4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125 getname_flags(const char __user *filename, int flags, int *empty)
127 struct filename *result;
131 result = audit_reusename(filename);
135 result = __getname();
136 if (unlikely(!result))
137 return ERR_PTR(-ENOMEM);
140 * First, try to embed the struct filename inside the names_cache
143 kname = (char *)result->iname;
144 result->name = kname;
146 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
147 if (unlikely(len < 0)) {
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
158 if (unlikely(len == EMBEDDED_NAME_MAX)) {
159 const size_t size = offsetof(struct filename, iname[1]);
160 kname = (char *)result;
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
167 result = kzalloc(size, GFP_KERNEL);
168 if (unlikely(!result)) {
170 return ERR_PTR(-ENOMEM);
172 result->name = kname;
173 len = strncpy_from_user(kname, filename, PATH_MAX);
174 if (unlikely(len < 0)) {
179 if (unlikely(len == PATH_MAX)) {
182 return ERR_PTR(-ENAMETOOLONG);
187 /* The empty path is special. */
188 if (unlikely(!len)) {
191 if (!(flags & LOOKUP_EMPTY)) {
193 return ERR_PTR(-ENOENT);
197 result->uptr = filename;
198 result->aname = NULL;
199 audit_getname(result);
204 getname(const char __user * filename)
206 return getname_flags(filename, 0, NULL);
210 getname_kernel(const char * filename)
212 struct filename *result;
213 int len = strlen(filename) + 1;
215 result = __getname();
216 if (unlikely(!result))
217 return ERR_PTR(-ENOMEM);
219 if (len <= EMBEDDED_NAME_MAX) {
220 result->name = (char *)result->iname;
221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
227 return ERR_PTR(-ENOMEM);
229 tmp->name = (char *)result;
233 return ERR_PTR(-ENAMETOOLONG);
235 memcpy((char *)result->name, filename, len);
237 result->aname = NULL;
239 audit_getname(result);
244 void putname(struct filename *name)
246 BUG_ON(name->refcnt <= 0);
248 if (--name->refcnt > 0)
251 if (name->name != name->iname) {
252 __putname(name->name);
258 static int check_acl(struct inode *inode, int mask)
260 #ifdef CONFIG_FS_POSIX_ACL
261 struct posix_acl *acl;
263 if (mask & MAY_NOT_BLOCK) {
264 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 /* no ->get_acl() calls in RCU mode... */
268 if (acl == ACL_NOT_CACHED)
270 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273 acl = get_acl(inode, ACL_TYPE_ACCESS);
277 int error = posix_acl_permission(inode, acl, mask);
278 posix_acl_release(acl);
287 * This does the basic permission checking
289 static int acl_permission_check(struct inode *inode, int mask)
291 unsigned int mode = inode->i_mode;
293 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
297 int error = check_acl(inode, mask);
298 if (error != -EAGAIN)
302 if (in_group_p(inode->i_gid))
307 * If the DACs are ok we don't need any capability check.
309 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
315 * generic_permission - check for access rights on a Posix-like filesystem
316 * @inode: inode to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
322 * are used for other things.
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
328 int generic_permission(struct inode *inode, int mask)
333 * Do the basic permission checks.
335 ret = acl_permission_check(inode, mask);
339 if (S_ISDIR(inode->i_mode)) {
340 /* DACs are overridable for directories */
341 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
343 if (!(mask & MAY_WRITE))
344 if (capable_wrt_inode_uidgid(inode,
345 CAP_DAC_READ_SEARCH))
350 * Read/write DACs are always overridable.
351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
354 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
355 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
359 * Searching includes executable on directories, else just read.
361 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
362 if (mask == MAY_READ)
363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
368 EXPORT_SYMBOL(generic_permission);
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
376 static inline int do_inode_permission(struct inode *inode, int mask)
378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379 if (likely(inode->i_op->permission))
380 return inode->i_op->permission(inode, mask);
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode->i_lock);
384 inode->i_opflags |= IOP_FASTPERM;
385 spin_unlock(&inode->i_lock);
387 return generic_permission(inode, mask);
391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395 * Check for read/write/execute permissions on an inode.
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
402 int __inode_permission(struct inode *inode, int mask)
406 if (unlikely(mask & MAY_WRITE)) {
408 * Nobody gets write access to an immutable file.
410 if (IS_IMMUTABLE(inode))
414 retval = do_inode_permission(inode, mask);
418 retval = devcgroup_inode_permission(inode, mask);
422 return security_inode_permission(inode, mask);
424 EXPORT_SYMBOL(__inode_permission);
427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Separate out file-system wide checks from inode-specific permission checks.
434 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
436 if (unlikely(mask & MAY_WRITE)) {
437 umode_t mode = inode->i_mode;
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb->s_flags & MS_RDONLY) &&
441 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 int inode_permission(struct inode *inode, int mask)
462 retval = sb_permission(inode->i_sb, inode, mask);
465 return __inode_permission(inode, mask);
467 EXPORT_SYMBOL(inode_permission);
470 * path_get - get a reference to a path
471 * @path: path to get the reference to
473 * Given a path increment the reference count to the dentry and the vfsmount.
475 void path_get(const struct path *path)
480 EXPORT_SYMBOL(path_get);
483 * path_put - put a reference to a path
484 * @path: path to put the reference to
486 * Given a path decrement the reference count to the dentry and the vfsmount.
488 void path_put(const struct path *path)
493 EXPORT_SYMBOL(path_put);
495 #define EMBEDDED_LEVELS 2
500 struct inode *inode; /* path.dentry.d_inode */
502 unsigned seq, m_seq, root_seq;
505 int total_link_count;
512 } *stack, internal[EMBEDDED_LEVELS];
515 static struct nameidata *set_nameidata(struct nameidata *p)
517 struct nameidata *old = current->nameidata;
518 p->stack = p->internal;
519 p->total_link_count = old ? old->total_link_count : 0;
520 current->nameidata = p;
524 static void restore_nameidata(struct nameidata *old)
526 struct nameidata *now = current->nameidata;
528 current->nameidata = old;
530 old->total_link_count = now->total_link_count;
531 if (now->stack != now->internal) {
533 now->stack = now->internal;
537 static int __nd_alloc_stack(struct nameidata *nd)
541 if (nd->flags & LOOKUP_RCU) {
542 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
547 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
552 memcpy(p, nd->internal, sizeof(nd->internal));
557 static inline int nd_alloc_stack(struct nameidata *nd)
559 if (likely(nd->depth != EMBEDDED_LEVELS))
561 if (likely(nd->stack != nd->internal))
563 return __nd_alloc_stack(nd);
566 static void drop_links(struct nameidata *nd)
570 struct saved *last = nd->stack + i;
571 struct inode *inode = last->inode;
572 if (last->cookie && inode->i_op->put_link) {
573 inode->i_op->put_link(inode, last->cookie);
579 static void terminate_walk(struct nameidata *nd)
582 if (!(nd->flags & LOOKUP_RCU)) {
585 for (i = 0; i < nd->depth; i++)
586 path_put(&nd->stack[i].link);
588 nd->flags &= ~LOOKUP_RCU;
589 if (!(nd->flags & LOOKUP_ROOT))
596 /* path_put is needed afterwards regardless of success or failure */
597 static bool legitimize_path(struct nameidata *nd,
598 struct path *path, unsigned seq)
600 int res = __legitimize_mnt(path->mnt, nd->m_seq);
607 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
611 return !read_seqcount_retry(&path->dentry->d_seq, seq);
614 static bool legitimize_links(struct nameidata *nd)
617 for (i = 0; i < nd->depth; i++) {
618 struct saved *last = nd->stack + i;
619 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
629 * Path walking has 2 modes, rcu-walk and ref-walk (see
630 * Documentation/filesystems/path-lookup.txt). In situations when we can't
631 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
632 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
633 * mode. Refcounts are grabbed at the last known good point before rcu-walk
634 * got stuck, so ref-walk may continue from there. If this is not successful
635 * (eg. a seqcount has changed), then failure is returned and it's up to caller
636 * to restart the path walk from the beginning in ref-walk mode.
640 * unlazy_walk - try to switch to ref-walk mode.
641 * @nd: nameidata pathwalk data
642 * @dentry: child of nd->path.dentry or NULL
643 * @seq: seq number to check dentry against
644 * Returns: 0 on success, -ECHILD on failure
646 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
647 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
648 * @nd or NULL. Must be called from rcu-walk context.
649 * Nothing should touch nameidata between unlazy_walk() failure and
652 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
654 struct dentry *parent = nd->path.dentry;
656 BUG_ON(!(nd->flags & LOOKUP_RCU));
658 nd->flags &= ~LOOKUP_RCU;
659 if (unlikely(!legitimize_links(nd)))
661 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
663 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
667 * For a negative lookup, the lookup sequence point is the parents
668 * sequence point, and it only needs to revalidate the parent dentry.
670 * For a positive lookup, we need to move both the parent and the
671 * dentry from the RCU domain to be properly refcounted. And the
672 * sequence number in the dentry validates *both* dentry counters,
673 * since we checked the sequence number of the parent after we got
674 * the child sequence number. So we know the parent must still
675 * be valid if the child sequence number is still valid.
678 if (read_seqcount_retry(&parent->d_seq, nd->seq))
680 BUG_ON(nd->inode != parent->d_inode);
682 if (!lockref_get_not_dead(&dentry->d_lockref))
684 if (read_seqcount_retry(&dentry->d_seq, seq))
689 * Sequence counts matched. Now make sure that the root is
690 * still valid and get it if required.
692 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
693 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
710 nd->path.dentry = NULL;
714 if (!(nd->flags & LOOKUP_ROOT))
719 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
721 if (unlikely(!legitimize_path(nd, link, seq))) {
724 nd->flags &= ~LOOKUP_RCU;
726 nd->path.dentry = NULL;
727 if (!(nd->flags & LOOKUP_ROOT))
730 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
737 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
739 return dentry->d_op->d_revalidate(dentry, flags);
743 * complete_walk - successful completion of path walk
744 * @nd: pointer nameidata
746 * If we had been in RCU mode, drop out of it and legitimize nd->path.
747 * Revalidate the final result, unless we'd already done that during
748 * the path walk or the filesystem doesn't ask for it. Return 0 on
749 * success, -error on failure. In case of failure caller does not
750 * need to drop nd->path.
752 static int complete_walk(struct nameidata *nd)
754 struct dentry *dentry = nd->path.dentry;
757 if (nd->flags & LOOKUP_RCU) {
758 if (!(nd->flags & LOOKUP_ROOT))
760 if (unlikely(unlazy_walk(nd, NULL, 0)))
764 if (likely(!(nd->flags & LOOKUP_JUMPED)))
767 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
770 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
780 static void set_root(struct nameidata *nd)
782 get_fs_root(current->fs, &nd->root);
785 static unsigned set_root_rcu(struct nameidata *nd)
787 struct fs_struct *fs = current->fs;
791 seq = read_seqcount_begin(&fs->seq);
793 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
794 } while (read_seqcount_retry(&fs->seq, seq));
798 static void path_put_conditional(struct path *path, struct nameidata *nd)
801 if (path->mnt != nd->path.mnt)
805 static inline void path_to_nameidata(const struct path *path,
806 struct nameidata *nd)
808 if (!(nd->flags & LOOKUP_RCU)) {
809 dput(nd->path.dentry);
810 if (nd->path.mnt != path->mnt)
811 mntput(nd->path.mnt);
813 nd->path.mnt = path->mnt;
814 nd->path.dentry = path->dentry;
818 * Helper to directly jump to a known parsed path from ->follow_link,
819 * caller must have taken a reference to path beforehand.
821 void nd_jump_link(struct path *path)
823 struct nameidata *nd = current->nameidata;
827 nd->inode = nd->path.dentry->d_inode;
828 nd->flags |= LOOKUP_JUMPED;
831 static inline void put_link(struct nameidata *nd)
833 struct saved *last = nd->stack + --nd->depth;
834 struct inode *inode = last->inode;
835 if (last->cookie && inode->i_op->put_link)
836 inode->i_op->put_link(inode, last->cookie);
837 if (!(nd->flags & LOOKUP_RCU))
838 path_put(&last->link);
841 int sysctl_protected_symlinks __read_mostly = 0;
842 int sysctl_protected_hardlinks __read_mostly = 0;
845 * may_follow_link - Check symlink following for unsafe situations
846 * @nd: nameidata pathwalk data
848 * In the case of the sysctl_protected_symlinks sysctl being enabled,
849 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
850 * in a sticky world-writable directory. This is to protect privileged
851 * processes from failing races against path names that may change out
852 * from under them by way of other users creating malicious symlinks.
853 * It will permit symlinks to be followed only when outside a sticky
854 * world-writable directory, or when the uid of the symlink and follower
855 * match, or when the directory owner matches the symlink's owner.
857 * Returns 0 if following the symlink is allowed, -ve on error.
859 static inline int may_follow_link(struct nameidata *nd)
861 const struct inode *inode;
862 const struct inode *parent;
864 if (!sysctl_protected_symlinks)
867 /* Allowed if owner and follower match. */
868 inode = nd->stack[0].inode;
869 if (uid_eq(current_cred()->fsuid, inode->i_uid))
872 /* Allowed if parent directory not sticky and world-writable. */
873 parent = nd->path.dentry->d_inode;
874 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
877 /* Allowed if parent directory and link owner match. */
878 if (uid_eq(parent->i_uid, inode->i_uid))
881 if (nd->flags & LOOKUP_RCU)
884 audit_log_link_denied("follow_link", &nd->stack[0].link);
889 * safe_hardlink_source - Check for safe hardlink conditions
890 * @inode: the source inode to hardlink from
892 * Return false if at least one of the following conditions:
893 * - inode is not a regular file
895 * - inode is setgid and group-exec
896 * - access failure for read and write
898 * Otherwise returns true.
900 static bool safe_hardlink_source(struct inode *inode)
902 umode_t mode = inode->i_mode;
904 /* Special files should not get pinned to the filesystem. */
908 /* Setuid files should not get pinned to the filesystem. */
912 /* Executable setgid files should not get pinned to the filesystem. */
913 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
916 /* Hardlinking to unreadable or unwritable sources is dangerous. */
917 if (inode_permission(inode, MAY_READ | MAY_WRITE))
924 * may_linkat - Check permissions for creating a hardlink
925 * @link: the source to hardlink from
927 * Block hardlink when all of:
928 * - sysctl_protected_hardlinks enabled
929 * - fsuid does not match inode
930 * - hardlink source is unsafe (see safe_hardlink_source() above)
933 * Returns 0 if successful, -ve on error.
935 static int may_linkat(struct path *link)
937 const struct cred *cred;
940 if (!sysctl_protected_hardlinks)
943 cred = current_cred();
944 inode = link->dentry->d_inode;
946 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
947 * otherwise, it must be a safe source.
949 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
953 audit_log_link_denied("linkat", link);
957 static __always_inline
958 const char *get_link(struct nameidata *nd)
960 struct saved *last = nd->stack + nd->depth - 1;
961 struct dentry *dentry = last->link.dentry;
962 struct inode *inode = last->inode;
966 if (!(nd->flags & LOOKUP_RCU)) {
967 touch_atime(&last->link);
969 } else if (atime_needs_update(&last->link, inode)) {
970 if (unlikely(unlazy_walk(nd, NULL, 0)))
971 return ERR_PTR(-ECHILD);
972 touch_atime(&last->link);
975 error = security_inode_follow_link(dentry, inode,
976 nd->flags & LOOKUP_RCU);
978 return ERR_PTR(error);
980 nd->last_type = LAST_BIND;
983 if (nd->flags & LOOKUP_RCU) {
984 if (unlikely(unlazy_walk(nd, NULL, 0)))
985 return ERR_PTR(-ECHILD);
987 res = inode->i_op->follow_link(dentry, &last->cookie);
988 if (IS_ERR_OR_NULL(res)) {
994 if (nd->flags & LOOKUP_RCU) {
1000 nd->inode = d->d_inode;
1001 nd->seq = nd->root_seq;
1002 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
1003 return ERR_PTR(-ECHILD);
1007 path_put(&nd->path);
1008 nd->path = nd->root;
1009 path_get(&nd->root);
1010 nd->inode = nd->path.dentry->d_inode;
1012 nd->flags |= LOOKUP_JUMPED;
1013 while (unlikely(*++res == '/'))
1022 * follow_up - Find the mountpoint of path's vfsmount
1024 * Given a path, find the mountpoint of its source file system.
1025 * Replace @path with the path of the mountpoint in the parent mount.
1028 * Return 1 if we went up a level and 0 if we were already at the
1031 int follow_up(struct path *path)
1033 struct mount *mnt = real_mount(path->mnt);
1034 struct mount *parent;
1035 struct dentry *mountpoint;
1037 read_seqlock_excl(&mount_lock);
1038 parent = mnt->mnt_parent;
1039 if (parent == mnt) {
1040 read_sequnlock_excl(&mount_lock);
1043 mntget(&parent->mnt);
1044 mountpoint = dget(mnt->mnt_mountpoint);
1045 read_sequnlock_excl(&mount_lock);
1047 path->dentry = mountpoint;
1049 path->mnt = &parent->mnt;
1052 EXPORT_SYMBOL(follow_up);
1055 * Perform an automount
1056 * - return -EISDIR to tell follow_managed() to stop and return the path we
1059 static int follow_automount(struct path *path, struct nameidata *nd,
1062 struct vfsmount *mnt;
1065 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1068 /* We don't want to mount if someone's just doing a stat -
1069 * unless they're stat'ing a directory and appended a '/' to
1072 * We do, however, want to mount if someone wants to open or
1073 * create a file of any type under the mountpoint, wants to
1074 * traverse through the mountpoint or wants to open the
1075 * mounted directory. Also, autofs may mark negative dentries
1076 * as being automount points. These will need the attentions
1077 * of the daemon to instantiate them before they can be used.
1079 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1080 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1081 path->dentry->d_inode)
1084 nd->total_link_count++;
1085 if (nd->total_link_count >= 40)
1088 mnt = path->dentry->d_op->d_automount(path);
1091 * The filesystem is allowed to return -EISDIR here to indicate
1092 * it doesn't want to automount. For instance, autofs would do
1093 * this so that its userspace daemon can mount on this dentry.
1095 * However, we can only permit this if it's a terminal point in
1096 * the path being looked up; if it wasn't then the remainder of
1097 * the path is inaccessible and we should say so.
1099 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1101 return PTR_ERR(mnt);
1104 if (!mnt) /* mount collision */
1107 if (!*need_mntput) {
1108 /* lock_mount() may release path->mnt on error */
1110 *need_mntput = true;
1112 err = finish_automount(mnt, path);
1116 /* Someone else made a mount here whilst we were busy */
1121 path->dentry = dget(mnt->mnt_root);
1130 * Handle a dentry that is managed in some way.
1131 * - Flagged for transit management (autofs)
1132 * - Flagged as mountpoint
1133 * - Flagged as automount point
1135 * This may only be called in refwalk mode.
1137 * Serialization is taken care of in namespace.c
1139 static int follow_managed(struct path *path, struct nameidata *nd)
1141 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1143 bool need_mntput = false;
1146 /* Given that we're not holding a lock here, we retain the value in a
1147 * local variable for each dentry as we look at it so that we don't see
1148 * the components of that value change under us */
1149 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1150 managed &= DCACHE_MANAGED_DENTRY,
1151 unlikely(managed != 0)) {
1152 /* Allow the filesystem to manage the transit without i_mutex
1154 if (managed & DCACHE_MANAGE_TRANSIT) {
1155 BUG_ON(!path->dentry->d_op);
1156 BUG_ON(!path->dentry->d_op->d_manage);
1157 ret = path->dentry->d_op->d_manage(path->dentry, false);
1162 /* Transit to a mounted filesystem. */
1163 if (managed & DCACHE_MOUNTED) {
1164 struct vfsmount *mounted = lookup_mnt(path);
1169 path->mnt = mounted;
1170 path->dentry = dget(mounted->mnt_root);
1175 /* Something is mounted on this dentry in another
1176 * namespace and/or whatever was mounted there in this
1177 * namespace got unmounted before lookup_mnt() could
1181 /* Handle an automount point */
1182 if (managed & DCACHE_NEED_AUTOMOUNT) {
1183 ret = follow_automount(path, nd, &need_mntput);
1189 /* We didn't change the current path point */
1193 if (need_mntput && path->mnt == mnt)
1198 nd->flags |= LOOKUP_JUMPED;
1199 if (unlikely(ret < 0))
1200 path_put_conditional(path, nd);
1204 int follow_down_one(struct path *path)
1206 struct vfsmount *mounted;
1208 mounted = lookup_mnt(path);
1212 path->mnt = mounted;
1213 path->dentry = dget(mounted->mnt_root);
1218 EXPORT_SYMBOL(follow_down_one);
1220 static inline int managed_dentry_rcu(struct dentry *dentry)
1222 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1223 dentry->d_op->d_manage(dentry, true) : 0;
1227 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1228 * we meet a managed dentry that would need blocking.
1230 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1231 struct inode **inode, unsigned *seqp)
1234 struct mount *mounted;
1236 * Don't forget we might have a non-mountpoint managed dentry
1237 * that wants to block transit.
1239 switch (managed_dentry_rcu(path->dentry)) {
1249 if (!d_mountpoint(path->dentry))
1250 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1252 mounted = __lookup_mnt(path->mnt, path->dentry);
1255 path->mnt = &mounted->mnt;
1256 path->dentry = mounted->mnt.mnt_root;
1257 nd->flags |= LOOKUP_JUMPED;
1258 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1260 * Update the inode too. We don't need to re-check the
1261 * dentry sequence number here after this d_inode read,
1262 * because a mount-point is always pinned.
1264 *inode = path->dentry->d_inode;
1266 return !read_seqretry(&mount_lock, nd->m_seq) &&
1267 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1270 static int follow_dotdot_rcu(struct nameidata *nd)
1272 struct inode *inode = nd->inode;
1277 if (path_equal(&nd->path, &nd->root))
1279 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1280 struct dentry *old = nd->path.dentry;
1281 struct dentry *parent = old->d_parent;
1284 inode = parent->d_inode;
1285 seq = read_seqcount_begin(&parent->d_seq);
1286 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1288 nd->path.dentry = parent;
1292 struct mount *mnt = real_mount(nd->path.mnt);
1293 struct mount *mparent = mnt->mnt_parent;
1294 struct dentry *mountpoint = mnt->mnt_mountpoint;
1295 struct inode *inode2 = mountpoint->d_inode;
1296 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1297 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1299 if (&mparent->mnt == nd->path.mnt)
1301 /* we know that mountpoint was pinned */
1302 nd->path.dentry = mountpoint;
1303 nd->path.mnt = &mparent->mnt;
1308 while (unlikely(d_mountpoint(nd->path.dentry))) {
1309 struct mount *mounted;
1310 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1311 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1315 nd->path.mnt = &mounted->mnt;
1316 nd->path.dentry = mounted->mnt.mnt_root;
1317 inode = nd->path.dentry->d_inode;
1318 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1325 * Follow down to the covering mount currently visible to userspace. At each
1326 * point, the filesystem owning that dentry may be queried as to whether the
1327 * caller is permitted to proceed or not.
1329 int follow_down(struct path *path)
1334 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1335 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1336 /* Allow the filesystem to manage the transit without i_mutex
1339 * We indicate to the filesystem if someone is trying to mount
1340 * something here. This gives autofs the chance to deny anyone
1341 * other than its daemon the right to mount on its
1344 * The filesystem may sleep at this point.
1346 if (managed & DCACHE_MANAGE_TRANSIT) {
1347 BUG_ON(!path->dentry->d_op);
1348 BUG_ON(!path->dentry->d_op->d_manage);
1349 ret = path->dentry->d_op->d_manage(
1350 path->dentry, false);
1352 return ret == -EISDIR ? 0 : ret;
1355 /* Transit to a mounted filesystem. */
1356 if (managed & DCACHE_MOUNTED) {
1357 struct vfsmount *mounted = lookup_mnt(path);
1362 path->mnt = mounted;
1363 path->dentry = dget(mounted->mnt_root);
1367 /* Don't handle automount points here */
1372 EXPORT_SYMBOL(follow_down);
1375 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1377 static void follow_mount(struct path *path)
1379 while (d_mountpoint(path->dentry)) {
1380 struct vfsmount *mounted = lookup_mnt(path);
1385 path->mnt = mounted;
1386 path->dentry = dget(mounted->mnt_root);
1390 static void follow_dotdot(struct nameidata *nd)
1396 struct dentry *old = nd->path.dentry;
1398 if (nd->path.dentry == nd->root.dentry &&
1399 nd->path.mnt == nd->root.mnt) {
1402 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1403 /* rare case of legitimate dget_parent()... */
1404 nd->path.dentry = dget_parent(nd->path.dentry);
1408 if (!follow_up(&nd->path))
1411 follow_mount(&nd->path);
1412 nd->inode = nd->path.dentry->d_inode;
1416 * This looks up the name in dcache, possibly revalidates the old dentry and
1417 * allocates a new one if not found or not valid. In the need_lookup argument
1418 * returns whether i_op->lookup is necessary.
1420 * dir->d_inode->i_mutex must be held
1422 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1423 unsigned int flags, bool *need_lookup)
1425 struct dentry *dentry;
1428 *need_lookup = false;
1429 dentry = d_lookup(dir, name);
1431 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1432 error = d_revalidate(dentry, flags);
1433 if (unlikely(error <= 0)) {
1436 return ERR_PTR(error);
1438 d_invalidate(dentry);
1447 dentry = d_alloc(dir, name);
1448 if (unlikely(!dentry))
1449 return ERR_PTR(-ENOMEM);
1451 *need_lookup = true;
1457 * Call i_op->lookup on the dentry. The dentry must be negative and
1460 * dir->d_inode->i_mutex must be held
1462 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1467 /* Don't create child dentry for a dead directory. */
1468 if (unlikely(IS_DEADDIR(dir))) {
1470 return ERR_PTR(-ENOENT);
1473 old = dir->i_op->lookup(dir, dentry, flags);
1474 if (unlikely(old)) {
1481 static struct dentry *__lookup_hash(struct qstr *name,
1482 struct dentry *base, unsigned int flags)
1485 struct dentry *dentry;
1487 dentry = lookup_dcache(name, base, flags, &need_lookup);
1491 return lookup_real(base->d_inode, dentry, flags);
1495 * It's more convoluted than I'd like it to be, but... it's still fairly
1496 * small and for now I'd prefer to have fast path as straight as possible.
1497 * It _is_ time-critical.
1499 static int lookup_fast(struct nameidata *nd,
1500 struct path *path, struct inode **inode,
1503 struct vfsmount *mnt = nd->path.mnt;
1504 struct dentry *dentry, *parent = nd->path.dentry;
1510 * Rename seqlock is not required here because in the off chance
1511 * of a false negative due to a concurrent rename, we're going to
1512 * do the non-racy lookup, below.
1514 if (nd->flags & LOOKUP_RCU) {
1517 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1522 * This sequence count validates that the inode matches
1523 * the dentry name information from lookup.
1525 *inode = d_backing_inode(dentry);
1526 negative = d_is_negative(dentry);
1527 if (read_seqcount_retry(&dentry->d_seq, seq))
1533 * This sequence count validates that the parent had no
1534 * changes while we did the lookup of the dentry above.
1536 * The memory barrier in read_seqcount_begin of child is
1537 * enough, we can use __read_seqcount_retry here.
1539 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1543 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1544 status = d_revalidate(dentry, nd->flags);
1545 if (unlikely(status <= 0)) {
1546 if (status != -ECHILD)
1552 path->dentry = dentry;
1553 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1556 if (unlazy_walk(nd, dentry, seq))
1559 dentry = __d_lookup(parent, &nd->last);
1562 if (unlikely(!dentry))
1565 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1566 status = d_revalidate(dentry, nd->flags);
1567 if (unlikely(status <= 0)) {
1572 d_invalidate(dentry);
1577 if (unlikely(d_is_negative(dentry))) {
1582 path->dentry = dentry;
1583 err = follow_managed(path, nd);
1585 *inode = d_backing_inode(path->dentry);
1592 /* Fast lookup failed, do it the slow way */
1593 static int lookup_slow(struct nameidata *nd, struct path *path)
1595 struct dentry *dentry, *parent;
1597 parent = nd->path.dentry;
1598 BUG_ON(nd->inode != parent->d_inode);
1600 mutex_lock(&parent->d_inode->i_mutex);
1601 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1602 mutex_unlock(&parent->d_inode->i_mutex);
1604 return PTR_ERR(dentry);
1605 path->mnt = nd->path.mnt;
1606 path->dentry = dentry;
1607 return follow_managed(path, nd);
1610 static inline int may_lookup(struct nameidata *nd)
1612 if (nd->flags & LOOKUP_RCU) {
1613 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1616 if (unlazy_walk(nd, NULL, 0))
1619 return inode_permission(nd->inode, MAY_EXEC);
1622 static inline int handle_dots(struct nameidata *nd, int type)
1624 if (type == LAST_DOTDOT) {
1625 if (nd->flags & LOOKUP_RCU) {
1626 return follow_dotdot_rcu(nd);
1633 static int pick_link(struct nameidata *nd, struct path *link,
1634 struct inode *inode, unsigned seq)
1638 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1639 path_to_nameidata(link, nd);
1642 if (!(nd->flags & LOOKUP_RCU)) {
1643 if (link->mnt == nd->path.mnt)
1646 error = nd_alloc_stack(nd);
1647 if (unlikely(error)) {
1648 if (error == -ECHILD) {
1649 if (unlikely(unlazy_link(nd, link, seq)))
1651 error = nd_alloc_stack(nd);
1659 last = nd->stack + nd->depth++;
1661 last->cookie = NULL;
1662 last->inode = inode;
1668 * Do we need to follow links? We _really_ want to be able
1669 * to do this check without having to look at inode->i_op,
1670 * so we keep a cache of "no, this doesn't need follow_link"
1671 * for the common case.
1673 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1675 struct inode *inode, unsigned seq)
1677 if (likely(!d_is_symlink(link->dentry)))
1681 return pick_link(nd, link, inode, seq);
1684 enum {WALK_GET = 1, WALK_PUT = 2};
1686 static int walk_component(struct nameidata *nd, int flags)
1689 struct inode *inode;
1693 * "." and ".." are special - ".." especially so because it has
1694 * to be able to know about the current root directory and
1695 * parent relationships.
1697 if (unlikely(nd->last_type != LAST_NORM)) {
1698 err = handle_dots(nd, nd->last_type);
1699 if (flags & WALK_PUT)
1703 err = lookup_fast(nd, &path, &inode, &seq);
1704 if (unlikely(err)) {
1708 err = lookup_slow(nd, &path);
1712 inode = d_backing_inode(path.dentry);
1713 seq = 0; /* we are already out of RCU mode */
1715 if (d_is_negative(path.dentry))
1719 if (flags & WALK_PUT)
1721 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1724 path_to_nameidata(&path, nd);
1730 path_to_nameidata(&path, nd);
1735 * We can do the critical dentry name comparison and hashing
1736 * operations one word at a time, but we are limited to:
1738 * - Architectures with fast unaligned word accesses. We could
1739 * do a "get_unaligned()" if this helps and is sufficiently
1742 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1743 * do not trap on the (extremely unlikely) case of a page
1744 * crossing operation.
1746 * - Furthermore, we need an efficient 64-bit compile for the
1747 * 64-bit case in order to generate the "number of bytes in
1748 * the final mask". Again, that could be replaced with a
1749 * efficient population count instruction or similar.
1751 #ifdef CONFIG_DCACHE_WORD_ACCESS
1753 #include <asm/word-at-a-time.h>
1757 static inline unsigned int fold_hash(unsigned long hash)
1759 return hash_64(hash, 32);
1762 #else /* 32-bit case */
1764 #define fold_hash(x) (x)
1768 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1770 unsigned long a, mask;
1771 unsigned long hash = 0;
1774 a = load_unaligned_zeropad(name);
1775 if (len < sizeof(unsigned long))
1779 name += sizeof(unsigned long);
1780 len -= sizeof(unsigned long);
1784 mask = bytemask_from_count(len);
1787 return fold_hash(hash);
1789 EXPORT_SYMBOL(full_name_hash);
1792 * Calculate the length and hash of the path component, and
1793 * return the "hash_len" as the result.
1795 static inline u64 hash_name(const char *name)
1797 unsigned long a, b, adata, bdata, mask, hash, len;
1798 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1801 len = -sizeof(unsigned long);
1803 hash = (hash + a) * 9;
1804 len += sizeof(unsigned long);
1805 a = load_unaligned_zeropad(name+len);
1806 b = a ^ REPEAT_BYTE('/');
1807 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1809 adata = prep_zero_mask(a, adata, &constants);
1810 bdata = prep_zero_mask(b, bdata, &constants);
1812 mask = create_zero_mask(adata | bdata);
1814 hash += a & zero_bytemask(mask);
1815 len += find_zero(mask);
1816 return hashlen_create(fold_hash(hash), len);
1821 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1823 unsigned long hash = init_name_hash();
1825 hash = partial_name_hash(*name++, hash);
1826 return end_name_hash(hash);
1828 EXPORT_SYMBOL(full_name_hash);
1831 * We know there's a real path component here of at least
1834 static inline u64 hash_name(const char *name)
1836 unsigned long hash = init_name_hash();
1837 unsigned long len = 0, c;
1839 c = (unsigned char)*name;
1842 hash = partial_name_hash(c, hash);
1843 c = (unsigned char)name[len];
1844 } while (c && c != '/');
1845 return hashlen_create(end_name_hash(hash), len);
1852 * This is the basic name resolution function, turning a pathname into
1853 * the final dentry. We expect 'base' to be positive and a directory.
1855 * Returns 0 and nd will have valid dentry and mnt on success.
1856 * Returns error and drops reference to input namei data on failure.
1858 static int link_path_walk(const char *name, struct nameidata *nd)
1867 /* At this point we know we have a real path component. */
1872 err = may_lookup(nd);
1876 hash_len = hash_name(name);
1879 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1881 if (name[1] == '.') {
1883 nd->flags |= LOOKUP_JUMPED;
1889 if (likely(type == LAST_NORM)) {
1890 struct dentry *parent = nd->path.dentry;
1891 nd->flags &= ~LOOKUP_JUMPED;
1892 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1893 struct qstr this = { { .hash_len = hash_len }, .name = name };
1894 err = parent->d_op->d_hash(parent, &this);
1897 hash_len = this.hash_len;
1902 nd->last.hash_len = hash_len;
1903 nd->last.name = name;
1904 nd->last_type = type;
1906 name += hashlen_len(hash_len);
1910 * If it wasn't NUL, we know it was '/'. Skip that
1911 * slash, and continue until no more slashes.
1915 } while (unlikely(*name == '/'));
1916 if (unlikely(!*name)) {
1918 /* pathname body, done */
1921 name = nd->stack[nd->depth - 1].name;
1922 /* trailing symlink, done */
1925 /* last component of nested symlink */
1926 err = walk_component(nd, WALK_GET | WALK_PUT);
1928 err = walk_component(nd, WALK_GET);
1934 const char *s = get_link(nd);
1936 if (unlikely(IS_ERR(s)))
1943 nd->stack[nd->depth - 1].name = name;
1948 if (unlikely(!d_can_lookup(nd->path.dentry)))
1953 static const char *path_init(int dfd, const struct filename *name,
1954 unsigned int flags, struct nameidata *nd)
1957 const char *s = name->name;
1959 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1960 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1962 nd->total_link_count = 0;
1963 if (flags & LOOKUP_ROOT) {
1964 struct dentry *root = nd->root.dentry;
1965 struct inode *inode = root->d_inode;
1967 if (!d_can_lookup(root))
1968 return ERR_PTR(-ENOTDIR);
1969 retval = inode_permission(inode, MAY_EXEC);
1971 return ERR_PTR(retval);
1973 nd->path = nd->root;
1975 if (flags & LOOKUP_RCU) {
1977 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1978 nd->root_seq = nd->seq;
1979 nd->m_seq = read_seqbegin(&mount_lock);
1981 path_get(&nd->path);
1986 nd->root.mnt = NULL;
1988 nd->m_seq = read_seqbegin(&mount_lock);
1990 if (flags & LOOKUP_RCU) {
1992 nd->seq = set_root_rcu(nd);
1995 path_get(&nd->root);
1997 nd->path = nd->root;
1998 } else if (dfd == AT_FDCWD) {
1999 if (flags & LOOKUP_RCU) {
2000 struct fs_struct *fs = current->fs;
2006 seq = read_seqcount_begin(&fs->seq);
2008 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2009 } while (read_seqcount_retry(&fs->seq, seq));
2011 get_fs_pwd(current->fs, &nd->path);
2014 /* Caller must check execute permissions on the starting path component */
2015 struct fd f = fdget_raw(dfd);
2016 struct dentry *dentry;
2019 return ERR_PTR(-EBADF);
2021 dentry = f.file->f_path.dentry;
2024 if (!d_can_lookup(dentry)) {
2026 return ERR_PTR(-ENOTDIR);
2030 nd->path = f.file->f_path;
2031 if (flags & LOOKUP_RCU) {
2033 nd->inode = nd->path.dentry->d_inode;
2034 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2036 path_get(&nd->path);
2037 nd->inode = nd->path.dentry->d_inode;
2043 nd->inode = nd->path.dentry->d_inode;
2044 if (!(flags & LOOKUP_RCU))
2046 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
2048 if (!(nd->flags & LOOKUP_ROOT))
2049 nd->root.mnt = NULL;
2051 return ERR_PTR(-ECHILD);
2054 static void path_cleanup(struct nameidata *nd)
2056 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2057 path_put(&nd->root);
2058 nd->root.mnt = NULL;
2062 static const char *trailing_symlink(struct nameidata *nd)
2065 int error = may_follow_link(nd);
2066 if (unlikely(error))
2067 return ERR_PTR(error);
2068 nd->flags |= LOOKUP_PARENT;
2069 nd->stack[0].name = NULL;
2074 static inline int lookup_last(struct nameidata *nd)
2076 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2077 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2079 nd->flags &= ~LOOKUP_PARENT;
2080 return walk_component(nd,
2081 nd->flags & LOOKUP_FOLLOW
2083 ? WALK_PUT | WALK_GET
2088 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2089 static int path_lookupat(int dfd, const struct filename *name, unsigned flags,
2090 struct nameidata *nd, struct path *path)
2092 const char *s = path_init(dfd, name, flags, nd);
2097 while (!(err = link_path_walk(s, nd))
2098 && ((err = lookup_last(nd)) > 0)) {
2099 s = trailing_symlink(nd);
2106 err = complete_walk(nd);
2108 if (!err && nd->flags & LOOKUP_DIRECTORY)
2109 if (!d_can_lookup(nd->path.dentry))
2113 nd->path.mnt = NULL;
2114 nd->path.dentry = NULL;
2121 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2122 struct path *path, struct path *root)
2125 struct nameidata nd, *saved_nd = set_nameidata(&nd);
2126 if (unlikely(root)) {
2128 flags |= LOOKUP_ROOT;
2130 retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, &nd, path);
2131 if (unlikely(retval == -ECHILD))
2132 retval = path_lookupat(dfd, name, flags, &nd, path);
2133 if (unlikely(retval == -ESTALE))
2134 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL,
2137 if (likely(!retval))
2138 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2139 restore_nameidata(saved_nd);
2144 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2145 static int path_parentat(int dfd, const struct filename *name,
2146 unsigned int flags, struct nameidata *nd)
2148 const char *s = path_init(dfd, name, flags, nd);
2152 err = link_path_walk(s, nd);
2154 err = complete_walk(nd);
2161 static int filename_parentat(int dfd, struct filename *name,
2162 unsigned int flags, struct nameidata *nd)
2165 struct nameidata *saved_nd = set_nameidata(nd);
2167 retval = path_parentat(dfd, name, flags | LOOKUP_RCU, nd);
2168 if (unlikely(retval == -ECHILD))
2169 retval = path_parentat(dfd, name, flags, nd);
2170 if (unlikely(retval == -ESTALE))
2171 retval = path_parentat(dfd, name, flags | LOOKUP_REVAL, nd);
2173 if (likely(!retval))
2174 audit_inode(name, nd->path.dentry, LOOKUP_PARENT);
2175 restore_nameidata(saved_nd);
2179 /* does lookup, returns the object with parent locked */
2180 struct dentry *kern_path_locked(const char *name, struct path *path)
2182 struct filename *filename = getname_kernel(name);
2183 struct nameidata nd;
2187 if (IS_ERR(filename))
2188 return ERR_CAST(filename);
2190 err = filename_parentat(AT_FDCWD, filename, 0, &nd);
2195 if (nd.last_type != LAST_NORM) {
2197 d = ERR_PTR(-EINVAL);
2200 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2201 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2203 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2213 int kern_path(const char *name, unsigned int flags, struct path *path)
2215 struct filename *filename = getname_kernel(name);
2216 if (IS_ERR(filename))
2217 return PTR_ERR(filename);
2218 return filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2220 EXPORT_SYMBOL(kern_path);
2223 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2224 * @dentry: pointer to dentry of the base directory
2225 * @mnt: pointer to vfs mount of the base directory
2226 * @name: pointer to file name
2227 * @flags: lookup flags
2228 * @path: pointer to struct path to fill
2230 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2231 const char *name, unsigned int flags,
2234 struct path root = {.mnt = mnt, .dentry = dentry};
2235 struct filename *filename = getname_kernel(name);
2237 BUG_ON(flags & LOOKUP_PARENT);
2239 if (IS_ERR(filename))
2240 return PTR_ERR(filename);
2242 /* the first argument of filename_lookup() is ignored with root */
2243 return filename_lookup(AT_FDCWD, filename, flags , path, &root);
2245 EXPORT_SYMBOL(vfs_path_lookup);
2248 * lookup_one_len - filesystem helper to lookup single pathname component
2249 * @name: pathname component to lookup
2250 * @base: base directory to lookup from
2251 * @len: maximum length @len should be interpreted to
2253 * Note that this routine is purely a helper for filesystem usage and should
2254 * not be called by generic code.
2256 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2262 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2266 this.hash = full_name_hash(name, len);
2268 return ERR_PTR(-EACCES);
2270 if (unlikely(name[0] == '.')) {
2271 if (len < 2 || (len == 2 && name[1] == '.'))
2272 return ERR_PTR(-EACCES);
2276 c = *(const unsigned char *)name++;
2277 if (c == '/' || c == '\0')
2278 return ERR_PTR(-EACCES);
2281 * See if the low-level filesystem might want
2282 * to use its own hash..
2284 if (base->d_flags & DCACHE_OP_HASH) {
2285 int err = base->d_op->d_hash(base, &this);
2287 return ERR_PTR(err);
2290 err = inode_permission(base->d_inode, MAY_EXEC);
2292 return ERR_PTR(err);
2294 return __lookup_hash(&this, base, 0);
2296 EXPORT_SYMBOL(lookup_one_len);
2298 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2299 struct path *path, int *empty)
2301 struct filename *tmp = getname_flags(name, flags, empty);
2303 return PTR_ERR(tmp);
2305 BUG_ON(flags & LOOKUP_PARENT);
2307 return filename_lookup(dfd, tmp, flags, path, NULL);
2310 int user_path_at(int dfd, const char __user *name, unsigned flags,
2313 return user_path_at_empty(dfd, name, flags, path, NULL);
2315 EXPORT_SYMBOL(user_path_at);
2318 * NB: most callers don't do anything directly with the reference to the
2319 * to struct filename, but the nd->last pointer points into the name string
2320 * allocated by getname. So we must hold the reference to it until all
2321 * path-walking is complete.
2323 static struct filename *
2324 user_path_parent(int dfd, const char __user *path,
2325 struct path *parent,
2330 struct nameidata nd;
2331 struct filename *s = getname(path);
2334 /* only LOOKUP_REVAL is allowed in extra flags */
2335 flags &= LOOKUP_REVAL;
2340 error = filename_parentat(dfd, s, flags, &nd);
2343 return ERR_PTR(error);
2347 *type = nd.last_type;
2353 * mountpoint_last - look up last component for umount
2354 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2355 * @path: pointer to container for result
2357 * This is a special lookup_last function just for umount. In this case, we
2358 * need to resolve the path without doing any revalidation.
2360 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2361 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2362 * in almost all cases, this lookup will be served out of the dcache. The only
2363 * cases where it won't are if nd->last refers to a symlink or the path is
2364 * bogus and it doesn't exist.
2367 * -error: if there was an error during lookup. This includes -ENOENT if the
2368 * lookup found a negative dentry. The nd->path reference will also be
2371 * 0: if we successfully resolved nd->path and found it to not to be a
2372 * symlink that needs to be followed. "path" will also be populated.
2373 * The nd->path reference will also be put.
2375 * 1: if we successfully resolved nd->last and found it to be a symlink
2376 * that needs to be followed. "path" will be populated with the path
2377 * to the link, and nd->path will *not* be put.
2380 mountpoint_last(struct nameidata *nd, struct path *path)
2383 struct dentry *dentry;
2384 struct dentry *dir = nd->path.dentry;
2386 /* If we're in rcuwalk, drop out of it to handle last component */
2387 if (nd->flags & LOOKUP_RCU) {
2388 if (unlazy_walk(nd, NULL, 0))
2392 nd->flags &= ~LOOKUP_PARENT;
2394 if (unlikely(nd->last_type != LAST_NORM)) {
2395 error = handle_dots(nd, nd->last_type);
2398 dentry = dget(nd->path.dentry);
2402 mutex_lock(&dir->d_inode->i_mutex);
2403 dentry = d_lookup(dir, &nd->last);
2406 * No cached dentry. Mounted dentries are pinned in the cache,
2407 * so that means that this dentry is probably a symlink or the
2408 * path doesn't actually point to a mounted dentry.
2410 dentry = d_alloc(dir, &nd->last);
2412 mutex_unlock(&dir->d_inode->i_mutex);
2415 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2416 if (IS_ERR(dentry)) {
2417 mutex_unlock(&dir->d_inode->i_mutex);
2418 return PTR_ERR(dentry);
2421 mutex_unlock(&dir->d_inode->i_mutex);
2424 if (d_is_negative(dentry)) {
2430 path->dentry = dentry;
2431 path->mnt = nd->path.mnt;
2432 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2433 d_backing_inode(dentry), 0);
2434 if (unlikely(error))
2442 * path_mountpoint - look up a path to be umounted
2443 * @dfd: directory file descriptor to start walk from
2444 * @name: full pathname to walk
2445 * @path: pointer to container for result
2446 * @flags: lookup flags
2448 * Look up the given name, but don't attempt to revalidate the last component.
2449 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2452 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2453 struct nameidata *nd, unsigned int flags)
2455 const char *s = path_init(dfd, name, flags, nd);
2459 while (!(err = link_path_walk(s, nd)) &&
2460 (err = mountpoint_last(nd, path)) > 0) {
2461 s = trailing_symlink(nd);
2473 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2476 struct nameidata nd, *saved;
2479 return PTR_ERR(name);
2480 saved = set_nameidata(&nd);
2481 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2482 if (unlikely(error == -ECHILD))
2483 error = path_mountpoint(dfd, name, path, &nd, flags);
2484 if (unlikely(error == -ESTALE))
2485 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2487 audit_inode(name, path->dentry, 0);
2488 restore_nameidata(saved);
2494 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2495 * @dfd: directory file descriptor
2496 * @name: pathname from userland
2497 * @flags: lookup flags
2498 * @path: pointer to container to hold result
2500 * A umount is a special case for path walking. We're not actually interested
2501 * in the inode in this situation, and ESTALE errors can be a problem. We
2502 * simply want track down the dentry and vfsmount attached at the mountpoint
2503 * and avoid revalidating the last component.
2505 * Returns 0 and populates "path" on success.
2508 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2511 return filename_mountpoint(dfd, getname(name), path, flags);
2515 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2518 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2520 EXPORT_SYMBOL(kern_path_mountpoint);
2522 int __check_sticky(struct inode *dir, struct inode *inode)
2524 kuid_t fsuid = current_fsuid();
2526 if (uid_eq(inode->i_uid, fsuid))
2528 if (uid_eq(dir->i_uid, fsuid))
2530 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2532 EXPORT_SYMBOL(__check_sticky);
2535 * Check whether we can remove a link victim from directory dir, check
2536 * whether the type of victim is right.
2537 * 1. We can't do it if dir is read-only (done in permission())
2538 * 2. We should have write and exec permissions on dir
2539 * 3. We can't remove anything from append-only dir
2540 * 4. We can't do anything with immutable dir (done in permission())
2541 * 5. If the sticky bit on dir is set we should either
2542 * a. be owner of dir, or
2543 * b. be owner of victim, or
2544 * c. have CAP_FOWNER capability
2545 * 6. If the victim is append-only or immutable we can't do antyhing with
2546 * links pointing to it.
2547 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2548 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2549 * 9. We can't remove a root or mountpoint.
2550 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2551 * nfs_async_unlink().
2553 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2555 struct inode *inode = d_backing_inode(victim);
2558 if (d_is_negative(victim))
2562 BUG_ON(victim->d_parent->d_inode != dir);
2563 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2565 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2571 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2572 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2575 if (!d_is_dir(victim))
2577 if (IS_ROOT(victim))
2579 } else if (d_is_dir(victim))
2581 if (IS_DEADDIR(dir))
2583 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2588 /* Check whether we can create an object with dentry child in directory
2590 * 1. We can't do it if child already exists (open has special treatment for
2591 * this case, but since we are inlined it's OK)
2592 * 2. We can't do it if dir is read-only (done in permission())
2593 * 3. We should have write and exec permissions on dir
2594 * 4. We can't do it if dir is immutable (done in permission())
2596 static inline int may_create(struct inode *dir, struct dentry *child)
2598 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2601 if (IS_DEADDIR(dir))
2603 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2607 * p1 and p2 should be directories on the same fs.
2609 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2614 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2618 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2620 p = d_ancestor(p2, p1);
2622 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2623 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2627 p = d_ancestor(p1, p2);
2629 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2630 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2634 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2635 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2638 EXPORT_SYMBOL(lock_rename);
2640 void unlock_rename(struct dentry *p1, struct dentry *p2)
2642 mutex_unlock(&p1->d_inode->i_mutex);
2644 mutex_unlock(&p2->d_inode->i_mutex);
2645 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2648 EXPORT_SYMBOL(unlock_rename);
2650 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2653 int error = may_create(dir, dentry);
2657 if (!dir->i_op->create)
2658 return -EACCES; /* shouldn't it be ENOSYS? */
2661 error = security_inode_create(dir, dentry, mode);
2664 error = dir->i_op->create(dir, dentry, mode, want_excl);
2666 fsnotify_create(dir, dentry);
2669 EXPORT_SYMBOL(vfs_create);
2671 static int may_open(struct path *path, int acc_mode, int flag)
2673 struct dentry *dentry = path->dentry;
2674 struct inode *inode = dentry->d_inode;
2684 switch (inode->i_mode & S_IFMT) {
2688 if (acc_mode & MAY_WRITE)
2693 if (path->mnt->mnt_flags & MNT_NODEV)
2702 error = inode_permission(inode, acc_mode);
2707 * An append-only file must be opened in append mode for writing.
2709 if (IS_APPEND(inode)) {
2710 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2716 /* O_NOATIME can only be set by the owner or superuser */
2717 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2723 static int handle_truncate(struct file *filp)
2725 struct path *path = &filp->f_path;
2726 struct inode *inode = path->dentry->d_inode;
2727 int error = get_write_access(inode);
2731 * Refuse to truncate files with mandatory locks held on them.
2733 error = locks_verify_locked(filp);
2735 error = security_path_truncate(path);
2737 error = do_truncate(path->dentry, 0,
2738 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2741 put_write_access(inode);
2745 static inline int open_to_namei_flags(int flag)
2747 if ((flag & O_ACCMODE) == 3)
2752 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2754 int error = security_path_mknod(dir, dentry, mode, 0);
2758 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2762 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2766 * Attempt to atomically look up, create and open a file from a negative
2769 * Returns 0 if successful. The file will have been created and attached to
2770 * @file by the filesystem calling finish_open().
2772 * Returns 1 if the file was looked up only or didn't need creating. The
2773 * caller will need to perform the open themselves. @path will have been
2774 * updated to point to the new dentry. This may be negative.
2776 * Returns an error code otherwise.
2778 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2779 struct path *path, struct file *file,
2780 const struct open_flags *op,
2781 bool got_write, bool need_lookup,
2784 struct inode *dir = nd->path.dentry->d_inode;
2785 unsigned open_flag = open_to_namei_flags(op->open_flag);
2789 int create_error = 0;
2790 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2793 BUG_ON(dentry->d_inode);
2795 /* Don't create child dentry for a dead directory. */
2796 if (unlikely(IS_DEADDIR(dir))) {
2802 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2803 mode &= ~current_umask();
2805 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2807 open_flag &= ~O_TRUNC;
2810 * Checking write permission is tricky, bacuse we don't know if we are
2811 * going to actually need it: O_CREAT opens should work as long as the
2812 * file exists. But checking existence breaks atomicity. The trick is
2813 * to check access and if not granted clear O_CREAT from the flags.
2815 * Another problem is returing the "right" error value (e.g. for an
2816 * O_EXCL open we want to return EEXIST not EROFS).
2818 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2819 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2820 if (!(open_flag & O_CREAT)) {
2822 * No O_CREATE -> atomicity not a requirement -> fall
2823 * back to lookup + open
2826 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2827 /* Fall back and fail with the right error */
2828 create_error = -EROFS;
2831 /* No side effects, safe to clear O_CREAT */
2832 create_error = -EROFS;
2833 open_flag &= ~O_CREAT;
2837 if (open_flag & O_CREAT) {
2838 error = may_o_create(&nd->path, dentry, mode);
2840 create_error = error;
2841 if (open_flag & O_EXCL)
2843 open_flag &= ~O_CREAT;
2847 if (nd->flags & LOOKUP_DIRECTORY)
2848 open_flag |= O_DIRECTORY;
2850 file->f_path.dentry = DENTRY_NOT_SET;
2851 file->f_path.mnt = nd->path.mnt;
2852 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2855 if (create_error && error == -ENOENT)
2856 error = create_error;
2860 if (error) { /* returned 1, that is */
2861 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2865 if (file->f_path.dentry) {
2867 dentry = file->f_path.dentry;
2869 if (*opened & FILE_CREATED)
2870 fsnotify_create(dir, dentry);
2871 if (!dentry->d_inode) {
2872 WARN_ON(*opened & FILE_CREATED);
2874 error = create_error;
2878 if (excl && !(*opened & FILE_CREATED)) {
2887 * We didn't have the inode before the open, so check open permission
2890 acc_mode = op->acc_mode;
2891 if (*opened & FILE_CREATED) {
2892 WARN_ON(!(open_flag & O_CREAT));
2893 fsnotify_create(dir, dentry);
2894 acc_mode = MAY_OPEN;
2896 error = may_open(&file->f_path, acc_mode, open_flag);
2906 dentry = lookup_real(dir, dentry, nd->flags);
2908 return PTR_ERR(dentry);
2911 int open_flag = op->open_flag;
2913 error = create_error;
2914 if ((open_flag & O_EXCL)) {
2915 if (!dentry->d_inode)
2917 } else if (!dentry->d_inode) {
2919 } else if ((open_flag & O_TRUNC) &&
2923 /* will fail later, go on to get the right error */
2927 path->dentry = dentry;
2928 path->mnt = nd->path.mnt;
2933 * Look up and maybe create and open the last component.
2935 * Must be called with i_mutex held on parent.
2937 * Returns 0 if the file was successfully atomically created (if necessary) and
2938 * opened. In this case the file will be returned attached to @file.
2940 * Returns 1 if the file was not completely opened at this time, though lookups
2941 * and creations will have been performed and the dentry returned in @path will
2942 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2943 * specified then a negative dentry may be returned.
2945 * An error code is returned otherwise.
2947 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2948 * cleared otherwise prior to returning.
2950 static int lookup_open(struct nameidata *nd, struct path *path,
2952 const struct open_flags *op,
2953 bool got_write, int *opened)
2955 struct dentry *dir = nd->path.dentry;
2956 struct inode *dir_inode = dir->d_inode;
2957 struct dentry *dentry;
2961 *opened &= ~FILE_CREATED;
2962 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2964 return PTR_ERR(dentry);
2966 /* Cached positive dentry: will open in f_op->open */
2967 if (!need_lookup && dentry->d_inode)
2970 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2971 return atomic_open(nd, dentry, path, file, op, got_write,
2972 need_lookup, opened);
2976 BUG_ON(dentry->d_inode);
2978 dentry = lookup_real(dir_inode, dentry, nd->flags);
2980 return PTR_ERR(dentry);
2983 /* Negative dentry, just create the file */
2984 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2985 umode_t mode = op->mode;
2986 if (!IS_POSIXACL(dir->d_inode))
2987 mode &= ~current_umask();
2989 * This write is needed to ensure that a
2990 * rw->ro transition does not occur between
2991 * the time when the file is created and when
2992 * a permanent write count is taken through
2993 * the 'struct file' in finish_open().
2999 *opened |= FILE_CREATED;
3000 error = security_path_mknod(&nd->path, dentry, mode, 0);
3003 error = vfs_create(dir->d_inode, dentry, mode,
3004 nd->flags & LOOKUP_EXCL);
3009 path->dentry = dentry;
3010 path->mnt = nd->path.mnt;
3019 * Handle the last step of open()
3021 static int do_last(struct nameidata *nd,
3022 struct file *file, const struct open_flags *op,
3023 int *opened, struct filename *name)
3025 struct dentry *dir = nd->path.dentry;
3026 int open_flag = op->open_flag;
3027 bool will_truncate = (open_flag & O_TRUNC) != 0;
3028 bool got_write = false;
3029 int acc_mode = op->acc_mode;
3031 struct inode *inode;
3032 struct path save_parent = { .dentry = NULL, .mnt = NULL };
3034 bool retried = false;
3037 nd->flags &= ~LOOKUP_PARENT;
3038 nd->flags |= op->intent;
3040 if (nd->last_type != LAST_NORM) {
3041 error = handle_dots(nd, nd->last_type);
3042 if (unlikely(error))
3047 if (!(open_flag & O_CREAT)) {
3048 if (nd->last.name[nd->last.len])
3049 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3050 /* we _can_ be in RCU mode here */
3051 error = lookup_fast(nd, &path, &inode, &seq);
3058 BUG_ON(nd->inode != dir->d_inode);
3060 /* create side of things */
3062 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3063 * has been cleared when we got to the last component we are
3066 error = complete_walk(nd);
3070 audit_inode(name, dir, LOOKUP_PARENT);
3071 /* trailing slashes? */
3072 if (unlikely(nd->last.name[nd->last.len]))
3077 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3078 error = mnt_want_write(nd->path.mnt);
3082 * do _not_ fail yet - we might not need that or fail with
3083 * a different error; let lookup_open() decide; we'll be
3084 * dropping this one anyway.
3087 mutex_lock(&dir->d_inode->i_mutex);
3088 error = lookup_open(nd, &path, file, op, got_write, opened);
3089 mutex_unlock(&dir->d_inode->i_mutex);
3095 if ((*opened & FILE_CREATED) ||
3096 !S_ISREG(file_inode(file)->i_mode))
3097 will_truncate = false;
3099 audit_inode(name, file->f_path.dentry, 0);
3103 if (*opened & FILE_CREATED) {
3104 /* Don't check for write permission, don't truncate */
3105 open_flag &= ~O_TRUNC;
3106 will_truncate = false;
3107 acc_mode = MAY_OPEN;
3108 path_to_nameidata(&path, nd);
3109 goto finish_open_created;
3113 * create/update audit record if it already exists.
3115 if (d_is_positive(path.dentry))
3116 audit_inode(name, path.dentry, 0);
3119 * If atomic_open() acquired write access it is dropped now due to
3120 * possible mount and symlink following (this might be optimized away if
3124 mnt_drop_write(nd->path.mnt);
3128 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3129 path_to_nameidata(&path, nd);
3133 error = follow_managed(&path, nd);
3134 if (unlikely(error < 0))
3137 BUG_ON(nd->flags & LOOKUP_RCU);
3138 inode = d_backing_inode(path.dentry);
3139 seq = 0; /* out of RCU mode, so the value doesn't matter */
3140 if (unlikely(d_is_negative(path.dentry))) {
3141 path_to_nameidata(&path, nd);
3147 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3149 if (unlikely(error))
3152 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3153 path_to_nameidata(&path, nd);
3157 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3158 path_to_nameidata(&path, nd);
3160 save_parent.dentry = nd->path.dentry;
3161 save_parent.mnt = mntget(path.mnt);
3162 nd->path.dentry = path.dentry;
3167 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3169 error = complete_walk(nd);
3171 path_put(&save_parent);
3174 audit_inode(name, nd->path.dentry, 0);
3176 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3179 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3181 if (!d_is_reg(nd->path.dentry))
3182 will_truncate = false;
3184 if (will_truncate) {
3185 error = mnt_want_write(nd->path.mnt);
3190 finish_open_created:
3191 error = may_open(&nd->path, acc_mode, open_flag);
3195 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3196 error = vfs_open(&nd->path, file, current_cred());
3198 *opened |= FILE_OPENED;
3200 if (error == -EOPENSTALE)
3205 error = open_check_o_direct(file);
3208 error = ima_file_check(file, op->acc_mode, *opened);
3212 if (will_truncate) {
3213 error = handle_truncate(file);
3219 mnt_drop_write(nd->path.mnt);
3220 path_put(&save_parent);
3228 /* If no saved parent or already retried then can't retry */
3229 if (!save_parent.dentry || retried)
3232 BUG_ON(save_parent.dentry != dir);
3233 path_put(&nd->path);
3234 nd->path = save_parent;
3235 nd->inode = dir->d_inode;
3236 save_parent.mnt = NULL;
3237 save_parent.dentry = NULL;
3239 mnt_drop_write(nd->path.mnt);
3246 static int do_tmpfile(int dfd, struct filename *pathname,
3247 struct nameidata *nd, int flags,
3248 const struct open_flags *op,
3249 struct file *file, int *opened)
3251 static const struct qstr name = QSTR_INIT("/", 1);
3252 struct dentry *child;
3255 int error = path_lookupat(dfd, pathname,
3256 flags | LOOKUP_DIRECTORY, nd, &path);
3257 if (unlikely(error))
3259 error = mnt_want_write(path.mnt);
3260 if (unlikely(error))
3262 dir = path.dentry->d_inode;
3263 /* we want directory to be writable */
3264 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3267 if (!dir->i_op->tmpfile) {
3268 error = -EOPNOTSUPP;
3271 child = d_alloc(path.dentry, &name);
3272 if (unlikely(!child)) {
3277 path.dentry = child;
3278 error = dir->i_op->tmpfile(dir, child, op->mode);
3281 audit_inode(pathname, child, 0);
3282 /* Don't check for other permissions, the inode was just created */
3283 error = may_open(&path, MAY_OPEN, op->open_flag);
3286 file->f_path.mnt = path.mnt;
3287 error = finish_open(file, child, NULL, opened);
3290 error = open_check_o_direct(file);
3293 } else if (!(op->open_flag & O_EXCL)) {
3294 struct inode *inode = file_inode(file);
3295 spin_lock(&inode->i_lock);
3296 inode->i_state |= I_LINKABLE;
3297 spin_unlock(&inode->i_lock);
3300 mnt_drop_write(path.mnt);
3306 static struct file *path_openat(int dfd, struct filename *pathname,
3307 struct nameidata *nd, const struct open_flags *op, int flags)
3314 file = get_empty_filp();
3318 file->f_flags = op->open_flag;
3320 if (unlikely(file->f_flags & __O_TMPFILE)) {
3321 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3325 s = path_init(dfd, pathname, flags, nd);
3330 while (!(error = link_path_walk(s, nd)) &&
3331 (error = do_last(nd, file, op, &opened, pathname)) > 0) {
3332 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3333 s = trailing_symlink(nd);
3342 if (!(opened & FILE_OPENED)) {
3346 if (unlikely(error)) {
3347 if (error == -EOPENSTALE) {
3348 if (flags & LOOKUP_RCU)
3353 file = ERR_PTR(error);
3358 struct file *do_filp_open(int dfd, struct filename *pathname,
3359 const struct open_flags *op)
3361 struct nameidata nd, *saved_nd = set_nameidata(&nd);
3362 int flags = op->lookup_flags;
3365 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3366 if (unlikely(filp == ERR_PTR(-ECHILD)))
3367 filp = path_openat(dfd, pathname, &nd, op, flags);
3368 if (unlikely(filp == ERR_PTR(-ESTALE)))
3369 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3370 restore_nameidata(saved_nd);
3374 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3375 const char *name, const struct open_flags *op)
3377 struct nameidata nd, *saved_nd;
3379 struct filename *filename;
3380 int flags = op->lookup_flags | LOOKUP_ROOT;
3383 nd.root.dentry = dentry;
3385 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3386 return ERR_PTR(-ELOOP);
3388 filename = getname_kernel(name);
3389 if (unlikely(IS_ERR(filename)))
3390 return ERR_CAST(filename);
3392 saved_nd = set_nameidata(&nd);
3393 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3394 if (unlikely(file == ERR_PTR(-ECHILD)))
3395 file = path_openat(-1, filename, &nd, op, flags);
3396 if (unlikely(file == ERR_PTR(-ESTALE)))
3397 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3398 restore_nameidata(saved_nd);
3403 static struct dentry *filename_create(int dfd, struct filename *name,
3404 struct path *path, unsigned int lookup_flags)
3406 struct dentry *dentry = ERR_PTR(-EEXIST);
3407 struct nameidata nd;
3410 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3413 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3414 * other flags passed in are ignored!
3416 lookup_flags &= LOOKUP_REVAL;
3418 error = filename_parentat(dfd, name, lookup_flags, &nd);
3420 return ERR_PTR(error);
3423 * Yucky last component or no last component at all?
3424 * (foo/., foo/.., /////)
3426 if (nd.last_type != LAST_NORM)
3428 nd.flags &= ~LOOKUP_PARENT;
3429 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3431 /* don't fail immediately if it's r/o, at least try to report other errors */
3432 err2 = mnt_want_write(nd.path.mnt);
3434 * Do the final lookup.
3436 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3437 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3442 if (d_is_positive(dentry))
3446 * Special case - lookup gave negative, but... we had foo/bar/
3447 * From the vfs_mknod() POV we just have a negative dentry -
3448 * all is fine. Let's be bastards - you had / on the end, you've
3449 * been asking for (non-existent) directory. -ENOENT for you.
3451 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3455 if (unlikely(err2)) {
3463 dentry = ERR_PTR(error);
3465 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3467 mnt_drop_write(nd.path.mnt);
3473 struct dentry *kern_path_create(int dfd, const char *pathname,
3474 struct path *path, unsigned int lookup_flags)
3476 struct filename *filename = getname_kernel(pathname);
3479 if (IS_ERR(filename))
3480 return ERR_CAST(filename);
3481 res = filename_create(dfd, filename, path, lookup_flags);
3485 EXPORT_SYMBOL(kern_path_create);
3487 void done_path_create(struct path *path, struct dentry *dentry)
3490 mutex_unlock(&path->dentry->d_inode->i_mutex);
3491 mnt_drop_write(path->mnt);
3494 EXPORT_SYMBOL(done_path_create);
3496 struct dentry *user_path_create(int dfd, const char __user *pathname,
3497 struct path *path, unsigned int lookup_flags)
3499 struct filename *tmp = getname(pathname);
3502 return ERR_CAST(tmp);
3503 res = filename_create(dfd, tmp, path, lookup_flags);
3507 EXPORT_SYMBOL(user_path_create);
3509 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3511 int error = may_create(dir, dentry);
3516 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3519 if (!dir->i_op->mknod)
3522 error = devcgroup_inode_mknod(mode, dev);
3526 error = security_inode_mknod(dir, dentry, mode, dev);
3530 error = dir->i_op->mknod(dir, dentry, mode, dev);
3532 fsnotify_create(dir, dentry);
3535 EXPORT_SYMBOL(vfs_mknod);
3537 static int may_mknod(umode_t mode)
3539 switch (mode & S_IFMT) {
3545 case 0: /* zero mode translates to S_IFREG */
3554 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3557 struct dentry *dentry;
3560 unsigned int lookup_flags = 0;
3562 error = may_mknod(mode);
3566 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3568 return PTR_ERR(dentry);
3570 if (!IS_POSIXACL(path.dentry->d_inode))
3571 mode &= ~current_umask();
3572 error = security_path_mknod(&path, dentry, mode, dev);
3575 switch (mode & S_IFMT) {
3576 case 0: case S_IFREG:
3577 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3579 case S_IFCHR: case S_IFBLK:
3580 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3581 new_decode_dev(dev));
3583 case S_IFIFO: case S_IFSOCK:
3584 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3588 done_path_create(&path, dentry);
3589 if (retry_estale(error, lookup_flags)) {
3590 lookup_flags |= LOOKUP_REVAL;
3596 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3598 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3601 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3603 int error = may_create(dir, dentry);
3604 unsigned max_links = dir->i_sb->s_max_links;
3609 if (!dir->i_op->mkdir)
3612 mode &= (S_IRWXUGO|S_ISVTX);
3613 error = security_inode_mkdir(dir, dentry, mode);
3617 if (max_links && dir->i_nlink >= max_links)
3620 error = dir->i_op->mkdir(dir, dentry, mode);
3622 fsnotify_mkdir(dir, dentry);
3625 EXPORT_SYMBOL(vfs_mkdir);
3627 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3629 struct dentry *dentry;
3632 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3635 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3637 return PTR_ERR(dentry);
3639 if (!IS_POSIXACL(path.dentry->d_inode))
3640 mode &= ~current_umask();
3641 error = security_path_mkdir(&path, dentry, mode);
3643 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3644 done_path_create(&path, dentry);
3645 if (retry_estale(error, lookup_flags)) {
3646 lookup_flags |= LOOKUP_REVAL;
3652 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3654 return sys_mkdirat(AT_FDCWD, pathname, mode);
3658 * The dentry_unhash() helper will try to drop the dentry early: we
3659 * should have a usage count of 1 if we're the only user of this
3660 * dentry, and if that is true (possibly after pruning the dcache),
3661 * then we drop the dentry now.
3663 * A low-level filesystem can, if it choses, legally
3666 * if (!d_unhashed(dentry))
3669 * if it cannot handle the case of removing a directory
3670 * that is still in use by something else..
3672 void dentry_unhash(struct dentry *dentry)
3674 shrink_dcache_parent(dentry);
3675 spin_lock(&dentry->d_lock);
3676 if (dentry->d_lockref.count == 1)
3678 spin_unlock(&dentry->d_lock);
3680 EXPORT_SYMBOL(dentry_unhash);
3682 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3684 int error = may_delete(dir, dentry, 1);
3689 if (!dir->i_op->rmdir)
3693 mutex_lock(&dentry->d_inode->i_mutex);
3696 if (is_local_mountpoint(dentry))
3699 error = security_inode_rmdir(dir, dentry);
3703 shrink_dcache_parent(dentry);
3704 error = dir->i_op->rmdir(dir, dentry);
3708 dentry->d_inode->i_flags |= S_DEAD;
3710 detach_mounts(dentry);
3713 mutex_unlock(&dentry->d_inode->i_mutex);
3719 EXPORT_SYMBOL(vfs_rmdir);
3721 static long do_rmdir(int dfd, const char __user *pathname)
3724 struct filename *name;
3725 struct dentry *dentry;
3729 unsigned int lookup_flags = 0;
3731 name = user_path_parent(dfd, pathname,
3732 &path, &last, &type, lookup_flags);
3734 return PTR_ERR(name);
3748 error = mnt_want_write(path.mnt);
3752 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3753 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3754 error = PTR_ERR(dentry);
3757 if (!dentry->d_inode) {
3761 error = security_path_rmdir(&path, dentry);
3764 error = vfs_rmdir(path.dentry->d_inode, dentry);
3768 mutex_unlock(&path.dentry->d_inode->i_mutex);
3769 mnt_drop_write(path.mnt);
3773 if (retry_estale(error, lookup_flags)) {
3774 lookup_flags |= LOOKUP_REVAL;
3780 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3782 return do_rmdir(AT_FDCWD, pathname);
3786 * vfs_unlink - unlink a filesystem object
3787 * @dir: parent directory
3789 * @delegated_inode: returns victim inode, if the inode is delegated.
3791 * The caller must hold dir->i_mutex.
3793 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3794 * return a reference to the inode in delegated_inode. The caller
3795 * should then break the delegation on that inode and retry. Because
3796 * breaking a delegation may take a long time, the caller should drop
3797 * dir->i_mutex before doing so.
3799 * Alternatively, a caller may pass NULL for delegated_inode. This may
3800 * be appropriate for callers that expect the underlying filesystem not
3801 * to be NFS exported.
3803 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3805 struct inode *target = dentry->d_inode;
3806 int error = may_delete(dir, dentry, 0);
3811 if (!dir->i_op->unlink)
3814 mutex_lock(&target->i_mutex);
3815 if (is_local_mountpoint(dentry))
3818 error = security_inode_unlink(dir, dentry);
3820 error = try_break_deleg(target, delegated_inode);
3823 error = dir->i_op->unlink(dir, dentry);
3826 detach_mounts(dentry);
3831 mutex_unlock(&target->i_mutex);
3833 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3834 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3835 fsnotify_link_count(target);
3841 EXPORT_SYMBOL(vfs_unlink);
3844 * Make sure that the actual truncation of the file will occur outside its
3845 * directory's i_mutex. Truncate can take a long time if there is a lot of
3846 * writeout happening, and we don't want to prevent access to the directory
3847 * while waiting on the I/O.
3849 static long do_unlinkat(int dfd, const char __user *pathname)
3852 struct filename *name;
3853 struct dentry *dentry;
3857 struct inode *inode = NULL;
3858 struct inode *delegated_inode = NULL;
3859 unsigned int lookup_flags = 0;
3861 name = user_path_parent(dfd, pathname,
3862 &path, &last, &type, lookup_flags);
3864 return PTR_ERR(name);
3867 if (type != LAST_NORM)
3870 error = mnt_want_write(path.mnt);
3874 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3875 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3876 error = PTR_ERR(dentry);
3877 if (!IS_ERR(dentry)) {
3878 /* Why not before? Because we want correct error value */
3879 if (last.name[last.len])
3881 inode = dentry->d_inode;
3882 if (d_is_negative(dentry))
3885 error = security_path_unlink(&path, dentry);
3888 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3892 mutex_unlock(&path.dentry->d_inode->i_mutex);
3894 iput(inode); /* truncate the inode here */
3896 if (delegated_inode) {
3897 error = break_deleg_wait(&delegated_inode);
3901 mnt_drop_write(path.mnt);
3905 if (retry_estale(error, lookup_flags)) {
3906 lookup_flags |= LOOKUP_REVAL;
3913 if (d_is_negative(dentry))
3915 else if (d_is_dir(dentry))
3922 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3924 if ((flag & ~AT_REMOVEDIR) != 0)
3927 if (flag & AT_REMOVEDIR)
3928 return do_rmdir(dfd, pathname);
3930 return do_unlinkat(dfd, pathname);
3933 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3935 return do_unlinkat(AT_FDCWD, pathname);
3938 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3940 int error = may_create(dir, dentry);
3945 if (!dir->i_op->symlink)
3948 error = security_inode_symlink(dir, dentry, oldname);
3952 error = dir->i_op->symlink(dir, dentry, oldname);
3954 fsnotify_create(dir, dentry);
3957 EXPORT_SYMBOL(vfs_symlink);
3959 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3960 int, newdfd, const char __user *, newname)
3963 struct filename *from;
3964 struct dentry *dentry;
3966 unsigned int lookup_flags = 0;
3968 from = getname(oldname);
3970 return PTR_ERR(from);
3972 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3973 error = PTR_ERR(dentry);
3977 error = security_path_symlink(&path, dentry, from->name);
3979 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3980 done_path_create(&path, dentry);
3981 if (retry_estale(error, lookup_flags)) {
3982 lookup_flags |= LOOKUP_REVAL;
3990 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3992 return sys_symlinkat(oldname, AT_FDCWD, newname);
3996 * vfs_link - create a new link
3997 * @old_dentry: object to be linked
3999 * @new_dentry: where to create the new link
4000 * @delegated_inode: returns inode needing a delegation break
4002 * The caller must hold dir->i_mutex
4004 * If vfs_link discovers a delegation on the to-be-linked file in need
4005 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4006 * inode in delegated_inode. The caller should then break the delegation
4007 * and retry. Because breaking a delegation may take a long time, the
4008 * caller should drop the i_mutex before doing so.
4010 * Alternatively, a caller may pass NULL for delegated_inode. This may
4011 * be appropriate for callers that expect the underlying filesystem not
4012 * to be NFS exported.
4014 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4016 struct inode *inode = old_dentry->d_inode;
4017 unsigned max_links = dir->i_sb->s_max_links;
4023 error = may_create(dir, new_dentry);
4027 if (dir->i_sb != inode->i_sb)
4031 * A link to an append-only or immutable file cannot be created.
4033 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4035 if (!dir->i_op->link)
4037 if (S_ISDIR(inode->i_mode))
4040 error = security_inode_link(old_dentry, dir, new_dentry);
4044 mutex_lock(&inode->i_mutex);
4045 /* Make sure we don't allow creating hardlink to an unlinked file */
4046 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4048 else if (max_links && inode->i_nlink >= max_links)
4051 error = try_break_deleg(inode, delegated_inode);
4053 error = dir->i_op->link(old_dentry, dir, new_dentry);
4056 if (!error && (inode->i_state & I_LINKABLE)) {
4057 spin_lock(&inode->i_lock);
4058 inode->i_state &= ~I_LINKABLE;
4059 spin_unlock(&inode->i_lock);
4061 mutex_unlock(&inode->i_mutex);
4063 fsnotify_link(dir, inode, new_dentry);
4066 EXPORT_SYMBOL(vfs_link);
4069 * Hardlinks are often used in delicate situations. We avoid
4070 * security-related surprises by not following symlinks on the
4073 * We don't follow them on the oldname either to be compatible
4074 * with linux 2.0, and to avoid hard-linking to directories
4075 * and other special files. --ADM
4077 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4078 int, newdfd, const char __user *, newname, int, flags)
4080 struct dentry *new_dentry;
4081 struct path old_path, new_path;
4082 struct inode *delegated_inode = NULL;
4086 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4089 * To use null names we require CAP_DAC_READ_SEARCH
4090 * This ensures that not everyone will be able to create
4091 * handlink using the passed filedescriptor.
4093 if (flags & AT_EMPTY_PATH) {
4094 if (!capable(CAP_DAC_READ_SEARCH))
4099 if (flags & AT_SYMLINK_FOLLOW)
4100 how |= LOOKUP_FOLLOW;
4102 error = user_path_at(olddfd, oldname, how, &old_path);
4106 new_dentry = user_path_create(newdfd, newname, &new_path,
4107 (how & LOOKUP_REVAL));
4108 error = PTR_ERR(new_dentry);
4109 if (IS_ERR(new_dentry))
4113 if (old_path.mnt != new_path.mnt)
4115 error = may_linkat(&old_path);
4116 if (unlikely(error))
4118 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4121 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4123 done_path_create(&new_path, new_dentry);
4124 if (delegated_inode) {
4125 error = break_deleg_wait(&delegated_inode);
4127 path_put(&old_path);
4131 if (retry_estale(error, how)) {
4132 path_put(&old_path);
4133 how |= LOOKUP_REVAL;
4137 path_put(&old_path);
4142 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4144 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4148 * vfs_rename - rename a filesystem object
4149 * @old_dir: parent of source
4150 * @old_dentry: source
4151 * @new_dir: parent of destination
4152 * @new_dentry: destination
4153 * @delegated_inode: returns an inode needing a delegation break
4154 * @flags: rename flags
4156 * The caller must hold multiple mutexes--see lock_rename()).
4158 * If vfs_rename discovers a delegation in need of breaking at either
4159 * the source or destination, it will return -EWOULDBLOCK and return a
4160 * reference to the inode in delegated_inode. The caller should then
4161 * break the delegation and retry. Because breaking a delegation may
4162 * take a long time, the caller should drop all locks before doing
4165 * Alternatively, a caller may pass NULL for delegated_inode. This may
4166 * be appropriate for callers that expect the underlying filesystem not
4167 * to be NFS exported.
4169 * The worst of all namespace operations - renaming directory. "Perverted"
4170 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4172 * a) we can get into loop creation.
4173 * b) race potential - two innocent renames can create a loop together.
4174 * That's where 4.4 screws up. Current fix: serialization on
4175 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4177 * c) we have to lock _four_ objects - parents and victim (if it exists),
4178 * and source (if it is not a directory).
4179 * And that - after we got ->i_mutex on parents (until then we don't know
4180 * whether the target exists). Solution: try to be smart with locking
4181 * order for inodes. We rely on the fact that tree topology may change
4182 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4183 * move will be locked. Thus we can rank directories by the tree
4184 * (ancestors first) and rank all non-directories after them.
4185 * That works since everybody except rename does "lock parent, lookup,
4186 * lock child" and rename is under ->s_vfs_rename_mutex.
4187 * HOWEVER, it relies on the assumption that any object with ->lookup()
4188 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4189 * we'd better make sure that there's no link(2) for them.
4190 * d) conversion from fhandle to dentry may come in the wrong moment - when
4191 * we are removing the target. Solution: we will have to grab ->i_mutex
4192 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4193 * ->i_mutex on parents, which works but leads to some truly excessive
4196 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4197 struct inode *new_dir, struct dentry *new_dentry,
4198 struct inode **delegated_inode, unsigned int flags)
4201 bool is_dir = d_is_dir(old_dentry);
4202 const unsigned char *old_name;
4203 struct inode *source = old_dentry->d_inode;
4204 struct inode *target = new_dentry->d_inode;
4205 bool new_is_dir = false;
4206 unsigned max_links = new_dir->i_sb->s_max_links;
4208 if (source == target)
4211 error = may_delete(old_dir, old_dentry, is_dir);
4216 error = may_create(new_dir, new_dentry);
4218 new_is_dir = d_is_dir(new_dentry);
4220 if (!(flags & RENAME_EXCHANGE))
4221 error = may_delete(new_dir, new_dentry, is_dir);
4223 error = may_delete(new_dir, new_dentry, new_is_dir);
4228 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4231 if (flags && !old_dir->i_op->rename2)
4235 * If we are going to change the parent - check write permissions,
4236 * we'll need to flip '..'.
4238 if (new_dir != old_dir) {
4240 error = inode_permission(source, MAY_WRITE);
4244 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4245 error = inode_permission(target, MAY_WRITE);
4251 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4256 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4258 if (!is_dir || (flags & RENAME_EXCHANGE))
4259 lock_two_nondirectories(source, target);
4261 mutex_lock(&target->i_mutex);
4264 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4267 if (max_links && new_dir != old_dir) {
4269 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4271 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4272 old_dir->i_nlink >= max_links)
4275 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4276 shrink_dcache_parent(new_dentry);
4278 error = try_break_deleg(source, delegated_inode);
4282 if (target && !new_is_dir) {
4283 error = try_break_deleg(target, delegated_inode);
4287 if (!old_dir->i_op->rename2) {
4288 error = old_dir->i_op->rename(old_dir, old_dentry,
4289 new_dir, new_dentry);
4291 WARN_ON(old_dir->i_op->rename != NULL);
4292 error = old_dir->i_op->rename2(old_dir, old_dentry,
4293 new_dir, new_dentry, flags);
4298 if (!(flags & RENAME_EXCHANGE) && target) {
4300 target->i_flags |= S_DEAD;
4301 dont_mount(new_dentry);
4302 detach_mounts(new_dentry);
4304 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4305 if (!(flags & RENAME_EXCHANGE))
4306 d_move(old_dentry, new_dentry);
4308 d_exchange(old_dentry, new_dentry);
4311 if (!is_dir || (flags & RENAME_EXCHANGE))
4312 unlock_two_nondirectories(source, target);
4314 mutex_unlock(&target->i_mutex);
4317 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4318 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4319 if (flags & RENAME_EXCHANGE) {
4320 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4321 new_is_dir, NULL, new_dentry);
4324 fsnotify_oldname_free(old_name);
4328 EXPORT_SYMBOL(vfs_rename);
4330 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4331 int, newdfd, const char __user *, newname, unsigned int, flags)
4333 struct dentry *old_dentry, *new_dentry;
4334 struct dentry *trap;
4335 struct path old_path, new_path;
4336 struct qstr old_last, new_last;
4337 int old_type, new_type;
4338 struct inode *delegated_inode = NULL;
4339 struct filename *from;
4340 struct filename *to;
4341 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4342 bool should_retry = false;
4345 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4348 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4349 (flags & RENAME_EXCHANGE))
4352 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4355 if (flags & RENAME_EXCHANGE)
4359 from = user_path_parent(olddfd, oldname,
4360 &old_path, &old_last, &old_type, lookup_flags);
4362 error = PTR_ERR(from);
4366 to = user_path_parent(newdfd, newname,
4367 &new_path, &new_last, &new_type, lookup_flags);
4369 error = PTR_ERR(to);
4374 if (old_path.mnt != new_path.mnt)
4378 if (old_type != LAST_NORM)
4381 if (flags & RENAME_NOREPLACE)
4383 if (new_type != LAST_NORM)
4386 error = mnt_want_write(old_path.mnt);
4391 trap = lock_rename(new_path.dentry, old_path.dentry);
4393 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4394 error = PTR_ERR(old_dentry);
4395 if (IS_ERR(old_dentry))
4397 /* source must exist */
4399 if (d_is_negative(old_dentry))
4401 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4402 error = PTR_ERR(new_dentry);
4403 if (IS_ERR(new_dentry))
4406 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4408 if (flags & RENAME_EXCHANGE) {
4410 if (d_is_negative(new_dentry))
4413 if (!d_is_dir(new_dentry)) {
4415 if (new_last.name[new_last.len])
4419 /* unless the source is a directory trailing slashes give -ENOTDIR */
4420 if (!d_is_dir(old_dentry)) {
4422 if (old_last.name[old_last.len])
4424 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4427 /* source should not be ancestor of target */
4429 if (old_dentry == trap)
4431 /* target should not be an ancestor of source */
4432 if (!(flags & RENAME_EXCHANGE))
4434 if (new_dentry == trap)
4437 error = security_path_rename(&old_path, old_dentry,
4438 &new_path, new_dentry, flags);
4441 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4442 new_path.dentry->d_inode, new_dentry,
4443 &delegated_inode, flags);
4449 unlock_rename(new_path.dentry, old_path.dentry);
4450 if (delegated_inode) {
4451 error = break_deleg_wait(&delegated_inode);
4455 mnt_drop_write(old_path.mnt);
4457 if (retry_estale(error, lookup_flags))
4458 should_retry = true;
4459 path_put(&new_path);
4462 path_put(&old_path);
4465 should_retry = false;
4466 lookup_flags |= LOOKUP_REVAL;
4473 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4474 int, newdfd, const char __user *, newname)
4476 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4479 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4481 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4484 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4486 int error = may_create(dir, dentry);
4490 if (!dir->i_op->mknod)
4493 return dir->i_op->mknod(dir, dentry,
4494 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4496 EXPORT_SYMBOL(vfs_whiteout);
4498 int readlink_copy(char __user *buffer, int buflen, const char *link)
4500 int len = PTR_ERR(link);
4505 if (len > (unsigned) buflen)
4507 if (copy_to_user(buffer, link, len))
4512 EXPORT_SYMBOL(readlink_copy);
4515 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4516 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4517 * using) it for any given inode is up to filesystem.
4519 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4522 struct inode *inode = d_inode(dentry);
4523 const char *link = inode->i_link;
4527 link = inode->i_op->follow_link(dentry, &cookie);
4529 return PTR_ERR(link);
4531 res = readlink_copy(buffer, buflen, link);
4532 if (inode->i_op->put_link)
4533 inode->i_op->put_link(inode, cookie);
4536 EXPORT_SYMBOL(generic_readlink);
4538 /* get the link contents into pagecache */
4539 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4543 struct address_space *mapping = dentry->d_inode->i_mapping;
4544 page = read_mapping_page(mapping, 0, NULL);
4549 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4553 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4555 struct page *page = NULL;
4556 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4559 page_cache_release(page);
4563 EXPORT_SYMBOL(page_readlink);
4565 const char *page_follow_link_light(struct dentry *dentry, void **cookie)
4567 struct page *page = NULL;
4568 char *res = page_getlink(dentry, &page);
4573 EXPORT_SYMBOL(page_follow_link_light);
4575 void page_put_link(struct inode *unused, void *cookie)
4577 struct page *page = cookie;
4579 page_cache_release(page);
4581 EXPORT_SYMBOL(page_put_link);
4584 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4586 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4588 struct address_space *mapping = inode->i_mapping;
4593 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4595 flags |= AOP_FLAG_NOFS;
4598 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4599 flags, &page, &fsdata);
4603 kaddr = kmap_atomic(page);
4604 memcpy(kaddr, symname, len-1);
4605 kunmap_atomic(kaddr);
4607 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4614 mark_inode_dirty(inode);
4619 EXPORT_SYMBOL(__page_symlink);
4621 int page_symlink(struct inode *inode, const char *symname, int len)
4623 return __page_symlink(inode, symname, len,
4624 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4626 EXPORT_SYMBOL(page_symlink);
4628 const struct inode_operations page_symlink_inode_operations = {
4629 .readlink = generic_readlink,
4630 .follow_link = page_follow_link_light,
4631 .put_link = page_put_link,
4633 EXPORT_SYMBOL(page_symlink_inode_operations);