projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'x86-platform-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[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
f52e0c11 139static char *getname_flags(const char __user * filename, int flags)
1da177e4
LT		 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) {
f52e0c11
AV		 150 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
1da177e4
LT		 154 }
155 }
156 audit_getname(result);
157 return result;
158}
159
f52e0c11
AV		 160char *getname(const char __user * filename)
161{
162 return getname_flags(filename, 0);
163}
164
1da177e4
LT		 165#ifdef CONFIG_AUDITSYSCALL
166void putname(const char *name)
167{
5ac3a9c2 168 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 169 audit_putname(name);
170 else
171 __putname(name);
172}
173EXPORT_SYMBOL(putname);
174#endif
175
5909ccaa
LT		 176/*
177 * This does basic POSIX ACL permission checking
1da177e4 178 */
b74c79e9
NP		 179static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
180 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
1da177e4
LT		 181{
182 umode_t mode = inode->i_mode;
183
e6305c43
AV		 184 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
185
da9592ed 186 if (current_fsuid() == inode->i_uid)
1da177e4
LT		 187 mode >>= 6;
188 else {
189 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
b74c79e9
NP		 190 int error = check_acl(inode, mask, flags);
191 if (error != -EAGAIN)
192 return error;
1da177e4
LT		 193 }
194
195 if (in_group_p(inode->i_gid))
196 mode >>= 3;
197 }
198
199 /*
200 * If the DACs are ok we don't need any capability check.
201 */
e6305c43 202 if ((mask & ~mode) == 0)
1da177e4 203 return 0;
5909ccaa
LT		 204 return -EACCES;
205}
206
207/**
b74c79e9 208 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa
LT		 209 * @inode: inode to check access rights for
210 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
211 * @check_acl: optional callback to check for Posix ACLs
39191628 212 * @flags: IPERM_FLAG_ flags.
5909ccaa
LT		 213 *
214 * Used to check for read/write/execute permissions on a file.
215 * We use "fsuid" for this, letting us set arbitrary permissions
216 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 217 * are used for other things.
218 *
219 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
220 * request cannot be satisfied (eg. requires blocking or too much complexity).
221 * It would then be called again in ref-walk mode.
5909ccaa 222 */
b74c79e9
NP		 223int generic_permission(struct inode *inode, int mask, unsigned int flags,
224 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
5909ccaa
LT		 225{
226 int ret;
227
228 /*
229 * Do the basic POSIX ACL permission checks.
230 */
b74c79e9 231 ret = acl_permission_check(inode, mask, flags, check_acl);
5909ccaa
LT		 232 if (ret != -EACCES)
233 return ret;
1da177e4 234
1da177e4
LT		 235 /*
236 * Read/write DACs are always overridable.
237 * Executable DACs are overridable if at least one exec bit is set.
238 */
f696a365 239 if (!(mask & MAY_EXEC) || execute_ok(inode))
1da177e4
LT		 240 if (capable(CAP_DAC_OVERRIDE))
241 return 0;
242
243 /*
244 * Searching includes executable on directories, else just read.
245 */
7ea66001 246 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
1da177e4
LT		 247 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
248 if (capable(CAP_DAC_READ_SEARCH))
249 return 0;
250
251 return -EACCES;
252}
253
cb23beb5
CH		 254/**
255 * inode_permission - check for access rights to a given inode
256 * @inode: inode to check permission on
257 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
258 *
259 * Used to check for read/write/execute permissions on an inode.
260 * We use "fsuid" for this, letting us set arbitrary permissions
261 * for filesystem access without changing the "normal" uids which
262 * are used for other things.
263 */
f419a2e3 264int inode_permission(struct inode *inode, int mask)
1da177e4 265{
e6305c43 266 int retval;
1da177e4
LT		 267
268 if (mask & MAY_WRITE) {
22590e41 269 umode_t mode = inode->i_mode;
1da177e4
LT		 270
271 /*
272 * Nobody gets write access to a read-only fs.
273 */
274 if (IS_RDONLY(inode) &&
275 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
276 return -EROFS;
277
278 /*
279 * Nobody gets write access to an immutable file.
280 */
281 if (IS_IMMUTABLE(inode))
282 return -EACCES;
283 }
284
acfa4380 285 if (inode->i_op->permission)
b74c79e9 286 retval = inode->i_op->permission(inode, mask, 0);
f696a365 287 else
b74c79e9
NP		 288 retval = generic_permission(inode, mask, 0,
289 inode->i_op->check_acl);
f696a365 290
1da177e4
LT		 291 if (retval)
292 return retval;
293
08ce5f16
SH		 294 retval = devcgroup_inode_permission(inode, mask);
295 if (retval)
296 return retval;
297
d09ca739 298 return security_inode_permission(inode, mask);
1da177e4
LT		 299}
300
8c744fb8
CH		 301/**
302 * file_permission - check for additional access rights to a given file
303 * @file: file to check access rights for
304 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
305 *
306 * Used to check for read/write/execute permissions on an already opened
307 * file.
308 *
309 * Note:
310 * Do not use this function in new code. All access checks should
cb23beb5 311 * be done using inode_permission().
8c744fb8
CH		 312 */
313int file_permission(struct file *file, int mask)
314{
f419a2e3 315 return inode_permission(file->f_path.dentry->d_inode, mask);
8c744fb8
CH		 316}
317
1da177e4
LT		 318/*
319 * get_write_access() gets write permission for a file.
320 * put_write_access() releases this write permission.
321 * This is used for regular files.
322 * We cannot support write (and maybe mmap read-write shared) accesses and
323 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
324 * can have the following values:
325 * 0: no writers, no VM_DENYWRITE mappings
326 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
327 * > 0: (i_writecount) users are writing to the file.
328 *
329 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
330 * except for the cases where we don't hold i_writecount yet. Then we need to
331 * use {get,deny}_write_access() - these functions check the sign and refuse
332 * to do the change if sign is wrong. Exclusion between them is provided by
333 * the inode->i_lock spinlock.
334 */
335
336int get_write_access(struct inode * inode)
337{
338 spin_lock(&inode->i_lock);
339 if (atomic_read(&inode->i_writecount) < 0) {
340 spin_unlock(&inode->i_lock);
341 return -ETXTBSY;
342 }
343 atomic_inc(&inode->i_writecount);
344 spin_unlock(&inode->i_lock);
345
346 return 0;
347}
348
349int deny_write_access(struct file * file)
350{
0f7fc9e4 351 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4
LT		 352
353 spin_lock(&inode->i_lock);
354 if (atomic_read(&inode->i_writecount) > 0) {
355 spin_unlock(&inode->i_lock);
356 return -ETXTBSY;
357 }
358 atomic_dec(&inode->i_writecount);
359 spin_unlock(&inode->i_lock);
360
361 return 0;
362}
363
5dd784d0
JB		 364/**
365 * path_get - get a reference to a path
366 * @path: path to get the reference to
367 *
368 * Given a path increment the reference count to the dentry and the vfsmount.
369 */
370void path_get(struct path *path)
371{
372 mntget(path->mnt);
373 dget(path->dentry);
374}
375EXPORT_SYMBOL(path_get);
376
1d957f9b
JB		 377/**
378 * path_put - put a reference to a path
379 * @path: path to put the reference to
380 *
381 * Given a path decrement the reference count to the dentry and the vfsmount.
382 */
383void path_put(struct path *path)
1da177e4 384{
1d957f9b
JB		 385 dput(path->dentry);
386 mntput(path->mnt);
1da177e4 387}
1d957f9b 388EXPORT_SYMBOL(path_put);
1da177e4 389
31e6b01f
NP		 390/**
391 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
392 * @nd: nameidata pathwalk data to drop
39191628 393 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 394 *
395 * Path walking has 2 modes, rcu-walk and ref-walk (see
396 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
397 * to drop out of rcu-walk mode and take normal reference counts on dentries
398 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
399 * refcounts at the last known good point before rcu-walk got stuck, so
400 * ref-walk may continue from there. If this is not successful (eg. a seqcount
401 * has changed), then failure is returned and path walk restarts from the
402 * beginning in ref-walk mode.
403 *
404 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
405 * ref-walk. Must be called from rcu-walk context.
406 */
407static int nameidata_drop_rcu(struct nameidata *nd)
408{
409 struct fs_struct *fs = current->fs;
410 struct dentry *dentry = nd->path.dentry;
5b6ca027 411 int want_root = 0;
31e6b01f
NP		 412
413 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 414 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
415 want_root = 1;
31e6b01f
NP		 416 spin_lock(&fs->lock);
417 if (nd->root.mnt != fs->root.mnt ||
418 nd->root.dentry != fs->root.dentry)
419 goto err_root;
420 }
421 spin_lock(&dentry->d_lock);
422 if (!__d_rcu_to_refcount(dentry, nd->seq))
423 goto err;
424 BUG_ON(nd->inode != dentry->d_inode);
425 spin_unlock(&dentry->d_lock);
5b6ca027 426 if (want_root) {
31e6b01f
NP		 427 path_get(&nd->root);
428 spin_unlock(&fs->lock);
429 }
430 mntget(nd->path.mnt);
431
432 rcu_read_unlock();
433 br_read_unlock(vfsmount_lock);
434 nd->flags &= ~LOOKUP_RCU;
435 return 0;
436err:
437 spin_unlock(&dentry->d_lock);
438err_root:
5b6ca027 439 if (want_root)
31e6b01f
NP		 440 spin_unlock(&fs->lock);
441 return -ECHILD;
442}
443
444/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
445static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
446{
447 if (nd->flags & LOOKUP_RCU)
448 return nameidata_drop_rcu(nd);
449 return 0;
450}
451
452/**
453 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
454 * @nd: nameidata pathwalk data to drop
455 * @dentry: dentry to drop
39191628 456 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 457 *
458 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
459 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
460 * @nd. Must be called from rcu-walk context.
461 */
462static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
463{
464 struct fs_struct *fs = current->fs;
465 struct dentry *parent = nd->path.dentry;
5b6ca027 466 int want_root = 0;
31e6b01f
NP		 467
468 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 469 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
470 want_root = 1;
31e6b01f
NP		 471 spin_lock(&fs->lock);
472 if (nd->root.mnt != fs->root.mnt ||
473 nd->root.dentry != fs->root.dentry)
474 goto err_root;
475 }
476 spin_lock(&parent->d_lock);
477 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
478 if (!__d_rcu_to_refcount(dentry, nd->seq))
479 goto err;
480 /*
481 * If the sequence check on the child dentry passed, then the child has
482 * not been removed from its parent. This means the parent dentry must
483 * be valid and able to take a reference at this point.
484 */
485 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
486 BUG_ON(!parent->d_count);
487 parent->d_count++;
488 spin_unlock(&dentry->d_lock);
489 spin_unlock(&parent->d_lock);
5b6ca027 490 if (want_root) {
31e6b01f
NP		 491 path_get(&nd->root);
492 spin_unlock(&fs->lock);
493 }
494 mntget(nd->path.mnt);
495
496 rcu_read_unlock();
497 br_read_unlock(vfsmount_lock);
498 nd->flags &= ~LOOKUP_RCU;
499 return 0;
500err:
501 spin_unlock(&dentry->d_lock);
502 spin_unlock(&parent->d_lock);
503err_root:
5b6ca027 504 if (want_root)
31e6b01f
NP		 505 spin_unlock(&fs->lock);
506 return -ECHILD;
507}
508
509/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
510static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
511{
a7472bab
AV		 512 if (nd->flags & LOOKUP_RCU) {
513 if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) {
514 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 515 if (!(nd->flags & LOOKUP_ROOT))
516 nd->root.mnt = NULL;
a7472bab
AV		 517 rcu_read_unlock();
518 br_read_unlock(vfsmount_lock);
519 return -ECHILD;
520 }
521 }
31e6b01f
NP		 522 return 0;
523}
524
525/**
526 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
527 * @nd: nameidata pathwalk data to drop
39191628 528 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 529 *
530 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
531 * nd->path should be the final element of the lookup, so nd->root is discarded.
532 * Must be called from rcu-walk context.
533 */
534static int nameidata_drop_rcu_last(struct nameidata *nd)
535{
536 struct dentry *dentry = nd->path.dentry;
537
538 BUG_ON(!(nd->flags & LOOKUP_RCU));
539 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 540 if (!(nd->flags & LOOKUP_ROOT))
541 nd->root.mnt = NULL;
31e6b01f
NP		 542 spin_lock(&dentry->d_lock);
543 if (!__d_rcu_to_refcount(dentry, nd->seq))
544 goto err_unlock;
545 BUG_ON(nd->inode != dentry->d_inode);
546 spin_unlock(&dentry->d_lock);
547
548 mntget(nd->path.mnt);
549
550 rcu_read_unlock();
551 br_read_unlock(vfsmount_lock);
552
553 return 0;
554
555err_unlock:
556 spin_unlock(&dentry->d_lock);
557 rcu_read_unlock();
558 br_read_unlock(vfsmount_lock);
559 return -ECHILD;
560}
561
834f2a4a
TM		 562/**
563 * release_open_intent - free up open intent resources
564 * @nd: pointer to nameidata
565 */
566void release_open_intent(struct nameidata *nd)
567{
2dab5974
LT		 568 struct file *file = nd->intent.open.file;
569
570 if (file && !IS_ERR(file)) {
571 if (file->f_path.dentry == NULL)
572 put_filp(file);
573 else
574 fput(file);
575 }
834f2a4a
TM		 576}
577
f60aef7e 578static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 579{
f60aef7e 580 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 581}
582
f5e1c1c1 583static struct dentry *
bcdc5e01
IK		 584do_revalidate(struct dentry *dentry, struct nameidata *nd)
585{
f5e1c1c1 586 int status = d_revalidate(dentry, nd);
bcdc5e01
IK		 587 if (unlikely(status <= 0)) {
588 /*
589 * The dentry failed validation.
590 * If d_revalidate returned 0 attempt to invalidate
591 * the dentry otherwise d_revalidate is asking us
592 * to return a fail status.
593 */
34286d66 594 if (status < 0) {
f5e1c1c1 595 dput(dentry);
34286d66 596 dentry = ERR_PTR(status);
f5e1c1c1
AV		 597 } else if (!d_invalidate(dentry)) {
598 dput(dentry);
599 dentry = NULL;
bcdc5e01
IK		 600 }
601 }
602 return dentry;
603}
604
39159de2 605/*
16c2cd71 606 * handle_reval_path - force revalidation of a dentry
39159de2
JL		 607 *
608 * In some situations the path walking code will trust dentries without
609 * revalidating them. This causes problems for filesystems that depend on
610 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
611 * (which indicates that it's possible for the dentry to go stale), force
612 * a d_revalidate call before proceeding.
613 *
614 * Returns 0 if the revalidation was successful. If the revalidation fails,
615 * either return the error returned by d_revalidate or -ESTALE if the
616 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
617 * invalidate the dentry. It's up to the caller to handle putting references
618 * to the path if necessary.
619 */
16c2cd71 620static inline int handle_reval_path(struct nameidata *nd)
39159de2 621{
16c2cd71 622 struct dentry *dentry = nd->path.dentry;
39159de2 623 int status;
39159de2 624
16c2cd71
AV		 625 if (likely(!(nd->flags & LOOKUP_JUMPED)))
626 return 0;
627
628 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 629 return 0;
630
16c2cd71
AV		 631 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
632 return 0;
633
634 /* Note: we do not d_invalidate() */
34286d66 635 status = d_revalidate(dentry, nd);
39159de2
JL		 636 if (status > 0)
637 return 0;
638
16c2cd71 639 if (!status)
39159de2 640 status = -ESTALE;
16c2cd71 641
39159de2
JL		 642 return status;
643}
644
1da177e4 645/*
b75b5086
AV		 646 * Short-cut version of permission(), for calling on directories
647 * during pathname resolution. Combines parts of permission()
648 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
1da177e4
LT		 649 *
650 * If appropriate, check DAC only. If not appropriate, or
b75b5086 651 * short-cut DAC fails, then call ->permission() to do more
1da177e4
LT		 652 * complete permission check.
653 */
b74c79e9 654static inline int exec_permission(struct inode *inode, unsigned int flags)
1da177e4 655{
5909ccaa 656 int ret;
1da177e4 657
cb9179ea 658 if (inode->i_op->permission) {
b74c79e9
NP		 659 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
660 } else {
661 ret = acl_permission_check(inode, MAY_EXEC, flags,
662 inode->i_op->check_acl);
cb9179ea 663 }
b74c79e9 664 if (likely(!ret))
1da177e4 665 goto ok;
b74c79e9 666 if (ret == -ECHILD)
31e6b01f 667 return ret;
1da177e4 668
f1ac9f6b 669 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
1da177e4
LT		 670 goto ok;
671
5909ccaa 672 return ret;
1da177e4 673ok:
b74c79e9 674 return security_inode_exec_permission(inode, flags);
1da177e4
LT		 675}
676
2a737871
AV		 677static __always_inline void set_root(struct nameidata *nd)
678{
f7ad3c6b
MS		 679 if (!nd->root.mnt)
680 get_fs_root(current->fs, &nd->root);
2a737871
AV		 681}
682
6de88d72
AV		 683static int link_path_walk(const char *, struct nameidata *);
684
31e6b01f
NP		 685static __always_inline void set_root_rcu(struct nameidata *nd)
686{
687 if (!nd->root.mnt) {
688 struct fs_struct *fs = current->fs;
c28cc364
NP		 689 unsigned seq;
690
691 do {
692 seq = read_seqcount_begin(&fs->seq);
693 nd->root = fs->root;
694 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 695 }
696}
697
f1662356 698static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 699{
31e6b01f
NP		 700 int ret;
701
1da177e4
LT		 702 if (IS_ERR(link))
703 goto fail;
704
705 if (*link == '/') {
2a737871 706 set_root(nd);
1d957f9b 707 path_put(&nd->path);
2a737871
AV		 708 nd->path = nd->root;
709 path_get(&nd->root);
16c2cd71 710 nd->flags |= LOOKUP_JUMPED;
1da177e4 711 }
31e6b01f 712 nd->inode = nd->path.dentry->d_inode;
b4091d5f 713
31e6b01f
NP		 714 ret = link_path_walk(link, nd);
715 return ret;
1da177e4 716fail:
1d957f9b 717 path_put(&nd->path);
1da177e4
LT		 718 return PTR_ERR(link);
719}
720
1d957f9b 721static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 722{
723 dput(path->dentry);
4ac91378 724 if (path->mnt != nd->path.mnt)
051d3812
IK		 725 mntput(path->mnt);
726}
727
7b9337aa
NP		 728static inline void path_to_nameidata(const struct path *path,
729 struct nameidata *nd)
051d3812 730{
31e6b01f
NP		 731 if (!(nd->flags & LOOKUP_RCU)) {
732 dput(nd->path.dentry);
733 if (nd->path.mnt != path->mnt)
734 mntput(nd->path.mnt);
9a229683 735 }
31e6b01f 736 nd->path.mnt = path->mnt;
4ac91378 737 nd->path.dentry = path->dentry;
051d3812
IK		 738}
739
574197e0
AV		 740static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
741{
742 struct inode *inode = link->dentry->d_inode;
743 if (!IS_ERR(cookie) && inode->i_op->put_link)
744 inode->i_op->put_link(link->dentry, nd, cookie);
745 path_put(link);
746}
747
def4af30 748static __always_inline int
574197e0 749follow_link(struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 750{
751 int error;
7b9337aa 752 struct dentry *dentry = link->dentry;
1da177e4 753
844a3917
AV		 754 BUG_ON(nd->flags & LOOKUP_RCU);
755
574197e0
AV		 756 if (unlikely(current->total_link_count >= 40)) {
757 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
758 path_put_conditional(link, nd);
759 path_put(&nd->path);
760 return -ELOOP;
761 }
762 cond_resched();
763 current->total_link_count++;
764
7b9337aa 765 touch_atime(link->mnt, dentry);
1da177e4 766 nd_set_link(nd, NULL);
cd4e91d3 767
87556ef1
DH		 768 if (link->mnt == nd->path.mnt)
769 mntget(link->mnt);
31e6b01f 770
36f3b4f6
AV		 771 error = security_inode_follow_link(link->dentry, nd);
772 if (error) {
773 *p = ERR_PTR(error); /* no ->put_link(), please */
774 path_put(&nd->path);
775 return error;
776 }
777
86acdca1 778 nd->last_type = LAST_BIND;
def4af30
AV		 779 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
780 error = PTR_ERR(*p);
781 if (!IS_ERR(*p)) {
1da177e4 782 char *s = nd_get_link(nd);
cc314eef 783 error = 0;
1da177e4
LT		 784 if (s)
785 error = __vfs_follow_link(nd, s);
bcda7652 786 else if (nd->last_type == LAST_BIND) {
16c2cd71 787 nd->flags |= LOOKUP_JUMPED;
b21041d0
AV		 788 nd->inode = nd->path.dentry->d_inode;
789 if (nd->inode->i_op->follow_link) {
bcda7652
AV		 790 /* stepped on a _really_ weird one */
791 path_put(&nd->path);
792 error = -ELOOP;
793 }
794 }
1da177e4 795 }
1da177e4
LT		 796 return error;
797}
798
31e6b01f
NP		 799static int follow_up_rcu(struct path *path)
800{
801 struct vfsmount *parent;
802 struct dentry *mountpoint;
803
804 parent = path->mnt->mnt_parent;
805 if (parent == path->mnt)
806 return 0;
807 mountpoint = path->mnt->mnt_mountpoint;
808 path->dentry = mountpoint;
809 path->mnt = parent;
810 return 1;
811}
812
bab77ebf 813int follow_up(struct path *path)
1da177e4
LT		 814{
815 struct vfsmount *parent;
816 struct dentry *mountpoint;
99b7db7b
NP		 817
818 br_read_lock(vfsmount_lock);
bab77ebf
AV		 819 parent = path->mnt->mnt_parent;
820 if (parent == path->mnt) {
99b7db7b 821 br_read_unlock(vfsmount_lock);
1da177e4
LT		 822 return 0;
823 }
824 mntget(parent);
bab77ebf 825 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 826 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 827 dput(path->dentry);
828 path->dentry = mountpoint;
829 mntput(path->mnt);
830 path->mnt = parent;
1da177e4
LT		 831 return 1;
832}
833
b5c84bf6 834/*
9875cf80
DH		 835 * Perform an automount
836 * - return -EISDIR to tell follow_managed() to stop and return the path we
837 * were called with.
1da177e4 838 */
9875cf80
DH		 839static int follow_automount(struct path *path, unsigned flags,
840 bool *need_mntput)
31e6b01f 841{
9875cf80 842 struct vfsmount *mnt;
ea5b778a 843 int err;
9875cf80
DH		 844
845 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
846 return -EREMOTE;
847
6f45b656
DH		 848 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
849 * and this is the terminal part of the path.
850 */
851 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
852 return -EISDIR; /* we actually want to stop here */
853
9875cf80
DH		 854 /* We want to mount if someone is trying to open/create a file of any
855 * type under the mountpoint, wants to traverse through the mountpoint
856 * or wants to open the mounted directory.
857 *
858 * We don't want to mount if someone's just doing a stat and they've
859 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
860 * appended a '/' to the name.
861 */
862 if (!(flags & LOOKUP_FOLLOW) &&
863 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
864 LOOKUP_OPEN | LOOKUP_CREATE)))
865 return -EISDIR;
866
867 current->total_link_count++;
868 if (current->total_link_count >= 40)
869 return -ELOOP;
870
871 mnt = path->dentry->d_op->d_automount(path);
872 if (IS_ERR(mnt)) {
873 /*
874 * The filesystem is allowed to return -EISDIR here to indicate
875 * it doesn't want to automount. For instance, autofs would do
876 * this so that its userspace daemon can mount on this dentry.
877 *
878 * However, we can only permit this if it's a terminal point in
879 * the path being looked up; if it wasn't then the remainder of
880 * the path is inaccessible and we should say so.
881 */
882 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
883 return -EREMOTE;
884 return PTR_ERR(mnt);
31e6b01f 885 }
ea5b778a 886
9875cf80
DH		 887 if (!mnt) /* mount collision */
888 return 0;
31e6b01f 889
19a167af 890 err = finish_automount(mnt, path);
9875cf80 891
ea5b778a
DH		 892 switch (err) {
893 case -EBUSY:
894 /* Someone else made a mount here whilst we were busy */
19a167af 895 return 0;
ea5b778a 896 case 0:
ea5b778a
DH		 897 dput(path->dentry);
898 if (*need_mntput)
899 mntput(path->mnt);
900 path->mnt = mnt;
901 path->dentry = dget(mnt->mnt_root);
902 *need_mntput = true;
903 return 0;
19a167af
AV		 904 default:
905 return err;
ea5b778a 906 }
19a167af 907
463ffb2e
AV		 908}
909
9875cf80
DH		 910/*
911 * Handle a dentry that is managed in some way.
cc53ce53 912 * - Flagged for transit management (autofs)
9875cf80
DH		 913 * - Flagged as mountpoint
914 * - Flagged as automount point
915 *
916 * This may only be called in refwalk mode.
917 *
918 * Serialization is taken care of in namespace.c
919 */
920static int follow_managed(struct path *path, unsigned flags)
1da177e4 921{
9875cf80
DH		 922 unsigned managed;
923 bool need_mntput = false;
924 int ret;
925
926 /* Given that we're not holding a lock here, we retain the value in a
927 * local variable for each dentry as we look at it so that we don't see
928 * the components of that value change under us */
929 while (managed = ACCESS_ONCE(path->dentry->d_flags),
930 managed &= DCACHE_MANAGED_DENTRY,
931 unlikely(managed != 0)) {
cc53ce53
DH		 932 /* Allow the filesystem to manage the transit without i_mutex
933 * being held. */
934 if (managed & DCACHE_MANAGE_TRANSIT) {
935 BUG_ON(!path->dentry->d_op);
936 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 937 ret = path->dentry->d_op->d_manage(path->dentry,
938 false, false);
cc53ce53
DH		 939 if (ret < 0)
940 return ret == -EISDIR ? 0 : ret;
941 }
942
9875cf80
DH		 943 /* Transit to a mounted filesystem. */
944 if (managed & DCACHE_MOUNTED) {
945 struct vfsmount *mounted = lookup_mnt(path);
946 if (mounted) {
947 dput(path->dentry);
948 if (need_mntput)
949 mntput(path->mnt);
950 path->mnt = mounted;
951 path->dentry = dget(mounted->mnt_root);
952 need_mntput = true;
953 continue;
954 }
955
956 /* Something is mounted on this dentry in another
957 * namespace and/or whatever was mounted there in this
958 * namespace got unmounted before we managed to get the
959 * vfsmount_lock */
960 }
961
962 /* Handle an automount point */
963 if (managed & DCACHE_NEED_AUTOMOUNT) {
964 ret = follow_automount(path, flags, &need_mntput);
965 if (ret < 0)
966 return ret == -EISDIR ? 0 : ret;
967 continue;
968 }
969
970 /* We didn't change the current path point */
971 break;
1da177e4 972 }
9875cf80 973 return 0;
1da177e4
LT		 974}
975
cc53ce53 976int follow_down_one(struct path *path)
1da177e4
LT		 977{
978 struct vfsmount *mounted;
979
1c755af4 980 mounted = lookup_mnt(path);
1da177e4 981 if (mounted) {
9393bd07
AV		 982 dput(path->dentry);
983 mntput(path->mnt);
984 path->mnt = mounted;
985 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 986 return 1;
987 }
988 return 0;
989}
990
9875cf80
DH		 991/*
992 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
cc53ce53 993 * meet a managed dentry and we're not walking to "..". True is returned to
9875cf80
DH		 994 * continue, false to abort.
995 */
996static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
997 struct inode **inode, bool reverse_transit)
998{
999 while (d_mountpoint(path->dentry)) {
1000 struct vfsmount *mounted;
ab90911f
DH		 1001 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1002 !reverse_transit &&
1003 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1004 return false;
9875cf80
DH		 1005 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1006 if (!mounted)
1007 break;
1008 path->mnt = mounted;
1009 path->dentry = mounted->mnt_root;
1010 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1011 *inode = path->dentry->d_inode;
1012 }
1013
1014 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1015 return reverse_transit;
1016 return true;
1017}
1018
31e6b01f
NP		 1019static int follow_dotdot_rcu(struct nameidata *nd)
1020{
1021 struct inode *inode = nd->inode;
1022
1023 set_root_rcu(nd);
1024
9875cf80 1025 while (1) {
31e6b01f
NP		 1026 if (nd->path.dentry == nd->root.dentry &&
1027 nd->path.mnt == nd->root.mnt) {
1028 break;
1029 }
1030 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1031 struct dentry *old = nd->path.dentry;
1032 struct dentry *parent = old->d_parent;
1033 unsigned seq;
1034
1035 seq = read_seqcount_begin(&parent->d_seq);
1036 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 1037 goto failed;
31e6b01f
NP		 1038 inode = parent->d_inode;
1039 nd->path.dentry = parent;
1040 nd->seq = seq;
1041 break;
1042 }
1043 if (!follow_up_rcu(&nd->path))
1044 break;
1045 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1046 inode = nd->path.dentry->d_inode;
1047 }
9875cf80 1048 __follow_mount_rcu(nd, &nd->path, &inode, true);
31e6b01f 1049 nd->inode = inode;
31e6b01f 1050 return 0;
ef7562d5
AV		 1051
1052failed:
1053 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1054 if (!(nd->flags & LOOKUP_ROOT))
1055 nd->root.mnt = NULL;
ef7562d5
AV		 1056 rcu_read_unlock();
1057 br_read_unlock(vfsmount_lock);
1058 return -ECHILD;
31e6b01f
NP		 1059}
1060
cc53ce53
DH		 1061/*
1062 * Follow down to the covering mount currently visible to userspace. At each
1063 * point, the filesystem owning that dentry may be queried as to whether the
1064 * caller is permitted to proceed or not.
1065 *
1066 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1067 * being true).
1068 */
1069int follow_down(struct path *path, bool mounting_here)
1070{
1071 unsigned managed;
1072 int ret;
1073
1074 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1075 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1076 /* Allow the filesystem to manage the transit without i_mutex
1077 * being held.
1078 *
1079 * We indicate to the filesystem if someone is trying to mount
1080 * something here. This gives autofs the chance to deny anyone
1081 * other than its daemon the right to mount on its
1082 * superstructure.
1083 *
1084 * The filesystem may sleep at this point.
1085 */
1086 if (managed & DCACHE_MANAGE_TRANSIT) {
1087 BUG_ON(!path->dentry->d_op);
1088 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 1089 ret = path->dentry->d_op->d_manage(
1090 path->dentry, mounting_here, false);
cc53ce53
DH		 1091 if (ret < 0)
1092 return ret == -EISDIR ? 0 : ret;
1093 }
1094
1095 /* Transit to a mounted filesystem. */
1096 if (managed & DCACHE_MOUNTED) {
1097 struct vfsmount *mounted = lookup_mnt(path);
1098 if (!mounted)
1099 break;
1100 dput(path->dentry);
1101 mntput(path->mnt);
1102 path->mnt = mounted;
1103 path->dentry = dget(mounted->mnt_root);
1104 continue;
1105 }
1106
1107 /* Don't handle automount points here */
1108 break;
1109 }
1110 return 0;
1111}
1112
9875cf80
DH		 1113/*
1114 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1115 */
1116static void follow_mount(struct path *path)
1117{
1118 while (d_mountpoint(path->dentry)) {
1119 struct vfsmount *mounted = lookup_mnt(path);
1120 if (!mounted)
1121 break;
1122 dput(path->dentry);
1123 mntput(path->mnt);
1124 path->mnt = mounted;
1125 path->dentry = dget(mounted->mnt_root);
1126 }
1127}
1128
31e6b01f 1129static void follow_dotdot(struct nameidata *nd)
1da177e4 1130{
2a737871 1131 set_root(nd);
e518ddb7 1132
1da177e4 1133 while(1) {
4ac91378 1134 struct dentry *old = nd->path.dentry;
1da177e4 1135
2a737871
AV		 1136 if (nd->path.dentry == nd->root.dentry &&
1137 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1138 break;
1139 }
4ac91378 1140 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1141 /* rare case of legitimate dget_parent()... */
1142 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1143 dput(old);
1144 break;
1145 }
3088dd70 1146 if (!follow_up(&nd->path))
1da177e4 1147 break;
1da177e4 1148 }
79ed0226 1149 follow_mount(&nd->path);
31e6b01f 1150 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1151}
1152
baa03890
NP		 1153/*
1154 * Allocate a dentry with name and parent, and perform a parent
1155 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1156 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1157 * have verified that no child exists while under i_mutex.
1158 */
1159static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1160 struct qstr *name, struct nameidata *nd)
1161{
1162 struct inode *inode = parent->d_inode;
1163 struct dentry *dentry;
1164 struct dentry *old;
1165
1166 /* Don't create child dentry for a dead directory. */
1167 if (unlikely(IS_DEADDIR(inode)))
1168 return ERR_PTR(-ENOENT);
1169
1170 dentry = d_alloc(parent, name);
1171 if (unlikely(!dentry))
1172 return ERR_PTR(-ENOMEM);
1173
1174 old = inode->i_op->lookup(inode, dentry, nd);
1175 if (unlikely(old)) {
1176 dput(dentry);
1177 dentry = old;
1178 }
1179 return dentry;
1180}
1181
1da177e4
LT		 1182/*
1183 * It's more convoluted than I'd like it to be, but... it's still fairly
1184 * small and for now I'd prefer to have fast path as straight as possible.
1185 * It _is_ time-critical.
1186 */
1187static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1188 struct path *path, struct inode **inode)
1da177e4 1189{
4ac91378 1190 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1191 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1192 int need_reval = 1;
1193 int status = 1;
9875cf80
DH		 1194 int err;
1195
b04f784e
NP		 1196 /*
1197 * Rename seqlock is not required here because in the off chance
1198 * of a false negative due to a concurrent rename, we're going to
1199 * do the non-racy lookup, below.
1200 */
31e6b01f
NP		 1201 if (nd->flags & LOOKUP_RCU) {
1202 unsigned seq;
31e6b01f
NP		 1203 *inode = nd->inode;
1204 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1205 if (!dentry)
1206 goto unlazy;
1207
31e6b01f
NP		 1208 /* Memory barrier in read_seqcount_begin of child is enough */
1209 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1210 return -ECHILD;
31e6b01f 1211 nd->seq = seq;
5a18fff2 1212
24643087 1213 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1214 status = d_revalidate(dentry, nd);
1215 if (unlikely(status <= 0)) {
1216 if (status != -ECHILD)
1217 need_reval = 0;
1218 goto unlazy;
1219 }
24643087 1220 }
31e6b01f
NP		 1221 path->mnt = mnt;
1222 path->dentry = dentry;
9875cf80
DH		 1223 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1224 return 0;
5a18fff2
AV		 1225unlazy:
1226 if (dentry) {
1227 if (nameidata_dentry_drop_rcu(nd, dentry))
1228 return -ECHILD;
1229 } else {
1230 if (nameidata_drop_rcu(nd))
1231 return -ECHILD;
1232 }
1233 } else {
1234 dentry = __d_lookup(parent, name);
9875cf80 1235 }
5a18fff2
AV		 1236
1237retry:
1238 if (unlikely(!dentry)) {
1239 struct inode *dir = parent->d_inode;
1240 BUG_ON(nd->inode != dir);
1241
1242 mutex_lock(&dir->i_mutex);
1243 dentry = d_lookup(parent, name);
1244 if (likely(!dentry)) {
1245 dentry = d_alloc_and_lookup(parent, name, nd);
1246 if (IS_ERR(dentry)) {
1247 mutex_unlock(&dir->i_mutex);
1248 return PTR_ERR(dentry);
1249 }
1250 /* known good */
1251 need_reval = 0;
1252 status = 1;
1253 }
1254 mutex_unlock(&dir->i_mutex);
1255 }
1256 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1257 status = d_revalidate(dentry, nd);
1258 if (unlikely(status <= 0)) {
1259 if (status < 0) {
1260 dput(dentry);
1261 return status;
1262 }
1263 if (!d_invalidate(dentry)) {
1264 dput(dentry);
1265 dentry = NULL;
1266 need_reval = 1;
1267 goto retry;
1268 }
24643087 1269 }
5a18fff2 1270
9875cf80
DH		 1271 path->mnt = mnt;
1272 path->dentry = dentry;
1273 err = follow_managed(path, nd->flags);
89312214
IK		 1274 if (unlikely(err < 0)) {
1275 path_put_conditional(path, nd);
9875cf80 1276 return err;
89312214 1277 }
9875cf80 1278 *inode = path->dentry->d_inode;
1da177e4 1279 return 0;
1da177e4
LT		 1280}
1281
52094c8a
AV		 1282static inline int may_lookup(struct nameidata *nd)
1283{
1284 if (nd->flags & LOOKUP_RCU) {
1285 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1286 if (err != -ECHILD)
1287 return err;
1288 if (nameidata_drop_rcu(nd))
1289 return -ECHILD;
1290 }
1291 return exec_permission(nd->inode, 0);
1292}
1293
9856fa1b
AV		 1294static inline int handle_dots(struct nameidata *nd, int type)
1295{
1296 if (type == LAST_DOTDOT) {
1297 if (nd->flags & LOOKUP_RCU) {
1298 if (follow_dotdot_rcu(nd))
1299 return -ECHILD;
1300 } else
1301 follow_dotdot(nd);
1302 }
1303 return 0;
1304}
1305
951361f9
AV		 1306static void terminate_walk(struct nameidata *nd)
1307{
1308 if (!(nd->flags & LOOKUP_RCU)) {
1309 path_put(&nd->path);
1310 } else {
1311 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1312 if (!(nd->flags & LOOKUP_ROOT))
1313 nd->root.mnt = NULL;
951361f9
AV		 1314 rcu_read_unlock();
1315 br_read_unlock(vfsmount_lock);
1316 }
1317}
1318
ce57dfc1
AV		 1319static inline int walk_component(struct nameidata *nd, struct path *path,
1320 struct qstr *name, int type, int follow)
1321{
1322 struct inode *inode;
1323 int err;
1324 /*
1325 * "." and ".." are special - ".." especially so because it has
1326 * to be able to know about the current root directory and
1327 * parent relationships.
1328 */
1329 if (unlikely(type != LAST_NORM))
1330 return handle_dots(nd, type);
1331 err = do_lookup(nd, name, path, &inode);
1332 if (unlikely(err)) {
1333 terminate_walk(nd);
1334 return err;
1335 }
1336 if (!inode) {
1337 path_to_nameidata(path, nd);
1338 terminate_walk(nd);
1339 return -ENOENT;
1340 }
1341 if (unlikely(inode->i_op->follow_link) && follow) {
1342 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
1343 return -ECHILD;
1344 BUG_ON(inode != path->dentry->d_inode);
1345 return 1;
1346 }
1347 path_to_nameidata(path, nd);
1348 nd->inode = inode;
1349 return 0;
1350}
1351
b356379a
AV		 1352/*
1353 * This limits recursive symlink follows to 8, while
1354 * limiting consecutive symlinks to 40.
1355 *
1356 * Without that kind of total limit, nasty chains of consecutive
1357 * symlinks can cause almost arbitrarily long lookups.
1358 */
1359static inline int nested_symlink(struct path *path, struct nameidata *nd)
1360{
1361 int res;
1362
1363 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1364 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1365 path_put_conditional(path, nd);
1366 path_put(&nd->path);
1367 return -ELOOP;
1368 }
1369
1370 nd->depth++;
1371 current->link_count++;
1372
1373 do {
1374 struct path link = *path;
1375 void *cookie;
574197e0
AV		 1376
1377 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1378 if (!res)
1379 res = walk_component(nd, path, &nd->last,
1380 nd->last_type, LOOKUP_FOLLOW);
574197e0 1381 put_link(nd, &link, cookie);
b356379a
AV		 1382 } while (res > 0);
1383
1384 current->link_count--;
1385 nd->depth--;
1386 return res;
1387}
1388
1da177e4
LT		 1389/*
1390 * Name resolution.
ea3834d9
PM		 1391 * This is the basic name resolution function, turning a pathname into
1392 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1393 *
ea3834d9
PM		 1394 * Returns 0 and nd will have valid dentry and mnt on success.
1395 * Returns error and drops reference to input namei data on failure.
1da177e4 1396 */
6de88d72 1397static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1398{
1399 struct path next;
1da177e4
LT		 1400 int err;
1401 unsigned int lookup_flags = nd->flags;
1402
1403 while (*name=='/')
1404 name++;
1405 if (!*name)
086e183a 1406 return 0;
1da177e4 1407
1da177e4
LT		 1408 /* At this point we know we have a real path component. */
1409 for(;;) {
1410 unsigned long hash;
1411 struct qstr this;
1412 unsigned int c;
fe479a58 1413 int type;
1da177e4 1414
cdce5d6b 1415 nd->flags |= LOOKUP_CONTINUE;
52094c8a
AV		 1416
1417 err = may_lookup(nd);
1da177e4
LT		 1418 if (err)
1419 break;
1420
1421 this.name = name;
1422 c = *(const unsigned char *)name;
1423
1424 hash = init_name_hash();
1425 do {
1426 name++;
1427 hash = partial_name_hash(c, hash);
1428 c = *(const unsigned char *)name;
1429 } while (c && (c != '/'));
1430 this.len = name - (const char *) this.name;
1431 this.hash = end_name_hash(hash);
1432
fe479a58
AV		 1433 type = LAST_NORM;
1434 if (this.name[0] == '.') switch (this.len) {
1435 case 2:
16c2cd71 1436 if (this.name[1] == '.') {
fe479a58 1437 type = LAST_DOTDOT;
16c2cd71
AV		 1438 nd->flags |= LOOKUP_JUMPED;
1439 }
fe479a58
AV		 1440 break;
1441 case 1:
1442 type = LAST_DOT;
1443 }
5a202bcd
AV		 1444 if (likely(type == LAST_NORM)) {
1445 struct dentry *parent = nd->path.dentry;
16c2cd71 1446 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1447 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1448 err = parent->d_op->d_hash(parent, nd->inode,
1449 &this);
1450 if (err < 0)
1451 break;
1452 }
1453 }
fe479a58 1454
1da177e4
LT		 1455 /* remove trailing slashes? */
1456 if (!c)
1457 goto last_component;
1458 while (*++name == '/');
1459 if (!*name)
b356379a 1460 goto last_component;
1da177e4 1461
ce57dfc1
AV		 1462 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1463 if (err < 0)
1464 return err;
1da177e4 1465
ce57dfc1 1466 if (err) {
b356379a 1467 err = nested_symlink(&next, nd);
1da177e4 1468 if (err)
a7472bab 1469 return err;
31e6b01f 1470 }
1da177e4 1471 err = -ENOTDIR;
31e6b01f 1472 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1473 break;
1474 continue;
1475 /* here ends the main loop */
1476
1da177e4 1477last_component:
f55eab82
TM		 1478 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1479 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
b356379a
AV		 1480 nd->last = this;
1481 nd->last_type = type;
086e183a 1482 return 0;
1da177e4 1483 }
951361f9 1484 terminate_walk(nd);
1da177e4
LT		 1485 return err;
1486}
1487
70e9b357
AV		 1488static int path_init(int dfd, const char *name, unsigned int flags,
1489 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1490{
1491 int retval = 0;
1492 int fput_needed;
1493 struct file *file;
1494
1495 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1496 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1497 nd->depth = 0;
5b6ca027
AV		 1498 if (flags & LOOKUP_ROOT) {
1499 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1500 if (*name) {
1501 if (!inode->i_op->lookup)
1502 return -ENOTDIR;
1503 retval = inode_permission(inode, MAY_EXEC);
1504 if (retval)
1505 return retval;
1506 }
5b6ca027
AV		 1507 nd->path = nd->root;
1508 nd->inode = inode;
1509 if (flags & LOOKUP_RCU) {
1510 br_read_lock(vfsmount_lock);
1511 rcu_read_lock();
1512 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1513 } else {
1514 path_get(&nd->path);
1515 }
1516 return 0;
1517 }
1518
31e6b01f 1519 nd->root.mnt = NULL;
31e6b01f
NP		 1520
1521 if (*name=='/') {
e41f7d4e
AV		 1522 if (flags & LOOKUP_RCU) {
1523 br_read_lock(vfsmount_lock);
1524 rcu_read_lock();
1525 set_root_rcu(nd);
1526 } else {
1527 set_root(nd);
1528 path_get(&nd->root);
1529 }
1530 nd->path = nd->root;
31e6b01f 1531 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1532 if (flags & LOOKUP_RCU) {
1533 struct fs_struct *fs = current->fs;
1534 unsigned seq;
31e6b01f 1535
e41f7d4e
AV		 1536 br_read_lock(vfsmount_lock);
1537 rcu_read_lock();
c28cc364 1538
e41f7d4e
AV		 1539 do {
1540 seq = read_seqcount_begin(&fs->seq);
1541 nd->path = fs->pwd;
1542 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1543 } while (read_seqcount_retry(&fs->seq, seq));
1544 } else {
1545 get_fs_pwd(current->fs, &nd->path);
1546 }
31e6b01f
NP		 1547 } else {
1548 struct dentry *dentry;
1549
1abf0c71 1550 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1551 retval = -EBADF;
1552 if (!file)
1553 goto out_fail;
1554
1555 dentry = file->f_path.dentry;
1556
f52e0c11
AV		 1557 if (*name) {
1558 retval = -ENOTDIR;
1559 if (!S_ISDIR(dentry->d_inode->i_mode))
1560 goto fput_fail;
31e6b01f 1561
f52e0c11
AV		 1562 retval = file_permission(file, MAY_EXEC);
1563 if (retval)
1564 goto fput_fail;
1565 }
31e6b01f
NP		 1566
1567 nd->path = file->f_path;
e41f7d4e
AV		 1568 if (flags & LOOKUP_RCU) {
1569 if (fput_needed)
70e9b357 1570 *fp = file;
e41f7d4e
AV		 1571 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1572 br_read_lock(vfsmount_lock);
1573 rcu_read_lock();
1574 } else {
1575 path_get(&file->f_path);
1576 fput_light(file, fput_needed);
1577 }
31e6b01f 1578 }
31e6b01f 1579
31e6b01f 1580 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1581 return 0;
2dfdd266 1582
9b4a9b14
AV		 1583fput_fail:
1584 fput_light(file, fput_needed);
1585out_fail:
1586 return retval;
1587}
1588
bd92d7fe
AV		 1589static inline int lookup_last(struct nameidata *nd, struct path *path)
1590{
1591 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1592 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1593
1594 nd->flags &= ~LOOKUP_PARENT;
1595 return walk_component(nd, path, &nd->last, nd->last_type,
1596 nd->flags & LOOKUP_FOLLOW);
1597}
1598
9b4a9b14 1599/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1600static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1601 unsigned int flags, struct nameidata *nd)
1602{
70e9b357 1603 struct file *base = NULL;
bd92d7fe
AV		 1604 struct path path;
1605 int err;
31e6b01f
NP		 1606
1607 /*
1608 * Path walking is largely split up into 2 different synchronisation
1609 * schemes, rcu-walk and ref-walk (explained in
1610 * Documentation/filesystems/path-lookup.txt). These share much of the
1611 * path walk code, but some things particularly setup, cleanup, and
1612 * following mounts are sufficiently divergent that functions are
1613 * duplicated. Typically there is a function foo(), and its RCU
1614 * analogue, foo_rcu().
1615 *
1616 * -ECHILD is the error number of choice (just to avoid clashes) that
1617 * is returned if some aspect of an rcu-walk fails. Such an error must
1618 * be handled by restarting a traditional ref-walk (which will always
1619 * be able to complete).
1620 */
bd92d7fe 1621 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1622
bd92d7fe
AV		 1623 if (unlikely(err))
1624 return err;
ee0827cd
AV		 1625
1626 current->total_link_count = 0;
bd92d7fe
AV		 1627 err = link_path_walk(name, nd);
1628
1629 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1630 err = lookup_last(nd, &path);
1631 while (err > 0) {
1632 void *cookie;
1633 struct path link = path;
bd92d7fe 1634 nd->flags |= LOOKUP_PARENT;
574197e0 1635 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1636 if (!err)
1637 err = lookup_last(nd, &path);
574197e0 1638 put_link(nd, &link, cookie);
bd92d7fe
AV		 1639 }
1640 }
ee0827cd
AV		 1641
1642 if (nd->flags & LOOKUP_RCU) {
4455ca62 1643 /* went all way through without dropping RCU */
bd92d7fe 1644 BUG_ON(err);
4455ca62 1645 if (nameidata_drop_rcu_last(nd))
bd92d7fe 1646 err = -ECHILD;
ee0827cd
AV		 1647 }
1648
bd92d7fe
AV		 1649 if (!err)
1650 err = handle_reval_path(nd);
1651
1652 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1653 if (!nd->inode->i_op->lookup) {
1654 path_put(&nd->path);
1655 return -ENOTDIR;
1656 }
1657 }
16c2cd71 1658
70e9b357
AV		 1659 if (base)
1660 fput(base);
ee0827cd 1661
5b6ca027 1662 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1663 path_put(&nd->root);
1664 nd->root.mnt = NULL;
1665 }
bd92d7fe 1666 return err;
ee0827cd 1667}
31e6b01f 1668
ee0827cd
AV		 1669static int do_path_lookup(int dfd, const char *name,
1670 unsigned int flags, struct nameidata *nd)
1671{
1672 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1673 if (unlikely(retval == -ECHILD))
1674 retval = path_lookupat(dfd, name, flags, nd);
1675 if (unlikely(retval == -ESTALE))
1676 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1677
1678 if (likely(!retval)) {
1679 if (unlikely(!audit_dummy_context())) {
1680 if (nd->path.dentry && nd->inode)
1681 audit_inode(name, nd->path.dentry);
1682 }
1683 }
170aa3d0 1684 return retval;
1da177e4
LT		 1685}
1686
c9c6cac0 1687int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1688{
c9c6cac0 1689 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1690}
1691
d1811465
AV		 1692int kern_path(const char *name, unsigned int flags, struct path *path)
1693{
1694 struct nameidata nd;
1695 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1696 if (!res)
1697 *path = nd.path;
1698 return res;
1699}
1700
16f18200
JJS		 1701/**
1702 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1703 * @dentry: pointer to dentry of the base directory
1704 * @mnt: pointer to vfs mount of the base directory
1705 * @name: pointer to file name
1706 * @flags: lookup flags
1707 * @nd: pointer to nameidata
1708 */
1709int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1710 const char *name, unsigned int flags,
1711 struct nameidata *nd)
1712{
5b6ca027
AV		 1713 nd->root.dentry = dentry;
1714 nd->root.mnt = mnt;
1715 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1716 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
16f18200
JJS		 1717}
1718
eead1911
CH		 1719static struct dentry *__lookup_hash(struct qstr *name,
1720 struct dentry *base, struct nameidata *nd)
1da177e4 1721{
81fca444 1722 struct inode *inode = base->d_inode;
057f6c01 1723 struct dentry *dentry;
1da177e4
LT		 1724 int err;
1725
b74c79e9 1726 err = exec_permission(inode, 0);
81fca444
CH		 1727 if (err)
1728 return ERR_PTR(err);
1da177e4 1729
b04f784e
NP		 1730 /*
1731 * Don't bother with __d_lookup: callers are for creat as
1732 * well as unlink, so a lot of the time it would cost
1733 * a double lookup.
6e6b1bd1 1734 */
b04f784e 1735 dentry = d_lookup(base, name);
6e6b1bd1 1736
fb045adb 1737 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1738 dentry = do_revalidate(dentry, nd);
1739
baa03890
NP		 1740 if (!dentry)
1741 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1742
1da177e4
LT		 1743 return dentry;
1744}
1745
057f6c01
JM		 1746/*
1747 * Restricted form of lookup. Doesn't follow links, single-component only,
1748 * needs parent already locked. Doesn't follow mounts.
1749 * SMP-safe.
1750 */
eead1911 1751static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1752{
4ac91378 1753 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1754}
1755
eead1911 1756/**
a6b91919 1757 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1758 * @name: pathname component to lookup
1759 * @base: base directory to lookup from
1760 * @len: maximum length @len should be interpreted to
1761 *
a6b91919
RD		 1762 * Note that this routine is purely a helper for filesystem usage and should
1763 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1764 * nameidata argument is passed to the filesystem methods and a filesystem
1765 * using this helper needs to be prepared for that.
1766 */
057f6c01
JM		 1767struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1768{
057f6c01 1769 struct qstr this;
6a96ba54
AV		 1770 unsigned long hash;
1771 unsigned int c;
057f6c01 1772
2f9092e1
DW		 1773 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1774
6a96ba54
AV		 1775 this.name = name;
1776 this.len = len;
1777 if (!len)
1778 return ERR_PTR(-EACCES);
1779
1780 hash = init_name_hash();
1781 while (len--) {
1782 c = *(const unsigned char *)name++;
1783 if (c == '/' || c == '\0')
1784 return ERR_PTR(-EACCES);
1785 hash = partial_name_hash(c, hash);
1786 }
1787 this.hash = end_name_hash(hash);
5a202bcd
AV		 1788 /*
1789 * See if the low-level filesystem might want
1790 * to use its own hash..
1791 */
1792 if (base->d_flags & DCACHE_OP_HASH) {
1793 int err = base->d_op->d_hash(base, base->d_inode, &this);
1794 if (err < 0)
1795 return ERR_PTR(err);
1796 }
eead1911 1797
49705b77 1798 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1799}
1800
2d8f3038
AV		 1801int user_path_at(int dfd, const char __user *name, unsigned flags,
1802 struct path *path)
1da177e4 1803{
2d8f3038 1804 struct nameidata nd;
f52e0c11 1805 char *tmp = getname_flags(name, flags);
1da177e4 1806 int err = PTR_ERR(tmp);
1da177e4 1807 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1808
1809 BUG_ON(flags & LOOKUP_PARENT);
1810
1811 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1812 putname(tmp);
2d8f3038
AV		 1813 if (!err)
1814 *path = nd.path;
1da177e4
LT		 1815 }
1816 return err;
1817}
1818
2ad94ae6
AV		 1819static int user_path_parent(int dfd, const char __user *path,
1820 struct nameidata *nd, char **name)
1821{
1822 char *s = getname(path);
1823 int error;
1824
1825 if (IS_ERR(s))
1826 return PTR_ERR(s);
1827
1828 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1829 if (error)
1830 putname(s);
1831 else
1832 *name = s;
1833
1834 return error;
1835}
1836
1da177e4
LT		 1837/*
1838 * It's inline, so penalty for filesystems that don't use sticky bit is
1839 * minimal.
1840 */
1841static inline int check_sticky(struct inode *dir, struct inode *inode)
1842{
da9592ed
DH		 1843 uid_t fsuid = current_fsuid();
1844
1da177e4
LT		 1845 if (!(dir->i_mode & S_ISVTX))
1846 return 0;
da9592ed 1847 if (inode->i_uid == fsuid)
1da177e4 1848 return 0;
da9592ed 1849 if (dir->i_uid == fsuid)
1da177e4
LT		 1850 return 0;
1851 return !capable(CAP_FOWNER);
1852}
1853
1854/*
1855 * Check whether we can remove a link victim from directory dir, check
1856 * whether the type of victim is right.
1857 * 1. We can't do it if dir is read-only (done in permission())
1858 * 2. We should have write and exec permissions on dir
1859 * 3. We can't remove anything from append-only dir
1860 * 4. We can't do anything with immutable dir (done in permission())
1861 * 5. If the sticky bit on dir is set we should either
1862 * a. be owner of dir, or
1863 * b. be owner of victim, or
1864 * c. have CAP_FOWNER capability
1865 * 6. If the victim is append-only or immutable we can't do antyhing with
1866 * links pointing to it.
1867 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1868 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1869 * 9. We can't remove a root or mountpoint.
1870 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1871 * nfs_async_unlink().
1872 */
858119e1 1873static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1874{
1875 int error;
1876
1877 if (!victim->d_inode)
1878 return -ENOENT;
1879
1880 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1881 audit_inode_child(victim, dir);
1da177e4 1882
f419a2e3 1883 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1884 if (error)
1885 return error;
1886 if (IS_APPEND(dir))
1887 return -EPERM;
1888 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1889 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1890 return -EPERM;
1891 if (isdir) {
1892 if (!S_ISDIR(victim->d_inode->i_mode))
1893 return -ENOTDIR;
1894 if (IS_ROOT(victim))
1895 return -EBUSY;
1896 } else if (S_ISDIR(victim->d_inode->i_mode))
1897 return -EISDIR;
1898 if (IS_DEADDIR(dir))
1899 return -ENOENT;
1900 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1901 return -EBUSY;
1902 return 0;
1903}
1904
1905/* Check whether we can create an object with dentry child in directory
1906 * dir.
1907 * 1. We can't do it if child already exists (open has special treatment for
1908 * this case, but since we are inlined it's OK)
1909 * 2. We can't do it if dir is read-only (done in permission())
1910 * 3. We should have write and exec permissions on dir
1911 * 4. We can't do it if dir is immutable (done in permission())
1912 */
a95164d9 1913static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1914{
1915 if (child->d_inode)
1916 return -EEXIST;
1917 if (IS_DEADDIR(dir))
1918 return -ENOENT;
f419a2e3 1919 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1920}
1921
1da177e4
LT		 1922/*
1923 * p1 and p2 should be directories on the same fs.
1924 */
1925struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1926{
1927 struct dentry *p;
1928
1929 if (p1 == p2) {
f2eace23 1930 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1931 return NULL;
1932 }
1933
a11f3a05 1934 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1935
e2761a11
OH		 1936 p = d_ancestor(p2, p1);
1937 if (p) {
1938 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1939 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1940 return p;
1da177e4
LT		 1941 }
1942
e2761a11
OH		 1943 p = d_ancestor(p1, p2);
1944 if (p) {
1945 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1946 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1947 return p;
1da177e4
LT		 1948 }
1949
f2eace23
IM		 1950 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1951 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1952 return NULL;
1953}
1954
1955void unlock_rename(struct dentry *p1, struct dentry *p2)
1956{
1b1dcc1b 1957 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1958 if (p1 != p2) {
1b1dcc1b 1959 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1960 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1961 }
1962}
1963
1964int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1965 struct nameidata *nd)
1966{
a95164d9 1967 int error = may_create(dir, dentry);
1da177e4
LT		 1968
1969 if (error)
1970 return error;
1971
acfa4380 1972 if (!dir->i_op->create)
1da177e4
LT		 1973 return -EACCES; /* shouldn't it be ENOSYS? */
1974 mode &= S_IALLUGO;
1975 mode |= S_IFREG;
1976 error = security_inode_create(dir, dentry, mode);
1977 if (error)
1978 return error;
1da177e4 1979 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1980 if (!error)
f38aa942 1981 fsnotify_create(dir, dentry);
1da177e4
LT		 1982 return error;
1983}
1984
73d049a4 1985static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1986{
3fb64190 1987 struct dentry *dentry = path->dentry;
1da177e4
LT		 1988 struct inode *inode = dentry->d_inode;
1989 int error;
1990
bcda7652
AV		 1991 /* O_PATH? */
1992 if (!acc_mode)
1993 return 0;
1994
1da177e4
LT		 1995 if (!inode)
1996 return -ENOENT;
1997
c8fe8f30
CH		 1998 switch (inode->i_mode & S_IFMT) {
1999 case S_IFLNK:
1da177e4 2000 return -ELOOP;
c8fe8f30
CH		 2001 case S_IFDIR:
2002 if (acc_mode & MAY_WRITE)
2003 return -EISDIR;
2004 break;
2005 case S_IFBLK:
2006 case S_IFCHR:
3fb64190 2007 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2008 return -EACCES;
c8fe8f30
CH		 2009 /*FALLTHRU*/
2010 case S_IFIFO:
2011 case S_IFSOCK:
1da177e4 2012 flag &= ~O_TRUNC;
c8fe8f30 2013 break;
4a3fd211 2014 }
b41572e9 2015
3fb64190 2016 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2017 if (error)
2018 return error;
6146f0d5 2019
1da177e4
LT		 2020 /*
2021 * An append-only file must be opened in append mode for writing.
2022 */
2023 if (IS_APPEND(inode)) {
8737c930 2024 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2025 return -EPERM;
1da177e4 2026 if (flag & O_TRUNC)
7715b521 2027 return -EPERM;
1da177e4
LT		 2028 }
2029
2030 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 2031 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2032 return -EPERM;
1da177e4
LT		 2033
2034 /*
2035 * Ensure there are no outstanding leases on the file.
2036 */
b65a9cfc 2037 return break_lease(inode, flag);
7715b521 2038}
1da177e4 2039
e1181ee6 2040static int handle_truncate(struct file *filp)
7715b521 2041{
e1181ee6 2042 struct path *path = &filp->f_path;
7715b521
AV		 2043 struct inode *inode = path->dentry->d_inode;
2044 int error = get_write_access(inode);
2045 if (error)
2046 return error;
2047 /*
2048 * Refuse to truncate files with mandatory locks held on them.
2049 */
2050 error = locks_verify_locked(inode);
2051 if (!error)
ea0d3ab2 2052 error = security_path_truncate(path);
7715b521
AV		 2053 if (!error) {
2054 error = do_truncate(path->dentry, 0,
2055 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2056 filp);
7715b521
AV		 2057 }
2058 put_write_access(inode);
acd0c935 2059 return error;
1da177e4
LT		 2060}
2061
d57999e1
DH		 2062/*
2063 * Note that while the flag value (low two bits) for sys_open means:
2064 * 00 - read-only
2065 * 01 - write-only
2066 * 10 - read-write
2067 * 11 - special
2068 * it is changed into
2069 * 00 - no permissions needed
2070 * 01 - read-permission
2071 * 10 - write-permission
2072 * 11 - read-write
2073 * for the internal routines (ie open_namei()/follow_link() etc)
2074 * This is more logical, and also allows the 00 "no perm needed"
2075 * to be used for symlinks (where the permissions are checked
2076 * later).
2077 *
2078*/
2079static inline int open_to_namei_flags(int flag)
2080{
2081 if ((flag+1) & O_ACCMODE)
2082 flag++;
2083 return flag;
2084}
2085
31e6b01f 2086/*
fe2d35ff 2087 * Handle the last step of open()
31e6b01f 2088 */
fb1cc555 2089static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2090 const struct open_flags *op, const char *pathname)
fb1cc555 2091{
a1e28038 2092 struct dentry *dir = nd->path.dentry;
6c0d46c4 2093 struct dentry *dentry;
ca344a89 2094 int open_flag = op->open_flag;
6c0d46c4 2095 int will_truncate = open_flag & O_TRUNC;
ca344a89 2096 int want_write = 0;
bcda7652 2097 int acc_mode = op->acc_mode;
fb1cc555 2098 struct file *filp;
16c2cd71 2099 int error;
1f36f774 2100
c3e380b0
AV		 2101 nd->flags &= ~LOOKUP_PARENT;
2102 nd->flags |= op->intent;
2103
1f36f774
AV		 2104 switch (nd->last_type) {
2105 case LAST_DOTDOT:
176306f5 2106 case LAST_DOT:
fe2d35ff
AV		 2107 error = handle_dots(nd, nd->last_type);
2108 if (error)
2109 return ERR_PTR(error);
1f36f774 2110 /* fallthrough */
1f36f774 2111 case LAST_ROOT:
fe2d35ff
AV		 2112 if (nd->flags & LOOKUP_RCU) {
2113 if (nameidata_drop_rcu_last(nd))
2114 return ERR_PTR(-ECHILD);
2115 }
16c2cd71
AV		 2116 error = handle_reval_path(nd);
2117 if (error)
2118 goto exit;
fe2d35ff 2119 audit_inode(pathname, nd->path.dentry);
ca344a89 2120 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2121 error = -EISDIR;
2122 goto exit;
2123 }
2124 goto ok;
1f36f774 2125 case LAST_BIND:
fe2d35ff 2126 /* can't be RCU mode here */
16c2cd71
AV		 2127 error = handle_reval_path(nd);
2128 if (error)
2129 goto exit;
1f36f774 2130 audit_inode(pathname, dir);
67ee3ad2 2131 goto ok;
1f36f774 2132 }
67ee3ad2 2133
ca344a89 2134 if (!(open_flag & O_CREAT)) {
bcda7652 2135 int symlink_ok = 0;
fe2d35ff
AV		 2136 if (nd->last.name[nd->last.len])
2137 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2138 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2139 symlink_ok = 1;
fe2d35ff 2140 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2141 error = walk_component(nd, path, &nd->last, LAST_NORM,
2142 !symlink_ok);
2143 if (error < 0)
fe2d35ff 2144 return ERR_PTR(error);
ce57dfc1 2145 if (error) /* symlink */
fe2d35ff 2146 return NULL;
fe2d35ff
AV		 2147 /* sayonara */
2148 if (nd->flags & LOOKUP_RCU) {
2149 if (nameidata_drop_rcu_last(nd))
2150 return ERR_PTR(-ECHILD);
2151 }
2152
2153 error = -ENOTDIR;
2154 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2155 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2156 goto exit;
2157 }
2158 audit_inode(pathname, nd->path.dentry);
2159 goto ok;
2160 }
2161
2162 /* create side of things */
2163
2164 if (nd->flags & LOOKUP_RCU) {
2165 if (nameidata_drop_rcu_last(nd))
2166 return ERR_PTR(-ECHILD);
2167 }
2168
2169 audit_inode(pathname, dir);
16c2cd71 2170 error = -EISDIR;
1f36f774 2171 /* trailing slashes? */
31e6b01f
NP		 2172 if (nd->last.name[nd->last.len])
2173 goto exit;
a2c36b45 2174
a1e28038
AV		 2175 mutex_lock(&dir->d_inode->i_mutex);
2176
6c0d46c4
AV		 2177 dentry = lookup_hash(nd);
2178 error = PTR_ERR(dentry);
2179 if (IS_ERR(dentry)) {
fb1cc555
AV		 2180 mutex_unlock(&dir->d_inode->i_mutex);
2181 goto exit;
2182 }
2183
6c0d46c4
AV		 2184 path->dentry = dentry;
2185 path->mnt = nd->path.mnt;
2186
fb1cc555 2187 /* Negative dentry, just create the file */
6c0d46c4
AV		 2188 if (!dentry->d_inode) {
2189 int mode = op->mode;
2190 if (!IS_POSIXACL(dir->d_inode))
2191 mode &= ~current_umask();
fb1cc555
AV		 2192 /*
2193 * This write is needed to ensure that a
6c0d46c4 2194 * rw->ro transition does not occur between
fb1cc555
AV		 2195 * the time when the file is created and when
2196 * a permanent write count is taken through
2197 * the 'struct file' in nameidata_to_filp().
2198 */
2199 error = mnt_want_write(nd->path.mnt);
2200 if (error)
2201 goto exit_mutex_unlock;
ca344a89 2202 want_write = 1;
9b44f1b3 2203 /* Don't check for write permission, don't truncate */
ca344a89 2204 open_flag &= ~O_TRUNC;
6c0d46c4 2205 will_truncate = 0;
bcda7652 2206 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2207 error = security_path_mknod(&nd->path, dentry, mode, 0);
2208 if (error)
2209 goto exit_mutex_unlock;
2210 error = vfs_create(dir->d_inode, dentry, mode, nd);
2211 if (error)
2212 goto exit_mutex_unlock;
2213 mutex_unlock(&dir->d_inode->i_mutex);
2214 dput(nd->path.dentry);
2215 nd->path.dentry = dentry;
ca344a89 2216 goto common;
fb1cc555
AV		 2217 }
2218
2219 /*
2220 * It already exists.
2221 */
2222 mutex_unlock(&dir->d_inode->i_mutex);
2223 audit_inode(pathname, path->dentry);
2224
2225 error = -EEXIST;
ca344a89 2226 if (open_flag & O_EXCL)
fb1cc555
AV		 2227 goto exit_dput;
2228
9875cf80
DH		 2229 error = follow_managed(path, nd->flags);
2230 if (error < 0)
2231 goto exit_dput;
fb1cc555
AV		 2232
2233 error = -ENOENT;
2234 if (!path->dentry->d_inode)
2235 goto exit_dput;
9e67f361
AV		 2236
2237 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2238 return NULL;
fb1cc555
AV		 2239
2240 path_to_nameidata(path, nd);
31e6b01f 2241 nd->inode = path->dentry->d_inode;
fb1cc555 2242 error = -EISDIR;
31e6b01f 2243 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2244 goto exit;
67ee3ad2 2245ok:
6c0d46c4
AV		 2246 if (!S_ISREG(nd->inode->i_mode))
2247 will_truncate = 0;
2248
0f9d1a10
AV		 2249 if (will_truncate) {
2250 error = mnt_want_write(nd->path.mnt);
2251 if (error)
2252 goto exit;
ca344a89 2253 want_write = 1;
0f9d1a10 2254 }
ca344a89 2255common:
bcda7652 2256 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2257 if (error)
0f9d1a10 2258 goto exit;
0f9d1a10
AV		 2259 filp = nameidata_to_filp(nd);
2260 if (!IS_ERR(filp)) {
2261 error = ima_file_check(filp, op->acc_mode);
2262 if (error) {
2263 fput(filp);
2264 filp = ERR_PTR(error);
2265 }
2266 }
2267 if (!IS_ERR(filp)) {
2268 if (will_truncate) {
2269 error = handle_truncate(filp);
2270 if (error) {
2271 fput(filp);
2272 filp = ERR_PTR(error);
2273 }
2274 }
2275 }
ca344a89
AV		 2276out:
2277 if (want_write)
0f9d1a10
AV		 2278 mnt_drop_write(nd->path.mnt);
2279 path_put(&nd->path);
fb1cc555
AV		 2280 return filp;
2281
2282exit_mutex_unlock:
2283 mutex_unlock(&dir->d_inode->i_mutex);
2284exit_dput:
2285 path_put_conditional(path, nd);
2286exit:
ca344a89
AV		 2287 filp = ERR_PTR(error);
2288 goto out;
fb1cc555
AV		 2289}
2290
13aab428 2291static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2292 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2293{
fe2d35ff 2294 struct file *base = NULL;
4a3fd211 2295 struct file *filp;
9850c056 2296 struct path path;
13aab428 2297 int error;
31e6b01f
NP		 2298
2299 filp = get_empty_filp();
2300 if (!filp)
2301 return ERR_PTR(-ENFILE);
2302
47c805dc 2303 filp->f_flags = op->open_flag;
73d049a4
AV		 2304 nd->intent.open.file = filp;
2305 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2306 nd->intent.open.create_mode = op->mode;
31e6b01f 2307
73d049a4 2308 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2309 if (unlikely(error))
13aab428 2310 goto out_filp;
31e6b01f 2311
fe2d35ff 2312 current->total_link_count = 0;
73d049a4 2313 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2314 if (unlikely(error))
2315 goto out_filp;
1da177e4 2316
73d049a4 2317 filp = do_last(nd, &path, op, pathname);
806b681c 2318 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2319 struct path link = path;
def4af30 2320 void *cookie;
574197e0 2321 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2322 path_put_conditional(&path, nd);
2323 path_put(&nd->path);
40b39136
AV		 2324 filp = ERR_PTR(-ELOOP);
2325 break;
2326 }
73d049a4
AV		 2327 nd->flags |= LOOKUP_PARENT;
2328 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2329 error = follow_link(&link, nd, &cookie);
c3e380b0 2330 if (unlikely(error))
f1afe9ef 2331 filp = ERR_PTR(error);
c3e380b0 2332 else
73d049a4 2333 filp = do_last(nd, &path, op, pathname);
574197e0 2334 put_link(nd, &link, cookie);
806b681c 2335 }
10fa8e62 2336out:
73d049a4
AV		 2337 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2338 path_put(&nd->root);
fe2d35ff
AV		 2339 if (base)
2340 fput(base);
73d049a4 2341 release_open_intent(nd);
10fa8e62 2342 return filp;
1da177e4 2343
31e6b01f 2344out_filp:
806b681c 2345 filp = ERR_PTR(error);
10fa8e62 2346 goto out;
1da177e4
LT		 2347}
2348
13aab428
AV		 2349struct file *do_filp_open(int dfd, const char *pathname,
2350 const struct open_flags *op, int flags)
2351{
73d049a4 2352 struct nameidata nd;
13aab428
AV		 2353 struct file *filp;
2354
73d049a4 2355 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2356 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2357 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2358 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2359 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2360 return filp;
2361}
2362
73d049a4
AV		 2363struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2364 const char *name, const struct open_flags *op, int flags)
2365{
2366 struct nameidata nd;
2367 struct file *file;
2368
2369 nd.root.mnt = mnt;
2370 nd.root.dentry = dentry;
2371
2372 flags |= LOOKUP_ROOT;
2373
bcda7652 2374 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2375 return ERR_PTR(-ELOOP);
2376
2377 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2378 if (unlikely(file == ERR_PTR(-ECHILD)))
2379 file = path_openat(-1, name, &nd, op, flags);
2380 if (unlikely(file == ERR_PTR(-ESTALE)))
2381 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2382 return file;
2383}
2384
1da177e4
LT		 2385/**
2386 * lookup_create - lookup a dentry, creating it if it doesn't exist
2387 * @nd: nameidata info
2388 * @is_dir: directory flag
2389 *
2390 * Simple function to lookup and return a dentry and create it
2391 * if it doesn't exist. Is SMP-safe.
c663e5d8 2392 *
4ac91378 2393 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2394 */
2395struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2396{
c663e5d8 2397 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2398
4ac91378 2399 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2400 /*
2401 * Yucky last component or no last component at all?
2402 * (foo/., foo/.., /////)
2403 */
1da177e4
LT		 2404 if (nd->last_type != LAST_NORM)
2405 goto fail;
2406 nd->flags &= ~LOOKUP_PARENT;
3516586a 2407 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2408 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2409
2410 /*
2411 * Do the final lookup.
2412 */
49705b77 2413 dentry = lookup_hash(nd);
1da177e4
LT		 2414 if (IS_ERR(dentry))
2415 goto fail;
c663e5d8 2416
e9baf6e5
AV		 2417 if (dentry->d_inode)
2418 goto eexist;
c663e5d8
CH		 2419 /*
2420 * Special case - lookup gave negative, but... we had foo/bar/
2421 * From the vfs_mknod() POV we just have a negative dentry -
2422 * all is fine. Let's be bastards - you had / on the end, you've
2423 * been asking for (non-existent) directory. -ENOENT for you.
2424 */
e9baf6e5
AV		 2425 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2426 dput(dentry);
2427 dentry = ERR_PTR(-ENOENT);
2428 }
1da177e4 2429 return dentry;
e9baf6e5 2430eexist:
1da177e4 2431 dput(dentry);
e9baf6e5 2432 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2433fail:
2434 return dentry;
2435}
f81a0bff 2436EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2437
2438int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2439{
a95164d9 2440 int error = may_create(dir, dentry);
1da177e4
LT		 2441
2442 if (error)
2443 return error;
2444
2445 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2446 return -EPERM;
2447
acfa4380 2448 if (!dir->i_op->mknod)
1da177e4
LT		 2449 return -EPERM;
2450
08ce5f16
SH		 2451 error = devcgroup_inode_mknod(mode, dev);
2452 if (error)
2453 return error;
2454
1da177e4
LT		 2455 error = security_inode_mknod(dir, dentry, mode, dev);
2456 if (error)
2457 return error;
2458
1da177e4 2459 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2460 if (!error)
f38aa942 2461 fsnotify_create(dir, dentry);
1da177e4
LT		 2462 return error;
2463}
2464
463c3197
DH		 2465static int may_mknod(mode_t mode)
2466{
2467 switch (mode & S_IFMT) {
2468 case S_IFREG:
2469 case S_IFCHR:
2470 case S_IFBLK:
2471 case S_IFIFO:
2472 case S_IFSOCK:
2473 case 0: /* zero mode translates to S_IFREG */
2474 return 0;
2475 case S_IFDIR:
2476 return -EPERM;
2477 default:
2478 return -EINVAL;
2479 }
2480}
2481
2e4d0924
HC		 2482SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2483 unsigned, dev)
1da177e4 2484{
2ad94ae6
AV		 2485 int error;
2486 char *tmp;
2487 struct dentry *dentry;
1da177e4
LT		 2488 struct nameidata nd;
2489
2490 if (S_ISDIR(mode))
2491 return -EPERM;
1da177e4 2492
2ad94ae6 2493 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2494 if (error)
2ad94ae6
AV		 2495 return error;
2496
1da177e4 2497 dentry = lookup_create(&nd, 0);
463c3197
DH		 2498 if (IS_ERR(dentry)) {
2499 error = PTR_ERR(dentry);
2500 goto out_unlock;
2501 }
4ac91378 2502 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2503 mode &= ~current_umask();
463c3197
DH		 2504 error = may_mknod(mode);
2505 if (error)
2506 goto out_dput;
2507 error = mnt_want_write(nd.path.mnt);
2508 if (error)
2509 goto out_dput;
be6d3e56
KT		 2510 error = security_path_mknod(&nd.path, dentry, mode, dev);
2511 if (error)
2512 goto out_drop_write;
463c3197 2513 switch (mode & S_IFMT) {
1da177e4 2514 case 0: case S_IFREG:
4ac91378 2515 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2516 break;
2517 case S_IFCHR: case S_IFBLK:
4ac91378 2518 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2519 new_decode_dev(dev));
2520 break;
2521 case S_IFIFO: case S_IFSOCK:
4ac91378 2522 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2523 break;
1da177e4 2524 }
be6d3e56 2525out_drop_write:
463c3197
DH		 2526 mnt_drop_write(nd.path.mnt);
2527out_dput:
2528 dput(dentry);
2529out_unlock:
4ac91378 2530 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2531 path_put(&nd.path);
1da177e4
LT		 2532 putname(tmp);
2533
2534 return error;
2535}
2536
3480b257 2537SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2538{
2539 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2540}
2541
1da177e4
LT		 2542int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2543{
a95164d9 2544 int error = may_create(dir, dentry);
1da177e4
LT		 2545
2546 if (error)
2547 return error;
2548
acfa4380 2549 if (!dir->i_op->mkdir)
1da177e4
LT		 2550 return -EPERM;
2551
2552 mode &= (S_IRWXUGO|S_ISVTX);
2553 error = security_inode_mkdir(dir, dentry, mode);
2554 if (error)
2555 return error;
2556
1da177e4 2557 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2558 if (!error)
f38aa942 2559 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2560 return error;
2561}
2562
2e4d0924 2563SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2564{
2565 int error = 0;
2566 char * tmp;
6902d925
DH		 2567 struct dentry *dentry;
2568 struct nameidata nd;
1da177e4 2569
2ad94ae6
AV		 2570 error = user_path_parent(dfd, pathname, &nd, &tmp);
2571 if (error)
6902d925 2572 goto out_err;
1da177e4 2573
6902d925
DH		 2574 dentry = lookup_create(&nd, 1);
2575 error = PTR_ERR(dentry);
2576 if (IS_ERR(dentry))
2577 goto out_unlock;
1da177e4 2578
4ac91378 2579 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2580 mode &= ~current_umask();
463c3197
DH		 2581 error = mnt_want_write(nd.path.mnt);
2582 if (error)
2583 goto out_dput;
be6d3e56
KT		 2584 error = security_path_mkdir(&nd.path, dentry, mode);
2585 if (error)
2586 goto out_drop_write;
4ac91378 2587 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2588out_drop_write:
463c3197
DH		 2589 mnt_drop_write(nd.path.mnt);
2590out_dput:
6902d925
DH		 2591 dput(dentry);
2592out_unlock:
4ac91378 2593 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2594 path_put(&nd.path);
6902d925
DH		 2595 putname(tmp);
2596out_err:
1da177e4
LT		 2597 return error;
2598}
2599
3cdad428 2600SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2601{
2602 return sys_mkdirat(AT_FDCWD, pathname, mode);
2603}
2604
1da177e4
LT		 2605/*
2606 * We try to drop the dentry early: we should have
2607 * a usage count of 2 if we're the only user of this
2608 * dentry, and if that is true (possibly after pruning
2609 * the dcache), then we drop the dentry now.
2610 *
2611 * A low-level filesystem can, if it choses, legally
2612 * do a
2613 *
2614 * if (!d_unhashed(dentry))
2615 * return -EBUSY;
2616 *
2617 * if it cannot handle the case of removing a directory
2618 * that is still in use by something else..
2619 */
2620void dentry_unhash(struct dentry *dentry)
2621{
2622 dget(dentry);
dc168427 2623 shrink_dcache_parent(dentry);
1da177e4 2624 spin_lock(&dentry->d_lock);
b7ab39f6 2625 if (dentry->d_count == 2)
1da177e4
LT		 2626 __d_drop(dentry);
2627 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2628}
2629
2630int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2631{
2632 int error = may_delete(dir, dentry, 1);
2633
2634 if (error)
2635 return error;
2636
acfa4380 2637 if (!dir->i_op->rmdir)
1da177e4
LT		 2638 return -EPERM;
2639
1b1dcc1b 2640 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2641 dentry_unhash(dentry);
2642 if (d_mountpoint(dentry))
2643 error = -EBUSY;
2644 else {
2645 error = security_inode_rmdir(dir, dentry);
2646 if (!error) {
2647 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2648 if (!error) {
1da177e4 2649 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2650 dont_mount(dentry);
2651 }
1da177e4
LT		 2652 }
2653 }
1b1dcc1b 2654 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2655 if (!error) {
1da177e4
LT		 2656 d_delete(dentry);
2657 }
2658 dput(dentry);
2659
2660 return error;
2661}
2662
5590ff0d 2663static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2664{
2665 int error = 0;
2666 char * name;
2667 struct dentry *dentry;
2668 struct nameidata nd;
2669
2ad94ae6 2670 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2671 if (error)
2ad94ae6 2672 return error;
1da177e4
LT		 2673
2674 switch(nd.last_type) {
0612d9fb
OH		 2675 case LAST_DOTDOT:
2676 error = -ENOTEMPTY;
2677 goto exit1;
2678 case LAST_DOT:
2679 error = -EINVAL;
2680 goto exit1;
2681 case LAST_ROOT:
2682 error = -EBUSY;
2683 goto exit1;
1da177e4 2684 }
0612d9fb
OH		 2685
2686 nd.flags &= ~LOOKUP_PARENT;
2687
4ac91378 2688 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2689 dentry = lookup_hash(&nd);
1da177e4 2690 error = PTR_ERR(dentry);
6902d925
DH		 2691 if (IS_ERR(dentry))
2692 goto exit2;
0622753b
DH		 2693 error = mnt_want_write(nd.path.mnt);
2694 if (error)
2695 goto exit3;
be6d3e56
KT		 2696 error = security_path_rmdir(&nd.path, dentry);
2697 if (error)
2698 goto exit4;
4ac91378 2699 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2700exit4:
0622753b
DH		 2701 mnt_drop_write(nd.path.mnt);
2702exit3:
6902d925
DH		 2703 dput(dentry);
2704exit2:
4ac91378 2705 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2706exit1:
1d957f9b 2707 path_put(&nd.path);
1da177e4
LT		 2708 putname(name);
2709 return error;
2710}
2711
3cdad428 2712SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2713{
2714 return do_rmdir(AT_FDCWD, pathname);
2715}
2716
1da177e4
LT		 2717int vfs_unlink(struct inode *dir, struct dentry *dentry)
2718{
2719 int error = may_delete(dir, dentry, 0);
2720
2721 if (error)
2722 return error;
2723
acfa4380 2724 if (!dir->i_op->unlink)
1da177e4
LT		 2725 return -EPERM;
2726
1b1dcc1b 2727 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2728 if (d_mountpoint(dentry))
2729 error = -EBUSY;
2730 else {
2731 error = security_inode_unlink(dir, dentry);
bec1052e 2732 if (!error) {
1da177e4 2733 error = dir->i_op->unlink(dir, dentry);
bec1052e 2734 if (!error)
d83c49f3 2735 dont_mount(dentry);
bec1052e 2736 }
1da177e4 2737 }
1b1dcc1b 2738 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2739
2740 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2741 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2742 fsnotify_link_count(dentry->d_inode);
e234f35c 2743 d_delete(dentry);
1da177e4 2744 }
0eeca283 2745
1da177e4
LT		 2746 return error;
2747}
2748
2749/*
2750 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2751 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2752 * writeout happening, and we don't want to prevent access to the directory
2753 * while waiting on the I/O.
2754 */
5590ff0d 2755static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2756{
2ad94ae6
AV		 2757 int error;
2758 char *name;
1da177e4
LT		 2759 struct dentry *dentry;
2760 struct nameidata nd;
2761 struct inode *inode = NULL;
2762
2ad94ae6 2763 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2764 if (error)
2ad94ae6
AV		 2765 return error;
2766
1da177e4
LT		 2767 error = -EISDIR;
2768 if (nd.last_type != LAST_NORM)
2769 goto exit1;
0612d9fb
OH		 2770
2771 nd.flags &= ~LOOKUP_PARENT;
2772
4ac91378 2773 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2774 dentry = lookup_hash(&nd);
1da177e4
LT		 2775 error = PTR_ERR(dentry);
2776 if (!IS_ERR(dentry)) {
2777 /* Why not before? Because we want correct error value */
2778 if (nd.last.name[nd.last.len])
2779 goto slashes;
2780 inode = dentry->d_inode;
2781 if (inode)
7de9c6ee 2782 ihold(inode);
0622753b
DH		 2783 error = mnt_want_write(nd.path.mnt);
2784 if (error)
2785 goto exit2;
be6d3e56
KT		 2786 error = security_path_unlink(&nd.path, dentry);
2787 if (error)
2788 goto exit3;
4ac91378 2789 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2790exit3:
0622753b 2791 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2792 exit2:
2793 dput(dentry);
2794 }
4ac91378 2795 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2796 if (inode)
2797 iput(inode); /* truncate the inode here */
2798exit1:
1d957f9b 2799 path_put(&nd.path);
1da177e4
LT		 2800 putname(name);
2801 return error;
2802
2803slashes:
2804 error = !dentry->d_inode ? -ENOENT :
2805 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2806 goto exit2;
2807}
2808
2e4d0924 2809SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2810{
2811 if ((flag & ~AT_REMOVEDIR) != 0)
2812 return -EINVAL;
2813
2814 if (flag & AT_REMOVEDIR)
2815 return do_rmdir(dfd, pathname);
2816
2817 return do_unlinkat(dfd, pathname);
2818}
2819
3480b257 2820SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2821{
2822 return do_unlinkat(AT_FDCWD, pathname);
2823}
2824
db2e747b 2825int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2826{
a95164d9 2827 int error = may_create(dir, dentry);
1da177e4
LT		 2828
2829 if (error)
2830 return error;
2831
acfa4380 2832 if (!dir->i_op->symlink)
1da177e4
LT		 2833 return -EPERM;
2834
2835 error = security_inode_symlink(dir, dentry, oldname);
2836 if (error)
2837 return error;
2838
1da177e4 2839 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2840 if (!error)
f38aa942 2841 fsnotify_create(dir, dentry);
1da177e4
LT		 2842 return error;
2843}
2844
2e4d0924
HC		 2845SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2846 int, newdfd, const char __user *, newname)
1da177e4 2847{
2ad94ae6
AV		 2848 int error;
2849 char *from;
2850 char *to;
6902d925
DH		 2851 struct dentry *dentry;
2852 struct nameidata nd;
1da177e4
LT		 2853
2854 from = getname(oldname);
2ad94ae6 2855 if (IS_ERR(from))
1da177e4 2856 return PTR_ERR(from);
1da177e4 2857
2ad94ae6 2858 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2859 if (error)
2ad94ae6
AV		 2860 goto out_putname;
2861
6902d925
DH		 2862 dentry = lookup_create(&nd, 0);
2863 error = PTR_ERR(dentry);
2864 if (IS_ERR(dentry))
2865 goto out_unlock;
2866
75c3f29d
DH		 2867 error = mnt_want_write(nd.path.mnt);
2868 if (error)
2869 goto out_dput;
be6d3e56
KT		 2870 error = security_path_symlink(&nd.path, dentry, from);
2871 if (error)
2872 goto out_drop_write;
db2e747b 2873 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2874out_drop_write:
75c3f29d
DH		 2875 mnt_drop_write(nd.path.mnt);
2876out_dput:
6902d925
DH		 2877 dput(dentry);
2878out_unlock:
4ac91378 2879 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2880 path_put(&nd.path);
6902d925
DH		 2881 putname(to);
2882out_putname:
1da177e4
LT		 2883 putname(from);
2884 return error;
2885}
2886
3480b257 2887SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2888{
2889 return sys_symlinkat(oldname, AT_FDCWD, newname);
2890}
2891
1da177e4
LT		 2892int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2893{
2894 struct inode *inode = old_dentry->d_inode;
2895 int error;
2896
2897 if (!inode)
2898 return -ENOENT;
2899
a95164d9 2900 error = may_create(dir, new_dentry);
1da177e4
LT		 2901 if (error)
2902 return error;
2903
2904 if (dir->i_sb != inode->i_sb)
2905 return -EXDEV;
2906
2907 /*
2908 * A link to an append-only or immutable file cannot be created.
2909 */
2910 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2911 return -EPERM;
acfa4380 2912 if (!dir->i_op->link)
1da177e4 2913 return -EPERM;
7e79eedb 2914 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2915 return -EPERM;
2916
2917 error = security_inode_link(old_dentry, dir, new_dentry);
2918 if (error)
2919 return error;
2920
7e79eedb 2921 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 2922 /* Make sure we don't allow creating hardlink to an unlinked file */
2923 if (inode->i_nlink == 0)
2924 error = -ENOENT;
2925 else
2926 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2927 mutex_unlock(&inode->i_mutex);
e31e14ec 2928 if (!error)
7e79eedb 2929 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2930 return error;
2931}
2932
2933/*
2934 * Hardlinks are often used in delicate situations. We avoid
2935 * security-related surprises by not following symlinks on the
2936 * newname. --KAB
2937 *
2938 * We don't follow them on the oldname either to be compatible
2939 * with linux 2.0, and to avoid hard-linking to directories
2940 * and other special files. --ADM
2941 */
2e4d0924
HC		 2942SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2943 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2944{
2945 struct dentry *new_dentry;
2d8f3038
AV		 2946 struct nameidata nd;
2947 struct path old_path;
11a7b371 2948 int how = 0;
1da177e4 2949 int error;
2ad94ae6 2950 char *to;
1da177e4 2951
11a7b371 2952 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 2953 return -EINVAL;
11a7b371
AK		 2954 /*
2955 * To use null names we require CAP_DAC_READ_SEARCH
2956 * This ensures that not everyone will be able to create
2957 * handlink using the passed filedescriptor.
2958 */
2959 if (flags & AT_EMPTY_PATH) {
2960 if (!capable(CAP_DAC_READ_SEARCH))
2961 return -ENOENT;
2962 how = LOOKUP_EMPTY;
2963 }
2964
2965 if (flags & AT_SYMLINK_FOLLOW)
2966 how |= LOOKUP_FOLLOW;
c04030e1 2967
11a7b371 2968 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 2969 if (error)
2ad94ae6
AV		 2970 return error;
2971
2972 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2973 if (error)
2974 goto out;
2975 error = -EXDEV;
2d8f3038 2976 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2977 goto out_release;
2978 new_dentry = lookup_create(&nd, 0);
2979 error = PTR_ERR(new_dentry);
6902d925
DH		 2980 if (IS_ERR(new_dentry))
2981 goto out_unlock;
75c3f29d
DH		 2982 error = mnt_want_write(nd.path.mnt);
2983 if (error)
2984 goto out_dput;
be6d3e56
KT		 2985 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2986 if (error)
2987 goto out_drop_write;
2d8f3038 2988 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2989out_drop_write:
75c3f29d
DH		 2990 mnt_drop_write(nd.path.mnt);
2991out_dput:
6902d925
DH		 2992 dput(new_dentry);
2993out_unlock:
4ac91378 2994 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2995out_release:
1d957f9b 2996 path_put(&nd.path);
2ad94ae6 2997 putname(to);
1da177e4 2998out:
2d8f3038 2999 path_put(&old_path);
1da177e4
LT		 3000
3001 return error;
3002}
3003
3480b257 3004SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3005{
c04030e1 3006 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3007}
3008
1da177e4
LT		 3009/*
3010 * The worst of all namespace operations - renaming directory. "Perverted"
3011 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3012 * Problems:
3013 * a) we can get into loop creation. Check is done in is_subdir().
3014 * b) race potential - two innocent renames can create a loop together.
3015 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3016 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3017 * story.
3018 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3019 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3020 * whether the target exists). Solution: try to be smart with locking
3021 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3022 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3023 * move will be locked. Thus we can rank directories by the tree
3024 * (ancestors first) and rank all non-directories after them.
3025 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3026 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3027 * HOWEVER, it relies on the assumption that any object with ->lookup()
3028 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3029 * we'd better make sure that there's no link(2) for them.
3030 * d) some filesystems don't support opened-but-unlinked directories,
3031 * either because of layout or because they are not ready to deal with
3032 * all cases correctly. The latter will be fixed (taking this sort of
3033 * stuff into VFS), but the former is not going away. Solution: the same
3034 * trick as in rmdir().
3035 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3036 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3037 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3038 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3039 * locking].
3040 */
75c96f85
AB		 3041static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3042 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3043{
3044 int error = 0;
3045 struct inode *target;
3046
3047 /*
3048 * If we are going to change the parent - check write permissions,
3049 * we'll need to flip '..'.
3050 */
3051 if (new_dir != old_dir) {
f419a2e3 3052 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3053 if (error)
3054 return error;
3055 }
3056
3057 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3058 if (error)
3059 return error;
3060
3061 target = new_dentry->d_inode;
d83c49f3 3062 if (target)
1b1dcc1b 3063 mutex_lock(&target->i_mutex);
1da177e4
LT		 3064 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3065 error = -EBUSY;
d83c49f3
AV		 3066 else {
3067 if (target)
3068 dentry_unhash(new_dentry);
1da177e4 3069 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3070 }
1da177e4 3071 if (target) {
d83c49f3 3072 if (!error) {
1da177e4 3073 target->i_flags |= S_DEAD;
d83c49f3
AV		 3074 dont_mount(new_dentry);
3075 }
1b1dcc1b 3076 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3077 if (d_unhashed(new_dentry))
3078 d_rehash(new_dentry);
3079 dput(new_dentry);
3080 }
e31e14ec 3081 if (!error)
349457cc
MF		 3082 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3083 d_move(old_dentry,new_dentry);
1da177e4
LT		 3084 return error;
3085}
3086
75c96f85
AB		 3087static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3088 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3089{
3090 struct inode *target;
3091 int error;
3092
3093 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3094 if (error)
3095 return error;
3096
3097 dget(new_dentry);
3098 target = new_dentry->d_inode;
3099 if (target)
1b1dcc1b 3100 mutex_lock(&target->i_mutex);
1da177e4
LT		 3101 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3102 error = -EBUSY;
3103 else
3104 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3105 if (!error) {
bec1052e 3106 if (target)
d83c49f3 3107 dont_mount(new_dentry);
349457cc 3108 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3109 d_move(old_dentry, new_dentry);
1da177e4
LT		 3110 }
3111 if (target)
1b1dcc1b 3112 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3113 dput(new_dentry);
3114 return error;
3115}
3116
3117int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3118 struct inode *new_dir, struct dentry *new_dentry)
3119{
3120 int error;
3121 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3122 const unsigned char *old_name;
1da177e4
LT		 3123
3124 if (old_dentry->d_inode == new_dentry->d_inode)
3125 return 0;
3126
3127 error = may_delete(old_dir, old_dentry, is_dir);
3128 if (error)
3129 return error;
3130
3131 if (!new_dentry->d_inode)
a95164d9 3132 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3133 else
3134 error = may_delete(new_dir, new_dentry, is_dir);
3135 if (error)
3136 return error;
3137
acfa4380 3138 if (!old_dir->i_op->rename)
1da177e4
LT		 3139 return -EPERM;
3140
0eeca283
RL		 3141 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3142
1da177e4
LT		 3143 if (is_dir)
3144 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3145 else
3146 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3147 if (!error)
3148 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3149 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3150 fsnotify_oldname_free(old_name);
3151
1da177e4
LT		 3152 return error;
3153}
3154
2e4d0924
HC		 3155SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3156 int, newdfd, const char __user *, newname)
1da177e4 3157{
2ad94ae6
AV		 3158 struct dentry *old_dir, *new_dir;
3159 struct dentry *old_dentry, *new_dentry;
3160 struct dentry *trap;
1da177e4 3161 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3162 char *from;
3163 char *to;
3164 int error;
1da177e4 3165
2ad94ae6 3166 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3167 if (error)
3168 goto exit;
3169
2ad94ae6 3170 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3171 if (error)
3172 goto exit1;
3173
3174 error = -EXDEV;
4ac91378 3175 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3176 goto exit2;
3177
4ac91378 3178 old_dir = oldnd.path.dentry;
1da177e4
LT		 3179 error = -EBUSY;
3180 if (oldnd.last_type != LAST_NORM)
3181 goto exit2;
3182
4ac91378 3183 new_dir = newnd.path.dentry;
1da177e4
LT		 3184 if (newnd.last_type != LAST_NORM)
3185 goto exit2;
3186
0612d9fb
OH		 3187 oldnd.flags &= ~LOOKUP_PARENT;
3188 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3189 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3190
1da177e4
LT		 3191 trap = lock_rename(new_dir, old_dir);
3192
49705b77 3193 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3194 error = PTR_ERR(old_dentry);
3195 if (IS_ERR(old_dentry))
3196 goto exit3;
3197 /* source must exist */
3198 error = -ENOENT;
3199 if (!old_dentry->d_inode)
3200 goto exit4;
3201 /* unless the source is a directory trailing slashes give -ENOTDIR */
3202 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3203 error = -ENOTDIR;
3204 if (oldnd.last.name[oldnd.last.len])
3205 goto exit4;
3206 if (newnd.last.name[newnd.last.len])
3207 goto exit4;
3208 }
3209 /* source should not be ancestor of target */
3210 error = -EINVAL;
3211 if (old_dentry == trap)
3212 goto exit4;
49705b77 3213 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3214 error = PTR_ERR(new_dentry);
3215 if (IS_ERR(new_dentry))
3216 goto exit4;
3217 /* target should not be an ancestor of source */
3218 error = -ENOTEMPTY;
3219 if (new_dentry == trap)
3220 goto exit5;
3221
9079b1eb
DH		 3222 error = mnt_want_write(oldnd.path.mnt);
3223 if (error)
3224 goto exit5;
be6d3e56
KT		 3225 error = security_path_rename(&oldnd.path, old_dentry,
3226 &newnd.path, new_dentry);
3227 if (error)
3228 goto exit6;
1da177e4
LT		 3229 error = vfs_rename(old_dir->d_inode, old_dentry,
3230 new_dir->d_inode, new_dentry);
be6d3e56 3231exit6:
9079b1eb 3232 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3233exit5:
3234 dput(new_dentry);
3235exit4:
3236 dput(old_dentry);
3237exit3:
3238 unlock_rename(new_dir, old_dir);
3239exit2:
1d957f9b 3240 path_put(&newnd.path);
2ad94ae6 3241 putname(to);
1da177e4 3242exit1:
1d957f9b 3243 path_put(&oldnd.path);
1da177e4 3244 putname(from);
2ad94ae6 3245exit:
1da177e4
LT		 3246 return error;
3247}
3248
a26eab24 3249SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3250{
3251 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3252}
3253
1da177e4
LT		 3254int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3255{
3256 int len;
3257
3258 len = PTR_ERR(link);
3259 if (IS_ERR(link))
3260 goto out;
3261
3262 len = strlen(link);
3263 if (len > (unsigned) buflen)
3264 len = buflen;
3265 if (copy_to_user(buffer, link, len))
3266 len = -EFAULT;
3267out:
3268 return len;
3269}
3270
3271/*
3272 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3273 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3274 * using) it for any given inode is up to filesystem.
3275 */
3276int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3277{
3278 struct nameidata nd;
cc314eef 3279 void *cookie;
694a1764 3280 int res;
cc314eef 3281
1da177e4 3282 nd.depth = 0;
cc314eef 3283 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3284 if (IS_ERR(cookie))
3285 return PTR_ERR(cookie);
3286
3287 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3288 if (dentry->d_inode->i_op->put_link)
3289 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3290 return res;
1da177e4
LT		 3291}
3292
3293int vfs_follow_link(struct nameidata *nd, const char *link)
3294{
3295 return __vfs_follow_link(nd, link);
3296}
3297
3298/* get the link contents into pagecache */
3299static char *page_getlink(struct dentry * dentry, struct page **ppage)
3300{
ebd09abb
DG		 3301 char *kaddr;
3302 struct page *page;
1da177e4 3303 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3304 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3305 if (IS_ERR(page))
6fe6900e 3306 return (char*)page;
1da177e4 3307 *ppage = page;
ebd09abb
DG		 3308 kaddr = kmap(page);
3309 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3310 return kaddr;
1da177e4
LT		 3311}
3312
3313int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3314{
3315 struct page *page = NULL;
3316 char *s = page_getlink(dentry, &page);
3317 int res = vfs_readlink(dentry,buffer,buflen,s);
3318 if (page) {
3319 kunmap(page);
3320 page_cache_release(page);
3321 }
3322 return res;
3323}
3324
cc314eef 3325void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3326{
cc314eef 3327 struct page *page = NULL;
1da177e4 3328 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3329 return page;
1da177e4
LT		 3330}
3331
cc314eef 3332void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3333{
cc314eef
LT		 3334 struct page *page = cookie;
3335
3336 if (page) {
1da177e4
LT		 3337 kunmap(page);
3338 page_cache_release(page);
1da177e4
LT		 3339 }
3340}
3341
54566b2c
NP		 3342/*
3343 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3344 */
3345int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3346{
3347 struct address_space *mapping = inode->i_mapping;
0adb25d2 3348 struct page *page;
afddba49 3349 void *fsdata;
beb497ab 3350 int err;
1da177e4 3351 char *kaddr;
54566b2c
NP		 3352 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3353 if (nofs)
3354 flags |= AOP_FLAG_NOFS;
1da177e4 3355
7e53cac4 3356retry:
afddba49 3357 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3358 flags, &page, &fsdata);
1da177e4 3359 if (err)
afddba49
NP		 3360 goto fail;
3361
1da177e4
LT		 3362 kaddr = kmap_atomic(page, KM_USER0);
3363 memcpy(kaddr, symname, len-1);
3364 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3365
3366 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3367 page, fsdata);
1da177e4
LT		 3368 if (err < 0)
3369 goto fail;
afddba49
NP		 3370 if (err < len-1)
3371 goto retry;
3372
1da177e4
LT		 3373 mark_inode_dirty(inode);
3374 return 0;
1da177e4
LT		 3375fail:
3376 return err;
3377}
3378
0adb25d2
KK		 3379int page_symlink(struct inode *inode, const char *symname, int len)
3380{
3381 return __page_symlink(inode, symname, len,
54566b2c 3382 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3383}
3384
92e1d5be 3385const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3386 .readlink = generic_readlink,
3387 .follow_link = page_follow_link_light,
3388 .put_link = page_put_link,
3389};
3390
2d8f3038 3391EXPORT_SYMBOL(user_path_at);
cc53ce53 3392EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3393EXPORT_SYMBOL(follow_down);
3394EXPORT_SYMBOL(follow_up);
3395EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3396EXPORT_SYMBOL(getname);
3397EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3398EXPORT_SYMBOL(lookup_one_len);
3399EXPORT_SYMBOL(page_follow_link_light);
3400EXPORT_SYMBOL(page_put_link);
3401EXPORT_SYMBOL(page_readlink);
0adb25d2 3402EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3403EXPORT_SYMBOL(page_symlink);
3404EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3405EXPORT_SYMBOL(kern_path_parent);
d1811465 3406EXPORT_SYMBOL(kern_path);
16f18200 3407EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3408EXPORT_SYMBOL(inode_permission);
8c744fb8 3409EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3410EXPORT_SYMBOL(unlock_rename);
3411EXPORT_SYMBOL(vfs_create);
3412EXPORT_SYMBOL(vfs_follow_link);
3413EXPORT_SYMBOL(vfs_link);
3414EXPORT_SYMBOL(vfs_mkdir);
3415EXPORT_SYMBOL(vfs_mknod);
3416EXPORT_SYMBOL(generic_permission);
3417EXPORT_SYMBOL(vfs_readlink);
3418EXPORT_SYMBOL(vfs_rename);
3419EXPORT_SYMBOL(vfs_rmdir);
3420EXPORT_SYMBOL(vfs_symlink);
3421EXPORT_SYMBOL(vfs_unlink);
3422EXPORT_SYMBOL(dentry_unhash);
3423EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)