projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
exportfs: Return the minimum required handle size
[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 */
b74c79e9
NP		 172static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
173 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
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) {
b74c79e9
NP		 183 int error = check_acl(inode, mask, flags);
184 if (error != -EAGAIN)
185 return error;
1da177e4
LT		 186 }
187
188 if (in_group_p(inode->i_gid))
189 mode >>= 3;
190 }
191
192 /*
193 * If the DACs are ok we don't need any capability check.
194 */
e6305c43 195 if ((mask & ~mode) == 0)
1da177e4 196 return 0;
5909ccaa
LT		 197 return -EACCES;
198}
199
200/**
b74c79e9 201 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa
LT		 202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
39191628 205 * @flags: IPERM_FLAG_ flags.
5909ccaa
LT		 206 *
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 210 * are used for other things.
211 *
212 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
213 * request cannot be satisfied (eg. requires blocking or too much complexity).
214 * It would then be called again in ref-walk mode.
5909ccaa 215 */
b74c79e9
NP		 216int generic_permission(struct inode *inode, int mask, unsigned int flags,
217 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
5909ccaa
LT		 218{
219 int ret;
220
221 /*
222 * Do the basic POSIX ACL permission checks.
223 */
b74c79e9 224 ret = acl_permission_check(inode, mask, flags, check_acl);
5909ccaa
LT		 225 if (ret != -EACCES)
226 return ret;
1da177e4 227
1da177e4
LT		 228 /*
229 * Read/write DACs are always overridable.
230 * Executable DACs are overridable if at least one exec bit is set.
231 */
f696a365 232 if (!(mask & MAY_EXEC) || execute_ok(inode))
1da177e4
LT		 233 if (capable(CAP_DAC_OVERRIDE))
234 return 0;
235
236 /*
237 * Searching includes executable on directories, else just read.
238 */
7ea66001 239 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
1da177e4
LT		 240 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
241 if (capable(CAP_DAC_READ_SEARCH))
242 return 0;
243
244 return -EACCES;
245}
246
cb23beb5
CH		 247/**
248 * inode_permission - check for access rights to a given inode
249 * @inode: inode to check permission on
250 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
251 *
252 * Used to check for read/write/execute permissions on an inode.
253 * We use "fsuid" for this, letting us set arbitrary permissions
254 * for filesystem access without changing the "normal" uids which
255 * are used for other things.
256 */
f419a2e3 257int inode_permission(struct inode *inode, int mask)
1da177e4 258{
e6305c43 259 int retval;
1da177e4
LT		 260
261 if (mask & MAY_WRITE) {
22590e41 262 umode_t mode = inode->i_mode;
1da177e4
LT		 263
264 /*
265 * Nobody gets write access to a read-only fs.
266 */
267 if (IS_RDONLY(inode) &&
268 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
269 return -EROFS;
270
271 /*
272 * Nobody gets write access to an immutable file.
273 */
274 if (IS_IMMUTABLE(inode))
275 return -EACCES;
276 }
277
acfa4380 278 if (inode->i_op->permission)
b74c79e9 279 retval = inode->i_op->permission(inode, mask, 0);
f696a365 280 else
b74c79e9
NP		 281 retval = generic_permission(inode, mask, 0,
282 inode->i_op->check_acl);
f696a365 283
1da177e4
LT		 284 if (retval)
285 return retval;
286
08ce5f16
SH		 287 retval = devcgroup_inode_permission(inode, mask);
288 if (retval)
289 return retval;
290
d09ca739 291 return security_inode_permission(inode, mask);
1da177e4
LT		 292}
293
8c744fb8
CH		 294/**
295 * file_permission - check for additional access rights to a given file
296 * @file: file to check access rights for
297 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
298 *
299 * Used to check for read/write/execute permissions on an already opened
300 * file.
301 *
302 * Note:
303 * Do not use this function in new code. All access checks should
cb23beb5 304 * be done using inode_permission().
8c744fb8
CH		 305 */
306int file_permission(struct file *file, int mask)
307{
f419a2e3 308 return inode_permission(file->f_path.dentry->d_inode, mask);
8c744fb8
CH		 309}
310
1da177e4
LT		 311/*
312 * get_write_access() gets write permission for a file.
313 * put_write_access() releases this write permission.
314 * This is used for regular files.
315 * We cannot support write (and maybe mmap read-write shared) accesses and
316 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
317 * can have the following values:
318 * 0: no writers, no VM_DENYWRITE mappings
319 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
320 * > 0: (i_writecount) users are writing to the file.
321 *
322 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
323 * except for the cases where we don't hold i_writecount yet. Then we need to
324 * use {get,deny}_write_access() - these functions check the sign and refuse
325 * to do the change if sign is wrong. Exclusion between them is provided by
326 * the inode->i_lock spinlock.
327 */
328
329int get_write_access(struct inode * inode)
330{
331 spin_lock(&inode->i_lock);
332 if (atomic_read(&inode->i_writecount) < 0) {
333 spin_unlock(&inode->i_lock);
334 return -ETXTBSY;
335 }
336 atomic_inc(&inode->i_writecount);
337 spin_unlock(&inode->i_lock);
338
339 return 0;
340}
341
342int deny_write_access(struct file * file)
343{
0f7fc9e4 344 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4
LT		 345
346 spin_lock(&inode->i_lock);
347 if (atomic_read(&inode->i_writecount) > 0) {
348 spin_unlock(&inode->i_lock);
349 return -ETXTBSY;
350 }
351 atomic_dec(&inode->i_writecount);
352 spin_unlock(&inode->i_lock);
353
354 return 0;
355}
356
5dd784d0
JB		 357/**
358 * path_get - get a reference to a path
359 * @path: path to get the reference to
360 *
361 * Given a path increment the reference count to the dentry and the vfsmount.
362 */
363void path_get(struct path *path)
364{
365 mntget(path->mnt);
366 dget(path->dentry);
367}
368EXPORT_SYMBOL(path_get);
369
1d957f9b
JB		 370/**
371 * path_put - put a reference to a path
372 * @path: path to put the reference to
373 *
374 * Given a path decrement the reference count to the dentry and the vfsmount.
375 */
376void path_put(struct path *path)
1da177e4 377{
1d957f9b
JB		 378 dput(path->dentry);
379 mntput(path->mnt);
1da177e4 380}
1d957f9b 381EXPORT_SYMBOL(path_put);
1da177e4 382
31e6b01f
NP		 383/**
384 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
385 * @nd: nameidata pathwalk data to drop
39191628 386 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 387 *
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
390 * to drop out of rcu-walk mode and take normal reference counts on dentries
391 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
392 * refcounts at the last known good point before rcu-walk got stuck, so
393 * ref-walk may continue from there. If this is not successful (eg. a seqcount
394 * has changed), then failure is returned and path walk restarts from the
395 * beginning in ref-walk mode.
396 *
397 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
398 * ref-walk. Must be called from rcu-walk context.
399 */
400static int nameidata_drop_rcu(struct nameidata *nd)
401{
402 struct fs_struct *fs = current->fs;
403 struct dentry *dentry = nd->path.dentry;
5b6ca027 404 int want_root = 0;
31e6b01f
NP		 405
406 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 407 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
408 want_root = 1;
31e6b01f
NP		 409 spin_lock(&fs->lock);
410 if (nd->root.mnt != fs->root.mnt ||
411 nd->root.dentry != fs->root.dentry)
412 goto err_root;
413 }
414 spin_lock(&dentry->d_lock);
415 if (!__d_rcu_to_refcount(dentry, nd->seq))
416 goto err;
417 BUG_ON(nd->inode != dentry->d_inode);
418 spin_unlock(&dentry->d_lock);
5b6ca027 419 if (want_root) {
31e6b01f
NP		 420 path_get(&nd->root);
421 spin_unlock(&fs->lock);
422 }
423 mntget(nd->path.mnt);
424
425 rcu_read_unlock();
426 br_read_unlock(vfsmount_lock);
427 nd->flags &= ~LOOKUP_RCU;
428 return 0;
429err:
430 spin_unlock(&dentry->d_lock);
431err_root:
5b6ca027 432 if (want_root)
31e6b01f
NP		 433 spin_unlock(&fs->lock);
434 return -ECHILD;
435}
436
437/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
438static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
439{
440 if (nd->flags & LOOKUP_RCU)
441 return nameidata_drop_rcu(nd);
442 return 0;
443}
444
445/**
446 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
447 * @nd: nameidata pathwalk data to drop
448 * @dentry: dentry to drop
39191628 449 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 450 *
451 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
452 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
453 * @nd. Must be called from rcu-walk context.
454 */
455static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
456{
457 struct fs_struct *fs = current->fs;
458 struct dentry *parent = nd->path.dentry;
5b6ca027 459 int want_root = 0;
31e6b01f
NP		 460
461 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 462 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
463 want_root = 1;
31e6b01f
NP		 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);
5b6ca027 483 if (want_root) {
31e6b01f
NP		 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:
5b6ca027 497 if (want_root)
31e6b01f
NP		 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{
a7472bab
AV		 505 if (nd->flags & LOOKUP_RCU) {
506 if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) {
507 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 508 if (!(nd->flags & LOOKUP_ROOT))
509 nd->root.mnt = NULL;
a7472bab
AV		 510 rcu_read_unlock();
511 br_read_unlock(vfsmount_lock);
512 return -ECHILD;
513 }
514 }
31e6b01f
NP		 515 return 0;
516}
517
518/**
519 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
520 * @nd: nameidata pathwalk data to drop
39191628 521 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 522 *
523 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
524 * nd->path should be the final element of the lookup, so nd->root is discarded.
525 * Must be called from rcu-walk context.
526 */
527static int nameidata_drop_rcu_last(struct nameidata *nd)
528{
529 struct dentry *dentry = nd->path.dentry;
530
531 BUG_ON(!(nd->flags & LOOKUP_RCU));
532 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 533 if (!(nd->flags & LOOKUP_ROOT))
534 nd->root.mnt = NULL;
31e6b01f
NP		 535 spin_lock(&dentry->d_lock);
536 if (!__d_rcu_to_refcount(dentry, nd->seq))
537 goto err_unlock;
538 BUG_ON(nd->inode != dentry->d_inode);
539 spin_unlock(&dentry->d_lock);
540
541 mntget(nd->path.mnt);
542
543 rcu_read_unlock();
544 br_read_unlock(vfsmount_lock);
545
546 return 0;
547
548err_unlock:
549 spin_unlock(&dentry->d_lock);
550 rcu_read_unlock();
551 br_read_unlock(vfsmount_lock);
552 return -ECHILD;
553}
554
834f2a4a
TM		 555/**
556 * release_open_intent - free up open intent resources
557 * @nd: pointer to nameidata
558 */
559void release_open_intent(struct nameidata *nd)
560{
2dab5974
LT		 561 struct file *file = nd->intent.open.file;
562
563 if (file && !IS_ERR(file)) {
564 if (file->f_path.dentry == NULL)
565 put_filp(file);
566 else
567 fput(file);
568 }
834f2a4a
TM		 569}
570
f60aef7e 571static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 572{
f60aef7e 573 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 574}
575
f5e1c1c1 576static struct dentry *
bcdc5e01
IK		 577do_revalidate(struct dentry *dentry, struct nameidata *nd)
578{
f5e1c1c1 579 int status = d_revalidate(dentry, nd);
bcdc5e01
IK		 580 if (unlikely(status <= 0)) {
581 /*
582 * The dentry failed validation.
583 * If d_revalidate returned 0 attempt to invalidate
584 * the dentry otherwise d_revalidate is asking us
585 * to return a fail status.
586 */
34286d66 587 if (status < 0) {
f5e1c1c1 588 dput(dentry);
34286d66 589 dentry = ERR_PTR(status);
f5e1c1c1
AV		 590 } else if (!d_invalidate(dentry)) {
591 dput(dentry);
592 dentry = NULL;
bcdc5e01
IK		 593 }
594 }
595 return dentry;
596}
597
39159de2 598/*
16c2cd71 599 * handle_reval_path - force revalidation of a dentry
39159de2
JL		 600 *
601 * In some situations the path walking code will trust dentries without
602 * revalidating them. This causes problems for filesystems that depend on
603 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
604 * (which indicates that it's possible for the dentry to go stale), force
605 * a d_revalidate call before proceeding.
606 *
607 * Returns 0 if the revalidation was successful. If the revalidation fails,
608 * either return the error returned by d_revalidate or -ESTALE if the
609 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
610 * invalidate the dentry. It's up to the caller to handle putting references
611 * to the path if necessary.
612 */
16c2cd71 613static inline int handle_reval_path(struct nameidata *nd)
39159de2 614{
16c2cd71 615 struct dentry *dentry = nd->path.dentry;
39159de2 616 int status;
39159de2 617
16c2cd71
AV		 618 if (likely(!(nd->flags & LOOKUP_JUMPED)))
619 return 0;
620
621 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 622 return 0;
623
16c2cd71
AV		 624 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
625 return 0;
626
627 /* Note: we do not d_invalidate() */
34286d66 628 status = d_revalidate(dentry, nd);
39159de2
JL		 629 if (status > 0)
630 return 0;
631
16c2cd71 632 if (!status)
39159de2 633 status = -ESTALE;
16c2cd71 634
39159de2
JL		 635 return status;
636}
637
1da177e4 638/*
b75b5086
AV		 639 * Short-cut version of permission(), for calling on directories
640 * during pathname resolution. Combines parts of permission()
641 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
1da177e4
LT		 642 *
643 * If appropriate, check DAC only. If not appropriate, or
b75b5086 644 * short-cut DAC fails, then call ->permission() to do more
1da177e4
LT		 645 * complete permission check.
646 */
b74c79e9 647static inline int exec_permission(struct inode *inode, unsigned int flags)
1da177e4 648{
5909ccaa 649 int ret;
1da177e4 650
cb9179ea 651 if (inode->i_op->permission) {
b74c79e9
NP		 652 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
653 } else {
654 ret = acl_permission_check(inode, MAY_EXEC, flags,
655 inode->i_op->check_acl);
cb9179ea 656 }
b74c79e9 657 if (likely(!ret))
1da177e4 658 goto ok;
b74c79e9 659 if (ret == -ECHILD)
31e6b01f 660 return ret;
1da177e4 661
f1ac9f6b 662 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
1da177e4
LT		 663 goto ok;
664
5909ccaa 665 return ret;
1da177e4 666ok:
b74c79e9 667 return security_inode_exec_permission(inode, flags);
1da177e4
LT		 668}
669
2a737871
AV		 670static __always_inline void set_root(struct nameidata *nd)
671{
f7ad3c6b
MS		 672 if (!nd->root.mnt)
673 get_fs_root(current->fs, &nd->root);
2a737871
AV		 674}
675
6de88d72
AV		 676static int link_path_walk(const char *, struct nameidata *);
677
31e6b01f
NP		 678static __always_inline void set_root_rcu(struct nameidata *nd)
679{
680 if (!nd->root.mnt) {
681 struct fs_struct *fs = current->fs;
c28cc364
NP		 682 unsigned seq;
683
684 do {
685 seq = read_seqcount_begin(&fs->seq);
686 nd->root = fs->root;
687 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 688 }
689}
690
f1662356 691static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 692{
31e6b01f
NP		 693 int ret;
694
1da177e4
LT		 695 if (IS_ERR(link))
696 goto fail;
697
698 if (*link == '/') {
2a737871 699 set_root(nd);
1d957f9b 700 path_put(&nd->path);
2a737871
AV		 701 nd->path = nd->root;
702 path_get(&nd->root);
16c2cd71 703 nd->flags |= LOOKUP_JUMPED;
1da177e4 704 }
31e6b01f 705 nd->inode = nd->path.dentry->d_inode;
b4091d5f 706
31e6b01f
NP		 707 ret = link_path_walk(link, nd);
708 return ret;
1da177e4 709fail:
1d957f9b 710 path_put(&nd->path);
1da177e4
LT		 711 return PTR_ERR(link);
712}
713
1d957f9b 714static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 715{
716 dput(path->dentry);
4ac91378 717 if (path->mnt != nd->path.mnt)
051d3812
IK		 718 mntput(path->mnt);
719}
720
7b9337aa
NP		 721static inline void path_to_nameidata(const struct path *path,
722 struct nameidata *nd)
051d3812 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 733static __always_inline int
7b9337aa 734__do_follow_link(const struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 735{
736 int error;
7b9337aa 737 struct dentry *dentry = link->dentry;
1da177e4 738
844a3917
AV		 739 BUG_ON(nd->flags & LOOKUP_RCU);
740
7b9337aa 741 touch_atime(link->mnt, dentry);
1da177e4 742 nd_set_link(nd, NULL);
cd4e91d3 743
87556ef1
DH		 744 if (link->mnt == nd->path.mnt)
745 mntget(link->mnt);
31e6b01f 746
36f3b4f6
AV		 747 error = security_inode_follow_link(link->dentry, nd);
748 if (error) {
749 *p = ERR_PTR(error); /* no ->put_link(), please */
750 path_put(&nd->path);
751 return error;
752 }
753
86acdca1 754 nd->last_type = LAST_BIND;
def4af30
AV		 755 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
756 error = PTR_ERR(*p);
757 if (!IS_ERR(*p)) {
1da177e4 758 char *s = nd_get_link(nd);
cc314eef 759 error = 0;
1da177e4
LT		 760 if (s)
761 error = __vfs_follow_link(nd, s);
16c2cd71
AV		 762 else if (nd->last_type == LAST_BIND)
763 nd->flags |= LOOKUP_JUMPED;
1da177e4 764 }
1da177e4
LT		 765 return error;
766}
767
768/*
769 * This limits recursive symlink follows to 8, while
770 * limiting consecutive symlinks to 40.
771 *
772 * Without that kind of total limit, nasty chains of consecutive
773 * symlinks can cause almost arbitrarily long lookups.
774 */
3abb17e8 775static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd)
1da177e4 776{
def4af30 777 void *cookie;
1da177e4 778 int err = -ELOOP;
844a3917
AV		 779
780 /* We drop rcu-walk here */
781 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
782 return -ECHILD;
3abb17e8 783 BUG_ON(inode != path->dentry->d_inode);
844a3917 784
1da177e4
LT		 785 if (current->link_count >= MAX_NESTED_LINKS)
786 goto loop;
787 if (current->total_link_count >= 40)
788 goto loop;
789 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
790 cond_resched();
1da177e4
LT		 791 current->link_count++;
792 current->total_link_count++;
793 nd->depth++;
def4af30
AV		 794 err = __do_follow_link(path, nd, &cookie);
795 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
796 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
258fa999 797 path_put(path);
839d9f93
AV		 798 current->link_count--;
799 nd->depth--;
1da177e4
LT		 800 return err;
801loop:
1d957f9b
JB		 802 path_put_conditional(path, nd);
803 path_put(&nd->path);
1da177e4
LT		 804 return err;
805}
806
31e6b01f
NP		 807static int follow_up_rcu(struct path *path)
808{
809 struct vfsmount *parent;
810 struct dentry *mountpoint;
811
812 parent = path->mnt->mnt_parent;
813 if (parent == path->mnt)
814 return 0;
815 mountpoint = path->mnt->mnt_mountpoint;
816 path->dentry = mountpoint;
817 path->mnt = parent;
818 return 1;
819}
820
bab77ebf 821int follow_up(struct path *path)
1da177e4
LT		 822{
823 struct vfsmount *parent;
824 struct dentry *mountpoint;
99b7db7b
NP		 825
826 br_read_lock(vfsmount_lock);
bab77ebf
AV		 827 parent = path->mnt->mnt_parent;
828 if (parent == path->mnt) {
99b7db7b 829 br_read_unlock(vfsmount_lock);
1da177e4
LT		 830 return 0;
831 }
832 mntget(parent);
bab77ebf 833 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 834 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 835 dput(path->dentry);
836 path->dentry = mountpoint;
837 mntput(path->mnt);
838 path->mnt = parent;
1da177e4
LT		 839 return 1;
840}
841
b5c84bf6 842/*
9875cf80
DH		 843 * Perform an automount
844 * - return -EISDIR to tell follow_managed() to stop and return the path we
845 * were called with.
1da177e4 846 */
9875cf80
DH		 847static int follow_automount(struct path *path, unsigned flags,
848 bool *need_mntput)
31e6b01f 849{
9875cf80 850 struct vfsmount *mnt;
ea5b778a 851 int err;
9875cf80
DH		 852
853 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
854 return -EREMOTE;
855
6f45b656
DH		 856 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
857 * and this is the terminal part of the path.
858 */
859 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
860 return -EISDIR; /* we actually want to stop here */
861
9875cf80
DH		 862 /* We want to mount if someone is trying to open/create a file of any
863 * type under the mountpoint, wants to traverse through the mountpoint
864 * or wants to open the mounted directory.
865 *
866 * We don't want to mount if someone's just doing a stat and they've
867 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
868 * appended a '/' to the name.
869 */
870 if (!(flags & LOOKUP_FOLLOW) &&
871 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
872 LOOKUP_OPEN | LOOKUP_CREATE)))
873 return -EISDIR;
874
875 current->total_link_count++;
876 if (current->total_link_count >= 40)
877 return -ELOOP;
878
879 mnt = path->dentry->d_op->d_automount(path);
880 if (IS_ERR(mnt)) {
881 /*
882 * The filesystem is allowed to return -EISDIR here to indicate
883 * it doesn't want to automount. For instance, autofs would do
884 * this so that its userspace daemon can mount on this dentry.
885 *
886 * However, we can only permit this if it's a terminal point in
887 * the path being looked up; if it wasn't then the remainder of
888 * the path is inaccessible and we should say so.
889 */
890 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
891 return -EREMOTE;
892 return PTR_ERR(mnt);
31e6b01f 893 }
ea5b778a 894
9875cf80
DH		 895 if (!mnt) /* mount collision */
896 return 0;
31e6b01f 897
19a167af 898 err = finish_automount(mnt, path);
9875cf80 899
ea5b778a
DH		 900 switch (err) {
901 case -EBUSY:
902 /* Someone else made a mount here whilst we were busy */
19a167af 903 return 0;
ea5b778a 904 case 0:
ea5b778a
DH		 905 dput(path->dentry);
906 if (*need_mntput)
907 mntput(path->mnt);
908 path->mnt = mnt;
909 path->dentry = dget(mnt->mnt_root);
910 *need_mntput = true;
911 return 0;
19a167af
AV		 912 default:
913 return err;
ea5b778a 914 }
19a167af 915
463ffb2e
AV		 916}
917
9875cf80
DH		 918/*
919 * Handle a dentry that is managed in some way.
cc53ce53 920 * - Flagged for transit management (autofs)
9875cf80
DH		 921 * - Flagged as mountpoint
922 * - Flagged as automount point
923 *
924 * This may only be called in refwalk mode.
925 *
926 * Serialization is taken care of in namespace.c
927 */
928static int follow_managed(struct path *path, unsigned flags)
1da177e4 929{
9875cf80
DH		 930 unsigned managed;
931 bool need_mntput = false;
932 int ret;
933
934 /* Given that we're not holding a lock here, we retain the value in a
935 * local variable for each dentry as we look at it so that we don't see
936 * the components of that value change under us */
937 while (managed = ACCESS_ONCE(path->dentry->d_flags),
938 managed &= DCACHE_MANAGED_DENTRY,
939 unlikely(managed != 0)) {
cc53ce53
DH		 940 /* Allow the filesystem to manage the transit without i_mutex
941 * being held. */
942 if (managed & DCACHE_MANAGE_TRANSIT) {
943 BUG_ON(!path->dentry->d_op);
944 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 945 ret = path->dentry->d_op->d_manage(path->dentry,
946 false, false);
cc53ce53
DH		 947 if (ret < 0)
948 return ret == -EISDIR ? 0 : ret;
949 }
950
9875cf80
DH		 951 /* Transit to a mounted filesystem. */
952 if (managed & DCACHE_MOUNTED) {
953 struct vfsmount *mounted = lookup_mnt(path);
954 if (mounted) {
955 dput(path->dentry);
956 if (need_mntput)
957 mntput(path->mnt);
958 path->mnt = mounted;
959 path->dentry = dget(mounted->mnt_root);
960 need_mntput = true;
961 continue;
962 }
963
964 /* Something is mounted on this dentry in another
965 * namespace and/or whatever was mounted there in this
966 * namespace got unmounted before we managed to get the
967 * vfsmount_lock */
968 }
969
970 /* Handle an automount point */
971 if (managed & DCACHE_NEED_AUTOMOUNT) {
972 ret = follow_automount(path, flags, &need_mntput);
973 if (ret < 0)
974 return ret == -EISDIR ? 0 : ret;
975 continue;
976 }
977
978 /* We didn't change the current path point */
979 break;
1da177e4 980 }
9875cf80 981 return 0;
1da177e4
LT		 982}
983
cc53ce53 984int follow_down_one(struct path *path)
1da177e4
LT		 985{
986 struct vfsmount *mounted;
987
1c755af4 988 mounted = lookup_mnt(path);
1da177e4 989 if (mounted) {
9393bd07
AV		 990 dput(path->dentry);
991 mntput(path->mnt);
992 path->mnt = mounted;
993 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 994 return 1;
995 }
996 return 0;
997}
998
9875cf80
DH		 999/*
1000 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
cc53ce53 1001 * meet a managed dentry and we're not walking to "..". True is returned to
9875cf80
DH		 1002 * continue, false to abort.
1003 */
1004static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1005 struct inode **inode, bool reverse_transit)
1006{
1007 while (d_mountpoint(path->dentry)) {
1008 struct vfsmount *mounted;
ab90911f
DH		 1009 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1010 !reverse_transit &&
1011 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1012 return false;
9875cf80
DH		 1013 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1014 if (!mounted)
1015 break;
1016 path->mnt = mounted;
1017 path->dentry = mounted->mnt_root;
1018 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1019 *inode = path->dentry->d_inode;
1020 }
1021
1022 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1023 return reverse_transit;
1024 return true;
1025}
1026
31e6b01f
NP		 1027static int follow_dotdot_rcu(struct nameidata *nd)
1028{
1029 struct inode *inode = nd->inode;
1030
1031 set_root_rcu(nd);
1032
9875cf80 1033 while (1) {
31e6b01f
NP		 1034 if (nd->path.dentry == nd->root.dentry &&
1035 nd->path.mnt == nd->root.mnt) {
1036 break;
1037 }
1038 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1039 struct dentry *old = nd->path.dentry;
1040 struct dentry *parent = old->d_parent;
1041 unsigned seq;
1042
1043 seq = read_seqcount_begin(&parent->d_seq);
1044 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 1045 goto failed;
31e6b01f
NP		 1046 inode = parent->d_inode;
1047 nd->path.dentry = parent;
1048 nd->seq = seq;
1049 break;
1050 }
1051 if (!follow_up_rcu(&nd->path))
1052 break;
1053 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1054 inode = nd->path.dentry->d_inode;
1055 }
9875cf80 1056 __follow_mount_rcu(nd, &nd->path, &inode, true);
31e6b01f 1057 nd->inode = inode;
31e6b01f 1058 return 0;
ef7562d5
AV		 1059
1060failed:
1061 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1062 if (!(nd->flags & LOOKUP_ROOT))
1063 nd->root.mnt = NULL;
ef7562d5
AV		 1064 rcu_read_unlock();
1065 br_read_unlock(vfsmount_lock);
1066 return -ECHILD;
31e6b01f
NP		 1067}
1068
cc53ce53
DH		 1069/*
1070 * Follow down to the covering mount currently visible to userspace. At each
1071 * point, the filesystem owning that dentry may be queried as to whether the
1072 * caller is permitted to proceed or not.
1073 *
1074 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1075 * being true).
1076 */
1077int follow_down(struct path *path, bool mounting_here)
1078{
1079 unsigned managed;
1080 int ret;
1081
1082 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1083 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1084 /* Allow the filesystem to manage the transit without i_mutex
1085 * being held.
1086 *
1087 * We indicate to the filesystem if someone is trying to mount
1088 * something here. This gives autofs the chance to deny anyone
1089 * other than its daemon the right to mount on its
1090 * superstructure.
1091 *
1092 * The filesystem may sleep at this point.
1093 */
1094 if (managed & DCACHE_MANAGE_TRANSIT) {
1095 BUG_ON(!path->dentry->d_op);
1096 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 1097 ret = path->dentry->d_op->d_manage(
1098 path->dentry, mounting_here, false);
cc53ce53
DH		 1099 if (ret < 0)
1100 return ret == -EISDIR ? 0 : ret;
1101 }
1102
1103 /* Transit to a mounted filesystem. */
1104 if (managed & DCACHE_MOUNTED) {
1105 struct vfsmount *mounted = lookup_mnt(path);
1106 if (!mounted)
1107 break;
1108 dput(path->dentry);
1109 mntput(path->mnt);
1110 path->mnt = mounted;
1111 path->dentry = dget(mounted->mnt_root);
1112 continue;
1113 }
1114
1115 /* Don't handle automount points here */
1116 break;
1117 }
1118 return 0;
1119}
1120
9875cf80
DH		 1121/*
1122 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1123 */
1124static void follow_mount(struct path *path)
1125{
1126 while (d_mountpoint(path->dentry)) {
1127 struct vfsmount *mounted = lookup_mnt(path);
1128 if (!mounted)
1129 break;
1130 dput(path->dentry);
1131 mntput(path->mnt);
1132 path->mnt = mounted;
1133 path->dentry = dget(mounted->mnt_root);
1134 }
1135}
1136
31e6b01f 1137static void follow_dotdot(struct nameidata *nd)
1da177e4 1138{
2a737871 1139 set_root(nd);
e518ddb7 1140
1da177e4 1141 while(1) {
4ac91378 1142 struct dentry *old = nd->path.dentry;
1da177e4 1143
2a737871
AV		 1144 if (nd->path.dentry == nd->root.dentry &&
1145 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1146 break;
1147 }
4ac91378 1148 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1149 /* rare case of legitimate dget_parent()... */
1150 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1151 dput(old);
1152 break;
1153 }
3088dd70 1154 if (!follow_up(&nd->path))
1da177e4 1155 break;
1da177e4 1156 }
79ed0226 1157 follow_mount(&nd->path);
31e6b01f 1158 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1159}
1160
baa03890
NP		 1161/*
1162 * Allocate a dentry with name and parent, and perform a parent
1163 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1164 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1165 * have verified that no child exists while under i_mutex.
1166 */
1167static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1168 struct qstr *name, struct nameidata *nd)
1169{
1170 struct inode *inode = parent->d_inode;
1171 struct dentry *dentry;
1172 struct dentry *old;
1173
1174 /* Don't create child dentry for a dead directory. */
1175 if (unlikely(IS_DEADDIR(inode)))
1176 return ERR_PTR(-ENOENT);
1177
1178 dentry = d_alloc(parent, name);
1179 if (unlikely(!dentry))
1180 return ERR_PTR(-ENOMEM);
1181
1182 old = inode->i_op->lookup(inode, dentry, nd);
1183 if (unlikely(old)) {
1184 dput(dentry);
1185 dentry = old;
1186 }
1187 return dentry;
1188}
1189
1da177e4
LT		 1190/*
1191 * It's more convoluted than I'd like it to be, but... it's still fairly
1192 * small and for now I'd prefer to have fast path as straight as possible.
1193 * It _is_ time-critical.
1194 */
1195static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1196 struct path *path, struct inode **inode)
1da177e4 1197{
4ac91378 1198 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1199 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1200 int need_reval = 1;
1201 int status = 1;
9875cf80
DH		 1202 int err;
1203
b04f784e
NP		 1204 /*
1205 * Rename seqlock is not required here because in the off chance
1206 * of a false negative due to a concurrent rename, we're going to
1207 * do the non-racy lookup, below.
1208 */
31e6b01f
NP		 1209 if (nd->flags & LOOKUP_RCU) {
1210 unsigned seq;
31e6b01f
NP		 1211 *inode = nd->inode;
1212 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1213 if (!dentry)
1214 goto unlazy;
1215
31e6b01f
NP		 1216 /* Memory barrier in read_seqcount_begin of child is enough */
1217 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1218 return -ECHILD;
31e6b01f 1219 nd->seq = seq;
5a18fff2 1220
24643087 1221 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1222 status = d_revalidate(dentry, nd);
1223 if (unlikely(status <= 0)) {
1224 if (status != -ECHILD)
1225 need_reval = 0;
1226 goto unlazy;
1227 }
24643087 1228 }
31e6b01f
NP		 1229 path->mnt = mnt;
1230 path->dentry = dentry;
9875cf80
DH		 1231 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1232 return 0;
5a18fff2
AV		 1233unlazy:
1234 if (dentry) {
1235 if (nameidata_dentry_drop_rcu(nd, dentry))
1236 return -ECHILD;
1237 } else {
1238 if (nameidata_drop_rcu(nd))
1239 return -ECHILD;
1240 }
1241 } else {
1242 dentry = __d_lookup(parent, name);
9875cf80 1243 }
5a18fff2
AV		 1244
1245retry:
1246 if (unlikely(!dentry)) {
1247 struct inode *dir = parent->d_inode;
1248 BUG_ON(nd->inode != dir);
1249
1250 mutex_lock(&dir->i_mutex);
1251 dentry = d_lookup(parent, name);
1252 if (likely(!dentry)) {
1253 dentry = d_alloc_and_lookup(parent, name, nd);
1254 if (IS_ERR(dentry)) {
1255 mutex_unlock(&dir->i_mutex);
1256 return PTR_ERR(dentry);
1257 }
1258 /* known good */
1259 need_reval = 0;
1260 status = 1;
1261 }
1262 mutex_unlock(&dir->i_mutex);
1263 }
1264 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1265 status = d_revalidate(dentry, nd);
1266 if (unlikely(status <= 0)) {
1267 if (status < 0) {
1268 dput(dentry);
1269 return status;
1270 }
1271 if (!d_invalidate(dentry)) {
1272 dput(dentry);
1273 dentry = NULL;
1274 need_reval = 1;
1275 goto retry;
1276 }
24643087 1277 }
5a18fff2 1278
9875cf80
DH		 1279 path->mnt = mnt;
1280 path->dentry = dentry;
1281 err = follow_managed(path, nd->flags);
89312214
IK		 1282 if (unlikely(err < 0)) {
1283 path_put_conditional(path, nd);
9875cf80 1284 return err;
89312214 1285 }
9875cf80 1286 *inode = path->dentry->d_inode;
1da177e4 1287 return 0;
1da177e4
LT		 1288}
1289
52094c8a
AV		 1290static inline int may_lookup(struct nameidata *nd)
1291{
1292 if (nd->flags & LOOKUP_RCU) {
1293 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1294 if (err != -ECHILD)
1295 return err;
1296 if (nameidata_drop_rcu(nd))
1297 return -ECHILD;
1298 }
1299 return exec_permission(nd->inode, 0);
1300}
1301
9856fa1b
AV		 1302static inline int handle_dots(struct nameidata *nd, int type)
1303{
1304 if (type == LAST_DOTDOT) {
1305 if (nd->flags & LOOKUP_RCU) {
1306 if (follow_dotdot_rcu(nd))
1307 return -ECHILD;
1308 } else
1309 follow_dotdot(nd);
1310 }
1311 return 0;
1312}
1313
951361f9
AV		 1314static void terminate_walk(struct nameidata *nd)
1315{
1316 if (!(nd->flags & LOOKUP_RCU)) {
1317 path_put(&nd->path);
1318 } else {
1319 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1320 if (!(nd->flags & LOOKUP_ROOT))
1321 nd->root.mnt = NULL;
951361f9
AV		 1322 rcu_read_unlock();
1323 br_read_unlock(vfsmount_lock);
1324 }
1325}
1326
1da177e4
LT		 1327/*
1328 * Name resolution.
ea3834d9
PM		 1329 * This is the basic name resolution function, turning a pathname into
1330 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1331 *
ea3834d9
PM		 1332 * Returns 0 and nd will have valid dentry and mnt on success.
1333 * Returns error and drops reference to input namei data on failure.
1da177e4 1334 */
6de88d72 1335static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1336{
1337 struct path next;
1da177e4
LT		 1338 int err;
1339 unsigned int lookup_flags = nd->flags;
1340
1341 while (*name=='/')
1342 name++;
1343 if (!*name)
086e183a 1344 return 0;
1da177e4 1345
1da177e4 1346 if (nd->depth)
f55eab82 1347 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 1348
1349 /* At this point we know we have a real path component. */
1350 for(;;) {
31e6b01f 1351 struct inode *inode;
1da177e4
LT		 1352 unsigned long hash;
1353 struct qstr this;
1354 unsigned int c;
fe479a58 1355 int type;
1da177e4 1356
cdce5d6b 1357 nd->flags |= LOOKUP_CONTINUE;
52094c8a
AV		 1358
1359 err = may_lookup(nd);
1da177e4
LT		 1360 if (err)
1361 break;
1362
1363 this.name = name;
1364 c = *(const unsigned char *)name;
1365
1366 hash = init_name_hash();
1367 do {
1368 name++;
1369 hash = partial_name_hash(c, hash);
1370 c = *(const unsigned char *)name;
1371 } while (c && (c != '/'));
1372 this.len = name - (const char *) this.name;
1373 this.hash = end_name_hash(hash);
1374
fe479a58
AV		 1375 type = LAST_NORM;
1376 if (this.name[0] == '.') switch (this.len) {
1377 case 2:
16c2cd71 1378 if (this.name[1] == '.') {
fe479a58 1379 type = LAST_DOTDOT;
16c2cd71
AV		 1380 nd->flags |= LOOKUP_JUMPED;
1381 }
fe479a58
AV		 1382 break;
1383 case 1:
1384 type = LAST_DOT;
1385 }
5a202bcd
AV		 1386 if (likely(type == LAST_NORM)) {
1387 struct dentry *parent = nd->path.dentry;
16c2cd71 1388 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1389 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1390 err = parent->d_op->d_hash(parent, nd->inode,
1391 &this);
1392 if (err < 0)
1393 break;
1394 }
1395 }
fe479a58 1396
1da177e4
LT		 1397 /* remove trailing slashes? */
1398 if (!c)
1399 goto last_component;
1400 while (*++name == '/');
1401 if (!*name)
1402 goto last_with_slashes;
1403
1404 /*
1405 * "." and ".." are special - ".." especially so because it has
1406 * to be able to know about the current root directory and
1407 * parent relationships.
1408 */
fe479a58 1409 if (unlikely(type != LAST_NORM)) {
ef7562d5
AV		 1410 if (handle_dots(nd, type))
1411 return -ECHILD;
fe479a58 1412 continue;
1da177e4 1413 }
fe479a58 1414
1da177e4 1415 /* This does the actual lookups.. */
31e6b01f 1416 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1417 if (err)
1418 break;
1da177e4 1419
7bc055d1 1420 if (inode && inode->i_op->follow_link) {
3abb17e8 1421 err = do_follow_link(inode, &next, nd);
1da177e4 1422 if (err)
a7472bab 1423 return err;
31e6b01f 1424 nd->inode = nd->path.dentry->d_inode;
31e6b01f 1425 } else {
09dd17d3 1426 path_to_nameidata(&next, nd);
31e6b01f
NP		 1427 nd->inode = inode;
1428 }
7bc055d1
AV		 1429 err = -ENOENT;
1430 if (!nd->inode)
1431 break;
1da177e4 1432 err = -ENOTDIR;
31e6b01f 1433 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1434 break;
1435 continue;
1436 /* here ends the main loop */
1437
1438last_with_slashes:
1439 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1440last_component:
f55eab82
TM		 1441 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1442 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 1443 if (lookup_flags & LOOKUP_PARENT)
1444 goto lookup_parent;
ef7562d5
AV		 1445 if (unlikely(type != LAST_NORM))
1446 return handle_dots(nd, type);
31e6b01f 1447 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1448 if (err)
1449 break;
db372915
DH		 1450 if (inode && unlikely(inode->i_op->follow_link) &&
1451 (lookup_flags & LOOKUP_FOLLOW)) {
3abb17e8 1452 err = do_follow_link(inode, &next, nd);
1da177e4 1453 if (err)
a7472bab 1454 return err;
31e6b01f
NP		 1455 nd->inode = nd->path.dentry->d_inode;
1456 } else {
09dd17d3 1457 path_to_nameidata(&next, nd);
31e6b01f
NP		 1458 nd->inode = inode;
1459 }
1da177e4 1460 err = -ENOENT;
31e6b01f 1461 if (!nd->inode)
1da177e4
LT		 1462 break;
1463 if (lookup_flags & LOOKUP_DIRECTORY) {
1464 err = -ENOTDIR;
31e6b01f 1465 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1466 break;
1467 }
086e183a 1468 return 0;
1da177e4
LT		 1469lookup_parent:
1470 nd->last = this;
fe479a58 1471 nd->last_type = type;
1da177e4 1472 return 0;
1da177e4 1473 }
951361f9 1474 terminate_walk(nd);
1da177e4
LT		 1475 return err;
1476}
1477
70e9b357
AV		 1478static int path_init(int dfd, const char *name, unsigned int flags,
1479 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1480{
1481 int retval = 0;
1482 int fput_needed;
1483 struct file *file;
1484
1485 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1486 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1487 nd->depth = 0;
5b6ca027
AV		 1488 if (flags & LOOKUP_ROOT) {
1489 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1490 if (*name) {
1491 if (!inode->i_op->lookup)
1492 return -ENOTDIR;
1493 retval = inode_permission(inode, MAY_EXEC);
1494 if (retval)
1495 return retval;
1496 }
5b6ca027
AV		 1497 nd->path = nd->root;
1498 nd->inode = inode;
1499 if (flags & LOOKUP_RCU) {
1500 br_read_lock(vfsmount_lock);
1501 rcu_read_lock();
1502 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1503 } else {
1504 path_get(&nd->path);
1505 }
1506 return 0;
1507 }
1508
31e6b01f 1509 nd->root.mnt = NULL;
31e6b01f
NP		 1510
1511 if (*name=='/') {
e41f7d4e
AV		 1512 if (flags & LOOKUP_RCU) {
1513 br_read_lock(vfsmount_lock);
1514 rcu_read_lock();
1515 set_root_rcu(nd);
1516 } else {
1517 set_root(nd);
1518 path_get(&nd->root);
1519 }
1520 nd->path = nd->root;
31e6b01f 1521 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1522 if (flags & LOOKUP_RCU) {
1523 struct fs_struct *fs = current->fs;
1524 unsigned seq;
31e6b01f 1525
e41f7d4e
AV		 1526 br_read_lock(vfsmount_lock);
1527 rcu_read_lock();
c28cc364 1528
e41f7d4e
AV		 1529 do {
1530 seq = read_seqcount_begin(&fs->seq);
1531 nd->path = fs->pwd;
1532 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1533 } while (read_seqcount_retry(&fs->seq, seq));
1534 } else {
1535 get_fs_pwd(current->fs, &nd->path);
1536 }
31e6b01f
NP		 1537 } else {
1538 struct dentry *dentry;
1539
1540 file = fget_light(dfd, &fput_needed);
1541 retval = -EBADF;
1542 if (!file)
1543 goto out_fail;
1544
1545 dentry = file->f_path.dentry;
1546
1547 retval = -ENOTDIR;
1548 if (!S_ISDIR(dentry->d_inode->i_mode))
1549 goto fput_fail;
1550
1551 retval = file_permission(file, MAY_EXEC);
1552 if (retval)
1553 goto fput_fail;
1554
1555 nd->path = file->f_path;
e41f7d4e
AV		 1556 if (flags & LOOKUP_RCU) {
1557 if (fput_needed)
70e9b357 1558 *fp = file;
e41f7d4e
AV		 1559 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1560 br_read_lock(vfsmount_lock);
1561 rcu_read_lock();
1562 } else {
1563 path_get(&file->f_path);
1564 fput_light(file, fput_needed);
1565 }
31e6b01f 1566 }
31e6b01f 1567
31e6b01f 1568 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1569 return 0;
2dfdd266 1570
9b4a9b14
AV		 1571fput_fail:
1572 fput_light(file, fput_needed);
1573out_fail:
1574 return retval;
1575}
1576
1577/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1578static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1579 unsigned int flags, struct nameidata *nd)
1580{
70e9b357 1581 struct file *base = NULL;
31e6b01f
NP		 1582 int retval;
1583
1584 /*
1585 * Path walking is largely split up into 2 different synchronisation
1586 * schemes, rcu-walk and ref-walk (explained in
1587 * Documentation/filesystems/path-lookup.txt). These share much of the
1588 * path walk code, but some things particularly setup, cleanup, and
1589 * following mounts are sufficiently divergent that functions are
1590 * duplicated. Typically there is a function foo(), and its RCU
1591 * analogue, foo_rcu().
1592 *
1593 * -ECHILD is the error number of choice (just to avoid clashes) that
1594 * is returned if some aspect of an rcu-walk fails. Such an error must
1595 * be handled by restarting a traditional ref-walk (which will always
1596 * be able to complete).
1597 */
70e9b357 1598 retval = path_init(dfd, name, flags, nd, &base);
ee0827cd 1599
31e6b01f
NP		 1600 if (unlikely(retval))
1601 return retval;
ee0827cd
AV		 1602
1603 current->total_link_count = 0;
1604 retval = link_path_walk(name, nd);
1605
1606 if (nd->flags & LOOKUP_RCU) {
4455ca62
AV		 1607 /* went all way through without dropping RCU */
1608 BUG_ON(retval);
1609 if (nameidata_drop_rcu_last(nd))
1610 retval = -ECHILD;
ee0827cd
AV		 1611 }
1612
16c2cd71
AV		 1613 if (!retval)
1614 retval = handle_reval_path(nd);
1615
70e9b357
AV		 1616 if (base)
1617 fput(base);
ee0827cd 1618
5b6ca027 1619 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1620 path_put(&nd->root);
1621 nd->root.mnt = NULL;
1622 }
ee0827cd
AV		 1623 return retval;
1624}
31e6b01f 1625
ee0827cd
AV		 1626static int do_path_lookup(int dfd, const char *name,
1627 unsigned int flags, struct nameidata *nd)
1628{
1629 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1630 if (unlikely(retval == -ECHILD))
1631 retval = path_lookupat(dfd, name, flags, nd);
1632 if (unlikely(retval == -ESTALE))
1633 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1634
1635 if (likely(!retval)) {
1636 if (unlikely(!audit_dummy_context())) {
1637 if (nd->path.dentry && nd->inode)
1638 audit_inode(name, nd->path.dentry);
1639 }
1640 }
170aa3d0 1641 return retval;
1da177e4
LT		 1642}
1643
c9c6cac0 1644int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1645{
c9c6cac0 1646 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1647}
1648
d1811465
AV		 1649int kern_path(const char *name, unsigned int flags, struct path *path)
1650{
1651 struct nameidata nd;
1652 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1653 if (!res)
1654 *path = nd.path;
1655 return res;
1656}
1657
16f18200
JJS		 1658/**
1659 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1660 * @dentry: pointer to dentry of the base directory
1661 * @mnt: pointer to vfs mount of the base directory
1662 * @name: pointer to file name
1663 * @flags: lookup flags
1664 * @nd: pointer to nameidata
1665 */
1666int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1667 const char *name, unsigned int flags,
1668 struct nameidata *nd)
1669{
5b6ca027
AV		 1670 nd->root.dentry = dentry;
1671 nd->root.mnt = mnt;
1672 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1673 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
16f18200
JJS		 1674}
1675
eead1911
CH		 1676static struct dentry *__lookup_hash(struct qstr *name,
1677 struct dentry *base, struct nameidata *nd)
1da177e4 1678{
81fca444 1679 struct inode *inode = base->d_inode;
057f6c01 1680 struct dentry *dentry;
1da177e4
LT		 1681 int err;
1682
b74c79e9 1683 err = exec_permission(inode, 0);
81fca444
CH		 1684 if (err)
1685 return ERR_PTR(err);
1da177e4 1686
b04f784e
NP		 1687 /*
1688 * Don't bother with __d_lookup: callers are for creat as
1689 * well as unlink, so a lot of the time it would cost
1690 * a double lookup.
6e6b1bd1 1691 */
b04f784e 1692 dentry = d_lookup(base, name);
6e6b1bd1 1693
fb045adb 1694 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1695 dentry = do_revalidate(dentry, nd);
1696
baa03890
NP		 1697 if (!dentry)
1698 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1699
1da177e4
LT		 1700 return dentry;
1701}
1702
057f6c01
JM		 1703/*
1704 * Restricted form of lookup. Doesn't follow links, single-component only,
1705 * needs parent already locked. Doesn't follow mounts.
1706 * SMP-safe.
1707 */
eead1911 1708static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1709{
4ac91378 1710 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1711}
1712
eead1911 1713/**
a6b91919 1714 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1715 * @name: pathname component to lookup
1716 * @base: base directory to lookup from
1717 * @len: maximum length @len should be interpreted to
1718 *
a6b91919
RD		 1719 * Note that this routine is purely a helper for filesystem usage and should
1720 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1721 * nameidata argument is passed to the filesystem methods and a filesystem
1722 * using this helper needs to be prepared for that.
1723 */
057f6c01
JM		 1724struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1725{
057f6c01 1726 struct qstr this;
6a96ba54
AV		 1727 unsigned long hash;
1728 unsigned int c;
057f6c01 1729
2f9092e1
DW		 1730 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1731
6a96ba54
AV		 1732 this.name = name;
1733 this.len = len;
1734 if (!len)
1735 return ERR_PTR(-EACCES);
1736
1737 hash = init_name_hash();
1738 while (len--) {
1739 c = *(const unsigned char *)name++;
1740 if (c == '/' || c == '\0')
1741 return ERR_PTR(-EACCES);
1742 hash = partial_name_hash(c, hash);
1743 }
1744 this.hash = end_name_hash(hash);
5a202bcd
AV		 1745 /*
1746 * See if the low-level filesystem might want
1747 * to use its own hash..
1748 */
1749 if (base->d_flags & DCACHE_OP_HASH) {
1750 int err = base->d_op->d_hash(base, base->d_inode, &this);
1751 if (err < 0)
1752 return ERR_PTR(err);
1753 }
eead1911 1754
49705b77 1755 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1756}
1757
2d8f3038
AV		 1758int user_path_at(int dfd, const char __user *name, unsigned flags,
1759 struct path *path)
1da177e4 1760{
2d8f3038 1761 struct nameidata nd;
1da177e4
LT		 1762 char *tmp = getname(name);
1763 int err = PTR_ERR(tmp);
1da177e4 1764 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1765
1766 BUG_ON(flags & LOOKUP_PARENT);
1767
1768 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1769 putname(tmp);
2d8f3038
AV		 1770 if (!err)
1771 *path = nd.path;
1da177e4
LT		 1772 }
1773 return err;
1774}
1775
2ad94ae6
AV		 1776static int user_path_parent(int dfd, const char __user *path,
1777 struct nameidata *nd, char **name)
1778{
1779 char *s = getname(path);
1780 int error;
1781
1782 if (IS_ERR(s))
1783 return PTR_ERR(s);
1784
1785 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1786 if (error)
1787 putname(s);
1788 else
1789 *name = s;
1790
1791 return error;
1792}
1793
1da177e4
LT		 1794/*
1795 * It's inline, so penalty for filesystems that don't use sticky bit is
1796 * minimal.
1797 */
1798static inline int check_sticky(struct inode *dir, struct inode *inode)
1799{
da9592ed
DH		 1800 uid_t fsuid = current_fsuid();
1801
1da177e4
LT		 1802 if (!(dir->i_mode & S_ISVTX))
1803 return 0;
da9592ed 1804 if (inode->i_uid == fsuid)
1da177e4 1805 return 0;
da9592ed 1806 if (dir->i_uid == fsuid)
1da177e4
LT		 1807 return 0;
1808 return !capable(CAP_FOWNER);
1809}
1810
1811/*
1812 * Check whether we can remove a link victim from directory dir, check
1813 * whether the type of victim is right.
1814 * 1. We can't do it if dir is read-only (done in permission())
1815 * 2. We should have write and exec permissions on dir
1816 * 3. We can't remove anything from append-only dir
1817 * 4. We can't do anything with immutable dir (done in permission())
1818 * 5. If the sticky bit on dir is set we should either
1819 * a. be owner of dir, or
1820 * b. be owner of victim, or
1821 * c. have CAP_FOWNER capability
1822 * 6. If the victim is append-only or immutable we can't do antyhing with
1823 * links pointing to it.
1824 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1825 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1826 * 9. We can't remove a root or mountpoint.
1827 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1828 * nfs_async_unlink().
1829 */
858119e1 1830static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1831{
1832 int error;
1833
1834 if (!victim->d_inode)
1835 return -ENOENT;
1836
1837 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1838 audit_inode_child(victim, dir);
1da177e4 1839
f419a2e3 1840 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1841 if (error)
1842 return error;
1843 if (IS_APPEND(dir))
1844 return -EPERM;
1845 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1846 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1847 return -EPERM;
1848 if (isdir) {
1849 if (!S_ISDIR(victim->d_inode->i_mode))
1850 return -ENOTDIR;
1851 if (IS_ROOT(victim))
1852 return -EBUSY;
1853 } else if (S_ISDIR(victim->d_inode->i_mode))
1854 return -EISDIR;
1855 if (IS_DEADDIR(dir))
1856 return -ENOENT;
1857 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1858 return -EBUSY;
1859 return 0;
1860}
1861
1862/* Check whether we can create an object with dentry child in directory
1863 * dir.
1864 * 1. We can't do it if child already exists (open has special treatment for
1865 * this case, but since we are inlined it's OK)
1866 * 2. We can't do it if dir is read-only (done in permission())
1867 * 3. We should have write and exec permissions on dir
1868 * 4. We can't do it if dir is immutable (done in permission())
1869 */
a95164d9 1870static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1871{
1872 if (child->d_inode)
1873 return -EEXIST;
1874 if (IS_DEADDIR(dir))
1875 return -ENOENT;
f419a2e3 1876 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1877}
1878
1da177e4
LT		 1879/*
1880 * p1 and p2 should be directories on the same fs.
1881 */
1882struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1883{
1884 struct dentry *p;
1885
1886 if (p1 == p2) {
f2eace23 1887 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1888 return NULL;
1889 }
1890
a11f3a05 1891 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1892
e2761a11
OH		 1893 p = d_ancestor(p2, p1);
1894 if (p) {
1895 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1896 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1897 return p;
1da177e4
LT		 1898 }
1899
e2761a11
OH		 1900 p = d_ancestor(p1, p2);
1901 if (p) {
1902 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1903 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1904 return p;
1da177e4
LT		 1905 }
1906
f2eace23
IM		 1907 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1908 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1909 return NULL;
1910}
1911
1912void unlock_rename(struct dentry *p1, struct dentry *p2)
1913{
1b1dcc1b 1914 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1915 if (p1 != p2) {
1b1dcc1b 1916 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1917 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1918 }
1919}
1920
1921int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1922 struct nameidata *nd)
1923{
a95164d9 1924 int error = may_create(dir, dentry);
1da177e4
LT		 1925
1926 if (error)
1927 return error;
1928
acfa4380 1929 if (!dir->i_op->create)
1da177e4
LT		 1930 return -EACCES; /* shouldn't it be ENOSYS? */
1931 mode &= S_IALLUGO;
1932 mode |= S_IFREG;
1933 error = security_inode_create(dir, dentry, mode);
1934 if (error)
1935 return error;
1da177e4 1936 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1937 if (!error)
f38aa942 1938 fsnotify_create(dir, dentry);
1da177e4
LT		 1939 return error;
1940}
1941
73d049a4 1942static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1943{
3fb64190 1944 struct dentry *dentry = path->dentry;
1da177e4
LT		 1945 struct inode *inode = dentry->d_inode;
1946 int error;
1947
1948 if (!inode)
1949 return -ENOENT;
1950
c8fe8f30
CH		 1951 switch (inode->i_mode & S_IFMT) {
1952 case S_IFLNK:
1da177e4 1953 return -ELOOP;
c8fe8f30
CH		 1954 case S_IFDIR:
1955 if (acc_mode & MAY_WRITE)
1956 return -EISDIR;
1957 break;
1958 case S_IFBLK:
1959 case S_IFCHR:
3fb64190 1960 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 1961 return -EACCES;
c8fe8f30
CH		 1962 /*FALLTHRU*/
1963 case S_IFIFO:
1964 case S_IFSOCK:
1da177e4 1965 flag &= ~O_TRUNC;
c8fe8f30 1966 break;
4a3fd211 1967 }
b41572e9 1968
3fb64190 1969 error = inode_permission(inode, acc_mode);
b41572e9
DH		 1970 if (error)
1971 return error;
6146f0d5 1972
1da177e4
LT		 1973 /*
1974 * An append-only file must be opened in append mode for writing.
1975 */
1976 if (IS_APPEND(inode)) {
8737c930 1977 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 1978 return -EPERM;
1da177e4 1979 if (flag & O_TRUNC)
7715b521 1980 return -EPERM;
1da177e4
LT		 1981 }
1982
1983 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 1984 if (flag & O_NOATIME && !is_owner_or_cap(inode))
1985 return -EPERM;
1da177e4
LT		 1986
1987 /*
1988 * Ensure there are no outstanding leases on the file.
1989 */
b65a9cfc 1990 return break_lease(inode, flag);
7715b521 1991}
1da177e4 1992
e1181ee6 1993static int handle_truncate(struct file *filp)
7715b521 1994{
e1181ee6 1995 struct path *path = &filp->f_path;
7715b521
AV		 1996 struct inode *inode = path->dentry->d_inode;
1997 int error = get_write_access(inode);
1998 if (error)
1999 return error;
2000 /*
2001 * Refuse to truncate files with mandatory locks held on them.
2002 */
2003 error = locks_verify_locked(inode);
2004 if (!error)
ea0d3ab2 2005 error = security_path_truncate(path);
7715b521
AV		 2006 if (!error) {
2007 error = do_truncate(path->dentry, 0,
2008 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2009 filp);
7715b521
AV		 2010 }
2011 put_write_access(inode);
acd0c935 2012 return error;
1da177e4
LT		 2013}
2014
d57999e1
DH		 2015/*
2016 * Note that while the flag value (low two bits) for sys_open means:
2017 * 00 - read-only
2018 * 01 - write-only
2019 * 10 - read-write
2020 * 11 - special
2021 * it is changed into
2022 * 00 - no permissions needed
2023 * 01 - read-permission
2024 * 10 - write-permission
2025 * 11 - read-write
2026 * for the internal routines (ie open_namei()/follow_link() etc)
2027 * This is more logical, and also allows the 00 "no perm needed"
2028 * to be used for symlinks (where the permissions are checked
2029 * later).
2030 *
2031*/
2032static inline int open_to_namei_flags(int flag)
2033{
2034 if ((flag+1) & O_ACCMODE)
2035 flag++;
2036 return flag;
2037}
2038
31e6b01f 2039/*
fe2d35ff 2040 * Handle the last step of open()
31e6b01f 2041 */
fb1cc555 2042static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2043 const struct open_flags *op, const char *pathname)
fb1cc555 2044{
a1e28038 2045 struct dentry *dir = nd->path.dentry;
6c0d46c4 2046 struct dentry *dentry;
ca344a89 2047 int open_flag = op->open_flag;
6c0d46c4 2048 int will_truncate = open_flag & O_TRUNC;
ca344a89
AV		 2049 int want_write = 0;
2050 int skip_perm = 0;
fb1cc555 2051 struct file *filp;
fe2d35ff 2052 struct inode *inode;
16c2cd71 2053 int error;
1f36f774 2054
c3e380b0
AV		 2055 nd->flags &= ~LOOKUP_PARENT;
2056 nd->flags |= op->intent;
2057
1f36f774
AV		 2058 switch (nd->last_type) {
2059 case LAST_DOTDOT:
176306f5 2060 case LAST_DOT:
fe2d35ff
AV		 2061 error = handle_dots(nd, nd->last_type);
2062 if (error)
2063 return ERR_PTR(error);
1f36f774 2064 /* fallthrough */
1f36f774 2065 case LAST_ROOT:
fe2d35ff
AV		 2066 if (nd->flags & LOOKUP_RCU) {
2067 if (nameidata_drop_rcu_last(nd))
2068 return ERR_PTR(-ECHILD);
2069 }
16c2cd71
AV		 2070 error = handle_reval_path(nd);
2071 if (error)
2072 goto exit;
fe2d35ff 2073 audit_inode(pathname, nd->path.dentry);
ca344a89 2074 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2075 error = -EISDIR;
2076 goto exit;
2077 }
2078 goto ok;
1f36f774 2079 case LAST_BIND:
fe2d35ff 2080 /* can't be RCU mode here */
16c2cd71
AV		 2081 error = handle_reval_path(nd);
2082 if (error)
2083 goto exit;
1f36f774 2084 audit_inode(pathname, dir);
67ee3ad2 2085 goto ok;
1f36f774 2086 }
67ee3ad2 2087
ca344a89 2088 if (!(open_flag & O_CREAT)) {
fe2d35ff
AV		 2089 if (nd->last.name[nd->last.len])
2090 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2091 /* we _can_ be in RCU mode here */
2092 error = do_lookup(nd, &nd->last, path, &inode);
2093 if (error) {
2094 terminate_walk(nd);
2095 return ERR_PTR(error);
2096 }
2097 if (!inode) {
2098 path_to_nameidata(path, nd);
2099 terminate_walk(nd);
2100 return ERR_PTR(-ENOENT);
2101 }
2102 if (unlikely(inode->i_op->follow_link)) {
2103 /* We drop rcu-walk here */
2104 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
2105 return ERR_PTR(-ECHILD);
2106 return NULL;
2107 }
2108 path_to_nameidata(path, nd);
2109 nd->inode = inode;
2110 /* sayonara */
2111 if (nd->flags & LOOKUP_RCU) {
2112 if (nameidata_drop_rcu_last(nd))
2113 return ERR_PTR(-ECHILD);
2114 }
2115
2116 error = -ENOTDIR;
2117 if (nd->flags & LOOKUP_DIRECTORY) {
2118 if (!inode->i_op->lookup)
2119 goto exit;
2120 }
2121 audit_inode(pathname, nd->path.dentry);
2122 goto ok;
2123 }
2124
2125 /* create side of things */
2126
2127 if (nd->flags & LOOKUP_RCU) {
2128 if (nameidata_drop_rcu_last(nd))
2129 return ERR_PTR(-ECHILD);
2130 }
2131
2132 audit_inode(pathname, dir);
16c2cd71 2133 error = -EISDIR;
1f36f774 2134 /* trailing slashes? */
31e6b01f
NP		 2135 if (nd->last.name[nd->last.len])
2136 goto exit;
a2c36b45 2137
a1e28038
AV		 2138 mutex_lock(&dir->d_inode->i_mutex);
2139
6c0d46c4
AV		 2140 dentry = lookup_hash(nd);
2141 error = PTR_ERR(dentry);
2142 if (IS_ERR(dentry)) {
fb1cc555
AV		 2143 mutex_unlock(&dir->d_inode->i_mutex);
2144 goto exit;
2145 }
2146
6c0d46c4
AV		 2147 path->dentry = dentry;
2148 path->mnt = nd->path.mnt;
2149
fb1cc555 2150 /* Negative dentry, just create the file */
6c0d46c4
AV		 2151 if (!dentry->d_inode) {
2152 int mode = op->mode;
2153 if (!IS_POSIXACL(dir->d_inode))
2154 mode &= ~current_umask();
fb1cc555
AV		 2155 /*
2156 * This write is needed to ensure that a
6c0d46c4 2157 * rw->ro transition does not occur between
fb1cc555
AV		 2158 * the time when the file is created and when
2159 * a permanent write count is taken through
2160 * the 'struct file' in nameidata_to_filp().
2161 */
2162 error = mnt_want_write(nd->path.mnt);
2163 if (error)
2164 goto exit_mutex_unlock;
ca344a89 2165 want_write = 1;
9b44f1b3 2166 /* Don't check for write permission, don't truncate */
ca344a89 2167 open_flag &= ~O_TRUNC;
6c0d46c4 2168 will_truncate = 0;
ca344a89 2169 skip_perm = 1;
6c0d46c4
AV		 2170 error = security_path_mknod(&nd->path, dentry, mode, 0);
2171 if (error)
2172 goto exit_mutex_unlock;
2173 error = vfs_create(dir->d_inode, dentry, mode, nd);
2174 if (error)
2175 goto exit_mutex_unlock;
2176 mutex_unlock(&dir->d_inode->i_mutex);
2177 dput(nd->path.dentry);
2178 nd->path.dentry = dentry;
ca344a89 2179 goto common;
fb1cc555
AV		 2180 }
2181
2182 /*
2183 * It already exists.
2184 */
2185 mutex_unlock(&dir->d_inode->i_mutex);
2186 audit_inode(pathname, path->dentry);
2187
2188 error = -EEXIST;
ca344a89 2189 if (open_flag & O_EXCL)
fb1cc555
AV		 2190 goto exit_dput;
2191
9875cf80
DH		 2192 error = follow_managed(path, nd->flags);
2193 if (error < 0)
2194 goto exit_dput;
fb1cc555
AV		 2195
2196 error = -ENOENT;
2197 if (!path->dentry->d_inode)
2198 goto exit_dput;
9e67f361
AV		 2199
2200 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2201 return NULL;
fb1cc555
AV		 2202
2203 path_to_nameidata(path, nd);
31e6b01f 2204 nd->inode = path->dentry->d_inode;
fb1cc555 2205 error = -EISDIR;
31e6b01f 2206 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2207 goto exit;
67ee3ad2 2208ok:
6c0d46c4
AV		 2209 if (!S_ISREG(nd->inode->i_mode))
2210 will_truncate = 0;
2211
0f9d1a10
AV		 2212 if (will_truncate) {
2213 error = mnt_want_write(nd->path.mnt);
2214 if (error)
2215 goto exit;
ca344a89 2216 want_write = 1;
0f9d1a10 2217 }
ca344a89
AV		 2218common:
2219 error = may_open(&nd->path, skip_perm ? 0 : op->acc_mode, open_flag);
2220 if (error)
0f9d1a10 2221 goto exit;
0f9d1a10
AV		 2222 filp = nameidata_to_filp(nd);
2223 if (!IS_ERR(filp)) {
2224 error = ima_file_check(filp, op->acc_mode);
2225 if (error) {
2226 fput(filp);
2227 filp = ERR_PTR(error);
2228 }
2229 }
2230 if (!IS_ERR(filp)) {
2231 if (will_truncate) {
2232 error = handle_truncate(filp);
2233 if (error) {
2234 fput(filp);
2235 filp = ERR_PTR(error);
2236 }
2237 }
2238 }
ca344a89
AV		 2239out:
2240 if (want_write)
0f9d1a10
AV		 2241 mnt_drop_write(nd->path.mnt);
2242 path_put(&nd->path);
fb1cc555
AV		 2243 return filp;
2244
2245exit_mutex_unlock:
2246 mutex_unlock(&dir->d_inode->i_mutex);
2247exit_dput:
2248 path_put_conditional(path, nd);
2249exit:
ca344a89
AV		 2250 filp = ERR_PTR(error);
2251 goto out;
fb1cc555
AV		 2252}
2253
13aab428 2254static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2255 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2256{
fe2d35ff 2257 struct file *base = NULL;
4a3fd211 2258 struct file *filp;
9850c056 2259 struct path path;
1da177e4 2260 int count = 0;
13aab428 2261 int error;
31e6b01f
NP		 2262
2263 filp = get_empty_filp();
2264 if (!filp)
2265 return ERR_PTR(-ENFILE);
2266
47c805dc 2267 filp->f_flags = op->open_flag;
73d049a4
AV		 2268 nd->intent.open.file = filp;
2269 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2270 nd->intent.open.create_mode = op->mode;
31e6b01f 2271
73d049a4 2272 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2273 if (unlikely(error))
13aab428 2274 goto out_filp;
31e6b01f 2275
fe2d35ff 2276 current->total_link_count = 0;
73d049a4 2277 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2278 if (unlikely(error))
2279 goto out_filp;
1da177e4 2280
73d049a4 2281 filp = do_last(nd, &path, op, pathname);
806b681c 2282 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa
NP		 2283 struct path link = path;
2284 struct inode *linki = link.dentry->d_inode;
def4af30 2285 void *cookie;
73d049a4
AV		 2286 if (!(nd->flags & LOOKUP_FOLLOW) || count++ == 32) {
2287 path_put_conditional(&path, nd);
2288 path_put(&nd->path);
40b39136
AV		 2289 filp = ERR_PTR(-ELOOP);
2290 break;
2291 }
806b681c
AV		 2292 /*
2293 * This is subtle. Instead of calling do_follow_link() we do
2294 * the thing by hands. The reason is that this way we have zero
2295 * link_count and path_walk() (called from ->follow_link)
2296 * honoring LOOKUP_PARENT. After that we have the parent and
2297 * last component, i.e. we are in the same situation as after
2298 * the first path_walk(). Well, almost - if the last component
2299 * is normal we get its copy stored in nd->last.name and we will
2300 * have to putname() it when we are done. Procfs-like symlinks
2301 * just set LAST_BIND.
2302 */
73d049a4
AV		 2303 nd->flags |= LOOKUP_PARENT;
2304 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2305 error = __do_follow_link(&link, nd, &cookie);
c3e380b0 2306 if (unlikely(error))
f1afe9ef 2307 filp = ERR_PTR(error);
c3e380b0 2308 else
73d049a4 2309 filp = do_last(nd, &path, op, pathname);
f1afe9ef 2310 if (!IS_ERR(cookie) && linki->i_op->put_link)
73d049a4 2311 linki->i_op->put_link(link.dentry, nd, cookie);
7b9337aa 2312 path_put(&link);
806b681c 2313 }
10fa8e62 2314out:
73d049a4
AV		 2315 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2316 path_put(&nd->root);
fe2d35ff
AV		 2317 if (base)
2318 fput(base);
73d049a4 2319 release_open_intent(nd);
10fa8e62 2320 return filp;
1da177e4 2321
31e6b01f 2322out_filp:
806b681c 2323 filp = ERR_PTR(error);
10fa8e62 2324 goto out;
1da177e4
LT		 2325}
2326
13aab428
AV		 2327struct file *do_filp_open(int dfd, const char *pathname,
2328 const struct open_flags *op, int flags)
2329{
73d049a4 2330 struct nameidata nd;
13aab428
AV		 2331 struct file *filp;
2332
73d049a4 2333 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2334 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2335 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2336 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2337 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2338 return filp;
2339}
2340
73d049a4
AV		 2341struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2342 const char *name, const struct open_flags *op, int flags)
2343{
2344 struct nameidata nd;
2345 struct file *file;
2346
2347 nd.root.mnt = mnt;
2348 nd.root.dentry = dentry;
2349
2350 flags |= LOOKUP_ROOT;
2351
2352 if (dentry->d_inode->i_op->follow_link)
2353 return ERR_PTR(-ELOOP);
2354
2355 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2356 if (unlikely(file == ERR_PTR(-ECHILD)))
2357 file = path_openat(-1, name, &nd, op, flags);
2358 if (unlikely(file == ERR_PTR(-ESTALE)))
2359 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2360 return file;
2361}
2362
1da177e4
LT		 2363/**
2364 * lookup_create - lookup a dentry, creating it if it doesn't exist
2365 * @nd: nameidata info
2366 * @is_dir: directory flag
2367 *
2368 * Simple function to lookup and return a dentry and create it
2369 * if it doesn't exist. Is SMP-safe.
c663e5d8 2370 *
4ac91378 2371 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2372 */
2373struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2374{
c663e5d8 2375 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2376
4ac91378 2377 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2378 /*
2379 * Yucky last component or no last component at all?
2380 * (foo/., foo/.., /////)
2381 */
1da177e4
LT		 2382 if (nd->last_type != LAST_NORM)
2383 goto fail;
2384 nd->flags &= ~LOOKUP_PARENT;
3516586a 2385 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2386 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2387
2388 /*
2389 * Do the final lookup.
2390 */
49705b77 2391 dentry = lookup_hash(nd);
1da177e4
LT		 2392 if (IS_ERR(dentry))
2393 goto fail;
c663e5d8 2394
e9baf6e5
AV		 2395 if (dentry->d_inode)
2396 goto eexist;
c663e5d8
CH		 2397 /*
2398 * Special case - lookup gave negative, but... we had foo/bar/
2399 * From the vfs_mknod() POV we just have a negative dentry -
2400 * all is fine. Let's be bastards - you had / on the end, you've
2401 * been asking for (non-existent) directory. -ENOENT for you.
2402 */
e9baf6e5
AV		 2403 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2404 dput(dentry);
2405 dentry = ERR_PTR(-ENOENT);
2406 }
1da177e4 2407 return dentry;
e9baf6e5 2408eexist:
1da177e4 2409 dput(dentry);
e9baf6e5 2410 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2411fail:
2412 return dentry;
2413}
f81a0bff 2414EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2415
2416int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2417{
a95164d9 2418 int error = may_create(dir, dentry);
1da177e4
LT		 2419
2420 if (error)
2421 return error;
2422
2423 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2424 return -EPERM;
2425
acfa4380 2426 if (!dir->i_op->mknod)
1da177e4
LT		 2427 return -EPERM;
2428
08ce5f16
SH		 2429 error = devcgroup_inode_mknod(mode, dev);
2430 if (error)
2431 return error;
2432
1da177e4
LT		 2433 error = security_inode_mknod(dir, dentry, mode, dev);
2434 if (error)
2435 return error;
2436
1da177e4 2437 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2438 if (!error)
f38aa942 2439 fsnotify_create(dir, dentry);
1da177e4
LT		 2440 return error;
2441}
2442
463c3197
DH		 2443static int may_mknod(mode_t mode)
2444{
2445 switch (mode & S_IFMT) {
2446 case S_IFREG:
2447 case S_IFCHR:
2448 case S_IFBLK:
2449 case S_IFIFO:
2450 case S_IFSOCK:
2451 case 0: /* zero mode translates to S_IFREG */
2452 return 0;
2453 case S_IFDIR:
2454 return -EPERM;
2455 default:
2456 return -EINVAL;
2457 }
2458}
2459
2e4d0924
HC		 2460SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2461 unsigned, dev)
1da177e4 2462{
2ad94ae6
AV		 2463 int error;
2464 char *tmp;
2465 struct dentry *dentry;
1da177e4
LT		 2466 struct nameidata nd;
2467
2468 if (S_ISDIR(mode))
2469 return -EPERM;
1da177e4 2470
2ad94ae6 2471 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2472 if (error)
2ad94ae6
AV		 2473 return error;
2474
1da177e4 2475 dentry = lookup_create(&nd, 0);
463c3197
DH		 2476 if (IS_ERR(dentry)) {
2477 error = PTR_ERR(dentry);
2478 goto out_unlock;
2479 }
4ac91378 2480 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2481 mode &= ~current_umask();
463c3197
DH		 2482 error = may_mknod(mode);
2483 if (error)
2484 goto out_dput;
2485 error = mnt_want_write(nd.path.mnt);
2486 if (error)
2487 goto out_dput;
be6d3e56
KT		 2488 error = security_path_mknod(&nd.path, dentry, mode, dev);
2489 if (error)
2490 goto out_drop_write;
463c3197 2491 switch (mode & S_IFMT) {
1da177e4 2492 case 0: case S_IFREG:
4ac91378 2493 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2494 break;
2495 case S_IFCHR: case S_IFBLK:
4ac91378 2496 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2497 new_decode_dev(dev));
2498 break;
2499 case S_IFIFO: case S_IFSOCK:
4ac91378 2500 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2501 break;
1da177e4 2502 }
be6d3e56 2503out_drop_write:
463c3197
DH		 2504 mnt_drop_write(nd.path.mnt);
2505out_dput:
2506 dput(dentry);
2507out_unlock:
4ac91378 2508 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2509 path_put(&nd.path);
1da177e4
LT		 2510 putname(tmp);
2511
2512 return error;
2513}
2514
3480b257 2515SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2516{
2517 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2518}
2519
1da177e4
LT		 2520int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2521{
a95164d9 2522 int error = may_create(dir, dentry);
1da177e4
LT		 2523
2524 if (error)
2525 return error;
2526
acfa4380 2527 if (!dir->i_op->mkdir)
1da177e4
LT		 2528 return -EPERM;
2529
2530 mode &= (S_IRWXUGO|S_ISVTX);
2531 error = security_inode_mkdir(dir, dentry, mode);
2532 if (error)
2533 return error;
2534
1da177e4 2535 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2536 if (!error)
f38aa942 2537 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2538 return error;
2539}
2540
2e4d0924 2541SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2542{
2543 int error = 0;
2544 char * tmp;
6902d925
DH		 2545 struct dentry *dentry;
2546 struct nameidata nd;
1da177e4 2547
2ad94ae6
AV		 2548 error = user_path_parent(dfd, pathname, &nd, &tmp);
2549 if (error)
6902d925 2550 goto out_err;
1da177e4 2551
6902d925
DH		 2552 dentry = lookup_create(&nd, 1);
2553 error = PTR_ERR(dentry);
2554 if (IS_ERR(dentry))
2555 goto out_unlock;
1da177e4 2556
4ac91378 2557 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2558 mode &= ~current_umask();
463c3197
DH		 2559 error = mnt_want_write(nd.path.mnt);
2560 if (error)
2561 goto out_dput;
be6d3e56
KT		 2562 error = security_path_mkdir(&nd.path, dentry, mode);
2563 if (error)
2564 goto out_drop_write;
4ac91378 2565 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2566out_drop_write:
463c3197
DH		 2567 mnt_drop_write(nd.path.mnt);
2568out_dput:
6902d925
DH		 2569 dput(dentry);
2570out_unlock:
4ac91378 2571 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2572 path_put(&nd.path);
6902d925
DH		 2573 putname(tmp);
2574out_err:
1da177e4
LT		 2575 return error;
2576}
2577
3cdad428 2578SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2579{
2580 return sys_mkdirat(AT_FDCWD, pathname, mode);
2581}
2582
1da177e4
LT		 2583/*
2584 * We try to drop the dentry early: we should have
2585 * a usage count of 2 if we're the only user of this
2586 * dentry, and if that is true (possibly after pruning
2587 * the dcache), then we drop the dentry now.
2588 *
2589 * A low-level filesystem can, if it choses, legally
2590 * do a
2591 *
2592 * if (!d_unhashed(dentry))
2593 * return -EBUSY;
2594 *
2595 * if it cannot handle the case of removing a directory
2596 * that is still in use by something else..
2597 */
2598void dentry_unhash(struct dentry *dentry)
2599{
2600 dget(dentry);
dc168427 2601 shrink_dcache_parent(dentry);
1da177e4 2602 spin_lock(&dentry->d_lock);
b7ab39f6 2603 if (dentry->d_count == 2)
1da177e4
LT		 2604 __d_drop(dentry);
2605 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2606}
2607
2608int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2609{
2610 int error = may_delete(dir, dentry, 1);
2611
2612 if (error)
2613 return error;
2614
acfa4380 2615 if (!dir->i_op->rmdir)
1da177e4
LT		 2616 return -EPERM;
2617
1b1dcc1b 2618 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2619 dentry_unhash(dentry);
2620 if (d_mountpoint(dentry))
2621 error = -EBUSY;
2622 else {
2623 error = security_inode_rmdir(dir, dentry);
2624 if (!error) {
2625 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2626 if (!error) {
1da177e4 2627 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2628 dont_mount(dentry);
2629 }
1da177e4
LT		 2630 }
2631 }
1b1dcc1b 2632 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2633 if (!error) {
1da177e4
LT		 2634 d_delete(dentry);
2635 }
2636 dput(dentry);
2637
2638 return error;
2639}
2640
5590ff0d 2641static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2642{
2643 int error = 0;
2644 char * name;
2645 struct dentry *dentry;
2646 struct nameidata nd;
2647
2ad94ae6 2648 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2649 if (error)
2ad94ae6 2650 return error;
1da177e4
LT		 2651
2652 switch(nd.last_type) {
0612d9fb
OH		 2653 case LAST_DOTDOT:
2654 error = -ENOTEMPTY;
2655 goto exit1;
2656 case LAST_DOT:
2657 error = -EINVAL;
2658 goto exit1;
2659 case LAST_ROOT:
2660 error = -EBUSY;
2661 goto exit1;
1da177e4 2662 }
0612d9fb
OH		 2663
2664 nd.flags &= ~LOOKUP_PARENT;
2665
4ac91378 2666 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2667 dentry = lookup_hash(&nd);
1da177e4 2668 error = PTR_ERR(dentry);
6902d925
DH		 2669 if (IS_ERR(dentry))
2670 goto exit2;
0622753b
DH		 2671 error = mnt_want_write(nd.path.mnt);
2672 if (error)
2673 goto exit3;
be6d3e56
KT		 2674 error = security_path_rmdir(&nd.path, dentry);
2675 if (error)
2676 goto exit4;
4ac91378 2677 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2678exit4:
0622753b
DH		 2679 mnt_drop_write(nd.path.mnt);
2680exit3:
6902d925
DH		 2681 dput(dentry);
2682exit2:
4ac91378 2683 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2684exit1:
1d957f9b 2685 path_put(&nd.path);
1da177e4
LT		 2686 putname(name);
2687 return error;
2688}
2689
3cdad428 2690SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2691{
2692 return do_rmdir(AT_FDCWD, pathname);
2693}
2694
1da177e4
LT		 2695int vfs_unlink(struct inode *dir, struct dentry *dentry)
2696{
2697 int error = may_delete(dir, dentry, 0);
2698
2699 if (error)
2700 return error;
2701
acfa4380 2702 if (!dir->i_op->unlink)
1da177e4
LT		 2703 return -EPERM;
2704
1b1dcc1b 2705 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2706 if (d_mountpoint(dentry))
2707 error = -EBUSY;
2708 else {
2709 error = security_inode_unlink(dir, dentry);
bec1052e 2710 if (!error) {
1da177e4 2711 error = dir->i_op->unlink(dir, dentry);
bec1052e 2712 if (!error)
d83c49f3 2713 dont_mount(dentry);
bec1052e 2714 }
1da177e4 2715 }
1b1dcc1b 2716 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2717
2718 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2719 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2720 fsnotify_link_count(dentry->d_inode);
e234f35c 2721 d_delete(dentry);
1da177e4 2722 }
0eeca283 2723
1da177e4
LT		 2724 return error;
2725}
2726
2727/*
2728 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2729 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2730 * writeout happening, and we don't want to prevent access to the directory
2731 * while waiting on the I/O.
2732 */
5590ff0d 2733static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2734{
2ad94ae6
AV		 2735 int error;
2736 char *name;
1da177e4
LT		 2737 struct dentry *dentry;
2738 struct nameidata nd;
2739 struct inode *inode = NULL;
2740
2ad94ae6 2741 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2742 if (error)
2ad94ae6
AV		 2743 return error;
2744
1da177e4
LT		 2745 error = -EISDIR;
2746 if (nd.last_type != LAST_NORM)
2747 goto exit1;
0612d9fb
OH		 2748
2749 nd.flags &= ~LOOKUP_PARENT;
2750
4ac91378 2751 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2752 dentry = lookup_hash(&nd);
1da177e4
LT		 2753 error = PTR_ERR(dentry);
2754 if (!IS_ERR(dentry)) {
2755 /* Why not before? Because we want correct error value */
2756 if (nd.last.name[nd.last.len])
2757 goto slashes;
2758 inode = dentry->d_inode;
2759 if (inode)
7de9c6ee 2760 ihold(inode);
0622753b
DH		 2761 error = mnt_want_write(nd.path.mnt);
2762 if (error)
2763 goto exit2;
be6d3e56
KT		 2764 error = security_path_unlink(&nd.path, dentry);
2765 if (error)
2766 goto exit3;
4ac91378 2767 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2768exit3:
0622753b 2769 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2770 exit2:
2771 dput(dentry);
2772 }
4ac91378 2773 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2774 if (inode)
2775 iput(inode); /* truncate the inode here */
2776exit1:
1d957f9b 2777 path_put(&nd.path);
1da177e4
LT		 2778 putname(name);
2779 return error;
2780
2781slashes:
2782 error = !dentry->d_inode ? -ENOENT :
2783 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2784 goto exit2;
2785}
2786
2e4d0924 2787SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2788{
2789 if ((flag & ~AT_REMOVEDIR) != 0)
2790 return -EINVAL;
2791
2792 if (flag & AT_REMOVEDIR)
2793 return do_rmdir(dfd, pathname);
2794
2795 return do_unlinkat(dfd, pathname);
2796}
2797
3480b257 2798SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2799{
2800 return do_unlinkat(AT_FDCWD, pathname);
2801}
2802
db2e747b 2803int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2804{
a95164d9 2805 int error = may_create(dir, dentry);
1da177e4
LT		 2806
2807 if (error)
2808 return error;
2809
acfa4380 2810 if (!dir->i_op->symlink)
1da177e4
LT		 2811 return -EPERM;
2812
2813 error = security_inode_symlink(dir, dentry, oldname);
2814 if (error)
2815 return error;
2816
1da177e4 2817 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2818 if (!error)
f38aa942 2819 fsnotify_create(dir, dentry);
1da177e4
LT		 2820 return error;
2821}
2822
2e4d0924
HC		 2823SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2824 int, newdfd, const char __user *, newname)
1da177e4 2825{
2ad94ae6
AV		 2826 int error;
2827 char *from;
2828 char *to;
6902d925
DH		 2829 struct dentry *dentry;
2830 struct nameidata nd;
1da177e4
LT		 2831
2832 from = getname(oldname);
2ad94ae6 2833 if (IS_ERR(from))
1da177e4 2834 return PTR_ERR(from);
1da177e4 2835
2ad94ae6 2836 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2837 if (error)
2ad94ae6
AV		 2838 goto out_putname;
2839
6902d925
DH		 2840 dentry = lookup_create(&nd, 0);
2841 error = PTR_ERR(dentry);
2842 if (IS_ERR(dentry))
2843 goto out_unlock;
2844
75c3f29d
DH		 2845 error = mnt_want_write(nd.path.mnt);
2846 if (error)
2847 goto out_dput;
be6d3e56
KT		 2848 error = security_path_symlink(&nd.path, dentry, from);
2849 if (error)
2850 goto out_drop_write;
db2e747b 2851 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2852out_drop_write:
75c3f29d
DH		 2853 mnt_drop_write(nd.path.mnt);
2854out_dput:
6902d925
DH		 2855 dput(dentry);
2856out_unlock:
4ac91378 2857 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2858 path_put(&nd.path);
6902d925
DH		 2859 putname(to);
2860out_putname:
1da177e4
LT		 2861 putname(from);
2862 return error;
2863}
2864
3480b257 2865SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2866{
2867 return sys_symlinkat(oldname, AT_FDCWD, newname);
2868}
2869
1da177e4
LT		 2870int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2871{
2872 struct inode *inode = old_dentry->d_inode;
2873 int error;
2874
2875 if (!inode)
2876 return -ENOENT;
2877
a95164d9 2878 error = may_create(dir, new_dentry);
1da177e4
LT		 2879 if (error)
2880 return error;
2881
2882 if (dir->i_sb != inode->i_sb)
2883 return -EXDEV;
2884
2885 /*
2886 * A link to an append-only or immutable file cannot be created.
2887 */
2888 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2889 return -EPERM;
acfa4380 2890 if (!dir->i_op->link)
1da177e4 2891 return -EPERM;
7e79eedb 2892 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2893 return -EPERM;
2894
2895 error = security_inode_link(old_dentry, dir, new_dentry);
2896 if (error)
2897 return error;
2898
7e79eedb 2899 mutex_lock(&inode->i_mutex);
1da177e4 2900 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2901 mutex_unlock(&inode->i_mutex);
e31e14ec 2902 if (!error)
7e79eedb 2903 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2904 return error;
2905}
2906
2907/*
2908 * Hardlinks are often used in delicate situations. We avoid
2909 * security-related surprises by not following symlinks on the
2910 * newname. --KAB
2911 *
2912 * We don't follow them on the oldname either to be compatible
2913 * with linux 2.0, and to avoid hard-linking to directories
2914 * and other special files. --ADM
2915 */
2e4d0924
HC		 2916SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2917 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2918{
2919 struct dentry *new_dentry;
2d8f3038
AV		 2920 struct nameidata nd;
2921 struct path old_path;
1da177e4 2922 int error;
2ad94ae6 2923 char *to;
1da177e4 2924
45c9b11a 2925 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 2926 return -EINVAL;
2927
2d8f3038
AV		 2928 error = user_path_at(olddfd, oldname,
2929 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2930 &old_path);
1da177e4 2931 if (error)
2ad94ae6
AV		 2932 return error;
2933
2934 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2935 if (error)
2936 goto out;
2937 error = -EXDEV;
2d8f3038 2938 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2939 goto out_release;
2940 new_dentry = lookup_create(&nd, 0);
2941 error = PTR_ERR(new_dentry);
6902d925
DH		 2942 if (IS_ERR(new_dentry))
2943 goto out_unlock;
75c3f29d
DH		 2944 error = mnt_want_write(nd.path.mnt);
2945 if (error)
2946 goto out_dput;
be6d3e56
KT		 2947 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2948 if (error)
2949 goto out_drop_write;
2d8f3038 2950 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2951out_drop_write:
75c3f29d
DH		 2952 mnt_drop_write(nd.path.mnt);
2953out_dput:
6902d925
DH		 2954 dput(new_dentry);
2955out_unlock:
4ac91378 2956 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2957out_release:
1d957f9b 2958 path_put(&nd.path);
2ad94ae6 2959 putname(to);
1da177e4 2960out:
2d8f3038 2961 path_put(&old_path);
1da177e4
LT		 2962
2963 return error;
2964}
2965
3480b257 2966SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 2967{
c04030e1 2968 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2969}
2970
1da177e4
LT		 2971/*
2972 * The worst of all namespace operations - renaming directory. "Perverted"
2973 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2974 * Problems:
2975 * a) we can get into loop creation. Check is done in is_subdir().
2976 * b) race potential - two innocent renames can create a loop together.
2977 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 2978 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 2979 * story.
2980 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 2981 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 2982 * whether the target exists). Solution: try to be smart with locking
2983 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 2984 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 2985 * move will be locked. Thus we can rank directories by the tree
2986 * (ancestors first) and rank all non-directories after them.
2987 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 2988 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 2989 * HOWEVER, it relies on the assumption that any object with ->lookup()
2990 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2991 * we'd better make sure that there's no link(2) for them.
2992 * d) some filesystems don't support opened-but-unlinked directories,
2993 * either because of layout or because they are not ready to deal with
2994 * all cases correctly. The latter will be fixed (taking this sort of
2995 * stuff into VFS), but the former is not going away. Solution: the same
2996 * trick as in rmdir().
2997 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 2998 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 2999 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3000 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3001 * locking].
3002 */
75c96f85
AB		 3003static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3004 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3005{
3006 int error = 0;
3007 struct inode *target;
3008
3009 /*
3010 * If we are going to change the parent - check write permissions,
3011 * we'll need to flip '..'.
3012 */
3013 if (new_dir != old_dir) {
f419a2e3 3014 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3015 if (error)
3016 return error;
3017 }
3018
3019 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3020 if (error)
3021 return error;
3022
3023 target = new_dentry->d_inode;
d83c49f3 3024 if (target)
1b1dcc1b 3025 mutex_lock(&target->i_mutex);
1da177e4
LT		 3026 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3027 error = -EBUSY;
d83c49f3
AV		 3028 else {
3029 if (target)
3030 dentry_unhash(new_dentry);
1da177e4 3031 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3032 }
1da177e4 3033 if (target) {
d83c49f3 3034 if (!error) {
1da177e4 3035 target->i_flags |= S_DEAD;
d83c49f3
AV		 3036 dont_mount(new_dentry);
3037 }
1b1dcc1b 3038 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3039 if (d_unhashed(new_dentry))
3040 d_rehash(new_dentry);
3041 dput(new_dentry);
3042 }
e31e14ec 3043 if (!error)
349457cc
MF		 3044 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3045 d_move(old_dentry,new_dentry);
1da177e4
LT		 3046 return error;
3047}
3048
75c96f85
AB		 3049static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3050 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3051{
3052 struct inode *target;
3053 int error;
3054
3055 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3056 if (error)
3057 return error;
3058
3059 dget(new_dentry);
3060 target = new_dentry->d_inode;
3061 if (target)
1b1dcc1b 3062 mutex_lock(&target->i_mutex);
1da177e4
LT		 3063 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3064 error = -EBUSY;
3065 else
3066 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3067 if (!error) {
bec1052e 3068 if (target)
d83c49f3 3069 dont_mount(new_dentry);
349457cc 3070 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3071 d_move(old_dentry, new_dentry);
1da177e4
LT		 3072 }
3073 if (target)
1b1dcc1b 3074 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3075 dput(new_dentry);
3076 return error;
3077}
3078
3079int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3080 struct inode *new_dir, struct dentry *new_dentry)
3081{
3082 int error;
3083 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3084 const unsigned char *old_name;
1da177e4
LT		 3085
3086 if (old_dentry->d_inode == new_dentry->d_inode)
3087 return 0;
3088
3089 error = may_delete(old_dir, old_dentry, is_dir);
3090 if (error)
3091 return error;
3092
3093 if (!new_dentry->d_inode)
a95164d9 3094 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3095 else
3096 error = may_delete(new_dir, new_dentry, is_dir);
3097 if (error)
3098 return error;
3099
acfa4380 3100 if (!old_dir->i_op->rename)
1da177e4
LT		 3101 return -EPERM;
3102
0eeca283
RL		 3103 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3104
1da177e4
LT		 3105 if (is_dir)
3106 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3107 else
3108 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3109 if (!error)
3110 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3111 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3112 fsnotify_oldname_free(old_name);
3113
1da177e4
LT		 3114 return error;
3115}
3116
2e4d0924
HC		 3117SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3118 int, newdfd, const char __user *, newname)
1da177e4 3119{
2ad94ae6
AV		 3120 struct dentry *old_dir, *new_dir;
3121 struct dentry *old_dentry, *new_dentry;
3122 struct dentry *trap;
1da177e4 3123 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3124 char *from;
3125 char *to;
3126 int error;
1da177e4 3127
2ad94ae6 3128 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3129 if (error)
3130 goto exit;
3131
2ad94ae6 3132 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3133 if (error)
3134 goto exit1;
3135
3136 error = -EXDEV;
4ac91378 3137 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3138 goto exit2;
3139
4ac91378 3140 old_dir = oldnd.path.dentry;
1da177e4
LT		 3141 error = -EBUSY;
3142 if (oldnd.last_type != LAST_NORM)
3143 goto exit2;
3144
4ac91378 3145 new_dir = newnd.path.dentry;
1da177e4
LT		 3146 if (newnd.last_type != LAST_NORM)
3147 goto exit2;
3148
0612d9fb
OH		 3149 oldnd.flags &= ~LOOKUP_PARENT;
3150 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3151 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3152
1da177e4
LT		 3153 trap = lock_rename(new_dir, old_dir);
3154
49705b77 3155 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3156 error = PTR_ERR(old_dentry);
3157 if (IS_ERR(old_dentry))
3158 goto exit3;
3159 /* source must exist */
3160 error = -ENOENT;
3161 if (!old_dentry->d_inode)
3162 goto exit4;
3163 /* unless the source is a directory trailing slashes give -ENOTDIR */
3164 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3165 error = -ENOTDIR;
3166 if (oldnd.last.name[oldnd.last.len])
3167 goto exit4;
3168 if (newnd.last.name[newnd.last.len])
3169 goto exit4;
3170 }
3171 /* source should not be ancestor of target */
3172 error = -EINVAL;
3173 if (old_dentry == trap)
3174 goto exit4;
49705b77 3175 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3176 error = PTR_ERR(new_dentry);
3177 if (IS_ERR(new_dentry))
3178 goto exit4;
3179 /* target should not be an ancestor of source */
3180 error = -ENOTEMPTY;
3181 if (new_dentry == trap)
3182 goto exit5;
3183
9079b1eb
DH		 3184 error = mnt_want_write(oldnd.path.mnt);
3185 if (error)
3186 goto exit5;
be6d3e56
KT		 3187 error = security_path_rename(&oldnd.path, old_dentry,
3188 &newnd.path, new_dentry);
3189 if (error)
3190 goto exit6;
1da177e4
LT		 3191 error = vfs_rename(old_dir->d_inode, old_dentry,
3192 new_dir->d_inode, new_dentry);
be6d3e56 3193exit6:
9079b1eb 3194 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3195exit5:
3196 dput(new_dentry);
3197exit4:
3198 dput(old_dentry);
3199exit3:
3200 unlock_rename(new_dir, old_dir);
3201exit2:
1d957f9b 3202 path_put(&newnd.path);
2ad94ae6 3203 putname(to);
1da177e4 3204exit1:
1d957f9b 3205 path_put(&oldnd.path);
1da177e4 3206 putname(from);
2ad94ae6 3207exit:
1da177e4
LT		 3208 return error;
3209}
3210
a26eab24 3211SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3212{
3213 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3214}
3215
1da177e4
LT		 3216int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3217{
3218 int len;
3219
3220 len = PTR_ERR(link);
3221 if (IS_ERR(link))
3222 goto out;
3223
3224 len = strlen(link);
3225 if (len > (unsigned) buflen)
3226 len = buflen;
3227 if (copy_to_user(buffer, link, len))
3228 len = -EFAULT;
3229out:
3230 return len;
3231}
3232
3233/*
3234 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3235 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3236 * using) it for any given inode is up to filesystem.
3237 */
3238int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3239{
3240 struct nameidata nd;
cc314eef 3241 void *cookie;
694a1764 3242 int res;
cc314eef 3243
1da177e4 3244 nd.depth = 0;
cc314eef 3245 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3246 if (IS_ERR(cookie))
3247 return PTR_ERR(cookie);
3248
3249 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3250 if (dentry->d_inode->i_op->put_link)
3251 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3252 return res;
1da177e4
LT		 3253}
3254
3255int vfs_follow_link(struct nameidata *nd, const char *link)
3256{
3257 return __vfs_follow_link(nd, link);
3258}
3259
3260/* get the link contents into pagecache */
3261static char *page_getlink(struct dentry * dentry, struct page **ppage)
3262{
ebd09abb
DG		 3263 char *kaddr;
3264 struct page *page;
1da177e4 3265 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3266 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3267 if (IS_ERR(page))
6fe6900e 3268 return (char*)page;
1da177e4 3269 *ppage = page;
ebd09abb
DG		 3270 kaddr = kmap(page);
3271 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3272 return kaddr;
1da177e4
LT		 3273}
3274
3275int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3276{
3277 struct page *page = NULL;
3278 char *s = page_getlink(dentry, &page);
3279 int res = vfs_readlink(dentry,buffer,buflen,s);
3280 if (page) {
3281 kunmap(page);
3282 page_cache_release(page);
3283 }
3284 return res;
3285}
3286
cc314eef 3287void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3288{
cc314eef 3289 struct page *page = NULL;
1da177e4 3290 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3291 return page;
1da177e4
LT		 3292}
3293
cc314eef 3294void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3295{
cc314eef
LT		 3296 struct page *page = cookie;
3297
3298 if (page) {
1da177e4
LT		 3299 kunmap(page);
3300 page_cache_release(page);
1da177e4
LT		 3301 }
3302}
3303
54566b2c
NP		 3304/*
3305 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3306 */
3307int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3308{
3309 struct address_space *mapping = inode->i_mapping;
0adb25d2 3310 struct page *page;
afddba49 3311 void *fsdata;
beb497ab 3312 int err;
1da177e4 3313 char *kaddr;
54566b2c
NP		 3314 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3315 if (nofs)
3316 flags |= AOP_FLAG_NOFS;
1da177e4 3317
7e53cac4 3318retry:
afddba49 3319 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3320 flags, &page, &fsdata);
1da177e4 3321 if (err)
afddba49
NP		 3322 goto fail;
3323
1da177e4
LT		 3324 kaddr = kmap_atomic(page, KM_USER0);
3325 memcpy(kaddr, symname, len-1);
3326 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3327
3328 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3329 page, fsdata);
1da177e4
LT		 3330 if (err < 0)
3331 goto fail;
afddba49
NP		 3332 if (err < len-1)
3333 goto retry;
3334
1da177e4
LT		 3335 mark_inode_dirty(inode);
3336 return 0;
1da177e4
LT		 3337fail:
3338 return err;
3339}
3340
0adb25d2
KK		 3341int page_symlink(struct inode *inode, const char *symname, int len)
3342{
3343 return __page_symlink(inode, symname, len,
54566b2c 3344 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3345}
3346
92e1d5be 3347const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3348 .readlink = generic_readlink,
3349 .follow_link = page_follow_link_light,
3350 .put_link = page_put_link,
3351};
3352
2d8f3038 3353EXPORT_SYMBOL(user_path_at);
cc53ce53 3354EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3355EXPORT_SYMBOL(follow_down);
3356EXPORT_SYMBOL(follow_up);
3357EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3358EXPORT_SYMBOL(getname);
3359EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3360EXPORT_SYMBOL(lookup_one_len);
3361EXPORT_SYMBOL(page_follow_link_light);
3362EXPORT_SYMBOL(page_put_link);
3363EXPORT_SYMBOL(page_readlink);
0adb25d2 3364EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3365EXPORT_SYMBOL(page_symlink);
3366EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3367EXPORT_SYMBOL(kern_path_parent);
d1811465 3368EXPORT_SYMBOL(kern_path);
16f18200 3369EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3370EXPORT_SYMBOL(inode_permission);
8c744fb8 3371EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3372EXPORT_SYMBOL(unlock_rename);
3373EXPORT_SYMBOL(vfs_create);
3374EXPORT_SYMBOL(vfs_follow_link);
3375EXPORT_SYMBOL(vfs_link);
3376EXPORT_SYMBOL(vfs_mkdir);
3377EXPORT_SYMBOL(vfs_mknod);
3378EXPORT_SYMBOL(generic_permission);
3379EXPORT_SYMBOL(vfs_readlink);
3380EXPORT_SYMBOL(vfs_rename);
3381EXPORT_SYMBOL(vfs_rmdir);
3382EXPORT_SYMBOL(vfs_symlink);
3383EXPORT_SYMBOL(vfs_unlink);
3384EXPORT_SYMBOL(dentry_unhash);
3385EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)