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 const size_t size = offsetof(struct filename, iname[1]);
223 struct filename *tmp;
225 tmp = kmalloc(size, GFP_KERNEL);
226 if (unlikely(!tmp)) {
228 return ERR_PTR(-ENOMEM);
230 tmp->name = (char *)result;
234 return ERR_PTR(-ENAMETOOLONG);
236 memcpy((char *)result->name, filename, len);
238 result->aname = NULL;
240 audit_getname(result);
245 void putname(struct filename *name)
247 BUG_ON(name->refcnt <= 0);
249 if (--name->refcnt > 0)
252 if (name->name != name->iname) {
253 __putname(name->name);
259 static int check_acl(struct inode *inode, int mask)
261 #ifdef CONFIG_FS_POSIX_ACL
262 struct posix_acl *acl;
264 if (mask & MAY_NOT_BLOCK) {
265 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
268 /* no ->get_acl() calls in RCU mode... */
269 if (acl == ACL_NOT_CACHED)
271 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
274 acl = get_acl(inode, ACL_TYPE_ACCESS);
278 int error = posix_acl_permission(inode, acl, mask);
279 posix_acl_release(acl);
288 * This does the basic permission checking
290 static int acl_permission_check(struct inode *inode, int mask)
292 unsigned int mode = inode->i_mode;
294 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
297 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
298 int error = check_acl(inode, mask);
299 if (error != -EAGAIN)
303 if (in_group_p(inode->i_gid))
308 * If the DACs are ok we don't need any capability check.
310 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
316 * generic_permission - check for access rights on a Posix-like filesystem
317 * @inode: inode to check access rights for
318 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
320 * Used to check for read/write/execute permissions on a file.
321 * We use "fsuid" for this, letting us set arbitrary permissions
322 * for filesystem access without changing the "normal" uids which
323 * are used for other things.
325 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
326 * request cannot be satisfied (eg. requires blocking or too much complexity).
327 * It would then be called again in ref-walk mode.
329 int generic_permission(struct inode *inode, int mask)
334 * Do the basic permission checks.
336 ret = acl_permission_check(inode, mask);
340 if (S_ISDIR(inode->i_mode)) {
341 /* DACs are overridable for directories */
342 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
344 if (!(mask & MAY_WRITE))
345 if (capable_wrt_inode_uidgid(inode,
346 CAP_DAC_READ_SEARCH))
351 * Read/write DACs are always overridable.
352 * Executable DACs are overridable when there is
353 * at least one exec bit set.
355 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
356 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
360 * Searching includes executable on directories, else just read.
362 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
363 if (mask == MAY_READ)
364 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
369 EXPORT_SYMBOL(generic_permission);
372 * We _really_ want to just do "generic_permission()" without
373 * even looking at the inode->i_op values. So we keep a cache
374 * flag in inode->i_opflags, that says "this has not special
375 * permission function, use the fast case".
377 static inline int do_inode_permission(struct inode *inode, int mask)
379 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
380 if (likely(inode->i_op->permission))
381 return inode->i_op->permission(inode, mask);
383 /* This gets set once for the inode lifetime */
384 spin_lock(&inode->i_lock);
385 inode->i_opflags |= IOP_FASTPERM;
386 spin_unlock(&inode->i_lock);
388 return generic_permission(inode, mask);
392 * __inode_permission - Check for access rights to a given inode
393 * @inode: Inode to check permission on
394 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
396 * Check for read/write/execute permissions on an inode.
398 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
400 * This does not check for a read-only file system. You probably want
401 * inode_permission().
403 int __inode_permission(struct inode *inode, int mask)
407 if (unlikely(mask & MAY_WRITE)) {
409 * Nobody gets write access to an immutable file.
411 if (IS_IMMUTABLE(inode))
415 retval = do_inode_permission(inode, mask);
419 retval = devcgroup_inode_permission(inode, mask);
423 return security_inode_permission(inode, mask);
425 EXPORT_SYMBOL(__inode_permission);
428 * sb_permission - Check superblock-level permissions
429 * @sb: Superblock of inode to check permission on
430 * @inode: Inode to check permission on
431 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
433 * Separate out file-system wide checks from inode-specific permission checks.
435 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
437 if (unlikely(mask & MAY_WRITE)) {
438 umode_t mode = inode->i_mode;
440 /* Nobody gets write access to a read-only fs. */
441 if ((sb->s_flags & MS_RDONLY) &&
442 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
449 * inode_permission - Check for access rights to a given inode
450 * @inode: Inode to check permission on
451 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
453 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
454 * this, letting us set arbitrary permissions for filesystem access without
455 * changing the "normal" UIDs which are used for other things.
457 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
459 int inode_permission(struct inode *inode, int mask)
463 retval = sb_permission(inode->i_sb, inode, mask);
466 return __inode_permission(inode, mask);
468 EXPORT_SYMBOL(inode_permission);
471 * path_get - get a reference to a path
472 * @path: path to get the reference to
474 * Given a path increment the reference count to the dentry and the vfsmount.
476 void path_get(const struct path *path)
481 EXPORT_SYMBOL(path_get);
484 * path_put - put a reference to a path
485 * @path: path to put the reference to
487 * Given a path decrement the reference count to the dentry and the vfsmount.
489 void path_put(const struct path *path)
494 EXPORT_SYMBOL(path_put);
496 #define EMBEDDED_LEVELS 2
501 struct inode *inode; /* path.dentry.d_inode */
506 int total_link_count;
513 } *stack, internal[EMBEDDED_LEVELS];
514 struct filename *name;
515 struct nameidata *saved;
520 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
522 struct nameidata *old = current->nameidata;
523 p->stack = p->internal;
526 p->total_link_count = old ? old->total_link_count : 0;
528 current->nameidata = p;
531 static void restore_nameidata(void)
533 struct nameidata *now = current->nameidata, *old = now->saved;
535 current->nameidata = old;
537 old->total_link_count = now->total_link_count;
538 if (now->stack != now->internal) {
540 now->stack = now->internal;
544 static int __nd_alloc_stack(struct nameidata *nd)
548 if (nd->flags & LOOKUP_RCU) {
549 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
554 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
559 memcpy(p, nd->internal, sizeof(nd->internal));
565 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
566 * @path: nameidate to verify
568 * Rename can sometimes move a file or directory outside of a bind
569 * mount, path_connected allows those cases to be detected.
571 static bool path_connected(const struct path *path)
573 struct vfsmount *mnt = path->mnt;
574 struct super_block *sb = mnt->mnt_sb;
576 /* Bind mounts and multi-root filesystems can have disconnected paths */
577 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
580 return is_subdir(path->dentry, mnt->mnt_root);
583 static inline int nd_alloc_stack(struct nameidata *nd)
585 if (likely(nd->depth != EMBEDDED_LEVELS))
587 if (likely(nd->stack != nd->internal))
589 return __nd_alloc_stack(nd);
592 static void drop_links(struct nameidata *nd)
596 struct saved *last = nd->stack + i;
597 struct inode *inode = last->inode;
598 if (last->cookie && inode->i_op->put_link) {
599 inode->i_op->put_link(inode, last->cookie);
605 static void terminate_walk(struct nameidata *nd)
608 if (!(nd->flags & LOOKUP_RCU)) {
611 for (i = 0; i < nd->depth; i++)
612 path_put(&nd->stack[i].link);
613 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
618 nd->flags &= ~LOOKUP_RCU;
619 if (!(nd->flags & LOOKUP_ROOT))
626 /* path_put is needed afterwards regardless of success or failure */
627 static bool legitimize_path(struct nameidata *nd,
628 struct path *path, unsigned seq)
630 int res = __legitimize_mnt(path->mnt, nd->m_seq);
637 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
641 return !read_seqcount_retry(&path->dentry->d_seq, seq);
644 static bool legitimize_links(struct nameidata *nd)
647 for (i = 0; i < nd->depth; i++) {
648 struct saved *last = nd->stack + i;
649 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
659 * Path walking has 2 modes, rcu-walk and ref-walk (see
660 * Documentation/filesystems/path-lookup.txt). In situations when we can't
661 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
662 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
663 * mode. Refcounts are grabbed at the last known good point before rcu-walk
664 * got stuck, so ref-walk may continue from there. If this is not successful
665 * (eg. a seqcount has changed), then failure is returned and it's up to caller
666 * to restart the path walk from the beginning in ref-walk mode.
670 * unlazy_walk - try to switch to ref-walk mode.
671 * @nd: nameidata pathwalk data
672 * @dentry: child of nd->path.dentry or NULL
673 * @seq: seq number to check dentry against
674 * Returns: 0 on success, -ECHILD on failure
676 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
677 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
678 * @nd or NULL. Must be called from rcu-walk context.
679 * Nothing should touch nameidata between unlazy_walk() failure and
682 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
684 struct dentry *parent = nd->path.dentry;
686 BUG_ON(!(nd->flags & LOOKUP_RCU));
688 nd->flags &= ~LOOKUP_RCU;
689 if (unlikely(!legitimize_links(nd)))
691 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
693 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
697 * For a negative lookup, the lookup sequence point is the parents
698 * sequence point, and it only needs to revalidate the parent dentry.
700 * For a positive lookup, we need to move both the parent and the
701 * dentry from the RCU domain to be properly refcounted. And the
702 * sequence number in the dentry validates *both* dentry counters,
703 * since we checked the sequence number of the parent after we got
704 * the child sequence number. So we know the parent must still
705 * be valid if the child sequence number is still valid.
708 if (read_seqcount_retry(&parent->d_seq, nd->seq))
710 BUG_ON(nd->inode != parent->d_inode);
712 if (!lockref_get_not_dead(&dentry->d_lockref))
714 if (read_seqcount_retry(&dentry->d_seq, seq))
719 * Sequence counts matched. Now make sure that the root is
720 * still valid and get it if required.
722 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
723 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
740 nd->path.dentry = NULL;
744 if (!(nd->flags & LOOKUP_ROOT))
749 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
751 if (unlikely(!legitimize_path(nd, link, seq))) {
754 nd->flags &= ~LOOKUP_RCU;
756 nd->path.dentry = NULL;
757 if (!(nd->flags & LOOKUP_ROOT))
760 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
767 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
769 return dentry->d_op->d_revalidate(dentry, flags);
773 * complete_walk - successful completion of path walk
774 * @nd: pointer nameidata
776 * If we had been in RCU mode, drop out of it and legitimize nd->path.
777 * Revalidate the final result, unless we'd already done that during
778 * the path walk or the filesystem doesn't ask for it. Return 0 on
779 * success, -error on failure. In case of failure caller does not
780 * need to drop nd->path.
782 static int complete_walk(struct nameidata *nd)
784 struct dentry *dentry = nd->path.dentry;
787 if (nd->flags & LOOKUP_RCU) {
788 if (!(nd->flags & LOOKUP_ROOT))
790 if (unlikely(unlazy_walk(nd, NULL, 0)))
794 if (likely(!(nd->flags & LOOKUP_JUMPED)))
797 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
800 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
810 static void set_root(struct nameidata *nd)
812 get_fs_root(current->fs, &nd->root);
815 static void set_root_rcu(struct nameidata *nd)
817 struct fs_struct *fs = current->fs;
821 seq = read_seqcount_begin(&fs->seq);
823 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
824 } while (read_seqcount_retry(&fs->seq, seq));
827 static void path_put_conditional(struct path *path, struct nameidata *nd)
830 if (path->mnt != nd->path.mnt)
834 static inline void path_to_nameidata(const struct path *path,
835 struct nameidata *nd)
837 if (!(nd->flags & LOOKUP_RCU)) {
838 dput(nd->path.dentry);
839 if (nd->path.mnt != path->mnt)
840 mntput(nd->path.mnt);
842 nd->path.mnt = path->mnt;
843 nd->path.dentry = path->dentry;
847 * Helper to directly jump to a known parsed path from ->follow_link,
848 * caller must have taken a reference to path beforehand.
850 void nd_jump_link(struct path *path)
852 struct nameidata *nd = current->nameidata;
856 nd->inode = nd->path.dentry->d_inode;
857 nd->flags |= LOOKUP_JUMPED;
860 static inline void put_link(struct nameidata *nd)
862 struct saved *last = nd->stack + --nd->depth;
863 struct inode *inode = last->inode;
864 if (last->cookie && inode->i_op->put_link)
865 inode->i_op->put_link(inode, last->cookie);
866 if (!(nd->flags & LOOKUP_RCU))
867 path_put(&last->link);
870 int sysctl_protected_symlinks __read_mostly = 0;
871 int sysctl_protected_hardlinks __read_mostly = 0;
872 int sysctl_protected_fifos __read_mostly;
873 int sysctl_protected_regular __read_mostly;
876 * may_follow_link - Check symlink following for unsafe situations
877 * @nd: nameidata pathwalk data
879 * In the case of the sysctl_protected_symlinks sysctl being enabled,
880 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
881 * in a sticky world-writable directory. This is to protect privileged
882 * processes from failing races against path names that may change out
883 * from under them by way of other users creating malicious symlinks.
884 * It will permit symlinks to be followed only when outside a sticky
885 * world-writable directory, or when the uid of the symlink and follower
886 * match, or when the directory owner matches the symlink's owner.
888 * Returns 0 if following the symlink is allowed, -ve on error.
890 static inline int may_follow_link(struct nameidata *nd)
892 const struct inode *inode;
893 const struct inode *parent;
896 if (!sysctl_protected_symlinks)
899 /* Allowed if owner and follower match. */
900 inode = nd->stack[0].inode;
901 if (uid_eq(current_cred()->fsuid, inode->i_uid))
904 /* Allowed if parent directory not sticky and world-writable. */
906 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
909 /* Allowed if parent directory and link owner match. */
910 puid = parent->i_uid;
911 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
914 if (nd->flags & LOOKUP_RCU)
917 audit_log_link_denied("follow_link", &nd->stack[0].link);
922 * safe_hardlink_source - Check for safe hardlink conditions
923 * @inode: the source inode to hardlink from
925 * Return false if at least one of the following conditions:
926 * - inode is not a regular file
928 * - inode is setgid and group-exec
929 * - access failure for read and write
931 * Otherwise returns true.
933 static bool safe_hardlink_source(struct inode *inode)
935 umode_t mode = inode->i_mode;
937 /* Special files should not get pinned to the filesystem. */
941 /* Setuid files should not get pinned to the filesystem. */
945 /* Executable setgid files should not get pinned to the filesystem. */
946 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
949 /* Hardlinking to unreadable or unwritable sources is dangerous. */
950 if (inode_permission(inode, MAY_READ | MAY_WRITE))
957 * may_linkat - Check permissions for creating a hardlink
958 * @link: the source to hardlink from
960 * Block hardlink when all of:
961 * - sysctl_protected_hardlinks enabled
962 * - fsuid does not match inode
963 * - hardlink source is unsafe (see safe_hardlink_source() above)
964 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
966 * Returns 0 if successful, -ve on error.
968 static int may_linkat(struct path *link)
972 if (!sysctl_protected_hardlinks)
975 inode = link->dentry->d_inode;
977 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
978 * otherwise, it must be a safe source.
980 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
983 audit_log_link_denied("linkat", link);
988 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
989 * should be allowed, or not, on files that already
991 * @dir_mode: mode bits of directory
992 * @dir_uid: owner of directory
993 * @inode: the inode of the file to open
995 * Block an O_CREAT open of a FIFO (or a regular file) when:
996 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
997 * - the file already exists
998 * - we are in a sticky directory
999 * - we don't own the file
1000 * - the owner of the directory doesn't own the file
1001 * - the directory is world writable
1002 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1003 * the directory doesn't have to be world writable: being group writable will
1006 * Returns 0 if the open is allowed, -ve on error.
1008 static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid,
1009 struct inode * const inode)
1011 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1012 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1013 likely(!(dir_mode & S_ISVTX)) ||
1014 uid_eq(inode->i_uid, dir_uid) ||
1015 uid_eq(current_fsuid(), inode->i_uid))
1018 if (likely(dir_mode & 0002) ||
1020 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1021 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1027 static __always_inline
1028 const char *get_link(struct nameidata *nd)
1030 struct saved *last = nd->stack + nd->depth - 1;
1031 struct dentry *dentry = last->link.dentry;
1032 struct inode *inode = last->inode;
1036 if (!(nd->flags & LOOKUP_RCU)) {
1037 touch_atime(&last->link);
1039 } else if (atime_needs_update(&last->link, inode)) {
1040 if (unlikely(unlazy_walk(nd, NULL, 0)))
1041 return ERR_PTR(-ECHILD);
1042 touch_atime(&last->link);
1045 error = security_inode_follow_link(dentry, inode,
1046 nd->flags & LOOKUP_RCU);
1047 if (unlikely(error))
1048 return ERR_PTR(error);
1050 nd->last_type = LAST_BIND;
1051 res = inode->i_link;
1053 if (nd->flags & LOOKUP_RCU) {
1054 if (unlikely(unlazy_walk(nd, NULL, 0)))
1055 return ERR_PTR(-ECHILD);
1057 res = inode->i_op->follow_link(dentry, &last->cookie);
1058 if (IS_ERR_OR_NULL(res)) {
1059 last->cookie = NULL;
1064 if (nd->flags & LOOKUP_RCU) {
1068 nd->path = nd->root;
1069 d = nd->path.dentry;
1070 nd->inode = d->d_inode;
1071 nd->seq = nd->root_seq;
1072 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
1073 return ERR_PTR(-ECHILD);
1077 path_put(&nd->path);
1078 nd->path = nd->root;
1079 path_get(&nd->root);
1080 nd->inode = nd->path.dentry->d_inode;
1082 nd->flags |= LOOKUP_JUMPED;
1083 while (unlikely(*++res == '/'))
1092 * follow_up - Find the mountpoint of path's vfsmount
1094 * Given a path, find the mountpoint of its source file system.
1095 * Replace @path with the path of the mountpoint in the parent mount.
1098 * Return 1 if we went up a level and 0 if we were already at the
1101 int follow_up(struct path *path)
1103 struct mount *mnt = real_mount(path->mnt);
1104 struct mount *parent;
1105 struct dentry *mountpoint;
1107 read_seqlock_excl(&mount_lock);
1108 parent = mnt->mnt_parent;
1109 if (parent == mnt) {
1110 read_sequnlock_excl(&mount_lock);
1113 mntget(&parent->mnt);
1114 mountpoint = dget(mnt->mnt_mountpoint);
1115 read_sequnlock_excl(&mount_lock);
1117 path->dentry = mountpoint;
1119 path->mnt = &parent->mnt;
1122 EXPORT_SYMBOL(follow_up);
1125 * Perform an automount
1126 * - return -EISDIR to tell follow_managed() to stop and return the path we
1129 static int follow_automount(struct path *path, struct nameidata *nd,
1132 struct vfsmount *mnt;
1135 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1138 /* We don't want to mount if someone's just doing a stat -
1139 * unless they're stat'ing a directory and appended a '/' to
1142 * We do, however, want to mount if someone wants to open or
1143 * create a file of any type under the mountpoint, wants to
1144 * traverse through the mountpoint or wants to open the
1145 * mounted directory. Also, autofs may mark negative dentries
1146 * as being automount points. These will need the attentions
1147 * of the daemon to instantiate them before they can be used.
1149 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1150 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1151 path->dentry->d_inode)
1154 nd->total_link_count++;
1155 if (nd->total_link_count >= 40)
1158 mnt = path->dentry->d_op->d_automount(path);
1161 * The filesystem is allowed to return -EISDIR here to indicate
1162 * it doesn't want to automount. For instance, autofs would do
1163 * this so that its userspace daemon can mount on this dentry.
1165 * However, we can only permit this if it's a terminal point in
1166 * the path being looked up; if it wasn't then the remainder of
1167 * the path is inaccessible and we should say so.
1169 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1171 return PTR_ERR(mnt);
1174 if (!mnt) /* mount collision */
1177 if (!*need_mntput) {
1178 /* lock_mount() may release path->mnt on error */
1180 *need_mntput = true;
1182 err = finish_automount(mnt, path);
1186 /* Someone else made a mount here whilst we were busy */
1191 path->dentry = dget(mnt->mnt_root);
1200 * Handle a dentry that is managed in some way.
1201 * - Flagged for transit management (autofs)
1202 * - Flagged as mountpoint
1203 * - Flagged as automount point
1205 * This may only be called in refwalk mode.
1207 * Serialization is taken care of in namespace.c
1209 static int follow_managed(struct path *path, struct nameidata *nd)
1211 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1213 bool need_mntput = false;
1216 /* Given that we're not holding a lock here, we retain the value in a
1217 * local variable for each dentry as we look at it so that we don't see
1218 * the components of that value change under us */
1219 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1220 managed &= DCACHE_MANAGED_DENTRY,
1221 unlikely(managed != 0)) {
1222 /* Allow the filesystem to manage the transit without i_mutex
1224 if (managed & DCACHE_MANAGE_TRANSIT) {
1225 BUG_ON(!path->dentry->d_op);
1226 BUG_ON(!path->dentry->d_op->d_manage);
1227 ret = path->dentry->d_op->d_manage(path->dentry, false);
1232 /* Transit to a mounted filesystem. */
1233 if (managed & DCACHE_MOUNTED) {
1234 struct vfsmount *mounted = lookup_mnt(path);
1239 path->mnt = mounted;
1240 path->dentry = dget(mounted->mnt_root);
1245 /* Something is mounted on this dentry in another
1246 * namespace and/or whatever was mounted there in this
1247 * namespace got unmounted before lookup_mnt() could
1251 /* Handle an automount point */
1252 if (managed & DCACHE_NEED_AUTOMOUNT) {
1253 ret = follow_automount(path, nd, &need_mntput);
1259 /* We didn't change the current path point */
1263 if (need_mntput && path->mnt == mnt)
1268 nd->flags |= LOOKUP_JUMPED;
1269 if (unlikely(ret < 0))
1270 path_put_conditional(path, nd);
1274 int follow_down_one(struct path *path)
1276 struct vfsmount *mounted;
1278 mounted = lookup_mnt(path);
1282 path->mnt = mounted;
1283 path->dentry = dget(mounted->mnt_root);
1288 EXPORT_SYMBOL(follow_down_one);
1290 static inline int managed_dentry_rcu(struct dentry *dentry)
1292 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1293 dentry->d_op->d_manage(dentry, true) : 0;
1297 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1298 * we meet a managed dentry that would need blocking.
1300 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1301 struct inode **inode, unsigned *seqp)
1304 struct mount *mounted;
1306 * Don't forget we might have a non-mountpoint managed dentry
1307 * that wants to block transit.
1309 switch (managed_dentry_rcu(path->dentry)) {
1319 if (!d_mountpoint(path->dentry))
1320 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1322 mounted = __lookup_mnt(path->mnt, path->dentry);
1325 path->mnt = &mounted->mnt;
1326 path->dentry = mounted->mnt.mnt_root;
1327 nd->flags |= LOOKUP_JUMPED;
1328 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1330 * Update the inode too. We don't need to re-check the
1331 * dentry sequence number here after this d_inode read,
1332 * because a mount-point is always pinned.
1334 *inode = path->dentry->d_inode;
1336 return !read_seqretry(&mount_lock, nd->m_seq) &&
1337 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1340 static int follow_dotdot_rcu(struct nameidata *nd)
1342 struct inode *inode = nd->inode;
1347 if (path_equal(&nd->path, &nd->root))
1349 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1350 struct dentry *old = nd->path.dentry;
1351 struct dentry *parent = old->d_parent;
1354 inode = parent->d_inode;
1355 seq = read_seqcount_begin(&parent->d_seq);
1356 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1358 nd->path.dentry = parent;
1360 if (unlikely(!path_connected(&nd->path)))
1364 struct mount *mnt = real_mount(nd->path.mnt);
1365 struct mount *mparent = mnt->mnt_parent;
1366 struct dentry *mountpoint = mnt->mnt_mountpoint;
1367 struct inode *inode2 = mountpoint->d_inode;
1368 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1369 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1371 if (&mparent->mnt == nd->path.mnt)
1373 /* we know that mountpoint was pinned */
1374 nd->path.dentry = mountpoint;
1375 nd->path.mnt = &mparent->mnt;
1380 while (unlikely(d_mountpoint(nd->path.dentry))) {
1381 struct mount *mounted;
1382 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1383 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1387 nd->path.mnt = &mounted->mnt;
1388 nd->path.dentry = mounted->mnt.mnt_root;
1389 inode = nd->path.dentry->d_inode;
1390 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1397 * Follow down to the covering mount currently visible to userspace. At each
1398 * point, the filesystem owning that dentry may be queried as to whether the
1399 * caller is permitted to proceed or not.
1401 int follow_down(struct path *path)
1406 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1407 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1408 /* Allow the filesystem to manage the transit without i_mutex
1411 * We indicate to the filesystem if someone is trying to mount
1412 * something here. This gives autofs the chance to deny anyone
1413 * other than its daemon the right to mount on its
1416 * The filesystem may sleep at this point.
1418 if (managed & DCACHE_MANAGE_TRANSIT) {
1419 BUG_ON(!path->dentry->d_op);
1420 BUG_ON(!path->dentry->d_op->d_manage);
1421 ret = path->dentry->d_op->d_manage(
1422 path->dentry, false);
1424 return ret == -EISDIR ? 0 : ret;
1427 /* Transit to a mounted filesystem. */
1428 if (managed & DCACHE_MOUNTED) {
1429 struct vfsmount *mounted = lookup_mnt(path);
1434 path->mnt = mounted;
1435 path->dentry = dget(mounted->mnt_root);
1439 /* Don't handle automount points here */
1444 EXPORT_SYMBOL(follow_down);
1447 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1449 static void follow_mount(struct path *path)
1451 while (d_mountpoint(path->dentry)) {
1452 struct vfsmount *mounted = lookup_mnt(path);
1457 path->mnt = mounted;
1458 path->dentry = dget(mounted->mnt_root);
1462 static int follow_dotdot(struct nameidata *nd)
1468 struct dentry *old = nd->path.dentry;
1470 if (nd->path.dentry == nd->root.dentry &&
1471 nd->path.mnt == nd->root.mnt) {
1474 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1475 /* rare case of legitimate dget_parent()... */
1476 nd->path.dentry = dget_parent(nd->path.dentry);
1478 if (unlikely(!path_connected(&nd->path)))
1482 if (!follow_up(&nd->path))
1485 follow_mount(&nd->path);
1486 nd->inode = nd->path.dentry->d_inode;
1491 * This looks up the name in dcache, possibly revalidates the old dentry and
1492 * allocates a new one if not found or not valid. In the need_lookup argument
1493 * returns whether i_op->lookup is necessary.
1495 * dir->d_inode->i_mutex must be held
1497 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1498 unsigned int flags, bool *need_lookup)
1500 struct dentry *dentry;
1503 *need_lookup = false;
1504 dentry = d_lookup(dir, name);
1506 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1507 error = d_revalidate(dentry, flags);
1508 if (unlikely(error <= 0)) {
1511 return ERR_PTR(error);
1513 d_invalidate(dentry);
1522 dentry = d_alloc(dir, name);
1523 if (unlikely(!dentry))
1524 return ERR_PTR(-ENOMEM);
1526 *need_lookup = true;
1532 * Call i_op->lookup on the dentry. The dentry must be negative and
1535 * dir->d_inode->i_mutex must be held
1537 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1542 /* Don't create child dentry for a dead directory. */
1543 if (unlikely(IS_DEADDIR(dir))) {
1545 return ERR_PTR(-ENOENT);
1548 old = dir->i_op->lookup(dir, dentry, flags);
1549 if (unlikely(old)) {
1556 static struct dentry *__lookup_hash(struct qstr *name,
1557 struct dentry *base, unsigned int flags)
1560 struct dentry *dentry;
1562 dentry = lookup_dcache(name, base, flags, &need_lookup);
1566 return lookup_real(base->d_inode, dentry, flags);
1570 * It's more convoluted than I'd like it to be, but... it's still fairly
1571 * small and for now I'd prefer to have fast path as straight as possible.
1572 * It _is_ time-critical.
1574 static int lookup_fast(struct nameidata *nd,
1575 struct path *path, struct inode **inode,
1578 struct vfsmount *mnt = nd->path.mnt;
1579 struct dentry *dentry, *parent = nd->path.dentry;
1585 * Rename seqlock is not required here because in the off chance
1586 * of a false negative due to a concurrent rename, we're going to
1587 * do the non-racy lookup, below.
1589 if (nd->flags & LOOKUP_RCU) {
1592 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1597 * This sequence count validates that the inode matches
1598 * the dentry name information from lookup.
1600 *inode = d_backing_inode(dentry);
1601 negative = d_is_negative(dentry);
1602 if (read_seqcount_retry(&dentry->d_seq, seq))
1606 * This sequence count validates that the parent had no
1607 * changes while we did the lookup of the dentry above.
1609 * The memory barrier in read_seqcount_begin of child is
1610 * enough, we can use __read_seqcount_retry here.
1612 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1616 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1617 status = d_revalidate(dentry, nd->flags);
1618 if (unlikely(status <= 0)) {
1619 if (status != -ECHILD)
1625 * Note: do negative dentry check after revalidation in
1626 * case that drops it.
1631 path->dentry = dentry;
1632 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1635 if (unlazy_walk(nd, dentry, seq))
1638 dentry = __d_lookup(parent, &nd->last);
1641 if (unlikely(!dentry))
1644 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1645 status = d_revalidate(dentry, nd->flags);
1646 if (unlikely(status <= 0)) {
1651 d_invalidate(dentry);
1656 if (unlikely(d_is_negative(dentry))) {
1661 path->dentry = dentry;
1662 err = follow_managed(path, nd);
1664 *inode = d_backing_inode(path->dentry);
1671 /* Fast lookup failed, do it the slow way */
1672 static int lookup_slow(struct nameidata *nd, struct path *path)
1674 struct dentry *dentry, *parent;
1676 parent = nd->path.dentry;
1677 BUG_ON(nd->inode != parent->d_inode);
1679 mutex_lock(&parent->d_inode->i_mutex);
1680 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1681 mutex_unlock(&parent->d_inode->i_mutex);
1683 return PTR_ERR(dentry);
1684 path->mnt = nd->path.mnt;
1685 path->dentry = dentry;
1686 return follow_managed(path, nd);
1689 static inline int may_lookup(struct nameidata *nd)
1691 if (nd->flags & LOOKUP_RCU) {
1692 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1695 if (unlazy_walk(nd, NULL, 0))
1698 return inode_permission(nd->inode, MAY_EXEC);
1701 static inline int handle_dots(struct nameidata *nd, int type)
1703 if (type == LAST_DOTDOT) {
1704 if (nd->flags & LOOKUP_RCU) {
1705 return follow_dotdot_rcu(nd);
1707 return follow_dotdot(nd);
1712 static int pick_link(struct nameidata *nd, struct path *link,
1713 struct inode *inode, unsigned seq)
1717 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1718 path_to_nameidata(link, nd);
1721 if (!(nd->flags & LOOKUP_RCU)) {
1722 if (link->mnt == nd->path.mnt)
1725 error = nd_alloc_stack(nd);
1726 if (unlikely(error)) {
1727 if (error == -ECHILD) {
1728 if (unlikely(unlazy_link(nd, link, seq)))
1730 error = nd_alloc_stack(nd);
1738 last = nd->stack + nd->depth++;
1740 last->cookie = NULL;
1741 last->inode = inode;
1747 * Do we need to follow links? We _really_ want to be able
1748 * to do this check without having to look at inode->i_op,
1749 * so we keep a cache of "no, this doesn't need follow_link"
1750 * for the common case.
1752 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1754 struct inode *inode, unsigned seq)
1756 if (likely(!d_is_symlink(link->dentry)))
1760 /* make sure that d_is_symlink above matches inode */
1761 if (nd->flags & LOOKUP_RCU) {
1762 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1765 return pick_link(nd, link, inode, seq);
1768 enum {WALK_GET = 1, WALK_PUT = 2};
1770 static int walk_component(struct nameidata *nd, int flags)
1773 struct inode *inode;
1777 * "." and ".." are special - ".." especially so because it has
1778 * to be able to know about the current root directory and
1779 * parent relationships.
1781 if (unlikely(nd->last_type != LAST_NORM)) {
1782 err = handle_dots(nd, nd->last_type);
1783 if (flags & WALK_PUT)
1787 err = lookup_fast(nd, &path, &inode, &seq);
1788 if (unlikely(err)) {
1792 err = lookup_slow(nd, &path);
1796 seq = 0; /* we are already out of RCU mode */
1798 if (d_is_negative(path.dentry))
1800 inode = d_backing_inode(path.dentry);
1803 if (flags & WALK_PUT)
1805 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1808 path_to_nameidata(&path, nd);
1814 path_to_nameidata(&path, nd);
1819 * We can do the critical dentry name comparison and hashing
1820 * operations one word at a time, but we are limited to:
1822 * - Architectures with fast unaligned word accesses. We could
1823 * do a "get_unaligned()" if this helps and is sufficiently
1826 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1827 * do not trap on the (extremely unlikely) case of a page
1828 * crossing operation.
1830 * - Furthermore, we need an efficient 64-bit compile for the
1831 * 64-bit case in order to generate the "number of bytes in
1832 * the final mask". Again, that could be replaced with a
1833 * efficient population count instruction or similar.
1835 #ifdef CONFIG_DCACHE_WORD_ACCESS
1837 #include <asm/word-at-a-time.h>
1841 static inline unsigned int fold_hash(unsigned long hash)
1843 return hash_64(hash, 32);
1846 #else /* 32-bit case */
1848 #define fold_hash(x) (x)
1852 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1854 unsigned long a, mask;
1855 unsigned long hash = 0;
1858 a = load_unaligned_zeropad(name);
1859 if (len < sizeof(unsigned long))
1863 name += sizeof(unsigned long);
1864 len -= sizeof(unsigned long);
1868 mask = bytemask_from_count(len);
1871 return fold_hash(hash);
1873 EXPORT_SYMBOL(full_name_hash);
1876 * Calculate the length and hash of the path component, and
1877 * return the "hash_len" as the result.
1879 static inline u64 hash_name(const char *name)
1881 unsigned long a, b, adata, bdata, mask, hash, len;
1882 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1885 len = -sizeof(unsigned long);
1887 hash = (hash + a) * 9;
1888 len += sizeof(unsigned long);
1889 a = load_unaligned_zeropad(name+len);
1890 b = a ^ REPEAT_BYTE('/');
1891 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1893 adata = prep_zero_mask(a, adata, &constants);
1894 bdata = prep_zero_mask(b, bdata, &constants);
1896 mask = create_zero_mask(adata | bdata);
1898 hash += a & zero_bytemask(mask);
1899 len += find_zero(mask);
1900 return hashlen_create(fold_hash(hash), len);
1905 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1907 unsigned long hash = init_name_hash();
1909 hash = partial_name_hash(*name++, hash);
1910 return end_name_hash(hash);
1912 EXPORT_SYMBOL(full_name_hash);
1915 * We know there's a real path component here of at least
1918 static inline u64 hash_name(const char *name)
1920 unsigned long hash = init_name_hash();
1921 unsigned long len = 0, c;
1923 c = (unsigned char)*name;
1926 hash = partial_name_hash(c, hash);
1927 c = (unsigned char)name[len];
1928 } while (c && c != '/');
1929 return hashlen_create(end_name_hash(hash), len);
1936 * This is the basic name resolution function, turning a pathname into
1937 * the final dentry. We expect 'base' to be positive and a directory.
1939 * Returns 0 and nd will have valid dentry and mnt on success.
1940 * Returns error and drops reference to input namei data on failure.
1942 static int link_path_walk(const char *name, struct nameidata *nd)
1951 /* At this point we know we have a real path component. */
1956 err = may_lookup(nd);
1960 hash_len = hash_name(name);
1963 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1965 if (name[1] == '.') {
1967 nd->flags |= LOOKUP_JUMPED;
1973 if (likely(type == LAST_NORM)) {
1974 struct dentry *parent = nd->path.dentry;
1975 nd->flags &= ~LOOKUP_JUMPED;
1976 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1977 struct qstr this = { { .hash_len = hash_len }, .name = name };
1978 err = parent->d_op->d_hash(parent, &this);
1981 hash_len = this.hash_len;
1986 nd->last.hash_len = hash_len;
1987 nd->last.name = name;
1988 nd->last_type = type;
1990 name += hashlen_len(hash_len);
1994 * If it wasn't NUL, we know it was '/'. Skip that
1995 * slash, and continue until no more slashes.
1999 } while (unlikely(*name == '/'));
2000 if (unlikely(!*name)) {
2002 /* pathname body, done */
2005 name = nd->stack[nd->depth - 1].name;
2006 /* trailing symlink, done */
2009 /* last component of nested symlink */
2010 err = walk_component(nd, WALK_GET | WALK_PUT);
2012 err = walk_component(nd, WALK_GET);
2018 const char *s = get_link(nd);
2027 nd->stack[nd->depth - 1].name = name;
2032 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2033 if (nd->flags & LOOKUP_RCU) {
2034 if (unlazy_walk(nd, NULL, 0))
2042 static const char *path_init(struct nameidata *nd, unsigned flags)
2045 const char *s = nd->name->name;
2048 flags &= ~LOOKUP_RCU;
2050 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2051 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2053 if (flags & LOOKUP_ROOT) {
2054 struct dentry *root = nd->root.dentry;
2055 struct inode *inode = root->d_inode;
2057 if (!d_can_lookup(root))
2058 return ERR_PTR(-ENOTDIR);
2059 retval = inode_permission(inode, MAY_EXEC);
2061 return ERR_PTR(retval);
2063 nd->path = nd->root;
2065 if (flags & LOOKUP_RCU) {
2067 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2068 nd->root_seq = nd->seq;
2069 nd->m_seq = read_seqbegin(&mount_lock);
2071 path_get(&nd->path);
2076 nd->root.mnt = NULL;
2078 nd->m_seq = read_seqbegin(&mount_lock);
2080 if (flags & LOOKUP_RCU) {
2083 nd->seq = nd->root_seq;
2086 path_get(&nd->root);
2088 nd->path = nd->root;
2089 } else if (nd->dfd == AT_FDCWD) {
2090 if (flags & LOOKUP_RCU) {
2091 struct fs_struct *fs = current->fs;
2097 seq = read_seqcount_begin(&fs->seq);
2099 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2100 } while (read_seqcount_retry(&fs->seq, seq));
2102 get_fs_pwd(current->fs, &nd->path);
2105 /* Caller must check execute permissions on the starting path component */
2106 struct fd f = fdget_raw(nd->dfd);
2107 struct dentry *dentry;
2110 return ERR_PTR(-EBADF);
2112 dentry = f.file->f_path.dentry;
2115 if (!d_can_lookup(dentry)) {
2117 return ERR_PTR(-ENOTDIR);
2121 nd->path = f.file->f_path;
2122 if (flags & LOOKUP_RCU) {
2124 nd->inode = nd->path.dentry->d_inode;
2125 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2127 path_get(&nd->path);
2128 nd->inode = nd->path.dentry->d_inode;
2134 nd->inode = nd->path.dentry->d_inode;
2135 if (!(flags & LOOKUP_RCU))
2137 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
2139 if (!(nd->flags & LOOKUP_ROOT))
2140 nd->root.mnt = NULL;
2142 return ERR_PTR(-ECHILD);
2145 static const char *trailing_symlink(struct nameidata *nd)
2148 int error = may_follow_link(nd);
2149 if (unlikely(error))
2150 return ERR_PTR(error);
2151 nd->flags |= LOOKUP_PARENT;
2152 nd->stack[0].name = NULL;
2157 static inline int lookup_last(struct nameidata *nd)
2159 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2160 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2162 nd->flags &= ~LOOKUP_PARENT;
2163 return walk_component(nd,
2164 nd->flags & LOOKUP_FOLLOW
2166 ? WALK_PUT | WALK_GET
2171 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2172 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2174 const char *s = path_init(nd, flags);
2179 while (!(err = link_path_walk(s, nd))
2180 && ((err = lookup_last(nd)) > 0)) {
2181 s = trailing_symlink(nd);
2188 err = complete_walk(nd);
2190 if (!err && nd->flags & LOOKUP_DIRECTORY)
2191 if (!d_can_lookup(nd->path.dentry))
2195 nd->path.mnt = NULL;
2196 nd->path.dentry = NULL;
2202 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2203 struct path *path, struct path *root)
2206 struct nameidata nd;
2208 return PTR_ERR(name);
2209 if (unlikely(root)) {
2211 flags |= LOOKUP_ROOT;
2213 set_nameidata(&nd, dfd, name);
2214 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2215 if (unlikely(retval == -ECHILD))
2216 retval = path_lookupat(&nd, flags, path);
2217 if (unlikely(retval == -ESTALE))
2218 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2220 if (likely(!retval))
2221 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2222 restore_nameidata();
2227 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2228 static int path_parentat(struct nameidata *nd, unsigned flags,
2229 struct path *parent)
2231 const char *s = path_init(nd, flags);
2235 err = link_path_walk(s, nd);
2237 err = complete_walk(nd);
2240 nd->path.mnt = NULL;
2241 nd->path.dentry = NULL;
2247 static struct filename *filename_parentat(int dfd, struct filename *name,
2248 unsigned int flags, struct path *parent,
2249 struct qstr *last, int *type)
2252 struct nameidata nd;
2256 set_nameidata(&nd, dfd, name);
2257 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2258 if (unlikely(retval == -ECHILD))
2259 retval = path_parentat(&nd, flags, parent);
2260 if (unlikely(retval == -ESTALE))
2261 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2262 if (likely(!retval)) {
2264 *type = nd.last_type;
2265 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2268 name = ERR_PTR(retval);
2270 restore_nameidata();
2274 /* does lookup, returns the object with parent locked */
2275 struct dentry *kern_path_locked(const char *name, struct path *path)
2277 struct filename *filename;
2282 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2284 if (IS_ERR(filename))
2285 return ERR_CAST(filename);
2286 if (unlikely(type != LAST_NORM)) {
2289 return ERR_PTR(-EINVAL);
2291 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2292 d = __lookup_hash(&last, path->dentry, 0);
2294 mutex_unlock(&path->dentry->d_inode->i_mutex);
2301 int kern_path(const char *name, unsigned int flags, struct path *path)
2303 return filename_lookup(AT_FDCWD, getname_kernel(name),
2306 EXPORT_SYMBOL(kern_path);
2309 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2310 * @dentry: pointer to dentry of the base directory
2311 * @mnt: pointer to vfs mount of the base directory
2312 * @name: pointer to file name
2313 * @flags: lookup flags
2314 * @path: pointer to struct path to fill
2316 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2317 const char *name, unsigned int flags,
2320 struct path root = {.mnt = mnt, .dentry = dentry};
2321 /* the first argument of filename_lookup() is ignored with root */
2322 return filename_lookup(AT_FDCWD, getname_kernel(name),
2323 flags , path, &root);
2325 EXPORT_SYMBOL(vfs_path_lookup);
2328 * lookup_one_len - filesystem helper to lookup single pathname component
2329 * @name: pathname component to lookup
2330 * @base: base directory to lookup from
2331 * @len: maximum length @len should be interpreted to
2333 * Note that this routine is purely a helper for filesystem usage and should
2334 * not be called by generic code.
2336 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2342 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2346 this.hash = full_name_hash(name, len);
2348 return ERR_PTR(-EACCES);
2350 if (unlikely(name[0] == '.')) {
2351 if (len < 2 || (len == 2 && name[1] == '.'))
2352 return ERR_PTR(-EACCES);
2356 c = *(const unsigned char *)name++;
2357 if (c == '/' || c == '\0')
2358 return ERR_PTR(-EACCES);
2361 * See if the low-level filesystem might want
2362 * to use its own hash..
2364 if (base->d_flags & DCACHE_OP_HASH) {
2365 int err = base->d_op->d_hash(base, &this);
2367 return ERR_PTR(err);
2370 err = inode_permission(base->d_inode, MAY_EXEC);
2372 return ERR_PTR(err);
2374 return __lookup_hash(&this, base, 0);
2376 EXPORT_SYMBOL(lookup_one_len);
2378 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2379 struct path *path, int *empty)
2381 return filename_lookup(dfd, getname_flags(name, flags, empty),
2384 EXPORT_SYMBOL(user_path_at_empty);
2387 * NB: most callers don't do anything directly with the reference to the
2388 * to struct filename, but the nd->last pointer points into the name string
2389 * allocated by getname. So we must hold the reference to it until all
2390 * path-walking is complete.
2392 static inline struct filename *
2393 user_path_parent(int dfd, const char __user *path,
2394 struct path *parent,
2399 /* only LOOKUP_REVAL is allowed in extra flags */
2400 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2401 parent, last, type);
2405 * mountpoint_last - look up last component for umount
2406 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2407 * @path: pointer to container for result
2409 * This is a special lookup_last function just for umount. In this case, we
2410 * need to resolve the path without doing any revalidation.
2412 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2413 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2414 * in almost all cases, this lookup will be served out of the dcache. The only
2415 * cases where it won't are if nd->last refers to a symlink or the path is
2416 * bogus and it doesn't exist.
2419 * -error: if there was an error during lookup. This includes -ENOENT if the
2420 * lookup found a negative dentry. The nd->path reference will also be
2423 * 0: if we successfully resolved nd->path and found it to not to be a
2424 * symlink that needs to be followed. "path" will also be populated.
2425 * The nd->path reference will also be put.
2427 * 1: if we successfully resolved nd->last and found it to be a symlink
2428 * that needs to be followed. "path" will be populated with the path
2429 * to the link, and nd->path will *not* be put.
2432 mountpoint_last(struct nameidata *nd, struct path *path)
2435 struct dentry *dentry;
2436 struct dentry *dir = nd->path.dentry;
2438 /* If we're in rcuwalk, drop out of it to handle last component */
2439 if (nd->flags & LOOKUP_RCU) {
2440 if (unlazy_walk(nd, NULL, 0))
2444 nd->flags &= ~LOOKUP_PARENT;
2446 if (unlikely(nd->last_type != LAST_NORM)) {
2447 error = handle_dots(nd, nd->last_type);
2450 dentry = dget(nd->path.dentry);
2454 mutex_lock(&dir->d_inode->i_mutex);
2455 dentry = d_lookup(dir, &nd->last);
2458 * No cached dentry. Mounted dentries are pinned in the cache,
2459 * so that means that this dentry is probably a symlink or the
2460 * path doesn't actually point to a mounted dentry.
2462 dentry = d_alloc(dir, &nd->last);
2464 mutex_unlock(&dir->d_inode->i_mutex);
2467 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2468 if (IS_ERR(dentry)) {
2469 mutex_unlock(&dir->d_inode->i_mutex);
2470 return PTR_ERR(dentry);
2473 mutex_unlock(&dir->d_inode->i_mutex);
2476 if (d_is_negative(dentry)) {
2482 path->dentry = dentry;
2483 path->mnt = nd->path.mnt;
2484 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2485 d_backing_inode(dentry), 0);
2486 if (unlikely(error))
2494 * path_mountpoint - look up a path to be umounted
2495 * @nd: lookup context
2496 * @flags: lookup flags
2497 * @path: pointer to container for result
2499 * Look up the given name, but don't attempt to revalidate the last component.
2500 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2503 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2505 const char *s = path_init(nd, flags);
2509 while (!(err = link_path_walk(s, nd)) &&
2510 (err = mountpoint_last(nd, path)) > 0) {
2511 s = trailing_symlink(nd);
2522 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2525 struct nameidata nd;
2528 return PTR_ERR(name);
2529 set_nameidata(&nd, dfd, name);
2530 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2531 if (unlikely(error == -ECHILD))
2532 error = path_mountpoint(&nd, flags, path);
2533 if (unlikely(error == -ESTALE))
2534 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2536 audit_inode(name, path->dentry, 0);
2537 restore_nameidata();
2543 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2544 * @dfd: directory file descriptor
2545 * @name: pathname from userland
2546 * @flags: lookup flags
2547 * @path: pointer to container to hold result
2549 * A umount is a special case for path walking. We're not actually interested
2550 * in the inode in this situation, and ESTALE errors can be a problem. We
2551 * simply want track down the dentry and vfsmount attached at the mountpoint
2552 * and avoid revalidating the last component.
2554 * Returns 0 and populates "path" on success.
2557 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2560 return filename_mountpoint(dfd, getname(name), path, flags);
2564 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2567 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2569 EXPORT_SYMBOL(kern_path_mountpoint);
2571 int __check_sticky(struct inode *dir, struct inode *inode)
2573 kuid_t fsuid = current_fsuid();
2575 if (uid_eq(inode->i_uid, fsuid))
2577 if (uid_eq(dir->i_uid, fsuid))
2579 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2581 EXPORT_SYMBOL(__check_sticky);
2584 * Check whether we can remove a link victim from directory dir, check
2585 * whether the type of victim is right.
2586 * 1. We can't do it if dir is read-only (done in permission())
2587 * 2. We should have write and exec permissions on dir
2588 * 3. We can't remove anything from append-only dir
2589 * 4. We can't do anything with immutable dir (done in permission())
2590 * 5. If the sticky bit on dir is set we should either
2591 * a. be owner of dir, or
2592 * b. be owner of victim, or
2593 * c. have CAP_FOWNER capability
2594 * 6. If the victim is append-only or immutable we can't do antyhing with
2595 * links pointing to it.
2596 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2597 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2598 * 9. We can't remove a root or mountpoint.
2599 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2600 * nfs_async_unlink().
2602 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2604 struct inode *inode = d_backing_inode(victim);
2607 if (d_is_negative(victim))
2611 BUG_ON(victim->d_parent->d_inode != dir);
2612 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2614 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2620 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2621 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2624 if (!d_is_dir(victim))
2626 if (IS_ROOT(victim))
2628 } else if (d_is_dir(victim))
2630 if (IS_DEADDIR(dir))
2632 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2637 /* Check whether we can create an object with dentry child in directory
2639 * 1. We can't do it if child already exists (open has special treatment for
2640 * this case, but since we are inlined it's OK)
2641 * 2. We can't do it if dir is read-only (done in permission())
2642 * 3. We should have write and exec permissions on dir
2643 * 4. We can't do it if dir is immutable (done in permission())
2645 static inline int may_create(struct inode *dir, struct dentry *child)
2647 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2650 if (IS_DEADDIR(dir))
2652 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2656 * p1 and p2 should be directories on the same fs.
2658 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2663 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2667 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2669 p = d_ancestor(p2, p1);
2671 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2672 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2676 p = d_ancestor(p1, p2);
2678 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2679 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2683 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2684 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2687 EXPORT_SYMBOL(lock_rename);
2689 void unlock_rename(struct dentry *p1, struct dentry *p2)
2691 mutex_unlock(&p1->d_inode->i_mutex);
2693 mutex_unlock(&p2->d_inode->i_mutex);
2694 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2697 EXPORT_SYMBOL(unlock_rename);
2699 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2702 int error = may_create(dir, dentry);
2706 if (!dir->i_op->create)
2707 return -EACCES; /* shouldn't it be ENOSYS? */
2710 error = security_inode_create(dir, dentry, mode);
2713 error = dir->i_op->create(dir, dentry, mode, want_excl);
2715 fsnotify_create(dir, dentry);
2718 EXPORT_SYMBOL(vfs_create);
2720 static int may_open(struct path *path, int acc_mode, int flag)
2722 struct dentry *dentry = path->dentry;
2723 struct inode *inode = dentry->d_inode;
2733 switch (inode->i_mode & S_IFMT) {
2737 if (acc_mode & MAY_WRITE)
2742 if (path->mnt->mnt_flags & MNT_NODEV)
2751 error = inode_permission(inode, acc_mode);
2756 * An append-only file must be opened in append mode for writing.
2758 if (IS_APPEND(inode)) {
2759 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2765 /* O_NOATIME can only be set by the owner or superuser */
2766 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2772 static int handle_truncate(struct file *filp)
2774 struct path *path = &filp->f_path;
2775 struct inode *inode = path->dentry->d_inode;
2776 int error = get_write_access(inode);
2780 * Refuse to truncate files with mandatory locks held on them.
2782 error = locks_verify_locked(filp);
2784 error = security_path_truncate(path);
2786 error = do_truncate(path->dentry, 0,
2787 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2790 put_write_access(inode);
2794 static inline int open_to_namei_flags(int flag)
2796 if ((flag & O_ACCMODE) == 3)
2801 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2803 int error = security_path_mknod(dir, dentry, mode, 0);
2807 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2811 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2815 * Attempt to atomically look up, create and open a file from a negative
2818 * Returns 0 if successful. The file will have been created and attached to
2819 * @file by the filesystem calling finish_open().
2821 * Returns 1 if the file was looked up only or didn't need creating. The
2822 * caller will need to perform the open themselves. @path will have been
2823 * updated to point to the new dentry. This may be negative.
2825 * Returns an error code otherwise.
2827 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2828 struct path *path, struct file *file,
2829 const struct open_flags *op,
2830 bool got_write, bool need_lookup,
2833 struct inode *dir = nd->path.dentry->d_inode;
2834 unsigned open_flag = open_to_namei_flags(op->open_flag);
2838 int create_error = 0;
2839 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2842 BUG_ON(dentry->d_inode);
2844 /* Don't create child dentry for a dead directory. */
2845 if (unlikely(IS_DEADDIR(dir))) {
2851 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2852 mode &= ~current_umask();
2854 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2856 open_flag &= ~O_TRUNC;
2859 * Checking write permission is tricky, bacuse we don't know if we are
2860 * going to actually need it: O_CREAT opens should work as long as the
2861 * file exists. But checking existence breaks atomicity. The trick is
2862 * to check access and if not granted clear O_CREAT from the flags.
2864 * Another problem is returing the "right" error value (e.g. for an
2865 * O_EXCL open we want to return EEXIST not EROFS).
2867 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2868 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2869 if (!(open_flag & O_CREAT)) {
2871 * No O_CREATE -> atomicity not a requirement -> fall
2872 * back to lookup + open
2875 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2876 /* Fall back and fail with the right error */
2877 create_error = -EROFS;
2880 /* No side effects, safe to clear O_CREAT */
2881 create_error = -EROFS;
2882 open_flag &= ~O_CREAT;
2886 if (open_flag & O_CREAT) {
2887 error = may_o_create(&nd->path, dentry, mode);
2889 create_error = error;
2890 if (open_flag & O_EXCL)
2892 open_flag &= ~O_CREAT;
2896 if (nd->flags & LOOKUP_DIRECTORY)
2897 open_flag |= O_DIRECTORY;
2899 file->f_path.dentry = DENTRY_NOT_SET;
2900 file->f_path.mnt = nd->path.mnt;
2901 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2904 if (create_error && error == -ENOENT)
2905 error = create_error;
2909 if (error) { /* returned 1, that is */
2910 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2914 if (file->f_path.dentry) {
2916 dentry = file->f_path.dentry;
2918 if (*opened & FILE_CREATED)
2919 fsnotify_create(dir, dentry);
2920 if (!dentry->d_inode) {
2921 WARN_ON(*opened & FILE_CREATED);
2923 error = create_error;
2927 if (excl && !(*opened & FILE_CREATED)) {
2936 * We didn't have the inode before the open, so check open permission
2939 acc_mode = op->acc_mode;
2940 if (*opened & FILE_CREATED) {
2941 WARN_ON(!(open_flag & O_CREAT));
2942 fsnotify_create(dir, dentry);
2943 acc_mode = MAY_OPEN;
2945 error = may_open(&file->f_path, acc_mode, open_flag);
2955 dentry = lookup_real(dir, dentry, nd->flags);
2957 return PTR_ERR(dentry);
2959 if (create_error && !dentry->d_inode) {
2960 error = create_error;
2964 path->dentry = dentry;
2965 path->mnt = nd->path.mnt;
2970 * Look up and maybe create and open the last component.
2972 * Must be called with i_mutex held on parent.
2974 * Returns 0 if the file was successfully atomically created (if necessary) and
2975 * opened. In this case the file will be returned attached to @file.
2977 * Returns 1 if the file was not completely opened at this time, though lookups
2978 * and creations will have been performed and the dentry returned in @path will
2979 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2980 * specified then a negative dentry may be returned.
2982 * An error code is returned otherwise.
2984 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2985 * cleared otherwise prior to returning.
2987 static int lookup_open(struct nameidata *nd, struct path *path,
2989 const struct open_flags *op,
2990 bool got_write, int *opened)
2992 struct dentry *dir = nd->path.dentry;
2993 struct inode *dir_inode = dir->d_inode;
2994 struct dentry *dentry;
2998 *opened &= ~FILE_CREATED;
2999 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
3001 return PTR_ERR(dentry);
3003 /* Cached positive dentry: will open in f_op->open */
3004 if (!need_lookup && dentry->d_inode)
3007 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
3008 return atomic_open(nd, dentry, path, file, op, got_write,
3009 need_lookup, opened);
3013 BUG_ON(dentry->d_inode);
3015 dentry = lookup_real(dir_inode, dentry, nd->flags);
3017 return PTR_ERR(dentry);
3020 /* Negative dentry, just create the file */
3021 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
3022 umode_t mode = op->mode;
3023 if (!IS_POSIXACL(dir->d_inode))
3024 mode &= ~current_umask();
3026 * This write is needed to ensure that a
3027 * rw->ro transition does not occur between
3028 * the time when the file is created and when
3029 * a permanent write count is taken through
3030 * the 'struct file' in finish_open().
3036 *opened |= FILE_CREATED;
3037 error = security_path_mknod(&nd->path, dentry, mode, 0);
3040 error = vfs_create(dir->d_inode, dentry, mode,
3041 nd->flags & LOOKUP_EXCL);
3046 path->dentry = dentry;
3047 path->mnt = nd->path.mnt;
3056 * Handle the last step of open()
3058 static int do_last(struct nameidata *nd,
3059 struct file *file, const struct open_flags *op,
3062 struct dentry *dir = nd->path.dentry;
3063 kuid_t dir_uid = nd->inode->i_uid;
3064 umode_t dir_mode = nd->inode->i_mode;
3065 int open_flag = op->open_flag;
3066 bool will_truncate = (open_flag & O_TRUNC) != 0;
3067 bool got_write = false;
3068 int acc_mode = op->acc_mode;
3070 struct inode *inode;
3071 struct path save_parent = { .dentry = NULL, .mnt = NULL };
3073 bool retried = false;
3076 nd->flags &= ~LOOKUP_PARENT;
3077 nd->flags |= op->intent;
3079 if (nd->last_type != LAST_NORM) {
3080 error = handle_dots(nd, nd->last_type);
3081 if (unlikely(error))
3086 if (!(open_flag & O_CREAT)) {
3087 if (nd->last.name[nd->last.len])
3088 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3089 /* we _can_ be in RCU mode here */
3090 error = lookup_fast(nd, &path, &inode, &seq);
3097 BUG_ON(nd->inode != dir->d_inode);
3099 /* create side of things */
3101 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3102 * has been cleared when we got to the last component we are
3105 error = complete_walk(nd);
3109 audit_inode(nd->name, dir, LOOKUP_PARENT);
3110 /* trailing slashes? */
3111 if (unlikely(nd->last.name[nd->last.len]))
3116 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3117 error = mnt_want_write(nd->path.mnt);
3121 * do _not_ fail yet - we might not need that or fail with
3122 * a different error; let lookup_open() decide; we'll be
3123 * dropping this one anyway.
3126 mutex_lock(&dir->d_inode->i_mutex);
3127 error = lookup_open(nd, &path, file, op, got_write, opened);
3128 mutex_unlock(&dir->d_inode->i_mutex);
3134 if ((*opened & FILE_CREATED) ||
3135 !S_ISREG(file_inode(file)->i_mode))
3136 will_truncate = false;
3138 audit_inode(nd->name, file->f_path.dentry, 0);
3142 if (*opened & FILE_CREATED) {
3143 /* Don't check for write permission, don't truncate */
3144 open_flag &= ~O_TRUNC;
3145 will_truncate = false;
3146 acc_mode = MAY_OPEN;
3147 path_to_nameidata(&path, nd);
3148 goto finish_open_created;
3152 * create/update audit record if it already exists.
3154 if (d_is_positive(path.dentry))
3155 audit_inode(nd->name, path.dentry, 0);
3158 * If atomic_open() acquired write access it is dropped now due to
3159 * possible mount and symlink following (this might be optimized away if
3163 mnt_drop_write(nd->path.mnt);
3167 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3168 path_to_nameidata(&path, nd);
3172 error = follow_managed(&path, nd);
3173 if (unlikely(error < 0))
3176 BUG_ON(nd->flags & LOOKUP_RCU);
3177 seq = 0; /* out of RCU mode, so the value doesn't matter */
3178 if (unlikely(d_is_negative(path.dentry))) {
3179 path_to_nameidata(&path, nd);
3182 inode = d_backing_inode(path.dentry);
3186 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3188 if (unlikely(error))
3191 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3192 path_to_nameidata(&path, nd);
3194 save_parent.dentry = nd->path.dentry;
3195 save_parent.mnt = mntget(path.mnt);
3196 nd->path.dentry = path.dentry;
3201 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3203 error = complete_walk(nd);
3205 path_put(&save_parent);
3208 audit_inode(nd->name, nd->path.dentry, 0);
3209 if (unlikely(d_is_symlink(nd->path.dentry)) && !(open_flag & O_PATH)) {
3213 if (open_flag & O_CREAT) {
3215 if (d_is_dir(nd->path.dentry))
3217 error = may_create_in_sticky(dir_mode, dir_uid,
3218 d_backing_inode(nd->path.dentry));
3219 if (unlikely(error))
3223 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3225 if (!d_is_reg(nd->path.dentry))
3226 will_truncate = false;
3228 if (will_truncate) {
3229 error = mnt_want_write(nd->path.mnt);
3234 finish_open_created:
3235 error = may_open(&nd->path, acc_mode, open_flag);
3239 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3240 error = vfs_open(&nd->path, file, current_cred());
3242 *opened |= FILE_OPENED;
3244 if (error == -EOPENSTALE)
3249 error = open_check_o_direct(file);
3252 error = ima_file_check(file, op->acc_mode, *opened);
3256 if (will_truncate) {
3257 error = handle_truncate(file);
3262 if (unlikely(error > 0)) {
3267 mnt_drop_write(nd->path.mnt);
3268 path_put(&save_parent);
3276 /* If no saved parent or already retried then can't retry */
3277 if (!save_parent.dentry || retried)
3280 BUG_ON(save_parent.dentry != dir);
3281 path_put(&nd->path);
3282 nd->path = save_parent;
3283 nd->inode = dir->d_inode;
3284 save_parent.mnt = NULL;
3285 save_parent.dentry = NULL;
3287 mnt_drop_write(nd->path.mnt);
3294 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3295 const struct open_flags *op,
3296 struct file *file, int *opened)
3298 static const struct qstr name = QSTR_INIT("/", 1);
3299 struct dentry *child;
3302 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3303 if (unlikely(error))
3305 error = mnt_want_write(path.mnt);
3306 if (unlikely(error))
3308 dir = path.dentry->d_inode;
3309 /* we want directory to be writable */
3310 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3313 if (!dir->i_op->tmpfile) {
3314 error = -EOPNOTSUPP;
3317 child = d_alloc(path.dentry, &name);
3318 if (unlikely(!child)) {
3323 path.dentry = child;
3324 error = dir->i_op->tmpfile(dir, child, op->mode);
3327 audit_inode(nd->name, child, 0);
3328 /* Don't check for other permissions, the inode was just created */
3329 error = may_open(&path, MAY_OPEN, op->open_flag);
3332 file->f_path.mnt = path.mnt;
3333 error = finish_open(file, child, NULL, opened);
3336 error = open_check_o_direct(file);
3339 } else if (!(op->open_flag & O_EXCL)) {
3340 struct inode *inode = file_inode(file);
3341 spin_lock(&inode->i_lock);
3342 inode->i_state |= I_LINKABLE;
3343 spin_unlock(&inode->i_lock);
3346 mnt_drop_write(path.mnt);
3352 static struct file *path_openat(struct nameidata *nd,
3353 const struct open_flags *op, unsigned flags)
3360 file = get_empty_filp();
3364 file->f_flags = op->open_flag;
3366 if (unlikely(file->f_flags & __O_TMPFILE)) {
3367 error = do_tmpfile(nd, flags, op, file, &opened);
3371 s = path_init(nd, flags);
3376 while (!(error = link_path_walk(s, nd)) &&
3377 (error = do_last(nd, file, op, &opened)) > 0) {
3378 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3379 s = trailing_symlink(nd);
3387 if (!(opened & FILE_OPENED)) {
3391 if (unlikely(error)) {
3392 if (error == -EOPENSTALE) {
3393 if (flags & LOOKUP_RCU)
3398 file = ERR_PTR(error);
3403 struct file *do_filp_open(int dfd, struct filename *pathname,
3404 const struct open_flags *op)
3406 struct nameidata nd;
3407 int flags = op->lookup_flags;
3410 set_nameidata(&nd, dfd, pathname);
3411 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3412 if (unlikely(filp == ERR_PTR(-ECHILD)))
3413 filp = path_openat(&nd, op, flags);
3414 if (unlikely(filp == ERR_PTR(-ESTALE)))
3415 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3416 restore_nameidata();
3420 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3421 const char *name, const struct open_flags *op)
3423 struct nameidata nd;
3425 struct filename *filename;
3426 int flags = op->lookup_flags | LOOKUP_ROOT;
3429 nd.root.dentry = dentry;
3431 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3432 return ERR_PTR(-ELOOP);
3434 filename = getname_kernel(name);
3435 if (IS_ERR(filename))
3436 return ERR_CAST(filename);
3438 set_nameidata(&nd, -1, filename);
3439 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3440 if (unlikely(file == ERR_PTR(-ECHILD)))
3441 file = path_openat(&nd, op, flags);
3442 if (unlikely(file == ERR_PTR(-ESTALE)))
3443 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3444 restore_nameidata();
3449 static struct dentry *filename_create(int dfd, struct filename *name,
3450 struct path *path, unsigned int lookup_flags)
3452 struct dentry *dentry = ERR_PTR(-EEXIST);
3457 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3460 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3461 * other flags passed in are ignored!
3463 lookup_flags &= LOOKUP_REVAL;
3465 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3467 return ERR_CAST(name);
3470 * Yucky last component or no last component at all?
3471 * (foo/., foo/.., /////)
3473 if (unlikely(type != LAST_NORM))
3476 /* don't fail immediately if it's r/o, at least try to report other errors */
3477 err2 = mnt_want_write(path->mnt);
3479 * Do the final lookup.
3481 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3482 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3483 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3488 if (d_is_positive(dentry))
3492 * Special case - lookup gave negative, but... we had foo/bar/
3493 * From the vfs_mknod() POV we just have a negative dentry -
3494 * all is fine. Let's be bastards - you had / on the end, you've
3495 * been asking for (non-existent) directory. -ENOENT for you.
3497 if (unlikely(!is_dir && last.name[last.len])) {
3501 if (unlikely(err2)) {
3509 dentry = ERR_PTR(error);
3511 mutex_unlock(&path->dentry->d_inode->i_mutex);
3513 mnt_drop_write(path->mnt);
3520 struct dentry *kern_path_create(int dfd, const char *pathname,
3521 struct path *path, unsigned int lookup_flags)
3523 return filename_create(dfd, getname_kernel(pathname),
3524 path, lookup_flags);
3526 EXPORT_SYMBOL(kern_path_create);
3528 void done_path_create(struct path *path, struct dentry *dentry)
3531 mutex_unlock(&path->dentry->d_inode->i_mutex);
3532 mnt_drop_write(path->mnt);
3535 EXPORT_SYMBOL(done_path_create);
3537 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3538 struct path *path, unsigned int lookup_flags)
3540 return filename_create(dfd, getname(pathname), path, lookup_flags);
3542 EXPORT_SYMBOL(user_path_create);
3544 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3546 int error = may_create(dir, dentry);
3551 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3554 if (!dir->i_op->mknod)
3557 error = devcgroup_inode_mknod(mode, dev);
3561 error = security_inode_mknod(dir, dentry, mode, dev);
3565 error = dir->i_op->mknod(dir, dentry, mode, dev);
3567 fsnotify_create(dir, dentry);
3570 EXPORT_SYMBOL(vfs_mknod);
3572 static int may_mknod(umode_t mode)
3574 switch (mode & S_IFMT) {
3580 case 0: /* zero mode translates to S_IFREG */
3589 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3592 struct dentry *dentry;
3595 unsigned int lookup_flags = 0;
3597 error = may_mknod(mode);
3601 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3603 return PTR_ERR(dentry);
3605 if (!IS_POSIXACL(path.dentry->d_inode))
3606 mode &= ~current_umask();
3607 error = security_path_mknod(&path, dentry, mode, dev);
3610 switch (mode & S_IFMT) {
3611 case 0: case S_IFREG:
3612 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3614 case S_IFCHR: case S_IFBLK:
3615 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3616 new_decode_dev(dev));
3618 case S_IFIFO: case S_IFSOCK:
3619 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3623 done_path_create(&path, dentry);
3624 if (retry_estale(error, lookup_flags)) {
3625 lookup_flags |= LOOKUP_REVAL;
3631 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3633 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3636 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3638 int error = may_create(dir, dentry);
3639 unsigned max_links = dir->i_sb->s_max_links;
3644 if (!dir->i_op->mkdir)
3647 mode &= (S_IRWXUGO|S_ISVTX);
3648 error = security_inode_mkdir(dir, dentry, mode);
3652 if (max_links && dir->i_nlink >= max_links)
3655 error = dir->i_op->mkdir(dir, dentry, mode);
3657 fsnotify_mkdir(dir, dentry);
3660 EXPORT_SYMBOL(vfs_mkdir);
3662 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3664 struct dentry *dentry;
3667 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3670 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3672 return PTR_ERR(dentry);
3674 if (!IS_POSIXACL(path.dentry->d_inode))
3675 mode &= ~current_umask();
3676 error = security_path_mkdir(&path, dentry, mode);
3678 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3679 done_path_create(&path, dentry);
3680 if (retry_estale(error, lookup_flags)) {
3681 lookup_flags |= LOOKUP_REVAL;
3687 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3689 return sys_mkdirat(AT_FDCWD, pathname, mode);
3693 * The dentry_unhash() helper will try to drop the dentry early: we
3694 * should have a usage count of 1 if we're the only user of this
3695 * dentry, and if that is true (possibly after pruning the dcache),
3696 * then we drop the dentry now.
3698 * A low-level filesystem can, if it choses, legally
3701 * if (!d_unhashed(dentry))
3704 * if it cannot handle the case of removing a directory
3705 * that is still in use by something else..
3707 void dentry_unhash(struct dentry *dentry)
3709 shrink_dcache_parent(dentry);
3710 spin_lock(&dentry->d_lock);
3711 if (dentry->d_lockref.count == 1)
3713 spin_unlock(&dentry->d_lock);
3715 EXPORT_SYMBOL(dentry_unhash);
3717 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3719 int error = may_delete(dir, dentry, 1);
3724 if (!dir->i_op->rmdir)
3728 mutex_lock(&dentry->d_inode->i_mutex);
3731 if (is_local_mountpoint(dentry))
3734 error = security_inode_rmdir(dir, dentry);
3738 shrink_dcache_parent(dentry);
3739 error = dir->i_op->rmdir(dir, dentry);
3743 dentry->d_inode->i_flags |= S_DEAD;
3745 detach_mounts(dentry);
3748 mutex_unlock(&dentry->d_inode->i_mutex);
3754 EXPORT_SYMBOL(vfs_rmdir);
3756 static long do_rmdir(int dfd, const char __user *pathname)
3759 struct filename *name;
3760 struct dentry *dentry;
3764 unsigned int lookup_flags = 0;
3766 name = user_path_parent(dfd, pathname,
3767 &path, &last, &type, lookup_flags);
3769 return PTR_ERR(name);
3783 error = mnt_want_write(path.mnt);
3787 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3788 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3789 error = PTR_ERR(dentry);
3792 if (!dentry->d_inode) {
3796 error = security_path_rmdir(&path, dentry);
3799 error = vfs_rmdir(path.dentry->d_inode, dentry);
3803 mutex_unlock(&path.dentry->d_inode->i_mutex);
3804 mnt_drop_write(path.mnt);
3808 if (retry_estale(error, lookup_flags)) {
3809 lookup_flags |= LOOKUP_REVAL;
3815 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3817 return do_rmdir(AT_FDCWD, pathname);
3821 * vfs_unlink - unlink a filesystem object
3822 * @dir: parent directory
3824 * @delegated_inode: returns victim inode, if the inode is delegated.
3826 * The caller must hold dir->i_mutex.
3828 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3829 * return a reference to the inode in delegated_inode. The caller
3830 * should then break the delegation on that inode and retry. Because
3831 * breaking a delegation may take a long time, the caller should drop
3832 * dir->i_mutex before doing so.
3834 * Alternatively, a caller may pass NULL for delegated_inode. This may
3835 * be appropriate for callers that expect the underlying filesystem not
3836 * to be NFS exported.
3838 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3840 struct inode *target = dentry->d_inode;
3841 int error = may_delete(dir, dentry, 0);
3846 if (!dir->i_op->unlink)
3849 mutex_lock(&target->i_mutex);
3850 if (is_local_mountpoint(dentry))
3853 error = security_inode_unlink(dir, dentry);
3855 error = try_break_deleg(target, delegated_inode);
3858 error = dir->i_op->unlink(dir, dentry);
3861 detach_mounts(dentry);
3866 mutex_unlock(&target->i_mutex);
3868 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3869 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3870 fsnotify_link_count(target);
3876 EXPORT_SYMBOL(vfs_unlink);
3879 * Make sure that the actual truncation of the file will occur outside its
3880 * directory's i_mutex. Truncate can take a long time if there is a lot of
3881 * writeout happening, and we don't want to prevent access to the directory
3882 * while waiting on the I/O.
3884 static long do_unlinkat(int dfd, const char __user *pathname)
3887 struct filename *name;
3888 struct dentry *dentry;
3892 struct inode *inode = NULL;
3893 struct inode *delegated_inode = NULL;
3894 unsigned int lookup_flags = 0;
3896 name = user_path_parent(dfd, pathname,
3897 &path, &last, &type, lookup_flags);
3899 return PTR_ERR(name);
3902 if (type != LAST_NORM)
3905 error = mnt_want_write(path.mnt);
3909 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3910 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3911 error = PTR_ERR(dentry);
3912 if (!IS_ERR(dentry)) {
3913 /* Why not before? Because we want correct error value */
3914 if (last.name[last.len])
3916 inode = dentry->d_inode;
3917 if (d_is_negative(dentry))
3920 error = security_path_unlink(&path, dentry);
3923 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3927 mutex_unlock(&path.dentry->d_inode->i_mutex);
3929 iput(inode); /* truncate the inode here */
3931 if (delegated_inode) {
3932 error = break_deleg_wait(&delegated_inode);
3936 mnt_drop_write(path.mnt);
3940 if (retry_estale(error, lookup_flags)) {
3941 lookup_flags |= LOOKUP_REVAL;
3948 if (d_is_negative(dentry))
3950 else if (d_is_dir(dentry))
3957 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3959 if ((flag & ~AT_REMOVEDIR) != 0)
3962 if (flag & AT_REMOVEDIR)
3963 return do_rmdir(dfd, pathname);
3965 return do_unlinkat(dfd, pathname);
3968 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3970 return do_unlinkat(AT_FDCWD, pathname);
3973 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3975 int error = may_create(dir, dentry);
3980 if (!dir->i_op->symlink)
3983 error = security_inode_symlink(dir, dentry, oldname);
3987 error = dir->i_op->symlink(dir, dentry, oldname);
3989 fsnotify_create(dir, dentry);
3992 EXPORT_SYMBOL(vfs_symlink);
3994 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3995 int, newdfd, const char __user *, newname)
3998 struct filename *from;
3999 struct dentry *dentry;
4001 unsigned int lookup_flags = 0;
4003 from = getname(oldname);
4005 return PTR_ERR(from);
4007 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4008 error = PTR_ERR(dentry);
4012 error = security_path_symlink(&path, dentry, from->name);
4014 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4015 done_path_create(&path, dentry);
4016 if (retry_estale(error, lookup_flags)) {
4017 lookup_flags |= LOOKUP_REVAL;
4025 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4027 return sys_symlinkat(oldname, AT_FDCWD, newname);
4031 * vfs_link - create a new link
4032 * @old_dentry: object to be linked
4034 * @new_dentry: where to create the new link
4035 * @delegated_inode: returns inode needing a delegation break
4037 * The caller must hold dir->i_mutex
4039 * If vfs_link discovers a delegation on the to-be-linked file in need
4040 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4041 * inode in delegated_inode. The caller should then break the delegation
4042 * and retry. Because breaking a delegation may take a long time, the
4043 * caller should drop the i_mutex before doing so.
4045 * Alternatively, a caller may pass NULL for delegated_inode. This may
4046 * be appropriate for callers that expect the underlying filesystem not
4047 * to be NFS exported.
4049 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4051 struct inode *inode = old_dentry->d_inode;
4052 unsigned max_links = dir->i_sb->s_max_links;
4058 error = may_create(dir, new_dentry);
4062 if (dir->i_sb != inode->i_sb)
4066 * A link to an append-only or immutable file cannot be created.
4068 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4070 if (!dir->i_op->link)
4072 if (S_ISDIR(inode->i_mode))
4075 error = security_inode_link(old_dentry, dir, new_dentry);
4079 mutex_lock(&inode->i_mutex);
4080 /* Make sure we don't allow creating hardlink to an unlinked file */
4081 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4083 else if (max_links && inode->i_nlink >= max_links)
4086 error = try_break_deleg(inode, delegated_inode);
4088 error = dir->i_op->link(old_dentry, dir, new_dentry);
4091 if (!error && (inode->i_state & I_LINKABLE)) {
4092 spin_lock(&inode->i_lock);
4093 inode->i_state &= ~I_LINKABLE;
4094 spin_unlock(&inode->i_lock);
4096 mutex_unlock(&inode->i_mutex);
4098 fsnotify_link(dir, inode, new_dentry);
4101 EXPORT_SYMBOL(vfs_link);
4104 * Hardlinks are often used in delicate situations. We avoid
4105 * security-related surprises by not following symlinks on the
4108 * We don't follow them on the oldname either to be compatible
4109 * with linux 2.0, and to avoid hard-linking to directories
4110 * and other special files. --ADM
4112 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4113 int, newdfd, const char __user *, newname, int, flags)
4115 struct dentry *new_dentry;
4116 struct path old_path, new_path;
4117 struct inode *delegated_inode = NULL;
4121 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4124 * To use null names we require CAP_DAC_READ_SEARCH
4125 * This ensures that not everyone will be able to create
4126 * handlink using the passed filedescriptor.
4128 if (flags & AT_EMPTY_PATH) {
4129 if (!capable(CAP_DAC_READ_SEARCH))
4134 if (flags & AT_SYMLINK_FOLLOW)
4135 how |= LOOKUP_FOLLOW;
4137 error = user_path_at(olddfd, oldname, how, &old_path);
4141 new_dentry = user_path_create(newdfd, newname, &new_path,
4142 (how & LOOKUP_REVAL));
4143 error = PTR_ERR(new_dentry);
4144 if (IS_ERR(new_dentry))
4148 if (old_path.mnt != new_path.mnt)
4150 error = may_linkat(&old_path);
4151 if (unlikely(error))
4153 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4156 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4158 done_path_create(&new_path, new_dentry);
4159 if (delegated_inode) {
4160 error = break_deleg_wait(&delegated_inode);
4162 path_put(&old_path);
4166 if (retry_estale(error, how)) {
4167 path_put(&old_path);
4168 how |= LOOKUP_REVAL;
4172 path_put(&old_path);
4177 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4179 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4183 * vfs_rename - rename a filesystem object
4184 * @old_dir: parent of source
4185 * @old_dentry: source
4186 * @new_dir: parent of destination
4187 * @new_dentry: destination
4188 * @delegated_inode: returns an inode needing a delegation break
4189 * @flags: rename flags
4191 * The caller must hold multiple mutexes--see lock_rename()).
4193 * If vfs_rename discovers a delegation in need of breaking at either
4194 * the source or destination, it will return -EWOULDBLOCK and return a
4195 * reference to the inode in delegated_inode. The caller should then
4196 * break the delegation and retry. Because breaking a delegation may
4197 * take a long time, the caller should drop all locks before doing
4200 * Alternatively, a caller may pass NULL for delegated_inode. This may
4201 * be appropriate for callers that expect the underlying filesystem not
4202 * to be NFS exported.
4204 * The worst of all namespace operations - renaming directory. "Perverted"
4205 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4207 * a) we can get into loop creation.
4208 * b) race potential - two innocent renames can create a loop together.
4209 * That's where 4.4 screws up. Current fix: serialization on
4210 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4212 * c) we have to lock _four_ objects - parents and victim (if it exists),
4213 * and source (if it is not a directory).
4214 * And that - after we got ->i_mutex on parents (until then we don't know
4215 * whether the target exists). Solution: try to be smart with locking
4216 * order for inodes. We rely on the fact that tree topology may change
4217 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4218 * move will be locked. Thus we can rank directories by the tree
4219 * (ancestors first) and rank all non-directories after them.
4220 * That works since everybody except rename does "lock parent, lookup,
4221 * lock child" and rename is under ->s_vfs_rename_mutex.
4222 * HOWEVER, it relies on the assumption that any object with ->lookup()
4223 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4224 * we'd better make sure that there's no link(2) for them.
4225 * d) conversion from fhandle to dentry may come in the wrong moment - when
4226 * we are removing the target. Solution: we will have to grab ->i_mutex
4227 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4228 * ->i_mutex on parents, which works but leads to some truly excessive
4231 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4232 struct inode *new_dir, struct dentry *new_dentry,
4233 struct inode **delegated_inode, unsigned int flags)
4236 bool is_dir = d_is_dir(old_dentry);
4237 struct inode *source = old_dentry->d_inode;
4238 struct inode *target = new_dentry->d_inode;
4239 bool new_is_dir = false;
4240 unsigned max_links = new_dir->i_sb->s_max_links;
4241 struct name_snapshot old_name;
4244 * Check source == target.
4245 * On overlayfs need to look at underlying inodes.
4247 if (vfs_select_inode(old_dentry, 0) == vfs_select_inode(new_dentry, 0))
4250 error = may_delete(old_dir, old_dentry, is_dir);
4255 error = may_create(new_dir, new_dentry);
4257 new_is_dir = d_is_dir(new_dentry);
4259 if (!(flags & RENAME_EXCHANGE))
4260 error = may_delete(new_dir, new_dentry, is_dir);
4262 error = may_delete(new_dir, new_dentry, new_is_dir);
4267 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4270 if (flags && !old_dir->i_op->rename2)
4274 * If we are going to change the parent - check write permissions,
4275 * we'll need to flip '..'.
4277 if (new_dir != old_dir) {
4279 error = inode_permission(source, MAY_WRITE);
4283 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4284 error = inode_permission(target, MAY_WRITE);
4290 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4295 take_dentry_name_snapshot(&old_name, old_dentry);
4297 if (!is_dir || (flags & RENAME_EXCHANGE))
4298 lock_two_nondirectories(source, target);
4300 mutex_lock(&target->i_mutex);
4303 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4306 if (max_links && new_dir != old_dir) {
4308 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4310 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4311 old_dir->i_nlink >= max_links)
4314 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4315 shrink_dcache_parent(new_dentry);
4317 error = try_break_deleg(source, delegated_inode);
4321 if (target && !new_is_dir) {
4322 error = try_break_deleg(target, delegated_inode);
4326 if (!old_dir->i_op->rename2) {
4327 error = old_dir->i_op->rename(old_dir, old_dentry,
4328 new_dir, new_dentry);
4330 WARN_ON(old_dir->i_op->rename != NULL);
4331 error = old_dir->i_op->rename2(old_dir, old_dentry,
4332 new_dir, new_dentry, flags);
4337 if (!(flags & RENAME_EXCHANGE) && target) {
4339 target->i_flags |= S_DEAD;
4340 dont_mount(new_dentry);
4341 detach_mounts(new_dentry);
4343 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4344 if (!(flags & RENAME_EXCHANGE))
4345 d_move(old_dentry, new_dentry);
4347 d_exchange(old_dentry, new_dentry);
4350 if (!is_dir || (flags & RENAME_EXCHANGE))
4351 unlock_two_nondirectories(source, target);
4353 mutex_unlock(&target->i_mutex);
4356 fsnotify_move(old_dir, new_dir, old_name.name, is_dir,
4357 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4358 if (flags & RENAME_EXCHANGE) {
4359 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4360 new_is_dir, NULL, new_dentry);
4363 release_dentry_name_snapshot(&old_name);
4367 EXPORT_SYMBOL(vfs_rename);
4369 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4370 int, newdfd, const char __user *, newname, unsigned int, flags)
4372 struct dentry *old_dentry, *new_dentry;
4373 struct dentry *trap;
4374 struct path old_path, new_path;
4375 struct qstr old_last, new_last;
4376 int old_type, new_type;
4377 struct inode *delegated_inode = NULL;
4378 struct filename *from;
4379 struct filename *to;
4380 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4381 bool should_retry = false;
4384 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4387 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4388 (flags & RENAME_EXCHANGE))
4391 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4394 if (flags & RENAME_EXCHANGE)
4398 from = user_path_parent(olddfd, oldname,
4399 &old_path, &old_last, &old_type, lookup_flags);
4401 error = PTR_ERR(from);
4405 to = user_path_parent(newdfd, newname,
4406 &new_path, &new_last, &new_type, lookup_flags);
4408 error = PTR_ERR(to);
4413 if (old_path.mnt != new_path.mnt)
4417 if (old_type != LAST_NORM)
4420 if (flags & RENAME_NOREPLACE)
4422 if (new_type != LAST_NORM)
4425 error = mnt_want_write(old_path.mnt);
4430 trap = lock_rename(new_path.dentry, old_path.dentry);
4432 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4433 error = PTR_ERR(old_dentry);
4434 if (IS_ERR(old_dentry))
4436 /* source must exist */
4438 if (d_is_negative(old_dentry))
4440 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4441 error = PTR_ERR(new_dentry);
4442 if (IS_ERR(new_dentry))
4445 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4447 if (flags & RENAME_EXCHANGE) {
4449 if (d_is_negative(new_dentry))
4452 if (!d_is_dir(new_dentry)) {
4454 if (new_last.name[new_last.len])
4458 /* unless the source is a directory trailing slashes give -ENOTDIR */
4459 if (!d_is_dir(old_dentry)) {
4461 if (old_last.name[old_last.len])
4463 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4466 /* source should not be ancestor of target */
4468 if (old_dentry == trap)
4470 /* target should not be an ancestor of source */
4471 if (!(flags & RENAME_EXCHANGE))
4473 if (new_dentry == trap)
4476 error = security_path_rename(&old_path, old_dentry,
4477 &new_path, new_dentry, flags);
4480 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4481 new_path.dentry->d_inode, new_dentry,
4482 &delegated_inode, flags);
4488 unlock_rename(new_path.dentry, old_path.dentry);
4489 if (delegated_inode) {
4490 error = break_deleg_wait(&delegated_inode);
4494 mnt_drop_write(old_path.mnt);
4496 if (retry_estale(error, lookup_flags))
4497 should_retry = true;
4498 path_put(&new_path);
4501 path_put(&old_path);
4504 should_retry = false;
4505 lookup_flags |= LOOKUP_REVAL;
4512 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4513 int, newdfd, const char __user *, newname)
4515 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4518 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4520 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4523 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4525 int error = may_create(dir, dentry);
4529 if (!dir->i_op->mknod)
4532 return dir->i_op->mknod(dir, dentry,
4533 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4535 EXPORT_SYMBOL(vfs_whiteout);
4537 int readlink_copy(char __user *buffer, int buflen, const char *link)
4539 int len = PTR_ERR(link);
4544 if (len > (unsigned) buflen)
4546 if (copy_to_user(buffer, link, len))
4551 EXPORT_SYMBOL(readlink_copy);
4554 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4555 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4556 * using) it for any given inode is up to filesystem.
4558 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4561 struct inode *inode = d_inode(dentry);
4562 const char *link = inode->i_link;
4566 link = inode->i_op->follow_link(dentry, &cookie);
4568 return PTR_ERR(link);
4570 res = readlink_copy(buffer, buflen, link);
4571 if (inode->i_op->put_link)
4572 inode->i_op->put_link(inode, cookie);
4575 EXPORT_SYMBOL(generic_readlink);
4577 /* get the link contents into pagecache */
4578 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4582 struct address_space *mapping = dentry->d_inode->i_mapping;
4583 page = read_mapping_page(mapping, 0, NULL);
4588 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4592 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4594 struct page *page = NULL;
4595 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4598 page_cache_release(page);
4602 EXPORT_SYMBOL(page_readlink);
4604 const char *page_follow_link_light(struct dentry *dentry, void **cookie)
4606 struct page *page = NULL;
4607 char *res = page_getlink(dentry, &page);
4612 EXPORT_SYMBOL(page_follow_link_light);
4614 void page_put_link(struct inode *unused, void *cookie)
4616 struct page *page = cookie;
4618 page_cache_release(page);
4620 EXPORT_SYMBOL(page_put_link);
4623 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4625 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4627 struct address_space *mapping = inode->i_mapping;
4632 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4634 flags |= AOP_FLAG_NOFS;
4637 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4638 flags, &page, &fsdata);
4642 kaddr = kmap_atomic(page);
4643 memcpy(kaddr, symname, len-1);
4644 kunmap_atomic(kaddr);
4646 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4653 mark_inode_dirty(inode);
4658 EXPORT_SYMBOL(__page_symlink);
4660 int page_symlink(struct inode *inode, const char *symname, int len)
4662 return __page_symlink(inode, symname, len,
4663 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4665 EXPORT_SYMBOL(page_symlink);
4667 const struct inode_operations page_symlink_inode_operations = {
4668 .readlink = generic_readlink,
4669 .follow_link = page_follow_link_light,
4670 .put_link = page_put_link,
4672 EXPORT_SYMBOL(page_symlink_inode_operations);