4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125 getname_flags(const char __user *filename, int flags, int *empty)
127 struct filename *result;
131 result = audit_reusename(filename);
135 result = __getname();
136 if (unlikely(!result))
137 return ERR_PTR(-ENOMEM);
140 * First, try to embed the struct filename inside the names_cache
143 kname = (char *)result->iname;
144 result->name = kname;
146 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
147 if (unlikely(len < 0)) {
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
158 if (unlikely(len == EMBEDDED_NAME_MAX)) {
159 const size_t size = offsetof(struct filename, iname[1]);
160 kname = (char *)result;
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
167 result = kzalloc(size, GFP_KERNEL);
168 if (unlikely(!result)) {
170 return ERR_PTR(-ENOMEM);
172 result->name = kname;
173 len = strncpy_from_user(kname, filename, PATH_MAX);
174 if (unlikely(len < 0)) {
179 if (unlikely(len == PATH_MAX)) {
182 return ERR_PTR(-ENAMETOOLONG);
187 /* The empty path is special. */
188 if (unlikely(!len)) {
191 if (!(flags & LOOKUP_EMPTY)) {
193 return ERR_PTR(-ENOENT);
197 result->uptr = filename;
198 result->aname = NULL;
199 audit_getname(result);
204 getname(const char __user * filename)
206 return getname_flags(filename, 0, NULL);
210 getname_kernel(const char * filename)
212 struct filename *result;
213 int len = strlen(filename) + 1;
215 result = __getname();
216 if (unlikely(!result))
217 return ERR_PTR(-ENOMEM);
219 if (len <= EMBEDDED_NAME_MAX) {
220 result->name = (char *)result->iname;
221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
227 return ERR_PTR(-ENOMEM);
229 tmp->name = (char *)result;
233 return ERR_PTR(-ENAMETOOLONG);
235 memcpy((char *)result->name, filename, len);
237 result->aname = NULL;
239 audit_getname(result);
244 void putname(struct filename *name)
246 BUG_ON(name->refcnt <= 0);
248 if (--name->refcnt > 0)
251 if (name->name != name->iname) {
252 __putname(name->name);
258 static int check_acl(struct inode *inode, int mask)
260 #ifdef CONFIG_FS_POSIX_ACL
261 struct posix_acl *acl;
263 if (mask & MAY_NOT_BLOCK) {
264 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 /* no ->get_acl() calls in RCU mode... */
268 if (acl == ACL_NOT_CACHED)
270 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273 acl = get_acl(inode, ACL_TYPE_ACCESS);
277 int error = posix_acl_permission(inode, acl, mask);
278 posix_acl_release(acl);
287 * This does the basic permission checking
289 static int acl_permission_check(struct inode *inode, int mask)
291 unsigned int mode = inode->i_mode;
293 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
297 int error = check_acl(inode, mask);
298 if (error != -EAGAIN)
302 if (in_group_p(inode->i_gid))
307 * If the DACs are ok we don't need any capability check.
309 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
315 * generic_permission - check for access rights on a Posix-like filesystem
316 * @inode: inode to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
322 * are used for other things.
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
328 int generic_permission(struct inode *inode, int mask)
333 * Do the basic permission checks.
335 ret = acl_permission_check(inode, mask);
339 if (S_ISDIR(inode->i_mode)) {
340 /* DACs are overridable for directories */
341 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
343 if (!(mask & MAY_WRITE))
344 if (capable_wrt_inode_uidgid(inode,
345 CAP_DAC_READ_SEARCH))
350 * Read/write DACs are always overridable.
351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
354 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
355 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
359 * Searching includes executable on directories, else just read.
361 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
362 if (mask == MAY_READ)
363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
368 EXPORT_SYMBOL(generic_permission);
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
376 static inline int do_inode_permission(struct inode *inode, int mask)
378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379 if (likely(inode->i_op->permission))
380 return inode->i_op->permission(inode, mask);
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode->i_lock);
384 inode->i_opflags |= IOP_FASTPERM;
385 spin_unlock(&inode->i_lock);
387 return generic_permission(inode, mask);
391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395 * Check for read/write/execute permissions on an inode.
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
402 int __inode_permission(struct inode *inode, int mask)
406 if (unlikely(mask & MAY_WRITE)) {
408 * Nobody gets write access to an immutable file.
410 if (IS_IMMUTABLE(inode))
414 retval = do_inode_permission(inode, mask);
418 retval = devcgroup_inode_permission(inode, mask);
422 return security_inode_permission(inode, mask);
424 EXPORT_SYMBOL(__inode_permission);
427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Separate out file-system wide checks from inode-specific permission checks.
434 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
436 if (unlikely(mask & MAY_WRITE)) {
437 umode_t mode = inode->i_mode;
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb->s_flags & MS_RDONLY) &&
441 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 int inode_permission(struct inode *inode, int mask)
462 retval = sb_permission(inode->i_sb, inode, mask);
465 return __inode_permission(inode, mask);
467 EXPORT_SYMBOL(inode_permission);
470 * path_get - get a reference to a path
471 * @path: path to get the reference to
473 * Given a path increment the reference count to the dentry and the vfsmount.
475 void path_get(const struct path *path)
480 EXPORT_SYMBOL(path_get);
483 * path_put - put a reference to a path
484 * @path: path to put the reference to
486 * Given a path decrement the reference count to the dentry and the vfsmount.
488 void path_put(const struct path *path)
493 EXPORT_SYMBOL(path_put);
495 #define EMBEDDED_LEVELS 2
503 struct inode *inode; /* path.dentry.d_inode */
513 } *stack, internal[EMBEDDED_LEVELS];
516 static void set_nameidata(struct nameidata *nd)
518 nd->stack = nd->internal;
521 static void restore_nameidata(struct nameidata *nd)
523 if (nd->stack != nd->internal) {
525 nd->stack = nd->internal;
529 static int __nd_alloc_stack(struct nameidata *nd)
531 struct saved *p = kmalloc((MAXSYMLINKS + 1) * sizeof(struct saved),
535 memcpy(p, nd->internal, sizeof(nd->internal));
540 static inline int nd_alloc_stack(struct nameidata *nd)
542 if (likely(nd->depth != EMBEDDED_LEVELS))
544 if (likely(nd->stack != nd->internal))
546 return __nd_alloc_stack(nd);
550 * Path walking has 2 modes, rcu-walk and ref-walk (see
551 * Documentation/filesystems/path-lookup.txt). In situations when we can't
552 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
553 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
554 * mode. Refcounts are grabbed at the last known good point before rcu-walk
555 * got stuck, so ref-walk may continue from there. If this is not successful
556 * (eg. a seqcount has changed), then failure is returned and it's up to caller
557 * to restart the path walk from the beginning in ref-walk mode.
561 * unlazy_walk - try to switch to ref-walk mode.
562 * @nd: nameidata pathwalk data
563 * @dentry: child of nd->path.dentry or NULL
564 * Returns: 0 on success, -ECHILD on failure
566 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
567 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
568 * @nd or NULL. Must be called from rcu-walk context.
570 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
572 struct fs_struct *fs = current->fs;
573 struct dentry *parent = nd->path.dentry;
575 BUG_ON(!(nd->flags & LOOKUP_RCU));
578 * After legitimizing the bastards, terminate_walk()
579 * will do the right thing for non-RCU mode, and all our
580 * subsequent exit cases should rcu_read_unlock()
581 * before returning. Do vfsmount first; if dentry
582 * can't be legitimized, just set nd->path.dentry to NULL
583 * and rely on dput(NULL) being a no-op.
585 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
587 nd->flags &= ~LOOKUP_RCU;
589 if (!lockref_get_not_dead(&parent->d_lockref)) {
590 nd->path.dentry = NULL;
595 * For a negative lookup, the lookup sequence point is the parents
596 * sequence point, and it only needs to revalidate the parent dentry.
598 * For a positive lookup, we need to move both the parent and the
599 * dentry from the RCU domain to be properly refcounted. And the
600 * sequence number in the dentry validates *both* dentry counters,
601 * since we checked the sequence number of the parent after we got
602 * the child sequence number. So we know the parent must still
603 * be valid if the child sequence number is still valid.
606 if (read_seqcount_retry(&parent->d_seq, nd->seq))
608 BUG_ON(nd->inode != parent->d_inode);
610 if (!lockref_get_not_dead(&dentry->d_lockref))
612 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
617 * Sequence counts matched. Now make sure that the root is
618 * still valid and get it if required.
620 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
621 spin_lock(&fs->lock);
622 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
623 goto unlock_and_drop_dentry;
625 spin_unlock(&fs->lock);
631 unlock_and_drop_dentry:
632 spin_unlock(&fs->lock);
640 if (!(nd->flags & LOOKUP_ROOT))
645 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
647 return dentry->d_op->d_revalidate(dentry, flags);
651 * complete_walk - successful completion of path walk
652 * @nd: pointer nameidata
654 * If we had been in RCU mode, drop out of it and legitimize nd->path.
655 * Revalidate the final result, unless we'd already done that during
656 * the path walk or the filesystem doesn't ask for it. Return 0 on
657 * success, -error on failure. In case of failure caller does not
658 * need to drop nd->path.
660 static int complete_walk(struct nameidata *nd)
662 struct dentry *dentry = nd->path.dentry;
665 if (nd->flags & LOOKUP_RCU) {
666 nd->flags &= ~LOOKUP_RCU;
667 if (!(nd->flags & LOOKUP_ROOT))
670 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
674 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
676 mntput(nd->path.mnt);
679 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
682 mntput(nd->path.mnt);
688 if (likely(!(nd->flags & LOOKUP_JUMPED)))
691 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
694 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
705 static __always_inline void set_root(struct nameidata *nd)
707 get_fs_root(current->fs, &nd->root);
710 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
712 struct fs_struct *fs = current->fs;
716 seq = read_seqcount_begin(&fs->seq);
718 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
719 } while (read_seqcount_retry(&fs->seq, seq));
723 static void path_put_conditional(struct path *path, struct nameidata *nd)
726 if (path->mnt != nd->path.mnt)
730 static inline void path_to_nameidata(const struct path *path,
731 struct nameidata *nd)
733 if (!(nd->flags & LOOKUP_RCU)) {
734 dput(nd->path.dentry);
735 if (nd->path.mnt != path->mnt)
736 mntput(nd->path.mnt);
738 nd->path.mnt = path->mnt;
739 nd->path.dentry = path->dentry;
743 * Helper to directly jump to a known parsed path from ->follow_link,
744 * caller must have taken a reference to path beforehand.
746 void nd_jump_link(struct nameidata *nd, struct path *path)
751 nd->inode = nd->path.dentry->d_inode;
752 nd->flags |= LOOKUP_JUMPED;
755 static inline void put_link(struct nameidata *nd)
757 struct saved *last = nd->stack + nd->depth;
758 struct inode *inode = last->link.dentry->d_inode;
759 if (last->cookie && inode->i_op->put_link)
760 inode->i_op->put_link(last->link.dentry, last->cookie);
761 path_put(&last->link);
764 int sysctl_protected_symlinks __read_mostly = 0;
765 int sysctl_protected_hardlinks __read_mostly = 0;
768 * may_follow_link - Check symlink following for unsafe situations
769 * @link: The path of the symlink
770 * @nd: nameidata pathwalk data
772 * In the case of the sysctl_protected_symlinks sysctl being enabled,
773 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
774 * in a sticky world-writable directory. This is to protect privileged
775 * processes from failing races against path names that may change out
776 * from under them by way of other users creating malicious symlinks.
777 * It will permit symlinks to be followed only when outside a sticky
778 * world-writable directory, or when the uid of the symlink and follower
779 * match, or when the directory owner matches the symlink's owner.
781 * Returns 0 if following the symlink is allowed, -ve on error.
783 static inline int may_follow_link(struct path *link, struct nameidata *nd)
785 const struct inode *inode;
786 const struct inode *parent;
788 if (!sysctl_protected_symlinks)
791 /* Allowed if owner and follower match. */
792 inode = link->dentry->d_inode;
793 if (uid_eq(current_cred()->fsuid, inode->i_uid))
796 /* Allowed if parent directory not sticky and world-writable. */
797 parent = nd->path.dentry->d_inode;
798 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
801 /* Allowed if parent directory and link owner match. */
802 if (uid_eq(parent->i_uid, inode->i_uid))
805 audit_log_link_denied("follow_link", link);
806 path_put_conditional(link, nd);
812 * safe_hardlink_source - Check for safe hardlink conditions
813 * @inode: the source inode to hardlink from
815 * Return false if at least one of the following conditions:
816 * - inode is not a regular file
818 * - inode is setgid and group-exec
819 * - access failure for read and write
821 * Otherwise returns true.
823 static bool safe_hardlink_source(struct inode *inode)
825 umode_t mode = inode->i_mode;
827 /* Special files should not get pinned to the filesystem. */
831 /* Setuid files should not get pinned to the filesystem. */
835 /* Executable setgid files should not get pinned to the filesystem. */
836 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
839 /* Hardlinking to unreadable or unwritable sources is dangerous. */
840 if (inode_permission(inode, MAY_READ | MAY_WRITE))
847 * may_linkat - Check permissions for creating a hardlink
848 * @link: the source to hardlink from
850 * Block hardlink when all of:
851 * - sysctl_protected_hardlinks enabled
852 * - fsuid does not match inode
853 * - hardlink source is unsafe (see safe_hardlink_source() above)
856 * Returns 0 if successful, -ve on error.
858 static int may_linkat(struct path *link)
860 const struct cred *cred;
863 if (!sysctl_protected_hardlinks)
866 cred = current_cred();
867 inode = link->dentry->d_inode;
869 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
870 * otherwise, it must be a safe source.
872 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
876 audit_log_link_denied("linkat", link);
880 static __always_inline
881 const char *get_link(struct nameidata *nd)
883 struct saved *last = nd->stack + nd->depth;
884 struct dentry *dentry = nd->link.dentry;
885 struct inode *inode = dentry->d_inode;
889 BUG_ON(nd->flags & LOOKUP_RCU);
891 if (nd->link.mnt == nd->path.mnt)
892 mntget(nd->link.mnt);
894 if (unlikely(current->total_link_count >= MAXSYMLINKS)) {
897 return ERR_PTR(-ELOOP);
900 last->link = nd->link;
904 current->total_link_count++;
906 touch_atime(&last->link);
908 error = security_inode_follow_link(dentry);
909 res = ERR_PTR(error);
913 nd->last_type = LAST_BIND;
916 res = inode->i_op->follow_link(dentry, &last->cookie, nd);
920 path_put(&last->link);
928 static int follow_up_rcu(struct path *path)
930 struct mount *mnt = real_mount(path->mnt);
931 struct mount *parent;
932 struct dentry *mountpoint;
934 parent = mnt->mnt_parent;
935 if (&parent->mnt == path->mnt)
937 mountpoint = mnt->mnt_mountpoint;
938 path->dentry = mountpoint;
939 path->mnt = &parent->mnt;
944 * follow_up - Find the mountpoint of path's vfsmount
946 * Given a path, find the mountpoint of its source file system.
947 * Replace @path with the path of the mountpoint in the parent mount.
950 * Return 1 if we went up a level and 0 if we were already at the
953 int follow_up(struct path *path)
955 struct mount *mnt = real_mount(path->mnt);
956 struct mount *parent;
957 struct dentry *mountpoint;
959 read_seqlock_excl(&mount_lock);
960 parent = mnt->mnt_parent;
962 read_sequnlock_excl(&mount_lock);
965 mntget(&parent->mnt);
966 mountpoint = dget(mnt->mnt_mountpoint);
967 read_sequnlock_excl(&mount_lock);
969 path->dentry = mountpoint;
971 path->mnt = &parent->mnt;
974 EXPORT_SYMBOL(follow_up);
977 * Perform an automount
978 * - return -EISDIR to tell follow_managed() to stop and return the path we
981 static int follow_automount(struct path *path, unsigned flags,
984 struct vfsmount *mnt;
987 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
990 /* We don't want to mount if someone's just doing a stat -
991 * unless they're stat'ing a directory and appended a '/' to
994 * We do, however, want to mount if someone wants to open or
995 * create a file of any type under the mountpoint, wants to
996 * traverse through the mountpoint or wants to open the
997 * mounted directory. Also, autofs may mark negative dentries
998 * as being automount points. These will need the attentions
999 * of the daemon to instantiate them before they can be used.
1001 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1002 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1003 path->dentry->d_inode)
1006 current->total_link_count++;
1007 if (current->total_link_count >= 40)
1010 mnt = path->dentry->d_op->d_automount(path);
1013 * The filesystem is allowed to return -EISDIR here to indicate
1014 * it doesn't want to automount. For instance, autofs would do
1015 * this so that its userspace daemon can mount on this dentry.
1017 * However, we can only permit this if it's a terminal point in
1018 * the path being looked up; if it wasn't then the remainder of
1019 * the path is inaccessible and we should say so.
1021 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
1023 return PTR_ERR(mnt);
1026 if (!mnt) /* mount collision */
1029 if (!*need_mntput) {
1030 /* lock_mount() may release path->mnt on error */
1032 *need_mntput = true;
1034 err = finish_automount(mnt, path);
1038 /* Someone else made a mount here whilst we were busy */
1043 path->dentry = dget(mnt->mnt_root);
1052 * Handle a dentry that is managed in some way.
1053 * - Flagged for transit management (autofs)
1054 * - Flagged as mountpoint
1055 * - Flagged as automount point
1057 * This may only be called in refwalk mode.
1059 * Serialization is taken care of in namespace.c
1061 static int follow_managed(struct path *path, unsigned flags)
1063 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1065 bool need_mntput = false;
1068 /* Given that we're not holding a lock here, we retain the value in a
1069 * local variable for each dentry as we look at it so that we don't see
1070 * the components of that value change under us */
1071 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1072 managed &= DCACHE_MANAGED_DENTRY,
1073 unlikely(managed != 0)) {
1074 /* Allow the filesystem to manage the transit without i_mutex
1076 if (managed & DCACHE_MANAGE_TRANSIT) {
1077 BUG_ON(!path->dentry->d_op);
1078 BUG_ON(!path->dentry->d_op->d_manage);
1079 ret = path->dentry->d_op->d_manage(path->dentry, false);
1084 /* Transit to a mounted filesystem. */
1085 if (managed & DCACHE_MOUNTED) {
1086 struct vfsmount *mounted = lookup_mnt(path);
1091 path->mnt = mounted;
1092 path->dentry = dget(mounted->mnt_root);
1097 /* Something is mounted on this dentry in another
1098 * namespace and/or whatever was mounted there in this
1099 * namespace got unmounted before lookup_mnt() could
1103 /* Handle an automount point */
1104 if (managed & DCACHE_NEED_AUTOMOUNT) {
1105 ret = follow_automount(path, flags, &need_mntput);
1111 /* We didn't change the current path point */
1115 if (need_mntput && path->mnt == mnt)
1119 return ret < 0 ? ret : need_mntput;
1122 int follow_down_one(struct path *path)
1124 struct vfsmount *mounted;
1126 mounted = lookup_mnt(path);
1130 path->mnt = mounted;
1131 path->dentry = dget(mounted->mnt_root);
1136 EXPORT_SYMBOL(follow_down_one);
1138 static inline int managed_dentry_rcu(struct dentry *dentry)
1140 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1141 dentry->d_op->d_manage(dentry, true) : 0;
1145 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1146 * we meet a managed dentry that would need blocking.
1148 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1149 struct inode **inode)
1152 struct mount *mounted;
1154 * Don't forget we might have a non-mountpoint managed dentry
1155 * that wants to block transit.
1157 switch (managed_dentry_rcu(path->dentry)) {
1167 if (!d_mountpoint(path->dentry))
1168 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1170 mounted = __lookup_mnt(path->mnt, path->dentry);
1173 path->mnt = &mounted->mnt;
1174 path->dentry = mounted->mnt.mnt_root;
1175 nd->flags |= LOOKUP_JUMPED;
1176 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1178 * Update the inode too. We don't need to re-check the
1179 * dentry sequence number here after this d_inode read,
1180 * because a mount-point is always pinned.
1182 *inode = path->dentry->d_inode;
1184 return !read_seqretry(&mount_lock, nd->m_seq) &&
1185 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1188 static int follow_dotdot_rcu(struct nameidata *nd)
1190 struct inode *inode = nd->inode;
1195 if (nd->path.dentry == nd->root.dentry &&
1196 nd->path.mnt == nd->root.mnt) {
1199 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1200 struct dentry *old = nd->path.dentry;
1201 struct dentry *parent = old->d_parent;
1204 inode = parent->d_inode;
1205 seq = read_seqcount_begin(&parent->d_seq);
1206 if (read_seqcount_retry(&old->d_seq, nd->seq))
1208 nd->path.dentry = parent;
1212 if (!follow_up_rcu(&nd->path))
1214 inode = nd->path.dentry->d_inode;
1215 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1217 while (d_mountpoint(nd->path.dentry)) {
1218 struct mount *mounted;
1219 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1222 nd->path.mnt = &mounted->mnt;
1223 nd->path.dentry = mounted->mnt.mnt_root;
1224 inode = nd->path.dentry->d_inode;
1225 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1226 if (read_seqretry(&mount_lock, nd->m_seq))
1233 nd->flags &= ~LOOKUP_RCU;
1234 if (!(nd->flags & LOOKUP_ROOT))
1235 nd->root.mnt = NULL;
1241 * Follow down to the covering mount currently visible to userspace. At each
1242 * point, the filesystem owning that dentry may be queried as to whether the
1243 * caller is permitted to proceed or not.
1245 int follow_down(struct path *path)
1250 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1251 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1252 /* Allow the filesystem to manage the transit without i_mutex
1255 * We indicate to the filesystem if someone is trying to mount
1256 * something here. This gives autofs the chance to deny anyone
1257 * other than its daemon the right to mount on its
1260 * The filesystem may sleep at this point.
1262 if (managed & DCACHE_MANAGE_TRANSIT) {
1263 BUG_ON(!path->dentry->d_op);
1264 BUG_ON(!path->dentry->d_op->d_manage);
1265 ret = path->dentry->d_op->d_manage(
1266 path->dentry, false);
1268 return ret == -EISDIR ? 0 : ret;
1271 /* Transit to a mounted filesystem. */
1272 if (managed & DCACHE_MOUNTED) {
1273 struct vfsmount *mounted = lookup_mnt(path);
1278 path->mnt = mounted;
1279 path->dentry = dget(mounted->mnt_root);
1283 /* Don't handle automount points here */
1288 EXPORT_SYMBOL(follow_down);
1291 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1293 static void follow_mount(struct path *path)
1295 while (d_mountpoint(path->dentry)) {
1296 struct vfsmount *mounted = lookup_mnt(path);
1301 path->mnt = mounted;
1302 path->dentry = dget(mounted->mnt_root);
1306 static void follow_dotdot(struct nameidata *nd)
1312 struct dentry *old = nd->path.dentry;
1314 if (nd->path.dentry == nd->root.dentry &&
1315 nd->path.mnt == nd->root.mnt) {
1318 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1319 /* rare case of legitimate dget_parent()... */
1320 nd->path.dentry = dget_parent(nd->path.dentry);
1324 if (!follow_up(&nd->path))
1327 follow_mount(&nd->path);
1328 nd->inode = nd->path.dentry->d_inode;
1332 * This looks up the name in dcache, possibly revalidates the old dentry and
1333 * allocates a new one if not found or not valid. In the need_lookup argument
1334 * returns whether i_op->lookup is necessary.
1336 * dir->d_inode->i_mutex must be held
1338 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1339 unsigned int flags, bool *need_lookup)
1341 struct dentry *dentry;
1344 *need_lookup = false;
1345 dentry = d_lookup(dir, name);
1347 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1348 error = d_revalidate(dentry, flags);
1349 if (unlikely(error <= 0)) {
1352 return ERR_PTR(error);
1354 d_invalidate(dentry);
1363 dentry = d_alloc(dir, name);
1364 if (unlikely(!dentry))
1365 return ERR_PTR(-ENOMEM);
1367 *need_lookup = true;
1373 * Call i_op->lookup on the dentry. The dentry must be negative and
1376 * dir->d_inode->i_mutex must be held
1378 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1383 /* Don't create child dentry for a dead directory. */
1384 if (unlikely(IS_DEADDIR(dir))) {
1386 return ERR_PTR(-ENOENT);
1389 old = dir->i_op->lookup(dir, dentry, flags);
1390 if (unlikely(old)) {
1397 static struct dentry *__lookup_hash(struct qstr *name,
1398 struct dentry *base, unsigned int flags)
1401 struct dentry *dentry;
1403 dentry = lookup_dcache(name, base, flags, &need_lookup);
1407 return lookup_real(base->d_inode, dentry, flags);
1411 * It's more convoluted than I'd like it to be, but... it's still fairly
1412 * small and for now I'd prefer to have fast path as straight as possible.
1413 * It _is_ time-critical.
1415 static int lookup_fast(struct nameidata *nd,
1416 struct path *path, struct inode **inode)
1418 struct vfsmount *mnt = nd->path.mnt;
1419 struct dentry *dentry, *parent = nd->path.dentry;
1425 * Rename seqlock is not required here because in the off chance
1426 * of a false negative due to a concurrent rename, we're going to
1427 * do the non-racy lookup, below.
1429 if (nd->flags & LOOKUP_RCU) {
1432 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1437 * This sequence count validates that the inode matches
1438 * the dentry name information from lookup.
1440 *inode = dentry->d_inode;
1441 negative = d_is_negative(dentry);
1442 if (read_seqcount_retry(&dentry->d_seq, seq))
1448 * This sequence count validates that the parent had no
1449 * changes while we did the lookup of the dentry above.
1451 * The memory barrier in read_seqcount_begin of child is
1452 * enough, we can use __read_seqcount_retry here.
1454 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1458 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1459 status = d_revalidate(dentry, nd->flags);
1460 if (unlikely(status <= 0)) {
1461 if (status != -ECHILD)
1467 path->dentry = dentry;
1468 if (likely(__follow_mount_rcu(nd, path, inode)))
1471 if (unlazy_walk(nd, dentry))
1474 dentry = __d_lookup(parent, &nd->last);
1477 if (unlikely(!dentry))
1480 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1481 status = d_revalidate(dentry, nd->flags);
1482 if (unlikely(status <= 0)) {
1487 d_invalidate(dentry);
1492 if (unlikely(d_is_negative(dentry))) {
1497 path->dentry = dentry;
1498 err = follow_managed(path, nd->flags);
1499 if (unlikely(err < 0)) {
1500 path_put_conditional(path, nd);
1504 nd->flags |= LOOKUP_JUMPED;
1505 *inode = path->dentry->d_inode;
1512 /* Fast lookup failed, do it the slow way */
1513 static int lookup_slow(struct nameidata *nd, struct path *path)
1515 struct dentry *dentry, *parent;
1518 parent = nd->path.dentry;
1519 BUG_ON(nd->inode != parent->d_inode);
1521 mutex_lock(&parent->d_inode->i_mutex);
1522 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1523 mutex_unlock(&parent->d_inode->i_mutex);
1525 return PTR_ERR(dentry);
1526 path->mnt = nd->path.mnt;
1527 path->dentry = dentry;
1528 err = follow_managed(path, nd->flags);
1529 if (unlikely(err < 0)) {
1530 path_put_conditional(path, nd);
1534 nd->flags |= LOOKUP_JUMPED;
1538 static inline int may_lookup(struct nameidata *nd)
1540 if (nd->flags & LOOKUP_RCU) {
1541 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1544 if (unlazy_walk(nd, NULL))
1547 return inode_permission(nd->inode, MAY_EXEC);
1550 static inline int handle_dots(struct nameidata *nd, int type)
1552 if (type == LAST_DOTDOT) {
1553 if (nd->flags & LOOKUP_RCU) {
1554 if (follow_dotdot_rcu(nd))
1562 static void terminate_walk(struct nameidata *nd)
1564 if (!(nd->flags & LOOKUP_RCU)) {
1565 path_put(&nd->path);
1567 nd->flags &= ~LOOKUP_RCU;
1568 if (!(nd->flags & LOOKUP_ROOT))
1569 nd->root.mnt = NULL;
1575 * Do we need to follow links? We _really_ want to be able
1576 * to do this check without having to look at inode->i_op,
1577 * so we keep a cache of "no, this doesn't need follow_link"
1578 * for the common case.
1580 static inline int should_follow_link(struct dentry *dentry, int follow)
1582 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1585 static int walk_component(struct nameidata *nd, int follow)
1588 struct inode *inode;
1591 * "." and ".." are special - ".." especially so because it has
1592 * to be able to know about the current root directory and
1593 * parent relationships.
1595 if (unlikely(nd->last_type != LAST_NORM))
1596 return handle_dots(nd, nd->last_type);
1597 err = lookup_fast(nd, &path, &inode);
1598 if (unlikely(err)) {
1602 err = lookup_slow(nd, &path);
1606 inode = path.dentry->d_inode;
1608 if (d_is_negative(path.dentry))
1612 if (should_follow_link(path.dentry, follow)) {
1613 if (nd->flags & LOOKUP_RCU) {
1614 if (unlikely(nd->path.mnt != path.mnt ||
1615 unlazy_walk(nd, path.dentry))) {
1620 BUG_ON(inode != path.dentry->d_inode);
1624 path_to_nameidata(&path, nd);
1629 path_to_nameidata(&path, nd);
1636 * We can do the critical dentry name comparison and hashing
1637 * operations one word at a time, but we are limited to:
1639 * - Architectures with fast unaligned word accesses. We could
1640 * do a "get_unaligned()" if this helps and is sufficiently
1643 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1644 * do not trap on the (extremely unlikely) case of a page
1645 * crossing operation.
1647 * - Furthermore, we need an efficient 64-bit compile for the
1648 * 64-bit case in order to generate the "number of bytes in
1649 * the final mask". Again, that could be replaced with a
1650 * efficient population count instruction or similar.
1652 #ifdef CONFIG_DCACHE_WORD_ACCESS
1654 #include <asm/word-at-a-time.h>
1658 static inline unsigned int fold_hash(unsigned long hash)
1660 return hash_64(hash, 32);
1663 #else /* 32-bit case */
1665 #define fold_hash(x) (x)
1669 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1671 unsigned long a, mask;
1672 unsigned long hash = 0;
1675 a = load_unaligned_zeropad(name);
1676 if (len < sizeof(unsigned long))
1680 name += sizeof(unsigned long);
1681 len -= sizeof(unsigned long);
1685 mask = bytemask_from_count(len);
1688 return fold_hash(hash);
1690 EXPORT_SYMBOL(full_name_hash);
1693 * Calculate the length and hash of the path component, and
1694 * return the "hash_len" as the result.
1696 static inline u64 hash_name(const char *name)
1698 unsigned long a, b, adata, bdata, mask, hash, len;
1699 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1702 len = -sizeof(unsigned long);
1704 hash = (hash + a) * 9;
1705 len += sizeof(unsigned long);
1706 a = load_unaligned_zeropad(name+len);
1707 b = a ^ REPEAT_BYTE('/');
1708 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1710 adata = prep_zero_mask(a, adata, &constants);
1711 bdata = prep_zero_mask(b, bdata, &constants);
1713 mask = create_zero_mask(adata | bdata);
1715 hash += a & zero_bytemask(mask);
1716 len += find_zero(mask);
1717 return hashlen_create(fold_hash(hash), len);
1722 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1724 unsigned long hash = init_name_hash();
1726 hash = partial_name_hash(*name++, hash);
1727 return end_name_hash(hash);
1729 EXPORT_SYMBOL(full_name_hash);
1732 * We know there's a real path component here of at least
1735 static inline u64 hash_name(const char *name)
1737 unsigned long hash = init_name_hash();
1738 unsigned long len = 0, c;
1740 c = (unsigned char)*name;
1743 hash = partial_name_hash(c, hash);
1744 c = (unsigned char)name[len];
1745 } while (c && c != '/');
1746 return hashlen_create(end_name_hash(hash), len);
1753 * This is the basic name resolution function, turning a pathname into
1754 * the final dentry. We expect 'base' to be positive and a directory.
1756 * Returns 0 and nd will have valid dentry and mnt on success.
1757 * Returns error and drops reference to input namei data on failure.
1759 static int link_path_walk(const char *name, struct nameidata *nd)
1768 /* At this point we know we have a real path component. */
1773 err = may_lookup(nd);
1777 hash_len = hash_name(name);
1780 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1782 if (name[1] == '.') {
1784 nd->flags |= LOOKUP_JUMPED;
1790 if (likely(type == LAST_NORM)) {
1791 struct dentry *parent = nd->path.dentry;
1792 nd->flags &= ~LOOKUP_JUMPED;
1793 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1794 struct qstr this = { { .hash_len = hash_len }, .name = name };
1795 err = parent->d_op->d_hash(parent, &this);
1798 hash_len = this.hash_len;
1803 nd->last.hash_len = hash_len;
1804 nd->last.name = name;
1805 nd->last_type = type;
1807 name += hashlen_len(hash_len);
1811 * If it wasn't NUL, we know it was '/'. Skip that
1812 * slash, and continue until no more slashes.
1816 } while (unlikely(*name == '/'));
1820 err = walk_component(nd, LOOKUP_FOLLOW);
1828 err = nd_alloc_stack(nd);
1829 if (unlikely(err)) {
1830 path_to_nameidata(&nd->link, nd);
1836 if (unlikely(IS_ERR(s))) {
1849 path_put(&nd->path);
1850 nd->path = nd->root;
1851 path_get(&nd->root);
1852 nd->flags |= LOOKUP_JUMPED;
1853 while (unlikely(*++s == '/'))
1856 nd->inode = nd->path.dentry->d_inode;
1857 nd->stack[nd->depth - 1].name = name;
1864 if (!d_can_lookup(nd->path.dentry)) {
1871 while (unlikely(nd->depth > 1)) {
1877 if (unlikely(nd->depth > 1)) {
1878 name = nd->stack[nd->depth - 1].name;
1879 err = walk_component(nd, LOOKUP_FOLLOW);
1887 static int path_init(int dfd, const struct filename *name, unsigned int flags,
1888 struct nameidata *nd)
1891 const char *s = name->name;
1893 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1894 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1897 if (flags & LOOKUP_ROOT) {
1898 struct dentry *root = nd->root.dentry;
1899 struct inode *inode = root->d_inode;
1901 if (!d_can_lookup(root))
1903 retval = inode_permission(inode, MAY_EXEC);
1907 nd->path = nd->root;
1909 if (flags & LOOKUP_RCU) {
1911 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1912 nd->m_seq = read_seqbegin(&mount_lock);
1914 path_get(&nd->path);
1919 nd->root.mnt = NULL;
1921 nd->m_seq = read_seqbegin(&mount_lock);
1923 if (flags & LOOKUP_RCU) {
1925 nd->seq = set_root_rcu(nd);
1928 path_get(&nd->root);
1930 nd->path = nd->root;
1931 } else if (dfd == AT_FDCWD) {
1932 if (flags & LOOKUP_RCU) {
1933 struct fs_struct *fs = current->fs;
1939 seq = read_seqcount_begin(&fs->seq);
1941 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1942 } while (read_seqcount_retry(&fs->seq, seq));
1944 get_fs_pwd(current->fs, &nd->path);
1947 /* Caller must check execute permissions on the starting path component */
1948 struct fd f = fdget_raw(dfd);
1949 struct dentry *dentry;
1954 dentry = f.file->f_path.dentry;
1957 if (!d_can_lookup(dentry)) {
1963 nd->path = f.file->f_path;
1964 if (flags & LOOKUP_RCU) {
1965 if (f.flags & FDPUT_FPUT)
1967 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1970 path_get(&nd->path);
1975 nd->inode = nd->path.dentry->d_inode;
1976 if (!(flags & LOOKUP_RCU))
1978 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1980 if (!(nd->flags & LOOKUP_ROOT))
1981 nd->root.mnt = NULL;
1985 current->total_link_count = 0;
1987 retval = link_path_walk(s, nd);
1992 static void path_cleanup(struct nameidata *nd)
1994 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1995 path_put(&nd->root);
1996 nd->root.mnt = NULL;
1998 if (unlikely(nd->base))
2002 static int trailing_symlink(struct nameidata *nd)
2005 int error = may_follow_link(&nd->link, nd);
2006 if (unlikely(error))
2008 nd->flags |= LOOKUP_PARENT;
2010 if (unlikely(IS_ERR(s)))
2017 path_put(&nd->path);
2018 nd->path = nd->root;
2019 path_get(&nd->root);
2020 nd->flags |= LOOKUP_JUMPED;
2022 nd->inode = nd->path.dentry->d_inode;
2023 error = link_path_walk(s, nd);
2024 if (unlikely(error)) {
2032 static inline int lookup_last(struct nameidata *nd)
2034 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2035 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2037 nd->flags &= ~LOOKUP_PARENT;
2038 return walk_component(nd, nd->flags & LOOKUP_FOLLOW);
2041 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2042 static int path_lookupat(int dfd, const struct filename *name,
2043 unsigned int flags, struct nameidata *nd)
2048 * Path walking is largely split up into 2 different synchronisation
2049 * schemes, rcu-walk and ref-walk (explained in
2050 * Documentation/filesystems/path-lookup.txt). These share much of the
2051 * path walk code, but some things particularly setup, cleanup, and
2052 * following mounts are sufficiently divergent that functions are
2053 * duplicated. Typically there is a function foo(), and its RCU
2054 * analogue, foo_rcu().
2056 * -ECHILD is the error number of choice (just to avoid clashes) that
2057 * is returned if some aspect of an rcu-walk fails. Such an error must
2058 * be handled by restarting a traditional ref-walk (which will always
2059 * be able to complete).
2061 err = path_init(dfd, name, flags, nd);
2062 if (!err && !(flags & LOOKUP_PARENT)) {
2063 err = lookup_last(nd);
2065 err = trailing_symlink(nd);
2068 err = lookup_last(nd);
2075 err = complete_walk(nd);
2077 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2078 if (!d_can_lookup(nd->path.dentry)) {
2079 path_put(&nd->path);
2088 static int filename_lookup(int dfd, struct filename *name,
2089 unsigned int flags, struct nameidata *nd)
2094 retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2096 if (unlikely(retval == -ECHILD))
2097 retval = path_lookupat(dfd, name, flags, nd);
2098 if (unlikely(retval == -ESTALE))
2099 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2101 if (likely(!retval))
2102 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2103 restore_nameidata(nd);
2107 /* does lookup, returns the object with parent locked */
2108 struct dentry *kern_path_locked(const char *name, struct path *path)
2110 struct filename *filename = getname_kernel(name);
2111 struct nameidata nd;
2115 if (IS_ERR(filename))
2116 return ERR_CAST(filename);
2118 err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd);
2123 if (nd.last_type != LAST_NORM) {
2125 d = ERR_PTR(-EINVAL);
2128 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2129 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2131 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2141 int kern_path(const char *name, unsigned int flags, struct path *path)
2143 struct nameidata nd;
2144 struct filename *filename = getname_kernel(name);
2145 int res = PTR_ERR(filename);
2147 if (!IS_ERR(filename)) {
2148 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2155 EXPORT_SYMBOL(kern_path);
2158 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2159 * @dentry: pointer to dentry of the base directory
2160 * @mnt: pointer to vfs mount of the base directory
2161 * @name: pointer to file name
2162 * @flags: lookup flags
2163 * @path: pointer to struct path to fill
2165 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2166 const char *name, unsigned int flags,
2169 struct filename *filename = getname_kernel(name);
2170 int err = PTR_ERR(filename);
2172 BUG_ON(flags & LOOKUP_PARENT);
2174 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2175 if (!IS_ERR(filename)) {
2176 struct nameidata nd;
2177 nd.root.dentry = dentry;
2179 err = filename_lookup(AT_FDCWD, filename,
2180 flags | LOOKUP_ROOT, &nd);
2187 EXPORT_SYMBOL(vfs_path_lookup);
2190 * lookup_one_len - filesystem helper to lookup single pathname component
2191 * @name: pathname component to lookup
2192 * @base: base directory to lookup from
2193 * @len: maximum length @len should be interpreted to
2195 * Note that this routine is purely a helper for filesystem usage and should
2196 * not be called by generic code.
2198 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2204 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2208 this.hash = full_name_hash(name, len);
2210 return ERR_PTR(-EACCES);
2212 if (unlikely(name[0] == '.')) {
2213 if (len < 2 || (len == 2 && name[1] == '.'))
2214 return ERR_PTR(-EACCES);
2218 c = *(const unsigned char *)name++;
2219 if (c == '/' || c == '\0')
2220 return ERR_PTR(-EACCES);
2223 * See if the low-level filesystem might want
2224 * to use its own hash..
2226 if (base->d_flags & DCACHE_OP_HASH) {
2227 int err = base->d_op->d_hash(base, &this);
2229 return ERR_PTR(err);
2232 err = inode_permission(base->d_inode, MAY_EXEC);
2234 return ERR_PTR(err);
2236 return __lookup_hash(&this, base, 0);
2238 EXPORT_SYMBOL(lookup_one_len);
2240 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2241 struct path *path, int *empty)
2243 struct nameidata nd;
2244 struct filename *tmp = getname_flags(name, flags, empty);
2245 int err = PTR_ERR(tmp);
2248 BUG_ON(flags & LOOKUP_PARENT);
2250 err = filename_lookup(dfd, tmp, flags, &nd);
2258 int user_path_at(int dfd, const char __user *name, unsigned flags,
2261 return user_path_at_empty(dfd, name, flags, path, NULL);
2263 EXPORT_SYMBOL(user_path_at);
2266 * NB: most callers don't do anything directly with the reference to the
2267 * to struct filename, but the nd->last pointer points into the name string
2268 * allocated by getname. So we must hold the reference to it until all
2269 * path-walking is complete.
2271 static struct filename *
2272 user_path_parent(int dfd, const char __user *path,
2273 struct path *parent,
2278 struct nameidata nd;
2279 struct filename *s = getname(path);
2282 /* only LOOKUP_REVAL is allowed in extra flags */
2283 flags &= LOOKUP_REVAL;
2288 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, &nd);
2291 return ERR_PTR(error);
2295 *type = nd.last_type;
2301 * mountpoint_last - look up last component for umount
2302 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2303 * @path: pointer to container for result
2305 * This is a special lookup_last function just for umount. In this case, we
2306 * need to resolve the path without doing any revalidation.
2308 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2309 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2310 * in almost all cases, this lookup will be served out of the dcache. The only
2311 * cases where it won't are if nd->last refers to a symlink or the path is
2312 * bogus and it doesn't exist.
2315 * -error: if there was an error during lookup. This includes -ENOENT if the
2316 * lookup found a negative dentry. The nd->path reference will also be
2319 * 0: if we successfully resolved nd->path and found it to not to be a
2320 * symlink that needs to be followed. "path" will also be populated.
2321 * The nd->path reference will also be put.
2323 * 1: if we successfully resolved nd->last and found it to be a symlink
2324 * that needs to be followed. "path" will be populated with the path
2325 * to the link, and nd->path will *not* be put.
2328 mountpoint_last(struct nameidata *nd, struct path *path)
2331 struct dentry *dentry;
2332 struct dentry *dir = nd->path.dentry;
2334 /* If we're in rcuwalk, drop out of it to handle last component */
2335 if (nd->flags & LOOKUP_RCU) {
2336 if (unlazy_walk(nd, NULL)) {
2342 nd->flags &= ~LOOKUP_PARENT;
2344 if (unlikely(nd->last_type != LAST_NORM)) {
2345 error = handle_dots(nd, nd->last_type);
2348 dentry = dget(nd->path.dentry);
2352 mutex_lock(&dir->d_inode->i_mutex);
2353 dentry = d_lookup(dir, &nd->last);
2356 * No cached dentry. Mounted dentries are pinned in the cache,
2357 * so that means that this dentry is probably a symlink or the
2358 * path doesn't actually point to a mounted dentry.
2360 dentry = d_alloc(dir, &nd->last);
2363 mutex_unlock(&dir->d_inode->i_mutex);
2366 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2367 error = PTR_ERR(dentry);
2368 if (IS_ERR(dentry)) {
2369 mutex_unlock(&dir->d_inode->i_mutex);
2373 mutex_unlock(&dir->d_inode->i_mutex);
2376 if (d_is_negative(dentry)) {
2381 path->dentry = dentry;
2382 path->mnt = nd->path.mnt;
2383 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) {
2396 * path_mountpoint - look up a path to be umounted
2397 * @dfd: directory file descriptor to start walk from
2398 * @name: full pathname to walk
2399 * @path: pointer to container for result
2400 * @flags: lookup flags
2402 * Look up the given name, but don't attempt to revalidate the last component.
2403 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2406 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2407 struct nameidata *nd, unsigned int flags)
2409 int err = path_init(dfd, name, flags, nd);
2413 err = mountpoint_last(nd, path);
2415 err = trailing_symlink(nd);
2418 err = mountpoint_last(nd, path);
2428 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2431 struct nameidata nd;
2434 return PTR_ERR(name);
2436 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2437 if (unlikely(error == -ECHILD))
2438 error = path_mountpoint(dfd, name, path, &nd, flags);
2439 if (unlikely(error == -ESTALE))
2440 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2442 audit_inode(name, path->dentry, 0);
2443 restore_nameidata(&nd);
2449 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2450 * @dfd: directory file descriptor
2451 * @name: pathname from userland
2452 * @flags: lookup flags
2453 * @path: pointer to container to hold result
2455 * A umount is a special case for path walking. We're not actually interested
2456 * in the inode in this situation, and ESTALE errors can be a problem. We
2457 * simply want track down the dentry and vfsmount attached at the mountpoint
2458 * and avoid revalidating the last component.
2460 * Returns 0 and populates "path" on success.
2463 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2466 return filename_mountpoint(dfd, getname(name), path, flags);
2470 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2473 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2475 EXPORT_SYMBOL(kern_path_mountpoint);
2477 int __check_sticky(struct inode *dir, struct inode *inode)
2479 kuid_t fsuid = current_fsuid();
2481 if (uid_eq(inode->i_uid, fsuid))
2483 if (uid_eq(dir->i_uid, fsuid))
2485 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2487 EXPORT_SYMBOL(__check_sticky);
2490 * Check whether we can remove a link victim from directory dir, check
2491 * whether the type of victim is right.
2492 * 1. We can't do it if dir is read-only (done in permission())
2493 * 2. We should have write and exec permissions on dir
2494 * 3. We can't remove anything from append-only dir
2495 * 4. We can't do anything with immutable dir (done in permission())
2496 * 5. If the sticky bit on dir is set we should either
2497 * a. be owner of dir, or
2498 * b. be owner of victim, or
2499 * c. have CAP_FOWNER capability
2500 * 6. If the victim is append-only or immutable we can't do antyhing with
2501 * links pointing to it.
2502 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2503 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2504 * 9. We can't remove a root or mountpoint.
2505 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2506 * nfs_async_unlink().
2508 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2510 struct inode *inode = victim->d_inode;
2513 if (d_is_negative(victim))
2517 BUG_ON(victim->d_parent->d_inode != dir);
2518 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2520 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2526 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2527 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2530 if (!d_is_dir(victim))
2532 if (IS_ROOT(victim))
2534 } else if (d_is_dir(victim))
2536 if (IS_DEADDIR(dir))
2538 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2543 /* Check whether we can create an object with dentry child in directory
2545 * 1. We can't do it if child already exists (open has special treatment for
2546 * this case, but since we are inlined it's OK)
2547 * 2. We can't do it if dir is read-only (done in permission())
2548 * 3. We should have write and exec permissions on dir
2549 * 4. We can't do it if dir is immutable (done in permission())
2551 static inline int may_create(struct inode *dir, struct dentry *child)
2553 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2556 if (IS_DEADDIR(dir))
2558 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2562 * p1 and p2 should be directories on the same fs.
2564 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2569 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2573 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2575 p = d_ancestor(p2, p1);
2577 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2578 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2582 p = d_ancestor(p1, p2);
2584 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2585 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2589 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2590 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2593 EXPORT_SYMBOL(lock_rename);
2595 void unlock_rename(struct dentry *p1, struct dentry *p2)
2597 mutex_unlock(&p1->d_inode->i_mutex);
2599 mutex_unlock(&p2->d_inode->i_mutex);
2600 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2603 EXPORT_SYMBOL(unlock_rename);
2605 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2608 int error = may_create(dir, dentry);
2612 if (!dir->i_op->create)
2613 return -EACCES; /* shouldn't it be ENOSYS? */
2616 error = security_inode_create(dir, dentry, mode);
2619 error = dir->i_op->create(dir, dentry, mode, want_excl);
2621 fsnotify_create(dir, dentry);
2624 EXPORT_SYMBOL(vfs_create);
2626 static int may_open(struct path *path, int acc_mode, int flag)
2628 struct dentry *dentry = path->dentry;
2629 struct inode *inode = dentry->d_inode;
2639 switch (inode->i_mode & S_IFMT) {
2643 if (acc_mode & MAY_WRITE)
2648 if (path->mnt->mnt_flags & MNT_NODEV)
2657 error = inode_permission(inode, acc_mode);
2662 * An append-only file must be opened in append mode for writing.
2664 if (IS_APPEND(inode)) {
2665 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2671 /* O_NOATIME can only be set by the owner or superuser */
2672 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2678 static int handle_truncate(struct file *filp)
2680 struct path *path = &filp->f_path;
2681 struct inode *inode = path->dentry->d_inode;
2682 int error = get_write_access(inode);
2686 * Refuse to truncate files with mandatory locks held on them.
2688 error = locks_verify_locked(filp);
2690 error = security_path_truncate(path);
2692 error = do_truncate(path->dentry, 0,
2693 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2696 put_write_access(inode);
2700 static inline int open_to_namei_flags(int flag)
2702 if ((flag & O_ACCMODE) == 3)
2707 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2709 int error = security_path_mknod(dir, dentry, mode, 0);
2713 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2717 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2721 * Attempt to atomically look up, create and open a file from a negative
2724 * Returns 0 if successful. The file will have been created and attached to
2725 * @file by the filesystem calling finish_open().
2727 * Returns 1 if the file was looked up only or didn't need creating. The
2728 * caller will need to perform the open themselves. @path will have been
2729 * updated to point to the new dentry. This may be negative.
2731 * Returns an error code otherwise.
2733 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2734 struct path *path, struct file *file,
2735 const struct open_flags *op,
2736 bool got_write, bool need_lookup,
2739 struct inode *dir = nd->path.dentry->d_inode;
2740 unsigned open_flag = open_to_namei_flags(op->open_flag);
2744 int create_error = 0;
2745 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2748 BUG_ON(dentry->d_inode);
2750 /* Don't create child dentry for a dead directory. */
2751 if (unlikely(IS_DEADDIR(dir))) {
2757 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2758 mode &= ~current_umask();
2760 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2762 open_flag &= ~O_TRUNC;
2765 * Checking write permission is tricky, bacuse we don't know if we are
2766 * going to actually need it: O_CREAT opens should work as long as the
2767 * file exists. But checking existence breaks atomicity. The trick is
2768 * to check access and if not granted clear O_CREAT from the flags.
2770 * Another problem is returing the "right" error value (e.g. for an
2771 * O_EXCL open we want to return EEXIST not EROFS).
2773 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2774 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2775 if (!(open_flag & O_CREAT)) {
2777 * No O_CREATE -> atomicity not a requirement -> fall
2778 * back to lookup + open
2781 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2782 /* Fall back and fail with the right error */
2783 create_error = -EROFS;
2786 /* No side effects, safe to clear O_CREAT */
2787 create_error = -EROFS;
2788 open_flag &= ~O_CREAT;
2792 if (open_flag & O_CREAT) {
2793 error = may_o_create(&nd->path, dentry, mode);
2795 create_error = error;
2796 if (open_flag & O_EXCL)
2798 open_flag &= ~O_CREAT;
2802 if (nd->flags & LOOKUP_DIRECTORY)
2803 open_flag |= O_DIRECTORY;
2805 file->f_path.dentry = DENTRY_NOT_SET;
2806 file->f_path.mnt = nd->path.mnt;
2807 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2810 if (create_error && error == -ENOENT)
2811 error = create_error;
2815 if (error) { /* returned 1, that is */
2816 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2820 if (file->f_path.dentry) {
2822 dentry = file->f_path.dentry;
2824 if (*opened & FILE_CREATED)
2825 fsnotify_create(dir, dentry);
2826 if (!dentry->d_inode) {
2827 WARN_ON(*opened & FILE_CREATED);
2829 error = create_error;
2833 if (excl && !(*opened & FILE_CREATED)) {
2842 * We didn't have the inode before the open, so check open permission
2845 acc_mode = op->acc_mode;
2846 if (*opened & FILE_CREATED) {
2847 WARN_ON(!(open_flag & O_CREAT));
2848 fsnotify_create(dir, dentry);
2849 acc_mode = MAY_OPEN;
2851 error = may_open(&file->f_path, acc_mode, open_flag);
2861 dentry = lookup_real(dir, dentry, nd->flags);
2863 return PTR_ERR(dentry);
2866 int open_flag = op->open_flag;
2868 error = create_error;
2869 if ((open_flag & O_EXCL)) {
2870 if (!dentry->d_inode)
2872 } else if (!dentry->d_inode) {
2874 } else if ((open_flag & O_TRUNC) &&
2878 /* will fail later, go on to get the right error */
2882 path->dentry = dentry;
2883 path->mnt = nd->path.mnt;
2888 * Look up and maybe create and open the last component.
2890 * Must be called with i_mutex held on parent.
2892 * Returns 0 if the file was successfully atomically created (if necessary) and
2893 * opened. In this case the file will be returned attached to @file.
2895 * Returns 1 if the file was not completely opened at this time, though lookups
2896 * and creations will have been performed and the dentry returned in @path will
2897 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2898 * specified then a negative dentry may be returned.
2900 * An error code is returned otherwise.
2902 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2903 * cleared otherwise prior to returning.
2905 static int lookup_open(struct nameidata *nd, struct path *path,
2907 const struct open_flags *op,
2908 bool got_write, int *opened)
2910 struct dentry *dir = nd->path.dentry;
2911 struct inode *dir_inode = dir->d_inode;
2912 struct dentry *dentry;
2916 *opened &= ~FILE_CREATED;
2917 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2919 return PTR_ERR(dentry);
2921 /* Cached positive dentry: will open in f_op->open */
2922 if (!need_lookup && dentry->d_inode)
2925 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2926 return atomic_open(nd, dentry, path, file, op, got_write,
2927 need_lookup, opened);
2931 BUG_ON(dentry->d_inode);
2933 dentry = lookup_real(dir_inode, dentry, nd->flags);
2935 return PTR_ERR(dentry);
2938 /* Negative dentry, just create the file */
2939 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2940 umode_t mode = op->mode;
2941 if (!IS_POSIXACL(dir->d_inode))
2942 mode &= ~current_umask();
2944 * This write is needed to ensure that a
2945 * rw->ro transition does not occur between
2946 * the time when the file is created and when
2947 * a permanent write count is taken through
2948 * the 'struct file' in finish_open().
2954 *opened |= FILE_CREATED;
2955 error = security_path_mknod(&nd->path, dentry, mode, 0);
2958 error = vfs_create(dir->d_inode, dentry, mode,
2959 nd->flags & LOOKUP_EXCL);
2964 path->dentry = dentry;
2965 path->mnt = nd->path.mnt;
2974 * Handle the last step of open()
2976 static int do_last(struct nameidata *nd,
2977 struct file *file, const struct open_flags *op,
2978 int *opened, struct filename *name)
2980 struct dentry *dir = nd->path.dentry;
2981 int open_flag = op->open_flag;
2982 bool will_truncate = (open_flag & O_TRUNC) != 0;
2983 bool got_write = false;
2984 int acc_mode = op->acc_mode;
2985 struct inode *inode;
2986 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2988 bool retried = false;
2991 nd->flags &= ~LOOKUP_PARENT;
2992 nd->flags |= op->intent;
2994 if (nd->last_type != LAST_NORM) {
2995 error = handle_dots(nd, nd->last_type);
3001 if (!(open_flag & O_CREAT)) {
3002 if (nd->last.name[nd->last.len])
3003 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3004 /* we _can_ be in RCU mode here */
3005 error = lookup_fast(nd, &path, &inode);
3012 BUG_ON(nd->inode != dir->d_inode);
3014 /* create side of things */
3016 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3017 * has been cleared when we got to the last component we are
3020 error = complete_walk(nd);
3024 audit_inode(name, dir, LOOKUP_PARENT);
3026 /* trailing slashes? */
3027 if (nd->last.name[nd->last.len])
3032 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3033 error = mnt_want_write(nd->path.mnt);
3037 * do _not_ fail yet - we might not need that or fail with
3038 * a different error; let lookup_open() decide; we'll be
3039 * dropping this one anyway.
3042 mutex_lock(&dir->d_inode->i_mutex);
3043 error = lookup_open(nd, &path, file, op, got_write, opened);
3044 mutex_unlock(&dir->d_inode->i_mutex);
3050 if ((*opened & FILE_CREATED) ||
3051 !S_ISREG(file_inode(file)->i_mode))
3052 will_truncate = false;
3054 audit_inode(name, file->f_path.dentry, 0);
3058 if (*opened & FILE_CREATED) {
3059 /* Don't check for write permission, don't truncate */
3060 open_flag &= ~O_TRUNC;
3061 will_truncate = false;
3062 acc_mode = MAY_OPEN;
3063 path_to_nameidata(&path, nd);
3064 goto finish_open_created;
3068 * create/update audit record if it already exists.
3070 if (d_is_positive(path.dentry))
3071 audit_inode(name, path.dentry, 0);
3074 * If atomic_open() acquired write access it is dropped now due to
3075 * possible mount and symlink following (this might be optimized away if
3079 mnt_drop_write(nd->path.mnt);
3084 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3087 error = follow_managed(&path, nd->flags);
3092 nd->flags |= LOOKUP_JUMPED;
3094 BUG_ON(nd->flags & LOOKUP_RCU);
3095 inode = path.dentry->d_inode;
3097 if (d_is_negative(path.dentry)) {
3098 path_to_nameidata(&path, nd);
3102 if (should_follow_link(path.dentry, nd->flags & LOOKUP_FOLLOW)) {
3103 if (nd->flags & LOOKUP_RCU) {
3104 if (unlikely(nd->path.mnt != path.mnt ||
3105 unlazy_walk(nd, path.dentry))) {
3110 BUG_ON(inode != path.dentry->d_inode);
3115 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3116 path_to_nameidata(&path, nd);
3121 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3122 path_to_nameidata(&path, nd);
3124 save_parent.dentry = nd->path.dentry;
3125 save_parent.mnt = mntget(path.mnt);
3126 nd->path.dentry = path.dentry;
3130 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3132 error = complete_walk(nd);
3134 path_put(&save_parent);
3137 audit_inode(name, nd->path.dentry, 0);
3139 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3142 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3144 if (!d_is_reg(nd->path.dentry))
3145 will_truncate = false;
3147 if (will_truncate) {
3148 error = mnt_want_write(nd->path.mnt);
3153 finish_open_created:
3154 error = may_open(&nd->path, acc_mode, open_flag);
3158 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3159 error = vfs_open(&nd->path, file, current_cred());
3161 *opened |= FILE_OPENED;
3163 if (error == -EOPENSTALE)
3168 error = open_check_o_direct(file);
3171 error = ima_file_check(file, op->acc_mode, *opened);
3175 if (will_truncate) {
3176 error = handle_truncate(file);
3182 mnt_drop_write(nd->path.mnt);
3183 path_put(&save_parent);
3188 path_put_conditional(&path, nd);
3195 /* If no saved parent or already retried then can't retry */
3196 if (!save_parent.dentry || retried)
3199 BUG_ON(save_parent.dentry != dir);
3200 path_put(&nd->path);
3201 nd->path = save_parent;
3202 nd->inode = dir->d_inode;
3203 save_parent.mnt = NULL;
3204 save_parent.dentry = NULL;
3206 mnt_drop_write(nd->path.mnt);
3213 static int do_tmpfile(int dfd, struct filename *pathname,
3214 struct nameidata *nd, int flags,
3215 const struct open_flags *op,
3216 struct file *file, int *opened)
3218 static const struct qstr name = QSTR_INIT("/", 1);
3219 struct dentry *dentry, *child;
3221 int error = path_lookupat(dfd, pathname,
3222 flags | LOOKUP_DIRECTORY, nd);
3223 if (unlikely(error))
3225 error = mnt_want_write(nd->path.mnt);
3226 if (unlikely(error))
3228 /* we want directory to be writable */
3229 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3232 dentry = nd->path.dentry;
3233 dir = dentry->d_inode;
3234 if (!dir->i_op->tmpfile) {
3235 error = -EOPNOTSUPP;
3238 child = d_alloc(dentry, &name);
3239 if (unlikely(!child)) {
3243 nd->flags &= ~LOOKUP_DIRECTORY;
3244 nd->flags |= op->intent;
3245 dput(nd->path.dentry);
3246 nd->path.dentry = child;
3247 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3250 audit_inode(pathname, nd->path.dentry, 0);
3251 /* Don't check for other permissions, the inode was just created */
3252 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3255 file->f_path.mnt = nd->path.mnt;
3256 error = finish_open(file, nd->path.dentry, NULL, opened);
3259 error = open_check_o_direct(file);
3262 } else if (!(op->open_flag & O_EXCL)) {
3263 struct inode *inode = file_inode(file);
3264 spin_lock(&inode->i_lock);
3265 inode->i_state |= I_LINKABLE;
3266 spin_unlock(&inode->i_lock);
3269 mnt_drop_write(nd->path.mnt);
3271 path_put(&nd->path);
3275 static struct file *path_openat(int dfd, struct filename *pathname,
3276 struct nameidata *nd, const struct open_flags *op, int flags)
3282 file = get_empty_filp();
3286 file->f_flags = op->open_flag;
3288 if (unlikely(file->f_flags & __O_TMPFILE)) {
3289 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3293 error = path_init(dfd, pathname, flags, nd);
3294 if (unlikely(error))
3297 error = do_last(nd, file, op, &opened, pathname);
3298 while (unlikely(error > 0)) { /* trailing symlink */
3299 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3300 error = trailing_symlink(nd);
3301 if (unlikely(error))
3303 error = do_last(nd, file, op, &opened, pathname);
3310 if (!(opened & FILE_OPENED)) {
3314 if (unlikely(error)) {
3315 if (error == -EOPENSTALE) {
3316 if (flags & LOOKUP_RCU)
3321 file = ERR_PTR(error);
3326 struct file *do_filp_open(int dfd, struct filename *pathname,
3327 const struct open_flags *op)
3329 struct nameidata nd;
3330 int flags = op->lookup_flags;
3334 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3335 if (unlikely(filp == ERR_PTR(-ECHILD)))
3336 filp = path_openat(dfd, pathname, &nd, op, flags);
3337 if (unlikely(filp == ERR_PTR(-ESTALE)))
3338 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3339 restore_nameidata(&nd);
3343 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3344 const char *name, const struct open_flags *op)
3346 struct nameidata nd;
3348 struct filename *filename;
3349 int flags = op->lookup_flags | LOOKUP_ROOT;
3352 nd.root.dentry = dentry;
3355 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3356 return ERR_PTR(-ELOOP);
3358 filename = getname_kernel(name);
3359 if (unlikely(IS_ERR(filename)))
3360 return ERR_CAST(filename);
3362 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3363 if (unlikely(file == ERR_PTR(-ECHILD)))
3364 file = path_openat(-1, filename, &nd, op, flags);
3365 if (unlikely(file == ERR_PTR(-ESTALE)))
3366 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3367 restore_nameidata(&nd);
3372 static struct dentry *filename_create(int dfd, struct filename *name,
3373 struct path *path, unsigned int lookup_flags)
3375 struct dentry *dentry = ERR_PTR(-EEXIST);
3376 struct nameidata nd;
3379 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3382 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3383 * other flags passed in are ignored!
3385 lookup_flags &= LOOKUP_REVAL;
3387 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3389 return ERR_PTR(error);
3392 * Yucky last component or no last component at all?
3393 * (foo/., foo/.., /////)
3395 if (nd.last_type != LAST_NORM)
3397 nd.flags &= ~LOOKUP_PARENT;
3398 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3400 /* don't fail immediately if it's r/o, at least try to report other errors */
3401 err2 = mnt_want_write(nd.path.mnt);
3403 * Do the final lookup.
3405 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3406 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3411 if (d_is_positive(dentry))
3415 * Special case - lookup gave negative, but... we had foo/bar/
3416 * From the vfs_mknod() POV we just have a negative dentry -
3417 * all is fine. Let's be bastards - you had / on the end, you've
3418 * been asking for (non-existent) directory. -ENOENT for you.
3420 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3424 if (unlikely(err2)) {
3432 dentry = ERR_PTR(error);
3434 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3436 mnt_drop_write(nd.path.mnt);
3442 struct dentry *kern_path_create(int dfd, const char *pathname,
3443 struct path *path, unsigned int lookup_flags)
3445 struct filename *filename = getname_kernel(pathname);
3448 if (IS_ERR(filename))
3449 return ERR_CAST(filename);
3450 res = filename_create(dfd, filename, path, lookup_flags);
3454 EXPORT_SYMBOL(kern_path_create);
3456 void done_path_create(struct path *path, struct dentry *dentry)
3459 mutex_unlock(&path->dentry->d_inode->i_mutex);
3460 mnt_drop_write(path->mnt);
3463 EXPORT_SYMBOL(done_path_create);
3465 struct dentry *user_path_create(int dfd, const char __user *pathname,
3466 struct path *path, unsigned int lookup_flags)
3468 struct filename *tmp = getname(pathname);
3471 return ERR_CAST(tmp);
3472 res = filename_create(dfd, tmp, path, lookup_flags);
3476 EXPORT_SYMBOL(user_path_create);
3478 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3480 int error = may_create(dir, dentry);
3485 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3488 if (!dir->i_op->mknod)
3491 error = devcgroup_inode_mknod(mode, dev);
3495 error = security_inode_mknod(dir, dentry, mode, dev);
3499 error = dir->i_op->mknod(dir, dentry, mode, dev);
3501 fsnotify_create(dir, dentry);
3504 EXPORT_SYMBOL(vfs_mknod);
3506 static int may_mknod(umode_t mode)
3508 switch (mode & S_IFMT) {
3514 case 0: /* zero mode translates to S_IFREG */
3523 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3526 struct dentry *dentry;
3529 unsigned int lookup_flags = 0;
3531 error = may_mknod(mode);
3535 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3537 return PTR_ERR(dentry);
3539 if (!IS_POSIXACL(path.dentry->d_inode))
3540 mode &= ~current_umask();
3541 error = security_path_mknod(&path, dentry, mode, dev);
3544 switch (mode & S_IFMT) {
3545 case 0: case S_IFREG:
3546 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3548 case S_IFCHR: case S_IFBLK:
3549 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3550 new_decode_dev(dev));
3552 case S_IFIFO: case S_IFSOCK:
3553 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3557 done_path_create(&path, dentry);
3558 if (retry_estale(error, lookup_flags)) {
3559 lookup_flags |= LOOKUP_REVAL;
3565 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3567 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3570 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3572 int error = may_create(dir, dentry);
3573 unsigned max_links = dir->i_sb->s_max_links;
3578 if (!dir->i_op->mkdir)
3581 mode &= (S_IRWXUGO|S_ISVTX);
3582 error = security_inode_mkdir(dir, dentry, mode);
3586 if (max_links && dir->i_nlink >= max_links)
3589 error = dir->i_op->mkdir(dir, dentry, mode);
3591 fsnotify_mkdir(dir, dentry);
3594 EXPORT_SYMBOL(vfs_mkdir);
3596 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3598 struct dentry *dentry;
3601 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3604 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3606 return PTR_ERR(dentry);
3608 if (!IS_POSIXACL(path.dentry->d_inode))
3609 mode &= ~current_umask();
3610 error = security_path_mkdir(&path, dentry, mode);
3612 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3613 done_path_create(&path, dentry);
3614 if (retry_estale(error, lookup_flags)) {
3615 lookup_flags |= LOOKUP_REVAL;
3621 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3623 return sys_mkdirat(AT_FDCWD, pathname, mode);
3627 * The dentry_unhash() helper will try to drop the dentry early: we
3628 * should have a usage count of 1 if we're the only user of this
3629 * dentry, and if that is true (possibly after pruning the dcache),
3630 * then we drop the dentry now.
3632 * A low-level filesystem can, if it choses, legally
3635 * if (!d_unhashed(dentry))
3638 * if it cannot handle the case of removing a directory
3639 * that is still in use by something else..
3641 void dentry_unhash(struct dentry *dentry)
3643 shrink_dcache_parent(dentry);
3644 spin_lock(&dentry->d_lock);
3645 if (dentry->d_lockref.count == 1)
3647 spin_unlock(&dentry->d_lock);
3649 EXPORT_SYMBOL(dentry_unhash);
3651 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3653 int error = may_delete(dir, dentry, 1);
3658 if (!dir->i_op->rmdir)
3662 mutex_lock(&dentry->d_inode->i_mutex);
3665 if (is_local_mountpoint(dentry))
3668 error = security_inode_rmdir(dir, dentry);
3672 shrink_dcache_parent(dentry);
3673 error = dir->i_op->rmdir(dir, dentry);
3677 dentry->d_inode->i_flags |= S_DEAD;
3679 detach_mounts(dentry);
3682 mutex_unlock(&dentry->d_inode->i_mutex);
3688 EXPORT_SYMBOL(vfs_rmdir);
3690 static long do_rmdir(int dfd, const char __user *pathname)
3693 struct filename *name;
3694 struct dentry *dentry;
3698 unsigned int lookup_flags = 0;
3700 name = user_path_parent(dfd, pathname,
3701 &path, &last, &type, lookup_flags);
3703 return PTR_ERR(name);
3717 error = mnt_want_write(path.mnt);
3721 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3722 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3723 error = PTR_ERR(dentry);
3726 if (!dentry->d_inode) {
3730 error = security_path_rmdir(&path, dentry);
3733 error = vfs_rmdir(path.dentry->d_inode, dentry);
3737 mutex_unlock(&path.dentry->d_inode->i_mutex);
3738 mnt_drop_write(path.mnt);
3742 if (retry_estale(error, lookup_flags)) {
3743 lookup_flags |= LOOKUP_REVAL;
3749 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3751 return do_rmdir(AT_FDCWD, pathname);
3755 * vfs_unlink - unlink a filesystem object
3756 * @dir: parent directory
3758 * @delegated_inode: returns victim inode, if the inode is delegated.
3760 * The caller must hold dir->i_mutex.
3762 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3763 * return a reference to the inode in delegated_inode. The caller
3764 * should then break the delegation on that inode and retry. Because
3765 * breaking a delegation may take a long time, the caller should drop
3766 * dir->i_mutex before doing so.
3768 * Alternatively, a caller may pass NULL for delegated_inode. This may
3769 * be appropriate for callers that expect the underlying filesystem not
3770 * to be NFS exported.
3772 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3774 struct inode *target = dentry->d_inode;
3775 int error = may_delete(dir, dentry, 0);
3780 if (!dir->i_op->unlink)
3783 mutex_lock(&target->i_mutex);
3784 if (is_local_mountpoint(dentry))
3787 error = security_inode_unlink(dir, dentry);
3789 error = try_break_deleg(target, delegated_inode);
3792 error = dir->i_op->unlink(dir, dentry);
3795 detach_mounts(dentry);
3800 mutex_unlock(&target->i_mutex);
3802 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3803 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3804 fsnotify_link_count(target);
3810 EXPORT_SYMBOL(vfs_unlink);
3813 * Make sure that the actual truncation of the file will occur outside its
3814 * directory's i_mutex. Truncate can take a long time if there is a lot of
3815 * writeout happening, and we don't want to prevent access to the directory
3816 * while waiting on the I/O.
3818 static long do_unlinkat(int dfd, const char __user *pathname)
3821 struct filename *name;
3822 struct dentry *dentry;
3826 struct inode *inode = NULL;
3827 struct inode *delegated_inode = NULL;
3828 unsigned int lookup_flags = 0;
3830 name = user_path_parent(dfd, pathname,
3831 &path, &last, &type, lookup_flags);
3833 return PTR_ERR(name);
3836 if (type != LAST_NORM)
3839 error = mnt_want_write(path.mnt);
3843 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3844 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3845 error = PTR_ERR(dentry);
3846 if (!IS_ERR(dentry)) {
3847 /* Why not before? Because we want correct error value */
3848 if (last.name[last.len])
3850 inode = dentry->d_inode;
3851 if (d_is_negative(dentry))
3854 error = security_path_unlink(&path, dentry);
3857 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3861 mutex_unlock(&path.dentry->d_inode->i_mutex);
3863 iput(inode); /* truncate the inode here */
3865 if (delegated_inode) {
3866 error = break_deleg_wait(&delegated_inode);
3870 mnt_drop_write(path.mnt);
3874 if (retry_estale(error, lookup_flags)) {
3875 lookup_flags |= LOOKUP_REVAL;
3882 if (d_is_negative(dentry))
3884 else if (d_is_dir(dentry))
3891 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3893 if ((flag & ~AT_REMOVEDIR) != 0)
3896 if (flag & AT_REMOVEDIR)
3897 return do_rmdir(dfd, pathname);
3899 return do_unlinkat(dfd, pathname);
3902 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3904 return do_unlinkat(AT_FDCWD, pathname);
3907 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3909 int error = may_create(dir, dentry);
3914 if (!dir->i_op->symlink)
3917 error = security_inode_symlink(dir, dentry, oldname);
3921 error = dir->i_op->symlink(dir, dentry, oldname);
3923 fsnotify_create(dir, dentry);
3926 EXPORT_SYMBOL(vfs_symlink);
3928 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3929 int, newdfd, const char __user *, newname)
3932 struct filename *from;
3933 struct dentry *dentry;
3935 unsigned int lookup_flags = 0;
3937 from = getname(oldname);
3939 return PTR_ERR(from);
3941 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3942 error = PTR_ERR(dentry);
3946 error = security_path_symlink(&path, dentry, from->name);
3948 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3949 done_path_create(&path, dentry);
3950 if (retry_estale(error, lookup_flags)) {
3951 lookup_flags |= LOOKUP_REVAL;
3959 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3961 return sys_symlinkat(oldname, AT_FDCWD, newname);
3965 * vfs_link - create a new link
3966 * @old_dentry: object to be linked
3968 * @new_dentry: where to create the new link
3969 * @delegated_inode: returns inode needing a delegation break
3971 * The caller must hold dir->i_mutex
3973 * If vfs_link discovers a delegation on the to-be-linked file in need
3974 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3975 * inode in delegated_inode. The caller should then break the delegation
3976 * and retry. Because breaking a delegation may take a long time, the
3977 * caller should drop the i_mutex before doing so.
3979 * Alternatively, a caller may pass NULL for delegated_inode. This may
3980 * be appropriate for callers that expect the underlying filesystem not
3981 * to be NFS exported.
3983 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3985 struct inode *inode = old_dentry->d_inode;
3986 unsigned max_links = dir->i_sb->s_max_links;
3992 error = may_create(dir, new_dentry);
3996 if (dir->i_sb != inode->i_sb)
4000 * A link to an append-only or immutable file cannot be created.
4002 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4004 if (!dir->i_op->link)
4006 if (S_ISDIR(inode->i_mode))
4009 error = security_inode_link(old_dentry, dir, new_dentry);
4013 mutex_lock(&inode->i_mutex);
4014 /* Make sure we don't allow creating hardlink to an unlinked file */
4015 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4017 else if (max_links && inode->i_nlink >= max_links)
4020 error = try_break_deleg(inode, delegated_inode);
4022 error = dir->i_op->link(old_dentry, dir, new_dentry);
4025 if (!error && (inode->i_state & I_LINKABLE)) {
4026 spin_lock(&inode->i_lock);
4027 inode->i_state &= ~I_LINKABLE;
4028 spin_unlock(&inode->i_lock);
4030 mutex_unlock(&inode->i_mutex);
4032 fsnotify_link(dir, inode, new_dentry);
4035 EXPORT_SYMBOL(vfs_link);
4038 * Hardlinks are often used in delicate situations. We avoid
4039 * security-related surprises by not following symlinks on the
4042 * We don't follow them on the oldname either to be compatible
4043 * with linux 2.0, and to avoid hard-linking to directories
4044 * and other special files. --ADM
4046 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4047 int, newdfd, const char __user *, newname, int, flags)
4049 struct dentry *new_dentry;
4050 struct path old_path, new_path;
4051 struct inode *delegated_inode = NULL;
4055 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4058 * To use null names we require CAP_DAC_READ_SEARCH
4059 * This ensures that not everyone will be able to create
4060 * handlink using the passed filedescriptor.
4062 if (flags & AT_EMPTY_PATH) {
4063 if (!capable(CAP_DAC_READ_SEARCH))
4068 if (flags & AT_SYMLINK_FOLLOW)
4069 how |= LOOKUP_FOLLOW;
4071 error = user_path_at(olddfd, oldname, how, &old_path);
4075 new_dentry = user_path_create(newdfd, newname, &new_path,
4076 (how & LOOKUP_REVAL));
4077 error = PTR_ERR(new_dentry);
4078 if (IS_ERR(new_dentry))
4082 if (old_path.mnt != new_path.mnt)
4084 error = may_linkat(&old_path);
4085 if (unlikely(error))
4087 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4090 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4092 done_path_create(&new_path, new_dentry);
4093 if (delegated_inode) {
4094 error = break_deleg_wait(&delegated_inode);
4096 path_put(&old_path);
4100 if (retry_estale(error, how)) {
4101 path_put(&old_path);
4102 how |= LOOKUP_REVAL;
4106 path_put(&old_path);
4111 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4113 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4117 * vfs_rename - rename a filesystem object
4118 * @old_dir: parent of source
4119 * @old_dentry: source
4120 * @new_dir: parent of destination
4121 * @new_dentry: destination
4122 * @delegated_inode: returns an inode needing a delegation break
4123 * @flags: rename flags
4125 * The caller must hold multiple mutexes--see lock_rename()).
4127 * If vfs_rename discovers a delegation in need of breaking at either
4128 * the source or destination, it will return -EWOULDBLOCK and return a
4129 * reference to the inode in delegated_inode. The caller should then
4130 * break the delegation and retry. Because breaking a delegation may
4131 * take a long time, the caller should drop all locks before doing
4134 * Alternatively, a caller may pass NULL for delegated_inode. This may
4135 * be appropriate for callers that expect the underlying filesystem not
4136 * to be NFS exported.
4138 * The worst of all namespace operations - renaming directory. "Perverted"
4139 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4141 * a) we can get into loop creation.
4142 * b) race potential - two innocent renames can create a loop together.
4143 * That's where 4.4 screws up. Current fix: serialization on
4144 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4146 * c) we have to lock _four_ objects - parents and victim (if it exists),
4147 * and source (if it is not a directory).
4148 * And that - after we got ->i_mutex on parents (until then we don't know
4149 * whether the target exists). Solution: try to be smart with locking
4150 * order for inodes. We rely on the fact that tree topology may change
4151 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4152 * move will be locked. Thus we can rank directories by the tree
4153 * (ancestors first) and rank all non-directories after them.
4154 * That works since everybody except rename does "lock parent, lookup,
4155 * lock child" and rename is under ->s_vfs_rename_mutex.
4156 * HOWEVER, it relies on the assumption that any object with ->lookup()
4157 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4158 * we'd better make sure that there's no link(2) for them.
4159 * d) conversion from fhandle to dentry may come in the wrong moment - when
4160 * we are removing the target. Solution: we will have to grab ->i_mutex
4161 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4162 * ->i_mutex on parents, which works but leads to some truly excessive
4165 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4166 struct inode *new_dir, struct dentry *new_dentry,
4167 struct inode **delegated_inode, unsigned int flags)
4170 bool is_dir = d_is_dir(old_dentry);
4171 const unsigned char *old_name;
4172 struct inode *source = old_dentry->d_inode;
4173 struct inode *target = new_dentry->d_inode;
4174 bool new_is_dir = false;
4175 unsigned max_links = new_dir->i_sb->s_max_links;
4177 if (source == target)
4180 error = may_delete(old_dir, old_dentry, is_dir);
4185 error = may_create(new_dir, new_dentry);
4187 new_is_dir = d_is_dir(new_dentry);
4189 if (!(flags & RENAME_EXCHANGE))
4190 error = may_delete(new_dir, new_dentry, is_dir);
4192 error = may_delete(new_dir, new_dentry, new_is_dir);
4197 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4200 if (flags && !old_dir->i_op->rename2)
4204 * If we are going to change the parent - check write permissions,
4205 * we'll need to flip '..'.
4207 if (new_dir != old_dir) {
4209 error = inode_permission(source, MAY_WRITE);
4213 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4214 error = inode_permission(target, MAY_WRITE);
4220 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4225 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4227 if (!is_dir || (flags & RENAME_EXCHANGE))
4228 lock_two_nondirectories(source, target);
4230 mutex_lock(&target->i_mutex);
4233 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4236 if (max_links && new_dir != old_dir) {
4238 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4240 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4241 old_dir->i_nlink >= max_links)
4244 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4245 shrink_dcache_parent(new_dentry);
4247 error = try_break_deleg(source, delegated_inode);
4251 if (target && !new_is_dir) {
4252 error = try_break_deleg(target, delegated_inode);
4256 if (!old_dir->i_op->rename2) {
4257 error = old_dir->i_op->rename(old_dir, old_dentry,
4258 new_dir, new_dentry);
4260 WARN_ON(old_dir->i_op->rename != NULL);
4261 error = old_dir->i_op->rename2(old_dir, old_dentry,
4262 new_dir, new_dentry, flags);
4267 if (!(flags & RENAME_EXCHANGE) && target) {
4269 target->i_flags |= S_DEAD;
4270 dont_mount(new_dentry);
4271 detach_mounts(new_dentry);
4273 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4274 if (!(flags & RENAME_EXCHANGE))
4275 d_move(old_dentry, new_dentry);
4277 d_exchange(old_dentry, new_dentry);
4280 if (!is_dir || (flags & RENAME_EXCHANGE))
4281 unlock_two_nondirectories(source, target);
4283 mutex_unlock(&target->i_mutex);
4286 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4287 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4288 if (flags & RENAME_EXCHANGE) {
4289 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4290 new_is_dir, NULL, new_dentry);
4293 fsnotify_oldname_free(old_name);
4297 EXPORT_SYMBOL(vfs_rename);
4299 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4300 int, newdfd, const char __user *, newname, unsigned int, flags)
4302 struct dentry *old_dentry, *new_dentry;
4303 struct dentry *trap;
4304 struct path old_path, new_path;
4305 struct qstr old_last, new_last;
4306 int old_type, new_type;
4307 struct inode *delegated_inode = NULL;
4308 struct filename *from;
4309 struct filename *to;
4310 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4311 bool should_retry = false;
4314 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4317 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4318 (flags & RENAME_EXCHANGE))
4321 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4324 if (flags & RENAME_EXCHANGE)
4328 from = user_path_parent(olddfd, oldname,
4329 &old_path, &old_last, &old_type, lookup_flags);
4331 error = PTR_ERR(from);
4335 to = user_path_parent(newdfd, newname,
4336 &new_path, &new_last, &new_type, lookup_flags);
4338 error = PTR_ERR(to);
4343 if (old_path.mnt != new_path.mnt)
4347 if (old_type != LAST_NORM)
4350 if (flags & RENAME_NOREPLACE)
4352 if (new_type != LAST_NORM)
4355 error = mnt_want_write(old_path.mnt);
4360 trap = lock_rename(new_path.dentry, old_path.dentry);
4362 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4363 error = PTR_ERR(old_dentry);
4364 if (IS_ERR(old_dentry))
4366 /* source must exist */
4368 if (d_is_negative(old_dentry))
4370 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4371 error = PTR_ERR(new_dentry);
4372 if (IS_ERR(new_dentry))
4375 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4377 if (flags & RENAME_EXCHANGE) {
4379 if (d_is_negative(new_dentry))
4382 if (!d_is_dir(new_dentry)) {
4384 if (new_last.name[new_last.len])
4388 /* unless the source is a directory trailing slashes give -ENOTDIR */
4389 if (!d_is_dir(old_dentry)) {
4391 if (old_last.name[old_last.len])
4393 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4396 /* source should not be ancestor of target */
4398 if (old_dentry == trap)
4400 /* target should not be an ancestor of source */
4401 if (!(flags & RENAME_EXCHANGE))
4403 if (new_dentry == trap)
4406 error = security_path_rename(&old_path, old_dentry,
4407 &new_path, new_dentry, flags);
4410 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4411 new_path.dentry->d_inode, new_dentry,
4412 &delegated_inode, flags);
4418 unlock_rename(new_path.dentry, old_path.dentry);
4419 if (delegated_inode) {
4420 error = break_deleg_wait(&delegated_inode);
4424 mnt_drop_write(old_path.mnt);
4426 if (retry_estale(error, lookup_flags))
4427 should_retry = true;
4428 path_put(&new_path);
4431 path_put(&old_path);
4434 should_retry = false;
4435 lookup_flags |= LOOKUP_REVAL;
4442 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4443 int, newdfd, const char __user *, newname)
4445 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4448 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4450 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4453 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4455 int error = may_create(dir, dentry);
4459 if (!dir->i_op->mknod)
4462 return dir->i_op->mknod(dir, dentry,
4463 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4465 EXPORT_SYMBOL(vfs_whiteout);
4467 int readlink_copy(char __user *buffer, int buflen, const char *link)
4469 int len = PTR_ERR(link);
4474 if (len > (unsigned) buflen)
4476 if (copy_to_user(buffer, link, len))
4481 EXPORT_SYMBOL(readlink_copy);
4484 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4485 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4486 * using) it for any given inode is up to filesystem.
4488 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4491 const char *link = dentry->d_inode->i_link;
4495 link = dentry->d_inode->i_op->follow_link(dentry, &cookie, NULL);
4497 return PTR_ERR(link);
4499 res = readlink_copy(buffer, buflen, link);
4500 if (cookie && dentry->d_inode->i_op->put_link)
4501 dentry->d_inode->i_op->put_link(dentry, cookie);
4504 EXPORT_SYMBOL(generic_readlink);
4506 /* get the link contents into pagecache */
4507 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4511 struct address_space *mapping = dentry->d_inode->i_mapping;
4512 page = read_mapping_page(mapping, 0, NULL);
4517 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4521 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4523 struct page *page = NULL;
4524 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4527 page_cache_release(page);
4531 EXPORT_SYMBOL(page_readlink);
4533 const char *page_follow_link_light(struct dentry *dentry, void **cookie, struct nameidata *nd)
4535 struct page *page = NULL;
4536 char *res = page_getlink(dentry, &page);
4541 EXPORT_SYMBOL(page_follow_link_light);
4543 void page_put_link(struct dentry *dentry, void *cookie)
4545 struct page *page = cookie;
4547 page_cache_release(page);
4549 EXPORT_SYMBOL(page_put_link);
4552 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4554 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4556 struct address_space *mapping = inode->i_mapping;
4561 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4563 flags |= AOP_FLAG_NOFS;
4566 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4567 flags, &page, &fsdata);
4571 kaddr = kmap_atomic(page);
4572 memcpy(kaddr, symname, len-1);
4573 kunmap_atomic(kaddr);
4575 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4582 mark_inode_dirty(inode);
4587 EXPORT_SYMBOL(__page_symlink);
4589 int page_symlink(struct inode *inode, const char *symname, int len)
4591 return __page_symlink(inode, symname, len,
4592 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4594 EXPORT_SYMBOL(page_symlink);
4596 const struct inode_operations page_symlink_inode_operations = {
4597 .readlink = generic_readlink,
4598 .follow_link = page_follow_link_light,
4599 .put_link = page_put_link,
4601 EXPORT_SYMBOL(page_symlink_inode_operations);