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)
2036 /* At this point we know we have a real path component. */
2041 err = may_lookup(nd);
2045 hash_len = hash_name(nd->path.dentry, name);
2048 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2050 if (name[1] == '.') {
2052 nd->flags |= LOOKUP_JUMPED;
2058 if (likely(type == LAST_NORM)) {
2059 struct dentry *parent = nd->path.dentry;
2060 nd->flags &= ~LOOKUP_JUMPED;
2061 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2062 struct qstr this = { { .hash_len = hash_len }, .name = name };
2063 err = parent->d_op->d_hash(parent, &this);
2066 hash_len = this.hash_len;
2071 nd->last.hash_len = hash_len;
2072 nd->last.name = name;
2073 nd->last_type = type;
2075 name += hashlen_len(hash_len);
2079 * If it wasn't NUL, we know it was '/'. Skip that
2080 * slash, and continue until no more slashes.
2084 } while (unlikely(*name == '/'));
2085 if (unlikely(!*name)) {
2087 /* pathname body, done */
2090 name = nd->stack[nd->depth - 1].name;
2091 /* trailing symlink, done */
2094 /* last component of nested symlink */
2095 err = walk_component(nd, WALK_FOLLOW);
2097 /* not the last component */
2098 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2104 const char *s = get_link(nd);
2113 nd->stack[nd->depth - 1].name = name;
2118 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2119 if (nd->flags & LOOKUP_RCU) {
2120 if (unlazy_walk(nd))
2128 static const char *path_init(struct nameidata *nd, unsigned flags)
2130 const char *s = nd->name->name;
2133 flags &= ~LOOKUP_RCU;
2135 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2136 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2138 if (flags & LOOKUP_ROOT) {
2139 struct dentry *root = nd->root.dentry;
2140 struct inode *inode = root->d_inode;
2141 if (*s && unlikely(!d_can_lookup(root)))
2142 return ERR_PTR(-ENOTDIR);
2143 nd->path = nd->root;
2145 if (flags & LOOKUP_RCU) {
2147 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2148 nd->root_seq = nd->seq;
2149 nd->m_seq = read_seqbegin(&mount_lock);
2151 path_get(&nd->path);
2156 nd->root.mnt = NULL;
2157 nd->path.mnt = NULL;
2158 nd->path.dentry = NULL;
2160 nd->m_seq = read_seqbegin(&mount_lock);
2162 if (flags & LOOKUP_RCU)
2165 if (likely(!nd_jump_root(nd)))
2167 nd->root.mnt = NULL;
2169 return ERR_PTR(-ECHILD);
2170 } else if (nd->dfd == AT_FDCWD) {
2171 if (flags & LOOKUP_RCU) {
2172 struct fs_struct *fs = current->fs;
2178 seq = read_seqcount_begin(&fs->seq);
2180 nd->inode = nd->path.dentry->d_inode;
2181 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2182 } while (read_seqcount_retry(&fs->seq, seq));
2184 get_fs_pwd(current->fs, &nd->path);
2185 nd->inode = nd->path.dentry->d_inode;
2189 /* Caller must check execute permissions on the starting path component */
2190 struct fd f = fdget_raw(nd->dfd);
2191 struct dentry *dentry;
2194 return ERR_PTR(-EBADF);
2196 dentry = f.file->f_path.dentry;
2199 if (!d_can_lookup(dentry)) {
2201 return ERR_PTR(-ENOTDIR);
2205 nd->path = f.file->f_path;
2206 if (flags & LOOKUP_RCU) {
2208 nd->inode = nd->path.dentry->d_inode;
2209 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2211 path_get(&nd->path);
2212 nd->inode = nd->path.dentry->d_inode;
2219 static const char *trailing_symlink(struct nameidata *nd)
2222 int error = may_follow_link(nd);
2223 if (unlikely(error))
2224 return ERR_PTR(error);
2225 nd->flags |= LOOKUP_PARENT;
2226 nd->stack[0].name = NULL;
2231 static inline int lookup_last(struct nameidata *nd)
2233 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2234 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2236 nd->flags &= ~LOOKUP_PARENT;
2237 return walk_component(nd, 0);
2240 static int handle_lookup_down(struct nameidata *nd)
2242 struct path path = nd->path;
2243 struct inode *inode = nd->inode;
2244 unsigned seq = nd->seq;
2247 if (nd->flags & LOOKUP_RCU) {
2249 * don't bother with unlazy_walk on failure - we are
2250 * at the very beginning of walk, so we lose nothing
2251 * if we simply redo everything in non-RCU mode
2253 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2257 err = follow_managed(&path, nd);
2258 if (unlikely(err < 0))
2260 inode = d_backing_inode(path.dentry);
2263 path_to_nameidata(&path, nd);
2269 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2270 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2272 const char *s = path_init(nd, flags);
2278 if (unlikely(flags & LOOKUP_DOWN)) {
2279 err = handle_lookup_down(nd);
2280 if (unlikely(err < 0)) {
2286 while (!(err = link_path_walk(s, nd))
2287 && ((err = lookup_last(nd)) > 0)) {
2288 s = trailing_symlink(nd);
2295 err = complete_walk(nd);
2297 if (!err && nd->flags & LOOKUP_DIRECTORY)
2298 if (!d_can_lookup(nd->path.dentry))
2302 nd->path.mnt = NULL;
2303 nd->path.dentry = NULL;
2309 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2310 struct path *path, struct path *root)
2313 struct nameidata nd;
2315 return PTR_ERR(name);
2316 if (unlikely(root)) {
2318 flags |= LOOKUP_ROOT;
2320 set_nameidata(&nd, dfd, name);
2321 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2322 if (unlikely(retval == -ECHILD))
2323 retval = path_lookupat(&nd, flags, path);
2324 if (unlikely(retval == -ESTALE))
2325 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2327 if (likely(!retval))
2328 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2329 restore_nameidata();
2334 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2335 static int path_parentat(struct nameidata *nd, unsigned flags,
2336 struct path *parent)
2338 const char *s = path_init(nd, flags);
2342 err = link_path_walk(s, nd);
2344 err = complete_walk(nd);
2347 nd->path.mnt = NULL;
2348 nd->path.dentry = NULL;
2354 static struct filename *filename_parentat(int dfd, struct filename *name,
2355 unsigned int flags, struct path *parent,
2356 struct qstr *last, int *type)
2359 struct nameidata nd;
2363 set_nameidata(&nd, dfd, name);
2364 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2365 if (unlikely(retval == -ECHILD))
2366 retval = path_parentat(&nd, flags, parent);
2367 if (unlikely(retval == -ESTALE))
2368 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2369 if (likely(!retval)) {
2371 *type = nd.last_type;
2372 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2375 name = ERR_PTR(retval);
2377 restore_nameidata();
2381 /* does lookup, returns the object with parent locked */
2382 struct dentry *kern_path_locked(const char *name, struct path *path)
2384 struct filename *filename;
2389 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2391 if (IS_ERR(filename))
2392 return ERR_CAST(filename);
2393 if (unlikely(type != LAST_NORM)) {
2396 return ERR_PTR(-EINVAL);
2398 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2399 d = __lookup_hash(&last, path->dentry, 0);
2401 inode_unlock(path->dentry->d_inode);
2408 int kern_path(const char *name, unsigned int flags, struct path *path)
2410 return filename_lookup(AT_FDCWD, getname_kernel(name),
2413 EXPORT_SYMBOL(kern_path);
2416 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2417 * @dentry: pointer to dentry of the base directory
2418 * @mnt: pointer to vfs mount of the base directory
2419 * @name: pointer to file name
2420 * @flags: lookup flags
2421 * @path: pointer to struct path to fill
2423 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2424 const char *name, unsigned int flags,
2427 struct path root = {.mnt = mnt, .dentry = dentry};
2428 /* the first argument of filename_lookup() is ignored with root */
2429 return filename_lookup(AT_FDCWD, getname_kernel(name),
2430 flags , path, &root);
2432 EXPORT_SYMBOL(vfs_path_lookup);
2434 static int lookup_one_len_common(const char *name, struct dentry *base,
2435 int len, struct qstr *this)
2439 this->hash = full_name_hash(base, name, len);
2443 if (unlikely(name[0] == '.')) {
2444 if (len < 2 || (len == 2 && name[1] == '.'))
2449 unsigned int c = *(const unsigned char *)name++;
2450 if (c == '/' || c == '\0')
2454 * See if the low-level filesystem might want
2455 * to use its own hash..
2457 if (base->d_flags & DCACHE_OP_HASH) {
2458 int err = base->d_op->d_hash(base, this);
2463 return inode_permission(base->d_inode, MAY_EXEC);
2467 * try_lookup_one_len - filesystem helper to lookup single pathname component
2468 * @name: pathname component to lookup
2469 * @base: base directory to lookup from
2470 * @len: maximum length @len should be interpreted to
2472 * Look up a dentry by name in the dcache, returning NULL if it does not
2473 * currently exist. The function does not try to create a dentry.
2475 * Note that this routine is purely a helper for filesystem usage and should
2476 * not be called by generic code.
2478 * The caller must hold base->i_mutex.
2480 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2485 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2487 err = lookup_one_len_common(name, base, len, &this);
2489 return ERR_PTR(err);
2491 return lookup_dcache(&this, base, 0);
2493 EXPORT_SYMBOL(try_lookup_one_len);
2496 * lookup_one_len - filesystem helper to lookup single pathname component
2497 * @name: pathname component to lookup
2498 * @base: base directory to lookup from
2499 * @len: maximum length @len should be interpreted to
2501 * Note that this routine is purely a helper for filesystem usage and should
2502 * not be called by generic code.
2504 * The caller must hold base->i_mutex.
2506 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2508 struct dentry *dentry;
2512 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2514 err = lookup_one_len_common(name, base, len, &this);
2516 return ERR_PTR(err);
2518 dentry = lookup_dcache(&this, base, 0);
2519 return dentry ? dentry : __lookup_slow(&this, base, 0);
2521 EXPORT_SYMBOL(lookup_one_len);
2524 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2525 * @name: pathname component to lookup
2526 * @base: base directory to lookup from
2527 * @len: maximum length @len should be interpreted to
2529 * Note that this routine is purely a helper for filesystem usage and should
2530 * not be called by generic code.
2532 * Unlike lookup_one_len, it should be called without the parent
2533 * i_mutex held, and will take the i_mutex itself if necessary.
2535 struct dentry *lookup_one_len_unlocked(const char *name,
2536 struct dentry *base, int len)
2542 err = lookup_one_len_common(name, base, len, &this);
2544 return ERR_PTR(err);
2546 ret = lookup_dcache(&this, base, 0);
2548 ret = lookup_slow(&this, base, 0);
2551 EXPORT_SYMBOL(lookup_one_len_unlocked);
2553 #ifdef CONFIG_UNIX98_PTYS
2554 int path_pts(struct path *path)
2556 /* Find something mounted on "pts" in the same directory as
2559 struct dentry *child, *parent;
2563 ret = path_parent_directory(path);
2567 parent = path->dentry;
2570 child = d_hash_and_lookup(parent, &this);
2574 path->dentry = child;
2581 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2582 struct path *path, int *empty)
2584 return filename_lookup(dfd, getname_flags(name, flags, empty),
2587 EXPORT_SYMBOL(user_path_at_empty);
2590 * mountpoint_last - look up last component for umount
2591 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2593 * This is a special lookup_last function just for umount. In this case, we
2594 * need to resolve the path without doing any revalidation.
2596 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2597 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2598 * in almost all cases, this lookup will be served out of the dcache. The only
2599 * cases where it won't are if nd->last refers to a symlink or the path is
2600 * bogus and it doesn't exist.
2603 * -error: if there was an error during lookup. This includes -ENOENT if the
2604 * lookup found a negative dentry.
2606 * 0: if we successfully resolved nd->last and found it to not to be a
2607 * symlink that needs to be followed.
2609 * 1: if we successfully resolved nd->last and found it to be a symlink
2610 * that needs to be followed.
2613 mountpoint_last(struct nameidata *nd)
2616 struct dentry *dir = nd->path.dentry;
2619 /* If we're in rcuwalk, drop out of it to handle last component */
2620 if (nd->flags & LOOKUP_RCU) {
2621 if (unlazy_walk(nd))
2625 nd->flags &= ~LOOKUP_PARENT;
2627 if (unlikely(nd->last_type != LAST_NORM)) {
2628 error = handle_dots(nd, nd->last_type);
2631 path.dentry = dget(nd->path.dentry);
2633 path.dentry = d_lookup(dir, &nd->last);
2636 * No cached dentry. Mounted dentries are pinned in the
2637 * cache, so that means that this dentry is probably
2638 * a symlink or the path doesn't actually point
2639 * to a mounted dentry.
2641 path.dentry = lookup_slow(&nd->last, dir,
2642 nd->flags | LOOKUP_NO_REVAL);
2643 if (IS_ERR(path.dentry))
2644 return PTR_ERR(path.dentry);
2647 if (d_is_negative(path.dentry)) {
2651 path.mnt = nd->path.mnt;
2652 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2656 * path_mountpoint - look up a path to be umounted
2657 * @nd: lookup context
2658 * @flags: lookup flags
2659 * @path: pointer to container for result
2661 * Look up the given name, but don't attempt to revalidate the last component.
2662 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2665 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2667 const char *s = path_init(nd, flags);
2671 while (!(err = link_path_walk(s, nd)) &&
2672 (err = mountpoint_last(nd)) > 0) {
2673 s = trailing_symlink(nd);
2681 nd->path.mnt = NULL;
2682 nd->path.dentry = NULL;
2690 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2693 struct nameidata nd;
2696 return PTR_ERR(name);
2697 set_nameidata(&nd, dfd, name);
2698 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2699 if (unlikely(error == -ECHILD))
2700 error = path_mountpoint(&nd, flags, path);
2701 if (unlikely(error == -ESTALE))
2702 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2704 audit_inode(name, path->dentry, 0);
2705 restore_nameidata();
2711 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2712 * @dfd: directory file descriptor
2713 * @name: pathname from userland
2714 * @flags: lookup flags
2715 * @path: pointer to container to hold result
2717 * A umount is a special case for path walking. We're not actually interested
2718 * in the inode in this situation, and ESTALE errors can be a problem. We
2719 * simply want track down the dentry and vfsmount attached at the mountpoint
2720 * and avoid revalidating the last component.
2722 * Returns 0 and populates "path" on success.
2725 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2728 return filename_mountpoint(dfd, getname(name), path, flags);
2732 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2735 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2737 EXPORT_SYMBOL(kern_path_mountpoint);
2739 int __check_sticky(struct inode *dir, struct inode *inode)
2741 kuid_t fsuid = current_fsuid();
2743 if (uid_eq(inode->i_uid, fsuid))
2745 if (uid_eq(dir->i_uid, fsuid))
2747 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2749 EXPORT_SYMBOL(__check_sticky);
2752 * Check whether we can remove a link victim from directory dir, check
2753 * whether the type of victim is right.
2754 * 1. We can't do it if dir is read-only (done in permission())
2755 * 2. We should have write and exec permissions on dir
2756 * 3. We can't remove anything from append-only dir
2757 * 4. We can't do anything with immutable dir (done in permission())
2758 * 5. If the sticky bit on dir is set we should either
2759 * a. be owner of dir, or
2760 * b. be owner of victim, or
2761 * c. have CAP_FOWNER capability
2762 * 6. If the victim is append-only or immutable we can't do antyhing with
2763 * links pointing to it.
2764 * 7. If the victim has an unknown uid or gid we can't change the inode.
2765 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2766 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2767 * 10. We can't remove a root or mountpoint.
2768 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2769 * nfs_async_unlink().
2771 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2773 struct inode *inode = d_backing_inode(victim);
2776 if (d_is_negative(victim))
2780 BUG_ON(victim->d_parent->d_inode != dir);
2782 /* Inode writeback is not safe when the uid or gid are invalid. */
2783 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2786 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2788 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2794 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2795 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2798 if (!d_is_dir(victim))
2800 if (IS_ROOT(victim))
2802 } else if (d_is_dir(victim))
2804 if (IS_DEADDIR(dir))
2806 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2811 /* Check whether we can create an object with dentry child in directory
2813 * 1. We can't do it if child already exists (open has special treatment for
2814 * this case, but since we are inlined it's OK)
2815 * 2. We can't do it if dir is read-only (done in permission())
2816 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2817 * 4. We should have write and exec permissions on dir
2818 * 5. We can't do it if dir is immutable (done in permission())
2820 static inline int may_create(struct inode *dir, struct dentry *child)
2822 struct user_namespace *s_user_ns;
2823 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2826 if (IS_DEADDIR(dir))
2828 s_user_ns = dir->i_sb->s_user_ns;
2829 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2830 !kgid_has_mapping(s_user_ns, current_fsgid()))
2832 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2836 * p1 and p2 should be directories on the same fs.
2838 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2843 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2847 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2849 p = d_ancestor(p2, p1);
2851 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2852 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2856 p = d_ancestor(p1, p2);
2858 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2859 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2863 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2864 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2867 EXPORT_SYMBOL(lock_rename);
2869 void unlock_rename(struct dentry *p1, struct dentry *p2)
2871 inode_unlock(p1->d_inode);
2873 inode_unlock(p2->d_inode);
2874 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2877 EXPORT_SYMBOL(unlock_rename);
2879 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2882 int error = may_create(dir, dentry);
2886 if (!dir->i_op->create)
2887 return -EACCES; /* shouldn't it be ENOSYS? */
2890 error = security_inode_create(dir, dentry, mode);
2893 error = dir->i_op->create(dir, dentry, mode, want_excl);
2895 fsnotify_create(dir, dentry);
2898 EXPORT_SYMBOL(vfs_create);
2900 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2901 int (*f)(struct dentry *, umode_t, void *),
2904 struct inode *dir = dentry->d_parent->d_inode;
2905 int error = may_create(dir, dentry);
2911 error = security_inode_create(dir, dentry, mode);
2914 error = f(dentry, mode, arg);
2916 fsnotify_create(dir, dentry);
2919 EXPORT_SYMBOL(vfs_mkobj);
2921 bool may_open_dev(const struct path *path)
2923 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2924 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2927 static int may_open(const struct path *path, int acc_mode, int flag)
2929 struct dentry *dentry = path->dentry;
2930 struct inode *inode = dentry->d_inode;
2936 switch (inode->i_mode & S_IFMT) {
2940 if (acc_mode & MAY_WRITE)
2945 if (!may_open_dev(path))
2954 error = inode_permission(inode, MAY_OPEN | acc_mode);
2959 * An append-only file must be opened in append mode for writing.
2961 if (IS_APPEND(inode)) {
2962 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2968 /* O_NOATIME can only be set by the owner or superuser */
2969 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2975 static int handle_truncate(struct file *filp)
2977 const struct path *path = &filp->f_path;
2978 struct inode *inode = path->dentry->d_inode;
2979 int error = get_write_access(inode);
2983 * Refuse to truncate files with mandatory locks held on them.
2985 error = locks_verify_locked(filp);
2987 error = security_path_truncate(path);
2989 error = do_truncate(path->dentry, 0,
2990 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2993 put_write_access(inode);
2997 static inline int open_to_namei_flags(int flag)
2999 if ((flag & O_ACCMODE) == 3)
3004 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
3006 struct user_namespace *s_user_ns;
3007 int error = security_path_mknod(dir, dentry, mode, 0);
3011 s_user_ns = dir->dentry->d_sb->s_user_ns;
3012 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
3013 !kgid_has_mapping(s_user_ns, current_fsgid()))
3016 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
3020 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3024 * Attempt to atomically look up, create and open a file from a negative
3027 * Returns 0 if successful. The file will have been created and attached to
3028 * @file by the filesystem calling finish_open().
3030 * Returns 1 if the file was looked up only or didn't need creating. The
3031 * caller will need to perform the open themselves. @path will have been
3032 * updated to point to the new dentry. This may be negative.
3034 * Returns an error code otherwise.
3036 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3037 struct path *path, struct file *file,
3038 const struct open_flags *op,
3039 int open_flag, umode_t mode)
3041 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3042 struct inode *dir = nd->path.dentry->d_inode;
3045 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3046 open_flag &= ~O_TRUNC;
3048 if (nd->flags & LOOKUP_DIRECTORY)
3049 open_flag |= O_DIRECTORY;
3051 file->f_path.dentry = DENTRY_NOT_SET;
3052 file->f_path.mnt = nd->path.mnt;
3053 error = dir->i_op->atomic_open(dir, dentry, file,
3054 open_to_namei_flags(open_flag), mode);
3055 d_lookup_done(dentry);
3058 * We didn't have the inode before the open, so check open
3061 int acc_mode = op->acc_mode;
3062 if (file->f_mode & FMODE_CREATED) {
3063 WARN_ON(!(open_flag & O_CREAT));
3064 fsnotify_create(dir, dentry);
3067 error = may_open(&file->f_path, acc_mode, open_flag);
3068 if (WARN_ON(error > 0))
3070 } else if (error > 0) {
3071 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3074 if (file->f_path.dentry) {
3076 dentry = file->f_path.dentry;
3078 if (file->f_mode & FMODE_CREATED)
3079 fsnotify_create(dir, dentry);
3080 if (unlikely(d_is_negative(dentry))) {
3083 path->dentry = dentry;
3084 path->mnt = nd->path.mnt;
3094 * Look up and maybe create and open the last component.
3096 * Must be called with i_mutex held on parent.
3098 * Returns 0 if the file was successfully atomically created (if necessary) and
3099 * opened. In this case the file will be returned attached to @file.
3101 * Returns 1 if the file was not completely opened at this time, though lookups
3102 * and creations will have been performed and the dentry returned in @path will
3103 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3104 * specified then a negative dentry may be returned.
3106 * An error code is returned otherwise.
3108 static int lookup_open(struct nameidata *nd, struct path *path,
3110 const struct open_flags *op,
3113 struct dentry *dir = nd->path.dentry;
3114 struct inode *dir_inode = dir->d_inode;
3115 int open_flag = op->open_flag;
3116 struct dentry *dentry;
3117 int error, create_error = 0;
3118 umode_t mode = op->mode;
3119 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3121 if (unlikely(IS_DEADDIR(dir_inode)))
3124 file->f_mode &= ~FMODE_CREATED;
3125 dentry = d_lookup(dir, &nd->last);
3128 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3130 return PTR_ERR(dentry);
3132 if (d_in_lookup(dentry))
3135 error = d_revalidate(dentry, nd->flags);
3136 if (likely(error > 0))
3140 d_invalidate(dentry);
3144 if (dentry->d_inode) {
3145 /* Cached positive dentry: will open in f_op->open */
3150 * Checking write permission is tricky, bacuse we don't know if we are
3151 * going to actually need it: O_CREAT opens should work as long as the
3152 * file exists. But checking existence breaks atomicity. The trick is
3153 * to check access and if not granted clear O_CREAT from the flags.
3155 * Another problem is returing the "right" error value (e.g. for an
3156 * O_EXCL open we want to return EEXIST not EROFS).
3158 if (open_flag & O_CREAT) {
3159 if (!IS_POSIXACL(dir->d_inode))
3160 mode &= ~current_umask();
3161 if (unlikely(!got_write)) {
3162 create_error = -EROFS;
3163 open_flag &= ~O_CREAT;
3164 if (open_flag & (O_EXCL | O_TRUNC))
3166 /* No side effects, safe to clear O_CREAT */
3168 create_error = may_o_create(&nd->path, dentry, mode);
3170 open_flag &= ~O_CREAT;
3171 if (open_flag & O_EXCL)
3175 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3176 unlikely(!got_write)) {
3178 * No O_CREATE -> atomicity not a requirement -> fall
3179 * back to lookup + open
3184 if (dir_inode->i_op->atomic_open) {
3185 error = atomic_open(nd, dentry, path, file, op, open_flag,
3187 if (unlikely(error == -ENOENT) && create_error)
3188 error = create_error;
3193 if (d_in_lookup(dentry)) {
3194 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3196 d_lookup_done(dentry);
3197 if (unlikely(res)) {
3199 error = PTR_ERR(res);
3207 /* Negative dentry, just create the file */
3208 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3209 file->f_mode |= FMODE_CREATED;
3210 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3211 if (!dir_inode->i_op->create) {
3215 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3216 open_flag & O_EXCL);
3219 fsnotify_create(dir_inode, dentry);
3221 if (unlikely(create_error) && !dentry->d_inode) {
3222 error = create_error;
3226 path->dentry = dentry;
3227 path->mnt = nd->path.mnt;
3236 * Handle the last step of open()
3238 static int do_last(struct nameidata *nd,
3239 struct file *file, const struct open_flags *op)
3241 struct dentry *dir = nd->path.dentry;
3242 int open_flag = op->open_flag;
3243 bool will_truncate = (open_flag & O_TRUNC) != 0;
3244 bool got_write = false;
3245 int acc_mode = op->acc_mode;
3247 struct inode *inode;
3251 nd->flags &= ~LOOKUP_PARENT;
3252 nd->flags |= op->intent;
3254 if (nd->last_type != LAST_NORM) {
3255 error = handle_dots(nd, nd->last_type);
3256 if (unlikely(error))
3261 if (!(open_flag & O_CREAT)) {
3262 if (nd->last.name[nd->last.len])
3263 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3264 /* we _can_ be in RCU mode here */
3265 error = lookup_fast(nd, &path, &inode, &seq);
3266 if (likely(error > 0))
3272 BUG_ON(nd->inode != dir->d_inode);
3273 BUG_ON(nd->flags & LOOKUP_RCU);
3275 /* create side of things */
3277 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3278 * has been cleared when we got to the last component we are
3281 error = complete_walk(nd);
3285 audit_inode(nd->name, dir, LOOKUP_PARENT);
3286 /* trailing slashes? */
3287 if (unlikely(nd->last.name[nd->last.len]))
3291 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3292 error = mnt_want_write(nd->path.mnt);
3296 * do _not_ fail yet - we might not need that or fail with
3297 * a different error; let lookup_open() decide; we'll be
3298 * dropping this one anyway.
3301 if (open_flag & O_CREAT)
3302 inode_lock(dir->d_inode);
3304 inode_lock_shared(dir->d_inode);
3305 error = lookup_open(nd, &path, file, op, got_write);
3306 if (open_flag & O_CREAT)
3307 inode_unlock(dir->d_inode);
3309 inode_unlock_shared(dir->d_inode);
3315 if ((file->f_mode & FMODE_CREATED) ||
3316 !S_ISREG(file_inode(file)->i_mode))
3317 will_truncate = false;
3319 audit_inode(nd->name, file->f_path.dentry, 0);
3323 if (file->f_mode & FMODE_CREATED) {
3324 /* Don't check for write permission, don't truncate */
3325 open_flag &= ~O_TRUNC;
3326 will_truncate = false;
3328 path_to_nameidata(&path, nd);
3329 goto finish_open_created;
3333 * If atomic_open() acquired write access it is dropped now due to
3334 * possible mount and symlink following (this might be optimized away if
3338 mnt_drop_write(nd->path.mnt);
3342 error = follow_managed(&path, nd);
3343 if (unlikely(error < 0))
3346 if (unlikely(d_is_negative(path.dentry))) {
3347 path_to_nameidata(&path, nd);
3352 * create/update audit record if it already exists.
3354 audit_inode(nd->name, path.dentry, 0);
3356 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3357 path_to_nameidata(&path, nd);
3361 seq = 0; /* out of RCU mode, so the value doesn't matter */
3362 inode = d_backing_inode(path.dentry);
3364 error = step_into(nd, &path, 0, inode, seq);
3365 if (unlikely(error))
3368 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3369 error = complete_walk(nd);
3372 audit_inode(nd->name, nd->path.dentry, 0);
3374 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3377 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3379 if (!d_is_reg(nd->path.dentry))
3380 will_truncate = false;
3382 if (will_truncate) {
3383 error = mnt_want_write(nd->path.mnt);
3388 finish_open_created:
3389 error = may_open(&nd->path, acc_mode, open_flag);
3392 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
3393 error = vfs_open(&nd->path, file);
3397 error = ima_file_check(file, op->acc_mode);
3398 if (!error && will_truncate)
3399 error = handle_truncate(file);
3401 if (unlikely(error > 0)) {
3406 mnt_drop_write(nd->path.mnt);
3410 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3412 struct dentry *child = NULL;
3413 struct inode *dir = dentry->d_inode;
3414 struct inode *inode;
3417 /* we want directory to be writable */
3418 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3421 error = -EOPNOTSUPP;
3422 if (!dir->i_op->tmpfile)
3425 child = d_alloc(dentry, &slash_name);
3426 if (unlikely(!child))
3428 error = dir->i_op->tmpfile(dir, child, mode);
3432 inode = child->d_inode;
3433 if (unlikely(!inode))
3435 if (!(open_flag & O_EXCL)) {
3436 spin_lock(&inode->i_lock);
3437 inode->i_state |= I_LINKABLE;
3438 spin_unlock(&inode->i_lock);
3444 return ERR_PTR(error);
3446 EXPORT_SYMBOL(vfs_tmpfile);
3448 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3449 const struct open_flags *op,
3452 struct dentry *child;
3454 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3455 if (unlikely(error))
3457 error = mnt_want_write(path.mnt);
3458 if (unlikely(error))
3460 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3461 error = PTR_ERR(child);
3465 path.dentry = child;
3466 audit_inode(nd->name, child, 0);
3467 /* Don't check for other permissions, the inode was just created */
3468 error = may_open(&path, 0, op->open_flag);
3471 file->f_path.mnt = path.mnt;
3472 error = finish_open(file, child, NULL);
3474 mnt_drop_write(path.mnt);
3480 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3483 int error = path_lookupat(nd, flags, &path);
3485 audit_inode(nd->name, path.dentry, 0);
3486 error = vfs_open(&path, file);
3492 static struct file *path_openat(struct nameidata *nd,
3493 const struct open_flags *op, unsigned flags)
3499 file = alloc_empty_file(op->open_flag, current_cred());
3503 if (unlikely(file->f_flags & __O_TMPFILE)) {
3504 error = do_tmpfile(nd, flags, op, file);
3508 if (unlikely(file->f_flags & O_PATH)) {
3509 error = do_o_path(nd, flags, file);
3513 s = path_init(nd, flags);
3518 while (!(error = link_path_walk(s, nd)) &&
3519 (error = do_last(nd, file, op)) > 0) {
3520 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3521 s = trailing_symlink(nd);
3529 if (likely(!error)) {
3530 if (likely(file->f_mode & FMODE_OPENED))
3536 if (error == -EOPENSTALE) {
3537 if (flags & LOOKUP_RCU)
3542 return ERR_PTR(error);
3545 struct file *do_filp_open(int dfd, struct filename *pathname,
3546 const struct open_flags *op)
3548 struct nameidata nd;
3549 int flags = op->lookup_flags;
3552 set_nameidata(&nd, dfd, pathname);
3553 filp = path_openat(&nd, op, flags | LOOKUP_RCU);