4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125 getname_flags(const char __user *filename, int flags, int *empty)
127 struct filename *result;
131 result = audit_reusename(filename);
135 result = __getname();
136 if (unlikely(!result))
137 return ERR_PTR(-ENOMEM);
140 * First, try to embed the struct filename inside the names_cache
143 kname = (char *)result->iname;
144 result->name = kname;
146 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
147 if (unlikely(len < 0)) {
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
158 if (unlikely(len == EMBEDDED_NAME_MAX)) {
159 const size_t size = offsetof(struct filename, iname[1]);
160 kname = (char *)result;
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
167 result = kzalloc(size, GFP_KERNEL);
168 if (unlikely(!result)) {
170 return ERR_PTR(-ENOMEM);
172 result->name = kname;
173 len = strncpy_from_user(kname, filename, PATH_MAX);
174 if (unlikely(len < 0)) {
179 if (unlikely(len == PATH_MAX)) {
182 return ERR_PTR(-ENAMETOOLONG);
187 /* The empty path is special. */
188 if (unlikely(!len)) {
191 if (!(flags & LOOKUP_EMPTY)) {
193 return ERR_PTR(-ENOENT);
197 result->uptr = filename;
198 result->aname = NULL;
199 audit_getname(result);
204 getname(const char __user * filename)
206 return getname_flags(filename, 0, NULL);
210 getname_kernel(const char * filename)
212 struct filename *result;
213 int len = strlen(filename) + 1;
215 result = __getname();
216 if (unlikely(!result))
217 return ERR_PTR(-ENOMEM);
219 if (len <= EMBEDDED_NAME_MAX) {
220 result->name = (char *)result->iname;
221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
227 return ERR_PTR(-ENOMEM);
229 tmp->name = (char *)result;
233 return ERR_PTR(-ENAMETOOLONG);
235 memcpy((char *)result->name, filename, len);
237 result->aname = NULL;
239 audit_getname(result);
244 void putname(struct filename *name)
246 BUG_ON(name->refcnt <= 0);
248 if (--name->refcnt > 0)
251 if (name->name != name->iname) {
252 __putname(name->name);
258 static int check_acl(struct inode *inode, int mask)
260 #ifdef CONFIG_FS_POSIX_ACL
261 struct posix_acl *acl;
263 if (mask & MAY_NOT_BLOCK) {
264 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 /* no ->get_acl() calls in RCU mode... */
268 if (acl == ACL_NOT_CACHED)
270 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273 acl = get_acl(inode, ACL_TYPE_ACCESS);
277 int error = posix_acl_permission(inode, acl, mask);
278 posix_acl_release(acl);
287 * This does the basic permission checking
289 static int acl_permission_check(struct inode *inode, int mask)
291 unsigned int mode = inode->i_mode;
293 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
297 int error = check_acl(inode, mask);
298 if (error != -EAGAIN)
302 if (in_group_p(inode->i_gid))
307 * If the DACs are ok we don't need any capability check.
309 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
315 * generic_permission - check for access rights on a Posix-like filesystem
316 * @inode: inode to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
322 * are used for other things.
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
328 int generic_permission(struct inode *inode, int mask)
333 * Do the basic permission checks.
335 ret = acl_permission_check(inode, mask);
339 if (S_ISDIR(inode->i_mode)) {
340 /* DACs are overridable for directories */
341 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
343 if (!(mask & MAY_WRITE))
344 if (capable_wrt_inode_uidgid(inode,
345 CAP_DAC_READ_SEARCH))
350 * Read/write DACs are always overridable.
351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
354 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
355 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
359 * Searching includes executable on directories, else just read.
361 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
362 if (mask == MAY_READ)
363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
368 EXPORT_SYMBOL(generic_permission);
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
376 static inline int do_inode_permission(struct inode *inode, int mask)
378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379 if (likely(inode->i_op->permission))
380 return inode->i_op->permission(inode, mask);
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode->i_lock);
384 inode->i_opflags |= IOP_FASTPERM;
385 spin_unlock(&inode->i_lock);
387 return generic_permission(inode, mask);
391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395 * Check for read/write/execute permissions on an inode.
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
402 int __inode_permission(struct inode *inode, int mask)
406 if (unlikely(mask & MAY_WRITE)) {
408 * Nobody gets write access to an immutable file.
410 if (IS_IMMUTABLE(inode))
414 retval = do_inode_permission(inode, mask);
418 retval = devcgroup_inode_permission(inode, mask);
422 return security_inode_permission(inode, mask);
424 EXPORT_SYMBOL(__inode_permission);
427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Separate out file-system wide checks from inode-specific permission checks.
434 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
436 if (unlikely(mask & MAY_WRITE)) {
437 umode_t mode = inode->i_mode;
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb->s_flags & MS_RDONLY) &&
441 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 int inode_permission(struct inode *inode, int mask)
462 retval = sb_permission(inode->i_sb, inode, mask);
465 return __inode_permission(inode, mask);
467 EXPORT_SYMBOL(inode_permission);
470 * path_get - get a reference to a path
471 * @path: path to get the reference to
473 * Given a path increment the reference count to the dentry and the vfsmount.
475 void path_get(const struct path *path)
480 EXPORT_SYMBOL(path_get);
483 * path_put - put a reference to a path
484 * @path: path to put the reference to
486 * Given a path decrement the reference count to the dentry and the vfsmount.
488 void path_put(const struct path *path)
493 EXPORT_SYMBOL(path_put);
495 #define EMBEDDED_LEVELS 2
500 struct inode *inode; /* path.dentry.d_inode */
505 int total_link_count;
511 } *stack, internal[EMBEDDED_LEVELS];
514 static struct nameidata *set_nameidata(struct nameidata *p)
516 struct nameidata *old = current->nameidata;
517 p->stack = p->internal;
518 p->total_link_count = old ? old->total_link_count : 0;
519 current->nameidata = p;
523 static void restore_nameidata(struct nameidata *old)
525 struct nameidata *now = current->nameidata;
527 current->nameidata = old;
529 old->total_link_count = now->total_link_count;
530 if (now->stack != now->internal) {
532 now->stack = now->internal;
536 static int __nd_alloc_stack(struct nameidata *nd)
538 struct saved *p = kmalloc(MAXSYMLINKS * sizeof(struct saved),
542 memcpy(p, nd->internal, sizeof(nd->internal));
547 static inline int nd_alloc_stack(struct nameidata *nd)
549 if (likely(nd->depth != EMBEDDED_LEVELS))
551 if (likely(nd->stack != nd->internal))
553 return __nd_alloc_stack(nd);
557 * Path walking has 2 modes, rcu-walk and ref-walk (see
558 * Documentation/filesystems/path-lookup.txt). In situations when we can't
559 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
560 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
561 * mode. Refcounts are grabbed at the last known good point before rcu-walk
562 * got stuck, so ref-walk may continue from there. If this is not successful
563 * (eg. a seqcount has changed), then failure is returned and it's up to caller
564 * to restart the path walk from the beginning in ref-walk mode.
568 * unlazy_walk - try to switch to ref-walk mode.
569 * @nd: nameidata pathwalk data
570 * @dentry: child of nd->path.dentry or NULL
571 * @seq: seq number to check dentry against
572 * Returns: 0 on success, -ECHILD on failure
574 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
575 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
576 * @nd or NULL. Must be called from rcu-walk context.
578 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
580 struct fs_struct *fs = current->fs;
581 struct dentry *parent = nd->path.dentry;
583 BUG_ON(!(nd->flags & LOOKUP_RCU));
586 * After legitimizing the bastards, terminate_walk()
587 * will do the right thing for non-RCU mode, and all our
588 * subsequent exit cases should rcu_read_unlock()
589 * before returning. Do vfsmount first; if dentry
590 * can't be legitimized, just set nd->path.dentry to NULL
591 * and rely on dput(NULL) being a no-op.
593 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
595 nd->flags &= ~LOOKUP_RCU;
597 if (!lockref_get_not_dead(&parent->d_lockref)) {
598 nd->path.dentry = NULL;
603 * For a negative lookup, the lookup sequence point is the parents
604 * sequence point, and it only needs to revalidate the parent dentry.
606 * For a positive lookup, we need to move both the parent and the
607 * dentry from the RCU domain to be properly refcounted. And the
608 * sequence number in the dentry validates *both* dentry counters,
609 * since we checked the sequence number of the parent after we got
610 * the child sequence number. So we know the parent must still
611 * be valid if the child sequence number is still valid.
614 if (read_seqcount_retry(&parent->d_seq, nd->seq))
616 BUG_ON(nd->inode != parent->d_inode);
618 if (!lockref_get_not_dead(&dentry->d_lockref))
620 if (read_seqcount_retry(&dentry->d_seq, seq))
625 * Sequence counts matched. Now make sure that the root is
626 * still valid and get it if required.
628 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
629 spin_lock(&fs->lock);
630 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
631 goto unlock_and_drop_dentry;
633 spin_unlock(&fs->lock);
639 unlock_and_drop_dentry:
640 spin_unlock(&fs->lock);
648 if (!(nd->flags & LOOKUP_ROOT))
653 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
655 return dentry->d_op->d_revalidate(dentry, flags);
659 * complete_walk - successful completion of path walk
660 * @nd: pointer nameidata
662 * If we had been in RCU mode, drop out of it and legitimize nd->path.
663 * Revalidate the final result, unless we'd already done that during
664 * the path walk or the filesystem doesn't ask for it. Return 0 on
665 * success, -error on failure. In case of failure caller does not
666 * need to drop nd->path.
668 static int complete_walk(struct nameidata *nd)
670 struct dentry *dentry = nd->path.dentry;
673 if (nd->flags & LOOKUP_RCU) {
674 if (!(nd->flags & LOOKUP_ROOT))
676 if (unlikely(unlazy_walk(nd, NULL, 0)))
680 if (likely(!(nd->flags & LOOKUP_JUMPED)))
683 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
686 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
696 static __always_inline void set_root(struct nameidata *nd)
698 get_fs_root(current->fs, &nd->root);
701 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
703 struct fs_struct *fs = current->fs;
707 seq = read_seqcount_begin(&fs->seq);
709 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
710 } while (read_seqcount_retry(&fs->seq, seq));
714 static void path_put_conditional(struct path *path, struct nameidata *nd)
717 if (path->mnt != nd->path.mnt)
721 static inline void path_to_nameidata(const struct path *path,
722 struct nameidata *nd)
724 if (!(nd->flags & LOOKUP_RCU)) {
725 dput(nd->path.dentry);
726 if (nd->path.mnt != path->mnt)
727 mntput(nd->path.mnt);
729 nd->path.mnt = path->mnt;
730 nd->path.dentry = path->dentry;
734 * Helper to directly jump to a known parsed path from ->follow_link,
735 * caller must have taken a reference to path beforehand.
737 void nd_jump_link(struct path *path)
739 struct nameidata *nd = current->nameidata;
743 nd->inode = nd->path.dentry->d_inode;
744 nd->flags |= LOOKUP_JUMPED;
747 static inline void put_link(struct nameidata *nd)
749 struct saved *last = nd->stack + --nd->depth;
750 struct inode *inode = last->inode;
751 if (last->cookie && inode->i_op->put_link)
752 inode->i_op->put_link(inode, last->cookie);
753 path_put(&last->link);
756 int sysctl_protected_symlinks __read_mostly = 0;
757 int sysctl_protected_hardlinks __read_mostly = 0;
760 * may_follow_link - Check symlink following for unsafe situations
761 * @nd: nameidata pathwalk data
763 * In the case of the sysctl_protected_symlinks sysctl being enabled,
764 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
765 * in a sticky world-writable directory. This is to protect privileged
766 * processes from failing races against path names that may change out
767 * from under them by way of other users creating malicious symlinks.
768 * It will permit symlinks to be followed only when outside a sticky
769 * world-writable directory, or when the uid of the symlink and follower
770 * match, or when the directory owner matches the symlink's owner.
772 * Returns 0 if following the symlink is allowed, -ve on error.
774 static inline int may_follow_link(struct nameidata *nd)
776 const struct inode *inode;
777 const struct inode *parent;
779 if (!sysctl_protected_symlinks)
782 /* Allowed if owner and follower match. */
783 inode = nd->stack[0].inode;
784 if (uid_eq(current_cred()->fsuid, inode->i_uid))
787 /* Allowed if parent directory not sticky and world-writable. */
788 parent = nd->path.dentry->d_inode;
789 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
792 /* Allowed if parent directory and link owner match. */
793 if (uid_eq(parent->i_uid, inode->i_uid))
796 audit_log_link_denied("follow_link", &nd->stack[0].link);
801 * safe_hardlink_source - Check for safe hardlink conditions
802 * @inode: the source inode to hardlink from
804 * Return false if at least one of the following conditions:
805 * - inode is not a regular file
807 * - inode is setgid and group-exec
808 * - access failure for read and write
810 * Otherwise returns true.
812 static bool safe_hardlink_source(struct inode *inode)
814 umode_t mode = inode->i_mode;
816 /* Special files should not get pinned to the filesystem. */
820 /* Setuid files should not get pinned to the filesystem. */
824 /* Executable setgid files should not get pinned to the filesystem. */
825 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
828 /* Hardlinking to unreadable or unwritable sources is dangerous. */
829 if (inode_permission(inode, MAY_READ | MAY_WRITE))
836 * may_linkat - Check permissions for creating a hardlink
837 * @link: the source to hardlink from
839 * Block hardlink when all of:
840 * - sysctl_protected_hardlinks enabled
841 * - fsuid does not match inode
842 * - hardlink source is unsafe (see safe_hardlink_source() above)
845 * Returns 0 if successful, -ve on error.
847 static int may_linkat(struct path *link)
849 const struct cred *cred;
852 if (!sysctl_protected_hardlinks)
855 cred = current_cred();
856 inode = link->dentry->d_inode;
858 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
859 * otherwise, it must be a safe source.
861 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
865 audit_log_link_denied("linkat", link);
869 static __always_inline
870 const char *get_link(struct nameidata *nd)
872 struct saved *last = nd->stack + nd->depth - 1;
873 struct dentry *dentry = last->link.dentry;
874 struct inode *inode = last->inode;
878 BUG_ON(nd->flags & LOOKUP_RCU);
882 touch_atime(&last->link);
884 error = security_inode_follow_link(dentry, inode,
885 nd->flags & LOOKUP_RCU);
887 return ERR_PTR(error);
889 nd->last_type = LAST_BIND;
892 res = inode->i_op->follow_link(dentry, &last->cookie);
893 if (IS_ERR_OR_NULL(res)) {
904 nd->inode = nd->path.dentry->d_inode;
905 nd->flags |= LOOKUP_JUMPED;
906 while (unlikely(*++res == '/'))
914 static int follow_up_rcu(struct path *path)
916 struct mount *mnt = real_mount(path->mnt);
917 struct mount *parent;
918 struct dentry *mountpoint;
920 parent = mnt->mnt_parent;
921 if (&parent->mnt == path->mnt)
923 mountpoint = mnt->mnt_mountpoint;
924 path->dentry = mountpoint;
925 path->mnt = &parent->mnt;
930 * follow_up - Find the mountpoint of path's vfsmount
932 * Given a path, find the mountpoint of its source file system.
933 * Replace @path with the path of the mountpoint in the parent mount.
936 * Return 1 if we went up a level and 0 if we were already at the
939 int follow_up(struct path *path)
941 struct mount *mnt = real_mount(path->mnt);
942 struct mount *parent;
943 struct dentry *mountpoint;
945 read_seqlock_excl(&mount_lock);
946 parent = mnt->mnt_parent;
948 read_sequnlock_excl(&mount_lock);
951 mntget(&parent->mnt);
952 mountpoint = dget(mnt->mnt_mountpoint);
953 read_sequnlock_excl(&mount_lock);
955 path->dentry = mountpoint;
957 path->mnt = &parent->mnt;
960 EXPORT_SYMBOL(follow_up);
963 * Perform an automount
964 * - return -EISDIR to tell follow_managed() to stop and return the path we
967 static int follow_automount(struct path *path, struct nameidata *nd,
970 struct vfsmount *mnt;
973 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
976 /* We don't want to mount if someone's just doing a stat -
977 * unless they're stat'ing a directory and appended a '/' to
980 * We do, however, want to mount if someone wants to open or
981 * create a file of any type under the mountpoint, wants to
982 * traverse through the mountpoint or wants to open the
983 * mounted directory. Also, autofs may mark negative dentries
984 * as being automount points. These will need the attentions
985 * of the daemon to instantiate them before they can be used.
987 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
988 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
989 path->dentry->d_inode)
992 nd->total_link_count++;
993 if (nd->total_link_count >= 40)
996 mnt = path->dentry->d_op->d_automount(path);
999 * The filesystem is allowed to return -EISDIR here to indicate
1000 * it doesn't want to automount. For instance, autofs would do
1001 * this so that its userspace daemon can mount on this dentry.
1003 * However, we can only permit this if it's a terminal point in
1004 * the path being looked up; if it wasn't then the remainder of
1005 * the path is inaccessible and we should say so.
1007 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1009 return PTR_ERR(mnt);
1012 if (!mnt) /* mount collision */
1015 if (!*need_mntput) {
1016 /* lock_mount() may release path->mnt on error */
1018 *need_mntput = true;
1020 err = finish_automount(mnt, path);
1024 /* Someone else made a mount here whilst we were busy */
1029 path->dentry = dget(mnt->mnt_root);
1038 * Handle a dentry that is managed in some way.
1039 * - Flagged for transit management (autofs)
1040 * - Flagged as mountpoint
1041 * - Flagged as automount point
1043 * This may only be called in refwalk mode.
1045 * Serialization is taken care of in namespace.c
1047 static int follow_managed(struct path *path, struct nameidata *nd)
1049 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1051 bool need_mntput = false;
1054 /* Given that we're not holding a lock here, we retain the value in a
1055 * local variable for each dentry as we look at it so that we don't see
1056 * the components of that value change under us */
1057 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1058 managed &= DCACHE_MANAGED_DENTRY,
1059 unlikely(managed != 0)) {
1060 /* Allow the filesystem to manage the transit without i_mutex
1062 if (managed & DCACHE_MANAGE_TRANSIT) {
1063 BUG_ON(!path->dentry->d_op);
1064 BUG_ON(!path->dentry->d_op->d_manage);
1065 ret = path->dentry->d_op->d_manage(path->dentry, false);
1070 /* Transit to a mounted filesystem. */
1071 if (managed & DCACHE_MOUNTED) {
1072 struct vfsmount *mounted = lookup_mnt(path);
1077 path->mnt = mounted;
1078 path->dentry = dget(mounted->mnt_root);
1083 /* Something is mounted on this dentry in another
1084 * namespace and/or whatever was mounted there in this
1085 * namespace got unmounted before lookup_mnt() could
1089 /* Handle an automount point */
1090 if (managed & DCACHE_NEED_AUTOMOUNT) {
1091 ret = follow_automount(path, nd, &need_mntput);
1097 /* We didn't change the current path point */
1101 if (need_mntput && path->mnt == mnt)
1106 nd->flags |= LOOKUP_JUMPED;
1107 if (unlikely(ret < 0))
1108 path_put_conditional(path, nd);
1112 int follow_down_one(struct path *path)
1114 struct vfsmount *mounted;
1116 mounted = lookup_mnt(path);
1120 path->mnt = mounted;
1121 path->dentry = dget(mounted->mnt_root);
1126 EXPORT_SYMBOL(follow_down_one);
1128 static inline int managed_dentry_rcu(struct dentry *dentry)
1130 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1131 dentry->d_op->d_manage(dentry, true) : 0;
1135 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1136 * we meet a managed dentry that would need blocking.
1138 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1139 struct inode **inode, unsigned *seqp)
1142 struct mount *mounted;
1144 * Don't forget we might have a non-mountpoint managed dentry
1145 * that wants to block transit.
1147 switch (managed_dentry_rcu(path->dentry)) {
1157 if (!d_mountpoint(path->dentry))
1158 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1160 mounted = __lookup_mnt(path->mnt, path->dentry);
1163 path->mnt = &mounted->mnt;
1164 path->dentry = mounted->mnt.mnt_root;
1165 nd->flags |= LOOKUP_JUMPED;
1166 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1168 * Update the inode too. We don't need to re-check the
1169 * dentry sequence number here after this d_inode read,
1170 * because a mount-point is always pinned.
1172 *inode = path->dentry->d_inode;
1174 return !read_seqretry(&mount_lock, nd->m_seq) &&
1175 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1178 static int follow_dotdot_rcu(struct nameidata *nd)
1180 struct inode *inode = nd->inode;
1185 if (nd->path.dentry == nd->root.dentry &&
1186 nd->path.mnt == nd->root.mnt) {
1189 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1190 struct dentry *old = nd->path.dentry;
1191 struct dentry *parent = old->d_parent;
1194 inode = parent->d_inode;
1195 seq = read_seqcount_begin(&parent->d_seq);
1196 if (read_seqcount_retry(&old->d_seq, nd->seq))
1198 nd->path.dentry = parent;
1202 if (!follow_up_rcu(&nd->path))
1204 inode = nd->path.dentry->d_inode;
1205 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1207 while (d_mountpoint(nd->path.dentry)) {
1208 struct mount *mounted;
1209 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1212 nd->path.mnt = &mounted->mnt;
1213 nd->path.dentry = mounted->mnt.mnt_root;
1214 inode = nd->path.dentry->d_inode;
1215 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1216 if (read_seqretry(&mount_lock, nd->m_seq))
1227 * Follow down to the covering mount currently visible to userspace. At each
1228 * point, the filesystem owning that dentry may be queried as to whether the
1229 * caller is permitted to proceed or not.
1231 int follow_down(struct path *path)
1236 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1237 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1238 /* Allow the filesystem to manage the transit without i_mutex
1241 * We indicate to the filesystem if someone is trying to mount
1242 * something here. This gives autofs the chance to deny anyone
1243 * other than its daemon the right to mount on its
1246 * The filesystem may sleep at this point.
1248 if (managed & DCACHE_MANAGE_TRANSIT) {
1249 BUG_ON(!path->dentry->d_op);
1250 BUG_ON(!path->dentry->d_op->d_manage);
1251 ret = path->dentry->d_op->d_manage(
1252 path->dentry, false);
1254 return ret == -EISDIR ? 0 : ret;
1257 /* Transit to a mounted filesystem. */
1258 if (managed & DCACHE_MOUNTED) {
1259 struct vfsmount *mounted = lookup_mnt(path);
1264 path->mnt = mounted;
1265 path->dentry = dget(mounted->mnt_root);
1269 /* Don't handle automount points here */
1274 EXPORT_SYMBOL(follow_down);
1277 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1279 static void follow_mount(struct path *path)
1281 while (d_mountpoint(path->dentry)) {
1282 struct vfsmount *mounted = lookup_mnt(path);
1287 path->mnt = mounted;
1288 path->dentry = dget(mounted->mnt_root);
1292 static void follow_dotdot(struct nameidata *nd)
1298 struct dentry *old = nd->path.dentry;
1300 if (nd->path.dentry == nd->root.dentry &&
1301 nd->path.mnt == nd->root.mnt) {
1304 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1305 /* rare case of legitimate dget_parent()... */
1306 nd->path.dentry = dget_parent(nd->path.dentry);
1310 if (!follow_up(&nd->path))
1313 follow_mount(&nd->path);
1314 nd->inode = nd->path.dentry->d_inode;
1318 * This looks up the name in dcache, possibly revalidates the old dentry and
1319 * allocates a new one if not found or not valid. In the need_lookup argument
1320 * returns whether i_op->lookup is necessary.
1322 * dir->d_inode->i_mutex must be held
1324 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1325 unsigned int flags, bool *need_lookup)
1327 struct dentry *dentry;
1330 *need_lookup = false;
1331 dentry = d_lookup(dir, name);
1333 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1334 error = d_revalidate(dentry, flags);
1335 if (unlikely(error <= 0)) {
1338 return ERR_PTR(error);
1340 d_invalidate(dentry);
1349 dentry = d_alloc(dir, name);
1350 if (unlikely(!dentry))
1351 return ERR_PTR(-ENOMEM);
1353 *need_lookup = true;
1359 * Call i_op->lookup on the dentry. The dentry must be negative and
1362 * dir->d_inode->i_mutex must be held
1364 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1369 /* Don't create child dentry for a dead directory. */
1370 if (unlikely(IS_DEADDIR(dir))) {
1372 return ERR_PTR(-ENOENT);
1375 old = dir->i_op->lookup(dir, dentry, flags);
1376 if (unlikely(old)) {
1383 static struct dentry *__lookup_hash(struct qstr *name,
1384 struct dentry *base, unsigned int flags)
1387 struct dentry *dentry;
1389 dentry = lookup_dcache(name, base, flags, &need_lookup);
1393 return lookup_real(base->d_inode, dentry, flags);
1397 * It's more convoluted than I'd like it to be, but... it's still fairly
1398 * small and for now I'd prefer to have fast path as straight as possible.
1399 * It _is_ time-critical.
1401 static int lookup_fast(struct nameidata *nd,
1402 struct path *path, struct inode **inode,
1405 struct vfsmount *mnt = nd->path.mnt;
1406 struct dentry *dentry, *parent = nd->path.dentry;
1412 * Rename seqlock is not required here because in the off chance
1413 * of a false negative due to a concurrent rename, we're going to
1414 * do the non-racy lookup, below.
1416 if (nd->flags & LOOKUP_RCU) {
1419 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1424 * This sequence count validates that the inode matches
1425 * the dentry name information from lookup.
1427 *inode = d_backing_inode(dentry);
1428 negative = d_is_negative(dentry);
1429 if (read_seqcount_retry(&dentry->d_seq, seq))
1435 * This sequence count validates that the parent had no
1436 * changes while we did the lookup of the dentry above.
1438 * The memory barrier in read_seqcount_begin of child is
1439 * enough, we can use __read_seqcount_retry here.
1441 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1445 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1446 status = d_revalidate(dentry, nd->flags);
1447 if (unlikely(status <= 0)) {
1448 if (status != -ECHILD)
1454 path->dentry = dentry;
1455 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1458 if (unlazy_walk(nd, dentry, seq))
1461 dentry = __d_lookup(parent, &nd->last);
1464 if (unlikely(!dentry))
1467 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1468 status = d_revalidate(dentry, nd->flags);
1469 if (unlikely(status <= 0)) {
1474 d_invalidate(dentry);
1479 if (unlikely(d_is_negative(dentry))) {
1484 path->dentry = dentry;
1485 err = follow_managed(path, nd);
1487 *inode = d_backing_inode(path->dentry);
1494 /* Fast lookup failed, do it the slow way */
1495 static int lookup_slow(struct nameidata *nd, struct path *path)
1497 struct dentry *dentry, *parent;
1499 parent = nd->path.dentry;
1500 BUG_ON(nd->inode != parent->d_inode);
1502 mutex_lock(&parent->d_inode->i_mutex);
1503 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1504 mutex_unlock(&parent->d_inode->i_mutex);
1506 return PTR_ERR(dentry);
1507 path->mnt = nd->path.mnt;
1508 path->dentry = dentry;
1509 return follow_managed(path, nd);
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, 0))
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 return follow_dotdot_rcu(nd);
1535 static void terminate_walk(struct nameidata *nd)
1537 if (!(nd->flags & LOOKUP_RCU)) {
1538 path_put(&nd->path);
1540 nd->flags &= ~LOOKUP_RCU;
1541 if (!(nd->flags & LOOKUP_ROOT))
1542 nd->root.mnt = NULL;
1545 while (unlikely(nd->depth))
1549 static int pick_link(struct nameidata *nd, struct path *link,
1550 struct inode *inode, unsigned seq)
1554 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1555 path_to_nameidata(link, nd);
1558 if (nd->flags & LOOKUP_RCU) {
1559 if (unlikely(nd->path.mnt != link->mnt ||
1560 unlazy_walk(nd, link->dentry, seq))) {
1564 if (link->mnt == nd->path.mnt)
1566 error = nd_alloc_stack(nd);
1567 if (unlikely(error)) {
1572 last = nd->stack + nd->depth++;
1574 last->cookie = NULL;
1575 last->inode = inode;
1580 * Do we need to follow links? We _really_ want to be able
1581 * to do this check without having to look at inode->i_op,
1582 * so we keep a cache of "no, this doesn't need follow_link"
1583 * for the common case.
1585 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1587 struct inode *inode, unsigned seq)
1589 if (likely(!d_is_symlink(link->dentry)))
1593 return pick_link(nd, link, inode, seq);
1596 enum {WALK_GET = 1, WALK_PUT = 2};
1598 static int walk_component(struct nameidata *nd, int flags)
1601 struct inode *inode;
1605 * "." and ".." are special - ".." especially so because it has
1606 * to be able to know about the current root directory and
1607 * parent relationships.
1609 if (unlikely(nd->last_type != LAST_NORM)) {
1610 err = handle_dots(nd, nd->last_type);
1611 if (flags & WALK_PUT)
1615 err = lookup_fast(nd, &path, &inode, &seq);
1616 if (unlikely(err)) {
1620 err = lookup_slow(nd, &path);
1624 inode = d_backing_inode(path.dentry);
1625 seq = 0; /* we are already out of RCU mode */
1627 if (d_is_negative(path.dentry))
1631 if (flags & WALK_PUT)
1633 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1636 path_to_nameidata(&path, nd);
1642 path_to_nameidata(&path, nd);
1647 * We can do the critical dentry name comparison and hashing
1648 * operations one word at a time, but we are limited to:
1650 * - Architectures with fast unaligned word accesses. We could
1651 * do a "get_unaligned()" if this helps and is sufficiently
1654 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1655 * do not trap on the (extremely unlikely) case of a page
1656 * crossing operation.
1658 * - Furthermore, we need an efficient 64-bit compile for the
1659 * 64-bit case in order to generate the "number of bytes in
1660 * the final mask". Again, that could be replaced with a
1661 * efficient population count instruction or similar.
1663 #ifdef CONFIG_DCACHE_WORD_ACCESS
1665 #include <asm/word-at-a-time.h>
1669 static inline unsigned int fold_hash(unsigned long hash)
1671 return hash_64(hash, 32);
1674 #else /* 32-bit case */
1676 #define fold_hash(x) (x)
1680 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1682 unsigned long a, mask;
1683 unsigned long hash = 0;
1686 a = load_unaligned_zeropad(name);
1687 if (len < sizeof(unsigned long))
1691 name += sizeof(unsigned long);
1692 len -= sizeof(unsigned long);
1696 mask = bytemask_from_count(len);
1699 return fold_hash(hash);
1701 EXPORT_SYMBOL(full_name_hash);
1704 * Calculate the length and hash of the path component, and
1705 * return the "hash_len" as the result.
1707 static inline u64 hash_name(const char *name)
1709 unsigned long a, b, adata, bdata, mask, hash, len;
1710 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1713 len = -sizeof(unsigned long);
1715 hash = (hash + a) * 9;
1716 len += sizeof(unsigned long);
1717 a = load_unaligned_zeropad(name+len);
1718 b = a ^ REPEAT_BYTE('/');
1719 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1721 adata = prep_zero_mask(a, adata, &constants);
1722 bdata = prep_zero_mask(b, bdata, &constants);
1724 mask = create_zero_mask(adata | bdata);
1726 hash += a & zero_bytemask(mask);
1727 len += find_zero(mask);
1728 return hashlen_create(fold_hash(hash), len);
1733 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1735 unsigned long hash = init_name_hash();
1737 hash = partial_name_hash(*name++, hash);
1738 return end_name_hash(hash);
1740 EXPORT_SYMBOL(full_name_hash);
1743 * We know there's a real path component here of at least
1746 static inline u64 hash_name(const char *name)
1748 unsigned long hash = init_name_hash();
1749 unsigned long len = 0, c;
1751 c = (unsigned char)*name;
1754 hash = partial_name_hash(c, hash);
1755 c = (unsigned char)name[len];
1756 } while (c && c != '/');
1757 return hashlen_create(end_name_hash(hash), len);
1764 * This is the basic name resolution function, turning a pathname into
1765 * the final dentry. We expect 'base' to be positive and a directory.
1767 * Returns 0 and nd will have valid dentry and mnt on success.
1768 * Returns error and drops reference to input namei data on failure.
1770 static int link_path_walk(const char *name, struct nameidata *nd)
1779 /* At this point we know we have a real path component. */
1784 err = may_lookup(nd);
1788 hash_len = hash_name(name);
1791 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1793 if (name[1] == '.') {
1795 nd->flags |= LOOKUP_JUMPED;
1801 if (likely(type == LAST_NORM)) {
1802 struct dentry *parent = nd->path.dentry;
1803 nd->flags &= ~LOOKUP_JUMPED;
1804 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1805 struct qstr this = { { .hash_len = hash_len }, .name = name };
1806 err = parent->d_op->d_hash(parent, &this);
1809 hash_len = this.hash_len;
1814 nd->last.hash_len = hash_len;
1815 nd->last.name = name;
1816 nd->last_type = type;
1818 name += hashlen_len(hash_len);
1822 * If it wasn't NUL, we know it was '/'. Skip that
1823 * slash, and continue until no more slashes.
1827 } while (unlikely(*name == '/'));
1828 if (unlikely(!*name)) {
1830 /* pathname body, done */
1833 name = nd->stack[nd->depth - 1].name;
1834 /* trailing symlink, done */
1837 /* last component of nested symlink */
1838 err = walk_component(nd, WALK_GET | WALK_PUT);
1840 err = walk_component(nd, WALK_GET);
1846 const char *s = get_link(nd);
1848 if (unlikely(IS_ERR(s)))
1855 nd->stack[nd->depth - 1].name = name;
1860 if (unlikely(!d_can_lookup(nd->path.dentry)))
1865 static const char *path_init(int dfd, const struct filename *name,
1866 unsigned int flags, struct nameidata *nd)
1869 const char *s = name->name;
1871 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1872 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1874 nd->total_link_count = 0;
1875 if (flags & LOOKUP_ROOT) {
1876 struct dentry *root = nd->root.dentry;
1877 struct inode *inode = root->d_inode;
1879 if (!d_can_lookup(root))
1880 return ERR_PTR(-ENOTDIR);
1881 retval = inode_permission(inode, MAY_EXEC);
1883 return ERR_PTR(retval);
1885 nd->path = nd->root;
1887 if (flags & LOOKUP_RCU) {
1889 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1890 nd->m_seq = read_seqbegin(&mount_lock);
1892 path_get(&nd->path);
1897 nd->root.mnt = NULL;
1899 nd->m_seq = read_seqbegin(&mount_lock);
1901 if (flags & LOOKUP_RCU) {
1903 nd->seq = set_root_rcu(nd);
1906 path_get(&nd->root);
1908 nd->path = nd->root;
1909 } else if (dfd == AT_FDCWD) {
1910 if (flags & LOOKUP_RCU) {
1911 struct fs_struct *fs = current->fs;
1917 seq = read_seqcount_begin(&fs->seq);
1919 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1920 } while (read_seqcount_retry(&fs->seq, seq));
1922 get_fs_pwd(current->fs, &nd->path);
1925 /* Caller must check execute permissions on the starting path component */
1926 struct fd f = fdget_raw(dfd);
1927 struct dentry *dentry;
1930 return ERR_PTR(-EBADF);
1932 dentry = f.file->f_path.dentry;
1935 if (!d_can_lookup(dentry)) {
1937 return ERR_PTR(-ENOTDIR);
1941 nd->path = f.file->f_path;
1942 if (flags & LOOKUP_RCU) {
1944 nd->inode = nd->path.dentry->d_inode;
1945 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1947 path_get(&nd->path);
1948 nd->inode = nd->path.dentry->d_inode;
1954 nd->inode = nd->path.dentry->d_inode;
1955 if (!(flags & LOOKUP_RCU))
1957 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1959 if (!(nd->flags & LOOKUP_ROOT))
1960 nd->root.mnt = NULL;
1962 return ERR_PTR(-ECHILD);
1965 static void path_cleanup(struct nameidata *nd)
1967 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1968 path_put(&nd->root);
1969 nd->root.mnt = NULL;
1973 static const char *trailing_symlink(struct nameidata *nd)
1976 int error = may_follow_link(nd);
1977 if (unlikely(error))
1978 return ERR_PTR(error);
1979 nd->flags |= LOOKUP_PARENT;
1980 nd->stack[0].name = NULL;
1985 static inline int lookup_last(struct nameidata *nd)
1987 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1988 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1990 nd->flags &= ~LOOKUP_PARENT;
1991 return walk_component(nd,
1992 nd->flags & LOOKUP_FOLLOW
1994 ? WALK_PUT | WALK_GET
1999 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2000 static int path_lookupat(int dfd, const struct filename *name,
2001 unsigned int flags, struct nameidata *nd)
2003 const char *s = path_init(dfd, name, flags, nd);
2008 while (!(err = link_path_walk(s, nd))
2009 && ((err = lookup_last(nd)) > 0)) {
2010 s = trailing_symlink(nd);
2017 err = complete_walk(nd);
2019 if (!err && nd->flags & LOOKUP_DIRECTORY)
2020 if (!d_can_lookup(nd->path.dentry))
2029 static int filename_lookup(int dfd, struct filename *name,
2030 unsigned int flags, struct nameidata *nd)
2033 struct nameidata *saved_nd = set_nameidata(nd);
2035 retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2036 if (unlikely(retval == -ECHILD))
2037 retval = path_lookupat(dfd, name, flags, nd);
2038 if (unlikely(retval == -ESTALE))
2039 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2041 if (likely(!retval))
2042 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2043 restore_nameidata(saved_nd);
2047 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2048 static int path_parentat(int dfd, const struct filename *name,
2049 unsigned int flags, struct nameidata *nd)
2051 const char *s = path_init(dfd, name, flags, nd);
2055 err = link_path_walk(s, nd);
2057 err = complete_walk(nd);
2064 static int filename_parentat(int dfd, struct filename *name,
2065 unsigned int flags, struct nameidata *nd)
2068 struct nameidata *saved_nd = set_nameidata(nd);
2070 retval = path_parentat(dfd, name, flags | LOOKUP_RCU, nd);
2071 if (unlikely(retval == -ECHILD))
2072 retval = path_parentat(dfd, name, flags, nd);
2073 if (unlikely(retval == -ESTALE))
2074 retval = path_parentat(dfd, name, flags | LOOKUP_REVAL, nd);
2076 if (likely(!retval))
2077 audit_inode(name, nd->path.dentry, LOOKUP_PARENT);
2078 restore_nameidata(saved_nd);
2082 /* does lookup, returns the object with parent locked */
2083 struct dentry *kern_path_locked(const char *name, struct path *path)
2085 struct filename *filename = getname_kernel(name);
2086 struct nameidata nd;
2090 if (IS_ERR(filename))
2091 return ERR_CAST(filename);
2093 err = filename_parentat(AT_FDCWD, filename, 0, &nd);
2098 if (nd.last_type != LAST_NORM) {
2100 d = ERR_PTR(-EINVAL);
2103 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2104 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2106 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2116 int kern_path(const char *name, unsigned int flags, struct path *path)
2118 struct nameidata nd;
2119 struct filename *filename = getname_kernel(name);
2120 int res = PTR_ERR(filename);
2122 if (!IS_ERR(filename)) {
2123 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2130 EXPORT_SYMBOL(kern_path);
2133 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2134 * @dentry: pointer to dentry of the base directory
2135 * @mnt: pointer to vfs mount of the base directory
2136 * @name: pointer to file name
2137 * @flags: lookup flags
2138 * @path: pointer to struct path to fill
2140 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2141 const char *name, unsigned int flags,
2144 struct filename *filename = getname_kernel(name);
2145 int err = PTR_ERR(filename);
2147 BUG_ON(flags & LOOKUP_PARENT);
2149 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2150 if (!IS_ERR(filename)) {
2151 struct nameidata nd;
2152 nd.root.dentry = dentry;
2154 err = filename_lookup(AT_FDCWD, filename,
2155 flags | LOOKUP_ROOT, &nd);
2162 EXPORT_SYMBOL(vfs_path_lookup);
2165 * lookup_one_len - filesystem helper to lookup single pathname component
2166 * @name: pathname component to lookup
2167 * @base: base directory to lookup from
2168 * @len: maximum length @len should be interpreted to
2170 * Note that this routine is purely a helper for filesystem usage and should
2171 * not be called by generic code.
2173 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2179 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2183 this.hash = full_name_hash(name, len);
2185 return ERR_PTR(-EACCES);
2187 if (unlikely(name[0] == '.')) {
2188 if (len < 2 || (len == 2 && name[1] == '.'))
2189 return ERR_PTR(-EACCES);
2193 c = *(const unsigned char *)name++;
2194 if (c == '/' || c == '\0')
2195 return ERR_PTR(-EACCES);
2198 * See if the low-level filesystem might want
2199 * to use its own hash..
2201 if (base->d_flags & DCACHE_OP_HASH) {
2202 int err = base->d_op->d_hash(base, &this);
2204 return ERR_PTR(err);
2207 err = inode_permission(base->d_inode, MAY_EXEC);
2209 return ERR_PTR(err);
2211 return __lookup_hash(&this, base, 0);
2213 EXPORT_SYMBOL(lookup_one_len);
2215 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2216 struct path *path, int *empty)
2218 struct nameidata nd;
2219 struct filename *tmp = getname_flags(name, flags, empty);
2220 int err = PTR_ERR(tmp);
2223 BUG_ON(flags & LOOKUP_PARENT);
2225 err = filename_lookup(dfd, tmp, flags, &nd);
2233 int user_path_at(int dfd, const char __user *name, unsigned flags,
2236 return user_path_at_empty(dfd, name, flags, path, NULL);
2238 EXPORT_SYMBOL(user_path_at);
2241 * NB: most callers don't do anything directly with the reference to the
2242 * to struct filename, but the nd->last pointer points into the name string
2243 * allocated by getname. So we must hold the reference to it until all
2244 * path-walking is complete.
2246 static struct filename *
2247 user_path_parent(int dfd, const char __user *path,
2248 struct path *parent,
2253 struct nameidata nd;
2254 struct filename *s = getname(path);
2257 /* only LOOKUP_REVAL is allowed in extra flags */
2258 flags &= LOOKUP_REVAL;
2263 error = filename_parentat(dfd, s, flags, &nd);
2266 return ERR_PTR(error);
2270 *type = nd.last_type;
2276 * mountpoint_last - look up last component for umount
2277 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2278 * @path: pointer to container for result
2280 * This is a special lookup_last function just for umount. In this case, we
2281 * need to resolve the path without doing any revalidation.
2283 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2284 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2285 * in almost all cases, this lookup will be served out of the dcache. The only
2286 * cases where it won't are if nd->last refers to a symlink or the path is
2287 * bogus and it doesn't exist.
2290 * -error: if there was an error during lookup. This includes -ENOENT if the
2291 * lookup found a negative dentry. The nd->path reference will also be
2294 * 0: if we successfully resolved nd->path and found it to not to be a
2295 * symlink that needs to be followed. "path" will also be populated.
2296 * The nd->path reference will also be put.
2298 * 1: if we successfully resolved nd->last and found it to be a symlink
2299 * that needs to be followed. "path" will be populated with the path
2300 * to the link, and nd->path will *not* be put.
2303 mountpoint_last(struct nameidata *nd, struct path *path)
2306 struct dentry *dentry;
2307 struct dentry *dir = nd->path.dentry;
2309 /* If we're in rcuwalk, drop out of it to handle last component */
2310 if (nd->flags & LOOKUP_RCU) {
2311 if (unlazy_walk(nd, NULL, 0))
2315 nd->flags &= ~LOOKUP_PARENT;
2317 if (unlikely(nd->last_type != LAST_NORM)) {
2318 error = handle_dots(nd, nd->last_type);
2321 dentry = dget(nd->path.dentry);
2325 mutex_lock(&dir->d_inode->i_mutex);
2326 dentry = d_lookup(dir, &nd->last);
2329 * No cached dentry. Mounted dentries are pinned in the cache,
2330 * so that means that this dentry is probably a symlink or the
2331 * path doesn't actually point to a mounted dentry.
2333 dentry = d_alloc(dir, &nd->last);
2335 mutex_unlock(&dir->d_inode->i_mutex);
2338 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2339 if (IS_ERR(dentry)) {
2340 mutex_unlock(&dir->d_inode->i_mutex);
2341 return PTR_ERR(dentry);
2344 mutex_unlock(&dir->d_inode->i_mutex);
2347 if (d_is_negative(dentry)) {
2353 path->dentry = dentry;
2354 path->mnt = nd->path.mnt;
2355 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2356 d_backing_inode(dentry), 0);
2357 if (unlikely(error))
2365 * path_mountpoint - look up a path to be umounted
2366 * @dfd: directory file descriptor to start walk from
2367 * @name: full pathname to walk
2368 * @path: pointer to container for result
2369 * @flags: lookup flags
2371 * Look up the given name, but don't attempt to revalidate the last component.
2372 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2375 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2376 struct nameidata *nd, unsigned int flags)
2378 const char *s = path_init(dfd, name, flags, nd);
2382 while (!(err = link_path_walk(s, nd)) &&
2383 (err = mountpoint_last(nd, path)) > 0) {
2384 s = trailing_symlink(nd);
2396 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2399 struct nameidata nd, *saved;
2402 return PTR_ERR(name);
2403 saved = set_nameidata(&nd);
2404 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2405 if (unlikely(error == -ECHILD))
2406 error = path_mountpoint(dfd, name, path, &nd, flags);
2407 if (unlikely(error == -ESTALE))
2408 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2410 audit_inode(name, path->dentry, 0);
2411 restore_nameidata(saved);
2417 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2418 * @dfd: directory file descriptor
2419 * @name: pathname from userland
2420 * @flags: lookup flags
2421 * @path: pointer to container to hold result
2423 * A umount is a special case for path walking. We're not actually interested
2424 * in the inode in this situation, and ESTALE errors can be a problem. We
2425 * simply want track down the dentry and vfsmount attached at the mountpoint
2426 * and avoid revalidating the last component.
2428 * Returns 0 and populates "path" on success.
2431 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2434 return filename_mountpoint(dfd, getname(name), path, flags);
2438 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2441 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2443 EXPORT_SYMBOL(kern_path_mountpoint);
2445 int __check_sticky(struct inode *dir, struct inode *inode)
2447 kuid_t fsuid = current_fsuid();
2449 if (uid_eq(inode->i_uid, fsuid))
2451 if (uid_eq(dir->i_uid, fsuid))
2453 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2455 EXPORT_SYMBOL(__check_sticky);
2458 * Check whether we can remove a link victim from directory dir, check
2459 * whether the type of victim is right.
2460 * 1. We can't do it if dir is read-only (done in permission())
2461 * 2. We should have write and exec permissions on dir
2462 * 3. We can't remove anything from append-only dir
2463 * 4. We can't do anything with immutable dir (done in permission())
2464 * 5. If the sticky bit on dir is set we should either
2465 * a. be owner of dir, or
2466 * b. be owner of victim, or
2467 * c. have CAP_FOWNER capability
2468 * 6. If the victim is append-only or immutable we can't do antyhing with
2469 * links pointing to it.
2470 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2471 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2472 * 9. We can't remove a root or mountpoint.
2473 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2474 * nfs_async_unlink().
2476 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2478 struct inode *inode = d_backing_inode(victim);
2481 if (d_is_negative(victim))
2485 BUG_ON(victim->d_parent->d_inode != dir);
2486 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2488 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2494 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2495 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2498 if (!d_is_dir(victim))
2500 if (IS_ROOT(victim))
2502 } else if (d_is_dir(victim))
2504 if (IS_DEADDIR(dir))
2506 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2511 /* Check whether we can create an object with dentry child in directory
2513 * 1. We can't do it if child already exists (open has special treatment for
2514 * this case, but since we are inlined it's OK)
2515 * 2. We can't do it if dir is read-only (done in permission())
2516 * 3. We should have write and exec permissions on dir
2517 * 4. We can't do it if dir is immutable (done in permission())
2519 static inline int may_create(struct inode *dir, struct dentry *child)
2521 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2524 if (IS_DEADDIR(dir))
2526 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2530 * p1 and p2 should be directories on the same fs.
2532 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2537 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2541 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2543 p = d_ancestor(p2, p1);
2545 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2546 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2550 p = d_ancestor(p1, p2);
2552 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2553 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2557 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2558 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2561 EXPORT_SYMBOL(lock_rename);
2563 void unlock_rename(struct dentry *p1, struct dentry *p2)
2565 mutex_unlock(&p1->d_inode->i_mutex);
2567 mutex_unlock(&p2->d_inode->i_mutex);
2568 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2571 EXPORT_SYMBOL(unlock_rename);
2573 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2576 int error = may_create(dir, dentry);
2580 if (!dir->i_op->create)
2581 return -EACCES; /* shouldn't it be ENOSYS? */
2584 error = security_inode_create(dir, dentry, mode);
2587 error = dir->i_op->create(dir, dentry, mode, want_excl);
2589 fsnotify_create(dir, dentry);
2592 EXPORT_SYMBOL(vfs_create);
2594 static int may_open(struct path *path, int acc_mode, int flag)
2596 struct dentry *dentry = path->dentry;
2597 struct inode *inode = dentry->d_inode;
2607 switch (inode->i_mode & S_IFMT) {
2611 if (acc_mode & MAY_WRITE)
2616 if (path->mnt->mnt_flags & MNT_NODEV)
2625 error = inode_permission(inode, acc_mode);
2630 * An append-only file must be opened in append mode for writing.
2632 if (IS_APPEND(inode)) {
2633 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2639 /* O_NOATIME can only be set by the owner or superuser */
2640 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2646 static int handle_truncate(struct file *filp)
2648 struct path *path = &filp->f_path;
2649 struct inode *inode = path->dentry->d_inode;
2650 int error = get_write_access(inode);
2654 * Refuse to truncate files with mandatory locks held on them.
2656 error = locks_verify_locked(filp);
2658 error = security_path_truncate(path);
2660 error = do_truncate(path->dentry, 0,
2661 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2664 put_write_access(inode);
2668 static inline int open_to_namei_flags(int flag)
2670 if ((flag & O_ACCMODE) == 3)
2675 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2677 int error = security_path_mknod(dir, dentry, mode, 0);
2681 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2685 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2689 * Attempt to atomically look up, create and open a file from a negative
2692 * Returns 0 if successful. The file will have been created and attached to
2693 * @file by the filesystem calling finish_open().
2695 * Returns 1 if the file was looked up only or didn't need creating. The
2696 * caller will need to perform the open themselves. @path will have been
2697 * updated to point to the new dentry. This may be negative.
2699 * Returns an error code otherwise.
2701 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2702 struct path *path, struct file *file,
2703 const struct open_flags *op,
2704 bool got_write, bool need_lookup,
2707 struct inode *dir = nd->path.dentry->d_inode;
2708 unsigned open_flag = open_to_namei_flags(op->open_flag);
2712 int create_error = 0;
2713 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2716 BUG_ON(dentry->d_inode);
2718 /* Don't create child dentry for a dead directory. */
2719 if (unlikely(IS_DEADDIR(dir))) {
2725 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2726 mode &= ~current_umask();
2728 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2730 open_flag &= ~O_TRUNC;
2733 * Checking write permission is tricky, bacuse we don't know if we are
2734 * going to actually need it: O_CREAT opens should work as long as the
2735 * file exists. But checking existence breaks atomicity. The trick is
2736 * to check access and if not granted clear O_CREAT from the flags.
2738 * Another problem is returing the "right" error value (e.g. for an
2739 * O_EXCL open we want to return EEXIST not EROFS).
2741 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2742 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2743 if (!(open_flag & O_CREAT)) {
2745 * No O_CREATE -> atomicity not a requirement -> fall
2746 * back to lookup + open
2749 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2750 /* Fall back and fail with the right error */
2751 create_error = -EROFS;
2754 /* No side effects, safe to clear O_CREAT */
2755 create_error = -EROFS;
2756 open_flag &= ~O_CREAT;
2760 if (open_flag & O_CREAT) {
2761 error = may_o_create(&nd->path, dentry, mode);
2763 create_error = error;
2764 if (open_flag & O_EXCL)
2766 open_flag &= ~O_CREAT;
2770 if (nd->flags & LOOKUP_DIRECTORY)
2771 open_flag |= O_DIRECTORY;
2773 file->f_path.dentry = DENTRY_NOT_SET;
2774 file->f_path.mnt = nd->path.mnt;
2775 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2778 if (create_error && error == -ENOENT)
2779 error = create_error;
2783 if (error) { /* returned 1, that is */
2784 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2788 if (file->f_path.dentry) {
2790 dentry = file->f_path.dentry;
2792 if (*opened & FILE_CREATED)
2793 fsnotify_create(dir, dentry);
2794 if (!dentry->d_inode) {
2795 WARN_ON(*opened & FILE_CREATED);
2797 error = create_error;
2801 if (excl && !(*opened & FILE_CREATED)) {
2810 * We didn't have the inode before the open, so check open permission
2813 acc_mode = op->acc_mode;
2814 if (*opened & FILE_CREATED) {
2815 WARN_ON(!(open_flag & O_CREAT));
2816 fsnotify_create(dir, dentry);
2817 acc_mode = MAY_OPEN;
2819 error = may_open(&file->f_path, acc_mode, open_flag);
2829 dentry = lookup_real(dir, dentry, nd->flags);
2831 return PTR_ERR(dentry);
2834 int open_flag = op->open_flag;
2836 error = create_error;
2837 if ((open_flag & O_EXCL)) {
2838 if (!dentry->d_inode)
2840 } else if (!dentry->d_inode) {
2842 } else if ((open_flag & O_TRUNC) &&
2846 /* will fail later, go on to get the right error */
2850 path->dentry = dentry;
2851 path->mnt = nd->path.mnt;
2856 * Look up and maybe create and open the last component.
2858 * Must be called with i_mutex held on parent.
2860 * Returns 0 if the file was successfully atomically created (if necessary) and
2861 * opened. In this case the file will be returned attached to @file.
2863 * Returns 1 if the file was not completely opened at this time, though lookups
2864 * and creations will have been performed and the dentry returned in @path will
2865 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2866 * specified then a negative dentry may be returned.
2868 * An error code is returned otherwise.
2870 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2871 * cleared otherwise prior to returning.
2873 static int lookup_open(struct nameidata *nd, struct path *path,
2875 const struct open_flags *op,
2876 bool got_write, int *opened)
2878 struct dentry *dir = nd->path.dentry;
2879 struct inode *dir_inode = dir->d_inode;
2880 struct dentry *dentry;
2884 *opened &= ~FILE_CREATED;
2885 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2887 return PTR_ERR(dentry);
2889 /* Cached positive dentry: will open in f_op->open */
2890 if (!need_lookup && dentry->d_inode)
2893 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2894 return atomic_open(nd, dentry, path, file, op, got_write,
2895 need_lookup, opened);
2899 BUG_ON(dentry->d_inode);
2901 dentry = lookup_real(dir_inode, dentry, nd->flags);
2903 return PTR_ERR(dentry);
2906 /* Negative dentry, just create the file */
2907 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2908 umode_t mode = op->mode;
2909 if (!IS_POSIXACL(dir->d_inode))
2910 mode &= ~current_umask();
2912 * This write is needed to ensure that a
2913 * rw->ro transition does not occur between
2914 * the time when the file is created and when
2915 * a permanent write count is taken through
2916 * the 'struct file' in finish_open().
2922 *opened |= FILE_CREATED;
2923 error = security_path_mknod(&nd->path, dentry, mode, 0);
2926 error = vfs_create(dir->d_inode, dentry, mode,
2927 nd->flags & LOOKUP_EXCL);
2932 path->dentry = dentry;
2933 path->mnt = nd->path.mnt;
2942 * Handle the last step of open()
2944 static int do_last(struct nameidata *nd,
2945 struct file *file, const struct open_flags *op,
2946 int *opened, struct filename *name)
2948 struct dentry *dir = nd->path.dentry;
2949 int open_flag = op->open_flag;
2950 bool will_truncate = (open_flag & O_TRUNC) != 0;
2951 bool got_write = false;
2952 int acc_mode = op->acc_mode;
2954 struct inode *inode;
2955 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2957 bool retried = false;
2960 nd->flags &= ~LOOKUP_PARENT;
2961 nd->flags |= op->intent;
2963 if (nd->last_type != LAST_NORM) {
2964 error = handle_dots(nd, nd->last_type);
2965 if (unlikely(error))
2970 if (!(open_flag & O_CREAT)) {
2971 if (nd->last.name[nd->last.len])
2972 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2973 /* we _can_ be in RCU mode here */
2974 error = lookup_fast(nd, &path, &inode, &seq);
2981 BUG_ON(nd->inode != dir->d_inode);
2983 /* create side of things */
2985 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2986 * has been cleared when we got to the last component we are
2989 error = complete_walk(nd);
2993 audit_inode(name, dir, LOOKUP_PARENT);
2994 /* trailing slashes? */
2995 if (unlikely(nd->last.name[nd->last.len]))
3000 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3001 error = mnt_want_write(nd->path.mnt);
3005 * do _not_ fail yet - we might not need that or fail with
3006 * a different error; let lookup_open() decide; we'll be
3007 * dropping this one anyway.
3010 mutex_lock(&dir->d_inode->i_mutex);
3011 error = lookup_open(nd, &path, file, op, got_write, opened);
3012 mutex_unlock(&dir->d_inode->i_mutex);
3018 if ((*opened & FILE_CREATED) ||
3019 !S_ISREG(file_inode(file)->i_mode))
3020 will_truncate = false;
3022 audit_inode(name, file->f_path.dentry, 0);
3026 if (*opened & FILE_CREATED) {
3027 /* Don't check for write permission, don't truncate */
3028 open_flag &= ~O_TRUNC;
3029 will_truncate = false;
3030 acc_mode = MAY_OPEN;
3031 path_to_nameidata(&path, nd);
3032 goto finish_open_created;
3036 * create/update audit record if it already exists.
3038 if (d_is_positive(path.dentry))
3039 audit_inode(name, path.dentry, 0);
3042 * If atomic_open() acquired write access it is dropped now due to
3043 * possible mount and symlink following (this might be optimized away if
3047 mnt_drop_write(nd->path.mnt);
3051 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3052 path_to_nameidata(&path, nd);
3056 error = follow_managed(&path, nd);
3057 if (unlikely(error < 0))
3060 BUG_ON(nd->flags & LOOKUP_RCU);
3061 inode = d_backing_inode(path.dentry);
3062 seq = 0; /* out of RCU mode, so the value doesn't matter */
3063 if (unlikely(d_is_negative(path.dentry))) {
3064 path_to_nameidata(&path, nd);
3070 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3072 if (unlikely(error))
3075 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3076 path_to_nameidata(&path, nd);
3080 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3081 path_to_nameidata(&path, nd);
3083 save_parent.dentry = nd->path.dentry;
3084 save_parent.mnt = mntget(path.mnt);
3085 nd->path.dentry = path.dentry;
3090 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3092 error = complete_walk(nd);
3094 path_put(&save_parent);
3097 audit_inode(name, nd->path.dentry, 0);
3099 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3102 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3104 if (!d_is_reg(nd->path.dentry))
3105 will_truncate = false;
3107 if (will_truncate) {
3108 error = mnt_want_write(nd->path.mnt);
3113 finish_open_created:
3114 error = may_open(&nd->path, acc_mode, open_flag);
3118 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3119 error = vfs_open(&nd->path, file, current_cred());
3121 *opened |= FILE_OPENED;
3123 if (error == -EOPENSTALE)
3128 error = open_check_o_direct(file);
3131 error = ima_file_check(file, op->acc_mode, *opened);
3135 if (will_truncate) {
3136 error = handle_truncate(file);
3142 mnt_drop_write(nd->path.mnt);
3143 path_put(&save_parent);
3151 /* If no saved parent or already retried then can't retry */
3152 if (!save_parent.dentry || retried)
3155 BUG_ON(save_parent.dentry != dir);
3156 path_put(&nd->path);
3157 nd->path = save_parent;
3158 nd->inode = dir->d_inode;
3159 save_parent.mnt = NULL;
3160 save_parent.dentry = NULL;
3162 mnt_drop_write(nd->path.mnt);
3169 static int do_tmpfile(int dfd, struct filename *pathname,
3170 struct nameidata *nd, int flags,
3171 const struct open_flags *op,
3172 struct file *file, int *opened)
3174 static const struct qstr name = QSTR_INIT("/", 1);
3175 struct dentry *dentry, *child;
3177 int error = path_lookupat(dfd, pathname,
3178 flags | LOOKUP_DIRECTORY, nd);
3179 if (unlikely(error))
3181 error = mnt_want_write(nd->path.mnt);
3182 if (unlikely(error))
3184 /* we want directory to be writable */
3185 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3188 dentry = nd->path.dentry;
3189 dir = dentry->d_inode;
3190 if (!dir->i_op->tmpfile) {
3191 error = -EOPNOTSUPP;
3194 child = d_alloc(dentry, &name);
3195 if (unlikely(!child)) {
3199 nd->flags &= ~LOOKUP_DIRECTORY;
3200 nd->flags |= op->intent;
3201 dput(nd->path.dentry);
3202 nd->path.dentry = child;
3203 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3206 audit_inode(pathname, nd->path.dentry, 0);
3207 /* Don't check for other permissions, the inode was just created */
3208 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3211 file->f_path.mnt = nd->path.mnt;
3212 error = finish_open(file, nd->path.dentry, NULL, opened);
3215 error = open_check_o_direct(file);
3218 } else if (!(op->open_flag & O_EXCL)) {
3219 struct inode *inode = file_inode(file);
3220 spin_lock(&inode->i_lock);
3221 inode->i_state |= I_LINKABLE;
3222 spin_unlock(&inode->i_lock);
3225 mnt_drop_write(nd->path.mnt);
3227 path_put(&nd->path);
3231 static struct file *path_openat(int dfd, struct filename *pathname,
3232 struct nameidata *nd, const struct open_flags *op, int flags)
3239 file = get_empty_filp();
3243 file->f_flags = op->open_flag;
3245 if (unlikely(file->f_flags & __O_TMPFILE)) {
3246 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3250 s = path_init(dfd, pathname, flags, nd);
3255 while (!(error = link_path_walk(s, nd)) &&
3256 (error = do_last(nd, file, op, &opened, pathname)) > 0) {
3257 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3258 s = trailing_symlink(nd);
3267 if (!(opened & FILE_OPENED)) {
3271 if (unlikely(error)) {
3272 if (error == -EOPENSTALE) {
3273 if (flags & LOOKUP_RCU)
3278 file = ERR_PTR(error);
3283 struct file *do_filp_open(int dfd, struct filename *pathname,
3284 const struct open_flags *op)
3286 struct nameidata nd, *saved_nd = set_nameidata(&nd);
3287 int flags = op->lookup_flags;
3290 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3291 if (unlikely(filp == ERR_PTR(-ECHILD)))
3292 filp = path_openat(dfd, pathname, &nd, op, flags);
3293 if (unlikely(filp == ERR_PTR(-ESTALE)))
3294 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3295 restore_nameidata(saved_nd);
3299 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3300 const char *name, const struct open_flags *op)
3302 struct nameidata nd, *saved_nd;
3304 struct filename *filename;
3305 int flags = op->lookup_flags | LOOKUP_ROOT;
3308 nd.root.dentry = dentry;
3310 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3311 return ERR_PTR(-ELOOP);
3313 filename = getname_kernel(name);
3314 if (unlikely(IS_ERR(filename)))
3315 return ERR_CAST(filename);
3317 saved_nd = set_nameidata(&nd);
3318 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3319 if (unlikely(file == ERR_PTR(-ECHILD)))
3320 file = path_openat(-1, filename, &nd, op, flags);
3321 if (unlikely(file == ERR_PTR(-ESTALE)))
3322 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3323 restore_nameidata(saved_nd);
3328 static struct dentry *filename_create(int dfd, struct filename *name,
3329 struct path *path, unsigned int lookup_flags)
3331 struct dentry *dentry = ERR_PTR(-EEXIST);
3332 struct nameidata nd;
3335 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3338 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3339 * other flags passed in are ignored!
3341 lookup_flags &= LOOKUP_REVAL;
3343 error = filename_parentat(dfd, name, lookup_flags, &nd);
3345 return ERR_PTR(error);
3348 * Yucky last component or no last component at all?
3349 * (foo/., foo/.., /////)
3351 if (nd.last_type != LAST_NORM)
3353 nd.flags &= ~LOOKUP_PARENT;
3354 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3356 /* don't fail immediately if it's r/o, at least try to report other errors */
3357 err2 = mnt_want_write(nd.path.mnt);
3359 * Do the final lookup.
3361 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3362 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3367 if (d_is_positive(dentry))
3371 * Special case - lookup gave negative, but... we had foo/bar/
3372 * From the vfs_mknod() POV we just have a negative dentry -
3373 * all is fine. Let's be bastards - you had / on the end, you've
3374 * been asking for (non-existent) directory. -ENOENT for you.
3376 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3380 if (unlikely(err2)) {
3388 dentry = ERR_PTR(error);
3390 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3392 mnt_drop_write(nd.path.mnt);
3398 struct dentry *kern_path_create(int dfd, const char *pathname,
3399 struct path *path, unsigned int lookup_flags)
3401 struct filename *filename = getname_kernel(pathname);
3404 if (IS_ERR(filename))
3405 return ERR_CAST(filename);
3406 res = filename_create(dfd, filename, path, lookup_flags);
3410 EXPORT_SYMBOL(kern_path_create);
3412 void done_path_create(struct path *path, struct dentry *dentry)
3415 mutex_unlock(&path->dentry->d_inode->i_mutex);
3416 mnt_drop_write(path->mnt);
3419 EXPORT_SYMBOL(done_path_create);
3421 struct dentry *user_path_create(int dfd, const char __user *pathname,
3422 struct path *path, unsigned int lookup_flags)
3424 struct filename *tmp = getname(pathname);
3427 return ERR_CAST(tmp);
3428 res = filename_create(dfd, tmp, path, lookup_flags);
3432 EXPORT_SYMBOL(user_path_create);
3434 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3436 int error = may_create(dir, dentry);
3441 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3444 if (!dir->i_op->mknod)
3447 error = devcgroup_inode_mknod(mode, dev);
3451 error = security_inode_mknod(dir, dentry, mode, dev);
3455 error = dir->i_op->mknod(dir, dentry, mode, dev);
3457 fsnotify_create(dir, dentry);
3460 EXPORT_SYMBOL(vfs_mknod);
3462 static int may_mknod(umode_t mode)
3464 switch (mode & S_IFMT) {
3470 case 0: /* zero mode translates to S_IFREG */
3479 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3482 struct dentry *dentry;
3485 unsigned int lookup_flags = 0;
3487 error = may_mknod(mode);
3491 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3493 return PTR_ERR(dentry);
3495 if (!IS_POSIXACL(path.dentry->d_inode))
3496 mode &= ~current_umask();
3497 error = security_path_mknod(&path, dentry, mode, dev);
3500 switch (mode & S_IFMT) {
3501 case 0: case S_IFREG:
3502 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3504 case S_IFCHR: case S_IFBLK:
3505 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3506 new_decode_dev(dev));
3508 case S_IFIFO: case S_IFSOCK:
3509 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3513 done_path_create(&path, dentry);
3514 if (retry_estale(error, lookup_flags)) {
3515 lookup_flags |= LOOKUP_REVAL;
3521 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3523 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3526 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3528 int error = may_create(dir, dentry);
3529 unsigned max_links = dir->i_sb->s_max_links;
3534 if (!dir->i_op->mkdir)
3537 mode &= (S_IRWXUGO|S_ISVTX);
3538 error = security_inode_mkdir(dir, dentry, mode);
3542 if (max_links && dir->i_nlink >= max_links)
3545 error = dir->i_op->mkdir(dir, dentry, mode);
3547 fsnotify_mkdir(dir, dentry);
3550 EXPORT_SYMBOL(vfs_mkdir);
3552 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3554 struct dentry *dentry;
3557 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3560 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3562 return PTR_ERR(dentry);
3564 if (!IS_POSIXACL(path.dentry->d_inode))
3565 mode &= ~current_umask();
3566 error = security_path_mkdir(&path, dentry, mode);
3568 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3569 done_path_create(&path, dentry);
3570 if (retry_estale(error, lookup_flags)) {
3571 lookup_flags |= LOOKUP_REVAL;
3577 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3579 return sys_mkdirat(AT_FDCWD, pathname, mode);
3583 * The dentry_unhash() helper will try to drop the dentry early: we
3584 * should have a usage count of 1 if we're the only user of this
3585 * dentry, and if that is true (possibly after pruning the dcache),
3586 * then we drop the dentry now.
3588 * A low-level filesystem can, if it choses, legally
3591 * if (!d_unhashed(dentry))
3594 * if it cannot handle the case of removing a directory
3595 * that is still in use by something else..
3597 void dentry_unhash(struct dentry *dentry)
3599 shrink_dcache_parent(dentry);
3600 spin_lock(&dentry->d_lock);
3601 if (dentry->d_lockref.count == 1)
3603 spin_unlock(&dentry->d_lock);
3605 EXPORT_SYMBOL(dentry_unhash);
3607 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3609 int error = may_delete(dir, dentry, 1);
3614 if (!dir->i_op->rmdir)
3618 mutex_lock(&dentry->d_inode->i_mutex);
3621 if (is_local_mountpoint(dentry))
3624 error = security_inode_rmdir(dir, dentry);
3628 shrink_dcache_parent(dentry);
3629 error = dir->i_op->rmdir(dir, dentry);
3633 dentry->d_inode->i_flags |= S_DEAD;
3635 detach_mounts(dentry);
3638 mutex_unlock(&dentry->d_inode->i_mutex);
3644 EXPORT_SYMBOL(vfs_rmdir);
3646 static long do_rmdir(int dfd, const char __user *pathname)
3649 struct filename *name;
3650 struct dentry *dentry;
3654 unsigned int lookup_flags = 0;
3656 name = user_path_parent(dfd, pathname,
3657 &path, &last, &type, lookup_flags);
3659 return PTR_ERR(name);
3673 error = mnt_want_write(path.mnt);
3677 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3678 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3679 error = PTR_ERR(dentry);
3682 if (!dentry->d_inode) {
3686 error = security_path_rmdir(&path, dentry);
3689 error = vfs_rmdir(path.dentry->d_inode, dentry);
3693 mutex_unlock(&path.dentry->d_inode->i_mutex);
3694 mnt_drop_write(path.mnt);
3698 if (retry_estale(error, lookup_flags)) {
3699 lookup_flags |= LOOKUP_REVAL;
3705 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3707 return do_rmdir(AT_FDCWD, pathname);
3711 * vfs_unlink - unlink a filesystem object
3712 * @dir: parent directory
3714 * @delegated_inode: returns victim inode, if the inode is delegated.
3716 * The caller must hold dir->i_mutex.
3718 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3719 * return a reference to the inode in delegated_inode. The caller
3720 * should then break the delegation on that inode and retry. Because
3721 * breaking a delegation may take a long time, the caller should drop
3722 * dir->i_mutex before doing so.
3724 * Alternatively, a caller may pass NULL for delegated_inode. This may
3725 * be appropriate for callers that expect the underlying filesystem not
3726 * to be NFS exported.
3728 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3730 struct inode *target = dentry->d_inode;
3731 int error = may_delete(dir, dentry, 0);
3736 if (!dir->i_op->unlink)
3739 mutex_lock(&target->i_mutex);
3740 if (is_local_mountpoint(dentry))
3743 error = security_inode_unlink(dir, dentry);
3745 error = try_break_deleg(target, delegated_inode);
3748 error = dir->i_op->unlink(dir, dentry);
3751 detach_mounts(dentry);
3756 mutex_unlock(&target->i_mutex);
3758 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3759 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3760 fsnotify_link_count(target);
3766 EXPORT_SYMBOL(vfs_unlink);
3769 * Make sure that the actual truncation of the file will occur outside its
3770 * directory's i_mutex. Truncate can take a long time if there is a lot of
3771 * writeout happening, and we don't want to prevent access to the directory
3772 * while waiting on the I/O.
3774 static long do_unlinkat(int dfd, const char __user *pathname)
3777 struct filename *name;
3778 struct dentry *dentry;
3782 struct inode *inode = NULL;
3783 struct inode *delegated_inode = NULL;
3784 unsigned int lookup_flags = 0;
3786 name = user_path_parent(dfd, pathname,
3787 &path, &last, &type, lookup_flags);
3789 return PTR_ERR(name);
3792 if (type != LAST_NORM)
3795 error = mnt_want_write(path.mnt);
3799 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3800 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3801 error = PTR_ERR(dentry);
3802 if (!IS_ERR(dentry)) {
3803 /* Why not before? Because we want correct error value */
3804 if (last.name[last.len])
3806 inode = dentry->d_inode;
3807 if (d_is_negative(dentry))
3810 error = security_path_unlink(&path, dentry);
3813 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3817 mutex_unlock(&path.dentry->d_inode->i_mutex);
3819 iput(inode); /* truncate the inode here */
3821 if (delegated_inode) {
3822 error = break_deleg_wait(&delegated_inode);
3826 mnt_drop_write(path.mnt);
3830 if (retry_estale(error, lookup_flags)) {
3831 lookup_flags |= LOOKUP_REVAL;
3838 if (d_is_negative(dentry))
3840 else if (d_is_dir(dentry))
3847 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3849 if ((flag & ~AT_REMOVEDIR) != 0)
3852 if (flag & AT_REMOVEDIR)
3853 return do_rmdir(dfd, pathname);
3855 return do_unlinkat(dfd, pathname);
3858 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3860 return do_unlinkat(AT_FDCWD, pathname);
3863 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3865 int error = may_create(dir, dentry);
3870 if (!dir->i_op->symlink)
3873 error = security_inode_symlink(dir, dentry, oldname);
3877 error = dir->i_op->symlink(dir, dentry, oldname);
3879 fsnotify_create(dir, dentry);
3882 EXPORT_SYMBOL(vfs_symlink);
3884 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3885 int, newdfd, const char __user *, newname)
3888 struct filename *from;
3889 struct dentry *dentry;
3891 unsigned int lookup_flags = 0;
3893 from = getname(oldname);
3895 return PTR_ERR(from);
3897 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3898 error = PTR_ERR(dentry);
3902 error = security_path_symlink(&path, dentry, from->name);
3904 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3905 done_path_create(&path, dentry);
3906 if (retry_estale(error, lookup_flags)) {
3907 lookup_flags |= LOOKUP_REVAL;
3915 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3917 return sys_symlinkat(oldname, AT_FDCWD, newname);
3921 * vfs_link - create a new link
3922 * @old_dentry: object to be linked
3924 * @new_dentry: where to create the new link
3925 * @delegated_inode: returns inode needing a delegation break
3927 * The caller must hold dir->i_mutex
3929 * If vfs_link discovers a delegation on the to-be-linked file in need
3930 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3931 * inode in delegated_inode. The caller should then break the delegation
3932 * and retry. Because breaking a delegation may take a long time, the
3933 * caller should drop the i_mutex before doing so.
3935 * Alternatively, a caller may pass NULL for delegated_inode. This may
3936 * be appropriate for callers that expect the underlying filesystem not
3937 * to be NFS exported.
3939 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3941 struct inode *inode = old_dentry->d_inode;
3942 unsigned max_links = dir->i_sb->s_max_links;
3948 error = may_create(dir, new_dentry);
3952 if (dir->i_sb != inode->i_sb)
3956 * A link to an append-only or immutable file cannot be created.
3958 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3960 if (!dir->i_op->link)
3962 if (S_ISDIR(inode->i_mode))
3965 error = security_inode_link(old_dentry, dir, new_dentry);
3969 mutex_lock(&inode->i_mutex);
3970 /* Make sure we don't allow creating hardlink to an unlinked file */
3971 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3973 else if (max_links && inode->i_nlink >= max_links)
3976 error = try_break_deleg(inode, delegated_inode);
3978 error = dir->i_op->link(old_dentry, dir, new_dentry);
3981 if (!error && (inode->i_state & I_LINKABLE)) {
3982 spin_lock(&inode->i_lock);
3983 inode->i_state &= ~I_LINKABLE;
3984 spin_unlock(&inode->i_lock);
3986 mutex_unlock(&inode->i_mutex);
3988 fsnotify_link(dir, inode, new_dentry);
3991 EXPORT_SYMBOL(vfs_link);
3994 * Hardlinks are often used in delicate situations. We avoid
3995 * security-related surprises by not following symlinks on the
3998 * We don't follow them on the oldname either to be compatible
3999 * with linux 2.0, and to avoid hard-linking to directories
4000 * and other special files. --ADM
4002 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4003 int, newdfd, const char __user *, newname, int, flags)
4005 struct dentry *new_dentry;
4006 struct path old_path, new_path;
4007 struct inode *delegated_inode = NULL;
4011 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4014 * To use null names we require CAP_DAC_READ_SEARCH
4015 * This ensures that not everyone will be able to create
4016 * handlink using the passed filedescriptor.
4018 if (flags & AT_EMPTY_PATH) {
4019 if (!capable(CAP_DAC_READ_SEARCH))
4024 if (flags & AT_SYMLINK_FOLLOW)
4025 how |= LOOKUP_FOLLOW;
4027 error = user_path_at(olddfd, oldname, how, &old_path);
4031 new_dentry = user_path_create(newdfd, newname, &new_path,
4032 (how & LOOKUP_REVAL));
4033 error = PTR_ERR(new_dentry);
4034 if (IS_ERR(new_dentry))
4038 if (old_path.mnt != new_path.mnt)
4040 error = may_linkat(&old_path);
4041 if (unlikely(error))
4043 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4046 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4048 done_path_create(&new_path, new_dentry);
4049 if (delegated_inode) {
4050 error = break_deleg_wait(&delegated_inode);
4052 path_put(&old_path);
4056 if (retry_estale(error, how)) {
4057 path_put(&old_path);
4058 how |= LOOKUP_REVAL;
4062 path_put(&old_path);
4067 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4069 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4073 * vfs_rename - rename a filesystem object
4074 * @old_dir: parent of source
4075 * @old_dentry: source
4076 * @new_dir: parent of destination
4077 * @new_dentry: destination
4078 * @delegated_inode: returns an inode needing a delegation break
4079 * @flags: rename flags
4081 * The caller must hold multiple mutexes--see lock_rename()).
4083 * If vfs_rename discovers a delegation in need of breaking at either
4084 * the source or destination, it will return -EWOULDBLOCK and return a
4085 * reference to the inode in delegated_inode. The caller should then
4086 * break the delegation and retry. Because breaking a delegation may
4087 * take a long time, the caller should drop all locks before doing
4090 * Alternatively, a caller may pass NULL for delegated_inode. This may
4091 * be appropriate for callers that expect the underlying filesystem not
4092 * to be NFS exported.
4094 * The worst of all namespace operations - renaming directory. "Perverted"
4095 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4097 * a) we can get into loop creation.
4098 * b) race potential - two innocent renames can create a loop together.
4099 * That's where 4.4 screws up. Current fix: serialization on
4100 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4102 * c) we have to lock _four_ objects - parents and victim (if it exists),
4103 * and source (if it is not a directory).
4104 * And that - after we got ->i_mutex on parents (until then we don't know
4105 * whether the target exists). Solution: try to be smart with locking
4106 * order for inodes. We rely on the fact that tree topology may change
4107 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4108 * move will be locked. Thus we can rank directories by the tree
4109 * (ancestors first) and rank all non-directories after them.
4110 * That works since everybody except rename does "lock parent, lookup,
4111 * lock child" and rename is under ->s_vfs_rename_mutex.
4112 * HOWEVER, it relies on the assumption that any object with ->lookup()
4113 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4114 * we'd better make sure that there's no link(2) for them.
4115 * d) conversion from fhandle to dentry may come in the wrong moment - when
4116 * we are removing the target. Solution: we will have to grab ->i_mutex
4117 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4118 * ->i_mutex on parents, which works but leads to some truly excessive
4121 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4122 struct inode *new_dir, struct dentry *new_dentry,
4123 struct inode **delegated_inode, unsigned int flags)
4126 bool is_dir = d_is_dir(old_dentry);
4127 const unsigned char *old_name;
4128 struct inode *source = old_dentry->d_inode;
4129 struct inode *target = new_dentry->d_inode;
4130 bool new_is_dir = false;
4131 unsigned max_links = new_dir->i_sb->s_max_links;
4133 if (source == target)
4136 error = may_delete(old_dir, old_dentry, is_dir);
4141 error = may_create(new_dir, new_dentry);
4143 new_is_dir = d_is_dir(new_dentry);
4145 if (!(flags & RENAME_EXCHANGE))
4146 error = may_delete(new_dir, new_dentry, is_dir);
4148 error = may_delete(new_dir, new_dentry, new_is_dir);
4153 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4156 if (flags && !old_dir->i_op->rename2)
4160 * If we are going to change the parent - check write permissions,
4161 * we'll need to flip '..'.
4163 if (new_dir != old_dir) {
4165 error = inode_permission(source, MAY_WRITE);
4169 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4170 error = inode_permission(target, MAY_WRITE);
4176 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4181 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4183 if (!is_dir || (flags & RENAME_EXCHANGE))
4184 lock_two_nondirectories(source, target);
4186 mutex_lock(&target->i_mutex);
4189 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4192 if (max_links && new_dir != old_dir) {
4194 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4196 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4197 old_dir->i_nlink >= max_links)
4200 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4201 shrink_dcache_parent(new_dentry);
4203 error = try_break_deleg(source, delegated_inode);
4207 if (target && !new_is_dir) {
4208 error = try_break_deleg(target, delegated_inode);
4212 if (!old_dir->i_op->rename2) {
4213 error = old_dir->i_op->rename(old_dir, old_dentry,
4214 new_dir, new_dentry);
4216 WARN_ON(old_dir->i_op->rename != NULL);
4217 error = old_dir->i_op->rename2(old_dir, old_dentry,
4218 new_dir, new_dentry, flags);
4223 if (!(flags & RENAME_EXCHANGE) && target) {
4225 target->i_flags |= S_DEAD;
4226 dont_mount(new_dentry);
4227 detach_mounts(new_dentry);
4229 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4230 if (!(flags & RENAME_EXCHANGE))
4231 d_move(old_dentry, new_dentry);
4233 d_exchange(old_dentry, new_dentry);
4236 if (!is_dir || (flags & RENAME_EXCHANGE))
4237 unlock_two_nondirectories(source, target);
4239 mutex_unlock(&target->i_mutex);
4242 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4243 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4244 if (flags & RENAME_EXCHANGE) {
4245 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4246 new_is_dir, NULL, new_dentry);
4249 fsnotify_oldname_free(old_name);
4253 EXPORT_SYMBOL(vfs_rename);
4255 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4256 int, newdfd, const char __user *, newname, unsigned int, flags)
4258 struct dentry *old_dentry, *new_dentry;
4259 struct dentry *trap;
4260 struct path old_path, new_path;
4261 struct qstr old_last, new_last;
4262 int old_type, new_type;
4263 struct inode *delegated_inode = NULL;
4264 struct filename *from;
4265 struct filename *to;
4266 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4267 bool should_retry = false;
4270 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4273 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4274 (flags & RENAME_EXCHANGE))
4277 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4280 if (flags & RENAME_EXCHANGE)
4284 from = user_path_parent(olddfd, oldname,
4285 &old_path, &old_last, &old_type, lookup_flags);
4287 error = PTR_ERR(from);
4291 to = user_path_parent(newdfd, newname,
4292 &new_path, &new_last, &new_type, lookup_flags);
4294 error = PTR_ERR(to);
4299 if (old_path.mnt != new_path.mnt)
4303 if (old_type != LAST_NORM)
4306 if (flags & RENAME_NOREPLACE)
4308 if (new_type != LAST_NORM)
4311 error = mnt_want_write(old_path.mnt);
4316 trap = lock_rename(new_path.dentry, old_path.dentry);
4318 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4319 error = PTR_ERR(old_dentry);
4320 if (IS_ERR(old_dentry))
4322 /* source must exist */
4324 if (d_is_negative(old_dentry))
4326 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4327 error = PTR_ERR(new_dentry);
4328 if (IS_ERR(new_dentry))
4331 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4333 if (flags & RENAME_EXCHANGE) {
4335 if (d_is_negative(new_dentry))
4338 if (!d_is_dir(new_dentry)) {
4340 if (new_last.name[new_last.len])
4344 /* unless the source is a directory trailing slashes give -ENOTDIR */
4345 if (!d_is_dir(old_dentry)) {
4347 if (old_last.name[old_last.len])
4349 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4352 /* source should not be ancestor of target */
4354 if (old_dentry == trap)
4356 /* target should not be an ancestor of source */
4357 if (!(flags & RENAME_EXCHANGE))
4359 if (new_dentry == trap)
4362 error = security_path_rename(&old_path, old_dentry,
4363 &new_path, new_dentry, flags);
4366 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4367 new_path.dentry->d_inode, new_dentry,
4368 &delegated_inode, flags);
4374 unlock_rename(new_path.dentry, old_path.dentry);
4375 if (delegated_inode) {
4376 error = break_deleg_wait(&delegated_inode);
4380 mnt_drop_write(old_path.mnt);
4382 if (retry_estale(error, lookup_flags))
4383 should_retry = true;
4384 path_put(&new_path);
4387 path_put(&old_path);
4390 should_retry = false;
4391 lookup_flags |= LOOKUP_REVAL;
4398 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4399 int, newdfd, const char __user *, newname)
4401 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4404 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4406 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4409 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4411 int error = may_create(dir, dentry);
4415 if (!dir->i_op->mknod)
4418 return dir->i_op->mknod(dir, dentry,
4419 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4421 EXPORT_SYMBOL(vfs_whiteout);
4423 int readlink_copy(char __user *buffer, int buflen, const char *link)
4425 int len = PTR_ERR(link);
4430 if (len > (unsigned) buflen)
4432 if (copy_to_user(buffer, link, len))
4437 EXPORT_SYMBOL(readlink_copy);
4440 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4441 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4442 * using) it for any given inode is up to filesystem.
4444 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4447 struct inode *inode = d_inode(dentry);
4448 const char *link = inode->i_link;
4452 link = inode->i_op->follow_link(dentry, &cookie);
4454 return PTR_ERR(link);
4456 res = readlink_copy(buffer, buflen, link);
4457 if (inode->i_op->put_link)
4458 inode->i_op->put_link(inode, cookie);
4461 EXPORT_SYMBOL(generic_readlink);
4463 /* get the link contents into pagecache */
4464 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4468 struct address_space *mapping = dentry->d_inode->i_mapping;
4469 page = read_mapping_page(mapping, 0, NULL);
4474 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4478 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4480 struct page *page = NULL;
4481 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4484 page_cache_release(page);
4488 EXPORT_SYMBOL(page_readlink);
4490 const char *page_follow_link_light(struct dentry *dentry, void **cookie)
4492 struct page *page = NULL;
4493 char *res = page_getlink(dentry, &page);
4498 EXPORT_SYMBOL(page_follow_link_light);
4500 void page_put_link(struct inode *unused, void *cookie)
4502 struct page *page = cookie;
4504 page_cache_release(page);
4506 EXPORT_SYMBOL(page_put_link);
4509 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4511 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4513 struct address_space *mapping = inode->i_mapping;
4518 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4520 flags |= AOP_FLAG_NOFS;
4523 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4524 flags, &page, &fsdata);
4528 kaddr = kmap_atomic(page);
4529 memcpy(kaddr, symname, len-1);
4530 kunmap_atomic(kaddr);
4532 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4539 mark_inode_dirty(inode);
4544 EXPORT_SYMBOL(__page_symlink);
4546 int page_symlink(struct inode *inode, const char *symname, int len)
4548 return __page_symlink(inode, symname, len,
4549 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4551 EXPORT_SYMBOL(page_symlink);
4553 const struct inode_operations page_symlink_inode_operations = {
4554 .readlink = generic_readlink,
4555 .follow_link = page_follow_link_light,
4556 .put_link = page_put_link,
4558 EXPORT_SYMBOL(page_symlink_inode_operations);