projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
expand finish_open() in its only caller
[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
b04f784e
NP		 1219 /*
1220 * Rename seqlock is not required here because in the off chance
1221 * of a false negative due to a concurrent rename, we're going to
1222 * do the non-racy lookup, below.
1223 */
31e6b01f
NP		 1224 if (nd->flags & LOOKUP_RCU) {
1225 unsigned seq;
1226
1227 *inode = nd->inode;
1228 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1229 if (!dentry) {
1230 if (nameidata_drop_rcu(nd))
1231 return -ECHILD;
1232 goto need_lookup;
1233 }
1234 /* Memory barrier in read_seqcount_begin of child is enough */
1235 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1236 return -ECHILD;
1237
1238 nd->seq = seq;
24643087 1239 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
f5e1c1c1 1240 dentry = do_revalidate_rcu(dentry, nd);
24643087
AV		 1241 if (!dentry)
1242 goto need_lookup;
1243 if (IS_ERR(dentry))
1244 goto fail;
1245 if (!(nd->flags & LOOKUP_RCU))
1246 goto done;
1247 }
31e6b01f
NP		 1248 path->mnt = mnt;
1249 path->dentry = dentry;
9875cf80
DH		 1250 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1251 return 0;
1252 if (nameidata_drop_rcu(nd))
1253 return -ECHILD;
1254 /* fallthru */
1255 }
1256 dentry = __d_lookup(parent, name);
1257 if (!dentry)
1258 goto need_lookup;
2e2e88ea 1259found:
24643087
AV		 1260 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1261 dentry = do_revalidate(dentry, nd);
1262 if (!dentry)
1263 goto need_lookup;
1264 if (IS_ERR(dentry))
1265 goto fail;
1266 }
1da177e4 1267done:
9875cf80
DH		 1268 path->mnt = mnt;
1269 path->dentry = dentry;
1270 err = follow_managed(path, nd->flags);
89312214
IK		 1271 if (unlikely(err < 0)) {
1272 path_put_conditional(path, nd);
9875cf80 1273 return err;
89312214 1274 }
9875cf80 1275 *inode = path->dentry->d_inode;
1da177e4
LT		 1276 return 0;
1277
1278need_lookup:
6e6b1bd1 1279 dir = parent->d_inode;
31e6b01f 1280 BUG_ON(nd->inode != dir);
6e6b1bd1
AV		 1281
1282 mutex_lock(&dir->i_mutex);
1283 /*
1284 * First re-do the cached lookup just in case it was created
b04f784e
NP		 1285 * while we waited for the directory semaphore, or the first
1286 * lookup failed due to an unrelated rename.
6e6b1bd1 1287 *
b04f784e
NP		 1288 * This could use version numbering or similar to avoid unnecessary
1289 * cache lookups, but then we'd have to do the first lookup in the
1290 * non-racy way. However in the common case here, everything should
1291 * be hot in cache, so would it be a big win?
6e6b1bd1
AV		 1292 */
1293 dentry = d_lookup(parent, name);
baa03890
NP		 1294 if (likely(!dentry)) {
1295 dentry = d_alloc_and_lookup(parent, name, nd);
6e6b1bd1
AV		 1296 mutex_unlock(&dir->i_mutex);
1297 if (IS_ERR(dentry))
1298 goto fail;
1299 goto done;
1300 }
6e6b1bd1
AV		 1301 /*
1302 * Uhhuh! Nasty case: the cache was re-populated while
1303 * we waited on the semaphore. Need to revalidate.
1304 */
1305 mutex_unlock(&dir->i_mutex);
2e2e88ea 1306 goto found;
1da177e4 1307
1da177e4
LT		 1308fail:
1309 return PTR_ERR(dentry);
1310}
1311
52094c8a
AV		 1312static inline int may_lookup(struct nameidata *nd)
1313{
1314 if (nd->flags & LOOKUP_RCU) {
1315 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1316 if (err != -ECHILD)
1317 return err;
1318 if (nameidata_drop_rcu(nd))
1319 return -ECHILD;
1320 }
1321 return exec_permission(nd->inode, 0);
1322}
1323
9856fa1b
AV		 1324static inline int handle_dots(struct nameidata *nd, int type)
1325{
1326 if (type == LAST_DOTDOT) {
1327 if (nd->flags & LOOKUP_RCU) {
1328 if (follow_dotdot_rcu(nd))
1329 return -ECHILD;
1330 } else
1331 follow_dotdot(nd);
1332 }
1333 return 0;
1334}
1335
951361f9
AV		 1336static void terminate_walk(struct nameidata *nd)
1337{
1338 if (!(nd->flags & LOOKUP_RCU)) {
1339 path_put(&nd->path);
1340 } else {
1341 nd->flags &= ~LOOKUP_RCU;
1342 nd->root.mnt = NULL;
1343 rcu_read_unlock();
1344 br_read_unlock(vfsmount_lock);
1345 }
1346}
1347
1da177e4
LT		 1348/*
1349 * Name resolution.
ea3834d9
PM		 1350 * This is the basic name resolution function, turning a pathname into
1351 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1352 *
ea3834d9
PM		 1353 * Returns 0 and nd will have valid dentry and mnt on success.
1354 * Returns error and drops reference to input namei data on failure.
1da177e4 1355 */
6de88d72 1356static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1357{
1358 struct path next;
1da177e4
LT		 1359 int err;
1360 unsigned int lookup_flags = nd->flags;
1361
1362 while (*name=='/')
1363 name++;
1364 if (!*name)
086e183a 1365 return 0;
1da177e4 1366
1da177e4 1367 if (nd->depth)
f55eab82 1368 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 1369
1370 /* At this point we know we have a real path component. */
1371 for(;;) {
31e6b01f 1372 struct inode *inode;
1da177e4
LT		 1373 unsigned long hash;
1374 struct qstr this;
1375 unsigned int c;
fe479a58 1376 int type;
1da177e4 1377
cdce5d6b 1378 nd->flags |= LOOKUP_CONTINUE;
52094c8a
AV		 1379
1380 err = may_lookup(nd);
1da177e4
LT		 1381 if (err)
1382 break;
1383
1384 this.name = name;
1385 c = *(const unsigned char *)name;
1386
1387 hash = init_name_hash();
1388 do {
1389 name++;
1390 hash = partial_name_hash(c, hash);
1391 c = *(const unsigned char *)name;
1392 } while (c && (c != '/'));
1393 this.len = name - (const char *) this.name;
1394 this.hash = end_name_hash(hash);
1395
fe479a58
AV		 1396 type = LAST_NORM;
1397 if (this.name[0] == '.') switch (this.len) {
1398 case 2:
16c2cd71 1399 if (this.name[1] == '.') {
fe479a58 1400 type = LAST_DOTDOT;
16c2cd71
AV		 1401 nd->flags |= LOOKUP_JUMPED;
1402 }
fe479a58
AV		 1403 break;
1404 case 1:
1405 type = LAST_DOT;
1406 }
5a202bcd
AV		 1407 if (likely(type == LAST_NORM)) {
1408 struct dentry *parent = nd->path.dentry;
16c2cd71 1409 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1410 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1411 err = parent->d_op->d_hash(parent, nd->inode,
1412 &this);
1413 if (err < 0)
1414 break;
1415 }
1416 }
fe479a58 1417
1da177e4
LT		 1418 /* remove trailing slashes? */
1419 if (!c)
1420 goto last_component;
1421 while (*++name == '/');
1422 if (!*name)
1423 goto last_with_slashes;
1424
1425 /*
1426 * "." and ".." are special - ".." especially so because it has
1427 * to be able to know about the current root directory and
1428 * parent relationships.
1429 */
fe479a58 1430 if (unlikely(type != LAST_NORM)) {
ef7562d5
AV		 1431 if (handle_dots(nd, type))
1432 return -ECHILD;
fe479a58 1433 continue;
1da177e4 1434 }
fe479a58 1435
1da177e4 1436 /* This does the actual lookups.. */
31e6b01f 1437 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1438 if (err)
1439 break;
1da177e4 1440
7bc055d1 1441 if (inode && inode->i_op->follow_link) {
3abb17e8 1442 err = do_follow_link(inode, &next, nd);
1da177e4 1443 if (err)
a7472bab 1444 return err;
31e6b01f 1445 nd->inode = nd->path.dentry->d_inode;
31e6b01f 1446 } else {
09dd17d3 1447 path_to_nameidata(&next, nd);
31e6b01f
NP		 1448 nd->inode = inode;
1449 }
7bc055d1
AV		 1450 err = -ENOENT;
1451 if (!nd->inode)
1452 break;
1da177e4 1453 err = -ENOTDIR;
31e6b01f 1454 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1455 break;
1456 continue;
1457 /* here ends the main loop */
1458
1459last_with_slashes:
1460 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1461last_component:
f55eab82
TM		 1462 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1463 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 1464 if (lookup_flags & LOOKUP_PARENT)
1465 goto lookup_parent;
ef7562d5
AV		 1466 if (unlikely(type != LAST_NORM))
1467 return handle_dots(nd, type);
31e6b01f 1468 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1469 if (err)
1470 break;
db372915
DH		 1471 if (inode && unlikely(inode->i_op->follow_link) &&
1472 (lookup_flags & LOOKUP_FOLLOW)) {
3abb17e8 1473 err = do_follow_link(inode, &next, nd);
1da177e4 1474 if (err)
a7472bab 1475 return err;
31e6b01f
NP		 1476 nd->inode = nd->path.dentry->d_inode;
1477 } else {
09dd17d3 1478 path_to_nameidata(&next, nd);
31e6b01f
NP		 1479 nd->inode = inode;
1480 }
1da177e4 1481 err = -ENOENT;
31e6b01f 1482 if (!nd->inode)
1da177e4
LT		 1483 break;
1484 if (lookup_flags & LOOKUP_DIRECTORY) {
1485 err = -ENOTDIR;
31e6b01f 1486 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1487 break;
1488 }
086e183a 1489 return 0;
1da177e4
LT		 1490lookup_parent:
1491 nd->last = this;
fe479a58 1492 nd->last_type = type;
1da177e4 1493 return 0;
1da177e4 1494 }
951361f9 1495 terminate_walk(nd);
1da177e4
LT		 1496 return err;
1497}
1498
70e9b357
AV		 1499static int path_init(int dfd, const char *name, unsigned int flags,
1500 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1501{
1502 int retval = 0;
1503 int fput_needed;
1504 struct file *file;
1505
1506 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1507 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f
NP		 1508 nd->depth = 0;
1509 nd->root.mnt = NULL;
31e6b01f
NP		 1510
1511 if (*name=='/') {
e41f7d4e
AV		 1512 if (flags & LOOKUP_RCU) {
1513 br_read_lock(vfsmount_lock);
1514 rcu_read_lock();
1515 set_root_rcu(nd);
1516 } else {
1517 set_root(nd);
1518 path_get(&nd->root);
1519 }
1520 nd->path = nd->root;
31e6b01f 1521 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1522 if (flags & LOOKUP_RCU) {
1523 struct fs_struct *fs = current->fs;
1524 unsigned seq;
31e6b01f 1525
e41f7d4e
AV		 1526 br_read_lock(vfsmount_lock);
1527 rcu_read_lock();
c28cc364 1528
e41f7d4e
AV		 1529 do {
1530 seq = read_seqcount_begin(&fs->seq);
1531 nd->path = fs->pwd;
1532 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1533 } while (read_seqcount_retry(&fs->seq, seq));
1534 } else {
1535 get_fs_pwd(current->fs, &nd->path);
1536 }
31e6b01f
NP		 1537 } else {
1538 struct dentry *dentry;
1539
1540 file = fget_light(dfd, &fput_needed);
1541 retval = -EBADF;
1542 if (!file)
1543 goto out_fail;
1544
1545 dentry = file->f_path.dentry;
1546
1547 retval = -ENOTDIR;
1548 if (!S_ISDIR(dentry->d_inode->i_mode))
1549 goto fput_fail;
1550
1551 retval = file_permission(file, MAY_EXEC);
1552 if (retval)
1553 goto fput_fail;
1554
1555 nd->path = file->f_path;
e41f7d4e
AV		 1556 if (flags & LOOKUP_RCU) {
1557 if (fput_needed)
70e9b357 1558 *fp = file;
e41f7d4e
AV		 1559 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1560 br_read_lock(vfsmount_lock);
1561 rcu_read_lock();
1562 } else {
1563 path_get(&file->f_path);
1564 fput_light(file, fput_needed);
1565 }
31e6b01f 1566 }
31e6b01f 1567
31e6b01f 1568 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1569 return 0;
2dfdd266 1570
9b4a9b14
AV		 1571fput_fail:
1572 fput_light(file, fput_needed);
1573out_fail:
1574 return retval;
1575}
1576
1577/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1578static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1579 unsigned int flags, struct nameidata *nd)
1580{
70e9b357 1581 struct file *base = NULL;
31e6b01f
NP		 1582 int retval;
1583
1584 /*
1585 * Path walking is largely split up into 2 different synchronisation
1586 * schemes, rcu-walk and ref-walk (explained in
1587 * Documentation/filesystems/path-lookup.txt). These share much of the
1588 * path walk code, but some things particularly setup, cleanup, and
1589 * following mounts are sufficiently divergent that functions are
1590 * duplicated. Typically there is a function foo(), and its RCU
1591 * analogue, foo_rcu().
1592 *
1593 * -ECHILD is the error number of choice (just to avoid clashes) that
1594 * is returned if some aspect of an rcu-walk fails. Such an error must
1595 * be handled by restarting a traditional ref-walk (which will always
1596 * be able to complete).
1597 */
70e9b357 1598 retval = path_init(dfd, name, flags, nd, &base);
ee0827cd 1599
31e6b01f
NP		 1600 if (unlikely(retval))
1601 return retval;
ee0827cd
AV		 1602
1603 current->total_link_count = 0;
1604 retval = link_path_walk(name, nd);
1605
1606 if (nd->flags & LOOKUP_RCU) {
4455ca62
AV		 1607 /* went all way through without dropping RCU */
1608 BUG_ON(retval);
1609 if (nameidata_drop_rcu_last(nd))
1610 retval = -ECHILD;
ee0827cd
AV		 1611 }
1612
16c2cd71
AV		 1613 if (!retval)
1614 retval = handle_reval_path(nd);
1615
70e9b357
AV		 1616 if (base)
1617 fput(base);
ee0827cd 1618
2a737871
AV		 1619 if (nd->root.mnt) {
1620 path_put(&nd->root);
1621 nd->root.mnt = NULL;
1622 }
ee0827cd
AV		 1623 return retval;
1624}
31e6b01f 1625
ee0827cd
AV		 1626static int do_path_lookup(int dfd, const char *name,
1627 unsigned int flags, struct nameidata *nd)
1628{
1629 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1630 if (unlikely(retval == -ECHILD))
1631 retval = path_lookupat(dfd, name, flags, nd);
1632 if (unlikely(retval == -ESTALE))
1633 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1634
1635 if (likely(!retval)) {
1636 if (unlikely(!audit_dummy_context())) {
1637 if (nd->path.dentry && nd->inode)
1638 audit_inode(name, nd->path.dentry);
1639 }
1640 }
170aa3d0 1641 return retval;
1da177e4
LT		 1642}
1643
c9c6cac0 1644int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1645{
c9c6cac0 1646 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1647}
1648
d1811465
AV		 1649int kern_path(const char *name, unsigned int flags, struct path *path)
1650{
1651 struct nameidata nd;
1652 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1653 if (!res)
1654 *path = nd.path;
1655 return res;
1656}
1657
16f18200
JJS		 1658/**
1659 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1660 * @dentry: pointer to dentry of the base directory
1661 * @mnt: pointer to vfs mount of the base directory
1662 * @name: pointer to file name
1663 * @flags: lookup flags
1664 * @nd: pointer to nameidata
1665 */
1666int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1667 const char *name, unsigned int flags,
1668 struct nameidata *nd)
1669{
ee0827cd 1670 int result;
16f18200
JJS		 1671
1672 /* same as do_path_lookup */
1673 nd->last_type = LAST_ROOT;
16c2cd71 1674 nd->flags = flags | LOOKUP_JUMPED;
16f18200
JJS		 1675 nd->depth = 0;
1676
c8e7f449
JB		 1677 nd->path.dentry = dentry;
1678 nd->path.mnt = mnt;
1679 path_get(&nd->path);
5b857119
AV		 1680 nd->root = nd->path;
1681 path_get(&nd->root);
31e6b01f 1682 nd->inode = nd->path.dentry->d_inode;
16f18200 1683
ee0827cd
AV		 1684 current->total_link_count = 0;
1685
1686 result = link_path_walk(name, nd);
16c2cd71
AV		 1687 if (!result)
1688 result = handle_reval_path(nd);
ee0827cd
AV		 1689 if (result == -ESTALE) {
1690 /* nd->path had been dropped */
1691 current->total_link_count = 0;
1692 nd->path.dentry = dentry;
1693 nd->path.mnt = mnt;
1694 nd->inode = dentry->d_inode;
1695 path_get(&nd->path);
16c2cd71
AV		 1696 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_REVAL;
1697
ee0827cd 1698 result = link_path_walk(name, nd);
16c2cd71
AV		 1699 if (!result)
1700 result = handle_reval_path(nd);
ee0827cd
AV		 1701 }
1702 if (unlikely(!result && !audit_dummy_context() && nd->path.dentry &&
31e6b01f 1703 nd->inode))
4ac91378 1704 audit_inode(name, nd->path.dentry);
16f18200 1705
5b857119
AV		 1706 path_put(&nd->root);
1707 nd->root.mnt = NULL;
16f18200 1708
ee0827cd 1709 return result;
16f18200
JJS		 1710}
1711
eead1911
CH		 1712static struct dentry *__lookup_hash(struct qstr *name,
1713 struct dentry *base, struct nameidata *nd)
1da177e4 1714{
81fca444 1715 struct inode *inode = base->d_inode;
057f6c01 1716 struct dentry *dentry;
1da177e4
LT		 1717 int err;
1718
b74c79e9 1719 err = exec_permission(inode, 0);
81fca444
CH		 1720 if (err)
1721 return ERR_PTR(err);
1da177e4 1722
b04f784e
NP		 1723 /*
1724 * Don't bother with __d_lookup: callers are for creat as
1725 * well as unlink, so a lot of the time it would cost
1726 * a double lookup.
6e6b1bd1 1727 */
b04f784e 1728 dentry = d_lookup(base, name);
6e6b1bd1 1729
fb045adb 1730 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1731 dentry = do_revalidate(dentry, nd);
1732
baa03890
NP		 1733 if (!dentry)
1734 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1735
1da177e4
LT		 1736 return dentry;
1737}
1738
057f6c01
JM		 1739/*
1740 * Restricted form of lookup. Doesn't follow links, single-component only,
1741 * needs parent already locked. Doesn't follow mounts.
1742 * SMP-safe.
1743 */
eead1911 1744static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1745{
4ac91378 1746 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1747}
1748
eead1911 1749/**
a6b91919 1750 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1751 * @name: pathname component to lookup
1752 * @base: base directory to lookup from
1753 * @len: maximum length @len should be interpreted to
1754 *
a6b91919
RD		 1755 * Note that this routine is purely a helper for filesystem usage and should
1756 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1757 * nameidata argument is passed to the filesystem methods and a filesystem
1758 * using this helper needs to be prepared for that.
1759 */
057f6c01
JM		 1760struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1761{
057f6c01 1762 struct qstr this;
6a96ba54
AV		 1763 unsigned long hash;
1764 unsigned int c;
057f6c01 1765
2f9092e1
DW		 1766 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1767
6a96ba54
AV		 1768 this.name = name;
1769 this.len = len;
1770 if (!len)
1771 return ERR_PTR(-EACCES);
1772
1773 hash = init_name_hash();
1774 while (len--) {
1775 c = *(const unsigned char *)name++;
1776 if (c == '/' || c == '\0')
1777 return ERR_PTR(-EACCES);
1778 hash = partial_name_hash(c, hash);
1779 }
1780 this.hash = end_name_hash(hash);
5a202bcd
AV		 1781 /*
1782 * See if the low-level filesystem might want
1783 * to use its own hash..
1784 */
1785 if (base->d_flags & DCACHE_OP_HASH) {
1786 int err = base->d_op->d_hash(base, base->d_inode, &this);
1787 if (err < 0)
1788 return ERR_PTR(err);
1789 }
eead1911 1790
49705b77 1791 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1792}
1793
2d8f3038
AV		 1794int user_path_at(int dfd, const char __user *name, unsigned flags,
1795 struct path *path)
1da177e4 1796{
2d8f3038 1797 struct nameidata nd;
1da177e4
LT		 1798 char *tmp = getname(name);
1799 int err = PTR_ERR(tmp);
1da177e4 1800 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1801
1802 BUG_ON(flags & LOOKUP_PARENT);
1803
1804 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1805 putname(tmp);
2d8f3038
AV		 1806 if (!err)
1807 *path = nd.path;
1da177e4
LT		 1808 }
1809 return err;
1810}
1811
2ad94ae6
AV		 1812static int user_path_parent(int dfd, const char __user *path,
1813 struct nameidata *nd, char **name)
1814{
1815 char *s = getname(path);
1816 int error;
1817
1818 if (IS_ERR(s))
1819 return PTR_ERR(s);
1820
1821 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1822 if (error)
1823 putname(s);
1824 else
1825 *name = s;
1826
1827 return error;
1828}
1829
1da177e4
LT		 1830/*
1831 * It's inline, so penalty for filesystems that don't use sticky bit is
1832 * minimal.
1833 */
1834static inline int check_sticky(struct inode *dir, struct inode *inode)
1835{
da9592ed
DH		 1836 uid_t fsuid = current_fsuid();
1837
1da177e4
LT		 1838 if (!(dir->i_mode & S_ISVTX))
1839 return 0;
da9592ed 1840 if (inode->i_uid == fsuid)
1da177e4 1841 return 0;
da9592ed 1842 if (dir->i_uid == fsuid)
1da177e4
LT		 1843 return 0;
1844 return !capable(CAP_FOWNER);
1845}
1846
1847/*
1848 * Check whether we can remove a link victim from directory dir, check
1849 * whether the type of victim is right.
1850 * 1. We can't do it if dir is read-only (done in permission())
1851 * 2. We should have write and exec permissions on dir
1852 * 3. We can't remove anything from append-only dir
1853 * 4. We can't do anything with immutable dir (done in permission())
1854 * 5. If the sticky bit on dir is set we should either
1855 * a. be owner of dir, or
1856 * b. be owner of victim, or
1857 * c. have CAP_FOWNER capability
1858 * 6. If the victim is append-only or immutable we can't do antyhing with
1859 * links pointing to it.
1860 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1861 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1862 * 9. We can't remove a root or mountpoint.
1863 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1864 * nfs_async_unlink().
1865 */
858119e1 1866static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1867{
1868 int error;
1869
1870 if (!victim->d_inode)
1871 return -ENOENT;
1872
1873 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1874 audit_inode_child(victim, dir);
1da177e4 1875
f419a2e3 1876 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1877 if (error)
1878 return error;
1879 if (IS_APPEND(dir))
1880 return -EPERM;
1881 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1882 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1883 return -EPERM;
1884 if (isdir) {
1885 if (!S_ISDIR(victim->d_inode->i_mode))
1886 return -ENOTDIR;
1887 if (IS_ROOT(victim))
1888 return -EBUSY;
1889 } else if (S_ISDIR(victim->d_inode->i_mode))
1890 return -EISDIR;
1891 if (IS_DEADDIR(dir))
1892 return -ENOENT;
1893 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1894 return -EBUSY;
1895 return 0;
1896}
1897
1898/* Check whether we can create an object with dentry child in directory
1899 * dir.
1900 * 1. We can't do it if child already exists (open has special treatment for
1901 * this case, but since we are inlined it's OK)
1902 * 2. We can't do it if dir is read-only (done in permission())
1903 * 3. We should have write and exec permissions on dir
1904 * 4. We can't do it if dir is immutable (done in permission())
1905 */
a95164d9 1906static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1907{
1908 if (child->d_inode)
1909 return -EEXIST;
1910 if (IS_DEADDIR(dir))
1911 return -ENOENT;
f419a2e3 1912 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1913}
1914
1da177e4
LT		 1915/*
1916 * p1 and p2 should be directories on the same fs.
1917 */
1918struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1919{
1920 struct dentry *p;
1921
1922 if (p1 == p2) {
f2eace23 1923 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1924 return NULL;
1925 }
1926
a11f3a05 1927 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1928
e2761a11
OH		 1929 p = d_ancestor(p2, p1);
1930 if (p) {
1931 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1932 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1933 return p;
1da177e4
LT		 1934 }
1935
e2761a11
OH		 1936 p = d_ancestor(p1, p2);
1937 if (p) {
1938 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1939 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1940 return p;
1da177e4
LT		 1941 }
1942
f2eace23
IM		 1943 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1944 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1945 return NULL;
1946}
1947
1948void unlock_rename(struct dentry *p1, struct dentry *p2)
1949{
1b1dcc1b 1950 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1951 if (p1 != p2) {
1b1dcc1b 1952 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1953 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1954 }
1955}
1956
1957int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1958 struct nameidata *nd)
1959{
a95164d9 1960 int error = may_create(dir, dentry);
1da177e4
LT		 1961
1962 if (error)
1963 return error;
1964
acfa4380 1965 if (!dir->i_op->create)
1da177e4
LT		 1966 return -EACCES; /* shouldn't it be ENOSYS? */
1967 mode &= S_IALLUGO;
1968 mode |= S_IFREG;
1969 error = security_inode_create(dir, dentry, mode);
1970 if (error)
1971 return error;
1da177e4 1972 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1973 if (!error)
f38aa942 1974 fsnotify_create(dir, dentry);
1da177e4
LT		 1975 return error;
1976}
1977
3fb64190 1978int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1979{
3fb64190 1980 struct dentry *dentry = path->dentry;
1da177e4
LT		 1981 struct inode *inode = dentry->d_inode;
1982 int error;
1983
1984 if (!inode)
1985 return -ENOENT;
1986
c8fe8f30
CH		 1987 switch (inode->i_mode & S_IFMT) {
1988 case S_IFLNK:
1da177e4 1989 return -ELOOP;
c8fe8f30
CH		 1990 case S_IFDIR:
1991 if (acc_mode & MAY_WRITE)
1992 return -EISDIR;
1993 break;
1994 case S_IFBLK:
1995 case S_IFCHR:
3fb64190 1996 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 1997 return -EACCES;
c8fe8f30
CH		 1998 /*FALLTHRU*/
1999 case S_IFIFO:
2000 case S_IFSOCK:
1da177e4 2001 flag &= ~O_TRUNC;
c8fe8f30 2002 break;
4a3fd211 2003 }
b41572e9 2004
3fb64190 2005 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2006 if (error)
2007 return error;
6146f0d5 2008
1da177e4
LT		 2009 /*
2010 * An append-only file must be opened in append mode for writing.
2011 */
2012 if (IS_APPEND(inode)) {
8737c930 2013 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2014 return -EPERM;
1da177e4 2015 if (flag & O_TRUNC)
7715b521 2016 return -EPERM;
1da177e4
LT		 2017 }
2018
2019 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 2020 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2021 return -EPERM;
1da177e4
LT		 2022
2023 /*
2024 * Ensure there are no outstanding leases on the file.
2025 */
b65a9cfc 2026 return break_lease(inode, flag);
7715b521 2027}
1da177e4 2028
e1181ee6 2029static int handle_truncate(struct file *filp)
7715b521 2030{
e1181ee6 2031 struct path *path = &filp->f_path;
7715b521
AV		 2032 struct inode *inode = path->dentry->d_inode;
2033 int error = get_write_access(inode);
2034 if (error)
2035 return error;
2036 /*
2037 * Refuse to truncate files with mandatory locks held on them.
2038 */
2039 error = locks_verify_locked(inode);
2040 if (!error)
ea0d3ab2 2041 error = security_path_truncate(path);
7715b521
AV		 2042 if (!error) {
2043 error = do_truncate(path->dentry, 0,
2044 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2045 filp);
7715b521
AV		 2046 }
2047 put_write_access(inode);
acd0c935 2048 return error;
1da177e4
LT		 2049}
2050
d57999e1
DH		 2051/*
2052 * Be careful about ever adding any more callers of this
2053 * function. Its flags must be in the namei format, not
2054 * what get passed to sys_open().
2055 */
2056static int __open_namei_create(struct nameidata *nd, struct path *path,
8737c930 2057 int open_flag, int mode)
aab520e2
DH		 2058{
2059 int error;
4ac91378 2060 struct dentry *dir = nd->path.dentry;
aab520e2
DH		 2061
2062 if (!IS_POSIXACL(dir->d_inode))
ce3b0f8d 2063 mode &= ~current_umask();
be6d3e56
KT		 2064 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
2065 if (error)
2066 goto out_unlock;
aab520e2 2067 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
be6d3e56 2068out_unlock:
aab520e2 2069 mutex_unlock(&dir->d_inode->i_mutex);
4ac91378
JB		 2070 dput(nd->path.dentry);
2071 nd->path.dentry = path->dentry;
31e6b01f 2072
aab520e2
DH		 2073 if (error)
2074 return error;
2075 /* Don't check for write permission, don't truncate */
8737c930 2076 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
aab520e2
DH		 2077}
2078
d57999e1
DH		 2079/*
2080 * Note that while the flag value (low two bits) for sys_open means:
2081 * 00 - read-only
2082 * 01 - write-only
2083 * 10 - read-write
2084 * 11 - special
2085 * it is changed into
2086 * 00 - no permissions needed
2087 * 01 - read-permission
2088 * 10 - write-permission
2089 * 11 - read-write
2090 * for the internal routines (ie open_namei()/follow_link() etc)
2091 * This is more logical, and also allows the 00 "no perm needed"
2092 * to be used for symlinks (where the permissions are checked
2093 * later).
2094 *
2095*/
2096static inline int open_to_namei_flags(int flag)
2097{
2098 if ((flag+1) & O_ACCMODE)
2099 flag++;
2100 return flag;
2101}
2102
7715b521 2103static int open_will_truncate(int flag, struct inode *inode)
4a3fd211
DH		 2104{
2105 /*
2106 * We'll never write to the fs underlying
2107 * a device file.
2108 */
2109 if (special_file(inode->i_mode))
2110 return 0;
2111 return (flag & O_TRUNC);
2112}
2113
31e6b01f 2114/*
fe2d35ff 2115 * Handle the last step of open()
31e6b01f 2116 */
fb1cc555 2117static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2118 const struct open_flags *op, const char *pathname)
fb1cc555 2119{
a1e28038 2120 struct dentry *dir = nd->path.dentry;
0f9d1a10 2121 int will_truncate;
fb1cc555 2122 struct file *filp;
fe2d35ff 2123 struct inode *inode;
16c2cd71 2124 int error;
1f36f774 2125
c3e380b0
AV		 2126 nd->flags &= ~LOOKUP_PARENT;
2127 nd->flags |= op->intent;
2128
1f36f774
AV		 2129 switch (nd->last_type) {
2130 case LAST_DOTDOT:
176306f5 2131 case LAST_DOT:
fe2d35ff
AV		 2132 error = handle_dots(nd, nd->last_type);
2133 if (error)
2134 return ERR_PTR(error);
1f36f774 2135 /* fallthrough */
1f36f774 2136 case LAST_ROOT:
fe2d35ff
AV		 2137 if (nd->flags & LOOKUP_RCU) {
2138 if (nameidata_drop_rcu_last(nd))
2139 return ERR_PTR(-ECHILD);
2140 }
16c2cd71
AV		 2141 error = handle_reval_path(nd);
2142 if (error)
2143 goto exit;
fe2d35ff
AV		 2144 audit_inode(pathname, nd->path.dentry);
2145 if (op->open_flag & O_CREAT) {
2146 error = -EISDIR;
2147 goto exit;
2148 }
2149 goto ok;
1f36f774 2150 case LAST_BIND:
fe2d35ff 2151 /* can't be RCU mode here */
16c2cd71
AV		 2152 error = handle_reval_path(nd);
2153 if (error)
2154 goto exit;
1f36f774 2155 audit_inode(pathname, dir);
67ee3ad2 2156 goto ok;
1f36f774 2157 }
67ee3ad2 2158
fe2d35ff
AV		 2159 if (!(op->open_flag & O_CREAT)) {
2160 if (nd->last.name[nd->last.len])
2161 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2162 /* we _can_ be in RCU mode here */
2163 error = do_lookup(nd, &nd->last, path, &inode);
2164 if (error) {
2165 terminate_walk(nd);
2166 return ERR_PTR(error);
2167 }
2168 if (!inode) {
2169 path_to_nameidata(path, nd);
2170 terminate_walk(nd);
2171 return ERR_PTR(-ENOENT);
2172 }
2173 if (unlikely(inode->i_op->follow_link)) {
2174 /* We drop rcu-walk here */
2175 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
2176 return ERR_PTR(-ECHILD);
2177 return NULL;
2178 }
2179 path_to_nameidata(path, nd);
2180 nd->inode = inode;
2181 /* sayonara */
2182 if (nd->flags & LOOKUP_RCU) {
2183 if (nameidata_drop_rcu_last(nd))
2184 return ERR_PTR(-ECHILD);
2185 }
2186
2187 error = -ENOTDIR;
2188 if (nd->flags & LOOKUP_DIRECTORY) {
2189 if (!inode->i_op->lookup)
2190 goto exit;
2191 }
2192 audit_inode(pathname, nd->path.dentry);
2193 goto ok;
2194 }
2195
2196 /* create side of things */
2197
2198 if (nd->flags & LOOKUP_RCU) {
2199 if (nameidata_drop_rcu_last(nd))
2200 return ERR_PTR(-ECHILD);
2201 }
2202
2203 audit_inode(pathname, dir);
16c2cd71 2204 error = -EISDIR;
1f36f774 2205 /* trailing slashes? */
31e6b01f
NP		 2206 if (nd->last.name[nd->last.len])
2207 goto exit;
a2c36b45 2208
a1e28038
AV		 2209 mutex_lock(&dir->d_inode->i_mutex);
2210
2211 path->dentry = lookup_hash(nd);
2212 path->mnt = nd->path.mnt;
2213
fb1cc555
AV		 2214 error = PTR_ERR(path->dentry);
2215 if (IS_ERR(path->dentry)) {
2216 mutex_unlock(&dir->d_inode->i_mutex);
2217 goto exit;
2218 }
2219
2220 if (IS_ERR(nd->intent.open.file)) {
2221 error = PTR_ERR(nd->intent.open.file);
2222 goto exit_mutex_unlock;
2223 }
2224
2225 /* Negative dentry, just create the file */
2226 if (!path->dentry->d_inode) {
2227 /*
2228 * This write is needed to ensure that a
2229 * ro->rw transition does not occur between
2230 * the time when the file is created and when
2231 * a permanent write count is taken through
2232 * the 'struct file' in nameidata_to_filp().
2233 */
2234 error = mnt_want_write(nd->path.mnt);
2235 if (error)
2236 goto exit_mutex_unlock;
c3e380b0 2237 error = __open_namei_create(nd, path, op->open_flag, op->mode);
fb1cc555
AV		 2238 if (error) {
2239 mnt_drop_write(nd->path.mnt);
2240 goto exit;
2241 }
2242 filp = nameidata_to_filp(nd);
2243 mnt_drop_write(nd->path.mnt);
d893f1bc 2244 path_put(&nd->path);
fb1cc555 2245 if (!IS_ERR(filp)) {
c3e380b0 2246 error = ima_file_check(filp, op->acc_mode);
fb1cc555
AV		 2247 if (error) {
2248 fput(filp);
2249 filp = ERR_PTR(error);
2250 }
2251 }
2252 return filp;
2253 }
2254
2255 /*
2256 * It already exists.
2257 */
2258 mutex_unlock(&dir->d_inode->i_mutex);
2259 audit_inode(pathname, path->dentry);
2260
2261 error = -EEXIST;
c3e380b0 2262 if (op->open_flag & O_EXCL)
fb1cc555
AV		 2263 goto exit_dput;
2264
9875cf80
DH		 2265 error = follow_managed(path, nd->flags);
2266 if (error < 0)
2267 goto exit_dput;
fb1cc555
AV		 2268
2269 error = -ENOENT;
2270 if (!path->dentry->d_inode)
2271 goto exit_dput;
9e67f361
AV		 2272
2273 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2274 return NULL;
fb1cc555
AV		 2275
2276 path_to_nameidata(path, nd);
31e6b01f 2277 nd->inode = path->dentry->d_inode;
fb1cc555 2278 error = -EISDIR;
31e6b01f 2279 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2280 goto exit;
67ee3ad2 2281ok:
0f9d1a10
AV		 2282 will_truncate = open_will_truncate(op->open_flag, nd->path.dentry->d_inode);
2283 if (will_truncate) {
2284 error = mnt_want_write(nd->path.mnt);
2285 if (error)
2286 goto exit;
2287 }
2288 error = may_open(&nd->path, op->acc_mode, op->open_flag);
2289 if (error) {
2290 if (will_truncate)
2291 mnt_drop_write(nd->path.mnt);
2292 goto exit;
2293 }
2294 filp = nameidata_to_filp(nd);
2295 if (!IS_ERR(filp)) {
2296 error = ima_file_check(filp, op->acc_mode);
2297 if (error) {
2298 fput(filp);
2299 filp = ERR_PTR(error);
2300 }
2301 }
2302 if (!IS_ERR(filp)) {
2303 if (will_truncate) {
2304 error = handle_truncate(filp);
2305 if (error) {
2306 fput(filp);
2307 filp = ERR_PTR(error);
2308 }
2309 }
2310 }
2311 /*
2312 * It is now safe to drop the mnt write
2313 * because the filp has had a write taken
2314 * on its behalf.
2315 */
2316 if (will_truncate)
2317 mnt_drop_write(nd->path.mnt);
2318 path_put(&nd->path);
fb1cc555
AV		 2319 return filp;
2320
2321exit_mutex_unlock:
2322 mutex_unlock(&dir->d_inode->i_mutex);
2323exit_dput:
2324 path_put_conditional(path, nd);
2325exit:
fb1cc555
AV		 2326 path_put(&nd->path);
2327 return ERR_PTR(error);
2328}
2329
13aab428 2330static struct file *path_openat(int dfd, const char *pathname,
47c805dc 2331 const struct open_flags *op, int flags)
1da177e4 2332{
fe2d35ff 2333 struct file *base = NULL;
4a3fd211 2334 struct file *filp;
a70e65df 2335 struct nameidata nd;
9850c056 2336 struct path path;
1da177e4 2337 int count = 0;
13aab428 2338 int error;
31e6b01f
NP		 2339
2340 filp = get_empty_filp();
2341 if (!filp)
2342 return ERR_PTR(-ENFILE);
2343
47c805dc 2344 filp->f_flags = op->open_flag;
31e6b01f 2345 nd.intent.open.file = filp;
47c805dc
AV		 2346 nd.intent.open.flags = open_to_namei_flags(op->open_flag);
2347 nd.intent.open.create_mode = op->mode;
31e6b01f 2348
fe2d35ff 2349 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, &nd, &base);
31e6b01f 2350 if (unlikely(error))
13aab428 2351 goto out_filp;
31e6b01f 2352
fe2d35ff
AV		 2353 current->total_link_count = 0;
2354 error = link_path_walk(pathname, &nd);
31e6b01f
NP		 2355 if (unlikely(error))
2356 goto out_filp;
1da177e4 2357
47c805dc 2358 filp = do_last(&nd, &path, op, pathname);
806b681c 2359 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa
NP		 2360 struct path link = path;
2361 struct inode *linki = link.dentry->d_inode;
def4af30 2362 void *cookie;
806b681c 2363 error = -ELOOP;
db372915 2364 if (!(nd.flags & LOOKUP_FOLLOW))
1f36f774
AV		 2365 goto exit_dput;
2366 if (count++ == 32)
806b681c
AV		 2367 goto exit_dput;
2368 /*
2369 * This is subtle. Instead of calling do_follow_link() we do
2370 * the thing by hands. The reason is that this way we have zero
2371 * link_count and path_walk() (called from ->follow_link)
2372 * honoring LOOKUP_PARENT. After that we have the parent and
2373 * last component, i.e. we are in the same situation as after
2374 * the first path_walk(). Well, almost - if the last component
2375 * is normal we get its copy stored in nd->last.name and we will
2376 * have to putname() it when we are done. Procfs-like symlinks
2377 * just set LAST_BIND.
2378 */
2379 nd.flags |= LOOKUP_PARENT;
c3e380b0 2380 nd.flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
7b9337aa 2381 error = __do_follow_link(&link, &nd, &cookie);
c3e380b0 2382 if (unlikely(error))
f1afe9ef 2383 filp = ERR_PTR(error);
c3e380b0 2384 else
47c805dc 2385 filp = do_last(&nd, &path, op, pathname);
f1afe9ef 2386 if (!IS_ERR(cookie) && linki->i_op->put_link)
7b9337aa
NP		 2387 linki->i_op->put_link(link.dentry, &nd, cookie);
2388 path_put(&link);
806b681c 2389 }
10fa8e62 2390out:
2a737871
AV		 2391 if (nd.root.mnt)
2392 path_put(&nd.root);
fe2d35ff
AV		 2393 if (base)
2394 fput(base);
2dab5974 2395 release_open_intent(&nd);
10fa8e62 2396 return filp;
1da177e4 2397
806b681c
AV		 2398exit_dput:
2399 path_put_conditional(&path, &nd);
31e6b01f
NP		 2400 path_put(&nd.path);
2401out_filp:
806b681c 2402 filp = ERR_PTR(error);
10fa8e62 2403 goto out;
1da177e4
LT		 2404}
2405
13aab428
AV		 2406struct file *do_filp_open(int dfd, const char *pathname,
2407 const struct open_flags *op, int flags)
2408{
2409 struct file *filp;
2410
2411 filp = path_openat(dfd, pathname, op, flags | LOOKUP_RCU);
2412 if (unlikely(filp == ERR_PTR(-ECHILD)))
2413 filp = path_openat(dfd, pathname, op, flags);
2414 if (unlikely(filp == ERR_PTR(-ESTALE)))
2415 filp = path_openat(dfd, pathname, op, flags | LOOKUP_REVAL);
2416 return filp;
2417}
2418
1da177e4
LT		 2419/**
2420 * lookup_create - lookup a dentry, creating it if it doesn't exist
2421 * @nd: nameidata info
2422 * @is_dir: directory flag
2423 *
2424 * Simple function to lookup and return a dentry and create it
2425 * if it doesn't exist. Is SMP-safe.
c663e5d8 2426 *
4ac91378 2427 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2428 */
2429struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2430{
c663e5d8 2431 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2432
4ac91378 2433 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2434 /*
2435 * Yucky last component or no last component at all?
2436 * (foo/., foo/.., /////)
2437 */
1da177e4
LT		 2438 if (nd->last_type != LAST_NORM)
2439 goto fail;
2440 nd->flags &= ~LOOKUP_PARENT;
3516586a 2441 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2442 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2443
2444 /*
2445 * Do the final lookup.
2446 */
49705b77 2447 dentry = lookup_hash(nd);
1da177e4
LT		 2448 if (IS_ERR(dentry))
2449 goto fail;
c663e5d8 2450
e9baf6e5
AV		 2451 if (dentry->d_inode)
2452 goto eexist;
c663e5d8
CH		 2453 /*
2454 * Special case - lookup gave negative, but... we had foo/bar/
2455 * From the vfs_mknod() POV we just have a negative dentry -
2456 * all is fine. Let's be bastards - you had / on the end, you've
2457 * been asking for (non-existent) directory. -ENOENT for you.
2458 */
e9baf6e5
AV		 2459 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2460 dput(dentry);
2461 dentry = ERR_PTR(-ENOENT);
2462 }
1da177e4 2463 return dentry;
e9baf6e5 2464eexist:
1da177e4 2465 dput(dentry);
e9baf6e5 2466 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2467fail:
2468 return dentry;
2469}
f81a0bff 2470EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2471
2472int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2473{
a95164d9 2474 int error = may_create(dir, dentry);
1da177e4
LT		 2475
2476 if (error)
2477 return error;
2478
2479 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2480 return -EPERM;
2481
acfa4380 2482 if (!dir->i_op->mknod)
1da177e4
LT		 2483 return -EPERM;
2484
08ce5f16
SH		 2485 error = devcgroup_inode_mknod(mode, dev);
2486 if (error)
2487 return error;
2488
1da177e4
LT		 2489 error = security_inode_mknod(dir, dentry, mode, dev);
2490 if (error)
2491 return error;
2492
1da177e4 2493 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2494 if (!error)
f38aa942 2495 fsnotify_create(dir, dentry);
1da177e4
LT		 2496 return error;
2497}
2498
463c3197
DH		 2499static int may_mknod(mode_t mode)
2500{
2501 switch (mode & S_IFMT) {
2502 case S_IFREG:
2503 case S_IFCHR:
2504 case S_IFBLK:
2505 case S_IFIFO:
2506 case S_IFSOCK:
2507 case 0: /* zero mode translates to S_IFREG */
2508 return 0;
2509 case S_IFDIR:
2510 return -EPERM;
2511 default:
2512 return -EINVAL;
2513 }
2514}
2515
2e4d0924
HC		 2516SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2517 unsigned, dev)
1da177e4 2518{
2ad94ae6
AV		 2519 int error;
2520 char *tmp;
2521 struct dentry *dentry;
1da177e4
LT		 2522 struct nameidata nd;
2523
2524 if (S_ISDIR(mode))
2525 return -EPERM;
1da177e4 2526
2ad94ae6 2527 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2528 if (error)
2ad94ae6
AV		 2529 return error;
2530
1da177e4 2531 dentry = lookup_create(&nd, 0);
463c3197
DH		 2532 if (IS_ERR(dentry)) {
2533 error = PTR_ERR(dentry);
2534 goto out_unlock;
2535 }
4ac91378 2536 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2537 mode &= ~current_umask();
463c3197
DH		 2538 error = may_mknod(mode);
2539 if (error)
2540 goto out_dput;
2541 error = mnt_want_write(nd.path.mnt);
2542 if (error)
2543 goto out_dput;
be6d3e56
KT		 2544 error = security_path_mknod(&nd.path, dentry, mode, dev);
2545 if (error)
2546 goto out_drop_write;
463c3197 2547 switch (mode & S_IFMT) {
1da177e4 2548 case 0: case S_IFREG:
4ac91378 2549 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2550 break;
2551 case S_IFCHR: case S_IFBLK:
4ac91378 2552 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2553 new_decode_dev(dev));
2554 break;
2555 case S_IFIFO: case S_IFSOCK:
4ac91378 2556 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2557 break;
1da177e4 2558 }
be6d3e56 2559out_drop_write:
463c3197
DH		 2560 mnt_drop_write(nd.path.mnt);
2561out_dput:
2562 dput(dentry);
2563out_unlock:
4ac91378 2564 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2565 path_put(&nd.path);
1da177e4
LT		 2566 putname(tmp);
2567
2568 return error;
2569}
2570
3480b257 2571SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2572{
2573 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2574}
2575
1da177e4
LT		 2576int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2577{
a95164d9 2578 int error = may_create(dir, dentry);
1da177e4
LT		 2579
2580 if (error)
2581 return error;
2582
acfa4380 2583 if (!dir->i_op->mkdir)
1da177e4
LT		 2584 return -EPERM;
2585
2586 mode &= (S_IRWXUGO|S_ISVTX);
2587 error = security_inode_mkdir(dir, dentry, mode);
2588 if (error)
2589 return error;
2590
1da177e4 2591 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2592 if (!error)
f38aa942 2593 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2594 return error;
2595}
2596
2e4d0924 2597SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2598{
2599 int error = 0;
2600 char * tmp;
6902d925
DH		 2601 struct dentry *dentry;
2602 struct nameidata nd;
1da177e4 2603
2ad94ae6
AV		 2604 error = user_path_parent(dfd, pathname, &nd, &tmp);
2605 if (error)
6902d925 2606 goto out_err;
1da177e4 2607
6902d925
DH		 2608 dentry = lookup_create(&nd, 1);
2609 error = PTR_ERR(dentry);
2610 if (IS_ERR(dentry))
2611 goto out_unlock;
1da177e4 2612
4ac91378 2613 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2614 mode &= ~current_umask();
463c3197
DH		 2615 error = mnt_want_write(nd.path.mnt);
2616 if (error)
2617 goto out_dput;
be6d3e56
KT		 2618 error = security_path_mkdir(&nd.path, dentry, mode);
2619 if (error)
2620 goto out_drop_write;
4ac91378 2621 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2622out_drop_write:
463c3197
DH		 2623 mnt_drop_write(nd.path.mnt);
2624out_dput:
6902d925
DH		 2625 dput(dentry);
2626out_unlock:
4ac91378 2627 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2628 path_put(&nd.path);
6902d925
DH		 2629 putname(tmp);
2630out_err:
1da177e4
LT		 2631 return error;
2632}
2633
3cdad428 2634SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2635{
2636 return sys_mkdirat(AT_FDCWD, pathname, mode);
2637}
2638
1da177e4
LT		 2639/*
2640 * We try to drop the dentry early: we should have
2641 * a usage count of 2 if we're the only user of this
2642 * dentry, and if that is true (possibly after pruning
2643 * the dcache), then we drop the dentry now.
2644 *
2645 * A low-level filesystem can, if it choses, legally
2646 * do a
2647 *
2648 * if (!d_unhashed(dentry))
2649 * return -EBUSY;
2650 *
2651 * if it cannot handle the case of removing a directory
2652 * that is still in use by something else..
2653 */
2654void dentry_unhash(struct dentry *dentry)
2655{
2656 dget(dentry);
dc168427 2657 shrink_dcache_parent(dentry);
1da177e4 2658 spin_lock(&dentry->d_lock);
b7ab39f6 2659 if (dentry->d_count == 2)
1da177e4
LT		 2660 __d_drop(dentry);
2661 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2662}
2663
2664int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2665{
2666 int error = may_delete(dir, dentry, 1);
2667
2668 if (error)
2669 return error;
2670
acfa4380 2671 if (!dir->i_op->rmdir)
1da177e4
LT		 2672 return -EPERM;
2673
1b1dcc1b 2674 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2675 dentry_unhash(dentry);
2676 if (d_mountpoint(dentry))
2677 error = -EBUSY;
2678 else {
2679 error = security_inode_rmdir(dir, dentry);
2680 if (!error) {
2681 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2682 if (!error) {
1da177e4 2683 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2684 dont_mount(dentry);
2685 }
1da177e4
LT		 2686 }
2687 }
1b1dcc1b 2688 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2689 if (!error) {
1da177e4
LT		 2690 d_delete(dentry);
2691 }
2692 dput(dentry);
2693
2694 return error;
2695}
2696
5590ff0d 2697static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2698{
2699 int error = 0;
2700 char * name;
2701 struct dentry *dentry;
2702 struct nameidata nd;
2703
2ad94ae6 2704 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2705 if (error)
2ad94ae6 2706 return error;
1da177e4
LT		 2707
2708 switch(nd.last_type) {
0612d9fb
OH		 2709 case LAST_DOTDOT:
2710 error = -ENOTEMPTY;
2711 goto exit1;
2712 case LAST_DOT:
2713 error = -EINVAL;
2714 goto exit1;
2715 case LAST_ROOT:
2716 error = -EBUSY;
2717 goto exit1;
1da177e4 2718 }
0612d9fb
OH		 2719
2720 nd.flags &= ~LOOKUP_PARENT;
2721
4ac91378 2722 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2723 dentry = lookup_hash(&nd);
1da177e4 2724 error = PTR_ERR(dentry);
6902d925
DH		 2725 if (IS_ERR(dentry))
2726 goto exit2;
0622753b
DH		 2727 error = mnt_want_write(nd.path.mnt);
2728 if (error)
2729 goto exit3;
be6d3e56
KT		 2730 error = security_path_rmdir(&nd.path, dentry);
2731 if (error)
2732 goto exit4;
4ac91378 2733 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2734exit4:
0622753b
DH		 2735 mnt_drop_write(nd.path.mnt);
2736exit3:
6902d925
DH		 2737 dput(dentry);
2738exit2:
4ac91378 2739 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2740exit1:
1d957f9b 2741 path_put(&nd.path);
1da177e4
LT		 2742 putname(name);
2743 return error;
2744}
2745
3cdad428 2746SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2747{
2748 return do_rmdir(AT_FDCWD, pathname);
2749}
2750
1da177e4
LT		 2751int vfs_unlink(struct inode *dir, struct dentry *dentry)
2752{
2753 int error = may_delete(dir, dentry, 0);
2754
2755 if (error)
2756 return error;
2757
acfa4380 2758 if (!dir->i_op->unlink)
1da177e4
LT		 2759 return -EPERM;
2760
1b1dcc1b 2761 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2762 if (d_mountpoint(dentry))
2763 error = -EBUSY;
2764 else {
2765 error = security_inode_unlink(dir, dentry);
bec1052e 2766 if (!error) {
1da177e4 2767 error = dir->i_op->unlink(dir, dentry);
bec1052e 2768 if (!error)
d83c49f3 2769 dont_mount(dentry);
bec1052e 2770 }
1da177e4 2771 }
1b1dcc1b 2772 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2773
2774 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2775 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2776 fsnotify_link_count(dentry->d_inode);
e234f35c 2777 d_delete(dentry);
1da177e4 2778 }
0eeca283 2779
1da177e4
LT		 2780 return error;
2781}
2782
2783/*
2784 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2785 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2786 * writeout happening, and we don't want to prevent access to the directory
2787 * while waiting on the I/O.
2788 */
5590ff0d 2789static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2790{
2ad94ae6
AV		 2791 int error;
2792 char *name;
1da177e4
LT		 2793 struct dentry *dentry;
2794 struct nameidata nd;
2795 struct inode *inode = NULL;
2796
2ad94ae6 2797 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2798 if (error)
2ad94ae6
AV		 2799 return error;
2800
1da177e4
LT		 2801 error = -EISDIR;
2802 if (nd.last_type != LAST_NORM)
2803 goto exit1;
0612d9fb
OH		 2804
2805 nd.flags &= ~LOOKUP_PARENT;
2806
4ac91378 2807 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2808 dentry = lookup_hash(&nd);
1da177e4
LT		 2809 error = PTR_ERR(dentry);
2810 if (!IS_ERR(dentry)) {
2811 /* Why not before? Because we want correct error value */
2812 if (nd.last.name[nd.last.len])
2813 goto slashes;
2814 inode = dentry->d_inode;
2815 if (inode)
7de9c6ee 2816 ihold(inode);
0622753b
DH		 2817 error = mnt_want_write(nd.path.mnt);
2818 if (error)
2819 goto exit2;
be6d3e56
KT		 2820 error = security_path_unlink(&nd.path, dentry);
2821 if (error)
2822 goto exit3;
4ac91378 2823 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2824exit3:
0622753b 2825 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2826 exit2:
2827 dput(dentry);
2828 }
4ac91378 2829 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2830 if (inode)
2831 iput(inode); /* truncate the inode here */
2832exit1:
1d957f9b 2833 path_put(&nd.path);
1da177e4
LT		 2834 putname(name);
2835 return error;
2836
2837slashes:
2838 error = !dentry->d_inode ? -ENOENT :
2839 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2840 goto exit2;
2841}
2842
2e4d0924 2843SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2844{
2845 if ((flag & ~AT_REMOVEDIR) != 0)
2846 return -EINVAL;
2847
2848 if (flag & AT_REMOVEDIR)
2849 return do_rmdir(dfd, pathname);
2850
2851 return do_unlinkat(dfd, pathname);
2852}
2853
3480b257 2854SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2855{
2856 return do_unlinkat(AT_FDCWD, pathname);
2857}
2858
db2e747b 2859int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2860{
a95164d9 2861 int error = may_create(dir, dentry);
1da177e4
LT		 2862
2863 if (error)
2864 return error;
2865
acfa4380 2866 if (!dir->i_op->symlink)
1da177e4
LT		 2867 return -EPERM;
2868
2869 error = security_inode_symlink(dir, dentry, oldname);
2870 if (error)
2871 return error;
2872
1da177e4 2873 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2874 if (!error)
f38aa942 2875 fsnotify_create(dir, dentry);
1da177e4
LT		 2876 return error;
2877}
2878
2e4d0924
HC		 2879SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2880 int, newdfd, const char __user *, newname)
1da177e4 2881{
2ad94ae6
AV		 2882 int error;
2883 char *from;
2884 char *to;
6902d925
DH		 2885 struct dentry *dentry;
2886 struct nameidata nd;
1da177e4
LT		 2887
2888 from = getname(oldname);
2ad94ae6 2889 if (IS_ERR(from))
1da177e4 2890 return PTR_ERR(from);
1da177e4 2891
2ad94ae6 2892 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2893 if (error)
2ad94ae6
AV		 2894 goto out_putname;
2895
6902d925
DH		 2896 dentry = lookup_create(&nd, 0);
2897 error = PTR_ERR(dentry);
2898 if (IS_ERR(dentry))
2899 goto out_unlock;
2900
75c3f29d
DH		 2901 error = mnt_want_write(nd.path.mnt);
2902 if (error)
2903 goto out_dput;
be6d3e56
KT		 2904 error = security_path_symlink(&nd.path, dentry, from);
2905 if (error)
2906 goto out_drop_write;
db2e747b 2907 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2908out_drop_write:
75c3f29d
DH		 2909 mnt_drop_write(nd.path.mnt);
2910out_dput:
6902d925
DH		 2911 dput(dentry);
2912out_unlock:
4ac91378 2913 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2914 path_put(&nd.path);
6902d925
DH		 2915 putname(to);
2916out_putname:
1da177e4
LT		 2917 putname(from);
2918 return error;
2919}
2920
3480b257 2921SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2922{
2923 return sys_symlinkat(oldname, AT_FDCWD, newname);
2924}
2925
1da177e4
LT		 2926int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2927{
2928 struct inode *inode = old_dentry->d_inode;
2929 int error;
2930
2931 if (!inode)
2932 return -ENOENT;
2933
a95164d9 2934 error = may_create(dir, new_dentry);
1da177e4
LT		 2935 if (error)
2936 return error;
2937
2938 if (dir->i_sb != inode->i_sb)
2939 return -EXDEV;
2940
2941 /*
2942 * A link to an append-only or immutable file cannot be created.
2943 */
2944 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2945 return -EPERM;
acfa4380 2946 if (!dir->i_op->link)
1da177e4 2947 return -EPERM;
7e79eedb 2948 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2949 return -EPERM;
2950
2951 error = security_inode_link(old_dentry, dir, new_dentry);
2952 if (error)
2953 return error;
2954
7e79eedb 2955 mutex_lock(&inode->i_mutex);
1da177e4 2956 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2957 mutex_unlock(&inode->i_mutex);
e31e14ec 2958 if (!error)
7e79eedb 2959 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2960 return error;
2961}
2962
2963/*
2964 * Hardlinks are often used in delicate situations. We avoid
2965 * security-related surprises by not following symlinks on the
2966 * newname. --KAB
2967 *
2968 * We don't follow them on the oldname either to be compatible
2969 * with linux 2.0, and to avoid hard-linking to directories
2970 * and other special files. --ADM
2971 */
2e4d0924
HC		 2972SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2973 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2974{
2975 struct dentry *new_dentry;
2d8f3038
AV		 2976 struct nameidata nd;
2977 struct path old_path;
1da177e4 2978 int error;
2ad94ae6 2979 char *to;
1da177e4 2980
45c9b11a 2981 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 2982 return -EINVAL;
2983
2d8f3038
AV		 2984 error = user_path_at(olddfd, oldname,
2985 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2986 &old_path);
1da177e4 2987 if (error)
2ad94ae6
AV		 2988 return error;
2989
2990 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2991 if (error)
2992 goto out;
2993 error = -EXDEV;
2d8f3038 2994 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2995 goto out_release;
2996 new_dentry = lookup_create(&nd, 0);
2997 error = PTR_ERR(new_dentry);
6902d925
DH		 2998 if (IS_ERR(new_dentry))
2999 goto out_unlock;
75c3f29d
DH		 3000 error = mnt_want_write(nd.path.mnt);
3001 if (error)
3002 goto out_dput;
be6d3e56
KT		 3003 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
3004 if (error)
3005 goto out_drop_write;
2d8f3038 3006 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 3007out_drop_write:
75c3f29d
DH		 3008 mnt_drop_write(nd.path.mnt);
3009out_dput:
6902d925
DH		 3010 dput(new_dentry);
3011out_unlock:
4ac91378 3012 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 3013out_release:
1d957f9b 3014 path_put(&nd.path);
2ad94ae6 3015 putname(to);
1da177e4 3016out:
2d8f3038 3017 path_put(&old_path);
1da177e4
LT		 3018
3019 return error;
3020}
3021
3480b257 3022SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3023{
c04030e1 3024 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3025}
3026
1da177e4
LT		 3027/*
3028 * The worst of all namespace operations - renaming directory. "Perverted"
3029 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3030 * Problems:
3031 * a) we can get into loop creation. Check is done in is_subdir().
3032 * b) race potential - two innocent renames can create a loop together.
3033 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3034 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3035 * story.
3036 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3037 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3038 * whether the target exists). Solution: try to be smart with locking
3039 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3040 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3041 * move will be locked. Thus we can rank directories by the tree
3042 * (ancestors first) and rank all non-directories after them.
3043 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3044 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3045 * HOWEVER, it relies on the assumption that any object with ->lookup()
3046 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3047 * we'd better make sure that there's no link(2) for them.
3048 * d) some filesystems don't support opened-but-unlinked directories,
3049 * either because of layout or because they are not ready to deal with
3050 * all cases correctly. The latter will be fixed (taking this sort of
3051 * stuff into VFS), but the former is not going away. Solution: the same
3052 * trick as in rmdir().
3053 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3054 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3055 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3056 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3057 * locking].
3058 */
75c96f85
AB		 3059static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3060 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3061{
3062 int error = 0;
3063 struct inode *target;
3064
3065 /*
3066 * If we are going to change the parent - check write permissions,
3067 * we'll need to flip '..'.
3068 */
3069 if (new_dir != old_dir) {
f419a2e3 3070 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3071 if (error)
3072 return error;
3073 }
3074
3075 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3076 if (error)
3077 return error;
3078
3079 target = new_dentry->d_inode;
d83c49f3 3080 if (target)
1b1dcc1b 3081 mutex_lock(&target->i_mutex);
1da177e4
LT		 3082 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3083 error = -EBUSY;
d83c49f3
AV		 3084 else {
3085 if (target)
3086 dentry_unhash(new_dentry);
1da177e4 3087 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3088 }
1da177e4 3089 if (target) {
d83c49f3 3090 if (!error) {
1da177e4 3091 target->i_flags |= S_DEAD;
d83c49f3
AV		 3092 dont_mount(new_dentry);
3093 }
1b1dcc1b 3094 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3095 if (d_unhashed(new_dentry))
3096 d_rehash(new_dentry);
3097 dput(new_dentry);
3098 }
e31e14ec 3099 if (!error)
349457cc
MF		 3100 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3101 d_move(old_dentry,new_dentry);
1da177e4
LT		 3102 return error;
3103}
3104
75c96f85
AB		 3105static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3106 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3107{
3108 struct inode *target;
3109 int error;
3110
3111 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3112 if (error)
3113 return error;
3114
3115 dget(new_dentry);
3116 target = new_dentry->d_inode;
3117 if (target)
1b1dcc1b 3118 mutex_lock(&target->i_mutex);
1da177e4
LT		 3119 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3120 error = -EBUSY;
3121 else
3122 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3123 if (!error) {
bec1052e 3124 if (target)
d83c49f3 3125 dont_mount(new_dentry);
349457cc 3126 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3127 d_move(old_dentry, new_dentry);
1da177e4
LT		 3128 }
3129 if (target)
1b1dcc1b 3130 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3131 dput(new_dentry);
3132 return error;
3133}
3134
3135int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3136 struct inode *new_dir, struct dentry *new_dentry)
3137{
3138 int error;
3139 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3140 const unsigned char *old_name;
1da177e4
LT		 3141
3142 if (old_dentry->d_inode == new_dentry->d_inode)
3143 return 0;
3144
3145 error = may_delete(old_dir, old_dentry, is_dir);
3146 if (error)
3147 return error;
3148
3149 if (!new_dentry->d_inode)
a95164d9 3150 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3151 else
3152 error = may_delete(new_dir, new_dentry, is_dir);
3153 if (error)
3154 return error;
3155
acfa4380 3156 if (!old_dir->i_op->rename)
1da177e4
LT		 3157 return -EPERM;
3158
0eeca283
RL		 3159 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3160
1da177e4
LT		 3161 if (is_dir)
3162 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3163 else
3164 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3165 if (!error)
3166 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3167 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3168 fsnotify_oldname_free(old_name);
3169
1da177e4
LT		 3170 return error;
3171}
3172
2e4d0924
HC		 3173SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3174 int, newdfd, const char __user *, newname)
1da177e4 3175{
2ad94ae6
AV		 3176 struct dentry *old_dir, *new_dir;
3177 struct dentry *old_dentry, *new_dentry;
3178 struct dentry *trap;
1da177e4 3179 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3180 char *from;
3181 char *to;
3182 int error;
1da177e4 3183
2ad94ae6 3184 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3185 if (error)
3186 goto exit;
3187
2ad94ae6 3188 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3189 if (error)
3190 goto exit1;
3191
3192 error = -EXDEV;
4ac91378 3193 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3194 goto exit2;
3195
4ac91378 3196 old_dir = oldnd.path.dentry;
1da177e4
LT		 3197 error = -EBUSY;
3198 if (oldnd.last_type != LAST_NORM)
3199 goto exit2;
3200
4ac91378 3201 new_dir = newnd.path.dentry;
1da177e4
LT		 3202 if (newnd.last_type != LAST_NORM)
3203 goto exit2;
3204
0612d9fb
OH		 3205 oldnd.flags &= ~LOOKUP_PARENT;
3206 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3207 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3208
1da177e4
LT		 3209 trap = lock_rename(new_dir, old_dir);
3210
49705b77 3211 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3212 error = PTR_ERR(old_dentry);
3213 if (IS_ERR(old_dentry))
3214 goto exit3;
3215 /* source must exist */
3216 error = -ENOENT;
3217 if (!old_dentry->d_inode)
3218 goto exit4;
3219 /* unless the source is a directory trailing slashes give -ENOTDIR */
3220 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3221 error = -ENOTDIR;
3222 if (oldnd.last.name[oldnd.last.len])
3223 goto exit4;
3224 if (newnd.last.name[newnd.last.len])
3225 goto exit4;
3226 }
3227 /* source should not be ancestor of target */
3228 error = -EINVAL;
3229 if (old_dentry == trap)
3230 goto exit4;
49705b77 3231 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3232 error = PTR_ERR(new_dentry);
3233 if (IS_ERR(new_dentry))
3234 goto exit4;
3235 /* target should not be an ancestor of source */
3236 error = -ENOTEMPTY;
3237 if (new_dentry == trap)
3238 goto exit5;
3239
9079b1eb
DH		 3240 error = mnt_want_write(oldnd.path.mnt);
3241 if (error)
3242 goto exit5;
be6d3e56
KT		 3243 error = security_path_rename(&oldnd.path, old_dentry,
3244 &newnd.path, new_dentry);
3245 if (error)
3246 goto exit6;
1da177e4
LT		 3247 error = vfs_rename(old_dir->d_inode, old_dentry,
3248 new_dir->d_inode, new_dentry);
be6d3e56 3249exit6:
9079b1eb 3250 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3251exit5:
3252 dput(new_dentry);
3253exit4:
3254 dput(old_dentry);
3255exit3:
3256 unlock_rename(new_dir, old_dir);
3257exit2:
1d957f9b 3258 path_put(&newnd.path);
2ad94ae6 3259 putname(to);
1da177e4 3260exit1:
1d957f9b 3261 path_put(&oldnd.path);
1da177e4 3262 putname(from);
2ad94ae6 3263exit:
1da177e4
LT		 3264 return error;
3265}
3266
a26eab24 3267SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3268{
3269 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3270}
3271
1da177e4
LT		 3272int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3273{
3274 int len;
3275
3276 len = PTR_ERR(link);
3277 if (IS_ERR(link))
3278 goto out;
3279
3280 len = strlen(link);
3281 if (len > (unsigned) buflen)
3282 len = buflen;
3283 if (copy_to_user(buffer, link, len))
3284 len = -EFAULT;
3285out:
3286 return len;
3287}
3288
3289/*
3290 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3291 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3292 * using) it for any given inode is up to filesystem.
3293 */
3294int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3295{
3296 struct nameidata nd;
cc314eef 3297 void *cookie;
694a1764 3298 int res;
cc314eef 3299
1da177e4 3300 nd.depth = 0;
cc314eef 3301 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3302 if (IS_ERR(cookie))
3303 return PTR_ERR(cookie);
3304
3305 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3306 if (dentry->d_inode->i_op->put_link)
3307 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3308 return res;
1da177e4
LT		 3309}
3310
3311int vfs_follow_link(struct nameidata *nd, const char *link)
3312{
3313 return __vfs_follow_link(nd, link);
3314}
3315
3316/* get the link contents into pagecache */
3317static char *page_getlink(struct dentry * dentry, struct page **ppage)
3318{
ebd09abb
DG		 3319 char *kaddr;
3320 struct page *page;
1da177e4 3321 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3322 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3323 if (IS_ERR(page))
6fe6900e 3324 return (char*)page;
1da177e4 3325 *ppage = page;
ebd09abb
DG		 3326 kaddr = kmap(page);
3327 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3328 return kaddr;
1da177e4
LT		 3329}
3330
3331int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3332{
3333 struct page *page = NULL;
3334 char *s = page_getlink(dentry, &page);
3335 int res = vfs_readlink(dentry,buffer,buflen,s);
3336 if (page) {
3337 kunmap(page);
3338 page_cache_release(page);
3339 }
3340 return res;
3341}
3342
cc314eef 3343void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3344{
cc314eef 3345 struct page *page = NULL;
1da177e4 3346 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3347 return page;
1da177e4
LT		 3348}
3349
cc314eef 3350void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3351{
cc314eef
LT		 3352 struct page *page = cookie;
3353
3354 if (page) {
1da177e4
LT		 3355 kunmap(page);
3356 page_cache_release(page);
1da177e4
LT		 3357 }
3358}
3359
54566b2c
NP		 3360/*
3361 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3362 */
3363int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3364{
3365 struct address_space *mapping = inode->i_mapping;
0adb25d2 3366 struct page *page;
afddba49 3367 void *fsdata;
beb497ab 3368 int err;
1da177e4 3369 char *kaddr;
54566b2c
NP		 3370 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3371 if (nofs)
3372 flags |= AOP_FLAG_NOFS;
1da177e4 3373
7e53cac4 3374retry:
afddba49 3375 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3376 flags, &page, &fsdata);
1da177e4 3377 if (err)
afddba49
NP		 3378 goto fail;
3379
1da177e4
LT		 3380 kaddr = kmap_atomic(page, KM_USER0);
3381 memcpy(kaddr, symname, len-1);
3382 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3383
3384 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3385 page, fsdata);
1da177e4
LT		 3386 if (err < 0)
3387 goto fail;
afddba49
NP		 3388 if (err < len-1)
3389 goto retry;
3390
1da177e4
LT		 3391 mark_inode_dirty(inode);
3392 return 0;
1da177e4
LT		 3393fail:
3394 return err;
3395}
3396
0adb25d2
KK		 3397int page_symlink(struct inode *inode, const char *symname, int len)
3398{
3399 return __page_symlink(inode, symname, len,
54566b2c 3400 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3401}
3402
92e1d5be 3403const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3404 .readlink = generic_readlink,
3405 .follow_link = page_follow_link_light,
3406 .put_link = page_put_link,
3407};
3408
2d8f3038 3409EXPORT_SYMBOL(user_path_at);
cc53ce53 3410EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3411EXPORT_SYMBOL(follow_down);
3412EXPORT_SYMBOL(follow_up);
3413EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3414EXPORT_SYMBOL(getname);
3415EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3416EXPORT_SYMBOL(lookup_one_len);
3417EXPORT_SYMBOL(page_follow_link_light);
3418EXPORT_SYMBOL(page_put_link);
3419EXPORT_SYMBOL(page_readlink);
0adb25d2 3420EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3421EXPORT_SYMBOL(page_symlink);
3422EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3423EXPORT_SYMBOL(kern_path_parent);
d1811465 3424EXPORT_SYMBOL(kern_path);
16f18200 3425EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3426EXPORT_SYMBOL(inode_permission);
8c744fb8 3427EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3428EXPORT_SYMBOL(unlock_rename);
3429EXPORT_SYMBOL(vfs_create);
3430EXPORT_SYMBOL(vfs_follow_link);
3431EXPORT_SYMBOL(vfs_link);
3432EXPORT_SYMBOL(vfs_mkdir);
3433EXPORT_SYMBOL(vfs_mknod);
3434EXPORT_SYMBOL(generic_permission);
3435EXPORT_SYMBOL(vfs_readlink);
3436EXPORT_SYMBOL(vfs_rename);
3437EXPORT_SYMBOL(vfs_rmdir);
3438EXPORT_SYMBOL(vfs_symlink);
3439EXPORT_SYMBOL(vfs_unlink);
3440EXPORT_SYMBOL(dentry_unhash);
3441EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)