projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
don't pass 'mounting_here' flag to follow_down()
[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);
ab90911f
DH		 936 ret = path->dentry->d_op->d_manage(path->dentry,
937 false, false);
cc53ce53
DH		 938 if (ret < 0)
939 return ret == -EISDIR ? 0 : ret;
940 }
941
9875cf80
DH		 942 /* Transit to a mounted filesystem. */
943 if (managed & DCACHE_MOUNTED) {
944 struct vfsmount *mounted = lookup_mnt(path);
945 if (mounted) {
946 dput(path->dentry);
947 if (need_mntput)
948 mntput(path->mnt);
949 path->mnt = mounted;
950 path->dentry = dget(mounted->mnt_root);
951 need_mntput = true;
952 continue;
953 }
954
955 /* Something is mounted on this dentry in another
956 * namespace and/or whatever was mounted there in this
957 * namespace got unmounted before we managed to get the
958 * vfsmount_lock */
959 }
960
961 /* Handle an automount point */
962 if (managed & DCACHE_NEED_AUTOMOUNT) {
963 ret = follow_automount(path, flags, &need_mntput);
964 if (ret < 0)
965 return ret == -EISDIR ? 0 : ret;
966 continue;
967 }
968
969 /* We didn't change the current path point */
970 break;
1da177e4 971 }
9875cf80 972 return 0;
1da177e4
LT		 973}
974
cc53ce53 975int follow_down_one(struct path *path)
1da177e4
LT		 976{
977 struct vfsmount *mounted;
978
1c755af4 979 mounted = lookup_mnt(path);
1da177e4 980 if (mounted) {
9393bd07
AV		 981 dput(path->dentry);
982 mntput(path->mnt);
983 path->mnt = mounted;
984 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 985 return 1;
986 }
987 return 0;
988}
989
9875cf80
DH		 990/*
991 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
cc53ce53 992 * meet a managed dentry and we're not walking to "..". True is returned to
9875cf80
DH		 993 * continue, false to abort.
994 */
995static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
996 struct inode **inode, bool reverse_transit)
997{
998 while (d_mountpoint(path->dentry)) {
999 struct vfsmount *mounted;
ab90911f
DH		 1000 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1001 !reverse_transit &&
1002 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1003 return false;
9875cf80
DH		 1004 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1005 if (!mounted)
1006 break;
1007 path->mnt = mounted;
1008 path->dentry = mounted->mnt_root;
1009 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1010 *inode = path->dentry->d_inode;
1011 }
1012
1013 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1014 return reverse_transit;
1015 return true;
1016}
1017
31e6b01f
NP		 1018static int follow_dotdot_rcu(struct nameidata *nd)
1019{
1020 struct inode *inode = nd->inode;
1021
1022 set_root_rcu(nd);
1023
9875cf80 1024 while (1) {
31e6b01f
NP		 1025 if (nd->path.dentry == nd->root.dentry &&
1026 nd->path.mnt == nd->root.mnt) {
1027 break;
1028 }
1029 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1030 struct dentry *old = nd->path.dentry;
1031 struct dentry *parent = old->d_parent;
1032 unsigned seq;
1033
1034 seq = read_seqcount_begin(&parent->d_seq);
1035 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 1036 goto failed;
31e6b01f
NP		 1037 inode = parent->d_inode;
1038 nd->path.dentry = parent;
1039 nd->seq = seq;
1040 break;
1041 }
1042 if (!follow_up_rcu(&nd->path))
1043 break;
1044 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1045 inode = nd->path.dentry->d_inode;
1046 }
9875cf80 1047 __follow_mount_rcu(nd, &nd->path, &inode, true);
31e6b01f 1048 nd->inode = inode;
31e6b01f 1049 return 0;
ef7562d5
AV		 1050
1051failed:
1052 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1053 if (!(nd->flags & LOOKUP_ROOT))
1054 nd->root.mnt = NULL;
ef7562d5
AV		 1055 rcu_read_unlock();
1056 br_read_unlock(vfsmount_lock);
1057 return -ECHILD;
31e6b01f
NP		 1058}
1059
cc53ce53
DH		 1060/*
1061 * Follow down to the covering mount currently visible to userspace. At each
1062 * point, the filesystem owning that dentry may be queried as to whether the
1063 * caller is permitted to proceed or not.
1064 *
1065 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1066 * being true).
1067 */
7cc90cc3 1068int follow_down(struct path *path)
cc53ce53
DH		 1069{
1070 unsigned managed;
1071 int ret;
1072
1073 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1074 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1075 /* Allow the filesystem to manage the transit without i_mutex
1076 * being held.
1077 *
1078 * We indicate to the filesystem if someone is trying to mount
1079 * something here. This gives autofs the chance to deny anyone
1080 * other than its daemon the right to mount on its
1081 * superstructure.
1082 *
1083 * The filesystem may sleep at this point.
1084 */
1085 if (managed & DCACHE_MANAGE_TRANSIT) {
1086 BUG_ON(!path->dentry->d_op);
1087 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 1088 ret = path->dentry->d_op->d_manage(
7cc90cc3 1089 path->dentry, false, false);
cc53ce53
DH		 1090 if (ret < 0)
1091 return ret == -EISDIR ? 0 : ret;
1092 }
1093
1094 /* Transit to a mounted filesystem. */
1095 if (managed & DCACHE_MOUNTED) {
1096 struct vfsmount *mounted = lookup_mnt(path);
1097 if (!mounted)
1098 break;
1099 dput(path->dentry);
1100 mntput(path->mnt);
1101 path->mnt = mounted;
1102 path->dentry = dget(mounted->mnt_root);
1103 continue;
1104 }
1105
1106 /* Don't handle automount points here */
1107 break;
1108 }
1109 return 0;
1110}
1111
9875cf80
DH		 1112/*
1113 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1114 */
1115static void follow_mount(struct path *path)
1116{
1117 while (d_mountpoint(path->dentry)) {
1118 struct vfsmount *mounted = lookup_mnt(path);
1119 if (!mounted)
1120 break;
1121 dput(path->dentry);
1122 mntput(path->mnt);
1123 path->mnt = mounted;
1124 path->dentry = dget(mounted->mnt_root);
1125 }
1126}
1127
31e6b01f 1128static void follow_dotdot(struct nameidata *nd)
1da177e4 1129{
2a737871 1130 set_root(nd);
e518ddb7 1131
1da177e4 1132 while(1) {
4ac91378 1133 struct dentry *old = nd->path.dentry;
1da177e4 1134
2a737871
AV		 1135 if (nd->path.dentry == nd->root.dentry &&
1136 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1137 break;
1138 }
4ac91378 1139 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1140 /* rare case of legitimate dget_parent()... */
1141 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1142 dput(old);
1143 break;
1144 }
3088dd70 1145 if (!follow_up(&nd->path))
1da177e4 1146 break;
1da177e4 1147 }
79ed0226 1148 follow_mount(&nd->path);
31e6b01f 1149 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1150}
1151
baa03890
NP		 1152/*
1153 * Allocate a dentry with name and parent, and perform a parent
1154 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1155 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1156 * have verified that no child exists while under i_mutex.
1157 */
1158static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1159 struct qstr *name, struct nameidata *nd)
1160{
1161 struct inode *inode = parent->d_inode;
1162 struct dentry *dentry;
1163 struct dentry *old;
1164
1165 /* Don't create child dentry for a dead directory. */
1166 if (unlikely(IS_DEADDIR(inode)))
1167 return ERR_PTR(-ENOENT);
1168
1169 dentry = d_alloc(parent, name);
1170 if (unlikely(!dentry))
1171 return ERR_PTR(-ENOMEM);
1172
1173 old = inode->i_op->lookup(inode, dentry, nd);
1174 if (unlikely(old)) {
1175 dput(dentry);
1176 dentry = old;
1177 }
1178 return dentry;
1179}
1180
1da177e4
LT		 1181/*
1182 * It's more convoluted than I'd like it to be, but... it's still fairly
1183 * small and for now I'd prefer to have fast path as straight as possible.
1184 * It _is_ time-critical.
1185 */
1186static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1187 struct path *path, struct inode **inode)
1da177e4 1188{
4ac91378 1189 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1190 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1191 int need_reval = 1;
1192 int status = 1;
9875cf80
DH		 1193 int err;
1194
b04f784e
NP		 1195 /*
1196 * Rename seqlock is not required here because in the off chance
1197 * of a false negative due to a concurrent rename, we're going to
1198 * do the non-racy lookup, below.
1199 */
31e6b01f
NP		 1200 if (nd->flags & LOOKUP_RCU) {
1201 unsigned seq;
31e6b01f
NP		 1202 *inode = nd->inode;
1203 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1204 if (!dentry)
1205 goto unlazy;
1206
31e6b01f
NP		 1207 /* Memory barrier in read_seqcount_begin of child is enough */
1208 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1209 return -ECHILD;
31e6b01f 1210 nd->seq = seq;
5a18fff2 1211
24643087 1212 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1213 status = d_revalidate(dentry, nd);
1214 if (unlikely(status <= 0)) {
1215 if (status != -ECHILD)
1216 need_reval = 0;
1217 goto unlazy;
1218 }
24643087 1219 }
31e6b01f
NP		 1220 path->mnt = mnt;
1221 path->dentry = dentry;
9875cf80
DH		 1222 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1223 return 0;
5a18fff2
AV		 1224unlazy:
1225 if (dentry) {
1226 if (nameidata_dentry_drop_rcu(nd, dentry))
1227 return -ECHILD;
1228 } else {
1229 if (nameidata_drop_rcu(nd))
1230 return -ECHILD;
1231 }
1232 } else {
1233 dentry = __d_lookup(parent, name);
9875cf80 1234 }
5a18fff2
AV		 1235
1236retry:
1237 if (unlikely(!dentry)) {
1238 struct inode *dir = parent->d_inode;
1239 BUG_ON(nd->inode != dir);
1240
1241 mutex_lock(&dir->i_mutex);
1242 dentry = d_lookup(parent, name);
1243 if (likely(!dentry)) {
1244 dentry = d_alloc_and_lookup(parent, name, nd);
1245 if (IS_ERR(dentry)) {
1246 mutex_unlock(&dir->i_mutex);
1247 return PTR_ERR(dentry);
1248 }
1249 /* known good */
1250 need_reval = 0;
1251 status = 1;
1252 }
1253 mutex_unlock(&dir->i_mutex);
1254 }
1255 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1256 status = d_revalidate(dentry, nd);
1257 if (unlikely(status <= 0)) {
1258 if (status < 0) {
1259 dput(dentry);
1260 return status;
1261 }
1262 if (!d_invalidate(dentry)) {
1263 dput(dentry);
1264 dentry = NULL;
1265 need_reval = 1;
1266 goto retry;
1267 }
24643087 1268 }
5a18fff2 1269
9875cf80
DH		 1270 path->mnt = mnt;
1271 path->dentry = dentry;
1272 err = follow_managed(path, nd->flags);
89312214
IK		 1273 if (unlikely(err < 0)) {
1274 path_put_conditional(path, nd);
9875cf80 1275 return err;
89312214 1276 }
9875cf80 1277 *inode = path->dentry->d_inode;
1da177e4 1278 return 0;
1da177e4
LT		 1279}
1280
52094c8a
AV		 1281static inline int may_lookup(struct nameidata *nd)
1282{
1283 if (nd->flags & LOOKUP_RCU) {
1284 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1285 if (err != -ECHILD)
1286 return err;
1287 if (nameidata_drop_rcu(nd))
1288 return -ECHILD;
1289 }
1290 return exec_permission(nd->inode, 0);
1291}
1292
9856fa1b
AV		 1293static inline int handle_dots(struct nameidata *nd, int type)
1294{
1295 if (type == LAST_DOTDOT) {
1296 if (nd->flags & LOOKUP_RCU) {
1297 if (follow_dotdot_rcu(nd))
1298 return -ECHILD;
1299 } else
1300 follow_dotdot(nd);
1301 }
1302 return 0;
1303}
1304
951361f9
AV		 1305static void terminate_walk(struct nameidata *nd)
1306{
1307 if (!(nd->flags & LOOKUP_RCU)) {
1308 path_put(&nd->path);
1309 } else {
1310 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1311 if (!(nd->flags & LOOKUP_ROOT))
1312 nd->root.mnt = NULL;
951361f9
AV		 1313 rcu_read_unlock();
1314 br_read_unlock(vfsmount_lock);
1315 }
1316}
1317
ce57dfc1
AV		 1318static inline int walk_component(struct nameidata *nd, struct path *path,
1319 struct qstr *name, int type, int follow)
1320{
1321 struct inode *inode;
1322 int err;
1323 /*
1324 * "." and ".." are special - ".." especially so because it has
1325 * to be able to know about the current root directory and
1326 * parent relationships.
1327 */
1328 if (unlikely(type != LAST_NORM))
1329 return handle_dots(nd, type);
1330 err = do_lookup(nd, name, path, &inode);
1331 if (unlikely(err)) {
1332 terminate_walk(nd);
1333 return err;
1334 }
1335 if (!inode) {
1336 path_to_nameidata(path, nd);
1337 terminate_walk(nd);
1338 return -ENOENT;
1339 }
1340 if (unlikely(inode->i_op->follow_link) && follow) {
1341 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
1342 return -ECHILD;
1343 BUG_ON(inode != path->dentry->d_inode);
1344 return 1;
1345 }
1346 path_to_nameidata(path, nd);
1347 nd->inode = inode;
1348 return 0;
1349}
1350
b356379a
AV		 1351/*
1352 * This limits recursive symlink follows to 8, while
1353 * limiting consecutive symlinks to 40.
1354 *
1355 * Without that kind of total limit, nasty chains of consecutive
1356 * symlinks can cause almost arbitrarily long lookups.
1357 */
1358static inline int nested_symlink(struct path *path, struct nameidata *nd)
1359{
1360 int res;
1361
1362 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1363 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1364 path_put_conditional(path, nd);
1365 path_put(&nd->path);
1366 return -ELOOP;
1367 }
1368
1369 nd->depth++;
1370 current->link_count++;
1371
1372 do {
1373 struct path link = *path;
1374 void *cookie;
574197e0
AV		 1375
1376 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1377 if (!res)
1378 res = walk_component(nd, path, &nd->last,
1379 nd->last_type, LOOKUP_FOLLOW);
574197e0 1380 put_link(nd, &link, cookie);
b356379a
AV		 1381 } while (res > 0);
1382
1383 current->link_count--;
1384 nd->depth--;
1385 return res;
1386}
1387
1da177e4
LT		 1388/*
1389 * Name resolution.
ea3834d9
PM		 1390 * This is the basic name resolution function, turning a pathname into
1391 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1392 *
ea3834d9
PM		 1393 * Returns 0 and nd will have valid dentry and mnt on success.
1394 * Returns error and drops reference to input namei data on failure.
1da177e4 1395 */
6de88d72 1396static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1397{
1398 struct path next;
1da177e4
LT		 1399 int err;
1400 unsigned int lookup_flags = nd->flags;
1401
1402 while (*name=='/')
1403 name++;
1404 if (!*name)
086e183a 1405 return 0;
1da177e4 1406
1da177e4
LT		 1407 /* At this point we know we have a real path component. */
1408 for(;;) {
1409 unsigned long hash;
1410 struct qstr this;
1411 unsigned int c;
fe479a58 1412 int type;
1da177e4 1413
cdce5d6b 1414 nd->flags |= LOOKUP_CONTINUE;
52094c8a
AV		 1415
1416 err = may_lookup(nd);
1da177e4
LT		 1417 if (err)
1418 break;
1419
1420 this.name = name;
1421 c = *(const unsigned char *)name;
1422
1423 hash = init_name_hash();
1424 do {
1425 name++;
1426 hash = partial_name_hash(c, hash);
1427 c = *(const unsigned char *)name;
1428 } while (c && (c != '/'));
1429 this.len = name - (const char *) this.name;
1430 this.hash = end_name_hash(hash);
1431
fe479a58
AV		 1432 type = LAST_NORM;
1433 if (this.name[0] == '.') switch (this.len) {
1434 case 2:
16c2cd71 1435 if (this.name[1] == '.') {
fe479a58 1436 type = LAST_DOTDOT;
16c2cd71
AV		 1437 nd->flags |= LOOKUP_JUMPED;
1438 }
fe479a58
AV		 1439 break;
1440 case 1:
1441 type = LAST_DOT;
1442 }
5a202bcd
AV		 1443 if (likely(type == LAST_NORM)) {
1444 struct dentry *parent = nd->path.dentry;
16c2cd71 1445 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1446 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1447 err = parent->d_op->d_hash(parent, nd->inode,
1448 &this);
1449 if (err < 0)
1450 break;
1451 }
1452 }
fe479a58 1453
1da177e4
LT		 1454 /* remove trailing slashes? */
1455 if (!c)
1456 goto last_component;
1457 while (*++name == '/');
1458 if (!*name)
b356379a 1459 goto last_component;
1da177e4 1460
ce57dfc1
AV		 1461 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1462 if (err < 0)
1463 return err;
1da177e4 1464
ce57dfc1 1465 if (err) {
b356379a 1466 err = nested_symlink(&next, nd);
1da177e4 1467 if (err)
a7472bab 1468 return err;
31e6b01f 1469 }
1da177e4 1470 err = -ENOTDIR;
31e6b01f 1471 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1472 break;
1473 continue;
1474 /* here ends the main loop */
1475
1da177e4 1476last_component:
f55eab82
TM		 1477 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1478 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
b356379a
AV		 1479 nd->last = this;
1480 nd->last_type = type;
086e183a 1481 return 0;
1da177e4 1482 }
951361f9 1483 terminate_walk(nd);
1da177e4
LT		 1484 return err;
1485}
1486
70e9b357
AV		 1487static int path_init(int dfd, const char *name, unsigned int flags,
1488 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1489{
1490 int retval = 0;
1491 int fput_needed;
1492 struct file *file;
1493
1494 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1495 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1496 nd->depth = 0;
5b6ca027
AV		 1497 if (flags & LOOKUP_ROOT) {
1498 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1499 if (*name) {
1500 if (!inode->i_op->lookup)
1501 return -ENOTDIR;
1502 retval = inode_permission(inode, MAY_EXEC);
1503 if (retval)
1504 return retval;
1505 }
5b6ca027
AV		 1506 nd->path = nd->root;
1507 nd->inode = inode;
1508 if (flags & LOOKUP_RCU) {
1509 br_read_lock(vfsmount_lock);
1510 rcu_read_lock();
1511 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1512 } else {
1513 path_get(&nd->path);
1514 }
1515 return 0;
1516 }
1517
31e6b01f 1518 nd->root.mnt = NULL;
31e6b01f
NP		 1519
1520 if (*name=='/') {
e41f7d4e
AV		 1521 if (flags & LOOKUP_RCU) {
1522 br_read_lock(vfsmount_lock);
1523 rcu_read_lock();
1524 set_root_rcu(nd);
1525 } else {
1526 set_root(nd);
1527 path_get(&nd->root);
1528 }
1529 nd->path = nd->root;
31e6b01f 1530 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1531 if (flags & LOOKUP_RCU) {
1532 struct fs_struct *fs = current->fs;
1533 unsigned seq;
31e6b01f 1534
e41f7d4e
AV		 1535 br_read_lock(vfsmount_lock);
1536 rcu_read_lock();
c28cc364 1537
e41f7d4e
AV		 1538 do {
1539 seq = read_seqcount_begin(&fs->seq);
1540 nd->path = fs->pwd;
1541 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1542 } while (read_seqcount_retry(&fs->seq, seq));
1543 } else {
1544 get_fs_pwd(current->fs, &nd->path);
1545 }
31e6b01f
NP		 1546 } else {
1547 struct dentry *dentry;
1548
1abf0c71 1549 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1550 retval = -EBADF;
1551 if (!file)
1552 goto out_fail;
1553
1554 dentry = file->f_path.dentry;
1555
f52e0c11
AV		 1556 if (*name) {
1557 retval = -ENOTDIR;
1558 if (!S_ISDIR(dentry->d_inode->i_mode))
1559 goto fput_fail;
31e6b01f 1560
f52e0c11
AV		 1561 retval = file_permission(file, MAY_EXEC);
1562 if (retval)
1563 goto fput_fail;
1564 }
31e6b01f
NP		 1565
1566 nd->path = file->f_path;
e41f7d4e
AV		 1567 if (flags & LOOKUP_RCU) {
1568 if (fput_needed)
70e9b357 1569 *fp = file;
e41f7d4e
AV		 1570 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1571 br_read_lock(vfsmount_lock);
1572 rcu_read_lock();
1573 } else {
1574 path_get(&file->f_path);
1575 fput_light(file, fput_needed);
1576 }
31e6b01f 1577 }
31e6b01f 1578
31e6b01f 1579 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1580 return 0;
2dfdd266 1581
9b4a9b14
AV		 1582fput_fail:
1583 fput_light(file, fput_needed);
1584out_fail:
1585 return retval;
1586}
1587
bd92d7fe
AV		 1588static inline int lookup_last(struct nameidata *nd, struct path *path)
1589{
1590 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1591 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1592
1593 nd->flags &= ~LOOKUP_PARENT;
1594 return walk_component(nd, path, &nd->last, nd->last_type,
1595 nd->flags & LOOKUP_FOLLOW);
1596}
1597
9b4a9b14 1598/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1599static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1600 unsigned int flags, struct nameidata *nd)
1601{
70e9b357 1602 struct file *base = NULL;
bd92d7fe
AV		 1603 struct path path;
1604 int err;
31e6b01f
NP		 1605
1606 /*
1607 * Path walking is largely split up into 2 different synchronisation
1608 * schemes, rcu-walk and ref-walk (explained in
1609 * Documentation/filesystems/path-lookup.txt). These share much of the
1610 * path walk code, but some things particularly setup, cleanup, and
1611 * following mounts are sufficiently divergent that functions are
1612 * duplicated. Typically there is a function foo(), and its RCU
1613 * analogue, foo_rcu().
1614 *
1615 * -ECHILD is the error number of choice (just to avoid clashes) that
1616 * is returned if some aspect of an rcu-walk fails. Such an error must
1617 * be handled by restarting a traditional ref-walk (which will always
1618 * be able to complete).
1619 */
bd92d7fe 1620 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1621
bd92d7fe
AV		 1622 if (unlikely(err))
1623 return err;
ee0827cd
AV		 1624
1625 current->total_link_count = 0;
bd92d7fe
AV		 1626 err = link_path_walk(name, nd);
1627
1628 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1629 err = lookup_last(nd, &path);
1630 while (err > 0) {
1631 void *cookie;
1632 struct path link = path;
bd92d7fe 1633 nd->flags |= LOOKUP_PARENT;
574197e0 1634 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1635 if (!err)
1636 err = lookup_last(nd, &path);
574197e0 1637 put_link(nd, &link, cookie);
bd92d7fe
AV		 1638 }
1639 }
ee0827cd
AV		 1640
1641 if (nd->flags & LOOKUP_RCU) {
4455ca62 1642 /* went all way through without dropping RCU */
bd92d7fe 1643 BUG_ON(err);
4455ca62 1644 if (nameidata_drop_rcu_last(nd))
bd92d7fe 1645 err = -ECHILD;
ee0827cd
AV		 1646 }
1647
bd92d7fe
AV		 1648 if (!err)
1649 err = handle_reval_path(nd);
1650
1651 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1652 if (!nd->inode->i_op->lookup) {
1653 path_put(&nd->path);
1654 return -ENOTDIR;
1655 }
1656 }
16c2cd71 1657
70e9b357
AV		 1658 if (base)
1659 fput(base);
ee0827cd 1660
5b6ca027 1661 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1662 path_put(&nd->root);
1663 nd->root.mnt = NULL;
1664 }
bd92d7fe 1665 return err;
ee0827cd 1666}
31e6b01f 1667
ee0827cd
AV		 1668static int do_path_lookup(int dfd, const char *name,
1669 unsigned int flags, struct nameidata *nd)
1670{
1671 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1672 if (unlikely(retval == -ECHILD))
1673 retval = path_lookupat(dfd, name, flags, nd);
1674 if (unlikely(retval == -ESTALE))
1675 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1676
1677 if (likely(!retval)) {
1678 if (unlikely(!audit_dummy_context())) {
1679 if (nd->path.dentry && nd->inode)
1680 audit_inode(name, nd->path.dentry);
1681 }
1682 }
170aa3d0 1683 return retval;
1da177e4
LT		 1684}
1685
c9c6cac0 1686int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1687{
c9c6cac0 1688 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1689}
1690
d1811465
AV		 1691int kern_path(const char *name, unsigned int flags, struct path *path)
1692{
1693 struct nameidata nd;
1694 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1695 if (!res)
1696 *path = nd.path;
1697 return res;
1698}
1699
16f18200
JJS		 1700/**
1701 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1702 * @dentry: pointer to dentry of the base directory
1703 * @mnt: pointer to vfs mount of the base directory
1704 * @name: pointer to file name
1705 * @flags: lookup flags
1706 * @nd: pointer to nameidata
1707 */
1708int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1709 const char *name, unsigned int flags,
1710 struct nameidata *nd)
1711{
5b6ca027
AV		 1712 nd->root.dentry = dentry;
1713 nd->root.mnt = mnt;
1714 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1715 return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
16f18200
JJS		 1716}
1717
eead1911
CH		 1718static struct dentry *__lookup_hash(struct qstr *name,
1719 struct dentry *base, struct nameidata *nd)
1da177e4 1720{
81fca444 1721 struct inode *inode = base->d_inode;
057f6c01 1722 struct dentry *dentry;
1da177e4
LT		 1723 int err;
1724
b74c79e9 1725 err = exec_permission(inode, 0);
81fca444
CH		 1726 if (err)
1727 return ERR_PTR(err);
1da177e4 1728
b04f784e
NP		 1729 /*
1730 * Don't bother with __d_lookup: callers are for creat as
1731 * well as unlink, so a lot of the time it would cost
1732 * a double lookup.
6e6b1bd1 1733 */
b04f784e 1734 dentry = d_lookup(base, name);
6e6b1bd1 1735
fb045adb 1736 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1737 dentry = do_revalidate(dentry, nd);
1738
baa03890
NP		 1739 if (!dentry)
1740 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1741
1da177e4
LT		 1742 return dentry;
1743}
1744
057f6c01
JM		 1745/*
1746 * Restricted form of lookup. Doesn't follow links, single-component only,
1747 * needs parent already locked. Doesn't follow mounts.
1748 * SMP-safe.
1749 */
eead1911 1750static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1751{
4ac91378 1752 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1753}
1754
eead1911 1755/**
a6b91919 1756 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1757 * @name: pathname component to lookup
1758 * @base: base directory to lookup from
1759 * @len: maximum length @len should be interpreted to
1760 *
a6b91919
RD		 1761 * Note that this routine is purely a helper for filesystem usage and should
1762 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1763 * nameidata argument is passed to the filesystem methods and a filesystem
1764 * using this helper needs to be prepared for that.
1765 */
057f6c01
JM		 1766struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1767{
057f6c01 1768 struct qstr this;
6a96ba54
AV		 1769 unsigned long hash;
1770 unsigned int c;
057f6c01 1771
2f9092e1
DW		 1772 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1773
6a96ba54
AV		 1774 this.name = name;
1775 this.len = len;
1776 if (!len)
1777 return ERR_PTR(-EACCES);
1778
1779 hash = init_name_hash();
1780 while (len--) {
1781 c = *(const unsigned char *)name++;
1782 if (c == '/' || c == '\0')
1783 return ERR_PTR(-EACCES);
1784 hash = partial_name_hash(c, hash);
1785 }
1786 this.hash = end_name_hash(hash);
5a202bcd
AV		 1787 /*
1788 * See if the low-level filesystem might want
1789 * to use its own hash..
1790 */
1791 if (base->d_flags & DCACHE_OP_HASH) {
1792 int err = base->d_op->d_hash(base, base->d_inode, &this);
1793 if (err < 0)
1794 return ERR_PTR(err);
1795 }
eead1911 1796
49705b77 1797 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1798}
1799
2d8f3038
AV		 1800int user_path_at(int dfd, const char __user *name, unsigned flags,
1801 struct path *path)
1da177e4 1802{
2d8f3038 1803 struct nameidata nd;
f52e0c11 1804 char *tmp = getname_flags(name, flags);
1da177e4 1805 int err = PTR_ERR(tmp);
1da177e4 1806 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1807
1808 BUG_ON(flags & LOOKUP_PARENT);
1809
1810 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1811 putname(tmp);
2d8f3038
AV		 1812 if (!err)
1813 *path = nd.path;
1da177e4
LT		 1814 }
1815 return err;
1816}
1817
2ad94ae6
AV		 1818static int user_path_parent(int dfd, const char __user *path,
1819 struct nameidata *nd, char **name)
1820{
1821 char *s = getname(path);
1822 int error;
1823
1824 if (IS_ERR(s))
1825 return PTR_ERR(s);
1826
1827 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1828 if (error)
1829 putname(s);
1830 else
1831 *name = s;
1832
1833 return error;
1834}
1835
1da177e4
LT		 1836/*
1837 * It's inline, so penalty for filesystems that don't use sticky bit is
1838 * minimal.
1839 */
1840static inline int check_sticky(struct inode *dir, struct inode *inode)
1841{
da9592ed
DH		 1842 uid_t fsuid = current_fsuid();
1843
1da177e4
LT		 1844 if (!(dir->i_mode & S_ISVTX))
1845 return 0;
da9592ed 1846 if (inode->i_uid == fsuid)
1da177e4 1847 return 0;
da9592ed 1848 if (dir->i_uid == fsuid)
1da177e4
LT		 1849 return 0;
1850 return !capable(CAP_FOWNER);
1851}
1852
1853/*
1854 * Check whether we can remove a link victim from directory dir, check
1855 * whether the type of victim is right.
1856 * 1. We can't do it if dir is read-only (done in permission())
1857 * 2. We should have write and exec permissions on dir
1858 * 3. We can't remove anything from append-only dir
1859 * 4. We can't do anything with immutable dir (done in permission())
1860 * 5. If the sticky bit on dir is set we should either
1861 * a. be owner of dir, or
1862 * b. be owner of victim, or
1863 * c. have CAP_FOWNER capability
1864 * 6. If the victim is append-only or immutable we can't do antyhing with
1865 * links pointing to it.
1866 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1867 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1868 * 9. We can't remove a root or mountpoint.
1869 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1870 * nfs_async_unlink().
1871 */
858119e1 1872static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1873{
1874 int error;
1875
1876 if (!victim->d_inode)
1877 return -ENOENT;
1878
1879 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1880 audit_inode_child(victim, dir);
1da177e4 1881
f419a2e3 1882 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1883 if (error)
1884 return error;
1885 if (IS_APPEND(dir))
1886 return -EPERM;
1887 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1888 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1889 return -EPERM;
1890 if (isdir) {
1891 if (!S_ISDIR(victim->d_inode->i_mode))
1892 return -ENOTDIR;
1893 if (IS_ROOT(victim))
1894 return -EBUSY;
1895 } else if (S_ISDIR(victim->d_inode->i_mode))
1896 return -EISDIR;
1897 if (IS_DEADDIR(dir))
1898 return -ENOENT;
1899 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1900 return -EBUSY;
1901 return 0;
1902}
1903
1904/* Check whether we can create an object with dentry child in directory
1905 * dir.
1906 * 1. We can't do it if child already exists (open has special treatment for
1907 * this case, but since we are inlined it's OK)
1908 * 2. We can't do it if dir is read-only (done in permission())
1909 * 3. We should have write and exec permissions on dir
1910 * 4. We can't do it if dir is immutable (done in permission())
1911 */
a95164d9 1912static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1913{
1914 if (child->d_inode)
1915 return -EEXIST;
1916 if (IS_DEADDIR(dir))
1917 return -ENOENT;
f419a2e3 1918 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1919}
1920
1da177e4
LT		 1921/*
1922 * p1 and p2 should be directories on the same fs.
1923 */
1924struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1925{
1926 struct dentry *p;
1927
1928 if (p1 == p2) {
f2eace23 1929 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1930 return NULL;
1931 }
1932
a11f3a05 1933 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1934
e2761a11
OH		 1935 p = d_ancestor(p2, p1);
1936 if (p) {
1937 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1938 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1939 return p;
1da177e4
LT		 1940 }
1941
e2761a11
OH		 1942 p = d_ancestor(p1, p2);
1943 if (p) {
1944 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1945 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1946 return p;
1da177e4
LT		 1947 }
1948
f2eace23
IM		 1949 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1950 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1951 return NULL;
1952}
1953
1954void unlock_rename(struct dentry *p1, struct dentry *p2)
1955{
1b1dcc1b 1956 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1957 if (p1 != p2) {
1b1dcc1b 1958 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1959 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1960 }
1961}
1962
1963int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1964 struct nameidata *nd)
1965{
a95164d9 1966 int error = may_create(dir, dentry);
1da177e4
LT		 1967
1968 if (error)
1969 return error;
1970
acfa4380 1971 if (!dir->i_op->create)
1da177e4
LT		 1972 return -EACCES; /* shouldn't it be ENOSYS? */
1973 mode &= S_IALLUGO;
1974 mode |= S_IFREG;
1975 error = security_inode_create(dir, dentry, mode);
1976 if (error)
1977 return error;
1da177e4 1978 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1979 if (!error)
f38aa942 1980 fsnotify_create(dir, dentry);
1da177e4
LT		 1981 return error;
1982}
1983
73d049a4 1984static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1985{
3fb64190 1986 struct dentry *dentry = path->dentry;
1da177e4
LT		 1987 struct inode *inode = dentry->d_inode;
1988 int error;
1989
bcda7652
AV		 1990 /* O_PATH? */
1991 if (!acc_mode)
1992 return 0;
1993
1da177e4
LT		 1994 if (!inode)
1995 return -ENOENT;
1996
c8fe8f30
CH		 1997 switch (inode->i_mode & S_IFMT) {
1998 case S_IFLNK:
1da177e4 1999 return -ELOOP;
c8fe8f30
CH		 2000 case S_IFDIR:
2001 if (acc_mode & MAY_WRITE)
2002 return -EISDIR;
2003 break;
2004 case S_IFBLK:
2005 case S_IFCHR:
3fb64190 2006 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2007 return -EACCES;
c8fe8f30
CH		 2008 /*FALLTHRU*/
2009 case S_IFIFO:
2010 case S_IFSOCK:
1da177e4 2011 flag &= ~O_TRUNC;
c8fe8f30 2012 break;
4a3fd211 2013 }
b41572e9 2014
3fb64190 2015 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2016 if (error)
2017 return error;
6146f0d5 2018
1da177e4
LT		 2019 /*
2020 * An append-only file must be opened in append mode for writing.
2021 */
2022 if (IS_APPEND(inode)) {
8737c930 2023 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2024 return -EPERM;
1da177e4 2025 if (flag & O_TRUNC)
7715b521 2026 return -EPERM;
1da177e4
LT		 2027 }
2028
2029 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 2030 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2031 return -EPERM;
1da177e4
LT		 2032
2033 /*
2034 * Ensure there are no outstanding leases on the file.
2035 */
b65a9cfc 2036 return break_lease(inode, flag);
7715b521 2037}
1da177e4 2038
e1181ee6 2039static int handle_truncate(struct file *filp)
7715b521 2040{
e1181ee6 2041 struct path *path = &filp->f_path;
7715b521
AV		 2042 struct inode *inode = path->dentry->d_inode;
2043 int error = get_write_access(inode);
2044 if (error)
2045 return error;
2046 /*
2047 * Refuse to truncate files with mandatory locks held on them.
2048 */
2049 error = locks_verify_locked(inode);
2050 if (!error)
ea0d3ab2 2051 error = security_path_truncate(path);
7715b521
AV		 2052 if (!error) {
2053 error = do_truncate(path->dentry, 0,
2054 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2055 filp);
7715b521
AV		 2056 }
2057 put_write_access(inode);
acd0c935 2058 return error;
1da177e4
LT		 2059}
2060
d57999e1
DH		 2061/*
2062 * Note that while the flag value (low two bits) for sys_open means:
2063 * 00 - read-only
2064 * 01 - write-only
2065 * 10 - read-write
2066 * 11 - special
2067 * it is changed into
2068 * 00 - no permissions needed
2069 * 01 - read-permission
2070 * 10 - write-permission
2071 * 11 - read-write
2072 * for the internal routines (ie open_namei()/follow_link() etc)
2073 * This is more logical, and also allows the 00 "no perm needed"
2074 * to be used for symlinks (where the permissions are checked
2075 * later).
2076 *
2077*/
2078static inline int open_to_namei_flags(int flag)
2079{
2080 if ((flag+1) & O_ACCMODE)
2081 flag++;
2082 return flag;
2083}
2084
31e6b01f 2085/*
fe2d35ff 2086 * Handle the last step of open()
31e6b01f 2087 */
fb1cc555 2088static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2089 const struct open_flags *op, const char *pathname)
fb1cc555 2090{
a1e28038 2091 struct dentry *dir = nd->path.dentry;
6c0d46c4 2092 struct dentry *dentry;
ca344a89 2093 int open_flag = op->open_flag;
6c0d46c4 2094 int will_truncate = open_flag & O_TRUNC;
ca344a89 2095 int want_write = 0;
bcda7652 2096 int acc_mode = op->acc_mode;
fb1cc555 2097 struct file *filp;
16c2cd71 2098 int error;
1f36f774 2099
c3e380b0
AV		 2100 nd->flags &= ~LOOKUP_PARENT;
2101 nd->flags |= op->intent;
2102
1f36f774
AV		 2103 switch (nd->last_type) {
2104 case LAST_DOTDOT:
176306f5 2105 case LAST_DOT:
fe2d35ff
AV		 2106 error = handle_dots(nd, nd->last_type);
2107 if (error)
2108 return ERR_PTR(error);
1f36f774 2109 /* fallthrough */
1f36f774 2110 case LAST_ROOT:
fe2d35ff
AV		 2111 if (nd->flags & LOOKUP_RCU) {
2112 if (nameidata_drop_rcu_last(nd))
2113 return ERR_PTR(-ECHILD);
2114 }
16c2cd71
AV		 2115 error = handle_reval_path(nd);
2116 if (error)
2117 goto exit;
fe2d35ff 2118 audit_inode(pathname, nd->path.dentry);
ca344a89 2119 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2120 error = -EISDIR;
2121 goto exit;
2122 }
2123 goto ok;
1f36f774 2124 case LAST_BIND:
fe2d35ff 2125 /* can't be RCU mode here */
16c2cd71
AV		 2126 error = handle_reval_path(nd);
2127 if (error)
2128 goto exit;
1f36f774 2129 audit_inode(pathname, dir);
67ee3ad2 2130 goto ok;
1f36f774 2131 }
67ee3ad2 2132
ca344a89 2133 if (!(open_flag & O_CREAT)) {
bcda7652 2134 int symlink_ok = 0;
fe2d35ff
AV		 2135 if (nd->last.name[nd->last.len])
2136 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2137 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2138 symlink_ok = 1;
fe2d35ff 2139 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2140 error = walk_component(nd, path, &nd->last, LAST_NORM,
2141 !symlink_ok);
2142 if (error < 0)
fe2d35ff 2143 return ERR_PTR(error);
ce57dfc1 2144 if (error) /* symlink */
fe2d35ff 2145 return NULL;
fe2d35ff
AV		 2146 /* sayonara */
2147 if (nd->flags & LOOKUP_RCU) {
2148 if (nameidata_drop_rcu_last(nd))
2149 return ERR_PTR(-ECHILD);
2150 }
2151
2152 error = -ENOTDIR;
2153 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2154 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2155 goto exit;
2156 }
2157 audit_inode(pathname, nd->path.dentry);
2158 goto ok;
2159 }
2160
2161 /* create side of things */
2162
2163 if (nd->flags & LOOKUP_RCU) {
2164 if (nameidata_drop_rcu_last(nd))
2165 return ERR_PTR(-ECHILD);
2166 }
2167
2168 audit_inode(pathname, dir);
16c2cd71 2169 error = -EISDIR;
1f36f774 2170 /* trailing slashes? */
31e6b01f
NP		 2171 if (nd->last.name[nd->last.len])
2172 goto exit;
a2c36b45 2173
a1e28038
AV		 2174 mutex_lock(&dir->d_inode->i_mutex);
2175
6c0d46c4
AV		 2176 dentry = lookup_hash(nd);
2177 error = PTR_ERR(dentry);
2178 if (IS_ERR(dentry)) {
fb1cc555
AV		 2179 mutex_unlock(&dir->d_inode->i_mutex);
2180 goto exit;
2181 }
2182
6c0d46c4
AV		 2183 path->dentry = dentry;
2184 path->mnt = nd->path.mnt;
2185
fb1cc555 2186 /* Negative dentry, just create the file */
6c0d46c4
AV		 2187 if (!dentry->d_inode) {
2188 int mode = op->mode;
2189 if (!IS_POSIXACL(dir->d_inode))
2190 mode &= ~current_umask();
fb1cc555
AV		 2191 /*
2192 * This write is needed to ensure that a
6c0d46c4 2193 * rw->ro transition does not occur between
fb1cc555
AV		 2194 * the time when the file is created and when
2195 * a permanent write count is taken through
2196 * the 'struct file' in nameidata_to_filp().
2197 */
2198 error = mnt_want_write(nd->path.mnt);
2199 if (error)
2200 goto exit_mutex_unlock;
ca344a89 2201 want_write = 1;
9b44f1b3 2202 /* Don't check for write permission, don't truncate */
ca344a89 2203 open_flag &= ~O_TRUNC;
6c0d46c4 2204 will_truncate = 0;
bcda7652 2205 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2206 error = security_path_mknod(&nd->path, dentry, mode, 0);
2207 if (error)
2208 goto exit_mutex_unlock;
2209 error = vfs_create(dir->d_inode, dentry, mode, nd);
2210 if (error)
2211 goto exit_mutex_unlock;
2212 mutex_unlock(&dir->d_inode->i_mutex);
2213 dput(nd->path.dentry);
2214 nd->path.dentry = dentry;
ca344a89 2215 goto common;
fb1cc555
AV		 2216 }
2217
2218 /*
2219 * It already exists.
2220 */
2221 mutex_unlock(&dir->d_inode->i_mutex);
2222 audit_inode(pathname, path->dentry);
2223
2224 error = -EEXIST;
ca344a89 2225 if (open_flag & O_EXCL)
fb1cc555
AV		 2226 goto exit_dput;
2227
9875cf80
DH		 2228 error = follow_managed(path, nd->flags);
2229 if (error < 0)
2230 goto exit_dput;
fb1cc555
AV		 2231
2232 error = -ENOENT;
2233 if (!path->dentry->d_inode)
2234 goto exit_dput;
9e67f361
AV		 2235
2236 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2237 return NULL;
fb1cc555
AV		 2238
2239 path_to_nameidata(path, nd);
31e6b01f 2240 nd->inode = path->dentry->d_inode;
fb1cc555 2241 error = -EISDIR;
31e6b01f 2242 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2243 goto exit;
67ee3ad2 2244ok:
6c0d46c4
AV		 2245 if (!S_ISREG(nd->inode->i_mode))
2246 will_truncate = 0;
2247
0f9d1a10
AV		 2248 if (will_truncate) {
2249 error = mnt_want_write(nd->path.mnt);
2250 if (error)
2251 goto exit;
ca344a89 2252 want_write = 1;
0f9d1a10 2253 }
ca344a89 2254common:
bcda7652 2255 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2256 if (error)
0f9d1a10 2257 goto exit;
0f9d1a10
AV		 2258 filp = nameidata_to_filp(nd);
2259 if (!IS_ERR(filp)) {
2260 error = ima_file_check(filp, op->acc_mode);
2261 if (error) {
2262 fput(filp);
2263 filp = ERR_PTR(error);
2264 }
2265 }
2266 if (!IS_ERR(filp)) {
2267 if (will_truncate) {
2268 error = handle_truncate(filp);
2269 if (error) {
2270 fput(filp);
2271 filp = ERR_PTR(error);
2272 }
2273 }
2274 }
ca344a89
AV		 2275out:
2276 if (want_write)
0f9d1a10
AV		 2277 mnt_drop_write(nd->path.mnt);
2278 path_put(&nd->path);
fb1cc555
AV		 2279 return filp;
2280
2281exit_mutex_unlock:
2282 mutex_unlock(&dir->d_inode->i_mutex);
2283exit_dput:
2284 path_put_conditional(path, nd);
2285exit:
ca344a89
AV		 2286 filp = ERR_PTR(error);
2287 goto out;
fb1cc555
AV		 2288}
2289
13aab428 2290static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2291 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2292{
fe2d35ff 2293 struct file *base = NULL;
4a3fd211 2294 struct file *filp;
9850c056 2295 struct path path;
13aab428 2296 int error;
31e6b01f
NP		 2297
2298 filp = get_empty_filp();
2299 if (!filp)
2300 return ERR_PTR(-ENFILE);
2301
47c805dc 2302 filp->f_flags = op->open_flag;
73d049a4
AV		 2303 nd->intent.open.file = filp;
2304 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2305 nd->intent.open.create_mode = op->mode;
31e6b01f 2306
73d049a4 2307 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2308 if (unlikely(error))
13aab428 2309 goto out_filp;
31e6b01f 2310
fe2d35ff 2311 current->total_link_count = 0;
73d049a4 2312 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2313 if (unlikely(error))
2314 goto out_filp;
1da177e4 2315
73d049a4 2316 filp = do_last(nd, &path, op, pathname);
806b681c 2317 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2318 struct path link = path;
def4af30 2319 void *cookie;
574197e0 2320 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2321 path_put_conditional(&path, nd);
2322 path_put(&nd->path);
40b39136
AV		 2323 filp = ERR_PTR(-ELOOP);
2324 break;
2325 }
73d049a4
AV		 2326 nd->flags |= LOOKUP_PARENT;
2327 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2328 error = follow_link(&link, nd, &cookie);
c3e380b0 2329 if (unlikely(error))
f1afe9ef 2330 filp = ERR_PTR(error);
c3e380b0 2331 else
73d049a4 2332 filp = do_last(nd, &path, op, pathname);
574197e0 2333 put_link(nd, &link, cookie);
806b681c 2334 }
10fa8e62 2335out:
73d049a4
AV		 2336 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2337 path_put(&nd->root);
fe2d35ff
AV		 2338 if (base)
2339 fput(base);
73d049a4 2340 release_open_intent(nd);
10fa8e62 2341 return filp;
1da177e4 2342
31e6b01f 2343out_filp:
806b681c 2344 filp = ERR_PTR(error);
10fa8e62 2345 goto out;
1da177e4
LT		 2346}
2347
13aab428
AV		 2348struct file *do_filp_open(int dfd, const char *pathname,
2349 const struct open_flags *op, int flags)
2350{
73d049a4 2351 struct nameidata nd;
13aab428
AV		 2352 struct file *filp;
2353
73d049a4 2354 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2355 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2356 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2357 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2358 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2359 return filp;
2360}
2361
73d049a4
AV		 2362struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2363 const char *name, const struct open_flags *op, int flags)
2364{
2365 struct nameidata nd;
2366 struct file *file;
2367
2368 nd.root.mnt = mnt;
2369 nd.root.dentry = dentry;
2370
2371 flags |= LOOKUP_ROOT;
2372
bcda7652 2373 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2374 return ERR_PTR(-ELOOP);
2375
2376 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2377 if (unlikely(file == ERR_PTR(-ECHILD)))
2378 file = path_openat(-1, name, &nd, op, flags);
2379 if (unlikely(file == ERR_PTR(-ESTALE)))
2380 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2381 return file;
2382}
2383
1da177e4
LT		 2384/**
2385 * lookup_create - lookup a dentry, creating it if it doesn't exist
2386 * @nd: nameidata info
2387 * @is_dir: directory flag
2388 *
2389 * Simple function to lookup and return a dentry and create it
2390 * if it doesn't exist. Is SMP-safe.
c663e5d8 2391 *
4ac91378 2392 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2393 */
2394struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2395{
c663e5d8 2396 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2397
4ac91378 2398 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2399 /*
2400 * Yucky last component or no last component at all?
2401 * (foo/., foo/.., /////)
2402 */
1da177e4
LT		 2403 if (nd->last_type != LAST_NORM)
2404 goto fail;
2405 nd->flags &= ~LOOKUP_PARENT;
3516586a 2406 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2407 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2408
2409 /*
2410 * Do the final lookup.
2411 */
49705b77 2412 dentry = lookup_hash(nd);
1da177e4
LT		 2413 if (IS_ERR(dentry))
2414 goto fail;
c663e5d8 2415
e9baf6e5
AV		 2416 if (dentry->d_inode)
2417 goto eexist;
c663e5d8
CH		 2418 /*
2419 * Special case - lookup gave negative, but... we had foo/bar/
2420 * From the vfs_mknod() POV we just have a negative dentry -
2421 * all is fine. Let's be bastards - you had / on the end, you've
2422 * been asking for (non-existent) directory. -ENOENT for you.
2423 */
e9baf6e5
AV		 2424 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2425 dput(dentry);
2426 dentry = ERR_PTR(-ENOENT);
2427 }
1da177e4 2428 return dentry;
e9baf6e5 2429eexist:
1da177e4 2430 dput(dentry);
e9baf6e5 2431 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2432fail:
2433 return dentry;
2434}
f81a0bff 2435EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2436
2437int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2438{
a95164d9 2439 int error = may_create(dir, dentry);
1da177e4
LT		 2440
2441 if (error)
2442 return error;
2443
2444 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2445 return -EPERM;
2446
acfa4380 2447 if (!dir->i_op->mknod)
1da177e4
LT		 2448 return -EPERM;
2449
08ce5f16
SH		 2450 error = devcgroup_inode_mknod(mode, dev);
2451 if (error)
2452 return error;
2453
1da177e4
LT		 2454 error = security_inode_mknod(dir, dentry, mode, dev);
2455 if (error)
2456 return error;
2457
1da177e4 2458 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2459 if (!error)
f38aa942 2460 fsnotify_create(dir, dentry);
1da177e4
LT		 2461 return error;
2462}
2463
463c3197
DH		 2464static int may_mknod(mode_t mode)
2465{
2466 switch (mode & S_IFMT) {
2467 case S_IFREG:
2468 case S_IFCHR:
2469 case S_IFBLK:
2470 case S_IFIFO:
2471 case S_IFSOCK:
2472 case 0: /* zero mode translates to S_IFREG */
2473 return 0;
2474 case S_IFDIR:
2475 return -EPERM;
2476 default:
2477 return -EINVAL;
2478 }
2479}
2480
2e4d0924
HC		 2481SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2482 unsigned, dev)
1da177e4 2483{
2ad94ae6
AV		 2484 int error;
2485 char *tmp;
2486 struct dentry *dentry;
1da177e4
LT		 2487 struct nameidata nd;
2488
2489 if (S_ISDIR(mode))
2490 return -EPERM;
1da177e4 2491
2ad94ae6 2492 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2493 if (error)
2ad94ae6
AV		 2494 return error;
2495
1da177e4 2496 dentry = lookup_create(&nd, 0);
463c3197
DH		 2497 if (IS_ERR(dentry)) {
2498 error = PTR_ERR(dentry);
2499 goto out_unlock;
2500 }
4ac91378 2501 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2502 mode &= ~current_umask();
463c3197
DH		 2503 error = may_mknod(mode);
2504 if (error)
2505 goto out_dput;
2506 error = mnt_want_write(nd.path.mnt);
2507 if (error)
2508 goto out_dput;
be6d3e56
KT		 2509 error = security_path_mknod(&nd.path, dentry, mode, dev);
2510 if (error)
2511 goto out_drop_write;
463c3197 2512 switch (mode & S_IFMT) {
1da177e4 2513 case 0: case S_IFREG:
4ac91378 2514 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2515 break;
2516 case S_IFCHR: case S_IFBLK:
4ac91378 2517 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2518 new_decode_dev(dev));
2519 break;
2520 case S_IFIFO: case S_IFSOCK:
4ac91378 2521 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2522 break;
1da177e4 2523 }
be6d3e56 2524out_drop_write:
463c3197
DH		 2525 mnt_drop_write(nd.path.mnt);
2526out_dput:
2527 dput(dentry);
2528out_unlock:
4ac91378 2529 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2530 path_put(&nd.path);
1da177e4
LT		 2531 putname(tmp);
2532
2533 return error;
2534}
2535
3480b257 2536SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2537{
2538 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2539}
2540
1da177e4
LT		 2541int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2542{
a95164d9 2543 int error = may_create(dir, dentry);
1da177e4
LT		 2544
2545 if (error)
2546 return error;
2547
acfa4380 2548 if (!dir->i_op->mkdir)
1da177e4
LT		 2549 return -EPERM;
2550
2551 mode &= (S_IRWXUGO|S_ISVTX);
2552 error = security_inode_mkdir(dir, dentry, mode);
2553 if (error)
2554 return error;
2555
1da177e4 2556 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2557 if (!error)
f38aa942 2558 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2559 return error;
2560}
2561
2e4d0924 2562SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2563{
2564 int error = 0;
2565 char * tmp;
6902d925
DH		 2566 struct dentry *dentry;
2567 struct nameidata nd;
1da177e4 2568
2ad94ae6
AV		 2569 error = user_path_parent(dfd, pathname, &nd, &tmp);
2570 if (error)
6902d925 2571 goto out_err;
1da177e4 2572
6902d925
DH		 2573 dentry = lookup_create(&nd, 1);
2574 error = PTR_ERR(dentry);
2575 if (IS_ERR(dentry))
2576 goto out_unlock;
1da177e4 2577
4ac91378 2578 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2579 mode &= ~current_umask();
463c3197
DH		 2580 error = mnt_want_write(nd.path.mnt);
2581 if (error)
2582 goto out_dput;
be6d3e56
KT		 2583 error = security_path_mkdir(&nd.path, dentry, mode);
2584 if (error)
2585 goto out_drop_write;
4ac91378 2586 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2587out_drop_write:
463c3197
DH		 2588 mnt_drop_write(nd.path.mnt);
2589out_dput:
6902d925
DH		 2590 dput(dentry);
2591out_unlock:
4ac91378 2592 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2593 path_put(&nd.path);
6902d925
DH		 2594 putname(tmp);
2595out_err:
1da177e4
LT		 2596 return error;
2597}
2598
3cdad428 2599SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2600{
2601 return sys_mkdirat(AT_FDCWD, pathname, mode);
2602}
2603
1da177e4
LT		 2604/*
2605 * We try to drop the dentry early: we should have
2606 * a usage count of 2 if we're the only user of this
2607 * dentry, and if that is true (possibly after pruning
2608 * the dcache), then we drop the dentry now.
2609 *
2610 * A low-level filesystem can, if it choses, legally
2611 * do a
2612 *
2613 * if (!d_unhashed(dentry))
2614 * return -EBUSY;
2615 *
2616 * if it cannot handle the case of removing a directory
2617 * that is still in use by something else..
2618 */
2619void dentry_unhash(struct dentry *dentry)
2620{
2621 dget(dentry);
dc168427 2622 shrink_dcache_parent(dentry);
1da177e4 2623 spin_lock(&dentry->d_lock);
b7ab39f6 2624 if (dentry->d_count == 2)
1da177e4
LT		 2625 __d_drop(dentry);
2626 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2627}
2628
2629int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2630{
2631 int error = may_delete(dir, dentry, 1);
2632
2633 if (error)
2634 return error;
2635
acfa4380 2636 if (!dir->i_op->rmdir)
1da177e4
LT		 2637 return -EPERM;
2638
1b1dcc1b 2639 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2640 dentry_unhash(dentry);
2641 if (d_mountpoint(dentry))
2642 error = -EBUSY;
2643 else {
2644 error = security_inode_rmdir(dir, dentry);
2645 if (!error) {
2646 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2647 if (!error) {
1da177e4 2648 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2649 dont_mount(dentry);
2650 }
1da177e4
LT		 2651 }
2652 }
1b1dcc1b 2653 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2654 if (!error) {
1da177e4
LT		 2655 d_delete(dentry);
2656 }
2657 dput(dentry);
2658
2659 return error;
2660}
2661
5590ff0d 2662static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2663{
2664 int error = 0;
2665 char * name;
2666 struct dentry *dentry;
2667 struct nameidata nd;
2668
2ad94ae6 2669 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2670 if (error)
2ad94ae6 2671 return error;
1da177e4
LT		 2672
2673 switch(nd.last_type) {
0612d9fb
OH		 2674 case LAST_DOTDOT:
2675 error = -ENOTEMPTY;
2676 goto exit1;
2677 case LAST_DOT:
2678 error = -EINVAL;
2679 goto exit1;
2680 case LAST_ROOT:
2681 error = -EBUSY;
2682 goto exit1;
1da177e4 2683 }
0612d9fb
OH		 2684
2685 nd.flags &= ~LOOKUP_PARENT;
2686
4ac91378 2687 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2688 dentry = lookup_hash(&nd);
1da177e4 2689 error = PTR_ERR(dentry);
6902d925
DH		 2690 if (IS_ERR(dentry))
2691 goto exit2;
0622753b
DH		 2692 error = mnt_want_write(nd.path.mnt);
2693 if (error)
2694 goto exit3;
be6d3e56
KT		 2695 error = security_path_rmdir(&nd.path, dentry);
2696 if (error)
2697 goto exit4;
4ac91378 2698 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2699exit4:
0622753b
DH		 2700 mnt_drop_write(nd.path.mnt);
2701exit3:
6902d925
DH		 2702 dput(dentry);
2703exit2:
4ac91378 2704 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2705exit1:
1d957f9b 2706 path_put(&nd.path);
1da177e4
LT		 2707 putname(name);
2708 return error;
2709}
2710
3cdad428 2711SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2712{
2713 return do_rmdir(AT_FDCWD, pathname);
2714}
2715
1da177e4
LT		 2716int vfs_unlink(struct inode *dir, struct dentry *dentry)
2717{
2718 int error = may_delete(dir, dentry, 0);
2719
2720 if (error)
2721 return error;
2722
acfa4380 2723 if (!dir->i_op->unlink)
1da177e4
LT		 2724 return -EPERM;
2725
1b1dcc1b 2726 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2727 if (d_mountpoint(dentry))
2728 error = -EBUSY;
2729 else {
2730 error = security_inode_unlink(dir, dentry);
bec1052e 2731 if (!error) {
1da177e4 2732 error = dir->i_op->unlink(dir, dentry);
bec1052e 2733 if (!error)
d83c49f3 2734 dont_mount(dentry);
bec1052e 2735 }
1da177e4 2736 }
1b1dcc1b 2737 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2738
2739 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2740 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2741 fsnotify_link_count(dentry->d_inode);
e234f35c 2742 d_delete(dentry);
1da177e4 2743 }
0eeca283 2744
1da177e4
LT		 2745 return error;
2746}
2747
2748/*
2749 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2750 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2751 * writeout happening, and we don't want to prevent access to the directory
2752 * while waiting on the I/O.
2753 */
5590ff0d 2754static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2755{
2ad94ae6
AV		 2756 int error;
2757 char *name;
1da177e4
LT		 2758 struct dentry *dentry;
2759 struct nameidata nd;
2760 struct inode *inode = NULL;
2761
2ad94ae6 2762 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2763 if (error)
2ad94ae6
AV		 2764 return error;
2765
1da177e4
LT		 2766 error = -EISDIR;
2767 if (nd.last_type != LAST_NORM)
2768 goto exit1;
0612d9fb
OH		 2769
2770 nd.flags &= ~LOOKUP_PARENT;
2771
4ac91378 2772 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2773 dentry = lookup_hash(&nd);
1da177e4
LT		 2774 error = PTR_ERR(dentry);
2775 if (!IS_ERR(dentry)) {
2776 /* Why not before? Because we want correct error value */
2777 if (nd.last.name[nd.last.len])
2778 goto slashes;
2779 inode = dentry->d_inode;
2780 if (inode)
7de9c6ee 2781 ihold(inode);
0622753b
DH		 2782 error = mnt_want_write(nd.path.mnt);
2783 if (error)
2784 goto exit2;
be6d3e56
KT		 2785 error = security_path_unlink(&nd.path, dentry);
2786 if (error)
2787 goto exit3;
4ac91378 2788 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2789exit3:
0622753b 2790 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2791 exit2:
2792 dput(dentry);
2793 }
4ac91378 2794 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2795 if (inode)
2796 iput(inode); /* truncate the inode here */
2797exit1:
1d957f9b 2798 path_put(&nd.path);
1da177e4
LT		 2799 putname(name);
2800 return error;
2801
2802slashes:
2803 error = !dentry->d_inode ? -ENOENT :
2804 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2805 goto exit2;
2806}
2807
2e4d0924 2808SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2809{
2810 if ((flag & ~AT_REMOVEDIR) != 0)
2811 return -EINVAL;
2812
2813 if (flag & AT_REMOVEDIR)
2814 return do_rmdir(dfd, pathname);
2815
2816 return do_unlinkat(dfd, pathname);
2817}
2818
3480b257 2819SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2820{
2821 return do_unlinkat(AT_FDCWD, pathname);
2822}
2823
db2e747b 2824int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2825{
a95164d9 2826 int error = may_create(dir, dentry);
1da177e4
LT		 2827
2828 if (error)
2829 return error;
2830
acfa4380 2831 if (!dir->i_op->symlink)
1da177e4
LT		 2832 return -EPERM;
2833
2834 error = security_inode_symlink(dir, dentry, oldname);
2835 if (error)
2836 return error;
2837
1da177e4 2838 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2839 if (!error)
f38aa942 2840 fsnotify_create(dir, dentry);
1da177e4
LT		 2841 return error;
2842}
2843
2e4d0924
HC		 2844SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2845 int, newdfd, const char __user *, newname)
1da177e4 2846{
2ad94ae6
AV		 2847 int error;
2848 char *from;
2849 char *to;
6902d925
DH		 2850 struct dentry *dentry;
2851 struct nameidata nd;
1da177e4
LT		 2852
2853 from = getname(oldname);
2ad94ae6 2854 if (IS_ERR(from))
1da177e4 2855 return PTR_ERR(from);
1da177e4 2856
2ad94ae6 2857 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2858 if (error)
2ad94ae6
AV		 2859 goto out_putname;
2860
6902d925
DH		 2861 dentry = lookup_create(&nd, 0);
2862 error = PTR_ERR(dentry);
2863 if (IS_ERR(dentry))
2864 goto out_unlock;
2865
75c3f29d
DH		 2866 error = mnt_want_write(nd.path.mnt);
2867 if (error)
2868 goto out_dput;
be6d3e56
KT		 2869 error = security_path_symlink(&nd.path, dentry, from);
2870 if (error)
2871 goto out_drop_write;
db2e747b 2872 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2873out_drop_write:
75c3f29d
DH		 2874 mnt_drop_write(nd.path.mnt);
2875out_dput:
6902d925
DH		 2876 dput(dentry);
2877out_unlock:
4ac91378 2878 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2879 path_put(&nd.path);
6902d925
DH		 2880 putname(to);
2881out_putname:
1da177e4
LT		 2882 putname(from);
2883 return error;
2884}
2885
3480b257 2886SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2887{
2888 return sys_symlinkat(oldname, AT_FDCWD, newname);
2889}
2890
1da177e4
LT		 2891int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2892{
2893 struct inode *inode = old_dentry->d_inode;
2894 int error;
2895
2896 if (!inode)
2897 return -ENOENT;
2898
a95164d9 2899 error = may_create(dir, new_dentry);
1da177e4
LT		 2900 if (error)
2901 return error;
2902
2903 if (dir->i_sb != inode->i_sb)
2904 return -EXDEV;
2905
2906 /*
2907 * A link to an append-only or immutable file cannot be created.
2908 */
2909 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2910 return -EPERM;
acfa4380 2911 if (!dir->i_op->link)
1da177e4 2912 return -EPERM;
7e79eedb 2913 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2914 return -EPERM;
2915
2916 error = security_inode_link(old_dentry, dir, new_dentry);
2917 if (error)
2918 return error;
2919
7e79eedb 2920 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 2921 /* Make sure we don't allow creating hardlink to an unlinked file */
2922 if (inode->i_nlink == 0)
2923 error = -ENOENT;
2924 else
2925 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2926 mutex_unlock(&inode->i_mutex);
e31e14ec 2927 if (!error)
7e79eedb 2928 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2929 return error;
2930}
2931
2932/*
2933 * Hardlinks are often used in delicate situations. We avoid
2934 * security-related surprises by not following symlinks on the
2935 * newname. --KAB
2936 *
2937 * We don't follow them on the oldname either to be compatible
2938 * with linux 2.0, and to avoid hard-linking to directories
2939 * and other special files. --ADM
2940 */
2e4d0924
HC		 2941SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2942 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2943{
2944 struct dentry *new_dentry;
2d8f3038
AV		 2945 struct nameidata nd;
2946 struct path old_path;
11a7b371 2947 int how = 0;
1da177e4 2948 int error;
2ad94ae6 2949 char *to;
1da177e4 2950
11a7b371 2951 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 2952 return -EINVAL;
11a7b371
AK		 2953 /*
2954 * To use null names we require CAP_DAC_READ_SEARCH
2955 * This ensures that not everyone will be able to create
2956 * handlink using the passed filedescriptor.
2957 */
2958 if (flags & AT_EMPTY_PATH) {
2959 if (!capable(CAP_DAC_READ_SEARCH))
2960 return -ENOENT;
2961 how = LOOKUP_EMPTY;
2962 }
2963
2964 if (flags & AT_SYMLINK_FOLLOW)
2965 how |= LOOKUP_FOLLOW;
c04030e1 2966
11a7b371 2967 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 2968 if (error)
2ad94ae6
AV		 2969 return error;
2970
2971 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2972 if (error)
2973 goto out;
2974 error = -EXDEV;
2d8f3038 2975 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2976 goto out_release;
2977 new_dentry = lookup_create(&nd, 0);
2978 error = PTR_ERR(new_dentry);
6902d925
DH		 2979 if (IS_ERR(new_dentry))
2980 goto out_unlock;
75c3f29d
DH		 2981 error = mnt_want_write(nd.path.mnt);
2982 if (error)
2983 goto out_dput;
be6d3e56
KT		 2984 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
2985 if (error)
2986 goto out_drop_write;
2d8f3038 2987 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 2988out_drop_write:
75c3f29d
DH		 2989 mnt_drop_write(nd.path.mnt);
2990out_dput:
6902d925
DH		 2991 dput(new_dentry);
2992out_unlock:
4ac91378 2993 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2994out_release:
1d957f9b 2995 path_put(&nd.path);
2ad94ae6 2996 putname(to);
1da177e4 2997out:
2d8f3038 2998 path_put(&old_path);
1da177e4
LT		 2999
3000 return error;
3001}
3002
3480b257 3003SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3004{
c04030e1 3005 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3006}
3007
1da177e4
LT		 3008/*
3009 * The worst of all namespace operations - renaming directory. "Perverted"
3010 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3011 * Problems:
3012 * a) we can get into loop creation. Check is done in is_subdir().
3013 * b) race potential - two innocent renames can create a loop together.
3014 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3015 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3016 * story.
3017 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3018 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3019 * whether the target exists). Solution: try to be smart with locking
3020 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3021 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3022 * move will be locked. Thus we can rank directories by the tree
3023 * (ancestors first) and rank all non-directories after them.
3024 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3025 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3026 * HOWEVER, it relies on the assumption that any object with ->lookup()
3027 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3028 * we'd better make sure that there's no link(2) for them.
3029 * d) some filesystems don't support opened-but-unlinked directories,
3030 * either because of layout or because they are not ready to deal with
3031 * all cases correctly. The latter will be fixed (taking this sort of
3032 * stuff into VFS), but the former is not going away. Solution: the same
3033 * trick as in rmdir().
3034 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3035 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3036 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3037 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3038 * locking].
3039 */
75c96f85
AB		 3040static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3041 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3042{
3043 int error = 0;
3044 struct inode *target;
3045
3046 /*
3047 * If we are going to change the parent - check write permissions,
3048 * we'll need to flip '..'.
3049 */
3050 if (new_dir != old_dir) {
f419a2e3 3051 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3052 if (error)
3053 return error;
3054 }
3055
3056 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3057 if (error)
3058 return error;
3059
3060 target = new_dentry->d_inode;
d83c49f3 3061 if (target)
1b1dcc1b 3062 mutex_lock(&target->i_mutex);
1da177e4
LT		 3063 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3064 error = -EBUSY;
d83c49f3
AV		 3065 else {
3066 if (target)
3067 dentry_unhash(new_dentry);
1da177e4 3068 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3069 }
1da177e4 3070 if (target) {
d83c49f3 3071 if (!error) {
1da177e4 3072 target->i_flags |= S_DEAD;
d83c49f3
AV		 3073 dont_mount(new_dentry);
3074 }
1b1dcc1b 3075 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3076 if (d_unhashed(new_dentry))
3077 d_rehash(new_dentry);
3078 dput(new_dentry);
3079 }
e31e14ec 3080 if (!error)
349457cc
MF		 3081 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3082 d_move(old_dentry,new_dentry);
1da177e4
LT		 3083 return error;
3084}
3085
75c96f85
AB		 3086static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3087 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3088{
3089 struct inode *target;
3090 int error;
3091
3092 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3093 if (error)
3094 return error;
3095
3096 dget(new_dentry);
3097 target = new_dentry->d_inode;
3098 if (target)
1b1dcc1b 3099 mutex_lock(&target->i_mutex);
1da177e4
LT		 3100 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3101 error = -EBUSY;
3102 else
3103 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3104 if (!error) {
bec1052e 3105 if (target)
d83c49f3 3106 dont_mount(new_dentry);
349457cc 3107 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3108 d_move(old_dentry, new_dentry);
1da177e4
LT		 3109 }
3110 if (target)
1b1dcc1b 3111 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3112 dput(new_dentry);
3113 return error;
3114}
3115
3116int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3117 struct inode *new_dir, struct dentry *new_dentry)
3118{
3119 int error;
3120 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3121 const unsigned char *old_name;
1da177e4
LT		 3122
3123 if (old_dentry->d_inode == new_dentry->d_inode)
3124 return 0;
3125
3126 error = may_delete(old_dir, old_dentry, is_dir);
3127 if (error)
3128 return error;
3129
3130 if (!new_dentry->d_inode)
a95164d9 3131 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3132 else
3133 error = may_delete(new_dir, new_dentry, is_dir);
3134 if (error)
3135 return error;
3136
acfa4380 3137 if (!old_dir->i_op->rename)
1da177e4
LT		 3138 return -EPERM;
3139
0eeca283
RL		 3140 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3141
1da177e4
LT		 3142 if (is_dir)
3143 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3144 else
3145 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3146 if (!error)
3147 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3148 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3149 fsnotify_oldname_free(old_name);
3150
1da177e4
LT		 3151 return error;
3152}
3153
2e4d0924
HC		 3154SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3155 int, newdfd, const char __user *, newname)
1da177e4 3156{
2ad94ae6
AV		 3157 struct dentry *old_dir, *new_dir;
3158 struct dentry *old_dentry, *new_dentry;
3159 struct dentry *trap;
1da177e4 3160 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3161 char *from;
3162 char *to;
3163 int error;
1da177e4 3164
2ad94ae6 3165 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3166 if (error)
3167 goto exit;
3168
2ad94ae6 3169 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3170 if (error)
3171 goto exit1;
3172
3173 error = -EXDEV;
4ac91378 3174 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3175 goto exit2;
3176
4ac91378 3177 old_dir = oldnd.path.dentry;
1da177e4
LT		 3178 error = -EBUSY;
3179 if (oldnd.last_type != LAST_NORM)
3180 goto exit2;
3181
4ac91378 3182 new_dir = newnd.path.dentry;
1da177e4
LT		 3183 if (newnd.last_type != LAST_NORM)
3184 goto exit2;
3185
0612d9fb
OH		 3186 oldnd.flags &= ~LOOKUP_PARENT;
3187 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3188 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3189
1da177e4
LT		 3190 trap = lock_rename(new_dir, old_dir);
3191
49705b77 3192 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3193 error = PTR_ERR(old_dentry);
3194 if (IS_ERR(old_dentry))
3195 goto exit3;
3196 /* source must exist */
3197 error = -ENOENT;
3198 if (!old_dentry->d_inode)
3199 goto exit4;
3200 /* unless the source is a directory trailing slashes give -ENOTDIR */
3201 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3202 error = -ENOTDIR;
3203 if (oldnd.last.name[oldnd.last.len])
3204 goto exit4;
3205 if (newnd.last.name[newnd.last.len])
3206 goto exit4;
3207 }
3208 /* source should not be ancestor of target */
3209 error = -EINVAL;
3210 if (old_dentry == trap)
3211 goto exit4;
49705b77 3212 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3213 error = PTR_ERR(new_dentry);
3214 if (IS_ERR(new_dentry))
3215 goto exit4;
3216 /* target should not be an ancestor of source */
3217 error = -ENOTEMPTY;
3218 if (new_dentry == trap)
3219 goto exit5;
3220
9079b1eb
DH		 3221 error = mnt_want_write(oldnd.path.mnt);
3222 if (error)
3223 goto exit5;
be6d3e56
KT		 3224 error = security_path_rename(&oldnd.path, old_dentry,
3225 &newnd.path, new_dentry);
3226 if (error)
3227 goto exit6;
1da177e4
LT		 3228 error = vfs_rename(old_dir->d_inode, old_dentry,
3229 new_dir->d_inode, new_dentry);
be6d3e56 3230exit6:
9079b1eb 3231 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3232exit5:
3233 dput(new_dentry);
3234exit4:
3235 dput(old_dentry);
3236exit3:
3237 unlock_rename(new_dir, old_dir);
3238exit2:
1d957f9b 3239 path_put(&newnd.path);
2ad94ae6 3240 putname(to);
1da177e4 3241exit1:
1d957f9b 3242 path_put(&oldnd.path);
1da177e4 3243 putname(from);
2ad94ae6 3244exit:
1da177e4
LT		 3245 return error;
3246}
3247
a26eab24 3248SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3249{
3250 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3251}
3252
1da177e4
LT		 3253int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3254{
3255 int len;
3256
3257 len = PTR_ERR(link);
3258 if (IS_ERR(link))
3259 goto out;
3260
3261 len = strlen(link);
3262 if (len > (unsigned) buflen)
3263 len = buflen;
3264 if (copy_to_user(buffer, link, len))
3265 len = -EFAULT;
3266out:
3267 return len;
3268}
3269
3270/*
3271 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3272 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3273 * using) it for any given inode is up to filesystem.
3274 */
3275int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3276{
3277 struct nameidata nd;
cc314eef 3278 void *cookie;
694a1764 3279 int res;
cc314eef 3280
1da177e4 3281 nd.depth = 0;
cc314eef 3282 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3283 if (IS_ERR(cookie))
3284 return PTR_ERR(cookie);
3285
3286 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3287 if (dentry->d_inode->i_op->put_link)
3288 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3289 return res;
1da177e4
LT		 3290}
3291
3292int vfs_follow_link(struct nameidata *nd, const char *link)
3293{
3294 return __vfs_follow_link(nd, link);
3295}
3296
3297/* get the link contents into pagecache */
3298static char *page_getlink(struct dentry * dentry, struct page **ppage)
3299{
ebd09abb
DG		 3300 char *kaddr;
3301 struct page *page;
1da177e4 3302 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3303 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3304 if (IS_ERR(page))
6fe6900e 3305 return (char*)page;
1da177e4 3306 *ppage = page;
ebd09abb
DG		 3307 kaddr = kmap(page);
3308 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3309 return kaddr;
1da177e4
LT		 3310}
3311
3312int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3313{
3314 struct page *page = NULL;
3315 char *s = page_getlink(dentry, &page);
3316 int res = vfs_readlink(dentry,buffer,buflen,s);
3317 if (page) {
3318 kunmap(page);
3319 page_cache_release(page);
3320 }
3321 return res;
3322}
3323
cc314eef 3324void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3325{
cc314eef 3326 struct page *page = NULL;
1da177e4 3327 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3328 return page;
1da177e4
LT		 3329}
3330
cc314eef 3331void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3332{
cc314eef
LT		 3333 struct page *page = cookie;
3334
3335 if (page) {
1da177e4
LT		 3336 kunmap(page);
3337 page_cache_release(page);
1da177e4
LT		 3338 }
3339}
3340
54566b2c
NP		 3341/*
3342 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3343 */
3344int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3345{
3346 struct address_space *mapping = inode->i_mapping;
0adb25d2 3347 struct page *page;
afddba49 3348 void *fsdata;
beb497ab 3349 int err;
1da177e4 3350 char *kaddr;
54566b2c
NP		 3351 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3352 if (nofs)
3353 flags |= AOP_FLAG_NOFS;
1da177e4 3354
7e53cac4 3355retry:
afddba49 3356 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3357 flags, &page, &fsdata);
1da177e4 3358 if (err)
afddba49
NP		 3359 goto fail;
3360
1da177e4
LT		 3361 kaddr = kmap_atomic(page, KM_USER0);
3362 memcpy(kaddr, symname, len-1);
3363 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3364
3365 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3366 page, fsdata);
1da177e4
LT		 3367 if (err < 0)
3368 goto fail;
afddba49
NP		 3369 if (err < len-1)
3370 goto retry;
3371
1da177e4
LT		 3372 mark_inode_dirty(inode);
3373 return 0;
1da177e4
LT		 3374fail:
3375 return err;
3376}
3377
0adb25d2
KK		 3378int page_symlink(struct inode *inode, const char *symname, int len)
3379{
3380 return __page_symlink(inode, symname, len,
54566b2c 3381 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3382}
3383
92e1d5be 3384const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3385 .readlink = generic_readlink,
3386 .follow_link = page_follow_link_light,
3387 .put_link = page_put_link,
3388};
3389
2d8f3038 3390EXPORT_SYMBOL(user_path_at);
cc53ce53 3391EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3392EXPORT_SYMBOL(follow_down);
3393EXPORT_SYMBOL(follow_up);
3394EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3395EXPORT_SYMBOL(getname);
3396EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3397EXPORT_SYMBOL(lookup_one_len);
3398EXPORT_SYMBOL(page_follow_link_light);
3399EXPORT_SYMBOL(page_put_link);
3400EXPORT_SYMBOL(page_readlink);
0adb25d2 3401EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3402EXPORT_SYMBOL(page_symlink);
3403EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3404EXPORT_SYMBOL(kern_path_parent);
d1811465 3405EXPORT_SYMBOL(kern_path);
16f18200 3406EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3407EXPORT_SYMBOL(inode_permission);
8c744fb8 3408EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3409EXPORT_SYMBOL(unlock_rename);
3410EXPORT_SYMBOL(vfs_create);
3411EXPORT_SYMBOL(vfs_follow_link);
3412EXPORT_SYMBOL(vfs_link);
3413EXPORT_SYMBOL(vfs_mkdir);
3414EXPORT_SYMBOL(vfs_mknod);
3415EXPORT_SYMBOL(generic_permission);
3416EXPORT_SYMBOL(vfs_readlink);
3417EXPORT_SYMBOL(vfs_rename);
3418EXPORT_SYMBOL(vfs_rmdir);
3419EXPORT_SYMBOL(vfs_symlink);
3420EXPORT_SYMBOL(vfs_unlink);
3421EXPORT_SYMBOL(dentry_unhash);
3422EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)