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->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
600 path_put(&nd->root);
601 nd->root.mnt = NULL;
602 }
603 } else {
604 nd->flags &= ~LOOKUP_RCU;
605 if (!(nd->flags & LOOKUP_ROOT))
606 nd->root.mnt = NULL;
607 rcu_read_unlock();
608 }
609 nd->depth = 0;
610}
611
612/* path_put is needed afterwards regardless of success or failure */
613static bool legitimize_path(struct nameidata *nd,
614 struct path *path, unsigned seq)
615{
616 int res = __legitimize_mnt(path->mnt, nd->m_seq);
617 if (unlikely(res)) {
618 if (res > 0)
619 path->mnt = NULL;
620 path->dentry = NULL;
621 return false;
622 }
623 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
624 path->dentry = NULL;
625 return false;
626 }
627 return !read_seqcount_retry(&path->dentry->d_seq, seq);
628}
629
630static bool legitimize_links(struct nameidata *nd)
631{
632 int i;
633 for (i = 0; i < nd->depth; i++) {
634 struct saved *last = nd->stack + i;
635 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
636 drop_links(nd);
637 nd->depth = i + 1;
638 return false;
639 }
640 }
641 return true;
642}
643
644/*
645 * Path walking has 2 modes, rcu-walk and ref-walk (see
646 * Documentation/filesystems/path-lookup.txt). In situations when we can't
647 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
648 * normal reference counts on dentries and vfsmounts to transition to ref-walk
649 * mode. Refcounts are grabbed at the last known good point before rcu-walk
650 * got stuck, so ref-walk may continue from there. If this is not successful
651 * (eg. a seqcount has changed), then failure is returned and it's up to caller
652 * to restart the path walk from the beginning in ref-walk mode.
653 */
654
655/**
656 * unlazy_walk - try to switch to ref-walk mode.
657 * @nd: nameidata pathwalk data
658 * Returns: 0 on success, -ECHILD on failure
659 *
660 * unlazy_walk attempts to legitimize the current nd->path and nd->root
661 * for ref-walk mode.
662 * Must be called from rcu-walk context.
663 * Nothing should touch nameidata between unlazy_walk() failure and
664 * terminate_walk().
665 */
666static int unlazy_walk(struct nameidata *nd)
667{
668 struct dentry *parent = nd->path.dentry;
669
670 BUG_ON(!(nd->flags & LOOKUP_RCU));
671
672 nd->flags &= ~LOOKUP_RCU;
673 if (unlikely(!legitimize_links(nd)))
674 goto out2;
675 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
676 goto out1;
677 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
678 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq)))
679 goto out;
680 }
681 rcu_read_unlock();
682 BUG_ON(nd->inode != parent->d_inode);
683 return 0;
684
685out2:
686 nd->path.mnt = NULL;
687 nd->path.dentry = NULL;
688out1:
689 if (!(nd->flags & LOOKUP_ROOT))
690 nd->root.mnt = NULL;
691out:
692 rcu_read_unlock();
693 return -ECHILD;
694}
695
696/**
697 * unlazy_child - try to switch to ref-walk mode.
698 * @nd: nameidata pathwalk data
699 * @dentry: child of nd->path.dentry
700 * @seq: seq number to check dentry against
701 * Returns: 0 on success, -ECHILD on failure
702 *
703 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
704 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
705 * @nd. Must be called from rcu-walk context.
706 * Nothing should touch nameidata between unlazy_child() failure and
707 * terminate_walk().
708 */
709static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
710{
711 BUG_ON(!(nd->flags & LOOKUP_RCU));
712
713 nd->flags &= ~LOOKUP_RCU;
714 if (unlikely(!legitimize_links(nd)))
715 goto out2;
716 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
717 goto out2;
718 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
719 goto out1;
720
721 /*
722 * We need to move both the parent and the dentry from the RCU domain
723 * to be properly refcounted. And the sequence number in the dentry
724 * validates *both* dentry counters, since we checked the sequence
725 * number of the parent after we got the child sequence number. So we
726 * know the parent must still be valid if the child sequence number is
727 */
728 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
729 goto out;
730 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) {
731 rcu_read_unlock();
732 dput(dentry);
733 goto drop_root_mnt;
734 }
735 /*
736 * Sequence counts matched. Now make sure that the root is
737 * still valid and get it if required.
738 */
739 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
740 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
741 rcu_read_unlock();
742 dput(dentry);
743 return -ECHILD;
744 }
745 }
746
747 rcu_read_unlock();
748 return 0;
749
750out2:
751 nd->path.mnt = NULL;
752out1:
753 nd->path.dentry = NULL;
754out:
755 rcu_read_unlock();
756drop_root_mnt:
757 if (!(nd->flags & LOOKUP_ROOT))
758 nd->root.mnt = NULL;
759 return -ECHILD;
760}
761
762static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
763{
764 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
765 return dentry->d_op->d_revalidate(dentry, flags);
766 else
767 return 1;
768}
769
770/**
771 * complete_walk - successful completion of path walk
772 * @nd: pointer nameidata
773 *
774 * If we had been in RCU mode, drop out of it and legitimize nd->path.
775 * Revalidate the final result, unless we'd already done that during
776 * the path walk or the filesystem doesn't ask for it. Return 0 on
777 * success, -error on failure. In case of failure caller does not
778 * need to drop nd->path.
779 */
780static int complete_walk(struct nameidata *nd)
781{
782 struct dentry *dentry = nd->path.dentry;
783 int status;
784
785 if (nd->flags & LOOKUP_RCU) {
786 if (!(nd->flags & LOOKUP_ROOT))
787 nd->root.mnt = NULL;
788 if (unlikely(unlazy_walk(nd)))
789 return -ECHILD;
790 }
791
792 if (likely(!(nd->flags & LOOKUP_JUMPED)))
793 return 0;
794
795 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
796 return 0;
797
798 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
799 if (status > 0)
800 return 0;
801
802 if (!status)
803 status = -ESTALE;
804
805 return status;
806}
807
808static void set_root(struct nameidata *nd)
809{
810 struct fs_struct *fs = current->fs;
811
812 if (nd->flags & LOOKUP_RCU) {
813 unsigned seq;
814
815 do {
816 seq = read_seqcount_begin(&fs->seq);
817 nd->root = fs->root;
818 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
819 } while (read_seqcount_retry(&fs->seq, seq));
820 } else {
821 get_fs_root(fs, &nd->root);
822 }
823}
824
825static void path_put_conditional(struct path *path, struct nameidata *nd)
826{
827 dput(path->dentry);
828 if (path->mnt != nd->path.mnt)
829 mntput(path->mnt);
830}
831
832static inline void path_to_nameidata(const struct path *path,
833 struct nameidata *nd)
834{
835 if (!(nd->flags & LOOKUP_RCU)) {
836 dput(nd->path.dentry);
837 if (nd->path.mnt != path->mnt)
838 mntput(nd->path.mnt);
839 }
840 nd->path.mnt = path->mnt;
841 nd->path.dentry = path->dentry;
842}
843
844static int nd_jump_root(struct nameidata *nd)
845{
846 if (nd->flags & LOOKUP_RCU) {
847 struct dentry *d;
848 nd->path = nd->root;
849 d = nd->path.dentry;
850 nd->inode = d->d_inode;
851 nd->seq = nd->root_seq;
852 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
853 return -ECHILD;
854 } else {
855 path_put(&nd->path);
856 nd->path = nd->root;
857 path_get(&nd->path);
858 nd->inode = nd->path.dentry->d_inode;
859 }
860 nd->flags |= LOOKUP_JUMPED;
861 return 0;
862}
863
864/*
865 * Helper to directly jump to a known parsed path from ->get_link,
866 * caller must have taken a reference to path beforehand.
867 */
868void nd_jump_link(struct path *path)
869{
870 struct nameidata *nd = current->nameidata;
871 path_put(&nd->path);
872
873 nd->path = *path;
874 nd->inode = nd->path.dentry->d_inode;
875 nd->flags |= LOOKUP_JUMPED;
876}
877
878static inline void put_link(struct nameidata *nd)
879{
880 struct saved *last = nd->stack + --nd->depth;
881 do_delayed_call(&last->done);
882 if (!(nd->flags & LOOKUP_RCU))
883 path_put(&last->link);
884}
885
886int sysctl_protected_symlinks __read_mostly = 0;
887int sysctl_protected_hardlinks __read_mostly = 0;
888int sysctl_protected_fifos __read_mostly;
889int sysctl_protected_regular __read_mostly;
890
891/**
892 * may_follow_link - Check symlink following for unsafe situations
893 * @nd: nameidata pathwalk data
894 *
895 * In the case of the sysctl_protected_symlinks sysctl being enabled,
896 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
897 * in a sticky world-writable directory. This is to protect privileged
898 * processes from failing races against path names that may change out
899 * from under them by way of other users creating malicious symlinks.
900 * It will permit symlinks to be followed only when outside a sticky
901 * world-writable directory, or when the uid of the symlink and follower
902 * match, or when the directory owner matches the symlink's owner.
903 *
904 * Returns 0 if following the symlink is allowed, -ve on error.
905 */
906static inline int may_follow_link(struct nameidata *nd)
907{
908 const struct inode *inode;
909 const struct inode *parent;
910 kuid_t puid;
911
912 if (!sysctl_protected_symlinks)
913 return 0;
914
915 /* Allowed if owner and follower match. */
916 inode = nd->link_inode;
917 if (uid_eq(current_cred()->fsuid, inode->i_uid))
918 return 0;
919
920 /* Allowed if parent directory not sticky and world-writable. */
921 parent = nd->inode;
922 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
923 return 0;
924
925 /* Allowed if parent directory and link owner match. */
926 puid = parent->i_uid;
927 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
928 return 0;
929
930 if (nd->flags & LOOKUP_RCU)
931 return -ECHILD;
932
933 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
934 audit_log_link_denied("follow_link");
935 return -EACCES;
936}
937
938/**
939 * safe_hardlink_source - Check for safe hardlink conditions
940 * @inode: the source inode to hardlink from
941 *
942 * Return false if at least one of the following conditions:
943 * - inode is not a regular file
944 * - inode is setuid
945 * - inode is setgid and group-exec
946 * - access failure for read and write
947 *
948 * Otherwise returns true.
949 */
950static bool safe_hardlink_source(struct inode *inode)
951{
952 umode_t mode = inode->i_mode;
953
954 /* Special files should not get pinned to the filesystem. */
955 if (!S_ISREG(mode))
956 return false;
957
958 /* Setuid files should not get pinned to the filesystem. */
959 if (mode & S_ISUID)
960 return false;
961
962 /* Executable setgid files should not get pinned to the filesystem. */
963 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
964 return false;
965
966 /* Hardlinking to unreadable or unwritable sources is dangerous. */
967 if (inode_permission(inode, MAY_READ | MAY_WRITE))
968 return false;
969
970 return true;
971}
972
973/**
974 * may_linkat - Check permissions for creating a hardlink
975 * @link: the source to hardlink from
976 *
977 * Block hardlink when all of:
978 * - sysctl_protected_hardlinks enabled
979 * - fsuid does not match inode
980 * - hardlink source is unsafe (see safe_hardlink_source() above)
981 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
982 *
983 * Returns 0 if successful, -ve on error.
984 */
985static int may_linkat(struct path *link)
986{
987 struct inode *inode = link->dentry->d_inode;
988
989 /* Inode writeback is not safe when the uid or gid are invalid. */
990 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
991 return -EOVERFLOW;
992
993 if (!sysctl_protected_hardlinks)
994 return 0;
995
996 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
997 * otherwise, it must be a safe source.
998 */
999 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1000 return 0;
1001
1002 audit_log_link_denied("linkat");
1003 return -EPERM;
1004}
1005
1006/**
1007 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1008 * should be allowed, or not, on files that already
1009 * exist.
1010 * @dir: the sticky parent directory
1011 * @inode: the inode of the file to open
1012 *
1013 * Block an O_CREAT open of a FIFO (or a regular file) when:
1014 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1015 * - the file already exists
1016 * - we are in a sticky directory
1017 * - we don't own the file
1018 * - the owner of the directory doesn't own the file
1019 * - the directory is world writable
1020 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1021 * the directory doesn't have to be world writable: being group writable will
1022 * be enough.
1023 *
1024 * Returns 0 if the open is allowed, -ve on error.
1025 */
1026static int may_create_in_sticky(struct dentry * const dir,
1027 struct inode * const inode)
1028{
1029 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1030 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1031 likely(!(dir->d_inode->i_mode & S_ISVTX)) ||
1032 uid_eq(inode->i_uid, dir->d_inode->i_uid) ||
1033 uid_eq(current_fsuid(), inode->i_uid))
1034 return 0;
1035
1036 if (likely(dir->d_inode->i_mode & 0002) ||
1037 (dir->d_inode->i_mode & 0020 &&
1038 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1039 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1040 return -EACCES;
1041 }
1042 return 0;
1043}
1044
1045static __always_inline
1046const char *get_link(struct nameidata *nd)
1047{
1048 struct saved *last = nd->stack + nd->depth - 1;
1049 struct dentry *dentry = last->link.dentry;
1050 struct inode *inode = nd->link_inode;
1051 int error;
1052 const char *res;
1053
1054 if (!(nd->flags & LOOKUP_RCU)) {
1055 touch_atime(&last->link);
1056 cond_resched();
1057 } else if (atime_needs_update(&last->link, inode)) {
1058 if (unlikely(unlazy_walk(nd)))
1059 return ERR_PTR(-ECHILD);
1060 touch_atime(&last->link);
1061 }
1062
1063 error = security_inode_follow_link(dentry, inode,
1064 nd->flags & LOOKUP_RCU);
1065 if (unlikely(error))
1066 return ERR_PTR(error);
1067
1068 nd->last_type = LAST_BIND;
1069 res = inode->i_link;
1070 if (!res) {
1071 const char * (*get)(struct dentry *, struct inode *,
1072 struct delayed_call *);
1073 get = inode->i_op->get_link;
1074 if (nd->flags & LOOKUP_RCU) {
1075 res = get(NULL, inode, &last->done);
1076 if (res == ERR_PTR(-ECHILD)) {
1077 if (unlikely(unlazy_walk(nd)))
1078 return ERR_PTR(-ECHILD);
1079 res = get(dentry, inode, &last->done);
1080 }
1081 } else {
1082 res = get(dentry, inode, &last->done);
1083 }
1084 if (IS_ERR_OR_NULL(res))
1085 return res;
1086 }
1087 if (*res == '/') {
1088 if (!nd->root.mnt)
1089 set_root(nd);
1090 if (unlikely(nd_jump_root(nd)))
1091 return ERR_PTR(-ECHILD);
1092 while (unlikely(*++res == '/'))
1093 ;
1094 }
1095 if (!*res)
1096 res = NULL;
1097 return res;
1098}
1099
1100/*
1101 * follow_up - Find the mountpoint of path's vfsmount
1102 *
1103 * Given a path, find the mountpoint of its source file system.
1104 * Replace @path with the path of the mountpoint in the parent mount.
1105 * Up is towards /.
1106 *
1107 * Return 1 if we went up a level and 0 if we were already at the
1108 * root.
1109 */
1110int follow_up(struct path *path)
1111{
1112 struct mount *mnt = real_mount(path->mnt);
1113 struct mount *parent;
1114 struct dentry *mountpoint;
1115
1116 read_seqlock_excl(&mount_lock);
1117 parent = mnt->mnt_parent;
1118 if (parent == mnt) {
1119 read_sequnlock_excl(&mount_lock);
1120 return 0;
1121 }
1122 mntget(&parent->mnt);
1123 mountpoint = dget(mnt->mnt_mountpoint);
1124 read_sequnlock_excl(&mount_lock);
1125 dput(path->dentry);
1126 path->dentry = mountpoint;
1127 mntput(path->mnt);
1128 path->mnt = &parent->mnt;
1129 return 1;
1130}
1131EXPORT_SYMBOL(follow_up);
1132
1133/*
1134 * Perform an automount
1135 * - return -EISDIR to tell follow_managed() to stop and return the path we
1136 * were called with.
1137 */
1138static int follow_automount(struct path *path, struct nameidata *nd,
1139 bool *need_mntput)
1140{
1141 struct vfsmount *mnt;
1142 int err;
1143
1144 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1145 return -EREMOTE;
1146
1147 /* We don't want to mount if someone's just doing a stat -
1148 * unless they're stat'ing a directory and appended a '/' to
1149 * the name.
1150 *
1151 * We do, however, want to mount if someone wants to open or
1152 * create a file of any type under the mountpoint, wants to
1153 * traverse through the mountpoint or wants to open the
1154 * mounted directory. Also, autofs may mark negative dentries
1155 * as being automount points. These will need the attentions
1156 * of the daemon to instantiate them before they can be used.
1157 */
1158 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1159 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1160 path->dentry->d_inode)
1161 return -EISDIR;
1162
1163 nd->total_link_count++;
1164 if (nd->total_link_count >= 40)
1165 return -ELOOP;
1166
1167 mnt = path->dentry->d_op->d_automount(path);
1168 if (IS_ERR(mnt)) {
1169 /*
1170 * The filesystem is allowed to return -EISDIR here to indicate
1171 * it doesn't want to automount. For instance, autofs would do
1172 * this so that its userspace daemon can mount on this dentry.
1173 *
1174 * However, we can only permit this if it's a terminal point in
1175 * the path being looked up; if it wasn't then the remainder of
1176 * the path is inaccessible and we should say so.
1177 */
1178 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1179 return -EREMOTE;
1180 return PTR_ERR(mnt);
1181 }
1182
1183 if (!mnt) /* mount collision */
1184 return 0;
1185
1186 if (!*need_mntput) {
1187 /* lock_mount() may release path->mnt on error */
1188 mntget(path->mnt);
1189 *need_mntput = true;
1190 }
1191 err = finish_automount(mnt, path);
1192
1193 switch (err) {
1194 case -EBUSY:
1195 /* Someone else made a mount here whilst we were busy */
1196 return 0;
1197 case 0:
1198 path_put(path);
1199 path->mnt = mnt;
1200 path->dentry = dget(mnt->mnt_root);
1201 return 0;
1202 default:
1203 return err;
1204 }
1205
1206}
1207
1208/*
1209 * Handle a dentry that is managed in some way.
1210 * - Flagged for transit management (autofs)
1211 * - Flagged as mountpoint
1212 * - Flagged as automount point
1213 *
1214 * This may only be called in refwalk mode.
1215 *
1216 * Serialization is taken care of in namespace.c
1217 */
1218static int follow_managed(struct path *path, struct nameidata *nd)
1219{
1220 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1221 unsigned managed;
1222 bool need_mntput = false;
1223 int ret = 0;
1224
1225 /* Given that we're not holding a lock here, we retain the value in a
1226 * local variable for each dentry as we look at it so that we don't see
1227 * the components of that value change under us */
1228 while (managed = READ_ONCE(path->dentry->d_flags),
1229 managed &= DCACHE_MANAGED_DENTRY,
1230 unlikely(managed != 0)) {
1231 /* Allow the filesystem to manage the transit without i_mutex
1232 * being held. */
1233 if (managed & DCACHE_MANAGE_TRANSIT) {
1234 BUG_ON(!path->dentry->d_op);
1235 BUG_ON(!path->dentry->d_op->d_manage);
1236 ret = path->dentry->d_op->d_manage(path, false);
1237 if (ret < 0)
1238 break;
1239 }
1240
1241 /* Transit to a mounted filesystem. */
1242 if (managed & DCACHE_MOUNTED) {
1243 struct vfsmount *mounted = lookup_mnt(path);
1244 if (mounted) {
1245 dput(path->dentry);
1246 if (need_mntput)
1247 mntput(path->mnt);
1248 path->mnt = mounted;
1249 path->dentry = dget(mounted->mnt_root);
1250 need_mntput = true;
1251 continue;
1252 }
1253
1254 /* Something is mounted on this dentry in another
1255 * namespace and/or whatever was mounted there in this
1256 * namespace got unmounted before lookup_mnt() could
1257 * get it */
1258 }
1259
1260 /* Handle an automount point */
1261 if (managed & DCACHE_NEED_AUTOMOUNT) {
1262 ret = follow_automount(path, nd, &need_mntput);
1263 if (ret < 0)
1264 break;
1265 continue;
1266 }
1267
1268 /* We didn't change the current path point */
1269 break;
1270 }
1271
1272 if (need_mntput && path->mnt == mnt)
1273 mntput(path->mnt);
1274 if (ret == -EISDIR || !ret)
1275 ret = 1;
1276 if (need_mntput)
1277 nd->flags |= LOOKUP_JUMPED;
1278 if (unlikely(ret < 0))
1279 path_put_conditional(path, nd);
1280 return ret;
1281}
1282
1283int follow_down_one(struct path *path)
1284{
1285 struct vfsmount *mounted;
1286
1287 mounted = lookup_mnt(path);
1288 if (mounted) {
1289 dput(path->dentry);
1290 mntput(path->mnt);
1291 path->mnt = mounted;
1292 path->dentry = dget(mounted->mnt_root);
1293 return 1;
1294 }
1295 return 0;
1296}
1297EXPORT_SYMBOL(follow_down_one);
1298
1299static inline int managed_dentry_rcu(const struct path *path)
1300{
1301 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1302 path->dentry->d_op->d_manage(path, true) : 0;
1303}
1304
1305/*
1306 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1307 * we meet a managed dentry that would need blocking.
1308 */
1309static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1310 struct inode **inode, unsigned *seqp)
1311{
1312 for (;;) {
1313 struct mount *mounted;
1314 /*
1315 * Don't forget we might have a non-mountpoint managed dentry
1316 * that wants to block transit.
1317 */
1318 switch (managed_dentry_rcu(path)) {
1319 case -ECHILD:
1320 default:
1321 return false;
1322 case -EISDIR:
1323 return true;
1324 case 0:
1325 break;
1326 }
1327
1328 if (!d_mountpoint(path->dentry))
1329 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1330
1331 mounted = __lookup_mnt(path->mnt, path->dentry);
1332 if (!mounted)
1333 break;
1334 path->mnt = &mounted->mnt;
1335 path->dentry = mounted->mnt.mnt_root;
1336 nd->flags |= LOOKUP_JUMPED;
1337 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1338 /*
1339 * Update the inode too. We don't need to re-check the
1340 * dentry sequence number here after this d_inode read,
1341 * because a mount-point is always pinned.
1342 */
1343 *inode = path->dentry->d_inode;
1344 }
1345 return !read_seqretry(&mount_lock, nd->m_seq) &&
1346 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1347}
1348
1349static int follow_dotdot_rcu(struct nameidata *nd)
1350{
1351 struct inode *inode = nd->inode;
1352
1353 while (1) {
1354 if (path_equal(&nd->path, &nd->root))
1355 break;
1356 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1357 struct dentry *old = nd->path.dentry;
1358 struct dentry *parent = old->d_parent;
1359 unsigned seq;
1360
1361 inode = parent->d_inode;
1362 seq = read_seqcount_begin(&parent->d_seq);
1363 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1364 return -ECHILD;
1365 nd->path.dentry = parent;
1366 nd->seq = seq;
1367 if (unlikely(!path_connected(&nd->path)))
1368 return -ENOENT;
1369 break;
1370 } else {
1371 struct mount *mnt = real_mount(nd->path.mnt);
1372 struct mount *mparent = mnt->mnt_parent;
1373 struct dentry *mountpoint = mnt->mnt_mountpoint;
1374 struct inode *inode2 = mountpoint->d_inode;
1375 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1376 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1377 return -ECHILD;
1378 if (&mparent->mnt == nd->path.mnt)
1379 break;
1380 /* we know that mountpoint was pinned */
1381 nd->path.dentry = mountpoint;
1382 nd->path.mnt = &mparent->mnt;
1383 inode = inode2;
1384 nd->seq = seq;
1385 }
1386 }
1387 while (unlikely(d_mountpoint(nd->path.dentry))) {
1388 struct mount *mounted;
1389 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1390 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1391 return -ECHILD;
1392 if (!mounted)
1393 break;
1394 nd->path.mnt = &mounted->mnt;
1395 nd->path.dentry = mounted->mnt.mnt_root;
1396 inode = nd->path.dentry->d_inode;
1397 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1398 }
1399 nd->inode = inode;
1400 return 0;
1401}
1402
1403/*
1404 * Follow down to the covering mount currently visible to userspace. At each
1405 * point, the filesystem owning that dentry may be queried as to whether the
1406 * caller is permitted to proceed or not.
1407 */
1408int follow_down(struct path *path)
1409{
1410 unsigned managed;
1411 int ret;
1412
1413 while (managed = READ_ONCE(path->dentry->d_flags),
1414 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1415 /* Allow the filesystem to manage the transit without i_mutex
1416 * being held.
1417 *
1418 * We indicate to the filesystem if someone is trying to mount
1419 * something here. This gives autofs the chance to deny anyone
1420 * other than its daemon the right to mount on its
1421 * superstructure.
1422 *
1423 * The filesystem may sleep at this point.
1424 */
1425 if (managed & DCACHE_MANAGE_TRANSIT) {
1426 BUG_ON(!path->dentry->d_op);
1427 BUG_ON(!path->dentry->d_op->d_manage);
1428 ret = path->dentry->d_op->d_manage(path, false);
1429 if (ret < 0)
1430 return ret == -EISDIR ? 0 : ret;
1431 }
1432
1433 /* Transit to a mounted filesystem. */
1434 if (managed & DCACHE_MOUNTED) {
1435 struct vfsmount *mounted = lookup_mnt(path);
1436 if (!mounted)
1437 break;
1438 dput(path->dentry);
1439 mntput(path->mnt);
1440 path->mnt = mounted;
1441 path->dentry = dget(mounted->mnt_root);
1442 continue;
1443 }
1444
1445 /* Don't handle automount points here */
1446 break;
1447 }
1448 return 0;
1449}
1450EXPORT_SYMBOL(follow_down);
1451
1452/*
1453 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1454 */
1455static void follow_mount(struct path *path)
1456{
1457 while (d_mountpoint(path->dentry)) {
1458 struct vfsmount *mounted = lookup_mnt(path);
1459 if (!mounted)
1460 break;
1461 dput(path->dentry);
1462 mntput(path->mnt);
1463 path->mnt = mounted;
1464 path->dentry = dget(mounted->mnt_root);
1465 }
1466}
1467
1468static int path_parent_directory(struct path *path)
1469{
1470 struct dentry *old = path->dentry;
1471 /* rare case of legitimate dget_parent()... */
1472 path->dentry = dget_parent(path->dentry);
1473 dput(old);
1474 if (unlikely(!path_connected(path)))
1475 return -ENOENT;
1476 return 0;
1477}
1478
1479static int follow_dotdot(struct nameidata *nd)
1480{
1481 while(1) {
1482 if (path_equal(&nd->path, &nd->root))
1483 break;
1484 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1485 int ret = path_parent_directory(&nd->path);
1486 if (ret)
1487 return ret;
1488 break;
1489 }
1490 if (!follow_up(&nd->path))
1491 break;
1492 }
1493 follow_mount(&nd->path);
1494 nd->inode = nd->path.dentry->d_inode;
1495 return 0;
1496}
1497
1498/*
1499 * This looks up the name in dcache and possibly revalidates the found dentry.
1500 * NULL is returned if the dentry does not exist in the cache.
1501 */
1502static struct dentry *lookup_dcache(const struct qstr *name,
1503 struct dentry *dir,
1504 unsigned int flags)
1505{
1506 struct dentry *dentry = d_lookup(dir, name);
1507 if (dentry) {
1508 int error = d_revalidate(dentry, flags);
1509 if (unlikely(error <= 0)) {
1510 if (!error)
1511 d_invalidate(dentry);
1512 dput(dentry);
1513 return ERR_PTR(error);
1514 }
1515 }
1516 return dentry;
1517}
1518
1519/*
1520 * Parent directory has inode locked exclusive. This is one
1521 * and only case when ->lookup() gets called on non in-lookup
1522 * dentries - as the matter of fact, this only gets called
1523 * when directory is guaranteed to have no in-lookup children
1524 * at all.
1525 */
1526static struct dentry *__lookup_hash(const struct qstr *name,
1527 struct dentry *base, unsigned int flags)
1528{
1529 struct dentry *dentry = lookup_dcache(name, base, flags);
1530 struct dentry *old;
1531 struct inode *dir = base->d_inode;
1532
1533 if (dentry)
1534 return dentry;
1535
1536 /* Don't create child dentry for a dead directory. */
1537 if (unlikely(IS_DEADDIR(dir)))
1538 return ERR_PTR(-ENOENT);
1539
1540 dentry = d_alloc(base, name);
1541 if (unlikely(!dentry))
1542 return ERR_PTR(-ENOMEM);
1543
1544 old = dir->i_op->lookup(dir, dentry, flags);
1545 if (unlikely(old)) {
1546 dput(dentry);
1547 dentry = old;
1548 }
1549 return dentry;
1550}
1551
1552static int lookup_fast(struct nameidata *nd,
1553 struct path *path, struct inode **inode,
1554 unsigned *seqp)
1555{
1556 struct vfsmount *mnt = nd->path.mnt;
1557 struct dentry *dentry, *parent = nd->path.dentry;
1558 int status = 1;
1559 int err;
1560
1561 /*
1562 * Rename seqlock is not required here because in the off chance
1563 * of a false negative due to a concurrent rename, the caller is
1564 * going to fall back to non-racy lookup.
1565 */
1566 if (nd->flags & LOOKUP_RCU) {
1567 unsigned seq;
1568 bool negative;
1569 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1570 if (unlikely(!dentry)) {
1571 if (unlazy_walk(nd))
1572 return -ECHILD;
1573 return 0;
1574 }
1575
1576 /*
1577 * This sequence count validates that the inode matches
1578 * the dentry name information from lookup.
1579 */
1580 *inode = d_backing_inode(dentry);
1581 negative = d_is_negative(dentry);
1582 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1583 return -ECHILD;
1584
1585 /*
1586 * This sequence count validates that the parent had no
1587 * changes while we did the lookup of the dentry above.
1588 *
1589 * The memory barrier in read_seqcount_begin of child is
1590 * enough, we can use __read_seqcount_retry here.
1591 */
1592 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1593 return -ECHILD;
1594
1595 *seqp = seq;
1596 status = d_revalidate(dentry, nd->flags);
1597 if (likely(status > 0)) {
1598 /*
1599 * Note: do negative dentry check after revalidation in
1600 * case that drops it.
1601 */
1602 if (unlikely(negative))
1603 return -ENOENT;
1604 path->mnt = mnt;
1605 path->dentry = dentry;
1606 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1607 return 1;
1608 }
1609 if (unlazy_child(nd, dentry, seq))
1610 return -ECHILD;
1611 if (unlikely(status == -ECHILD))
1612 /* we'd been told to redo it in non-rcu mode */
1613 status = d_revalidate(dentry, nd->flags);
1614 } else {
1615 dentry = __d_lookup(parent, &nd->last);
1616 if (unlikely(!dentry))
1617 return 0;
1618 status = d_revalidate(dentry, nd->flags);
1619 }
1620 if (unlikely(status <= 0)) {
1621 if (!status)
1622 d_invalidate(dentry);
1623 dput(dentry);
1624 return status;
1625 }
1626 if (unlikely(d_is_negative(dentry))) {
1627 dput(dentry);
1628 return -ENOENT;
1629 }
1630
1631 path->mnt = mnt;
1632 path->dentry = dentry;
1633 err = follow_managed(path, nd);
1634 if (likely(err > 0))
1635 *inode = d_backing_inode(path->dentry);
1636 return err;
1637}
1638
1639/* Fast lookup failed, do it the slow way */
1640static struct dentry *__lookup_slow(const struct qstr *name,
1641 struct dentry *dir,
1642 unsigned int flags)
1643{
1644 struct dentry *dentry, *old;
1645 struct inode *inode = dir->d_inode;
1646 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1647
1648 /* Don't go there if it's already dead */
1649 if (unlikely(IS_DEADDIR(inode)))
1650 return ERR_PTR(-ENOENT);
1651again:
1652 dentry = d_alloc_parallel(dir, name, &wq);
1653 if (IS_ERR(dentry))
1654 return dentry;
1655 if (unlikely(!d_in_lookup(dentry))) {
1656 if (!(flags & LOOKUP_NO_REVAL)) {
1657 int error = d_revalidate(dentry, flags);
1658 if (unlikely(error <= 0)) {
1659 if (!error) {
1660 d_invalidate(dentry);
1661 dput(dentry);
1662 goto again;
1663 }
1664 dput(dentry);
1665 dentry = ERR_PTR(error);
1666 }
1667 }
1668 } else {
1669 old = inode->i_op->lookup(inode, dentry, flags);
1670 d_lookup_done(dentry);
1671 if (unlikely(old)) {
1672 dput(dentry);
1673 dentry = old;
1674 }
1675 }
1676 return dentry;
1677}
1678
1679static struct dentry *lookup_slow(const struct qstr *name,
1680 struct dentry *dir,
1681 unsigned int flags)
1682{
1683 struct inode *inode = dir->d_inode;
1684 struct dentry *res;
1685 inode_lock_shared(inode);
1686 res = __lookup_slow(name, dir, flags);
1687 inode_unlock_shared(inode);
1688 return res;
1689}
1690
1691static inline int may_lookup(struct nameidata *nd)
1692{
1693 if (nd->flags & LOOKUP_RCU) {
1694 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1695 if (err != -ECHILD)
1696 return err;
1697 if (unlazy_walk(nd))
1698 return -ECHILD;
1699 }
1700 return inode_permission(nd->inode, MAY_EXEC);
1701}
1702
1703static inline int handle_dots(struct nameidata *nd, int type)
1704{
1705 if (type == LAST_DOTDOT) {
1706 if (!nd->root.mnt)
1707 set_root(nd);
1708 if (nd->flags & LOOKUP_RCU) {
1709 return follow_dotdot_rcu(nd);
1710 } else
1711 return follow_dotdot(nd);
1712 }
1713 return 0;
1714}
1715
1716static int pick_link(struct nameidata *nd, struct path *link,
1717 struct inode *inode, unsigned seq)
1718{
1719 int error;
1720 struct saved *last;
1721 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1722 path_to_nameidata(link, nd);
1723 return -ELOOP;
1724 }
1725 if (!(nd->flags & LOOKUP_RCU)) {
1726 if (link->mnt == nd->path.mnt)
1727 mntget(link->mnt);
1728 }
1729 error = nd_alloc_stack(nd);
1730 if (unlikely(error)) {
1731 if (error == -ECHILD) {
1732 if (unlikely(!legitimize_path(nd, link, seq))) {
1733 drop_links(nd);
1734 nd->depth = 0;
1735 nd->flags &= ~LOOKUP_RCU;
1736 nd->path.mnt = NULL;
1737 nd->path.dentry = NULL;
1738 if (!(nd->flags & LOOKUP_ROOT))
1739 nd->root.mnt = NULL;
1740 rcu_read_unlock();
1741 } else if (likely(unlazy_walk(nd)) == 0)
1742 error = nd_alloc_stack(nd);
1743 }
1744 if (error) {
1745 path_put(link);
1746 return error;
1747 }
1748 }
1749
1750 last = nd->stack + nd->depth++;
1751 last->link = *link;
1752 clear_delayed_call(&last->done);
1753 nd->link_inode = inode;
1754 last->seq = seq;
1755 return 1;
1756}
1757
1758enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1759
1760/*
1761 * Do we need to follow links? We _really_ want to be able
1762 * to do this check without having to look at inode->i_op,
1763 * so we keep a cache of "no, this doesn't need follow_link"
1764 * for the common case.
1765 */
1766static inline int step_into(struct nameidata *nd, struct path *path,
1767 int flags, struct inode *inode, unsigned seq)
1768{
1769 if (!(flags & WALK_MORE) && nd->depth)
1770 put_link(nd);
1771 if (likely(!d_is_symlink(path->dentry)) ||
1772 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1773 /* not a symlink or should not follow */
1774 path_to_nameidata(path, nd);
1775 nd->inode = inode;
1776 nd->seq = seq;
1777 return 0;
1778 }
1779 /* make sure that d_is_symlink above matches inode */
1780 if (nd->flags & LOOKUP_RCU) {
1781 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1782 return -ECHILD;
1783 }
1784 return pick_link(nd, path, inode, seq);
1785}
1786
1787static int walk_component(struct nameidata *nd, int flags)
1788{
1789 struct path path;
1790 struct inode *inode;
1791 unsigned seq;
1792 int err;
1793 /*
1794 * "." and ".." are special - ".." especially so because it has
1795 * to be able to know about the current root directory and
1796 * parent relationships.
1797 */
1798 if (unlikely(nd->last_type != LAST_NORM)) {
1799 err = handle_dots(nd, nd->last_type);
1800 if (!(flags & WALK_MORE) && nd->depth)
1801 put_link(nd);
1802 return err;
1803 }
1804 err = lookup_fast(nd, &path, &inode, &seq);
1805 if (unlikely(err <= 0)) {
1806 if (err < 0)
1807 return err;
1808 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1809 nd->flags);
1810 if (IS_ERR(path.dentry))
1811 return PTR_ERR(path.dentry);
1812
1813 path.mnt = nd->path.mnt;
1814 err = follow_managed(&path, nd);
1815 if (unlikely(err < 0))
1816 return err;
1817
1818 if (unlikely(d_is_negative(path.dentry))) {
1819 path_to_nameidata(&path, nd);
1820 return -ENOENT;
1821 }
1822
1823 seq = 0; /* we are already out of RCU mode */
1824 inode = d_backing_inode(path.dentry);
1825 }
1826
1827 return step_into(nd, &path, flags, inode, seq);
1828}
1829
1830/*
1831 * We can do the critical dentry name comparison and hashing
1832 * operations one word at a time, but we are limited to:
1833 *
1834 * - Architectures with fast unaligned word accesses. We could
1835 * do a "get_unaligned()" if this helps and is sufficiently
1836 * fast.
1837 *
1838 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1839 * do not trap on the (extremely unlikely) case of a page
1840 * crossing operation.
1841 *
1842 * - Furthermore, we need an efficient 64-bit compile for the
1843 * 64-bit case in order to generate the "number of bytes in
1844 * the final mask". Again, that could be replaced with a
1845 * efficient population count instruction or similar.
1846 */
1847#ifdef CONFIG_DCACHE_WORD_ACCESS
1848
1849#include <asm/word-at-a-time.h>
1850
1851#ifdef HASH_MIX
1852
1853/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1854
1855#elif defined(CONFIG_64BIT)
1856/*
1857 * Register pressure in the mixing function is an issue, particularly
1858 * on 32-bit x86, but almost any function requires one state value and
1859 * one temporary. Instead, use a function designed for two state values
1860 * and no temporaries.
1861 *
1862 * This function cannot create a collision in only two iterations, so
1863 * we have two iterations to achieve avalanche. In those two iterations,
1864 * we have six layers of mixing, which is enough to spread one bit's
1865 * influence out to 2^6 = 64 state bits.
1866 *
1867 * Rotate constants are scored by considering either 64 one-bit input
1868 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1869 * probability of that delta causing a change to each of the 128 output
1870 * bits, using a sample of random initial states.
1871 *
1872 * The Shannon entropy of the computed probabilities is then summed
1873 * to produce a score. Ideally, any input change has a 50% chance of
1874 * toggling any given output bit.
1875 *
1876 * Mixing scores (in bits) for (12,45):
1877 * Input delta: 1-bit 2-bit
1878 * 1 round: 713.3 42542.6
1879 * 2 rounds: 2753.7 140389.8
1880 * 3 rounds: 5954.1 233458.2
1881 * 4 rounds: 7862.6 256672.2
1882 * Perfect: 8192 258048
1883 * (64*128) (64*63/2 * 128)
1884 */
1885#define HASH_MIX(x, y, a) \
1886 ( x ^= (a), \
1887 y ^= x, x = rol64(x,12),\
1888 x += y, y = rol64(y,45),\
1889 y *= 9 )
1890
1891/*
1892 * Fold two longs into one 32-bit hash value. This must be fast, but
1893 * latency isn't quite as critical, as there is a fair bit of additional
1894 * work done before the hash value is used.
1895 */
1896static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1897{
1898 y ^= x * GOLDEN_RATIO_64;
1899 y *= GOLDEN_RATIO_64;
1900 return y >> 32;
1901}
1902
1903#else /* 32-bit case */
1904
1905/*
1906 * Mixing scores (in bits) for (7,20):
1907 * Input delta: 1-bit 2-bit
1908 * 1 round: 330.3 9201.6
1909 * 2 rounds: 1246.4 25475.4
1910 * 3 rounds: 1907.1 31295.1
1911 * 4 rounds: 2042.3 31718.6
1912 * Perfect: 2048 31744
1913 * (32*64) (32*31/2 * 64)
1914 */
1915#define HASH_MIX(x, y, a) \
1916 ( x ^= (a), \
1917 y ^= x, x = rol32(x, 7),\
1918 x += y, y = rol32(y,20),\
1919 y *= 9 )
1920
1921static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1922{
1923 /* Use arch-optimized multiply if one exists */
1924 return __hash_32(y ^ __hash_32(x));
1925}
1926
1927#endif
1928
1929/*
1930 * Return the hash of a string of known length. This is carfully
1931 * designed to match hash_name(), which is the more critical function.
1932 * In particular, we must end by hashing a final word containing 0..7
1933 * payload bytes, to match the way that hash_name() iterates until it
1934 * finds the delimiter after the name.
1935 */
1936unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1937{
1938 unsigned long a, x = 0, y = (unsigned long)salt;
1939
1940 for (;;) {
1941 if (!len)
1942 goto done;
1943 a = load_unaligned_zeropad(name);
1944 if (len < sizeof(unsigned long))
1945 break;
1946 HASH_MIX(x, y, a);
1947 name += sizeof(unsigned long);
1948 len -= sizeof(unsigned long);
1949 }
1950 x ^= a & bytemask_from_count(len);
1951done:
1952 return fold_hash(x, y);
1953}
1954EXPORT_SYMBOL(full_name_hash);
1955
1956/* Return the "hash_len" (hash and length) of a null-terminated string */
1957u64 hashlen_string(const void *salt, const char *name)
1958{
1959 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1960 unsigned long adata, mask, len;
1961 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1962
1963 len = 0;
1964 goto inside;
1965
1966 do {
1967 HASH_MIX(x, y, a);
1968 len += sizeof(unsigned long);
1969inside:
1970 a = load_unaligned_zeropad(name+len);
1971 } while (!has_zero(a, &adata, &constants));
1972
1973 adata = prep_zero_mask(a, adata, &constants);
1974 mask = create_zero_mask(adata);
1975 x ^= a & zero_bytemask(mask);
1976
1977 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1978}
1979EXPORT_SYMBOL(hashlen_string);
1980
1981/*
1982 * Calculate the length and hash of the path component, and
1983 * return the "hash_len" as the result.
1984 */
1985static inline u64 hash_name(const void *salt, const char *name)
1986{
1987 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1988 unsigned long adata, bdata, mask, len;
1989 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1990
1991 len = 0;
1992 goto inside;
1993
1994 do {
1995 HASH_MIX(x, y, a);
1996 len += sizeof(unsigned long);
1997inside:
1998 a = load_unaligned_zeropad(name+len);
1999 b = a ^ REPEAT_BYTE('/');
2000 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2001
2002 adata = prep_zero_mask(a, adata, &constants);
2003 bdata = prep_zero_mask(b, bdata, &constants);
2004 mask = create_zero_mask(adata | bdata);
2005 x ^= a & zero_bytemask(mask);
2006
2007 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2008}
2009
2010#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2011
2012/* Return the hash of a string of known length */
2013unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2014{
2015 unsigned long hash = init_name_hash(salt);
2016 while (len--)
2017 hash = partial_name_hash((unsigned char)*name++, hash);
2018 return end_name_hash(hash);
2019}
2020EXPORT_SYMBOL(full_name_hash);
2021
2022/* Return the "hash_len" (hash and length) of a null-terminated string */
2023u64 hashlen_string(const void *salt, const char *name)
2024{
2025 unsigned long hash = init_name_hash(salt);
2026 unsigned long len = 0, c;
2027
2028 c = (unsigned char)*name;
2029 while (c) {
2030 len++;
2031 hash = partial_name_hash(c, hash);
2032 c = (unsigned char)name[len];
2033 }
2034 return hashlen_create(end_name_hash(hash), len);
2035}
2036EXPORT_SYMBOL(hashlen_string);
2037
2038/*
2039 * We know there's a real path component here of at least
2040 * one character.
2041 */
2042static inline u64 hash_name(const void *salt, const char *name)
2043{
2044 unsigned long hash = init_name_hash(salt);
2045 unsigned long len = 0, c;
2046
2047 c = (unsigned char)*name;
2048 do {
2049 len++;
2050 hash = partial_name_hash(c, hash);
2051 c = (unsigned char)name[len];
2052 } while (c && c != '/');
2053 return hashlen_create(end_name_hash(hash), len);
2054}
2055
2056#endif
2057
2058/*
2059 * Name resolution.
2060 * This is the basic name resolution function, turning a pathname into
2061 * the final dentry. We expect 'base' to be positive and a directory.
2062 *
2063 * Returns 0 and nd will have valid dentry and mnt on success.
2064 * Returns error and drops reference to input namei data on failure.
2065 */
2066static int link_path_walk(const char *name, struct nameidata *nd)
2067{
2068 int err;
2069
2070 if (IS_ERR(name))
2071 return PTR_ERR(name);
2072 while (*name=='/')
2073 name++;
2074 if (!*name)
2075 return 0;
2076
2077 /* At this point we know we have a real path component. */
2078 for(;;) {
2079 u64 hash_len;
2080 int type;
2081
2082 err = may_lookup(nd);
2083 if (err)
2084 return err;
2085
2086 hash_len = hash_name(nd->path.dentry, name);
2087
2088 type = LAST_NORM;
2089 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2090 case 2:
2091 if (name[1] == '.') {
2092 type = LAST_DOTDOT;
2093 nd->flags |= LOOKUP_JUMPED;
2094 }
2095 break;
2096 case 1:
2097 type = LAST_DOT;
2098 }
2099 if (likely(type == LAST_NORM)) {
2100 struct dentry *parent = nd->path.dentry;
2101 nd->flags &= ~LOOKUP_JUMPED;
2102 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2103 struct qstr this = { { .hash_len = hash_len }, .name = name };
2104 err = parent->d_op->d_hash(parent, &this);
2105 if (err < 0)
2106 return err;
2107 hash_len = this.hash_len;
2108 name = this.name;
2109 }
2110 }
2111
2112 nd->last.hash_len = hash_len;
2113 nd->last.name = name;
2114 nd->last_type = type;
2115
2116 name += hashlen_len(hash_len);
2117 if (!*name)
2118 goto OK;
2119 /*
2120 * If it wasn't NUL, we know it was '/'. Skip that
2121 * slash, and continue until no more slashes.
2122 */
2123 do {
2124 name++;
2125 } while (unlikely(*name == '/'));
2126 if (unlikely(!*name)) {
2127OK:
2128 /* pathname body, done */
2129 if (!nd->depth)
2130 return 0;
2131 name = nd->stack[nd->depth - 1].name;
2132 /* trailing symlink, done */
2133 if (!name)
2134 return 0;
2135 /* last component of nested symlink */
2136 err = walk_component(nd, WALK_FOLLOW);
2137 } else {
2138 /* not the last component */
2139 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2140 }
2141 if (err < 0)
2142 return err;
2143
2144 if (err) {
2145 const char *s = get_link(nd);
2146
2147 if (IS_ERR(s))
2148 return PTR_ERR(s);
2149 err = 0;
2150 if (unlikely(!s)) {
2151 /* jumped */
2152 put_link(nd);
2153 } else {
2154 nd->stack[nd->depth - 1].name = name;
2155 name = s;
2156 continue;
2157 }
2158 }
2159 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2160 if (nd->flags & LOOKUP_RCU) {
2161 if (unlazy_walk(nd))
2162 return -ECHILD;
2163 }
2164 return -ENOTDIR;
2165 }
2166 }
2167}
2168
2169/* must be paired with terminate_walk() */
2170static const char *path_init(struct nameidata *nd, unsigned flags)
2171{
2172 const char *s = nd->name->name;
2173
2174 if (!*s)
2175 flags &= ~LOOKUP_RCU;
2176 if (flags & LOOKUP_RCU)
2177 rcu_read_lock();
2178
2179 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2180 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2181 nd->depth = 0;
2182 if (flags & LOOKUP_ROOT) {
2183 struct dentry *root = nd->root.dentry;
2184 struct inode *inode = root->d_inode;
2185 if (*s && unlikely(!d_can_lookup(root)))
2186 return ERR_PTR(-ENOTDIR);
2187 nd->path = nd->root;
2188 nd->inode = inode;
2189 if (flags & LOOKUP_RCU) {
2190 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2191 nd->root_seq = nd->seq;
2192 nd->m_seq = read_seqbegin(&mount_lock);
2193 } else {
2194 path_get(&nd->path);
2195 }
2196 return s;
2197 }
2198
2199 nd->root.mnt = NULL;
2200 nd->path.mnt = NULL;
2201 nd->path.dentry = NULL;
2202
2203 nd->m_seq = read_seqbegin(&mount_lock);
2204 if (*s == '/') {
2205 set_root(nd);
2206 if (likely(!nd_jump_root(nd)))
2207 return s;
2208 return ERR_PTR(-ECHILD);
2209 } else if (nd->dfd == AT_FDCWD) {
2210 if (flags & LOOKUP_RCU) {
2211 struct fs_struct *fs = current->fs;
2212 unsigned seq;
2213
2214 do {
2215 seq = read_seqcount_begin(&fs->seq);
2216 nd->path = fs->pwd;
2217 nd->inode = nd->path.dentry->d_inode;
2218 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2219 } while (read_seqcount_retry(&fs->seq, seq));
2220 } else {
2221 get_fs_pwd(current->fs, &nd->path);
2222 nd->inode = nd->path.dentry->d_inode;
2223 }
2224 return s;
2225 } else {
2226 /* Caller must check execute permissions on the starting path component */
2227 struct fd f = fdget_raw(nd->dfd);
2228 struct dentry *dentry;
2229
2230 if (!f.file)
2231 return ERR_PTR(-EBADF);
2232
2233 dentry = f.file->f_path.dentry;
2234
2235 if (*s && unlikely(!d_can_lookup(dentry))) {
2236 fdput(f);
2237 return ERR_PTR(-ENOTDIR);
2238 }
2239
2240 nd->path = f.file->f_path;
2241 if (flags & LOOKUP_RCU) {
2242 nd->inode = nd->path.dentry->d_inode;
2243 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2244 } else {
2245 path_get(&nd->path);
2246 nd->inode = nd->path.dentry->d_inode;
2247 }
2248 fdput(f);
2249 return s;
2250 }
2251}
2252
2253static const char *trailing_symlink(struct nameidata *nd)
2254{
2255 const char *s;
2256 int error = may_follow_link(nd);
2257 if (unlikely(error))
2258 return ERR_PTR(error);
2259 nd->flags |= LOOKUP_PARENT;
2260 nd->stack[0].name = NULL;
2261 s = get_link(nd);
2262 return s ? s : "";
2263}
2264
2265static inline int lookup_last(struct nameidata *nd)
2266{
2267 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2268 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2269
2270 nd->flags &= ~LOOKUP_PARENT;
2271 return walk_component(nd, 0);
2272}
2273
2274static int handle_lookup_down(struct nameidata *nd)
2275{
2276 struct path path = nd->path;
2277 struct inode *inode = nd->inode;
2278 unsigned seq = nd->seq;
2279 int err;
2280
2281 if (nd->flags & LOOKUP_RCU) {
2282 /*
2283 * don't bother with unlazy_walk on failure - we are
2284 * at the very beginning of walk, so we lose nothing
2285 * if we simply redo everything in non-RCU mode
2286 */
2287 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2288 return -ECHILD;
2289 } else {
2290 dget(path.dentry);
2291 err = follow_managed(&path, nd);
2292 if (unlikely(err < 0))
2293 return err;
2294 inode = d_backing_inode(path.dentry);
2295 seq = 0;
2296 }
2297 path_to_nameidata(&path, nd);
2298 nd->inode = inode;
2299 nd->seq = seq;
2300 return 0;
2301}
2302
2303/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2304static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2305{
2306 const char *s = path_init(nd, flags);
2307 int err;
2308
2309 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2310 err = handle_lookup_down(nd);
2311 if (unlikely(err < 0))
2312 s = ERR_PTR(err);
2313 }
2314
2315 while (!(err = link_path_walk(s, nd))
2316 && ((err = lookup_last(nd)) > 0)) {
2317 s = trailing_symlink(nd);
2318 }
2319 if (!err)
2320 err = complete_walk(nd);
2321
2322 if (!err && nd->flags & LOOKUP_DIRECTORY)
2323 if (!d_can_lookup(nd->path.dentry))
2324 err = -ENOTDIR;
2325 if (!err) {
2326 *path = nd->path;
2327 nd->path.mnt = NULL;
2328 nd->path.dentry = NULL;
2329 }
2330 terminate_walk(nd);
2331 return err;
2332}
2333
2334int filename_lookup(int dfd, struct filename *name, unsigned flags,
2335 struct path *path, struct path *root)
2336{
2337 int retval;
2338 struct nameidata nd;
2339 if (IS_ERR(name))
2340 return PTR_ERR(name);
2341 if (unlikely(root)) {
2342 nd.root = *root;
2343 flags |= LOOKUP_ROOT;
2344 }
2345 set_nameidata(&nd, dfd, name);
2346 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2347 if (unlikely(retval == -ECHILD))
2348 retval = path_lookupat(&nd, flags, path);
2349 if (unlikely(retval == -ESTALE))
2350 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2351
2352 if (likely(!retval))
2353 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2354 restore_nameidata();
2355 putname(name);
2356 return retval;
2357}
2358
2359/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2360static int path_parentat(struct nameidata *nd, unsigned flags,
2361 struct path *parent)
2362{
2363 const char *s = path_init(nd, flags);
2364 int err = link_path_walk(s, nd);
2365 if (!err)
2366 err = complete_walk(nd);
2367 if (!err) {
2368 *parent = nd->path;
2369 nd->path.mnt = NULL;
2370 nd->path.dentry = NULL;
2371 }
2372 terminate_walk(nd);
2373 return err;
2374}
2375
2376static struct filename *filename_parentat(int dfd, struct filename *name,
2377 unsigned int flags, struct path *parent,
2378 struct qstr *last, int *type)
2379{
2380 int retval;
2381 struct nameidata nd;
2382
2383 if (IS_ERR(name))
2384 return name;
2385 set_nameidata(&nd, dfd, name);
2386 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2387 if (unlikely(retval == -ECHILD))
2388 retval = path_parentat(&nd, flags, parent);
2389 if (unlikely(retval == -ESTALE))
2390 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2391 if (likely(!retval)) {
2392 *last = nd.last;
2393 *type = nd.last_type;
2394 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2395 } else {
2396 putname(name);
2397 name = ERR_PTR(retval);
2398 }
2399 restore_nameidata();
2400 return name;
2401}
2402
2403/* does lookup, returns the object with parent locked */
2404struct dentry *kern_path_locked(const char *name, struct path *path)
2405{
2406 struct filename *filename;
2407 struct dentry *d;
2408 struct qstr last;
2409 int type;
2410
2411 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2412 &last, &type);
2413 if (IS_ERR(filename))
2414 return ERR_CAST(filename);
2415 if (unlikely(type != LAST_NORM)) {
2416 path_put(path);
2417 putname(filename);
2418 return ERR_PTR(-EINVAL);
2419 }
2420 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2421 d = __lookup_hash(&last, path->dentry, 0);
2422 if (IS_ERR(d)) {
2423 inode_unlock(path->dentry->d_inode);
2424 path_put(path);
2425 }
2426 putname(filename);
2427 return d;
2428}
2429
2430int kern_path(const char *name, unsigned int flags, struct path *path)
2431{
2432 return filename_lookup(AT_FDCWD, getname_kernel(name),
2433 flags, path, NULL);
2434}
2435EXPORT_SYMBOL(kern_path);
2436
2437/**
2438 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2439 * @dentry: pointer to dentry of the base directory
2440 * @mnt: pointer to vfs mount of the base directory
2441 * @name: pointer to file name
2442 * @flags: lookup flags
2443 * @path: pointer to struct path to fill
2444 */
2445int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2446 const char *name, unsigned int flags,
2447 struct path *path)
2448{
2449 struct path root = {.mnt = mnt, .dentry = dentry};
2450 /* the first argument of filename_lookup() is ignored with root */
2451 return filename_lookup(AT_FDCWD, getname_kernel(name),
2452 flags , path, &root);
2453}
2454EXPORT_SYMBOL(vfs_path_lookup);
2455
2456static int lookup_one_len_common(const char *name, struct dentry *base,
2457 int len, struct qstr *this)
2458{
2459 this->name = name;
2460 this->len = len;
2461 this->hash = full_name_hash(base, name, len);
2462 if (!len)
2463 return -EACCES;
2464
2465 if (unlikely(name[0] == '.')) {
2466 if (len < 2 || (len == 2 && name[1] == '.'))
2467 return -EACCES;
2468 }
2469
2470 while (len--) {
2471 unsigned int c = *(const unsigned char *)name++;
2472 if (c == '/' || c == '\0')
2473 return -EACCES;
2474 }
2475 /*
2476 * See if the low-level filesystem might want
2477 * to use its own hash..
2478 */
2479 if (base->d_flags & DCACHE_OP_HASH) {
2480 int err = base->d_op->d_hash(base, this);
2481 if (err < 0)
2482 return err;
2483 }
2484
2485 return inode_permission(base->d_inode, MAY_EXEC);
2486}
2487
2488/**
2489 * try_lookup_one_len - filesystem helper to lookup single pathname component
2490 * @name: pathname component to lookup
2491 * @base: base directory to lookup from
2492 * @len: maximum length @len should be interpreted to
2493 *
2494 * Look up a dentry by name in the dcache, returning NULL if it does not
2495 * currently exist. The function does not try to create a dentry.
2496 *
2497 * Note that this routine is purely a helper for filesystem usage and should
2498 * not be called by generic code.
2499 *
2500 * The caller must hold base->i_mutex.
2501 */
2502struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2503{
2504 struct qstr this;
2505 int err;
2506
2507 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2508
2509 err = lookup_one_len_common(name, base, len, &this);
2510 if (err)
2511 return ERR_PTR(err);
2512
2513 return lookup_dcache(&this, base, 0);
2514}
2515EXPORT_SYMBOL(try_lookup_one_len);
2516
2517/**
2518 * lookup_one_len - filesystem helper to lookup single pathname component
2519 * @name: pathname component to lookup
2520 * @base: base directory to lookup from
2521 * @len: maximum length @len should be interpreted to
2522 *
2523 * Note that this routine is purely a helper for filesystem usage and should
2524 * not be called by generic code.
2525 *
2526 * The caller must hold base->i_mutex.
2527 */
2528struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2529{
2530 struct dentry *dentry;
2531 struct qstr this;
2532 int err;
2533
2534 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2535
2536 err = lookup_one_len_common(name, base, len, &this);
2537 if (err)
2538 return ERR_PTR(err);
2539
2540 dentry = lookup_dcache(&this, base, 0);
2541 return dentry ? dentry : __lookup_slow(&this, base, 0);
2542}
2543EXPORT_SYMBOL(lookup_one_len);
2544
2545/**
2546 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2547 * @name: pathname component to lookup
2548 * @base: base directory to lookup from
2549 * @len: maximum length @len should be interpreted to
2550 *
2551 * Note that this routine is purely a helper for filesystem usage and should
2552 * not be called by generic code.
2553 *
2554 * Unlike lookup_one_len, it should be called without the parent
2555 * i_mutex held, and will take the i_mutex itself if necessary.
2556 */
2557struct dentry *lookup_one_len_unlocked(const char *name,
2558 struct dentry *base, int len)
2559{
2560 struct qstr this;
2561 int err;
2562 struct dentry *ret;
2563
2564 err = lookup_one_len_common(name, base, len, &this);
2565 if (err)
2566 return ERR_PTR(err);
2567
2568 ret = lookup_dcache(&this, base, 0);
2569 if (!ret)
2570 ret = lookup_slow(&this, base, 0);
2571 return ret;
2572}
2573EXPORT_SYMBOL(lookup_one_len_unlocked);
2574
2575#ifdef CONFIG_UNIX98_PTYS
2576int path_pts(struct path *path)
2577{
2578 /* Find something mounted on "pts" in the same directory as
2579 * the input path.
2580 */
2581 struct dentry *child, *parent;
2582 struct qstr this;
2583 int ret;
2584
2585 ret = path_parent_directory(path);
2586 if (ret)
2587 return ret;
2588
2589 parent = path->dentry;
2590 this.name = "pts";
2591 this.len = 3;
2592 child = d_hash_and_lookup(parent, &this);
2593 if (!child)
2594 return -ENOENT;
2595
2596 path->dentry = child;
2597 dput(parent);
2598 follow_mount(path);
2599 return 0;
2600}
2601#endif
2602
2603int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2604 struct path *path, int *empty)
2605{
2606 return filename_lookup(dfd, getname_flags(name, flags, empty),
2607 flags, path, NULL);
2608}
2609EXPORT_SYMBOL(user_path_at_empty);
2610
2611/**
2612 * mountpoint_last - look up last component for umount
2613 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2614 *
2615 * This is a special lookup_last function just for umount. In this case, we
2616 * need to resolve the path without doing any revalidation.
2617 *
2618 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2619 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2620 * in almost all cases, this lookup will be served out of the dcache. The only
2621 * cases where it won't are if nd->last refers to a symlink or the path is
2622 * bogus and it doesn't exist.
2623 *
2624 * Returns:
2625 * -error: if there was an error during lookup. This includes -ENOENT if the
2626 * lookup found a negative dentry.
2627 *
2628 * 0: if we successfully resolved nd->last and found it to not to be a
2629 * symlink that needs to be followed.
2630 *
2631 * 1: if we successfully resolved nd->last and found it to be a symlink
2632 * that needs to be followed.
2633 */
2634static int
2635mountpoint_last(struct nameidata *nd)
2636{
2637 int error = 0;
2638 struct dentry *dir = nd->path.dentry;
2639 struct path path;
2640
2641 /* If we're in rcuwalk, drop out of it to handle last component */
2642 if (nd->flags & LOOKUP_RCU) {
2643 if (unlazy_walk(nd))
2644 return -ECHILD;
2645 }
2646
2647 nd->flags &= ~LOOKUP_PARENT;
2648
2649 if (unlikely(nd->last_type != LAST_NORM)) {
2650 error = handle_dots(nd, nd->last_type);
2651 if (error)
2652 return error;
2653 path.dentry = dget(nd->path.dentry);
2654 } else {
2655 path.dentry = d_lookup(dir, &nd->last);
2656 if (!path.dentry) {
2657 /*
2658 * No cached dentry. Mounted dentries are pinned in the
2659 * cache, so that means that this dentry is probably
2660 * a symlink or the path doesn't actually point
2661 * to a mounted dentry.
2662 */
2663 path.dentry = lookup_slow(&nd->last, dir,
2664 nd->flags | LOOKUP_NO_REVAL);
2665 if (IS_ERR(path.dentry))
2666 return PTR_ERR(path.dentry);
2667 }
2668 }
2669 if (d_is_negative(path.dentry)) {
2670 dput(path.dentry);
2671 return -ENOENT;
2672 }
2673 path.mnt = nd->path.mnt;
2674 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2675}
2676
2677/**
2678 * path_mountpoint - look up a path to be umounted
2679 * @nd: lookup context
2680 * @flags: lookup flags
2681 * @path: pointer to container for result
2682 *
2683 * Look up the given name, but don't attempt to revalidate the last component.
2684 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2685 */
2686static int
2687path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2688{
2689 const char *s = path_init(nd, flags);
2690 int err;
2691
2692 while (!(err = link_path_walk(s, nd)) &&
2693 (err = mountpoint_last(nd)) > 0) {
2694 s = trailing_symlink(nd);
2695 }
2696 if (!err) {
2697 *path = nd->path;
2698 nd->path.mnt = NULL;
2699 nd->path.dentry = NULL;
2700 follow_mount(path);
2701 }
2702 terminate_walk(nd);
2703 return err;
2704}
2705
2706static int
2707filename_mountpoint(int dfd, struct filename *name, struct path *path,
2708 unsigned int flags)
2709{
2710 struct nameidata nd;
2711 int error;
2712 if (IS_ERR(name))
2713 return PTR_ERR(name);
2714 set_nameidata(&nd, dfd, name);
2715 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2716 if (unlikely(error == -ECHILD))
2717 error = path_mountpoint(&nd, flags, path);
2718 if (unlikely(error == -ESTALE))
2719 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2720 if (likely(!error))
2721 audit_inode(name, path->dentry, flags & LOOKUP_NO_EVAL);
2722 restore_nameidata();
2723 putname(name);
2724 return error;
2725}
2726
2727/**
2728 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2729 * @dfd: directory file descriptor
2730 * @name: pathname from userland
2731 * @flags: lookup flags
2732 * @path: pointer to container to hold result
2733 *
2734 * A umount is a special case for path walking. We're not actually interested
2735 * in the inode in this situation, and ESTALE errors can be a problem. We
2736 * simply want track down the dentry and vfsmount attached at the mountpoint
2737 * and avoid revalidating the last component.
2738 *
2739 * Returns 0 and populates "path" on success.
2740 */
2741int
2742user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2743 struct path *path)
2744{
2745 return filename_mountpoint(dfd, getname(name), path, flags);
2746}
2747
2748int
2749kern_path_mountpoint(int dfd, const char *name, struct path *path,
2750 unsigned int flags)
2751{
2752 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2753}
2754EXPORT_SYMBOL(kern_path_mountpoint);
2755
2756int __check_sticky(struct inode *dir, struct inode *inode)
2757{
2758 kuid_t fsuid = current_fsuid();
2759
2760 if (uid_eq(inode->i_uid, fsuid))
2761 return 0;
2762 if (uid_eq(dir->i_uid, fsuid))
2763 return 0;
2764 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2765}
2766EXPORT_SYMBOL(__check_sticky);
2767
2768/*
2769 * Check whether we can remove a link victim from directory dir, check
2770 * whether the type of victim is right.
2771 * 1. We can't do it if dir is read-only (done in permission())
2772 * 2. We should have write and exec permissions on dir
2773 * 3. We can't remove anything from append-only dir
2774 * 4. We can't do anything with immutable dir (done in permission())
2775 * 5. If the sticky bit on dir is set we should either
2776 * a. be owner of dir, or
2777 * b. be owner of victim, or
2778 * c. have CAP_FOWNER capability
2779 * 6. If the victim is append-only or immutable we can't do antyhing with
2780 * links pointing to it.
2781 * 7. If the victim has an unknown uid or gid we can't change the inode.
2782 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2783 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2784 * 10. We can't remove a root or mountpoint.
2785 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2786 * nfs_async_unlink().
2787 */
2788static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2789{
2790 struct inode *inode = d_backing_inode(victim);
2791 int error;
2792
2793 if (d_is_negative(victim))
2794 return -ENOENT;
2795 BUG_ON(!inode);
2796
2797 BUG_ON(victim->d_parent->d_inode != dir);
2798
2799 /* Inode writeback is not safe when the uid or gid are invalid. */
2800 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2801 return -EOVERFLOW;
2802
2803 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2804
2805 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2806 if (error)
2807 return error;
2808 if (IS_APPEND(dir))
2809 return -EPERM;
2810
2811 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2812 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2813 return -EPERM;
2814 if (isdir) {
2815 if (!d_is_dir(victim))
2816 return -ENOTDIR;
2817 if (IS_ROOT(victim))
2818 return -EBUSY;
2819 } else if (d_is_dir(victim))
2820 return -EISDIR;
2821 if (IS_DEADDIR(dir))
2822 return -ENOENT;
2823 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2824 return -EBUSY;
2825 return 0;
2826}
2827
2828/* Check whether we can create an object with dentry child in directory
2829 * dir.
2830 * 1. We can't do it if child already exists (open has special treatment for
2831 * this case, but since we are inlined it's OK)
2832 * 2. We can't do it if dir is read-only (done in permission())
2833 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2834 * 4. We should have write and exec permissions on dir
2835 * 5. We can't do it if dir is immutable (done in permission())
2836 */
2837static inline int may_create(struct inode *dir, struct dentry *child)
2838{
2839 struct user_namespace *s_user_ns;
2840 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2841 if (child->d_inode)
2842 return -EEXIST;
2843 if (IS_DEADDIR(dir))
2844 return -ENOENT;
2845 s_user_ns = dir->i_sb->s_user_ns;
2846 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2847 !kgid_has_mapping(s_user_ns, current_fsgid()))
2848 return -EOVERFLOW;
2849 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2850}
2851
2852/*
2853 * p1 and p2 should be directories on the same fs.
2854 */
2855struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2856{
2857 struct dentry *p;
2858
2859 if (p1 == p2) {
2860 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2861 return NULL;
2862 }
2863
2864 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2865
2866 p = d_ancestor(p2, p1);
2867 if (p) {
2868 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2869 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2870 return p;
2871 }
2872
2873 p = d_ancestor(p1, p2);
2874 if (p) {
2875 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2876 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2877 return p;
2878 }
2879
2880 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2881 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2882 return NULL;
2883}
2884EXPORT_SYMBOL(lock_rename);
2885
2886void unlock_rename(struct dentry *p1, struct dentry *p2)
2887{
2888 inode_unlock(p1->d_inode);
2889 if (p1 != p2) {
2890 inode_unlock(p2->d_inode);
2891 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2892 }
2893}
2894EXPORT_SYMBOL(unlock_rename);
2895
2896int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2897 bool want_excl)
2898{
2899 int error = may_create(dir, dentry);
2900 if (error)
2901 return error;
2902
2903 if (!dir->i_op->create)
2904 return -EACCES; /* shouldn't it be ENOSYS? */
2905 mode &= S_IALLUGO;
2906 mode |= S_IFREG;
2907 error = security_inode_create(dir, dentry, mode);
2908 if (error)
2909 return error;
2910 error = dir->i_op->create(dir, dentry, mode, want_excl);
2911 if (!error)
2912 fsnotify_create(dir, dentry);
2913 return error;
2914}
2915EXPORT_SYMBOL(vfs_create);
2916
2917int vfs_mkobj(struct dentry *dentry, umode_t mode,
2918 int (*f)(struct dentry *, umode_t, void *),
2919 void *arg)
2920{
2921 struct inode *dir = dentry->d_parent->d_inode;
2922 int error = may_create(dir, dentry);
2923 if (error)
2924 return error;
2925
2926 mode &= S_IALLUGO;
2927 mode |= S_IFREG;
2928 error = security_inode_create(dir, dentry, mode);
2929 if (error)
2930 return error;
2931 error = f(dentry, mode, arg);
2932 if (!error)
2933 fsnotify_create(dir, dentry);
2934 return error;
2935}
2936EXPORT_SYMBOL(vfs_mkobj);
2937
2938bool may_open_dev(const struct path *path)
2939{
2940 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2941 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2942}
2943
2944static int may_open(const struct path *path, int acc_mode, int flag)
2945{
2946 struct dentry *dentry = path->dentry;
2947 struct inode *inode = dentry->d_inode;
2948 int error;
2949
2950 if (!inode)
2951 return -ENOENT;
2952
2953 switch (inode->i_mode & S_IFMT) {
2954 case S_IFLNK:
2955 return -ELOOP;
2956 case S_IFDIR:
2957 if (acc_mode & MAY_WRITE)
2958 return -EISDIR;
2959 break;
2960 case S_IFBLK:
2961 case S_IFCHR:
2962 if (!may_open_dev(path))
2963 return -EACCES;
2964 /*FALLTHRU*/
2965 case S_IFIFO:
2966 case S_IFSOCK:
2967 flag &= ~O_TRUNC;
2968 break;
2969 }
2970
2971 error = inode_permission(inode, MAY_OPEN | acc_mode);
2972 if (error)
2973 return error;
2974
2975 /*
2976 * An append-only file must be opened in append mode for writing.
2977 */
2978 if (IS_APPEND(inode)) {
2979 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2980 return -EPERM;
2981 if (flag & O_TRUNC)
2982 return -EPERM;
2983 }
2984
2985 /* O_NOATIME can only be set by the owner or superuser */
2986 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2987 return -EPERM;
2988
2989 return 0;
2990}
2991
2992static int handle_truncate(struct file *filp)
2993{
2994 const struct path *path = &filp->f_path;
2995 struct inode *inode = path->dentry->d_inode;
2996 int error = get_write_access(inode);
2997 if (error)
2998 return error;
2999 /*
3000 * Refuse to truncate files with mandatory locks held on them.
3001 */
3002 error = locks_verify_locked(filp);
3003 if (!error)
3004 error = security_path_truncate(path);
3005 if (!error) {
3006 error = do_truncate(path->dentry, 0,
3007 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3008 filp);
3009 }
3010 put_write_access(inode);
3011 return error;
3012}
3013
3014static inline int open_to_namei_flags(int flag)
3015{
3016 if ((flag & O_ACCMODE) == 3)
3017 flag--;
3018 return flag;
3019}
3020
3021static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
3022{
3023 struct user_namespace *s_user_ns;
3024 int error = security_path_mknod(dir, dentry, mode, 0);
3025 if (error)
3026 return error;
3027
3028 s_user_ns = dir->dentry->d_sb->s_user_ns;
3029 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
3030 !kgid_has_mapping(s_user_ns, current_fsgid()))
3031 return -EOVERFLOW;
3032
3033 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
3034 if (error)
3035 return error;
3036
3037 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3038}
3039
3040/*
3041 * Attempt to atomically look up, create and open a file from a negative
3042 * dentry.
3043 *
3044 * Returns 0 if successful. The file will have been created and attached to
3045 * @file by the filesystem calling finish_open().
3046 *
3047 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3048 * be set. The caller will need to perform the open themselves. @path will
3049 * have been updated to point to the new dentry. This may be negative.
3050 *
3051 * Returns an error code otherwise.
3052 */
3053static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3054 struct path *path, struct file *file,
3055 const struct open_flags *op,
3056 int open_flag, umode_t mode)
3057{
3058 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3059 struct inode *dir = nd->path.dentry->d_inode;
3060 int error;
3061
3062 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3063 open_flag &= ~O_TRUNC;
3064
3065 if (nd->flags & LOOKUP_DIRECTORY)
3066 open_flag |= O_DIRECTORY;
3067
3068 file->f_path.dentry = DENTRY_NOT_SET;
3069 file->f_path.mnt = nd->path.mnt;
3070 error = dir->i_op->atomic_open(dir, dentry, file,
3071 open_to_namei_flags(open_flag), mode);
3072 d_lookup_done(dentry);
3073 if (!error) {
3074 if (file->f_mode & FMODE_OPENED) {
3075 /*
3076 * We didn't have the inode before the open, so check open
3077 * permission here.
3078 */
3079 int acc_mode = op->acc_mode;
3080 if (file->f_mode & FMODE_CREATED) {
3081 WARN_ON(!(open_flag & O_CREAT));
3082 fsnotify_create(dir, dentry);
3083 acc_mode = 0;
3084 }
3085 error = may_open(&file->f_path, acc_mode, open_flag);
3086 if (WARN_ON(error > 0))
3087 error = -EINVAL;
3088 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3089 error = -EIO;
3090 } else {
3091 if (file->f_path.dentry) {
3092 dput(dentry);
3093 dentry = file->f_path.dentry;
3094 }
3095 if (file->f_mode & FMODE_CREATED)
3096 fsnotify_create(dir, dentry);
3097 if (unlikely(d_is_negative(dentry))) {
3098 error = -ENOENT;
3099 } else {
3100 path->dentry = dentry;
3101 path->mnt = nd->path.mnt;
3102 return 0;
3103 }
3104 }
3105 }
3106 dput(dentry);
3107 return error;
3108}
3109
3110/*
3111 * Look up and maybe create and open the last component.
3112 *
3113 * Must be called with parent locked (exclusive in O_CREAT case).
3114 *
3115 * Returns 0 on success, that is, if
3116 * the file was successfully atomically created (if necessary) and opened, or
3117 * the file was not completely opened at this time, though lookups and
3118 * creations were performed.
3119 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3120 * In the latter case dentry returned in @path might be negative if O_CREAT
3121 * hadn't been specified.
3122 *
3123 * An error code is returned on failure.
3124 */
3125static int lookup_open(struct nameidata *nd, struct path *path,
3126 struct file *file,
3127 const struct open_flags *op,
3128 bool got_write)
3129{
3130 struct dentry *dir = nd->path.dentry;
3131 struct inode *dir_inode = dir->d_inode;
3132 int open_flag = op->open_flag;
3133 struct dentry *dentry;
3134 int error, create_error = 0;
3135 umode_t mode = op->mode;
3136 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3137
3138 if (unlikely(IS_DEADDIR(dir_inode)))
3139 return -ENOENT;
3140
3141 file->f_mode &= ~FMODE_CREATED;
3142 dentry = d_lookup(dir, &nd->last);
3143 for (;;) {
3144 if (!dentry) {
3145 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3146 if (IS_ERR(dentry))
3147 return PTR_ERR(dentry);
3148 }
3149 if (d_in_lookup(dentry))
3150 break;
3151
3152 error = d_revalidate(dentry, nd->flags);
3153 if (likely(error > 0))
3154 break;
3155 if (error)
3156 goto out_dput;
3157 d_invalidate(dentry);
3158 dput(dentry);
3159 dentry = NULL;
3160 }
3161 if (dentry->d_inode) {
3162 /* Cached positive dentry: will open in f_op->open */
3163 goto out_no_open;
3164 }
3165
3166 /*
3167 * Checking write permission is tricky, bacuse we don't know if we are
3168 * going to actually need it: O_CREAT opens should work as long as the
3169 * file exists. But checking existence breaks atomicity. The trick is
3170 * to check access and if not granted clear O_CREAT from the flags.
3171 *
3172 * Another problem is returing the "right" error value (e.g. for an
3173 * O_EXCL open we want to return EEXIST not EROFS).
3174 */
3175 if (open_flag & O_CREAT) {
3176 if (!IS_POSIXACL(dir->d_inode))
3177 mode &= ~current_umask();
3178 if (unlikely(!got_write)) {
3179 create_error = -EROFS;
3180 open_flag &= ~O_CREAT;
3181 if (open_flag & (O_EXCL | O_TRUNC))
3182 goto no_open;
3183 /* No side effects, safe to clear O_CREAT */
3184 } else {
3185 create_error = may_o_create(&nd->path, dentry, mode);
3186 if (create_error) {
3187 open_flag &= ~O_CREAT;
3188 if (open_flag & O_EXCL)
3189 goto no_open;
3190 }
3191 }
3192 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3193 unlikely(!got_write)) {
3194 /*
3195 * No O_CREATE -> atomicity not a requirement -> fall
3196 * back to lookup + open
3197 */
3198 goto no_open;
3199 }
3200
3201 if (dir_inode->i_op->atomic_open) {
3202 error = atomic_open(nd, dentry, path, file, op, open_flag,
3203 mode);
3204 if (unlikely(error == -ENOENT) && create_error)
3205 error = create_error;
3206 return error;
3207 }
3208
3209no_open:
3210 if (d_in_lookup(dentry)) {
3211 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3212 nd->flags);
3213 d_lookup_done(dentry);
3214 if (unlikely(res)) {
3215 if (IS_ERR(res)) {
3216 error = PTR_ERR(res);
3217 goto out_dput;
3218 }
3219 dput(dentry);
3220 dentry = res;
3221 }
3222 }
3223
3224 /* Negative dentry, just create the file */
3225 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3226 file->f_mode |= FMODE_CREATED;
3227 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3228 if (!dir_inode->i_op->create) {
3229 error = -EACCES;
3230 goto out_dput;
3231 }
3232 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3233 open_flag & O_EXCL);
3234 if (error)
3235 goto out_dput;
3236 fsnotify_create(dir_inode, dentry);
3237 }
3238 if (unlikely(create_error) && !dentry->d_inode) {
3239 error = create_error;
3240 goto out_dput;
3241 }
3242out_no_open:
3243 path->dentry = dentry;
3244 path->mnt = nd->path.mnt;
3245 return 0;
3246
3247out_dput:
3248 dput(dentry);
3249 return error;
3250}
3251
3252/*
3253 * Handle the last step of open()
3254 */
3255static int do_last(struct nameidata *nd,
3256 struct file *file, const struct open_flags *op)
3257{
3258 struct dentry *dir = nd->path.dentry;
3259 int open_flag = op->open_flag;
3260 bool will_truncate = (open_flag & O_TRUNC) != 0;
3261 bool got_write = false;
3262 int acc_mode = op->acc_mode;
3263 unsigned seq;
3264 struct inode *inode;
3265 struct path path;
3266 int error;
3267
3268 nd->flags &= ~LOOKUP_PARENT;
3269 nd->flags |= op->intent;
3270
3271 if (nd->last_type != LAST_NORM) {
3272 error = handle_dots(nd, nd->last_type);
3273 if (unlikely(error))
3274 return error;
3275 goto finish_open;
3276 }
3277
3278 if (!(open_flag & O_CREAT)) {
3279 if (nd->last.name[nd->last.len])
3280 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3281 /* we _can_ be in RCU mode here */
3282 error = lookup_fast(nd, &path, &inode, &seq);
3283 if (likely(error > 0))
3284 goto finish_lookup;
3285
3286 if (error < 0)
3287 return error;
3288
3289 BUG_ON(nd->inode != dir->d_inode);
3290 BUG_ON(nd->flags & LOOKUP_RCU);
3291 } else {
3292 /* create side of things */
3293 /*
3294 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3295 * has been cleared when we got to the last component we are
3296 * about to look up
3297 */
3298 error = complete_walk(nd);
3299 if (error)
3300 return error;
3301
3302 audit_inode(nd->name, dir, LOOKUP_PARENT);
3303 /* trailing slashes? */
3304 if (unlikely(nd->last.name[nd->last.len]))
3305 return -EISDIR;
3306 }
3307
3308 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3309 error = mnt_want_write(nd->path.mnt);
3310 if (!error)
3311 got_write = true;
3312 /*
3313 * do _not_ fail yet - we might not need that or fail with
3314 * a different error; let lookup_open() decide; we'll be
3315 * dropping this one anyway.
3316 */
3317 }
3318 if (open_flag & O_CREAT)
3319 inode_lock(dir->d_inode);
3320 else
3321 inode_lock_shared(dir->d_inode);
3322 error = lookup_open(nd, &path, file, op, got_write);
3323 if (open_flag & O_CREAT)
3324 inode_unlock(dir->d_inode);
3325 else
3326 inode_unlock_shared(dir->d_inode);
3327
3328 if (error)
3329 goto out;
3330
3331 if (file->f_mode & FMODE_OPENED) {
3332 if ((file->f_mode & FMODE_CREATED) ||
3333 !S_ISREG(file_inode(file)->i_mode))
3334 will_truncate = false;
3335
3336 audit_inode(nd->name, file->f_path.dentry, 0);
3337 goto opened;
3338 }
3339
3340 if (file->f_mode & FMODE_CREATED) {
3341 /* Don't check for write permission, don't truncate */
3342 open_flag &= ~O_TRUNC;
3343 will_truncate = false;
3344 acc_mode = 0;
3345 path_to_nameidata(&path, nd);
3346 goto finish_open_created;
3347 }
3348
3349 /*
3350 * If atomic_open() acquired write access it is dropped now due to
3351 * possible mount and symlink following (this might be optimized away if
3352 * necessary...)
3353 */
3354 if (got_write) {
3355 mnt_drop_write(nd->path.mnt);
3356 got_write = false;
3357 }
3358
3359 error = follow_managed(&path, nd);
3360 if (unlikely(error < 0))
3361 return error;
3362
3363 if (unlikely(d_is_negative(path.dentry))) {
3364 path_to_nameidata(&path, nd);
3365 return -ENOENT;
3366 }
3367
3368 /*
3369 * create/update audit record if it already exists.
3370 */
3371 audit_inode(nd->name, path.dentry, 0);
3372
3373 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3374 path_to_nameidata(&path, nd);
3375 return -EEXIST;
3376 }
3377
3378 seq = 0; /* out of RCU mode, so the value doesn't matter */
3379 inode = d_backing_inode(path.dentry);
3380finish_lookup:
3381 error = step_into(nd, &path, 0, inode, seq);
3382 if (unlikely(error))
3383 return error;
3384finish_open:
3385 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3386 error = complete_walk(nd);
3387 if (error)
3388 return error;
3389 audit_inode(nd->name, nd->path.dentry, 0);
3390 if (open_flag & O_CREAT) {
3391 error = -EISDIR;
3392 if (d_is_dir(nd->path.dentry))
3393 goto out;
3394 error = may_create_in_sticky(dir,
3395 d_backing_inode(nd->path.dentry));
3396 if (unlikely(error))
3397 goto out;
3398 }
3399 error = -ENOTDIR;
3400 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3401 goto out;
3402 if (!d_is_reg(nd->path.dentry))
3403 will_truncate = false;
3404
3405 if (will_truncate) {
3406 error = mnt_want_write(nd->path.mnt);
3407 if (error)
3408 goto out;
3409 got_write = true;
3410 }
3411finish_open_created:
3412 error = may_open(&nd->path, acc_mode, open_flag);
3413 if (error)
3414 goto out;
3415 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
3416 error = vfs_open(&nd->path, file);
3417 if (error)
3418 goto out;
3419opened:
3420 error = ima_file_check(file, op->acc_mode);
3421 if (!error && will_truncate)
3422 error = handle_truncate(file);
3423out:
3424 if (unlikely(error > 0)) {
3425 WARN_ON(1);
3426 error = -EINVAL;
3427 }
3428 if (got_write)
3429 mnt_drop_write(nd->path.mnt);
3430 return error;
3431}
3432
3433struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3434{
3435 struct dentry *child = NULL;
3436 struct inode *dir = dentry->d_inode;
3437 struct inode *inode;
3438 int error;
3439
3440 /* we want directory to be writable */
3441 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3442 if (error)
3443 goto out_err;
3444 error = -EOPNOTSUPP;
3445 if (!dir->i_op->tmpfile)
3446 goto out_err;
3447 error = -ENOMEM;
3448 child = d_alloc(dentry, &slash_name);
3449 if (unlikely(!child))
3450 goto out_err;
3451 error = dir->i_op->tmpfile(dir, child, mode);
3452 if (error)
3453 goto out_err;
3454 error = -ENOENT;
3455 inode = child->d_inode;
3456 if (unlikely(!inode))
3457 goto out_err;
3458 if (!(open_flag & O_EXCL)) {
3459 spin_lock(&inode->i_lock);
3460 inode->i_state |= I_LINKABLE;
3461 spin_unlock(&inode->i_lock);
3462 }
3463 ima_post_create_tmpfile(inode);
3464 return child;
3465
3466out_err:
3467 dput(child);
3468 return ERR_PTR(error);
3469}
3470EXPORT_SYMBOL(vfs_tmpfile);
3471
3472static int do_tmpfile(struct nameidata *nd, unsigned flags,
3473 const struct open_flags *op,
3474 struct file *file)
3475{
3476 struct dentry *child;
3477 struct path path;
3478 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3479 if (unlikely(error))
3480 return error;
3481 error = mnt_want_write(path.mnt);
3482 if (unlikely(error))
3483 goto out;
3484 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3485 error = PTR_ERR(child);
3486 if (IS_ERR(child))
3487 goto out2;
3488 dput(path.dentry);
3489 path.dentry = child;
3490 audit_inode(nd->name, child, 0);
3491 /* Don't check for other permissions, the inode was just created */
3492 error = may_open(&path, 0, op->open_flag);
3493 if (error)
3494 goto out2;
3495 file->f_path.mnt = path.mnt;
3496 error = finish_open(file, child, NULL);
3497out2:
3498 mnt_drop_write(path.mnt);
3499out:
3500 path_put(&path);
3501 return error;
3502}
3503
3504static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3505{
3506 struct path path;
3507 int error = path_lookupat(nd, flags, &path);
3508 if (!error) {
3509 audit_inode(nd->name, path.dentry, 0);
3510 error = vfs_open(&path, file);
3511 path_put(&path);
3512 }
3513 return error;
3514}
3515
3516static struct file *path_openat(struct nameidata *nd,
3517 const struct open_flags *op, unsigned flags)
3518{
3519 struct file *file;
3520 int error;
3521
3522 file = alloc_empty_file(op->open_flag, current_cred());
3523 if (IS_ERR(file))
3524 return file;
3525
3526 if (unlikely(file->f_flags & __O_TMPFILE)) {
3527 error = do_tmpfile(nd, flags, op, file);
3528 } else if (unlikely(file->f_flags & O_PATH)) {
3529 error = do_o_path(nd, flags, file);
3530 } else {
3531 const char *s = path_init(nd, flags);
3532 while (!(error = link_path_walk(s, nd)) &&
3533 (error = do_last(nd, file, op)) > 0) {
3534 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3535 s = trailing_symlink(nd);
3536 }
3537 terminate_walk(nd);
3538 }
3539 if (likely(!error)) {
3540 if (likely(file->f_mode & FMODE_OPENED))
3541 return file;
3542 WARN_ON(1);
3543 error = -EINVAL;
3544 }
3545 fput(file);
3546 if (error == -EOPENSTALE) {
3547 if (flags & LOOKUP_RCU)
3548 error = -ECHILD;
3549 else
3550 error = -ESTALE;
3551 }
3552 return ERR_PTR(error);
3553}
3554
3555struct file *do_filp_open(int dfd, struct filename *pathname,
3556 const struct open_flags *op)
3557{
3558 struct nameidata nd;
3559 int flags = op->lookup_flags;
3560 struct file *filp;
3561
3562 set_nameidata(&nd, dfd, pathname);
3563 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3564 if (unlikely(filp == ERR_PTR(-ECHILD)))
3565 filp = path_openat(&nd, op, flags);
3566 if (unlikely(filp == ERR_PTR(-ESTALE)))
3567 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3568 restore_nameidata();
3569 return filp;
3570}
3571
3572struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3573 const char *name, const struct open_flags *op)
3574{
3575 struct nameidata nd;
3576 struct file *file;
3577 struct filename *filename;
3578 int flags = op->lookup_flags | LOOKUP_ROOT;
3579
3580 nd.root.mnt = mnt;
3581 nd.root.dentry = dentry;
3582
3583 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3584 return ERR_PTR(-ELOOP);
3585
3586 filename = getname_kernel(name);
3587 if (IS_ERR(filename))
3588 return ERR_CAST(filename);
3589
3590 set_nameidata(&nd, -1, filename);
3591 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3592 if (unlikely(file == ERR_PTR(-ECHILD)))
3593 file = path_openat(&nd, op, flags);
3594 if (unlikely(file == ERR_PTR(-ESTALE)))
3595 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3596 restore_nameidata();
3597 putname(filename);
3598 return file;
3599}
3600
3601static struct dentry *filename_create(int dfd, struct filename *name,
3602 struct path *path, unsigned int lookup_flags)
3603{
3604 struct dentry *dentry = ERR_PTR(-EEXIST);
3605 struct qstr last;
3606 int type;
3607 int err2;
3608 int error;
3609 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3610
3611 /*
3612 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3613 * other flags passed in are ignored!
3614 */
3615 lookup_flags &= LOOKUP_REVAL;
3616
3617 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3618 if (IS_ERR(name))
3619 return ERR_CAST(name);
3620
3621 /*
3622 * Yucky last component or no last component at all?
3623 * (foo/., foo/.., /////)
3624 */
3625 if (unlikely(type != LAST_NORM))
3626 goto out;
3627
3628 /* don't fail immediately if it's r/o, at least try to report other errors */
3629 err2 = mnt_want_write(path->mnt);
3630 /*
3631 * Do the final lookup.
3632 */
3633 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3634 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3635 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3636 if (IS_ERR(dentry))
3637 goto unlock;
3638
3639 error = -EEXIST;
3640 if (d_is_positive(dentry))
3641 goto fail;
3642
3643 /*
3644 * Special case - lookup gave negative, but... we had foo/bar/
3645 * From the vfs_mknod() POV we just have a negative dentry -
3646 * all is fine. Let's be bastards - you had / on the end, you've
3647 * been asking for (non-existent) directory. -ENOENT for you.
3648 */
3649 if (unlikely(!is_dir && last.name[last.len])) {
3650 error = -ENOENT;
3651 goto fail;
3652 }
3653 if (unlikely(err2)) {
3654 error = err2;
3655 goto fail;
3656 }
3657 putname(name);
3658 return dentry;
3659fail:
3660 dput(dentry);
3661 dentry = ERR_PTR(error);
3662unlock:
3663 inode_unlock(path->dentry->d_inode);
3664 if (!err2)
3665 mnt_drop_write(path->mnt);
3666out:
3667 path_put(path);
3668 putname(name);
3669 return dentry;
3670}
3671
3672struct dentry *kern_path_create(int dfd, const char *pathname,
3673 struct path *path, unsigned int lookup_flags)
3674{
3675 return filename_create(dfd, getname_kernel(pathname),
3676 path, lookup_flags);
3677}
3678EXPORT_SYMBOL(kern_path_create);
3679
3680void done_path_create(struct path *path, struct dentry *dentry)
3681{
3682 dput(dentry);
3683 inode_unlock(path->dentry->d_inode);
3684 mnt_drop_write(path->mnt);
3685 path_put(path);
3686}
3687EXPORT_SYMBOL(done_path_create);
3688
3689inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3690 struct path *path, unsigned int lookup_flags)
3691{
3692 return filename_create(dfd, getname(pathname), path, lookup_flags);
3693}
3694EXPORT_SYMBOL(user_path_create);
3695
3696int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3697{
3698 int error = may_create(dir, dentry);
3699
3700 if (error)
3701 return error;
3702
3703 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3704 return -EPERM;
3705
3706 if (!dir->i_op->mknod)
3707 return -EPERM;
3708
3709 error = devcgroup_inode_mknod(mode, dev);
3710 if (error)
3711 return error;
3712
3713 error = security_inode_mknod(dir, dentry, mode, dev);
3714 if (error)
3715 return error;
3716
3717 error = dir->i_op->mknod(dir, dentry, mode, dev);
3718 if (!error)
3719 fsnotify_create(dir, dentry);
3720 return error;
3721}
3722EXPORT_SYMBOL(vfs_mknod);
3723
3724static int may_mknod(umode_t mode)
3725{
3726 switch (mode & S_IFMT) {
3727 case S_IFREG:
3728 case S_IFCHR:
3729 case S_IFBLK:
3730 case S_IFIFO:
3731 case S_IFSOCK:
3732 case 0: /* zero mode translates to S_IFREG */
3733 return 0;
3734 case S_IFDIR:
3735 return -EPERM;
3736 default:
3737 return -EINVAL;
3738 }
3739}
3740
3741long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3742 unsigned int dev)
3743{
3744 struct dentry *dentry;
3745 struct path path;
3746 int error;
3747 unsigned int lookup_flags = 0;
3748
3749 error = may_mknod(mode);
3750 if (error)
3751 return error;
3752retry:
3753 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3754 if (IS_ERR(dentry))
3755 return PTR_ERR(dentry);
3756
3757 if (!IS_POSIXACL(path.dentry->d_inode))
3758 mode &= ~current_umask();
3759 error = security_path_mknod(&path, dentry, mode, dev);
3760 if (error)
3761 goto out;
3762 switch (mode & S_IFMT) {
3763 case 0: case S_IFREG:
3764 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3765 if (!error)
3766 ima_post_path_mknod(dentry);
3767 break;
3768 case S_IFCHR: case S_IFBLK:
3769 error = vfs_mknod(path.dentry->