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))
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);
419 EXPORT_SYMBOL(__inode_permission);
422 * sb_permission - Check superblock-level permissions
423 * @sb: Superblock of inode to check permission on
424 * @inode: Inode to check permission on
425 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
427 * Separate out file-system wide checks from inode-specific permission checks.
429 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
431 if (unlikely(mask & MAY_WRITE)) {
432 umode_t mode = inode->i_mode;
434 /* Nobody gets write access to a read-only fs. */
435 if ((sb->s_flags & MS_RDONLY) &&
436 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
443 * inode_permission - Check for access rights to a given inode
444 * @inode: Inode to check permission on
445 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
447 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
448 * this, letting us set arbitrary permissions for filesystem access without
449 * changing the "normal" UIDs which are used for other things.
451 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
453 int inode_permission(struct inode *inode, int mask)
457 retval = sb_permission(inode->i_sb, inode, mask);
460 return __inode_permission(inode, mask);
462 EXPORT_SYMBOL(inode_permission);
465 * path_get - get a reference to a path
466 * @path: path to get the reference to
468 * Given a path increment the reference count to the dentry and the vfsmount.
470 void path_get(const struct path *path)
475 EXPORT_SYMBOL(path_get);
478 * path_put - put a reference to a path
479 * @path: path to put the reference to
481 * Given a path decrement the reference count to the dentry and the vfsmount.
483 void path_put(const struct path *path)
488 EXPORT_SYMBOL(path_put);
494 struct inode *inode; /* path.dentry.d_inode */
500 char *saved_names[MAX_NESTED_LINKS + 1];
504 * Path walking has 2 modes, rcu-walk and ref-walk (see
505 * Documentation/filesystems/path-lookup.txt). In situations when we can't
506 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
507 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
508 * mode. Refcounts are grabbed at the last known good point before rcu-walk
509 * got stuck, so ref-walk may continue from there. If this is not successful
510 * (eg. a seqcount has changed), then failure is returned and it's up to caller
511 * to restart the path walk from the beginning in ref-walk mode.
515 * unlazy_walk - try to switch to ref-walk mode.
516 * @nd: nameidata pathwalk data
517 * @dentry: child of nd->path.dentry or NULL
518 * Returns: 0 on success, -ECHILD on failure
520 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
521 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
522 * @nd or NULL. Must be called from rcu-walk context.
524 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
526 struct fs_struct *fs = current->fs;
527 struct dentry *parent = nd->path.dentry;
529 BUG_ON(!(nd->flags & LOOKUP_RCU));
532 * After legitimizing the bastards, terminate_walk()
533 * will do the right thing for non-RCU mode, and all our
534 * subsequent exit cases should rcu_read_unlock()
535 * before returning. Do vfsmount first; if dentry
536 * can't be legitimized, just set nd->path.dentry to NULL
537 * and rely on dput(NULL) being a no-op.
539 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
541 nd->flags &= ~LOOKUP_RCU;
543 if (!lockref_get_not_dead(&parent->d_lockref)) {
544 nd->path.dentry = NULL;
549 * For a negative lookup, the lookup sequence point is the parents
550 * sequence point, and it only needs to revalidate the parent dentry.
552 * For a positive lookup, we need to move both the parent and the
553 * dentry from the RCU domain to be properly refcounted. And the
554 * sequence number in the dentry validates *both* dentry counters,
555 * since we checked the sequence number of the parent after we got
556 * the child sequence number. So we know the parent must still
557 * be valid if the child sequence number is still valid.
560 if (read_seqcount_retry(&parent->d_seq, nd->seq))
562 BUG_ON(nd->inode != parent->d_inode);
564 if (!lockref_get_not_dead(&dentry->d_lockref))
566 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
571 * Sequence counts matched. Now make sure that the root is
572 * still valid and get it if required.
574 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
575 spin_lock(&fs->lock);
576 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
577 goto unlock_and_drop_dentry;
579 spin_unlock(&fs->lock);
585 unlock_and_drop_dentry:
586 spin_unlock(&fs->lock);
594 if (!(nd->flags & LOOKUP_ROOT))
599 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
601 return dentry->d_op->d_revalidate(dentry, flags);
605 * complete_walk - successful completion of path walk
606 * @nd: pointer nameidata
608 * If we had been in RCU mode, drop out of it and legitimize nd->path.
609 * Revalidate the final result, unless we'd already done that during
610 * the path walk or the filesystem doesn't ask for it. Return 0 on
611 * success, -error on failure. In case of failure caller does not
612 * need to drop nd->path.
614 static int complete_walk(struct nameidata *nd)
616 struct dentry *dentry = nd->path.dentry;
619 if (nd->flags & LOOKUP_RCU) {
620 nd->flags &= ~LOOKUP_RCU;
621 if (!(nd->flags & LOOKUP_ROOT))
624 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
628 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
630 mntput(nd->path.mnt);
633 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
636 mntput(nd->path.mnt);
642 if (likely(!(nd->flags & LOOKUP_JUMPED)))
645 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
648 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
659 static __always_inline void set_root(struct nameidata *nd)
661 get_fs_root(current->fs, &nd->root);
664 static int link_path_walk(const char *, struct nameidata *);
666 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
668 struct fs_struct *fs = current->fs;
672 seq = read_seqcount_begin(&fs->seq);
674 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
675 } while (read_seqcount_retry(&fs->seq, seq));
679 static void path_put_conditional(struct path *path, struct nameidata *nd)
682 if (path->mnt != nd->path.mnt)
686 static inline void path_to_nameidata(const struct path *path,
687 struct nameidata *nd)
689 if (!(nd->flags & LOOKUP_RCU)) {
690 dput(nd->path.dentry);
691 if (nd->path.mnt != path->mnt)
692 mntput(nd->path.mnt);
694 nd->path.mnt = path->mnt;
695 nd->path.dentry = path->dentry;
699 * Helper to directly jump to a known parsed path from ->follow_link,
700 * caller must have taken a reference to path beforehand.
702 void nd_jump_link(struct nameidata *nd, struct path *path)
707 nd->inode = nd->path.dentry->d_inode;
708 nd->flags |= LOOKUP_JUMPED;
711 void nd_set_link(struct nameidata *nd, char *path)
713 nd->saved_names[nd->depth] = path;
715 EXPORT_SYMBOL(nd_set_link);
717 char *nd_get_link(struct nameidata *nd)
719 return nd->saved_names[nd->depth];
721 EXPORT_SYMBOL(nd_get_link);
723 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
725 struct inode *inode = link->dentry->d_inode;
726 if (inode->i_op->put_link)
727 inode->i_op->put_link(link->dentry, nd, cookie);
731 int sysctl_protected_symlinks __read_mostly = 0;
732 int sysctl_protected_hardlinks __read_mostly = 0;
735 * may_follow_link - Check symlink following for unsafe situations
736 * @link: The path of the symlink
737 * @nd: nameidata pathwalk data
739 * In the case of the sysctl_protected_symlinks sysctl being enabled,
740 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
741 * in a sticky world-writable directory. This is to protect privileged
742 * processes from failing races against path names that may change out
743 * from under them by way of other users creating malicious symlinks.
744 * It will permit symlinks to be followed only when outside a sticky
745 * world-writable directory, or when the uid of the symlink and follower
746 * match, or when the directory owner matches the symlink's owner.
748 * Returns 0 if following the symlink is allowed, -ve on error.
750 static inline int may_follow_link(struct path *link, struct nameidata *nd)
752 const struct inode *inode;
753 const struct inode *parent;
755 if (!sysctl_protected_symlinks)
758 /* Allowed if owner and follower match. */
759 inode = link->dentry->d_inode;
760 if (uid_eq(current_cred()->fsuid, inode->i_uid))
763 /* Allowed if parent directory not sticky and world-writable. */
764 parent = nd->path.dentry->d_inode;
765 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
768 /* Allowed if parent directory and link owner match. */
769 if (uid_eq(parent->i_uid, inode->i_uid))
772 audit_log_link_denied("follow_link", link);
773 path_put_conditional(link, nd);
779 * safe_hardlink_source - Check for safe hardlink conditions
780 * @inode: the source inode to hardlink from
782 * Return false if at least one of the following conditions:
783 * - inode is not a regular file
785 * - inode is setgid and group-exec
786 * - access failure for read and write
788 * Otherwise returns true.
790 static bool safe_hardlink_source(struct inode *inode)
792 umode_t mode = inode->i_mode;
794 /* Special files should not get pinned to the filesystem. */
798 /* Setuid files should not get pinned to the filesystem. */
802 /* Executable setgid files should not get pinned to the filesystem. */
803 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
806 /* Hardlinking to unreadable or unwritable sources is dangerous. */
807 if (inode_permission(inode, MAY_READ | MAY_WRITE))
814 * may_linkat - Check permissions for creating a hardlink
815 * @link: the source to hardlink from
817 * Block hardlink when all of:
818 * - sysctl_protected_hardlinks enabled
819 * - fsuid does not match inode
820 * - hardlink source is unsafe (see safe_hardlink_source() above)
823 * Returns 0 if successful, -ve on error.
825 static int may_linkat(struct path *link)
827 const struct cred *cred;
830 if (!sysctl_protected_hardlinks)
833 cred = current_cred();
834 inode = link->dentry->d_inode;
836 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
837 * otherwise, it must be a safe source.
839 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
843 audit_log_link_denied("linkat", link);
847 static __always_inline int
848 follow_link(struct path *link, struct nameidata *nd, void **p)
850 struct dentry *dentry = link->dentry;
854 BUG_ON(nd->flags & LOOKUP_RCU);
856 if (link->mnt == nd->path.mnt)
860 if (unlikely(current->total_link_count >= 40))
861 goto out_put_nd_path;
864 current->total_link_count++;
867 nd_set_link(nd, NULL);
869 error = security_inode_follow_link(link->dentry, nd);
871 goto out_put_nd_path;
873 nd->last_type = LAST_BIND;
874 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
877 goto out_put_nd_path;
882 if (unlikely(IS_ERR(s))) {
884 put_link(nd, link, *p);
893 nd->flags |= LOOKUP_JUMPED;
895 nd->inode = nd->path.dentry->d_inode;
896 error = link_path_walk(s, nd);
898 put_link(nd, link, *p);
910 static int follow_up_rcu(struct path *path)
912 struct mount *mnt = real_mount(path->mnt);
913 struct mount *parent;
914 struct dentry *mountpoint;
916 parent = mnt->mnt_parent;
917 if (&parent->mnt == path->mnt)
919 mountpoint = mnt->mnt_mountpoint;
920 path->dentry = mountpoint;
921 path->mnt = &parent->mnt;
926 * follow_up - Find the mountpoint of path's vfsmount
928 * Given a path, find the mountpoint of its source file system.
929 * Replace @path with the path of the mountpoint in the parent mount.
932 * Return 1 if we went up a level and 0 if we were already at the
935 int follow_up(struct path *path)
937 struct mount *mnt = real_mount(path->mnt);
938 struct mount *parent;
939 struct dentry *mountpoint;
941 read_seqlock_excl(&mount_lock);
942 parent = mnt->mnt_parent;
944 read_sequnlock_excl(&mount_lock);
947 mntget(&parent->mnt);
948 mountpoint = dget(mnt->mnt_mountpoint);
949 read_sequnlock_excl(&mount_lock);
951 path->dentry = mountpoint;
953 path->mnt = &parent->mnt;
956 EXPORT_SYMBOL(follow_up);
959 * Perform an automount
960 * - return -EISDIR to tell follow_managed() to stop and return the path we
963 static int follow_automount(struct path *path, unsigned flags,
966 struct vfsmount *mnt;
969 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
972 /* We don't want to mount if someone's just doing a stat -
973 * unless they're stat'ing a directory and appended a '/' to
976 * We do, however, want to mount if someone wants to open or
977 * create a file of any type under the mountpoint, wants to
978 * traverse through the mountpoint or wants to open the
979 * mounted directory. Also, autofs may mark negative dentries
980 * as being automount points. These will need the attentions
981 * of the daemon to instantiate them before they can be used.
983 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
984 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
985 path->dentry->d_inode)
988 current->total_link_count++;
989 if (current->total_link_count >= 40)
992 mnt = path->dentry->d_op->d_automount(path);
995 * The filesystem is allowed to return -EISDIR here to indicate
996 * it doesn't want to automount. For instance, autofs would do
997 * this so that its userspace daemon can mount on this dentry.
999 * However, we can only permit this if it's a terminal point in
1000 * the path being looked up; if it wasn't then the remainder of
1001 * the path is inaccessible and we should say so.
1003 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
1005 return PTR_ERR(mnt);
1008 if (!mnt) /* mount collision */
1011 if (!*need_mntput) {
1012 /* lock_mount() may release path->mnt on error */
1014 *need_mntput = true;
1016 err = finish_automount(mnt, path);
1020 /* Someone else made a mount here whilst we were busy */
1025 path->dentry = dget(mnt->mnt_root);
1034 * Handle a dentry that is managed in some way.
1035 * - Flagged for transit management (autofs)
1036 * - Flagged as mountpoint
1037 * - Flagged as automount point
1039 * This may only be called in refwalk mode.
1041 * Serialization is taken care of in namespace.c
1043 static int follow_managed(struct path *path, unsigned flags)
1045 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1047 bool need_mntput = false;
1050 /* Given that we're not holding a lock here, we retain the value in a
1051 * local variable for each dentry as we look at it so that we don't see
1052 * the components of that value change under us */
1053 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1054 managed &= DCACHE_MANAGED_DENTRY,
1055 unlikely(managed != 0)) {
1056 /* Allow the filesystem to manage the transit without i_mutex
1058 if (managed & DCACHE_MANAGE_TRANSIT) {
1059 BUG_ON(!path->dentry->d_op);
1060 BUG_ON(!path->dentry->d_op->d_manage);
1061 ret = path->dentry->d_op->d_manage(path->dentry, false);
1066 /* Transit to a mounted filesystem. */
1067 if (managed & DCACHE_MOUNTED) {
1068 struct vfsmount *mounted = lookup_mnt(path);
1073 path->mnt = mounted;
1074 path->dentry = dget(mounted->mnt_root);
1079 /* Something is mounted on this dentry in another
1080 * namespace and/or whatever was mounted there in this
1081 * namespace got unmounted before lookup_mnt() could
1085 /* Handle an automount point */
1086 if (managed & DCACHE_NEED_AUTOMOUNT) {
1087 ret = follow_automount(path, flags, &need_mntput);
1093 /* We didn't change the current path point */
1097 if (need_mntput && path->mnt == mnt)
1101 return ret < 0 ? ret : need_mntput;
1104 int follow_down_one(struct path *path)
1106 struct vfsmount *mounted;
1108 mounted = lookup_mnt(path);
1112 path->mnt = mounted;
1113 path->dentry = dget(mounted->mnt_root);
1118 EXPORT_SYMBOL(follow_down_one);
1120 static inline int managed_dentry_rcu(struct dentry *dentry)
1122 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1123 dentry->d_op->d_manage(dentry, true) : 0;
1127 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1128 * we meet a managed dentry that would need blocking.
1130 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1131 struct inode **inode)
1134 struct mount *mounted;
1136 * Don't forget we might have a non-mountpoint managed dentry
1137 * that wants to block transit.
1139 switch (managed_dentry_rcu(path->dentry)) {
1149 if (!d_mountpoint(path->dentry))
1150 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1152 mounted = __lookup_mnt(path->mnt, path->dentry);
1155 path->mnt = &mounted->mnt;
1156 path->dentry = mounted->mnt.mnt_root;
1157 nd->flags |= LOOKUP_JUMPED;
1158 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1160 * Update the inode too. We don't need to re-check the
1161 * dentry sequence number here after this d_inode read,
1162 * because a mount-point is always pinned.
1164 *inode = path->dentry->d_inode;
1166 return !read_seqretry(&mount_lock, nd->m_seq) &&
1167 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1170 static int follow_dotdot_rcu(struct nameidata *nd)
1172 struct inode *inode = nd->inode;
1177 if (nd->path.dentry == nd->root.dentry &&
1178 nd->path.mnt == nd->root.mnt) {
1181 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1182 struct dentry *old = nd->path.dentry;
1183 struct dentry *parent = old->d_parent;
1186 inode = parent->d_inode;
1187 seq = read_seqcount_begin(&parent->d_seq);
1188 if (read_seqcount_retry(&old->d_seq, nd->seq))
1190 nd->path.dentry = parent;
1194 if (!follow_up_rcu(&nd->path))
1196 inode = nd->path.dentry->d_inode;
1197 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1199 while (d_mountpoint(nd->path.dentry)) {
1200 struct mount *mounted;
1201 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1204 nd->path.mnt = &mounted->mnt;
1205 nd->path.dentry = mounted->mnt.mnt_root;
1206 inode = nd->path.dentry->d_inode;
1207 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1208 if (read_seqretry(&mount_lock, nd->m_seq))
1215 nd->flags &= ~LOOKUP_RCU;
1216 if (!(nd->flags & LOOKUP_ROOT))
1217 nd->root.mnt = NULL;
1223 * Follow down to the covering mount currently visible to userspace. At each
1224 * point, the filesystem owning that dentry may be queried as to whether the
1225 * caller is permitted to proceed or not.
1227 int follow_down(struct path *path)
1232 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1233 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1234 /* Allow the filesystem to manage the transit without i_mutex
1237 * We indicate to the filesystem if someone is trying to mount
1238 * something here. This gives autofs the chance to deny anyone
1239 * other than its daemon the right to mount on its
1242 * The filesystem may sleep at this point.
1244 if (managed & DCACHE_MANAGE_TRANSIT) {
1245 BUG_ON(!path->dentry->d_op);
1246 BUG_ON(!path->dentry->d_op->d_manage);
1247 ret = path->dentry->d_op->d_manage(
1248 path->dentry, false);
1250 return ret == -EISDIR ? 0 : ret;
1253 /* Transit to a mounted filesystem. */
1254 if (managed & DCACHE_MOUNTED) {
1255 struct vfsmount *mounted = lookup_mnt(path);
1260 path->mnt = mounted;
1261 path->dentry = dget(mounted->mnt_root);
1265 /* Don't handle automount points here */
1270 EXPORT_SYMBOL(follow_down);
1273 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1275 static void follow_mount(struct path *path)
1277 while (d_mountpoint(path->dentry)) {
1278 struct vfsmount *mounted = lookup_mnt(path);
1283 path->mnt = mounted;
1284 path->dentry = dget(mounted->mnt_root);
1288 static void follow_dotdot(struct nameidata *nd)
1294 struct dentry *old = nd->path.dentry;
1296 if (nd->path.dentry == nd->root.dentry &&
1297 nd->path.mnt == nd->root.mnt) {
1300 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1301 /* rare case of legitimate dget_parent()... */
1302 nd->path.dentry = dget_parent(nd->path.dentry);
1306 if (!follow_up(&nd->path))
1309 follow_mount(&nd->path);
1310 nd->inode = nd->path.dentry->d_inode;
1314 * This looks up the name in dcache, possibly revalidates the old dentry and
1315 * allocates a new one if not found or not valid. In the need_lookup argument
1316 * returns whether i_op->lookup is necessary.
1318 * dir->d_inode->i_mutex must be held
1320 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1321 unsigned int flags, bool *need_lookup)
1323 struct dentry *dentry;
1326 *need_lookup = false;
1327 dentry = d_lookup(dir, name);
1329 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1330 error = d_revalidate(dentry, flags);
1331 if (unlikely(error <= 0)) {
1334 return ERR_PTR(error);
1336 d_invalidate(dentry);
1345 dentry = d_alloc(dir, name);
1346 if (unlikely(!dentry))
1347 return ERR_PTR(-ENOMEM);
1349 *need_lookup = true;
1355 * Call i_op->lookup on the dentry. The dentry must be negative and
1358 * dir->d_inode->i_mutex must be held
1360 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1365 /* Don't create child dentry for a dead directory. */
1366 if (unlikely(IS_DEADDIR(dir))) {
1368 return ERR_PTR(-ENOENT);
1371 old = dir->i_op->lookup(dir, dentry, flags);
1372 if (unlikely(old)) {
1379 static struct dentry *__lookup_hash(struct qstr *name,
1380 struct dentry *base, unsigned int flags)
1383 struct dentry *dentry;
1385 dentry = lookup_dcache(name, base, flags, &need_lookup);
1389 return lookup_real(base->d_inode, dentry, flags);
1393 * It's more convoluted than I'd like it to be, but... it's still fairly
1394 * small and for now I'd prefer to have fast path as straight as possible.
1395 * It _is_ time-critical.
1397 static int lookup_fast(struct nameidata *nd,
1398 struct path *path, struct inode **inode)
1400 struct vfsmount *mnt = nd->path.mnt;
1401 struct dentry *dentry, *parent = nd->path.dentry;
1407 * Rename seqlock is not required here because in the off chance
1408 * of a false negative due to a concurrent rename, we're going to
1409 * do the non-racy lookup, below.
1411 if (nd->flags & LOOKUP_RCU) {
1413 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1418 * This sequence count validates that the inode matches
1419 * the dentry name information from lookup.
1421 *inode = dentry->d_inode;
1422 if (read_seqcount_retry(&dentry->d_seq, seq))
1426 * This sequence count validates that the parent had no
1427 * changes while we did the lookup of the dentry above.
1429 * The memory barrier in read_seqcount_begin of child is
1430 * enough, we can use __read_seqcount_retry here.
1432 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1436 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1437 status = d_revalidate(dentry, nd->flags);
1438 if (unlikely(status <= 0)) {
1439 if (status != -ECHILD)
1445 path->dentry = dentry;
1446 if (likely(__follow_mount_rcu(nd, path, inode)))
1449 if (unlazy_walk(nd, dentry))
1452 dentry = __d_lookup(parent, &nd->last);
1455 if (unlikely(!dentry))
1458 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1459 status = d_revalidate(dentry, nd->flags);
1460 if (unlikely(status <= 0)) {
1465 d_invalidate(dentry);
1471 path->dentry = dentry;
1472 err = follow_managed(path, nd->flags);
1473 if (unlikely(err < 0)) {
1474 path_put_conditional(path, nd);
1478 nd->flags |= LOOKUP_JUMPED;
1479 *inode = path->dentry->d_inode;
1486 /* Fast lookup failed, do it the slow way */
1487 static int lookup_slow(struct nameidata *nd, struct path *path)
1489 struct dentry *dentry, *parent;
1492 parent = nd->path.dentry;
1493 BUG_ON(nd->inode != parent->d_inode);
1495 mutex_lock(&parent->d_inode->i_mutex);
1496 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1497 mutex_unlock(&parent->d_inode->i_mutex);
1499 return PTR_ERR(dentry);
1500 path->mnt = nd->path.mnt;
1501 path->dentry = dentry;
1502 err = follow_managed(path, nd->flags);
1503 if (unlikely(err < 0)) {
1504 path_put_conditional(path, nd);
1508 nd->flags |= LOOKUP_JUMPED;
1512 static inline int may_lookup(struct nameidata *nd)
1514 if (nd->flags & LOOKUP_RCU) {
1515 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1518 if (unlazy_walk(nd, NULL))
1521 return inode_permission(nd->inode, MAY_EXEC);
1524 static inline int handle_dots(struct nameidata *nd, int type)
1526 if (type == LAST_DOTDOT) {
1527 if (nd->flags & LOOKUP_RCU) {
1528 if (follow_dotdot_rcu(nd))
1536 static void terminate_walk(struct nameidata *nd)
1538 if (!(nd->flags & LOOKUP_RCU)) {
1539 path_put(&nd->path);
1541 nd->flags &= ~LOOKUP_RCU;
1542 if (!(nd->flags & LOOKUP_ROOT))
1543 nd->root.mnt = NULL;
1549 * Do we need to follow links? We _really_ want to be able
1550 * to do this check without having to look at inode->i_op,
1551 * so we keep a cache of "no, this doesn't need follow_link"
1552 * for the common case.
1554 static inline int should_follow_link(struct dentry *dentry, int follow)
1556 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1559 static inline int walk_component(struct nameidata *nd, struct path *path,
1562 struct inode *inode;
1565 * "." and ".." are special - ".." especially so because it has
1566 * to be able to know about the current root directory and
1567 * parent relationships.
1569 if (unlikely(nd->last_type != LAST_NORM))
1570 return handle_dots(nd, nd->last_type);
1571 err = lookup_fast(nd, path, &inode);
1572 if (unlikely(err)) {
1576 err = lookup_slow(nd, path);
1580 inode = path->dentry->d_inode;
1583 if (!inode || d_is_negative(path->dentry))
1586 if (should_follow_link(path->dentry, follow)) {
1587 if (nd->flags & LOOKUP_RCU) {
1588 if (unlikely(unlazy_walk(nd, path->dentry))) {
1593 BUG_ON(inode != path->dentry->d_inode);
1596 path_to_nameidata(path, nd);
1601 path_to_nameidata(path, nd);
1608 * This limits recursive symlink follows to 8, while
1609 * limiting consecutive symlinks to 40.
1611 * Without that kind of total limit, nasty chains of consecutive
1612 * symlinks can cause almost arbitrarily long lookups.
1614 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1618 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1619 path_put_conditional(path, nd);
1620 path_put(&nd->path);
1623 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1626 current->link_count++;
1629 struct path link = *path;
1632 res = follow_link(&link, nd, &cookie);
1635 res = walk_component(nd, path, LOOKUP_FOLLOW);
1636 put_link(nd, &link, cookie);
1639 current->link_count--;
1645 * We can do the critical dentry name comparison and hashing
1646 * operations one word at a time, but we are limited to:
1648 * - Architectures with fast unaligned word accesses. We could
1649 * do a "get_unaligned()" if this helps and is sufficiently
1652 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1653 * do not trap on the (extremely unlikely) case of a page
1654 * crossing operation.
1656 * - Furthermore, we need an efficient 64-bit compile for the
1657 * 64-bit case in order to generate the "number of bytes in
1658 * the final mask". Again, that could be replaced with a
1659 * efficient population count instruction or similar.
1661 #ifdef CONFIG_DCACHE_WORD_ACCESS
1663 #include <asm/word-at-a-time.h>
1667 static inline unsigned int fold_hash(unsigned long hash)
1669 return hash_64(hash, 32);
1672 #else /* 32-bit case */
1674 #define fold_hash(x) (x)
1678 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1680 unsigned long a, mask;
1681 unsigned long hash = 0;
1684 a = load_unaligned_zeropad(name);
1685 if (len < sizeof(unsigned long))
1689 name += sizeof(unsigned long);
1690 len -= sizeof(unsigned long);
1694 mask = bytemask_from_count(len);
1697 return fold_hash(hash);
1699 EXPORT_SYMBOL(full_name_hash);
1702 * Calculate the length and hash of the path component, and
1703 * return the "hash_len" as the result.
1705 static inline u64 hash_name(const char *name)
1707 unsigned long a, b, adata, bdata, mask, hash, len;
1708 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1711 len = -sizeof(unsigned long);
1713 hash = (hash + a) * 9;
1714 len += sizeof(unsigned long);
1715 a = load_unaligned_zeropad(name+len);
1716 b = a ^ REPEAT_BYTE('/');
1717 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1719 adata = prep_zero_mask(a, adata, &constants);
1720 bdata = prep_zero_mask(b, bdata, &constants);
1722 mask = create_zero_mask(adata | bdata);
1724 hash += a & zero_bytemask(mask);
1725 len += find_zero(mask);
1726 return hashlen_create(fold_hash(hash), len);
1731 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1733 unsigned long hash = init_name_hash();
1735 hash = partial_name_hash(*name++, hash);
1736 return end_name_hash(hash);
1738 EXPORT_SYMBOL(full_name_hash);
1741 * We know there's a real path component here of at least
1744 static inline u64 hash_name(const char *name)
1746 unsigned long hash = init_name_hash();
1747 unsigned long len = 0, c;
1749 c = (unsigned char)*name;
1752 hash = partial_name_hash(c, hash);
1753 c = (unsigned char)name[len];
1754 } while (c && c != '/');
1755 return hashlen_create(end_name_hash(hash), len);
1762 * This is the basic name resolution function, turning a pathname into
1763 * the final dentry. We expect 'base' to be positive and a directory.
1765 * Returns 0 and nd will have valid dentry and mnt on success.
1766 * Returns error and drops reference to input namei data on failure.
1768 static int link_path_walk(const char *name, struct nameidata *nd)
1778 /* At this point we know we have a real path component. */
1783 err = may_lookup(nd);
1787 hash_len = hash_name(name);
1790 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1792 if (name[1] == '.') {
1794 nd->flags |= LOOKUP_JUMPED;
1800 if (likely(type == LAST_NORM)) {
1801 struct dentry *parent = nd->path.dentry;
1802 nd->flags &= ~LOOKUP_JUMPED;
1803 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1804 struct qstr this = { { .hash_len = hash_len }, .name = name };
1805 err = parent->d_op->d_hash(parent, &this);
1808 hash_len = this.hash_len;
1813 nd->last.hash_len = hash_len;
1814 nd->last.name = name;
1815 nd->last_type = type;
1817 name += hashlen_len(hash_len);
1821 * If it wasn't NUL, we know it was '/'. Skip that
1822 * slash, and continue until no more slashes.
1826 } while (unlikely(*name == '/'));
1830 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1835 err = nested_symlink(&next, nd);
1839 if (!d_can_lookup(nd->path.dentry)) {
1848 static int path_init(int dfd, const char *name, unsigned int flags,
1849 struct nameidata *nd)
1853 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1854 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1857 if (flags & LOOKUP_ROOT) {
1858 struct dentry *root = nd->root.dentry;
1859 struct inode *inode = root->d_inode;
1861 if (!d_can_lookup(root))
1863 retval = inode_permission(inode, MAY_EXEC);
1867 nd->path = nd->root;
1869 if (flags & LOOKUP_RCU) {
1871 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1872 nd->m_seq = read_seqbegin(&mount_lock);
1874 path_get(&nd->path);
1879 nd->root.mnt = NULL;
1881 nd->m_seq = read_seqbegin(&mount_lock);
1883 if (flags & LOOKUP_RCU) {
1885 nd->seq = set_root_rcu(nd);
1888 path_get(&nd->root);
1890 nd->path = nd->root;
1891 } else if (dfd == AT_FDCWD) {
1892 if (flags & LOOKUP_RCU) {
1893 struct fs_struct *fs = current->fs;
1899 seq = read_seqcount_begin(&fs->seq);
1901 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1902 } while (read_seqcount_retry(&fs->seq, seq));
1904 get_fs_pwd(current->fs, &nd->path);
1907 /* Caller must check execute permissions on the starting path component */
1908 struct fd f = fdget_raw(dfd);
1909 struct dentry *dentry;
1914 dentry = f.file->f_path.dentry;
1917 if (!d_can_lookup(dentry)) {
1923 nd->path = f.file->f_path;
1924 if (flags & LOOKUP_RCU) {
1925 if (f.flags & FDPUT_FPUT)
1927 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1930 path_get(&nd->path);
1935 nd->inode = nd->path.dentry->d_inode;
1936 if (!(flags & LOOKUP_RCU))
1938 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1940 if (!(nd->flags & LOOKUP_ROOT))
1941 nd->root.mnt = NULL;
1945 current->total_link_count = 0;
1946 return link_path_walk(name, nd);
1949 static void path_cleanup(struct nameidata *nd)
1951 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1952 path_put(&nd->root);
1953 nd->root.mnt = NULL;
1955 if (unlikely(nd->base))
1959 static inline int lookup_last(struct nameidata *nd, struct path *path)
1961 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1962 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1964 nd->flags &= ~LOOKUP_PARENT;
1965 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1968 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1969 static int path_lookupat(int dfd, const char *name,
1970 unsigned int flags, struct nameidata *nd)
1976 * Path walking is largely split up into 2 different synchronisation
1977 * schemes, rcu-walk and ref-walk (explained in
1978 * Documentation/filesystems/path-lookup.txt). These share much of the
1979 * path walk code, but some things particularly setup, cleanup, and
1980 * following mounts are sufficiently divergent that functions are
1981 * duplicated. Typically there is a function foo(), and its RCU
1982 * analogue, foo_rcu().
1984 * -ECHILD is the error number of choice (just to avoid clashes) that
1985 * is returned if some aspect of an rcu-walk fails. Such an error must
1986 * be handled by restarting a traditional ref-walk (which will always
1987 * be able to complete).
1989 err = path_init(dfd, name, flags, nd);
1990 if (!err && !(flags & LOOKUP_PARENT)) {
1991 err = lookup_last(nd, &path);
1994 struct path link = path;
1995 err = may_follow_link(&link, nd);
1998 nd->flags |= LOOKUP_PARENT;
1999 err = follow_link(&link, nd, &cookie);
2002 err = lookup_last(nd, &path);
2003 put_link(nd, &link, cookie);
2008 err = complete_walk(nd);
2010 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2011 if (!d_can_lookup(nd->path.dentry)) {
2012 path_put(&nd->path);
2021 static int filename_lookup(int dfd, struct filename *name,
2022 unsigned int flags, struct nameidata *nd)
2024 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2025 if (unlikely(retval == -ECHILD))
2026 retval = path_lookupat(dfd, name->name, flags, nd);
2027 if (unlikely(retval == -ESTALE))
2028 retval = path_lookupat(dfd, name->name,
2029 flags | LOOKUP_REVAL, nd);
2031 if (likely(!retval))
2032 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2036 static int do_path_lookup(int dfd, const char *name,
2037 unsigned int flags, struct nameidata *nd)
2039 struct filename filename = { .name = name };
2041 return filename_lookup(dfd, &filename, flags, nd);
2044 /* does lookup, returns the object with parent locked */
2045 struct dentry *kern_path_locked(const char *name, struct path *path)
2047 struct nameidata nd;
2049 struct filename filename = {.name = name};
2050 int err = filename_lookup(AT_FDCWD, &filename, LOOKUP_PARENT, &nd);
2052 return ERR_PTR(err);
2053 if (nd.last_type != LAST_NORM) {
2055 return ERR_PTR(-EINVAL);
2057 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2058 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2060 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2068 int kern_path(const char *name, unsigned int flags, struct path *path)
2070 struct nameidata nd;
2071 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2076 EXPORT_SYMBOL(kern_path);
2079 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2080 * @dentry: pointer to dentry of the base directory
2081 * @mnt: pointer to vfs mount of the base directory
2082 * @name: pointer to file name
2083 * @flags: lookup flags
2084 * @path: pointer to struct path to fill
2086 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2087 const char *name, unsigned int flags,
2090 struct nameidata nd;
2092 nd.root.dentry = dentry;
2094 BUG_ON(flags & LOOKUP_PARENT);
2095 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2096 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2101 EXPORT_SYMBOL(vfs_path_lookup);
2104 * Restricted form of lookup. Doesn't follow links, single-component only,
2105 * needs parent already locked. Doesn't follow mounts.
2108 static struct dentry *lookup_hash(struct nameidata *nd)
2110 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2114 * lookup_one_len - filesystem helper to lookup single pathname component
2115 * @name: pathname component to lookup
2116 * @base: base directory to lookup from
2117 * @len: maximum length @len should be interpreted to
2119 * Note that this routine is purely a helper for filesystem usage and should
2120 * not be called by generic code. Also note that by using this function the
2121 * nameidata argument is passed to the filesystem methods and a filesystem
2122 * using this helper needs to be prepared for that.
2124 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2130 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2134 this.hash = full_name_hash(name, len);
2136 return ERR_PTR(-EACCES);
2138 if (unlikely(name[0] == '.')) {
2139 if (len < 2 || (len == 2 && name[1] == '.'))
2140 return ERR_PTR(-EACCES);
2144 c = *(const unsigned char *)name++;
2145 if (c == '/' || c == '\0')
2146 return ERR_PTR(-EACCES);
2149 * See if the low-level filesystem might want
2150 * to use its own hash..
2152 if (base->d_flags & DCACHE_OP_HASH) {
2153 int err = base->d_op->d_hash(base, &this);
2155 return ERR_PTR(err);
2158 err = inode_permission(base->d_inode, MAY_EXEC);
2160 return ERR_PTR(err);
2162 return __lookup_hash(&this, base, 0);
2164 EXPORT_SYMBOL(lookup_one_len);
2166 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2167 struct path *path, int *empty)
2169 struct nameidata nd;
2170 struct filename *tmp = getname_flags(name, flags, empty);
2171 int err = PTR_ERR(tmp);
2174 BUG_ON(flags & LOOKUP_PARENT);
2176 err = filename_lookup(dfd, tmp, flags, &nd);
2184 int user_path_at(int dfd, const char __user *name, unsigned flags,
2187 return user_path_at_empty(dfd, name, flags, path, NULL);
2189 EXPORT_SYMBOL(user_path_at);
2192 * NB: most callers don't do anything directly with the reference to the
2193 * to struct filename, but the nd->last pointer points into the name string
2194 * allocated by getname. So we must hold the reference to it until all
2195 * path-walking is complete.
2197 static struct filename *
2198 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2201 struct filename *s = getname(path);
2204 /* only LOOKUP_REVAL is allowed in extra flags */
2205 flags &= LOOKUP_REVAL;
2210 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2213 return ERR_PTR(error);
2220 * mountpoint_last - look up last component for umount
2221 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2222 * @path: pointer to container for result
2224 * This is a special lookup_last function just for umount. In this case, we
2225 * need to resolve the path without doing any revalidation.
2227 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2228 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2229 * in almost all cases, this lookup will be served out of the dcache. The only
2230 * cases where it won't are if nd->last refers to a symlink or the path is
2231 * bogus and it doesn't exist.
2234 * -error: if there was an error during lookup. This includes -ENOENT if the
2235 * lookup found a negative dentry. The nd->path reference will also be
2238 * 0: if we successfully resolved nd->path and found it to not to be a
2239 * symlink that needs to be followed. "path" will also be populated.
2240 * The nd->path reference will also be put.
2242 * 1: if we successfully resolved nd->last and found it to be a symlink
2243 * that needs to be followed. "path" will be populated with the path
2244 * to the link, and nd->path will *not* be put.
2247 mountpoint_last(struct nameidata *nd, struct path *path)
2250 struct dentry *dentry;
2251 struct dentry *dir = nd->path.dentry;
2253 /* If we're in rcuwalk, drop out of it to handle last component */
2254 if (nd->flags & LOOKUP_RCU) {
2255 if (unlazy_walk(nd, NULL)) {
2261 nd->flags &= ~LOOKUP_PARENT;
2263 if (unlikely(nd->last_type != LAST_NORM)) {
2264 error = handle_dots(nd, nd->last_type);
2267 dentry = dget(nd->path.dentry);
2271 mutex_lock(&dir->d_inode->i_mutex);
2272 dentry = d_lookup(dir, &nd->last);
2275 * No cached dentry. Mounted dentries are pinned in the cache,
2276 * so that means that this dentry is probably a symlink or the
2277 * path doesn't actually point to a mounted dentry.
2279 dentry = d_alloc(dir, &nd->last);
2282 mutex_unlock(&dir->d_inode->i_mutex);
2285 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2286 error = PTR_ERR(dentry);
2287 if (IS_ERR(dentry)) {
2288 mutex_unlock(&dir->d_inode->i_mutex);
2292 mutex_unlock(&dir->d_inode->i_mutex);
2295 if (!dentry->d_inode || d_is_negative(dentry)) {
2300 path->dentry = dentry;
2301 path->mnt = nd->path.mnt;
2302 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2313 * path_mountpoint - look up a path to be umounted
2314 * @dfd: directory file descriptor to start walk from
2315 * @name: full pathname to walk
2316 * @path: pointer to container for result
2317 * @flags: lookup flags
2319 * Look up the given name, but don't attempt to revalidate the last component.
2320 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2323 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2325 struct nameidata nd;
2328 err = path_init(dfd, name, flags, &nd);
2332 err = mountpoint_last(&nd, path);
2335 struct path link = *path;
2336 err = may_follow_link(&link, &nd);
2339 nd.flags |= LOOKUP_PARENT;
2340 err = follow_link(&link, &nd, &cookie);
2343 err = mountpoint_last(&nd, path);
2344 put_link(&nd, &link, cookie);
2352 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2355 int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2356 if (unlikely(error == -ECHILD))
2357 error = path_mountpoint(dfd, s->name, path, flags);
2358 if (unlikely(error == -ESTALE))
2359 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2361 audit_inode(s, path->dentry, 0);
2366 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2367 * @dfd: directory file descriptor
2368 * @name: pathname from userland
2369 * @flags: lookup flags
2370 * @path: pointer to container to hold result
2372 * A umount is a special case for path walking. We're not actually interested
2373 * in the inode in this situation, and ESTALE errors can be a problem. We
2374 * simply want track down the dentry and vfsmount attached at the mountpoint
2375 * and avoid revalidating the last component.
2377 * Returns 0 and populates "path" on success.
2380 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2383 struct filename *s = getname(name);
2387 error = filename_mountpoint(dfd, s, path, flags);
2393 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2396 struct filename s = {.name = name};
2397 return filename_mountpoint(dfd, &s, path, flags);
2399 EXPORT_SYMBOL(kern_path_mountpoint);
2401 int __check_sticky(struct inode *dir, struct inode *inode)
2403 kuid_t fsuid = current_fsuid();
2405 if (uid_eq(inode->i_uid, fsuid))
2407 if (uid_eq(dir->i_uid, fsuid))
2409 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2411 EXPORT_SYMBOL(__check_sticky);
2414 * Check whether we can remove a link victim from directory dir, check
2415 * whether the type of victim is right.
2416 * 1. We can't do it if dir is read-only (done in permission())
2417 * 2. We should have write and exec permissions on dir
2418 * 3. We can't remove anything from append-only dir
2419 * 4. We can't do anything with immutable dir (done in permission())
2420 * 5. If the sticky bit on dir is set we should either
2421 * a. be owner of dir, or
2422 * b. be owner of victim, or
2423 * c. have CAP_FOWNER capability
2424 * 6. If the victim is append-only or immutable we can't do antyhing with
2425 * links pointing to it.
2426 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2427 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2428 * 9. We can't remove a root or mountpoint.
2429 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2430 * nfs_async_unlink().
2432 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2434 struct inode *inode = victim->d_inode;
2437 if (d_is_negative(victim))
2441 BUG_ON(victim->d_parent->d_inode != dir);
2442 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2444 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2450 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2451 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2454 if (!d_is_dir(victim))
2456 if (IS_ROOT(victim))
2458 } else if (d_is_dir(victim))
2460 if (IS_DEADDIR(dir))
2462 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2467 /* Check whether we can create an object with dentry child in directory
2469 * 1. We can't do it if child already exists (open has special treatment for
2470 * this case, but since we are inlined it's OK)
2471 * 2. We can't do it if dir is read-only (done in permission())
2472 * 3. We should have write and exec permissions on dir
2473 * 4. We can't do it if dir is immutable (done in permission())
2475 static inline int may_create(struct inode *dir, struct dentry *child)
2477 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2480 if (IS_DEADDIR(dir))
2482 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2486 * p1 and p2 should be directories on the same fs.
2488 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2493 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2497 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2499 p = d_ancestor(p2, p1);
2501 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2502 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2506 p = d_ancestor(p1, p2);
2508 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2509 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2513 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2514 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2517 EXPORT_SYMBOL(lock_rename);
2519 void unlock_rename(struct dentry *p1, struct dentry *p2)
2521 mutex_unlock(&p1->d_inode->i_mutex);
2523 mutex_unlock(&p2->d_inode->i_mutex);
2524 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2527 EXPORT_SYMBOL(unlock_rename);
2529 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2532 int error = may_create(dir, dentry);
2536 if (!dir->i_op->create)
2537 return -EACCES; /* shouldn't it be ENOSYS? */
2540 error = security_inode_create(dir, dentry, mode);
2543 error = dir->i_op->create(dir, dentry, mode, want_excl);
2545 fsnotify_create(dir, dentry);
2548 EXPORT_SYMBOL(vfs_create);
2550 static int may_open(struct path *path, int acc_mode, int flag)
2552 struct dentry *dentry = path->dentry;
2553 struct inode *inode = dentry->d_inode;
2563 switch (inode->i_mode & S_IFMT) {
2567 if (acc_mode & MAY_WRITE)
2572 if (path->mnt->mnt_flags & MNT_NODEV)
2581 error = inode_permission(inode, acc_mode);
2586 * An append-only file must be opened in append mode for writing.
2588 if (IS_APPEND(inode)) {
2589 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2595 /* O_NOATIME can only be set by the owner or superuser */
2596 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2602 static int handle_truncate(struct file *filp)
2604 struct path *path = &filp->f_path;
2605 struct inode *inode = path->dentry->d_inode;
2606 int error = get_write_access(inode);
2610 * Refuse to truncate files with mandatory locks held on them.
2612 error = locks_verify_locked(filp);
2614 error = security_path_truncate(path);
2616 error = do_truncate(path->dentry, 0,
2617 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2620 put_write_access(inode);
2624 static inline int open_to_namei_flags(int flag)
2626 if ((flag & O_ACCMODE) == 3)
2631 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2633 int error = security_path_mknod(dir, dentry, mode, 0);
2637 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2641 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2645 * Attempt to atomically look up, create and open a file from a negative
2648 * Returns 0 if successful. The file will have been created and attached to
2649 * @file by the filesystem calling finish_open().
2651 * Returns 1 if the file was looked up only or didn't need creating. The
2652 * caller will need to perform the open themselves. @path will have been
2653 * updated to point to the new dentry. This may be negative.
2655 * Returns an error code otherwise.
2657 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2658 struct path *path, struct file *file,
2659 const struct open_flags *op,
2660 bool got_write, bool need_lookup,
2663 struct inode *dir = nd->path.dentry->d_inode;
2664 unsigned open_flag = open_to_namei_flags(op->open_flag);
2668 int create_error = 0;
2669 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2672 BUG_ON(dentry->d_inode);
2674 /* Don't create child dentry for a dead directory. */
2675 if (unlikely(IS_DEADDIR(dir))) {
2681 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2682 mode &= ~current_umask();
2684 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2686 open_flag &= ~O_TRUNC;
2689 * Checking write permission is tricky, bacuse we don't know if we are
2690 * going to actually need it: O_CREAT opens should work as long as the
2691 * file exists. But checking existence breaks atomicity. The trick is
2692 * to check access and if not granted clear O_CREAT from the flags.
2694 * Another problem is returing the "right" error value (e.g. for an
2695 * O_EXCL open we want to return EEXIST not EROFS).
2697 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2698 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2699 if (!(open_flag & O_CREAT)) {
2701 * No O_CREATE -> atomicity not a requirement -> fall
2702 * back to lookup + open
2705 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2706 /* Fall back and fail with the right error */
2707 create_error = -EROFS;
2710 /* No side effects, safe to clear O_CREAT */
2711 create_error = -EROFS;
2712 open_flag &= ~O_CREAT;
2716 if (open_flag & O_CREAT) {
2717 error = may_o_create(&nd->path, dentry, mode);
2719 create_error = error;
2720 if (open_flag & O_EXCL)
2722 open_flag &= ~O_CREAT;
2726 if (nd->flags & LOOKUP_DIRECTORY)
2727 open_flag |= O_DIRECTORY;
2729 file->f_path.dentry = DENTRY_NOT_SET;
2730 file->f_path.mnt = nd->path.mnt;
2731 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2734 if (create_error && error == -ENOENT)
2735 error = create_error;
2739 if (error) { /* returned 1, that is */
2740 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2744 if (file->f_path.dentry) {
2746 dentry = file->f_path.dentry;
2748 if (*opened & FILE_CREATED)
2749 fsnotify_create(dir, dentry);
2750 if (!dentry->d_inode) {
2751 WARN_ON(*opened & FILE_CREATED);
2753 error = create_error;
2757 if (excl && !(*opened & FILE_CREATED)) {
2766 * We didn't have the inode before the open, so check open permission
2769 acc_mode = op->acc_mode;
2770 if (*opened & FILE_CREATED) {
2771 WARN_ON(!(open_flag & O_CREAT));
2772 fsnotify_create(dir, dentry);
2773 acc_mode = MAY_OPEN;
2775 error = may_open(&file->f_path, acc_mode, open_flag);
2785 dentry = lookup_real(dir, dentry, nd->flags);
2787 return PTR_ERR(dentry);
2790 int open_flag = op->open_flag;
2792 error = create_error;
2793 if ((open_flag & O_EXCL)) {
2794 if (!dentry->d_inode)
2796 } else if (!dentry->d_inode) {
2798 } else if ((open_flag & O_TRUNC) &&
2799 S_ISREG(dentry->d_inode->i_mode)) {
2802 /* will fail later, go on to get the right error */
2806 path->dentry = dentry;
2807 path->mnt = nd->path.mnt;
2812 * Look up and maybe create and open the last component.
2814 * Must be called with i_mutex held on parent.
2816 * Returns 0 if the file was successfully atomically created (if necessary) and
2817 * opened. In this case the file will be returned attached to @file.
2819 * Returns 1 if the file was not completely opened at this time, though lookups
2820 * and creations will have been performed and the dentry returned in @path will
2821 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2822 * specified then a negative dentry may be returned.
2824 * An error code is returned otherwise.
2826 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2827 * cleared otherwise prior to returning.
2829 static int lookup_open(struct nameidata *nd, struct path *path,
2831 const struct open_flags *op,
2832 bool got_write, int *opened)
2834 struct dentry *dir = nd->path.dentry;
2835 struct inode *dir_inode = dir->d_inode;
2836 struct dentry *dentry;
2840 *opened &= ~FILE_CREATED;
2841 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2843 return PTR_ERR(dentry);
2845 /* Cached positive dentry: will open in f_op->open */
2846 if (!need_lookup && dentry->d_inode)
2849 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2850 return atomic_open(nd, dentry, path, file, op, got_write,
2851 need_lookup, opened);
2855 BUG_ON(dentry->d_inode);
2857 dentry = lookup_real(dir_inode, dentry, nd->flags);
2859 return PTR_ERR(dentry);
2862 /* Negative dentry, just create the file */
2863 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2864 umode_t mode = op->mode;
2865 if (!IS_POSIXACL(dir->d_inode))
2866 mode &= ~current_umask();
2868 * This write is needed to ensure that a
2869 * rw->ro transition does not occur between
2870 * the time when the file is created and when
2871 * a permanent write count is taken through
2872 * the 'struct file' in finish_open().
2878 *opened |= FILE_CREATED;
2879 error = security_path_mknod(&nd->path, dentry, mode, 0);
2882 error = vfs_create(dir->d_inode, dentry, mode,
2883 nd->flags & LOOKUP_EXCL);
2888 path->dentry = dentry;
2889 path->mnt = nd->path.mnt;
2898 * Handle the last step of open()
2900 static int do_last(struct nameidata *nd, struct path *path,
2901 struct file *file, const struct open_flags *op,
2902 int *opened, struct filename *name)
2904 struct dentry *dir = nd->path.dentry;
2905 int open_flag = op->open_flag;
2906 bool will_truncate = (open_flag & O_TRUNC) != 0;
2907 bool got_write = false;
2908 int acc_mode = op->acc_mode;
2909 struct inode *inode;
2910 bool symlink_ok = false;
2911 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2912 bool retried = false;
2915 nd->flags &= ~LOOKUP_PARENT;
2916 nd->flags |= op->intent;
2918 if (nd->last_type != LAST_NORM) {
2919 error = handle_dots(nd, nd->last_type);
2925 if (!(open_flag & O_CREAT)) {
2926 if (nd->last.name[nd->last.len])
2927 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2928 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2930 /* we _can_ be in RCU mode here */
2931 error = lookup_fast(nd, path, &inode);
2938 BUG_ON(nd->inode != dir->d_inode);
2940 /* create side of things */
2942 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2943 * has been cleared when we got to the last component we are
2946 error = complete_walk(nd);
2950 audit_inode(name, dir, LOOKUP_PARENT);
2952 /* trailing slashes? */
2953 if (nd->last.name[nd->last.len])
2958 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2959 error = mnt_want_write(nd->path.mnt);
2963 * do _not_ fail yet - we might not need that or fail with
2964 * a different error; let lookup_open() decide; we'll be
2965 * dropping this one anyway.
2968 mutex_lock(&dir->d_inode->i_mutex);
2969 error = lookup_open(nd, path, file, op, got_write, opened);
2970 mutex_unlock(&dir->d_inode->i_mutex);
2976 if ((*opened & FILE_CREATED) ||
2977 !S_ISREG(file_inode(file)->i_mode))
2978 will_truncate = false;
2980 audit_inode(name, file->f_path.dentry, 0);
2984 if (*opened & FILE_CREATED) {
2985 /* Don't check for write permission, don't truncate */
2986 open_flag &= ~O_TRUNC;
2987 will_truncate = false;
2988 acc_mode = MAY_OPEN;
2989 path_to_nameidata(path, nd);
2990 goto finish_open_created;
2994 * create/update audit record if it already exists.
2996 if (d_is_positive(path->dentry))
2997 audit_inode(name, path->dentry, 0);
3000 * If atomic_open() acquired write access it is dropped now due to
3001 * possible mount and symlink following (this might be optimized away if
3005 mnt_drop_write(nd->path.mnt);
3010 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3013 error = follow_managed(path, nd->flags);
3018 nd->flags |= LOOKUP_JUMPED;
3020 BUG_ON(nd->flags & LOOKUP_RCU);
3021 inode = path->dentry->d_inode;
3023 /* we _can_ be in RCU mode here */
3025 if (!inode || d_is_negative(path->dentry)) {
3026 path_to_nameidata(path, nd);
3030 if (should_follow_link(path->dentry, !symlink_ok)) {
3031 if (nd->flags & LOOKUP_RCU) {
3032 if (unlikely(unlazy_walk(nd, path->dentry))) {
3037 BUG_ON(inode != path->dentry->d_inode);
3041 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3042 path_to_nameidata(path, nd);
3044 save_parent.dentry = nd->path.dentry;
3045 save_parent.mnt = mntget(path->mnt);
3046 nd->path.dentry = path->dentry;
3050 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3052 error = complete_walk(nd);
3054 path_put(&save_parent);
3057 audit_inode(name, nd->path.dentry, 0);
3059 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3062 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3064 if (!S_ISREG(nd->inode->i_mode))
3065 will_truncate = false;
3067 if (will_truncate) {
3068 error = mnt_want_write(nd->path.mnt);
3073 finish_open_created:
3074 error = may_open(&nd->path, acc_mode, open_flag);
3078 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3079 error = vfs_open(&nd->path, file, current_cred());
3081 *opened |= FILE_OPENED;
3083 if (error == -EOPENSTALE)
3088 error = open_check_o_direct(file);
3091 error = ima_file_check(file, op->acc_mode, *opened);
3095 if (will_truncate) {
3096 error = handle_truncate(file);
3102 mnt_drop_write(nd->path.mnt);
3103 path_put(&save_parent);
3108 path_put_conditional(path, nd);
3115 /* If no saved parent or already retried then can't retry */
3116 if (!save_parent.dentry || retried)
3119 BUG_ON(save_parent.dentry != dir);
3120 path_put(&nd->path);
3121 nd->path = save_parent;
3122 nd->inode = dir->d_inode;
3123 save_parent.mnt = NULL;
3124 save_parent.dentry = NULL;
3126 mnt_drop_write(nd->path.mnt);
3133 static int do_tmpfile(int dfd, struct filename *pathname,
3134 struct nameidata *nd, int flags,
3135 const struct open_flags *op,
3136 struct file *file, int *opened)
3138 static const struct qstr name = QSTR_INIT("/", 1);
3139 struct dentry *dentry, *child;
3141 int error = path_lookupat(dfd, pathname->name,
3142 flags | LOOKUP_DIRECTORY, nd);
3143 if (unlikely(error))
3145 error = mnt_want_write(nd->path.mnt);
3146 if (unlikely(error))
3148 /* we want directory to be writable */
3149 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3152 dentry = nd->path.dentry;
3153 dir = dentry->d_inode;
3154 if (!dir->i_op->tmpfile) {
3155 error = -EOPNOTSUPP;
3158 child = d_alloc(dentry, &name);
3159 if (unlikely(!child)) {
3163 nd->flags &= ~LOOKUP_DIRECTORY;
3164 nd->flags |= op->intent;
3165 dput(nd->path.dentry);
3166 nd->path.dentry = child;
3167 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3170 audit_inode(pathname, nd->path.dentry, 0);
3171 /* Don't check for other permissions, the inode was just created */
3172 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3175 file->f_path.mnt = nd->path.mnt;
3176 error = finish_open(file, nd->path.dentry, NULL, opened);
3179 error = open_check_o_direct(file);
3182 } else if (!(op->open_flag & O_EXCL)) {
3183 struct inode *inode = file_inode(file);
3184 spin_lock(&inode->i_lock);
3185 inode->i_state |= I_LINKABLE;
3186 spin_unlock(&inode->i_lock);
3189 mnt_drop_write(nd->path.mnt);
3191 path_put(&nd->path);
3195 static struct file *path_openat(int dfd, struct filename *pathname,
3196 struct nameidata *nd, const struct open_flags *op, int flags)
3203 file = get_empty_filp();
3207 file->f_flags = op->open_flag;
3209 if (unlikely(file->f_flags & __O_TMPFILE)) {
3210 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3214 error = path_init(dfd, pathname->name, flags, nd);
3215 if (unlikely(error))
3218 error = do_last(nd, &path, file, op, &opened, pathname);
3219 while (unlikely(error > 0)) { /* trailing symlink */
3220 struct path link = path;
3222 if (!(nd->flags & LOOKUP_FOLLOW)) {
3223 path_put_conditional(&path, nd);
3224 path_put(&nd->path);
3228 error = may_follow_link(&link, nd);
3229 if (unlikely(error))
3231 nd->flags |= LOOKUP_PARENT;
3232 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3233 error = follow_link(&link, nd, &cookie);
3234 if (unlikely(error))
3236 error = do_last(nd, &path, file, op, &opened, pathname);
3237 put_link(nd, &link, cookie);
3241 if (!(opened & FILE_OPENED)) {
3245 if (unlikely(error)) {
3246 if (error == -EOPENSTALE) {
3247 if (flags & LOOKUP_RCU)
3252 file = ERR_PTR(error);
3257 struct file *do_filp_open(int dfd, struct filename *pathname,
3258 const struct open_flags *op)
3260 struct nameidata nd;
3261 int flags = op->lookup_flags;
3264 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3265 if (unlikely(filp == ERR_PTR(-ECHILD)))
3266 filp = path_openat(dfd, pathname, &nd, op, flags);
3267 if (unlikely(filp == ERR_PTR(-ESTALE)))
3268 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3272 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3273 const char *name, const struct open_flags *op)
3275 struct nameidata nd;
3277 struct filename filename = { .name = name };
3278 int flags = op->lookup_flags | LOOKUP_ROOT;
3281 nd.root.dentry = dentry;
3283 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3284 return ERR_PTR(-ELOOP);
3286 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3287 if (unlikely(file == ERR_PTR(-ECHILD)))
3288 file = path_openat(-1, &filename, &nd, op, flags);
3289 if (unlikely(file == ERR_PTR(-ESTALE)))
3290 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3294 static struct dentry *filename_create(int dfd, struct filename *name,
3295 struct path *path, unsigned int lookup_flags)
3297 struct dentry *dentry = ERR_PTR(-EEXIST);
3298 struct nameidata nd;
3301 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3304 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3305 * other flags passed in are ignored!
3307 lookup_flags &= LOOKUP_REVAL;
3309 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3311 return ERR_PTR(error);
3314 * Yucky last component or no last component at all?
3315 * (foo/., foo/.., /////)
3317 if (nd.last_type != LAST_NORM)
3319 nd.flags &= ~LOOKUP_PARENT;
3320 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3322 /* don't fail immediately if it's r/o, at least try to report other errors */
3323 err2 = mnt_want_write(nd.path.mnt);
3325 * Do the final lookup.
3327 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3328 dentry = lookup_hash(&nd);
3333 if (d_is_positive(dentry))
3337 * Special case - lookup gave negative, but... we had foo/bar/
3338 * From the vfs_mknod() POV we just have a negative dentry -
3339 * all is fine. Let's be bastards - you had / on the end, you've
3340 * been asking for (non-existent) directory. -ENOENT for you.
3342 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3346 if (unlikely(err2)) {
3354 dentry = ERR_PTR(error);
3356 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3358 mnt_drop_write(nd.path.mnt);
3364 struct dentry *kern_path_create(int dfd, const char *pathname,
3365 struct path *path, unsigned int lookup_flags)
3367 struct filename filename = {.name = pathname};
3368 return filename_create(dfd, &filename, path, lookup_flags);
3370 EXPORT_SYMBOL(kern_path_create);
3372 void done_path_create(struct path *path, struct dentry *dentry)
3375 mutex_unlock(&path->dentry->d_inode->i_mutex);
3376 mnt_drop_write(path->mnt);
3379 EXPORT_SYMBOL(done_path_create);
3381 struct dentry *user_path_create(int dfd, const char __user *pathname,
3382 struct path *path, unsigned int lookup_flags)
3384 struct filename *tmp = getname(pathname);
3387 return ERR_CAST(tmp);
3388 res = filename_create(dfd, tmp, path, lookup_flags);
3392 EXPORT_SYMBOL(user_path_create);
3394 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3396 int error = may_create(dir, dentry);
3401 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3404 if (!dir->i_op->mknod)
3407 error = devcgroup_inode_mknod(mode, dev);
3411 error = security_inode_mknod(dir, dentry, mode, dev);
3415 error = dir->i_op->mknod(dir, dentry, mode, dev);
3417 fsnotify_create(dir, dentry);
3420 EXPORT_SYMBOL(vfs_mknod);
3422 static int may_mknod(umode_t mode)
3424 switch (mode & S_IFMT) {
3430 case 0: /* zero mode translates to S_IFREG */
3439 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3442 struct dentry *dentry;
3445 unsigned int lookup_flags = 0;
3447 error = may_mknod(mode);
3451 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3453 return PTR_ERR(dentry);
3455 if (!IS_POSIXACL(path.dentry->d_inode))
3456 mode &= ~current_umask();
3457 error = security_path_mknod(&path, dentry, mode, dev);
3460 switch (mode & S_IFMT) {
3461 case 0: case S_IFREG:
3462 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3464 case S_IFCHR: case S_IFBLK:
3465 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3466 new_decode_dev(dev));
3468 case S_IFIFO: case S_IFSOCK:
3469 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3473 done_path_create(&path, dentry);
3474 if (retry_estale(error, lookup_flags)) {
3475 lookup_flags |= LOOKUP_REVAL;
3481 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3483 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3486 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3488 int error = may_create(dir, dentry);
3489 unsigned max_links = dir->i_sb->s_max_links;
3494 if (!dir->i_op->mkdir)
3497 mode &= (S_IRWXUGO|S_ISVTX);
3498 error = security_inode_mkdir(dir, dentry, mode);
3502 if (max_links && dir->i_nlink >= max_links)
3505 error = dir->i_op->mkdir(dir, dentry, mode);
3507 fsnotify_mkdir(dir, dentry);
3510 EXPORT_SYMBOL(vfs_mkdir);
3512 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3514 struct dentry *dentry;
3517 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3520 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3522 return PTR_ERR(dentry);
3524 if (!IS_POSIXACL(path.dentry->d_inode))
3525 mode &= ~current_umask();
3526 error = security_path_mkdir(&path, dentry, mode);
3528 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3529 done_path_create(&path, dentry);
3530 if (retry_estale(error, lookup_flags)) {
3531 lookup_flags |= LOOKUP_REVAL;
3537 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3539 return sys_mkdirat(AT_FDCWD, pathname, mode);
3543 * The dentry_unhash() helper will try to drop the dentry early: we
3544 * should have a usage count of 1 if we're the only user of this
3545 * dentry, and if that is true (possibly after pruning the dcache),
3546 * then we drop the dentry now.
3548 * A low-level filesystem can, if it choses, legally
3551 * if (!d_unhashed(dentry))
3554 * if it cannot handle the case of removing a directory
3555 * that is still in use by something else..
3557 void dentry_unhash(struct dentry *dentry)
3559 shrink_dcache_parent(dentry);
3560 spin_lock(&dentry->d_lock);
3561 if (dentry->d_lockref.count == 1)
3563 spin_unlock(&dentry->d_lock);
3565 EXPORT_SYMBOL(dentry_unhash);
3567 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3569 int error = may_delete(dir, dentry, 1);
3574 if (!dir->i_op->rmdir)
3578 mutex_lock(&dentry->d_inode->i_mutex);
3581 if (is_local_mountpoint(dentry))
3584 error = security_inode_rmdir(dir, dentry);
3588 shrink_dcache_parent(dentry);
3589 error = dir->i_op->rmdir(dir, dentry);
3593 dentry->d_inode->i_flags |= S_DEAD;
3595 detach_mounts(dentry);
3598 mutex_unlock(&dentry->d_inode->i_mutex);
3604 EXPORT_SYMBOL(vfs_rmdir);
3606 static long do_rmdir(int dfd, const char __user *pathname)
3609 struct filename *name;
3610 struct dentry *dentry;
3611 struct nameidata nd;
3612 unsigned int lookup_flags = 0;
3614 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3616 return PTR_ERR(name);
3618 switch(nd.last_type) {
3630 nd.flags &= ~LOOKUP_PARENT;
3631 error = mnt_want_write(nd.path.mnt);
3635 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3636 dentry = lookup_hash(&nd);
3637 error = PTR_ERR(dentry);
3640 if (!dentry->d_inode) {
3644 error = security_path_rmdir(&nd.path, dentry);
3647 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3651 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3652 mnt_drop_write(nd.path.mnt);
3656 if (retry_estale(error, lookup_flags)) {
3657 lookup_flags |= LOOKUP_REVAL;
3663 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3665 return do_rmdir(AT_FDCWD, pathname);
3669 * vfs_unlink - unlink a filesystem object
3670 * @dir: parent directory
3672 * @delegated_inode: returns victim inode, if the inode is delegated.
3674 * The caller must hold dir->i_mutex.
3676 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3677 * return a reference to the inode in delegated_inode. The caller
3678 * should then break the delegation on that inode and retry. Because
3679 * breaking a delegation may take a long time, the caller should drop
3680 * dir->i_mutex before doing so.
3682 * Alternatively, a caller may pass NULL for delegated_inode. This may
3683 * be appropriate for callers that expect the underlying filesystem not
3684 * to be NFS exported.
3686 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3688 struct inode *target = dentry->d_inode;
3689 int error = may_delete(dir, dentry, 0);
3694 if (!dir->i_op->unlink)
3697 mutex_lock(&target->i_mutex);
3698 if (is_local_mountpoint(dentry))
3701 error = security_inode_unlink(dir, dentry);
3703 error = try_break_deleg(target, delegated_inode);
3706 error = dir->i_op->unlink(dir, dentry);
3709 detach_mounts(dentry);
3714 mutex_unlock(&target->i_mutex);
3716 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3717 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3718 fsnotify_link_count(target);
3724 EXPORT_SYMBOL(vfs_unlink);
3727 * Make sure that the actual truncation of the file will occur outside its
3728 * directory's i_mutex. Truncate can take a long time if there is a lot of
3729 * writeout happening, and we don't want to prevent access to the directory
3730 * while waiting on the I/O.
3732 static long do_unlinkat(int dfd, const char __user *pathname)
3735 struct filename *name;
3736 struct dentry *dentry;
3737 struct nameidata nd;
3738 struct inode *inode = NULL;
3739 struct inode *delegated_inode = NULL;
3740 unsigned int lookup_flags = 0;
3742 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3744 return PTR_ERR(name);
3747 if (nd.last_type != LAST_NORM)
3750 nd.flags &= ~LOOKUP_PARENT;
3751 error = mnt_want_write(nd.path.mnt);
3755 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3756 dentry = lookup_hash(&nd);
3757 error = PTR_ERR(dentry);
3758 if (!IS_ERR(dentry)) {
3759 /* Why not before? Because we want correct error value */
3760 if (nd.last.name[nd.last.len])
3762 inode = dentry->d_inode;
3763 if (d_is_negative(dentry))
3766 error = security_path_unlink(&nd.path, dentry);
3769 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3773 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3775 iput(inode); /* truncate the inode here */
3777 if (delegated_inode) {
3778 error = break_deleg_wait(&delegated_inode);
3782 mnt_drop_write(nd.path.mnt);
3786 if (retry_estale(error, lookup_flags)) {
3787 lookup_flags |= LOOKUP_REVAL;
3794 if (d_is_negative(dentry))
3796 else if (d_is_dir(dentry))
3803 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3805 if ((flag & ~AT_REMOVEDIR) != 0)
3808 if (flag & AT_REMOVEDIR)
3809 return do_rmdir(dfd, pathname);
3811 return do_unlinkat(dfd, pathname);
3814 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3816 return do_unlinkat(AT_FDCWD, pathname);
3819 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3821 int error = may_create(dir, dentry);
3826 if (!dir->i_op->symlink)
3829 error = security_inode_symlink(dir, dentry, oldname);
3833 error = dir->i_op->symlink(dir, dentry, oldname);
3835 fsnotify_create(dir, dentry);
3838 EXPORT_SYMBOL(vfs_symlink);
3840 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3841 int, newdfd, const char __user *, newname)
3844 struct filename *from;
3845 struct dentry *dentry;
3847 unsigned int lookup_flags = 0;
3849 from = getname(oldname);
3851 return PTR_ERR(from);
3853 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3854 error = PTR_ERR(dentry);
3858 error = security_path_symlink(&path, dentry, from->name);
3860 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3861 done_path_create(&path, dentry);
3862 if (retry_estale(error, lookup_flags)) {
3863 lookup_flags |= LOOKUP_REVAL;
3871 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3873 return sys_symlinkat(oldname, AT_FDCWD, newname);
3877 * vfs_link - create a new link
3878 * @old_dentry: object to be linked
3880 * @new_dentry: where to create the new link
3881 * @delegated_inode: returns inode needing a delegation break
3883 * The caller must hold dir->i_mutex
3885 * If vfs_link discovers a delegation on the to-be-linked file in need
3886 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3887 * inode in delegated_inode. The caller should then break the delegation
3888 * and retry. Because breaking a delegation may take a long time, the
3889 * caller should drop the i_mutex before doing so.
3891 * Alternatively, a caller may pass NULL for delegated_inode. This may
3892 * be appropriate for callers that expect the underlying filesystem not
3893 * to be NFS exported.
3895 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3897 struct inode *inode = old_dentry->d_inode;
3898 unsigned max_links = dir->i_sb->s_max_links;
3904 error = may_create(dir, new_dentry);
3908 if (dir->i_sb != inode->i_sb)
3912 * A link to an append-only or immutable file cannot be created.
3914 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3916 if (!dir->i_op->link)
3918 if (S_ISDIR(inode->i_mode))
3921 error = security_inode_link(old_dentry, dir, new_dentry);
3925 mutex_lock(&inode->i_mutex);
3926 /* Make sure we don't allow creating hardlink to an unlinked file */
3927 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3929 else if (max_links && inode->i_nlink >= max_links)
3932 error = try_break_deleg(inode, delegated_inode);
3934 error = dir->i_op->link(old_dentry, dir, new_dentry);
3937 if (!error && (inode->i_state & I_LINKABLE)) {
3938 spin_lock(&inode->i_lock);
3939 inode->i_state &= ~I_LINKABLE;
3940 spin_unlock(&inode->i_lock);
3942 mutex_unlock(&inode->i_mutex);
3944 fsnotify_link(dir, inode, new_dentry);
3947 EXPORT_SYMBOL(vfs_link);
3950 * Hardlinks are often used in delicate situations. We avoid
3951 * security-related surprises by not following symlinks on the
3954 * We don't follow them on the oldname either to be compatible
3955 * with linux 2.0, and to avoid hard-linking to directories
3956 * and other special files. --ADM
3958 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3959 int, newdfd, const char __user *, newname, int, flags)
3961 struct dentry *new_dentry;
3962 struct path old_path, new_path;
3963 struct inode *delegated_inode = NULL;
3967 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3970 * To use null names we require CAP_DAC_READ_SEARCH
3971 * This ensures that not everyone will be able to create
3972 * handlink using the passed filedescriptor.
3974 if (flags & AT_EMPTY_PATH) {
3975 if (!capable(CAP_DAC_READ_SEARCH))
3980 if (flags & AT_SYMLINK_FOLLOW)
3981 how |= LOOKUP_FOLLOW;
3983 error = user_path_at(olddfd, oldname, how, &old_path);
3987 new_dentry = user_path_create(newdfd, newname, &new_path,
3988 (how & LOOKUP_REVAL));
3989 error = PTR_ERR(new_dentry);
3990 if (IS_ERR(new_dentry))
3994 if (old_path.mnt != new_path.mnt)
3996 error = may_linkat(&old_path);
3997 if (unlikely(error))
3999 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4002 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4004 done_path_create(&new_path, new_dentry);
4005 if (delegated_inode) {
4006 error = break_deleg_wait(&delegated_inode);
4008 path_put(&old_path);
4012 if (retry_estale(error, how)) {
4013 path_put(&old_path);
4014 how |= LOOKUP_REVAL;
4018 path_put(&old_path);
4023 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4025 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4029 * vfs_rename - rename a filesystem object
4030 * @old_dir: parent of source
4031 * @old_dentry: source
4032 * @new_dir: parent of destination
4033 * @new_dentry: destination
4034 * @delegated_inode: returns an inode needing a delegation break
4035 * @flags: rename flags
4037 * The caller must hold multiple mutexes--see lock_rename()).
4039 * If vfs_rename discovers a delegation in need of breaking at either
4040 * the source or destination, it will return -EWOULDBLOCK and return a
4041 * reference to the inode in delegated_inode. The caller should then
4042 * break the delegation and retry. Because breaking a delegation may
4043 * take a long time, the caller should drop all locks before doing
4046 * Alternatively, a caller may pass NULL for delegated_inode. This may
4047 * be appropriate for callers that expect the underlying filesystem not
4048 * to be NFS exported.
4050 * The worst of all namespace operations - renaming directory. "Perverted"
4051 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4053 * a) we can get into loop creation.
4054 * b) race potential - two innocent renames can create a loop together.
4055 * That's where 4.4 screws up. Current fix: serialization on
4056 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4058 * c) we have to lock _four_ objects - parents and victim (if it exists),
4059 * and source (if it is not a directory).
4060 * And that - after we got ->i_mutex on parents (until then we don't know
4061 * whether the target exists). Solution: try to be smart with locking
4062 * order for inodes. We rely on the fact that tree topology may change
4063 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4064 * move will be locked. Thus we can rank directories by the tree
4065 * (ancestors first) and rank all non-directories after them.
4066 * That works since everybody except rename does "lock parent, lookup,
4067 * lock child" and rename is under ->s_vfs_rename_mutex.
4068 * HOWEVER, it relies on the assumption that any object with ->lookup()
4069 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4070 * we'd better make sure that there's no link(2) for them.
4071 * d) conversion from fhandle to dentry may come in the wrong moment - when
4072 * we are removing the target. Solution: we will have to grab ->i_mutex
4073 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4074 * ->i_mutex on parents, which works but leads to some truly excessive
4077 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4078 struct inode *new_dir, struct dentry *new_dentry,
4079 struct inode **delegated_inode, unsigned int flags)
4082 bool is_dir = d_is_dir(old_dentry);
4083 const unsigned char *old_name;
4084 struct inode *source = old_dentry->d_inode;
4085 struct inode *target = new_dentry->d_inode;
4086 bool new_is_dir = false;
4087 unsigned max_links = new_dir->i_sb->s_max_links;
4089 if (source == target)
4092 error = may_delete(old_dir, old_dentry, is_dir);
4097 error = may_create(new_dir, new_dentry);
4099 new_is_dir = d_is_dir(new_dentry);
4101 if (!(flags & RENAME_EXCHANGE))
4102 error = may_delete(new_dir, new_dentry, is_dir);
4104 error = may_delete(new_dir, new_dentry, new_is_dir);
4109 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4112 if (flags && !old_dir->i_op->rename2)
4116 * If we are going to change the parent - check write permissions,
4117 * we'll need to flip '..'.
4119 if (new_dir != old_dir) {
4121 error = inode_permission(source, MAY_WRITE);
4125 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4126 error = inode_permission(target, MAY_WRITE);
4132 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4137 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4139 if (!is_dir || (flags & RENAME_EXCHANGE))
4140 lock_two_nondirectories(source, target);
4142 mutex_lock(&target->i_mutex);
4145 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4148 if (max_links && new_dir != old_dir) {
4150 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4152 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4153 old_dir->i_nlink >= max_links)
4156 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4157 shrink_dcache_parent(new_dentry);
4159 error = try_break_deleg(source, delegated_inode);
4163 if (target && !new_is_dir) {
4164 error = try_break_deleg(target, delegated_inode);
4168 if (!old_dir->i_op->rename2) {
4169 error = old_dir->i_op->rename(old_dir, old_dentry,
4170 new_dir, new_dentry);
4172 WARN_ON(old_dir->i_op->rename != NULL);
4173 error = old_dir->i_op->rename2(old_dir, old_dentry,
4174 new_dir, new_dentry, flags);
4179 if (!(flags & RENAME_EXCHANGE) && target) {
4181 target->i_flags |= S_DEAD;
4182 dont_mount(new_dentry);
4183 detach_mounts(new_dentry);
4185 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4186 if (!(flags & RENAME_EXCHANGE))
4187 d_move(old_dentry, new_dentry);
4189 d_exchange(old_dentry, new_dentry);
4192 if (!is_dir || (flags & RENAME_EXCHANGE))
4193 unlock_two_nondirectories(source, target);
4195 mutex_unlock(&target->i_mutex);
4198 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4199 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4200 if (flags & RENAME_EXCHANGE) {
4201 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4202 new_is_dir, NULL, new_dentry);
4205 fsnotify_oldname_free(old_name);
4209 EXPORT_SYMBOL(vfs_rename);
4211 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4212 int, newdfd, const char __user *, newname, unsigned int, flags)
4214 struct dentry *old_dir, *new_dir;
4215 struct dentry *old_dentry, *new_dentry;
4216 struct dentry *trap;
4217 struct nameidata oldnd, newnd;
4218 struct inode *delegated_inode = NULL;
4219 struct filename *from;
4220 struct filename *to;
4221 unsigned int lookup_flags = 0;
4222 bool should_retry = false;
4225 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4228 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4229 (flags & RENAME_EXCHANGE))
4232 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4236 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4238 error = PTR_ERR(from);
4242 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4244 error = PTR_ERR(to);
4249 if (oldnd.path.mnt != newnd.path.mnt)
4252 old_dir = oldnd.path.dentry;
4254 if (oldnd.last_type != LAST_NORM)
4257 new_dir = newnd.path.dentry;
4258 if (flags & RENAME_NOREPLACE)
4260 if (newnd.last_type != LAST_NORM)
4263 error = mnt_want_write(oldnd.path.mnt);
4267 oldnd.flags &= ~LOOKUP_PARENT;
4268 newnd.flags &= ~LOOKUP_PARENT;
4269 if (!(flags & RENAME_EXCHANGE))
4270 newnd.flags |= LOOKUP_RENAME_TARGET;
4273 trap = lock_rename(new_dir, old_dir);
4275 old_dentry = lookup_hash(&oldnd);
4276 error = PTR_ERR(old_dentry);
4277 if (IS_ERR(old_dentry))
4279 /* source must exist */
4281 if (d_is_negative(old_dentry))
4283 new_dentry = lookup_hash(&newnd);
4284 error = PTR_ERR(new_dentry);
4285 if (IS_ERR(new_dentry))
4288 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4290 if (flags & RENAME_EXCHANGE) {
4292 if (d_is_negative(new_dentry))
4295 if (!d_is_dir(new_dentry)) {
4297 if (newnd.last.name[newnd.last.len])
4301 /* unless the source is a directory trailing slashes give -ENOTDIR */
4302 if (!d_is_dir(old_dentry)) {
4304 if (oldnd.last.name[oldnd.last.len])
4306 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
4309 /* source should not be ancestor of target */
4311 if (old_dentry == trap)
4313 /* target should not be an ancestor of source */
4314 if (!(flags & RENAME_EXCHANGE))
4316 if (new_dentry == trap)
4319 error = security_path_rename(&oldnd.path, old_dentry,
4320 &newnd.path, new_dentry, flags);
4323 error = vfs_rename(old_dir->d_inode, old_dentry,
4324 new_dir->d_inode, new_dentry,
4325 &delegated_inode, flags);
4331 unlock_rename(new_dir, old_dir);
4332 if (delegated_inode) {
4333 error = break_deleg_wait(&delegated_inode);
4337 mnt_drop_write(oldnd.path.mnt);
4339 if (retry_estale(error, lookup_flags))
4340 should_retry = true;
4341 path_put(&newnd.path);
4344 path_put(&oldnd.path);
4347 should_retry = false;
4348 lookup_flags |= LOOKUP_REVAL;
4355 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4356 int, newdfd, const char __user *, newname)
4358 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4361 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4363 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4366 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4368 int error = may_create(dir, dentry);
4372 if (!dir->i_op->mknod)
4375 return dir->i_op->mknod(dir, dentry,
4376 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4378 EXPORT_SYMBOL(vfs_whiteout);
4380 int readlink_copy(char __user *buffer, int buflen, const char *link)
4382 int len = PTR_ERR(link);
4387 if (len > (unsigned) buflen)
4389 if (copy_to_user(buffer, link, len))
4394 EXPORT_SYMBOL(readlink_copy);
4397 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4398 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4399 * using) it for any given inode is up to filesystem.
4401 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4403 struct nameidata nd;
4408 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4410 return PTR_ERR(cookie);
4412 res = readlink_copy(buffer, buflen, nd_get_link(&nd));
4413 if (dentry->d_inode->i_op->put_link)
4414 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4417 EXPORT_SYMBOL(generic_readlink);
4419 /* get the link contents into pagecache */
4420 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4424 struct address_space *mapping = dentry->d_inode->i_mapping;
4425 page = read_mapping_page(mapping, 0, NULL);
4430 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4434 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4436 struct page *page = NULL;
4437 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4440 page_cache_release(page);
4444 EXPORT_SYMBOL(page_readlink);
4446 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4448 struct page *page = NULL;
4449 nd_set_link(nd, page_getlink(dentry, &page));
4452 EXPORT_SYMBOL(page_follow_link_light);
4454 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4456 struct page *page = cookie;
4460 page_cache_release(page);
4463 EXPORT_SYMBOL(page_put_link);
4466 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4468 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4470 struct address_space *mapping = inode->i_mapping;
4475 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4477 flags |= AOP_FLAG_NOFS;
4480 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4481 flags, &page, &fsdata);
4485 kaddr = kmap_atomic(page);
4486 memcpy(kaddr, symname, len-1);
4487 kunmap_atomic(kaddr);
4489 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4496 mark_inode_dirty(inode);
4501 EXPORT_SYMBOL(__page_symlink);
4503 int page_symlink(struct inode *inode, const char *symname, int len)
4505 return __page_symlink(inode, symname, len,
4506 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4508 EXPORT_SYMBOL(page_symlink);
4510 const struct inode_operations page_symlink_inode_operations = {
4511 .readlink = generic_readlink,
4512 .follow_link = page_follow_link_light,
4513 .put_link = page_put_link,
4515 EXPORT_SYMBOL(page_symlink_inode_operations);