projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'kbuild' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild-2.6
[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
0e794589
AV		 756 if (link->mnt == nd->path.mnt)
757 mntget(link->mnt);
758
574197e0
AV		 759 if (unlikely(current->total_link_count >= 40)) {
760 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
574197e0
AV		 761 path_put(&nd->path);
762 return -ELOOP;
763 }
764 cond_resched();
765 current->total_link_count++;
766
7b9337aa 767 touch_atime(link->mnt, dentry);
1da177e4 768 nd_set_link(nd, NULL);
cd4e91d3 769
36f3b4f6
AV		 770 error = security_inode_follow_link(link->dentry, nd);
771 if (error) {
772 *p = ERR_PTR(error); /* no ->put_link(), please */
773 path_put(&nd->path);
774 return error;
775 }
776
86acdca1 777 nd->last_type = LAST_BIND;
def4af30
AV		 778 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
779 error = PTR_ERR(*p);
780 if (!IS_ERR(*p)) {
1da177e4 781 char *s = nd_get_link(nd);
cc314eef 782 error = 0;
1da177e4
LT		 783 if (s)
784 error = __vfs_follow_link(nd, s);
bcda7652 785 else if (nd->last_type == LAST_BIND) {
16c2cd71 786 nd->flags |= LOOKUP_JUMPED;
b21041d0
AV		 787 nd->inode = nd->path.dentry->d_inode;
788 if (nd->inode->i_op->follow_link) {
bcda7652
AV		 789 /* stepped on a _really_ weird one */
790 path_put(&nd->path);
791 error = -ELOOP;
792 }
793 }
1da177e4 794 }
1da177e4
LT		 795 return error;
796}
797
31e6b01f
NP		 798static int follow_up_rcu(struct path *path)
799{
800 struct vfsmount *parent;
801 struct dentry *mountpoint;
802
803 parent = path->mnt->mnt_parent;
804 if (parent == path->mnt)
805 return 0;
806 mountpoint = path->mnt->mnt_mountpoint;
807 path->dentry = mountpoint;
808 path->mnt = parent;
809 return 1;
810}
811
bab77ebf 812int follow_up(struct path *path)
1da177e4
LT		 813{
814 struct vfsmount *parent;
815 struct dentry *mountpoint;
99b7db7b
NP		 816
817 br_read_lock(vfsmount_lock);
bab77ebf
AV		 818 parent = path->mnt->mnt_parent;
819 if (parent == path->mnt) {
99b7db7b 820 br_read_unlock(vfsmount_lock);
1da177e4
LT		 821 return 0;
822 }
823 mntget(parent);
bab77ebf 824 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 825 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 826 dput(path->dentry);
827 path->dentry = mountpoint;
828 mntput(path->mnt);
829 path->mnt = parent;
1da177e4
LT		 830 return 1;
831}
832
b5c84bf6 833/*
9875cf80
DH		 834 * Perform an automount
835 * - return -EISDIR to tell follow_managed() to stop and return the path we
836 * were called with.
1da177e4 837 */
9875cf80
DH		 838static int follow_automount(struct path *path, unsigned flags,
839 bool *need_mntput)
31e6b01f 840{
9875cf80 841 struct vfsmount *mnt;
ea5b778a 842 int err;
9875cf80
DH		 843
844 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
845 return -EREMOTE;
846
6f45b656
DH		 847 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
848 * and this is the terminal part of the path.
849 */
850 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
851 return -EISDIR; /* we actually want to stop here */
852
9875cf80
DH		 853 /* We want to mount if someone is trying to open/create a file of any
854 * type under the mountpoint, wants to traverse through the mountpoint
855 * or wants to open the mounted directory.
856 *
857 * We don't want to mount if someone's just doing a stat and they've
858 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
859 * appended a '/' to the name.
860 */
861 if (!(flags & LOOKUP_FOLLOW) &&
862 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
863 LOOKUP_OPEN | LOOKUP_CREATE)))
864 return -EISDIR;
865
866 current->total_link_count++;
867 if (current->total_link_count >= 40)
868 return -ELOOP;
869
870 mnt = path->dentry->d_op->d_automount(path);
871 if (IS_ERR(mnt)) {
872 /*
873 * The filesystem is allowed to return -EISDIR here to indicate
874 * it doesn't want to automount. For instance, autofs would do
875 * this so that its userspace daemon can mount on this dentry.
876 *
877 * However, we can only permit this if it's a terminal point in
878 * the path being looked up; if it wasn't then the remainder of
879 * the path is inaccessible and we should say so.
880 */
881 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
882 return -EREMOTE;
883 return PTR_ERR(mnt);
31e6b01f 884 }
ea5b778a 885
9875cf80
DH		 886 if (!mnt) /* mount collision */
887 return 0;
31e6b01f 888
19a167af 889 err = finish_automount(mnt, path);
9875cf80 890
ea5b778a
DH		 891 switch (err) {
892 case -EBUSY:
893 /* Someone else made a mount here whilst we were busy */
19a167af 894 return 0;
ea5b778a 895 case 0:
ea5b778a
DH		 896 dput(path->dentry);
897 if (*need_mntput)
898 mntput(path->mnt);
899 path->mnt = mnt;
900 path->dentry = dget(mnt->mnt_root);
901 *need_mntput = true;
902 return 0;
19a167af
AV		 903 default:
904 return err;
ea5b778a 905 }
19a167af 906
463ffb2e
AV		 907}
908
9875cf80
DH		 909/*
910 * Handle a dentry that is managed in some way.
cc53ce53 911 * - Flagged for transit management (autofs)
9875cf80
DH		 912 * - Flagged as mountpoint
913 * - Flagged as automount point
914 *
915 * This may only be called in refwalk mode.
916 *
917 * Serialization is taken care of in namespace.c
918 */
919static int follow_managed(struct path *path, unsigned flags)
1da177e4 920{
9875cf80
DH		 921 unsigned managed;
922 bool need_mntput = false;
923 int ret;
924
925 /* Given that we're not holding a lock here, we retain the value in a
926 * local variable for each dentry as we look at it so that we don't see
927 * the components of that value change under us */
928 while (managed = ACCESS_ONCE(path->dentry->d_flags),
929 managed &= DCACHE_MANAGED_DENTRY,
930 unlikely(managed != 0)) {
cc53ce53
DH		 931 /* Allow the filesystem to manage the transit without i_mutex
932 * being held. */
933 if (managed & DCACHE_MANAGE_TRANSIT) {
934 BUG_ON(!path->dentry->d_op);
935 BUG_ON(!path->dentry->d_op->d_manage);
1aed3e42 936 ret = path->dentry->d_op->d_manage(path->dentry, false);
cc53ce53
DH		 937 if (ret < 0)
938 return ret == -EISDIR ? 0 : ret;
939 }
940
9875cf80
DH		 941 /* Transit to a mounted filesystem. */
942 if (managed & DCACHE_MOUNTED) {
943 struct vfsmount *mounted = lookup_mnt(path);
944 if (mounted) {
945 dput(path->dentry);
946 if (need_mntput)
947 mntput(path->mnt);
948 path->mnt = mounted;
949 path->dentry = dget(mounted->mnt_root);
950 need_mntput = true;
951 continue;
952 }
953
954 /* Something is mounted on this dentry in another
955 * namespace and/or whatever was mounted there in this
956 * namespace got unmounted before we managed to get the
957 * vfsmount_lock */
958 }
959
960 /* Handle an automount point */
961 if (managed & DCACHE_NEED_AUTOMOUNT) {
962 ret = follow_automount(path, flags, &need_mntput);
963 if (ret < 0)
964 return ret == -EISDIR ? 0 : ret;
965 continue;
966 }
967
968 /* We didn't change the current path point */
969 break;
1da177e4 970 }
9875cf80 971 return 0;
1da177e4
LT		 972}
973
cc53ce53 974int follow_down_one(struct path *path)
1da177e4
LT		 975{
976 struct vfsmount *mounted;
977
1c755af4 978 mounted = lookup_mnt(path);
1da177e4 979 if (mounted) {
9393bd07
AV		 980 dput(path->dentry);
981 mntput(path->mnt);
982 path->mnt = mounted;
983 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 984 return 1;
985 }
986 return 0;
987}
988
9875cf80
DH		 989/*
990 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
cc53ce53 991 * meet a managed dentry and we're not walking to "..". True is returned to
9875cf80
DH		 992 * continue, false to abort.
993 */
994static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
995 struct inode **inode, bool reverse_transit)
996{
997 while (d_mountpoint(path->dentry)) {
998 struct vfsmount *mounted;
ab90911f
DH		 999 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1000 !reverse_transit &&
1aed3e42 1001 path->dentry->d_op->d_manage(path->dentry, true) < 0)
ab90911f 1002 return false;
9875cf80
DH		 1003 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1004 if (!mounted)
1005 break;
1006 path->mnt = mounted;
1007 path->dentry = mounted->mnt_root;
1008 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1009 *inode = path->dentry->d_inode;
1010 }
1011
1012 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1013 return reverse_transit;
1014 return true;
1015}
1016
31e6b01f
NP		 1017static int follow_dotdot_rcu(struct nameidata *nd)
1018{
1019 struct inode *inode = nd->inode;
1020
1021 set_root_rcu(nd);
1022
9875cf80 1023 while (1) {
31e6b01f
NP		 1024 if (nd->path.dentry == nd->root.dentry &&
1025 nd->path.mnt == nd->root.mnt) {
1026 break;
1027 }
1028 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1029 struct dentry *old = nd->path.dentry;
1030 struct dentry *parent = old->d_parent;
1031 unsigned seq;
1032
1033 seq = read_seqcount_begin(&parent->d_seq);
1034 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 1035 goto failed;
31e6b01f
NP		 1036 inode = parent->d_inode;
1037 nd->path.dentry = parent;
1038 nd->seq = seq;
1039 break;
1040 }
1041 if (!follow_up_rcu(&nd->path))
1042 break;
1043 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1044 inode = nd->path.dentry->d_inode;
1045 }
9875cf80 1046 __follow_mount_rcu(nd, &nd->path, &inode, true);
31e6b01f 1047 nd->inode = inode;
31e6b01f 1048 return 0;
ef7562d5
AV		 1049
1050failed:
1051 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1052 if (!(nd->flags & LOOKUP_ROOT))
1053 nd->root.mnt = NULL;
ef7562d5
AV		 1054 rcu_read_unlock();
1055 br_read_unlock(vfsmount_lock);
1056 return -ECHILD;
31e6b01f
NP		 1057}
1058
cc53ce53
DH		 1059/*
1060 * Follow down to the covering mount currently visible to userspace. At each
1061 * point, the filesystem owning that dentry may be queried as to whether the
1062 * caller is permitted to proceed or not.
1063 *
1064 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1065 * being true).
1066 */
7cc90cc3 1067int follow_down(struct path *path)
cc53ce53
DH		 1068{
1069 unsigned managed;
1070 int ret;
1071
1072 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1073 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1074 /* Allow the filesystem to manage the transit without i_mutex
1075 * being held.
1076 *
1077 * We indicate to the filesystem if someone is trying to mount
1078 * something here. This gives autofs the chance to deny anyone
1079 * other than its daemon the right to mount on its
1080 * superstructure.
1081 *
1082 * The filesystem may sleep at this point.
1083 */
1084 if (managed & DCACHE_MANAGE_TRANSIT) {
1085 BUG_ON(!path->dentry->d_op);
1086 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 1087 ret = path->dentry->d_op->d_manage(
1aed3e42 1088 path->dentry, false);
cc53ce53
DH		 1089 if (ret < 0)
1090 return ret == -EISDIR ? 0 : ret;
1091 }
1092
1093 /* Transit to a mounted filesystem. */
1094 if (managed & DCACHE_MOUNTED) {
1095 struct vfsmount *mounted = lookup_mnt(path);
1096 if (!mounted)
1097 break;
1098 dput(path->dentry);
1099 mntput(path->mnt);
1100 path->mnt = mounted;
1101 path->dentry = dget(mounted->mnt_root);
1102 continue;
1103 }
1104
1105 /* Don't handle automount points here */
1106 break;
1107 }
1108 return 0;
1109}
1110
9875cf80
DH		 1111/*
1112 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1113 */
1114static void follow_mount(struct path *path)
1115{
1116 while (d_mountpoint(path->dentry)) {
1117 struct vfsmount *mounted = lookup_mnt(path);
1118 if (!mounted)
1119 break;
1120 dput(path->dentry);
1121 mntput(path->mnt);
1122 path->mnt = mounted;
1123 path->dentry = dget(mounted->mnt_root);
1124 }
1125}
1126
31e6b01f 1127static void follow_dotdot(struct nameidata *nd)
1da177e4 1128{
2a737871 1129 set_root(nd);
e518ddb7 1130
1da177e4 1131 while(1) {
4ac91378 1132 struct dentry *old = nd->path.dentry;
1da177e4 1133
2a737871
AV		 1134 if (nd->path.dentry == nd->root.dentry &&
1135 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1136 break;
1137 }
4ac91378 1138 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1139 /* rare case of legitimate dget_parent()... */
1140 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1141 dput(old);
1142 break;
1143 }
3088dd70 1144 if (!follow_up(&nd->path))
1da177e4 1145 break;
1da177e4 1146 }
79ed0226 1147 follow_mount(&nd->path);
31e6b01f 1148 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1149}
1150
baa03890
NP		 1151/*
1152 * Allocate a dentry with name and parent, and perform a parent
1153 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1154 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1155 * have verified that no child exists while under i_mutex.
1156 */
1157static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1158 struct qstr *name, struct nameidata *nd)
1159{
1160 struct inode *inode = parent->d_inode;
1161 struct dentry *dentry;
1162 struct dentry *old;
1163
1164 /* Don't create child dentry for a dead directory. */
1165 if (unlikely(IS_DEADDIR(inode)))
1166 return ERR_PTR(-ENOENT);
1167
1168 dentry = d_alloc(parent, name);
1169 if (unlikely(!dentry))
1170 return ERR_PTR(-ENOMEM);
1171
1172 old = inode->i_op->lookup(inode, dentry, nd);
1173 if (unlikely(old)) {
1174 dput(dentry);
1175 dentry = old;
1176 }
1177 return dentry;
1178}
1179
1da177e4
LT		 1180/*
1181 * It's more convoluted than I'd like it to be, but... it's still fairly
1182 * small and for now I'd prefer to have fast path as straight as possible.
1183 * It _is_ time-critical.
1184 */
1185static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1186 struct path *path, struct inode **inode)
1da177e4 1187{
4ac91378 1188 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1189 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1190 int need_reval = 1;
1191 int status = 1;
9875cf80
DH		 1192 int err;
1193
b04f784e
NP		 1194 /*
1195 * Rename seqlock is not required here because in the off chance
1196 * of a false negative due to a concurrent rename, we're going to
1197 * do the non-racy lookup, below.
1198 */
31e6b01f
NP		 1199 if (nd->flags & LOOKUP_RCU) {
1200 unsigned seq;
31e6b01f
NP		 1201 *inode = nd->inode;
1202 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1203 if (!dentry)
1204 goto unlazy;
1205
31e6b01f
NP		 1206 /* Memory barrier in read_seqcount_begin of child is enough */
1207 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1208 return -ECHILD;
31e6b01f 1209 nd->seq = seq;
5a18fff2 1210
24643087 1211 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1212 status = d_revalidate(dentry, nd);
1213 if (unlikely(status <= 0)) {
1214 if (status != -ECHILD)
1215 need_reval = 0;
1216 goto unlazy;
1217 }
24643087 1218 }
31e6b01f
NP		 1219 path->mnt = mnt;
1220 path->dentry = dentry;
9875cf80
DH		 1221 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1222 return 0;
5a18fff2
AV		 1223unlazy:
1224 if (dentry) {
1225 if (nameidata_dentry_drop_rcu(nd, dentry))
1226 return -ECHILD;
1227 } else {
1228 if (nameidata_drop_rcu(nd))
1229 return -ECHILD;
1230 }
1231 } else {
1232 dentry = __d_lookup(parent, name);
9875cf80 1233 }
5a18fff2
AV		 1234
1235retry:
1236 if (unlikely(!dentry)) {
1237 struct inode *dir = parent->d_inode;
1238 BUG_ON(nd->inode != dir);
1239
1240 mutex_lock(&dir->i_mutex);
1241 dentry = d_lookup(parent, name);
1242 if (likely(!dentry)) {
1243 dentry = d_alloc_and_lookup(parent, name, nd);
1244 if (IS_ERR(dentry)) {
1245 mutex_unlock(&dir->i_mutex);
1246 return PTR_ERR(dentry);
1247 }
1248 /* known good */
1249 need_reval = 0;
1250 status = 1;
1251 }
1252 mutex_unlock(&dir->i_mutex);
1253 }
1254 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1255 status = d_revalidate(dentry, nd);
1256 if (unlikely(status <= 0)) {
1257 if (status < 0) {
1258 dput(dentry);
1259 return status;
1260 }
1261 if (!d_invalidate(dentry)) {
1262 dput(dentry);
1263 dentry = NULL;
1264 need_reval = 1;
1265 goto retry;
1266 }
24643087 1267 }
5a18fff2 1268
9875cf80
DH		 1269 path->mnt = mnt;
1270 path->dentry = dentry;
1271 err = follow_managed(path, nd->flags);
89312214
IK		 1272 if (unlikely(err < 0)) {
1273 path_put_conditional(path, nd);
9875cf80 1274 return err;
89312214 1275 }
9875cf80 1276 *inode = path->dentry->d_inode;
1da177e4 1277 return 0;
1da177e4
LT		 1278}
1279
52094c8a
AV		 1280static inline int may_lookup(struct nameidata *nd)
1281{
1282 if (nd->flags & LOOKUP_RCU) {
1283 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1284 if (err != -ECHILD)
1285 return err;
1286 if (nameidata_drop_rcu(nd))
1287 return -ECHILD;
1288 }
1289 return exec_permission(nd->inode, 0);
1290}
1291
9856fa1b
AV		 1292static inline int handle_dots(struct nameidata *nd, int type)
1293{
1294 if (type == LAST_DOTDOT) {
1295 if (nd->flags & LOOKUP_RCU) {
1296 if (follow_dotdot_rcu(nd))
1297 return -ECHILD;
1298 } else
1299 follow_dotdot(nd);
1300 }
1301 return 0;
1302}
1303
951361f9
AV		 1304static void terminate_walk(struct nameidata *nd)
1305{
1306 if (!(nd->flags & LOOKUP_RCU)) {
1307 path_put(&nd->path);
1308 } else {
1309 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1310 if (!(nd->flags & LOOKUP_ROOT))
1311 nd->root.mnt = NULL;
951361f9
AV		 1312 rcu_read_unlock();
1313 br_read_unlock(vfsmount_lock);
1314 }
1315}
1316
ce57dfc1
AV		 1317static inline int walk_component(struct nameidata *nd, struct path *path,
1318 struct qstr *name, int type, int follow)
1319{
1320 struct inode *inode;
1321 int err;
1322 /*
1323 * "." and ".." are special - ".." especially so because it has
1324 * to be able to know about the current root directory and
1325 * parent relationships.
1326 */
1327 if (unlikely(type != LAST_NORM))
1328 return handle_dots(nd, type);
1329 err = do_lookup(nd, name, path, &inode);
1330 if (unlikely(err)) {
1331 terminate_walk(nd);
1332 return err;
1333 }
1334 if (!inode) {
1335 path_to_nameidata(path, nd);
1336 terminate_walk(nd);
1337 return -ENOENT;
1338 }
1339 if (unlikely(inode->i_op->follow_link) && follow) {
1340 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
1341 return -ECHILD;
1342 BUG_ON(inode != path->dentry->d_inode);
1343 return 1;
1344 }
1345 path_to_nameidata(path, nd);
1346 nd->inode = inode;
1347 return 0;
1348}
1349
b356379a
AV		 1350/*
1351 * This limits recursive symlink follows to 8, while
1352 * limiting consecutive symlinks to 40.
1353 *
1354 * Without that kind of total limit, nasty chains of consecutive
1355 * symlinks can cause almost arbitrarily long lookups.
1356 */
1357static inline int nested_symlink(struct path *path, struct nameidata *nd)
1358{
1359 int res;
1360
1361 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1362 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1363 path_put_conditional(path, nd);
1364 path_put(&nd->path);
1365 return -ELOOP;
1366 }
1367
1368 nd->depth++;
1369 current->link_count++;
1370
1371 do {
1372 struct path link = *path;
1373 void *cookie;
574197e0
AV		 1374
1375 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1376 if (!res)
1377 res = walk_component(nd, path, &nd->last,
1378 nd->last_type, LOOKUP_FOLLOW);
574197e0 1379 put_link(nd, &link, cookie);
b356379a
AV		 1380 } while (res > 0);
1381
1382 current->link_count--;
1383 nd->depth--;
1384 return res;
1385}
1386
1da177e4
LT		 1387/*
1388 * Name resolution.
ea3834d9
PM		 1389 * This is the basic name resolution function, turning a pathname into
1390 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1391 *
ea3834d9
PM		 1392 * Returns 0 and nd will have valid dentry and mnt on success.
1393 * Returns error and drops reference to input namei data on failure.
1da177e4 1394 */
6de88d72 1395static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1396{
1397 struct path next;
1da177e4
LT		 1398 int err;
1399 unsigned int lookup_flags = nd->flags;
1400
1401 while (*name=='/')
1402 name++;
1403 if (!*name)
086e183a 1404 return 0;
1da177e4 1405
1da177e4
LT		 1406 /* At this point we know we have a real path component. */
1407 for(;;) {
1408 unsigned long hash;
1409 struct qstr this;
1410 unsigned int c;
fe479a58 1411 int type;
1da177e4 1412
cdce5d6b 1413 nd->flags |= LOOKUP_CONTINUE;
52094c8a
AV		 1414
1415 err = may_lookup(nd);
1da177e4
LT		 1416 if (err)
1417 break;
1418
1419 this.name = name;
1420 c = *(const unsigned char *)name;
1421
1422 hash = init_name_hash();
1423 do {
1424 name++;
1425 hash = partial_name_hash(c, hash);
1426 c = *(const unsigned char *)name;
1427 } while (c && (c != '/'));
1428 this.len = name - (const char *) this.name;
1429 this.hash = end_name_hash(hash);
1430
fe479a58
AV		 1431 type = LAST_NORM;
1432 if (this.name[0] == '.') switch (this.len) {
1433 case 2:
16c2cd71 1434 if (this.name[1] == '.') {
fe479a58 1435 type = LAST_DOTDOT;
16c2cd71
AV		 1436 nd->flags |= LOOKUP_JUMPED;
1437 }
fe479a58
AV		 1438 break;
1439 case 1:
1440 type = LAST_DOT;
1441 }
5a202bcd
AV		 1442 if (likely(type == LAST_NORM)) {
1443 struct dentry *parent = nd->path.dentry;
16c2cd71 1444 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1445 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1446 err = parent->d_op->d_hash(parent, nd->inode,
1447 &this);
1448 if (err < 0)
1449 break;
1450 }
1451 }
fe479a58 1452
1da177e4
LT		 1453 /* remove trailing slashes? */
1454 if (!c)
1455 goto last_component;
1456 while (*++name == '/');
1457 if (!*name)
b356379a 1458 goto last_component;
1da177e4 1459
ce57dfc1
AV		 1460 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1461 if (err < 0)
1462 return err;
1da177e4 1463
ce57dfc1 1464 if (err) {
b356379a 1465 err = nested_symlink(&next, nd);
1da177e4 1466 if (err)
a7472bab 1467 return err;
31e6b01f 1468 }
1da177e4 1469 err = -ENOTDIR;
31e6b01f 1470 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1471 break;
1472 continue;
1473 /* here ends the main loop */
1474
1da177e4 1475last_component:
f55eab82
TM		 1476 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1477 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
b356379a
AV		 1478 nd->last = this;
1479 nd->last_type = type;
086e183a 1480 return 0;
1da177e4 1481 }
951361f9 1482 terminate_walk(nd);
1da177e4
LT		 1483 return err;
1484}
1485
70e9b357
AV		 1486static int path_init(int dfd, const char *name, unsigned int flags,
1487 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1488{
1489 int retval = 0;
1490 int fput_needed;
1491 struct file *file;
1492
1493 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1494 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1495 nd->depth = 0;
5b6ca027
AV		 1496 if (flags & LOOKUP_ROOT) {
1497 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1498 if (*name) {
1499 if (!inode->i_op->lookup)
1500 return -ENOTDIR;
1501 retval = inode_permission(inode, MAY_EXEC);
1502 if (retval)
1503 return retval;
1504 }
5b6ca027
AV		 1505 nd->path = nd->root;
1506 nd->inode = inode;
1507 if (flags & LOOKUP_RCU) {
1508 br_read_lock(vfsmount_lock);
1509 rcu_read_lock();
1510 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1511 } else {
1512 path_get(&nd->path);
1513 }
1514 return 0;
1515 }
1516
31e6b01f 1517 nd->root.mnt = NULL;
31e6b01f
NP		 1518
1519 if (*name=='/') {
e41f7d4e
AV		 1520 if (flags & LOOKUP_RCU) {
1521 br_read_lock(vfsmount_lock);
1522 rcu_read_lock();
1523 set_root_rcu(nd);
1524 } else {
1525 set_root(nd);
1526 path_get(&nd->root);
1527 }
1528 nd->path = nd->root;
31e6b01f 1529 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1530 if (flags & LOOKUP_RCU) {
1531 struct fs_struct *fs = current->fs;
1532 unsigned seq;
31e6b01f 1533
e41f7d4e
AV		 1534 br_read_lock(vfsmount_lock);
1535 rcu_read_lock();
c28cc364 1536
e41f7d4e
AV		 1537 do {
1538 seq = read_seqcount_begin(&fs->seq);
1539 nd->path = fs->pwd;
1540 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1541 } while (read_seqcount_retry(&fs->seq, seq));
1542 } else {
1543 get_fs_pwd(current->fs, &nd->path);
1544 }
31e6b01f
NP		 1545 } else {
1546 struct dentry *dentry;
1547
1abf0c71 1548 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1549 retval = -EBADF;
1550 if (!file)
1551 goto out_fail;
1552
1553 dentry = file->f_path.dentry;
1554
f52e0c11
AV		 1555 if (*name) {
1556 retval = -ENOTDIR;
1557 if (!S_ISDIR(dentry->d_inode->i_mode))
1558 goto fput_fail;
31e6b01f 1559
f52e0c11
AV		 1560 retval = file_permission(file, MAY_EXEC);
1561 if (retval)
1562 goto fput_fail;
1563 }
31e6b01f
NP		 1564
1565 nd->path = file->f_path;
e41f7d4e
AV		 1566 if (flags & LOOKUP_RCU) {
1567 if (fput_needed)
70e9b357 1568 *fp = file;
e41f7d4e
AV		 1569 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1570 br_read_lock(vfsmount_lock);
1571 rcu_read_lock();
1572 } else {
1573 path_get(&file->f_path);
1574 fput_light(file, fput_needed);
1575 }
31e6b01f 1576 }
31e6b01f 1577
31e6b01f 1578 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1579 return 0;
2dfdd266 1580
9b4a9b14
AV		 1581fput_fail:
1582 fput_light(file, fput_needed);
1583out_fail:
1584 return retval;
1585}
1586
bd92d7fe
AV		 1587static inline int lookup_last(struct nameidata *nd, struct path *path)
1588{
1589 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1590 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1591
1592 nd->flags &= ~LOOKUP_PARENT;
1593 return walk_component(nd, path, &nd->last, nd->last_type,
1594 nd->flags & LOOKUP_FOLLOW);
1595}
1596
9b4a9b14 1597/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1598static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1599 unsigned int flags, struct nameidata *nd)
1600{
70e9b357 1601 struct file *base = NULL;
bd92d7fe
AV		 1602 struct path path;
1603 int err;
31e6b01f
NP		 1604
1605 /*
1606 * Path walking is largely split up into 2 different synchronisation
1607 * schemes, rcu-walk and ref-walk (explained in
1608 * Documentation/filesystems/path-lookup.txt). These share much of the
1609 * path walk code, but some things particularly setup, cleanup, and
1610 * following mounts are sufficiently divergent that functions are
1611 * duplicated. Typically there is a function foo(), and its RCU
1612 * analogue, foo_rcu().
1613 *
1614 * -ECHILD is the error number of choice (just to avoid clashes) that
1615 * is returned if some aspect of an rcu-walk fails. Such an error must
1616 * be handled by restarting a traditional ref-walk (which will always
1617 * be able to complete).
1618 */
bd92d7fe 1619 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1620
bd92d7fe
AV		 1621 if (unlikely(err))
1622 return err;
ee0827cd
AV		 1623
1624 current->total_link_count = 0;
bd92d7fe
AV		 1625 err = link_path_walk(name, nd);
1626
1627 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1628 err = lookup_last(nd, &path);
1629 while (err > 0) {
1630 void *cookie;
1631 struct path link = path;
bd92d7fe 1632 nd->flags |= LOOKUP_PARENT;
574197e0 1633 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1634 if (!err)
1635 err = lookup_last(nd, &path);
574197e0 1636 put_link(nd, &link, cookie);
bd92d7fe
AV		 1637 }
1638 }
ee0827cd
AV		 1639
1640 if (nd->flags & LOOKUP_RCU) {
4455ca62 1641 /* went all way through without dropping RCU */
bd92d7fe 1642 BUG_ON(err);
4455ca62 1643 if (nameidata_drop_rcu_last(nd))
bd92d7fe 1644 err = -ECHILD;
ee0827cd
AV		 1645 }
1646
bd92d7fe
AV		 1647 if (!err)
1648 err = handle_reval_path(nd);
1649
1650 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1651 if (!nd->inode->i_op->lookup) {
1652 path_put(&nd->path);
1653 return -ENOTDIR;
1654 }
1655 }
16c2cd71 1656
70e9b357
AV		 1657 if (base)
1658 fput(base);
ee0827cd 1659
5b6ca027 1660 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1661 path_put(&nd->root);
1662 nd->root.mnt = NULL;
1663 }
bd92d7fe 1664 return err;
ee0827cd 1665}
31e6b01f 1666
ee0827cd
AV		 1667static int do_path_lookup(int dfd, const char *name,
1668 unsigned int flags, struct nameidata *nd)
1669{
1670 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1671 if (unlikely(retval == -ECHILD))
1672 retval = path_lookupat(dfd, name, flags, nd);
1673 if (unlikely(retval == -ESTALE))
1674 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1675
1676 if (likely(!retval)) {
1677 if (unlikely(!audit_dummy_context())) {
1678 if (nd->path.dentry && nd->inode)
1679 audit_inode(name, nd->path.dentry);
1680 }
1681 }
170aa3d0 1682 return retval;
1da177e4
LT		 1683}
1684
c9c6cac0 1685int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1686{
c9c6cac0 1687 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1688}
1689
d1811465
AV		 1690int kern_path(const char *name, unsigned int flags, struct path *path)
1691{
1692 struct nameidata nd;
1693 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1694 if (!res)
1695 *path = nd.path;
1696 return res;
1697}
1698
16f18200
JJS		 1699/**
1700 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1701 * @dentry: pointer to dentry of the base directory
1702 * @mnt: pointer to vfs mount of the base directory
1703 * @name: pointer to file name
1704 * @flags: lookup flags
1705 * @nd: pointer to nameidata
1706 */
1707int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1708 const char *name, unsigned int flags,
1709 struct nameidata *nd)
1710{
5b6ca027
AV		 1711 nd->root.dentry = dentry;
1712 nd->root.mnt = mnt;
1713 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1714 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
16f18200
JJS		 1715}
1716
eead1911
CH		 1717static struct dentry *__lookup_hash(struct qstr *name,
1718 struct dentry *base, struct nameidata *nd)
1da177e4 1719{
81fca444 1720 struct inode *inode = base->d_inode;
057f6c01 1721 struct dentry *dentry;
1da177e4
LT		 1722 int err;
1723
b74c79e9 1724 err = exec_permission(inode, 0);
81fca444
CH		 1725 if (err)
1726 return ERR_PTR(err);
1da177e4 1727
b04f784e
NP		 1728 /*
1729 * Don't bother with __d_lookup: callers are for creat as
1730 * well as unlink, so a lot of the time it would cost
1731 * a double lookup.
6e6b1bd1 1732 */
b04f784e 1733 dentry = d_lookup(base, name);
6e6b1bd1 1734
fb045adb 1735 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1736 dentry = do_revalidate(dentry, nd);
1737
baa03890
NP		 1738 if (!dentry)
1739 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1740
1da177e4
LT		 1741 return dentry;
1742}
1743
057f6c01
JM		 1744/*
1745 * Restricted form of lookup. Doesn't follow links, single-component only,
1746 * needs parent already locked. Doesn't follow mounts.
1747 * SMP-safe.
1748 */
eead1911 1749static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1750{
4ac91378 1751 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1752}
1753
eead1911 1754/**
a6b91919 1755 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1756 * @name: pathname component to lookup
1757 * @base: base directory to lookup from
1758 * @len: maximum length @len should be interpreted to
1759 *
a6b91919
RD		 1760 * Note that this routine is purely a helper for filesystem usage and should
1761 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1762 * nameidata argument is passed to the filesystem methods and a filesystem
1763 * using this helper needs to be prepared for that.
1764 */
057f6c01
JM		 1765struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1766{
057f6c01 1767 struct qstr this;
6a96ba54
AV		 1768 unsigned long hash;
1769 unsigned int c;
057f6c01 1770
2f9092e1
DW		 1771 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1772
6a96ba54
AV		 1773 this.name = name;
1774 this.len = len;
1775 if (!len)
1776 return ERR_PTR(-EACCES);
1777
1778 hash = init_name_hash();
1779 while (len--) {
1780 c = *(const unsigned char *)name++;
1781 if (c == '/' || c == '\0')
1782 return ERR_PTR(-EACCES);
1783 hash = partial_name_hash(c, hash);
1784 }
1785 this.hash = end_name_hash(hash);
5a202bcd
AV		 1786 /*
1787 * See if the low-level filesystem might want
1788 * to use its own hash..
1789 */
1790 if (base->d_flags & DCACHE_OP_HASH) {
1791 int err = base->d_op->d_hash(base, base->d_inode, &this);
1792 if (err < 0)
1793 return ERR_PTR(err);
1794 }
eead1911 1795
49705b77 1796 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1797}
1798
2d8f3038
AV		 1799int user_path_at(int dfd, const char __user *name, unsigned flags,
1800 struct path *path)
1da177e4 1801{
2d8f3038 1802 struct nameidata nd;
f52e0c11 1803 char *tmp = getname_flags(name, flags);
1da177e4 1804 int err = PTR_ERR(tmp);
1da177e4 1805 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1806
1807 BUG_ON(flags & LOOKUP_PARENT);
1808
1809 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1810 putname(tmp);
2d8f3038
AV		 1811 if (!err)
1812 *path = nd.path;
1da177e4
LT		 1813 }
1814 return err;
1815}
1816
2ad94ae6
AV		 1817static int user_path_parent(int dfd, const char __user *path,
1818 struct nameidata *nd, char **name)
1819{
1820 char *s = getname(path);
1821 int error;
1822
1823 if (IS_ERR(s))
1824 return PTR_ERR(s);
1825
1826 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1827 if (error)
1828 putname(s);
1829 else
1830 *name = s;
1831
1832 return error;
1833}
1834
1da177e4
LT		 1835/*
1836 * It's inline, so penalty for filesystems that don't use sticky bit is
1837 * minimal.
1838 */
1839static inline int check_sticky(struct inode *dir, struct inode *inode)
1840{
da9592ed
DH		 1841 uid_t fsuid = current_fsuid();
1842
1da177e4
LT		 1843 if (!(dir->i_mode & S_ISVTX))
1844 return 0;
da9592ed 1845 if (inode->i_uid == fsuid)
1da177e4 1846 return 0;
da9592ed 1847 if (dir->i_uid == fsuid)
1da177e4
LT		 1848 return 0;
1849 return !capable(CAP_FOWNER);
1850}
1851
1852/*
1853 * Check whether we can remove a link victim from directory dir, check
1854 * whether the type of victim is right.
1855 * 1. We can't do it if dir is read-only (done in permission())
1856 * 2. We should have write and exec permissions on dir
1857 * 3. We can't remove anything from append-only dir
1858 * 4. We can't do anything with immutable dir (done in permission())
1859 * 5. If the sticky bit on dir is set we should either
1860 * a. be owner of dir, or
1861 * b. be owner of victim, or
1862 * c. have CAP_FOWNER capability
1863 * 6. If the victim is append-only or immutable we can't do antyhing with
1864 * links pointing to it.
1865 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1866 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1867 * 9. We can't remove a root or mountpoint.
1868 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1869 * nfs_async_unlink().
1870 */
858119e1 1871static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1872{
1873 int error;
1874
1875 if (!victim->d_inode)
1876 return -ENOENT;
1877
1878 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1879 audit_inode_child(victim, dir);
1da177e4 1880
f419a2e3 1881 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1882 if (error)
1883 return error;
1884 if (IS_APPEND(dir))
1885 return -EPERM;
1886 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1887 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1888 return -EPERM;
1889 if (isdir) {
1890 if (!S_ISDIR(victim->d_inode->i_mode))
1891 return -ENOTDIR;
1892 if (IS_ROOT(victim))
1893 return -EBUSY;
1894 } else if (S_ISDIR(victim->d_inode->i_mode))
1895 return -EISDIR;
1896 if (IS_DEADDIR(dir))
1897 return -ENOENT;
1898 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1899 return -EBUSY;
1900 return 0;
1901}
1902
1903/* Check whether we can create an object with dentry child in directory
1904 * dir.
1905 * 1. We can't do it if child already exists (open has special treatment for
1906 * this case, but since we are inlined it's OK)
1907 * 2. We can't do it if dir is read-only (done in permission())
1908 * 3. We should have write and exec permissions on dir
1909 * 4. We can't do it if dir is immutable (done in permission())
1910 */
a95164d9 1911static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1912{
1913 if (child->d_inode)
1914 return -EEXIST;
1915 if (IS_DEADDIR(dir))
1916 return -ENOENT;
f419a2e3 1917 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1918}
1919
1da177e4
LT		 1920/*
1921 * p1 and p2 should be directories on the same fs.
1922 */
1923struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1924{
1925 struct dentry *p;
1926
1927 if (p1 == p2) {
f2eace23 1928 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1929 return NULL;
1930 }
1931
a11f3a05 1932 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1933
e2761a11
OH		 1934 p = d_ancestor(p2, p1);
1935 if (p) {
1936 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1937 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1938 return p;
1da177e4
LT		 1939 }
1940
e2761a11
OH		 1941 p = d_ancestor(p1, p2);
1942 if (p) {
1943 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1944 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1945 return p;
1da177e4
LT		 1946 }
1947
f2eace23
IM		 1948 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1949 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1950 return NULL;
1951}
1952
1953void unlock_rename(struct dentry *p1, struct dentry *p2)
1954{
1b1dcc1b 1955 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1956 if (p1 != p2) {
1b1dcc1b 1957 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1958 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1959 }
1960}
1961
1962int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1963 struct nameidata *nd)
1964{
a95164d9 1965 int error = may_create(dir, dentry);
1da177e4
LT		 1966
1967 if (error)
1968 return error;
1969
acfa4380 1970 if (!dir->i_op->create)
1da177e4
LT		 1971 return -EACCES; /* shouldn't it be ENOSYS? */
1972 mode &= S_IALLUGO;
1973 mode |= S_IFREG;
1974 error = security_inode_create(dir, dentry, mode);
1975 if (error)
1976 return error;
1da177e4 1977 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1978 if (!error)
f38aa942 1979 fsnotify_create(dir, dentry);
1da177e4
LT		 1980 return error;
1981}
1982
73d049a4 1983static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1984{
3fb64190 1985 struct dentry *dentry = path->dentry;
1da177e4
LT		 1986 struct inode *inode = dentry->d_inode;
1987 int error;
1988
bcda7652
AV		 1989 /* O_PATH? */
1990 if (!acc_mode)
1991 return 0;
1992
1da177e4
LT		 1993 if (!inode)
1994 return -ENOENT;
1995
c8fe8f30
CH		 1996 switch (inode->i_mode & S_IFMT) {
1997 case S_IFLNK:
1da177e4 1998 return -ELOOP;
c8fe8f30
CH		 1999 case S_IFDIR:
2000 if (acc_mode & MAY_WRITE)
2001 return -EISDIR;
2002 break;
2003 case S_IFBLK:
2004 case S_IFCHR:
3fb64190 2005 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2006 return -EACCES;
c8fe8f30
CH		 2007 /*FALLTHRU*/
2008 case S_IFIFO:
2009 case S_IFSOCK:
1da177e4 2010 flag &= ~O_TRUNC;
c8fe8f30 2011 break;
4a3fd211 2012 }
b41572e9 2013
3fb64190 2014 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2015 if (error)
2016 return error;
6146f0d5 2017
1da177e4
LT		 2018 /*
2019 * An append-only file must be opened in append mode for writing.
2020 */
2021 if (IS_APPEND(inode)) {
8737c930 2022 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2023 return -EPERM;
1da177e4 2024 if (flag & O_TRUNC)
7715b521 2025 return -EPERM;
1da177e4
LT		 2026 }
2027
2028 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 2029 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2030 return -EPERM;
1da177e4
LT		 2031
2032 /*
2033 * Ensure there are no outstanding leases on the file.
2034 */
b65a9cfc 2035 return break_lease(inode, flag);
7715b521 2036}
1da177e4 2037
e1181ee6 2038static int handle_truncate(struct file *filp)
7715b521 2039{
e1181ee6 2040 struct path *path = &filp->f_path;
7715b521
AV		 2041 struct inode *inode = path->dentry->d_inode;
2042 int error = get_write_access(inode);
2043 if (error)
2044 return error;
2045 /*
2046 * Refuse to truncate files with mandatory locks held on them.
2047 */
2048 error = locks_verify_locked(inode);
2049 if (!error)
ea0d3ab2 2050 error = security_path_truncate(path);
7715b521
AV		 2051 if (!error) {
2052 error = do_truncate(path->dentry, 0,
2053 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2054 filp);
7715b521
AV		 2055 }
2056 put_write_access(inode);
acd0c935 2057 return error;
1da177e4
LT		 2058}
2059
d57999e1
DH		 2060/*
2061 * Note that while the flag value (low two bits) for sys_open means:
2062 * 00 - read-only
2063 * 01 - write-only
2064 * 10 - read-write
2065 * 11 - special
2066 * it is changed into
2067 * 00 - no permissions needed
2068 * 01 - read-permission
2069 * 10 - write-permission
2070 * 11 - read-write
2071 * for the internal routines (ie open_namei()/follow_link() etc)
2072 * This is more logical, and also allows the 00 "no perm needed"
2073 * to be used for symlinks (where the permissions are checked
2074 * later).
2075 *
2076*/
2077static inline int open_to_namei_flags(int flag)
2078{
2079 if ((flag+1) & O_ACCMODE)
2080 flag++;
2081 return flag;
2082}
2083
31e6b01f 2084/*
fe2d35ff 2085 * Handle the last step of open()
31e6b01f 2086 */
fb1cc555 2087static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2088 const struct open_flags *op, const char *pathname)
fb1cc555 2089{
a1e28038 2090 struct dentry *dir = nd->path.dentry;
6c0d46c4 2091 struct dentry *dentry;
ca344a89 2092 int open_flag = op->open_flag;
6c0d46c4 2093 int will_truncate = open_flag & O_TRUNC;
ca344a89 2094 int want_write = 0;
bcda7652 2095 int acc_mode = op->acc_mode;
fb1cc555 2096 struct file *filp;
16c2cd71 2097 int error;
1f36f774 2098
c3e380b0
AV		 2099 nd->flags &= ~LOOKUP_PARENT;
2100 nd->flags |= op->intent;
2101
1f36f774
AV		 2102 switch (nd->last_type) {
2103 case LAST_DOTDOT:
176306f5 2104 case LAST_DOT:
fe2d35ff
AV		 2105 error = handle_dots(nd, nd->last_type);
2106 if (error)
2107 return ERR_PTR(error);
1f36f774 2108 /* fallthrough */
1f36f774 2109 case LAST_ROOT:
fe2d35ff
AV		 2110 if (nd->flags & LOOKUP_RCU) {
2111 if (nameidata_drop_rcu_last(nd))
2112 return ERR_PTR(-ECHILD);
2113 }
16c2cd71
AV		 2114 error = handle_reval_path(nd);
2115 if (error)
2116 goto exit;
fe2d35ff 2117 audit_inode(pathname, nd->path.dentry);
ca344a89 2118 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2119 error = -EISDIR;
2120 goto exit;
2121 }
2122 goto ok;
1f36f774 2123 case LAST_BIND:
fe2d35ff 2124 /* can't be RCU mode here */
16c2cd71
AV		 2125 error = handle_reval_path(nd);
2126 if (error)
2127 goto exit;
1f36f774 2128 audit_inode(pathname, dir);
67ee3ad2 2129 goto ok;
1f36f774 2130 }
67ee3ad2 2131
ca344a89 2132 if (!(open_flag & O_CREAT)) {
bcda7652 2133 int symlink_ok = 0;
fe2d35ff
AV		 2134 if (nd->last.name[nd->last.len])
2135 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2136 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2137 symlink_ok = 1;
fe2d35ff 2138 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2139 error = walk_component(nd, path, &nd->last, LAST_NORM,
2140 !symlink_ok);
2141 if (error < 0)
fe2d35ff 2142 return ERR_PTR(error);
ce57dfc1 2143 if (error) /* symlink */
fe2d35ff 2144 return NULL;
fe2d35ff
AV		 2145 /* sayonara */
2146 if (nd->flags & LOOKUP_RCU) {
2147 if (nameidata_drop_rcu_last(nd))
2148 return ERR_PTR(-ECHILD);
2149 }
2150
2151 error = -ENOTDIR;
2152 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2153 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2154 goto exit;
2155 }
2156 audit_inode(pathname, nd->path.dentry);
2157 goto ok;
2158 }
2159
2160 /* create side of things */
2161
2162 if (nd->flags & LOOKUP_RCU) {
2163 if (nameidata_drop_rcu_last(nd))
2164 return ERR_PTR(-ECHILD);
2165 }
2166
2167 audit_inode(pathname, dir);
16c2cd71 2168 error = -EISDIR;
1f36f774 2169 /* trailing slashes? */
31e6b01f
NP		 2170 if (nd->last.name[nd->last.len])
2171 goto exit;
a2c36b45 2172
a1e28038
AV		 2173 mutex_lock(&dir->d_inode->i_mutex);
2174
6c0d46c4
AV		 2175 dentry = lookup_hash(nd);
2176 error = PTR_ERR(dentry);
2177 if (IS_ERR(dentry)) {
fb1cc555
AV		 2178 mutex_unlock(&dir->d_inode->i_mutex);
2179 goto exit;
2180 }
2181
6c0d46c4
AV		 2182 path->dentry = dentry;
2183 path->mnt = nd->path.mnt;
2184
fb1cc555 2185 /* Negative dentry, just create the file */
6c0d46c4
AV		 2186 if (!dentry->d_inode) {
2187 int mode = op->mode;
2188 if (!IS_POSIXACL(dir->d_inode))
2189 mode &= ~current_umask();
fb1cc555
AV		 2190 /*
2191 * This write is needed to ensure that a
6c0d46c4 2192 * rw->ro transition does not occur between
fb1cc555
AV		 2193 * the time when the file is created and when
2194 * a permanent write count is taken through
2195 * the 'struct file' in nameidata_to_filp().
2196 */
2197 error = mnt_want_write(nd->path.mnt);
2198 if (error)
2199 goto exit_mutex_unlock;
ca344a89 2200 want_write = 1;
9b44f1b3 2201 /* Don't check for write permission, don't truncate */
ca344a89 2202 open_flag &= ~O_TRUNC;
6c0d46c4 2203 will_truncate = 0;
bcda7652 2204 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2205 error = security_path_mknod(&nd->path, dentry, mode, 0);
2206 if (error)
2207 goto exit_mutex_unlock;
2208 error = vfs_create(dir->d_inode, dentry, mode, nd);
2209 if (error)
2210 goto exit_mutex_unlock;
2211 mutex_unlock(&dir->d_inode->i_mutex);
2212 dput(nd->path.dentry);
2213 nd->path.dentry = dentry;
ca344a89 2214 goto common;
fb1cc555
AV		 2215 }
2216
2217 /*
2218 * It already exists.
2219 */
2220 mutex_unlock(&dir->d_inode->i_mutex);
2221 audit_inode(pathname, path->dentry);
2222
2223 error = -EEXIST;
ca344a89 2224 if (open_flag & O_EXCL)
fb1cc555
AV		 2225 goto exit_dput;
2226
9875cf80
DH		 2227 error = follow_managed(path, nd->flags);
2228 if (error < 0)
2229 goto exit_dput;
fb1cc555
AV		 2230
2231 error = -ENOENT;
2232 if (!path->dentry->d_inode)
2233 goto exit_dput;
9e67f361
AV		 2234
2235 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2236 return NULL;
fb1cc555
AV		 2237
2238 path_to_nameidata(path, nd);
31e6b01f 2239 nd->inode = path->dentry->d_inode;
fb1cc555 2240 error = -EISDIR;
31e6b01f 2241 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2242 goto exit;
67ee3ad2 2243ok:
6c0d46c4
AV		 2244 if (!S_ISREG(nd->inode->i_mode))
2245 will_truncate = 0;
2246
0f9d1a10
AV		 2247 if (will_truncate) {
2248 error = mnt_want_write(nd->path.mnt);
2249 if (error)
2250 goto exit;
ca344a89 2251 want_write = 1;
0f9d1a10 2252 }
ca344a89 2253common:
bcda7652 2254 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2255 if (error)
0f9d1a10 2256 goto exit;
0f9d1a10
AV		 2257 filp = nameidata_to_filp(nd);
2258 if (!IS_ERR(filp)) {
2259 error = ima_file_check(filp, op->acc_mode);
2260 if (error) {
2261 fput(filp);
2262 filp = ERR_PTR(error);
2263 }
2264 }
2265 if (!IS_ERR(filp)) {
2266 if (will_truncate) {
2267 error = handle_truncate(filp);
2268 if (error) {
2269 fput(filp);
2270 filp = ERR_PTR(error);
2271 }
2272 }
2273 }
ca344a89
AV		 2274out:
2275 if (want_write)
0f9d1a10
AV		 2276 mnt_drop_write(nd->path.mnt);
2277 path_put(&nd->path);
fb1cc555
AV		 2278 return filp;
2279
2280exit_mutex_unlock:
2281 mutex_unlock(&dir->d_inode->i_mutex);
2282exit_dput:
2283 path_put_conditional(path, nd);
2284exit:
ca344a89
AV		 2285 filp = ERR_PTR(error);
2286 goto out;
fb1cc555
AV		 2287}
2288
13aab428 2289static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2290 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2291{
fe2d35ff 2292 struct file *base = NULL;
4a3fd211 2293 struct file *filp;
9850c056 2294 struct path path;
13aab428 2295 int error;
31e6b01f
NP		 2296
2297 filp = get_empty_filp();
2298 if (!filp)
2299 return ERR_PTR(-ENFILE);
2300
47c805dc 2301 filp->f_flags = op->open_flag;
73d049a4
AV		 2302 nd->intent.open.file = filp;
2303 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2304 nd->intent.open.create_mode = op->mode;
31e6b01f 2305
73d049a4 2306 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2307 if (unlikely(error))
13aab428 2308 goto out_filp;
31e6b01f 2309
fe2d35ff 2310 current->total_link_count = 0;
73d049a4 2311 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2312 if (unlikely(error))
2313 goto out_filp;
1da177e4 2314
73d049a4 2315 filp = do_last(nd, &path, op, pathname);
806b681c 2316 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2317 struct path link = path;
def4af30 2318 void *cookie;
574197e0 2319 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2320 path_put_conditional(&path, nd);
2321 path_put(&nd->path);
40b39136
AV		 2322 filp = ERR_PTR(-ELOOP);
2323 break;
2324 }
73d049a4
AV		 2325 nd->flags |= LOOKUP_PARENT;
2326 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2327 error = follow_link(&link, nd, &cookie);
c3e380b0 2328 if (unlikely(error))
f1afe9ef 2329 filp = ERR_PTR(error);
c3e380b0 2330 else
73d049a4 2331 filp = do_last(nd, &path, op, pathname);
574197e0 2332 put_link(nd, &link, cookie);
806b681c 2333 }
10fa8e62 2334out:
73d049a4
AV		 2335 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2336 path_put(&nd->root);
fe2d35ff
AV		 2337 if (base)
2338 fput(base);
73d049a4 2339 release_open_intent(nd);
10fa8e62 2340 return filp;
1da177e4 2341
31e6b01f 2342out_filp:
806b681c 2343 filp = ERR_PTR(error);
10fa8e62 2344 goto out;
1da177e4
LT		 2345}
2346
13aab428
AV		 2347struct file *do_filp_open(int dfd, const char *pathname,
2348 const struct open_flags *op, int flags)
2349{
73d049a4 2350 struct nameidata nd;
13aab428
AV		 2351 struct file *filp;
2352
73d049a4 2353 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2354 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2355 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2356 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2357 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2358 return filp;
2359}
2360
73d049a4
AV		 2361struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2362 const char *name, const struct open_flags *op, int flags)
2363{
2364 struct nameidata nd;
2365 struct file *file;
2366
2367 nd.root.mnt = mnt;
2368 nd.root.dentry = dentry;
2369
2370 flags |= LOOKUP_ROOT;
2371
bcda7652 2372 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2373 return ERR_PTR(-ELOOP);
2374
2375 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2376 if (unlikely(file == ERR_PTR(-ECHILD)))
2377 file = path_openat(-1, name, &nd, op, flags);
2378 if (unlikely(file == ERR_PTR(-ESTALE)))
2379 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2380 return file;
2381}
2382
1da177e4
LT		 2383/**
2384 * lookup_create - lookup a dentry, creating it if it doesn't exist
2385 * @nd: nameidata info
2386 * @is_dir: directory flag
2387 *
2388 * Simple function to lookup and return a dentry and create it
2389 * if it doesn't exist. Is SMP-safe.
c663e5d8 2390 *
4ac91378 2391 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2392 */
2393struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2394{
c663e5d8 2395 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2396
4ac91378 2397 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2398 /*
2399 * Yucky last component or no last component at all?
2400 * (foo/., foo/.., /////)
2401 */
1da177e4
LT		 2402 if (nd->last_type != LAST_NORM)
2403 goto fail;
2404 nd->flags &= ~LOOKUP_PARENT;
3516586a 2405 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2406 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2407
2408 /*
2409 * Do the final lookup.
2410 */
49705b77 2411 dentry = lookup_hash(nd);
1da177e4
LT		 2412 if (IS_ERR(dentry))
2413 goto fail;
c663e5d8 2414
e9baf6e5
AV		 2415 if (dentry->d_inode)
2416 goto eexist;
c663e5d8
CH		 2417 /*
2418 * Special case - lookup gave negative, but... we had foo/bar/
2419 * From the vfs_mknod() POV we just have a negative dentry -
2420 * all is fine. Let's be bastards - you had / on the end, you've
2421 * been asking for (non-existent) directory. -ENOENT for you.
2422 */
e9baf6e5
AV		 2423 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2424 dput(dentry);
2425 dentry = ERR_PTR(-ENOENT);
2426 }
1da177e4 2427 return dentry;
e9baf6e5 2428eexist:
1da177e4 2429 dput(dentry);
e9baf6e5 2430 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2431fail:
2432 return dentry;
2433}
f81a0bff 2434EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2435
2436int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2437{
a95164d9 2438 int error = may_create(dir, dentry);
1da177e4
LT		 2439
2440 if (error)
2441 return error;
2442
2443 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2444 return -EPERM;
2445
acfa4380 2446 if (!dir->i_op->mknod)
1da177e4
LT		 2447 return -EPERM;
2448
08ce5f16
SH		 2449 error = devcgroup_inode_mknod(mode, dev);
2450 if (error)
2451 return error;
2452
1da177e4
LT		 2453 error = security_inode_mknod(dir, dentry, mode, dev);
2454 if (error)
2455 return error;
2456
1da177e4 2457 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2458 if (!error)
f38aa942 2459 fsnotify_create(dir, dentry);
1da177e4
LT		 2460 return error;
2461}
2462
463c3197
DH		 2463static int may_mknod(mode_t mode)
2464{
2465 switch (mode & S_IFMT) {
2466 case S_IFREG:
2467 case S_IFCHR:
2468 case S_IFBLK:
2469 case S_IFIFO:
2470 case S_IFSOCK:
2471 case 0: /* zero mode translates to S_IFREG */
2472 return 0;
2473 case S_IFDIR:
2474 return -EPERM;
2475 default:
2476 return -EINVAL;
2477 }
2478}
2479
2e4d0924
HC		 2480SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2481 unsigned, dev)
1da177e4 2482{
2ad94ae6
AV		 2483 int error;
2484 char *tmp;
2485 struct dentry *dentry;
1da177e4
LT		 2486 struct nameidata nd;
2487
2488 if (S_ISDIR(mode))
2489 return -EPERM;
1da177e4 2490
2ad94ae6 2491 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2492 if (error)
2ad94ae6
AV		 2493 return error;
2494
1da177e4 2495 dentry = lookup_create(&nd, 0);
463c3197
DH		 2496 if (IS_ERR(dentry)) {
2497 error = PTR_ERR(dentry);
2498 goto out_unlock;
2499 }
4ac91378 2500 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2501 mode &= ~current_umask();
463c3197
DH		 2502 error = may_mknod(mode);
2503 if (error)
2504 goto out_dput;
2505 error = mnt_want_write(nd.path.mnt);
2506 if (error)
2507 goto out_dput;
be6d3e56
KT		 2508 error = security_path_mknod(&nd.path, dentry, mode, dev);
2509 if (error)
2510 goto out_drop_write;
463c3197 2511 switch (mode & S_IFMT) {
1da177e4 2512 case 0: case S_IFREG:
4ac91378 2513 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2514 break;
2515 case S_IFCHR: case S_IFBLK:
4ac91378 2516 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2517 new_decode_dev(dev));
2518 break;
2519 case S_IFIFO: case S_IFSOCK:
4ac91378 2520 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2521 break;
1da177e4 2522 }
be6d3e56 2523out_drop_write:
463c3197
DH		 2524 mnt_drop_write(nd.path.mnt);
2525out_dput:
2526 dput(dentry);
2527out_unlock:
4ac91378 2528 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2529 path_put(&nd.path);
1da177e4
LT		 2530 putname(tmp);
2531
2532 return error;
2533}
2534
3480b257 2535SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2536{
2537 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2538}
2539
1da177e4
LT		 2540int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2541{
a95164d9 2542 int error = may_create(dir, dentry);
1da177e4
LT		 2543
2544 if (error)
2545 return error;
2546
acfa4380 2547 if (!dir->i_op->mkdir)
1da177e4
LT		 2548 return -EPERM;
2549
2550 mode &= (S_IRWXUGO|S_ISVTX);
2551 error = security_inode_mkdir(dir, dentry, mode);
2552 if (error)
2553 return error;
2554
1da177e4 2555 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2556 if (!error)
f38aa942 2557 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2558 return error;
2559}
2560
2e4d0924 2561SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2562{
2563 int error = 0;
2564 char * tmp;
6902d925
DH		 2565 struct dentry *dentry;
2566 struct nameidata nd;
1da177e4 2567
2ad94ae6
AV		 2568 error = user_path_parent(dfd, pathname, &nd, &tmp);
2569 if (error)
6902d925 2570 goto out_err;
1da177e4 2571
6902d925
DH		 2572 dentry = lookup_create(&nd, 1);
2573 error = PTR_ERR(dentry);
2574 if (IS_ERR(dentry))
2575 goto out_unlock;
1da177e4 2576
4ac91378 2577 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2578 mode &= ~current_umask();
463c3197
DH		 2579 error = mnt_want_write(nd.path.mnt);
2580 if (error)
2581 goto out_dput;
be6d3e56
KT		 2582 error = security_path_mkdir(&nd.path, dentry, mode);
2583 if (error)
2584 goto out_drop_write;
4ac91378 2585 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2586out_drop_write:
463c3197
DH		 2587 mnt_drop_write(nd.path.mnt);
2588out_dput:
6902d925
DH		 2589 dput(dentry);
2590out_unlock:
4ac91378 2591 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2592 path_put(&nd.path);
6902d925
DH		 2593 putname(tmp);
2594out_err:
1da177e4
LT		 2595 return error;
2596}
2597
3cdad428 2598SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2599{
2600 return sys_mkdirat(AT_FDCWD, pathname, mode);
2601}
2602
1da177e4
LT		 2603/*
2604 * We try to drop the dentry early: we should have
2605 * a usage count of 2 if we're the only user of this
2606 * dentry, and if that is true (possibly after pruning
2607 * the dcache), then we drop the dentry now.
2608 *
2609 * A low-level filesystem can, if it choses, legally
2610 * do a
2611 *
2612 * if (!d_unhashed(dentry))
2613 * return -EBUSY;
2614 *
2615 * if it cannot handle the case of removing a directory
2616 * that is still in use by something else..
2617 */
2618void dentry_unhash(struct dentry *dentry)
2619{
2620 dget(dentry);
dc168427 2621 shrink_dcache_parent(dentry);
1da177e4 2622 spin_lock(&dentry->d_lock);
b7ab39f6 2623 if (dentry->d_count == 2)
1da177e4
LT		 2624 __d_drop(dentry);
2625 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2626}
2627
2628int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2629{
2630 int error = may_delete(dir, dentry, 1);
2631
2632 if (error)
2633 return error;
2634
acfa4380 2635 if (!dir->i_op->rmdir)
1da177e4
LT		 2636 return -EPERM;
2637
1b1dcc1b 2638 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2639 dentry_unhash(dentry);
2640 if (d_mountpoint(dentry))
2641 error = -EBUSY;
2642 else {
2643 error = security_inode_rmdir(dir, dentry);
2644 if (!error) {
2645 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2646 if (!error) {
1da177e4 2647 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2648 dont_mount(dentry);
2649 }
1da177e4
LT		 2650 }
2651 }
1b1dcc1b 2652 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2653 if (!error) {
1da177e4
LT		 2654 d_delete(dentry);
2655 }
2656 dput(dentry);
2657
2658 return error;
2659}
2660
5590ff0d 2661static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2662{
2663 int error = 0;
2664 char * name;
2665 struct dentry *dentry;
2666 struct nameidata nd;
2667
2ad94ae6 2668 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2669 if (error)
2ad94ae6 2670 return error;
1da177e4
LT		 2671
2672 switch(nd.last_type) {
0612d9fb
OH		 2673 case LAST_DOTDOT:
2674 error = -ENOTEMPTY;
2675 goto exit1;
2676 case LAST_DOT:
2677 error = -EINVAL;
2678 goto exit1;
2679 case LAST_ROOT:
2680 error = -EBUSY;
2681 goto exit1;
1da177e4 2682 }
0612d9fb
OH		 2683
2684 nd.flags &= ~LOOKUP_PARENT;
2685
4ac91378 2686 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2687 dentry = lookup_hash(&nd);
1da177e4 2688 error = PTR_ERR(dentry);
6902d925
DH		 2689 if (IS_ERR(dentry))
2690 goto exit2;
0622753b
DH		 2691 error = mnt_want_write(nd.path.mnt);
2692 if (error)
2693 goto exit3;
be6d3e56
KT		 2694 error = security_path_rmdir(&nd.path, dentry);
2695 if (error)
2696 goto exit4;
4ac91378 2697 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2698exit4:
0622753b
DH		 2699 mnt_drop_write(nd.path.mnt);
2700exit3:
6902d925
DH		 2701 dput(dentry);
2702exit2:
4ac91378 2703 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2704exit1:
1d957f9b 2705 path_put(&nd.path);
1da177e4
LT		 2706 putname(name);
2707 return error;
2708}
2709
3cdad428 2710SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2711{
2712 return do_rmdir(AT_FDCWD, pathname);
2713}
2714
1da177e4
LT		 2715int vfs_unlink(struct inode *dir, struct dentry *dentry)
2716{
2717 int error = may_delete(dir, dentry, 0);
2718
2719 if (error)
2720 return error;
2721
acfa4380 2722 if (!dir->i_op->unlink)
1da177e4
LT		 2723 return -EPERM;
2724
1b1dcc1b 2725 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2726 if (d_mountpoint(dentry))
2727 error = -EBUSY;
2728 else {
2729 error = security_inode_unlink(dir, dentry);
bec1052e 2730 if (!error) {
1da177e4 2731 error = dir->i_op->unlink(dir, dentry);
bec1052e 2732 if (!error)
d83c49f3 2733 dont_mount(dentry);
bec1052e 2734 }
1da177e4 2735 }
1b1dcc1b 2736 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2737
2738 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2739 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2740 fsnotify_link_count(dentry->d_inode);
e234f35c 2741 d_delete(dentry);
1da177e4 2742 }
0eeca283 2743
1da177e4
LT		 2744 return error;
2745}
2746
2747/*
2748 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2749 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2750 * writeout happening, and we don't want to prevent access to the directory
2751 * while waiting on the I/O.
2752 */
5590ff0d 2753static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2754{
2ad94ae6
AV		 2755 int error;
2756 char *name;
1da177e4
LT		 2757 struct dentry *dentry;
2758 struct nameidata nd;
2759 struct inode *inode = NULL;
2760
2ad94ae6 2761 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2762 if (error)
2ad94ae6
AV		 2763 return error;
2764
1da177e4
LT		 2765 error = -EISDIR;
2766 if (nd.last_type != LAST_NORM)
2767 goto exit1;
0612d9fb
OH		 2768
2769 nd.flags &= ~LOOKUP_PARENT;
2770
4ac91378 2771 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2772 dentry = lookup_hash(&nd);
1da177e4
LT		 2773 error = PTR_ERR(dentry);
2774 if (!IS_ERR(dentry)) {
2775 /* Why not before? Because we want correct error value */
2776 if (nd.last.name[nd.last.len])
2777 goto slashes;
2778 inode = dentry->d_inode;
2779 if (inode)
7de9c6ee 2780 ihold(inode);
0622753b
DH		 2781 error = mnt_want_write(nd.path.mnt);
2782 if (error)
2783 goto exit2;
be6d3e56
KT		 2784 error = security_path_unlink(&nd.path, dentry);
2785 if (error)
2786 goto exit3;
4ac91378 2787 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2788exit3:
0622753b 2789 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2790 exit2:
2791 dput(dentry);
2792 }
4ac91378 2793 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2794 if (inode)
2795 iput(inode); /* truncate the inode here */
2796exit1:
1d957f9b 2797 path_put(&nd.path);
1da177e4
LT		 2798 putname(name);
2799 return error;
2800
2801slashes:
2802 error = !dentry->d_inode ? -ENOENT :
2803 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2804 goto exit2;
2805}
2806
2e4d0924 2807SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2808{
2809 if ((flag & ~AT_REMOVEDIR) != 0)
2810 return -EINVAL;
2811
2812 if (flag & AT_REMOVEDIR)
2813 return do_rmdir(dfd, pathname);
2814
2815 return do_unlinkat(dfd, pathname);
2816}
2817
3480b257 2818SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2819{
2820 return do_unlinkat(AT_FDCWD, pathname);
2821}
2822
db2e747b 2823int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2824{
a95164d9 2825 int error = may_create(dir, dentry);
1da177e4
LT		 2826
2827 if (error)
2828 return error;
2829
acfa4380 2830 if (!dir->i_op->symlink)
1da177e4
LT		 2831 return -EPERM;
2832
2833 error = security_inode_symlink(dir, dentry, oldname);
2834 if (error)
2835 return error;
2836
1da177e4 2837 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2838 if (!error)
f38aa942 2839 fsnotify_create(dir, dentry);
1da177e4
LT		 2840 return error;
2841}
2842
2e4d0924
HC		 2843SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2844 int, newdfd, const char __user *, newname)
1da177e4 2845{
2ad94ae6
AV		 2846 int error;
2847 char *from;
2848 char *to;
6902d925
DH		 2849 struct dentry *dentry;
2850 struct nameidata nd;
1da177e4
LT		 2851
2852 from = getname(oldname);
2ad94ae6 2853 if (IS_ERR(from))
1da177e4 2854 return PTR_ERR(from);
1da177e4 2855
2ad94ae6 2856 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2857 if (error)
2ad94ae6
AV		 2858 goto out_putname;
2859
6902d925
DH		 2860 dentry = lookup_create(&nd, 0);
2861 error = PTR_ERR(dentry);
2862 if (IS_ERR(dentry))
2863 goto out_unlock;
2864
75c3f29d
DH		 2865 error = mnt_want_write(nd.path.mnt);
2866 if (error)
2867 goto out_dput;
be6d3e56
KT		 2868 error = security_path_symlink(&nd.path, dentry, from);
2869 if (error)
2870 goto out_drop_write;
db2e747b 2871 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2872out_drop_write:
75c3f29d
DH		 2873 mnt_drop_write(nd.path.mnt);
2874out_dput:
6902d925
DH		 2875 dput(dentry);
2876out_unlock:
4ac91378 2877 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2878 path_put(&nd.path);
6902d925
DH		 2879 putname(to);
2880out_putname:
1da177e4
LT		 2881 putname(from);
2882 return error;
2883}
2884
3480b257 2885SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2886{
2887 return sys_symlinkat(oldname, AT_FDCWD, newname);
2888}
2889
1da177e4
LT		 2890int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2891{
2892 struct inode *inode = old_dentry->d_inode;
2893 int error;
2894
2895 if (!inode)
2896 return -ENOENT;
2897
a95164d9 2898 error = may_create(dir, new_dentry);
1da177e4
LT		 2899 if (error)
2900 return error;
2901
2902 if (dir->i_sb != inode->i_sb)
2903 return -EXDEV;
2904
2905 /*
2906 * A link to an append-only or immutable file cannot be created.
2907 */
2908 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2909 return -EPERM;
acfa4380 2910 if (!dir->i_op->link)
1da177e4 2911 return -EPERM;
7e79eedb 2912 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2913 return -EPERM;
2914
2915 error = security_inode_link(old_dentry, dir, new_dentry);
2916 if (error)
2917 return error;
2918
7e79eedb 2919 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 2920 /* Make sure we don't allow creating hardlink to an unlinked file */
2921 if (inode->i_nlink == 0)
2922 error = -ENOENT;
2923 else
2924 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2925 mutex_unlock(&inode->i_mutex);
e31e14ec 2926 if (!error)
7e79eedb 2927 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2928 return error;
2929}
2930
2931/*
2932 * Hardlinks are often used in delicate situations. We avoid
2933 * security-related surprises by not following symlinks on the
2934 * newname. --KAB
2935 *
2936 * We don't follow them on the oldname either to be compatible
2937 * with linux 2.0, and to avoid hard-linking to directories
2938 * and other special files. --ADM
2939 */
2e4d0924
HC		 2940SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2941 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2942{
2943 struct dentry *new_dentry;
2d8f3038
AV		 2944 struct nameidata nd;
2945 struct path old_path;
11a7b371 2946 int how = 0;
1da177e4 2947 int error;
2ad94ae6 2948 char *to;
1da177e4 2949
11a7b371 2950 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 2951 return -EINVAL;
11a7b371
AK		 2952 /*
2953 * To use null names we require CAP_DAC_READ_SEARCH
2954 * This ensures that not everyone will be able to create
2955 * handlink using the passed filedescriptor.
2956 */
2957 if (flags & AT_EMPTY_PATH) {
2958 if (!capable(CAP_DAC_READ_SEARCH))
2959 return -ENOENT;
2960 how = LOOKUP_EMPTY;
2961 }
2962
2963 if (flags & AT_SYMLINK_FOLLOW)
2964 how |= LOOKUP_FOLLOW;
c04030e1 2965
11a7b371 2966 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 2967 if (error)
2ad94ae6
AV		 2968 return error;
2969
2970 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2971 if (error)
2972 goto out;
2973 error = -EXDEV;
2d8f3038 2974 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2975 goto out_release;
2976 new_dentry = lookup_create(&nd, 0);
2977 error = PTR_ERR(new_dentry);
6902d925
DH		 2978 if (IS_ERR(new_dentry))
2979 goto out_unlock;
75c3f29d
DH		 2980 error = mnt_want_write(nd.path.mnt);
2981 if (error)
2982 goto out_dput;
be6d3e56
KT		 2983 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2984 if (error)
2985 goto out_drop_write;
2d8f3038 2986 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2987out_drop_write:
75c3f29d
DH		 2988 mnt_drop_write(nd.path.mnt);
2989out_dput:
6902d925
DH		 2990 dput(new_dentry);
2991out_unlock:
4ac91378 2992 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2993out_release:
1d957f9b 2994 path_put(&nd.path);
2ad94ae6 2995 putname(to);
1da177e4 2996out:
2d8f3038 2997 path_put(&old_path);
1da177e4
LT		 2998
2999 return error;
3000}
3001
3480b257 3002SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3003{
c04030e1 3004 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3005}
3006
1da177e4
LT		 3007/*
3008 * The worst of all namespace operations - renaming directory. "Perverted"
3009 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3010 * Problems:
3011 * a) we can get into loop creation. Check is done in is_subdir().
3012 * b) race potential - two innocent renames can create a loop together.
3013 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3014 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3015 * story.
3016 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3017 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3018 * whether the target exists). Solution: try to be smart with locking
3019 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3020 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3021 * move will be locked. Thus we can rank directories by the tree
3022 * (ancestors first) and rank all non-directories after them.
3023 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3024 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3025 * HOWEVER, it relies on the assumption that any object with ->lookup()
3026 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3027 * we'd better make sure that there's no link(2) for them.
3028 * d) some filesystems don't support opened-but-unlinked directories,
3029 * either because of layout or because they are not ready to deal with
3030 * all cases correctly. The latter will be fixed (taking this sort of
3031 * stuff into VFS), but the former is not going away. Solution: the same
3032 * trick as in rmdir().
3033 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3034 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3035 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3036 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3037 * locking].
3038 */
75c96f85
AB		 3039static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3040 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3041{
3042 int error = 0;
3043 struct inode *target;
3044
3045 /*
3046 * If we are going to change the parent - check write permissions,
3047 * we'll need to flip '..'.
3048 */
3049 if (new_dir != old_dir) {
f419a2e3 3050 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3051 if (error)
3052 return error;
3053 }
3054
3055 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3056 if (error)
3057 return error;
3058
3059 target = new_dentry->d_inode;
d83c49f3 3060 if (target)
1b1dcc1b 3061 mutex_lock(&target->i_mutex);
1da177e4
LT		 3062 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3063 error = -EBUSY;
d83c49f3
AV		 3064 else {
3065 if (target)
3066 dentry_unhash(new_dentry);
1da177e4 3067 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3068 }
1da177e4 3069 if (target) {
d83c49f3 3070 if (!error) {
1da177e4 3071 target->i_flags |= S_DEAD;
d83c49f3
AV		 3072 dont_mount(new_dentry);
3073 }
1b1dcc1b 3074 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3075 if (d_unhashed(new_dentry))
3076 d_rehash(new_dentry);
3077 dput(new_dentry);
3078 }
e31e14ec 3079 if (!error)
349457cc
MF		 3080 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3081 d_move(old_dentry,new_dentry);
1da177e4
LT		 3082 return error;
3083}
3084
75c96f85
AB		 3085static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3086 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3087{
3088 struct inode *target;
3089 int error;
3090
3091 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3092 if (error)
3093 return error;
3094
3095 dget(new_dentry);
3096 target = new_dentry->d_inode;
3097 if (target)
1b1dcc1b 3098 mutex_lock(&target->i_mutex);
1da177e4
LT		 3099 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3100 error = -EBUSY;
3101 else
3102 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3103 if (!error) {
bec1052e 3104 if (target)
d83c49f3 3105 dont_mount(new_dentry);
349457cc 3106 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3107 d_move(old_dentry, new_dentry);
1da177e4
LT		 3108 }
3109 if (target)
1b1dcc1b 3110 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3111 dput(new_dentry);
3112 return error;
3113}
3114
3115int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3116 struct inode *new_dir, struct dentry *new_dentry)
3117{
3118 int error;
3119 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3120 const unsigned char *old_name;
1da177e4
LT		 3121
3122 if (old_dentry->d_inode == new_dentry->d_inode)
3123 return 0;
3124
3125 error = may_delete(old_dir, old_dentry, is_dir);
3126 if (error)
3127 return error;
3128
3129 if (!new_dentry->d_inode)
a95164d9 3130 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3131 else
3132 error = may_delete(new_dir, new_dentry, is_dir);
3133 if (error)
3134 return error;
3135
acfa4380 3136 if (!old_dir->i_op->rename)
1da177e4
LT		 3137 return -EPERM;
3138
0eeca283
RL		 3139 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3140
1da177e4
LT		 3141 if (is_dir)
3142 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3143 else
3144 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3145 if (!error)
3146 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3147 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3148 fsnotify_oldname_free(old_name);
3149
1da177e4
LT		 3150 return error;
3151}
3152
2e4d0924
HC		 3153SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3154 int, newdfd, const char __user *, newname)
1da177e4 3155{
2ad94ae6
AV		 3156 struct dentry *old_dir, *new_dir;
3157 struct dentry *old_dentry, *new_dentry;
3158 struct dentry *trap;
1da177e4 3159 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3160 char *from;
3161 char *to;
3162 int error;
1da177e4 3163
2ad94ae6 3164 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3165 if (error)
3166 goto exit;
3167
2ad94ae6 3168 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3169 if (error)
3170 goto exit1;
3171
3172 error = -EXDEV;
4ac91378 3173 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3174 goto exit2;
3175
4ac91378 3176 old_dir = oldnd.path.dentry;
1da177e4
LT		 3177 error = -EBUSY;
3178 if (oldnd.last_type != LAST_NORM)
3179 goto exit2;
3180
4ac91378 3181 new_dir = newnd.path.dentry;
1da177e4
LT		 3182 if (newnd.last_type != LAST_NORM)
3183 goto exit2;
3184
0612d9fb
OH		 3185 oldnd.flags &= ~LOOKUP_PARENT;
3186 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3187 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3188
1da177e4
LT		 3189 trap = lock_rename(new_dir, old_dir);
3190
49705b77 3191 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3192 error = PTR_ERR(old_dentry);
3193 if (IS_ERR(old_dentry))
3194 goto exit3;
3195 /* source must exist */
3196 error = -ENOENT;
3197 if (!old_dentry->d_inode)
3198 goto exit4;
3199 /* unless the source is a directory trailing slashes give -ENOTDIR */
3200 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3201 error = -ENOTDIR;
3202 if (oldnd.last.name[oldnd.last.len])
3203 goto exit4;
3204 if (newnd.last.name[newnd.last.len])
3205 goto exit4;
3206 }
3207 /* source should not be ancestor of target */
3208 error = -EINVAL;
3209 if (old_dentry == trap)
3210 goto exit4;
49705b77 3211 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3212 error = PTR_ERR(new_dentry);
3213 if (IS_ERR(new_dentry))
3214 goto exit4;
3215 /* target should not be an ancestor of source */
3216 error = -ENOTEMPTY;
3217 if (new_dentry == trap)
3218 goto exit5;
3219
9079b1eb
DH		 3220 error = mnt_want_write(oldnd.path.mnt);
3221 if (error)
3222 goto exit5;
be6d3e56
KT		 3223 error = security_path_rename(&oldnd.path, old_dentry,
3224 &newnd.path, new_dentry);
3225 if (error)
3226 goto exit6;
1da177e4
LT		 3227 error = vfs_rename(old_dir->d_inode, old_dentry,
3228 new_dir->d_inode, new_dentry);
be6d3e56 3229exit6:
9079b1eb 3230 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3231exit5:
3232 dput(new_dentry);
3233exit4:
3234 dput(old_dentry);
3235exit3:
3236 unlock_rename(new_dir, old_dir);
3237exit2:
1d957f9b 3238 path_put(&newnd.path);
2ad94ae6 3239 putname(to);
1da177e4 3240exit1:
1d957f9b 3241 path_put(&oldnd.path);
1da177e4 3242 putname(from);
2ad94ae6 3243exit:
1da177e4
LT		 3244 return error;
3245}
3246
a26eab24 3247SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3248{
3249 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3250}
3251
1da177e4
LT		 3252int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3253{
3254 int len;
3255
3256 len = PTR_ERR(link);
3257 if (IS_ERR(link))
3258 goto out;
3259
3260 len = strlen(link);
3261 if (len > (unsigned) buflen)
3262 len = buflen;
3263 if (copy_to_user(buffer, link, len))
3264 len = -EFAULT;
3265out:
3266 return len;
3267}
3268
3269/*
3270 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3271 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3272 * using) it for any given inode is up to filesystem.
3273 */
3274int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3275{
3276 struct nameidata nd;
cc314eef 3277 void *cookie;
694a1764 3278 int res;
cc314eef 3279
1da177e4 3280 nd.depth = 0;
cc314eef 3281 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3282 if (IS_ERR(cookie))
3283 return PTR_ERR(cookie);
3284
3285 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3286 if (dentry->d_inode->i_op->put_link)
3287 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3288 return res;
1da177e4
LT		 3289}
3290
3291int vfs_follow_link(struct nameidata *nd, const char *link)
3292{
3293 return __vfs_follow_link(nd, link);
3294}
3295
3296/* get the link contents into pagecache */
3297static char *page_getlink(struct dentry * dentry, struct page **ppage)
3298{
ebd09abb
DG		 3299 char *kaddr;
3300 struct page *page;
1da177e4 3301 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3302 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3303 if (IS_ERR(page))
6fe6900e 3304 return (char*)page;
1da177e4 3305 *ppage = page;
ebd09abb
DG		 3306 kaddr = kmap(page);
3307 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3308 return kaddr;
1da177e4
LT		 3309}
3310
3311int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3312{
3313 struct page *page = NULL;
3314 char *s = page_getlink(dentry, &page);
3315 int res = vfs_readlink(dentry,buffer,buflen,s);
3316 if (page) {
3317 kunmap(page);
3318 page_cache_release(page);
3319 }
3320 return res;
3321}
3322
cc314eef 3323void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3324{
cc314eef 3325 struct page *page = NULL;
1da177e4 3326 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3327 return page;
1da177e4
LT		 3328}
3329
cc314eef 3330void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3331{
cc314eef
LT		 3332 struct page *page = cookie;
3333
3334 if (page) {
1da177e4
LT		 3335 kunmap(page);
3336 page_cache_release(page);
1da177e4
LT		 3337 }
3338}
3339
54566b2c
NP		 3340/*
3341 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3342 */
3343int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3344{
3345 struct address_space *mapping = inode->i_mapping;
0adb25d2 3346 struct page *page;
afddba49 3347 void *fsdata;
beb497ab 3348 int err;
1da177e4 3349 char *kaddr;
54566b2c
NP		 3350 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3351 if (nofs)
3352 flags |= AOP_FLAG_NOFS;
1da177e4 3353
7e53cac4 3354retry:
afddba49 3355 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3356 flags, &page, &fsdata);
1da177e4 3357 if (err)
afddba49
NP		 3358 goto fail;
3359
1da177e4
LT		 3360 kaddr = kmap_atomic(page, KM_USER0);
3361 memcpy(kaddr, symname, len-1);
3362 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3363
3364 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3365 page, fsdata);
1da177e4
LT		 3366 if (err < 0)
3367 goto fail;
afddba49
NP		 3368 if (err < len-1)
3369 goto retry;
3370
1da177e4
LT		 3371 mark_inode_dirty(inode);
3372 return 0;
1da177e4
LT		 3373fail:
3374 return err;
3375}
3376
0adb25d2
KK		 3377int page_symlink(struct inode *inode, const char *symname, int len)
3378{
3379 return __page_symlink(inode, symname, len,
54566b2c 3380 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3381}
3382
92e1d5be 3383const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3384 .readlink = generic_readlink,
3385 .follow_link = page_follow_link_light,
3386 .put_link = page_put_link,
3387};
3388
2d8f3038 3389EXPORT_SYMBOL(user_path_at);
cc53ce53 3390EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3391EXPORT_SYMBOL(follow_down);
3392EXPORT_SYMBOL(follow_up);
3393EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3394EXPORT_SYMBOL(getname);
3395EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3396EXPORT_SYMBOL(lookup_one_len);
3397EXPORT_SYMBOL(page_follow_link_light);
3398EXPORT_SYMBOL(page_put_link);
3399EXPORT_SYMBOL(page_readlink);
0adb25d2 3400EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3401EXPORT_SYMBOL(page_symlink);
3402EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3403EXPORT_SYMBOL(kern_path_parent);
d1811465 3404EXPORT_SYMBOL(kern_path);
16f18200 3405EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3406EXPORT_SYMBOL(inode_permission);
8c744fb8 3407EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3408EXPORT_SYMBOL(unlock_rename);
3409EXPORT_SYMBOL(vfs_create);
3410EXPORT_SYMBOL(vfs_follow_link);
3411EXPORT_SYMBOL(vfs_link);
3412EXPORT_SYMBOL(vfs_mkdir);
3413EXPORT_SYMBOL(vfs_mknod);
3414EXPORT_SYMBOL(generic_permission);
3415EXPORT_SYMBOL(vfs_readlink);
3416EXPORT_SYMBOL(vfs_rename);
3417EXPORT_SYMBOL(vfs_rmdir);
3418EXPORT_SYMBOL(vfs_symlink);
3419EXPORT_SYMBOL(vfs_unlink);
3420EXPORT_SYMBOL(dentry_unhash);
3421EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)