projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
kill __lookup_one_len()
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / namei.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
1da177e4 22#include <linux/pagemap.h>
0eeca283 23#include <linux/fsnotify.h>
1da177e4
LT		 24#include <linux/personality.h>
25#include <linux/security.h>
6146f0d5 26#include <linux/ima.h>
1da177e4
LT		 27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
16f7e0fe 30#include <linux/capability.h>
834f2a4a 31#include <linux/file.h>
5590ff0d 32#include <linux/fcntl.h>
08ce5f16 33#include <linux/device_cgroup.h>
5ad4e53b 34#include <linux/fs_struct.h>
1da177e4
LT		 35#include <asm/uaccess.h>
36
e81e3f4d
EP		 37#include "internal.h"
38
1da177e4
LT		 39/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 107 * any extra contention...
108 */
109
110/* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
113 *
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
116 */
858119e1 117static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 118{
119 int retval;
120 unsigned long len = PATH_MAX;
121
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
127 }
128
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
137}
138
139char * getname(const char __user * filename)
140{
141 char *tmp, *result;
142
143 result = ERR_PTR(-ENOMEM);
144 tmp = __getname();
145 if (tmp) {
146 int retval = do_getname(filename, tmp);
147
148 result = tmp;
149 if (retval < 0) {
150 __putname(tmp);
151 result = ERR_PTR(retval);
152 }
153 }
154 audit_getname(result);
155 return result;
156}
157
158#ifdef CONFIG_AUDITSYSCALL
159void putname(const char *name)
160{
5ac3a9c2 161 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 162 audit_putname(name);
163 else
164 __putname(name);
165}
166EXPORT_SYMBOL(putname);
167#endif
168
5909ccaa
LT		 169/*
170 * This does basic POSIX ACL permission checking
1da177e4 171 */
b74c79e9
NP		 172static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
173 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
1da177e4
LT		 174{
175 umode_t mode = inode->i_mode;
176
e6305c43
AV		 177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
178
da9592ed 179 if (current_fsuid() == inode->i_uid)
1da177e4
LT		 180 mode >>= 6;
181 else {
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
b74c79e9
NP		 183 int error = check_acl(inode, mask, flags);
184 if (error != -EAGAIN)
185 return error;
1da177e4
LT		 186 }
187
188 if (in_group_p(inode->i_gid))
189 mode >>= 3;
190 }
191
192 /*
193 * If the DACs are ok we don't need any capability check.
194 */
e6305c43 195 if ((mask & ~mode) == 0)
1da177e4 196 return 0;
5909ccaa
LT		 197 return -EACCES;
198}
199
200/**
b74c79e9 201 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa
LT		 202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
39191628 205 * @flags: IPERM_FLAG_ flags.
5909ccaa
LT		 206 *
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 210 * are used for other things.
211 *
212 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
213 * request cannot be satisfied (eg. requires blocking or too much complexity).
214 * It would then be called again in ref-walk mode.
5909ccaa 215 */
b74c79e9
NP		 216int generic_permission(struct inode *inode, int mask, unsigned int flags,
217 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
5909ccaa
LT		 218{
219 int ret;
220
221 /*
222 * Do the basic POSIX ACL permission checks.
223 */
b74c79e9 224 ret = acl_permission_check(inode, mask, flags, check_acl);
5909ccaa
LT		 225 if (ret != -EACCES)
226 return ret;
1da177e4 227
1da177e4
LT		 228 /*
229 * Read/write DACs are always overridable.
230 * Executable DACs are overridable if at least one exec bit is set.
231 */
f696a365 232 if (!(mask & MAY_EXEC) || execute_ok(inode))
1da177e4
LT		 233 if (capable(CAP_DAC_OVERRIDE))
234 return 0;
235
236 /*
237 * Searching includes executable on directories, else just read.
238 */
7ea66001 239 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
1da177e4
LT		 240 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
241 if (capable(CAP_DAC_READ_SEARCH))
242 return 0;
243
244 return -EACCES;
245}
246
cb23beb5
CH		 247/**
248 * inode_permission - check for access rights to a given inode
249 * @inode: inode to check permission on
250 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
251 *
252 * Used to check for read/write/execute permissions on an inode.
253 * We use "fsuid" for this, letting us set arbitrary permissions
254 * for filesystem access without changing the "normal" uids which
255 * are used for other things.
256 */
f419a2e3 257int inode_permission(struct inode *inode, int mask)
1da177e4 258{
e6305c43 259 int retval;
1da177e4
LT		 260
261 if (mask & MAY_WRITE) {
22590e41 262 umode_t mode = inode->i_mode;
1da177e4
LT		 263
264 /*
265 * Nobody gets write access to a read-only fs.
266 */
267 if (IS_RDONLY(inode) &&
268 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
269 return -EROFS;
270
271 /*
272 * Nobody gets write access to an immutable file.
273 */
274 if (IS_IMMUTABLE(inode))
275 return -EACCES;
276 }
277
acfa4380 278 if (inode->i_op->permission)
b74c79e9 279 retval = inode->i_op->permission(inode, mask, 0);
f696a365 280 else
b74c79e9
NP		 281 retval = generic_permission(inode, mask, 0,
282 inode->i_op->check_acl);
f696a365 283
1da177e4
LT		 284 if (retval)
285 return retval;
286
08ce5f16
SH		 287 retval = devcgroup_inode_permission(inode, mask);
288 if (retval)
289 return retval;
290
d09ca739 291 return security_inode_permission(inode, mask);
1da177e4
LT		 292}
293
8c744fb8
CH		 294/**
295 * file_permission - check for additional access rights to a given file
296 * @file: file to check access rights for
297 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
298 *
299 * Used to check for read/write/execute permissions on an already opened
300 * file.
301 *
302 * Note:
303 * Do not use this function in new code. All access checks should
cb23beb5 304 * be done using inode_permission().
8c744fb8
CH		 305 */
306int file_permission(struct file *file, int mask)
307{
f419a2e3 308 return inode_permission(file->f_path.dentry->d_inode, mask);
8c744fb8
CH		 309}
310
1da177e4
LT		 311/*
312 * get_write_access() gets write permission for a file.
313 * put_write_access() releases this write permission.
314 * This is used for regular files.
315 * We cannot support write (and maybe mmap read-write shared) accesses and
316 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
317 * can have the following values:
318 * 0: no writers, no VM_DENYWRITE mappings
319 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
320 * > 0: (i_writecount) users are writing to the file.
321 *
322 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
323 * except for the cases where we don't hold i_writecount yet. Then we need to
324 * use {get,deny}_write_access() - these functions check the sign and refuse
325 * to do the change if sign is wrong. Exclusion between them is provided by
326 * the inode->i_lock spinlock.
327 */
328
329int get_write_access(struct inode * inode)
330{
331 spin_lock(&inode->i_lock);
332 if (atomic_read(&inode->i_writecount) < 0) {
333 spin_unlock(&inode->i_lock);
334 return -ETXTBSY;
335 }
336 atomic_inc(&inode->i_writecount);
337 spin_unlock(&inode->i_lock);
338
339 return 0;
340}
341
342int deny_write_access(struct file * file)
343{
0f7fc9e4 344 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4
LT		 345
346 spin_lock(&inode->i_lock);
347 if (atomic_read(&inode->i_writecount) > 0) {
348 spin_unlock(&inode->i_lock);
349 return -ETXTBSY;
350 }
351 atomic_dec(&inode->i_writecount);
352 spin_unlock(&inode->i_lock);
353
354 return 0;
355}
356
5dd784d0
JB		 357/**
358 * path_get - get a reference to a path
359 * @path: path to get the reference to
360 *
361 * Given a path increment the reference count to the dentry and the vfsmount.
362 */
363void path_get(struct path *path)
364{
365 mntget(path->mnt);
366 dget(path->dentry);
367}
368EXPORT_SYMBOL(path_get);
369
1d957f9b
JB		 370/**
371 * path_put - put a reference to a path
372 * @path: path to put the reference to
373 *
374 * Given a path decrement the reference count to the dentry and the vfsmount.
375 */
376void path_put(struct path *path)
1da177e4 377{
1d957f9b
JB		 378 dput(path->dentry);
379 mntput(path->mnt);
1da177e4 380}
1d957f9b 381EXPORT_SYMBOL(path_put);
1da177e4 382
31e6b01f
NP		 383/**
384 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
385 * @nd: nameidata pathwalk data to drop
39191628 386 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 387 *
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
390 * to drop out of rcu-walk mode and take normal reference counts on dentries
391 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
392 * refcounts at the last known good point before rcu-walk got stuck, so
393 * ref-walk may continue from there. If this is not successful (eg. a seqcount
394 * has changed), then failure is returned and path walk restarts from the
395 * beginning in ref-walk mode.
396 *
397 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
398 * ref-walk. Must be called from rcu-walk context.
399 */
400static int nameidata_drop_rcu(struct nameidata *nd)
401{
402 struct fs_struct *fs = current->fs;
403 struct dentry *dentry = nd->path.dentry;
404
405 BUG_ON(!(nd->flags & LOOKUP_RCU));
406 if (nd->root.mnt) {
407 spin_lock(&fs->lock);
408 if (nd->root.mnt != fs->root.mnt ||
409 nd->root.dentry != fs->root.dentry)
410 goto err_root;
411 }
412 spin_lock(&dentry->d_lock);
413 if (!__d_rcu_to_refcount(dentry, nd->seq))
414 goto err;
415 BUG_ON(nd->inode != dentry->d_inode);
416 spin_unlock(&dentry->d_lock);
417 if (nd->root.mnt) {
418 path_get(&nd->root);
419 spin_unlock(&fs->lock);
420 }
421 mntget(nd->path.mnt);
422
423 rcu_read_unlock();
424 br_read_unlock(vfsmount_lock);
425 nd->flags &= ~LOOKUP_RCU;
426 return 0;
427err:
428 spin_unlock(&dentry->d_lock);
429err_root:
430 if (nd->root.mnt)
431 spin_unlock(&fs->lock);
432 return -ECHILD;
433}
434
435/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
436static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
437{
438 if (nd->flags & LOOKUP_RCU)
439 return nameidata_drop_rcu(nd);
440 return 0;
441}
442
443/**
444 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
445 * @nd: nameidata pathwalk data to drop
446 * @dentry: dentry to drop
39191628 447 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 448 *
449 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
450 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
451 * @nd. Must be called from rcu-walk context.
452 */
453static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
454{
455 struct fs_struct *fs = current->fs;
456 struct dentry *parent = nd->path.dentry;
457
458 BUG_ON(!(nd->flags & LOOKUP_RCU));
459 if (nd->root.mnt) {
460 spin_lock(&fs->lock);
461 if (nd->root.mnt != fs->root.mnt ||
462 nd->root.dentry != fs->root.dentry)
463 goto err_root;
464 }
465 spin_lock(&parent->d_lock);
466 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
467 if (!__d_rcu_to_refcount(dentry, nd->seq))
468 goto err;
469 /*
470 * If the sequence check on the child dentry passed, then the child has
471 * not been removed from its parent. This means the parent dentry must
472 * be valid and able to take a reference at this point.
473 */
474 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
475 BUG_ON(!parent->d_count);
476 parent->d_count++;
477 spin_unlock(&dentry->d_lock);
478 spin_unlock(&parent->d_lock);
479 if (nd->root.mnt) {
480 path_get(&nd->root);
481 spin_unlock(&fs->lock);
482 }
483 mntget(nd->path.mnt);
484
485 rcu_read_unlock();
486 br_read_unlock(vfsmount_lock);
487 nd->flags &= ~LOOKUP_RCU;
488 return 0;
489err:
490 spin_unlock(&dentry->d_lock);
491 spin_unlock(&parent->d_lock);
492err_root:
493 if (nd->root.mnt)
494 spin_unlock(&fs->lock);
495 return -ECHILD;
496}
497
498/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
499static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
500{
a7472bab
AV		 501 if (nd->flags & LOOKUP_RCU) {
502 if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) {
503 nd->flags &= ~LOOKUP_RCU;
504 nd->root.mnt = NULL;
505 rcu_read_unlock();
506 br_read_unlock(vfsmount_lock);
507 return -ECHILD;
508 }
509 }
31e6b01f
NP		 510 return 0;
511}
512
513/**
514 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
515 * @nd: nameidata pathwalk data to drop
39191628 516 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 517 *
518 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
519 * nd->path should be the final element of the lookup, so nd->root is discarded.
520 * Must be called from rcu-walk context.
521 */
522static int nameidata_drop_rcu_last(struct nameidata *nd)
523{
524 struct dentry *dentry = nd->path.dentry;
525
526 BUG_ON(!(nd->flags & LOOKUP_RCU));
527 nd->flags &= ~LOOKUP_RCU;
528 nd->root.mnt = NULL;
529 spin_lock(&dentry->d_lock);
530 if (!__d_rcu_to_refcount(dentry, nd->seq))
531 goto err_unlock;
532 BUG_ON(nd->inode != dentry->d_inode);
533 spin_unlock(&dentry->d_lock);
534
535 mntget(nd->path.mnt);
536
537 rcu_read_unlock();
538 br_read_unlock(vfsmount_lock);
539
540 return 0;
541
542err_unlock:
543 spin_unlock(&dentry->d_lock);
544 rcu_read_unlock();
545 br_read_unlock(vfsmount_lock);
546 return -ECHILD;
547}
548
834f2a4a
TM		 549/**
550 * release_open_intent - free up open intent resources
551 * @nd: pointer to nameidata
552 */
553void release_open_intent(struct nameidata *nd)
554{
2dab5974
LT		 555 struct file *file = nd->intent.open.file;
556
557 if (file && !IS_ERR(file)) {
558 if (file->f_path.dentry == NULL)
559 put_filp(file);
560 else
561 fput(file);
562 }
834f2a4a
TM		 563}
564
f60aef7e 565static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 566{
f60aef7e 567 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 568}
569
f5e1c1c1 570static struct dentry *
bcdc5e01
IK		 571do_revalidate(struct dentry *dentry, struct nameidata *nd)
572{
f5e1c1c1 573 int status = d_revalidate(dentry, nd);
bcdc5e01
IK		 574 if (unlikely(status <= 0)) {
575 /*
576 * The dentry failed validation.
577 * If d_revalidate returned 0 attempt to invalidate
578 * the dentry otherwise d_revalidate is asking us
579 * to return a fail status.
580 */
34286d66 581 if (status < 0) {
f5e1c1c1 582 dput(dentry);
34286d66 583 dentry = ERR_PTR(status);
f5e1c1c1
AV		 584 } else if (!d_invalidate(dentry)) {
585 dput(dentry);
586 dentry = NULL;
bcdc5e01
IK		 587 }
588 }
589 return dentry;
590}
591
f5e1c1c1
AV		 592static inline struct dentry *
593do_revalidate_rcu(struct dentry *dentry, struct nameidata *nd)
594{
f60aef7e 595 int status = d_revalidate(dentry, nd);
f5e1c1c1
AV		 596 if (likely(status > 0))
597 return dentry;
598 if (status == -ECHILD) {
599 if (nameidata_dentry_drop_rcu(nd, dentry))
600 return ERR_PTR(-ECHILD);
601 return do_revalidate(dentry, nd);
602 }
603 if (status < 0)
604 return ERR_PTR(status);
605 /* Don't d_invalidate in rcu-walk mode */
606 if (nameidata_dentry_drop_rcu(nd, dentry))
607 return ERR_PTR(-ECHILD);
608 if (!d_invalidate(dentry)) {
609 dput(dentry);
610 dentry = NULL;
611 }
612 return dentry;
613}
614
39159de2 615/*
16c2cd71 616 * handle_reval_path - force revalidation of a dentry
39159de2
JL		 617 *
618 * In some situations the path walking code will trust dentries without
619 * revalidating them. This causes problems for filesystems that depend on
620 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
621 * (which indicates that it's possible for the dentry to go stale), force
622 * a d_revalidate call before proceeding.
623 *
624 * Returns 0 if the revalidation was successful. If the revalidation fails,
625 * either return the error returned by d_revalidate or -ESTALE if the
626 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
627 * invalidate the dentry. It's up to the caller to handle putting references
628 * to the path if necessary.
629 */
16c2cd71 630static inline int handle_reval_path(struct nameidata *nd)
39159de2 631{
16c2cd71 632 struct dentry *dentry = nd->path.dentry;
39159de2 633 int status;
39159de2 634
16c2cd71
AV		 635 if (likely(!(nd->flags & LOOKUP_JUMPED)))
636 return 0;
637
638 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 639 return 0;
640
16c2cd71
AV		 641 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
642 return 0;
643
644 /* Note: we do not d_invalidate() */
34286d66 645 status = d_revalidate(dentry, nd);
39159de2
JL		 646 if (status > 0)
647 return 0;
648
16c2cd71 649 if (!status)
39159de2 650 status = -ESTALE;
16c2cd71 651
39159de2
JL		 652 return status;
653}
654
1da177e4 655/*
b75b5086
AV		 656 * Short-cut version of permission(), for calling on directories
657 * during pathname resolution. Combines parts of permission()
658 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
1da177e4
LT		 659 *
660 * If appropriate, check DAC only. If not appropriate, or
b75b5086 661 * short-cut DAC fails, then call ->permission() to do more
1da177e4
LT		 662 * complete permission check.
663 */
b74c79e9 664static inline int exec_permission(struct inode *inode, unsigned int flags)
1da177e4 665{
5909ccaa 666 int ret;
1da177e4 667
cb9179ea 668 if (inode->i_op->permission) {
b74c79e9
NP		 669 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
670 } else {
671 ret = acl_permission_check(inode, MAY_EXEC, flags,
672 inode->i_op->check_acl);
cb9179ea 673 }
b74c79e9 674 if (likely(!ret))
1da177e4 675 goto ok;
b74c79e9 676 if (ret == -ECHILD)
31e6b01f 677 return ret;
1da177e4 678
f1ac9f6b 679 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
1da177e4
LT		 680 goto ok;
681
5909ccaa 682 return ret;
1da177e4 683ok:
b74c79e9 684 return security_inode_exec_permission(inode, flags);
1da177e4
LT		 685}
686
2a737871
AV		 687static __always_inline void set_root(struct nameidata *nd)
688{
f7ad3c6b
MS		 689 if (!nd->root.mnt)
690 get_fs_root(current->fs, &nd->root);
2a737871
AV		 691}
692
6de88d72
AV		 693static int link_path_walk(const char *, struct nameidata *);
694
31e6b01f
NP		 695static __always_inline void set_root_rcu(struct nameidata *nd)
696{
697 if (!nd->root.mnt) {
698 struct fs_struct *fs = current->fs;
c28cc364
NP		 699 unsigned seq;
700
701 do {
702 seq = read_seqcount_begin(&fs->seq);
703 nd->root = fs->root;
704 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 705 }
706}
707
f1662356 708static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 709{
31e6b01f
NP		 710 int ret;
711
1da177e4
LT		 712 if (IS_ERR(link))
713 goto fail;
714
715 if (*link == '/') {
2a737871 716 set_root(nd);
1d957f9b 717 path_put(&nd->path);
2a737871
AV		 718 nd->path = nd->root;
719 path_get(&nd->root);
16c2cd71 720 nd->flags |= LOOKUP_JUMPED;
1da177e4 721 }
31e6b01f 722 nd->inode = nd->path.dentry->d_inode;
b4091d5f 723
31e6b01f
NP		 724 ret = link_path_walk(link, nd);
725 return ret;
1da177e4 726fail:
1d957f9b 727 path_put(&nd->path);
1da177e4
LT		 728 return PTR_ERR(link);
729}
730
1d957f9b 731static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 732{
733 dput(path->dentry);
4ac91378 734 if (path->mnt != nd->path.mnt)
051d3812
IK		 735 mntput(path->mnt);
736}
737
7b9337aa
NP		 738static inline void path_to_nameidata(const struct path *path,
739 struct nameidata *nd)
051d3812 740{
31e6b01f
NP		 741 if (!(nd->flags & LOOKUP_RCU)) {
742 dput(nd->path.dentry);
743 if (nd->path.mnt != path->mnt)
744 mntput(nd->path.mnt);
9a229683 745 }
31e6b01f 746 nd->path.mnt = path->mnt;
4ac91378 747 nd->path.dentry = path->dentry;
051d3812
IK		 748}
749
def4af30 750static __always_inline int
7b9337aa 751__do_follow_link(const struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 752{
753 int error;
7b9337aa 754 struct dentry *dentry = link->dentry;
1da177e4 755
844a3917
AV		 756 BUG_ON(nd->flags & LOOKUP_RCU);
757
7b9337aa 758 touch_atime(link->mnt, dentry);
1da177e4 759 nd_set_link(nd, NULL);
cd4e91d3 760
87556ef1
DH		 761 if (link->mnt == nd->path.mnt)
762 mntget(link->mnt);
31e6b01f 763
36f3b4f6
AV		 764 error = security_inode_follow_link(link->dentry, nd);
765 if (error) {
766 *p = ERR_PTR(error); /* no ->put_link(), please */
767 path_put(&nd->path);
768 return error;
769 }
770
86acdca1 771 nd->last_type = LAST_BIND;
def4af30
AV		 772 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
773 error = PTR_ERR(*p);
774 if (!IS_ERR(*p)) {
1da177e4 775 char *s = nd_get_link(nd);
cc314eef 776 error = 0;
1da177e4
LT		 777 if (s)
778 error = __vfs_follow_link(nd, s);
16c2cd71
AV		 779 else if (nd->last_type == LAST_BIND)
780 nd->flags |= LOOKUP_JUMPED;
1da177e4 781 }
1da177e4
LT		 782 return error;
783}
784
785/*
786 * This limits recursive symlink follows to 8, while
787 * limiting consecutive symlinks to 40.
788 *
789 * Without that kind of total limit, nasty chains of consecutive
790 * symlinks can cause almost arbitrarily long lookups.
791 */
3abb17e8 792static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd)
1da177e4 793{
def4af30 794 void *cookie;
1da177e4 795 int err = -ELOOP;
844a3917
AV		 796
797 /* We drop rcu-walk here */
798 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
799 return -ECHILD;
3abb17e8 800 BUG_ON(inode != path->dentry->d_inode);
844a3917 801
1da177e4
LT		 802 if (current->link_count >= MAX_NESTED_LINKS)
803 goto loop;
804 if (current->total_link_count >= 40)
805 goto loop;
806 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
807 cond_resched();
1da177e4
LT		 808 current->link_count++;
809 current->total_link_count++;
810 nd->depth++;
def4af30
AV		 811 err = __do_follow_link(path, nd, &cookie);
812 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
813 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
258fa999 814 path_put(path);
839d9f93
AV		 815 current->link_count--;
816 nd->depth--;
1da177e4
LT		 817 return err;
818loop:
1d957f9b
JB		 819 path_put_conditional(path, nd);
820 path_put(&nd->path);
1da177e4
LT		 821 return err;
822}
823
31e6b01f
NP		 824static int follow_up_rcu(struct path *path)
825{
826 struct vfsmount *parent;
827 struct dentry *mountpoint;
828
829 parent = path->mnt->mnt_parent;
830 if (parent == path->mnt)
831 return 0;
832 mountpoint = path->mnt->mnt_mountpoint;
833 path->dentry = mountpoint;
834 path->mnt = parent;
835 return 1;
836}
837
bab77ebf 838int follow_up(struct path *path)
1da177e4
LT		 839{
840 struct vfsmount *parent;
841 struct dentry *mountpoint;
99b7db7b
NP		 842
843 br_read_lock(vfsmount_lock);
bab77ebf
AV		 844 parent = path->mnt->mnt_parent;
845 if (parent == path->mnt) {
99b7db7b 846 br_read_unlock(vfsmount_lock);
1da177e4
LT		 847 return 0;
848 }
849 mntget(parent);
bab77ebf 850 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 851 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 852 dput(path->dentry);
853 path->dentry = mountpoint;
854 mntput(path->mnt);
855 path->mnt = parent;
1da177e4
LT		 856 return 1;
857}
858
b5c84bf6 859/*
9875cf80
DH		 860 * Perform an automount
861 * - return -EISDIR to tell follow_managed() to stop and return the path we
862 * were called with.
1da177e4 863 */
9875cf80
DH		 864static int follow_automount(struct path *path, unsigned flags,
865 bool *need_mntput)
31e6b01f 866{
9875cf80 867 struct vfsmount *mnt;
ea5b778a 868 int err;
9875cf80
DH		 869
870 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
871 return -EREMOTE;
872
6f45b656
DH		 873 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
874 * and this is the terminal part of the path.
875 */
876 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
877 return -EISDIR; /* we actually want to stop here */
878
9875cf80
DH		 879 /* We want to mount if someone is trying to open/create a file of any
880 * type under the mountpoint, wants to traverse through the mountpoint
881 * or wants to open the mounted directory.
882 *
883 * We don't want to mount if someone's just doing a stat and they've
884 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
885 * appended a '/' to the name.
886 */
887 if (!(flags & LOOKUP_FOLLOW) &&
888 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
889 LOOKUP_OPEN | LOOKUP_CREATE)))
890 return -EISDIR;
891
892 current->total_link_count++;
893 if (current->total_link_count >= 40)
894 return -ELOOP;
895
896 mnt = path->dentry->d_op->d_automount(path);
897 if (IS_ERR(mnt)) {
898 /*
899 * The filesystem is allowed to return -EISDIR here to indicate
900 * it doesn't want to automount. For instance, autofs would do
901 * this so that its userspace daemon can mount on this dentry.
902 *
903 * However, we can only permit this if it's a terminal point in
904 * the path being looked up; if it wasn't then the remainder of
905 * the path is inaccessible and we should say so.
906 */
907 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
908 return -EREMOTE;
909 return PTR_ERR(mnt);
31e6b01f 910 }
ea5b778a 911
9875cf80
DH		 912 if (!mnt) /* mount collision */
913 return 0;
31e6b01f 914
19a167af 915 err = finish_automount(mnt, path);
9875cf80 916
ea5b778a
DH		 917 switch (err) {
918 case -EBUSY:
919 /* Someone else made a mount here whilst we were busy */
19a167af 920 return 0;
ea5b778a 921 case 0:
ea5b778a
DH		 922 dput(path->dentry);
923 if (*need_mntput)
924 mntput(path->mnt);
925 path->mnt = mnt;
926 path->dentry = dget(mnt->mnt_root);
927 *need_mntput = true;
928 return 0;
19a167af
AV		 929 default:
930 return err;
ea5b778a 931 }
19a167af 932
463ffb2e
AV		 933}
934
9875cf80
DH		 935/*
936 * Handle a dentry that is managed in some way.
cc53ce53 937 * - Flagged for transit management (autofs)
9875cf80
DH		 938 * - Flagged as mountpoint
939 * - Flagged as automount point
940 *
941 * This may only be called in refwalk mode.
942 *
943 * Serialization is taken care of in namespace.c
944 */
945static int follow_managed(struct path *path, unsigned flags)
1da177e4 946{
9875cf80
DH		 947 unsigned managed;
948 bool need_mntput = false;
949 int ret;
950
951 /* Given that we're not holding a lock here, we retain the value in a
952 * local variable for each dentry as we look at it so that we don't see
953 * the components of that value change under us */
954 while (managed = ACCESS_ONCE(path->dentry->d_flags),
955 managed &= DCACHE_MANAGED_DENTRY,
956 unlikely(managed != 0)) {
cc53ce53
DH		 957 /* Allow the filesystem to manage the transit without i_mutex
958 * being held. */
959 if (managed & DCACHE_MANAGE_TRANSIT) {
960 BUG_ON(!path->dentry->d_op);
961 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 962 ret = path->dentry->d_op->d_manage(path->dentry,
963 false, false);
cc53ce53
DH		 964 if (ret < 0)
965 return ret == -EISDIR ? 0 : ret;
966 }
967
9875cf80
DH		 968 /* Transit to a mounted filesystem. */
969 if (managed & DCACHE_MOUNTED) {
970 struct vfsmount *mounted = lookup_mnt(path);
971 if (mounted) {
972 dput(path->dentry);
973 if (need_mntput)
974 mntput(path->mnt);
975 path->mnt = mounted;
976 path->dentry = dget(mounted->mnt_root);
977 need_mntput = true;
978 continue;
979 }
980
981 /* Something is mounted on this dentry in another
982 * namespace and/or whatever was mounted there in this
983 * namespace got unmounted before we managed to get the
984 * vfsmount_lock */
985 }
986
987 /* Handle an automount point */
988 if (managed & DCACHE_NEED_AUTOMOUNT) {
989 ret = follow_automount(path, flags, &need_mntput);
990 if (ret < 0)
991 return ret == -EISDIR ? 0 : ret;
992 continue;
993 }
994
995 /* We didn't change the current path point */
996 break;
1da177e4 997 }
9875cf80 998 return 0;
1da177e4
LT		 999}
1000
cc53ce53 1001int follow_down_one(struct path *path)
1da177e4
LT		 1002{
1003 struct vfsmount *mounted;
1004
1c755af4 1005 mounted = lookup_mnt(path);
1da177e4 1006 if (mounted) {
9393bd07
AV		 1007 dput(path->dentry);
1008 mntput(path->mnt);
1009 path->mnt = mounted;
1010 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 1011 return 1;
1012 }
1013 return 0;
1014}
1015
9875cf80
DH		 1016/*
1017 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
cc53ce53 1018 * meet a managed dentry and we're not walking to "..". True is returned to
9875cf80
DH		 1019 * continue, false to abort.
1020 */
1021static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1022 struct inode **inode, bool reverse_transit)
1023{
1024 while (d_mountpoint(path->dentry)) {
1025 struct vfsmount *mounted;
ab90911f
DH		 1026 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1027 !reverse_transit &&
1028 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1029 return false;
9875cf80
DH		 1030 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1031 if (!mounted)
1032 break;
1033 path->mnt = mounted;
1034 path->dentry = mounted->mnt_root;
1035 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1036 *inode = path->dentry->d_inode;
1037 }
1038
1039 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1040 return reverse_transit;
1041 return true;
1042}
1043
31e6b01f
NP		 1044static int follow_dotdot_rcu(struct nameidata *nd)
1045{
1046 struct inode *inode = nd->inode;
1047
1048 set_root_rcu(nd);
1049
9875cf80 1050 while (1) {
31e6b01f
NP		 1051 if (nd->path.dentry == nd->root.dentry &&
1052 nd->path.mnt == nd->root.mnt) {
1053 break;
1054 }
1055 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1056 struct dentry *old = nd->path.dentry;
1057 struct dentry *parent = old->d_parent;
1058 unsigned seq;
1059
1060 seq = read_seqcount_begin(&parent->d_seq);
1061 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 1062 goto failed;
31e6b01f
NP		 1063 inode = parent->d_inode;
1064 nd->path.dentry = parent;
1065 nd->seq = seq;
1066 break;
1067 }
1068 if (!follow_up_rcu(&nd->path))
1069 break;
1070 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1071 inode = nd->path.dentry->d_inode;
1072 }
9875cf80 1073 __follow_mount_rcu(nd, &nd->path, &inode, true);
31e6b01f 1074 nd->inode = inode;
31e6b01f 1075 return 0;
ef7562d5
AV		 1076
1077failed:
1078 nd->flags &= ~LOOKUP_RCU;
1079 nd->root.mnt = NULL;
1080 rcu_read_unlock();
1081 br_read_unlock(vfsmount_lock);
1082 return -ECHILD;
31e6b01f
NP		 1083}
1084
cc53ce53
DH		 1085/*
1086 * Follow down to the covering mount currently visible to userspace. At each
1087 * point, the filesystem owning that dentry may be queried as to whether the
1088 * caller is permitted to proceed or not.
1089 *
1090 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1091 * being true).
1092 */
1093int follow_down(struct path *path, bool mounting_here)
1094{
1095 unsigned managed;
1096 int ret;
1097
1098 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1099 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1100 /* Allow the filesystem to manage the transit without i_mutex
1101 * being held.
1102 *
1103 * We indicate to the filesystem if someone is trying to mount
1104 * something here. This gives autofs the chance to deny anyone
1105 * other than its daemon the right to mount on its
1106 * superstructure.
1107 *
1108 * The filesystem may sleep at this point.
1109 */
1110 if (managed & DCACHE_MANAGE_TRANSIT) {
1111 BUG_ON(!path->dentry->d_op);
1112 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 1113 ret = path->dentry->d_op->d_manage(
1114 path->dentry, mounting_here, false);
cc53ce53
DH		 1115 if (ret < 0)
1116 return ret == -EISDIR ? 0 : ret;
1117 }
1118
1119 /* Transit to a mounted filesystem. */
1120 if (managed & DCACHE_MOUNTED) {
1121 struct vfsmount *mounted = lookup_mnt(path);
1122 if (!mounted)
1123 break;
1124 dput(path->dentry);
1125 mntput(path->mnt);
1126 path->mnt = mounted;
1127 path->dentry = dget(mounted->mnt_root);
1128 continue;
1129 }
1130
1131 /* Don't handle automount points here */
1132 break;
1133 }
1134 return 0;
1135}
1136
9875cf80
DH		 1137/*
1138 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1139 */
1140static void follow_mount(struct path *path)
1141{
1142 while (d_mountpoint(path->dentry)) {
1143 struct vfsmount *mounted = lookup_mnt(path);
1144 if (!mounted)
1145 break;
1146 dput(path->dentry);
1147 mntput(path->mnt);
1148 path->mnt = mounted;
1149 path->dentry = dget(mounted->mnt_root);
1150 }
1151}
1152
31e6b01f 1153static void follow_dotdot(struct nameidata *nd)
1da177e4 1154{
2a737871 1155 set_root(nd);
e518ddb7 1156
1da177e4 1157 while(1) {
4ac91378 1158 struct dentry *old = nd->path.dentry;
1da177e4 1159
2a737871
AV		 1160 if (nd->path.dentry == nd->root.dentry &&
1161 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1162 break;
1163 }
4ac91378 1164 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1165 /* rare case of legitimate dget_parent()... */
1166 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1167 dput(old);
1168 break;
1169 }
3088dd70 1170 if (!follow_up(&nd->path))
1da177e4 1171 break;
1da177e4 1172 }
79ed0226 1173 follow_mount(&nd->path);
31e6b01f 1174 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1175}
1176
baa03890
NP		 1177/*
1178 * Allocate a dentry with name and parent, and perform a parent
1179 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1180 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1181 * have verified that no child exists while under i_mutex.
1182 */
1183static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1184 struct qstr *name, struct nameidata *nd)
1185{
1186 struct inode *inode = parent->d_inode;
1187 struct dentry *dentry;
1188 struct dentry *old;
1189
1190 /* Don't create child dentry for a dead directory. */
1191 if (unlikely(IS_DEADDIR(inode)))
1192 return ERR_PTR(-ENOENT);
1193
1194 dentry = d_alloc(parent, name);
1195 if (unlikely(!dentry))
1196 return ERR_PTR(-ENOMEM);
1197
1198 old = inode->i_op->lookup(inode, dentry, nd);
1199 if (unlikely(old)) {
1200 dput(dentry);
1201 dentry = old;
1202 }
1203 return dentry;
1204}
1205
1da177e4
LT		 1206/*
1207 * It's more convoluted than I'd like it to be, but... it's still fairly
1208 * small and for now I'd prefer to have fast path as straight as possible.
1209 * It _is_ time-critical.
1210 */
1211static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1212 struct path *path, struct inode **inode)
1da177e4 1213{
4ac91378 1214 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1215 struct dentry *dentry, *parent = nd->path.dentry;
6e6b1bd1 1216 struct inode *dir;
9875cf80
DH		 1217 int err;
1218
3cac260a
AV		 1219 /*
1220 * See if the low-level filesystem might want
1221 * to use its own hash..
1222 */
fb045adb 1223 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
9875cf80 1224 err = parent->d_op->d_hash(parent, nd->inode, name);
3cac260a
AV		 1225 if (err < 0)
1226 return err;
1227 }
1da177e4 1228
b04f784e
NP		 1229 /*
1230 * Rename seqlock is not required here because in the off chance
1231 * of a false negative due to a concurrent rename, we're going to
1232 * do the non-racy lookup, below.
1233 */
31e6b01f
NP		 1234 if (nd->flags & LOOKUP_RCU) {
1235 unsigned seq;
1236
1237 *inode = nd->inode;
1238 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1239 if (!dentry) {
1240 if (nameidata_drop_rcu(nd))
1241 return -ECHILD;
1242 goto need_lookup;
1243 }
1244 /* Memory barrier in read_seqcount_begin of child is enough */
1245 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1246 return -ECHILD;
1247
1248 nd->seq = seq;
24643087 1249 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
f5e1c1c1 1250 dentry = do_revalidate_rcu(dentry, nd);
24643087
AV		 1251 if (!dentry)
1252 goto need_lookup;
1253 if (IS_ERR(dentry))
1254 goto fail;
1255 if (!(nd->flags & LOOKUP_RCU))
1256 goto done;
1257 }
31e6b01f
NP		 1258 path->mnt = mnt;
1259 path->dentry = dentry;
9875cf80
DH		 1260 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1261 return 0;
1262 if (nameidata_drop_rcu(nd))
1263 return -ECHILD;
1264 /* fallthru */
1265 }
1266 dentry = __d_lookup(parent, name);
1267 if (!dentry)
1268 goto need_lookup;
2e2e88ea 1269found:
24643087
AV		 1270 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1271 dentry = do_revalidate(dentry, nd);
1272 if (!dentry)
1273 goto need_lookup;
1274 if (IS_ERR(dentry))
1275 goto fail;
1276 }
1da177e4 1277done:
9875cf80
DH		 1278 path->mnt = mnt;
1279 path->dentry = dentry;
1280 err = follow_managed(path, nd->flags);
89312214
IK		 1281 if (unlikely(err < 0)) {
1282 path_put_conditional(path, nd);
9875cf80 1283 return err;
89312214 1284 }
9875cf80 1285 *inode = path->dentry->d_inode;
1da177e4
LT		 1286 return 0;
1287
1288need_lookup:
6e6b1bd1 1289 dir = parent->d_inode;
31e6b01f 1290 BUG_ON(nd->inode != dir);
6e6b1bd1
AV		 1291
1292 mutex_lock(&dir->i_mutex);
1293 /*
1294 * First re-do the cached lookup just in case it was created
b04f784e
NP		 1295 * while we waited for the directory semaphore, or the first
1296 * lookup failed due to an unrelated rename.
6e6b1bd1 1297 *
b04f784e
NP		 1298 * This could use version numbering or similar to avoid unnecessary
1299 * cache lookups, but then we'd have to do the first lookup in the
1300 * non-racy way. However in the common case here, everything should
1301 * be hot in cache, so would it be a big win?
6e6b1bd1
AV		 1302 */
1303 dentry = d_lookup(parent, name);
baa03890
NP		 1304 if (likely(!dentry)) {
1305 dentry = d_alloc_and_lookup(parent, name, nd);
6e6b1bd1
AV		 1306 mutex_unlock(&dir->i_mutex);
1307 if (IS_ERR(dentry))
1308 goto fail;
1309 goto done;
1310 }
6e6b1bd1
AV		 1311 /*
1312 * Uhhuh! Nasty case: the cache was re-populated while
1313 * we waited on the semaphore. Need to revalidate.
1314 */
1315 mutex_unlock(&dir->i_mutex);
2e2e88ea 1316 goto found;
1da177e4 1317
1da177e4
LT		 1318fail:
1319 return PTR_ERR(dentry);
1320}
1321
52094c8a
AV		 1322static inline int may_lookup(struct nameidata *nd)
1323{
1324 if (nd->flags & LOOKUP_RCU) {
1325 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1326 if (err != -ECHILD)
1327 return err;
1328 if (nameidata_drop_rcu(nd))
1329 return -ECHILD;
1330 }
1331 return exec_permission(nd->inode, 0);
1332}
1333
9856fa1b
AV		 1334static inline int handle_dots(struct nameidata *nd, int type)
1335{
1336 if (type == LAST_DOTDOT) {
1337 if (nd->flags & LOOKUP_RCU) {
1338 if (follow_dotdot_rcu(nd))
1339 return -ECHILD;
1340 } else
1341 follow_dotdot(nd);
1342 }
1343 return 0;
1344}
1345
951361f9
AV		 1346static void terminate_walk(struct nameidata *nd)
1347{
1348 if (!(nd->flags & LOOKUP_RCU)) {
1349 path_put(&nd->path);
1350 } else {
1351 nd->flags &= ~LOOKUP_RCU;
1352 nd->root.mnt = NULL;
1353 rcu_read_unlock();
1354 br_read_unlock(vfsmount_lock);
1355 }
1356}
1357
1da177e4
LT		 1358/*
1359 * Name resolution.
ea3834d9
PM		 1360 * This is the basic name resolution function, turning a pathname into
1361 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1362 *
ea3834d9
PM		 1363 * Returns 0 and nd will have valid dentry and mnt on success.
1364 * Returns error and drops reference to input namei data on failure.
1da177e4 1365 */
6de88d72 1366static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1367{
1368 struct path next;
1da177e4
LT		 1369 int err;
1370 unsigned int lookup_flags = nd->flags;
1371
1372 while (*name=='/')
1373 name++;
1374 if (!*name)
086e183a 1375 return 0;
1da177e4 1376
1da177e4 1377 if (nd->depth)
f55eab82 1378 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 1379
1380 /* At this point we know we have a real path component. */
1381 for(;;) {
31e6b01f 1382 struct inode *inode;
1da177e4
LT		 1383 unsigned long hash;
1384 struct qstr this;
1385 unsigned int c;
fe479a58 1386 int type;
1da177e4 1387
cdce5d6b 1388 nd->flags |= LOOKUP_CONTINUE;
52094c8a
AV		 1389
1390 err = may_lookup(nd);
1da177e4
LT		 1391 if (err)
1392 break;
1393
1394 this.name = name;
1395 c = *(const unsigned char *)name;
1396
1397 hash = init_name_hash();
1398 do {
1399 name++;
1400 hash = partial_name_hash(c, hash);
1401 c = *(const unsigned char *)name;
1402 } while (c && (c != '/'));
1403 this.len = name - (const char *) this.name;
1404 this.hash = end_name_hash(hash);
1405
fe479a58
AV		 1406 type = LAST_NORM;
1407 if (this.name[0] == '.') switch (this.len) {
1408 case 2:
16c2cd71 1409 if (this.name[1] == '.') {
fe479a58 1410 type = LAST_DOTDOT;
16c2cd71
AV		 1411 nd->flags |= LOOKUP_JUMPED;
1412 }
fe479a58
AV		 1413 break;
1414 case 1:
1415 type = LAST_DOT;
1416 }
16c2cd71
AV		 1417 if (likely(type == LAST_NORM))
1418 nd->flags &= ~LOOKUP_JUMPED;
fe479a58 1419
1da177e4
LT		 1420 /* remove trailing slashes? */
1421 if (!c)
1422 goto last_component;
1423 while (*++name == '/');
1424 if (!*name)
1425 goto last_with_slashes;
1426
1427 /*
1428 * "." and ".." are special - ".." especially so because it has
1429 * to be able to know about the current root directory and
1430 * parent relationships.
1431 */
fe479a58 1432 if (unlikely(type != LAST_NORM)) {
ef7562d5
AV		 1433 if (handle_dots(nd, type))
1434 return -ECHILD;
fe479a58 1435 continue;
1da177e4 1436 }
fe479a58 1437
1da177e4 1438 /* This does the actual lookups.. */
31e6b01f 1439 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1440 if (err)
1441 break;
1da177e4 1442
7bc055d1 1443 if (inode && inode->i_op->follow_link) {
3abb17e8 1444 err = do_follow_link(inode, &next, nd);
1da177e4 1445 if (err)
a7472bab 1446 return err;
31e6b01f 1447 nd->inode = nd->path.dentry->d_inode;
31e6b01f 1448 } else {
09dd17d3 1449 path_to_nameidata(&next, nd);
31e6b01f
NP		 1450 nd->inode = inode;
1451 }
7bc055d1
AV		 1452 err = -ENOENT;
1453 if (!nd->inode)
1454 break;
1da177e4 1455 err = -ENOTDIR;
31e6b01f 1456 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1457 break;
1458 continue;
1459 /* here ends the main loop */
1460
1461last_with_slashes:
1462 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1463last_component:
f55eab82
TM		 1464 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1465 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 1466 if (lookup_flags & LOOKUP_PARENT)
1467 goto lookup_parent;
ef7562d5
AV		 1468 if (unlikely(type != LAST_NORM))
1469 return handle_dots(nd, type);
31e6b01f 1470 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1471 if (err)
1472 break;
db372915
DH		 1473 if (inode && unlikely(inode->i_op->follow_link) &&
1474 (lookup_flags & LOOKUP_FOLLOW)) {
3abb17e8 1475 err = do_follow_link(inode, &next, nd);
1da177e4 1476 if (err)
a7472bab 1477 return err;
31e6b01f
NP		 1478 nd->inode = nd->path.dentry->d_inode;
1479 } else {
09dd17d3 1480 path_to_nameidata(&next, nd);
31e6b01f
NP		 1481 nd->inode = inode;
1482 }
1da177e4 1483 err = -ENOENT;
31e6b01f 1484 if (!nd->inode)
1da177e4
LT		 1485 break;
1486 if (lookup_flags & LOOKUP_DIRECTORY) {
1487 err = -ENOTDIR;
31e6b01f 1488 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1489 break;
1490 }
086e183a 1491 return 0;
1da177e4
LT		 1492lookup_parent:
1493 nd->last = this;
fe479a58 1494 nd->last_type = type;
1da177e4 1495 return 0;
1da177e4 1496 }
951361f9 1497 terminate_walk(nd);
1da177e4
LT		 1498 return err;
1499}
1500
70e9b357
AV		 1501static int path_init(int dfd, const char *name, unsigned int flags,
1502 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1503{
1504 int retval = 0;
1505 int fput_needed;
1506 struct file *file;
1507
1508 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1509 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f
NP		 1510 nd->depth = 0;
1511 nd->root.mnt = NULL;
31e6b01f
NP		 1512
1513 if (*name=='/') {
e41f7d4e
AV		 1514 if (flags & LOOKUP_RCU) {
1515 br_read_lock(vfsmount_lock);
1516 rcu_read_lock();
1517 set_root_rcu(nd);
1518 } else {
1519 set_root(nd);
1520 path_get(&nd->root);
1521 }
1522 nd->path = nd->root;
31e6b01f 1523 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1524 if (flags & LOOKUP_RCU) {
1525 struct fs_struct *fs = current->fs;
1526 unsigned seq;
31e6b01f 1527
e41f7d4e
AV		 1528 br_read_lock(vfsmount_lock);
1529 rcu_read_lock();
c28cc364 1530
e41f7d4e
AV		 1531 do {
1532 seq = read_seqcount_begin(&fs->seq);
1533 nd->path = fs->pwd;
1534 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1535 } while (read_seqcount_retry(&fs->seq, seq));
1536 } else {
1537 get_fs_pwd(current->fs, &nd->path);
1538 }
31e6b01f
NP		 1539 } else {
1540 struct dentry *dentry;
1541
1542 file = fget_light(dfd, &fput_needed);
1543 retval = -EBADF;
1544 if (!file)
1545 goto out_fail;
1546
1547 dentry = file->f_path.dentry;
1548
1549 retval = -ENOTDIR;
1550 if (!S_ISDIR(dentry->d_inode->i_mode))
1551 goto fput_fail;
1552
1553 retval = file_permission(file, MAY_EXEC);
1554 if (retval)
1555 goto fput_fail;
1556
1557 nd->path = file->f_path;
e41f7d4e
AV		 1558 if (flags & LOOKUP_RCU) {
1559 if (fput_needed)
70e9b357 1560 *fp = file;
e41f7d4e
AV		 1561 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1562 br_read_lock(vfsmount_lock);
1563 rcu_read_lock();
1564 } else {
1565 path_get(&file->f_path);
1566 fput_light(file, fput_needed);
1567 }
31e6b01f 1568 }
31e6b01f 1569
31e6b01f 1570 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1571 return 0;
2dfdd266 1572
9b4a9b14
AV		 1573fput_fail:
1574 fput_light(file, fput_needed);
1575out_fail:
1576 return retval;
1577}
1578
1579/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1580static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1581 unsigned int flags, struct nameidata *nd)
1582{
70e9b357 1583 struct file *base = NULL;
31e6b01f
NP		 1584 int retval;
1585
1586 /*
1587 * Path walking is largely split up into 2 different synchronisation
1588 * schemes, rcu-walk and ref-walk (explained in
1589 * Documentation/filesystems/path-lookup.txt). These share much of the
1590 * path walk code, but some things particularly setup, cleanup, and
1591 * following mounts are sufficiently divergent that functions are
1592 * duplicated. Typically there is a function foo(), and its RCU
1593 * analogue, foo_rcu().
1594 *
1595 * -ECHILD is the error number of choice (just to avoid clashes) that
1596 * is returned if some aspect of an rcu-walk fails. Such an error must
1597 * be handled by restarting a traditional ref-walk (which will always
1598 * be able to complete).
1599 */
70e9b357 1600 retval = path_init(dfd, name, flags, nd, &base);
ee0827cd 1601
31e6b01f
NP		 1602 if (unlikely(retval))
1603 return retval;
ee0827cd
AV		 1604
1605 current->total_link_count = 0;
1606 retval = link_path_walk(name, nd);
1607
1608 if (nd->flags & LOOKUP_RCU) {
4455ca62
AV		 1609 /* went all way through without dropping RCU */
1610 BUG_ON(retval);
1611 if (nameidata_drop_rcu_last(nd))
1612 retval = -ECHILD;
ee0827cd
AV		 1613 }
1614
16c2cd71
AV		 1615 if (!retval)
1616 retval = handle_reval_path(nd);
1617
70e9b357
AV		 1618 if (base)
1619 fput(base);
ee0827cd 1620
2a737871
AV		 1621 if (nd->root.mnt) {
1622 path_put(&nd->root);
1623 nd->root.mnt = NULL;
1624 }
ee0827cd
AV		 1625 return retval;
1626}
31e6b01f 1627
ee0827cd
AV		 1628static int do_path_lookup(int dfd, const char *name,
1629 unsigned int flags, struct nameidata *nd)
1630{
1631 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1632 if (unlikely(retval == -ECHILD))
1633 retval = path_lookupat(dfd, name, flags, nd);
1634 if (unlikely(retval == -ESTALE))
1635 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1636
1637 if (likely(!retval)) {
1638 if (unlikely(!audit_dummy_context())) {
1639 if (nd->path.dentry && nd->inode)
1640 audit_inode(name, nd->path.dentry);
1641 }
1642 }
170aa3d0 1643 return retval;
1da177e4
LT		 1644}
1645
c9c6cac0 1646int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1647{
c9c6cac0 1648 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1649}
1650
d1811465
AV		 1651int kern_path(const char *name, unsigned int flags, struct path *path)
1652{
1653 struct nameidata nd;
1654 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1655 if (!res)
1656 *path = nd.path;
1657 return res;
1658}
1659
16f18200
JJS		 1660/**
1661 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1662 * @dentry: pointer to dentry of the base directory
1663 * @mnt: pointer to vfs mount of the base directory
1664 * @name: pointer to file name
1665 * @flags: lookup flags
1666 * @nd: pointer to nameidata
1667 */
1668int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1669 const char *name, unsigned int flags,
1670 struct nameidata *nd)
1671{
ee0827cd 1672 int result;
16f18200
JJS		 1673
1674 /* same as do_path_lookup */
1675 nd->last_type = LAST_ROOT;
16c2cd71 1676 nd->flags = flags | LOOKUP_JUMPED;
16f18200
JJS		 1677 nd->depth = 0;
1678
c8e7f449
JB		 1679 nd->path.dentry = dentry;
1680 nd->path.mnt = mnt;
1681 path_get(&nd->path);
5b857119
AV		 1682 nd->root = nd->path;
1683 path_get(&nd->root);
31e6b01f 1684 nd->inode = nd->path.dentry->d_inode;
16f18200 1685
ee0827cd
AV		 1686 current->total_link_count = 0;
1687
1688 result = link_path_walk(name, nd);
16c2cd71
AV		 1689 if (!result)
1690 result = handle_reval_path(nd);
ee0827cd
AV		 1691 if (result == -ESTALE) {
1692 /* nd->path had been dropped */
1693 current->total_link_count = 0;
1694 nd->path.dentry = dentry;
1695 nd->path.mnt = mnt;
1696 nd->inode = dentry->d_inode;
1697 path_get(&nd->path);
16c2cd71
AV		 1698 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_REVAL;
1699
ee0827cd 1700 result = link_path_walk(name, nd);
16c2cd71
AV		 1701 if (!result)
1702 result = handle_reval_path(nd);
ee0827cd
AV		 1703 }
1704 if (unlikely(!result && !audit_dummy_context() && nd->path.dentry &&
31e6b01f 1705 nd->inode))
4ac91378 1706 audit_inode(name, nd->path.dentry);
16f18200 1707
5b857119
AV		 1708 path_put(&nd->root);
1709 nd->root.mnt = NULL;
16f18200 1710
ee0827cd 1711 return result;
16f18200
JJS		 1712}
1713
eead1911
CH		 1714static struct dentry *__lookup_hash(struct qstr *name,
1715 struct dentry *base, struct nameidata *nd)
1da177e4 1716{
81fca444 1717 struct inode *inode = base->d_inode;
057f6c01 1718 struct dentry *dentry;
1da177e4
LT		 1719 int err;
1720
b74c79e9 1721 err = exec_permission(inode, 0);
81fca444
CH		 1722 if (err)
1723 return ERR_PTR(err);
1da177e4
LT		 1724
1725 /*
1726 * See if the low-level filesystem might want
1727 * to use its own hash..
1728 */
fb045adb 1729 if (base->d_flags & DCACHE_OP_HASH) {
b1e6a015 1730 err = base->d_op->d_hash(base, inode, name);
1da177e4
LT		 1731 dentry = ERR_PTR(err);
1732 if (err < 0)
1733 goto out;
1734 }
1735
b04f784e
NP		 1736 /*
1737 * Don't bother with __d_lookup: callers are for creat as
1738 * well as unlink, so a lot of the time it would cost
1739 * a double lookup.
6e6b1bd1 1740 */
b04f784e 1741 dentry = d_lookup(base, name);
6e6b1bd1 1742
fb045adb 1743 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1744 dentry = do_revalidate(dentry, nd);
1745
baa03890
NP		 1746 if (!dentry)
1747 dentry = d_alloc_and_lookup(base, name, nd);
1da177e4
LT		 1748out:
1749 return dentry;
1750}
1751
057f6c01
JM		 1752/*
1753 * Restricted form of lookup. Doesn't follow links, single-component only,
1754 * needs parent already locked. Doesn't follow mounts.
1755 * SMP-safe.
1756 */
eead1911 1757static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1758{
4ac91378 1759 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1760}
1761
eead1911 1762/**
a6b91919 1763 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1764 * @name: pathname component to lookup
1765 * @base: base directory to lookup from
1766 * @len: maximum length @len should be interpreted to
1767 *
a6b91919
RD		 1768 * Note that this routine is purely a helper for filesystem usage and should
1769 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1770 * nameidata argument is passed to the filesystem methods and a filesystem
1771 * using this helper needs to be prepared for that.
1772 */
057f6c01
JM		 1773struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1774{
057f6c01 1775 struct qstr this;
6a96ba54
AV		 1776 unsigned long hash;
1777 unsigned int c;
057f6c01 1778
2f9092e1
DW		 1779 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1780
6a96ba54
AV		 1781 this.name = name;
1782 this.len = len;
1783 if (!len)
1784 return ERR_PTR(-EACCES);
1785
1786 hash = init_name_hash();
1787 while (len--) {
1788 c = *(const unsigned char *)name++;
1789 if (c == '/' || c == '\0')
1790 return ERR_PTR(-EACCES);
1791 hash = partial_name_hash(c, hash);
1792 }
1793 this.hash = end_name_hash(hash);
eead1911 1794
49705b77 1795 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1796}
1797
2d8f3038
AV		 1798int user_path_at(int dfd, const char __user *name, unsigned flags,
1799 struct path *path)
1da177e4 1800{
2d8f3038 1801 struct nameidata nd;
1da177e4
LT		 1802 char *tmp = getname(name);
1803 int err = PTR_ERR(tmp);
1da177e4 1804 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1805
1806 BUG_ON(flags & LOOKUP_PARENT);
1807
1808 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1809 putname(tmp);
2d8f3038
AV		 1810 if (!err)
1811 *path = nd.path;
1da177e4
LT		 1812 }
1813 return err;
1814}
1815
2ad94ae6
AV		 1816static int user_path_parent(int dfd, const char __user *path,
1817 struct nameidata *nd, char **name)
1818{
1819 char *s = getname(path);
1820 int error;
1821
1822 if (IS_ERR(s))
1823 return PTR_ERR(s);
1824
1825 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1826 if (error)
1827 putname(s);
1828 else
1829 *name = s;
1830
1831 return error;
1832}
1833
1da177e4
LT		 1834/*
1835 * It's inline, so penalty for filesystems that don't use sticky bit is
1836 * minimal.
1837 */
1838static inline int check_sticky(struct inode *dir, struct inode *inode)
1839{
da9592ed
DH		 1840 uid_t fsuid = current_fsuid();
1841
1da177e4
LT		 1842 if (!(dir->i_mode & S_ISVTX))
1843 return 0;
da9592ed 1844 if (inode->i_uid == fsuid)
1da177e4 1845 return 0;
da9592ed 1846 if (dir->i_uid == fsuid)
1da177e4
LT		 1847 return 0;
1848 return !capable(CAP_FOWNER);
1849}
1850
1851/*
1852 * Check whether we can remove a link victim from directory dir, check
1853 * whether the type of victim is right.
1854 * 1. We can't do it if dir is read-only (done in permission())
1855 * 2. We should have write and exec permissions on dir
1856 * 3. We can't remove anything from append-only dir
1857 * 4. We can't do anything with immutable dir (done in permission())
1858 * 5. If the sticky bit on dir is set we should either
1859 * a. be owner of dir, or
1860 * b. be owner of victim, or
1861 * c. have CAP_FOWNER capability
1862 * 6. If the victim is append-only or immutable we can't do antyhing with
1863 * links pointing to it.
1864 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1865 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1866 * 9. We can't remove a root or mountpoint.
1867 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1868 * nfs_async_unlink().
1869 */
858119e1 1870static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1871{
1872 int error;
1873
1874 if (!victim->d_inode)
1875 return -ENOENT;
1876
1877 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1878 audit_inode_child(victim, dir);
1da177e4 1879
f419a2e3 1880 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1881 if (error)
1882 return error;
1883 if (IS_APPEND(dir))
1884 return -EPERM;
1885 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1886 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1887 return -EPERM;
1888 if (isdir) {
1889 if (!S_ISDIR(victim->d_inode->i_mode))
1890 return -ENOTDIR;
1891 if (IS_ROOT(victim))
1892 return -EBUSY;
1893 } else if (S_ISDIR(victim->d_inode->i_mode))
1894 return -EISDIR;
1895 if (IS_DEADDIR(dir))
1896 return -ENOENT;
1897 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1898 return -EBUSY;
1899 return 0;
1900}
1901
1902/* Check whether we can create an object with dentry child in directory
1903 * dir.
1904 * 1. We can't do it if child already exists (open has special treatment for
1905 * this case, but since we are inlined it's OK)
1906 * 2. We can't do it if dir is read-only (done in permission())
1907 * 3. We should have write and exec permissions on dir
1908 * 4. We can't do it if dir is immutable (done in permission())
1909 */
a95164d9 1910static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1911{
1912 if (child->d_inode)
1913 return -EEXIST;
1914 if (IS_DEADDIR(dir))
1915 return -ENOENT;
f419a2e3 1916 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1917}
1918
1da177e4
LT		 1919/*
1920 * p1 and p2 should be directories on the same fs.
1921 */
1922struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1923{
1924 struct dentry *p;
1925
1926 if (p1 == p2) {
f2eace23 1927 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1928 return NULL;
1929 }
1930
a11f3a05 1931 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1932
e2761a11
OH		 1933 p = d_ancestor(p2, p1);
1934 if (p) {
1935 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1936 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1937 return p;
1da177e4
LT		 1938 }
1939
e2761a11
OH		 1940 p = d_ancestor(p1, p2);
1941 if (p) {
1942 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1943 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1944 return p;
1da177e4
LT		 1945 }
1946
f2eace23
IM		 1947 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1948 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1949 return NULL;
1950}
1951
1952void unlock_rename(struct dentry *p1, struct dentry *p2)
1953{
1b1dcc1b 1954 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1955 if (p1 != p2) {
1b1dcc1b 1956 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1957 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1958 }
1959}
1960
1961int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1962 struct nameidata *nd)
1963{
a95164d9 1964 int error = may_create(dir, dentry);
1da177e4
LT		 1965
1966 if (error)
1967 return error;
1968
acfa4380 1969 if (!dir->i_op->create)
1da177e4
LT		 1970 return -EACCES; /* shouldn't it be ENOSYS? */
1971 mode &= S_IALLUGO;
1972 mode |= S_IFREG;
1973 error = security_inode_create(dir, dentry, mode);
1974 if (error)
1975 return error;
1da177e4 1976 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1977 if (!error)
f38aa942 1978 fsnotify_create(dir, dentry);
1da177e4
LT		 1979 return error;
1980}
1981
3fb64190 1982int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1983{
3fb64190 1984 struct dentry *dentry = path->dentry;
1da177e4
LT		 1985 struct inode *inode = dentry->d_inode;
1986 int error;
1987
1988 if (!inode)
1989 return -ENOENT;
1990
c8fe8f30
CH		 1991 switch (inode->i_mode & S_IFMT) {
1992 case S_IFLNK:
1da177e4 1993 return -ELOOP;
c8fe8f30
CH		 1994 case S_IFDIR:
1995 if (acc_mode & MAY_WRITE)
1996 return -EISDIR;
1997 break;
1998 case S_IFBLK:
1999 case S_IFCHR:
3fb64190 2000 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2001 return -EACCES;
c8fe8f30
CH		 2002 /*FALLTHRU*/
2003 case S_IFIFO:
2004 case S_IFSOCK:
1da177e4 2005 flag &= ~O_TRUNC;
c8fe8f30 2006 break;
4a3fd211 2007 }
b41572e9 2008
3fb64190 2009 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2010 if (error)
2011 return error;
6146f0d5 2012
1da177e4
LT		 2013 /*
2014 * An append-only file must be opened in append mode for writing.
2015 */
2016 if (IS_APPEND(inode)) {
8737c930 2017 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2018 return -EPERM;
1da177e4 2019 if (flag & O_TRUNC)
7715b521 2020 return -EPERM;
1da177e4
LT		 2021 }
2022
2023 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 2024 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2025 return -EPERM;
1da177e4
LT		 2026
2027 /*
2028 * Ensure there are no outstanding leases on the file.
2029 */
b65a9cfc 2030 return break_lease(inode, flag);
7715b521 2031}
1da177e4 2032
e1181ee6 2033static int handle_truncate(struct file *filp)
7715b521 2034{
e1181ee6 2035 struct path *path = &filp->f_path;
7715b521
AV		 2036 struct inode *inode = path->dentry->d_inode;
2037 int error = get_write_access(inode);
2038 if (error)
2039 return error;
2040 /*
2041 * Refuse to truncate files with mandatory locks held on them.
2042 */
2043 error = locks_verify_locked(inode);
2044 if (!error)
ea0d3ab2 2045 error = security_path_truncate(path);
7715b521
AV		 2046 if (!error) {
2047 error = do_truncate(path->dentry, 0,
2048 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2049 filp);
7715b521
AV		 2050 }
2051 put_write_access(inode);
acd0c935 2052 return error;
1da177e4
LT		 2053}
2054
d57999e1
DH		 2055/*
2056 * Be careful about ever adding any more callers of this
2057 * function. Its flags must be in the namei format, not
2058 * what get passed to sys_open().
2059 */
2060static int __open_namei_create(struct nameidata *nd, struct path *path,
8737c930 2061 int open_flag, int mode)
aab520e2
DH		 2062{
2063 int error;
4ac91378 2064 struct dentry *dir = nd->path.dentry;
aab520e2
DH		 2065
2066 if (!IS_POSIXACL(dir->d_inode))
ce3b0f8d 2067 mode &= ~current_umask();
be6d3e56
KT		 2068 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
2069 if (error)
2070 goto out_unlock;
aab520e2 2071 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
be6d3e56 2072out_unlock:
aab520e2 2073 mutex_unlock(&dir->d_inode->i_mutex);
4ac91378
JB		 2074 dput(nd->path.dentry);
2075 nd->path.dentry = path->dentry;
31e6b01f 2076
aab520e2
DH		 2077 if (error)
2078 return error;
2079 /* Don't check for write permission, don't truncate */
8737c930 2080 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
aab520e2
DH		 2081}
2082
d57999e1
DH		 2083/*
2084 * Note that while the flag value (low two bits) for sys_open means:
2085 * 00 - read-only
2086 * 01 - write-only
2087 * 10 - read-write
2088 * 11 - special
2089 * it is changed into
2090 * 00 - no permissions needed
2091 * 01 - read-permission
2092 * 10 - write-permission
2093 * 11 - read-write
2094 * for the internal routines (ie open_namei()/follow_link() etc)
2095 * This is more logical, and also allows the 00 "no perm needed"
2096 * to be used for symlinks (where the permissions are checked
2097 * later).
2098 *
2099*/
2100static inline int open_to_namei_flags(int flag)
2101{
2102 if ((flag+1) & O_ACCMODE)
2103 flag++;
2104 return flag;
2105}
2106
7715b521 2107static int open_will_truncate(int flag, struct inode *inode)
4a3fd211
DH		 2108{
2109 /*
2110 * We'll never write to the fs underlying
2111 * a device file.
2112 */
2113 if (special_file(inode->i_mode))
2114 return 0;
2115 return (flag & O_TRUNC);
2116}
2117
648fa861 2118static struct file *finish_open(struct nameidata *nd,
9a66179e 2119 int open_flag, int acc_mode)
648fa861
AV		 2120{
2121 struct file *filp;
2122 int will_truncate;
2123 int error;
2124
9a66179e 2125 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
648fa861
AV		 2126 if (will_truncate) {
2127 error = mnt_want_write(nd->path.mnt);
2128 if (error)
2129 goto exit;
2130 }
2131 error = may_open(&nd->path, acc_mode, open_flag);
2132 if (error) {
2133 if (will_truncate)
2134 mnt_drop_write(nd->path.mnt);
2135 goto exit;
2136 }
2137 filp = nameidata_to_filp(nd);
2138 if (!IS_ERR(filp)) {
2139 error = ima_file_check(filp, acc_mode);
2140 if (error) {
2141 fput(filp);
2142 filp = ERR_PTR(error);
2143 }
2144 }
2145 if (!IS_ERR(filp)) {
648fa861 2146 if (will_truncate) {
e1181ee6 2147 error = handle_truncate(filp);
648fa861
AV		 2148 if (error) {
2149 fput(filp);
2150 filp = ERR_PTR(error);
2151 }
2152 }
2153 }
2154 /*
2155 * It is now safe to drop the mnt write
2156 * because the filp has had a write taken
2157 * on its behalf.
2158 */
2159 if (will_truncate)
2160 mnt_drop_write(nd->path.mnt);
d893f1bc 2161 path_put(&nd->path);
648fa861
AV		 2162 return filp;
2163
2164exit:
648fa861
AV		 2165 path_put(&nd->path);
2166 return ERR_PTR(error);
2167}
2168
31e6b01f 2169/*
fe2d35ff 2170 * Handle the last step of open()
31e6b01f 2171 */
fb1cc555 2172static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2173 const struct open_flags *op, const char *pathname)
fb1cc555 2174{
a1e28038 2175 struct dentry *dir = nd->path.dentry;
fb1cc555 2176 struct file *filp;
fe2d35ff 2177 struct inode *inode;
16c2cd71 2178 int error;
1f36f774 2179
c3e380b0
AV		 2180 nd->flags &= ~LOOKUP_PARENT;
2181 nd->flags |= op->intent;
2182
1f36f774
AV		 2183 switch (nd->last_type) {
2184 case LAST_DOTDOT:
176306f5 2185 case LAST_DOT:
fe2d35ff
AV		 2186 error = handle_dots(nd, nd->last_type);
2187 if (error)
2188 return ERR_PTR(error);
1f36f774 2189 /* fallthrough */
1f36f774 2190 case LAST_ROOT:
fe2d35ff
AV		 2191 if (nd->flags & LOOKUP_RCU) {
2192 if (nameidata_drop_rcu_last(nd))
2193 return ERR_PTR(-ECHILD);
2194 }
16c2cd71
AV		 2195 error = handle_reval_path(nd);
2196 if (error)
2197 goto exit;
fe2d35ff
AV		 2198 audit_inode(pathname, nd->path.dentry);
2199 if (op->open_flag & O_CREAT) {
2200 error = -EISDIR;
2201 goto exit;
2202 }
2203 goto ok;
1f36f774 2204 case LAST_BIND:
fe2d35ff 2205 /* can't be RCU mode here */
16c2cd71
AV		 2206 error = handle_reval_path(nd);
2207 if (error)
2208 goto exit;
1f36f774 2209 audit_inode(pathname, dir);
67ee3ad2 2210 goto ok;
1f36f774 2211 }
67ee3ad2 2212
fe2d35ff
AV		 2213 if (!(op->open_flag & O_CREAT)) {
2214 if (nd->last.name[nd->last.len])
2215 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2216 /* we _can_ be in RCU mode here */
2217 error = do_lookup(nd, &nd->last, path, &inode);
2218 if (error) {
2219 terminate_walk(nd);
2220 return ERR_PTR(error);
2221 }
2222 if (!inode) {
2223 path_to_nameidata(path, nd);
2224 terminate_walk(nd);
2225 return ERR_PTR(-ENOENT);
2226 }
2227 if (unlikely(inode->i_op->follow_link)) {
2228 /* We drop rcu-walk here */
2229 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
2230 return ERR_PTR(-ECHILD);
2231 return NULL;
2232 }
2233 path_to_nameidata(path, nd);
2234 nd->inode = inode;
2235 /* sayonara */
2236 if (nd->flags & LOOKUP_RCU) {
2237 if (nameidata_drop_rcu_last(nd))
2238 return ERR_PTR(-ECHILD);
2239 }
2240
2241 error = -ENOTDIR;
2242 if (nd->flags & LOOKUP_DIRECTORY) {
2243 if (!inode->i_op->lookup)
2244 goto exit;
2245 }
2246 audit_inode(pathname, nd->path.dentry);
2247 goto ok;
2248 }
2249
2250 /* create side of things */
2251
2252 if (nd->flags & LOOKUP_RCU) {
2253 if (nameidata_drop_rcu_last(nd))
2254 return ERR_PTR(-ECHILD);
2255 }
2256
2257 audit_inode(pathname, dir);
16c2cd71 2258 error = -EISDIR;
1f36f774 2259 /* trailing slashes? */
31e6b01f
NP		 2260 if (nd->last.name[nd->last.len])
2261 goto exit;
a2c36b45 2262
a1e28038
AV		 2263 mutex_lock(&dir->d_inode->i_mutex);
2264
2265 path->dentry = lookup_hash(nd);
2266 path->mnt = nd->path.mnt;
2267
fb1cc555
AV		 2268 error = PTR_ERR(path->dentry);
2269 if (IS_ERR(path->dentry)) {
2270 mutex_unlock(&dir->d_inode->i_mutex);
2271 goto exit;
2272 }
2273
2274 if (IS_ERR(nd->intent.open.file)) {
2275 error = PTR_ERR(nd->intent.open.file);
2276 goto exit_mutex_unlock;
2277 }
2278
2279 /* Negative dentry, just create the file */
2280 if (!path->dentry->d_inode) {
2281 /*
2282 * This write is needed to ensure that a
2283 * ro->rw transition does not occur between
2284 * the time when the file is created and when
2285 * a permanent write count is taken through
2286 * the 'struct file' in nameidata_to_filp().
2287 */
2288 error = mnt_want_write(nd->path.mnt);
2289 if (error)
2290 goto exit_mutex_unlock;
c3e380b0 2291 error = __open_namei_create(nd, path, op->open_flag, op->mode);
fb1cc555
AV		 2292 if (error) {
2293 mnt_drop_write(nd->path.mnt);
2294 goto exit;
2295 }
2296 filp = nameidata_to_filp(nd);
2297 mnt_drop_write(nd->path.mnt);
d893f1bc 2298 path_put(&nd->path);
fb1cc555 2299 if (!IS_ERR(filp)) {
c3e380b0 2300 error = ima_file_check(filp, op->acc_mode);
fb1cc555
AV		 2301 if (error) {
2302 fput(filp);
2303 filp = ERR_PTR(error);
2304 }
2305 }
2306 return filp;
2307 }
2308
2309 /*
2310 * It already exists.
2311 */
2312 mutex_unlock(&dir->d_inode->i_mutex);
2313 audit_inode(pathname, path->dentry);
2314
2315 error = -EEXIST;
c3e380b0 2316 if (op->open_flag & O_EXCL)
fb1cc555
AV		 2317 goto exit_dput;
2318
9875cf80
DH		 2319 error = follow_managed(path, nd->flags);
2320 if (error < 0)
2321 goto exit_dput;
fb1cc555
AV		 2322
2323 error = -ENOENT;
2324 if (!path->dentry->d_inode)
2325 goto exit_dput;
9e67f361
AV		 2326
2327 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2328 return NULL;
fb1cc555
AV		 2329
2330 path_to_nameidata(path, nd);
31e6b01f 2331 nd->inode = path->dentry->d_inode;
fb1cc555 2332 error = -EISDIR;
31e6b01f 2333 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2334 goto exit;
67ee3ad2 2335ok:
c3e380b0 2336 filp = finish_open(nd, op->open_flag, op->acc_mode);
fb1cc555
AV		 2337 return filp;
2338
2339exit_mutex_unlock:
2340 mutex_unlock(&dir->d_inode->i_mutex);
2341exit_dput:
2342 path_put_conditional(path, nd);
2343exit:
fb1cc555
AV		 2344 path_put(&nd->path);
2345 return ERR_PTR(error);
2346}
2347
13aab428 2348static struct file *path_openat(int dfd, const char *pathname,
47c805dc 2349 const struct open_flags *op, int flags)
1da177e4 2350{
fe2d35ff 2351 struct file *base = NULL;
4a3fd211 2352 struct file *filp;
a70e65df 2353 struct nameidata nd;
9850c056 2354 struct path path;
1da177e4 2355 int count = 0;
13aab428 2356 int error;
31e6b01f
NP		 2357
2358 filp = get_empty_filp();
2359 if (!filp)
2360 return ERR_PTR(-ENFILE);
2361
47c805dc 2362 filp->f_flags = op->open_flag;
31e6b01f 2363 nd.intent.open.file = filp;
47c805dc
AV		 2364 nd.intent.open.flags = open_to_namei_flags(op->open_flag);
2365 nd.intent.open.create_mode = op->mode;
31e6b01f 2366
fe2d35ff 2367 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, &nd, &base);
31e6b01f 2368 if (unlikely(error))
13aab428 2369 goto out_filp;
31e6b01f 2370
fe2d35ff
AV		 2371 current->total_link_count = 0;
2372 error = link_path_walk(pathname, &nd);
31e6b01f
NP		 2373 if (unlikely(error))
2374 goto out_filp;
1da177e4 2375
47c805dc 2376 filp = do_last(&nd, &path, op, pathname);
806b681c 2377 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa
NP		 2378 struct path link = path;
2379 struct inode *linki = link.dentry->d_inode;
def4af30 2380 void *cookie;
806b681c 2381 error = -ELOOP;
db372915 2382 if (!(nd.flags & LOOKUP_FOLLOW))
1f36f774
AV		 2383 goto exit_dput;
2384 if (count++ == 32)
806b681c
AV		 2385 goto exit_dput;
2386 /*
2387 * This is subtle. Instead of calling do_follow_link() we do
2388 * the thing by hands. The reason is that this way we have zero
2389 * link_count and path_walk() (called from ->follow_link)
2390 * honoring LOOKUP_PARENT. After that we have the parent and
2391 * last component, i.e. we are in the same situation as after
2392 * the first path_walk(). Well, almost - if the last component
2393 * is normal we get its copy stored in nd->last.name and we will
2394 * have to putname() it when we are done. Procfs-like symlinks
2395 * just set LAST_BIND.
2396 */
2397 nd.flags |= LOOKUP_PARENT;
c3e380b0 2398 nd.flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
7b9337aa 2399 error = __do_follow_link(&link, &nd, &cookie);
c3e380b0 2400 if (unlikely(error))
f1afe9ef 2401 filp = ERR_PTR(error);
c3e380b0 2402 else
47c805dc 2403 filp = do_last(&nd, &path, op, pathname);
f1afe9ef 2404 if (!IS_ERR(cookie) && linki->i_op->put_link)
7b9337aa
NP		 2405 linki->i_op->put_link(link.dentry, &nd, cookie);
2406 path_put(&link);
806b681c 2407 }
10fa8e62 2408out:
2a737871
AV		 2409 if (nd.root.mnt)
2410 path_put(&nd.root);
fe2d35ff
AV		 2411 if (base)
2412 fput(base);
2dab5974 2413 release_open_intent(&nd);
10fa8e62 2414 return filp;
1da177e4 2415
806b681c
AV		 2416exit_dput:
2417 path_put_conditional(&path, &nd);
31e6b01f
NP		 2418 path_put(&nd.path);
2419out_filp:
806b681c 2420 filp = ERR_PTR(error);
10fa8e62 2421 goto out;
1da177e4
LT		 2422}
2423
13aab428
AV		 2424struct file *do_filp_open(int dfd, const char *pathname,
2425 const struct open_flags *op, int flags)
2426{
2427 struct file *filp;
2428
2429 filp = path_openat(dfd, pathname, op, flags | LOOKUP_RCU);
2430 if (unlikely(filp == ERR_PTR(-ECHILD)))
2431 filp = path_openat(dfd, pathname, op, flags);
2432 if (unlikely(filp == ERR_PTR(-ESTALE)))
2433 filp = path_openat(dfd, pathname, op, flags | LOOKUP_REVAL);
2434 return filp;
2435}
2436
1da177e4
LT		 2437/**
2438 * lookup_create - lookup a dentry, creating it if it doesn't exist
2439 * @nd: nameidata info
2440 * @is_dir: directory flag
2441 *
2442 * Simple function to lookup and return a dentry and create it
2443 * if it doesn't exist. Is SMP-safe.
c663e5d8 2444 *
4ac91378 2445 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2446 */
2447struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2448{
c663e5d8 2449 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2450
4ac91378 2451 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2452 /*
2453 * Yucky last component or no last component at all?
2454 * (foo/., foo/.., /////)
2455 */
1da177e4
LT		 2456 if (nd->last_type != LAST_NORM)
2457 goto fail;
2458 nd->flags &= ~LOOKUP_PARENT;
3516586a 2459 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2460 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2461
2462 /*
2463 * Do the final lookup.
2464 */
49705b77 2465 dentry = lookup_hash(nd);
1da177e4
LT		 2466 if (IS_ERR(dentry))
2467 goto fail;
c663e5d8 2468
e9baf6e5
AV		 2469 if (dentry->d_inode)
2470 goto eexist;
c663e5d8
CH		 2471 /*
2472 * Special case - lookup gave negative, but... we had foo/bar/
2473 * From the vfs_mknod() POV we just have a negative dentry -
2474 * all is fine. Let's be bastards - you had / on the end, you've
2475 * been asking for (non-existent) directory. -ENOENT for you.
2476 */
e9baf6e5
AV		 2477 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2478 dput(dentry);
2479 dentry = ERR_PTR(-ENOENT);
2480 }
1da177e4 2481 return dentry;
e9baf6e5 2482eexist:
1da177e4 2483 dput(dentry);
e9baf6e5 2484 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2485fail:
2486 return dentry;
2487}
f81a0bff 2488EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2489
2490int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2491{
a95164d9 2492 int error = may_create(dir, dentry);
1da177e4
LT		 2493
2494 if (error)
2495 return error;
2496
2497 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2498 return -EPERM;
2499
acfa4380 2500 if (!dir->i_op->mknod)
1da177e4
LT		 2501 return -EPERM;
2502
08ce5f16
SH		 2503 error = devcgroup_inode_mknod(mode, dev);
2504 if (error)
2505 return error;
2506
1da177e4
LT		 2507 error = security_inode_mknod(dir, dentry, mode, dev);
2508 if (error)
2509 return error;
2510
1da177e4 2511 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2512 if (!error)
f38aa942 2513 fsnotify_create(dir, dentry);
1da177e4
LT		 2514 return error;
2515}
2516
463c3197
DH		 2517static int may_mknod(mode_t mode)
2518{
2519 switch (mode & S_IFMT) {
2520 case S_IFREG:
2521 case S_IFCHR:
2522 case S_IFBLK:
2523 case S_IFIFO:
2524 case S_IFSOCK:
2525 case 0: /* zero mode translates to S_IFREG */
2526 return 0;
2527 case S_IFDIR:
2528 return -EPERM;
2529 default:
2530 return -EINVAL;
2531 }
2532}
2533
2e4d0924
HC		 2534SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2535 unsigned, dev)
1da177e4 2536{
2ad94ae6
AV		 2537 int error;
2538 char *tmp;
2539 struct dentry *dentry;
1da177e4
LT		 2540 struct nameidata nd;
2541
2542 if (S_ISDIR(mode))
2543 return -EPERM;
1da177e4 2544
2ad94ae6 2545 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2546 if (error)
2ad94ae6
AV		 2547 return error;
2548
1da177e4 2549 dentry = lookup_create(&nd, 0);
463c3197
DH		 2550 if (IS_ERR(dentry)) {
2551 error = PTR_ERR(dentry);
2552 goto out_unlock;
2553 }
4ac91378 2554 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2555 mode &= ~current_umask();
463c3197
DH		 2556 error = may_mknod(mode);
2557 if (error)
2558 goto out_dput;
2559 error = mnt_want_write(nd.path.mnt);
2560 if (error)
2561 goto out_dput;
be6d3e56
KT		 2562 error = security_path_mknod(&nd.path, dentry, mode, dev);
2563 if (error)
2564 goto out_drop_write;
463c3197 2565 switch (mode & S_IFMT) {
1da177e4 2566 case 0: case S_IFREG:
4ac91378 2567 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2568 break;
2569 case S_IFCHR: case S_IFBLK:
4ac91378 2570 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2571 new_decode_dev(dev));
2572 break;
2573 case S_IFIFO: case S_IFSOCK:
4ac91378 2574 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2575 break;
1da177e4 2576 }
be6d3e56 2577out_drop_write:
463c3197
DH		 2578 mnt_drop_write(nd.path.mnt);
2579out_dput:
2580 dput(dentry);
2581out_unlock:
4ac91378 2582 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2583 path_put(&nd.path);
1da177e4
LT		 2584 putname(tmp);
2585
2586 return error;
2587}
2588
3480b257 2589SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2590{
2591 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2592}
2593
1da177e4
LT		 2594int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2595{
a95164d9 2596 int error = may_create(dir, dentry);
1da177e4
LT		 2597
2598 if (error)
2599 return error;
2600
acfa4380 2601 if (!dir->i_op->mkdir)
1da177e4
LT		 2602 return -EPERM;
2603
2604 mode &= (S_IRWXUGO|S_ISVTX);
2605 error = security_inode_mkdir(dir, dentry, mode);
2606 if (error)
2607 return error;
2608
1da177e4 2609 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2610 if (!error)
f38aa942 2611 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2612 return error;
2613}
2614
2e4d0924 2615SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2616{
2617 int error = 0;
2618 char * tmp;
6902d925
DH		 2619 struct dentry *dentry;
2620 struct nameidata nd;
1da177e4 2621
2ad94ae6
AV		 2622 error = user_path_parent(dfd, pathname, &nd, &tmp);
2623 if (error)
6902d925 2624 goto out_err;
1da177e4 2625
6902d925
DH		 2626 dentry = lookup_create(&nd, 1);
2627 error = PTR_ERR(dentry);
2628 if (IS_ERR(dentry))
2629 goto out_unlock;
1da177e4 2630
4ac91378 2631 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2632 mode &= ~current_umask();
463c3197
DH		 2633 error = mnt_want_write(nd.path.mnt);
2634 if (error)
2635 goto out_dput;
be6d3e56
KT		 2636 error = security_path_mkdir(&nd.path, dentry, mode);
2637 if (error)
2638 goto out_drop_write;
4ac91378 2639 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2640out_drop_write:
463c3197
DH		 2641 mnt_drop_write(nd.path.mnt);
2642out_dput:
6902d925
DH		 2643 dput(dentry);
2644out_unlock:
4ac91378 2645 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2646 path_put(&nd.path);
6902d925
DH		 2647 putname(tmp);
2648out_err:
1da177e4
LT		 2649 return error;
2650}
2651
3cdad428 2652SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2653{
2654 return sys_mkdirat(AT_FDCWD, pathname, mode);
2655}
2656
1da177e4
LT		 2657/*
2658 * We try to drop the dentry early: we should have
2659 * a usage count of 2 if we're the only user of this
2660 * dentry, and if that is true (possibly after pruning
2661 * the dcache), then we drop the dentry now.
2662 *
2663 * A low-level filesystem can, if it choses, legally
2664 * do a
2665 *
2666 * if (!d_unhashed(dentry))
2667 * return -EBUSY;
2668 *
2669 * if it cannot handle the case of removing a directory
2670 * that is still in use by something else..
2671 */
2672void dentry_unhash(struct dentry *dentry)
2673{
2674 dget(dentry);
dc168427 2675 shrink_dcache_parent(dentry);
1da177e4 2676 spin_lock(&dentry->d_lock);
b7ab39f6 2677 if (dentry->d_count == 2)
1da177e4
LT		 2678 __d_drop(dentry);
2679 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2680}
2681
2682int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2683{
2684 int error = may_delete(dir, dentry, 1);
2685
2686 if (error)
2687 return error;
2688
acfa4380 2689 if (!dir->i_op->rmdir)
1da177e4
LT		 2690 return -EPERM;
2691
1b1dcc1b 2692 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2693 dentry_unhash(dentry);
2694 if (d_mountpoint(dentry))
2695 error = -EBUSY;
2696 else {
2697 error = security_inode_rmdir(dir, dentry);
2698 if (!error) {
2699 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2700 if (!error) {
1da177e4 2701 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2702 dont_mount(dentry);
2703 }
1da177e4
LT		 2704 }
2705 }
1b1dcc1b 2706 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2707 if (!error) {
1da177e4
LT		 2708 d_delete(dentry);
2709 }
2710 dput(dentry);
2711
2712 return error;
2713}
2714
5590ff0d 2715static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2716{
2717 int error = 0;
2718 char * name;
2719 struct dentry *dentry;
2720 struct nameidata nd;
2721
2ad94ae6 2722 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2723 if (error)
2ad94ae6 2724 return error;
1da177e4
LT		 2725
2726 switch(nd.last_type) {
0612d9fb
OH		 2727 case LAST_DOTDOT:
2728 error = -ENOTEMPTY;
2729 goto exit1;
2730 case LAST_DOT:
2731 error = -EINVAL;
2732 goto exit1;
2733 case LAST_ROOT:
2734 error = -EBUSY;
2735 goto exit1;
1da177e4 2736 }
0612d9fb
OH		 2737
2738 nd.flags &= ~LOOKUP_PARENT;
2739
4ac91378 2740 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2741 dentry = lookup_hash(&nd);
1da177e4 2742 error = PTR_ERR(dentry);
6902d925
DH		 2743 if (IS_ERR(dentry))
2744 goto exit2;
0622753b
DH		 2745 error = mnt_want_write(nd.path.mnt);
2746 if (error)
2747 goto exit3;
be6d3e56
KT		 2748 error = security_path_rmdir(&nd.path, dentry);
2749 if (error)
2750 goto exit4;
4ac91378 2751 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2752exit4:
0622753b
DH		 2753 mnt_drop_write(nd.path.mnt);
2754exit3:
6902d925
DH		 2755 dput(dentry);
2756exit2:
4ac91378 2757 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2758exit1:
1d957f9b 2759 path_put(&nd.path);
1da177e4
LT		 2760 putname(name);
2761 return error;
2762}
2763
3cdad428 2764SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2765{
2766 return do_rmdir(AT_FDCWD, pathname);
2767}
2768
1da177e4
LT		 2769int vfs_unlink(struct inode *dir, struct dentry *dentry)
2770{
2771 int error = may_delete(dir, dentry, 0);
2772
2773 if (error)
2774 return error;
2775
acfa4380 2776 if (!dir->i_op->unlink)
1da177e4
LT		 2777 return -EPERM;
2778
1b1dcc1b 2779 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2780 if (d_mountpoint(dentry))
2781 error = -EBUSY;
2782 else {
2783 error = security_inode_unlink(dir, dentry);
bec1052e 2784 if (!error) {
1da177e4 2785 error = dir->i_op->unlink(dir, dentry);
bec1052e 2786 if (!error)
d83c49f3 2787 dont_mount(dentry);
bec1052e 2788 }
1da177e4 2789 }
1b1dcc1b 2790 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2791
2792 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2793 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2794 fsnotify_link_count(dentry->d_inode);
e234f35c 2795 d_delete(dentry);
1da177e4 2796 }
0eeca283 2797
1da177e4
LT		 2798 return error;
2799}
2800
2801/*
2802 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2803 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2804 * writeout happening, and we don't want to prevent access to the directory
2805 * while waiting on the I/O.
2806 */
5590ff0d 2807static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2808{
2ad94ae6
AV		 2809 int error;
2810 char *name;
1da177e4
LT		 2811 struct dentry *dentry;
2812 struct nameidata nd;
2813 struct inode *inode = NULL;
2814
2ad94ae6 2815 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2816 if (error)
2ad94ae6
AV		 2817 return error;
2818
1da177e4
LT		 2819 error = -EISDIR;
2820 if (nd.last_type != LAST_NORM)
2821 goto exit1;
0612d9fb
OH		 2822
2823 nd.flags &= ~LOOKUP_PARENT;
2824
4ac91378 2825 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2826 dentry = lookup_hash(&nd);
1da177e4
LT		 2827 error = PTR_ERR(dentry);
2828 if (!IS_ERR(dentry)) {
2829 /* Why not before? Because we want correct error value */
2830 if (nd.last.name[nd.last.len])
2831 goto slashes;
2832 inode = dentry->d_inode;
2833 if (inode)
7de9c6ee 2834 ihold(inode);
0622753b
DH		 2835 error = mnt_want_write(nd.path.mnt);
2836 if (error)
2837 goto exit2;
be6d3e56
KT		 2838 error = security_path_unlink(&nd.path, dentry);
2839 if (error)
2840 goto exit3;
4ac91378 2841 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2842exit3:
0622753b 2843 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2844 exit2:
2845 dput(dentry);
2846 }
4ac91378 2847 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2848 if (inode)
2849 iput(inode); /* truncate the inode here */
2850exit1:
1d957f9b 2851 path_put(&nd.path);
1da177e4
LT		 2852 putname(name);
2853 return error;
2854
2855slashes:
2856 error = !dentry->d_inode ? -ENOENT :
2857 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2858 goto exit2;
2859}
2860
2e4d0924 2861SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2862{
2863 if ((flag & ~AT_REMOVEDIR) != 0)
2864 return -EINVAL;
2865
2866 if (flag & AT_REMOVEDIR)
2867 return do_rmdir(dfd, pathname);
2868
2869 return do_unlinkat(dfd, pathname);
2870}
2871
3480b257 2872SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2873{
2874 return do_unlinkat(AT_FDCWD, pathname);
2875}
2876
db2e747b 2877int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2878{
a95164d9 2879 int error = may_create(dir, dentry);
1da177e4
LT		 2880
2881 if (error)
2882 return error;
2883
acfa4380 2884 if (!dir->i_op->symlink)
1da177e4
LT		 2885 return -EPERM;
2886
2887 error = security_inode_symlink(dir, dentry, oldname);
2888 if (error)
2889 return error;
2890
1da177e4 2891 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2892 if (!error)
f38aa942 2893 fsnotify_create(dir, dentry);
1da177e4
LT		 2894 return error;
2895}
2896
2e4d0924
HC		 2897SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2898 int, newdfd, const char __user *, newname)
1da177e4 2899{
2ad94ae6
AV		 2900 int error;
2901 char *from;
2902 char *to;
6902d925
DH		 2903 struct dentry *dentry;
2904 struct nameidata nd;
1da177e4
LT		 2905
2906 from = getname(oldname);
2ad94ae6 2907 if (IS_ERR(from))
1da177e4 2908 return PTR_ERR(from);
1da177e4 2909
2ad94ae6 2910 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2911 if (error)
2ad94ae6
AV		 2912 goto out_putname;
2913
6902d925
DH		 2914 dentry = lookup_create(&nd, 0);
2915 error = PTR_ERR(dentry);
2916 if (IS_ERR(dentry))
2917 goto out_unlock;
2918
75c3f29d
DH		 2919 error = mnt_want_write(nd.path.mnt);
2920 if (error)
2921 goto out_dput;
be6d3e56
KT		 2922 error = security_path_symlink(&nd.path, dentry, from);
2923 if (error)
2924 goto out_drop_write;
db2e747b 2925 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2926out_drop_write:
75c3f29d
DH		 2927 mnt_drop_write(nd.path.mnt);
2928out_dput:
6902d925
DH		 2929 dput(dentry);
2930out_unlock:
4ac91378 2931 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2932 path_put(&nd.path);
6902d925
DH		 2933 putname(to);
2934out_putname:
1da177e4
LT		 2935 putname(from);
2936 return error;
2937}
2938
3480b257 2939SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2940{
2941 return sys_symlinkat(oldname, AT_FDCWD, newname);
2942}
2943
1da177e4
LT		 2944int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2945{
2946 struct inode *inode = old_dentry->d_inode;
2947 int error;
2948
2949 if (!inode)
2950 return -ENOENT;
2951
a95164d9 2952 error = may_create(dir, new_dentry);
1da177e4
LT		 2953 if (error)
2954 return error;
2955
2956 if (dir->i_sb != inode->i_sb)
2957 return -EXDEV;
2958
2959 /*
2960 * A link to an append-only or immutable file cannot be created.
2961 */
2962 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2963 return -EPERM;
acfa4380 2964 if (!dir->i_op->link)
1da177e4 2965 return -EPERM;
7e79eedb 2966 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2967 return -EPERM;
2968
2969 error = security_inode_link(old_dentry, dir, new_dentry);
2970 if (error)
2971 return error;
2972
7e79eedb 2973 mutex_lock(&inode->i_mutex);
1da177e4 2974 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2975 mutex_unlock(&inode->i_mutex);
e31e14ec 2976 if (!error)
7e79eedb 2977 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2978 return error;
2979}
2980
2981/*
2982 * Hardlinks are often used in delicate situations. We avoid
2983 * security-related surprises by not following symlinks on the
2984 * newname. --KAB
2985 *
2986 * We don't follow them on the oldname either to be compatible
2987 * with linux 2.0, and to avoid hard-linking to directories
2988 * and other special files. --ADM
2989 */
2e4d0924
HC		 2990SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2991 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2992{
2993 struct dentry *new_dentry;
2d8f3038
AV		 2994 struct nameidata nd;
2995 struct path old_path;
1da177e4 2996 int error;
2ad94ae6 2997 char *to;
1da177e4 2998
45c9b11a 2999 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 3000 return -EINVAL;
3001
2d8f3038
AV		 3002 error = user_path_at(olddfd, oldname,
3003 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
3004 &old_path);
1da177e4 3005 if (error)
2ad94ae6
AV		 3006 return error;
3007
3008 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 3009 if (error)
3010 goto out;
3011 error = -EXDEV;
2d8f3038 3012 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 3013 goto out_release;
3014 new_dentry = lookup_create(&nd, 0);
3015 error = PTR_ERR(new_dentry);
6902d925
DH		 3016 if (IS_ERR(new_dentry))
3017 goto out_unlock;
75c3f29d
DH		 3018 error = mnt_want_write(nd.path.mnt);
3019 if (error)
3020 goto out_dput;
be6d3e56
KT		 3021 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
3022 if (error)
3023 goto out_drop_write;
2d8f3038 3024 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 3025out_drop_write:
75c3f29d
DH		 3026 mnt_drop_write(nd.path.mnt);
3027out_dput:
6902d925
DH		 3028 dput(new_dentry);
3029out_unlock:
4ac91378 3030 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 3031out_release:
1d957f9b 3032 path_put(&nd.path);
2ad94ae6 3033 putname(to);
1da177e4 3034out:
2d8f3038 3035 path_put(&old_path);
1da177e4
LT		 3036
3037 return error;
3038}
3039
3480b257 3040SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3041{
c04030e1 3042 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3043}
3044
1da177e4
LT		 3045/*
3046 * The worst of all namespace operations - renaming directory. "Perverted"
3047 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3048 * Problems:
3049 * a) we can get into loop creation. Check is done in is_subdir().
3050 * b) race potential - two innocent renames can create a loop together.
3051 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3052 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3053 * story.
3054 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3055 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3056 * whether the target exists). Solution: try to be smart with locking
3057 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3058 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3059 * move will be locked. Thus we can rank directories by the tree
3060 * (ancestors first) and rank all non-directories after them.
3061 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3062 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3063 * HOWEVER, it relies on the assumption that any object with ->lookup()
3064 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3065 * we'd better make sure that there's no link(2) for them.
3066 * d) some filesystems don't support opened-but-unlinked directories,
3067 * either because of layout or because they are not ready to deal with
3068 * all cases correctly. The latter will be fixed (taking this sort of
3069 * stuff into VFS), but the former is not going away. Solution: the same
3070 * trick as in rmdir().
3071 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3072 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3073 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3074 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3075 * locking].
3076 */
75c96f85
AB		 3077static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3078 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3079{
3080 int error = 0;
3081 struct inode *target;
3082
3083 /*
3084 * If we are going to change the parent - check write permissions,
3085 * we'll need to flip '..'.
3086 */
3087 if (new_dir != old_dir) {
f419a2e3 3088 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3089 if (error)
3090 return error;
3091 }
3092
3093 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3094 if (error)
3095 return error;
3096
3097 target = new_dentry->d_inode;
d83c49f3 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;
d83c49f3
AV		 3102 else {
3103 if (target)
3104 dentry_unhash(new_dentry);
1da177e4 3105 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3106 }
1da177e4 3107 if (target) {
d83c49f3 3108 if (!error) {
1da177e4 3109 target->i_flags |= S_DEAD;
d83c49f3
AV		 3110 dont_mount(new_dentry);
3111 }
1b1dcc1b 3112 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3113 if (d_unhashed(new_dentry))
3114 d_rehash(new_dentry);
3115 dput(new_dentry);
3116 }
e31e14ec 3117 if (!error)
349457cc
MF		 3118 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3119 d_move(old_dentry,new_dentry);
1da177e4
LT		 3120 return error;
3121}
3122
75c96f85
AB		 3123static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3124 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3125{
3126 struct inode *target;
3127 int error;
3128
3129 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3130 if (error)
3131 return error;
3132
3133 dget(new_dentry);
3134 target = new_dentry->d_inode;
3135 if (target)
1b1dcc1b 3136 mutex_lock(&target->i_mutex);
1da177e4
LT		 3137 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3138 error = -EBUSY;
3139 else
3140 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3141 if (!error) {
bec1052e 3142 if (target)
d83c49f3 3143 dont_mount(new_dentry);
349457cc 3144 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3145 d_move(old_dentry, new_dentry);
1da177e4
LT		 3146 }
3147 if (target)
1b1dcc1b 3148 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3149 dput(new_dentry);
3150 return error;
3151}
3152
3153int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3154 struct inode *new_dir, struct dentry *new_dentry)
3155{
3156 int error;
3157 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3158 const unsigned char *old_name;
1da177e4
LT		 3159
3160 if (old_dentry->d_inode == new_dentry->d_inode)
3161 return 0;
3162
3163 error = may_delete(old_dir, old_dentry, is_dir);
3164 if (error)
3165 return error;
3166
3167 if (!new_dentry->d_inode)
a95164d9 3168 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3169 else
3170 error = may_delete(new_dir, new_dentry, is_dir);
3171 if (error)
3172 return error;
3173
acfa4380 3174 if (!old_dir->i_op->rename)
1da177e4
LT		 3175 return -EPERM;
3176
0eeca283
RL		 3177 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3178
1da177e4
LT		 3179 if (is_dir)
3180 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3181 else
3182 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3183 if (!error)
3184 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3185 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3186 fsnotify_oldname_free(old_name);
3187
1da177e4
LT		 3188 return error;
3189}
3190
2e4d0924
HC		 3191SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3192 int, newdfd, const char __user *, newname)
1da177e4 3193{
2ad94ae6
AV		 3194 struct dentry *old_dir, *new_dir;
3195 struct dentry *old_dentry, *new_dentry;
3196 struct dentry *trap;
1da177e4 3197 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3198 char *from;
3199 char *to;
3200 int error;
1da177e4 3201
2ad94ae6 3202 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3203 if (error)
3204 goto exit;
3205
2ad94ae6 3206 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3207 if (error)
3208 goto exit1;
3209
3210 error = -EXDEV;
4ac91378 3211 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3212 goto exit2;
3213
4ac91378 3214 old_dir = oldnd.path.dentry;
1da177e4
LT		 3215 error = -EBUSY;
3216 if (oldnd.last_type != LAST_NORM)
3217 goto exit2;
3218
4ac91378 3219 new_dir = newnd.path.dentry;
1da177e4
LT		 3220 if (newnd.last_type != LAST_NORM)
3221 goto exit2;
3222
0612d9fb
OH		 3223 oldnd.flags &= ~LOOKUP_PARENT;
3224 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3225 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3226
1da177e4
LT		 3227 trap = lock_rename(new_dir, old_dir);
3228
49705b77 3229 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3230 error = PTR_ERR(old_dentry);
3231 if (IS_ERR(old_dentry))
3232 goto exit3;
3233 /* source must exist */
3234 error = -ENOENT;
3235 if (!old_dentry->d_inode)
3236 goto exit4;
3237 /* unless the source is a directory trailing slashes give -ENOTDIR */
3238 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3239 error = -ENOTDIR;
3240 if (oldnd.last.name[oldnd.last.len])
3241 goto exit4;
3242 if (newnd.last.name[newnd.last.len])
3243 goto exit4;
3244 }
3245 /* source should not be ancestor of target */
3246 error = -EINVAL;
3247 if (old_dentry == trap)
3248 goto exit4;
49705b77 3249 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3250 error = PTR_ERR(new_dentry);
3251 if (IS_ERR(new_dentry))
3252 goto exit4;
3253 /* target should not be an ancestor of source */
3254 error = -ENOTEMPTY;
3255 if (new_dentry == trap)
3256 goto exit5;
3257
9079b1eb
DH		 3258 error = mnt_want_write(oldnd.path.mnt);
3259 if (error)
3260 goto exit5;
be6d3e56
KT		 3261 error = security_path_rename(&oldnd.path, old_dentry,
3262 &newnd.path, new_dentry);
3263 if (error)
3264 goto exit6;
1da177e4
LT		 3265 error = vfs_rename(old_dir->d_inode, old_dentry,
3266 new_dir->d_inode, new_dentry);
be6d3e56 3267exit6:
9079b1eb 3268 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3269exit5:
3270 dput(new_dentry);
3271exit4:
3272 dput(old_dentry);
3273exit3:
3274 unlock_rename(new_dir, old_dir);
3275exit2:
1d957f9b 3276 path_put(&newnd.path);
2ad94ae6 3277 putname(to);
1da177e4 3278exit1:
1d957f9b 3279 path_put(&oldnd.path);
1da177e4 3280 putname(from);
2ad94ae6 3281exit:
1da177e4
LT		 3282 return error;
3283}
3284
a26eab24 3285SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3286{
3287 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3288}
3289
1da177e4
LT		 3290int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3291{
3292 int len;
3293
3294 len = PTR_ERR(link);
3295 if (IS_ERR(link))
3296 goto out;
3297
3298 len = strlen(link);
3299 if (len > (unsigned) buflen)
3300 len = buflen;
3301 if (copy_to_user(buffer, link, len))
3302 len = -EFAULT;
3303out:
3304 return len;
3305}
3306
3307/*
3308 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3309 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3310 * using) it for any given inode is up to filesystem.
3311 */
3312int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3313{
3314 struct nameidata nd;
cc314eef 3315 void *cookie;
694a1764 3316 int res;
cc314eef 3317
1da177e4 3318 nd.depth = 0;
cc314eef 3319 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3320 if (IS_ERR(cookie))
3321 return PTR_ERR(cookie);
3322
3323 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3324 if (dentry->d_inode->i_op->put_link)
3325 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3326 return res;
1da177e4
LT		 3327}
3328
3329int vfs_follow_link(struct nameidata *nd, const char *link)
3330{
3331 return __vfs_follow_link(nd, link);
3332}
3333
3334/* get the link contents into pagecache */
3335static char *page_getlink(struct dentry * dentry, struct page **ppage)
3336{
ebd09abb
DG		 3337 char *kaddr;
3338 struct page *page;
1da177e4 3339 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3340 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3341 if (IS_ERR(page))
6fe6900e 3342 return (char*)page;
1da177e4 3343 *ppage = page;
ebd09abb
DG		 3344 kaddr = kmap(page);
3345 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3346 return kaddr;
1da177e4
LT		 3347}
3348
3349int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3350{
3351 struct page *page = NULL;
3352 char *s = page_getlink(dentry, &page);
3353 int res = vfs_readlink(dentry,buffer,buflen,s);
3354 if (page) {
3355 kunmap(page);
3356 page_cache_release(page);
3357 }
3358 return res;
3359}
3360
cc314eef 3361void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3362{
cc314eef 3363 struct page *page = NULL;
1da177e4 3364 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3365 return page;
1da177e4
LT		 3366}
3367
cc314eef 3368void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3369{
cc314eef
LT		 3370 struct page *page = cookie;
3371
3372 if (page) {
1da177e4
LT		 3373 kunmap(page);
3374 page_cache_release(page);
1da177e4
LT		 3375 }
3376}
3377
54566b2c
NP		 3378/*
3379 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3380 */
3381int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3382{
3383 struct address_space *mapping = inode->i_mapping;
0adb25d2 3384 struct page *page;
afddba49 3385 void *fsdata;
beb497ab 3386 int err;
1da177e4 3387 char *kaddr;
54566b2c
NP		 3388 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3389 if (nofs)
3390 flags |= AOP_FLAG_NOFS;
1da177e4 3391
7e53cac4 3392retry:
afddba49 3393 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3394 flags, &page, &fsdata);
1da177e4 3395 if (err)
afddba49
NP		 3396 goto fail;
3397
1da177e4
LT		 3398 kaddr = kmap_atomic(page, KM_USER0);
3399 memcpy(kaddr, symname, len-1);
3400 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3401
3402 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3403 page, fsdata);
1da177e4
LT		 3404 if (err < 0)
3405 goto fail;
afddba49
NP		 3406 if (err < len-1)
3407 goto retry;
3408
1da177e4
LT		 3409 mark_inode_dirty(inode);
3410 return 0;
1da177e4
LT		 3411fail:
3412 return err;
3413}
3414
0adb25d2
KK		 3415int page_symlink(struct inode *inode, const char *symname, int len)
3416{
3417 return __page_symlink(inode, symname, len,
54566b2c 3418 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3419}
3420
92e1d5be 3421const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3422 .readlink = generic_readlink,
3423 .follow_link = page_follow_link_light,
3424 .put_link = page_put_link,
3425};
3426
2d8f3038 3427EXPORT_SYMBOL(user_path_at);
cc53ce53 3428EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3429EXPORT_SYMBOL(follow_down);
3430EXPORT_SYMBOL(follow_up);
3431EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3432EXPORT_SYMBOL(getname);
3433EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3434EXPORT_SYMBOL(lookup_one_len);
3435EXPORT_SYMBOL(page_follow_link_light);
3436EXPORT_SYMBOL(page_put_link);
3437EXPORT_SYMBOL(page_readlink);
0adb25d2 3438EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3439EXPORT_SYMBOL(page_symlink);
3440EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3441EXPORT_SYMBOL(kern_path_parent);
d1811465 3442EXPORT_SYMBOL(kern_path);
16f18200 3443EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3444EXPORT_SYMBOL(inode_permission);
8c744fb8 3445EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3446EXPORT_SYMBOL(unlock_rename);
3447EXPORT_SYMBOL(vfs_create);
3448EXPORT_SYMBOL(vfs_follow_link);
3449EXPORT_SYMBOL(vfs_link);
3450EXPORT_SYMBOL(vfs_mkdir);
3451EXPORT_SYMBOL(vfs_mknod);
3452EXPORT_SYMBOL(generic_permission);
3453EXPORT_SYMBOL(vfs_readlink);
3454EXPORT_SYMBOL(vfs_rename);
3455EXPORT_SYMBOL(vfs_rmdir);
3456EXPORT_SYMBOL(vfs_symlink);
3457EXPORT_SYMBOL(vfs_unlink);
3458EXPORT_SYMBOL(dentry_unhash);
3459EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)