in finish_open().
2868 *opened |= FILE_CREATED;
2869 error = security_path_mknod(&nd->path, dentry, mode, 0);
2872 error = vfs_create(dir->d_inode, dentry, mode,
2873 nd->flags & LOOKUP_EXCL);
2878 path->dentry = dentry;
2879 path->mnt = nd->path.mnt;
2888 * Handle the last step of open()
2890 static int do_last(struct nameidata *nd, struct path *path,
2891 struct file *file, const struct open_flags *op,
2892 int *opened, struct filename *name)
2894 struct dentry *dir = nd->path.dentry;
2895 int open_flag = op->open_flag;
2896 bool will_truncate = (open_flag & O_TRUNC) != 0;
2897 bool got_write = false;
2898 int acc_mode = op->acc_mode;
2899 struct inode *inode;
2900 bool symlink_ok = false;
2901 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2902 bool retried = false;
2905 nd->flags &= ~LOOKUP_PARENT;
2906 nd->flags |= op->intent;
2908 if (nd->last_type != LAST_NORM) {
2909 error = handle_dots(nd, nd->last_type);
2915 if (!(open_flag & O_CREAT)) {
2916 if (nd->last.name[nd->last.len])
2917 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2918 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2920 /* we _can_ be in RCU mode here */
2921 error = lookup_fast(nd, path, &inode);
2928 BUG_ON(nd->inode != dir->d_inode);
2930 /* create side of things */
2932 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2933 * has been cleared when we got to the last component we are
2936 error = complete_walk(nd);
2940 audit_inode(name, dir, LOOKUP_PARENT);
2942 /* trailing slashes? */
2943 if (nd->last.name[nd->last.len])
2948 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2949 error = mnt_want_write(nd->path.mnt);
2953 * do _not_ fail yet - we might not need that or fail with
2954 * a different error; let lookup_open() decide; we'll be
2955 * dropping this one anyway.
2958 mutex_lock(&dir->d_inode->i_mutex);
2959 error = lookup_open(nd, path, file, op, got_write, opened);
2960 mutex_unlock(&dir->d_inode->i_mutex);
2966 if ((*opened & FILE_CREATED) ||
2967 !S_ISREG(file_inode(file)->i_mode))
2968 will_truncate = false;
2970 audit_inode(name, file->f_path.dentry, 0);
2974 if (*opened & FILE_CREATED) {
2975 /* Don't check for write permission, don't truncate */
2976 open_flag &= ~O_TRUNC;
2977 will_truncate = false;
2978 acc_mode = MAY_OPEN;
2979 path_to_nameidata(path, nd);
2980 goto finish_open_created;
2984 * create/update audit record if it already exists.
2986 if (d_is_positive(path->dentry))
2987 audit_inode(name, path->dentry, 0);
2990 * If atomic_open() acquired write access it is dropped now due to
2991 * possible mount and symlink following (this might be optimized away if
2995 mnt_drop_write(nd->path.mnt);
3000 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3003 error = follow_managed(path, nd->flags);
3008 nd->flags |= LOOKUP_JUMPED;
3010 BUG_ON(nd->flags & LOOKUP_RCU);
3011 inode = path->dentry->d_inode;
3013 /* we _can_ be in RCU mode here */
3015 if (!inode || d_is_negative(path->dentry)) {
3016 path_to_nameidata(path, nd);
3020 if (should_follow_link(path->dentry, !symlink_ok)) {
3021 if (nd->flags & LOOKUP_RCU) {
3022 if (unlikely(unlazy_walk(nd, path->dentry))) {
3027 BUG_ON(inode != path->dentry->d_inode);
3031 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3032 path_to_nameidata(path, nd);
3034 save_parent.dentry = nd->path.dentry;
3035 save_parent.mnt = mntget(path->mnt);
3036 nd->path.dentry = path->dentry;
3040 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3042 error = complete_walk(nd);
3044 path_put(&save_parent);
3047 audit_inode(name, nd->path.dentry, 0);
3049 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3052 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3054 if (!S_ISREG(nd->inode->i_mode))
3055 will_truncate = false;
3057 if (will_truncate) {
3058 error = mnt_want_write(nd->path.mnt);
3063 finish_open_created:
3064 error = may_open(&nd->path, acc_mode, open_flag);
3067 file->f_path.mnt = nd->path.mnt;
3068 error = finish_open(file, nd->path.dentry, NULL, opened);
3070 if (error == -EOPENSTALE)
3075 error = open_check_o_direct(file);
3078 error = ima_file_check(file, op->acc_mode, *opened);
3082 if (will_truncate) {
3083 error = handle_truncate(file);
3089 mnt_drop_write(nd->path.mnt);
3090 path_put(&save_parent);
3095 path_put_conditional(path, nd);
3102 /* If no saved parent or already retried then can't retry */
3103 if (!save_parent.dentry || retried)
3106 BUG_ON(save_parent.dentry != dir);
3107 path_put(&nd->path);
3108 nd->path = save_parent;
3109 nd->inode = dir->d_inode;
3110 save_parent.mnt = NULL;
3111 save_parent.dentry = NULL;
3113 mnt_drop_write(nd->path.mnt);
3120 static int do_tmpfile(int dfd, struct filename *pathname,
3121 struct nameidata *nd, int flags,
3122 const struct open_flags *op,
3123 struct file *file, int *opened)
3125 static const struct qstr name = QSTR_INIT("/", 1);
3126 struct dentry *dentry, *child;
3128 int error = path_lookupat(dfd, pathname->name,
3129 flags | LOOKUP_DIRECTORY, nd);
3130 if (unlikely(error))
3132 error = mnt_want_write(nd->path.mnt);
3133 if (unlikely(error))
3135 /* we want directory to be writable */
3136 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3139 dentry = nd->path.dentry;
3140 dir = dentry->d_inode;
3141 if (!dir->i_op->tmpfile) {
3142 error = -EOPNOTSUPP;
3145 child = d_alloc(dentry, &name);
3146 if (unlikely(!child)) {
3150 nd->flags &= ~LOOKUP_DIRECTORY;
3151 nd->flags |= op->intent;
3152 dput(nd->path.dentry);
3153 nd->path.dentry = child;
3154 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3157 audit_inode(pathname, nd->path.dentry, 0);
3158 error = may_open(&nd->path, op->acc_mode, op->open_flag);
3161 file->f_path.mnt = nd->path.mnt;
3162 error = finish_open(file, nd->path.dentry, NULL, opened);
3165 error = open_check_o_direct(file);
3168 } else if (!(op->open_flag & O_EXCL)) {
3169 struct inode *inode = file_inode(file);
3170 spin_lock(&inode->i_lock);
3171 inode->i_state |= I_LINKABLE;
3172 spin_unlock(&inode->i_lock);
3175 mnt_drop_write(nd->path.mnt);
3177 path_put(&nd->path);
3181 static struct file *path_openat(int dfd, struct filename *pathname,
3182 struct nameidata *nd, const struct open_flags *op, int flags)
3184 struct file *base = NULL;
3190 file = get_empty_filp();
3194 file->f_flags = op->open_flag;
3196 if (unlikely(file->f_flags & __O_TMPFILE)) {
3197 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3201 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
3202 if (unlikely(error))
3205 current->total_link_count = 0;
3206 error = link_path_walk(pathname->name, nd);
3207 if (unlikely(error))
3210 error = do_last(nd, &path, file, op, &opened, pathname);
3211 while (unlikely(error > 0)) { /* trailing symlink */
3212 struct path link = path;
3214 if (!(nd->flags & LOOKUP_FOLLOW)) {
3215 path_put_conditional(&path, nd);
3216 path_put(&nd->path);
3220 error = may_follow_link(&link, nd);
3221 if (unlikely(error))
3223 nd->flags |= LOOKUP_PARENT;
3224 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3225 error = follow_link(&link, nd, &cookie);
3226 if (unlikely(error))
3228 error = do_last(nd, &path, file, op, &opened, pathname);
3229 put_link(nd, &link, cookie);
3232 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3233 path_put(&nd->root);
3236 if (!(opened & FILE_OPENED)) {
3240 if (unlikely(error)) {
3241 if (error == -EOPENSTALE) {
3242 if (flags & LOOKUP_RCU)
3247 file = ERR_PTR(error);
3252 struct file *do_filp_open(int dfd, struct filename *pathname,
3253 const struct open_flags *op)
3255 struct nameidata nd;
3256 int flags = op->lookup_flags;
3259 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3260 if (unlikely(filp == ERR_PTR(-ECHILD)))
3261 filp = path_openat(dfd, pathname, &nd, op, flags);
3262 if (unlikely(filp == ERR_PTR(-ESTALE)))
3263 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3267 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3268 const char *name, const struct open_flags *op)
3270 struct nameidata nd;
3272 struct filename filename = { .name = name };
3273 int flags = op->lookup_flags | LOOKUP_ROOT;
3276 nd.root.dentry = dentry;
3278 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3279 return ERR_PTR(-ELOOP);
3281 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3282 if (unlikely(file == ERR_PTR(-ECHILD)))
3283 file = path_openat(-1, &filename, &nd, op, flags);
3284 if (unlikely(file == ERR_PTR(-ESTALE)))
3285 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3289 struct dentry *kern_path_create(int dfd, const char *pathname,
3290 struct path *path, unsigned int lookup_flags)
3292 struct dentry *dentry = ERR_PTR(-EEXIST);
3293 struct nameidata nd;
3296 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3299 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3300 * other flags passed in are ignored!
3302 lookup_flags &= LOOKUP_REVAL;
3304 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3306 return ERR_PTR(error);
3309 * Yucky last component or no last component at all?
3310 * (foo/., foo/.., /////)
3312 if (nd.last_type != LAST_NORM)
3314 nd.flags &= ~LOOKUP_PARENT;
3315 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3317 /* don't fail immediately if it's r/o, at least try to report other errors */
3318 err2 = mnt_want_write(nd.path.mnt);
3320 * Do the final lookup.
3322 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3323 dentry = lookup_hash(&nd);
3328 if (d_is_positive(dentry))
3332 * Special case - lookup gave negative, but... we had foo/bar/
3333 * From the vfs_mknod() POV we just have a negative dentry -
3334 * all is fine. Let's be bastards - you had / on the end, you've
3335 * been asking for (non-existent) directory. -ENOENT for you.
3337 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3341 if (unlikely(err2)) {
3349 dentry = ERR_PTR(error);
3351 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3353 mnt_drop_write(nd.path.mnt);
3358 EXPORT_SYMBOL(kern_path_create);
3360 void done_path_create(struct path *path, struct dentry *dentry)
3363 mutex_unlock(&path->dentry->d_inode->i_mutex);
3364 mnt_drop_write(path->mnt);
3367 EXPORT_SYMBOL(done_path_create);
3369 struct dentry *user_path_create(int dfd, const char __user *pathname,
3370 struct path *path, unsigned int lookup_flags)
3372 struct filename *tmp = getname(pathname);
3375 return ERR_CAST(tmp);
3376 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3380 EXPORT_SYMBOL(user_path_create);
3382 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3384 int error = may_create(dir, dentry);
3389 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3392 if (!dir->i_op->mknod)
3395 error = devcgroup_inode_mknod(mode, dev);
3399 error = security_inode_mknod(dir, dentry, mode, dev);
3403 error = dir->i_op->mknod(dir, dentry, mode, dev);
3405 fsnotify_create(dir, dentry);
3408 EXPORT_SYMBOL(vfs_mknod);
3410 static int may_mknod(umode_t mode)
3412 switch (mode & S_IFMT) {
3418 case 0: /* zero mode translates to S_IFREG */
3427 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3430 struct dentry *dentry;
3433 unsigned int lookup_flags = 0;
3435 error = may_mknod(mode);
3439 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3441 return PTR_ERR(dentry);
3443 if (!IS_POSIXACL(path.dentry->d_inode))
3444 mode &= ~current_umask();
3445 error = security_path_mknod(&path, dentry, mode, dev);
3448 switch (mode & S_IFMT) {
3449 case 0: case S_IFREG:
3450 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3452 case S_IFCHR: case S_IFBLK:
3453 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3454 new_decode_dev(dev));
3456 case S_IFIFO: case S_IFSOCK:
3457 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3461 done_path_create(&path, dentry);
3462 if (retry_estale(error, lookup_flags)) {
3463 lookup_flags |= LOOKUP_REVAL;
3469 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3471 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3474 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3476 int error = may_create(dir, dentry);
3477 unsigned max_links = dir->i_sb->s_max_links;
3482 if (!dir->i_op->mkdir)
3485 mode &= (S_IRWXUGO|S_ISVTX);
3486 error = security_inode_mkdir(dir, dentry, mode);
3490 if (max_links && dir->i_nlink >= max_links)
3493 error = dir->i_op->mkdir(dir, dentry, mode);
3495 fsnotify_mkdir(dir, dentry);
3498 EXPORT_SYMBOL(vfs_mkdir);
3500 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3502 struct dentry *dentry;
3505 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3508 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3510 return PTR_ERR(dentry);
3512 if (!IS_POSIXACL(path.dentry->d_inode))
3513 mode &= ~current_umask();
3514 error = security_path_mkdir(&path, dentry, mode);
3516 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3517 done_path_create(&path, dentry);
3518 if (retry_estale(error, lookup_flags)) {
3519 lookup_flags |= LOOKUP_REVAL;
3525 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3527 return sys_mkdirat(AT_FDCWD, pathname, mode);
3531 * The dentry_unhash() helper will try to drop the dentry early: we
3532 * should have a usage count of 1 if we're the only user of this
3533 * dentry, and if that is true (possibly after pruning the dcache),
3534 * then we drop the dentry now.
3536 * A low-level filesystem can, if it choses, legally
3539 * if (!d_unhashed(dentry))
3542 * if it cannot handle the case of removing a directory
3543 * that is still in use by something else..
3545 void dentry_unhash(struct dentry *dentry)
3547 shrink_dcache_parent(dentry);
3548 spin_lock(&dentry->d_lock);
3549 if (dentry->d_lockref.count == 1)
3551 spin_unlock(&dentry->d_lock);
3553 EXPORT_SYMBOL(dentry_unhash);
3555 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3557 int error = may_delete(dir, dentry, 1);
3562 if (!dir->i_op->rmdir)
3566 mutex_lock(&dentry->d_inode->i_mutex);
3569 if (is_local_mountpoint(dentry))
3572 error = security_inode_rmdir(dir, dentry);
3576 shrink_dcache_parent(dentry);
3577 error = dir->i_op->rmdir(dir, dentry);
3581 dentry->d_inode->i_flags |= S_DEAD;
3583 detach_mounts(dentry);
3586 mutex_unlock(&dentry->d_inode->i_mutex);
3592 EXPORT_SYMBOL(vfs_rmdir);
3594 static long do_rmdir(int dfd, const char __user *pathname)
3597 struct filename *name;
3598 struct dentry *dentry;
3599 struct nameidata nd;
3600 unsigned int lookup_flags = 0;
3602 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3604 return PTR_ERR(name);
3606 switch(nd.last_type) {
3618 nd.flags &= ~LOOKUP_PARENT;
3619 error = mnt_want_write(nd.path.mnt);
3623 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3624 dentry = lookup_hash(&nd);
3625 error = PTR_ERR(dentry);
3628 if (!dentry->d_inode) {
3632 error = security_path_rmdir(&nd.path, dentry);
3635 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3639 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3640 mnt_drop_write(nd.path.mnt);
3644 if (retry_estale(error, lookup_flags)) {
3645 lookup_flags |= LOOKUP_REVAL;
3651 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3653 return do_rmdir(AT_FDCWD, pathname);
3657 * vfs_unlink - unlink a filesystem object
3658 * @dir: parent directory
3660 * @delegated_inode: returns victim inode, if the inode is delegated.
3662 * The caller must hold dir->i_mutex.
3664 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3665 * return a reference to the inode in delegated_inode. The caller
3666 * should then break the delegation on that inode and retry. Because
3667 * breaking a delegation may take a long time, the caller should drop
3668 * dir->i_mutex before doing so.
3670 * Alternatively, a caller may pass NULL for delegated_inode. This may
3671 * be appropriate for callers that expect the underlying filesystem not
3672 * to be NFS exported.
3674 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3676 struct inode *target = dentry->d_inode;
3677 int error = may_delete(dir, dentry, 0);
3682 if (!dir->i_op->unlink)
3685 mutex_lock(&target->i_mutex);
3686 if (is_local_mountpoint(dentry))
3689 error = security_inode_unlink(dir, dentry);
3691 error = try_break_deleg(target, delegated_inode);
3694 error = dir->i_op->unlink(dir, dentry);
3697 detach_mounts(dentry);
3702 mutex_unlock(&target->i_mutex);
3704 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3705 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3706 fsnotify_link_count(target);
3712 EXPORT_SYMBOL(vfs_unlink);
3715 * Make sure that the actual truncation of the file will occur outside its
3716 * directory's i_mutex. Truncate can take a long time if there is a lot of
3717 * writeout happening, and we don't want to prevent access to the directory
3718 * while waiting on the I/O.
3720 static long do_unlinkat(int dfd, const char __user *pathname)
3723 struct filename *name;
3724 struct dentry *dentry;
3725 struct nameidata nd;
3726 struct inode *inode = NULL;
3727 struct inode *delegated_inode = NULL;
3728 unsigned int lookup_flags = 0;
3730 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3732 return PTR_ERR(name);
3735 if (nd.last_type != LAST_NORM)
3738 nd.flags &= ~LOOKUP_PARENT;
3739 error = mnt_want_write(nd.path.mnt);
3743 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3744 dentry = lookup_hash(&nd);
3745 error = PTR_ERR(dentry);
3746 if (!IS_ERR(dentry)) {
3747 /* Why not before? Because we want correct error value */
3748 if (nd.last.name[nd.last.len])
3750 inode = dentry->d_inode;
3751 if (d_is_negative(dentry))
3754 error = security_path_unlink(&nd.path, dentry);
3757 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3761 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3763 iput(inode); /* truncate the inode here */
3765 if (delegated_inode) {
3766 error = break_deleg_wait(&delegated_inode);
3770 mnt_drop_write(nd.path.mnt);
3774 if (retry_estale(error, lookup_flags)) {
3775 lookup_flags |= LOOKUP_REVAL;
3782 if (d_is_negative(dentry))
3784 else if (d_is_dir(dentry))
3791 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3793 if ((flag & ~AT_REMOVEDIR) != 0)
3796 if (flag & AT_REMOVEDIR)
3797 return do_rmdir(dfd, pathname);
3799 return do_unlinkat(dfd, pathname);
3802 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3804 return do_unlinkat(AT_FDCWD, pathname);
3807 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3809 int error = may_create(dir, dentry);
3814 if (!dir->i_op->symlink)
3817 error = security_inode_symlink(dir, dentry, oldname);
3821 error = dir->i_op->symlink(dir, dentry, oldname);
3823 fsnotify_create(dir, dentry);
3826 EXPORT_SYMBOL(vfs_symlink);
3828 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3829 int, newdfd, const char __user *, newname)
3832 struct filename *from;
3833 struct dentry *dentry;
3835 unsigned int lookup_flags = 0;
3837 from = getname(oldname);
3839 return PTR_ERR(from);
3841 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3842 error = PTR_ERR(dentry);
3846 error = security_path_symlink(&path, dentry, from->name);
3848 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3849 done_path_create(&path, dentry);
3850 if (retry_estale(error, lookup_flags)) {
3851 lookup_flags |= LOOKUP_REVAL;
3859 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3861 return sys_symlinkat(oldname, AT_FDCWD, newname);
3865 * vfs_link - create a new link
3866 * @old_dentry: object to be linked
3868 * @new_dentry: where to create the new link
3869 * @delegated_inode: returns inode needing a delegation break
3871 * The caller must hold dir->i_mutex
3873 * If vfs_link discovers a delegation on the to-be-linked file in need
3874 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3875 * inode in delegated_inode. The caller should then break the delegation
3876 * and retry. Because breaking a delegation may take a long time, the
3877 * caller should drop the i_mutex before doing so.
3879 * Alternatively, a caller may pass NULL for delegated_inode. This may
3880 * be appropriate for callers that expect the underlying filesystem not
3881 * to be NFS exported.
3883 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3885 struct inode *inode = old_dentry->d_inode;
3886 unsigned max_links = dir->i_sb->s_max_links;
3892 error = may_create(dir, new_dentry);
3896 if (dir->i_sb != inode->i_sb)
3900 * A link to an append-only or immutable file cannot be created.
3902 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3904 if (!dir->i_op->link)
3906 if (S_ISDIR(inode->i_mode))
3909 error = security_inode_link(old_dentry, dir, new_dentry);
3913 mutex_lock(&inode->i_mutex);
3914 /* Make sure we don't allow creating hardlink to an unlinked file */
3915 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3917 else if (max_links && inode->i_nlink >= max_links)
3920 error = try_break_deleg(inode, delegated_inode);
3922 error = dir->i_op->link(old_dentry, dir, new_dentry);
3925 if (!error && (inode->i_state & I_LINKABLE)) {
3926 spin_lock(&inode->i_lock);
3927 inode->i_state &= ~I_LINKABLE;
3928 spin_unlock(&inode->i_lock);
3930 mutex_unlock(&inode->i_mutex);
3932 fsnotify_link(dir, inode, new_dentry);
3935 EXPORT_SYMBOL(vfs_link);
3938 * Hardlinks are often used in delicate situations. We avoid
3939 * security-related surprises by not following symlinks on the
3942 * We don't follow them on the oldname either to be compatible
3943 * with linux 2.0, and to avoid hard-linking to directories
3944 * and other special files. --ADM
3946 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3947 int, newdfd, const char __user *, newname, int, flags)
3949 struct dentry *new_dentry;
3950 struct path old_path, new_path;
3951 struct inode *delegated_inode = NULL;
3955 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3958 * To use null names we require CAP_DAC_READ_SEARCH
3959 * This ensures that not everyone will be able to create
3960 * handlink using the passed filedescriptor.
3962 if (flags & AT_EMPTY_PATH) {
3963 if (!capable(CAP_DAC_READ_SEARCH))
3968 if (flags & AT_SYMLINK_FOLLOW)
3969 how |= LOOKUP_FOLLOW;
3971 error = user_path_at(olddfd, oldname, how, &old_path);
3975 new_dentry = user_path_create(newdfd, newname, &new_path,
3976 (how & LOOKUP_REVAL));
3977 error = PTR_ERR(new_dentry);
3978 if (IS_ERR(new_dentry))
3982 if (old_path.mnt != new_path.mnt)
3984 error = may_linkat(&old_path);
3985 if (unlikely(error))
3987 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3990 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
3992 done_path_create(&new_path, new_dentry);
3993 if (delegated_inode) {
3994 error = break_deleg_wait(&delegated_inode);
3996 path_put(&old_path);
4000 if (retry_estale(error, how)) {
4001 path_put(&old_path);
4002 how |= LOOKUP_REVAL;
4006 path_put(&old_path);
4011 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4013 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4017 * vfs_rename - rename a filesystem object
4018 * @old_dir: parent of source
4019 * @old_dentry: source
4020 * @new_dir: parent of destination
4021 * @new_dentry: destination
4022 * @delegated_inode: returns an inode needing a delegation break
4023 * @flags: rename flags
4025 * The caller must hold multiple mutexes--see lock_rename()).
4027 * If vfs_rename discovers a delegation in need of breaking at either
4028 * the source or destination, it will return -EWOULDBLOCK and return a
4029 * reference to the inode in delegated_inode. The caller should then
4030 * break the delegation and retry. Because breaking a delegation may
4031 * take a long time, the caller should drop all locks before doing
4034 * Alternatively, a caller may pass NULL for delegated_inode. This may
4035 * be appropriate for callers that expect the underlying filesystem not
4036 * to be NFS exported.
4038 * The worst of all namespace operations - renaming directory. "Perverted"
4039 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4041 * a) we can get into loop creation.
4042 * b) race potential - two innocent renames can create a loop together.
4043 * That's where 4.4 screws up. Current fix: serialization on
4044 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4046 * c) we have to lock _four_ objects - parents and victim (if it exists),
4047 * and source (if it is not a directory).
4048 * And that - after we got ->i_mutex on parents (until then we don't know
4049 * whether the target exists). Solution: try to be smart with locking
4050 * order for inodes. We rely on the fact that tree topology may change
4051 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4052 * move will be locked. Thus we can rank directories by the tree
4053 * (ancestors first) and rank all non-directories after them.
4054 * That works since everybody except rename does "lock parent, lookup,
4055 * lock child" and rename is under ->s_vfs_rename_mutex.
4056 * HOWEVER, it relies on the assumption that any object with ->lookup()
4057 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4058 * we'd better make sure that there's no link(2) for them.
4059 * d) conversion from fhandle to dentry may come in the wrong moment - when
4060 * we are removing the target. Solution: we will have to grab ->i_mutex
4061 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4062 * ->i_mutex on parents, which works but leads to some truly excessive
4065 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4066 struct inode *new_dir, struct dentry *new_dentry,
4067 struct inode **delegated_inode, unsigned int flags)
4070 bool is_dir = d_is_dir(old_dentry);
4071 const unsigned char *old_name;
4072 struct inode *source = old_dentry->d_inode;
4073 struct inode *target = new_dentry->d_inode;
4074 bool new_is_dir = false;
4075 unsigned max_links = new_dir->i_sb->s_max_links;
4077 if (source == target)
4080 error = may_delete(old_dir, old_dentry, is_dir);
4085 error = may_create(new_dir, new_dentry);
4087 new_is_dir = d_is_dir(new_dentry);
4089 if (!(flags & RENAME_EXCHANGE))
4090 error = may_delete(new_dir, new_dentry, is_dir);
4092 error = may_delete(new_dir, new_dentry, new_is_dir);
4097 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4100 if (flags && !old_dir->i_op->rename2)
4104 * If we are going to change the parent - check write permissions,
4105 * we'll need to flip '..'.
4107 if (new_dir != old_dir) {
4109 error = inode_permission(source, MAY_WRITE);
4113 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4114 error = inode_permission(target, MAY_WRITE);
4120 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4125 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4127 if (!is_dir || (flags & RENAME_EXCHANGE))
4128 lock_two_nondirectories(source, target);
4130 mutex_lock(&target->i_mutex);
4133 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4136 if (max_links && new_dir != old_dir) {
4138 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4140 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4141 old_dir->i_nlink >= max_links)
4144 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4145 shrink_dcache_parent(new_dentry);
4147 error = try_break_deleg(source, delegated_inode);
4151 if (target && !new_is_dir) {
4152 error = try_break_deleg(target, delegated_inode);
4156 if (!old_dir->i_op->rename2) {
4157 error = old_dir->i_op->rename(old_dir, old_dentry,
4158 new_dir, new_dentry);
4160 WARN_ON(old_dir->i_op->rename != NULL);
4161 error = old_dir->i_op->rename2(old_dir, old_dentry,
4162 new_dir, new_dentry, flags);
4167 if (!(flags & RENAME_EXCHANGE) && target) {
4169 target->i_flags |= S_DEAD;
4170 dont_mount(new_dentry);
4171 detach_mounts(new_dentry);
4173 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4174 if (!(flags & RENAME_EXCHANGE))
4175 d_move(old_dentry, new_dentry);
4177 d_exchange(old_dentry, new_dentry);
4180 if (!is_dir || (flags & RENAME_EXCHANGE))
4181 unlock_two_nondirectories(source, target);
4183 mutex_unlock(&target->i_mutex);
4186 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4187 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4188 if (flags & RENAME_EXCHANGE) {
4189 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4190 new_is_dir, NULL, new_dentry);
4193 fsnotify_oldname_free(old_name);
4197 EXPORT_SYMBOL(vfs_rename);
4199 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4200 int, newdfd, const char __user *, newname, unsigned int, flags)
4202 struct dentry *old_dir, *new_dir;
4203 struct dentry *old_dentry, *new_dentry;
4204 struct dentry *trap;
4205 struct nameidata oldnd, newnd;
4206 struct inode *delegated_inode = NULL;
4207 struct filename *from;
4208 struct filename *to;
4209 unsigned int lookup_flags = 0;
4210 bool should_retry = false;
4213 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
4216 if ((flags & RENAME_NOREPLACE) && (flags & RENAME_EXCHANGE))
4220 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4222 error = PTR_ERR(from);
4226 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4228 error = PTR_ERR(to);
4233 if (oldnd.path.mnt != newnd.path.mnt)
4236 old_dir = oldnd.path.dentry;
4238 if (oldnd.last_type != LAST_NORM)
4241 new_dir = newnd.path.dentry;
4242 if (flags & RENAME_NOREPLACE)
4244 if (newnd.last_type != LAST_NORM)
4247 error = mnt_want_write(oldnd.path.mnt);
4251 oldnd.flags &= ~LOOKUP_PARENT;
4252 newnd.flags &= ~LOOKUP_PARENT;
4253 if (!(flags & RENAME_EXCHANGE))
4254 newnd.flags |= LOOKUP_RENAME_TARGET;
4257 trap = lock_rename(new_dir, old_dir);
4259 old_dentry = lookup_hash(&oldnd);
4260 error = PTR_ERR(old_dentry);
4261 if (IS_ERR(old_dentry))
4263 /* source must exist */
4265 if (d_is_negative(old_dentry))
4267 new_dentry = lookup_hash(&newnd);
4268 error = PTR_ERR(new_dentry);
4269 if (IS_ERR(new_dentry))
4272 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4274 if (flags & RENAME_EXCHANGE) {
4276 if (d_is_negative(new_dentry))
4279 if (!d_is_dir(new_dentry)) {
4281 if (newnd.last.name[newnd.last.len])
4285 /* unless the source is a directory trailing slashes give -ENOTDIR */
4286 if (!d_is_dir(old_dentry)) {
4288 if (oldnd.last.name[oldnd.last.len])
4290 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
4293 /* source should not be ancestor of target */
4295 if (old_dentry == trap)
4297 /* target should not be an ancestor of source */
4298 if (!(flags & RENAME_EXCHANGE))
4300 if (new_dentry == trap)
4303 error = security_path_rename(&oldnd.path, old_dentry,
4304 &newnd.path, new_dentry, flags);
4307 error = vfs_rename(old_dir->d_inode, old_dentry,
4308 new_dir->d_inode, new_dentry,
4309 &delegated_inode, flags);
4315 unlock_rename(new_dir, old_dir);
4316 if (delegated_inode) {
4317 error = break_deleg_wait(&delegated_inode);
4321 mnt_drop_write(oldnd.path.mnt);
4323 if (retry_estale(error, lookup_flags))
4324 should_retry = true;
4325 path_put(&newnd.path);
4328 path_put(&oldnd.path);
4331 should_retry = false;
4332 lookup_flags |= LOOKUP_REVAL;
4339 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4340 int, newdfd, const char __user *, newname)
4342 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4345 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4347 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4350 int readlink_copy(char __user *buffer, int buflen, const char *link)
4352 int len = PTR_ERR(link);
4357 if (len > (unsigned) buflen)
4359 if (copy_to_user(buffer, link, len))
4364 EXPORT_SYMBOL(readlink_copy);
4367 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4368 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4369 * using) it for any given inode is up to filesystem.
4371 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4373 struct nameidata nd;
4378 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4380 return PTR_ERR(cookie);
4382 res = readlink_copy(buffer, buflen, nd_get_link(&nd));
4383 if (dentry->d_inode->i_op->put_link)
4384 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4387 EXPORT_SYMBOL(generic_readlink);
4389 /* get the link contents into pagecache */
4390 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4394 struct address_space *mapping = dentry->d_inode->i_mapping;
4395 page = read_mapping_page(mapping, 0, NULL);
4400 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4404 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4406 struct page *page = NULL;
4407 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4410 page_cache_release(page);
4414 EXPORT_SYMBOL(page_readlink);
4416 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4418 struct page *page = NULL;
4419 nd_set_link(nd, page_getlink(dentry, &page));
4422 EXPORT_SYMBOL(page_follow_link_light);
4424 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4426 struct page *page = cookie;
4430 page_cache_release(page);
4433 EXPORT_SYMBOL(page_put_link);
4436 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4438 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4440 struct address_space *mapping = inode->i_mapping;
4445 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4447 flags |= AOP_FLAG_NOFS;
4450 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4451 flags, &page, &fsdata);
4455 kaddr = kmap_atomic(page);
4456 memcpy(kaddr, symname, len-1);
4457 kunmap_atomic(kaddr);
4459 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4466 mark_inode_dirty(inode);
4471 EXPORT_SYMBOL(__page_symlink);
4473 int page_symlink(struct inode *inode, const char *symname, int len)
4475 return __page_symlink(inode, symname, len,
4476 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4478 EXPORT_SYMBOL(page_symlink);
4480 const struct inode_operations page_symlink_inode_operations = {
4481 .readlink = generic_readlink,
4482 .follow_link = page_follow_link_light,
4483 .put_link = page_put_link,
4485 EXPORT_SYMBOL(page_symlink_inode_operations);