projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
get rid of the last LOOKUP_RCU dependencies in link_path_walk()
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / namei.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
1da177e4 22#include <linux/pagemap.h>
0eeca283 23#include <linux/fsnotify.h>
1da177e4
LT		 24#include <linux/personality.h>
25#include <linux/security.h>
6146f0d5 26#include <linux/ima.h>
1da177e4
LT		 27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
16f7e0fe 30#include <linux/capability.h>
834f2a4a 31#include <linux/file.h>
5590ff0d 32#include <linux/fcntl.h>
08ce5f16 33#include <linux/device_cgroup.h>
5ad4e53b 34#include <linux/fs_struct.h>
1da177e4
LT		 35#include <asm/uaccess.h>
36
e81e3f4d
EP		 37#include "internal.h"
38
1da177e4
LT		 39/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 107 * any extra contention...
108 */
109
110/* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
113 *
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
116 */
858119e1 117static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 118{
119 int retval;
120 unsigned long len = PATH_MAX;
121
122 if (!segment_eq(get_fs(), KERNEL_DS)) {
123 if ((unsigned long) filename >= TASK_SIZE)
124 return -EFAULT;
125 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126 len = TASK_SIZE - (unsigned long) filename;
127 }
128
129 retval = strncpy_from_user(page, filename, len);
130 if (retval > 0) {
131 if (retval < len)
132 return 0;
133 return -ENAMETOOLONG;
134 } else if (!retval)
135 retval = -ENOENT;
136 return retval;
137}
138
139char * getname(const char __user * filename)
140{
141 char *tmp, *result;
142
143 result = ERR_PTR(-ENOMEM);
144 tmp = __getname();
145 if (tmp) {
146 int retval = do_getname(filename, tmp);
147
148 result = tmp;
149 if (retval < 0) {
150 __putname(tmp);
151 result = ERR_PTR(retval);
152 }
153 }
154 audit_getname(result);
155 return result;
156}
157
158#ifdef CONFIG_AUDITSYSCALL
159void putname(const char *name)
160{
5ac3a9c2 161 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 162 audit_putname(name);
163 else
164 __putname(name);
165}
166EXPORT_SYMBOL(putname);
167#endif
168
5909ccaa
LT		 169/*
170 * This does basic POSIX ACL permission checking
1da177e4 171 */
b74c79e9
NP		 172static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
173 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
1da177e4
LT		 174{
175 umode_t mode = inode->i_mode;
176
e6305c43
AV		 177 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
178
da9592ed 179 if (current_fsuid() == inode->i_uid)
1da177e4
LT		 180 mode >>= 6;
181 else {
182 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
b74c79e9
NP		 183 int error = check_acl(inode, mask, flags);
184 if (error != -EAGAIN)
185 return error;
1da177e4
LT		 186 }
187
188 if (in_group_p(inode->i_gid))
189 mode >>= 3;
190 }
191
192 /*
193 * If the DACs are ok we don't need any capability check.
194 */
e6305c43 195 if ((mask & ~mode) == 0)
1da177e4 196 return 0;
5909ccaa
LT		 197 return -EACCES;
198}
199
200/**
b74c79e9 201 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa
LT		 202 * @inode: inode to check access rights for
203 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
204 * @check_acl: optional callback to check for Posix ACLs
39191628 205 * @flags: IPERM_FLAG_ flags.
5909ccaa
LT		 206 *
207 * Used to check for read/write/execute permissions on a file.
208 * We use "fsuid" for this, letting us set arbitrary permissions
209 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 210 * are used for other things.
211 *
212 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
213 * request cannot be satisfied (eg. requires blocking or too much complexity).
214 * It would then be called again in ref-walk mode.
5909ccaa 215 */
b74c79e9
NP		 216int generic_permission(struct inode *inode, int mask, unsigned int flags,
217 int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
5909ccaa
LT		 218{
219 int ret;
220
221 /*
222 * Do the basic POSIX ACL permission checks.
223 */
b74c79e9 224 ret = acl_permission_check(inode, mask, flags, check_acl);
5909ccaa
LT		 225 if (ret != -EACCES)
226 return ret;
1da177e4 227
1da177e4
LT		 228 /*
229 * Read/write DACs are always overridable.
230 * Executable DACs are overridable if at least one exec bit is set.
231 */
f696a365 232 if (!(mask & MAY_EXEC) || execute_ok(inode))
1da177e4
LT		 233 if (capable(CAP_DAC_OVERRIDE))
234 return 0;
235
236 /*
237 * Searching includes executable on directories, else just read.
238 */
7ea66001 239 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
1da177e4
LT		 240 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
241 if (capable(CAP_DAC_READ_SEARCH))
242 return 0;
243
244 return -EACCES;
245}
246
cb23beb5
CH		 247/**
248 * inode_permission - check for access rights to a given inode
249 * @inode: inode to check permission on
250 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
251 *
252 * Used to check for read/write/execute permissions on an inode.
253 * We use "fsuid" for this, letting us set arbitrary permissions
254 * for filesystem access without changing the "normal" uids which
255 * are used for other things.
256 */
f419a2e3 257int inode_permission(struct inode *inode, int mask)
1da177e4 258{
e6305c43 259 int retval;
1da177e4
LT		 260
261 if (mask & MAY_WRITE) {
22590e41 262 umode_t mode = inode->i_mode;
1da177e4
LT		 263
264 /*
265 * Nobody gets write access to a read-only fs.
266 */
267 if (IS_RDONLY(inode) &&
268 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
269 return -EROFS;
270
271 /*
272 * Nobody gets write access to an immutable file.
273 */
274 if (IS_IMMUTABLE(inode))
275 return -EACCES;
276 }
277
acfa4380 278 if (inode->i_op->permission)
b74c79e9 279 retval = inode->i_op->permission(inode, mask, 0);
f696a365 280 else
b74c79e9
NP		 281 retval = generic_permission(inode, mask, 0,
282 inode->i_op->check_acl);
f696a365 283
1da177e4
LT		 284 if (retval)
285 return retval;
286
08ce5f16
SH		 287 retval = devcgroup_inode_permission(inode, mask);
288 if (retval)
289 return retval;
290
d09ca739 291 return security_inode_permission(inode, mask);
1da177e4
LT		 292}
293
8c744fb8
CH		 294/**
295 * file_permission - check for additional access rights to a given file
296 * @file: file to check access rights for
297 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
298 *
299 * Used to check for read/write/execute permissions on an already opened
300 * file.
301 *
302 * Note:
303 * Do not use this function in new code. All access checks should
cb23beb5 304 * be done using inode_permission().
8c744fb8
CH		 305 */
306int file_permission(struct file *file, int mask)
307{
f419a2e3 308 return inode_permission(file->f_path.dentry->d_inode, mask);
8c744fb8
CH		 309}
310
1da177e4
LT		 311/*
312 * get_write_access() gets write permission for a file.
313 * put_write_access() releases this write permission.
314 * This is used for regular files.
315 * We cannot support write (and maybe mmap read-write shared) accesses and
316 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
317 * can have the following values:
318 * 0: no writers, no VM_DENYWRITE mappings
319 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
320 * > 0: (i_writecount) users are writing to the file.
321 *
322 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
323 * except for the cases where we don't hold i_writecount yet. Then we need to
324 * use {get,deny}_write_access() - these functions check the sign and refuse
325 * to do the change if sign is wrong. Exclusion between them is provided by
326 * the inode->i_lock spinlock.
327 */
328
329int get_write_access(struct inode * inode)
330{
331 spin_lock(&inode->i_lock);
332 if (atomic_read(&inode->i_writecount) < 0) {
333 spin_unlock(&inode->i_lock);
334 return -ETXTBSY;
335 }
336 atomic_inc(&inode->i_writecount);
337 spin_unlock(&inode->i_lock);
338
339 return 0;
340}
341
342int deny_write_access(struct file * file)
343{
0f7fc9e4 344 struct inode *inode = file->f_path.dentry->d_inode;
1da177e4
LT		 345
346 spin_lock(&inode->i_lock);
347 if (atomic_read(&inode->i_writecount) > 0) {
348 spin_unlock(&inode->i_lock);
349 return -ETXTBSY;
350 }
351 atomic_dec(&inode->i_writecount);
352 spin_unlock(&inode->i_lock);
353
354 return 0;
355}
356
5dd784d0
JB		 357/**
358 * path_get - get a reference to a path
359 * @path: path to get the reference to
360 *
361 * Given a path increment the reference count to the dentry and the vfsmount.
362 */
363void path_get(struct path *path)
364{
365 mntget(path->mnt);
366 dget(path->dentry);
367}
368EXPORT_SYMBOL(path_get);
369
1d957f9b
JB		 370/**
371 * path_put - put a reference to a path
372 * @path: path to put the reference to
373 *
374 * Given a path decrement the reference count to the dentry and the vfsmount.
375 */
376void path_put(struct path *path)
1da177e4 377{
1d957f9b
JB		 378 dput(path->dentry);
379 mntput(path->mnt);
1da177e4 380}
1d957f9b 381EXPORT_SYMBOL(path_put);
1da177e4 382
31e6b01f
NP		 383/**
384 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
385 * @nd: nameidata pathwalk data to drop
39191628 386 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 387 *
388 * Path walking has 2 modes, rcu-walk and ref-walk (see
389 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
390 * to drop out of rcu-walk mode and take normal reference counts on dentries
391 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
392 * refcounts at the last known good point before rcu-walk got stuck, so
393 * ref-walk may continue from there. If this is not successful (eg. a seqcount
394 * has changed), then failure is returned and path walk restarts from the
395 * beginning in ref-walk mode.
396 *
397 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
398 * ref-walk. Must be called from rcu-walk context.
399 */
400static int nameidata_drop_rcu(struct nameidata *nd)
401{
402 struct fs_struct *fs = current->fs;
403 struct dentry *dentry = nd->path.dentry;
404
405 BUG_ON(!(nd->flags & LOOKUP_RCU));
406 if (nd->root.mnt) {
407 spin_lock(&fs->lock);
408 if (nd->root.mnt != fs->root.mnt ||
409 nd->root.dentry != fs->root.dentry)
410 goto err_root;
411 }
412 spin_lock(&dentry->d_lock);
413 if (!__d_rcu_to_refcount(dentry, nd->seq))
414 goto err;
415 BUG_ON(nd->inode != dentry->d_inode);
416 spin_unlock(&dentry->d_lock);
417 if (nd->root.mnt) {
418 path_get(&nd->root);
419 spin_unlock(&fs->lock);
420 }
421 mntget(nd->path.mnt);
422
423 rcu_read_unlock();
424 br_read_unlock(vfsmount_lock);
425 nd->flags &= ~LOOKUP_RCU;
426 return 0;
427err:
428 spin_unlock(&dentry->d_lock);
429err_root:
430 if (nd->root.mnt)
431 spin_unlock(&fs->lock);
432 return -ECHILD;
433}
434
435/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
436static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
437{
438 if (nd->flags & LOOKUP_RCU)
439 return nameidata_drop_rcu(nd);
440 return 0;
441}
442
443/**
444 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
445 * @nd: nameidata pathwalk data to drop
446 * @dentry: dentry to drop
39191628 447 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 448 *
449 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
450 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
451 * @nd. Must be called from rcu-walk context.
452 */
453static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry)
454{
455 struct fs_struct *fs = current->fs;
456 struct dentry *parent = nd->path.dentry;
457
458 BUG_ON(!(nd->flags & LOOKUP_RCU));
459 if (nd->root.mnt) {
460 spin_lock(&fs->lock);
461 if (nd->root.mnt != fs->root.mnt ||
462 nd->root.dentry != fs->root.dentry)
463 goto err_root;
464 }
465 spin_lock(&parent->d_lock);
466 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
467 if (!__d_rcu_to_refcount(dentry, nd->seq))
468 goto err;
469 /*
470 * If the sequence check on the child dentry passed, then the child has
471 * not been removed from its parent. This means the parent dentry must
472 * be valid and able to take a reference at this point.
473 */
474 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
475 BUG_ON(!parent->d_count);
476 parent->d_count++;
477 spin_unlock(&dentry->d_lock);
478 spin_unlock(&parent->d_lock);
479 if (nd->root.mnt) {
480 path_get(&nd->root);
481 spin_unlock(&fs->lock);
482 }
483 mntget(nd->path.mnt);
484
485 rcu_read_unlock();
486 br_read_unlock(vfsmount_lock);
487 nd->flags &= ~LOOKUP_RCU;
488 return 0;
489err:
490 spin_unlock(&dentry->d_lock);
491 spin_unlock(&parent->d_lock);
492err_root:
493 if (nd->root.mnt)
494 spin_unlock(&fs->lock);
495 return -ECHILD;
496}
497
498/* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
499static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry)
500{
a7472bab
AV		 501 if (nd->flags & LOOKUP_RCU) {
502 if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) {
503 nd->flags &= ~LOOKUP_RCU;
504 nd->root.mnt = NULL;
505 rcu_read_unlock();
506 br_read_unlock(vfsmount_lock);
507 return -ECHILD;
508 }
509 }
31e6b01f
NP		 510 return 0;
511}
512
513/**
514 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
515 * @nd: nameidata pathwalk data to drop
39191628 516 * Returns: 0 on success, -ECHILD on failure
31e6b01f
NP		 517 *
518 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
519 * nd->path should be the final element of the lookup, so nd->root is discarded.
520 * Must be called from rcu-walk context.
521 */
522static int nameidata_drop_rcu_last(struct nameidata *nd)
523{
524 struct dentry *dentry = nd->path.dentry;
525
526 BUG_ON(!(nd->flags & LOOKUP_RCU));
527 nd->flags &= ~LOOKUP_RCU;
528 nd->root.mnt = NULL;
529 spin_lock(&dentry->d_lock);
530 if (!__d_rcu_to_refcount(dentry, nd->seq))
531 goto err_unlock;
532 BUG_ON(nd->inode != dentry->d_inode);
533 spin_unlock(&dentry->d_lock);
534
535 mntget(nd->path.mnt);
536
537 rcu_read_unlock();
538 br_read_unlock(vfsmount_lock);
539
540 return 0;
541
542err_unlock:
543 spin_unlock(&dentry->d_lock);
544 rcu_read_unlock();
545 br_read_unlock(vfsmount_lock);
546 return -ECHILD;
547}
548
834f2a4a
TM		 549/**
550 * release_open_intent - free up open intent resources
551 * @nd: pointer to nameidata
552 */
553void release_open_intent(struct nameidata *nd)
554{
2dab5974
LT		 555 struct file *file = nd->intent.open.file;
556
557 if (file && !IS_ERR(file)) {
558 if (file->f_path.dentry == NULL)
559 put_filp(file);
560 else
561 fput(file);
562 }
834f2a4a
TM		 563}
564
f60aef7e 565static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 566{
f60aef7e 567 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 568}
569
f5e1c1c1 570static struct dentry *
bcdc5e01
IK		 571do_revalidate(struct dentry *dentry, struct nameidata *nd)
572{
f5e1c1c1 573 int status = d_revalidate(dentry, nd);
bcdc5e01
IK		 574 if (unlikely(status <= 0)) {
575 /*
576 * The dentry failed validation.
577 * If d_revalidate returned 0 attempt to invalidate
578 * the dentry otherwise d_revalidate is asking us
579 * to return a fail status.
580 */
34286d66 581 if (status < 0) {
f5e1c1c1 582 dput(dentry);
34286d66 583 dentry = ERR_PTR(status);
f5e1c1c1
AV		 584 } else if (!d_invalidate(dentry)) {
585 dput(dentry);
586 dentry = NULL;
bcdc5e01
IK		 587 }
588 }
589 return dentry;
590}
591
f5e1c1c1
AV		 592static inline struct dentry *
593do_revalidate_rcu(struct dentry *dentry, struct nameidata *nd)
594{
f60aef7e 595 int status = d_revalidate(dentry, nd);
f5e1c1c1
AV		 596 if (likely(status > 0))
597 return dentry;
598 if (status == -ECHILD) {
599 if (nameidata_dentry_drop_rcu(nd, dentry))
600 return ERR_PTR(-ECHILD);
601 return do_revalidate(dentry, nd);
602 }
603 if (status < 0)
604 return ERR_PTR(status);
605 /* Don't d_invalidate in rcu-walk mode */
606 if (nameidata_dentry_drop_rcu(nd, dentry))
607 return ERR_PTR(-ECHILD);
608 if (!d_invalidate(dentry)) {
609 dput(dentry);
610 dentry = NULL;
611 }
612 return dentry;
613}
614
39159de2 615/*
16c2cd71 616 * handle_reval_path - force revalidation of a dentry
39159de2
JL		 617 *
618 * In some situations the path walking code will trust dentries without
619 * revalidating them. This causes problems for filesystems that depend on
620 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
621 * (which indicates that it's possible for the dentry to go stale), force
622 * a d_revalidate call before proceeding.
623 *
624 * Returns 0 if the revalidation was successful. If the revalidation fails,
625 * either return the error returned by d_revalidate or -ESTALE if the
626 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
627 * invalidate the dentry. It's up to the caller to handle putting references
628 * to the path if necessary.
629 */
16c2cd71 630static inline int handle_reval_path(struct nameidata *nd)
39159de2 631{
16c2cd71 632 struct dentry *dentry = nd->path.dentry;
39159de2 633 int status;
39159de2 634
16c2cd71
AV		 635 if (likely(!(nd->flags & LOOKUP_JUMPED)))
636 return 0;
637
638 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 639 return 0;
640
16c2cd71
AV		 641 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
642 return 0;
643
644 /* Note: we do not d_invalidate() */
34286d66 645 status = d_revalidate(dentry, nd);
39159de2
JL		 646 if (status > 0)
647 return 0;
648
16c2cd71 649 if (!status)
39159de2 650 status = -ESTALE;
16c2cd71 651
39159de2
JL		 652 return status;
653}
654
1da177e4 655/*
b75b5086
AV		 656 * Short-cut version of permission(), for calling on directories
657 * during pathname resolution. Combines parts of permission()
658 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
1da177e4
LT		 659 *
660 * If appropriate, check DAC only. If not appropriate, or
b75b5086 661 * short-cut DAC fails, then call ->permission() to do more
1da177e4
LT		 662 * complete permission check.
663 */
b74c79e9 664static inline int exec_permission(struct inode *inode, unsigned int flags)
1da177e4 665{
5909ccaa 666 int ret;
1da177e4 667
cb9179ea 668 if (inode->i_op->permission) {
b74c79e9
NP		 669 ret = inode->i_op->permission(inode, MAY_EXEC, flags);
670 } else {
671 ret = acl_permission_check(inode, MAY_EXEC, flags,
672 inode->i_op->check_acl);
cb9179ea 673 }
b74c79e9 674 if (likely(!ret))
1da177e4 675 goto ok;
b74c79e9 676 if (ret == -ECHILD)
31e6b01f 677 return ret;
1da177e4 678
f1ac9f6b 679 if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
1da177e4
LT		 680 goto ok;
681
5909ccaa 682 return ret;
1da177e4 683ok:
b74c79e9 684 return security_inode_exec_permission(inode, flags);
1da177e4
LT		 685}
686
2a737871
AV		 687static __always_inline void set_root(struct nameidata *nd)
688{
f7ad3c6b
MS		 689 if (!nd->root.mnt)
690 get_fs_root(current->fs, &nd->root);
2a737871
AV		 691}
692
6de88d72
AV		 693static int link_path_walk(const char *, struct nameidata *);
694
31e6b01f
NP		 695static __always_inline void set_root_rcu(struct nameidata *nd)
696{
697 if (!nd->root.mnt) {
698 struct fs_struct *fs = current->fs;
c28cc364
NP		 699 unsigned seq;
700
701 do {
702 seq = read_seqcount_begin(&fs->seq);
703 nd->root = fs->root;
704 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 705 }
706}
707
f1662356 708static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 709{
31e6b01f
NP		 710 int ret;
711
1da177e4
LT		 712 if (IS_ERR(link))
713 goto fail;
714
715 if (*link == '/') {
2a737871 716 set_root(nd);
1d957f9b 717 path_put(&nd->path);
2a737871
AV		 718 nd->path = nd->root;
719 path_get(&nd->root);
16c2cd71 720 nd->flags |= LOOKUP_JUMPED;
1da177e4 721 }
31e6b01f 722 nd->inode = nd->path.dentry->d_inode;
b4091d5f 723
31e6b01f
NP		 724 ret = link_path_walk(link, nd);
725 return ret;
1da177e4 726fail:
1d957f9b 727 path_put(&nd->path);
1da177e4
LT		 728 return PTR_ERR(link);
729}
730
1d957f9b 731static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 732{
733 dput(path->dentry);
4ac91378 734 if (path->mnt != nd->path.mnt)
051d3812
IK		 735 mntput(path->mnt);
736}
737
7b9337aa
NP		 738static inline void path_to_nameidata(const struct path *path,
739 struct nameidata *nd)
051d3812 740{
31e6b01f
NP		 741 if (!(nd->flags & LOOKUP_RCU)) {
742 dput(nd->path.dentry);
743 if (nd->path.mnt != path->mnt)
744 mntput(nd->path.mnt);
9a229683 745 }
31e6b01f 746 nd->path.mnt = path->mnt;
4ac91378 747 nd->path.dentry = path->dentry;
051d3812
IK		 748}
749
def4af30 750static __always_inline int
7b9337aa 751__do_follow_link(const struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 752{
753 int error;
7b9337aa 754 struct dentry *dentry = link->dentry;
1da177e4 755
844a3917
AV		 756 BUG_ON(nd->flags & LOOKUP_RCU);
757
7b9337aa 758 touch_atime(link->mnt, dentry);
1da177e4 759 nd_set_link(nd, NULL);
cd4e91d3 760
87556ef1
DH		 761 if (link->mnt == nd->path.mnt)
762 mntget(link->mnt);
31e6b01f 763
36f3b4f6
AV		 764 error = security_inode_follow_link(link->dentry, nd);
765 if (error) {
766 *p = ERR_PTR(error); /* no ->put_link(), please */
767 path_put(&nd->path);
768 return error;
769 }
770
86acdca1 771 nd->last_type = LAST_BIND;
def4af30
AV		 772 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
773 error = PTR_ERR(*p);
774 if (!IS_ERR(*p)) {
1da177e4 775 char *s = nd_get_link(nd);
cc314eef 776 error = 0;
1da177e4
LT		 777 if (s)
778 error = __vfs_follow_link(nd, s);
16c2cd71
AV		 779 else if (nd->last_type == LAST_BIND)
780 nd->flags |= LOOKUP_JUMPED;
1da177e4 781 }
1da177e4
LT		 782 return error;
783}
784
785/*
786 * This limits recursive symlink follows to 8, while
787 * limiting consecutive symlinks to 40.
788 *
789 * Without that kind of total limit, nasty chains of consecutive
790 * symlinks can cause almost arbitrarily long lookups.
791 */
3abb17e8 792static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd)
1da177e4 793{
def4af30 794 void *cookie;
1da177e4 795 int err = -ELOOP;
844a3917
AV		 796
797 /* We drop rcu-walk here */
798 if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
799 return -ECHILD;
3abb17e8 800 BUG_ON(inode != path->dentry->d_inode);
844a3917 801
1da177e4
LT		 802 if (current->link_count >= MAX_NESTED_LINKS)
803 goto loop;
804 if (current->total_link_count >= 40)
805 goto loop;
806 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
807 cond_resched();
1da177e4
LT		 808 current->link_count++;
809 current->total_link_count++;
810 nd->depth++;
def4af30
AV		 811 err = __do_follow_link(path, nd, &cookie);
812 if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link)
813 path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie);
258fa999 814 path_put(path);
839d9f93
AV		 815 current->link_count--;
816 nd->depth--;
1da177e4
LT		 817 return err;
818loop:
1d957f9b
JB		 819 path_put_conditional(path, nd);
820 path_put(&nd->path);
1da177e4
LT		 821 return err;
822}
823
31e6b01f
NP		 824static int follow_up_rcu(struct path *path)
825{
826 struct vfsmount *parent;
827 struct dentry *mountpoint;
828
829 parent = path->mnt->mnt_parent;
830 if (parent == path->mnt)
831 return 0;
832 mountpoint = path->mnt->mnt_mountpoint;
833 path->dentry = mountpoint;
834 path->mnt = parent;
835 return 1;
836}
837
bab77ebf 838int follow_up(struct path *path)
1da177e4
LT		 839{
840 struct vfsmount *parent;
841 struct dentry *mountpoint;
99b7db7b
NP		 842
843 br_read_lock(vfsmount_lock);
bab77ebf
AV		 844 parent = path->mnt->mnt_parent;
845 if (parent == path->mnt) {
99b7db7b 846 br_read_unlock(vfsmount_lock);
1da177e4
LT		 847 return 0;
848 }
849 mntget(parent);
bab77ebf 850 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 851 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 852 dput(path->dentry);
853 path->dentry = mountpoint;
854 mntput(path->mnt);
855 path->mnt = parent;
1da177e4
LT		 856 return 1;
857}
858
b5c84bf6 859/*
9875cf80
DH		 860 * Perform an automount
861 * - return -EISDIR to tell follow_managed() to stop and return the path we
862 * were called with.
1da177e4 863 */
9875cf80
DH		 864static int follow_automount(struct path *path, unsigned flags,
865 bool *need_mntput)
31e6b01f 866{
9875cf80 867 struct vfsmount *mnt;
ea5b778a 868 int err;
9875cf80
DH		 869
870 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
871 return -EREMOTE;
872
6f45b656
DH		 873 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
874 * and this is the terminal part of the path.
875 */
876 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
877 return -EISDIR; /* we actually want to stop here */
878
9875cf80
DH		 879 /* We want to mount if someone is trying to open/create a file of any
880 * type under the mountpoint, wants to traverse through the mountpoint
881 * or wants to open the mounted directory.
882 *
883 * We don't want to mount if someone's just doing a stat and they've
884 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
885 * appended a '/' to the name.
886 */
887 if (!(flags & LOOKUP_FOLLOW) &&
888 !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
889 LOOKUP_OPEN | LOOKUP_CREATE)))
890 return -EISDIR;
891
892 current->total_link_count++;
893 if (current->total_link_count >= 40)
894 return -ELOOP;
895
896 mnt = path->dentry->d_op->d_automount(path);
897 if (IS_ERR(mnt)) {
898 /*
899 * The filesystem is allowed to return -EISDIR here to indicate
900 * it doesn't want to automount. For instance, autofs would do
901 * this so that its userspace daemon can mount on this dentry.
902 *
903 * However, we can only permit this if it's a terminal point in
904 * the path being looked up; if it wasn't then the remainder of
905 * the path is inaccessible and we should say so.
906 */
907 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
908 return -EREMOTE;
909 return PTR_ERR(mnt);
31e6b01f 910 }
ea5b778a 911
9875cf80
DH		 912 if (!mnt) /* mount collision */
913 return 0;
31e6b01f 914
19a167af 915 err = finish_automount(mnt, path);
9875cf80 916
ea5b778a
DH		 917 switch (err) {
918 case -EBUSY:
919 /* Someone else made a mount here whilst we were busy */
19a167af 920 return 0;
ea5b778a 921 case 0:
ea5b778a
DH		 922 dput(path->dentry);
923 if (*need_mntput)
924 mntput(path->mnt);
925 path->mnt = mnt;
926 path->dentry = dget(mnt->mnt_root);
927 *need_mntput = true;
928 return 0;
19a167af
AV		 929 default:
930 return err;
ea5b778a 931 }
19a167af 932
463ffb2e
AV		 933}
934
9875cf80
DH		 935/*
936 * Handle a dentry that is managed in some way.
cc53ce53 937 * - Flagged for transit management (autofs)
9875cf80
DH		 938 * - Flagged as mountpoint
939 * - Flagged as automount point
940 *
941 * This may only be called in refwalk mode.
942 *
943 * Serialization is taken care of in namespace.c
944 */
945static int follow_managed(struct path *path, unsigned flags)
1da177e4 946{
9875cf80
DH		 947 unsigned managed;
948 bool need_mntput = false;
949 int ret;
950
951 /* Given that we're not holding a lock here, we retain the value in a
952 * local variable for each dentry as we look at it so that we don't see
953 * the components of that value change under us */
954 while (managed = ACCESS_ONCE(path->dentry->d_flags),
955 managed &= DCACHE_MANAGED_DENTRY,
956 unlikely(managed != 0)) {
cc53ce53
DH		 957 /* Allow the filesystem to manage the transit without i_mutex
958 * being held. */
959 if (managed & DCACHE_MANAGE_TRANSIT) {
960 BUG_ON(!path->dentry->d_op);
961 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 962 ret = path->dentry->d_op->d_manage(path->dentry,
963 false, false);
cc53ce53
DH		 964 if (ret < 0)
965 return ret == -EISDIR ? 0 : ret;
966 }
967
9875cf80
DH		 968 /* Transit to a mounted filesystem. */
969 if (managed & DCACHE_MOUNTED) {
970 struct vfsmount *mounted = lookup_mnt(path);
971 if (mounted) {
972 dput(path->dentry);
973 if (need_mntput)
974 mntput(path->mnt);
975 path->mnt = mounted;
976 path->dentry = dget(mounted->mnt_root);
977 need_mntput = true;
978 continue;
979 }
980
981 /* Something is mounted on this dentry in another
982 * namespace and/or whatever was mounted there in this
983 * namespace got unmounted before we managed to get the
984 * vfsmount_lock */
985 }
986
987 /* Handle an automount point */
988 if (managed & DCACHE_NEED_AUTOMOUNT) {
989 ret = follow_automount(path, flags, &need_mntput);
990 if (ret < 0)
991 return ret == -EISDIR ? 0 : ret;
992 continue;
993 }
994
995 /* We didn't change the current path point */
996 break;
1da177e4 997 }
9875cf80 998 return 0;
1da177e4
LT		 999}
1000
cc53ce53 1001int follow_down_one(struct path *path)
1da177e4
LT		 1002{
1003 struct vfsmount *mounted;
1004
1c755af4 1005 mounted = lookup_mnt(path);
1da177e4 1006 if (mounted) {
9393bd07
AV		 1007 dput(path->dentry);
1008 mntput(path->mnt);
1009 path->mnt = mounted;
1010 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 1011 return 1;
1012 }
1013 return 0;
1014}
1015
9875cf80
DH		 1016/*
1017 * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we
cc53ce53 1018 * meet a managed dentry and we're not walking to "..". True is returned to
9875cf80
DH		 1019 * continue, false to abort.
1020 */
1021static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1022 struct inode **inode, bool reverse_transit)
1023{
1024 while (d_mountpoint(path->dentry)) {
1025 struct vfsmount *mounted;
ab90911f
DH		 1026 if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
1027 !reverse_transit &&
1028 path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
1029 return false;
9875cf80
DH		 1030 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1031 if (!mounted)
1032 break;
1033 path->mnt = mounted;
1034 path->dentry = mounted->mnt_root;
1035 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1036 *inode = path->dentry->d_inode;
1037 }
1038
1039 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1040 return reverse_transit;
1041 return true;
1042}
1043
31e6b01f
NP		 1044static int follow_dotdot_rcu(struct nameidata *nd)
1045{
1046 struct inode *inode = nd->inode;
1047
1048 set_root_rcu(nd);
1049
9875cf80 1050 while (1) {
31e6b01f
NP		 1051 if (nd->path.dentry == nd->root.dentry &&
1052 nd->path.mnt == nd->root.mnt) {
1053 break;
1054 }
1055 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1056 struct dentry *old = nd->path.dentry;
1057 struct dentry *parent = old->d_parent;
1058 unsigned seq;
1059
1060 seq = read_seqcount_begin(&parent->d_seq);
1061 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 1062 goto failed;
31e6b01f
NP		 1063 inode = parent->d_inode;
1064 nd->path.dentry = parent;
1065 nd->seq = seq;
1066 break;
1067 }
1068 if (!follow_up_rcu(&nd->path))
1069 break;
1070 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1071 inode = nd->path.dentry->d_inode;
1072 }
9875cf80 1073 __follow_mount_rcu(nd, &nd->path, &inode, true);
31e6b01f 1074 nd->inode = inode;
31e6b01f 1075 return 0;
ef7562d5
AV		 1076
1077failed:
1078 nd->flags &= ~LOOKUP_RCU;
1079 nd->root.mnt = NULL;
1080 rcu_read_unlock();
1081 br_read_unlock(vfsmount_lock);
1082 return -ECHILD;
31e6b01f
NP		 1083}
1084
cc53ce53
DH		 1085/*
1086 * Follow down to the covering mount currently visible to userspace. At each
1087 * point, the filesystem owning that dentry may be queried as to whether the
1088 * caller is permitted to proceed or not.
1089 *
1090 * Care must be taken as namespace_sem may be held (indicated by mounting_here
1091 * being true).
1092 */
1093int follow_down(struct path *path, bool mounting_here)
1094{
1095 unsigned managed;
1096 int ret;
1097
1098 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1099 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1100 /* Allow the filesystem to manage the transit without i_mutex
1101 * being held.
1102 *
1103 * We indicate to the filesystem if someone is trying to mount
1104 * something here. This gives autofs the chance to deny anyone
1105 * other than its daemon the right to mount on its
1106 * superstructure.
1107 *
1108 * The filesystem may sleep at this point.
1109 */
1110 if (managed & DCACHE_MANAGE_TRANSIT) {
1111 BUG_ON(!path->dentry->d_op);
1112 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f
DH		 1113 ret = path->dentry->d_op->d_manage(
1114 path->dentry, mounting_here, false);
cc53ce53
DH		 1115 if (ret < 0)
1116 return ret == -EISDIR ? 0 : ret;
1117 }
1118
1119 /* Transit to a mounted filesystem. */
1120 if (managed & DCACHE_MOUNTED) {
1121 struct vfsmount *mounted = lookup_mnt(path);
1122 if (!mounted)
1123 break;
1124 dput(path->dentry);
1125 mntput(path->mnt);
1126 path->mnt = mounted;
1127 path->dentry = dget(mounted->mnt_root);
1128 continue;
1129 }
1130
1131 /* Don't handle automount points here */
1132 break;
1133 }
1134 return 0;
1135}
1136
9875cf80
DH		 1137/*
1138 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1139 */
1140static void follow_mount(struct path *path)
1141{
1142 while (d_mountpoint(path->dentry)) {
1143 struct vfsmount *mounted = lookup_mnt(path);
1144 if (!mounted)
1145 break;
1146 dput(path->dentry);
1147 mntput(path->mnt);
1148 path->mnt = mounted;
1149 path->dentry = dget(mounted->mnt_root);
1150 }
1151}
1152
31e6b01f 1153static void follow_dotdot(struct nameidata *nd)
1da177e4 1154{
2a737871 1155 set_root(nd);
e518ddb7 1156
1da177e4 1157 while(1) {
4ac91378 1158 struct dentry *old = nd->path.dentry;
1da177e4 1159
2a737871
AV		 1160 if (nd->path.dentry == nd->root.dentry &&
1161 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1162 break;
1163 }
4ac91378 1164 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1165 /* rare case of legitimate dget_parent()... */
1166 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1167 dput(old);
1168 break;
1169 }
3088dd70 1170 if (!follow_up(&nd->path))
1da177e4 1171 break;
1da177e4 1172 }
79ed0226 1173 follow_mount(&nd->path);
31e6b01f 1174 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1175}
1176
baa03890
NP		 1177/*
1178 * Allocate a dentry with name and parent, and perform a parent
1179 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1180 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1181 * have verified that no child exists while under i_mutex.
1182 */
1183static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1184 struct qstr *name, struct nameidata *nd)
1185{
1186 struct inode *inode = parent->d_inode;
1187 struct dentry *dentry;
1188 struct dentry *old;
1189
1190 /* Don't create child dentry for a dead directory. */
1191 if (unlikely(IS_DEADDIR(inode)))
1192 return ERR_PTR(-ENOENT);
1193
1194 dentry = d_alloc(parent, name);
1195 if (unlikely(!dentry))
1196 return ERR_PTR(-ENOMEM);
1197
1198 old = inode->i_op->lookup(inode, dentry, nd);
1199 if (unlikely(old)) {
1200 dput(dentry);
1201 dentry = old;
1202 }
1203 return dentry;
1204}
1205
1da177e4
LT		 1206/*
1207 * It's more convoluted than I'd like it to be, but... it's still fairly
1208 * small and for now I'd prefer to have fast path as straight as possible.
1209 * It _is_ time-critical.
1210 */
1211static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1212 struct path *path, struct inode **inode)
1da177e4 1213{
4ac91378 1214 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1215 struct dentry *dentry, *parent = nd->path.dentry;
6e6b1bd1 1216 struct inode *dir;
9875cf80
DH		 1217 int err;
1218
3cac260a
AV		 1219 /*
1220 * See if the low-level filesystem might want
1221 * to use its own hash..
1222 */
fb045adb 1223 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
9875cf80 1224 err = parent->d_op->d_hash(parent, nd->inode, name);
3cac260a
AV		 1225 if (err < 0)
1226 return err;
1227 }
1da177e4 1228
b04f784e
NP		 1229 /*
1230 * Rename seqlock is not required here because in the off chance
1231 * of a false negative due to a concurrent rename, we're going to
1232 * do the non-racy lookup, below.
1233 */
31e6b01f
NP		 1234 if (nd->flags & LOOKUP_RCU) {
1235 unsigned seq;
1236
1237 *inode = nd->inode;
1238 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1239 if (!dentry) {
1240 if (nameidata_drop_rcu(nd))
1241 return -ECHILD;
1242 goto need_lookup;
1243 }
1244 /* Memory barrier in read_seqcount_begin of child is enough */
1245 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1246 return -ECHILD;
1247
1248 nd->seq = seq;
24643087 1249 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
f5e1c1c1 1250 dentry = do_revalidate_rcu(dentry, nd);
24643087
AV		 1251 if (!dentry)
1252 goto need_lookup;
1253 if (IS_ERR(dentry))
1254 goto fail;
1255 if (!(nd->flags & LOOKUP_RCU))
1256 goto done;
1257 }
31e6b01f
NP		 1258 path->mnt = mnt;
1259 path->dentry = dentry;
9875cf80
DH		 1260 if (likely(__follow_mount_rcu(nd, path, inode, false)))
1261 return 0;
1262 if (nameidata_drop_rcu(nd))
1263 return -ECHILD;
1264 /* fallthru */
1265 }
1266 dentry = __d_lookup(parent, name);
1267 if (!dentry)
1268 goto need_lookup;
2e2e88ea 1269found:
24643087
AV		 1270 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1271 dentry = do_revalidate(dentry, nd);
1272 if (!dentry)
1273 goto need_lookup;
1274 if (IS_ERR(dentry))
1275 goto fail;
1276 }
1da177e4 1277done:
9875cf80
DH		 1278 path->mnt = mnt;
1279 path->dentry = dentry;
1280 err = follow_managed(path, nd->flags);
89312214
IK		 1281 if (unlikely(err < 0)) {
1282 path_put_conditional(path, nd);
9875cf80 1283 return err;
89312214 1284 }
9875cf80 1285 *inode = path->dentry->d_inode;
1da177e4
LT		 1286 return 0;
1287
1288need_lookup:
6e6b1bd1 1289 dir = parent->d_inode;
31e6b01f 1290 BUG_ON(nd->inode != dir);
6e6b1bd1
AV		 1291
1292 mutex_lock(&dir->i_mutex);
1293 /*
1294 * First re-do the cached lookup just in case it was created
b04f784e
NP		 1295 * while we waited for the directory semaphore, or the first
1296 * lookup failed due to an unrelated rename.
6e6b1bd1 1297 *
b04f784e
NP		 1298 * This could use version numbering or similar to avoid unnecessary
1299 * cache lookups, but then we'd have to do the first lookup in the
1300 * non-racy way. However in the common case here, everything should
1301 * be hot in cache, so would it be a big win?
6e6b1bd1
AV		 1302 */
1303 dentry = d_lookup(parent, name);
baa03890
NP		 1304 if (likely(!dentry)) {
1305 dentry = d_alloc_and_lookup(parent, name, nd);
6e6b1bd1
AV		 1306 mutex_unlock(&dir->i_mutex);
1307 if (IS_ERR(dentry))
1308 goto fail;
1309 goto done;
1310 }
6e6b1bd1
AV		 1311 /*
1312 * Uhhuh! Nasty case: the cache was re-populated while
1313 * we waited on the semaphore. Need to revalidate.
1314 */
1315 mutex_unlock(&dir->i_mutex);
2e2e88ea 1316 goto found;
1da177e4 1317
1da177e4
LT		 1318fail:
1319 return PTR_ERR(dentry);
1320}
1321
52094c8a
AV		 1322static inline int may_lookup(struct nameidata *nd)
1323{
1324 if (nd->flags & LOOKUP_RCU) {
1325 int err = exec_permission(nd->inode, IPERM_FLAG_RCU);
1326 if (err != -ECHILD)
1327 return err;
1328 if (nameidata_drop_rcu(nd))
1329 return -ECHILD;
1330 }
1331 return exec_permission(nd->inode, 0);
1332}
1333
9856fa1b
AV		 1334static inline int handle_dots(struct nameidata *nd, int type)
1335{
1336 if (type == LAST_DOTDOT) {
1337 if (nd->flags & LOOKUP_RCU) {
1338 if (follow_dotdot_rcu(nd))
1339 return -ECHILD;
1340 } else
1341 follow_dotdot(nd);
1342 }
1343 return 0;
1344}
1345
951361f9
AV		 1346static void terminate_walk(struct nameidata *nd)
1347{
1348 if (!(nd->flags & LOOKUP_RCU)) {
1349 path_put(&nd->path);
1350 } else {
1351 nd->flags &= ~LOOKUP_RCU;
1352 nd->root.mnt = NULL;
1353 rcu_read_unlock();
1354 br_read_unlock(vfsmount_lock);
1355 }
1356}
1357
1da177e4
LT		 1358/*
1359 * Name resolution.
ea3834d9
PM		 1360 * This is the basic name resolution function, turning a pathname into
1361 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1362 *
ea3834d9
PM		 1363 * Returns 0 and nd will have valid dentry and mnt on success.
1364 * Returns error and drops reference to input namei data on failure.
1da177e4 1365 */
6de88d72 1366static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1367{
1368 struct path next;
1da177e4
LT		 1369 int err;
1370 unsigned int lookup_flags = nd->flags;
1371
1372 while (*name=='/')
1373 name++;
1374 if (!*name)
086e183a 1375 return 0;
1da177e4 1376
1da177e4 1377 if (nd->depth)
f55eab82 1378 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
1da177e4
LT		 1379
1380 /* At this point we know we have a real path component. */
1381 for(;;) {
31e6b01f 1382 struct inode *inode;
1da177e4
LT		 1383 unsigned long hash;
1384 struct qstr this;
1385 unsigned int c;
fe479a58 1386 int type;
1da177e4 1387
cdce5d6b 1388 nd->flags |= LOOKUP_CONTINUE;
52094c8a
AV		 1389
1390 err = may_lookup(nd);
1da177e4
LT		 1391 if (err)
1392 break;
1393
1394 this.name = name;
1395 c = *(const unsigned char *)name;
1396
1397 hash = init_name_hash();
1398 do {
1399 name++;
1400 hash = partial_name_hash(c, hash);
1401 c = *(const unsigned char *)name;
1402 } while (c && (c != '/'));
1403 this.len = name - (const char *) this.name;
1404 this.hash = end_name_hash(hash);
1405
fe479a58
AV		 1406 type = LAST_NORM;
1407 if (this.name[0] == '.') switch (this.len) {
1408 case 2:
16c2cd71 1409 if (this.name[1] == '.') {
fe479a58 1410 type = LAST_DOTDOT;
16c2cd71
AV		 1411 nd->flags |= LOOKUP_JUMPED;
1412 }
fe479a58
AV		 1413 break;
1414 case 1:
1415 type = LAST_DOT;
1416 }
16c2cd71
AV		 1417 if (likely(type == LAST_NORM))
1418 nd->flags &= ~LOOKUP_JUMPED;
fe479a58 1419
1da177e4
LT		 1420 /* remove trailing slashes? */
1421 if (!c)
1422 goto last_component;
1423 while (*++name == '/');
1424 if (!*name)
1425 goto last_with_slashes;
1426
1427 /*
1428 * "." and ".." are special - ".." especially so because it has
1429 * to be able to know about the current root directory and
1430 * parent relationships.
1431 */
fe479a58 1432 if (unlikely(type != LAST_NORM)) {
ef7562d5
AV		 1433 if (handle_dots(nd, type))
1434 return -ECHILD;
fe479a58 1435 continue;
1da177e4 1436 }
fe479a58 1437
1da177e4 1438 /* This does the actual lookups.. */
31e6b01f 1439 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1440 if (err)
1441 break;
1da177e4 1442
7bc055d1 1443 if (inode && inode->i_op->follow_link) {
3abb17e8 1444 err = do_follow_link(inode, &next, nd);
1da177e4 1445 if (err)
a7472bab 1446 return err;
31e6b01f 1447 nd->inode = nd->path.dentry->d_inode;
31e6b01f 1448 } else {
09dd17d3 1449 path_to_nameidata(&next, nd);
31e6b01f
NP		 1450 nd->inode = inode;
1451 }
7bc055d1
AV		 1452 err = -ENOENT;
1453 if (!nd->inode)
1454 break;
1da177e4 1455 err = -ENOTDIR;
31e6b01f 1456 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1457 break;
1458 continue;
1459 /* here ends the main loop */
1460
1461last_with_slashes:
1462 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1463last_component:
f55eab82
TM		 1464 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1465 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
1da177e4
LT		 1466 if (lookup_flags & LOOKUP_PARENT)
1467 goto lookup_parent;
ef7562d5
AV		 1468 if (unlikely(type != LAST_NORM))
1469 return handle_dots(nd, type);
31e6b01f 1470 err = do_lookup(nd, &this, &next, &inode);
1da177e4
LT		 1471 if (err)
1472 break;
db372915
DH		 1473 if (inode && unlikely(inode->i_op->follow_link) &&
1474 (lookup_flags & LOOKUP_FOLLOW)) {
3abb17e8 1475 err = do_follow_link(inode, &next, nd);
1da177e4 1476 if (err)
a7472bab 1477 return err;
31e6b01f
NP		 1478 nd->inode = nd->path.dentry->d_inode;
1479 } else {
09dd17d3 1480 path_to_nameidata(&next, nd);
31e6b01f
NP		 1481 nd->inode = inode;
1482 }
1da177e4 1483 err = -ENOENT;
31e6b01f 1484 if (!nd->inode)
1da177e4
LT		 1485 break;
1486 if (lookup_flags & LOOKUP_DIRECTORY) {
1487 err = -ENOTDIR;
31e6b01f 1488 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1489 break;
1490 }
086e183a 1491 return 0;
1da177e4
LT		 1492lookup_parent:
1493 nd->last = this;
fe479a58 1494 nd->last_type = type;
1da177e4 1495 return 0;
1da177e4 1496 }
951361f9 1497 terminate_walk(nd);
1da177e4
LT		 1498 return err;
1499}
1500
e41f7d4e 1501static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd)
31e6b01f
NP		 1502{
1503 int retval = 0;
1504 int fput_needed;
1505 struct file *file;
1506
1507 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1508 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f
NP		 1509 nd->depth = 0;
1510 nd->root.mnt = NULL;
1511 nd->file = NULL;
1512
1513 if (*name=='/') {
e41f7d4e
AV		 1514 if (flags & LOOKUP_RCU) {
1515 br_read_lock(vfsmount_lock);
1516 rcu_read_lock();
1517 set_root_rcu(nd);
1518 } else {
1519 set_root(nd);
1520 path_get(&nd->root);
1521 }
1522 nd->path = nd->root;
31e6b01f 1523 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1524 if (flags & LOOKUP_RCU) {
1525 struct fs_struct *fs = current->fs;
1526 unsigned seq;
31e6b01f 1527
e41f7d4e
AV		 1528 br_read_lock(vfsmount_lock);
1529 rcu_read_lock();
c28cc364 1530
e41f7d4e
AV		 1531 do {
1532 seq = read_seqcount_begin(&fs->seq);
1533 nd->path = fs->pwd;
1534 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1535 } while (read_seqcount_retry(&fs->seq, seq));
1536 } else {
1537 get_fs_pwd(current->fs, &nd->path);
1538 }
31e6b01f
NP		 1539 } else {
1540 struct dentry *dentry;
1541
1542 file = fget_light(dfd, &fput_needed);
1543 retval = -EBADF;
1544 if (!file)
1545 goto out_fail;
1546
1547 dentry = file->f_path.dentry;
1548
1549 retval = -ENOTDIR;
1550 if (!S_ISDIR(dentry->d_inode->i_mode))
1551 goto fput_fail;
1552
1553 retval = file_permission(file, MAY_EXEC);
1554 if (retval)
1555 goto fput_fail;
1556
1557 nd->path = file->f_path;
e41f7d4e
AV		 1558 if (flags & LOOKUP_RCU) {
1559 if (fput_needed)
1560 nd->file = file;
1561 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1562 br_read_lock(vfsmount_lock);
1563 rcu_read_lock();
1564 } else {
1565 path_get(&file->f_path);
1566 fput_light(file, fput_needed);
1567 }
31e6b01f 1568 }
31e6b01f 1569
31e6b01f 1570 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1571 return 0;
2dfdd266 1572
9b4a9b14
AV		 1573fput_fail:
1574 fput_light(file, fput_needed);
1575out_fail:
1576 return retval;
1577}
1578
1579/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1580static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1581 unsigned int flags, struct nameidata *nd)
1582{
31e6b01f
NP		 1583 int retval;
1584
1585 /*
1586 * Path walking is largely split up into 2 different synchronisation
1587 * schemes, rcu-walk and ref-walk (explained in
1588 * Documentation/filesystems/path-lookup.txt). These share much of the
1589 * path walk code, but some things particularly setup, cleanup, and
1590 * following mounts are sufficiently divergent that functions are
1591 * duplicated. Typically there is a function foo(), and its RCU
1592 * analogue, foo_rcu().
1593 *
1594 * -ECHILD is the error number of choice (just to avoid clashes) that
1595 * is returned if some aspect of an rcu-walk fails. Such an error must
1596 * be handled by restarting a traditional ref-walk (which will always
1597 * be able to complete).
1598 */
e41f7d4e 1599 retval = path_init(dfd, name, flags, nd);
ee0827cd 1600
31e6b01f
NP		 1601 if (unlikely(retval))
1602 return retval;
ee0827cd
AV		 1603
1604 current->total_link_count = 0;
1605 retval = link_path_walk(name, nd);
1606
1607 if (nd->flags & LOOKUP_RCU) {
4455ca62
AV		 1608 /* went all way through without dropping RCU */
1609 BUG_ON(retval);
1610 if (nameidata_drop_rcu_last(nd))
1611 retval = -ECHILD;
ee0827cd
AV		 1612 }
1613
16c2cd71
AV		 1614 if (!retval)
1615 retval = handle_reval_path(nd);
1616
ee0827cd
AV		 1617 if (nd->file) {
1618 fput(nd->file);
1619 nd->file = NULL;
1620 }
1621
2a737871
AV		 1622 if (nd->root.mnt) {
1623 path_put(&nd->root);
1624 nd->root.mnt = NULL;
1625 }
ee0827cd
AV		 1626 return retval;
1627}
31e6b01f 1628
ee0827cd
AV		 1629static int do_path_lookup(int dfd, const char *name,
1630 unsigned int flags, struct nameidata *nd)
1631{
1632 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1633 if (unlikely(retval == -ECHILD))
1634 retval = path_lookupat(dfd, name, flags, nd);
1635 if (unlikely(retval == -ESTALE))
1636 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1637
1638 if (likely(!retval)) {
1639 if (unlikely(!audit_dummy_context())) {
1640 if (nd->path.dentry && nd->inode)
1641 audit_inode(name, nd->path.dentry);
1642 }
1643 }
170aa3d0 1644 return retval;
1da177e4
LT		 1645}
1646
c9c6cac0 1647int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1648{
c9c6cac0 1649 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1650}
1651
d1811465
AV		 1652int kern_path(const char *name, unsigned int flags, struct path *path)
1653{
1654 struct nameidata nd;
1655 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1656 if (!res)
1657 *path = nd.path;
1658 return res;
1659}
1660
16f18200
JJS		 1661/**
1662 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1663 * @dentry: pointer to dentry of the base directory
1664 * @mnt: pointer to vfs mount of the base directory
1665 * @name: pointer to file name
1666 * @flags: lookup flags
1667 * @nd: pointer to nameidata
1668 */
1669int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1670 const char *name, unsigned int flags,
1671 struct nameidata *nd)
1672{
ee0827cd 1673 int result;
16f18200
JJS		 1674
1675 /* same as do_path_lookup */
1676 nd->last_type = LAST_ROOT;
16c2cd71 1677 nd->flags = flags | LOOKUP_JUMPED;
16f18200
JJS		 1678 nd->depth = 0;
1679
c8e7f449
JB		 1680 nd->path.dentry = dentry;
1681 nd->path.mnt = mnt;
1682 path_get(&nd->path);
5b857119
AV		 1683 nd->root = nd->path;
1684 path_get(&nd->root);
31e6b01f 1685 nd->inode = nd->path.dentry->d_inode;
16f18200 1686
ee0827cd
AV		 1687 current->total_link_count = 0;
1688
1689 result = link_path_walk(name, nd);
16c2cd71
AV		 1690 if (!result)
1691 result = handle_reval_path(nd);
ee0827cd
AV		 1692 if (result == -ESTALE) {
1693 /* nd->path had been dropped */
1694 current->total_link_count = 0;
1695 nd->path.dentry = dentry;
1696 nd->path.mnt = mnt;
1697 nd->inode = dentry->d_inode;
1698 path_get(&nd->path);
16c2cd71
AV		 1699 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_REVAL;
1700
ee0827cd 1701 result = link_path_walk(name, nd);
16c2cd71
AV		 1702 if (!result)
1703 result = handle_reval_path(nd);
ee0827cd
AV		 1704 }
1705 if (unlikely(!result && !audit_dummy_context() && nd->path.dentry &&
31e6b01f 1706 nd->inode))
4ac91378 1707 audit_inode(name, nd->path.dentry);
16f18200 1708
5b857119
AV		 1709 path_put(&nd->root);
1710 nd->root.mnt = NULL;
16f18200 1711
ee0827cd 1712 return result;
16f18200
JJS		 1713}
1714
eead1911
CH		 1715static struct dentry *__lookup_hash(struct qstr *name,
1716 struct dentry *base, struct nameidata *nd)
1da177e4 1717{
81fca444 1718 struct inode *inode = base->d_inode;
057f6c01 1719 struct dentry *dentry;
1da177e4
LT		 1720 int err;
1721
b74c79e9 1722 err = exec_permission(inode, 0);
81fca444
CH		 1723 if (err)
1724 return ERR_PTR(err);
1da177e4
LT		 1725
1726 /*
1727 * See if the low-level filesystem might want
1728 * to use its own hash..
1729 */
fb045adb 1730 if (base->d_flags & DCACHE_OP_HASH) {
b1e6a015 1731 err = base->d_op->d_hash(base, inode, name);
1da177e4
LT		 1732 dentry = ERR_PTR(err);
1733 if (err < 0)
1734 goto out;
1735 }
1736
b04f784e
NP		 1737 /*
1738 * Don't bother with __d_lookup: callers are for creat as
1739 * well as unlink, so a lot of the time it would cost
1740 * a double lookup.
6e6b1bd1 1741 */
b04f784e 1742 dentry = d_lookup(base, name);
6e6b1bd1 1743
fb045adb 1744 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
6e6b1bd1
AV		 1745 dentry = do_revalidate(dentry, nd);
1746
baa03890
NP		 1747 if (!dentry)
1748 dentry = d_alloc_and_lookup(base, name, nd);
1da177e4
LT		 1749out:
1750 return dentry;
1751}
1752
057f6c01
JM		 1753/*
1754 * Restricted form of lookup. Doesn't follow links, single-component only,
1755 * needs parent already locked. Doesn't follow mounts.
1756 * SMP-safe.
1757 */
eead1911 1758static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1759{
4ac91378 1760 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1761}
1762
eead1911
CH		 1763static int __lookup_one_len(const char *name, struct qstr *this,
1764 struct dentry *base, int len)
1da177e4
LT		 1765{
1766 unsigned long hash;
1da177e4
LT		 1767 unsigned int c;
1768
057f6c01
JM		 1769 this->name = name;
1770 this->len = len;
1da177e4 1771 if (!len)
057f6c01 1772 return -EACCES;
1da177e4
LT		 1773
1774 hash = init_name_hash();
1775 while (len--) {
1776 c = *(const unsigned char *)name++;
1777 if (c == '/' || c == '\0')
057f6c01 1778 return -EACCES;
1da177e4
LT		 1779 hash = partial_name_hash(c, hash);
1780 }
057f6c01
JM		 1781 this->hash = end_name_hash(hash);
1782 return 0;
1783}
1da177e4 1784
eead1911 1785/**
a6b91919 1786 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1787 * @name: pathname component to lookup
1788 * @base: base directory to lookup from
1789 * @len: maximum length @len should be interpreted to
1790 *
a6b91919
RD		 1791 * Note that this routine is purely a helper for filesystem usage and should
1792 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1793 * nameidata argument is passed to the filesystem methods and a filesystem
1794 * using this helper needs to be prepared for that.
1795 */
057f6c01
JM		 1796struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1797{
1798 int err;
1799 struct qstr this;
1800
2f9092e1
DW		 1801 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1802
057f6c01 1803 err = __lookup_one_len(name, &this, base, len);
eead1911
CH		 1804 if (err)
1805 return ERR_PTR(err);
1806
49705b77 1807 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1808}
1809
2d8f3038
AV		 1810int user_path_at(int dfd, const char __user *name, unsigned flags,
1811 struct path *path)
1da177e4 1812{
2d8f3038 1813 struct nameidata nd;
1da177e4
LT		 1814 char *tmp = getname(name);
1815 int err = PTR_ERR(tmp);
1da177e4 1816 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1817
1818 BUG_ON(flags & LOOKUP_PARENT);
1819
1820 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1821 putname(tmp);
2d8f3038
AV		 1822 if (!err)
1823 *path = nd.path;
1da177e4
LT		 1824 }
1825 return err;
1826}
1827
2ad94ae6
AV		 1828static int user_path_parent(int dfd, const char __user *path,
1829 struct nameidata *nd, char **name)
1830{
1831 char *s = getname(path);
1832 int error;
1833
1834 if (IS_ERR(s))
1835 return PTR_ERR(s);
1836
1837 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1838 if (error)
1839 putname(s);
1840 else
1841 *name = s;
1842
1843 return error;
1844}
1845
1da177e4
LT		 1846/*
1847 * It's inline, so penalty for filesystems that don't use sticky bit is
1848 * minimal.
1849 */
1850static inline int check_sticky(struct inode *dir, struct inode *inode)
1851{
da9592ed
DH		 1852 uid_t fsuid = current_fsuid();
1853
1da177e4
LT		 1854 if (!(dir->i_mode & S_ISVTX))
1855 return 0;
da9592ed 1856 if (inode->i_uid == fsuid)
1da177e4 1857 return 0;
da9592ed 1858 if (dir->i_uid == fsuid)
1da177e4
LT		 1859 return 0;
1860 return !capable(CAP_FOWNER);
1861}
1862
1863/*
1864 * Check whether we can remove a link victim from directory dir, check
1865 * whether the type of victim is right.
1866 * 1. We can't do it if dir is read-only (done in permission())
1867 * 2. We should have write and exec permissions on dir
1868 * 3. We can't remove anything from append-only dir
1869 * 4. We can't do anything with immutable dir (done in permission())
1870 * 5. If the sticky bit on dir is set we should either
1871 * a. be owner of dir, or
1872 * b. be owner of victim, or
1873 * c. have CAP_FOWNER capability
1874 * 6. If the victim is append-only or immutable we can't do antyhing with
1875 * links pointing to it.
1876 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1877 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1878 * 9. We can't remove a root or mountpoint.
1879 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1880 * nfs_async_unlink().
1881 */
858119e1 1882static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1883{
1884 int error;
1885
1886 if (!victim->d_inode)
1887 return -ENOENT;
1888
1889 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1890 audit_inode_child(victim, dir);
1da177e4 1891
f419a2e3 1892 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1893 if (error)
1894 return error;
1895 if (IS_APPEND(dir))
1896 return -EPERM;
1897 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1898 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1899 return -EPERM;
1900 if (isdir) {
1901 if (!S_ISDIR(victim->d_inode->i_mode))
1902 return -ENOTDIR;
1903 if (IS_ROOT(victim))
1904 return -EBUSY;
1905 } else if (S_ISDIR(victim->d_inode->i_mode))
1906 return -EISDIR;
1907 if (IS_DEADDIR(dir))
1908 return -ENOENT;
1909 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1910 return -EBUSY;
1911 return 0;
1912}
1913
1914/* Check whether we can create an object with dentry child in directory
1915 * dir.
1916 * 1. We can't do it if child already exists (open has special treatment for
1917 * this case, but since we are inlined it's OK)
1918 * 2. We can't do it if dir is read-only (done in permission())
1919 * 3. We should have write and exec permissions on dir
1920 * 4. We can't do it if dir is immutable (done in permission())
1921 */
a95164d9 1922static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1923{
1924 if (child->d_inode)
1925 return -EEXIST;
1926 if (IS_DEADDIR(dir))
1927 return -ENOENT;
f419a2e3 1928 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1929}
1930
1da177e4
LT		 1931/*
1932 * p1 and p2 should be directories on the same fs.
1933 */
1934struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1935{
1936 struct dentry *p;
1937
1938 if (p1 == p2) {
f2eace23 1939 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1940 return NULL;
1941 }
1942
a11f3a05 1943 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1944
e2761a11
OH		 1945 p = d_ancestor(p2, p1);
1946 if (p) {
1947 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1948 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1949 return p;
1da177e4
LT		 1950 }
1951
e2761a11
OH		 1952 p = d_ancestor(p1, p2);
1953 if (p) {
1954 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1955 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1956 return p;
1da177e4
LT		 1957 }
1958
f2eace23
IM		 1959 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1960 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1961 return NULL;
1962}
1963
1964void unlock_rename(struct dentry *p1, struct dentry *p2)
1965{
1b1dcc1b 1966 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1967 if (p1 != p2) {
1b1dcc1b 1968 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1969 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1970 }
1971}
1972
1973int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1974 struct nameidata *nd)
1975{
a95164d9 1976 int error = may_create(dir, dentry);
1da177e4
LT		 1977
1978 if (error)
1979 return error;
1980
acfa4380 1981 if (!dir->i_op->create)
1da177e4
LT		 1982 return -EACCES; /* shouldn't it be ENOSYS? */
1983 mode &= S_IALLUGO;
1984 mode |= S_IFREG;
1985 error = security_inode_create(dir, dentry, mode);
1986 if (error)
1987 return error;
1da177e4 1988 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1989 if (!error)
f38aa942 1990 fsnotify_create(dir, dentry);
1da177e4
LT		 1991 return error;
1992}
1993
3fb64190 1994int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1995{
3fb64190 1996 struct dentry *dentry = path->dentry;
1da177e4
LT		 1997 struct inode *inode = dentry->d_inode;
1998 int error;
1999
2000 if (!inode)
2001 return -ENOENT;
2002
c8fe8f30
CH		 2003 switch (inode->i_mode & S_IFMT) {
2004 case S_IFLNK:
1da177e4 2005 return -ELOOP;
c8fe8f30
CH		 2006 case S_IFDIR:
2007 if (acc_mode & MAY_WRITE)
2008 return -EISDIR;
2009 break;
2010 case S_IFBLK:
2011 case S_IFCHR:
3fb64190 2012 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2013 return -EACCES;
c8fe8f30
CH		 2014 /*FALLTHRU*/
2015 case S_IFIFO:
2016 case S_IFSOCK:
1da177e4 2017 flag &= ~O_TRUNC;
c8fe8f30 2018 break;
4a3fd211 2019 }
b41572e9 2020
3fb64190 2021 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2022 if (error)
2023 return error;
6146f0d5 2024
1da177e4
LT		 2025 /*
2026 * An append-only file must be opened in append mode for writing.
2027 */
2028 if (IS_APPEND(inode)) {
8737c930 2029 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2030 return -EPERM;
1da177e4 2031 if (flag & O_TRUNC)
7715b521 2032 return -EPERM;
1da177e4
LT		 2033 }
2034
2035 /* O_NOATIME can only be set by the owner or superuser */
7715b521
AV		 2036 if (flag & O_NOATIME && !is_owner_or_cap(inode))
2037 return -EPERM;
1da177e4
LT		 2038
2039 /*
2040 * Ensure there are no outstanding leases on the file.
2041 */
b65a9cfc 2042 return break_lease(inode, flag);
7715b521 2043}
1da177e4 2044
e1181ee6 2045static int handle_truncate(struct file *filp)
7715b521 2046{
e1181ee6 2047 struct path *path = &filp->f_path;
7715b521
AV		 2048 struct inode *inode = path->dentry->d_inode;
2049 int error = get_write_access(inode);
2050 if (error)
2051 return error;
2052 /*
2053 * Refuse to truncate files with mandatory locks held on them.
2054 */
2055 error = locks_verify_locked(inode);
2056 if (!error)
ea0d3ab2 2057 error = security_path_truncate(path);
7715b521
AV		 2058 if (!error) {
2059 error = do_truncate(path->dentry, 0,
2060 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2061 filp);
7715b521
AV		 2062 }
2063 put_write_access(inode);
acd0c935 2064 return error;
1da177e4
LT		 2065}
2066
d57999e1
DH		 2067/*
2068 * Be careful about ever adding any more callers of this
2069 * function. Its flags must be in the namei format, not
2070 * what get passed to sys_open().
2071 */
2072static int __open_namei_create(struct nameidata *nd, struct path *path,
8737c930 2073 int open_flag, int mode)
aab520e2
DH		 2074{
2075 int error;
4ac91378 2076 struct dentry *dir = nd->path.dentry;
aab520e2
DH		 2077
2078 if (!IS_POSIXACL(dir->d_inode))
ce3b0f8d 2079 mode &= ~current_umask();
be6d3e56
KT		 2080 error = security_path_mknod(&nd->path, path->dentry, mode, 0);
2081 if (error)
2082 goto out_unlock;
aab520e2 2083 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
be6d3e56 2084out_unlock:
aab520e2 2085 mutex_unlock(&dir->d_inode->i_mutex);
4ac91378
JB		 2086 dput(nd->path.dentry);
2087 nd->path.dentry = path->dentry;
31e6b01f 2088
aab520e2
DH		 2089 if (error)
2090 return error;
2091 /* Don't check for write permission, don't truncate */
8737c930 2092 return may_open(&nd->path, 0, open_flag & ~O_TRUNC);
aab520e2
DH		 2093}
2094
d57999e1
DH		 2095/*
2096 * Note that while the flag value (low two bits) for sys_open means:
2097 * 00 - read-only
2098 * 01 - write-only
2099 * 10 - read-write
2100 * 11 - special
2101 * it is changed into
2102 * 00 - no permissions needed
2103 * 01 - read-permission
2104 * 10 - write-permission
2105 * 11 - read-write
2106 * for the internal routines (ie open_namei()/follow_link() etc)
2107 * This is more logical, and also allows the 00 "no perm needed"
2108 * to be used for symlinks (where the permissions are checked
2109 * later).
2110 *
2111*/
2112static inline int open_to_namei_flags(int flag)
2113{
2114 if ((flag+1) & O_ACCMODE)
2115 flag++;
2116 return flag;
2117}
2118
7715b521 2119static int open_will_truncate(int flag, struct inode *inode)
4a3fd211
DH		 2120{
2121 /*
2122 * We'll never write to the fs underlying
2123 * a device file.
2124 */
2125 if (special_file(inode->i_mode))
2126 return 0;
2127 return (flag & O_TRUNC);
2128}
2129
648fa861 2130static struct file *finish_open(struct nameidata *nd,
9a66179e 2131 int open_flag, int acc_mode)
648fa861
AV		 2132{
2133 struct file *filp;
2134 int will_truncate;
2135 int error;
2136
9a66179e 2137 will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode);
648fa861
AV		 2138 if (will_truncate) {
2139 error = mnt_want_write(nd->path.mnt);
2140 if (error)
2141 goto exit;
2142 }
2143 error = may_open(&nd->path, acc_mode, open_flag);
2144 if (error) {
2145 if (will_truncate)
2146 mnt_drop_write(nd->path.mnt);
2147 goto exit;
2148 }
2149 filp = nameidata_to_filp(nd);
2150 if (!IS_ERR(filp)) {
2151 error = ima_file_check(filp, acc_mode);
2152 if (error) {
2153 fput(filp);
2154 filp = ERR_PTR(error);
2155 }
2156 }
2157 if (!IS_ERR(filp)) {
648fa861 2158 if (will_truncate) {
e1181ee6 2159 error = handle_truncate(filp);
648fa861
AV		 2160 if (error) {
2161 fput(filp);
2162 filp = ERR_PTR(error);
2163 }
2164 }
2165 }
2166 /*
2167 * It is now safe to drop the mnt write
2168 * because the filp has had a write taken
2169 * on its behalf.
2170 */
2171 if (will_truncate)
2172 mnt_drop_write(nd->path.mnt);
d893f1bc 2173 path_put(&nd->path);
648fa861
AV		 2174 return filp;
2175
2176exit:
648fa861
AV		 2177 path_put(&nd->path);
2178 return ERR_PTR(error);
2179}
2180
31e6b01f
NP		 2181/*
2182 * Handle O_CREAT case for do_filp_open
2183 */
fb1cc555 2184static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2185 const struct open_flags *op, const char *pathname)
fb1cc555 2186{
a1e28038 2187 struct dentry *dir = nd->path.dentry;
fb1cc555 2188 struct file *filp;
16c2cd71 2189 int error;
1f36f774 2190
c3e380b0
AV		 2191 nd->flags &= ~LOOKUP_PARENT;
2192 nd->flags |= op->intent;
2193
1f36f774
AV		 2194 switch (nd->last_type) {
2195 case LAST_DOTDOT:
2196 follow_dotdot(nd);
2197 dir = nd->path.dentry;
176306f5 2198 case LAST_DOT:
1f36f774 2199 /* fallthrough */
1f36f774 2200 case LAST_ROOT:
16c2cd71
AV		 2201 error = handle_reval_path(nd);
2202 if (error)
2203 goto exit;
2204 error = -EISDIR;
31e6b01f 2205 goto exit;
1f36f774 2206 case LAST_BIND:
16c2cd71
AV		 2207 error = handle_reval_path(nd);
2208 if (error)
2209 goto exit;
1f36f774 2210 audit_inode(pathname, dir);
67ee3ad2 2211 goto ok;
1f36f774 2212 }
67ee3ad2 2213
16c2cd71 2214 error = -EISDIR;
1f36f774 2215 /* trailing slashes? */
31e6b01f
NP		 2216 if (nd->last.name[nd->last.len])
2217 goto exit;
a2c36b45 2218
a1e28038
AV		 2219 mutex_lock(&dir->d_inode->i_mutex);
2220
2221 path->dentry = lookup_hash(nd);
2222 path->mnt = nd->path.mnt;
2223
fb1cc555
AV		 2224 error = PTR_ERR(path->dentry);
2225 if (IS_ERR(path->dentry)) {
2226 mutex_unlock(&dir->d_inode->i_mutex);
2227 goto exit;
2228 }
2229
2230 if (IS_ERR(nd->intent.open.file)) {
2231 error = PTR_ERR(nd->intent.open.file);
2232 goto exit_mutex_unlock;
2233 }
2234
2235 /* Negative dentry, just create the file */
2236 if (!path->dentry->d_inode) {
2237 /*
2238 * This write is needed to ensure that a
2239 * ro->rw transition does not occur between
2240 * the time when the file is created and when
2241 * a permanent write count is taken through
2242 * the 'struct file' in nameidata_to_filp().
2243 */
2244 error = mnt_want_write(nd->path.mnt);
2245 if (error)
2246 goto exit_mutex_unlock;
c3e380b0 2247 error = __open_namei_create(nd, path, op->open_flag, op->mode);
fb1cc555
AV		 2248 if (error) {
2249 mnt_drop_write(nd->path.mnt);
2250 goto exit;
2251 }
2252 filp = nameidata_to_filp(nd);
2253 mnt_drop_write(nd->path.mnt);
d893f1bc 2254 path_put(&nd->path);
fb1cc555 2255 if (!IS_ERR(filp)) {
c3e380b0 2256 error = ima_file_check(filp, op->acc_mode);
fb1cc555
AV		 2257 if (error) {
2258 fput(filp);
2259 filp = ERR_PTR(error);
2260 }
2261 }
2262 return filp;
2263 }
2264
2265 /*
2266 * It already exists.
2267 */
2268 mutex_unlock(&dir->d_inode->i_mutex);
2269 audit_inode(pathname, path->dentry);
2270
2271 error = -EEXIST;
c3e380b0 2272 if (op->open_flag & O_EXCL)
fb1cc555
AV		 2273 goto exit_dput;
2274
9875cf80
DH		 2275 error = follow_managed(path, nd->flags);
2276 if (error < 0)
2277 goto exit_dput;
fb1cc555
AV		 2278
2279 error = -ENOENT;
2280 if (!path->dentry->d_inode)
2281 goto exit_dput;
9e67f361
AV		 2282
2283 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2284 return NULL;
fb1cc555
AV		 2285
2286 path_to_nameidata(path, nd);
31e6b01f 2287 nd->inode = path->dentry->d_inode;
fb1cc555 2288 error = -EISDIR;
31e6b01f 2289 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2290 goto exit;
67ee3ad2 2291ok:
c3e380b0 2292 filp = finish_open(nd, op->open_flag, op->acc_mode);
fb1cc555
AV		 2293 return filp;
2294
2295exit_mutex_unlock:
2296 mutex_unlock(&dir->d_inode->i_mutex);
2297exit_dput:
2298 path_put_conditional(path, nd);
2299exit:
fb1cc555
AV		 2300 path_put(&nd->path);
2301 return ERR_PTR(error);
2302}
2303
13aab428 2304static struct file *path_openat(int dfd, const char *pathname,
47c805dc 2305 const struct open_flags *op, int flags)
1da177e4 2306{
4a3fd211 2307 struct file *filp;
a70e65df 2308 struct nameidata nd;
9850c056 2309 struct path path;
1da177e4 2310 int count = 0;
13aab428 2311 int error;
31e6b01f
NP		 2312
2313 filp = get_empty_filp();
2314 if (!filp)
2315 return ERR_PTR(-ENFILE);
2316
47c805dc 2317 filp->f_flags = op->open_flag;
31e6b01f 2318 nd.intent.open.file = filp;
47c805dc
AV		 2319 nd.intent.open.flags = open_to_namei_flags(op->open_flag);
2320 nd.intent.open.create_mode = op->mode;
31e6b01f 2321
47c805dc 2322 if (op->open_flag & O_CREAT)
31e6b01f
NP		 2323 goto creat;
2324
2325 /* !O_CREAT, simple open */
13aab428 2326 error = path_lookupat(dfd, pathname, flags | op->intent, &nd);
31e6b01f 2327 if (unlikely(error))
13aab428 2328 goto out_filp;
31e6b01f
NP		 2329 error = -ELOOP;
2330 if (!(nd.flags & LOOKUP_FOLLOW)) {
2331 if (nd.inode->i_op->follow_link)
13aab428 2332 goto out_path;
31e6b01f
NP		 2333 }
2334 error = -ENOTDIR;
2335 if (nd.flags & LOOKUP_DIRECTORY) {
2336 if (!nd.inode->i_op->lookup)
13aab428 2337 goto out_path;
31e6b01f
NP		 2338 }
2339 audit_inode(pathname, nd.path.dentry);
47c805dc 2340 filp = finish_open(&nd, op->open_flag, op->acc_mode);
2dab5974 2341 release_open_intent(&nd);
31e6b01f
NP		 2342 return filp;
2343
2344creat:
2345 /* OK, have to create the file. Find the parent. */
13aab428 2346 error = path_lookupat(dfd, pathname, LOOKUP_PARENT | flags, &nd);
31e6b01f
NP		 2347 if (unlikely(error))
2348 goto out_filp;
2349 if (unlikely(!audit_dummy_context()))
9b4a9b14 2350 audit_inode(pathname, nd.path.dentry);
1da177e4
LT		 2351
2352 /*
a2c36b45 2353 * We have the parent and last component.
1da177e4 2354 */
47c805dc 2355 filp = do_last(&nd, &path, op, pathname);
806b681c 2356 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa
NP		 2357 struct path link = path;
2358 struct inode *linki = link.dentry->d_inode;
def4af30 2359 void *cookie;
806b681c 2360 error = -ELOOP;
db372915 2361 if (!(nd.flags & LOOKUP_FOLLOW))
1f36f774
AV		 2362 goto exit_dput;
2363 if (count++ == 32)
806b681c
AV		 2364 goto exit_dput;
2365 /*
2366 * This is subtle. Instead of calling do_follow_link() we do
2367 * the thing by hands. The reason is that this way we have zero
2368 * link_count and path_walk() (called from ->follow_link)
2369 * honoring LOOKUP_PARENT. After that we have the parent and
2370 * last component, i.e. we are in the same situation as after
2371 * the first path_walk(). Well, almost - if the last component
2372 * is normal we get its copy stored in nd->last.name and we will
2373 * have to putname() it when we are done. Procfs-like symlinks
2374 * just set LAST_BIND.
2375 */
2376 nd.flags |= LOOKUP_PARENT;
c3e380b0 2377 nd.flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
7b9337aa 2378 error = __do_follow_link(&link, &nd, &cookie);
c3e380b0 2379 if (unlikely(error))
f1afe9ef 2380 filp = ERR_PTR(error);
c3e380b0 2381 else
47c805dc 2382 filp = do_last(&nd, &path, op, pathname);
f1afe9ef 2383 if (!IS_ERR(cookie) && linki->i_op->put_link)
7b9337aa
NP		 2384 linki->i_op->put_link(link.dentry, &nd, cookie);
2385 path_put(&link);
806b681c 2386 }
10fa8e62 2387out:
2a737871
AV		 2388 if (nd.root.mnt)
2389 path_put(&nd.root);
2dab5974 2390 release_open_intent(&nd);
10fa8e62 2391 return filp;
1da177e4 2392
806b681c
AV		 2393exit_dput:
2394 path_put_conditional(&path, &nd);
31e6b01f
NP		 2395out_path:
2396 path_put(&nd.path);
2397out_filp:
806b681c 2398 filp = ERR_PTR(error);
10fa8e62 2399 goto out;
1da177e4
LT		 2400}
2401
13aab428
AV		 2402struct file *do_filp_open(int dfd, const char *pathname,
2403 const struct open_flags *op, int flags)
2404{
2405 struct file *filp;
2406
2407 filp = path_openat(dfd, pathname, op, flags | LOOKUP_RCU);
2408 if (unlikely(filp == ERR_PTR(-ECHILD)))
2409 filp = path_openat(dfd, pathname, op, flags);
2410 if (unlikely(filp == ERR_PTR(-ESTALE)))
2411 filp = path_openat(dfd, pathname, op, flags | LOOKUP_REVAL);
2412 return filp;
2413}
2414
1da177e4
LT		 2415/**
2416 * lookup_create - lookup a dentry, creating it if it doesn't exist
2417 * @nd: nameidata info
2418 * @is_dir: directory flag
2419 *
2420 * Simple function to lookup and return a dentry and create it
2421 * if it doesn't exist. Is SMP-safe.
c663e5d8 2422 *
4ac91378 2423 * Returns with nd->path.dentry->d_inode->i_mutex locked.
1da177e4
LT		 2424 */
2425struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2426{
c663e5d8 2427 struct dentry *dentry = ERR_PTR(-EEXIST);
1da177e4 2428
4ac91378 2429 mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
c663e5d8
CH		 2430 /*
2431 * Yucky last component or no last component at all?
2432 * (foo/., foo/.., /////)
2433 */
1da177e4
LT		 2434 if (nd->last_type != LAST_NORM)
2435 goto fail;
2436 nd->flags &= ~LOOKUP_PARENT;
3516586a 2437 nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
a634904a 2438 nd->intent.open.flags = O_EXCL;
c663e5d8
CH		 2439
2440 /*
2441 * Do the final lookup.
2442 */
49705b77 2443 dentry = lookup_hash(nd);
1da177e4
LT		 2444 if (IS_ERR(dentry))
2445 goto fail;
c663e5d8 2446
e9baf6e5
AV		 2447 if (dentry->d_inode)
2448 goto eexist;
c663e5d8
CH		 2449 /*
2450 * Special case - lookup gave negative, but... we had foo/bar/
2451 * From the vfs_mknod() POV we just have a negative dentry -
2452 * all is fine. Let's be bastards - you had / on the end, you've
2453 * been asking for (non-existent) directory. -ENOENT for you.
2454 */
e9baf6e5
AV		 2455 if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2456 dput(dentry);
2457 dentry = ERR_PTR(-ENOENT);
2458 }
1da177e4 2459 return dentry;
e9baf6e5 2460eexist:
1da177e4 2461 dput(dentry);
e9baf6e5 2462 dentry = ERR_PTR(-EEXIST);
1da177e4
LT		 2463fail:
2464 return dentry;
2465}
f81a0bff 2466EXPORT_SYMBOL_GPL(lookup_create);
1da177e4
LT		 2467
2468int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2469{
a95164d9 2470 int error = may_create(dir, dentry);
1da177e4
LT		 2471
2472 if (error)
2473 return error;
2474
2475 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2476 return -EPERM;
2477
acfa4380 2478 if (!dir->i_op->mknod)
1da177e4
LT		 2479 return -EPERM;
2480
08ce5f16
SH		 2481 error = devcgroup_inode_mknod(mode, dev);
2482 if (error)
2483 return error;
2484
1da177e4
LT		 2485 error = security_inode_mknod(dir, dentry, mode, dev);
2486 if (error)
2487 return error;
2488
1da177e4 2489 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2490 if (!error)
f38aa942 2491 fsnotify_create(dir, dentry);
1da177e4
LT		 2492 return error;
2493}
2494
463c3197
DH		 2495static int may_mknod(mode_t mode)
2496{
2497 switch (mode & S_IFMT) {
2498 case S_IFREG:
2499 case S_IFCHR:
2500 case S_IFBLK:
2501 case S_IFIFO:
2502 case S_IFSOCK:
2503 case 0: /* zero mode translates to S_IFREG */
2504 return 0;
2505 case S_IFDIR:
2506 return -EPERM;
2507 default:
2508 return -EINVAL;
2509 }
2510}
2511
2e4d0924
HC		 2512SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2513 unsigned, dev)
1da177e4 2514{
2ad94ae6
AV		 2515 int error;
2516 char *tmp;
2517 struct dentry *dentry;
1da177e4
LT		 2518 struct nameidata nd;
2519
2520 if (S_ISDIR(mode))
2521 return -EPERM;
1da177e4 2522
2ad94ae6 2523 error = user_path_parent(dfd, filename, &nd, &tmp);
1da177e4 2524 if (error)
2ad94ae6
AV		 2525 return error;
2526
1da177e4 2527 dentry = lookup_create(&nd, 0);
463c3197
DH		 2528 if (IS_ERR(dentry)) {
2529 error = PTR_ERR(dentry);
2530 goto out_unlock;
2531 }
4ac91378 2532 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2533 mode &= ~current_umask();
463c3197
DH		 2534 error = may_mknod(mode);
2535 if (error)
2536 goto out_dput;
2537 error = mnt_want_write(nd.path.mnt);
2538 if (error)
2539 goto out_dput;
be6d3e56
KT		 2540 error = security_path_mknod(&nd.path, dentry, mode, dev);
2541 if (error)
2542 goto out_drop_write;
463c3197 2543 switch (mode & S_IFMT) {
1da177e4 2544 case 0: case S_IFREG:
4ac91378 2545 error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
1da177e4
LT		 2546 break;
2547 case S_IFCHR: case S_IFBLK:
4ac91378 2548 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2549 new_decode_dev(dev));
2550 break;
2551 case S_IFIFO: case S_IFSOCK:
4ac91378 2552 error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
1da177e4 2553 break;
1da177e4 2554 }
be6d3e56 2555out_drop_write:
463c3197
DH		 2556 mnt_drop_write(nd.path.mnt);
2557out_dput:
2558 dput(dentry);
2559out_unlock:
4ac91378 2560 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2561 path_put(&nd.path);
1da177e4
LT		 2562 putname(tmp);
2563
2564 return error;
2565}
2566
3480b257 2567SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2568{
2569 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2570}
2571
1da177e4
LT		 2572int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2573{
a95164d9 2574 int error = may_create(dir, dentry);
1da177e4
LT		 2575
2576 if (error)
2577 return error;
2578
acfa4380 2579 if (!dir->i_op->mkdir)
1da177e4
LT		 2580 return -EPERM;
2581
2582 mode &= (S_IRWXUGO|S_ISVTX);
2583 error = security_inode_mkdir(dir, dentry, mode);
2584 if (error)
2585 return error;
2586
1da177e4 2587 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2588 if (!error)
f38aa942 2589 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2590 return error;
2591}
2592
2e4d0924 2593SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4
LT		 2594{
2595 int error = 0;
2596 char * tmp;
6902d925
DH		 2597 struct dentry *dentry;
2598 struct nameidata nd;
1da177e4 2599
2ad94ae6
AV		 2600 error = user_path_parent(dfd, pathname, &nd, &tmp);
2601 if (error)
6902d925 2602 goto out_err;
1da177e4 2603
6902d925
DH		 2604 dentry = lookup_create(&nd, 1);
2605 error = PTR_ERR(dentry);
2606 if (IS_ERR(dentry))
2607 goto out_unlock;
1da177e4 2608
4ac91378 2609 if (!IS_POSIXACL(nd.path.dentry->d_inode))
ce3b0f8d 2610 mode &= ~current_umask();
463c3197
DH		 2611 error = mnt_want_write(nd.path.mnt);
2612 if (error)
2613 goto out_dput;
be6d3e56
KT		 2614 error = security_path_mkdir(&nd.path, dentry, mode);
2615 if (error)
2616 goto out_drop_write;
4ac91378 2617 error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
be6d3e56 2618out_drop_write:
463c3197
DH		 2619 mnt_drop_write(nd.path.mnt);
2620out_dput:
6902d925
DH		 2621 dput(dentry);
2622out_unlock:
4ac91378 2623 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2624 path_put(&nd.path);
6902d925
DH		 2625 putname(tmp);
2626out_err:
1da177e4
LT		 2627 return error;
2628}
2629
3cdad428 2630SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2631{
2632 return sys_mkdirat(AT_FDCWD, pathname, mode);
2633}
2634
1da177e4
LT		 2635/*
2636 * We try to drop the dentry early: we should have
2637 * a usage count of 2 if we're the only user of this
2638 * dentry, and if that is true (possibly after pruning
2639 * the dcache), then we drop the dentry now.
2640 *
2641 * A low-level filesystem can, if it choses, legally
2642 * do a
2643 *
2644 * if (!d_unhashed(dentry))
2645 * return -EBUSY;
2646 *
2647 * if it cannot handle the case of removing a directory
2648 * that is still in use by something else..
2649 */
2650void dentry_unhash(struct dentry *dentry)
2651{
2652 dget(dentry);
dc168427 2653 shrink_dcache_parent(dentry);
1da177e4 2654 spin_lock(&dentry->d_lock);
b7ab39f6 2655 if (dentry->d_count == 2)
1da177e4
LT		 2656 __d_drop(dentry);
2657 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2658}
2659
2660int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2661{
2662 int error = may_delete(dir, dentry, 1);
2663
2664 if (error)
2665 return error;
2666
acfa4380 2667 if (!dir->i_op->rmdir)
1da177e4
LT		 2668 return -EPERM;
2669
1b1dcc1b 2670 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2671 dentry_unhash(dentry);
2672 if (d_mountpoint(dentry))
2673 error = -EBUSY;
2674 else {
2675 error = security_inode_rmdir(dir, dentry);
2676 if (!error) {
2677 error = dir->i_op->rmdir(dir, dentry);
d83c49f3 2678 if (!error) {
1da177e4 2679 dentry->d_inode->i_flags |= S_DEAD;
d83c49f3
AV		 2680 dont_mount(dentry);
2681 }
1da177e4
LT		 2682 }
2683 }
1b1dcc1b 2684 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4 2685 if (!error) {
1da177e4
LT		 2686 d_delete(dentry);
2687 }
2688 dput(dentry);
2689
2690 return error;
2691}
2692
5590ff0d 2693static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2694{
2695 int error = 0;
2696 char * name;
2697 struct dentry *dentry;
2698 struct nameidata nd;
2699
2ad94ae6 2700 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2701 if (error)
2ad94ae6 2702 return error;
1da177e4
LT		 2703
2704 switch(nd.last_type) {
0612d9fb
OH		 2705 case LAST_DOTDOT:
2706 error = -ENOTEMPTY;
2707 goto exit1;
2708 case LAST_DOT:
2709 error = -EINVAL;
2710 goto exit1;
2711 case LAST_ROOT:
2712 error = -EBUSY;
2713 goto exit1;
1da177e4 2714 }
0612d9fb
OH		 2715
2716 nd.flags &= ~LOOKUP_PARENT;
2717
4ac91378 2718 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2719 dentry = lookup_hash(&nd);
1da177e4 2720 error = PTR_ERR(dentry);
6902d925
DH		 2721 if (IS_ERR(dentry))
2722 goto exit2;
0622753b
DH		 2723 error = mnt_want_write(nd.path.mnt);
2724 if (error)
2725 goto exit3;
be6d3e56
KT		 2726 error = security_path_rmdir(&nd.path, dentry);
2727 if (error)
2728 goto exit4;
4ac91378 2729 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2730exit4:
0622753b
DH		 2731 mnt_drop_write(nd.path.mnt);
2732exit3:
6902d925
DH		 2733 dput(dentry);
2734exit2:
4ac91378 2735 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2736exit1:
1d957f9b 2737 path_put(&nd.path);
1da177e4
LT		 2738 putname(name);
2739 return error;
2740}
2741
3cdad428 2742SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2743{
2744 return do_rmdir(AT_FDCWD, pathname);
2745}
2746
1da177e4
LT		 2747int vfs_unlink(struct inode *dir, struct dentry *dentry)
2748{
2749 int error = may_delete(dir, dentry, 0);
2750
2751 if (error)
2752 return error;
2753
acfa4380 2754 if (!dir->i_op->unlink)
1da177e4
LT		 2755 return -EPERM;
2756
1b1dcc1b 2757 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2758 if (d_mountpoint(dentry))
2759 error = -EBUSY;
2760 else {
2761 error = security_inode_unlink(dir, dentry);
bec1052e 2762 if (!error) {
1da177e4 2763 error = dir->i_op->unlink(dir, dentry);
bec1052e 2764 if (!error)
d83c49f3 2765 dont_mount(dentry);
bec1052e 2766 }
1da177e4 2767 }
1b1dcc1b 2768 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2769
2770 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2771 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2772 fsnotify_link_count(dentry->d_inode);
e234f35c 2773 d_delete(dentry);
1da177e4 2774 }
0eeca283 2775
1da177e4
LT		 2776 return error;
2777}
2778
2779/*
2780 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2781 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2782 * writeout happening, and we don't want to prevent access to the directory
2783 * while waiting on the I/O.
2784 */
5590ff0d 2785static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2786{
2ad94ae6
AV		 2787 int error;
2788 char *name;
1da177e4
LT		 2789 struct dentry *dentry;
2790 struct nameidata nd;
2791 struct inode *inode = NULL;
2792
2ad94ae6 2793 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2794 if (error)
2ad94ae6
AV		 2795 return error;
2796
1da177e4
LT		 2797 error = -EISDIR;
2798 if (nd.last_type != LAST_NORM)
2799 goto exit1;
0612d9fb
OH		 2800
2801 nd.flags &= ~LOOKUP_PARENT;
2802
4ac91378 2803 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2804 dentry = lookup_hash(&nd);
1da177e4
LT		 2805 error = PTR_ERR(dentry);
2806 if (!IS_ERR(dentry)) {
2807 /* Why not before? Because we want correct error value */
2808 if (nd.last.name[nd.last.len])
2809 goto slashes;
2810 inode = dentry->d_inode;
2811 if (inode)
7de9c6ee 2812 ihold(inode);
0622753b
DH		 2813 error = mnt_want_write(nd.path.mnt);
2814 if (error)
2815 goto exit2;
be6d3e56
KT		 2816 error = security_path_unlink(&nd.path, dentry);
2817 if (error)
2818 goto exit3;
4ac91378 2819 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2820exit3:
0622753b 2821 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2822 exit2:
2823 dput(dentry);
2824 }
4ac91378 2825 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2826 if (inode)
2827 iput(inode); /* truncate the inode here */
2828exit1:
1d957f9b 2829 path_put(&nd.path);
1da177e4
LT		 2830 putname(name);
2831 return error;
2832
2833slashes:
2834 error = !dentry->d_inode ? -ENOENT :
2835 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2836 goto exit2;
2837}
2838
2e4d0924 2839SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2840{
2841 if ((flag & ~AT_REMOVEDIR) != 0)
2842 return -EINVAL;
2843
2844 if (flag & AT_REMOVEDIR)
2845 return do_rmdir(dfd, pathname);
2846
2847 return do_unlinkat(dfd, pathname);
2848}
2849
3480b257 2850SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2851{
2852 return do_unlinkat(AT_FDCWD, pathname);
2853}
2854
db2e747b 2855int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2856{
a95164d9 2857 int error = may_create(dir, dentry);
1da177e4
LT		 2858
2859 if (error)
2860 return error;
2861
acfa4380 2862 if (!dir->i_op->symlink)
1da177e4
LT		 2863 return -EPERM;
2864
2865 error = security_inode_symlink(dir, dentry, oldname);
2866 if (error)
2867 return error;
2868
1da177e4 2869 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2870 if (!error)
f38aa942 2871 fsnotify_create(dir, dentry);
1da177e4
LT		 2872 return error;
2873}
2874
2e4d0924
HC		 2875SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2876 int, newdfd, const char __user *, newname)
1da177e4 2877{
2ad94ae6
AV		 2878 int error;
2879 char *from;
2880 char *to;
6902d925
DH		 2881 struct dentry *dentry;
2882 struct nameidata nd;
1da177e4
LT		 2883
2884 from = getname(oldname);
2ad94ae6 2885 if (IS_ERR(from))
1da177e4 2886 return PTR_ERR(from);
1da177e4 2887
2ad94ae6 2888 error = user_path_parent(newdfd, newname, &nd, &to);
6902d925 2889 if (error)
2ad94ae6
AV		 2890 goto out_putname;
2891
6902d925
DH		 2892 dentry = lookup_create(&nd, 0);
2893 error = PTR_ERR(dentry);
2894 if (IS_ERR(dentry))
2895 goto out_unlock;
2896
75c3f29d
DH		 2897 error = mnt_want_write(nd.path.mnt);
2898 if (error)
2899 goto out_dput;
be6d3e56
KT		 2900 error = security_path_symlink(&nd.path, dentry, from);
2901 if (error)
2902 goto out_drop_write;
db2e747b 2903 error = vfs_symlink(nd.path.dentry->d_inode, dentry, from);
be6d3e56 2904out_drop_write:
75c3f29d
DH		 2905 mnt_drop_write(nd.path.mnt);
2906out_dput:
6902d925
DH		 2907 dput(dentry);
2908out_unlock:
4ac91378 2909 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1d957f9b 2910 path_put(&nd.path);
6902d925
DH		 2911 putname(to);
2912out_putname:
1da177e4
LT		 2913 putname(from);
2914 return error;
2915}
2916
3480b257 2917SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2918{
2919 return sys_symlinkat(oldname, AT_FDCWD, newname);
2920}
2921
1da177e4
LT		 2922int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2923{
2924 struct inode *inode = old_dentry->d_inode;
2925 int error;
2926
2927 if (!inode)
2928 return -ENOENT;
2929
a95164d9 2930 error = may_create(dir, new_dentry);
1da177e4
LT		 2931 if (error)
2932 return error;
2933
2934 if (dir->i_sb != inode->i_sb)
2935 return -EXDEV;
2936
2937 /*
2938 * A link to an append-only or immutable file cannot be created.
2939 */
2940 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2941 return -EPERM;
acfa4380 2942 if (!dir->i_op->link)
1da177e4 2943 return -EPERM;
7e79eedb 2944 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2945 return -EPERM;
2946
2947 error = security_inode_link(old_dentry, dir, new_dentry);
2948 if (error)
2949 return error;
2950
7e79eedb 2951 mutex_lock(&inode->i_mutex);
1da177e4 2952 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2953 mutex_unlock(&inode->i_mutex);
e31e14ec 2954 if (!error)
7e79eedb 2955 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2956 return error;
2957}
2958
2959/*
2960 * Hardlinks are often used in delicate situations. We avoid
2961 * security-related surprises by not following symlinks on the
2962 * newname. --KAB
2963 *
2964 * We don't follow them on the oldname either to be compatible
2965 * with linux 2.0, and to avoid hard-linking to directories
2966 * and other special files. --ADM
2967 */
2e4d0924
HC		 2968SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2969 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2970{
2971 struct dentry *new_dentry;
2d8f3038
AV		 2972 struct nameidata nd;
2973 struct path old_path;
1da177e4 2974 int error;
2ad94ae6 2975 char *to;
1da177e4 2976
45c9b11a 2977 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
c04030e1
UD		 2978 return -EINVAL;
2979
2d8f3038
AV		 2980 error = user_path_at(olddfd, oldname,
2981 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2982 &old_path);
1da177e4 2983 if (error)
2ad94ae6
AV		 2984 return error;
2985
2986 error = user_path_parent(newdfd, newname, &nd, &to);
1da177e4
LT		 2987 if (error)
2988 goto out;
2989 error = -EXDEV;
2d8f3038 2990 if (old_path.mnt != nd.path.mnt)
1da177e4
LT		 2991 goto out_release;
2992 new_dentry = lookup_create(&nd, 0);
2993 error = PTR_ERR(new_dentry);
6902d925
DH		 2994 if (IS_ERR(new_dentry))
2995 goto out_unlock;
75c3f29d
DH		 2996 error = mnt_want_write(nd.path.mnt);
2997 if (error)
2998 goto out_dput;
be6d3e56
KT		 2999 error = security_path_link(old_path.dentry, &nd.path, new_dentry);
3000 if (error)
3001 goto out_drop_write;
2d8f3038 3002 error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry);
be6d3e56 3003out_drop_write:
75c3f29d
DH		 3004 mnt_drop_write(nd.path.mnt);
3005out_dput:
6902d925
DH		 3006 dput(new_dentry);
3007out_unlock:
4ac91378 3008 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 3009out_release:
1d957f9b 3010 path_put(&nd.path);
2ad94ae6 3011 putname(to);
1da177e4 3012out:
2d8f3038 3013 path_put(&old_path);
1da177e4
LT		 3014
3015 return error;
3016}
3017
3480b257 3018SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3019{
c04030e1 3020 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3021}
3022
1da177e4
LT		 3023/*
3024 * The worst of all namespace operations - renaming directory. "Perverted"
3025 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3026 * Problems:
3027 * a) we can get into loop creation. Check is done in is_subdir().
3028 * b) race potential - two innocent renames can create a loop together.
3029 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3030 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3031 * story.
3032 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3033 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3034 * whether the target exists). Solution: try to be smart with locking
3035 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3036 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3037 * move will be locked. Thus we can rank directories by the tree
3038 * (ancestors first) and rank all non-directories after them.
3039 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3040 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3041 * HOWEVER, it relies on the assumption that any object with ->lookup()
3042 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3043 * we'd better make sure that there's no link(2) for them.
3044 * d) some filesystems don't support opened-but-unlinked directories,
3045 * either because of layout or because they are not ready to deal with
3046 * all cases correctly. The latter will be fixed (taking this sort of
3047 * stuff into VFS), but the former is not going away. Solution: the same
3048 * trick as in rmdir().
3049 * e) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3050 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3051 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3052 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3053 * locking].
3054 */
75c96f85
AB		 3055static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3056 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3057{
3058 int error = 0;
3059 struct inode *target;
3060
3061 /*
3062 * If we are going to change the parent - check write permissions,
3063 * we'll need to flip '..'.
3064 */
3065 if (new_dir != old_dir) {
f419a2e3 3066 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3067 if (error)
3068 return error;
3069 }
3070
3071 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3072 if (error)
3073 return error;
3074
3075 target = new_dentry->d_inode;
d83c49f3 3076 if (target)
1b1dcc1b 3077 mutex_lock(&target->i_mutex);
1da177e4
LT		 3078 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3079 error = -EBUSY;
d83c49f3
AV		 3080 else {
3081 if (target)
3082 dentry_unhash(new_dentry);
1da177e4 3083 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
d83c49f3 3084 }
1da177e4 3085 if (target) {
d83c49f3 3086 if (!error) {
1da177e4 3087 target->i_flags |= S_DEAD;
d83c49f3
AV		 3088 dont_mount(new_dentry);
3089 }
1b1dcc1b 3090 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3091 if (d_unhashed(new_dentry))
3092 d_rehash(new_dentry);
3093 dput(new_dentry);
3094 }
e31e14ec 3095 if (!error)
349457cc
MF		 3096 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3097 d_move(old_dentry,new_dentry);
1da177e4
LT		 3098 return error;
3099}
3100
75c96f85
AB		 3101static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3102 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3103{
3104 struct inode *target;
3105 int error;
3106
3107 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3108 if (error)
3109 return error;
3110
3111 dget(new_dentry);
3112 target = new_dentry->d_inode;
3113 if (target)
1b1dcc1b 3114 mutex_lock(&target->i_mutex);
1da177e4
LT		 3115 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3116 error = -EBUSY;
3117 else
3118 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3119 if (!error) {
bec1052e 3120 if (target)
d83c49f3 3121 dont_mount(new_dentry);
349457cc 3122 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
1da177e4 3123 d_move(old_dentry, new_dentry);
1da177e4
LT		 3124 }
3125 if (target)
1b1dcc1b 3126 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3127 dput(new_dentry);
3128 return error;
3129}
3130
3131int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3132 struct inode *new_dir, struct dentry *new_dentry)
3133{
3134 int error;
3135 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3136 const unsigned char *old_name;
1da177e4
LT		 3137
3138 if (old_dentry->d_inode == new_dentry->d_inode)
3139 return 0;
3140
3141 error = may_delete(old_dir, old_dentry, is_dir);
3142 if (error)
3143 return error;
3144
3145 if (!new_dentry->d_inode)
a95164d9 3146 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3147 else
3148 error = may_delete(new_dir, new_dentry, is_dir);
3149 if (error)
3150 return error;
3151
acfa4380 3152 if (!old_dir->i_op->rename)
1da177e4
LT		 3153 return -EPERM;
3154
0eeca283
RL		 3155 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3156
1da177e4
LT		 3157 if (is_dir)
3158 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3159 else
3160 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3161 if (!error)
3162 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3163 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3164 fsnotify_oldname_free(old_name);
3165
1da177e4
LT		 3166 return error;
3167}
3168
2e4d0924
HC		 3169SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3170 int, newdfd, const char __user *, newname)
1da177e4 3171{
2ad94ae6
AV		 3172 struct dentry *old_dir, *new_dir;
3173 struct dentry *old_dentry, *new_dentry;
3174 struct dentry *trap;
1da177e4 3175 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3176 char *from;
3177 char *to;
3178 int error;
1da177e4 3179
2ad94ae6 3180 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3181 if (error)
3182 goto exit;
3183
2ad94ae6 3184 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3185 if (error)
3186 goto exit1;
3187
3188 error = -EXDEV;
4ac91378 3189 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3190 goto exit2;
3191
4ac91378 3192 old_dir = oldnd.path.dentry;
1da177e4
LT		 3193 error = -EBUSY;
3194 if (oldnd.last_type != LAST_NORM)
3195 goto exit2;
3196
4ac91378 3197 new_dir = newnd.path.dentry;
1da177e4
LT		 3198 if (newnd.last_type != LAST_NORM)
3199 goto exit2;
3200
0612d9fb
OH		 3201 oldnd.flags &= ~LOOKUP_PARENT;
3202 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3203 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3204
1da177e4
LT		 3205 trap = lock_rename(new_dir, old_dir);
3206
49705b77 3207 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3208 error = PTR_ERR(old_dentry);
3209 if (IS_ERR(old_dentry))
3210 goto exit3;
3211 /* source must exist */
3212 error = -ENOENT;
3213 if (!old_dentry->d_inode)
3214 goto exit4;
3215 /* unless the source is a directory trailing slashes give -ENOTDIR */
3216 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3217 error = -ENOTDIR;
3218 if (oldnd.last.name[oldnd.last.len])
3219 goto exit4;
3220 if (newnd.last.name[newnd.last.len])
3221 goto exit4;
3222 }
3223 /* source should not be ancestor of target */
3224 error = -EINVAL;
3225 if (old_dentry == trap)
3226 goto exit4;
49705b77 3227 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3228 error = PTR_ERR(new_dentry);
3229 if (IS_ERR(new_dentry))
3230 goto exit4;
3231 /* target should not be an ancestor of source */
3232 error = -ENOTEMPTY;
3233 if (new_dentry == trap)
3234 goto exit5;
3235
9079b1eb
DH		 3236 error = mnt_want_write(oldnd.path.mnt);
3237 if (error)
3238 goto exit5;
be6d3e56
KT		 3239 error = security_path_rename(&oldnd.path, old_dentry,
3240 &newnd.path, new_dentry);
3241 if (error)
3242 goto exit6;
1da177e4
LT		 3243 error = vfs_rename(old_dir->d_inode, old_dentry,
3244 new_dir->d_inode, new_dentry);
be6d3e56 3245exit6:
9079b1eb 3246 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3247exit5:
3248 dput(new_dentry);
3249exit4:
3250 dput(old_dentry);
3251exit3:
3252 unlock_rename(new_dir, old_dir);
3253exit2:
1d957f9b 3254 path_put(&newnd.path);
2ad94ae6 3255 putname(to);
1da177e4 3256exit1:
1d957f9b 3257 path_put(&oldnd.path);
1da177e4 3258 putname(from);
2ad94ae6 3259exit:
1da177e4
LT		 3260 return error;
3261}
3262
a26eab24 3263SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3264{
3265 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3266}
3267
1da177e4
LT		 3268int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3269{
3270 int len;
3271
3272 len = PTR_ERR(link);
3273 if (IS_ERR(link))
3274 goto out;
3275
3276 len = strlen(link);
3277 if (len > (unsigned) buflen)
3278 len = buflen;
3279 if (copy_to_user(buffer, link, len))
3280 len = -EFAULT;
3281out:
3282 return len;
3283}
3284
3285/*
3286 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3287 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3288 * using) it for any given inode is up to filesystem.
3289 */
3290int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3291{
3292 struct nameidata nd;
cc314eef 3293 void *cookie;
694a1764 3294 int res;
cc314eef 3295
1da177e4 3296 nd.depth = 0;
cc314eef 3297 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3298 if (IS_ERR(cookie))
3299 return PTR_ERR(cookie);
3300
3301 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3302 if (dentry->d_inode->i_op->put_link)
3303 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3304 return res;
1da177e4
LT		 3305}
3306
3307int vfs_follow_link(struct nameidata *nd, const char *link)
3308{
3309 return __vfs_follow_link(nd, link);
3310}
3311
3312/* get the link contents into pagecache */
3313static char *page_getlink(struct dentry * dentry, struct page **ppage)
3314{
ebd09abb
DG		 3315 char *kaddr;
3316 struct page *page;
1da177e4 3317 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3318 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3319 if (IS_ERR(page))
6fe6900e 3320 return (char*)page;
1da177e4 3321 *ppage = page;
ebd09abb
DG		 3322 kaddr = kmap(page);
3323 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3324 return kaddr;
1da177e4
LT		 3325}
3326
3327int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3328{
3329 struct page *page = NULL;
3330 char *s = page_getlink(dentry, &page);
3331 int res = vfs_readlink(dentry,buffer,buflen,s);
3332 if (page) {
3333 kunmap(page);
3334 page_cache_release(page);
3335 }
3336 return res;
3337}
3338
cc314eef 3339void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3340{
cc314eef 3341 struct page *page = NULL;
1da177e4 3342 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3343 return page;
1da177e4
LT		 3344}
3345
cc314eef 3346void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3347{
cc314eef
LT		 3348 struct page *page = cookie;
3349
3350 if (page) {
1da177e4
LT		 3351 kunmap(page);
3352 page_cache_release(page);
1da177e4
LT		 3353 }
3354}
3355
54566b2c
NP		 3356/*
3357 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3358 */
3359int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3360{
3361 struct address_space *mapping = inode->i_mapping;
0adb25d2 3362 struct page *page;
afddba49 3363 void *fsdata;
beb497ab 3364 int err;
1da177e4 3365 char *kaddr;
54566b2c
NP		 3366 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3367 if (nofs)
3368 flags |= AOP_FLAG_NOFS;
1da177e4 3369
7e53cac4 3370retry:
afddba49 3371 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3372 flags, &page, &fsdata);
1da177e4 3373 if (err)
afddba49
NP		 3374 goto fail;
3375
1da177e4
LT		 3376 kaddr = kmap_atomic(page, KM_USER0);
3377 memcpy(kaddr, symname, len-1);
3378 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3379
3380 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3381 page, fsdata);
1da177e4
LT		 3382 if (err < 0)
3383 goto fail;
afddba49
NP		 3384 if (err < len-1)
3385 goto retry;
3386
1da177e4
LT		 3387 mark_inode_dirty(inode);
3388 return 0;
1da177e4
LT		 3389fail:
3390 return err;
3391}
3392
0adb25d2
KK		 3393int page_symlink(struct inode *inode, const char *symname, int len)
3394{
3395 return __page_symlink(inode, symname, len,
54566b2c 3396 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3397}
3398
92e1d5be 3399const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3400 .readlink = generic_readlink,
3401 .follow_link = page_follow_link_light,
3402 .put_link = page_put_link,
3403};
3404
2d8f3038 3405EXPORT_SYMBOL(user_path_at);
cc53ce53 3406EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3407EXPORT_SYMBOL(follow_down);
3408EXPORT_SYMBOL(follow_up);
3409EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3410EXPORT_SYMBOL(getname);
3411EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3412EXPORT_SYMBOL(lookup_one_len);
3413EXPORT_SYMBOL(page_follow_link_light);
3414EXPORT_SYMBOL(page_put_link);
3415EXPORT_SYMBOL(page_readlink);
0adb25d2 3416EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3417EXPORT_SYMBOL(page_symlink);
3418EXPORT_SYMBOL(page_symlink_inode_operations);
c9c6cac0 3419EXPORT_SYMBOL(kern_path_parent);
d1811465 3420EXPORT_SYMBOL(kern_path);
16f18200 3421EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3422EXPORT_SYMBOL(inode_permission);
8c744fb8 3423EXPORT_SYMBOL(file_permission);
1da177e4
LT		 3424EXPORT_SYMBOL(unlock_rename);
3425EXPORT_SYMBOL(vfs_create);
3426EXPORT_SYMBOL(vfs_follow_link);
3427EXPORT_SYMBOL(vfs_link);
3428EXPORT_SYMBOL(vfs_mkdir);
3429EXPORT_SYMBOL(vfs_mknod);
3430EXPORT_SYMBOL(generic_permission);
3431EXPORT_SYMBOL(vfs_readlink);
3432EXPORT_SYMBOL(vfs_rename);
3433EXPORT_SYMBOL(vfs_rmdir);
3434EXPORT_SYMBOL(vfs_symlink);
3435EXPORT_SYMBOL(vfs_unlink);
3436EXPORT_SYMBOL(dentry_unhash);
3437EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)