1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * Some corrections by tytso.
12 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
15 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
18 #include <linux/init.h>
19 #include <linux/export.h>
20 #include <linux/kernel.h>
21 #include <linux/slab.h>
23 #include <linux/namei.h>
24 #include <linux/pagemap.h>
25 #include <linux/fsnotify.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/ima.h>
29 #include <linux/syscalls.h>
30 #include <linux/mount.h>
31 #include <linux/audit.h>
32 #include <linux/capability.h>
33 #include <linux/file.h>
34 #include <linux/fcntl.h>
35 #include <linux/device_cgroup.h>
36 #include <linux/fs_struct.h>
37 #include <linux/posix_acl.h>
38 #include <linux/hash.h>
39 #include <linux/bitops.h>
40 #include <linux/init_task.h>
41 #include <linux/uaccess.h>
42 #include <linux/build_bug.h>
47 /* [Feb-1997 T. Schoebel-Theuer]
48 * Fundamental changes in the pathname lookup mechanisms (namei)
49 * were necessary because of omirr. The reason is that omirr needs
50 * to know the _real_ pathname, not the user-supplied one, in case
51 * of symlinks (and also when transname replacements occur).
53 * The new code replaces the old recursive symlink resolution with
54 * an iterative one (in case of non-nested symlink chains). It does
55 * this with calls to <fs>_follow_link().
56 * As a side effect, dir_namei(), _namei() and follow_link() are now
57 * replaced with a single function lookup_dentry() that can handle all
58 * the special cases of the former code.
60 * With the new dcache, the pathname is stored at each inode, at least as
61 * long as the refcount of the inode is positive. As a side effect, the
62 * size of the dcache depends on the inode cache and thus is dynamic.
64 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
65 * resolution to correspond with current state of the code.
67 * Note that the symlink resolution is not *completely* iterative.
68 * There is still a significant amount of tail- and mid- recursion in
69 * the algorithm. Also, note that <fs>_readlink() is not used in
70 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
71 * may return different results than <fs>_follow_link(). Many virtual
72 * filesystems (including /proc) exhibit this behavior.
75 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
76 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
77 * and the name already exists in form of a symlink, try to create the new
78 * name indicated by the symlink. The old code always complained that the
79 * name already exists, due to not following the symlink even if its target
80 * is nonexistent. The new semantics affects also mknod() and link() when
81 * the name is a symlink pointing to a non-existent name.
83 * I don't know which semantics is the right one, since I have no access
84 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
85 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
86 * "old" one. Personally, I think the new semantics is much more logical.
87 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
88 * file does succeed in both HP-UX and SunOs, but not in Solaris
89 * and in the old Linux semantics.
92 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
93 * semantics. See the comments in "open_namei" and "do_link" below.
95 * [10-Sep-98 Alan Modra] Another symlink change.
98 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
99 * inside the path - always follow.
100 * in the last component in creation/removal/renaming - never follow.
101 * if LOOKUP_FOLLOW passed - follow.
102 * if the pathname has trailing slashes - follow.
103 * otherwise - don't follow.
104 * (applied in that order).
106 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
107 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
108 * During the 2.4 we need to fix the userland stuff depending on it -
109 * hopefully we will be able to get rid of that wart in 2.5. So far only
110 * XEmacs seems to be relying on it...
113 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
114 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
115 * any extra contention...
118 /* In order to reduce some races, while at the same time doing additional
119 * checking and hopefully speeding things up, we copy filenames to the
120 * kernel data space before using them..
122 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
123 * PATH_MAX includes the nul terminator --RR.
126 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
129 getname_flags(const char __user *filename, int flags, int *empty)
131 struct filename *result;
134 BUILD_BUG_ON(offsetof(struct filename, iname) % sizeof(long) != 0);
136 result = audit_reusename(filename);
140 result = __getname();
141 if (unlikely(!result))
142 return ERR_PTR(-ENOMEM);
145 * First, try to embed the struct filename inside the names_cache
148 kname = (char *)result->iname;
149 result->name = kname;
151 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
152 if (unlikely(len < 0)) {
158 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
159 * separate struct filename so we can dedicate the entire
160 * names_cache allocation for the pathname, and re-do the copy from
163 if (unlikely(len == EMBEDDED_NAME_MAX)) {
164 const size_t size = offsetof(struct filename, iname[1]);
165 kname = (char *)result;
168 * size is chosen that way we to guarantee that
169 * result->iname[0] is within the same object and that
170 * kname can't be equal to result->iname, no matter what.
172 result = kzalloc(size, GFP_KERNEL);
173 if (unlikely(!result)) {
175 return ERR_PTR(-ENOMEM);
177 result->name = kname;
178 len = strncpy_from_user(kname, filename, PATH_MAX);
179 if (unlikely(len < 0)) {
184 if (unlikely(len == PATH_MAX)) {
187 return ERR_PTR(-ENAMETOOLONG);
192 /* The empty path is special. */
193 if (unlikely(!len)) {
196 if (!(flags & LOOKUP_EMPTY)) {
198 return ERR_PTR(-ENOENT);
202 result->uptr = filename;
203 result->aname = NULL;
204 audit_getname(result);
209 getname(const char __user * filename)
211 return getname_flags(filename, 0, NULL);
215 getname_kernel(const char * filename)
217 struct filename *result;
218 int len = strlen(filename) + 1;
220 result = __getname();
221 if (unlikely(!result))
222 return ERR_PTR(-ENOMEM);
224 if (len <= EMBEDDED_NAME_MAX) {
225 result->name = (char *)result->iname;
226 } else if (len <= PATH_MAX) {
227 const size_t size = offsetof(struct filename, iname[1]);
228 struct filename *tmp;
230 tmp = kmalloc(size, GFP_KERNEL);
231 if (unlikely(!tmp)) {
233 return ERR_PTR(-ENOMEM);
235 tmp->name = (char *)result;
239 return ERR_PTR(-ENAMETOOLONG);
241 memcpy((char *)result->name, filename, len);
243 result->aname = NULL;
245 audit_getname(result);
250 void putname(struct filename *name)
252 BUG_ON(name->refcnt <= 0);
254 if (--name->refcnt > 0)
257 if (name->name != name->iname) {
258 __putname(name->name);
264 static int check_acl(struct inode *inode, int mask)
266 #ifdef CONFIG_FS_POSIX_ACL
267 struct posix_acl *acl;
269 if (mask & MAY_NOT_BLOCK) {
270 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
273 /* no ->get_acl() calls in RCU mode... */
274 if (is_uncached_acl(acl))
276 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
279 acl = get_acl(inode, ACL_TYPE_ACCESS);
283 int error = posix_acl_permission(inode, acl, mask);
284 posix_acl_release(acl);
293 * This does the basic permission checking
295 static int acl_permission_check(struct inode *inode, int mask)
297 unsigned int mode = inode->i_mode;
299 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
302 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
303 int error = check_acl(inode, mask);
304 if (error != -EAGAIN)
308 if (in_group_p(inode->i_gid))
313 * If the DACs are ok we don't need any capability check.
315 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
321 * generic_permission - check for access rights on a Posix-like filesystem
322 * @inode: inode to check access rights for
323 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
325 * Used to check for read/write/execute permissions on a file.
326 * We use "fsuid" for this, letting us set arbitrary permissions
327 * for filesystem access without changing the "normal" uids which
328 * are used for other things.
330 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
331 * request cannot be satisfied (eg. requires blocking or too much complexity).
332 * It would then be called again in ref-walk mode.
334 int generic_permission(struct inode *inode, int mask)
339 * Do the basic permission checks.
341 ret = acl_permission_check(inode, mask);
345 if (S_ISDIR(inode->i_mode)) {
346 /* DACs are overridable for directories */
347 if (!(mask & MAY_WRITE))
348 if (capable_wrt_inode_uidgid(inode,
349 CAP_DAC_READ_SEARCH))
351 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
357 * Searching includes executable on directories, else just read.
359 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
360 if (mask == MAY_READ)
361 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
364 * Read/write DACs are always overridable.
365 * Executable DACs are overridable when there is
366 * at least one exec bit set.
368 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
369 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
374 EXPORT_SYMBOL(generic_permission);
377 * We _really_ want to just do "generic_permission()" without
378 * even looking at the inode->i_op values. So we keep a cache
379 * flag in inode->i_opflags, that says "this has not special
380 * permission function, use the fast case".
382 static inline int do_inode_permission(struct inode *inode, int mask)
384 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
385 if (likely(inode->i_op->permission))
386 return inode->i_op->permission(inode, mask);
388 /* This gets set once for the inode lifetime */
389 spin_lock(&inode->i_lock);
390 inode->i_opflags |= IOP_FASTPERM;
391 spin_unlock(&inode->i_lock);
393 return generic_permission(inode, mask);
397 * sb_permission - Check superblock-level permissions
398 * @sb: Superblock of inode to check permission on
399 * @inode: Inode to check permission on
400 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
402 * Separate out file-system wide checks from inode-specific permission checks.
404 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
406 if (unlikely(mask & MAY_WRITE)) {
407 umode_t mode = inode->i_mode;
409 /* Nobody gets write access to a read-only fs. */
410 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
417 * inode_permission - Check for access rights to a given inode
418 * @inode: Inode to check permission on
419 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
421 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
422 * this, letting us set arbitrary permissions for filesystem access without
423 * changing the "normal" UIDs which are used for other things.
425 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
427 int inode_permission(struct inode *inode, int mask)
431 retval = sb_permission(inode->i_sb, inode, mask);
435 if (unlikely(mask & MAY_WRITE)) {
437 * Nobody gets write access to an immutable file.
439 if (IS_IMMUTABLE(inode))
443 * Updating mtime will likely cause i_uid and i_gid to be
444 * written back improperly if their true value is unknown
447 if (HAS_UNMAPPED_ID(inode))
451 retval = do_inode_permission(inode, mask);
455 retval = devcgroup_inode_permission(inode, mask);
459 return security_inode_permission(inode, mask);
461 EXPORT_SYMBOL(inode_permission);
464 * path_get - get a reference to a path
465 * @path: path to get the reference to
467 * Given a path increment the reference count to the dentry and the vfsmount.
469 void path_get(const struct path *path)
474 EXPORT_SYMBOL(path_get);
477 * path_put - put a reference to a path
478 * @path: path to put the reference to
480 * Given a path decrement the reference count to the dentry and the vfsmount.
482 void path_put(const struct path *path)
487 EXPORT_SYMBOL(path_put);
489 #define EMBEDDED_LEVELS 2
494 struct inode *inode; /* path.dentry.d_inode */
499 int total_link_count;
502 struct delayed_call done;
505 } *stack, internal[EMBEDDED_LEVELS];
506 struct filename *name;
507 struct nameidata *saved;
508 struct inode *link_inode;
511 } __randomize_layout;
513 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
515 struct nameidata *old = current->nameidata;
516 p->stack = p->internal;
519 p->total_link_count = old ? old->total_link_count : 0;
521 current->nameidata = p;
524 static void restore_nameidata(void)
526 struct nameidata *now = current->nameidata, *old = now->saved;
528 current->nameidata = old;
530 old->total_link_count = now->total_link_count;
531 if (now->stack != now->internal)
535 static int __nd_alloc_stack(struct nameidata *nd)
539 if (nd->flags & LOOKUP_RCU) {
540 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
545 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
550 memcpy(p, nd->internal, sizeof(nd->internal));
556 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
557 * @path: nameidate to verify
559 * Rename can sometimes move a file or directory outside of a bind
560 * mount, path_connected allows those cases to be detected.
562 static bool path_connected(const struct path *path)
564 struct vfsmount *mnt = path->mnt;
565 struct super_block *sb = mnt->mnt_sb;
567 /* Bind mounts and multi-root filesystems can have disconnected paths */
568 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
571 return is_subdir(path->dentry, mnt->mnt_root);
574 static inline int nd_alloc_stack(struct nameidata *nd)
576 if (likely(nd->depth != EMBEDDED_LEVELS))
578 if (likely(nd->stack != nd->internal))
580 return __nd_alloc_stack(nd);
583 static void drop_links(struct nameidata *nd)
587 struct saved *last = nd->stack + i;
588 do_delayed_call(&last->done);
589 clear_delayed_call(&last->done);
593 static void terminate_walk(struct nameidata *nd)
596 if (!(nd->flags & LOOKUP_RCU)) {
599 for (i = 0; i < nd->depth; i++)
600 path_put(&nd->stack[i].link);
601 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
606 nd->flags &= ~LOOKUP_RCU;
607 if (!(nd->flags & LOOKUP_ROOT))
614 /* path_put is needed afterwards regardless of success or failure */
615 static bool legitimize_path(struct nameidata *nd,
616 struct path *path, unsigned seq)
618 int res = __legitimize_mnt(path->mnt, nd->m_seq);
625 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
629 return !read_seqcount_retry(&path->dentry->d_seq, seq);
632 static bool legitimize_links(struct nameidata *nd)
635 for (i = 0; i < nd->depth; i++) {
636 struct saved *last = nd->stack + i;
637 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
647 * Path walking has 2 modes, rcu-walk and ref-walk (see
648 * Documentation/filesystems/path-lookup.txt). In situations when we can't
649 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
650 * normal reference counts on dentries and vfsmounts to transition to ref-walk
651 * mode. Refcounts are grabbed at the last known good point before rcu-walk
652 * got stuck, so ref-walk may continue from there. If this is not successful
653 * (eg. a seqcount has changed), then failure is returned and it's up to caller
654 * to restart the path walk from the beginning in ref-walk mode.
658 * unlazy_walk - try to switch to ref-walk mode.
659 * @nd: nameidata pathwalk data
660 * Returns: 0 on success, -ECHILD on failure
662 * unlazy_walk attempts to legitimize the current nd->path and nd->root
664 * Must be called from rcu-walk context.
665 * Nothing should touch nameidata between unlazy_walk() failure and
668 static int unlazy_walk(struct nameidata *nd)
670 struct dentry *parent = nd->path.dentry;
672 BUG_ON(!(nd->flags & LOOKUP_RCU));
674 nd->flags &= ~LOOKUP_RCU;
675 if (unlikely(!legitimize_links(nd)))
677 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
679 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
680 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq)))
684 BUG_ON(nd->inode != parent->d_inode);
689 nd->path.dentry = NULL;
691 if (!(nd->flags & LOOKUP_ROOT))
699 * unlazy_child - try to switch to ref-walk mode.
700 * @nd: nameidata pathwalk data
701 * @dentry: child of nd->path.dentry
702 * @seq: seq number to check dentry against
703 * Returns: 0 on success, -ECHILD on failure
705 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
706 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
707 * @nd. Must be called from rcu-walk context.
708 * Nothing should touch nameidata between unlazy_child() failure and
711 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
713 BUG_ON(!(nd->flags & LOOKUP_RCU));
715 nd->flags &= ~LOOKUP_RCU;
716 if (unlikely(!legitimize_links(nd)))
718 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
720 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
724 * We need to move both the parent and the dentry from the RCU domain
725 * to be properly refcounted. And the sequence number in the dentry
726 * validates *both* dentry counters, since we checked the sequence
727 * number of the parent after we got the child sequence number. So we
728 * know the parent must still be valid if the child sequence number is
730 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
732 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) {
738 * Sequence counts matched. Now make sure that the root is
739 * still valid and get it if required.
741 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
742 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
755 nd->path.dentry = NULL;
759 if (!(nd->flags & LOOKUP_ROOT))
764 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
766 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
767 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)))
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 struct fs_struct *fs = current->fs;
814 if (nd->flags & LOOKUP_RCU) {
818 seq = read_seqcount_begin(&fs->seq);
820 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
821 } while (read_seqcount_retry(&fs->seq, seq));
823 get_fs_root(fs, &nd->root);
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;
846 static int nd_jump_root(struct nameidata *nd)
848 if (nd->flags & LOOKUP_RCU) {
852 nd->inode = d->d_inode;
853 nd->seq = nd->root_seq;
854 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
860 nd->inode = nd->path.dentry->d_inode;
862 nd->flags |= LOOKUP_JUMPED;
867 * Helper to directly jump to a known parsed path from ->get_link,
868 * caller must have taken a reference to path beforehand.
870 void nd_jump_link(struct path *path)
872 struct nameidata *nd = current->nameidata;
876 nd->inode = nd->path.dentry->d_inode;
877 nd->flags |= LOOKUP_JUMPED;
880 static inline void put_link(struct nameidata *nd)
882 struct saved *last = nd->stack + --nd->depth;
883 do_delayed_call(&last->done);
884 if (!(nd->flags & LOOKUP_RCU))
885 path_put(&last->link);
888 int sysctl_protected_symlinks __read_mostly = 0;
889 int sysctl_protected_hardlinks __read_mostly = 0;
892 * may_follow_link - Check symlink following for unsafe situations
893 * @nd: nameidata pathwalk data
895 * In the case of the sysctl_protected_symlinks sysctl being enabled,
896 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
897 * in a sticky world-writable directory. This is to protect privileged
898 * processes from failing races against path names that may change out
899 * from under them by way of other users creating malicious symlinks.
900 * It will permit symlinks to be followed only when outside a sticky
901 * world-writable directory, or when the uid of the symlink and follower
902 * match, or when the directory owner matches the symlink's owner.
904 * Returns 0 if following the symlink is allowed, -ve on error.
906 static inline int may_follow_link(struct nameidata *nd)
908 const struct inode *inode;
909 const struct inode *parent;
912 if (!sysctl_protected_symlinks)
915 /* Allowed if owner and follower match. */
916 inode = nd->link_inode;
917 if (uid_eq(current_cred()->fsuid, inode->i_uid))
920 /* Allowed if parent directory not sticky and world-writable. */
922 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
925 /* Allowed if parent directory and link owner match. */
926 puid = parent->i_uid;
927 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
930 if (nd->flags & LOOKUP_RCU)
933 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
934 audit_log_link_denied("follow_link");
939 * safe_hardlink_source - Check for safe hardlink conditions
940 * @inode: the source inode to hardlink from
942 * Return false if at least one of the following conditions:
943 * - inode is not a regular file
945 * - inode is setgid and group-exec
946 * - access failure for read and write
948 * Otherwise returns true.
950 static bool safe_hardlink_source(struct inode *inode)
952 umode_t mode = inode->i_mode;
954 /* Special files should not get pinned to the filesystem. */
958 /* Setuid files should not get pinned to the filesystem. */
962 /* Executable setgid files should not get pinned to the filesystem. */
963 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
966 /* Hardlinking to unreadable or unwritable sources is dangerous. */
967 if (inode_permission(inode, MAY_READ | MAY_WRITE))
974 * may_linkat - Check permissions for creating a hardlink
975 * @link: the source to hardlink from
977 * Block hardlink when all of:
978 * - sysctl_protected_hardlinks enabled
979 * - fsuid does not match inode
980 * - hardlink source is unsafe (see safe_hardlink_source() above)
981 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
983 * Returns 0 if successful, -ve on error.
985 static int may_linkat(struct path *link)
987 struct inode *inode = link->dentry->d_inode;
989 /* Inode writeback is not safe when the uid or gid are invalid. */
990 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
993 if (!sysctl_protected_hardlinks)
996 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
997 * otherwise, it must be a safe source.
999 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1002 audit_log_link_denied("linkat");
1006 static __always_inline
1007 const char *get_link(struct nameidata *nd)
1009 struct saved *last = nd->stack + nd->depth - 1;
1010 struct dentry *dentry = last->link.dentry;
1011 struct inode *inode = nd->link_inode;
1015 if (!(nd->flags & LOOKUP_RCU)) {
1016 touch_atime(&last->link);
1018 } else if (atime_needs_update_rcu(&last->link, inode)) {
1019 if (unlikely(unlazy_walk(nd)))
1020 return ERR_PTR(-ECHILD);
1021 touch_atime(&last->link);
1024 error = security_inode_follow_link(dentry, inode,
1025 nd->flags & LOOKUP_RCU);
1026 if (unlikely(error))
1027 return ERR_PTR(error);
1029 nd->last_type = LAST_BIND;
1030 res = inode->i_link;
1032 const char * (*get)(struct dentry *, struct inode *,
1033 struct delayed_call *);
1034 get = inode->i_op->get_link;
1035 if (nd->flags & LOOKUP_RCU) {
1036 res = get(NULL, inode, &last->done);
1037 if (res == ERR_PTR(-ECHILD)) {
1038 if (unlikely(unlazy_walk(nd)))
1039 return ERR_PTR(-ECHILD);
1040 res = get(dentry, inode, &last->done);
1043 res = get(dentry, inode, &last->done);
1045 if (IS_ERR_OR_NULL(res))
1051 if (unlikely(nd_jump_root(nd)))
1052 return ERR_PTR(-ECHILD);
1053 while (unlikely(*++res == '/'))
1062 * follow_up - Find the mountpoint of path's vfsmount
1064 * Given a path, find the mountpoint of its source file system.
1065 * Replace @path with the path of the mountpoint in the parent mount.
1068 * Return 1 if we went up a level and 0 if we were already at the
1071 int follow_up(struct path *path)
1073 struct mount *mnt = real_mount(path->mnt);
1074 struct mount *parent;
1075 struct dentry *mountpoint;
1077 read_seqlock_excl(&mount_lock);
1078 parent = mnt->mnt_parent;
1079 if (parent == mnt) {
1080 read_sequnlock_excl(&mount_lock);
1083 mntget(&parent->mnt);
1084 mountpoint = dget(mnt->mnt_mountpoint);
1085 read_sequnlock_excl(&mount_lock);
1087 path->dentry = mountpoint;
1089 path->mnt = &parent->mnt;
1092 EXPORT_SYMBOL(follow_up);
1095 * Perform an automount
1096 * - return -EISDIR to tell follow_managed() to stop and return the path we
1099 static int follow_automount(struct path *path, struct nameidata *nd,
1102 struct vfsmount *mnt;
1105 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1108 /* We don't want to mount if someone's just doing a stat -
1109 * unless they're stat'ing a directory and appended a '/' to
1112 * We do, however, want to mount if someone wants to open or
1113 * create a file of any type under the mountpoint, wants to
1114 * traverse through the mountpoint or wants to open the
1115 * mounted directory. Also, autofs may mark negative dentries
1116 * as being automount points. These will need the attentions
1117 * of the daemon to instantiate them before they can be used.
1119 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1120 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1121 path->dentry->d_inode)
1124 nd->total_link_count++;
1125 if (nd->total_link_count >= 40)
1128 mnt = path->dentry->d_op->d_automount(path);
1131 * The filesystem is allowed to return -EISDIR here to indicate
1132 * it doesn't want to automount. For instance, autofs would do
1133 * this so that its userspace daemon can mount on this dentry.
1135 * However, we can only permit this if it's a terminal point in
1136 * the path being looked up; if it wasn't then the remainder of
1137 * the path is inaccessible and we should say so.
1139 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1141 return PTR_ERR(mnt);
1144 if (!mnt) /* mount collision */
1147 if (!*need_mntput) {
1148 /* lock_mount() may release path->mnt on error */
1150 *need_mntput = true;
1152 err = finish_automount(mnt, path);
1156 /* Someone else made a mount here whilst we were busy */
1161 path->dentry = dget(mnt->mnt_root);
1170 * Handle a dentry that is managed in some way.
1171 * - Flagged for transit management (autofs)
1172 * - Flagged as mountpoint
1173 * - Flagged as automount point
1175 * This may only be called in refwalk mode.
1177 * Serialization is taken care of in namespace.c
1179 static int follow_managed(struct path *path, struct nameidata *nd)
1181 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1183 bool need_mntput = false;
1186 /* Given that we're not holding a lock here, we retain the value in a
1187 * local variable for each dentry as we look at it so that we don't see
1188 * the components of that value change under us */
1189 while (managed = READ_ONCE(path->dentry->d_flags),
1190 managed &= DCACHE_MANAGED_DENTRY,
1191 unlikely(managed != 0)) {
1192 /* Allow the filesystem to manage the transit without i_mutex
1194 if (managed & DCACHE_MANAGE_TRANSIT) {
1195 BUG_ON(!path->dentry->d_op);
1196 BUG_ON(!path->dentry->d_op->d_manage);
1197 ret = path->dentry->d_op->d_manage(path, false);
1202 /* Transit to a mounted filesystem. */
1203 if (managed & DCACHE_MOUNTED) {
1204 struct vfsmount *mounted = lookup_mnt(path);
1209 path->mnt = mounted;
1210 path->dentry = dget(mounted->mnt_root);
1215 /* Something is mounted on this dentry in another
1216 * namespace and/or whatever was mounted there in this
1217 * namespace got unmounted before lookup_mnt() could
1221 /* Handle an automount point */
1222 if (managed & DCACHE_NEED_AUTOMOUNT) {
1223 ret = follow_automount(path, nd, &need_mntput);
1229 /* We didn't change the current path point */
1233 if (need_mntput && path->mnt == mnt)
1235 if (ret == -EISDIR || !ret)
1238 nd->flags |= LOOKUP_JUMPED;
1239 if (unlikely(ret < 0))
1240 path_put_conditional(path, nd);
1244 int follow_down_one(struct path *path)
1246 struct vfsmount *mounted;
1248 mounted = lookup_mnt(path);
1252 path->mnt = mounted;
1253 path->dentry = dget(mounted->mnt_root);
1258 EXPORT_SYMBOL(follow_down_one);
1260 static inline int managed_dentry_rcu(const struct path *path)
1262 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1263 path->dentry->d_op->d_manage(path, true) : 0;
1267 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1268 * we meet a managed dentry that would need blocking.
1270 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1271 struct inode **inode, unsigned *seqp)
1274 struct mount *mounted;
1276 * Don't forget we might have a non-mountpoint managed dentry
1277 * that wants to block transit.
1279 switch (managed_dentry_rcu(path)) {
1289 if (!d_mountpoint(path->dentry))
1290 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1292 mounted = __lookup_mnt(path->mnt, path->dentry);
1295 path->mnt = &mounted->mnt;
1296 path->dentry = mounted->mnt.mnt_root;
1297 nd->flags |= LOOKUP_JUMPED;
1298 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1300 * Update the inode too. We don't need to re-check the
1301 * dentry sequence number here after this d_inode read,
1302 * because a mount-point is always pinned.
1304 *inode = path->dentry->d_inode;
1306 return !read_seqretry(&mount_lock, nd->m_seq) &&
1307 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1310 static int follow_dotdot_rcu(struct nameidata *nd)
1312 struct inode *inode = nd->inode;
1315 if (path_equal(&nd->path, &nd->root))
1317 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1318 struct dentry *old = nd->path.dentry;
1319 struct dentry *parent = old->d_parent;
1322 inode = parent->d_inode;
1323 seq = read_seqcount_begin(&parent->d_seq);
1324 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1326 nd->path.dentry = parent;
1328 if (unlikely(!path_connected(&nd->path)))
1332 struct mount *mnt = real_mount(nd->path.mnt);
1333 struct mount *mparent = mnt->mnt_parent;
1334 struct dentry *mountpoint = mnt->mnt_mountpoint;
1335 struct inode *inode2 = mountpoint->d_inode;
1336 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1337 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1339 if (&mparent->mnt == nd->path.mnt)
1341 /* we know that mountpoint was pinned */
1342 nd->path.dentry = mountpoint;
1343 nd->path.mnt = &mparent->mnt;
1348 while (unlikely(d_mountpoint(nd->path.dentry))) {
1349 struct mount *mounted;
1350 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1351 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1355 nd->path.mnt = &mounted->mnt;
1356 nd->path.dentry = mounted->mnt.mnt_root;
1357 inode = nd->path.dentry->d_inode;
1358 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1365 * Follow down to the covering mount currently visible to userspace. At each
1366 * point, the filesystem owning that dentry may be queried as to whether the
1367 * caller is permitted to proceed or not.
1369 int follow_down(struct path *path)
1374 while (managed = READ_ONCE(path->dentry->d_flags),
1375 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1376 /* Allow the filesystem to manage the transit without i_mutex
1379 * We indicate to the filesystem if someone is trying to mount
1380 * something here. This gives autofs the chance to deny anyone
1381 * other than its daemon the right to mount on its
1384 * The filesystem may sleep at this point.
1386 if (managed & DCACHE_MANAGE_TRANSIT) {
1387 BUG_ON(!path->dentry->d_op);
1388 BUG_ON(!path->dentry->d_op->d_manage);
1389 ret = path->dentry->d_op->d_manage(path, false);
1391 return ret == -EISDIR ? 0 : ret;
1394 /* Transit to a mounted filesystem. */
1395 if (managed & DCACHE_MOUNTED) {
1396 struct vfsmount *mounted = lookup_mnt(path);
1401 path->mnt = mounted;
1402 path->dentry = dget(mounted->mnt_root);
1406 /* Don't handle automount points here */
1411 EXPORT_SYMBOL(follow_down);
1414 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1416 static void follow_mount(struct path *path)
1418 while (d_mountpoint(path->dentry)) {
1419 struct vfsmount *mounted = lookup_mnt(path);
1424 path->mnt = mounted;
1425 path->dentry = dget(mounted->mnt_root);
1429 static int path_parent_directory(struct path *path)
1431 struct dentry *old = path->dentry;
1432 /* rare case of legitimate dget_parent()... */
1433 path->dentry = dget_parent(path->dentry);
1435 if (unlikely(!path_connected(path)))
1440 static int follow_dotdot(struct nameidata *nd)
1443 if (path_equal(&nd->path, &nd->root))
1445 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1446 int ret = path_parent_directory(&nd->path);
1451 if (!follow_up(&nd->path))
1454 follow_mount(&nd->path);
1455 nd->inode = nd->path.dentry->d_inode;
1460 * This looks up the name in dcache and possibly revalidates the found dentry.
1461 * NULL is returned if the dentry does not exist in the cache.
1463 static struct dentry *lookup_dcache(const struct qstr *name,
1467 struct dentry *dentry = d_lookup(dir, name);
1469 int error = d_revalidate(dentry, flags);
1470 if (unlikely(error <= 0)) {
1472 d_invalidate(dentry);
1474 return ERR_PTR(error);
1481 * Parent directory has inode locked exclusive. This is one
1482 * and only case when ->lookup() gets called on non in-lookup
1483 * dentries - as the matter of fact, this only gets called
1484 * when directory is guaranteed to have no in-lookup children
1487 static struct dentry *__lookup_hash(const struct qstr *name,
1488 struct dentry *base, unsigned int flags)
1490 struct dentry *dentry = lookup_dcache(name, base, flags);
1492 struct inode *dir = base->d_inode;
1497 /* Don't create child dentry for a dead directory. */
1498 if (unlikely(IS_DEADDIR(dir)))
1499 return ERR_PTR(-ENOENT);
1501 dentry = d_alloc(base, name);
1502 if (unlikely(!dentry))
1503 return ERR_PTR(-ENOMEM);
1505 old = dir->i_op->lookup(dir, dentry, flags);
1506 if (unlikely(old)) {
1513 static int lookup_fast(struct nameidata *nd,
1514 struct path *path, struct inode **inode,
1517 struct vfsmount *mnt = nd->path.mnt;
1518 struct dentry *dentry, *parent = nd->path.dentry;
1523 * Rename seqlock is not required here because in the off chance
1524 * of a false negative due to a concurrent rename, the caller is
1525 * going to fall back to non-racy lookup.
1527 if (nd->flags & LOOKUP_RCU) {
1530 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1531 if (unlikely(!dentry)) {
1532 if (unlazy_walk(nd))
1538 * This sequence count validates that the inode matches
1539 * the dentry name information from lookup.
1541 *inode = d_backing_inode(dentry);
1542 negative = d_is_negative(dentry);
1543 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1547 * This sequence count validates that the parent had no
1548 * changes while we did the lookup of the dentry above.
1550 * The memory barrier in read_seqcount_begin of child is
1551 * enough, we can use __read_seqcount_retry here.
1553 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1557 status = d_revalidate(dentry, nd->flags);
1558 if (likely(status > 0)) {
1560 * Note: do negative dentry check after revalidation in
1561 * case that drops it.
1563 if (unlikely(negative))
1566 path->dentry = dentry;
1567 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1570 if (unlazy_child(nd, dentry, seq))
1572 if (unlikely(status == -ECHILD))
1573 /* we'd been told to redo it in non-rcu mode */
1574 status = d_revalidate(dentry, nd->flags);
1576 dentry = __d_lookup(parent, &nd->last);
1577 if (unlikely(!dentry))
1579 status = d_revalidate(dentry, nd->flags);
1581 if (unlikely(status <= 0)) {
1583 d_invalidate(dentry);
1587 if (unlikely(d_is_negative(dentry))) {
1593 path->dentry = dentry;
1594 err = follow_managed(path, nd);
1595 if (likely(err > 0))
1596 *inode = d_backing_inode(path->dentry);
1600 /* Fast lookup failed, do it the slow way */
1601 static struct dentry *__lookup_slow(const struct qstr *name,
1605 struct dentry *dentry, *old;
1606 struct inode *inode = dir->d_inode;
1607 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1609 /* Don't go there if it's already dead */
1610 if (unlikely(IS_DEADDIR(inode)))
1611 return ERR_PTR(-ENOENT);
1613 dentry = d_alloc_parallel(dir, name, &wq);
1616 if (unlikely(!d_in_lookup(dentry))) {
1617 if (!(flags & LOOKUP_NO_REVAL)) {
1618 int error = d_revalidate(dentry, flags);
1619 if (unlikely(error <= 0)) {
1621 d_invalidate(dentry);
1626 dentry = ERR_PTR(error);
1630 old = inode->i_op->lookup(inode, dentry, flags);
1631 d_lookup_done(dentry);
1632 if (unlikely(old)) {
1640 static struct dentry *lookup_slow(const struct qstr *name,
1644 struct inode *inode = dir->d_inode;
1646 inode_lock_shared(inode);
1647 res = __lookup_slow(name, dir, flags);
1648 inode_unlock_shared(inode);
1652 static inline int may_lookup(struct nameidata *nd)
1654 if (nd->flags & LOOKUP_RCU) {
1655 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1658 if (unlazy_walk(nd))
1661 return inode_permission(nd->inode, MAY_EXEC);
1664 static inline int handle_dots(struct nameidata *nd, int type)
1666 if (type == LAST_DOTDOT) {
1669 if (nd->flags & LOOKUP_RCU) {
1670 return follow_dotdot_rcu(nd);
1672 return follow_dotdot(nd);
1677 static int pick_link(struct nameidata *nd, struct path *link,
1678 struct inode *inode, unsigned seq)
1682 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1683 path_to_nameidata(link, nd);
1686 if (!(nd->flags & LOOKUP_RCU)) {
1687 if (link->mnt == nd->path.mnt)
1690 error = nd_alloc_stack(nd);
1691 if (unlikely(error)) {
1692 if (error == -ECHILD) {
1693 if (unlikely(!legitimize_path(nd, link, seq))) {
1696 nd->flags &= ~LOOKUP_RCU;
1697 nd->path.mnt = NULL;
1698 nd->path.dentry = NULL;
1699 if (!(nd->flags & LOOKUP_ROOT))
1700 nd->root.mnt = NULL;
1702 } else if (likely(unlazy_walk(nd)) == 0)
1703 error = nd_alloc_stack(nd);
1711 last = nd->stack + nd->depth++;
1713 clear_delayed_call(&last->done);
1714 nd->link_inode = inode;
1719 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1722 * Do we need to follow links? We _really_ want to be able
1723 * to do this check without having to look at inode->i_op,
1724 * so we keep a cache of "no, this doesn't need follow_link"
1725 * for the common case.
1727 static inline int step_into(struct nameidata *nd, struct path *path,
1728 int flags, struct inode *inode, unsigned seq)
1730 if (!(flags & WALK_MORE) && nd->depth)
1732 if (likely(!d_is_symlink(path->dentry)) ||
1733 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1734 /* not a symlink or should not follow */
1735 path_to_nameidata(path, nd);
1740 /* make sure that d_is_symlink above matches inode */
1741 if (nd->flags & LOOKUP_RCU) {
1742 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1745 return pick_link(nd, path, inode, seq);
1748 static int walk_component(struct nameidata *nd, int flags)
1751 struct inode *inode;
1755 * "." and ".." are special - ".." especially so because it has
1756 * to be able to know about the current root directory and
1757 * parent relationships.
1759 if (unlikely(nd->last_type != LAST_NORM)) {
1760 err = handle_dots(nd, nd->last_type);
1761 if (!(flags & WALK_MORE) && nd->depth)
1765 err = lookup_fast(nd, &path, &inode, &seq);
1766 if (unlikely(err <= 0)) {
1769 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1771 if (IS_ERR(path.dentry))
1772 return PTR_ERR(path.dentry);
1774 path.mnt = nd->path.mnt;
1775 err = follow_managed(&path, nd);
1776 if (unlikely(err < 0))
1779 if (unlikely(d_is_negative(path.dentry))) {
1780 path_to_nameidata(&path, nd);
1784 seq = 0; /* we are already out of RCU mode */
1785 inode = d_backing_inode(path.dentry);
1788 return step_into(nd, &path, flags, inode, seq);
1792 * We can do the critical dentry name comparison and hashing
1793 * operations one word at a time, but we are limited to:
1795 * - Architectures with fast unaligned word accesses. We could
1796 * do a "get_unaligned()" if this helps and is sufficiently
1799 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1800 * do not trap on the (extremely unlikely) case of a page
1801 * crossing operation.
1803 * - Furthermore, we need an efficient 64-bit compile for the
1804 * 64-bit case in order to generate the "number of bytes in
1805 * the final mask". Again, that could be replaced with a
1806 * efficient population count instruction or similar.
1808 #ifdef CONFIG_DCACHE_WORD_ACCESS
1810 #include <asm/word-at-a-time.h>
1814 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1816 #elif defined(CONFIG_64BIT)
1818 * Register pressure in the mixing function is an issue, particularly
1819 * on 32-bit x86, but almost any function requires one state value and
1820 * one temporary. Instead, use a function designed for two state values
1821 * and no temporaries.
1823 * This function cannot create a collision in only two iterations, so
1824 * we have two iterations to achieve avalanche. In those two iterations,
1825 * we have six layers of mixing, which is enough to spread one bit's
1826 * influence out to 2^6 = 64 state bits.
1828 * Rotate constants are scored by considering either 64 one-bit input
1829 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1830 * probability of that delta causing a change to each of the 128 output
1831 * bits, using a sample of random initial states.
1833 * The Shannon entropy of the computed probabilities is then summed
1834 * to produce a score. Ideally, any input change has a 50% chance of
1835 * toggling any given output bit.
1837 * Mixing scores (in bits) for (12,45):
1838 * Input delta: 1-bit 2-bit
1839 * 1 round: 713.3 42542.6
1840 * 2 rounds: 2753.7 140389.8
1841 * 3 rounds: 5954.1 233458.2
1842 * 4 rounds: 7862.6 256672.2
1843 * Perfect: 8192 258048
1844 * (64*128) (64*63/2 * 128)
1846 #define HASH_MIX(x, y, a) \
1848 y ^= x, x = rol64(x,12),\
1849 x += y, y = rol64(y,45),\
1853 * Fold two longs into one 32-bit hash value. This must be fast, but
1854 * latency isn't quite as critical, as there is a fair bit of additional
1855 * work done before the hash value is used.
1857 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1859 y ^= x * GOLDEN_RATIO_64;
1860 y *= GOLDEN_RATIO_64;
1864 #else /* 32-bit case */
1867 * Mixing scores (in bits) for (7,20):
1868 * Input delta: 1-bit 2-bit
1869 * 1 round: 330.3 9201.6
1870 * 2 rounds: 1246.4 25475.4
1871 * 3 rounds: 1907.1 31295.1
1872 * 4 rounds: 2042.3 31718.6
1873 * Perfect: 2048 31744
1874 * (32*64) (32*31/2 * 64)
1876 #define HASH_MIX(x, y, a) \
1878 y ^= x, x = rol32(x, 7),\
1879 x += y, y = rol32(y,20),\
1882 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1884 /* Use arch-optimized multiply if one exists */
1885 return __hash_32(y ^ __hash_32(x));
1891 * Return the hash of a string of known length. This is carfully
1892 * designed to match hash_name(), which is the more critical function.
1893 * In particular, we must end by hashing a final word containing 0..7
1894 * payload bytes, to match the way that hash_name() iterates until it
1895 * finds the delimiter after the name.
1897 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1899 unsigned long a, x = 0, y = (unsigned long)salt;
1904 a = load_unaligned_zeropad(name);
1905 if (len < sizeof(unsigned long))
1908 name += sizeof(unsigned long);
1909 len -= sizeof(unsigned long);
1911 x ^= a & bytemask_from_count(len);
1913 return fold_hash(x, y);
1915 EXPORT_SYMBOL(full_name_hash);
1917 /* Return the "hash_len" (hash and length) of a null-terminated string */
1918 u64 hashlen_string(const void *salt, const char *name)
1920 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1921 unsigned long adata, mask, len;
1922 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1929 len += sizeof(unsigned long);
1931 a = load_unaligned_zeropad(name+len);
1932 } while (!has_zero(a, &adata, &constants));
1934 adata = prep_zero_mask(a, adata, &constants);
1935 mask = create_zero_mask(adata);
1936 x ^= a & zero_bytemask(mask);
1938 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1940 EXPORT_SYMBOL(hashlen_string);
1943 * Calculate the length and hash of the path component, and
1944 * return the "hash_len" as the result.
1946 static inline u64 hash_name(const void *salt, const char *name)
1948 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1949 unsigned long adata, bdata, mask, len;
1950 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1957 len += sizeof(unsigned long);
1959 a = load_unaligned_zeropad(name+len);
1960 b = a ^ REPEAT_BYTE('/');
1961 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1963 adata = prep_zero_mask(a, adata, &constants);
1964 bdata = prep_zero_mask(b, bdata, &constants);
1965 mask = create_zero_mask(adata | bdata);
1966 x ^= a & zero_bytemask(mask);
1968 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1971 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1973 /* Return the hash of a string of known length */
1974 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1976 unsigned long hash = init_name_hash(salt);
1978 hash = partial_name_hash((unsigned char)*name++, hash);
1979 return end_name_hash(hash);
1981 EXPORT_SYMBOL(full_name_hash);
1983 /* Return the "hash_len" (hash and length) of a null-terminated string */
1984 u64 hashlen_string(const void *salt, const char *name)
1986 unsigned long hash = init_name_hash(salt);
1987 unsigned long len = 0, c;
1989 c = (unsigned char)*name;
1992 hash = partial_name_hash(c, hash);
1993 c = (unsigned char)name[len];
1995 return hashlen_create(end_name_hash(hash), len);
1997 EXPORT_SYMBOL(hashlen_string);
2000 * We know there's a real path component here of at least
2003 static inline u64 hash_name(const void *salt, const char *name)
2005 unsigned long hash = init_name_hash(salt);
2006 unsigned long len = 0, c;
2008 c = (unsigned char)*name;
2011 hash = partial_name_hash(c, hash);
2012 c = (unsigned char)name[len];
2013 } while (c && c != '/');
2014 return hashlen_create(end_name_hash(hash), len);
2021 * This is the basic name resolution function, turning a pathname into
2022 * the final dentry. We expect 'base' to be positive and a directory.
2024 * Returns 0 and nd will have valid dentry and mnt on success.
2025 * Returns error and drops reference to input namei data on failure.
2027 static int link_path_walk(const char *name, struct nameidata *nd)
2032 return PTR_ERR(name);
2038 /* At this point we know we have a real path component. */
2043 err = may_lookup(nd);
2047 hash_len = hash_name(nd->path.dentry, name);
2050 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2052 if (name[1] == '.') {
2054 nd->flags |= LOOKUP_JUMPED;
2060 if (likely(type == LAST_NORM)) {
2061 struct dentry *parent = nd->path.dentry;
2062 nd->flags &= ~LOOKUP_JUMPED;
2063 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2064 struct qstr this = { { .hash_len = hash_len }, .name = name };
2065 err = parent->d_op->d_hash(parent, &this);
2068 hash_len = this.hash_len;
2073 nd->last.hash_len = hash_len;
2074 nd->last.name = name;
2075 nd->last_type = type;
2077 name += hashlen_len(hash_len);
2081 * If it wasn't NUL, we know it was '/'. Skip that
2082 * slash, and continue until no more slashes.
2086 } while (unlikely(*name == '/'));
2087 if (unlikely(!*name)) {
2089 /* pathname body, done */
2092 name = nd->stack[nd->depth - 1].name;
2093 /* trailing symlink, done */
2096 /* last component of nested symlink */
2097 err = walk_component(nd, WALK_FOLLOW);
2099 /* not the last component */
2100 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2106 const char *s = get_link(nd);
2115 nd->stack[nd->depth - 1].name = name;
2120 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2121 if (nd->flags & LOOKUP_RCU) {
2122 if (unlazy_walk(nd))
2130 /* must be paired with terminate_walk() */
2131 static const char *path_init(struct nameidata *nd, unsigned flags)
2133 const char *s = nd->name->name;
2136 flags &= ~LOOKUP_RCU;
2137 if (flags & LOOKUP_RCU)
2140 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2141 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2143 if (flags & LOOKUP_ROOT) {
2144 struct dentry *root = nd->root.dentry;
2145 struct inode *inode = root->d_inode;
2146 if (*s && unlikely(!d_can_lookup(root)))
2147 return ERR_PTR(-ENOTDIR);
2148 nd->path = nd->root;
2150 if (flags & LOOKUP_RCU) {
2151 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2152 nd->root_seq = nd->seq;
2153 nd->m_seq = read_seqbegin(&mount_lock);
2155 path_get(&nd->path);
2160 nd->root.mnt = NULL;
2161 nd->path.mnt = NULL;
2162 nd->path.dentry = NULL;
2164 nd->m_seq = read_seqbegin(&mount_lock);
2167 if (likely(!nd_jump_root(nd)))
2169 return ERR_PTR(-ECHILD);
2170 } else if (nd->dfd == AT_FDCWD) {
2171 if (flags & LOOKUP_RCU) {
2172 struct fs_struct *fs = current->fs;
2176 seq = read_seqcount_begin(&fs->seq);
2178 nd->inode = nd->path.dentry->d_inode;
2179 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2180 } while (read_seqcount_retry(&fs->seq, seq));
2182 get_fs_pwd(current->fs, &nd->path);
2183 nd->inode = nd->path.dentry->d_inode;
2187 /* Caller must check execute permissions on the starting path component */
2188 struct fd f = fdget_raw(nd->dfd);
2189 struct dentry *dentry;
2192 return ERR_PTR(-EBADF);
2194 dentry = f.file->f_path.dentry;
2196 if (*s && unlikely(!d_can_lookup(dentry))) {
2198 return ERR_PTR(-ENOTDIR);
2201 nd->path = f.file->f_path;
2202 if (flags & LOOKUP_RCU) {
2203 nd->inode = nd->path.dentry->d_inode;
2204 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2206 path_get(&nd->path);
2207 nd->inode = nd->path.dentry->d_inode;
2214 static const char *trailing_symlink(struct nameidata *nd)
2217 int error = may_follow_link(nd);
2218 if (unlikely(error))
2219 return ERR_PTR(error);
2220 nd->flags |= LOOKUP_PARENT;
2221 nd->stack[0].name = NULL;
2226 static inline int lookup_last(struct nameidata *nd)
2228 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2229 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2231 nd->flags &= ~LOOKUP_PARENT;
2232 return walk_component(nd, 0);
2235 static int handle_lookup_down(struct nameidata *nd)
2237 struct path path = nd->path;
2238 struct inode *inode = nd->inode;
2239 unsigned seq = nd->seq;
2242 if (nd->flags & LOOKUP_RCU) {
2244 * don't bother with unlazy_walk on failure - we are
2245 * at the very beginning of walk, so we lose nothing
2246 * if we simply redo everything in non-RCU mode
2248 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2252 err = follow_managed(&path, nd);
2253 if (unlikely(err < 0))
2255 inode = d_backing_inode(path.dentry);
2258 path_to_nameidata(&path, nd);
2264 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2265 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2267 const char *s = path_init(nd, flags);
2270 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2271 err = handle_lookup_down(nd);
2272 if (unlikely(err < 0))
2276 while (!(err = link_path_walk(s, nd))
2277 && ((err = lookup_last(nd)) > 0)) {
2278 s = trailing_symlink(nd);
2281 err = complete_walk(nd);
2283 if (!err && nd->flags & LOOKUP_DIRECTORY)
2284 if (!d_can_lookup(nd->path.dentry))
2288 nd->path.mnt = NULL;
2289 nd->path.dentry = NULL;
2295 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2296 struct path *path, struct path *root)
2299 struct nameidata nd;
2301 return PTR_ERR(name);
2302 if (unlikely(root)) {
2304 flags |= LOOKUP_ROOT;
2306 set_nameidata(&nd, dfd, name);
2307 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2308 if (unlikely(retval == -ECHILD))
2309 retval = path_lookupat(&nd, flags, path);
2310 if (unlikely(retval == -ESTALE))
2311 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2313 if (likely(!retval))
2314 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2315 restore_nameidata();
2320 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2321 static int path_parentat(struct nameidata *nd, unsigned flags,
2322 struct path *parent)
2324 const char *s = path_init(nd, flags);
2325 int err = link_path_walk(s, nd);
2327 err = complete_walk(nd);
2330 nd->path.mnt = NULL;
2331 nd->path.dentry = NULL;
2337 static struct filename *filename_parentat(int dfd, struct filename *name,
2338 unsigned int flags, struct path *parent,
2339 struct qstr *last, int *type)
2342 struct nameidata nd;
2346 set_nameidata(&nd, dfd, name);
2347 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2348 if (unlikely(retval == -ECHILD))
2349 retval = path_parentat(&nd, flags, parent);
2350 if (unlikely(retval == -ESTALE))
2351 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2352 if (likely(!retval)) {
2354 *type = nd.last_type;
2355 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2358 name = ERR_PTR(retval);
2360 restore_nameidata();
2364 /* does lookup, returns the object with parent locked */
2365 struct dentry *kern_path_locked(const char *name, struct path *path)
2367 struct filename *filename;
2372 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2374 if (IS_ERR(filename))
2375 return ERR_CAST(filename);
2376 if (unlikely(type != LAST_NORM)) {
2379 return ERR_PTR(-EINVAL);
2381 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2382 d = __lookup_hash(&last, path->dentry, 0);
2384 inode_unlock(path->dentry->d_inode);
2391 int kern_path(const char *name, unsigned int flags, struct path *path)
2393 return filename_lookup(AT_FDCWD, getname_kernel(name),
2396 EXPORT_SYMBOL(kern_path);
2399 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2400 * @dentry: pointer to dentry of the base directory
2401 * @mnt: pointer to vfs mount of the base directory
2402 * @name: pointer to file name
2403 * @flags: lookup flags
2404 * @path: pointer to struct path to fill
2406 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2407 const char *name, unsigned int flags,
2410 struct path root = {.mnt = mnt, .dentry = dentry};
2411 /* the first argument of filename_lookup() is ignored with root */
2412 return filename_lookup(AT_FDCWD, getname_kernel(name),
2413 flags , path, &root);
2415 EXPORT_SYMBOL(vfs_path_lookup);
2417 static int lookup_one_len_common(const char *name, struct dentry *base,
2418 int len, struct qstr *this)
2422 this->hash = full_name_hash(base, name, len);
2426 if (unlikely(name[0] == '.')) {
2427 if (len < 2 || (len == 2 && name[1] == '.'))
2432 unsigned int c = *(const unsigned char *)name++;
2433 if (c == '/' || c == '\0')
2437 * See if the low-level filesystem might want
2438 * to use its own hash..
2440 if (base->d_flags & DCACHE_OP_HASH) {
2441 int err = base->d_op->d_hash(base, this);
2446 return inode_permission(base->d_inode, MAY_EXEC);
2450 * try_lookup_one_len - filesystem helper to lookup single pathname component
2451 * @name: pathname component to lookup
2452 * @base: base directory to lookup from
2453 * @len: maximum length @len should be interpreted to
2455 * Look up a dentry by name in the dcache, returning NULL if it does not
2456 * currently exist. The function does not try to create a dentry.
2458 * Note that this routine is purely a helper for filesystem usage and should
2459 * not be called by generic code.
2461 * The caller must hold base->i_mutex.
2463 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2468 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2470 err = lookup_one_len_common(name, base, len, &this);
2472 return ERR_PTR(err);
2474 return lookup_dcache(&this, base, 0);
2476 EXPORT_SYMBOL(try_lookup_one_len);
2479 * lookup_one_len - filesystem helper to lookup single pathname component
2480 * @name: pathname component to lookup
2481 * @base: base directory to lookup from
2482 * @len: maximum length @len should be interpreted to
2484 * Note that this routine is purely a helper for filesystem usage and should
2485 * not be called by generic code.
2487 * The caller must hold base->i_mutex.
2489 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2491 struct dentry *dentry;
2495 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2497 err = lookup_one_len_common(name, base, len, &this);
2499 return ERR_PTR(err);
2501 dentry = lookup_dcache(&this, base, 0);
2502 return dentry ? dentry : __lookup_slow(&this, base, 0);
2504 EXPORT_SYMBOL(lookup_one_len);
2507 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2508 * @name: pathname component to lookup
2509 * @base: base directory to lookup from
2510 * @len: maximum length @len should be interpreted to
2512 * Note that this routine is purely a helper for filesystem usage and should
2513 * not be called by generic code.
2515 * Unlike lookup_one_len, it should be called without the parent
2516 * i_mutex held, and will take the i_mutex itself if necessary.
2518 struct dentry *lookup_one_len_unlocked(const char *name,
2519 struct dentry *base, int len)
2525 err = lookup_one_len_common(name, base, len, &this);
2527 return ERR_PTR(err);
2529 ret = lookup_dcache(&this, base, 0);
2531 ret = lookup_slow(&this, base, 0);
2534 EXPORT_SYMBOL(lookup_one_len_unlocked);
2536 #ifdef CONFIG_UNIX98_PTYS
2537 int path_pts(struct path *path)
2539 /* Find something mounted on "pts" in the same directory as
2542 struct dentry *child, *parent;
2546 ret = path_parent_directory(path);
2550 parent = path->dentry;
2553 child = d_hash_and_lookup(parent, &this);
2557 path->dentry = child;
2564 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2565 struct path *path, int *empty)
2567 return filename_lookup(dfd, getname_flags(name, flags, empty),
2570 EXPORT_SYMBOL(user_path_at_empty);
2573 * mountpoint_last - look up last component for umount
2574 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2576 * This is a special lookup_last function just for umount. In this case, we
2577 * need to resolve the path without doing any revalidation.
2579 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2580 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2581 * in almost all cases, this lookup will be served out of the dcache. The only
2582 * cases where it won't are if nd->last refers to a symlink or the path is
2583 * bogus and it doesn't exist.
2586 * -error: if there was an error during lookup. This includes -ENOENT if the
2587 * lookup found a negative dentry.
2589 * 0: if we successfully resolved nd->last and found it to not to be a
2590 * symlink that needs to be followed.
2592 * 1: if we successfully resolved nd->last and found it to be a symlink
2593 * that needs to be followed.
2596 mountpoint_last(struct nameidata *nd)
2599 struct dentry *dir = nd->path.dentry;
2602 /* If we're in rcuwalk, drop out of it to handle last component */
2603 if (nd->flags & LOOKUP_RCU) {
2604 if (unlazy_walk(nd))
2608 nd->flags &= ~LOOKUP_PARENT;
2610 if (unlikely(nd->last_type != LAST_NORM)) {
2611 error = handle_dots(nd, nd->last_type);
2614 path.dentry = dget(nd->path.dentry);
2616 path.dentry = d_lookup(dir, &nd->last);
2619 * No cached dentry. Mounted dentries are pinned in the
2620 * cache, so that means that this dentry is probably
2621 * a symlink or the path doesn't actually point
2622 * to a mounted dentry.
2624 path.dentry = lookup_slow(&nd->last, dir,
2625 nd->flags | LOOKUP_NO_REVAL);
2626 if (IS_ERR(path.dentry))
2627 return PTR_ERR(path.dentry);
2630 if (d_is_negative(path.dentry)) {
2634 path.mnt = nd->path.mnt;
2635 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2639 * path_mountpoint - look up a path to be umounted
2640 * @nd: lookup context
2641 * @flags: lookup flags
2642 * @path: pointer to container for result
2644 * Look up the given name, but don't attempt to revalidate the last component.
2645 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2648 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2650 const char *s = path_init(nd, flags);
2653 while (!(err = link_path_walk(s, nd)) &&
2654 (err = mountpoint_last(nd)) > 0) {
2655 s = trailing_symlink(nd);
2659 nd->path.mnt = NULL;
2660 nd->path.dentry = NULL;
2668 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2671 struct nameidata nd;
2674 return PTR_ERR(name);
2675 set_nameidata(&nd, dfd, name);
2676 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2677 if (unlikely(error == -ECHILD))
2678 error = path_mountpoint(&nd, flags, path);
2679 if (unlikely(error == -ESTALE))
2680 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2682 audit_inode(name, path->dentry, 0);
2683 restore_nameidata();
2689 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2690 * @dfd: directory file descriptor
2691 * @name: pathname from userland
2692 * @flags: lookup flags
2693 * @path: pointer to container to hold result
2695 * A umount is a special case for path walking. We're not actually interested
2696 * in the inode in this situation, and ESTALE errors can be a problem. We
2697 * simply want track down the dentry and vfsmount attached at the mountpoint
2698 * and avoid revalidating the last component.
2700 * Returns 0 and populates "path" on success.
2703 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2706 return filename_mountpoint(dfd, getname(name), path, flags);
2710 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2713 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2715 EXPORT_SYMBOL(kern_path_mountpoint);
2717 int __check_sticky(struct inode *dir, struct inode *inode)
2719 kuid_t fsuid = current_fsuid();
2721 if (uid_eq(inode->i_uid, fsuid))
2723 if (uid_eq(dir->i_uid, fsuid))
2725 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2727 EXPORT_SYMBOL(__check_sticky);
2730 * Check whether we can remove a link victim from directory dir, check
2731 * whether the type of victim is right.
2732 * 1. We can't do it if dir is read-only (done in permission())
2733 * 2. We should have write and exec permissions on dir
2734 * 3. We can't remove anything from append-only dir
2735 * 4. We can't do anything with immutable dir (done in permission())
2736 * 5. If the sticky bit on dir is set we should either
2737 * a. be owner of dir, or
2738 * b. be owner of victim, or
2739 * c. have CAP_FOWNER capability
2740 * 6. If the victim is append-only or immutable we can't do antyhing with
2741 * links pointing to it.
2742 * 7. If the victim has an unknown uid or gid we can't change the inode.
2743 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2744 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2745 * 10. We can't remove a root or mountpoint.
2746 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2747 * nfs_async_unlink().
2749 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2751 struct inode *inode = d_backing_inode(victim);
2754 if (d_is_negative(victim))
2758 BUG_ON(victim->d_parent->d_inode != dir);
2760 /* Inode writeback is not safe when the uid or gid are invalid. */
2761 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2764 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2766 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2772 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2773 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2776 if (!d_is_dir(victim))
2778 if (IS_ROOT(victim))
2780 } else if (d_is_dir(victim))
2782 if (IS_DEADDIR(dir))
2784 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2789 /* Check whether we can create an object with dentry child in directory
2791 * 1. We can't do it if child already exists (open has special treatment for
2792 * this case, but since we are inlined it's OK)
2793 * 2. We can't do it if dir is read-only (done in permission())
2794 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2795 * 4. We should have write and exec permissions on dir
2796 * 5. We can't do it if dir is immutable (done in permission())
2798 static inline int may_create(struct inode *dir, struct dentry *child)
2800 struct user_namespace *s_user_ns;
2801 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2804 if (IS_DEADDIR(dir))
2806 s_user_ns = dir->i_sb->s_user_ns;
2807 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2808 !kgid_has_mapping(s_user_ns, current_fsgid()))
2810 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2814 * p1 and p2 should be directories on the same fs.
2816 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2821 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2825 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2827 p = d_ancestor(p2, p1);
2829 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2830 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2834 p = d_ancestor(p1, p2);
2836 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2837 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2841 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2842 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2845 EXPORT_SYMBOL(lock_rename);
2847 void unlock_rename(struct dentry *p1, struct dentry *p2)
2849 inode_unlock(p1->d_inode);
2851 inode_unlock(p2->d_inode);
2852 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2855 EXPORT_SYMBOL(unlock_rename);
2857 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2860 int error = may_create(dir, dentry);
2864 if (!dir->i_op->create)
2865 return -EACCES; /* shouldn't it be ENOSYS? */
2868 error = security_inode_create(dir, dentry, mode);
2871 error = dir->i_op->create(dir, dentry, mode, want_excl);
2873 fsnotify_create(dir, dentry);
2876 EXPORT_SYMBOL(vfs_create);
2878 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2879 int (*f)(struct dentry *, umode_t, void *),
2882 struct inode *dir = dentry->d_parent->d_inode;
2883 int error = may_create(dir, dentry);
2889 error = security_inode_create(dir, dentry, mode);
2892 error = f(dentry, mode, arg);
2894 fsnotify_create(dir, dentry);
2897 EXPORT_SYMBOL(vfs_mkobj);
2899 bool may_open_dev(const struct path *path)
2901 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2902 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2905 static int may_open(const struct path *path, int acc_mode, int flag)
2907 struct dentry *dentry = path->dentry;
2908 struct inode *inode = dentry->d_inode;
2914 switch (inode->i_mode & S_IFMT) {
2918 if (acc_mode & MAY_WRITE)
2923 if (!may_open_dev(path))
2932 error = inode_permission(inode, MAY_OPEN | acc_mode);
2937 * An append-only file must be opened in append mode for writing.
2939 if (IS_APPEND(inode)) {
2940 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2946 /* O_NOATIME can only be set by the owner or superuser */
2947 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2953 static int handle_truncate(struct file *filp)
2955 const struct path *path = &filp->f_path;
2956 struct inode *inode = path->dentry->d_inode;
2957 int error = get_write_access(inode);
2961 * Refuse to truncate files with mandatory locks held on them.
2963 error = locks_verify_locked(filp);
2965 error = security_path_truncate(path);
2967 error = do_truncate(path->dentry, 0,
2968 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2971 put_write_access(inode);
2975 static inline int open_to_namei_flags(int flag)
2977 if ((flag & O_ACCMODE) == 3)
2982 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2984 struct user_namespace *s_user_ns;
2985 int error = security_path_mknod(dir, dentry, mode, 0);
2989 s_user_ns = dir->dentry->d_sb->s_user_ns;
2990 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2991 !kgid_has_mapping(s_user_ns, current_fsgid()))
2994 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2998 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3002 * Attempt to atomically look up, create and open a file from a negative
3005 * Returns 0 if successful. The file will have been created and attached to
3006 * @file by the filesystem calling finish_open().
3008 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3009 * be set. The caller will need to perform the open themselves. @path will
3010 * have been updated to point to the new dentry. This may be negative.
3012 * Returns an error code otherwise.
3014 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3015 struct path *path, struct file *file,
3016 const struct open_flags *op,
3017 int open_flag, umode_t mode)
3019 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3020 struct inode *dir = nd->path.dentry->d_inode;
3023 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3024 open_flag &= ~O_TRUNC;
3026 if (nd->flags & LOOKUP_DIRECTORY)
3027 open_flag |= O_DIRECTORY;
3029 file->f_path.dentry = DENTRY_NOT_SET;
3030 file->f_path.mnt = nd->path.mnt;
3031 error = dir->i_op->atomic_open(dir, dentry, file,
3032 open_to_namei_flags(open_flag), mode);
3033 d_lookup_done(dentry);
3035 if (file->f_mode & FMODE_OPENED) {
3037 * We didn't have the inode before the open, so check open
3040 int acc_mode = op->acc_mode;
3041 if (file->f_mode & FMODE_CREATED) {
3042 WARN_ON(!(open_flag & O_CREAT));
3043 fsnotify_create(dir, dentry);
3046 error = may_open(&file->f_path, acc_mode, open_flag);
3047 if (WARN_ON(error > 0))
3049 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3052 if (file->f_path.dentry) {
3054 dentry = file->f_path.dentry;
3056 if (file->f_mode & FMODE_CREATED)
3057 fsnotify_create(dir, dentry);
3058 if (unlikely(d_is_negative(dentry))) {
3061 path->dentry = dentry;
3062 path->mnt = nd->path.mnt;
3072 * Look up and maybe create and open the last component.
3074 * Must be called with parent locked (exclusive in O_CREAT case).
3076 * Returns 0 on success, that is, if
3077 * the file was successfully atomically created (if necessary) and opened, or
3078 * the file was not completely opened at this time, though lookups and
3079 * creations were performed.
3080 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3081 * In the latter case dentry returned in @path might be negative if O_CREAT
3082 * hadn't been specified.
3084 * An error code is returned on failure.
3086 static int lookup_open(struct nameidata *nd, struct path *path,
3088 const struct open_flags *op,
3091 struct dentry *dir = nd->path.dentry;
3092 struct inode *dir_inode = dir->d_inode;
3093 int open_flag = op->open_flag;
3094 struct dentry *dentry;
3095 int error, create_error = 0;
3096 umode_t mode = op->mode;
3097 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3099 if (unlikely(IS_DEADDIR(dir_inode)))
3102 file->f_mode &= ~FMODE_CREATED;
3103 dentry = d_lookup(dir, &nd->last);
3106 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3108 return PTR_ERR(dentry);
3110 if (d_in_lookup(dentry))
3113 error = d_revalidate(dentry, nd->flags);
3114 if (likely(error > 0))
3118 d_invalidate(dentry);
3122 if (dentry->d_inode) {
3123 /* Cached positive dentry: will open in f_op->open */
3128 * Checking write permission is tricky, bacuse we don't know if we are
3129 * going to actually need it: O_CREAT opens should work as long as the
3130 * file exists. But checking existence breaks atomicity. The trick is
3131 * to check access and if not granted clear O_CREAT from the flags.
3133 * Another problem is returing the "right" error value (e.g. for an
3134 * O_EXCL open we want to return EEXIST not EROFS).
3136 if (open_flag & O_CREAT) {
3137 if (!IS_POSIXACL(dir->d_inode))
3138 mode &= ~current_umask();
3139 if (unlikely(!got_write)) {
3140 create_error = -EROFS;
3141 open_flag &= ~O_CREAT;
3142 if (open_flag & (O_EXCL | O_TRUNC))
3144 /* No side effects, safe to clear O_CREAT */
3146 create_error = may_o_create(&nd->path, dentry, mode);
3148 open_flag &= ~O_CREAT;
3149 if (open_flag & O_EXCL)
3153 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3154 unlikely(!got_write)) {
3156 * No O_CREATE -> atomicity not a requirement -> fall
3157 * back to lookup + open
3162 if (dir_inode->i_op->atomic_open) {
3163 error = atomic_open(nd, dentry, path, file, op, open_flag,
3165 if (unlikely(error == -ENOENT) && create_error)
3166 error = create_error;
3171 if (d_in_lookup(dentry)) {
3172 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3174 d_lookup_done(dentry);
3175 if (unlikely(res)) {
3177 error = PTR_ERR(res);
3185 /* Negative dentry, just create the file */
3186 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3187 file->f_mode |= FMODE_CREATED;
3188 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3189 if (!dir_inode->i_op->create) {
3193 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3194 open_flag & O_EXCL);
3197 fsnotify_create(dir_inode, dentry);
3199 if (unlikely(create_error) && !dentry->d_inode) {
3200 error = create_error;
3204 path->dentry = dentry;
3205 path->mnt = nd->path.mnt;
3214 * Handle the last step of open()
3216 static int do_last(struct nameidata *nd,
3217 struct file *file, const struct open_flags *op)
3219 struct dentry *dir = nd->path.dentry;
3220 int open_flag = op->open_flag;
3221 bool will_truncate = (open_flag & O_TRUNC) != 0;
3222 bool got_write = false;
3223 int acc_mode = op->acc_mode;
3225 struct inode *inode;
3229 nd->flags &= ~LOOKUP_PARENT;
3230 nd->flags |= op->intent;
3232 if (nd->last_type != LAST_NORM) {
3233 error = handle_dots(nd, nd->last_type);
3234 if (unlikely(error))
3239 if (!(open_flag & O_CREAT)) {
3240 if (nd->last.name[nd->last.len])
3241 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3242 /* we _can_ be in RCU mode here */
3243 error = lookup_fast(nd, &path, &inode, &seq);
3244 if (likely(error > 0))
3250 BUG_ON(nd->inode != dir->d_inode);
3251 BUG_ON(nd->flags & LOOKUP_RCU);
3253 /* create side of things */
3255 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3256 * has been cleared when we got to the last component we are
3259 error = complete_walk(nd);
3263 audit_inode(nd->name, dir, LOOKUP_PARENT);
3264 /* trailing slashes? */
3265 if (unlikely(nd->last.name[nd->last.len]))
3269 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3270 error = mnt_want_write(nd->path.mnt);
3274 * do _not_ fail yet - we might not need that or fail with
3275 * a different error; let lookup_open() decide; we'll be
3276 * dropping this one anyway.
3279 if (open_flag & O_CREAT)
3280 inode_lock(dir->d_inode);
3282 inode_lock_shared(dir->d_inode);
3283 error = lookup_open(nd, &path, file, op, got_write);
3284 if (open_flag & O_CREAT)
3285 inode_unlock(dir->d_inode);
3287 inode_unlock_shared(dir->d_inode);
3292 if (file->f_mode & FMODE_OPENED) {
3293 if ((file->f_mode & FMODE_CREATED) ||
3294 !S_ISREG(file_inode(file)->i_mode))
3295 will_truncate = false;
3297 audit_inode(nd->name, file->f_path.dentry, 0);
3301 if (file->f_mode & FMODE_CREATED) {
3302 /* Don't check for write permission, don't truncate */
3303 open_flag &= ~O_TRUNC;
3304 will_truncate = false;
3306 path_to_nameidata(&path, nd);
3307 goto finish_open_created;
3311 * If atomic_open() acquired write access it is dropped now due to
3312 * possible mount and symlink following (this might be optimized away if
3316 mnt_drop_write(nd->path.mnt);
3320 error = follow_managed(&path, nd);
3321 if (unlikely(error < 0))
3324 if (unlikely(d_is_negative(path.dentry))) {
3325 path_to_nameidata(&path, nd);
3330 * create/update audit record if it already exists.
3332 audit_inode(nd->name, path.dentry, 0);
3334 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3335 path_to_nameidata(&path, nd);
3339 seq = 0; /* out of RCU mode, so the value doesn't matter */
3340 inode = d_backing_inode(path.dentry);
3342 error = step_into(nd, &path, 0, inode, seq);
3343 if (unlikely(error))
3346 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3347 error = complete_walk(nd);
3350 audit_inode(nd->name, nd->path.dentry, 0);
3352 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3355 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3357 if (!d_is_reg(nd->path.dentry))
3358 will_truncate = false;
3360 if (will_truncate) {
3361 error = mnt_want_write(nd->path.mnt);
3366 finish_open_created:
3367 error = may_open(&nd->path, acc_mode, open_flag);
3370 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
3371 error = vfs_open(&nd->path, file);
3375 error = ima_file_check(file, op->acc_mode);
3376 if (!error && will_truncate)
3377 error = handle_truncate(file);
3379 if (unlikely(error > 0)) {
3384 mnt_drop_write(nd->path.mnt);
3388 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3390 struct dentry *child = NULL;
3391 struct inode *dir = dentry->d_inode;
3392 struct inode *inode;
3395 /* we want directory to be writable */
3396 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3399 error = -EOPNOTSUPP;
3400 if (!dir->i_op->tmpfile)
3403 child = d_alloc(dentry, &slash_name);
3404 if (unlikely(!child))
3406 error = dir->i_op->tmpfile(dir, child, mode);
3410 inode = child->d_inode;
3411 if (unlikely(!inode))
3413 if (!(open_flag & O_EXCL)) {
3414 spin_lock(&inode->i_lock);
3415 inode->i_state |= I_LINKABLE;
3416 spin_unlock(&inode->i_lock);
3422 return ERR_PTR(error);
3424 EXPORT_SYMBOL(vfs_tmpfile);
3426 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3427 const struct open_flags *op,
3430 struct dentry *child;
3432 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3433 if (unlikely(error))
3435 error = mnt_want_write(path.mnt);
3436 if (unlikely(error))
3438 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3439 error = PTR_ERR(child);
3443 path.dentry = child;
3444 audit_inode(nd->name, child, 0);
3445 /* Don't check for other permissions, the inode was just created */
3446 error = may_open(&path, 0, op->open_flag);
3449 file->f_path.mnt = path.mnt;
3450 error = finish_open(file, child, NULL);
3452 mnt_drop_write(path.mnt);
3458 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3461 int error = path_lookupat(nd, flags, &path);
3463 audit_inode(nd->name, path.dentry, 0);
3464 error = vfs_open(&path, file);
3470 static struct file *path_openat(struct nameidata *nd,
3471 const struct open_flags *op, unsigned flags)
3476 file = alloc_empty_file(op->open_flag, current_cred());
3480 if (unlikely(file->f_flags & __O_TMPFILE)) {
3481 error = do_tmpfile(nd, flags, op, file);
3482 } else if (unlikely(file->f_flags & O_PATH)) {
3483 error = do_o_path(nd, flags, file);
3485 const char *s = path_init(nd, flags);
3486 while (!(error = link_path_walk(s, nd)) &&
3487 (error = do_last(nd, file, op)) > 0) {
3488 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3489 s = trailing_symlink(nd);
3493 if (likely(!error)) {
3494 if (likely(file->f_mode & FMODE_OPENED))
3500 if (error == -EOPENSTALE) {
3501 if (flags & LOOKUP_RCU)
3506 return ERR_PTR(error);
3509 struct file *do_filp_open(int dfd, struct filename *pathname,
3510 const struct open_flags *op)
3512 struct nameidata nd;
3513 int flags = op->lookup_flags;
3516 set_nameidata(&nd, dfd, pathname);
3517 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3518 if (unlikely(filp == ERR_PTR(-ECHILD)))
3519 filp = path_openat(&nd, op, flags);
3520 if (unlikely(filp == ERR_PTR(-ESTALE)))
3521 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3522 restore_nameidata();
3526 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3527 const char *name, const struct open_flags *op)
3529 struct nameidata nd;
3531 struct filename *filename;
3532 int flags = op->lookup_flags | LOOKUP_ROOT;
3535 nd.root.dentry = dentry;
3537 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3538 return ERR_PTR(-ELOOP);
3540 filename = getname_kernel(name);
3541 if (IS_ERR(filename))
3542 return ERR_CAST(filename);
3544 set_nameidata(&nd, -1, filename);
3545 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3546 if (unlikely(file == ERR_PTR(-ECHILD)))
3547 file = path_openat(&nd, op, flags);
3548 if (unlikely(file == ERR_PTR(-ESTALE)))
3549 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3550 restore_nameidata();
3555 static struct dentry *filename_create(int dfd, struct filename *name,
3556 struct path *path, unsigned int lookup_flags)
3558 struct dentry *dentry = ERR_PTR(-EEXIST);
3563 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3566 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3567 * other flags passed in are ignored!
3569 lookup_flags &= LOOKUP_REVAL;
3571 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3573 return ERR_CAST(name);
3576 * Yucky last component or no last component at all?
3577 * (foo/., foo/.., /////)
3579 if (unlikely(type != LAST_NORM))
3582 /* don't fail immediately if it's r/o, at least try to report other errors */
3583 err2 = mnt_want_write(path->mnt);
3585 * Do the final lookup.
3587 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3588 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3589 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3594 if (d_is_positive(dentry))
3598 * Special case - lookup gave negative, but... we had foo/bar/
3599 * From the vfs_mknod() POV we just have a negative dentry -
3600 * all is fine. Let's be bastards - you had / on the end, you've
3601 * been asking for (non-existent) directory. -ENOENT for you.
3603 if (unlikely(!is_dir && last.name[last.len])) {
3607 if (unlikely(err2)) {
3615 dentry = ERR_PTR(error);
3617 inode_unlock(path->dentry->d_inode);
3619 mnt_drop_write(path->mnt);
3626 struct dentry *kern_path_create(int dfd, const char *pathname,
3627 struct path *path, unsigned int lookup_flags)
3629 return filename_create(dfd, getname_kernel(pathname),
3630 path, lookup_flags);
3632 EXPORT_SYMBOL(kern_path_create);
3634 void done_path_create(struct path *path, struct dentry *dentry)
3637 inode_unlock(path->dentry->d_inode);
3638 mnt_drop_write(path->mnt);
3641 EXPORT_SYMBOL(done_path_create);
3643 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3644 struct path *path, unsigned int lookup_flags)
3646 return filename_create(dfd, getname(pathname), path, lookup_flags);
3648 EXPORT_SYMBOL(user_path_create);
3650 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3652 int error = may_create(dir, dentry);
3657 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
3658 !ns_capable(dentry->d_sb->s_user_ns, CAP_MKNOD))
3661 if (!dir->i_op->mknod)
3664 error = devcgroup_inode_mknod(mode, dev);
3668 error = security_inode_mknod(dir, dentry, mode, dev);
3672 error = dir->i_op->mknod(dir, dentry, mode, dev);
3674 fsnotify_create(dir, dentry);
3677 EXPORT_SYMBOL(vfs_mknod);
3679 static int may_mknod(umode_t mode)
3681 switch (mode & S_IFMT) {
3687 case 0: /* zero mode translates to S_IFREG */
3696 long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3699 struct dentry *dentry;
3702 unsigned int lookup_flags = 0;
3704 error = may_mknod(mode);
3708 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3710 return PTR_ERR(dentry);
3712 if (!IS_POSIXACL(path.dentry->d_inode))
3713 mode &= ~current_umask();
3714 error = security_path_mknod(&path, dentry, mode, dev);
3717 switch (mode & S_IFMT) {
3718 case 0: case S_IFREG:
3719 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3721 ima_post_path_mknod(dentry);
3723 case S_IFCHR: case S_IFBLK:
3724 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3725 new_decode_dev(dev));
3727 case S_IFIFO: case S_IFSOCK:
3728 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3732 done_path_create(&path, dentry);
3733 if (retry_estale(error, lookup_flags)) {
3734 lookup_flags |= LOOKUP_REVAL;
3740 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3743 return do_mknodat(dfd, filename, mode, dev);
3746 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3748 return do_mknodat(AT_FDCWD, filename, mode, dev);
3751 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3753 int error = may_create(dir, dentry);
3754 unsigned max_links = dir->i_sb->s_max_links;
3759 if (!dir->i_op->mkdir)
3762 mode &= (S_IRWXUGO|S_ISVTX);
3763 error = security_inode_mkdir(dir, dentry, mode);
3767 if (max_links && dir->i_nlink >= max_links)
3770 error = dir->i_op->mkdir(dir, dentry, mode);
3772 fsnotify_mkdir(dir, dentry);
3775 EXPORT_SYMBOL(vfs_mkdir);
3777 long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3779 struct dentry *dentry;
3782 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3785 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3787 return PTR_ERR(dentry);
3789 if (!IS_POSIXACL(path.dentry->d_inode))
3790 mode &= ~current_umask();
3791 error = security_path_mkdir(&path, dentry, mode);
3793 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3794 done_path_create(&path, dentry);
3795 if (retry_estale(error, lookup_flags)) {
3796 lookup_flags |= LOOKUP_REVAL;
3802 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3804 return do_mkdirat(dfd, pathname, mode);
3807 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3809 return do_mkdirat(AT_FDCWD, pathname, mode);
3812 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3814 int error = may_delete(dir, dentry, 1);
3819 if (!dir->i_op->rmdir)
3823 inode_lock(dentry->d_inode);
3826 if (is_local_mountpoint(dentry))
3829 error = security_inode_rmdir(dir, dentry);
3833 error = dir->i_op->rmdir(dir, dentry);
3837 shrink_dcache_parent(dentry);
3838 dentry->d_inode->i_flags |= S_DEAD;
3840 detach_mounts(dentry);
3843 inode_unlock(dentry->d_inode);
3849 EXPORT_SYMBOL(vfs_rmdir);
3851 long do_rmdir(int dfd, const char __user *pathname)
3854 struct filename *name;
3855 struct dentry *dentry;
3859 unsigned int lookup_flags = 0;
3861 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3862 &path, &last, &type);
3864 return PTR_ERR(name);
3878 error = mnt_want_write(path.mnt);
3882 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3883 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3884 error = PTR_ERR(dentry);
3887 if (!dentry->d_inode) {
3891 error = security_path_rmdir(&path, dentry);
3894 error = vfs_rmdir(path.dentry->d_inode, dentry);
3898 inode_unlock(path.dentry->d_inode);
3899 mnt_drop_write(path.mnt);
3903 if (retry_estale(error, lookup_flags)) {
3904 lookup_flags |= LOOKUP_REVAL;
3910 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3912 return do_rmdir(AT_FDCWD, pathname);
3916 * vfs_unlink - unlink a filesystem object
3917 * @dir: parent directory
3919 * @delegated_inode: returns victim inode, if the inode is delegated.
3921 * The caller must hold dir->i_mutex.
3923 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3924 * return a reference to the inode in delegated_inode. The caller
3925 * should then break the delegation on that inode and retry. Because
3926 * breaking a delegation may take a long time, the caller should drop
3927 * dir->i_mutex before doing so.
3929 * Alternatively, a caller may pass NULL for delegated_inode. This may
3930 * be appropriate for callers that expect the underlying filesystem not
3931 * to be NFS exported.
3933 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3935 struct inode *target = dentry->d_inode;
3936 int error = may_delete(dir, dentry, 0);
3941 if (!dir->i_op->unlink)
3945 if (is_local_mountpoint(dentry))
3948 error = security_inode_unlink(dir, dentry);
3950 error = try_break_deleg(target, delegated_inode);
3953 error = dir->i_op->unlink(dir, dentry);
3956 detach_mounts(dentry);
3961 inode_unlock(target);
3963 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3964 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3965 fsnotify_link_count(target);
3971 EXPORT_SYMBOL(vfs_unlink);
3974 * Make sure that the actual truncation of the file will occur outside its
3975 * directory's i_mutex. Truncate can take a long time if there is a lot of
3976 * writeout happening, and we don't want to prevent access to the directory
3977 * while waiting on the I/O.
3979 long do_unlinkat(int dfd, struct filename *name)
3982 struct dentry *dentry;
3986 struct inode *inode = NULL;
3987 struct inode *delegated_inode = NULL;
3988 unsigned int lookup_flags = 0;
3990 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
3992 return PTR_ERR(name);
3995 if (type != LAST_NORM)
3998 error = mnt_want_write(path.mnt);
4002 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4003 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4004 error = PTR_ERR(dentry);
4005 if (!IS_ERR(dentry)) {
4006 /* Why not before? Because we want correct error value */
4007 if (last.name[last.len])
4009 inode = dentry->d_inode;
4010 if (d_is_negative(dentry))
4013 error = security_path_unlink(&path, dentry);
4016 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4020 inode_unlock(path.dentry->d_inode);
4022 iput(inode); /* truncate the inode here */
4024 if (delegated_inode) {
4025 error = break_deleg_wait(&delegated_inode);
4029 mnt_drop_write(path.mnt);
4032 if (retry_estale(error, lookup_flags)) {
4033 lookup_flags |= LOOKUP_REVAL;
4041 if (d_is_negative(dentry))
4043 else if (d_is_dir(dentry))
4050 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4052 if ((flag & ~AT_REMOVEDIR) != 0)
4055 if (flag & AT_REMOVEDIR)
4056 return do_rmdir(dfd, pathname);
4058 return do_unlinkat(dfd, getname(pathname));
4061 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4063 return do_unlinkat(AT_FDCWD, getname(pathname));
4066 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4068 int error = may_create(dir, dentry);
4073 if (!dir->i_op->symlink)
4076 error = security_inode_symlink(dir, dentry, oldname);
4080 error = dir->i_op->symlink(dir, dentry, oldname);
4082 fsnotify_create(dir, dentry);
4085 EXPORT_SYMBOL(vfs_symlink);
4087 long do_symlinkat(const char __user *oldname, int newdfd,
4088 const char __user *newname)
4091 struct filename *from;
4092 struct dentry *dentry;
4094 unsigned int lookup_flags = 0;
4096 from = getname(oldname);
4098 return PTR_ERR(from);
4100 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4101 error = PTR_ERR(dentry);
4105 error = security_path_symlink(&path, dentry, from->name);
4107 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4108 done_path_create(&path, dentry);
4109 if (retry_estale(error, lookup_flags)) {
4110 lookup_flags |= LOOKUP_REVAL;
4118 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4119 int, newdfd, const char __user *, newname)
4121 return do_symlinkat(oldname, newdfd, newname);
4124 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4126 return do_symlinkat(oldname, AT_FDCWD, newname);
4130 * vfs_link - create a new link
4131 * @old_dentry: object to be linked
4133 * @new_dentry: where to create the new link
4134 * @delegated_inode: returns inode needing a delegation break
4136 * The caller must hold dir->i_mutex
4138 * If vfs_link discovers a delegation on the to-be-linked file in need
4139 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4140 * inode in delegated_inode. The caller should then break the delegation
4141 * and retry. Because breaking a delegation may take a long time, the
4142 * caller should drop the i_mutex before doing so.
4144 * Alternatively, a caller may pass NULL for delegated_inode. This may
4145 * be appropriate for callers that expect the underlying filesystem not
4146 * to be NFS exported.
4148 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4150 struct inode *inode = old_dentry->d_inode;
4151 unsigned max_links = dir->i_sb->s_max_links;
4157 error = may_create(dir, new_dentry);
4161 if (dir->i_sb != inode->i_sb)
4165 * A link to an append-only or immutable file cannot be created.
4167 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4170 * Updating the link count will likely cause i_uid and i_gid to
4171 * be writen back improperly if their true value is unknown to
4174 if (HAS_UNMAPPED_ID(inode))
4176 if (!dir->i_op->link)
4178 if (S_ISDIR(inode->i_mode))
4181 error = security_inode_link(old_dentry, dir, new_dentry);
4186 /* Make sure we don't allow creating hardlink to an unlinked file */
4187 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4189 else if (max_links && inode->i_nlink >= max_links)
4192 error = try_break_deleg(inode, delegated_inode);
4194 error = dir->i_op->link(old_dentry, dir, new_dentry);
4197 if (!error && (inode->i_state & I_LINKABLE)) {
4198 spin_lock(&inode->i_lock);
4199 inode->i_state &= ~I_LINKABLE;
4200 spin_unlock(&inode->i_lock);
4202 inode_unlock(inode);
4204 fsnotify_link(dir, inode, new_dentry);
4207 EXPORT_SYMBOL(vfs_link);
4210 * Hardlinks are often used in delicate situations. We avoid
4211 * security-related surprises by not following symlinks on the
4214 * We don't follow them on the oldname either to be compatible
4215 * with linux 2.0, and to avoid hard-linking to directories
4216 * and other special files. --ADM
4218 int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4219 const char __user *newname, int flags)
4221 struct dentry *new_dentry;
4222 struct path old_path, new_path;
4223 struct inode *delegated_inode = NULL;
4227 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4230 * To use null names we require CAP_DAC_READ_SEARCH
4231 * This ensures that not everyone will be able to create
4232 * handlink using the passed filedescriptor.
4234 if (flags & AT_EMPTY_PATH) {
4235 if (!capable(CAP_DAC_READ_SEARCH))
4240 if (flags & AT_SYMLINK_FOLLOW)
4241 how |= LOOKUP_FOLLOW;
4243 error = user_path_at(olddfd, oldname, how, &old_path);
4247 new_dentry = user_path_create(newdfd, newname, &new_path,
4248 (how & LOOKUP_REVAL));
4249 error = PTR_ERR(new_dentry);
4250 if (IS_ERR(new_dentry))
4254 if (old_path.mnt != new_path.mnt)
4256 error = may_linkat(&old_path);
4257 if (unlikely(error))
4259 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4262 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4264 done_path_create(&new_path, new_dentry);
4265 if (delegated_inode) {
4266 error = break_deleg_wait(&delegated_inode);
4268 path_put(&old_path);
4272 if (retry_estale(error, how)) {
4273 path_put(&old_path);
4274 how |= LOOKUP_REVAL;
4278 path_put(&old_path);
4283 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4284 int, newdfd, const char __user *, newname, int, flags)
4286 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4289 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4291 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4295 * vfs_rename - rename a filesystem object
4296 * @old_dir: parent of source
4297 * @old_dentry: source
4298 * @new_dir: parent of destination
4299 * @new_dentry: destination
4300 * @delegated_inode: returns an inode needing a delegation break
4301 * @flags: rename flags
4303 * The caller must hold multiple mutexes--see lock_rename()).
4305 * If vfs_rename discovers a delegation in need of breaking at either
4306 * the source or destination, it will return -EWOULDBLOCK and return a
4307 * reference to the inode in delegated_inode. The caller should then
4308 * break the delegation and retry. Because breaking a delegation may
4309 * take a long time, the caller should drop all locks before doing
4312 * Alternatively, a caller may pass NULL for delegated_inode. This may
4313 * be appropriate for callers that expect the underlying filesystem not
4314 * to be NFS exported.
4316 * The worst of all namespace operations - renaming directory. "Perverted"
4317 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4320 * a) we can get into loop creation.
4321 * b) race potential - two innocent renames can create a loop together.
4322 * That's where 4.4 screws up. Current fix: serialization on
4323 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4325 * c) we have to lock _four_ objects - parents and victim (if it exists),
4326 * and source (if it is not a directory).
4327 * And that - after we got ->i_mutex on parents (until then we don't know
4328 * whether the target exists). Solution: try to be smart with locking
4329 * order for inodes. We rely on the fact that tree topology may change
4330 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4331 * move will be locked. Thus we can rank directories by the tree
4332 * (ancestors first) and rank all non-directories after them.
4333 * That works since everybody except rename does "lock parent, lookup,
4334 * lock child" and rename is under ->s_vfs_rename_mutex.
4335 * HOWEVER, it relies on the assumption that any object with ->lookup()
4336 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4337 * we'd better make sure that there's no link(2) for them.
4338 * d) conversion from fhandle to dentry may come in the wrong moment - when
4339 * we are removing the target. Solution: we will have to grab ->i_mutex
4340 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4341 * ->i_mutex on parents, which works but leads to some truly excessive
4344 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4345 struct inode *new_dir, struct dentry *new_dentry,
4346 struct inode **delegated_inode, unsigned int flags)
4349 bool is_dir = d_is_dir(old_dentry);
4350 struct inode *source = old_dentry->d_inode;
4351 struct inode *target = new_dentry->d_inode;
4352 bool new_is_dir = false;
4353 unsigned max_links = new_dir->i_sb->s_max_links;
4354 struct name_snapshot old_name;
4356 if (source == target)
4359 error = may_delete(old_dir, old_dentry, is_dir);
4364 error = may_create(new_dir, new_dentry);
4366 new_is_dir = d_is_dir(new_dentry);
4368 if (!(flags & RENAME_EXCHANGE))
4369 error = may_delete(new_dir, new_dentry, is_dir);
4371 error = may_delete(new_dir, new_dentry, new_is_dir);
4376 if (!old_dir->i_op->rename)
4380 * If we are going to change the parent - check write permissions,
4381 * we'll need to flip '..'.
4383 if (new_dir != old_dir) {
4385 error = inode_permission(source, MAY_WRITE);
4389 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4390 error = inode_permission(target, MAY_WRITE);
4396 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4401 take_dentry_name_snapshot(&old_name, old_dentry);
4403 if (!is_dir || (flags & RENAME_EXCHANGE))
4404 lock_two_nondirectories(source, target);
4409 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4412 if (max_links && new_dir != old_dir) {
4414 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4416 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4417 old_dir->i_nlink >= max_links)
4421 error = try_break_deleg(source, delegated_inode);
4425 if (target && !new_is_dir) {
4426 error = try_break_deleg(target, delegated_inode);
4430 error = old_dir->i_op->rename(old_dir, old_dentry,
4431 new_dir, new_dentry, flags);
4435 if (!(flags & RENAME_EXCHANGE) && target) {
4437 shrink_dcache_parent(new_dentry);
4438 target->i_flags |= S_DEAD;
4440 dont_mount(new_dentry);
4441 detach_mounts(new_dentry);
4443 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4444 if (!(flags & RENAME_EXCHANGE))
4445 d_move(old_dentry, new_dentry);
4447 d_exchange(old_dentry, new_dentry);
4450 if (!is_dir || (flags & RENAME_EXCHANGE))
4451 unlock_two_nondirectories(source, target);
4453 inode_unlock(target);
4456 fsnotify_move(old_dir, new_dir, old_name.name, is_dir,
4457 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4458 if (flags & RENAME_EXCHANGE) {
4459 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4460 new_is_dir, NULL, new_dentry);
4463 release_dentry_name_snapshot(&old_name);
4467 EXPORT_SYMBOL(vfs_rename);
4469 static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4470 const char __user *newname, unsigned int flags)
4472 struct dentry *old_dentry, *new_dentry;
4473 struct dentry *trap;
4474 struct path old_path, new_path;
4475 struct qstr old_last, new_last;
4476 int old_type, new_type;
4477 struct inode *delegated_inode = NULL;
4478 struct filename *from;
4479 struct filename *to;
4480 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4481 bool should_retry = false;
4484 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4487 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4488 (flags & RENAME_EXCHANGE))
4491 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4494 if (flags & RENAME_EXCHANGE)
4498 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4499 &old_path, &old_last, &old_type);
4501 error = PTR_ERR(from);
4505 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4506 &new_path, &new_last, &new_type);
4508 error = PTR_ERR(to);
4513 if (old_path.mnt != new_path.mnt)
4517 if (old_type != LAST_NORM)
4520 if (flags & RENAME_NOREPLACE)
4522 if (new_type != LAST_NORM)
4525 error = mnt_want_write(old_path.mnt);
4530 trap = lock_rename(new_path.dentry, old_path.dentry);
4532 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4533 error = PTR_ERR(old_dentry);
4534 if (IS_ERR(old_dentry))
4536 /* source must exist */
4538 if (d_is_negative(old_dentry))
4540 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4541 error = PTR_ERR(new_dentry);
4542 if (IS_ERR(new_dentry))
4545 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4547 if (flags & RENAME_EXCHANGE) {
4549 if (d_is_negative(new_dentry))
4552 if (!d_is_dir(new_dentry)) {
4554 if (new_last.name[new_last.len])
4558 /* unless the source is a directory trailing slashes give -ENOTDIR */
4559 if (!d_is_dir(old_dentry)) {
4561 if (old_last.name[old_last.len])
4563 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4566 /* source should not be ancestor of target */
4568 if (old_dentry == trap)
4570 /* target should not be an ancestor of source */
4571 if (!(flags & RENAME_EXCHANGE))
4573 if (new_dentry == trap)
4576 error = security_path_rename(&old_path, old_dentry,
4577 &new_path, new_dentry, flags);
4580 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4581 new_path.dentry->d_inode, new_dentry,
4582 &delegated_inode, flags);
4588 unlock_rename(new_path.dentry, old_path.dentry);
4589 if (delegated_inode) {
4590 error = break_deleg_wait(&delegated_inode);
4594 mnt_drop_write(old_path.mnt);
4596 if (retry_estale(error, lookup_flags))
4597 should_retry = true;
4598 path_put(&new_path);
4601 path_put(&old_path);
4604 should_retry = false;
4605 lookup_flags |= LOOKUP_REVAL;
4612 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4613 int, newdfd, const char __user *, newname, unsigned int, flags)
4615 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4618 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4619 int, newdfd, const char __user *, newname)
4621 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
4624 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4626 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4629 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4631 int error = may_create(dir, dentry);
4635 if (!dir->i_op->mknod)
4638 return dir->i_op->mknod(dir, dentry,
4639 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4641 EXPORT_SYMBOL(vfs_whiteout);
4643 int readlink_copy(char __user *buffer, int buflen, const char *link)
4645 int len = PTR_ERR(link);
4650 if (len > (unsigned) buflen)
4652 if (copy_to_user(buffer, link, len))
4659 * vfs_readlink - copy symlink body into userspace buffer
4660 * @dentry: dentry on which to get symbolic link
4661 * @buffer: user memory pointer
4662 * @buflen: size of buffer
4664 * Does not touch atime. That's up to the caller if necessary
4666 * Does not call security hook.
4668 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4670 struct inode *inode = d_inode(dentry);
4671 DEFINE_DELAYED_CALL(done);
4675 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4676 if (unlikely(inode->i_op->readlink))
4677 return inode->i_op->readlink(dentry, buffer, buflen);
4679 if (!d_is_symlink(dentry))
4682 spin_lock(&inode->i_lock);
4683 inode->i_opflags |= IOP_DEFAULT_READLINK;
4684 spin_unlock(&inode->i_lock);
4687 link = inode->i_link;
4689 link = inode->i_op->get_link(dentry, inode, &done);
4691 return PTR_ERR(link);
4693 res = readlink_copy(buffer, buflen, link);
4694 do_delayed_call(&done);
4697 EXPORT_SYMBOL(vfs_readlink);
4700 * vfs_get_link - get symlink body
4701 * @dentry: dentry on which to get symbolic link
4702 * @done: caller needs to free returned data with this
4704 * Calls security hook and i_op->get_link() on the supplied inode.
4706 * It does not touch atime. That's up to the caller if necessary.
4708 * Does not work on "special" symlinks like /proc/$$/fd/N
4710 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4712 const char *res = ERR_PTR(-EINVAL);
4713 struct inode *inode = d_inode(dentry);
4715 if (d_is_symlink(dentry)) {
4716 res = ERR_PTR(security_inode_readlink(dentry));
4718 res = inode->i_op->get_link(dentry, inode, done);
4722 EXPORT_SYMBOL(vfs_get_link);
4724 /* get the link contents into pagecache */
4725 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4726 struct delayed_call *callback)
4730 struct address_space *mapping = inode->i_mapping;
4733 page = find_get_page(mapping, 0);
4735 return ERR_PTR(-ECHILD);
4736 if (!PageUptodate(page)) {
4738 return ERR_PTR(-ECHILD);
4741 page = read_mapping_page(mapping, 0, NULL);
4745 set_delayed_call(callback, page_put_link, page);
4746 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4747 kaddr = page_address(page);
4748 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4752 EXPORT_SYMBOL(page_get_link);
4754 void page_put_link(void *arg)
4758 EXPORT_SYMBOL(page_put_link);
4760 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4762 DEFINE_DELAYED_CALL(done);
4763 int res = readlink_copy(buffer, buflen,
4764 page_get_link(dentry, d_inode(dentry),
4766 do_delayed_call(&done);
4769 EXPORT_SYMBOL(page_readlink);
4772 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4774 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4776 struct address_space *mapping = inode->i_mapping;
4780 unsigned int flags = 0;
4782 flags |= AOP_FLAG_NOFS;
4785 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4786 flags, &page, &fsdata);
4790 memcpy(page_address(page), symname, len-1);
4792 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4799 mark_inode_dirty(inode);
4804 EXPORT_SYMBOL(__page_symlink);
4806 int page_symlink(struct inode *inode, const char *symname, int len)
4808 return __page_symlink(inode, symname, len,
4809 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4811 EXPORT_SYMBOL(page_symlink);
4813 const struct inode_operations page_symlink_inode_operations = {
4814 .get_link = page_get_link,
4816 EXPORT_SYMBOL(page_symlink_inode_operations);