]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blob - fs/namei.c
Merge tag '4.19-rc2-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6
[mirror_ubuntu-eoan-kernel.git] / fs / namei.c
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 #include <linux/build_bug.h>
43
44 #include "internal.h"
45 #include "mount.h"
46
47 /* [Feb-1997 T. Schoebel-Theuer]
48 * Fundamental changes in the pathname lookup mechanisms (namei)
49 * were necessary because of omirr. The reason is that omirr needs
50 * to know the _real_ pathname, not the user-supplied one, in case
51 * of symlinks (and also when transname replacements occur).
52 *
53 * The new code replaces the old recursive symlink resolution with
54 * an iterative one (in case of non-nested symlink chains). It does
55 * this with calls to <fs>_follow_link().
56 * As a side effect, dir_namei(), _namei() and follow_link() are now
57 * replaced with a single function lookup_dentry() that can handle all
58 * the special cases of the former code.
59 *
60 * With the new dcache, the pathname is stored at each inode, at least as
61 * long as the refcount of the inode is positive. As a side effect, the
62 * size of the dcache depends on the inode cache and thus is dynamic.
63 *
64 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
65 * resolution to correspond with current state of the code.
66 *
67 * Note that the symlink resolution is not *completely* iterative.
68 * There is still a significant amount of tail- and mid- recursion in
69 * the algorithm. Also, note that <fs>_readlink() is not used in
70 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
71 * may return different results than <fs>_follow_link(). Many virtual
72 * filesystems (including /proc) exhibit this behavior.
73 */
74
75 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
76 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
77 * and the name already exists in form of a symlink, try to create the new
78 * name indicated by the symlink. The old code always complained that the
79 * name already exists, due to not following the symlink even if its target
80 * is nonexistent. The new semantics affects also mknod() and link() when
81 * the name is a symlink pointing to a non-existent name.
82 *
83 * I don't know which semantics is the right one, since I have no access
84 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
85 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
86 * "old" one. Personally, I think the new semantics is much more logical.
87 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
88 * file does succeed in both HP-UX and SunOs, but not in Solaris
89 * and in the old Linux semantics.
90 */
91
92 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
93 * semantics. See the comments in "open_namei" and "do_link" below.
94 *
95 * [10-Sep-98 Alan Modra] Another symlink change.
96 */
97
98 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
99 * inside the path - always follow.
100 * in the last component in creation/removal/renaming - never follow.
101 * if LOOKUP_FOLLOW passed - follow.
102 * if the pathname has trailing slashes - follow.
103 * otherwise - don't follow.
104 * (applied in that order).
105 *
106 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
107 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
108 * During the 2.4 we need to fix the userland stuff depending on it -
109 * hopefully we will be able to get rid of that wart in 2.5. So far only
110 * XEmacs seems to be relying on it...
111 */
112 /*
113 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
114 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
115 * any extra contention...
116 */
117
118 /* In order to reduce some races, while at the same time doing additional
119 * checking and hopefully speeding things up, we copy filenames to the
120 * kernel data space before using them..
121 *
122 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
123 * PATH_MAX includes the nul terminator --RR.
124 */
125
126 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
127
128 struct filename *
129 getname_flags(const char __user *filename, int flags, int *empty)
130 {
131 struct filename *result;
132 char *kname;
133 int len;
134 BUILD_BUG_ON(offsetof(struct filename, iname) % sizeof(long) != 0);
135
136 result = audit_reusename(filename);
137 if (result)
138 return result;
139
140 result = __getname();
141 if (unlikely(!result))
142 return ERR_PTR(-ENOMEM);
143
144 /*
145 * First, try to embed the struct filename inside the names_cache
146 * allocation
147 */
148 kname = (char *)result->iname;
149 result->name = kname;
150
151 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
152 if (unlikely(len < 0)) {
153 __putname(result);
154 return ERR_PTR(len);
155 }
156
157 /*
158 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
159 * separate struct filename so we can dedicate the entire
160 * names_cache allocation for the pathname, and re-do the copy from
161 * userland.
162 */
163 if (unlikely(len == EMBEDDED_NAME_MAX)) {
164 const size_t size = offsetof(struct filename, iname[1]);
165 kname = (char *)result;
166
167 /*
168 * size is chosen that way we to guarantee that
169 * result->iname[0] is within the same object and that
170 * kname can't be equal to result->iname, no matter what.
171 */
172 result = kzalloc(size, GFP_KERNEL);
173 if (unlikely(!result)) {
174 __putname(kname);
175 return ERR_PTR(-ENOMEM);
176 }
177 result->name = kname;
178 len = strncpy_from_user(kname, filename, PATH_MAX);
179 if (unlikely(len < 0)) {
180 __putname(kname);
181 kfree(result);
182 return ERR_PTR(len);
183 }
184 if (unlikely(len == PATH_MAX)) {
185 __putname(kname);
186 kfree(result);
187 return ERR_PTR(-ENAMETOOLONG);
188 }
189 }
190
191 result->refcnt = 1;
192 /* The empty path is special. */
193 if (unlikely(!len)) {
194 if (empty)
195 *empty = 1;
196 if (!(flags & LOOKUP_EMPTY)) {
197 putname(result);
198 return ERR_PTR(-ENOENT);
199 }
200 }
201
202 result->uptr = filename;
203 result->aname = NULL;
204 audit_getname(result);
205 return result;
206 }
207
208 struct filename *
209 getname(const char __user * filename)
210 {
211 return getname_flags(filename, 0, NULL);
212 }
213
214 struct filename *
215 getname_kernel(const char * filename)
216 {
217 struct filename *result;
218 int len = strlen(filename) + 1;
219
220 result = __getname();
221 if (unlikely(!result))
222 return ERR_PTR(-ENOMEM);
223
224 if (len <= EMBEDDED_NAME_MAX) {
225 result->name = (char *)result->iname;
226 } else if (len <= PATH_MAX) {
227 const size_t size = offsetof(struct filename, iname[1]);
228 struct filename *tmp;
229
230 tmp = kmalloc(size, GFP_KERNEL);
231 if (unlikely(!tmp)) {
232 __putname(result);
233 return ERR_PTR(-ENOMEM);
234 }
235 tmp->name = (char *)result;
236 result = tmp;
237 } else {
238 __putname(result);
239 return ERR_PTR(-ENAMETOOLONG);
240 }
241 memcpy((char *)result->name, filename, len);
242 result->uptr = NULL;
243 result->aname = NULL;
244 result->refcnt = 1;
245 audit_getname(result);
246
247 return result;
248 }
249
250 void putname(struct filename *name)
251 {
252 BUG_ON(name->refcnt <= 0);
253
254 if (--name->refcnt > 0)
255 return;
256
257 if (name->name != name->iname) {
258 __putname(name->name);
259 kfree(name);
260 } else
261 __putname(name);
262 }
263
264 static int check_acl(struct inode *inode, int mask)
265 {
266 #ifdef CONFIG_FS_POSIX_ACL
267 struct posix_acl *acl;
268
269 if (mask & MAY_NOT_BLOCK) {
270 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
271 if (!acl)
272 return -EAGAIN;
273 /* no ->get_acl() calls in RCU mode... */
274 if (is_uncached_acl(acl))
275 return -ECHILD;
276 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
277 }
278
279 acl = get_acl(inode, ACL_TYPE_ACCESS);
280 if (IS_ERR(acl))
281 return PTR_ERR(acl);
282 if (acl) {
283 int error = posix_acl_permission(inode, acl, mask);
284 posix_acl_release(acl);
285 return error;
286 }
287 #endif
288
289 return -EAGAIN;
290 }
291
292 /*
293 * This does the basic permission checking
294 */
295 static int acl_permission_check(struct inode *inode, int mask)
296 {
297 unsigned int mode = inode->i_mode;
298
299 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
300 mode >>= 6;
301 else {
302 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
303 int error = check_acl(inode, mask);
304 if (error != -EAGAIN)
305 return error;
306 }
307
308 if (in_group_p(inode->i_gid))
309 mode >>= 3;
310 }
311
312 /*
313 * If the DACs are ok we don't need any capability check.
314 */
315 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
316 return 0;
317 return -EACCES;
318 }
319
320 /**
321 * generic_permission - check for access rights on a Posix-like filesystem
322 * @inode: inode to check access rights for
323 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
324 *
325 * Used to check for read/write/execute permissions on a file.
326 * We use "fsuid" for this, letting us set arbitrary permissions
327 * for filesystem access without changing the "normal" uids which
328 * are used for other things.
329 *
330 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
331 * request cannot be satisfied (eg. requires blocking or too much complexity).
332 * It would then be called again in ref-walk mode.
333 */
334 int generic_permission(struct inode *inode, int mask)
335 {
336 int ret;
337
338 /*
339 * Do the basic permission checks.
340 */
341 ret = acl_permission_check(inode, mask);
342 if (ret != -EACCES)
343 return ret;
344
345 if (S_ISDIR(inode->i_mode)) {
346 /* DACs are overridable for directories */
347 if (!(mask & MAY_WRITE))
348 if (capable_wrt_inode_uidgid(inode,
349 CAP_DAC_READ_SEARCH))
350 return 0;
351 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
352 return 0;
353 return -EACCES;
354 }
355
356 /*
357 * Searching includes executable on directories, else just read.
358 */
359 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
360 if (mask == MAY_READ)
361 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
362 return 0;
363 /*
364 * Read/write DACs are always overridable.
365 * Executable DACs are overridable when there is
366 * at least one exec bit set.
367 */
368 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
369 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
370 return 0;
371
372 return -EACCES;
373 }
374 EXPORT_SYMBOL(generic_permission);
375
376 /*
377 * We _really_ want to just do "generic_permission()" without
378 * even looking at the inode->i_op values. So we keep a cache
379 * flag in inode->i_opflags, that says "this has not special
380 * permission function, use the fast case".
381 */
382 static inline int do_inode_permission(struct inode *inode, int mask)
383 {
384 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
385 if (likely(inode->i_op->permission))
386 return inode->i_op->permission(inode, mask);
387
388 /* This gets set once for the inode lifetime */
389 spin_lock(&inode->i_lock);
390 inode->i_opflags |= IOP_FASTPERM;
391 spin_unlock(&inode->i_lock);
392 }
393 return generic_permission(inode, mask);
394 }
395
396 /**
397 * sb_permission - Check superblock-level permissions
398 * @sb: Superblock of inode to check permission on
399 * @inode: Inode to check permission on
400 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
401 *
402 * Separate out file-system wide checks from inode-specific permission checks.
403 */
404 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
405 {
406 if (unlikely(mask & MAY_WRITE)) {
407 umode_t mode = inode->i_mode;
408
409 /* Nobody gets write access to a read-only fs. */
410 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
411 return -EROFS;
412 }
413 return 0;
414 }
415
416 /**
417 * inode_permission - Check for access rights to a given inode
418 * @inode: Inode to check permission on
419 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
420 *
421 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
422 * this, letting us set arbitrary permissions for filesystem access without
423 * changing the "normal" UIDs which are used for other things.
424 *
425 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
426 */
427 int inode_permission(struct inode *inode, int mask)
428 {
429 int retval;
430
431 retval = sb_permission(inode->i_sb, inode, mask);
432 if (retval)
433 return retval;
434
435 if (unlikely(mask & MAY_WRITE)) {
436 /*
437 * Nobody gets write access to an immutable file.
438 */
439 if (IS_IMMUTABLE(inode))
440 return -EPERM;
441
442 /*
443 * Updating mtime will likely cause i_uid and i_gid to be
444 * written back improperly if their true value is unknown
445 * to the vfs.
446 */
447 if (HAS_UNMAPPED_ID(inode))
448 return -EACCES;
449 }
450
451 retval = do_inode_permission(inode, mask);
452 if (retval)
453 return retval;
454
455 retval = devcgroup_inode_permission(inode, mask);
456 if (retval)
457 return retval;
458
459 return security_inode_permission(inode, mask);
460 }
461 EXPORT_SYMBOL(inode_permission);
462
463 /**
464 * path_get - get a reference to a path
465 * @path: path to get the reference to
466 *
467 * Given a path increment the reference count to the dentry and the vfsmount.
468 */
469 void path_get(const struct path *path)
470 {
471 mntget(path->mnt);
472 dget(path->dentry);
473 }
474 EXPORT_SYMBOL(path_get);
475
476 /**
477 * path_put - put a reference to a path
478 * @path: path to put the reference to
479 *
480 * Given a path decrement the reference count to the dentry and the vfsmount.
481 */
482 void path_put(const struct path *path)
483 {
484 dput(path->dentry);
485 mntput(path->mnt);
486 }
487 EXPORT_SYMBOL(path_put);
488
489 #define EMBEDDED_LEVELS 2
490 struct nameidata {
491 struct path path;
492 struct qstr last;
493 struct path root;
494 struct inode *inode; /* path.dentry.d_inode */
495 unsigned int flags;
496 unsigned seq, m_seq;
497 int last_type;
498 unsigned depth;
499 int total_link_count;
500 struct saved {
501 struct path link;
502 struct delayed_call done;
503 const char *name;
504 unsigned seq;
505 } *stack, internal[EMBEDDED_LEVELS];
506 struct filename *name;
507 struct nameidata *saved;
508 struct inode *link_inode;
509 unsigned root_seq;
510 int dfd;
511 } __randomize_layout;
512
513 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
514 {
515 struct nameidata *old = current->nameidata;
516 p->stack = p->internal;
517 p->dfd = dfd;
518 p->name = name;
519 p->total_link_count = old ? old->total_link_count : 0;
520 p->saved = old;
521 current->nameidata = p;
522 }
523
524 static void restore_nameidata(void)
525 {
526 struct nameidata *now = current->nameidata, *old = now->saved;
527
528 current->nameidata = old;
529 if (old)
530 old->total_link_count = now->total_link_count;
531 if (now->stack != now->internal)
532 kfree(now->stack);
533 }
534
535 static int __nd_alloc_stack(struct nameidata *nd)
536 {
537 struct saved *p;
538
539 if (nd->flags & LOOKUP_RCU) {
540 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
541 GFP_ATOMIC);
542 if (unlikely(!p))
543 return -ECHILD;
544 } else {
545 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
546 GFP_KERNEL);
547 if (unlikely(!p))
548 return -ENOMEM;
549 }
550 memcpy(p, nd->internal, sizeof(nd->internal));
551 nd->stack = p;
552 return 0;
553 }
554
555 /**
556 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
557 * @path: nameidate to verify
558 *
559 * Rename can sometimes move a file or directory outside of a bind
560 * mount, path_connected allows those cases to be detected.
561 */
562 static bool path_connected(const struct path *path)
563 {
564 struct vfsmount *mnt = path->mnt;
565 struct super_block *sb = mnt->mnt_sb;
566
567 /* Bind mounts and multi-root filesystems can have disconnected paths */
568 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
569 return true;
570
571 return is_subdir(path->dentry, mnt->mnt_root);
572 }
573
574 static inline int nd_alloc_stack(struct nameidata *nd)
575 {
576 if (likely(nd->depth != EMBEDDED_LEVELS))
577 return 0;
578 if (likely(nd->stack != nd->internal))
579 return 0;
580 return __nd_alloc_stack(nd);
581 }
582
583 static void drop_links(struct nameidata *nd)
584 {
585 int i = nd->depth;
586 while (i--) {
587 struct saved *last = nd->stack + i;
588 do_delayed_call(&last->done);
589 clear_delayed_call(&last->done);
590 }
591 }
592
593 static void terminate_walk(struct nameidata *nd)
594 {
595 drop_links(nd);
596 if (!(nd->flags & LOOKUP_RCU)) {
597 int i;
598 path_put(&nd->path);
599 for (i = 0; i < nd->depth; i++)
600 path_put(&nd->stack[i].link);
601 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
602 path_put(&nd->root);
603 nd->root.mnt = NULL;
604 }
605 } else {
606 nd->flags &= ~LOOKUP_RCU;
607 if (!(nd->flags & LOOKUP_ROOT))
608 nd->root.mnt = NULL;
609 rcu_read_unlock();
610 }
611 nd->depth = 0;
612 }
613
614 /* path_put is needed afterwards regardless of success or failure */
615 static bool legitimize_path(struct nameidata *nd,
616 struct path *path, unsigned seq)
617 {
618 int res = __legitimize_mnt(path->mnt, nd->m_seq);
619 if (unlikely(res)) {
620 if (res > 0)
621 path->mnt = NULL;
622 path->dentry = NULL;
623 return false;
624 }
625 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
626 path->dentry = NULL;
627 return false;
628 }
629 return !read_seqcount_retry(&path->dentry->d_seq, seq);
630 }
631
632 static bool legitimize_links(struct nameidata *nd)
633 {
634 int i;
635 for (i = 0; i < nd->depth; i++) {
636 struct saved *last = nd->stack + i;
637 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
638 drop_links(nd);
639 nd->depth = i + 1;
640 return false;
641 }
642 }
643 return true;
644 }
645
646 /*
647 * Path walking has 2 modes, rcu-walk and ref-walk (see
648 * Documentation/filesystems/path-lookup.txt). In situations when we can't
649 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
650 * normal reference counts on dentries and vfsmounts to transition to ref-walk
651 * mode. Refcounts are grabbed at the last known good point before rcu-walk
652 * got stuck, so ref-walk may continue from there. If this is not successful
653 * (eg. a seqcount has changed), then failure is returned and it's up to caller
654 * to restart the path walk from the beginning in ref-walk mode.
655 */
656
657 /**
658 * unlazy_walk - try to switch to ref-walk mode.
659 * @nd: nameidata pathwalk data
660 * Returns: 0 on success, -ECHILD on failure
661 *
662 * unlazy_walk attempts to legitimize the current nd->path and nd->root
663 * for ref-walk mode.
664 * Must be called from rcu-walk context.
665 * Nothing should touch nameidata between unlazy_walk() failure and
666 * terminate_walk().
667 */
668 static int unlazy_walk(struct nameidata *nd)
669 {
670 struct dentry *parent = nd->path.dentry;
671
672 BUG_ON(!(nd->flags & LOOKUP_RCU));
673
674 nd->flags &= ~LOOKUP_RCU;
675 if (unlikely(!legitimize_links(nd)))
676 goto out2;
677 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
678 goto out1;
679 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
680 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq)))
681 goto out;
682 }
683 rcu_read_unlock();
684 BUG_ON(nd->inode != parent->d_inode);
685 return 0;
686
687 out2:
688 nd->path.mnt = NULL;
689 nd->path.dentry = NULL;
690 out1:
691 if (!(nd->flags & LOOKUP_ROOT))
692 nd->root.mnt = NULL;
693 out:
694 rcu_read_unlock();
695 return -ECHILD;
696 }
697
698 /**
699 * unlazy_child - try to switch to ref-walk mode.
700 * @nd: nameidata pathwalk data
701 * @dentry: child of nd->path.dentry
702 * @seq: seq number to check dentry against
703 * Returns: 0 on success, -ECHILD on failure
704 *
705 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
706 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
707 * @nd. Must be called from rcu-walk context.
708 * Nothing should touch nameidata between unlazy_child() failure and
709 * terminate_walk().
710 */
711 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
712 {
713 BUG_ON(!(nd->flags & LOOKUP_RCU));
714
715 nd->flags &= ~LOOKUP_RCU;
716 if (unlikely(!legitimize_links(nd)))
717 goto out2;
718 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
719 goto out2;
720 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
721 goto out1;
722
723 /*
724 * We need to move both the parent and the dentry from the RCU domain
725 * to be properly refcounted. And the sequence number in the dentry
726 * validates *both* dentry counters, since we checked the sequence
727 * number of the parent after we got the child sequence number. So we
728 * know the parent must still be valid if the child sequence number is
729 */
730 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
731 goto out;
732 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) {
733 rcu_read_unlock();
734 dput(dentry);
735 goto drop_root_mnt;
736 }
737 /*
738 * Sequence counts matched. Now make sure that the root is
739 * still valid and get it if required.
740 */
741 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
742 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
743 rcu_read_unlock();
744 dput(dentry);
745 return -ECHILD;
746 }
747 }
748
749 rcu_read_unlock();
750 return 0;
751
752 out2:
753 nd->path.mnt = NULL;
754 out1:
755 nd->path.dentry = NULL;
756 out:
757 rcu_read_unlock();
758 drop_root_mnt:
759 if (!(nd->flags & LOOKUP_ROOT))
760 nd->root.mnt = NULL;
761 return -ECHILD;
762 }
763
764 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
765 {
766 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
767 return dentry->d_op->d_revalidate(dentry, flags);
768 else
769 return 1;
770 }
771
772 /**
773 * complete_walk - successful completion of path walk
774 * @nd: pointer nameidata
775 *
776 * If we had been in RCU mode, drop out of it and legitimize nd->path.
777 * Revalidate the final result, unless we'd already done that during
778 * the path walk or the filesystem doesn't ask for it. Return 0 on
779 * success, -error on failure. In case of failure caller does not
780 * need to drop nd->path.
781 */
782 static int complete_walk(struct nameidata *nd)
783 {
784 struct dentry *dentry = nd->path.dentry;
785 int status;
786
787 if (nd->flags & LOOKUP_RCU) {
788 if (!(nd->flags & LOOKUP_ROOT))
789 nd->root.mnt = NULL;
790 if (unlikely(unlazy_walk(nd)))
791 return -ECHILD;
792 }
793
794 if (likely(!(nd->flags & LOOKUP_JUMPED)))
795 return 0;
796
797 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
798 return 0;
799
800 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
801 if (status > 0)
802 return 0;
803
804 if (!status)
805 status = -ESTALE;
806
807 return status;
808 }
809
810 static void set_root(struct nameidata *nd)
811 {
812 struct fs_struct *fs = current->fs;
813
814 if (nd->flags & LOOKUP_RCU) {
815 unsigned seq;
816
817 do {
818 seq = read_seqcount_begin(&fs->seq);
819 nd->root = fs->root;
820 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
821 } while (read_seqcount_retry(&fs->seq, seq));
822 } else {
823 get_fs_root(fs, &nd->root);
824 }
825 }
826
827 static void path_put_conditional(struct path *path, struct nameidata *nd)
828 {
829 dput(path->dentry);
830 if (path->mnt != nd->path.mnt)
831 mntput(path->mnt);
832 }
833
834 static inline void path_to_nameidata(const struct path *path,
835 struct nameidata *nd)
836 {
837 if (!(nd->flags & LOOKUP_RCU)) {
838 dput(nd->path.dentry);
839 if (nd->path.mnt != path->mnt)
840 mntput(nd->path.mnt);
841 }
842 nd->path.mnt = path->mnt;
843 nd->path.dentry = path->dentry;
844 }
845
846 static int nd_jump_root(struct nameidata *nd)
847 {
848 if (nd->flags & LOOKUP_RCU) {
849 struct dentry *d;
850 nd->path = nd->root;
851 d = nd->path.dentry;
852 nd->inode = d->d_inode;
853 nd->seq = nd->root_seq;
854 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
855 return -ECHILD;
856 } else {
857 path_put(&nd->path);
858 nd->path = nd->root;
859 path_get(&nd->path);
860 nd->inode = nd->path.dentry->d_inode;
861 }
862 nd->flags |= LOOKUP_JUMPED;
863 return 0;
864 }
865
866 /*
867 * Helper to directly jump to a known parsed path from ->get_link,
868 * caller must have taken a reference to path beforehand.
869 */
870 void nd_jump_link(struct path *path)
871 {
872 struct nameidata *nd = current->nameidata;
873 path_put(&nd->path);
874
875 nd->path = *path;
876 nd->inode = nd->path.dentry->d_inode;
877 nd->flags |= LOOKUP_JUMPED;
878 }
879
880 static inline void put_link(struct nameidata *nd)
881 {
882 struct saved *last = nd->stack + --nd->depth;
883 do_delayed_call(&last->done);
884 if (!(nd->flags & LOOKUP_RCU))
885 path_put(&last->link);
886 }
887
888 int sysctl_protected_symlinks __read_mostly = 0;
889 int sysctl_protected_hardlinks __read_mostly = 0;
890 int sysctl_protected_fifos __read_mostly;
891 int sysctl_protected_regular __read_mostly;
892
893 /**
894 * may_follow_link - Check symlink following for unsafe situations
895 * @nd: nameidata pathwalk data
896 *
897 * In the case of the sysctl_protected_symlinks sysctl being enabled,
898 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
899 * in a sticky world-writable directory. This is to protect privileged
900 * processes from failing races against path names that may change out
901 * from under them by way of other users creating malicious symlinks.
902 * It will permit symlinks to be followed only when outside a sticky
903 * world-writable directory, or when the uid of the symlink and follower
904 * match, or when the directory owner matches the symlink's owner.
905 *
906 * Returns 0 if following the symlink is allowed, -ve on error.
907 */
908 static inline int may_follow_link(struct nameidata *nd)
909 {
910 const struct inode *inode;
911 const struct inode *parent;
912 kuid_t puid;
913
914 if (!sysctl_protected_symlinks)
915 return 0;
916
917 /* Allowed if owner and follower match. */
918 inode = nd->link_inode;
919 if (uid_eq(current_cred()->fsuid, inode->i_uid))
920 return 0;
921
922 /* Allowed if parent directory not sticky and world-writable. */
923 parent = nd->inode;
924 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
925 return 0;
926
927 /* Allowed if parent directory and link owner match. */
928 puid = parent->i_uid;
929 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
930 return 0;
931
932 if (nd->flags & LOOKUP_RCU)
933 return -ECHILD;
934
935 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
936 audit_log_link_denied("follow_link");
937 return -EACCES;
938 }
939
940 /**
941 * safe_hardlink_source - Check for safe hardlink conditions
942 * @inode: the source inode to hardlink from
943 *
944 * Return false if at least one of the following conditions:
945 * - inode is not a regular file
946 * - inode is setuid
947 * - inode is setgid and group-exec
948 * - access failure for read and write
949 *
950 * Otherwise returns true.
951 */
952 static bool safe_hardlink_source(struct inode *inode)
953 {
954 umode_t mode = inode->i_mode;
955
956 /* Special files should not get pinned to the filesystem. */
957 if (!S_ISREG(mode))
958 return false;
959
960 /* Setuid files should not get pinned to the filesystem. */
961 if (mode & S_ISUID)
962 return false;
963
964 /* Executable setgid files should not get pinned to the filesystem. */
965 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
966 return false;
967
968 /* Hardlinking to unreadable or unwritable sources is dangerous. */
969 if (inode_permission(inode, MAY_READ | MAY_WRITE))
970 return false;
971
972 return true;
973 }
974
975 /**
976 * may_linkat - Check permissions for creating a hardlink
977 * @link: the source to hardlink from
978 *
979 * Block hardlink when all of:
980 * - sysctl_protected_hardlinks enabled
981 * - fsuid does not match inode
982 * - hardlink source is unsafe (see safe_hardlink_source() above)
983 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
984 *
985 * Returns 0 if successful, -ve on error.
986 */
987 static int may_linkat(struct path *link)
988 {
989 struct inode *inode = link->dentry->d_inode;
990
991 /* Inode writeback is not safe when the uid or gid are invalid. */
992 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
993 return -EOVERFLOW;
994
995 if (!sysctl_protected_hardlinks)
996 return 0;
997
998 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
999 * otherwise, it must be a safe source.
1000 */
1001 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1002 return 0;
1003
1004 audit_log_link_denied("linkat");
1005 return -EPERM;
1006 }
1007
1008 /**
1009 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1010 * should be allowed, or not, on files that already
1011 * exist.
1012 * @dir: the sticky parent directory
1013 * @inode: the inode of the file to open
1014 *
1015 * Block an O_CREAT open of a FIFO (or a regular file) when:
1016 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1017 * - the file already exists
1018 * - we are in a sticky directory
1019 * - we don't own the file
1020 * - the owner of the directory doesn't own the file
1021 * - the directory is world writable
1022 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1023 * the directory doesn't have to be world writable: being group writable will
1024 * be enough.
1025 *
1026 * Returns 0 if the open is allowed, -ve on error.
1027 */
1028 static int may_create_in_sticky(struct dentry * const dir,
1029 struct inode * const inode)
1030 {
1031 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1032 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1033 likely(!(dir->d_inode->i_mode & S_ISVTX)) ||
1034 uid_eq(inode->i_uid, dir->d_inode->i_uid) ||
1035 uid_eq(current_fsuid(), inode->i_uid))
1036 return 0;
1037
1038 if (likely(dir->d_inode->i_mode & 0002) ||
1039 (dir->d_inode->i_mode & 0020 &&
1040 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1041 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1042 return -EACCES;
1043 }
1044 return 0;
1045 }
1046
1047 static __always_inline
1048 const char *get_link(struct nameidata *nd)
1049 {
1050 struct saved *last = nd->stack + nd->depth - 1;
1051 struct dentry *dentry = last->link.dentry;
1052 struct inode *inode = nd->link_inode;
1053 int error;
1054 const char *res;
1055
1056 if (!(nd->flags & LOOKUP_RCU)) {
1057 touch_atime(&last->link);
1058 cond_resched();
1059 } else if (atime_needs_update(&last->link, inode)) {
1060 if (unlikely(unlazy_walk(nd)))
1061 return ERR_PTR(-ECHILD);
1062 touch_atime(&last->link);
1063 }
1064
1065 error = security_inode_follow_link(dentry, inode,
1066 nd->flags & LOOKUP_RCU);
1067 if (unlikely(error))
1068 return ERR_PTR(error);
1069
1070 nd->last_type = LAST_BIND;
1071 res = inode->i_link;
1072 if (!res) {
1073 const char * (*get)(struct dentry *, struct inode *,
1074 struct delayed_call *);
1075 get = inode->i_op->get_link;
1076 if (nd->flags & LOOKUP_RCU) {
1077 res = get(NULL, inode, &last->done);
1078 if (res == ERR_PTR(-ECHILD)) {
1079 if (unlikely(unlazy_walk(nd)))
1080 return ERR_PTR(-ECHILD);
1081 res = get(dentry, inode, &last->done);
1082 }
1083 } else {
1084 res = get(dentry, inode, &last->done);
1085 }
1086 if (IS_ERR_OR_NULL(res))
1087 return res;
1088 }
1089 if (*res == '/') {
1090 if (!nd->root.mnt)
1091 set_root(nd);
1092 if (unlikely(nd_jump_root(nd)))
1093 return ERR_PTR(-ECHILD);
1094 while (unlikely(*++res == '/'))
1095 ;
1096 }
1097 if (!*res)
1098 res = NULL;
1099 return res;
1100 }
1101
1102 /*
1103 * follow_up - Find the mountpoint of path's vfsmount
1104 *
1105 * Given a path, find the mountpoint of its source file system.
1106 * Replace @path with the path of the mountpoint in the parent mount.
1107 * Up is towards /.
1108 *
1109 * Return 1 if we went up a level and 0 if we were already at the
1110 * root.
1111 */
1112 int follow_up(struct path *path)
1113 {
1114 struct mount *mnt = real_mount(path->mnt);
1115 struct mount *parent;
1116 struct dentry *mountpoint;
1117
1118 read_seqlock_excl(&mount_lock);
1119 parent = mnt->mnt_parent;
1120 if (parent == mnt) {
1121 read_sequnlock_excl(&mount_lock);
1122 return 0;
1123 }
1124 mntget(&parent->mnt);
1125 mountpoint = dget(mnt->mnt_mountpoint);
1126 read_sequnlock_excl(&mount_lock);
1127 dput(path->dentry);
1128 path->dentry = mountpoint;
1129 mntput(path->mnt);
1130 path->mnt = &parent->mnt;
1131 return 1;
1132 }
1133 EXPORT_SYMBOL(follow_up);
1134
1135 /*
1136 * Perform an automount
1137 * - return -EISDIR to tell follow_managed() to stop and return the path we
1138 * were called with.
1139 */
1140 static int follow_automount(struct path *path, struct nameidata *nd,
1141 bool *need_mntput)
1142 {
1143 struct vfsmount *mnt;
1144 int err;
1145
1146 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1147 return -EREMOTE;
1148
1149 /* We don't want to mount if someone's just doing a stat -
1150 * unless they're stat'ing a directory and appended a '/' to
1151 * the name.
1152 *
1153 * We do, however, want to mount if someone wants to open or
1154 * create a file of any type under the mountpoint, wants to
1155 * traverse through the mountpoint or wants to open the
1156 * mounted directory. Also, autofs may mark negative dentries
1157 * as being automount points. These will need the attentions
1158 * of the daemon to instantiate them before they can be used.
1159 */
1160 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1161 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1162 path->dentry->d_inode)
1163 return -EISDIR;
1164
1165 nd->total_link_count++;
1166 if (nd->total_link_count >= 40)
1167 return -ELOOP;
1168
1169 mnt = path->dentry->d_op->d_automount(path);
1170 if (IS_ERR(mnt)) {
1171 /*
1172 * The filesystem is allowed to return -EISDIR here to indicate
1173 * it doesn't want to automount. For instance, autofs would do
1174 * this so that its userspace daemon can mount on this dentry.
1175 *
1176 * However, we can only permit this if it's a terminal point in
1177 * the path being looked up; if it wasn't then the remainder of
1178 * the path is inaccessible and we should say so.
1179 */
1180 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1181 return -EREMOTE;
1182 return PTR_ERR(mnt);
1183 }
1184
1185 if (!mnt) /* mount collision */
1186 return 0;
1187
1188 if (!*need_mntput) {
1189 /* lock_mount() may release path->mnt on error */
1190 mntget(path->mnt);
1191 *need_mntput = true;
1192 }
1193 err = finish_automount(mnt, path);
1194
1195 switch (err) {
1196 case -EBUSY:
1197 /* Someone else made a mount here whilst we were busy */
1198 return 0;
1199 case 0:
1200 path_put(path);
1201 path->mnt = mnt;
1202 path->dentry = dget(mnt->mnt_root);
1203 return 0;
1204 default:
1205 return err;
1206 }
1207
1208 }
1209
1210 /*
1211 * Handle a dentry that is managed in some way.
1212 * - Flagged for transit management (autofs)
1213 * - Flagged as mountpoint
1214 * - Flagged as automount point
1215 *
1216 * This may only be called in refwalk mode.
1217 *
1218 * Serialization is taken care of in namespace.c
1219 */
1220 static int follow_managed(struct path *path, struct nameidata *nd)
1221 {
1222 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1223 unsigned managed;
1224 bool need_mntput = false;
1225 int ret = 0;
1226
1227 /* Given that we're not holding a lock here, we retain the value in a
1228 * local variable for each dentry as we look at it so that we don't see
1229 * the components of that value change under us */
1230 while (managed = READ_ONCE(path->dentry->d_flags),
1231 managed &= DCACHE_MANAGED_DENTRY,
1232 unlikely(managed != 0)) {
1233 /* Allow the filesystem to manage the transit without i_mutex
1234 * being held. */
1235 if (managed & DCACHE_MANAGE_TRANSIT) {
1236 BUG_ON(!path->dentry->d_op);
1237 BUG_ON(!path->dentry->d_op->d_manage);
1238 ret = path->dentry->d_op->d_manage(path, false);
1239 if (ret < 0)
1240 break;
1241 }
1242
1243 /* Transit to a mounted filesystem. */
1244 if (managed & DCACHE_MOUNTED) {
1245 struct vfsmount *mounted = lookup_mnt(path);
1246 if (mounted) {
1247 dput(path->dentry);
1248 if (need_mntput)
1249 mntput(path->mnt);
1250 path->mnt = mounted;
1251 path->dentry = dget(mounted->mnt_root);
1252 need_mntput = true;
1253 continue;
1254 }
1255
1256 /* Something is mounted on this dentry in another
1257 * namespace and/or whatever was mounted there in this
1258 * namespace got unmounted before lookup_mnt() could
1259 * get it */
1260 }
1261
1262 /* Handle an automount point */
1263 if (managed & DCACHE_NEED_AUTOMOUNT) {
1264 ret = follow_automount(path, nd, &need_mntput);
1265 if (ret < 0)
1266 break;
1267 continue;
1268 }
1269
1270 /* We didn't change the current path point */
1271 break;
1272 }
1273
1274 if (need_mntput && path->mnt == mnt)
1275 mntput(path->mnt);
1276 if (ret == -EISDIR || !ret)
1277 ret = 1;
1278 if (need_mntput)
1279 nd->flags |= LOOKUP_JUMPED;
1280 if (unlikely(ret < 0))
1281 path_put_conditional(path, nd);
1282 return ret;
1283 }
1284
1285 int follow_down_one(struct path *path)
1286 {
1287 struct vfsmount *mounted;
1288
1289 mounted = lookup_mnt(path);
1290 if (mounted) {
1291 dput(path->dentry);
1292 mntput(path->mnt);
1293 path->mnt = mounted;
1294 path->dentry = dget(mounted->mnt_root);
1295 return 1;
1296 }
1297 return 0;
1298 }
1299 EXPORT_SYMBOL(follow_down_one);
1300
1301 static inline int managed_dentry_rcu(const struct path *path)
1302 {
1303 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1304 path->dentry->d_op->d_manage(path, true) : 0;
1305 }
1306
1307 /*
1308 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1309 * we meet a managed dentry that would need blocking.
1310 */
1311 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1312 struct inode **inode, unsigned *seqp)
1313 {
1314 for (;;) {
1315 struct mount *mounted;
1316 /*
1317 * Don't forget we might have a non-mountpoint managed dentry
1318 * that wants to block transit.
1319 */
1320 switch (managed_dentry_rcu(path)) {
1321 case -ECHILD:
1322 default:
1323 return false;
1324 case -EISDIR:
1325 return true;
1326 case 0:
1327 break;
1328 }
1329
1330 if (!d_mountpoint(path->dentry))
1331 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1332
1333 mounted = __lookup_mnt(path->mnt, path->dentry);
1334 if (!mounted)
1335 break;
1336 path->mnt = &mounted->mnt;
1337 path->dentry = mounted->mnt.mnt_root;
1338 nd->flags |= LOOKUP_JUMPED;
1339 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1340 /*
1341 * Update the inode too. We don't need to re-check the
1342 * dentry sequence number here after this d_inode read,
1343 * because a mount-point is always pinned.
1344 */
1345 *inode = path->dentry->d_inode;
1346 }
1347 return !read_seqretry(&mount_lock, nd->m_seq) &&
1348 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1349 }
1350
1351 static int follow_dotdot_rcu(struct nameidata *nd)
1352 {
1353 struct inode *inode = nd->inode;
1354
1355 while (1) {
1356 if (path_equal(&nd->path, &nd->root))
1357 break;
1358 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1359 struct dentry *old = nd->path.dentry;
1360 struct dentry *parent = old->d_parent;
1361 unsigned seq;
1362
1363 inode = parent->d_inode;
1364 seq = read_seqcount_begin(&parent->d_seq);
1365 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1366 return -ECHILD;
1367 nd->path.dentry = parent;
1368 nd->seq = seq;
1369 if (unlikely(!path_connected(&nd->path)))
1370 return -ENOENT;
1371 break;
1372 } else {
1373 struct mount *mnt = real_mount(nd->path.mnt);
1374 struct mount *mparent = mnt->mnt_parent;
1375 struct dentry *mountpoint = mnt->mnt_mountpoint;
1376 struct inode *inode2 = mountpoint->d_inode;
1377 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1378 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1379 return -ECHILD;
1380 if (&mparent->mnt == nd->path.mnt)
1381 break;
1382 /* we know that mountpoint was pinned */
1383 nd->path.dentry = mountpoint;
1384 nd->path.mnt = &mparent->mnt;
1385 inode = inode2;
1386 nd->seq = seq;
1387 }
1388 }
1389 while (unlikely(d_mountpoint(nd->path.dentry))) {
1390 struct mount *mounted;
1391 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1392 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1393 return -ECHILD;
1394 if (!mounted)
1395 break;
1396 nd->path.mnt = &mounted->mnt;
1397 nd->path.dentry = mounted->mnt.mnt_root;
1398 inode = nd->path.dentry->d_inode;
1399 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1400 }
1401 nd->inode = inode;
1402 return 0;
1403 }
1404
1405 /*
1406 * Follow down to the covering mount currently visible to userspace. At each
1407 * point, the filesystem owning that dentry may be queried as to whether the
1408 * caller is permitted to proceed or not.
1409 */
1410 int follow_down(struct path *path)
1411 {
1412 unsigned managed;
1413 int ret;
1414
1415 while (managed = READ_ONCE(path->dentry->d_flags),
1416 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1417 /* Allow the filesystem to manage the transit without i_mutex
1418 * being held.
1419 *
1420 * We indicate to the filesystem if someone is trying to mount
1421 * something here. This gives autofs the chance to deny anyone
1422 * other than its daemon the right to mount on its
1423 * superstructure.
1424 *
1425 * The filesystem may sleep at this point.
1426 */
1427 if (managed & DCACHE_MANAGE_TRANSIT) {
1428 BUG_ON(!path->dentry->d_op);
1429 BUG_ON(!path->dentry->d_op->d_manage);
1430 ret = path->dentry->d_op->d_manage(path, false);
1431 if (ret < 0)
1432 return ret == -EISDIR ? 0 : ret;
1433 }
1434
1435 /* Transit to a mounted filesystem. */
1436 if (managed & DCACHE_MOUNTED) {
1437 struct vfsmount *mounted = lookup_mnt(path);
1438 if (!mounted)
1439 break;
1440 dput(path->dentry);
1441 mntput(path->mnt);
1442 path->mnt = mounted;
1443 path->dentry = dget(mounted->mnt_root);
1444 continue;
1445 }
1446
1447 /* Don't handle automount points here */
1448 break;
1449 }
1450 return 0;
1451 }
1452 EXPORT_SYMBOL(follow_down);
1453
1454 /*
1455 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1456 */
1457 static void follow_mount(struct path *path)
1458 {
1459 while (d_mountpoint(path->dentry)) {
1460 struct vfsmount *mounted = lookup_mnt(path);
1461 if (!mounted)
1462 break;
1463 dput(path->dentry);
1464 mntput(path->mnt);
1465 path->mnt = mounted;
1466 path->dentry = dget(mounted->mnt_root);
1467 }
1468 }
1469
1470 static int path_parent_directory(struct path *path)
1471 {
1472 struct dentry *old = path->dentry;
1473 /* rare case of legitimate dget_parent()... */
1474 path->dentry = dget_parent(path->dentry);
1475 dput(old);
1476 if (unlikely(!path_connected(path)))
1477 return -ENOENT;
1478 return 0;
1479 }
1480
1481 static int follow_dotdot(struct nameidata *nd)
1482 {
1483 while(1) {
1484 if (path_equal(&nd->path, &nd->root))
1485 break;
1486 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1487 int ret = path_parent_directory(&nd->path);
1488 if (ret)
1489 return ret;
1490 break;
1491 }
1492 if (!follow_up(&nd->path))
1493 break;
1494 }
1495 follow_mount(&nd->path);
1496 nd->inode = nd->path.dentry->d_inode;
1497 return 0;
1498 }
1499
1500 /*
1501 * This looks up the name in dcache and possibly revalidates the found dentry.
1502 * NULL is returned if the dentry does not exist in the cache.
1503 */
1504 static struct dentry *lookup_dcache(const struct qstr *name,
1505 struct dentry *dir,
1506 unsigned int flags)
1507 {
1508 struct dentry *dentry = d_lookup(dir, name);
1509 if (dentry) {
1510 int error = d_revalidate(dentry, flags);
1511 if (unlikely(error <= 0)) {
1512 if (!error)
1513 d_invalidate(dentry);
1514 dput(dentry);
1515 return ERR_PTR(error);
1516 }
1517 }
1518 return dentry;
1519 }
1520
1521 /*
1522 * Parent directory has inode locked exclusive. This is one
1523 * and only case when ->lookup() gets called on non in-lookup
1524 * dentries - as the matter of fact, this only gets called
1525 * when directory is guaranteed to have no in-lookup children
1526 * at all.
1527 */
1528 static struct dentry *__lookup_hash(const struct qstr *name,
1529 struct dentry *base, unsigned int flags)
1530 {
1531 struct dentry *dentry = lookup_dcache(name, base, flags);
1532 struct dentry *old;
1533 struct inode *dir = base->d_inode;
1534
1535 if (dentry)
1536 return dentry;
1537
1538 /* Don't create child dentry for a dead directory. */
1539 if (unlikely(IS_DEADDIR(dir)))
1540 return ERR_PTR(-ENOENT);
1541
1542 dentry = d_alloc(base, name);
1543 if (unlikely(!dentry))
1544 return ERR_PTR(-ENOMEM);
1545
1546 old = dir->i_op->lookup(dir, dentry, flags);
1547 if (unlikely(old)) {
1548 dput(dentry);
1549 dentry = old;
1550 }
1551 return dentry;
1552 }
1553
1554 static int lookup_fast(struct nameidata *nd,
1555 struct path *path, struct inode **inode,
1556 unsigned *seqp)
1557 {
1558 struct vfsmount *mnt = nd->path.mnt;
1559 struct dentry *dentry, *parent = nd->path.dentry;
1560 int status = 1;
1561 int err;
1562
1563 /*
1564 * Rename seqlock is not required here because in the off chance
1565 * of a false negative due to a concurrent rename, the caller is
1566 * going to fall back to non-racy lookup.
1567 */
1568 if (nd->flags & LOOKUP_RCU) {
1569 unsigned seq;
1570 bool negative;
1571 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1572 if (unlikely(!dentry)) {
1573 if (unlazy_walk(nd))
1574 return -ECHILD;
1575 return 0;
1576 }
1577
1578 /*
1579 * This sequence count validates that the inode matches
1580 * the dentry name information from lookup.
1581 */
1582 *inode = d_backing_inode(dentry);
1583 negative = d_is_negative(dentry);
1584 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1585 return -ECHILD;
1586
1587 /*
1588 * This sequence count validates that the parent had no
1589 * changes while we did the lookup of the dentry above.
1590 *
1591 * The memory barrier in read_seqcount_begin of child is
1592 * enough, we can use __read_seqcount_retry here.
1593 */
1594 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1595 return -ECHILD;
1596
1597 *seqp = seq;
1598 status = d_revalidate(dentry, nd->flags);
1599 if (likely(status > 0)) {
1600 /*
1601 * Note: do negative dentry check after revalidation in
1602 * case that drops it.
1603 */
1604 if (unlikely(negative))
1605 return -ENOENT;
1606 path->mnt = mnt;
1607 path->dentry = dentry;
1608 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1609 return 1;
1610 }
1611 if (unlazy_child(nd, dentry, seq))
1612 return -ECHILD;
1613 if (unlikely(status == -ECHILD))
1614 /* we'd been told to redo it in non-rcu mode */
1615 status = d_revalidate(dentry, nd->flags);
1616 } else {
1617 dentry = __d_lookup(parent, &nd->last);
1618 if (unlikely(!dentry))
1619 return 0;
1620 status = d_revalidate(dentry, nd->flags);
1621 }
1622 if (unlikely(status <= 0)) {
1623 if (!status)
1624 d_invalidate(dentry);
1625 dput(dentry);
1626 return status;
1627 }
1628 if (unlikely(d_is_negative(dentry))) {
1629 dput(dentry);
1630 return -ENOENT;
1631 }
1632
1633 path->mnt = mnt;
1634 path->dentry = dentry;
1635 err = follow_managed(path, nd);
1636 if (likely(err > 0))
1637 *inode = d_backing_inode(path->dentry);
1638 return err;
1639 }
1640
1641 /* Fast lookup failed, do it the slow way */
1642 static struct dentry *__lookup_slow(const struct qstr *name,
1643 struct dentry *dir,
1644 unsigned int flags)
1645 {
1646 struct dentry *dentry, *old;
1647 struct inode *inode = dir->d_inode;
1648 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1649
1650 /* Don't go there if it's already dead */
1651 if (unlikely(IS_DEADDIR(inode)))
1652 return ERR_PTR(-ENOENT);
1653 again:
1654 dentry = d_alloc_parallel(dir, name, &wq);
1655 if (IS_ERR(dentry))
1656 return dentry;
1657 if (unlikely(!d_in_lookup(dentry))) {
1658 if (!(flags & LOOKUP_NO_REVAL)) {
1659 int error = d_revalidate(dentry, flags);
1660 if (unlikely(error <= 0)) {
1661 if (!error) {
1662 d_invalidate(dentry);
1663 dput(dentry);
1664 goto again;
1665 }
1666 dput(dentry);
1667 dentry = ERR_PTR(error);
1668 }
1669 }
1670 } else {
1671 old = inode->i_op->lookup(inode, dentry, flags);
1672 d_lookup_done(dentry);
1673 if (unlikely(old)) {
1674 dput(dentry);
1675 dentry = old;
1676 }
1677 }
1678 return dentry;
1679 }
1680
1681 static struct dentry *lookup_slow(const struct qstr *name,
1682 struct dentry *dir,
1683 unsigned int flags)
1684 {
1685 struct inode *inode = dir->d_inode;
1686 struct dentry *res;
1687 inode_lock_shared(inode);
1688 res = __lookup_slow(name, dir, flags);
1689 inode_unlock_shared(inode);
1690 return res;
1691 }
1692
1693 static inline int may_lookup(struct nameidata *nd)
1694 {
1695 if (nd->flags & LOOKUP_RCU) {
1696 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1697 if (err != -ECHILD)
1698 return err;
1699 if (unlazy_walk(nd))
1700 return -ECHILD;
1701 }
1702 return inode_permission(nd->inode, MAY_EXEC);
1703 }
1704
1705 static inline int handle_dots(struct nameidata *nd, int type)
1706 {
1707 if (type == LAST_DOTDOT) {
1708 if (!nd->root.mnt)
1709 set_root(nd);
1710 if (nd->flags & LOOKUP_RCU) {
1711 return follow_dotdot_rcu(nd);
1712 } else
1713 return follow_dotdot(nd);
1714 }
1715 return 0;
1716 }
1717
1718 static int pick_link(struct nameidata *nd, struct path *link,
1719 struct inode *inode, unsigned seq)
1720 {
1721 int error;
1722 struct saved *last;
1723 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1724 path_to_nameidata(link, nd);
1725 return -ELOOP;
1726 }
1727 if (!(nd->flags & LOOKUP_RCU)) {
1728 if (link->mnt == nd->path.mnt)
1729 mntget(link->mnt);
1730 }
1731 error = nd_alloc_stack(nd);
1732 if (unlikely(error)) {
1733 if (error == -ECHILD) {
1734 if (unlikely(!legitimize_path(nd, link, seq))) {
1735 drop_links(nd);
1736 nd->depth = 0;
1737 nd->flags &= ~LOOKUP_RCU;
1738 nd->path.mnt = NULL;
1739 nd->path.dentry = NULL;
1740 if (!(nd->flags & LOOKUP_ROOT))
1741 nd->root.mnt = NULL;
1742 rcu_read_unlock();
1743 } else if (likely(unlazy_walk(nd)) == 0)
1744 error = nd_alloc_stack(nd);
1745 }
1746 if (error) {
1747 path_put(link);
1748 return error;
1749 }
1750 }
1751
1752 last = nd->stack + nd->depth++;
1753 last->link = *link;
1754 clear_delayed_call(&last->done);
1755 nd->link_inode = inode;
1756 last->seq = seq;
1757 return 1;
1758 }
1759
1760 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1761
1762 /*
1763 * Do we need to follow links? We _really_ want to be able
1764 * to do this check without having to look at inode->i_op,
1765 * so we keep a cache of "no, this doesn't need follow_link"
1766 * for the common case.
1767 */
1768 static inline int step_into(struct nameidata *nd, struct path *path,
1769 int flags, struct inode *inode, unsigned seq)
1770 {
1771 if (!(flags & WALK_MORE) && nd->depth)
1772 put_link(nd);
1773 if (likely(!d_is_symlink(path->dentry)) ||
1774 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1775 /* not a symlink or should not follow */
1776 path_to_nameidata(path, nd);
1777 nd->inode = inode;
1778 nd->seq = seq;
1779 return 0;
1780 }
1781 /* make sure that d_is_symlink above matches inode */
1782 if (nd->flags & LOOKUP_RCU) {
1783 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1784 return -ECHILD;
1785 }
1786 return pick_link(nd, path, inode, seq);
1787 }
1788
1789 static int walk_component(struct nameidata *nd, int flags)
1790 {
1791 struct path path;
1792 struct inode *inode;
1793 unsigned seq;
1794 int err;
1795 /*
1796 * "." and ".." are special - ".." especially so because it has
1797 * to be able to know about the current root directory and
1798 * parent relationships.
1799 */
1800 if (unlikely(nd->last_type != LAST_NORM)) {
1801 err = handle_dots(nd, nd->last_type);
1802 if (!(flags & WALK_MORE) && nd->depth)
1803 put_link(nd);
1804 return err;
1805 }
1806 err = lookup_fast(nd, &path, &inode, &seq);
1807 if (unlikely(err <= 0)) {
1808 if (err < 0)
1809 return err;
1810 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1811 nd->flags);
1812 if (IS_ERR(path.dentry))
1813 return PTR_ERR(path.dentry);
1814
1815 path.mnt = nd->path.mnt;
1816 err = follow_managed(&path, nd);
1817 if (unlikely(err < 0))
1818 return err;
1819
1820 if (unlikely(d_is_negative(path.dentry))) {
1821 path_to_nameidata(&path, nd);
1822 return -ENOENT;
1823 }
1824
1825 seq = 0; /* we are already out of RCU mode */
1826 inode = d_backing_inode(path.dentry);
1827 }
1828
1829 return step_into(nd, &path, flags, inode, seq);
1830 }
1831
1832 /*
1833 * We can do the critical dentry name comparison and hashing
1834 * operations one word at a time, but we are limited to:
1835 *
1836 * - Architectures with fast unaligned word accesses. We could
1837 * do a "get_unaligned()" if this helps and is sufficiently
1838 * fast.
1839 *
1840 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1841 * do not trap on the (extremely unlikely) case of a page
1842 * crossing operation.
1843 *
1844 * - Furthermore, we need an efficient 64-bit compile for the
1845 * 64-bit case in order to generate the "number of bytes in
1846 * the final mask". Again, that could be replaced with a
1847 * efficient population count instruction or similar.
1848 */
1849 #ifdef CONFIG_DCACHE_WORD_ACCESS
1850
1851 #include <asm/word-at-a-time.h>
1852
1853 #ifdef HASH_MIX
1854
1855 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1856
1857 #elif defined(CONFIG_64BIT)
1858 /*
1859 * Register pressure in the mixing function is an issue, particularly
1860 * on 32-bit x86, but almost any function requires one state value and
1861 * one temporary. Instead, use a function designed for two state values
1862 * and no temporaries.
1863 *
1864 * This function cannot create a collision in only two iterations, so
1865 * we have two iterations to achieve avalanche. In those two iterations,
1866 * we have six layers of mixing, which is enough to spread one bit's
1867 * influence out to 2^6 = 64 state bits.
1868 *
1869 * Rotate constants are scored by considering either 64 one-bit input
1870 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1871 * probability of that delta causing a change to each of the 128 output
1872 * bits, using a sample of random initial states.
1873 *
1874 * The Shannon entropy of the computed probabilities is then summed
1875 * to produce a score. Ideally, any input change has a 50% chance of
1876 * toggling any given output bit.
1877 *
1878 * Mixing scores (in bits) for (12,45):
1879 * Input delta: 1-bit 2-bit
1880 * 1 round: 713.3 42542.6
1881 * 2 rounds: 2753.7 140389.8
1882 * 3 rounds: 5954.1 233458.2
1883 * 4 rounds: 7862.6 256672.2
1884 * Perfect: 8192 258048
1885 * (64*128) (64*63/2 * 128)
1886 */
1887 #define HASH_MIX(x, y, a) \
1888 ( x ^= (a), \
1889 y ^= x, x = rol64(x,12),\
1890 x += y, y = rol64(y,45),\
1891 y *= 9 )
1892
1893 /*
1894 * Fold two longs into one 32-bit hash value. This must be fast, but
1895 * latency isn't quite as critical, as there is a fair bit of additional
1896 * work done before the hash value is used.
1897 */
1898 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1899 {
1900 y ^= x * GOLDEN_RATIO_64;
1901 y *= GOLDEN_RATIO_64;
1902 return y >> 32;
1903 }
1904
1905 #else /* 32-bit case */
1906
1907 /*
1908 * Mixing scores (in bits) for (7,20):
1909 * Input delta: 1-bit 2-bit
1910 * 1 round: 330.3 9201.6
1911 * 2 rounds: 1246.4 25475.4
1912 * 3 rounds: 1907.1 31295.1
1913 * 4 rounds: 2042.3 31718.6
1914 * Perfect: 2048 31744
1915 * (32*64) (32*31/2 * 64)
1916 */
1917 #define HASH_MIX(x, y, a) \
1918 ( x ^= (a), \
1919 y ^= x, x = rol32(x, 7),\
1920 x += y, y = rol32(y,20),\
1921 y *= 9 )
1922
1923 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1924 {
1925 /* Use arch-optimized multiply if one exists */
1926 return __hash_32(y ^ __hash_32(x));
1927 }
1928
1929 #endif
1930
1931 /*
1932 * Return the hash of a string of known length. This is carfully
1933 * designed to match hash_name(), which is the more critical function.
1934 * In particular, we must end by hashing a final word containing 0..7
1935 * payload bytes, to match the way that hash_name() iterates until it
1936 * finds the delimiter after the name.
1937 */
1938 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1939 {
1940 unsigned long a, x = 0, y = (unsigned long)salt;
1941
1942 for (;;) {
1943 if (!len)
1944 goto done;
1945 a = load_unaligned_zeropad(name);
1946 if (len < sizeof(unsigned long))
1947 break;
1948 HASH_MIX(x, y, a);
1949 name += sizeof(unsigned long);
1950 len -= sizeof(unsigned long);
1951 }
1952 x ^= a & bytemask_from_count(len);
1953 done:
1954 return fold_hash(x, y);
1955 }
1956 EXPORT_SYMBOL(full_name_hash);
1957
1958 /* Return the "hash_len" (hash and length) of a null-terminated string */
1959 u64 hashlen_string(const void *salt, const char *name)
1960 {
1961 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1962 unsigned long adata, mask, len;
1963 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1964
1965 len = 0;
1966 goto inside;
1967
1968 do {
1969 HASH_MIX(x, y, a);
1970 len += sizeof(unsigned long);
1971 inside:
1972 a = load_unaligned_zeropad(name+len);
1973 } while (!has_zero(a, &adata, &constants));
1974
1975 adata = prep_zero_mask(a, adata, &constants);
1976 mask = create_zero_mask(adata);
1977 x ^= a & zero_bytemask(mask);
1978
1979 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1980 }
1981 EXPORT_SYMBOL(hashlen_string);
1982
1983 /*
1984 * Calculate the length and hash of the path component, and
1985 * return the "hash_len" as the result.
1986 */
1987 static inline u64 hash_name(const void *salt, const char *name)
1988 {
1989 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1990 unsigned long adata, bdata, mask, len;
1991 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1992
1993 len = 0;
1994 goto inside;
1995
1996 do {
1997 HASH_MIX(x, y, a);
1998 len += sizeof(unsigned long);
1999 inside:
2000 a = load_unaligned_zeropad(name+len);
2001 b = a ^ REPEAT_BYTE('/');
2002 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2003
2004 adata = prep_zero_mask(a, adata, &constants);
2005 bdata = prep_zero_mask(b, bdata, &constants);
2006 mask = create_zero_mask(adata | bdata);
2007 x ^= a & zero_bytemask(mask);
2008
2009 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2010 }
2011
2012 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2013
2014 /* Return the hash of a string of known length */
2015 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2016 {
2017 unsigned long hash = init_name_hash(salt);
2018 while (len--)
2019 hash = partial_name_hash((unsigned char)*name++, hash);
2020 return end_name_hash(hash);
2021 }
2022 EXPORT_SYMBOL(full_name_hash);
2023
2024 /* Return the "hash_len" (hash and length) of a null-terminated string */
2025 u64 hashlen_string(const void *salt, const char *name)
2026 {
2027 unsigned long hash = init_name_hash(salt);
2028 unsigned long len = 0, c;
2029
2030 c = (unsigned char)*name;
2031 while (c) {
2032 len++;
2033 hash = partial_name_hash(c, hash);
2034 c = (unsigned char)name[len];
2035 }
2036 return hashlen_create(end_name_hash(hash), len);
2037 }
2038 EXPORT_SYMBOL(hashlen_string);
2039
2040 /*
2041 * We know there's a real path component here of at least
2042 * one character.
2043 */
2044 static inline u64 hash_name(const void *salt, const char *name)
2045 {
2046 unsigned long hash = init_name_hash(salt);
2047 unsigned long len = 0, c;
2048
2049 c = (unsigned char)*name;
2050 do {
2051 len++;
2052 hash = partial_name_hash(c, hash);
2053 c = (unsigned char)name[len];
2054 } while (c && c != '/');
2055 return hashlen_create(end_name_hash(hash), len);
2056 }
2057
2058 #endif
2059
2060 /*
2061 * Name resolution.
2062 * This is the basic name resolution function, turning a pathname into
2063 * the final dentry. We expect 'base' to be positive and a directory.
2064 *
2065 * Returns 0 and nd will have valid dentry and mnt on success.
2066 * Returns error and drops reference to input namei data on failure.
2067 */
2068 static int link_path_walk(const char *name, struct nameidata *nd)
2069 {
2070 int err;
2071
2072 if (IS_ERR(name))
2073 return PTR_ERR(name);
2074 while (*name=='/')
2075 name++;
2076 if (!*name)
2077 return 0;
2078
2079 /* At this point we know we have a real path component. */
2080 for(;;) {
2081 u64 hash_len;
2082 int type;
2083
2084 err = may_lookup(nd);
2085 if (err)
2086 return err;
2087
2088 hash_len = hash_name(nd->path.dentry, name);
2089
2090 type = LAST_NORM;
2091 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2092 case 2:
2093 if (name[1] == '.') {
2094 type = LAST_DOTDOT;
2095 nd->flags |= LOOKUP_JUMPED;
2096 }
2097 break;
2098 case 1:
2099 type = LAST_DOT;
2100 }
2101 if (likely(type == LAST_NORM)) {
2102 struct dentry *parent = nd->path.dentry;
2103 nd->flags &= ~LOOKUP_JUMPED;
2104 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2105 struct qstr this = { { .hash_len = hash_len }, .name = name };
2106 err = parent->d_op->d_hash(parent, &this);
2107 if (err < 0)
2108 return err;
2109 hash_len = this.hash_len;
2110 name = this.name;
2111 }
2112 }
2113
2114 nd->last.hash_len = hash_len;
2115 nd->last.name = name;
2116 nd->last_type = type;
2117
2118 name += hashlen_len(hash_len);
2119 if (!*name)
2120 goto OK;
2121 /*
2122 * If it wasn't NUL, we know it was '/'. Skip that
2123 * slash, and continue until no more slashes.
2124 */
2125 do {
2126 name++;
2127 } while (unlikely(*name == '/'));
2128 if (unlikely(!*name)) {
2129 OK:
2130 /* pathname body, done */
2131 if (!nd->depth)
2132 return 0;
2133 name = nd->stack[nd->depth - 1].name;
2134 /* trailing symlink, done */
2135 if (!name)
2136 return 0;
2137 /* last component of nested symlink */
2138 err = walk_component(nd, WALK_FOLLOW);
2139 } else {
2140 /* not the last component */
2141 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2142 }
2143 if (err < 0)
2144 return err;
2145
2146 if (err) {
2147 const char *s = get_link(nd);
2148
2149 if (IS_ERR(s))
2150 return PTR_ERR(s);
2151 err = 0;
2152 if (unlikely(!s)) {
2153 /* jumped */
2154 put_link(nd);
2155 } else {
2156 nd->stack[nd->depth - 1].name = name;
2157 name = s;
2158 continue;
2159 }
2160 }
2161 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2162 if (nd->flags & LOOKUP_RCU) {
2163 if (unlazy_walk(nd))
2164 return -ECHILD;
2165 }
2166 return -ENOTDIR;
2167 }
2168 }
2169 }
2170
2171 /* must be paired with terminate_walk() */
2172 static const char *path_init(struct nameidata *nd, unsigned flags)
2173 {
2174 const char *s = nd->name->name;
2175
2176 if (!*s)
2177 flags &= ~LOOKUP_RCU;
2178 if (flags & LOOKUP_RCU)
2179 rcu_read_lock();
2180
2181 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2182 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2183 nd->depth = 0;
2184 if (flags & LOOKUP_ROOT) {
2185 struct dentry *root = nd->root.dentry;
2186 struct inode *inode = root->d_inode;
2187 if (*s && unlikely(!d_can_lookup(root)))
2188 return ERR_PTR(-ENOTDIR);
2189 nd->path = nd->root;
2190 nd->inode = inode;
2191 if (flags & LOOKUP_RCU) {
2192 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2193 nd->root_seq = nd->seq;
2194 nd->m_seq = read_seqbegin(&mount_lock);
2195 } else {
2196 path_get(&nd->path);
2197 }
2198 return s;
2199 }
2200
2201 nd->root.mnt = NULL;
2202 nd->path.mnt = NULL;
2203 nd->path.dentry = NULL;
2204
2205 nd->m_seq = read_seqbegin(&mount_lock);
2206 if (*s == '/') {
2207 set_root(nd);
2208 if (likely(!nd_jump_root(nd)))
2209 return s;
2210 return ERR_PTR(-ECHILD);
2211 } else if (nd->dfd == AT_FDCWD) {
2212 if (flags & LOOKUP_RCU) {
2213 struct fs_struct *fs = current->fs;
2214 unsigned seq;
2215
2216 do {
2217 seq = read_seqcount_begin(&fs->seq);
2218 nd->path = fs->pwd;
2219 nd->inode = nd->path.dentry->d_inode;
2220 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2221 } while (read_seqcount_retry(&fs->seq, seq));
2222 } else {
2223 get_fs_pwd(current->fs, &nd->path);
2224 nd->inode = nd->path.dentry->d_inode;
2225 }
2226 return s;
2227 } else {
2228 /* Caller must check execute permissions on the starting path component */
2229 struct fd f = fdget_raw(nd->dfd);
2230 struct dentry *dentry;
2231
2232 if (!f.file)
2233 return ERR_PTR(-EBADF);
2234
2235 dentry = f.file->f_path.dentry;
2236
2237 if (*s && unlikely(!d_can_lookup(dentry))) {
2238 fdput(f);
2239 return ERR_PTR(-ENOTDIR);
2240 }
2241
2242 nd->path = f.file->f_path;
2243 if (flags & LOOKUP_RCU) {
2244 nd->inode = nd->path.dentry->d_inode;
2245 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2246 } else {
2247 path_get(&nd->path);
2248 nd->inode = nd->path.dentry->d_inode;
2249 }
2250 fdput(f);
2251 return s;
2252 }
2253 }
2254
2255 static const char *trailing_symlink(struct nameidata *nd)
2256 {
2257 const char *s;
2258 int error = may_follow_link(nd);
2259 if (unlikely(error))
2260 return ERR_PTR(error);
2261 nd->flags |= LOOKUP_PARENT;
2262 nd->stack[0].name = NULL;
2263 s = get_link(nd);
2264 return s ? s : "";
2265 }
2266
2267 static inline int lookup_last(struct nameidata *nd)
2268 {
2269 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2270 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2271
2272 nd->flags &= ~LOOKUP_PARENT;
2273 return walk_component(nd, 0);
2274 }
2275
2276 static int handle_lookup_down(struct nameidata *nd)
2277 {
2278 struct path path = nd->path;
2279 struct inode *inode = nd->inode;
2280 unsigned seq = nd->seq;
2281 int err;
2282
2283 if (nd->flags & LOOKUP_RCU) {
2284 /*
2285 * don't bother with unlazy_walk on failure - we are
2286 * at the very beginning of walk, so we lose nothing
2287 * if we simply redo everything in non-RCU mode
2288 */
2289 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2290 return -ECHILD;
2291 } else {
2292 dget(path.dentry);
2293 err = follow_managed(&path, nd);
2294 if (unlikely(err < 0))
2295 return err;
2296 inode = d_backing_inode(path.dentry);
2297 seq = 0;
2298 }
2299 path_to_nameidata(&path, nd);
2300 nd->inode = inode;
2301 nd->seq = seq;
2302 return 0;
2303 }
2304
2305 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2306 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2307 {
2308 const char *s = path_init(nd, flags);
2309 int err;
2310
2311 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2312 err = handle_lookup_down(nd);
2313 if (unlikely(err < 0))
2314 s = ERR_PTR(err);
2315 }
2316
2317 while (!(err = link_path_walk(s, nd))
2318 && ((err = lookup_last(nd)) > 0)) {
2319 s = trailing_symlink(nd);
2320 }
2321 if (!err)
2322 err = complete_walk(nd);
2323
2324 if (!err && nd->flags & LOOKUP_DIRECTORY)
2325 if (!d_can_lookup(nd->path.dentry))
2326 err = -ENOTDIR;
2327 if (!err) {
2328 *path = nd->path;
2329 nd->path.mnt = NULL;
2330 nd->path.dentry = NULL;
2331 }
2332 terminate_walk(nd);
2333 return err;
2334 }
2335
2336 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2337 struct path *path, struct path *root)
2338 {
2339 int retval;
2340 struct nameidata nd;
2341 if (IS_ERR(name))
2342 return PTR_ERR(name);
2343 if (unlikely(root)) {
2344 nd.root = *root;
2345 flags |= LOOKUP_ROOT;
2346 }
2347 set_nameidata(&nd, dfd, name);
2348 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2349 if (unlikely(retval == -ECHILD))
2350 retval = path_lookupat(&nd, flags, path);
2351 if (unlikely(retval == -ESTALE))
2352 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2353
2354 if (likely(!retval))
2355 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2356 restore_nameidata();
2357 putname(name);
2358 return retval;
2359 }
2360
2361 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2362 static int path_parentat(struct nameidata *nd, unsigned flags,
2363 struct path *parent)
2364 {
2365 const char *s = path_init(nd, flags);
2366 int err = link_path_walk(s, nd);
2367 if (!err)
2368 err = complete_walk(nd);
2369 if (!err) {
2370 *parent = nd->path;
2371 nd->path.mnt = NULL;
2372 nd->path.dentry = NULL;
2373 }
2374 terminate_walk(nd);
2375 return err;
2376 }
2377
2378 static struct filename *filename_parentat(int dfd, struct filename *name,
2379 unsigned int flags, struct path *parent,
2380 struct qstr *last, int *type)
2381 {
2382 int retval;
2383 struct nameidata nd;
2384
2385 if (IS_ERR(name))
2386 return name;
2387 set_nameidata(&nd, dfd, name);
2388 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2389 if (unlikely(retval == -ECHILD))
2390 retval = path_parentat(&nd, flags, parent);
2391 if (unlikely(retval == -ESTALE))
2392 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2393 if (likely(!retval)) {
2394 *last = nd.last;
2395 *type = nd.last_type;
2396 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2397 } else {
2398 putname(name);
2399 name = ERR_PTR(retval);
2400 }
2401 restore_nameidata();
2402 return name;
2403 }
2404
2405 /* does lookup, returns the object with parent locked */
2406 struct dentry *kern_path_locked(const char *name, struct path *path)
2407 {
2408 struct filename *filename;
2409 struct dentry *d;
2410 struct qstr last;
2411 int type;
2412
2413 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2414 &last, &type);
2415 if (IS_ERR(filename))
2416 return ERR_CAST(filename);
2417 if (unlikely(type != LAST_NORM)) {
2418 path_put(path);
2419 putname(filename);
2420 return ERR_PTR(-EINVAL);
2421 }
2422 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2423 d = __lookup_hash(&last, path->dentry, 0);
2424 if (IS_ERR(d)) {
2425 inode_unlock(path->dentry->d_inode);
2426 path_put(path);
2427 }
2428 putname(filename);
2429 return d;
2430 }
2431
2432 int kern_path(const char *name, unsigned int flags, struct path *path)
2433 {
2434 return filename_lookup(AT_FDCWD, getname_kernel(name),
2435 flags, path, NULL);
2436 }
2437 EXPORT_SYMBOL(kern_path);
2438
2439 /**
2440 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2441 * @dentry: pointer to dentry of the base directory
2442 * @mnt: pointer to vfs mount of the base directory
2443 * @name: pointer to file name
2444 * @flags: lookup flags
2445 * @path: pointer to struct path to fill
2446 */
2447 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2448 const char *name, unsigned int flags,
2449 struct path *path)
2450 {
2451 struct path root = {.mnt = mnt, .dentry = dentry};
2452 /* the first argument of filename_lookup() is ignored with root */
2453 return filename_lookup(AT_FDCWD, getname_kernel(name),
2454 flags , path, &root);
2455 }
2456 EXPORT_SYMBOL(vfs_path_lookup);
2457
2458 static int lookup_one_len_common(const char *name, struct dentry *base,
2459 int len, struct qstr *this)
2460 {
2461 this->name = name;
2462 this->len = len;
2463 this->hash = full_name_hash(base, name, len);
2464 if (!len)
2465 return -EACCES;
2466
2467 if (unlikely(name[0] == '.')) {
2468 if (len < 2 || (len == 2 && name[1] == '.'))
2469 return -EACCES;
2470 }
2471
2472 while (len--) {
2473 unsigned int c = *(const unsigned char *)name++;
2474 if (c == '/' || c == '\0')
2475 return -EACCES;
2476 }
2477 /*
2478 * See if the low-level filesystem might want
2479 * to use its own hash..
2480 */
2481 if (base->d_flags & DCACHE_OP_HASH) {
2482 int err = base->d_op->d_hash(base, this);
2483 if (err < 0)
2484 return err;
2485 }
2486
2487 return inode_permission(base->d_inode, MAY_EXEC);
2488 }
2489
2490 /**
2491 * try_lookup_one_len - filesystem helper to lookup single pathname component
2492 * @name: pathname component to lookup
2493 * @base: base directory to lookup from
2494 * @len: maximum length @len should be interpreted to
2495 *
2496 * Look up a dentry by name in the dcache, returning NULL if it does not
2497 * currently exist. The function does not try to create a dentry.
2498 *
2499 * Note that this routine is purely a helper for filesystem usage and should
2500 * not be called by generic code.
2501 *
2502 * The caller must hold base->i_mutex.
2503 */
2504 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2505 {
2506 struct qstr this;
2507 int err;
2508
2509 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2510
2511 err = lookup_one_len_common(name, base, len, &this);
2512 if (err)
2513 return ERR_PTR(err);
2514
2515 return lookup_dcache(&this, base, 0);
2516 }
2517 EXPORT_SYMBOL(try_lookup_one_len);
2518
2519 /**
2520 * lookup_one_len - filesystem helper to lookup single pathname component
2521 * @name: pathname component to lookup
2522 * @base: base directory to lookup from
2523 * @len: maximum length @len should be interpreted to
2524 *
2525 * Note that this routine is purely a helper for filesystem usage and should
2526 * not be called by generic code.
2527 *
2528 * The caller must hold base->i_mutex.
2529 */
2530 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2531 {
2532 struct dentry *dentry;
2533 struct qstr this;
2534 int err;
2535
2536 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2537
2538 err = lookup_one_len_common(name, base, len, &this);
2539 if (err)
2540 return ERR_PTR(err);
2541
2542 dentry = lookup_dcache(&this, base, 0);
2543 return dentry ? dentry : __lookup_slow(&this, base, 0);
2544 }
2545 EXPORT_SYMBOL(lookup_one_len);
2546
2547 /**
2548 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2549 * @name: pathname component to lookup
2550 * @base: base directory to lookup from
2551 * @len: maximum length @len should be interpreted to
2552 *
2553 * Note that this routine is purely a helper for filesystem usage and should
2554 * not be called by generic code.
2555 *
2556 * Unlike lookup_one_len, it should be called without the parent
2557 * i_mutex held, and will take the i_mutex itself if necessary.
2558 */
2559 struct dentry *lookup_one_len_unlocked(const char *name,
2560 struct dentry *base, int len)
2561 {
2562 struct qstr this;
2563 int err;
2564 struct dentry *ret;
2565
2566 err = lookup_one_len_common(name, base, len, &this);
2567 if (err)
2568 return ERR_PTR(err);
2569
2570 ret = lookup_dcache(&this, base, 0);
2571 if (!ret)
2572 ret = lookup_slow(&this, base, 0);
2573 return ret;
2574 }
2575 EXPORT_SYMBOL(lookup_one_len_unlocked);
2576
2577 #ifdef CONFIG_UNIX98_PTYS
2578 int path_pts(struct path *path)
2579 {
2580 /* Find something mounted on "pts" in the same directory as
2581 * the input path.
2582 */
2583 struct dentry *child, *parent;
2584 struct qstr this;
2585 int ret;
2586
2587 ret = path_parent_directory(path);
2588 if (ret)
2589 return ret;
2590
2591 parent = path->dentry;
2592 this.name = "pts";
2593 this.len = 3;
2594 child = d_hash_and_lookup(parent, &this);
2595 if (!child)
2596 return -ENOENT;
2597
2598 path->dentry = child;
2599 dput(parent);
2600 follow_mount(path);
2601 return 0;
2602 }
2603 #endif
2604
2605 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2606 struct path *path, int *empty)
2607 {
2608 return filename_lookup(dfd, getname_flags(name, flags, empty),
2609 flags, path, NULL);
2610 }
2611 EXPORT_SYMBOL(user_path_at_empty);
2612
2613 /**
2614 * mountpoint_last - look up last component for umount
2615 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2616 *
2617 * This is a special lookup_last function just for umount. In this case, we
2618 * need to resolve the path without doing any revalidation.
2619 *
2620 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2621 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2622 * in almost all cases, this lookup will be served out of the dcache. The only
2623 * cases where it won't are if nd->last refers to a symlink or the path is
2624 * bogus and it doesn't exist.
2625 *
2626 * Returns:
2627 * -error: if there was an error during lookup. This includes -ENOENT if the
2628 * lookup found a negative dentry.
2629 *
2630 * 0: if we successfully resolved nd->last and found it to not to be a
2631 * symlink that needs to be followed.
2632 *
2633 * 1: if we successfully resolved nd->last and found it to be a symlink
2634 * that needs to be followed.
2635 */
2636 static int
2637 mountpoint_last(struct nameidata *nd)
2638 {
2639 int error = 0;
2640 struct dentry *dir = nd->path.dentry;
2641 struct path path;
2642
2643 /* If we're in rcuwalk, drop out of it to handle last component */
2644 if (nd->flags & LOOKUP_RCU) {
2645 if (unlazy_walk(nd))
2646 return -ECHILD;
2647 }
2648
2649 nd->flags &= ~LOOKUP_PARENT;
2650
2651 if (unlikely(nd->last_type != LAST_NORM)) {
2652 error = handle_dots(nd, nd->last_type);
2653 if (error)
2654 return error;
2655 path.dentry = dget(nd->path.dentry);
2656 } else {
2657 path.dentry = d_lookup(dir, &nd->last);
2658 if (!path.dentry) {
2659 /*
2660 * No cached dentry. Mounted dentries are pinned in the
2661 * cache, so that means that this dentry is probably
2662 * a symlink or the path doesn't actually point
2663 * to a mounted dentry.
2664 */
2665 path.dentry = lookup_slow(&nd->last, dir,
2666 nd->flags | LOOKUP_NO_REVAL);
2667 if (IS_ERR(path.dentry))
2668 return PTR_ERR(path.dentry);
2669 }
2670 }
2671 if (d_is_negative(path.dentry)) {
2672 dput(path.dentry);
2673 return -ENOENT;
2674 }
2675 path.mnt = nd->path.mnt;
2676 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2677 }
2678
2679 /**
2680 * path_mountpoint - look up a path to be umounted
2681 * @nd: lookup context
2682 * @flags: lookup flags
2683 * @path: pointer to container for result
2684 *
2685 * Look up the given name, but don't attempt to revalidate the last component.
2686 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2687 */
2688 static int
2689 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2690 {
2691 const char *s = path_init(nd, flags);
2692 int err;
2693
2694 while (!(err = link_path_walk(s, nd)) &&
2695 (err = mountpoint_last(nd)) > 0) {
2696 s = trailing_symlink(nd);
2697 }
2698 if (!err) {
2699 *path = nd->path;
2700 nd->path.mnt = NULL;
2701 nd->path.dentry = NULL;
2702 follow_mount(path);
2703 }
2704 terminate_walk(nd);
2705 return err;
2706 }
2707
2708 static int
2709 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2710 unsigned int flags)
2711 {
2712 struct nameidata nd;
2713 int error;
2714 if (IS_ERR(name))
2715 return PTR_ERR(name);
2716 set_nameidata(&nd, dfd, name);
2717 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2718 if (unlikely(error == -ECHILD))
2719 error = path_mountpoint(&nd, flags, path);
2720 if (unlikely(error == -ESTALE))
2721 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2722 if (likely(!error))
2723 audit_inode(name, path->dentry, 0);
2724 restore_nameidata();
2725 putname(name);
2726 return error;
2727 }
2728
2729 /**
2730 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2731 * @dfd: directory file descriptor
2732 * @name: pathname from userland
2733 * @flags: lookup flags
2734 * @path: pointer to container to hold result
2735 *
2736 * A umount is a special case for path walking. We're not actually interested
2737 * in the inode in this situation, and ESTALE errors can be a problem. We
2738 * simply want track down the dentry and vfsmount attached at the mountpoint
2739 * and avoid revalidating the last component.
2740 *
2741 * Returns 0 and populates "path" on success.
2742 */
2743 int
2744 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2745 struct path *path)
2746 {
2747 return filename_mountpoint(dfd, getname(name), path, flags);
2748 }
2749
2750 int
2751 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2752 unsigned int flags)
2753 {
2754 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2755 }
2756 EXPORT_SYMBOL(kern_path_mountpoint);
2757
2758 int __check_sticky(struct inode *dir, struct inode *inode)
2759 {
2760 kuid_t fsuid = current_fsuid();
2761
2762 if (uid_eq(inode->i_uid, fsuid))
2763 return 0;
2764 if (uid_eq(dir->i_uid, fsuid))
2765 return 0;
2766 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2767 }
2768 EXPORT_SYMBOL(__check_sticky);
2769
2770 /*
2771 * Check whether we can remove a link victim from directory dir, check
2772 * whether the type of victim is right.
2773 * 1. We can't do it if dir is read-only (done in permission())
2774 * 2. We should have write and exec permissions on dir
2775 * 3. We can't remove anything from append-only dir
2776 * 4. We can't do anything with immutable dir (done in permission())
2777 * 5. If the sticky bit on dir is set we should either
2778 * a. be owner of dir, or
2779 * b. be owner of victim, or
2780 * c. have CAP_FOWNER capability
2781 * 6. If the victim is append-only or immutable we can't do antyhing with
2782 * links pointing to it.
2783 * 7. If the victim has an unknown uid or gid we can't change the inode.
2784 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2785 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2786 * 10. We can't remove a root or mountpoint.
2787 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2788 * nfs_async_unlink().
2789 */
2790 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2791 {
2792 struct inode *inode = d_backing_inode(victim);
2793 int error;
2794
2795 if (d_is_negative(victim))
2796 return -ENOENT;
2797 BUG_ON(!inode);
2798
2799 BUG_ON(victim->d_parent->d_inode != dir);
2800
2801 /* Inode writeback is not safe when the uid or gid are invalid. */
2802 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2803 return -EOVERFLOW;
2804
2805 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2806
2807 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2808 if (error)
2809 return error;
2810 if (IS_APPEND(dir))
2811 return -EPERM;
2812
2813 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2814 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2815 return -EPERM;
2816 if (isdir) {
2817 if (!d_is_dir(victim))
2818 return -ENOTDIR;
2819 if (IS_ROOT(victim))
2820 return -EBUSY;
2821 } else if (d_is_dir(victim))
2822 return -EISDIR;
2823 if (IS_DEADDIR(dir))
2824 return -ENOENT;
2825 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2826 return -EBUSY;
2827 return 0;
2828 }
2829
2830 /* Check whether we can create an object with dentry child in directory
2831 * dir.
2832 * 1. We can't do it if child already exists (open has special treatment for
2833 * this case, but since we are inlined it's OK)
2834 * 2. We can't do it if dir is read-only (done in permission())
2835 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2836 * 4. We should have write and exec permissions on dir
2837 * 5. We can't do it if dir is immutable (done in permission())
2838 */
2839 static inline int may_create(struct inode *dir, struct dentry *child)
2840 {
2841 struct user_namespace *s_user_ns;
2842 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2843 if (child->d_inode)
2844 return -EEXIST;
2845 if (IS_DEADDIR(dir))
2846 return -ENOENT;
2847 s_user_ns = dir->i_sb->s_user_ns;
2848 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2849 !kgid_has_mapping(s_user_ns, current_fsgid()))
2850 return -EOVERFLOW;
2851 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2852 }
2853
2854 /*
2855 * p1 and p2 should be directories on the same fs.
2856 */
2857 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2858 {
2859 struct dentry *p;
2860
2861 if (p1 == p2) {
2862 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2863 return NULL;
2864 }
2865
2866 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2867
2868 p = d_ancestor(p2, p1);
2869 if (p) {
2870 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2871 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2872 return p;
2873 }
2874
2875 p = d_ancestor(p1, p2);
2876 if (p) {
2877 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2878 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2879 return p;
2880 }
2881
2882 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2883 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2884 return NULL;
2885 }
2886 EXPORT_SYMBOL(lock_rename);
2887
2888 void unlock_rename(struct dentry *p1, struct dentry *p2)
2889 {
2890 inode_unlock(p1->d_inode);
2891 if (p1 != p2) {
2892 inode_unlock(p2->d_inode);
2893 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2894 }
2895 }
2896 EXPORT_SYMBOL(unlock_rename);
2897
2898 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2899 bool want_excl)
2900 {
2901 int error = may_create(dir, dentry);
2902 if (error)
2903 return error;
2904
2905 if (!dir->i_op->create)
2906 return -EACCES; /* shouldn't it be ENOSYS? */
2907 mode &= S_IALLUGO;
2908 mode |= S_IFREG;
2909 error = security_inode_create(dir, dentry, mode);
2910 if (error)
2911 return error;
2912 error = dir->i_op->create(dir, dentry, mode, want_excl);
2913 if (!error)
2914 fsnotify_create(dir, dentry);
2915 return error;
2916 }
2917 EXPORT_SYMBOL(vfs_create);
2918
2919 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2920 int (*f)(struct dentry *, umode_t, void *),
2921 void *arg)
2922 {
2923 struct inode *dir = dentry->d_parent->d_inode;
2924 int error = may_create(dir, dentry);
2925 if (error)
2926 return error;
2927
2928 mode &= S_IALLUGO;
2929 mode |= S_IFREG;
2930 error = security_inode_create(dir, dentry, mode);
2931 if (error)
2932 return error;
2933 error = f(dentry, mode, arg);
2934 if (!error)
2935 fsnotify_create(dir, dentry);
2936 return error;
2937 }
2938 EXPORT_SYMBOL(vfs_mkobj);
2939
2940 bool may_open_dev(const struct path *path)
2941 {
2942 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2943 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2944 }
2945
2946 static int may_open(const struct path *path, int acc_mode, int flag)
2947 {
2948 struct dentry *dentry = path->dentry;
2949 struct inode *inode = dentry->d_inode;
2950 int error;
2951
2952 if (!inode)
2953 return -ENOENT;
2954
2955 switch (inode->i_mode & S_IFMT) {
2956 case S_IFLNK:
2957 return -ELOOP;
2958 case S_IFDIR:
2959 if (acc_mode & MAY_WRITE)
2960 return -EISDIR;
2961 break;
2962 case S_IFBLK:
2963 case S_IFCHR:
2964 if (!may_open_dev(path))
2965 return -EACCES;
2966 /*FALLTHRU*/
2967 case S_IFIFO:
2968 case S_IFSOCK:
2969 flag &= ~O_TRUNC;
2970 break;
2971 }
2972
2973 error = inode_permission(inode, MAY_OPEN | acc_mode);
2974 if (error)
2975 return error;
2976
2977 /*
2978 * An append-only file must be opened in append mode for writing.
2979 */
2980 if (IS_APPEND(inode)) {
2981 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2982 return -EPERM;
2983 if (flag & O_TRUNC)
2984 return -EPERM;
2985 }
2986
2987 /* O_NOATIME can only be set by the owner or superuser */
2988 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2989 return -EPERM;
2990
2991 return 0;
2992 }
2993
2994 static int handle_truncate(struct file *filp)
2995 {
2996 const struct path *path = &filp->f_path;
2997 struct inode *inode = path->dentry->d_inode;
2998 int error = get_write_access(inode);
2999 if (error)
3000 return error;
3001 /*
3002 * Refuse to truncate files with mandatory locks held on them.
3003 */
3004 error = locks_verify_locked(filp);
3005 if (!error)
3006 error = security_path_truncate(path);
3007 if (!error) {
3008 error = do_truncate(path->dentry, 0,
3009 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3010 filp);
3011 }
3012 put_write_access(inode);
3013 return error;
3014 }
3015
3016 static inline int open_to_namei_flags(int flag)
3017 {
3018 if ((flag & O_ACCMODE) == 3)
3019 flag--;
3020 return flag;
3021 }
3022
3023 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
3024 {
3025 struct user_namespace *s_user_ns;
3026 int error = security_path_mknod(dir, dentry, mode, 0);
3027 if (error)
3028 return error;
3029
3030 s_user_ns = dir->dentry->d_sb->s_user_ns;
3031 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
3032 !kgid_has_mapping(s_user_ns, current_fsgid()))
3033 return -EOVERFLOW;
3034
3035 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
3036 if (error)
3037 return error;
3038
3039 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3040 }
3041
3042 /*
3043 * Attempt to atomically look up, create and open a file from a negative
3044 * dentry.
3045 *
3046 * Returns 0 if successful. The file will have been created and attached to
3047 * @file by the filesystem calling finish_open().
3048 *
3049 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3050 * be set. The caller will need to perform the open themselves. @path will
3051 * have been updated to point to the new dentry. This may be negative.
3052 *
3053 * Returns an error code otherwise.
3054 */
3055 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3056 struct path *path, struct file *file,
3057 const struct open_flags *op,
3058 int open_flag, umode_t mode)
3059 {
3060 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3061 struct inode *dir = nd->path.dentry->d_inode;
3062 int error;
3063
3064 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3065 open_flag &= ~O_TRUNC;
3066
3067 if (nd->flags & LOOKUP_DIRECTORY)
3068 open_flag |= O_DIRECTORY;
3069
3070 file->f_path.dentry = DENTRY_NOT_SET;
3071 file->f_path.mnt = nd->path.mnt;
3072 error = dir->i_op->atomic_open(dir, dentry, file,
3073 open_to_namei_flags(open_flag), mode);
3074 d_lookup_done(dentry);
3075 if (!error) {
3076 if (file->f_mode & FMODE_OPENED) {
3077 /*
3078 * We didn't have the inode before the open, so check open
3079 * permission here.
3080 */
3081 int acc_mode = op->acc_mode;
3082 if (file->f_mode & FMODE_CREATED) {
3083 WARN_ON(!(open_flag & O_CREAT));
3084 fsnotify_create(dir, dentry);
3085 acc_mode = 0;
3086 }
3087 error = may_open(&file->f_path, acc_mode, open_flag);
3088 if (WARN_ON(error > 0))
3089 error = -EINVAL;
3090 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3091 error = -EIO;
3092 } else {
3093 if (file->f_path.dentry) {
3094 dput(dentry);
3095 dentry = file->f_path.dentry;
3096 }
3097 if (file->f_mode & FMODE_CREATED)
3098 fsnotify_create(dir, dentry);
3099 if (unlikely(d_is_negative(dentry))) {
3100 error = -ENOENT;
3101 } else {
3102 path->dentry = dentry;
3103 path->mnt = nd->path.mnt;
3104 return 0;
3105 }
3106 }
3107 }
3108 dput(dentry);
3109 return error;
3110 }
3111
3112 /*
3113 * Look up and maybe create and open the last component.
3114 *
3115 * Must be called with parent locked (exclusive in O_CREAT case).
3116 *
3117 * Returns 0 on success, that is, if
3118 * the file was successfully atomically created (if necessary) and opened, or
3119 * the file was not completely opened at this time, though lookups and
3120 * creations were performed.
3121 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3122 * In the latter case dentry returned in @path might be negative if O_CREAT
3123 * hadn't been specified.
3124 *
3125 * An error code is returned on failure.
3126 */
3127 static int lookup_open(struct nameidata *nd, struct path *path,
3128 struct file *file,
3129 const struct open_flags *op,
3130 bool got_write)
3131 {
3132 struct dentry *dir = nd->path.dentry;
3133 struct inode *dir_inode = dir->d_inode;
3134 int open_flag = op->open_flag;
3135 struct dentry *dentry;
3136 int error, create_error = 0;
3137 umode_t mode = op->mode;
3138 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3139
3140 if (unlikely(IS_DEADDIR(dir_inode)))
3141 return -ENOENT;
3142
3143 file->f_mode &= ~FMODE_CREATED;
3144 dentry = d_lookup(dir, &nd->last);
3145 for (;;) {
3146 if (!dentry) {
3147 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3148 if (IS_ERR(dentry))
3149 return PTR_ERR(dentry);
3150 }
3151 if (d_in_lookup(dentry))
3152 break;
3153
3154 error = d_revalidate(dentry, nd->flags);
3155 if (likely(error > 0))
3156 break;
3157 if (error)
3158 goto out_dput;
3159 d_invalidate(dentry);
3160 dput(dentry);
3161 dentry = NULL;
3162 }
3163 if (dentry->d_inode) {
3164 /* Cached positive dentry: will open in f_op->open */
3165 goto out_no_open;
3166 }
3167
3168 /*
3169 * Checking write permission is tricky, bacuse we don't know if we are
3170 * going to actually need it: O_CREAT opens should work as long as the
3171 * file exists. But checking existence breaks atomicity. The trick is
3172 * to check access and if not granted clear O_CREAT from the flags.
3173 *
3174 * Another problem is returing the "right" error value (e.g. for an
3175 * O_EXCL open we want to return EEXIST not EROFS).
3176 */
3177 if (open_flag & O_CREAT) {
3178 if (!IS_POSIXACL(dir->d_inode))
3179 mode &= ~current_umask();
3180 if (unlikely(!got_write)) {
3181 create_error = -EROFS;
3182 open_flag &= ~O_CREAT;
3183 if (open_flag & (O_EXCL | O_TRUNC))
3184 goto no_open;
3185 /* No side effects, safe to clear O_CREAT */
3186 } else {
3187 create_error = may_o_create(&nd->path, dentry, mode);
3188 if (create_error) {
3189 open_flag &= ~O_CREAT;
3190 if (open_flag & O_EXCL)
3191 goto no_open;
3192 }
3193 }
3194 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3195 unlikely(!got_write)) {
3196 /*
3197 * No O_CREATE -> atomicity not a requirement -> fall
3198 * back to lookup + open
3199 */
3200 goto no_open;
3201 }
3202
3203 if (dir_inode->i_op->atomic_open) {
3204 error = atomic_open(nd, dentry, path, file, op, open_flag,
3205 mode);
3206 if (unlikely(error == -ENOENT) && create_error)
3207 error = create_error;
3208 return error;
3209 }
3210
3211 no_open:
3212 if (d_in_lookup(dentry)) {
3213 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3214 nd->flags);
3215 d_lookup_done(dentry);
3216 if (unlikely(res)) {
3217 if (IS_ERR(res)) {
3218 error = PTR_ERR(res);
3219 goto out_dput;
3220 }
3221 dput(dentry);
3222 dentry = res;
3223 }
3224 }
3225
3226 /* Negative dentry, just create the file */
3227 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3228 file->f_mode |= FMODE_CREATED;
3229 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3230 if (!dir_inode->i_op->create) {
3231 error = -EACCES;
3232 goto out_dput;
3233 }
3234 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3235 open_flag & O_EXCL);
3236 if (error)
3237 goto out_dput;
3238 fsnotify_create(dir_inode, dentry);
3239 }
3240 if (unlikely(create_error) && !dentry->d_inode) {
3241 error = create_error;
3242 goto out_dput;
3243 }
3244 out_no_open:
3245 path->dentry = dentry;
3246 path->mnt = nd->path.mnt;
3247 return 0;
3248
3249 out_dput:
3250 dput(dentry);
3251 return error;
3252 }
3253
3254 /*
3255 * Handle the last step of open()
3256 */
3257 static int do_last(struct nameidata *nd,
3258 struct file *file, const struct open_flags *op)
3259 {
3260 struct dentry *dir = nd->path.dentry;
3261 int open_flag = op->open_flag;
3262 bool will_truncate = (open_flag & O_TRUNC) != 0;
3263 bool got_write = false;
3264 int acc_mode = op->acc_mode;
3265 unsigned seq;
3266 struct inode *inode;
3267 struct path path;
3268 int error;
3269
3270 nd->flags &= ~LOOKUP_PARENT;
3271 nd->flags |= op->intent;
3272
3273 if (nd->last_type != LAST_NORM) {
3274 error = handle_dots(nd, nd->last_type);
3275 if (unlikely(error))
3276 return error;
3277 goto finish_open;
3278 }
3279
3280 if (!(open_flag & O_CREAT)) {
3281 if (nd->last.name[nd->last.len])
3282 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3283 /* we _can_ be in RCU mode here */
3284 error = lookup_fast(nd, &path, &inode, &seq);
3285 if (likely(error > 0))
3286 goto finish_lookup;
3287
3288 if (error < 0)
3289 return error;
3290
3291 BUG_ON(nd->inode != dir->d_inode);
3292 BUG_ON(nd->flags & LOOKUP_RCU);
3293 } else {
3294 /* create side of things */
3295 /*
3296 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3297 * has been cleared when we got to the last component we are
3298 * about to look up
3299 */
3300 error = complete_walk(nd);
3301 if (error)
3302 return error;
3303
3304 audit_inode(nd->name, dir, LOOKUP_PARENT);
3305 /* trailing slashes? */
3306 if (unlikely(nd->last.name[nd->last.len]))
3307 return -EISDIR;
3308 }
3309
3310 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3311 error = mnt_want_write(nd->path.mnt);
3312 if (!error)
3313 got_write = true;
3314 /*
3315 * do _not_ fail yet - we might not need that or fail with
3316 * a different error; let lookup_open() decide; we'll be
3317 * dropping this one anyway.
3318 */
3319 }
3320 if (open_flag & O_CREAT)
3321 inode_lock(dir->d_inode);
3322 else
3323 inode_lock_shared(dir->d_inode);
3324 error = lookup_open(nd, &path, file, op, got_write);
3325 if (open_flag & O_CREAT)
3326 inode_unlock(dir->d_inode);
3327 else
3328 inode_unlock_shared(dir->d_inode);
3329
3330 if (error)
3331 goto out;
3332
3333 if (file->f_mode & FMODE_OPENED) {
3334 if ((file->f_mode & FMODE_CREATED) ||
3335 !S_ISREG(file_inode(file)->i_mode))
3336 will_truncate = false;
3337
3338 audit_inode(nd->name, file->f_path.dentry, 0);
3339 goto opened;
3340 }
3341
3342 if (file->f_mode & FMODE_CREATED) {
3343 /* Don't check for write permission, don't truncate */
3344 open_flag &= ~O_TRUNC;
3345 will_truncate = false;
3346 acc_mode = 0;
3347 path_to_nameidata(&path, nd);
3348 goto finish_open_created;
3349 }
3350
3351 /*
3352 * If atomic_open() acquired write access it is dropped now due to
3353 * possible mount and symlink following (this might be optimized away if
3354 * necessary...)
3355 */
3356 if (got_write) {
3357 mnt_drop_write(nd->path.mnt);
3358 got_write = false;
3359 }
3360
3361 error = follow_managed(&path, nd);
3362 if (unlikely(error < 0))
3363 return error;
3364
3365 if (unlikely(d_is_negative(path.dentry))) {
3366 path_to_nameidata(&path, nd);
3367 return -ENOENT;
3368 }
3369
3370 /*
3371 * create/update audit record if it already exists.
3372 */
3373 audit_inode(nd->name, path.dentry, 0);
3374
3375 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3376 path_to_nameidata(&path, nd);
3377 return -EEXIST;
3378 }
3379
3380 seq = 0; /* out of RCU mode, so the value doesn't matter */
3381 inode = d_backing_inode(path.dentry);
3382 finish_lookup:
3383 error = step_into(nd, &path, 0, inode, seq);
3384 if (unlikely(error))
3385 return error;
3386 finish_open:
3387 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3388 error = complete_walk(nd);
3389 if (error)
3390 return error;
3391 audit_inode(nd->name, nd->path.dentry, 0);
3392 if (open_flag & O_CREAT) {
3393 error = -EISDIR;
3394 if (d_is_dir(nd->path.dentry))
3395 goto out;
3396 error = may_create_in_sticky(dir,
3397 d_backing_inode(nd->path.dentry));
3398 if (unlikely(error))
3399 goto out;
3400 }
3401 error = -ENOTDIR;
3402 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3403 goto out;
3404 if (!d_is_reg(nd->path.dentry))
3405 will_truncate = false;
3406
3407 if (will_truncate) {
3408 error = mnt_want_write(nd->path.mnt);
3409 if (error)
3410 goto out;
3411 got_write = true;
3412 }
3413 finish_open_created:
3414 error = may_open(&nd->path, acc_mode, open_flag);
3415 if (error)
3416 goto out;
3417 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
3418 error = vfs_open(&nd->path, file);
3419 if (error)
3420 goto out;
3421 opened:
3422 error = ima_file_check(file, op->acc_mode);
3423 if (!error && will_truncate)
3424 error = handle_truncate(file);
3425 out:
3426 if (unlikely(error > 0)) {
3427 WARN_ON(1);
3428 error = -EINVAL;
3429 }
3430 if (got_write)
3431 mnt_drop_write(nd->path.mnt);
3432 return error;
3433 }
3434
3435 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3436 {
3437 struct dentry *child = NULL;
3438 struct inode *dir = dentry->d_inode;
3439 struct inode *inode;
3440 int error;
3441
3442 /* we want directory to be writable */
3443 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3444 if (error)
3445 goto out_err;
3446 error = -EOPNOTSUPP;
3447 if (!dir->i_op->tmpfile)
3448 goto out_err;
3449 error = -ENOMEM;
3450 child = d_alloc(dentry, &slash_name);
3451 if (unlikely(!child))
3452 goto out_err;
3453 error = dir->i_op->tmpfile(dir, child, mode);
3454 if (error)
3455 goto out_err;
3456 error = -ENOENT;
3457 inode = child->d_inode;
3458 if (unlikely(!inode))
3459 goto out_err;
3460 if (!(open_flag & O_EXCL)) {
3461 spin_lock(&inode->i_lock);
3462 inode->i_state |= I_LINKABLE;
3463 spin_unlock(&inode->i_lock);
3464 }
3465 return child;
3466
3467 out_err:
3468 dput(child);
3469 return ERR_PTR(error);
3470 }
3471 EXPORT_SYMBOL(vfs_tmpfile);
3472
3473 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3474 const struct open_flags *op,
3475 struct file *file)
3476 {
3477 struct dentry *child;
3478 struct path path;
3479 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3480 if (unlikely(error))
3481 return error;
3482 error = mnt_want_write(path.mnt);
3483 if (unlikely(error))
3484 goto out;
3485 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3486 error = PTR_ERR(child);
3487 if (IS_ERR(child))
3488 goto out2;
3489 dput(path.dentry);
3490 path.dentry = child;
3491 audit_inode(nd->name, child, 0);
3492 /* Don't check for other permissions, the inode was just created */
3493 error = may_open(&path, 0, op->open_flag);
3494 if (error)
3495 goto out2;
3496 file->f_path.mnt = path.mnt;
3497 error = finish_open(file, child, NULL);
3498 out2:
3499 mnt_drop_write(path.mnt);
3500 out:
3501 path_put(&path);
3502 return error;
3503 }
3504
3505 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3506 {
3507 struct path path;
3508 int error = path_lookupat(nd, flags, &path);
3509 if (!error) {
3510 audit_inode(nd->name, path.dentry, 0);
3511 error = vfs_open(&path, file);
3512 path_put(&path);
3513 }
3514 return error;
3515 }
3516
3517 static struct file *path_openat(struct nameidata *nd,
3518 const struct open_flags *op, unsigned flags)
3519 {
3520 struct file *file;
3521 int error;
3522
3523 file = alloc_empty_file(op->open_flag, current_cred());
3524 if (IS_ERR(file))
3525 return file;
3526
3527 if (unlikely(file->f_flags & __O_TMPFILE)) {
3528 error = do_tmpfile(nd, flags, op, file);
3529 } else if (unlikely(file->f_flags & O_PATH)) {
3530 error = do_o_path(nd, flags, file);
3531 } else {
3532 const char *s = path_init(nd, flags);
3533 while (!(error = link_path_walk(s, nd)) &&
3534 (error = do_last(nd, file, op)) > 0) {
3535 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3536 s = trailing_symlink(nd);
3537 }
3538 terminate_walk(nd);
3539 }
3540 if (likely(!error)) {
3541 if (likely(file->f_mode & FMODE_OPENED))
3542 return file;
3543 WARN_ON(1);
3544 error = -EINVAL;
3545 }
3546 fput(file);
3547 if (error == -EOPENSTALE) {
3548 if (flags & LOOKUP_RCU)
3549 error = -ECHILD;
3550 else
3551 error = -ESTALE;
3552 }
3553 return ERR_PTR(error);
3554 }
3555
3556 struct file *do_filp_open(int dfd, struct filename *pathname,
3557 const struct open_flags *op)
3558 {
3559 struct nameidata nd;
3560 int flags = op->lookup_flags;
3561 struct file *filp;
3562
3563 set_nameidata(&nd, dfd, pathname);
3564 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3565 if (unlikely(filp == ERR_PTR(-ECHILD)))
3566 filp = path_openat(&nd, op, flags);
3567 if (unlikely(filp == ERR_PTR(-ESTALE)))
3568 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3569 restore_nameidata();
3570 return filp;
3571 }
3572
3573 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3574 const char *name, const struct open_flags *op)
3575 {
3576 struct nameidata nd;
3577 struct file *file;
3578 struct filename *filename;
3579 int flags = op->lookup_flags | LOOKUP_ROOT;
3580
3581 nd.root.mnt = mnt;
3582 nd.root.dentry = dentry;
3583
3584 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3585 return ERR_PTR(-ELOOP);
3586
3587 filename = getname_kernel(name);
3588 if (IS_ERR(filename))
3589 return ERR_CAST(filename);
3590
3591 set_nameidata(&nd, -1, filename);
3592 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3593 if (unlikely(file == ERR_PTR(-ECHILD)))
3594 file = path_openat(&nd, op, flags);
3595 if (unlikely(file == ERR_PTR(-ESTALE)))
3596 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3597 restore_nameidata();
3598 putname(filename);
3599 return file;
3600 }
3601
3602 static struct dentry *filename_create(int dfd, struct filename *name,
3603 struct path *path, unsigned int lookup_flags)
3604 {
3605 struct dentry *dentry = ERR_PTR(-EEXIST);
3606 struct qstr last;
3607 int type;
3608 int err2;
3609 int error;
3610 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3611
3612 /*
3613 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3614 * other flags passed in are ignored!
3615 */
3616 lookup_flags &= LOOKUP_REVAL;
3617
3618 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3619 if (IS_ERR(name))
3620 return ERR_CAST(name);
3621
3622 /*
3623 * Yucky last component or no last component at all?
3624 * (foo/., foo/.., /////)
3625 */
3626 if (unlikely(type != LAST_NORM))
3627 goto out;
3628
3629 /* don't fail immediately if it's r/o, at least try to report other errors */
3630 err2 = mnt_want_write(path->mnt);
3631 /*
3632 * Do the final lookup.
3633 */
3634 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3635 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3636 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3637 if (IS_ERR(dentry))
3638 goto unlock;
3639
3640 error = -EEXIST;
3641 if (d_is_positive(dentry))
3642 goto fail;
3643
3644 /*
3645 * Special case - lookup gave negative, but... we had foo/bar/
3646 * From the vfs_mknod() POV we just have a negative dentry -
3647 * all is fine. Let's be bastards - you had / on the end, you've
3648 * been asking for (non-existent) directory. -ENOENT for you.
3649 */
3650 if (unlikely(!is_dir && last.name[last.len])) {
3651 error = -ENOENT;
3652 goto fail;
3653 }
3654 if (unlikely(err2)) {
3655 error = err2;
3656 goto fail;
3657 }
3658 putname(name);
3659 return dentry;
3660 fail:
3661 dput(dentry);
3662 dentry = ERR_PTR(error);
3663 unlock:
3664 inode_unlock(path->dentry->d_inode);
3665 if (!err2)
3666 mnt_drop_write(path->mnt);
3667 out:
3668 path_put(path);
3669 putname(name);
3670 return dentry;
3671 }
3672
3673 struct dentry *kern_path_create(int dfd, const char *pathname,
3674 struct path *path, unsigned int lookup_flags)
3675 {
3676 return filename_create(dfd, getname_kernel(pathname),
3677 path, lookup_flags);
3678 }
3679 EXPORT_SYMBOL(kern_path_create);
3680
3681 void done_path_create(struct path *path, struct dentry *dentry)
3682 {
3683 dput(dentry);
3684 inode_unlock(path->dentry->d_inode);
3685 mnt_drop_write(path->mnt);
3686 path_put(path);
3687 }
3688 EXPORT_SYMBOL(done_path_create);
3689
3690 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3691 struct path *path, unsigned int lookup_flags)
3692 {
3693 return filename_create(dfd, getname(pathname), path, lookup_flags);
3694 }
3695 EXPORT_SYMBOL(user_path_create);
3696
3697 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3698 {
3699 int error = may_create(dir, dentry);
3700
3701 if (error)
3702 return error;
3703
3704 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
3705 !ns_capable(dentry->d_sb->s_user_ns, CAP_MKNOD))
3706 return -EPERM;
3707
3708 if (!dir->i_op->mknod)
3709 return -EPERM;
3710
3711 error = devcgroup_inode_mknod(mode, dev);
3712 if (error)
3713 return error;
3714
3715 error = security_inode_mknod(dir, dentry, mode, dev);
3716 if (error)
3717 return error;
3718
3719 error = dir->i_op->mknod(dir, dentry, mode, dev);
3720 if (!error)
3721 fsnotify_create(dir, dentry);
3722 return error;
3723 }
3724 EXPORT_SYMBOL(vfs_mknod);
3725
3726 static int may_mknod(umode_t mode)
3727 {
3728 switch (mode & S_IFMT) {
3729 case S_IFREG:
3730 case S_IFCHR:
3731 case S_IFBLK:
3732 case S_IFIFO:
3733 case S_IFSOCK:
3734 case 0: /* zero mode translates to S_IFREG */
3735 return 0;
3736 case S_IFDIR:
3737 return -EPERM;
3738 default:
3739 return -EINVAL;
3740 }
3741 }
3742
3743 long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3744 unsigned int dev)
3745 {
3746 struct dentry *dentry;
3747 struct path path;
3748 int error;
3749 unsigned int lookup_flags = 0;
3750
3751 error = may_mknod(mode);
3752 if (error)
3753 return error;
3754 retry:
3755 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3756 if (IS_ERR(dentry))
3757 return PTR_ERR(dentry);
3758
3759 if (!IS_POSIXACL(path.dentry->d_inode))
3760 mode &= ~current_umask();
3761 error = security_path_mknod(&path, dentry, mode, dev);
3762 if (error)
3763 goto out;
3764 switch (mode & S_IFMT) {
3765 case 0: case S_IFREG:
3766 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3767 if (!error)
3768 ima_post_path_mknod(dentry);
3769 break;
3770 case S_IFCHR: case S_IFBLK:
3771 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3772 new_decode_dev(dev));
3773 break;
3774 case S_IFIFO: case S_IFSOCK:
3775 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3776 break;
3777 }
3778 out:
3779 done_path_create(&path, dentry);
3780 if (retry_estale(error, lookup_flags)) {
3781 lookup_flags |= LOOKUP_REVAL;
3782 goto retry;
3783 }
3784 return error;
3785 }
3786
3787 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3788 unsigned int, dev)
3789 {
3790 return do_mknodat(dfd, filename, mode, dev);
3791 }
3792
3793 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3794 {
3795 return do_mknodat(AT_FDCWD, filename, mode, dev);
3796 }
3797
3798 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3799 {
3800 int error = may_create(dir, dentry);
3801 unsigned max_links = dir->i_sb->s_max_links;
3802
3803 if (error)
3804 return error;
3805
3806 if (!dir->i_op->mkdir)
3807 return -EPERM;
3808
3809 mode &= (S_IRWXUGO|S_ISVTX);
3810 error = security_inode_mkdir(dir, dentry, mode);
3811 if (error)
3812 return error;
3813
3814 if (max_links && dir->i_nlink >= max_links)
3815 return -EMLINK;
3816
3817 error = dir->i_op->mkdir(dir, dentry, mode);
3818 if (!error)
3819 fsnotify_mkdir(dir, dentry);
3820 return error;
3821 }
3822 EXPORT_SYMBOL(vfs_mkdir);
3823
3824 long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3825 {
3826 struct dentry *dentry;
3827 struct path path;
3828 int error;
3829 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3830
3831 retry:
3832 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3833 if (IS_ERR(dentry))
3834 return PTR_ERR(dentry);
3835
3836 if (!IS_POSIXACL(path.dentry->d_inode))
3837 mode &= ~current_umask();
3838 error = security_path_mkdir(&path, dentry, mode);
3839 if (!error)
3840 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3841 done_path_create(&path, dentry);
3842 if (retry_estale(error, lookup_flags)) {
3843 lookup_flags |= LOOKUP_REVAL;
3844 goto retry;
3845 }
3846 return error;
3847 }
3848
3849 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3850 {
3851 return do_mkdirat(dfd, pathname, mode);
3852 }
3853
3854 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3855 {
3856 return do_mkdirat(AT_FDCWD, pathname, mode);
3857 }
3858
3859 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3860 {
3861 int error = may_delete(dir, dentry, 1);
3862
3863 if (error)
3864 return error;
3865
3866 if (!dir->i_op->rmdir)
3867 return -EPERM;
3868
3869 dget(dentry);
3870 inode_lock(dentry->d_inode);
3871
3872 error = -EBUSY;
3873 if (is_local_mountpoint(dentry))
3874 goto out;
3875
3876 error = security_inode_rmdir(dir, dentry);
3877 if (error)
3878 goto out;
3879
3880 error = dir->i_op->rmdir(dir, dentry);
3881 if (error)
3882 goto out;
3883
3884 shrink_dcache_parent(dentry);
3885 dentry->d_inode->i_flags |= S_DEAD;
3886 dont_mount(dentry);
3887 detach_mounts(dentry);
3888
3889 out:
3890 inode_unlock(dentry->d_inode);
3891 dput(dentry);
3892 if (!error)
3893 d_delete(dentry);
3894 return error;
3895 }
3896 EXPORT_SYMBOL(vfs_rmdir);
3897
3898 long do_rmdir(int dfd, const char __user *pathname)
3899 {
3900 int error = 0;
3901 struct filename *name;
3902 struct dentry *dentry;
3903 struct path path;
3904 struct qstr last;
3905 int type;
3906 unsigned int lookup_flags = 0;
3907 retry:
3908 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3909 &path, &last, &type);
3910 if (IS_ERR(name))
3911 return PTR_ERR(name);
3912
3913 switch (type) {
3914 case LAST_DOTDOT:
3915 error = -ENOTEMPTY;
3916 goto exit1;
3917 case LAST_DOT:
3918 error = -EINVAL;
3919 goto exit1;
3920 case LAST_ROOT:
3921 error = -EBUSY;
3922 goto exit1;
3923 }
3924
3925 error = mnt_want_write(path.mnt);
3926 if (error)
3927 goto exit1;
3928
3929 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3930 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3931 error = PTR_ERR(dentry);
3932 if (IS_ERR(dentry))
3933 goto exit2;
3934 if (!dentry->d_inode) {
3935 error = -ENOENT;
3936 goto exit3;
3937 }
3938 error = security_path_rmdir(&path, dentry);
3939 if (error)
3940 goto exit3;
3941 error = vfs_rmdir(path.dentry->d_inode, dentry);
3942 exit3:
3943 dput(dentry);
3944 exit2:
3945 inode_unlock(path.dentry->d_inode);
3946 mnt_drop_write(path.mnt);
3947 exit1:
3948 path_put(&path);
3949 putname(name);
3950 if (retry_estale(error, lookup_flags)) {
3951 lookup_flags |= LOOKUP_REVAL;
3952 goto retry;
3953 }
3954 return error;
3955 }
3956
3957 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3958 {
3959 return do_rmdir(AT_FDCWD, pathname);
3960 }
3961
3962 /**
3963 * vfs_unlink - unlink a filesystem object
3964 * @dir: parent directory
3965 * @dentry: victim
3966 * @delegated_inode: returns victim inode, if the inode is delegated.
3967 *
3968 * The caller must hold dir->i_mutex.
3969 *
3970 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3971 * return a reference to the inode in delegated_inode. The caller
3972 * should then break the delegation on that inode and retry. Because
3973 * breaking a delegation may take a long time, the caller should drop
3974 * dir->i_mutex before doing so.
3975 *
3976 * Alternatively, a caller may pass NULL for delegated_inode. This may
3977 * be appropriate for callers that expect the underlying filesystem not
3978 * to be NFS exported.
3979 */
3980 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3981 {
3982 struct inode *target = dentry->d_inode;
3983 int error = may_delete(dir, dentry, 0);
3984
3985 if (error)
3986 return error;
3987
3988 if (!dir->i_op->unlink)
3989 return -EPERM;
3990
3991 inode_lock(target);
3992 if (is_local_mountpoint(dentry))
3993 error = -EBUSY;
3994 else {
3995 error = security_inode_unlink(dir, dentry);
3996 if (!error) {
3997 error = try_break_deleg(target, delegated_inode);
3998 if (error)
3999 goto out;
4000 error = dir->i_op->unlink(dir, dentry);
4001 if (!error) {
4002 dont_mount(dentry);
4003 detach_mounts(dentry);
4004 }
4005 }
4006 }
4007 out:
4008 inode_unlock(target);
4009
4010 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4011 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
4012 fsnotify_link_count(target);
4013 d_delete(dentry);
4014 }
4015
4016 return error;
4017 }
4018 EXPORT_SYMBOL(vfs_unlink);
4019
4020 /*
4021 * Make sure that the actual truncation of the file will occur outside its
4022 * directory's i_mutex. Truncate can take a long time if there is a lot of
4023 * writeout happening, and we don't want to prevent access to the directory
4024 * while waiting on the I/O.
4025 */
4026 long do_unlinkat(int dfd, struct filename *name)
4027 {
4028 int error;
4029 struct dentry *dentry;
4030 struct path path;
4031 struct qstr last;
4032 int type;
4033 struct inode *inode = NULL;
4034 struct inode *delegated_inode = NULL;
4035 unsigned int lookup_flags = 0;
4036 retry:
4037 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4038 if (IS_ERR(name))
4039 return PTR_ERR(name);
4040
4041 error = -EISDIR;
4042 if (type != LAST_NORM)
4043 goto exit1;
4044
4045 error = mnt_want_write(path.mnt);
4046 if (error)
4047 goto exit1;
4048 retry_deleg:
4049 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4050 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4051 error = PTR_ERR(dentry);
4052 if (!IS_ERR(dentry)) {
4053 /* Why not before? Because we want correct error value */
4054 if (last.name[last.len])
4055 goto slashes;
4056 inode = dentry->d_inode;
4057 if (d_is_negative(dentry))
4058 goto slashes;
4059 ihold(inode);
4060 error = security_path_unlink(&path, dentry);
4061 if (error)
4062 goto exit2;
4063 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4064 exit2:
4065 dput(dentry);
4066 }
4067 inode_unlock(path.dentry->d_inode);
4068 if (inode)
4069 iput(inode); /* truncate the inode here */
4070 inode = NULL;
4071 if (delegated_inode) {
4072 error = break_deleg_wait(&delegated_inode);
4073 if (!error)
4074 goto retry_deleg;
4075 }
4076 mnt_drop_write(path.mnt);
4077 exit1:
4078 path_put(&path);
4079 if (retry_estale(error, lookup_flags)) {
4080 lookup_flags |= LOOKUP_REVAL;
4081 inode = NULL;
4082 goto retry;
4083 }
4084 putname(name);
4085 return error;
4086
4087 slashes:
4088 if (d_is_negative(dentry))
4089 error = -ENOENT;
4090 else if (d_is_dir(dentry))
4091 error = -EISDIR;
4092 else
4093 error = -ENOTDIR;
4094 goto exit2;
4095 }
4096
4097 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4098 {
4099 if ((flag & ~AT_REMOVEDIR) != 0)
4100 return -EINVAL;
4101
4102 if (flag & AT_REMOVEDIR)
4103 return do_rmdir(dfd, pathname);
4104
4105 return do_unlinkat(dfd, getname(pathname));
4106 }
4107
4108 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4109 {
4110 return do_unlinkat(AT_FDCWD, getname(pathname));
4111 }
4112
4113 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4114 {
4115 int error = may_create(dir, dentry);
4116
4117 if (error)
4118 return error;
4119
4120 if (!dir->i_op->symlink)
4121 return -EPERM;
4122
4123 error = security_inode_symlink(dir, dentry, oldname);
4124 if (error)
4125 return error;
4126
4127 error = dir->i_op->symlink(dir, dentry, oldname);
4128 if (!error)
4129 fsnotify_create(dir, dentry);
4130 return error;
4131 }
4132 EXPORT_SYMBOL(vfs_symlink);
4133
4134 long do_symlinkat(const char __user *oldname, int newdfd,
4135 const char __user *newname)
4136 {
4137 int error;
4138 struct filename *from;
4139 struct dentry *dentry;
4140 struct path path;
4141 unsigned int lookup_flags = 0;
4142
4143 from = getname(oldname);
4144 if (IS_ERR(from))
4145 return PTR_ERR(from);
4146 retry:
4147 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4148 error = PTR_ERR(dentry);
4149 if (IS_ERR(dentry))
4150 goto out_putname;
4151
4152 error = security_path_symlink(&path, dentry, from->name);
4153 if (!error)
4154 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4155 done_path_create(&path, dentry);
4156 if (retry_estale(error, lookup_flags)) {
4157 lookup_flags |= LOOKUP_REVAL;
4158 goto retry;
4159 }
4160 out_putname:
4161 putname(from);
4162 return error;
4163 }
4164
4165 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4166 int, newdfd, const char __user *, newname)
4167 {
4168 return do_symlinkat(oldname, newdfd, newname);
4169 }
4170
4171 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4172 {
4173 return do_symlinkat(oldname, AT_FDCWD, newname);
4174 }
4175
4176 /**
4177 * vfs_link - create a new link
4178 * @old_dentry: object to be linked
4179 * @dir: new parent
4180 * @new_dentry: where to create the new link
4181 * @delegated_inode: returns inode needing a delegation break
4182 *
4183 * The caller must hold dir->i_mutex
4184 *
4185 * If vfs_link discovers a delegation on the to-be-linked file in need
4186 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4187 * inode in delegated_inode. The caller should then break the delegation
4188 * and retry. Because breaking a delegation may take a long time, the
4189 * caller should drop the i_mutex before doing so.
4190 *
4191 * Alternatively, a caller may pass NULL for delegated_inode. This may
4192 * be appropriate for callers that expect the underlying filesystem not
4193 * to be NFS exported.
4194 */
4195 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4196 {
4197 struct inode *inode = old_dentry->d_inode;
4198 unsigned max_links = dir->i_sb->s_max_links;
4199 int error;
4200
4201 if (!inode)
4202 return -ENOENT;
4203
4204 error = may_create(dir, new_dentry);
4205 if (error)
4206 return error;
4207
4208 if (dir->i_sb != inode->i_sb)
4209 return -EXDEV;
4210
4211 /*
4212 * A link to an append-only or immutable file cannot be created.
4213 */
4214 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4215 return -EPERM;
4216 /*
4217 * Updating the link count will likely cause i_uid and i_gid to
4218 * be writen back improperly if their true value is unknown to
4219 * the vfs.
4220 */
4221 if (HAS_UNMAPPED_ID(inode))
4222 return -EPERM;
4223 if (!dir->i_op->link)
4224 return -EPERM;
4225 if (S_ISDIR(inode->i_mode))
4226 return -EPERM;
4227
4228 error = security_inode_link(old_dentry, dir, new_dentry);
4229 if (error)
4230 return error;
4231
4232 inode_lock(inode);
4233 /* Make sure we don't allow creating hardlink to an unlinked file */
4234 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4235 error = -ENOENT;
4236 else if (max_links && inode->i_nlink >= max_links)
4237 error = -EMLINK;
4238 else {
4239 error = try_break_deleg(inode, delegated_inode);
4240 if (!error)
4241 error = dir->i_op->link(old_dentry, dir, new_dentry);
4242 }
4243
4244 if (!error && (inode->i_state & I_LINKABLE)) {
4245 spin_lock(&inode->i_lock);
4246 inode->i_state &= ~I_LINKABLE;
4247 spin_unlock(&inode->i_lock);
4248 }
4249 inode_unlock(inode);
4250 if (!error)
4251 fsnotify_link(dir, inode, new_dentry);
4252 return error;
4253 }
4254 EXPORT_SYMBOL(vfs_link);
4255
4256 /*
4257 * Hardlinks are often used in delicate situations. We avoid
4258 * security-related surprises by not following symlinks on the
4259 * newname. --KAB
4260 *
4261 * We don't follow them on the oldname either to be compatible
4262 * with linux 2.0, and to avoid hard-linking to directories
4263 * and other special files. --ADM
4264 */
4265 int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4266 const char __user *newname, int flags)
4267 {
4268 struct dentry *new_dentry;
4269 struct path old_path, new_path;
4270 struct inode *delegated_inode = NULL;
4271 int how = 0;
4272 int error;
4273
4274 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4275 return -EINVAL;
4276 /*
4277 * To use null names we require CAP_DAC_READ_SEARCH
4278 * This ensures that not everyone will be able to create
4279 * handlink using the passed filedescriptor.
4280 */
4281 if (flags & AT_EMPTY_PATH) {
4282 if (!capable(CAP_DAC_READ_SEARCH))
4283 return -ENOENT;
4284 how = LOOKUP_EMPTY;
4285 }
4286
4287 if (flags & AT_SYMLINK_FOLLOW)
4288 how |= LOOKUP_FOLLOW;
4289 retry:
4290 error = user_path_at(olddfd, oldname, how, &old_path);
4291 if (error)
4292 return error;
4293
4294 new_dentry = user_path_create(newdfd, newname, &new_path,
4295 (how & LOOKUP_REVAL));
4296 error = PTR_ERR(new_dentry);
4297 if (IS_ERR(new_dentry))
4298 goto out;
4299
4300 error = -EXDEV;
4301 if (old_path.mnt != new_path.mnt)
4302 goto out_dput;
4303 error = may_linkat(&old_path);
4304 if (unlikely(error))
4305 goto out_dput;
4306 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4307 if (error)
4308 goto out_dput;
4309 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4310 out_dput:
4311 done_path_create(&new_path, new_dentry);
4312 if (delegated_inode) {
4313 error = break_deleg_wait(&delegated_inode);
4314 if (!error) {
4315 path_put(&old_path);
4316 goto retry;
4317 }
4318 }
4319 if (retry_estale(error, how)) {
4320 path_put(&old_path);
4321 how |= LOOKUP_REVAL;
4322 goto retry;
4323 }
4324 out:
4325 path_put(&old_path);
4326
4327 return error;
4328 }
4329
4330 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4331 int, newdfd, const char __user *, newname, int, flags)
4332 {
4333 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4334 }
4335
4336 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4337 {
4338 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4339 }
4340
4341 /**
4342 * vfs_rename - rename a filesystem object
4343 * @old_dir: parent of source
4344 * @old_dentry: source
4345 * @new_dir: parent of destination
4346 * @new_dentry: destination
4347 * @delegated_inode: returns an inode needing a delegation break
4348 * @flags: rename flags
4349 *
4350 * The caller must hold multiple mutexes--see lock_rename()).
4351 *
4352 * If vfs_rename discovers a delegation in need of breaking at either
4353 * the source or destination, it will return -EWOULDBLOCK and return a
4354 * reference to the inode in delegated_inode. The caller should then
4355 * break the delegation and retry. Because breaking a delegation may
4356 * take a long time, the caller should drop all locks before doing
4357 * so.
4358 *
4359 * Alternatively, a caller may pass NULL for delegated_inode. This may
4360 * be appropriate for callers that expect the underlying filesystem not
4361 * to be NFS exported.
4362 *
4363 * The worst of all namespace operations - renaming directory. "Perverted"
4364 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4365 * Problems:
4366 *
4367 * a) we can get into loop creation.
4368 * b) race potential - two innocent renames can create a loop together.
4369 * That's where 4.4 screws up. Current fix: serialization on
4370 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4371 * story.
4372 * c) we have to lock _four_ objects - parents and victim (if it exists),
4373 * and source (if it is not a directory).
4374 * And that - after we got ->i_mutex on parents (until then we don't know
4375 * whether the target exists). Solution: try to be smart with locking
4376 * order for inodes. We rely on the fact that tree topology may change
4377 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4378 * move will be locked. Thus we can rank directories by the tree
4379 * (ancestors first) and rank all non-directories after them.
4380 * That works since everybody except rename does "lock parent, lookup,
4381 * lock child" and rename is under ->s_vfs_rename_mutex.
4382 * HOWEVER, it relies on the assumption that any object with ->lookup()
4383 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4384 * we'd better make sure that there's no link(2) for them.
4385 * d) conversion from fhandle to dentry may come in the wrong moment - when
4386 * we are removing the target. Solution: we will have to grab ->i_mutex
4387 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4388 * ->i_mutex on parents, which works but leads to some truly excessive
4389 * locking].
4390 */
4391 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4392 struct inode *new_dir, struct dentry *new_dentry,
4393 struct inode **delegated_inode, unsigned int flags)
4394 {
4395 int error;
4396 bool is_dir = d_is_dir(old_dentry);
4397 struct inode *source = old_dentry->d_inode;
4398 struct inode *target = new_dentry->d_inode;
4399 bool new_is_dir = false;
4400 unsigned max_links = new_dir->i_sb->s_max_links;
4401 struct name_snapshot old_name;
4402
4403 if (source == target)
4404 return 0;
4405
4406 error = may_delete(old_dir, old_dentry, is_dir);
4407 if (error)
4408 return error;
4409
4410 if (!target) {
4411 error = may_create(new_dir, new_dentry);
4412 } else {
4413 new_is_dir = d_is_dir(new_dentry);
4414
4415 if (!(flags & RENAME_EXCHANGE))
4416 error = may_delete(new_dir, new_dentry, is_dir);
4417 else
4418 error = may_delete(new_dir, new_dentry, new_is_dir);
4419 }
4420 if (error)
4421 return error;
4422
4423 if (!old_dir->i_op->rename)
4424 return -EPERM;
4425
4426 /*
4427 * If we are going to change the parent - check write permissions,
4428 * we'll need to flip '..'.
4429 */
4430 if (new_dir != old_dir) {
4431 if (is_dir) {
4432 error = inode_permission(source, MAY_WRITE);
4433 if (error)
4434 return error;
4435 }
4436 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4437 error = inode_permission(target, MAY_WRITE);
4438 if (error)
4439 return error;
4440 }
4441 }
4442
4443 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4444 flags);
4445 if (error)
4446 return error;
4447
4448 take_dentry_name_snapshot(&old_name, old_dentry);
4449 dget(new_dentry);
4450 if (!is_dir || (flags & RENAME_EXCHANGE))
4451 lock_two_nondirectories(source, target);
4452 else if (target)
4453 inode_lock(target);
4454
4455 error = -EBUSY;
4456 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4457 goto out;
4458
4459 if (max_links && new_dir != old_dir) {
4460 error = -EMLINK;
4461 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4462 goto out;
4463 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4464 old_dir->i_nlink >= max_links)
4465 goto out;
4466 }
4467 if (!is_dir) {
4468 error = try_break_deleg(source, delegated_inode);
4469 if (error)
4470 goto out;
4471 }
4472 if (target && !new_is_dir) {
4473 error = try_break_deleg(target, delegated_inode);
4474 if (error)
4475 goto out;
4476 }
4477 error = old_dir->i_op->rename(old_dir, old_dentry,
4478 new_dir, new_dentry, flags);
4479 if (error)
4480 goto out;
4481
4482 if (!(flags & RENAME_EXCHANGE) && target) {
4483 if (is_dir) {
4484 shrink_dcache_parent(new_dentry);
4485 target->i_flags |= S_DEAD;
4486 }
4487 dont_mount(new_dentry);
4488 detach_mounts(new_dentry);
4489 }
4490 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4491 if (!(flags & RENAME_EXCHANGE))
4492 d_move(old_dentry, new_dentry);
4493 else
4494 d_exchange(old_dentry, new_dentry);
4495 }
4496 out:
4497 if (!is_dir || (flags & RENAME_EXCHANGE))
4498 unlock_two_nondirectories(source, target);
4499 else if (target)
4500 inode_unlock(target);
4501 dput(new_dentry);
4502 if (!error) {
4503 fsnotify_move(old_dir, new_dir, old_name.name, is_dir,
4504 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4505 if (flags & RENAME_EXCHANGE) {
4506 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4507 new_is_dir, NULL, new_dentry);
4508 }
4509 }
4510 release_dentry_name_snapshot(&old_name);
4511
4512 return error;
4513 }
4514 EXPORT_SYMBOL(vfs_rename);
4515
4516 static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4517 const char __user *newname, unsigned int flags)
4518 {
4519 struct dentry *old_dentry, *new_dentry;
4520 struct dentry *trap;
4521 struct path old_path, new_path;
4522 struct qstr old_last, new_last;
4523 int old_type, new_type;
4524 struct inode *delegated_inode = NULL;
4525 struct filename *from;
4526 struct filename *to;
4527 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4528 bool should_retry = false;
4529 int error;
4530
4531 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4532 return -EINVAL;
4533
4534 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4535 (flags & RENAME_EXCHANGE))
4536 return -EINVAL;
4537
4538 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4539 return -EPERM;
4540
4541 if (flags & RENAME_EXCHANGE)
4542 target_flags = 0;
4543
4544 retry:
4545 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4546 &old_path, &old_last, &old_type);
4547 if (IS_ERR(from)) {
4548 error = PTR_ERR(from);
4549 goto exit;
4550 }
4551
4552 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4553 &new_path, &new_last, &new_type);
4554 if (IS_ERR(to)) {
4555 error = PTR_ERR(to);
4556 goto exit1;
4557 }
4558
4559 error = -EXDEV;
4560 if (old_path.mnt != new_path.mnt)
4561 goto exit2;
4562
4563 error = -EBUSY;
4564 if (old_type != LAST_NORM)
4565 goto exit2;
4566
4567 if (flags & RENAME_NOREPLACE)
4568 error = -EEXIST;
4569 if (new_type != LAST_NORM)
4570 goto exit2;
4571
4572 error = mnt_want_write(old_path.mnt);
4573 if (error)
4574 goto exit2;
4575
4576 retry_deleg:
4577 trap = lock_rename(new_path.dentry, old_path.dentry);
4578
4579 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4580 error = PTR_ERR(old_dentry);
4581 if (IS_ERR(old_dentry))
4582 goto exit3;
4583 /* source must exist */
4584 error = -ENOENT;
4585 if (d_is_negative(old_dentry))
4586 goto exit4;
4587 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4588 error = PTR_ERR(new_dentry);
4589 if (IS_ERR(new_dentry))
4590 goto exit4;
4591 error = -EEXIST;
4592 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4593 goto exit5;
4594 if (flags & RENAME_EXCHANGE) {
4595 error = -ENOENT;
4596 if (d_is_negative(new_dentry))
4597 goto exit5;
4598
4599 if (!d_is_dir(new_dentry)) {
4600 error = -ENOTDIR;
4601 if (new_last.name[new_last.len])
4602 goto exit5;
4603 }
4604 }
4605 /* unless the source is a directory trailing slashes give -ENOTDIR */
4606 if (!d_is_dir(old_dentry)) {
4607 error = -ENOTDIR;
4608 if (old_last.name[old_last.len])
4609 goto exit5;
4610 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4611 goto exit5;
4612 }
4613 /* source should not be ancestor of target */
4614 error = -EINVAL;
4615 if (old_dentry == trap)
4616 goto exit5;
4617 /* target should not be an ancestor of source */
4618 if (!(flags & RENAME_EXCHANGE))
4619 error = -ENOTEMPTY;
4620 if (new_dentry == trap)
4621 goto exit5;
4622
4623 error = security_path_rename(&old_path, old_dentry,
4624 &new_path, new_dentry, flags);
4625 if (error)
4626 goto exit5;
4627 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4628 new_path.dentry->d_inode, new_dentry,
4629 &delegated_inode, flags);
4630 exit5:
4631 dput(new_dentry);
4632 exit4:
4633 dput(old_dentry);
4634 exit3:
4635 unlock_rename(new_path.dentry, old_path.dentry);
4636 if (delegated_inode) {
4637 error = break_deleg_wait(&delegated_inode);
4638 if (!error)
4639 goto retry_deleg;
4640 }
4641 mnt_drop_write(old_path.mnt);
4642 exit2:
4643 if (retry_estale(error, lookup_flags))
4644 should_retry = true;
4645 path_put(&new_path);
4646 putname(to);
4647 exit1:
4648 path_put(&old_path);
4649 putname(from);
4650 if (should_retry) {
4651 should_retry = false;
4652 lookup_flags |= LOOKUP_REVAL;
4653 goto retry;
4654 }
4655 exit:
4656 return error;
4657 }
4658
4659 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4660 int, newdfd, const char __user *, newname, unsigned int, flags)
4661 {
4662 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4663 }
4664
4665 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4666 int, newdfd, const char __user *, newname)
4667 {
4668 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
4669 }
4670
4671 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4672 {
4673 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4674 }
4675
4676 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4677 {
4678 int error = may_create(dir, dentry);
4679 if (error)
4680 return error;
4681
4682 if (!dir->i_op->mknod)
4683 return -EPERM;
4684
4685 return dir->i_op->mknod(dir, dentry,
4686 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4687 }
4688 EXPORT_SYMBOL(vfs_whiteout);
4689
4690 int readlink_copy(char __user *buffer, int buflen, const char *link)
4691 {
4692 int len = PTR_ERR(link);
4693 if (IS_ERR(link))
4694 goto out;
4695
4696 len = strlen(link);
4697 if (len > (unsigned) buflen)
4698 len = buflen;
4699 if (copy_to_user(buffer, link, len))
4700 len = -EFAULT;
4701 out:
4702 return len;
4703 }
4704
4705 /**
4706 * vfs_readlink - copy symlink body into userspace buffer
4707 * @dentry: dentry on which to get symbolic link
4708 * @buffer: user memory pointer
4709 * @buflen: size of buffer
4710 *
4711 * Does not touch atime. That's up to the caller if necessary
4712 *
4713 * Does not call security hook.
4714 */
4715 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4716 {
4717 struct inode *inode = d_inode(dentry);
4718 DEFINE_DELAYED_CALL(done);
4719 const char *link;
4720 int res;
4721
4722 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4723 if (unlikely(inode->i_op->readlink))
4724 return inode->i_op->readlink(dentry, buffer, buflen);
4725
4726 if (!d_is_symlink(dentry))
4727 return -EINVAL;
4728
4729 spin_lock(&inode->i_lock);
4730 inode->i_opflags |= IOP_DEFAULT_READLINK;
4731 spin_unlock(&inode->i_lock);
4732 }
4733
4734 link = inode->i_link;
4735 if (!link) {
4736 link = inode->i_op->get_link(dentry, inode, &done);
4737 if (IS_ERR(link))
4738 return PTR_ERR(link);
4739 }
4740 res = readlink_copy(buffer, buflen, link);
4741 do_delayed_call(&done);
4742 return res;
4743 }
4744 EXPORT_SYMBOL(vfs_readlink);
4745
4746 /**
4747 * vfs_get_link - get symlink body
4748 * @dentry: dentry on which to get symbolic link
4749 * @done: caller needs to free returned data with this
4750 *
4751 * Calls security hook and i_op->get_link() on the supplied inode.
4752 *
4753 * It does not touch atime. That's up to the caller if necessary.
4754 *
4755 * Does not work on "special" symlinks like /proc/$$/fd/N
4756 */
4757 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4758 {
4759 const char *res = ERR_PTR(-EINVAL);
4760 struct inode *inode = d_inode(dentry);
4761
4762 if (d_is_symlink(dentry)) {
4763 res = ERR_PTR(security_inode_readlink(dentry));
4764 if (!res)
4765 res = inode->i_op->get_link(dentry, inode, done);
4766 }
4767 return res;
4768 }
4769 EXPORT_SYMBOL(vfs_get_link);
4770
4771 /* get the link contents into pagecache */
4772 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4773 struct delayed_call *callback)
4774 {
4775 char *kaddr;
4776 struct page *page;
4777 struct address_space *mapping = inode->i_mapping;
4778
4779 if (!dentry) {
4780 page = find_get_page(mapping, 0);
4781 if (!page)
4782 return ERR_PTR(-ECHILD);
4783 if (!PageUptodate(page)) {
4784 put_page(page);
4785 return ERR_PTR(-ECHILD);
4786 }
4787 } else {
4788 page = read_mapping_page(mapping, 0, NULL);
4789 if (IS_ERR(page))
4790 return (char*)page;
4791 }
4792 set_delayed_call(callback, page_put_link, page);
4793 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4794 kaddr = page_address(page);
4795 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4796 return kaddr;
4797 }
4798
4799 EXPORT_SYMBOL(page_get_link);
4800
4801 void page_put_link(void *arg)
4802 {
4803 put_page(arg);
4804 }
4805 EXPORT_SYMBOL(page_put_link);
4806
4807 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4808 {
4809 DEFINE_DELAYED_CALL(done);
4810 int res = readlink_copy(buffer, buflen,
4811 page_get_link(dentry, d_inode(dentry),
4812 &done));
4813 do_delayed_call(&done);
4814 return res;
4815 }
4816 EXPORT_SYMBOL(page_readlink);
4817
4818 /*
4819 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4820 */
4821 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4822 {
4823 struct address_space *mapping = inode->i_mapping;
4824 struct page *page;
4825 void *fsdata;
4826 int err;
4827 unsigned int flags = 0;
4828 if (nofs)
4829 flags |= AOP_FLAG_NOFS;
4830
4831 retry:
4832 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4833 flags, &page, &fsdata);
4834 if (err)
4835 goto fail;
4836
4837 memcpy(page_address(page), symname, len-1);
4838
4839 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4840 page, fsdata);
4841 if (err < 0)
4842 goto fail;
4843 if (err < len-1)
4844 goto retry;
4845
4846 mark_inode_dirty(inode);
4847 return 0;
4848 fail:
4849 return err;
4850 }
4851 EXPORT_SYMBOL(__page_symlink);
4852
4853 int page_symlink(struct inode *inode, const char *symname, int len)
4854 {
4855 return __page_symlink(inode, symname, len,
4856 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4857 }
4858 EXPORT_SYMBOL(page_symlink);
4859
4860 const struct inode_operations page_symlink_inode_operations = {
4861 .get_link = page_get_link,
4862 };
4863 EXPORT_SYMBOL(page_symlink_inode_operations);