projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
fs: rcu-walk for path lookup
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / namei.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
1da177e4 22#include <linux/pagemap.h>
0eeca283 23#include <linux/fsnotify.h>
1da177e4
LT		 24#include <linux/personality.h>
25#include <linux/security.h>
6146f0d5 26#include <linux/ima.h>
1da177e4
LT		 27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
16f7e0fe 30#include <linux/capability.h>
834f2a4a 31#include <linux/file.h>
5590ff0d 32#include <linux/fcntl.h>
08ce5f16 33#include <linux/device_cgroup.h>
5ad4e53b 34#include <linux/fs_struct.h>
1da177e4
LT		 35#include <asm/uaccess.h>
36
e81e3f4d
EP		 37#include "internal.h"
38
1da177e4
LT		 39/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 107 * any extra contention...
108 */
109
110/* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
113 *
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
116 */
858119e1 117static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 118{
119 int retval;
120 unsigned long len = PATH_MAX;
121
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
127 }
128
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
137}
138
139char * getname(const char __user * filename)
140{
141 char *tmp, *result;
142
143 result = ERR_PTR(-ENOMEM);
144 tmp = __getname();
145 if (tmp) {
146 int retval = do_getname(filename, tmp);
147
148 result = tmp;
149 if (retval < 0) {
150 __putname(tmp);
151 result = ERR_PTR(retval);
152 }
153 }
154 audit_getname(result);
155 return result;
156}
157
158#ifdef CONFIG_AUDITSYSCALL
159void putname(const char *name)
160{
5ac3a9c2 161 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 162 audit_putname(name);
163 else
164 __putname(name);
165}
166EXPORT_SYMBOL(putname);
167#endif
168
5909ccaa
LT		 169/*
170 * This does basic POSIX ACL permission checking
1da177e4 171 */
31e6b01f
NP		 172static inline int __acl_permission_check(struct inode *inode, int mask,
173 int (*check_acl)(struct inode *inode, int mask), int rcu)
1da177e4
LT		 174{
175 umode_t mode = inode->i_mode;
176
e6305c43
AV		 177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
178
da9592ed 179 if (current_fsuid() == inode->i_uid)
1da177e4
LT		 180 mode >>= 6;
181 else {
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
31e6b01f
NP		 183 if (rcu) {
184 return -ECHILD;
185 } else {
186 int error = check_acl(inode, mask);
187 if (error != -EAGAIN)
188 return error;
189 }
1da177e4
LT		 190 }
191
192 if (in_group_p(inode->i_gid))
193 mode >>= 3;
194 }
195
196 /*
197 * If the DACs are ok we don't need any capability check.
198 */
e6305c43 199 if ((mask & ~mode) == 0)
1da177e4 200 return 0;
5909ccaa
LT		 201 return -EACCES;
202}
203
31e6b01f
NP		 204static inline int acl_permission_check(struct inode *inode, int mask,
205 int (*check_acl)(struct inode *inode, int mask))
206{
207 return __acl_permission_check(inode, mask, check_acl, 0);
208}
209
5909ccaa
LT		 210/**
211 * generic_permission - check for access rights on a Posix-like filesystem
212 * @inode: inode to check access rights for
213 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
214 * @check_acl: optional callback to check for Posix ACLs
215 *
216 * Used to check for read/write/execute permissions on a file.
217 * We use "fsuid" for this, letting us set arbitrary permissions
218 * for filesystem access without changing the "normal" uids which
219 * are used for other things..
220 */
221int generic_permission(struct inode *inode, int mask,
222 int (*check_acl)(struct inode *inode, int mask))
223{
224 int ret;
225
226 /*
227 * Do the basic POSIX ACL permission checks.
228 */
229 ret = acl_permission_check(inode, mask, check_acl);
230 if (ret != -EACCES)
231 return ret;
1da177e4 232
1da177e4
LT		 233 /*
234 * Read/write DACs are always overridable.
235 * Executable DACs are overridable if at least one exec bit is set.
236 */
f696a365 237 if (!(mask & MAY_EXEC) || execute_ok(inode))
1da177e4
LT		 238 if (capable(CAP_DAC_OVERRIDE))
239 return 0;
240
241 /*
242 * Searching includes executable on directories, else just read.
243 */
7ea66001 244 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
1da177e4
LT		 245 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
246 if (capable(CAP_DAC_READ_SEARCH))
247 return 0;
248
249 return -EACCES;
250}
251
cb23beb5
CH		 252/**
253 * inode_permission - check for access rights to a given inode
254 * @inode: inode to check permission on
255 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
256 *
257 * Used to check for read/write/execute permissions on an inode.
258 * We use "fsuid" for this, letting us set arbitrary permissions
259 * for filesystem access without changing the "normal" uids which
260 * are used for other things.
261 */
f419a2e3 262int inode_permission(struct inode *inode, int mask)
1da177e4 263{
e6305c43 264 int retval;
1da177e4
LT		 265
266 if (mask & MAY_WRITE) {
22590e41 267 umode_t mode = inode->i_mode;
1da177e4
LT		 268
269 /*
270 * Nobody gets write access to a read-only fs.
271 */
272 if (IS_RDONLY(inode) &&
273 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
274 return -EROFS;
275
276 /*
277 * Nobody gets write access to an immutable file.
278 */
279 if (IS_IMMUTABLE(inode))
280 return -EACCES;
281 }
282
acfa4380 283 if (inode->i_op->permission)
b77b0646 284 retval = inode->i_op->permission(inode, mask);
f696a365 285 else
5909ccaa 286 retval = generic_permission(inode, mask, inode->i_op->check_acl);
f696a365 287
1da177e4
LT		 288 if (retval)
289 return retval;
290
08ce5f16
SH		 291 retval = devcgroup_inode_permission(inode, mask);
292 if (retval)
293 return retval;
294
d09ca739 295 return security_inode_permission(inode, mask);
1da177e4
LT		 296}
297
8c744fb8
CH		 298/**
299 * file_permission - check for additional access rights to a given file
300 * @file: file to check access rights for
301 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
302 *
303 * Used to check for read/write/execute permissions on an already opened
304 * file.
305 *
306 * Note:
307 * Do not use this function in new code. All access checks should
cb23beb5 308 * be done using inode_permission().
8c744fb8
CH		 309 */
310int file_permission(struct file *file, int mask)
311{
f419a2e3 312 return inode_permission(file->f_path.dentry->d_inode, mask);
8c744fb8
CH		 313}
314
1da177e4
LT		 315/*
316 * get_write_access() gets write permission for a file.
317 * put_write_access() releases this write permission.
318 * This is used for regular files.
319 * We cannot support write (and maybe mmap read-write shared) accesses and
320 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
321 * can have the following values:
322 * 0: no writers, no VM_DENYWRITE mappings
323 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
324 * > 0: (i_writecount) users are writing to the file.
325 *
326 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
327 * except for the cases where we don't hold i_writecount yet. Then we need to
328 * use {get,deny}_write_access() - these functions check the sign and refuse
329 * to do the change if sign is wrong. Exclusion between them is provided by
330 * the inode->i_lock spinlock.
331 */
332
333int get_write_access(struct inode * inode)
334{
335 spin_lock(&inode->i_lock);
336 if (atomic_read(&inode->i_writecount) < 0) {
337 spin_unlock(&inode->i_lock);
338 return -ETXTBSY;
339 }
340 atomic_inc(&inode->i_writecount);
341 spin_unlock(&inode->i_lock);
342
343 return 0;
344}
345
346int deny_write_access(struct file * file)
347{
0f7fc9e4 348 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4
LT		 349
350 spin_lock(&inode->i_lock);
351 if (atomic_read(&inode->i_writecount) > 0) {
352 spin_unlock(&inode->i_lock);
353 return -ETXTBSY;
354 }
355 atomic_dec(&inode->i_writecount);
356 spin_unlock(&inode->i_lock);
357
358 return 0;
359}
360
5dd784d0
JB		 361/**
362 * path_get - get a reference to a path
363 * @path: path to get the reference to
364 *
365 * Given a path increment the reference count to the dentry and the vfsmount.
366 */
367void path_get(struct path *path)
368{
369 mntget(path->mnt);
370 dget(path->dentry);
371}
372EXPORT_SYMBOL(path_get);
373
1d957f9b
JB		 374/**
375 * path_put - put a reference to a path
376 * @path: path to put the reference to
377 *
378 * Given a path decrement the reference count to the dentry and the vfsmount.
379 */
380void path_put(struct path *path)
1da177e4 381{
1d957f9b
JB		 382 dput(path->dentry);
383 mntput(path->mnt);
1da177e4 384}
1d957f9b 385EXPORT_SYMBOL(path_put);
1da177e4 386
31e6b01f
NP		 387/**
388 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
389 * @nd: nameidata pathwalk data to drop
390 * @Returns: 0 on success, -ECHLID on failure
391 *
392 * Path walking has 2 modes, rcu-walk and ref-walk (see
393 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
394 * to drop out of rcu-walk mode and take normal reference counts on dentries
395 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
396 * refcounts at the last known good point before rcu-walk got stuck, so
397 * ref-walk may continue from there. If this is not successful (eg. a seqcount
398 * has changed), then failure is returned and path walk restarts from the
399 * beginning in ref-walk mode.
400 *
401 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
402 * ref-walk. Must be called from rcu-walk context.
403 */
404static int nameidata_drop_rcu(struct nameidata *nd)
405{
406 struct fs_struct *fs = current->fs;
407 struct dentry *dentry = nd->path.dentry;
408
409 BUG_ON(!(nd->flags & LOOKUP_RCU));
410 if (nd->root.mnt) {
411 spin_lock(&fs->lock);
412 if (nd->root.mnt != fs->root.mnt ||
413 nd->root.dentry != fs->root.dentry)
414 goto err_root;
415 }
416 spin_lock(&dentry->d_lock);
417 if (!__d_rcu_to_refcount(dentry, nd->seq))
418 goto err;
419 BUG_ON(nd->inode != dentry->d_inode);
420 spin_unlock(&dentry->d_lock);
421 if (nd->root.mnt) {
422 path_get(&nd->root);
423 spin_unlock(&fs->lock);
424 }
425 mntget(nd->path.mnt);
426
427 rcu_read_unlock();
428 br_read_unlock(vfsmount_lock);
429 nd->flags &= ~LOOKUP_RCU;
430 return 0;
431err:
432 spin_unlock(&dentry->d_lock);
433err_root:
434 if (nd->root.mnt)
435 spin_unlock(&fs->lock);
436 return -ECHILD;
437}
438
439/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
440static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
441{
442 if (nd->flags & LOOKUP_RCU)
443 return nameidata_drop_rcu(nd);
444 return 0;
445}
446
447/**
448 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
449 * @nd: nameidata pathwalk data to drop
450 * @dentry: dentry to drop
451 * @Returns: 0 on success, -ECHLID on failure
452 *
453 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
454 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
455 * @nd. Must be called from rcu-walk context.
456 */
457static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
458{
459 struct fs_struct *fs = current->fs;
460 struct dentry *parent = nd->path.dentry;
461
462 BUG_ON(!(nd->flags & LOOKUP_RCU));
463 if (nd->root.mnt) {
464 spin_lock(&fs->lock);
465 if (nd->root.mnt != fs->root.mnt ||
466 nd->root.dentry != fs->root.dentry)
467 goto err_root;
468 }
469 spin_lock(&parent->d_lock);
470 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
471 if (!__d_rcu_to_refcount(dentry, nd->seq))
472 goto err;
473 /*
474 * If the sequence check on the child dentry passed, then the child has
475 * not been removed from its parent. This means the parent dentry must
476 * be valid and able to take a reference at this point.
477 */
478 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
479 BUG_ON(!parent->d_count);
480 parent->d_count++;
481 spin_unlock(&dentry->d_lock);
482 spin_unlock(&parent->d_lock);
483 if (nd->root.mnt) {
484 path_get(&nd->root);
485 spin_unlock(&fs->lock);
486 }
487 mntget(nd->path.mnt);
488
489 rcu_read_unlock();
490 br_read_unlock(vfsmount_lock);
491 nd->flags &= ~LOOKUP_RCU;
492 return 0;
493err:
494 spin_unlock(&dentry->d_lock);
495 spin_unlock(&parent->d_lock);
496err_root:
497 if (nd->root.mnt)
498 spin_unlock(&fs->lock);
499 return -ECHILD;
500}
501
502/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
503static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
504{
505 if (nd->flags & LOOKUP_RCU)
506 return nameidata_dentry_drop_rcu(nd, dentry);
507 return 0;
508}
509
510/**
511 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
512 * @nd: nameidata pathwalk data to drop
513 * @Returns: 0 on success, -ECHLID on failure
514 *
515 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
516 * nd->path should be the final element of the lookup, so nd->root is discarded.
517 * Must be called from rcu-walk context.
518 */
519static int nameidata_drop_rcu_last(struct nameidata *nd)
520{
521 struct dentry *dentry = nd->path.dentry;
522
523 BUG_ON(!(nd->flags & LOOKUP_RCU));
524 nd->flags &= ~LOOKUP_RCU;
525 nd->root.mnt = NULL;
526 spin_lock(&dentry->d_lock);
527 if (!__d_rcu_to_refcount(dentry, nd->seq))
528 goto err_unlock;
529 BUG_ON(nd->inode != dentry->d_inode);
530 spin_unlock(&dentry->d_lock);
531
532 mntget(nd->path.mnt);
533
534 rcu_read_unlock();
535 br_read_unlock(vfsmount_lock);
536
537 return 0;
538
539err_unlock:
540 spin_unlock(&dentry->d_lock);
541 rcu_read_unlock();
542 br_read_unlock(vfsmount_lock);
543 return -ECHILD;
544}
545
546/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
547static inline int nameidata_drop_rcu_last_maybe(struct nameidata *nd)
548{
549 if (likely(nd->flags & LOOKUP_RCU))
550 return nameidata_drop_rcu_last(nd);
551 return 0;
552}
553
834f2a4a
TM		 554/**
555 * release_open_intent - free up open intent resources
556 * @nd: pointer to nameidata
557 */
558void release_open_intent(struct nameidata *nd)
559{
0f7fc9e4 560 if (nd->intent.open.file->f_path.dentry == NULL)
834f2a4a
TM		 561 put_filp(nd->intent.open.file);
562 else
563 fput(nd->intent.open.file);
564}
565
bcdc5e01
IK		 566static inline struct dentry *
567do_revalidate(struct dentry *dentry, struct nameidata *nd)
568{
569 int status = dentry->d_op->d_revalidate(dentry, nd);
570 if (unlikely(status <= 0)) {
571 /*
572 * The dentry failed validation.
573 * If d_revalidate returned 0 attempt to invalidate
574 * the dentry otherwise d_revalidate is asking us
575 * to return a fail status.
576 */
577 if (!status) {
578 if (!d_invalidate(dentry)) {
579 dput(dentry);
580 dentry = NULL;
581 }
582 } else {
583 dput(dentry);
584 dentry = ERR_PTR(status);
585 }
586 }
587 return dentry;
588}
589
39159de2
JL		 590/*
591 * force_reval_path - force revalidation of a dentry
592 *
593 * In some situations the path walking code will trust dentries without
594 * revalidating them. This causes problems for filesystems that depend on
595 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
596 * (which indicates that it's possible for the dentry to go stale), force
597 * a d_revalidate call before proceeding.
598 *
599 * Returns 0 if the revalidation was successful. If the revalidation fails,
600 * either return the error returned by d_revalidate or -ESTALE if the
601 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
602 * invalidate the dentry. It's up to the caller to handle putting references
603 * to the path if necessary.
604 */
605static int
606force_reval_path(struct path *path, struct nameidata *nd)
607{
608 int status;
609 struct dentry *dentry = path->dentry;
610
611 /*
612 * only check on filesystems where it's possible for the dentry to
613 * become stale. It's assumed that if this flag is set then the
614 * d_revalidate op will also be defined.
615 */
616 if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))
617 return 0;
618
619 status = dentry->d_op->d_revalidate(dentry, nd);
620 if (status > 0)
621 return 0;
622
623 if (!status) {
624 d_invalidate(dentry);
625 status = -ESTALE;
626 }
627 return status;
628}
629
1da177e4 630/*
b75b5086
AV		 631 * Short-cut version of permission(), for calling on directories
632 * during pathname resolution. Combines parts of permission()
633 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
1da177e4
LT		 634 *
635 * If appropriate, check DAC only. If not appropriate, or
b75b5086 636 * short-cut DAC fails, then call ->permission() to do more
1da177e4
LT		 637 * complete permission check.
638 */
31e6b01f 639static inline int __exec_permission(struct inode *inode, int rcu)
1da177e4 640{
5909ccaa 641 int ret;
1da177e4 642
cb9179ea 643 if (inode->i_op->permission) {
31e6b01f
NP		 644 if (rcu)
645 return -ECHILD;
5909ccaa 646 ret = inode->i_op->permission(inode, MAY_EXEC);
cb9179ea
LT		 647 if (!ret)
648 goto ok;
649 return ret;
650 }
31e6b01f 651 ret = __acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl, rcu);
5909ccaa 652 if (!ret)
1da177e4 653 goto ok;
31e6b01f
NP		 654 if (rcu && ret == -ECHILD)
655 return ret;
1da177e4 656
f1ac9f6b 657 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
1da177e4
LT		 658 goto ok;
659
5909ccaa 660 return ret;
1da177e4 661ok:
31e6b01f
NP		 662 return security_inode_exec_permission(inode, rcu);
663}
664
665static int exec_permission(struct inode *inode)
666{
667 return __exec_permission(inode, 0);
668}
669
670static int exec_permission_rcu(struct inode *inode)
671{
672 return __exec_permission(inode, 1);
1da177e4
LT		 673}
674
2a737871
AV		 675static __always_inline void set_root(struct nameidata *nd)
676{
f7ad3c6b
MS		 677 if (!nd->root.mnt)
678 get_fs_root(current->fs, &nd->root);
2a737871
AV		 679}
680
6de88d72
AV		 681static int link_path_walk(const char *, struct nameidata *);
682
31e6b01f
NP		 683static __always_inline void set_root_rcu(struct nameidata *nd)
684{
685 if (!nd->root.mnt) {
686 struct fs_struct *fs = current->fs;
687 spin_lock(&fs->lock);
688 nd->root = fs->root;
689 spin_unlock(&fs->lock);
690 }
691}
692
f1662356 693static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 694{
31e6b01f
NP		 695 int ret;
696
1da177e4
LT		 697 if (IS_ERR(link))
698 goto fail;
699
700 if (*link == '/') {
2a737871 701 set_root(nd);
1d957f9b 702 path_put(&nd->path);
2a737871
AV		 703 nd->path = nd->root;
704 path_get(&nd->root);
1da177e4 705 }
31e6b01f 706 nd->inode = nd->path.dentry->d_inode;
b4091d5f 707
31e6b01f
NP		 708 ret = link_path_walk(link, nd);
709 return ret;
1da177e4 710fail:
1d957f9b 711 path_put(&nd->path);
1da177e4
LT		 712 return PTR_ERR(link);
713}
714
1d957f9b 715static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 716{
717 dput(path->dentry);
4ac91378 718 if (path->mnt != nd->path.mnt)
051d3812
IK		 719 mntput(path->mnt);
720}
721
722static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
723{
31e6b01f
NP		 724 if (!(nd->flags & LOOKUP_RCU)) {
725 dput(nd->path.dentry);
726 if (nd->path.mnt != path->mnt)
727 mntput(nd->path.mnt);
9a229683 728 }
31e6b01f 729 nd->path.mnt = path->mnt;
4ac91378 730 nd->path.dentry = path->dentry;
051d3812
IK		 731}
732
def4af30
AV		 733static __always_inline int
734__do_follow_link(struct path *path, struct nameidata *nd, void **p)
1da177e4
LT		 735{
736 int error;
cd4e91d3 737 struct dentry *dentry = path->dentry;
1da177e4 738
d671a1cb 739 touch_atime(path->mnt, dentry);
1da177e4 740 nd_set_link(nd, NULL);
cd4e91d3 741
4ac91378 742 if (path->mnt != nd->path.mnt) {
051d3812 743 path_to_nameidata(path, nd);
31e6b01f 744 nd->inode = nd->path.dentry->d_inode;
051d3812
IK		 745 dget(dentry);
746 }
747 mntget(path->mnt);
31e6b01f 748
86acdca1 749 nd->last_type = LAST_BIND;
def4af30
AV		 750 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
751 error = PTR_ERR(*p);
752 if (!IS_ERR(*p)) {
1da177e4 753 char *s = nd_get_link(nd);
cc314eef 754 error = 0;
1da177e4
LT		 755 if (s)
756 error = __vfs_follow_link(nd, s);
39159de2
JL		 757 else if (nd->last_type == LAST_BIND) {
758 error = force_reval_path(&nd->path, nd);
759 if (error)
760 path_put(&nd->path);
761 }
1da177e4 762 }
1da177e4
LT		 763 return error;
764}
765
766/*
767 * This limits recursive symlink follows to 8, while
768 * limiting consecutive symlinks to 40.
769 *
770 * Without that kind of total limit, nasty chains of consecutive
771 * symlinks can cause almost arbitrarily long lookups.
772 */
90ebe565 773static inline int do_follow_link(struct path *path, struct nameidata *nd)
1da177e4 774{
def4af30 775 void *cookie;
1da177e4
LT		 776 int err = -ELOOP;
777 if (current->link_count >= MAX_NESTED_LINKS)
778 goto loop;
779 if (current->total_link_count >= 40)
780 goto loop;
781 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
782 cond_resched();
90ebe565 783 err = security_inode_follow_link(path->dentry, nd);
1da177e4
LT		 784 if (err)
785 goto loop;
786 current->link_count++;
787 current->total_link_count++;
788 nd->depth++;
def4af30
AV		 789 err = __do_follow_link(path, nd, &cookie);
790 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
791 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
258fa999 792 path_put(path);
839d9f93
AV		 793 current->link_count--;
794 nd->depth--;
1da177e4
LT		 795 return err;
796loop:
1d957f9b
JB		 797 path_put_conditional(path, nd);
798 path_put(&nd->path);
1da177e4
LT		 799 return err;
800}
801
31e6b01f
NP		 802static int follow_up_rcu(struct path *path)
803{
804 struct vfsmount *parent;
805 struct dentry *mountpoint;
806
807 parent = path->mnt->mnt_parent;
808 if (parent == path->mnt)
809 return 0;
810 mountpoint = path->mnt->mnt_mountpoint;
811 path->dentry = mountpoint;
812 path->mnt = parent;
813 return 1;
814}
815
bab77ebf 816int follow_up(struct path *path)
1da177e4
LT		 817{
818 struct vfsmount *parent;
819 struct dentry *mountpoint;
99b7db7b
NP		 820
821 br_read_lock(vfsmount_lock);
bab77ebf
AV		 822 parent = path->mnt->mnt_parent;
823 if (parent == path->mnt) {
99b7db7b 824 br_read_unlock(vfsmount_lock);
1da177e4
LT		 825 return 0;
826 }
827 mntget(parent);
bab77ebf 828 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 829 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 830 dput(path->dentry);
831 path->dentry = mountpoint;
832 mntput(path->mnt);
833 path->mnt = parent;
1da177e4
LT		 834 return 1;
835}
836
b5c84bf6
NP		 837/*
838 * serialization is taken care of in namespace.c
1da177e4 839 */
31e6b01f
NP		 840static void __follow_mount_rcu(struct nameidata *nd, struct path *path,
841 struct inode **inode)
842{
843 while (d_mountpoint(path->dentry)) {
844 struct vfsmount *mounted;
845 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
846 if (!mounted)
847 return;
848 path->mnt = mounted;
849 path->dentry = mounted->mnt_root;
850 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
851 *inode = path->dentry->d_inode;
852 }
853}
854
463ffb2e
AV		 855static int __follow_mount(struct path *path)
856{
857 int res = 0;
858 while (d_mountpoint(path->dentry)) {
1c755af4 859 struct vfsmount *mounted = lookup_mnt(path);
463ffb2e
AV		 860 if (!mounted)
861 break;
862 dput(path->dentry);
863 if (res)
864 mntput(path->mnt);
865 path->mnt = mounted;
866 path->dentry = dget(mounted->mnt_root);
867 res = 1;
868 }
869 return res;
870}
871
79ed0226 872static void follow_mount(struct path *path)
1da177e4 873{
79ed0226 874 while (d_mountpoint(path->dentry)) {
1c755af4 875 struct vfsmount *mounted = lookup_mnt(path);
1da177e4
LT		 876 if (!mounted)
877 break;
79ed0226
AV		 878 dput(path->dentry);
879 mntput(path->mnt);
880 path->mnt = mounted;
881 path->dentry = dget(mounted->mnt_root);
1da177e4 882 }
1da177e4
LT		 883}
884
9393bd07 885int follow_down(struct path *path)
1da177e4
LT		 886{
887 struct vfsmount *mounted;
888
1c755af4 889 mounted = lookup_mnt(path);
1da177e4 890 if (mounted) {
9393bd07
AV		 891 dput(path->dentry);
892 mntput(path->mnt);
893 path->mnt = mounted;
894 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 895 return 1;
896 }
897 return 0;
898}
899
31e6b01f
NP		 900static int follow_dotdot_rcu(struct nameidata *nd)
901{
902 struct inode *inode = nd->inode;
903
904 set_root_rcu(nd);
905
906 while(1) {
907 if (nd->path.dentry == nd->root.dentry &&
908 nd->path.mnt == nd->root.mnt) {
909 break;
910 }
911 if (nd->path.dentry != nd->path.mnt->mnt_root) {
912 struct dentry *old = nd->path.dentry;
913 struct dentry *parent = old->d_parent;
914 unsigned seq;
915
916 seq = read_seqcount_begin(&parent->d_seq);
917 if (read_seqcount_retry(&old->d_seq, nd->seq))
918 return -ECHILD;
919 inode = parent->d_inode;
920 nd->path.dentry = parent;
921 nd->seq = seq;
922 break;
923 }
924 if (!follow_up_rcu(&nd->path))
925 break;
926 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
927 inode = nd->path.dentry->d_inode;
928 }
929 __follow_mount_rcu(nd, &nd->path, &inode);
930 nd->inode = inode;
931
932 return 0;
933}
934
935static void follow_dotdot(struct nameidata *nd)
1da177e4 936{
2a737871 937 set_root(nd);
e518ddb7 938
1da177e4 939 while(1) {
4ac91378 940 struct dentry *old = nd->path.dentry;
1da177e4 941
2a737871
AV		 942 if (nd->path.dentry == nd->root.dentry &&
943 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 944 break;
945 }
4ac91378 946 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 947 /* rare case of legitimate dget_parent()... */
948 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 949 dput(old);
950 break;
951 }
3088dd70 952 if (!follow_up(&nd->path))
1da177e4 953 break;
1da177e4 954 }
79ed0226 955 follow_mount(&nd->path);
31e6b01f 956 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 957}
958
baa03890
NP		 959/*
960 * Allocate a dentry with name and parent, and perform a parent
961 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
962 * on error. parent->d_inode->i_mutex must be held. d_lookup must
963 * have verified that no child exists while under i_mutex.
964 */
965static struct dentry *d_alloc_and_lookup(struct dentry *parent,
966 struct qstr *name, struct nameidata *nd)
967{
968 struct inode *inode = parent->d_inode;
969 struct dentry *dentry;
970 struct dentry *old;
971
972 /* Don't create child dentry for a dead directory. */
973 if (unlikely(IS_DEADDIR(inode)))
974 return ERR_PTR(-ENOENT);
975
976 dentry = d_alloc(parent, name);
977 if (unlikely(!dentry))
978 return ERR_PTR(-ENOMEM);
979
980 old = inode->i_op->lookup(inode, dentry, nd);
981 if (unlikely(old)) {
982 dput(dentry);
983 dentry = old;
984 }
985 return dentry;
986}
987
1da177e4
LT		 988/*
989 * It's more convoluted than I'd like it to be, but... it's still fairly
990 * small and for now I'd prefer to have fast path as straight as possible.
991 * It _is_ time-critical.
992 */
993static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 994 struct path *path, struct inode **inode)
1da177e4 995{
4ac91378 996 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 997 struct dentry *dentry, *parent = nd->path.dentry;
6e6b1bd1 998 struct inode *dir;
3cac260a
AV		 999 /*
1000 * See if the low-level filesystem might want
1001 * to use its own hash..
1002 */
31e6b01f
NP		 1003 if (parent->d_op && parent->d_op->d_hash) {
1004 int err = parent->d_op->d_hash(parent, nd->inode, name);
3cac260a
AV		 1005 if (err < 0)
1006 return err;
1007 }
1da177e4 1008
b04f784e
NP		 1009 /*
1010 * Rename seqlock is not required here because in the off chance
1011 * of a false negative due to a concurrent rename, we're going to
1012 * do the non-racy lookup, below.
1013 */
31e6b01f
NP		 1014 if (nd->flags & LOOKUP_RCU) {
1015 unsigned seq;
1016
1017 *inode = nd->inode;
1018 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1019 if (!dentry) {
1020 if (nameidata_drop_rcu(nd))
1021 return -ECHILD;
1022 goto need_lookup;
1023 }
1024 /* Memory barrier in read_seqcount_begin of child is enough */
1025 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1026 return -ECHILD;
1027
1028 nd->seq = seq;
1029 if (dentry->d_op && dentry->d_op->d_revalidate) {
1030 /* We commonly drop rcu-walk here */
1031 if (nameidata_dentry_drop_rcu(nd, dentry))
1032 return -ECHILD;
1033 goto need_revalidate;
1034 }
1035 path->mnt = mnt;
1036 path->dentry = dentry;
1037 __follow_mount_rcu(nd, path, inode);
1038 } else {
1039 dentry = __d_lookup(parent, name);
1040 if (!dentry)
1041 goto need_lookup;
2e2e88ea 1042found:
31e6b01f
NP		 1043 if (dentry->d_op && dentry->d_op->d_revalidate)
1044 goto need_revalidate;
1da177e4 1045done:
31e6b01f
NP		 1046 path->mnt = mnt;
1047 path->dentry = dentry;
1048 __follow_mount(path);
1049 *inode = path->dentry->d_inode;
1050 }
1da177e4
LT		 1051 return 0;
1052
1053need_lookup:
6e6b1bd1 1054 dir = parent->d_inode;
31e6b01f 1055 BUG_ON(nd->inode != dir);
6e6b1bd1
AV		 1056
1057 mutex_lock(&dir->i_mutex);
1058 /*
1059 * First re-do the cached lookup just in case it was created
b04f784e
NP		 1060 * while we waited for the directory semaphore, or the first
1061 * lookup failed due to an unrelated rename.
6e6b1bd1 1062 *
b04f784e
NP		 1063 * This could use version numbering or similar to avoid unnecessary
1064 * cache lookups, but then we'd have to do the first lookup in the
1065 * non-racy way. However in the common case here, everything should
1066 * be hot in cache, so would it be a big win?
6e6b1bd1
AV		 1067 */
1068 dentry = d_lookup(parent, name);
baa03890
NP		 1069 if (likely(!dentry)) {
1070 dentry = d_alloc_and_lookup(parent, name, nd);
6e6b1bd1
AV		 1071 mutex_unlock(&dir->i_mutex);
1072 if (IS_ERR(dentry))
1073 goto fail;
1074 goto done;
1075 }
6e6b1bd1
AV		 1076 /*
1077 * Uhhuh! Nasty case: the cache was re-populated while
1078 * we waited on the semaphore. Need to revalidate.
1079 */
1080 mutex_unlock(&dir->i_mutex);
2e2e88ea 1081 goto found;
1da177e4
LT		 1082
1083need_revalidate:
bcdc5e01
IK		 1084 dentry = do_revalidate(dentry, nd);
1085 if (!dentry)
1086 goto need_lookup;
1087 if (IS_ERR(dentry))
1088 goto fail;
1089 goto done;
1da177e4
LT		 1090
1091fail:
1092 return PTR_ERR(dentry);
1093}
1094
ac278a9c
AV		 1095/*
1096 * This is a temporary kludge to deal with "automount" symlinks; proper
1097 * solution is to trigger them on follow_mount(), so that do_lookup()
1098 * would DTRT. To be killed before 2.6.34-final.
1099 */
1100static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
1101{
1102 return inode && unlikely(inode->i_op->follow_link) &&
1103 ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
1104}
1105
1da177e4
LT		 1106/*
1107 * Name resolution.
ea3834d9
PM		 1108 * This is the basic name resolution function, turning a pathname into
1109 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1110 *
ea3834d9
PM		 1111 * Returns 0 and nd will have valid dentry and mnt on success.
1112 * Returns error and drops reference to input namei data on failure.
1da177e4 1113 */
6de88d72 1114static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1115{
1116 struct path next;
1da177e4
LT		 1117 int err;
1118 unsigned int lookup_flags = nd->flags;
1119
1120 while (*name=='/')
1121 name++;
1122 if (!*name)
1123 goto return_reval;
1124
1da177e4 1125 if (nd->depth)
f55eab82 1126 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 1127
1128 /* At this point we know we have a real path component. */
1129 for(;;) {
31e6b01f 1130 struct inode *inode;
1da177e4
LT		 1131 unsigned long hash;
1132 struct qstr this;
1133 unsigned int c;
1134
cdce5d6b 1135 nd->flags |= LOOKUP_CONTINUE;
31e6b01f
NP		 1136 if (nd->flags & LOOKUP_RCU) {
1137 err = exec_permission_rcu(nd->inode);
1138 if (err == -ECHILD) {
1139 if (nameidata_drop_rcu(nd))
1140 return -ECHILD;
1141 goto exec_again;
1142 }
1143 } else {
1144exec_again:
1145 err = exec_permission(nd->inode);
1146 }
1da177e4
LT		 1147 if (err)
1148 break;
1149
1150 this.name = name;
1151 c = *(const unsigned char *)name;
1152
1153 hash = init_name_hash();
1154 do {
1155 name++;
1156 hash = partial_name_hash(c, hash);
1157 c = *(const unsigned char *)name;
1158 } while (c && (c != '/'));
1159 this.len = name - (const char *) this.name;
1160 this.hash = end_name_hash(hash);
1161
1162 /* remove trailing slashes? */
1163 if (!c)
1164 goto last_component;
1165 while (*++name == '/');
1166 if (!*name)
1167 goto last_with_slashes;
1168
1169 /*
1170 * "." and ".." are special - ".." especially so because it has
1171 * to be able to know about the current root directory and
1172 * parent relationships.
1173 */
1174 if (this.name[0] == '.') switch (this.len) {
1175 default:
1176 break;
31e6b01f 1177 case 2:
1da177e4
LT		 1178 if (this.name[1] != '.')
1179 break;
31e6b01f
NP		 1180 if (nd->flags & LOOKUP_RCU) {
1181 if (follow_dotdot_rcu(nd))
1182 return -ECHILD;
1183 } else
1184 follow_dotdot(nd);
1da177e4
LT		 1185 /* fallthrough */
1186 case 1:
1187 continue;
1188 }
1da177e4 1189 /* This does the actual lookups.. */
31e6b01f 1190 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1191 if (err)
1192 break;
1da177e4 1193 err = -ENOENT;
1da177e4
LT		 1194 if (!inode)
1195 goto out_dput;
1da177e4
LT		 1196
1197 if (inode->i_op->follow_link) {
31e6b01f
NP		 1198 /* We commonly drop rcu-walk here */
1199 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1200 return -ECHILD;
1201 BUG_ON(inode != next.dentry->d_inode);
90ebe565 1202 err = do_follow_link(&next, nd);
1da177e4
LT		 1203 if (err)
1204 goto return_err;
31e6b01f 1205 nd->inode = nd->path.dentry->d_inode;
1da177e4 1206 err = -ENOENT;
31e6b01f 1207 if (!nd->inode)
1da177e4 1208 break;
31e6b01f 1209 } else {
09dd17d3 1210 path_to_nameidata(&next, nd);
31e6b01f
NP		 1211 nd->inode = inode;
1212 }
1da177e4 1213 err = -ENOTDIR;
31e6b01f 1214 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1215 break;
1216 continue;
1217 /* here ends the main loop */
1218
1219last_with_slashes:
1220 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1221last_component:
f55eab82
TM		 1222 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1223 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 1224 if (lookup_flags & LOOKUP_PARENT)
1225 goto lookup_parent;
1226 if (this.name[0] == '.') switch (this.len) {
1227 default:
1228 break;
31e6b01f 1229 case 2:
1da177e4
LT		 1230 if (this.name[1] != '.')
1231 break;
31e6b01f
NP		 1232 if (nd->flags & LOOKUP_RCU) {
1233 if (follow_dotdot_rcu(nd))
1234 return -ECHILD;
1235 } else
1236 follow_dotdot(nd);
1da177e4
LT		 1237 /* fallthrough */
1238 case 1:
1239 goto return_reval;
1240 }
31e6b01f 1241 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1242 if (err)
1243 break;
ac278a9c 1244 if (follow_on_final(inode, lookup_flags)) {
31e6b01f
NP		 1245 if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
1246 return -ECHILD;
1247 BUG_ON(inode != next.dentry->d_inode);
90ebe565 1248 err = do_follow_link(&next, nd);
1da177e4
LT		 1249 if (err)
1250 goto return_err;
31e6b01f
NP		 1251 nd->inode = nd->path.dentry->d_inode;
1252 } else {
09dd17d3 1253 path_to_nameidata(&next, nd);
31e6b01f
NP		 1254 nd->inode = inode;
1255 }
1da177e4 1256 err = -ENOENT;
31e6b01f 1257 if (!nd->inode)
1da177e4
LT		 1258 break;
1259 if (lookup_flags & LOOKUP_DIRECTORY) {
1260 err = -ENOTDIR;
31e6b01f 1261 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1262 break;
1263 }
1264 goto return_base;
1265lookup_parent:
1266 nd->last = this;
1267 nd->last_type = LAST_NORM;
1268 if (this.name[0] != '.')
1269 goto return_base;
1270 if (this.len == 1)
1271 nd->last_type = LAST_DOT;
1272 else if (this.len == 2 && this.name[1] == '.')
1273 nd->last_type = LAST_DOTDOT;
1274 else
1275 goto return_base;
1276return_reval:
1277 /*
1278 * We bypassed the ordinary revalidation routines.
1279 * We may need to check the cached dentry for staleness.
1280 */
4ac91378
JB		 1281 if (nd->path.dentry && nd->path.dentry->d_sb &&
1282 (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
31e6b01f
NP		 1283 if (nameidata_drop_rcu_maybe(nd))
1284 return -ECHILD;
1da177e4
LT		 1285 err = -ESTALE;
1286 /* Note: we do not d_invalidate() */
4ac91378
JB		 1287 if (!nd->path.dentry->d_op->d_revalidate(
1288 nd->path.dentry, nd))
1da177e4
LT		 1289 break;
1290 }
1291return_base:
31e6b01f
NP		 1292 if (nameidata_drop_rcu_last_maybe(nd))
1293 return -ECHILD;
1da177e4
LT		 1294 return 0;
1295out_dput:
31e6b01f
NP		 1296 if (!(nd->flags & LOOKUP_RCU))
1297 path_put_conditional(&next, nd);
1da177e4
LT		 1298 break;
1299 }
31e6b01f
NP		 1300 if (!(nd->flags & LOOKUP_RCU))
1301 path_put(&nd->path);
1da177e4
LT		 1302return_err:
1303 return err;
1304}
1305
31e6b01f
NP		 1306static inline int path_walk_rcu(const char *name, struct nameidata *nd)
1307{
1308 current->total_link_count = 0;
1309
1310 return link_path_walk(name, nd);
1311}
1312
1313static inline int path_walk_simple(const char *name, struct nameidata *nd)
1314{
1315 current->total_link_count = 0;
1316
1317 return link_path_walk(name, nd);
1318}
1319
fc9b52cd 1320static int path_walk(const char *name, struct nameidata *nd)
1da177e4 1321{
6de88d72
AV		 1322 struct path save = nd->path;
1323 int result;
1324
1da177e4 1325 current->total_link_count = 0;
6de88d72
AV		 1326
1327 /* make sure the stuff we saved doesn't go away */
1328 path_get(&save);
1329
1330 result = link_path_walk(name, nd);
1331 if (result == -ESTALE) {
1332 /* nd->path had been dropped */
1333 current->total_link_count = 0;
1334 nd->path = save;
1335 path_get(&nd->path);
1336 nd->flags |= LOOKUP_REVAL;
1337 result = link_path_walk(name, nd);
1338 }
1339
1340 path_put(&save);
1341
1342 return result;
1da177e4
LT		 1343}
1344
31e6b01f
NP		 1345static void path_finish_rcu(struct nameidata *nd)
1346{
1347 if (nd->flags & LOOKUP_RCU) {
1348 /* RCU dangling. Cancel it. */
1349 nd->flags &= ~LOOKUP_RCU;
1350 nd->root.mnt = NULL;
1351 rcu_read_unlock();
1352 br_read_unlock(vfsmount_lock);
1353 }
1354 if (nd->file)
1355 fput(nd->file);
1356}
1357
1358static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1359{
1360 int retval = 0;
1361 int fput_needed;
1362 struct file *file;
1363
1364 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1365 nd->flags = flags | LOOKUP_RCU;
1366 nd->depth = 0;
1367 nd->root.mnt = NULL;
1368 nd->file = NULL;
1369
1370 if (*name=='/') {
1371 struct fs_struct *fs = current->fs;
1372
1373 br_read_lock(vfsmount_lock);
1374 rcu_read_lock();
1375
1376 spin_lock(&fs->lock);
1377 nd->root = fs->root;
1378 nd->path = nd->root;
1379 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1380 spin_unlock(&fs->lock);
1381
1382 } else if (dfd == AT_FDCWD) {
1383 struct fs_struct *fs = current->fs;
1384
1385 br_read_lock(vfsmount_lock);
1386 rcu_read_lock();
1387
1388 spin_lock(&fs->lock);
1389 nd->path = fs->pwd;
1390 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1391 spin_unlock(&fs->lock);
1392 } else {
1393 struct dentry *dentry;
1394
1395 file = fget_light(dfd, &fput_needed);
1396 retval = -EBADF;
1397 if (!file)
1398 goto out_fail;
1399
1400 dentry = file->f_path.dentry;
1401
1402 retval = -ENOTDIR;
1403 if (!S_ISDIR(dentry->d_inode->i_mode))
1404 goto fput_fail;
1405
1406 retval = file_permission(file, MAY_EXEC);
1407 if (retval)
1408 goto fput_fail;
1409
1410 nd->path = file->f_path;
1411 if (fput_needed)
1412 nd->file = file;
1413
1414 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1415 br_read_lock(vfsmount_lock);
1416 rcu_read_lock();
1417 }
1418 nd->inode = nd->path.dentry->d_inode;
1419 return 0;
1420
1421fput_fail:
1422 fput_light(file, fput_needed);
1423out_fail:
1424 return retval;
1425}
1426
9b4a9b14 1427static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
1da177e4 1428{
ea3834d9 1429 int retval = 0;
170aa3d0
UD		 1430 int fput_needed;
1431 struct file *file;
1da177e4
LT		 1432
1433 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1434 nd->flags = flags;
1435 nd->depth = 0;
2a737871 1436 nd->root.mnt = NULL;
1da177e4 1437
1da177e4 1438 if (*name=='/') {
2a737871
AV		 1439 set_root(nd);
1440 nd->path = nd->root;
1441 path_get(&nd->root);
5590ff0d 1442 } else if (dfd == AT_FDCWD) {
f7ad3c6b 1443 get_fs_pwd(current->fs, &nd->path);
5590ff0d 1444 } else {
5590ff0d
UD		 1445 struct dentry *dentry;
1446
1447 file = fget_light(dfd, &fput_needed);
170aa3d0
UD		 1448 retval = -EBADF;
1449 if (!file)
6d09bb62 1450 goto out_fail;
5590ff0d 1451
0f7fc9e4 1452 dentry = file->f_path.dentry;
5590ff0d 1453
170aa3d0
UD		 1454 retval = -ENOTDIR;
1455 if (!S_ISDIR(dentry->d_inode->i_mode))
6d09bb62 1456 goto fput_fail;
5590ff0d
UD		 1457
1458 retval = file_permission(file, MAY_EXEC);
170aa3d0 1459 if (retval)
6d09bb62 1460 goto fput_fail;
5590ff0d 1461
5dd784d0
JB		 1462 nd->path = file->f_path;
1463 path_get(&file->f_path);
5590ff0d
UD		 1464
1465 fput_light(file, fput_needed);
1da177e4 1466 }
31e6b01f 1467 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1468 return 0;
2dfdd266 1469
9b4a9b14
AV		 1470fput_fail:
1471 fput_light(file, fput_needed);
1472out_fail:
1473 return retval;
1474}
1475
1476/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1477static int do_path_lookup(int dfd, const char *name,
1478 unsigned int flags, struct nameidata *nd)
1479{
31e6b01f
NP		 1480 int retval;
1481
1482 /*
1483 * Path walking is largely split up into 2 different synchronisation
1484 * schemes, rcu-walk and ref-walk (explained in
1485 * Documentation/filesystems/path-lookup.txt). These share much of the
1486 * path walk code, but some things particularly setup, cleanup, and
1487 * following mounts are sufficiently divergent that functions are
1488 * duplicated. Typically there is a function foo(), and its RCU
1489 * analogue, foo_rcu().
1490 *
1491 * -ECHILD is the error number of choice (just to avoid clashes) that
1492 * is returned if some aspect of an rcu-walk fails. Such an error must
1493 * be handled by restarting a traditional ref-walk (which will always
1494 * be able to complete).
1495 */
1496 retval = path_init_rcu(dfd, name, flags, nd);
1497 if (unlikely(retval))
1498 return retval;
1499 retval = path_walk_rcu(name, nd);
1500 path_finish_rcu(nd);
2a737871
AV		 1501 if (nd->root.mnt) {
1502 path_put(&nd->root);
1503 nd->root.mnt = NULL;
1504 }
31e6b01f
NP		 1505
1506 if (unlikely(retval == -ECHILD || retval == -ESTALE)) {
1507 /* slower, locked walk */
1508 if (retval == -ESTALE)
1509 flags |= LOOKUP_REVAL;
1510 retval = path_init(dfd, name, flags, nd);
1511 if (unlikely(retval))
1512 return retval;
1513 retval = path_walk(name, nd);
1514 if (nd->root.mnt) {
1515 path_put(&nd->root);
1516 nd->root.mnt = NULL;
1517 }
1518 }
1519
1520 if (likely(!retval)) {
1521 if (unlikely(!audit_dummy_context())) {
1522 if (nd->path.dentry && nd->inode)
1523 audit_inode(name, nd->path.dentry);
1524 }
1525 }
1526
170aa3d0 1527 return retval;
1da177e4
LT		 1528}
1529
fc9b52cd 1530int path_lookup(const char *name, unsigned int flags,
5590ff0d
UD		 1531 struct nameidata *nd)
1532{
1533 return do_path_lookup(AT_FDCWD, name, flags, nd);
1534}
1535
d1811465
AV		 1536int kern_path(const char *name, unsigned int flags, struct path *path)
1537{
1538 struct nameidata nd;
1539 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1540 if (!res)
1541 *path = nd.path;
1542 return res;
1543}
1544
16f18200
JJS		 1545/**
1546 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1547 * @dentry: pointer to dentry of the base directory
1548 * @mnt: pointer to vfs mount of the base directory
1549 * @name: pointer to file name
1550 * @flags: lookup flags
1551 * @nd: pointer to nameidata
1552 */
1553int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1554 const char *name, unsigned int flags,
1555 struct nameidata *nd)
1556{
1557 int retval;
1558
1559 /* same as do_path_lookup */
1560 nd->last_type = LAST_ROOT;
1561 nd->flags = flags;
1562 nd->depth = 0;
1563
c8e7f449
JB		 1564 nd->path.dentry = dentry;
1565 nd->path.mnt = mnt;
1566 path_get(&nd->path);
5b857119
AV		 1567 nd->root = nd->path;
1568 path_get(&nd->root);
31e6b01f 1569 nd->inode = nd->path.dentry->d_inode;
16f18200
JJS		 1570
1571 retval = path_walk(name, nd);
4ac91378 1572 if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
31e6b01f 1573 nd->inode))
4ac91378 1574 audit_inode(name, nd->path.dentry);
16f18200 1575
5b857119
AV		 1576 path_put(&nd->root);
1577 nd->root.mnt = NULL;
16f18200 1578
2a737871 1579 return retval;
16f18200
JJS		 1580}
1581
eead1911
CH		 1582static struct dentry *__lookup_hash(struct qstr *name,
1583 struct dentry *base, struct nameidata *nd)
1da177e4 1584{
81fca444 1585 struct inode *inode = base->d_inode;
057f6c01 1586 struct dentry *dentry;
1da177e4
LT		 1587 int err;
1588
81fca444
CH		 1589 err = exec_permission(inode);
1590 if (err)
1591 return ERR_PTR(err);
1da177e4
LT		 1592
1593 /*
1594 * See if the low-level filesystem might want
1595 * to use its own hash..
1596 */
1597 if (base->d_op && base->d_op->d_hash) {
b1e6a015 1598 err = base->d_op->d_hash(base, inode, name);
1da177e4
LT		 1599 dentry = ERR_PTR(err);
1600 if (err < 0)
1601 goto out;
1602 }
1603
b04f784e
NP		 1604 /*
1605 * Don't bother with __d_lookup: callers are for creat as
1606 * well as unlink, so a lot of the time it would cost
1607 * a double lookup.
6e6b1bd1 1608 */
b04f784e 1609 dentry = d_lookup(base, name);
6e6b1bd1
AV		 1610
1611 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
1612 dentry = do_revalidate(dentry, nd);
1613
baa03890
NP		 1614 if (!dentry)
1615 dentry = d_alloc_and_lookup(base, name, nd);
1da177e4
LT		 1616out:
1617 return dentry;
1618}
1619
057f6c01
JM		 1620/*
1621 * Restricted form of lookup. Doesn't follow links, single-component only,
1622 * needs parent already locked. Doesn't follow mounts.
1623 * SMP-safe.
1624 */
eead1911 1625static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1626{
4ac91378 1627 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1628}
1629
eead1911
CH		 1630static int __lookup_one_len(const char *name, struct qstr *this,
1631 struct dentry *base, int len)
1da177e4
LT		 1632{
1633 unsigned long hash;
1da177e4
LT		 1634 unsigned int c;
1635
057f6c01
JM		 1636 this->name = name;
1637 this->len = len;
1da177e4 1638 if (!len)
057f6c01 1639 return -EACCES;
1da177e4
LT		 1640
1641 hash = init_name_hash();
1642 while (len--) {
1643 c = *(const unsigned char *)name++;
1644 if (c == '/' || c == '\0')
057f6c01 1645 return -EACCES;
1da177e4
LT		 1646 hash = partial_name_hash(c, hash);
1647 }
057f6c01
JM		 1648 this->hash = end_name_hash(hash);
1649 return 0;
1650}
1da177e4 1651
eead1911 1652/**
a6b91919 1653 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1654 * @name: pathname component to lookup
1655 * @base: base directory to lookup from
1656 * @len: maximum length @len should be interpreted to
1657 *
a6b91919
RD		 1658 * Note that this routine is purely a helper for filesystem usage and should
1659 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1660 * nameidata argument is passed to the filesystem methods and a filesystem
1661 * using this helper needs to be prepared for that.
1662 */
057f6c01
JM		 1663struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1664{
1665 int err;
1666 struct qstr this;
1667
2f9092e1
DW		 1668 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1669
057f6c01 1670 err = __lookup_one_len(name, &this, base, len);
eead1911
CH		 1671 if (err)
1672 return ERR_PTR(err);
1673
49705b77 1674 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1675}
1676
2d8f3038
AV		 1677int user_path_at(int dfd, const char __user *name, unsigned flags,
1678 struct path *path)
1da177e4 1679{
2d8f3038 1680 struct nameidata nd;
1da177e4
LT		 1681 char *tmp = getname(name);
1682 int err = PTR_ERR(tmp);
1da177e4 1683 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1684
1685 BUG_ON(flags & LOOKUP_PARENT);
1686
1687 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1688 putname(tmp);
2d8f3038
AV		 1689 if (!err)
1690 *path = nd.path;
1da177e4
LT		 1691 }
1692 return err;
1693}
1694
2ad94ae6
AV		 1695static int user_path_parent(int dfd, const char __user *path,
1696 struct nameidata *nd, char **name)
1697{
1698 char *s = getname(path);
1699 int error;
1700
1701 if (IS_ERR(s))
1702 return PTR_ERR(s);
1703
1704 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1705 if (error)
1706 putname(s);
1707 else
1708 *name = s;
1709
1710 return error;
1711}
1712
1da177e4
LT		 1713/*
1714 * It's inline, so penalty for filesystems that don't use sticky bit is
1715 * minimal.
1716 */
1717static inline int check_sticky(struct inode *dir, struct inode *inode)
1718{
da9592ed
DH		 1719 uid_t fsuid = current_fsuid();
1720
1da177e4
LT		 1721 if (!(dir->i_mode & S_ISVTX))
1722 return 0;
da9592ed 1723 if (inode->i_uid == fsuid)
1da177e4 1724 return 0;
da9592ed 1725 if (dir->i_uid == fsuid)
1da177e4
LT		 1726 return 0;
1727 return !capable(CAP_FOWNER);
1728}
1729
1730/*
1731 * Check whether we can remove a link victim from directory dir, check
1732 * whether the type of victim is right.
1733 * 1. We can't do it if dir is read-only (done in permission())
1734 * 2. We should have write and exec permissions on dir
1735 * 3. We can't remove anything from append-only dir
1736 * 4. We can't do anything with immutable dir (done in permission())
1737 * 5. If the sticky bit on dir is set we should either
1738 * a. be owner of dir, or
1739 * b. be owner of victim, or
1740 * c. have CAP_FOWNER capability
1741 * 6. If the victim is append-only or immutable we can't do antyhing with
1742 * links pointing to it.
1743 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1744 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1745 * 9. We can't remove a root or mountpoint.
1746 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1747 * nfs_async_unlink().
1748 */
858119e1 1749static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1750{
1751 int error;
1752
1753 if (!victim->d_inode)
1754 return -ENOENT;
1755
1756 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1757 audit_inode_child(victim, dir);
1da177e4 1758
f419a2e3 1759 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1760 if (error)
1761 return error;
1762 if (IS_APPEND(dir))
1763 return -EPERM;
1764 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1765 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1766 return -EPERM;
1767 if (isdir) {
1768 if (!S_ISDIR(victim->d_inode->i_mode))
1769 return -ENOTDIR;
1770 if (IS_ROOT(victim))
1771 return -EBUSY;
1772 } else if (S_ISDIR(victim->d_inode->i_mode))
1773 return -EISDIR;
1774 if (IS_DEADDIR(dir))
1775 return -ENOENT;
1776 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1777 return -EBUSY;
1778 return 0;
1779}
1780
1781/* Check whether we can create an object with dentry child in directory
1782 * dir.
1783 * 1. We can't do it if child already exists (open has special treatment for
1784 * this case, but since we are inlined it's OK)
1785 * 2. We can't do it if dir is read-only (done in permission())
1786 * 3. We should have write and exec permissions on dir
1787 * 4. We can't do it if dir is immutable (done in permission())
1788 */
a95164d9 1789static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1790{
1791 if (child->d_inode)
1792 return -EEXIST;
1793 if (IS_DEADDIR(dir))
1794 return -ENOENT;
f419a2e3 1795 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1796}
1797
1da177e4
LT		 1798/*
1799 * p1 and p2 should be directories on the same fs.
1800 */
1801struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1802{
1803 struct dentry *p;
1804
1805 if (p1 == p2) {
f2eace23 1806 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1807 return NULL;
1808 }
1809
a11f3a05 1810 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1811
e2761a11
OH		 1812 p = d_ancestor(p2, p1);
1813 if (p) {
1814 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1815 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1816 return p;
1da177e4
LT		 1817 }
1818
e2761a11
OH		 1819 p = d_ancestor(p1, p2);
1820 if (p) {
1821 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1822 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1823 return p;
1da177e4
LT		 1824 }
1825
f2eace23
IM		 1826 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1827 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1828 return NULL;
1829}
1830
1831void unlock_rename(struct dentry *p1, struct dentry *p2)
1832{
1b1dcc1b 1833 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1834 if (p1 != p2) {
1b1dcc1b 1835 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1836 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1837 }
1838}
1839
1840int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1841 struct nameidata *nd)
1842{
a95164d9 1843 int error = may_create(dir, dentry);
1da177e4
LT		 1844
1845 if (error)
1846 return error;
1847
acfa4380 1848 if (!dir->i_op->create)
1da177e4
LT		 1849 return -EACCES; /* shouldn't it be ENOSYS? */
1850 mode &= S_IALLUGO;
1851 mode |= S_IFREG;
1852 error = security_inode_create(dir, dentry, mode);
1853 if (error)
1854 return error;
1da177e4 1855 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1856 if (!error)
f38aa942 1857 fsnotify_create(dir, dentry);
1da177e4
LT		 1858 return error;
1859}
1860
3fb64190 1861int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1862{
3fb64190 1863 struct dentry *dentry = path->dentry;
1da177e4
LT		 1864 struct inode *inode = dentry->d_inode;
1865 int error;
1866
1867 if (!inode)
1868 return -ENOENT;
1869
c8fe8f30
CH		 1870 switch (inode->i_mode & S_IFMT) {
1871 case S_IFLNK:
1da177e4 1872 return -ELOOP;
c8fe8f30
CH		 1873 case S_IFDIR:
1874 if (acc_mode & MAY_WRITE)
1875 return -EISDIR;
1876 break;
1877 case S_IFBLK:
1878 case S_IFCHR:
3fb64190 1879 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 1880 return -EACCES;
c8fe8f30
CH		 1881 /*FALLTHRU*/
1882 case S_IFIFO:
1883 case S_IFSOCK:
1da177e4 1884 flag &= ~O_TRUNC;
c8fe8f30 1885 break;
4a3fd211 1886 }
b41572e9 1887
3fb64190 1888 error = inode_permission(inode, acc_mode);
b41572e9
DH		 1889 if (error)
1890 return error;
6146f0d5 1891
1da177e4
LT		 1892 /*
1893 * An append-only file must be opened in append mode for writing.
1894 */
1895 if (IS_APPEND(inode)) {
8737c930 1896 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 1897 return -EPERM;
1da177e4 1898 if (flag & O_TRUNC)
7715b521 1899 return -EPERM;
1da177e4
LT		 1900 }
1901
1902 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 1903 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1904 return -EPERM;
1da177e4
LT		 1905
1906 /*
1907 * Ensure there are no outstanding leases on the file.
1908 */
b65a9cfc 1909 return break_lease(inode, flag);
7715b521 1910}
1da177e4 1911
7715b521
AV		 1912static int handle_truncate(struct path *path)
1913{
1914 struct inode *inode = path->dentry->d_inode;
1915 int error = get_write_access(inode);
1916 if (error)
1917 return error;
1918 /*
1919 * Refuse to truncate files with mandatory locks held on them.
1920 */
1921 error = locks_verify_locked(inode);
1922 if (!error)
ea0d3ab2 1923 error = security_path_truncate(path);
7715b521
AV		 1924 if (!error) {
1925 error = do_truncate(path->dentry, 0,
1926 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1927 NULL);
1928 }
1929 put_write_access(inode);
acd0c935 1930 return error;
1da177e4
LT		 1931}
1932
d57999e1
DH		 1933/*
1934 * Be careful about ever adding any more callers of this
1935 * function. Its flags must be in the namei format, not
1936 * what get passed to sys_open().
1937 */
1938static int __open_namei_create(struct nameidata *nd, struct path *path,
8737c930 1939 int open_flag, int mode)
aab520e2
DH		 1940{
1941 int error;
4ac91378 1942 struct dentry *dir = nd->path.dentry;
aab520e2
DH		 1943
1944 if (!IS_POSIXACL(dir->d_inode))
ce3b0f8d 1945 mode &= ~current_umask();
be6d3e56
KT		 1946 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
1947 if (error)
1948 goto out_unlock;
aab520e2 1949 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
be6d3e56 1950out_unlock:
aab520e2 1951 mutex_unlock(&dir->d_inode->i_mutex);
4ac91378
JB		 1952 dput(nd->path.dentry);
1953 nd->path.dentry = path->dentry;
31e6b01f 1954
aab520e2
DH		 1955 if (error)
1956 return error;
1957 /* Don't check for write permission, don't truncate */
8737c930 1958 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
aab520e2
DH		 1959}
1960
d57999e1
DH		 1961/*
1962 * Note that while the flag value (low two bits) for sys_open means:
1963 * 00 - read-only
1964 * 01 - write-only
1965 * 10 - read-write
1966 * 11 - special
1967 * it is changed into
1968 * 00 - no permissions needed
1969 * 01 - read-permission
1970 * 10 - write-permission
1971 * 11 - read-write
1972 * for the internal routines (ie open_namei()/follow_link() etc)
1973 * This is more logical, and also allows the 00 "no perm needed"
1974 * to be used for symlinks (where the permissions are checked
1975 * later).
1976 *
1977*/
1978static inline int open_to_namei_flags(int flag)
1979{
1980 if ((flag+1) & O_ACCMODE)
1981 flag++;
1982 return flag;
1983}
1984
7715b521 1985static int open_will_truncate(int flag, struct inode *inode)
4a3fd211
DH		 1986{
1987 /*
1988 * We'll never write to the fs underlying
1989 * a device file.
1990 */
1991 if (special_file(inode->i_mode))
1992 return 0;
1993 return (flag & O_TRUNC);
1994}
1995
648fa861 1996static struct file *finish_open(struct nameidata *nd,
9a66179e 1997 int open_flag, int acc_mode)
648fa861
AV		 1998{
1999 struct file *filp;
2000 int will_truncate;
2001 int error;
2002
9a66179e 2003 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
648fa861
AV		 2004 if (will_truncate) {
2005 error = mnt_want_write(nd->path.mnt);
2006 if (error)
2007 goto exit;
2008 }
2009 error = may_open(&nd->path, acc_mode, open_flag);
2010 if (error) {
2011 if (will_truncate)
2012 mnt_drop_write(nd->path.mnt);
2013 goto exit;
2014 }
2015 filp = nameidata_to_filp(nd);
2016 if (!IS_ERR(filp)) {
2017 error = ima_file_check(filp, acc_mode);
2018 if (error) {
2019 fput(filp);
2020 filp = ERR_PTR(error);
2021 }
2022 }
2023 if (!IS_ERR(filp)) {
648fa861
AV		 2024 if (will_truncate) {
2025 error = handle_truncate(&nd->path);
2026 if (error) {
2027 fput(filp);
2028 filp = ERR_PTR(error);
2029 }
2030 }
2031 }
2032 /*
2033 * It is now safe to drop the mnt write
2034 * because the filp has had a write taken
2035 * on its behalf.
2036 */
2037 if (will_truncate)
2038 mnt_drop_write(nd->path.mnt);
d893f1bc 2039 path_put(&nd->path);
648fa861
AV		 2040 return filp;
2041
2042exit:
2043 if (!IS_ERR(nd->intent.open.file))
2044 release_open_intent(nd);
2045 path_put(&nd->path);
2046 return ERR_PTR(error);
2047}
2048
31e6b01f
NP		 2049/*
2050 * Handle O_CREAT case for do_filp_open
2051 */
fb1cc555 2052static struct file *do_last(struct nameidata *nd, struct path *path,
5b369df8 2053 int open_flag, int acc_mode,
3e297b61 2054 int mode, const char *pathname)
fb1cc555 2055{
a1e28038 2056 struct dentry *dir = nd->path.dentry;
fb1cc555 2057 struct file *filp;
1f36f774
AV		 2058 int error = -EISDIR;
2059
2060 switch (nd->last_type) {
2061 case LAST_DOTDOT:
2062 follow_dotdot(nd);
2063 dir = nd->path.dentry;
176306f5 2064 case LAST_DOT:
1f36f774
AV		 2065 if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) {
2066 if (!dir->d_op->d_revalidate(dir, nd)) {
2067 error = -ESTALE;
2068 goto exit;
2069 }
2070 }
2071 /* fallthrough */
1f36f774 2072 case LAST_ROOT:
31e6b01f 2073 goto exit;
1f36f774
AV		 2074 case LAST_BIND:
2075 audit_inode(pathname, dir);
67ee3ad2 2076 goto ok;
1f36f774 2077 }
67ee3ad2 2078
1f36f774 2079 /* trailing slashes? */
31e6b01f
NP		 2080 if (nd->last.name[nd->last.len])
2081 goto exit;
a2c36b45 2082
a1e28038
AV		 2083 mutex_lock(&dir->d_inode->i_mutex);
2084
2085 path->dentry = lookup_hash(nd);
2086 path->mnt = nd->path.mnt;
2087
fb1cc555
AV		 2088 error = PTR_ERR(path->dentry);
2089 if (IS_ERR(path->dentry)) {
2090 mutex_unlock(&dir->d_inode->i_mutex);
2091 goto exit;
2092 }
2093
2094 if (IS_ERR(nd->intent.open.file)) {
2095 error = PTR_ERR(nd->intent.open.file);
2096 goto exit_mutex_unlock;
2097 }
2098
2099 /* Negative dentry, just create the file */
2100 if (!path->dentry->d_inode) {
2101 /*
2102 * This write is needed to ensure that a
2103 * ro->rw transition does not occur between
2104 * the time when the file is created and when
2105 * a permanent write count is taken through
2106 * the 'struct file' in nameidata_to_filp().
2107 */
2108 error = mnt_want_write(nd->path.mnt);
2109 if (error)
2110 goto exit_mutex_unlock;
2111 error = __open_namei_create(nd, path, open_flag, mode);
2112 if (error) {
2113 mnt_drop_write(nd->path.mnt);
2114 goto exit;
2115 }
2116 filp = nameidata_to_filp(nd);
2117 mnt_drop_write(nd->path.mnt);
d893f1bc 2118 path_put(&nd->path);
fb1cc555
AV		 2119 if (!IS_ERR(filp)) {
2120 error = ima_file_check(filp, acc_mode);
2121 if (error) {
2122 fput(filp);
2123 filp = ERR_PTR(error);
2124 }
2125 }
2126 return filp;
2127 }
2128
2129 /*
2130 * It already exists.
2131 */
2132 mutex_unlock(&dir->d_inode->i_mutex);
2133 audit_inode(pathname, path->dentry);
2134
2135 error = -EEXIST;
5b369df8 2136 if (open_flag & O_EXCL)
fb1cc555
AV		 2137 goto exit_dput;
2138
2139 if (__follow_mount(path)) {
2140 error = -ELOOP;
5b369df8 2141 if (open_flag & O_NOFOLLOW)
fb1cc555
AV		 2142 goto exit_dput;
2143 }
2144
2145 error = -ENOENT;
2146 if (!path->dentry->d_inode)
2147 goto exit_dput;
9e67f361
AV		 2148
2149 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2150 return NULL;
fb1cc555
AV		 2151
2152 path_to_nameidata(path, nd);
31e6b01f 2153 nd->inode = path->dentry->d_inode;
fb1cc555 2154 error = -EISDIR;
31e6b01f 2155 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2156 goto exit;
67ee3ad2 2157ok:
9a66179e 2158 filp = finish_open(nd, open_flag, acc_mode);
fb1cc555
AV		 2159 return filp;
2160
2161exit_mutex_unlock:
2162 mutex_unlock(&dir->d_inode->i_mutex);
2163exit_dput:
2164 path_put_conditional(path, nd);
2165exit:
2166 if (!IS_ERR(nd->intent.open.file))
2167 release_open_intent(nd);
fb1cc555
AV		 2168 path_put(&nd->path);
2169 return ERR_PTR(error);
2170}
2171
1da177e4 2172/*
4a3fd211
DH		 2173 * Note that the low bits of the passed in "open_flag"
2174 * are not the same as in the local variable "flag". See
2175 * open_to_namei_flags() for more details.
1da177e4 2176 */
a70e65df 2177struct file *do_filp_open(int dfd, const char *pathname,
6e8341a1 2178 int open_flag, int mode, int acc_mode)
1da177e4 2179{
4a3fd211 2180 struct file *filp;
a70e65df 2181 struct nameidata nd;
6e8341a1 2182 int error;
9850c056 2183 struct path path;
1da177e4 2184 int count = 0;
d57999e1 2185 int flag = open_to_namei_flags(open_flag);
31e6b01f 2186 int flags;
1f36f774
AV		 2187
2188 if (!(open_flag & O_CREAT))
2189 mode = 0;
1da177e4 2190
b1085ba8
LS		 2191 /* Must never be set by userspace */
2192 open_flag &= ~FMODE_NONOTIFY;
2193
6b2f3d1f
CH		 2194 /*
2195 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2196 * check for O_DSYNC if the need any syncing at all we enforce it's
2197 * always set instead of having to deal with possibly weird behaviour
2198 * for malicious applications setting only __O_SYNC.
2199 */
2200 if (open_flag & __O_SYNC)
2201 open_flag |= O_DSYNC;
2202
6e8341a1 2203 if (!acc_mode)
6d125529 2204 acc_mode = MAY_OPEN | ACC_MODE(open_flag);
1da177e4 2205
834f2a4a 2206 /* O_TRUNC implies we need access checks for write permissions */
4296e2cb 2207 if (open_flag & O_TRUNC)
834f2a4a
TM		 2208 acc_mode |= MAY_WRITE;
2209
1da177e4
LT		 2210 /* Allow the LSM permission hook to distinguish append
2211 access from general write access. */
4296e2cb 2212 if (open_flag & O_APPEND)
1da177e4
LT		 2213 acc_mode |= MAY_APPEND;
2214
31e6b01f
NP		 2215 flags = LOOKUP_OPEN;
2216 if (open_flag & O_CREAT) {
2217 flags |= LOOKUP_CREATE;
2218 if (open_flag & O_EXCL)
2219 flags |= LOOKUP_EXCL;
2220 }
2221 if (open_flag & O_DIRECTORY)
2222 flags |= LOOKUP_DIRECTORY;
2223 if (!(open_flag & O_NOFOLLOW))
2224 flags |= LOOKUP_FOLLOW;
2225
2226 filp = get_empty_filp();
2227 if (!filp)
2228 return ERR_PTR(-ENFILE);
2229
2230 filp->f_flags = open_flag;
2231 nd.intent.open.file = filp;
2232 nd.intent.open.flags = flag;
2233 nd.intent.open.create_mode = mode;
2234
2235 if (open_flag & O_CREAT)
2236 goto creat;
2237
2238 /* !O_CREAT, simple open */
2239 error = do_path_lookup(dfd, pathname, flags, &nd);
2240 if (unlikely(error))
2241 goto out_filp;
2242 error = -ELOOP;
2243 if (!(nd.flags & LOOKUP_FOLLOW)) {
2244 if (nd.inode->i_op->follow_link)
2245 goto out_path;
2246 }
2247 error = -ENOTDIR;
2248 if (nd.flags & LOOKUP_DIRECTORY) {
2249 if (!nd.inode->i_op->lookup)
2250 goto out_path;
2251 }
2252 audit_inode(pathname, nd.path.dentry);
2253 filp = finish_open(&nd, open_flag, acc_mode);
2254 return filp;
2255
2256creat:
2257 /* OK, have to create the file. Find the parent. */
2258 error = path_init_rcu(dfd, pathname,
2259 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
1da177e4 2260 if (error)
31e6b01f
NP		 2261 goto out_filp;
2262 error = path_walk_rcu(pathname, &nd);
2263 path_finish_rcu(&nd);
2264 if (unlikely(error == -ECHILD || error == -ESTALE)) {
2265 /* slower, locked walk */
2266 if (error == -ESTALE) {
2267reval:
2268 flags |= LOOKUP_REVAL;
2269 }
2270 error = path_init(dfd, pathname,
2271 LOOKUP_PARENT | (flags & LOOKUP_REVAL), &nd);
2272 if (error)
2273 goto out_filp;
3866248e 2274
31e6b01f 2275 error = path_walk_simple(pathname, &nd);
654f562c 2276 }
31e6b01f
NP		 2277 if (unlikely(error))
2278 goto out_filp;
2279 if (unlikely(!audit_dummy_context()))
9b4a9b14 2280 audit_inode(pathname, nd.path.dentry);
1da177e4
LT		 2281
2282 /*
a2c36b45 2283 * We have the parent and last component.
1da177e4 2284 */
31e6b01f 2285 nd.flags = flags;
3e297b61 2286 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
806b681c 2287 while (unlikely(!filp)) { /* trailing symlink */
def4af30 2288 struct path holder;
def4af30 2289 void *cookie;
806b681c 2290 error = -ELOOP;
1f36f774 2291 /* S_ISDIR part is a temporary automount kludge */
31e6b01f 2292 if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(nd.inode->i_mode))
1f36f774
AV		 2293 goto exit_dput;
2294 if (count++ == 32)
806b681c
AV		 2295 goto exit_dput;
2296 /*
2297 * This is subtle. Instead of calling do_follow_link() we do
2298 * the thing by hands. The reason is that this way we have zero
2299 * link_count and path_walk() (called from ->follow_link)
2300 * honoring LOOKUP_PARENT. After that we have the parent and
2301 * last component, i.e. we are in the same situation as after
2302 * the first path_walk(). Well, almost - if the last component
2303 * is normal we get its copy stored in nd->last.name and we will
2304 * have to putname() it when we are done. Procfs-like symlinks
2305 * just set LAST_BIND.
2306 */
2307 nd.flags |= LOOKUP_PARENT;
2308 error = security_inode_follow_link(path.dentry, &nd);
2309 if (error)
2310 goto exit_dput;
def4af30
AV		 2311 error = __do_follow_link(&path, &nd, &cookie);
2312 if (unlikely(error)) {
31e6b01f
NP		 2313 if (!IS_ERR(cookie) && nd.inode->i_op->put_link)
2314 nd.inode->i_op->put_link(path.dentry, &nd, cookie);
806b681c 2315 /* nd.path had been dropped */
31e6b01f
NP		 2316 nd.path = path;
2317 goto out_path;
806b681c 2318 }
def4af30 2319 holder = path;
806b681c 2320 nd.flags &= ~LOOKUP_PARENT;
3e297b61 2321 filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
31e6b01f
NP		 2322 if (nd.inode->i_op->put_link)
2323 nd.inode->i_op->put_link(holder.dentry, &nd, cookie);
def4af30 2324 path_put(&holder);
806b681c 2325 }
10fa8e62 2326out:
2a737871
AV		 2327 if (nd.root.mnt)
2328 path_put(&nd.root);
31e6b01f 2329 if (filp == ERR_PTR(-ESTALE) && !(flags & LOOKUP_REVAL))
10fa8e62 2330 goto reval;
10fa8e62 2331 return filp;
1da177e4 2332
806b681c
AV		 2333exit_dput:
2334 path_put_conditional(&path, &nd);
31e6b01f
NP		 2335out_path:
2336 path_put(&nd.path);
2337out_filp:
806b681c 2338 if (!IS_ERR(nd.intent.open.file))
a70e65df 2339 release_open_intent(&nd);
806b681c 2340 filp = ERR_PTR(error);
10fa8e62 2341 goto out;
1da177e4
LT		 2342}
2343
a70e65df
CH		 2344/**
2345 * filp_open - open file and return file pointer
2346 *
2347 * @filename: path to open
2348 * @flags: open flags as per the open(2) second argument
2349 * @mode: mode for the new file if O_CREAT is set, else ignored
2350 *
2351 * This is the helper to open a file from kernelspace if you really
2352 * have to. But in generally you should not do this, so please move
2353 * along, nothing to see here..
2354 */
2355struct file *filp_open(const char *filename, int flags, int mode)
2356{
6e8341a1 2357 return do_filp_open(AT_FDCWD, filename, flags, mode, 0);
a70e65df
CH		 2358}
2359EXPORT_SYMBOL(filp_open);
2360
1da177e4
LT		 2361/**
2362 * lookup_create - lookup a dentry, creating it if it doesn't exist
2363 * @nd: nameidata info
2364 * @is_dir: directory flag
2365 *
2366 * Simple function to lookup and return a dentry and create it
2367 * if it doesn't exist. Is SMP-safe.
c663e5d8 2368 *
4ac91378 2369 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2370 */
2371struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2372{
c663e5d8 2373 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2374
4ac91378 2375 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2376 /*
2377 * Yucky last component or no last component at all?
2378 * (foo/., foo/.., /////)
2379 */
1da177e4
LT		 2380 if (nd->last_type != LAST_NORM)
2381 goto fail;
2382 nd->flags &= ~LOOKUP_PARENT;
3516586a 2383 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2384 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2385
2386 /*
2387 * Do the final lookup.
2388 */
49705b77 2389 dentry = lookup_hash(nd);
1da177e4
LT		 2390 if (IS_ERR(dentry))
2391 goto fail;
c663e5d8 2392
e9baf6e5
AV		 2393 if (dentry->d_inode)
2394 goto eexist;
c663e5d8
CH		 2395 /*
2396 * Special case - lookup gave negative, but... we had foo/bar/
2397 * From the vfs_mknod() POV we just have a negative dentry -
2398 * all is fine. Let's be bastards - you had / on the end, you've
2399 * been asking for (non-existent) directory. -ENOENT for you.
2400 */
e9baf6e5
AV		 2401 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2402 dput(dentry);
2403 dentry = ERR_PTR(-ENOENT);
2404 }
1da177e4 2405 return dentry;
e9baf6e5 2406eexist:
1da177e4 2407 dput(dentry);
e9baf6e5 2408 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2409fail:
2410 return dentry;
2411}
f81a0bff 2412EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2413
2414int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2415{
a95164d9 2416 int error = may_create(dir, dentry);
1da177e4
LT		 2417
2418 if (error)
2419 return error;
2420
2421 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2422 return -EPERM;
2423
acfa4380 2424 if (!dir->i_op->mknod)
1da177e4
LT		 2425 return -EPERM;
2426
08ce5f16
SH		 2427 error = devcgroup_inode_mknod(mode, dev);
2428 if (error)
2429 return error;
2430
1da177e4
LT		 2431 error = security_inode_mknod(dir, dentry, mode, dev);
2432 if (error)
2433 return error;
2434
1da177e4 2435 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2436 if (!error)
f38aa942 2437 fsnotify_create(dir, dentry);
1da177e4
LT		 2438 return error;
2439}
2440
463c3197
DH		 2441static int may_mknod(mode_t mode)
2442{
2443 switch (mode & S_IFMT) {
2444 case S_IFREG:
2445 case S_IFCHR:
2446 case S_IFBLK:
2447 case S_IFIFO:
2448 case S_IFSOCK:
2449 case 0: /* zero mode translates to S_IFREG */
2450 return 0;
2451 case S_IFDIR:
2452 return -EPERM;
2453 default:
2454 return -EINVAL;
2455 }
2456}
2457
2e4d0924
HC		 2458SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2459 unsigned, dev)
1da177e4 2460{
2ad94ae6
AV		 2461 int error;
2462 char *tmp;
2463 struct dentry *dentry;
1da177e4
LT		 2464 struct nameidata nd;
2465
2466 if (S_ISDIR(mode))
2467 return -EPERM;
1da177e4 2468
2ad94ae6 2469 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2470 if (error)
2ad94ae6
AV		 2471 return error;
2472
1da177e4 2473 dentry = lookup_create(&nd, 0);
463c3197
DH		 2474 if (IS_ERR(dentry)) {
2475 error = PTR_ERR(dentry);
2476 goto out_unlock;
2477 }
4ac91378 2478 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2479 mode &= ~current_umask();
463c3197
DH		 2480 error = may_mknod(mode);
2481 if (error)
2482 goto out_dput;
2483 error = mnt_want_write(nd.path.mnt);
2484 if (error)
2485 goto out_dput;
be6d3e56
KT		 2486 error = security_path_mknod(&nd.path, dentry, mode, dev);
2487 if (error)
2488 goto out_drop_write;
463c3197 2489 switch (mode & S_IFMT) {
1da177e4 2490 case 0: case S_IFREG:
4ac91378 2491 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2492 break;
2493 case S_IFCHR: case S_IFBLK:
4ac91378 2494 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2495 new_decode_dev(dev));
2496 break;
2497 case S_IFIFO: case S_IFSOCK:
4ac91378 2498 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2499 break;
1da177e4 2500 }
be6d3e56 2501out_drop_write:
463c3197
DH		 2502 mnt_drop_write(nd.path.mnt);
2503out_dput:
2504 dput(dentry);
2505out_unlock:
4ac91378 2506 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2507 path_put(&nd.path);
1da177e4
LT		 2508 putname(tmp);
2509
2510 return error;
2511}
2512
3480b257 2513SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2514{
2515 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2516}
2517
1da177e4
LT		 2518int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2519{
a95164d9 2520 int error = may_create(dir, dentry);
1da177e4
LT		 2521
2522 if (error)
2523 return error;
2524
acfa4380 2525 if (!dir->i_op->mkdir)
1da177e4
LT		 2526 return -EPERM;
2527
2528 mode &= (S_IRWXUGO|S_ISVTX);
2529 error = security_inode_mkdir(dir, dentry, mode);
2530 if (error)
2531 return error;
2532
1da177e4 2533 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2534 if (!error)
f38aa942 2535 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2536 return error;
2537}
2538
2e4d0924 2539SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2540{
2541 int error = 0;
2542 char * tmp;
6902d925
DH		 2543 struct dentry *dentry;
2544 struct nameidata nd;
1da177e4 2545
2ad94ae6
AV		 2546 error = user_path_parent(dfd, pathname, &nd, &tmp);
2547 if (error)
6902d925 2548 goto out_err;
1da177e4 2549
6902d925
DH		 2550 dentry = lookup_create(&nd, 1);
2551 error = PTR_ERR(dentry);
2552 if (IS_ERR(dentry))
2553 goto out_unlock;
1da177e4 2554
4ac91378 2555 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2556 mode &= ~current_umask();
463c3197
DH		 2557 error = mnt_want_write(nd.path.mnt);
2558 if (error)
2559 goto out_dput;
be6d3e56
KT		 2560 error = security_path_mkdir(&nd.path, dentry, mode);
2561 if (error)
2562 goto out_drop_write;
4ac91378 2563 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2564out_drop_write:
463c3197
DH		 2565 mnt_drop_write(nd.path.mnt);
2566out_dput:
6902d925
DH		 2567 dput(dentry);
2568out_unlock:
4ac91378 2569 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2570 path_put(&nd.path);
6902d925
DH		 2571 putname(tmp);
2572out_err:
1da177e4
LT		 2573 return error;
2574}
2575
3cdad428 2576SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2577{
2578 return sys_mkdirat(AT_FDCWD, pathname, mode);
2579}
2580
1da177e4
LT		 2581/*
2582 * We try to drop the dentry early: we should have
2583 * a usage count of 2 if we're the only user of this
2584 * dentry, and if that is true (possibly after pruning
2585 * the dcache), then we drop the dentry now.
2586 *
2587 * A low-level filesystem can, if it choses, legally
2588 * do a
2589 *
2590 * if (!d_unhashed(dentry))
2591 * return -EBUSY;
2592 *
2593 * if it cannot handle the case of removing a directory
2594 * that is still in use by something else..
2595 */
2596void dentry_unhash(struct dentry *dentry)
2597{
2598 dget(dentry);
dc168427 2599 shrink_dcache_parent(dentry);
1da177e4 2600 spin_lock(&dentry->d_lock);
b7ab39f6 2601 if (dentry->d_count == 2)
1da177e4
LT		 2602 __d_drop(dentry);
2603 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2604}
2605
2606int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2607{
2608 int error = may_delete(dir, dentry, 1);
2609
2610 if (error)
2611 return error;
2612
acfa4380 2613 if (!dir->i_op->rmdir)
1da177e4
LT		 2614 return -EPERM;
2615
1b1dcc1b 2616 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2617 dentry_unhash(dentry);
2618 if (d_mountpoint(dentry))
2619 error = -EBUSY;
2620 else {
2621 error = security_inode_rmdir(dir, dentry);
2622 if (!error) {
2623 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2624 if (!error) {
1da177e4 2625 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2626 dont_mount(dentry);
2627 }
1da177e4
LT		 2628 }
2629 }
1b1dcc1b 2630 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2631 if (!error) {
1da177e4
LT		 2632 d_delete(dentry);
2633 }
2634 dput(dentry);
2635
2636 return error;
2637}
2638
5590ff0d 2639static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2640{
2641 int error = 0;
2642 char * name;
2643 struct dentry *dentry;
2644 struct nameidata nd;
2645
2ad94ae6 2646 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2647 if (error)
2ad94ae6 2648 return error;
1da177e4
LT		 2649
2650 switch(nd.last_type) {
0612d9fb
OH		 2651 case LAST_DOTDOT:
2652 error = -ENOTEMPTY;
2653 goto exit1;
2654 case LAST_DOT:
2655 error = -EINVAL;
2656 goto exit1;
2657 case LAST_ROOT:
2658 error = -EBUSY;
2659 goto exit1;
1da177e4 2660 }
0612d9fb
OH		 2661
2662 nd.flags &= ~LOOKUP_PARENT;
2663
4ac91378 2664 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2665 dentry = lookup_hash(&nd);
1da177e4 2666 error = PTR_ERR(dentry);
6902d925
DH		 2667 if (IS_ERR(dentry))
2668 goto exit2;
0622753b
DH		 2669 error = mnt_want_write(nd.path.mnt);
2670 if (error)
2671 goto exit3;
be6d3e56
KT		 2672 error = security_path_rmdir(&nd.path, dentry);
2673 if (error)
2674 goto exit4;
4ac91378 2675 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2676exit4:
0622753b
DH		 2677 mnt_drop_write(nd.path.mnt);
2678exit3:
6902d925
DH		 2679 dput(dentry);
2680exit2:
4ac91378 2681 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2682exit1:
1d957f9b 2683 path_put(&nd.path);
1da177e4
LT		 2684 putname(name);
2685 return error;
2686}
2687
3cdad428 2688SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2689{
2690 return do_rmdir(AT_FDCWD, pathname);
2691}
2692
1da177e4
LT		 2693int vfs_unlink(struct inode *dir, struct dentry *dentry)
2694{
2695 int error = may_delete(dir, dentry, 0);
2696
2697 if (error)
2698 return error;
2699
acfa4380 2700 if (!dir->i_op->unlink)
1da177e4
LT		 2701 return -EPERM;
2702
1b1dcc1b 2703 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2704 if (d_mountpoint(dentry))
2705 error = -EBUSY;
2706 else {
2707 error = security_inode_unlink(dir, dentry);
bec1052e 2708 if (!error) {
1da177e4 2709 error = dir->i_op->unlink(dir, dentry);
bec1052e 2710 if (!error)
d83c49f3 2711 dont_mount(dentry);
bec1052e 2712 }
1da177e4 2713 }
1b1dcc1b 2714 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2715
2716 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2717 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2718 fsnotify_link_count(dentry->d_inode);
e234f35c 2719 d_delete(dentry);
1da177e4 2720 }
0eeca283 2721
1da177e4
LT		 2722 return error;
2723}
2724
2725/*
2726 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2727 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2728 * writeout happening, and we don't want to prevent access to the directory
2729 * while waiting on the I/O.
2730 */
5590ff0d 2731static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2732{
2ad94ae6
AV		 2733 int error;
2734 char *name;
1da177e4
LT		 2735 struct dentry *dentry;
2736 struct nameidata nd;
2737 struct inode *inode = NULL;
2738
2ad94ae6 2739 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2740 if (error)
2ad94ae6
AV		 2741 return error;
2742
1da177e4
LT		 2743 error = -EISDIR;
2744 if (nd.last_type != LAST_NORM)
2745 goto exit1;
0612d9fb
OH		 2746
2747 nd.flags &= ~LOOKUP_PARENT;
2748
4ac91378 2749 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2750 dentry = lookup_hash(&nd);
1da177e4
LT		 2751 error = PTR_ERR(dentry);
2752 if (!IS_ERR(dentry)) {
2753 /* Why not before? Because we want correct error value */
2754 if (nd.last.name[nd.last.len])
2755 goto slashes;
2756 inode = dentry->d_inode;
2757 if (inode)
7de9c6ee 2758 ihold(inode);
0622753b
DH		 2759 error = mnt_want_write(nd.path.mnt);
2760 if (error)
2761 goto exit2;
be6d3e56
KT		 2762 error = security_path_unlink(&nd.path, dentry);
2763 if (error)
2764 goto exit3;
4ac91378 2765 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2766exit3:
0622753b 2767 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2768 exit2:
2769 dput(dentry);
2770 }
4ac91378 2771 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2772 if (inode)
2773 iput(inode); /* truncate the inode here */
2774exit1:
1d957f9b 2775 path_put(&nd.path);
1da177e4
LT		 2776 putname(name);
2777 return error;
2778
2779slashes:
2780 error = !dentry->d_inode ? -ENOENT :
2781 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2782 goto exit2;
2783}
2784
2e4d0924 2785SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2786{
2787 if ((flag & ~AT_REMOVEDIR) != 0)
2788 return -EINVAL;
2789
2790 if (flag & AT_REMOVEDIR)
2791 return do_rmdir(dfd, pathname);
2792
2793 return do_unlinkat(dfd, pathname);
2794}
2795
3480b257 2796SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2797{
2798 return do_unlinkat(AT_FDCWD, pathname);
2799}
2800
db2e747b 2801int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2802{
a95164d9 2803 int error = may_create(dir, dentry);
1da177e4
LT		 2804
2805 if (error)
2806 return error;
2807
acfa4380 2808 if (!dir->i_op->symlink)
1da177e4
LT		 2809 return -EPERM;
2810
2811 error = security_inode_symlink(dir, dentry, oldname);
2812 if (error)
2813 return error;
2814
1da177e4 2815 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2816 if (!error)
f38aa942 2817 fsnotify_create(dir, dentry);
1da177e4
LT		 2818 return error;
2819}
2820
2e4d0924
HC		 2821SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2822 int, newdfd, const char __user *, newname)
1da177e4 2823{
2ad94ae6
AV		 2824 int error;
2825 char *from;
2826 char *to;
6902d925
DH		 2827 struct dentry *dentry;
2828 struct nameidata nd;
1da177e4
LT		 2829
2830 from = getname(oldname);
2ad94ae6 2831 if (IS_ERR(from))
1da177e4 2832 return PTR_ERR(from);
1da177e4 2833
2ad94ae6 2834 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2835 if (error)
2ad94ae6
AV		 2836 goto out_putname;
2837
6902d925
DH		 2838 dentry = lookup_create(&nd, 0);
2839 error = PTR_ERR(dentry);
2840 if (IS_ERR(dentry))
2841 goto out_unlock;
2842
75c3f29d
DH		 2843 error = mnt_want_write(nd.path.mnt);
2844 if (error)
2845 goto out_dput;
be6d3e56
KT		 2846 error = security_path_symlink(&nd.path, dentry, from);
2847 if (error)
2848 goto out_drop_write;
db2e747b 2849 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2850out_drop_write:
75c3f29d
DH		 2851 mnt_drop_write(nd.path.mnt);
2852out_dput:
6902d925
DH		 2853 dput(dentry);
2854out_unlock:
4ac91378 2855 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2856 path_put(&nd.path);
6902d925
DH		 2857 putname(to);
2858out_putname:
1da177e4
LT		 2859 putname(from);
2860 return error;
2861}
2862
3480b257 2863SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2864{
2865 return sys_symlinkat(oldname, AT_FDCWD, newname);
2866}
2867
1da177e4
LT		 2868int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2869{
2870 struct inode *inode = old_dentry->d_inode;
2871 int error;
2872
2873 if (!inode)
2874 return -ENOENT;
2875
a95164d9 2876 error = may_create(dir, new_dentry);
1da177e4
LT		 2877 if (error)
2878 return error;
2879
2880 if (dir->i_sb != inode->i_sb)
2881 return -EXDEV;
2882
2883 /*
2884 * A link to an append-only or immutable file cannot be created.
2885 */
2886 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2887 return -EPERM;
acfa4380 2888 if (!dir->i_op->link)
1da177e4 2889 return -EPERM;
7e79eedb 2890 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2891 return -EPERM;
2892
2893 error = security_inode_link(old_dentry, dir, new_dentry);
2894 if (error)
2895 return error;
2896
7e79eedb 2897 mutex_lock(&inode->i_mutex);
1da177e4 2898 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2899 mutex_unlock(&inode->i_mutex);
e31e14ec 2900 if (!error)
7e79eedb 2901 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2902 return error;
2903}
2904
2905/*
2906 * Hardlinks are often used in delicate situations. We avoid
2907 * security-related surprises by not following symlinks on the
2908 * newname. --KAB
2909 *
2910 * We don't follow them on the oldname either to be compatible
2911 * with linux 2.0, and to avoid hard-linking to directories
2912 * and other special files. --ADM
2913 */
2e4d0924
HC		 2914SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2915 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2916{
2917 struct dentry *new_dentry;
2d8f3038
AV		 2918 struct nameidata nd;
2919 struct path old_path;
1da177e4 2920 int error;
2ad94ae6 2921 char *to;
1da177e4 2922
45c9b11a 2923 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 2924 return -EINVAL;
2925
2d8f3038
AV		 2926 error = user_path_at(olddfd, oldname,
2927 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2928 &old_path);
1da177e4 2929 if (error)
2ad94ae6
AV		 2930 return error;
2931
2932 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2933 if (error)
2934 goto out;
2935 error = -EXDEV;
2d8f3038 2936 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2937 goto out_release;
2938 new_dentry = lookup_create(&nd, 0);
2939 error = PTR_ERR(new_dentry);
6902d925
DH		 2940 if (IS_ERR(new_dentry))
2941 goto out_unlock;
75c3f29d
DH		 2942 error = mnt_want_write(nd.path.mnt);
2943 if (error)
2944 goto out_dput;
be6d3e56
KT		 2945 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2946 if (error)
2947 goto out_drop_write;
2d8f3038 2948 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2949out_drop_write:
75c3f29d
DH		 2950 mnt_drop_write(nd.path.mnt);
2951out_dput:
6902d925
DH		 2952 dput(new_dentry);
2953out_unlock:
4ac91378 2954 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2955out_release:
1d957f9b 2956 path_put(&nd.path);
2ad94ae6 2957 putname(to);
1da177e4 2958out:
2d8f3038 2959 path_put(&old_path);
1da177e4
LT		 2960
2961 return error;
2962}
2963
3480b257 2964SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 2965{
c04030e1 2966 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2967}
2968
1da177e4
LT		 2969/*
2970 * The worst of all namespace operations - renaming directory. "Perverted"
2971 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2972 * Problems:
2973 * a) we can get into loop creation. Check is done in is_subdir().
2974 * b) race potential - two innocent renames can create a loop together.
2975 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 2976 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 2977 * story.
2978 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 2979 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 2980 * whether the target exists). Solution: try to be smart with locking
2981 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 2982 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 2983 * move will be locked. Thus we can rank directories by the tree
2984 * (ancestors first) and rank all non-directories after them.
2985 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 2986 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 2987 * HOWEVER, it relies on the assumption that any object with ->lookup()
2988 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2989 * we'd better make sure that there's no link(2) for them.
2990 * d) some filesystems don't support opened-but-unlinked directories,
2991 * either because of layout or because they are not ready to deal with
2992 * all cases correctly. The latter will be fixed (taking this sort of
2993 * stuff into VFS), but the former is not going away. Solution: the same
2994 * trick as in rmdir().
2995 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 2996 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 2997 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 2998 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 2999 * locking].
3000 */
75c96f85
AB		 3001static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3002 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3003{
3004 int error = 0;
3005 struct inode *target;
3006
3007 /*
3008 * If we are going to change the parent - check write permissions,
3009 * we'll need to flip '..'.
3010 */
3011 if (new_dir != old_dir) {
f419a2e3 3012 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3013 if (error)
3014 return error;
3015 }
3016
3017 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3018 if (error)
3019 return error;
3020
3021 target = new_dentry->d_inode;
d83c49f3 3022 if (target)
1b1dcc1b 3023 mutex_lock(&target->i_mutex);
1da177e4
LT		 3024 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3025 error = -EBUSY;
d83c49f3
AV		 3026 else {
3027 if (target)
3028 dentry_unhash(new_dentry);
1da177e4 3029 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3030 }
1da177e4 3031 if (target) {
d83c49f3 3032 if (!error) {
1da177e4 3033 target->i_flags |= S_DEAD;
d83c49f3
AV		 3034 dont_mount(new_dentry);
3035 }
1b1dcc1b 3036 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3037 if (d_unhashed(new_dentry))
3038 d_rehash(new_dentry);
3039 dput(new_dentry);
3040 }
e31e14ec 3041 if (!error)
349457cc
MF		 3042 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3043 d_move(old_dentry,new_dentry);
1da177e4
LT		 3044 return error;
3045}
3046
75c96f85
AB		 3047static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3048 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3049{
3050 struct inode *target;
3051 int error;
3052
3053 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3054 if (error)
3055 return error;
3056
3057 dget(new_dentry);
3058 target = new_dentry->d_inode;
3059 if (target)
1b1dcc1b 3060 mutex_lock(&target->i_mutex);
1da177e4
LT		 3061 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3062 error = -EBUSY;
3063 else
3064 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3065 if (!error) {
bec1052e 3066 if (target)
d83c49f3 3067 dont_mount(new_dentry);
349457cc 3068 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3069 d_move(old_dentry, new_dentry);
1da177e4
LT		 3070 }
3071 if (target)
1b1dcc1b 3072 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3073 dput(new_dentry);
3074 return error;
3075}
3076
3077int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3078 struct inode *new_dir, struct dentry *new_dentry)
3079{
3080 int error;
3081 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3082 const unsigned char *old_name;
1da177e4
LT		 3083
3084 if (old_dentry->d_inode == new_dentry->d_inode)
3085 return 0;
3086
3087 error = may_delete(old_dir, old_dentry, is_dir);
3088 if (error)
3089 return error;
3090
3091 if (!new_dentry->d_inode)
a95164d9 3092 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3093 else
3094 error = may_delete(new_dir, new_dentry, is_dir);
3095 if (error)
3096 return error;
3097
acfa4380 3098 if (!old_dir->i_op->rename)
1da177e4
LT		 3099 return -EPERM;
3100
0eeca283
RL		 3101 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3102
1da177e4
LT		 3103 if (is_dir)
3104 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3105 else
3106 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3107 if (!error)
3108 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3109 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3110 fsnotify_oldname_free(old_name);
3111
1da177e4
LT		 3112 return error;
3113}
3114
2e4d0924
HC		 3115SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3116 int, newdfd, const char __user *, newname)
1da177e4 3117{
2ad94ae6
AV		 3118 struct dentry *old_dir, *new_dir;
3119 struct dentry *old_dentry, *new_dentry;
3120 struct dentry *trap;
1da177e4 3121 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3122 char *from;
3123 char *to;
3124 int error;
1da177e4 3125
2ad94ae6 3126 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3127 if (error)
3128 goto exit;
3129
2ad94ae6 3130 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3131 if (error)
3132 goto exit1;
3133
3134 error = -EXDEV;
4ac91378 3135 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3136 goto exit2;
3137
4ac91378 3138 old_dir = oldnd.path.dentry;
1da177e4
LT		 3139 error = -EBUSY;
3140 if (oldnd.last_type != LAST_NORM)
3141 goto exit2;
3142
4ac91378 3143 new_dir = newnd.path.dentry;
1da177e4
LT		 3144 if (newnd.last_type != LAST_NORM)
3145 goto exit2;
3146
0612d9fb
OH		 3147 oldnd.flags &= ~LOOKUP_PARENT;
3148 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3149 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3150
1da177e4
LT		 3151 trap = lock_rename(new_dir, old_dir);
3152
49705b77 3153 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3154 error = PTR_ERR(old_dentry);
3155 if (IS_ERR(old_dentry))
3156 goto exit3;
3157 /* source must exist */
3158 error = -ENOENT;
3159 if (!old_dentry->d_inode)
3160 goto exit4;
3161 /* unless the source is a directory trailing slashes give -ENOTDIR */
3162 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3163 error = -ENOTDIR;
3164 if (oldnd.last.name[oldnd.last.len])
3165 goto exit4;
3166 if (newnd.last.name[newnd.last.len])
3167 goto exit4;
3168 }
3169 /* source should not be ancestor of target */
3170 error = -EINVAL;
3171 if (old_dentry == trap)
3172 goto exit4;
49705b77 3173 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3174 error = PTR_ERR(new_dentry);
3175 if (IS_ERR(new_dentry))
3176 goto exit4;
3177 /* target should not be an ancestor of source */
3178 error = -ENOTEMPTY;
3179 if (new_dentry == trap)
3180 goto exit5;
3181
9079b1eb
DH		 3182 error = mnt_want_write(oldnd.path.mnt);
3183 if (error)
3184 goto exit5;
be6d3e56
KT		 3185 error = security_path_rename(&oldnd.path, old_dentry,
3186 &newnd.path, new_dentry);
3187 if (error)
3188 goto exit6;
1da177e4
LT		 3189 error = vfs_rename(old_dir->d_inode, old_dentry,
3190 new_dir->d_inode, new_dentry);
be6d3e56 3191exit6:
9079b1eb 3192 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3193exit5:
3194 dput(new_dentry);
3195exit4:
3196 dput(old_dentry);
3197exit3:
3198 unlock_rename(new_dir, old_dir);
3199exit2:
1d957f9b 3200 path_put(&newnd.path);
2ad94ae6 3201 putname(to);
1da177e4 3202exit1:
1d957f9b 3203 path_put(&oldnd.path);
1da177e4 3204 putname(from);
2ad94ae6 3205exit:
1da177e4
LT		 3206 return error;
3207}
3208
a26eab24 3209SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3210{
3211 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3212}
3213
1da177e4
LT		 3214int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3215{
3216 int len;
3217
3218 len = PTR_ERR(link);
3219 if (IS_ERR(link))
3220 goto out;
3221
3222 len = strlen(link);
3223 if (len > (unsigned) buflen)
3224 len = buflen;
3225 if (copy_to_user(buffer, link, len))
3226 len = -EFAULT;
3227out:
3228 return len;
3229}
3230
3231/*
3232 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3233 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3234 * using) it for any given inode is up to filesystem.
3235 */
3236int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3237{
3238 struct nameidata nd;
cc314eef 3239 void *cookie;
694a1764 3240 int res;
cc314eef 3241
1da177e4 3242 nd.depth = 0;
cc314eef 3243 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3244 if (IS_ERR(cookie))
3245 return PTR_ERR(cookie);
3246
3247 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3248 if (dentry->d_inode->i_op->put_link)
3249 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3250 return res;
1da177e4
LT		 3251}
3252
3253int vfs_follow_link(struct nameidata *nd, const char *link)
3254{
3255 return __vfs_follow_link(nd, link);
3256}
3257
3258/* get the link contents into pagecache */
3259static char *page_getlink(struct dentry * dentry, struct page **ppage)
3260{
ebd09abb
DG		 3261 char *kaddr;
3262 struct page *page;
1da177e4 3263 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3264 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3265 if (IS_ERR(page))
6fe6900e 3266 return (char*)page;
1da177e4 3267 *ppage = page;
ebd09abb
DG		 3268 kaddr = kmap(page);
3269 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3270 return kaddr;
1da177e4
LT		 3271}
3272
3273int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3274{
3275 struct page *page = NULL;
3276 char *s = page_getlink(dentry, &page);
3277 int res = vfs_readlink(dentry,buffer,buflen,s);
3278 if (page) {
3279 kunmap(page);
3280 page_cache_release(page);
3281 }
3282 return res;
3283}
3284
cc314eef 3285void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3286{
cc314eef 3287 struct page *page = NULL;
1da177e4 3288 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3289 return page;
1da177e4
LT		 3290}
3291
cc314eef 3292void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3293{
cc314eef
LT		 3294 struct page *page = cookie;
3295
3296 if (page) {
1da177e4
LT		 3297 kunmap(page);
3298 page_cache_release(page);
1da177e4
LT		 3299 }
3300}
3301
54566b2c
NP		 3302/*
3303 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3304 */
3305int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3306{
3307 struct address_space *mapping = inode->i_mapping;
0adb25d2 3308 struct page *page;
afddba49 3309 void *fsdata;
beb497ab 3310 int err;
1da177e4 3311 char *kaddr;
54566b2c
NP		 3312 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3313 if (nofs)
3314 flags |= AOP_FLAG_NOFS;
1da177e4 3315
7e53cac4 3316retry:
afddba49 3317 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3318 flags, &page, &fsdata);
1da177e4 3319 if (err)
afddba49
NP		 3320 goto fail;
3321
1da177e4
LT		 3322 kaddr = kmap_atomic(page, KM_USER0);
3323 memcpy(kaddr, symname, len-1);
3324 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3325
3326 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3327 page, fsdata);
1da177e4
LT		 3328 if (err < 0)
3329 goto fail;
afddba49
NP		 3330 if (err < len-1)
3331 goto retry;
3332
1da177e4
LT		 3333 mark_inode_dirty(inode);
3334 return 0;
1da177e4
LT		 3335fail:
3336 return err;
3337}
3338
0adb25d2
KK		 3339int page_symlink(struct inode *inode, const char *symname, int len)
3340{
3341 return __page_symlink(inode, symname, len,
54566b2c 3342 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3343}
3344
92e1d5be 3345const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3346 .readlink = generic_readlink,
3347 .follow_link = page_follow_link_light,
3348 .put_link = page_put_link,
3349};
3350
2d8f3038 3351EXPORT_SYMBOL(user_path_at);
1da177e4
LT		 3352EXPORT_SYMBOL(follow_down);
3353EXPORT_SYMBOL(follow_up);
3354EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3355EXPORT_SYMBOL(getname);
3356EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3357EXPORT_SYMBOL(lookup_one_len);
3358EXPORT_SYMBOL(page_follow_link_light);
3359EXPORT_SYMBOL(page_put_link);
3360EXPORT_SYMBOL(page_readlink);
0adb25d2 3361EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3362EXPORT_SYMBOL(page_symlink);
3363EXPORT_SYMBOL(page_symlink_inode_operations);
3364EXPORT_SYMBOL(path_lookup);
d1811465 3365EXPORT_SYMBOL(kern_path);
16f18200 3366EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3367EXPORT_SYMBOL(inode_permission);
8c744fb8 3368EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3369EXPORT_SYMBOL(unlock_rename);
3370EXPORT_SYMBOL(vfs_create);
3371EXPORT_SYMBOL(vfs_follow_link);
3372EXPORT_SYMBOL(vfs_link);
3373EXPORT_SYMBOL(vfs_mkdir);
3374EXPORT_SYMBOL(vfs_mknod);
3375EXPORT_SYMBOL(generic_permission);
3376EXPORT_SYMBOL(vfs_readlink);
3377EXPORT_SYMBOL(vfs_rename);
3378EXPORT_SYMBOL(vfs_rmdir);
3379EXPORT_SYMBOL(vfs_symlink);
3380EXPORT_SYMBOL(vfs_unlink);
3381EXPORT_SYMBOL(dentry_unhash);
3382EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)