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// 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/sched/mm.h>
26#include <linux/fsnotify.h>
27#include <linux/personality.h>
28#include <linux/security.h>
29#include <linux/ima.h>
30#include <linux/syscalls.h>
31#include <linux/mount.h>
32#include <linux/audit.h>
33#include <linux/capability.h>
34#include <linux/file.h>
35#include <linux/fcntl.h>
36#include <linux/device_cgroup.h>
37#include <linux/fs_struct.h>
38#include <linux/posix_acl.h>
39#include <linux/hash.h>
40#include <linux/bitops.h>
41#include <linux/init_task.h>
42#include <linux/uaccess.h>
43
44#include "internal.h"
45#include "mount.h"
46
47/* [Feb-1997 T. Schoebel-Theuer]
48 * Fundamental changes in the pathname lookup mechanisms (namei)
49 * were necessary because of omirr. The reason is that omirr needs
50 * to know the _real_ pathname, not the user-supplied one, in case
51 * of symlinks (and also when transname replacements occur).
52 *
53 * The new code replaces the old recursive symlink resolution with
54 * an iterative one (in case of non-nested symlink chains). It does
55 * this with calls to <fs>_follow_link().
56 * As a side effect, dir_namei(), _namei() and follow_link() are now
57 * replaced with a single function lookup_dentry() that can handle all
58 * the special cases of the former code.
59 *
60 * With the new dcache, the pathname is stored at each inode, at least as
61 * long as the refcount of the inode is positive. As a side effect, the
62 * size of the dcache depends on the inode cache and thus is dynamic.
63 *
64 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
65 * resolution to correspond with current state of the code.
66 *
67 * Note that the symlink resolution is not *completely* iterative.
68 * There is still a significant amount of tail- and mid- recursion in
69 * the algorithm. Also, note that <fs>_readlink() is not used in
70 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
71 * may return different results than <fs>_follow_link(). Many virtual
72 * filesystems (including /proc) exhibit this behavior.
73 */
74
75/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
76 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
77 * and the name already exists in form of a symlink, try to create the new
78 * name indicated by the symlink. The old code always complained that the
79 * name already exists, due to not following the symlink even if its target
80 * is nonexistent. The new semantics affects also mknod() and link() when
81 * the name is a symlink pointing to a non-existent name.
82 *
83 * I don't know which semantics is the right one, since I have no access
84 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
85 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
86 * "old" one. Personally, I think the new semantics is much more logical.
87 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
88 * file does succeed in both HP-UX and SunOs, but not in Solaris
89 * and in the old Linux semantics.
90 */
91
92/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
93 * semantics. See the comments in "open_namei" and "do_link" below.
94 *
95 * [10-Sep-98 Alan Modra] Another symlink change.
96 */
97
98/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
99 * inside the path - always follow.
100 * in the last component in creation/removal/renaming - never follow.
101 * if LOOKUP_FOLLOW passed - follow.
102 * if the pathname has trailing slashes - follow.
103 * otherwise - don't follow.
104 * (applied in that order).
105 *
106 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
107 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
108 * During the 2.4 we need to fix the userland stuff depending on it -
109 * hopefully we will be able to get rid of that wart in 2.5. So far only
110 * XEmacs seems to be relying on it...
111 */
112/*
113 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
114 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
115 * any extra contention...
116 */
117
118/* In order to reduce some races, while at the same time doing additional
119 * checking and hopefully speeding things up, we copy filenames to the
120 * kernel data space before using them..
121 *
122 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
123 * PATH_MAX includes the nul terminator --RR.
124 */
125
126#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
127
128struct filename *
129getname_flags(const char __user *filename, int flags, int *empty)
130{
131 struct filename *result;
132 char *kname;
133 int len;
134
135 result = audit_reusename(filename);
136 if (result)
137 return result;
138
139 result = __getname();
140 if (unlikely(!result))
141 return ERR_PTR(-ENOMEM);
142
143 /*
144 * First, try to embed the struct filename inside the names_cache
145 * allocation
146 */
147 kname = (char *)result->iname;
148 result->name = kname;
149
150 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
151 if (unlikely(len < 0)) {
152 __putname(result);
153 return ERR_PTR(len);
154 }
155
156 /*
157 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
158 * separate struct filename so we can dedicate the entire
159 * names_cache allocation for the pathname, and re-do the copy from
160 * userland.
161 */
162 if (unlikely(len == EMBEDDED_NAME_MAX)) {
163 const size_t size = offsetof(struct filename, iname[1]);
164 kname = (char *)result;
165
166 /*
167 * size is chosen that way we to guarantee that
168 * result->iname[0] is within the same object and that
169 * kname can't be equal to result->iname, no matter what.
170 */
171 result = kzalloc(size, GFP_KERNEL);
172 if (unlikely(!result)) {
173 __putname(kname);
174 return ERR_PTR(-ENOMEM);
175 }
176 result->name = kname;
177 len = strncpy_from_user(kname, filename, PATH_MAX);
178 if (unlikely(len < 0)) {
179 __putname(kname);
180 kfree(result);
181 return ERR_PTR(len);
182 }
183 if (unlikely(len == PATH_MAX)) {
184 __putname(kname);
185 kfree(result);
186 return ERR_PTR(-ENAMETOOLONG);
187 }
188 }
189
190 result->refcnt = 1;
191 /* The empty path is special. */
192 if (unlikely(!len)) {
193 if (empty)
194 *empty = 1;
195 if (!(flags & LOOKUP_EMPTY)) {
196 putname(result);
197 return ERR_PTR(-ENOENT);
198 }
199 }
200
201 result->uptr = filename;
202 result->aname = NULL;
203 audit_getname(result);
204 return result;
205}
206
207struct filename *
208getname_uflags(const char __user *filename, int uflags)
209{
210 int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
211
212 return getname_flags(filename, flags, NULL);
213}
214
215struct filename *
216getname(const char __user * filename)
217{
218 return getname_flags(filename, 0, NULL);
219}
220
221struct filename *
222getname_kernel(const char * filename)
223{
224 struct filename *result;
225 int len = strlen(filename) + 1;
226
227 result = __getname();
228 if (unlikely(!result))
229 return ERR_PTR(-ENOMEM);
230
231 if (len <= EMBEDDED_NAME_MAX) {
232 result->name = (char *)result->iname;
233 } else if (len <= PATH_MAX) {
234 const size_t size = offsetof(struct filename, iname[1]);
235 struct filename *tmp;
236
237 tmp = kmalloc(size, GFP_KERNEL);
238 if (unlikely(!tmp)) {
239 __putname(result);
240 return ERR_PTR(-ENOMEM);
241 }
242 tmp->name = (char *)result;
243 result = tmp;
244 } else {
245 __putname(result);
246 return ERR_PTR(-ENAMETOOLONG);
247 }
248 memcpy((char *)result->name, filename, len);
249 result->uptr = NULL;
250 result->aname = NULL;
251 result->refcnt = 1;
252 audit_getname(result);
253
254 return result;
255}
256
257void putname(struct filename *name)
258{
259 if (IS_ERR(name))
260 return;
261
262 BUG_ON(name->refcnt <= 0);
263
264 if (--name->refcnt > 0)
265 return;
266
267 if (name->name != name->iname) {
268 __putname(name->name);
269 kfree(name);
270 } else
271 __putname(name);
272}
273
274/**
275 * check_acl - perform ACL permission checking
276 * @mnt_userns: user namespace of the mount the inode was found from
277 * @inode: inode to check permissions on
278 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
279 *
280 * This function performs the ACL permission checking. Since this function
281 * retrieve POSIX acls it needs to know whether it is called from a blocking or
282 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
283 *
284 * If the inode has been found through an idmapped mount the user namespace of
285 * the vfsmount must be passed through @mnt_userns. This function will then take
286 * care to map the inode according to @mnt_userns before checking permissions.
287 * On non-idmapped mounts or if permission checking is to be performed on the
288 * raw inode simply passs init_user_ns.
289 */
290static int check_acl(struct user_namespace *mnt_userns,
291 struct inode *inode, int mask)
292{
293#ifdef CONFIG_FS_POSIX_ACL
294 struct posix_acl *acl;
295
296 if (mask & MAY_NOT_BLOCK) {
297 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
298 if (!acl)
299 return -EAGAIN;
300 /* no ->get_inode_acl() calls in RCU mode... */
301 if (is_uncached_acl(acl))
302 return -ECHILD;
303 return posix_acl_permission(mnt_userns, inode, acl, mask);
304 }
305
306 acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
307 if (IS_ERR(acl))
308 return PTR_ERR(acl);
309 if (acl) {
310 int error = posix_acl_permission(mnt_userns, inode, acl, mask);
311 posix_acl_release(acl);
312 return error;
313 }
314#endif
315
316 return -EAGAIN;
317}
318
319/**
320 * acl_permission_check - perform basic UNIX permission checking
321 * @mnt_userns: user namespace of the mount the inode was found from
322 * @inode: inode to check permissions on
323 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
324 *
325 * This function performs the basic UNIX permission checking. Since this
326 * function may retrieve POSIX acls it needs to know whether it is called from a
327 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
328 *
329 * If the inode has been found through an idmapped mount the user namespace of
330 * the vfsmount must be passed through @mnt_userns. This function will then take
331 * care to map the inode according to @mnt_userns before checking permissions.
332 * On non-idmapped mounts or if permission checking is to be performed on the
333 * raw inode simply passs init_user_ns.
334 */
335static int acl_permission_check(struct user_namespace *mnt_userns,
336 struct inode *inode, int mask)
337{
338 unsigned int mode = inode->i_mode;
339 vfsuid_t vfsuid;
340
341 /* Are we the owner? If so, ACL's don't matter */
342 vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
343 if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) {
344 mask &= 7;
345 mode >>= 6;
346 return (mask & ~mode) ? -EACCES : 0;
347 }
348
349 /* Do we have ACL's? */
350 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
351 int error = check_acl(mnt_userns, inode, mask);
352 if (error != -EAGAIN)
353 return error;
354 }
355
356 /* Only RWX matters for group/other mode bits */
357 mask &= 7;
358
359 /*
360 * Are the group permissions different from
361 * the other permissions in the bits we care
362 * about? Need to check group ownership if so.
363 */
364 if (mask & (mode ^ (mode >> 3))) {
365 vfsgid_t vfsgid = i_gid_into_vfsgid(mnt_userns, inode);
366 if (vfsgid_in_group_p(vfsgid))
367 mode >>= 3;
368 }
369
370 /* Bits in 'mode' clear that we require? */
371 return (mask & ~mode) ? -EACCES : 0;
372}
373
374/**
375 * generic_permission - check for access rights on a Posix-like filesystem
376 * @mnt_userns: user namespace of the mount the inode was found from
377 * @inode: inode to check access rights for
378 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
379 * %MAY_NOT_BLOCK ...)
380 *
381 * Used to check for read/write/execute permissions on a file.
382 * We use "fsuid" for this, letting us set arbitrary permissions
383 * for filesystem access without changing the "normal" uids which
384 * are used for other things.
385 *
386 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
387 * request cannot be satisfied (eg. requires blocking or too much complexity).
388 * It would then be called again in ref-walk mode.
389 *
390 * If the inode has been found through an idmapped mount the user namespace of
391 * the vfsmount must be passed through @mnt_userns. This function will then take
392 * care to map the inode according to @mnt_userns before checking permissions.
393 * On non-idmapped mounts or if permission checking is to be performed on the
394 * raw inode simply passs init_user_ns.
395 */
396int generic_permission(struct user_namespace *mnt_userns, struct inode *inode,
397 int mask)
398{
399 int ret;
400
401 /*
402 * Do the basic permission checks.
403 */
404 ret = acl_permission_check(mnt_userns, inode, mask);
405 if (ret != -EACCES)
406 return ret;
407
408 if (S_ISDIR(inode->i_mode)) {
409 /* DACs are overridable for directories */
410 if (!(mask & MAY_WRITE))
411 if (capable_wrt_inode_uidgid(mnt_userns, inode,
412 CAP_DAC_READ_SEARCH))
413 return 0;
414 if (capable_wrt_inode_uidgid(mnt_userns, inode,
415 CAP_DAC_OVERRIDE))
416 return 0;
417 return -EACCES;
418 }
419
420 /*
421 * Searching includes executable on directories, else just read.
422 */
423 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
424 if (mask == MAY_READ)
425 if (capable_wrt_inode_uidgid(mnt_userns, inode,
426 CAP_DAC_READ_SEARCH))
427 return 0;
428 /*
429 * Read/write DACs are always overridable.
430 * Executable DACs are overridable when there is
431 * at least one exec bit set.
432 */
433 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
434 if (capable_wrt_inode_uidgid(mnt_userns, inode,
435 CAP_DAC_OVERRIDE))
436 return 0;
437
438 return -EACCES;
439}
440EXPORT_SYMBOL(generic_permission);
441
442/**
443 * do_inode_permission - UNIX permission checking
444 * @mnt_userns: user namespace of the mount the inode was found from
445 * @inode: inode to check permissions on
446 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
447 *
448 * We _really_ want to just do "generic_permission()" without
449 * even looking at the inode->i_op values. So we keep a cache
450 * flag in inode->i_opflags, that says "this has not special
451 * permission function, use the fast case".
452 */
453static inline int do_inode_permission(struct user_namespace *mnt_userns,
454 struct inode *inode, int mask)
455{
456 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
457 if (likely(inode->i_op->permission))
458 return inode->i_op->permission(mnt_userns, inode, mask);
459
460 /* This gets set once for the inode lifetime */
461 spin_lock(&inode->i_lock);
462 inode->i_opflags |= IOP_FASTPERM;
463 spin_unlock(&inode->i_lock);
464 }
465 return generic_permission(mnt_userns, inode, mask);
466}
467
468/**
469 * sb_permission - Check superblock-level permissions
470 * @sb: Superblock of inode to check permission on
471 * @inode: Inode to check permission on
472 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
473 *
474 * Separate out file-system wide checks from inode-specific permission checks.
475 */
476static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
477{
478 if (unlikely(mask & MAY_WRITE)) {
479 umode_t mode = inode->i_mode;
480
481 /* Nobody gets write access to a read-only fs. */
482 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
483 return -EROFS;
484 }
485 return 0;
486}
487
488/**
489 * inode_permission - Check for access rights to a given inode
490 * @mnt_userns: User namespace of the mount the inode was found from
491 * @inode: Inode to check permission on
492 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
493 *
494 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
495 * this, letting us set arbitrary permissions for filesystem access without
496 * changing the "normal" UIDs which are used for other things.
497 *
498 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
499 */
500int inode_permission(struct user_namespace *mnt_userns,
501 struct inode *inode, int mask)
502{
503 int retval;
504
505 retval = sb_permission(inode->i_sb, inode, mask);
506 if (retval)
507 return retval;
508
509 if (unlikely(mask & MAY_WRITE)) {
510 /*
511 * Nobody gets write access to an immutable file.
512 */
513 if (IS_IMMUTABLE(inode))
514 return -EPERM;
515
516 /*
517 * Updating mtime will likely cause i_uid and i_gid to be
518 * written back improperly if their true value is unknown
519 * to the vfs.
520 */
521 if (HAS_UNMAPPED_ID(mnt_userns, inode))
522 return -EACCES;
523 }
524
525 retval = do_inode_permission(mnt_userns, inode, mask);
526 if (retval)
527 return retval;
528
529 retval = devcgroup_inode_permission(inode, mask);
530 if (retval)
531 return retval;
532
533 return security_inode_permission(inode, mask);
534}
535EXPORT_SYMBOL(inode_permission);
536
537/**
538 * path_get - get a reference to a path
539 * @path: path to get the reference to
540 *
541 * Given a path increment the reference count to the dentry and the vfsmount.
542 */
543void path_get(const struct path *path)
544{
545 mntget(path->mnt);
546 dget(path->dentry);
547}
548EXPORT_SYMBOL(path_get);
549
550/**
551 * path_put - put a reference to a path
552 * @path: path to put the reference to
553 *
554 * Given a path decrement the reference count to the dentry and the vfsmount.
555 */
556void path_put(const struct path *path)
557{
558 dput(path->dentry);
559 mntput(path->mnt);
560}
561EXPORT_SYMBOL(path_put);
562
563#define EMBEDDED_LEVELS 2
564struct nameidata {
565 struct path path;
566 struct qstr last;
567 struct path root;
568 struct inode *inode; /* path.dentry.d_inode */
569 unsigned int flags, state;
570 unsigned seq, next_seq, m_seq, r_seq;
571 int last_type;
572 unsigned depth;
573 int total_link_count;
574 struct saved {
575 struct path link;
576 struct delayed_call done;
577 const char *name;
578 unsigned seq;
579 } *stack, internal[EMBEDDED_LEVELS];
580 struct filename *name;
581 struct nameidata *saved;
582 unsigned root_seq;
583 int dfd;
584 vfsuid_t dir_vfsuid;
585 umode_t dir_mode;
586} __randomize_layout;
587
588#define ND_ROOT_PRESET 1
589#define ND_ROOT_GRABBED 2
590#define ND_JUMPED 4
591
592static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
593{
594 struct nameidata *old = current->nameidata;
595 p->stack = p->internal;
596 p->depth = 0;
597 p->dfd = dfd;
598 p->name = name;
599 p->path.mnt = NULL;
600 p->path.dentry = NULL;
601 p->total_link_count = old ? old->total_link_count : 0;
602 p->saved = old;
603 current->nameidata = p;
604}
605
606static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
607 const struct path *root)
608{
609 __set_nameidata(p, dfd, name);
610 p->state = 0;
611 if (unlikely(root)) {
612 p->state = ND_ROOT_PRESET;
613 p->root = *root;
614 }
615}
616
617static void restore_nameidata(void)
618{
619 struct nameidata *now = current->nameidata, *old = now->saved;
620
621 current->nameidata = old;
622 if (old)
623 old->total_link_count = now->total_link_count;
624 if (now->stack != now->internal)
625 kfree(now->stack);
626}
627
628static bool nd_alloc_stack(struct nameidata *nd)
629{
630 struct saved *p;
631
632 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
633 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
634 if (unlikely(!p))
635 return false;
636 memcpy(p, nd->internal, sizeof(nd->internal));
637 nd->stack = p;
638 return true;
639}
640
641/**
642 * path_connected - Verify that a dentry is below mnt.mnt_root
643 *
644 * Rename can sometimes move a file or directory outside of a bind
645 * mount, path_connected allows those cases to be detected.
646 */
647static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
648{
649 struct super_block *sb = mnt->mnt_sb;
650
651 /* Bind mounts can have disconnected paths */
652 if (mnt->mnt_root == sb->s_root)
653 return true;
654
655 return is_subdir(dentry, mnt->mnt_root);
656}
657
658static void drop_links(struct nameidata *nd)
659{
660 int i = nd->depth;
661 while (i--) {
662 struct saved *last = nd->stack + i;
663 do_delayed_call(&last->done);
664 clear_delayed_call(&last->done);
665 }
666}
667
668static void leave_rcu(struct nameidata *nd)
669{
670 nd->flags &= ~LOOKUP_RCU;
671 nd->seq = nd->next_seq = 0;
672 rcu_read_unlock();
673}
674
675static void terminate_walk(struct nameidata *nd)
676{
677 drop_links(nd);
678 if (!(nd->flags & LOOKUP_RCU)) {
679 int i;
680 path_put(&nd->path);
681 for (i = 0; i < nd->depth; i++)
682 path_put(&nd->stack[i].link);
683 if (nd->state & ND_ROOT_GRABBED) {
684 path_put(&nd->root);
685 nd->state &= ~ND_ROOT_GRABBED;
686 }
687 } else {
688 leave_rcu(nd);
689 }
690 nd->depth = 0;
691 nd->path.mnt = NULL;
692 nd->path.dentry = NULL;
693}
694
695/* path_put is needed afterwards regardless of success or failure */
696static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
697{
698 int res = __legitimize_mnt(path->mnt, mseq);
699 if (unlikely(res)) {
700 if (res > 0)
701 path->mnt = NULL;
702 path->dentry = NULL;
703 return false;
704 }
705 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
706 path->dentry = NULL;
707 return false;
708 }
709 return !read_seqcount_retry(&path->dentry->d_seq, seq);
710}
711
712static inline bool legitimize_path(struct nameidata *nd,
713 struct path *path, unsigned seq)
714{
715 return __legitimize_path(path, seq, nd->m_seq);
716}
717
718static bool legitimize_links(struct nameidata *nd)
719{
720 int i;
721 if (unlikely(nd->flags & LOOKUP_CACHED)) {
722 drop_links(nd);
723 nd->depth = 0;
724 return false;
725 }
726 for (i = 0; i < nd->depth; i++) {
727 struct saved *last = nd->stack + i;
728 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
729 drop_links(nd);
730 nd->depth = i + 1;
731 return false;
732 }
733 }
734 return true;
735}
736
737static bool legitimize_root(struct nameidata *nd)
738{
739 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
740 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
741 return true;
742 nd->state |= ND_ROOT_GRABBED;
743 return legitimize_path(nd, &nd->root, nd->root_seq);
744}
745
746/*
747 * Path walking has 2 modes, rcu-walk and ref-walk (see
748 * Documentation/filesystems/path-lookup.txt). In situations when we can't
749 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
750 * normal reference counts on dentries and vfsmounts to transition to ref-walk
751 * mode. Refcounts are grabbed at the last known good point before rcu-walk
752 * got stuck, so ref-walk may continue from there. If this is not successful
753 * (eg. a seqcount has changed), then failure is returned and it's up to caller
754 * to restart the path walk from the beginning in ref-walk mode.
755 */
756
757/**
758 * try_to_unlazy - try to switch to ref-walk mode.
759 * @nd: nameidata pathwalk data
760 * Returns: true on success, false on failure
761 *
762 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
763 * for ref-walk mode.
764 * Must be called from rcu-walk context.
765 * Nothing should touch nameidata between try_to_unlazy() failure and
766 * terminate_walk().
767 */
768static bool try_to_unlazy(struct nameidata *nd)
769{
770 struct dentry *parent = nd->path.dentry;
771
772 BUG_ON(!(nd->flags & LOOKUP_RCU));
773
774 if (unlikely(!legitimize_links(nd)))
775 goto out1;
776 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
777 goto out;
778 if (unlikely(!legitimize_root(nd)))
779 goto out;
780 leave_rcu(nd);
781 BUG_ON(nd->inode != parent->d_inode);
782 return true;
783
784out1:
785 nd->path.mnt = NULL;
786 nd->path.dentry = NULL;
787out:
788 leave_rcu(nd);
789 return false;
790}
791
792/**
793 * try_to_unlazy_next - try to switch to ref-walk mode.
794 * @nd: nameidata pathwalk data
795 * @dentry: next dentry to step into
796 * Returns: true on success, false on failure
797 *
798 * Similar to try_to_unlazy(), but here we have the next dentry already
799 * picked by rcu-walk and want to legitimize that in addition to the current
800 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
801 * Nothing should touch nameidata between try_to_unlazy_next() failure and
802 * terminate_walk().
803 */
804static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry)
805{
806 int res;
807 BUG_ON(!(nd->flags & LOOKUP_RCU));
808
809 if (unlikely(!legitimize_links(nd)))
810 goto out2;
811 res = __legitimize_mnt(nd->path.mnt, nd->m_seq);
812 if (unlikely(res)) {
813 if (res > 0)
814 goto out2;
815 goto out1;
816 }
817 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
818 goto out1;
819
820 /*
821 * We need to move both the parent and the dentry from the RCU domain
822 * to be properly refcounted. And the sequence number in the dentry
823 * validates *both* dentry counters, since we checked the sequence
824 * number of the parent after we got the child sequence number. So we
825 * know the parent must still be valid if the child sequence number is
826 */
827 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
828 goto out;
829 if (read_seqcount_retry(&dentry->d_seq, nd->next_seq))
830 goto out_dput;
831 /*
832 * Sequence counts matched. Now make sure that the root is
833 * still valid and get it if required.
834 */
835 if (unlikely(!legitimize_root(nd)))
836 goto out_dput;
837 leave_rcu(nd);
838 return true;
839
840out2:
841 nd->path.mnt = NULL;
842out1:
843 nd->path.dentry = NULL;
844out:
845 leave_rcu(nd);
846 return false;
847out_dput:
848 leave_rcu(nd);
849 dput(dentry);
850 return false;
851}
852
853static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
854{
855 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
856 return dentry->d_op->d_revalidate(dentry, flags);
857 else
858 return 1;
859}
860
861/**
862 * complete_walk - successful completion of path walk
863 * @nd: pointer nameidata
864 *
865 * If we had been in RCU mode, drop out of it and legitimize nd->path.
866 * Revalidate the final result, unless we'd already done that during
867 * the path walk or the filesystem doesn't ask for it. Return 0 on
868 * success, -error on failure. In case of failure caller does not
869 * need to drop nd->path.
870 */
871static int complete_walk(struct nameidata *nd)
872{
873 struct dentry *dentry = nd->path.dentry;
874 int status;
875
876 if (nd->flags & LOOKUP_RCU) {
877 /*
878 * We don't want to zero nd->root for scoped-lookups or
879 * externally-managed nd->root.
880 */
881 if (!(nd->state & ND_ROOT_PRESET))
882 if (!(nd->flags & LOOKUP_IS_SCOPED))
883 nd->root.mnt = NULL;
884 nd->flags &= ~LOOKUP_CACHED;
885 if (!try_to_unlazy(nd))
886 return -ECHILD;
887 }
888
889 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
890 /*
891 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
892 * ever step outside the root during lookup" and should already
893 * be guaranteed by the rest of namei, we want to avoid a namei
894 * BUG resulting in userspace being given a path that was not
895 * scoped within the root at some point during the lookup.
896 *
897 * So, do a final sanity-check to make sure that in the
898 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
899 * we won't silently return an fd completely outside of the
900 * requested root to userspace.
901 *
902 * Userspace could move the path outside the root after this
903 * check, but as discussed elsewhere this is not a concern (the
904 * resolved file was inside the root at some point).
905 */
906 if (!path_is_under(&nd->path, &nd->root))
907 return -EXDEV;
908 }
909
910 if (likely(!(nd->state & ND_JUMPED)))
911 return 0;
912
913 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
914 return 0;
915
916 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
917 if (status > 0)
918 return 0;
919
920 if (!status)
921 status = -ESTALE;
922
923 return status;
924}
925
926static int set_root(struct nameidata *nd)
927{
928 struct fs_struct *fs = current->fs;
929
930 /*
931 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
932 * still have to ensure it doesn't happen because it will cause a breakout
933 * from the dirfd.
934 */
935 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
936 return -ENOTRECOVERABLE;
937
938 if (nd->flags & LOOKUP_RCU) {
939 unsigned seq;
940
941 do {
942 seq = read_seqcount_begin(&fs->seq);
943 nd->root = fs->root;
944 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
945 } while (read_seqcount_retry(&fs->seq, seq));
946 } else {
947 get_fs_root(fs, &nd->root);
948 nd->state |= ND_ROOT_GRABBED;
949 }
950 return 0;
951}
952
953static int nd_jump_root(struct nameidata *nd)
954{
955 if (unlikely(nd->flags & LOOKUP_BENEATH))
956 return -EXDEV;
957 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
958 /* Absolute path arguments to path_init() are allowed. */
959 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
960 return -EXDEV;
961 }
962 if (!nd->root.mnt) {
963 int error = set_root(nd);
964 if (error)
965 return error;
966 }
967 if (nd->flags & LOOKUP_RCU) {
968 struct dentry *d;
969 nd->path = nd->root;
970 d = nd->path.dentry;
971 nd->inode = d->d_inode;
972 nd->seq = nd->root_seq;
973 if (read_seqcount_retry(&d->d_seq, nd->seq))
974 return -ECHILD;
975 } else {
976 path_put(&nd->path);
977 nd->path = nd->root;
978 path_get(&nd->path);
979 nd->inode = nd->path.dentry->d_inode;
980 }
981 nd->state |= ND_JUMPED;
982 return 0;
983}
984
985/*
986 * Helper to directly jump to a known parsed path from ->get_link,
987 * caller must have taken a reference to path beforehand.
988 */
989int nd_jump_link(const struct path *path)
990{
991 int error = -ELOOP;
992 struct nameidata *nd = current->nameidata;
993
994 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
995 goto err;
996
997 error = -EXDEV;
998 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
999 if (nd->path.mnt != path->mnt)
1000 goto err;
1001 }
1002 /* Not currently safe for scoped-lookups. */
1003 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1004 goto err;
1005
1006 path_put(&nd->path);
1007 nd->path = *path;
1008 nd->inode = nd->path.dentry->d_inode;
1009 nd->state |= ND_JUMPED;
1010 return 0;
1011
1012err:
1013 path_put(path);
1014 return error;
1015}
1016
1017static inline void put_link(struct nameidata *nd)
1018{
1019 struct saved *last = nd->stack + --nd->depth;
1020 do_delayed_call(&last->done);
1021 if (!(nd->flags & LOOKUP_RCU))
1022 path_put(&last->link);
1023}
1024
1025static int sysctl_protected_symlinks __read_mostly;
1026static int sysctl_protected_hardlinks __read_mostly;
1027static int sysctl_protected_fifos __read_mostly;
1028static int sysctl_protected_regular __read_mostly;
1029
1030#ifdef CONFIG_SYSCTL
1031static struct ctl_table namei_sysctls[] = {
1032 {
1033 .procname = "protected_symlinks",
1034 .data = &sysctl_protected_symlinks,
1035 .maxlen = sizeof(int),
1036 .mode = 0644,
1037 .proc_handler = proc_dointvec_minmax,
1038 .extra1 = SYSCTL_ZERO,
1039 .extra2 = SYSCTL_ONE,
1040 },
1041 {
1042 .procname = "protected_hardlinks",
1043 .data = &sysctl_protected_hardlinks,
1044 .maxlen = sizeof(int),
1045 .mode = 0644,
1046 .proc_handler = proc_dointvec_minmax,
1047 .extra1 = SYSCTL_ZERO,
1048 .extra2 = SYSCTL_ONE,
1049 },
1050 {
1051 .procname = "protected_fifos",
1052 .data = &sysctl_protected_fifos,
1053 .maxlen = sizeof(int),
1054 .mode = 0644,
1055 .proc_handler = proc_dointvec_minmax,
1056 .extra1 = SYSCTL_ZERO,
1057 .extra2 = SYSCTL_TWO,
1058 },
1059 {
1060 .procname = "protected_regular",
1061 .data = &sysctl_protected_regular,
1062 .maxlen = sizeof(int),
1063 .mode = 0644,
1064 .proc_handler = proc_dointvec_minmax,
1065 .extra1 = SYSCTL_ZERO,
1066 .extra2 = SYSCTL_TWO,
1067 },
1068 { }
1069};
1070
1071static int __init init_fs_namei_sysctls(void)
1072{
1073 register_sysctl_init("fs", namei_sysctls);
1074 return 0;
1075}
1076fs_initcall(init_fs_namei_sysctls);
1077
1078#endif /* CONFIG_SYSCTL */
1079
1080/**
1081 * may_follow_link - Check symlink following for unsafe situations
1082 * @nd: nameidata pathwalk data
1083 *
1084 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1085 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1086 * in a sticky world-writable directory. This is to protect privileged
1087 * processes from failing races against path names that may change out
1088 * from under them by way of other users creating malicious symlinks.
1089 * It will permit symlinks to be followed only when outside a sticky
1090 * world-writable directory, or when the uid of the symlink and follower
1091 * match, or when the directory owner matches the symlink's owner.
1092 *
1093 * Returns 0 if following the symlink is allowed, -ve on error.
1094 */
1095static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1096{
1097 struct user_namespace *mnt_userns;
1098 vfsuid_t vfsuid;
1099
1100 if (!sysctl_protected_symlinks)
1101 return 0;
1102
1103 mnt_userns = mnt_user_ns(nd->path.mnt);
1104 vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
1105 /* Allowed if owner and follower match. */
1106 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1107 return 0;
1108
1109 /* Allowed if parent directory not sticky and world-writable. */
1110 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1111 return 0;
1112
1113 /* Allowed if parent directory and link owner match. */
1114 if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid))
1115 return 0;
1116
1117 if (nd->flags & LOOKUP_RCU)
1118 return -ECHILD;
1119
1120 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1121 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1122 return -EACCES;
1123}
1124
1125/**
1126 * safe_hardlink_source - Check for safe hardlink conditions
1127 * @mnt_userns: user namespace of the mount the inode was found from
1128 * @inode: the source inode to hardlink from
1129 *
1130 * Return false if at least one of the following conditions:
1131 * - inode is not a regular file
1132 * - inode is setuid
1133 * - inode is setgid and group-exec
1134 * - access failure for read and write
1135 *
1136 * Otherwise returns true.
1137 */
1138static bool safe_hardlink_source(struct user_namespace *mnt_userns,
1139 struct inode *inode)
1140{
1141 umode_t mode = inode->i_mode;
1142
1143 /* Special files should not get pinned to the filesystem. */
1144 if (!S_ISREG(mode))
1145 return false;
1146
1147 /* Setuid files should not get pinned to the filesystem. */
1148 if (mode & S_ISUID)
1149 return false;
1150
1151 /* Executable setgid files should not get pinned to the filesystem. */
1152 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1153 return false;
1154
1155 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1156 if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE))
1157 return false;
1158
1159 return true;
1160}
1161
1162/**
1163 * may_linkat - Check permissions for creating a hardlink
1164 * @mnt_userns: user namespace of the mount the inode was found from
1165 * @link: the source to hardlink from
1166 *
1167 * Block hardlink when all of:
1168 * - sysctl_protected_hardlinks enabled
1169 * - fsuid does not match inode
1170 * - hardlink source is unsafe (see safe_hardlink_source() above)
1171 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1172 *
1173 * If the inode has been found through an idmapped mount the user namespace of
1174 * the vfsmount must be passed through @mnt_userns. This function will then take
1175 * care to map the inode according to @mnt_userns before checking permissions.
1176 * On non-idmapped mounts or if permission checking is to be performed on the
1177 * raw inode simply passs init_user_ns.
1178 *
1179 * Returns 0 if successful, -ve on error.
1180 */
1181int may_linkat(struct user_namespace *mnt_userns, const struct path *link)
1182{
1183 struct inode *inode = link->dentry->d_inode;
1184
1185 /* Inode writeback is not safe when the uid or gid are invalid. */
1186 if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) ||
1187 !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode)))
1188 return -EOVERFLOW;
1189
1190 if (!sysctl_protected_hardlinks)
1191 return 0;
1192
1193 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1194 * otherwise, it must be a safe source.
1195 */
1196 if (safe_hardlink_source(mnt_userns, inode) ||
1197 inode_owner_or_capable(mnt_userns, inode))
1198 return 0;
1199
1200 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1201 return -EPERM;
1202}
1203
1204/**
1205 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1206 * should be allowed, or not, on files that already
1207 * exist.
1208 * @mnt_userns: user namespace of the mount the inode was found from
1209 * @nd: nameidata pathwalk data
1210 * @inode: the inode of the file to open
1211 *
1212 * Block an O_CREAT open of a FIFO (or a regular file) when:
1213 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1214 * - the file already exists
1215 * - we are in a sticky directory
1216 * - we don't own the file
1217 * - the owner of the directory doesn't own the file
1218 * - the directory is world writable
1219 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1220 * the directory doesn't have to be world writable: being group writable will
1221 * be enough.
1222 *
1223 * If the inode has been found through an idmapped mount the user namespace of
1224 * the vfsmount must be passed through @mnt_userns. This function will then take
1225 * care to map the inode according to @mnt_userns before checking permissions.
1226 * On non-idmapped mounts or if permission checking is to be performed on the
1227 * raw inode simply passs init_user_ns.
1228 *
1229 * Returns 0 if the open is allowed, -ve on error.
1230 */
1231static int may_create_in_sticky(struct user_namespace *mnt_userns,
1232 struct nameidata *nd, struct inode *const inode)
1233{
1234 umode_t dir_mode = nd->dir_mode;
1235 vfsuid_t dir_vfsuid = nd->dir_vfsuid;
1236
1237 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1238 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1239 likely(!(dir_mode & S_ISVTX)) ||
1240 vfsuid_eq(i_uid_into_vfsuid(mnt_userns, inode), dir_vfsuid) ||
1241 vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), current_fsuid()))
1242 return 0;
1243
1244 if (likely(dir_mode & 0002) ||
1245 (dir_mode & 0020 &&
1246 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1247 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1248 const char *operation = S_ISFIFO(inode->i_mode) ?
1249 "sticky_create_fifo" :
1250 "sticky_create_regular";
1251 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1252 return -EACCES;
1253 }
1254 return 0;
1255}
1256
1257/*
1258 * follow_up - Find the mountpoint of path's vfsmount
1259 *
1260 * Given a path, find the mountpoint of its source file system.
1261 * Replace @path with the path of the mountpoint in the parent mount.
1262 * Up is towards /.
1263 *
1264 * Return 1 if we went up a level and 0 if we were already at the
1265 * root.
1266 */
1267int follow_up(struct path *path)
1268{
1269 struct mount *mnt = real_mount(path->mnt);
1270 struct mount *parent;
1271 struct dentry *mountpoint;
1272
1273 read_seqlock_excl(&mount_lock);
1274 parent = mnt->mnt_parent;
1275 if (parent == mnt) {
1276 read_sequnlock_excl(&mount_lock);
1277 return 0;
1278 }
1279 mntget(&parent->mnt);
1280 mountpoint = dget(mnt->mnt_mountpoint);
1281 read_sequnlock_excl(&mount_lock);
1282 dput(path->dentry);
1283 path->dentry = mountpoint;
1284 mntput(path->mnt);
1285 path->mnt = &parent->mnt;
1286 return 1;
1287}
1288EXPORT_SYMBOL(follow_up);
1289
1290static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1291 struct path *path, unsigned *seqp)
1292{
1293 while (mnt_has_parent(m)) {
1294 struct dentry *mountpoint = m->mnt_mountpoint;
1295
1296 m = m->mnt_parent;
1297 if (unlikely(root->dentry == mountpoint &&
1298 root->mnt == &m->mnt))
1299 break;
1300 if (mountpoint != m->mnt.mnt_root) {
1301 path->mnt = &m->mnt;
1302 path->dentry = mountpoint;
1303 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1304 return true;
1305 }
1306 }
1307 return false;
1308}
1309
1310static bool choose_mountpoint(struct mount *m, const struct path *root,
1311 struct path *path)
1312{
1313 bool found;
1314
1315 rcu_read_lock();
1316 while (1) {
1317 unsigned seq, mseq = read_seqbegin(&mount_lock);
1318
1319 found = choose_mountpoint_rcu(m, root, path, &seq);
1320 if (unlikely(!found)) {
1321 if (!read_seqretry(&mount_lock, mseq))
1322 break;
1323 } else {
1324 if (likely(__legitimize_path(path, seq, mseq)))
1325 break;
1326 rcu_read_unlock();
1327 path_put(path);
1328 rcu_read_lock();
1329 }
1330 }
1331 rcu_read_unlock();
1332 return found;
1333}
1334
1335/*
1336 * Perform an automount
1337 * - return -EISDIR to tell follow_managed() to stop and return the path we
1338 * were called with.
1339 */
1340static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1341{
1342 struct dentry *dentry = path->dentry;
1343
1344 /* We don't want to mount if someone's just doing a stat -
1345 * unless they're stat'ing a directory and appended a '/' to
1346 * the name.
1347 *
1348 * We do, however, want to mount if someone wants to open or
1349 * create a file of any type under the mountpoint, wants to
1350 * traverse through the mountpoint or wants to open the
1351 * mounted directory. Also, autofs may mark negative dentries
1352 * as being automount points. These will need the attentions
1353 * of the daemon to instantiate them before they can be used.
1354 */
1355 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1356 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1357 dentry->d_inode)
1358 return -EISDIR;
1359
1360 if (count && (*count)++ >= MAXSYMLINKS)
1361 return -ELOOP;
1362
1363 return finish_automount(dentry->d_op->d_automount(path), path);
1364}
1365
1366/*
1367 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1368 * dentries are pinned but not locked here, so negative dentry can go
1369 * positive right under us. Use of smp_load_acquire() provides a barrier
1370 * sufficient for ->d_inode and ->d_flags consistency.
1371 */
1372static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1373 int *count, unsigned lookup_flags)
1374{
1375 struct vfsmount *mnt = path->mnt;
1376 bool need_mntput = false;
1377 int ret = 0;
1378
1379 while (flags & DCACHE_MANAGED_DENTRY) {
1380 /* Allow the filesystem to manage the transit without i_mutex
1381 * being held. */
1382 if (flags & DCACHE_MANAGE_TRANSIT) {
1383 ret = path->dentry->d_op->d_manage(path, false);
1384 flags = smp_load_acquire(&path->dentry->d_flags);
1385 if (ret < 0)
1386 break;
1387 }
1388
1389 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1390 struct vfsmount *mounted = lookup_mnt(path);
1391 if (mounted) { // ... in our namespace
1392 dput(path->dentry);
1393 if (need_mntput)
1394 mntput(path->mnt);
1395 path->mnt = mounted;
1396 path->dentry = dget(mounted->mnt_root);
1397 // here we know it's positive
1398 flags = path->dentry->d_flags;
1399 need_mntput = true;
1400 continue;
1401 }
1402 }
1403
1404 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1405 break;
1406
1407 // uncovered automount point
1408 ret = follow_automount(path, count, lookup_flags);
1409 flags = smp_load_acquire(&path->dentry->d_flags);
1410 if (ret < 0)
1411 break;
1412 }
1413
1414 if (ret == -EISDIR)
1415 ret = 0;
1416 // possible if you race with several mount --move
1417 if (need_mntput && path->mnt == mnt)
1418 mntput(path->mnt);
1419 if (!ret && unlikely(d_flags_negative(flags)))
1420 ret = -ENOENT;
1421 *jumped = need_mntput;
1422 return ret;
1423}
1424
1425static inline int traverse_mounts(struct path *path, bool *jumped,
1426 int *count, unsigned lookup_flags)
1427{
1428 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1429
1430 /* fastpath */
1431 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1432 *jumped = false;
1433 if (unlikely(d_flags_negative(flags)))
1434 return -ENOENT;
1435 return 0;
1436 }
1437 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1438}
1439
1440int follow_down_one(struct path *path)
1441{
1442 struct vfsmount *mounted;
1443
1444 mounted = lookup_mnt(path);
1445 if (mounted) {
1446 dput(path->dentry);
1447 mntput(path->mnt);
1448 path->mnt = mounted;
1449 path->dentry = dget(mounted->mnt_root);
1450 return 1;
1451 }
1452 return 0;
1453}
1454EXPORT_SYMBOL(follow_down_one);
1455
1456/*
1457 * Follow down to the covering mount currently visible to userspace. At each
1458 * point, the filesystem owning that dentry may be queried as to whether the
1459 * caller is permitted to proceed or not.
1460 */
1461int follow_down(struct path *path)
1462{
1463 struct vfsmount *mnt = path->mnt;
1464 bool jumped;
1465 int ret = traverse_mounts(path, &jumped, NULL, 0);
1466
1467 if (path->mnt != mnt)
1468 mntput(mnt);
1469 return ret;
1470}
1471EXPORT_SYMBOL(follow_down);
1472
1473/*
1474 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1475 * we meet a managed dentry that would need blocking.
1476 */
1477static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1478{
1479 struct dentry *dentry = path->dentry;
1480 unsigned int flags = dentry->d_flags;
1481
1482 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1483 return true;
1484
1485 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1486 return false;
1487
1488 for (;;) {
1489 /*
1490 * Don't forget we might have a non-mountpoint managed dentry
1491 * that wants to block transit.
1492 */
1493 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1494 int res = dentry->d_op->d_manage(path, true);
1495 if (res)
1496 return res == -EISDIR;
1497 flags = dentry->d_flags;
1498 }
1499
1500 if (flags & DCACHE_MOUNTED) {
1501 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1502 if (mounted) {
1503 path->mnt = &mounted->mnt;
1504 dentry = path->dentry = mounted->mnt.mnt_root;
1505 nd->state |= ND_JUMPED;
1506 nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1507 flags = dentry->d_flags;
1508 // makes sure that non-RCU pathwalk could reach
1509 // this state.
1510 if (read_seqretry(&mount_lock, nd->m_seq))
1511 return false;
1512 continue;
1513 }
1514 if (read_seqretry(&mount_lock, nd->m_seq))
1515 return false;
1516 }
1517 return !(flags & DCACHE_NEED_AUTOMOUNT);
1518 }
1519}
1520
1521static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1522 struct path *path)
1523{
1524 bool jumped;
1525 int ret;
1526
1527 path->mnt = nd->path.mnt;
1528 path->dentry = dentry;
1529 if (nd->flags & LOOKUP_RCU) {
1530 unsigned int seq = nd->next_seq;
1531 if (likely(__follow_mount_rcu(nd, path)))
1532 return 0;
1533 // *path and nd->next_seq might've been clobbered
1534 path->mnt = nd->path.mnt;
1535 path->dentry = dentry;
1536 nd->next_seq = seq;
1537 if (!try_to_unlazy_next(nd, dentry))
1538 return -ECHILD;
1539 }
1540 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1541 if (jumped) {
1542 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1543 ret = -EXDEV;
1544 else
1545 nd->state |= ND_JUMPED;
1546 }
1547 if (unlikely(ret)) {
1548 dput(path->dentry);
1549 if (path->mnt != nd->path.mnt)
1550 mntput(path->mnt);
1551 }
1552 return ret;
1553}
1554
1555/*
1556 * This looks up the name in dcache and possibly revalidates the found dentry.
1557 * NULL is returned if the dentry does not exist in the cache.
1558 */
1559static struct dentry *lookup_dcache(const struct qstr *name,
1560 struct dentry *dir,
1561 unsigned int flags)
1562{
1563 struct dentry *dentry = d_lookup(dir, name);
1564 if (dentry) {
1565 int error = d_revalidate(dentry, flags);
1566 if (unlikely(error <= 0)) {
1567 if (!error)
1568 d_invalidate(dentry);
1569 dput(dentry);
1570 return ERR_PTR(error);
1571 }
1572 }
1573 return dentry;
1574}
1575
1576/*
1577 * Parent directory has inode locked exclusive. This is one
1578 * and only case when ->lookup() gets called on non in-lookup
1579 * dentries - as the matter of fact, this only gets called
1580 * when directory is guaranteed to have no in-lookup children
1581 * at all.
1582 */
1583static struct dentry *__lookup_hash(const struct qstr *name,
1584 struct dentry *base, unsigned int flags)
1585{
1586 struct dentry *dentry = lookup_dcache(name, base, flags);
1587 struct dentry *old;
1588 struct inode *dir = base->d_inode;
1589
1590 if (dentry)
1591 return dentry;
1592
1593 /* Don't create child dentry for a dead directory. */
1594 if (unlikely(IS_DEADDIR(dir)))
1595 return ERR_PTR(-ENOENT);
1596
1597 dentry = d_alloc(base, name);
1598 if (unlikely(!dentry))
1599 return ERR_PTR(-ENOMEM);
1600
1601 old = dir->i_op->lookup(dir, dentry, flags);
1602 if (unlikely(old)) {
1603 dput(dentry);
1604 dentry = old;
1605 }
1606 return dentry;
1607}
1608
1609static struct dentry *lookup_fast(struct nameidata *nd)
1610{
1611 struct dentry *dentry, *parent = nd->path.dentry;
1612 int status = 1;
1613
1614 /*
1615 * Rename seqlock is not required here because in the off chance
1616 * of a false negative due to a concurrent rename, the caller is
1617 * going to fall back to non-racy lookup.
1618 */
1619 if (nd->flags & LOOKUP_RCU) {
1620 dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq);
1621 if (unlikely(!dentry)) {
1622 if (!try_to_unlazy(nd))
1623 return ERR_PTR(-ECHILD);
1624 return NULL;
1625 }
1626
1627 /*
1628 * This sequence count validates that the parent had no
1629 * changes while we did the lookup of the dentry above.
1630 */
1631 if (read_seqcount_retry(&parent->d_seq, nd->seq))
1632 return ERR_PTR(-ECHILD);
1633
1634 status = d_revalidate(dentry, nd->flags);
1635 if (likely(status > 0))
1636 return dentry;
1637 if (!try_to_unlazy_next(nd, dentry))
1638 return ERR_PTR(-ECHILD);
1639 if (status == -ECHILD)
1640 /* we'd been told to redo it in non-rcu mode */
1641 status = d_revalidate(dentry, nd->flags);
1642 } else {
1643 dentry = __d_lookup(parent, &nd->last);
1644 if (unlikely(!dentry))
1645 return NULL;
1646 status = d_revalidate(dentry, nd->flags);
1647 }
1648 if (unlikely(status <= 0)) {
1649 if (!status)
1650 d_invalidate(dentry);
1651 dput(dentry);
1652 return ERR_PTR(status);
1653 }
1654 return dentry;
1655}
1656
1657/* Fast lookup failed, do it the slow way */
1658static struct dentry *__lookup_slow(const struct qstr *name,
1659 struct dentry *dir,
1660 unsigned int flags)
1661{
1662 struct dentry *dentry, *old;
1663 struct inode *inode = dir->d_inode;
1664 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1665
1666 /* Don't go there if it's already dead */
1667 if (unlikely(IS_DEADDIR(inode)))
1668 return ERR_PTR(-ENOENT);
1669again:
1670 dentry = d_alloc_parallel(dir, name, &wq);
1671 if (IS_ERR(dentry))
1672 return dentry;
1673 if (unlikely(!d_in_lookup(dentry))) {
1674 int error = d_revalidate(dentry, flags);
1675 if (unlikely(error <= 0)) {
1676 if (!error) {
1677 d_invalidate(dentry);
1678 dput(dentry);
1679 goto again;
1680 }
1681 dput(dentry);
1682 dentry = ERR_PTR(error);
1683 }
1684 } else {
1685 old = inode->i_op->lookup(inode, dentry, flags);
1686 d_lookup_done(dentry);
1687 if (unlikely(old)) {
1688 dput(dentry);
1689 dentry = old;
1690 }
1691 }
1692 return dentry;
1693}
1694
1695static struct dentry *lookup_slow(const struct qstr *name,
1696 struct dentry *dir,
1697 unsigned int flags)
1698{
1699 struct inode *inode = dir->d_inode;
1700 struct dentry *res;
1701 inode_lock_shared(inode);
1702 res = __lookup_slow(name, dir, flags);
1703 inode_unlock_shared(inode);
1704 return res;
1705}
1706
1707static inline int may_lookup(struct user_namespace *mnt_userns,
1708 struct nameidata *nd)
1709{
1710 if (nd->flags & LOOKUP_RCU) {
1711 int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1712 if (err != -ECHILD || !try_to_unlazy(nd))
1713 return err;
1714 }
1715 return inode_permission(mnt_userns, nd->inode, MAY_EXEC);
1716}
1717
1718static int reserve_stack(struct nameidata *nd, struct path *link)
1719{
1720 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1721 return -ELOOP;
1722
1723 if (likely(nd->depth != EMBEDDED_LEVELS))
1724 return 0;
1725 if (likely(nd->stack != nd->internal))
1726 return 0;
1727 if (likely(nd_alloc_stack(nd)))
1728 return 0;
1729
1730 if (nd->flags & LOOKUP_RCU) {
1731 // we need to grab link before we do unlazy. And we can't skip
1732 // unlazy even if we fail to grab the link - cleanup needs it
1733 bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1734
1735 if (!try_to_unlazy(nd) || !grabbed_link)
1736 return -ECHILD;
1737
1738 if (nd_alloc_stack(nd))
1739 return 0;
1740 }
1741 return -ENOMEM;
1742}
1743
1744enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1745
1746static const char *pick_link(struct nameidata *nd, struct path *link,
1747 struct inode *inode, int flags)
1748{
1749 struct saved *last;
1750 const char *res;
1751 int error = reserve_stack(nd, link);
1752
1753 if (unlikely(error)) {
1754 if (!(nd->flags & LOOKUP_RCU))
1755 path_put(link);
1756 return ERR_PTR(error);
1757 }
1758 last = nd->stack + nd->depth++;
1759 last->link = *link;
1760 clear_delayed_call(&last->done);
1761 last->seq = nd->next_seq;
1762
1763 if (flags & WALK_TRAILING) {
1764 error = may_follow_link(nd, inode);
1765 if (unlikely(error))
1766 return ERR_PTR(error);
1767 }
1768
1769 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1770 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1771 return ERR_PTR(-ELOOP);
1772
1773 if (!(nd->flags & LOOKUP_RCU)) {
1774 touch_atime(&last->link);
1775 cond_resched();
1776 } else if (atime_needs_update(&last->link, inode)) {
1777 if (!try_to_unlazy(nd))
1778 return ERR_PTR(-ECHILD);
1779 touch_atime(&last->link);
1780 }
1781
1782 error = security_inode_follow_link(link->dentry, inode,
1783 nd->flags & LOOKUP_RCU);
1784 if (unlikely(error))
1785 return ERR_PTR(error);
1786
1787 res = READ_ONCE(inode->i_link);
1788 if (!res) {
1789 const char * (*get)(struct dentry *, struct inode *,
1790 struct delayed_call *);
1791 get = inode->i_op->get_link;
1792 if (nd->flags & LOOKUP_RCU) {
1793 res = get(NULL, inode, &last->done);
1794 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1795 res = get(link->dentry, inode, &last->done);
1796 } else {
1797 res = get(link->dentry, inode, &last->done);
1798 }
1799 if (!res)
1800 goto all_done;
1801 if (IS_ERR(res))
1802 return res;
1803 }
1804 if (*res == '/') {
1805 error = nd_jump_root(nd);
1806 if (unlikely(error))
1807 return ERR_PTR(error);
1808 while (unlikely(*++res == '/'))
1809 ;
1810 }
1811 if (*res)
1812 return res;
1813all_done: // pure jump
1814 put_link(nd);
1815 return NULL;
1816}
1817
1818/*
1819 * Do we need to follow links? We _really_ want to be able
1820 * to do this check without having to look at inode->i_op,
1821 * so we keep a cache of "no, this doesn't need follow_link"
1822 * for the common case.
1823 *
1824 * NOTE: dentry must be what nd->next_seq had been sampled from.
1825 */
1826static const char *step_into(struct nameidata *nd, int flags,
1827 struct dentry *dentry)
1828{
1829 struct path path;
1830 struct inode *inode;
1831 int err = handle_mounts(nd, dentry, &path);
1832
1833 if (err < 0)
1834 return ERR_PTR(err);
1835 inode = path.dentry->d_inode;
1836 if (likely(!d_is_symlink(path.dentry)) ||
1837 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1838 (flags & WALK_NOFOLLOW)) {
1839 /* not a symlink or should not follow */
1840 if (nd->flags & LOOKUP_RCU) {
1841 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1842 return ERR_PTR(-ECHILD);
1843 if (unlikely(!inode))
1844 return ERR_PTR(-ENOENT);
1845 } else {
1846 dput(nd->path.dentry);
1847 if (nd->path.mnt != path.mnt)
1848 mntput(nd->path.mnt);
1849 }
1850 nd->path = path;
1851 nd->inode = inode;
1852 nd->seq = nd->next_seq;
1853 return NULL;
1854 }
1855 if (nd->flags & LOOKUP_RCU) {
1856 /* make sure that d_is_symlink above matches inode */
1857 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1858 return ERR_PTR(-ECHILD);
1859 } else {
1860 if (path.mnt == nd->path.mnt)
1861 mntget(path.mnt);
1862 }
1863 return pick_link(nd, &path, inode, flags);
1864}
1865
1866static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1867{
1868 struct dentry *parent, *old;
1869
1870 if (path_equal(&nd->path, &nd->root))
1871 goto in_root;
1872 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1873 struct path path;
1874 unsigned seq;
1875 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1876 &nd->root, &path, &seq))
1877 goto in_root;
1878 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1879 return ERR_PTR(-ECHILD);
1880 nd->path = path;
1881 nd->inode = path.dentry->d_inode;
1882 nd->seq = seq;
1883 // makes sure that non-RCU pathwalk could reach this state
1884 if (read_seqretry(&mount_lock, nd->m_seq))
1885 return ERR_PTR(-ECHILD);
1886 /* we know that mountpoint was pinned */
1887 }
1888 old = nd->path.dentry;
1889 parent = old->d_parent;
1890 nd->next_seq = read_seqcount_begin(&parent->d_seq);
1891 // makes sure that non-RCU pathwalk could reach this state
1892 if (read_seqcount_retry(&old->d_seq, nd->seq))
1893 return ERR_PTR(-ECHILD);
1894 if (unlikely(!path_connected(nd->path.mnt, parent)))
1895 return ERR_PTR(-ECHILD);
1896 return parent;
1897in_root:
1898 if (read_seqretry(&mount_lock, nd->m_seq))
1899 return ERR_PTR(-ECHILD);
1900 if (unlikely(nd->flags & LOOKUP_BENEATH))
1901 return ERR_PTR(-ECHILD);
1902 nd->next_seq = nd->seq;
1903 return nd->path.dentry;
1904}
1905
1906static struct dentry *follow_dotdot(struct nameidata *nd)
1907{
1908 struct dentry *parent;
1909
1910 if (path_equal(&nd->path, &nd->root))
1911 goto in_root;
1912 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1913 struct path path;
1914
1915 if (!choose_mountpoint(real_mount(nd->path.mnt),
1916 &nd->root, &path))
1917 goto in_root;
1918 path_put(&nd->path);
1919 nd->path = path;
1920 nd->inode = path.dentry->d_inode;
1921 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1922 return ERR_PTR(-EXDEV);
1923 }
1924 /* rare case of legitimate dget_parent()... */
1925 parent = dget_parent(nd->path.dentry);
1926 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1927 dput(parent);
1928 return ERR_PTR(-ENOENT);
1929 }
1930 return parent;
1931
1932in_root:
1933 if (unlikely(nd->flags & LOOKUP_BENEATH))
1934 return ERR_PTR(-EXDEV);
1935 return dget(nd->path.dentry);
1936}
1937
1938static const char *handle_dots(struct nameidata *nd, int type)
1939{
1940 if (type == LAST_DOTDOT) {
1941 const char *error = NULL;
1942 struct dentry *parent;
1943
1944 if (!nd->root.mnt) {
1945 error = ERR_PTR(set_root(nd));
1946 if (error)
1947 return error;
1948 }
1949 if (nd->flags & LOOKUP_RCU)
1950 parent = follow_dotdot_rcu(nd);
1951 else
1952 parent = follow_dotdot(nd);
1953 if (IS_ERR(parent))
1954 return ERR_CAST(parent);
1955 error = step_into(nd, WALK_NOFOLLOW, parent);
1956 if (unlikely(error))
1957 return error;
1958
1959 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1960 /*
1961 * If there was a racing rename or mount along our
1962 * path, then we can't be sure that ".." hasn't jumped
1963 * above nd->root (and so userspace should retry or use
1964 * some fallback).
1965 */
1966 smp_rmb();
1967 if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1968 return ERR_PTR(-EAGAIN);
1969 if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1970 return ERR_PTR(-EAGAIN);
1971 }
1972 }
1973 return NULL;
1974}
1975
1976static const char *walk_component(struct nameidata *nd, int flags)
1977{
1978 struct dentry *dentry;
1979 /*
1980 * "." and ".." are special - ".." especially so because it has
1981 * to be able to know about the current root directory and
1982 * parent relationships.
1983 */
1984 if (unlikely(nd->last_type != LAST_NORM)) {
1985 if (!(flags & WALK_MORE) && nd->depth)
1986 put_link(nd);
1987 return handle_dots(nd, nd->last_type);
1988 }
1989 dentry = lookup_fast(nd);
1990 if (IS_ERR(dentry))
1991 return ERR_CAST(dentry);
1992 if (unlikely(!dentry)) {
1993 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1994 if (IS_ERR(dentry))
1995 return ERR_CAST(dentry);
1996 }
1997 if (!(flags & WALK_MORE) && nd->depth)
1998 put_link(nd);
1999 return step_into(nd, flags, dentry);
2000}
2001
2002/*
2003 * We can do the critical dentry name comparison and hashing
2004 * operations one word at a time, but we are limited to:
2005 *
2006 * - Architectures with fast unaligned word accesses. We could
2007 * do a "get_unaligned()" if this helps and is sufficiently
2008 * fast.
2009 *
2010 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2011 * do not trap on the (extremely unlikely) case of a page
2012 * crossing operation.
2013 *
2014 * - Furthermore, we need an efficient 64-bit compile for the
2015 * 64-bit case in order to generate the "number of bytes in
2016 * the final mask". Again, that could be replaced with a
2017 * efficient population count instruction or similar.
2018 */
2019#ifdef CONFIG_DCACHE_WORD_ACCESS
2020
2021#include <asm/word-at-a-time.h>
2022
2023#ifdef HASH_MIX
2024
2025/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2026
2027#elif defined(CONFIG_64BIT)
2028/*
2029 * Register pressure in the mixing function is an issue, particularly
2030 * on 32-bit x86, but almost any function requires one state value and
2031 * one temporary. Instead, use a function designed for two state values
2032 * and no temporaries.
2033 *
2034 * This function cannot create a collision in only two iterations, so
2035 * we have two iterations to achieve avalanche. In those two iterations,
2036 * we have six layers of mixing, which is enough to spread one bit's
2037 * influence out to 2^6 = 64 state bits.
2038 *
2039 * Rotate constants are scored by considering either 64 one-bit input
2040 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2041 * probability of that delta causing a change to each of the 128 output
2042 * bits, using a sample of random initial states.
2043 *
2044 * The Shannon entropy of the computed probabilities is then summed
2045 * to produce a score. Ideally, any input change has a 50% chance of
2046 * toggling any given output bit.
2047 *
2048 * Mixing scores (in bits) for (12,45):
2049 * Input delta: 1-bit 2-bit
2050 * 1 round: 713.3 42542.6
2051 * 2 rounds: 2753.7 140389.8
2052 * 3 rounds: 5954.1 233458.2
2053 * 4 rounds: 7862.6 256672.2
2054 * Perfect: 8192 258048
2055 * (64*128) (64*63/2 * 128)
2056 */
2057#define HASH_MIX(x, y, a) \
2058 ( x ^= (a), \
2059 y ^= x, x = rol64(x,12),\
2060 x += y, y = rol64(y,45),\
2061 y *= 9 )
2062
2063/*
2064 * Fold two longs into one 32-bit hash value. This must be fast, but
2065 * latency isn't quite as critical, as there is a fair bit of additional
2066 * work done before the hash value is used.
2067 */
2068static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2069{
2070 y ^= x * GOLDEN_RATIO_64;
2071 y *= GOLDEN_RATIO_64;
2072 return y >> 32;
2073}
2074
2075#else /* 32-bit case */
2076
2077/*
2078 * Mixing scores (in bits) for (7,20):
2079 * Input delta: 1-bit 2-bit
2080 * 1 round: 330.3 9201.6
2081 * 2 rounds: 1246.4 25475.4
2082 * 3 rounds: 1907.1 31295.1
2083 * 4 rounds: 2042.3 31718.6
2084 * Perfect: 2048 31744
2085 * (32*64) (32*31/2 * 64)
2086 */
2087#define HASH_MIX(x, y, a) \
2088 ( x ^= (a), \
2089 y ^= x, x = rol32(x, 7),\
2090 x += y, y = rol32(y,20),\
2091 y *= 9 )
2092
2093static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2094{
2095 /* Use arch-optimized multiply if one exists */
2096 return __hash_32(y ^ __hash_32(x));
2097}
2098
2099#endif
2100
2101/*
2102 * Return the hash of a string of known length. This is carfully
2103 * designed to match hash_name(), which is the more critical function.
2104 * In particular, we must end by hashing a final word containing 0..7
2105 * payload bytes, to match the way that hash_name() iterates until it
2106 * finds the delimiter after the name.
2107 */
2108unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2109{
2110 unsigned long a, x = 0, y = (unsigned long)salt;
2111
2112 for (;;) {
2113 if (!len)
2114 goto done;
2115 a = load_unaligned_zeropad(name);
2116 if (len < sizeof(unsigned long))
2117 break;
2118 HASH_MIX(x, y, a);
2119 name += sizeof(unsigned long);
2120 len -= sizeof(unsigned long);
2121 }
2122 x ^= a & bytemask_from_count(len);
2123done:
2124 return fold_hash(x, y);
2125}
2126EXPORT_SYMBOL(full_name_hash);
2127
2128/* Return the "hash_len" (hash and length) of a null-terminated string */
2129u64 hashlen_string(const void *salt, const char *name)
2130{
2131 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2132 unsigned long adata, mask, len;
2133 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2134
2135 len = 0;
2136 goto inside;
2137
2138 do {
2139 HASH_MIX(x, y, a);
2140 len += sizeof(unsigned long);
2141inside:
2142 a = load_unaligned_zeropad(name+len);
2143 } while (!has_zero(a, &adata, &constants));
2144
2145 adata = prep_zero_mask(a, adata, &constants);
2146 mask = create_zero_mask(adata);
2147 x ^= a & zero_bytemask(mask);
2148
2149 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2150}
2151EXPORT_SYMBOL(hashlen_string);
2152
2153/*
2154 * Calculate the length and hash of the path component, and
2155 * return the "hash_len" as the result.
2156 */
2157static inline u64 hash_name(const void *salt, const char *name)
2158{
2159 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2160 unsigned long adata, bdata, mask, len;
2161 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2162
2163 len = 0;
2164 goto inside;
2165
2166 do {
2167 HASH_MIX(x, y, a);
2168 len += sizeof(unsigned long);
2169inside:
2170 a = load_unaligned_zeropad(name+len);
2171 b = a ^ REPEAT_BYTE('/');
2172 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2173
2174 adata = prep_zero_mask(a, adata, &constants);
2175 bdata = prep_zero_mask(b, bdata, &constants);
2176 mask = create_zero_mask(adata | bdata);
2177 x ^= a & zero_bytemask(mask);
2178
2179 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2180}
2181
2182#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2183
2184/* Return the hash of a string of known length */
2185unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2186{
2187 unsigned long hash = init_name_hash(salt);
2188 while (len--)
2189 hash = partial_name_hash((unsigned char)*name++, hash);
2190 return end_name_hash(hash);
2191}
2192EXPORT_SYMBOL(full_name_hash);
2193
2194/* Return the "hash_len" (hash and length) of a null-terminated string */
2195u64 hashlen_string(const void *salt, const char *name)
2196{
2197 unsigned long hash = init_name_hash(salt);
2198 unsigned long len = 0, c;
2199
2200 c = (unsigned char)*name;
2201 while (c) {
2202 len++;
2203 hash = partial_name_hash(c, hash);
2204 c = (unsigned char)name[len];
2205 }
2206 return hashlen_create(end_name_hash(hash), len);
2207}
2208EXPORT_SYMBOL(hashlen_string);
2209
2210/*
2211 * We know there's a real path component here of at least
2212 * one character.
2213 */
2214static inline u64 hash_name(const void *salt, const char *name)
2215{
2216 unsigned long hash = init_name_hash(salt);
2217 unsigned long len = 0, c;
2218
2219 c = (unsigned char)*name;
2220 do {
2221 len++;
2222 hash = partial_name_hash(c, hash);
2223 c = (unsigned char)name[len];
2224 } while (c && c != '/');
2225 return hashlen_create(end_name_hash(hash), len);
2226}
2227
2228#endif
2229
2230/*
2231 * Name resolution.
2232 * This is the basic name resolution function, turning a pathname into
2233 * the final dentry. We expect 'base' to be positive and a directory.
2234 *
2235 * Returns 0 and nd will have valid dentry and mnt on success.
2236 * Returns error and drops reference to input namei data on failure.
2237 */
2238static int link_path_walk(const char *name, struct nameidata *nd)
2239{
2240 int depth = 0; // depth <= nd->depth
2241 int err;
2242
2243 nd->last_type = LAST_ROOT;
2244 nd->flags |= LOOKUP_PARENT;
2245 if (IS_ERR(name))
2246 return PTR_ERR(name);
2247 while (*name=='/')
2248 name++;
2249 if (!*name) {
2250 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2251 return 0;
2252 }
2253
2254 /* At this point we know we have a real path component. */
2255 for(;;) {
2256 struct user_namespace *mnt_userns;
2257 const char *link;
2258 u64 hash_len;
2259 int type;
2260
2261 mnt_userns = mnt_user_ns(nd->path.mnt);
2262 err = may_lookup(mnt_userns, nd);
2263 if (err)
2264 return err;
2265
2266 hash_len = hash_name(nd->path.dentry, name);
2267
2268 type = LAST_NORM;
2269 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2270 case 2:
2271 if (name[1] == '.') {
2272 type = LAST_DOTDOT;
2273 nd->state |= ND_JUMPED;
2274 }
2275 break;
2276 case 1:
2277 type = LAST_DOT;
2278 }
2279 if (likely(type == LAST_NORM)) {
2280 struct dentry *parent = nd->path.dentry;
2281 nd->state &= ~ND_JUMPED;
2282 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2283 struct qstr this = { { .hash_len = hash_len }, .name = name };
2284 err = parent->d_op->d_hash(parent, &this);
2285 if (err < 0)
2286 return err;
2287 hash_len = this.hash_len;
2288 name = this.name;
2289 }
2290 }
2291
2292 nd->last.hash_len = hash_len;
2293 nd->last.name = name;
2294 nd->last_type = type;
2295
2296 name += hashlen_len(hash_len);
2297 if (!*name)
2298 goto OK;
2299 /*
2300 * If it wasn't NUL, we know it was '/'. Skip that
2301 * slash, and continue until no more slashes.
2302 */
2303 do {
2304 name++;
2305 } while (unlikely(*name == '/'));
2306 if (unlikely(!*name)) {
2307OK:
2308 /* pathname or trailing symlink, done */
2309 if (!depth) {
2310 nd->dir_vfsuid = i_uid_into_vfsuid(mnt_userns, nd->inode);
2311 nd->dir_mode = nd->inode->i_mode;
2312 nd->flags &= ~LOOKUP_PARENT;
2313 return 0;
2314 }
2315 /* last component of nested symlink */
2316 name = nd->stack[--depth].name;
2317 link = walk_component(nd, 0);
2318 } else {
2319 /* not the last component */
2320 link = walk_component(nd, WALK_MORE);
2321 }
2322 if (unlikely(link)) {
2323 if (IS_ERR(link))
2324 return PTR_ERR(link);
2325 /* a symlink to follow */
2326 nd->stack[depth++].name = name;
2327 name = link;
2328 continue;
2329 }
2330 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2331 if (nd->flags & LOOKUP_RCU) {
2332 if (!try_to_unlazy(nd))
2333 return -ECHILD;
2334 }
2335 return -ENOTDIR;
2336 }
2337 }
2338}
2339
2340/* must be paired with terminate_walk() */
2341static const char *path_init(struct nameidata *nd, unsigned flags)
2342{
2343 int error;
2344 const char *s = nd->name->name;
2345
2346 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2347 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2348 return ERR_PTR(-EAGAIN);
2349
2350 if (!*s)
2351 flags &= ~LOOKUP_RCU;
2352 if (flags & LOOKUP_RCU)
2353 rcu_read_lock();
2354 else
2355 nd->seq = nd->next_seq = 0;
2356
2357 nd->flags = flags;
2358 nd->state |= ND_JUMPED;
2359
2360 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2361 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2362 smp_rmb();
2363
2364 if (nd->state & ND_ROOT_PRESET) {
2365 struct dentry *root = nd->root.dentry;
2366 struct inode *inode = root->d_inode;
2367 if (*s && unlikely(!d_can_lookup(root)))
2368 return ERR_PTR(-ENOTDIR);
2369 nd->path = nd->root;
2370 nd->inode = inode;
2371 if (flags & LOOKUP_RCU) {
2372 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2373 nd->root_seq = nd->seq;
2374 } else {
2375 path_get(&nd->path);
2376 }
2377 return s;
2378 }
2379
2380 nd->root.mnt = NULL;
2381
2382 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2383 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2384 error = nd_jump_root(nd);
2385 if (unlikely(error))
2386 return ERR_PTR(error);
2387 return s;
2388 }
2389
2390 /* Relative pathname -- get the starting-point it is relative to. */
2391 if (nd->dfd == AT_FDCWD) {
2392 if (flags & LOOKUP_RCU) {
2393 struct fs_struct *fs = current->fs;
2394 unsigned seq;
2395
2396 do {
2397 seq = read_seqcount_begin(&fs->seq);
2398 nd->path = fs->pwd;
2399 nd->inode = nd->path.dentry->d_inode;
2400 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2401 } while (read_seqcount_retry(&fs->seq, seq));
2402 } else {
2403 get_fs_pwd(current->fs, &nd->path);
2404 nd->inode = nd->path.dentry->d_inode;
2405 }
2406 } else {
2407 /* Caller must check execute permissions on the starting path component */
2408 struct fd f = fdget_raw(nd->dfd);
2409 struct dentry *dentry;
2410
2411 if (!f.file)
2412 return ERR_PTR(-EBADF);
2413
2414 dentry = f.file->f_path.dentry;
2415
2416 if (*s && unlikely(!d_can_lookup(dentry))) {
2417 fdput(f);
2418 return ERR_PTR(-ENOTDIR);
2419 }
2420
2421 nd->path = f.file->f_path;
2422 if (flags & LOOKUP_RCU) {
2423 nd->inode = nd->path.dentry->d_inode;
2424 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2425 } else {
2426 path_get(&nd->path);
2427 nd->inode = nd->path.dentry->d_inode;
2428 }
2429 fdput(f);
2430 }
2431
2432 /* For scoped-lookups we need to set the root to the dirfd as well. */
2433 if (flags & LOOKUP_IS_SCOPED) {
2434 nd->root = nd->path;
2435 if (flags & LOOKUP_RCU) {
2436 nd->root_seq = nd->seq;
2437 } else {
2438 path_get(&nd->root);
2439 nd->state |= ND_ROOT_GRABBED;
2440 }
2441 }
2442 return s;
2443}
2444
2445static inline const char *lookup_last(struct nameidata *nd)
2446{
2447 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2448 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2449
2450 return walk_component(nd, WALK_TRAILING);
2451}
2452
2453static int handle_lookup_down(struct nameidata *nd)
2454{
2455 if (!(nd->flags & LOOKUP_RCU))
2456 dget(nd->path.dentry);
2457 nd->next_seq = nd->seq;
2458 return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2459}
2460
2461/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2462static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2463{
2464 const char *s = path_init(nd, flags);
2465 int err;
2466
2467 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2468 err = handle_lookup_down(nd);
2469 if (unlikely(err < 0))
2470 s = ERR_PTR(err);
2471 }
2472
2473 while (!(err = link_path_walk(s, nd)) &&
2474 (s = lookup_last(nd)) != NULL)
2475 ;
2476 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2477 err = handle_lookup_down(nd);
2478 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2479 }
2480 if (!err)
2481 err = complete_walk(nd);
2482
2483 if (!err && nd->flags & LOOKUP_DIRECTORY)
2484 if (!d_can_lookup(nd->path.dentry))
2485 err = -ENOTDIR;
2486 if (!err) {
2487 *path = nd->path;
2488 nd->path.mnt = NULL;
2489 nd->path.dentry = NULL;
2490 }
2491 terminate_walk(nd);
2492 return err;
2493}
2494
2495int filename_lookup(int dfd, struct filename *name, unsigned flags,
2496 struct path *path, struct path *root)
2497{
2498 int retval;
2499 struct nameidata nd;
2500 if (IS_ERR(name))
2501 return PTR_ERR(name);
2502 set_nameidata(&nd, dfd, name, root);
2503 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2504 if (unlikely(retval == -ECHILD))
2505 retval = path_lookupat(&nd, flags, path);
2506 if (unlikely(retval == -ESTALE))
2507 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2508
2509 if (likely(!retval))
2510 audit_inode(name, path->dentry,
2511 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2512 restore_nameidata();
2513 return retval;
2514}
2515
2516/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2517static int path_parentat(struct nameidata *nd, unsigned flags,
2518 struct path *parent)
2519{
2520 const char *s = path_init(nd, flags);
2521 int err = link_path_walk(s, nd);
2522 if (!err)
2523 err = complete_walk(nd);
2524 if (!err) {
2525 *parent = nd->path;
2526 nd->path.mnt = NULL;
2527 nd->path.dentry = NULL;
2528 }
2529 terminate_walk(nd);
2530 return err;
2531}
2532
2533/* Note: this does not consume "name" */
2534static int filename_parentat(int dfd, struct filename *name,
2535 unsigned int flags, struct path *parent,
2536 struct qstr *last, int *type)
2537{
2538 int retval;
2539 struct nameidata nd;
2540
2541 if (IS_ERR(name))
2542 return PTR_ERR(name);
2543 set_nameidata(&nd, dfd, name, NULL);
2544 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2545 if (unlikely(retval == -ECHILD))
2546 retval = path_parentat(&nd, flags, parent);
2547 if (unlikely(retval == -ESTALE))
2548 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2549 if (likely(!retval)) {
2550 *last = nd.last;
2551 *type = nd.last_type;
2552 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2553 }
2554 restore_nameidata();
2555 return retval;
2556}
2557
2558/* does lookup, returns the object with parent locked */
2559static struct dentry *__kern_path_locked(struct filename *name, struct path *path)
2560{
2561 struct dentry *d;
2562 struct qstr last;
2563 int type, error;
2564
2565 error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type);
2566 if (error)
2567 return ERR_PTR(error);
2568 if (unlikely(type != LAST_NORM)) {
2569 path_put(path);
2570 return ERR_PTR(-EINVAL);
2571 }
2572 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2573 d = __lookup_hash(&last, path->dentry, 0);
2574 if (IS_ERR(d)) {
2575 inode_unlock(path->dentry->d_inode);
2576 path_put(path);
2577 }
2578 return d;
2579}
2580
2581struct dentry *kern_path_locked(const char *name, struct path *path)
2582{
2583 struct filename *filename = getname_kernel(name);
2584 struct dentry *res = __kern_path_locked(filename, path);
2585
2586 putname(filename);
2587 return res;
2588}
2589
2590int kern_path(const char *name, unsigned int flags, struct path *path)
2591{
2592 struct filename *filename = getname_kernel(name);
2593 int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2594
2595 putname(filename);
2596 return ret;
2597
2598}
2599EXPORT_SYMBOL(kern_path);
2600
2601/**
2602 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2603 * @dentry: pointer to dentry of the base directory
2604 * @mnt: pointer to vfs mount of the base directory
2605 * @name: pointer to file name
2606 * @flags: lookup flags
2607 * @path: pointer to struct path to fill
2608 */
2609int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2610 const char *name, unsigned int flags,
2611 struct path *path)
2612{
2613 struct filename *filename;
2614 struct path root = {.mnt = mnt, .dentry = dentry};
2615 int ret;
2616
2617 filename = getname_kernel(name);
2618 /* the first argument of filename_lookup() is ignored with root */
2619 ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2620 putname(filename);
2621 return ret;
2622}
2623EXPORT_SYMBOL(vfs_path_lookup);
2624
2625static int lookup_one_common(struct user_namespace *mnt_userns,
2626 const char *name, struct dentry *base, int len,
2627 struct qstr *this)
2628{
2629 this->name = name;
2630 this->len = len;
2631 this->hash = full_name_hash(base, name, len);
2632 if (!len)
2633 return -EACCES;
2634
2635 if (unlikely(name[0] == '.')) {
2636 if (len < 2 || (len == 2 && name[1] == '.'))
2637 return -EACCES;
2638 }
2639
2640 while (len--) {
2641 unsigned int c = *(const unsigned char *)name++;
2642 if (c == '/' || c == '\0')
2643 return -EACCES;
2644 }
2645 /*
2646 * See if the low-level filesystem might want
2647 * to use its own hash..
2648 */
2649 if (base->d_flags & DCACHE_OP_HASH) {
2650 int err = base->d_op->d_hash(base, this);
2651 if (err < 0)
2652 return err;
2653 }
2654
2655 return inode_permission(mnt_userns, base->d_inode, MAY_EXEC);
2656}
2657
2658/**
2659 * try_lookup_one_len - filesystem helper to lookup single pathname component
2660 * @name: pathname component to lookup
2661 * @base: base directory to lookup from
2662 * @len: maximum length @len should be interpreted to
2663 *
2664 * Look up a dentry by name in the dcache, returning NULL if it does not
2665 * currently exist. The function does not try to create a dentry.
2666 *
2667 * Note that this routine is purely a helper for filesystem usage and should
2668 * not be called by generic code.
2669 *
2670 * The caller must hold base->i_mutex.
2671 */
2672struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2673{
2674 struct qstr this;
2675 int err;
2676
2677 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2678
2679 err = lookup_one_common(&init_user_ns, name, base, len, &this);
2680 if (err)
2681 return ERR_PTR(err);
2682
2683 return lookup_dcache(&this, base, 0);
2684}
2685EXPORT_SYMBOL(try_lookup_one_len);
2686
2687/**
2688 * lookup_one_len - filesystem helper to lookup single pathname component
2689 * @name: pathname component to lookup
2690 * @base: base directory to lookup from
2691 * @len: maximum length @len should be interpreted to
2692 *
2693 * Note that this routine is purely a helper for filesystem usage and should
2694 * not be called by generic code.
2695 *
2696 * The caller must hold base->i_mutex.
2697 */
2698struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2699{
2700 struct dentry *dentry;
2701 struct qstr this;
2702 int err;
2703
2704 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2705
2706 err = lookup_one_common(&init_user_ns, name, base, len, &this);
2707 if (err)
2708 return ERR_PTR(err);
2709
2710 dentry = lookup_dcache(&this, base, 0);
2711 return dentry ? dentry : __lookup_slow(&this, base, 0);
2712}
2713EXPORT_SYMBOL(lookup_one_len);
2714
2715/**
2716 * lookup_one - filesystem helper to lookup single pathname component
2717 * @mnt_userns: user namespace of the mount the lookup is performed from
2718 * @name: pathname component to lookup
2719 * @base: base directory to lookup from
2720 * @len: maximum length @len should be interpreted to
2721 *
2722 * Note that this routine is purely a helper for filesystem usage and should
2723 * not be called by generic code.
2724 *
2725 * The caller must hold base->i_mutex.
2726 */
2727struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name,
2728 struct dentry *base, int len)
2729{
2730 struct dentry *dentry;
2731 struct qstr this;
2732 int err;
2733
2734 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2735
2736 err = lookup_one_common(mnt_userns, name, base, len, &this);
2737 if (err)
2738 return ERR_PTR(err);
2739
2740 dentry = lookup_dcache(&this, base, 0);
2741 return dentry ? dentry : __lookup_slow(&this, base, 0);
2742}
2743EXPORT_SYMBOL(lookup_one);
2744
2745/**
2746 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2747 * @mnt_userns: idmapping of the mount the lookup is performed from
2748 * @name: pathname component to lookup
2749 * @base: base directory to lookup from
2750 * @len: maximum length @len should be interpreted to
2751 *
2752 * Note that this routine is purely a helper for filesystem usage and should
2753 * not be called by generic code.
2754 *
2755 * Unlike lookup_one_len, it should be called without the parent
2756 * i_mutex held, and will take the i_mutex itself if necessary.
2757 */
2758struct dentry *lookup_one_unlocked(struct user_namespace *mnt_userns,
2759 const char *name, struct dentry *base,
2760 int len)
2761{
2762 struct qstr this;
2763 int err;
2764 struct dentry *ret;
2765
2766 err = lookup_one_common(mnt_userns, name, base, len, &this);
2767 if (err)
2768 return ERR_PTR(err);
2769
2770 ret = lookup_dcache(&this, base, 0);
2771 if (!ret)
2772 ret = lookup_slow(&this, base, 0);
2773 return ret;
2774}
2775EXPORT_SYMBOL(lookup_one_unlocked);
2776
2777/**
2778 * lookup_one_positive_unlocked - filesystem helper to lookup single
2779 * pathname component
2780 * @mnt_userns: idmapping of the mount the lookup is performed from
2781 * @name: pathname component to lookup
2782 * @base: base directory to lookup from
2783 * @len: maximum length @len should be interpreted to
2784 *
2785 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2786 * known positive or ERR_PTR(). This is what most of the users want.
2787 *
2788 * Note that pinned negative with unlocked parent _can_ become positive at any
2789 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2790 * positives have >d_inode stable, so this one avoids such problems.
2791 *
2792 * Note that this routine is purely a helper for filesystem usage and should
2793 * not be called by generic code.
2794 *
2795 * The helper should be called without i_mutex held.
2796 */
2797struct dentry *lookup_one_positive_unlocked(struct user_namespace *mnt_userns,
2798 const char *name,
2799 struct dentry *base, int len)
2800{
2801 struct dentry *ret = lookup_one_unlocked(mnt_userns, name, base, len);
2802
2803 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2804 dput(ret);
2805 ret = ERR_PTR(-ENOENT);
2806 }
2807 return ret;
2808}
2809EXPORT_SYMBOL(lookup_one_positive_unlocked);
2810
2811/**
2812 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2813 * @name: pathname component to lookup
2814 * @base: base directory to lookup from
2815 * @len: maximum length @len should be interpreted to
2816 *
2817 * Note that this routine is purely a helper for filesystem usage and should
2818 * not be called by generic code.
2819 *
2820 * Unlike lookup_one_len, it should be called without the parent
2821 * i_mutex held, and will take the i_mutex itself if necessary.
2822 */
2823struct dentry *lookup_one_len_unlocked(const char *name,
2824 struct dentry *base, int len)
2825{
2826 return lookup_one_unlocked(&init_user_ns, name, base, len);
2827}
2828EXPORT_SYMBOL(lookup_one_len_unlocked);
2829
2830/*
2831 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2832 * on negatives. Returns known positive or ERR_PTR(); that's what
2833 * most of the users want. Note that pinned negative with unlocked parent
2834 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2835 * need to be very careful; pinned positives have ->d_inode stable, so
2836 * this one avoids such problems.
2837 */
2838struct dentry *lookup_positive_unlocked(const char *name,
2839 struct dentry *base, int len)
2840{
2841 return lookup_one_positive_unlocked(&init_user_ns, name, base, len);
2842}
2843EXPORT_SYMBOL(lookup_positive_unlocked);
2844
2845#ifdef CONFIG_UNIX98_PTYS
2846int path_pts(struct path *path)
2847{
2848 /* Find something mounted on "pts" in the same directory as
2849 * the input path.
2850 */
2851 struct dentry *parent = dget_parent(path->dentry);
2852 struct dentry *child;
2853 struct qstr this = QSTR_INIT("pts", 3);
2854
2855 if (unlikely(!path_connected(path->mnt, parent))) {
2856 dput(parent);
2857 return -ENOENT;
2858 }
2859 dput(path->dentry);
2860 path->dentry = parent;
2861 child = d_hash_and_lookup(parent, &this);
2862 if (!child)
2863 return -ENOENT;
2864
2865 path->dentry = child;
2866 dput(parent);
2867 follow_down(path);
2868 return 0;
2869}
2870#endif
2871
2872int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2873 struct path *path, int *empty)
2874{
2875 struct filename *filename = getname_flags(name, flags, empty);
2876 int ret = filename_lookup(dfd, filename, flags, path, NULL);
2877
2878 putname(filename);
2879 return ret;
2880}
2881EXPORT_SYMBOL(user_path_at_empty);
2882
2883int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2884 struct inode *inode)
2885{
2886 kuid_t fsuid = current_fsuid();
2887
2888 if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), fsuid))
2889 return 0;
2890 if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, dir), fsuid))
2891 return 0;
2892 return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
2893}
2894EXPORT_SYMBOL(__check_sticky);
2895
2896/*
2897 * Check whether we can remove a link victim from directory dir, check
2898 * whether the type of victim is right.
2899 * 1. We can't do it if dir is read-only (done in permission())
2900 * 2. We should have write and exec permissions on dir
2901 * 3. We can't remove anything from append-only dir
2902 * 4. We can't do anything with immutable dir (done in permission())
2903 * 5. If the sticky bit on dir is set we should either
2904 * a. be owner of dir, or
2905 * b. be owner of victim, or
2906 * c. have CAP_FOWNER capability
2907 * 6. If the victim is append-only or immutable we can't do antyhing with
2908 * links pointing to it.
2909 * 7. If the victim has an unknown uid or gid we can't change the inode.
2910 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2911 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2912 * 10. We can't remove a root or mountpoint.
2913 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2914 * nfs_async_unlink().
2915 */
2916static int may_delete(struct user_namespace *mnt_userns, struct inode *dir,
2917 struct dentry *victim, bool isdir)
2918{
2919 struct inode *inode = d_backing_inode(victim);
2920 int error;
2921
2922 if (d_is_negative(victim))
2923 return -ENOENT;
2924 BUG_ON(!inode);
2925
2926 BUG_ON(victim->d_parent->d_inode != dir);
2927
2928 /* Inode writeback is not safe when the uid or gid are invalid. */
2929 if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) ||
2930 !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode)))
2931 return -EOVERFLOW;
2932
2933 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2934
2935 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2936 if (error)
2937 return error;
2938 if (IS_APPEND(dir))
2939 return -EPERM;
2940
2941 if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2942 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2943 HAS_UNMAPPED_ID(mnt_userns, inode))
2944 return -EPERM;
2945 if (isdir) {
2946 if (!d_is_dir(victim))
2947 return -ENOTDIR;
2948 if (IS_ROOT(victim))
2949 return -EBUSY;
2950 } else if (d_is_dir(victim))
2951 return -EISDIR;
2952 if (IS_DEADDIR(dir))
2953 return -ENOENT;
2954 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2955 return -EBUSY;
2956 return 0;
2957}
2958
2959/* Check whether we can create an object with dentry child in directory
2960 * dir.
2961 * 1. We can't do it if child already exists (open has special treatment for
2962 * this case, but since we are inlined it's OK)
2963 * 2. We can't do it if dir is read-only (done in permission())
2964 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2965 * 4. We should have write and exec permissions on dir
2966 * 5. We can't do it if dir is immutable (done in permission())
2967 */
2968static inline int may_create(struct user_namespace *mnt_userns,
2969 struct inode *dir, struct dentry *child)
2970{
2971 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2972 if (child->d_inode)
2973 return -EEXIST;
2974 if (IS_DEADDIR(dir))
2975 return -ENOENT;
2976 if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
2977 return -EOVERFLOW;
2978
2979 return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2980}
2981
2982/*
2983 * p1 and p2 should be directories on the same fs.
2984 */
2985struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2986{
2987 struct dentry *p;
2988
2989 if (p1 == p2) {
2990 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2991 return NULL;
2992 }
2993
2994 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2995
2996 p = d_ancestor(p2, p1);
2997 if (p) {
2998 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2999 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
3000 return p;
3001 }
3002
3003 p = d_ancestor(p1, p2);
3004 if (p) {
3005 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3006 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
3007 return p;
3008 }
3009
3010 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3011 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3012 return NULL;
3013}
3014EXPORT_SYMBOL(lock_rename);
3015
3016void unlock_rename(struct dentry *p1, struct dentry *p2)
3017{
3018 inode_unlock(p1->d_inode);
3019 if (p1 != p2) {
3020 inode_unlock(p2->d_inode);
3021 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3022 }
3023}
3024EXPORT_SYMBOL(unlock_rename);
3025
3026/**
3027 * mode_strip_umask - handle vfs umask stripping
3028 * @dir: parent directory of the new inode
3029 * @mode: mode of the new inode to be created in @dir
3030 *
3031 * Umask stripping depends on whether or not the filesystem supports POSIX
3032 * ACLs. If the filesystem doesn't support it umask stripping is done directly
3033 * in here. If the filesystem does support POSIX ACLs umask stripping is
3034 * deferred until the filesystem calls posix_acl_create().
3035 *
3036 * Returns: mode
3037 */
3038static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode)
3039{
3040 if (!IS_POSIXACL(dir))
3041 mode &= ~current_umask();
3042 return mode;
3043}
3044
3045/**
3046 * vfs_prepare_mode - prepare the mode to be used for a new inode
3047 * @mnt_userns: user namespace of the mount the inode was found from
3048 * @dir: parent directory of the new inode
3049 * @mode: mode of the new inode
3050 * @mask_perms: allowed permission by the vfs
3051 * @type: type of file to be created
3052 *
3053 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3054 * object to be created.
3055 *
3056 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3057 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3058 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3059 * POSIX ACL supporting filesystems.
3060 *
3061 * Note that it's currently valid for @type to be 0 if a directory is created.
3062 * Filesystems raise that flag individually and we need to check whether each
3063 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3064 * non-zero type.
3065 *
3066 * Returns: mode to be passed to the filesystem
3067 */
3068static inline umode_t vfs_prepare_mode(struct user_namespace *mnt_userns,
3069 const struct inode *dir, umode_t mode,
3070 umode_t mask_perms, umode_t type)
3071{
3072 mode = mode_strip_sgid(mnt_userns, dir, mode);
3073 mode = mode_strip_umask(dir, mode);
3074
3075 /*
3076 * Apply the vfs mandated allowed permission mask and set the type of
3077 * file to be created before we call into the filesystem.
3078 */
3079 mode &= (mask_perms & ~S_IFMT);
3080 mode |= (type & S_IFMT);
3081
3082 return mode;
3083}
3084
3085/**
3086 * vfs_create - create new file
3087 * @mnt_userns: user namespace of the mount the inode was found from
3088 * @dir: inode of @dentry
3089 * @dentry: pointer to dentry of the base directory
3090 * @mode: mode of the new file
3091 * @want_excl: whether the file must not yet exist
3092 *
3093 * Create a new file.
3094 *
3095 * If the inode has been found through an idmapped mount the user namespace of
3096 * the vfsmount must be passed through @mnt_userns. This function will then take
3097 * care to map the inode according to @mnt_userns before checking permissions.
3098 * On non-idmapped mounts or if permission checking is to be performed on the
3099 * raw inode simply passs init_user_ns.
3100 */
3101int vfs_create(struct user_namespace *mnt_userns, struct inode *dir,
3102 struct dentry *dentry, umode_t mode, bool want_excl)
3103{
3104 int error = may_create(mnt_userns, dir, dentry);
3105 if (error)
3106 return error;
3107
3108 if (!dir->i_op->create)
3109 return -EACCES; /* shouldn't it be ENOSYS? */
3110
3111 mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IALLUGO, S_IFREG);
3112 error = security_inode_create(dir, dentry, mode);
3113 if (error)
3114 return error;
3115 error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl);
3116 if (!error)
3117 fsnotify_create(dir, dentry);
3118 return error;
3119}
3120EXPORT_SYMBOL(vfs_create);
3121
3122int vfs_mkobj(struct dentry *dentry, umode_t mode,
3123 int (*f)(struct dentry *, umode_t, void *),
3124 void *arg)
3125{
3126 struct inode *dir = dentry->d_parent->d_inode;
3127 int error = may_create(&init_user_ns, dir, dentry);
3128 if (error)
3129 return error;
3130
3131 mode &= S_IALLUGO;
3132 mode |= S_IFREG;
3133 error = security_inode_create(dir, dentry, mode);
3134 if (error)
3135 return error;
3136 error = f(dentry, mode, arg);
3137 if (!error)
3138 fsnotify_create(dir, dentry);
3139 return error;
3140}
3141EXPORT_SYMBOL(vfs_mkobj);
3142
3143bool may_open_dev(const struct path *path)
3144{
3145 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3146 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3147}
3148
3149static int may_open(struct user_namespace *mnt_userns, const struct path *path,
3150 int acc_mode, int flag)
3151{
3152 struct dentry *dentry = path->dentry;
3153 struct inode *inode = dentry->d_inode;
3154 int error;
3155
3156 if (!inode)
3157 return -ENOENT;
3158
3159 switch (inode->i_mode & S_IFMT) {
3160 case S_IFLNK:
3161 return -ELOOP;
3162 case S_IFDIR:
3163 if (acc_mode & MAY_WRITE)
3164 return -EISDIR;
3165 if (acc_mode & MAY_EXEC)
3166 return -EACCES;
3167 break;
3168 case S_IFBLK:
3169 case S_IFCHR:
3170 if (!may_open_dev(path))
3171 return -EACCES;
3172 fallthrough;
3173 case S_IFIFO:
3174 case S_IFSOCK:
3175 if (acc_mode & MAY_EXEC)
3176 return -EACCES;
3177 flag &= ~O_TRUNC;
3178 break;
3179 case S_IFREG:
3180 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3181 return -EACCES;
3182 break;
3183 }
3184
3185 error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode);
3186 if (error)
3187 return error;
3188
3189 /*
3190 * An append-only file must be opened in append mode for writing.
3191 */
3192 if (IS_APPEND(inode)) {
3193 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3194 return -EPERM;
3195 if (flag & O_TRUNC)
3196 return -EPERM;
3197 }
3198
3199 /* O_NOATIME can only be set by the owner or superuser */
3200 if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
3201 return -EPERM;
3202
3203 return 0;
3204}
3205
3206static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp)
3207{
3208 const struct path *path = &filp->f_path;
3209 struct inode *inode = path->dentry->d_inode;
3210 int error = get_write_access(inode);
3211 if (error)
3212 return error;
3213
3214 error = security_file_truncate(filp);
3215 if (!error) {
3216 error = do_truncate(mnt_userns, path->dentry, 0,
3217 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3218 filp);
3219 }
3220 put_write_access(inode);
3221 return error;
3222}
3223
3224static inline int open_to_namei_flags(int flag)
3225{
3226 if ((flag & O_ACCMODE) == 3)
3227 flag--;
3228 return flag;
3229}
3230
3231static int may_o_create(struct user_namespace *mnt_userns,
3232 const struct path *dir, struct dentry *dentry,
3233 umode_t mode)
3234{
3235 int error = security_path_mknod(dir, dentry, mode, 0);
3236 if (error)
3237 return error;
3238
3239 if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns))
3240 return -EOVERFLOW;
3241
3242 error = inode_permission(mnt_userns, dir->dentry->d_inode,
3243 MAY_WRITE | MAY_EXEC);
3244 if (error)
3245 return error;
3246
3247 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3248}
3249
3250/*
3251 * Attempt to atomically look up, create and open a file from a negative
3252 * dentry.
3253 *
3254 * Returns 0 if successful. The file will have been created and attached to
3255 * @file by the filesystem calling finish_open().
3256 *
3257 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3258 * be set. The caller will need to perform the open themselves. @path will
3259 * have been updated to point to the new dentry. This may be negative.
3260 *
3261 * Returns an error code otherwise.
3262 */
3263static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3264 struct file *file,
3265 int open_flag, umode_t mode)
3266{
3267 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3268 struct inode *dir = nd->path.dentry->d_inode;
3269 int error;
3270
3271 if (nd->flags & LOOKUP_DIRECTORY)
3272 open_flag |= O_DIRECTORY;
3273
3274 file->f_path.dentry = DENTRY_NOT_SET;
3275 file->f_path.mnt = nd->path.mnt;
3276 error = dir->i_op->atomic_open(dir, dentry, file,
3277 open_to_namei_flags(open_flag), mode);
3278 d_lookup_done(dentry);
3279 if (!error) {
3280 if (file->f_mode & FMODE_OPENED) {
3281 if (unlikely(dentry != file->f_path.dentry)) {
3282 dput(dentry);
3283 dentry = dget(file->f_path.dentry);
3284 }
3285 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3286 error = -EIO;
3287 } else {
3288 if (file->f_path.dentry) {
3289 dput(dentry);
3290 dentry = file->f_path.dentry;
3291 }
3292 if (unlikely(d_is_negative(dentry)))
3293 error = -ENOENT;
3294 }
3295 }
3296 if (error) {
3297 dput(dentry);
3298 dentry = ERR_PTR(error);
3299 }
3300 return dentry;
3301}
3302
3303/*
3304 * Look up and maybe create and open the last component.
3305 *
3306 * Must be called with parent locked (exclusive in O_CREAT case).
3307 *
3308 * Returns 0 on success, that is, if
3309 * the file was successfully atomically created (if necessary) and opened, or
3310 * the file was not completely opened at this time, though lookups and
3311 * creations were performed.
3312 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3313 * In the latter case dentry returned in @path might be negative if O_CREAT
3314 * hadn't been specified.
3315 *
3316 * An error code is returned on failure.
3317 */
3318static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3319 const struct open_flags *op,
3320 bool got_write)
3321{
3322 struct user_namespace *mnt_userns;
3323 struct dentry *dir = nd->path.dentry;
3324 struct inode *dir_inode = dir->d_inode;
3325 int open_flag = op->open_flag;
3326 struct dentry *dentry;
3327 int error, create_error = 0;
3328 umode_t mode = op->mode;
3329 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3330
3331 if (unlikely(IS_DEADDIR(dir_inode)))
3332 return ERR_PTR(-ENOENT);
3333
3334 file->f_mode &= ~FMODE_CREATED;
3335 dentry = d_lookup(dir, &nd->last);
3336 for (;;) {
3337 if (!dentry) {
3338 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3339 if (IS_ERR(dentry))
3340 return dentry;
3341 }
3342 if (d_in_lookup(dentry))
3343 break;
3344
3345 error = d_revalidate(dentry, nd->flags);
3346 if (likely(error > 0))
3347 break;
3348 if (error)
3349 goto out_dput;
3350 d_invalidate(dentry);
3351 dput(dentry);
3352 dentry = NULL;
3353 }
3354 if (dentry->d_inode) {
3355 /* Cached positive dentry: will open in f_op->open */
3356 return dentry;
3357 }
3358
3359 /*
3360 * Checking write permission is tricky, bacuse we don't know if we are
3361 * going to actually need it: O_CREAT opens should work as long as the
3362 * file exists. But checking existence breaks atomicity. The trick is
3363 * to check access and if not granted clear O_CREAT from the flags.
3364 *
3365 * Another problem is returing the "right" error value (e.g. for an
3366 * O_EXCL open we want to return EEXIST not EROFS).
3367 */
3368 if (unlikely(!got_write))
3369 open_flag &= ~O_TRUNC;
3370 mnt_userns = mnt_user_ns(nd->path.mnt);
3371 if (open_flag & O_CREAT) {
3372 if (open_flag & O_EXCL)
3373 open_flag &= ~O_TRUNC;
3374 mode = vfs_prepare_mode(mnt_userns, dir->d_inode, mode, mode, mode);
3375 if (likely(got_write))
3376 create_error = may_o_create(mnt_userns, &nd->path,
3377 dentry, mode);
3378 else
3379 create_error = -EROFS;
3380 }
3381 if (create_error)
3382 open_flag &= ~O_CREAT;
3383 if (dir_inode->i_op->atomic_open) {
3384 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3385 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3386 dentry = ERR_PTR(create_error);
3387 return dentry;
3388 }
3389
3390 if (d_in_lookup(dentry)) {
3391 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3392 nd->flags);
3393 d_lookup_done(dentry);
3394 if (unlikely(res)) {
3395 if (IS_ERR(res)) {
3396 error = PTR_ERR(res);
3397 goto out_dput;
3398 }
3399 dput(dentry);
3400 dentry = res;
3401 }
3402 }
3403
3404 /* Negative dentry, just create the file */
3405 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3406 file->f_mode |= FMODE_CREATED;
3407 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3408 if (!dir_inode->i_op->create) {
3409 error = -EACCES;
3410 goto out_dput;
3411 }
3412
3413 error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry,
3414 mode, open_flag & O_EXCL);
3415 if (error)
3416 goto out_dput;
3417 }
3418 if (unlikely(create_error) && !dentry->d_inode) {
3419 error = create_error;
3420 goto out_dput;
3421 }
3422 return dentry;
3423
3424out_dput:
3425 dput(dentry);
3426 return ERR_PTR(error);
3427}
3428
3429static const char *open_last_lookups(struct nameidata *nd,
3430 struct file *file, const struct open_flags *op)
3431{
3432 struct dentry *dir = nd->path.dentry;
3433 int open_flag = op->open_flag;
3434 bool got_write = false;
3435 struct dentry *dentry;
3436 const char *res;
3437
3438 nd->flags |= op->intent;
3439
3440 if (nd->last_type != LAST_NORM) {
3441 if (nd->depth)
3442 put_link(nd);
3443 return handle_dots(nd, nd->last_type);
3444 }
3445
3446 if (!(open_flag & O_CREAT)) {
3447 if (nd->last.name[nd->last.len])
3448 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3449 /* we _can_ be in RCU mode here */
3450 dentry = lookup_fast(nd);
3451 if (IS_ERR(dentry))
3452 return ERR_CAST(dentry);
3453 if (likely(dentry))
3454 goto finish_lookup;
3455
3456 BUG_ON(nd->flags & LOOKUP_RCU);
3457 } else {
3458 /* create side of things */
3459 if (nd->flags & LOOKUP_RCU) {
3460 if (!try_to_unlazy(nd))
3461 return ERR_PTR(-ECHILD);
3462 }
3463 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3464 /* trailing slashes? */
3465 if (unlikely(nd->last.name[nd->last.len]))
3466 return ERR_PTR(-EISDIR);
3467 }
3468
3469 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3470 got_write = !mnt_want_write(nd->path.mnt);
3471 /*
3472 * do _not_ fail yet - we might not need that or fail with
3473 * a different error; let lookup_open() decide; we'll be
3474 * dropping this one anyway.
3475 */
3476 }
3477 if (open_flag & O_CREAT)
3478 inode_lock(dir->d_inode);
3479 else
3480 inode_lock_shared(dir->d_inode);
3481 dentry = lookup_open(nd, file, op, got_write);
3482 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3483 fsnotify_create(dir->d_inode, dentry);
3484 if (open_flag & O_CREAT)
3485 inode_unlock(dir->d_inode);
3486 else
3487 inode_unlock_shared(dir->d_inode);
3488
3489 if (got_write)
3490 mnt_drop_write(nd->path.mnt);
3491
3492 if (IS_ERR(dentry))
3493 return ERR_CAST(dentry);
3494
3495 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3496 dput(nd->path.dentry);
3497 nd->path.dentry = dentry;
3498 return NULL;
3499 }
3500
3501finish_lookup:
3502 if (nd->depth)
3503 put_link(nd);
3504 res = step_into(nd, WALK_TRAILING, dentry);
3505 if (unlikely(res))
3506 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3507 return res;
3508}
3509
3510/*
3511 * Handle the last step of open()
3512 */
3513static int do_open(struct nameidata *nd,
3514 struct file *file, const struct open_flags *op)
3515{
3516 struct user_namespace *mnt_userns;
3517 int open_flag = op->open_flag;
3518 bool do_truncate;
3519 int acc_mode;
3520 int error;
3521
3522 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3523 error = complete_walk(nd);
3524 if (error)
3525 return error;
3526 }
3527 if (!(file->f_mode & FMODE_CREATED))
3528 audit_inode(nd->name, nd->path.dentry, 0);
3529 mnt_userns = mnt_user_ns(nd->path.mnt);
3530 if (open_flag & O_CREAT) {
3531 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3532 return -EEXIST;
3533 if (d_is_dir(nd->path.dentry))
3534 return -EISDIR;
3535 error = may_create_in_sticky(mnt_userns, nd,
3536 d_backing_inode(nd->path.dentry));
3537 if (unlikely(error))
3538 return error;
3539 }
3540 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3541 return -ENOTDIR;
3542
3543 do_truncate = false;
3544 acc_mode = op->acc_mode;
3545 if (file->f_mode & FMODE_CREATED) {
3546 /* Don't check for write permission, don't truncate */
3547 open_flag &= ~O_TRUNC;
3548 acc_mode = 0;
3549 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3550 error = mnt_want_write(nd->path.mnt);
3551 if (error)
3552 return error;
3553 do_truncate = true;
3554 }
3555 error = may_open(mnt_userns, &nd->path, acc_mode, open_flag);
3556 if (!error && !(file->f_mode & FMODE_OPENED))
3557 error = vfs_open(&nd->path, file);
3558 if (!error)
3559 error = ima_file_check(file, op->acc_mode);
3560 if (!error && do_truncate)
3561 error = handle_truncate(mnt_userns, file);
3562 if (unlikely(error > 0)) {
3563 WARN_ON(1);
3564 error = -EINVAL;
3565 }
3566 if (do_truncate)
3567 mnt_drop_write(nd->path.mnt);
3568 return error;
3569}
3570
3571/**
3572 * vfs_tmpfile - create tmpfile
3573 * @mnt_userns: user namespace of the mount the inode was found from
3574 * @dentry: pointer to dentry of the base directory
3575 * @mode: mode of the new tmpfile
3576 * @open_flag: flags
3577 *
3578 * Create a temporary file.
3579 *
3580 * If the inode has been found through an idmapped mount the user namespace of
3581 * the vfsmount must be passed through @mnt_userns. This function will then take
3582 * care to map the inode according to @mnt_userns before checking permissions.
3583 * On non-idmapped mounts or if permission checking is to be performed on the
3584 * raw inode simply passs init_user_ns.
3585 */
3586static int vfs_tmpfile(struct user_namespace *mnt_userns,
3587 const struct path *parentpath,
3588 struct file *file, umode_t mode)
3589{
3590 struct dentry *child;
3591 struct inode *dir = d_inode(parentpath->dentry);
3592 struct inode *inode;
3593 int error;
3594 int open_flag = file->f_flags;
3595
3596 /* we want directory to be writable */
3597 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
3598 if (error)
3599 return error;
3600 if (!dir->i_op->tmpfile)
3601 return -EOPNOTSUPP;
3602 child = d_alloc(parentpath->dentry, &slash_name);
3603 if (unlikely(!child))
3604 return -ENOMEM;
3605 file->f_path.mnt = parentpath->mnt;
3606 file->f_path.dentry = child;
3607 mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode);
3608 error = dir->i_op->tmpfile(mnt_userns, dir, file, mode);
3609 dput(child);
3610 if (error)
3611 return error;
3612 /* Don't check for other permissions, the inode was just created */
3613 error = may_open(mnt_userns, &file->f_path, 0, file->f_flags);
3614 if (error)
3615 return error;
3616 inode = file_inode(file);
3617 if (!(open_flag & O_EXCL)) {
3618 spin_lock(&inode->i_lock);
3619 inode->i_state |= I_LINKABLE;
3620 spin_unlock(&inode->i_lock);
3621 }
3622 ima_post_create_tmpfile(mnt_userns, inode);
3623 return 0;
3624}
3625
3626/**
3627 * vfs_tmpfile_open - open a tmpfile for kernel internal use
3628 * @mnt_userns: user namespace of the mount the inode was found from
3629 * @parentpath: path of the base directory
3630 * @mode: mode of the new tmpfile
3631 * @open_flag: flags
3632 * @cred: credentials for open
3633 *
3634 * Create and open a temporary file. The file is not accounted in nr_files,
3635 * hence this is only for kernel internal use, and must not be installed into
3636 * file tables or such.
3637 */
3638struct file *vfs_tmpfile_open(struct user_namespace *mnt_userns,
3639 const struct path *parentpath,
3640 umode_t mode, int open_flag, const struct cred *cred)
3641{
3642 struct file *file;
3643 int error;
3644
3645 file = alloc_empty_file_noaccount(open_flag, cred);
3646 if (!IS_ERR(file)) {
3647 error = vfs_tmpfile(mnt_userns, parentpath, file, mode);
3648 if (error) {
3649 fput(file);
3650 file = ERR_PTR(error);
3651 }
3652 }
3653 return file;
3654}
3655EXPORT_SYMBOL(vfs_tmpfile_open);
3656
3657static int do_tmpfile(struct nameidata *nd, unsigned flags,
3658 const struct open_flags *op,
3659 struct file *file)
3660{
3661 struct user_namespace *mnt_userns;
3662 struct path path;
3663 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3664
3665 if (unlikely(error))
3666 return error;
3667 error = mnt_want_write(path.mnt);
3668 if (unlikely(error))
3669 goto out;
3670 mnt_userns = mnt_user_ns(path.mnt);
3671 error = vfs_tmpfile(mnt_userns, &path, file, op->mode);
3672 if (error)
3673 goto out2;
3674 audit_inode(nd->name, file->f_path.dentry, 0);
3675out2:
3676 mnt_drop_write(path.mnt);
3677out:
3678 path_put(&path);
3679 return error;
3680}
3681
3682static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3683{
3684 struct path path;
3685 int error = path_lookupat(nd, flags, &path);
3686 if (!error) {
3687 audit_inode(nd->name, path.dentry, 0);
3688 error = vfs_open(&path, file);
3689 path_put(&path);
3690 }
3691 return error;
3692}
3693
3694static struct file *path_openat(struct nameidata *nd,
3695 const struct open_flags *op, unsigned flags)
3696{
3697 struct file *file;
3698 int error;
3699
3700 file = alloc_empty_file(op->open_flag, current_cred());
3701 if (IS_ERR(file))
3702 return file;
3703
3704 if (unlikely(file->f_flags & __O_TMPFILE)) {
3705 error = do_tmpfile(nd, flags, op, file);
3706 } else if (unlikely(file->f_flags & O_PATH)) {
3707 error = do_o_path(nd, flags, file);
3708 } else {
3709 const char *s = path_init(nd, flags);
3710 while (!(error = link_path_walk(s, nd)) &&
3711 (s = open_last_lookups(nd, file, op)) != NULL)
3712 ;
3713 if (!error)
3714 error = do_open(nd, file, op);
3715 terminate_walk(nd);
3716 }
3717 if (likely(!error)) {
3718 if (likely(file->f_mode & FMODE_OPENED))
3719 return file;
3720 WARN_ON(1);
3721 error = -EINVAL;
3722 }
3723 fput(file);
3724 if (error == -EOPENSTALE) {
3725 if (flags & LOOKUP_RCU)
3726 error = -ECHILD;
3727 else
3728 error = -ESTALE;
3729 }
3730 return ERR_PTR(error);
3731}
3732
3733struct file *do_filp_open(int dfd, struct filename *pathname,
3734 const struct open_flags *op)
3735{
3736 struct nameidata nd;
3737 int flags = op->lookup_flags;
3738 struct file *filp;
3739
3740 set_nameidata(&nd, dfd, pathname, NULL);
3741 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3742 if (unlikely(filp == ERR_PTR(-ECHILD)))
3743 filp = path_openat(&nd, op, flags);
3744 if (unlikely(filp == ERR_PTR(-ESTALE)))
3745 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3746 restore_nameidata();
3747 return filp;
3748}
3749
3750struct file *do_file_open_root(const struct path *root,
3751 const char *name, const struct open_flags *op)
3752{
3753 struct nameidata nd;
3754 struct file *file;
3755 struct filename *filename;
3756 int flags = op->lookup_flags;
3757
3758 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3759 return ERR_PTR(-ELOOP);
3760
3761 filename = getname_kernel(name);
3762 if (IS_ERR(filename))
3763 return ERR_CAST(filename);
3764
3765 set_nameidata(&nd, -1, filename, root);
3766 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3767 if (unlikely(file == ERR_PTR(-ECHILD)))
3768 file = path_openat(&nd, op, flags);
3769 if (unlikely(file == ERR_PTR(-ESTALE)))
3770 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3771 restore_nameidata();
3772 putname(filename);
3773 return file;
3774}
3775
3776static struct dentry *filename_create(int dfd, struct filename *name,
3777 struct path *path, unsigned int lookup_flags)
3778{
3779 struct dentry *dentry = ERR_PTR(-EEXIST);
3780 struct qstr last;
3781 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3782 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3783 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3784 int type;
3785 int err2;
3786 int error;
3787
3788 error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3789 if (error)
3790 return ERR_PTR(error);
3791
3792 /*
3793 * Yucky last component or no last component at all?
3794 * (foo/., foo/.., /////)
3795 */
3796 if (unlikely(type != LAST_NORM))
3797 goto out;
3798
3799 /* don't fail immediately if it's r/o, at least try to report other errors */
3800 err2 = mnt_want_write(path->mnt);
3801 /*
3802 * Do the final lookup. Suppress 'create' if there is a trailing
3803 * '/', and a directory wasn't requested.
3804 */
3805 if (last.name[last.len] && !want_dir)
3806 create_flags = 0;
3807 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3808 dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags);
3809 if (IS_ERR(dentry))
3810 goto unlock;
3811
3812 error = -EEXIST;
3813 if (d_is_positive(dentry))
3814 goto fail;
3815
3816 /*
3817 * Special case - lookup gave negative, but... we had foo/bar/
3818 * From the vfs_mknod() POV we just have a negative dentry -
3819 * all is fine. Let's be bastards - you had / on the end, you've
3820 * been asking for (non-existent) directory. -ENOENT for you.
3821 */
3822 if (unlikely(!create_flags)) {
3823 error = -ENOENT;
3824 goto fail;
3825 }
3826 if (unlikely(err2)) {
3827 error = err2;
3828 goto fail;
3829 }
3830 return dentry;
3831fail:
3832 dput(dentry);
3833 dentry = ERR_PTR(error);
3834unlock:
3835 inode_unlock(path->dentry->d_inode);
3836 if (!err2)
3837 mnt_drop_write(path->mnt);
3838out:
3839 path_put(path);
3840 return dentry;
3841}
3842
3843struct dentry *kern_path_create(int dfd, const char *pathname,
3844 struct path *path, unsigned int lookup_flags)
3845{
3846 struct filename *filename = getname_kernel(pathname);
3847 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3848
3849 putname(filename);
3850 return res;
3851}
3852EXPORT_SYMBOL(kern_path_create);
3853
3854void done_path_create(struct path *path, struct dentry *dentry)
3855{
3856 dput(dentry);
3857 inode_unlock(path->dentry->d_inode);
3858 mnt_drop_write(path->mnt);
3859 path_put(path);
3860}
3861EXPORT_SYMBOL(done_path_create);
3862
3863inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3864 struct path *path, unsigned int lookup_flags)
3865{
3866 struct filename *filename = getname(pathname);
3867 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3868
3869 putname(filename);
3870 return res;
3871}
3872EXPORT_SYMBOL(user_path_create);
3873
3874/**
3875 * vfs_mknod - create device node or file
3876 * @mnt_userns: user namespace of the mount the inode was found from
3877 * @dir: inode of @dentry
3878 * @dentry: pointer to dentry of the base directory
3879 * @mode: mode of the new device node or file
3880 * @dev: device number of device to create
3881 *
3882 * Create a device node or file.
3883 *
3884 * If the inode has been found through an idmapped mount the user namespace of
3885 * the vfsmount must be passed through @mnt_userns. This function will then take
3886 * care to map the inode according to @mnt_userns before checking permissions.
3887 * On non-idmapped mounts or if permission checking is to be performed on the
3888 * raw inode simply passs init_user_ns.
3889 */
3890int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
3891 struct dentry *dentry, umode_t mode, dev_t dev)
3892{
3893 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3894 int error = may_create(mnt_userns, dir, dentry);
3895
3896 if (error)
3897 return error;
3898
3899 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3900 !capable(CAP_MKNOD))
3901 return -EPERM;
3902
3903 if (!dir->i_op->mknod)
3904 return -EPERM;
3905
3906 mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode);
3907 error = devcgroup_inode_mknod(mode, dev);
3908 if (error)
3909 return error;
3910
3911 error = security_inode_mknod(dir, dentry, mode, dev);
3912 if (error)
3913 return error;
3914
3915 error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev);
3916 if (!error)
3917 fsnotify_create(dir, dentry);
3918 return error;
3919}
3920EXPORT_SYMBOL(vfs_mknod);
3921
3922static int may_mknod(umode_t mode)
3923{
3924 switch (mode & S_IFMT) {
3925 case S_IFREG:
3926 case S_IFCHR:
3927 case S_IFBLK:
3928 case S_IFIFO:
3929 case S_IFSOCK:
3930 case 0: /* zero mode translates to S_IFREG */
3931 return 0;
3932 case S_IFDIR:
3933 return -EPERM;
3934 default:
3935 return -EINVAL;
3936 }
3937}
3938
3939static int do_mknodat(int dfd, struct filename *name, umode_t mode,
3940 unsigned int dev)
3941{
3942 struct user_namespace *mnt_userns;
3943 struct dentry *dentry;
3944 struct path path;
3945 int error;
3946 unsigned int lookup_flags = 0;
3947
3948 error = may_mknod(mode);
3949 if (error)
3950 goto out1;
3951retry:
3952 dentry = filename_create(dfd, name, &path, lookup_flags);
3953 error = PTR_ERR(dentry);
3954 if (IS_ERR(dentry))
3955 goto out1;
3956
3957 error = security_path_mknod(&path, dentry,
3958 mode_strip_umask(path.dentry->d_inode, mode), dev);
3959 if (error)
3960 goto out2;
3961
3962 mnt_userns = mnt_user_ns(path.mnt);
3963 switch (mode & S_IFMT) {
3964 case 0: case S_IFREG:
3965 error = vfs_create(mnt_userns, path.dentry->d_inode,
3966 dentry, mode, true);
3967 if (!error)
3968 ima_post_path_mknod(mnt_userns, dentry);
3969 break;
3970 case S_IFCHR: case S_IFBLK:
3971 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3972 dentry, mode, new_decode_dev(dev));
3973 break;
3974 case S_IFIFO: case S_IFSOCK:
3975 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3976 dentry, mode, 0);
3977 break;
3978 }
3979out2:
3980 done_path_create(&path, dentry);
3981 if (retry_estale(error, lookup_flags)) {
3982 lookup_flags |= LOOKUP_REVAL;
3983 goto retry;
3984 }
3985out1:
3986 putname(name);
3987 return error;
3988}
3989
3990SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3991 unsigned int, dev)
3992{
3993 return do_mknodat(dfd, getname(filename), mode, dev);
3994}
3995
3996SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3997{
3998 return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
3999}
4000
4001/**
4002 * vfs_mkdir - create directory
4003 * @mnt_userns: user namespace of the mount the inode was found from
4004 * @dir: inode of @dentry
4005 * @dentry: pointer to dentry of the base directory
4006 * @mode: mode of the new directory
4007 *
4008 * Create a directory.
4009 *
4010 * If the inode has been found through an idmapped mount the user namespace of
4011 * the vfsmount must be passed through @mnt_userns. This function will then take
4012 * care to map the inode according to @mnt_userns before checking permissions.
4013 * On non-idmapped mounts or if permission checking is to be performed on the
4014 * raw inode simply passs init_user_ns.
4015 */
4016int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
4017 struct dentry *dentry, umode_t mode)
4018{
4019 int error = may_create(mnt_userns, dir, dentry);
4020 unsigned max_links = dir->i_sb->s_max_links;
4021
4022 if (error)
4023 return error;
4024
4025 if (!dir->i_op->mkdir)
4026 return -EPERM;
4027
4028 mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4029 error = security_inode_mkdir(dir, dentry, mode);
4030 if (error)
4031 return error;
4032
4033 if (max_links && dir->i_nlink >= max_links)
4034 return -EMLINK;
4035
4036 error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode);
4037 if (!error)
4038 fsnotify_mkdir(dir, dentry);
4039 return error;
4040}
4041EXPORT_SYMBOL(vfs_mkdir);
4042
4043int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4044{
4045 struct dentry *dentry;
4046 struct path path;
4047 int error;
4048 unsigned int lookup_flags = LOOKUP_DIRECTORY;
4049
4050retry:
4051 dentry = filename_create(dfd, name, &path, lookup_flags);
4052 error = PTR_ERR(dentry);
4053 if (IS_ERR(dentry))
4054 goto out_putname;
4055
4056 error = security_path_mkdir(&path, dentry,
4057 mode_strip_umask(path.dentry->d_inode, mode));
4058 if (!error) {
4059 struct user_namespace *mnt_userns;
4060 mnt_userns = mnt_user_ns(path.mnt);
4061 error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry,
4062 mode);
4063 }
4064 done_path_create(&path, dentry);
4065 if (retry_estale(error, lookup_flags)) {
4066 lookup_flags |= LOOKUP_REVAL;
4067 goto retry;
4068 }
4069out_putname:
4070 putname(name);
4071 return error;
4072}
4073
4074SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4075{
4076 return do_mkdirat(dfd, getname(pathname), mode);
4077}
4078
4079SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4080{
4081 return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4082}
4083
4084/**
4085 * vfs_rmdir - remove directory
4086 * @mnt_userns: user namespace of the mount the inode was found from
4087 * @dir: inode of @dentry
4088 * @dentry: pointer to dentry of the base directory
4089 *
4090 * Remove a directory.
4091 *
4092 * If the inode has been found through an idmapped mount the user namespace of
4093 * the vfsmount must be passed through @mnt_userns. This function will then take
4094 * care to map the inode according to @mnt_userns before checking permissions.
4095 * On non-idmapped mounts or if permission checking is to be performed on the
4096 * raw inode simply passs init_user_ns.
4097 */
4098int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir,
4099 struct dentry *dentry)
4100{
4101 int error = may_delete(mnt_userns, dir, dentry, 1);
4102
4103 if (error)
4104 return error;
4105
4106 if (!dir->i_op->rmdir)
4107 return -EPERM;
4108
4109 dget(dentry);
4110 inode_lock(dentry->d_inode);
4111
4112 error = -EBUSY;
4113 if (is_local_mountpoint(dentry) ||
4114 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4115 goto out;
4116
4117 error = security_inode_rmdir(dir, dentry);
4118 if (error)
4119 goto out;
4120
4121 error = dir->i_op->rmdir(dir, dentry);
4122 if (error)
4123 goto out;
4124
4125 shrink_dcache_parent(dentry);
4126 dentry->d_inode->i_flags |= S_DEAD;
4127 dont_mount(dentry);
4128 detach_mounts(dentry);
4129
4130out:
4131 inode_unlock(dentry->d_inode);
4132 dput(dentry);
4133 if (!error)
4134 d_delete_notify(dir, dentry);
4135 return error;
4136}
4137EXPORT_SYMBOL(vfs_rmdir);
4138
4139int do_rmdir(int dfd, struct filename *name)
4140{
4141 struct user_namespace *mnt_userns;
4142 int error;
4143 struct dentry *dentry;
4144 struct path path;
4145 struct qstr last;
4146 int type;
4147 unsigned int lookup_flags = 0;
4148retry:
4149 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4150 if (error)
4151 goto exit1;
4152
4153 switch (type) {
4154 case LAST_DOTDOT:
4155 error = -ENOTEMPTY;
4156 goto exit2;
4157 case LAST_DOT:
4158 error = -EINVAL;
4159 goto exit2;
4160 case LAST_ROOT:
4161 error = -EBUSY;
4162 goto exit2;
4163 }
4164
4165 error = mnt_want_write(path.mnt);
4166 if (error)
4167 goto exit2;
4168
4169 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4170 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4171 error = PTR_ERR(dentry);
4172 if (IS_ERR(dentry))
4173 goto exit3;
4174 if (!dentry->d_inode) {
4175 error = -ENOENT;
4176 goto exit4;
4177 }
4178 error = security_path_rmdir(&path, dentry);
4179 if (error)
4180 goto exit4;
4181 mnt_userns = mnt_user_ns(path.mnt);
4182 error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry);
4183exit4:
4184 dput(dentry);
4185exit3:
4186 inode_unlock(path.dentry->d_inode);
4187 mnt_drop_write(path.mnt);
4188exit2:
4189 path_put(&path);
4190 if (retry_estale(error, lookup_flags)) {
4191 lookup_flags |= LOOKUP_REVAL;
4192 goto retry;
4193 }
4194exit1:
4195 putname(name);
4196 return error;
4197}
4198
4199SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4200{
4201 return do_rmdir(AT_FDCWD, getname(pathname));
4202}
4203
4204/**
4205 * vfs_unlink - unlink a filesystem object
4206 * @mnt_userns: user namespace of the mount the inode was found from
4207 * @dir: parent directory
4208 * @dentry: victim
4209 * @delegated_inode: returns victim inode, if the inode is delegated.
4210 *
4211 * The caller must hold dir->i_mutex.
4212 *
4213 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4214 * return a reference to the inode in delegated_inode. The caller
4215 * should then break the delegation on that inode and retry. Because
4216 * breaking a delegation may take a long time, the caller should drop
4217 * dir->i_mutex before doing so.
4218 *
4219 * Alternatively, a caller may pass NULL for delegated_inode. This may
4220 * be appropriate for callers that expect the underlying filesystem not
4221 * to be NFS exported.
4222 *
4223 * If the inode has been found through an idmapped mount the user namespace of
4224 * the vfsmount must be passed through @mnt_userns. This function will then take
4225 * care to map the inode according to @mnt_userns before checking permissions.
4226 * On non-idmapped mounts or if permission checking is to be performed on the
4227 * raw inode simply passs init_user_ns.
4228 */
4229int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir,
4230 struct dentry *dentry, struct inode **delegated_inode)
4231{
4232 struct inode *target = dentry->d_inode;
4233 int error = may_delete(mnt_userns, dir, dentry, 0);
4234
4235 if (error)
4236 return error;
4237
4238 if (!dir->i_op->unlink)
4239 return -EPERM;
4240
4241 inode_lock(target);
4242 if (IS_SWAPFILE(target))
4243 error = -EPERM;
4244 else if (is_local_mountpoint(dentry))
4245 error = -EBUSY;
4246 else {
4247 error = security_inode_unlink(dir, dentry);
4248 if (!error) {
4249 error = try_break_deleg(target, delegated_inode);
4250 if (error)
4251 goto out;
4252 error = dir->i_op->unlink(dir, dentry);
4253 if (!error) {
4254 dont_mount(dentry);
4255 detach_mounts(dentry);
4256 }
4257 }
4258 }
4259out:
4260 inode_unlock(target);
4261
4262 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4263 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4264 fsnotify_unlink(dir, dentry);
4265 } else if (!error) {
4266 fsnotify_link_count(target);
4267 d_delete_notify(dir, dentry);
4268 }
4269
4270 return error;
4271}
4272EXPORT_SYMBOL(vfs_unlink);
4273
4274/*
4275 * Make sure that the actual truncation of the file will occur outside its
4276 * directory's i_mutex. Truncate can take a long time if there is a lot of
4277 * writeout happening, and we don't want to prevent access to the directory
4278 * while waiting on the I/O.
4279 */
4280int do_unlinkat(int dfd, struct filename *name)
4281{
4282 int error;
4283 struct dentry *dentry;
4284 struct path path;
4285 struct qstr last;
4286 int type;
4287 struct inode *inode = NULL;
4288 struct inode *delegated_inode = NULL;
4289 unsigned int lookup_flags = 0;
4290retry:
4291 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4292 if (error)
4293 goto exit1;
4294
4295 error = -EISDIR;
4296 if (type != LAST_NORM)
4297 goto exit2;
4298
4299 error = mnt_want_write(path.mnt);
4300 if (error)
4301 goto exit2;
4302retry_deleg:
4303 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4304 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4305 error = PTR_ERR(dentry);
4306 if (!IS_ERR(dentry)) {
4307 struct user_namespace *mnt_userns;
4308
4309 /* Why not before? Because we want correct error value */
4310 if (last.name[last.len])
4311 goto slashes;
4312 inode = dentry->d_inode;
4313 if (d_is_negative(dentry))
4314 goto slashes;
4315 ihold(inode);
4316 error = security_path_unlink(&path, dentry);
4317 if (error)
4318 goto exit3;
4319 mnt_userns = mnt_user_ns(path.mnt);
4320 error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry,
4321 &delegated_inode);
4322exit3:
4323 dput(dentry);
4324 }
4325 inode_unlock(path.dentry->d_inode);
4326 if (inode)
4327 iput(inode); /* truncate the inode here */
4328 inode = NULL;
4329 if (delegated_inode) {
4330 error = break_deleg_wait(&delegated_inode);
4331 if (!error)
4332 goto retry_deleg;
4333 }
4334 mnt_drop_write(path.mnt);
4335exit2:
4336 path_put(&path);
4337 if (retry_estale(error, lookup_flags)) {
4338 lookup_flags |= LOOKUP_REVAL;
4339 inode = NULL;
4340 goto retry;
4341 }
4342exit1:
4343 putname(name);
4344 return error;
4345
4346slashes:
4347 if (d_is_negative(dentry))
4348 error = -ENOENT;
4349 else if (d_is_dir(dentry))
4350 error = -EISDIR;
4351 else
4352 error = -ENOTDIR;
4353 goto exit3;
4354}
4355
4356SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4357{
4358 if ((flag & ~AT_REMOVEDIR) != 0)
4359 return -EINVAL;
4360
4361 if (flag & AT_REMOVEDIR)
4362 return do_rmdir(dfd, getname(pathname));
4363 return do_unlinkat(dfd, getname(pathname));
4364}
4365
4366SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4367{
4368 return do_unlinkat(AT_FDCWD, getname(pathname));
4369}
4370
4371/**
4372 * vfs_symlink - create symlink
4373 * @mnt_userns: user namespace of the mount the inode was found from
4374 * @dir: inode of @dentry
4375 * @dentry: pointer to dentry of the base directory
4376 * @oldname: name of the file to link to
4377 *
4378 * Create a symlink.
4379 *
4380 * If the inode has been found through an idmapped mount the user namespace of
4381 * the vfsmount must be passed through @mnt_userns. This function will then take
4382 * care to map the inode according to @mnt_userns before checking permissions.
4383 * On non-idmapped mounts or if permission checking is to be performed on the
4384 * raw inode simply passs init_user_ns.
4385 */
4386int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
4387 struct dentry *dentry, const char *oldname)
4388{
4389 int error = may_create(mnt_userns, dir, dentry);
4390
4391 if (error)
4392 return error;
4393
4394 if (!dir->i_op->symlink)
4395 return -EPERM;
4396
4397 error = security_inode_symlink(dir, dentry, oldname);
4398 if (error)
4399 return error;
4400
4401 error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname);
4402 if (!error)
4403 fsnotify_create(dir, dentry);
4404 return error;
4405}
4406EXPORT_SYMBOL(vfs_symlink);
4407
4408int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4409{
4410 int error;
4411 struct dentry *dentry;
4412 struct path path;
4413 unsigned int lookup_flags = 0;
4414
4415 if (IS_ERR(from)) {
4416 error = PTR_ERR(from);
4417 goto out_putnames;
4418 }
4419retry:
4420 dentry = filename_create(newdfd, to, &path, lookup_flags);
4421 error = PTR_ERR(dentry);
4422 if (IS_ERR(dentry))
4423 goto out_putnames;
4424
4425 error = security_path_symlink(&path, dentry, from->name);
4426 if (!error) {
4427 struct user_namespace *mnt_userns;
4428
4429 mnt_userns = mnt_user_ns(path.mnt);
4430 error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry,
4431 from->name);
4432 }
4433 done_path_create(&path, dentry);
4434 if (retry_estale(error, lookup_flags)) {
4435 lookup_flags |= LOOKUP_REVAL;
4436 goto retry;
4437 }
4438out_putnames:
4439 putname(to);
4440 putname(from);
4441 return error;
4442}
4443
4444SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4445 int, newdfd, const char __user *, newname)
4446{
4447 return do_symlinkat(getname(oldname), newdfd, getname(newname));
4448}
4449
4450SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4451{
4452 return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4453}
4454
4455/**
4456 * vfs_link - create a new link
4457 * @old_dentry: object to be linked
4458 * @mnt_userns: the user namespace of the mount
4459 * @dir: new parent
4460 * @new_dentry: where to create the new link
4461 * @delegated_inode: returns inode needing a delegation break
4462 *
4463 * The caller must hold dir->i_mutex
4464 *
4465 * If vfs_link discovers a delegation on the to-be-linked file in need
4466 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4467 * inode in delegated_inode. The caller should then break the delegation
4468 * and retry. Because breaking a delegation may take a long time, the
4469 * caller should drop the i_mutex before doing so.
4470 *
4471 * Alternatively, a caller may pass NULL for delegated_inode. This may
4472 * be appropriate for callers that expect the underlying filesystem not
4473 * to be NFS exported.
4474 *
4475 * If the inode has been found through an idmapped mount the user namespace of
4476 * the vfsmount must be passed through @mnt_userns. This function will then take
4477 * care to map the inode according to @mnt_userns before checking permissions.
4478 * On non-idmapped mounts or if permission checking is to be performed on the
4479 * raw inode simply passs init_user_ns.
4480 */
4481int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns,
4482 struct inode *dir, struct dentry *new_dentry,
4483 struct inode **delegated_inode)
4484{
4485 struct inode *inode = old_dentry->d_inode;
4486 unsigned max_links = dir->i_sb->s_max_links;
4487 int error;
4488
4489 if (!inode)
4490 return -ENOENT;
4491
4492 error = may_create(mnt_userns, dir, new_dentry);
4493 if (error)
4494 return error;
4495
4496 if (dir->i_sb != inode->i_sb)
4497 return -EXDEV;
4498
4499 /*
4500 * A link to an append-only or immutable file cannot be created.
4501 */
4502 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4503 return -EPERM;
4504 /*
4505 * Updating the link count will likely cause i_uid and i_gid to
4506 * be writen back improperly if their true value is unknown to
4507 * the vfs.
4508 */
4509 if (HAS_UNMAPPED_ID(mnt_userns, inode))
4510 return -EPERM;
4511 if (!dir->i_op->link)
4512 return -EPERM;
4513 if (S_ISDIR(inode->i_mode))
4514 return -EPERM;
4515
4516 error = security_inode_link(old_dentry, dir, new_dentry);
4517 if (error)
4518 return error;
4519
4520 inode_lock(inode);
4521 /* Make sure we don't allow creating hardlink to an unlinked file */
4522 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4523 error = -ENOENT;
4524 else if (max_links && inode->i_nlink >= max_links)
4525 error = -EMLINK;
4526 else {
4527 error = try_break_deleg(inode, delegated_inode);
4528 if (!error)
4529 error = dir->i_op->link(old_dentry, dir, new_dentry);
4530 }
4531
4532 if (!error && (inode->i_state & I_LINKABLE)) {
4533 spin_lock(&inode->i_lock);
4534 inode->i_state &= ~I_LINKABLE;
4535 spin_unlock(&inode->i_lock);
4536 }
4537 inode_unlock(inode);
4538 if (!error)
4539 fsnotify_link(dir, inode, new_dentry);
4540 return error;
4541}
4542EXPORT_SYMBOL(vfs_link);
4543
4544/*
4545 * Hardlinks are often used in delicate situations. We avoid
4546 * security-related surprises by not following symlinks on the
4547 * newname. --KAB
4548 *
4549 * We don't follow them on the oldname either to be compatible
4550 * with linux 2.0, and to avoid hard-linking to directories
4551 * and other special files. --ADM
4552 */
4553int do_linkat(int olddfd, struct filename *old, int newdfd,
4554 struct filename *new, int flags)
4555{
4556 struct user_namespace *mnt_userns;
4557 struct dentry *new_dentry;
4558 struct path old_path, new_path;
4559 struct inode *delegated_inode = NULL;
4560 int how = 0;
4561 int error;
4562
4563 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4564 error = -EINVAL;
4565 goto out_putnames;
4566 }
4567 /*
4568 * To use null names we require CAP_DAC_READ_SEARCH
4569 * This ensures that not everyone will be able to create
4570 * handlink using the passed filedescriptor.
4571 */
4572 if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4573 error = -ENOENT;
4574 goto out_putnames;
4575 }
4576
4577 if (flags & AT_SYMLINK_FOLLOW)
4578 how |= LOOKUP_FOLLOW;
4579retry:
4580 error = filename_lookup(olddfd, old, how, &old_path, NULL);
4581 if (error)
4582 goto out_putnames;
4583
4584 new_dentry = filename_create(newdfd, new, &new_path,
4585 (how & LOOKUP_REVAL));
4586 error = PTR_ERR(new_dentry);
4587 if (IS_ERR(new_dentry))
4588 goto out_putpath;
4589
4590 error = -EXDEV;
4591 if (old_path.mnt != new_path.mnt)
4592 goto out_dput;
4593 mnt_userns = mnt_user_ns(new_path.mnt);
4594 error = may_linkat(mnt_userns, &old_path);
4595 if (unlikely(error))
4596 goto out_dput;
4597 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4598 if (error)
4599 goto out_dput;
4600 error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode,
4601 new_dentry, &delegated_inode);
4602out_dput:
4603 done_path_create(&new_path, new_dentry);
4604 if (delegated_inode) {
4605 error = break_deleg_wait(&delegated_inode);
4606 if (!error) {
4607 path_put(&old_path);
4608 goto retry;
4609 }
4610 }
4611 if (retry_estale(error, how)) {
4612 path_put(&old_path);
4613 how |= LOOKUP_REVAL;
4614 goto retry;
4615 }
4616out_putpath:
4617 path_put(&old_path);
4618out_putnames:
4619 putname(old);
4620 putname(new);
4621
4622 return error;
4623}
4624
4625SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4626 int, newdfd, const char __user *, newname, int, flags)
4627{
4628 return do_linkat(olddfd, getname_uflags(oldname, flags),
4629 newdfd, getname(newname), flags);
4630}
4631
4632SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4633{
4634 return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4635}
4636
4637/**
4638 * vfs_rename - rename a filesystem object
4639 * @rd: pointer to &struct renamedata info
4640 *
4641 * The caller must hold multiple mutexes--see lock_rename()).
4642 *
4643 * If vfs_rename discovers a delegation in need of breaking at either
4644 * the source or destination, it will return -EWOULDBLOCK and return a
4645 * reference to the inode in delegated_inode. The caller should then
4646 * break the delegation and retry. Because breaking a delegation may
4647 * take a long time, the caller should drop all locks before doing
4648 * so.
4649 *
4650 * Alternatively, a caller may pass NULL for delegated_inode. This may
4651 * be appropriate for callers that expect the underlying filesystem not
4652 * to be NFS exported.
4653 *
4654 * The worst of all namespace operations - renaming directory. "Perverted"
4655 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4656 * Problems:
4657 *
4658 * a) we can get into loop creation.
4659 * b) race potential - two innocent renames can create a loop together.
4660 * That's where 4.4 screws up. Current fix: serialization on
4661 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4662 * story.
4663 * c) we have to lock _four_ objects - parents and victim (if it exists),
4664 * and source (if it is not a directory).
4665 * And that - after we got ->i_mutex on parents (until then we don't know
4666 * whether the target exists). Solution: try to be smart with locking
4667 * order for inodes. We rely on the fact that tree topology may change
4668 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4669 * move will be locked. Thus we can rank directories by the tree
4670 * (ancestors first) and rank all non-directories after them.
4671 * That works since everybody except rename does "lock parent, lookup,
4672 * lock child" and rename is under ->s_vfs_rename_mutex.
4673 * HOWEVER, it relies on the assumption that any object with ->lookup()
4674 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4675 * we'd better make sure that there's no link(2) for them.
4676 * d) conversion from fhandle to dentry may come in the wrong moment - when
4677 * we are removing the target. Solution: we will have to grab ->i_mutex
4678 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4679 * ->i_mutex on parents, which works but leads to some truly excessive
4680 * locking].
4681 */
4682int vfs_rename(struct renamedata *rd)
4683{
4684 int error;
4685 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4686 struct dentry *old_dentry = rd->old_dentry;
4687 struct dentry *new_dentry = rd->new_dentry;
4688 struct inode **delegated_inode = rd->delegated_inode;
4689 unsigned int flags = rd->flags;
4690 bool is_dir = d_is_dir(old_dentry);
4691 struct inode *source = old_dentry->d_inode;
4692 struct inode *target = new_dentry->d_inode;
4693 bool new_is_dir = false;
4694 unsigned max_links = new_dir->i_sb->s_max_links;
4695 struct name_snapshot old_name;
4696
4697 if (source == target)
4698 return 0;
4699
4700 error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir);
4701 if (error)
4702 return error;
4703
4704 if (!target) {
4705 error = may_create(rd->new_mnt_userns, new_dir, new_dentry);
4706 } else {
4707 new_is_dir = d_is_dir(new_dentry);
4708
4709 if (!(flags & RENAME_EXCHANGE))
4710 error = may_delete(rd->new_mnt_userns, new_dir,
4711 new_dentry, is_dir);
4712 else
4713 error = may_delete(rd->new_mnt_userns, new_dir,
4714 new_dentry, new_is_dir);
4715 }
4716 if (error)
4717 return error;
4718
4719 if (!old_dir->i_op->rename)
4720 return -EPERM;
4721
4722 /*
4723 * If we are going to change the parent - check write permissions,
4724 * we'll need to flip '..'.
4725 */
4726 if (new_dir != old_dir) {
4727 if (is_dir) {
4728 error = inode_permission(rd->old_mnt_userns, source,
4729 MAY_WRITE);
4730 if (error)
4731 return error;
4732 }
4733 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4734 error = inode_permission(rd->new_mnt_userns, target,
4735 MAY_WRITE);
4736 if (error)
4737 return error;
4738 }
4739 }
4740
4741 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4742 flags);
4743 if (error)
4744 return error;
4745
4746 take_dentry_name_snapshot(&old_name, old_dentry);
4747 dget(new_dentry);
4748 if (!is_dir || (flags & RENAME_EXCHANGE))
4749 lock_two_nondirectories(source, target);
4750 else if (target)
4751 inode_lock(target);
4752
4753 error = -EPERM;
4754 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4755 goto out;
4756
4757 error = -EBUSY;
4758 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4759 goto out;
4760
4761 if (max_links && new_dir != old_dir) {
4762 error = -EMLINK;
4763 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4764 goto out;
4765 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4766 old_dir->i_nlink >= max_links)
4767 goto out;
4768 }
4769 if (!is_dir) {
4770 error = try_break_deleg(source, delegated_inode);
4771 if (error)
4772 goto out;
4773 }
4774 if (target && !new_is_dir) {
4775 error = try_break_deleg(target, delegated_inode);
4776 if (error)
4777 goto out;
4778 }
4779 error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry,
4780 new_dir, new_dentry, flags);
4781 if (error)
4782 goto out;
4783
4784 if (!(flags & RENAME_EXCHANGE) && target) {
4785 if (is_dir) {
4786 shrink_dcache_parent(new_dentry);
4787 target->i_flags |= S_DEAD;
4788 }
4789 dont_mount(new_dentry);
4790 detach_mounts(new_dentry);
4791 }
4792 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4793 if (!(flags & RENAME_EXCHANGE))
4794 d_move(old_dentry, new_dentry);
4795 else
4796 d_exchange(old_dentry, new_dentry);
4797 }
4798out:
4799 if (!is_dir || (flags & RENAME_EXCHANGE))
4800 unlock_two_nondirectories(source, target);
4801 else if (target)
4802 inode_unlock(target);
4803 dput(new_dentry);
4804 if (!error) {
4805 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4806 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4807 if (flags & RENAME_EXCHANGE) {
4808 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4809 new_is_dir, NULL, new_dentry);
4810 }
4811 }
4812 release_dentry_name_snapshot(&old_name);
4813
4814 return error;
4815}
4816EXPORT_SYMBOL(vfs_rename);
4817
4818int do_renameat2(int olddfd, struct filename *from, int newdfd,
4819 struct filename *to, unsigned int flags)
4820{
4821 struct renamedata rd;
4822 struct dentry *old_dentry, *new_dentry;
4823 struct dentry *trap;
4824 struct path old_path, new_path;
4825 struct qstr old_last, new_last;
4826 int old_type, new_type;
4827 struct inode *delegated_inode = NULL;
4828 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4829 bool should_retry = false;
4830 int error = -EINVAL;
4831
4832 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4833 goto put_names;
4834
4835 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4836 (flags & RENAME_EXCHANGE))
4837 goto put_names;
4838
4839 if (flags & RENAME_EXCHANGE)
4840 target_flags = 0;
4841
4842retry:
4843 error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4844 &old_last, &old_type);
4845 if (error)
4846 goto put_names;
4847
4848 error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4849 &new_type);
4850 if (error)
4851 goto exit1;
4852
4853 error = -EXDEV;
4854 if (old_path.mnt != new_path.mnt)
4855 goto exit2;
4856
4857 error = -EBUSY;
4858 if (old_type != LAST_NORM)
4859 goto exit2;
4860
4861 if (flags & RENAME_NOREPLACE)
4862 error = -EEXIST;
4863 if (new_type != LAST_NORM)
4864 goto exit2;
4865
4866 error = mnt_want_write(old_path.mnt);
4867 if (error)
4868 goto exit2;
4869
4870retry_deleg:
4871 trap = lock_rename(new_path.dentry, old_path.dentry);
4872
4873 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4874 error = PTR_ERR(old_dentry);
4875 if (IS_ERR(old_dentry))
4876 goto exit3;
4877 /* source must exist */
4878 error = -ENOENT;
4879 if (d_is_negative(old_dentry))
4880 goto exit4;
4881 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4882 error = PTR_ERR(new_dentry);
4883 if (IS_ERR(new_dentry))
4884 goto exit4;
4885 error = -EEXIST;
4886 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4887 goto exit5;
4888 if (flags & RENAME_EXCHANGE) {
4889 error = -ENOENT;
4890 if (d_is_negative(new_dentry))
4891 goto exit5;
4892
4893 if (!d_is_dir(new_dentry)) {
4894 error = -ENOTDIR;
4895 if (new_last.name[new_last.len])
4896 goto exit5;
4897 }
4898 }
4899 /* unless the source is a directory trailing slashes give -ENOTDIR */
4900 if (!d_is_dir(old_dentry)) {
4901 error = -ENOTDIR;
4902 if (old_last.name[old_last.len])
4903 goto exit5;
4904 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4905 goto exit5;
4906 }
4907 /* source should not be ancestor of target */
4908 error = -EINVAL;
4909 if (old_dentry == trap)
4910 goto exit5;
4911 /* target should not be an ancestor of source */
4912 if (!(flags & RENAME_EXCHANGE))
4913 error = -ENOTEMPTY;
4914 if (new_dentry == trap)
4915 goto exit5;
4916
4917 error = security_path_rename(&old_path, old_dentry,
4918 &new_path, new_dentry, flags);
4919 if (error)
4920 goto exit5;
4921
4922 rd.old_dir = old_path.dentry->d_inode;
4923 rd.old_dentry = old_dentry;
4924 rd.old_mnt_userns = mnt_user_ns(old_path.mnt);
4925 rd.new_dir = new_path.dentry->d_inode;
4926 rd.new_dentry = new_dentry;
4927 rd.new_mnt_userns = mnt_user_ns(new_path.mnt);
4928 rd.delegated_inode = &delegated_inode;
4929 rd.flags = flags;
4930 error = vfs_rename(&rd);
4931exit5:
4932 dput(new_dentry);
4933exit4:
4934 dput(old_dentry);
4935exit3:
4936 unlock_rename(new_path.dentry, old_path.dentry);
4937 if (delegated_inode) {
4938 error = break_deleg_wait(&delegated_inode);
4939 if (!error)
4940 goto retry_deleg;
4941 }
4942 mnt_drop_write(old_path.mnt);
4943exit2:
4944 if (retry_estale(error, lookup_flags))
4945 should_retry = true;
4946 path_put(&new_path);
4947exit1:
4948 path_put(&old_path);
4949 if (should_retry) {
4950 should_retry = false;
4951 lookup_flags |= LOOKUP_REVAL;
4952 goto retry;
4953 }
4954put_names:
4955 putname(from);
4956 putname(to);
4957 return error;
4958}
4959
4960SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4961 int, newdfd, const char __user *, newname, unsigned int, flags)
4962{
4963 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4964 flags);
4965}
4966
4967SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4968 int, newdfd, const char __user *, newname)
4969{
4970 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4971 0);
4972}
4973
4974SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4975{
4976 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4977 getname(newname), 0);
4978}
4979
4980int readlink_copy(char __user *buffer, int buflen, const char *link)
4981{
4982 int len = PTR_ERR(link);
4983 if (IS_ERR(link))
4984 goto out;
4985
4986 len = strlen(link);
4987 if (len > (unsigned) buflen)
4988 len = buflen;
4989 if (copy_to_user(buffer, link, len))
4990 len = -EFAULT;
4991out:
4992 return len;
4993}
4994
4995/**
4996 * vfs_readlink - copy symlink body into userspace buffer
4997 * @dentry: dentry on which to get symbolic link
4998 * @buffer: user memory pointer
4999 * @buflen: size of buffer
5000 *
5001 * Does not touch atime. That's up to the caller if necessary
5002 *
5003 * Does not call security hook.
5004 */
5005int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5006{
5007 struct inode *inode = d_inode(dentry);
5008 DEFINE_DELAYED_CALL(done);
5009 const char *link;
5010 int res;
5011
5012 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5013 if (unlikely(inode->i_op->readlink))
5014 return inode->i_op->readlink(dentry, buffer, buflen);
5015
5016 if (!d_is_symlink(dentry))
5017 return -EINVAL;
5018
5019 spin_lock(&inode->i_lock);
5020 inode->i_opflags |= IOP_DEFAULT_READLINK;
5021 spin_unlock(&inode->i_lock);
5022 }
5023
5024 link = READ_ONCE(inode->i_link);
5025 if (!link) {
5026 link = inode->i_op->get_link(dentry, inode, &done);
5027 if (IS_ERR(link))
5028 return PTR_ERR(link);
5029 }
5030 res = readlink_copy(buffer, buflen, link);
5031 do_delayed_call(&done);
5032 return res;
5033}
5034EXPORT_SYMBOL(vfs_readlink);
5035
5036/**
5037 * vfs_get_link - get symlink body
5038 * @dentry: dentry on which to get symbolic link
5039 * @done: caller needs to free returned data with this
5040 *
5041 * Calls security hook and i_op->get_link() on the supplied inode.
5042 *
5043 * It does not touch atime. That's up to the caller if necessary.
5044 *
5045 * Does not work on "special" symlinks like /proc/$$/fd/N
5046 */
5047const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5048{
5049 const char *res = ERR_PTR(-EINVAL);
5050 struct inode *inode = d_inode(dentry);
5051
5052 if (d_is_symlink(dentry)) {
5053 res = ERR_PTR(security_inode_readlink(dentry));
5054 if (!res)
5055 res = inode->i_op->get_link(dentry, inode, done);
5056 }
5057 return res;
5058}
5059EXPORT_SYMBOL(vfs_get_link);
5060
5061/* get the link contents into pagecache */
5062const char *page_get_link(struct dentry *dentry, struct inode *inode,
5063 struct delayed_call *callback)
5064{
5065 char *kaddr;
5066 struct page *page;
5067 struct address_space *mapping = inode->i_mapping;
5068
5069 if (!dentry) {
5070 page = find_get_page(mapping, 0);
5071 if (!page)
5072 return ERR_PTR(-ECHILD);
5073 if (!PageUptodate(page)) {
5074 put_page(page);
5075 return ERR_PTR(-ECHILD);
5076 }
5077 } else {
5078 page = read_mapping_page(mapping, 0, NULL);
5079 if (IS_ERR(page))
5080 return (char*)page;
5081 }
5082 set_delayed_call(callback, page_put_link, page);
5083 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5084 kaddr = page_address(page);
5085 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5086 return kaddr;
5087}
5088
5089EXPORT_SYMBOL(page_get_link);
5090
5091void page_put_link(void *arg)
5092{
5093 put_page(arg);
5094}
5095EXPORT_SYMBOL(page_put_link);
5096
5097int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5098{
5099 DEFINE_DELAYED_CALL(done);
5100 int res = readlink_copy(buffer, buflen,
5101 page_get_link(dentry, d_inode(dentry),
5102 &done));
5103 do_delayed_call(&done);
5104 return res;
5105}
5106EXPORT_SYMBOL(page_readlink);
5107
5108int page_symlink(struct inode *inode, const char *symname, int len)
5109{
5110 struct address_space *mapping = inode->i_mapping;
5111 const struct address_space_operations *aops = mapping->a_ops;
5112 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5113 struct page *page;
5114 void *fsdata = NULL;
5115 int err;
5116 unsigned int flags;
5117
5118retry:
5119 if (nofs)
5120 flags = memalloc_nofs_save();
5121 err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5122 if (nofs)
5123 memalloc_nofs_restore(flags);
5124 if (err)
5125 goto fail;
5126
5127 memcpy(page_address(page), symname, len-1);
5128
5129 err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5130 page, fsdata);
5131 if (err < 0)
5132 goto fail;
5133 if (err < len-1)
5134 goto retry;
5135
5136 mark_inode_dirty(inode);
5137 return 0;
5138fail:
5139 return err;
5140}
5141EXPORT_SYMBOL(page_symlink);
5142
5143const struct inode_operations page_symlink_inode_operations = {
5144 .get_link = page_get_link,
5145};
5146EXPORT_SYMBOL(page_symlink_inode_operations);