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/export.h>
19#include <linux/kernel.h>
20#include <linux/slab.h>
21#include <linux/fs.h>
22#include <linux/namei.h>
23#include <linux/pagemap.h>
24#include <linux/fsnotify.h>
25#include <linux/personality.h>
26#include <linux/security.h>
27#include <linux/ima.h>
28#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
31#include <linux/capability.h>
32#include <linux/file.h>
33#include <linux/fcntl.h>
34#include <linux/device_cgroup.h>
35#include <linux/fs_struct.h>
36#include <linux/posix_acl.h>
37#include <linux/hash.h>
38#include <linux/bitops.h>
39#include <linux/init_task.h>
40#include <linux/uaccess.h>
41
42#include "internal.h"
43#include "mount.h"
44
45/* [Feb-1997 T. Schoebel-Theuer]
46 * Fundamental changes in the pathname lookup mechanisms (namei)
47 * were necessary because of omirr. The reason is that omirr needs
48 * to know the _real_ pathname, not the user-supplied one, in case
49 * of symlinks (and also when transname replacements occur).
50 *
51 * The new code replaces the old recursive symlink resolution with
52 * an iterative one (in case of non-nested symlink chains). It does
53 * this with calls to <fs>_follow_link().
54 * As a side effect, dir_namei(), _namei() and follow_link() are now
55 * replaced with a single function lookup_dentry() that can handle all
56 * the special cases of the former code.
57 *
58 * With the new dcache, the pathname is stored at each inode, at least as
59 * long as the refcount of the inode is positive. As a side effect, the
60 * size of the dcache depends on the inode cache and thus is dynamic.
61 *
62 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
63 * resolution to correspond with current state of the code.
64 *
65 * Note that the symlink resolution is not *completely* iterative.
66 * There is still a significant amount of tail- and mid- recursion in
67 * the algorithm. Also, note that <fs>_readlink() is not used in
68 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
69 * may return different results than <fs>_follow_link(). Many virtual
70 * filesystems (including /proc) exhibit this behavior.
71 */
72
73/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
74 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
75 * and the name already exists in form of a symlink, try to create the new
76 * name indicated by the symlink. The old code always complained that the
77 * name already exists, due to not following the symlink even if its target
78 * is nonexistent. The new semantics affects also mknod() and link() when
79 * the name is a symlink pointing to a non-existent name.
80 *
81 * I don't know which semantics is the right one, since I have no access
82 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
83 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
84 * "old" one. Personally, I think the new semantics is much more logical.
85 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
86 * file does succeed in both HP-UX and SunOs, but not in Solaris
87 * and in the old Linux semantics.
88 */
89
90/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
91 * semantics. See the comments in "open_namei" and "do_link" below.
92 *
93 * [10-Sep-98 Alan Modra] Another symlink change.
94 */
95
96/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
97 * inside the path - always follow.
98 * in the last component in creation/removal/renaming - never follow.
99 * if LOOKUP_FOLLOW passed - follow.
100 * if the pathname has trailing slashes - follow.
101 * otherwise - don't follow.
102 * (applied in that order).
103 *
104 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
105 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
106 * During the 2.4 we need to fix the userland stuff depending on it -
107 * hopefully we will be able to get rid of that wart in 2.5. So far only
108 * XEmacs seems to be relying on it...
109 */
110/*
111 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
112 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
113 * any extra contention...
114 */
115
116/* In order to reduce some races, while at the same time doing additional
117 * checking and hopefully speeding things up, we copy filenames to the
118 * kernel data space before using them..
119 *
120 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
121 * PATH_MAX includes the nul terminator --RR.
122 */
123
124#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125
126struct filename *
127getname_flags(const char __user *filename, int flags, int *empty)
128{
129 struct filename *result;
130 char *kname;
131 int len;
132
133 result = audit_reusename(filename);
134 if (result)
135 return result;
136
137 result = __getname();
138 if (unlikely(!result))
139 return ERR_PTR(-ENOMEM);
140
141 /*
142 * First, try to embed the struct filename inside the names_cache
143 * allocation
144 */
145 kname = (char *)result->iname;
146 result->name = kname;
147
148 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
149 if (unlikely(len < 0)) {
150 __putname(result);
151 return ERR_PTR(len);
152 }
153
154 /*
155 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
156 * separate struct filename so we can dedicate the entire
157 * names_cache allocation for the pathname, and re-do the copy from
158 * userland.
159 */
160 if (unlikely(len == EMBEDDED_NAME_MAX)) {
161 const size_t size = offsetof(struct filename, iname[1]);
162 kname = (char *)result;
163
164 /*
165 * size is chosen that way we to guarantee that
166 * result->iname[0] is within the same object and that
167 * kname can't be equal to result->iname, no matter what.
168 */
169 result = kzalloc(size, GFP_KERNEL);
170 if (unlikely(!result)) {
171 __putname(kname);
172 return ERR_PTR(-ENOMEM);
173 }
174 result->name = kname;
175 len = strncpy_from_user(kname, filename, PATH_MAX);
176 if (unlikely(len < 0)) {
177 __putname(kname);
178 kfree(result);
179 return ERR_PTR(len);
180 }
181 if (unlikely(len == PATH_MAX)) {
182 __putname(kname);
183 kfree(result);
184 return ERR_PTR(-ENAMETOOLONG);
185 }
186 }
187
188 result->refcnt = 1;
189 /* The empty path is special. */
190 if (unlikely(!len)) {
191 if (empty)
192 *empty = 1;
193 if (!(flags & LOOKUP_EMPTY)) {
194 putname(result);
195 return ERR_PTR(-ENOENT);
196 }
197 }
198
199 result->uptr = filename;
200 result->aname = NULL;
201 audit_getname(result);
202 return result;
203}
204
205struct filename *
206getname(const char __user * filename)
207{
208 return getname_flags(filename, 0, NULL);
209}
210
211struct filename *
212getname_kernel(const char * filename)
213{
214 struct filename *result;
215 int len = strlen(filename) + 1;
216
217 result = __getname();
218 if (unlikely(!result))
219 return ERR_PTR(-ENOMEM);
220
221 if (len <= EMBEDDED_NAME_MAX) {
222 result->name = (char *)result->iname;
223 } else if (len <= PATH_MAX) {
224 struct filename *tmp;
225
226 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
227 if (unlikely(!tmp)) {
228 __putname(result);
229 return ERR_PTR(-ENOMEM);
230 }
231 tmp->name = (char *)result;
232 result = tmp;
233 } else {
234 __putname(result);
235 return ERR_PTR(-ENAMETOOLONG);
236 }
237 memcpy((char *)result->name, filename, len);
238 result->uptr = NULL;
239 result->aname = NULL;
240 result->refcnt = 1;
241 audit_getname(result);
242
243 return result;
244}
245
246void putname(struct filename *name)
247{
248 BUG_ON(name->refcnt <= 0);
249
250 if (--name->refcnt > 0)
251 return;
252
253 if (name->name != name->iname) {
254 __putname(name->name);
255 kfree(name);
256 } else
257 __putname(name);
258}
259
260static int check_acl(struct inode *inode, int mask)
261{
262#ifdef CONFIG_FS_POSIX_ACL
263 struct posix_acl *acl;
264
265 if (mask & MAY_NOT_BLOCK) {
266 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 if (!acl)
268 return -EAGAIN;
269 /* no ->get_acl() calls in RCU mode... */
270 if (is_uncached_acl(acl))
271 return -ECHILD;
272 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273 }
274
275 acl = get_acl(inode, ACL_TYPE_ACCESS);
276 if (IS_ERR(acl))
277 return PTR_ERR(acl);
278 if (acl) {
279 int error = posix_acl_permission(inode, acl, mask);
280 posix_acl_release(acl);
281 return error;
282 }
283#endif
284
285 return -EAGAIN;
286}
287
288/*
289 * This does the basic permission checking
290 */
291static int acl_permission_check(struct inode *inode, int mask)
292{
293 unsigned int mode = inode->i_mode;
294
295 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296 mode >>= 6;
297 else {
298 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
299 int error = check_acl(inode, mask);
300 if (error != -EAGAIN)
301 return error;
302 }
303
304 if (in_group_p(inode->i_gid))
305 mode >>= 3;
306 }
307
308 /*
309 * If the DACs are ok we don't need any capability check.
310 */
311 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
312 return 0;
313 return -EACCES;
314}
315
316/**
317 * generic_permission - check for access rights on a Posix-like filesystem
318 * @inode: inode to check access rights for
319 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
320 *
321 * Used to check for read/write/execute permissions on a file.
322 * We use "fsuid" for this, letting us set arbitrary permissions
323 * for filesystem access without changing the "normal" uids which
324 * are used for other things.
325 *
326 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
327 * request cannot be satisfied (eg. requires blocking or too much complexity).
328 * It would then be called again in ref-walk mode.
329 */
330int generic_permission(struct inode *inode, int mask)
331{
332 int ret;
333
334 /*
335 * Do the basic permission checks.
336 */
337 ret = acl_permission_check(inode, mask);
338 if (ret != -EACCES)
339 return ret;
340
341 if (S_ISDIR(inode->i_mode)) {
342 /* DACs are overridable for directories */
343 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
344 return 0;
345 if (!(mask & MAY_WRITE))
346 if (capable_wrt_inode_uidgid(inode,
347 CAP_DAC_READ_SEARCH))
348 return 0;
349 return -EACCES;
350 }
351 /*
352 * Read/write DACs are always overridable.
353 * Executable DACs are overridable when there is
354 * at least one exec bit set.
355 */
356 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
357 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
358 return 0;
359
360 /*
361 * Searching includes executable on directories, else just read.
362 */
363 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
364 if (mask == MAY_READ)
365 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
366 return 0;
367
368 return -EACCES;
369}
370EXPORT_SYMBOL(generic_permission);
371
372/*
373 * We _really_ want to just do "generic_permission()" without
374 * even looking at the inode->i_op values. So we keep a cache
375 * flag in inode->i_opflags, that says "this has not special
376 * permission function, use the fast case".
377 */
378static inline int do_inode_permission(struct inode *inode, int mask)
379{
380 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
381 if (likely(inode->i_op->permission))
382 return inode->i_op->permission(inode, mask);
383
384 /* This gets set once for the inode lifetime */
385 spin_lock(&inode->i_lock);
386 inode->i_opflags |= IOP_FASTPERM;
387 spin_unlock(&inode->i_lock);
388 }
389 return generic_permission(inode, mask);
390}
391
392/**
393 * __inode_permission - Check for access rights to a given inode
394 * @inode: Inode to check permission on
395 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
396 *
397 * Check for read/write/execute permissions on an inode.
398 *
399 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
400 *
401 * This does not check for a read-only file system. You probably want
402 * inode_permission().
403 */
404int __inode_permission(struct inode *inode, int mask)
405{
406 int retval;
407
408 if (unlikely(mask & MAY_WRITE)) {
409 /*
410 * Nobody gets write access to an immutable file.
411 */
412 if (IS_IMMUTABLE(inode))
413 return -EPERM;
414
415 /*
416 * Updating mtime will likely cause i_uid and i_gid to be
417 * written back improperly if their true value is unknown
418 * to the vfs.
419 */
420 if (HAS_UNMAPPED_ID(inode))
421 return -EACCES;
422 }
423
424 retval = do_inode_permission(inode, mask);
425 if (retval)
426 return retval;
427
428 retval = devcgroup_inode_permission(inode, mask);
429 if (retval)
430 return retval;
431
432 return security_inode_permission(inode, mask);
433}
434EXPORT_SYMBOL(__inode_permission);
435
436/**
437 * sb_permission - Check superblock-level permissions
438 * @sb: Superblock of inode to check permission on
439 * @inode: Inode to check permission on
440 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
441 *
442 * Separate out file-system wide checks from inode-specific permission checks.
443 */
444static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
445{
446 if (unlikely(mask & MAY_WRITE)) {
447 umode_t mode = inode->i_mode;
448
449 /* Nobody gets write access to a read-only fs. */
450 if ((sb->s_flags & MS_RDONLY) &&
451 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
452 return -EROFS;
453 }
454 return 0;
455}
456
457/**
458 * inode_permission - Check for access rights to a given inode
459 * @inode: Inode to check permission on
460 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
461 *
462 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
463 * this, letting us set arbitrary permissions for filesystem access without
464 * changing the "normal" UIDs which are used for other things.
465 *
466 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
467 */
468int inode_permission(struct inode *inode, int mask)
469{
470 int retval;
471
472 retval = sb_permission(inode->i_sb, inode, mask);
473 if (retval)
474 return retval;
475 return __inode_permission(inode, mask);
476}
477EXPORT_SYMBOL(inode_permission);
478
479/**
480 * path_get - get a reference to a path
481 * @path: path to get the reference to
482 *
483 * Given a path increment the reference count to the dentry and the vfsmount.
484 */
485void path_get(const struct path *path)
486{
487 mntget(path->mnt);
488 dget(path->dentry);
489}
490EXPORT_SYMBOL(path_get);
491
492/**
493 * path_put - put a reference to a path
494 * @path: path to put the reference to
495 *
496 * Given a path decrement the reference count to the dentry and the vfsmount.
497 */
498void path_put(const struct path *path)
499{
500 dput(path->dentry);
501 mntput(path->mnt);
502}
503EXPORT_SYMBOL(path_put);
504
505#define EMBEDDED_LEVELS 2
506struct nameidata {
507 struct path path;
508 struct qstr last;
509 struct path root;
510 struct inode *inode; /* path.dentry.d_inode */
511 unsigned int flags;
512 unsigned seq, m_seq;
513 int last_type;
514 unsigned depth;
515 int total_link_count;
516 struct saved {
517 struct path link;
518 struct delayed_call done;
519 const char *name;
520 unsigned seq;
521 } *stack, internal[EMBEDDED_LEVELS];
522 struct filename *name;
523 struct nameidata *saved;
524 struct inode *link_inode;
525 unsigned root_seq;
526 int dfd;
527};
528
529static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
530{
531 struct nameidata *old = current->nameidata;
532 p->stack = p->internal;
533 p->dfd = dfd;
534 p->name = name;
535 p->total_link_count = old ? old->total_link_count : 0;
536 p->saved = old;
537 current->nameidata = p;
538}
539
540static void restore_nameidata(void)
541{
542 struct nameidata *now = current->nameidata, *old = now->saved;
543
544 current->nameidata = old;
545 if (old)
546 old->total_link_count = now->total_link_count;
547 if (now->stack != now->internal)
548 kfree(now->stack);
549}
550
551static int __nd_alloc_stack(struct nameidata *nd)
552{
553 struct saved *p;
554
555 if (nd->flags & LOOKUP_RCU) {
556 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
557 GFP_ATOMIC);
558 if (unlikely(!p))
559 return -ECHILD;
560 } else {
561 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
562 GFP_KERNEL);
563 if (unlikely(!p))
564 return -ENOMEM;
565 }
566 memcpy(p, nd->internal, sizeof(nd->internal));
567 nd->stack = p;
568 return 0;
569}
570
571/**
572 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
573 * @path: nameidate to verify
574 *
575 * Rename can sometimes move a file or directory outside of a bind
576 * mount, path_connected allows those cases to be detected.
577 */
578static bool path_connected(const struct path *path)
579{
580 struct vfsmount *mnt = path->mnt;
581
582 /* Only bind mounts can have disconnected paths */
583 if (mnt->mnt_root == mnt->mnt_sb->s_root)
584 return true;
585
586 return is_subdir(path->dentry, mnt->mnt_root);
587}
588
589static inline int nd_alloc_stack(struct nameidata *nd)
590{
591 if (likely(nd->depth != EMBEDDED_LEVELS))
592 return 0;
593 if (likely(nd->stack != nd->internal))
594 return 0;
595 return __nd_alloc_stack(nd);
596}
597
598static void drop_links(struct nameidata *nd)
599{
600 int i = nd->depth;
601 while (i--) {
602 struct saved *last = nd->stack + i;
603 do_delayed_call(&last->done);
604 clear_delayed_call(&last->done);
605 }
606}
607
608static void terminate_walk(struct nameidata *nd)
609{
610 drop_links(nd);
611 if (!(nd->flags & LOOKUP_RCU)) {
612 int i;
613 path_put(&nd->path);
614 for (i = 0; i < nd->depth; i++)
615 path_put(&nd->stack[i].link);
616 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
617 path_put(&nd->root);
618 nd->root.mnt = NULL;
619 }
620 } else {
621 nd->flags &= ~LOOKUP_RCU;
622 if (!(nd->flags & LOOKUP_ROOT))
623 nd->root.mnt = NULL;
624 rcu_read_unlock();
625 }
626 nd->depth = 0;
627}
628
629/* path_put is needed afterwards regardless of success or failure */
630static bool legitimize_path(struct nameidata *nd,
631 struct path *path, unsigned seq)
632{
633 int res = __legitimize_mnt(path->mnt, nd->m_seq);
634 if (unlikely(res)) {
635 if (res > 0)
636 path->mnt = NULL;
637 path->dentry = NULL;
638 return false;
639 }
640 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
641 path->dentry = NULL;
642 return false;
643 }
644 return !read_seqcount_retry(&path->dentry->d_seq, seq);
645}
646
647static bool legitimize_links(struct nameidata *nd)
648{
649 int i;
650 for (i = 0; i < nd->depth; i++) {
651 struct saved *last = nd->stack + i;
652 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
653 drop_links(nd);
654 nd->depth = i + 1;
655 return false;
656 }
657 }
658 return true;
659}
660
661/*
662 * Path walking has 2 modes, rcu-walk and ref-walk (see
663 * Documentation/filesystems/path-lookup.txt). In situations when we can't
664 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
665 * normal reference counts on dentries and vfsmounts to transition to ref-walk
666 * mode. Refcounts are grabbed at the last known good point before rcu-walk
667 * got stuck, so ref-walk may continue from there. If this is not successful
668 * (eg. a seqcount has changed), then failure is returned and it's up to caller
669 * to restart the path walk from the beginning in ref-walk mode.
670 */
671
672/**
673 * unlazy_walk - try to switch to ref-walk mode.
674 * @nd: nameidata pathwalk data
675 * @dentry: child of nd->path.dentry or NULL
676 * @seq: seq number to check dentry against
677 * Returns: 0 on success, -ECHILD on failure
678 *
679 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
680 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
681 * @nd or NULL. Must be called from rcu-walk context.
682 * Nothing should touch nameidata between unlazy_walk() failure and
683 * terminate_walk().
684 */
685static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
686{
687 struct dentry *parent = nd->path.dentry;
688
689 BUG_ON(!(nd->flags & LOOKUP_RCU));
690
691 nd->flags &= ~LOOKUP_RCU;
692 if (unlikely(!legitimize_links(nd)))
693 goto out2;
694 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
695 goto out2;
696 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
697 goto out1;
698
699 /*
700 * For a negative lookup, the lookup sequence point is the parents
701 * sequence point, and it only needs to revalidate the parent dentry.
702 *
703 * For a positive lookup, we need to move both the parent and the
704 * dentry from the RCU domain to be properly refcounted. And the
705 * sequence number in the dentry validates *both* dentry counters,
706 * since we checked the sequence number of the parent after we got
707 * the child sequence number. So we know the parent must still
708 * be valid if the child sequence number is still valid.
709 */
710 if (!dentry) {
711 if (read_seqcount_retry(&parent->d_seq, nd->seq))
712 goto out;
713 BUG_ON(nd->inode != parent->d_inode);
714 } else {
715 if (!lockref_get_not_dead(&dentry->d_lockref))
716 goto out;
717 if (read_seqcount_retry(&dentry->d_seq, seq))
718 goto drop_dentry;
719 }
720
721 /*
722 * Sequence counts matched. Now make sure that the root is
723 * still valid and get it if required.
724 */
725 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
726 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
727 rcu_read_unlock();
728 dput(dentry);
729 return -ECHILD;
730 }
731 }
732
733 rcu_read_unlock();
734 return 0;
735
736drop_dentry:
737 rcu_read_unlock();
738 dput(dentry);
739 goto drop_root_mnt;
740out2:
741 nd->path.mnt = NULL;
742out1:
743 nd->path.dentry = NULL;
744out:
745 rcu_read_unlock();
746drop_root_mnt:
747 if (!(nd->flags & LOOKUP_ROOT))
748 nd->root.mnt = NULL;
749 return -ECHILD;
750}
751
752static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
753{
754 if (unlikely(!legitimize_path(nd, link, seq))) {
755 drop_links(nd);
756 nd->depth = 0;
757 nd->flags &= ~LOOKUP_RCU;
758 nd->path.mnt = NULL;
759 nd->path.dentry = NULL;
760 if (!(nd->flags & LOOKUP_ROOT))
761 nd->root.mnt = NULL;
762 rcu_read_unlock();
763 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
764 return 0;
765 }
766 path_put(link);
767 return -ECHILD;
768}
769
770static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
771{
772 return dentry->d_op->d_revalidate(dentry, flags);
773}
774
775/**
776 * complete_walk - successful completion of path walk
777 * @nd: pointer nameidata
778 *
779 * If we had been in RCU mode, drop out of it and legitimize nd->path.
780 * Revalidate the final result, unless we'd already done that during
781 * the path walk or the filesystem doesn't ask for it. Return 0 on
782 * success, -error on failure. In case of failure caller does not
783 * need to drop nd->path.
784 */
785static int complete_walk(struct nameidata *nd)
786{
787 struct dentry *dentry = nd->path.dentry;
788 int status;
789
790 if (nd->flags & LOOKUP_RCU) {
791 if (!(nd->flags & LOOKUP_ROOT))
792 nd->root.mnt = NULL;
793 if (unlikely(unlazy_walk(nd, NULL, 0)))
794 return -ECHILD;
795 }
796
797 if (likely(!(nd->flags & LOOKUP_JUMPED)))
798 return 0;
799
800 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
801 return 0;
802
803 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
804 if (status > 0)
805 return 0;
806
807 if (!status)
808 status = -ESTALE;
809
810 return status;
811}
812
813static void set_root(struct nameidata *nd)
814{
815 struct fs_struct *fs = current->fs;
816
817 if (nd->flags & LOOKUP_RCU) {
818 unsigned seq;
819
820 do {
821 seq = read_seqcount_begin(&fs->seq);
822 nd->root = fs->root;
823 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
824 } while (read_seqcount_retry(&fs->seq, seq));
825 } else {
826 get_fs_root(fs, &nd->root);
827 }
828}
829
830static void path_put_conditional(struct path *path, struct nameidata *nd)
831{
832 dput(path->dentry);
833 if (path->mnt != nd->path.mnt)
834 mntput(path->mnt);
835}
836
837static inline void path_to_nameidata(const struct path *path,
838 struct nameidata *nd)
839{
840 if (!(nd->flags & LOOKUP_RCU)) {
841 dput(nd->path.dentry);
842 if (nd->path.mnt != path->mnt)
843 mntput(nd->path.mnt);
844 }
845 nd->path.mnt = path->mnt;
846 nd->path.dentry = path->dentry;
847}
848
849static int nd_jump_root(struct nameidata *nd)
850{
851 if (nd->flags & LOOKUP_RCU) {
852 struct dentry *d;
853 nd->path = nd->root;
854 d = nd->path.dentry;
855 nd->inode = d->d_inode;
856 nd->seq = nd->root_seq;
857 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
858 return -ECHILD;
859 } else {
860 path_put(&nd->path);
861 nd->path = nd->root;
862 path_get(&nd->path);
863 nd->inode = nd->path.dentry->d_inode;
864 }
865 nd->flags |= LOOKUP_JUMPED;
866 return 0;
867}
868
869/*
870 * Helper to directly jump to a known parsed path from ->get_link,
871 * caller must have taken a reference to path beforehand.
872 */
873void nd_jump_link(struct path *path)
874{
875 struct nameidata *nd = current->nameidata;
876 path_put(&nd->path);
877
878 nd->path = *path;
879 nd->inode = nd->path.dentry->d_inode;
880 nd->flags |= LOOKUP_JUMPED;
881}
882
883static inline void put_link(struct nameidata *nd)
884{
885 struct saved *last = nd->stack + --nd->depth;
886 do_delayed_call(&last->done);
887 if (!(nd->flags & LOOKUP_RCU))
888 path_put(&last->link);
889}
890
891int sysctl_protected_symlinks __read_mostly = 0;
892int sysctl_protected_hardlinks __read_mostly = 0;
893
894/**
895 * may_follow_link - Check symlink following for unsafe situations
896 * @nd: nameidata pathwalk data
897 *
898 * In the case of the sysctl_protected_symlinks sysctl being enabled,
899 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
900 * in a sticky world-writable directory. This is to protect privileged
901 * processes from failing races against path names that may change out
902 * from under them by way of other users creating malicious symlinks.
903 * It will permit symlinks to be followed only when outside a sticky
904 * world-writable directory, or when the uid of the symlink and follower
905 * match, or when the directory owner matches the symlink's owner.
906 *
907 * Returns 0 if following the symlink is allowed, -ve on error.
908 */
909static inline int may_follow_link(struct nameidata *nd)
910{
911 const struct inode *inode;
912 const struct inode *parent;
913 kuid_t puid;
914
915 if (!sysctl_protected_symlinks)
916 return 0;
917
918 /* Allowed if owner and follower match. */
919 inode = nd->link_inode;
920 if (uid_eq(current_cred()->fsuid, inode->i_uid))
921 return 0;
922
923 /* Allowed if parent directory not sticky and world-writable. */
924 parent = nd->inode;
925 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
926 return 0;
927
928 /* Allowed if parent directory and link owner match. */
929 puid = parent->i_uid;
930 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
931 return 0;
932
933 if (nd->flags & LOOKUP_RCU)
934 return -ECHILD;
935
936 audit_log_link_denied("follow_link", &nd->stack[0].link);
937 return -EACCES;
938}
939
940/**
941 * safe_hardlink_source - Check for safe hardlink conditions
942 * @inode: the source inode to hardlink from
943 *
944 * Return false if at least one of the following conditions:
945 * - inode is not a regular file
946 * - inode is setuid
947 * - inode is setgid and group-exec
948 * - access failure for read and write
949 *
950 * Otherwise returns true.
951 */
952static bool safe_hardlink_source(struct inode *inode)
953{
954 umode_t mode = inode->i_mode;
955
956 /* Special files should not get pinned to the filesystem. */
957 if (!S_ISREG(mode))
958 return false;
959
960 /* Setuid files should not get pinned to the filesystem. */
961 if (mode & S_ISUID)
962 return false;
963
964 /* Executable setgid files should not get pinned to the filesystem. */
965 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
966 return false;
967
968 /* Hardlinking to unreadable or unwritable sources is dangerous. */
969 if (inode_permission(inode, MAY_READ | MAY_WRITE))
970 return false;
971
972 return true;
973}
974
975/**
976 * may_linkat - Check permissions for creating a hardlink
977 * @link: the source to hardlink from
978 *
979 * Block hardlink when all of:
980 * - sysctl_protected_hardlinks enabled
981 * - fsuid does not match inode
982 * - hardlink source is unsafe (see safe_hardlink_source() above)
983 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
984 *
985 * Returns 0 if successful, -ve on error.
986 */
987static int may_linkat(struct path *link)
988{
989 struct inode *inode;
990
991 if (!sysctl_protected_hardlinks)
992 return 0;
993
994 inode = link->dentry->d_inode;
995
996 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
997 * otherwise, it must be a safe source.
998 */
999 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
1000 return 0;
1001
1002 audit_log_link_denied("linkat", link);
1003 return -EPERM;
1004}
1005
1006static __always_inline
1007const char *get_link(struct nameidata *nd)
1008{
1009 struct saved *last = nd->stack + nd->depth - 1;
1010 struct dentry *dentry = last->link.dentry;
1011 struct inode *inode = nd->link_inode;
1012 int error;
1013 const char *res;
1014
1015 if (!(nd->flags & LOOKUP_RCU)) {
1016 touch_atime(&last->link);
1017 cond_resched();
1018 } else if (atime_needs_update_rcu(&last->link, inode)) {
1019 if (unlikely(unlazy_walk(nd, NULL, 0)))
1020 return ERR_PTR(-ECHILD);
1021 touch_atime(&last->link);
1022 }
1023
1024 error = security_inode_follow_link(dentry, inode,
1025 nd->flags & LOOKUP_RCU);
1026 if (unlikely(error))
1027 return ERR_PTR(error);
1028
1029 nd->last_type = LAST_BIND;
1030 res = inode->i_link;
1031 if (!res) {
1032 const char * (*get)(struct dentry *, struct inode *,
1033 struct delayed_call *);
1034 get = inode->i_op->get_link;
1035 if (nd->flags & LOOKUP_RCU) {
1036 res = get(NULL, inode, &last->done);
1037 if (res == ERR_PTR(-ECHILD)) {
1038 if (unlikely(unlazy_walk(nd, NULL, 0)))
1039 return ERR_PTR(-ECHILD);
1040 res = get(dentry, inode, &last->done);
1041 }
1042 } else {
1043 res = get(dentry, inode, &last->done);
1044 }
1045 if (IS_ERR_OR_NULL(res))
1046 return res;
1047 }
1048 if (*res == '/') {
1049 if (!nd->root.mnt)
1050 set_root(nd);
1051 if (unlikely(nd_jump_root(nd)))
1052 return ERR_PTR(-ECHILD);
1053 while (unlikely(*++res == '/'))
1054 ;
1055 }
1056 if (!*res)
1057 res = NULL;
1058 return res;
1059}
1060
1061/*
1062 * follow_up - Find the mountpoint of path's vfsmount
1063 *
1064 * Given a path, find the mountpoint of its source file system.
1065 * Replace @path with the path of the mountpoint in the parent mount.
1066 * Up is towards /.
1067 *
1068 * Return 1 if we went up a level and 0 if we were already at the
1069 * root.
1070 */
1071int follow_up(struct path *path)
1072{
1073 struct mount *mnt = real_mount(path->mnt);
1074 struct mount *parent;
1075 struct dentry *mountpoint;
1076
1077 read_seqlock_excl(&mount_lock);
1078 parent = mnt->mnt_parent;
1079 if (parent == mnt) {
1080 read_sequnlock_excl(&mount_lock);
1081 return 0;
1082 }
1083 mntget(&parent->mnt);
1084 mountpoint = dget(mnt->mnt_mountpoint);
1085 read_sequnlock_excl(&mount_lock);
1086 dput(path->dentry);
1087 path->dentry = mountpoint;
1088 mntput(path->mnt);
1089 path->mnt = &parent->mnt;
1090 return 1;
1091}
1092EXPORT_SYMBOL(follow_up);
1093
1094/*
1095 * Perform an automount
1096 * - return -EISDIR to tell follow_managed() to stop and return the path we
1097 * were called with.
1098 */
1099static int follow_automount(struct path *path, struct nameidata *nd,
1100 bool *need_mntput)
1101{
1102 struct vfsmount *mnt;
1103 int err;
1104
1105 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1106 return -EREMOTE;
1107
1108 /* We don't want to mount if someone's just doing a stat -
1109 * unless they're stat'ing a directory and appended a '/' to
1110 * the name.
1111 *
1112 * We do, however, want to mount if someone wants to open or
1113 * create a file of any type under the mountpoint, wants to
1114 * traverse through the mountpoint or wants to open the
1115 * mounted directory. Also, autofs may mark negative dentries
1116 * as being automount points. These will need the attentions
1117 * of the daemon to instantiate them before they can be used.
1118 */
1119 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1120 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1121 path->dentry->d_inode)
1122 return -EISDIR;
1123
1124 if (path->dentry->d_sb->s_user_ns != &init_user_ns)
1125 return -EACCES;
1126
1127 nd->total_link_count++;
1128 if (nd->total_link_count >= 40)
1129 return -ELOOP;
1130
1131 mnt = path->dentry->d_op->d_automount(path);
1132 if (IS_ERR(mnt)) {
1133 /*
1134 * The filesystem is allowed to return -EISDIR here to indicate
1135 * it doesn't want to automount. For instance, autofs would do
1136 * this so that its userspace daemon can mount on this dentry.
1137 *
1138 * However, we can only permit this if it's a terminal point in
1139 * the path being looked up; if it wasn't then the remainder of
1140 * the path is inaccessible and we should say so.
1141 */
1142 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1143 return -EREMOTE;
1144 return PTR_ERR(mnt);
1145 }
1146
1147 if (!mnt) /* mount collision */
1148 return 0;
1149
1150 if (!*need_mntput) {
1151 /* lock_mount() may release path->mnt on error */
1152 mntget(path->mnt);
1153 *need_mntput = true;
1154 }
1155 err = finish_automount(mnt, path);
1156
1157 switch (err) {
1158 case -EBUSY:
1159 /* Someone else made a mount here whilst we were busy */
1160 return 0;
1161 case 0:
1162 path_put(path);
1163 path->mnt = mnt;
1164 path->dentry = dget(mnt->mnt_root);
1165 return 0;
1166 default:
1167 return err;
1168 }
1169
1170}
1171
1172/*
1173 * Handle a dentry that is managed in some way.
1174 * - Flagged for transit management (autofs)
1175 * - Flagged as mountpoint
1176 * - Flagged as automount point
1177 *
1178 * This may only be called in refwalk mode.
1179 *
1180 * Serialization is taken care of in namespace.c
1181 */
1182static int follow_managed(struct path *path, struct nameidata *nd)
1183{
1184 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1185 unsigned managed;
1186 bool need_mntput = false;
1187 int ret = 0;
1188
1189 /* Given that we're not holding a lock here, we retain the value in a
1190 * local variable for each dentry as we look at it so that we don't see
1191 * the components of that value change under us */
1192 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1193 managed &= DCACHE_MANAGED_DENTRY,
1194 unlikely(managed != 0)) {
1195 /* Allow the filesystem to manage the transit without i_mutex
1196 * being held. */
1197 if (managed & DCACHE_MANAGE_TRANSIT) {
1198 BUG_ON(!path->dentry->d_op);
1199 BUG_ON(!path->dentry->d_op->d_manage);
1200 ret = path->dentry->d_op->d_manage(path, false);
1201 if (ret < 0)
1202 break;
1203 }
1204
1205 /* Transit to a mounted filesystem. */
1206 if (managed & DCACHE_MOUNTED) {
1207 struct vfsmount *mounted = lookup_mnt(path);
1208 if (mounted) {
1209 dput(path->dentry);
1210 if (need_mntput)
1211 mntput(path->mnt);
1212 path->mnt = mounted;
1213 path->dentry = dget(mounted->mnt_root);
1214 need_mntput = true;
1215 continue;
1216 }
1217
1218 /* Something is mounted on this dentry in another
1219 * namespace and/or whatever was mounted there in this
1220 * namespace got unmounted before lookup_mnt() could
1221 * get it */
1222 }
1223
1224 /* Handle an automount point */
1225 if (managed & DCACHE_NEED_AUTOMOUNT) {
1226 ret = follow_automount(path, nd, &need_mntput);
1227 if (ret < 0)
1228 break;
1229 continue;
1230 }
1231
1232 /* We didn't change the current path point */
1233 break;
1234 }
1235
1236 if (need_mntput && path->mnt == mnt)
1237 mntput(path->mnt);
1238 if (ret == -EISDIR || !ret)
1239 ret = 1;
1240 if (need_mntput)
1241 nd->flags |= LOOKUP_JUMPED;
1242 if (unlikely(ret < 0))
1243 path_put_conditional(path, nd);
1244 return ret;
1245}
1246
1247int follow_down_one(struct path *path)
1248{
1249 struct vfsmount *mounted;
1250
1251 mounted = lookup_mnt(path);
1252 if (mounted) {
1253 dput(path->dentry);
1254 mntput(path->mnt);
1255 path->mnt = mounted;
1256 path->dentry = dget(mounted->mnt_root);
1257 return 1;
1258 }
1259 return 0;
1260}
1261EXPORT_SYMBOL(follow_down_one);
1262
1263static inline int managed_dentry_rcu(const struct path *path)
1264{
1265 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1266 path->dentry->d_op->d_manage(path, true) : 0;
1267}
1268
1269/*
1270 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1271 * we meet a managed dentry that would need blocking.
1272 */
1273static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1274 struct inode **inode, unsigned *seqp)
1275{
1276 for (;;) {
1277 struct mount *mounted;
1278 /*
1279 * Don't forget we might have a non-mountpoint managed dentry
1280 * that wants to block transit.
1281 */
1282 switch (managed_dentry_rcu(path)) {
1283 case -ECHILD:
1284 default:
1285 return false;
1286 case -EISDIR:
1287 return true;
1288 case 0:
1289 break;
1290 }
1291
1292 if (!d_mountpoint(path->dentry))
1293 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1294
1295 mounted = __lookup_mnt(path->mnt, path->dentry);
1296 if (!mounted)
1297 break;
1298 path->mnt = &mounted->mnt;
1299 path->dentry = mounted->mnt.mnt_root;
1300 nd->flags |= LOOKUP_JUMPED;
1301 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1302 /*
1303 * Update the inode too. We don't need to re-check the
1304 * dentry sequence number here after this d_inode read,
1305 * because a mount-point is always pinned.
1306 */
1307 *inode = path->dentry->d_inode;
1308 }
1309 return !read_seqretry(&mount_lock, nd->m_seq) &&
1310 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1311}
1312
1313static int follow_dotdot_rcu(struct nameidata *nd)
1314{
1315 struct inode *inode = nd->inode;
1316
1317 while (1) {
1318 if (path_equal(&nd->path, &nd->root))
1319 break;
1320 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1321 struct dentry *old = nd->path.dentry;
1322 struct dentry *parent = old->d_parent;
1323 unsigned seq;
1324
1325 inode = parent->d_inode;
1326 seq = read_seqcount_begin(&parent->d_seq);
1327 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1328 return -ECHILD;
1329 nd->path.dentry = parent;
1330 nd->seq = seq;
1331 if (unlikely(!path_connected(&nd->path)))
1332 return -ENOENT;
1333 break;
1334 } else {
1335 struct mount *mnt = real_mount(nd->path.mnt);
1336 struct mount *mparent = mnt->mnt_parent;
1337 struct dentry *mountpoint = mnt->mnt_mountpoint;
1338 struct inode *inode2 = mountpoint->d_inode;
1339 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1340 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1341 return -ECHILD;
1342 if (&mparent->mnt == nd->path.mnt)
1343 break;
1344 /* we know that mountpoint was pinned */
1345 nd->path.dentry = mountpoint;
1346 nd->path.mnt = &mparent->mnt;
1347 inode = inode2;
1348 nd->seq = seq;
1349 }
1350 }
1351 while (unlikely(d_mountpoint(nd->path.dentry))) {
1352 struct mount *mounted;
1353 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1354 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1355 return -ECHILD;
1356 if (!mounted)
1357 break;
1358 nd->path.mnt = &mounted->mnt;
1359 nd->path.dentry = mounted->mnt.mnt_root;
1360 inode = nd->path.dentry->d_inode;
1361 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1362 }
1363 nd->inode = inode;
1364 return 0;
1365}
1366
1367/*
1368 * Follow down to the covering mount currently visible to userspace. At each
1369 * point, the filesystem owning that dentry may be queried as to whether the
1370 * caller is permitted to proceed or not.
1371 */
1372int follow_down(struct path *path)
1373{
1374 unsigned managed;
1375 int ret;
1376
1377 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1378 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1379 /* Allow the filesystem to manage the transit without i_mutex
1380 * being held.
1381 *
1382 * We indicate to the filesystem if someone is trying to mount
1383 * something here. This gives autofs the chance to deny anyone
1384 * other than its daemon the right to mount on its
1385 * superstructure.
1386 *
1387 * The filesystem may sleep at this point.
1388 */
1389 if (managed & DCACHE_MANAGE_TRANSIT) {
1390 BUG_ON(!path->dentry->d_op);
1391 BUG_ON(!path->dentry->d_op->d_manage);
1392 ret = path->dentry->d_op->d_manage(path, false);
1393 if (ret < 0)
1394 return ret == -EISDIR ? 0 : ret;
1395 }
1396
1397 /* Transit to a mounted filesystem. */
1398 if (managed & DCACHE_MOUNTED) {
1399 struct vfsmount *mounted = lookup_mnt(path);
1400 if (!mounted)
1401 break;
1402 dput(path->dentry);
1403 mntput(path->mnt);
1404 path->mnt = mounted;
1405 path->dentry = dget(mounted->mnt_root);
1406 continue;
1407 }
1408
1409 /* Don't handle automount points here */
1410 break;
1411 }
1412 return 0;
1413}
1414EXPORT_SYMBOL(follow_down);
1415
1416/*
1417 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1418 */
1419static void follow_mount(struct path *path)
1420{
1421 while (d_mountpoint(path->dentry)) {
1422 struct vfsmount *mounted = lookup_mnt(path);
1423 if (!mounted)
1424 break;
1425 dput(path->dentry);
1426 mntput(path->mnt);
1427 path->mnt = mounted;
1428 path->dentry = dget(mounted->mnt_root);
1429 }
1430}
1431
1432static int path_parent_directory(struct path *path)
1433{
1434 struct dentry *old = path->dentry;
1435 /* rare case of legitimate dget_parent()... */
1436 path->dentry = dget_parent(path->dentry);
1437 dput(old);
1438 if (unlikely(!path_connected(path)))
1439 return -ENOENT;
1440 return 0;
1441}
1442
1443static int follow_dotdot(struct nameidata *nd)
1444{
1445 while(1) {
1446 if (nd->path.dentry == nd->root.dentry &&
1447 nd->path.mnt == nd->root.mnt) {
1448 break;
1449 }
1450 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1451 int ret = path_parent_directory(&nd->path);
1452 if (ret)
1453 return ret;
1454 break;
1455 }
1456 if (!follow_up(&nd->path))
1457 break;
1458 }
1459 follow_mount(&nd->path);
1460 nd->inode = nd->path.dentry->d_inode;
1461 return 0;
1462}
1463
1464/*
1465 * This looks up the name in dcache and possibly revalidates the found dentry.
1466 * NULL is returned if the dentry does not exist in the cache.
1467 */
1468static struct dentry *lookup_dcache(const struct qstr *name,
1469 struct dentry *dir,
1470 unsigned int flags)
1471{
1472 struct dentry *dentry;
1473 int error;
1474
1475 dentry = d_lookup(dir, name);
1476 if (dentry) {
1477 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1478 error = d_revalidate(dentry, flags);
1479 if (unlikely(error <= 0)) {
1480 if (!error)
1481 d_invalidate(dentry);
1482 dput(dentry);
1483 return ERR_PTR(error);
1484 }
1485 }
1486 }
1487 return dentry;
1488}
1489
1490/*
1491 * Call i_op->lookup on the dentry. The dentry must be negative and
1492 * unhashed.
1493 *
1494 * dir->d_inode->i_mutex must be held
1495 */
1496static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1497 unsigned int flags)
1498{
1499 struct dentry *old;
1500
1501 /* Don't create child dentry for a dead directory. */
1502 if (unlikely(IS_DEADDIR(dir))) {
1503 dput(dentry);
1504 return ERR_PTR(-ENOENT);
1505 }
1506
1507 old = dir->i_op->lookup(dir, dentry, flags);
1508 if (unlikely(old)) {
1509 dput(dentry);
1510 dentry = old;
1511 }
1512 return dentry;
1513}
1514
1515static struct dentry *__lookup_hash(const struct qstr *name,
1516 struct dentry *base, unsigned int flags)
1517{
1518 struct dentry *dentry = lookup_dcache(name, base, flags);
1519
1520 if (dentry)
1521 return dentry;
1522
1523 dentry = d_alloc(base, name);
1524 if (unlikely(!dentry))
1525 return ERR_PTR(-ENOMEM);
1526
1527 return lookup_real(base->d_inode, dentry, flags);
1528}
1529
1530static int lookup_fast(struct nameidata *nd,
1531 struct path *path, struct inode **inode,
1532 unsigned *seqp)
1533{
1534 struct vfsmount *mnt = nd->path.mnt;
1535 struct dentry *dentry, *parent = nd->path.dentry;
1536 int status = 1;
1537 int err;
1538
1539 /*
1540 * Rename seqlock is not required here because in the off chance
1541 * of a false negative due to a concurrent rename, the caller is
1542 * going to fall back to non-racy lookup.
1543 */
1544 if (nd->flags & LOOKUP_RCU) {
1545 unsigned seq;
1546 bool negative;
1547 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1548 if (unlikely(!dentry)) {
1549 if (unlazy_walk(nd, NULL, 0))
1550 return -ECHILD;
1551 return 0;
1552 }
1553
1554 /*
1555 * This sequence count validates that the inode matches
1556 * the dentry name information from lookup.
1557 */
1558 *inode = d_backing_inode(dentry);
1559 negative = d_is_negative(dentry);
1560 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1561 return -ECHILD;
1562
1563 /*
1564 * This sequence count validates that the parent had no
1565 * changes while we did the lookup of the dentry above.
1566 *
1567 * The memory barrier in read_seqcount_begin of child is
1568 * enough, we can use __read_seqcount_retry here.
1569 */
1570 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1571 return -ECHILD;
1572
1573 *seqp = seq;
1574 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1575 status = d_revalidate(dentry, nd->flags);
1576 if (unlikely(status <= 0)) {
1577 if (unlazy_walk(nd, dentry, seq))
1578 return -ECHILD;
1579 if (status == -ECHILD)
1580 status = d_revalidate(dentry, nd->flags);
1581 } else {
1582 /*
1583 * Note: do negative dentry check after revalidation in
1584 * case that drops it.
1585 */
1586 if (unlikely(negative))
1587 return -ENOENT;
1588 path->mnt = mnt;
1589 path->dentry = dentry;
1590 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1591 return 1;
1592 if (unlazy_walk(nd, dentry, seq))
1593 return -ECHILD;
1594 }
1595 } else {
1596 dentry = __d_lookup(parent, &nd->last);
1597 if (unlikely(!dentry))
1598 return 0;
1599 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1600 status = d_revalidate(dentry, nd->flags);
1601 }
1602 if (unlikely(status <= 0)) {
1603 if (!status)
1604 d_invalidate(dentry);
1605 dput(dentry);
1606 return status;
1607 }
1608 if (unlikely(d_is_negative(dentry))) {
1609 dput(dentry);
1610 return -ENOENT;
1611 }
1612
1613 path->mnt = mnt;
1614 path->dentry = dentry;
1615 err = follow_managed(path, nd);
1616 if (likely(err > 0))
1617 *inode = d_backing_inode(path->dentry);
1618 return err;
1619}
1620
1621/* Fast lookup failed, do it the slow way */
1622static struct dentry *lookup_slow(const struct qstr *name,
1623 struct dentry *dir,
1624 unsigned int flags)
1625{
1626 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1627 struct inode *inode = dir->d_inode;
1628 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1629
1630 inode_lock_shared(inode);
1631 /* Don't go there if it's already dead */
1632 if (unlikely(IS_DEADDIR(inode)))
1633 goto out;
1634again:
1635 dentry = d_alloc_parallel(dir, name, &wq);
1636 if (IS_ERR(dentry))
1637 goto out;
1638 if (unlikely(!d_in_lookup(dentry))) {
1639 if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1640 !(flags & LOOKUP_NO_REVAL)) {
1641 int error = d_revalidate(dentry, flags);
1642 if (unlikely(error <= 0)) {
1643 if (!error) {
1644 d_invalidate(dentry);
1645 dput(dentry);
1646 goto again;
1647 }
1648 dput(dentry);
1649 dentry = ERR_PTR(error);
1650 }
1651 }
1652 } else {
1653 old = inode->i_op->lookup(inode, dentry, flags);
1654 d_lookup_done(dentry);
1655 if (unlikely(old)) {
1656 dput(dentry);
1657 dentry = old;
1658 }
1659 }
1660out:
1661 inode_unlock_shared(inode);
1662 return dentry;
1663}
1664
1665static inline int may_lookup(struct nameidata *nd)
1666{
1667 if (nd->flags & LOOKUP_RCU) {
1668 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1669 if (err != -ECHILD)
1670 return err;
1671 if (unlazy_walk(nd, NULL, 0))
1672 return -ECHILD;
1673 }
1674 return inode_permission(nd->inode, MAY_EXEC);
1675}
1676
1677static inline int handle_dots(struct nameidata *nd, int type)
1678{
1679 if (type == LAST_DOTDOT) {
1680 if (!nd->root.mnt)
1681 set_root(nd);
1682 if (nd->flags & LOOKUP_RCU) {
1683 return follow_dotdot_rcu(nd);
1684 } else
1685 return follow_dotdot(nd);
1686 }
1687 return 0;
1688}
1689
1690static int pick_link(struct nameidata *nd, struct path *link,
1691 struct inode *inode, unsigned seq)
1692{
1693 int error;
1694 struct saved *last;
1695 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1696 path_to_nameidata(link, nd);
1697 return -ELOOP;
1698 }
1699 if (!(nd->flags & LOOKUP_RCU)) {
1700 if (link->mnt == nd->path.mnt)
1701 mntget(link->mnt);
1702 }
1703 error = nd_alloc_stack(nd);
1704 if (unlikely(error)) {
1705 if (error == -ECHILD) {
1706 if (unlikely(unlazy_link(nd, link, seq)))
1707 return -ECHILD;
1708 error = nd_alloc_stack(nd);
1709 }
1710 if (error) {
1711 path_put(link);
1712 return error;
1713 }
1714 }
1715
1716 last = nd->stack + nd->depth++;
1717 last->link = *link;
1718 clear_delayed_call(&last->done);
1719 nd->link_inode = inode;
1720 last->seq = seq;
1721 return 1;
1722}
1723
1724enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1725
1726/*
1727 * Do we need to follow links? We _really_ want to be able
1728 * to do this check without having to look at inode->i_op,
1729 * so we keep a cache of "no, this doesn't need follow_link"
1730 * for the common case.
1731 */
1732static inline int step_into(struct nameidata *nd, struct path *path,
1733 int flags, struct inode *inode, unsigned seq)
1734{
1735 if (!(flags & WALK_MORE) && nd->depth)
1736 put_link(nd);
1737 if (likely(!d_is_symlink(path->dentry)) ||
1738 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1739 /* not a symlink or should not follow */
1740 path_to_nameidata(path, nd);
1741 nd->inode = inode;
1742 nd->seq = seq;
1743 return 0;
1744 }
1745 /* make sure that d_is_symlink above matches inode */
1746 if (nd->flags & LOOKUP_RCU) {
1747 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1748 return -ECHILD;
1749 }
1750 return pick_link(nd, path, inode, seq);
1751}
1752
1753static int walk_component(struct nameidata *nd, int flags)
1754{
1755 struct path path;
1756 struct inode *inode;
1757 unsigned seq;
1758 int err;
1759 /*
1760 * "." and ".." are special - ".." especially so because it has
1761 * to be able to know about the current root directory and
1762 * parent relationships.
1763 */
1764 if (unlikely(nd->last_type != LAST_NORM)) {
1765 err = handle_dots(nd, nd->last_type);
1766 if (!(flags & WALK_MORE) && nd->depth)
1767 put_link(nd);
1768 return err;
1769 }
1770 err = lookup_fast(nd, &path, &inode, &seq);
1771 if (unlikely(err <= 0)) {
1772 if (err < 0)
1773 return err;
1774 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1775 nd->flags);
1776 if (IS_ERR(path.dentry))
1777 return PTR_ERR(path.dentry);
1778
1779 path.mnt = nd->path.mnt;
1780 err = follow_managed(&path, nd);
1781 if (unlikely(err < 0))
1782 return err;
1783
1784 if (unlikely(d_is_negative(path.dentry))) {
1785 path_to_nameidata(&path, nd);
1786 return -ENOENT;
1787 }
1788
1789 seq = 0; /* we are already out of RCU mode */
1790 inode = d_backing_inode(path.dentry);
1791 }
1792
1793 return step_into(nd, &path, flags, inode, seq);
1794}
1795
1796/*
1797 * We can do the critical dentry name comparison and hashing
1798 * operations one word at a time, but we are limited to:
1799 *
1800 * - Architectures with fast unaligned word accesses. We could
1801 * do a "get_unaligned()" if this helps and is sufficiently
1802 * fast.
1803 *
1804 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1805 * do not trap on the (extremely unlikely) case of a page
1806 * crossing operation.
1807 *
1808 * - Furthermore, we need an efficient 64-bit compile for the
1809 * 64-bit case in order to generate the "number of bytes in
1810 * the final mask". Again, that could be replaced with a
1811 * efficient population count instruction or similar.
1812 */
1813#ifdef CONFIG_DCACHE_WORD_ACCESS
1814
1815#include <asm/word-at-a-time.h>
1816
1817#ifdef HASH_MIX
1818
1819/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1820
1821#elif defined(CONFIG_64BIT)
1822/*
1823 * Register pressure in the mixing function is an issue, particularly
1824 * on 32-bit x86, but almost any function requires one state value and
1825 * one temporary. Instead, use a function designed for two state values
1826 * and no temporaries.
1827 *
1828 * This function cannot create a collision in only two iterations, so
1829 * we have two iterations to achieve avalanche. In those two iterations,
1830 * we have six layers of mixing, which is enough to spread one bit's
1831 * influence out to 2^6 = 64 state bits.
1832 *
1833 * Rotate constants are scored by considering either 64 one-bit input
1834 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1835 * probability of that delta causing a change to each of the 128 output
1836 * bits, using a sample of random initial states.
1837 *
1838 * The Shannon entropy of the computed probabilities is then summed
1839 * to produce a score. Ideally, any input change has a 50% chance of
1840 * toggling any given output bit.
1841 *
1842 * Mixing scores (in bits) for (12,45):
1843 * Input delta: 1-bit 2-bit
1844 * 1 round: 713.3 42542.6
1845 * 2 rounds: 2753.7 140389.8
1846 * 3 rounds: 5954.1 233458.2
1847 * 4 rounds: 7862.6 256672.2
1848 * Perfect: 8192 258048
1849 * (64*128) (64*63/2 * 128)
1850 */
1851#define HASH_MIX(x, y, a) \
1852 ( x ^= (a), \
1853 y ^= x, x = rol64(x,12),\
1854 x += y, y = rol64(y,45),\
1855 y *= 9 )
1856
1857/*
1858 * Fold two longs into one 32-bit hash value. This must be fast, but
1859 * latency isn't quite as critical, as there is a fair bit of additional
1860 * work done before the hash value is used.
1861 */
1862static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1863{
1864 y ^= x * GOLDEN_RATIO_64;
1865 y *= GOLDEN_RATIO_64;
1866 return y >> 32;
1867}
1868
1869#else /* 32-bit case */
1870
1871/*
1872 * Mixing scores (in bits) for (7,20):
1873 * Input delta: 1-bit 2-bit
1874 * 1 round: 330.3 9201.6
1875 * 2 rounds: 1246.4 25475.4
1876 * 3 rounds: 1907.1 31295.1
1877 * 4 rounds: 2042.3 31718.6
1878 * Perfect: 2048 31744
1879 * (32*64) (32*31/2 * 64)
1880 */
1881#define HASH_MIX(x, y, a) \
1882 ( x ^= (a), \
1883 y ^= x, x = rol32(x, 7),\
1884 x += y, y = rol32(y,20),\
1885 y *= 9 )
1886
1887static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1888{
1889 /* Use arch-optimized multiply if one exists */
1890 return __hash_32(y ^ __hash_32(x));
1891}
1892
1893#endif
1894
1895/*
1896 * Return the hash of a string of known length. This is carfully
1897 * designed to match hash_name(), which is the more critical function.
1898 * In particular, we must end by hashing a final word containing 0..7
1899 * payload bytes, to match the way that hash_name() iterates until it
1900 * finds the delimiter after the name.
1901 */
1902unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1903{
1904 unsigned long a, x = 0, y = (unsigned long)salt;
1905
1906 for (;;) {
1907 if (!len)
1908 goto done;
1909 a = load_unaligned_zeropad(name);
1910 if (len < sizeof(unsigned long))
1911 break;
1912 HASH_MIX(x, y, a);
1913 name += sizeof(unsigned long);
1914 len -= sizeof(unsigned long);
1915 }
1916 x ^= a & bytemask_from_count(len);
1917done:
1918 return fold_hash(x, y);
1919}
1920EXPORT_SYMBOL(full_name_hash);
1921
1922/* Return the "hash_len" (hash and length) of a null-terminated string */
1923u64 hashlen_string(const void *salt, const char *name)
1924{
1925 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1926 unsigned long adata, mask, len;
1927 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1928
1929 len = 0;
1930 goto inside;
1931
1932 do {
1933 HASH_MIX(x, y, a);
1934 len += sizeof(unsigned long);
1935inside:
1936 a = load_unaligned_zeropad(name+len);
1937 } while (!has_zero(a, &adata, &constants));
1938
1939 adata = prep_zero_mask(a, adata, &constants);
1940 mask = create_zero_mask(adata);
1941 x ^= a & zero_bytemask(mask);
1942
1943 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1944}
1945EXPORT_SYMBOL(hashlen_string);
1946
1947/*
1948 * Calculate the length and hash of the path component, and
1949 * return the "hash_len" as the result.
1950 */
1951static inline u64 hash_name(const void *salt, const char *name)
1952{
1953 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1954 unsigned long adata, bdata, mask, len;
1955 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1956
1957 len = 0;
1958 goto inside;
1959
1960 do {
1961 HASH_MIX(x, y, a);
1962 len += sizeof(unsigned long);
1963inside:
1964 a = load_unaligned_zeropad(name+len);
1965 b = a ^ REPEAT_BYTE('/');
1966 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1967
1968 adata = prep_zero_mask(a, adata, &constants);
1969 bdata = prep_zero_mask(b, bdata, &constants);
1970 mask = create_zero_mask(adata | bdata);
1971 x ^= a & zero_bytemask(mask);
1972
1973 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1974}
1975
1976#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1977
1978/* Return the hash of a string of known length */
1979unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1980{
1981 unsigned long hash = init_name_hash(salt);
1982 while (len--)
1983 hash = partial_name_hash((unsigned char)*name++, hash);
1984 return end_name_hash(hash);
1985}
1986EXPORT_SYMBOL(full_name_hash);
1987
1988/* Return the "hash_len" (hash and length) of a null-terminated string */
1989u64 hashlen_string(const void *salt, const char *name)
1990{
1991 unsigned long hash = init_name_hash(salt);
1992 unsigned long len = 0, c;
1993
1994 c = (unsigned char)*name;
1995 while (c) {
1996 len++;
1997 hash = partial_name_hash(c, hash);
1998 c = (unsigned char)name[len];
1999 }
2000 return hashlen_create(end_name_hash(hash), len);
2001}
2002EXPORT_SYMBOL(hashlen_string);
2003
2004/*
2005 * We know there's a real path component here of at least
2006 * one character.
2007 */
2008static inline u64 hash_name(const void *salt, const char *name)
2009{
2010 unsigned long hash = init_name_hash(salt);
2011 unsigned long len = 0, c;
2012
2013 c = (unsigned char)*name;
2014 do {
2015 len++;
2016 hash = partial_name_hash(c, hash);
2017 c = (unsigned char)name[len];
2018 } while (c && c != '/');
2019 return hashlen_create(end_name_hash(hash), len);
2020}
2021
2022#endif
2023
2024/*
2025 * Name resolution.
2026 * This is the basic name resolution function, turning a pathname into
2027 * the final dentry. We expect 'base' to be positive and a directory.
2028 *
2029 * Returns 0 and nd will have valid dentry and mnt on success.
2030 * Returns error and drops reference to input namei data on failure.
2031 */
2032static int link_path_walk(const char *name, struct nameidata *nd)
2033{
2034 int err;
2035
2036 while (*name=='/')
2037 name++;
2038 if (!*name)
2039 return 0;
2040
2041 /* At this point we know we have a real path component. */
2042 for(;;) {
2043 u64 hash_len;
2044 int type;
2045
2046 err = may_lookup(nd);
2047 if (err)
2048 return err;
2049
2050 hash_len = hash_name(nd->path.dentry, name);
2051
2052 type = LAST_NORM;
2053 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2054 case 2:
2055 if (name[1] == '.') {
2056 type = LAST_DOTDOT;
2057 nd->flags |= LOOKUP_JUMPED;
2058 }
2059 break;
2060 case 1:
2061 type = LAST_DOT;
2062 }
2063 if (likely(type == LAST_NORM)) {
2064 struct dentry *parent = nd->path.dentry;
2065 nd->flags &= ~LOOKUP_JUMPED;
2066 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2067 struct qstr this = { { .hash_len = hash_len }, .name = name };
2068 err = parent->d_op->d_hash(parent, &this);
2069 if (err < 0)
2070 return err;
2071 hash_len = this.hash_len;
2072 name = this.name;
2073 }
2074 }
2075
2076 nd->last.hash_len = hash_len;
2077 nd->last.name = name;
2078 nd->last_type = type;
2079
2080 name += hashlen_len(hash_len);
2081 if (!*name)
2082 goto OK;
2083 /*
2084 * If it wasn't NUL, we know it was '/'. Skip that
2085 * slash, and continue until no more slashes.
2086 */
2087 do {
2088 name++;
2089 } while (unlikely(*name == '/'));
2090 if (unlikely(!*name)) {
2091OK:
2092 /* pathname body, done */
2093 if (!nd->depth)
2094 return 0;
2095 name = nd->stack[nd->depth - 1].name;
2096 /* trailing symlink, done */
2097 if (!name)
2098 return 0;
2099 /* last component of nested symlink */
2100 err = walk_component(nd, WALK_FOLLOW);
2101 } else {
2102 /* not the last component */
2103 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2104 }
2105 if (err < 0)
2106 return err;
2107
2108 if (err) {
2109 const char *s = get_link(nd);
2110
2111 if (IS_ERR(s))
2112 return PTR_ERR(s);
2113 err = 0;
2114 if (unlikely(!s)) {
2115 /* jumped */
2116 put_link(nd);
2117 } else {
2118 nd->stack[nd->depth - 1].name = name;
2119 name = s;
2120 continue;
2121 }
2122 }
2123 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2124 if (nd->flags & LOOKUP_RCU) {
2125 if (unlazy_walk(nd, NULL, 0))
2126 return -ECHILD;
2127 }
2128 return -ENOTDIR;
2129 }
2130 }
2131}
2132
2133static const char *path_init(struct nameidata *nd, unsigned flags)
2134{
2135 int retval = 0;
2136 const char *s = nd->name->name;
2137
2138 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2139 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2140 nd->depth = 0;
2141 if (flags & LOOKUP_ROOT) {
2142 struct dentry *root = nd->root.dentry;
2143 struct inode *inode = root->d_inode;
2144 if (*s) {
2145 if (!d_can_lookup(root))
2146 return ERR_PTR(-ENOTDIR);
2147 retval = inode_permission(inode, MAY_EXEC);
2148 if (retval)
2149 return ERR_PTR(retval);
2150 }
2151 nd->path = nd->root;
2152 nd->inode = inode;
2153 if (flags & LOOKUP_RCU) {
2154 rcu_read_lock();
2155 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2156 nd->root_seq = nd->seq;
2157 nd->m_seq = read_seqbegin(&mount_lock);
2158 } else {
2159 path_get(&nd->path);
2160 }
2161 return s;
2162 }
2163
2164 nd->root.mnt = NULL;
2165 nd->path.mnt = NULL;
2166 nd->path.dentry = NULL;
2167
2168 nd->m_seq = read_seqbegin(&mount_lock);
2169 if (*s == '/') {
2170 if (flags & LOOKUP_RCU)
2171 rcu_read_lock();
2172 set_root(nd);
2173 if (likely(!nd_jump_root(nd)))
2174 return s;
2175 nd->root.mnt = NULL;
2176 rcu_read_unlock();
2177 return ERR_PTR(-ECHILD);
2178 } else if (nd->dfd == AT_FDCWD) {
2179 if (flags & LOOKUP_RCU) {
2180 struct fs_struct *fs = current->fs;
2181 unsigned seq;
2182
2183 rcu_read_lock();
2184
2185 do {
2186 seq = read_seqcount_begin(&fs->seq);
2187 nd->path = fs->pwd;
2188 nd->inode = nd->path.dentry->d_inode;
2189 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2190 } while (read_seqcount_retry(&fs->seq, seq));
2191 } else {
2192 get_fs_pwd(current->fs, &nd->path);
2193 nd->inode = nd->path.dentry->d_inode;
2194 }
2195 return s;
2196 } else {
2197 /* Caller must check execute permissions on the starting path component */
2198 struct fd f = fdget_raw(nd->dfd);
2199 struct dentry *dentry;
2200
2201 if (!f.file)
2202 return ERR_PTR(-EBADF);
2203
2204 dentry = f.file->f_path.dentry;
2205
2206 if (*s) {
2207 if (!d_can_lookup(dentry)) {
2208 fdput(f);
2209 return ERR_PTR(-ENOTDIR);
2210 }
2211 }
2212
2213 nd->path = f.file->f_path;
2214 if (flags & LOOKUP_RCU) {
2215 rcu_read_lock();
2216 nd->inode = nd->path.dentry->d_inode;
2217 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2218 } else {
2219 path_get(&nd->path);
2220 nd->inode = nd->path.dentry->d_inode;
2221 }
2222 fdput(f);
2223 return s;
2224 }
2225}
2226
2227static const char *trailing_symlink(struct nameidata *nd)
2228{
2229 const char *s;
2230 int error = may_follow_link(nd);
2231 if (unlikely(error))
2232 return ERR_PTR(error);
2233 nd->flags |= LOOKUP_PARENT;
2234 nd->stack[0].name = NULL;
2235 s = get_link(nd);
2236 return s ? s : "";
2237}
2238
2239static inline int lookup_last(struct nameidata *nd)
2240{
2241 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2242 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2243
2244 nd->flags &= ~LOOKUP_PARENT;
2245 return walk_component(nd, 0);
2246}
2247
2248/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2249static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2250{
2251 const char *s = path_init(nd, flags);
2252 int err;
2253
2254 if (IS_ERR(s))
2255 return PTR_ERR(s);
2256 while (!(err = link_path_walk(s, nd))
2257 && ((err = lookup_last(nd)) > 0)) {
2258 s = trailing_symlink(nd);
2259 if (IS_ERR(s)) {
2260 err = PTR_ERR(s);
2261 break;
2262 }
2263 }
2264 if (!err)
2265 err = complete_walk(nd);
2266
2267 if (!err && nd->flags & LOOKUP_DIRECTORY)
2268 if (!d_can_lookup(nd->path.dentry))
2269 err = -ENOTDIR;
2270 if (!err) {
2271 *path = nd->path;
2272 nd->path.mnt = NULL;
2273 nd->path.dentry = NULL;
2274 }
2275 terminate_walk(nd);
2276 return err;
2277}
2278
2279static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2280 struct path *path, struct path *root)
2281{
2282 int retval;
2283 struct nameidata nd;
2284 if (IS_ERR(name))
2285 return PTR_ERR(name);
2286 if (unlikely(root)) {
2287 nd.root = *root;
2288 flags |= LOOKUP_ROOT;
2289 }
2290 set_nameidata(&nd, dfd, name);
2291 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2292 if (unlikely(retval == -ECHILD))
2293 retval = path_lookupat(&nd, flags, path);
2294 if (unlikely(retval == -ESTALE))
2295 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2296
2297 if (likely(!retval))
2298 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2299 restore_nameidata();
2300 putname(name);
2301 return retval;
2302}
2303
2304/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2305static int path_parentat(struct nameidata *nd, unsigned flags,
2306 struct path *parent)
2307{
2308 const char *s = path_init(nd, flags);
2309 int err;
2310 if (IS_ERR(s))
2311 return PTR_ERR(s);
2312 err = link_path_walk(s, nd);
2313 if (!err)
2314 err = complete_walk(nd);
2315 if (!err) {
2316 *parent = nd->path;
2317 nd->path.mnt = NULL;
2318 nd->path.dentry = NULL;
2319 }
2320 terminate_walk(nd);
2321 return err;
2322}
2323
2324static struct filename *filename_parentat(int dfd, struct filename *name,
2325 unsigned int flags, struct path *parent,
2326 struct qstr *last, int *type)
2327{
2328 int retval;
2329 struct nameidata nd;
2330
2331 if (IS_ERR(name))
2332 return name;
2333 set_nameidata(&nd, dfd, name);
2334 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2335 if (unlikely(retval == -ECHILD))
2336 retval = path_parentat(&nd, flags, parent);
2337 if (unlikely(retval == -ESTALE))
2338 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2339 if (likely(!retval)) {
2340 *last = nd.last;
2341 *type = nd.last_type;
2342 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2343 } else {
2344 putname(name);
2345 name = ERR_PTR(retval);
2346 }
2347 restore_nameidata();
2348 return name;
2349}
2350
2351/* does lookup, returns the object with parent locked */
2352struct dentry *kern_path_locked(const char *name, struct path *path)
2353{
2354 struct filename *filename;
2355 struct dentry *d;
2356 struct qstr last;
2357 int type;
2358
2359 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2360 &last, &type);
2361 if (IS_ERR(filename))
2362 return ERR_CAST(filename);
2363 if (unlikely(type != LAST_NORM)) {
2364 path_put(path);
2365 putname(filename);
2366 return ERR_PTR(-EINVAL);
2367 }
2368 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2369 d = __lookup_hash(&last, path->dentry, 0);
2370 if (IS_ERR(d)) {
2371 inode_unlock(path->dentry->d_inode);
2372 path_put(path);
2373 }
2374 putname(filename);
2375 return d;
2376}
2377
2378int kern_path(const char *name, unsigned int flags, struct path *path)
2379{
2380 return filename_lookup(AT_FDCWD, getname_kernel(name),
2381 flags, path, NULL);
2382}
2383EXPORT_SYMBOL(kern_path);
2384
2385/**
2386 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2387 * @dentry: pointer to dentry of the base directory
2388 * @mnt: pointer to vfs mount of the base directory
2389 * @name: pointer to file name
2390 * @flags: lookup flags
2391 * @path: pointer to struct path to fill
2392 */
2393int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2394 const char *name, unsigned int flags,
2395 struct path *path)
2396{
2397 struct path root = {.mnt = mnt, .dentry = dentry};
2398 /* the first argument of filename_lookup() is ignored with root */
2399 return filename_lookup(AT_FDCWD, getname_kernel(name),
2400 flags , path, &root);
2401}
2402EXPORT_SYMBOL(vfs_path_lookup);
2403
2404/**
2405 * lookup_one_len - filesystem helper to lookup single pathname component
2406 * @name: pathname component to lookup
2407 * @base: base directory to lookup from
2408 * @len: maximum length @len should be interpreted to
2409 *
2410 * Note that this routine is purely a helper for filesystem usage and should
2411 * not be called by generic code.
2412 *
2413 * The caller must hold base->i_mutex.
2414 */
2415struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2416{
2417 struct qstr this;
2418 unsigned int c;
2419 int err;
2420
2421 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2422
2423 this.name = name;
2424 this.len = len;
2425 this.hash = full_name_hash(base, name, len);
2426 if (!len)
2427 return ERR_PTR(-EACCES);
2428
2429 if (unlikely(name[0] == '.')) {
2430 if (len < 2 || (len == 2 && name[1] == '.'))
2431 return ERR_PTR(-EACCES);
2432 }
2433
2434 while (len--) {
2435 c = *(const unsigned char *)name++;
2436 if (c == '/' || c == '\0')
2437 return ERR_PTR(-EACCES);
2438 }
2439 /*
2440 * See if the low-level filesystem might want
2441 * to use its own hash..
2442 */
2443 if (base->d_flags & DCACHE_OP_HASH) {
2444 int err = base->d_op->d_hash(base, &this);
2445 if (err < 0)
2446 return ERR_PTR(err);
2447 }
2448
2449 err = inode_permission(base->d_inode, MAY_EXEC);
2450 if (err)
2451 return ERR_PTR(err);
2452
2453 return __lookup_hash(&this, base, 0);
2454}
2455EXPORT_SYMBOL(lookup_one_len);
2456
2457/**
2458 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2459 * @name: pathname component to lookup
2460 * @base: base directory to lookup from
2461 * @len: maximum length @len should be interpreted to
2462 *
2463 * Note that this routine is purely a helper for filesystem usage and should
2464 * not be called by generic code.
2465 *
2466 * Unlike lookup_one_len, it should be called without the parent
2467 * i_mutex held, and will take the i_mutex itself if necessary.
2468 */
2469struct dentry *lookup_one_len_unlocked(const char *name,
2470 struct dentry *base, int len)
2471{
2472 struct qstr this;
2473 unsigned int c;
2474 int err;
2475 struct dentry *ret;
2476
2477 this.name = name;
2478 this.len = len;
2479 this.hash = full_name_hash(base, name, len);
2480 if (!len)
2481 return ERR_PTR(-EACCES);
2482
2483 if (unlikely(name[0] == '.')) {
2484 if (len < 2 || (len == 2 && name[1] == '.'))
2485 return ERR_PTR(-EACCES);
2486 }
2487
2488 while (len--) {
2489 c = *(const unsigned char *)name++;
2490 if (c == '/' || c == '\0')
2491 return ERR_PTR(-EACCES);
2492 }
2493 /*
2494 * See if the low-level filesystem might want
2495 * to use its own hash..
2496 */
2497 if (base->d_flags & DCACHE_OP_HASH) {
2498 int err = base->d_op->d_hash(base, &this);
2499 if (err < 0)
2500 return ERR_PTR(err);
2501 }
2502
2503 err = inode_permission(base->d_inode, MAY_EXEC);
2504 if (err)
2505 return ERR_PTR(err);
2506
2507 ret = lookup_dcache(&this, base, 0);
2508 if (!ret)
2509 ret = lookup_slow(&this, base, 0);
2510 return ret;
2511}
2512EXPORT_SYMBOL(lookup_one_len_unlocked);
2513
2514#ifdef CONFIG_UNIX98_PTYS
2515int path_pts(struct path *path)
2516{
2517 /* Find something mounted on "pts" in the same directory as
2518 * the input path.
2519 */
2520 struct dentry *child, *parent;
2521 struct qstr this;
2522 int ret;
2523
2524 ret = path_parent_directory(path);
2525 if (ret)
2526 return ret;
2527
2528 parent = path->dentry;
2529 this.name = "pts";
2530 this.len = 3;
2531 child = d_hash_and_lookup(parent, &this);
2532 if (!child)
2533 return -ENOENT;
2534
2535 path->dentry = child;
2536 dput(parent);
2537 follow_mount(path);
2538 return 0;
2539}
2540#endif
2541
2542int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2543 struct path *path, int *empty)
2544{
2545 return filename_lookup(dfd, getname_flags(name, flags, empty),
2546 flags, path, NULL);
2547}
2548EXPORT_SYMBOL(user_path_at_empty);
2549
2550/**
2551 * mountpoint_last - look up last component for umount
2552 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2553 *
2554 * This is a special lookup_last function just for umount. In this case, we
2555 * need to resolve the path without doing any revalidation.
2556 *
2557 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2558 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2559 * in almost all cases, this lookup will be served out of the dcache. The only
2560 * cases where it won't are if nd->last refers to a symlink or the path is
2561 * bogus and it doesn't exist.
2562 *
2563 * Returns:
2564 * -error: if there was an error during lookup. This includes -ENOENT if the
2565 * lookup found a negative dentry.
2566 *
2567 * 0: if we successfully resolved nd->last and found it to not to be a
2568 * symlink that needs to be followed.
2569 *
2570 * 1: if we successfully resolved nd->last and found it to be a symlink
2571 * that needs to be followed.
2572 */
2573static int
2574mountpoint_last(struct nameidata *nd)
2575{
2576 int error = 0;
2577 struct dentry *dir = nd->path.dentry;
2578 struct path path;
2579
2580 /* If we're in rcuwalk, drop out of it to handle last component */
2581 if (nd->flags & LOOKUP_RCU) {
2582 if (unlazy_walk(nd, NULL, 0))
2583 return -ECHILD;
2584 }
2585
2586 nd->flags &= ~LOOKUP_PARENT;
2587
2588 if (unlikely(nd->last_type != LAST_NORM)) {
2589 error = handle_dots(nd, nd->last_type);
2590 if (error)
2591 return error;
2592 path.dentry = dget(nd->path.dentry);
2593 } else {
2594 path.dentry = d_lookup(dir, &nd->last);
2595 if (!path.dentry) {
2596 /*
2597 * No cached dentry. Mounted dentries are pinned in the
2598 * cache, so that means that this dentry is probably
2599 * a symlink or the path doesn't actually point
2600 * to a mounted dentry.
2601 */
2602 path.dentry = lookup_slow(&nd->last, dir,
2603 nd->flags | LOOKUP_NO_REVAL);
2604 if (IS_ERR(path.dentry))
2605 return PTR_ERR(path.dentry);
2606 }
2607 }
2608 if (d_is_negative(path.dentry)) {
2609 dput(path.dentry);
2610 return -ENOENT;
2611 }
2612 path.mnt = nd->path.mnt;
2613 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2614}
2615
2616/**
2617 * path_mountpoint - look up a path to be umounted
2618 * @nd: lookup context
2619 * @flags: lookup flags
2620 * @path: pointer to container for result
2621 *
2622 * Look up the given name, but don't attempt to revalidate the last component.
2623 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2624 */
2625static int
2626path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2627{
2628 const char *s = path_init(nd, flags);
2629 int err;
2630 if (IS_ERR(s))
2631 return PTR_ERR(s);
2632 while (!(err = link_path_walk(s, nd)) &&
2633 (err = mountpoint_last(nd)) > 0) {
2634 s = trailing_symlink(nd);
2635 if (IS_ERR(s)) {
2636 err = PTR_ERR(s);
2637 break;
2638 }
2639 }
2640 if (!err) {
2641 *path = nd->path;
2642 nd->path.mnt = NULL;
2643 nd->path.dentry = NULL;
2644 follow_mount(path);
2645 }
2646 terminate_walk(nd);
2647 return err;
2648}
2649
2650static int
2651filename_mountpoint(int dfd, struct filename *name, struct path *path,
2652 unsigned int flags)
2653{
2654 struct nameidata nd;
2655 int error;
2656 if (IS_ERR(name))
2657 return PTR_ERR(name);
2658 set_nameidata(&nd, dfd, name);
2659 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2660 if (unlikely(error == -ECHILD))
2661 error = path_mountpoint(&nd, flags, path);
2662 if (unlikely(error == -ESTALE))
2663 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2664 if (likely(!error))
2665 audit_inode(name, path->dentry, 0);
2666 restore_nameidata();
2667 putname(name);
2668 return error;
2669}
2670
2671/**
2672 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2673 * @dfd: directory file descriptor
2674 * @name: pathname from userland
2675 * @flags: lookup flags
2676 * @path: pointer to container to hold result
2677 *
2678 * A umount is a special case for path walking. We're not actually interested
2679 * in the inode in this situation, and ESTALE errors can be a problem. We
2680 * simply want track down the dentry and vfsmount attached at the mountpoint
2681 * and avoid revalidating the last component.
2682 *
2683 * Returns 0 and populates "path" on success.
2684 */
2685int
2686user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2687 struct path *path)
2688{
2689 return filename_mountpoint(dfd, getname(name), path, flags);
2690}
2691
2692int
2693kern_path_mountpoint(int dfd, const char *name, struct path *path,
2694 unsigned int flags)
2695{
2696 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2697}
2698EXPORT_SYMBOL(kern_path_mountpoint);
2699
2700int __check_sticky(struct inode *dir, struct inode *inode)
2701{
2702 kuid_t fsuid = current_fsuid();
2703
2704 if (uid_eq(inode->i_uid, fsuid))
2705 return 0;
2706 if (uid_eq(dir->i_uid, fsuid))
2707 return 0;
2708 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2709}
2710EXPORT_SYMBOL(__check_sticky);
2711
2712/*
2713 * Check whether we can remove a link victim from directory dir, check
2714 * whether the type of victim is right.
2715 * 1. We can't do it if dir is read-only (done in permission())
2716 * 2. We should have write and exec permissions on dir
2717 * 3. We can't remove anything from append-only dir
2718 * 4. We can't do anything with immutable dir (done in permission())
2719 * 5. If the sticky bit on dir is set we should either
2720 * a. be owner of dir, or
2721 * b. be owner of victim, or
2722 * c. have CAP_FOWNER capability
2723 * 6. If the victim is append-only or immutable we can't do antyhing with
2724 * links pointing to it.
2725 * 7. If the victim has an unknown uid or gid we can't change the inode.
2726 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2727 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2728 * 10. We can't remove a root or mountpoint.
2729 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2730 * nfs_async_unlink().
2731 */
2732static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2733{
2734 struct inode *inode = d_backing_inode(victim);
2735 int error;
2736
2737 if (d_is_negative(victim))
2738 return -ENOENT;
2739 BUG_ON(!inode);
2740
2741 BUG_ON(victim->d_parent->d_inode != dir);
2742 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2743
2744 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2745 if (error)
2746 return error;
2747 if (IS_APPEND(dir))
2748 return -EPERM;
2749
2750 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2751 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2752 return -EPERM;
2753 if (isdir) {
2754 if (!d_is_dir(victim))
2755 return -ENOTDIR;
2756 if (IS_ROOT(victim))
2757 return -EBUSY;
2758 } else if (d_is_dir(victim))
2759 return -EISDIR;
2760 if (IS_DEADDIR(dir))
2761 return -ENOENT;
2762 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2763 return -EBUSY;
2764 return 0;
2765}
2766
2767/* Check whether we can create an object with dentry child in directory
2768 * dir.
2769 * 1. We can't do it if child already exists (open has special treatment for
2770 * this case, but since we are inlined it's OK)
2771 * 2. We can't do it if dir is read-only (done in permission())
2772 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2773 * 4. We should have write and exec permissions on dir
2774 * 5. We can't do it if dir is immutable (done in permission())
2775 */
2776static inline int may_create(struct inode *dir, struct dentry *child)
2777{
2778 struct user_namespace *s_user_ns;
2779 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2780 if (child->d_inode)
2781 return -EEXIST;
2782 if (IS_DEADDIR(dir))
2783 return -ENOENT;
2784 s_user_ns = dir->i_sb->s_user_ns;
2785 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2786 !kgid_has_mapping(s_user_ns, current_fsgid()))
2787 return -EOVERFLOW;
2788 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2789}
2790
2791/*
2792 * p1 and p2 should be directories on the same fs.
2793 */
2794struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2795{
2796 struct dentry *p;
2797
2798 if (p1 == p2) {
2799 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2800 return NULL;
2801 }
2802
2803 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2804
2805 p = d_ancestor(p2, p1);
2806 if (p) {
2807 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2808 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2809 return p;
2810 }
2811
2812 p = d_ancestor(p1, p2);
2813 if (p) {
2814 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2815 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2816 return p;
2817 }
2818
2819 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2820 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2821 return NULL;
2822}
2823EXPORT_SYMBOL(lock_rename);
2824
2825void unlock_rename(struct dentry *p1, struct dentry *p2)
2826{
2827 inode_unlock(p1->d_inode);
2828 if (p1 != p2) {
2829 inode_unlock(p2->d_inode);
2830 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2831 }
2832}
2833EXPORT_SYMBOL(unlock_rename);
2834
2835int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2836 bool want_excl)
2837{
2838 int error = may_create(dir, dentry);
2839 if (error)
2840 return error;
2841
2842 if (!dir->i_op->create)
2843 return -EACCES; /* shouldn't it be ENOSYS? */
2844 mode &= S_IALLUGO;
2845 mode |= S_IFREG;
2846 error = security_inode_create(dir, dentry, mode);
2847 if (error)
2848 return error;
2849 error = dir->i_op->create(dir, dentry, mode, want_excl);
2850 if (!error)
2851 fsnotify_create(dir, dentry);
2852 return error;
2853}
2854EXPORT_SYMBOL(vfs_create);
2855
2856bool may_open_dev(const struct path *path)
2857{
2858 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2859 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2860}
2861
2862static int may_open(const struct path *path, int acc_mode, int flag)
2863{
2864 struct dentry *dentry = path->dentry;
2865 struct inode *inode = dentry->d_inode;
2866 int error;
2867
2868 if (!inode)
2869 return -ENOENT;
2870
2871 switch (inode->i_mode & S_IFMT) {
2872 case S_IFLNK:
2873 return -ELOOP;
2874 case S_IFDIR:
2875 if (acc_mode & MAY_WRITE)
2876 return -EISDIR;
2877 break;
2878 case S_IFBLK:
2879 case S_IFCHR:
2880 if (!may_open_dev(path))
2881 return -EACCES;
2882 /*FALLTHRU*/
2883 case S_IFIFO:
2884 case S_IFSOCK:
2885 flag &= ~O_TRUNC;
2886 break;
2887 }
2888
2889 error = inode_permission(inode, MAY_OPEN | acc_mode);
2890 if (error)
2891 return error;
2892
2893 /*
2894 * An append-only file must be opened in append mode for writing.
2895 */
2896 if (IS_APPEND(inode)) {
2897 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2898 return -EPERM;
2899 if (flag & O_TRUNC)
2900 return -EPERM;
2901 }
2902
2903 /* O_NOATIME can only be set by the owner or superuser */
2904 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2905 return -EPERM;
2906
2907 return 0;
2908}
2909
2910static int handle_truncate(struct file *filp)
2911{
2912 const struct path *path = &filp->f_path;
2913 struct inode *inode = path->dentry->d_inode;
2914 int error = get_write_access(inode);
2915 if (error)
2916 return error;
2917 /*
2918 * Refuse to truncate files with mandatory locks held on them.
2919 */
2920 error = locks_verify_locked(filp);
2921 if (!error)
2922 error = security_path_truncate(path);
2923 if (!error) {
2924 error = do_truncate(path->dentry, 0,
2925 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2926 filp);
2927 }
2928 put_write_access(inode);
2929 return error;
2930}
2931
2932static inline int open_to_namei_flags(int flag)
2933{
2934 if ((flag & O_ACCMODE) == 3)
2935 flag--;
2936 return flag;
2937}
2938
2939static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2940{
2941 int error = security_path_mknod(dir, dentry, mode, 0);
2942 if (error)
2943 return error;
2944
2945 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2946 if (error)
2947 return error;
2948
2949 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2950}
2951
2952/*
2953 * Attempt to atomically look up, create and open a file from a negative
2954 * dentry.
2955 *
2956 * Returns 0 if successful. The file will have been created and attached to
2957 * @file by the filesystem calling finish_open().
2958 *
2959 * Returns 1 if the file was looked up only or didn't need creating. The
2960 * caller will need to perform the open themselves. @path will have been
2961 * updated to point to the new dentry. This may be negative.
2962 *
2963 * Returns an error code otherwise.
2964 */
2965static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2966 struct path *path, struct file *file,
2967 const struct open_flags *op,
2968 int open_flag, umode_t mode,
2969 int *opened)
2970{
2971 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2972 struct inode *dir = nd->path.dentry->d_inode;
2973 int error;
2974
2975 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
2976 open_flag &= ~O_TRUNC;
2977
2978 if (nd->flags & LOOKUP_DIRECTORY)
2979 open_flag |= O_DIRECTORY;
2980
2981 file->f_path.dentry = DENTRY_NOT_SET;
2982 file->f_path.mnt = nd->path.mnt;
2983 error = dir->i_op->atomic_open(dir, dentry, file,
2984 open_to_namei_flags(open_flag),
2985 mode, opened);
2986 d_lookup_done(dentry);
2987 if (!error) {
2988 /*
2989 * We didn't have the inode before the open, so check open
2990 * permission here.
2991 */
2992 int acc_mode = op->acc_mode;
2993 if (*opened & FILE_CREATED) {
2994 WARN_ON(!(open_flag & O_CREAT));
2995 fsnotify_create(dir, dentry);
2996 acc_mode = 0;
2997 }
2998 error = may_open(&file->f_path, acc_mode, open_flag);
2999 if (WARN_ON(error > 0))
3000 error = -EINVAL;
3001 } else if (error > 0) {
3002 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3003 error = -EIO;
3004 } else {
3005 if (file->f_path.dentry) {
3006 dput(dentry);
3007 dentry = file->f_path.dentry;
3008 }
3009 if (*opened & FILE_CREATED)
3010 fsnotify_create(dir, dentry);
3011 if (unlikely(d_is_negative(dentry))) {
3012 error = -ENOENT;
3013 } else {
3014 path->dentry = dentry;
3015 path->mnt = nd->path.mnt;
3016 return 1;
3017 }
3018 }
3019 }
3020 dput(dentry);
3021 return error;
3022}
3023
3024/*
3025 * Look up and maybe create and open the last component.
3026 *
3027 * Must be called with i_mutex held on parent.
3028 *
3029 * Returns 0 if the file was successfully atomically created (if necessary) and
3030 * opened. In this case the file will be returned attached to @file.
3031 *
3032 * Returns 1 if the file was not completely opened at this time, though lookups
3033 * and creations will have been performed and the dentry returned in @path will
3034 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3035 * specified then a negative dentry may be returned.
3036 *
3037 * An error code is returned otherwise.
3038 *
3039 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3040 * cleared otherwise prior to returning.
3041 */
3042static int lookup_open(struct nameidata *nd, struct path *path,
3043 struct file *file,
3044 const struct open_flags *op,
3045 bool got_write, int *opened)
3046{
3047 struct dentry *dir = nd->path.dentry;
3048 struct inode *dir_inode = dir->d_inode;
3049 int open_flag = op->open_flag;
3050 struct dentry *dentry;
3051 int error, create_error = 0;
3052 umode_t mode = op->mode;
3053 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3054
3055 if (unlikely(IS_DEADDIR(dir_inode)))
3056 return -ENOENT;
3057
3058 *opened &= ~FILE_CREATED;
3059 dentry = d_lookup(dir, &nd->last);
3060 for (;;) {
3061 if (!dentry) {
3062 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3063 if (IS_ERR(dentry))
3064 return PTR_ERR(dentry);
3065 }
3066 if (d_in_lookup(dentry))
3067 break;
3068
3069 if (!(dentry->d_flags & DCACHE_OP_REVALIDATE))
3070 break;
3071
3072 error = d_revalidate(dentry, nd->flags);
3073 if (likely(error > 0))
3074 break;
3075 if (error)
3076 goto out_dput;
3077 d_invalidate(dentry);
3078 dput(dentry);
3079 dentry = NULL;
3080 }
3081 if (dentry->d_inode) {
3082 /* Cached positive dentry: will open in f_op->open */
3083 goto out_no_open;
3084 }
3085
3086 /*
3087 * Checking write permission is tricky, bacuse we don't know if we are
3088 * going to actually need it: O_CREAT opens should work as long as the
3089 * file exists. But checking existence breaks atomicity. The trick is
3090 * to check access and if not granted clear O_CREAT from the flags.
3091 *
3092 * Another problem is returing the "right" error value (e.g. for an
3093 * O_EXCL open we want to return EEXIST not EROFS).
3094 */
3095 if (open_flag & O_CREAT) {
3096 if (!IS_POSIXACL(dir->d_inode))
3097 mode &= ~current_umask();
3098 if (unlikely(!got_write)) {
3099 create_error = -EROFS;
3100 open_flag &= ~O_CREAT;
3101 if (open_flag & (O_EXCL | O_TRUNC))
3102 goto no_open;
3103 /* No side effects, safe to clear O_CREAT */
3104 } else {
3105 create_error = may_o_create(&nd->path, dentry, mode);
3106 if (create_error) {
3107 open_flag &= ~O_CREAT;
3108 if (open_flag & O_EXCL)
3109 goto no_open;
3110 }
3111 }
3112 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3113 unlikely(!got_write)) {
3114 /*
3115 * No O_CREATE -> atomicity not a requirement -> fall
3116 * back to lookup + open
3117 */
3118 goto no_open;
3119 }
3120
3121 if (dir_inode->i_op->atomic_open) {
3122 error = atomic_open(nd, dentry, path, file, op, open_flag,
3123 mode, opened);
3124 if (unlikely(error == -ENOENT) && create_error)
3125 error = create_error;
3126 return error;
3127 }
3128
3129no_open:
3130 if (d_in_lookup(dentry)) {
3131 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3132 nd->flags);
3133 d_lookup_done(dentry);
3134 if (unlikely(res)) {
3135 if (IS_ERR(res)) {
3136 error = PTR_ERR(res);
3137 goto out_dput;
3138 }
3139 dput(dentry);
3140 dentry = res;
3141 }
3142 }
3143
3144 /* Negative dentry, just create the file */
3145 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3146 *opened |= FILE_CREATED;
3147 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3148 if (!dir_inode->i_op->create) {
3149 error = -EACCES;
3150 goto out_dput;
3151 }
3152 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3153 open_flag & O_EXCL);
3154 if (error)
3155 goto out_dput;
3156 fsnotify_create(dir_inode, dentry);
3157 }
3158 if (unlikely(create_error) && !dentry->d_inode) {
3159 error = create_error;
3160 goto out_dput;
3161 }
3162out_no_open:
3163 path->dentry = dentry;
3164 path->mnt = nd->path.mnt;
3165 return 1;
3166
3167out_dput:
3168 dput(dentry);
3169 return error;
3170}
3171
3172/*
3173 * Handle the last step of open()
3174 */
3175static int do_last(struct nameidata *nd,
3176 struct file *file, const struct open_flags *op,
3177 int *opened)
3178{
3179 struct dentry *dir = nd->path.dentry;
3180 int open_flag = op->open_flag;
3181 bool will_truncate = (open_flag & O_TRUNC) != 0;
3182 bool got_write = false;
3183 int acc_mode = op->acc_mode;
3184 unsigned seq;
3185 struct inode *inode;
3186 struct path path;
3187 int error;
3188
3189 nd->flags &= ~LOOKUP_PARENT;
3190 nd->flags |= op->intent;
3191
3192 if (nd->last_type != LAST_NORM) {
3193 error = handle_dots(nd, nd->last_type);
3194 if (unlikely(error))
3195 return error;
3196 goto finish_open;
3197 }
3198
3199 if (!(open_flag & O_CREAT)) {
3200 if (nd->last.name[nd->last.len])
3201 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3202 /* we _can_ be in RCU mode here */
3203 error = lookup_fast(nd, &path, &inode, &seq);
3204 if (likely(error > 0))
3205 goto finish_lookup;
3206
3207 if (error < 0)
3208 return error;
3209
3210 BUG_ON(nd->inode != dir->d_inode);
3211 BUG_ON(nd->flags & LOOKUP_RCU);
3212 } else {
3213 /* create side of things */
3214 /*
3215 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3216 * has been cleared when we got to the last component we are
3217 * about to look up
3218 */
3219 error = complete_walk(nd);
3220 if (error)
3221 return error;
3222
3223 audit_inode(nd->name, dir, LOOKUP_PARENT);
3224 /* trailing slashes? */
3225 if (unlikely(nd->last.name[nd->last.len]))
3226 return -EISDIR;
3227 }
3228
3229 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3230 error = mnt_want_write(nd->path.mnt);
3231 if (!error)
3232 got_write = true;
3233 /*
3234 * do _not_ fail yet - we might not need that or fail with
3235 * a different error; let lookup_open() decide; we'll be
3236 * dropping this one anyway.
3237 */
3238 }
3239 if (open_flag & O_CREAT)
3240 inode_lock(dir->d_inode);
3241 else
3242 inode_lock_shared(dir->d_inode);
3243 error = lookup_open(nd, &path, file, op, got_write, opened);
3244 if (open_flag & O_CREAT)
3245 inode_unlock(dir->d_inode);
3246 else
3247 inode_unlock_shared(dir->d_inode);
3248
3249 if (error <= 0) {
3250 if (error)
3251 goto out;
3252
3253 if ((*opened & FILE_CREATED) ||
3254 !S_ISREG(file_inode(file)->i_mode))
3255 will_truncate = false;
3256
3257 audit_inode(nd->name, file->f_path.dentry, 0);
3258 goto opened;
3259 }
3260
3261 if (*opened & FILE_CREATED) {
3262 /* Don't check for write permission, don't truncate */
3263 open_flag &= ~O_TRUNC;
3264 will_truncate = false;
3265 acc_mode = 0;
3266 path_to_nameidata(&path, nd);
3267 goto finish_open_created;
3268 }
3269
3270 /*
3271 * If atomic_open() acquired write access it is dropped now due to
3272 * possible mount and symlink following (this might be optimized away if
3273 * necessary...)
3274 */
3275 if (got_write) {
3276 mnt_drop_write(nd->path.mnt);
3277 got_write = false;
3278 }
3279
3280 error = follow_managed(&path, nd);
3281 if (unlikely(error < 0))
3282 return error;
3283
3284 if (unlikely(d_is_negative(path.dentry))) {
3285 path_to_nameidata(&path, nd);
3286 return -ENOENT;
3287 }
3288
3289 /*
3290 * create/update audit record if it already exists.
3291 */
3292 audit_inode(nd->name, path.dentry, 0);
3293
3294 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3295 path_to_nameidata(&path, nd);
3296 return -EEXIST;
3297 }
3298
3299 seq = 0; /* out of RCU mode, so the value doesn't matter */
3300 inode = d_backing_inode(path.dentry);
3301finish_lookup:
3302 error = step_into(nd, &path, 0, inode, seq);
3303 if (unlikely(error))
3304 return error;
3305finish_open:
3306 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3307 error = complete_walk(nd);
3308 if (error)
3309 return error;
3310 audit_inode(nd->name, nd->path.dentry, 0);
3311 error = -EISDIR;
3312 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3313 goto out;
3314 error = -ENOTDIR;
3315 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3316 goto out;
3317 if (!d_is_reg(nd->path.dentry))
3318 will_truncate = false;
3319
3320 if (will_truncate) {
3321 error = mnt_want_write(nd->path.mnt);
3322 if (error)
3323 goto out;
3324 got_write = true;
3325 }
3326finish_open_created:
3327 error = may_open(&nd->path, acc_mode, open_flag);
3328 if (error)
3329 goto out;
3330 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3331 error = vfs_open(&nd->path, file, current_cred());
3332 if (error)
3333 goto out;
3334 *opened |= FILE_OPENED;
3335opened:
3336 error = open_check_o_direct(file);
3337 if (!error)
3338 error = ima_file_check(file, op->acc_mode, *opened);
3339 if (!error && will_truncate)
3340 error = handle_truncate(file);
3341out:
3342 if (unlikely(error) && (*opened & FILE_OPENED))
3343 fput(file);
3344 if (unlikely(error > 0)) {
3345 WARN_ON(1);
3346 error = -EINVAL;
3347 }
3348 if (got_write)
3349 mnt_drop_write(nd->path.mnt);
3350 return error;
3351}
3352
3353static int do_tmpfile(struct nameidata *nd, unsigned flags,
3354 const struct open_flags *op,
3355 struct file *file, int *opened)
3356{
3357 static const struct qstr name = QSTR_INIT("/", 1);
3358 struct dentry *child;
3359 struct inode *dir;
3360 struct path path;
3361 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3362 if (unlikely(error))
3363 return error;
3364 error = mnt_want_write(path.mnt);
3365 if (unlikely(error))
3366 goto out;
3367 dir = path.dentry->d_inode;
3368 /* we want directory to be writable */
3369 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3370 if (error)
3371 goto out2;
3372 if (!dir->i_op->tmpfile) {
3373 error = -EOPNOTSUPP;
3374 goto out2;
3375 }
3376 child = d_alloc(path.dentry, &name);
3377 if (unlikely(!child)) {
3378 error = -ENOMEM;
3379 goto out2;
3380 }
3381 dput(path.dentry);
3382 path.dentry = child;
3383 error = dir->i_op->tmpfile(dir, child, op->mode);
3384 if (error)
3385 goto out2;
3386 audit_inode(nd->name, child, 0);
3387 /* Don't check for other permissions, the inode was just created */
3388 error = may_open(&path, 0, op->open_flag);
3389 if (error)
3390 goto out2;
3391 file->f_path.mnt = path.mnt;
3392 error = finish_open(file, child, NULL, opened);
3393 if (error)
3394 goto out2;
3395 error = open_check_o_direct(file);
3396 if (error) {
3397 fput(file);
3398 } else if (!(op->open_flag & O_EXCL)) {
3399 struct inode *inode = file_inode(file);
3400 spin_lock(&inode->i_lock);
3401 inode->i_state |= I_LINKABLE;
3402 spin_unlock(&inode->i_lock);
3403 }
3404out2:
3405 mnt_drop_write(path.mnt);
3406out:
3407 path_put(&path);
3408 return error;
3409}
3410
3411static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3412{
3413 struct path path;
3414 int error = path_lookupat(nd, flags, &path);
3415 if (!error) {
3416 audit_inode(nd->name, path.dentry, 0);
3417 error = vfs_open(&path, file, current_cred());
3418 path_put(&path);
3419 }
3420 return error;
3421}
3422
3423static struct file *path_openat(struct nameidata *nd,
3424 const struct open_flags *op, unsigned flags)
3425{
3426 const char *s;
3427 struct file *file;
3428 int opened = 0;
3429 int error;
3430
3431 file = get_empty_filp();
3432 if (IS_ERR(file))
3433 return file;
3434
3435 file->f_flags = op->open_flag;
3436
3437 if (unlikely(file->f_flags & __O_TMPFILE)) {
3438 error = do_tmpfile(nd, flags, op, file, &opened);
3439 goto out2;
3440 }
3441
3442 if (unlikely(file->f_flags & O_PATH)) {
3443 error = do_o_path(nd, flags, file);
3444 if (!error)
3445 opened |= FILE_OPENED;
3446 goto out2;
3447 }
3448
3449 s = path_init(nd, flags);
3450 if (IS_ERR(s)) {
3451 put_filp(file);
3452 return ERR_CAST(s);
3453 }
3454 while (!(error = link_path_walk(s, nd)) &&
3455 (error = do_last(nd, file, op, &opened)) > 0) {
3456 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3457 s = trailing_symlink(nd);
3458 if (IS_ERR(s)) {
3459 error = PTR_ERR(s);
3460 break;
3461 }
3462 }
3463 terminate_walk(nd);
3464out2:
3465 if (!(opened & FILE_OPENED)) {
3466 BUG_ON(!error);
3467 put_filp(file);
3468 }
3469 if (unlikely(error)) {
3470 if (error == -EOPENSTALE) {
3471 if (flags & LOOKUP_RCU)
3472 error = -ECHILD;
3473 else
3474 error = -ESTALE;
3475 }
3476 file = ERR_PTR(error);
3477 }
3478 return file;
3479}
3480
3481struct file *do_filp_open(int dfd, struct filename *pathname,
3482 const struct open_flags *op)
3483{
3484 struct nameidata nd;
3485 int flags = op->lookup_flags;
3486 struct file *filp;
3487
3488 set_nameidata(&nd, dfd, pathname);
3489 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3490 if (unlikely(filp == ERR_PTR(-ECHILD)))
3491 filp = path_openat(&nd, op, flags);
3492 if (unlikely(filp == ERR_PTR(-ESTALE)))
3493 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3494 restore_nameidata();
3495 return filp;
3496}
3497
3498struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3499 const char *name, const struct open_flags *op)
3500{
3501 struct nameidata nd;
3502 struct file *file;
3503 struct filename *filename;
3504 int flags = op->lookup_flags | LOOKUP_ROOT;
3505
3506 nd.root.mnt = mnt;
3507 nd.root.dentry = dentry;
3508
3509 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3510 return ERR_PTR(-ELOOP);
3511
3512 filename = getname_kernel(name);
3513 if (IS_ERR(filename))
3514 return ERR_CAST(filename);
3515
3516 set_nameidata(&nd, -1, filename);
3517 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3518 if (unlikely(file == ERR_PTR(-ECHILD)))
3519 file = path_openat(&nd, op, flags);
3520 if (unlikely(file == ERR_PTR(-ESTALE)))
3521 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3522 restore_nameidata();
3523 putname(filename);
3524 return file;
3525}
3526
3527static struct dentry *filename_create(int dfd, struct filename *name,
3528 struct path *path, unsigned int lookup_flags)
3529{
3530 struct dentry *dentry = ERR_PTR(-EEXIST);
3531 struct qstr last;
3532 int type;
3533 int err2;
3534 int error;
3535 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3536
3537 /*
3538 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3539 * other flags passed in are ignored!
3540 */
3541 lookup_flags &= LOOKUP_REVAL;
3542
3543 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3544 if (IS_ERR(name))
3545 return ERR_CAST(name);
3546
3547 /*
3548 * Yucky last component or no last component at all?
3549 * (foo/., foo/.., /////)
3550 */
3551 if (unlikely(type != LAST_NORM))
3552 goto out;
3553
3554 /* don't fail immediately if it's r/o, at least try to report other errors */
3555 err2 = mnt_want_write(path->mnt);
3556 /*
3557 * Do the final lookup.
3558 */
3559 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3560 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3561 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3562 if (IS_ERR(dentry))
3563 goto unlock;
3564
3565 error = -EEXIST;
3566 if (d_is_positive(dentry))
3567 goto fail;
3568
3569 /*
3570 * Special case - lookup gave negative, but... we had foo/bar/
3571 * From the vfs_mknod() POV we just have a negative dentry -
3572 * all is fine. Let's be bastards - you had / on the end, you've
3573 * been asking for (non-existent) directory. -ENOENT for you.
3574 */
3575 if (unlikely(!is_dir && last.name[last.len])) {
3576 error = -ENOENT;
3577 goto fail;
3578 }
3579 if (unlikely(err2)) {
3580 error = err2;
3581 goto fail;
3582 }
3583 putname(name);
3584 return dentry;
3585fail:
3586 dput(dentry);
3587 dentry = ERR_PTR(error);
3588unlock:
3589 inode_unlock(path->dentry->d_inode);
3590 if (!err2)
3591 mnt_drop_write(path->mnt);
3592out:
3593 path_put(path);
3594 putname(name);
3595 return dentry;
3596}
3597
3598struct dentry *kern_path_create(int dfd, const char *pathname,
3599 struct path *path, unsigned int lookup_flags)
3600{
3601 return filename_create(dfd, getname_kernel(pathname),
3602 path, lookup_flags);
3603}
3604EXPORT_SYMBOL(kern_path_create);
3605
3606void done_path_create(struct path *path, struct dentry *dentry)
3607{
3608 dput(dentry);
3609 inode_unlock(path->dentry->d_inode);
3610 mnt_drop_write(path->mnt);
3611 path_put(path);
3612}
3613EXPORT_SYMBOL(done_path_create);
3614
3615inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3616 struct path *path, unsigned int lookup_flags)
3617{
3618 return filename_create(dfd, getname(pathname), path, lookup_flags);
3619}
3620EXPORT_SYMBOL(user_path_create);
3621
3622int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3623{
3624 int error = may_create(dir, dentry);
3625
3626 if (error)
3627 return error;
3628
3629 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3630 return -EPERM;
3631
3632 if (!dir->i_op->mknod)
3633 return -EPERM;
3634
3635 error = devcgroup_inode_mknod(mode, dev);
3636 if (error)
3637 return error;
3638
3639 error = security_inode_mknod(dir, dentry, mode, dev);
3640 if (error)
3641 return error;
3642
3643 error = dir->i_op->mknod(dir, dentry, mode, dev);
3644 if (!error)
3645 fsnotify_create(dir, dentry);
3646 return error;
3647}
3648EXPORT_SYMBOL(vfs_mknod);
3649
3650static int may_mknod(umode_t mode)
3651{
3652 switch (mode & S_IFMT) {
3653 case S_IFREG:
3654 case S_IFCHR:
3655 case S_IFBLK:
3656 case S_IFIFO:
3657 case S_IFSOCK:
3658 case 0: /* zero mode translates to S_IFREG */
3659 return 0;
3660 case S_IFDIR:
3661 return -EPERM;
3662 default:
3663 return -EINVAL;
3664 }
3665}
3666
3667SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3668 unsigned, dev)
3669{
3670 struct dentry *dentry;
3671 struct path path;
3672 int error;
3673 unsigned int lookup_flags = 0;
3674
3675 error = may_mknod(mode);
3676 if (error)
3677 return error;
3678retry:
3679 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3680 if (IS_ERR(dentry))
3681 return PTR_ERR(dentry);
3682
3683 if (!IS_POSIXACL(path.dentry->d_inode))
3684 mode &= ~current_umask();
3685 error = security_path_mknod(&path, dentry, mode, dev);
3686 if (error)
3687 goto out;
3688 switch (mode & S_IFMT) {
3689 case 0: case S_IFREG:
3690 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3691 if (!error)
3692 ima_post_path_mknod(dentry);
3693 break;
3694 case S_IFCHR: case S_IFBLK:
3695 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3696 new_decode_dev(dev));
3697 break;
3698 case S_IFIFO: case S_IFSOCK:
3699 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3700 break;
3701 }
3702out:
3703 done_path_create(&path, dentry);
3704 if (retry_estale(error, lookup_flags)) {
3705 lookup_flags |= LOOKUP_REVAL;
3706 goto retry;
3707 }
3708 return error;
3709}
3710
3711SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3712{
3713 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3714}
3715
3716int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3717{
3718 int error = may_create(dir, dentry);
3719 unsigned max_links = dir->i_sb->s_max_links;
3720
3721 if (error)
3722 return error;
3723
3724 if (!dir->i_op->mkdir)
3725 return -EPERM;
3726
3727 mode &= (S_IRWXUGO|S_ISVTX);
3728 error = security_inode_mkdir(dir, dentry, mode);
3729 if (error)
3730 return error;
3731
3732 if (max_links && dir->i_nlink >= max_links)
3733 return -EMLINK;
3734
3735 error = dir->i_op->mkdir(dir, dentry, mode);
3736 if (!error)
3737 fsnotify_mkdir(dir, dentry);
3738 return error;
3739}
3740EXPORT_SYMBOL(vfs_mkdir);
3741
3742SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3743{
3744 struct dentry *dentry;
3745 struct path path;
3746 int error;
3747 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3748
3749retry:
3750 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3751 if (IS_ERR(dentry))
3752 return PTR_ERR(dentry);
3753
3754 if (!IS_POSIXACL(path.dentry->d_inode))
3755 mode &= ~current_umask();
3756 error = security_path_mkdir(&path, dentry, mode);
3757 if (!error)
3758 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3759 done_path_create(&path, dentry);
3760 if (retry_estale(error, lookup_flags)) {
3761 lookup_flags |= LOOKUP_REVAL;
3762 goto retry;
3763 }
3764 return error;
3765}
3766
3767SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3768{
3769 return sys_mkdirat(AT_FDCWD, pathname, mode);
3770}
3771
3772int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3773{
3774 int error = may_delete(dir, dentry, 1);
3775
3776 if (error)
3777 return error;
3778
3779 if (!dir->i_op->rmdir)
3780 return -EPERM;
3781
3782 dget(dentry);
3783 inode_lock(dentry->d_inode);
3784
3785 error = -EBUSY;
3786 if (is_local_mountpoint(dentry))
3787 goto out;
3788
3789 error = security_inode_rmdir(dir, dentry);
3790 if (error)
3791 goto out;
3792
3793 shrink_dcache_parent(dentry);
3794 error = dir->i_op->rmdir(dir, dentry);
3795 if (error)
3796 goto out;
3797
3798 dentry->d_inode->i_flags |= S_DEAD;
3799 dont_mount(dentry);
3800 detach_mounts(dentry);
3801
3802out:
3803 inode_unlock(dentry->d_inode);
3804 dput(dentry);
3805 if (!error)
3806 d_delete(dentry);
3807 return error;
3808}
3809EXPORT_SYMBOL(vfs_rmdir);
3810
3811static long do_rmdir(int dfd, const char __user *pathname)
3812{
3813 int error = 0;
3814 struct filename *name;
3815 struct dentry *dentry;
3816 struct path path;
3817 struct qstr last;
3818 int type;
3819 unsigned int lookup_flags = 0;
3820retry:
3821 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3822 &path, &last, &type);
3823 if (IS_ERR(name))
3824 return PTR_ERR(name);
3825
3826 switch (type) {
3827 case LAST_DOTDOT:
3828 error = -ENOTEMPTY;
3829 goto exit1;
3830 case LAST_DOT:
3831 error = -EINVAL;
3832 goto exit1;
3833 case LAST_ROOT:
3834 error = -EBUSY;
3835 goto exit1;
3836 }
3837
3838 error = mnt_want_write(path.mnt);
3839 if (error)
3840 goto exit1;
3841
3842 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3843 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3844 error = PTR_ERR(dentry);
3845 if (IS_ERR(dentry))
3846 goto exit2;
3847 if (!dentry->d_inode) {
3848 error = -ENOENT;
3849 goto exit3;
3850 }
3851 error = security_path_rmdir(&path, dentry);
3852 if (error)
3853 goto exit3;
3854 error = vfs_rmdir(path.dentry->d_inode, dentry);
3855exit3:
3856 dput(dentry);
3857exit2:
3858 inode_unlock(path.dentry->d_inode);
3859 mnt_drop_write(path.mnt);
3860exit1:
3861 path_put(&path);
3862 putname(name);
3863 if (retry_estale(error, lookup_flags)) {
3864 lookup_flags |= LOOKUP_REVAL;
3865 goto retry;
3866 }
3867 return error;
3868}
3869
3870SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3871{
3872 return do_rmdir(AT_FDCWD, pathname);
3873}
3874
3875/**
3876 * vfs_unlink - unlink a filesystem object
3877 * @dir: parent directory
3878 * @dentry: victim
3879 * @delegated_inode: returns victim inode, if the inode is delegated.
3880 *
3881 * The caller must hold dir->i_mutex.
3882 *
3883 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3884 * return a reference to the inode in delegated_inode. The caller
3885 * should then break the delegation on that inode and retry. Because
3886 * breaking a delegation may take a long time, the caller should drop
3887 * dir->i_mutex before doing so.
3888 *
3889 * Alternatively, a caller may pass NULL for delegated_inode. This may
3890 * be appropriate for callers that expect the underlying filesystem not
3891 * to be NFS exported.
3892 */
3893int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3894{
3895 struct inode *target = dentry->d_inode;
3896 int error = may_delete(dir, dentry, 0);
3897
3898 if (error)
3899 return error;
3900
3901 if (!dir->i_op->unlink)
3902 return -EPERM;
3903
3904 inode_lock(target);
3905 if (is_local_mountpoint(dentry))
3906 error = -EBUSY;
3907 else {
3908 error = security_inode_unlink(dir, dentry);
3909 if (!error) {
3910 error = try_break_deleg(target, delegated_inode);
3911 if (error)
3912 goto out;
3913 error = dir->i_op->unlink(dir, dentry);
3914 if (!error) {
3915 dont_mount(dentry);
3916 detach_mounts(dentry);
3917 }
3918 }
3919 }
3920out:
3921 inode_unlock(target);
3922
3923 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3924 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3925 fsnotify_link_count(target);
3926 d_delete(dentry);
3927 }
3928
3929 return error;
3930}
3931EXPORT_SYMBOL(vfs_unlink);
3932
3933/*
3934 * Make sure that the actual truncation of the file will occur outside its
3935 * directory's i_mutex. Truncate can take a long time if there is a lot of
3936 * writeout happening, and we don't want to prevent access to the directory
3937 * while waiting on the I/O.
3938 */
3939static long do_unlinkat(int dfd, const char __user *pathname)
3940{
3941 int error;
3942 struct filename *name;
3943 struct dentry *dentry;
3944 struct path path;
3945 struct qstr last;
3946 int type;
3947 struct inode *inode = NULL;
3948 struct inode *delegated_inode = NULL;
3949 unsigned int lookup_flags = 0;
3950retry:
3951 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3952 &path, &last, &type);
3953 if (IS_ERR(name))
3954 return PTR_ERR(name);
3955
3956 error = -EISDIR;
3957 if (type != LAST_NORM)
3958 goto exit1;
3959
3960 error = mnt_want_write(path.mnt);
3961 if (error)
3962 goto exit1;
3963retry_deleg:
3964 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3965 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3966 error = PTR_ERR(dentry);
3967 if (!IS_ERR(dentry)) {
3968 /* Why not before? Because we want correct error value */
3969 if (last.name[last.len])
3970 goto slashes;
3971 inode = dentry->d_inode;
3972 if (d_is_negative(dentry))
3973 goto slashes;
3974 ihold(inode);
3975 error = security_path_unlink(&path, dentry);
3976 if (error)
3977 goto exit2;
3978 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3979exit2:
3980 dput(dentry);
3981 }
3982 inode_unlock(path.dentry->d_inode);
3983 if (inode)
3984 iput(inode); /* truncate the inode here */
3985 inode = NULL;
3986 if (delegated_inode) {
3987 error = break_deleg_wait(&delegated_inode);
3988 if (!error)
3989 goto retry_deleg;
3990 }
3991 mnt_drop_write(path.mnt);
3992exit1:
3993 path_put(&path);
3994 putname(name);
3995 if (retry_estale(error, lookup_flags)) {
3996 lookup_flags |= LOOKUP_REVAL;
3997 inode = NULL;
3998 goto retry;
3999 }
4000 return error;
4001
4002slashes:
4003 if (d_is_negative(dentry))
4004 error = -ENOENT;
4005 else if (d_is_dir(dentry))
4006 error = -EISDIR;
4007 else
4008 error = -ENOTDIR;
4009 goto exit2;
4010}
4011
4012SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4013{
4014 if ((flag & ~AT_REMOVEDIR) != 0)
4015 return -EINVAL;
4016
4017 if (flag & AT_REMOVEDIR)
4018 return do_rmdir(dfd, pathname);
4019
4020 return do_unlinkat(dfd, pathname);
4021}
4022
4023SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4024{
4025 return do_unlinkat(AT_FDCWD, pathname);
4026}
4027
4028int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4029{
4030 int error = may_create(dir, dentry);
4031
4032 if (error)
4033 return error;
4034
4035 if (!dir->i_op->symlink)
4036 return -EPERM;
4037
4038 error = security_inode_symlink(dir, dentry, oldname);
4039 if (error)
4040 return error;
4041
4042 error = dir->i_op->symlink(dir, dentry, oldname);
4043 if (!error)
4044 fsnotify_create(dir, dentry);
4045 return error;
4046}
4047EXPORT_SYMBOL(vfs_symlink);
4048
4049SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4050 int, newdfd, const char __user *, newname)
4051{
4052 int error;
4053 struct filename *from;
4054 struct dentry *dentry;
4055 struct path path;
4056 unsigned int lookup_flags = 0;
4057
4058 from = getname(oldname);
4059 if (IS_ERR(from))
4060 return PTR_ERR(from);
4061retry:
4062 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4063 error = PTR_ERR(dentry);
4064 if (IS_ERR(dentry))
4065 goto out_putname;
4066
4067 error = security_path_symlink(&path, dentry, from->name);
4068 if (!error)
4069 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4070 done_path_create(&path, dentry);
4071 if (retry_estale(error, lookup_flags)) {
4072 lookup_flags |= LOOKUP_REVAL;
4073 goto retry;
4074 }
4075out_putname:
4076 putname(from);
4077 return error;
4078}
4079
4080SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4081{
4082 return sys_symlinkat(oldname, AT_FDCWD, newname);
4083}
4084
4085/**
4086 * vfs_link - create a new link
4087 * @old_dentry: object to be linked
4088 * @dir: new parent
4089 * @new_dentry: where to create the new link
4090 * @delegated_inode: returns inode needing a delegation break
4091 *
4092 * The caller must hold dir->i_mutex
4093 *
4094 * If vfs_link discovers a delegation on the to-be-linked file in need
4095 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4096 * inode in delegated_inode. The caller should then break the delegation
4097 * and retry. Because breaking a delegation may take a long time, the
4098 * caller should drop the i_mutex before doing so.
4099 *
4100 * Alternatively, a caller may pass NULL for delegated_inode. This may
4101 * be appropriate for callers that expect the underlying filesystem not
4102 * to be NFS exported.
4103 */
4104int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4105{
4106 struct inode *inode = old_dentry->d_inode;
4107 unsigned max_links = dir->i_sb->s_max_links;
4108 int error;
4109
4110 if (!inode)
4111 return -ENOENT;
4112
4113 error = may_create(dir, new_dentry);
4114 if (error)
4115 return error;
4116
4117 if (dir->i_sb != inode->i_sb)
4118 return -EXDEV;
4119
4120 /*
4121 * A link to an append-only or immutable file cannot be created.
4122 */
4123 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4124 return -EPERM;
4125 /*
4126 * Updating the link count will likely cause i_uid and i_gid to
4127 * be writen back improperly if their true value is unknown to
4128 * the vfs.
4129 */
4130 if (HAS_UNMAPPED_ID(inode))
4131 return -EPERM;
4132 if (!dir->i_op->link)
4133 return -EPERM;
4134 if (S_ISDIR(inode->i_mode))
4135 return -EPERM;
4136
4137 error = security_inode_link(old_dentry, dir, new_dentry);
4138 if (error)
4139 return error;
4140
4141 inode_lock(inode);
4142 /* Make sure we don't allow creating hardlink to an unlinked file */
4143 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4144 error = -ENOENT;
4145 else if (max_links && inode->i_nlink >= max_links)
4146 error = -EMLINK;
4147 else {
4148 error = try_break_deleg(inode, delegated_inode);
4149 if (!error)
4150 error = dir->i_op->link(old_dentry, dir, new_dentry);
4151 }
4152
4153 if (!error && (inode->i_state & I_LINKABLE)) {
4154 spin_lock(&inode->i_lock);
4155 inode->i_state &= ~I_LINKABLE;
4156 spin_unlock(&inode->i_lock);
4157 }
4158 inode_unlock(inode);
4159 if (!error)
4160 fsnotify_link(dir, inode, new_dentry);
4161 return error;
4162}
4163EXPORT_SYMBOL(vfs_link);
4164
4165/*
4166 * Hardlinks are often used in delicate situations. We avoid
4167 * security-related surprises by not following symlinks on the
4168 * newname. --KAB
4169 *
4170 * We don't follow them on the oldname either to be compatible
4171 * with linux 2.0, and to avoid hard-linking to directories
4172 * and other special files. --ADM
4173 */
4174SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4175 int, newdfd, const char __user *, newname, int, flags)
4176{
4177 struct dentry *new_dentry;
4178 struct path old_path, new_path;
4179 struct inode *delegated_inode = NULL;
4180 int how = 0;
4181 int error;
4182
4183 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4184 return -EINVAL;
4185 /*
4186 * To use null names we require CAP_DAC_READ_SEARCH
4187 * This ensures that not everyone will be able to create
4188 * handlink using the passed filedescriptor.
4189 */
4190 if (flags & AT_EMPTY_PATH) {
4191 if (!capable(CAP_DAC_READ_SEARCH))
4192 return -ENOENT;
4193 how = LOOKUP_EMPTY;
4194 }
4195
4196 if (flags & AT_SYMLINK_FOLLOW)
4197 how |= LOOKUP_FOLLOW;
4198retry:
4199 error = user_path_at(olddfd, oldname, how, &old_path);
4200 if (error)
4201 return error;
4202
4203 new_dentry = user_path_create(newdfd, newname, &new_path,
4204 (how & LOOKUP_REVAL));
4205 error = PTR_ERR(new_dentry);
4206 if (IS_ERR(new_dentry))
4207 goto out;
4208
4209 error = -EXDEV;
4210 if (old_path.mnt != new_path.mnt)
4211 goto out_dput;
4212 error = may_linkat(&old_path);
4213 if (unlikely(error))
4214 goto out_dput;
4215 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4216 if (error)
4217 goto out_dput;
4218 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4219out_dput:
4220 done_path_create(&new_path, new_dentry);
4221 if (delegated_inode) {
4222 error = break_deleg_wait(&delegated_inode);
4223 if (!error) {
4224 path_put(&old_path);
4225 goto retry;
4226 }
4227 }
4228 if (retry_estale(error, how)) {
4229 path_put(&old_path);
4230 how |= LOOKUP_REVAL;
4231 goto retry;
4232 }
4233out:
4234 path_put(&old_path);
4235
4236 return error;
4237}
4238
4239SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4240{
4241 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4242}
4243
4244/**
4245 * vfs_rename - rename a filesystem object
4246 * @old_dir: parent of source
4247 * @old_dentry: source
4248 * @new_dir: parent of destination
4249 * @new_dentry: destination
4250 * @delegated_inode: returns an inode needing a delegation break
4251 * @flags: rename flags
4252 *
4253 * The caller must hold multiple mutexes--see lock_rename()).
4254 *
4255 * If vfs_rename discovers a delegation in need of breaking at either
4256 * the source or destination, it will return -EWOULDBLOCK and return a
4257 * reference to the inode in delegated_inode. The caller should then
4258 * break the delegation and retry. Because breaking a delegation may
4259 * take a long time, the caller should drop all locks before doing
4260 * so.
4261 *
4262 * Alternatively, a caller may pass NULL for delegated_inode. This may
4263 * be appropriate for callers that expect the underlying filesystem not
4264 * to be NFS exported.
4265 *
4266 * The worst of all namespace operations - renaming directory. "Perverted"
4267 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4268 * Problems:
4269 * a) we can get into loop creation.
4270 * b) race potential - two innocent renames can create a loop together.
4271 * That's where 4.4 screws up. Current fix: serialization on
4272 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4273 * story.
4274 * c) we have to lock _four_ objects - parents and victim (if it exists),
4275 * and source (if it is not a directory).
4276 * And that - after we got ->i_mutex on parents (until then we don't know
4277 * whether the target exists). Solution: try to be smart with locking
4278 * order for inodes. We rely on the fact that tree topology may change
4279 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4280 * move will be locked. Thus we can rank directories by the tree
4281 * (ancestors first) and rank all non-directories after them.
4282 * That works since everybody except rename does "lock parent, lookup,
4283 * lock child" and rename is under ->s_vfs_rename_mutex.
4284 * HOWEVER, it relies on the assumption that any object with ->lookup()
4285 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4286 * we'd better make sure that there's no link(2) for them.
4287 * d) conversion from fhandle to dentry may come in the wrong moment - when
4288 * we are removing the target. Solution: we will have to grab ->i_mutex
4289 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4290 * ->i_mutex on parents, which works but leads to some truly excessive
4291 * locking].
4292 */
4293int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4294 struct inode *new_dir, struct dentry *new_dentry,
4295 struct inode **delegated_inode, unsigned int flags)
4296{
4297 int error;
4298 bool is_dir = d_is_dir(old_dentry);
4299 const unsigned char *old_name;
4300 struct inode *source = old_dentry->d_inode;
4301 struct inode *target = new_dentry->d_inode;
4302 bool new_is_dir = false;
4303 unsigned max_links = new_dir->i_sb->s_max_links;
4304
4305 if (source == target)
4306 return 0;
4307
4308 error = may_delete(old_dir, old_dentry, is_dir);
4309 if (error)
4310 return error;
4311
4312 if (!target) {
4313 error = may_create(new_dir, new_dentry);
4314 } else {
4315 new_is_dir = d_is_dir(new_dentry);
4316
4317 if (!(flags & RENAME_EXCHANGE))
4318 error = may_delete(new_dir, new_dentry, is_dir);
4319 else
4320 error = may_delete(new_dir, new_dentry, new_is_dir);
4321 }
4322 if (error)
4323 return error;
4324
4325 if (!old_dir->i_op->rename)
4326 return -EPERM;
4327
4328 /*
4329 * If we are going to change the parent - check write permissions,
4330 * we'll need to flip '..'.
4331 */
4332 if (new_dir != old_dir) {
4333 if (is_dir) {
4334 error = inode_permission(source, MAY_WRITE);
4335 if (error)
4336 return error;
4337 }
4338 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4339 error = inode_permission(target, MAY_WRITE);
4340 if (error)
4341 return error;
4342 }
4343 }
4344
4345 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4346 flags);
4347 if (error)
4348 return error;
4349
4350 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4351 dget(new_dentry);
4352 if (!is_dir || (flags & RENAME_EXCHANGE))
4353 lock_two_nondirectories(source, target);
4354 else if (target)
4355 inode_lock(target);
4356
4357 error = -EBUSY;
4358 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4359 goto out;
4360
4361 if (max_links && new_dir != old_dir) {
4362 error = -EMLINK;
4363 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4364 goto out;
4365 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4366 old_dir->i_nlink >= max_links)
4367 goto out;
4368 }
4369 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4370 shrink_dcache_parent(new_dentry);
4371 if (!is_dir) {
4372 error = try_break_deleg(source, delegated_inode);
4373 if (error)
4374 goto out;
4375 }
4376 if (target && !new_is_dir) {
4377 error = try_break_deleg(target, delegated_inode);
4378 if (error)
4379 goto out;
4380 }
4381 error = old_dir->i_op->rename(old_dir, old_dentry,
4382 new_dir, new_dentry, flags);
4383 if (error)
4384 goto out;
4385
4386 if (!(flags & RENAME_EXCHANGE) && target) {
4387 if (is_dir)
4388 target->i_flags |= S_DEAD;
4389 dont_mount(new_dentry);
4390 detach_mounts(new_dentry);
4391 }
4392 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4393 if (!(flags & RENAME_EXCHANGE))
4394 d_move(old_dentry, new_dentry);
4395 else
4396 d_exchange(old_dentry, new_dentry);
4397 }
4398out:
4399 if (!is_dir || (flags & RENAME_EXCHANGE))
4400 unlock_two_nondirectories(source, target);
4401 else if (target)
4402 inode_unlock(target);
4403 dput(new_dentry);
4404 if (!error) {
4405 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4406 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4407 if (flags & RENAME_EXCHANGE) {
4408 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4409 new_is_dir, NULL, new_dentry);
4410 }
4411 }
4412 fsnotify_oldname_free(old_name);
4413
4414 return error;
4415}
4416EXPORT_SYMBOL(vfs_rename);
4417
4418SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4419 int, newdfd, const char __user *, newname, unsigned int, flags)
4420{
4421 struct dentry *old_dentry, *new_dentry;
4422 struct dentry *trap;
4423 struct path old_path, new_path;
4424 struct qstr old_last, new_last;
4425 int old_type, new_type;
4426 struct inode *delegated_inode = NULL;
4427 struct filename *from;
4428 struct filename *to;
4429 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4430 bool should_retry = false;
4431 int error;
4432
4433 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4434 return -EINVAL;
4435
4436 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4437 (flags & RENAME_EXCHANGE))
4438 return -EINVAL;
4439
4440 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4441 return -EPERM;
4442
4443 if (flags & RENAME_EXCHANGE)
4444 target_flags = 0;
4445
4446retry:
4447 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4448 &old_path, &old_last, &old_type);
4449 if (IS_ERR(from)) {
4450 error = PTR_ERR(from);
4451 goto exit;
4452 }
4453
4454 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4455 &new_path, &new_last, &new_type);
4456 if (IS_ERR(to)) {
4457 error = PTR_ERR(to);
4458 goto exit1;
4459 }
4460
4461 error = -EXDEV;
4462 if (old_path.mnt != new_path.mnt)
4463 goto exit2;
4464
4465 error = -EBUSY;
4466 if (old_type != LAST_NORM)
4467 goto exit2;
4468
4469 if (flags & RENAME_NOREPLACE)
4470 error = -EEXIST;
4471 if (new_type != LAST_NORM)
4472 goto exit2;
4473
4474 error = mnt_want_write(old_path.mnt);
4475 if (error)
4476 goto exit2;
4477
4478retry_deleg:
4479 trap = lock_rename(new_path.dentry, old_path.dentry);
4480
4481 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4482 error = PTR_ERR(old_dentry);
4483 if (IS_ERR(old_dentry))
4484 goto exit3;
4485 /* source must exist */
4486 error = -ENOENT;
4487 if (d_is_negative(old_dentry))
4488 goto exit4;
4489 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4490 error = PTR_ERR(new_dentry);
4491 if (IS_ERR(new_dentry))
4492 goto exit4;
4493 error = -EEXIST;
4494 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4495 goto exit5;
4496 if (flags & RENAME_EXCHANGE) {
4497 error = -ENOENT;
4498 if (d_is_negative(new_dentry))
4499 goto exit5;
4500
4501 if (!d_is_dir(new_dentry)) {
4502 error = -ENOTDIR;
4503 if (new_last.name[new_last.len])
4504 goto exit5;
4505 }
4506 }
4507 /* unless the source is a directory trailing slashes give -ENOTDIR */
4508 if (!d_is_dir(old_dentry)) {
4509 error = -ENOTDIR;
4510 if (old_last.name[old_last.len])
4511 goto exit5;
4512 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4513 goto exit5;
4514 }
4515 /* source should not be ancestor of target */
4516 error = -EINVAL;
4517 if (old_dentry == trap)
4518 goto exit5;
4519 /* target should not be an ancestor of source */
4520 if (!(flags & RENAME_EXCHANGE))
4521 error = -ENOTEMPTY;
4522 if (new_dentry == trap)
4523 goto exit5;
4524
4525 error = security_path_rename(&old_path, old_dentry,
4526 &new_path, new_dentry, flags);
4527 if (error)
4528 goto exit5;
4529 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4530 new_path.dentry->d_inode, new_dentry,
4531 &delegated_inode, flags);
4532exit5:
4533 dput(new_dentry);
4534exit4:
4535 dput(old_dentry);
4536exit3:
4537 unlock_rename(new_path.dentry, old_path.dentry);
4538 if (delegated_inode) {
4539 error = break_deleg_wait(&delegated_inode);
4540 if (!error)
4541 goto retry_deleg;
4542 }
4543 mnt_drop_write(old_path.mnt);
4544exit2:
4545 if (retry_estale(error, lookup_flags))
4546 should_retry = true;
4547 path_put(&new_path);
4548 putname(to);
4549exit1:
4550 path_put(&old_path);
4551 putname(from);
4552 if (should_retry) {
4553 should_retry = false;
4554 lookup_flags |= LOOKUP_REVAL;
4555 goto retry;
4556 }
4557exit:
4558 return error;
4559}
4560
4561SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4562 int, newdfd, const char __user *, newname)
4563{
4564 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4565}
4566
4567SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4568{
4569 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4570}
4571
4572int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4573{
4574 int error = may_create(dir, dentry);
4575 if (error)
4576 return error;
4577
4578 if (!dir->i_op->mknod)
4579 return -EPERM;
4580
4581 return dir->i_op->mknod(dir, dentry,
4582 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4583}
4584EXPORT_SYMBOL(vfs_whiteout);
4585
4586int readlink_copy(char __user *buffer, int buflen, const char *link)
4587{
4588 int len = PTR_ERR(link);
4589 if (IS_ERR(link))
4590 goto out;
4591
4592 len = strlen(link);
4593 if (len > (unsigned) buflen)
4594 len = buflen;
4595 if (copy_to_user(buffer, link, len))
4596 len = -EFAULT;
4597out:
4598 return len;
4599}
4600
4601/*
4602 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4603 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4604 * for any given inode is up to filesystem.
4605 */
4606static int generic_readlink(struct dentry *dentry, char __user *buffer,
4607 int buflen)
4608{
4609 DEFINE_DELAYED_CALL(done);
4610 struct inode *inode = d_inode(dentry);
4611 const char *link = inode->i_link;
4612 int res;
4613
4614 if (!link) {
4615 link = inode->i_op->get_link(dentry, inode, &done);
4616 if (IS_ERR(link))
4617 return PTR_ERR(link);
4618 }
4619 res = readlink_copy(buffer, buflen, link);
4620 do_delayed_call(&done);
4621 return res;
4622}
4623
4624/**
4625 * vfs_readlink - copy symlink body into userspace buffer
4626 * @dentry: dentry on which to get symbolic link
4627 * @buffer: user memory pointer
4628 * @buflen: size of buffer
4629 *
4630 * Does not touch atime. That's up to the caller if necessary
4631 *
4632 * Does not call security hook.
4633 */
4634int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4635{
4636 struct inode *inode = d_inode(dentry);
4637
4638 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4639 if (unlikely(inode->i_op->readlink))
4640 return inode->i_op->readlink(dentry, buffer, buflen);
4641
4642 if (!d_is_symlink(dentry))
4643 return -EINVAL;
4644
4645 spin_lock(&inode->i_lock);
4646 inode->i_opflags |= IOP_DEFAULT_READLINK;
4647 spin_unlock(&inode->i_lock);
4648 }
4649
4650 return generic_readlink(dentry, buffer, buflen);
4651}
4652EXPORT_SYMBOL(vfs_readlink);
4653
4654/**
4655 * vfs_get_link - get symlink body
4656 * @dentry: dentry on which to get symbolic link
4657 * @done: caller needs to free returned data with this
4658 *
4659 * Calls security hook and i_op->get_link() on the supplied inode.
4660 *
4661 * It does not touch atime. That's up to the caller if necessary.
4662 *
4663 * Does not work on "special" symlinks like /proc/$$/fd/N
4664 */
4665const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4666{
4667 const char *res = ERR_PTR(-EINVAL);
4668 struct inode *inode = d_inode(dentry);
4669
4670 if (d_is_symlink(dentry)) {
4671 res = ERR_PTR(security_inode_readlink(dentry));
4672 if (!res)
4673 res = inode->i_op->get_link(dentry, inode, done);
4674 }
4675 return res;
4676}
4677EXPORT_SYMBOL(vfs_get_link);
4678
4679/* get the link contents into pagecache */
4680const char *page_get_link(struct dentry *dentry, struct inode *inode,
4681 struct delayed_call *callback)
4682{
4683 char *kaddr;
4684 struct page *page;
4685 struct address_space *mapping = inode->i_mapping;
4686
4687 if (!dentry) {
4688 page = find_get_page(mapping, 0);
4689 if (!page)
4690 return ERR_PTR(-ECHILD);
4691 if (!PageUptodate(page)) {
4692 put_page(page);
4693 return ERR_PTR(-ECHILD);
4694 }
4695 } else {
4696 page = read_mapping_page(mapping, 0, NULL);
4697 if (IS_ERR(page))
4698 return (char*)page;
4699 }
4700 set_delayed_call(callback, page_put_link, page);
4701 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4702 kaddr = page_address(page);
4703 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4704 return kaddr;
4705}
4706
4707EXPORT_SYMBOL(page_get_link);
4708
4709void page_put_link(void *arg)
4710{
4711 put_page(arg);
4712}
4713EXPORT_SYMBOL(page_put_link);
4714
4715int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4716{
4717 DEFINE_DELAYED_CALL(done);
4718 int res = readlink_copy(buffer, buflen,
4719 page_get_link(dentry, d_inode(dentry),
4720 &done));
4721 do_delayed_call(&done);
4722 return res;
4723}
4724EXPORT_SYMBOL(page_readlink);
4725
4726/*
4727 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4728 */
4729int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4730{
4731 struct address_space *mapping = inode->i_mapping;
4732 struct page *page;
4733 void *fsdata;
4734 int err;
4735 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4736 if (nofs)
4737 flags |= AOP_FLAG_NOFS;
4738
4739retry:
4740 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4741 flags, &page, &fsdata);
4742 if (err)
4743 goto fail;
4744
4745 memcpy(page_address(page), symname, len-1);
4746
4747 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4748 page, fsdata);
4749 if (err < 0)
4750 goto fail;
4751 if (err < len-1)
4752 goto retry;
4753
4754 mark_inode_dirty(inode);
4755 return 0;
4756fail:
4757 return err;
4758}
4759EXPORT_SYMBOL(__page_symlink);
4760
4761int page_symlink(struct inode *inode, const char *symname, int len)
4762{
4763 return __page_symlink(inode, symname, len,
4764 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4765}
4766EXPORT_SYMBOL(page_symlink);
4767
4768const struct inode_operations page_symlink_inode_operations = {
4769 .get_link = page_get_link,
4770};
4771EXPORT_SYMBOL(page_symlink_inode_operations);