Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/namei.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * Some corrections by tytso.
10 */
11
12/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
13 * lookup logic.
14 */
15/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
16 */
17
18#include <linux/init.h>
19#include <linux/export.h>
20#include <linux/kernel.h>
21#include <linux/slab.h>
22#include <linux/fs.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
43#include "internal.h"
44#include "mount.h"
45
46/* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
51 *
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
58 *
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
62 *
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
65 *
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
72 */
73
74/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
80 * the name is a symlink pointing to a non-existent name.
81 *
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
89 */
90
91/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
93 *
94 * [10-Sep-98 Alan Modra] Another symlink change.
95 */
96
97/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
104 *
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
110 */
111/*
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 * any extra contention...
115 */
116
117/* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
120 *
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
123 */
124
125#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
126
127struct filename *
128getname_flags(const char __user *filename, int flags, int *empty)
129{
130 struct filename *result;
131 char *kname;
132 int len;
133
134 result = audit_reusename(filename);
135 if (result)
136 return result;
137
138 result = __getname();
139 if (unlikely(!result))
140 return ERR_PTR(-ENOMEM);
141
142 /*
143 * First, try to embed the struct filename inside the names_cache
144 * allocation
145 */
146 kname = (char *)result->iname;
147 result->name = kname;
148
149 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 if (unlikely(len < 0)) {
151 __putname(result);
152 return ERR_PTR(len);
153 }
154
155 /*
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
159 * userland.
160 */
161 if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 const size_t size = offsetof(struct filename, iname[1]);
163 kname = (char *)result;
164
165 /*
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
169 */
170 result = kzalloc(size, GFP_KERNEL);
171 if (unlikely(!result)) {
172 __putname(kname);
173 return ERR_PTR(-ENOMEM);
174 }
175 result->name = kname;
176 len = strncpy_from_user(kname, filename, PATH_MAX);
177 if (unlikely(len < 0)) {
178 __putname(kname);
179 kfree(result);
180 return ERR_PTR(len);
181 }
182 if (unlikely(len == PATH_MAX)) {
183 __putname(kname);
184 kfree(result);
185 return ERR_PTR(-ENAMETOOLONG);
186 }
187 }
188
189 result->refcnt = 1;
190 /* The empty path is special. */
191 if (unlikely(!len)) {
192 if (empty)
193 *empty = 1;
194 if (!(flags & LOOKUP_EMPTY)) {
195 putname(result);
196 return ERR_PTR(-ENOENT);
197 }
198 }
199
200 result->uptr = filename;
201 result->aname = NULL;
202 audit_getname(result);
203 return result;
204}
205
206struct filename *
207getname(const char __user * filename)
208{
209 return getname_flags(filename, 0, NULL);
210}
211
212struct filename *
213getname_kernel(const char * filename)
214{
215 struct filename *result;
216 int len = strlen(filename) + 1;
217
218 result = __getname();
219 if (unlikely(!result))
220 return ERR_PTR(-ENOMEM);
221
222 if (len <= EMBEDDED_NAME_MAX) {
223 result->name = (char *)result->iname;
224 } else if (len <= PATH_MAX) {
225 const size_t size = offsetof(struct filename, iname[1]);
226 struct filename *tmp;
227
228 tmp = kmalloc(size, GFP_KERNEL);
229 if (unlikely(!tmp)) {
230 __putname(result);
231 return ERR_PTR(-ENOMEM);
232 }
233 tmp->name = (char *)result;
234 result = tmp;
235 } else {
236 __putname(result);
237 return ERR_PTR(-ENAMETOOLONG);
238 }
239 memcpy((char *)result->name, filename, len);
240 result->uptr = NULL;
241 result->aname = NULL;
242 result->refcnt = 1;
243 audit_getname(result);
244
245 return result;
246}
247
248void putname(struct filename *name)
249{
250 BUG_ON(name->refcnt <= 0);
251
252 if (--name->refcnt > 0)
253 return;
254
255 if (name->name != name->iname) {
256 __putname(name->name);
257 kfree(name);
258 } else
259 __putname(name);
260}
261
262static int check_acl(struct inode *inode, int mask)
263{
264#ifdef CONFIG_FS_POSIX_ACL
265 struct posix_acl *acl;
266
267 if (mask & MAY_NOT_BLOCK) {
268 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
269 if (!acl)
270 return -EAGAIN;
271 /* no ->get_acl() calls in RCU mode... */
272 if (is_uncached_acl(acl))
273 return -ECHILD;
274 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
275 }
276
277 acl = get_acl(inode, ACL_TYPE_ACCESS);
278 if (IS_ERR(acl))
279 return PTR_ERR(acl);
280 if (acl) {
281 int error = posix_acl_permission(inode, acl, mask);
282 posix_acl_release(acl);
283 return error;
284 }
285#endif
286
287 return -EAGAIN;
288}
289
290/*
291 * This does the basic permission checking
292 */
293static int acl_permission_check(struct inode *inode, int mask)
294{
295 unsigned int mode = inode->i_mode;
296
297 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
298 mode >>= 6;
299 else {
300 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
301 int error = check_acl(inode, mask);
302 if (error != -EAGAIN)
303 return error;
304 }
305
306 if (in_group_p(inode->i_gid))
307 mode >>= 3;
308 }
309
310 /*
311 * If the DACs are ok we don't need any capability check.
312 */
313 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
314 return 0;
315 return -EACCES;
316}
317
318/**
319 * generic_permission - check for access rights on a Posix-like filesystem
320 * @inode: inode to check access rights for
321 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
322 *
323 * Used to check for read/write/execute permissions on a file.
324 * We use "fsuid" for this, letting us set arbitrary permissions
325 * for filesystem access without changing the "normal" uids which
326 * are used for other things.
327 *
328 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
329 * request cannot be satisfied (eg. requires blocking or too much complexity).
330 * It would then be called again in ref-walk mode.
331 */
332int generic_permission(struct inode *inode, int mask)
333{
334 int ret;
335
336 /*
337 * Do the basic permission checks.
338 */
339 ret = acl_permission_check(inode, mask);
340 if (ret != -EACCES)
341 return ret;
342
343 if (S_ISDIR(inode->i_mode)) {
344 /* DACs are overridable for directories */
345 if (!(mask & MAY_WRITE))
346 if (capable_wrt_inode_uidgid(inode,
347 CAP_DAC_READ_SEARCH))
348 return 0;
349 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
350 return 0;
351 return -EACCES;
352 }
353
354 /*
355 * Searching includes executable on directories, else just read.
356 */
357 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
358 if (mask == MAY_READ)
359 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
360 return 0;
361 /*
362 * Read/write DACs are always overridable.
363 * Executable DACs are overridable when there is
364 * at least one exec bit set.
365 */
366 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
367 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
368 return 0;
369
370 return -EACCES;
371}
372EXPORT_SYMBOL(generic_permission);
373
374/*
375 * We _really_ want to just do "generic_permission()" without
376 * even looking at the inode->i_op values. So we keep a cache
377 * flag in inode->i_opflags, that says "this has not special
378 * permission function, use the fast case".
379 */
380static inline int do_inode_permission(struct inode *inode, int mask)
381{
382 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
383 if (likely(inode->i_op->permission))
384 return inode->i_op->permission(inode, mask);
385
386 /* This gets set once for the inode lifetime */
387 spin_lock(&inode->i_lock);
388 inode->i_opflags |= IOP_FASTPERM;
389 spin_unlock(&inode->i_lock);
390 }
391 return generic_permission(inode, mask);
392}
393
394/**
395 * sb_permission - Check superblock-level permissions
396 * @sb: Superblock of inode to check permission on
397 * @inode: Inode to check permission on
398 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
399 *
400 * Separate out file-system wide checks from inode-specific permission checks.
401 */
402static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
403{
404 if (unlikely(mask & MAY_WRITE)) {
405 umode_t mode = inode->i_mode;
406
407 /* Nobody gets write access to a read-only fs. */
408 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
409 return -EROFS;
410 }
411 return 0;
412}
413
414/**
415 * inode_permission - Check for access rights to a given inode
416 * @inode: Inode to check permission on
417 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
418 *
419 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
420 * this, letting us set arbitrary permissions for filesystem access without
421 * changing the "normal" UIDs which are used for other things.
422 *
423 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
424 */
425int inode_permission(struct inode *inode, int mask)
426{
427 int retval;
428
429 retval = sb_permission(inode->i_sb, inode, mask);
430 if (retval)
431 return retval;
432
433 if (unlikely(mask & MAY_WRITE)) {
434 /*
435 * Nobody gets write access to an immutable file.
436 */
437 if (IS_IMMUTABLE(inode))
438 return -EPERM;
439
440 /*
441 * Updating mtime will likely cause i_uid and i_gid to be
442 * written back improperly if their true value is unknown
443 * to the vfs.
444 */
445 if (HAS_UNMAPPED_ID(inode))
446 return -EACCES;
447 }
448
449 retval = do_inode_permission(inode, mask);
450 if (retval)
451 return retval;
452
453 retval = devcgroup_inode_permission(inode, mask);
454 if (retval)
455 return retval;
456
457 return security_inode_permission(inode, mask);
458}
459EXPORT_SYMBOL(inode_permission);
460
461/**
462 * path_get - get a reference to a path
463 * @path: path to get the reference to
464 *
465 * Given a path increment the reference count to the dentry and the vfsmount.
466 */
467void path_get(const struct path *path)
468{
469 mntget(path->mnt);
470 dget(path->dentry);
471}
472EXPORT_SYMBOL(path_get);
473
474/**
475 * path_put - put a reference to a path
476 * @path: path to put the reference to
477 *
478 * Given a path decrement the reference count to the dentry and the vfsmount.
479 */
480void path_put(const struct path *path)
481{
482 dput(path->dentry);
483 mntput(path->mnt);
484}
485EXPORT_SYMBOL(path_put);
486
487#define EMBEDDED_LEVELS 2
488struct nameidata {
489 struct path path;
490 struct qstr last;
491 struct path root;
492 struct inode *inode; /* path.dentry.d_inode */
493 unsigned int flags;
494 unsigned seq, m_seq;
495 int last_type;
496 unsigned depth;
497 int total_link_count;
498 struct saved {
499 struct path link;
500 struct delayed_call done;
501 const char *name;
502 unsigned seq;
503 } *stack, internal[EMBEDDED_LEVELS];
504 struct filename *name;
505 struct nameidata *saved;
506 struct inode *link_inode;
507 unsigned root_seq;
508 int dfd;
509} __randomize_layout;
510
511static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
512{
513 struct nameidata *old = current->nameidata;
514 p->stack = p->internal;
515 p->dfd = dfd;
516 p->name = name;
517 p->total_link_count = old ? old->total_link_count : 0;
518 p->saved = old;
519 current->nameidata = p;
520}
521
522static void restore_nameidata(void)
523{
524 struct nameidata *now = current->nameidata, *old = now->saved;
525
526 current->nameidata = old;
527 if (old)
528 old->total_link_count = now->total_link_count;
529 if (now->stack != now->internal)
530 kfree(now->stack);
531}
532
533static int __nd_alloc_stack(struct nameidata *nd)
534{
535 struct saved *p;
536
537 if (nd->flags & LOOKUP_RCU) {
538 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
539 GFP_ATOMIC);
540 if (unlikely(!p))
541 return -ECHILD;
542 } else {
543 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
544 GFP_KERNEL);
545 if (unlikely(!p))
546 return -ENOMEM;
547 }
548 memcpy(p, nd->internal, sizeof(nd->internal));
549 nd->stack = p;
550 return 0;
551}
552
553/**
554 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
555 * @path: nameidate to verify
556 *
557 * Rename can sometimes move a file or directory outside of a bind
558 * mount, path_connected allows those cases to be detected.
559 */
560static bool path_connected(const struct path *path)
561{
562 struct vfsmount *mnt = path->mnt;
563 struct super_block *sb = mnt->mnt_sb;
564
565 /* Bind mounts and multi-root filesystems can have disconnected paths */
566 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
567 return true;
568
569 return is_subdir(path->dentry, mnt->mnt_root);
570}
571
572static inline int nd_alloc_stack(struct nameidata *nd)
573{
574 if (likely(nd->depth != EMBEDDED_LEVELS))
575 return 0;
576 if (likely(nd->stack != nd->internal))
577 return 0;
578 return __nd_alloc_stack(nd);
579}
580
581static void drop_links(struct nameidata *nd)
582{
583 int i = nd->depth;
584 while (i--) {
585 struct saved *last = nd->stack + i;
586 do_delayed_call(&last->done);
587 clear_delayed_call(&last->done);
588 }
589}
590
591static void terminate_walk(struct nameidata *nd)
592{
593 drop_links(nd);
594 if (!(nd->flags & LOOKUP_RCU)) {
595 int i;
596 path_put(&nd->path);
597 for (i = 0; i < nd->depth; i++)
598 path_put(&nd->stack[i].link);
599 if (nd->flags & LOOKUP_ROOT_GRABBED) {
600 path_put(&nd->root);
601 nd->flags &= ~LOOKUP_ROOT_GRABBED;
602 }
603 } else {
604 nd->flags &= ~LOOKUP_RCU;
605 rcu_read_unlock();
606 }
607 nd->depth = 0;
608}
609
610/* path_put is needed afterwards regardless of success or failure */
611static bool legitimize_path(struct nameidata *nd,
612 struct path *path, unsigned seq)
613{
614 int res = __legitimize_mnt(path->mnt, nd->m_seq);
615 if (unlikely(res)) {
616 if (res > 0)
617 path->mnt = NULL;
618 path->dentry = NULL;
619 return false;
620 }
621 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
622 path->dentry = NULL;
623 return false;
624 }
625 return !read_seqcount_retry(&path->dentry->d_seq, seq);
626}
627
628static bool legitimize_links(struct nameidata *nd)
629{
630 int i;
631 for (i = 0; i < nd->depth; i++) {
632 struct saved *last = nd->stack + i;
633 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
634 drop_links(nd);
635 nd->depth = i + 1;
636 return false;
637 }
638 }
639 return true;
640}
641
642static bool legitimize_root(struct nameidata *nd)
643{
644 if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
645 return true;
646 nd->flags |= LOOKUP_ROOT_GRABBED;
647 return legitimize_path(nd, &nd->root, nd->root_seq);
648}
649
650/*
651 * Path walking has 2 modes, rcu-walk and ref-walk (see
652 * Documentation/filesystems/path-lookup.txt). In situations when we can't
653 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
654 * normal reference counts on dentries and vfsmounts to transition to ref-walk
655 * mode. Refcounts are grabbed at the last known good point before rcu-walk
656 * got stuck, so ref-walk may continue from there. If this is not successful
657 * (eg. a seqcount has changed), then failure is returned and it's up to caller
658 * to restart the path walk from the beginning in ref-walk mode.
659 */
660
661/**
662 * unlazy_walk - try to switch to ref-walk mode.
663 * @nd: nameidata pathwalk data
664 * Returns: 0 on success, -ECHILD on failure
665 *
666 * unlazy_walk attempts to legitimize the current nd->path and nd->root
667 * for ref-walk mode.
668 * Must be called from rcu-walk context.
669 * Nothing should touch nameidata between unlazy_walk() failure and
670 * terminate_walk().
671 */
672static int unlazy_walk(struct nameidata *nd)
673{
674 struct dentry *parent = nd->path.dentry;
675
676 BUG_ON(!(nd->flags & LOOKUP_RCU));
677
678 nd->flags &= ~LOOKUP_RCU;
679 if (unlikely(!legitimize_links(nd)))
680 goto out1;
681 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
682 goto out;
683 if (unlikely(!legitimize_root(nd)))
684 goto out;
685 rcu_read_unlock();
686 BUG_ON(nd->inode != parent->d_inode);
687 return 0;
688
689out1:
690 nd->path.mnt = NULL;
691 nd->path.dentry = NULL;
692out:
693 rcu_read_unlock();
694 return -ECHILD;
695}
696
697/**
698 * unlazy_child - try to switch to ref-walk mode.
699 * @nd: nameidata pathwalk data
700 * @dentry: child of nd->path.dentry
701 * @seq: seq number to check dentry against
702 * Returns: 0 on success, -ECHILD on failure
703 *
704 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
705 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
706 * @nd. Must be called from rcu-walk context.
707 * Nothing should touch nameidata between unlazy_child() failure and
708 * terminate_walk().
709 */
710static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
711{
712 BUG_ON(!(nd->flags & LOOKUP_RCU));
713
714 nd->flags &= ~LOOKUP_RCU;
715 if (unlikely(!legitimize_links(nd)))
716 goto out2;
717 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
718 goto out2;
719 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
720 goto out1;
721
722 /*
723 * We need to move both the parent and the dentry from the RCU domain
724 * to be properly refcounted. And the sequence number in the dentry
725 * validates *both* dentry counters, since we checked the sequence
726 * number of the parent after we got the child sequence number. So we
727 * know the parent must still be valid if the child sequence number is
728 */
729 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
730 goto out;
731 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
732 goto out_dput;
733 /*
734 * Sequence counts matched. Now make sure that the root is
735 * still valid and get it if required.
736 */
737 if (unlikely(!legitimize_root(nd)))
738 goto out_dput;
739 rcu_read_unlock();
740 return 0;
741
742out2:
743 nd->path.mnt = NULL;
744out1:
745 nd->path.dentry = NULL;
746out:
747 rcu_read_unlock();
748 return -ECHILD;
749out_dput:
750 rcu_read_unlock();
751 dput(dentry);
752 return -ECHILD;
753}
754
755static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
756{
757 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
758 return dentry->d_op->d_revalidate(dentry, flags);
759 else
760 return 1;
761}
762
763/**
764 * complete_walk - successful completion of path walk
765 * @nd: pointer nameidata
766 *
767 * If we had been in RCU mode, drop out of it and legitimize nd->path.
768 * Revalidate the final result, unless we'd already done that during
769 * the path walk or the filesystem doesn't ask for it. Return 0 on
770 * success, -error on failure. In case of failure caller does not
771 * need to drop nd->path.
772 */
773static int complete_walk(struct nameidata *nd)
774{
775 struct dentry *dentry = nd->path.dentry;
776 int status;
777
778 if (nd->flags & LOOKUP_RCU) {
779 if (!(nd->flags & LOOKUP_ROOT))
780 nd->root.mnt = NULL;
781 if (unlikely(unlazy_walk(nd)))
782 return -ECHILD;
783 }
784
785 if (likely(!(nd->flags & LOOKUP_JUMPED)))
786 return 0;
787
788 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
789 return 0;
790
791 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
792 if (status > 0)
793 return 0;
794
795 if (!status)
796 status = -ESTALE;
797
798 return status;
799}
800
801static void set_root(struct nameidata *nd)
802{
803 struct fs_struct *fs = current->fs;
804
805 if (nd->flags & LOOKUP_RCU) {
806 unsigned seq;
807
808 do {
809 seq = read_seqcount_begin(&fs->seq);
810 nd->root = fs->root;
811 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
812 } while (read_seqcount_retry(&fs->seq, seq));
813 } else {
814 get_fs_root(fs, &nd->root);
815 nd->flags |= LOOKUP_ROOT_GRABBED;
816 }
817}
818
819static void path_put_conditional(struct path *path, struct nameidata *nd)
820{
821 dput(path->dentry);
822 if (path->mnt != nd->path.mnt)
823 mntput(path->mnt);
824}
825
826static inline void path_to_nameidata(const struct path *path,
827 struct nameidata *nd)
828{
829 if (!(nd->flags & LOOKUP_RCU)) {
830 dput(nd->path.dentry);
831 if (nd->path.mnt != path->mnt)
832 mntput(nd->path.mnt);
833 }
834 nd->path.mnt = path->mnt;
835 nd->path.dentry = path->dentry;
836}
837
838static int nd_jump_root(struct nameidata *nd)
839{
840 if (nd->flags & LOOKUP_RCU) {
841 struct dentry *d;
842 nd->path = nd->root;
843 d = nd->path.dentry;
844 nd->inode = d->d_inode;
845 nd->seq = nd->root_seq;
846 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
847 return -ECHILD;
848 } else {
849 path_put(&nd->path);
850 nd->path = nd->root;
851 path_get(&nd->path);
852 nd->inode = nd->path.dentry->d_inode;
853 }
854 nd->flags |= LOOKUP_JUMPED;
855 return 0;
856}
857
858/*
859 * Helper to directly jump to a known parsed path from ->get_link,
860 * caller must have taken a reference to path beforehand.
861 */
862void nd_jump_link(struct path *path)
863{
864 struct nameidata *nd = current->nameidata;
865 path_put(&nd->path);
866
867 nd->path = *path;
868 nd->inode = nd->path.dentry->d_inode;
869 nd->flags |= LOOKUP_JUMPED;
870}
871
872static inline void put_link(struct nameidata *nd)
873{
874 struct saved *last = nd->stack + --nd->depth;
875 do_delayed_call(&last->done);
876 if (!(nd->flags & LOOKUP_RCU))
877 path_put(&last->link);
878}
879
880int sysctl_protected_symlinks __read_mostly = 0;
881int sysctl_protected_hardlinks __read_mostly = 0;
882int sysctl_protected_fifos __read_mostly;
883int sysctl_protected_regular __read_mostly;
884
885/**
886 * may_follow_link - Check symlink following for unsafe situations
887 * @nd: nameidata pathwalk data
888 *
889 * In the case of the sysctl_protected_symlinks sysctl being enabled,
890 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
891 * in a sticky world-writable directory. This is to protect privileged
892 * processes from failing races against path names that may change out
893 * from under them by way of other users creating malicious symlinks.
894 * It will permit symlinks to be followed only when outside a sticky
895 * world-writable directory, or when the uid of the symlink and follower
896 * match, or when the directory owner matches the symlink's owner.
897 *
898 * Returns 0 if following the symlink is allowed, -ve on error.
899 */
900static inline int may_follow_link(struct nameidata *nd)
901{
902 const struct inode *inode;
903 const struct inode *parent;
904 kuid_t puid;
905
906 if (!sysctl_protected_symlinks)
907 return 0;
908
909 /* Allowed if owner and follower match. */
910 inode = nd->link_inode;
911 if (uid_eq(current_cred()->fsuid, inode->i_uid))
912 return 0;
913
914 /* Allowed if parent directory not sticky and world-writable. */
915 parent = nd->inode;
916 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
917 return 0;
918
919 /* Allowed if parent directory and link owner match. */
920 puid = parent->i_uid;
921 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
922 return 0;
923
924 if (nd->flags & LOOKUP_RCU)
925 return -ECHILD;
926
927 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
928 audit_log_link_denied("follow_link");
929 return -EACCES;
930}
931
932/**
933 * safe_hardlink_source - Check for safe hardlink conditions
934 * @inode: the source inode to hardlink from
935 *
936 * Return false if at least one of the following conditions:
937 * - inode is not a regular file
938 * - inode is setuid
939 * - inode is setgid and group-exec
940 * - access failure for read and write
941 *
942 * Otherwise returns true.
943 */
944static bool safe_hardlink_source(struct inode *inode)
945{
946 umode_t mode = inode->i_mode;
947
948 /* Special files should not get pinned to the filesystem. */
949 if (!S_ISREG(mode))
950 return false;
951
952 /* Setuid files should not get pinned to the filesystem. */
953 if (mode & S_ISUID)
954 return false;
955
956 /* Executable setgid files should not get pinned to the filesystem. */
957 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
958 return false;
959
960 /* Hardlinking to unreadable or unwritable sources is dangerous. */
961 if (inode_permission(inode, MAY_READ | MAY_WRITE))
962 return false;
963
964 return true;
965}
966
967/**
968 * may_linkat - Check permissions for creating a hardlink
969 * @link: the source to hardlink from
970 *
971 * Block hardlink when all of:
972 * - sysctl_protected_hardlinks enabled
973 * - fsuid does not match inode
974 * - hardlink source is unsafe (see safe_hardlink_source() above)
975 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
976 *
977 * Returns 0 if successful, -ve on error.
978 */
979static int may_linkat(struct path *link)
980{
981 struct inode *inode = link->dentry->d_inode;
982
983 /* Inode writeback is not safe when the uid or gid are invalid. */
984 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
985 return -EOVERFLOW;
986
987 if (!sysctl_protected_hardlinks)
988 return 0;
989
990 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
991 * otherwise, it must be a safe source.
992 */
993 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
994 return 0;
995
996 audit_log_link_denied("linkat");
997 return -EPERM;
998}
999
1000/**
1001 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1002 * should be allowed, or not, on files that already
1003 * exist.
1004 * @dir: the sticky parent directory
1005 * @inode: the inode of the file to open
1006 *
1007 * Block an O_CREAT open of a FIFO (or a regular file) when:
1008 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1009 * - the file already exists
1010 * - we are in a sticky directory
1011 * - we don't own the file
1012 * - the owner of the directory doesn't own the file
1013 * - the directory is world writable
1014 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1015 * the directory doesn't have to be world writable: being group writable will
1016 * be enough.
1017 *
1018 * Returns 0 if the open is allowed, -ve on error.
1019 */
1020static int may_create_in_sticky(struct dentry * const dir,
1021 struct inode * const inode)
1022{
1023 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1024 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1025 likely(!(dir->d_inode->i_mode & S_ISVTX)) ||
1026 uid_eq(inode->i_uid, dir->d_inode->i_uid) ||
1027 uid_eq(current_fsuid(), inode->i_uid))
1028 return 0;
1029
1030 if (likely(dir->d_inode->i_mode & 0002) ||
1031 (dir->d_inode->i_mode & 0020 &&
1032 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1033 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1034 return -EACCES;
1035 }
1036 return 0;
1037}
1038
1039static __always_inline
1040const char *get_link(struct nameidata *nd)
1041{
1042 struct saved *last = nd->stack + nd->depth - 1;
1043 struct dentry *dentry = last->link.dentry;
1044 struct inode *inode = nd->link_inode;
1045 int error;
1046 const char *res;
1047
1048 if (!(nd->flags & LOOKUP_RCU)) {
1049 touch_atime(&last->link);
1050 cond_resched();
1051 } else if (atime_needs_update(&last->link, inode)) {
1052 if (unlikely(unlazy_walk(nd)))
1053 return ERR_PTR(-ECHILD);
1054 touch_atime(&last->link);
1055 }
1056
1057 error = security_inode_follow_link(dentry, inode,
1058 nd->flags & LOOKUP_RCU);
1059 if (unlikely(error))
1060 return ERR_PTR(error);
1061
1062 nd->last_type = LAST_BIND;
1063 res = READ_ONCE(inode->i_link);
1064 if (!res) {
1065 const char * (*get)(struct dentry *, struct inode *,
1066 struct delayed_call *);
1067 get = inode->i_op->get_link;
1068 if (nd->flags & LOOKUP_RCU) {
1069 res = get(NULL, inode, &last->done);
1070 if (res == ERR_PTR(-ECHILD)) {
1071 if (unlikely(unlazy_walk(nd)))
1072 return ERR_PTR(-ECHILD);
1073 res = get(dentry, inode, &last->done);
1074 }
1075 } else {
1076 res = get(dentry, inode, &last->done);
1077 }
1078 if (IS_ERR_OR_NULL(res))
1079 return res;
1080 }
1081 if (*res == '/') {
1082 if (!nd->root.mnt)
1083 set_root(nd);
1084 if (unlikely(nd_jump_root(nd)))
1085 return ERR_PTR(-ECHILD);
1086 while (unlikely(*++res == '/'))
1087 ;
1088 }
1089 if (!*res)
1090 res = NULL;
1091 return res;
1092}
1093
1094/*
1095 * follow_up - Find the mountpoint of path's vfsmount
1096 *
1097 * Given a path, find the mountpoint of its source file system.
1098 * Replace @path with the path of the mountpoint in the parent mount.
1099 * Up is towards /.
1100 *
1101 * Return 1 if we went up a level and 0 if we were already at the
1102 * root.
1103 */
1104int follow_up(struct path *path)
1105{
1106 struct mount *mnt = real_mount(path->mnt);
1107 struct mount *parent;
1108 struct dentry *mountpoint;
1109
1110 read_seqlock_excl(&mount_lock);
1111 parent = mnt->mnt_parent;
1112 if (parent == mnt) {
1113 read_sequnlock_excl(&mount_lock);
1114 return 0;
1115 }
1116 mntget(&parent->mnt);
1117 mountpoint = dget(mnt->mnt_mountpoint);
1118 read_sequnlock_excl(&mount_lock);
1119 dput(path->dentry);
1120 path->dentry = mountpoint;
1121 mntput(path->mnt);
1122 path->mnt = &parent->mnt;
1123 return 1;
1124}
1125EXPORT_SYMBOL(follow_up);
1126
1127/*
1128 * Perform an automount
1129 * - return -EISDIR to tell follow_managed() to stop and return the path we
1130 * were called with.
1131 */
1132static int follow_automount(struct path *path, struct nameidata *nd,
1133 bool *need_mntput)
1134{
1135 struct vfsmount *mnt;
1136 int err;
1137
1138 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1139 return -EREMOTE;
1140
1141 /* We don't want to mount if someone's just doing a stat -
1142 * unless they're stat'ing a directory and appended a '/' to
1143 * the name.
1144 *
1145 * We do, however, want to mount if someone wants to open or
1146 * create a file of any type under the mountpoint, wants to
1147 * traverse through the mountpoint or wants to open the
1148 * mounted directory. Also, autofs may mark negative dentries
1149 * as being automount points. These will need the attentions
1150 * of the daemon to instantiate them before they can be used.
1151 */
1152 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1153 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1154 path->dentry->d_inode)
1155 return -EISDIR;
1156
1157 nd->total_link_count++;
1158 if (nd->total_link_count >= 40)
1159 return -ELOOP;
1160
1161 mnt = path->dentry->d_op->d_automount(path);
1162 if (IS_ERR(mnt)) {
1163 /*
1164 * The filesystem is allowed to return -EISDIR here to indicate
1165 * it doesn't want to automount. For instance, autofs would do
1166 * this so that its userspace daemon can mount on this dentry.
1167 *
1168 * However, we can only permit this if it's a terminal point in
1169 * the path being looked up; if it wasn't then the remainder of
1170 * the path is inaccessible and we should say so.
1171 */
1172 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1173 return -EREMOTE;
1174 return PTR_ERR(mnt);
1175 }
1176
1177 if (!mnt) /* mount collision */
1178 return 0;
1179
1180 if (!*need_mntput) {
1181 /* lock_mount() may release path->mnt on error */
1182 mntget(path->mnt);
1183 *need_mntput = true;
1184 }
1185 err = finish_automount(mnt, path);
1186
1187 switch (err) {
1188 case -EBUSY:
1189 /* Someone else made a mount here whilst we were busy */
1190 return 0;
1191 case 0:
1192 path_put(path);
1193 path->mnt = mnt;
1194 path->dentry = dget(mnt->mnt_root);
1195 return 0;
1196 default:
1197 return err;
1198 }
1199
1200}
1201
1202/*
1203 * Handle a dentry that is managed in some way.
1204 * - Flagged for transit management (autofs)
1205 * - Flagged as mountpoint
1206 * - Flagged as automount point
1207 *
1208 * This may only be called in refwalk mode.
1209 *
1210 * Serialization is taken care of in namespace.c
1211 */
1212static int follow_managed(struct path *path, struct nameidata *nd)
1213{
1214 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1215 unsigned managed;
1216 bool need_mntput = false;
1217 int ret = 0;
1218
1219 /* Given that we're not holding a lock here, we retain the value in a
1220 * local variable for each dentry as we look at it so that we don't see
1221 * the components of that value change under us */
1222 while (managed = READ_ONCE(path->dentry->d_flags),
1223 managed &= DCACHE_MANAGED_DENTRY,
1224 unlikely(managed != 0)) {
1225 /* Allow the filesystem to manage the transit without i_mutex
1226 * being held. */
1227 if (managed & DCACHE_MANAGE_TRANSIT) {
1228 BUG_ON(!path->dentry->d_op);
1229 BUG_ON(!path->dentry->d_op->d_manage);
1230 ret = path->dentry->d_op->d_manage(path, false);
1231 if (ret < 0)
1232 break;
1233 }
1234
1235 /* Transit to a mounted filesystem. */
1236 if (managed & DCACHE_MOUNTED) {
1237 struct vfsmount *mounted = lookup_mnt(path);
1238 if (mounted) {
1239 dput(path->dentry);
1240 if (need_mntput)
1241 mntput(path->mnt);
1242 path->mnt = mounted;
1243 path->dentry = dget(mounted->mnt_root);
1244 need_mntput = true;
1245 continue;
1246 }
1247
1248 /* Something is mounted on this dentry in another
1249 * namespace and/or whatever was mounted there in this
1250 * namespace got unmounted before lookup_mnt() could
1251 * get it */
1252 }
1253
1254 /* Handle an automount point */
1255 if (managed & DCACHE_NEED_AUTOMOUNT) {
1256 ret = follow_automount(path, nd, &need_mntput);
1257 if (ret < 0)
1258 break;
1259 continue;
1260 }
1261
1262 /* We didn't change the current path point */
1263 break;
1264 }
1265
1266 if (need_mntput && path->mnt == mnt)
1267 mntput(path->mnt);
1268 if (ret == -EISDIR || !ret)
1269 ret = 1;
1270 if (need_mntput)
1271 nd->flags |= LOOKUP_JUMPED;
1272 if (unlikely(ret < 0))
1273 path_put_conditional(path, nd);
1274 return ret;
1275}
1276
1277int follow_down_one(struct path *path)
1278{
1279 struct vfsmount *mounted;
1280
1281 mounted = lookup_mnt(path);
1282 if (mounted) {
1283 dput(path->dentry);
1284 mntput(path->mnt);
1285 path->mnt = mounted;
1286 path->dentry = dget(mounted->mnt_root);
1287 return 1;
1288 }
1289 return 0;
1290}
1291EXPORT_SYMBOL(follow_down_one);
1292
1293static inline int managed_dentry_rcu(const struct path *path)
1294{
1295 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1296 path->dentry->d_op->d_manage(path, true) : 0;
1297}
1298
1299/*
1300 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1301 * we meet a managed dentry that would need blocking.
1302 */
1303static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1304 struct inode **inode, unsigned *seqp)
1305{
1306 for (;;) {
1307 struct mount *mounted;
1308 /*
1309 * Don't forget we might have a non-mountpoint managed dentry
1310 * that wants to block transit.
1311 */
1312 switch (managed_dentry_rcu(path)) {
1313 case -ECHILD:
1314 default:
1315 return false;
1316 case -EISDIR:
1317 return true;
1318 case 0:
1319 break;
1320 }
1321
1322 if (!d_mountpoint(path->dentry))
1323 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1324
1325 mounted = __lookup_mnt(path->mnt, path->dentry);
1326 if (!mounted)
1327 break;
1328 path->mnt = &mounted->mnt;
1329 path->dentry = mounted->mnt.mnt_root;
1330 nd->flags |= LOOKUP_JUMPED;
1331 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1332 /*
1333 * Update the inode too. We don't need to re-check the
1334 * dentry sequence number here after this d_inode read,
1335 * because a mount-point is always pinned.
1336 */
1337 *inode = path->dentry->d_inode;
1338 }
1339 return !read_seqretry(&mount_lock, nd->m_seq) &&
1340 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1341}
1342
1343static int follow_dotdot_rcu(struct nameidata *nd)
1344{
1345 struct inode *inode = nd->inode;
1346
1347 while (1) {
1348 if (path_equal(&nd->path, &nd->root))
1349 break;
1350 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1351 struct dentry *old = nd->path.dentry;
1352 struct dentry *parent = old->d_parent;
1353 unsigned seq;
1354
1355 inode = parent->d_inode;
1356 seq = read_seqcount_begin(&parent->d_seq);
1357 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1358 return -ECHILD;
1359 nd->path.dentry = parent;
1360 nd->seq = seq;
1361 if (unlikely(!path_connected(&nd->path)))
1362 return -ENOENT;
1363 break;
1364 } else {
1365 struct mount *mnt = real_mount(nd->path.mnt);
1366 struct mount *mparent = mnt->mnt_parent;
1367 struct dentry *mountpoint = mnt->mnt_mountpoint;
1368 struct inode *inode2 = mountpoint->d_inode;
1369 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1370 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1371 return -ECHILD;
1372 if (&mparent->mnt == nd->path.mnt)
1373 break;
1374 /* we know that mountpoint was pinned */
1375 nd->path.dentry = mountpoint;
1376 nd->path.mnt = &mparent->mnt;
1377 inode = inode2;
1378 nd->seq = seq;
1379 }
1380 }
1381 while (unlikely(d_mountpoint(nd->path.dentry))) {
1382 struct mount *mounted;
1383 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1384 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1385 return -ECHILD;
1386 if (!mounted)
1387 break;
1388 nd->path.mnt = &mounted->mnt;
1389 nd->path.dentry = mounted->mnt.mnt_root;
1390 inode = nd->path.dentry->d_inode;
1391 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1392 }
1393 nd->inode = inode;
1394 return 0;
1395}
1396
1397/*
1398 * Follow down to the covering mount currently visible to userspace. At each
1399 * point, the filesystem owning that dentry may be queried as to whether the
1400 * caller is permitted to proceed or not.
1401 */
1402int follow_down(struct path *path)
1403{
1404 unsigned managed;
1405 int ret;
1406
1407 while (managed = READ_ONCE(path->dentry->d_flags),
1408 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1409 /* Allow the filesystem to manage the transit without i_mutex
1410 * being held.
1411 *
1412 * We indicate to the filesystem if someone is trying to mount
1413 * something here. This gives autofs the chance to deny anyone
1414 * other than its daemon the right to mount on its
1415 * superstructure.
1416 *
1417 * The filesystem may sleep at this point.
1418 */
1419 if (managed & DCACHE_MANAGE_TRANSIT) {
1420 BUG_ON(!path->dentry->d_op);
1421 BUG_ON(!path->dentry->d_op->d_manage);
1422 ret = path->dentry->d_op->d_manage(path, false);
1423 if (ret < 0)
1424 return ret == -EISDIR ? 0 : ret;
1425 }
1426
1427 /* Transit to a mounted filesystem. */
1428 if (managed & DCACHE_MOUNTED) {
1429 struct vfsmount *mounted = lookup_mnt(path);
1430 if (!mounted)
1431 break;
1432 dput(path->dentry);
1433 mntput(path->mnt);
1434 path->mnt = mounted;
1435 path->dentry = dget(mounted->mnt_root);
1436 continue;
1437 }
1438
1439 /* Don't handle automount points here */
1440 break;
1441 }
1442 return 0;
1443}
1444EXPORT_SYMBOL(follow_down);
1445
1446/*
1447 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1448 */
1449static void follow_mount(struct path *path)
1450{
1451 while (d_mountpoint(path->dentry)) {
1452 struct vfsmount *mounted = lookup_mnt(path);
1453 if (!mounted)
1454 break;
1455 dput(path->dentry);
1456 mntput(path->mnt);
1457 path->mnt = mounted;
1458 path->dentry = dget(mounted->mnt_root);
1459 }
1460}
1461
1462static int path_parent_directory(struct path *path)
1463{
1464 struct dentry *old = path->dentry;
1465 /* rare case of legitimate dget_parent()... */
1466 path->dentry = dget_parent(path->dentry);
1467 dput(old);
1468 if (unlikely(!path_connected(path)))
1469 return -ENOENT;
1470 return 0;
1471}
1472
1473static int follow_dotdot(struct nameidata *nd)
1474{
1475 while(1) {
1476 if (path_equal(&nd->path, &nd->root))
1477 break;
1478 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1479 int ret = path_parent_directory(&nd->path);
1480 if (ret)
1481 return ret;
1482 break;
1483 }
1484 if (!follow_up(&nd->path))
1485 break;
1486 }
1487 follow_mount(&nd->path);
1488 nd->inode = nd->path.dentry->d_inode;
1489 return 0;
1490}
1491
1492/*
1493 * This looks up the name in dcache and possibly revalidates the found dentry.
1494 * NULL is returned if the dentry does not exist in the cache.
1495 */
1496static struct dentry *lookup_dcache(const struct qstr *name,
1497 struct dentry *dir,
1498 unsigned int flags)
1499{
1500 struct dentry *dentry = d_lookup(dir, name);
1501 if (dentry) {
1502 int error = d_revalidate(dentry, flags);
1503 if (unlikely(error <= 0)) {
1504 if (!error)
1505 d_invalidate(dentry);
1506 dput(dentry);
1507 return ERR_PTR(error);
1508 }
1509 }
1510 return dentry;
1511}
1512
1513/*
1514 * Parent directory has inode locked exclusive. This is one
1515 * and only case when ->lookup() gets called on non in-lookup
1516 * dentries - as the matter of fact, this only gets called
1517 * when directory is guaranteed to have no in-lookup children
1518 * at all.
1519 */
1520static struct dentry *__lookup_hash(const struct qstr *name,
1521 struct dentry *base, unsigned int flags)
1522{
1523 struct dentry *dentry = lookup_dcache(name, base, flags);
1524 struct dentry *old;
1525 struct inode *dir = base->d_inode;
1526
1527 if (dentry)
1528 return dentry;
1529
1530 /* Don't create child dentry for a dead directory. */
1531 if (unlikely(IS_DEADDIR(dir)))
1532 return ERR_PTR(-ENOENT);
1533
1534 dentry = d_alloc(base, name);
1535 if (unlikely(!dentry))
1536 return ERR_PTR(-ENOMEM);
1537
1538 old = dir->i_op->lookup(dir, dentry, flags);
1539 if (unlikely(old)) {
1540 dput(dentry);
1541 dentry = old;
1542 }
1543 return dentry;
1544}
1545
1546static int lookup_fast(struct nameidata *nd,
1547 struct path *path, struct inode **inode,
1548 unsigned *seqp)
1549{
1550 struct vfsmount *mnt = nd->path.mnt;
1551 struct dentry *dentry, *parent = nd->path.dentry;
1552 int status = 1;
1553 int err;
1554
1555 /*
1556 * Rename seqlock is not required here because in the off chance
1557 * of a false negative due to a concurrent rename, the caller is
1558 * going to fall back to non-racy lookup.
1559 */
1560 if (nd->flags & LOOKUP_RCU) {
1561 unsigned seq;
1562 bool negative;
1563 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1564 if (unlikely(!dentry)) {
1565 if (unlazy_walk(nd))
1566 return -ECHILD;
1567 return 0;
1568 }
1569
1570 /*
1571 * This sequence count validates that the inode matches
1572 * the dentry name information from lookup.
1573 */
1574 *inode = d_backing_inode(dentry);
1575 negative = d_is_negative(dentry);
1576 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1577 return -ECHILD;
1578
1579 /*
1580 * This sequence count validates that the parent had no
1581 * changes while we did the lookup of the dentry above.
1582 *
1583 * The memory barrier in read_seqcount_begin of child is
1584 * enough, we can use __read_seqcount_retry here.
1585 */
1586 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1587 return -ECHILD;
1588
1589 *seqp = seq;
1590 status = d_revalidate(dentry, nd->flags);
1591 if (likely(status > 0)) {
1592 /*
1593 * Note: do negative dentry check after revalidation in
1594 * case that drops it.
1595 */
1596 if (unlikely(negative))
1597 return -ENOENT;
1598 path->mnt = mnt;
1599 path->dentry = dentry;
1600 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1601 return 1;
1602 }
1603 if (unlazy_child(nd, dentry, seq))
1604 return -ECHILD;
1605 if (unlikely(status == -ECHILD))
1606 /* we'd been told to redo it in non-rcu mode */
1607 status = d_revalidate(dentry, nd->flags);
1608 } else {
1609 dentry = __d_lookup(parent, &nd->last);
1610 if (unlikely(!dentry))
1611 return 0;
1612 status = d_revalidate(dentry, nd->flags);
1613 }
1614 if (unlikely(status <= 0)) {
1615 if (!status)
1616 d_invalidate(dentry);
1617 dput(dentry);
1618 return status;
1619 }
1620 if (unlikely(d_is_negative(dentry))) {
1621 dput(dentry);
1622 return -ENOENT;
1623 }
1624
1625 path->mnt = mnt;
1626 path->dentry = dentry;
1627 err = follow_managed(path, nd);
1628 if (likely(err > 0))
1629 *inode = d_backing_inode(path->dentry);
1630 return err;
1631}
1632
1633/* Fast lookup failed, do it the slow way */
1634static struct dentry *__lookup_slow(const struct qstr *name,
1635 struct dentry *dir,
1636 unsigned int flags)
1637{
1638 struct dentry *dentry, *old;
1639 struct inode *inode = dir->d_inode;
1640 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1641
1642 /* Don't go there if it's already dead */
1643 if (unlikely(IS_DEADDIR(inode)))
1644 return ERR_PTR(-ENOENT);
1645again:
1646 dentry = d_alloc_parallel(dir, name, &wq);
1647 if (IS_ERR(dentry))
1648 return dentry;
1649 if (unlikely(!d_in_lookup(dentry))) {
1650 if (!(flags & LOOKUP_NO_REVAL)) {
1651 int error = d_revalidate(dentry, flags);
1652 if (unlikely(error <= 0)) {
1653 if (!error) {
1654 d_invalidate(dentry);
1655 dput(dentry);
1656 goto again;
1657 }
1658 dput(dentry);
1659 dentry = ERR_PTR(error);
1660 }
1661 }
1662 } else {
1663 old = inode->i_op->lookup(inode, dentry, flags);
1664 d_lookup_done(dentry);
1665 if (unlikely(old)) {
1666 dput(dentry);
1667 dentry = old;
1668 }
1669 }
1670 return dentry;
1671}
1672
1673static struct dentry *lookup_slow(const struct qstr *name,
1674 struct dentry *dir,
1675 unsigned int flags)
1676{
1677 struct inode *inode = dir->d_inode;
1678 struct dentry *res;
1679 inode_lock_shared(inode);
1680 res = __lookup_slow(name, dir, flags);
1681 inode_unlock_shared(inode);
1682 return res;
1683}
1684
1685static inline int may_lookup(struct nameidata *nd)
1686{
1687 if (nd->flags & LOOKUP_RCU) {
1688 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1689 if (err != -ECHILD)
1690 return err;
1691 if (unlazy_walk(nd))
1692 return -ECHILD;
1693 }
1694 return inode_permission(nd->inode, MAY_EXEC);
1695}
1696
1697static inline int handle_dots(struct nameidata *nd, int type)
1698{
1699 if (type == LAST_DOTDOT) {
1700 if (!nd->root.mnt)
1701 set_root(nd);
1702 if (nd->flags & LOOKUP_RCU) {
1703 return follow_dotdot_rcu(nd);
1704 } else
1705 return follow_dotdot(nd);
1706 }
1707 return 0;
1708}
1709
1710static int pick_link(struct nameidata *nd, struct path *link,
1711 struct inode *inode, unsigned seq)
1712{
1713 int error;
1714 struct saved *last;
1715 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1716 path_to_nameidata(link, nd);
1717 return -ELOOP;
1718 }
1719 if (!(nd->flags & LOOKUP_RCU)) {
1720 if (link->mnt == nd->path.mnt)
1721 mntget(link->mnt);
1722 }
1723 error = nd_alloc_stack(nd);
1724 if (unlikely(error)) {
1725 if (error == -ECHILD) {
1726 if (unlikely(!legitimize_path(nd, link, seq))) {
1727 drop_links(nd);
1728 nd->depth = 0;
1729 nd->flags &= ~LOOKUP_RCU;
1730 nd->path.mnt = NULL;
1731 nd->path.dentry = NULL;
1732 rcu_read_unlock();
1733 } else if (likely(unlazy_walk(nd)) == 0)
1734 error = nd_alloc_stack(nd);
1735 }
1736 if (error) {
1737 path_put(link);
1738 return error;
1739 }
1740 }
1741
1742 last = nd->stack + nd->depth++;
1743 last->link = *link;
1744 clear_delayed_call(&last->done);
1745 nd->link_inode = inode;
1746 last->seq = seq;
1747 return 1;
1748}
1749
1750enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1751
1752/*
1753 * Do we need to follow links? We _really_ want to be able
1754 * to do this check without having to look at inode->i_op,
1755 * so we keep a cache of "no, this doesn't need follow_link"
1756 * for the common case.
1757 */
1758static inline int step_into(struct nameidata *nd, struct path *path,
1759 int flags, struct inode *inode, unsigned seq)
1760{
1761 if (!(flags & WALK_MORE) && nd->depth)
1762 put_link(nd);
1763 if (likely(!d_is_symlink(path->dentry)) ||
1764 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1765 /* not a symlink or should not follow */
1766 path_to_nameidata(path, nd);
1767 nd->inode = inode;
1768 nd->seq = seq;
1769 return 0;
1770 }
1771 /* make sure that d_is_symlink above matches inode */
1772 if (nd->flags & LOOKUP_RCU) {
1773 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1774 return -ECHILD;
1775 }
1776 return pick_link(nd, path, inode, seq);
1777}
1778
1779static int walk_component(struct nameidata *nd, int flags)
1780{
1781 struct path path;
1782 struct inode *inode;
1783 unsigned seq;
1784 int err;
1785 /*
1786 * "." and ".." are special - ".." especially so because it has
1787 * to be able to know about the current root directory and
1788 * parent relationships.
1789 */
1790 if (unlikely(nd->last_type != LAST_NORM)) {
1791 err = handle_dots(nd, nd->last_type);
1792 if (!(flags & WALK_MORE) && nd->depth)
1793 put_link(nd);
1794 return err;
1795 }
1796 err = lookup_fast(nd, &path, &inode, &seq);
1797 if (unlikely(err <= 0)) {
1798 if (err < 0)
1799 return err;
1800 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1801 nd->flags);
1802 if (IS_ERR(path.dentry))
1803 return PTR_ERR(path.dentry);
1804
1805 path.mnt = nd->path.mnt;
1806 err = follow_managed(&path, nd);
1807 if (unlikely(err < 0))
1808 return err;
1809
1810 if (unlikely(d_is_negative(path.dentry))) {
1811 path_to_nameidata(&path, nd);
1812 return -ENOENT;
1813 }
1814
1815 seq = 0; /* we are already out of RCU mode */
1816 inode = d_backing_inode(path.dentry);
1817 }
1818
1819 return step_into(nd, &path, flags, inode, seq);
1820}
1821
1822/*
1823 * We can do the critical dentry name comparison and hashing
1824 * operations one word at a time, but we are limited to:
1825 *
1826 * - Architectures with fast unaligned word accesses. We could
1827 * do a "get_unaligned()" if this helps and is sufficiently
1828 * fast.
1829 *
1830 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1831 * do not trap on the (extremely unlikely) case of a page
1832 * crossing operation.
1833 *
1834 * - Furthermore, we need an efficient 64-bit compile for the
1835 * 64-bit case in order to generate the "number of bytes in
1836 * the final mask". Again, that could be replaced with a
1837 * efficient population count instruction or similar.
1838 */
1839#ifdef CONFIG_DCACHE_WORD_ACCESS
1840
1841#include <asm/word-at-a-time.h>
1842
1843#ifdef HASH_MIX
1844
1845/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1846
1847#elif defined(CONFIG_64BIT)
1848/*
1849 * Register pressure in the mixing function is an issue, particularly
1850 * on 32-bit x86, but almost any function requires one state value and
1851 * one temporary. Instead, use a function designed for two state values
1852 * and no temporaries.
1853 *
1854 * This function cannot create a collision in only two iterations, so
1855 * we have two iterations to achieve avalanche. In those two iterations,
1856 * we have six layers of mixing, which is enough to spread one bit's
1857 * influence out to 2^6 = 64 state bits.
1858 *
1859 * Rotate constants are scored by considering either 64 one-bit input
1860 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1861 * probability of that delta causing a change to each of the 128 output
1862 * bits, using a sample of random initial states.
1863 *
1864 * The Shannon entropy of the computed probabilities is then summed
1865 * to produce a score. Ideally, any input change has a 50% chance of
1866 * toggling any given output bit.
1867 *
1868 * Mixing scores (in bits) for (12,45):
1869 * Input delta: 1-bit 2-bit
1870 * 1 round: 713.3 42542.6
1871 * 2 rounds: 2753.7 140389.8
1872 * 3 rounds: 5954.1 233458.2
1873 * 4 rounds: 7862.6 256672.2
1874 * Perfect: 8192 258048
1875 * (64*128) (64*63/2 * 128)
1876 */
1877#define HASH_MIX(x, y, a) \
1878 ( x ^= (a), \
1879 y ^= x, x = rol64(x,12),\
1880 x += y, y = rol64(y,45),\
1881 y *= 9 )
1882
1883/*
1884 * Fold two longs into one 32-bit hash value. This must be fast, but
1885 * latency isn't quite as critical, as there is a fair bit of additional
1886 * work done before the hash value is used.
1887 */
1888static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1889{
1890 y ^= x * GOLDEN_RATIO_64;
1891 y *= GOLDEN_RATIO_64;
1892 return y >> 32;
1893}
1894
1895#else /* 32-bit case */
1896
1897/*
1898 * Mixing scores (in bits) for (7,20):
1899 * Input delta: 1-bit 2-bit
1900 * 1 round: 330.3 9201.6
1901 * 2 rounds: 1246.4 25475.4
1902 * 3 rounds: 1907.1 31295.1
1903 * 4 rounds: 2042.3 31718.6
1904 * Perfect: 2048 31744
1905 * (32*64) (32*31/2 * 64)
1906 */
1907#define HASH_MIX(x, y, a) \
1908 ( x ^= (a), \
1909 y ^= x, x = rol32(x, 7),\
1910 x += y, y = rol32(y,20),\
1911 y *= 9 )
1912
1913static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1914{
1915 /* Use arch-optimized multiply if one exists */
1916 return __hash_32(y ^ __hash_32(x));
1917}
1918
1919#endif
1920
1921/*
1922 * Return the hash of a string of known length. This is carfully
1923 * designed to match hash_name(), which is the more critical function.
1924 * In particular, we must end by hashing a final word containing 0..7
1925 * payload bytes, to match the way that hash_name() iterates until it
1926 * finds the delimiter after the name.
1927 */
1928unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1929{
1930 unsigned long a, x = 0, y = (unsigned long)salt;
1931
1932 for (;;) {
1933 if (!len)
1934 goto done;
1935 a = load_unaligned_zeropad(name);
1936 if (len < sizeof(unsigned long))
1937 break;
1938 HASH_MIX(x, y, a);
1939 name += sizeof(unsigned long);
1940 len -= sizeof(unsigned long);
1941 }
1942 x ^= a & bytemask_from_count(len);
1943done:
1944 return fold_hash(x, y);
1945}
1946EXPORT_SYMBOL(full_name_hash);
1947
1948/* Return the "hash_len" (hash and length) of a null-terminated string */
1949u64 hashlen_string(const void *salt, const char *name)
1950{
1951 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1952 unsigned long adata, mask, len;
1953 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1954
1955 len = 0;
1956 goto inside;
1957
1958 do {
1959 HASH_MIX(x, y, a);
1960 len += sizeof(unsigned long);
1961inside:
1962 a = load_unaligned_zeropad(name+len);
1963 } while (!has_zero(a, &adata, &constants));
1964
1965 adata = prep_zero_mask(a, adata, &constants);
1966 mask = create_zero_mask(adata);
1967 x ^= a & zero_bytemask(mask);
1968
1969 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1970}
1971EXPORT_SYMBOL(hashlen_string);
1972
1973/*
1974 * Calculate the length and hash of the path component, and
1975 * return the "hash_len" as the result.
1976 */
1977static inline u64 hash_name(const void *salt, const char *name)
1978{
1979 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1980 unsigned long adata, bdata, mask, len;
1981 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1982
1983 len = 0;
1984 goto inside;
1985
1986 do {
1987 HASH_MIX(x, y, a);
1988 len += sizeof(unsigned long);
1989inside:
1990 a = load_unaligned_zeropad(name+len);
1991 b = a ^ REPEAT_BYTE('/');
1992 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1993
1994 adata = prep_zero_mask(a, adata, &constants);
1995 bdata = prep_zero_mask(b, bdata, &constants);
1996 mask = create_zero_mask(adata | bdata);
1997 x ^= a & zero_bytemask(mask);
1998
1999 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2000}
2001
2002#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2003
2004/* Return the hash of a string of known length */
2005unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2006{
2007 unsigned long hash = init_name_hash(salt);
2008 while (len--)
2009 hash = partial_name_hash((unsigned char)*name++, hash);
2010 return end_name_hash(hash);
2011}
2012EXPORT_SYMBOL(full_name_hash);
2013
2014/* Return the "hash_len" (hash and length) of a null-terminated string */
2015u64 hashlen_string(const void *salt, const char *name)
2016{
2017 unsigned long hash = init_name_hash(salt);
2018 unsigned long len = 0, c;
2019
2020 c = (unsigned char)*name;
2021 while (c) {
2022 len++;
2023 hash = partial_name_hash(c, hash);
2024 c = (unsigned char)name[len];
2025 }
2026 return hashlen_create(end_name_hash(hash), len);
2027}
2028EXPORT_SYMBOL(hashlen_string);
2029
2030/*
2031 * We know there's a real path component here of at least
2032 * one character.
2033 */
2034static inline u64 hash_name(const void *salt, const char *name)
2035{
2036 unsigned long hash = init_name_hash(salt);
2037 unsigned long len = 0, c;
2038
2039 c = (unsigned char)*name;
2040 do {
2041 len++;
2042 hash = partial_name_hash(c, hash);
2043 c = (unsigned char)name[len];
2044 } while (c && c != '/');
2045 return hashlen_create(end_name_hash(hash), len);
2046}
2047
2048#endif
2049
2050/*
2051 * Name resolution.
2052 * This is the basic name resolution function, turning a pathname into
2053 * the final dentry. We expect 'base' to be positive and a directory.
2054 *
2055 * Returns 0 and nd will have valid dentry and mnt on success.
2056 * Returns error and drops reference to input namei data on failure.
2057 */
2058static int link_path_walk(const char *name, struct nameidata *nd)
2059{
2060 int err;
2061
2062 if (IS_ERR(name))
2063 return PTR_ERR(name);
2064 while (*name=='/')
2065 name++;
2066 if (!*name)
2067 return 0;
2068
2069 /* At this point we know we have a real path component. */
2070 for(;;) {
2071 u64 hash_len;
2072 int type;
2073
2074 err = may_lookup(nd);
2075 if (err)
2076 return err;
2077
2078 hash_len = hash_name(nd->path.dentry, name);
2079
2080 type = LAST_NORM;
2081 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2082 case 2:
2083 if (name[1] == '.') {
2084 type = LAST_DOTDOT;
2085 nd->flags |= LOOKUP_JUMPED;
2086 }
2087 break;
2088 case 1:
2089 type = LAST_DOT;
2090 }
2091 if (likely(type == LAST_NORM)) {
2092 struct dentry *parent = nd->path.dentry;
2093 nd->flags &= ~LOOKUP_JUMPED;
2094 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2095 struct qstr this = { { .hash_len = hash_len }, .name = name };
2096 err = parent->d_op->d_hash(parent, &this);
2097 if (err < 0)
2098 return err;
2099 hash_len = this.hash_len;
2100 name = this.name;
2101 }
2102 }
2103
2104 nd->last.hash_len = hash_len;
2105 nd->last.name = name;
2106 nd->last_type = type;
2107
2108 name += hashlen_len(hash_len);
2109 if (!*name)
2110 goto OK;
2111 /*
2112 * If it wasn't NUL, we know it was '/'. Skip that
2113 * slash, and continue until no more slashes.
2114 */
2115 do {
2116 name++;
2117 } while (unlikely(*name == '/'));
2118 if (unlikely(!*name)) {
2119OK:
2120 /* pathname body, done */
2121 if (!nd->depth)
2122 return 0;
2123 name = nd->stack[nd->depth - 1].name;
2124 /* trailing symlink, done */
2125 if (!name)
2126 return 0;
2127 /* last component of nested symlink */
2128 err = walk_component(nd, WALK_FOLLOW);
2129 } else {
2130 /* not the last component */
2131 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2132 }
2133 if (err < 0)
2134 return err;
2135
2136 if (err) {
2137 const char *s = get_link(nd);
2138
2139 if (IS_ERR(s))
2140 return PTR_ERR(s);
2141 err = 0;
2142 if (unlikely(!s)) {
2143 /* jumped */
2144 put_link(nd);
2145 } else {
2146 nd->stack[nd->depth - 1].name = name;
2147 name = s;
2148 continue;
2149 }
2150 }
2151 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2152 if (nd->flags & LOOKUP_RCU) {
2153 if (unlazy_walk(nd))
2154 return -ECHILD;
2155 }
2156 return -ENOTDIR;
2157 }
2158 }
2159}
2160
2161/* must be paired with terminate_walk() */
2162static const char *path_init(struct nameidata *nd, unsigned flags)
2163{
2164 const char *s = nd->name->name;
2165
2166 if (!*s)
2167 flags &= ~LOOKUP_RCU;
2168 if (flags & LOOKUP_RCU)
2169 rcu_read_lock();
2170
2171 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2172 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2173 nd->depth = 0;
2174 if (flags & LOOKUP_ROOT) {
2175 struct dentry *root = nd->root.dentry;
2176 struct inode *inode = root->d_inode;
2177 if (*s && unlikely(!d_can_lookup(root)))
2178 return ERR_PTR(-ENOTDIR);
2179 nd->path = nd->root;
2180 nd->inode = inode;
2181 if (flags & LOOKUP_RCU) {
2182 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2183 nd->root_seq = nd->seq;
2184 nd->m_seq = read_seqbegin(&mount_lock);
2185 } else {
2186 path_get(&nd->path);
2187 }
2188 return s;
2189 }
2190
2191 nd->root.mnt = NULL;
2192 nd->path.mnt = NULL;
2193 nd->path.dentry = NULL;
2194
2195 nd->m_seq = read_seqbegin(&mount_lock);
2196 if (*s == '/') {
2197 set_root(nd);
2198 if (likely(!nd_jump_root(nd)))
2199 return s;
2200 return ERR_PTR(-ECHILD);
2201 } else if (nd->dfd == AT_FDCWD) {
2202 if (flags & LOOKUP_RCU) {
2203 struct fs_struct *fs = current->fs;
2204 unsigned seq;
2205
2206 do {
2207 seq = read_seqcount_begin(&fs->seq);
2208 nd->path = fs->pwd;
2209 nd->inode = nd->path.dentry->d_inode;
2210 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2211 } while (read_seqcount_retry(&fs->seq, seq));
2212 } else {
2213 get_fs_pwd(current->fs, &nd->path);
2214 nd->inode = nd->path.dentry->d_inode;
2215 }
2216 return s;
2217 } else {
2218 /* Caller must check execute permissions on the starting path component */
2219 struct fd f = fdget_raw(nd->dfd);
2220 struct dentry *dentry;
2221
2222 if (!f.file)
2223 return ERR_PTR(-EBADF);
2224
2225 dentry = f.file->f_path.dentry;
2226
2227 if (*s && unlikely(!d_can_lookup(dentry))) {
2228 fdput(f);
2229 return ERR_PTR(-ENOTDIR);
2230 }
2231
2232 nd->path = f.file->f_path;
2233 if (flags & LOOKUP_RCU) {
2234 nd->inode = nd->path.dentry->d_inode;
2235 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2236 } else {
2237 path_get(&nd->path);
2238 nd->inode = nd->path.dentry->d_inode;
2239 }
2240 fdput(f);
2241 return s;
2242 }
2243}
2244
2245static const char *trailing_symlink(struct nameidata *nd)
2246{
2247 const char *s;
2248 int error = may_follow_link(nd);
2249 if (unlikely(error))
2250 return ERR_PTR(error);
2251 nd->flags |= LOOKUP_PARENT;
2252 nd->stack[0].name = NULL;
2253 s = get_link(nd);
2254 return s ? s : "";
2255}
2256
2257static inline int lookup_last(struct nameidata *nd)
2258{
2259 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2260 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2261
2262 nd->flags &= ~LOOKUP_PARENT;
2263 return walk_component(nd, 0);
2264}
2265
2266static int handle_lookup_down(struct nameidata *nd)
2267{
2268 struct path path = nd->path;
2269 struct inode *inode = nd->inode;
2270 unsigned seq = nd->seq;
2271 int err;
2272
2273 if (nd->flags & LOOKUP_RCU) {
2274 /*
2275 * don't bother with unlazy_walk on failure - we are
2276 * at the very beginning of walk, so we lose nothing
2277 * if we simply redo everything in non-RCU mode
2278 */
2279 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2280 return -ECHILD;
2281 } else {
2282 dget(path.dentry);
2283 err = follow_managed(&path, nd);
2284 if (unlikely(err < 0))
2285 return err;
2286 inode = d_backing_inode(path.dentry);
2287 seq = 0;
2288 }
2289 path_to_nameidata(&path, nd);
2290 nd->inode = inode;
2291 nd->seq = seq;
2292 return 0;
2293}
2294
2295/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2296static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2297{
2298 const char *s = path_init(nd, flags);
2299 int err;
2300
2301 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2302 err = handle_lookup_down(nd);
2303 if (unlikely(err < 0))
2304 s = ERR_PTR(err);
2305 }
2306
2307 while (!(err = link_path_walk(s, nd))
2308 && ((err = lookup_last(nd)) > 0)) {
2309 s = trailing_symlink(nd);
2310 }
2311 if (!err)
2312 err = complete_walk(nd);
2313
2314 if (!err && nd->flags & LOOKUP_DIRECTORY)
2315 if (!d_can_lookup(nd->path.dentry))
2316 err = -ENOTDIR;
2317 if (!err) {
2318 *path = nd->path;
2319 nd->path.mnt = NULL;
2320 nd->path.dentry = NULL;
2321 }
2322 terminate_walk(nd);
2323 return err;
2324}
2325
2326int filename_lookup(int dfd, struct filename *name, unsigned flags,
2327 struct path *path, struct path *root)
2328{
2329 int retval;
2330 struct nameidata nd;
2331 if (IS_ERR(name))
2332 return PTR_ERR(name);
2333 if (unlikely(root)) {
2334 nd.root = *root;
2335 flags |= LOOKUP_ROOT;
2336 }
2337 set_nameidata(&nd, dfd, name);
2338 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2339 if (unlikely(retval == -ECHILD))
2340 retval = path_lookupat(&nd, flags, path);
2341 if (unlikely(retval == -ESTALE))
2342 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2343
2344 if (likely(!retval))
2345 audit_inode(name, path->dentry, 0);
2346 restore_nameidata();
2347 putname(name);
2348 return retval;
2349}
2350
2351/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2352static int path_parentat(struct nameidata *nd, unsigned flags,
2353 struct path *parent)
2354{
2355 const char *s = path_init(nd, flags);
2356 int err = link_path_walk(s, nd);
2357 if (!err)
2358 err = complete_walk(nd);
2359 if (!err) {
2360 *parent = nd->path;
2361 nd->path.mnt = NULL;
2362 nd->path.dentry = NULL;
2363 }
2364 terminate_walk(nd);
2365 return err;
2366}
2367
2368static struct filename *filename_parentat(int dfd, struct filename *name,
2369 unsigned int flags, struct path *parent,
2370 struct qstr *last, int *type)
2371{
2372 int retval;
2373 struct nameidata nd;
2374
2375 if (IS_ERR(name))
2376 return name;
2377 set_nameidata(&nd, dfd, name);
2378 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2379 if (unlikely(retval == -ECHILD))
2380 retval = path_parentat(&nd, flags, parent);
2381 if (unlikely(retval == -ESTALE))
2382 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2383 if (likely(!retval)) {
2384 *last = nd.last;
2385 *type = nd.last_type;
2386 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2387 } else {
2388 putname(name);
2389 name = ERR_PTR(retval);
2390 }
2391 restore_nameidata();
2392 return name;
2393}
2394
2395/* does lookup, returns the object with parent locked */
2396struct dentry *kern_path_locked(const char *name, struct path *path)
2397{
2398 struct filename *filename;
2399 struct dentry *d;
2400 struct qstr last;
2401 int type;
2402
2403 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2404 &last, &type);
2405 if (IS_ERR(filename))
2406 return ERR_CAST(filename);
2407 if (unlikely(type != LAST_NORM)) {
2408 path_put(path);
2409 putname(filename);
2410 return ERR_PTR(-EINVAL);
2411 }
2412 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2413 d = __lookup_hash(&last, path->dentry, 0);
2414 if (IS_ERR(d)) {
2415 inode_unlock(path->dentry->d_inode);
2416 path_put(path);
2417 }
2418 putname(filename);
2419 return d;
2420}
2421
2422int kern_path(const char *name, unsigned int flags, struct path *path)
2423{
2424 return filename_lookup(AT_FDCWD, getname_kernel(name),
2425 flags, path, NULL);
2426}
2427EXPORT_SYMBOL(kern_path);
2428
2429/**
2430 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2431 * @dentry: pointer to dentry of the base directory
2432 * @mnt: pointer to vfs mount of the base directory
2433 * @name: pointer to file name
2434 * @flags: lookup flags
2435 * @path: pointer to struct path to fill
2436 */
2437int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2438 const char *name, unsigned int flags,
2439 struct path *path)
2440{
2441 struct path root = {.mnt = mnt, .dentry = dentry};
2442 /* the first argument of filename_lookup() is ignored with root */
2443 return filename_lookup(AT_FDCWD, getname_kernel(name),
2444 flags , path, &root);
2445}
2446EXPORT_SYMBOL(vfs_path_lookup);
2447
2448static int lookup_one_len_common(const char *name, struct dentry *base,
2449 int len, struct qstr *this)
2450{
2451 this->name = name;
2452 this->len = len;
2453 this->hash = full_name_hash(base, name, len);
2454 if (!len)
2455 return -EACCES;
2456
2457 if (unlikely(name[0] == '.')) {
2458 if (len < 2 || (len == 2 && name[1] == '.'))
2459 return -EACCES;
2460 }
2461
2462 while (len--) {
2463 unsigned int c = *(const unsigned char *)name++;
2464 if (c == '/' || c == '\0')
2465 return -EACCES;
2466 }
2467 /*
2468 * See if the low-level filesystem might want
2469 * to use its own hash..
2470 */
2471 if (base->d_flags & DCACHE_OP_HASH) {
2472 int err = base->d_op->d_hash(base, this);
2473 if (err < 0)
2474 return err;
2475 }
2476
2477 return inode_permission(base->d_inode, MAY_EXEC);
2478}
2479
2480/**
2481 * try_lookup_one_len - filesystem helper to lookup single pathname component
2482 * @name: pathname component to lookup
2483 * @base: base directory to lookup from
2484 * @len: maximum length @len should be interpreted to
2485 *
2486 * Look up a dentry by name in the dcache, returning NULL if it does not
2487 * currently exist. The function does not try to create a dentry.
2488 *
2489 * Note that this routine is purely a helper for filesystem usage and should
2490 * not be called by generic code.
2491 *
2492 * The caller must hold base->i_mutex.
2493 */
2494struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2495{
2496 struct qstr this;
2497 int err;
2498
2499 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2500
2501 err = lookup_one_len_common(name, base, len, &this);
2502 if (err)
2503 return ERR_PTR(err);
2504
2505 return lookup_dcache(&this, base, 0);
2506}
2507EXPORT_SYMBOL(try_lookup_one_len);
2508
2509/**
2510 * lookup_one_len - filesystem helper to lookup single pathname component
2511 * @name: pathname component to lookup
2512 * @base: base directory to lookup from
2513 * @len: maximum length @len should be interpreted to
2514 *
2515 * Note that this routine is purely a helper for filesystem usage and should
2516 * not be called by generic code.
2517 *
2518 * The caller must hold base->i_mutex.
2519 */
2520struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2521{
2522 struct dentry *dentry;
2523 struct qstr this;
2524 int err;
2525
2526 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2527
2528 err = lookup_one_len_common(name, base, len, &this);
2529 if (err)
2530 return ERR_PTR(err);
2531
2532 dentry = lookup_dcache(&this, base, 0);
2533 return dentry ? dentry : __lookup_slow(&this, base, 0);
2534}
2535EXPORT_SYMBOL(lookup_one_len);
2536
2537/**
2538 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2539 * @name: pathname component to lookup
2540 * @base: base directory to lookup from
2541 * @len: maximum length @len should be interpreted to
2542 *
2543 * Note that this routine is purely a helper for filesystem usage and should
2544 * not be called by generic code.
2545 *
2546 * Unlike lookup_one_len, it should be called without the parent
2547 * i_mutex held, and will take the i_mutex itself if necessary.
2548 */
2549struct dentry *lookup_one_len_unlocked(const char *name,
2550 struct dentry *base, int len)
2551{
2552 struct qstr this;
2553 int err;
2554 struct dentry *ret;
2555
2556 err = lookup_one_len_common(name, base, len, &this);
2557 if (err)
2558 return ERR_PTR(err);
2559
2560 ret = lookup_dcache(&this, base, 0);
2561 if (!ret)
2562 ret = lookup_slow(&this, base, 0);
2563 return ret;
2564}
2565EXPORT_SYMBOL(lookup_one_len_unlocked);
2566
2567#ifdef CONFIG_UNIX98_PTYS
2568int path_pts(struct path *path)
2569{
2570 /* Find something mounted on "pts" in the same directory as
2571 * the input path.
2572 */
2573 struct dentry *child, *parent;
2574 struct qstr this;
2575 int ret;
2576
2577 ret = path_parent_directory(path);
2578 if (ret)
2579 return ret;
2580
2581 parent = path->dentry;
2582 this.name = "pts";
2583 this.len = 3;
2584 child = d_hash_and_lookup(parent, &this);
2585 if (!child)
2586 return -ENOENT;
2587
2588 path->dentry = child;
2589 dput(parent);
2590 follow_mount(path);
2591 return 0;
2592}
2593#endif
2594
2595int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2596 struct path *path, int *empty)
2597{
2598 return filename_lookup(dfd, getname_flags(name, flags, empty),
2599 flags, path, NULL);
2600}
2601EXPORT_SYMBOL(user_path_at_empty);
2602
2603/**
2604 * mountpoint_last - look up last component for umount
2605 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2606 *
2607 * This is a special lookup_last function just for umount. In this case, we
2608 * need to resolve the path without doing any revalidation.
2609 *
2610 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2611 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2612 * in almost all cases, this lookup will be served out of the dcache. The only
2613 * cases where it won't are if nd->last refers to a symlink or the path is
2614 * bogus and it doesn't exist.
2615 *
2616 * Returns:
2617 * -error: if there was an error during lookup. This includes -ENOENT if the
2618 * lookup found a negative dentry.
2619 *
2620 * 0: if we successfully resolved nd->last and found it to not to be a
2621 * symlink that needs to be followed.
2622 *
2623 * 1: if we successfully resolved nd->last and found it to be a symlink
2624 * that needs to be followed.
2625 */
2626static int
2627mountpoint_last(struct nameidata *nd)
2628{
2629 int error = 0;
2630 struct dentry *dir = nd->path.dentry;
2631 struct path path;
2632
2633 /* If we're in rcuwalk, drop out of it to handle last component */
2634 if (nd->flags & LOOKUP_RCU) {
2635 if (unlazy_walk(nd))
2636 return -ECHILD;
2637 }
2638
2639 nd->flags &= ~LOOKUP_PARENT;
2640
2641 if (unlikely(nd->last_type != LAST_NORM)) {
2642 error = handle_dots(nd, nd->last_type);
2643 if (error)
2644 return error;
2645 path.dentry = dget(nd->path.dentry);
2646 } else {
2647 path.dentry = d_lookup(dir, &nd->last);
2648 if (!path.dentry) {
2649 /*
2650 * No cached dentry. Mounted dentries are pinned in the
2651 * cache, so that means that this dentry is probably
2652 * a symlink or the path doesn't actually point
2653 * to a mounted dentry.
2654 */
2655 path.dentry = lookup_slow(&nd->last, dir,
2656 nd->flags | LOOKUP_NO_REVAL);
2657 if (IS_ERR(path.dentry))
2658 return PTR_ERR(path.dentry);
2659 }
2660 }
2661 if (d_is_negative(path.dentry)) {
2662 dput(path.dentry);
2663 return -ENOENT;
2664 }
2665 path.mnt = nd->path.mnt;
2666 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2667}
2668
2669/**
2670 * path_mountpoint - look up a path to be umounted
2671 * @nd: lookup context
2672 * @flags: lookup flags
2673 * @path: pointer to container for result
2674 *
2675 * Look up the given name, but don't attempt to revalidate the last component.
2676 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2677 */
2678static int
2679path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2680{
2681 const char *s = path_init(nd, flags);
2682 int err;
2683
2684 while (!(err = link_path_walk(s, nd)) &&
2685 (err = mountpoint_last(nd)) > 0) {
2686 s = trailing_symlink(nd);
2687 }
2688 if (!err) {
2689 *path = nd->path;
2690 nd->path.mnt = NULL;
2691 nd->path.dentry = NULL;
2692 follow_mount(path);
2693 }
2694 terminate_walk(nd);
2695 return err;
2696}
2697
2698static int
2699filename_mountpoint(int dfd, struct filename *name, struct path *path,
2700 unsigned int flags)
2701{
2702 struct nameidata nd;
2703 int error;
2704 if (IS_ERR(name))
2705 return PTR_ERR(name);
2706 set_nameidata(&nd, dfd, name);
2707 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2708 if (unlikely(error == -ECHILD))
2709 error = path_mountpoint(&nd, flags, path);
2710 if (unlikely(error == -ESTALE))
2711 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2712 if (likely(!error))
2713 audit_inode(name, path->dentry, AUDIT_INODE_NOEVAL);
2714 restore_nameidata();
2715 putname(name);
2716 return error;
2717}
2718
2719/**
2720 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2721 * @dfd: directory file descriptor
2722 * @name: pathname from userland
2723 * @flags: lookup flags
2724 * @path: pointer to container to hold result
2725 *
2726 * A umount is a special case for path walking. We're not actually interested
2727 * in the inode in this situation, and ESTALE errors can be a problem. We
2728 * simply want track down the dentry and vfsmount attached at the mountpoint
2729 * and avoid revalidating the last component.
2730 *
2731 * Returns 0 and populates "path" on success.
2732 */
2733int
2734user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2735 struct path *path)
2736{
2737 return filename_mountpoint(dfd, getname(name), path, flags);
2738}
2739
2740int
2741kern_path_mountpoint(int dfd, const char *name, struct path *path,
2742 unsigned int flags)
2743{
2744 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2745}
2746EXPORT_SYMBOL(kern_path_mountpoint);
2747
2748int __check_sticky(struct inode *dir, struct inode *inode)
2749{
2750 kuid_t fsuid = current_fsuid();
2751
2752 if (uid_eq(inode->i_uid, fsuid))
2753 return 0;
2754 if (uid_eq(dir->i_uid, fsuid))
2755 return 0;
2756 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2757}
2758EXPORT_SYMBOL(__check_sticky);
2759
2760/*
2761 * Check whether we can remove a link victim from directory dir, check
2762 * whether the type of victim is right.
2763 * 1. We can't do it if dir is read-only (done in permission())
2764 * 2. We should have write and exec permissions on dir
2765 * 3. We can't remove anything from append-only dir
2766 * 4. We can't do anything with immutable dir (done in permission())
2767 * 5. If the sticky bit on dir is set we should either
2768 * a. be owner of dir, or
2769 * b. be owner of victim, or
2770 * c. have CAP_FOWNER capability
2771 * 6. If the victim is append-only or immutable we can't do antyhing with
2772 * links pointing to it.
2773 * 7. If the victim has an unknown uid or gid we can't change the inode.
2774 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2775 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2776 * 10. We can't remove a root or mountpoint.
2777 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2778 * nfs_async_unlink().
2779 */
2780static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2781{
2782 struct inode *inode = d_backing_inode(victim);
2783 int error;
2784
2785 if (d_is_negative(victim))
2786 return -ENOENT;
2787 BUG_ON(!inode);
2788
2789 BUG_ON(victim->d_parent->d_inode != dir);
2790
2791 /* Inode writeback is not safe when the uid or gid are invalid. */
2792 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2793 return -EOVERFLOW;
2794
2795 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2796
2797 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2798 if (error)
2799 return error;
2800 if (IS_APPEND(dir))
2801 return -EPERM;
2802
2803 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2804 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2805 return -EPERM;
2806 if (isdir) {
2807 if (!d_is_dir(victim))
2808 return -ENOTDIR;
2809 if (IS_ROOT(victim))
2810 return -EBUSY;
2811 } else if (d_is_dir(victim))
2812 return -EISDIR;
2813 if (IS_DEADDIR(dir))
2814 return -ENOENT;
2815 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2816 return -EBUSY;
2817 return 0;
2818}
2819
2820/* Check whether we can create an object with dentry child in directory
2821 * dir.
2822 * 1. We can't do it if child already exists (open has special treatment for
2823 * this case, but since we are inlined it's OK)
2824 * 2. We can't do it if dir is read-only (done in permission())
2825 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2826 * 4. We should have write and exec permissions on dir
2827 * 5. We can't do it if dir is immutable (done in permission())
2828 */
2829static inline int may_create(struct inode *dir, struct dentry *child)
2830{
2831 struct user_namespace *s_user_ns;
2832 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2833 if (child->d_inode)
2834 return -EEXIST;
2835 if (IS_DEADDIR(dir))
2836 return -ENOENT;
2837 s_user_ns = dir->i_sb->s_user_ns;
2838 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2839 !kgid_has_mapping(s_user_ns, current_fsgid()))
2840 return -EOVERFLOW;
2841 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2842}
2843
2844/*
2845 * p1 and p2 should be directories on the same fs.
2846 */
2847struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2848{
2849 struct dentry *p;
2850
2851 if (p1 == p2) {
2852 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2853 return NULL;
2854 }
2855
2856 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2857
2858 p = d_ancestor(p2, p1);
2859 if (p) {
2860 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2861 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2862 return p;
2863 }
2864
2865 p = d_ancestor(p1, p2);
2866 if (p) {
2867 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2868 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2869 return p;
2870 }
2871
2872 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2873 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2874 return NULL;
2875}
2876EXPORT_SYMBOL(lock_rename);
2877
2878void unlock_rename(struct dentry *p1, struct dentry *p2)
2879{
2880 inode_unlock(p1->d_inode);
2881 if (p1 != p2) {
2882 inode_unlock(p2->d_inode);
2883 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2884 }
2885}
2886EXPORT_SYMBOL(unlock_rename);
2887
2888int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2889 bool want_excl)
2890{
2891 int error = may_create(dir, dentry);
2892 if (error)
2893 return error;
2894
2895 if (!dir->i_op->create)
2896 return -EACCES; /* shouldn't it be ENOSYS? */
2897 mode &= S_IALLUGO;
2898 mode |= S_IFREG;
2899 error = security_inode_create(dir, dentry, mode);
2900 if (error)
2901 return error;
2902 error = dir->i_op->create(dir, dentry, mode, want_excl);
2903 if (!error)
2904 fsnotify_create(dir, dentry);
2905 return error;
2906}
2907EXPORT_SYMBOL(vfs_create);
2908
2909int vfs_mkobj(struct dentry *dentry, umode_t mode,
2910 int (*f)(struct dentry *, umode_t, void *),
2911 void *arg)
2912{
2913 struct inode *dir = dentry->d_parent->d_inode;
2914 int error = may_create(dir, dentry);
2915 if (error)
2916 return error;
2917
2918 mode &= S_IALLUGO;
2919 mode |= S_IFREG;
2920 error = security_inode_create(dir, dentry, mode);
2921 if (error)
2922 return error;
2923 error = f(dentry, mode, arg);
2924 if (!error)
2925 fsnotify_create(dir, dentry);
2926 return error;
2927}
2928EXPORT_SYMBOL(vfs_mkobj);
2929
2930bool may_open_dev(const struct path *path)
2931{
2932 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2933 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2934}
2935
2936static int may_open(const struct path *path, int acc_mode, int flag)
2937{
2938 struct dentry *dentry = path->dentry;
2939 struct inode *inode = dentry->d_inode;
2940 int error;
2941
2942 if (!inode)
2943 return -ENOENT;
2944
2945 switch (inode->i_mode & S_IFMT) {
2946 case S_IFLNK:
2947 return -ELOOP;
2948 case S_IFDIR:
2949 if (acc_mode & MAY_WRITE)
2950 return -EISDIR;
2951 break;
2952 case S_IFBLK:
2953 case S_IFCHR:
2954 if (!may_open_dev(path))
2955 return -EACCES;
2956 /*FALLTHRU*/
2957 case S_IFIFO:
2958 case S_IFSOCK:
2959 flag &= ~O_TRUNC;
2960 break;
2961 }
2962
2963 error = inode_permission(inode, MAY_OPEN | acc_mode);
2964 if (error)
2965 return error;
2966
2967 /*
2968 * An append-only file must be opened in append mode for writing.
2969 */
2970 if (IS_APPEND(inode)) {
2971 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2972 return -EPERM;
2973 if (flag & O_TRUNC)
2974 return -EPERM;
2975 }
2976
2977 /* O_NOATIME can only be set by the owner or superuser */
2978 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2979 return -EPERM;
2980
2981 return 0;
2982}
2983
2984static int handle_truncate(struct file *filp)
2985{
2986 const struct path *path = &filp->f_path;
2987 struct inode *inode = path->dentry->d_inode;
2988 int error = get_write_access(inode);
2989 if (error)
2990 return error;
2991 /*
2992 * Refuse to truncate files with mandatory locks held on them.
2993 */
2994 error = locks_verify_locked(filp);
2995 if (!error)
2996 error = security_path_truncate(path);
2997 if (!error) {
2998 error = do_truncate(path->dentry, 0,
2999 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3000 filp);
3001 }
3002 put_write_access(inode);
3003 return error;
3004}
3005
3006static inline int open_to_namei_flags(int flag)
3007{
3008 if ((flag & O_ACCMODE) == 3)
3009 flag--;
3010 return flag;
3011}
3012
3013static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
3014{
3015 struct user_namespace *s_user_ns;
3016 int error = security_path_mknod(dir, dentry, mode, 0);
3017 if (error)
3018 return error;
3019
3020 s_user_ns = dir->dentry->d_sb->s_user_ns;
3021 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
3022 !kgid_has_mapping(s_user_ns, current_fsgid()))
3023 return -EOVERFLOW;
3024
3025 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
3026 if (error)
3027 return error;
3028
3029 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3030}
3031
3032/*
3033 * Attempt to atomically look up, create and open a file from a negative
3034 * dentry.
3035 *
3036 * Returns 0 if successful. The file will have been created and attached to
3037 * @file by the filesystem calling finish_open().
3038 *
3039 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3040 * be set. The caller will need to perform the open themselves. @path will
3041 * have been updated to point to the new dentry. This may be negative.
3042 *
3043 * Returns an error code otherwise.
3044 */
3045static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3046 struct path *path, struct file *file,
3047 const struct open_flags *op,
3048 int open_flag, umode_t mode)
3049{
3050 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3051 struct inode *dir = nd->path.dentry->d_inode;
3052 int error;
3053
3054 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3055 open_flag &= ~O_TRUNC;
3056
3057 if (nd->flags & LOOKUP_DIRECTORY)
3058 open_flag |= O_DIRECTORY;
3059
3060 file->f_path.dentry = DENTRY_NOT_SET;
3061 file->f_path.mnt = nd->path.mnt;
3062 error = dir->i_op->atomic_open(dir, dentry, file,
3063 open_to_namei_flags(open_flag), mode);
3064 d_lookup_done(dentry);
3065 if (!error) {
3066 if (file->f_mode & FMODE_OPENED) {
3067 /*
3068 * We didn't have the inode before the open, so check open
3069 * permission here.
3070 */
3071 int acc_mode = op->acc_mode;
3072 if (file->f_mode & FMODE_CREATED) {
3073 WARN_ON(!(open_flag & O_CREAT));
3074 fsnotify_create(dir, dentry);
3075 acc_mode = 0;
3076 }
3077 error = may_open(&file->f_path, acc_mode, open_flag);
3078 if (WARN_ON(error > 0))
3079 error = -EINVAL;
3080 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3081 error = -EIO;
3082 } else {
3083 if (file->f_path.dentry) {
3084 dput(dentry);
3085 dentry = file->f_path.dentry;
3086 }
3087 if (file->f_mode & FMODE_CREATED)
3088 fsnotify_create(dir, dentry);
3089 if (unlikely(d_is_negative(dentry))) {
3090 error = -ENOENT;
3091 } else {
3092 path->dentry = dentry;
3093 path->mnt = nd->path.mnt;
3094 return 0;
3095 }
3096 }
3097 }
3098 dput(dentry);
3099 return error;
3100}
3101
3102/*
3103 * Look up and maybe create and open the last component.
3104 *
3105 * Must be called with parent locked (exclusive in O_CREAT case).
3106 *
3107 * Returns 0 on success, that is, if
3108 * the file was successfully atomically created (if necessary) and opened, or
3109 * the file was not completely opened at this time, though lookups and
3110 * creations were performed.
3111 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3112 * In the latter case dentry returned in @path might be negative if O_CREAT
3113 * hadn't been specified.
3114 *
3115 * An error code is returned on failure.
3116 */
3117static int lookup_open(struct nameidata *nd, struct path *path,
3118 struct file *file,
3119 const struct open_flags *op,
3120 bool got_write)
3121{
3122 struct dentry *dir = nd->path.dentry;
3123 struct inode *dir_inode = dir->d_inode;
3124 int open_flag = op->open_flag;
3125 struct dentry *dentry;
3126 int error, create_error = 0;
3127 umode_t mode = op->mode;
3128 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3129
3130 if (unlikely(IS_DEADDIR(dir_inode)))
3131 return -ENOENT;
3132
3133 file->f_mode &= ~FMODE_CREATED;
3134 dentry = d_lookup(dir, &nd->last);
3135 for (;;) {
3136 if (!dentry) {
3137 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3138 if (IS_ERR(dentry))
3139 return PTR_ERR(dentry);
3140 }
3141 if (d_in_lookup(dentry))
3142 break;
3143
3144 error = d_revalidate(dentry, nd->flags);
3145 if (likely(error > 0))
3146 break;
3147 if (error)
3148 goto out_dput;
3149 d_invalidate(dentry);
3150 dput(dentry);
3151 dentry = NULL;
3152 }
3153 if (dentry->d_inode) {
3154 /* Cached positive dentry: will open in f_op->open */
3155 goto out_no_open;
3156 }
3157
3158 /*
3159 * Checking write permission is tricky, bacuse we don't know if we are
3160 * going to actually need it: O_CREAT opens should work as long as the
3161 * file exists. But checking existence breaks atomicity. The trick is
3162 * to check access and if not granted clear O_CREAT from the flags.
3163 *
3164 * Another problem is returing the "right" error value (e.g. for an
3165 * O_EXCL open we want to return EEXIST not EROFS).
3166 */
3167 if (open_flag & O_CREAT) {
3168 if (!IS_POSIXACL(dir->d_inode))
3169 mode &= ~current_umask();
3170 if (unlikely(!got_write)) {
3171 create_error = -EROFS;
3172 open_flag &= ~O_CREAT;
3173 if (open_flag & (O_EXCL | O_TRUNC))
3174 goto no_open;
3175 /* No side effects, safe to clear O_CREAT */
3176 } else {
3177 create_error = may_o_create(&nd->path, dentry, mode);
3178 if (create_error) {
3179 open_flag &= ~O_CREAT;
3180 if (open_flag & O_EXCL)
3181 goto no_open;
3182 }
3183 }
3184 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3185 unlikely(!got_write)) {
3186 /*
3187 * No O_CREATE -> atomicity not a requirement -> fall
3188 * back to lookup + open
3189 */
3190 goto no_open;
3191 }
3192
3193 if (dir_inode->i_op->atomic_open) {
3194 error = atomic_open(nd, dentry, path, file, op, open_flag,
3195 mode);
3196 if (unlikely(error == -ENOENT) && create_error)
3197 error = create_error;
3198 return error;
3199 }
3200
3201no_open:
3202 if (d_in_lookup(dentry)) {
3203 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3204 nd->flags);
3205 d_lookup_done(dentry);
3206 if (unlikely(res)) {
3207 if (IS_ERR(res)) {
3208 error = PTR_ERR(res);
3209 goto out_dput;
3210 }
3211 dput(dentry);
3212 dentry = res;
3213 }
3214 }
3215
3216 /* Negative dentry, just create the file */
3217 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3218 file->f_mode |= FMODE_CREATED;
3219 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3220 if (!dir_inode->i_op->create) {
3221 error = -EACCES;
3222 goto out_dput;
3223 }
3224 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3225 open_flag & O_EXCL);
3226 if (error)
3227 goto out_dput;
3228 fsnotify_create(dir_inode, dentry);
3229 }
3230 if (unlikely(create_error) && !dentry->d_inode) {
3231 error = create_error;
3232 goto out_dput;
3233 }
3234out_no_open:
3235 path->dentry = dentry;
3236 path->mnt = nd->path.mnt;
3237 return 0;
3238
3239out_dput:
3240 dput(dentry);
3241 return error;
3242}
3243
3244/*
3245 * Handle the last step of open()
3246 */
3247static int do_last(struct nameidata *nd,
3248 struct file *file, const struct open_flags *op)
3249{
3250 struct dentry *dir = nd->path.dentry;
3251 int open_flag = op->open_flag;
3252 bool will_truncate = (open_flag & O_TRUNC) != 0;
3253 bool got_write = false;
3254 int acc_mode = op->acc_mode;
3255 unsigned seq;
3256 struct inode *inode;
3257 struct path path;
3258 int error;
3259
3260 nd->flags &= ~LOOKUP_PARENT;
3261 nd->flags |= op->intent;
3262
3263 if (nd->last_type != LAST_NORM) {
3264 error = handle_dots(nd, nd->last_type);
3265 if (unlikely(error))
3266 return error;
3267 goto finish_open;
3268 }
3269
3270 if (!(open_flag & O_CREAT)) {
3271 if (nd->last.name[nd->last.len])
3272 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3273 /* we _can_ be in RCU mode here */
3274 error = lookup_fast(nd, &path, &inode, &seq);
3275 if (likely(error > 0))
3276 goto finish_lookup;
3277
3278 if (error < 0)
3279 return error;
3280
3281 BUG_ON(nd->inode != dir->d_inode);
3282 BUG_ON(nd->flags & LOOKUP_RCU);
3283 } else {
3284 /* create side of things */
3285 /*
3286 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3287 * has been cleared when we got to the last component we are
3288 * about to look up
3289 */
3290 error = complete_walk(nd);
3291 if (error)
3292 return error;
3293
3294 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3295 /* trailing slashes? */
3296 if (unlikely(nd->last.name[nd->last.len]))
3297 return -EISDIR;
3298 }
3299
3300 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3301 error = mnt_want_write(nd->path.mnt);
3302 if (!error)
3303 got_write = true;
3304 /*
3305 * do _not_ fail yet - we might not need that or fail with
3306 * a different error; let lookup_open() decide; we'll be
3307 * dropping this one anyway.
3308 */
3309 }
3310 if (open_flag & O_CREAT)
3311 inode_lock(dir->d_inode);
3312 else
3313 inode_lock_shared(dir->d_inode);
3314 error = lookup_open(nd, &path, file, op, got_write);
3315 if (open_flag & O_CREAT)
3316 inode_unlock(dir->d_inode);
3317 else
3318 inode_unlock_shared(dir->d_inode);
3319
3320 if (error)
3321 goto out;
3322
3323 if (file->f_mode & FMODE_OPENED) {
3324 if ((file->f_mode & FMODE_CREATED) ||
3325 !S_ISREG(file_inode(file)->i_mode))
3326 will_truncate = false;
3327
3328 audit_inode(nd->name, file->f_path.dentry, 0);
3329 goto opened;
3330 }
3331
3332 if (file->f_mode & FMODE_CREATED) {
3333 /* Don't check for write permission, don't truncate */
3334 open_flag &= ~O_TRUNC;
3335 will_truncate = false;
3336 acc_mode = 0;
3337 path_to_nameidata(&path, nd);
3338 goto finish_open_created;
3339 }
3340
3341 /*
3342 * If atomic_open() acquired write access it is dropped now due to
3343 * possible mount and symlink following (this might be optimized away if
3344 * necessary...)
3345 */
3346 if (got_write) {
3347 mnt_drop_write(nd->path.mnt);
3348 got_write = false;
3349 }
3350
3351 error = follow_managed(&path, nd);
3352 if (unlikely(error < 0))
3353 return error;
3354
3355 if (unlikely(d_is_negative(path.dentry))) {
3356 path_to_nameidata(&path, nd);
3357 return -ENOENT;
3358 }
3359
3360 /*
3361 * create/update audit record if it already exists.
3362 */
3363 audit_inode(nd->name, path.dentry, 0);
3364
3365 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3366 path_to_nameidata(&path, nd);
3367 return -EEXIST;
3368 }
3369
3370 seq = 0; /* out of RCU mode, so the value doesn't matter */
3371 inode = d_backing_inode(path.dentry);
3372finish_lookup:
3373 error = step_into(nd, &path, 0, inode, seq);
3374 if (unlikely(error))
3375 return error;
3376finish_open:
3377 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3378 error = complete_walk(nd);
3379 if (error)
3380 return error;
3381 audit_inode(nd->name, nd->path.dentry, 0);
3382 if (open_flag & O_CREAT) {
3383 error = -EISDIR;
3384 if (d_is_dir(nd->path.dentry))
3385 goto out;
3386 error = may_create_in_sticky(dir,
3387 d_backing_inode(nd->path.dentry));
3388 if (unlikely(error))
3389 goto out;
3390 }
3391 error = -ENOTDIR;
3392 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3393 goto out;
3394 if (!d_is_reg(nd->path.dentry))
3395 will_truncate = false;
3396
3397 if (will_truncate) {
3398 error = mnt_want_write(nd->path.mnt);
3399 if (error)
3400 goto out;
3401 got_write = true;
3402 }
3403finish_open_created:
3404 error = may_open(&nd->path, acc_mode, open_flag);
3405 if (error)
3406 goto out;
3407 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
3408 error = vfs_open(&nd->path, file);
3409 if (error)
3410 goto out;
3411opened:
3412 error = ima_file_check(file, op->acc_mode);
3413 if (!error && will_truncate)
3414 error = handle_truncate(file);
3415out:
3416 if (unlikely(error > 0)) {
3417 WARN_ON(1);
3418 error = -EINVAL;
3419 }
3420 if (got_write)
3421 mnt_drop_write(nd->path.mnt);
3422 return error;
3423}
3424
3425struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3426{
3427 struct dentry *child = NULL;
3428 struct inode *dir = dentry->d_inode;
3429 struct inode *inode;
3430 int error;
3431
3432 /* we want directory to be writable */
3433 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3434 if (error)
3435 goto out_err;
3436 error = -EOPNOTSUPP;
3437 if (!dir->i_op->tmpfile)
3438 goto out_err;
3439 error = -ENOMEM;
3440 child = d_alloc(dentry, &slash_name);
3441 if (unlikely(!child))
3442 goto out_err;
3443 error = dir->i_op->tmpfile(dir, child, mode);
3444 if (error)
3445 goto out_err;
3446 error = -ENOENT;
3447 inode = child->d_inode;
3448 if (unlikely(!inode))
3449 goto out_err;
3450 if (!(open_flag & O_EXCL)) {
3451 spin_lock(&inode->i_lock);
3452 inode->i_state |= I_LINKABLE;
3453 spin_unlock(&inode->i_lock);
3454 }
3455 ima_post_create_tmpfile(inode);
3456 return child;
3457
3458out_err:
3459 dput(child);
3460 return ERR_PTR(error);
3461}
3462EXPORT_SYMBOL(vfs_tmpfile);
3463
3464static int do_tmpfile(struct nameidata *nd, unsigned flags,
3465 const struct open_flags *op,
3466 struct file *file)
3467{
3468 struct dentry *child;
3469 struct path path;
3470 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3471 if (unlikely(error))
3472 return error;
3473 error = mnt_want_write(path.mnt);
3474 if (unlikely(error))
3475 goto out;
3476 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3477 error = PTR_ERR(child);
3478 if (IS_ERR(child))
3479 goto out2;
3480 dput(path.dentry);
3481 path.dentry = child;
3482 audit_inode(nd->name, child, 0);
3483 /* Don't check for other permissions, the inode was just created */
3484 error = may_open(&path, 0, op->open_flag);
3485 if (error)
3486 goto out2;
3487 file->f_path.mnt = path.mnt;
3488 error = finish_open(file, child, NULL);
3489out2:
3490 mnt_drop_write(path.mnt);
3491out:
3492 path_put(&path);
3493 return error;
3494}
3495
3496static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3497{
3498 struct path path;
3499 int error = path_lookupat(nd, flags, &path);
3500 if (!error) {
3501 audit_inode(nd->name, path.dentry, 0);
3502 error = vfs_open(&path, file);
3503 path_put(&path);
3504 }
3505 return error;
3506}
3507
3508static struct file *path_openat(struct nameidata *nd,
3509 const struct open_flags *op, unsigned flags)
3510{
3511 struct file *file;
3512 int error;
3513
3514 file = alloc_empty_file(op->open_flag, current_cred());
3515 if (IS_ERR(file))
3516 return file;
3517
3518 if (unlikely(file->f_flags & __O_TMPFILE)) {
3519 error = do_tmpfile(nd, flags, op, file);
3520 } else if (unlikely(file->f_flags & O_PATH)) {
3521 error = do_o_path(nd, flags, file);
3522 } else {
3523 const char *s = path_init(nd, flags);
3524 while (!(error = link_path_walk(s, nd)) &&
3525 (error = do_last(nd, file, op)) > 0) {
3526 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3527 s = trailing_symlink(nd);
3528 }
3529 terminate_walk(nd);
3530 }
3531 if (likely(!error)) {
3532 if (likely(file->f_mode & FMODE_OPENED))
3533 return file;
3534 WARN_ON(1);
3535 error = -EINVAL;
3536 }
3537 fput(file);
3538 if (error == -EOPENSTALE) {
3539 if (flags & LOOKUP_RCU)
3540 error = -ECHILD;
3541 else
3542 error = -ESTALE;
3543 }
3544 return ERR_PTR(error);
3545}
3546
3547struct file *do_filp_open(int dfd, struct filename *pathname,
3548 const struct open_flags *op)
3549{
3550 struct nameidata nd;
3551 int flags = op->lookup_flags;
3552 struct file *filp;
3553
3554 set_nameidata(&nd, dfd, pathname);
3555 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3556 if (unlikely(filp == ERR_PTR(-ECHILD)))
3557 filp = path_openat(&nd, op, flags);
3558 if (unlikely(filp == ERR_PTR(-ESTALE)))
3559 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3560 restore_nameidata();
3561 return filp;
3562}
3563
3564struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3565 const char *name, const struct open_flags *op)
3566{
3567 struct nameidata nd;
3568 struct file *file;
3569 struct filename *filename;
3570 int flags = op->lookup_flags | LOOKUP_ROOT;
3571
3572 nd.root.mnt = mnt;
3573 nd.root.dentry = dentry;
3574
3575 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3576 return ERR_PTR(-ELOOP);
3577
3578 filename = getname_kernel(name);
3579 if (IS_ERR(filename))
3580 return ERR_CAST(filename);
3581
3582 set_nameidata(&nd, -1, filename);
3583 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3584 if (unlikely(file == ERR_PTR(-ECHILD)))
3585 file = path_openat(&nd, op, flags);
3586 if (unlikely(file == ERR_PTR(-ESTALE)))
3587 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3588 restore_nameidata();
3589 putname(filename);
3590 return file;
3591}
3592
3593static struct dentry *filename_create(int dfd, struct filename *name,
3594 struct path *path, unsigned int lookup_flags)
3595{
3596 struct dentry *dentry = ERR_PTR(-EEXIST);
3597 struct qstr last;
3598 int type;
3599 int err2;
3600 int error;
3601 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3602
3603 /*
3604 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3605 * other flags passed in are ignored!
3606 */
3607 lookup_flags &= LOOKUP_REVAL;
3608
3609 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3610 if (IS_ERR(name))
3611 return ERR_CAST(name);
3612
3613 /*
3614 * Yucky last component or no last component at all?
3615 * (foo/., foo/.., /////)
3616 */
3617 if (unlikely(type != LAST_NORM))
3618 goto out;
3619
3620 /* don't fail immediately if it's r/o, at least try to report other errors */
3621 err2 = mnt_want_write(path->mnt);
3622 /*
3623 * Do the final lookup.
3624 */
3625 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3626 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3627 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3628 if (IS_ERR(dentry))
3629 goto unlock;
3630
3631 error = -EEXIST;
3632 if (d_is_positive(dentry))
3633 goto fail;
3634
3635 /*
3636 * Special case - lookup gave negative, but... we had foo/bar/
3637 * From the vfs_mknod() POV we just have a negative dentry -
3638 * all is fine. Let's be bastards - you had / on the end, you've
3639 * been asking for (non-existent) directory. -ENOENT for you.
3640 */
3641 if (unlikely(!is_dir && last.name[last.len])) {
3642 error = -ENOENT;
3643 goto fail;
3644 }
3645 if (unlikely(err2)) {
3646 error = err2;
3647 goto fail;
3648 }
3649 putname(name);
3650 return dentry;
3651fail:
3652 dput(dentry);
3653 dentry = ERR_PTR(error);
3654unlock:
3655 inode_unlock(path->dentry->d_inode);
3656 if (!err2)
3657 mnt_drop_write(path->mnt);
3658out:
3659 path_put(path);
3660 putname(name);
3661 return dentry;
3662}
3663
3664struct dentry *kern_path_create(int dfd, const char *pathname,
3665 struct path *path, unsigned int lookup_flags)
3666{
3667 return filename_create(dfd, getname_kernel(pathname),
3668 path, lookup_flags);
3669}
3670EXPORT_SYMBOL(kern_path_create);
3671
3672void done_path_create(struct path *path, struct dentry *dentry)
3673{
3674 dput(dentry);
3675 inode_unlock(path->dentry->d_inode);
3676 mnt_drop_write(path->mnt);
3677 path_put(path);
3678}
3679EXPORT_SYMBOL(done_path_create);
3680
3681inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3682 struct path *path, unsigned int lookup_flags)
3683{
3684 return filename_create(dfd, getname(pathname), path, lookup_flags);
3685}
3686EXPORT_SYMBOL(user_path_create);
3687
3688int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3689{
3690 int error = may_create(dir, dentry);
3691
3692 if (error)
3693 return error;
3694
3695 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3696 return -EPERM;
3697
3698 if (!dir->i_op->mknod)
3699 return -EPERM;
3700
3701 error = devcgroup_inode_mknod(mode, dev);
3702 if (error)
3703 return error;
3704
3705 error = security_inode_mknod(dir, dentry, mode, dev);
3706 if (error)
3707 return error;
3708
3709 error = dir->i_op->mknod(dir, dentry, mode, dev);
3710 if (!error)
3711 fsnotify_create(dir, dentry);
3712 return error;
3713}
3714EXPORT_SYMBOL(vfs_mknod);
3715
3716static int may_mknod(umode_t mode)
3717{
3718 switch (mode & S_IFMT) {
3719 case S_IFREG:
3720 case S_IFCHR:
3721 case S_IFBLK:
3722 case S_IFIFO:
3723 case S_IFSOCK:
3724 case 0: /* zero mode translates to S_IFREG */
3725 return 0;
3726 case S_IFDIR:
3727 return -EPERM;
3728 default:
3729 return -EINVAL;
3730 }
3731}
3732
3733long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3734 unsigned int dev)
3735{
3736 struct dentry *dentry;
3737 struct path path;
3738 int error;
3739 unsigned int lookup_flags = 0;
3740
3741 error = may_mknod(mode);
3742 if (error)
3743 return error;
3744retry:
3745 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3746 if (IS_ERR(dentry))
3747 return PTR_ERR(dentry);
3748
3749 if (!IS_POSIXACL(path.dentry->d_inode))
3750 mode &= ~current_umask();
3751 error = security_path_mknod(&path, dentry, mode, dev);
3752 if (error)
3753 goto out;
3754 switch (mode & S_IFMT) {
3755 case 0: case S_IFREG:
3756 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3757 if (!error)
3758 ima_post_path_mknod(dentry);
3759 break;
3760 case S_IFCHR: case S_IFBLK:
3761 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3762 new_decode_dev(dev));
3763 break;
3764 case S_IFIFO: case S_IFSOCK:
3765 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3766 break;
3767 }
3768out:
3769 done_path_create(&path, dentry);
3770 if (retry_estale(error, lookup_flags)) {
3771 lookup_flags |= LOOKUP_REVAL;
3772 goto retry;
3773 }
3774 return error;
3775}
3776
3777SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3778 unsigned int, dev)
3779{
3780 return do_mknodat(dfd, filename, mode, dev);
3781}
3782
3783SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3784{
3785 return do_mknodat(AT_FDCWD, filename, mode, dev);
3786}
3787
3788int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3789{
3790 int error = may_create(dir, dentry);
3791 unsigned max_links = dir->i_sb->s_max_links;
3792
3793 if (error)
3794 return error;
3795
3796 if (!dir->i_op->mkdir)
3797 return -EPERM;
3798
3799 mode &= (S_IRWXUGO|S_ISVTX);
3800 error = security_inode_mkdir(dir, dentry, mode);
3801 if (error)
3802 return error;
3803
3804 if (max_links && dir->i_nlink >= max_links)
3805 return -EMLINK;
3806
3807 error = dir->i_op->mkdir(dir, dentry, mode);
3808 if (!error)
3809 fsnotify_mkdir(dir, dentry);
3810 return error;
3811}
3812EXPORT_SYMBOL(vfs_mkdir);
3813
3814long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3815{
3816 struct dentry *dentry;
3817 struct path path;
3818 int error;
3819 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3820
3821retry:
3822 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3823 if (IS_ERR(dentry))
3824 return PTR_ERR(dentry);
3825
3826 if (!IS_POSIXACL(path.dentry->d_inode))
3827 mode &= ~current_umask();
3828 error = security_path_mkdir(&path, dentry, mode);
3829 if (!error)
3830 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3831 done_path_create(&path, dentry);
3832 if (retry_estale(error, lookup_flags)) {
3833 lookup_flags |= LOOKUP_REVAL;
3834 goto retry;
3835 }
3836 return error;
3837}
3838
3839SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3840{
3841 return do_mkdirat(dfd, pathname, mode);
3842}
3843
3844SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3845{
3846 return do_mkdirat(AT_FDCWD, pathname, mode);
3847}
3848
3849int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3850{
3851 int error = may_delete(dir, dentry, 1);
3852
3853 if (error)
3854 return error;
3855
3856 if (!dir->i_op->rmdir)
3857 return -EPERM;
3858
3859 dget(dentry);
3860 inode_lock(dentry->d_inode);
3861
3862 error = -EBUSY;
3863 if (is_local_mountpoint(dentry))
3864 goto out;
3865
3866 error = security_inode_rmdir(dir, dentry);
3867 if (error)
3868 goto out;
3869
3870 error = dir->i_op->rmdir(dir, dentry);
3871 if (error)
3872 goto out;
3873
3874 shrink_dcache_parent(dentry);
3875 dentry->d_inode->i_flags |= S_DEAD;
3876 dont_mount(dentry);
3877 detach_mounts(dentry);
3878 fsnotify_rmdir(dir, dentry);
3879
3880out:
3881 inode_unlock(dentry->d_inode);
3882 dput(dentry);
3883 if (!error)
3884 d_delete(dentry);
3885 return error;
3886}
3887EXPORT_SYMBOL(vfs_rmdir);
3888
3889long do_rmdir(int dfd, const char __user *pathname)
3890{
3891 int error = 0;
3892 struct filename *name;
3893 struct dentry *dentry;
3894 struct path path;
3895 struct qstr last;
3896 int type;
3897 unsigned int lookup_flags = 0;
3898retry:
3899 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3900 &path, &last, &type);
3901 if (IS_ERR(name))
3902 return PTR_ERR(name);
3903
3904 switch (type) {
3905 case LAST_DOTDOT:
3906 error = -ENOTEMPTY;
3907 goto exit1;
3908 case LAST_DOT:
3909 error = -EINVAL;
3910 goto exit1;
3911 case LAST_ROOT:
3912 error = -EBUSY;
3913 goto exit1;
3914 }
3915
3916 error = mnt_want_write(path.mnt);
3917 if (error)
3918 goto exit1;
3919
3920 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3921 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3922 error = PTR_ERR(dentry);
3923 if (IS_ERR(dentry))
3924 goto exit2;
3925 if (!dentry->d_inode) {
3926 error = -ENOENT;
3927 goto exit3;
3928 }
3929 error = security_path_rmdir(&path, dentry);
3930 if (error)
3931 goto exit3;
3932 error = vfs_rmdir(path.dentry->d_inode, dentry);
3933exit3:
3934 dput(dentry);
3935exit2:
3936 inode_unlock(path.dentry->d_inode);
3937 mnt_drop_write(path.mnt);
3938exit1:
3939 path_put(&path);
3940 putname(name);
3941 if (retry_estale(error, lookup_flags)) {
3942 lookup_flags |= LOOKUP_REVAL;
3943 goto retry;
3944 }
3945 return error;
3946}
3947
3948SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3949{
3950 return do_rmdir(AT_FDCWD, pathname);
3951}
3952
3953/**
3954 * vfs_unlink - unlink a filesystem object
3955 * @dir: parent directory
3956 * @dentry: victim
3957 * @delegated_inode: returns victim inode, if the inode is delegated.
3958 *
3959 * The caller must hold dir->i_mutex.
3960 *
3961 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3962 * return a reference to the inode in delegated_inode. The caller
3963 * should then break the delegation on that inode and retry. Because
3964 * breaking a delegation may take a long time, the caller should drop
3965 * dir->i_mutex before doing so.
3966 *
3967 * Alternatively, a caller may pass NULL for delegated_inode. This may
3968 * be appropriate for callers that expect the underlying filesystem not
3969 * to be NFS exported.
3970 */
3971int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3972{
3973 struct inode *target = dentry->d_inode;
3974 int error = may_delete(dir, dentry, 0);
3975
3976 if (error)
3977 return error;
3978
3979 if (!dir->i_op->unlink)
3980 return -EPERM;
3981
3982 inode_lock(target);
3983 if (is_local_mountpoint(dentry))
3984 error = -EBUSY;
3985 else {
3986 error = security_inode_unlink(dir, dentry);
3987 if (!error) {
3988 error = try_break_deleg(target, delegated_inode);
3989 if (error)
3990 goto out;
3991 error = dir->i_op->unlink(dir, dentry);
3992 if (!error) {
3993 dont_mount(dentry);
3994 detach_mounts(dentry);
3995 fsnotify_unlink(dir, dentry);
3996 }
3997 }
3998 }
3999out:
4000 inode_unlock(target);
4001
4002 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4003 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
4004 fsnotify_link_count(target);
4005 d_delete(dentry);
4006 }
4007
4008 return error;
4009}
4010EXPORT_SYMBOL(vfs_unlink);
4011
4012/*
4013 * Make sure that the actual truncation of the file will occur outside its
4014 * directory's i_mutex. Truncate can take a long time if there is a lot of
4015 * writeout happening, and we don't want to prevent access to the directory
4016 * while waiting on the I/O.
4017 */
4018long do_unlinkat(int dfd, struct filename *name)
4019{
4020 int error;
4021 struct dentry *dentry;
4022 struct path path;
4023 struct qstr last;
4024 int type;
4025 struct inode *inode = NULL;
4026 struct inode *delegated_inode = NULL;
4027 unsigned int lookup_flags = 0;
4028retry:
4029 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4030 if (IS_ERR(name))
4031 return PTR_ERR(name);
4032
4033 error = -EISDIR;
4034 if (type != LAST_NORM)
4035 goto exit1;
4036
4037 error = mnt_want_write(path.mnt);
4038 if (error)
4039 goto exit1;
4040retry_deleg:
4041 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4042 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4043 error = PTR_ERR(dentry);
4044 if (!IS_ERR(dentry)) {
4045 /* Why not before? Because we want correct error value */
4046 if (last.name[last.len])
4047 goto slashes;
4048 inode = dentry->d_inode;
4049 if (d_is_negative(dentry))
4050 goto slashes;
4051 ihold(inode);
4052 error = security_path_unlink(&path, dentry);
4053 if (error)
4054 goto exit2;
4055 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4056exit2:
4057 dput(dentry);
4058 }
4059 inode_unlock(path.dentry->d_inode);
4060 if (inode)
4061 iput(inode); /* truncate the inode here */
4062 inode = NULL;
4063 if (delegated_inode) {
4064 error = break_deleg_wait(&delegated_inode);
4065 if (!error)
4066 goto retry_deleg;
4067 }
4068 mnt_drop_write(path.mnt);
4069exit1:
4070 path_put(&path);
4071 if (retry_estale(error, lookup_flags)) {
4072 lookup_flags |= LOOKUP_REVAL;
4073 inode = NULL;
4074 goto retry;
4075 }
4076 putname(name);
4077 return error;
4078
4079slashes:
4080 if (d_is_negative(dentry))
4081 error = -ENOENT;
4082 else if (d_is_dir(dentry))
4083 error = -EISDIR;
4084 else
4085 error = -ENOTDIR;
4086 goto exit2;
4087}
4088
4089SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4090{
4091 if ((flag & ~AT_REMOVEDIR) != 0)
4092 return -EINVAL;
4093
4094 if (flag & AT_REMOVEDIR)
4095 return do_rmdir(dfd, pathname);
4096
4097 return do_unlinkat(dfd, getname(pathname));
4098}
4099
4100SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4101{
4102 return do_unlinkat(AT_FDCWD, getname(pathname));
4103}
4104
4105int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4106{
4107 int error = may_create(dir, dentry);
4108
4109 if (error)
4110 return error;
4111
4112 if (!dir->i_op->symlink)
4113 return -EPERM;
4114
4115 error = security_inode_symlink(dir, dentry, oldname);
4116 if (error)
4117 return error;
4118
4119 error = dir->i_op->symlink(dir, dentry, oldname);
4120 if (!error)
4121 fsnotify_create(dir, dentry);
4122 return error;
4123}
4124EXPORT_SYMBOL(vfs_symlink);
4125
4126long do_symlinkat(const char __user *oldname, int newdfd,
4127 const char __user *newname)
4128{
4129 int error;
4130 struct filename *from;
4131 struct dentry *dentry;
4132 struct path path;
4133 unsigned int lookup_flags = 0;
4134
4135 from = getname(oldname);
4136 if (IS_ERR(from))
4137 return PTR_ERR(from);
4138retry:
4139 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4140 error = PTR_ERR(dentry);
4141 if (IS_ERR(dentry))
4142 goto out_putname;
4143
4144 error = security_path_symlink(&path, dentry, from->name);
4145 if (!error)
4146 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4147 done_path_create(&path, dentry);
4148 if (retry_estale(error, lookup_flags)) {
4149 lookup_flags |= LOOKUP_REVAL;
4150 goto retry;
4151 }
4152out_putname:
4153 putname(from);
4154 return error;
4155}
4156
4157SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4158 int, newdfd, const char __user *, newname)
4159{
4160 return do_symlinkat(oldname, newdfd, newname);
4161}
4162
4163SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4164{
4165 return do_symlinkat(oldname, AT_FDCWD, newname);
4166}
4167
4168/**
4169 * vfs_link - create a new link
4170 * @old_dentry: object to be linked
4171 * @dir: new parent
4172 * @new_dentry: where to create the new link
4173 * @delegated_inode: returns inode needing a delegation break
4174 *
4175 * The caller must hold dir->i_mutex
4176 *
4177 * If vfs_link discovers a delegation on the to-be-linked file in need
4178 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4179 * inode in delegated_inode. The caller should then break the delegation
4180 * and retry. Because breaking a delegation may take a long time, the
4181 * caller should drop the i_mutex before doing so.
4182 *
4183 * Alternatively, a caller may pass NULL for delegated_inode. This may
4184 * be appropriate for callers that expect the underlying filesystem not
4185 * to be NFS exported.
4186 */
4187int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4188{
4189 struct inode *inode = old_dentry->d_inode;
4190 unsigned max_links = dir->i_sb->s_max_links;
4191 int error;
4192
4193 if (!inode)
4194 return -ENOENT;
4195
4196 error = may_create(dir, new_dentry);
4197 if (error)
4198 return error;
4199
4200 if (dir->i_sb != inode->i_sb)
4201 return -EXDEV;
4202
4203 /*
4204 * A link to an append-only or immutable file cannot be created.
4205 */
4206 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4207 return -EPERM;
4208 /*
4209 * Updating the link count will likely cause i_uid and i_gid to
4210 * be writen back improperly if their true value is unknown to
4211 * the vfs.
4212 */
4213 if (HAS_UNMAPPED_ID(inode))
4214 return -EPERM;
4215 if (!dir->i_op->link)
4216 return -EPERM;
4217 if (S_ISDIR(inode->i_mode))
4218 return -EPERM;
4219
4220 error = security_inode_link(old_dentry, dir, new_dentry);
4221 if (error)
4222 return error;
4223
4224 inode_lock(inode);
4225 /* Make sure we don't allow creating hardlink to an unlinked file */
4226 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4227 error = -ENOENT;
4228 else if (max_links && inode->i_nlink >= max_links)
4229 error = -EMLINK;
4230 else {
4231 error = try_break_deleg(inode, delegated_inode);
4232 if (!error)
4233 error = dir->i_op->link(old_dentry, dir, new_dentry);
4234 }
4235
4236 if (!error && (inode->i_state & I_LINKABLE)) {
4237 spin_lock(&inode->i_lock);
4238 inode->i_state &= ~I_LINKABLE;
4239 spin_unlock(&inode->i_lock);
4240 }
4241 inode_unlock(inode);
4242 if (!error)
4243 fsnotify_link(dir, inode, new_dentry);
4244 return error;
4245}
4246EXPORT_SYMBOL(vfs_link);
4247
4248/*
4249 * Hardlinks are often used in delicate situations. We avoid
4250 * security-related surprises by not following symlinks on the
4251 * newname. --KAB
4252 *
4253 * We don't follow them on the oldname either to be compatible
4254 * with linux 2.0, and to avoid hard-linking to directories
4255 * and other special files. --ADM
4256 */
4257int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4258 const char __user *newname, int flags)
4259{
4260 struct dentry *new_dentry;
4261 struct path old_path, new_path;
4262 struct inode *delegated_inode = NULL;
4263 int how = 0;
4264 int error;
4265
4266 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4267 return -EINVAL;
4268 /*
4269 * To use null names we require CAP_DAC_READ_SEARCH
4270 * This ensures that not everyone will be able to create
4271 * handlink using the passed filedescriptor.
4272 */
4273 if (flags & AT_EMPTY_PATH) {
4274 if (!capable(CAP_DAC_READ_SEARCH))
4275 return -ENOENT;
4276 how = LOOKUP_EMPTY;
4277 }
4278
4279 if (flags & AT_SYMLINK_FOLLOW)
4280 how |= LOOKUP_FOLLOW;
4281retry:
4282 error = user_path_at(olddfd, oldname, how, &old_path);
4283 if (error)
4284 return error;
4285
4286 new_dentry = user_path_create(newdfd, newname, &new_path,
4287 (how & LOOKUP_REVAL));
4288 error = PTR_ERR(new_dentry);
4289 if (IS_ERR(new_dentry))
4290 goto out;
4291
4292 error = -EXDEV;
4293 if (old_path.mnt != new_path.mnt)
4294 goto out_dput;
4295 error = may_linkat(&old_path);
4296 if (unlikely(error))
4297 goto out_dput;
4298 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4299 if (error)
4300 goto out_dput;
4301 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4302out_dput:
4303 done_path_create(&new_path, new_dentry);
4304 if (delegated_inode) {
4305 error = break_deleg_wait(&delegated_inode);
4306 if (!error) {
4307 path_put(&old_path);
4308 goto retry;
4309 }
4310 }
4311 if (retry_estale(error, how)) {
4312 path_put(&old_path);
4313 how |= LOOKUP_REVAL;
4314 goto retry;
4315 }
4316out:
4317 path_put(&old_path);
4318
4319 return error;
4320}
4321
4322SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4323 int, newdfd, const char __user *, newname, int, flags)
4324{
4325 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4326}
4327
4328SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4329{
4330 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4331}
4332
4333/**
4334 * vfs_rename - rename a filesystem object
4335 * @old_dir: parent of source
4336 * @old_dentry: source
4337 * @new_dir: parent of destination
4338 * @new_dentry: destination
4339 * @delegated_inode: returns an inode needing a delegation break
4340 * @flags: rename flags
4341 *
4342 * The caller must hold multiple mutexes--see lock_rename()).
4343 *
4344 * If vfs_rename discovers a delegation in need of breaking at either
4345 * the source or destination, it will return -EWOULDBLOCK and return a
4346 * reference to the inode in delegated_inode. The caller should then
4347 * break the delegation and retry. Because breaking a delegation may
4348 * take a long time, the caller should drop all locks before doing
4349 * so.
4350 *
4351 * Alternatively, a caller may pass NULL for delegated_inode. This may
4352 * be appropriate for callers that expect the underlying filesystem not
4353 * to be NFS exported.
4354 *
4355 * The worst of all namespace operations - renaming directory. "Perverted"
4356 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4357 * Problems:
4358 *
4359 * a) we can get into loop creation.
4360 * b) race potential - two innocent renames can create a loop together.
4361 * That's where 4.4 screws up. Current fix: serialization on
4362 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4363 * story.
4364 * c) we have to lock _four_ objects - parents and victim (if it exists),
4365 * and source (if it is not a directory).
4366 * And that - after we got ->i_mutex on parents (until then we don't know
4367 * whether the target exists). Solution: try to be smart with locking
4368 * order for inodes. We rely on the fact that tree topology may change
4369 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4370 * move will be locked. Thus we can rank directories by the tree
4371 * (ancestors first) and rank all non-directories after them.
4372 * That works since everybody except rename does "lock parent, lookup,
4373 * lock child" and rename is under ->s_vfs_rename_mutex.
4374 * HOWEVER, it relies on the assumption that any object with ->lookup()
4375 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4376 * we'd better make sure that there's no link(2) for them.
4377 * d) conversion from fhandle to dentry may come in the wrong moment - when
4378 * we are removing the target. Solution: we will have to grab ->i_mutex
4379 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4380 * ->i_mutex on parents, which works but leads to some truly excessive
4381 * locking].
4382 */
4383int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4384 struct inode *new_dir, struct dentry *new_dentry,
4385 struct inode **delegated_inode, unsigned int flags)
4386{
4387 int error;
4388 bool is_dir = d_is_dir(old_dentry);
4389 struct inode *source = old_dentry->d_inode;
4390 struct inode *target = new_dentry->d_inode;
4391 bool new_is_dir = false;
4392 unsigned max_links = new_dir->i_sb->s_max_links;
4393 struct name_snapshot old_name;
4394
4395 if (source == target)
4396 return 0;
4397
4398 error = may_delete(old_dir, old_dentry, is_dir);
4399 if (error)
4400 return error;
4401
4402 if (!target) {
4403 error = may_create(new_dir, new_dentry);
4404 } else {
4405 new_is_dir = d_is_dir(new_dentry);
4406
4407 if (!(flags & RENAME_EXCHANGE))
4408 error = may_delete(new_dir, new_dentry, is_dir);
4409 else
4410 error = may_delete(new_dir, new_dentry, new_is_dir);
4411 }
4412 if (error)
4413 return error;
4414
4415 if (!old_dir->i_op->rename)
4416 return -EPERM;
4417
4418 /*
4419 * If we are going to change the parent - check write permissions,
4420 * we'll need to flip '..'.
4421 */
4422 if (new_dir != old_dir) {
4423 if (is_dir) {
4424 error = inode_permission(source, MAY_WRITE);
4425 if (error)
4426 return error;
4427 }
4428 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4429 error = inode_permission(target, MAY_WRITE);
4430 if (error)
4431 return error;
4432 }
4433 }
4434
4435 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4436 flags);
4437 if (error)
4438 return error;
4439
4440 take_dentry_name_snapshot(&old_name, old_dentry);
4441 dget(new_dentry);
4442 if (!is_dir || (flags & RENAME_EXCHANGE))
4443 lock_two_nondirectories(source, target);
4444 else if (target)
4445 inode_lock(target);
4446
4447 error = -EBUSY;
4448 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4449 goto out;
4450
4451 if (max_links && new_dir != old_dir) {
4452 error = -EMLINK;
4453 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4454 goto out;
4455 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4456 old_dir->i_nlink >= max_links)
4457 goto out;
4458 }
4459 if (!is_dir) {
4460 error = try_break_deleg(source, delegated_inode);
4461 if (error)
4462 goto out;
4463 }
4464 if (target && !new_is_dir) {
4465 error = try_break_deleg(target, delegated_inode);
4466 if (error)
4467 goto out;
4468 }
4469 error = old_dir->i_op->rename(old_dir, old_dentry,
4470 new_dir, new_dentry, flags);
4471 if (error)
4472 goto out;
4473
4474 if (!(flags & RENAME_EXCHANGE) && target) {
4475 if (is_dir) {
4476 shrink_dcache_parent(new_dentry);
4477 target->i_flags |= S_DEAD;
4478 }
4479 dont_mount(new_dentry);
4480 detach_mounts(new_dentry);
4481 }
4482 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4483 if (!(flags & RENAME_EXCHANGE))
4484 d_move(old_dentry, new_dentry);
4485 else
4486 d_exchange(old_dentry, new_dentry);
4487 }
4488out:
4489 if (!is_dir || (flags & RENAME_EXCHANGE))
4490 unlock_two_nondirectories(source, target);
4491 else if (target)
4492 inode_unlock(target);
4493 dput(new_dentry);
4494 if (!error) {
4495 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4496 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4497 if (flags & RENAME_EXCHANGE) {
4498 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4499 new_is_dir, NULL, new_dentry);
4500 }
4501 }
4502 release_dentry_name_snapshot(&old_name);
4503
4504 return error;
4505}
4506EXPORT_SYMBOL(vfs_rename);
4507
4508static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4509 const char __user *newname, unsigned int flags)
4510{
4511 struct dentry *old_dentry, *new_dentry;
4512 struct dentry *trap;
4513 struct path old_path, new_path;
4514 struct qstr old_last, new_last;
4515 int old_type, new_type;
4516 struct inode *delegated_inode = NULL;
4517 struct filename *from;
4518 struct filename *to;
4519 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4520 bool should_retry = false;
4521 int error;
4522
4523 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4524 return -EINVAL;
4525
4526 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4527 (flags & RENAME_EXCHANGE))
4528 return -EINVAL;
4529
4530 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4531 return -EPERM;
4532
4533 if (flags & RENAME_EXCHANGE)
4534 target_flags = 0;
4535
4536retry:
4537 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4538 &old_path, &old_last, &old_type);
4539 if (IS_ERR(from)) {
4540 error = PTR_ERR(from);
4541 goto exit;
4542 }
4543
4544 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4545 &new_path, &new_last, &new_type);
4546 if (IS_ERR(to)) {
4547 error = PTR_ERR(to);
4548 goto exit1;
4549 }
4550
4551 error = -EXDEV;
4552 if (old_path.mnt != new_path.mnt)
4553 goto exit2;
4554
4555 error = -EBUSY;
4556 if (old_type != LAST_NORM)
4557 goto exit2;
4558
4559 if (flags & RENAME_NOREPLACE)
4560 error = -EEXIST;
4561 if (new_type != LAST_NORM)
4562 goto exit2;
4563
4564 error = mnt_want_write(old_path.mnt);
4565 if (error)
4566 goto exit2;
4567
4568retry_deleg:
4569 trap = lock_rename(new_path.dentry, old_path.dentry);
4570
4571 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4572 error = PTR_ERR(old_dentry);
4573 if (IS_ERR(old_dentry))
4574 goto exit3;
4575 /* source must exist */
4576 error = -ENOENT;
4577 if (d_is_negative(old_dentry))
4578 goto exit4;
4579 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4580 error = PTR_ERR(new_dentry);
4581 if (IS_ERR(new_dentry))
4582 goto exit4;
4583 error = -EEXIST;
4584 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4585 goto exit5;
4586 if (flags & RENAME_EXCHANGE) {
4587 error = -ENOENT;
4588 if (d_is_negative(new_dentry))
4589 goto exit5;
4590
4591 if (!d_is_dir(new_dentry)) {
4592 error = -ENOTDIR;
4593 if (new_last.name[new_last.len])
4594 goto exit5;
4595 }
4596 }
4597 /* unless the source is a directory trailing slashes give -ENOTDIR */
4598 if (!d_is_dir(old_dentry)) {
4599 error = -ENOTDIR;
4600 if (old_last.name[old_last.len])
4601 goto exit5;
4602 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4603 goto exit5;
4604 }
4605 /* source should not be ancestor of target */
4606 error = -EINVAL;
4607 if (old_dentry == trap)
4608 goto exit5;
4609 /* target should not be an ancestor of source */
4610 if (!(flags & RENAME_EXCHANGE))
4611 error = -ENOTEMPTY;
4612 if (new_dentry == trap)
4613 goto exit5;
4614
4615 error = security_path_rename(&old_path, old_dentry,
4616 &new_path, new_dentry, flags);
4617 if (error)
4618 goto exit5;
4619 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4620 new_path.dentry->d_inode, new_dentry,
4621 &delegated_inode, flags);
4622exit5:
4623 dput(new_dentry);
4624exit4:
4625 dput(old_dentry);
4626exit3:
4627 unlock_rename(new_path.dentry, old_path.dentry);
4628 if (delegated_inode) {
4629 error = break_deleg_wait(&delegated_inode);
4630 if (!error)
4631 goto retry_deleg;
4632 }
4633 mnt_drop_write(old_path.mnt);
4634exit2:
4635 if (retry_estale(error, lookup_flags))
4636 should_retry = true;
4637 path_put(&new_path);
4638 putname(to);
4639exit1:
4640 path_put(&old_path);
4641 putname(from);
4642 if (should_retry) {
4643 should_retry = false;
4644 lookup_flags |= LOOKUP_REVAL;
4645 goto retry;
4646 }
4647exit:
4648 return error;
4649}
4650
4651SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4652 int, newdfd, const char __user *, newname, unsigned int, flags)
4653{
4654 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4655}
4656
4657SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4658 int, newdfd, const char __user *, newname)
4659{
4660 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
4661}
4662
4663SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4664{
4665 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4666}
4667
4668int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4669{
4670 int error = may_create(dir, dentry);
4671 if (error)
4672 return error;
4673
4674 if (!dir->i_op->mknod)
4675 return -EPERM;
4676
4677 return dir->i_op->mknod(dir, dentry,
4678 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4679}
4680EXPORT_SYMBOL(vfs_whiteout);
4681
4682int readlink_copy(char __user *buffer, int buflen, const char *link)
4683{
4684 int len = PTR_ERR(link);
4685 if (IS_ERR(link))
4686 goto out;
4687
4688 len = strlen(link);
4689 if (len > (unsigned) buflen)
4690 len = buflen;
4691 if (copy_to_user(buffer, link, len))
4692 len = -EFAULT;
4693out:
4694 return len;
4695}
4696
4697/**
4698 * vfs_readlink - copy symlink body into userspace buffer
4699 * @dentry: dentry on which to get symbolic link
4700 * @buffer: user memory pointer
4701 * @buflen: size of buffer
4702 *
4703 * Does not touch atime. That's up to the caller if necessary
4704 *
4705 * Does not call security hook.
4706 */
4707int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4708{
4709 struct inode *inode = d_inode(dentry);
4710 DEFINE_DELAYED_CALL(done);
4711 const char *link;
4712 int res;
4713
4714 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4715 if (unlikely(inode->i_op->readlink))
4716 return inode->i_op->readlink(dentry, buffer, buflen);
4717
4718 if (!d_is_symlink(dentry))
4719 return -EINVAL;
4720
4721 spin_lock(&inode->i_lock);
4722 inode->i_opflags |= IOP_DEFAULT_READLINK;
4723 spin_unlock(&inode->i_lock);
4724 }
4725
4726 link = READ_ONCE(inode->i_link);
4727 if (!link) {
4728 link = inode->i_op->get_link(dentry, inode, &done);
4729 if (IS_ERR(link))
4730 return PTR_ERR(link);
4731 }
4732 res = readlink_copy(buffer, buflen, link);
4733 do_delayed_call(&done);
4734 return res;
4735}
4736EXPORT_SYMBOL(vfs_readlink);
4737
4738/**
4739 * vfs_get_link - get symlink body
4740 * @dentry: dentry on which to get symbolic link
4741 * @done: caller needs to free returned data with this
4742 *
4743 * Calls security hook and i_op->get_link() on the supplied inode.
4744 *
4745 * It does not touch atime. That's up to the caller if necessary.
4746 *
4747 * Does not work on "special" symlinks like /proc/$$/fd/N
4748 */
4749const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4750{
4751 const char *res = ERR_PTR(-EINVAL);
4752 struct inode *inode = d_inode(dentry);
4753
4754 if (d_is_symlink(dentry)) {
4755 res = ERR_PTR(security_inode_readlink(dentry));
4756 if (!res)
4757 res = inode->i_op->get_link(dentry, inode, done);
4758 }
4759 return res;
4760}
4761EXPORT_SYMBOL(vfs_get_link);
4762
4763/* get the link contents into pagecache */
4764const char *page_get_link(struct dentry *dentry, struct inode *inode,
4765 struct delayed_call *callback)
4766{
4767 char *kaddr;
4768 struct page *page;
4769 struct address_space *mapping = inode->i_mapping;
4770
4771 if (!dentry) {
4772 page = find_get_page(mapping, 0);
4773 if (!page)
4774 return ERR_PTR(-ECHILD);
4775 if (!PageUptodate(page)) {
4776 put_page(page);
4777 return ERR_PTR(-ECHILD);
4778 }
4779 } else {
4780 page = read_mapping_page(mapping, 0, NULL);
4781 if (IS_ERR(page))
4782 return (char*)page;
4783 }
4784 set_delayed_call(callback, page_put_link, page);
4785 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4786 kaddr = page_address(page);
4787 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4788 return kaddr;
4789}
4790
4791EXPORT_SYMBOL(page_get_link);
4792
4793void page_put_link(void *arg)
4794{
4795 put_page(arg);
4796}
4797EXPORT_SYMBOL(page_put_link);
4798
4799int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4800{
4801 DEFINE_DELAYED_CALL(done);
4802 int res = readlink_copy(buffer, buflen,
4803 page_get_link(dentry, d_inode(dentry),
4804 &done));
4805 do_delayed_call(&done);
4806 return res;
4807}
4808EXPORT_SYMBOL(page_readlink);
4809
4810/*
4811 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4812 */
4813int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4814{
4815 struct address_space *mapping = inode->i_mapping;
4816 struct page *page;
4817 void *fsdata;
4818 int err;
4819 unsigned int flags = 0;
4820 if (nofs)
4821 flags |= AOP_FLAG_NOFS;
4822
4823retry:
4824 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4825 flags, &page, &fsdata);
4826 if (err)
4827 goto fail;
4828
4829 memcpy(page_address(page), symname, len-1);
4830
4831 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4832 page, fsdata);
4833 if (err < 0)
4834 goto fail;
4835 if (err < len-1)
4836 goto retry;
4837
4838 mark_inode_dirty(inode);
4839 return 0;
4840fail:
4841 return err;
4842}
4843EXPORT_SYMBOL(__page_symlink);
4844
4845int page_symlink(struct inode *inode, const char *symname, int len)
4846{
4847 return __page_symlink(inode, symname, len,
4848 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4849}
4850EXPORT_SYMBOL(page_symlink);
4851
4852const struct inode_operations page_symlink_inode_operations = {
4853 .get_link = page_get_link,
4854};
4855EXPORT_SYMBOL(page_symlink_inode_operations);
1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
22#include <linux/pagemap.h>
23#include <linux/fsnotify.h>
24#include <linux/personality.h>
25#include <linux/security.h>
26#include <linux/ima.h>
27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
30#include <linux/capability.h>
31#include <linux/file.h>
32#include <linux/fcntl.h>
33#include <linux/device_cgroup.h>
34#include <linux/fs_struct.h>
35#include <linux/posix_acl.h>
36#include <asm/uaccess.h>
37
38#include "internal.h"
39
40/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existent name.
75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
118static int do_getname(const char __user *filename, char *page)
119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
140static char *getname_flags(const char __user * filename, int flags)
141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
151 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
152 __putname(tmp);
153 result = ERR_PTR(retval);
154 }
155 }
156 }
157 audit_getname(result);
158 return result;
159}
160
161char *getname(const char __user * filename)
162{
163 return getname_flags(filename, 0);
164}
165
166#ifdef CONFIG_AUDITSYSCALL
167void putname(const char *name)
168{
169 if (unlikely(!audit_dummy_context()))
170 audit_putname(name);
171 else
172 __putname(name);
173}
174EXPORT_SYMBOL(putname);
175#endif
176
177static int check_acl(struct inode *inode, int mask)
178{
179#ifdef CONFIG_FS_POSIX_ACL
180 struct posix_acl *acl;
181
182 if (mask & MAY_NOT_BLOCK) {
183 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
184 if (!acl)
185 return -EAGAIN;
186 /* no ->get_acl() calls in RCU mode... */
187 if (acl == ACL_NOT_CACHED)
188 return -ECHILD;
189 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
190 }
191
192 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
193
194 /*
195 * A filesystem can force a ACL callback by just never filling the
196 * ACL cache. But normally you'd fill the cache either at inode
197 * instantiation time, or on the first ->get_acl call.
198 *
199 * If the filesystem doesn't have a get_acl() function at all, we'll
200 * just create the negative cache entry.
201 */
202 if (acl == ACL_NOT_CACHED) {
203 if (inode->i_op->get_acl) {
204 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
205 if (IS_ERR(acl))
206 return PTR_ERR(acl);
207 } else {
208 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
209 return -EAGAIN;
210 }
211 }
212
213 if (acl) {
214 int error = posix_acl_permission(inode, acl, mask);
215 posix_acl_release(acl);
216 return error;
217 }
218#endif
219
220 return -EAGAIN;
221}
222
223/*
224 * This does basic POSIX ACL permission checking
225 */
226static int acl_permission_check(struct inode *inode, int mask)
227{
228 unsigned int mode = inode->i_mode;
229
230 mask &= MAY_READ | MAY_WRITE | MAY_EXEC | MAY_NOT_BLOCK;
231
232 if (current_user_ns() != inode_userns(inode))
233 goto other_perms;
234
235 if (likely(current_fsuid() == inode->i_uid))
236 mode >>= 6;
237 else {
238 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
239 int error = check_acl(inode, mask);
240 if (error != -EAGAIN)
241 return error;
242 }
243
244 if (in_group_p(inode->i_gid))
245 mode >>= 3;
246 }
247
248other_perms:
249 /*
250 * If the DACs are ok we don't need any capability check.
251 */
252 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
253 return 0;
254 return -EACCES;
255}
256
257/**
258 * generic_permission - check for access rights on a Posix-like filesystem
259 * @inode: inode to check access rights for
260 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
261 *
262 * Used to check for read/write/execute permissions on a file.
263 * We use "fsuid" for this, letting us set arbitrary permissions
264 * for filesystem access without changing the "normal" uids which
265 * are used for other things.
266 *
267 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
268 * request cannot be satisfied (eg. requires blocking or too much complexity).
269 * It would then be called again in ref-walk mode.
270 */
271int generic_permission(struct inode *inode, int mask)
272{
273 int ret;
274
275 /*
276 * Do the basic POSIX ACL permission checks.
277 */
278 ret = acl_permission_check(inode, mask);
279 if (ret != -EACCES)
280 return ret;
281
282 if (S_ISDIR(inode->i_mode)) {
283 /* DACs are overridable for directories */
284 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
285 return 0;
286 if (!(mask & MAY_WRITE))
287 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
288 return 0;
289 return -EACCES;
290 }
291 /*
292 * Read/write DACs are always overridable.
293 * Executable DACs are overridable when there is
294 * at least one exec bit set.
295 */
296 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
297 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
298 return 0;
299
300 /*
301 * Searching includes executable on directories, else just read.
302 */
303 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
304 if (mask == MAY_READ)
305 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
306 return 0;
307
308 return -EACCES;
309}
310
311/*
312 * We _really_ want to just do "generic_permission()" without
313 * even looking at the inode->i_op values. So we keep a cache
314 * flag in inode->i_opflags, that says "this has not special
315 * permission function, use the fast case".
316 */
317static inline int do_inode_permission(struct inode *inode, int mask)
318{
319 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
320 if (likely(inode->i_op->permission))
321 return inode->i_op->permission(inode, mask);
322
323 /* This gets set once for the inode lifetime */
324 spin_lock(&inode->i_lock);
325 inode->i_opflags |= IOP_FASTPERM;
326 spin_unlock(&inode->i_lock);
327 }
328 return generic_permission(inode, mask);
329}
330
331/**
332 * inode_permission - check for access rights to a given inode
333 * @inode: inode to check permission on
334 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
335 *
336 * Used to check for read/write/execute permissions on an inode.
337 * We use "fsuid" for this, letting us set arbitrary permissions
338 * for filesystem access without changing the "normal" uids which
339 * are used for other things.
340 */
341int inode_permission(struct inode *inode, int mask)
342{
343 int retval;
344
345 if (unlikely(mask & MAY_WRITE)) {
346 umode_t mode = inode->i_mode;
347
348 /*
349 * Nobody gets write access to a read-only fs.
350 */
351 if (IS_RDONLY(inode) &&
352 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
353 return -EROFS;
354
355 /*
356 * Nobody gets write access to an immutable file.
357 */
358 if (IS_IMMUTABLE(inode))
359 return -EACCES;
360 }
361
362 retval = do_inode_permission(inode, mask);
363 if (retval)
364 return retval;
365
366 retval = devcgroup_inode_permission(inode, mask);
367 if (retval)
368 return retval;
369
370 return security_inode_permission(inode, mask);
371}
372
373/**
374 * path_get - get a reference to a path
375 * @path: path to get the reference to
376 *
377 * Given a path increment the reference count to the dentry and the vfsmount.
378 */
379void path_get(struct path *path)
380{
381 mntget(path->mnt);
382 dget(path->dentry);
383}
384EXPORT_SYMBOL(path_get);
385
386/**
387 * path_put - put a reference to a path
388 * @path: path to put the reference to
389 *
390 * Given a path decrement the reference count to the dentry and the vfsmount.
391 */
392void path_put(struct path *path)
393{
394 dput(path->dentry);
395 mntput(path->mnt);
396}
397EXPORT_SYMBOL(path_put);
398
399/*
400 * Path walking has 2 modes, rcu-walk and ref-walk (see
401 * Documentation/filesystems/path-lookup.txt). In situations when we can't
402 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
403 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
404 * mode. Refcounts are grabbed at the last known good point before rcu-walk
405 * got stuck, so ref-walk may continue from there. If this is not successful
406 * (eg. a seqcount has changed), then failure is returned and it's up to caller
407 * to restart the path walk from the beginning in ref-walk mode.
408 */
409
410/**
411 * unlazy_walk - try to switch to ref-walk mode.
412 * @nd: nameidata pathwalk data
413 * @dentry: child of nd->path.dentry or NULL
414 * Returns: 0 on success, -ECHILD on failure
415 *
416 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
417 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
418 * @nd or NULL. Must be called from rcu-walk context.
419 */
420static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
421{
422 struct fs_struct *fs = current->fs;
423 struct dentry *parent = nd->path.dentry;
424 int want_root = 0;
425
426 BUG_ON(!(nd->flags & LOOKUP_RCU));
427 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
428 want_root = 1;
429 spin_lock(&fs->lock);
430 if (nd->root.mnt != fs->root.mnt ||
431 nd->root.dentry != fs->root.dentry)
432 goto err_root;
433 }
434 spin_lock(&parent->d_lock);
435 if (!dentry) {
436 if (!__d_rcu_to_refcount(parent, nd->seq))
437 goto err_parent;
438 BUG_ON(nd->inode != parent->d_inode);
439 } else {
440 if (dentry->d_parent != parent)
441 goto err_parent;
442 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
443 if (!__d_rcu_to_refcount(dentry, nd->seq))
444 goto err_child;
445 /*
446 * If the sequence check on the child dentry passed, then
447 * the child has not been removed from its parent. This
448 * means the parent dentry must be valid and able to take
449 * a reference at this point.
450 */
451 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
452 BUG_ON(!parent->d_count);
453 parent->d_count++;
454 spin_unlock(&dentry->d_lock);
455 }
456 spin_unlock(&parent->d_lock);
457 if (want_root) {
458 path_get(&nd->root);
459 spin_unlock(&fs->lock);
460 }
461 mntget(nd->path.mnt);
462
463 rcu_read_unlock();
464 br_read_unlock(vfsmount_lock);
465 nd->flags &= ~LOOKUP_RCU;
466 return 0;
467
468err_child:
469 spin_unlock(&dentry->d_lock);
470err_parent:
471 spin_unlock(&parent->d_lock);
472err_root:
473 if (want_root)
474 spin_unlock(&fs->lock);
475 return -ECHILD;
476}
477
478/**
479 * release_open_intent - free up open intent resources
480 * @nd: pointer to nameidata
481 */
482void release_open_intent(struct nameidata *nd)
483{
484 struct file *file = nd->intent.open.file;
485
486 if (file && !IS_ERR(file)) {
487 if (file->f_path.dentry == NULL)
488 put_filp(file);
489 else
490 fput(file);
491 }
492}
493
494static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
495{
496 return dentry->d_op->d_revalidate(dentry, nd);
497}
498
499/**
500 * complete_walk - successful completion of path walk
501 * @nd: pointer nameidata
502 *
503 * If we had been in RCU mode, drop out of it and legitimize nd->path.
504 * Revalidate the final result, unless we'd already done that during
505 * the path walk or the filesystem doesn't ask for it. Return 0 on
506 * success, -error on failure. In case of failure caller does not
507 * need to drop nd->path.
508 */
509static int complete_walk(struct nameidata *nd)
510{
511 struct dentry *dentry = nd->path.dentry;
512 int status;
513
514 if (nd->flags & LOOKUP_RCU) {
515 nd->flags &= ~LOOKUP_RCU;
516 if (!(nd->flags & LOOKUP_ROOT))
517 nd->root.mnt = NULL;
518 spin_lock(&dentry->d_lock);
519 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
520 spin_unlock(&dentry->d_lock);
521 rcu_read_unlock();
522 br_read_unlock(vfsmount_lock);
523 return -ECHILD;
524 }
525 BUG_ON(nd->inode != dentry->d_inode);
526 spin_unlock(&dentry->d_lock);
527 mntget(nd->path.mnt);
528 rcu_read_unlock();
529 br_read_unlock(vfsmount_lock);
530 }
531
532 if (likely(!(nd->flags & LOOKUP_JUMPED)))
533 return 0;
534
535 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
536 return 0;
537
538 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
539 return 0;
540
541 /* Note: we do not d_invalidate() */
542 status = d_revalidate(dentry, nd);
543 if (status > 0)
544 return 0;
545
546 if (!status)
547 status = -ESTALE;
548
549 path_put(&nd->path);
550 return status;
551}
552
553static __always_inline void set_root(struct nameidata *nd)
554{
555 if (!nd->root.mnt)
556 get_fs_root(current->fs, &nd->root);
557}
558
559static int link_path_walk(const char *, struct nameidata *);
560
561static __always_inline void set_root_rcu(struct nameidata *nd)
562{
563 if (!nd->root.mnt) {
564 struct fs_struct *fs = current->fs;
565 unsigned seq;
566
567 do {
568 seq = read_seqcount_begin(&fs->seq);
569 nd->root = fs->root;
570 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
571 } while (read_seqcount_retry(&fs->seq, seq));
572 }
573}
574
575static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
576{
577 int ret;
578
579 if (IS_ERR(link))
580 goto fail;
581
582 if (*link == '/') {
583 set_root(nd);
584 path_put(&nd->path);
585 nd->path = nd->root;
586 path_get(&nd->root);
587 nd->flags |= LOOKUP_JUMPED;
588 }
589 nd->inode = nd->path.dentry->d_inode;
590
591 ret = link_path_walk(link, nd);
592 return ret;
593fail:
594 path_put(&nd->path);
595 return PTR_ERR(link);
596}
597
598static void path_put_conditional(struct path *path, struct nameidata *nd)
599{
600 dput(path->dentry);
601 if (path->mnt != nd->path.mnt)
602 mntput(path->mnt);
603}
604
605static inline void path_to_nameidata(const struct path *path,
606 struct nameidata *nd)
607{
608 if (!(nd->flags & LOOKUP_RCU)) {
609 dput(nd->path.dentry);
610 if (nd->path.mnt != path->mnt)
611 mntput(nd->path.mnt);
612 }
613 nd->path.mnt = path->mnt;
614 nd->path.dentry = path->dentry;
615}
616
617static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
618{
619 struct inode *inode = link->dentry->d_inode;
620 if (!IS_ERR(cookie) && inode->i_op->put_link)
621 inode->i_op->put_link(link->dentry, nd, cookie);
622 path_put(link);
623}
624
625static __always_inline int
626follow_link(struct path *link, struct nameidata *nd, void **p)
627{
628 int error;
629 struct dentry *dentry = link->dentry;
630
631 BUG_ON(nd->flags & LOOKUP_RCU);
632
633 if (link->mnt == nd->path.mnt)
634 mntget(link->mnt);
635
636 if (unlikely(current->total_link_count >= 40)) {
637 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
638 path_put(&nd->path);
639 return -ELOOP;
640 }
641 cond_resched();
642 current->total_link_count++;
643
644 touch_atime(link->mnt, dentry);
645 nd_set_link(nd, NULL);
646
647 error = security_inode_follow_link(link->dentry, nd);
648 if (error) {
649 *p = ERR_PTR(error); /* no ->put_link(), please */
650 path_put(&nd->path);
651 return error;
652 }
653
654 nd->last_type = LAST_BIND;
655 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
656 error = PTR_ERR(*p);
657 if (!IS_ERR(*p)) {
658 char *s = nd_get_link(nd);
659 error = 0;
660 if (s)
661 error = __vfs_follow_link(nd, s);
662 else if (nd->last_type == LAST_BIND) {
663 nd->flags |= LOOKUP_JUMPED;
664 nd->inode = nd->path.dentry->d_inode;
665 if (nd->inode->i_op->follow_link) {
666 /* stepped on a _really_ weird one */
667 path_put(&nd->path);
668 error = -ELOOP;
669 }
670 }
671 }
672 return error;
673}
674
675static int follow_up_rcu(struct path *path)
676{
677 struct vfsmount *parent;
678 struct dentry *mountpoint;
679
680 parent = path->mnt->mnt_parent;
681 if (parent == path->mnt)
682 return 0;
683 mountpoint = path->mnt->mnt_mountpoint;
684 path->dentry = mountpoint;
685 path->mnt = parent;
686 return 1;
687}
688
689int follow_up(struct path *path)
690{
691 struct vfsmount *parent;
692 struct dentry *mountpoint;
693
694 br_read_lock(vfsmount_lock);
695 parent = path->mnt->mnt_parent;
696 if (parent == path->mnt) {
697 br_read_unlock(vfsmount_lock);
698 return 0;
699 }
700 mntget(parent);
701 mountpoint = dget(path->mnt->mnt_mountpoint);
702 br_read_unlock(vfsmount_lock);
703 dput(path->dentry);
704 path->dentry = mountpoint;
705 mntput(path->mnt);
706 path->mnt = parent;
707 return 1;
708}
709
710/*
711 * Perform an automount
712 * - return -EISDIR to tell follow_managed() to stop and return the path we
713 * were called with.
714 */
715static int follow_automount(struct path *path, unsigned flags,
716 bool *need_mntput)
717{
718 struct vfsmount *mnt;
719 int err;
720
721 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
722 return -EREMOTE;
723
724 /* We don't want to mount if someone's just doing a stat -
725 * unless they're stat'ing a directory and appended a '/' to
726 * the name.
727 *
728 * We do, however, want to mount if someone wants to open or
729 * create a file of any type under the mountpoint, wants to
730 * traverse through the mountpoint or wants to open the
731 * mounted directory. Also, autofs may mark negative dentries
732 * as being automount points. These will need the attentions
733 * of the daemon to instantiate them before they can be used.
734 */
735 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
736 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
737 path->dentry->d_inode)
738 return -EISDIR;
739
740 current->total_link_count++;
741 if (current->total_link_count >= 40)
742 return -ELOOP;
743
744 mnt = path->dentry->d_op->d_automount(path);
745 if (IS_ERR(mnt)) {
746 /*
747 * The filesystem is allowed to return -EISDIR here to indicate
748 * it doesn't want to automount. For instance, autofs would do
749 * this so that its userspace daemon can mount on this dentry.
750 *
751 * However, we can only permit this if it's a terminal point in
752 * the path being looked up; if it wasn't then the remainder of
753 * the path is inaccessible and we should say so.
754 */
755 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
756 return -EREMOTE;
757 return PTR_ERR(mnt);
758 }
759
760 if (!mnt) /* mount collision */
761 return 0;
762
763 if (!*need_mntput) {
764 /* lock_mount() may release path->mnt on error */
765 mntget(path->mnt);
766 *need_mntput = true;
767 }
768 err = finish_automount(mnt, path);
769
770 switch (err) {
771 case -EBUSY:
772 /* Someone else made a mount here whilst we were busy */
773 return 0;
774 case 0:
775 path_put(path);
776 path->mnt = mnt;
777 path->dentry = dget(mnt->mnt_root);
778 return 0;
779 default:
780 return err;
781 }
782
783}
784
785/*
786 * Handle a dentry that is managed in some way.
787 * - Flagged for transit management (autofs)
788 * - Flagged as mountpoint
789 * - Flagged as automount point
790 *
791 * This may only be called in refwalk mode.
792 *
793 * Serialization is taken care of in namespace.c
794 */
795static int follow_managed(struct path *path, unsigned flags)
796{
797 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
798 unsigned managed;
799 bool need_mntput = false;
800 int ret = 0;
801
802 /* Given that we're not holding a lock here, we retain the value in a
803 * local variable for each dentry as we look at it so that we don't see
804 * the components of that value change under us */
805 while (managed = ACCESS_ONCE(path->dentry->d_flags),
806 managed &= DCACHE_MANAGED_DENTRY,
807 unlikely(managed != 0)) {
808 /* Allow the filesystem to manage the transit without i_mutex
809 * being held. */
810 if (managed & DCACHE_MANAGE_TRANSIT) {
811 BUG_ON(!path->dentry->d_op);
812 BUG_ON(!path->dentry->d_op->d_manage);
813 ret = path->dentry->d_op->d_manage(path->dentry, false);
814 if (ret < 0)
815 break;
816 }
817
818 /* Transit to a mounted filesystem. */
819 if (managed & DCACHE_MOUNTED) {
820 struct vfsmount *mounted = lookup_mnt(path);
821 if (mounted) {
822 dput(path->dentry);
823 if (need_mntput)
824 mntput(path->mnt);
825 path->mnt = mounted;
826 path->dentry = dget(mounted->mnt_root);
827 need_mntput = true;
828 continue;
829 }
830
831 /* Something is mounted on this dentry in another
832 * namespace and/or whatever was mounted there in this
833 * namespace got unmounted before we managed to get the
834 * vfsmount_lock */
835 }
836
837 /* Handle an automount point */
838 if (managed & DCACHE_NEED_AUTOMOUNT) {
839 ret = follow_automount(path, flags, &need_mntput);
840 if (ret < 0)
841 break;
842 continue;
843 }
844
845 /* We didn't change the current path point */
846 break;
847 }
848
849 if (need_mntput && path->mnt == mnt)
850 mntput(path->mnt);
851 if (ret == -EISDIR)
852 ret = 0;
853 return ret;
854}
855
856int follow_down_one(struct path *path)
857{
858 struct vfsmount *mounted;
859
860 mounted = lookup_mnt(path);
861 if (mounted) {
862 dput(path->dentry);
863 mntput(path->mnt);
864 path->mnt = mounted;
865 path->dentry = dget(mounted->mnt_root);
866 return 1;
867 }
868 return 0;
869}
870
871static inline bool managed_dentry_might_block(struct dentry *dentry)
872{
873 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
874 dentry->d_op->d_manage(dentry, true) < 0);
875}
876
877/*
878 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
879 * we meet a managed dentry that would need blocking.
880 */
881static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
882 struct inode **inode)
883{
884 for (;;) {
885 struct vfsmount *mounted;
886 /*
887 * Don't forget we might have a non-mountpoint managed dentry
888 * that wants to block transit.
889 */
890 if (unlikely(managed_dentry_might_block(path->dentry)))
891 return false;
892
893 if (!d_mountpoint(path->dentry))
894 break;
895
896 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
897 if (!mounted)
898 break;
899 path->mnt = mounted;
900 path->dentry = mounted->mnt_root;
901 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
902 /*
903 * Update the inode too. We don't need to re-check the
904 * dentry sequence number here after this d_inode read,
905 * because a mount-point is always pinned.
906 */
907 *inode = path->dentry->d_inode;
908 }
909 return true;
910}
911
912static void follow_mount_rcu(struct nameidata *nd)
913{
914 while (d_mountpoint(nd->path.dentry)) {
915 struct vfsmount *mounted;
916 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
917 if (!mounted)
918 break;
919 nd->path.mnt = mounted;
920 nd->path.dentry = mounted->mnt_root;
921 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
922 }
923}
924
925static int follow_dotdot_rcu(struct nameidata *nd)
926{
927 set_root_rcu(nd);
928
929 while (1) {
930 if (nd->path.dentry == nd->root.dentry &&
931 nd->path.mnt == nd->root.mnt) {
932 break;
933 }
934 if (nd->path.dentry != nd->path.mnt->mnt_root) {
935 struct dentry *old = nd->path.dentry;
936 struct dentry *parent = old->d_parent;
937 unsigned seq;
938
939 seq = read_seqcount_begin(&parent->d_seq);
940 if (read_seqcount_retry(&old->d_seq, nd->seq))
941 goto failed;
942 nd->path.dentry = parent;
943 nd->seq = seq;
944 break;
945 }
946 if (!follow_up_rcu(&nd->path))
947 break;
948 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
949 }
950 follow_mount_rcu(nd);
951 nd->inode = nd->path.dentry->d_inode;
952 return 0;
953
954failed:
955 nd->flags &= ~LOOKUP_RCU;
956 if (!(nd->flags & LOOKUP_ROOT))
957 nd->root.mnt = NULL;
958 rcu_read_unlock();
959 br_read_unlock(vfsmount_lock);
960 return -ECHILD;
961}
962
963/*
964 * Follow down to the covering mount currently visible to userspace. At each
965 * point, the filesystem owning that dentry may be queried as to whether the
966 * caller is permitted to proceed or not.
967 */
968int follow_down(struct path *path)
969{
970 unsigned managed;
971 int ret;
972
973 while (managed = ACCESS_ONCE(path->dentry->d_flags),
974 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
975 /* Allow the filesystem to manage the transit without i_mutex
976 * being held.
977 *
978 * We indicate to the filesystem if someone is trying to mount
979 * something here. This gives autofs the chance to deny anyone
980 * other than its daemon the right to mount on its
981 * superstructure.
982 *
983 * The filesystem may sleep at this point.
984 */
985 if (managed & DCACHE_MANAGE_TRANSIT) {
986 BUG_ON(!path->dentry->d_op);
987 BUG_ON(!path->dentry->d_op->d_manage);
988 ret = path->dentry->d_op->d_manage(
989 path->dentry, false);
990 if (ret < 0)
991 return ret == -EISDIR ? 0 : ret;
992 }
993
994 /* Transit to a mounted filesystem. */
995 if (managed & DCACHE_MOUNTED) {
996 struct vfsmount *mounted = lookup_mnt(path);
997 if (!mounted)
998 break;
999 dput(path->dentry);
1000 mntput(path->mnt);
1001 path->mnt = mounted;
1002 path->dentry = dget(mounted->mnt_root);
1003 continue;
1004 }
1005
1006 /* Don't handle automount points here */
1007 break;
1008 }
1009 return 0;
1010}
1011
1012/*
1013 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1014 */
1015static void follow_mount(struct path *path)
1016{
1017 while (d_mountpoint(path->dentry)) {
1018 struct vfsmount *mounted = lookup_mnt(path);
1019 if (!mounted)
1020 break;
1021 dput(path->dentry);
1022 mntput(path->mnt);
1023 path->mnt = mounted;
1024 path->dentry = dget(mounted->mnt_root);
1025 }
1026}
1027
1028static void follow_dotdot(struct nameidata *nd)
1029{
1030 set_root(nd);
1031
1032 while(1) {
1033 struct dentry *old = nd->path.dentry;
1034
1035 if (nd->path.dentry == nd->root.dentry &&
1036 nd->path.mnt == nd->root.mnt) {
1037 break;
1038 }
1039 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1040 /* rare case of legitimate dget_parent()... */
1041 nd->path.dentry = dget_parent(nd->path.dentry);
1042 dput(old);
1043 break;
1044 }
1045 if (!follow_up(&nd->path))
1046 break;
1047 }
1048 follow_mount(&nd->path);
1049 nd->inode = nd->path.dentry->d_inode;
1050}
1051
1052/*
1053 * Allocate a dentry with name and parent, and perform a parent
1054 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1055 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1056 * have verified that no child exists while under i_mutex.
1057 */
1058static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1059 struct qstr *name, struct nameidata *nd)
1060{
1061 struct inode *inode = parent->d_inode;
1062 struct dentry *dentry;
1063 struct dentry *old;
1064
1065 /* Don't create child dentry for a dead directory. */
1066 if (unlikely(IS_DEADDIR(inode)))
1067 return ERR_PTR(-ENOENT);
1068
1069 dentry = d_alloc(parent, name);
1070 if (unlikely(!dentry))
1071 return ERR_PTR(-ENOMEM);
1072
1073 old = inode->i_op->lookup(inode, dentry, nd);
1074 if (unlikely(old)) {
1075 dput(dentry);
1076 dentry = old;
1077 }
1078 return dentry;
1079}
1080
1081/*
1082 * We already have a dentry, but require a lookup to be performed on the parent
1083 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1084 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1085 * child exists while under i_mutex.
1086 */
1087static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1088 struct nameidata *nd)
1089{
1090 struct inode *inode = parent->d_inode;
1091 struct dentry *old;
1092
1093 /* Don't create child dentry for a dead directory. */
1094 if (unlikely(IS_DEADDIR(inode)))
1095 return ERR_PTR(-ENOENT);
1096
1097 old = inode->i_op->lookup(inode, dentry, nd);
1098 if (unlikely(old)) {
1099 dput(dentry);
1100 dentry = old;
1101 }
1102 return dentry;
1103}
1104
1105/*
1106 * It's more convoluted than I'd like it to be, but... it's still fairly
1107 * small and for now I'd prefer to have fast path as straight as possible.
1108 * It _is_ time-critical.
1109 */
1110static int do_lookup(struct nameidata *nd, struct qstr *name,
1111 struct path *path, struct inode **inode)
1112{
1113 struct vfsmount *mnt = nd->path.mnt;
1114 struct dentry *dentry, *parent = nd->path.dentry;
1115 int need_reval = 1;
1116 int status = 1;
1117 int err;
1118
1119 /*
1120 * Rename seqlock is not required here because in the off chance
1121 * of a false negative due to a concurrent rename, we're going to
1122 * do the non-racy lookup, below.
1123 */
1124 if (nd->flags & LOOKUP_RCU) {
1125 unsigned seq;
1126 *inode = nd->inode;
1127 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1128 if (!dentry)
1129 goto unlazy;
1130
1131 /* Memory barrier in read_seqcount_begin of child is enough */
1132 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1133 return -ECHILD;
1134 nd->seq = seq;
1135
1136 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1137 status = d_revalidate(dentry, nd);
1138 if (unlikely(status <= 0)) {
1139 if (status != -ECHILD)
1140 need_reval = 0;
1141 goto unlazy;
1142 }
1143 }
1144 if (unlikely(d_need_lookup(dentry)))
1145 goto unlazy;
1146 path->mnt = mnt;
1147 path->dentry = dentry;
1148 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1149 goto unlazy;
1150 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1151 goto unlazy;
1152 return 0;
1153unlazy:
1154 if (unlazy_walk(nd, dentry))
1155 return -ECHILD;
1156 } else {
1157 dentry = __d_lookup(parent, name);
1158 }
1159
1160 if (dentry && unlikely(d_need_lookup(dentry))) {
1161 dput(dentry);
1162 dentry = NULL;
1163 }
1164retry:
1165 if (unlikely(!dentry)) {
1166 struct inode *dir = parent->d_inode;
1167 BUG_ON(nd->inode != dir);
1168
1169 mutex_lock(&dir->i_mutex);
1170 dentry = d_lookup(parent, name);
1171 if (likely(!dentry)) {
1172 dentry = d_alloc_and_lookup(parent, name, nd);
1173 if (IS_ERR(dentry)) {
1174 mutex_unlock(&dir->i_mutex);
1175 return PTR_ERR(dentry);
1176 }
1177 /* known good */
1178 need_reval = 0;
1179 status = 1;
1180 } else if (unlikely(d_need_lookup(dentry))) {
1181 dentry = d_inode_lookup(parent, dentry, nd);
1182 if (IS_ERR(dentry)) {
1183 mutex_unlock(&dir->i_mutex);
1184 return PTR_ERR(dentry);
1185 }
1186 /* known good */
1187 need_reval = 0;
1188 status = 1;
1189 }
1190 mutex_unlock(&dir->i_mutex);
1191 }
1192 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1193 status = d_revalidate(dentry, nd);
1194 if (unlikely(status <= 0)) {
1195 if (status < 0) {
1196 dput(dentry);
1197 return status;
1198 }
1199 if (!d_invalidate(dentry)) {
1200 dput(dentry);
1201 dentry = NULL;
1202 need_reval = 1;
1203 goto retry;
1204 }
1205 }
1206
1207 path->mnt = mnt;
1208 path->dentry = dentry;
1209 err = follow_managed(path, nd->flags);
1210 if (unlikely(err < 0)) {
1211 path_put_conditional(path, nd);
1212 return err;
1213 }
1214 *inode = path->dentry->d_inode;
1215 return 0;
1216}
1217
1218static inline int may_lookup(struct nameidata *nd)
1219{
1220 if (nd->flags & LOOKUP_RCU) {
1221 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1222 if (err != -ECHILD)
1223 return err;
1224 if (unlazy_walk(nd, NULL))
1225 return -ECHILD;
1226 }
1227 return inode_permission(nd->inode, MAY_EXEC);
1228}
1229
1230static inline int handle_dots(struct nameidata *nd, int type)
1231{
1232 if (type == LAST_DOTDOT) {
1233 if (nd->flags & LOOKUP_RCU) {
1234 if (follow_dotdot_rcu(nd))
1235 return -ECHILD;
1236 } else
1237 follow_dotdot(nd);
1238 }
1239 return 0;
1240}
1241
1242static void terminate_walk(struct nameidata *nd)
1243{
1244 if (!(nd->flags & LOOKUP_RCU)) {
1245 path_put(&nd->path);
1246 } else {
1247 nd->flags &= ~LOOKUP_RCU;
1248 if (!(nd->flags & LOOKUP_ROOT))
1249 nd->root.mnt = NULL;
1250 rcu_read_unlock();
1251 br_read_unlock(vfsmount_lock);
1252 }
1253}
1254
1255/*
1256 * Do we need to follow links? We _really_ want to be able
1257 * to do this check without having to look at inode->i_op,
1258 * so we keep a cache of "no, this doesn't need follow_link"
1259 * for the common case.
1260 */
1261static inline int should_follow_link(struct inode *inode, int follow)
1262{
1263 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1264 if (likely(inode->i_op->follow_link))
1265 return follow;
1266
1267 /* This gets set once for the inode lifetime */
1268 spin_lock(&inode->i_lock);
1269 inode->i_opflags |= IOP_NOFOLLOW;
1270 spin_unlock(&inode->i_lock);
1271 }
1272 return 0;
1273}
1274
1275static inline int walk_component(struct nameidata *nd, struct path *path,
1276 struct qstr *name, int type, int follow)
1277{
1278 struct inode *inode;
1279 int err;
1280 /*
1281 * "." and ".." are special - ".." especially so because it has
1282 * to be able to know about the current root directory and
1283 * parent relationships.
1284 */
1285 if (unlikely(type != LAST_NORM))
1286 return handle_dots(nd, type);
1287 err = do_lookup(nd, name, path, &inode);
1288 if (unlikely(err)) {
1289 terminate_walk(nd);
1290 return err;
1291 }
1292 if (!inode) {
1293 path_to_nameidata(path, nd);
1294 terminate_walk(nd);
1295 return -ENOENT;
1296 }
1297 if (should_follow_link(inode, follow)) {
1298 if (nd->flags & LOOKUP_RCU) {
1299 if (unlikely(unlazy_walk(nd, path->dentry))) {
1300 terminate_walk(nd);
1301 return -ECHILD;
1302 }
1303 }
1304 BUG_ON(inode != path->dentry->d_inode);
1305 return 1;
1306 }
1307 path_to_nameidata(path, nd);
1308 nd->inode = inode;
1309 return 0;
1310}
1311
1312/*
1313 * This limits recursive symlink follows to 8, while
1314 * limiting consecutive symlinks to 40.
1315 *
1316 * Without that kind of total limit, nasty chains of consecutive
1317 * symlinks can cause almost arbitrarily long lookups.
1318 */
1319static inline int nested_symlink(struct path *path, struct nameidata *nd)
1320{
1321 int res;
1322
1323 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1324 path_put_conditional(path, nd);
1325 path_put(&nd->path);
1326 return -ELOOP;
1327 }
1328 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1329
1330 nd->depth++;
1331 current->link_count++;
1332
1333 do {
1334 struct path link = *path;
1335 void *cookie;
1336
1337 res = follow_link(&link, nd, &cookie);
1338 if (!res)
1339 res = walk_component(nd, path, &nd->last,
1340 nd->last_type, LOOKUP_FOLLOW);
1341 put_link(nd, &link, cookie);
1342 } while (res > 0);
1343
1344 current->link_count--;
1345 nd->depth--;
1346 return res;
1347}
1348
1349/*
1350 * We really don't want to look at inode->i_op->lookup
1351 * when we don't have to. So we keep a cache bit in
1352 * the inode ->i_opflags field that says "yes, we can
1353 * do lookup on this inode".
1354 */
1355static inline int can_lookup(struct inode *inode)
1356{
1357 if (likely(inode->i_opflags & IOP_LOOKUP))
1358 return 1;
1359 if (likely(!inode->i_op->lookup))
1360 return 0;
1361
1362 /* We do this once for the lifetime of the inode */
1363 spin_lock(&inode->i_lock);
1364 inode->i_opflags |= IOP_LOOKUP;
1365 spin_unlock(&inode->i_lock);
1366 return 1;
1367}
1368
1369/*
1370 * Name resolution.
1371 * This is the basic name resolution function, turning a pathname into
1372 * the final dentry. We expect 'base' to be positive and a directory.
1373 *
1374 * Returns 0 and nd will have valid dentry and mnt on success.
1375 * Returns error and drops reference to input namei data on failure.
1376 */
1377static int link_path_walk(const char *name, struct nameidata *nd)
1378{
1379 struct path next;
1380 int err;
1381
1382 while (*name=='/')
1383 name++;
1384 if (!*name)
1385 return 0;
1386
1387 /* At this point we know we have a real path component. */
1388 for(;;) {
1389 unsigned long hash;
1390 struct qstr this;
1391 unsigned int c;
1392 int type;
1393
1394 err = may_lookup(nd);
1395 if (err)
1396 break;
1397
1398 this.name = name;
1399 c = *(const unsigned char *)name;
1400
1401 hash = init_name_hash();
1402 do {
1403 name++;
1404 hash = partial_name_hash(c, hash);
1405 c = *(const unsigned char *)name;
1406 } while (c && (c != '/'));
1407 this.len = name - (const char *) this.name;
1408 this.hash = end_name_hash(hash);
1409
1410 type = LAST_NORM;
1411 if (this.name[0] == '.') switch (this.len) {
1412 case 2:
1413 if (this.name[1] == '.') {
1414 type = LAST_DOTDOT;
1415 nd->flags |= LOOKUP_JUMPED;
1416 }
1417 break;
1418 case 1:
1419 type = LAST_DOT;
1420 }
1421 if (likely(type == LAST_NORM)) {
1422 struct dentry *parent = nd->path.dentry;
1423 nd->flags &= ~LOOKUP_JUMPED;
1424 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1425 err = parent->d_op->d_hash(parent, nd->inode,
1426 &this);
1427 if (err < 0)
1428 break;
1429 }
1430 }
1431
1432 /* remove trailing slashes? */
1433 if (!c)
1434 goto last_component;
1435 while (*++name == '/');
1436 if (!*name)
1437 goto last_component;
1438
1439 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1440 if (err < 0)
1441 return err;
1442
1443 if (err) {
1444 err = nested_symlink(&next, nd);
1445 if (err)
1446 return err;
1447 }
1448 if (can_lookup(nd->inode))
1449 continue;
1450 err = -ENOTDIR;
1451 break;
1452 /* here ends the main loop */
1453
1454last_component:
1455 nd->last = this;
1456 nd->last_type = type;
1457 return 0;
1458 }
1459 terminate_walk(nd);
1460 return err;
1461}
1462
1463static int path_init(int dfd, const char *name, unsigned int flags,
1464 struct nameidata *nd, struct file **fp)
1465{
1466 int retval = 0;
1467 int fput_needed;
1468 struct file *file;
1469
1470 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1471 nd->flags = flags | LOOKUP_JUMPED;
1472 nd->depth = 0;
1473 if (flags & LOOKUP_ROOT) {
1474 struct inode *inode = nd->root.dentry->d_inode;
1475 if (*name) {
1476 if (!inode->i_op->lookup)
1477 return -ENOTDIR;
1478 retval = inode_permission(inode, MAY_EXEC);
1479 if (retval)
1480 return retval;
1481 }
1482 nd->path = nd->root;
1483 nd->inode = inode;
1484 if (flags & LOOKUP_RCU) {
1485 br_read_lock(vfsmount_lock);
1486 rcu_read_lock();
1487 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1488 } else {
1489 path_get(&nd->path);
1490 }
1491 return 0;
1492 }
1493
1494 nd->root.mnt = NULL;
1495
1496 if (*name=='/') {
1497 if (flags & LOOKUP_RCU) {
1498 br_read_lock(vfsmount_lock);
1499 rcu_read_lock();
1500 set_root_rcu(nd);
1501 } else {
1502 set_root(nd);
1503 path_get(&nd->root);
1504 }
1505 nd->path = nd->root;
1506 } else if (dfd == AT_FDCWD) {
1507 if (flags & LOOKUP_RCU) {
1508 struct fs_struct *fs = current->fs;
1509 unsigned seq;
1510
1511 br_read_lock(vfsmount_lock);
1512 rcu_read_lock();
1513
1514 do {
1515 seq = read_seqcount_begin(&fs->seq);
1516 nd->path = fs->pwd;
1517 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1518 } while (read_seqcount_retry(&fs->seq, seq));
1519 } else {
1520 get_fs_pwd(current->fs, &nd->path);
1521 }
1522 } else {
1523 struct dentry *dentry;
1524
1525 file = fget_raw_light(dfd, &fput_needed);
1526 retval = -EBADF;
1527 if (!file)
1528 goto out_fail;
1529
1530 dentry = file->f_path.dentry;
1531
1532 if (*name) {
1533 retval = -ENOTDIR;
1534 if (!S_ISDIR(dentry->d_inode->i_mode))
1535 goto fput_fail;
1536
1537 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1538 if (retval)
1539 goto fput_fail;
1540 }
1541
1542 nd->path = file->f_path;
1543 if (flags & LOOKUP_RCU) {
1544 if (fput_needed)
1545 *fp = file;
1546 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1547 br_read_lock(vfsmount_lock);
1548 rcu_read_lock();
1549 } else {
1550 path_get(&file->f_path);
1551 fput_light(file, fput_needed);
1552 }
1553 }
1554
1555 nd->inode = nd->path.dentry->d_inode;
1556 return 0;
1557
1558fput_fail:
1559 fput_light(file, fput_needed);
1560out_fail:
1561 return retval;
1562}
1563
1564static inline int lookup_last(struct nameidata *nd, struct path *path)
1565{
1566 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1567 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1568
1569 nd->flags &= ~LOOKUP_PARENT;
1570 return walk_component(nd, path, &nd->last, nd->last_type,
1571 nd->flags & LOOKUP_FOLLOW);
1572}
1573
1574/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1575static int path_lookupat(int dfd, const char *name,
1576 unsigned int flags, struct nameidata *nd)
1577{
1578 struct file *base = NULL;
1579 struct path path;
1580 int err;
1581
1582 /*
1583 * Path walking is largely split up into 2 different synchronisation
1584 * schemes, rcu-walk and ref-walk (explained in
1585 * Documentation/filesystems/path-lookup.txt). These share much of the
1586 * path walk code, but some things particularly setup, cleanup, and
1587 * following mounts are sufficiently divergent that functions are
1588 * duplicated. Typically there is a function foo(), and its RCU
1589 * analogue, foo_rcu().
1590 *
1591 * -ECHILD is the error number of choice (just to avoid clashes) that
1592 * is returned if some aspect of an rcu-walk fails. Such an error must
1593 * be handled by restarting a traditional ref-walk (which will always
1594 * be able to complete).
1595 */
1596 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1597
1598 if (unlikely(err))
1599 return err;
1600
1601 current->total_link_count = 0;
1602 err = link_path_walk(name, nd);
1603
1604 if (!err && !(flags & LOOKUP_PARENT)) {
1605 err = lookup_last(nd, &path);
1606 while (err > 0) {
1607 void *cookie;
1608 struct path link = path;
1609 nd->flags |= LOOKUP_PARENT;
1610 err = follow_link(&link, nd, &cookie);
1611 if (!err)
1612 err = lookup_last(nd, &path);
1613 put_link(nd, &link, cookie);
1614 }
1615 }
1616
1617 if (!err)
1618 err = complete_walk(nd);
1619
1620 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1621 if (!nd->inode->i_op->lookup) {
1622 path_put(&nd->path);
1623 err = -ENOTDIR;
1624 }
1625 }
1626
1627 if (base)
1628 fput(base);
1629
1630 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1631 path_put(&nd->root);
1632 nd->root.mnt = NULL;
1633 }
1634 return err;
1635}
1636
1637static int do_path_lookup(int dfd, const char *name,
1638 unsigned int flags, struct nameidata *nd)
1639{
1640 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1641 if (unlikely(retval == -ECHILD))
1642 retval = path_lookupat(dfd, name, flags, nd);
1643 if (unlikely(retval == -ESTALE))
1644 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1645
1646 if (likely(!retval)) {
1647 if (unlikely(!audit_dummy_context())) {
1648 if (nd->path.dentry && nd->inode)
1649 audit_inode(name, nd->path.dentry);
1650 }
1651 }
1652 return retval;
1653}
1654
1655int kern_path_parent(const char *name, struct nameidata *nd)
1656{
1657 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1658}
1659
1660int kern_path(const char *name, unsigned int flags, struct path *path)
1661{
1662 struct nameidata nd;
1663 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1664 if (!res)
1665 *path = nd.path;
1666 return res;
1667}
1668
1669/**
1670 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1671 * @dentry: pointer to dentry of the base directory
1672 * @mnt: pointer to vfs mount of the base directory
1673 * @name: pointer to file name
1674 * @flags: lookup flags
1675 * @path: pointer to struct path to fill
1676 */
1677int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1678 const char *name, unsigned int flags,
1679 struct path *path)
1680{
1681 struct nameidata nd;
1682 int err;
1683 nd.root.dentry = dentry;
1684 nd.root.mnt = mnt;
1685 BUG_ON(flags & LOOKUP_PARENT);
1686 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1687 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1688 if (!err)
1689 *path = nd.path;
1690 return err;
1691}
1692
1693static struct dentry *__lookup_hash(struct qstr *name,
1694 struct dentry *base, struct nameidata *nd)
1695{
1696 struct inode *inode = base->d_inode;
1697 struct dentry *dentry;
1698 int err;
1699
1700 err = inode_permission(inode, MAY_EXEC);
1701 if (err)
1702 return ERR_PTR(err);
1703
1704 /*
1705 * Don't bother with __d_lookup: callers are for creat as
1706 * well as unlink, so a lot of the time it would cost
1707 * a double lookup.
1708 */
1709 dentry = d_lookup(base, name);
1710
1711 if (dentry && d_need_lookup(dentry)) {
1712 /*
1713 * __lookup_hash is called with the parent dir's i_mutex already
1714 * held, so we are good to go here.
1715 */
1716 dentry = d_inode_lookup(base, dentry, nd);
1717 if (IS_ERR(dentry))
1718 return dentry;
1719 }
1720
1721 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1722 int status = d_revalidate(dentry, nd);
1723 if (unlikely(status <= 0)) {
1724 /*
1725 * The dentry failed validation.
1726 * If d_revalidate returned 0 attempt to invalidate
1727 * the dentry otherwise d_revalidate is asking us
1728 * to return a fail status.
1729 */
1730 if (status < 0) {
1731 dput(dentry);
1732 return ERR_PTR(status);
1733 } else if (!d_invalidate(dentry)) {
1734 dput(dentry);
1735 dentry = NULL;
1736 }
1737 }
1738 }
1739
1740 if (!dentry)
1741 dentry = d_alloc_and_lookup(base, name, nd);
1742
1743 return dentry;
1744}
1745
1746/*
1747 * Restricted form of lookup. Doesn't follow links, single-component only,
1748 * needs parent already locked. Doesn't follow mounts.
1749 * SMP-safe.
1750 */
1751static struct dentry *lookup_hash(struct nameidata *nd)
1752{
1753 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1754}
1755
1756/**
1757 * lookup_one_len - filesystem helper to lookup single pathname component
1758 * @name: pathname component to lookup
1759 * @base: base directory to lookup from
1760 * @len: maximum length @len should be interpreted to
1761 *
1762 * Note that this routine is purely a helper for filesystem usage and should
1763 * not be called by generic code. Also note that by using this function the
1764 * nameidata argument is passed to the filesystem methods and a filesystem
1765 * using this helper needs to be prepared for that.
1766 */
1767struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1768{
1769 struct qstr this;
1770 unsigned long hash;
1771 unsigned int c;
1772
1773 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1774
1775 this.name = name;
1776 this.len = len;
1777 if (!len)
1778 return ERR_PTR(-EACCES);
1779
1780 hash = init_name_hash();
1781 while (len--) {
1782 c = *(const unsigned char *)name++;
1783 if (c == '/' || c == '\0')
1784 return ERR_PTR(-EACCES);
1785 hash = partial_name_hash(c, hash);
1786 }
1787 this.hash = end_name_hash(hash);
1788 /*
1789 * See if the low-level filesystem might want
1790 * to use its own hash..
1791 */
1792 if (base->d_flags & DCACHE_OP_HASH) {
1793 int err = base->d_op->d_hash(base, base->d_inode, &this);
1794 if (err < 0)
1795 return ERR_PTR(err);
1796 }
1797
1798 return __lookup_hash(&this, base, NULL);
1799}
1800
1801int user_path_at(int dfd, const char __user *name, unsigned flags,
1802 struct path *path)
1803{
1804 struct nameidata nd;
1805 char *tmp = getname_flags(name, flags);
1806 int err = PTR_ERR(tmp);
1807 if (!IS_ERR(tmp)) {
1808
1809 BUG_ON(flags & LOOKUP_PARENT);
1810
1811 err = do_path_lookup(dfd, tmp, flags, &nd);
1812 putname(tmp);
1813 if (!err)
1814 *path = nd.path;
1815 }
1816 return err;
1817}
1818
1819static int user_path_parent(int dfd, const char __user *path,
1820 struct nameidata *nd, char **name)
1821{
1822 char *s = getname(path);
1823 int error;
1824
1825 if (IS_ERR(s))
1826 return PTR_ERR(s);
1827
1828 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1829 if (error)
1830 putname(s);
1831 else
1832 *name = s;
1833
1834 return error;
1835}
1836
1837/*
1838 * It's inline, so penalty for filesystems that don't use sticky bit is
1839 * minimal.
1840 */
1841static inline int check_sticky(struct inode *dir, struct inode *inode)
1842{
1843 uid_t fsuid = current_fsuid();
1844
1845 if (!(dir->i_mode & S_ISVTX))
1846 return 0;
1847 if (current_user_ns() != inode_userns(inode))
1848 goto other_userns;
1849 if (inode->i_uid == fsuid)
1850 return 0;
1851 if (dir->i_uid == fsuid)
1852 return 0;
1853
1854other_userns:
1855 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1856}
1857
1858/*
1859 * Check whether we can remove a link victim from directory dir, check
1860 * whether the type of victim is right.
1861 * 1. We can't do it if dir is read-only (done in permission())
1862 * 2. We should have write and exec permissions on dir
1863 * 3. We can't remove anything from append-only dir
1864 * 4. We can't do anything with immutable dir (done in permission())
1865 * 5. If the sticky bit on dir is set we should either
1866 * a. be owner of dir, or
1867 * b. be owner of victim, or
1868 * c. have CAP_FOWNER capability
1869 * 6. If the victim is append-only or immutable we can't do antyhing with
1870 * links pointing to it.
1871 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1872 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1873 * 9. We can't remove a root or mountpoint.
1874 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1875 * nfs_async_unlink().
1876 */
1877static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1878{
1879 int error;
1880
1881 if (!victim->d_inode)
1882 return -ENOENT;
1883
1884 BUG_ON(victim->d_parent->d_inode != dir);
1885 audit_inode_child(victim, dir);
1886
1887 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1888 if (error)
1889 return error;
1890 if (IS_APPEND(dir))
1891 return -EPERM;
1892 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1893 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1894 return -EPERM;
1895 if (isdir) {
1896 if (!S_ISDIR(victim->d_inode->i_mode))
1897 return -ENOTDIR;
1898 if (IS_ROOT(victim))
1899 return -EBUSY;
1900 } else if (S_ISDIR(victim->d_inode->i_mode))
1901 return -EISDIR;
1902 if (IS_DEADDIR(dir))
1903 return -ENOENT;
1904 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1905 return -EBUSY;
1906 return 0;
1907}
1908
1909/* Check whether we can create an object with dentry child in directory
1910 * dir.
1911 * 1. We can't do it if child already exists (open has special treatment for
1912 * this case, but since we are inlined it's OK)
1913 * 2. We can't do it if dir is read-only (done in permission())
1914 * 3. We should have write and exec permissions on dir
1915 * 4. We can't do it if dir is immutable (done in permission())
1916 */
1917static inline int may_create(struct inode *dir, struct dentry *child)
1918{
1919 if (child->d_inode)
1920 return -EEXIST;
1921 if (IS_DEADDIR(dir))
1922 return -ENOENT;
1923 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1924}
1925
1926/*
1927 * p1 and p2 should be directories on the same fs.
1928 */
1929struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1930{
1931 struct dentry *p;
1932
1933 if (p1 == p2) {
1934 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1935 return NULL;
1936 }
1937
1938 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1939
1940 p = d_ancestor(p2, p1);
1941 if (p) {
1942 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1943 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1944 return p;
1945 }
1946
1947 p = d_ancestor(p1, p2);
1948 if (p) {
1949 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1950 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1951 return p;
1952 }
1953
1954 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1955 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1956 return NULL;
1957}
1958
1959void unlock_rename(struct dentry *p1, struct dentry *p2)
1960{
1961 mutex_unlock(&p1->d_inode->i_mutex);
1962 if (p1 != p2) {
1963 mutex_unlock(&p2->d_inode->i_mutex);
1964 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1965 }
1966}
1967
1968int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1969 struct nameidata *nd)
1970{
1971 int error = may_create(dir, dentry);
1972
1973 if (error)
1974 return error;
1975
1976 if (!dir->i_op->create)
1977 return -EACCES; /* shouldn't it be ENOSYS? */
1978 mode &= S_IALLUGO;
1979 mode |= S_IFREG;
1980 error = security_inode_create(dir, dentry, mode);
1981 if (error)
1982 return error;
1983 error = dir->i_op->create(dir, dentry, mode, nd);
1984 if (!error)
1985 fsnotify_create(dir, dentry);
1986 return error;
1987}
1988
1989static int may_open(struct path *path, int acc_mode, int flag)
1990{
1991 struct dentry *dentry = path->dentry;
1992 struct inode *inode = dentry->d_inode;
1993 int error;
1994
1995 /* O_PATH? */
1996 if (!acc_mode)
1997 return 0;
1998
1999 if (!inode)
2000 return -ENOENT;
2001
2002 switch (inode->i_mode & S_IFMT) {
2003 case S_IFLNK:
2004 return -ELOOP;
2005 case S_IFDIR:
2006 if (acc_mode & MAY_WRITE)
2007 return -EISDIR;
2008 break;
2009 case S_IFBLK:
2010 case S_IFCHR:
2011 if (path->mnt->mnt_flags & MNT_NODEV)
2012 return -EACCES;
2013 /*FALLTHRU*/
2014 case S_IFIFO:
2015 case S_IFSOCK:
2016 flag &= ~O_TRUNC;
2017 break;
2018 }
2019
2020 error = inode_permission(inode, acc_mode);
2021 if (error)
2022 return error;
2023
2024 /*
2025 * An append-only file must be opened in append mode for writing.
2026 */
2027 if (IS_APPEND(inode)) {
2028 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2029 return -EPERM;
2030 if (flag & O_TRUNC)
2031 return -EPERM;
2032 }
2033
2034 /* O_NOATIME can only be set by the owner or superuser */
2035 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2036 return -EPERM;
2037
2038 /*
2039 * Ensure there are no outstanding leases on the file.
2040 */
2041 return break_lease(inode, flag);
2042}
2043
2044static int handle_truncate(struct file *filp)
2045{
2046 struct path *path = &filp->f_path;
2047 struct inode *inode = path->dentry->d_inode;
2048 int error = get_write_access(inode);
2049 if (error)
2050 return error;
2051 /*
2052 * Refuse to truncate files with mandatory locks held on them.
2053 */
2054 error = locks_verify_locked(inode);
2055 if (!error)
2056 error = security_path_truncate(path);
2057 if (!error) {
2058 error = do_truncate(path->dentry, 0,
2059 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2060 filp);
2061 }
2062 put_write_access(inode);
2063 return error;
2064}
2065
2066static inline int open_to_namei_flags(int flag)
2067{
2068 if ((flag & O_ACCMODE) == 3)
2069 flag--;
2070 return flag;
2071}
2072
2073/*
2074 * Handle the last step of open()
2075 */
2076static struct file *do_last(struct nameidata *nd, struct path *path,
2077 const struct open_flags *op, const char *pathname)
2078{
2079 struct dentry *dir = nd->path.dentry;
2080 struct dentry *dentry;
2081 int open_flag = op->open_flag;
2082 int will_truncate = open_flag & O_TRUNC;
2083 int want_write = 0;
2084 int acc_mode = op->acc_mode;
2085 struct file *filp;
2086 int error;
2087
2088 nd->flags &= ~LOOKUP_PARENT;
2089 nd->flags |= op->intent;
2090
2091 switch (nd->last_type) {
2092 case LAST_DOTDOT:
2093 case LAST_DOT:
2094 error = handle_dots(nd, nd->last_type);
2095 if (error)
2096 return ERR_PTR(error);
2097 /* fallthrough */
2098 case LAST_ROOT:
2099 error = complete_walk(nd);
2100 if (error)
2101 return ERR_PTR(error);
2102 audit_inode(pathname, nd->path.dentry);
2103 if (open_flag & O_CREAT) {
2104 error = -EISDIR;
2105 goto exit;
2106 }
2107 goto ok;
2108 case LAST_BIND:
2109 error = complete_walk(nd);
2110 if (error)
2111 return ERR_PTR(error);
2112 audit_inode(pathname, dir);
2113 goto ok;
2114 }
2115
2116 if (!(open_flag & O_CREAT)) {
2117 int symlink_ok = 0;
2118 if (nd->last.name[nd->last.len])
2119 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2120 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2121 symlink_ok = 1;
2122 /* we _can_ be in RCU mode here */
2123 error = walk_component(nd, path, &nd->last, LAST_NORM,
2124 !symlink_ok);
2125 if (error < 0)
2126 return ERR_PTR(error);
2127 if (error) /* symlink */
2128 return NULL;
2129 /* sayonara */
2130 error = complete_walk(nd);
2131 if (error)
2132 return ERR_PTR(-ECHILD);
2133
2134 error = -ENOTDIR;
2135 if (nd->flags & LOOKUP_DIRECTORY) {
2136 if (!nd->inode->i_op->lookup)
2137 goto exit;
2138 }
2139 audit_inode(pathname, nd->path.dentry);
2140 goto ok;
2141 }
2142
2143 /* create side of things */
2144 error = complete_walk(nd);
2145 if (error)
2146 return ERR_PTR(error);
2147
2148 audit_inode(pathname, dir);
2149 error = -EISDIR;
2150 /* trailing slashes? */
2151 if (nd->last.name[nd->last.len])
2152 goto exit;
2153
2154 mutex_lock(&dir->d_inode->i_mutex);
2155
2156 dentry = lookup_hash(nd);
2157 error = PTR_ERR(dentry);
2158 if (IS_ERR(dentry)) {
2159 mutex_unlock(&dir->d_inode->i_mutex);
2160 goto exit;
2161 }
2162
2163 path->dentry = dentry;
2164 path->mnt = nd->path.mnt;
2165
2166 /* Negative dentry, just create the file */
2167 if (!dentry->d_inode) {
2168 int mode = op->mode;
2169 if (!IS_POSIXACL(dir->d_inode))
2170 mode &= ~current_umask();
2171 /*
2172 * This write is needed to ensure that a
2173 * rw->ro transition does not occur between
2174 * the time when the file is created and when
2175 * a permanent write count is taken through
2176 * the 'struct file' in nameidata_to_filp().
2177 */
2178 error = mnt_want_write(nd->path.mnt);
2179 if (error)
2180 goto exit_mutex_unlock;
2181 want_write = 1;
2182 /* Don't check for write permission, don't truncate */
2183 open_flag &= ~O_TRUNC;
2184 will_truncate = 0;
2185 acc_mode = MAY_OPEN;
2186 error = security_path_mknod(&nd->path, dentry, mode, 0);
2187 if (error)
2188 goto exit_mutex_unlock;
2189 error = vfs_create(dir->d_inode, dentry, mode, nd);
2190 if (error)
2191 goto exit_mutex_unlock;
2192 mutex_unlock(&dir->d_inode->i_mutex);
2193 dput(nd->path.dentry);
2194 nd->path.dentry = dentry;
2195 goto common;
2196 }
2197
2198 /*
2199 * It already exists.
2200 */
2201 mutex_unlock(&dir->d_inode->i_mutex);
2202 audit_inode(pathname, path->dentry);
2203
2204 error = -EEXIST;
2205 if (open_flag & O_EXCL)
2206 goto exit_dput;
2207
2208 error = follow_managed(path, nd->flags);
2209 if (error < 0)
2210 goto exit_dput;
2211
2212 error = -ENOENT;
2213 if (!path->dentry->d_inode)
2214 goto exit_dput;
2215
2216 if (path->dentry->d_inode->i_op->follow_link)
2217 return NULL;
2218
2219 path_to_nameidata(path, nd);
2220 nd->inode = path->dentry->d_inode;
2221 error = -EISDIR;
2222 if (S_ISDIR(nd->inode->i_mode))
2223 goto exit;
2224ok:
2225 if (!S_ISREG(nd->inode->i_mode))
2226 will_truncate = 0;
2227
2228 if (will_truncate) {
2229 error = mnt_want_write(nd->path.mnt);
2230 if (error)
2231 goto exit;
2232 want_write = 1;
2233 }
2234common:
2235 error = may_open(&nd->path, acc_mode, open_flag);
2236 if (error)
2237 goto exit;
2238 filp = nameidata_to_filp(nd);
2239 if (!IS_ERR(filp)) {
2240 error = ima_file_check(filp, op->acc_mode);
2241 if (error) {
2242 fput(filp);
2243 filp = ERR_PTR(error);
2244 }
2245 }
2246 if (!IS_ERR(filp)) {
2247 if (will_truncate) {
2248 error = handle_truncate(filp);
2249 if (error) {
2250 fput(filp);
2251 filp = ERR_PTR(error);
2252 }
2253 }
2254 }
2255out:
2256 if (want_write)
2257 mnt_drop_write(nd->path.mnt);
2258 path_put(&nd->path);
2259 return filp;
2260
2261exit_mutex_unlock:
2262 mutex_unlock(&dir->d_inode->i_mutex);
2263exit_dput:
2264 path_put_conditional(path, nd);
2265exit:
2266 filp = ERR_PTR(error);
2267 goto out;
2268}
2269
2270static struct file *path_openat(int dfd, const char *pathname,
2271 struct nameidata *nd, const struct open_flags *op, int flags)
2272{
2273 struct file *base = NULL;
2274 struct file *filp;
2275 struct path path;
2276 int error;
2277
2278 filp = get_empty_filp();
2279 if (!filp)
2280 return ERR_PTR(-ENFILE);
2281
2282 filp->f_flags = op->open_flag;
2283 nd->intent.open.file = filp;
2284 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2285 nd->intent.open.create_mode = op->mode;
2286
2287 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2288 if (unlikely(error))
2289 goto out_filp;
2290
2291 current->total_link_count = 0;
2292 error = link_path_walk(pathname, nd);
2293 if (unlikely(error))
2294 goto out_filp;
2295
2296 filp = do_last(nd, &path, op, pathname);
2297 while (unlikely(!filp)) { /* trailing symlink */
2298 struct path link = path;
2299 void *cookie;
2300 if (!(nd->flags & LOOKUP_FOLLOW)) {
2301 path_put_conditional(&path, nd);
2302 path_put(&nd->path);
2303 filp = ERR_PTR(-ELOOP);
2304 break;
2305 }
2306 nd->flags |= LOOKUP_PARENT;
2307 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2308 error = follow_link(&link, nd, &cookie);
2309 if (unlikely(error))
2310 filp = ERR_PTR(error);
2311 else
2312 filp = do_last(nd, &path, op, pathname);
2313 put_link(nd, &link, cookie);
2314 }
2315out:
2316 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2317 path_put(&nd->root);
2318 if (base)
2319 fput(base);
2320 release_open_intent(nd);
2321 return filp;
2322
2323out_filp:
2324 filp = ERR_PTR(error);
2325 goto out;
2326}
2327
2328struct file *do_filp_open(int dfd, const char *pathname,
2329 const struct open_flags *op, int flags)
2330{
2331 struct nameidata nd;
2332 struct file *filp;
2333
2334 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2335 if (unlikely(filp == ERR_PTR(-ECHILD)))
2336 filp = path_openat(dfd, pathname, &nd, op, flags);
2337 if (unlikely(filp == ERR_PTR(-ESTALE)))
2338 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2339 return filp;
2340}
2341
2342struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2343 const char *name, const struct open_flags *op, int flags)
2344{
2345 struct nameidata nd;
2346 struct file *file;
2347
2348 nd.root.mnt = mnt;
2349 nd.root.dentry = dentry;
2350
2351 flags |= LOOKUP_ROOT;
2352
2353 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2354 return ERR_PTR(-ELOOP);
2355
2356 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2357 if (unlikely(file == ERR_PTR(-ECHILD)))
2358 file = path_openat(-1, name, &nd, op, flags);
2359 if (unlikely(file == ERR_PTR(-ESTALE)))
2360 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2361 return file;
2362}
2363
2364struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2365{
2366 struct dentry *dentry = ERR_PTR(-EEXIST);
2367 struct nameidata nd;
2368 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2369 if (error)
2370 return ERR_PTR(error);
2371
2372 /*
2373 * Yucky last component or no last component at all?
2374 * (foo/., foo/.., /////)
2375 */
2376 if (nd.last_type != LAST_NORM)
2377 goto out;
2378 nd.flags &= ~LOOKUP_PARENT;
2379 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2380 nd.intent.open.flags = O_EXCL;
2381
2382 /*
2383 * Do the final lookup.
2384 */
2385 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2386 dentry = lookup_hash(&nd);
2387 if (IS_ERR(dentry))
2388 goto fail;
2389
2390 if (dentry->d_inode)
2391 goto eexist;
2392 /*
2393 * Special case - lookup gave negative, but... we had foo/bar/
2394 * From the vfs_mknod() POV we just have a negative dentry -
2395 * all is fine. Let's be bastards - you had / on the end, you've
2396 * been asking for (non-existent) directory. -ENOENT for you.
2397 */
2398 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2399 dput(dentry);
2400 dentry = ERR_PTR(-ENOENT);
2401 goto fail;
2402 }
2403 *path = nd.path;
2404 return dentry;
2405eexist:
2406 dput(dentry);
2407 dentry = ERR_PTR(-EEXIST);
2408fail:
2409 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2410out:
2411 path_put(&nd.path);
2412 return dentry;
2413}
2414EXPORT_SYMBOL(kern_path_create);
2415
2416struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2417{
2418 char *tmp = getname(pathname);
2419 struct dentry *res;
2420 if (IS_ERR(tmp))
2421 return ERR_CAST(tmp);
2422 res = kern_path_create(dfd, tmp, path, is_dir);
2423 putname(tmp);
2424 return res;
2425}
2426EXPORT_SYMBOL(user_path_create);
2427
2428int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2429{
2430 int error = may_create(dir, dentry);
2431
2432 if (error)
2433 return error;
2434
2435 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2436 !ns_capable(inode_userns(dir), CAP_MKNOD))
2437 return -EPERM;
2438
2439 if (!dir->i_op->mknod)
2440 return -EPERM;
2441
2442 error = devcgroup_inode_mknod(mode, dev);
2443 if (error)
2444 return error;
2445
2446 error = security_inode_mknod(dir, dentry, mode, dev);
2447 if (error)
2448 return error;
2449
2450 error = dir->i_op->mknod(dir, dentry, mode, dev);
2451 if (!error)
2452 fsnotify_create(dir, dentry);
2453 return error;
2454}
2455
2456static int may_mknod(mode_t mode)
2457{
2458 switch (mode & S_IFMT) {
2459 case S_IFREG:
2460 case S_IFCHR:
2461 case S_IFBLK:
2462 case S_IFIFO:
2463 case S_IFSOCK:
2464 case 0: /* zero mode translates to S_IFREG */
2465 return 0;
2466 case S_IFDIR:
2467 return -EPERM;
2468 default:
2469 return -EINVAL;
2470 }
2471}
2472
2473SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2474 unsigned, dev)
2475{
2476 struct dentry *dentry;
2477 struct path path;
2478 int error;
2479
2480 if (S_ISDIR(mode))
2481 return -EPERM;
2482
2483 dentry = user_path_create(dfd, filename, &path, 0);
2484 if (IS_ERR(dentry))
2485 return PTR_ERR(dentry);
2486
2487 if (!IS_POSIXACL(path.dentry->d_inode))
2488 mode &= ~current_umask();
2489 error = may_mknod(mode);
2490 if (error)
2491 goto out_dput;
2492 error = mnt_want_write(path.mnt);
2493 if (error)
2494 goto out_dput;
2495 error = security_path_mknod(&path, dentry, mode, dev);
2496 if (error)
2497 goto out_drop_write;
2498 switch (mode & S_IFMT) {
2499 case 0: case S_IFREG:
2500 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2501 break;
2502 case S_IFCHR: case S_IFBLK:
2503 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2504 new_decode_dev(dev));
2505 break;
2506 case S_IFIFO: case S_IFSOCK:
2507 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2508 break;
2509 }
2510out_drop_write:
2511 mnt_drop_write(path.mnt);
2512out_dput:
2513 dput(dentry);
2514 mutex_unlock(&path.dentry->d_inode->i_mutex);
2515 path_put(&path);
2516
2517 return error;
2518}
2519
2520SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2521{
2522 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2523}
2524
2525int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2526{
2527 int error = may_create(dir, dentry);
2528
2529 if (error)
2530 return error;
2531
2532 if (!dir->i_op->mkdir)
2533 return -EPERM;
2534
2535 mode &= (S_IRWXUGO|S_ISVTX);
2536 error = security_inode_mkdir(dir, dentry, mode);
2537 if (error)
2538 return error;
2539
2540 error = dir->i_op->mkdir(dir, dentry, mode);
2541 if (!error)
2542 fsnotify_mkdir(dir, dentry);
2543 return error;
2544}
2545
2546SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2547{
2548 struct dentry *dentry;
2549 struct path path;
2550 int error;
2551
2552 dentry = user_path_create(dfd, pathname, &path, 1);
2553 if (IS_ERR(dentry))
2554 return PTR_ERR(dentry);
2555
2556 if (!IS_POSIXACL(path.dentry->d_inode))
2557 mode &= ~current_umask();
2558 error = mnt_want_write(path.mnt);
2559 if (error)
2560 goto out_dput;
2561 error = security_path_mkdir(&path, dentry, mode);
2562 if (error)
2563 goto out_drop_write;
2564 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2565out_drop_write:
2566 mnt_drop_write(path.mnt);
2567out_dput:
2568 dput(dentry);
2569 mutex_unlock(&path.dentry->d_inode->i_mutex);
2570 path_put(&path);
2571 return error;
2572}
2573
2574SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2575{
2576 return sys_mkdirat(AT_FDCWD, pathname, mode);
2577}
2578
2579/*
2580 * The dentry_unhash() helper will try to drop the dentry early: we
2581 * should have a usage count of 2 if we're the only user of this
2582 * dentry, and if that is true (possibly after pruning the dcache),
2583 * then we drop the dentry now.
2584 *
2585 * A low-level filesystem can, if it choses, legally
2586 * do a
2587 *
2588 * if (!d_unhashed(dentry))
2589 * return -EBUSY;
2590 *
2591 * if it cannot handle the case of removing a directory
2592 * that is still in use by something else..
2593 */
2594void dentry_unhash(struct dentry *dentry)
2595{
2596 shrink_dcache_parent(dentry);
2597 spin_lock(&dentry->d_lock);
2598 if (dentry->d_count == 1)
2599 __d_drop(dentry);
2600 spin_unlock(&dentry->d_lock);
2601}
2602
2603int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2604{
2605 int error = may_delete(dir, dentry, 1);
2606
2607 if (error)
2608 return error;
2609
2610 if (!dir->i_op->rmdir)
2611 return -EPERM;
2612
2613 dget(dentry);
2614 mutex_lock(&dentry->d_inode->i_mutex);
2615
2616 error = -EBUSY;
2617 if (d_mountpoint(dentry))
2618 goto out;
2619
2620 error = security_inode_rmdir(dir, dentry);
2621 if (error)
2622 goto out;
2623
2624 shrink_dcache_parent(dentry);
2625 error = dir->i_op->rmdir(dir, dentry);
2626 if (error)
2627 goto out;
2628
2629 dentry->d_inode->i_flags |= S_DEAD;
2630 dont_mount(dentry);
2631
2632out:
2633 mutex_unlock(&dentry->d_inode->i_mutex);
2634 dput(dentry);
2635 if (!error)
2636 d_delete(dentry);
2637 return error;
2638}
2639
2640static long do_rmdir(int dfd, const char __user *pathname)
2641{
2642 int error = 0;
2643 char * name;
2644 struct dentry *dentry;
2645 struct nameidata nd;
2646
2647 error = user_path_parent(dfd, pathname, &nd, &name);
2648 if (error)
2649 return error;
2650
2651 switch(nd.last_type) {
2652 case LAST_DOTDOT:
2653 error = -ENOTEMPTY;
2654 goto exit1;
2655 case LAST_DOT:
2656 error = -EINVAL;
2657 goto exit1;
2658 case LAST_ROOT:
2659 error = -EBUSY;
2660 goto exit1;
2661 }
2662
2663 nd.flags &= ~LOOKUP_PARENT;
2664
2665 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2666 dentry = lookup_hash(&nd);
2667 error = PTR_ERR(dentry);
2668 if (IS_ERR(dentry))
2669 goto exit2;
2670 if (!dentry->d_inode) {
2671 error = -ENOENT;
2672 goto exit3;
2673 }
2674 error = mnt_want_write(nd.path.mnt);
2675 if (error)
2676 goto exit3;
2677 error = security_path_rmdir(&nd.path, dentry);
2678 if (error)
2679 goto exit4;
2680 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2681exit4:
2682 mnt_drop_write(nd.path.mnt);
2683exit3:
2684 dput(dentry);
2685exit2:
2686 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2687exit1:
2688 path_put(&nd.path);
2689 putname(name);
2690 return error;
2691}
2692
2693SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2694{
2695 return do_rmdir(AT_FDCWD, pathname);
2696}
2697
2698int vfs_unlink(struct inode *dir, struct dentry *dentry)
2699{
2700 int error = may_delete(dir, dentry, 0);
2701
2702 if (error)
2703 return error;
2704
2705 if (!dir->i_op->unlink)
2706 return -EPERM;
2707
2708 mutex_lock(&dentry->d_inode->i_mutex);
2709 if (d_mountpoint(dentry))
2710 error = -EBUSY;
2711 else {
2712 error = security_inode_unlink(dir, dentry);
2713 if (!error) {
2714 error = dir->i_op->unlink(dir, dentry);
2715 if (!error)
2716 dont_mount(dentry);
2717 }
2718 }
2719 mutex_unlock(&dentry->d_inode->i_mutex);
2720
2721 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2722 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2723 fsnotify_link_count(dentry->d_inode);
2724 d_delete(dentry);
2725 }
2726
2727 return error;
2728}
2729
2730/*
2731 * Make sure that the actual truncation of the file will occur outside its
2732 * directory's i_mutex. Truncate can take a long time if there is a lot of
2733 * writeout happening, and we don't want to prevent access to the directory
2734 * while waiting on the I/O.
2735 */
2736static long do_unlinkat(int dfd, const char __user *pathname)
2737{
2738 int error;
2739 char *name;
2740 struct dentry *dentry;
2741 struct nameidata nd;
2742 struct inode *inode = NULL;
2743
2744 error = user_path_parent(dfd, pathname, &nd, &name);
2745 if (error)
2746 return error;
2747
2748 error = -EISDIR;
2749 if (nd.last_type != LAST_NORM)
2750 goto exit1;
2751
2752 nd.flags &= ~LOOKUP_PARENT;
2753
2754 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2755 dentry = lookup_hash(&nd);
2756 error = PTR_ERR(dentry);
2757 if (!IS_ERR(dentry)) {
2758 /* Why not before? Because we want correct error value */
2759 if (nd.last.name[nd.last.len])
2760 goto slashes;
2761 inode = dentry->d_inode;
2762 if (!inode)
2763 goto slashes;
2764 ihold(inode);
2765 error = mnt_want_write(nd.path.mnt);
2766 if (error)
2767 goto exit2;
2768 error = security_path_unlink(&nd.path, dentry);
2769 if (error)
2770 goto exit3;
2771 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2772exit3:
2773 mnt_drop_write(nd.path.mnt);
2774 exit2:
2775 dput(dentry);
2776 }
2777 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2778 if (inode)
2779 iput(inode); /* truncate the inode here */
2780exit1:
2781 path_put(&nd.path);
2782 putname(name);
2783 return error;
2784
2785slashes:
2786 error = !dentry->d_inode ? -ENOENT :
2787 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2788 goto exit2;
2789}
2790
2791SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2792{
2793 if ((flag & ~AT_REMOVEDIR) != 0)
2794 return -EINVAL;
2795
2796 if (flag & AT_REMOVEDIR)
2797 return do_rmdir(dfd, pathname);
2798
2799 return do_unlinkat(dfd, pathname);
2800}
2801
2802SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2803{
2804 return do_unlinkat(AT_FDCWD, pathname);
2805}
2806
2807int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2808{
2809 int error = may_create(dir, dentry);
2810
2811 if (error)
2812 return error;
2813
2814 if (!dir->i_op->symlink)
2815 return -EPERM;
2816
2817 error = security_inode_symlink(dir, dentry, oldname);
2818 if (error)
2819 return error;
2820
2821 error = dir->i_op->symlink(dir, dentry, oldname);
2822 if (!error)
2823 fsnotify_create(dir, dentry);
2824 return error;
2825}
2826
2827SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2828 int, newdfd, const char __user *, newname)
2829{
2830 int error;
2831 char *from;
2832 struct dentry *dentry;
2833 struct path path;
2834
2835 from = getname(oldname);
2836 if (IS_ERR(from))
2837 return PTR_ERR(from);
2838
2839 dentry = user_path_create(newdfd, newname, &path, 0);
2840 error = PTR_ERR(dentry);
2841 if (IS_ERR(dentry))
2842 goto out_putname;
2843
2844 error = mnt_want_write(path.mnt);
2845 if (error)
2846 goto out_dput;
2847 error = security_path_symlink(&path, dentry, from);
2848 if (error)
2849 goto out_drop_write;
2850 error = vfs_symlink(path.dentry->d_inode, dentry, from);
2851out_drop_write:
2852 mnt_drop_write(path.mnt);
2853out_dput:
2854 dput(dentry);
2855 mutex_unlock(&path.dentry->d_inode->i_mutex);
2856 path_put(&path);
2857out_putname:
2858 putname(from);
2859 return error;
2860}
2861
2862SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2863{
2864 return sys_symlinkat(oldname, AT_FDCWD, newname);
2865}
2866
2867int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2868{
2869 struct inode *inode = old_dentry->d_inode;
2870 int error;
2871
2872 if (!inode)
2873 return -ENOENT;
2874
2875 error = may_create(dir, new_dentry);
2876 if (error)
2877 return error;
2878
2879 if (dir->i_sb != inode->i_sb)
2880 return -EXDEV;
2881
2882 /*
2883 * A link to an append-only or immutable file cannot be created.
2884 */
2885 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2886 return -EPERM;
2887 if (!dir->i_op->link)
2888 return -EPERM;
2889 if (S_ISDIR(inode->i_mode))
2890 return -EPERM;
2891
2892 error = security_inode_link(old_dentry, dir, new_dentry);
2893 if (error)
2894 return error;
2895
2896 mutex_lock(&inode->i_mutex);
2897 /* Make sure we don't allow creating hardlink to an unlinked file */
2898 if (inode->i_nlink == 0)
2899 error = -ENOENT;
2900 else
2901 error = dir->i_op->link(old_dentry, dir, new_dentry);
2902 mutex_unlock(&inode->i_mutex);
2903 if (!error)
2904 fsnotify_link(dir, inode, new_dentry);
2905 return error;
2906}
2907
2908/*
2909 * Hardlinks are often used in delicate situations. We avoid
2910 * security-related surprises by not following symlinks on the
2911 * newname. --KAB
2912 *
2913 * We don't follow them on the oldname either to be compatible
2914 * with linux 2.0, and to avoid hard-linking to directories
2915 * and other special files. --ADM
2916 */
2917SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2918 int, newdfd, const char __user *, newname, int, flags)
2919{
2920 struct dentry *new_dentry;
2921 struct path old_path, new_path;
2922 int how = 0;
2923 int error;
2924
2925 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
2926 return -EINVAL;
2927 /*
2928 * To use null names we require CAP_DAC_READ_SEARCH
2929 * This ensures that not everyone will be able to create
2930 * handlink using the passed filedescriptor.
2931 */
2932 if (flags & AT_EMPTY_PATH) {
2933 if (!capable(CAP_DAC_READ_SEARCH))
2934 return -ENOENT;
2935 how = LOOKUP_EMPTY;
2936 }
2937
2938 if (flags & AT_SYMLINK_FOLLOW)
2939 how |= LOOKUP_FOLLOW;
2940
2941 error = user_path_at(olddfd, oldname, how, &old_path);
2942 if (error)
2943 return error;
2944
2945 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
2946 error = PTR_ERR(new_dentry);
2947 if (IS_ERR(new_dentry))
2948 goto out;
2949
2950 error = -EXDEV;
2951 if (old_path.mnt != new_path.mnt)
2952 goto out_dput;
2953 error = mnt_want_write(new_path.mnt);
2954 if (error)
2955 goto out_dput;
2956 error = security_path_link(old_path.dentry, &new_path, new_dentry);
2957 if (error)
2958 goto out_drop_write;
2959 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
2960out_drop_write:
2961 mnt_drop_write(new_path.mnt);
2962out_dput:
2963 dput(new_dentry);
2964 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2965 path_put(&new_path);
2966out:
2967 path_put(&old_path);
2968
2969 return error;
2970}
2971
2972SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2973{
2974 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2975}
2976
2977/*
2978 * The worst of all namespace operations - renaming directory. "Perverted"
2979 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2980 * Problems:
2981 * a) we can get into loop creation. Check is done in is_subdir().
2982 * b) race potential - two innocent renames can create a loop together.
2983 * That's where 4.4 screws up. Current fix: serialization on
2984 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2985 * story.
2986 * c) we have to lock _three_ objects - parents and victim (if it exists).
2987 * And that - after we got ->i_mutex on parents (until then we don't know
2988 * whether the target exists). Solution: try to be smart with locking
2989 * order for inodes. We rely on the fact that tree topology may change
2990 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2991 * move will be locked. Thus we can rank directories by the tree
2992 * (ancestors first) and rank all non-directories after them.
2993 * That works since everybody except rename does "lock parent, lookup,
2994 * lock child" and rename is under ->s_vfs_rename_mutex.
2995 * HOWEVER, it relies on the assumption that any object with ->lookup()
2996 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2997 * we'd better make sure that there's no link(2) for them.
2998 * d) conversion from fhandle to dentry may come in the wrong moment - when
2999 * we are removing the target. Solution: we will have to grab ->i_mutex
3000 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3001 * ->i_mutex on parents, which works but leads to some truly excessive
3002 * locking].
3003 */
3004static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3005 struct inode *new_dir, struct dentry *new_dentry)
3006{
3007 int error = 0;
3008 struct inode *target = new_dentry->d_inode;
3009
3010 /*
3011 * If we are going to change the parent - check write permissions,
3012 * we'll need to flip '..'.
3013 */
3014 if (new_dir != old_dir) {
3015 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3016 if (error)
3017 return error;
3018 }
3019
3020 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3021 if (error)
3022 return error;
3023
3024 dget(new_dentry);
3025 if (target)
3026 mutex_lock(&target->i_mutex);
3027
3028 error = -EBUSY;
3029 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3030 goto out;
3031
3032 if (target)
3033 shrink_dcache_parent(new_dentry);
3034 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3035 if (error)
3036 goto out;
3037
3038 if (target) {
3039 target->i_flags |= S_DEAD;
3040 dont_mount(new_dentry);
3041 }
3042out:
3043 if (target)
3044 mutex_unlock(&target->i_mutex);
3045 dput(new_dentry);
3046 if (!error)
3047 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3048 d_move(old_dentry,new_dentry);
3049 return error;
3050}
3051
3052static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3053 struct inode *new_dir, struct dentry *new_dentry)
3054{
3055 struct inode *target = new_dentry->d_inode;
3056 int error;
3057
3058 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3059 if (error)
3060 return error;
3061
3062 dget(new_dentry);
3063 if (target)
3064 mutex_lock(&target->i_mutex);
3065
3066 error = -EBUSY;
3067 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3068 goto out;
3069
3070 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3071 if (error)
3072 goto out;
3073
3074 if (target)
3075 dont_mount(new_dentry);
3076 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3077 d_move(old_dentry, new_dentry);
3078out:
3079 if (target)
3080 mutex_unlock(&target->i_mutex);
3081 dput(new_dentry);
3082 return error;
3083}
3084
3085int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3086 struct inode *new_dir, struct dentry *new_dentry)
3087{
3088 int error;
3089 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3090 const unsigned char *old_name;
3091
3092 if (old_dentry->d_inode == new_dentry->d_inode)
3093 return 0;
3094
3095 error = may_delete(old_dir, old_dentry, is_dir);
3096 if (error)
3097 return error;
3098
3099 if (!new_dentry->d_inode)
3100 error = may_create(new_dir, new_dentry);
3101 else
3102 error = may_delete(new_dir, new_dentry, is_dir);
3103 if (error)
3104 return error;
3105
3106 if (!old_dir->i_op->rename)
3107 return -EPERM;
3108
3109 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3110
3111 if (is_dir)
3112 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3113 else
3114 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3115 if (!error)
3116 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3117 new_dentry->d_inode, old_dentry);
3118 fsnotify_oldname_free(old_name);
3119
3120 return error;
3121}
3122
3123SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3124 int, newdfd, const char __user *, newname)
3125{
3126 struct dentry *old_dir, *new_dir;
3127 struct dentry *old_dentry, *new_dentry;
3128 struct dentry *trap;
3129 struct nameidata oldnd, newnd;
3130 char *from;
3131 char *to;
3132 int error;
3133
3134 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3135 if (error)
3136 goto exit;
3137
3138 error = user_path_parent(newdfd, newname, &newnd, &to);
3139 if (error)
3140 goto exit1;
3141
3142 error = -EXDEV;
3143 if (oldnd.path.mnt != newnd.path.mnt)
3144 goto exit2;
3145
3146 old_dir = oldnd.path.dentry;
3147 error = -EBUSY;
3148 if (oldnd.last_type != LAST_NORM)
3149 goto exit2;
3150
3151 new_dir = newnd.path.dentry;
3152 if (newnd.last_type != LAST_NORM)
3153 goto exit2;
3154
3155 oldnd.flags &= ~LOOKUP_PARENT;
3156 newnd.flags &= ~LOOKUP_PARENT;
3157 newnd.flags |= LOOKUP_RENAME_TARGET;
3158
3159 trap = lock_rename(new_dir, old_dir);
3160
3161 old_dentry = lookup_hash(&oldnd);
3162 error = PTR_ERR(old_dentry);
3163 if (IS_ERR(old_dentry))
3164 goto exit3;
3165 /* source must exist */
3166 error = -ENOENT;
3167 if (!old_dentry->d_inode)
3168 goto exit4;
3169 /* unless the source is a directory trailing slashes give -ENOTDIR */
3170 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3171 error = -ENOTDIR;
3172 if (oldnd.last.name[oldnd.last.len])
3173 goto exit4;
3174 if (newnd.last.name[newnd.last.len])
3175 goto exit4;
3176 }
3177 /* source should not be ancestor of target */
3178 error = -EINVAL;
3179 if (old_dentry == trap)
3180 goto exit4;
3181 new_dentry = lookup_hash(&newnd);
3182 error = PTR_ERR(new_dentry);
3183 if (IS_ERR(new_dentry))
3184 goto exit4;
3185 /* target should not be an ancestor of source */
3186 error = -ENOTEMPTY;
3187 if (new_dentry == trap)
3188 goto exit5;
3189
3190 error = mnt_want_write(oldnd.path.mnt);
3191 if (error)
3192 goto exit5;
3193 error = security_path_rename(&oldnd.path, old_dentry,
3194 &newnd.path, new_dentry);
3195 if (error)
3196 goto exit6;
3197 error = vfs_rename(old_dir->d_inode, old_dentry,
3198 new_dir->d_inode, new_dentry);
3199exit6:
3200 mnt_drop_write(oldnd.path.mnt);
3201exit5:
3202 dput(new_dentry);
3203exit4:
3204 dput(old_dentry);
3205exit3:
3206 unlock_rename(new_dir, old_dir);
3207exit2:
3208 path_put(&newnd.path);
3209 putname(to);
3210exit1:
3211 path_put(&oldnd.path);
3212 putname(from);
3213exit:
3214 return error;
3215}
3216
3217SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3218{
3219 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3220}
3221
3222int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3223{
3224 int len;
3225
3226 len = PTR_ERR(link);
3227 if (IS_ERR(link))
3228 goto out;
3229
3230 len = strlen(link);
3231 if (len > (unsigned) buflen)
3232 len = buflen;
3233 if (copy_to_user(buffer, link, len))
3234 len = -EFAULT;
3235out:
3236 return len;
3237}
3238
3239/*
3240 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3241 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3242 * using) it for any given inode is up to filesystem.
3243 */
3244int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3245{
3246 struct nameidata nd;
3247 void *cookie;
3248 int res;
3249
3250 nd.depth = 0;
3251 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3252 if (IS_ERR(cookie))
3253 return PTR_ERR(cookie);
3254
3255 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3256 if (dentry->d_inode->i_op->put_link)
3257 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3258 return res;
3259}
3260
3261int vfs_follow_link(struct nameidata *nd, const char *link)
3262{
3263 return __vfs_follow_link(nd, link);
3264}
3265
3266/* get the link contents into pagecache */
3267static char *page_getlink(struct dentry * dentry, struct page **ppage)
3268{
3269 char *kaddr;
3270 struct page *page;
3271 struct address_space *mapping = dentry->d_inode->i_mapping;
3272 page = read_mapping_page(mapping, 0, NULL);
3273 if (IS_ERR(page))
3274 return (char*)page;
3275 *ppage = page;
3276 kaddr = kmap(page);
3277 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3278 return kaddr;
3279}
3280
3281int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3282{
3283 struct page *page = NULL;
3284 char *s = page_getlink(dentry, &page);
3285 int res = vfs_readlink(dentry,buffer,buflen,s);
3286 if (page) {
3287 kunmap(page);
3288 page_cache_release(page);
3289 }
3290 return res;
3291}
3292
3293void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3294{
3295 struct page *page = NULL;
3296 nd_set_link(nd, page_getlink(dentry, &page));
3297 return page;
3298}
3299
3300void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3301{
3302 struct page *page = cookie;
3303
3304 if (page) {
3305 kunmap(page);
3306 page_cache_release(page);
3307 }
3308}
3309
3310/*
3311 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3312 */
3313int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3314{
3315 struct address_space *mapping = inode->i_mapping;
3316 struct page *page;
3317 void *fsdata;
3318 int err;
3319 char *kaddr;
3320 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3321 if (nofs)
3322 flags |= AOP_FLAG_NOFS;
3323
3324retry:
3325 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3326 flags, &page, &fsdata);
3327 if (err)
3328 goto fail;
3329
3330 kaddr = kmap_atomic(page, KM_USER0);
3331 memcpy(kaddr, symname, len-1);
3332 kunmap_atomic(kaddr, KM_USER0);
3333
3334 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3335 page, fsdata);
3336 if (err < 0)
3337 goto fail;
3338 if (err < len-1)
3339 goto retry;
3340
3341 mark_inode_dirty(inode);
3342 return 0;
3343fail:
3344 return err;
3345}
3346
3347int page_symlink(struct inode *inode, const char *symname, int len)
3348{
3349 return __page_symlink(inode, symname, len,
3350 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3351}
3352
3353const struct inode_operations page_symlink_inode_operations = {
3354 .readlink = generic_readlink,
3355 .follow_link = page_follow_link_light,
3356 .put_link = page_put_link,
3357};
3358
3359EXPORT_SYMBOL(user_path_at);
3360EXPORT_SYMBOL(follow_down_one);
3361EXPORT_SYMBOL(follow_down);
3362EXPORT_SYMBOL(follow_up);
3363EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3364EXPORT_SYMBOL(getname);
3365EXPORT_SYMBOL(lock_rename);
3366EXPORT_SYMBOL(lookup_one_len);
3367EXPORT_SYMBOL(page_follow_link_light);
3368EXPORT_SYMBOL(page_put_link);
3369EXPORT_SYMBOL(page_readlink);
3370EXPORT_SYMBOL(__page_symlink);
3371EXPORT_SYMBOL(page_symlink);
3372EXPORT_SYMBOL(page_symlink_inode_operations);
3373EXPORT_SYMBOL(kern_path);
3374EXPORT_SYMBOL(vfs_path_lookup);
3375EXPORT_SYMBOL(inode_permission);
3376EXPORT_SYMBOL(unlock_rename);
3377EXPORT_SYMBOL(vfs_create);
3378EXPORT_SYMBOL(vfs_follow_link);
3379EXPORT_SYMBOL(vfs_link);
3380EXPORT_SYMBOL(vfs_mkdir);
3381EXPORT_SYMBOL(vfs_mknod);
3382EXPORT_SYMBOL(generic_permission);
3383EXPORT_SYMBOL(vfs_readlink);
3384EXPORT_SYMBOL(vfs_rename);
3385EXPORT_SYMBOL(vfs_rmdir);
3386EXPORT_SYMBOL(vfs_symlink);
3387EXPORT_SYMBOL(vfs_unlink);
3388EXPORT_SYMBOL(dentry_unhash);
3389EXPORT_SYMBOL(generic_readlink);