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 vfsmount *mnt, struct inode *inode, int mask)
378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379 if (likely(mnt && inode->i_op->permission2))
380 return inode->i_op->permission2(mnt, inode, mask);
381 if (likely(inode->i_op->permission))
382 return inode->i_op->permission(inode, mask);
384 /* This gets set once for the inode lifetime */
385 spin_lock(&inode->i_lock);
386 inode->i_opflags |= IOP_FASTPERM;
387 spin_unlock(&inode->i_lock);
389 return generic_permission(inode, mask);
393 * __inode_permission - Check for access rights to a given inode
394 * @inode: Inode to check permission on
395 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
397 * Check for read/write/execute permissions on an inode.
399 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
401 * This does not check for a read-only file system. You probably want
402 * inode_permission().
404 int __inode_permission2(struct vfsmount *mnt, struct inode *inode, int mask)
408 if (unlikely(mask & MAY_WRITE)) {
410 * Nobody gets write access to an immutable file.
412 if (IS_IMMUTABLE(inode))
416 retval = do_inode_permission(mnt, inode, mask);
420 retval = devcgroup_inode_permission(inode, mask);
424 retval = security_inode_permission(inode, mask);
427 EXPORT_SYMBOL(__inode_permission2);
429 int __inode_permission(struct inode *inode, int mask)
431 return __inode_permission2(NULL, inode, mask);
433 EXPORT_SYMBOL(__inode_permission);
436 * sb_permission - Check superblock-level permissions
437 * @sb: Superblock of inode to check permission on
438 * @inode: Inode to check permission on
439 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
441 * Separate out file-system wide checks from inode-specific permission checks.
443 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
445 if (unlikely(mask & MAY_WRITE)) {
446 umode_t mode = inode->i_mode;
448 /* Nobody gets write access to a read-only fs. */
449 if ((sb->s_flags & MS_RDONLY) &&
450 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
457 * inode_permission - Check for access rights to a given inode
458 * @inode: Inode to check permission on
459 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
461 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
462 * this, letting us set arbitrary permissions for filesystem access without
463 * changing the "normal" UIDs which are used for other things.
465 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
467 int inode_permission2(struct vfsmount *mnt, struct inode *inode, int mask)
471 retval = sb_permission(inode->i_sb, inode, mask);
474 return __inode_permission2(mnt, inode, mask);
476 EXPORT_SYMBOL(inode_permission2);
478 int inode_permission(struct inode *inode, int mask)
480 return inode_permission2(NULL, inode, mask);
482 EXPORT_SYMBOL(inode_permission);
485 * path_get - get a reference to a path
486 * @path: path to get the reference to
488 * Given a path increment the reference count to the dentry and the vfsmount.
490 void path_get(const struct path *path)
495 EXPORT_SYMBOL(path_get);
498 * path_put - put a reference to a path
499 * @path: path to put the reference to
501 * Given a path decrement the reference count to the dentry and the vfsmount.
503 void path_put(const struct path *path)
508 EXPORT_SYMBOL(path_put);
510 #define EMBEDDED_LEVELS 2
515 struct inode *inode; /* path.dentry.d_inode */
520 int total_link_count;
527 } *stack, internal[EMBEDDED_LEVELS];
528 struct filename *name;
529 struct nameidata *saved;
534 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
536 struct nameidata *old = current->nameidata;
537 p->stack = p->internal;
540 p->total_link_count = old ? old->total_link_count : 0;
542 current->nameidata = p;
545 static void restore_nameidata(void)
547 struct nameidata *now = current->nameidata, *old = now->saved;
549 current->nameidata = old;
551 old->total_link_count = now->total_link_count;
552 if (now->stack != now->internal) {
554 now->stack = now->internal;
558 static int __nd_alloc_stack(struct nameidata *nd)
562 if (nd->flags & LOOKUP_RCU) {
563 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
568 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
573 memcpy(p, nd->internal, sizeof(nd->internal));
579 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
580 * @path: nameidate to verify
582 * Rename can sometimes move a file or directory outside of a bind
583 * mount, path_connected allows those cases to be detected.
585 static bool path_connected(const struct path *path)
587 struct vfsmount *mnt = path->mnt;
589 /* Only bind mounts can have disconnected paths */
590 if (mnt->mnt_root == mnt->mnt_sb->s_root)
593 return is_subdir(path->dentry, mnt->mnt_root);
596 static inline int nd_alloc_stack(struct nameidata *nd)
598 if (likely(nd->depth != EMBEDDED_LEVELS))
600 if (likely(nd->stack != nd->internal))
602 return __nd_alloc_stack(nd);
605 static void drop_links(struct nameidata *nd)
609 struct saved *last = nd->stack + i;
610 struct inode *inode = last->inode;
611 if (last->cookie && inode->i_op->put_link) {
612 inode->i_op->put_link(inode, last->cookie);
618 static void terminate_walk(struct nameidata *nd)
621 if (!(nd->flags & LOOKUP_RCU)) {
624 for (i = 0; i < nd->depth; i++)
625 path_put(&nd->stack[i].link);
626 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
631 nd->flags &= ~LOOKUP_RCU;
632 if (!(nd->flags & LOOKUP_ROOT))
639 /* path_put is needed afterwards regardless of success or failure */
640 static bool legitimize_path(struct nameidata *nd,
641 struct path *path, unsigned seq)
643 int res = __legitimize_mnt(path->mnt, nd->m_seq);
650 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
654 return !read_seqcount_retry(&path->dentry->d_seq, seq);
657 static bool legitimize_links(struct nameidata *nd)
660 for (i = 0; i < nd->depth; i++) {
661 struct saved *last = nd->stack + i;
662 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
672 * Path walking has 2 modes, rcu-walk and ref-walk (see
673 * Documentation/filesystems/path-lookup.txt). In situations when we can't
674 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
675 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
676 * mode. Refcounts are grabbed at the last known good point before rcu-walk
677 * got stuck, so ref-walk may continue from there. If this is not successful
678 * (eg. a seqcount has changed), then failure is returned and it's up to caller
679 * to restart the path walk from the beginning in ref-walk mode.
683 * unlazy_walk - try to switch to ref-walk mode.
684 * @nd: nameidata pathwalk data
685 * @dentry: child of nd->path.dentry or NULL
686 * @seq: seq number to check dentry against
687 * Returns: 0 on success, -ECHILD on failure
689 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
690 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
691 * @nd or NULL. Must be called from rcu-walk context.
692 * Nothing should touch nameidata between unlazy_walk() failure and
695 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
697 struct dentry *parent = nd->path.dentry;
699 BUG_ON(!(nd->flags & LOOKUP_RCU));
701 nd->flags &= ~LOOKUP_RCU;
702 if (unlikely(!legitimize_links(nd)))
704 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
706 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
710 * For a negative lookup, the lookup sequence point is the parents
711 * sequence point, and it only needs to revalidate the parent dentry.
713 * For a positive lookup, we need to move both the parent and the
714 * dentry from the RCU domain to be properly refcounted. And the
715 * sequence number in the dentry validates *both* dentry counters,
716 * since we checked the sequence number of the parent after we got
717 * the child sequence number. So we know the parent must still
718 * be valid if the child sequence number is still valid.
721 if (read_seqcount_retry(&parent->d_seq, nd->seq))
723 BUG_ON(nd->inode != parent->d_inode);
725 if (!lockref_get_not_dead(&dentry->d_lockref))
727 if (read_seqcount_retry(&dentry->d_seq, seq))
732 * Sequence counts matched. Now make sure that the root is
733 * still valid and get it if required.
735 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
736 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
753 nd->path.dentry = NULL;
757 if (!(nd->flags & LOOKUP_ROOT))
762 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
764 if (unlikely(!legitimize_path(nd, link, seq))) {
767 nd->flags &= ~LOOKUP_RCU;
769 nd->path.dentry = NULL;
770 if (!(nd->flags & LOOKUP_ROOT))
773 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
780 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
782 return dentry->d_op->d_revalidate(dentry, flags);
786 * complete_walk - successful completion of path walk
787 * @nd: pointer nameidata
789 * If we had been in RCU mode, drop out of it and legitimize nd->path.
790 * Revalidate the final result, unless we'd already done that during
791 * the path walk or the filesystem doesn't ask for it. Return 0 on
792 * success, -error on failure. In case of failure caller does not
793 * need to drop nd->path.
795 static int complete_walk(struct nameidata *nd)
797 struct dentry *dentry = nd->path.dentry;
800 if (nd->flags & LOOKUP_RCU) {
801 if (!(nd->flags & LOOKUP_ROOT))
803 if (unlikely(unlazy_walk(nd, NULL, 0)))
807 if (likely(!(nd->flags & LOOKUP_JUMPED)))
810 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
813 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
823 static void set_root(struct nameidata *nd)
825 get_fs_root(current->fs, &nd->root);
828 static void set_root_rcu(struct nameidata *nd)
830 struct fs_struct *fs = current->fs;
834 seq = read_seqcount_begin(&fs->seq);
836 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
837 } while (read_seqcount_retry(&fs->seq, seq));
840 static void path_put_conditional(struct path *path, struct nameidata *nd)
843 if (path->mnt != nd->path.mnt)
847 static inline void path_to_nameidata(const struct path *path,
848 struct nameidata *nd)
850 if (!(nd->flags & LOOKUP_RCU)) {
851 dput(nd->path.dentry);
852 if (nd->path.mnt != path->mnt)
853 mntput(nd->path.mnt);
855 nd->path.mnt = path->mnt;
856 nd->path.dentry = path->dentry;
860 * Helper to directly jump to a known parsed path from ->follow_link,
861 * caller must have taken a reference to path beforehand.
863 void nd_jump_link(struct path *path)
865 struct nameidata *nd = current->nameidata;
869 nd->inode = nd->path.dentry->d_inode;
870 nd->flags |= LOOKUP_JUMPED;
873 static inline void put_link(struct nameidata *nd)
875 struct saved *last = nd->stack + --nd->depth;
876 struct inode *inode = last->inode;
877 if (last->cookie && inode->i_op->put_link)
878 inode->i_op->put_link(inode, last->cookie);
879 if (!(nd->flags & LOOKUP_RCU))
880 path_put(&last->link);
883 int sysctl_protected_symlinks __read_mostly = 0;
884 int sysctl_protected_hardlinks __read_mostly = 0;
887 * may_follow_link - Check symlink following for unsafe situations
888 * @nd: nameidata pathwalk data
890 * In the case of the sysctl_protected_symlinks sysctl being enabled,
891 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
892 * in a sticky world-writable directory. This is to protect privileged
893 * processes from failing races against path names that may change out
894 * from under them by way of other users creating malicious symlinks.
895 * It will permit symlinks to be followed only when outside a sticky
896 * world-writable directory, or when the uid of the symlink and follower
897 * match, or when the directory owner matches the symlink's owner.
899 * Returns 0 if following the symlink is allowed, -ve on error.
901 static inline int may_follow_link(struct nameidata *nd)
903 const struct inode *inode;
904 const struct inode *parent;
907 if (!sysctl_protected_symlinks)
910 /* Allowed if owner and follower match. */
911 inode = nd->stack[0].inode;
912 if (uid_eq(current_cred()->fsuid, inode->i_uid))
915 /* Allowed if parent directory not sticky and world-writable. */
917 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
920 /* Allowed if parent directory and link owner match. */
921 puid = parent->i_uid;
922 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
925 if (nd->flags & LOOKUP_RCU)
928 audit_log_link_denied("follow_link", &nd->stack[0].link);
933 * safe_hardlink_source - Check for safe hardlink conditions
934 * @inode: the source inode to hardlink from
936 * Return false if at least one of the following conditions:
937 * - inode is not a regular file
939 * - inode is setgid and group-exec
940 * - access failure for read and write
942 * Otherwise returns true.
944 static bool safe_hardlink_source(struct inode *inode)
946 umode_t mode = inode->i_mode;
948 /* Special files should not get pinned to the filesystem. */
952 /* Setuid files should not get pinned to the filesystem. */
956 /* Executable setgid files should not get pinned to the filesystem. */
957 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
960 /* Hardlinking to unreadable or unwritable sources is dangerous. */
961 if (inode_permission(inode, MAY_READ | MAY_WRITE))
968 * may_linkat - Check permissions for creating a hardlink
969 * @link: the source to hardlink from
971 * Block hardlink when all of:
972 * - sysctl_protected_hardlinks enabled
973 * - fsuid does not match inode
974 * - hardlink source is unsafe (see safe_hardlink_source() above)
975 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
977 * Returns 0 if successful, -ve on error.
979 static int may_linkat(struct path *link)
983 if (!sysctl_protected_hardlinks)
986 inode = link->dentry->d_inode;
988 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
989 * otherwise, it must be a safe source.
991 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
994 audit_log_link_denied("linkat", link);
998 static __always_inline
999 const char *get_link(struct nameidata *nd)
1001 struct saved *last = nd->stack + nd->depth - 1;
1002 struct dentry *dentry = last->link.dentry;
1003 struct inode *inode = last->inode;
1007 if (!(nd->flags & LOOKUP_RCU)) {
1008 touch_atime(&last->link);
1010 } else if (atime_needs_update(&last->link, inode)) {
1011 if (unlikely(unlazy_walk(nd, NULL, 0)))
1012 return ERR_PTR(-ECHILD);
1013 touch_atime(&last->link);
1016 error = security_inode_follow_link(dentry, inode,
1017 nd->flags & LOOKUP_RCU);
1018 if (unlikely(error))
1019 return ERR_PTR(error);
1021 nd->last_type = LAST_BIND;
1022 res = inode->i_link;
1024 if (nd->flags & LOOKUP_RCU) {
1025 if (unlikely(unlazy_walk(nd, NULL, 0)))
1026 return ERR_PTR(-ECHILD);
1028 res = inode->i_op->follow_link(dentry, &last->cookie);
1029 if (IS_ERR_OR_NULL(res)) {
1030 last->cookie = NULL;
1035 if (nd->flags & LOOKUP_RCU) {
1039 nd->path = nd->root;
1040 d = nd->path.dentry;
1041 nd->inode = d->d_inode;
1042 nd->seq = nd->root_seq;
1043 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
1044 return ERR_PTR(-ECHILD);
1048 path_put(&nd->path);
1049 nd->path = nd->root;
1050 path_get(&nd->root);
1051 nd->inode = nd->path.dentry->d_inode;
1053 nd->flags |= LOOKUP_JUMPED;
1054 while (unlikely(*++res == '/'))
1063 * follow_up - Find the mountpoint of path's vfsmount
1065 * Given a path, find the mountpoint of its source file system.
1066 * Replace @path with the path of the mountpoint in the parent mount.
1069 * Return 1 if we went up a level and 0 if we were already at the
1072 int follow_up(struct path *path)
1074 struct mount *mnt = real_mount(path->mnt);
1075 struct mount *parent;
1076 struct dentry *mountpoint;
1078 read_seqlock_excl(&mount_lock);
1079 parent = mnt->mnt_parent;
1080 if (parent == mnt) {
1081 read_sequnlock_excl(&mount_lock);
1084 mntget(&parent->mnt);
1085 mountpoint = dget(mnt->mnt_mountpoint);
1086 read_sequnlock_excl(&mount_lock);
1088 path->dentry = mountpoint;
1090 path->mnt = &parent->mnt;
1093 EXPORT_SYMBOL(follow_up);
1096 * Perform an automount
1097 * - return -EISDIR to tell follow_managed() to stop and return the path we
1100 static int follow_automount(struct path *path, struct nameidata *nd,
1103 struct vfsmount *mnt;
1106 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1109 /* We don't want to mount if someone's just doing a stat -
1110 * unless they're stat'ing a directory and appended a '/' to
1113 * We do, however, want to mount if someone wants to open or
1114 * create a file of any type under the mountpoint, wants to
1115 * traverse through the mountpoint or wants to open the
1116 * mounted directory. Also, autofs may mark negative dentries
1117 * as being automount points. These will need the attentions
1118 * of the daemon to instantiate them before they can be used.
1120 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1121 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1122 path->dentry->d_inode)
1125 nd->total_link_count++;
1126 if (nd->total_link_count >= 40)
1129 mnt = path->dentry->d_op->d_automount(path);
1132 * The filesystem is allowed to return -EISDIR here to indicate
1133 * it doesn't want to automount. For instance, autofs would do
1134 * this so that its userspace daemon can mount on this dentry.
1136 * However, we can only permit this if it's a terminal point in
1137 * the path being looked up; if it wasn't then the remainder of
1138 * the path is inaccessible and we should say so.
1140 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1142 return PTR_ERR(mnt);
1145 if (!mnt) /* mount collision */
1148 if (!*need_mntput) {
1149 /* lock_mount() may release path->mnt on error */
1151 *need_mntput = true;
1153 err = finish_automount(mnt, path);
1157 /* Someone else made a mount here whilst we were busy */
1162 path->dentry = dget(mnt->mnt_root);
1171 * Handle a dentry that is managed in some way.
1172 * - Flagged for transit management (autofs)
1173 * - Flagged as mountpoint
1174 * - Flagged as automount point
1176 * This may only be called in refwalk mode.
1178 * Serialization is taken care of in namespace.c
1180 static int follow_managed(struct path *path, struct nameidata *nd)
1182 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1184 bool need_mntput = false;
1187 /* Given that we're not holding a lock here, we retain the value in a
1188 * local variable for each dentry as we look at it so that we don't see
1189 * the components of that value change under us */
1190 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1191 managed &= DCACHE_MANAGED_DENTRY,
1192 unlikely(managed != 0)) {
1193 /* Allow the filesystem to manage the transit without i_mutex
1195 if (managed & DCACHE_MANAGE_TRANSIT) {
1196 BUG_ON(!path->dentry->d_op);
1197 BUG_ON(!path->dentry->d_op->d_manage);
1198 ret = path->dentry->d_op->d_manage(path->dentry, false);
1203 /* Transit to a mounted filesystem. */
1204 if (managed & DCACHE_MOUNTED) {
1205 struct vfsmount *mounted = lookup_mnt(path);
1210 path->mnt = mounted;
1211 path->dentry = dget(mounted->mnt_root);
1216 /* Something is mounted on this dentry in another
1217 * namespace and/or whatever was mounted there in this
1218 * namespace got unmounted before lookup_mnt() could
1222 /* Handle an automount point */
1223 if (managed & DCACHE_NEED_AUTOMOUNT) {
1224 ret = follow_automount(path, nd, &need_mntput);
1230 /* We didn't change the current path point */
1234 if (need_mntput && path->mnt == mnt)
1239 nd->flags |= LOOKUP_JUMPED;
1240 if (unlikely(ret < 0))
1241 path_put_conditional(path, nd);
1245 int follow_down_one(struct path *path)
1247 struct vfsmount *mounted;
1249 mounted = lookup_mnt(path);
1253 path->mnt = mounted;
1254 path->dentry = dget(mounted->mnt_root);
1259 EXPORT_SYMBOL(follow_down_one);
1261 static inline int managed_dentry_rcu(struct dentry *dentry)
1263 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1264 dentry->d_op->d_manage(dentry, true) : 0;
1268 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1269 * we meet a managed dentry that would need blocking.
1271 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1272 struct inode **inode, unsigned *seqp)
1275 struct mount *mounted;
1277 * Don't forget we might have a non-mountpoint managed dentry
1278 * that wants to block transit.
1280 switch (managed_dentry_rcu(path->dentry)) {
1290 if (!d_mountpoint(path->dentry))
1291 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1293 mounted = __lookup_mnt(path->mnt, path->dentry);
1296 path->mnt = &mounted->mnt;
1297 path->dentry = mounted->mnt.mnt_root;
1298 nd->flags |= LOOKUP_JUMPED;
1299 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1301 * Update the inode too. We don't need to re-check the
1302 * dentry sequence number here after this d_inode read,
1303 * because a mount-point is always pinned.
1305 *inode = path->dentry->d_inode;
1307 return !read_seqretry(&mount_lock, nd->m_seq) &&
1308 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1311 static int follow_dotdot_rcu(struct nameidata *nd)
1313 struct inode *inode = nd->inode;
1318 if (path_equal(&nd->path, &nd->root))
1320 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1321 struct dentry *old = nd->path.dentry;
1322 struct dentry *parent = old->d_parent;
1325 inode = parent->d_inode;
1326 seq = read_seqcount_begin(&parent->d_seq);
1327 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1329 nd->path.dentry = parent;
1331 if (unlikely(!path_connected(&nd->path)))
1335 struct mount *mnt = real_mount(nd->path.mnt);
1336 struct mount *mparent = mnt->mnt_parent;
1337 struct dentry *mountpoint = mnt->mnt_mountpoint;
1338 struct inode *inode2 = mountpoint->d_inode;
1339 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1340 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1342 if (&mparent->mnt == nd->path.mnt)
1344 /* we know that mountpoint was pinned */
1345 nd->path.dentry = mountpoint;
1346 nd->path.mnt = &mparent->mnt;
1351 while (unlikely(d_mountpoint(nd->path.dentry))) {
1352 struct mount *mounted;
1353 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1354 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1358 nd->path.mnt = &mounted->mnt;
1359 nd->path.dentry = mounted->mnt.mnt_root;
1360 inode = nd->path.dentry->d_inode;
1361 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1368 * Follow down to the covering mount currently visible to userspace. At each
1369 * point, the filesystem owning that dentry may be queried as to whether the
1370 * caller is permitted to proceed or not.
1372 int follow_down(struct path *path)
1377 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1378 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1379 /* Allow the filesystem to manage the transit without i_mutex
1382 * We indicate to the filesystem if someone is trying to mount
1383 * something here. This gives autofs the chance to deny anyone
1384 * other than its daemon the right to mount on its
1387 * The filesystem may sleep at this point.
1389 if (managed & DCACHE_MANAGE_TRANSIT) {
1390 BUG_ON(!path->dentry->d_op);
1391 BUG_ON(!path->dentry->d_op->d_manage);
1392 ret = path->dentry->d_op->d_manage(
1393 path->dentry, false);
1395 return ret == -EISDIR ? 0 : ret;
1398 /* Transit to a mounted filesystem. */
1399 if (managed & DCACHE_MOUNTED) {
1400 struct vfsmount *mounted = lookup_mnt(path);
1405 path->mnt = mounted;
1406 path->dentry = dget(mounted->mnt_root);
1410 /* Don't handle automount points here */
1415 EXPORT_SYMBOL(follow_down);
1418 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1420 static void follow_mount(struct path *path)
1422 while (d_mountpoint(path->dentry)) {
1423 struct vfsmount *mounted = lookup_mnt(path);
1428 path->mnt = mounted;
1429 path->dentry = dget(mounted->mnt_root);
1433 static int follow_dotdot(struct nameidata *nd)
1439 struct dentry *old = nd->path.dentry;
1441 if (nd->path.dentry == nd->root.dentry &&
1442 nd->path.mnt == nd->root.mnt) {
1445 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1446 /* rare case of legitimate dget_parent()... */
1447 nd->path.dentry = dget_parent(nd->path.dentry);
1449 if (unlikely(!path_connected(&nd->path)))
1453 if (!follow_up(&nd->path))
1456 follow_mount(&nd->path);
1457 nd->inode = nd->path.dentry->d_inode;
1462 * This looks up the name in dcache, possibly revalidates the old dentry and
1463 * allocates a new one if not found or not valid. In the need_lookup argument
1464 * returns whether i_op->lookup is necessary.
1466 * dir->d_inode->i_mutex must be held
1468 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1469 unsigned int flags, bool *need_lookup)
1471 struct dentry *dentry;
1474 *need_lookup = false;
1475 dentry = d_lookup(dir, name);
1477 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1478 error = d_revalidate(dentry, flags);
1479 if (unlikely(error <= 0)) {
1482 return ERR_PTR(error);
1484 d_invalidate(dentry);
1493 dentry = d_alloc(dir, name);
1494 if (unlikely(!dentry))
1495 return ERR_PTR(-ENOMEM);
1497 *need_lookup = true;
1503 * Call i_op->lookup on the dentry. The dentry must be negative and
1506 * dir->d_inode->i_mutex must be held
1508 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1513 /* Don't create child dentry for a dead directory. */
1514 if (unlikely(IS_DEADDIR(dir))) {
1516 return ERR_PTR(-ENOENT);
1519 old = dir->i_op->lookup(dir, dentry, flags);
1520 if (unlikely(old)) {
1527 static struct dentry *__lookup_hash(struct qstr *name,
1528 struct dentry *base, unsigned int flags)
1531 struct dentry *dentry;
1533 dentry = lookup_dcache(name, base, flags, &need_lookup);
1537 return lookup_real(base->d_inode, dentry, flags);
1541 * It's more convoluted than I'd like it to be, but... it's still fairly
1542 * small and for now I'd prefer to have fast path as straight as possible.
1543 * It _is_ time-critical.
1545 static int lookup_fast(struct nameidata *nd,
1546 struct path *path, struct inode **inode,
1549 struct vfsmount *mnt = nd->path.mnt;
1550 struct dentry *dentry, *parent = nd->path.dentry;
1556 * Rename seqlock is not required here because in the off chance
1557 * of a false negative due to a concurrent rename, we're going to
1558 * do the non-racy lookup, below.
1560 if (nd->flags & LOOKUP_RCU) {
1563 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1568 * This sequence count validates that the inode matches
1569 * the dentry name information from lookup.
1571 *inode = d_backing_inode(dentry);
1572 negative = d_is_negative(dentry);
1573 if (read_seqcount_retry(&dentry->d_seq, seq))
1577 * This sequence count validates that the parent had no
1578 * changes while we did the lookup of the dentry above.
1580 * The memory barrier in read_seqcount_begin of child is
1581 * enough, we can use __read_seqcount_retry here.
1583 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1587 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1588 status = d_revalidate(dentry, nd->flags);
1589 if (unlikely(status <= 0)) {
1590 if (status != -ECHILD)
1596 * Note: do negative dentry check after revalidation in
1597 * case that drops it.
1602 path->dentry = dentry;
1603 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1606 if (unlazy_walk(nd, dentry, seq))
1609 dentry = __d_lookup(parent, &nd->last);
1612 if (unlikely(!dentry))
1615 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1616 status = d_revalidate(dentry, nd->flags);
1617 if (unlikely(status <= 0)) {
1622 d_invalidate(dentry);
1627 if (unlikely(d_is_negative(dentry))) {
1632 path->dentry = dentry;
1633 err = follow_managed(path, nd);
1635 *inode = d_backing_inode(path->dentry);
1642 /* Fast lookup failed, do it the slow way */
1643 static int lookup_slow(struct nameidata *nd, struct path *path)
1645 struct dentry *dentry, *parent;
1647 parent = nd->path.dentry;
1648 BUG_ON(nd->inode != parent->d_inode);
1650 mutex_lock(&parent->d_inode->i_mutex);
1651 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1652 mutex_unlock(&parent->d_inode->i_mutex);
1654 return PTR_ERR(dentry);
1655 path->mnt = nd->path.mnt;
1656 path->dentry = dentry;
1657 return follow_managed(path, nd);
1660 static inline int may_lookup(struct nameidata *nd)
1662 if (nd->flags & LOOKUP_RCU) {
1663 int err = inode_permission2(nd->path.mnt, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1666 if (unlazy_walk(nd, NULL, 0))
1669 return inode_permission2(nd->path.mnt, nd->inode, MAY_EXEC);
1672 static inline int handle_dots(struct nameidata *nd, int type)
1674 if (type == LAST_DOTDOT) {
1675 if (nd->flags & LOOKUP_RCU) {
1676 return follow_dotdot_rcu(nd);
1678 return follow_dotdot(nd);
1683 static int pick_link(struct nameidata *nd, struct path *link,
1684 struct inode *inode, unsigned seq)
1688 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1689 path_to_nameidata(link, nd);
1692 if (!(nd->flags & LOOKUP_RCU)) {
1693 if (link->mnt == nd->path.mnt)
1696 error = nd_alloc_stack(nd);
1697 if (unlikely(error)) {
1698 if (error == -ECHILD) {
1699 if (unlikely(unlazy_link(nd, link, seq)))
1701 error = nd_alloc_stack(nd);
1709 last = nd->stack + nd->depth++;
1711 last->cookie = NULL;
1712 last->inode = inode;
1718 * Do we need to follow links? We _really_ want to be able
1719 * to do this check without having to look at inode->i_op,
1720 * so we keep a cache of "no, this doesn't need follow_link"
1721 * for the common case.
1723 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1725 struct inode *inode, unsigned seq)
1727 if (likely(!d_is_symlink(link->dentry)))
1731 /* make sure that d_is_symlink above matches inode */
1732 if (nd->flags & LOOKUP_RCU) {
1733 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1736 return pick_link(nd, link, inode, seq);
1739 enum {WALK_GET = 1, WALK_PUT = 2};
1741 static int walk_component(struct nameidata *nd, int flags)
1744 struct inode *inode;
1748 * "." and ".." are special - ".." especially so because it has
1749 * to be able to know about the current root directory and
1750 * parent relationships.
1752 if (unlikely(nd->last_type != LAST_NORM)) {
1753 err = handle_dots(nd, nd->last_type);
1754 if (flags & WALK_PUT)
1758 err = lookup_fast(nd, &path, &inode, &seq);
1759 if (unlikely(err)) {
1763 err = lookup_slow(nd, &path);
1767 seq = 0; /* we are already out of RCU mode */
1769 if (d_is_negative(path.dentry))
1771 inode = d_backing_inode(path.dentry);
1774 if (flags & WALK_PUT)
1776 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1779 path_to_nameidata(&path, nd);
1785 path_to_nameidata(&path, nd);
1790 * We can do the critical dentry name comparison and hashing
1791 * operations one word at a time, but we are limited to:
1793 * - Architectures with fast unaligned word accesses. We could
1794 * do a "get_unaligned()" if this helps and is sufficiently
1797 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1798 * do not trap on the (extremely unlikely) case of a page
1799 * crossing operation.
1801 * - Furthermore, we need an efficient 64-bit compile for the
1802 * 64-bit case in order to generate the "number of bytes in
1803 * the final mask". Again, that could be replaced with a
1804 * efficient population count instruction or similar.
1806 #ifdef CONFIG_DCACHE_WORD_ACCESS
1808 #include <asm/word-at-a-time.h>
1812 static inline unsigned int fold_hash(unsigned long hash)
1814 return hash_64(hash, 32);
1817 #else /* 32-bit case */
1819 #define fold_hash(x) (x)
1823 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1825 unsigned long a, mask;
1826 unsigned long hash = 0;
1829 a = load_unaligned_zeropad(name);
1830 if (len < sizeof(unsigned long))
1834 name += sizeof(unsigned long);
1835 len -= sizeof(unsigned long);
1839 mask = bytemask_from_count(len);
1842 return fold_hash(hash);
1844 EXPORT_SYMBOL(full_name_hash);
1847 * Calculate the length and hash of the path component, and
1848 * return the "hash_len" as the result.
1850 static inline u64 hash_name(const char *name)
1852 unsigned long a, b, adata, bdata, mask, hash, len;
1853 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1856 len = -sizeof(unsigned long);
1858 hash = (hash + a) * 9;
1859 len += sizeof(unsigned long);
1860 a = load_unaligned_zeropad(name+len);
1861 b = a ^ REPEAT_BYTE('/');
1862 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1864 adata = prep_zero_mask(a, adata, &constants);
1865 bdata = prep_zero_mask(b, bdata, &constants);
1867 mask = create_zero_mask(adata | bdata);
1869 hash += a & zero_bytemask(mask);
1870 len += find_zero(mask);
1871 return hashlen_create(fold_hash(hash), len);
1876 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1878 unsigned long hash = init_name_hash();
1880 hash = partial_name_hash(*name++, hash);
1881 return end_name_hash(hash);
1883 EXPORT_SYMBOL(full_name_hash);
1886 * We know there's a real path component here of at least
1889 static inline u64 hash_name(const char *name)
1891 unsigned long hash = init_name_hash();
1892 unsigned long len = 0, c;
1894 c = (unsigned char)*name;
1897 hash = partial_name_hash(c, hash);
1898 c = (unsigned char)name[len];
1899 } while (c && c != '/');
1900 return hashlen_create(end_name_hash(hash), len);
1907 * This is the basic name resolution function, turning a pathname into
1908 * the final dentry. We expect 'base' to be positive and a directory.
1910 * Returns 0 and nd will have valid dentry and mnt on success.
1911 * Returns error and drops reference to input namei data on failure.
1913 static int link_path_walk(const char *name, struct nameidata *nd)
1922 /* At this point we know we have a real path component. */
1927 err = may_lookup(nd);
1931 hash_len = hash_name(name);
1934 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1936 if (name[1] == '.') {
1938 nd->flags |= LOOKUP_JUMPED;
1944 if (likely(type == LAST_NORM)) {
1945 struct dentry *parent = nd->path.dentry;
1946 nd->flags &= ~LOOKUP_JUMPED;
1947 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1948 struct qstr this = { { .hash_len = hash_len }, .name = name };
1949 err = parent->d_op->d_hash(parent, &this);
1952 hash_len = this.hash_len;
1957 nd->last.hash_len = hash_len;
1958 nd->last.name = name;
1959 nd->last_type = type;
1961 name += hashlen_len(hash_len);
1965 * If it wasn't NUL, we know it was '/'. Skip that
1966 * slash, and continue until no more slashes.
1970 } while (unlikely(*name == '/'));
1971 if (unlikely(!*name)) {
1973 /* pathname body, done */
1976 name = nd->stack[nd->depth - 1].name;
1977 /* trailing symlink, done */
1980 /* last component of nested symlink */
1981 err = walk_component(nd, WALK_GET | WALK_PUT);
1983 err = walk_component(nd, WALK_GET);
1989 const char *s = get_link(nd);
1998 nd->stack[nd->depth - 1].name = name;
2003 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2004 if (nd->flags & LOOKUP_RCU) {
2005 if (unlazy_walk(nd, NULL, 0))
2013 static const char *path_init(struct nameidata *nd, unsigned flags)
2016 const char *s = nd->name->name;
2018 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2019 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2021 if (flags & LOOKUP_ROOT) {
2022 struct dentry *root = nd->root.dentry;
2023 struct vfsmount *mnt = nd->root.mnt;
2024 struct inode *inode = root->d_inode;
2026 if (!d_can_lookup(root))
2027 return ERR_PTR(-ENOTDIR);
2028 retval = inode_permission2(mnt, inode, MAY_EXEC);
2030 return ERR_PTR(retval);
2032 nd->path = nd->root;
2034 if (flags & LOOKUP_RCU) {
2036 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2037 nd->root_seq = nd->seq;
2038 nd->m_seq = read_seqbegin(&mount_lock);
2040 path_get(&nd->path);
2045 nd->root.mnt = NULL;
2047 nd->m_seq = read_seqbegin(&mount_lock);
2049 if (flags & LOOKUP_RCU) {
2052 nd->seq = nd->root_seq;
2055 path_get(&nd->root);
2057 nd->path = nd->root;
2058 } else if (nd->dfd == AT_FDCWD) {
2059 if (flags & LOOKUP_RCU) {
2060 struct fs_struct *fs = current->fs;
2066 seq = read_seqcount_begin(&fs->seq);
2068 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2069 } while (read_seqcount_retry(&fs->seq, seq));
2071 get_fs_pwd(current->fs, &nd->path);
2074 /* Caller must check execute permissions on the starting path component */
2075 struct fd f = fdget_raw(nd->dfd);
2076 struct dentry *dentry;
2079 return ERR_PTR(-EBADF);
2081 dentry = f.file->f_path.dentry;
2084 if (!d_can_lookup(dentry)) {
2086 return ERR_PTR(-ENOTDIR);
2090 nd->path = f.file->f_path;
2091 if (flags & LOOKUP_RCU) {
2093 nd->inode = nd->path.dentry->d_inode;
2094 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2096 path_get(&nd->path);
2097 nd->inode = nd->path.dentry->d_inode;
2103 nd->inode = nd->path.dentry->d_inode;
2104 if (!(flags & LOOKUP_RCU))
2106 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
2108 if (!(nd->flags & LOOKUP_ROOT))
2109 nd->root.mnt = NULL;
2111 return ERR_PTR(-ECHILD);
2114 static const char *trailing_symlink(struct nameidata *nd)
2117 int error = may_follow_link(nd);
2118 if (unlikely(error))
2119 return ERR_PTR(error);
2120 nd->flags |= LOOKUP_PARENT;
2121 nd->stack[0].name = NULL;
2126 static inline int lookup_last(struct nameidata *nd)
2128 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2129 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2131 nd->flags &= ~LOOKUP_PARENT;
2132 return walk_component(nd,
2133 nd->flags & LOOKUP_FOLLOW
2135 ? WALK_PUT | WALK_GET
2140 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2141 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2143 const char *s = path_init(nd, flags);
2148 while (!(err = link_path_walk(s, nd))
2149 && ((err = lookup_last(nd)) > 0)) {
2150 s = trailing_symlink(nd);
2157 err = complete_walk(nd);
2159 if (!err && nd->flags & LOOKUP_DIRECTORY)
2160 if (!d_can_lookup(nd->path.dentry))
2164 nd->path.mnt = NULL;
2165 nd->path.dentry = NULL;
2171 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2172 struct path *path, struct path *root)
2175 struct nameidata nd;
2177 return PTR_ERR(name);
2178 if (unlikely(root)) {
2180 flags |= LOOKUP_ROOT;
2182 set_nameidata(&nd, dfd, name);
2183 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2184 if (unlikely(retval == -ECHILD))
2185 retval = path_lookupat(&nd, flags, path);
2186 if (unlikely(retval == -ESTALE))
2187 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2189 if (likely(!retval))
2190 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2191 restore_nameidata();
2196 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2197 static int path_parentat(struct nameidata *nd, unsigned flags,
2198 struct path *parent)
2200 const char *s = path_init(nd, flags);
2204 err = link_path_walk(s, nd);
2206 err = complete_walk(nd);
2209 nd->path.mnt = NULL;
2210 nd->path.dentry = NULL;
2216 static struct filename *filename_parentat(int dfd, struct filename *name,
2217 unsigned int flags, struct path *parent,
2218 struct qstr *last, int *type)
2221 struct nameidata nd;
2225 set_nameidata(&nd, dfd, name);
2226 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2227 if (unlikely(retval == -ECHILD))
2228 retval = path_parentat(&nd, flags, parent);
2229 if (unlikely(retval == -ESTALE))
2230 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2231 if (likely(!retval)) {
2233 *type = nd.last_type;
2234 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2237 name = ERR_PTR(retval);
2239 restore_nameidata();
2243 /* does lookup, returns the object with parent locked */
2244 struct dentry *kern_path_locked(const char *name, struct path *path)
2246 struct filename *filename;
2251 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2253 if (IS_ERR(filename))
2254 return ERR_CAST(filename);
2255 if (unlikely(type != LAST_NORM)) {
2258 return ERR_PTR(-EINVAL);
2260 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2261 d = __lookup_hash(&last, path->dentry, 0);
2263 mutex_unlock(&path->dentry->d_inode->i_mutex);
2270 int kern_path(const char *name, unsigned int flags, struct path *path)
2272 return filename_lookup(AT_FDCWD, getname_kernel(name),
2275 EXPORT_SYMBOL(kern_path);
2278 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2279 * @dentry: pointer to dentry of the base directory
2280 * @mnt: pointer to vfs mount of the base directory
2281 * @name: pointer to file name
2282 * @flags: lookup flags
2283 * @path: pointer to struct path to fill
2285 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2286 const char *name, unsigned int flags,
2289 struct path root = {.mnt = mnt, .dentry = dentry};
2290 /* the first argument of filename_lookup() is ignored with root */
2291 return filename_lookup(AT_FDCWD, getname_kernel(name),
2292 flags , path, &root);
2294 EXPORT_SYMBOL(vfs_path_lookup);
2297 * lookup_one_len - filesystem helper to lookup single pathname component
2298 * @name: pathname component to lookup
2299 * @mnt: mount we are looking up on
2300 * @base: base directory to lookup from
2301 * @len: maximum length @len should be interpreted to
2303 * Note that this routine is purely a helper for filesystem usage and should
2304 * not be called by generic code.
2306 struct dentry *lookup_one_len2(const char *name, struct vfsmount *mnt, struct dentry *base, int len)
2312 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2316 this.hash = full_name_hash(name, len);
2318 return ERR_PTR(-EACCES);
2320 if (unlikely(name[0] == '.')) {
2321 if (len < 2 || (len == 2 && name[1] == '.'))
2322 return ERR_PTR(-EACCES);
2326 c = *(const unsigned char *)name++;
2327 if (c == '/' || c == '\0')
2328 return ERR_PTR(-EACCES);
2331 * See if the low-level filesystem might want
2332 * to use its own hash..
2334 if (base->d_flags & DCACHE_OP_HASH) {
2335 int err = base->d_op->d_hash(base, &this);
2337 return ERR_PTR(err);
2340 err = inode_permission2(mnt, base->d_inode, MAY_EXEC);
2342 return ERR_PTR(err);
2344 return __lookup_hash(&this, base, 0);
2346 EXPORT_SYMBOL(lookup_one_len2);
2348 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2350 return lookup_one_len2(name, NULL, base, len);
2352 EXPORT_SYMBOL(lookup_one_len);
2354 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2355 struct path *path, int *empty)
2357 return filename_lookup(dfd, getname_flags(name, flags, empty),
2360 EXPORT_SYMBOL(user_path_at_empty);
2363 * NB: most callers don't do anything directly with the reference to the
2364 * to struct filename, but the nd->last pointer points into the name string
2365 * allocated by getname. So we must hold the reference to it until all
2366 * path-walking is complete.
2368 static inline struct filename *
2369 user_path_parent(int dfd, const char __user *path,
2370 struct path *parent,
2375 /* only LOOKUP_REVAL is allowed in extra flags */
2376 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2377 parent, last, type);
2381 * mountpoint_last - look up last component for umount
2382 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2383 * @path: pointer to container for result
2385 * This is a special lookup_last function just for umount. In this case, we
2386 * need to resolve the path without doing any revalidation.
2388 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2389 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2390 * in almost all cases, this lookup will be served out of the dcache. The only
2391 * cases where it won't are if nd->last refers to a symlink or the path is
2392 * bogus and it doesn't exist.
2395 * -error: if there was an error during lookup. This includes -ENOENT if the
2396 * lookup found a negative dentry. The nd->path reference will also be
2399 * 0: if we successfully resolved nd->path and found it to not to be a
2400 * symlink that needs to be followed. "path" will also be populated.
2401 * The nd->path reference will also be put.
2403 * 1: if we successfully resolved nd->last and found it to be a symlink
2404 * that needs to be followed. "path" will be populated with the path
2405 * to the link, and nd->path will *not* be put.
2408 mountpoint_last(struct nameidata *nd, struct path *path)
2411 struct dentry *dentry;
2412 struct dentry *dir = nd->path.dentry;
2414 /* If we're in rcuwalk, drop out of it to handle last component */
2415 if (nd->flags & LOOKUP_RCU) {
2416 if (unlazy_walk(nd, NULL, 0))
2420 nd->flags &= ~LOOKUP_PARENT;
2422 if (unlikely(nd->last_type != LAST_NORM)) {
2423 error = handle_dots(nd, nd->last_type);
2426 dentry = dget(nd->path.dentry);
2430 mutex_lock(&dir->d_inode->i_mutex);
2431 dentry = d_lookup(dir, &nd->last);
2434 * No cached dentry. Mounted dentries are pinned in the cache,
2435 * so that means that this dentry is probably a symlink or the
2436 * path doesn't actually point to a mounted dentry.
2438 dentry = d_alloc(dir, &nd->last);
2440 mutex_unlock(&dir->d_inode->i_mutex);
2443 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2444 if (IS_ERR(dentry)) {
2445 mutex_unlock(&dir->d_inode->i_mutex);
2446 return PTR_ERR(dentry);
2449 mutex_unlock(&dir->d_inode->i_mutex);
2452 if (d_is_negative(dentry)) {
2458 path->dentry = dentry;
2459 path->mnt = nd->path.mnt;
2460 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2461 d_backing_inode(dentry), 0);
2462 if (unlikely(error))
2470 * path_mountpoint - look up a path to be umounted
2471 * @nd: lookup context
2472 * @flags: lookup flags
2473 * @path: pointer to container for result
2475 * Look up the given name, but don't attempt to revalidate the last component.
2476 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2479 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2481 const char *s = path_init(nd, flags);
2485 while (!(err = link_path_walk(s, nd)) &&
2486 (err = mountpoint_last(nd, path)) > 0) {
2487 s = trailing_symlink(nd);
2498 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2501 struct nameidata nd;
2504 return PTR_ERR(name);
2505 set_nameidata(&nd, dfd, name);
2506 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2507 if (unlikely(error == -ECHILD))
2508 error = path_mountpoint(&nd, flags, path);
2509 if (unlikely(error == -ESTALE))
2510 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2512 audit_inode(name, path->dentry, 0);
2513 restore_nameidata();
2519 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2520 * @dfd: directory file descriptor
2521 * @name: pathname from userland
2522 * @flags: lookup flags
2523 * @path: pointer to container to hold result
2525 * A umount is a special case for path walking. We're not actually interested
2526 * in the inode in this situation, and ESTALE errors can be a problem. We
2527 * simply want track down the dentry and vfsmount attached at the mountpoint
2528 * and avoid revalidating the last component.
2530 * Returns 0 and populates "path" on success.
2533 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2536 return filename_mountpoint(dfd, getname(name), path, flags);
2540 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2543 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2545 EXPORT_SYMBOL(kern_path_mountpoint);
2547 int __check_sticky(struct inode *dir, struct inode *inode)
2549 kuid_t fsuid = current_fsuid();
2551 if (uid_eq(inode->i_uid, fsuid))
2553 if (uid_eq(dir->i_uid, fsuid))
2555 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2557 EXPORT_SYMBOL(__check_sticky);
2560 * Check whether we can remove a link victim from directory dir, check
2561 * whether the type of victim is right.
2562 * 1. We can't do it if dir is read-only (done in permission())
2563 * 2. We should have write and exec permissions on dir
2564 * 3. We can't remove anything from append-only dir
2565 * 4. We can't do anything with immutable dir (done in permission())
2566 * 5. If the sticky bit on dir is set we should either
2567 * a. be owner of dir, or
2568 * b. be owner of victim, or
2569 * c. have CAP_FOWNER capability
2570 * 6. If the victim is append-only or immutable we can't do antyhing with
2571 * links pointing to it.
2572 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2573 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2574 * 9. We can't remove a root or mountpoint.
2575 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2576 * nfs_async_unlink().
2578 static int may_delete(struct vfsmount *mnt, struct inode *dir, struct dentry *victim, bool isdir)
2580 struct inode *inode = d_backing_inode(victim);
2583 if (d_is_negative(victim))
2587 BUG_ON(victim->d_parent->d_inode != dir);
2588 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2590 error = inode_permission2(mnt, dir, MAY_WRITE | MAY_EXEC);
2596 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2597 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2600 if (!d_is_dir(victim))
2602 if (IS_ROOT(victim))
2604 } else if (d_is_dir(victim))
2606 if (IS_DEADDIR(dir))
2608 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2613 /* Check whether we can create an object with dentry child in directory
2615 * 1. We can't do it if child already exists (open has special treatment for
2616 * this case, but since we are inlined it's OK)
2617 * 2. We can't do it if dir is read-only (done in permission())
2618 * 3. We should have write and exec permissions on dir
2619 * 4. We can't do it if dir is immutable (done in permission())
2621 static inline int may_create(struct vfsmount *mnt, struct inode *dir, struct dentry *child)
2623 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2626 if (IS_DEADDIR(dir))
2628 return inode_permission2(mnt, dir, MAY_WRITE | MAY_EXEC);
2632 * p1 and p2 should be directories on the same fs.
2634 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2639 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2643 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2645 p = d_ancestor(p2, p1);
2647 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2648 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2652 p = d_ancestor(p1, p2);
2654 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2655 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2659 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2660 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2663 EXPORT_SYMBOL(lock_rename);
2665 void unlock_rename(struct dentry *p1, struct dentry *p2)
2667 mutex_unlock(&p1->d_inode->i_mutex);
2669 mutex_unlock(&p2->d_inode->i_mutex);
2670 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2673 EXPORT_SYMBOL(unlock_rename);
2675 int vfs_create2(struct vfsmount *mnt, struct inode *dir, struct dentry *dentry,
2676 umode_t mode, bool want_excl)
2678 int error = may_create(mnt, dir, dentry);
2682 if (!dir->i_op->create)
2683 return -EACCES; /* shouldn't it be ENOSYS? */
2686 error = security_inode_create(dir, dentry, mode);
2689 error = dir->i_op->create(dir, dentry, mode, want_excl);
2691 fsnotify_create(dir, dentry);
2694 EXPORT_SYMBOL(vfs_create2);
2696 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2699 return vfs_create2(NULL, dir, dentry, mode, want_excl);
2701 EXPORT_SYMBOL(vfs_create);
2703 static int may_open(struct path *path, int acc_mode, int flag)
2705 struct dentry *dentry = path->dentry;
2706 struct vfsmount *mnt = path->mnt;
2707 struct inode *inode = dentry->d_inode;
2717 switch (inode->i_mode & S_IFMT) {
2721 if (acc_mode & MAY_WRITE)
2726 if (path->mnt->mnt_flags & MNT_NODEV)
2735 error = inode_permission2(mnt, inode, acc_mode);
2740 * An append-only file must be opened in append mode for writing.
2742 if (IS_APPEND(inode)) {
2743 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2749 /* O_NOATIME can only be set by the owner or superuser */
2750 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2756 static int handle_truncate(struct file *filp)
2758 struct path *path = &filp->f_path;
2759 struct inode *inode = path->dentry->d_inode;
2760 int error = get_write_access(inode);
2764 * Refuse to truncate files with mandatory locks held on them.
2766 error = locks_verify_locked(filp);
2768 error = security_path_truncate(path);
2770 error = do_truncate2(path->mnt, path->dentry, 0,
2771 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2774 put_write_access(inode);
2778 static inline int open_to_namei_flags(int flag)
2780 if ((flag & O_ACCMODE) == 3)
2785 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2787 int error = security_path_mknod(dir, dentry, mode, 0);
2791 error = inode_permission2(dir->mnt, dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2795 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2799 * Attempt to atomically look up, create and open a file from a negative
2802 * Returns 0 if successful. The file will have been created and attached to
2803 * @file by the filesystem calling finish_open().
2805 * Returns 1 if the file was looked up only or didn't need creating. The
2806 * caller will need to perform the open themselves. @path will have been
2807 * updated to point to the new dentry. This may be negative.
2809 * Returns an error code otherwise.
2811 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2812 struct path *path, struct file *file,
2813 const struct open_flags *op,
2814 bool got_write, bool need_lookup,
2817 struct inode *dir = nd->path.dentry->d_inode;
2818 unsigned open_flag = open_to_namei_flags(op->open_flag);
2822 int create_error = 0;
2823 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2826 BUG_ON(dentry->d_inode);
2828 /* Don't create child dentry for a dead directory. */
2829 if (unlikely(IS_DEADDIR(dir))) {
2835 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2836 mode &= ~current_umask();
2838 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2840 open_flag &= ~O_TRUNC;
2843 * Checking write permission is tricky, bacuse we don't know if we are
2844 * going to actually need it: O_CREAT opens should work as long as the
2845 * file exists. But checking existence breaks atomicity. The trick is
2846 * to check access and if not granted clear O_CREAT from the flags.
2848 * Another problem is returing the "right" error value (e.g. for an
2849 * O_EXCL open we want to return EEXIST not EROFS).
2851 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2852 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2853 if (!(open_flag & O_CREAT)) {
2855 * No O_CREATE -> atomicity not a requirement -> fall
2856 * back to lookup + open
2859 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2860 /* Fall back and fail with the right error */
2861 create_error = -EROFS;
2864 /* No side effects, safe to clear O_CREAT */
2865 create_error = -EROFS;
2866 open_flag &= ~O_CREAT;
2870 if (open_flag & O_CREAT) {
2871 error = may_o_create(&nd->path, dentry, mode);
2873 create_error = error;
2874 if (open_flag & O_EXCL)
2876 open_flag &= ~O_CREAT;
2880 if (nd->flags & LOOKUP_DIRECTORY)
2881 open_flag |= O_DIRECTORY;
2883 file->f_path.dentry = DENTRY_NOT_SET;
2884 file->f_path.mnt = nd->path.mnt;
2885 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2888 if (create_error && error == -ENOENT)
2889 error = create_error;
2893 if (error) { /* returned 1, that is */
2894 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2898 if (file->f_path.dentry) {
2900 dentry = file->f_path.dentry;
2902 if (*opened & FILE_CREATED)
2903 fsnotify_create(dir, dentry);
2904 if (!dentry->d_inode) {
2905 WARN_ON(*opened & FILE_CREATED);
2907 error = create_error;
2911 if (excl && !(*opened & FILE_CREATED)) {
2920 * We didn't have the inode before the open, so check open permission
2923 acc_mode = op->acc_mode;
2924 if (*opened & FILE_CREATED) {
2925 WARN_ON(!(open_flag & O_CREAT));
2926 fsnotify_create(dir, dentry);
2927 acc_mode = MAY_OPEN;
2929 error = may_open(&file->f_path, acc_mode, open_flag);
2939 dentry = lookup_real(dir, dentry, nd->flags);
2941 return PTR_ERR(dentry);
2943 if (create_error && !dentry->d_inode) {
2944 error = create_error;
2948 path->dentry = dentry;
2949 path->mnt = nd->path.mnt;
2954 * Look up and maybe create and open the last component.
2956 * Must be called with i_mutex held on parent.
2958 * Returns 0 if the file was successfully atomically created (if necessary) and
2959 * opened. In this case the file will be returned attached to @file.
2961 * Returns 1 if the file was not completely opened at this time, though lookups
2962 * and creations will have been performed and the dentry returned in @path will
2963 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2964 * specified then a negative dentry may be returned.
2966 * An error code is returned otherwise.
2968 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2969 * cleared otherwise prior to returning.
2971 static int lookup_open(struct nameidata *nd, struct path *path,
2973 const struct open_flags *op,
2974 bool got_write, int *opened)
2976 struct dentry *dir = nd->path.dentry;
2977 struct vfsmount *mnt = nd->path.mnt;
2978 struct inode *dir_inode = dir->d_inode;
2979 struct dentry *dentry;
2983 *opened &= ~FILE_CREATED;
2984 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2986 return PTR_ERR(dentry);
2988 /* Cached positive dentry: will open in f_op->open */
2989 if (!need_lookup && dentry->d_inode)
2992 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2993 return atomic_open(nd, dentry, path, file, op, got_write,
2994 need_lookup, opened);
2998 BUG_ON(dentry->d_inode);
3000 dentry = lookup_real(dir_inode, dentry, nd->flags);
3002 return PTR_ERR(dentry);
3005 /* Negative dentry, just create the file */
3006 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
3007 umode_t mode = op->mode;
3008 if (!IS_POSIXACL(dir->d_inode))
3009 mode &= ~current_umask();
3011 * This write is needed to ensure that a
3012 * rw->ro transition does not occur between
3013 * the time when the file is created and when
3014 * a permanent write count is taken through
3015 * the 'struct file' in finish_open().
3021 *opened |= FILE_CREATED;
3022 error = security_path_mknod(&nd->path, dentry, mode, 0);
3025 error = vfs_create2(mnt, dir->d_inode, dentry, mode,
3026 nd->flags & LOOKUP_EXCL);
3031 path->dentry = dentry;
3032 path->mnt = nd->path.mnt;
3041 * Handle the last step of open()
3043 static int do_last(struct nameidata *nd,
3044 struct file *file, const struct open_flags *op,
3047 struct dentry *dir = nd->path.dentry;
3048 int open_flag = op->open_flag;
3049 bool will_truncate = (open_flag & O_TRUNC) != 0;
3050 bool got_write = false;
3051 int acc_mode = op->acc_mode;
3053 struct inode *inode;
3054 struct path save_parent = { .dentry = NULL, .mnt = NULL };
3056 bool retried = false;
3059 nd->flags &= ~LOOKUP_PARENT;
3060 nd->flags |= op->intent;
3062 if (nd->last_type != LAST_NORM) {
3063 error = handle_dots(nd, nd->last_type);
3064 if (unlikely(error))
3069 if (!(open_flag & O_CREAT)) {
3070 if (nd->last.name[nd->last.len])
3071 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3072 /* we _can_ be in RCU mode here */
3073 error = lookup_fast(nd, &path, &inode, &seq);
3080 BUG_ON(nd->inode != dir->d_inode);
3082 /* create side of things */
3084 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3085 * has been cleared when we got to the last component we are
3088 error = complete_walk(nd);
3092 audit_inode(nd->name, dir, LOOKUP_PARENT);
3093 /* trailing slashes? */
3094 if (unlikely(nd->last.name[nd->last.len]))
3099 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3100 error = mnt_want_write(nd->path.mnt);
3104 * do _not_ fail yet - we might not need that or fail with
3105 * a different error; let lookup_open() decide; we'll be
3106 * dropping this one anyway.
3109 mutex_lock(&dir->d_inode->i_mutex);
3110 error = lookup_open(nd, &path, file, op, got_write, opened);
3111 mutex_unlock(&dir->d_inode->i_mutex);
3117 if ((*opened & FILE_CREATED) ||
3118 !S_ISREG(file_inode(file)->i_mode))
3119 will_truncate = false;
3121 audit_inode(nd->name, file->f_path.dentry, 0);
3125 if (*opened & FILE_CREATED) {
3126 /* Don't check for write permission, don't truncate */
3127 open_flag &= ~O_TRUNC;
3128 will_truncate = false;
3129 acc_mode = MAY_OPEN;
3130 path_to_nameidata(&path, nd);
3131 goto finish_open_created;
3135 * create/update audit record if it already exists.
3137 if (d_is_positive(path.dentry))
3138 audit_inode(nd->name, path.dentry, 0);
3141 * If atomic_open() acquired write access it is dropped now due to
3142 * possible mount and symlink following (this might be optimized away if
3146 mnt_drop_write(nd->path.mnt);
3150 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3151 path_to_nameidata(&path, nd);
3155 error = follow_managed(&path, nd);
3156 if (unlikely(error < 0))
3159 BUG_ON(nd->flags & LOOKUP_RCU);
3160 seq = 0; /* out of RCU mode, so the value doesn't matter */
3161 if (unlikely(d_is_negative(path.dentry))) {
3162 path_to_nameidata(&path, nd);
3165 inode = d_backing_inode(path.dentry);
3169 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3171 if (unlikely(error))
3174 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3175 path_to_nameidata(&path, nd);
3177 save_parent.dentry = nd->path.dentry;
3178 save_parent.mnt = mntget(path.mnt);
3179 nd->path.dentry = path.dentry;
3184 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3186 error = complete_walk(nd);
3188 path_put(&save_parent);
3191 audit_inode(nd->name, nd->path.dentry, 0);
3192 if (unlikely(d_is_symlink(nd->path.dentry)) && !(open_flag & O_PATH)) {
3197 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3200 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3202 if (!d_is_reg(nd->path.dentry))
3203 will_truncate = false;
3205 if (will_truncate) {
3206 error = mnt_want_write(nd->path.mnt);
3211 finish_open_created:
3212 error = may_open(&nd->path, acc_mode, open_flag);
3216 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3217 error = vfs_open(&nd->path, file, current_cred());
3219 *opened |= FILE_OPENED;
3221 if (error == -EOPENSTALE)
3226 error = open_check_o_direct(file);
3229 error = ima_file_check(file, op->acc_mode, *opened);
3233 if (will_truncate) {
3234 error = handle_truncate(file);
3239 if (unlikely(error > 0)) {
3244 mnt_drop_write(nd->path.mnt);
3245 path_put(&save_parent);
3253 /* If no saved parent or already retried then can't retry */
3254 if (!save_parent.dentry || retried)
3257 BUG_ON(save_parent.dentry != dir);
3258 path_put(&nd->path);
3259 nd->path = save_parent;
3260 nd->inode = dir->d_inode;
3261 save_parent.mnt = NULL;
3262 save_parent.dentry = NULL;
3264 mnt_drop_write(nd->path.mnt);
3271 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3272 const struct open_flags *op,
3273 struct file *file, int *opened)
3275 static const struct qstr name = QSTR_INIT("/", 1);
3276 struct dentry *child;
3279 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3280 if (unlikely(error))
3282 error = mnt_want_write(path.mnt);
3283 if (unlikely(error))
3285 dir = path.dentry->d_inode;
3286 /* we want directory to be writable */
3287 error = inode_permission2(path.mnt, dir, MAY_WRITE | MAY_EXEC);
3290 if (!dir->i_op->tmpfile) {
3291 error = -EOPNOTSUPP;
3294 child = d_alloc(path.dentry, &name);
3295 if (unlikely(!child)) {
3300 path.dentry = child;
3301 error = dir->i_op->tmpfile(dir, child, op->mode);
3304 audit_inode(nd->name, child, 0);
3305 /* Don't check for other permissions, the inode was just created */
3306 error = may_open(&path, MAY_OPEN, op->open_flag);
3309 file->f_path.mnt = path.mnt;
3310 error = finish_open(file, child, NULL, opened);
3313 error = open_check_o_direct(file);
3316 } else if (!(op->open_flag & O_EXCL)) {
3317 struct inode *inode = file_inode(file);
3318 spin_lock(&inode->i_lock);
3319 inode->i_state |= I_LINKABLE;
3320 spin_unlock(&inode->i_lock);
3323 mnt_drop_write(path.mnt);
3329 static struct file *path_openat(struct nameidata *nd,
3330 const struct open_flags *op, unsigned flags)
3337 file = get_empty_filp();
3341 file->f_flags = op->open_flag;
3343 if (unlikely(file->f_flags & __O_TMPFILE)) {
3344 error = do_tmpfile(nd, flags, op, file, &opened);
3348 s = path_init(nd, flags);
3353 while (!(error = link_path_walk(s, nd)) &&
3354 (error = do_last(nd, file, op, &opened)) > 0) {
3355 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3356 s = trailing_symlink(nd);
3364 if (!(opened & FILE_OPENED)) {
3368 if (unlikely(error)) {
3369 if (error == -EOPENSTALE) {
3370 if (flags & LOOKUP_RCU)
3375 file = ERR_PTR(error);
3380 struct file *do_filp_open(int dfd, struct filename *pathname,
3381 const struct open_flags *op)
3383 struct nameidata nd;
3384 int flags = op->lookup_flags;
3387 set_nameidata(&nd, dfd, pathname);
3388 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3389 if (unlikely(filp == ERR_PTR(-ECHILD)))
3390 filp = path_openat(&nd, op, flags);
3391 if (unlikely(filp == ERR_PTR(-ESTALE)))
3392 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3393 restore_nameidata();
3397 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3398 const char *name, const struct open_flags *op)
3400 struct nameidata nd;
3402 struct filename *filename;
3403 int flags = op->lookup_flags | LOOKUP_ROOT;
3406 nd.root.dentry = dentry;
3408 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3409 return ERR_PTR(-ELOOP);
3411 filename = getname_kernel(name);
3412 if (IS_ERR(filename))
3413 return ERR_CAST(filename);
3415 set_nameidata(&nd, -1, filename);
3416 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3417 if (unlikely(file == ERR_PTR(-ECHILD)))
3418 file = path_openat(&nd, op, flags);
3419 if (unlikely(file == ERR_PTR(-ESTALE)))
3420 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3421 restore_nameidata();
3426 static struct dentry *filename_create(int dfd, struct filename *name,
3427 struct path *path, unsigned int lookup_flags)
3429 struct dentry *dentry = ERR_PTR(-EEXIST);
3434 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3437 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3438 * other flags passed in are ignored!
3440 lookup_flags &= LOOKUP_REVAL;
3442 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3444 return ERR_CAST(name);
3447 * Yucky last component or no last component at all?
3448 * (foo/., foo/.., /////)
3450 if (unlikely(type != LAST_NORM))
3453 /* don't fail immediately if it's r/o, at least try to report other errors */
3454 err2 = mnt_want_write(path->mnt);
3456 * Do the final lookup.
3458 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3459 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3460 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3465 if (d_is_positive(dentry))
3469 * Special case - lookup gave negative, but... we had foo/bar/
3470 * From the vfs_mknod() POV we just have a negative dentry -
3471 * all is fine. Let's be bastards - you had / on the end, you've
3472 * been asking for (non-existent) directory. -ENOENT for you.
3474 if (unlikely(!is_dir && last.name[last.len])) {
3478 if (unlikely(err2)) {
3486 dentry = ERR_PTR(error);
3488 mutex_unlock(&path->dentry->d_inode->i_mutex);
3490 mnt_drop_write(path->mnt);
3497 struct dentry *kern_path_create(int dfd, const char *pathname,
3498 struct path *path, unsigned int lookup_flags)
3500 return filename_create(dfd, getname_kernel(pathname),
3501 path, lookup_flags);
3503 EXPORT_SYMBOL(kern_path_create);
3505 void done_path_create(struct path *path, struct dentry *dentry)
3508 mutex_unlock(&path->dentry->d_inode->i_mutex);
3509 mnt_drop_write(path->mnt);
3512 EXPORT_SYMBOL(done_path_create);
3514 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3515 struct path *path, unsigned int lookup_flags)
3517 return filename_create(dfd, getname(pathname), path, lookup_flags);
3519 EXPORT_SYMBOL(user_path_create);
3521 int vfs_mknod2(struct vfsmount *mnt, struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3523 int error = may_create(mnt, dir, dentry);
3528 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3531 if (!dir->i_op->mknod)
3534 error = devcgroup_inode_mknod(mode, dev);
3538 error = security_inode_mknod(dir, dentry, mode, dev);
3542 error = dir->i_op->mknod(dir, dentry, mode, dev);
3544 fsnotify_create(dir, dentry);
3547 EXPORT_SYMBOL(vfs_mknod2);
3549 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3551 return vfs_mknod2(NULL, dir, dentry, mode, dev);
3553 EXPORT_SYMBOL(vfs_mknod);
3555 static int may_mknod(umode_t mode)
3557 switch (mode & S_IFMT) {
3563 case 0: /* zero mode translates to S_IFREG */
3572 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3575 struct dentry *dentry;
3578 unsigned int lookup_flags = 0;
3580 error = may_mknod(mode);
3584 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3586 return PTR_ERR(dentry);
3588 if (!IS_POSIXACL(path.dentry->d_inode))
3589 mode &= ~current_umask();
3590 error = security_path_mknod(&path, dentry, mode, dev);
3593 switch (mode & S_IFMT) {
3594 case 0: case S_IFREG:
3595 error = vfs_create2(path.mnt, path.dentry->d_inode,dentry,mode,true);
3597 case S_IFCHR: case S_IFBLK:
3598 error = vfs_mknod2(path.mnt, path.dentry->d_inode,dentry,mode,
3599 new_decode_dev(dev));
3601 case S_IFIFO: case S_IFSOCK:
3602 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3606 done_path_create(&path, dentry);
3607 if (retry_estale(error, lookup_flags)) {
3608 lookup_flags |= LOOKUP_REVAL;
3614 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3616 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3619 int vfs_mkdir2(struct vfsmount *mnt, struct inode *dir, struct dentry *dentry, umode_t mode)
3621 int error = may_create(mnt, dir, dentry);
3622 unsigned max_links = dir->i_sb->s_max_links;
3627 if (!dir->i_op->mkdir)
3630 mode &= (S_IRWXUGO|S_ISVTX);
3631 error = security_inode_mkdir(dir, dentry, mode);
3635 if (max_links && dir->i_nlink >= max_links)
3638 error = dir->i_op->mkdir(dir, dentry, mode);
3640 fsnotify_mkdir(dir, dentry);
3643 EXPORT_SYMBOL(vfs_mkdir2);
3645 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3647 return vfs_mkdir2(NULL, dir, dentry, mode);
3649 EXPORT_SYMBOL(vfs_mkdir);
3651 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3653 struct dentry *dentry;
3656 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3659 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3661 return PTR_ERR(dentry);
3663 if (!IS_POSIXACL(path.dentry->d_inode))
3664 mode &= ~current_umask();
3665 error = security_path_mkdir(&path, dentry, mode);
3667 error = vfs_mkdir2(path.mnt, path.dentry->d_inode, dentry, mode);
3668 done_path_create(&path, dentry);
3669 if (retry_estale(error, lookup_flags)) {
3670 lookup_flags |= LOOKUP_REVAL;
3676 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3678 return sys_mkdirat(AT_FDCWD, pathname, mode);
3682 * The dentry_unhash() helper will try to drop the dentry early: we
3683 * should have a usage count of 1 if we're the only user of this
3684 * dentry, and if that is true (possibly after pruning the dcache),
3685 * then we drop the dentry now.
3687 * A low-level filesystem can, if it choses, legally
3690 * if (!d_unhashed(dentry))
3693 * if it cannot handle the case of removing a directory
3694 * that is still in use by something else..
3696 void dentry_unhash(struct dentry *dentry)
3698 shrink_dcache_parent(dentry);
3699 spin_lock(&dentry->d_lock);
3700 if (dentry->d_lockref.count == 1)
3702 spin_unlock(&dentry->d_lock);
3704 EXPORT_SYMBOL(dentry_unhash);
3706 int vfs_rmdir2(struct vfsmount *mnt, struct inode *dir, struct dentry *dentry)
3708 int error = may_delete(mnt, dir, dentry, 1);
3713 if (!dir->i_op->rmdir)
3717 mutex_lock(&dentry->d_inode->i_mutex);
3720 if (is_local_mountpoint(dentry))
3723 error = security_inode_rmdir(dir, dentry);
3727 shrink_dcache_parent(dentry);
3728 error = dir->i_op->rmdir(dir, dentry);
3732 dentry->d_inode->i_flags |= S_DEAD;
3734 detach_mounts(dentry);
3737 mutex_unlock(&dentry->d_inode->i_mutex);
3743 EXPORT_SYMBOL(vfs_rmdir2);
3745 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3747 return vfs_rmdir2(NULL, dir, dentry);
3749 EXPORT_SYMBOL(vfs_rmdir);
3751 static long do_rmdir(int dfd, const char __user *pathname)
3754 struct filename *name;
3755 struct dentry *dentry;
3759 unsigned int lookup_flags = 0;
3761 name = user_path_parent(dfd, pathname,
3762 &path, &last, &type, lookup_flags);
3764 return PTR_ERR(name);
3778 error = mnt_want_write(path.mnt);
3782 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3783 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3784 error = PTR_ERR(dentry);
3787 if (!dentry->d_inode) {
3791 error = security_path_rmdir(&path, dentry);
3794 error = vfs_rmdir2(path.mnt, path.dentry->d_inode, dentry);
3798 mutex_unlock(&path.dentry->d_inode->i_mutex);
3799 mnt_drop_write(path.mnt);
3803 if (retry_estale(error, lookup_flags)) {
3804 lookup_flags |= LOOKUP_REVAL;
3810 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3812 return do_rmdir(AT_FDCWD, pathname);
3816 * vfs_unlink - unlink a filesystem object
3817 * @dir: parent directory
3819 * @delegated_inode: returns victim inode, if the inode is delegated.
3821 * The caller must hold dir->i_mutex.
3823 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3824 * return a reference to the inode in delegated_inode. The caller
3825 * should then break the delegation on that inode and retry. Because
3826 * breaking a delegation may take a long time, the caller should drop
3827 * dir->i_mutex before doing so.
3829 * Alternatively, a caller may pass NULL for delegated_inode. This may
3830 * be appropriate for callers that expect the underlying filesystem not
3831 * to be NFS exported.
3833 int vfs_unlink2(struct vfsmount *mnt, struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3835 struct inode *target = dentry->d_inode;
3836 int error = may_delete(mnt, dir, dentry, 0);
3841 if (!dir->i_op->unlink)
3844 mutex_lock(&target->i_mutex);
3845 if (is_local_mountpoint(dentry))
3848 error = security_inode_unlink(dir, dentry);
3850 error = try_break_deleg(target, delegated_inode);
3853 error = dir->i_op->unlink(dir, dentry);
3856 detach_mounts(dentry);
3861 mutex_unlock(&target->i_mutex);
3863 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3864 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3865 fsnotify_link_count(target);
3871 EXPORT_SYMBOL(vfs_unlink2);
3873 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3875 return vfs_unlink2(NULL, dir, dentry, delegated_inode);
3877 EXPORT_SYMBOL(vfs_unlink);
3880 * Make sure that the actual truncation of the file will occur outside its
3881 * directory's i_mutex. Truncate can take a long time if there is a lot of
3882 * writeout happening, and we don't want to prevent access to the directory
3883 * while waiting on the I/O.
3885 static long do_unlinkat(int dfd, const char __user *pathname)
3888 struct filename *name;
3889 struct dentry *dentry;
3893 struct inode *inode = NULL;
3894 struct inode *delegated_inode = NULL;
3895 unsigned int lookup_flags = 0;
3897 name = user_path_parent(dfd, pathname,
3898 &path, &last, &type, lookup_flags);
3900 return PTR_ERR(name);
3903 if (type != LAST_NORM)
3906 error = mnt_want_write(path.mnt);
3910 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3911 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3912 error = PTR_ERR(dentry);
3913 if (!IS_ERR(dentry)) {
3914 /* Why not before? Because we want correct error value */
3915 if (last.name[last.len])
3917 inode = dentry->d_inode;
3918 if (d_is_negative(dentry))
3921 error = security_path_unlink(&path, dentry);
3924 error = vfs_unlink2(path.mnt, path.dentry->d_inode, dentry, &delegated_inode);
3928 mutex_unlock(&path.dentry->d_inode->i_mutex);
3930 iput(inode); /* truncate the inode here */
3932 if (delegated_inode) {
3933 error = break_deleg_wait(&delegated_inode);
3937 mnt_drop_write(path.mnt);
3941 if (retry_estale(error, lookup_flags)) {
3942 lookup_flags |= LOOKUP_REVAL;
3949 if (d_is_negative(dentry))
3951 else if (d_is_dir(dentry))
3958 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3960 if ((flag & ~AT_REMOVEDIR) != 0)
3963 if (flag & AT_REMOVEDIR)
3964 return do_rmdir(dfd, pathname);
3966 return do_unlinkat(dfd, pathname);
3969 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3971 return do_unlinkat(AT_FDCWD, pathname);
3974 int vfs_symlink2(struct vfsmount *mnt, struct inode *dir, struct dentry *dentry, const char *oldname)
3976 int error = may_create(mnt, dir, dentry);
3981 if (!dir->i_op->symlink)
3984 error = security_inode_symlink(dir, dentry, oldname);
3988 error = dir->i_op->symlink(dir, dentry, oldname);
3990 fsnotify_create(dir, dentry);
3993 EXPORT_SYMBOL(vfs_symlink2);
3995 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3997 return vfs_symlink2(NULL, dir, dentry, oldname);
3999 EXPORT_SYMBOL(vfs_symlink);
4001 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4002 int, newdfd, const char __user *, newname)
4005 struct filename *from;
4006 struct dentry *dentry;
4008 unsigned int lookup_flags = 0;
4010 from = getname(oldname);
4012 return PTR_ERR(from);
4014 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4015 error = PTR_ERR(dentry);
4019 error = security_path_symlink(&path, dentry, from->name);
4021 error = vfs_symlink2(path.mnt, path.dentry->d_inode, dentry, from->name);
4022 done_path_create(&path, dentry);
4023 if (retry_estale(error, lookup_flags)) {
4024 lookup_flags |= LOOKUP_REVAL;
4032 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4034 return sys_symlinkat(oldname, AT_FDCWD, newname);
4038 * vfs_link - create a new link
4039 * @old_dentry: object to be linked
4041 * @new_dentry: where to create the new link
4042 * @delegated_inode: returns inode needing a delegation break
4044 * The caller must hold dir->i_mutex
4046 * If vfs_link discovers a delegation on the to-be-linked file in need
4047 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4048 * inode in delegated_inode. The caller should then break the delegation
4049 * and retry. Because breaking a delegation may take a long time, the
4050 * caller should drop the i_mutex before doing so.
4052 * Alternatively, a caller may pass NULL for delegated_inode. This may
4053 * be appropriate for callers that expect the underlying filesystem not
4054 * to be NFS exported.
4056 int vfs_link2(struct vfsmount *mnt, struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4058 struct inode *inode = old_dentry->d_inode;
4059 unsigned max_links = dir->i_sb->s_max_links;
4065 error = may_create(mnt, dir, new_dentry);
4069 if (dir->i_sb != inode->i_sb)
4073 * A link to an append-only or immutable file cannot be created.
4075 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4077 if (!dir->i_op->link)
4079 if (S_ISDIR(inode->i_mode))
4082 error = security_inode_link(old_dentry, dir, new_dentry);
4086 mutex_lock(&inode->i_mutex);
4087 /* Make sure we don't allow creating hardlink to an unlinked file */
4088 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4090 else if (max_links && inode->i_nlink >= max_links)
4093 error = try_break_deleg(inode, delegated_inode);
4095 error = dir->i_op->link(old_dentry, dir, new_dentry);
4098 if (!error && (inode->i_state & I_LINKABLE)) {
4099 spin_lock(&inode->i_lock);
4100 inode->i_state &= ~I_LINKABLE;
4101 spin_unlock(&inode->i_lock);
4103 mutex_unlock(&inode->i_mutex);
4105 fsnotify_link(dir, inode, new_dentry);
4108 EXPORT_SYMBOL(vfs_link2);
4110 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4112 return vfs_link2(NULL, old_dentry, dir, new_dentry, delegated_inode);
4114 EXPORT_SYMBOL(vfs_link);
4117 * Hardlinks are often used in delicate situations. We avoid
4118 * security-related surprises by not following symlinks on the
4121 * We don't follow them on the oldname either to be compatible
4122 * with linux 2.0, and to avoid hard-linking to directories
4123 * and other special files. --ADM
4125 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4126 int, newdfd, const char __user *, newname, int, flags)
4128 struct dentry *new_dentry;
4129 struct path old_path, new_path;
4130 struct inode *delegated_inode = NULL;
4134 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4137 * To use null names we require CAP_DAC_READ_SEARCH
4138 * This ensures that not everyone will be able to create
4139 * handlink using the passed filedescriptor.
4141 if (flags & AT_EMPTY_PATH) {
4142 if (!capable(CAP_DAC_READ_SEARCH))
4147 if (flags & AT_SYMLINK_FOLLOW)
4148 how |= LOOKUP_FOLLOW;
4150 error = user_path_at(olddfd, oldname, how, &old_path);
4154 new_dentry = user_path_create(newdfd, newname, &new_path,
4155 (how & LOOKUP_REVAL));
4156 error = PTR_ERR(new_dentry);
4157 if (IS_ERR(new_dentry))
4161 if (old_path.mnt != new_path.mnt)
4163 error = may_linkat(&old_path);
4164 if (unlikely(error))
4166 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4169 error = vfs_link2(old_path.mnt, old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4171 done_path_create(&new_path, new_dentry);
4172 if (delegated_inode) {
4173 error = break_deleg_wait(&delegated_inode);
4175 path_put(&old_path);
4179 if (retry_estale(error, how)) {
4180 path_put(&old_path);
4181 how |= LOOKUP_REVAL;
4185 path_put(&old_path);
4190 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4192 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4196 * vfs_rename - rename a filesystem object
4197 * @old_dir: parent of source
4198 * @old_dentry: source
4199 * @new_dir: parent of destination
4200 * @new_dentry: destination
4201 * @delegated_inode: returns an inode needing a delegation break
4202 * @flags: rename flags
4204 * The caller must hold multiple mutexes--see lock_rename()).
4206 * If vfs_rename discovers a delegation in need of breaking at either
4207 * the source or destination, it will return -EWOULDBLOCK and return a
4208 * reference to the inode in delegated_inode. The caller should then
4209 * break the delegation and retry. Because breaking a delegation may
4210 * take a long time, the caller should drop all locks before doing
4213 * Alternatively, a caller may pass NULL for delegated_inode. This may
4214 * be appropriate for callers that expect the underlying filesystem not
4215 * to be NFS exported.
4217 * The worst of all namespace operations - renaming directory. "Perverted"
4218 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4220 * a) we can get into loop creation.
4221 * b) race potential - two innocent renames can create a loop together.
4222 * That's where 4.4 screws up. Current fix: serialization on
4223 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4225 * c) we have to lock _four_ objects - parents and victim (if it exists),
4226 * and source (if it is not a directory).
4227 * And that - after we got ->i_mutex on parents (until then we don't know
4228 * whether the target exists). Solution: try to be smart with locking
4229 * order for inodes. We rely on the fact that tree topology may change
4230 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4231 * move will be locked. Thus we can rank directories by the tree
4232 * (ancestors first) and rank all non-directories after them.
4233 * That works since everybody except rename does "lock parent, lookup,
4234 * lock child" and rename is under ->s_vfs_rename_mutex.
4235 * HOWEVER, it relies on the assumption that any object with ->lookup()
4236 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4237 * we'd better make sure that there's no link(2) for them.
4238 * d) conversion from fhandle to dentry may come in the wrong moment - when
4239 * we are removing the target. Solution: we will have to grab ->i_mutex
4240 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4241 * ->i_mutex on parents, which works but leads to some truly excessive
4244 int vfs_rename2(struct vfsmount *mnt,
4245 struct inode *old_dir, struct dentry *old_dentry,
4246 struct inode *new_dir, struct dentry *new_dentry,
4247 struct inode **delegated_inode, unsigned int flags)
4250 bool is_dir = d_is_dir(old_dentry);
4251 const unsigned char *old_name;
4252 struct inode *source = old_dentry->d_inode;
4253 struct inode *target = new_dentry->d_inode;
4254 bool new_is_dir = false;
4255 unsigned max_links = new_dir->i_sb->s_max_links;
4258 * Check source == target.
4259 * On overlayfs need to look at underlying inodes.
4261 if (vfs_select_inode(old_dentry, 0) == vfs_select_inode(new_dentry, 0))
4264 error = may_delete(mnt, old_dir, old_dentry, is_dir);
4269 error = may_create(mnt, new_dir, new_dentry);
4271 new_is_dir = d_is_dir(new_dentry);
4273 if (!(flags & RENAME_EXCHANGE))
4274 error = may_delete(mnt, new_dir, new_dentry, is_dir);
4276 error = may_delete(mnt, new_dir, new_dentry, new_is_dir);
4281 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4284 if (flags && !old_dir->i_op->rename2)
4288 * If we are going to change the parent - check write permissions,
4289 * we'll need to flip '..'.
4291 if (new_dir != old_dir) {
4293 error = inode_permission2(mnt, source, MAY_WRITE);
4297 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4298 error = inode_permission2(mnt, target, MAY_WRITE);
4304 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4309 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4311 if (!is_dir || (flags & RENAME_EXCHANGE))
4312 lock_two_nondirectories(source, target);
4314 mutex_lock(&target->i_mutex);
4317 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4320 if (max_links && new_dir != old_dir) {
4322 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4324 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4325 old_dir->i_nlink >= max_links)
4328 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4329 shrink_dcache_parent(new_dentry);
4331 error = try_break_deleg(source, delegated_inode);
4335 if (target && !new_is_dir) {
4336 error = try_break_deleg(target, delegated_inode);
4340 if (!old_dir->i_op->rename2) {
4341 error = old_dir->i_op->rename(old_dir, old_dentry,
4342 new_dir, new_dentry);
4344 WARN_ON(old_dir->i_op->rename != NULL);
4345 error = old_dir->i_op->rename2(old_dir, old_dentry,
4346 new_dir, new_dentry, flags);
4351 if (!(flags & RENAME_EXCHANGE) && target) {
4353 target->i_flags |= S_DEAD;
4354 dont_mount(new_dentry);
4355 detach_mounts(new_dentry);
4357 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4358 if (!(flags & RENAME_EXCHANGE))
4359 d_move(old_dentry, new_dentry);
4361 d_exchange(old_dentry, new_dentry);
4364 if (!is_dir || (flags & RENAME_EXCHANGE))
4365 unlock_two_nondirectories(source, target);
4367 mutex_unlock(&target->i_mutex);
4370 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4371 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4372 if (flags & RENAME_EXCHANGE) {
4373 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4374 new_is_dir, NULL, new_dentry);
4377 fsnotify_oldname_free(old_name);
4381 EXPORT_SYMBOL(vfs_rename2);
4383 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4384 struct inode *new_dir, struct dentry *new_dentry,
4385 struct inode **delegated_inode, unsigned int flags)
4387 return vfs_rename2(NULL, old_dir, old_dentry, new_dir, new_dentry, delegated_inode, flags);
4389 EXPORT_SYMBOL(vfs_rename);
4391 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4392 int, newdfd, const char __user *, newname, unsigned int, flags)
4394 struct dentry *old_dentry, *new_dentry;
4395 struct dentry *trap;
4396 struct path old_path, new_path;
4397 struct qstr old_last, new_last;
4398 int old_type, new_type;
4399 struct inode *delegated_inode = NULL;
4400 struct filename *from;
4401 struct filename *to;
4402 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4403 bool should_retry = false;
4406 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4409 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4410 (flags & RENAME_EXCHANGE))
4413 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4416 if (flags & RENAME_EXCHANGE)
4420 from = user_path_parent(olddfd, oldname,
4421 &old_path, &old_last, &old_type, lookup_flags);
4423 error = PTR_ERR(from);
4427 to = user_path_parent(newdfd, newname,
4428 &new_path, &new_last, &new_type, lookup_flags);
4430 error = PTR_ERR(to);
4435 if (old_path.mnt != new_path.mnt)
4439 if (old_type != LAST_NORM)
4442 if (flags & RENAME_NOREPLACE)
4444 if (new_type != LAST_NORM)
4447 error = mnt_want_write(old_path.mnt);
4452 trap = lock_rename(new_path.dentry, old_path.dentry);
4454 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4455 error = PTR_ERR(old_dentry);
4456 if (IS_ERR(old_dentry))
4458 /* source must exist */
4460 if (d_is_negative(old_dentry))
4462 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4463 error = PTR_ERR(new_dentry);
4464 if (IS_ERR(new_dentry))
4467 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4469 if (flags & RENAME_EXCHANGE) {
4471 if (d_is_negative(new_dentry))
4474 if (!d_is_dir(new_dentry)) {
4476 if (new_last.name[new_last.len])
4480 /* unless the source is a directory trailing slashes give -ENOTDIR */
4481 if (!d_is_dir(old_dentry)) {
4483 if (old_last.name[old_last.len])
4485 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4488 /* source should not be ancestor of target */
4490 if (old_dentry == trap)
4492 /* target should not be an ancestor of source */
4493 if (!(flags & RENAME_EXCHANGE))
4495 if (new_dentry == trap)
4498 error = security_path_rename(&old_path, old_dentry,
4499 &new_path, new_dentry, flags);
4502 error = vfs_rename2(old_path.mnt, old_path.dentry->d_inode, old_dentry,
4503 new_path.dentry->d_inode, new_dentry,
4504 &delegated_inode, flags);
4510 unlock_rename(new_path.dentry, old_path.dentry);
4511 if (delegated_inode) {
4512 error = break_deleg_wait(&delegated_inode);
4516 mnt_drop_write(old_path.mnt);
4518 if (retry_estale(error, lookup_flags))
4519 should_retry = true;
4520 path_put(&new_path);
4523 path_put(&old_path);
4526 should_retry = false;
4527 lookup_flags |= LOOKUP_REVAL;
4534 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4535 int, newdfd, const char __user *, newname)
4537 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4540 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4542 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4545 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4547 int error = may_create(NULL, dir, dentry);
4551 if (!dir->i_op->mknod)
4554 return dir->i_op->mknod(dir, dentry,
4555 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4557 EXPORT_SYMBOL(vfs_whiteout);
4559 int readlink_copy(char __user *buffer, int buflen, const char *link)
4561 int len = PTR_ERR(link);
4566 if (len > (unsigned) buflen)
4568 if (copy_to_user(buffer, link, len))
4573 EXPORT_SYMBOL(readlink_copy);
4576 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4577 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4578 * using) it for any given inode is up to filesystem.
4580 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4583 struct inode *inode = d_inode(dentry);
4584 const char *link = inode->i_link;
4588 link = inode->i_op->follow_link(dentry, &cookie);
4590 return PTR_ERR(link);
4592 res = readlink_copy(buffer, buflen, link);
4593 if (inode->i_op->put_link)
4594 inode->i_op->put_link(inode, cookie);
4597 EXPORT_SYMBOL(generic_readlink);
4599 /* get the link contents into pagecache */
4600 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4604 struct address_space *mapping = dentry->d_inode->i_mapping;
4605 page = read_mapping_page(mapping, 0, NULL);
4610 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4614 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4616 struct page *page = NULL;
4617 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4620 page_cache_release(page);
4624 EXPORT_SYMBOL(page_readlink);
4626 const char *page_follow_link_light(struct dentry *dentry, void **cookie)
4628 struct page *page = NULL;
4629 char *res = page_getlink(dentry, &page);
4634 EXPORT_SYMBOL(page_follow_link_light);
4636 void page_put_link(struct inode *unused, void *cookie)
4638 struct page *page = cookie;
4640 page_cache_release(page);
4642 EXPORT_SYMBOL(page_put_link);
4645 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4647 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4649 struct address_space *mapping = inode->i_mapping;
4654 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4656 flags |= AOP_FLAG_NOFS;
4659 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4660 flags, &page, &fsdata);
4664 kaddr = kmap_atomic(page);
4665 memcpy(kaddr, symname, len-1);
4666 kunmap_atomic(kaddr);
4668 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4675 mark_inode_dirty(inode);
4680 EXPORT_SYMBOL(__page_symlink);
4682 int page_symlink(struct inode *inode, const char *symname, int len)
4684 return __page_symlink(inode, symname, len,
4685 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4687 EXPORT_SYMBOL(page_symlink);
4689 const struct inode_operations page_symlink_inode_operations = {
4690 .readlink = generic_readlink,
4691 .follow_link = page_follow_link_light,
4692 .put_link = page_put_link,
4694 EXPORT_SYMBOL(page_symlink_inode_operations);